diff --git "a/exp/log/log-train-2023-02-28-10-38-34-0" "b/exp/log/log-train-2023-02-28-10-38-34-0"
new file mode 100644--- /dev/null
+++ "b/exp/log/log-train-2023-02-28-10-38-34-0"
@@ -0,0 +1,96647 @@
+2023-02-28 10:38:34,447 INFO [train.py:1094] (0/2) Training started
+2023-02-28 10:38:34,450 INFO [train.py:1104] (0/2) Device: cuda:0
+2023-02-28 10:38:34,452 INFO [train.py:1113] (0/2) {'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 50, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.22', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '96c9a2aece2a3a7633da07740e24fa3d96f5498c', 'k2-git-date': 'Thu Nov 10 08:14:02 2022', 'lhotse-version': '1.13.0.dev+git.527d964.clean', 'torch-version': '1.12.1', 'torch-cuda-available': True, 'torch-cuda-version': '11.6', 'python-version': '3.8', 'icefall-git-branch': 'zipformer_libri_small_models', 'icefall-git-sha1': 'd3145cd-dirty', 'icefall-git-date': 'Thu Feb 16 15:24:55 2023', 'icefall-path': '/ceph-data4/yangxiaoyu/softwares/icefall_development/icefall_small_models', 'k2-path': '/ceph-data4/yangxiaoyu/softwares/anaconda3/envs/k2_latest/lib/python3.8/site-packages/k2/__init__.py', 'lhotse-path': '/ceph-data4/yangxiaoyu/softwares/lhotse_development/lhotse_random_padding_left/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-2-1216192652-5bcf7587b4-n6q9m', 'IP address': '10.177.74.211'}, 'world_size': 2, 'master_port': 12346, 'tensorboard': True, 'full_libri': True, 'num_epochs': 30, 'start_epoch': 1, 'start_batch': 0, 'exp_dir': PosixPath('pruned_transducer_stateless7_streaming_multi/exp-small-6M'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'base_lr': 0.05, 'lr_batches': 5000, 'lr_epochs': 3.5, 'context_size': 2, 'prune_range': 5, 'lm_scale': 0.25, 'am_scale': 0.0, 'simple_loss_scale': 0.5, 'seed': 42, 'print_diagnostics': False, 'inf_check': False, 'save_every_n': 2000, 'keep_last_k': 30, 'average_period': 200, 'use_fp16': True, 'giga_prob': 0.9, 'num_encoder_layers': '2,2,2,2,2', 'feedforward_dims': '256,256,512,512,256', 'nhead': '4,4,4,4,4', 'encoder_dims': '128,128,128,128,128', 'attention_dims': '96,96,96,96,96', 'encoder_unmasked_dims': '96,96,96,96,96', 'zipformer_downsampling_factors': '1,2,4,8,2', 'cnn_module_kernels': '31,31,31,31,31', 'decoder_dim': 512, 'joiner_dim': 512, 'short_chunk_size': 50, 'num_left_chunks': 4, 'decode_chunk_len': 32, 'max_duration': 1200, 'bucketing_sampler': True, 'num_buckets': 30, 'shuffle': True, 'return_cuts': True, 'num_workers': 2, 'on_the_fly_num_workers': 0, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'manifest_dir': PosixPath('data/fbank'), 'on_the_fly_feats': False, 'blank_id': 0, 'vocab_size': 500}
+2023-02-28 10:38:34,452 INFO [train.py:1115] (0/2) About to create model
+2023-02-28 10:38:34,598 INFO [zipformer.py:405] (0/2) At encoder stack 4, which has downsampling_factor=2, we will combine the outputs of layers 1 and 3, with downsampling_factors=2 and 8.
+2023-02-28 10:38:34,606 INFO [train.py:536] (0/2) Use giga
+2023-02-28 10:38:34,611 INFO [train.py:1119] (0/2) Number of model parameters: 6061029
+2023-02-28 10:38:36,330 INFO [train.py:1134] (0/2) Using DDP
+2023-02-28 10:38:36,475 INFO [librispeech.py:43] (0/2) About to get train-clean-100 cuts from data/fbank/librispeech_cuts_train-clean-100.jsonl.gz
+2023-02-28 10:38:36,477 INFO [librispeech.py:48] (0/2) About to get train-clean-360 cuts from data/fbank/librispeech_cuts_train-clean-360.jsonl.gz
+2023-02-28 10:38:36,479 INFO [librispeech.py:53] (0/2) About to get train-other-500 cuts from data/fbank/librispeech_cuts_train-other-500.jsonl.gz
+2023-02-28 10:38:36,480 INFO [train.py:1189] (0/2) Using the XL subset of GigaSpeech (10k hours)
+2023-02-28 10:38:36,480 INFO [gigaspeech.py:46] (0/2) About to get train-XL cuts
+2023-02-28 10:38:36,666 INFO [gigaspeech.py:60] (0/2) Loading 1998 splits
+2023-02-28 10:38:45,671 INFO [asr_datamodule.py:165] (0/2) Enable MUSAN
+2023-02-28 10:38:45,671 INFO [asr_datamodule.py:175] (0/2) Enable SpecAugment
+2023-02-28 10:38:45,672 INFO [asr_datamodule.py:176] (0/2) Time warp factor: 80
+2023-02-28 10:38:45,672 INFO [asr_datamodule.py:189] (0/2) About to create train dataset
+2023-02-28 10:38:45,672 INFO [asr_datamodule.py:220] (0/2) Using DynamicBucketingSampler.
+2023-02-28 10:38:49,809 INFO [asr_datamodule.py:229] (0/2) About to create train dataloader
+2023-02-28 10:38:49,810 INFO [asr_datamodule.py:165] (0/2) Enable MUSAN
+2023-02-28 10:38:49,811 INFO [asr_datamodule.py:175] (0/2) Enable SpecAugment
+2023-02-28 10:38:49,811 INFO [asr_datamodule.py:176] (0/2) Time warp factor: 80
+2023-02-28 10:38:49,811 INFO [asr_datamodule.py:189] (0/2) About to create train dataset
+2023-02-28 10:38:49,811 INFO [asr_datamodule.py:220] (0/2) Using DynamicBucketingSampler.
+2023-02-28 10:38:53,560 INFO [asr_datamodule.py:229] (0/2) About to create train dataloader
+2023-02-28 10:38:53,560 INFO [librispeech.py:68] (0/2) About to get dev-clean cuts from data/fbank/librispeech_cuts_dev-clean.jsonl.gz
+2023-02-28 10:38:53,562 INFO [librispeech.py:73] (0/2) About to get dev-other cuts from data/fbank/librispeech_cuts_dev-other.jsonl.gz
+2023-02-28 10:38:53,563 INFO [asr_datamodule.py:242] (0/2) About to create dev dataset
+2023-02-28 10:38:53,754 INFO [asr_datamodule.py:261] (0/2) About to create dev dataloader
+2023-02-28 10:39:29,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=6.39 vs. limit=2.0
+2023-02-28 10:39:31,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5677, 4.5521, 4.5633, 4.5621], device='cuda:0'), covar=tensor([0.0020, 0.0019, 0.0022, 0.0019], device='cuda:0'), in_proj_covar=tensor([0.0027, 0.0026, 0.0027, 0.0027], device='cuda:0'), out_proj_covar=tensor([1.8915e-05, 1.8449e-05, 1.7936e-05, 1.7283e-05], device='cuda:0')
+2023-02-28 10:39:58,431 INFO [train.py:968] (0/2) Epoch 1, batch 50, giga_loss[loss=1.536, simple_loss=1.369, pruned_loss=1.501, over 27904.00 frames. ], tot_loss[loss=2.485, simple_loss=2.245, pruned_loss=2.297, over 1267296.38 frames. ], libri_tot_loss[loss=2.606, simple_loss=2.356, pruned_loss=2.426, over 259425.43 frames. ], giga_tot_loss[loss=2.49, simple_loss=2.251, pruned_loss=2.298, over 1055457.99 frames. ], batch size: 412, lr: 2.75e-02, grad_scale: 1.0
+2023-02-28 10:40:33,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:40:43,135 INFO [train.py:968] (0/2) Epoch 1, batch 100, giga_loss[loss=1.131, simple_loss=0.969, pruned_loss=1.285, over 28868.00 frames. ], tot_loss[loss=1.837, simple_loss=1.636, pruned_loss=1.813, over 2249613.87 frames. ], libri_tot_loss[loss=2.153, simple_loss=1.933, pruned_loss=2.064, over 370699.53 frames. ], giga_tot_loss[loss=1.824, simple_loss=1.624, pruned_loss=1.805, over 2008358.21 frames. ], batch size: 99, lr: 3.00e-02, grad_scale: 1.0
+2023-02-28 10:40:43,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 2.013e+02 4.874e+02 1.026e+03 2.164e+03 3.307e+04, threshold=2.052e+03, percent-clipped=0.0
+2023-02-28 10:41:03,202 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=4.91 vs. limit=2.0
+2023-02-28 10:41:21,765 INFO [train.py:968] (0/2) Epoch 1, batch 150, giga_loss[loss=1.037, simple_loss=0.8815, pruned_loss=1.129, over 28623.00 frames. ], tot_loss[loss=1.523, simple_loss=1.342, pruned_loss=1.544, over 3016591.02 frames. ], libri_tot_loss[loss=1.803, simple_loss=1.605, pruned_loss=1.765, over 534569.36 frames. ], giga_tot_loss[loss=1.523, simple_loss=1.341, pruned_loss=1.546, over 2735646.86 frames. ], batch size: 78, lr: 3.25e-02, grad_scale: 1.0
+2023-02-28 10:41:50,446 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=4.56 vs. limit=2.0
+2023-02-28 10:41:59,838 INFO [train.py:968] (0/2) Epoch 1, batch 200, giga_loss[loss=0.9087, simple_loss=0.7647, pruned_loss=0.9593, over 28666.00 frames. ], tot_loss[loss=1.333, simple_loss=1.165, pruned_loss=1.353, over 3618292.42 frames. ], libri_tot_loss[loss=1.638, simple_loss=1.451, pruned_loss=1.612, over 666488.57 frames. ], giga_tot_loss[loss=1.333, simple_loss=1.165, pruned_loss=1.355, over 3337467.72 frames. ], batch size: 85, lr: 3.50e-02, grad_scale: 1.0
+2023-02-28 10:42:00,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.600e+02 5.047e+02 7.414e+02 1.221e+03 4.600e+03, threshold=1.483e+03, percent-clipped=4.0
+2023-02-28 10:42:03,795 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=4.54 vs. limit=2.0
+2023-02-28 10:42:22,165 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:42:24,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:42:38,385 INFO [train.py:968] (0/2) Epoch 1, batch 250, giga_loss[loss=0.8994, simple_loss=0.7665, pruned_loss=0.8508, over 28588.00 frames. ], tot_loss[loss=1.206, simple_loss=1.047, pruned_loss=1.218, over 4082811.19 frames. ], libri_tot_loss[loss=1.568, simple_loss=1.384, pruned_loss=1.544, over 743621.83 frames. ], giga_tot_loss[loss=1.203, simple_loss=1.044, pruned_loss=1.216, over 3834660.44 frames. ], batch size: 307, lr: 3.75e-02, grad_scale: 1.0
+2023-02-28 10:42:47,390 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:43:03,228 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=5.07 vs. limit=2.0
+2023-02-28 10:43:16,653 INFO [train.py:968] (0/2) Epoch 1, batch 300, giga_loss[loss=0.8353, simple_loss=0.6915, pruned_loss=0.8327, over 28531.00 frames. ], tot_loss[loss=1.122, simple_loss=0.967, pruned_loss=1.12, over 4432213.15 frames. ], libri_tot_loss[loss=1.462, simple_loss=1.283, pruned_loss=1.433, over 906705.44 frames. ], giga_tot_loss[loss=1.119, simple_loss=0.9652, pruned_loss=1.119, over 4195342.28 frames. ], batch size: 85, lr: 4.00e-02, grad_scale: 1.0
+2023-02-28 10:43:17,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.699e+02 3.926e+02 4.914e+02 8.129e+02 2.623e+03, threshold=9.828e+02, percent-clipped=7.0
+2023-02-28 10:43:49,605 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:43:59,245 INFO [train.py:968] (0/2) Epoch 1, batch 350, giga_loss[loss=0.8262, simple_loss=0.6882, pruned_loss=0.7697, over 28381.00 frames. ], tot_loss[loss=1.052, simple_loss=0.9018, pruned_loss=1.035, over 4714324.43 frames. ], libri_tot_loss[loss=1.435, simple_loss=1.256, pruned_loss=1.403, over 956444.67 frames. ], giga_tot_loss[loss=1.048, simple_loss=0.8986, pruned_loss=1.033, over 4516377.53 frames. ], batch size: 78, lr: 4.24e-02, grad_scale: 1.0
+2023-02-28 10:44:36,892 INFO [train.py:968] (0/2) Epoch 1, batch 400, giga_loss[loss=0.8592, simple_loss=0.7181, pruned_loss=0.7602, over 28901.00 frames. ], tot_loss[loss=1.004, simple_loss=0.8559, pruned_loss=0.9701, over 4931102.73 frames. ], libri_tot_loss[loss=1.405, simple_loss=1.227, pruned_loss=1.367, over 1020573.33 frames. ], giga_tot_loss[loss=0.9992, simple_loss=0.8519, pruned_loss=0.9671, over 4768999.64 frames. ], batch size: 227, lr: 4.49e-02, grad_scale: 2.0
+2023-02-28 10:44:37,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 2.669e+02 3.675e+02 5.055e+02 1.089e+03, threshold=7.349e+02, percent-clipped=1.0
+2023-02-28 10:45:12,290 INFO [train.py:968] (0/2) Epoch 1, batch 450, giga_loss[loss=0.8302, simple_loss=0.6959, pruned_loss=0.7027, over 28945.00 frames. ], tot_loss[loss=0.9694, simple_loss=0.8226, pruned_loss=0.9156, over 5105835.77 frames. ], libri_tot_loss[loss=1.356, simple_loss=1.179, pruned_loss=1.306, over 1141476.76 frames. ], giga_tot_loss[loss=0.9636, simple_loss=0.8182, pruned_loss=0.9131, over 4956810.65 frames. ], batch size: 164, lr: 4.74e-02, grad_scale: 2.0
+2023-02-28 10:45:33,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=476.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:45:53,885 INFO [train.py:968] (0/2) Epoch 1, batch 500, giga_loss[loss=0.7765, simple_loss=0.6603, pruned_loss=0.6116, over 23533.00 frames. ], tot_loss[loss=0.9399, simple_loss=0.795, pruned_loss=0.8658, over 5241824.53 frames. ], libri_tot_loss[loss=1.328, simple_loss=1.151, pruned_loss=1.267, over 1235503.55 frames. ], giga_tot_loss[loss=0.9327, simple_loss=0.7896, pruned_loss=0.8627, over 5111046.45 frames. ], batch size: 705, lr: 4.99e-02, grad_scale: 2.0
+2023-02-28 10:45:54,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 2.178e+02 3.339e+02 4.102e+02 6.024e+02 1.143e+03, threshold=8.203e+02, percent-clipped=12.0
+2023-02-28 10:46:22,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1303, 1.4819, 1.3136, 1.0938], device='cuda:0'), covar=tensor([0.6211, 0.4270, 0.7028, 0.4947], device='cuda:0'), in_proj_covar=tensor([0.0086, 0.0072, 0.0081, 0.0073], device='cuda:0'), out_proj_covar=tensor([5.3994e-05, 4.6333e-05, 4.7202e-05, 5.3995e-05], device='cuda:0')
+2023-02-28 10:46:30,326 INFO [train.py:968] (0/2) Epoch 1, batch 550, libri_loss[loss=0.9856, simple_loss=0.8253, pruned_loss=0.7879, over 29539.00 frames. ], tot_loss[loss=0.9201, simple_loss=0.776, pruned_loss=0.8265, over 5349453.31 frames. ], libri_tot_loss[loss=1.283, simple_loss=1.106, pruned_loss=1.203, over 1416606.29 frames. ], giga_tot_loss[loss=0.9091, simple_loss=0.7677, pruned_loss=0.8224, over 5223807.20 frames. ], batch size: 80, lr: 4.98e-02, grad_scale: 2.0
+2023-02-28 10:47:12,066 INFO [train.py:968] (0/2) Epoch 1, batch 600, giga_loss[loss=0.7461, simple_loss=0.6279, pruned_loss=0.5741, over 29010.00 frames. ], tot_loss[loss=0.8899, simple_loss=0.7499, pruned_loss=0.778, over 5427869.16 frames. ], libri_tot_loss[loss=1.265, simple_loss=1.089, pruned_loss=1.177, over 1483536.68 frames. ], giga_tot_loss[loss=0.8795, simple_loss=0.7421, pruned_loss=0.7741, over 5321580.71 frames. ], batch size: 145, lr: 4.98e-02, grad_scale: 2.0
+2023-02-28 10:47:12,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 3.594e+02 5.899e+02 8.462e+02 1.137e+03 3.775e+03, threshold=1.692e+03, percent-clipped=54.0
+2023-02-28 10:47:56,292 INFO [train.py:968] (0/2) Epoch 1, batch 650, giga_loss[loss=0.7048, simple_loss=0.5956, pruned_loss=0.5247, over 28444.00 frames. ], tot_loss[loss=0.8603, simple_loss=0.7254, pruned_loss=0.7308, over 5479142.97 frames. ], libri_tot_loss[loss=1.247, simple_loss=1.072, pruned_loss=1.152, over 1547941.12 frames. ], giga_tot_loss[loss=0.8507, simple_loss=0.7181, pruned_loss=0.7274, over 5390503.23 frames. ], batch size: 65, lr: 4.98e-02, grad_scale: 2.0
+2023-02-28 10:48:26,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:48:36,489 INFO [train.py:968] (0/2) Epoch 1, batch 700, giga_loss[loss=0.7171, simple_loss=0.6109, pruned_loss=0.5135, over 28504.00 frames. ], tot_loss[loss=0.8325, simple_loss=0.7028, pruned_loss=0.6871, over 5525009.20 frames. ], libri_tot_loss[loss=1.221, simple_loss=1.048, pruned_loss=1.114, over 1650597.30 frames. ], giga_tot_loss[loss=0.8226, simple_loss=0.6952, pruned_loss=0.684, over 5450139.12 frames. ], batch size: 336, lr: 4.98e-02, grad_scale: 2.0
+2023-02-28 10:48:37,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.242e+02 9.634e+02 1.298e+03 1.890e+03 3.872e+03, threshold=2.595e+03, percent-clipped=32.0
+2023-02-28 10:48:46,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:49:16,907 INFO [train.py:968] (0/2) Epoch 1, batch 750, giga_loss[loss=0.7075, simple_loss=0.6074, pruned_loss=0.4891, over 29006.00 frames. ], tot_loss[loss=0.8052, simple_loss=0.6813, pruned_loss=0.6453, over 5571264.98 frames. ], libri_tot_loss[loss=1.191, simple_loss=1.021, pruned_loss=1.07, over 1773426.77 frames. ], giga_tot_loss[loss=0.7947, simple_loss=0.673, pruned_loss=0.6426, over 5504649.17 frames. ], batch size: 164, lr: 4.97e-02, grad_scale: 2.0
+2023-02-28 10:49:41,969 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:49:58,860 INFO [train.py:968] (0/2) Epoch 1, batch 800, giga_loss[loss=0.6541, simple_loss=0.5654, pruned_loss=0.4387, over 28896.00 frames. ], tot_loss[loss=0.7742, simple_loss=0.6568, pruned_loss=0.6033, over 5595088.30 frames. ], libri_tot_loss[loss=1.186, simple_loss=1.016, pruned_loss=1.063, over 1794170.64 frames. ], giga_tot_loss[loss=0.7653, simple_loss=0.6498, pruned_loss=0.6007, over 5541500.81 frames. ], batch size: 213, lr: 4.97e-02, grad_scale: 4.0
+2023-02-28 10:49:59,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.179e+02 1.281e+03 1.604e+03 1.966e+03 3.315e+03, threshold=3.208e+03, percent-clipped=9.0
+2023-02-28 10:50:31,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0527, 2.2487, 2.7848, 1.8236], device='cuda:0'), covar=tensor([0.3615, 0.4142, 0.4561, 0.4255], device='cuda:0'), in_proj_covar=tensor([0.0151, 0.0161, 0.0217, 0.0155], device='cuda:0'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001], device='cuda:0')
+2023-02-28 10:50:44,170 INFO [train.py:968] (0/2) Epoch 1, batch 850, giga_loss[loss=0.9182, simple_loss=0.7536, pruned_loss=0.6729, over 28268.00 frames. ], tot_loss[loss=0.7757, simple_loss=0.6582, pruned_loss=0.5892, over 5607915.22 frames. ], libri_tot_loss[loss=1.173, simple_loss=1.005, pruned_loss=1.047, over 1835297.50 frames. ], giga_tot_loss[loss=0.7684, simple_loss=0.6525, pruned_loss=0.5873, over 5563002.38 frames. ], batch size: 368, lr: 4.96e-02, grad_scale: 4.0
+2023-02-28 10:50:44,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:50:47,929 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=855.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:50:49,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:51:12,489 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=887.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:51:21,723 INFO [train.py:968] (0/2) Epoch 1, batch 900, giga_loss[loss=0.8074, simple_loss=0.6863, pruned_loss=0.5457, over 28906.00 frames. ], tot_loss[loss=0.7913, simple_loss=0.6709, pruned_loss=0.5872, over 5632523.68 frames. ], libri_tot_loss[loss=1.146, simple_loss=0.9806, pruned_loss=1.008, over 1955605.13 frames. ], giga_tot_loss[loss=0.7832, simple_loss=0.6646, pruned_loss=0.5852, over 5589432.07 frames. ], batch size: 164, lr: 4.96e-02, grad_scale: 4.0
+2023-02-28 10:51:22,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.815e+02 1.416e+03 1.889e+03 2.839e+03 7.286e+03, threshold=3.778e+03, percent-clipped=13.0
+2023-02-28 10:52:03,261 INFO [train.py:968] (0/2) Epoch 1, batch 950, giga_loss[loss=0.7711, simple_loss=0.6467, pruned_loss=0.527, over 23658.00 frames. ], tot_loss[loss=0.7954, simple_loss=0.6756, pruned_loss=0.576, over 5649033.07 frames. ], libri_tot_loss[loss=1.135, simple_loss=0.9709, pruned_loss=0.9934, over 1994954.56 frames. ], giga_tot_loss[loss=0.7894, simple_loss=0.671, pruned_loss=0.5751, over 5612934.39 frames. ], batch size: 705, lr: 4.96e-02, grad_scale: 4.0
+2023-02-28 10:52:08,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=955.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:52:12,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=961.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 10:52:26,908 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=983.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:52:37,454 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=994.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:52:39,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=997.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:52:40,567 INFO [train.py:968] (0/2) Epoch 1, batch 1000, giga_loss[loss=0.7662, simple_loss=0.6649, pruned_loss=0.4863, over 29045.00 frames. ], tot_loss[loss=0.7941, simple_loss=0.6764, pruned_loss=0.562, over 5655262.20 frames. ], libri_tot_loss[loss=1.112, simple_loss=0.9508, pruned_loss=0.9588, over 2099300.76 frames. ], giga_tot_loss[loss=0.7883, simple_loss=0.6717, pruned_loss=0.5617, over 5630678.83 frames. ], batch size: 155, lr: 4.95e-02, grad_scale: 4.0
+2023-02-28 10:52:41,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.165e+02 1.290e+03 1.606e+03 2.215e+03 6.761e+03, threshold=3.212e+03, percent-clipped=5.0
+2023-02-28 10:52:57,864 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:53:14,858 INFO [train.py:968] (0/2) Epoch 1, batch 1050, libri_loss[loss=0.7609, simple_loss=0.6526, pruned_loss=0.4884, over 29656.00 frames. ], tot_loss[loss=0.7917, simple_loss=0.6751, pruned_loss=0.5495, over 5667894.08 frames. ], libri_tot_loss[loss=1.083, simple_loss=0.925, pruned_loss=0.9169, over 2245321.84 frames. ], giga_tot_loss[loss=0.7857, simple_loss=0.6705, pruned_loss=0.5496, over 5642626.94 frames. ], batch size: 73, lr: 4.95e-02, grad_scale: 4.0
+2023-02-28 10:53:27,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1064.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:53:52,901 INFO [train.py:968] (0/2) Epoch 1, batch 1100, giga_loss[loss=0.7127, simple_loss=0.6244, pruned_loss=0.437, over 29039.00 frames. ], tot_loss[loss=0.7795, simple_loss=0.6678, pruned_loss=0.5292, over 5668550.52 frames. ], libri_tot_loss[loss=1.05, simple_loss=0.8983, pruned_loss=0.8721, over 2397822.65 frames. ], giga_tot_loss[loss=0.7751, simple_loss=0.6641, pruned_loss=0.5313, over 5646370.10 frames. ], batch size: 136, lr: 4.94e-02, grad_scale: 4.0
+2023-02-28 10:53:54,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.463e+02 1.460e+03 1.974e+03 2.511e+03 1.683e+04, threshold=3.948e+03, percent-clipped=9.0
+2023-02-28 10:54:32,743 INFO [train.py:968] (0/2) Epoch 1, batch 1150, libri_loss[loss=0.7918, simple_loss=0.6826, pruned_loss=0.4946, over 29167.00 frames. ], tot_loss[loss=0.7676, simple_loss=0.6597, pruned_loss=0.5113, over 5676197.64 frames. ], libri_tot_loss[loss=1.038, simple_loss=0.8871, pruned_loss=0.8532, over 2471417.08 frames. ], giga_tot_loss[loss=0.763, simple_loss=0.656, pruned_loss=0.5127, over 5665826.76 frames. ], batch size: 101, lr: 4.94e-02, grad_scale: 4.0
+2023-02-28 10:54:36,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:54:47,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4401, 2.4354, 3.9682, 1.5363], device='cuda:0'), covar=tensor([0.2065, 0.2381, 0.1410, 0.3286], device='cuda:0'), in_proj_covar=tensor([0.0208, 0.0218, 0.0298, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 10:54:59,188 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-02-28 10:55:11,828 INFO [train.py:968] (0/2) Epoch 1, batch 1200, libri_loss[loss=0.731, simple_loss=0.6173, pruned_loss=0.467, over 29659.00 frames. ], tot_loss[loss=0.7581, simple_loss=0.653, pruned_loss=0.4967, over 5659764.41 frames. ], libri_tot_loss[loss=1.017, simple_loss=0.8699, pruned_loss=0.8235, over 2592843.85 frames. ], giga_tot_loss[loss=0.753, simple_loss=0.649, pruned_loss=0.4979, over 5654810.10 frames. ], batch size: 73, lr: 4.93e-02, grad_scale: 8.0
+2023-02-28 10:55:12,299 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.267e+02 1.592e+03 2.086e+03 2.875e+03 6.911e+03, threshold=4.171e+03, percent-clipped=6.0
+2023-02-28 10:55:15,896 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1207.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:55:19,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1210.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:55:43,683 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:55:44,080 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1240.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 10:55:50,445 INFO [train.py:968] (0/2) Epoch 1, batch 1250, giga_loss[loss=0.7393, simple_loss=0.6312, pruned_loss=0.4609, over 27514.00 frames. ], tot_loss[loss=0.7441, simple_loss=0.6441, pruned_loss=0.4784, over 5664272.05 frames. ], libri_tot_loss[loss=1.006, simple_loss=0.8607, pruned_loss=0.8078, over 2654753.98 frames. ], giga_tot_loss[loss=0.7399, simple_loss=0.6408, pruned_loss=0.4797, over 5659461.27 frames. ], batch size: 472, lr: 4.92e-02, grad_scale: 8.0
+2023-02-28 10:56:04,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.7010, 1.0287, 0.7013, 0.3013], device='cuda:0'), covar=tensor([0.7859, 0.9728, 1.0319, 1.8006], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0227, 0.0222, 0.0297], device='cuda:0'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-02-28 10:56:09,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3645, 2.2601, 3.0458, 1.9215], device='cuda:0'), covar=tensor([0.1384, 0.1815, 0.1267, 0.2010], device='cuda:0'), in_proj_covar=tensor([0.0230, 0.0234, 0.0313, 0.0265], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 10:56:18,753 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=3.31 vs. limit=2.0
+2023-02-28 10:56:31,633 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1298.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:56:32,433 INFO [train.py:968] (0/2) Epoch 1, batch 1300, giga_loss[loss=0.7425, simple_loss=0.6534, pruned_loss=0.4401, over 28793.00 frames. ], tot_loss[loss=0.7402, simple_loss=0.6422, pruned_loss=0.469, over 5673116.51 frames. ], libri_tot_loss[loss=0.9999, simple_loss=0.8555, pruned_loss=0.798, over 2703063.94 frames. ], giga_tot_loss[loss=0.7359, simple_loss=0.6388, pruned_loss=0.4696, over 5666265.71 frames. ], batch size: 99, lr: 4.92e-02, grad_scale: 8.0
+2023-02-28 10:56:33,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.898e+02 1.482e+03 1.871e+03 2.500e+03 7.793e+03, threshold=3.742e+03, percent-clipped=5.0
+2023-02-28 10:56:33,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1301.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:56:54,911 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:56:54,917 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:56:58,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:57:09,571 INFO [train.py:968] (0/2) Epoch 1, batch 1350, giga_loss[loss=0.79, simple_loss=0.6643, pruned_loss=0.4946, over 26579.00 frames. ], tot_loss[loss=0.7288, simple_loss=0.636, pruned_loss=0.4534, over 5689913.90 frames. ], libri_tot_loss[loss=0.9933, simple_loss=0.85, pruned_loss=0.7898, over 2735169.69 frames. ], giga_tot_loss[loss=0.7257, simple_loss=0.6336, pruned_loss=0.4543, over 5682444.15 frames. ], batch size: 555, lr: 4.91e-02, grad_scale: 8.0
+2023-02-28 10:57:16,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:57:47,816 INFO [train.py:968] (0/2) Epoch 1, batch 1400, giga_loss[loss=0.7364, simple_loss=0.6323, pruned_loss=0.4457, over 28607.00 frames. ], tot_loss[loss=0.7213, simple_loss=0.6315, pruned_loss=0.4427, over 5687508.61 frames. ], libri_tot_loss[loss=0.9797, simple_loss=0.8391, pruned_loss=0.7715, over 2809508.31 frames. ], giga_tot_loss[loss=0.7189, simple_loss=0.6296, pruned_loss=0.4438, over 5680270.51 frames. ], batch size: 92, lr: 4.91e-02, grad_scale: 8.0
+2023-02-28 10:57:48,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.598e+02 1.493e+03 1.874e+03 2.832e+03 1.835e+04, threshold=3.748e+03, percent-clipped=5.0
+2023-02-28 10:57:50,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-02-28 10:58:24,307 INFO [train.py:968] (0/2) Epoch 1, batch 1450, giga_loss[loss=0.6607, simple_loss=0.6005, pruned_loss=0.3693, over 28991.00 frames. ], tot_loss[loss=0.7163, simple_loss=0.6286, pruned_loss=0.4344, over 5696359.08 frames. ], libri_tot_loss[loss=0.9664, simple_loss=0.8276, pruned_loss=0.7512, over 2915240.61 frames. ], giga_tot_loss[loss=0.7122, simple_loss=0.6257, pruned_loss=0.4341, over 5684689.00 frames. ], batch size: 128, lr: 4.90e-02, grad_scale: 4.0
+2023-02-28 10:58:26,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7096, 2.1502, 2.9953, 1.7436], device='cuda:0'), covar=tensor([0.2513, 0.2887, 0.1307, 0.3884], device='cuda:0'), in_proj_covar=tensor([0.0225, 0.0225, 0.0250, 0.0310], device='cuda:0'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 10:58:38,050 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-02-28 10:58:39,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1473.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:58:41,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1476.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:58:44,226 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1479.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:58:45,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1482.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:58:58,032 INFO [train.py:968] (0/2) Epoch 1, batch 1500, giga_loss[loss=0.7176, simple_loss=0.622, pruned_loss=0.4243, over 27954.00 frames. ], tot_loss[loss=0.7011, simple_loss=0.6197, pruned_loss=0.4183, over 5705599.16 frames. ], libri_tot_loss[loss=0.9517, simple_loss=0.816, pruned_loss=0.7317, over 3003511.29 frames. ], giga_tot_loss[loss=0.6976, simple_loss=0.6172, pruned_loss=0.4182, over 5691075.60 frames. ], batch size: 412, lr: 4.89e-02, grad_scale: 4.0
+2023-02-28 10:58:58,988 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1501.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:58:59,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.440e+02 1.623e+03 2.062e+03 2.705e+03 7.771e+03, threshold=4.125e+03, percent-clipped=10.0
+2023-02-28 10:59:02,149 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1504.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:59:02,657 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1505.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:59:06,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1511.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:59:22,245 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1533.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:59:32,256 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1546.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 10:59:34,980 INFO [train.py:968] (0/2) Epoch 1, batch 1550, giga_loss[loss=0.6541, simple_loss=0.5924, pruned_loss=0.3653, over 28905.00 frames. ], tot_loss[loss=0.6883, simple_loss=0.6113, pruned_loss=0.4054, over 5706386.88 frames. ], libri_tot_loss[loss=0.9423, simple_loss=0.8085, pruned_loss=0.7193, over 3059491.67 frames. ], giga_tot_loss[loss=0.6853, simple_loss=0.6093, pruned_loss=0.4055, over 5692606.27 frames. ], batch size: 174, lr: 4.89e-02, grad_scale: 4.0
+2023-02-28 10:59:54,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-02-28 11:00:15,342 INFO [train.py:968] (0/2) Epoch 1, batch 1600, giga_loss[loss=0.6601, simple_loss=0.5834, pruned_loss=0.3781, over 29057.00 frames. ], tot_loss[loss=0.6833, simple_loss=0.6069, pruned_loss=0.3997, over 5717106.12 frames. ], libri_tot_loss[loss=0.9337, simple_loss=0.8018, pruned_loss=0.7078, over 3115800.14 frames. ], giga_tot_loss[loss=0.6804, simple_loss=0.6049, pruned_loss=0.3995, over 5702920.86 frames. ], batch size: 155, lr: 4.88e-02, grad_scale: 8.0
+2023-02-28 11:00:16,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.188e+03 1.868e+03 2.434e+03 2.910e+03 9.281e+03, threshold=4.868e+03, percent-clipped=8.0
+2023-02-28 11:00:23,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1807, 0.8447, 0.7766, 1.0100], device='cuda:0'), covar=tensor([1.4381, 1.4249, 1.9392, 2.4030], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0400, 0.0550, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 11:00:24,918 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4484, 1.5309, 1.2666, 1.2791], device='cuda:0'), covar=tensor([0.5890, 0.4561, 0.4458, 0.4070], device='cuda:0'), in_proj_covar=tensor([0.0153, 0.0167, 0.0148, 0.0138], device='cuda:0'), out_proj_covar=tensor([1.0100e-04, 1.0741e-04, 9.0694e-05, 8.3753e-05], device='cuda:0')
+2023-02-28 11:00:25,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1615.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:00:53,079 INFO [train.py:968] (0/2) Epoch 1, batch 1650, giga_loss[loss=0.6754, simple_loss=0.599, pruned_loss=0.3839, over 29000.00 frames. ], tot_loss[loss=0.6862, simple_loss=0.6061, pruned_loss=0.4013, over 5714179.74 frames. ], libri_tot_loss[loss=0.9243, simple_loss=0.7941, pruned_loss=0.6948, over 3184631.50 frames. ], giga_tot_loss[loss=0.683, simple_loss=0.604, pruned_loss=0.4007, over 5698884.26 frames. ], batch size: 145, lr: 4.87e-02, grad_scale: 4.0
+2023-02-28 11:01:31,801 INFO [train.py:968] (0/2) Epoch 1, batch 1700, giga_loss[loss=0.5979, simple_loss=0.5466, pruned_loss=0.3279, over 28606.00 frames. ], tot_loss[loss=0.6842, simple_loss=0.6038, pruned_loss=0.3983, over 5696622.29 frames. ], libri_tot_loss[loss=0.9183, simple_loss=0.7892, pruned_loss=0.686, over 3228755.27 frames. ], giga_tot_loss[loss=0.6806, simple_loss=0.6013, pruned_loss=0.3972, over 5689327.48 frames. ], batch size: 85, lr: 4.87e-02, grad_scale: 4.0
+2023-02-28 11:01:33,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.439e+02 1.838e+03 2.367e+03 3.126e+03 5.811e+03, threshold=4.734e+03, percent-clipped=5.0
+2023-02-28 11:02:14,746 INFO [train.py:968] (0/2) Epoch 1, batch 1750, giga_loss[loss=0.5983, simple_loss=0.5447, pruned_loss=0.3289, over 28504.00 frames. ], tot_loss[loss=0.6709, simple_loss=0.5948, pruned_loss=0.3868, over 5703383.20 frames. ], libri_tot_loss[loss=0.9094, simple_loss=0.7823, pruned_loss=0.675, over 3281799.64 frames. ], giga_tot_loss[loss=0.6681, simple_loss=0.5929, pruned_loss=0.3862, over 5693767.16 frames. ], batch size: 85, lr: 4.86e-02, grad_scale: 4.0
+2023-02-28 11:02:20,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1758.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:02:23,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1761.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:02:24,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0706, 1.0219, 1.1436, 1.1737], device='cuda:0'), covar=tensor([0.4954, 0.4615, 0.4002, 0.6666], device='cuda:0'), in_proj_covar=tensor([0.0226, 0.0218, 0.0237, 0.0261], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 11:02:30,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0191, 2.1220, 3.1537, 1.3967], device='cuda:0'), covar=tensor([0.0833, 0.0858, 0.0565, 0.1410], device='cuda:0'), in_proj_covar=tensor([0.0258, 0.0234, 0.0367, 0.0295], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002], device='cuda:0')
+2023-02-28 11:02:45,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1790.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:02:47,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0658, 1.0522, 1.2575, 1.0345], device='cuda:0'), covar=tensor([0.8958, 0.7461, 0.6364, 1.2704], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0463, 0.0399, 0.0510], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 11:02:50,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0774, 0.8308, 0.7380, 1.1300], device='cuda:0'), covar=tensor([1.0123, 0.8703, 1.6827, 1.5303], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0389, 0.0526, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 11:02:53,569 INFO [train.py:968] (0/2) Epoch 1, batch 1800, giga_loss[loss=0.633, simple_loss=0.5837, pruned_loss=0.3428, over 28704.00 frames. ], tot_loss[loss=0.6584, simple_loss=0.5871, pruned_loss=0.3759, over 5706777.16 frames. ], libri_tot_loss[loss=0.9094, simple_loss=0.7823, pruned_loss=0.675, over 3281799.64 frames. ], giga_tot_loss[loss=0.6563, simple_loss=0.5856, pruned_loss=0.3754, over 5699292.87 frames. ], batch size: 242, lr: 4.85e-02, grad_scale: 4.0
+2023-02-28 11:02:55,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.697e+03 2.098e+03 2.759e+03 6.576e+03, threshold=4.196e+03, percent-clipped=4.0
+2023-02-28 11:03:12,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1827.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:03:31,127 INFO [train.py:968] (0/2) Epoch 1, batch 1850, giga_loss[loss=0.5433, simple_loss=0.523, pruned_loss=0.2811, over 28389.00 frames. ], tot_loss[loss=0.6507, simple_loss=0.5834, pruned_loss=0.368, over 5714186.28 frames. ], libri_tot_loss[loss=0.901, simple_loss=0.776, pruned_loss=0.6644, over 3333725.04 frames. ], giga_tot_loss[loss=0.6487, simple_loss=0.5819, pruned_loss=0.3676, over 5705021.80 frames. ], batch size: 65, lr: 4.84e-02, grad_scale: 4.0
+2023-02-28 11:03:52,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2892, 2.3113, 3.4458, 2.0696], device='cuda:0'), covar=tensor([0.1283, 0.1596, 0.0449, 0.1880], device='cuda:0'), in_proj_covar=tensor([0.0293, 0.0278, 0.0327, 0.0360], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003], device='cuda:0')
+2023-02-28 11:04:11,943 INFO [train.py:968] (0/2) Epoch 1, batch 1900, giga_loss[loss=0.6111, simple_loss=0.5661, pruned_loss=0.3287, over 28897.00 frames. ], tot_loss[loss=0.6397, simple_loss=0.5768, pruned_loss=0.3584, over 5704526.06 frames. ], libri_tot_loss[loss=0.8952, simple_loss=0.7716, pruned_loss=0.6569, over 3370066.81 frames. ], giga_tot_loss[loss=0.6377, simple_loss=0.5753, pruned_loss=0.358, over 5696601.95 frames. ], batch size: 199, lr: 4.83e-02, grad_scale: 4.0
+2023-02-28 11:04:15,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.144e+03 1.899e+03 2.403e+03 3.422e+03 6.556e+03, threshold=4.805e+03, percent-clipped=10.0
+2023-02-28 11:04:30,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1921.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:04:51,054 INFO [train.py:968] (0/2) Epoch 1, batch 1950, giga_loss[loss=0.5805, simple_loss=0.5387, pruned_loss=0.3115, over 28217.00 frames. ], tot_loss[loss=0.6261, simple_loss=0.5684, pruned_loss=0.3476, over 5694012.62 frames. ], libri_tot_loss[loss=0.8781, simple_loss=0.7587, pruned_loss=0.6356, over 3471706.93 frames. ], giga_tot_loss[loss=0.6236, simple_loss=0.5664, pruned_loss=0.3469, over 5688535.54 frames. ], batch size: 368, lr: 4.83e-02, grad_scale: 4.0
+2023-02-28 11:05:00,815 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1959.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:05:34,887 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-2000.pt
+2023-02-28 11:05:35,168 INFO [train.py:968] (0/2) Epoch 1, batch 2000, giga_loss[loss=0.4533, simple_loss=0.4532, pruned_loss=0.2267, over 28818.00 frames. ], tot_loss[loss=0.6053, simple_loss=0.5543, pruned_loss=0.3326, over 5688450.35 frames. ], libri_tot_loss[loss=0.8673, simple_loss=0.7506, pruned_loss=0.6222, over 3543030.13 frames. ], giga_tot_loss[loss=0.602, simple_loss=0.5517, pruned_loss=0.3314, over 5678951.41 frames. ], batch size: 285, lr: 4.82e-02, grad_scale: 8.0
+2023-02-28 11:05:37,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.314e+02 1.813e+03 2.453e+03 3.086e+03 7.215e+03, threshold=4.906e+03, percent-clipped=9.0
+2023-02-28 11:05:38,232 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=2004.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:06:00,116 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.42 vs. limit=5.0
+2023-02-28 11:06:19,135 INFO [train.py:968] (0/2) Epoch 1, batch 2050, giga_loss[loss=0.5337, simple_loss=0.5149, pruned_loss=0.2763, over 28606.00 frames. ], tot_loss[loss=0.5792, simple_loss=0.5376, pruned_loss=0.3138, over 5671131.27 frames. ], libri_tot_loss[loss=0.865, simple_loss=0.7489, pruned_loss=0.619, over 3556645.76 frames. ], giga_tot_loss[loss=0.5758, simple_loss=0.535, pruned_loss=0.3124, over 5671025.13 frames. ], batch size: 85, lr: 4.81e-02, grad_scale: 8.0
+2023-02-28 11:06:33,235 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2064.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:06:36,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2067.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:07:00,034 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2096.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:07:02,276 INFO [train.py:968] (0/2) Epoch 1, batch 2100, giga_loss[loss=0.527, simple_loss=0.509, pruned_loss=0.2725, over 28017.00 frames. ], tot_loss[loss=0.5691, simple_loss=0.5321, pruned_loss=0.3058, over 5674586.50 frames. ], libri_tot_loss[loss=0.8626, simple_loss=0.7469, pruned_loss=0.6158, over 3569715.43 frames. ], giga_tot_loss[loss=0.5657, simple_loss=0.5296, pruned_loss=0.3041, over 5680821.72 frames. ], batch size: 77, lr: 4.80e-02, grad_scale: 8.0
+2023-02-28 11:07:04,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.123e+02 1.430e+03 1.949e+03 2.626e+03 7.392e+03, threshold=3.898e+03, percent-clipped=3.0
+2023-02-28 11:07:40,170 INFO [train.py:968] (0/2) Epoch 1, batch 2150, libri_loss[loss=0.7086, simple_loss=0.6493, pruned_loss=0.384, over 28665.00 frames. ], tot_loss[loss=0.5637, simple_loss=0.5302, pruned_loss=0.3007, over 5686431.84 frames. ], libri_tot_loss[loss=0.8508, simple_loss=0.7385, pruned_loss=0.6002, over 3660190.86 frames. ], giga_tot_loss[loss=0.5569, simple_loss=0.5251, pruned_loss=0.297, over 5684786.89 frames. ], batch size: 106, lr: 4.79e-02, grad_scale: 8.0
+2023-02-28 11:07:50,612 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1920, 1.2498, 1.1367, 0.5059], device='cuda:0'), covar=tensor([0.3445, 0.3089, 0.2332, 0.8367], device='cuda:0'), in_proj_covar=tensor([0.0302, 0.0273, 0.0234, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003], device='cuda:0')
+2023-02-28 11:08:18,361 INFO [train.py:968] (0/2) Epoch 1, batch 2200, giga_loss[loss=0.4794, simple_loss=0.4772, pruned_loss=0.2408, over 29017.00 frames. ], tot_loss[loss=0.556, simple_loss=0.5258, pruned_loss=0.2947, over 5687482.81 frames. ], libri_tot_loss[loss=0.8356, simple_loss=0.7274, pruned_loss=0.5827, over 3747910.77 frames. ], giga_tot_loss[loss=0.548, simple_loss=0.52, pruned_loss=0.2901, over 5687464.43 frames. ], batch size: 164, lr: 4.78e-02, grad_scale: 4.0
+2023-02-28 11:08:19,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2202.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:08:20,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.804e+02 1.667e+03 2.192e+03 3.083e+03 6.873e+03, threshold=4.383e+03, percent-clipped=13.0
+2023-02-28 11:08:41,032 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=2231.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:08:54,928 INFO [train.py:968] (0/2) Epoch 1, batch 2250, giga_loss[loss=0.4613, simple_loss=0.4705, pruned_loss=0.226, over 29003.00 frames. ], tot_loss[loss=0.5476, simple_loss=0.5207, pruned_loss=0.2885, over 5690033.34 frames. ], libri_tot_loss[loss=0.8258, simple_loss=0.7203, pruned_loss=0.5701, over 3818790.58 frames. ], giga_tot_loss[loss=0.5373, simple_loss=0.5132, pruned_loss=0.2824, over 5694786.43 frames. ], batch size: 213, lr: 4.77e-02, grad_scale: 4.0
+2023-02-28 11:08:55,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=2250.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:09:03,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=2263.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:09:34,743 INFO [train.py:968] (0/2) Epoch 1, batch 2300, giga_loss[loss=0.439, simple_loss=0.4465, pruned_loss=0.2158, over 28895.00 frames. ], tot_loss[loss=0.533, simple_loss=0.511, pruned_loss=0.2785, over 5693948.51 frames. ], libri_tot_loss[loss=0.8241, simple_loss=0.719, pruned_loss=0.5683, over 3829072.63 frames. ], giga_tot_loss[loss=0.5246, simple_loss=0.505, pruned_loss=0.2735, over 5696938.74 frames. ], batch size: 112, lr: 4.77e-02, grad_scale: 4.0
+2023-02-28 11:09:37,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.979e+02 1.533e+03 2.066e+03 2.734e+03 7.125e+03, threshold=4.133e+03, percent-clipped=8.0
+2023-02-28 11:09:52,848 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-02-28 11:09:59,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2334.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:10:09,183 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2345.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:10:10,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2348.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:10:12,098 INFO [train.py:968] (0/2) Epoch 1, batch 2350, giga_loss[loss=0.5329, simple_loss=0.5096, pruned_loss=0.2781, over 27973.00 frames. ], tot_loss[loss=0.5224, simple_loss=0.5042, pruned_loss=0.271, over 5706985.04 frames. ], libri_tot_loss[loss=0.8164, simple_loss=0.7136, pruned_loss=0.5593, over 3880760.01 frames. ], giga_tot_loss[loss=0.5136, simple_loss=0.4978, pruned_loss=0.2658, over 5704598.69 frames. ], batch size: 412, lr: 4.76e-02, grad_scale: 4.0
+2023-02-28 11:10:33,176 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2377.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:10:35,215 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2379.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:10:41,839 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-02-28 11:10:50,452 INFO [train.py:968] (0/2) Epoch 1, batch 2400, giga_loss[loss=0.4256, simple_loss=0.4465, pruned_loss=0.2024, over 28682.00 frames. ], tot_loss[loss=0.5116, simple_loss=0.4969, pruned_loss=0.2637, over 5717481.20 frames. ], libri_tot_loss[loss=0.8146, simple_loss=0.7125, pruned_loss=0.5567, over 3900953.81 frames. ], giga_tot_loss[loss=0.5029, simple_loss=0.4906, pruned_loss=0.2585, over 5713859.34 frames. ], batch size: 242, lr: 4.75e-02, grad_scale: 8.0
+2023-02-28 11:10:53,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.171e+02 1.546e+03 2.148e+03 3.184e+03 6.518e+03, threshold=4.296e+03, percent-clipped=8.0
+2023-02-28 11:11:27,202 INFO [train.py:968] (0/2) Epoch 1, batch 2450, giga_loss[loss=0.4942, simple_loss=0.4851, pruned_loss=0.2516, over 28684.00 frames. ], tot_loss[loss=0.5049, simple_loss=0.4929, pruned_loss=0.2589, over 5717405.66 frames. ], libri_tot_loss[loss=0.8113, simple_loss=0.7103, pruned_loss=0.551, over 3940415.76 frames. ], giga_tot_loss[loss=0.4935, simple_loss=0.4845, pruned_loss=0.2519, over 5719500.86 frames. ], batch size: 284, lr: 4.74e-02, grad_scale: 8.0
+2023-02-28 11:11:48,483 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2477.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:11:50,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2480.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:12:05,856 INFO [train.py:968] (0/2) Epoch 1, batch 2500, giga_loss[loss=0.4201, simple_loss=0.4303, pruned_loss=0.205, over 28561.00 frames. ], tot_loss[loss=0.496, simple_loss=0.487, pruned_loss=0.2528, over 5717290.45 frames. ], libri_tot_loss[loss=0.8074, simple_loss=0.7075, pruned_loss=0.5464, over 3970190.51 frames. ], giga_tot_loss[loss=0.485, simple_loss=0.479, pruned_loss=0.246, over 5716062.11 frames. ], batch size: 85, lr: 4.73e-02, grad_scale: 8.0
+2023-02-28 11:12:08,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.188e+02 1.511e+03 2.015e+03 2.800e+03 9.447e+03, threshold=4.029e+03, percent-clipped=9.0
+2023-02-28 11:12:12,219 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2509.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:12:21,624 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2522.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:12:23,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2525.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:12:42,081 INFO [train.py:968] (0/2) Epoch 1, batch 2550, giga_loss[loss=0.4936, simple_loss=0.4804, pruned_loss=0.2534, over 28924.00 frames. ], tot_loss[loss=0.4949, simple_loss=0.4864, pruned_loss=0.252, over 5716527.30 frames. ], libri_tot_loss[loss=0.8021, simple_loss=0.704, pruned_loss=0.5389, over 4024092.78 frames. ], giga_tot_loss[loss=0.4808, simple_loss=0.4762, pruned_loss=0.2432, over 5713391.04 frames. ], batch size: 106, lr: 4.72e-02, grad_scale: 8.0
+2023-02-28 11:12:45,211 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2554.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:13:20,223 INFO [train.py:968] (0/2) Epoch 1, batch 2600, giga_loss[loss=0.4672, simple_loss=0.4739, pruned_loss=0.2302, over 28946.00 frames. ], tot_loss[loss=0.4915, simple_loss=0.4845, pruned_loss=0.2495, over 5725252.08 frames. ], libri_tot_loss[loss=0.7984, simple_loss=0.7015, pruned_loss=0.5332, over 4070966.78 frames. ], giga_tot_loss[loss=0.4755, simple_loss=0.4729, pruned_loss=0.2394, over 5718295.96 frames. ], batch size: 227, lr: 4.71e-02, grad_scale: 8.0
+2023-02-28 11:13:22,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.416e+02 1.742e+03 2.523e+03 3.419e+03 7.588e+03, threshold=5.047e+03, percent-clipped=16.0
+2023-02-28 11:13:24,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2606.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:13:39,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2625.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:13:48,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2638.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:13:57,771 INFO [train.py:968] (0/2) Epoch 1, batch 2650, giga_loss[loss=0.4571, simple_loss=0.4646, pruned_loss=0.2248, over 28551.00 frames. ], tot_loss[loss=0.4876, simple_loss=0.4826, pruned_loss=0.2464, over 5726389.67 frames. ], libri_tot_loss[loss=0.7897, simple_loss=0.6957, pruned_loss=0.5232, over 4133694.38 frames. ], giga_tot_loss[loss=0.4701, simple_loss=0.4698, pruned_loss=0.2355, over 5715482.41 frames. ], batch size: 85, lr: 4.70e-02, grad_scale: 8.0
+2023-02-28 11:14:28,719 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-02-28 11:14:35,945 INFO [train.py:968] (0/2) Epoch 1, batch 2700, giga_loss[loss=0.4853, simple_loss=0.4875, pruned_loss=0.2416, over 28878.00 frames. ], tot_loss[loss=0.4917, simple_loss=0.4858, pruned_loss=0.2489, over 5719589.03 frames. ], libri_tot_loss[loss=0.7821, simple_loss=0.6905, pruned_loss=0.515, over 4181371.62 frames. ], giga_tot_loss[loss=0.4744, simple_loss=0.4732, pruned_loss=0.2381, over 5710422.64 frames. ], batch size: 213, lr: 4.69e-02, grad_scale: 8.0
+2023-02-28 11:14:39,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.677e+02 1.504e+03 2.079e+03 2.737e+03 4.915e+03, threshold=4.157e+03, percent-clipped=0.0
+2023-02-28 11:15:02,616 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=2733.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:15:15,366 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2749.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:15:15,677 INFO [train.py:968] (0/2) Epoch 1, batch 2750, giga_loss[loss=0.6025, simple_loss=0.5554, pruned_loss=0.3249, over 28568.00 frames. ], tot_loss[loss=0.5009, simple_loss=0.4925, pruned_loss=0.2547, over 5717693.76 frames. ], libri_tot_loss[loss=0.7755, simple_loss=0.6859, pruned_loss=0.508, over 4223743.91 frames. ], giga_tot_loss[loss=0.4849, simple_loss=0.4809, pruned_loss=0.2447, over 5706445.39 frames. ], batch size: 336, lr: 4.68e-02, grad_scale: 8.0
+2023-02-28 11:15:18,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2752.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:15:19,600 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.32 vs. limit=5.0
+2023-02-28 11:15:29,840 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2768.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:15:31,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2771.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:15:40,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2781.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:15:40,837 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2781.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:15:43,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2784.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:15:56,925 INFO [train.py:968] (0/2) Epoch 1, batch 2800, giga_loss[loss=0.5817, simple_loss=0.5517, pruned_loss=0.3058, over 28963.00 frames. ], tot_loss[loss=0.5148, simple_loss=0.5025, pruned_loss=0.2637, over 5713947.43 frames. ], libri_tot_loss[loss=0.7669, simple_loss=0.68, pruned_loss=0.4987, over 4280138.47 frames. ], giga_tot_loss[loss=0.4993, simple_loss=0.491, pruned_loss=0.2539, over 5700193.36 frames. ], batch size: 145, lr: 4.67e-02, grad_scale: 8.0
+2023-02-28 11:15:57,397 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2800.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:15:59,595 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.036e+02 1.421e+03 1.878e+03 2.889e+03 7.423e+03, threshold=3.757e+03, percent-clipped=10.0
+2023-02-28 11:16:09,201 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8333, 2.0281, 3.2161, 1.5577], device='cuda:0'), covar=tensor([0.1282, 0.1124, 0.0294, 0.1427], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0341, 0.0393, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 11:16:09,205 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2813.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:16:38,316 INFO [train.py:968] (0/2) Epoch 1, batch 2850, giga_loss[loss=0.555, simple_loss=0.5501, pruned_loss=0.28, over 28917.00 frames. ], tot_loss[loss=0.5269, simple_loss=0.5113, pruned_loss=0.2713, over 5701039.40 frames. ], libri_tot_loss[loss=0.7575, simple_loss=0.6736, pruned_loss=0.4892, over 4336661.67 frames. ], giga_tot_loss[loss=0.5128, simple_loss=0.5009, pruned_loss=0.2625, over 5683178.32 frames. ], batch size: 213, lr: 4.66e-02, grad_scale: 8.0
+2023-02-28 11:16:41,971 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9721, 2.2259, 3.5515, 1.8089], device='cuda:0'), covar=tensor([0.1489, 0.1257, 0.0329, 0.1508], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0347, 0.0410, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 11:17:25,616 INFO [train.py:968] (0/2) Epoch 1, batch 2900, giga_loss[loss=0.5441, simple_loss=0.5389, pruned_loss=0.2747, over 28456.00 frames. ], tot_loss[loss=0.5379, simple_loss=0.5197, pruned_loss=0.2781, over 5672842.78 frames. ], libri_tot_loss[loss=0.753, simple_loss=0.6704, pruned_loss=0.4841, over 4365514.94 frames. ], giga_tot_loss[loss=0.5248, simple_loss=0.5101, pruned_loss=0.2699, over 5663813.04 frames. ], batch size: 71, lr: 4.65e-02, grad_scale: 4.0
+2023-02-28 11:17:29,063 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.364e+02 1.844e+03 2.565e+03 3.261e+03 8.769e+03, threshold=5.131e+03, percent-clipped=14.0
+2023-02-28 11:17:53,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-02-28 11:18:04,347 INFO [train.py:968] (0/2) Epoch 1, batch 2950, giga_loss[loss=0.5158, simple_loss=0.5156, pruned_loss=0.2581, over 28477.00 frames. ], tot_loss[loss=0.5415, simple_loss=0.5241, pruned_loss=0.2795, over 5685902.51 frames. ], libri_tot_loss[loss=0.7484, simple_loss=0.6672, pruned_loss=0.4793, over 4395650.45 frames. ], giga_tot_loss[loss=0.5305, simple_loss=0.516, pruned_loss=0.2726, over 5675074.37 frames. ], batch size: 71, lr: 4.64e-02, grad_scale: 4.0
+2023-02-28 11:18:52,033 INFO [train.py:968] (0/2) Epoch 1, batch 3000, giga_loss[loss=0.5753, simple_loss=0.5237, pruned_loss=0.3134, over 26764.00 frames. ], tot_loss[loss=0.5455, simple_loss=0.5272, pruned_loss=0.2819, over 5678104.15 frames. ], libri_tot_loss[loss=0.7438, simple_loss=0.6642, pruned_loss=0.4746, over 4421799.96 frames. ], giga_tot_loss[loss=0.5363, simple_loss=0.5203, pruned_loss=0.2762, over 5669057.61 frames. ], batch size: 555, lr: 4.63e-02, grad_scale: 4.0
+2023-02-28 11:18:52,038 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 11:19:00,219 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.5338, simple_loss=0.5608, pruned_loss=0.2534, over 944034.00 frames. 
+2023-02-28 11:19:00,219 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 17390MB
+2023-02-28 11:19:04,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.704e+02 1.639e+03 2.144e+03 2.977e+03 6.357e+03, threshold=4.288e+03, percent-clipped=4.0
+2023-02-28 11:19:38,810 INFO [train.py:968] (0/2) Epoch 1, batch 3050, giga_loss[loss=0.4009, simple_loss=0.4234, pruned_loss=0.1892, over 28304.00 frames. ], tot_loss[loss=0.5266, simple_loss=0.5143, pruned_loss=0.2695, over 5687663.91 frames. ], libri_tot_loss[loss=0.7356, simple_loss=0.6587, pruned_loss=0.4662, over 4472362.12 frames. ], giga_tot_loss[loss=0.5183, simple_loss=0.508, pruned_loss=0.2644, over 5673911.93 frames. ], batch size: 368, lr: 4.62e-02, grad_scale: 4.0
+2023-02-28 11:19:41,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4839, 1.5670, 1.5960, 1.4706], device='cuda:0'), covar=tensor([0.2147, 0.2044, 0.1693, 0.3367], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0564, 0.0439, 0.0572], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005], device='cuda:0')
+2023-02-28 11:20:01,508 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3078.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:20:18,651 INFO [train.py:968] (0/2) Epoch 1, batch 3100, giga_loss[loss=0.5385, simple_loss=0.5307, pruned_loss=0.2731, over 28968.00 frames. ], tot_loss[loss=0.5169, simple_loss=0.5074, pruned_loss=0.2632, over 5691811.09 frames. ], libri_tot_loss[loss=0.7306, simple_loss=0.655, pruned_loss=0.4616, over 4499324.31 frames. ], giga_tot_loss[loss=0.51, simple_loss=0.5024, pruned_loss=0.2588, over 5678574.76 frames. ], batch size: 136, lr: 4.61e-02, grad_scale: 4.0
+2023-02-28 11:20:21,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.050e+02 1.635e+03 2.188e+03 2.854e+03 7.300e+03, threshold=4.376e+03, percent-clipped=6.0
+2023-02-28 11:20:23,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3108.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:20:59,213 INFO [train.py:968] (0/2) Epoch 1, batch 3150, giga_loss[loss=0.4857, simple_loss=0.4963, pruned_loss=0.2376, over 28259.00 frames. ], tot_loss[loss=0.5166, simple_loss=0.5078, pruned_loss=0.2628, over 5685902.02 frames. ], libri_tot_loss[loss=0.7253, simple_loss=0.6514, pruned_loss=0.4562, over 4534224.87 frames. ], giga_tot_loss[loss=0.5099, simple_loss=0.5029, pruned_loss=0.2585, over 5670149.71 frames. ], batch size: 65, lr: 4.60e-02, grad_scale: 4.0
+2023-02-28 11:21:40,557 INFO [train.py:968] (0/2) Epoch 1, batch 3200, giga_loss[loss=0.5478, simple_loss=0.5426, pruned_loss=0.2765, over 28864.00 frames. ], tot_loss[loss=0.5181, simple_loss=0.5098, pruned_loss=0.2632, over 5685984.02 frames. ], libri_tot_loss[loss=0.7229, simple_loss=0.6497, pruned_loss=0.4539, over 4548090.63 frames. ], giga_tot_loss[loss=0.5126, simple_loss=0.5059, pruned_loss=0.2597, over 5671436.09 frames. ], batch size: 112, lr: 4.59e-02, grad_scale: 8.0
+2023-02-28 11:21:43,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.790e+03 2.255e+03 2.823e+03 8.056e+03, threshold=4.510e+03, percent-clipped=9.0
+2023-02-28 11:21:51,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5433, 2.1729, 4.0380, 1.7003], device='cuda:0'), covar=tensor([0.1000, 0.1578, 0.0610, 0.1773], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0382, 0.0510, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 11:21:52,872 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3217.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:22:01,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4137, 1.3223, 1.8921, 1.2591], device='cuda:0'), covar=tensor([0.2499, 0.2753, 0.1845, 0.3748], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0589, 0.0464, 0.0591], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005], device='cuda:0')
+2023-02-28 11:22:18,342 INFO [train.py:968] (0/2) Epoch 1, batch 3250, giga_loss[loss=0.506, simple_loss=0.5084, pruned_loss=0.2519, over 28877.00 frames. ], tot_loss[loss=0.5227, simple_loss=0.5134, pruned_loss=0.266, over 5687051.68 frames. ], libri_tot_loss[loss=0.7176, simple_loss=0.646, pruned_loss=0.4491, over 4572495.41 frames. ], giga_tot_loss[loss=0.5184, simple_loss=0.5104, pruned_loss=0.2632, over 5674143.90 frames. ], batch size: 227, lr: 4.58e-02, grad_scale: 8.0
+2023-02-28 11:22:19,640 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3251.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:22:21,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3254.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:22:42,349 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-02-28 11:22:45,790 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3283.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:22:59,154 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9534, 0.7542, 1.0642, 0.3287], device='cuda:0'), covar=tensor([0.0454, 0.0788, 0.0332, 0.0924], device='cuda:0'), in_proj_covar=tensor([0.0238, 0.0239, 0.0210, 0.0257], device='cuda:0'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-02-28 11:22:59,555 INFO [train.py:968] (0/2) Epoch 1, batch 3300, giga_loss[loss=0.5238, simple_loss=0.5193, pruned_loss=0.2641, over 28974.00 frames. ], tot_loss[loss=0.5244, simple_loss=0.5152, pruned_loss=0.2668, over 5690759.37 frames. ], libri_tot_loss[loss=0.7131, simple_loss=0.643, pruned_loss=0.4444, over 4600595.94 frames. ], giga_tot_loss[loss=0.5199, simple_loss=0.512, pruned_loss=0.2639, over 5680452.47 frames. ], batch size: 136, lr: 4.57e-02, grad_scale: 8.0
+2023-02-28 11:23:02,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.745e+02 1.694e+03 2.151e+03 2.944e+03 4.665e+03, threshold=4.303e+03, percent-clipped=1.0
+2023-02-28 11:23:23,770 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3330.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:23:37,901 INFO [train.py:968] (0/2) Epoch 1, batch 3350, giga_loss[loss=0.5051, simple_loss=0.502, pruned_loss=0.2541, over 28586.00 frames. ], tot_loss[loss=0.5225, simple_loss=0.5141, pruned_loss=0.2654, over 5683527.65 frames. ], libri_tot_loss[loss=0.7103, simple_loss=0.6412, pruned_loss=0.4414, over 4613534.02 frames. ], giga_tot_loss[loss=0.5178, simple_loss=0.5108, pruned_loss=0.2624, over 5683057.60 frames. ], batch size: 307, lr: 4.56e-02, grad_scale: 8.0
+2023-02-28 11:23:47,864 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.15 vs. limit=2.0
+2023-02-28 11:23:58,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5143, 2.5644, 4.4538, 2.2949], device='cuda:0'), covar=tensor([0.1237, 0.1076, 0.0179, 0.0922], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0373, 0.0427, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 11:24:19,788 INFO [train.py:968] (0/2) Epoch 1, batch 3400, giga_loss[loss=0.5989, simple_loss=0.5464, pruned_loss=0.3257, over 26469.00 frames. ], tot_loss[loss=0.5216, simple_loss=0.5135, pruned_loss=0.2649, over 5682630.21 frames. ], libri_tot_loss[loss=0.7075, simple_loss=0.6394, pruned_loss=0.4384, over 4632728.52 frames. ], giga_tot_loss[loss=0.5174, simple_loss=0.5104, pruned_loss=0.2622, over 5679440.41 frames. ], batch size: 555, lr: 4.55e-02, grad_scale: 8.0
+2023-02-28 11:24:23,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.206e+03 1.855e+03 2.546e+03 3.469e+03 5.795e+03, threshold=5.092e+03, percent-clipped=8.0
+2023-02-28 11:24:28,630 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3412.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:24:57,751 INFO [train.py:968] (0/2) Epoch 1, batch 3450, giga_loss[loss=0.5388, simple_loss=0.5399, pruned_loss=0.2688, over 28850.00 frames. ], tot_loss[loss=0.525, simple_loss=0.5157, pruned_loss=0.2671, over 5685531.78 frames. ], libri_tot_loss[loss=0.7014, simple_loss=0.6354, pruned_loss=0.4318, over 4681584.09 frames. ], giga_tot_loss[loss=0.5193, simple_loss=0.5117, pruned_loss=0.2635, over 5676186.16 frames. ], batch size: 199, lr: 4.54e-02, grad_scale: 4.0
+2023-02-28 11:25:01,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3453.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:25:34,851 INFO [train.py:968] (0/2) Epoch 1, batch 3500, giga_loss[loss=0.4826, simple_loss=0.5038, pruned_loss=0.2307, over 28591.00 frames. ], tot_loss[loss=0.5227, simple_loss=0.516, pruned_loss=0.2647, over 5695119.26 frames. ], libri_tot_loss[loss=0.6969, simple_loss=0.6325, pruned_loss=0.427, over 4717880.25 frames. ], giga_tot_loss[loss=0.5166, simple_loss=0.5116, pruned_loss=0.2608, over 5681761.79 frames. ], batch size: 78, lr: 4.53e-02, grad_scale: 4.0
+2023-02-28 11:25:38,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.216e+03 1.902e+03 2.368e+03 3.144e+03 1.358e+04, threshold=4.737e+03, percent-clipped=6.0
+2023-02-28 11:26:11,899 INFO [train.py:968] (0/2) Epoch 1, batch 3550, giga_loss[loss=0.5002, simple_loss=0.5157, pruned_loss=0.2424, over 28898.00 frames. ], tot_loss[loss=0.5184, simple_loss=0.5146, pruned_loss=0.2611, over 5701772.31 frames. ], libri_tot_loss[loss=0.6928, simple_loss=0.6298, pruned_loss=0.4228, over 4747453.00 frames. ], giga_tot_loss[loss=0.5125, simple_loss=0.5104, pruned_loss=0.2573, over 5686393.90 frames. ], batch size: 227, lr: 4.51e-02, grad_scale: 4.0
+2023-02-28 11:26:38,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-02-28 11:26:47,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3592.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:26:47,451 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.14 vs. limit=2.0
+2023-02-28 11:26:51,067 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3596.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:26:54,035 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3599.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:26:54,422 INFO [train.py:968] (0/2) Epoch 1, batch 3600, giga_loss[loss=0.5236, simple_loss=0.5202, pruned_loss=0.2635, over 28850.00 frames. ], tot_loss[loss=0.5139, simple_loss=0.5122, pruned_loss=0.2578, over 5700878.94 frames. ], libri_tot_loss[loss=0.6923, simple_loss=0.6295, pruned_loss=0.4222, over 4753379.74 frames. ], giga_tot_loss[loss=0.5088, simple_loss=0.5086, pruned_loss=0.2545, over 5687847.32 frames. ], batch size: 106, lr: 4.50e-02, grad_scale: 8.0
+2023-02-28 11:26:58,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.555e+02 1.630e+03 2.177e+03 3.063e+03 8.097e+03, threshold=4.355e+03, percent-clipped=7.0
+2023-02-28 11:27:14,588 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3628.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:27:33,572 INFO [train.py:968] (0/2) Epoch 1, batch 3650, giga_loss[loss=0.446, simple_loss=0.4352, pruned_loss=0.2284, over 23602.00 frames. ], tot_loss[loss=0.5048, simple_loss=0.5058, pruned_loss=0.2519, over 5693414.84 frames. ], libri_tot_loss[loss=0.6907, simple_loss=0.6285, pruned_loss=0.4205, over 4764894.06 frames. ], giga_tot_loss[loss=0.5004, simple_loss=0.5026, pruned_loss=0.2491, over 5681307.98 frames. ], batch size: 705, lr: 4.49e-02, grad_scale: 8.0
+2023-02-28 11:27:45,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5483, 1.3879, 1.7448, 0.7101], device='cuda:0'), covar=tensor([0.0695, 0.0743, 0.0441, 0.1663], device='cuda:0'), in_proj_covar=tensor([0.0296, 0.0277, 0.0254, 0.0350], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-02-28 11:27:58,485 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3681.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:28:12,489 INFO [train.py:968] (0/2) Epoch 1, batch 3700, giga_loss[loss=0.5345, simple_loss=0.5333, pruned_loss=0.2679, over 28791.00 frames. ], tot_loss[loss=0.5064, simple_loss=0.506, pruned_loss=0.2534, over 5707059.62 frames. ], libri_tot_loss[loss=0.6895, simple_loss=0.6278, pruned_loss=0.4189, over 4781315.07 frames. ], giga_tot_loss[loss=0.5011, simple_loss=0.5022, pruned_loss=0.25, over 5695216.23 frames. ], batch size: 284, lr: 4.48e-02, grad_scale: 8.0
+2023-02-28 11:28:16,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3705.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:28:16,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.482e+02 1.581e+03 2.445e+03 3.469e+03 1.889e+04, threshold=4.889e+03, percent-clipped=15.0
+2023-02-28 11:28:16,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4889, 1.4591, 1.0267, 1.3301], device='cuda:0'), covar=tensor([0.1710, 0.1563, 0.2146, 0.2485], device='cuda:0'), in_proj_covar=tensor([0.0507, 0.0583, 0.0582, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0006], device='cuda:0')
+2023-02-28 11:28:38,448 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3735.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:28:40,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3738.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:28:48,307 INFO [train.py:968] (0/2) Epoch 1, batch 3750, giga_loss[loss=0.4869, simple_loss=0.4937, pruned_loss=0.2401, over 28756.00 frames. ], tot_loss[loss=0.5019, simple_loss=0.503, pruned_loss=0.2504, over 5710322.32 frames. ], libri_tot_loss[loss=0.6854, simple_loss=0.6251, pruned_loss=0.4148, over 4807520.94 frames. ], giga_tot_loss[loss=0.4965, simple_loss=0.4992, pruned_loss=0.2469, over 5697805.24 frames. ], batch size: 284, lr: 4.47e-02, grad_scale: 8.0
+2023-02-28 11:29:03,421 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3767.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:29:19,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3787.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:29:29,603 INFO [train.py:968] (0/2) Epoch 1, batch 3800, giga_loss[loss=0.6036, simple_loss=0.5511, pruned_loss=0.3281, over 26691.00 frames. ], tot_loss[loss=0.5035, simple_loss=0.5043, pruned_loss=0.2514, over 5704981.20 frames. ], libri_tot_loss[loss=0.6817, simple_loss=0.6227, pruned_loss=0.4111, over 4830884.84 frames. ], giga_tot_loss[loss=0.498, simple_loss=0.5004, pruned_loss=0.2479, over 5694049.29 frames. ], batch size: 555, lr: 4.46e-02, grad_scale: 8.0
+2023-02-28 11:29:35,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.764e+02 1.828e+03 2.341e+03 3.297e+03 7.298e+03, threshold=4.682e+03, percent-clipped=6.0
+2023-02-28 11:29:43,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5968, 1.7942, 1.4367, 1.5083], device='cuda:0'), covar=tensor([0.1857, 0.2637, 0.1783, 0.3244], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0503, 0.0403, 0.0597], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004], device='cuda:0')
+2023-02-28 11:30:01,355 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-02-28 11:30:05,473 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3848.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:30:07,385 INFO [train.py:968] (0/2) Epoch 1, batch 3850, giga_loss[loss=0.5162, simple_loss=0.5145, pruned_loss=0.259, over 27938.00 frames. ], tot_loss[loss=0.5025, simple_loss=0.5039, pruned_loss=0.2505, over 5696823.66 frames. ], libri_tot_loss[loss=0.6793, simple_loss=0.621, pruned_loss=0.4086, over 4842148.03 frames. ], giga_tot_loss[loss=0.4969, simple_loss=0.5001, pruned_loss=0.2468, over 5694420.29 frames. ], batch size: 412, lr: 4.45e-02, grad_scale: 8.0
+2023-02-28 11:30:08,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3851.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:30:31,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3880.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:30:47,204 INFO [train.py:968] (0/2) Epoch 1, batch 3900, giga_loss[loss=0.4374, simple_loss=0.4608, pruned_loss=0.207, over 28659.00 frames. ], tot_loss[loss=0.4969, simple_loss=0.5013, pruned_loss=0.2463, over 5707932.00 frames. ], libri_tot_loss[loss=0.6757, simple_loss=0.6186, pruned_loss=0.4053, over 4863426.59 frames. ], giga_tot_loss[loss=0.4919, simple_loss=0.498, pruned_loss=0.243, over 5702392.19 frames. ], batch size: 92, lr: 4.44e-02, grad_scale: 8.0
+2023-02-28 11:30:52,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.546e+02 1.658e+03 2.144e+03 2.840e+03 5.109e+03, threshold=4.287e+03, percent-clipped=4.0
+2023-02-28 11:30:59,619 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-02-28 11:31:10,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3930.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:31:11,693 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-02-28 11:31:13,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3933.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:31:18,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3939.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:31:26,084 INFO [train.py:968] (0/2) Epoch 1, batch 3950, giga_loss[loss=0.4334, simple_loss=0.4591, pruned_loss=0.2038, over 29034.00 frames. ], tot_loss[loss=0.4963, simple_loss=0.5003, pruned_loss=0.2461, over 5706630.37 frames. ], libri_tot_loss[loss=0.6727, simple_loss=0.6163, pruned_loss=0.4025, over 4883841.09 frames. ], giga_tot_loss[loss=0.4914, simple_loss=0.4973, pruned_loss=0.2428, over 5699087.88 frames. ], batch size: 128, lr: 4.43e-02, grad_scale: 4.0
+2023-02-28 11:31:35,902 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3962.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:32:04,554 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-4000.pt
+2023-02-28 11:32:04,867 INFO [train.py:968] (0/2) Epoch 1, batch 4000, giga_loss[loss=0.4677, simple_loss=0.4813, pruned_loss=0.227, over 28869.00 frames. ], tot_loss[loss=0.4957, simple_loss=0.4992, pruned_loss=0.2461, over 5712743.44 frames. ], libri_tot_loss[loss=0.6658, simple_loss=0.6114, pruned_loss=0.3962, over 4923353.24 frames. ], giga_tot_loss[loss=0.4907, simple_loss=0.4963, pruned_loss=0.2426, over 5700253.38 frames. ], batch size: 199, lr: 4.42e-02, grad_scale: 8.0
+2023-02-28 11:32:09,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.523e+02 1.777e+03 2.366e+03 3.540e+03 1.495e+04, threshold=4.732e+03, percent-clipped=16.0
+2023-02-28 11:32:43,297 INFO [train.py:968] (0/2) Epoch 1, batch 4050, giga_loss[loss=0.4782, simple_loss=0.4872, pruned_loss=0.2346, over 28988.00 frames. ], tot_loss[loss=0.49, simple_loss=0.4944, pruned_loss=0.2427, over 5715008.21 frames. ], libri_tot_loss[loss=0.6626, simple_loss=0.609, pruned_loss=0.3936, over 4937829.23 frames. ], giga_tot_loss[loss=0.4861, simple_loss=0.4924, pruned_loss=0.2398, over 5702820.50 frames. ], batch size: 136, lr: 4.41e-02, grad_scale: 8.0
+2023-02-28 11:32:48,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4056.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:32:57,354 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=4069.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:33:19,601 INFO [train.py:968] (0/2) Epoch 1, batch 4100, libri_loss[loss=0.5195, simple_loss=0.5024, pruned_loss=0.2682, over 29559.00 frames. ], tot_loss[loss=0.4871, simple_loss=0.4915, pruned_loss=0.2413, over 5723550.37 frames. ], libri_tot_loss[loss=0.6577, simple_loss=0.6055, pruned_loss=0.389, over 4969433.02 frames. ], giga_tot_loss[loss=0.4821, simple_loss=0.4888, pruned_loss=0.2377, over 5709281.44 frames. ], batch size: 75, lr: 4.40e-02, grad_scale: 4.0
+2023-02-28 11:33:25,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.743e+02 1.677e+03 2.228e+03 3.075e+03 6.170e+03, threshold=4.455e+03, percent-clipped=6.0
+2023-02-28 11:33:39,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9977, 1.6259, 1.6162, 1.4163], device='cuda:0'), covar=tensor([0.1139, 0.1663, 0.1226, 0.2389], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0545, 0.0437, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005], device='cuda:0')
+2023-02-28 11:33:58,407 INFO [train.py:968] (0/2) Epoch 1, batch 4150, giga_loss[loss=0.4437, simple_loss=0.4585, pruned_loss=0.2144, over 28623.00 frames. ], tot_loss[loss=0.4864, simple_loss=0.4906, pruned_loss=0.2411, over 5712719.01 frames. ], libri_tot_loss[loss=0.6541, simple_loss=0.6031, pruned_loss=0.3855, over 4986252.29 frames. ], giga_tot_loss[loss=0.4808, simple_loss=0.4873, pruned_loss=0.2371, over 5705455.11 frames. ], batch size: 85, lr: 4.39e-02, grad_scale: 4.0
+2023-02-28 11:34:10,974 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-02-28 11:34:16,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8608, 2.2870, 3.5735, 1.9408], device='cuda:0'), covar=tensor([0.1459, 0.1002, 0.0289, 0.1040], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0374, 0.0450, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 11:34:17,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2758, 1.4655, 1.6850, 1.3584], device='cuda:0'), covar=tensor([0.2114, 0.2078, 0.1784, 0.3270], device='cuda:0'), in_proj_covar=tensor([0.0510, 0.0583, 0.0473, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0006], device='cuda:0')
+2023-02-28 11:34:36,969 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4199.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:34:37,334 INFO [train.py:968] (0/2) Epoch 1, batch 4200, giga_loss[loss=0.4528, simple_loss=0.4678, pruned_loss=0.2189, over 28856.00 frames. ], tot_loss[loss=0.4858, simple_loss=0.4896, pruned_loss=0.241, over 5720312.14 frames. ], libri_tot_loss[loss=0.6496, simple_loss=0.6, pruned_loss=0.3816, over 5009044.71 frames. ], giga_tot_loss[loss=0.4809, simple_loss=0.4868, pruned_loss=0.2375, over 5710297.85 frames. ], batch size: 199, lr: 4.38e-02, grad_scale: 4.0
+2023-02-28 11:34:39,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4202.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:34:44,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.904e+02 1.497e+03 2.080e+03 2.883e+03 7.543e+03, threshold=4.160e+03, percent-clipped=5.0
+2023-02-28 11:35:02,336 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4231.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:35:17,646 INFO [train.py:968] (0/2) Epoch 1, batch 4250, giga_loss[loss=0.4277, simple_loss=0.4538, pruned_loss=0.2008, over 28422.00 frames. ], tot_loss[loss=0.4847, simple_loss=0.4879, pruned_loss=0.2408, over 5716420.06 frames. ], libri_tot_loss[loss=0.6484, simple_loss=0.5993, pruned_loss=0.3804, over 5017729.34 frames. ], giga_tot_loss[loss=0.4804, simple_loss=0.4853, pruned_loss=0.2377, over 5707118.24 frames. ], batch size: 60, lr: 4.36e-02, grad_scale: 4.0
+2023-02-28 11:35:55,559 INFO [train.py:968] (0/2) Epoch 1, batch 4300, libri_loss[loss=0.4934, simple_loss=0.4878, pruned_loss=0.2495, over 28505.00 frames. ], tot_loss[loss=0.4811, simple_loss=0.4846, pruned_loss=0.2388, over 5717074.59 frames. ], libri_tot_loss[loss=0.6464, simple_loss=0.598, pruned_loss=0.3783, over 5033638.57 frames. ], giga_tot_loss[loss=0.4763, simple_loss=0.4814, pruned_loss=0.2355, over 5707932.97 frames. ], batch size: 63, lr: 4.35e-02, grad_scale: 4.0
+2023-02-28 11:35:57,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9611, 2.1889, 2.9902, 1.8769], device='cuda:0'), covar=tensor([0.1095, 0.0795, 0.0279, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0368, 0.0441, 0.0362], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 11:36:01,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.699e+02 1.550e+03 2.005e+03 2.714e+03 6.640e+03, threshold=4.010e+03, percent-clipped=5.0
+2023-02-28 11:36:06,030 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4314.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:36:09,326 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=4319.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:36:26,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5524, 1.4737, 1.9490, 0.3745], device='cuda:0'), covar=tensor([0.1115, 0.1467, 0.0609, 0.3517], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0320, 0.0305, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003], device='cuda:0')
+2023-02-28 11:36:32,003 INFO [train.py:968] (0/2) Epoch 1, batch 4350, giga_loss[loss=0.4135, simple_loss=0.4402, pruned_loss=0.1934, over 28972.00 frames. ], tot_loss[loss=0.4774, simple_loss=0.4809, pruned_loss=0.237, over 5711689.45 frames. ], libri_tot_loss[loss=0.6427, simple_loss=0.5954, pruned_loss=0.3751, over 5045299.78 frames. ], giga_tot_loss[loss=0.4729, simple_loss=0.478, pruned_loss=0.2339, over 5708717.10 frames. ], batch size: 136, lr: 4.34e-02, grad_scale: 4.0
+2023-02-28 11:37:10,563 INFO [train.py:968] (0/2) Epoch 1, batch 4400, giga_loss[loss=0.6352, simple_loss=0.5698, pruned_loss=0.3502, over 26774.00 frames. ], tot_loss[loss=0.4757, simple_loss=0.4799, pruned_loss=0.2358, over 5709298.48 frames. ], libri_tot_loss[loss=0.6394, simple_loss=0.5934, pruned_loss=0.3717, over 5070357.27 frames. ], giga_tot_loss[loss=0.4703, simple_loss=0.476, pruned_loss=0.2323, over 5702174.83 frames. ], batch size: 555, lr: 4.33e-02, grad_scale: 8.0
+2023-02-28 11:37:17,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.554e+02 1.687e+03 2.082e+03 2.903e+03 9.780e+03, threshold=4.163e+03, percent-clipped=11.0
+2023-02-28 11:37:18,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1451, 1.3470, 0.8525, 1.3027], device='cuda:0'), covar=tensor([0.1093, 0.2279, 0.1810, 0.1067], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0576, 0.0526, 0.0372], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002], device='cuda:0')
+2023-02-28 11:37:46,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4444.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:37:48,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4686, 1.5305, 1.7625, 1.2845], device='cuda:0'), covar=tensor([0.2316, 0.1758, 0.1398, 0.3501], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0530, 0.0533, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0006], device='cuda:0')
+2023-02-28 11:37:50,950 INFO [train.py:968] (0/2) Epoch 1, batch 4450, giga_loss[loss=0.5388, simple_loss=0.5272, pruned_loss=0.2752, over 28740.00 frames. ], tot_loss[loss=0.4771, simple_loss=0.4815, pruned_loss=0.2364, over 5704997.10 frames. ], libri_tot_loss[loss=0.6379, simple_loss=0.5924, pruned_loss=0.37, over 5082505.77 frames. ], giga_tot_loss[loss=0.4711, simple_loss=0.4773, pruned_loss=0.2325, over 5699791.15 frames. ], batch size: 262, lr: 4.32e-02, grad_scale: 8.0
+2023-02-28 11:37:56,714 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4457.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:37:58,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4460.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:38:07,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0998, 2.4635, 3.9819, 1.9667], device='cuda:0'), covar=tensor([0.1441, 0.1053, 0.0239, 0.0939], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0364, 0.0440, 0.0357], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 11:38:22,598 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4489.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:38:30,107 INFO [train.py:968] (0/2) Epoch 1, batch 4500, giga_loss[loss=0.4717, simple_loss=0.4839, pruned_loss=0.2298, over 28870.00 frames. ], tot_loss[loss=0.479, simple_loss=0.4837, pruned_loss=0.2371, over 5711071.33 frames. ], libri_tot_loss[loss=0.6335, simple_loss=0.5892, pruned_loss=0.3662, over 5101797.89 frames. ], giga_tot_loss[loss=0.4727, simple_loss=0.4795, pruned_loss=0.233, over 5708286.99 frames. ], batch size: 186, lr: 4.31e-02, grad_scale: 4.0
+2023-02-28 11:38:37,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.349e+02 1.605e+03 2.058e+03 2.668e+03 1.102e+04, threshold=4.116e+03, percent-clipped=6.0
+2023-02-28 11:38:57,715 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=4534.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:39:01,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2837, 1.1524, 1.4070, 1.0942], device='cuda:0'), covar=tensor([0.1009, 0.2254, 0.1049, 0.1082], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0308, 0.0237, 0.0237], device='cuda:0'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001], device='cuda:0')
+2023-02-28 11:39:08,435 INFO [train.py:968] (0/2) Epoch 1, batch 4550, giga_loss[loss=0.4458, simple_loss=0.4727, pruned_loss=0.2095, over 28989.00 frames. ], tot_loss[loss=0.478, simple_loss=0.4845, pruned_loss=0.2357, over 5711765.49 frames. ], libri_tot_loss[loss=0.6299, simple_loss=0.587, pruned_loss=0.3625, over 5123209.35 frames. ], giga_tot_loss[loss=0.4703, simple_loss=0.4791, pruned_loss=0.2307, over 5712170.23 frames. ], batch size: 227, lr: 4.30e-02, grad_scale: 4.0
+2023-02-28 11:39:17,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3570, 1.8764, 1.7573, 1.3646], device='cuda:0'), covar=tensor([0.1911, 0.1393, 0.1364, 0.3104], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0582, 0.0479, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0006], device='cuda:0')
+2023-02-28 11:39:42,420 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4587.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:39:44,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4590.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:39:52,986 INFO [train.py:968] (0/2) Epoch 1, batch 4600, giga_loss[loss=0.4163, simple_loss=0.4519, pruned_loss=0.1903, over 28984.00 frames. ], tot_loss[loss=0.4736, simple_loss=0.4825, pruned_loss=0.2323, over 5701855.89 frames. ], libri_tot_loss[loss=0.6293, simple_loss=0.5866, pruned_loss=0.3619, over 5127371.30 frames. ], giga_tot_loss[loss=0.4674, simple_loss=0.4782, pruned_loss=0.2283, over 5701208.49 frames. ], batch size: 164, lr: 4.29e-02, grad_scale: 4.0
+2023-02-28 11:39:59,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.039e+02 1.439e+03 1.886e+03 2.305e+03 1.013e+04, threshold=3.772e+03, percent-clipped=7.0
+2023-02-28 11:40:08,577 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4619.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:40:31,302 INFO [train.py:968] (0/2) Epoch 1, batch 4650, giga_loss[loss=0.4594, simple_loss=0.4779, pruned_loss=0.2204, over 28653.00 frames. ], tot_loss[loss=0.4766, simple_loss=0.4846, pruned_loss=0.2343, over 5700237.39 frames. ], libri_tot_loss[loss=0.6277, simple_loss=0.5856, pruned_loss=0.3599, over 5152375.27 frames. ], giga_tot_loss[loss=0.4679, simple_loss=0.4786, pruned_loss=0.2286, over 5694832.04 frames. ], batch size: 336, lr: 4.28e-02, grad_scale: 4.0
+2023-02-28 11:40:54,592 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=4681.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:41:04,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4694.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:41:08,877 INFO [train.py:968] (0/2) Epoch 1, batch 4700, giga_loss[loss=0.4624, simple_loss=0.4771, pruned_loss=0.2239, over 28811.00 frames. ], tot_loss[loss=0.4805, simple_loss=0.4877, pruned_loss=0.2367, over 5706401.32 frames. ], libri_tot_loss[loss=0.6245, simple_loss=0.5835, pruned_loss=0.3566, over 5178922.97 frames. ], giga_tot_loss[loss=0.4706, simple_loss=0.4808, pruned_loss=0.2302, over 5697601.33 frames. ], batch size: 145, lr: 4.27e-02, grad_scale: 4.0
+2023-02-28 11:41:15,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.775e+02 1.860e+03 2.656e+03 3.670e+03 1.258e+04, threshold=5.311e+03, percent-clipped=21.0
+2023-02-28 11:41:24,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2775, 2.7918, 2.7998, 1.8405], device='cuda:0'), covar=tensor([0.0853, 0.1612, 0.0683, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0222, 0.0352, 0.0274, 0.0263], device='cuda:0'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001], device='cuda:0')
+2023-02-28 11:41:47,058 INFO [train.py:968] (0/2) Epoch 1, batch 4750, giga_loss[loss=0.4642, simple_loss=0.4809, pruned_loss=0.2237, over 28997.00 frames. ], tot_loss[loss=0.4814, simple_loss=0.4883, pruned_loss=0.2372, over 5716535.68 frames. ], libri_tot_loss[loss=0.6189, simple_loss=0.5799, pruned_loss=0.3516, over 5209073.62 frames. ], giga_tot_loss[loss=0.4721, simple_loss=0.4818, pruned_loss=0.2312, over 5703296.28 frames. ], batch size: 164, lr: 4.26e-02, grad_scale: 4.0
+2023-02-28 11:42:26,119 INFO [train.py:968] (0/2) Epoch 1, batch 4800, giga_loss[loss=0.4701, simple_loss=0.4777, pruned_loss=0.2313, over 28565.00 frames. ], tot_loss[loss=0.4805, simple_loss=0.4877, pruned_loss=0.2366, over 5718740.76 frames. ], libri_tot_loss[loss=0.6172, simple_loss=0.5789, pruned_loss=0.3501, over 5219511.08 frames. ], giga_tot_loss[loss=0.4726, simple_loss=0.4821, pruned_loss=0.2315, over 5705869.25 frames. ], batch size: 85, lr: 4.25e-02, grad_scale: 8.0
+2023-02-28 11:42:35,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.662e+03 2.151e+03 2.715e+03 5.435e+03, threshold=4.301e+03, percent-clipped=2.0
+2023-02-28 11:42:56,983 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4837.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:42:58,891 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4840.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:43:06,450 INFO [train.py:968] (0/2) Epoch 1, batch 4850, giga_loss[loss=0.445, simple_loss=0.4723, pruned_loss=0.2089, over 28931.00 frames. ], tot_loss[loss=0.4841, simple_loss=0.4904, pruned_loss=0.2389, over 5707253.77 frames. ], libri_tot_loss[loss=0.616, simple_loss=0.5781, pruned_loss=0.3489, over 5217852.66 frames. ], giga_tot_loss[loss=0.4772, simple_loss=0.4855, pruned_loss=0.2345, over 5705648.80 frames. ], batch size: 164, lr: 4.24e-02, grad_scale: 8.0
+2023-02-28 11:43:13,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3235, 0.9253, 1.6785, 0.2712], device='cuda:0'), covar=tensor([0.0736, 0.0759, 0.0327, 0.2333], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0328, 0.0314, 0.0506], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003], device='cuda:0')
+2023-02-28 11:43:22,013 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4869.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:43:22,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.04 vs. limit=2.0
+2023-02-28 11:43:31,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3460, 1.2782, 1.9042, 0.6358], device='cuda:0'), covar=tensor([0.1056, 0.0783, 0.0407, 0.2672], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0334, 0.0314, 0.0511], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003], device='cuda:0')
+2023-02-28 11:43:45,539 INFO [train.py:968] (0/2) Epoch 1, batch 4900, giga_loss[loss=0.4732, simple_loss=0.4923, pruned_loss=0.2271, over 28935.00 frames. ], tot_loss[loss=0.4889, simple_loss=0.4936, pruned_loss=0.2421, over 5708384.04 frames. ], libri_tot_loss[loss=0.6149, simple_loss=0.5775, pruned_loss=0.3476, over 5230896.59 frames. ], giga_tot_loss[loss=0.482, simple_loss=0.4887, pruned_loss=0.2376, over 5704245.53 frames. ], batch size: 213, lr: 4.23e-02, grad_scale: 8.0
+2023-02-28 11:43:51,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.129e+03 1.862e+03 2.414e+03 3.079e+03 6.069e+03, threshold=4.828e+03, percent-clipped=9.0
+2023-02-28 11:43:51,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4909.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:44:14,107 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-02-28 11:44:25,360 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3738, 1.3959, 0.9681, 1.5328], device='cuda:0'), covar=tensor([0.0954, 0.1723, 0.1440, 0.1064], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0586, 0.0556, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 11:44:26,302 INFO [train.py:968] (0/2) Epoch 1, batch 4950, giga_loss[loss=0.4035, simple_loss=0.4396, pruned_loss=0.1837, over 28831.00 frames. ], tot_loss[loss=0.4844, simple_loss=0.4916, pruned_loss=0.2386, over 5709386.10 frames. ], libri_tot_loss[loss=0.6142, simple_loss=0.5772, pruned_loss=0.3468, over 5237319.65 frames. ], giga_tot_loss[loss=0.4783, simple_loss=0.4871, pruned_loss=0.2347, over 5704525.67 frames. ], batch size: 112, lr: 4.22e-02, grad_scale: 8.0
+2023-02-28 11:45:05,780 INFO [train.py:968] (0/2) Epoch 1, batch 5000, giga_loss[loss=0.5428, simple_loss=0.5343, pruned_loss=0.2756, over 28598.00 frames. ], tot_loss[loss=0.4839, simple_loss=0.4916, pruned_loss=0.2381, over 5709720.03 frames. ], libri_tot_loss[loss=0.6122, simple_loss=0.576, pruned_loss=0.3448, over 5253198.51 frames. ], giga_tot_loss[loss=0.4778, simple_loss=0.4871, pruned_loss=0.2342, over 5702330.23 frames. ], batch size: 336, lr: 4.20e-02, grad_scale: 8.0
+2023-02-28 11:45:11,724 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-02-28 11:45:12,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.666e+02 1.619e+03 1.991e+03 2.395e+03 4.418e+03, threshold=3.983e+03, percent-clipped=0.0
+2023-02-28 11:45:43,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5140, 1.6584, 2.0625, 1.4186], device='cuda:0'), covar=tensor([0.1551, 0.1527, 0.1288, 0.2644], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0599, 0.0483, 0.0627], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0006], device='cuda:0')
+2023-02-28 11:45:43,896 INFO [train.py:968] (0/2) Epoch 1, batch 5050, giga_loss[loss=0.4338, simple_loss=0.4675, pruned_loss=0.2001, over 28940.00 frames. ], tot_loss[loss=0.482, simple_loss=0.4908, pruned_loss=0.2366, over 5713706.88 frames. ], libri_tot_loss[loss=0.6102, simple_loss=0.5749, pruned_loss=0.3428, over 5269219.98 frames. ], giga_tot_loss[loss=0.4758, simple_loss=0.4863, pruned_loss=0.2327, over 5703646.88 frames. ], batch size: 227, lr: 4.19e-02, grad_scale: 8.0
+2023-02-28 11:45:45,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5052.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:45:48,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5055.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:45:48,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=5056.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:46:10,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5084.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:46:23,690 INFO [train.py:968] (0/2) Epoch 1, batch 5100, giga_loss[loss=0.4151, simple_loss=0.4481, pruned_loss=0.1911, over 28655.00 frames. ], tot_loss[loss=0.476, simple_loss=0.487, pruned_loss=0.2325, over 5707682.09 frames. ], libri_tot_loss[loss=0.6098, simple_loss=0.5746, pruned_loss=0.3422, over 5268683.79 frames. ], giga_tot_loss[loss=0.47, simple_loss=0.4826, pruned_loss=0.2286, over 5706511.40 frames. ], batch size: 262, lr: 4.18e-02, grad_scale: 8.0
+2023-02-28 11:46:31,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.399e+02 1.452e+03 1.925e+03 2.569e+03 5.360e+03, threshold=3.849e+03, percent-clipped=3.0
+2023-02-28 11:46:55,063 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5495, 1.6488, 2.1783, 0.3065], device='cuda:0'), covar=tensor([0.1497, 0.1396, 0.0708, 0.5287], device='cuda:0'), in_proj_covar=tensor([0.0439, 0.0348, 0.0337, 0.0544], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003], device='cuda:0')
+2023-02-28 11:47:06,168 INFO [train.py:968] (0/2) Epoch 1, batch 5150, giga_loss[loss=0.4185, simple_loss=0.4448, pruned_loss=0.1961, over 28834.00 frames. ], tot_loss[loss=0.4726, simple_loss=0.4836, pruned_loss=0.2308, over 5703925.03 frames. ], libri_tot_loss[loss=0.6073, simple_loss=0.5729, pruned_loss=0.34, over 5287310.16 frames. ], giga_tot_loss[loss=0.4661, simple_loss=0.4791, pruned_loss=0.2266, over 5697965.67 frames. ], batch size: 199, lr: 4.17e-02, grad_scale: 8.0
+2023-02-28 11:47:32,471 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-02-28 11:47:48,380 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5199.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:47:48,763 INFO [train.py:968] (0/2) Epoch 1, batch 5200, giga_loss[loss=0.4114, simple_loss=0.4437, pruned_loss=0.1895, over 28917.00 frames. ], tot_loss[loss=0.4648, simple_loss=0.4776, pruned_loss=0.226, over 5708568.09 frames. ], libri_tot_loss[loss=0.6073, simple_loss=0.573, pruned_loss=0.3398, over 5290090.61 frames. ], giga_tot_loss[loss=0.4591, simple_loss=0.4736, pruned_loss=0.2223, over 5703318.30 frames. ], batch size: 145, lr: 4.16e-02, grad_scale: 8.0
+2023-02-28 11:47:50,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5202.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:47:56,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.160e+02 1.484e+03 1.874e+03 2.502e+03 5.736e+03, threshold=3.747e+03, percent-clipped=8.0
+2023-02-28 11:48:12,185 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5231.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:48:28,782 INFO [train.py:968] (0/2) Epoch 1, batch 5250, giga_loss[loss=0.4029, simple_loss=0.4406, pruned_loss=0.1826, over 28812.00 frames. ], tot_loss[loss=0.4654, simple_loss=0.4787, pruned_loss=0.2261, over 5715413.93 frames. ], libri_tot_loss[loss=0.6049, simple_loss=0.5711, pruned_loss=0.3377, over 5308437.93 frames. ], giga_tot_loss[loss=0.4589, simple_loss=0.4744, pruned_loss=0.2217, over 5705823.21 frames. ], batch size: 119, lr: 4.15e-02, grad_scale: 8.0
+2023-02-28 11:49:02,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8255, 1.2753, 1.3899, 1.4050], device='cuda:0'), covar=tensor([0.0561, 0.1327, 0.1011, 0.1406], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0704, 0.0539, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0006], device='cuda:0')
+2023-02-28 11:49:12,202 INFO [train.py:968] (0/2) Epoch 1, batch 5300, giga_loss[loss=0.5773, simple_loss=0.5528, pruned_loss=0.3009, over 27604.00 frames. ], tot_loss[loss=0.4674, simple_loss=0.4819, pruned_loss=0.2265, over 5706917.79 frames. ], libri_tot_loss[loss=0.6032, simple_loss=0.5699, pruned_loss=0.3362, over 5311068.12 frames. ], giga_tot_loss[loss=0.4613, simple_loss=0.4779, pruned_loss=0.2223, over 5704370.15 frames. ], batch size: 472, lr: 4.14e-02, grad_scale: 8.0
+2023-02-28 11:49:19,527 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.648e+02 1.537e+03 1.976e+03 2.474e+03 6.872e+03, threshold=3.951e+03, percent-clipped=5.0
+2023-02-28 11:49:50,464 INFO [train.py:968] (0/2) Epoch 1, batch 5350, giga_loss[loss=0.4299, simple_loss=0.4424, pruned_loss=0.2087, over 28653.00 frames. ], tot_loss[loss=0.4731, simple_loss=0.4855, pruned_loss=0.2304, over 5720055.69 frames. ], libri_tot_loss[loss=0.6016, simple_loss=0.5689, pruned_loss=0.3344, over 5332842.79 frames. ], giga_tot_loss[loss=0.4645, simple_loss=0.4799, pruned_loss=0.2245, over 5712515.72 frames. ], batch size: 78, lr: 4.13e-02, grad_scale: 4.0
+2023-02-28 11:50:30,226 INFO [train.py:968] (0/2) Epoch 1, batch 5400, giga_loss[loss=0.3943, simple_loss=0.428, pruned_loss=0.1803, over 29020.00 frames. ], tot_loss[loss=0.4729, simple_loss=0.4836, pruned_loss=0.2311, over 5721403.79 frames. ], libri_tot_loss[loss=0.6012, simple_loss=0.5687, pruned_loss=0.3338, over 5340421.80 frames. ], giga_tot_loss[loss=0.4646, simple_loss=0.4781, pruned_loss=0.2255, over 5713774.86 frames. ], batch size: 164, lr: 4.12e-02, grad_scale: 4.0
+2023-02-28 11:50:38,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.298e+02 1.515e+03 2.139e+03 3.075e+03 1.222e+04, threshold=4.279e+03, percent-clipped=10.0
+2023-02-28 11:50:51,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2550, 0.6023, 1.6665, 0.2509], device='cuda:0'), covar=tensor([0.0673, 0.0766, 0.0337, 0.1911], device='cuda:0'), in_proj_covar=tensor([0.0472, 0.0388, 0.0343, 0.0564], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003], device='cuda:0')
+2023-02-28 11:50:52,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9456, 1.1491, 1.1916, 1.0293], device='cuda:0'), covar=tensor([0.1883, 0.1521, 0.1437, 0.2889], device='cuda:0'), in_proj_covar=tensor([0.0507, 0.0608, 0.0489, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 11:51:11,243 INFO [train.py:968] (0/2) Epoch 1, batch 5450, giga_loss[loss=0.3996, simple_loss=0.4224, pruned_loss=0.1884, over 28862.00 frames. ], tot_loss[loss=0.4725, simple_loss=0.4819, pruned_loss=0.2315, over 5729190.07 frames. ], libri_tot_loss[loss=0.5991, simple_loss=0.5675, pruned_loss=0.3318, over 5357327.51 frames. ], giga_tot_loss[loss=0.4641, simple_loss=0.4762, pruned_loss=0.226, over 5718163.49 frames. ], batch size: 119, lr: 4.11e-02, grad_scale: 4.0
+2023-02-28 11:51:20,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1661, 2.0626, 3.8150, 1.6366], device='cuda:0'), covar=tensor([0.0652, 0.1200, 0.0593, 0.1618], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0423, 0.0582, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 11:51:54,002 INFO [train.py:968] (0/2) Epoch 1, batch 5500, giga_loss[loss=0.4074, simple_loss=0.4411, pruned_loss=0.1868, over 28947.00 frames. ], tot_loss[loss=0.4666, simple_loss=0.4765, pruned_loss=0.2283, over 5733356.94 frames. ], libri_tot_loss[loss=0.5983, simple_loss=0.5671, pruned_loss=0.3309, over 5364483.81 frames. ], giga_tot_loss[loss=0.4589, simple_loss=0.4712, pruned_loss=0.2233, over 5723275.39 frames. ], batch size: 174, lr: 4.10e-02, grad_scale: 4.0
+2023-02-28 11:52:02,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.364e+02 1.386e+03 1.760e+03 2.224e+03 3.493e+03, threshold=3.519e+03, percent-clipped=0.0
+2023-02-28 11:52:33,354 INFO [train.py:968] (0/2) Epoch 1, batch 5550, giga_loss[loss=0.4495, simple_loss=0.4761, pruned_loss=0.2115, over 28855.00 frames. ], tot_loss[loss=0.4681, simple_loss=0.4766, pruned_loss=0.2298, over 5735951.95 frames. ], libri_tot_loss[loss=0.5949, simple_loss=0.5648, pruned_loss=0.3279, over 5385906.99 frames. ], giga_tot_loss[loss=0.4593, simple_loss=0.4705, pruned_loss=0.2241, over 5723070.25 frames. ], batch size: 285, lr: 4.09e-02, grad_scale: 4.0
+2023-02-28 11:53:16,128 INFO [train.py:968] (0/2) Epoch 1, batch 5600, giga_loss[loss=0.5008, simple_loss=0.496, pruned_loss=0.2529, over 27685.00 frames. ], tot_loss[loss=0.4652, simple_loss=0.4743, pruned_loss=0.2281, over 5726564.63 frames. ], libri_tot_loss[loss=0.5926, simple_loss=0.5633, pruned_loss=0.3259, over 5398800.08 frames. ], giga_tot_loss[loss=0.4572, simple_loss=0.4686, pruned_loss=0.2229, over 5712356.45 frames. ], batch size: 472, lr: 4.08e-02, grad_scale: 8.0
+2023-02-28 11:53:23,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.193e+02 1.471e+03 2.065e+03 2.787e+03 6.142e+03, threshold=4.130e+03, percent-clipped=15.0
+2023-02-28 11:53:56,154 INFO [train.py:968] (0/2) Epoch 1, batch 5650, giga_loss[loss=0.4224, simple_loss=0.4474, pruned_loss=0.1987, over 28602.00 frames. ], tot_loss[loss=0.4577, simple_loss=0.4683, pruned_loss=0.2235, over 5721443.87 frames. ], libri_tot_loss[loss=0.5894, simple_loss=0.5611, pruned_loss=0.3232, over 5412135.26 frames. ], giga_tot_loss[loss=0.4498, simple_loss=0.4627, pruned_loss=0.2185, over 5708333.53 frames. ], batch size: 336, lr: 4.07e-02, grad_scale: 8.0
+2023-02-28 11:54:36,580 INFO [train.py:968] (0/2) Epoch 1, batch 5700, libri_loss[loss=0.4792, simple_loss=0.4809, pruned_loss=0.2387, over 29655.00 frames. ], tot_loss[loss=0.4505, simple_loss=0.4622, pruned_loss=0.2194, over 5723498.28 frames. ], libri_tot_loss[loss=0.5878, simple_loss=0.56, pruned_loss=0.3218, over 5422040.28 frames. ], giga_tot_loss[loss=0.4431, simple_loss=0.4568, pruned_loss=0.2147, over 5709677.86 frames. ], batch size: 69, lr: 4.06e-02, grad_scale: 8.0
+2023-02-28 11:54:45,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.088e+03 1.725e+03 2.072e+03 2.605e+03 5.482e+03, threshold=4.143e+03, percent-clipped=5.0
+2023-02-28 11:54:46,892 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 11:54:53,193 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0243, 2.4857, 2.3583, 2.0291], device='cuda:0'), covar=tensor([0.1654, 0.0941, 0.1002, 0.2278], device='cuda:0'), in_proj_covar=tensor([0.0555, 0.0572, 0.0543, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0008], device='cuda:0')
+2023-02-28 11:55:13,222 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-02-28 11:55:16,892 INFO [train.py:968] (0/2) Epoch 1, batch 5750, giga_loss[loss=0.3877, simple_loss=0.4041, pruned_loss=0.1857, over 28593.00 frames. ], tot_loss[loss=0.4483, simple_loss=0.4603, pruned_loss=0.2181, over 5716704.11 frames. ], libri_tot_loss[loss=0.5858, simple_loss=0.5586, pruned_loss=0.3202, over 5424086.79 frames. ], giga_tot_loss[loss=0.4404, simple_loss=0.4546, pruned_loss=0.2131, over 5711552.91 frames. ], batch size: 85, lr: 4.05e-02, grad_scale: 8.0
+2023-02-28 11:55:53,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8399, 0.7311, 1.0982, 0.2334], device='cuda:0'), covar=tensor([0.0764, 0.0495, 0.0514, 0.1152], device='cuda:0'), in_proj_covar=tensor([0.0626, 0.0454, 0.0498, 0.0599], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 11:55:54,346 INFO [train.py:968] (0/2) Epoch 1, batch 5800, giga_loss[loss=0.4242, simple_loss=0.4572, pruned_loss=0.1956, over 28828.00 frames. ], tot_loss[loss=0.4493, simple_loss=0.4621, pruned_loss=0.2183, over 5716728.87 frames. ], libri_tot_loss[loss=0.5832, simple_loss=0.5569, pruned_loss=0.318, over 5438219.68 frames. ], giga_tot_loss[loss=0.4415, simple_loss=0.4563, pruned_loss=0.2133, over 5707822.51 frames. ], batch size: 112, lr: 4.04e-02, grad_scale: 8.0
+2023-02-28 11:56:01,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.442e+02 1.662e+03 2.034e+03 2.541e+03 4.357e+03, threshold=4.068e+03, percent-clipped=2.0
+2023-02-28 11:56:30,939 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-02-28 11:56:36,513 INFO [train.py:968] (0/2) Epoch 1, batch 5850, giga_loss[loss=0.5363, simple_loss=0.5244, pruned_loss=0.2741, over 27606.00 frames. ], tot_loss[loss=0.4551, simple_loss=0.4674, pruned_loss=0.2214, over 5712036.24 frames. ], libri_tot_loss[loss=0.5816, simple_loss=0.5558, pruned_loss=0.3166, over 5445928.91 frames. ], giga_tot_loss[loss=0.448, simple_loss=0.4621, pruned_loss=0.2169, over 5702933.09 frames. ], batch size: 472, lr: 4.03e-02, grad_scale: 8.0
+2023-02-28 11:57:18,955 INFO [train.py:968] (0/2) Epoch 1, batch 5900, giga_loss[loss=0.4634, simple_loss=0.4845, pruned_loss=0.2212, over 28688.00 frames. ], tot_loss[loss=0.4592, simple_loss=0.4721, pruned_loss=0.2232, over 5719907.39 frames. ], libri_tot_loss[loss=0.5808, simple_loss=0.5554, pruned_loss=0.3158, over 5452472.43 frames. ], giga_tot_loss[loss=0.4526, simple_loss=0.4671, pruned_loss=0.2191, over 5710526.55 frames. ], batch size: 242, lr: 4.02e-02, grad_scale: 8.0
+2023-02-28 11:57:27,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.970e+02 1.440e+03 1.932e+03 2.293e+03 4.240e+03, threshold=3.865e+03, percent-clipped=1.0
+2023-02-28 11:57:29,026 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-02-28 11:57:57,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1691, 1.0368, 1.5008, 0.5035], device='cuda:0'), covar=tensor([0.0766, 0.0506, 0.0424, 0.1109], device='cuda:0'), in_proj_covar=tensor([0.0609, 0.0445, 0.0480, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 11:57:59,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3273, 2.3280, 4.7787, 1.8233], device='cuda:0'), covar=tensor([0.0360, 0.1061, 0.0456, 0.1530], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0434, 0.0606, 0.0466], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 11:58:01,370 INFO [train.py:968] (0/2) Epoch 1, batch 5950, libri_loss[loss=0.4804, simple_loss=0.4722, pruned_loss=0.2443, over 29446.00 frames. ], tot_loss[loss=0.4645, simple_loss=0.4762, pruned_loss=0.2264, over 5718929.01 frames. ], libri_tot_loss[loss=0.5767, simple_loss=0.5526, pruned_loss=0.3126, over 5470418.16 frames. ], giga_tot_loss[loss=0.4582, simple_loss=0.4716, pruned_loss=0.2224, over 5704502.32 frames. ], batch size: 70, lr: 4.01e-02, grad_scale: 8.0
+2023-02-28 11:58:13,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4053, 1.3463, 0.8726, 1.4992], device='cuda:0'), covar=tensor([0.1061, 0.1718, 0.1372, 0.1202], device='cuda:0'), in_proj_covar=tensor([0.0569, 0.0714, 0.0676, 0.0572], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 11:58:43,586 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-6000.pt
+2023-02-28 11:58:43,884 INFO [train.py:968] (0/2) Epoch 1, batch 6000, giga_loss[loss=0.4421, simple_loss=0.4582, pruned_loss=0.213, over 28744.00 frames. ], tot_loss[loss=0.4692, simple_loss=0.4802, pruned_loss=0.2291, over 5716861.31 frames. ], libri_tot_loss[loss=0.5751, simple_loss=0.5516, pruned_loss=0.3111, over 5477197.92 frames. ], giga_tot_loss[loss=0.4632, simple_loss=0.4757, pruned_loss=0.2253, over 5704930.44 frames. ], batch size: 92, lr: 4.00e-02, grad_scale: 8.0
+2023-02-28 11:58:43,890 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 11:58:51,740 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.4087, simple_loss=0.4648, pruned_loss=0.1763, over 944034.00 frames. 
+2023-02-28 11:58:51,741 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 18855MB
+2023-02-28 11:58:58,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.104e+02 1.839e+03 2.225e+03 2.757e+03 5.207e+03, threshold=4.449e+03, percent-clipped=6.0
+2023-02-28 11:59:29,417 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6043.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:59:34,523 INFO [train.py:968] (0/2) Epoch 1, batch 6050, giga_loss[loss=0.4978, simple_loss=0.4856, pruned_loss=0.2551, over 28262.00 frames. ], tot_loss[loss=0.4825, simple_loss=0.4883, pruned_loss=0.2383, over 5707957.01 frames. ], libri_tot_loss[loss=0.5745, simple_loss=0.5513, pruned_loss=0.3105, over 5484383.99 frames. ], giga_tot_loss[loss=0.4764, simple_loss=0.4838, pruned_loss=0.2345, over 5696358.55 frames. ], batch size: 77, lr: 3.99e-02, grad_scale: 8.0
+2023-02-28 12:00:19,899 INFO [train.py:968] (0/2) Epoch 1, batch 6100, giga_loss[loss=0.5106, simple_loss=0.5132, pruned_loss=0.254, over 28710.00 frames. ], tot_loss[loss=0.4953, simple_loss=0.4957, pruned_loss=0.2475, over 5704087.04 frames. ], libri_tot_loss[loss=0.5701, simple_loss=0.5478, pruned_loss=0.3072, over 5499302.96 frames. ], giga_tot_loss[loss=0.4908, simple_loss=0.4927, pruned_loss=0.2445, over 5689896.45 frames. ], batch size: 242, lr: 3.98e-02, grad_scale: 4.0
+2023-02-28 12:00:31,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.380e+03 2.258e+03 2.754e+03 3.685e+03 1.059e+04, threshold=5.508e+03, percent-clipped=15.0
+2023-02-28 12:01:07,648 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6148.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:01:08,849 INFO [train.py:968] (0/2) Epoch 1, batch 6150, giga_loss[loss=0.4547, simple_loss=0.4788, pruned_loss=0.2153, over 28894.00 frames. ], tot_loss[loss=0.5045, simple_loss=0.5027, pruned_loss=0.2532, over 5705953.21 frames. ], libri_tot_loss[loss=0.5691, simple_loss=0.5472, pruned_loss=0.3063, over 5505594.31 frames. ], giga_tot_loss[loss=0.5006, simple_loss=0.5, pruned_loss=0.2506, over 5692686.81 frames. ], batch size: 112, lr: 3.97e-02, grad_scale: 4.0
+2023-02-28 12:01:16,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7486, 1.6555, 1.8790, 1.3251], device='cuda:0'), covar=tensor([0.1663, 0.1171, 0.0979, 0.2755], device='cuda:0'), in_proj_covar=tensor([0.0574, 0.0563, 0.0543, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0009], device='cuda:0')
+2023-02-28 12:01:54,181 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-02-28 12:01:57,834 INFO [train.py:968] (0/2) Epoch 1, batch 6200, libri_loss[loss=0.4774, simple_loss=0.4702, pruned_loss=0.2423, over 29486.00 frames. ], tot_loss[loss=0.5132, simple_loss=0.5082, pruned_loss=0.2591, over 5694566.01 frames. ], libri_tot_loss[loss=0.5675, simple_loss=0.5462, pruned_loss=0.3048, over 5503550.02 frames. ], giga_tot_loss[loss=0.5097, simple_loss=0.5056, pruned_loss=0.2569, over 5691528.95 frames. ], batch size: 70, lr: 3.96e-02, grad_scale: 4.0
+2023-02-28 12:02:02,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9320, 2.2790, 1.6665, 1.7102], device='cuda:0'), covar=tensor([0.0580, 0.1285, 0.0659, 0.0515], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0559, 0.0391, 0.0366], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 12:02:06,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 2.168e+03 2.742e+03 3.618e+03 6.021e+03, threshold=5.484e+03, percent-clipped=3.0
+2023-02-28 12:02:41,409 INFO [train.py:968] (0/2) Epoch 1, batch 6250, giga_loss[loss=0.57, simple_loss=0.5518, pruned_loss=0.2941, over 28842.00 frames. ], tot_loss[loss=0.5255, simple_loss=0.516, pruned_loss=0.2675, over 5694425.77 frames. ], libri_tot_loss[loss=0.5651, simple_loss=0.545, pruned_loss=0.3028, over 5513311.75 frames. ], giga_tot_loss[loss=0.5235, simple_loss=0.5141, pruned_loss=0.2664, over 5688267.59 frames. ], batch size: 119, lr: 3.95e-02, grad_scale: 4.0
+2023-02-28 12:02:42,372 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6251.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:02:48,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8648, 1.9145, 3.1566, 1.7363], device='cuda:0'), covar=tensor([0.1205, 0.0910, 0.0326, 0.0997], device='cuda:0'), in_proj_covar=tensor([0.0520, 0.0422, 0.0509, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 12:03:28,567 INFO [train.py:968] (0/2) Epoch 1, batch 6300, giga_loss[loss=0.6364, simple_loss=0.5717, pruned_loss=0.3506, over 27600.00 frames. ], tot_loss[loss=0.5336, simple_loss=0.5213, pruned_loss=0.2729, over 5689504.10 frames. ], libri_tot_loss[loss=0.5632, simple_loss=0.5439, pruned_loss=0.3011, over 5522896.83 frames. ], giga_tot_loss[loss=0.5328, simple_loss=0.5202, pruned_loss=0.2727, over 5679856.56 frames. ], batch size: 473, lr: 3.94e-02, grad_scale: 4.0
+2023-02-28 12:03:39,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.349e+03 2.079e+03 2.522e+03 3.180e+03 7.632e+03, threshold=5.043e+03, percent-clipped=1.0
+2023-02-28 12:04:21,669 INFO [train.py:968] (0/2) Epoch 1, batch 6350, giga_loss[loss=0.4904, simple_loss=0.4991, pruned_loss=0.2408, over 28994.00 frames. ], tot_loss[loss=0.5347, simple_loss=0.5216, pruned_loss=0.2739, over 5675049.99 frames. ], libri_tot_loss[loss=0.5624, simple_loss=0.5435, pruned_loss=0.3004, over 5527117.08 frames. ], giga_tot_loss[loss=0.5344, simple_loss=0.5207, pruned_loss=0.274, over 5665612.70 frames. ], batch size: 213, lr: 3.93e-02, grad_scale: 4.0
+2023-02-28 12:04:35,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9282, 2.0785, 1.9257, 1.7201], device='cuda:0'), covar=tensor([0.0560, 0.0997, 0.0492, 0.0448], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0570, 0.0408, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 12:04:57,424 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6383.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:05:16,474 INFO [train.py:968] (0/2) Epoch 1, batch 6400, giga_loss[loss=0.652, simple_loss=0.5638, pruned_loss=0.37, over 23619.00 frames. ], tot_loss[loss=0.5403, simple_loss=0.5246, pruned_loss=0.278, over 5670356.30 frames. ], libri_tot_loss[loss=0.5614, simple_loss=0.5429, pruned_loss=0.2995, over 5532910.89 frames. ], giga_tot_loss[loss=0.5406, simple_loss=0.5241, pruned_loss=0.2785, over 5659956.92 frames. ], batch size: 705, lr: 3.92e-02, grad_scale: 8.0
+2023-02-28 12:05:18,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1485, 0.9464, 1.2175, 0.3752], device='cuda:0'), covar=tensor([0.0473, 0.0407, 0.0387, 0.0682], device='cuda:0'), in_proj_covar=tensor([0.0615, 0.0453, 0.0507, 0.0598], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 12:05:27,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.195e+03 1.808e+03 2.358e+03 3.192e+03 5.398e+03, threshold=4.716e+03, percent-clipped=1.0
+2023-02-28 12:05:34,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6418.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:05:55,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6436.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:06:00,263 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-02-28 12:06:09,329 INFO [train.py:968] (0/2) Epoch 1, batch 6450, giga_loss[loss=0.5255, simple_loss=0.5173, pruned_loss=0.2668, over 28555.00 frames. ], tot_loss[loss=0.5469, simple_loss=0.528, pruned_loss=0.2829, over 5661379.56 frames. ], libri_tot_loss[loss=0.5604, simple_loss=0.5424, pruned_loss=0.2986, over 5534750.02 frames. ], giga_tot_loss[loss=0.5477, simple_loss=0.5279, pruned_loss=0.2838, over 5653535.49 frames. ], batch size: 60, lr: 3.91e-02, grad_scale: 8.0
+2023-02-28 12:06:40,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6111, 0.8713, 2.0251, 0.1135], device='cuda:0'), covar=tensor([0.0767, 0.0781, 0.0303, 0.2241], device='cuda:0'), in_proj_covar=tensor([0.0549, 0.0439, 0.0391, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004], device='cuda:0')
+2023-02-28 12:07:01,416 INFO [train.py:968] (0/2) Epoch 1, batch 6500, giga_loss[loss=0.5504, simple_loss=0.5213, pruned_loss=0.2897, over 28632.00 frames. ], tot_loss[loss=0.5524, simple_loss=0.5314, pruned_loss=0.2867, over 5644111.95 frames. ], libri_tot_loss[loss=0.5589, simple_loss=0.5414, pruned_loss=0.2974, over 5531837.29 frames. ], giga_tot_loss[loss=0.5542, simple_loss=0.532, pruned_loss=0.2883, over 5641855.21 frames. ], batch size: 307, lr: 3.90e-02, grad_scale: 8.0
+2023-02-28 12:07:13,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.947e+03 2.366e+03 3.174e+03 5.633e+03, threshold=4.733e+03, percent-clipped=4.0
+2023-02-28 12:07:27,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6523.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:07:32,758 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6528.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:07:37,951 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-02-28 12:07:53,985 INFO [train.py:968] (0/2) Epoch 1, batch 6550, giga_loss[loss=0.687, simple_loss=0.5945, pruned_loss=0.3897, over 28008.00 frames. ], tot_loss[loss=0.5548, simple_loss=0.5323, pruned_loss=0.2887, over 5642815.39 frames. ], libri_tot_loss[loss=0.5581, simple_loss=0.5411, pruned_loss=0.2966, over 5537353.05 frames. ], giga_tot_loss[loss=0.5569, simple_loss=0.5327, pruned_loss=0.2905, over 5637865.61 frames. ], batch size: 412, lr: 3.89e-02, grad_scale: 4.0
+2023-02-28 12:08:05,503 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6561.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:08:07,026 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6563.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:08:08,002 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6564.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:08:37,699 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6593.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:08:39,596 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6596.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:08:42,734 INFO [train.py:968] (0/2) Epoch 1, batch 6600, giga_loss[loss=0.6184, simple_loss=0.5471, pruned_loss=0.3449, over 23731.00 frames. ], tot_loss[loss=0.5529, simple_loss=0.5306, pruned_loss=0.2876, over 5644568.63 frames. ], libri_tot_loss[loss=0.5574, simple_loss=0.5407, pruned_loss=0.2959, over 5543597.53 frames. ], giga_tot_loss[loss=0.5552, simple_loss=0.5312, pruned_loss=0.2896, over 5636710.66 frames. ], batch size: 705, lr: 3.89e-02, grad_scale: 4.0
+2023-02-28 12:08:46,848 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6604.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:08:56,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 2.203e+03 2.545e+03 3.170e+03 1.090e+04, threshold=5.090e+03, percent-clipped=6.0
+2023-02-28 12:09:09,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6626.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:09:09,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2523, 1.2546, 0.8527, 1.4270], device='cuda:0'), covar=tensor([0.1181, 0.1819, 0.1503, 0.1190], device='cuda:0'), in_proj_covar=tensor([0.0616, 0.0772, 0.0731, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 12:09:31,592 INFO [train.py:968] (0/2) Epoch 1, batch 6650, giga_loss[loss=0.5167, simple_loss=0.521, pruned_loss=0.2562, over 28914.00 frames. ], tot_loss[loss=0.5502, simple_loss=0.5296, pruned_loss=0.2853, over 5639674.57 frames. ], libri_tot_loss[loss=0.5554, simple_loss=0.5397, pruned_loss=0.2941, over 5552742.72 frames. ], giga_tot_loss[loss=0.5536, simple_loss=0.5307, pruned_loss=0.2882, over 5627964.08 frames. ], batch size: 174, lr: 3.88e-02, grad_scale: 4.0
+2023-02-28 12:09:41,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6659.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:09:46,638 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6666.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:09:49,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6669.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:10:06,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2341, 1.3563, 1.6685, 1.1669], device='cuda:0'), covar=tensor([0.1712, 0.1506, 0.1304, 0.2973], device='cuda:0'), in_proj_covar=tensor([0.0528, 0.0627, 0.0517, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 12:10:13,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4061, 1.6009, 1.5467, 1.1921], device='cuda:0'), covar=tensor([0.2304, 0.1341, 0.1248, 0.3405], device='cuda:0'), in_proj_covar=tensor([0.0593, 0.0564, 0.0558, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0009], device='cuda:0')
+2023-02-28 12:10:20,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6698.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:10:22,848 INFO [train.py:968] (0/2) Epoch 1, batch 6700, giga_loss[loss=0.606, simple_loss=0.5614, pruned_loss=0.3252, over 27878.00 frames. ], tot_loss[loss=0.5469, simple_loss=0.5288, pruned_loss=0.2826, over 5649176.39 frames. ], libri_tot_loss[loss=0.5547, simple_loss=0.5392, pruned_loss=0.2935, over 5556418.81 frames. ], giga_tot_loss[loss=0.5502, simple_loss=0.5299, pruned_loss=0.2852, over 5637581.24 frames. ], batch size: 412, lr: 3.87e-02, grad_scale: 4.0
+2023-02-28 12:10:31,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.207e+03 1.780e+03 2.217e+03 3.484e+03 1.468e+04, threshold=4.434e+03, percent-clipped=4.0
+2023-02-28 12:10:35,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0471, 2.4798, 2.3601, 1.9559], device='cuda:0'), covar=tensor([0.0663, 0.0907, 0.0418, 0.0415], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0616, 0.0440, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002], device='cuda:0')
+2023-02-28 12:11:10,056 INFO [train.py:968] (0/2) Epoch 1, batch 6750, giga_loss[loss=0.512, simple_loss=0.516, pruned_loss=0.254, over 29067.00 frames. ], tot_loss[loss=0.5471, simple_loss=0.5295, pruned_loss=0.2823, over 5642768.00 frames. ], libri_tot_loss[loss=0.5528, simple_loss=0.5381, pruned_loss=0.2918, over 5568309.72 frames. ], giga_tot_loss[loss=0.5511, simple_loss=0.531, pruned_loss=0.2856, over 5626093.66 frames. ], batch size: 155, lr: 3.86e-02, grad_scale: 4.0
+2023-02-28 12:11:18,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6758.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:11:28,927 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6769.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:11:32,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6772.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:11:58,011 INFO [train.py:968] (0/2) Epoch 1, batch 6800, libri_loss[loss=0.42, simple_loss=0.4394, pruned_loss=0.2003, over 29638.00 frames. ], tot_loss[loss=0.5399, simple_loss=0.5251, pruned_loss=0.2774, over 5641304.37 frames. ], libri_tot_loss[loss=0.5509, simple_loss=0.537, pruned_loss=0.2902, over 5575542.93 frames. ], giga_tot_loss[loss=0.5446, simple_loss=0.5271, pruned_loss=0.2811, over 5623142.63 frames. ], batch size: 69, lr: 3.85e-02, grad_scale: 8.0
+2023-02-28 12:12:00,031 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6801.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:12:12,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6811.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:12:12,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.855e+03 2.338e+03 2.950e+03 7.898e+03, threshold=4.675e+03, percent-clipped=7.0
+2023-02-28 12:12:50,805 INFO [train.py:968] (0/2) Epoch 1, batch 6850, giga_loss[loss=0.4457, simple_loss=0.4727, pruned_loss=0.2094, over 28895.00 frames. ], tot_loss[loss=0.5338, simple_loss=0.5221, pruned_loss=0.2728, over 5650612.75 frames. ], libri_tot_loss[loss=0.5491, simple_loss=0.5358, pruned_loss=0.2888, over 5582512.79 frames. ], giga_tot_loss[loss=0.5389, simple_loss=0.5244, pruned_loss=0.2767, over 5631140.90 frames. ], batch size: 112, lr: 3.84e-02, grad_scale: 8.0
+2023-02-28 12:12:55,958 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6855.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:12:59,754 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6860.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:13:38,029 INFO [train.py:968] (0/2) Epoch 1, batch 6900, giga_loss[loss=0.5168, simple_loss=0.5124, pruned_loss=0.2607, over 28925.00 frames. ], tot_loss[loss=0.5265, simple_loss=0.5177, pruned_loss=0.2676, over 5657602.62 frames. ], libri_tot_loss[loss=0.5467, simple_loss=0.5341, pruned_loss=0.2872, over 5587969.59 frames. ], giga_tot_loss[loss=0.5322, simple_loss=0.5208, pruned_loss=0.2718, over 5638460.37 frames. ], batch size: 186, lr: 3.83e-02, grad_scale: 8.0
+2023-02-28 12:13:39,935 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6901.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:13:41,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6903.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:13:44,814 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6904.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:13:52,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.298e+03 1.879e+03 2.305e+03 3.141e+03 7.033e+03, threshold=4.609e+03, percent-clipped=8.0
+2023-02-28 12:13:58,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-02-28 12:14:10,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6933.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:14:14,005 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3710, 1.2950, 1.1046, 1.6615], device='cuda:0'), covar=tensor([0.1015, 0.1375, 0.1078, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0631, 0.0783, 0.0734, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 12:14:15,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6938.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:14:28,903 INFO [train.py:968] (0/2) Epoch 1, batch 6950, giga_loss[loss=0.5358, simple_loss=0.5013, pruned_loss=0.2851, over 23779.00 frames. ], tot_loss[loss=0.5218, simple_loss=0.5146, pruned_loss=0.2645, over 5661846.25 frames. ], libri_tot_loss[loss=0.5449, simple_loss=0.533, pruned_loss=0.2856, over 5596226.07 frames. ], giga_tot_loss[loss=0.5275, simple_loss=0.5176, pruned_loss=0.2687, over 5640957.50 frames. ], batch size: 705, lr: 3.82e-02, grad_scale: 4.0
+2023-02-28 12:14:34,075 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6954.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:14:36,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6957.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:14:50,610 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6971.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:14:56,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6979.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:15:01,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6986.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:15:14,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8537, 2.0744, 3.2615, 1.6510], device='cuda:0'), covar=tensor([0.1454, 0.0947, 0.0334, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0524, 0.0425, 0.0503, 0.0375], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 12:15:17,948 INFO [train.py:968] (0/2) Epoch 1, batch 7000, giga_loss[loss=0.5124, simple_loss=0.5149, pruned_loss=0.255, over 28863.00 frames. ], tot_loss[loss=0.5181, simple_loss=0.5118, pruned_loss=0.2622, over 5663106.02 frames. ], libri_tot_loss[loss=0.5433, simple_loss=0.5319, pruned_loss=0.2844, over 5603768.51 frames. ], giga_tot_loss[loss=0.5235, simple_loss=0.5147, pruned_loss=0.2661, over 5641324.32 frames. ], batch size: 199, lr: 3.81e-02, grad_scale: 4.0
+2023-02-28 12:15:28,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.339e+03 2.237e+03 2.831e+03 3.603e+03 7.232e+03, threshold=5.662e+03, percent-clipped=6.0
+2023-02-28 12:15:39,177 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7022.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:15:51,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7034.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:02,350 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7046.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:04,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7049.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:04,954 INFO [train.py:968] (0/2) Epoch 1, batch 7050, giga_loss[loss=0.5569, simple_loss=0.5364, pruned_loss=0.2887, over 29071.00 frames. ], tot_loss[loss=0.5134, simple_loss=0.509, pruned_loss=0.2589, over 5652644.53 frames. ], libri_tot_loss[loss=0.5422, simple_loss=0.5313, pruned_loss=0.2834, over 5600061.42 frames. ], giga_tot_loss[loss=0.5181, simple_loss=0.5114, pruned_loss=0.2624, over 5639228.79 frames. ], batch size: 128, lr: 3.80e-02, grad_scale: 4.0
+2023-02-28 12:16:41,199 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7078.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:44,358 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7081.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:46,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7084.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:57,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8352, 2.0587, 3.3744, 1.8287], device='cuda:0'), covar=tensor([0.1436, 0.0919, 0.0271, 0.1036], device='cuda:0'), in_proj_covar=tensor([0.0517, 0.0424, 0.0498, 0.0374], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 12:17:02,600 INFO [train.py:968] (0/2) Epoch 1, batch 7100, giga_loss[loss=0.4352, simple_loss=0.4734, pruned_loss=0.1985, over 29016.00 frames. ], tot_loss[loss=0.513, simple_loss=0.5089, pruned_loss=0.2585, over 5654550.80 frames. ], libri_tot_loss[loss=0.5423, simple_loss=0.5314, pruned_loss=0.2835, over 5602746.06 frames. ], giga_tot_loss[loss=0.5162, simple_loss=0.5105, pruned_loss=0.261, over 5642118.33 frames. ], batch size: 128, lr: 3.79e-02, grad_scale: 4.0
+2023-02-28 12:17:09,521 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-02-28 12:17:19,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.640e+02 1.767e+03 2.136e+03 2.564e+03 4.468e+03, threshold=4.273e+03, percent-clipped=0.0
+2023-02-28 12:17:19,294 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7113.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:21,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7114.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:23,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7117.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:27,728 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7122.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:31,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7125.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:51,257 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7146.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:54,192 INFO [train.py:968] (0/2) Epoch 1, batch 7150, giga_loss[loss=0.4854, simple_loss=0.4974, pruned_loss=0.2367, over 28762.00 frames. ], tot_loss[loss=0.5025, simple_loss=0.5028, pruned_loss=0.2511, over 5660980.24 frames. ], libri_tot_loss[loss=0.539, simple_loss=0.5292, pruned_loss=0.281, over 5613779.55 frames. ], giga_tot_loss[loss=0.5068, simple_loss=0.5051, pruned_loss=0.2543, over 5642752.62 frames. ], batch size: 99, lr: 3.79e-02, grad_scale: 4.0
+2023-02-28 12:17:58,517 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7154.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:18:25,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7177.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:18:28,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7180.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:18:50,278 INFO [train.py:968] (0/2) Epoch 1, batch 7200, giga_loss[loss=0.5033, simple_loss=0.5202, pruned_loss=0.2431, over 28863.00 frames. ], tot_loss[loss=0.4999, simple_loss=0.5032, pruned_loss=0.2483, over 5665444.39 frames. ], libri_tot_loss[loss=0.5374, simple_loss=0.5281, pruned_loss=0.2798, over 5619791.15 frames. ], giga_tot_loss[loss=0.504, simple_loss=0.5055, pruned_loss=0.2513, over 5646435.89 frames. ], batch size: 186, lr: 3.78e-02, grad_scale: 8.0
+2023-02-28 12:19:03,428 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7209.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:19:05,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.200e+03 1.676e+03 2.199e+03 2.998e+03 8.798e+03, threshold=4.398e+03, percent-clipped=6.0
+2023-02-28 12:19:19,442 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7230.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:19:24,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7235.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:19:38,393 INFO [train.py:968] (0/2) Epoch 1, batch 7250, giga_loss[loss=0.5287, simple_loss=0.5247, pruned_loss=0.2664, over 29063.00 frames. ], tot_loss[loss=0.5041, simple_loss=0.507, pruned_loss=0.2506, over 5677234.54 frames. ], libri_tot_loss[loss=0.5364, simple_loss=0.5274, pruned_loss=0.2789, over 5619996.74 frames. ], giga_tot_loss[loss=0.5073, simple_loss=0.5088, pruned_loss=0.2529, over 5662886.14 frames. ], batch size: 155, lr: 3.77e-02, grad_scale: 4.0
+2023-02-28 12:19:53,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4318, 1.3536, 1.3001, 0.6285], device='cuda:0'), covar=tensor([0.0670, 0.0459, 0.0541, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0618, 0.0456, 0.0528, 0.0584], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 12:20:01,994 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-02-28 12:20:03,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7271.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:20:30,932 INFO [train.py:968] (0/2) Epoch 1, batch 7300, giga_loss[loss=0.5483, simple_loss=0.5324, pruned_loss=0.2821, over 28017.00 frames. ], tot_loss[loss=0.5104, simple_loss=0.5103, pruned_loss=0.2552, over 5666430.53 frames. ], libri_tot_loss[loss=0.5364, simple_loss=0.5275, pruned_loss=0.2788, over 5622122.82 frames. ], giga_tot_loss[loss=0.5125, simple_loss=0.5114, pruned_loss=0.2568, over 5653740.97 frames. ], batch size: 412, lr: 3.76e-02, grad_scale: 4.0
+2023-02-28 12:20:32,818 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7302.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:20:42,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.297e+03 2.213e+03 2.750e+03 3.892e+03 9.794e+03, threshold=5.501e+03, percent-clipped=18.0
+2023-02-28 12:21:15,320 INFO [train.py:968] (0/2) Epoch 1, batch 7350, giga_loss[loss=0.4755, simple_loss=0.4833, pruned_loss=0.2339, over 28753.00 frames. ], tot_loss[loss=0.5088, simple_loss=0.5091, pruned_loss=0.2543, over 5679919.73 frames. ], libri_tot_loss[loss=0.533, simple_loss=0.5255, pruned_loss=0.276, over 5636521.47 frames. ], giga_tot_loss[loss=0.5124, simple_loss=0.511, pruned_loss=0.2569, over 5658587.35 frames. ], batch size: 243, lr: 3.75e-02, grad_scale: 4.0
+2023-02-28 12:21:36,450 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7373.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:21:41,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7376.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:21:45,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7378.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:21:47,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7381.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:22:00,919 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7397.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:22:03,947 INFO [train.py:968] (0/2) Epoch 1, batch 7400, giga_loss[loss=0.4588, simple_loss=0.4798, pruned_loss=0.2189, over 28649.00 frames. ], tot_loss[loss=0.5062, simple_loss=0.5062, pruned_loss=0.2531, over 5683668.59 frames. ], libri_tot_loss[loss=0.5308, simple_loss=0.524, pruned_loss=0.2744, over 5640684.08 frames. ], giga_tot_loss[loss=0.5103, simple_loss=0.5085, pruned_loss=0.2561, over 5664235.98 frames. ], batch size: 92, lr: 3.74e-02, grad_scale: 4.0
+2023-02-28 12:22:09,686 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7405.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:22:10,292 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6135, 1.5657, 1.1277, 1.2637], device='cuda:0'), covar=tensor([0.0939, 0.0877, 0.1326, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0550, 0.0680, 0.0661, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 12:22:13,946 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7410.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:22:18,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.735e+03 2.229e+03 3.022e+03 5.084e+03, threshold=4.458e+03, percent-clipped=0.0
+2023-02-28 12:22:25,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6421, 1.5302, 1.4832, 1.4890], device='cuda:0'), covar=tensor([0.0394, 0.0756, 0.0397, 0.0299], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0635, 0.0461, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0005, 0.0003, 0.0002], device='cuda:0')
+2023-02-28 12:22:26,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9714, 2.5110, 4.5148, 2.5769], device='cuda:0'), covar=tensor([0.1511, 0.0971, 0.0222, 0.0644], device='cuda:0'), in_proj_covar=tensor([0.0523, 0.0431, 0.0510, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 12:22:42,085 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7441.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:22:48,725 INFO [train.py:968] (0/2) Epoch 1, batch 7450, giga_loss[loss=0.4715, simple_loss=0.4755, pruned_loss=0.2337, over 28345.00 frames. ], tot_loss[loss=0.5071, simple_loss=0.5056, pruned_loss=0.2543, over 5672297.42 frames. ], libri_tot_loss[loss=0.5299, simple_loss=0.5235, pruned_loss=0.2735, over 5638745.90 frames. ], giga_tot_loss[loss=0.5105, simple_loss=0.5073, pruned_loss=0.2569, over 5659013.88 frames. ], batch size: 60, lr: 3.73e-02, grad_scale: 4.0
+2023-02-28 12:23:40,181 INFO [train.py:968] (0/2) Epoch 1, batch 7500, giga_loss[loss=0.5688, simple_loss=0.5265, pruned_loss=0.3056, over 26660.00 frames. ], tot_loss[loss=0.5049, simple_loss=0.5046, pruned_loss=0.2526, over 5672886.44 frames. ], libri_tot_loss[loss=0.5281, simple_loss=0.5225, pruned_loss=0.2721, over 5645280.49 frames. ], giga_tot_loss[loss=0.5086, simple_loss=0.5065, pruned_loss=0.2554, over 5657209.34 frames. ], batch size: 555, lr: 3.72e-02, grad_scale: 4.0
+2023-02-28 12:23:46,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4272, 2.5602, 4.2255, 1.7672], device='cuda:0'), covar=tensor([0.0475, 0.1196, 0.0911, 0.1940], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0432, 0.0638, 0.0486], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 12:23:51,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.159e+03 1.910e+03 2.495e+03 3.316e+03 7.663e+03, threshold=4.989e+03, percent-clipped=11.0
+2023-02-28 12:24:01,020 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-02-28 12:24:15,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8599, 2.3327, 2.0622, 1.6985], device='cuda:0'), covar=tensor([0.0639, 0.0852, 0.0472, 0.0438], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0651, 0.0475, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0005, 0.0003, 0.0002], device='cuda:0')
+2023-02-28 12:24:17,683 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7540.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:24:19,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7543.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:24:25,994 INFO [train.py:968] (0/2) Epoch 1, batch 7550, giga_loss[loss=0.4979, simple_loss=0.4983, pruned_loss=0.2488, over 28642.00 frames. ], tot_loss[loss=0.4998, simple_loss=0.5018, pruned_loss=0.2489, over 5673984.64 frames. ], libri_tot_loss[loss=0.5256, simple_loss=0.5206, pruned_loss=0.2703, over 5651676.63 frames. ], giga_tot_loss[loss=0.5044, simple_loss=0.5045, pruned_loss=0.2522, over 5656110.51 frames. ], batch size: 262, lr: 3.72e-02, grad_scale: 4.0
+2023-02-28 12:24:32,652 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7556.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:24:45,944 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7572.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:25:15,635 INFO [train.py:968] (0/2) Epoch 1, batch 7600, giga_loss[loss=0.4989, simple_loss=0.5076, pruned_loss=0.2451, over 28671.00 frames. ], tot_loss[loss=0.4998, simple_loss=0.5022, pruned_loss=0.2487, over 5678008.56 frames. ], libri_tot_loss[loss=0.5254, simple_loss=0.5204, pruned_loss=0.2702, over 5654279.40 frames. ], giga_tot_loss[loss=0.5034, simple_loss=0.5043, pruned_loss=0.2512, over 5661841.68 frames. ], batch size: 307, lr: 3.71e-02, grad_scale: 8.0
+2023-02-28 12:25:26,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.194e+03 1.795e+03 2.176e+03 2.875e+03 5.564e+03, threshold=4.352e+03, percent-clipped=1.0
+2023-02-28 12:25:55,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7646.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:26:00,603 INFO [train.py:968] (0/2) Epoch 1, batch 7650, giga_loss[loss=0.4482, simple_loss=0.4628, pruned_loss=0.2168, over 28832.00 frames. ], tot_loss[loss=0.4973, simple_loss=0.5007, pruned_loss=0.2469, over 5687107.71 frames. ], libri_tot_loss[loss=0.524, simple_loss=0.5198, pruned_loss=0.269, over 5652498.25 frames. ], giga_tot_loss[loss=0.5005, simple_loss=0.5024, pruned_loss=0.2493, over 5676861.46 frames. ], batch size: 119, lr: 3.70e-02, grad_scale: 4.0
+2023-02-28 12:26:19,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7668.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:26:27,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7677.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:26:49,260 INFO [train.py:968] (0/2) Epoch 1, batch 7700, giga_loss[loss=0.4957, simple_loss=0.4897, pruned_loss=0.2508, over 28645.00 frames. ], tot_loss[loss=0.4941, simple_loss=0.4978, pruned_loss=0.2452, over 5678996.15 frames. ], libri_tot_loss[loss=0.5234, simple_loss=0.5195, pruned_loss=0.2684, over 5654315.27 frames. ], giga_tot_loss[loss=0.4968, simple_loss=0.4991, pruned_loss=0.2473, over 5669914.77 frames. ], batch size: 85, lr: 3.69e-02, grad_scale: 4.0
+2023-02-28 12:27:06,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.788e+03 2.143e+03 2.857e+03 6.600e+03, threshold=4.285e+03, percent-clipped=4.0
+2023-02-28 12:27:36,627 INFO [train.py:968] (0/2) Epoch 1, batch 7750, libri_loss[loss=0.3994, simple_loss=0.4208, pruned_loss=0.189, over 29377.00 frames. ], tot_loss[loss=0.4955, simple_loss=0.4977, pruned_loss=0.2466, over 5678858.96 frames. ], libri_tot_loss[loss=0.5217, simple_loss=0.5184, pruned_loss=0.2671, over 5662989.25 frames. ], giga_tot_loss[loss=0.4983, simple_loss=0.4991, pruned_loss=0.2488, over 5664094.82 frames. ], batch size: 67, lr: 3.68e-02, grad_scale: 4.0
+2023-02-28 12:27:39,753 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-02-28 12:28:08,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3990, 0.9164, 1.9268, 0.2749], device='cuda:0'), covar=tensor([0.0744, 0.0708, 0.0314, 0.1529], device='cuda:0'), in_proj_covar=tensor([0.0631, 0.0495, 0.0402, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0002, 0.0004], device='cuda:0')
+2023-02-28 12:28:15,169 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7789.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:28:18,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7792.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:28:26,841 INFO [train.py:968] (0/2) Epoch 1, batch 7800, giga_loss[loss=0.4834, simple_loss=0.4902, pruned_loss=0.2383, over 28742.00 frames. ], tot_loss[loss=0.4992, simple_loss=0.4992, pruned_loss=0.2496, over 5658329.01 frames. ], libri_tot_loss[loss=0.522, simple_loss=0.5186, pruned_loss=0.2672, over 5652021.18 frames. ], giga_tot_loss[loss=0.5009, simple_loss=0.4997, pruned_loss=0.251, over 5655987.70 frames. ], batch size: 119, lr: 3.67e-02, grad_scale: 4.0
+2023-02-28 12:28:27,778 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7800.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:28:27,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0236, 1.0217, 0.8868, 0.8943], device='cuda:0'), covar=tensor([0.1053, 0.1554, 0.1251, 0.0943], device='cuda:0'), in_proj_covar=tensor([0.0674, 0.0821, 0.0755, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0005], device='cuda:0')
+2023-02-28 12:28:42,143 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.986e+02 2.195e+03 3.025e+03 4.147e+03 1.202e+04, threshold=6.050e+03, percent-clipped=22.0
+2023-02-28 12:28:44,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7816.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:28:48,273 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7820.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:28:49,634 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7821.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:28:50,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7823.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:29:03,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3006, 1.5433, 1.6374, 1.3351], device='cuda:0'), covar=tensor([0.1482, 0.1272, 0.1125, 0.2332], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0573, 0.0477, 0.0618], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 12:29:17,598 INFO [train.py:968] (0/2) Epoch 1, batch 7850, giga_loss[loss=0.6082, simple_loss=0.548, pruned_loss=0.3342, over 26671.00 frames. ], tot_loss[loss=0.5005, simple_loss=0.499, pruned_loss=0.251, over 5651750.16 frames. ], libri_tot_loss[loss=0.522, simple_loss=0.5188, pruned_loss=0.2671, over 5655559.71 frames. ], giga_tot_loss[loss=0.5015, simple_loss=0.499, pruned_loss=0.2519, over 5646611.67 frames. ], batch size: 555, lr: 3.66e-02, grad_scale: 4.0
+2023-02-28 12:29:19,800 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7852.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:29:35,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9742, 0.8342, 1.0116, 0.2439], device='cuda:0'), covar=tensor([0.0435, 0.0328, 0.0403, 0.0684], device='cuda:0'), in_proj_covar=tensor([0.0649, 0.0473, 0.0548, 0.0590], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 12:29:59,772 INFO [train.py:968] (0/2) Epoch 1, batch 7900, giga_loss[loss=0.4653, simple_loss=0.4749, pruned_loss=0.2279, over 28529.00 frames. ], tot_loss[loss=0.4976, simple_loss=0.4976, pruned_loss=0.2488, over 5662615.74 frames. ], libri_tot_loss[loss=0.5208, simple_loss=0.5183, pruned_loss=0.2659, over 5663152.92 frames. ], giga_tot_loss[loss=0.4989, simple_loss=0.4975, pruned_loss=0.2501, over 5651760.49 frames. ], batch size: 78, lr: 3.66e-02, grad_scale: 4.0
+2023-02-28 12:30:15,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.813e+03 2.172e+03 2.761e+03 4.539e+03, threshold=4.344e+03, percent-clipped=0.0
+2023-02-28 12:30:29,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7931.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:30:49,803 INFO [train.py:968] (0/2) Epoch 1, batch 7950, giga_loss[loss=0.5789, simple_loss=0.5299, pruned_loss=0.3139, over 26615.00 frames. ], tot_loss[loss=0.4975, simple_loss=0.4979, pruned_loss=0.2486, over 5664776.81 frames. ], libri_tot_loss[loss=0.5191, simple_loss=0.5172, pruned_loss=0.2647, over 5667139.06 frames. ], giga_tot_loss[loss=0.4995, simple_loss=0.4984, pruned_loss=0.2503, over 5652632.43 frames. ], batch size: 555, lr: 3.65e-02, grad_scale: 4.0
+2023-02-28 12:30:59,639 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7959.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:31:02,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7962.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:31:28,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7991.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:31:39,198 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-8000.pt
+2023-02-28 12:31:39,512 INFO [train.py:968] (0/2) Epoch 1, batch 8000, giga_loss[loss=0.5273, simple_loss=0.5279, pruned_loss=0.2633, over 29024.00 frames. ], tot_loss[loss=0.4969, simple_loss=0.4989, pruned_loss=0.2474, over 5665149.42 frames. ], libri_tot_loss[loss=0.5187, simple_loss=0.5172, pruned_loss=0.2642, over 5664877.63 frames. ], giga_tot_loss[loss=0.4984, simple_loss=0.4988, pruned_loss=0.2489, over 5657407.28 frames. ], batch size: 128, lr: 3.64e-02, grad_scale: 8.0
+2023-02-28 12:31:54,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.805e+02 1.869e+03 2.375e+03 3.089e+03 8.046e+03, threshold=4.749e+03, percent-clipped=10.0
+2023-02-28 12:31:56,230 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=8017.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:32:18,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8043.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:32:23,987 INFO [train.py:968] (0/2) Epoch 1, batch 8050, giga_loss[loss=0.4702, simple_loss=0.4849, pruned_loss=0.2277, over 29094.00 frames. ], tot_loss[loss=0.4926, simple_loss=0.4971, pruned_loss=0.244, over 5679702.90 frames. ], libri_tot_loss[loss=0.5167, simple_loss=0.5159, pruned_loss=0.2628, over 5668802.10 frames. ], giga_tot_loss[loss=0.495, simple_loss=0.4979, pruned_loss=0.2461, over 5670078.58 frames. ], batch size: 128, lr: 3.63e-02, grad_scale: 8.0
+2023-02-28 12:32:49,604 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8074.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:32:51,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8077.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:32:56,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6023, 1.7876, 1.3364, 0.8936], device='cuda:0'), covar=tensor([0.0554, 0.0342, 0.0573, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0629, 0.0458, 0.0543, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 12:33:15,914 INFO [train.py:968] (0/2) Epoch 1, batch 8100, giga_loss[loss=0.4441, simple_loss=0.4631, pruned_loss=0.2125, over 28185.00 frames. ], tot_loss[loss=0.4944, simple_loss=0.4982, pruned_loss=0.2453, over 5682488.85 frames. ], libri_tot_loss[loss=0.516, simple_loss=0.5155, pruned_loss=0.2622, over 5673338.48 frames. ], giga_tot_loss[loss=0.4965, simple_loss=0.4988, pruned_loss=0.2471, over 5671022.19 frames. ], batch size: 77, lr: 3.62e-02, grad_scale: 4.0
+2023-02-28 12:33:22,205 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8106.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:33:32,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.335e+03 1.887e+03 2.376e+03 3.225e+03 1.246e+04, threshold=4.752e+03, percent-clipped=9.0
+2023-02-28 12:33:34,184 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-02-28 12:33:49,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=8136.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:34:04,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3505, 1.0825, 1.3969, 1.1473], device='cuda:0'), covar=tensor([0.0636, 0.0422, 0.0412, 0.0590], device='cuda:0'), in_proj_covar=tensor([0.0659, 0.0481, 0.0553, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 12:34:04,847 INFO [train.py:968] (0/2) Epoch 1, batch 8150, giga_loss[loss=0.5259, simple_loss=0.52, pruned_loss=0.2659, over 29023.00 frames. ], tot_loss[loss=0.4991, simple_loss=0.5005, pruned_loss=0.2489, over 5680944.36 frames. ], libri_tot_loss[loss=0.5143, simple_loss=0.5143, pruned_loss=0.261, over 5678697.51 frames. ], giga_tot_loss[loss=0.502, simple_loss=0.5017, pruned_loss=0.2512, over 5667060.76 frames. ], batch size: 155, lr: 3.62e-02, grad_scale: 4.0
+2023-02-28 12:34:27,338 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-02-28 12:34:32,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8175.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:34:44,333 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8186.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:34:48,549 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8189.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:34:55,800 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=8195.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:35:01,160 INFO [train.py:968] (0/2) Epoch 1, batch 8200, giga_loss[loss=0.5261, simple_loss=0.5153, pruned_loss=0.2685, over 28658.00 frames. ], tot_loss[loss=0.5049, simple_loss=0.5031, pruned_loss=0.2533, over 5665042.24 frames. ], libri_tot_loss[loss=0.5144, simple_loss=0.5145, pruned_loss=0.261, over 5679471.02 frames. ], giga_tot_loss[loss=0.507, simple_loss=0.5039, pruned_loss=0.2551, over 5653445.72 frames. ], batch size: 307, lr: 3.61e-02, grad_scale: 4.0
+2023-02-28 12:35:17,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.892e+02 2.048e+03 2.507e+03 3.539e+03 8.389e+03, threshold=5.014e+03, percent-clipped=8.0
+2023-02-28 12:35:22,913 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8218.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:35:53,061 INFO [train.py:968] (0/2) Epoch 1, batch 8250, giga_loss[loss=0.5787, simple_loss=0.5371, pruned_loss=0.3101, over 27975.00 frames. ], tot_loss[loss=0.5075, simple_loss=0.5039, pruned_loss=0.2555, over 5666397.62 frames. ], libri_tot_loss[loss=0.5136, simple_loss=0.5141, pruned_loss=0.2603, over 5684168.17 frames. ], giga_tot_loss[loss=0.5097, simple_loss=0.5046, pruned_loss=0.2574, over 5652560.33 frames. ], batch size: 412, lr: 3.60e-02, grad_scale: 4.0
+2023-02-28 12:35:58,930 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2833, 0.9330, 1.9517, 0.1509], device='cuda:0'), covar=tensor([0.0914, 0.0821, 0.0455, 0.1897], device='cuda:0'), in_proj_covar=tensor([0.0674, 0.0512, 0.0433, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004], device='cuda:0')
+2023-02-28 12:36:43,137 INFO [train.py:968] (0/2) Epoch 1, batch 8300, giga_loss[loss=0.461, simple_loss=0.4807, pruned_loss=0.2206, over 29082.00 frames. ], tot_loss[loss=0.5093, simple_loss=0.5046, pruned_loss=0.2569, over 5670999.16 frames. ], libri_tot_loss[loss=0.5132, simple_loss=0.5139, pruned_loss=0.2598, over 5689680.22 frames. ], giga_tot_loss[loss=0.5114, simple_loss=0.5052, pruned_loss=0.2588, over 5654964.14 frames. ], batch size: 128, lr: 3.59e-02, grad_scale: 4.0
+2023-02-28 12:36:59,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.202e+03 2.013e+03 2.456e+03 3.249e+03 1.080e+04, threshold=4.912e+03, percent-clipped=5.0
+2023-02-28 12:37:01,776 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8318.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:37:04,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8321.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:37:31,434 INFO [train.py:968] (0/2) Epoch 1, batch 8350, giga_loss[loss=0.4789, simple_loss=0.4942, pruned_loss=0.2318, over 28931.00 frames. ], tot_loss[loss=0.5067, simple_loss=0.5029, pruned_loss=0.2553, over 5674476.91 frames. ], libri_tot_loss[loss=0.5111, simple_loss=0.5126, pruned_loss=0.2584, over 5695743.27 frames. ], giga_tot_loss[loss=0.5101, simple_loss=0.5043, pruned_loss=0.258, over 5655557.57 frames. ], batch size: 174, lr: 3.58e-02, grad_scale: 4.0
+2023-02-28 12:37:31,685 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8350.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:37:41,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4771, 1.6913, 2.7718, 1.3772], device='cuda:0'), covar=tensor([0.1376, 0.0941, 0.0412, 0.0796], device='cuda:0'), in_proj_covar=tensor([0.0525, 0.0441, 0.0521, 0.0373], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 12:38:09,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8392.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:38:16,722 INFO [train.py:968] (0/2) Epoch 1, batch 8400, giga_loss[loss=0.4566, simple_loss=0.4772, pruned_loss=0.218, over 28761.00 frames. ], tot_loss[loss=0.5033, simple_loss=0.5011, pruned_loss=0.2528, over 5685260.03 frames. ], libri_tot_loss[loss=0.5103, simple_loss=0.5121, pruned_loss=0.2577, over 5700894.02 frames. ], giga_tot_loss[loss=0.5066, simple_loss=0.5023, pruned_loss=0.2555, over 5665244.90 frames. ], batch size: 186, lr: 3.58e-02, grad_scale: 8.0
+2023-02-28 12:38:30,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 2.005e+03 2.456e+03 3.070e+03 6.349e+03, threshold=4.912e+03, percent-clipped=4.0
+2023-02-28 12:38:32,474 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3767, 1.5537, 1.6724, 1.4161], device='cuda:0'), covar=tensor([0.1610, 0.1348, 0.1278, 0.2381], device='cuda:0'), in_proj_covar=tensor([0.0514, 0.0588, 0.0493, 0.0609], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0009], device='cuda:0')
+2023-02-28 12:38:43,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9127, 2.2492, 4.2108, 2.6674], device='cuda:0'), covar=tensor([0.1530, 0.0974, 0.0242, 0.0598], device='cuda:0'), in_proj_covar=tensor([0.0535, 0.0440, 0.0537, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 12:38:59,231 INFO [train.py:968] (0/2) Epoch 1, batch 8450, giga_loss[loss=0.4283, simple_loss=0.4786, pruned_loss=0.189, over 28475.00 frames. ], tot_loss[loss=0.4924, simple_loss=0.4954, pruned_loss=0.2448, over 5690843.67 frames. ], libri_tot_loss[loss=0.5092, simple_loss=0.5113, pruned_loss=0.2569, over 5702880.22 frames. ], giga_tot_loss[loss=0.4959, simple_loss=0.4969, pruned_loss=0.2475, over 5672839.72 frames. ], batch size: 71, lr: 3.57e-02, grad_scale: 8.0
+2023-02-28 12:39:04,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9067, 2.3212, 1.9609, 1.8051], device='cuda:0'), covar=tensor([0.0677, 0.0791, 0.0459, 0.0363], device='cuda:0'), in_proj_covar=tensor([0.0547, 0.0710, 0.0535, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0005, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 12:39:45,078 INFO [train.py:968] (0/2) Epoch 1, batch 8500, giga_loss[loss=0.5193, simple_loss=0.5075, pruned_loss=0.2656, over 28617.00 frames. ], tot_loss[loss=0.4874, simple_loss=0.4913, pruned_loss=0.2417, over 5685698.91 frames. ], libri_tot_loss[loss=0.5093, simple_loss=0.5115, pruned_loss=0.2569, over 5706082.38 frames. ], giga_tot_loss[loss=0.4897, simple_loss=0.492, pruned_loss=0.2436, over 5668254.19 frames. ], batch size: 92, lr: 3.56e-02, grad_scale: 8.0
+2023-02-28 12:39:55,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8511.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:40:02,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.885e+03 2.408e+03 3.196e+03 7.629e+03, threshold=4.815e+03, percent-clipped=6.0
+2023-02-28 12:40:21,528 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8535.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:40:23,332 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8538.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:40:34,917 INFO [train.py:968] (0/2) Epoch 1, batch 8550, giga_loss[loss=0.4819, simple_loss=0.4921, pruned_loss=0.2359, over 28793.00 frames. ], tot_loss[loss=0.487, simple_loss=0.4901, pruned_loss=0.2419, over 5683605.29 frames. ], libri_tot_loss[loss=0.5098, simple_loss=0.5119, pruned_loss=0.2571, over 5709385.18 frames. ], giga_tot_loss[loss=0.488, simple_loss=0.4901, pruned_loss=0.243, over 5666551.31 frames. ], batch size: 284, lr: 3.55e-02, grad_scale: 4.0
+2023-02-28 12:40:52,278 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8567.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:40:55,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8570.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:41:28,384 INFO [train.py:968] (0/2) Epoch 1, batch 8600, giga_loss[loss=0.4176, simple_loss=0.4442, pruned_loss=0.1955, over 28340.00 frames. ], tot_loss[loss=0.4867, simple_loss=0.4894, pruned_loss=0.242, over 5678795.38 frames. ], libri_tot_loss[loss=0.5093, simple_loss=0.5116, pruned_loss=0.2568, over 5712385.07 frames. ], giga_tot_loss[loss=0.4876, simple_loss=0.4893, pruned_loss=0.2429, over 5662171.91 frames. ], batch size: 65, lr: 3.54e-02, grad_scale: 4.0
+2023-02-28 12:41:42,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.139e+02 1.819e+03 2.475e+03 3.052e+03 7.773e+03, threshold=4.950e+03, percent-clipped=3.0
+2023-02-28 12:42:01,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1505, 1.0885, 1.0786, 1.0254], device='cuda:0'), covar=tensor([0.1753, 0.0981, 0.1062, 0.2718], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0532, 0.0511, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0008, 0.0011], device='cuda:0')
+2023-02-28 12:42:20,647 INFO [train.py:968] (0/2) Epoch 1, batch 8650, giga_loss[loss=0.4823, simple_loss=0.4942, pruned_loss=0.2352, over 28594.00 frames. ], tot_loss[loss=0.4934, simple_loss=0.4947, pruned_loss=0.2461, over 5678666.08 frames. ], libri_tot_loss[loss=0.5078, simple_loss=0.5106, pruned_loss=0.2556, over 5708257.23 frames. ], giga_tot_loss[loss=0.495, simple_loss=0.4948, pruned_loss=0.2475, over 5667308.26 frames. ], batch size: 85, lr: 3.54e-02, grad_scale: 4.0
+2023-02-28 12:42:24,251 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8654.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:42:27,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8657.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:42:27,822 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=8658.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:42:39,502 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.85 vs. limit=5.0
+2023-02-28 12:42:57,217 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8686.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:43:08,289 INFO [train.py:968] (0/2) Epoch 1, batch 8700, giga_loss[loss=0.4606, simple_loss=0.4972, pruned_loss=0.212, over 28520.00 frames. ], tot_loss[loss=0.4953, simple_loss=0.499, pruned_loss=0.2458, over 5683260.97 frames. ], libri_tot_loss[loss=0.5071, simple_loss=0.5103, pruned_loss=0.255, over 5712486.61 frames. ], giga_tot_loss[loss=0.4968, simple_loss=0.4991, pruned_loss=0.2472, over 5669825.61 frames. ], batch size: 71, lr: 3.53e-02, grad_scale: 4.0
+2023-02-28 12:43:21,907 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8713.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:43:23,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8716.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:43:24,106 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.926e+02 1.779e+03 2.105e+03 2.731e+03 5.321e+03, threshold=4.209e+03, percent-clipped=1.0
+2023-02-28 12:43:54,459 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8745.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:43:58,914 INFO [train.py:968] (0/2) Epoch 1, batch 8750, giga_loss[loss=0.4853, simple_loss=0.5014, pruned_loss=0.2347, over 28941.00 frames. ], tot_loss[loss=0.4975, simple_loss=0.5021, pruned_loss=0.2464, over 5673119.82 frames. ], libri_tot_loss[loss=0.5066, simple_loss=0.5099, pruned_loss=0.2546, over 5707244.80 frames. ], giga_tot_loss[loss=0.499, simple_loss=0.5024, pruned_loss=0.2478, over 5665985.17 frames. ], batch size: 145, lr: 3.52e-02, grad_scale: 4.0
+2023-02-28 12:44:45,434 INFO [train.py:968] (0/2) Epoch 1, batch 8800, giga_loss[loss=0.5096, simple_loss=0.5151, pruned_loss=0.2521, over 28667.00 frames. ], tot_loss[loss=0.5007, simple_loss=0.5043, pruned_loss=0.2486, over 5678962.52 frames. ], libri_tot_loss[loss=0.5046, simple_loss=0.5086, pruned_loss=0.2531, over 5711817.08 frames. ], giga_tot_loss[loss=0.5035, simple_loss=0.5054, pruned_loss=0.2508, over 5668319.48 frames. ], batch size: 262, lr: 3.51e-02, grad_scale: 8.0
+2023-02-28 12:44:58,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.184e+03 1.956e+03 2.527e+03 3.310e+03 6.887e+03, threshold=5.055e+03, percent-clipped=14.0
+2023-02-28 12:45:28,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3317, 1.1540, 1.1854, 0.8071], device='cuda:0'), covar=tensor([0.0416, 0.0337, 0.0445, 0.0578], device='cuda:0'), in_proj_covar=tensor([0.0614, 0.0495, 0.0554, 0.0592], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 12:45:31,718 INFO [train.py:968] (0/2) Epoch 1, batch 8850, giga_loss[loss=0.4718, simple_loss=0.4903, pruned_loss=0.2266, over 29058.00 frames. ], tot_loss[loss=0.5005, simple_loss=0.504, pruned_loss=0.2485, over 5688261.51 frames. ], libri_tot_loss[loss=0.5036, simple_loss=0.508, pruned_loss=0.2524, over 5715528.72 frames. ], giga_tot_loss[loss=0.5035, simple_loss=0.5053, pruned_loss=0.2508, over 5675873.73 frames. ], batch size: 128, lr: 3.51e-02, grad_scale: 8.0
+2023-02-28 12:45:38,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4213, 1.3761, 1.8458, 1.5722], device='cuda:0'), covar=tensor([0.1327, 0.1505, 0.1191, 0.2082], device='cuda:0'), in_proj_covar=tensor([0.0492, 0.0574, 0.0470, 0.0576], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 12:46:15,102 INFO [train.py:968] (0/2) Epoch 1, batch 8900, giga_loss[loss=0.5582, simple_loss=0.5365, pruned_loss=0.29, over 28854.00 frames. ], tot_loss[loss=0.5052, simple_loss=0.5064, pruned_loss=0.252, over 5674540.28 frames. ], libri_tot_loss[loss=0.5041, simple_loss=0.5085, pruned_loss=0.2526, over 5699555.27 frames. ], giga_tot_loss[loss=0.5071, simple_loss=0.5069, pruned_loss=0.2536, over 5678388.00 frames. ], batch size: 199, lr: 3.50e-02, grad_scale: 8.0
+2023-02-28 12:46:29,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.907e+03 2.450e+03 3.238e+03 7.385e+03, threshold=4.900e+03, percent-clipped=6.0
+2023-02-28 12:47:03,587 INFO [train.py:968] (0/2) Epoch 1, batch 8950, giga_loss[loss=0.4503, simple_loss=0.4754, pruned_loss=0.2126, over 28948.00 frames. ], tot_loss[loss=0.5047, simple_loss=0.5053, pruned_loss=0.2521, over 5681541.82 frames. ], libri_tot_loss[loss=0.5027, simple_loss=0.5073, pruned_loss=0.2517, over 5703150.65 frames. ], giga_tot_loss[loss=0.5075, simple_loss=0.5067, pruned_loss=0.2541, over 5680781.90 frames. ], batch size: 227, lr: 3.49e-02, grad_scale: 4.0
+2023-02-28 12:47:39,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2870, 0.7776, 1.5797, 0.2401], device='cuda:0'), covar=tensor([0.0686, 0.0671, 0.0508, 0.1443], device='cuda:0'), in_proj_covar=tensor([0.0682, 0.0552, 0.0492, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004], device='cuda:0')
+2023-02-28 12:47:50,071 INFO [train.py:968] (0/2) Epoch 1, batch 9000, giga_loss[loss=0.4622, simple_loss=0.4581, pruned_loss=0.2331, over 23655.00 frames. ], tot_loss[loss=0.4998, simple_loss=0.5016, pruned_loss=0.249, over 5680268.39 frames. ], libri_tot_loss[loss=0.5017, simple_loss=0.5067, pruned_loss=0.2509, over 5705062.03 frames. ], giga_tot_loss[loss=0.5029, simple_loss=0.5031, pruned_loss=0.2514, over 5677238.86 frames. ], batch size: 705, lr: 3.48e-02, grad_scale: 4.0
+2023-02-28 12:47:50,076 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 12:47:57,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4021, 0.9309, 1.6344, 0.2861], device='cuda:0'), covar=tensor([0.0605, 0.0563, 0.0394, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.0684, 0.0554, 0.0493, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004], device='cuda:0')
+2023-02-28 12:47:57,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6713, 2.0830, 1.6844, 1.6869], device='cuda:0'), covar=tensor([0.0663, 0.0917, 0.0611, 0.0408], device='cuda:0'), in_proj_covar=tensor([0.0547, 0.0724, 0.0551, 0.0471], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 12:47:58,245 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.3693, simple_loss=0.4409, pruned_loss=0.1489, over 944034.00 frames. 
+2023-02-28 12:47:58,245 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 18855MB
+2023-02-28 12:48:16,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.648e+03 2.139e+03 2.825e+03 6.349e+03, threshold=4.278e+03, percent-clipped=2.0
+2023-02-28 12:48:31,165 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9033.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:48:44,637 INFO [train.py:968] (0/2) Epoch 1, batch 9050, giga_loss[loss=0.4412, simple_loss=0.4521, pruned_loss=0.2151, over 28704.00 frames. ], tot_loss[loss=0.4956, simple_loss=0.4982, pruned_loss=0.2466, over 5679199.68 frames. ], libri_tot_loss[loss=0.5006, simple_loss=0.5061, pruned_loss=0.25, over 5710275.77 frames. ], giga_tot_loss[loss=0.4991, simple_loss=0.4998, pruned_loss=0.2492, over 5671173.11 frames. ], batch size: 99, lr: 3.48e-02, grad_scale: 4.0
+2023-02-28 12:48:49,367 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1064, 1.3860, 1.5711, 1.2548], device='cuda:0'), covar=tensor([0.1415, 0.1294, 0.1055, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0579, 0.0473, 0.0601], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0009], device='cuda:0')
+2023-02-28 12:49:25,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3235, 3.3584, 2.2303, 2.2832], device='cuda:0'), covar=tensor([0.0576, 0.0476, 0.0454, 0.0276], device='cuda:0'), in_proj_covar=tensor([0.0535, 0.0707, 0.0535, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0005, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 12:49:36,348 INFO [train.py:968] (0/2) Epoch 1, batch 9100, giga_loss[loss=0.4733, simple_loss=0.4902, pruned_loss=0.2282, over 28787.00 frames. ], tot_loss[loss=0.4956, simple_loss=0.4978, pruned_loss=0.2467, over 5684751.27 frames. ], libri_tot_loss[loss=0.4997, simple_loss=0.5056, pruned_loss=0.2493, over 5714106.71 frames. ], giga_tot_loss[loss=0.4991, simple_loss=0.4994, pruned_loss=0.2494, over 5674098.04 frames. ], batch size: 119, lr: 3.47e-02, grad_scale: 2.0
+2023-02-28 12:49:40,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4596, 2.4661, 4.2557, 1.7687], device='cuda:0'), covar=tensor([0.0380, 0.0959, 0.0694, 0.1759], device='cuda:0'), in_proj_covar=tensor([0.0573, 0.0436, 0.0724, 0.0513], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0004, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 12:49:52,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 2.189e+03 2.813e+03 3.485e+03 3.474e+04, threshold=5.625e+03, percent-clipped=14.0
+2023-02-28 12:50:23,451 INFO [train.py:968] (0/2) Epoch 1, batch 9150, giga_loss[loss=0.4792, simple_loss=0.4885, pruned_loss=0.2349, over 28831.00 frames. ], tot_loss[loss=0.497, simple_loss=0.4987, pruned_loss=0.2477, over 5677975.74 frames. ], libri_tot_loss[loss=0.4995, simple_loss=0.5055, pruned_loss=0.249, over 5710805.48 frames. ], giga_tot_loss[loss=0.5, simple_loss=0.4998, pruned_loss=0.2501, over 5670994.44 frames. ], batch size: 227, lr: 3.46e-02, grad_scale: 2.0
+2023-02-28 12:50:27,161 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9153.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:50:49,196 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9176.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:50:51,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9179.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:51:01,154 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 12:51:10,971 INFO [train.py:968] (0/2) Epoch 1, batch 9200, giga_loss[loss=0.4947, simple_loss=0.4967, pruned_loss=0.2464, over 28832.00 frames. ], tot_loss[loss=0.4944, simple_loss=0.4962, pruned_loss=0.2463, over 5677867.60 frames. ], libri_tot_loss[loss=0.4989, simple_loss=0.5053, pruned_loss=0.2485, over 5713685.19 frames. ], giga_tot_loss[loss=0.4972, simple_loss=0.4972, pruned_loss=0.2486, over 5669098.53 frames. ], batch size: 99, lr: 3.46e-02, grad_scale: 4.0
+2023-02-28 12:51:19,294 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9208.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:51:28,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 1.806e+03 2.216e+03 2.836e+03 6.190e+03, threshold=4.432e+03, percent-clipped=2.0
+2023-02-28 12:51:58,941 INFO [train.py:968] (0/2) Epoch 1, batch 9250, libri_loss[loss=0.4623, simple_loss=0.4823, pruned_loss=0.2212, over 29543.00 frames. ], tot_loss[loss=0.4937, simple_loss=0.4953, pruned_loss=0.2461, over 5685979.42 frames. ], libri_tot_loss[loss=0.4983, simple_loss=0.505, pruned_loss=0.2481, over 5717707.09 frames. ], giga_tot_loss[loss=0.4964, simple_loss=0.4961, pruned_loss=0.2483, over 5674689.72 frames. ], batch size: 78, lr: 3.45e-02, grad_scale: 4.0
+2023-02-28 12:52:21,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3464, 2.0106, 1.4369, 1.2345], device='cuda:0'), covar=tensor([0.0908, 0.1262, 0.1505, 0.2128], device='cuda:0'), in_proj_covar=tensor([0.0554, 0.0712, 0.0694, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 12:52:47,494 INFO [train.py:968] (0/2) Epoch 1, batch 9300, giga_loss[loss=0.5469, simple_loss=0.5388, pruned_loss=0.2775, over 28951.00 frames. ], tot_loss[loss=0.495, simple_loss=0.4969, pruned_loss=0.2466, over 5680937.11 frames. ], libri_tot_loss[loss=0.4975, simple_loss=0.5044, pruned_loss=0.2474, over 5719431.98 frames. ], giga_tot_loss[loss=0.4979, simple_loss=0.4979, pruned_loss=0.2489, over 5669842.06 frames. ], batch size: 213, lr: 3.44e-02, grad_scale: 4.0
+2023-02-28 12:53:08,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.731e+02 1.824e+03 2.236e+03 2.791e+03 7.077e+03, threshold=4.472e+03, percent-clipped=5.0
+2023-02-28 12:53:09,994 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9320.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:53:34,120 INFO [train.py:968] (0/2) Epoch 1, batch 9350, giga_loss[loss=0.6236, simple_loss=0.5643, pruned_loss=0.3415, over 26573.00 frames. ], tot_loss[loss=0.4955, simple_loss=0.4977, pruned_loss=0.2466, over 5685003.25 frames. ], libri_tot_loss[loss=0.4956, simple_loss=0.5031, pruned_loss=0.2462, over 5726740.16 frames. ], giga_tot_loss[loss=0.4995, simple_loss=0.4997, pruned_loss=0.2496, over 5667706.09 frames. ], batch size: 555, lr: 3.43e-02, grad_scale: 4.0
+2023-02-28 12:54:03,123 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9379.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:54:21,832 INFO [train.py:968] (0/2) Epoch 1, batch 9400, giga_loss[loss=0.4527, simple_loss=0.4742, pruned_loss=0.2156, over 28981.00 frames. ], tot_loss[loss=0.4959, simple_loss=0.4976, pruned_loss=0.2471, over 5680056.18 frames. ], libri_tot_loss[loss=0.4959, simple_loss=0.5033, pruned_loss=0.2463, over 5723875.28 frames. ], giga_tot_loss[loss=0.4987, simple_loss=0.4988, pruned_loss=0.2493, over 5668772.68 frames. ], batch size: 164, lr: 3.43e-02, grad_scale: 4.0
+2023-02-28 12:54:26,802 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9406.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:54:39,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.914e+03 2.509e+03 3.425e+03 7.510e+03, threshold=5.018e+03, percent-clipped=11.0
+2023-02-28 12:55:11,350 INFO [train.py:968] (0/2) Epoch 1, batch 9450, giga_loss[loss=0.4819, simple_loss=0.5115, pruned_loss=0.2261, over 28899.00 frames. ], tot_loss[loss=0.4973, simple_loss=0.5001, pruned_loss=0.2472, over 5681403.26 frames. ], libri_tot_loss[loss=0.4957, simple_loss=0.5031, pruned_loss=0.2462, over 5726571.94 frames. ], giga_tot_loss[loss=0.4997, simple_loss=0.5011, pruned_loss=0.2492, over 5669429.39 frames. ], batch size: 174, lr: 3.42e-02, grad_scale: 4.0
+2023-02-28 12:55:28,942 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5846, 1.5058, 1.0107, 1.1818], device='cuda:0'), covar=tensor([0.0809, 0.0867, 0.1245, 0.1366], device='cuda:0'), in_proj_covar=tensor([0.0559, 0.0695, 0.0688, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0007, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 12:55:56,873 INFO [train.py:968] (0/2) Epoch 1, batch 9500, giga_loss[loss=0.4682, simple_loss=0.4944, pruned_loss=0.221, over 28923.00 frames. ], tot_loss[loss=0.4966, simple_loss=0.5017, pruned_loss=0.2457, over 5687786.79 frames. ], libri_tot_loss[loss=0.495, simple_loss=0.5028, pruned_loss=0.2456, over 5731009.35 frames. ], giga_tot_loss[loss=0.4991, simple_loss=0.5027, pruned_loss=0.2478, over 5673060.07 frames. ], batch size: 227, lr: 3.41e-02, grad_scale: 4.0
+2023-02-28 12:56:15,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.492e+02 1.899e+03 2.569e+03 3.472e+03 1.171e+04, threshold=5.138e+03, percent-clipped=14.0
+2023-02-28 12:56:22,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9528.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:56:44,664 INFO [train.py:968] (0/2) Epoch 1, batch 9550, giga_loss[loss=0.5179, simple_loss=0.5222, pruned_loss=0.2568, over 28743.00 frames. ], tot_loss[loss=0.4923, simple_loss=0.5007, pruned_loss=0.242, over 5686265.08 frames. ], libri_tot_loss[loss=0.4941, simple_loss=0.5023, pruned_loss=0.2448, over 5733043.75 frames. ], giga_tot_loss[loss=0.4952, simple_loss=0.5019, pruned_loss=0.2443, over 5672014.43 frames. ], batch size: 262, lr: 3.41e-02, grad_scale: 4.0
+2023-02-28 12:57:30,485 INFO [train.py:968] (0/2) Epoch 1, batch 9600, giga_loss[loss=0.4623, simple_loss=0.4799, pruned_loss=0.2223, over 28652.00 frames. ], tot_loss[loss=0.496, simple_loss=0.5027, pruned_loss=0.2446, over 5688762.63 frames. ], libri_tot_loss[loss=0.4928, simple_loss=0.5013, pruned_loss=0.244, over 5737622.61 frames. ], giga_tot_loss[loss=0.4995, simple_loss=0.5047, pruned_loss=0.2472, over 5671691.30 frames. ], batch size: 92, lr: 3.40e-02, grad_scale: 8.0
+2023-02-28 12:57:48,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.893e+02 1.591e+03 2.129e+03 2.663e+03 6.080e+03, threshold=4.259e+03, percent-clipped=1.0
+2023-02-28 12:58:14,414 INFO [train.py:968] (0/2) Epoch 1, batch 9650, giga_loss[loss=0.6179, simple_loss=0.5761, pruned_loss=0.3299, over 28060.00 frames. ], tot_loss[loss=0.4986, simple_loss=0.5038, pruned_loss=0.2468, over 5679258.42 frames. ], libri_tot_loss[loss=0.4915, simple_loss=0.5005, pruned_loss=0.2431, over 5730509.36 frames. ], giga_tot_loss[loss=0.5027, simple_loss=0.5061, pruned_loss=0.2497, over 5670263.10 frames. ], batch size: 412, lr: 3.39e-02, grad_scale: 4.0
+2023-02-28 12:58:26,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-02-28 12:58:27,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3767, 1.7281, 1.7252, 1.2864], device='cuda:0'), covar=tensor([0.1960, 0.0990, 0.1004, 0.2672], device='cuda:0'), in_proj_covar=tensor([0.0556, 0.0512, 0.0490, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0008, 0.0011], device='cuda:0')
+2023-02-28 12:58:38,963 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9671.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:58:41,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9674.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:59:01,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9695.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:59:06,560 INFO [train.py:968] (0/2) Epoch 1, batch 9700, giga_loss[loss=0.4584, simple_loss=0.4752, pruned_loss=0.2209, over 28964.00 frames. ], tot_loss[loss=0.5011, simple_loss=0.5042, pruned_loss=0.249, over 5665626.57 frames. ], libri_tot_loss[loss=0.4915, simple_loss=0.5005, pruned_loss=0.243, over 5721954.03 frames. ], giga_tot_loss[loss=0.5044, simple_loss=0.5061, pruned_loss=0.2514, over 5666026.56 frames. ], batch size: 106, lr: 3.38e-02, grad_scale: 4.0
+2023-02-28 12:59:11,016 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9703.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:59:12,071 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-02-28 12:59:24,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 2.056e+03 2.513e+03 3.234e+03 6.731e+03, threshold=5.026e+03, percent-clipped=8.0
+2023-02-28 12:59:29,592 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-02-28 12:59:51,594 INFO [train.py:968] (0/2) Epoch 1, batch 9750, giga_loss[loss=0.4938, simple_loss=0.5017, pruned_loss=0.2429, over 28617.00 frames. ], tot_loss[loss=0.499, simple_loss=0.502, pruned_loss=0.248, over 5637563.34 frames. ], libri_tot_loss[loss=0.4894, simple_loss=0.499, pruned_loss=0.2417, over 5709016.10 frames. ], giga_tot_loss[loss=0.5039, simple_loss=0.5051, pruned_loss=0.2514, over 5646122.79 frames. ], batch size: 336, lr: 3.38e-02, grad_scale: 4.0
+2023-02-28 12:59:56,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9754.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:00:03,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1528, 1.0843, 1.3695, 0.5743], device='cuda:0'), covar=tensor([0.0439, 0.0266, 0.0297, 0.0512], device='cuda:0'), in_proj_covar=tensor([0.0653, 0.0492, 0.0554, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 13:00:19,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9781.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:00:35,531 INFO [train.py:968] (0/2) Epoch 1, batch 9800, giga_loss[loss=0.4021, simple_loss=0.4605, pruned_loss=0.1719, over 28472.00 frames. ], tot_loss[loss=0.4943, simple_loss=0.5003, pruned_loss=0.2441, over 5641523.90 frames. ], libri_tot_loss[loss=0.4886, simple_loss=0.4983, pruned_loss=0.2411, over 5702218.29 frames. ], giga_tot_loss[loss=0.4992, simple_loss=0.5034, pruned_loss=0.2475, over 5652480.42 frames. ], batch size: 65, lr: 3.37e-02, grad_scale: 2.0
+2023-02-28 13:00:42,818 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9807.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:00:53,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.886e+02 1.860e+03 2.193e+03 2.766e+03 1.109e+04, threshold=4.386e+03, percent-clipped=4.0
+2023-02-28 13:01:09,187 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9838.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:01:11,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9841.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:01:22,052 INFO [train.py:968] (0/2) Epoch 1, batch 9850, giga_loss[loss=0.483, simple_loss=0.5043, pruned_loss=0.2308, over 29084.00 frames. ], tot_loss[loss=0.4903, simple_loss=0.4992, pruned_loss=0.2407, over 5642772.79 frames. ], libri_tot_loss[loss=0.4884, simple_loss=0.4982, pruned_loss=0.2409, over 5693838.38 frames. ], giga_tot_loss[loss=0.4946, simple_loss=0.5018, pruned_loss=0.2437, over 5658558.19 frames. ], batch size: 128, lr: 3.36e-02, grad_scale: 2.0
+2023-02-28 13:01:37,965 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9870.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:02:07,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9897.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:02:09,654 INFO [train.py:968] (0/2) Epoch 1, batch 9900, giga_loss[loss=0.5806, simple_loss=0.558, pruned_loss=0.3016, over 28830.00 frames. ], tot_loss[loss=0.4931, simple_loss=0.5012, pruned_loss=0.2425, over 5653576.49 frames. ], libri_tot_loss[loss=0.4878, simple_loss=0.4979, pruned_loss=0.2404, over 5698876.96 frames. ], giga_tot_loss[loss=0.497, simple_loss=0.5035, pruned_loss=0.2453, over 5660683.16 frames. ], batch size: 186, lr: 3.36e-02, grad_scale: 2.0
+2023-02-28 13:02:09,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9900.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:02:28,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.765e+02 1.701e+03 2.206e+03 2.777e+03 7.661e+03, threshold=4.412e+03, percent-clipped=8.0
+2023-02-28 13:02:34,055 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9924.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:02:36,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9927.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:02:38,291 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9929.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:03:00,601 INFO [train.py:968] (0/2) Epoch 1, batch 9950, giga_loss[loss=0.4265, simple_loss=0.4651, pruned_loss=0.1939, over 28962.00 frames. ], tot_loss[loss=0.4955, simple_loss=0.5021, pruned_loss=0.2444, over 5644496.07 frames. ], libri_tot_loss[loss=0.4874, simple_loss=0.4977, pruned_loss=0.2401, over 5692705.51 frames. ], giga_tot_loss[loss=0.4989, simple_loss=0.5041, pruned_loss=0.2469, over 5655744.32 frames. ], batch size: 164, lr: 3.35e-02, grad_scale: 2.0
+2023-02-28 13:03:08,238 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9956.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:03:20,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0496, 2.0912, 3.8778, 2.3455], device='cuda:0'), covar=tensor([0.1447, 0.1009, 0.0269, 0.0582], device='cuda:0'), in_proj_covar=tensor([0.0522, 0.0434, 0.0535, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 13:03:26,229 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9977.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:03:46,677 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-10000.pt
+2023-02-28 13:03:46,978 INFO [train.py:968] (0/2) Epoch 1, batch 10000, libri_loss[loss=0.5214, simple_loss=0.5212, pruned_loss=0.2608, over 29540.00 frames. ], tot_loss[loss=0.4928, simple_loss=0.5, pruned_loss=0.2428, over 5664827.73 frames. ], libri_tot_loss[loss=0.486, simple_loss=0.497, pruned_loss=0.239, over 5701469.59 frames. ], giga_tot_loss[loss=0.4971, simple_loss=0.5024, pruned_loss=0.2459, over 5664460.14 frames. ], batch size: 82, lr: 3.34e-02, grad_scale: 4.0
+2023-02-28 13:03:49,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2546, 2.9425, 5.0275, 2.1730], device='cuda:0'), covar=tensor([0.0292, 0.0858, 0.0597, 0.1496], device='cuda:0'), in_proj_covar=tensor([0.0615, 0.0435, 0.0738, 0.0517], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0004, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 13:04:04,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.245e+03 2.042e+03 2.521e+03 3.418e+03 6.660e+03, threshold=5.042e+03, percent-clipped=11.0
+2023-02-28 13:04:19,862 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 13:04:32,249 INFO [train.py:968] (0/2) Epoch 1, batch 10050, giga_loss[loss=0.48, simple_loss=0.4788, pruned_loss=0.2406, over 28941.00 frames. ], tot_loss[loss=0.4925, simple_loss=0.4986, pruned_loss=0.2432, over 5654971.31 frames. ], libri_tot_loss[loss=0.4858, simple_loss=0.4968, pruned_loss=0.2389, over 5699288.13 frames. ], giga_tot_loss[loss=0.4965, simple_loss=0.5009, pruned_loss=0.246, over 5655050.26 frames. ], batch size: 227, lr: 3.34e-02, grad_scale: 4.0
+2023-02-28 13:04:34,288 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=10051.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:05:01,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8206, 1.5079, 1.2307, 1.2441], device='cuda:0'), covar=tensor([0.0815, 0.1048, 0.1294, 0.1479], device='cuda:0'), in_proj_covar=tensor([0.0566, 0.0711, 0.0687, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 13:05:11,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2008, 1.1988, 1.1260, 1.1047], device='cuda:0'), covar=tensor([0.1723, 0.0884, 0.0960, 0.2507], device='cuda:0'), in_proj_covar=tensor([0.0559, 0.0501, 0.0483, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0008, 0.0012], device='cuda:0')
+2023-02-28 13:05:20,479 INFO [train.py:968] (0/2) Epoch 1, batch 10100, giga_loss[loss=0.4907, simple_loss=0.5019, pruned_loss=0.2397, over 28708.00 frames. ], tot_loss[loss=0.4871, simple_loss=0.4944, pruned_loss=0.2399, over 5655745.25 frames. ], libri_tot_loss[loss=0.4846, simple_loss=0.4959, pruned_loss=0.238, over 5690510.29 frames. ], giga_tot_loss[loss=0.4914, simple_loss=0.4969, pruned_loss=0.2429, over 5661950.06 frames. ], batch size: 262, lr: 3.33e-02, grad_scale: 4.0
+2023-02-28 13:05:36,492 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-02-28 13:05:41,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.782e+02 1.908e+03 2.347e+03 3.309e+03 1.020e+04, threshold=4.694e+03, percent-clipped=7.0
+2023-02-28 13:05:56,484 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=10136.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:06:14,557 INFO [train.py:968] (0/2) Epoch 1, batch 10150, giga_loss[loss=0.5161, simple_loss=0.482, pruned_loss=0.2751, over 23433.00 frames. ], tot_loss[loss=0.4862, simple_loss=0.4927, pruned_loss=0.2399, over 5657223.86 frames. ], libri_tot_loss[loss=0.484, simple_loss=0.4955, pruned_loss=0.2376, over 5693959.75 frames. ], giga_tot_loss[loss=0.4903, simple_loss=0.4951, pruned_loss=0.2427, over 5658347.66 frames. ], batch size: 705, lr: 3.32e-02, grad_scale: 4.0
+2023-02-28 13:06:41,351 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3749, 1.3778, 1.4369, 1.3007], device='cuda:0'), covar=tensor([0.1472, 0.1345, 0.1283, 0.2389], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0574, 0.0472, 0.0579], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0009], device='cuda:0')
+2023-02-28 13:06:41,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10182.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:06:59,935 INFO [train.py:968] (0/2) Epoch 1, batch 10200, giga_loss[loss=0.5401, simple_loss=0.524, pruned_loss=0.2781, over 28608.00 frames. ], tot_loss[loss=0.4852, simple_loss=0.4919, pruned_loss=0.2392, over 5667199.75 frames. ], libri_tot_loss[loss=0.4836, simple_loss=0.4956, pruned_loss=0.2371, over 5701774.10 frames. ], giga_tot_loss[loss=0.4889, simple_loss=0.4937, pruned_loss=0.2421, over 5659582.14 frames. ], batch size: 307, lr: 3.32e-02, grad_scale: 4.0
+2023-02-28 13:07:17,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.711e+03 2.195e+03 2.783e+03 5.904e+03, threshold=4.390e+03, percent-clipped=2.0
+2023-02-28 13:07:33,600 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-02-28 13:07:43,341 INFO [train.py:968] (0/2) Epoch 1, batch 10250, giga_loss[loss=0.4414, simple_loss=0.467, pruned_loss=0.2079, over 28681.00 frames. ], tot_loss[loss=0.4807, simple_loss=0.489, pruned_loss=0.2362, over 5672329.50 frames. ], libri_tot_loss[loss=0.4825, simple_loss=0.495, pruned_loss=0.2363, over 5707797.49 frames. ], giga_tot_loss[loss=0.4848, simple_loss=0.4909, pruned_loss=0.2393, over 5659599.07 frames. ], batch size: 92, lr: 3.31e-02, grad_scale: 4.0
+2023-02-28 13:08:30,186 INFO [train.py:968] (0/2) Epoch 1, batch 10300, giga_loss[loss=0.4087, simple_loss=0.4435, pruned_loss=0.1869, over 29043.00 frames. ], tot_loss[loss=0.4755, simple_loss=0.4863, pruned_loss=0.2323, over 5669794.32 frames. ], libri_tot_loss[loss=0.4822, simple_loss=0.4948, pruned_loss=0.236, over 5708237.51 frames. ], giga_tot_loss[loss=0.4789, simple_loss=0.4878, pruned_loss=0.235, over 5659162.66 frames. ], batch size: 164, lr: 3.30e-02, grad_scale: 4.0
+2023-02-28 13:08:50,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.614e+03 2.093e+03 2.809e+03 9.740e+03, threshold=4.186e+03, percent-clipped=9.0
+2023-02-28 13:08:57,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10325.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:09:00,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10328.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:09:20,285 INFO [train.py:968] (0/2) Epoch 1, batch 10350, libri_loss[loss=0.5066, simple_loss=0.5174, pruned_loss=0.2478, over 29358.00 frames. ], tot_loss[loss=0.4722, simple_loss=0.4843, pruned_loss=0.23, over 5657404.69 frames. ], libri_tot_loss[loss=0.4818, simple_loss=0.4946, pruned_loss=0.2357, over 5703585.13 frames. ], giga_tot_loss[loss=0.475, simple_loss=0.4854, pruned_loss=0.2323, over 5652157.76 frames. ], batch size: 92, lr: 3.30e-02, grad_scale: 4.0
+2023-02-28 13:09:22,582 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10352.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:09:27,673 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10357.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:10:08,237 INFO [train.py:968] (0/2) Epoch 1, batch 10400, giga_loss[loss=0.4345, simple_loss=0.4584, pruned_loss=0.2053, over 28847.00 frames. ], tot_loss[loss=0.4713, simple_loss=0.4831, pruned_loss=0.2298, over 5663655.57 frames. ], libri_tot_loss[loss=0.4815, simple_loss=0.4945, pruned_loss=0.2354, over 5704197.68 frames. ], giga_tot_loss[loss=0.4737, simple_loss=0.484, pruned_loss=0.2317, over 5657956.11 frames. ], batch size: 174, lr: 3.29e-02, grad_scale: 8.0
+2023-02-28 13:10:21,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5131, 2.6831, 4.2634, 1.6652], device='cuda:0'), covar=tensor([0.0425, 0.0939, 0.0761, 0.1996], device='cuda:0'), in_proj_covar=tensor([0.0632, 0.0446, 0.0744, 0.0533], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 13:10:34,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.875e+03 2.275e+03 2.868e+03 5.272e+03, threshold=4.550e+03, percent-clipped=4.0
+2023-02-28 13:10:39,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10426.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:10:44,477 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=10432.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:11:01,217 INFO [train.py:968] (0/2) Epoch 1, batch 10450, giga_loss[loss=0.4651, simple_loss=0.4712, pruned_loss=0.2295, over 28867.00 frames. ], tot_loss[loss=0.4667, simple_loss=0.4786, pruned_loss=0.2274, over 5670423.24 frames. ], libri_tot_loss[loss=0.4813, simple_loss=0.4943, pruned_loss=0.2353, over 5708866.71 frames. ], giga_tot_loss[loss=0.4684, simple_loss=0.479, pruned_loss=0.2289, over 5660540.19 frames. ], batch size: 263, lr: 3.28e-02, grad_scale: 4.0
+2023-02-28 13:11:02,318 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-02-28 13:11:05,171 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=10455.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:11:45,189 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10495.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:11:47,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10498.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:11:50,092 INFO [train.py:968] (0/2) Epoch 1, batch 10500, giga_loss[loss=0.4743, simple_loss=0.496, pruned_loss=0.2263, over 28883.00 frames. ], tot_loss[loss=0.4682, simple_loss=0.4798, pruned_loss=0.2283, over 5676350.61 frames. ], libri_tot_loss[loss=0.4807, simple_loss=0.4939, pruned_loss=0.2349, over 5711016.37 frames. ], giga_tot_loss[loss=0.47, simple_loss=0.4804, pruned_loss=0.2298, over 5666286.33 frames. ], batch size: 227, lr: 3.28e-02, grad_scale: 4.0
+2023-02-28 13:12:00,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10511.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:12:10,212 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.677e+03 2.324e+03 3.468e+03 1.095e+04, threshold=4.648e+03, percent-clipped=10.0
+2023-02-28 13:12:14,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10527.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:12:36,498 INFO [train.py:968] (0/2) Epoch 1, batch 10550, giga_loss[loss=0.5431, simple_loss=0.5262, pruned_loss=0.28, over 28774.00 frames. ], tot_loss[loss=0.4708, simple_loss=0.4822, pruned_loss=0.2297, over 5677282.00 frames. ], libri_tot_loss[loss=0.4809, simple_loss=0.4942, pruned_loss=0.2349, over 5714785.00 frames. ], giga_tot_loss[loss=0.4718, simple_loss=0.4822, pruned_loss=0.2307, over 5665237.60 frames. ], batch size: 284, lr: 3.27e-02, grad_scale: 4.0
+2023-02-28 13:12:55,150 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10569.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:12:58,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10572.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:13:22,100 INFO [train.py:968] (0/2) Epoch 1, batch 10600, libri_loss[loss=0.5207, simple_loss=0.5345, pruned_loss=0.2534, over 29376.00 frames. ], tot_loss[loss=0.4672, simple_loss=0.48, pruned_loss=0.2272, over 5665144.49 frames. ], libri_tot_loss[loss=0.4818, simple_loss=0.495, pruned_loss=0.2354, over 5720843.37 frames. ], giga_tot_loss[loss=0.4669, simple_loss=0.4789, pruned_loss=0.2274, over 5648662.29 frames. ], batch size: 92, lr: 3.27e-02, grad_scale: 1.0
+2023-02-28 13:13:23,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10601.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:13:47,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.951e+02 1.614e+03 2.207e+03 3.444e+03 3.720e+04, threshold=4.413e+03, percent-clipped=10.0
+2023-02-28 13:14:12,808 INFO [train.py:968] (0/2) Epoch 1, batch 10650, giga_loss[loss=0.4336, simple_loss=0.4411, pruned_loss=0.213, over 28711.00 frames. ], tot_loss[loss=0.4665, simple_loss=0.4787, pruned_loss=0.2271, over 5646314.77 frames. ], libri_tot_loss[loss=0.482, simple_loss=0.4953, pruned_loss=0.2355, over 5723515.23 frames. ], giga_tot_loss[loss=0.4658, simple_loss=0.4773, pruned_loss=0.2271, over 5629821.07 frames. ], batch size: 92, lr: 3.26e-02, grad_scale: 1.0
+2023-02-28 13:14:17,234 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10654.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:14:20,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10657.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:14:21,310 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3518, 1.4993, 1.5967, 1.2864], device='cuda:0'), covar=tensor([0.1795, 0.1042, 0.0934, 0.2439], device='cuda:0'), in_proj_covar=tensor([0.0561, 0.0506, 0.0475, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0010, 0.0008, 0.0012], device='cuda:0')
+2023-02-28 13:14:35,381 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-02-28 13:14:47,067 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10686.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:15:00,128 INFO [train.py:968] (0/2) Epoch 1, batch 10700, giga_loss[loss=0.4812, simple_loss=0.4949, pruned_loss=0.2338, over 28903.00 frames. ], tot_loss[loss=0.4698, simple_loss=0.481, pruned_loss=0.2293, over 5658978.36 frames. ], libri_tot_loss[loss=0.4822, simple_loss=0.4955, pruned_loss=0.2355, over 5726577.66 frames. ], giga_tot_loss[loss=0.469, simple_loss=0.4795, pruned_loss=0.2292, over 5642195.22 frames. ], batch size: 186, lr: 3.25e-02, grad_scale: 1.0
+2023-02-28 13:15:25,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.761e+03 2.116e+03 2.513e+03 6.078e+03, threshold=4.231e+03, percent-clipped=5.0
+2023-02-28 13:15:39,934 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-02-28 13:15:50,771 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-02-28 13:15:52,290 INFO [train.py:968] (0/2) Epoch 1, batch 10750, giga_loss[loss=0.4858, simple_loss=0.5012, pruned_loss=0.2352, over 28837.00 frames. ], tot_loss[loss=0.4733, simple_loss=0.4837, pruned_loss=0.2314, over 5655933.23 frames. ], libri_tot_loss[loss=0.4817, simple_loss=0.4952, pruned_loss=0.2351, over 5729211.43 frames. ], giga_tot_loss[loss=0.4729, simple_loss=0.4826, pruned_loss=0.2316, over 5639445.13 frames. ], batch size: 112, lr: 3.25e-02, grad_scale: 1.0
+2023-02-28 13:16:36,628 INFO [train.py:968] (0/2) Epoch 1, batch 10800, giga_loss[loss=0.5446, simple_loss=0.5234, pruned_loss=0.2829, over 28241.00 frames. ], tot_loss[loss=0.4744, simple_loss=0.4847, pruned_loss=0.2321, over 5653840.32 frames. ], libri_tot_loss[loss=0.481, simple_loss=0.4947, pruned_loss=0.2347, over 5725152.05 frames. ], giga_tot_loss[loss=0.4745, simple_loss=0.4839, pruned_loss=0.2325, over 5641219.65 frames. ], batch size: 368, lr: 3.24e-02, grad_scale: 2.0
+2023-02-28 13:16:38,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=10802.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:16:41,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10807.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:16:56,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.371e+02 1.799e+03 2.128e+03 3.063e+03 6.754e+03, threshold=4.256e+03, percent-clipped=9.0
+2023-02-28 13:17:00,408 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10830.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:17:17,518 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=10847.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:17:19,420 INFO [train.py:968] (0/2) Epoch 1, batch 10850, giga_loss[loss=0.5159, simple_loss=0.5084, pruned_loss=0.2617, over 28211.00 frames. ], tot_loss[loss=0.4769, simple_loss=0.4864, pruned_loss=0.2337, over 5647957.00 frames. ], libri_tot_loss[loss=0.4807, simple_loss=0.4946, pruned_loss=0.2344, over 5718441.17 frames. ], giga_tot_loss[loss=0.4771, simple_loss=0.4856, pruned_loss=0.2343, over 5641629.17 frames. ], batch size: 368, lr: 3.23e-02, grad_scale: 2.0
+2023-02-28 13:18:09,536 INFO [train.py:968] (0/2) Epoch 1, batch 10900, giga_loss[loss=0.4667, simple_loss=0.4845, pruned_loss=0.2244, over 28929.00 frames. ], tot_loss[loss=0.4815, simple_loss=0.4893, pruned_loss=0.2369, over 5653877.37 frames. ], libri_tot_loss[loss=0.4805, simple_loss=0.4946, pruned_loss=0.2342, over 5721532.95 frames. ], giga_tot_loss[loss=0.4818, simple_loss=0.4885, pruned_loss=0.2376, over 5644857.92 frames. ], batch size: 136, lr: 3.23e-02, grad_scale: 2.0
+2023-02-28 13:18:31,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.831e+02 1.766e+03 2.210e+03 2.818e+03 7.846e+03, threshold=4.421e+03, percent-clipped=8.0
+2023-02-28 13:18:58,363 INFO [train.py:968] (0/2) Epoch 1, batch 10950, giga_loss[loss=0.5026, simple_loss=0.504, pruned_loss=0.2506, over 27497.00 frames. ], tot_loss[loss=0.4785, simple_loss=0.4887, pruned_loss=0.2342, over 5662142.92 frames. ], libri_tot_loss[loss=0.4794, simple_loss=0.4938, pruned_loss=0.2334, over 5726342.23 frames. ], giga_tot_loss[loss=0.4798, simple_loss=0.4887, pruned_loss=0.2355, over 5649263.72 frames. ], batch size: 472, lr: 3.22e-02, grad_scale: 2.0
+2023-02-28 13:18:58,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10950.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:19:02,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10953.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:19:23,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10973.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 13:19:25,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10976.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:19:31,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10982.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:19:49,439 INFO [train.py:968] (0/2) Epoch 1, batch 11000, giga_loss[loss=0.4937, simple_loss=0.4884, pruned_loss=0.2495, over 28863.00 frames. ], tot_loss[loss=0.4757, simple_loss=0.4863, pruned_loss=0.2326, over 5654851.62 frames. ], libri_tot_loss[loss=0.4773, simple_loss=0.4924, pruned_loss=0.2321, over 5730003.61 frames. ], giga_tot_loss[loss=0.4785, simple_loss=0.4874, pruned_loss=0.2348, over 5639706.77 frames. ], batch size: 199, lr: 3.22e-02, grad_scale: 2.0
+2023-02-28 13:19:56,053 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11005.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 13:20:14,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.606e+03 2.333e+03 3.278e+03 8.828e+03, threshold=4.665e+03, percent-clipped=9.0
+2023-02-28 13:20:23,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8851, 1.7680, 1.2763, 1.4004], device='cuda:0'), covar=tensor([0.0736, 0.0874, 0.1173, 0.1312], device='cuda:0'), in_proj_covar=tensor([0.0540, 0.0709, 0.0675, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 13:20:42,323 INFO [train.py:968] (0/2) Epoch 1, batch 11050, giga_loss[loss=0.436, simple_loss=0.4668, pruned_loss=0.2026, over 28931.00 frames. ], tot_loss[loss=0.4743, simple_loss=0.4844, pruned_loss=0.2321, over 5653758.97 frames. ], libri_tot_loss[loss=0.4772, simple_loss=0.4922, pruned_loss=0.232, over 5722076.49 frames. ], giga_tot_loss[loss=0.4766, simple_loss=0.4853, pruned_loss=0.234, over 5647931.49 frames. ], batch size: 199, lr: 3.21e-02, grad_scale: 2.0
+2023-02-28 13:20:57,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0458, 1.6557, 1.5570, 1.5131], device='cuda:0'), covar=tensor([0.0557, 0.1337, 0.0861, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0545, 0.0910, 0.0682, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 13:21:37,733 INFO [train.py:968] (0/2) Epoch 1, batch 11100, libri_loss[loss=0.4565, simple_loss=0.4899, pruned_loss=0.2115, over 29625.00 frames. ], tot_loss[loss=0.4754, simple_loss=0.4848, pruned_loss=0.233, over 5654593.71 frames. ], libri_tot_loss[loss=0.4757, simple_loss=0.4912, pruned_loss=0.231, over 5727641.66 frames. ], giga_tot_loss[loss=0.4786, simple_loss=0.4863, pruned_loss=0.2355, over 5642821.49 frames. ], batch size: 91, lr: 3.20e-02, grad_scale: 2.0
+2023-02-28 13:22:01,016 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 2.022e+03 2.338e+03 3.082e+03 6.948e+03, threshold=4.675e+03, percent-clipped=3.0
+2023-02-28 13:22:15,718 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=11137.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:22:26,355 INFO [train.py:968] (0/2) Epoch 1, batch 11150, libri_loss[loss=0.4951, simple_loss=0.5076, pruned_loss=0.2413, over 28688.00 frames. ], tot_loss[loss=0.4723, simple_loss=0.4829, pruned_loss=0.2308, over 5662044.70 frames. ], libri_tot_loss[loss=0.4752, simple_loss=0.4909, pruned_loss=0.2306, over 5720095.51 frames. ], giga_tot_loss[loss=0.4754, simple_loss=0.4843, pruned_loss=0.2333, over 5657197.30 frames. ], batch size: 106, lr: 3.20e-02, grad_scale: 2.0
+2023-02-28 13:22:43,330 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=11167.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:22:54,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11177.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:22:54,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4997, 2.7294, 4.2285, 1.6272], device='cuda:0'), covar=tensor([0.0421, 0.0893, 0.0788, 0.1812], device='cuda:0'), in_proj_covar=tensor([0.0646, 0.0450, 0.0754, 0.0524], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 13:23:08,967 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-02-28 13:23:12,617 INFO [train.py:968] (0/2) Epoch 1, batch 11200, giga_loss[loss=0.6054, simple_loss=0.5557, pruned_loss=0.3275, over 28294.00 frames. ], tot_loss[loss=0.4748, simple_loss=0.4837, pruned_loss=0.233, over 5665933.59 frames. ], libri_tot_loss[loss=0.4743, simple_loss=0.4902, pruned_loss=0.23, over 5724559.02 frames. ], giga_tot_loss[loss=0.4781, simple_loss=0.4852, pruned_loss=0.2355, over 5656734.32 frames. ], batch size: 368, lr: 3.19e-02, grad_scale: 4.0
+2023-02-28 13:23:36,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11222.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:23:38,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.889e+03 2.241e+03 3.073e+03 8.284e+03, threshold=4.482e+03, percent-clipped=9.0
+2023-02-28 13:23:46,909 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-02-28 13:24:03,445 INFO [train.py:968] (0/2) Epoch 1, batch 11250, giga_loss[loss=0.4457, simple_loss=0.465, pruned_loss=0.2132, over 28966.00 frames. ], tot_loss[loss=0.4766, simple_loss=0.4846, pruned_loss=0.2343, over 5663262.32 frames. ], libri_tot_loss[loss=0.4745, simple_loss=0.4904, pruned_loss=0.2301, over 5723899.23 frames. ], giga_tot_loss[loss=0.479, simple_loss=0.4856, pruned_loss=0.2362, over 5656411.27 frames. ], batch size: 227, lr: 3.19e-02, grad_scale: 4.0
+2023-02-28 13:24:35,110 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=11278.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 13:24:52,400 INFO [train.py:968] (0/2) Epoch 1, batch 11300, giga_loss[loss=0.4586, simple_loss=0.4738, pruned_loss=0.2218, over 28816.00 frames. ], tot_loss[loss=0.4764, simple_loss=0.4846, pruned_loss=0.2341, over 5666297.07 frames. ], libri_tot_loss[loss=0.4739, simple_loss=0.4899, pruned_loss=0.2297, over 5727077.65 frames. ], giga_tot_loss[loss=0.4789, simple_loss=0.4857, pruned_loss=0.236, over 5656639.65 frames. ], batch size: 119, lr: 3.18e-02, grad_scale: 4.0
+2023-02-28 13:25:12,579 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11320.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:25:14,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11323.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:25:15,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.941e+02 1.836e+03 2.485e+03 3.397e+03 5.887e+03, threshold=4.971e+03, percent-clipped=8.0
+2023-02-28 13:25:26,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0888, 0.5152, 1.0585, 0.0161], device='cuda:0'), covar=tensor([0.0449, 0.0568, 0.0485, 0.1171], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0622, 0.0635, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 13:25:41,587 INFO [train.py:968] (0/2) Epoch 1, batch 11350, giga_loss[loss=0.4824, simple_loss=0.4911, pruned_loss=0.2369, over 28614.00 frames. ], tot_loss[loss=0.4788, simple_loss=0.4864, pruned_loss=0.2356, over 5660913.09 frames. ], libri_tot_loss[loss=0.474, simple_loss=0.4901, pruned_loss=0.2297, over 5720387.97 frames. ], giga_tot_loss[loss=0.4807, simple_loss=0.487, pruned_loss=0.2372, over 5659416.07 frames. ], batch size: 307, lr: 3.17e-02, grad_scale: 4.0
+2023-02-28 13:25:43,989 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11352.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:25:57,514 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11365.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:26:00,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11368.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:26:27,811 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11397.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:26:29,640 INFO [train.py:968] (0/2) Epoch 1, batch 11400, libri_loss[loss=0.4771, simple_loss=0.5048, pruned_loss=0.2247, over 29550.00 frames. ], tot_loss[loss=0.4785, simple_loss=0.4869, pruned_loss=0.235, over 5667508.10 frames. ], libri_tot_loss[loss=0.4741, simple_loss=0.4904, pruned_loss=0.2297, over 5723483.61 frames. ], giga_tot_loss[loss=0.4799, simple_loss=0.487, pruned_loss=0.2364, over 5662211.14 frames. ], batch size: 84, lr: 3.17e-02, grad_scale: 4.0
+2023-02-28 13:26:39,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4658, 1.5533, 1.0126, 1.9635], device='cuda:0'), covar=tensor([0.1140, 0.1444, 0.1133, 0.1252], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0807, 0.0818, 0.0779], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0005], device='cuda:0')
+2023-02-28 13:26:45,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2398, 1.5136, 1.5661, 1.3265], device='cuda:0'), covar=tensor([0.1074, 0.1130, 0.0872, 0.1539], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0548, 0.0454, 0.0556], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0009], device='cuda:0')
+2023-02-28 13:26:52,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 2.144e+03 2.587e+03 3.559e+03 9.216e+03, threshold=5.173e+03, percent-clipped=7.0
+2023-02-28 13:26:57,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0793, 1.1744, 0.9214, 1.3749], device='cuda:0'), covar=tensor([0.1011, 0.1182, 0.0923, 0.1095], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0803, 0.0825, 0.0788], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0005], device='cuda:0')
+2023-02-28 13:27:13,674 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-02-28 13:27:17,945 INFO [train.py:968] (0/2) Epoch 1, batch 11450, giga_loss[loss=0.4751, simple_loss=0.4771, pruned_loss=0.2365, over 28869.00 frames. ], tot_loss[loss=0.4802, simple_loss=0.4873, pruned_loss=0.2365, over 5666031.64 frames. ], libri_tot_loss[loss=0.4728, simple_loss=0.4896, pruned_loss=0.2288, over 5727939.93 frames. ], giga_tot_loss[loss=0.4827, simple_loss=0.4881, pruned_loss=0.2387, over 5656273.03 frames. ], batch size: 99, lr: 3.16e-02, grad_scale: 4.0
+2023-02-28 13:27:51,521 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 13:28:05,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0775, 1.2201, 0.9807, 0.9042], device='cuda:0'), covar=tensor([0.1010, 0.1263, 0.0955, 0.1105], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0806, 0.0833, 0.0796], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0005], device='cuda:0')
+2023-02-28 13:28:06,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9711, 0.8236, 1.0611, 0.5772], device='cuda:0'), covar=tensor([0.0430, 0.0281, 0.0286, 0.0450], device='cuda:0'), in_proj_covar=tensor([0.0684, 0.0499, 0.0549, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 13:28:07,984 INFO [train.py:968] (0/2) Epoch 1, batch 11500, giga_loss[loss=0.4696, simple_loss=0.4807, pruned_loss=0.2293, over 28906.00 frames. ], tot_loss[loss=0.4792, simple_loss=0.486, pruned_loss=0.2362, over 5661041.81 frames. ], libri_tot_loss[loss=0.4727, simple_loss=0.4895, pruned_loss=0.2287, over 5728854.23 frames. ], giga_tot_loss[loss=0.4813, simple_loss=0.4867, pruned_loss=0.2379, over 5652332.29 frames. ], batch size: 145, lr: 3.16e-02, grad_scale: 4.0
+2023-02-28 13:28:20,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11512.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:28:32,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.543e+02 1.563e+03 1.924e+03 2.329e+03 5.313e+03, threshold=3.849e+03, percent-clipped=1.0
+2023-02-28 13:28:50,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11542.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:28:59,261 INFO [train.py:968] (0/2) Epoch 1, batch 11550, giga_loss[loss=0.5419, simple_loss=0.5076, pruned_loss=0.2881, over 23568.00 frames. ], tot_loss[loss=0.4766, simple_loss=0.485, pruned_loss=0.2342, over 5667396.74 frames. ], libri_tot_loss[loss=0.4724, simple_loss=0.4893, pruned_loss=0.2285, over 5730666.98 frames. ], giga_tot_loss[loss=0.4786, simple_loss=0.4856, pruned_loss=0.2358, over 5658372.75 frames. ], batch size: 705, lr: 3.15e-02, grad_scale: 4.0
+2023-02-28 13:29:15,408 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=11567.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:29:35,589 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.91 vs. limit=5.0
+2023-02-28 13:29:43,768 INFO [train.py:968] (0/2) Epoch 1, batch 11600, giga_loss[loss=0.427, simple_loss=0.4617, pruned_loss=0.1961, over 28406.00 frames. ], tot_loss[loss=0.476, simple_loss=0.4856, pruned_loss=0.2332, over 5672284.18 frames. ], libri_tot_loss[loss=0.4722, simple_loss=0.4894, pruned_loss=0.2282, over 5734938.18 frames. ], giga_tot_loss[loss=0.4779, simple_loss=0.4859, pruned_loss=0.235, over 5659094.63 frames. ], batch size: 71, lr: 3.15e-02, grad_scale: 8.0
+2023-02-28 13:29:52,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-02-28 13:30:09,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.655e+02 1.840e+03 2.294e+03 2.848e+03 1.432e+04, threshold=4.588e+03, percent-clipped=11.0
+2023-02-28 13:30:35,056 INFO [train.py:968] (0/2) Epoch 1, batch 11650, giga_loss[loss=0.4227, simple_loss=0.4544, pruned_loss=0.1955, over 28798.00 frames. ], tot_loss[loss=0.4784, simple_loss=0.4872, pruned_loss=0.2349, over 5685282.27 frames. ], libri_tot_loss[loss=0.4715, simple_loss=0.489, pruned_loss=0.2277, over 5739382.79 frames. ], giga_tot_loss[loss=0.4807, simple_loss=0.4877, pruned_loss=0.2369, over 5669363.15 frames. ], batch size: 66, lr: 3.14e-02, grad_scale: 2.0
+2023-02-28 13:30:38,046 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11653.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:30:39,270 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11655.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:30:41,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11658.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:31:08,759 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-02-28 13:31:12,634 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11685.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:31:16,494 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11687.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:31:17,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11688.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:31:28,200 INFO [train.py:968] (0/2) Epoch 1, batch 11700, libri_loss[loss=0.475, simple_loss=0.5003, pruned_loss=0.2249, over 25508.00 frames. ], tot_loss[loss=0.4858, simple_loss=0.4913, pruned_loss=0.2401, over 5675407.89 frames. ], libri_tot_loss[loss=0.4714, simple_loss=0.4891, pruned_loss=0.2276, over 5737685.39 frames. ], giga_tot_loss[loss=0.4879, simple_loss=0.4917, pruned_loss=0.242, over 5663936.62 frames. ], batch size: 136, lr: 3.13e-02, grad_scale: 2.0
+2023-02-28 13:31:43,666 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11717.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:31:51,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.188e+02 1.955e+03 2.494e+03 3.275e+03 6.949e+03, threshold=4.987e+03, percent-clipped=11.0
+2023-02-28 13:31:54,487 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-02-28 13:32:06,372 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6298, 1.4185, 1.3603, 1.2467], device='cuda:0'), covar=tensor([0.0621, 0.0932, 0.0869, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0576, 0.0938, 0.0694, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 13:32:16,498 INFO [train.py:968] (0/2) Epoch 1, batch 11750, giga_loss[loss=0.5076, simple_loss=0.5107, pruned_loss=0.2523, over 28893.00 frames. ], tot_loss[loss=0.4836, simple_loss=0.4902, pruned_loss=0.2385, over 5685067.96 frames. ], libri_tot_loss[loss=0.4709, simple_loss=0.4888, pruned_loss=0.2272, over 5739257.19 frames. ], giga_tot_loss[loss=0.4857, simple_loss=0.4907, pruned_loss=0.2404, over 5674199.47 frames. ], batch size: 174, lr: 3.13e-02, grad_scale: 2.0
+2023-02-28 13:32:59,352 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11796.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 13:33:03,388 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11799.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:33:03,807 INFO [train.py:968] (0/2) Epoch 1, batch 11800, giga_loss[loss=0.4315, simple_loss=0.4651, pruned_loss=0.1989, over 28881.00 frames. ], tot_loss[loss=0.4833, simple_loss=0.4906, pruned_loss=0.238, over 5677307.49 frames. ], libri_tot_loss[loss=0.4703, simple_loss=0.4883, pruned_loss=0.2268, over 5734295.68 frames. ], giga_tot_loss[loss=0.4859, simple_loss=0.4915, pruned_loss=0.2402, over 5670903.63 frames. ], batch size: 186, lr: 3.12e-02, grad_scale: 2.0
+2023-02-28 13:33:27,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.113e+03 1.774e+03 2.184e+03 2.821e+03 6.595e+03, threshold=4.367e+03, percent-clipped=1.0
+2023-02-28 13:33:29,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11828.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:33:52,777 INFO [train.py:968] (0/2) Epoch 1, batch 11850, giga_loss[loss=0.5035, simple_loss=0.4761, pruned_loss=0.2654, over 23547.00 frames. ], tot_loss[loss=0.4797, simple_loss=0.4889, pruned_loss=0.2353, over 5667582.80 frames. ], libri_tot_loss[loss=0.4701, simple_loss=0.4882, pruned_loss=0.2266, over 5737133.90 frames. ], giga_tot_loss[loss=0.4822, simple_loss=0.4897, pruned_loss=0.2374, over 5658921.37 frames. ], batch size: 705, lr: 3.12e-02, grad_scale: 2.0
+2023-02-28 13:34:17,191 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-02-28 13:34:32,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9748, 2.0093, 3.6693, 2.1033], device='cuda:0'), covar=tensor([0.1297, 0.0918, 0.0272, 0.0577], device='cuda:0'), in_proj_covar=tensor([0.0507, 0.0439, 0.0552, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 13:34:41,192 INFO [train.py:968] (0/2) Epoch 1, batch 11900, giga_loss[loss=0.4913, simple_loss=0.491, pruned_loss=0.2457, over 28858.00 frames. ], tot_loss[loss=0.4753, simple_loss=0.4861, pruned_loss=0.2322, over 5671101.64 frames. ], libri_tot_loss[loss=0.4698, simple_loss=0.4882, pruned_loss=0.2264, over 5730329.04 frames. ], giga_tot_loss[loss=0.4776, simple_loss=0.4868, pruned_loss=0.2342, over 5668737.65 frames. ], batch size: 199, lr: 3.11e-02, grad_scale: 2.0
+2023-02-28 13:34:44,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8473, 2.3880, 3.6090, 1.6673], device='cuda:0'), covar=tensor([0.0541, 0.0836, 0.0970, 0.1703], device='cuda:0'), in_proj_covar=tensor([0.0670, 0.0447, 0.0770, 0.0524], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 13:34:47,425 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-02-28 13:35:04,309 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=11925.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:35:04,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.213e+02 1.628e+03 1.988e+03 2.797e+03 6.976e+03, threshold=3.977e+03, percent-clipped=3.0
+2023-02-28 13:35:17,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11942.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:35:27,178 INFO [train.py:968] (0/2) Epoch 1, batch 11950, giga_loss[loss=0.5409, simple_loss=0.5265, pruned_loss=0.2776, over 27894.00 frames. ], tot_loss[loss=0.4746, simple_loss=0.4855, pruned_loss=0.2319, over 5669101.06 frames. ], libri_tot_loss[loss=0.47, simple_loss=0.4884, pruned_loss=0.2264, over 5731965.95 frames. ], giga_tot_loss[loss=0.4764, simple_loss=0.4858, pruned_loss=0.2335, over 5664709.43 frames. ], batch size: 412, lr: 3.11e-02, grad_scale: 2.0
+2023-02-28 13:35:54,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0707, 1.1426, 1.2067, 1.0873], device='cuda:0'), covar=tensor([0.1243, 0.0946, 0.1059, 0.1853], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0551, 0.0457, 0.0561], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-02-28 13:36:19,093 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-12000.pt
+2023-02-28 13:36:19,391 INFO [train.py:968] (0/2) Epoch 1, batch 12000, giga_loss[loss=0.4589, simple_loss=0.4749, pruned_loss=0.2215, over 28759.00 frames. ], tot_loss[loss=0.4763, simple_loss=0.4864, pruned_loss=0.2331, over 5665143.30 frames. ], libri_tot_loss[loss=0.4697, simple_loss=0.4881, pruned_loss=0.2263, over 5730547.12 frames. ], giga_tot_loss[loss=0.478, simple_loss=0.4869, pruned_loss=0.2345, over 5662663.80 frames. ], batch size: 284, lr: 3.10e-02, grad_scale: 4.0
+2023-02-28 13:36:19,396 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 13:36:27,811 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.3639, simple_loss=0.4389, pruned_loss=0.1445, over 944034.00 frames. 
+2023-02-28 13:36:27,812 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 13:36:52,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.740e+03 2.107e+03 2.845e+03 9.088e+03, threshold=4.214e+03, percent-clipped=11.0
+2023-02-28 13:36:54,729 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12028.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:37:16,253 INFO [train.py:968] (0/2) Epoch 1, batch 12050, giga_loss[loss=0.4506, simple_loss=0.4659, pruned_loss=0.2176, over 28550.00 frames. ], tot_loss[loss=0.4779, simple_loss=0.4873, pruned_loss=0.2343, over 5667703.39 frames. ], libri_tot_loss[loss=0.4692, simple_loss=0.4879, pruned_loss=0.2258, over 5732196.24 frames. ], giga_tot_loss[loss=0.4798, simple_loss=0.4878, pruned_loss=0.2359, over 5663194.58 frames. ], batch size: 336, lr: 3.10e-02, grad_scale: 4.0
+2023-02-28 13:37:54,788 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12085.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:37:57,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12088.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:38:00,994 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12092.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:38:07,660 INFO [train.py:968] (0/2) Epoch 1, batch 12100, giga_loss[loss=0.5888, simple_loss=0.5346, pruned_loss=0.3214, over 26552.00 frames. ], tot_loss[loss=0.476, simple_loss=0.4851, pruned_loss=0.2334, over 5665999.43 frames. ], libri_tot_loss[loss=0.469, simple_loss=0.4878, pruned_loss=0.2256, over 5733660.79 frames. ], giga_tot_loss[loss=0.4778, simple_loss=0.4856, pruned_loss=0.235, over 5660105.94 frames. ], batch size: 555, lr: 3.09e-02, grad_scale: 4.0
+2023-02-28 13:38:20,688 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-02-28 13:38:24,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12117.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:38:33,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.846e+02 1.644e+03 1.963e+03 2.534e+03 8.412e+03, threshold=3.926e+03, percent-clipped=2.0
+2023-02-28 13:38:58,410 INFO [train.py:968] (0/2) Epoch 1, batch 12150, giga_loss[loss=0.5262, simple_loss=0.517, pruned_loss=0.2677, over 28664.00 frames. ], tot_loss[loss=0.4769, simple_loss=0.4852, pruned_loss=0.2343, over 5665610.87 frames. ], libri_tot_loss[loss=0.4687, simple_loss=0.4876, pruned_loss=0.2255, over 5736276.72 frames. ], giga_tot_loss[loss=0.4786, simple_loss=0.4857, pruned_loss=0.2357, over 5657728.19 frames. ], batch size: 262, lr: 3.08e-02, grad_scale: 4.0
+2023-02-28 13:39:33,631 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-02-28 13:39:49,223 INFO [train.py:968] (0/2) Epoch 1, batch 12200, giga_loss[loss=0.5237, simple_loss=0.4953, pruned_loss=0.2761, over 23706.00 frames. ], tot_loss[loss=0.4804, simple_loss=0.4873, pruned_loss=0.2367, over 5658524.11 frames. ], libri_tot_loss[loss=0.4685, simple_loss=0.4874, pruned_loss=0.2254, over 5738608.95 frames. ], giga_tot_loss[loss=0.4821, simple_loss=0.4879, pruned_loss=0.2381, over 5649391.78 frames. ], batch size: 705, lr: 3.08e-02, grad_scale: 4.0
+2023-02-28 13:39:57,648 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12209.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:40:06,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.33 vs. limit=5.0
+2023-02-28 13:40:10,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1810, 0.9376, 1.0364, 1.2084], device='cuda:0'), covar=tensor([0.1635, 0.0877, 0.0919, 0.2232], device='cuda:0'), in_proj_covar=tensor([0.0565, 0.0494, 0.0471, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0011, 0.0009, 0.0014], device='cuda:0')
+2023-02-28 13:40:11,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.904e+02 1.805e+03 2.204e+03 2.949e+03 1.001e+04, threshold=4.408e+03, percent-clipped=10.0
+2023-02-28 13:40:24,501 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12239.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:40:35,291 INFO [train.py:968] (0/2) Epoch 1, batch 12250, giga_loss[loss=0.4518, simple_loss=0.4733, pruned_loss=0.2151, over 28630.00 frames. ], tot_loss[loss=0.4787, simple_loss=0.4864, pruned_loss=0.2355, over 5660568.06 frames. ], libri_tot_loss[loss=0.4682, simple_loss=0.4872, pruned_loss=0.2251, over 5742643.55 frames. ], giga_tot_loss[loss=0.4807, simple_loss=0.487, pruned_loss=0.2371, over 5647606.93 frames. ], batch size: 262, lr: 3.07e-02, grad_scale: 4.0
+2023-02-28 13:40:48,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0997, 1.1948, 0.8499, 1.0089], device='cuda:0'), covar=tensor([0.0733, 0.0579, 0.1031, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0558, 0.0719, 0.0691, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 13:41:26,157 INFO [train.py:968] (0/2) Epoch 1, batch 12300, giga_loss[loss=0.5311, simple_loss=0.5124, pruned_loss=0.2749, over 27654.00 frames. ], tot_loss[loss=0.477, simple_loss=0.4853, pruned_loss=0.2344, over 5648881.44 frames. ], libri_tot_loss[loss=0.4679, simple_loss=0.4871, pruned_loss=0.2249, over 5744951.19 frames. ], giga_tot_loss[loss=0.479, simple_loss=0.4859, pruned_loss=0.236, over 5635552.24 frames. ], batch size: 472, lr: 3.07e-02, grad_scale: 4.0
+2023-02-28 13:41:26,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12300.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:41:36,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2113, 1.4665, 1.3494, 1.1845], device='cuda:0'), covar=tensor([0.1694, 0.0861, 0.0920, 0.2257], device='cuda:0'), in_proj_covar=tensor([0.0560, 0.0489, 0.0465, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0011, 0.0009, 0.0014], device='cuda:0')
+2023-02-28 13:41:38,851 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12313.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:41:51,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.162e+03 1.743e+03 2.283e+03 3.155e+03 1.008e+04, threshold=4.565e+03, percent-clipped=11.0
+2023-02-28 13:42:05,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12341.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:42:13,471 INFO [train.py:968] (0/2) Epoch 1, batch 12350, giga_loss[loss=0.469, simple_loss=0.4898, pruned_loss=0.2241, over 28882.00 frames. ], tot_loss[loss=0.477, simple_loss=0.4859, pruned_loss=0.234, over 5646031.23 frames. ], libri_tot_loss[loss=0.4674, simple_loss=0.487, pruned_loss=0.2244, over 5740401.32 frames. ], giga_tot_loss[loss=0.4794, simple_loss=0.4864, pruned_loss=0.2361, over 5636422.24 frames. ], batch size: 186, lr: 3.06e-02, grad_scale: 4.0
+2023-02-28 13:42:36,399 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12377.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:43:00,495 INFO [train.py:968] (0/2) Epoch 1, batch 12400, giga_loss[loss=0.4663, simple_loss=0.4715, pruned_loss=0.2306, over 28931.00 frames. ], tot_loss[loss=0.4766, simple_loss=0.4857, pruned_loss=0.2337, over 5651293.79 frames. ], libri_tot_loss[loss=0.4677, simple_loss=0.4873, pruned_loss=0.2246, over 5741616.77 frames. ], giga_tot_loss[loss=0.4783, simple_loss=0.4859, pruned_loss=0.2353, over 5642012.74 frames. ], batch size: 112, lr: 3.06e-02, grad_scale: 8.0
+2023-02-28 13:43:03,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12403.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:43:28,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.949e+03 2.248e+03 2.937e+03 1.034e+04, threshold=4.497e+03, percent-clipped=6.0
+2023-02-28 13:43:45,195 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12443.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:43:47,903 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12446.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:43:50,241 INFO [train.py:968] (0/2) Epoch 1, batch 12450, giga_loss[loss=0.4113, simple_loss=0.4383, pruned_loss=0.1921, over 28675.00 frames. ], tot_loss[loss=0.4725, simple_loss=0.4832, pruned_loss=0.2309, over 5649389.61 frames. ], libri_tot_loss[loss=0.4677, simple_loss=0.4873, pruned_loss=0.2245, over 5730663.48 frames. ], giga_tot_loss[loss=0.474, simple_loss=0.4833, pruned_loss=0.2324, over 5651450.19 frames. ], batch size: 92, lr: 3.05e-02, grad_scale: 8.0
+2023-02-28 13:44:06,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12467.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:44:08,211 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12469.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:44:14,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12475.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:44:40,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9462, 1.7119, 1.3578, 1.3088], device='cuda:0'), covar=tensor([0.0938, 0.0978, 0.1235, 0.1416], device='cuda:0'), in_proj_covar=tensor([0.0537, 0.0716, 0.0678, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 13:44:41,273 INFO [train.py:968] (0/2) Epoch 1, batch 12500, giga_loss[loss=0.567, simple_loss=0.5294, pruned_loss=0.3023, over 26557.00 frames. ], tot_loss[loss=0.4689, simple_loss=0.4804, pruned_loss=0.2287, over 5647878.90 frames. ], libri_tot_loss[loss=0.4679, simple_loss=0.4875, pruned_loss=0.2247, over 5721958.35 frames. ], giga_tot_loss[loss=0.4698, simple_loss=0.4802, pruned_loss=0.2297, over 5656035.99 frames. ], batch size: 555, lr: 3.05e-02, grad_scale: 8.0
+2023-02-28 13:45:06,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.191e+03 1.855e+03 2.511e+03 3.187e+03 5.170e+03, threshold=5.023e+03, percent-clipped=4.0
+2023-02-28 13:45:27,897 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12546.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:45:30,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12549.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:45:30,583 INFO [train.py:968] (0/2) Epoch 1, batch 12550, giga_loss[loss=0.373, simple_loss=0.403, pruned_loss=0.1715, over 28515.00 frames. ], tot_loss[loss=0.4657, simple_loss=0.4768, pruned_loss=0.2273, over 5649789.50 frames. ], libri_tot_loss[loss=0.4675, simple_loss=0.4872, pruned_loss=0.2244, over 5722849.56 frames. ], giga_tot_loss[loss=0.4668, simple_loss=0.4769, pruned_loss=0.2284, over 5655038.97 frames. ], batch size: 85, lr: 3.04e-02, grad_scale: 4.0
+2023-02-28 13:45:58,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2264, 1.5486, 1.4830, 1.2495], device='cuda:0'), covar=tensor([0.0806, 0.1270, 0.1215, 0.1768], device='cuda:0'), in_proj_covar=tensor([0.0535, 0.0704, 0.0661, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 13:46:00,074 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12578.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:46:06,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12584.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:46:20,584 INFO [train.py:968] (0/2) Epoch 1, batch 12600, giga_loss[loss=0.4352, simple_loss=0.4514, pruned_loss=0.2095, over 28917.00 frames. ], tot_loss[loss=0.4656, simple_loss=0.4754, pruned_loss=0.2279, over 5635304.90 frames. ], libri_tot_loss[loss=0.4672, simple_loss=0.487, pruned_loss=0.2242, over 5714757.14 frames. ], giga_tot_loss[loss=0.4667, simple_loss=0.4755, pruned_loss=0.229, over 5645321.89 frames. ], batch size: 227, lr: 3.04e-02, grad_scale: 4.0
+2023-02-28 13:46:29,301 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12610.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:46:35,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12613.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:46:35,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12614.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:46:47,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.233e+03 1.896e+03 2.473e+03 3.330e+03 1.090e+04, threshold=4.945e+03, percent-clipped=9.0
+2023-02-28 13:47:03,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12642.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:47:08,979 INFO [train.py:968] (0/2) Epoch 1, batch 12650, libri_loss[loss=0.4625, simple_loss=0.4837, pruned_loss=0.2207, over 19992.00 frames. ], tot_loss[loss=0.4663, simple_loss=0.4752, pruned_loss=0.2287, over 5629155.12 frames. ], libri_tot_loss[loss=0.4671, simple_loss=0.4871, pruned_loss=0.2241, over 5709433.13 frames. ], giga_tot_loss[loss=0.4672, simple_loss=0.4749, pruned_loss=0.2298, over 5640576.43 frames. ], batch size: 187, lr: 3.03e-02, grad_scale: 4.0
+2023-02-28 13:47:31,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7906, 2.3825, 3.5835, 1.7727], device='cuda:0'), covar=tensor([0.0629, 0.0857, 0.0964, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.0680, 0.0440, 0.0787, 0.0528], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 13:47:45,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12688.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:47:49,032 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-02-28 13:47:55,340 INFO [train.py:968] (0/2) Epoch 1, batch 12700, giga_loss[loss=0.4744, simple_loss=0.4836, pruned_loss=0.2326, over 28605.00 frames. ], tot_loss[loss=0.4633, simple_loss=0.4735, pruned_loss=0.2266, over 5637754.25 frames. ], libri_tot_loss[loss=0.4659, simple_loss=0.4862, pruned_loss=0.2232, over 5710864.62 frames. ], giga_tot_loss[loss=0.4652, simple_loss=0.4737, pruned_loss=0.2284, over 5643032.50 frames. ], batch size: 336, lr: 3.03e-02, grad_scale: 4.0
+2023-02-28 13:48:10,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12716.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:21,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.775e+03 2.050e+03 2.717e+03 9.742e+03, threshold=4.100e+03, percent-clipped=4.0
+2023-02-28 13:48:22,057 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12727.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:24,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12730.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:44,295 INFO [train.py:968] (0/2) Epoch 1, batch 12750, giga_loss[loss=0.4327, simple_loss=0.4424, pruned_loss=0.2115, over 26707.00 frames. ], tot_loss[loss=0.4565, simple_loss=0.4699, pruned_loss=0.2215, over 5644545.15 frames. ], libri_tot_loss[loss=0.4639, simple_loss=0.4845, pruned_loss=0.2221, over 5717206.48 frames. ], giga_tot_loss[loss=0.4596, simple_loss=0.471, pruned_loss=0.2241, over 5640751.88 frames. ], batch size: 555, lr: 3.02e-02, grad_scale: 4.0
+2023-02-28 13:48:44,761 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-02-28 13:48:47,293 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12752.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:50,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12757.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:51,864 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12759.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:52,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12760.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:49:21,412 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12789.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:49:32,110 INFO [train.py:968] (0/2) Epoch 1, batch 12800, giga_loss[loss=0.4261, simple_loss=0.4593, pruned_loss=0.1965, over 28912.00 frames. ], tot_loss[loss=0.4476, simple_loss=0.4646, pruned_loss=0.2152, over 5649947.46 frames. ], libri_tot_loss[loss=0.4625, simple_loss=0.4833, pruned_loss=0.2213, over 5723434.49 frames. ], giga_tot_loss[loss=0.451, simple_loss=0.4661, pruned_loss=0.2179, over 5638913.96 frames. ], batch size: 186, lr: 3.02e-02, grad_scale: 8.0
+2023-02-28 13:49:58,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.716e+02 1.818e+03 2.555e+03 3.556e+03 8.421e+03, threshold=5.111e+03, percent-clipped=12.0
+2023-02-28 13:50:01,160 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12831.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:06,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12834.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:09,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12839.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:14,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12844.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:21,591 INFO [train.py:968] (0/2) Epoch 1, batch 12850, libri_loss[loss=0.4281, simple_loss=0.4623, pruned_loss=0.197, over 29379.00 frames. ], tot_loss[loss=0.4378, simple_loss=0.4586, pruned_loss=0.2085, over 5653035.43 frames. ], libri_tot_loss[loss=0.4612, simple_loss=0.4822, pruned_loss=0.2205, over 5727866.56 frames. ], giga_tot_loss[loss=0.4411, simple_loss=0.4601, pruned_loss=0.2111, over 5637924.13 frames. ], batch size: 92, lr: 3.01e-02, grad_scale: 4.0
+2023-02-28 13:50:31,783 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12859.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:34,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12862.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:36,494 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12863.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:51:04,299 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12891.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:51:09,246 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12895.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:51:11,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12898.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:51:14,600 INFO [train.py:968] (0/2) Epoch 1, batch 12900, giga_loss[loss=0.3357, simple_loss=0.3661, pruned_loss=0.1527, over 24106.00 frames. ], tot_loss[loss=0.4294, simple_loss=0.4528, pruned_loss=0.203, over 5632835.15 frames. ], libri_tot_loss[loss=0.4608, simple_loss=0.4818, pruned_loss=0.2203, over 5711065.70 frames. ], giga_tot_loss[loss=0.4319, simple_loss=0.4539, pruned_loss=0.205, over 5635257.39 frames. ], batch size: 705, lr: 3.01e-02, grad_scale: 4.0
+2023-02-28 13:51:41,332 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12927.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:51:41,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.470e+02 1.603e+03 1.948e+03 2.582e+03 4.131e+03, threshold=3.897e+03, percent-clipped=1.0
+2023-02-28 13:51:43,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1861, 1.6721, 1.3930, 1.3016], device='cuda:0'), covar=tensor([0.0872, 0.1172, 0.1149, 0.1553], device='cuda:0'), in_proj_covar=tensor([0.0551, 0.0705, 0.0679, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 13:52:06,606 INFO [train.py:968] (0/2) Epoch 1, batch 12950, giga_loss[loss=0.3969, simple_loss=0.4476, pruned_loss=0.1731, over 28830.00 frames. ], tot_loss[loss=0.4191, simple_loss=0.4474, pruned_loss=0.1954, over 5637512.81 frames. ], libri_tot_loss[loss=0.4598, simple_loss=0.4812, pruned_loss=0.2197, over 5714164.70 frames. ], giga_tot_loss[loss=0.4215, simple_loss=0.4484, pruned_loss=0.1973, over 5635704.27 frames. ], batch size: 284, lr: 3.00e-02, grad_scale: 4.0
+2023-02-28 13:52:42,513 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12987.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:52:45,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12990.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:52:55,014 INFO [train.py:968] (0/2) Epoch 1, batch 13000, giga_loss[loss=0.3855, simple_loss=0.4332, pruned_loss=0.1689, over 28825.00 frames. ], tot_loss[loss=0.4129, simple_loss=0.4448, pruned_loss=0.1905, over 5658782.52 frames. ], libri_tot_loss[loss=0.4573, simple_loss=0.4791, pruned_loss=0.2181, over 5719852.39 frames. ], giga_tot_loss[loss=0.4157, simple_loss=0.4464, pruned_loss=0.1925, over 5650285.49 frames. ], batch size: 227, lr: 3.00e-02, grad_scale: 2.0
+2023-02-28 13:53:16,628 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13019.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:53:26,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.678e+03 2.069e+03 2.498e+03 6.596e+03, threshold=4.137e+03, percent-clipped=8.0
+2023-02-28 13:53:40,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0738, 1.1523, 0.9750, 1.2262], device='cuda:0'), covar=tensor([0.1436, 0.1388, 0.1200, 0.1561], device='cuda:0'), in_proj_covar=tensor([0.0833, 0.0772, 0.0851, 0.0837], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 13:53:46,865 INFO [train.py:968] (0/2) Epoch 1, batch 13050, giga_loss[loss=0.4173, simple_loss=0.4502, pruned_loss=0.1922, over 28541.00 frames. ], tot_loss[loss=0.4118, simple_loss=0.4438, pruned_loss=0.1899, over 5651290.34 frames. ], libri_tot_loss[loss=0.456, simple_loss=0.478, pruned_loss=0.2174, over 5721060.72 frames. ], giga_tot_loss[loss=0.4142, simple_loss=0.4452, pruned_loss=0.1915, over 5642100.44 frames. ], batch size: 336, lr: 2.99e-02, grad_scale: 2.0
+2023-02-28 13:54:34,885 INFO [train.py:968] (0/2) Epoch 1, batch 13100, giga_loss[loss=0.3567, simple_loss=0.4118, pruned_loss=0.1508, over 28664.00 frames. ], tot_loss[loss=0.4082, simple_loss=0.4412, pruned_loss=0.1876, over 5664579.51 frames. ], libri_tot_loss[loss=0.4543, simple_loss=0.4764, pruned_loss=0.2164, over 5726875.30 frames. ], giga_tot_loss[loss=0.4101, simple_loss=0.4426, pruned_loss=0.1888, over 5649468.57 frames. ], batch size: 242, lr: 2.99e-02, grad_scale: 2.0
+2023-02-28 13:54:43,399 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.09 vs. limit=2.0
+2023-02-28 13:54:53,563 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.90 vs. limit=5.0
+2023-02-28 13:55:03,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.146e+02 1.801e+03 2.283e+03 3.050e+03 5.742e+03, threshold=4.566e+03, percent-clipped=8.0
+2023-02-28 13:55:24,010 INFO [train.py:968] (0/2) Epoch 1, batch 13150, giga_loss[loss=0.4756, simple_loss=0.4835, pruned_loss=0.2339, over 28563.00 frames. ], tot_loss[loss=0.4025, simple_loss=0.4361, pruned_loss=0.1845, over 5650932.20 frames. ], libri_tot_loss[loss=0.4518, simple_loss=0.4744, pruned_loss=0.215, over 5731746.82 frames. ], giga_tot_loss[loss=0.4049, simple_loss=0.4379, pruned_loss=0.1859, over 5632565.75 frames. ], batch size: 336, lr: 2.98e-02, grad_scale: 2.0
+2023-02-28 13:55:31,748 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=13156.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:56:18,329 INFO [train.py:968] (0/2) Epoch 1, batch 13200, giga_loss[loss=0.4225, simple_loss=0.45, pruned_loss=0.1975, over 27893.00 frames. ], tot_loss[loss=0.4037, simple_loss=0.4369, pruned_loss=0.1853, over 5648260.09 frames. ], libri_tot_loss[loss=0.4517, simple_loss=0.4743, pruned_loss=0.2149, over 5732610.08 frames. ], giga_tot_loss[loss=0.4054, simple_loss=0.4383, pruned_loss=0.1863, over 5632846.12 frames. ], batch size: 412, lr: 2.98e-02, grad_scale: 4.0
+2023-02-28 13:56:32,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13214.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:56:46,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.726e+03 2.220e+03 2.959e+03 7.474e+03, threshold=4.440e+03, percent-clipped=8.0
+2023-02-28 13:57:07,150 INFO [train.py:968] (0/2) Epoch 1, batch 13250, giga_loss[loss=0.4355, simple_loss=0.4412, pruned_loss=0.2149, over 26831.00 frames. ], tot_loss[loss=0.4042, simple_loss=0.4369, pruned_loss=0.1858, over 5645405.39 frames. ], libri_tot_loss[loss=0.4505, simple_loss=0.473, pruned_loss=0.2143, over 5728827.86 frames. ], giga_tot_loss[loss=0.405, simple_loss=0.4379, pruned_loss=0.186, over 5634016.06 frames. ], batch size: 555, lr: 2.97e-02, grad_scale: 4.0
+2023-02-28 13:57:53,737 INFO [train.py:968] (0/2) Epoch 1, batch 13300, giga_loss[loss=0.3623, simple_loss=0.4148, pruned_loss=0.1549, over 28948.00 frames. ], tot_loss[loss=0.3977, simple_loss=0.4326, pruned_loss=0.1814, over 5652853.10 frames. ], libri_tot_loss[loss=0.4484, simple_loss=0.4711, pruned_loss=0.2131, over 5732785.89 frames. ], giga_tot_loss[loss=0.3985, simple_loss=0.4337, pruned_loss=0.1816, over 5637195.35 frames. ], batch size: 186, lr: 2.97e-02, grad_scale: 4.0
+2023-02-28 13:58:24,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.756e+02 1.528e+03 2.341e+03 3.001e+03 1.204e+04, threshold=4.682e+03, percent-clipped=10.0
+2023-02-28 13:58:46,284 INFO [train.py:968] (0/2) Epoch 1, batch 13350, giga_loss[loss=0.3349, simple_loss=0.3981, pruned_loss=0.1358, over 28944.00 frames. ], tot_loss[loss=0.391, simple_loss=0.4283, pruned_loss=0.1769, over 5655061.96 frames. ], libri_tot_loss[loss=0.4474, simple_loss=0.4705, pruned_loss=0.2125, over 5736215.44 frames. ], giga_tot_loss[loss=0.3915, simple_loss=0.429, pruned_loss=0.177, over 5638193.25 frames. ], batch size: 164, lr: 2.96e-02, grad_scale: 4.0
+2023-02-28 13:58:54,550 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13357.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:58:58,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13360.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:59:08,165 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9752, 1.5093, 1.5734, 1.4642], device='cuda:0'), covar=tensor([0.0634, 0.1472, 0.0959, 0.1151], device='cuda:0'), in_proj_covar=tensor([0.0562, 0.0878, 0.0665, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 13:59:25,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4967, 1.2865, 1.8199, 0.3768], device='cuda:0'), covar=tensor([0.0483, 0.0505, 0.0494, 0.1012], device='cuda:0'), in_proj_covar=tensor([0.0730, 0.0680, 0.0711, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 13:59:26,102 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13389.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:59:30,879 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.02 vs. limit=2.0
+2023-02-28 13:59:37,502 INFO [train.py:968] (0/2) Epoch 1, batch 13400, giga_loss[loss=0.3512, simple_loss=0.3958, pruned_loss=0.1533, over 27996.00 frames. ], tot_loss[loss=0.3869, simple_loss=0.4248, pruned_loss=0.1745, over 5662653.82 frames. ], libri_tot_loss[loss=0.4465, simple_loss=0.4697, pruned_loss=0.2119, over 5740170.47 frames. ], giga_tot_loss[loss=0.3863, simple_loss=0.4246, pruned_loss=0.174, over 5642926.22 frames. ], batch size: 412, lr: 2.96e-02, grad_scale: 4.0
+2023-02-28 13:59:46,207 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-02-28 14:00:03,018 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 14:00:08,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.556e+02 1.647e+03 2.194e+03 2.853e+03 1.301e+04, threshold=4.387e+03, percent-clipped=6.0
+2023-02-28 14:00:31,340 INFO [train.py:968] (0/2) Epoch 1, batch 13450, giga_loss[loss=0.4254, simple_loss=0.4358, pruned_loss=0.2075, over 26432.00 frames. ], tot_loss[loss=0.3874, simple_loss=0.4242, pruned_loss=0.1753, over 5661700.85 frames. ], libri_tot_loss[loss=0.446, simple_loss=0.4693, pruned_loss=0.2117, over 5738592.63 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4235, pruned_loss=0.1743, over 5645362.80 frames. ], batch size: 555, lr: 2.95e-02, grad_scale: 4.0
+2023-02-28 14:00:36,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 14:01:26,360 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8180, 1.9169, 3.6161, 2.1477], device='cuda:0'), covar=tensor([0.1433, 0.0958, 0.0280, 0.0585], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0432, 0.0531, 0.0373], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 14:01:27,287 INFO [train.py:968] (0/2) Epoch 1, batch 13500, libri_loss[loss=0.4218, simple_loss=0.4535, pruned_loss=0.195, over 29400.00 frames. ], tot_loss[loss=0.3865, simple_loss=0.4231, pruned_loss=0.175, over 5648874.21 frames. ], libri_tot_loss[loss=0.4459, simple_loss=0.4692, pruned_loss=0.2116, over 5739299.66 frames. ], giga_tot_loss[loss=0.3853, simple_loss=0.4224, pruned_loss=0.1741, over 5634932.77 frames. ], batch size: 92, lr: 2.95e-02, grad_scale: 4.0
+2023-02-28 14:02:05,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.619e+03 2.070e+03 2.771e+03 6.489e+03, threshold=4.140e+03, percent-clipped=7.0
+2023-02-28 14:02:07,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13531.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:02:28,125 INFO [train.py:968] (0/2) Epoch 1, batch 13550, giga_loss[loss=0.4061, simple_loss=0.4483, pruned_loss=0.1819, over 28580.00 frames. ], tot_loss[loss=0.3883, simple_loss=0.4259, pruned_loss=0.1753, over 5657508.12 frames. ], libri_tot_loss[loss=0.4449, simple_loss=0.4684, pruned_loss=0.211, over 5740824.19 frames. ], giga_tot_loss[loss=0.3874, simple_loss=0.4255, pruned_loss=0.1746, over 5643979.10 frames. ], batch size: 336, lr: 2.94e-02, grad_scale: 4.0
+2023-02-28 14:03:15,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7667, 2.5288, 3.4757, 1.9420], device='cuda:0'), covar=tensor([0.0586, 0.0807, 0.0957, 0.1680], device='cuda:0'), in_proj_covar=tensor([0.0620, 0.0428, 0.0719, 0.0502], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0004, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 14:03:20,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0831, 1.2027, 0.9805, 1.4091], device='cuda:0'), covar=tensor([0.1247, 0.1320, 0.1085, 0.1322], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0751, 0.0819, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 14:03:31,259 INFO [train.py:968] (0/2) Epoch 1, batch 13600, giga_loss[loss=0.3432, simple_loss=0.4, pruned_loss=0.1432, over 28925.00 frames. ], tot_loss[loss=0.3905, simple_loss=0.4287, pruned_loss=0.1762, over 5649458.87 frames. ], libri_tot_loss[loss=0.4444, simple_loss=0.468, pruned_loss=0.2107, over 5731233.46 frames. ], giga_tot_loss[loss=0.3896, simple_loss=0.4282, pruned_loss=0.1755, over 5645961.48 frames. ], batch size: 199, lr: 2.94e-02, grad_scale: 8.0
+2023-02-28 14:04:04,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.213e+03 1.895e+03 2.422e+03 3.360e+03 8.495e+03, threshold=4.843e+03, percent-clipped=16.0
+2023-02-28 14:04:26,064 INFO [train.py:968] (0/2) Epoch 1, batch 13650, giga_loss[loss=0.4044, simple_loss=0.4408, pruned_loss=0.184, over 29049.00 frames. ], tot_loss[loss=0.3925, simple_loss=0.4303, pruned_loss=0.1774, over 5656683.65 frames. ], libri_tot_loss[loss=0.4433, simple_loss=0.4672, pruned_loss=0.21, over 5726593.53 frames. ], giga_tot_loss[loss=0.3908, simple_loss=0.4292, pruned_loss=0.1762, over 5655482.34 frames. ], batch size: 285, lr: 2.93e-02, grad_scale: 4.0
+2023-02-28 14:04:58,601 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13674.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:05:00,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13677.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:05:17,642 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 14:05:23,614 INFO [train.py:968] (0/2) Epoch 1, batch 13700, giga_loss[loss=0.3776, simple_loss=0.4183, pruned_loss=0.1685, over 28726.00 frames. ], tot_loss[loss=0.3868, simple_loss=0.426, pruned_loss=0.1738, over 5662106.95 frames. ], libri_tot_loss[loss=0.4412, simple_loss=0.4655, pruned_loss=0.2087, over 5729508.89 frames. ], giga_tot_loss[loss=0.3856, simple_loss=0.4254, pruned_loss=0.1729, over 5656420.75 frames. ], batch size: 60, lr: 2.93e-02, grad_scale: 4.0
+2023-02-28 14:05:31,743 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13706.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:06:04,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.099e+03 1.472e+03 1.852e+03 2.445e+03 5.916e+03, threshold=3.703e+03, percent-clipped=2.0
+2023-02-28 14:06:25,967 INFO [train.py:968] (0/2) Epoch 1, batch 13750, giga_loss[loss=0.3701, simple_loss=0.4295, pruned_loss=0.1554, over 28740.00 frames. ], tot_loss[loss=0.3818, simple_loss=0.4232, pruned_loss=0.1702, over 5665579.64 frames. ], libri_tot_loss[loss=0.4399, simple_loss=0.4645, pruned_loss=0.2079, over 5733184.24 frames. ], giga_tot_loss[loss=0.381, simple_loss=0.423, pruned_loss=0.1695, over 5656280.98 frames. ], batch size: 243, lr: 2.92e-02, grad_scale: 4.0
+2023-02-28 14:06:41,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0138, 2.8803, 2.0060, 1.8921], device='cuda:0'), covar=tensor([0.0665, 0.0499, 0.0510, 0.0349], device='cuda:0'), in_proj_covar=tensor([0.0712, 0.0786, 0.0656, 0.0572], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 14:07:24,865 INFO [train.py:968] (0/2) Epoch 1, batch 13800, libri_loss[loss=0.4376, simple_loss=0.466, pruned_loss=0.2046, over 29547.00 frames. ], tot_loss[loss=0.38, simple_loss=0.422, pruned_loss=0.169, over 5666831.93 frames. ], libri_tot_loss[loss=0.439, simple_loss=0.4637, pruned_loss=0.2074, over 5737215.89 frames. ], giga_tot_loss[loss=0.3782, simple_loss=0.4211, pruned_loss=0.1677, over 5653426.58 frames. ], batch size: 89, lr: 2.92e-02, grad_scale: 4.0
+2023-02-28 14:07:51,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1925, 1.3350, 1.0666, 1.2014], device='cuda:0'), covar=tensor([0.1174, 0.1278, 0.1052, 0.1462], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0784, 0.0836, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 14:08:02,375 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.938e+02 1.907e+03 2.777e+03 3.928e+03 9.906e+03, threshold=5.555e+03, percent-clipped=27.0
+2023-02-28 14:08:26,303 INFO [train.py:968] (0/2) Epoch 1, batch 13850, giga_loss[loss=0.3876, simple_loss=0.4289, pruned_loss=0.1732, over 28639.00 frames. ], tot_loss[loss=0.3794, simple_loss=0.4197, pruned_loss=0.1695, over 5672729.56 frames. ], libri_tot_loss[loss=0.4384, simple_loss=0.4631, pruned_loss=0.2071, over 5738508.21 frames. ], giga_tot_loss[loss=0.3771, simple_loss=0.4186, pruned_loss=0.1679, over 5659266.81 frames. ], batch size: 307, lr: 2.91e-02, grad_scale: 2.0
+2023-02-28 14:09:21,721 INFO [train.py:968] (0/2) Epoch 1, batch 13900, giga_loss[loss=0.3477, simple_loss=0.3979, pruned_loss=0.1487, over 28683.00 frames. ], tot_loss[loss=0.3781, simple_loss=0.4184, pruned_loss=0.1689, over 5666858.53 frames. ], libri_tot_loss[loss=0.4371, simple_loss=0.4622, pruned_loss=0.2062, over 5735310.36 frames. ], giga_tot_loss[loss=0.3753, simple_loss=0.4167, pruned_loss=0.167, over 5656848.40 frames. ], batch size: 262, lr: 2.91e-02, grad_scale: 2.0
+2023-02-28 14:09:57,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.113e+02 1.467e+03 1.854e+03 2.758e+03 2.618e+04, threshold=3.708e+03, percent-clipped=9.0
+2023-02-28 14:10:17,121 INFO [train.py:968] (0/2) Epoch 1, batch 13950, giga_loss[loss=0.3473, simple_loss=0.4054, pruned_loss=0.1446, over 28977.00 frames. ], tot_loss[loss=0.3768, simple_loss=0.4173, pruned_loss=0.1682, over 5666954.10 frames. ], libri_tot_loss[loss=0.4361, simple_loss=0.4612, pruned_loss=0.2058, over 5737652.55 frames. ], giga_tot_loss[loss=0.3738, simple_loss=0.4156, pruned_loss=0.1659, over 5654922.38 frames. ], batch size: 213, lr: 2.90e-02, grad_scale: 2.0
+2023-02-28 14:11:17,406 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-14000.pt
+2023-02-28 14:11:17,706 INFO [train.py:968] (0/2) Epoch 1, batch 14000, giga_loss[loss=0.5072, simple_loss=0.4918, pruned_loss=0.2614, over 26674.00 frames. ], tot_loss[loss=0.3782, simple_loss=0.4193, pruned_loss=0.1686, over 5659021.06 frames. ], libri_tot_loss[loss=0.435, simple_loss=0.4602, pruned_loss=0.2051, over 5740974.78 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.4176, pruned_loss=0.1664, over 5644523.36 frames. ], batch size: 555, lr: 2.90e-02, grad_scale: 4.0
+2023-02-28 14:11:18,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1013, 2.4775, 1.9450, 1.8808], device='cuda:0'), covar=tensor([0.0537, 0.0484, 0.0407, 0.0293], device='cuda:0'), in_proj_covar=tensor([0.0708, 0.0765, 0.0651, 0.0569], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 14:11:37,757 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-02-28 14:11:56,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.889e+02 1.805e+03 2.208e+03 2.967e+03 6.070e+03, threshold=4.417e+03, percent-clipped=13.0
+2023-02-28 14:12:22,404 INFO [train.py:968] (0/2) Epoch 1, batch 14050, giga_loss[loss=0.4308, simple_loss=0.4632, pruned_loss=0.1992, over 28934.00 frames. ], tot_loss[loss=0.3759, simple_loss=0.4181, pruned_loss=0.1669, over 5666524.34 frames. ], libri_tot_loss[loss=0.4347, simple_loss=0.4601, pruned_loss=0.2049, over 5742797.80 frames. ], giga_tot_loss[loss=0.3733, simple_loss=0.4166, pruned_loss=0.165, over 5652771.89 frames. ], batch size: 213, lr: 2.90e-02, grad_scale: 4.0
+2023-02-28 14:13:25,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7156, 1.7962, 1.1594, 1.5849], device='cuda:0'), covar=tensor([0.0912, 0.0913, 0.1415, 0.1156], device='cuda:0'), in_proj_covar=tensor([0.0533, 0.0689, 0.0668, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 14:13:29,433 INFO [train.py:968] (0/2) Epoch 1, batch 14100, giga_loss[loss=0.3129, simple_loss=0.3711, pruned_loss=0.1273, over 29191.00 frames. ], tot_loss[loss=0.3736, simple_loss=0.4159, pruned_loss=0.1656, over 5677584.79 frames. ], libri_tot_loss[loss=0.4339, simple_loss=0.4594, pruned_loss=0.2044, over 5746894.21 frames. ], giga_tot_loss[loss=0.3708, simple_loss=0.4142, pruned_loss=0.1637, over 5661299.87 frames. ], batch size: 107, lr: 2.89e-02, grad_scale: 4.0
+2023-02-28 14:14:04,859 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14130.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:14:05,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.931e+02 1.526e+03 2.000e+03 2.710e+03 7.271e+03, threshold=3.999e+03, percent-clipped=4.0
+2023-02-28 14:14:33,957 INFO [train.py:968] (0/2) Epoch 1, batch 14150, giga_loss[loss=0.3388, simple_loss=0.3577, pruned_loss=0.16, over 24410.00 frames. ], tot_loss[loss=0.3766, simple_loss=0.4189, pruned_loss=0.1671, over 5683376.40 frames. ], libri_tot_loss[loss=0.4331, simple_loss=0.4589, pruned_loss=0.2038, over 5747721.39 frames. ], giga_tot_loss[loss=0.374, simple_loss=0.4173, pruned_loss=0.1654, over 5668618.08 frames. ], batch size: 705, lr: 2.89e-02, grad_scale: 4.0
+2023-02-28 14:15:09,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2664, 1.3142, 1.1246, 1.4527], device='cuda:0'), covar=tensor([0.1263, 0.1137, 0.0983, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0821, 0.0784, 0.0844, 0.0822], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 14:15:37,513 INFO [train.py:968] (0/2) Epoch 1, batch 14200, giga_loss[loss=0.3497, simple_loss=0.4255, pruned_loss=0.1369, over 28951.00 frames. ], tot_loss[loss=0.3787, simple_loss=0.4231, pruned_loss=0.1672, over 5678188.36 frames. ], libri_tot_loss[loss=0.4322, simple_loss=0.4581, pruned_loss=0.2033, over 5748071.83 frames. ], giga_tot_loss[loss=0.376, simple_loss=0.4215, pruned_loss=0.1652, over 5664138.05 frames. ], batch size: 164, lr: 2.88e-02, grad_scale: 4.0
+2023-02-28 14:16:06,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.99 vs. limit=2.0
+2023-02-28 14:16:12,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.754e+03 2.204e+03 2.944e+03 6.102e+03, threshold=4.408e+03, percent-clipped=9.0
+2023-02-28 14:16:18,797 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-02-28 14:16:35,770 INFO [train.py:968] (0/2) Epoch 1, batch 14250, giga_loss[loss=0.4161, simple_loss=0.4515, pruned_loss=0.1903, over 28105.00 frames. ], tot_loss[loss=0.3782, simple_loss=0.4241, pruned_loss=0.1662, over 5682848.20 frames. ], libri_tot_loss[loss=0.4312, simple_loss=0.4575, pruned_loss=0.2026, over 5752566.75 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.4222, pruned_loss=0.1641, over 5665287.16 frames. ], batch size: 412, lr: 2.88e-02, grad_scale: 4.0
+2023-02-28 14:17:08,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8885, 1.8336, 4.3400, 2.6404], device='cuda:0'), covar=tensor([0.1421, 0.0940, 0.0198, 0.0359], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0419, 0.0536, 0.0373], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0007, 0.0004], device='cuda:0')
+2023-02-28 14:17:33,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7858, 2.2356, 1.6543, 1.6390], device='cuda:0'), covar=tensor([0.0803, 0.0760, 0.0601, 0.0429], device='cuda:0'), in_proj_covar=tensor([0.0728, 0.0801, 0.0660, 0.0590], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 14:17:34,477 INFO [train.py:968] (0/2) Epoch 1, batch 14300, giga_loss[loss=0.3429, simple_loss=0.406, pruned_loss=0.1399, over 28901.00 frames. ], tot_loss[loss=0.3769, simple_loss=0.424, pruned_loss=0.165, over 5679960.76 frames. ], libri_tot_loss[loss=0.4305, simple_loss=0.4568, pruned_loss=0.2022, over 5754346.87 frames. ], giga_tot_loss[loss=0.3742, simple_loss=0.4225, pruned_loss=0.1629, over 5663237.50 frames. ], batch size: 174, lr: 2.87e-02, grad_scale: 4.0
+2023-02-28 14:18:13,685 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.179e+02 1.555e+03 2.142e+03 2.610e+03 6.839e+03, threshold=4.284e+03, percent-clipped=6.0
+2023-02-28 14:18:41,593 INFO [train.py:968] (0/2) Epoch 1, batch 14350, giga_loss[loss=0.3588, simple_loss=0.4133, pruned_loss=0.1522, over 29114.00 frames. ], tot_loss[loss=0.375, simple_loss=0.4227, pruned_loss=0.1637, over 5666582.24 frames. ], libri_tot_loss[loss=0.4306, simple_loss=0.457, pruned_loss=0.2023, over 5741907.26 frames. ], giga_tot_loss[loss=0.3723, simple_loss=0.4212, pruned_loss=0.1618, over 5663379.03 frames. ], batch size: 214, lr: 2.87e-02, grad_scale: 4.0
+2023-02-28 14:19:44,219 INFO [train.py:968] (0/2) Epoch 1, batch 14400, giga_loss[loss=0.3869, simple_loss=0.4243, pruned_loss=0.1748, over 28911.00 frames. ], tot_loss[loss=0.375, simple_loss=0.4216, pruned_loss=0.1642, over 5675411.82 frames. ], libri_tot_loss[loss=0.4299, simple_loss=0.4565, pruned_loss=0.2019, over 5745094.43 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4202, pruned_loss=0.1624, over 5668625.30 frames. ], batch size: 186, lr: 2.86e-02, grad_scale: 8.0
+2023-02-28 14:20:29,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.296e+02 1.571e+03 1.969e+03 2.470e+03 6.789e+03, threshold=3.938e+03, percent-clipped=4.0
+2023-02-28 14:21:01,019 INFO [train.py:968] (0/2) Epoch 1, batch 14450, giga_loss[loss=0.3747, simple_loss=0.4233, pruned_loss=0.1631, over 28898.00 frames. ], tot_loss[loss=0.3781, simple_loss=0.4234, pruned_loss=0.1664, over 5687515.15 frames. ], libri_tot_loss[loss=0.4293, simple_loss=0.4561, pruned_loss=0.2014, over 5747763.75 frames. ], giga_tot_loss[loss=0.3759, simple_loss=0.4222, pruned_loss=0.1648, over 5678896.25 frames. ], batch size: 284, lr: 2.86e-02, grad_scale: 8.0
+2023-02-28 14:21:19,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14465.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:21:36,738 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-02-28 14:22:11,438 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14496.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:22:17,094 INFO [train.py:968] (0/2) Epoch 1, batch 14500, giga_loss[loss=0.3288, simple_loss=0.3878, pruned_loss=0.1349, over 29024.00 frames. ], tot_loss[loss=0.3746, simple_loss=0.42, pruned_loss=0.1646, over 5685258.90 frames. ], libri_tot_loss[loss=0.4281, simple_loss=0.4552, pruned_loss=0.2007, over 5749069.05 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4189, pruned_loss=0.1631, over 5675133.94 frames. ], batch size: 285, lr: 2.85e-02, grad_scale: 8.0
+2023-02-28 14:22:24,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14505.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:22:29,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3067, 1.2127, 1.1878, 0.9056], device='cuda:0'), covar=tensor([0.0444, 0.0321, 0.0310, 0.0426], device='cuda:0'), in_proj_covar=tensor([0.0645, 0.0480, 0.0548, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:0')
+2023-02-28 14:22:37,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14513.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 14:23:03,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.485e+02 1.814e+03 2.290e+03 3.148e+03 1.183e+04, threshold=4.580e+03, percent-clipped=14.0
+2023-02-28 14:23:27,451 INFO [train.py:968] (0/2) Epoch 1, batch 14550, libri_loss[loss=0.387, simple_loss=0.414, pruned_loss=0.18, over 29471.00 frames. ], tot_loss[loss=0.3726, simple_loss=0.418, pruned_loss=0.1637, over 5678058.62 frames. ], libri_tot_loss[loss=0.4273, simple_loss=0.4545, pruned_loss=0.2002, over 5748933.00 frames. ], giga_tot_loss[loss=0.3705, simple_loss=0.4169, pruned_loss=0.162, over 5668856.25 frames. ], batch size: 70, lr: 2.85e-02, grad_scale: 4.0
+2023-02-28 14:24:02,656 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-02-28 14:24:23,716 INFO [train.py:968] (0/2) Epoch 1, batch 14600, giga_loss[loss=0.3841, simple_loss=0.4233, pruned_loss=0.1725, over 28081.00 frames. ], tot_loss[loss=0.3729, simple_loss=0.4176, pruned_loss=0.164, over 5689845.34 frames. ], libri_tot_loss[loss=0.4244, simple_loss=0.4525, pruned_loss=0.1983, over 5753971.59 frames. ], giga_tot_loss[loss=0.3706, simple_loss=0.4164, pruned_loss=0.1624, over 5673770.11 frames. ], batch size: 412, lr: 2.85e-02, grad_scale: 4.0
+2023-02-28 14:24:38,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1563, 1.2091, 1.0336, 1.0956], device='cuda:0'), covar=tensor([0.0897, 0.0752, 0.1245, 0.0939], device='cuda:0'), in_proj_covar=tensor([0.0538, 0.0691, 0.0678, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 14:25:05,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.942e+02 1.535e+03 1.985e+03 2.728e+03 6.859e+03, threshold=3.971e+03, percent-clipped=4.0
+2023-02-28 14:25:25,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14648.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:25:26,897 INFO [train.py:968] (0/2) Epoch 1, batch 14650, giga_loss[loss=0.3949, simple_loss=0.4344, pruned_loss=0.1777, over 29068.00 frames. ], tot_loss[loss=0.3754, simple_loss=0.419, pruned_loss=0.1659, over 5687073.57 frames. ], libri_tot_loss[loss=0.423, simple_loss=0.4515, pruned_loss=0.1974, over 5758453.08 frames. ], giga_tot_loss[loss=0.3734, simple_loss=0.418, pruned_loss=0.1644, over 5668310.66 frames. ], batch size: 128, lr: 2.84e-02, grad_scale: 4.0
+2023-02-28 14:25:27,609 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.94 vs. limit=5.0
+2023-02-28 14:25:28,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14651.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:26:03,748 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14680.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:26:28,341 INFO [train.py:968] (0/2) Epoch 1, batch 14700, giga_loss[loss=0.3837, simple_loss=0.4311, pruned_loss=0.1681, over 28928.00 frames. ], tot_loss[loss=0.3795, simple_loss=0.4228, pruned_loss=0.1681, over 5692158.16 frames. ], libri_tot_loss[loss=0.4212, simple_loss=0.45, pruned_loss=0.1964, over 5761569.72 frames. ], giga_tot_loss[loss=0.3779, simple_loss=0.4224, pruned_loss=0.1667, over 5671471.14 frames. ], batch size: 186, lr: 2.84e-02, grad_scale: 4.0
+2023-02-28 14:27:09,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.182e+03 1.787e+03 2.165e+03 2.793e+03 7.151e+03, threshold=4.330e+03, percent-clipped=6.0
+2023-02-28 14:27:13,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2057, 1.2239, 1.0831, 1.1565], device='cuda:0'), covar=tensor([0.1325, 0.1415, 0.1116, 0.1720], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0764, 0.0828, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 14:27:30,900 INFO [train.py:968] (0/2) Epoch 1, batch 14750, giga_loss[loss=0.3618, simple_loss=0.4111, pruned_loss=0.1563, over 28897.00 frames. ], tot_loss[loss=0.3808, simple_loss=0.4224, pruned_loss=0.1696, over 5678957.74 frames. ], libri_tot_loss[loss=0.4207, simple_loss=0.4495, pruned_loss=0.1961, over 5751533.68 frames. ], giga_tot_loss[loss=0.3793, simple_loss=0.422, pruned_loss=0.1683, over 5670423.07 frames. ], batch size: 227, lr: 2.83e-02, grad_scale: 4.0
+2023-02-28 14:27:53,581 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-02-28 14:28:40,362 INFO [train.py:968] (0/2) Epoch 1, batch 14800, giga_loss[loss=0.3824, simple_loss=0.4273, pruned_loss=0.1687, over 28948.00 frames. ], tot_loss[loss=0.3803, simple_loss=0.4218, pruned_loss=0.1694, over 5685499.80 frames. ], libri_tot_loss[loss=0.4201, simple_loss=0.4491, pruned_loss=0.1957, over 5754134.83 frames. ], giga_tot_loss[loss=0.3789, simple_loss=0.4213, pruned_loss=0.1682, over 5675323.67 frames. ], batch size: 199, lr: 2.83e-02, grad_scale: 8.0
+2023-02-28 14:28:47,516 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14806.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:29:18,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.857e+02 1.652e+03 2.107e+03 2.634e+03 7.675e+03, threshold=4.215e+03, percent-clipped=4.0
+2023-02-28 14:29:29,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14840.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:29:42,803 INFO [train.py:968] (0/2) Epoch 1, batch 14850, giga_loss[loss=0.4089, simple_loss=0.4521, pruned_loss=0.1829, over 28593.00 frames. ], tot_loss[loss=0.3818, simple_loss=0.4228, pruned_loss=0.1704, over 5687971.98 frames. ], libri_tot_loss[loss=0.4192, simple_loss=0.4484, pruned_loss=0.1951, over 5755701.94 frames. ], giga_tot_loss[loss=0.3807, simple_loss=0.4225, pruned_loss=0.1694, over 5676988.91 frames. ], batch size: 307, lr: 2.82e-02, grad_scale: 8.0
+2023-02-28 14:30:15,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14871.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:30:21,138 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-02-28 14:30:24,819 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14880.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:30:32,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14888.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 14:30:49,212 INFO [train.py:968] (0/2) Epoch 1, batch 14900, giga_loss[loss=0.4078, simple_loss=0.429, pruned_loss=0.1933, over 26825.00 frames. ], tot_loss[loss=0.3831, simple_loss=0.4249, pruned_loss=0.1707, over 5688830.66 frames. ], libri_tot_loss[loss=0.4181, simple_loss=0.4476, pruned_loss=0.1944, over 5759300.66 frames. ], giga_tot_loss[loss=0.3819, simple_loss=0.4245, pruned_loss=0.1696, over 5674179.92 frames. ], batch size: 555, lr: 2.82e-02, grad_scale: 8.0
+2023-02-28 14:31:46,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.341e+03 2.111e+03 2.681e+03 3.425e+03 8.569e+03, threshold=5.361e+03, percent-clipped=11.0
+2023-02-28 14:32:11,960 INFO [train.py:968] (0/2) Epoch 1, batch 14950, giga_loss[loss=0.3884, simple_loss=0.4168, pruned_loss=0.18, over 26911.00 frames. ], tot_loss[loss=0.3805, simple_loss=0.4237, pruned_loss=0.1687, over 5678250.92 frames. ], libri_tot_loss[loss=0.418, simple_loss=0.4475, pruned_loss=0.1944, over 5755392.13 frames. ], giga_tot_loss[loss=0.3793, simple_loss=0.4232, pruned_loss=0.1676, over 5669647.66 frames. ], batch size: 555, lr: 2.82e-02, grad_scale: 4.0
+2023-02-28 14:33:11,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14983.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:33:18,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14986.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:33:36,482 INFO [train.py:968] (0/2) Epoch 1, batch 15000, giga_loss[loss=0.325, simple_loss=0.3758, pruned_loss=0.1371, over 28919.00 frames. ], tot_loss[loss=0.3779, simple_loss=0.4201, pruned_loss=0.1678, over 5667461.03 frames. ], libri_tot_loss[loss=0.417, simple_loss=0.4467, pruned_loss=0.1938, over 5757541.46 frames. ], giga_tot_loss[loss=0.3772, simple_loss=0.4201, pruned_loss=0.1672, over 5657519.19 frames. ], batch size: 213, lr: 2.81e-02, grad_scale: 4.0
+2023-02-28 14:33:36,487 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 14:33:44,993 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.3082, simple_loss=0.3863, pruned_loss=0.115, over 944034.00 frames. 
+2023-02-28 14:33:44,993 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 14:34:02,957 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15014.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:34:03,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15015.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:34:05,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15017.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:34:26,037 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15031.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 14:34:27,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.592e+02 1.870e+03 2.363e+03 3.393e+03 1.219e+04, threshold=4.726e+03, percent-clipped=7.0
+2023-02-28 14:34:30,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15034.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 14:34:47,202 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15046.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:34:52,422 INFO [train.py:968] (0/2) Epoch 1, batch 15050, giga_loss[loss=0.3047, simple_loss=0.3595, pruned_loss=0.1249, over 28603.00 frames. ], tot_loss[loss=0.3701, simple_loss=0.4127, pruned_loss=0.1638, over 5665097.79 frames. ], libri_tot_loss[loss=0.4169, simple_loss=0.4465, pruned_loss=0.1938, over 5751505.32 frames. ], giga_tot_loss[loss=0.3688, simple_loss=0.4123, pruned_loss=0.1627, over 5660506.36 frames. ], batch size: 307, lr: 2.81e-02, grad_scale: 4.0
+2023-02-28 14:35:10,890 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15063.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 14:35:21,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7330, 1.8967, 3.9029, 2.4781], device='cuda:0'), covar=tensor([0.1411, 0.0911, 0.0235, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0442, 0.0544, 0.0375], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0007, 0.0004], device='cuda:0')
+2023-02-28 14:35:51,702 INFO [train.py:968] (0/2) Epoch 1, batch 15100, libri_loss[loss=0.3649, simple_loss=0.4099, pruned_loss=0.1599, over 29566.00 frames. ], tot_loss[loss=0.3676, simple_loss=0.4101, pruned_loss=0.1625, over 5663890.51 frames. ], libri_tot_loss[loss=0.4161, simple_loss=0.4458, pruned_loss=0.1933, over 5748566.39 frames. ], giga_tot_loss[loss=0.3654, simple_loss=0.409, pruned_loss=0.1609, over 5659713.14 frames. ], batch size: 78, lr: 2.80e-02, grad_scale: 4.0
+2023-02-28 14:36:32,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.216e+02 1.520e+03 2.000e+03 2.720e+03 1.342e+04, threshold=4.000e+03, percent-clipped=4.0
+2023-02-28 14:36:32,812 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=15133.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:36:53,121 INFO [train.py:968] (0/2) Epoch 1, batch 15150, giga_loss[loss=0.3967, simple_loss=0.4338, pruned_loss=0.1798, over 28879.00 frames. ], tot_loss[loss=0.3683, simple_loss=0.4105, pruned_loss=0.1631, over 5662046.38 frames. ], libri_tot_loss[loss=0.4147, simple_loss=0.4447, pruned_loss=0.1924, over 5751184.65 frames. ], giga_tot_loss[loss=0.3668, simple_loss=0.4097, pruned_loss=0.1619, over 5654347.18 frames. ], batch size: 284, lr: 2.80e-02, grad_scale: 4.0
+2023-02-28 14:37:26,437 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15181.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:37:51,135 INFO [train.py:968] (0/2) Epoch 1, batch 15200, giga_loss[loss=0.3165, simple_loss=0.3693, pruned_loss=0.1319, over 27559.00 frames. ], tot_loss[loss=0.3676, simple_loss=0.4104, pruned_loss=0.1624, over 5670732.84 frames. ], libri_tot_loss[loss=0.4142, simple_loss=0.4444, pruned_loss=0.1921, over 5754449.06 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.4095, pruned_loss=0.1612, over 5660169.11 frames. ], batch size: 472, lr: 2.79e-02, grad_scale: 8.0
+2023-02-28 14:38:35,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.233e+02 1.547e+03 2.030e+03 2.651e+03 5.153e+03, threshold=4.061e+03, percent-clipped=6.0
+2023-02-28 14:38:39,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4106, 1.7301, 1.4290, 1.5127], device='cuda:0'), covar=tensor([0.1452, 0.0611, 0.0779, 0.1766], device='cuda:0'), in_proj_covar=tensor([0.0531, 0.0433, 0.0423, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0015], device='cuda:0')
+2023-02-28 14:38:57,272 INFO [train.py:968] (0/2) Epoch 1, batch 15250, giga_loss[loss=0.3254, simple_loss=0.3617, pruned_loss=0.1446, over 24445.00 frames. ], tot_loss[loss=0.3618, simple_loss=0.4063, pruned_loss=0.1587, over 5660319.77 frames. ], libri_tot_loss[loss=0.413, simple_loss=0.4433, pruned_loss=0.1914, over 5758689.43 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4058, pruned_loss=0.1576, over 5645883.53 frames. ], batch size: 705, lr: 2.79e-02, grad_scale: 8.0
+2023-02-28 14:39:01,789 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.40 vs. limit=5.0
+2023-02-28 14:39:02,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15255.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:40:01,064 INFO [train.py:968] (0/2) Epoch 1, batch 15300, giga_loss[loss=0.3145, simple_loss=0.3765, pruned_loss=0.1263, over 29107.00 frames. ], tot_loss[loss=0.3596, simple_loss=0.4051, pruned_loss=0.157, over 5672514.94 frames. ], libri_tot_loss[loss=0.4125, simple_loss=0.4431, pruned_loss=0.1911, over 5759798.29 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.4039, pruned_loss=0.1556, over 5657720.71 frames. ], batch size: 146, lr: 2.79e-02, grad_scale: 8.0
+2023-02-28 14:40:36,442 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15324.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:40:42,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15327.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:40:48,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.076e+02 1.551e+03 2.040e+03 2.779e+03 4.532e+03, threshold=4.080e+03, percent-clipped=4.0
+2023-02-28 14:41:09,795 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-02-28 14:41:12,616 INFO [train.py:968] (0/2) Epoch 1, batch 15350, giga_loss[loss=0.3448, simple_loss=0.3973, pruned_loss=0.1461, over 28864.00 frames. ], tot_loss[loss=0.3585, simple_loss=0.4041, pruned_loss=0.1565, over 5655606.45 frames. ], libri_tot_loss[loss=0.4126, simple_loss=0.4431, pruned_loss=0.1911, over 5750595.95 frames. ], giga_tot_loss[loss=0.3559, simple_loss=0.4024, pruned_loss=0.1547, over 5651155.04 frames. ], batch size: 227, lr: 2.78e-02, grad_scale: 8.0
+2023-02-28 14:41:17,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4596, 2.8405, 4.1160, 1.9639], device='cuda:0'), covar=tensor([0.0431, 0.0747, 0.0877, 0.1892], device='cuda:0'), in_proj_covar=tensor([0.0622, 0.0433, 0.0714, 0.0500], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 14:41:23,357 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15356.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:42:09,661 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-02-28 14:42:23,958 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15398.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:42:26,102 INFO [train.py:968] (0/2) Epoch 1, batch 15400, giga_loss[loss=0.358, simple_loss=0.4125, pruned_loss=0.1518, over 28709.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4041, pruned_loss=0.1559, over 5650224.49 frames. ], libri_tot_loss[loss=0.4126, simple_loss=0.4431, pruned_loss=0.1911, over 5750595.95 frames. ], giga_tot_loss[loss=0.356, simple_loss=0.4029, pruned_loss=0.1545, over 5646759.90 frames. ], batch size: 262, lr: 2.78e-02, grad_scale: 8.0
+2023-02-28 14:42:27,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15401.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:43:04,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15430.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:43:08,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.485e+03 1.869e+03 2.480e+03 5.265e+03, threshold=3.739e+03, percent-clipped=1.0
+2023-02-28 14:43:31,764 INFO [train.py:968] (0/2) Epoch 1, batch 15450, giga_loss[loss=0.3506, simple_loss=0.3985, pruned_loss=0.1514, over 28961.00 frames. ], tot_loss[loss=0.3604, simple_loss=0.4057, pruned_loss=0.1576, over 5662134.12 frames. ], libri_tot_loss[loss=0.4121, simple_loss=0.4428, pruned_loss=0.1908, over 5753856.34 frames. ], giga_tot_loss[loss=0.358, simple_loss=0.4041, pruned_loss=0.156, over 5654316.01 frames. ], batch size: 284, lr: 2.77e-02, grad_scale: 8.0
+2023-02-28 14:44:42,369 INFO [train.py:968] (0/2) Epoch 1, batch 15500, giga_loss[loss=0.3299, simple_loss=0.3986, pruned_loss=0.1305, over 28835.00 frames. ], tot_loss[loss=0.36, simple_loss=0.4049, pruned_loss=0.1576, over 5648662.11 frames. ], libri_tot_loss[loss=0.4122, simple_loss=0.4428, pruned_loss=0.1909, over 5746396.91 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.4032, pruned_loss=0.1559, over 5647195.06 frames. ], batch size: 174, lr: 2.77e-02, grad_scale: 4.0
+2023-02-28 14:44:53,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15508.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:45:15,378 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-02-28 14:45:21,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.799e+02 1.521e+03 1.930e+03 2.552e+03 6.791e+03, threshold=3.859e+03, percent-clipped=7.0
+2023-02-28 14:45:43,547 INFO [train.py:968] (0/2) Epoch 1, batch 15550, giga_loss[loss=0.3842, simple_loss=0.4348, pruned_loss=0.1668, over 28909.00 frames. ], tot_loss[loss=0.3595, simple_loss=0.4065, pruned_loss=0.1563, over 5664708.71 frames. ], libri_tot_loss[loss=0.412, simple_loss=0.4427, pruned_loss=0.1908, over 5747873.41 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4045, pruned_loss=0.1544, over 5660681.69 frames. ], batch size: 199, lr: 2.77e-02, grad_scale: 4.0
+2023-02-28 14:46:16,584 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.46 vs. limit=5.0
+2023-02-28 14:46:25,865 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=15588.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:46:40,778 INFO [train.py:968] (0/2) Epoch 1, batch 15600, giga_loss[loss=0.3825, simple_loss=0.4229, pruned_loss=0.1711, over 29001.00 frames. ], tot_loss[loss=0.3648, simple_loss=0.4109, pruned_loss=0.1593, over 5660948.60 frames. ], libri_tot_loss[loss=0.411, simple_loss=0.4419, pruned_loss=0.1902, over 5743719.84 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.4088, pruned_loss=0.1571, over 5658462.11 frames. ], batch size: 128, lr: 2.76e-02, grad_scale: 8.0
+2023-02-28 14:47:15,134 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-02-28 14:47:28,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.731e+02 1.786e+03 2.105e+03 3.212e+03 8.100e+03, threshold=4.210e+03, percent-clipped=11.0
+2023-02-28 14:47:47,557 INFO [train.py:968] (0/2) Epoch 1, batch 15650, giga_loss[loss=0.3339, simple_loss=0.4006, pruned_loss=0.1336, over 28849.00 frames. ], tot_loss[loss=0.3661, simple_loss=0.4124, pruned_loss=0.1599, over 5657855.66 frames. ], libri_tot_loss[loss=0.411, simple_loss=0.442, pruned_loss=0.1901, over 5744292.24 frames. ], giga_tot_loss[loss=0.3633, simple_loss=0.4105, pruned_loss=0.1581, over 5655072.32 frames. ], batch size: 174, lr: 2.76e-02, grad_scale: 4.0
+2023-02-28 14:47:49,976 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15651.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:47:53,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15654.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:48:29,800 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15683.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:48:50,386 INFO [train.py:968] (0/2) Epoch 1, batch 15700, giga_loss[loss=0.354, simple_loss=0.4063, pruned_loss=0.1509, over 28375.00 frames. ], tot_loss[loss=0.365, simple_loss=0.4118, pruned_loss=0.1591, over 5671903.13 frames. ], libri_tot_loss[loss=0.4105, simple_loss=0.4414, pruned_loss=0.1899, over 5746812.70 frames. ], giga_tot_loss[loss=0.3627, simple_loss=0.4104, pruned_loss=0.1575, over 5666331.73 frames. ], batch size: 368, lr: 2.75e-02, grad_scale: 4.0
+2023-02-28 14:49:35,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.711e+03 2.273e+03 3.161e+03 6.328e+03, threshold=4.546e+03, percent-clipped=10.0
+2023-02-28 14:49:47,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6762, 2.2601, 1.6193, 1.6331], device='cuda:0'), covar=tensor([0.0870, 0.0848, 0.0747, 0.0488], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0793, 0.0673, 0.0585], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 14:49:52,246 INFO [train.py:968] (0/2) Epoch 1, batch 15750, giga_loss[loss=0.2751, simple_loss=0.3474, pruned_loss=0.1014, over 28942.00 frames. ], tot_loss[loss=0.362, simple_loss=0.4098, pruned_loss=0.1572, over 5679206.34 frames. ], libri_tot_loss[loss=0.4099, simple_loss=0.441, pruned_loss=0.1895, over 5746459.95 frames. ], giga_tot_loss[loss=0.3599, simple_loss=0.4086, pruned_loss=0.1556, over 5673904.92 frames. ], batch size: 145, lr: 2.75e-02, grad_scale: 4.0
+2023-02-28 14:50:27,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1210, 1.3414, 1.1822, 1.1223], device='cuda:0'), covar=tensor([0.1449, 0.1123, 0.1279, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.0473, 0.0498, 0.0427, 0.0527], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0008, 0.0010], device='cuda:0')
+2023-02-28 14:50:55,991 INFO [train.py:968] (0/2) Epoch 1, batch 15800, giga_loss[loss=0.3359, simple_loss=0.3974, pruned_loss=0.1372, over 28724.00 frames. ], tot_loss[loss=0.3586, simple_loss=0.4071, pruned_loss=0.1551, over 5685029.39 frames. ], libri_tot_loss[loss=0.4087, simple_loss=0.44, pruned_loss=0.1888, over 5750853.79 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4061, pruned_loss=0.1536, over 5675382.53 frames. ], batch size: 307, lr: 2.75e-02, grad_scale: 4.0
+2023-02-28 14:51:39,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.724e+02 1.560e+03 1.896e+03 2.495e+03 5.517e+03, threshold=3.792e+03, percent-clipped=1.0
+2023-02-28 14:51:56,941 INFO [train.py:968] (0/2) Epoch 1, batch 15850, giga_loss[loss=0.2791, simple_loss=0.3409, pruned_loss=0.1087, over 28809.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.4044, pruned_loss=0.1539, over 5684422.29 frames. ], libri_tot_loss[loss=0.4076, simple_loss=0.4392, pruned_loss=0.1881, over 5752002.79 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4034, pruned_loss=0.1524, over 5673584.85 frames. ], batch size: 119, lr: 2.74e-02, grad_scale: 4.0
+2023-02-28 14:52:58,607 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-02-28 14:53:01,297 INFO [train.py:968] (0/2) Epoch 1, batch 15900, giga_loss[loss=0.3786, simple_loss=0.4206, pruned_loss=0.1683, over 28916.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4051, pruned_loss=0.1544, over 5682179.27 frames. ], libri_tot_loss[loss=0.407, simple_loss=0.4388, pruned_loss=0.1877, over 5754242.53 frames. ], giga_tot_loss[loss=0.3549, simple_loss=0.404, pruned_loss=0.153, over 5670583.33 frames. ], batch size: 164, lr: 2.74e-02, grad_scale: 4.0
+2023-02-28 14:53:43,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.587e+02 1.638e+03 1.974e+03 2.647e+03 5.884e+03, threshold=3.947e+03, percent-clipped=6.0
+2023-02-28 14:54:00,677 INFO [train.py:968] (0/2) Epoch 1, batch 15950, giga_loss[loss=0.3739, simple_loss=0.4058, pruned_loss=0.171, over 27631.00 frames. ], tot_loss[loss=0.3604, simple_loss=0.4077, pruned_loss=0.1565, over 5685649.59 frames. ], libri_tot_loss[loss=0.4054, simple_loss=0.4376, pruned_loss=0.1867, over 5759906.55 frames. ], giga_tot_loss[loss=0.3584, simple_loss=0.4068, pruned_loss=0.1551, over 5668466.27 frames. ], batch size: 472, lr: 2.73e-02, grad_scale: 4.0
+2023-02-28 14:54:19,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15963.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:54:35,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9714, 1.7867, 1.2998, 1.3340], device='cuda:0'), covar=tensor([0.0819, 0.1004, 0.1217, 0.1450], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0652, 0.0651, 0.0615], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 14:55:06,902 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-16000.pt
+2023-02-28 14:55:07,246 INFO [train.py:968] (0/2) Epoch 1, batch 16000, giga_loss[loss=0.3231, simple_loss=0.386, pruned_loss=0.1301, over 28957.00 frames. ], tot_loss[loss=0.361, simple_loss=0.4081, pruned_loss=0.157, over 5685218.97 frames. ], libri_tot_loss[loss=0.4044, simple_loss=0.4368, pruned_loss=0.1861, over 5763047.27 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4069, pruned_loss=0.1553, over 5665747.89 frames. ], batch size: 186, lr: 2.73e-02, grad_scale: 8.0
+2023-02-28 14:55:44,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3945, 1.4152, 1.6118, 0.7493], device='cuda:0'), covar=tensor([0.0390, 0.0227, 0.0219, 0.0425], device='cuda:0'), in_proj_covar=tensor([0.0600, 0.0466, 0.0516, 0.0550], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 14:55:52,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.848e+02 1.532e+03 1.908e+03 2.512e+03 5.421e+03, threshold=3.815e+03, percent-clipped=5.0
+2023-02-28 14:56:08,287 INFO [train.py:968] (0/2) Epoch 1, batch 16050, giga_loss[loss=0.3604, simple_loss=0.4117, pruned_loss=0.1545, over 28939.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.4123, pruned_loss=0.1603, over 5680998.85 frames. ], libri_tot_loss[loss=0.4036, simple_loss=0.4362, pruned_loss=0.1856, over 5755157.89 frames. ], giga_tot_loss[loss=0.3645, simple_loss=0.4114, pruned_loss=0.1588, over 5669999.36 frames. ], batch size: 199, lr: 2.73e-02, grad_scale: 4.0
+2023-02-28 14:57:09,570 INFO [train.py:968] (0/2) Epoch 1, batch 16100, giga_loss[loss=0.353, simple_loss=0.4129, pruned_loss=0.1466, over 29027.00 frames. ], tot_loss[loss=0.3707, simple_loss=0.4168, pruned_loss=0.1623, over 5678378.65 frames. ], libri_tot_loss[loss=0.4029, simple_loss=0.4357, pruned_loss=0.1851, over 5747316.22 frames. ], giga_tot_loss[loss=0.3691, simple_loss=0.416, pruned_loss=0.1611, over 5674549.18 frames. ], batch size: 285, lr: 2.72e-02, grad_scale: 4.0
+2023-02-28 14:57:16,578 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16106.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:57:19,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8835, 1.9307, 4.6052, 2.8129], device='cuda:0'), covar=tensor([0.1522, 0.1003, 0.0213, 0.0469], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0438, 0.0551, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 14:57:19,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16109.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:57:48,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.821e+02 1.726e+03 2.374e+03 3.066e+03 7.912e+03, threshold=4.747e+03, percent-clipped=10.0
+2023-02-28 14:57:50,752 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16138.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:58:05,640 INFO [train.py:968] (0/2) Epoch 1, batch 16150, giga_loss[loss=0.3416, simple_loss=0.4029, pruned_loss=0.1402, over 28995.00 frames. ], tot_loss[loss=0.3737, simple_loss=0.4192, pruned_loss=0.1641, over 5678175.61 frames. ], libri_tot_loss[loss=0.4016, simple_loss=0.4349, pruned_loss=0.1843, over 5744442.19 frames. ], giga_tot_loss[loss=0.3723, simple_loss=0.4186, pruned_loss=0.163, over 5674659.19 frames. ], batch size: 164, lr: 2.72e-02, grad_scale: 4.0
+2023-02-28 14:59:20,006 INFO [train.py:968] (0/2) Epoch 1, batch 16200, giga_loss[loss=0.3695, simple_loss=0.4154, pruned_loss=0.1618, over 28959.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.4153, pruned_loss=0.1608, over 5684455.91 frames. ], libri_tot_loss[loss=0.4008, simple_loss=0.4342, pruned_loss=0.1838, over 5746862.18 frames. ], giga_tot_loss[loss=0.3676, simple_loss=0.4152, pruned_loss=0.16, over 5678729.93 frames. ], batch size: 155, lr: 2.72e-02, grad_scale: 4.0
+2023-02-28 14:59:30,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6227, 2.4720, 2.4710, 1.6069], device='cuda:0'), covar=tensor([0.1791, 0.0717, 0.0689, 0.2243], device='cuda:0'), in_proj_covar=tensor([0.0532, 0.0434, 0.0413, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0016], device='cuda:0')
+2023-02-28 14:59:53,468 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-02-28 15:00:04,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.308e+02 1.567e+03 2.083e+03 2.828e+03 9.487e+03, threshold=4.167e+03, percent-clipped=4.0
+2023-02-28 15:00:10,141 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-02-28 15:00:18,804 INFO [train.py:968] (0/2) Epoch 1, batch 16250, giga_loss[loss=0.3498, simple_loss=0.3994, pruned_loss=0.1502, over 28949.00 frames. ], tot_loss[loss=0.3676, simple_loss=0.4138, pruned_loss=0.1607, over 5696393.37 frames. ], libri_tot_loss[loss=0.3999, simple_loss=0.4334, pruned_loss=0.1832, over 5751991.51 frames. ], giga_tot_loss[loss=0.3665, simple_loss=0.4137, pruned_loss=0.1597, over 5684876.63 frames. ], batch size: 164, lr: 2.71e-02, grad_scale: 4.0
+2023-02-28 15:00:57,132 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=16276.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:01:03,130 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.00 vs. limit=2.0
+2023-02-28 15:01:10,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4545, 1.6504, 1.3377, 1.3358], device='cuda:0'), covar=tensor([0.0513, 0.0574, 0.0496, 0.0311], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0794, 0.0678, 0.0577], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 15:01:30,042 INFO [train.py:968] (0/2) Epoch 1, batch 16300, giga_loss[loss=0.3194, simple_loss=0.3551, pruned_loss=0.1418, over 24416.00 frames. ], tot_loss[loss=0.367, simple_loss=0.4128, pruned_loss=0.1606, over 5668064.27 frames. ], libri_tot_loss[loss=0.3998, simple_loss=0.4333, pruned_loss=0.1832, over 5745704.71 frames. ], giga_tot_loss[loss=0.3657, simple_loss=0.4124, pruned_loss=0.1595, over 5663703.26 frames. ], batch size: 705, lr: 2.71e-02, grad_scale: 4.0
+2023-02-28 15:01:31,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4704, 2.0667, 1.7985, 1.3196], device='cuda:0'), covar=tensor([0.1972, 0.0806, 0.0846, 0.2906], device='cuda:0'), in_proj_covar=tensor([0.0521, 0.0418, 0.0406, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0016], device='cuda:0')
+2023-02-28 15:02:14,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.067e+02 1.728e+03 2.170e+03 2.689e+03 1.097e+04, threshold=4.340e+03, percent-clipped=11.0
+2023-02-28 15:02:32,044 INFO [train.py:968] (0/2) Epoch 1, batch 16350, giga_loss[loss=0.3435, simple_loss=0.393, pruned_loss=0.147, over 28385.00 frames. ], tot_loss[loss=0.3652, simple_loss=0.411, pruned_loss=0.1597, over 5677910.09 frames. ], libri_tot_loss[loss=0.3989, simple_loss=0.4328, pruned_loss=0.1826, over 5748374.57 frames. ], giga_tot_loss[loss=0.3639, simple_loss=0.4106, pruned_loss=0.1586, over 5669923.06 frames. ], batch size: 369, lr: 2.70e-02, grad_scale: 4.0
+2023-02-28 15:03:28,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=16394.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:03:34,411 INFO [train.py:968] (0/2) Epoch 1, batch 16400, libri_loss[loss=0.3761, simple_loss=0.4156, pruned_loss=0.1683, over 29559.00 frames. ], tot_loss[loss=0.3645, simple_loss=0.4096, pruned_loss=0.1598, over 5675984.28 frames. ], libri_tot_loss[loss=0.3983, simple_loss=0.4323, pruned_loss=0.1822, over 5752270.32 frames. ], giga_tot_loss[loss=0.3632, simple_loss=0.4091, pruned_loss=0.1587, over 5664383.53 frames. ], batch size: 79, lr: 2.70e-02, grad_scale: 8.0
+2023-02-28 15:04:16,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.451e+02 1.617e+03 1.983e+03 2.617e+03 5.219e+03, threshold=3.966e+03, percent-clipped=2.0
+2023-02-28 15:04:34,749 INFO [train.py:968] (0/2) Epoch 1, batch 16450, giga_loss[loss=0.3429, simple_loss=0.4061, pruned_loss=0.1399, over 29052.00 frames. ], tot_loss[loss=0.3637, simple_loss=0.4098, pruned_loss=0.1588, over 5679354.42 frames. ], libri_tot_loss[loss=0.3981, simple_loss=0.4322, pruned_loss=0.1821, over 5754297.19 frames. ], giga_tot_loss[loss=0.3621, simple_loss=0.4091, pruned_loss=0.1575, over 5666643.39 frames. ], batch size: 155, lr: 2.70e-02, grad_scale: 8.0
+2023-02-28 15:05:08,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2699, 1.5681, 1.4876, 1.1365], device='cuda:0'), covar=tensor([0.1074, 0.1106, 0.0953, 0.1569], device='cuda:0'), in_proj_covar=tensor([0.0476, 0.0509, 0.0429, 0.0526], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0010], device='cuda:0')
+2023-02-28 15:05:10,968 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=16478.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:05:32,927 INFO [train.py:968] (0/2) Epoch 1, batch 16500, giga_loss[loss=0.357, simple_loss=0.4283, pruned_loss=0.1428, over 28698.00 frames. ], tot_loss[loss=0.3606, simple_loss=0.4084, pruned_loss=0.1564, over 5681100.42 frames. ], libri_tot_loss[loss=0.3983, simple_loss=0.4325, pruned_loss=0.1821, over 5755567.82 frames. ], giga_tot_loss[loss=0.3582, simple_loss=0.4069, pruned_loss=0.1547, over 5667560.86 frames. ], batch size: 307, lr: 2.69e-02, grad_scale: 4.0
+2023-02-28 15:05:57,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1541, 1.3930, 1.3024, 1.2345], device='cuda:0'), covar=tensor([0.1731, 0.0834, 0.0851, 0.2192], device='cuda:0'), in_proj_covar=tensor([0.0516, 0.0421, 0.0403, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0016], device='cuda:0')
+2023-02-28 15:06:17,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.795e+02 1.495e+03 1.882e+03 2.443e+03 8.206e+03, threshold=3.764e+03, percent-clipped=5.0
+2023-02-28 15:06:25,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=16542.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:06:34,139 INFO [train.py:968] (0/2) Epoch 1, batch 16550, giga_loss[loss=0.3648, simple_loss=0.4136, pruned_loss=0.1579, over 27524.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.4106, pruned_loss=0.1556, over 5683649.09 frames. ], libri_tot_loss[loss=0.3975, simple_loss=0.4318, pruned_loss=0.1817, over 5758496.07 frames. ], giga_tot_loss[loss=0.359, simple_loss=0.4096, pruned_loss=0.1542, over 5669031.73 frames. ], batch size: 472, lr: 2.69e-02, grad_scale: 2.0
+2023-02-28 15:07:33,172 INFO [train.py:968] (0/2) Epoch 1, batch 16600, giga_loss[loss=0.313, simple_loss=0.3854, pruned_loss=0.1203, over 29189.00 frames. ], tot_loss[loss=0.3619, simple_loss=0.4123, pruned_loss=0.1557, over 5685904.72 frames. ], libri_tot_loss[loss=0.3966, simple_loss=0.431, pruned_loss=0.1811, over 5759622.46 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4118, pruned_loss=0.1546, over 5671848.40 frames. ], batch size: 113, lr: 2.69e-02, grad_scale: 2.0
+2023-02-28 15:07:38,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4323, 1.5985, 1.1306, 1.2612], device='cuda:0'), covar=tensor([0.1227, 0.1280, 0.1090, 0.1534], device='cuda:0'), in_proj_covar=tensor([0.0847, 0.0772, 0.0841, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 15:08:17,095 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.644e+02 1.810e+03 2.256e+03 3.067e+03 6.727e+03, threshold=4.512e+03, percent-clipped=9.0
+2023-02-28 15:08:32,935 INFO [train.py:968] (0/2) Epoch 1, batch 16650, giga_loss[loss=0.3775, simple_loss=0.4232, pruned_loss=0.1659, over 28990.00 frames. ], tot_loss[loss=0.3622, simple_loss=0.4125, pruned_loss=0.156, over 5691826.65 frames. ], libri_tot_loss[loss=0.3955, simple_loss=0.4303, pruned_loss=0.1804, over 5761032.84 frames. ], giga_tot_loss[loss=0.3608, simple_loss=0.4121, pruned_loss=0.1548, over 5676765.70 frames. ], batch size: 106, lr: 2.68e-02, grad_scale: 2.0
+2023-02-28 15:08:35,573 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16651.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:09:40,123 INFO [train.py:968] (0/2) Epoch 1, batch 16700, giga_loss[loss=0.3958, simple_loss=0.4237, pruned_loss=0.1839, over 26815.00 frames. ], tot_loss[loss=0.3635, simple_loss=0.4128, pruned_loss=0.1571, over 5677289.02 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.4295, pruned_loss=0.1799, over 5756278.63 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.4127, pruned_loss=0.1558, over 5666860.03 frames. ], batch size: 555, lr: 2.68e-02, grad_scale: 2.0
+2023-02-28 15:10:30,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.588e+02 1.664e+03 2.082e+03 2.938e+03 6.388e+03, threshold=4.164e+03, percent-clipped=5.0
+2023-02-28 15:10:48,176 INFO [train.py:968] (0/2) Epoch 1, batch 16750, giga_loss[loss=0.3374, simple_loss=0.3988, pruned_loss=0.138, over 28563.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.4125, pruned_loss=0.1561, over 5678181.71 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4294, pruned_loss=0.1797, over 5756584.51 frames. ], giga_tot_loss[loss=0.3606, simple_loss=0.412, pruned_loss=0.1547, over 5667434.01 frames. ], batch size: 78, lr: 2.67e-02, grad_scale: 2.0
+2023-02-28 15:11:09,153 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=16764.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:11:16,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16769.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:11:56,474 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16794.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:12:00,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16797.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:12:04,606 INFO [train.py:968] (0/2) Epoch 1, batch 16800, giga_loss[loss=0.3581, simple_loss=0.4108, pruned_loss=0.1527, over 29006.00 frames. ], tot_loss[loss=0.3622, simple_loss=0.413, pruned_loss=0.1557, over 5684331.93 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.4295, pruned_loss=0.1798, over 5758815.74 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4123, pruned_loss=0.1541, over 5672802.52 frames. ], batch size: 285, lr: 2.67e-02, grad_scale: 4.0
+2023-02-28 15:12:41,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16826.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:12:56,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.550e+02 1.646e+03 2.151e+03 3.233e+03 2.270e+04, threshold=4.302e+03, percent-clipped=16.0
+2023-02-28 15:13:05,728 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-02-28 15:13:09,798 INFO [train.py:968] (0/2) Epoch 1, batch 16850, giga_loss[loss=0.4102, simple_loss=0.4583, pruned_loss=0.181, over 28645.00 frames. ], tot_loss[loss=0.3657, simple_loss=0.4154, pruned_loss=0.158, over 5688712.43 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4281, pruned_loss=0.1788, over 5761710.14 frames. ], giga_tot_loss[loss=0.3644, simple_loss=0.4155, pruned_loss=0.1567, over 5673543.04 frames. ], batch size: 307, lr: 2.67e-02, grad_scale: 4.0
+2023-02-28 15:13:15,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16853.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:14:18,415 INFO [train.py:968] (0/2) Epoch 1, batch 16900, giga_loss[loss=0.4431, simple_loss=0.4759, pruned_loss=0.2051, over 28510.00 frames. ], tot_loss[loss=0.3687, simple_loss=0.4181, pruned_loss=0.1596, over 5684294.27 frames. ], libri_tot_loss[loss=0.3922, simple_loss=0.4275, pruned_loss=0.1785, over 5755431.02 frames. ], giga_tot_loss[loss=0.3676, simple_loss=0.4184, pruned_loss=0.1585, over 5676476.04 frames. ], batch size: 336, lr: 2.66e-02, grad_scale: 4.0
+2023-02-28 15:14:34,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16912.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:14:40,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16915.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:14:41,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16917.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:15:12,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.170e+03 1.677e+03 2.122e+03 2.684e+03 5.233e+03, threshold=4.244e+03, percent-clipped=2.0
+2023-02-28 15:15:21,838 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16944.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:15:30,095 INFO [train.py:968] (0/2) Epoch 1, batch 16950, giga_loss[loss=0.3298, simple_loss=0.3881, pruned_loss=0.1358, over 28404.00 frames. ], tot_loss[loss=0.3662, simple_loss=0.4158, pruned_loss=0.1583, over 5686184.47 frames. ], libri_tot_loss[loss=0.3916, simple_loss=0.4272, pruned_loss=0.1781, over 5757139.83 frames. ], giga_tot_loss[loss=0.3656, simple_loss=0.4163, pruned_loss=0.1574, over 5677524.24 frames. ], batch size: 336, lr: 2.66e-02, grad_scale: 4.0
+2023-02-28 15:16:27,087 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16996.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:16:30,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16999.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:16:30,954 INFO [train.py:968] (0/2) Epoch 1, batch 17000, giga_loss[loss=0.3683, simple_loss=0.4151, pruned_loss=0.1608, over 28411.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4133, pruned_loss=0.1569, over 5702514.71 frames. ], libri_tot_loss[loss=0.3912, simple_loss=0.4273, pruned_loss=0.1776, over 5765087.47 frames. ], giga_tot_loss[loss=0.3621, simple_loss=0.4129, pruned_loss=0.1557, over 5685134.65 frames. ], batch size: 368, lr: 2.66e-02, grad_scale: 4.0
+2023-02-28 15:17:02,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1845, 1.6132, 1.4477, 1.2011], device='cuda:0'), covar=tensor([0.1876, 0.0786, 0.0845, 0.2240], device='cuda:0'), in_proj_covar=tensor([0.0532, 0.0429, 0.0408, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0016], device='cuda:0')
+2023-02-28 15:17:10,371 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17028.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:17:24,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.450e+02 1.733e+03 2.226e+03 2.916e+03 5.286e+03, threshold=4.453e+03, percent-clipped=4.0
+2023-02-28 15:17:43,303 INFO [train.py:968] (0/2) Epoch 1, batch 17050, giga_loss[loss=0.3072, simple_loss=0.3793, pruned_loss=0.1176, over 28915.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.4104, pruned_loss=0.1541, over 5706132.58 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.4269, pruned_loss=0.1774, over 5768936.34 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.41, pruned_loss=0.1525, over 5686766.35 frames. ], batch size: 145, lr: 2.65e-02, grad_scale: 4.0
+2023-02-28 15:17:56,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17060.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:18:00,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17063.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:18:05,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2511, 1.2237, 1.2781, 0.1224], device='cuda:0'), covar=tensor([0.0459, 0.0527, 0.0557, 0.0908], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0791, 0.0801, 0.0779], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 15:18:39,104 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17092.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:18:47,554 INFO [train.py:968] (0/2) Epoch 1, batch 17100, libri_loss[loss=0.3646, simple_loss=0.4206, pruned_loss=0.1543, over 29674.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.4077, pruned_loss=0.1516, over 5711371.74 frames. ], libri_tot_loss[loss=0.3901, simple_loss=0.4264, pruned_loss=0.1769, over 5771070.40 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4072, pruned_loss=0.15, over 5691834.79 frames. ], batch size: 91, lr: 2.65e-02, grad_scale: 4.0
+2023-02-28 15:19:35,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.507e+02 1.498e+03 1.694e+03 2.404e+03 8.291e+03, threshold=3.388e+03, percent-clipped=3.0
+2023-02-28 15:19:36,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17139.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:19:48,676 INFO [train.py:968] (0/2) Epoch 1, batch 17150, giga_loss[loss=0.3712, simple_loss=0.4236, pruned_loss=0.1594, over 28039.00 frames. ], tot_loss[loss=0.3585, simple_loss=0.4096, pruned_loss=0.1537, over 5691996.81 frames. ], libri_tot_loss[loss=0.39, simple_loss=0.4264, pruned_loss=0.1768, over 5763942.10 frames. ], giga_tot_loss[loss=0.3561, simple_loss=0.4086, pruned_loss=0.1518, over 5680222.09 frames. ], batch size: 412, lr: 2.65e-02, grad_scale: 4.0
+2023-02-28 15:20:13,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5139, 1.9966, 1.6069, 1.4177], device='cuda:0'), covar=tensor([0.1582, 0.0799, 0.0825, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.0517, 0.0417, 0.0404, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0016], device='cuda:0')
+2023-02-28 15:20:44,877 INFO [train.py:968] (0/2) Epoch 1, batch 17200, giga_loss[loss=0.3837, simple_loss=0.4317, pruned_loss=0.1679, over 28894.00 frames. ], tot_loss[loss=0.3649, simple_loss=0.4141, pruned_loss=0.1578, over 5691747.40 frames. ], libri_tot_loss[loss=0.389, simple_loss=0.4256, pruned_loss=0.1762, over 5762866.34 frames. ], giga_tot_loss[loss=0.3626, simple_loss=0.4134, pruned_loss=0.1559, over 5680307.21 frames. ], batch size: 186, lr: 2.64e-02, grad_scale: 8.0
+2023-02-28 15:21:06,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0480, 3.0148, 4.7486, 1.5815], device='cuda:0'), covar=tensor([0.0419, 0.0773, 0.0905, 0.2349], device='cuda:0'), in_proj_covar=tensor([0.0644, 0.0439, 0.0751, 0.0518], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0006], device='cuda:0')
+2023-02-28 15:21:28,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.524e+02 2.036e+03 2.580e+03 3.377e+03 7.576e+03, threshold=5.160e+03, percent-clipped=22.0
+2023-02-28 15:21:40,832 INFO [train.py:968] (0/2) Epoch 1, batch 17250, libri_loss[loss=0.3941, simple_loss=0.4235, pruned_loss=0.1824, over 19735.00 frames. ], tot_loss[loss=0.3666, simple_loss=0.4151, pruned_loss=0.1591, over 5679136.45 frames. ], libri_tot_loss[loss=0.3884, simple_loss=0.4253, pruned_loss=0.1758, over 5753436.58 frames. ], giga_tot_loss[loss=0.3647, simple_loss=0.4145, pruned_loss=0.1574, over 5677526.53 frames. ], batch size: 186, lr: 2.64e-02, grad_scale: 4.0
+2023-02-28 15:21:53,824 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=17261.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:22:14,756 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-02-28 15:22:17,056 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17282.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:22:20,023 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4708, 2.0434, 1.8222, 1.2937], device='cuda:0'), covar=tensor([0.1532, 0.0665, 0.0719, 0.1998], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0398, 0.0393, 0.0628], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0016], device='cuda:0')
+2023-02-28 15:22:20,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17285.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:22:34,090 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=17299.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:22:34,470 INFO [train.py:968] (0/2) Epoch 1, batch 17300, giga_loss[loss=0.3987, simple_loss=0.4272, pruned_loss=0.1851, over 27626.00 frames. ], tot_loss[loss=0.3645, simple_loss=0.412, pruned_loss=0.1585, over 5678918.70 frames. ], libri_tot_loss[loss=0.3877, simple_loss=0.4248, pruned_loss=0.1753, over 5754368.10 frames. ], giga_tot_loss[loss=0.3629, simple_loss=0.4117, pruned_loss=0.1571, over 5674566.84 frames. ], batch size: 472, lr: 2.63e-02, grad_scale: 4.0
+2023-02-28 15:22:53,305 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17314.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:23:22,493 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.162e+03 1.841e+03 2.461e+03 3.033e+03 7.160e+03, threshold=4.922e+03, percent-clipped=4.0
+2023-02-28 15:23:35,448 INFO [train.py:968] (0/2) Epoch 1, batch 17350, giga_loss[loss=0.3411, simple_loss=0.3923, pruned_loss=0.145, over 28453.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4113, pruned_loss=0.1586, over 5682257.28 frames. ], libri_tot_loss[loss=0.3871, simple_loss=0.4243, pruned_loss=0.175, over 5756515.40 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.4113, pruned_loss=0.1575, over 5675907.42 frames. ], batch size: 336, lr: 2.63e-02, grad_scale: 4.0
+2023-02-28 15:24:34,901 INFO [train.py:968] (0/2) Epoch 1, batch 17400, libri_loss[loss=0.3698, simple_loss=0.4065, pruned_loss=0.1666, over 29562.00 frames. ], tot_loss[loss=0.3784, simple_loss=0.4213, pruned_loss=0.1677, over 5683491.66 frames. ], libri_tot_loss[loss=0.388, simple_loss=0.4249, pruned_loss=0.1756, over 5758599.91 frames. ], giga_tot_loss[loss=0.3765, simple_loss=0.4206, pruned_loss=0.1662, over 5675451.13 frames. ], batch size: 75, lr: 2.63e-02, grad_scale: 4.0
+2023-02-28 15:25:12,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6699, 2.0149, 3.8077, 2.4074], device='cuda:0'), covar=tensor([0.1455, 0.0840, 0.0240, 0.0459], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0419, 0.0538, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 15:25:16,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.824e+03 2.174e+03 2.789e+03 9.476e+03, threshold=4.348e+03, percent-clipped=7.0
+2023-02-28 15:25:26,036 INFO [train.py:968] (0/2) Epoch 1, batch 17450, giga_loss[loss=0.5401, simple_loss=0.5337, pruned_loss=0.2733, over 28543.00 frames. ], tot_loss[loss=0.3958, simple_loss=0.435, pruned_loss=0.1783, over 5690203.81 frames. ], libri_tot_loss[loss=0.3886, simple_loss=0.4254, pruned_loss=0.176, over 5760484.33 frames. ], giga_tot_loss[loss=0.3938, simple_loss=0.4341, pruned_loss=0.1767, over 5681040.87 frames. ], batch size: 65, lr: 2.62e-02, grad_scale: 4.0
+2023-02-28 15:26:13,104 INFO [train.py:968] (0/2) Epoch 1, batch 17500, giga_loss[loss=0.4331, simple_loss=0.4591, pruned_loss=0.2036, over 28995.00 frames. ], tot_loss[loss=0.402, simple_loss=0.4403, pruned_loss=0.1819, over 5694446.56 frames. ], libri_tot_loss[loss=0.389, simple_loss=0.4257, pruned_loss=0.1761, over 5762053.36 frames. ], giga_tot_loss[loss=0.4003, simple_loss=0.4395, pruned_loss=0.1806, over 5684748.67 frames. ], batch size: 155, lr: 2.62e-02, grad_scale: 4.0
+2023-02-28 15:26:48,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.465e+02 1.302e+03 1.575e+03 2.055e+03 6.128e+03, threshold=3.149e+03, percent-clipped=3.0
+2023-02-28 15:26:57,783 INFO [train.py:968] (0/2) Epoch 1, batch 17550, giga_loss[loss=0.3224, simple_loss=0.3646, pruned_loss=0.14, over 28799.00 frames. ], tot_loss[loss=0.3972, simple_loss=0.435, pruned_loss=0.1797, over 5684744.47 frames. ], libri_tot_loss[loss=0.3892, simple_loss=0.4261, pruned_loss=0.1762, over 5756298.87 frames. ], giga_tot_loss[loss=0.3959, simple_loss=0.4343, pruned_loss=0.1787, over 5680738.85 frames. ], batch size: 92, lr: 2.62e-02, grad_scale: 4.0
+2023-02-28 15:27:11,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0161, 1.1388, 1.0259, 0.8198], device='cuda:0'), covar=tensor([0.1368, 0.1539, 0.1103, 0.1414], device='cuda:0'), in_proj_covar=tensor([0.0834, 0.0772, 0.0815, 0.0839], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 15:27:27,475 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=17582.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:27:44,764 INFO [train.py:968] (0/2) Epoch 1, batch 17600, giga_loss[loss=0.3066, simple_loss=0.3636, pruned_loss=0.1248, over 28926.00 frames. ], tot_loss[loss=0.3863, simple_loss=0.4251, pruned_loss=0.1737, over 5686427.15 frames. ], libri_tot_loss[loss=0.3895, simple_loss=0.4264, pruned_loss=0.1763, over 5756443.03 frames. ], giga_tot_loss[loss=0.385, simple_loss=0.4243, pruned_loss=0.1729, over 5682106.16 frames. ], batch size: 136, lr: 2.61e-02, grad_scale: 8.0
+2023-02-28 15:28:16,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17636.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 15:28:20,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.205e+02 1.274e+03 1.496e+03 1.981e+03 5.001e+03, threshold=2.991e+03, percent-clipped=8.0
+2023-02-28 15:28:28,313 INFO [train.py:968] (0/2) Epoch 1, batch 17650, giga_loss[loss=0.3156, simple_loss=0.3611, pruned_loss=0.1351, over 28680.00 frames. ], tot_loss[loss=0.3769, simple_loss=0.4164, pruned_loss=0.1687, over 5689896.38 frames. ], libri_tot_loss[loss=0.39, simple_loss=0.4271, pruned_loss=0.1765, over 5760514.97 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.4151, pruned_loss=0.1677, over 5680883.22 frames. ], batch size: 92, lr: 2.61e-02, grad_scale: 4.0
+2023-02-28 15:28:35,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2743, 1.3017, 1.4887, 1.0643], device='cuda:0'), covar=tensor([0.1325, 0.1134, 0.1030, 0.2030], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0507, 0.0427, 0.0529], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0011], device='cuda:0')
+2023-02-28 15:28:51,871 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17674.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:29:16,074 INFO [train.py:968] (0/2) Epoch 1, batch 17700, giga_loss[loss=0.3284, simple_loss=0.3755, pruned_loss=0.1406, over 28844.00 frames. ], tot_loss[loss=0.3663, simple_loss=0.4064, pruned_loss=0.1631, over 5688148.41 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.4276, pruned_loss=0.177, over 5762080.03 frames. ], giga_tot_loss[loss=0.364, simple_loss=0.4046, pruned_loss=0.1617, over 5678861.63 frames. ], batch size: 174, lr: 2.61e-02, grad_scale: 4.0
+2023-02-28 15:29:47,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.307e+02 1.276e+03 1.570e+03 2.046e+03 8.200e+03, threshold=3.141e+03, percent-clipped=11.0
+2023-02-28 15:29:56,184 INFO [train.py:968] (0/2) Epoch 1, batch 17750, giga_loss[loss=0.3497, simple_loss=0.3815, pruned_loss=0.159, over 28003.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.4016, pruned_loss=0.1607, over 5697887.84 frames. ], libri_tot_loss[loss=0.3915, simple_loss=0.4283, pruned_loss=0.1774, over 5768821.74 frames. ], giga_tot_loss[loss=0.3576, simple_loss=0.3983, pruned_loss=0.1585, over 5680823.39 frames. ], batch size: 412, lr: 2.60e-02, grad_scale: 4.0
+2023-02-28 15:29:59,623 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=17755.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:30:04,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5430, 2.2627, 3.2968, 1.6997], device='cuda:0'), covar=tensor([0.0655, 0.0840, 0.0911, 0.1820], device='cuda:0'), in_proj_covar=tensor([0.0655, 0.0436, 0.0741, 0.0509], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0006], device='cuda:0')
+2023-02-28 15:30:16,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=17776.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:30:19,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17779.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:30:21,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17782.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:30:36,173 INFO [train.py:968] (0/2) Epoch 1, batch 17800, giga_loss[loss=0.3175, simple_loss=0.3705, pruned_loss=0.1322, over 28833.00 frames. ], tot_loss[loss=0.356, simple_loss=0.3971, pruned_loss=0.1575, over 5703915.69 frames. ], libri_tot_loss[loss=0.393, simple_loss=0.4297, pruned_loss=0.1782, over 5771077.38 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.3924, pruned_loss=0.1544, over 5686711.05 frames. ], batch size: 174, lr: 2.60e-02, grad_scale: 4.0
+2023-02-28 15:30:45,759 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17811.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:30:50,720 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17817.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:30:53,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17820.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:31:07,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.610e+02 1.364e+03 1.809e+03 2.685e+03 7.974e+03, threshold=3.618e+03, percent-clipped=15.0
+2023-02-28 15:31:16,989 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17849.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:31:17,328 INFO [train.py:968] (0/2) Epoch 1, batch 17850, giga_loss[loss=0.3315, simple_loss=0.3659, pruned_loss=0.1486, over 27653.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.393, pruned_loss=0.155, over 5709451.47 frames. ], libri_tot_loss[loss=0.3932, simple_loss=0.4299, pruned_loss=0.1783, over 5774059.51 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3882, pruned_loss=0.1519, over 5691592.23 frames. ], batch size: 472, lr: 2.60e-02, grad_scale: 4.0
+2023-02-28 15:31:55,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8235, 1.8037, 1.2972, 1.4628], device='cuda:0'), covar=tensor([0.0761, 0.0854, 0.1163, 0.1214], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0663, 0.0629, 0.0602], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 15:32:02,924 INFO [train.py:968] (0/2) Epoch 1, batch 17900, giga_loss[loss=0.2774, simple_loss=0.3347, pruned_loss=0.1101, over 28838.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3865, pruned_loss=0.1505, over 5703110.37 frames. ], libri_tot_loss[loss=0.3933, simple_loss=0.43, pruned_loss=0.1783, over 5774470.48 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3823, pruned_loss=0.1478, over 5688048.07 frames. ], batch size: 186, lr: 2.59e-02, grad_scale: 4.0
+2023-02-28 15:32:35,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.565e+02 1.148e+03 1.399e+03 2.088e+03 9.885e+03, threshold=2.799e+03, percent-clipped=8.0
+2023-02-28 15:32:47,951 INFO [train.py:968] (0/2) Epoch 1, batch 17950, giga_loss[loss=0.2792, simple_loss=0.3378, pruned_loss=0.1103, over 28560.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3836, pruned_loss=0.1492, over 5702900.25 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4308, pruned_loss=0.1789, over 5775100.80 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3789, pruned_loss=0.1462, over 5689524.45 frames. ], batch size: 78, lr: 2.59e-02, grad_scale: 4.0
+2023-02-28 15:32:48,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7906, 2.0392, 1.5962, 1.1938], device='cuda:0'), covar=tensor([0.0478, 0.0252, 0.0328, 0.0455], device='cuda:0'), in_proj_covar=tensor([0.0672, 0.0505, 0.0554, 0.0605], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 15:32:53,474 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17957.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:33:32,312 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-18000.pt
+2023-02-28 15:33:32,621 INFO [train.py:968] (0/2) Epoch 1, batch 18000, giga_loss[loss=0.3032, simple_loss=0.3561, pruned_loss=0.1252, over 28846.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3792, pruned_loss=0.1462, over 5698667.41 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4309, pruned_loss=0.1789, over 5776160.13 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3747, pruned_loss=0.1434, over 5686057.26 frames. ], batch size: 199, lr: 2.59e-02, grad_scale: 8.0
+2023-02-28 15:33:32,625 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 15:33:41,882 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.3425, simple_loss=0.4197, pruned_loss=0.1326, over 944034.00 frames. 
+2023-02-28 15:33:41,883 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 15:33:55,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3661, 1.3767, 1.2540, 1.1917], device='cuda:0'), covar=tensor([0.0739, 0.0914, 0.1155, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0564, 0.0902, 0.0676, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 15:34:16,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.284e+02 1.045e+03 1.376e+03 1.985e+03 4.039e+03, threshold=2.753e+03, percent-clipped=6.0
+2023-02-28 15:34:27,051 INFO [train.py:968] (0/2) Epoch 1, batch 18050, giga_loss[loss=0.3282, simple_loss=0.368, pruned_loss=0.1442, over 28873.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3772, pruned_loss=0.1454, over 5699700.87 frames. ], libri_tot_loss[loss=0.3954, simple_loss=0.4317, pruned_loss=0.1795, over 5778083.78 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3718, pruned_loss=0.1419, over 5686652.47 frames. ], batch size: 199, lr: 2.58e-02, grad_scale: 4.0
+2023-02-28 15:35:02,251 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18093.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:07,821 INFO [train.py:968] (0/2) Epoch 1, batch 18100, giga_loss[loss=0.2842, simple_loss=0.3405, pruned_loss=0.1139, over 28387.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3758, pruned_loss=0.1448, over 5702790.77 frames. ], libri_tot_loss[loss=0.3966, simple_loss=0.4329, pruned_loss=0.1801, over 5781145.63 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3686, pruned_loss=0.1404, over 5687122.26 frames. ], batch size: 71, lr: 2.58e-02, grad_scale: 4.0
+2023-02-28 15:35:08,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18100.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:11,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18103.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:30,718 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18123.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:39,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18130.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:43,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18132.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:49,402 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.690e+02 1.230e+03 1.654e+03 2.339e+03 6.277e+03, threshold=3.308e+03, percent-clipped=17.0
+2023-02-28 15:35:50,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5164, 1.6569, 1.6294, 1.2379], device='cuda:0'), covar=tensor([0.1507, 0.0715, 0.0724, 0.2035], device='cuda:0'), in_proj_covar=tensor([0.0496, 0.0397, 0.0384, 0.0625], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0016], device='cuda:0')
+2023-02-28 15:35:56,376 INFO [train.py:968] (0/2) Epoch 1, batch 18150, giga_loss[loss=0.2828, simple_loss=0.3463, pruned_loss=0.1096, over 29103.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3725, pruned_loss=0.1427, over 5686815.04 frames. ], libri_tot_loss[loss=0.3974, simple_loss=0.4336, pruned_loss=0.1806, over 5779479.87 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3643, pruned_loss=0.1377, over 5673242.66 frames. ], batch size: 155, lr: 2.58e-02, grad_scale: 4.0
+2023-02-28 15:35:57,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18151.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:36:38,834 INFO [train.py:968] (0/2) Epoch 1, batch 18200, libri_loss[loss=0.4335, simple_loss=0.4705, pruned_loss=0.1982, over 29373.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3726, pruned_loss=0.1435, over 5691272.85 frames. ], libri_tot_loss[loss=0.3985, simple_loss=0.4346, pruned_loss=0.1812, over 5781393.13 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3637, pruned_loss=0.138, over 5676765.68 frames. ], batch size: 92, lr: 2.57e-02, grad_scale: 4.0
+2023-02-28 15:37:24,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.925e+02 1.424e+03 1.887e+03 2.619e+03 6.850e+03, threshold=3.774e+03, percent-clipped=12.0
+2023-02-28 15:37:32,279 INFO [train.py:968] (0/2) Epoch 1, batch 18250, giga_loss[loss=0.485, simple_loss=0.4949, pruned_loss=0.2375, over 28858.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.3873, pruned_loss=0.1542, over 5686076.43 frames. ], libri_tot_loss[loss=0.3985, simple_loss=0.4346, pruned_loss=0.1812, over 5783359.70 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3795, pruned_loss=0.1494, over 5671484.03 frames. ], batch size: 186, lr: 2.57e-02, grad_scale: 4.0
+2023-02-28 15:37:55,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18273.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:37:57,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18276.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:38:12,551 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18294.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:38:15,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18297.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:38:16,768 INFO [train.py:968] (0/2) Epoch 1, batch 18300, libri_loss[loss=0.3499, simple_loss=0.3963, pruned_loss=0.1518, over 29636.00 frames. ], tot_loss[loss=0.3673, simple_loss=0.4042, pruned_loss=0.1652, over 5693779.74 frames. ], libri_tot_loss[loss=0.399, simple_loss=0.435, pruned_loss=0.1815, over 5786960.93 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.3965, pruned_loss=0.1605, over 5676062.68 frames. ], batch size: 69, lr: 2.57e-02, grad_scale: 4.0
+2023-02-28 15:38:20,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18305.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:38:21,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3979, 1.2504, 1.1269, 1.3707], device='cuda:0'), covar=tensor([0.1193, 0.1427, 0.1011, 0.1420], device='cuda:0'), in_proj_covar=tensor([0.0837, 0.0757, 0.0808, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 15:38:27,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1310, 1.1426, 1.2125, 1.1793], device='cuda:0'), covar=tensor([0.1167, 0.0906, 0.0976, 0.1697], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0494, 0.0420, 0.0528], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0010, 0.0008, 0.0011], device='cuda:0')
+2023-02-28 15:38:37,254 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18326.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:38:49,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.500e+02 1.525e+03 1.824e+03 2.570e+03 7.826e+03, threshold=3.648e+03, percent-clipped=7.0
+2023-02-28 15:38:56,782 INFO [train.py:968] (0/2) Epoch 1, batch 18350, giga_loss[loss=0.4257, simple_loss=0.4651, pruned_loss=0.1931, over 28737.00 frames. ], tot_loss[loss=0.3818, simple_loss=0.4169, pruned_loss=0.1733, over 5691562.37 frames. ], libri_tot_loss[loss=0.399, simple_loss=0.4351, pruned_loss=0.1815, over 5779794.99 frames. ], giga_tot_loss[loss=0.3747, simple_loss=0.4104, pruned_loss=0.1695, over 5681585.87 frames. ], batch size: 242, lr: 2.56e-02, grad_scale: 4.0
+2023-02-28 15:38:59,790 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18353.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:39:39,104 INFO [train.py:968] (0/2) Epoch 1, batch 18400, giga_loss[loss=0.4352, simple_loss=0.4525, pruned_loss=0.209, over 26626.00 frames. ], tot_loss[loss=0.3897, simple_loss=0.4249, pruned_loss=0.1772, over 5684877.29 frames. ], libri_tot_loss[loss=0.3999, simple_loss=0.4358, pruned_loss=0.182, over 5773427.81 frames. ], giga_tot_loss[loss=0.3827, simple_loss=0.4185, pruned_loss=0.1734, over 5680166.80 frames. ], batch size: 555, lr: 2.56e-02, grad_scale: 8.0
+2023-02-28 15:40:15,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.844e+02 1.338e+03 1.699e+03 2.147e+03 6.155e+03, threshold=3.399e+03, percent-clipped=4.0
+2023-02-28 15:40:17,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4467, 1.4582, 1.3152, 1.2425], device='cuda:0'), covar=tensor([0.0516, 0.0676, 0.0797, 0.0675], device='cuda:0'), in_proj_covar=tensor([0.0560, 0.0880, 0.0669, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 15:40:22,548 INFO [train.py:968] (0/2) Epoch 1, batch 18450, giga_loss[loss=0.3459, simple_loss=0.4004, pruned_loss=0.1457, over 28399.00 frames. ], tot_loss[loss=0.3884, simple_loss=0.426, pruned_loss=0.1754, over 5689490.13 frames. ], libri_tot_loss[loss=0.4003, simple_loss=0.4362, pruned_loss=0.1823, over 5775070.46 frames. ], giga_tot_loss[loss=0.3822, simple_loss=0.4204, pruned_loss=0.172, over 5682925.10 frames. ], batch size: 71, lr: 2.56e-02, grad_scale: 8.0
+2023-02-28 15:40:42,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18468.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:41:11,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18498.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:41:12,321 INFO [train.py:968] (0/2) Epoch 1, batch 18500, giga_loss[loss=0.3511, simple_loss=0.406, pruned_loss=0.1481, over 28370.00 frames. ], tot_loss[loss=0.3893, simple_loss=0.4274, pruned_loss=0.1755, over 5672654.69 frames. ], libri_tot_loss[loss=0.4006, simple_loss=0.4363, pruned_loss=0.1825, over 5775470.13 frames. ], giga_tot_loss[loss=0.384, simple_loss=0.4228, pruned_loss=0.1726, over 5666248.33 frames. ], batch size: 65, lr: 2.55e-02, grad_scale: 8.0
+2023-02-28 15:41:46,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.099e+02 1.274e+03 1.622e+03 2.041e+03 4.179e+03, threshold=3.244e+03, percent-clipped=2.0
+2023-02-28 15:41:56,727 INFO [train.py:968] (0/2) Epoch 1, batch 18550, giga_loss[loss=0.4209, simple_loss=0.4507, pruned_loss=0.1956, over 28377.00 frames. ], tot_loss[loss=0.3911, simple_loss=0.4288, pruned_loss=0.1767, over 5678113.88 frames. ], libri_tot_loss[loss=0.4005, simple_loss=0.4362, pruned_loss=0.1824, over 5777945.03 frames. ], giga_tot_loss[loss=0.3869, simple_loss=0.4251, pruned_loss=0.1744, over 5669254.82 frames. ], batch size: 65, lr: 2.55e-02, grad_scale: 8.0
+2023-02-28 15:42:11,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18567.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:42:12,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18568.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:42:36,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8993, 1.0016, 1.1002, 0.3943], device='cuda:0'), covar=tensor([0.0302, 0.0230, 0.0201, 0.0360], device='cuda:0'), in_proj_covar=tensor([0.0692, 0.0527, 0.0551, 0.0615], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 15:42:41,772 INFO [train.py:968] (0/2) Epoch 1, batch 18600, giga_loss[loss=0.407, simple_loss=0.4519, pruned_loss=0.181, over 28851.00 frames. ], tot_loss[loss=0.3966, simple_loss=0.4328, pruned_loss=0.1802, over 5671545.80 frames. ], libri_tot_loss[loss=0.4006, simple_loss=0.4364, pruned_loss=0.1824, over 5770426.16 frames. ], giga_tot_loss[loss=0.393, simple_loss=0.4295, pruned_loss=0.1783, over 5668976.94 frames. ], batch size: 174, lr: 2.55e-02, grad_scale: 8.0
+2023-02-28 15:42:52,392 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18611.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:42:55,574 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18614.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:43:16,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.962e+02 1.407e+03 1.879e+03 2.448e+03 6.619e+03, threshold=3.758e+03, percent-clipped=10.0
+2023-02-28 15:43:16,944 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18641.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:43:20,348 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18643.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:43:21,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18644.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:43:25,638 INFO [train.py:968] (0/2) Epoch 1, batch 18650, giga_loss[loss=0.3731, simple_loss=0.4222, pruned_loss=0.162, over 28537.00 frames. ], tot_loss[loss=0.4009, simple_loss=0.4366, pruned_loss=0.1826, over 5678319.74 frames. ], libri_tot_loss[loss=0.402, simple_loss=0.4375, pruned_loss=0.1833, over 5773263.73 frames. ], giga_tot_loss[loss=0.3967, simple_loss=0.4329, pruned_loss=0.1802, over 5671506.73 frames. ], batch size: 78, lr: 2.54e-02, grad_scale: 8.0
+2023-02-28 15:43:45,738 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18673.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:44:06,035 INFO [train.py:968] (0/2) Epoch 1, batch 18700, giga_loss[loss=0.4242, simple_loss=0.4643, pruned_loss=0.192, over 28893.00 frames. ], tot_loss[loss=0.4025, simple_loss=0.4393, pruned_loss=0.1828, over 5685230.79 frames. ], libri_tot_loss[loss=0.4025, simple_loss=0.4381, pruned_loss=0.1835, over 5773228.68 frames. ], giga_tot_loss[loss=0.3986, simple_loss=0.4358, pruned_loss=0.1807, over 5677843.02 frames. ], batch size: 186, lr: 2.54e-02, grad_scale: 8.0
+2023-02-28 15:44:30,962 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18728.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:44:42,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.749e+02 1.267e+03 1.580e+03 2.007e+03 4.709e+03, threshold=3.160e+03, percent-clipped=3.0
+2023-02-28 15:44:50,190 INFO [train.py:968] (0/2) Epoch 1, batch 18750, giga_loss[loss=0.3462, simple_loss=0.406, pruned_loss=0.1432, over 28898.00 frames. ], tot_loss[loss=0.4058, simple_loss=0.4426, pruned_loss=0.1845, over 5685101.96 frames. ], libri_tot_loss[loss=0.4039, simple_loss=0.4391, pruned_loss=0.1843, over 5773359.47 frames. ], giga_tot_loss[loss=0.4015, simple_loss=0.439, pruned_loss=0.182, over 5677059.67 frames. ], batch size: 174, lr: 2.54e-02, grad_scale: 4.0
+2023-02-28 15:45:07,831 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-02-28 15:45:30,656 INFO [train.py:968] (0/2) Epoch 1, batch 18800, giga_loss[loss=0.391, simple_loss=0.4359, pruned_loss=0.1731, over 28709.00 frames. ], tot_loss[loss=0.4045, simple_loss=0.4426, pruned_loss=0.1832, over 5689781.15 frames. ], libri_tot_loss[loss=0.4037, simple_loss=0.4391, pruned_loss=0.1842, over 5776103.60 frames. ], giga_tot_loss[loss=0.4013, simple_loss=0.4399, pruned_loss=0.1813, over 5679217.09 frames. ], batch size: 242, lr: 2.53e-02, grad_scale: 8.0
+2023-02-28 15:45:37,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7503, 2.0779, 1.8014, 1.5907], device='cuda:0'), covar=tensor([0.0781, 0.0811, 0.0602, 0.0429], device='cuda:0'), in_proj_covar=tensor([0.0744, 0.0796, 0.0666, 0.0589], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0004], device='cuda:0')
+2023-02-28 15:46:02,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1646, 1.8911, 2.9362, 1.3121], device='cuda:0'), covar=tensor([0.0786, 0.0992, 0.1177, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.0662, 0.0444, 0.0761, 0.0531], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 15:46:04,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.576e+02 1.295e+03 1.496e+03 1.984e+03 3.957e+03, threshold=2.991e+03, percent-clipped=5.0
+2023-02-28 15:46:13,746 INFO [train.py:968] (0/2) Epoch 1, batch 18850, giga_loss[loss=0.37, simple_loss=0.425, pruned_loss=0.1575, over 28921.00 frames. ], tot_loss[loss=0.3995, simple_loss=0.4407, pruned_loss=0.1792, over 5704165.60 frames. ], libri_tot_loss[loss=0.4037, simple_loss=0.4391, pruned_loss=0.1842, over 5776812.08 frames. ], giga_tot_loss[loss=0.397, simple_loss=0.4386, pruned_loss=0.1777, over 5695010.33 frames. ], batch size: 106, lr: 2.53e-02, grad_scale: 8.0
+2023-02-28 15:46:31,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18871.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:46:33,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1866, 1.1647, 1.0279, 0.8102], device='cuda:0'), covar=tensor([0.0296, 0.0290, 0.0310, 0.0354], device='cuda:0'), in_proj_covar=tensor([0.0676, 0.0533, 0.0554, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 15:46:33,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18874.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:46:40,029 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18883.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:46:53,426 INFO [train.py:968] (0/2) Epoch 1, batch 18900, giga_loss[loss=0.3429, simple_loss=0.4059, pruned_loss=0.14, over 29011.00 frames. ], tot_loss[loss=0.3952, simple_loss=0.4382, pruned_loss=0.1761, over 5699167.28 frames. ], libri_tot_loss[loss=0.4052, simple_loss=0.4403, pruned_loss=0.185, over 5767035.08 frames. ], giga_tot_loss[loss=0.3917, simple_loss=0.4354, pruned_loss=0.174, over 5699039.52 frames. ], batch size: 106, lr: 2.53e-02, grad_scale: 4.0
+2023-02-28 15:46:54,171 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18901.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:46:56,276 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18903.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:47:23,804 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 15:47:29,378 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18942.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:47:29,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.082e+02 1.252e+03 1.617e+03 2.180e+03 5.816e+03, threshold=3.235e+03, percent-clipped=11.0
+2023-02-28 15:47:30,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18943.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:47:34,926 INFO [train.py:968] (0/2) Epoch 1, batch 18950, giga_loss[loss=0.4358, simple_loss=0.4583, pruned_loss=0.2067, over 28651.00 frames. ], tot_loss[loss=0.3985, simple_loss=0.4401, pruned_loss=0.1785, over 5704912.16 frames. ], libri_tot_loss[loss=0.4063, simple_loss=0.4413, pruned_loss=0.1857, over 5769487.14 frames. ], giga_tot_loss[loss=0.3944, simple_loss=0.4368, pruned_loss=0.176, over 5700792.62 frames. ], batch size: 262, lr: 2.53e-02, grad_scale: 4.0
+2023-02-28 15:48:18,565 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-02-28 15:48:20,970 INFO [train.py:968] (0/2) Epoch 1, batch 19000, giga_loss[loss=0.4459, simple_loss=0.4581, pruned_loss=0.2168, over 28815.00 frames. ], tot_loss[loss=0.4058, simple_loss=0.443, pruned_loss=0.1843, over 5711834.11 frames. ], libri_tot_loss[loss=0.4067, simple_loss=0.4416, pruned_loss=0.1859, over 5772476.22 frames. ], giga_tot_loss[loss=0.402, simple_loss=0.4401, pruned_loss=0.182, over 5704892.33 frames. ], batch size: 285, lr: 2.52e-02, grad_scale: 4.0
+2023-02-28 15:48:59,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.311e+02 1.466e+03 1.939e+03 2.512e+03 1.310e+04, threshold=3.878e+03, percent-clipped=12.0
+2023-02-28 15:49:04,839 INFO [train.py:968] (0/2) Epoch 1, batch 19050, giga_loss[loss=0.5008, simple_loss=0.4879, pruned_loss=0.2569, over 26554.00 frames. ], tot_loss[loss=0.4108, simple_loss=0.4447, pruned_loss=0.1884, over 5706470.85 frames. ], libri_tot_loss[loss=0.4075, simple_loss=0.4422, pruned_loss=0.1864, over 5765876.59 frames. ], giga_tot_loss[loss=0.4072, simple_loss=0.442, pruned_loss=0.1862, over 5704896.22 frames. ], batch size: 555, lr: 2.52e-02, grad_scale: 4.0
+2023-02-28 15:49:16,614 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7123, 2.5470, 3.4188, 1.5413], device='cuda:0'), covar=tensor([0.0659, 0.0846, 0.1130, 0.2130], device='cuda:0'), in_proj_covar=tensor([0.0674, 0.0438, 0.0775, 0.0521], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 15:49:19,474 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=19067.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:49:22,172 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-02-28 15:49:32,894 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19085.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:49:33,579 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19086.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:49:35,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19088.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:49:36,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19089.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:49:46,681 INFO [train.py:968] (0/2) Epoch 1, batch 19100, giga_loss[loss=0.468, simple_loss=0.4733, pruned_loss=0.2314, over 28883.00 frames. ], tot_loss[loss=0.4112, simple_loss=0.4441, pruned_loss=0.1892, over 5706197.14 frames. ], libri_tot_loss[loss=0.4082, simple_loss=0.4428, pruned_loss=0.1868, over 5769586.24 frames. ], giga_tot_loss[loss=0.4077, simple_loss=0.4414, pruned_loss=0.187, over 5700194.74 frames. ], batch size: 66, lr: 2.52e-02, grad_scale: 4.0
+2023-02-28 15:50:00,793 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19117.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:50:01,508 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19118.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:50:23,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.244e+02 1.362e+03 1.831e+03 2.488e+03 9.704e+03, threshold=3.663e+03, percent-clipped=7.0
+2023-02-28 15:50:29,143 INFO [train.py:968] (0/2) Epoch 1, batch 19150, giga_loss[loss=0.3691, simple_loss=0.4104, pruned_loss=0.1639, over 28440.00 frames. ], tot_loss[loss=0.4097, simple_loss=0.4423, pruned_loss=0.1886, over 5704591.79 frames. ], libri_tot_loss[loss=0.4089, simple_loss=0.4435, pruned_loss=0.1872, over 5772609.43 frames. ], giga_tot_loss[loss=0.4064, simple_loss=0.4394, pruned_loss=0.1866, over 5695090.36 frames. ], batch size: 65, lr: 2.51e-02, grad_scale: 4.0
+2023-02-28 15:50:47,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8128, 2.6541, 3.5653, 1.7465], device='cuda:0'), covar=tensor([0.0556, 0.0708, 0.0850, 0.1768], device='cuda:0'), in_proj_covar=tensor([0.0672, 0.0439, 0.0769, 0.0517], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 15:50:53,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3895, 1.4155, 1.0623, 1.2564], device='cuda:0'), covar=tensor([0.0705, 0.0718, 0.1060, 0.0972], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0640, 0.0633, 0.0593], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 15:51:07,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6970, 1.8574, 3.6835, 2.6607], device='cuda:0'), covar=tensor([0.1890, 0.1227, 0.0421, 0.0563], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0426, 0.0537, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 15:51:15,281 INFO [train.py:968] (0/2) Epoch 1, batch 19200, giga_loss[loss=0.3959, simple_loss=0.4309, pruned_loss=0.1804, over 28501.00 frames. ], tot_loss[loss=0.4078, simple_loss=0.4412, pruned_loss=0.1872, over 5714037.06 frames. ], libri_tot_loss[loss=0.4093, simple_loss=0.4439, pruned_loss=0.1873, over 5775283.16 frames. ], giga_tot_loss[loss=0.4048, simple_loss=0.4386, pruned_loss=0.1856, over 5702978.95 frames. ], batch size: 336, lr: 2.51e-02, grad_scale: 8.0
+2023-02-28 15:51:31,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3679, 1.2010, 1.1707, 1.3151], device='cuda:0'), covar=tensor([0.1441, 0.1782, 0.1222, 0.1856], device='cuda:0'), in_proj_covar=tensor([0.0843, 0.0780, 0.0819, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 15:51:49,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.436e+02 1.286e+03 1.710e+03 2.072e+03 4.590e+03, threshold=3.420e+03, percent-clipped=3.0
+2023-02-28 15:51:55,177 INFO [train.py:968] (0/2) Epoch 1, batch 19250, giga_loss[loss=0.3636, simple_loss=0.4222, pruned_loss=0.1526, over 28931.00 frames. ], tot_loss[loss=0.4049, simple_loss=0.4398, pruned_loss=0.185, over 5724089.27 frames. ], libri_tot_loss[loss=0.4085, simple_loss=0.4435, pruned_loss=0.1868, over 5781382.29 frames. ], giga_tot_loss[loss=0.4029, simple_loss=0.4378, pruned_loss=0.184, over 5707095.30 frames. ], batch size: 145, lr: 2.51e-02, grad_scale: 8.0
+2023-02-28 15:52:01,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19258.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:52:08,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6836, 1.6075, 1.2623, 1.0555], device='cuda:0'), covar=tensor([0.0446, 0.0382, 0.0327, 0.0418], device='cuda:0'), in_proj_covar=tensor([0.0706, 0.0552, 0.0578, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 15:52:18,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19276.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:52:39,657 INFO [train.py:968] (0/2) Epoch 1, batch 19300, giga_loss[loss=0.383, simple_loss=0.4201, pruned_loss=0.173, over 28609.00 frames. ], tot_loss[loss=0.4003, simple_loss=0.4365, pruned_loss=0.182, over 5710836.41 frames. ], libri_tot_loss[loss=0.4091, simple_loss=0.444, pruned_loss=0.1871, over 5784668.86 frames. ], giga_tot_loss[loss=0.398, simple_loss=0.4343, pruned_loss=0.1808, over 5692416.19 frames. ], batch size: 307, lr: 2.50e-02, grad_scale: 8.0
+2023-02-28 15:53:01,651 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=19323.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:53:17,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.797e+02 1.299e+03 1.717e+03 2.166e+03 5.758e+03, threshold=3.434e+03, percent-clipped=9.0
+2023-02-28 15:53:23,153 INFO [train.py:968] (0/2) Epoch 1, batch 19350, giga_loss[loss=0.3779, simple_loss=0.3993, pruned_loss=0.1782, over 26663.00 frames. ], tot_loss[loss=0.3897, simple_loss=0.4283, pruned_loss=0.1756, over 5699525.83 frames. ], libri_tot_loss[loss=0.4098, simple_loss=0.4444, pruned_loss=0.1876, over 5776907.54 frames. ], giga_tot_loss[loss=0.387, simple_loss=0.4259, pruned_loss=0.174, over 5690361.88 frames. ], batch size: 555, lr: 2.50e-02, grad_scale: 4.0
+2023-02-28 15:54:07,804 INFO [train.py:968] (0/2) Epoch 1, batch 19400, giga_loss[loss=0.3977, simple_loss=0.4355, pruned_loss=0.1799, over 28474.00 frames. ], tot_loss[loss=0.3816, simple_loss=0.4215, pruned_loss=0.1709, over 5690094.31 frames. ], libri_tot_loss[loss=0.4099, simple_loss=0.4447, pruned_loss=0.1875, over 5780423.69 frames. ], giga_tot_loss[loss=0.3785, simple_loss=0.4186, pruned_loss=0.1692, over 5676742.49 frames. ], batch size: 78, lr: 2.50e-02, grad_scale: 4.0
+2023-02-28 15:54:09,297 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19401.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:54:12,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19404.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:54:27,531 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19419.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:54:29,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19422.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:54:38,685 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19433.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:54:40,671 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8147, 1.9472, 3.4310, 2.7565], device='cuda:0'), covar=tensor([0.1301, 0.0857, 0.0246, 0.0538], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0428, 0.0531, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 15:54:49,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19442.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:54:50,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.205e+02 1.111e+03 1.498e+03 1.983e+03 5.784e+03, threshold=2.996e+03, percent-clipped=7.0
+2023-02-28 15:54:55,778 INFO [train.py:968] (0/2) Epoch 1, batch 19450, giga_loss[loss=0.3797, simple_loss=0.4164, pruned_loss=0.1715, over 28837.00 frames. ], tot_loss[loss=0.3719, simple_loss=0.4134, pruned_loss=0.1652, over 5678292.01 frames. ], libri_tot_loss[loss=0.4102, simple_loss=0.4449, pruned_loss=0.1877, over 5781438.00 frames. ], giga_tot_loss[loss=0.3681, simple_loss=0.4101, pruned_loss=0.1631, over 5664087.90 frames. ], batch size: 145, lr: 2.49e-02, grad_scale: 4.0
+2023-02-28 15:54:57,981 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19451.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:55:20,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2467, 1.2967, 1.1755, 1.1316], device='cuda:0'), covar=tensor([0.1264, 0.1415, 0.1047, 0.1396], device='cuda:0'), in_proj_covar=tensor([0.0847, 0.0781, 0.0813, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 15:55:24,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4776, 1.7616, 1.7741, 1.2421], device='cuda:0'), covar=tensor([0.1411, 0.0736, 0.0645, 0.1971], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0386, 0.0367, 0.0597], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0017], device='cuda:0')
+2023-02-28 15:55:43,532 INFO [train.py:968] (0/2) Epoch 1, batch 19500, giga_loss[loss=0.3624, simple_loss=0.4049, pruned_loss=0.1599, over 28573.00 frames. ], tot_loss[loss=0.3727, simple_loss=0.4136, pruned_loss=0.1659, over 5663637.25 frames. ], libri_tot_loss[loss=0.411, simple_loss=0.4457, pruned_loss=0.1881, over 5780755.35 frames. ], giga_tot_loss[loss=0.3679, simple_loss=0.4093, pruned_loss=0.1632, over 5649848.12 frames. ], batch size: 71, lr: 2.49e-02, grad_scale: 4.0
+2023-02-28 15:55:59,998 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=19522.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:56:06,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6092, 3.6495, 5.3808, 2.1488], device='cuda:0'), covar=tensor([0.0370, 0.0562, 0.0564, 0.1851], device='cuda:0'), in_proj_covar=tensor([0.0642, 0.0420, 0.0716, 0.0499], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 15:56:22,685 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.007e+02 1.420e+03 1.962e+03 3.051e+03 1.203e+04, threshold=3.924e+03, percent-clipped=25.0
+2023-02-28 15:56:26,853 INFO [train.py:968] (0/2) Epoch 1, batch 19550, giga_loss[loss=0.3704, simple_loss=0.407, pruned_loss=0.1669, over 28439.00 frames. ], tot_loss[loss=0.3761, simple_loss=0.4159, pruned_loss=0.1682, over 5672235.20 frames. ], libri_tot_loss[loss=0.4125, simple_loss=0.4471, pruned_loss=0.189, over 5781807.57 frames. ], giga_tot_loss[loss=0.37, simple_loss=0.4105, pruned_loss=0.1647, over 5657507.92 frames. ], batch size: 65, lr: 2.49e-02, grad_scale: 4.0
+2023-02-28 15:56:44,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1800, 1.3341, 1.1006, 1.0543], device='cuda:0'), covar=tensor([0.1762, 0.0828, 0.0884, 0.2730], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0386, 0.0365, 0.0596], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0017], device='cuda:0')
+2023-02-28 15:56:58,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19585.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:57:01,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19588.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:57:11,001 INFO [train.py:968] (0/2) Epoch 1, batch 19600, giga_loss[loss=0.3356, simple_loss=0.3819, pruned_loss=0.1447, over 28927.00 frames. ], tot_loss[loss=0.373, simple_loss=0.4134, pruned_loss=0.1663, over 5679727.43 frames. ], libri_tot_loss[loss=0.4134, simple_loss=0.4478, pruned_loss=0.1896, over 5782774.59 frames. ], giga_tot_loss[loss=0.3671, simple_loss=0.4083, pruned_loss=0.163, over 5666520.05 frames. ], batch size: 199, lr: 2.49e-02, grad_scale: 8.0
+2023-02-28 15:57:26,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19617.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:57:40,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3734, 1.2738, 1.2679, 1.2670], device='cuda:0'), covar=tensor([0.1408, 0.1749, 0.1102, 0.1801], device='cuda:0'), in_proj_covar=tensor([0.0837, 0.0792, 0.0808, 0.0854], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 15:57:46,810 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6491, 1.7310, 4.0106, 2.6247], device='cuda:0'), covar=tensor([0.1454, 0.1033, 0.0232, 0.0453], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0424, 0.0525, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 15:57:51,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.920e+02 1.184e+03 1.487e+03 1.914e+03 7.054e+03, threshold=2.974e+03, percent-clipped=5.0
+2023-02-28 15:57:55,246 INFO [train.py:968] (0/2) Epoch 1, batch 19650, giga_loss[loss=0.3476, simple_loss=0.3878, pruned_loss=0.1537, over 28616.00 frames. ], tot_loss[loss=0.3697, simple_loss=0.4104, pruned_loss=0.1645, over 5685673.55 frames. ], libri_tot_loss[loss=0.414, simple_loss=0.4484, pruned_loss=0.1899, over 5784700.16 frames. ], giga_tot_loss[loss=0.3639, simple_loss=0.4054, pruned_loss=0.1612, over 5672224.42 frames. ], batch size: 85, lr: 2.48e-02, grad_scale: 4.0
+2023-02-28 15:57:59,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0777, 1.2283, 1.1414, 1.0473], device='cuda:0'), covar=tensor([0.1301, 0.0927, 0.1045, 0.1885], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0503, 0.0415, 0.0520], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0010, 0.0009, 0.0011], device='cuda:0')
+2023-02-28 15:58:34,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19698.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:58:35,780 INFO [train.py:968] (0/2) Epoch 1, batch 19700, giga_loss[loss=0.3216, simple_loss=0.375, pruned_loss=0.1341, over 28972.00 frames. ], tot_loss[loss=0.3655, simple_loss=0.4072, pruned_loss=0.1619, over 5691255.36 frames. ], libri_tot_loss[loss=0.4156, simple_loss=0.4497, pruned_loss=0.1907, over 5785535.32 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.401, pruned_loss=0.1578, over 5677606.97 frames. ], batch size: 136, lr: 2.48e-02, grad_scale: 4.0
+2023-02-28 15:59:01,280 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4344, 1.5175, 1.4754, 1.3988], device='cuda:0'), covar=tensor([0.1355, 0.0731, 0.0675, 0.1760], device='cuda:0'), in_proj_covar=tensor([0.0465, 0.0379, 0.0360, 0.0585], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0017], device='cuda:0')
+2023-02-28 15:59:11,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.728e+02 1.149e+03 1.496e+03 2.029e+03 6.606e+03, threshold=2.993e+03, percent-clipped=7.0
+2023-02-28 15:59:14,852 INFO [train.py:968] (0/2) Epoch 1, batch 19750, giga_loss[loss=0.3001, simple_loss=0.3581, pruned_loss=0.121, over 28931.00 frames. ], tot_loss[loss=0.3617, simple_loss=0.4046, pruned_loss=0.1594, over 5704313.23 frames. ], libri_tot_loss[loss=0.417, simple_loss=0.4511, pruned_loss=0.1915, over 5786544.79 frames. ], giga_tot_loss[loss=0.3531, simple_loss=0.397, pruned_loss=0.1546, over 5690102.90 frames. ], batch size: 164, lr: 2.48e-02, grad_scale: 4.0
+2023-02-28 15:59:40,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6103, 1.4901, 1.0926, 1.0962], device='cuda:0'), covar=tensor([0.0678, 0.0769, 0.1161, 0.1037], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0652, 0.0638, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 15:59:47,961 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=19787.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:59:54,165 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-02-28 15:59:56,408 INFO [train.py:968] (0/2) Epoch 1, batch 19800, giga_loss[loss=0.2985, simple_loss=0.3534, pruned_loss=0.1219, over 28942.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.401, pruned_loss=0.1572, over 5706414.83 frames. ], libri_tot_loss[loss=0.4178, simple_loss=0.4518, pruned_loss=0.1919, over 5788915.14 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.3934, pruned_loss=0.1524, over 5691337.09 frames. ], batch size: 106, lr: 2.47e-02, grad_scale: 4.0
+2023-02-28 15:59:59,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5867, 1.4493, 1.1220, 1.2835], device='cuda:0'), covar=tensor([0.0740, 0.0836, 0.1079, 0.1031], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0647, 0.0626, 0.0605], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 16:00:20,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7684, 2.2243, 1.9293, 1.6934], device='cuda:0'), covar=tensor([0.0909, 0.0852, 0.0637, 0.0450], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0830, 0.0671, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 16:00:31,089 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19841.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:00:33,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19844.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:00:34,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.763e+02 1.095e+03 1.461e+03 1.978e+03 5.720e+03, threshold=2.921e+03, percent-clipped=6.0
+2023-02-28 16:00:38,254 INFO [train.py:968] (0/2) Epoch 1, batch 19850, giga_loss[loss=0.4543, simple_loss=0.452, pruned_loss=0.2283, over 26773.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.3966, pruned_loss=0.1541, over 5716002.47 frames. ], libri_tot_loss[loss=0.4186, simple_loss=0.4526, pruned_loss=0.1923, over 5790581.22 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.3891, pruned_loss=0.1493, over 5701485.31 frames. ], batch size: 555, lr: 2.47e-02, grad_scale: 4.0
+2023-02-28 16:00:40,654 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0582, 1.7359, 1.6081, 1.7072], device='cuda:0'), covar=tensor([0.0788, 0.1572, 0.1209, 0.1267], device='cuda:0'), in_proj_covar=tensor([0.0557, 0.0915, 0.0679, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 16:00:57,390 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19873.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:01:11,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6266, 2.2785, 3.3090, 1.4412], device='cuda:0'), covar=tensor([0.0621, 0.0836, 0.0893, 0.2187], device='cuda:0'), in_proj_covar=tensor([0.0687, 0.0441, 0.0758, 0.0526], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 16:01:17,091 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19897.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 16:01:18,869 INFO [train.py:968] (0/2) Epoch 1, batch 19900, giga_loss[loss=0.3176, simple_loss=0.3757, pruned_loss=0.1298, over 28835.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.3929, pruned_loss=0.1512, over 5714202.02 frames. ], libri_tot_loss[loss=0.4196, simple_loss=0.4535, pruned_loss=0.1928, over 5781323.79 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3855, pruned_loss=0.1465, over 5710239.33 frames. ], batch size: 199, lr: 2.47e-02, grad_scale: 4.0
+2023-02-28 16:01:56,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.993e+02 1.158e+03 1.341e+03 1.966e+03 7.245e+03, threshold=2.681e+03, percent-clipped=12.0
+2023-02-28 16:02:00,086 INFO [train.py:968] (0/2) Epoch 1, batch 19950, giga_loss[loss=0.3155, simple_loss=0.3692, pruned_loss=0.1309, over 28869.00 frames. ], tot_loss[loss=0.347, simple_loss=0.392, pruned_loss=0.151, over 5712344.05 frames. ], libri_tot_loss[loss=0.4213, simple_loss=0.4549, pruned_loss=0.1939, over 5782674.46 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3833, pruned_loss=0.1452, over 5706104.24 frames. ], batch size: 112, lr: 2.47e-02, grad_scale: 4.0
+2023-02-28 16:02:40,122 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-20000.pt
+2023-02-28 16:02:40,424 INFO [train.py:968] (0/2) Epoch 1, batch 20000, giga_loss[loss=0.291, simple_loss=0.3544, pruned_loss=0.1138, over 28958.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3904, pruned_loss=0.1496, over 5709256.90 frames. ], libri_tot_loss[loss=0.422, simple_loss=0.4556, pruned_loss=0.1943, over 5775725.92 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3817, pruned_loss=0.1439, over 5710031.07 frames. ], batch size: 155, lr: 2.46e-02, grad_scale: 8.0
+2023-02-28 16:02:48,378 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20011.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:03:11,145 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20040.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 16:03:13,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20043.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 16:03:14,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.434e+02 1.304e+03 1.540e+03 2.127e+03 1.104e+04, threshold=3.080e+03, percent-clipped=17.0
+2023-02-28 16:03:18,505 INFO [train.py:968] (0/2) Epoch 1, batch 20050, giga_loss[loss=0.3594, simple_loss=0.3915, pruned_loss=0.1636, over 28352.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.3926, pruned_loss=0.1514, over 5715613.73 frames. ], libri_tot_loss[loss=0.4236, simple_loss=0.4568, pruned_loss=0.1952, over 5777620.09 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3832, pruned_loss=0.145, over 5713289.37 frames. ], batch size: 65, lr: 2.46e-02, grad_scale: 8.0
+2023-02-28 16:03:23,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.81 vs. limit=5.0
+2023-02-28 16:03:36,884 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20072.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 16:03:47,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0905, 1.7968, 1.6318, 1.6156], device='cuda:0'), covar=tensor([0.0713, 0.1470, 0.1093, 0.1111], device='cuda:0'), in_proj_covar=tensor([0.0553, 0.0900, 0.0683, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 16:04:03,457 INFO [train.py:968] (0/2) Epoch 1, batch 20100, giga_loss[loss=0.4779, simple_loss=0.477, pruned_loss=0.2394, over 27933.00 frames. ], tot_loss[loss=0.3597, simple_loss=0.4016, pruned_loss=0.1589, over 5712263.53 frames. ], libri_tot_loss[loss=0.4239, simple_loss=0.4572, pruned_loss=0.1954, over 5778303.50 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.3925, pruned_loss=0.1528, over 5708595.87 frames. ], batch size: 412, lr: 2.46e-02, grad_scale: 8.0
+2023-02-28 16:04:47,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.917e+02 1.388e+03 1.810e+03 2.332e+03 6.353e+03, threshold=3.620e+03, percent-clipped=13.0
+2023-02-28 16:04:52,596 INFO [train.py:968] (0/2) Epoch 1, batch 20150, giga_loss[loss=0.505, simple_loss=0.4937, pruned_loss=0.2581, over 26666.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.4108, pruned_loss=0.1662, over 5705444.10 frames. ], libri_tot_loss[loss=0.4255, simple_loss=0.4584, pruned_loss=0.1963, over 5782411.65 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4005, pruned_loss=0.1592, over 5696302.91 frames. ], batch size: 555, lr: 2.45e-02, grad_scale: 8.0
+2023-02-28 16:05:03,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20162.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:05:44,035 INFO [train.py:968] (0/2) Epoch 1, batch 20200, giga_loss[loss=0.4532, simple_loss=0.4493, pruned_loss=0.2285, over 23545.00 frames. ], tot_loss[loss=0.3829, simple_loss=0.4194, pruned_loss=0.1732, over 5697942.23 frames. ], libri_tot_loss[loss=0.4253, simple_loss=0.4581, pruned_loss=0.1962, over 5782069.13 frames. ], giga_tot_loss[loss=0.3723, simple_loss=0.4105, pruned_loss=0.1671, over 5689135.14 frames. ], batch size: 705, lr: 2.45e-02, grad_scale: 8.0
+2023-02-28 16:06:07,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7306, 2.2990, 3.4252, 1.6475], device='cuda:0'), covar=tensor([0.0582, 0.0807, 0.0832, 0.1893], device='cuda:0'), in_proj_covar=tensor([0.0667, 0.0443, 0.0762, 0.0517], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 16:06:22,454 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-02-28 16:06:22,485 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-02-28 16:06:24,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.959e+02 1.345e+03 1.603e+03 2.073e+03 4.387e+03, threshold=3.205e+03, percent-clipped=2.0
+2023-02-28 16:06:29,860 INFO [train.py:968] (0/2) Epoch 1, batch 20250, giga_loss[loss=0.3875, simple_loss=0.4426, pruned_loss=0.1661, over 28558.00 frames. ], tot_loss[loss=0.3894, simple_loss=0.4259, pruned_loss=0.1765, over 5689749.13 frames. ], libri_tot_loss[loss=0.4252, simple_loss=0.458, pruned_loss=0.1962, over 5772176.30 frames. ], giga_tot_loss[loss=0.3804, simple_loss=0.4182, pruned_loss=0.1713, over 5689480.33 frames. ], batch size: 336, lr: 2.45e-02, grad_scale: 8.0
+2023-02-28 16:06:53,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1287, 0.9322, 0.9710, 0.1058], device='cuda:0'), covar=tensor([0.0414, 0.0494, 0.0652, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0842, 0.0858, 0.0887, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 16:07:14,138 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7969, 2.4159, 1.8341, 1.8543], device='cuda:0'), covar=tensor([0.0887, 0.0746, 0.0700, 0.0433], device='cuda:0'), in_proj_covar=tensor([0.0760, 0.0824, 0.0676, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 16:07:20,212 INFO [train.py:968] (0/2) Epoch 1, batch 20300, giga_loss[loss=0.4113, simple_loss=0.4507, pruned_loss=0.186, over 28470.00 frames. ], tot_loss[loss=0.3953, simple_loss=0.4316, pruned_loss=0.1794, over 5686506.01 frames. ], libri_tot_loss[loss=0.425, simple_loss=0.4579, pruned_loss=0.1961, over 5771756.30 frames. ], giga_tot_loss[loss=0.3879, simple_loss=0.4253, pruned_loss=0.1752, over 5685406.37 frames. ], batch size: 85, lr: 2.45e-02, grad_scale: 8.0
+2023-02-28 16:07:25,945 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20305.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:07:29,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20308.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:07:57,696 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20337.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:08:05,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.663e+02 1.251e+03 1.576e+03 2.008e+03 5.916e+03, threshold=3.152e+03, percent-clipped=8.0
+2023-02-28 16:08:09,647 INFO [train.py:968] (0/2) Epoch 1, batch 20350, giga_loss[loss=0.4316, simple_loss=0.4646, pruned_loss=0.1993, over 29138.00 frames. ], tot_loss[loss=0.4028, simple_loss=0.4382, pruned_loss=0.1838, over 5688262.96 frames. ], libri_tot_loss[loss=0.4252, simple_loss=0.4581, pruned_loss=0.1962, over 5773617.94 frames. ], giga_tot_loss[loss=0.3965, simple_loss=0.4327, pruned_loss=0.1801, over 5684445.21 frames. ], batch size: 128, lr: 2.44e-02, grad_scale: 4.0
+2023-02-28 16:08:34,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3833, 1.7865, 1.5231, 1.3522], device='cuda:0'), covar=tensor([0.1300, 0.0736, 0.0674, 0.1768], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0365, 0.0350, 0.0586], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0017], device='cuda:0')
+2023-02-28 16:08:41,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20386.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:08:53,593 INFO [train.py:968] (0/2) Epoch 1, batch 20400, giga_loss[loss=0.3396, simple_loss=0.3897, pruned_loss=0.1448, over 28815.00 frames. ], tot_loss[loss=0.4027, simple_loss=0.4381, pruned_loss=0.1837, over 5697398.51 frames. ], libri_tot_loss[loss=0.4258, simple_loss=0.4586, pruned_loss=0.1965, over 5776639.75 frames. ], giga_tot_loss[loss=0.3965, simple_loss=0.4328, pruned_loss=0.1801, over 5689515.38 frames. ], batch size: 112, lr: 2.44e-02, grad_scale: 4.0
+2023-02-28 16:09:04,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20413.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:09:34,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.397e+02 1.466e+03 2.015e+03 3.128e+03 2.378e+04, threshold=4.030e+03, percent-clipped=24.0
+2023-02-28 16:09:35,661 INFO [train.py:968] (0/2) Epoch 1, batch 20450, libri_loss[loss=0.46, simple_loss=0.4775, pruned_loss=0.2213, over 19644.00 frames. ], tot_loss[loss=0.3951, simple_loss=0.4317, pruned_loss=0.1792, over 5690286.03 frames. ], libri_tot_loss[loss=0.4266, simple_loss=0.459, pruned_loss=0.1972, over 5770050.67 frames. ], giga_tot_loss[loss=0.388, simple_loss=0.4258, pruned_loss=0.1751, over 5686800.56 frames. ], batch size: 186, lr: 2.44e-02, grad_scale: 2.0
+2023-02-28 16:10:03,315 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20482.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:10:19,499 INFO [train.py:968] (0/2) Epoch 1, batch 20500, giga_loss[loss=0.3855, simple_loss=0.4255, pruned_loss=0.1728, over 28322.00 frames. ], tot_loss[loss=0.3886, simple_loss=0.4277, pruned_loss=0.1748, over 5699276.49 frames. ], libri_tot_loss[loss=0.4267, simple_loss=0.459, pruned_loss=0.1971, over 5772712.46 frames. ], giga_tot_loss[loss=0.3824, simple_loss=0.4226, pruned_loss=0.1711, over 5692838.65 frames. ], batch size: 368, lr: 2.43e-02, grad_scale: 2.0
+2023-02-28 16:10:25,120 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20505.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:10:25,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3461, 1.4015, 1.4753, 0.1521], device='cuda:0'), covar=tensor([0.0659, 0.0649, 0.0737, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0829, 0.0859, 0.0860, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 16:10:33,145 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-02-28 16:10:44,499 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20529.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:10:46,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20532.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:11:01,104 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.624e+02 1.178e+03 1.580e+03 2.181e+03 4.169e+03, threshold=3.160e+03, percent-clipped=2.0
+2023-02-28 16:11:03,801 INFO [train.py:968] (0/2) Epoch 1, batch 20550, giga_loss[loss=0.3841, simple_loss=0.4353, pruned_loss=0.1664, over 28488.00 frames. ], tot_loss[loss=0.3878, simple_loss=0.4274, pruned_loss=0.1741, over 5693912.56 frames. ], libri_tot_loss[loss=0.426, simple_loss=0.4585, pruned_loss=0.1967, over 5774869.33 frames. ], giga_tot_loss[loss=0.383, simple_loss=0.4233, pruned_loss=0.1713, over 5685986.36 frames. ], batch size: 71, lr: 2.43e-02, grad_scale: 2.0
+2023-02-28 16:11:13,967 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20561.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:11:33,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5081, 1.5997, 1.6329, 0.3500], device='cuda:0'), covar=tensor([0.0563, 0.0558, 0.0672, 0.1078], device='cuda:0'), in_proj_covar=tensor([0.0827, 0.0866, 0.0867, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 16:11:42,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20597.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:11:44,807 INFO [train.py:968] (0/2) Epoch 1, batch 20600, giga_loss[loss=0.3849, simple_loss=0.4247, pruned_loss=0.1726, over 28525.00 frames. ], tot_loss[loss=0.3952, simple_loss=0.4331, pruned_loss=0.1787, over 5701047.78 frames. ], libri_tot_loss[loss=0.427, simple_loss=0.4591, pruned_loss=0.1974, over 5777752.88 frames. ], giga_tot_loss[loss=0.3893, simple_loss=0.4284, pruned_loss=0.1751, over 5689803.28 frames. ], batch size: 85, lr: 2.43e-02, grad_scale: 2.0
+2023-02-28 16:11:56,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7588, 1.9632, 2.1996, 1.8948], device='cuda:0'), covar=tensor([0.0691, 0.1731, 0.1016, 0.1186], device='cuda:0'), in_proj_covar=tensor([0.0542, 0.0900, 0.0683, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 16:12:28,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.836e+02 1.478e+03 1.965e+03 2.613e+03 7.137e+03, threshold=3.930e+03, percent-clipped=20.0
+2023-02-28 16:12:29,933 INFO [train.py:968] (0/2) Epoch 1, batch 20650, giga_loss[loss=0.4074, simple_loss=0.442, pruned_loss=0.1864, over 28688.00 frames. ], tot_loss[loss=0.399, simple_loss=0.4361, pruned_loss=0.181, over 5701913.02 frames. ], libri_tot_loss[loss=0.4266, simple_loss=0.4588, pruned_loss=0.1972, over 5780038.49 frames. ], giga_tot_loss[loss=0.3942, simple_loss=0.4322, pruned_loss=0.178, over 5689343.67 frames. ], batch size: 92, lr: 2.43e-02, grad_scale: 2.0
+2023-02-28 16:13:12,481 INFO [train.py:968] (0/2) Epoch 1, batch 20700, giga_loss[loss=0.4151, simple_loss=0.4522, pruned_loss=0.1891, over 28631.00 frames. ], tot_loss[loss=0.3996, simple_loss=0.4369, pruned_loss=0.1811, over 5706151.68 frames. ], libri_tot_loss[loss=0.4262, simple_loss=0.4586, pruned_loss=0.1969, over 5782135.33 frames. ], giga_tot_loss[loss=0.3951, simple_loss=0.4333, pruned_loss=0.1785, over 5691800.01 frames. ], batch size: 336, lr: 2.42e-02, grad_scale: 2.0
+2023-02-28 16:13:54,477 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-02-28 16:13:57,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.511e+03 1.817e+03 2.420e+03 4.652e+03, threshold=3.634e+03, percent-clipped=1.0
+2023-02-28 16:13:58,662 INFO [train.py:968] (0/2) Epoch 1, batch 20750, giga_loss[loss=0.385, simple_loss=0.4259, pruned_loss=0.1721, over 28912.00 frames. ], tot_loss[loss=0.4011, simple_loss=0.4383, pruned_loss=0.1819, over 5713828.92 frames. ], libri_tot_loss[loss=0.4266, simple_loss=0.459, pruned_loss=0.1971, over 5781173.72 frames. ], giga_tot_loss[loss=0.3965, simple_loss=0.4344, pruned_loss=0.1793, over 5701327.97 frames. ], batch size: 227, lr: 2.42e-02, grad_scale: 2.0
+2023-02-28 16:14:02,833 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 16:14:25,643 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20781.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:14:33,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20788.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:14:41,066 INFO [train.py:968] (0/2) Epoch 1, batch 20800, giga_loss[loss=0.3742, simple_loss=0.4189, pruned_loss=0.1648, over 28487.00 frames. ], tot_loss[loss=0.4035, simple_loss=0.4399, pruned_loss=0.1835, over 5708584.80 frames. ], libri_tot_loss[loss=0.4274, simple_loss=0.46, pruned_loss=0.1975, over 5778578.75 frames. ], giga_tot_loss[loss=0.3981, simple_loss=0.4352, pruned_loss=0.1805, over 5698468.85 frames. ], batch size: 65, lr: 2.42e-02, grad_scale: 4.0
+2023-02-28 16:14:50,995 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20811.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:15:20,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.173e+02 1.503e+03 1.755e+03 2.454e+03 7.005e+03, threshold=3.510e+03, percent-clipped=11.0
+2023-02-28 16:15:21,289 INFO [train.py:968] (0/2) Epoch 1, batch 20850, libri_loss[loss=0.4359, simple_loss=0.4727, pruned_loss=0.1996, over 26274.00 frames. ], tot_loss[loss=0.4034, simple_loss=0.4406, pruned_loss=0.1831, over 5711184.57 frames. ], libri_tot_loss[loss=0.4283, simple_loss=0.4607, pruned_loss=0.1979, over 5776314.59 frames. ], giga_tot_loss[loss=0.3978, simple_loss=0.4357, pruned_loss=0.18, over 5703905.05 frames. ], batch size: 136, lr: 2.41e-02, grad_scale: 4.0
+2023-02-28 16:15:26,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20857.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:15:44,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20880.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:16:03,845 INFO [train.py:968] (0/2) Epoch 1, batch 20900, giga_loss[loss=0.4005, simple_loss=0.4474, pruned_loss=0.1769, over 28969.00 frames. ], tot_loss[loss=0.4006, simple_loss=0.4398, pruned_loss=0.1807, over 5699845.37 frames. ], libri_tot_loss[loss=0.4287, simple_loss=0.4611, pruned_loss=0.1982, over 5766681.02 frames. ], giga_tot_loss[loss=0.3955, simple_loss=0.4353, pruned_loss=0.1778, over 5701770.64 frames. ], batch size: 136, lr: 2.41e-02, grad_scale: 4.0
+2023-02-28 16:16:29,900 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20931.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:16:31,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20934.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:16:43,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.178e+02 1.239e+03 1.487e+03 2.052e+03 5.661e+03, threshold=2.974e+03, percent-clipped=6.0
+2023-02-28 16:16:44,774 INFO [train.py:968] (0/2) Epoch 1, batch 20950, giga_loss[loss=0.4082, simple_loss=0.4446, pruned_loss=0.1859, over 28623.00 frames. ], tot_loss[loss=0.4016, simple_loss=0.4413, pruned_loss=0.1809, over 5694316.07 frames. ], libri_tot_loss[loss=0.4287, simple_loss=0.4609, pruned_loss=0.1982, over 5751747.34 frames. ], giga_tot_loss[loss=0.3967, simple_loss=0.4372, pruned_loss=0.178, over 5707164.05 frames. ], batch size: 336, lr: 2.41e-02, grad_scale: 4.0
+2023-02-28 16:16:56,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20963.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:04,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20972.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:12,773 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20981.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:27,550 INFO [train.py:968] (0/2) Epoch 1, batch 21000, giga_loss[loss=0.3711, simple_loss=0.4227, pruned_loss=0.1598, over 28823.00 frames. ], tot_loss[loss=0.3992, simple_loss=0.4397, pruned_loss=0.1793, over 5703227.35 frames. ], libri_tot_loss[loss=0.4284, simple_loss=0.4607, pruned_loss=0.1981, over 5752314.78 frames. ], giga_tot_loss[loss=0.3952, simple_loss=0.4364, pruned_loss=0.177, over 5712420.29 frames. ], batch size: 199, lr: 2.41e-02, grad_scale: 4.0
+2023-02-28 16:17:27,555 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 16:17:37,401 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.3389, simple_loss=0.4169, pruned_loss=0.1304, over 944034.00 frames. 
+2023-02-28 16:17:37,402 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 16:17:37,682 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21000.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:39,634 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21003.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:40,445 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-02-28 16:17:54,642 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21023.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:57,793 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21026.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:18:02,697 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21032.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:18:15,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.102e+02 1.264e+03 1.657e+03 2.275e+03 6.839e+03, threshold=3.313e+03, percent-clipped=12.0
+2023-02-28 16:18:16,939 INFO [train.py:968] (0/2) Epoch 1, batch 21050, giga_loss[loss=0.3459, simple_loss=0.3946, pruned_loss=0.1486, over 28646.00 frames. ], tot_loss[loss=0.395, simple_loss=0.4359, pruned_loss=0.177, over 5705611.46 frames. ], libri_tot_loss[loss=0.4279, simple_loss=0.4602, pruned_loss=0.1978, over 5756993.29 frames. ], giga_tot_loss[loss=0.3913, simple_loss=0.4329, pruned_loss=0.1748, over 5707110.51 frames. ], batch size: 85, lr: 2.40e-02, grad_scale: 4.0
+2023-02-28 16:18:21,148 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21055.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:18:55,971 INFO [train.py:968] (0/2) Epoch 1, batch 21100, giga_loss[loss=0.3734, simple_loss=0.427, pruned_loss=0.1599, over 29037.00 frames. ], tot_loss[loss=0.3921, simple_loss=0.4337, pruned_loss=0.1753, over 5711248.76 frames. ], libri_tot_loss[loss=0.4283, simple_loss=0.4605, pruned_loss=0.198, over 5759929.86 frames. ], giga_tot_loss[loss=0.388, simple_loss=0.4305, pruned_loss=0.1728, over 5708749.90 frames. ], batch size: 155, lr: 2.40e-02, grad_scale: 4.0
+2023-02-28 16:19:09,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6774, 1.6782, 3.5108, 2.4976], device='cuda:0'), covar=tensor([0.1148, 0.0873, 0.0242, 0.0417], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0425, 0.0533, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 16:19:09,899 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21115.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:19:12,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21118.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:19:32,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21147.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:19:33,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.773e+02 1.473e+03 1.841e+03 2.295e+03 5.322e+03, threshold=3.682e+03, percent-clipped=11.0
+2023-02-28 16:19:34,987 INFO [train.py:968] (0/2) Epoch 1, batch 21150, giga_loss[loss=0.3713, simple_loss=0.4192, pruned_loss=0.1617, over 29004.00 frames. ], tot_loss[loss=0.3928, simple_loss=0.4337, pruned_loss=0.1759, over 5716937.32 frames. ], libri_tot_loss[loss=0.4287, simple_loss=0.4607, pruned_loss=0.1984, over 5761602.67 frames. ], giga_tot_loss[loss=0.3875, simple_loss=0.4298, pruned_loss=0.1726, over 5711248.66 frames. ], batch size: 155, lr: 2.40e-02, grad_scale: 4.0
+2023-02-28 16:19:40,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21156.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:20:06,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21186.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:20:18,671 INFO [train.py:968] (0/2) Epoch 1, batch 21200, libri_loss[loss=0.399, simple_loss=0.4458, pruned_loss=0.1761, over 29661.00 frames. ], tot_loss[loss=0.3935, simple_loss=0.4339, pruned_loss=0.1766, over 5711056.38 frames. ], libri_tot_loss[loss=0.4283, simple_loss=0.4602, pruned_loss=0.1982, over 5757026.48 frames. ], giga_tot_loss[loss=0.3885, simple_loss=0.4303, pruned_loss=0.1734, over 5709337.03 frames. ], batch size: 91, lr: 2.40e-02, grad_scale: 8.0
+2023-02-28 16:20:59,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.645e+02 1.271e+03 1.592e+03 2.066e+03 4.641e+03, threshold=3.184e+03, percent-clipped=3.0
+2023-02-28 16:21:00,782 INFO [train.py:968] (0/2) Epoch 1, batch 21250, giga_loss[loss=0.3786, simple_loss=0.4303, pruned_loss=0.1634, over 29050.00 frames. ], tot_loss[loss=0.3919, simple_loss=0.4331, pruned_loss=0.1754, over 5708894.14 frames. ], libri_tot_loss[loss=0.4285, simple_loss=0.4603, pruned_loss=0.1983, over 5758828.15 frames. ], giga_tot_loss[loss=0.3874, simple_loss=0.4299, pruned_loss=0.1724, over 5705299.52 frames. ], batch size: 128, lr: 2.39e-02, grad_scale: 8.0
+2023-02-28 16:21:18,887 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-02-28 16:21:41,529 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21299.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:21:42,023 INFO [train.py:968] (0/2) Epoch 1, batch 21300, giga_loss[loss=0.3993, simple_loss=0.4309, pruned_loss=0.1838, over 28671.00 frames. ], tot_loss[loss=0.3935, simple_loss=0.4346, pruned_loss=0.1762, over 5714052.50 frames. ], libri_tot_loss[loss=0.4291, simple_loss=0.4605, pruned_loss=0.1989, over 5758907.45 frames. ], giga_tot_loss[loss=0.3881, simple_loss=0.431, pruned_loss=0.1726, over 5709543.34 frames. ], batch size: 78, lr: 2.39e-02, grad_scale: 4.0
+2023-02-28 16:21:43,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21302.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:22:06,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21329.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:22:07,291 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21331.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:22:08,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21332.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:22:22,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.462e+02 1.181e+03 1.503e+03 2.274e+03 1.134e+04, threshold=3.007e+03, percent-clipped=12.0
+2023-02-28 16:22:23,956 INFO [train.py:968] (0/2) Epoch 1, batch 21350, giga_loss[loss=0.451, simple_loss=0.4658, pruned_loss=0.2181, over 27593.00 frames. ], tot_loss[loss=0.3895, simple_loss=0.4321, pruned_loss=0.1735, over 5709617.94 frames. ], libri_tot_loss[loss=0.4294, simple_loss=0.4607, pruned_loss=0.1991, over 5760590.77 frames. ], giga_tot_loss[loss=0.3846, simple_loss=0.4287, pruned_loss=0.1702, over 5704145.50 frames. ], batch size: 472, lr: 2.39e-02, grad_scale: 4.0
+2023-02-28 16:22:29,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21356.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:22:33,429 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21361.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:23:02,472 INFO [train.py:968] (0/2) Epoch 1, batch 21400, giga_loss[loss=0.317, simple_loss=0.3779, pruned_loss=0.1281, over 28457.00 frames. ], tot_loss[loss=0.3926, simple_loss=0.4334, pruned_loss=0.1759, over 5713247.70 frames. ], libri_tot_loss[loss=0.4301, simple_loss=0.461, pruned_loss=0.1996, over 5766815.08 frames. ], giga_tot_loss[loss=0.3859, simple_loss=0.429, pruned_loss=0.1714, over 5700610.93 frames. ], batch size: 60, lr: 2.38e-02, grad_scale: 4.0
+2023-02-28 16:23:38,776 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=21445.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:23:42,295 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.686e+02 1.302e+03 1.529e+03 1.966e+03 5.570e+03, threshold=3.059e+03, percent-clipped=11.0
+2023-02-28 16:23:42,922 INFO [train.py:968] (0/2) Epoch 1, batch 21450, giga_loss[loss=0.4512, simple_loss=0.4583, pruned_loss=0.222, over 26845.00 frames. ], tot_loss[loss=0.3918, simple_loss=0.4319, pruned_loss=0.1758, over 5708978.12 frames. ], libri_tot_loss[loss=0.4307, simple_loss=0.4612, pruned_loss=0.2001, over 5768283.71 frames. ], giga_tot_loss[loss=0.3854, simple_loss=0.4278, pruned_loss=0.1715, over 5696742.99 frames. ], batch size: 555, lr: 2.38e-02, grad_scale: 4.0
+2023-02-28 16:24:24,151 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21499.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:24:24,543 INFO [train.py:968] (0/2) Epoch 1, batch 21500, giga_loss[loss=0.3569, simple_loss=0.3975, pruned_loss=0.1581, over 28686.00 frames. ], tot_loss[loss=0.3871, simple_loss=0.4278, pruned_loss=0.1732, over 5710129.60 frames. ], libri_tot_loss[loss=0.4308, simple_loss=0.4613, pruned_loss=0.2002, over 5771759.28 frames. ], giga_tot_loss[loss=0.3809, simple_loss=0.4237, pruned_loss=0.1691, over 5695951.92 frames. ], batch size: 92, lr: 2.38e-02, grad_scale: 2.0
+2023-02-28 16:24:26,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21502.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:24:34,244 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=21511.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:24:49,080 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21531.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:24:54,179 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4297, 1.3934, 1.2197, 0.9531], device='cuda:0'), covar=tensor([0.0375, 0.0417, 0.0311, 0.0382], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0547, 0.0553, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 16:25:05,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.555e+02 1.398e+03 1.676e+03 2.224e+03 1.589e+04, threshold=3.353e+03, percent-clipped=16.0
+2023-02-28 16:25:05,476 INFO [train.py:968] (0/2) Epoch 1, batch 21550, giga_loss[loss=0.4061, simple_loss=0.4449, pruned_loss=0.1836, over 28626.00 frames. ], tot_loss[loss=0.3871, simple_loss=0.427, pruned_loss=0.1736, over 5695081.24 frames. ], libri_tot_loss[loss=0.4321, simple_loss=0.4619, pruned_loss=0.2012, over 5764282.11 frames. ], giga_tot_loss[loss=0.3795, simple_loss=0.4221, pruned_loss=0.1685, over 5688673.11 frames. ], batch size: 284, lr: 2.38e-02, grad_scale: 2.0
+2023-02-28 16:25:48,579 INFO [train.py:968] (0/2) Epoch 1, batch 21600, giga_loss[loss=0.3891, simple_loss=0.4253, pruned_loss=0.1765, over 28982.00 frames. ], tot_loss[loss=0.3891, simple_loss=0.428, pruned_loss=0.1751, over 5699260.30 frames. ], libri_tot_loss[loss=0.4324, simple_loss=0.4622, pruned_loss=0.2013, over 5765278.18 frames. ], giga_tot_loss[loss=0.3826, simple_loss=0.4236, pruned_loss=0.1708, over 5692803.72 frames. ], batch size: 145, lr: 2.37e-02, grad_scale: 4.0
+2023-02-28 16:26:30,044 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.095e+02 1.265e+03 1.606e+03 2.039e+03 5.354e+03, threshold=3.212e+03, percent-clipped=4.0
+2023-02-28 16:26:30,057 INFO [train.py:968] (0/2) Epoch 1, batch 21650, giga_loss[loss=0.3908, simple_loss=0.4213, pruned_loss=0.1801, over 28791.00 frames. ], tot_loss[loss=0.3864, simple_loss=0.4252, pruned_loss=0.1738, over 5696215.96 frames. ], libri_tot_loss[loss=0.433, simple_loss=0.4626, pruned_loss=0.2017, over 5763616.05 frames. ], giga_tot_loss[loss=0.3801, simple_loss=0.4208, pruned_loss=0.1697, over 5691914.72 frames. ], batch size: 284, lr: 2.37e-02, grad_scale: 4.0
+2023-02-28 16:27:08,282 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-02-28 16:27:10,441 INFO [train.py:968] (0/2) Epoch 1, batch 21700, giga_loss[loss=0.3462, simple_loss=0.3878, pruned_loss=0.1523, over 28680.00 frames. ], tot_loss[loss=0.3825, simple_loss=0.4217, pruned_loss=0.1717, over 5705696.16 frames. ], libri_tot_loss[loss=0.4331, simple_loss=0.4624, pruned_loss=0.2019, over 5767493.77 frames. ], giga_tot_loss[loss=0.3757, simple_loss=0.417, pruned_loss=0.1672, over 5696727.12 frames. ], batch size: 85, lr: 2.37e-02, grad_scale: 4.0
+2023-02-28 16:27:14,456 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=21705.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:27:28,289 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=21721.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:27:34,517 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2114, 1.0792, 1.1207, 0.6548], device='cuda:0'), covar=tensor([0.0303, 0.0274, 0.0200, 0.0344], device='cuda:0'), in_proj_covar=tensor([0.0718, 0.0536, 0.0552, 0.0600], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 16:27:52,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.958e+02 1.203e+03 1.473e+03 1.889e+03 1.514e+04, threshold=2.946e+03, percent-clipped=4.0
+2023-02-28 16:27:52,666 INFO [train.py:968] (0/2) Epoch 1, batch 21750, giga_loss[loss=0.3397, simple_loss=0.3886, pruned_loss=0.1454, over 29019.00 frames. ], tot_loss[loss=0.3786, simple_loss=0.4182, pruned_loss=0.1695, over 5705682.11 frames. ], libri_tot_loss[loss=0.4334, simple_loss=0.4626, pruned_loss=0.2021, over 5760945.18 frames. ], giga_tot_loss[loss=0.3718, simple_loss=0.4135, pruned_loss=0.1651, over 5703970.40 frames. ], batch size: 136, lr: 2.37e-02, grad_scale: 4.0
+2023-02-28 16:28:09,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3705, 1.6718, 1.5662, 1.4541], device='cuda:0'), covar=tensor([0.1194, 0.1027, 0.1083, 0.1766], device='cuda:0'), in_proj_covar=tensor([0.0439, 0.0493, 0.0393, 0.0495], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0008, 0.0011], device='cuda:0')
+2023-02-28 16:28:32,491 INFO [train.py:968] (0/2) Epoch 1, batch 21800, giga_loss[loss=0.3937, simple_loss=0.4301, pruned_loss=0.1787, over 28950.00 frames. ], tot_loss[loss=0.3767, simple_loss=0.4168, pruned_loss=0.1683, over 5715872.67 frames. ], libri_tot_loss[loss=0.4342, simple_loss=0.4632, pruned_loss=0.2026, over 5764109.89 frames. ], giga_tot_loss[loss=0.3686, simple_loss=0.411, pruned_loss=0.1631, over 5710085.30 frames. ], batch size: 227, lr: 2.36e-02, grad_scale: 4.0
+2023-02-28 16:28:49,008 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-02-28 16:28:49,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21820.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:28:50,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4132, 1.8723, 1.5945, 1.3539], device='cuda:0'), covar=tensor([0.1241, 0.0778, 0.0732, 0.1639], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0370, 0.0355, 0.0576], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0012, 0.0018], device='cuda:0')
+2023-02-28 16:28:57,870 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-02-28 16:29:16,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.020e+02 1.169e+03 1.542e+03 1.936e+03 4.439e+03, threshold=3.084e+03, percent-clipped=4.0
+2023-02-28 16:29:16,082 INFO [train.py:968] (0/2) Epoch 1, batch 21850, giga_loss[loss=0.4088, simple_loss=0.4388, pruned_loss=0.1894, over 29044.00 frames. ], tot_loss[loss=0.3763, simple_loss=0.4171, pruned_loss=0.1678, over 5712607.13 frames. ], libri_tot_loss[loss=0.4341, simple_loss=0.463, pruned_loss=0.2026, over 5765353.83 frames. ], giga_tot_loss[loss=0.369, simple_loss=0.4118, pruned_loss=0.1631, over 5706088.11 frames. ], batch size: 155, lr: 2.36e-02, grad_scale: 4.0
+2023-02-28 16:29:48,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21886.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:30:01,858 INFO [train.py:968] (0/2) Epoch 1, batch 21900, giga_loss[loss=0.4014, simple_loss=0.4363, pruned_loss=0.1833, over 27605.00 frames. ], tot_loss[loss=0.3793, simple_loss=0.4198, pruned_loss=0.1694, over 5704851.23 frames. ], libri_tot_loss[loss=0.4343, simple_loss=0.463, pruned_loss=0.2028, over 5766847.59 frames. ], giga_tot_loss[loss=0.3728, simple_loss=0.4152, pruned_loss=0.1652, over 5697846.82 frames. ], batch size: 472, lr: 2.36e-02, grad_scale: 4.0
+2023-02-28 16:30:46,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.374e+02 1.237e+03 1.517e+03 2.196e+03 7.590e+03, threshold=3.034e+03, percent-clipped=14.0
+2023-02-28 16:30:46,656 INFO [train.py:968] (0/2) Epoch 1, batch 21950, giga_loss[loss=0.3794, simple_loss=0.4268, pruned_loss=0.166, over 28883.00 frames. ], tot_loss[loss=0.3816, simple_loss=0.4226, pruned_loss=0.1703, over 5702011.55 frames. ], libri_tot_loss[loss=0.4343, simple_loss=0.4627, pruned_loss=0.203, over 5769437.25 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.4182, pruned_loss=0.1661, over 5692717.68 frames. ], batch size: 199, lr: 2.36e-02, grad_scale: 4.0
+2023-02-28 16:30:48,943 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-02-28 16:31:00,649 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21963.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:02,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21966.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:21,414 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=21988.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:26,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21995.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:31,085 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-22000.pt
+2023-02-28 16:31:31,399 INFO [train.py:968] (0/2) Epoch 1, batch 22000, giga_loss[loss=0.3176, simple_loss=0.3925, pruned_loss=0.1214, over 28932.00 frames. ], tot_loss[loss=0.3807, simple_loss=0.4226, pruned_loss=0.1694, over 5709510.78 frames. ], libri_tot_loss[loss=0.435, simple_loss=0.463, pruned_loss=0.2035, over 5770680.41 frames. ], giga_tot_loss[loss=0.3743, simple_loss=0.4183, pruned_loss=0.1651, over 5700214.25 frames. ], batch size: 174, lr: 2.35e-02, grad_scale: 8.0
+2023-02-28 16:31:38,621 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22006.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:57,886 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22029.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:59,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22032.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:32:15,503 INFO [train.py:968] (0/2) Epoch 1, batch 22050, giga_loss[loss=0.3459, simple_loss=0.4054, pruned_loss=0.1432, over 28869.00 frames. ], tot_loss[loss=0.3796, simple_loss=0.4222, pruned_loss=0.1685, over 5703202.51 frames. ], libri_tot_loss[loss=0.4363, simple_loss=0.4637, pruned_loss=0.2044, over 5761624.29 frames. ], giga_tot_loss[loss=0.3717, simple_loss=0.417, pruned_loss=0.1632, over 5702006.98 frames. ], batch size: 199, lr: 2.35e-02, grad_scale: 4.0
+2023-02-28 16:32:16,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.537e+02 1.320e+03 1.782e+03 2.376e+03 7.911e+03, threshold=3.565e+03, percent-clipped=14.0
+2023-02-28 16:32:27,517 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22061.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:32:45,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22080.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:32:58,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22096.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:33:00,462 INFO [train.py:968] (0/2) Epoch 1, batch 22100, giga_loss[loss=0.3905, simple_loss=0.4344, pruned_loss=0.1733, over 28575.00 frames. ], tot_loss[loss=0.3819, simple_loss=0.4234, pruned_loss=0.1702, over 5685853.26 frames. ], libri_tot_loss[loss=0.437, simple_loss=0.4641, pruned_loss=0.205, over 5753368.55 frames. ], giga_tot_loss[loss=0.3739, simple_loss=0.4182, pruned_loss=0.1648, over 5691399.82 frames. ], batch size: 336, lr: 2.35e-02, grad_scale: 4.0
+2023-02-28 16:33:18,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8228, 1.5850, 1.1833, 1.2407], device='cuda:0'), covar=tensor([0.0734, 0.0957, 0.1160, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0462, 0.0637, 0.0614, 0.0601], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 16:33:42,516 INFO [train.py:968] (0/2) Epoch 1, batch 22150, giga_loss[loss=0.3677, simple_loss=0.4087, pruned_loss=0.1634, over 28743.00 frames. ], tot_loss[loss=0.3852, simple_loss=0.4255, pruned_loss=0.1724, over 5686316.43 frames. ], libri_tot_loss[loss=0.438, simple_loss=0.4645, pruned_loss=0.2057, over 5746420.68 frames. ], giga_tot_loss[loss=0.3764, simple_loss=0.4198, pruned_loss=0.1665, over 5696096.71 frames. ], batch size: 119, lr: 2.35e-02, grad_scale: 4.0
+2023-02-28 16:33:43,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.856e+02 1.347e+03 1.658e+03 2.204e+03 1.103e+04, threshold=3.315e+03, percent-clipped=9.0
+2023-02-28 16:33:47,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22156.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:34:26,006 INFO [train.py:968] (0/2) Epoch 1, batch 22200, giga_loss[loss=0.3807, simple_loss=0.4263, pruned_loss=0.1675, over 28826.00 frames. ], tot_loss[loss=0.3863, simple_loss=0.4264, pruned_loss=0.1731, over 5689274.67 frames. ], libri_tot_loss[loss=0.4386, simple_loss=0.465, pruned_loss=0.2061, over 5747483.33 frames. ], giga_tot_loss[loss=0.3781, simple_loss=0.4209, pruned_loss=0.1676, over 5695380.04 frames. ], batch size: 112, lr: 2.34e-02, grad_scale: 4.0
+2023-02-28 16:34:44,552 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22223.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:34:46,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22226.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:01,154 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22239.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:03,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22242.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:03,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22243.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:08,509 INFO [train.py:968] (0/2) Epoch 1, batch 22250, libri_loss[loss=0.4393, simple_loss=0.4631, pruned_loss=0.2078, over 29514.00 frames. ], tot_loss[loss=0.3903, simple_loss=0.43, pruned_loss=0.1753, over 5701076.21 frames. ], libri_tot_loss[loss=0.4398, simple_loss=0.4656, pruned_loss=0.207, over 5750515.31 frames. ], giga_tot_loss[loss=0.3815, simple_loss=0.4242, pruned_loss=0.1694, over 5701964.42 frames. ], batch size: 80, lr: 2.34e-02, grad_scale: 4.0
+2023-02-28 16:35:09,205 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.947e+02 1.286e+03 1.583e+03 2.206e+03 4.759e+03, threshold=3.166e+03, percent-clipped=7.0
+2023-02-28 16:35:13,101 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22255.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:26,552 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22271.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:47,610 INFO [train.py:968] (0/2) Epoch 1, batch 22300, libri_loss[loss=0.4895, simple_loss=0.4974, pruned_loss=0.2408, over 29078.00 frames. ], tot_loss[loss=0.3939, simple_loss=0.433, pruned_loss=0.1774, over 5704765.37 frames. ], libri_tot_loss[loss=0.4407, simple_loss=0.466, pruned_loss=0.2077, over 5752207.69 frames. ], giga_tot_loss[loss=0.3838, simple_loss=0.4265, pruned_loss=0.1706, over 5701896.81 frames. ], batch size: 101, lr: 2.34e-02, grad_scale: 4.0
+2023-02-28 16:36:25,070 INFO [train.py:968] (0/2) Epoch 1, batch 22350, giga_loss[loss=0.4894, simple_loss=0.4903, pruned_loss=0.2442, over 26658.00 frames. ], tot_loss[loss=0.3986, simple_loss=0.436, pruned_loss=0.1806, over 5701671.14 frames. ], libri_tot_loss[loss=0.4416, simple_loss=0.4661, pruned_loss=0.2086, over 5740529.47 frames. ], giga_tot_loss[loss=0.3866, simple_loss=0.4286, pruned_loss=0.1723, over 5706617.08 frames. ], batch size: 555, lr: 2.34e-02, grad_scale: 4.0
+2023-02-28 16:36:25,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.085e+02 1.643e+03 2.075e+03 2.817e+03 6.280e+03, threshold=4.150e+03, percent-clipped=18.0
+2023-02-28 16:36:34,308 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22363.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:36:34,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22363.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:36:48,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22381.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:36:56,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2868, 1.6313, 1.3849, 1.2579], device='cuda:0'), covar=tensor([0.1451, 0.0700, 0.0730, 0.1966], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0356, 0.0345, 0.0576], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0012, 0.0018], device='cuda:0')
+2023-02-28 16:37:03,620 INFO [train.py:968] (0/2) Epoch 1, batch 22400, giga_loss[loss=0.3717, simple_loss=0.4287, pruned_loss=0.1574, over 28907.00 frames. ], tot_loss[loss=0.3964, simple_loss=0.435, pruned_loss=0.1789, over 5713734.99 frames. ], libri_tot_loss[loss=0.4414, simple_loss=0.4657, pruned_loss=0.2086, over 5743744.05 frames. ], giga_tot_loss[loss=0.3853, simple_loss=0.4282, pruned_loss=0.1712, over 5713818.29 frames. ], batch size: 174, lr: 2.33e-02, grad_scale: 8.0
+2023-02-28 16:37:05,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9863, 2.4502, 2.0260, 1.9055], device='cuda:0'), covar=tensor([0.0827, 0.0809, 0.0658, 0.0415], device='cuda:0'), in_proj_covar=tensor([0.0769, 0.0820, 0.0681, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0006, 0.0004], device='cuda:0')
+2023-02-28 16:37:49,153 INFO [train.py:968] (0/2) Epoch 1, batch 22450, giga_loss[loss=0.3583, simple_loss=0.4167, pruned_loss=0.1499, over 28952.00 frames. ], tot_loss[loss=0.3935, simple_loss=0.4334, pruned_loss=0.1768, over 5708514.43 frames. ], libri_tot_loss[loss=0.4418, simple_loss=0.466, pruned_loss=0.2088, over 5744561.33 frames. ], giga_tot_loss[loss=0.3843, simple_loss=0.4277, pruned_loss=0.1704, over 5707812.40 frames. ], batch size: 145, lr: 2.33e-02, grad_scale: 8.0
+2023-02-28 16:37:51,104 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.350e+02 1.284e+03 1.558e+03 2.022e+03 9.202e+03, threshold=3.116e+03, percent-clipped=4.0
+2023-02-28 16:38:32,581 INFO [train.py:968] (0/2) Epoch 1, batch 22500, giga_loss[loss=0.3577, simple_loss=0.4103, pruned_loss=0.1525, over 29044.00 frames. ], tot_loss[loss=0.3901, simple_loss=0.431, pruned_loss=0.1746, over 5715862.95 frames. ], libri_tot_loss[loss=0.4423, simple_loss=0.4663, pruned_loss=0.2091, over 5745383.52 frames. ], giga_tot_loss[loss=0.3823, simple_loss=0.4261, pruned_loss=0.1692, over 5714366.99 frames. ], batch size: 128, lr: 2.33e-02, grad_scale: 4.0
+2023-02-28 16:38:33,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2524, 1.2273, 1.1630, 0.8377], device='cuda:0'), covar=tensor([0.0394, 0.0299, 0.0243, 0.0350], device='cuda:0'), in_proj_covar=tensor([0.0730, 0.0536, 0.0569, 0.0615], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 16:38:37,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22506.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:38:41,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22509.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:38:51,757 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22523.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:38:52,500 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22524.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:38:54,451 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22527.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:38:56,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22531.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:39:02,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22538.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:39:15,605 INFO [train.py:968] (0/2) Epoch 1, batch 22550, giga_loss[loss=0.322, simple_loss=0.3839, pruned_loss=0.1301, over 29041.00 frames. ], tot_loss[loss=0.3883, simple_loss=0.4292, pruned_loss=0.1737, over 5721858.22 frames. ], libri_tot_loss[loss=0.4444, simple_loss=0.4676, pruned_loss=0.2106, over 5749807.77 frames. ], giga_tot_loss[loss=0.3788, simple_loss=0.4233, pruned_loss=0.1672, over 5716118.43 frames. ], batch size: 164, lr: 2.33e-02, grad_scale: 4.0
+2023-02-28 16:39:17,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.445e+02 1.174e+03 1.431e+03 1.804e+03 8.974e+03, threshold=2.863e+03, percent-clipped=9.0
+2023-02-28 16:39:22,234 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22556.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:39:22,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4560, 1.5112, 1.5410, 1.2596], device='cuda:0'), covar=tensor([0.1039, 0.1010, 0.0952, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0483, 0.0392, 0.0489], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0010, 0.0009, 0.0011], device='cuda:0')
+2023-02-28 16:39:25,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4009, 1.2831, 1.1519, 0.7801], device='cuda:0'), covar=tensor([0.0491, 0.0342, 0.0330, 0.0468], device='cuda:0'), in_proj_covar=tensor([0.0736, 0.0542, 0.0581, 0.0628], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 16:39:32,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22569.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:39:59,531 INFO [train.py:968] (0/2) Epoch 1, batch 22600, giga_loss[loss=0.3114, simple_loss=0.3697, pruned_loss=0.1266, over 28607.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.4248, pruned_loss=0.1708, over 5722420.38 frames. ], libri_tot_loss[loss=0.4448, simple_loss=0.4676, pruned_loss=0.211, over 5752851.11 frames. ], giga_tot_loss[loss=0.3742, simple_loss=0.4192, pruned_loss=0.1646, over 5714606.92 frames. ], batch size: 85, lr: 2.32e-02, grad_scale: 4.0
+2023-02-28 16:40:04,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3324, 1.6047, 1.4791, 0.2624], device='cuda:0'), covar=tensor([0.0822, 0.0664, 0.0765, 0.1564], device='cuda:0'), in_proj_covar=tensor([0.0870, 0.0894, 0.0916, 0.0844], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 16:40:05,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22608.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:40:14,770 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22618.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:40:24,537 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22630.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:40:30,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4788, 3.5416, 5.2329, 2.7863], device='cuda:0'), covar=tensor([0.0372, 0.0546, 0.0756, 0.1417], device='cuda:0'), in_proj_covar=tensor([0.0694, 0.0442, 0.0766, 0.0519], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 16:40:39,252 INFO [train.py:968] (0/2) Epoch 1, batch 22650, libri_loss[loss=0.5636, simple_loss=0.541, pruned_loss=0.2931, over 29550.00 frames. ], tot_loss[loss=0.3828, simple_loss=0.4242, pruned_loss=0.1707, over 5726180.87 frames. ], libri_tot_loss[loss=0.4458, simple_loss=0.468, pruned_loss=0.2117, over 5754876.55 frames. ], giga_tot_loss[loss=0.3726, simple_loss=0.4179, pruned_loss=0.1636, over 5717003.20 frames. ], batch size: 83, lr: 2.32e-02, grad_scale: 4.0
+2023-02-28 16:40:40,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.315e+02 1.300e+03 1.618e+03 2.246e+03 8.899e+03, threshold=3.237e+03, percent-clipped=16.0
+2023-02-28 16:40:59,074 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22674.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:41:00,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22677.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:41:05,825 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-02-28 16:41:22,410 INFO [train.py:968] (0/2) Epoch 1, batch 22700, libri_loss[loss=0.4711, simple_loss=0.4867, pruned_loss=0.2277, over 29777.00 frames. ], tot_loss[loss=0.3878, simple_loss=0.4285, pruned_loss=0.1736, over 5711217.16 frames. ], libri_tot_loss[loss=0.4473, simple_loss=0.4688, pruned_loss=0.2129, over 5749045.66 frames. ], giga_tot_loss[loss=0.3751, simple_loss=0.4206, pruned_loss=0.1648, over 5707068.20 frames. ], batch size: 87, lr: 2.32e-02, grad_scale: 4.0
+2023-02-28 16:41:26,163 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22706.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:41:54,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22738.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:42:04,664 INFO [train.py:968] (0/2) Epoch 1, batch 22750, giga_loss[loss=0.3608, simple_loss=0.4149, pruned_loss=0.1534, over 28874.00 frames. ], tot_loss[loss=0.3869, simple_loss=0.429, pruned_loss=0.1723, over 5716185.12 frames. ], libri_tot_loss[loss=0.447, simple_loss=0.4684, pruned_loss=0.2128, over 5751191.50 frames. ], giga_tot_loss[loss=0.3759, simple_loss=0.4224, pruned_loss=0.1647, over 5710268.66 frames. ], batch size: 186, lr: 2.32e-02, grad_scale: 4.0
+2023-02-28 16:42:05,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.968e+02 1.272e+03 1.758e+03 2.203e+03 7.319e+03, threshold=3.515e+03, percent-clipped=12.0
+2023-02-28 16:42:14,759 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22761.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:42:16,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22764.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:42:26,728 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22778.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:42:28,200 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-02-28 16:42:37,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4837, 1.8017, 1.5802, 0.5295], device='cuda:0'), covar=tensor([0.0974, 0.0750, 0.0976, 0.1589], device='cuda:0'), in_proj_covar=tensor([0.0898, 0.0891, 0.0936, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 16:42:38,476 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22793.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:42:43,166 INFO [train.py:968] (0/2) Epoch 1, batch 22800, giga_loss[loss=0.3164, simple_loss=0.3633, pruned_loss=0.1348, over 28327.00 frames. ], tot_loss[loss=0.3871, simple_loss=0.4286, pruned_loss=0.1728, over 5728431.08 frames. ], libri_tot_loss[loss=0.4474, simple_loss=0.4686, pruned_loss=0.2131, over 5754683.91 frames. ], giga_tot_loss[loss=0.3767, simple_loss=0.4222, pruned_loss=0.1656, over 5719877.45 frames. ], batch size: 77, lr: 2.31e-02, grad_scale: 8.0
+2023-02-28 16:42:47,238 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-02-28 16:43:26,382 INFO [train.py:968] (0/2) Epoch 1, batch 22850, giga_loss[loss=0.3636, simple_loss=0.4063, pruned_loss=0.1605, over 28946.00 frames. ], tot_loss[loss=0.3863, simple_loss=0.4268, pruned_loss=0.1729, over 5726916.98 frames. ], libri_tot_loss[loss=0.4475, simple_loss=0.4686, pruned_loss=0.2132, over 5755259.97 frames. ], giga_tot_loss[loss=0.3768, simple_loss=0.4211, pruned_loss=0.1663, over 5719091.00 frames. ], batch size: 136, lr: 2.31e-02, grad_scale: 8.0
+2023-02-28 16:43:29,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.590e+02 1.195e+03 1.512e+03 2.025e+03 5.618e+03, threshold=3.024e+03, percent-clipped=5.0
+2023-02-28 16:43:51,616 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22881.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:43:53,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22884.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:43:54,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4993, 1.3903, 1.3254, 1.7996], device='cuda:0'), covar=tensor([0.1383, 0.1454, 0.1093, 0.1676], device='cuda:0'), in_proj_covar=tensor([0.0835, 0.0739, 0.0792, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 16:44:02,462 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.42 vs. limit=5.0
+2023-02-28 16:44:05,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22898.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:44:06,950 INFO [train.py:968] (0/2) Epoch 1, batch 22900, giga_loss[loss=0.3511, simple_loss=0.4038, pruned_loss=0.1492, over 28966.00 frames. ], tot_loss[loss=0.3854, simple_loss=0.4247, pruned_loss=0.1731, over 5725321.87 frames. ], libri_tot_loss[loss=0.4473, simple_loss=0.4684, pruned_loss=0.2131, over 5759020.68 frames. ], giga_tot_loss[loss=0.3763, simple_loss=0.419, pruned_loss=0.1668, over 5714946.96 frames. ], batch size: 164, lr: 2.31e-02, grad_scale: 8.0
+2023-02-28 16:44:17,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22913.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:44:42,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22944.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:44:46,233 INFO [train.py:968] (0/2) Epoch 1, batch 22950, giga_loss[loss=0.3079, simple_loss=0.3629, pruned_loss=0.1264, over 28432.00 frames. ], tot_loss[loss=0.3867, simple_loss=0.424, pruned_loss=0.1747, over 5729796.39 frames. ], libri_tot_loss[loss=0.448, simple_loss=0.4685, pruned_loss=0.2138, over 5764428.31 frames. ], giga_tot_loss[loss=0.3763, simple_loss=0.4176, pruned_loss=0.1675, over 5715194.81 frames. ], batch size: 65, lr: 2.31e-02, grad_scale: 4.0
+2023-02-28 16:44:48,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.239e+02 1.364e+03 1.696e+03 2.233e+03 4.832e+03, threshold=3.392e+03, percent-clipped=8.0
+2023-02-28 16:44:52,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9149, 1.7048, 4.1678, 2.9189], device='cuda:0'), covar=tensor([0.1225, 0.0977, 0.0239, 0.0379], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0432, 0.0562, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 16:45:12,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22983.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:45:26,970 INFO [train.py:968] (0/2) Epoch 1, batch 23000, giga_loss[loss=0.3663, simple_loss=0.3976, pruned_loss=0.1675, over 28752.00 frames. ], tot_loss[loss=0.3867, simple_loss=0.4236, pruned_loss=0.1749, over 5729190.39 frames. ], libri_tot_loss[loss=0.4489, simple_loss=0.469, pruned_loss=0.2143, over 5764537.59 frames. ], giga_tot_loss[loss=0.3763, simple_loss=0.417, pruned_loss=0.1678, over 5716698.25 frames. ], batch size: 92, lr: 2.30e-02, grad_scale: 4.0
+2023-02-28 16:45:31,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23005.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:45:37,296 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.85 vs. limit=5.0
+2023-02-28 16:45:50,042 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-02-28 16:45:52,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23032.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:45:59,969 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23041.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:46:01,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23044.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:46:06,366 INFO [train.py:968] (0/2) Epoch 1, batch 23050, giga_loss[loss=0.3411, simple_loss=0.3906, pruned_loss=0.1458, over 28620.00 frames. ], tot_loss[loss=0.3839, simple_loss=0.4208, pruned_loss=0.1735, over 5725920.51 frames. ], libri_tot_loss[loss=0.4491, simple_loss=0.469, pruned_loss=0.2146, over 5765940.99 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.414, pruned_loss=0.1662, over 5713137.63 frames. ], batch size: 242, lr: 2.30e-02, grad_scale: 4.0
+2023-02-28 16:46:08,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.032e+02 1.447e+03 1.712e+03 2.292e+03 4.197e+03, threshold=3.423e+03, percent-clipped=6.0
+2023-02-28 16:46:25,089 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23073.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:46:35,295 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23087.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:46:38,673 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23090.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:46:47,659 INFO [train.py:968] (0/2) Epoch 1, batch 23100, giga_loss[loss=0.3318, simple_loss=0.3651, pruned_loss=0.1492, over 28648.00 frames. ], tot_loss[loss=0.3748, simple_loss=0.4134, pruned_loss=0.1681, over 5731249.33 frames. ], libri_tot_loss[loss=0.4491, simple_loss=0.4689, pruned_loss=0.2147, over 5767890.31 frames. ], giga_tot_loss[loss=0.3655, simple_loss=0.4075, pruned_loss=0.1618, over 5719042.10 frames. ], batch size: 78, lr: 2.30e-02, grad_scale: 4.0
+2023-02-28 16:47:01,809 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23119.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:07,374 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23126.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:09,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23129.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:23,927 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23148.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:24,827 INFO [train.py:968] (0/2) Epoch 1, batch 23150, giga_loss[loss=0.3585, simple_loss=0.4049, pruned_loss=0.1561, over 28919.00 frames. ], tot_loss[loss=0.3719, simple_loss=0.4106, pruned_loss=0.1666, over 5721174.98 frames. ], libri_tot_loss[loss=0.45, simple_loss=0.4694, pruned_loss=0.2153, over 5758782.58 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4028, pruned_loss=0.1588, over 5717667.25 frames. ], batch size: 186, lr: 2.30e-02, grad_scale: 4.0
+2023-02-28 16:47:25,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23151.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:26,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.370e+02 1.257e+03 1.811e+03 2.634e+03 1.055e+04, threshold=3.622e+03, percent-clipped=10.0
+2023-02-28 16:47:27,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23153.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:29,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4307, 1.2831, 1.2518, 1.2645], device='cuda:0'), covar=tensor([0.0648, 0.0971, 0.1062, 0.0829], device='cuda:0'), in_proj_covar=tensor([0.0543, 0.0897, 0.0673, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 16:47:30,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23158.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:47,327 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-02-28 16:47:50,365 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23180.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:55,871 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23187.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:48:07,777 INFO [train.py:968] (0/2) Epoch 1, batch 23200, giga_loss[loss=0.3844, simple_loss=0.4239, pruned_loss=0.1724, over 29069.00 frames. ], tot_loss[loss=0.373, simple_loss=0.4117, pruned_loss=0.1671, over 5716347.98 frames. ], libri_tot_loss[loss=0.45, simple_loss=0.4694, pruned_loss=0.2153, over 5760283.96 frames. ], giga_tot_loss[loss=0.3629, simple_loss=0.405, pruned_loss=0.1604, over 5711915.62 frames. ], batch size: 136, lr: 2.29e-02, grad_scale: 8.0
+2023-02-28 16:48:24,983 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23218.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:48:50,537 INFO [train.py:968] (0/2) Epoch 1, batch 23250, giga_loss[loss=0.3928, simple_loss=0.4295, pruned_loss=0.178, over 28963.00 frames. ], tot_loss[loss=0.3776, simple_loss=0.4161, pruned_loss=0.1696, over 5700139.47 frames. ], libri_tot_loss[loss=0.4508, simple_loss=0.4698, pruned_loss=0.2159, over 5743394.60 frames. ], giga_tot_loss[loss=0.3675, simple_loss=0.4093, pruned_loss=0.1628, over 5711780.94 frames. ], batch size: 106, lr: 2.29e-02, grad_scale: 8.0
+2023-02-28 16:48:53,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.676e+02 1.255e+03 1.498e+03 1.969e+03 6.451e+03, threshold=2.995e+03, percent-clipped=2.0
+2023-02-28 16:49:11,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8219, 1.6873, 1.3113, 1.5155], device='cuda:0'), covar=tensor([0.0697, 0.0762, 0.1079, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0636, 0.0640, 0.0596], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 16:49:30,252 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23296.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:49:32,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23299.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:49:32,488 INFO [train.py:968] (0/2) Epoch 1, batch 23300, giga_loss[loss=0.4015, simple_loss=0.4422, pruned_loss=0.1804, over 28977.00 frames. ], tot_loss[loss=0.3841, simple_loss=0.4218, pruned_loss=0.1732, over 5707692.67 frames. ], libri_tot_loss[loss=0.4516, simple_loss=0.4703, pruned_loss=0.2164, over 5748028.48 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.4144, pruned_loss=0.1659, over 5711464.84 frames. ], batch size: 164, lr: 2.29e-02, grad_scale: 4.0
+2023-02-28 16:49:32,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6954, 1.3813, 1.4489, 1.1989], device='cuda:0'), covar=tensor([0.0392, 0.0318, 0.0241, 0.0323], device='cuda:0'), in_proj_covar=tensor([0.0746, 0.0552, 0.0583, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 16:49:56,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23328.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:50:13,224 INFO [train.py:968] (0/2) Epoch 1, batch 23350, giga_loss[loss=0.3772, simple_loss=0.4249, pruned_loss=0.1648, over 28702.00 frames. ], tot_loss[loss=0.3879, simple_loss=0.4253, pruned_loss=0.1752, over 5709672.37 frames. ], libri_tot_loss[loss=0.4513, simple_loss=0.4698, pruned_loss=0.2164, over 5751991.25 frames. ], giga_tot_loss[loss=0.3778, simple_loss=0.4187, pruned_loss=0.1684, over 5707889.30 frames. ], batch size: 242, lr: 2.29e-02, grad_scale: 4.0
+2023-02-28 16:50:16,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.358e+02 1.496e+03 1.937e+03 2.743e+03 5.008e+03, threshold=3.874e+03, percent-clipped=18.0
+2023-02-28 16:50:27,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7700, 2.4775, 1.7089, 1.8261], device='cuda:0'), covar=tensor([0.0570, 0.0797, 0.1228, 0.1153], device='cuda:0'), in_proj_covar=tensor([0.0451, 0.0629, 0.0619, 0.0591], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 16:50:39,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 16:50:46,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4222, 1.7374, 1.5271, 1.3934], device='cuda:0'), covar=tensor([0.0796, 0.0974, 0.0725, 0.0487], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0805, 0.0687, 0.0605], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0006, 0.0005], device='cuda:0')
+2023-02-28 16:50:54,344 INFO [train.py:968] (0/2) Epoch 1, batch 23400, libri_loss[loss=0.5204, simple_loss=0.5202, pruned_loss=0.2603, over 29625.00 frames. ], tot_loss[loss=0.3916, simple_loss=0.429, pruned_loss=0.1771, over 5723024.79 frames. ], libri_tot_loss[loss=0.452, simple_loss=0.47, pruned_loss=0.217, over 5757716.85 frames. ], giga_tot_loss[loss=0.3805, simple_loss=0.4219, pruned_loss=0.1695, over 5714972.97 frames. ], batch size: 88, lr: 2.29e-02, grad_scale: 4.0
+2023-02-28 16:51:00,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23407.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:51:17,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4738, 1.3923, 1.3415, 1.5628], device='cuda:0'), covar=tensor([0.1465, 0.1545, 0.1121, 0.1765], device='cuda:0'), in_proj_covar=tensor([0.0867, 0.0785, 0.0834, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 16:51:40,306 INFO [train.py:968] (0/2) Epoch 1, batch 23450, libri_loss[loss=0.4164, simple_loss=0.4331, pruned_loss=0.1999, over 29673.00 frames. ], tot_loss[loss=0.3933, simple_loss=0.4307, pruned_loss=0.178, over 5726072.14 frames. ], libri_tot_loss[loss=0.4518, simple_loss=0.4699, pruned_loss=0.2169, over 5759623.91 frames. ], giga_tot_loss[loss=0.3839, simple_loss=0.4247, pruned_loss=0.1715, over 5717470.71 frames. ], batch size: 73, lr: 2.28e-02, grad_scale: 4.0
+2023-02-28 16:51:45,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.910e+02 1.273e+03 1.652e+03 2.367e+03 7.325e+03, threshold=3.304e+03, percent-clipped=4.0
+2023-02-28 16:51:57,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8266, 1.5526, 1.5221, 1.4621], device='cuda:0'), covar=tensor([0.0487, 0.0956, 0.0692, 0.0815], device='cuda:0'), in_proj_covar=tensor([0.0536, 0.0898, 0.0673, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 16:52:21,436 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-02-28 16:52:32,649 INFO [train.py:968] (0/2) Epoch 1, batch 23500, giga_loss[loss=0.4714, simple_loss=0.4885, pruned_loss=0.2272, over 28686.00 frames. ], tot_loss[loss=0.4048, simple_loss=0.4381, pruned_loss=0.1858, over 5711493.34 frames. ], libri_tot_loss[loss=0.4512, simple_loss=0.4693, pruned_loss=0.2165, over 5762560.74 frames. ], giga_tot_loss[loss=0.3969, simple_loss=0.4332, pruned_loss=0.1803, over 5701416.49 frames. ], batch size: 242, lr: 2.28e-02, grad_scale: 4.0
+2023-02-28 16:52:33,573 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23501.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:52:47,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2764, 2.6481, 5.5395, 3.4989], device='cuda:0'), covar=tensor([0.1212, 0.0831, 0.0187, 0.0318], device='cuda:0'), in_proj_covar=tensor([0.0490, 0.0449, 0.0593, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0005], device='cuda:0')
+2023-02-28 16:53:25,764 INFO [train.py:968] (0/2) Epoch 1, batch 23550, giga_loss[loss=0.4314, simple_loss=0.4683, pruned_loss=0.1973, over 28929.00 frames. ], tot_loss[loss=0.4161, simple_loss=0.4461, pruned_loss=0.1931, over 5698698.60 frames. ], libri_tot_loss[loss=0.4513, simple_loss=0.4693, pruned_loss=0.2166, over 5763761.88 frames. ], giga_tot_loss[loss=0.4092, simple_loss=0.4418, pruned_loss=0.1883, over 5689071.29 frames. ], batch size: 136, lr: 2.28e-02, grad_scale: 4.0
+2023-02-28 16:53:26,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23550.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:53:28,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23553.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:53:28,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.383e+02 1.629e+03 2.091e+03 2.982e+03 6.189e+03, threshold=4.181e+03, percent-clipped=19.0
+2023-02-28 16:53:36,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23562.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:53:56,442 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23582.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:54:06,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23593.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:54:13,339 INFO [train.py:968] (0/2) Epoch 1, batch 23600, giga_loss[loss=0.4846, simple_loss=0.495, pruned_loss=0.2371, over 28588.00 frames. ], tot_loss[loss=0.4274, simple_loss=0.4539, pruned_loss=0.2005, over 5694042.07 frames. ], libri_tot_loss[loss=0.4519, simple_loss=0.4696, pruned_loss=0.2171, over 5766611.84 frames. ], giga_tot_loss[loss=0.4202, simple_loss=0.4495, pruned_loss=0.1955, over 5681123.37 frames. ], batch size: 336, lr: 2.28e-02, grad_scale: 8.0
+2023-02-28 16:54:23,884 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23610.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:54:26,020 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-02-28 16:54:35,819 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23622.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:54:52,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7600, 1.3190, 1.5500, 1.1332], device='cuda:0'), covar=tensor([0.0281, 0.0259, 0.0173, 0.0261], device='cuda:0'), in_proj_covar=tensor([0.0762, 0.0557, 0.0604, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 16:54:53,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2976, 1.8089, 1.7328, 1.7815], device='cuda:0'), covar=tensor([0.0540, 0.1256, 0.0901, 0.0992], device='cuda:0'), in_proj_covar=tensor([0.0557, 0.0911, 0.0685, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 16:55:04,721 INFO [train.py:968] (0/2) Epoch 1, batch 23650, giga_loss[loss=0.5811, simple_loss=0.5288, pruned_loss=0.3167, over 23336.00 frames. ], tot_loss[loss=0.4367, simple_loss=0.4601, pruned_loss=0.2066, over 5688757.85 frames. ], libri_tot_loss[loss=0.4525, simple_loss=0.4702, pruned_loss=0.2174, over 5766818.75 frames. ], giga_tot_loss[loss=0.4299, simple_loss=0.4558, pruned_loss=0.2021, over 5676691.41 frames. ], batch size: 705, lr: 2.27e-02, grad_scale: 8.0
+2023-02-28 16:55:07,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.615e+02 1.820e+03 2.174e+03 3.309e+03 9.253e+03, threshold=4.348e+03, percent-clipped=7.0
+2023-02-28 16:55:37,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8896, 0.9528, 1.0203, 0.4077], device='cuda:0'), covar=tensor([0.0271, 0.0218, 0.0188, 0.0307], device='cuda:0'), in_proj_covar=tensor([0.0762, 0.0544, 0.0596, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 16:55:55,643 INFO [train.py:968] (0/2) Epoch 1, batch 23700, giga_loss[loss=0.4661, simple_loss=0.4872, pruned_loss=0.2225, over 28844.00 frames. ], tot_loss[loss=0.4487, simple_loss=0.4682, pruned_loss=0.2146, over 5686603.22 frames. ], libri_tot_loss[loss=0.4533, simple_loss=0.4707, pruned_loss=0.218, over 5765902.99 frames. ], giga_tot_loss[loss=0.4423, simple_loss=0.4641, pruned_loss=0.2102, over 5675309.31 frames. ], batch size: 186, lr: 2.27e-02, grad_scale: 4.0
+2023-02-28 16:55:59,235 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23705.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:56:03,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23708.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:56:32,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23736.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:56:32,746 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23737.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:56:34,378 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23739.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:56:42,831 INFO [train.py:968] (0/2) Epoch 1, batch 23750, giga_loss[loss=0.4758, simple_loss=0.4821, pruned_loss=0.2348, over 28602.00 frames. ], tot_loss[loss=0.4531, simple_loss=0.4709, pruned_loss=0.2177, over 5687808.81 frames. ], libri_tot_loss[loss=0.4537, simple_loss=0.4707, pruned_loss=0.2183, over 5769154.54 frames. ], giga_tot_loss[loss=0.4475, simple_loss=0.4675, pruned_loss=0.2138, over 5674169.88 frames. ], batch size: 307, lr: 2.27e-02, grad_scale: 4.0
+2023-02-28 16:56:49,483 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.980e+03 2.421e+03 3.463e+03 9.409e+03, threshold=4.841e+03, percent-clipped=12.0
+2023-02-28 16:57:02,724 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23768.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:57:34,527 INFO [train.py:968] (0/2) Epoch 1, batch 23800, giga_loss[loss=0.449, simple_loss=0.4707, pruned_loss=0.2136, over 28692.00 frames. ], tot_loss[loss=0.4582, simple_loss=0.4734, pruned_loss=0.2215, over 5672427.01 frames. ], libri_tot_loss[loss=0.454, simple_loss=0.4709, pruned_loss=0.2186, over 5770135.82 frames. ], giga_tot_loss[loss=0.4535, simple_loss=0.4706, pruned_loss=0.2182, over 5658922.70 frames. ], batch size: 242, lr: 2.27e-02, grad_scale: 4.0
+2023-02-28 16:58:29,182 INFO [train.py:968] (0/2) Epoch 1, batch 23850, giga_loss[loss=0.5372, simple_loss=0.503, pruned_loss=0.2857, over 23525.00 frames. ], tot_loss[loss=0.4672, simple_loss=0.4787, pruned_loss=0.2279, over 5661370.08 frames. ], libri_tot_loss[loss=0.4546, simple_loss=0.4713, pruned_loss=0.2189, over 5771754.08 frames. ], giga_tot_loss[loss=0.4632, simple_loss=0.4762, pruned_loss=0.2251, over 5646521.82 frames. ], batch size: 705, lr: 2.26e-02, grad_scale: 4.0
+2023-02-28 16:58:34,452 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.262e+02 1.821e+03 2.407e+03 3.267e+03 7.528e+03, threshold=4.814e+03, percent-clipped=6.0
+2023-02-28 16:58:54,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23876.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:59:01,262 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23881.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:59:22,371 INFO [train.py:968] (0/2) Epoch 1, batch 23900, giga_loss[loss=0.5163, simple_loss=0.5149, pruned_loss=0.2589, over 28620.00 frames. ], tot_loss[loss=0.4746, simple_loss=0.4833, pruned_loss=0.2329, over 5646395.78 frames. ], libri_tot_loss[loss=0.4549, simple_loss=0.4715, pruned_loss=0.2192, over 5761875.77 frames. ], giga_tot_loss[loss=0.4714, simple_loss=0.4813, pruned_loss=0.2307, over 5640956.49 frames. ], batch size: 336, lr: 2.26e-02, grad_scale: 4.0
+2023-02-28 16:59:42,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2519, 1.2631, 1.1976, 1.4248], device='cuda:0'), covar=tensor([0.1415, 0.1445, 0.1127, 0.1239], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0801, 0.0859, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 17:00:26,473 INFO [train.py:968] (0/2) Epoch 1, batch 23950, giga_loss[loss=0.4711, simple_loss=0.4707, pruned_loss=0.2358, over 28940.00 frames. ], tot_loss[loss=0.4767, simple_loss=0.4851, pruned_loss=0.2342, over 5652177.82 frames. ], libri_tot_loss[loss=0.4556, simple_loss=0.4718, pruned_loss=0.2197, over 5764091.06 frames. ], giga_tot_loss[loss=0.4738, simple_loss=0.4835, pruned_loss=0.2321, over 5644386.49 frames. ], batch size: 106, lr: 2.26e-02, grad_scale: 4.0
+2023-02-28 17:00:32,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.884e+02 2.036e+03 2.576e+03 3.614e+03 8.022e+03, threshold=5.151e+03, percent-clipped=13.0
+2023-02-28 17:01:04,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23985.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:01:16,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23997.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:01:18,535 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-24000.pt
+2023-02-28 17:01:18,854 INFO [train.py:968] (0/2) Epoch 1, batch 24000, libri_loss[loss=0.4666, simple_loss=0.4793, pruned_loss=0.227, over 29526.00 frames. ], tot_loss[loss=0.4767, simple_loss=0.4843, pruned_loss=0.2345, over 5643702.32 frames. ], libri_tot_loss[loss=0.4558, simple_loss=0.4719, pruned_loss=0.2199, over 5768131.72 frames. ], giga_tot_loss[loss=0.4748, simple_loss=0.4833, pruned_loss=0.2331, over 5630305.11 frames. ], batch size: 82, lr: 2.26e-02, grad_scale: 8.0
+2023-02-28 17:01:18,859 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 17:01:25,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6921, 1.5351, 1.4430, 1.3821], device='cuda:0'), covar=tensor([0.0702, 0.1217, 0.1119, 0.0990], device='cuda:0'), in_proj_covar=tensor([0.0556, 0.0922, 0.0685, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 17:01:26,963 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.3173, simple_loss=0.4031, pruned_loss=0.1157, over 944034.00 frames. 
+2023-02-28 17:01:26,964 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 17:01:43,444 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24019.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:01:47,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24022.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:02:02,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4290, 1.8116, 1.6337, 0.8192], device='cuda:0'), covar=tensor([0.0954, 0.0892, 0.0589, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0944, 0.0955, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 17:02:14,853 INFO [train.py:968] (0/2) Epoch 1, batch 24050, giga_loss[loss=0.4072, simple_loss=0.4466, pruned_loss=0.1839, over 28990.00 frames. ], tot_loss[loss=0.4732, simple_loss=0.4819, pruned_loss=0.2322, over 5647668.22 frames. ], libri_tot_loss[loss=0.4559, simple_loss=0.4719, pruned_loss=0.22, over 5760227.49 frames. ], giga_tot_loss[loss=0.4718, simple_loss=0.4812, pruned_loss=0.2312, over 5641729.28 frames. ], batch size: 213, lr: 2.26e-02, grad_scale: 4.0
+2023-02-28 17:02:18,058 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24051.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:02:21,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.768e+02 2.077e+03 2.683e+03 3.436e+03 5.095e+03, threshold=5.365e+03, percent-clipped=0.0
+2023-02-28 17:02:22,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3432, 1.1817, 1.1649, 1.2870], device='cuda:0'), covar=tensor([0.1489, 0.1789, 0.1288, 0.1875], device='cuda:0'), in_proj_covar=tensor([0.0862, 0.0803, 0.0856, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 17:02:24,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4609, 3.1400, 4.2096, 2.1564], device='cuda:0'), covar=tensor([0.0488, 0.0583, 0.0815, 0.1587], device='cuda:0'), in_proj_covar=tensor([0.0743, 0.0483, 0.0826, 0.0549], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 17:02:38,382 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.21 vs. limit=2.0
+2023-02-28 17:03:01,126 INFO [train.py:968] (0/2) Epoch 1, batch 24100, giga_loss[loss=0.4298, simple_loss=0.4652, pruned_loss=0.1972, over 28824.00 frames. ], tot_loss[loss=0.4712, simple_loss=0.4817, pruned_loss=0.2304, over 5653520.57 frames. ], libri_tot_loss[loss=0.4564, simple_loss=0.4722, pruned_loss=0.2203, over 5760857.82 frames. ], giga_tot_loss[loss=0.4702, simple_loss=0.4812, pruned_loss=0.2296, over 5644586.27 frames. ], batch size: 145, lr: 2.25e-02, grad_scale: 4.0
+2023-02-28 17:03:04,455 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7562, 1.9562, 4.1434, 2.7215], device='cuda:0'), covar=tensor([0.1408, 0.1037, 0.0247, 0.0473], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0454, 0.0590, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0005], device='cuda:0')
+2023-02-28 17:03:32,807 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24128.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:03:36,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24131.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:03:37,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0588, 1.7076, 1.4932, 1.5759], device='cuda:0'), covar=tensor([0.0777, 0.1556, 0.1101, 0.1124], device='cuda:0'), in_proj_covar=tensor([0.0560, 0.0905, 0.0674, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 17:03:46,146 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24140.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:03:48,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24143.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:03:54,917 INFO [train.py:968] (0/2) Epoch 1, batch 24150, libri_loss[loss=0.5203, simple_loss=0.5144, pruned_loss=0.2631, over 29539.00 frames. ], tot_loss[loss=0.473, simple_loss=0.4829, pruned_loss=0.2316, over 5624849.71 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4731, pruned_loss=0.2213, over 5744784.86 frames. ], giga_tot_loss[loss=0.4712, simple_loss=0.4818, pruned_loss=0.2303, over 5627882.97 frames. ], batch size: 84, lr: 2.25e-02, grad_scale: 4.0
+2023-02-28 17:03:59,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.989e+03 2.608e+03 3.737e+03 7.721e+03, threshold=5.216e+03, percent-clipped=7.0
+2023-02-28 17:04:04,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24160.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:04:18,703 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24172.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:04:46,979 INFO [train.py:968] (0/2) Epoch 1, batch 24200, libri_loss[loss=0.4539, simple_loss=0.4799, pruned_loss=0.2139, over 29246.00 frames. ], tot_loss[loss=0.4741, simple_loss=0.484, pruned_loss=0.2321, over 5611088.95 frames. ], libri_tot_loss[loss=0.4583, simple_loss=0.4734, pruned_loss=0.2216, over 5727935.42 frames. ], giga_tot_loss[loss=0.4726, simple_loss=0.4832, pruned_loss=0.231, over 5625538.79 frames. ], batch size: 94, lr: 2.25e-02, grad_scale: 4.0
+2023-02-28 17:05:14,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2866, 1.5560, 1.3484, 1.2871], device='cuda:0'), covar=tensor([0.1578, 0.0687, 0.0749, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0356, 0.0342, 0.0569], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0020], device='cuda:0')
+2023-02-28 17:05:39,494 INFO [train.py:968] (0/2) Epoch 1, batch 24250, giga_loss[loss=0.4728, simple_loss=0.4755, pruned_loss=0.235, over 27583.00 frames. ], tot_loss[loss=0.4643, simple_loss=0.478, pruned_loss=0.2253, over 5613207.43 frames. ], libri_tot_loss[loss=0.4584, simple_loss=0.4732, pruned_loss=0.2218, over 5728987.08 frames. ], giga_tot_loss[loss=0.4633, simple_loss=0.4777, pruned_loss=0.2245, over 5619923.94 frames. ], batch size: 472, lr: 2.25e-02, grad_scale: 4.0
+2023-02-28 17:05:40,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2990, 1.4661, 1.4429, 0.2588], device='cuda:0'), covar=tensor([0.0781, 0.0701, 0.0867, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0912, 0.0923, 0.0920, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 17:05:44,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.598e+02 1.792e+03 2.350e+03 2.947e+03 7.313e+03, threshold=4.701e+03, percent-clipped=3.0
+2023-02-28 17:05:45,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24256.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:06:25,205 INFO [train.py:968] (0/2) Epoch 1, batch 24300, giga_loss[loss=0.4378, simple_loss=0.4682, pruned_loss=0.2037, over 28839.00 frames. ], tot_loss[loss=0.456, simple_loss=0.4734, pruned_loss=0.2194, over 5621229.96 frames. ], libri_tot_loss[loss=0.4585, simple_loss=0.473, pruned_loss=0.222, over 5724017.92 frames. ], giga_tot_loss[loss=0.4552, simple_loss=0.4733, pruned_loss=0.2186, over 5628180.43 frames. ], batch size: 284, lr: 2.24e-02, grad_scale: 4.0
+2023-02-28 17:07:16,094 INFO [train.py:968] (0/2) Epoch 1, batch 24350, giga_loss[loss=0.4422, simple_loss=0.4671, pruned_loss=0.2086, over 28784.00 frames. ], tot_loss[loss=0.4507, simple_loss=0.47, pruned_loss=0.2156, over 5643933.78 frames. ], libri_tot_loss[loss=0.4581, simple_loss=0.4728, pruned_loss=0.2217, over 5726358.97 frames. ], giga_tot_loss[loss=0.4503, simple_loss=0.4703, pruned_loss=0.2152, over 5646019.23 frames. ], batch size: 284, lr: 2.24e-02, grad_scale: 2.0
+2023-02-28 17:07:23,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.676e+03 2.392e+03 3.438e+03 1.440e+04, threshold=4.783e+03, percent-clipped=13.0
+2023-02-28 17:08:06,035 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24399.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:08:06,439 INFO [train.py:968] (0/2) Epoch 1, batch 24400, giga_loss[loss=0.3732, simple_loss=0.4217, pruned_loss=0.1624, over 28891.00 frames. ], tot_loss[loss=0.4453, simple_loss=0.4668, pruned_loss=0.212, over 5651038.51 frames. ], libri_tot_loss[loss=0.4575, simple_loss=0.4722, pruned_loss=0.2215, over 5720437.35 frames. ], giga_tot_loss[loss=0.4454, simple_loss=0.4674, pruned_loss=0.2117, over 5656052.88 frames. ], batch size: 174, lr: 2.24e-02, grad_scale: 4.0
+2023-02-28 17:08:09,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24402.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:08:36,485 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=24430.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:08:37,061 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24431.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:08:55,665 INFO [train.py:968] (0/2) Epoch 1, batch 24450, giga_loss[loss=0.3757, simple_loss=0.4225, pruned_loss=0.1645, over 28526.00 frames. ], tot_loss[loss=0.4435, simple_loss=0.465, pruned_loss=0.2111, over 5644843.64 frames. ], libri_tot_loss[loss=0.4581, simple_loss=0.4725, pruned_loss=0.2219, over 5715381.83 frames. ], giga_tot_loss[loss=0.4426, simple_loss=0.465, pruned_loss=0.2101, over 5650832.57 frames. ], batch size: 65, lr: 2.24e-02, grad_scale: 4.0
+2023-02-28 17:08:55,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3101, 1.4366, 1.2702, 1.0749], device='cuda:0'), covar=tensor([0.1126, 0.1038, 0.1033, 0.1771], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0499, 0.0402, 0.0512], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0011, 0.0009, 0.0012], device='cuda:0')
+2023-02-28 17:09:02,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.065e+02 1.769e+03 2.080e+03 2.978e+03 6.103e+03, threshold=4.161e+03, percent-clipped=6.0
+2023-02-28 17:09:51,040 INFO [train.py:968] (0/2) Epoch 1, batch 24500, giga_loss[loss=0.4468, simple_loss=0.4675, pruned_loss=0.2131, over 28221.00 frames. ], tot_loss[loss=0.442, simple_loss=0.4641, pruned_loss=0.21, over 5650185.50 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4722, pruned_loss=0.2217, over 5718354.48 frames. ], giga_tot_loss[loss=0.4413, simple_loss=0.4643, pruned_loss=0.2092, over 5651279.43 frames. ], batch size: 368, lr: 2.24e-02, grad_scale: 4.0
+2023-02-28 17:10:21,002 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-02-28 17:10:43,650 INFO [train.py:968] (0/2) Epoch 1, batch 24550, giga_loss[loss=0.4099, simple_loss=0.447, pruned_loss=0.1864, over 28827.00 frames. ], tot_loss[loss=0.4377, simple_loss=0.4619, pruned_loss=0.2068, over 5667928.07 frames. ], libri_tot_loss[loss=0.458, simple_loss=0.4723, pruned_loss=0.2218, over 5722068.59 frames. ], giga_tot_loss[loss=0.4367, simple_loss=0.4618, pruned_loss=0.2058, over 5664396.34 frames. ], batch size: 99, lr: 2.23e-02, grad_scale: 4.0
+2023-02-28 17:10:51,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.475e+02 1.544e+03 1.853e+03 2.206e+03 5.246e+03, threshold=3.705e+03, percent-clipped=4.0
+2023-02-28 17:11:36,004 INFO [train.py:968] (0/2) Epoch 1, batch 24600, giga_loss[loss=0.4275, simple_loss=0.4707, pruned_loss=0.1922, over 28693.00 frames. ], tot_loss[loss=0.434, simple_loss=0.4607, pruned_loss=0.2036, over 5671208.22 frames. ], libri_tot_loss[loss=0.4581, simple_loss=0.4723, pruned_loss=0.2219, over 5722233.49 frames. ], giga_tot_loss[loss=0.4324, simple_loss=0.4602, pruned_loss=0.2023, over 5666838.86 frames. ], batch size: 262, lr: 2.23e-02, grad_scale: 4.0
+2023-02-28 17:11:53,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7542, 1.9180, 4.7730, 3.1213], device='cuda:0'), covar=tensor([0.1586, 0.1155, 0.0229, 0.0467], device='cuda:0'), in_proj_covar=tensor([0.0492, 0.0453, 0.0590, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0005], device='cuda:0')
+2023-02-28 17:12:31,093 INFO [train.py:968] (0/2) Epoch 1, batch 24650, giga_loss[loss=0.4234, simple_loss=0.4651, pruned_loss=0.1908, over 28830.00 frames. ], tot_loss[loss=0.433, simple_loss=0.4616, pruned_loss=0.2022, over 5670068.83 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4721, pruned_loss=0.2219, over 5725621.74 frames. ], giga_tot_loss[loss=0.4314, simple_loss=0.4612, pruned_loss=0.2008, over 5662751.98 frames. ], batch size: 119, lr: 2.23e-02, grad_scale: 4.0
+2023-02-28 17:12:37,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.737e+03 2.417e+03 3.106e+03 1.084e+04, threshold=4.834e+03, percent-clipped=11.0
+2023-02-28 17:13:23,144 INFO [train.py:968] (0/2) Epoch 1, batch 24700, giga_loss[loss=0.4386, simple_loss=0.4383, pruned_loss=0.2194, over 23829.00 frames. ], tot_loss[loss=0.4342, simple_loss=0.4624, pruned_loss=0.2031, over 5658981.34 frames. ], libri_tot_loss[loss=0.4575, simple_loss=0.4719, pruned_loss=0.2216, over 5720595.24 frames. ], giga_tot_loss[loss=0.4327, simple_loss=0.462, pruned_loss=0.2017, over 5656682.39 frames. ], batch size: 705, lr: 2.23e-02, grad_scale: 4.0
+2023-02-28 17:14:14,051 INFO [train.py:968] (0/2) Epoch 1, batch 24750, giga_loss[loss=0.517, simple_loss=0.5022, pruned_loss=0.2659, over 26642.00 frames. ], tot_loss[loss=0.4353, simple_loss=0.4627, pruned_loss=0.204, over 5651243.09 frames. ], libri_tot_loss[loss=0.4571, simple_loss=0.4715, pruned_loss=0.2214, over 5714506.99 frames. ], giga_tot_loss[loss=0.4341, simple_loss=0.4625, pruned_loss=0.2029, over 5653200.01 frames. ], batch size: 555, lr: 2.22e-02, grad_scale: 4.0
+2023-02-28 17:14:18,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.749e+03 2.265e+03 2.865e+03 8.240e+03, threshold=4.529e+03, percent-clipped=6.0
+2023-02-28 17:14:18,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8177, 2.0473, 4.7273, 3.1259], device='cuda:0'), covar=tensor([0.1595, 0.1114, 0.0217, 0.0434], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0459, 0.0607, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0005], device='cuda:0')
+2023-02-28 17:15:03,078 INFO [train.py:968] (0/2) Epoch 1, batch 24800, giga_loss[loss=0.3898, simple_loss=0.4292, pruned_loss=0.1752, over 28570.00 frames. ], tot_loss[loss=0.4332, simple_loss=0.46, pruned_loss=0.2032, over 5646955.93 frames. ], libri_tot_loss[loss=0.4574, simple_loss=0.4717, pruned_loss=0.2215, over 5716731.71 frames. ], giga_tot_loss[loss=0.4316, simple_loss=0.4595, pruned_loss=0.2018, over 5645351.98 frames. ], batch size: 242, lr: 2.22e-02, grad_scale: 8.0
+2023-02-28 17:15:07,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24805.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:15:47,103 INFO [train.py:968] (0/2) Epoch 1, batch 24850, giga_loss[loss=0.467, simple_loss=0.4749, pruned_loss=0.2295, over 28374.00 frames. ], tot_loss[loss=0.4335, simple_loss=0.4591, pruned_loss=0.204, over 5655959.22 frames. ], libri_tot_loss[loss=0.4573, simple_loss=0.4716, pruned_loss=0.2215, over 5709980.32 frames. ], giga_tot_loss[loss=0.4316, simple_loss=0.4584, pruned_loss=0.2023, over 5658275.83 frames. ], batch size: 368, lr: 2.22e-02, grad_scale: 4.0
+2023-02-28 17:15:54,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.639e+02 1.661e+03 2.184e+03 2.694e+03 5.458e+03, threshold=4.368e+03, percent-clipped=5.0
+2023-02-28 17:16:32,105 INFO [train.py:968] (0/2) Epoch 1, batch 24900, giga_loss[loss=0.3933, simple_loss=0.4318, pruned_loss=0.1775, over 28284.00 frames. ], tot_loss[loss=0.4336, simple_loss=0.4587, pruned_loss=0.2042, over 5651256.69 frames. ], libri_tot_loss[loss=0.4564, simple_loss=0.4709, pruned_loss=0.2209, over 5703864.44 frames. ], giga_tot_loss[loss=0.432, simple_loss=0.4584, pruned_loss=0.2028, over 5657145.41 frames. ], batch size: 65, lr: 2.22e-02, grad_scale: 4.0
+2023-02-28 17:16:47,237 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-02-28 17:17:06,063 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=24940.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:17:13,016 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24948.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:17:16,610 INFO [train.py:968] (0/2) Epoch 1, batch 24950, giga_loss[loss=0.5539, simple_loss=0.5195, pruned_loss=0.2941, over 26749.00 frames. ], tot_loss[loss=0.4313, simple_loss=0.4584, pruned_loss=0.2021, over 5665271.63 frames. ], libri_tot_loss[loss=0.4563, simple_loss=0.4708, pruned_loss=0.2209, over 5701852.85 frames. ], giga_tot_loss[loss=0.4298, simple_loss=0.458, pruned_loss=0.2008, over 5670896.84 frames. ], batch size: 555, lr: 2.22e-02, grad_scale: 4.0
+2023-02-28 17:17:18,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6951, 1.5128, 1.2474, 1.3052], device='cuda:0'), covar=tensor([0.0751, 0.0940, 0.1096, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0632, 0.0629, 0.0587], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 17:17:18,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24951.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:17:26,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.605e+03 2.068e+03 2.897e+03 6.278e+03, threshold=4.135e+03, percent-clipped=4.0
+2023-02-28 17:17:29,735 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=24962.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:17:49,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24980.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:18:05,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0434, 2.5036, 2.0512, 1.9145], device='cuda:0'), covar=tensor([0.1305, 0.0547, 0.0668, 0.1636], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0354, 0.0342, 0.0557], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0020], device='cuda:0')
+2023-02-28 17:18:07,235 INFO [train.py:968] (0/2) Epoch 1, batch 25000, giga_loss[loss=0.4403, simple_loss=0.47, pruned_loss=0.2053, over 28417.00 frames. ], tot_loss[loss=0.4299, simple_loss=0.4572, pruned_loss=0.2013, over 5657646.90 frames. ], libri_tot_loss[loss=0.4567, simple_loss=0.4709, pruned_loss=0.2212, over 5703535.78 frames. ], giga_tot_loss[loss=0.4277, simple_loss=0.4565, pruned_loss=0.1994, over 5659536.24 frames. ], batch size: 71, lr: 2.21e-02, grad_scale: 4.0
+2023-02-28 17:18:17,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2935, 1.5771, 1.3577, 1.3100], device='cuda:0'), covar=tensor([0.1548, 0.0718, 0.0808, 0.1984], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0351, 0.0342, 0.0557], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0020], device='cuda:0')
+2023-02-28 17:18:41,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3610, 1.8621, 1.9662, 1.6868], device='cuda:0'), covar=tensor([0.0675, 0.1318, 0.0831, 0.1054], device='cuda:0'), in_proj_covar=tensor([0.0555, 0.0919, 0.0669, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 17:18:43,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4735, 1.6881, 1.8114, 1.6264], device='cuda:0'), covar=tensor([0.0731, 0.2027, 0.1202, 0.1469], device='cuda:0'), in_proj_covar=tensor([0.0556, 0.0920, 0.0669, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 17:18:52,485 INFO [train.py:968] (0/2) Epoch 1, batch 25050, giga_loss[loss=0.4424, simple_loss=0.4567, pruned_loss=0.214, over 28740.00 frames. ], tot_loss[loss=0.43, simple_loss=0.4576, pruned_loss=0.2012, over 5664553.01 frames. ], libri_tot_loss[loss=0.4574, simple_loss=0.4714, pruned_loss=0.2217, over 5699229.36 frames. ], giga_tot_loss[loss=0.4268, simple_loss=0.4562, pruned_loss=0.1987, over 5668779.45 frames. ], batch size: 92, lr: 2.21e-02, grad_scale: 4.0
+2023-02-28 17:19:01,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.148e+02 1.747e+03 2.250e+03 3.448e+03 8.440e+03, threshold=4.500e+03, percent-clipped=13.0
+2023-02-28 17:19:37,732 INFO [train.py:968] (0/2) Epoch 1, batch 25100, giga_loss[loss=0.4861, simple_loss=0.4922, pruned_loss=0.24, over 27964.00 frames. ], tot_loss[loss=0.4307, simple_loss=0.4576, pruned_loss=0.2019, over 5670185.68 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4716, pruned_loss=0.2219, over 5697326.87 frames. ], giga_tot_loss[loss=0.4266, simple_loss=0.4556, pruned_loss=0.1988, over 5674201.17 frames. ], batch size: 412, lr: 2.21e-02, grad_scale: 4.0
+2023-02-28 17:20:18,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6891, 2.5446, 2.1966, 0.4786], device='cuda:0'), covar=tensor([0.1106, 0.0692, 0.0707, 0.1828], device='cuda:0'), in_proj_covar=tensor([0.0951, 0.0965, 0.0957, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 17:20:21,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4222, 1.3044, 1.3153, 1.5828], device='cuda:0'), covar=tensor([0.1273, 0.1218, 0.0950, 0.1478], device='cuda:0'), in_proj_covar=tensor([0.0852, 0.0787, 0.0842, 0.0899], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 17:20:29,422 INFO [train.py:968] (0/2) Epoch 1, batch 25150, giga_loss[loss=0.405, simple_loss=0.4438, pruned_loss=0.183, over 28993.00 frames. ], tot_loss[loss=0.4297, simple_loss=0.4562, pruned_loss=0.2016, over 5672184.20 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.4715, pruned_loss=0.2219, over 5697719.44 frames. ], giga_tot_loss[loss=0.426, simple_loss=0.4544, pruned_loss=0.1988, over 5674537.09 frames. ], batch size: 145, lr: 2.21e-02, grad_scale: 4.0
+2023-02-28 17:20:38,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.842e+03 2.428e+03 3.204e+03 6.123e+03, threshold=4.856e+03, percent-clipped=5.0
+2023-02-28 17:20:38,392 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=25158.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:21:15,121 INFO [train.py:968] (0/2) Epoch 1, batch 25200, giga_loss[loss=0.3929, simple_loss=0.4292, pruned_loss=0.1782, over 28931.00 frames. ], tot_loss[loss=0.4296, simple_loss=0.4555, pruned_loss=0.2019, over 5681198.87 frames. ], libri_tot_loss[loss=0.4575, simple_loss=0.4715, pruned_loss=0.2218, over 5698802.38 frames. ], giga_tot_loss[loss=0.4261, simple_loss=0.4537, pruned_loss=0.1992, over 5681759.23 frames. ], batch size: 227, lr: 2.21e-02, grad_scale: 8.0
+2023-02-28 17:21:52,915 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=25241.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:22:02,161 INFO [train.py:968] (0/2) Epoch 1, batch 25250, giga_loss[loss=0.3668, simple_loss=0.4121, pruned_loss=0.1607, over 28809.00 frames. ], tot_loss[loss=0.4296, simple_loss=0.4551, pruned_loss=0.202, over 5682741.41 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4717, pruned_loss=0.2221, over 5693113.20 frames. ], giga_tot_loss[loss=0.4254, simple_loss=0.453, pruned_loss=0.1989, over 5687880.21 frames. ], batch size: 92, lr: 2.20e-02, grad_scale: 4.0
+2023-02-28 17:22:09,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 2.154e+03 2.619e+03 3.447e+03 7.047e+03, threshold=5.238e+03, percent-clipped=4.0
+2023-02-28 17:22:43,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3269, 1.5365, 1.3335, 1.1829], device='cuda:0'), covar=tensor([0.1122, 0.0851, 0.0980, 0.1755], device='cuda:0'), in_proj_covar=tensor([0.0462, 0.0497, 0.0401, 0.0505], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0011, 0.0009, 0.0012], device='cuda:0')
+2023-02-28 17:22:47,783 INFO [train.py:968] (0/2) Epoch 1, batch 25300, giga_loss[loss=0.3855, simple_loss=0.4262, pruned_loss=0.1724, over 28983.00 frames. ], tot_loss[loss=0.4259, simple_loss=0.4519, pruned_loss=0.2, over 5682266.09 frames. ], libri_tot_loss[loss=0.458, simple_loss=0.4715, pruned_loss=0.2223, over 5694936.44 frames. ], giga_tot_loss[loss=0.4216, simple_loss=0.4497, pruned_loss=0.1967, over 5684777.39 frames. ], batch size: 213, lr: 2.20e-02, grad_scale: 2.0
+2023-02-28 17:23:04,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25315.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:23:04,513 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-02-28 17:23:25,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25337.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:23:39,396 INFO [train.py:968] (0/2) Epoch 1, batch 25350, giga_loss[loss=0.43, simple_loss=0.4562, pruned_loss=0.202, over 28652.00 frames. ], tot_loss[loss=0.4259, simple_loss=0.4515, pruned_loss=0.2002, over 5679639.05 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4715, pruned_loss=0.2221, over 5695388.75 frames. ], giga_tot_loss[loss=0.4221, simple_loss=0.4494, pruned_loss=0.1974, over 5681072.22 frames. ], batch size: 307, lr: 2.20e-02, grad_scale: 2.0
+2023-02-28 17:23:49,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.878e+02 1.643e+03 2.178e+03 3.194e+03 1.158e+04, threshold=4.357e+03, percent-clipped=5.0
+2023-02-28 17:23:49,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7936, 2.8785, 5.6152, 3.5774], device='cuda:0'), covar=tensor([0.1175, 0.0839, 0.0208, 0.0323], device='cuda:0'), in_proj_covar=tensor([0.0490, 0.0449, 0.0592, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0005], device='cuda:0')
+2023-02-28 17:24:05,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1950, 1.6161, 1.4693, 1.5640], device='cuda:0'), covar=tensor([0.0681, 0.1631, 0.1151, 0.1223], device='cuda:0'), in_proj_covar=tensor([0.0560, 0.0920, 0.0682, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 17:24:14,444 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=25388.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:24:15,265 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-02-28 17:24:27,249 INFO [train.py:968] (0/2) Epoch 1, batch 25400, giga_loss[loss=0.5274, simple_loss=0.5015, pruned_loss=0.2767, over 26549.00 frames. ], tot_loss[loss=0.4282, simple_loss=0.4534, pruned_loss=0.2015, over 5681633.34 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.4714, pruned_loss=0.222, over 5702146.75 frames. ], giga_tot_loss[loss=0.4242, simple_loss=0.4513, pruned_loss=0.1986, over 5676037.14 frames. ], batch size: 555, lr: 2.20e-02, grad_scale: 2.0
+2023-02-28 17:25:14,203 INFO [train.py:968] (0/2) Epoch 1, batch 25450, giga_loss[loss=0.3759, simple_loss=0.4283, pruned_loss=0.1618, over 28902.00 frames. ], tot_loss[loss=0.4248, simple_loss=0.4526, pruned_loss=0.1986, over 5690062.45 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4716, pruned_loss=0.222, over 5703315.02 frames. ], giga_tot_loss[loss=0.4214, simple_loss=0.4506, pruned_loss=0.1961, over 5684593.51 frames. ], batch size: 213, lr: 2.20e-02, grad_scale: 2.0
+2023-02-28 17:25:24,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7786, 1.4293, 1.2469, 1.2406], device='cuda:0'), covar=tensor([0.0563, 0.0736, 0.0905, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0619, 0.0612, 0.0567], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 17:25:24,498 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25458.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:25:25,621 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.909e+02 1.709e+03 2.146e+03 2.560e+03 4.304e+03, threshold=4.291e+03, percent-clipped=0.0
+2023-02-28 17:25:27,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25461.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:25:47,038 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25480.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:25:49,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25483.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:25:55,264 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25490.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:26:04,917 INFO [train.py:968] (0/2) Epoch 1, batch 25500, giga_loss[loss=0.4706, simple_loss=0.4885, pruned_loss=0.2264, over 28906.00 frames. ], tot_loss[loss=0.4246, simple_loss=0.4529, pruned_loss=0.1982, over 5691082.67 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4716, pruned_loss=0.2221, over 5705386.68 frames. ], giga_tot_loss[loss=0.4215, simple_loss=0.4512, pruned_loss=0.1959, over 5684706.16 frames. ], batch size: 145, lr: 2.19e-02, grad_scale: 2.0
+2023-02-28 17:26:19,904 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25512.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:26:38,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25533.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:26:54,260 INFO [train.py:968] (0/2) Epoch 1, batch 25550, libri_loss[loss=0.4992, simple_loss=0.4951, pruned_loss=0.2517, over 19292.00 frames. ], tot_loss[loss=0.427, simple_loss=0.4541, pruned_loss=0.2, over 5677971.48 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4714, pruned_loss=0.2222, over 5697464.96 frames. ], giga_tot_loss[loss=0.4241, simple_loss=0.4526, pruned_loss=0.1978, over 5680526.23 frames. ], batch size: 187, lr: 2.19e-02, grad_scale: 2.0
+2023-02-28 17:27:03,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.861e+03 2.490e+03 3.725e+03 9.272e+03, threshold=4.980e+03, percent-clipped=15.0
+2023-02-28 17:27:44,110 INFO [train.py:968] (0/2) Epoch 1, batch 25600, libri_loss[loss=0.5492, simple_loss=0.5405, pruned_loss=0.2789, over 28599.00 frames. ], tot_loss[loss=0.434, simple_loss=0.4586, pruned_loss=0.2047, over 5679910.75 frames. ], libri_tot_loss[loss=0.4586, simple_loss=0.4719, pruned_loss=0.2227, over 5699055.08 frames. ], giga_tot_loss[loss=0.4302, simple_loss=0.4565, pruned_loss=0.2019, over 5679990.15 frames. ], batch size: 106, lr: 2.19e-02, grad_scale: 4.0
+2023-02-28 17:27:59,954 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25616.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:28:15,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6394, 1.7501, 4.0674, 2.8542], device='cuda:0'), covar=tensor([0.1419, 0.1066, 0.0240, 0.0389], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0448, 0.0586, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0005], device='cuda:0')
+2023-02-28 17:28:36,576 INFO [train.py:968] (0/2) Epoch 1, batch 25650, giga_loss[loss=0.5298, simple_loss=0.4998, pruned_loss=0.2799, over 26686.00 frames. ], tot_loss[loss=0.4368, simple_loss=0.4595, pruned_loss=0.2071, over 5677665.95 frames. ], libri_tot_loss[loss=0.4583, simple_loss=0.4717, pruned_loss=0.2224, over 5701843.18 frames. ], giga_tot_loss[loss=0.4337, simple_loss=0.4578, pruned_loss=0.2048, over 5674949.26 frames. ], batch size: 555, lr: 2.19e-02, grad_scale: 2.0
+2023-02-28 17:28:47,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.930e+03 2.513e+03 3.566e+03 9.883e+03, threshold=5.027e+03, percent-clipped=8.0
+2023-02-28 17:29:04,667 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25676.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:29:10,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25679.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:29:32,659 INFO [train.py:968] (0/2) Epoch 1, batch 25700, giga_loss[loss=0.3808, simple_loss=0.4292, pruned_loss=0.1662, over 28894.00 frames. ], tot_loss[loss=0.4387, simple_loss=0.4597, pruned_loss=0.2088, over 5675525.98 frames. ], libri_tot_loss[loss=0.4583, simple_loss=0.4717, pruned_loss=0.2224, over 5701843.18 frames. ], giga_tot_loss[loss=0.4363, simple_loss=0.4584, pruned_loss=0.2071, over 5673411.55 frames. ], batch size: 174, lr: 2.19e-02, grad_scale: 2.0
+2023-02-28 17:29:42,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25708.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:30:21,056 INFO [train.py:968] (0/2) Epoch 1, batch 25750, giga_loss[loss=0.3911, simple_loss=0.4301, pruned_loss=0.176, over 28951.00 frames. ], tot_loss[loss=0.4395, simple_loss=0.4604, pruned_loss=0.2093, over 5682529.13 frames. ], libri_tot_loss[loss=0.4586, simple_loss=0.472, pruned_loss=0.2226, over 5703980.72 frames. ], giga_tot_loss[loss=0.4372, simple_loss=0.459, pruned_loss=0.2077, over 5678808.61 frames. ], batch size: 213, lr: 2.18e-02, grad_scale: 2.0
+2023-02-28 17:30:31,470 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25759.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:30:32,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.088e+03 1.894e+03 2.500e+03 3.351e+03 7.380e+03, threshold=5.000e+03, percent-clipped=11.0
+2023-02-28 17:30:34,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25762.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:30:35,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25763.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:31:01,196 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25791.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:31:06,268 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-02-28 17:31:09,335 INFO [train.py:968] (0/2) Epoch 1, batch 25800, giga_loss[loss=0.3801, simple_loss=0.4249, pruned_loss=0.1676, over 29012.00 frames. ], tot_loss[loss=0.4375, simple_loss=0.4583, pruned_loss=0.2083, over 5672754.99 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4713, pruned_loss=0.2221, over 5705974.79 frames. ], giga_tot_loss[loss=0.4359, simple_loss=0.4576, pruned_loss=0.2071, over 5667257.37 frames. ], batch size: 155, lr: 2.18e-02, grad_scale: 2.0
+2023-02-28 17:31:30,463 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=25821.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 17:31:35,848 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-02-28 17:31:40,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8367, 2.8501, 3.5651, 1.8913], device='cuda:0'), covar=tensor([0.0514, 0.0577, 0.0815, 0.1485], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0472, 0.0824, 0.0533], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 17:31:56,011 INFO [train.py:968] (0/2) Epoch 1, batch 25850, giga_loss[loss=0.3745, simple_loss=0.433, pruned_loss=0.158, over 29112.00 frames. ], tot_loss[loss=0.4353, simple_loss=0.4582, pruned_loss=0.2062, over 5678355.45 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4713, pruned_loss=0.2221, over 5708058.70 frames. ], giga_tot_loss[loss=0.4339, simple_loss=0.4575, pruned_loss=0.2051, over 5671904.47 frames. ], batch size: 128, lr: 2.18e-02, grad_scale: 2.0
+2023-02-28 17:32:08,300 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.135e+03 1.891e+03 2.396e+03 3.250e+03 6.363e+03, threshold=4.791e+03, percent-clipped=7.0
+2023-02-28 17:32:36,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9510, 1.6700, 1.5293, 1.6342], device='cuda:0'), covar=tensor([0.0625, 0.1214, 0.1062, 0.1000], device='cuda:0'), in_proj_covar=tensor([0.0539, 0.0868, 0.0670, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 17:32:51,240 INFO [train.py:968] (0/2) Epoch 1, batch 25900, giga_loss[loss=0.4364, simple_loss=0.4602, pruned_loss=0.2063, over 28484.00 frames. ], tot_loss[loss=0.4277, simple_loss=0.4524, pruned_loss=0.2015, over 5659807.86 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4713, pruned_loss=0.2221, over 5708058.70 frames. ], giga_tot_loss[loss=0.4266, simple_loss=0.4518, pruned_loss=0.2007, over 5654786.98 frames. ], batch size: 71, lr: 2.18e-02, grad_scale: 2.0
+2023-02-28 17:32:56,480 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25906.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:32:58,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25909.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:33:27,831 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25938.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:33:37,622 INFO [train.py:968] (0/2) Epoch 1, batch 25950, giga_loss[loss=0.3935, simple_loss=0.4291, pruned_loss=0.179, over 28912.00 frames. ], tot_loss[loss=0.4252, simple_loss=0.4502, pruned_loss=0.2001, over 5665950.16 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.4712, pruned_loss=0.222, over 5709695.99 frames. ], giga_tot_loss[loss=0.4234, simple_loss=0.4491, pruned_loss=0.1988, over 5658902.99 frames. ], batch size: 145, lr: 2.17e-02, grad_scale: 2.0
+2023-02-28 17:33:46,279 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-02-28 17:33:46,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.746e+03 2.239e+03 3.051e+03 6.714e+03, threshold=4.477e+03, percent-clipped=8.0
+2023-02-28 17:34:11,066 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4570, 1.6365, 1.4516, 0.4421], device='cuda:0'), covar=tensor([0.0732, 0.0593, 0.0645, 0.1230], device='cuda:0'), in_proj_covar=tensor([0.0950, 0.0954, 0.0949, 0.0899], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 17:34:24,972 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-26000.pt
+2023-02-28 17:34:25,293 INFO [train.py:968] (0/2) Epoch 1, batch 26000, giga_loss[loss=0.5286, simple_loss=0.4956, pruned_loss=0.2808, over 26623.00 frames. ], tot_loss[loss=0.4253, simple_loss=0.4495, pruned_loss=0.2005, over 5668752.10 frames. ], libri_tot_loss[loss=0.458, simple_loss=0.4715, pruned_loss=0.2223, over 5715850.25 frames. ], giga_tot_loss[loss=0.4225, simple_loss=0.4478, pruned_loss=0.1986, over 5656444.79 frames. ], batch size: 555, lr: 2.17e-02, grad_scale: 4.0
+2023-02-28 17:35:17,768 INFO [train.py:968] (0/2) Epoch 1, batch 26050, giga_loss[loss=0.4053, simple_loss=0.4357, pruned_loss=0.1874, over 28948.00 frames. ], tot_loss[loss=0.429, simple_loss=0.4521, pruned_loss=0.2029, over 5665151.65 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4715, pruned_loss=0.2222, over 5717449.11 frames. ], giga_tot_loss[loss=0.4262, simple_loss=0.4504, pruned_loss=0.201, over 5652546.38 frames. ], batch size: 112, lr: 2.17e-02, grad_scale: 4.0
+2023-02-28 17:35:28,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.104e+03 1.522e+03 2.149e+03 2.814e+03 8.033e+03, threshold=4.298e+03, percent-clipped=5.0
+2023-02-28 17:36:03,252 INFO [train.py:968] (0/2) Epoch 1, batch 26100, giga_loss[loss=0.3814, simple_loss=0.4369, pruned_loss=0.1629, over 28893.00 frames. ], tot_loss[loss=0.4317, simple_loss=0.4554, pruned_loss=0.2039, over 5673734.06 frames. ], libri_tot_loss[loss=0.4573, simple_loss=0.471, pruned_loss=0.2218, over 5724831.41 frames. ], giga_tot_loss[loss=0.4289, simple_loss=0.4537, pruned_loss=0.2021, over 5655130.20 frames. ], batch size: 186, lr: 2.17e-02, grad_scale: 4.0
+2023-02-28 17:36:21,851 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-02-28 17:36:54,434 INFO [train.py:968] (0/2) Epoch 1, batch 26150, giga_loss[loss=0.5343, simple_loss=0.5162, pruned_loss=0.2762, over 26611.00 frames. ], tot_loss[loss=0.4331, simple_loss=0.4596, pruned_loss=0.2033, over 5676303.39 frames. ], libri_tot_loss[loss=0.4571, simple_loss=0.4709, pruned_loss=0.2216, over 5727802.28 frames. ], giga_tot_loss[loss=0.4308, simple_loss=0.4581, pruned_loss=0.2017, over 5658068.17 frames. ], batch size: 555, lr: 2.17e-02, grad_scale: 4.0
+2023-02-28 17:37:05,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.723e+03 2.116e+03 3.158e+03 8.601e+03, threshold=4.233e+03, percent-clipped=7.0
+2023-02-28 17:37:41,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26196.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:37:45,332 INFO [train.py:968] (0/2) Epoch 1, batch 26200, giga_loss[loss=0.4152, simple_loss=0.4501, pruned_loss=0.1902, over 28858.00 frames. ], tot_loss[loss=0.4339, simple_loss=0.4608, pruned_loss=0.2035, over 5670677.76 frames. ], libri_tot_loss[loss=0.457, simple_loss=0.471, pruned_loss=0.2216, over 5728447.27 frames. ], giga_tot_loss[loss=0.4319, simple_loss=0.4595, pruned_loss=0.2022, over 5655660.73 frames. ], batch size: 119, lr: 2.16e-02, grad_scale: 4.0
+2023-02-28 17:38:15,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7052, 2.5393, 3.3914, 1.7373], device='cuda:0'), covar=tensor([0.0800, 0.0986, 0.1477, 0.1847], device='cuda:0'), in_proj_covar=tensor([0.0757, 0.0481, 0.0831, 0.0549], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 17:38:36,239 INFO [train.py:968] (0/2) Epoch 1, batch 26250, giga_loss[loss=0.3548, simple_loss=0.4119, pruned_loss=0.1488, over 28579.00 frames. ], tot_loss[loss=0.4375, simple_loss=0.4631, pruned_loss=0.206, over 5663810.13 frames. ], libri_tot_loss[loss=0.4575, simple_loss=0.4713, pruned_loss=0.2219, over 5728378.23 frames. ], giga_tot_loss[loss=0.4353, simple_loss=0.4617, pruned_loss=0.2045, over 5651219.99 frames. ], batch size: 60, lr: 2.16e-02, grad_scale: 2.0
+2023-02-28 17:38:37,946 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=26251.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:38:49,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.634e+02 1.879e+03 2.617e+03 3.240e+03 9.634e+03, threshold=5.233e+03, percent-clipped=11.0
+2023-02-28 17:38:59,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2013, 1.3252, 1.2929, 1.2856], device='cuda:0'), covar=tensor([0.1291, 0.0662, 0.0659, 0.1646], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0342, 0.0328, 0.0552], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0021], device='cuda:0')
+2023-02-28 17:39:24,942 INFO [train.py:968] (0/2) Epoch 1, batch 26300, giga_loss[loss=0.3751, simple_loss=0.4218, pruned_loss=0.1642, over 29032.00 frames. ], tot_loss[loss=0.4403, simple_loss=0.4643, pruned_loss=0.2081, over 5656681.06 frames. ], libri_tot_loss[loss=0.4575, simple_loss=0.4713, pruned_loss=0.2218, over 5727303.28 frames. ], giga_tot_loss[loss=0.4382, simple_loss=0.463, pruned_loss=0.2067, over 5647017.93 frames. ], batch size: 155, lr: 2.16e-02, grad_scale: 2.0
+2023-02-28 17:39:30,496 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-02-28 17:39:54,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4898, 2.0510, 1.7871, 1.4961], device='cuda:0'), covar=tensor([0.1374, 0.0583, 0.0688, 0.1743], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0343, 0.0330, 0.0550], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0021], device='cuda:0')
+2023-02-28 17:40:03,170 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26339.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:40:06,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26342.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:40:14,631 INFO [train.py:968] (0/2) Epoch 1, batch 26350, giga_loss[loss=0.3913, simple_loss=0.4219, pruned_loss=0.1803, over 28909.00 frames. ], tot_loss[loss=0.4396, simple_loss=0.463, pruned_loss=0.2081, over 5655965.05 frames. ], libri_tot_loss[loss=0.4575, simple_loss=0.4713, pruned_loss=0.2219, over 5730057.34 frames. ], giga_tot_loss[loss=0.4377, simple_loss=0.4619, pruned_loss=0.2067, over 5644580.33 frames. ], batch size: 119, lr: 2.16e-02, grad_scale: 2.0
+2023-02-28 17:40:16,417 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4991, 1.7959, 1.4932, 1.3555], device='cuda:0'), covar=tensor([0.0930, 0.0812, 0.0899, 0.1527], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0488, 0.0394, 0.0487], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0011, 0.0009, 0.0012], device='cuda:0')
+2023-02-28 17:40:26,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.700e+03 2.138e+03 2.853e+03 1.070e+04, threshold=4.277e+03, percent-clipped=2.0
+2023-02-28 17:40:34,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0506, 2.9076, 2.3423, 0.6946], device='cuda:0'), covar=tensor([0.1177, 0.0654, 0.0764, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.0963, 0.0963, 0.0962, 0.0900], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-02-28 17:40:36,106 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26371.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 17:41:00,360 INFO [train.py:968] (0/2) Epoch 1, batch 26400, giga_loss[loss=0.5028, simple_loss=0.4907, pruned_loss=0.2574, over 26647.00 frames. ], tot_loss[loss=0.4392, simple_loss=0.4622, pruned_loss=0.2082, over 5654488.87 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4715, pruned_loss=0.222, over 5728124.54 frames. ], giga_tot_loss[loss=0.4367, simple_loss=0.4607, pruned_loss=0.2064, over 5644103.31 frames. ], batch size: 555, lr: 2.16e-02, grad_scale: 4.0
+2023-02-28 17:41:16,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9749, 2.8819, 2.3678, 0.6825], device='cuda:0'), covar=tensor([0.0833, 0.0528, 0.0631, 0.1382], device='cuda:0'), in_proj_covar=tensor([0.0945, 0.0949, 0.0956, 0.0876], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 17:41:47,649 INFO [train.py:968] (0/2) Epoch 1, batch 26450, giga_loss[loss=0.3864, simple_loss=0.4272, pruned_loss=0.1728, over 28718.00 frames. ], tot_loss[loss=0.4372, simple_loss=0.4598, pruned_loss=0.2073, over 5663358.14 frames. ], libri_tot_loss[loss=0.4581, simple_loss=0.4717, pruned_loss=0.2222, over 5731422.38 frames. ], giga_tot_loss[loss=0.4345, simple_loss=0.4583, pruned_loss=0.2054, over 5650885.02 frames. ], batch size: 262, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:42:00,388 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 2.091e+03 2.742e+03 3.600e+03 7.491e+03, threshold=5.484e+03, percent-clipped=18.0
+2023-02-28 17:42:25,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7990, 1.5643, 1.3234, 1.3871], device='cuda:0'), covar=tensor([0.0671, 0.0820, 0.1104, 0.1001], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0618, 0.0627, 0.0575], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-02-28 17:42:39,266 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=26495.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:42:44,161 INFO [train.py:968] (0/2) Epoch 1, batch 26500, giga_loss[loss=0.4304, simple_loss=0.4586, pruned_loss=0.2011, over 28959.00 frames. ], tot_loss[loss=0.4365, simple_loss=0.4587, pruned_loss=0.2072, over 5640177.15 frames. ], libri_tot_loss[loss=0.4582, simple_loss=0.4718, pruned_loss=0.2223, over 5722440.27 frames. ], giga_tot_loss[loss=0.4342, simple_loss=0.4573, pruned_loss=0.2056, over 5637438.95 frames. ], batch size: 164, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:42:48,643 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 17:43:30,909 INFO [train.py:968] (0/2) Epoch 1, batch 26550, giga_loss[loss=0.3745, simple_loss=0.4092, pruned_loss=0.1699, over 28697.00 frames. ], tot_loss[loss=0.4364, simple_loss=0.4584, pruned_loss=0.2071, over 5646199.87 frames. ], libri_tot_loss[loss=0.4585, simple_loss=0.472, pruned_loss=0.2225, over 5724533.16 frames. ], giga_tot_loss[loss=0.4339, simple_loss=0.457, pruned_loss=0.2054, over 5640614.46 frames. ], batch size: 99, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:43:41,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=26561.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:43:41,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.064e+02 1.673e+03 2.328e+03 2.856e+03 6.607e+03, threshold=4.656e+03, percent-clipped=4.0
+2023-02-28 17:44:19,231 INFO [train.py:968] (0/2) Epoch 1, batch 26600, giga_loss[loss=0.3516, simple_loss=0.3974, pruned_loss=0.1529, over 28468.00 frames. ], tot_loss[loss=0.4323, simple_loss=0.4552, pruned_loss=0.2047, over 5656140.49 frames. ], libri_tot_loss[loss=0.4584, simple_loss=0.4719, pruned_loss=0.2225, over 5717944.01 frames. ], giga_tot_loss[loss=0.4301, simple_loss=0.4539, pruned_loss=0.2031, over 5656882.07 frames. ], batch size: 85, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:44:23,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7841, 1.5184, 1.6187, 0.9165], device='cuda:0'), covar=tensor([0.0449, 0.0312, 0.0194, 0.0427], device='cuda:0'), in_proj_covar=tensor([0.0766, 0.0557, 0.0568, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 17:44:42,793 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26626.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 17:45:02,573 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=26643.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:45:07,121 INFO [train.py:968] (0/2) Epoch 1, batch 26650, libri_loss[loss=0.4862, simple_loss=0.5019, pruned_loss=0.2353, over 29061.00 frames. ], tot_loss[loss=0.4289, simple_loss=0.4527, pruned_loss=0.2026, over 5668139.75 frames. ], libri_tot_loss[loss=0.4583, simple_loss=0.4718, pruned_loss=0.2224, over 5721061.35 frames. ], giga_tot_loss[loss=0.4267, simple_loss=0.4514, pruned_loss=0.201, over 5664821.54 frames. ], batch size: 101, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:45:18,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.287e+02 1.660e+03 2.289e+03 3.128e+03 6.011e+03, threshold=4.579e+03, percent-clipped=5.0
+2023-02-28 17:45:52,812 INFO [train.py:968] (0/2) Epoch 1, batch 26700, giga_loss[loss=0.4805, simple_loss=0.4916, pruned_loss=0.2347, over 28267.00 frames. ], tot_loss[loss=0.4309, simple_loss=0.4548, pruned_loss=0.2035, over 5674927.65 frames. ], libri_tot_loss[loss=0.4586, simple_loss=0.4719, pruned_loss=0.2226, over 5725866.94 frames. ], giga_tot_loss[loss=0.4279, simple_loss=0.453, pruned_loss=0.2014, over 5666189.37 frames. ], batch size: 368, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:46:42,433 INFO [train.py:968] (0/2) Epoch 1, batch 26750, libri_loss[loss=0.5104, simple_loss=0.5089, pruned_loss=0.256, over 29511.00 frames. ], tot_loss[loss=0.436, simple_loss=0.4592, pruned_loss=0.2064, over 5662126.99 frames. ], libri_tot_loss[loss=0.4593, simple_loss=0.4725, pruned_loss=0.2231, over 5715758.18 frames. ], giga_tot_loss[loss=0.4324, simple_loss=0.457, pruned_loss=0.2039, over 5663734.65 frames. ], batch size: 84, lr: 2.14e-02, grad_scale: 4.0
+2023-02-28 17:46:55,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.141e+03 1.818e+03 2.398e+03 3.171e+03 9.228e+03, threshold=4.795e+03, percent-clipped=6.0
+2023-02-28 17:47:04,184 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26769.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:47:09,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26772.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:47:33,536 INFO [train.py:968] (0/2) Epoch 1, batch 26800, giga_loss[loss=0.4147, simple_loss=0.4498, pruned_loss=0.1898, over 28688.00 frames. ], tot_loss[loss=0.4366, simple_loss=0.459, pruned_loss=0.207, over 5647418.01 frames. ], libri_tot_loss[loss=0.4593, simple_loss=0.4723, pruned_loss=0.2231, over 5709019.49 frames. ], giga_tot_loss[loss=0.4334, simple_loss=0.4572, pruned_loss=0.2048, over 5653686.34 frames. ], batch size: 262, lr: 2.14e-02, grad_scale: 8.0
+2023-02-28 17:47:36,072 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26801.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:47:42,295 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=26807.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:48:18,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-02-28 17:48:18,991 INFO [train.py:968] (0/2) Epoch 1, batch 26850, giga_loss[loss=0.3592, simple_loss=0.4269, pruned_loss=0.1457, over 29012.00 frames. ], tot_loss[loss=0.4357, simple_loss=0.4604, pruned_loss=0.2055, over 5646941.03 frames. ], libri_tot_loss[loss=0.4592, simple_loss=0.4721, pruned_loss=0.2231, over 5693703.26 frames. ], giga_tot_loss[loss=0.4328, simple_loss=0.4588, pruned_loss=0.2034, over 5664369.02 frames. ], batch size: 106, lr: 2.14e-02, grad_scale: 4.0
+2023-02-28 17:48:19,427 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-02-28 17:48:29,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.594e+02 1.625e+03 2.024e+03 2.489e+03 7.687e+03, threshold=4.047e+03, percent-clipped=2.0
+2023-02-28 17:48:35,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26870.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:48:51,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2977, 1.5368, 1.3890, 1.2357], device='cuda:0'), covar=tensor([0.1501, 0.0711, 0.0792, 0.1846], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0342, 0.0332, 0.0541], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0021], device='cuda:0')
+2023-02-28 17:49:05,423 INFO [train.py:968] (0/2) Epoch 1, batch 26900, giga_loss[loss=0.4868, simple_loss=0.4904, pruned_loss=0.2417, over 27956.00 frames. ], tot_loss[loss=0.4329, simple_loss=0.4598, pruned_loss=0.203, over 5655568.29 frames. ], libri_tot_loss[loss=0.4581, simple_loss=0.4713, pruned_loss=0.2225, over 5696581.81 frames. ], giga_tot_loss[loss=0.4309, simple_loss=0.459, pruned_loss=0.2014, over 5665504.14 frames. ], batch size: 412, lr: 2.14e-02, grad_scale: 4.0
+2023-02-28 17:49:07,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9073, 2.4501, 1.8735, 1.5507], device='cuda:0'), covar=tensor([0.1053, 0.0756, 0.0994, 0.1690], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0496, 0.0402, 0.0496], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0009, 0.0012], device='cuda:0')
+2023-02-28 17:49:27,310 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 17:49:39,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26936.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:49:50,979 INFO [train.py:968] (0/2) Epoch 1, batch 26950, giga_loss[loss=0.5439, simple_loss=0.5294, pruned_loss=0.2792, over 27922.00 frames. ], tot_loss[loss=0.4315, simple_loss=0.4605, pruned_loss=0.2013, over 5663843.67 frames. ], libri_tot_loss[loss=0.4582, simple_loss=0.4713, pruned_loss=0.2226, over 5701529.24 frames. ], giga_tot_loss[loss=0.429, simple_loss=0.4595, pruned_loss=0.1992, over 5666317.62 frames. ], batch size: 412, lr: 2.14e-02, grad_scale: 4.0
+2023-02-28 17:49:56,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9810, 1.5945, 1.5107, 1.5700], device='cuda:0'), covar=tensor([0.0586, 0.1213, 0.0932, 0.0920], device='cuda:0'), in_proj_covar=tensor([0.0545, 0.0880, 0.0664, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 17:50:01,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.627e+03 2.044e+03 2.644e+03 7.409e+03, threshold=4.088e+03, percent-clipped=5.0
+2023-02-28 17:50:36,369 INFO [train.py:968] (0/2) Epoch 1, batch 27000, giga_loss[loss=0.4008, simple_loss=0.4376, pruned_loss=0.182, over 28669.00 frames. ], tot_loss[loss=0.4379, simple_loss=0.4644, pruned_loss=0.2057, over 5673640.25 frames. ], libri_tot_loss[loss=0.4572, simple_loss=0.4705, pruned_loss=0.2219, over 5707249.19 frames. ], giga_tot_loss[loss=0.4361, simple_loss=0.4641, pruned_loss=0.204, over 5669638.24 frames. ], batch size: 92, lr: 2.13e-02, grad_scale: 4.0
+2023-02-28 17:50:36,374 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 17:50:44,997 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.3094, simple_loss=0.3959, pruned_loss=0.1115, over 944034.00 frames. 
+2023-02-28 17:50:44,998 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 17:50:57,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27013.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:51:00,426 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27016.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:51:02,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27018.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:51:29,778 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27045.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:51:33,706 INFO [train.py:968] (0/2) Epoch 1, batch 27050, giga_loss[loss=0.4496, simple_loss=0.4708, pruned_loss=0.2142, over 28832.00 frames. ], tot_loss[loss=0.4451, simple_loss=0.4681, pruned_loss=0.211, over 5676600.12 frames. ], libri_tot_loss[loss=0.4569, simple_loss=0.4701, pruned_loss=0.2219, over 5712367.03 frames. ], giga_tot_loss[loss=0.4435, simple_loss=0.4681, pruned_loss=0.2094, over 5667663.43 frames. ], batch size: 145, lr: 2.13e-02, grad_scale: 4.0
+2023-02-28 17:51:46,300 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 2.256e+03 2.815e+03 3.973e+03 1.042e+04, threshold=5.630e+03, percent-clipped=24.0
+2023-02-28 17:52:02,387 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27079.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:52:05,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27082.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:52:24,564 INFO [train.py:968] (0/2) Epoch 1, batch 27100, giga_loss[loss=0.4768, simple_loss=0.4889, pruned_loss=0.2323, over 28655.00 frames. ], tot_loss[loss=0.4454, simple_loss=0.4678, pruned_loss=0.2115, over 5684275.65 frames. ], libri_tot_loss[loss=0.4561, simple_loss=0.4694, pruned_loss=0.2214, over 5716249.67 frames. ], giga_tot_loss[loss=0.4445, simple_loss=0.4683, pruned_loss=0.2104, over 5673101.39 frames. ], batch size: 262, lr: 2.13e-02, grad_scale: 4.0
+2023-02-28 17:52:35,965 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27111.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:53:10,852 INFO [train.py:968] (0/2) Epoch 1, batch 27150, libri_loss[loss=0.5203, simple_loss=0.5135, pruned_loss=0.2635, over 29674.00 frames. ], tot_loss[loss=0.4456, simple_loss=0.4674, pruned_loss=0.2119, over 5674220.91 frames. ], libri_tot_loss[loss=0.4546, simple_loss=0.4682, pruned_loss=0.2205, over 5713918.99 frames. ], giga_tot_loss[loss=0.4458, simple_loss=0.4689, pruned_loss=0.2114, over 5665907.66 frames. ], batch size: 88, lr: 2.13e-02, grad_scale: 4.0
+2023-02-28 17:53:22,656 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27161.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:53:24,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.804e+03 2.208e+03 3.267e+03 1.029e+04, threshold=4.416e+03, percent-clipped=5.0
+2023-02-28 17:53:26,415 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27164.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:53:42,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27182.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:53:55,333 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27193.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:54:01,102 INFO [train.py:968] (0/2) Epoch 1, batch 27200, giga_loss[loss=0.3824, simple_loss=0.4402, pruned_loss=0.1623, over 28558.00 frames. ], tot_loss[loss=0.4408, simple_loss=0.4652, pruned_loss=0.2082, over 5680920.45 frames. ], libri_tot_loss[loss=0.455, simple_loss=0.4684, pruned_loss=0.2208, over 5716490.98 frames. ], giga_tot_loss[loss=0.4405, simple_loss=0.4662, pruned_loss=0.2074, over 5671662.55 frames. ], batch size: 71, lr: 2.13e-02, grad_scale: 8.0
+2023-02-28 17:54:25,568 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-02-28 17:54:55,081 INFO [train.py:968] (0/2) Epoch 1, batch 27250, giga_loss[loss=0.4685, simple_loss=0.4576, pruned_loss=0.2397, over 23445.00 frames. ], tot_loss[loss=0.4374, simple_loss=0.4642, pruned_loss=0.2053, over 5663001.80 frames. ], libri_tot_loss[loss=0.4552, simple_loss=0.4686, pruned_loss=0.2209, over 5717450.52 frames. ], giga_tot_loss[loss=0.4368, simple_loss=0.4648, pruned_loss=0.2045, over 5654578.78 frames. ], batch size: 705, lr: 2.12e-02, grad_scale: 2.0
+2023-02-28 17:55:09,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.207e+02 1.570e+03 2.101e+03 3.006e+03 1.355e+04, threshold=4.202e+03, percent-clipped=9.0
+2023-02-28 17:55:14,758 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0849, 1.0597, 0.8532, 1.1877], device='cuda:0'), covar=tensor([0.1389, 0.0624, 0.0822, 0.1800], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0341, 0.0333, 0.0555], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0014, 0.0021], device='cuda:0')
+2023-02-28 17:55:21,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-02-28 17:55:41,021 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=27297.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:55:42,912 INFO [train.py:968] (0/2) Epoch 1, batch 27300, libri_loss[loss=0.4592, simple_loss=0.4743, pruned_loss=0.222, over 28833.00 frames. ], tot_loss[loss=0.4395, simple_loss=0.466, pruned_loss=0.2065, over 5662679.54 frames. ], libri_tot_loss[loss=0.4556, simple_loss=0.4689, pruned_loss=0.2211, over 5711135.85 frames. ], giga_tot_loss[loss=0.4384, simple_loss=0.466, pruned_loss=0.2054, over 5661317.50 frames. ], batch size: 107, lr: 2.12e-02, grad_scale: 2.0
+2023-02-28 17:56:05,355 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27325.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:56:07,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27328.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:56:24,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=27343.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:56:29,997 INFO [train.py:968] (0/2) Epoch 1, batch 27350, giga_loss[loss=0.3819, simple_loss=0.4312, pruned_loss=0.1663, over 28821.00 frames. ], tot_loss[loss=0.4429, simple_loss=0.4675, pruned_loss=0.2091, over 5663695.33 frames. ], libri_tot_loss[loss=0.4553, simple_loss=0.4685, pruned_loss=0.221, over 5718122.02 frames. ], giga_tot_loss[loss=0.4416, simple_loss=0.4678, pruned_loss=0.2077, over 5654317.80 frames. ], batch size: 112, lr: 2.12e-02, grad_scale: 2.0
+2023-02-28 17:56:39,847 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27357.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:56:46,062 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.757e+03 2.295e+03 3.359e+03 6.119e+03, threshold=4.589e+03, percent-clipped=10.0
+2023-02-28 17:57:14,588 INFO [train.py:968] (0/2) Epoch 1, batch 27400, giga_loss[loss=0.407, simple_loss=0.4423, pruned_loss=0.1858, over 28592.00 frames. ], tot_loss[loss=0.4418, simple_loss=0.4665, pruned_loss=0.2085, over 5664740.01 frames. ], libri_tot_loss[loss=0.4553, simple_loss=0.4687, pruned_loss=0.2209, over 5714224.49 frames. ], giga_tot_loss[loss=0.4402, simple_loss=0.4665, pruned_loss=0.207, over 5658632.27 frames. ], batch size: 307, lr: 2.12e-02, grad_scale: 2.0
+2023-02-28 17:57:49,682 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 17:58:10,696 INFO [train.py:968] (0/2) Epoch 1, batch 27450, libri_loss[loss=0.4765, simple_loss=0.4966, pruned_loss=0.2282, over 29233.00 frames. ], tot_loss[loss=0.4391, simple_loss=0.4639, pruned_loss=0.2072, over 5669710.73 frames. ], libri_tot_loss[loss=0.4556, simple_loss=0.4691, pruned_loss=0.221, over 5716102.92 frames. ], giga_tot_loss[loss=0.4374, simple_loss=0.4635, pruned_loss=0.2056, over 5662160.54 frames. ], batch size: 94, lr: 2.12e-02, grad_scale: 2.0
+2023-02-28 17:58:23,650 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.585e+02 1.646e+03 2.087e+03 2.784e+03 6.492e+03, threshold=4.174e+03, percent-clipped=3.0
+2023-02-28 17:59:02,130 INFO [train.py:968] (0/2) Epoch 1, batch 27500, giga_loss[loss=0.4579, simple_loss=0.4722, pruned_loss=0.2218, over 27847.00 frames. ], tot_loss[loss=0.4357, simple_loss=0.4608, pruned_loss=0.2053, over 5663799.92 frames. ], libri_tot_loss[loss=0.4556, simple_loss=0.4692, pruned_loss=0.221, over 5709089.18 frames. ], giga_tot_loss[loss=0.4339, simple_loss=0.4602, pruned_loss=0.2038, over 5662769.37 frames. ], batch size: 412, lr: 2.11e-02, grad_scale: 2.0
+2023-02-28 17:59:45,748 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=27541.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:59:52,526 INFO [train.py:968] (0/2) Epoch 1, batch 27550, giga_loss[loss=0.4156, simple_loss=0.4513, pruned_loss=0.19, over 28859.00 frames. ], tot_loss[loss=0.4344, simple_loss=0.4592, pruned_loss=0.2048, over 5662579.27 frames. ], libri_tot_loss[loss=0.4559, simple_loss=0.4695, pruned_loss=0.2212, over 5712424.04 frames. ], giga_tot_loss[loss=0.4323, simple_loss=0.4584, pruned_loss=0.2032, over 5658105.65 frames. ], batch size: 186, lr: 2.11e-02, grad_scale: 2.0
+2023-02-28 17:59:56,634 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-02-28 18:00:04,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6819, 2.6010, 2.0185, 0.4510], device='cuda:0'), covar=tensor([0.1205, 0.0642, 0.0834, 0.1817], device='cuda:0'), in_proj_covar=tensor([0.0971, 0.0966, 0.0974, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 18:00:05,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.823e+02 1.869e+03 2.361e+03 3.103e+03 1.375e+04, threshold=4.722e+03, percent-clipped=8.0
+2023-02-28 18:00:39,582 INFO [train.py:968] (0/2) Epoch 1, batch 27600, giga_loss[loss=0.423, simple_loss=0.4512, pruned_loss=0.1974, over 28659.00 frames. ], tot_loss[loss=0.4378, simple_loss=0.4607, pruned_loss=0.2074, over 5661805.86 frames. ], libri_tot_loss[loss=0.4562, simple_loss=0.4697, pruned_loss=0.2213, over 5707883.05 frames. ], giga_tot_loss[loss=0.4354, simple_loss=0.4596, pruned_loss=0.2056, over 5660948.61 frames. ], batch size: 307, lr: 2.11e-02, grad_scale: 4.0
+2023-02-28 18:01:23,393 INFO [train.py:968] (0/2) Epoch 1, batch 27650, giga_loss[loss=0.3871, simple_loss=0.43, pruned_loss=0.1721, over 28897.00 frames. ], tot_loss[loss=0.4354, simple_loss=0.4588, pruned_loss=0.206, over 5636906.79 frames. ], libri_tot_loss[loss=0.4564, simple_loss=0.4698, pruned_loss=0.2215, over 5682083.97 frames. ], giga_tot_loss[loss=0.4328, simple_loss=0.4575, pruned_loss=0.204, over 5657618.82 frames. ], batch size: 186, lr: 2.11e-02, grad_scale: 4.0
+2023-02-28 18:01:39,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.817e+02 1.770e+03 2.281e+03 2.866e+03 8.131e+03, threshold=4.562e+03, percent-clipped=9.0
+2023-02-28 18:01:46,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27672.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:01:49,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2757, 2.3256, 5.1528, 3.6558], device='cuda:0'), covar=tensor([0.1214, 0.0957, 0.0229, 0.0341], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0450, 0.0597, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0005], device='cuda:0')
+2023-02-28 18:02:10,255 INFO [train.py:968] (0/2) Epoch 1, batch 27700, giga_loss[loss=0.4954, simple_loss=0.4912, pruned_loss=0.2498, over 26712.00 frames. ], tot_loss[loss=0.4297, simple_loss=0.456, pruned_loss=0.2017, over 5641566.42 frames. ], libri_tot_loss[loss=0.4564, simple_loss=0.4699, pruned_loss=0.2214, over 5681147.73 frames. ], giga_tot_loss[loss=0.4271, simple_loss=0.4546, pruned_loss=0.1998, over 5658055.80 frames. ], batch size: 555, lr: 2.11e-02, grad_scale: 4.0
+2023-02-28 18:02:28,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27718.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:02:31,202 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-02-28 18:03:00,139 INFO [train.py:968] (0/2) Epoch 1, batch 27750, giga_loss[loss=0.3906, simple_loss=0.4355, pruned_loss=0.1728, over 28923.00 frames. ], tot_loss[loss=0.4245, simple_loss=0.4524, pruned_loss=0.1983, over 5645169.09 frames. ], libri_tot_loss[loss=0.4556, simple_loss=0.4693, pruned_loss=0.2209, over 5684543.63 frames. ], giga_tot_loss[loss=0.4226, simple_loss=0.4516, pruned_loss=0.1968, over 5654512.12 frames. ], batch size: 213, lr: 2.11e-02, grad_scale: 4.0
+2023-02-28 18:03:18,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.625e+02 1.549e+03 1.954e+03 2.386e+03 4.881e+03, threshold=3.909e+03, percent-clipped=1.0
+2023-02-28 18:03:52,934 INFO [train.py:968] (0/2) Epoch 1, batch 27800, libri_loss[loss=0.4541, simple_loss=0.471, pruned_loss=0.2186, over 29654.00 frames. ], tot_loss[loss=0.4213, simple_loss=0.45, pruned_loss=0.1963, over 5650962.47 frames. ], libri_tot_loss[loss=0.4556, simple_loss=0.4694, pruned_loss=0.2209, over 5687090.58 frames. ], giga_tot_loss[loss=0.4194, simple_loss=0.4491, pruned_loss=0.1948, over 5655548.66 frames. ], batch size: 88, lr: 2.10e-02, grad_scale: 4.0
+2023-02-28 18:04:04,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-02-28 18:04:07,059 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 18:04:10,177 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27815.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:04:12,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27818.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:04:49,471 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27847.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:04:52,715 INFO [train.py:968] (0/2) Epoch 1, batch 27850, giga_loss[loss=0.3766, simple_loss=0.421, pruned_loss=0.166, over 29080.00 frames. ], tot_loss[loss=0.4169, simple_loss=0.4454, pruned_loss=0.1942, over 5643076.57 frames. ], libri_tot_loss[loss=0.4556, simple_loss=0.4694, pruned_loss=0.2209, over 5687017.33 frames. ], giga_tot_loss[loss=0.4151, simple_loss=0.4444, pruned_loss=0.1929, over 5646188.96 frames. ], batch size: 128, lr: 2.10e-02, grad_scale: 2.0
+2023-02-28 18:05:03,536 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27861.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:05:09,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27864.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:05:10,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.851e+02 1.893e+03 2.420e+03 3.290e+03 1.282e+04, threshold=4.840e+03, percent-clipped=14.0
+2023-02-28 18:05:36,378 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27893.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:05:41,880 INFO [train.py:968] (0/2) Epoch 1, batch 27900, giga_loss[loss=0.4035, simple_loss=0.4533, pruned_loss=0.1769, over 28901.00 frames. ], tot_loss[loss=0.4175, simple_loss=0.4459, pruned_loss=0.1946, over 5641121.41 frames. ], libri_tot_loss[loss=0.456, simple_loss=0.4696, pruned_loss=0.2212, over 5681647.48 frames. ], giga_tot_loss[loss=0.4149, simple_loss=0.4444, pruned_loss=0.1927, over 5646882.51 frames. ], batch size: 145, lr: 2.10e-02, grad_scale: 2.0
+2023-02-28 18:05:56,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27916.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:06:29,423 INFO [train.py:968] (0/2) Epoch 1, batch 27950, giga_loss[loss=0.3784, simple_loss=0.4265, pruned_loss=0.1652, over 28867.00 frames. ], tot_loss[loss=0.417, simple_loss=0.4464, pruned_loss=0.1938, over 5652820.66 frames. ], libri_tot_loss[loss=0.4559, simple_loss=0.4695, pruned_loss=0.2212, over 5687442.50 frames. ], giga_tot_loss[loss=0.4138, simple_loss=0.4447, pruned_loss=0.1915, over 5651384.61 frames. ], batch size: 145, lr: 2.10e-02, grad_scale: 2.0
+2023-02-28 18:06:46,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.536e+02 1.558e+03 1.942e+03 2.620e+03 5.358e+03, threshold=3.885e+03, percent-clipped=2.0
+2023-02-28 18:07:23,111 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-28000.pt
+2023-02-28 18:07:23,431 INFO [train.py:968] (0/2) Epoch 1, batch 28000, giga_loss[loss=0.4099, simple_loss=0.4372, pruned_loss=0.1913, over 28334.00 frames. ], tot_loss[loss=0.4132, simple_loss=0.4442, pruned_loss=0.1911, over 5646680.50 frames. ], libri_tot_loss[loss=0.4562, simple_loss=0.4698, pruned_loss=0.2213, over 5686048.52 frames. ], giga_tot_loss[loss=0.4101, simple_loss=0.4424, pruned_loss=0.1889, over 5646235.55 frames. ], batch size: 368, lr: 2.10e-02, grad_scale: 4.0
+2023-02-28 18:08:06,012 INFO [train.py:968] (0/2) Epoch 1, batch 28050, giga_loss[loss=0.3938, simple_loss=0.4316, pruned_loss=0.178, over 28737.00 frames. ], tot_loss[loss=0.4159, simple_loss=0.446, pruned_loss=0.193, over 5641719.14 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4712, pruned_loss=0.2223, over 5680289.58 frames. ], giga_tot_loss[loss=0.4104, simple_loss=0.4423, pruned_loss=0.1893, over 5644753.13 frames. ], batch size: 99, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:08:14,275 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28059.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:08:17,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28062.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:08:20,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.586e+02 1.740e+03 2.170e+03 2.940e+03 6.695e+03, threshold=4.339e+03, percent-clipped=12.0
+2023-02-28 18:08:42,466 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28091.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:08:50,300 INFO [train.py:968] (0/2) Epoch 1, batch 28100, giga_loss[loss=0.3752, simple_loss=0.4178, pruned_loss=0.1663, over 28785.00 frames. ], tot_loss[loss=0.4177, simple_loss=0.4468, pruned_loss=0.1943, over 5651506.87 frames. ], libri_tot_loss[loss=0.4585, simple_loss=0.4717, pruned_loss=0.2226, over 5686774.62 frames. ], giga_tot_loss[loss=0.4117, simple_loss=0.4428, pruned_loss=0.1903, over 5647260.29 frames. ], batch size: 119, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:08:57,190 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28107.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 18:09:29,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6975, 1.6865, 1.1161, 1.3062], device='cuda:0'), covar=tensor([0.0641, 0.0806, 0.1057, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0615, 0.0604, 0.0568], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-02-28 18:09:36,774 INFO [train.py:968] (0/2) Epoch 1, batch 28150, giga_loss[loss=0.433, simple_loss=0.4609, pruned_loss=0.2026, over 28905.00 frames. ], tot_loss[loss=0.4222, simple_loss=0.4499, pruned_loss=0.1973, over 5642355.98 frames. ], libri_tot_loss[loss=0.4593, simple_loss=0.4722, pruned_loss=0.2232, over 5679511.57 frames. ], giga_tot_loss[loss=0.4161, simple_loss=0.4459, pruned_loss=0.1931, over 5644042.47 frames. ], batch size: 186, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:09:51,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.646e+03 2.382e+03 3.008e+03 6.760e+03, threshold=4.764e+03, percent-clipped=4.0
+2023-02-28 18:10:22,090 INFO [train.py:968] (0/2) Epoch 1, batch 28200, libri_loss[loss=0.4302, simple_loss=0.4554, pruned_loss=0.2025, over 29526.00 frames. ], tot_loss[loss=0.4248, simple_loss=0.452, pruned_loss=0.1988, over 5658816.07 frames. ], libri_tot_loss[loss=0.458, simple_loss=0.4713, pruned_loss=0.2224, over 5689796.01 frames. ], giga_tot_loss[loss=0.4192, simple_loss=0.4486, pruned_loss=0.1949, over 5648919.01 frames. ], batch size: 84, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:11:15,744 INFO [train.py:968] (0/2) Epoch 1, batch 28250, giga_loss[loss=0.4555, simple_loss=0.4447, pruned_loss=0.2332, over 23547.00 frames. ], tot_loss[loss=0.4316, simple_loss=0.4562, pruned_loss=0.2035, over 5653346.04 frames. ], libri_tot_loss[loss=0.4587, simple_loss=0.4716, pruned_loss=0.2228, over 5692239.54 frames. ], giga_tot_loss[loss=0.4263, simple_loss=0.4529, pruned_loss=0.1998, over 5643179.45 frames. ], batch size: 705, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:11:31,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.696e+02 1.835e+03 2.347e+03 3.218e+03 9.503e+03, threshold=4.694e+03, percent-clipped=7.0
+2023-02-28 18:11:35,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6164, 1.7313, 3.9431, 2.8172], device='cuda:0'), covar=tensor([0.1441, 0.1194, 0.0280, 0.0412], device='cuda:0'), in_proj_covar=tensor([0.0476, 0.0454, 0.0582, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 18:12:03,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6111, 2.5809, 3.3363, 1.8671], device='cuda:0'), covar=tensor([0.0693, 0.0763, 0.1046, 0.1763], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0500, 0.0845, 0.0556], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0006], device='cuda:0')
+2023-02-28 18:12:04,771 INFO [train.py:968] (0/2) Epoch 1, batch 28300, giga_loss[loss=0.4789, simple_loss=0.4931, pruned_loss=0.2323, over 27542.00 frames. ], tot_loss[loss=0.4333, simple_loss=0.4569, pruned_loss=0.2049, over 5654672.56 frames. ], libri_tot_loss[loss=0.4583, simple_loss=0.4714, pruned_loss=0.2226, over 5697634.50 frames. ], giga_tot_loss[loss=0.4289, simple_loss=0.4542, pruned_loss=0.2018, over 5640863.71 frames. ], batch size: 472, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:12:12,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5672, 3.3893, 4.2102, 2.1089], device='cuda:0'), covar=tensor([0.0597, 0.0721, 0.1221, 0.1740], device='cuda:0'), in_proj_covar=tensor([0.0768, 0.0497, 0.0839, 0.0549], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0005, 0.0009, 0.0006], device='cuda:0')
+2023-02-28 18:12:57,616 INFO [train.py:968] (0/2) Epoch 1, batch 28350, giga_loss[loss=0.4002, simple_loss=0.4476, pruned_loss=0.1764, over 28829.00 frames. ], tot_loss[loss=0.4311, simple_loss=0.4567, pruned_loss=0.2027, over 5656148.64 frames. ], libri_tot_loss[loss=0.458, simple_loss=0.4713, pruned_loss=0.2224, over 5700271.19 frames. ], giga_tot_loss[loss=0.4272, simple_loss=0.4543, pruned_loss=0.2001, over 5641823.42 frames. ], batch size: 119, lr: 2.08e-02, grad_scale: 4.0
+2023-02-28 18:13:12,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.197e+03 1.898e+03 2.396e+03 3.388e+03 6.608e+03, threshold=4.792e+03, percent-clipped=9.0
+2023-02-28 18:13:32,136 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-02-28 18:13:48,737 INFO [train.py:968] (0/2) Epoch 1, batch 28400, giga_loss[loss=0.4817, simple_loss=0.4817, pruned_loss=0.2409, over 26722.00 frames. ], tot_loss[loss=0.4309, simple_loss=0.4567, pruned_loss=0.2025, over 5649511.29 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4711, pruned_loss=0.2223, over 5692923.51 frames. ], giga_tot_loss[loss=0.4275, simple_loss=0.4548, pruned_loss=0.2001, over 5644886.22 frames. ], batch size: 555, lr: 2.08e-02, grad_scale: 8.0
+2023-02-28 18:14:41,379 INFO [train.py:968] (0/2) Epoch 1, batch 28450, giga_loss[loss=0.3736, simple_loss=0.4157, pruned_loss=0.1657, over 28931.00 frames. ], tot_loss[loss=0.4312, simple_loss=0.456, pruned_loss=0.2032, over 5639869.24 frames. ], libri_tot_loss[loss=0.4572, simple_loss=0.4705, pruned_loss=0.2219, over 5696061.99 frames. ], giga_tot_loss[loss=0.4287, simple_loss=0.4548, pruned_loss=0.2013, over 5632487.74 frames. ], batch size: 213, lr: 2.08e-02, grad_scale: 4.0
+2023-02-28 18:15:02,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.954e+02 1.799e+03 2.423e+03 3.132e+03 6.327e+03, threshold=4.845e+03, percent-clipped=6.0
+2023-02-28 18:15:15,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9264, 2.3334, 1.9871, 1.7885], device='cuda:0'), covar=tensor([0.0779, 0.0724, 0.0815, 0.1196], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0482, 0.0391, 0.0479], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0009, 0.0012], device='cuda:0')
+2023-02-28 18:15:18,824 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28482.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 18:15:22,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0307, 1.5574, 1.4644, 1.5647], device='cuda:0'), covar=tensor([0.0690, 0.1736, 0.1187, 0.1132], device='cuda:0'), in_proj_covar=tensor([0.0533, 0.0887, 0.0671, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 18:15:40,615 INFO [train.py:968] (0/2) Epoch 1, batch 28500, giga_loss[loss=0.4663, simple_loss=0.4636, pruned_loss=0.2345, over 26653.00 frames. ], tot_loss[loss=0.4329, simple_loss=0.4566, pruned_loss=0.2046, over 5632444.01 frames. ], libri_tot_loss[loss=0.4572, simple_loss=0.4705, pruned_loss=0.2219, over 5690359.59 frames. ], giga_tot_loss[loss=0.4305, simple_loss=0.4554, pruned_loss=0.2028, over 5630647.43 frames. ], batch size: 555, lr: 2.08e-02, grad_scale: 4.0
+2023-02-28 18:16:07,332 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-02-28 18:16:38,449 INFO [train.py:968] (0/2) Epoch 1, batch 28550, libri_loss[loss=0.4382, simple_loss=0.4676, pruned_loss=0.2044, over 29666.00 frames. ], tot_loss[loss=0.4301, simple_loss=0.4539, pruned_loss=0.2032, over 5625092.05 frames. ], libri_tot_loss[loss=0.4569, simple_loss=0.4704, pruned_loss=0.2217, over 5692169.32 frames. ], giga_tot_loss[loss=0.4279, simple_loss=0.4527, pruned_loss=0.2016, over 5620363.29 frames. ], batch size: 88, lr: 2.08e-02, grad_scale: 2.0
+2023-02-28 18:16:56,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.452e+02 1.620e+03 2.218e+03 2.881e+03 1.440e+04, threshold=4.437e+03, percent-clipped=7.0
+2023-02-28 18:17:05,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0629, 2.4268, 2.0324, 1.8360], device='cuda:0'), covar=tensor([0.1025, 0.0813, 0.0963, 0.1672], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0488, 0.0393, 0.0486], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0012], device='cuda:0')
+2023-02-28 18:17:27,572 INFO [train.py:968] (0/2) Epoch 1, batch 28600, giga_loss[loss=0.3808, simple_loss=0.4204, pruned_loss=0.1706, over 28905.00 frames. ], tot_loss[loss=0.4294, simple_loss=0.4535, pruned_loss=0.2026, over 5644878.80 frames. ], libri_tot_loss[loss=0.4567, simple_loss=0.4704, pruned_loss=0.2216, over 5694339.11 frames. ], giga_tot_loss[loss=0.4275, simple_loss=0.4524, pruned_loss=0.2013, over 5638799.72 frames. ], batch size: 227, lr: 2.07e-02, grad_scale: 2.0
+2023-02-28 18:17:50,049 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28622.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:17:55,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28625.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:17:57,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28628.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:18:15,470 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28648.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:18:16,452 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1512, 1.1905, 1.0088, 0.9630], device='cuda:0'), covar=tensor([0.0517, 0.0511, 0.0759, 0.0777], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0621, 0.0608, 0.0572], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-02-28 18:18:18,639 INFO [train.py:968] (0/2) Epoch 1, batch 28650, giga_loss[loss=0.4995, simple_loss=0.4692, pruned_loss=0.2649, over 23600.00 frames. ], tot_loss[loss=0.4292, simple_loss=0.4531, pruned_loss=0.2027, over 5646139.69 frames. ], libri_tot_loss[loss=0.4568, simple_loss=0.4705, pruned_loss=0.2216, over 5698178.18 frames. ], giga_tot_loss[loss=0.4272, simple_loss=0.4518, pruned_loss=0.2013, over 5636960.06 frames. ], batch size: 705, lr: 2.07e-02, grad_scale: 2.0
+2023-02-28 18:18:25,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28657.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:18:34,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.652e+03 2.042e+03 2.851e+03 5.516e+03, threshold=4.084e+03, percent-clipped=3.0
+2023-02-28 18:18:45,260 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28680.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:19:03,478 INFO [train.py:968] (0/2) Epoch 1, batch 28700, giga_loss[loss=0.5053, simple_loss=0.48, pruned_loss=0.2653, over 23439.00 frames. ], tot_loss[loss=0.4308, simple_loss=0.4542, pruned_loss=0.2036, over 5641541.21 frames. ], libri_tot_loss[loss=0.4565, simple_loss=0.4702, pruned_loss=0.2214, over 5686974.10 frames. ], giga_tot_loss[loss=0.4284, simple_loss=0.4528, pruned_loss=0.202, over 5643240.75 frames. ], batch size: 705, lr: 2.07e-02, grad_scale: 2.0
+2023-02-28 18:19:55,839 INFO [train.py:968] (0/2) Epoch 1, batch 28750, giga_loss[loss=0.4612, simple_loss=0.4819, pruned_loss=0.2202, over 28659.00 frames. ], tot_loss[loss=0.4289, simple_loss=0.4533, pruned_loss=0.2022, over 5646769.76 frames. ], libri_tot_loss[loss=0.4561, simple_loss=0.4699, pruned_loss=0.2211, over 5681091.07 frames. ], giga_tot_loss[loss=0.4269, simple_loss=0.4522, pruned_loss=0.2008, over 5653005.44 frames. ], batch size: 307, lr: 2.07e-02, grad_scale: 2.0
+2023-02-28 18:20:08,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6021, 1.9976, 1.6897, 0.5789], device='cuda:0'), covar=tensor([0.0921, 0.0700, 0.0893, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0985, 0.0986, 0.0995, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 18:20:11,867 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.713e+02 1.723e+03 2.161e+03 2.942e+03 1.077e+04, threshold=4.322e+03, percent-clipped=11.0
+2023-02-28 18:20:44,124 INFO [train.py:968] (0/2) Epoch 1, batch 28800, giga_loss[loss=0.345, simple_loss=0.4002, pruned_loss=0.1449, over 28488.00 frames. ], tot_loss[loss=0.4306, simple_loss=0.4548, pruned_loss=0.2032, over 5650674.68 frames. ], libri_tot_loss[loss=0.4559, simple_loss=0.4697, pruned_loss=0.2211, over 5682787.54 frames. ], giga_tot_loss[loss=0.429, simple_loss=0.4539, pruned_loss=0.202, over 5653816.03 frames. ], batch size: 71, lr: 2.07e-02, grad_scale: 4.0
+2023-02-28 18:21:29,987 INFO [train.py:968] (0/2) Epoch 1, batch 28850, giga_loss[loss=0.3853, simple_loss=0.4197, pruned_loss=0.1754, over 28621.00 frames. ], tot_loss[loss=0.4341, simple_loss=0.4566, pruned_loss=0.2058, over 5655088.49 frames. ], libri_tot_loss[loss=0.455, simple_loss=0.469, pruned_loss=0.2205, over 5680418.55 frames. ], giga_tot_loss[loss=0.4326, simple_loss=0.456, pruned_loss=0.2046, over 5657910.33 frames. ], batch size: 78, lr: 2.07e-02, grad_scale: 4.0
+2023-02-28 18:21:47,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.296e+02 1.918e+03 2.523e+03 3.063e+03 8.018e+03, threshold=5.047e+03, percent-clipped=11.0
+2023-02-28 18:21:56,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28879.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:22:03,442 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28885.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:22:17,284 INFO [train.py:968] (0/2) Epoch 1, batch 28900, giga_loss[loss=0.4212, simple_loss=0.4529, pruned_loss=0.1948, over 28980.00 frames. ], tot_loss[loss=0.4335, simple_loss=0.4561, pruned_loss=0.2054, over 5654287.66 frames. ], libri_tot_loss[loss=0.4543, simple_loss=0.4684, pruned_loss=0.22, over 5673050.84 frames. ], giga_tot_loss[loss=0.4324, simple_loss=0.4558, pruned_loss=0.2046, over 5662563.76 frames. ], batch size: 128, lr: 2.06e-02, grad_scale: 4.0
+2023-02-28 18:22:18,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2074, 1.1126, 1.3368, 0.6891], device='cuda:0'), covar=tensor([0.0341, 0.0293, 0.0178, 0.0340], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0585, 0.0607, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 18:22:25,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-02-28 18:23:03,672 INFO [train.py:968] (0/2) Epoch 1, batch 28950, giga_loss[loss=0.3601, simple_loss=0.4138, pruned_loss=0.1532, over 28342.00 frames. ], tot_loss[loss=0.4338, simple_loss=0.4567, pruned_loss=0.2055, over 5667297.96 frames. ], libri_tot_loss[loss=0.4544, simple_loss=0.4685, pruned_loss=0.2202, over 5674866.03 frames. ], giga_tot_loss[loss=0.4323, simple_loss=0.4561, pruned_loss=0.2042, over 5671949.52 frames. ], batch size: 65, lr: 2.06e-02, grad_scale: 4.0
+2023-02-28 18:23:22,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.207e+03 1.911e+03 2.499e+03 3.684e+03 7.695e+03, threshold=4.998e+03, percent-clipped=10.0
+2023-02-28 18:23:52,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28997.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:23:55,486 INFO [train.py:968] (0/2) Epoch 1, batch 29000, giga_loss[loss=0.4176, simple_loss=0.4462, pruned_loss=0.1945, over 28744.00 frames. ], tot_loss[loss=0.4361, simple_loss=0.4586, pruned_loss=0.2067, over 5661256.07 frames. ], libri_tot_loss[loss=0.4543, simple_loss=0.4685, pruned_loss=0.2201, over 5677057.45 frames. ], giga_tot_loss[loss=0.4347, simple_loss=0.458, pruned_loss=0.2057, over 5663098.93 frames. ], batch size: 99, lr: 2.06e-02, grad_scale: 4.0
+2023-02-28 18:24:17,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29023.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:24:42,539 INFO [train.py:968] (0/2) Epoch 1, batch 29050, giga_loss[loss=0.3976, simple_loss=0.4309, pruned_loss=0.1822, over 28976.00 frames. ], tot_loss[loss=0.4359, simple_loss=0.4587, pruned_loss=0.2065, over 5669217.35 frames. ], libri_tot_loss[loss=0.454, simple_loss=0.4682, pruned_loss=0.2199, over 5678509.52 frames. ], giga_tot_loss[loss=0.4346, simple_loss=0.4582, pruned_loss=0.2055, over 5668895.43 frames. ], batch size: 106, lr: 2.06e-02, grad_scale: 4.0
+2023-02-28 18:24:47,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29055.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:25:00,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.631e+03 2.082e+03 2.687e+03 4.756e+03, threshold=4.165e+03, percent-clipped=0.0
+2023-02-28 18:25:29,617 INFO [train.py:968] (0/2) Epoch 1, batch 29100, giga_loss[loss=0.4639, simple_loss=0.4801, pruned_loss=0.2239, over 28875.00 frames. ], tot_loss[loss=0.4382, simple_loss=0.4603, pruned_loss=0.2081, over 5667819.68 frames. ], libri_tot_loss[loss=0.4539, simple_loss=0.4682, pruned_loss=0.2198, over 5680228.76 frames. ], giga_tot_loss[loss=0.437, simple_loss=0.4598, pruned_loss=0.2071, over 5665891.74 frames. ], batch size: 199, lr: 2.06e-02, grad_scale: 4.0
+2023-02-28 18:25:34,307 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 18:25:41,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3559, 1.7534, 1.4061, 1.4362], device='cuda:0'), covar=tensor([0.1189, 0.0502, 0.0637, 0.1395], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0331, 0.0321, 0.0537], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0014, 0.0022], device='cuda:0')
+2023-02-28 18:25:45,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-02-28 18:26:05,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29140.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:26:07,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29143.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:26:13,073 INFO [train.py:968] (0/2) Epoch 1, batch 29150, giga_loss[loss=0.4139, simple_loss=0.4398, pruned_loss=0.194, over 28860.00 frames. ], tot_loss[loss=0.4385, simple_loss=0.4602, pruned_loss=0.2084, over 5656812.69 frames. ], libri_tot_loss[loss=0.4544, simple_loss=0.4686, pruned_loss=0.2201, over 5672939.29 frames. ], giga_tot_loss[loss=0.4367, simple_loss=0.4593, pruned_loss=0.2071, over 5661116.06 frames. ], batch size: 213, lr: 2.06e-02, grad_scale: 2.0
+2023-02-28 18:26:24,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6028, 1.4625, 1.5050, 0.7988], device='cuda:0'), covar=tensor([0.0399, 0.0338, 0.0267, 0.0419], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0588, 0.0610, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 18:26:27,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29166.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:26:31,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.247e+02 1.805e+03 2.245e+03 2.882e+03 6.457e+03, threshold=4.491e+03, percent-clipped=13.0
+2023-02-28 18:26:31,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29169.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:26:33,265 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29172.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:26:59,339 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29198.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:26:59,376 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29198.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:27:00,374 INFO [train.py:968] (0/2) Epoch 1, batch 29200, giga_loss[loss=0.4628, simple_loss=0.454, pruned_loss=0.2358, over 23386.00 frames. ], tot_loss[loss=0.4362, simple_loss=0.4593, pruned_loss=0.2066, over 5648134.17 frames. ], libri_tot_loss[loss=0.4539, simple_loss=0.4683, pruned_loss=0.2198, over 5673435.85 frames. ], giga_tot_loss[loss=0.4348, simple_loss=0.4586, pruned_loss=0.2055, over 5650785.78 frames. ], batch size: 705, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:27:03,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29201.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:27:36,762 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29230.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:27:54,787 INFO [train.py:968] (0/2) Epoch 1, batch 29250, giga_loss[loss=0.4039, simple_loss=0.4396, pruned_loss=0.1841, over 28802.00 frames. ], tot_loss[loss=0.4347, simple_loss=0.4589, pruned_loss=0.2053, over 5635740.54 frames. ], libri_tot_loss[loss=0.4537, simple_loss=0.4681, pruned_loss=0.2197, over 5665163.59 frames. ], giga_tot_loss[loss=0.4334, simple_loss=0.4583, pruned_loss=0.2043, over 5645643.98 frames. ], batch size: 99, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:27:58,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=29254.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:27:58,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29254.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:28:04,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29260.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:28:13,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.673e+02 1.563e+03 2.046e+03 2.913e+03 6.289e+03, threshold=4.092e+03, percent-clipped=5.0
+2023-02-28 18:28:15,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3271, 1.9426, 1.4924, 1.2886], device='cuda:0'), covar=tensor([0.1400, 0.0908, 0.1080, 0.2041], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0486, 0.0387, 0.0479], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0012], device='cuda:0')
+2023-02-28 18:28:39,957 INFO [train.py:968] (0/2) Epoch 1, batch 29300, giga_loss[loss=0.3814, simple_loss=0.4211, pruned_loss=0.1709, over 29000.00 frames. ], tot_loss[loss=0.4326, simple_loss=0.4577, pruned_loss=0.2037, over 5643687.36 frames. ], libri_tot_loss[loss=0.4528, simple_loss=0.4676, pruned_loss=0.219, over 5670215.71 frames. ], giga_tot_loss[loss=0.4319, simple_loss=0.4575, pruned_loss=0.2032, over 5646628.56 frames. ], batch size: 106, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:28:43,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4104, 3.1765, 4.1780, 1.6390], device='cuda:0'), covar=tensor([0.0450, 0.0569, 0.0754, 0.1876], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0502, 0.0865, 0.0555], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-02-28 18:29:05,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4718, 1.3746, 1.4029, 1.5708], device='cuda:0'), covar=tensor([0.1475, 0.1685, 0.1198, 0.1987], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0799, 0.0862, 0.0917], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 18:29:25,806 INFO [train.py:968] (0/2) Epoch 1, batch 29350, giga_loss[loss=0.4219, simple_loss=0.4245, pruned_loss=0.2097, over 23531.00 frames. ], tot_loss[loss=0.4308, simple_loss=0.4563, pruned_loss=0.2027, over 5649691.89 frames. ], libri_tot_loss[loss=0.453, simple_loss=0.4678, pruned_loss=0.2191, over 5670166.85 frames. ], giga_tot_loss[loss=0.4296, simple_loss=0.4557, pruned_loss=0.2018, over 5651651.52 frames. ], batch size: 705, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:29:45,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.795e+03 2.364e+03 3.042e+03 6.822e+03, threshold=4.728e+03, percent-clipped=7.0
+2023-02-28 18:30:08,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=29395.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:11,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29397.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:13,234 INFO [train.py:968] (0/2) Epoch 1, batch 29400, giga_loss[loss=0.511, simple_loss=0.499, pruned_loss=0.2615, over 27862.00 frames. ], tot_loss[loss=0.431, simple_loss=0.4562, pruned_loss=0.2029, over 5652454.70 frames. ], libri_tot_loss[loss=0.4531, simple_loss=0.468, pruned_loss=0.2191, over 5672014.00 frames. ], giga_tot_loss[loss=0.4296, simple_loss=0.4554, pruned_loss=0.2019, over 5652214.12 frames. ], batch size: 412, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:30:15,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29400.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:18,094 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29403.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:21,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29406.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:44,394 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29429.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:49,976 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29435.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:59,143 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-02-28 18:31:07,666 INFO [train.py:968] (0/2) Epoch 1, batch 29450, giga_loss[loss=0.4059, simple_loss=0.4446, pruned_loss=0.1835, over 28543.00 frames. ], tot_loss[loss=0.4325, simple_loss=0.4573, pruned_loss=0.2038, over 5652432.79 frames. ], libri_tot_loss[loss=0.4527, simple_loss=0.4676, pruned_loss=0.2188, over 5674119.06 frames. ], giga_tot_loss[loss=0.4314, simple_loss=0.4567, pruned_loss=0.203, over 5650000.13 frames. ], batch size: 336, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:31:16,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3425, 1.7656, 1.4129, 1.1515], device='cuda:0'), covar=tensor([0.1120, 0.0872, 0.0987, 0.1690], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0477, 0.0382, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0009, 0.0012], device='cuda:0')
+2023-02-28 18:31:23,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.680e+03 2.073e+03 2.806e+03 5.581e+03, threshold=4.146e+03, percent-clipped=3.0
+2023-02-28 18:31:55,898 INFO [train.py:968] (0/2) Epoch 1, batch 29500, giga_loss[loss=0.4157, simple_loss=0.4416, pruned_loss=0.195, over 28508.00 frames. ], tot_loss[loss=0.4321, simple_loss=0.4564, pruned_loss=0.2039, over 5654390.82 frames. ], libri_tot_loss[loss=0.4522, simple_loss=0.4675, pruned_loss=0.2185, over 5679389.32 frames. ], giga_tot_loss[loss=0.4313, simple_loss=0.4558, pruned_loss=0.2033, over 5647436.36 frames. ], batch size: 307, lr: 2.04e-02, grad_scale: 4.0
+2023-02-28 18:32:44,253 INFO [train.py:968] (0/2) Epoch 1, batch 29550, giga_loss[loss=0.4123, simple_loss=0.4488, pruned_loss=0.1879, over 29015.00 frames. ], tot_loss[loss=0.4291, simple_loss=0.4542, pruned_loss=0.202, over 5668780.37 frames. ], libri_tot_loss[loss=0.4521, simple_loss=0.4675, pruned_loss=0.2183, over 5682870.54 frames. ], giga_tot_loss[loss=0.4282, simple_loss=0.4536, pruned_loss=0.2014, over 5659803.61 frames. ], batch size: 128, lr: 2.04e-02, grad_scale: 4.0
+2023-02-28 18:33:00,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3528, 1.1554, 1.0852, 1.4433], device='cuda:0'), covar=tensor([0.1644, 0.1811, 0.1451, 0.1980], device='cuda:0'), in_proj_covar=tensor([0.0892, 0.0787, 0.0864, 0.0920], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 18:33:01,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.673e+03 2.161e+03 2.657e+03 6.401e+03, threshold=4.321e+03, percent-clipped=10.0
+2023-02-28 18:33:06,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4590, 2.0008, 1.7647, 1.6419], device='cuda:0'), covar=tensor([0.1431, 0.0567, 0.0622, 0.1659], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0322, 0.0313, 0.0528], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0014, 0.0022], device='cuda:0')
+2023-02-28 18:33:32,534 INFO [train.py:968] (0/2) Epoch 1, batch 29600, giga_loss[loss=0.4021, simple_loss=0.4449, pruned_loss=0.1796, over 28941.00 frames. ], tot_loss[loss=0.4326, simple_loss=0.4565, pruned_loss=0.2044, over 5664335.18 frames. ], libri_tot_loss[loss=0.4526, simple_loss=0.468, pruned_loss=0.2186, over 5684732.03 frames. ], giga_tot_loss[loss=0.4309, simple_loss=0.4552, pruned_loss=0.2033, over 5655020.10 frames. ], batch size: 136, lr: 2.04e-02, grad_scale: 8.0
+2023-02-28 18:34:01,603 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=29629.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:34:01,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29629.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:34:22,468 INFO [train.py:968] (0/2) Epoch 1, batch 29650, libri_loss[loss=0.5041, simple_loss=0.4986, pruned_loss=0.2548, over 19293.00 frames. ], tot_loss[loss=0.4337, simple_loss=0.4574, pruned_loss=0.205, over 5649478.86 frames. ], libri_tot_loss[loss=0.4527, simple_loss=0.4682, pruned_loss=0.2187, over 5678636.27 frames. ], giga_tot_loss[loss=0.4319, simple_loss=0.4561, pruned_loss=0.2038, over 5648013.28 frames. ], batch size: 186, lr: 2.04e-02, grad_scale: 4.0
+2023-02-28 18:34:42,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.692e+03 1.971e+03 2.394e+03 5.553e+03, threshold=3.943e+03, percent-clipped=4.0
+2023-02-28 18:35:09,764 INFO [train.py:968] (0/2) Epoch 1, batch 29700, giga_loss[loss=0.3861, simple_loss=0.4266, pruned_loss=0.1728, over 28937.00 frames. ], tot_loss[loss=0.4309, simple_loss=0.4554, pruned_loss=0.2032, over 5649114.78 frames. ], libri_tot_loss[loss=0.4522, simple_loss=0.4678, pruned_loss=0.2183, over 5680583.74 frames. ], giga_tot_loss[loss=0.4295, simple_loss=0.4545, pruned_loss=0.2023, over 5645816.05 frames. ], batch size: 213, lr: 2.04e-02, grad_scale: 4.0
+2023-02-28 18:35:59,987 INFO [train.py:968] (0/2) Epoch 1, batch 29750, giga_loss[loss=0.3988, simple_loss=0.4368, pruned_loss=0.1804, over 28956.00 frames. ], tot_loss[loss=0.4319, simple_loss=0.4565, pruned_loss=0.2036, over 5651705.36 frames. ], libri_tot_loss[loss=0.4521, simple_loss=0.468, pruned_loss=0.2181, over 5683087.72 frames. ], giga_tot_loss[loss=0.4304, simple_loss=0.4553, pruned_loss=0.2027, over 5646316.32 frames. ], batch size: 227, lr: 2.04e-02, grad_scale: 4.0
+2023-02-28 18:36:14,476 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-02-28 18:36:18,806 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.417e+02 1.698e+03 2.210e+03 2.939e+03 7.355e+03, threshold=4.421e+03, percent-clipped=11.0
+2023-02-28 18:36:21,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29770.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:36:22,401 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29772.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:36:24,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29775.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:36:46,728 INFO [train.py:968] (0/2) Epoch 1, batch 29800, giga_loss[loss=0.4543, simple_loss=0.4741, pruned_loss=0.2173, over 28652.00 frames. ], tot_loss[loss=0.4326, simple_loss=0.4575, pruned_loss=0.2039, over 5664121.00 frames. ], libri_tot_loss[loss=0.4521, simple_loss=0.4679, pruned_loss=0.2182, over 5686887.00 frames. ], giga_tot_loss[loss=0.4307, simple_loss=0.4562, pruned_loss=0.2025, over 5655389.72 frames. ], batch size: 85, lr: 2.03e-02, grad_scale: 4.0
+2023-02-28 18:36:51,460 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29804.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:36:53,257 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 18:37:34,395 INFO [train.py:968] (0/2) Epoch 1, batch 29850, giga_loss[loss=0.3687, simple_loss=0.4199, pruned_loss=0.1588, over 29026.00 frames. ], tot_loss[loss=0.4319, simple_loss=0.4569, pruned_loss=0.2034, over 5664066.47 frames. ], libri_tot_loss[loss=0.4525, simple_loss=0.4681, pruned_loss=0.2184, over 5694230.80 frames. ], giga_tot_loss[loss=0.4293, simple_loss=0.4553, pruned_loss=0.2017, over 5649348.86 frames. ], batch size: 136, lr: 2.03e-02, grad_scale: 4.0
+2023-02-28 18:37:53,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.862e+03 2.657e+03 3.623e+03 7.374e+03, threshold=5.314e+03, percent-clipped=12.0
+2023-02-28 18:38:21,845 INFO [train.py:968] (0/2) Epoch 1, batch 29900, giga_loss[loss=0.4067, simple_loss=0.4389, pruned_loss=0.1872, over 28903.00 frames. ], tot_loss[loss=0.4308, simple_loss=0.4561, pruned_loss=0.2027, over 5678736.81 frames. ], libri_tot_loss[loss=0.4523, simple_loss=0.468, pruned_loss=0.2183, over 5697607.56 frames. ], giga_tot_loss[loss=0.4286, simple_loss=0.4549, pruned_loss=0.2012, over 5663841.56 frames. ], batch size: 186, lr: 2.03e-02, grad_scale: 4.0
+2023-02-28 18:38:36,790 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29913.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:38:39,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29916.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:39:03,847 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=29944.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:39:04,594 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29945.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:39:08,398 INFO [train.py:968] (0/2) Epoch 1, batch 29950, giga_loss[loss=0.3647, simple_loss=0.4041, pruned_loss=0.1626, over 28472.00 frames. ], tot_loss[loss=0.428, simple_loss=0.4538, pruned_loss=0.2011, over 5681087.55 frames. ], libri_tot_loss[loss=0.4526, simple_loss=0.4683, pruned_loss=0.2184, over 5702917.67 frames. ], giga_tot_loss[loss=0.4253, simple_loss=0.452, pruned_loss=0.1993, over 5663617.91 frames. ], batch size: 85, lr: 2.03e-02, grad_scale: 4.0
+2023-02-28 18:39:27,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.683e+02 1.770e+03 2.293e+03 2.942e+03 6.616e+03, threshold=4.587e+03, percent-clipped=2.0
+2023-02-28 18:39:31,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4474, 1.8388, 1.5798, 0.4036], device='cuda:0'), covar=tensor([0.1052, 0.0799, 0.1044, 0.1556], device='cuda:0'), in_proj_covar=tensor([0.0980, 0.0987, 0.0977, 0.0885], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 18:39:59,586 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-30000.pt
+2023-02-28 18:39:59,908 INFO [train.py:968] (0/2) Epoch 1, batch 30000, giga_loss[loss=0.5287, simple_loss=0.4971, pruned_loss=0.2802, over 26640.00 frames. ], tot_loss[loss=0.4229, simple_loss=0.4489, pruned_loss=0.1985, over 5669512.64 frames. ], libri_tot_loss[loss=0.4517, simple_loss=0.4676, pruned_loss=0.2179, over 5707284.62 frames. ], giga_tot_loss[loss=0.421, simple_loss=0.4478, pruned_loss=0.1971, over 5651023.13 frames. ], batch size: 555, lr: 2.03e-02, grad_scale: 8.0
+2023-02-28 18:39:59,915 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 18:40:08,175 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.3067, simple_loss=0.3955, pruned_loss=0.109, over 944034.00 frames. 
+2023-02-28 18:40:08,175 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 18:40:12,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30004.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:40:53,662 INFO [train.py:968] (0/2) Epoch 1, batch 30050, giga_loss[loss=0.3991, simple_loss=0.4362, pruned_loss=0.181, over 28882.00 frames. ], tot_loss[loss=0.4217, simple_loss=0.447, pruned_loss=0.1982, over 5672196.29 frames. ], libri_tot_loss[loss=0.452, simple_loss=0.4679, pruned_loss=0.218, over 5711010.52 frames. ], giga_tot_loss[loss=0.4193, simple_loss=0.4454, pruned_loss=0.1967, over 5653392.65 frames. ], batch size: 199, lr: 2.03e-02, grad_scale: 4.0
+2023-02-28 18:40:59,372 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=30055.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:41:12,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.595e+02 1.897e+03 2.460e+03 3.535e+03 8.312e+03, threshold=4.919e+03, percent-clipped=9.0
+2023-02-28 18:41:42,110 INFO [train.py:968] (0/2) Epoch 1, batch 30100, giga_loss[loss=0.3823, simple_loss=0.4213, pruned_loss=0.1716, over 28894.00 frames. ], tot_loss[loss=0.4205, simple_loss=0.4452, pruned_loss=0.1979, over 5661825.92 frames. ], libri_tot_loss[loss=0.4513, simple_loss=0.4673, pruned_loss=0.2177, over 5707044.67 frames. ], giga_tot_loss[loss=0.4184, simple_loss=0.4439, pruned_loss=0.1965, over 5648742.30 frames. ], batch size: 186, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:42:12,058 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0522, 1.8385, 1.1879, 0.8693], device='cuda:0'), covar=tensor([0.0449, 0.0392, 0.0364, 0.0458], device='cuda:0'), in_proj_covar=tensor([0.0831, 0.0599, 0.0623, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 18:42:32,634 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30147.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:42:34,343 INFO [train.py:968] (0/2) Epoch 1, batch 30150, libri_loss[loss=0.4415, simple_loss=0.457, pruned_loss=0.2131, over 29523.00 frames. ], tot_loss[loss=0.4158, simple_loss=0.4426, pruned_loss=0.1945, over 5656020.64 frames. ], libri_tot_loss[loss=0.4508, simple_loss=0.4669, pruned_loss=0.2174, over 5709878.62 frames. ], giga_tot_loss[loss=0.4141, simple_loss=0.4417, pruned_loss=0.1933, over 5642495.66 frames. ], batch size: 82, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:42:35,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30150.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:42:43,742 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.40 vs. limit=5.0
+2023-02-28 18:42:55,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.996e+02 1.629e+03 2.117e+03 3.111e+03 6.876e+03, threshold=4.235e+03, percent-clipped=5.0
+2023-02-28 18:42:58,264 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-02-28 18:43:05,686 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30179.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:43:21,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=30195.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:43:27,773 INFO [train.py:968] (0/2) Epoch 1, batch 30200, giga_loss[loss=0.3831, simple_loss=0.4248, pruned_loss=0.1707, over 28084.00 frames. ], tot_loss[loss=0.4085, simple_loss=0.4386, pruned_loss=0.1892, over 5650126.98 frames. ], libri_tot_loss[loss=0.4503, simple_loss=0.4664, pruned_loss=0.2171, over 5711008.27 frames. ], giga_tot_loss[loss=0.4066, simple_loss=0.4375, pruned_loss=0.1878, over 5636607.83 frames. ], batch size: 412, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:44:23,848 INFO [train.py:968] (0/2) Epoch 1, batch 30250, giga_loss[loss=0.3513, simple_loss=0.4085, pruned_loss=0.147, over 28926.00 frames. ], tot_loss[loss=0.4009, simple_loss=0.4342, pruned_loss=0.1837, over 5653101.71 frames. ], libri_tot_loss[loss=0.4501, simple_loss=0.4662, pruned_loss=0.2171, over 5709821.32 frames. ], giga_tot_loss[loss=0.3988, simple_loss=0.4332, pruned_loss=0.1822, over 5642740.40 frames. ], batch size: 213, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:44:44,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.341e+02 1.452e+03 1.763e+03 2.340e+03 8.106e+03, threshold=3.526e+03, percent-clipped=8.0
+2023-02-28 18:45:16,753 INFO [train.py:968] (0/2) Epoch 1, batch 30300, giga_loss[loss=0.374, simple_loss=0.4216, pruned_loss=0.1632, over 28945.00 frames. ], tot_loss[loss=0.3939, simple_loss=0.4297, pruned_loss=0.179, over 5651420.70 frames. ], libri_tot_loss[loss=0.4491, simple_loss=0.4652, pruned_loss=0.2165, over 5711098.74 frames. ], giga_tot_loss[loss=0.3922, simple_loss=0.4291, pruned_loss=0.1777, over 5641124.06 frames. ], batch size: 164, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:45:35,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30319.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:46:02,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8594, 2.3913, 1.8613, 1.5954], device='cuda:0'), covar=tensor([0.1106, 0.1154, 0.0898, 0.0585], device='cuda:0'), in_proj_covar=tensor([0.0833, 0.0853, 0.0720, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 18:46:06,784 INFO [train.py:968] (0/2) Epoch 1, batch 30350, giga_loss[loss=0.3267, simple_loss=0.3807, pruned_loss=0.1363, over 28910.00 frames. ], tot_loss[loss=0.3852, simple_loss=0.424, pruned_loss=0.1733, over 5655053.52 frames. ], libri_tot_loss[loss=0.4483, simple_loss=0.4646, pruned_loss=0.216, over 5711594.27 frames. ], giga_tot_loss[loss=0.3835, simple_loss=0.4233, pruned_loss=0.1719, over 5645432.56 frames. ], batch size: 213, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:46:19,524 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-02-28 18:46:24,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.826e+02 1.556e+03 2.088e+03 2.663e+03 7.245e+03, threshold=4.176e+03, percent-clipped=10.0
+2023-02-28 18:46:57,087 INFO [train.py:968] (0/2) Epoch 1, batch 30400, giga_loss[loss=0.3132, simple_loss=0.3924, pruned_loss=0.117, over 28905.00 frames. ], tot_loss[loss=0.3768, simple_loss=0.4203, pruned_loss=0.1666, over 5661231.99 frames. ], libri_tot_loss[loss=0.4473, simple_loss=0.4637, pruned_loss=0.2154, over 5702136.13 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.4199, pruned_loss=0.1653, over 5660692.71 frames. ], batch size: 112, lr: 2.01e-02, grad_scale: 8.0
+2023-02-28 18:47:18,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1613, 1.1814, 1.1957, 0.6450], device='cuda:0'), covar=tensor([0.0288, 0.0257, 0.0167, 0.0305], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0558, 0.0583, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 18:47:28,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30430.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:47:49,248 INFO [train.py:968] (0/2) Epoch 1, batch 30450, giga_loss[loss=0.3588, simple_loss=0.4131, pruned_loss=0.1523, over 28308.00 frames. ], tot_loss[loss=0.3766, simple_loss=0.4203, pruned_loss=0.1664, over 5667946.84 frames. ], libri_tot_loss[loss=0.446, simple_loss=0.4626, pruned_loss=0.2147, over 5707705.93 frames. ], giga_tot_loss[loss=0.3742, simple_loss=0.4196, pruned_loss=0.1644, over 5661303.70 frames. ], batch size: 369, lr: 2.01e-02, grad_scale: 4.0
+2023-02-28 18:48:02,002 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30462.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:48:06,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30465.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:48:12,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.317e+02 1.552e+03 1.907e+03 2.694e+03 7.893e+03, threshold=3.813e+03, percent-clipped=10.0
+2023-02-28 18:48:34,198 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30494.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:48:39,788 INFO [train.py:968] (0/2) Epoch 1, batch 30500, giga_loss[loss=0.3033, simple_loss=0.3704, pruned_loss=0.1181, over 28630.00 frames. ], tot_loss[loss=0.3755, simple_loss=0.4192, pruned_loss=0.1659, over 5672773.51 frames. ], libri_tot_loss[loss=0.4455, simple_loss=0.4618, pruned_loss=0.2146, over 5713328.19 frames. ], giga_tot_loss[loss=0.372, simple_loss=0.4181, pruned_loss=0.1629, over 5661244.49 frames. ], batch size: 66, lr: 2.01e-02, grad_scale: 4.0
+2023-02-28 18:48:59,314 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=30518.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:49:32,525 INFO [train.py:968] (0/2) Epoch 1, batch 30550, giga_loss[loss=0.3312, simple_loss=0.3949, pruned_loss=0.1338, over 28787.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.4137, pruned_loss=0.1607, over 5671833.63 frames. ], libri_tot_loss[loss=0.4449, simple_loss=0.4613, pruned_loss=0.2142, over 5715269.17 frames. ], giga_tot_loss[loss=0.3645, simple_loss=0.4128, pruned_loss=0.1582, over 5660492.10 frames. ], batch size: 243, lr: 2.01e-02, grad_scale: 2.0
+2023-02-28 18:49:50,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30570.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:49:53,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.308e+02 1.412e+03 1.800e+03 2.666e+03 8.430e+03, threshold=3.599e+03, percent-clipped=13.0
+2023-02-28 18:49:54,727 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30573.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:49:57,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3771, 1.7777, 1.4099, 1.2435], device='cuda:0'), covar=tensor([0.0871, 0.0852, 0.0821, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0467, 0.0390, 0.0484], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0012], device='cuda:0')
+2023-02-28 18:49:57,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30576.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:50:05,672 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-02-28 18:50:20,815 INFO [train.py:968] (0/2) Epoch 1, batch 30600, giga_loss[loss=0.3888, simple_loss=0.4093, pruned_loss=0.1842, over 26518.00 frames. ], tot_loss[loss=0.3652, simple_loss=0.4111, pruned_loss=0.1596, over 5667922.24 frames. ], libri_tot_loss[loss=0.4435, simple_loss=0.46, pruned_loss=0.2135, over 5721800.61 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.4098, pruned_loss=0.1562, over 5651059.39 frames. ], batch size: 555, lr: 2.01e-02, grad_scale: 2.0
+2023-02-28 18:50:22,953 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=30602.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:50:27,122 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30605.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:51:11,691 INFO [train.py:968] (0/2) Epoch 1, batch 30650, giga_loss[loss=0.3545, simple_loss=0.3934, pruned_loss=0.1578, over 26632.00 frames. ], tot_loss[loss=0.3653, simple_loss=0.4118, pruned_loss=0.1594, over 5674987.25 frames. ], libri_tot_loss[loss=0.4429, simple_loss=0.4595, pruned_loss=0.2132, over 5724598.42 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.4102, pruned_loss=0.1559, over 5658172.22 frames. ], batch size: 555, lr: 2.01e-02, grad_scale: 2.0
+2023-02-28 18:51:31,556 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.734e+02 1.626e+03 2.129e+03 2.988e+03 8.207e+03, threshold=4.258e+03, percent-clipped=19.0
+2023-02-28 18:51:59,928 INFO [train.py:968] (0/2) Epoch 1, batch 30700, giga_loss[loss=0.327, simple_loss=0.388, pruned_loss=0.133, over 28299.00 frames. ], tot_loss[loss=0.363, simple_loss=0.41, pruned_loss=0.158, over 5667662.92 frames. ], libri_tot_loss[loss=0.4422, simple_loss=0.4588, pruned_loss=0.2128, over 5723926.76 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4078, pruned_loss=0.1538, over 5653186.85 frames. ], batch size: 368, lr: 2.00e-02, grad_scale: 2.0
+2023-02-28 18:52:09,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7039, 1.7908, 1.6186, 1.8739], device='cuda:0'), covar=tensor([0.1425, 0.1376, 0.1126, 0.1682], device='cuda:0'), in_proj_covar=tensor([0.0877, 0.0758, 0.0851, 0.0914], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 18:52:10,523 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30713.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:52:13,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30716.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:52:44,737 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30745.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:52:48,740 INFO [train.py:968] (0/2) Epoch 1, batch 30750, giga_loss[loss=0.3563, simple_loss=0.4084, pruned_loss=0.1521, over 28899.00 frames. ], tot_loss[loss=0.3575, simple_loss=0.4061, pruned_loss=0.1544, over 5670127.32 frames. ], libri_tot_loss[loss=0.4414, simple_loss=0.458, pruned_loss=0.2124, over 5724695.49 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4038, pruned_loss=0.15, over 5656522.58 frames. ], batch size: 227, lr: 2.00e-02, grad_scale: 2.0
+2023-02-28 18:52:52,414 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=30753.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:52:57,616 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-02-28 18:53:01,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7007, 2.0429, 1.6616, 0.7809], device='cuda:0'), covar=tensor([0.0790, 0.0656, 0.0826, 0.1108], device='cuda:0'), in_proj_covar=tensor([0.0982, 0.1048, 0.0993, 0.0905], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 18:53:11,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.472e+02 1.398e+03 1.804e+03 2.523e+03 8.610e+03, threshold=3.608e+03, percent-clipped=4.0
+2023-02-28 18:53:39,574 INFO [train.py:968] (0/2) Epoch 1, batch 30800, giga_loss[loss=0.3301, simple_loss=0.381, pruned_loss=0.1396, over 27956.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.4019, pruned_loss=0.151, over 5678043.35 frames. ], libri_tot_loss[loss=0.4405, simple_loss=0.4572, pruned_loss=0.2119, over 5727692.08 frames. ], giga_tot_loss[loss=0.3463, simple_loss=0.3995, pruned_loss=0.1466, over 5663369.94 frames. ], batch size: 412, lr: 2.00e-02, grad_scale: 4.0
+2023-02-28 18:54:31,570 INFO [train.py:968] (0/2) Epoch 1, batch 30850, giga_loss[loss=0.3365, simple_loss=0.3913, pruned_loss=0.1408, over 28802.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.3992, pruned_loss=0.1497, over 5670658.31 frames. ], libri_tot_loss[loss=0.4398, simple_loss=0.4567, pruned_loss=0.2115, over 5722348.14 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3967, pruned_loss=0.1454, over 5662315.87 frames. ], batch size: 119, lr: 2.00e-02, grad_scale: 4.0
+2023-02-28 18:54:53,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.173e+02 1.602e+03 1.963e+03 2.668e+03 5.418e+03, threshold=3.925e+03, percent-clipped=9.0
+2023-02-28 18:55:00,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7025, 2.1769, 1.6863, 1.6293], device='cuda:0'), covar=tensor([0.1153, 0.1122, 0.0929, 0.0642], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0820, 0.0721, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 18:55:17,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30893.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:55:23,604 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-02-28 18:55:25,890 INFO [train.py:968] (0/2) Epoch 1, batch 30900, giga_loss[loss=0.3536, simple_loss=0.4059, pruned_loss=0.1506, over 28014.00 frames. ], tot_loss[loss=0.3491, simple_loss=0.3983, pruned_loss=0.15, over 5656278.82 frames. ], libri_tot_loss[loss=0.4394, simple_loss=0.4562, pruned_loss=0.2113, over 5714758.20 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.396, pruned_loss=0.146, over 5655051.11 frames. ], batch size: 412, lr: 2.00e-02, grad_scale: 4.0
+2023-02-28 18:56:22,315 INFO [train.py:968] (0/2) Epoch 1, batch 30950, giga_loss[loss=0.3815, simple_loss=0.4215, pruned_loss=0.1708, over 28855.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.3987, pruned_loss=0.1502, over 5635278.66 frames. ], libri_tot_loss[loss=0.4398, simple_loss=0.4564, pruned_loss=0.2116, over 5703029.44 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.3963, pruned_loss=0.1463, over 5644254.31 frames. ], batch size: 99, lr: 2.00e-02, grad_scale: 4.0
+2023-02-28 18:56:49,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.346e+02 1.563e+03 2.095e+03 3.056e+03 7.060e+03, threshold=4.189e+03, percent-clipped=10.0
+2023-02-28 18:56:55,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30977.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:57:14,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6601, 1.9003, 1.6593, 1.4665], device='cuda:0'), covar=tensor([0.0905, 0.0716, 0.0902, 0.1429], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0464, 0.0387, 0.0489], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0013], device='cuda:0')
+2023-02-28 18:57:22,136 INFO [train.py:968] (0/2) Epoch 1, batch 31000, giga_loss[loss=0.3559, simple_loss=0.4124, pruned_loss=0.1497, over 28693.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.4033, pruned_loss=0.1526, over 5636552.95 frames. ], libri_tot_loss[loss=0.439, simple_loss=0.4558, pruned_loss=0.2112, over 5706084.45 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.4012, pruned_loss=0.1491, over 5640179.09 frames. ], batch size: 262, lr: 2.00e-02, grad_scale: 2.0
+2023-02-28 18:58:04,810 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31036.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:58:06,924 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31039.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:58:16,351 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31048.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:58:17,715 INFO [train.py:968] (0/2) Epoch 1, batch 31050, giga_loss[loss=0.3038, simple_loss=0.3762, pruned_loss=0.1157, over 28811.00 frames. ], tot_loss[loss=0.3548, simple_loss=0.4038, pruned_loss=0.1529, over 5636367.44 frames. ], libri_tot_loss[loss=0.4383, simple_loss=0.4552, pruned_loss=0.2107, over 5713461.12 frames. ], giga_tot_loss[loss=0.3486, simple_loss=0.4007, pruned_loss=0.1483, over 5629979.95 frames. ], batch size: 119, lr: 1.99e-02, grad_scale: 2.0
+2023-02-28 18:58:32,463 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-02-28 18:58:38,534 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8450, 1.7487, 3.8474, 2.9839], device='cuda:0'), covar=tensor([0.1292, 0.1146, 0.0268, 0.1056], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0470, 0.0603, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 18:58:45,764 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31068.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:58:52,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.060e+02 1.678e+03 2.162e+03 3.056e+03 7.968e+03, threshold=4.325e+03, percent-clipped=6.0
+2023-02-28 18:59:27,822 INFO [train.py:968] (0/2) Epoch 1, batch 31100, giga_loss[loss=0.3132, simple_loss=0.3772, pruned_loss=0.1246, over 28640.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4026, pruned_loss=0.152, over 5626121.60 frames. ], libri_tot_loss[loss=0.4379, simple_loss=0.4548, pruned_loss=0.2105, over 5696448.71 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.3998, pruned_loss=0.1476, over 5634014.31 frames. ], batch size: 307, lr: 1.99e-02, grad_scale: 2.0
+2023-02-28 18:59:55,123 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31120.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:59:58,481 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31123.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:00:05,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31128.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:00:16,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3675, 3.2455, 4.0821, 1.7995], device='cuda:0'), covar=tensor([0.0456, 0.0531, 0.0781, 0.1892], device='cuda:0'), in_proj_covar=tensor([0.0735, 0.0492, 0.0782, 0.0521], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 19:00:19,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5109, 2.1105, 1.8875, 1.9614], device='cuda:0'), covar=tensor([0.0637, 0.1591, 0.1062, 0.1093], device='cuda:0'), in_proj_covar=tensor([0.0530, 0.0864, 0.0659, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 19:00:33,753 INFO [train.py:968] (0/2) Epoch 1, batch 31150, giga_loss[loss=0.2942, simple_loss=0.3686, pruned_loss=0.1099, over 28849.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3985, pruned_loss=0.1476, over 5625089.26 frames. ], libri_tot_loss[loss=0.438, simple_loss=0.4548, pruned_loss=0.2106, over 5687664.47 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3962, pruned_loss=0.144, over 5638646.00 frames. ], batch size: 112, lr: 1.99e-02, grad_scale: 2.0
+2023-02-28 19:00:37,823 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31152.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:01:07,812 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-02-28 19:01:07,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.809e+02 1.248e+03 1.576e+03 2.124e+03 5.048e+03, threshold=3.152e+03, percent-clipped=4.0
+2023-02-28 19:01:41,120 INFO [train.py:968] (0/2) Epoch 1, batch 31200, giga_loss[loss=0.3425, simple_loss=0.3962, pruned_loss=0.1443, over 28791.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3951, pruned_loss=0.1434, over 5628777.25 frames. ], libri_tot_loss[loss=0.4373, simple_loss=0.4542, pruned_loss=0.2102, over 5691204.61 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3929, pruned_loss=0.14, over 5635542.25 frames. ], batch size: 243, lr: 1.99e-02, grad_scale: 4.0
+2023-02-28 19:02:36,881 INFO [train.py:968] (0/2) Epoch 1, batch 31250, giga_loss[loss=0.3305, simple_loss=0.3886, pruned_loss=0.1362, over 28863.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3937, pruned_loss=0.1442, over 5624196.02 frames. ], libri_tot_loss[loss=0.4354, simple_loss=0.4527, pruned_loss=0.2091, over 5673024.62 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3908, pruned_loss=0.1396, over 5642359.30 frames. ], batch size: 174, lr: 1.99e-02, grad_scale: 4.0
+2023-02-28 19:03:02,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31271.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:03:03,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2568, 1.5444, 1.2743, 0.3012], device='cuda:0'), covar=tensor([0.0691, 0.0595, 0.0776, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0961, 0.1021, 0.0968, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 19:03:05,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.026e+02 1.523e+03 1.957e+03 2.548e+03 5.391e+03, threshold=3.914e+03, percent-clipped=8.0
+2023-02-28 19:03:05,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31274.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:03:33,927 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31296.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:03:37,944 INFO [train.py:968] (0/2) Epoch 1, batch 31300, giga_loss[loss=0.3051, simple_loss=0.3608, pruned_loss=0.1247, over 28718.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3935, pruned_loss=0.1449, over 5643174.02 frames. ], libri_tot_loss[loss=0.4351, simple_loss=0.4524, pruned_loss=0.2089, over 5677559.15 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3899, pruned_loss=0.1396, over 5652666.53 frames. ], batch size: 99, lr: 1.99e-02, grad_scale: 4.0
+2023-02-28 19:03:40,673 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31303.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:03:57,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31318.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:04:06,718 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31327.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:04:14,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2534, 1.2304, 1.1879, 1.3339], device='cuda:0'), covar=tensor([0.1664, 0.1689, 0.1382, 0.1963], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0759, 0.0854, 0.0903], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 19:04:24,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4333, 3.0522, 2.6026, 1.9983], device='cuda:0'), covar=tensor([0.0853, 0.0691, 0.0608, 0.0427], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0798, 0.0702, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 19:04:29,918 INFO [train.py:968] (0/2) Epoch 1, batch 31350, giga_loss[loss=0.3131, simple_loss=0.374, pruned_loss=0.1261, over 28943.00 frames. ], tot_loss[loss=0.3457, simple_loss=0.3953, pruned_loss=0.148, over 5646322.04 frames. ], libri_tot_loss[loss=0.436, simple_loss=0.4526, pruned_loss=0.2098, over 5670525.61 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3898, pruned_loss=0.1404, over 5659169.72 frames. ], batch size: 227, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:05:00,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.065e+02 1.808e+03 2.422e+03 3.518e+03 2.593e+04, threshold=4.844e+03, percent-clipped=20.0
+2023-02-28 19:05:27,671 INFO [train.py:968] (0/2) Epoch 1, batch 31400, giga_loss[loss=0.3567, simple_loss=0.4184, pruned_loss=0.1475, over 28922.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3964, pruned_loss=0.1476, over 5650753.51 frames. ], libri_tot_loss[loss=0.4348, simple_loss=0.4516, pruned_loss=0.209, over 5672650.86 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3916, pruned_loss=0.1407, over 5658619.49 frames. ], batch size: 199, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:05:32,829 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31404.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 19:05:57,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31423.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:06:31,640 INFO [train.py:968] (0/2) Epoch 1, batch 31450, giga_loss[loss=0.3164, simple_loss=0.3804, pruned_loss=0.1262, over 28926.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3977, pruned_loss=0.1479, over 5647582.21 frames. ], libri_tot_loss[loss=0.4339, simple_loss=0.4508, pruned_loss=0.2085, over 5678638.83 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.393, pruned_loss=0.141, over 5647734.75 frames. ], batch size: 164, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:06:58,537 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-02-28 19:07:01,497 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.137e+02 1.671e+03 2.284e+03 3.072e+03 9.041e+03, threshold=4.568e+03, percent-clipped=4.0
+2023-02-28 19:07:35,073 INFO [train.py:968] (0/2) Epoch 1, batch 31500, giga_loss[loss=0.2942, simple_loss=0.3534, pruned_loss=0.1175, over 28999.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3925, pruned_loss=0.1438, over 5669367.15 frames. ], libri_tot_loss[loss=0.4325, simple_loss=0.4496, pruned_loss=0.2077, over 5685328.21 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3883, pruned_loss=0.1374, over 5663007.16 frames. ], batch size: 186, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:08:41,011 INFO [train.py:968] (0/2) Epoch 1, batch 31550, giga_loss[loss=0.3128, simple_loss=0.3785, pruned_loss=0.1236, over 29027.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3953, pruned_loss=0.1463, over 5673191.46 frames. ], libri_tot_loss[loss=0.432, simple_loss=0.4492, pruned_loss=0.2075, over 5684415.25 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3905, pruned_loss=0.1394, over 5668399.76 frames. ], batch size: 136, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:09:04,023 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31566.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:09:08,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31569.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:09:09,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1476, 1.1544, 0.9522, 0.9727], device='cuda:0'), covar=tensor([0.0742, 0.0645, 0.1210, 0.0850], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0575, 0.0610, 0.0552], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 19:09:15,071 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.080e+02 1.483e+03 1.901e+03 2.514e+03 7.619e+03, threshold=3.801e+03, percent-clipped=5.0
+2023-02-28 19:09:47,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31598.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:09:49,129 INFO [train.py:968] (0/2) Epoch 1, batch 31600, giga_loss[loss=0.3186, simple_loss=0.4046, pruned_loss=0.1163, over 28679.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3982, pruned_loss=0.1464, over 5667999.12 frames. ], libri_tot_loss[loss=0.4319, simple_loss=0.4491, pruned_loss=0.2074, over 5683560.17 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3943, pruned_loss=0.1408, over 5664922.96 frames. ], batch size: 242, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:10:56,090 INFO [train.py:968] (0/2) Epoch 1, batch 31650, giga_loss[loss=0.3252, simple_loss=0.399, pruned_loss=0.1257, over 28725.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3994, pruned_loss=0.1442, over 5665678.13 frames. ], libri_tot_loss[loss=0.432, simple_loss=0.4492, pruned_loss=0.2074, over 5687868.28 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3954, pruned_loss=0.1387, over 5659440.42 frames. ], batch size: 99, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:11:26,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31671.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:11:31,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.649e+02 1.449e+03 2.034e+03 2.852e+03 2.955e+04, threshold=4.068e+03, percent-clipped=13.0
+2023-02-28 19:11:51,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31693.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:12:00,531 INFO [train.py:968] (0/2) Epoch 1, batch 31700, giga_loss[loss=0.3306, simple_loss=0.399, pruned_loss=0.1311, over 28815.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3987, pruned_loss=0.1416, over 5657597.00 frames. ], libri_tot_loss[loss=0.4316, simple_loss=0.4488, pruned_loss=0.2072, over 5681066.71 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.395, pruned_loss=0.1364, over 5658156.03 frames. ], batch size: 99, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:12:02,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31702.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:12:51,291 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-02-28 19:13:01,571 INFO [train.py:968] (0/2) Epoch 1, batch 31750, giga_loss[loss=0.3438, simple_loss=0.4012, pruned_loss=0.1432, over 28948.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3962, pruned_loss=0.1387, over 5664165.65 frames. ], libri_tot_loss[loss=0.4318, simple_loss=0.4487, pruned_loss=0.2074, over 5680729.46 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3927, pruned_loss=0.1335, over 5664530.77 frames. ], batch size: 284, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:13:36,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.367e+02 1.494e+03 1.815e+03 2.387e+03 8.831e+03, threshold=3.630e+03, percent-clipped=4.0
+2023-02-28 19:13:40,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31779.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:13:56,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4254, 1.8108, 1.5036, 0.2394], device='cuda:0'), covar=tensor([0.0655, 0.0571, 0.0776, 0.0996], device='cuda:0'), in_proj_covar=tensor([0.0994, 0.1055, 0.0989, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 19:14:03,393 INFO [train.py:968] (0/2) Epoch 1, batch 31800, giga_loss[loss=0.3238, simple_loss=0.3818, pruned_loss=0.1329, over 28944.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3982, pruned_loss=0.1417, over 5679862.46 frames. ], libri_tot_loss[loss=0.4314, simple_loss=0.4483, pruned_loss=0.2073, over 5687604.30 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.394, pruned_loss=0.1355, over 5673493.36 frames. ], batch size: 136, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:14:07,788 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31802.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:14:23,455 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31814.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:14:29,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31817.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:14:54,564 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31836.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:00,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31839.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:10,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31845.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:11,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31846.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:14,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31848.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:16,416 INFO [train.py:968] (0/2) Epoch 1, batch 31850, giga_loss[loss=0.3208, simple_loss=0.3824, pruned_loss=0.1296, over 28683.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3981, pruned_loss=0.1432, over 5671470.45 frames. ], libri_tot_loss[loss=0.4312, simple_loss=0.4482, pruned_loss=0.2071, over 5681173.72 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3941, pruned_loss=0.1374, over 5671862.01 frames. ], batch size: 262, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:15:46,363 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31868.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:57,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.897e+02 1.440e+03 1.858e+03 2.434e+03 7.644e+03, threshold=3.716e+03, percent-clipped=11.0
+2023-02-28 19:15:59,242 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31877.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:16:16,236 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31884.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 19:16:38,956 INFO [train.py:968] (0/2) Epoch 1, batch 31900, giga_loss[loss=0.3024, simple_loss=0.3705, pruned_loss=0.1172, over 29140.00 frames. ], tot_loss[loss=0.3457, simple_loss=0.4001, pruned_loss=0.1456, over 5676963.48 frames. ], libri_tot_loss[loss=0.4308, simple_loss=0.4478, pruned_loss=0.2069, over 5685505.20 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3964, pruned_loss=0.1401, over 5673319.67 frames. ], batch size: 120, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:17:10,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31922.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 19:17:17,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31925.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 19:17:48,392 INFO [train.py:968] (0/2) Epoch 1, batch 31950, giga_loss[loss=0.3085, simple_loss=0.3744, pruned_loss=0.1213, over 28864.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3953, pruned_loss=0.1427, over 5667955.60 frames. ], libri_tot_loss[loss=0.4302, simple_loss=0.4474, pruned_loss=0.2065, over 5679970.95 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3914, pruned_loss=0.1369, over 5670350.68 frames. ], batch size: 164, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:17:55,196 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31954.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:18:21,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.701e+02 1.479e+03 1.956e+03 2.640e+03 9.727e+03, threshold=3.911e+03, percent-clipped=9.0
+2023-02-28 19:18:55,784 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-32000.pt
+2023-02-28 19:18:56,103 INFO [train.py:968] (0/2) Epoch 1, batch 32000, giga_loss[loss=0.2957, simple_loss=0.3665, pruned_loss=0.1125, over 28867.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3932, pruned_loss=0.141, over 5661820.02 frames. ], libri_tot_loss[loss=0.4293, simple_loss=0.4468, pruned_loss=0.2059, over 5673331.35 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3899, pruned_loss=0.1359, over 5669116.95 frames. ], batch size: 227, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:19:01,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4772, 1.7677, 1.4826, 1.3245], device='cuda:0'), covar=tensor([0.1134, 0.0861, 0.1066, 0.1685], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0459, 0.0383, 0.0485], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0012, 0.0010, 0.0013], device='cuda:0')
+2023-02-28 19:19:30,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7523, 2.1330, 1.6434, 1.5790], device='cuda:0'), covar=tensor([0.1153, 0.1051, 0.0928, 0.0593], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0831, 0.0709, 0.0623], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 19:19:58,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4420, 1.9946, 1.5853, 0.3506], device='cuda:0'), covar=tensor([0.1028, 0.0853, 0.1214, 0.1661], device='cuda:0'), in_proj_covar=tensor([0.0976, 0.1055, 0.1002, 0.0908], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 19:19:59,199 INFO [train.py:968] (0/2) Epoch 1, batch 32050, giga_loss[loss=0.3236, simple_loss=0.3794, pruned_loss=0.134, over 28943.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3929, pruned_loss=0.1423, over 5678795.42 frames. ], libri_tot_loss[loss=0.4282, simple_loss=0.4458, pruned_loss=0.2053, over 5679877.74 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3893, pruned_loss=0.1366, over 5678429.40 frames. ], batch size: 106, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:20:34,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.986e+02 1.538e+03 1.977e+03 2.767e+03 7.446e+03, threshold=3.953e+03, percent-clipped=11.0
+2023-02-28 19:20:46,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7434, 1.5012, 1.3522, 1.4388], device='cuda:0'), covar=tensor([0.0640, 0.1261, 0.0908, 0.0978], device='cuda:0'), in_proj_covar=tensor([0.0513, 0.0861, 0.0648, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 19:20:50,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.90 vs. limit=5.0
+2023-02-28 19:21:03,731 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32097.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:21:05,556 INFO [train.py:968] (0/2) Epoch 1, batch 32100, libri_loss[loss=0.3884, simple_loss=0.4006, pruned_loss=0.1881, over 29575.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3969, pruned_loss=0.1441, over 5678070.05 frames. ], libri_tot_loss[loss=0.4276, simple_loss=0.4453, pruned_loss=0.2049, over 5683175.27 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3933, pruned_loss=0.1385, over 5674643.47 frames. ], batch size: 76, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:21:08,813 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-02-28 19:21:32,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3163, 1.6238, 1.4195, 1.1869], device='cuda:0'), covar=tensor([0.1060, 0.0887, 0.0942, 0.1513], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0451, 0.0374, 0.0478], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0012], device='cuda:0')
+2023-02-28 19:22:05,568 INFO [train.py:968] (0/2) Epoch 1, batch 32150, giga_loss[loss=0.321, simple_loss=0.3778, pruned_loss=0.1321, over 28888.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3976, pruned_loss=0.1455, over 5690508.41 frames. ], libri_tot_loss[loss=0.427, simple_loss=0.445, pruned_loss=0.2045, over 5686937.79 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3937, pruned_loss=0.1397, over 5684404.46 frames. ], batch size: 164, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:22:39,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.734e+03 2.123e+03 2.961e+03 7.751e+03, threshold=4.246e+03, percent-clipped=11.0
+2023-02-28 19:22:40,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32177.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:22:43,759 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32181.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:23:00,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3081, 1.3983, 1.1379, 1.2189], device='cuda:0'), covar=tensor([0.0703, 0.0812, 0.1116, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0857, 0.0635, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 19:23:07,360 INFO [train.py:968] (0/2) Epoch 1, batch 32200, giga_loss[loss=0.3201, simple_loss=0.3814, pruned_loss=0.1294, over 28873.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.3972, pruned_loss=0.1465, over 5694494.14 frames. ], libri_tot_loss[loss=0.4256, simple_loss=0.4439, pruned_loss=0.2036, over 5693972.65 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3933, pruned_loss=0.1405, over 5683305.74 frames. ], batch size: 106, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:23:26,177 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32216.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:24:06,996 INFO [train.py:968] (0/2) Epoch 1, batch 32250, giga_loss[loss=0.3477, simple_loss=0.4058, pruned_loss=0.1448, over 28675.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3981, pruned_loss=0.1478, over 5691208.87 frames. ], libri_tot_loss[loss=0.4254, simple_loss=0.4439, pruned_loss=0.2035, over 5694797.68 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3937, pruned_loss=0.1415, over 5680807.71 frames. ], batch size: 262, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:24:20,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32259.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:24:39,437 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32271.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:24:44,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.104e+02 1.291e+03 1.655e+03 1.946e+03 5.024e+03, threshold=3.310e+03, percent-clipped=3.0
+2023-02-28 19:25:20,107 INFO [train.py:968] (0/2) Epoch 1, batch 32300, libri_loss[loss=0.3946, simple_loss=0.4142, pruned_loss=0.1875, over 29563.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.3978, pruned_loss=0.1465, over 5684716.16 frames. ], libri_tot_loss[loss=0.4248, simple_loss=0.4433, pruned_loss=0.2032, over 5697551.24 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3939, pruned_loss=0.1407, over 5673736.55 frames. ], batch size: 77, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:25:29,308 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32308.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:25:48,620 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32320.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:25:53,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32323.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:26:28,225 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32343.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:26:37,416 INFO [train.py:968] (0/2) Epoch 1, batch 32350, libri_loss[loss=0.477, simple_loss=0.4769, pruned_loss=0.2385, over 29260.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.3998, pruned_loss=0.1472, over 5678314.00 frames. ], libri_tot_loss[loss=0.4245, simple_loss=0.443, pruned_loss=0.203, over 5698132.50 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3959, pruned_loss=0.1414, over 5668433.29 frames. ], batch size: 94, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:26:39,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32352.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:27:17,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.670e+02 1.699e+03 2.306e+03 3.260e+03 6.517e+03, threshold=4.611e+03, percent-clipped=22.0
+2023-02-28 19:27:29,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3840, 1.7106, 1.4265, 1.3375], device='cuda:0'), covar=tensor([0.1268, 0.0534, 0.0670, 0.1668], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0306, 0.0304, 0.0514], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0015, 0.0024], device='cuda:0')
+2023-02-28 19:27:33,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7412, 2.6668, 3.4825, 1.7704], device='cuda:0'), covar=tensor([0.0589, 0.0569, 0.0860, 0.1676], device='cuda:0'), in_proj_covar=tensor([0.0728, 0.0485, 0.0779, 0.0524], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 19:27:53,172 INFO [train.py:968] (0/2) Epoch 1, batch 32400, giga_loss[loss=0.2881, simple_loss=0.3559, pruned_loss=0.1102, over 29049.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3982, pruned_loss=0.1464, over 5674689.02 frames. ], libri_tot_loss[loss=0.4234, simple_loss=0.4422, pruned_loss=0.2023, over 5700654.11 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3949, pruned_loss=0.1411, over 5664039.38 frames. ], batch size: 136, lr: 1.95e-02, grad_scale: 8.0
+2023-02-28 19:27:55,807 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32402.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 19:27:58,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32405.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 19:28:38,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32434.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:28:57,867 INFO [train.py:968] (0/2) Epoch 1, batch 32450, giga_loss[loss=0.2823, simple_loss=0.3481, pruned_loss=0.1082, over 29023.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3929, pruned_loss=0.1443, over 5682954.27 frames. ], libri_tot_loss[loss=0.4231, simple_loss=0.442, pruned_loss=0.2021, over 5705017.21 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.389, pruned_loss=0.1386, over 5669768.23 frames. ], batch size: 155, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:29:28,574 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32472.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:29:33,187 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.043e+02 1.522e+03 2.005e+03 2.699e+03 4.778e+03, threshold=4.011e+03, percent-clipped=1.0
+2023-02-28 19:29:36,100 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 19:30:06,128 INFO [train.py:968] (0/2) Epoch 1, batch 32500, giga_loss[loss=0.2778, simple_loss=0.3446, pruned_loss=0.1055, over 29059.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3854, pruned_loss=0.1397, over 5686754.36 frames. ], libri_tot_loss[loss=0.4227, simple_loss=0.4416, pruned_loss=0.2019, over 5708228.34 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3817, pruned_loss=0.1343, over 5673199.84 frames. ], batch size: 120, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:31:11,964 INFO [train.py:968] (0/2) Epoch 1, batch 32550, giga_loss[loss=0.3347, simple_loss=0.3924, pruned_loss=0.1385, over 29017.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3858, pruned_loss=0.1401, over 5669340.81 frames. ], libri_tot_loss[loss=0.4228, simple_loss=0.4416, pruned_loss=0.202, over 5700167.87 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3824, pruned_loss=0.1352, over 5665975.63 frames. ], batch size: 285, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:31:19,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32556.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:31:42,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.402e+02 1.503e+03 1.919e+03 2.380e+03 9.721e+03, threshold=3.838e+03, percent-clipped=8.0
+2023-02-28 19:31:59,248 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32591.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:32:10,000 INFO [train.py:968] (0/2) Epoch 1, batch 32600, libri_loss[loss=0.4509, simple_loss=0.4501, pruned_loss=0.2259, over 29539.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3887, pruned_loss=0.1423, over 5679873.63 frames. ], libri_tot_loss[loss=0.4217, simple_loss=0.4407, pruned_loss=0.2014, over 5705442.96 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.385, pruned_loss=0.137, over 5671323.60 frames. ], batch size: 81, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:32:27,583 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32615.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:32:31,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32618.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:32:37,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2871, 1.7455, 1.5371, 1.3679], device='cuda:0'), covar=tensor([0.1415, 0.0562, 0.0685, 0.1728], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0300, 0.0301, 0.0503], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0016, 0.0015, 0.0023], device='cuda:0')
+2023-02-28 19:33:08,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32646.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:33:11,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32647.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:33:14,354 INFO [train.py:968] (0/2) Epoch 1, batch 32650, giga_loss[loss=0.2898, simple_loss=0.3607, pruned_loss=0.1094, over 28917.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3843, pruned_loss=0.1379, over 5670619.95 frames. ], libri_tot_loss[loss=0.4217, simple_loss=0.4407, pruned_loss=0.2014, over 5705442.96 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3815, pruned_loss=0.1337, over 5663965.36 frames. ], batch size: 213, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:33:48,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.041e+02 1.591e+03 2.005e+03 3.001e+03 6.488e+03, threshold=4.010e+03, percent-clipped=9.0
+2023-02-28 19:33:55,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32683.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:34:18,041 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32699.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:34:18,425 INFO [train.py:968] (0/2) Epoch 1, batch 32700, giga_loss[loss=0.3224, simple_loss=0.3732, pruned_loss=0.1358, over 28042.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3834, pruned_loss=0.1372, over 5667798.24 frames. ], libri_tot_loss[loss=0.4206, simple_loss=0.4398, pruned_loss=0.2006, over 5707073.48 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3804, pruned_loss=0.1328, over 5660083.53 frames. ], batch size: 412, lr: 1.94e-02, grad_scale: 4.0
+2023-02-28 19:34:20,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32702.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:34:43,585 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32718.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:35:01,015 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32731.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:35:05,160 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32734.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:35:08,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32737.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:35:23,864 INFO [train.py:968] (0/2) Epoch 1, batch 32750, giga_loss[loss=0.2963, simple_loss=0.3683, pruned_loss=0.1122, over 28994.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3825, pruned_loss=0.1373, over 5665293.16 frames. ], libri_tot_loss[loss=0.4195, simple_loss=0.4391, pruned_loss=0.2, over 5700284.70 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3787, pruned_loss=0.1321, over 5663332.43 frames. ], batch size: 199, lr: 1.94e-02, grad_scale: 4.0
+2023-02-28 19:35:44,211 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32766.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:35:57,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.316e+02 1.642e+03 2.183e+03 2.808e+03 7.631e+03, threshold=4.366e+03, percent-clipped=7.0
+2023-02-28 19:36:15,466 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32789.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:36:17,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32792.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:36:27,090 INFO [train.py:968] (0/2) Epoch 1, batch 32800, giga_loss[loss=0.3071, simple_loss=0.3796, pruned_loss=0.1174, over 29032.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3854, pruned_loss=0.1389, over 5661275.10 frames. ], libri_tot_loss[loss=0.4199, simple_loss=0.4392, pruned_loss=0.2003, over 5685014.93 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3808, pruned_loss=0.1328, over 5673132.08 frames. ], batch size: 136, lr: 1.94e-02, grad_scale: 8.0
+2023-02-28 19:36:52,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32821.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:36:58,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32826.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:37:03,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32829.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:37:31,543 INFO [train.py:968] (0/2) Epoch 1, batch 32850, giga_loss[loss=0.3137, simple_loss=0.3768, pruned_loss=0.1253, over 28799.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3867, pruned_loss=0.1403, over 5650705.37 frames. ], libri_tot_loss[loss=0.4204, simple_loss=0.4395, pruned_loss=0.2006, over 5668539.17 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3814, pruned_loss=0.1335, over 5673461.14 frames. ], batch size: 243, lr: 1.94e-02, grad_scale: 8.0
+2023-02-28 19:37:42,272 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32858.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:37:46,872 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32861.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:37:49,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4972, 1.5220, 1.0657, 1.2533], device='cuda:0'), covar=tensor([0.0544, 0.0545, 0.0991, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0560, 0.0576, 0.0534], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 19:37:49,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32864.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:38:08,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.937e+02 1.510e+03 1.883e+03 2.614e+03 5.756e+03, threshold=3.765e+03, percent-clipped=3.0
+2023-02-28 19:38:30,511 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32893.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:38:36,663 INFO [train.py:968] (0/2) Epoch 1, batch 32900, giga_loss[loss=0.3309, simple_loss=0.3851, pruned_loss=0.1384, over 28935.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3883, pruned_loss=0.142, over 5660844.88 frames. ], libri_tot_loss[loss=0.421, simple_loss=0.4399, pruned_loss=0.2011, over 5671149.90 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3831, pruned_loss=0.1354, over 5676185.69 frames. ], batch size: 112, lr: 1.94e-02, grad_scale: 4.0
+2023-02-28 19:39:38,862 INFO [train.py:968] (0/2) Epoch 1, batch 32950, giga_loss[loss=0.3295, simple_loss=0.374, pruned_loss=0.1425, over 26872.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3867, pruned_loss=0.1402, over 5658099.31 frames. ], libri_tot_loss[loss=0.4204, simple_loss=0.4393, pruned_loss=0.2007, over 5671966.56 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3818, pruned_loss=0.1339, over 5669435.44 frames. ], batch size: 555, lr: 1.94e-02, grad_scale: 4.0
+2023-02-28 19:40:10,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.590e+02 1.391e+03 1.904e+03 2.891e+03 5.866e+03, threshold=3.809e+03, percent-clipped=11.0
+2023-02-28 19:40:34,624 INFO [train.py:968] (0/2) Epoch 1, batch 33000, giga_loss[loss=0.3331, simple_loss=0.3965, pruned_loss=0.1348, over 28544.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3889, pruned_loss=0.1404, over 5656401.77 frames. ], libri_tot_loss[loss=0.4201, simple_loss=0.4391, pruned_loss=0.2006, over 5675497.88 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3832, pruned_loss=0.1332, over 5661361.92 frames. ], batch size: 85, lr: 1.94e-02, grad_scale: 4.0
+2023-02-28 19:40:34,629 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 19:40:40,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3302, 2.0249, 1.7410, 1.5873], device='cuda:0'), covar=tensor([0.1444, 0.1354, 0.1181, 0.0745], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0827, 0.0718, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 19:40:42,967 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.2718, simple_loss=0.3588, pruned_loss=0.0924, over 944034.00 frames. 
+2023-02-28 19:40:42,968 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 19:41:44,283 INFO [train.py:968] (0/2) Epoch 1, batch 33050, giga_loss[loss=0.3439, simple_loss=0.3972, pruned_loss=0.1453, over 28167.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.392, pruned_loss=0.1417, over 5661058.40 frames. ], libri_tot_loss[loss=0.4191, simple_loss=0.4382, pruned_loss=0.2, over 5679730.84 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3874, pruned_loss=0.1354, over 5660905.04 frames. ], batch size: 412, lr: 1.93e-02, grad_scale: 4.0
+2023-02-28 19:42:20,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.790e+03 2.234e+03 2.975e+03 5.479e+03, threshold=4.468e+03, percent-clipped=11.0
+2023-02-28 19:42:49,317 INFO [train.py:968] (0/2) Epoch 1, batch 33100, giga_loss[loss=0.3384, simple_loss=0.3995, pruned_loss=0.1387, over 28950.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3925, pruned_loss=0.1417, over 5670143.19 frames. ], libri_tot_loss[loss=0.4181, simple_loss=0.4373, pruned_loss=0.1994, over 5684158.99 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3888, pruned_loss=0.1362, over 5665837.89 frames. ], batch size: 213, lr: 1.93e-02, grad_scale: 4.0
+2023-02-28 19:43:52,856 INFO [train.py:968] (0/2) Epoch 1, batch 33150, giga_loss[loss=0.3665, simple_loss=0.4227, pruned_loss=0.1551, over 28931.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.393, pruned_loss=0.1423, over 5670701.38 frames. ], libri_tot_loss[loss=0.4181, simple_loss=0.4374, pruned_loss=0.1994, over 5686185.23 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.389, pruned_loss=0.1365, over 5665014.33 frames. ], batch size: 213, lr: 1.93e-02, grad_scale: 4.0
+2023-02-28 19:44:25,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.788e+02 1.600e+03 2.156e+03 2.963e+03 4.992e+03, threshold=4.311e+03, percent-clipped=4.0
+2023-02-28 19:44:53,703 INFO [train.py:968] (0/2) Epoch 1, batch 33200, giga_loss[loss=0.3575, simple_loss=0.3988, pruned_loss=0.158, over 28102.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3888, pruned_loss=0.1385, over 5676448.14 frames. ], libri_tot_loss[loss=0.4182, simple_loss=0.4375, pruned_loss=0.1995, over 5687942.81 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3851, pruned_loss=0.1334, over 5670357.61 frames. ], batch size: 412, lr: 1.93e-02, grad_scale: 8.0
+2023-02-28 19:45:58,708 INFO [train.py:968] (0/2) Epoch 1, batch 33250, giga_loss[loss=0.2922, simple_loss=0.3549, pruned_loss=0.1147, over 28714.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3875, pruned_loss=0.138, over 5677473.24 frames. ], libri_tot_loss[loss=0.4177, simple_loss=0.4371, pruned_loss=0.1992, over 5691932.09 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.384, pruned_loss=0.133, over 5668806.29 frames. ], batch size: 307, lr: 1.93e-02, grad_scale: 8.0
+2023-02-28 19:46:32,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.952e+02 1.521e+03 2.047e+03 2.967e+03 6.399e+03, threshold=4.093e+03, percent-clipped=5.0
+2023-02-28 19:46:59,390 INFO [train.py:968] (0/2) Epoch 1, batch 33300, giga_loss[loss=0.3142, simple_loss=0.3542, pruned_loss=0.1371, over 24694.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3855, pruned_loss=0.1371, over 5658311.62 frames. ], libri_tot_loss[loss=0.4179, simple_loss=0.4372, pruned_loss=0.1993, over 5673968.22 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3815, pruned_loss=0.1318, over 5667579.25 frames. ], batch size: 705, lr: 1.93e-02, grad_scale: 8.0
+2023-02-28 19:48:00,014 INFO [train.py:968] (0/2) Epoch 1, batch 33350, giga_loss[loss=0.3229, simple_loss=0.3906, pruned_loss=0.1276, over 28537.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3865, pruned_loss=0.1372, over 5668239.54 frames. ], libri_tot_loss[loss=0.417, simple_loss=0.4366, pruned_loss=0.1987, over 5680430.71 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3825, pruned_loss=0.1317, over 5669763.02 frames. ], batch size: 370, lr: 1.93e-02, grad_scale: 8.0
+2023-02-28 19:48:09,295 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=33359.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:48:16,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1894, 1.7517, 1.5517, 1.5787], device='cuda:0'), covar=tensor([0.0790, 0.1798, 0.1197, 0.1345], device='cuda:0'), in_proj_covar=tensor([0.0520, 0.0865, 0.0643, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 19:48:39,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.669e+02 1.486e+03 1.720e+03 2.270e+03 8.154e+03, threshold=3.440e+03, percent-clipped=2.0
+2023-02-28 19:49:03,293 INFO [train.py:968] (0/2) Epoch 1, batch 33400, giga_loss[loss=0.3524, simple_loss=0.3988, pruned_loss=0.153, over 28918.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3892, pruned_loss=0.1396, over 5676137.30 frames. ], libri_tot_loss[loss=0.4162, simple_loss=0.4361, pruned_loss=0.1982, over 5686194.25 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3851, pruned_loss=0.134, over 5672047.19 frames. ], batch size: 213, lr: 1.92e-02, grad_scale: 4.0
+2023-02-28 19:49:06,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5993, 1.9787, 1.7200, 1.5816], device='cuda:0'), covar=tensor([0.0947, 0.1061, 0.0833, 0.0571], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0816, 0.0704, 0.0605], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 19:49:24,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6720, 1.3672, 1.4429, 0.9838], device='cuda:0'), covar=tensor([0.0424, 0.0299, 0.0263, 0.0388], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0535, 0.0607, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 19:50:03,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5795, 1.9605, 1.5512, 0.5760], device='cuda:0'), covar=tensor([0.0969, 0.0762, 0.1012, 0.1476], device='cuda:0'), in_proj_covar=tensor([0.0984, 0.1057, 0.1006, 0.0913], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 19:50:09,142 INFO [train.py:968] (0/2) Epoch 1, batch 33450, giga_loss[loss=0.4414, simple_loss=0.442, pruned_loss=0.2204, over 26763.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3906, pruned_loss=0.1418, over 5671880.29 frames. ], libri_tot_loss[loss=0.4153, simple_loss=0.4354, pruned_loss=0.1977, over 5692844.75 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3867, pruned_loss=0.1362, over 5662442.37 frames. ], batch size: 555, lr: 1.92e-02, grad_scale: 4.0
+2023-02-28 19:50:29,342 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=33462.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:50:41,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6643, 2.1433, 1.6629, 1.6259], device='cuda:0'), covar=tensor([0.1259, 0.1161, 0.0960, 0.0595], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0833, 0.0714, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0010, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 19:50:52,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.728e+02 1.507e+03 2.041e+03 2.653e+03 6.688e+03, threshold=4.083e+03, percent-clipped=10.0
+2023-02-28 19:51:18,846 INFO [train.py:968] (0/2) Epoch 1, batch 33500, giga_loss[loss=0.3752, simple_loss=0.4341, pruned_loss=0.1582, over 28681.00 frames. ], tot_loss[loss=0.3444, simple_loss=0.3967, pruned_loss=0.1461, over 5662415.55 frames. ], libri_tot_loss[loss=0.4152, simple_loss=0.4352, pruned_loss=0.1976, over 5694578.58 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3932, pruned_loss=0.141, over 5652907.28 frames. ], batch size: 242, lr: 1.92e-02, grad_scale: 4.0
+2023-02-28 19:51:52,179 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0759, 1.1640, 1.0951, 0.8769], device='cuda:0'), covar=tensor([0.1575, 0.1510, 0.1345, 0.1666], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0747, 0.0837, 0.0908], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 19:52:10,388 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=33548.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:52:12,022 INFO [train.py:968] (0/2) Epoch 1, batch 33550, giga_loss[loss=0.3501, simple_loss=0.4162, pruned_loss=0.142, over 28771.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3992, pruned_loss=0.1473, over 5669878.36 frames. ], libri_tot_loss[loss=0.4145, simple_loss=0.4345, pruned_loss=0.1973, over 5697575.55 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3956, pruned_loss=0.1416, over 5658189.15 frames. ], batch size: 243, lr: 1.92e-02, grad_scale: 4.0
+2023-02-28 19:52:13,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1419, 1.3393, 1.2048, 1.1127], device='cuda:0'), covar=tensor([0.1304, 0.1289, 0.1167, 0.1294], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0740, 0.0837, 0.0903], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 19:52:52,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.728e+03 2.110e+03 2.945e+03 5.528e+03, threshold=4.220e+03, percent-clipped=10.0
+2023-02-28 19:53:22,823 INFO [train.py:968] (0/2) Epoch 1, batch 33600, giga_loss[loss=0.3392, simple_loss=0.3885, pruned_loss=0.145, over 27644.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3996, pruned_loss=0.1471, over 5664785.23 frames. ], libri_tot_loss[loss=0.4147, simple_loss=0.4347, pruned_loss=0.1974, over 5690066.32 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3961, pruned_loss=0.1417, over 5661104.97 frames. ], batch size: 472, lr: 1.92e-02, grad_scale: 8.0
+2023-02-28 19:54:38,633 INFO [train.py:968] (0/2) Epoch 1, batch 33650, giga_loss[loss=0.3246, simple_loss=0.3886, pruned_loss=0.1303, over 28905.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3957, pruned_loss=0.1442, over 5671862.48 frames. ], libri_tot_loss[loss=0.4153, simple_loss=0.4351, pruned_loss=0.1977, over 5690897.99 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3925, pruned_loss=0.1395, over 5668209.95 frames. ], batch size: 284, lr: 1.92e-02, grad_scale: 8.0
+2023-02-28 19:54:54,699 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5471, 2.1703, 1.7792, 1.7305], device='cuda:0'), covar=tensor([0.1299, 0.1300, 0.1005, 0.0626], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0845, 0.0717, 0.0623], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0010, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 19:55:17,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.822e+02 1.678e+03 2.084e+03 2.771e+03 5.505e+03, threshold=4.169e+03, percent-clipped=6.0
+2023-02-28 19:55:40,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5870, 1.5948, 3.3016, 2.4068], device='cuda:0'), covar=tensor([0.1413, 0.1179, 0.0351, 0.0705], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0457, 0.0577, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0005], device='cuda:0')
+2023-02-28 19:55:43,871 INFO [train.py:968] (0/2) Epoch 1, batch 33700, libri_loss[loss=0.3609, simple_loss=0.4055, pruned_loss=0.1582, over 29512.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3946, pruned_loss=0.1437, over 5677409.23 frames. ], libri_tot_loss[loss=0.4145, simple_loss=0.4343, pruned_loss=0.1973, over 5695463.97 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3916, pruned_loss=0.1389, over 5669610.15 frames. ], batch size: 76, lr: 1.92e-02, grad_scale: 8.0
+2023-02-28 19:56:27,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33734.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:56:28,921 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-02-28 19:56:50,260 INFO [train.py:968] (0/2) Epoch 1, batch 33750, giga_loss[loss=0.4298, simple_loss=0.4453, pruned_loss=0.2071, over 28544.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3935, pruned_loss=0.1434, over 5679603.66 frames. ], libri_tot_loss[loss=0.414, simple_loss=0.434, pruned_loss=0.197, over 5697108.00 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3905, pruned_loss=0.1386, over 5671383.79 frames. ], batch size: 336, lr: 1.91e-02, grad_scale: 4.0
+2023-02-28 19:57:29,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.980e+02 1.482e+03 1.816e+03 2.606e+03 1.003e+04, threshold=3.632e+03, percent-clipped=4.0
+2023-02-28 19:57:55,077 INFO [train.py:968] (0/2) Epoch 1, batch 33800, libri_loss[loss=0.4629, simple_loss=0.4713, pruned_loss=0.2273, over 29216.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3917, pruned_loss=0.1435, over 5674973.75 frames. ], libri_tot_loss[loss=0.4138, simple_loss=0.4337, pruned_loss=0.1969, over 5691953.15 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3884, pruned_loss=0.1382, over 5673189.00 frames. ], batch size: 94, lr: 1.91e-02, grad_scale: 4.0
+2023-02-28 19:58:11,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9196, 2.0213, 4.0963, 2.8537], device='cuda:0'), covar=tensor([0.1827, 0.1317, 0.0507, 0.0488], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0459, 0.0571, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 19:58:29,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0983, 1.1475, 1.1956, 0.8648], device='cuda:0'), covar=tensor([0.0331, 0.0267, 0.0213, 0.0277], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0550, 0.0615, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 19:58:41,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33837.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:58:49,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6883, 2.1905, 1.6331, 1.6016], device='cuda:0'), covar=tensor([0.1406, 0.1353, 0.1073, 0.0641], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0830, 0.0713, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0010, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 19:58:57,513 INFO [train.py:968] (0/2) Epoch 1, batch 33850, giga_loss[loss=0.2981, simple_loss=0.3738, pruned_loss=0.1112, over 28952.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3905, pruned_loss=0.1416, over 5684998.81 frames. ], libri_tot_loss[loss=0.4139, simple_loss=0.4338, pruned_loss=0.197, over 5696090.64 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3869, pruned_loss=0.1364, over 5679672.89 frames. ], batch size: 145, lr: 1.91e-02, grad_scale: 2.0
+2023-02-28 19:59:16,096 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=33865.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:59:31,780 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33877.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:59:36,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33880.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:59:36,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.330e+02 1.542e+03 2.097e+03 2.813e+03 1.051e+04, threshold=4.194e+03, percent-clipped=12.0
+2023-02-28 19:59:59,561 INFO [train.py:968] (0/2) Epoch 1, batch 33900, giga_loss[loss=0.2952, simple_loss=0.3646, pruned_loss=0.1129, over 28823.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3876, pruned_loss=0.1387, over 5677979.85 frames. ], libri_tot_loss[loss=0.4136, simple_loss=0.4335, pruned_loss=0.1968, over 5699318.29 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.384, pruned_loss=0.1333, over 5670401.56 frames. ], batch size: 284, lr: 1.91e-02, grad_scale: 2.0
+2023-02-28 20:00:12,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33909.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:00:27,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33923.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:00:56,150 INFO [train.py:968] (0/2) Epoch 1, batch 33950, giga_loss[loss=0.2995, simple_loss=0.3822, pruned_loss=0.1084, over 28840.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3901, pruned_loss=0.1383, over 5674814.29 frames. ], libri_tot_loss[loss=0.4126, simple_loss=0.4328, pruned_loss=0.1962, over 5693454.23 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3869, pruned_loss=0.1332, over 5673317.02 frames. ], batch size: 174, lr: 1.91e-02, grad_scale: 2.0
+2023-02-28 20:01:34,055 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6411, 1.4738, 1.2455, 1.3062], device='cuda:0'), covar=tensor([0.0693, 0.1137, 0.1275, 0.1038], device='cuda:0'), in_proj_covar=tensor([0.0514, 0.0844, 0.0638, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 20:01:37,292 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33980.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:01:37,595 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.721e+02 1.565e+03 2.048e+03 2.597e+03 1.299e+04, threshold=4.097e+03, percent-clipped=8.0
+2023-02-28 20:01:40,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33983.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:02:01,295 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-34000.pt
+2023-02-28 20:02:01,626 INFO [train.py:968] (0/2) Epoch 1, batch 34000, giga_loss[loss=0.4107, simple_loss=0.4519, pruned_loss=0.1847, over 28810.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3912, pruned_loss=0.137, over 5675353.17 frames. ], libri_tot_loss[loss=0.4129, simple_loss=0.433, pruned_loss=0.1964, over 5691873.53 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3882, pruned_loss=0.1325, over 5675101.48 frames. ], batch size: 119, lr: 1.91e-02, grad_scale: 4.0
+2023-02-28 20:02:14,332 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34012.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:03:05,142 INFO [train.py:968] (0/2) Epoch 1, batch 34050, giga_loss[loss=0.2949, simple_loss=0.3426, pruned_loss=0.1236, over 24218.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3913, pruned_loss=0.1371, over 5672023.72 frames. ], libri_tot_loss[loss=0.4121, simple_loss=0.4323, pruned_loss=0.196, over 5693491.38 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3885, pruned_loss=0.1325, over 5669967.19 frames. ], batch size: 705, lr: 1.91e-02, grad_scale: 4.0
+2023-02-28 20:03:28,403 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34066.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:03:31,347 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34069.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:03:35,886 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=34070.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:03:46,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7075, 1.9756, 1.5789, 1.4531], device='cuda:0'), covar=tensor([0.0920, 0.0733, 0.0942, 0.1422], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0448, 0.0376, 0.0470], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0013], device='cuda:0')
+2023-02-28 20:03:49,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5360, 1.7575, 1.2881, 1.1323], device='cuda:0'), covar=tensor([0.0534, 0.0336, 0.0337, 0.0394], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0541, 0.0600, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 20:03:49,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.648e+02 1.397e+03 1.892e+03 2.450e+03 5.142e+03, threshold=3.783e+03, percent-clipped=2.0
+2023-02-28 20:04:08,937 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-02-28 20:04:16,252 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34098.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:04:17,885 INFO [train.py:968] (0/2) Epoch 1, batch 34100, giga_loss[loss=0.3589, simple_loss=0.4137, pruned_loss=0.1521, over 28464.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3927, pruned_loss=0.1386, over 5665806.41 frames. ], libri_tot_loss[loss=0.4124, simple_loss=0.4325, pruned_loss=0.1961, over 5692180.49 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3898, pruned_loss=0.1341, over 5665404.65 frames. ], batch size: 336, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:05:25,360 INFO [train.py:968] (0/2) Epoch 1, batch 34150, giga_loss[loss=0.3246, simple_loss=0.3724, pruned_loss=0.1384, over 26853.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3921, pruned_loss=0.1385, over 5652530.33 frames. ], libri_tot_loss[loss=0.412, simple_loss=0.4322, pruned_loss=0.1959, over 5679684.21 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3889, pruned_loss=0.1333, over 5663375.88 frames. ], batch size: 555, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:06:07,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.908e+02 1.653e+03 2.128e+03 2.954e+03 9.145e+03, threshold=4.257e+03, percent-clipped=12.0
+2023-02-28 20:06:33,546 INFO [train.py:968] (0/2) Epoch 1, batch 34200, giga_loss[loss=0.3151, simple_loss=0.3775, pruned_loss=0.1263, over 27660.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3934, pruned_loss=0.139, over 5663609.45 frames. ], libri_tot_loss[loss=0.412, simple_loss=0.432, pruned_loss=0.196, over 5684835.97 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3896, pruned_loss=0.1327, over 5666541.17 frames. ], batch size: 472, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:07:04,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3169, 1.2040, 1.2115, 1.2328], device='cuda:0'), covar=tensor([0.1733, 0.1863, 0.1474, 0.1970], device='cuda:0'), in_proj_covar=tensor([0.0880, 0.0756, 0.0838, 0.0900], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 20:07:33,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34240.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:07:44,998 INFO [train.py:968] (0/2) Epoch 1, batch 34250, giga_loss[loss=0.2931, simple_loss=0.3682, pruned_loss=0.109, over 28764.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3952, pruned_loss=0.1402, over 5666784.58 frames. ], libri_tot_loss[loss=0.4122, simple_loss=0.4321, pruned_loss=0.1961, over 5687092.36 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3915, pruned_loss=0.1343, over 5666822.90 frames. ], batch size: 119, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:07:50,974 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3647, 1.1169, 1.1302, 1.3655], device='cuda:0'), covar=tensor([0.1648, 0.1864, 0.1498, 0.2232], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0747, 0.0829, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 20:08:25,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.773e+02 1.713e+03 2.339e+03 3.280e+03 6.281e+03, threshold=4.678e+03, percent-clipped=11.0
+2023-02-28 20:08:28,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5834, 2.0849, 1.3875, 0.4549], device='cuda:0'), covar=tensor([0.0857, 0.0770, 0.1258, 0.1628], device='cuda:0'), in_proj_covar=tensor([0.0996, 0.1079, 0.1010, 0.0926], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 20:08:28,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6546, 2.2109, 1.6950, 1.6212], device='cuda:0'), covar=tensor([0.1227, 0.1174, 0.1044, 0.0690], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0818, 0.0702, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0010, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 20:08:37,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4943, 2.2451, 1.6220, 0.4177], device='cuda:0'), covar=tensor([0.1034, 0.0732, 0.1058, 0.1656], device='cuda:0'), in_proj_covar=tensor([0.1000, 0.1083, 0.1013, 0.0929], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 20:08:52,318 INFO [train.py:968] (0/2) Epoch 1, batch 34300, giga_loss[loss=0.4141, simple_loss=0.4574, pruned_loss=0.1853, over 28460.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3977, pruned_loss=0.141, over 5672509.65 frames. ], libri_tot_loss[loss=0.4114, simple_loss=0.4315, pruned_loss=0.1956, over 5689033.57 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3945, pruned_loss=0.1355, over 5670418.29 frames. ], batch size: 336, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:09:39,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=34332.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:09:58,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-02-28 20:09:59,890 INFO [train.py:968] (0/2) Epoch 1, batch 34350, giga_loss[loss=0.283, simple_loss=0.3543, pruned_loss=0.1058, over 28635.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3963, pruned_loss=0.1409, over 5668340.73 frames. ], libri_tot_loss[loss=0.4101, simple_loss=0.4307, pruned_loss=0.1948, over 5685352.27 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3935, pruned_loss=0.1354, over 5669912.87 frames. ], batch size: 92, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:10:46,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.481e+02 1.478e+03 1.824e+03 2.495e+03 6.883e+03, threshold=3.648e+03, percent-clipped=3.0
+2023-02-28 20:10:51,169 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34383.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:10:54,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34386.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:10:58,218 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-02-28 20:11:10,638 INFO [train.py:968] (0/2) Epoch 1, batch 34400, giga_loss[loss=0.2673, simple_loss=0.3424, pruned_loss=0.09609, over 28908.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3937, pruned_loss=0.1397, over 5677149.19 frames. ], libri_tot_loss[loss=0.409, simple_loss=0.43, pruned_loss=0.1941, over 5687225.45 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3913, pruned_loss=0.1348, over 5676044.15 frames. ], batch size: 106, lr: 1.90e-02, grad_scale: 8.0
+2023-02-28 20:11:11,238 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-02-28 20:11:33,382 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34415.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:12:08,442 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=34437.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:12:21,094 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34445.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:12:30,235 INFO [train.py:968] (0/2) Epoch 1, batch 34450, giga_loss[loss=0.2889, simple_loss=0.3427, pruned_loss=0.1175, over 25050.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.391, pruned_loss=0.1371, over 5674213.65 frames. ], libri_tot_loss[loss=0.4094, simple_loss=0.4304, pruned_loss=0.1942, over 5688250.61 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3878, pruned_loss=0.1318, over 5672120.36 frames. ], batch size: 705, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:13:07,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.679e+02 1.327e+03 1.631e+03 2.379e+03 5.452e+03, threshold=3.262e+03, percent-clipped=5.0
+2023-02-28 20:13:33,425 INFO [train.py:968] (0/2) Epoch 1, batch 34500, giga_loss[loss=0.3514, simple_loss=0.4084, pruned_loss=0.1472, over 28663.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3895, pruned_loss=0.1358, over 5671607.51 frames. ], libri_tot_loss[loss=0.4093, simple_loss=0.4304, pruned_loss=0.1941, over 5683446.47 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3861, pruned_loss=0.1304, over 5673623.59 frames. ], batch size: 262, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:13:41,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7553, 1.6989, 3.6159, 2.6976], device='cuda:0'), covar=tensor([0.1404, 0.1241, 0.0323, 0.0606], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0465, 0.0591, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 20:14:23,861 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=34538.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:14:38,508 INFO [train.py:968] (0/2) Epoch 1, batch 34550, giga_loss[loss=0.4143, simple_loss=0.4591, pruned_loss=0.1847, over 28370.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3923, pruned_loss=0.1381, over 5669955.04 frames. ], libri_tot_loss[loss=0.4096, simple_loss=0.4306, pruned_loss=0.1943, over 5686755.80 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3888, pruned_loss=0.1329, over 5668320.36 frames. ], batch size: 368, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:15:17,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.947e+02 1.657e+03 2.121e+03 2.875e+03 7.244e+03, threshold=4.242e+03, percent-clipped=15.0
+2023-02-28 20:15:25,861 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34588.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:15:29,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34591.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:15:41,712 INFO [train.py:968] (0/2) Epoch 1, batch 34600, giga_loss[loss=0.3548, simple_loss=0.4109, pruned_loss=0.1493, over 28887.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3963, pruned_loss=0.1406, over 5675064.33 frames. ], libri_tot_loss[loss=0.4094, simple_loss=0.4307, pruned_loss=0.1941, over 5691688.84 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3927, pruned_loss=0.1354, over 5669022.36 frames. ], batch size: 174, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:15:59,569 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 20:15:59,606 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-02-28 20:16:03,776 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34620.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:16:43,487 INFO [train.py:968] (0/2) Epoch 1, batch 34650, giga_loss[loss=0.3311, simple_loss=0.3918, pruned_loss=0.1352, over 28948.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3953, pruned_loss=0.1409, over 5682959.76 frames. ], libri_tot_loss[loss=0.4091, simple_loss=0.4304, pruned_loss=0.1938, over 5694615.82 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3921, pruned_loss=0.1363, over 5675419.44 frames. ], batch size: 284, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:17:04,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8940, 1.0492, 0.8997, 0.5066], device='cuda:0'), covar=tensor([0.0255, 0.0253, 0.0265, 0.0329], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0533, 0.0607, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 20:17:22,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.180e+03 1.788e+03 2.166e+03 3.018e+03 7.184e+03, threshold=4.333e+03, percent-clipped=8.0
+2023-02-28 20:17:28,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3117, 1.6357, 1.1872, 1.3388], device='cuda:0'), covar=tensor([0.1411, 0.1993, 0.1524, 0.0961], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0825, 0.0709, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0010, 0.0007, 0.0005], device='cuda:0')
+2023-02-28 20:17:43,516 INFO [train.py:968] (0/2) Epoch 1, batch 34700, giga_loss[loss=0.3923, simple_loss=0.4281, pruned_loss=0.1782, over 28480.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3937, pruned_loss=0.1421, over 5665753.73 frames. ], libri_tot_loss[loss=0.4096, simple_loss=0.4306, pruned_loss=0.1943, over 5689720.69 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3902, pruned_loss=0.1368, over 5664248.10 frames. ], batch size: 336, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:17:54,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34707.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:18:42,098 INFO [train.py:968] (0/2) Epoch 1, batch 34750, giga_loss[loss=0.3761, simple_loss=0.4295, pruned_loss=0.1614, over 28907.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.393, pruned_loss=0.1421, over 5674240.68 frames. ], libri_tot_loss[loss=0.4088, simple_loss=0.4301, pruned_loss=0.1938, over 5694734.81 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3899, pruned_loss=0.1371, over 5667908.27 frames. ], batch size: 213, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:19:08,568 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-02-28 20:19:17,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.374e+02 1.673e+03 2.049e+03 2.654e+03 7.487e+03, threshold=4.098e+03, percent-clipped=6.0
+2023-02-28 20:19:32,695 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.40 vs. limit=5.0
+2023-02-28 20:19:34,401 INFO [train.py:968] (0/2) Epoch 1, batch 34800, giga_loss[loss=0.3968, simple_loss=0.4457, pruned_loss=0.1739, over 28931.00 frames. ], tot_loss[loss=0.3514, simple_loss=0.403, pruned_loss=0.1498, over 5666806.10 frames. ], libri_tot_loss[loss=0.4083, simple_loss=0.4299, pruned_loss=0.1934, over 5689577.79 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3997, pruned_loss=0.1448, over 5665571.12 frames. ], batch size: 213, lr: 1.89e-02, grad_scale: 8.0
+2023-02-28 20:19:44,813 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34812.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:19:54,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7154, 1.4360, 1.3236, 1.2959], device='cuda:0'), covar=tensor([0.0676, 0.1199, 0.1240, 0.1023], device='cuda:0'), in_proj_covar=tensor([0.0526, 0.0881, 0.0652, 0.0711], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 20:20:24,148 INFO [train.py:968] (0/2) Epoch 1, batch 34850, giga_loss[loss=0.4281, simple_loss=0.4583, pruned_loss=0.1989, over 27651.00 frames. ], tot_loss[loss=0.3641, simple_loss=0.4136, pruned_loss=0.1573, over 5668498.32 frames. ], libri_tot_loss[loss=0.4079, simple_loss=0.4297, pruned_loss=0.1931, over 5692688.06 frames. ], giga_tot_loss[loss=0.3584, simple_loss=0.4108, pruned_loss=0.1529, over 5664651.07 frames. ], batch size: 472, lr: 1.88e-02, grad_scale: 8.0
+2023-02-28 20:20:24,410 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34850.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:20:26,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34853.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:20:49,588 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-02-28 20:20:54,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.297e+02 1.269e+03 1.608e+03 2.303e+03 4.402e+03, threshold=3.216e+03, percent-clipped=2.0
+2023-02-28 20:20:54,451 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34882.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:21:09,143 INFO [train.py:968] (0/2) Epoch 1, batch 34900, giga_loss[loss=0.3579, simple_loss=0.4049, pruned_loss=0.1554, over 28980.00 frames. ], tot_loss[loss=0.3709, simple_loss=0.418, pruned_loss=0.1619, over 5655741.46 frames. ], libri_tot_loss[loss=0.4085, simple_loss=0.43, pruned_loss=0.1935, over 5678349.91 frames. ], giga_tot_loss[loss=0.365, simple_loss=0.4152, pruned_loss=0.1574, over 5665935.05 frames. ], batch size: 106, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:21:10,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6451, 1.5128, 3.5287, 2.6122], device='cuda:0'), covar=tensor([0.1453, 0.1357, 0.0345, 0.0608], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0465, 0.0588, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 20:21:19,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34913.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:21:38,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3684, 1.9489, 1.5657, 1.3921], device='cuda:0'), covar=tensor([0.1351, 0.0534, 0.0622, 0.1621], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0291, 0.0289, 0.0495], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0017, 0.0015, 0.0025], device='cuda:0')
+2023-02-28 20:21:51,778 INFO [train.py:968] (0/2) Epoch 1, batch 34950, giga_loss[loss=0.3595, simple_loss=0.3994, pruned_loss=0.1597, over 28826.00 frames. ], tot_loss[loss=0.3659, simple_loss=0.4124, pruned_loss=0.1597, over 5673041.94 frames. ], libri_tot_loss[loss=0.4085, simple_loss=0.4302, pruned_loss=0.1934, over 5683596.39 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.4097, pruned_loss=0.1554, over 5676084.75 frames. ], batch size: 199, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:21:55,536 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34955.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:21:57,582 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34958.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:22:17,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.621e+02 1.397e+03 1.753e+03 2.423e+03 4.102e+03, threshold=3.506e+03, percent-clipped=2.0
+2023-02-28 20:22:19,418 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=34986.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:22:20,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34987.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:22:32,582 INFO [train.py:968] (0/2) Epoch 1, batch 35000, giga_loss[loss=0.3108, simple_loss=0.3685, pruned_loss=0.1266, over 29027.00 frames. ], tot_loss[loss=0.3611, simple_loss=0.4069, pruned_loss=0.1576, over 5688129.76 frames. ], libri_tot_loss[loss=0.409, simple_loss=0.4308, pruned_loss=0.1937, over 5691350.76 frames. ], giga_tot_loss[loss=0.3542, simple_loss=0.4034, pruned_loss=0.1525, over 5683413.32 frames. ], batch size: 164, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:22:45,922 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35015.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:22:50,204 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35019.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:23:14,922 INFO [train.py:968] (0/2) Epoch 1, batch 35050, giga_loss[loss=0.3018, simple_loss=0.3542, pruned_loss=0.1247, over 28906.00 frames. ], tot_loss[loss=0.3547, simple_loss=0.4, pruned_loss=0.1547, over 5680739.62 frames. ], libri_tot_loss[loss=0.4099, simple_loss=0.4314, pruned_loss=0.1942, over 5687200.25 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.3958, pruned_loss=0.1491, over 5679703.88 frames. ], batch size: 199, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:23:20,607 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35056.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:23:22,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35059.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:23:42,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.490e+02 1.255e+03 1.598e+03 2.354e+03 6.445e+03, threshold=3.195e+03, percent-clipped=9.0
+2023-02-28 20:23:47,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35088.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:23:57,796 INFO [train.py:968] (0/2) Epoch 1, batch 35100, giga_loss[loss=0.2951, simple_loss=0.3491, pruned_loss=0.1205, over 28829.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.3924, pruned_loss=0.1509, over 5684816.77 frames. ], libri_tot_loss[loss=0.411, simple_loss=0.4323, pruned_loss=0.1948, over 5689009.24 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3872, pruned_loss=0.1446, over 5682091.75 frames. ], batch size: 186, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:24:00,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35103.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:24:36,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1030, 3.0661, 3.8214, 1.8414], device='cuda:0'), covar=tensor([0.0557, 0.0564, 0.0732, 0.1915], device='cuda:0'), in_proj_covar=tensor([0.0715, 0.0480, 0.0763, 0.0526], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 20:24:41,820 INFO [train.py:968] (0/2) Epoch 1, batch 35150, giga_loss[loss=0.299, simple_loss=0.3529, pruned_loss=0.1225, over 29033.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.384, pruned_loss=0.1462, over 5685564.64 frames. ], libri_tot_loss[loss=0.4106, simple_loss=0.4319, pruned_loss=0.1946, over 5689961.17 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3795, pruned_loss=0.1406, over 5682480.42 frames. ], batch size: 128, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:25:10,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.994e+02 1.208e+03 1.691e+03 2.482e+03 6.743e+03, threshold=3.383e+03, percent-clipped=9.0
+2023-02-28 20:25:26,443 INFO [train.py:968] (0/2) Epoch 1, batch 35200, giga_loss[loss=0.2808, simple_loss=0.3383, pruned_loss=0.1116, over 28500.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3797, pruned_loss=0.1441, over 5676787.92 frames. ], libri_tot_loss[loss=0.4116, simple_loss=0.4328, pruned_loss=0.1952, over 5684136.76 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3745, pruned_loss=0.1384, over 5679612.75 frames. ], batch size: 336, lr: 1.88e-02, grad_scale: 8.0
+2023-02-28 20:25:49,997 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-02-28 20:26:11,576 INFO [train.py:968] (0/2) Epoch 1, batch 35250, giga_loss[loss=0.3036, simple_loss=0.3554, pruned_loss=0.1259, over 28913.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3745, pruned_loss=0.1402, over 5679095.14 frames. ], libri_tot_loss[loss=0.4124, simple_loss=0.4334, pruned_loss=0.1957, over 5676977.74 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3691, pruned_loss=0.1345, over 5687362.29 frames. ], batch size: 145, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:26:42,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.549e+02 1.186e+03 1.459e+03 2.032e+03 3.723e+03, threshold=2.919e+03, percent-clipped=1.0
+2023-02-28 20:26:54,892 INFO [train.py:968] (0/2) Epoch 1, batch 35300, giga_loss[loss=0.2495, simple_loss=0.3144, pruned_loss=0.09232, over 28942.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3701, pruned_loss=0.137, over 5691057.86 frames. ], libri_tot_loss[loss=0.4129, simple_loss=0.4339, pruned_loss=0.1959, over 5680090.06 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3642, pruned_loss=0.1311, over 5694856.84 frames. ], batch size: 136, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:27:43,439 INFO [train.py:968] (0/2) Epoch 1, batch 35350, giga_loss[loss=0.2883, simple_loss=0.3518, pruned_loss=0.1124, over 29044.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3659, pruned_loss=0.1341, over 5700573.77 frames. ], libri_tot_loss[loss=0.4138, simple_loss=0.4346, pruned_loss=0.1965, over 5682648.04 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3597, pruned_loss=0.1281, over 5701525.10 frames. ], batch size: 155, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:27:52,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35361.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:27:57,515 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-02-28 20:28:12,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.306e+02 1.196e+03 1.421e+03 1.925e+03 8.924e+03, threshold=2.842e+03, percent-clipped=12.0
+2023-02-28 20:28:16,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35390.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:28:20,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35394.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:28:25,026 INFO [train.py:968] (0/2) Epoch 1, batch 35400, giga_loss[loss=0.2337, simple_loss=0.2986, pruned_loss=0.08438, over 28491.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3642, pruned_loss=0.1333, over 5687755.29 frames. ], libri_tot_loss[loss=0.4152, simple_loss=0.4357, pruned_loss=0.1973, over 5677993.69 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3556, pruned_loss=0.1255, over 5693354.91 frames. ], batch size: 60, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:28:53,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0215, 1.0867, 1.0039, 0.6175], device='cuda:0'), covar=tensor([0.0316, 0.0228, 0.0208, 0.0286], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0536, 0.0601, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 20:28:58,075 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-02-28 20:29:05,154 INFO [train.py:968] (0/2) Epoch 1, batch 35450, giga_loss[loss=0.2695, simple_loss=0.3319, pruned_loss=0.1036, over 28873.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3635, pruned_loss=0.1334, over 5694852.06 frames. ], libri_tot_loss[loss=0.4159, simple_loss=0.4363, pruned_loss=0.1977, over 5684083.75 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3533, pruned_loss=0.1243, over 5694131.54 frames. ], batch size: 174, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:29:27,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35478.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:29:32,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.445e+02 1.279e+03 1.846e+03 2.561e+03 5.870e+03, threshold=3.692e+03, percent-clipped=14.0
+2023-02-28 20:29:34,538 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35485.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:29:46,926 INFO [train.py:968] (0/2) Epoch 1, batch 35500, giga_loss[loss=0.2956, simple_loss=0.3558, pruned_loss=0.1177, over 28786.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3607, pruned_loss=0.1316, over 5692352.12 frames. ], libri_tot_loss[loss=0.416, simple_loss=0.4366, pruned_loss=0.1977, over 5687859.84 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3506, pruned_loss=0.1229, over 5688618.79 frames. ], batch size: 262, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:29:48,949 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-02-28 20:29:51,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35504.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:29:55,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35507.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:17,027 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35533.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:18,769 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35536.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:18,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35536.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:19,450 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35537.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:22,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35540.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:31,740 INFO [train.py:968] (0/2) Epoch 1, batch 35550, libri_loss[loss=0.5249, simple_loss=0.4948, pruned_loss=0.2775, over 29613.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3595, pruned_loss=0.1315, over 5690807.78 frames. ], libri_tot_loss[loss=0.4178, simple_loss=0.438, pruned_loss=0.1989, over 5682233.04 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3468, pruned_loss=0.1205, over 5693301.62 frames. ], batch size: 69, lr: 1.87e-02, grad_scale: 1.0
+2023-02-28 20:30:43,521 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35565.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:46,213 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4366, 1.9669, 1.5654, 1.6186], device='cuda:0'), covar=tensor([0.1299, 0.1347, 0.1121, 0.0663], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0831, 0.0716, 0.0625], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0010, 0.0007, 0.0006], device='cuda:0')
+2023-02-28 20:30:47,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35569.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:31:03,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.154e+02 1.276e+03 1.709e+03 2.547e+03 2.858e+04, threshold=3.419e+03, percent-clipped=14.0
+2023-02-28 20:31:16,169 INFO [train.py:968] (0/2) Epoch 1, batch 35600, giga_loss[loss=0.4798, simple_loss=0.4622, pruned_loss=0.2488, over 26711.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3593, pruned_loss=0.1325, over 5689117.43 frames. ], libri_tot_loss[loss=0.4188, simple_loss=0.4387, pruned_loss=0.1995, over 5684927.30 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3475, pruned_loss=0.1224, over 5688993.69 frames. ], batch size: 555, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:31:31,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5482, 1.4451, 1.1521, 1.1920], device='cuda:0'), covar=tensor([0.0654, 0.0655, 0.1020, 0.1000], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0568, 0.0582, 0.0522], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-02-28 20:31:36,943 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35621.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:31:39,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35624.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:32:04,901 INFO [train.py:968] (0/2) Epoch 1, batch 35650, giga_loss[loss=0.3954, simple_loss=0.4433, pruned_loss=0.1737, over 29105.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3714, pruned_loss=0.1402, over 5686764.94 frames. ], libri_tot_loss[loss=0.4197, simple_loss=0.4394, pruned_loss=0.2, over 5687051.54 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3609, pruned_loss=0.1311, over 5684802.88 frames. ], batch size: 155, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:32:07,808 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35653.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:32:41,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.868e+02 1.321e+03 1.710e+03 2.057e+03 5.015e+03, threshold=3.419e+03, percent-clipped=4.0
+2023-02-28 20:32:54,028 INFO [train.py:968] (0/2) Epoch 1, batch 35700, giga_loss[loss=0.3765, simple_loss=0.4322, pruned_loss=0.1604, over 28295.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3875, pruned_loss=0.1499, over 5684115.88 frames. ], libri_tot_loss[loss=0.4201, simple_loss=0.4397, pruned_loss=0.2003, over 5684336.29 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3788, pruned_loss=0.1424, over 5684948.10 frames. ], batch size: 65, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:33:17,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0352, 1.0838, 0.8025, 1.1596], device='cuda:0'), covar=tensor([0.1232, 0.0638, 0.0654, 0.1525], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0283, 0.0280, 0.0481], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0017, 0.0015, 0.0024], device='cuda:0')
+2023-02-28 20:33:28,096 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35738.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:33:40,453 INFO [train.py:968] (0/2) Epoch 1, batch 35750, giga_loss[loss=0.3524, simple_loss=0.4168, pruned_loss=0.1439, over 28700.00 frames. ], tot_loss[loss=0.3603, simple_loss=0.4022, pruned_loss=0.1592, over 5688345.63 frames. ], libri_tot_loss[loss=0.4211, simple_loss=0.4404, pruned_loss=0.2009, over 5686328.72 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.3943, pruned_loss=0.1524, over 5687360.71 frames. ], batch size: 242, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:33:56,241 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35767.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:34:12,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.987e+02 1.401e+03 1.781e+03 2.486e+03 4.764e+03, threshold=3.563e+03, percent-clipped=5.0
+2023-02-28 20:34:25,046 INFO [train.py:968] (0/2) Epoch 1, batch 35800, giga_loss[loss=0.3441, simple_loss=0.4007, pruned_loss=0.1438, over 28813.00 frames. ], tot_loss[loss=0.3634, simple_loss=0.4071, pruned_loss=0.1599, over 5694429.85 frames. ], libri_tot_loss[loss=0.4214, simple_loss=0.4408, pruned_loss=0.201, over 5688746.70 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.4001, pruned_loss=0.1539, over 5691648.41 frames. ], batch size: 99, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:35:12,526 INFO [train.py:968] (0/2) Epoch 1, batch 35850, giga_loss[loss=0.3456, simple_loss=0.3977, pruned_loss=0.1468, over 28726.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4084, pruned_loss=0.1583, over 5693333.52 frames. ], libri_tot_loss[loss=0.4217, simple_loss=0.4411, pruned_loss=0.2011, over 5690818.42 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4023, pruned_loss=0.153, over 5689159.85 frames. ], batch size: 92, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:35:14,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3340, 1.6308, 1.6000, 1.3874], device='cuda:0'), covar=tensor([0.0960, 0.2012, 0.1319, 0.1675], device='cuda:0'), in_proj_covar=tensor([0.0532, 0.0866, 0.0655, 0.0710], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 20:35:24,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35860.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:35:25,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8620, 2.3016, 1.9269, 1.8135], device='cuda:0'), covar=tensor([0.1215, 0.1062, 0.0933, 0.0585], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0816, 0.0707, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0010, 0.0007, 0.0006], device='cuda:0')
+2023-02-28 20:35:49,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.765e+02 1.173e+03 1.362e+03 1.707e+03 6.638e+03, threshold=2.723e+03, percent-clipped=3.0
+2023-02-28 20:36:01,381 INFO [train.py:968] (0/2) Epoch 1, batch 35900, giga_loss[loss=0.3522, simple_loss=0.4063, pruned_loss=0.1491, over 28902.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4104, pruned_loss=0.1591, over 5693273.95 frames. ], libri_tot_loss[loss=0.4222, simple_loss=0.4415, pruned_loss=0.2015, over 5692989.76 frames. ], giga_tot_loss[loss=0.3566, simple_loss=0.4049, pruned_loss=0.1542, over 5688023.04 frames. ], batch size: 145, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:36:20,610 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35923.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:36:43,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5027, 1.9265, 1.4218, 0.7299], device='cuda:0'), covar=tensor([0.0936, 0.0633, 0.1140, 0.1332], device='cuda:0'), in_proj_covar=tensor([0.1017, 0.1018, 0.1017, 0.0918], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 20:36:46,679 INFO [train.py:968] (0/2) Epoch 1, batch 35950, giga_loss[loss=0.5123, simple_loss=0.4889, pruned_loss=0.2679, over 26657.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.413, pruned_loss=0.1619, over 5679313.57 frames. ], libri_tot_loss[loss=0.4233, simple_loss=0.4422, pruned_loss=0.2022, over 5689977.42 frames. ], giga_tot_loss[loss=0.3599, simple_loss=0.4072, pruned_loss=0.1563, over 5678867.34 frames. ], batch size: 555, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:36:52,805 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-02-28 20:37:12,257 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 20:37:16,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.059e+02 1.522e+03 1.832e+03 2.349e+03 7.886e+03, threshold=3.665e+03, percent-clipped=15.0
+2023-02-28 20:37:27,473 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-36000.pt
+2023-02-28 20:37:27,785 INFO [train.py:968] (0/2) Epoch 1, batch 36000, giga_loss[loss=0.3504, simple_loss=0.4097, pruned_loss=0.1455, over 28756.00 frames. ], tot_loss[loss=0.373, simple_loss=0.4169, pruned_loss=0.1645, over 5677703.95 frames. ], libri_tot_loss[loss=0.4242, simple_loss=0.443, pruned_loss=0.2027, over 5685304.76 frames. ], giga_tot_loss[loss=0.3642, simple_loss=0.411, pruned_loss=0.1588, over 5681071.62 frames. ], batch size: 284, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:37:27,789 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 20:37:35,987 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.3098, simple_loss=0.3985, pruned_loss=0.1105, over 944034.00 frames. 
+2023-02-28 20:37:35,988 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 20:37:38,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1621, 1.3920, 1.4628, 1.1726], device='cuda:0'), covar=tensor([0.1037, 0.2325, 0.1472, 0.1800], device='cuda:0'), in_proj_covar=tensor([0.0535, 0.0887, 0.0666, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 20:37:39,634 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36003.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:37:41,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36006.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:38:05,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36035.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:38:17,298 INFO [train.py:968] (0/2) Epoch 1, batch 36050, giga_loss[loss=0.3603, simple_loss=0.4136, pruned_loss=0.1535, over 28957.00 frames. ], tot_loss[loss=0.3757, simple_loss=0.4195, pruned_loss=0.166, over 5681758.94 frames. ], libri_tot_loss[loss=0.4242, simple_loss=0.4432, pruned_loss=0.2026, over 5686480.89 frames. ], giga_tot_loss[loss=0.3674, simple_loss=0.4139, pruned_loss=0.1605, over 5683035.34 frames. ], batch size: 106, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:38:46,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.147e+02 1.204e+03 1.645e+03 2.482e+03 5.549e+03, threshold=3.290e+03, percent-clipped=7.0
+2023-02-28 20:38:57,242 INFO [train.py:968] (0/2) Epoch 1, batch 36100, giga_loss[loss=0.4186, simple_loss=0.4549, pruned_loss=0.1911, over 28576.00 frames. ], tot_loss[loss=0.3781, simple_loss=0.4224, pruned_loss=0.1669, over 5692760.72 frames. ], libri_tot_loss[loss=0.4249, simple_loss=0.4437, pruned_loss=0.203, over 5691071.23 frames. ], giga_tot_loss[loss=0.3695, simple_loss=0.4168, pruned_loss=0.1611, over 5689520.33 frames. ], batch size: 336, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:39:08,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36113.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:39:24,979 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-02-28 20:39:34,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36142.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:39:42,286 INFO [train.py:968] (0/2) Epoch 1, batch 36150, libri_loss[loss=0.5269, simple_loss=0.5182, pruned_loss=0.2678, over 27985.00 frames. ], tot_loss[loss=0.3823, simple_loss=0.4259, pruned_loss=0.1694, over 5673342.40 frames. ], libri_tot_loss[loss=0.4268, simple_loss=0.4451, pruned_loss=0.2042, over 5684445.80 frames. ], giga_tot_loss[loss=0.373, simple_loss=0.4199, pruned_loss=0.1631, over 5675976.61 frames. ], batch size: 116, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:39:52,805 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=36164.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:40:08,556 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.564e+02 1.342e+03 1.823e+03 2.679e+03 1.607e+04, threshold=3.646e+03, percent-clipped=18.0
+2023-02-28 20:40:21,762 INFO [train.py:968] (0/2) Epoch 1, batch 36200, giga_loss[loss=0.3248, simple_loss=0.3946, pruned_loss=0.1275, over 28619.00 frames. ], tot_loss[loss=0.3817, simple_loss=0.4268, pruned_loss=0.1683, over 5685781.76 frames. ], libri_tot_loss[loss=0.428, simple_loss=0.4463, pruned_loss=0.2049, over 5684658.11 frames. ], giga_tot_loss[loss=0.3721, simple_loss=0.4206, pruned_loss=0.1619, over 5687427.65 frames. ], batch size: 60, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:40:26,116 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=36204.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:41:02,930 INFO [train.py:968] (0/2) Epoch 1, batch 36250, libri_loss[loss=0.4689, simple_loss=0.4868, pruned_loss=0.2255, over 29159.00 frames. ], tot_loss[loss=0.3792, simple_loss=0.426, pruned_loss=0.1662, over 5697731.74 frames. ], libri_tot_loss[loss=0.4291, simple_loss=0.4472, pruned_loss=0.2056, over 5686781.45 frames. ], giga_tot_loss[loss=0.3695, simple_loss=0.4198, pruned_loss=0.1597, over 5697170.67 frames. ], batch size: 97, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:41:07,864 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36256.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:41:10,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36259.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:41:31,892 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36285.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:41:32,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.069e+02 1.290e+03 1.548e+03 2.071e+03 6.559e+03, threshold=3.096e+03, percent-clipped=6.0
+2023-02-28 20:41:33,868 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36288.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:41:33,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36288.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:41:42,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36298.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:41:43,987 INFO [train.py:968] (0/2) Epoch 1, batch 36300, giga_loss[loss=0.3777, simple_loss=0.4276, pruned_loss=0.1639, over 28736.00 frames. ], tot_loss[loss=0.3762, simple_loss=0.4244, pruned_loss=0.164, over 5700819.04 frames. ], libri_tot_loss[loss=0.4294, simple_loss=0.4474, pruned_loss=0.2057, over 5685656.95 frames. ], giga_tot_loss[loss=0.3674, simple_loss=0.4189, pruned_loss=0.158, over 5701576.83 frames. ], batch size: 284, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:41:57,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0785, 1.3513, 1.0966, 1.0364], device='cuda:0'), covar=tensor([0.1124, 0.0694, 0.0919, 0.1676], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0449, 0.0366, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0013], device='cuda:0')
+2023-02-28 20:41:58,051 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36317.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:42:27,030 INFO [train.py:968] (0/2) Epoch 1, batch 36350, giga_loss[loss=0.3702, simple_loss=0.4239, pruned_loss=0.1582, over 28816.00 frames. ], tot_loss[loss=0.3734, simple_loss=0.4224, pruned_loss=0.1622, over 5706904.97 frames. ], libri_tot_loss[loss=0.4297, simple_loss=0.4477, pruned_loss=0.2059, over 5686891.67 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.4177, pruned_loss=0.1572, over 5706498.11 frames. ], batch size: 119, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:42:57,836 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-02-28 20:43:03,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.158e+02 1.267e+03 1.528e+03 1.884e+03 4.026e+03, threshold=3.055e+03, percent-clipped=3.0
+2023-02-28 20:43:15,439 INFO [train.py:968] (0/2) Epoch 1, batch 36400, giga_loss[loss=0.3918, simple_loss=0.4272, pruned_loss=0.1782, over 28586.00 frames. ], tot_loss[loss=0.3801, simple_loss=0.425, pruned_loss=0.1676, over 5701733.25 frames. ], libri_tot_loss[loss=0.4305, simple_loss=0.4484, pruned_loss=0.2063, over 5688363.07 frames. ], giga_tot_loss[loss=0.3728, simple_loss=0.4203, pruned_loss=0.1627, over 5700280.34 frames. ], batch size: 242, lr: 1.84e-02, grad_scale: 8.0
+2023-02-28 20:43:51,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36441.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:43:53,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36444.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:43:58,364 INFO [train.py:968] (0/2) Epoch 1, batch 36450, giga_loss[loss=0.3771, simple_loss=0.415, pruned_loss=0.1696, over 28899.00 frames. ], tot_loss[loss=0.3865, simple_loss=0.4278, pruned_loss=0.1726, over 5688054.34 frames. ], libri_tot_loss[loss=0.4306, simple_loss=0.4487, pruned_loss=0.2063, over 5681472.36 frames. ], giga_tot_loss[loss=0.3793, simple_loss=0.4232, pruned_loss=0.1677, over 5694052.43 frames. ], batch size: 112, lr: 1.84e-02, grad_scale: 8.0
+2023-02-28 20:44:18,798 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36473.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:44:33,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.972e+02 1.261e+03 1.579e+03 2.352e+03 7.126e+03, threshold=3.158e+03, percent-clipped=10.0
+2023-02-28 20:44:44,148 INFO [train.py:968] (0/2) Epoch 1, batch 36500, giga_loss[loss=0.3923, simple_loss=0.4228, pruned_loss=0.1809, over 28609.00 frames. ], tot_loss[loss=0.3849, simple_loss=0.4259, pruned_loss=0.1719, over 5696444.07 frames. ], libri_tot_loss[loss=0.431, simple_loss=0.4489, pruned_loss=0.2066, over 5682742.19 frames. ], giga_tot_loss[loss=0.3784, simple_loss=0.4218, pruned_loss=0.1675, over 5700073.65 frames. ], batch size: 71, lr: 1.84e-02, grad_scale: 8.0
+2023-02-28 20:45:10,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8194, 2.7948, 3.5152, 1.4541], device='cuda:0'), covar=tensor([0.0665, 0.0653, 0.1076, 0.2193], device='cuda:0'), in_proj_covar=tensor([0.0720, 0.0482, 0.0784, 0.0545], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0005, 0.0008, 0.0007], device='cuda:0')
+2023-02-28 20:45:21,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36539.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:45:29,985 INFO [train.py:968] (0/2) Epoch 1, batch 36550, giga_loss[loss=0.3703, simple_loss=0.4168, pruned_loss=0.1619, over 29009.00 frames. ], tot_loss[loss=0.3804, simple_loss=0.422, pruned_loss=0.1694, over 5698182.15 frames. ], libri_tot_loss[loss=0.4311, simple_loss=0.449, pruned_loss=0.2066, over 5685043.04 frames. ], giga_tot_loss[loss=0.3747, simple_loss=0.4185, pruned_loss=0.1655, over 5699011.90 frames. ], batch size: 164, lr: 1.84e-02, grad_scale: 8.0
+2023-02-28 20:45:53,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=36574.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:45:58,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36579.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:46:02,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9255, 1.7652, 1.2254, 1.4737], device='cuda:0'), covar=tensor([0.0615, 0.0672, 0.1081, 0.0968], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0567, 0.0589, 0.0541], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 20:46:04,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.569e+02 1.233e+03 1.549e+03 2.235e+03 4.608e+03, threshold=3.098e+03, percent-clipped=8.0
+2023-02-28 20:46:13,683 INFO [train.py:968] (0/2) Epoch 1, batch 36600, giga_loss[loss=0.4328, simple_loss=0.4562, pruned_loss=0.2047, over 27972.00 frames. ], tot_loss[loss=0.3809, simple_loss=0.4227, pruned_loss=0.1696, over 5699276.97 frames. ], libri_tot_loss[loss=0.4316, simple_loss=0.4494, pruned_loss=0.2069, over 5687769.86 frames. ], giga_tot_loss[loss=0.3747, simple_loss=0.4188, pruned_loss=0.1653, over 5697598.65 frames. ], batch size: 412, lr: 1.84e-02, grad_scale: 4.0
+2023-02-28 20:47:03,381 INFO [train.py:968] (0/2) Epoch 1, batch 36650, giga_loss[loss=0.343, simple_loss=0.3976, pruned_loss=0.1442, over 28970.00 frames. ], tot_loss[loss=0.3751, simple_loss=0.4192, pruned_loss=0.1655, over 5689259.08 frames. ], libri_tot_loss[loss=0.4317, simple_loss=0.4497, pruned_loss=0.2069, over 5689843.92 frames. ], giga_tot_loss[loss=0.3696, simple_loss=0.4157, pruned_loss=0.1618, over 5686177.96 frames. ], batch size: 106, lr: 1.84e-02, grad_scale: 4.0
+2023-02-28 20:47:27,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7713, 1.8933, 3.4068, 2.9333], device='cuda:0'), covar=tensor([0.1276, 0.1013, 0.0270, 0.0510], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0466, 0.0570, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:0')
+2023-02-28 20:47:32,749 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36682.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:47:32,870 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-02-28 20:47:36,964 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36685.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:47:39,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.158e+02 1.225e+03 1.556e+03 2.101e+03 5.681e+03, threshold=3.111e+03, percent-clipped=10.0
+2023-02-28 20:47:51,815 INFO [train.py:968] (0/2) Epoch 1, batch 36700, giga_loss[loss=0.3225, simple_loss=0.3808, pruned_loss=0.1321, over 28493.00 frames. ], tot_loss[loss=0.3687, simple_loss=0.4145, pruned_loss=0.1615, over 5681835.61 frames. ], libri_tot_loss[loss=0.4326, simple_loss=0.4504, pruned_loss=0.2074, over 5683028.07 frames. ], giga_tot_loss[loss=0.3629, simple_loss=0.4107, pruned_loss=0.1576, over 5685535.23 frames. ], batch size: 71, lr: 1.84e-02, grad_scale: 4.0
+2023-02-28 20:47:52,935 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2355, 1.1586, 1.1179, 1.1081], device='cuda:0'), covar=tensor([0.1778, 0.1831, 0.1466, 0.1817], device='cuda:0'), in_proj_covar=tensor([0.0877, 0.0762, 0.0842, 0.0898], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 20:48:05,098 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36714.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:48:12,118 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36722.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:48:15,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36725.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:48:38,450 INFO [train.py:968] (0/2) Epoch 1, batch 36750, giga_loss[loss=0.2882, simple_loss=0.3593, pruned_loss=0.1085, over 29095.00 frames. ], tot_loss[loss=0.3644, simple_loss=0.4097, pruned_loss=0.1595, over 5674223.73 frames. ], libri_tot_loss[loss=0.4342, simple_loss=0.4516, pruned_loss=0.2084, over 5688919.66 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4044, pruned_loss=0.1542, over 5671654.42 frames. ], batch size: 128, lr: 1.84e-02, grad_scale: 4.0
+2023-02-28 20:48:42,298 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36754.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:48:49,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5059, 1.8593, 1.5235, 1.2789], device='cuda:0'), covar=tensor([0.1115, 0.0822, 0.0997, 0.1708], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0448, 0.0365, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0012, 0.0010, 0.0013], device='cuda:0')
+2023-02-28 20:49:18,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.242e+02 1.120e+03 1.503e+03 2.194e+03 8.137e+03, threshold=3.006e+03, percent-clipped=10.0
+2023-02-28 20:49:32,519 INFO [train.py:968] (0/2) Epoch 1, batch 36800, giga_loss[loss=0.3333, simple_loss=0.3806, pruned_loss=0.143, over 27885.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.4038, pruned_loss=0.1565, over 5645180.98 frames. ], libri_tot_loss[loss=0.4349, simple_loss=0.4522, pruned_loss=0.2088, over 5673142.21 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.398, pruned_loss=0.1507, over 5657292.77 frames. ], batch size: 412, lr: 1.83e-02, grad_scale: 8.0
+2023-02-28 20:50:23,403 INFO [train.py:968] (0/2) Epoch 1, batch 36850, giga_loss[loss=0.3607, simple_loss=0.4073, pruned_loss=0.157, over 28564.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.3986, pruned_loss=0.1526, over 5646928.34 frames. ], libri_tot_loss[loss=0.4353, simple_loss=0.4525, pruned_loss=0.2091, over 5675095.87 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.3933, pruned_loss=0.1473, over 5654323.45 frames. ], batch size: 336, lr: 1.83e-02, grad_scale: 8.0
+2023-02-28 20:50:56,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.358e+02 1.103e+03 1.338e+03 1.779e+03 5.232e+03, threshold=2.677e+03, percent-clipped=5.0
+2023-02-28 20:51:08,468 INFO [train.py:968] (0/2) Epoch 1, batch 36900, giga_loss[loss=0.3086, simple_loss=0.3719, pruned_loss=0.1226, over 29093.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.3972, pruned_loss=0.1503, over 5660975.68 frames. ], libri_tot_loss[loss=0.4359, simple_loss=0.453, pruned_loss=0.2094, over 5675015.34 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3919, pruned_loss=0.1451, over 5666423.32 frames. ], batch size: 155, lr: 1.83e-02, grad_scale: 8.0
+2023-02-28 20:51:47,645 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-02-28 20:51:50,695 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36949.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:51:51,066 INFO [train.py:968] (0/2) Epoch 1, batch 36950, giga_loss[loss=0.2882, simple_loss=0.35, pruned_loss=0.1132, over 28845.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.3946, pruned_loss=0.1481, over 5669632.85 frames. ], libri_tot_loss[loss=0.4369, simple_loss=0.4538, pruned_loss=0.21, over 5674650.28 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3893, pruned_loss=0.1429, over 5674113.55 frames. ], batch size: 119, lr: 1.83e-02, grad_scale: 8.0
+2023-02-28 20:52:05,912 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-02-28 20:52:24,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.133e+02 1.081e+03 1.367e+03 1.887e+03 4.500e+03, threshold=2.733e+03, percent-clipped=6.0
+2023-02-28 20:52:25,590 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-02-28 20:52:32,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=36998.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:52:33,984 INFO [train.py:968] (0/2) Epoch 1, batch 37000, giga_loss[loss=0.3316, simple_loss=0.3783, pruned_loss=0.1424, over 28962.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.3937, pruned_loss=0.1483, over 5679144.77 frames. ], libri_tot_loss[loss=0.4384, simple_loss=0.455, pruned_loss=0.2109, over 5670659.93 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3865, pruned_loss=0.1415, over 5686943.03 frames. ], batch size: 106, lr: 1.83e-02, grad_scale: 4.0
+2023-02-28 20:52:35,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37001.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:52:36,291 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6478, 1.6764, 4.4724, 3.2824], device='cuda:0'), covar=tensor([0.1756, 0.1479, 0.0266, 0.0421], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0452, 0.0556, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:0')
+2023-02-28 20:52:38,179 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37005.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:53:13,895 INFO [train.py:968] (0/2) Epoch 1, batch 37050, giga_loss[loss=0.3323, simple_loss=0.3839, pruned_loss=0.1403, over 28798.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3899, pruned_loss=0.1458, over 5686960.56 frames. ], libri_tot_loss[loss=0.4391, simple_loss=0.4556, pruned_loss=0.2113, over 5661159.99 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3828, pruned_loss=0.1391, over 5701732.01 frames. ], batch size: 119, lr: 1.83e-02, grad_scale: 4.0
+2023-02-28 20:53:32,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6894, 1.7635, 4.1409, 2.9990], device='cuda:0'), covar=tensor([0.1597, 0.1359, 0.0256, 0.0454], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0464, 0.0573, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:0')
+2023-02-28 20:53:44,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.651e+02 1.136e+03 1.617e+03 2.482e+03 1.067e+04, threshold=3.234e+03, percent-clipped=18.0
+2023-02-28 20:53:46,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8422, 1.6779, 1.3231, 1.3599], device='cuda:0'), covar=tensor([0.0656, 0.0850, 0.1005, 0.1088], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0565, 0.0568, 0.0522], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-02-28 20:53:48,233 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37092.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:53:48,807 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37093.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:53:50,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37095.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:53:54,923 INFO [train.py:968] (0/2) Epoch 1, batch 37100, giga_loss[loss=0.3786, simple_loss=0.4053, pruned_loss=0.176, over 24150.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3879, pruned_loss=0.1452, over 5688346.06 frames. ], libri_tot_loss[loss=0.4404, simple_loss=0.4567, pruned_loss=0.212, over 5666124.19 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3798, pruned_loss=0.1378, over 5696240.64 frames. ], batch size: 705, lr: 1.83e-02, grad_scale: 4.0
+2023-02-28 20:54:00,333 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-02-28 20:54:12,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37124.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:54:33,807 INFO [train.py:968] (0/2) Epoch 1, batch 37150, giga_loss[loss=0.3032, simple_loss=0.3579, pruned_loss=0.1243, over 28905.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3862, pruned_loss=0.1443, over 5700592.95 frames. ], libri_tot_loss[loss=0.4416, simple_loss=0.4579, pruned_loss=0.2127, over 5671553.80 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3767, pruned_loss=0.136, over 5702847.89 frames. ], batch size: 136, lr: 1.83e-02, grad_scale: 4.0
+2023-02-28 20:54:44,816 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.06 vs. limit=2.0
+2023-02-28 20:55:06,114 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.205e+02 1.106e+03 1.418e+03 1.784e+03 5.027e+03, threshold=2.835e+03, percent-clipped=11.0
+2023-02-28 20:55:15,247 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37198.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:55:16,441 INFO [train.py:968] (0/2) Epoch 1, batch 37200, giga_loss[loss=0.2722, simple_loss=0.3414, pruned_loss=0.1015, over 29043.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3826, pruned_loss=0.1417, over 5705986.73 frames. ], libri_tot_loss[loss=0.4429, simple_loss=0.459, pruned_loss=0.2134, over 5665321.49 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3733, pruned_loss=0.1337, over 5713258.69 frames. ], batch size: 213, lr: 1.82e-02, grad_scale: 8.0
+2023-02-28 20:55:56,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3217, 1.4528, 1.3152, 1.1664], device='cuda:0'), covar=tensor([0.0928, 0.0984, 0.0871, 0.1429], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0447, 0.0361, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0013], device='cuda:0')
+2023-02-28 20:55:57,212 INFO [train.py:968] (0/2) Epoch 1, batch 37250, giga_loss[loss=0.393, simple_loss=0.4132, pruned_loss=0.1864, over 26664.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3812, pruned_loss=0.1408, over 5688506.21 frames. ], libri_tot_loss[loss=0.4443, simple_loss=0.4602, pruned_loss=0.2142, over 5646681.21 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3711, pruned_loss=0.1323, over 5712078.97 frames. ], batch size: 555, lr: 1.82e-02, grad_scale: 8.0
+2023-02-28 20:56:22,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3874, 1.7691, 1.4657, 0.5379], device='cuda:0'), covar=tensor([0.1070, 0.0832, 0.1074, 0.1629], device='cuda:0'), in_proj_covar=tensor([0.1015, 0.1018, 0.1045, 0.0927], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 20:56:27,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.572e+02 1.132e+03 1.405e+03 1.778e+03 5.423e+03, threshold=2.810e+03, percent-clipped=6.0
+2023-02-28 20:56:37,845 INFO [train.py:968] (0/2) Epoch 1, batch 37300, libri_loss[loss=0.4831, simple_loss=0.4833, pruned_loss=0.2414, over 29462.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3786, pruned_loss=0.1391, over 5696914.18 frames. ], libri_tot_loss[loss=0.445, simple_loss=0.4609, pruned_loss=0.2146, over 5651639.81 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.368, pruned_loss=0.1301, over 5712270.82 frames. ], batch size: 70, lr: 1.82e-02, grad_scale: 8.0
+2023-02-28 20:56:40,762 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37304.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:56:52,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5511, 1.5509, 1.2434, 1.2831], device='cuda:0'), covar=tensor([0.0782, 0.0724, 0.1051, 0.1058], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0570, 0.0577, 0.0534], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 20:56:58,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 20:57:01,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6600, 1.9702, 1.7178, 1.6550], device='cuda:0'), covar=tensor([0.1302, 0.1370, 0.1021, 0.0656], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0839, 0.0705, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0007, 0.0006], device='cuda:0')
+2023-02-28 20:57:19,704 INFO [train.py:968] (0/2) Epoch 1, batch 37350, giga_loss[loss=0.2987, simple_loss=0.3538, pruned_loss=0.1218, over 28786.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3769, pruned_loss=0.1377, over 5697244.68 frames. ], libri_tot_loss[loss=0.4457, simple_loss=0.4614, pruned_loss=0.215, over 5651659.73 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3671, pruned_loss=0.1293, over 5709894.70 frames. ], batch size: 119, lr: 1.82e-02, grad_scale: 8.0
+2023-02-28 20:57:20,568 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37351.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:57:39,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37373.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:57:41,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37376.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:57:44,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37380.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:57:52,736 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.420e+02 1.208e+03 1.563e+03 1.995e+03 5.395e+03, threshold=3.126e+03, percent-clipped=10.0
+2023-02-28 20:57:58,211 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-02-28 20:57:59,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37397.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:58:02,666 INFO [train.py:968] (0/2) Epoch 1, batch 37400, libri_loss[loss=0.435, simple_loss=0.4465, pruned_loss=0.2117, over 28513.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3774, pruned_loss=0.1377, over 5694290.19 frames. ], libri_tot_loss[loss=0.4471, simple_loss=0.4628, pruned_loss=0.2157, over 5648573.75 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3659, pruned_loss=0.1283, over 5708841.74 frames. ], batch size: 63, lr: 1.82e-02, grad_scale: 4.0
+2023-02-28 20:58:24,075 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-02-28 20:58:43,622 INFO [train.py:968] (0/2) Epoch 1, batch 37450, giga_loss[loss=0.3875, simple_loss=0.4184, pruned_loss=0.1783, over 28880.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3805, pruned_loss=0.1397, over 5707604.21 frames. ], libri_tot_loss[loss=0.4477, simple_loss=0.4635, pruned_loss=0.216, over 5651220.15 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3693, pruned_loss=0.1305, over 5717669.21 frames. ], batch size: 213, lr: 1.82e-02, grad_scale: 4.0
+2023-02-28 20:59:00,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37468.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:59:18,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.965e+02 1.205e+03 1.469e+03 1.855e+03 6.110e+03, threshold=2.938e+03, percent-clipped=5.0
+2023-02-28 20:59:26,369 INFO [train.py:968] (0/2) Epoch 1, batch 37500, giga_loss[loss=0.3826, simple_loss=0.4275, pruned_loss=0.1689, over 28916.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3888, pruned_loss=0.1459, over 5708871.82 frames. ], libri_tot_loss[loss=0.4476, simple_loss=0.4635, pruned_loss=0.2158, over 5657776.75 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3776, pruned_loss=0.1366, over 5712332.41 frames. ], batch size: 186, lr: 1.82e-02, grad_scale: 4.0
+2023-02-28 20:59:35,631 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-02-28 20:59:39,566 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37516.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:43,395 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37519.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:43,409 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37519.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:47,056 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37522.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:47,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37523.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:49,935 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37526.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:50,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3905, 1.4462, 1.3565, 1.3212], device='cuda:0'), covar=tensor([0.1205, 0.0566, 0.0563, 0.1517], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0272, 0.0268, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0017, 0.0015, 0.0025], device='cuda:0')
+2023-02-28 21:00:15,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37548.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:00:16,264 INFO [train.py:968] (0/2) Epoch 1, batch 37550, giga_loss[loss=0.4435, simple_loss=0.4636, pruned_loss=0.2116, over 28337.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.4005, pruned_loss=0.1559, over 5698661.70 frames. ], libri_tot_loss[loss=0.4477, simple_loss=0.4636, pruned_loss=0.2159, over 5661678.47 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3895, pruned_loss=0.1466, over 5699373.71 frames. ], batch size: 368, lr: 1.82e-02, grad_scale: 4.0
+2023-02-28 21:00:17,206 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37551.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:00:21,793 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37555.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:00:21,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0827, 1.8274, 1.7280, 1.7723], device='cuda:0'), covar=tensor([0.0768, 0.1331, 0.1063, 0.1062], device='cuda:0'), in_proj_covar=tensor([0.0517, 0.0860, 0.0658, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 21:00:26,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1427, 0.9180, 0.8212, 1.2795], device='cuda:0'), covar=tensor([0.1180, 0.0530, 0.0619, 0.1482], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0268, 0.0266, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0017, 0.0015, 0.0025], device='cuda:0')
+2023-02-28 21:00:37,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37573.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:00:52,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.984e+02 1.598e+03 2.075e+03 2.775e+03 7.257e+03, threshold=4.150e+03, percent-clipped=22.0
+2023-02-28 21:01:02,133 INFO [train.py:968] (0/2) Epoch 1, batch 37600, giga_loss[loss=0.379, simple_loss=0.4309, pruned_loss=0.1636, over 28597.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.4099, pruned_loss=0.1626, over 5691175.75 frames. ], libri_tot_loss[loss=0.448, simple_loss=0.4639, pruned_loss=0.216, over 5661870.10 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.3995, pruned_loss=0.1538, over 5692268.06 frames. ], batch size: 336, lr: 1.82e-02, grad_scale: 8.0
+2023-02-28 21:01:17,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37611.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:01:19,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37614.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:01:45,236 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37643.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:01:51,008 INFO [train.py:968] (0/2) Epoch 1, batch 37650, giga_loss[loss=0.3305, simple_loss=0.3975, pruned_loss=0.1318, over 28916.00 frames. ], tot_loss[loss=0.3738, simple_loss=0.4159, pruned_loss=0.1658, over 5689557.05 frames. ], libri_tot_loss[loss=0.4481, simple_loss=0.464, pruned_loss=0.2161, over 5670300.06 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4058, pruned_loss=0.157, over 5683561.21 frames. ], batch size: 145, lr: 1.81e-02, grad_scale: 8.0
+2023-02-28 21:01:52,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0791, 1.9192, 1.3681, 0.9847], device='cuda:0'), covar=tensor([0.0445, 0.0345, 0.0332, 0.0538], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0572, 0.0650, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 21:01:59,594 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37661.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:02:18,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37679.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:02:28,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.339e+02 1.235e+03 1.571e+03 2.191e+03 6.080e+03, threshold=3.143e+03, percent-clipped=7.0
+2023-02-28 21:02:37,466 INFO [train.py:968] (0/2) Epoch 1, batch 37700, libri_loss[loss=0.4354, simple_loss=0.461, pruned_loss=0.2049, over 28630.00 frames. ], tot_loss[loss=0.3792, simple_loss=0.4214, pruned_loss=0.1685, over 5680002.47 frames. ], libri_tot_loss[loss=0.4487, simple_loss=0.4645, pruned_loss=0.2164, over 5662521.77 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.4119, pruned_loss=0.1601, over 5681975.26 frames. ], batch size: 106, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:02:39,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6415, 2.5947, 3.4078, 1.5148], device='cuda:0'), covar=tensor([0.0618, 0.0717, 0.0779, 0.2067], device='cuda:0'), in_proj_covar=tensor([0.0706, 0.0478, 0.0761, 0.0522], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:02:55,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37716.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:02:57,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37719.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:03:02,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37726.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:03:04,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5396, 1.5041, 3.3329, 2.6346], device='cuda:0'), covar=tensor([0.2117, 0.1720, 0.0618, 0.1090], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0462, 0.0582, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:0')
+2023-02-28 21:03:18,788 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37748.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:03:19,725 INFO [train.py:968] (0/2) Epoch 1, batch 37750, libri_loss[loss=0.4501, simple_loss=0.453, pruned_loss=0.2236, over 19732.00 frames. ], tot_loss[loss=0.3886, simple_loss=0.4282, pruned_loss=0.1745, over 5680640.29 frames. ], libri_tot_loss[loss=0.4485, simple_loss=0.4643, pruned_loss=0.2164, over 5663047.92 frames. ], giga_tot_loss[loss=0.3753, simple_loss=0.419, pruned_loss=0.1658, over 5682978.97 frames. ], batch size: 187, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:03:26,093 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37757.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:03:39,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37772.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:03:55,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.617e+02 1.244e+03 1.547e+03 2.291e+03 6.019e+03, threshold=3.093e+03, percent-clipped=9.0
+2023-02-28 21:04:04,496 INFO [train.py:968] (0/2) Epoch 1, batch 37800, libri_loss[loss=0.3641, simple_loss=0.3893, pruned_loss=0.1694, over 29379.00 frames. ], tot_loss[loss=0.3798, simple_loss=0.4209, pruned_loss=0.1694, over 5682093.47 frames. ], libri_tot_loss[loss=0.4479, simple_loss=0.4637, pruned_loss=0.2161, over 5668073.96 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.4132, pruned_loss=0.1616, over 5679784.14 frames. ], batch size: 67, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:04:13,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9069, 1.5972, 1.7331, 1.4888], device='cuda:0'), covar=tensor([0.0772, 0.1762, 0.1161, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0865, 0.0657, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 21:04:23,335 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37822.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:04:25,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37825.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:04:45,914 INFO [train.py:968] (0/2) Epoch 1, batch 37850, giga_loss[loss=0.3486, simple_loss=0.402, pruned_loss=0.1475, over 28435.00 frames. ], tot_loss[loss=0.3702, simple_loss=0.4144, pruned_loss=0.1629, over 5697797.94 frames. ], libri_tot_loss[loss=0.448, simple_loss=0.4638, pruned_loss=0.2161, over 5672566.84 frames. ], giga_tot_loss[loss=0.3594, simple_loss=0.4072, pruned_loss=0.1558, over 5692224.04 frames. ], batch size: 65, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:04:49,058 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37854.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:04:58,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3919, 1.7852, 1.4735, 1.5217], device='cuda:0'), covar=tensor([0.1257, 0.1484, 0.1107, 0.0642], device='cuda:0'), in_proj_covar=tensor([0.0826, 0.0852, 0.0716, 0.0620], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:05:04,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37869.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:05:07,009 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37872.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:05:22,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.401e+02 1.168e+03 1.496e+03 2.189e+03 5.584e+03, threshold=2.992e+03, percent-clipped=13.0
+2023-02-28 21:05:31,212 INFO [train.py:968] (0/2) Epoch 1, batch 37900, giga_loss[loss=0.331, simple_loss=0.3892, pruned_loss=0.1364, over 29047.00 frames. ], tot_loss[loss=0.368, simple_loss=0.4134, pruned_loss=0.1613, over 5692264.32 frames. ], libri_tot_loss[loss=0.448, simple_loss=0.4639, pruned_loss=0.2161, over 5667413.05 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4065, pruned_loss=0.1544, over 5693171.97 frames. ], batch size: 136, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:05:32,216 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37901.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:05:42,826 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37915.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:05:45,581 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37918.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:06:10,254 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37947.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:06:13,495 INFO [train.py:968] (0/2) Epoch 1, batch 37950, giga_loss[loss=0.4153, simple_loss=0.4329, pruned_loss=0.1988, over 23675.00 frames. ], tot_loss[loss=0.3678, simple_loss=0.4136, pruned_loss=0.161, over 5688344.88 frames. ], libri_tot_loss[loss=0.4473, simple_loss=0.4632, pruned_loss=0.2157, over 5662779.55 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4079, pruned_loss=0.1549, over 5692798.07 frames. ], batch size: 705, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:06:27,376 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6044, 2.0640, 1.7190, 0.5542], device='cuda:0'), covar=tensor([0.1233, 0.0914, 0.1067, 0.1937], device='cuda:0'), in_proj_covar=tensor([0.1035, 0.1043, 0.1061, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 21:06:48,059 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.612e+02 1.211e+03 1.619e+03 2.109e+03 5.733e+03, threshold=3.238e+03, percent-clipped=8.0
+2023-02-28 21:06:57,769 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-38000.pt
+2023-02-28 21:06:58,077 INFO [train.py:968] (0/2) Epoch 1, batch 38000, giga_loss[loss=0.3865, simple_loss=0.4328, pruned_loss=0.1701, over 28937.00 frames. ], tot_loss[loss=0.3729, simple_loss=0.418, pruned_loss=0.1639, over 5696163.79 frames. ], libri_tot_loss[loss=0.4468, simple_loss=0.4629, pruned_loss=0.2154, over 5666958.58 frames. ], giga_tot_loss[loss=0.3648, simple_loss=0.4129, pruned_loss=0.1584, over 5696525.43 frames. ], batch size: 136, lr: 1.81e-02, grad_scale: 8.0
+2023-02-28 21:07:23,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1468, 3.1007, 3.8039, 1.9795], device='cuda:0'), covar=tensor([0.0502, 0.0588, 0.0916, 0.1732], device='cuda:0'), in_proj_covar=tensor([0.0711, 0.0486, 0.0775, 0.0517], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:07:29,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38036.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:07:42,112 INFO [train.py:968] (0/2) Epoch 1, batch 38050, giga_loss[loss=0.3547, simple_loss=0.4109, pruned_loss=0.1492, over 28289.00 frames. ], tot_loss[loss=0.3772, simple_loss=0.4213, pruned_loss=0.1666, over 5684799.68 frames. ], libri_tot_loss[loss=0.4483, simple_loss=0.4639, pruned_loss=0.2164, over 5654102.06 frames. ], giga_tot_loss[loss=0.3673, simple_loss=0.415, pruned_loss=0.1598, over 5697797.06 frames. ], batch size: 65, lr: 1.80e-02, grad_scale: 8.0
+2023-02-28 21:08:20,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.548e+02 1.311e+03 1.648e+03 2.295e+03 3.984e+03, threshold=3.297e+03, percent-clipped=7.0
+2023-02-28 21:08:28,421 INFO [train.py:968] (0/2) Epoch 1, batch 38100, giga_loss[loss=0.3894, simple_loss=0.4329, pruned_loss=0.173, over 28886.00 frames. ], tot_loss[loss=0.3808, simple_loss=0.4236, pruned_loss=0.169, over 5674965.05 frames. ], libri_tot_loss[loss=0.449, simple_loss=0.4644, pruned_loss=0.2168, over 5645711.89 frames. ], giga_tot_loss[loss=0.3715, simple_loss=0.4177, pruned_loss=0.1627, over 5692205.44 frames. ], batch size: 199, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:08:56,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38132.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:09:05,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6767, 2.0110, 1.7421, 1.5996], device='cuda:0'), covar=tensor([0.1191, 0.1256, 0.0911, 0.0610], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0851, 0.0719, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:09:12,473 INFO [train.py:968] (0/2) Epoch 1, batch 38150, libri_loss[loss=0.4111, simple_loss=0.4296, pruned_loss=0.1963, over 29483.00 frames. ], tot_loss[loss=0.3811, simple_loss=0.4236, pruned_loss=0.1693, over 5682194.09 frames. ], libri_tot_loss[loss=0.4488, simple_loss=0.4644, pruned_loss=0.2166, over 5650236.06 frames. ], giga_tot_loss[loss=0.372, simple_loss=0.4178, pruned_loss=0.1631, over 5692676.72 frames. ], batch size: 70, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:09:35,150 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38179.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:09:37,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38182.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:09:46,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.756e+02 1.417e+03 1.887e+03 2.475e+03 8.072e+03, threshold=3.775e+03, percent-clipped=11.0
+2023-02-28 21:09:54,056 INFO [train.py:968] (0/2) Epoch 1, batch 38200, giga_loss[loss=0.4022, simple_loss=0.4427, pruned_loss=0.1809, over 28875.00 frames. ], tot_loss[loss=0.3818, simple_loss=0.4241, pruned_loss=0.1697, over 5677458.09 frames. ], libri_tot_loss[loss=0.4486, simple_loss=0.4643, pruned_loss=0.2165, over 5647295.49 frames. ], giga_tot_loss[loss=0.3731, simple_loss=0.4186, pruned_loss=0.1639, over 5689164.10 frames. ], batch size: 186, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:10:04,619 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38211.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:10:04,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7849, 2.1628, 1.9073, 1.7272], device='cuda:0'), covar=tensor([0.1266, 0.1466, 0.1029, 0.0668], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0857, 0.0719, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:10:38,676 INFO [train.py:968] (0/2) Epoch 1, batch 38250, giga_loss[loss=0.3759, simple_loss=0.4389, pruned_loss=0.1565, over 28866.00 frames. ], tot_loss[loss=0.3795, simple_loss=0.4237, pruned_loss=0.1677, over 5686450.57 frames. ], libri_tot_loss[loss=0.4485, simple_loss=0.4641, pruned_loss=0.2164, over 5648640.01 frames. ], giga_tot_loss[loss=0.3726, simple_loss=0.4194, pruned_loss=0.1629, over 5694603.60 frames. ], batch size: 227, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:10:59,491 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38275.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:11:01,573 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38278.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:11:11,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.742e+02 1.235e+03 1.608e+03 2.333e+03 9.251e+03, threshold=3.216e+03, percent-clipped=6.0
+2023-02-28 21:11:19,012 INFO [train.py:968] (0/2) Epoch 1, batch 38300, giga_loss[loss=0.3215, simple_loss=0.3919, pruned_loss=0.1256, over 28556.00 frames. ], tot_loss[loss=0.3812, simple_loss=0.4255, pruned_loss=0.1684, over 5689404.29 frames. ], libri_tot_loss[loss=0.449, simple_loss=0.4644, pruned_loss=0.2168, over 5651347.85 frames. ], giga_tot_loss[loss=0.3719, simple_loss=0.4199, pruned_loss=0.1619, over 5695638.21 frames. ], batch size: 60, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:11:26,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38307.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:11:48,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3527, 1.7062, 1.5470, 1.4319], device='cuda:0'), covar=tensor([0.1239, 0.1479, 0.1102, 0.0681], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0867, 0.0729, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:11:57,581 INFO [train.py:968] (0/2) Epoch 1, batch 38350, giga_loss[loss=0.3701, simple_loss=0.4211, pruned_loss=0.1595, over 28678.00 frames. ], tot_loss[loss=0.3801, simple_loss=0.4255, pruned_loss=0.1674, over 5707601.04 frames. ], libri_tot_loss[loss=0.4479, simple_loss=0.4635, pruned_loss=0.2161, over 5661816.21 frames. ], giga_tot_loss[loss=0.3712, simple_loss=0.4204, pruned_loss=0.161, over 5704891.81 frames. ], batch size: 92, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:12:16,004 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=38372.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:12:32,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.697e+02 1.242e+03 1.459e+03 2.003e+03 1.096e+04, threshold=2.917e+03, percent-clipped=10.0
+2023-02-28 21:12:38,189 INFO [train.py:968] (0/2) Epoch 1, batch 38400, giga_loss[loss=0.3405, simple_loss=0.3881, pruned_loss=0.1464, over 29008.00 frames. ], tot_loss[loss=0.3764, simple_loss=0.4224, pruned_loss=0.1652, over 5707773.43 frames. ], libri_tot_loss[loss=0.4471, simple_loss=0.4629, pruned_loss=0.2156, over 5665670.32 frames. ], giga_tot_loss[loss=0.3687, simple_loss=0.4182, pruned_loss=0.1596, over 5703146.98 frames. ], batch size: 106, lr: 1.80e-02, grad_scale: 8.0
+2023-02-28 21:13:02,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4614, 1.2416, 1.3146, 0.8171], device='cuda:0'), covar=tensor([0.0462, 0.0329, 0.0249, 0.0421], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0573, 0.0605, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 21:13:21,419 INFO [train.py:968] (0/2) Epoch 1, batch 38450, giga_loss[loss=0.3074, simple_loss=0.373, pruned_loss=0.1209, over 28534.00 frames. ], tot_loss[loss=0.3719, simple_loss=0.419, pruned_loss=0.1624, over 5707586.40 frames. ], libri_tot_loss[loss=0.4473, simple_loss=0.4632, pruned_loss=0.2157, over 5667544.21 frames. ], giga_tot_loss[loss=0.3647, simple_loss=0.4148, pruned_loss=0.1573, over 5702642.10 frames. ], batch size: 92, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:13:47,470 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-02-28 21:13:58,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.566e+02 1.159e+03 1.481e+03 1.981e+03 6.538e+03, threshold=2.963e+03, percent-clipped=10.0
+2023-02-28 21:14:03,579 INFO [train.py:968] (0/2) Epoch 1, batch 38500, giga_loss[loss=0.3504, simple_loss=0.4022, pruned_loss=0.1493, over 28582.00 frames. ], tot_loss[loss=0.368, simple_loss=0.4157, pruned_loss=0.1601, over 5708679.38 frames. ], libri_tot_loss[loss=0.447, simple_loss=0.4629, pruned_loss=0.2156, over 5670970.82 frames. ], giga_tot_loss[loss=0.3615, simple_loss=0.412, pruned_loss=0.1555, over 5702311.13 frames. ], batch size: 336, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:14:49,049 INFO [train.py:968] (0/2) Epoch 1, batch 38550, giga_loss[loss=0.341, simple_loss=0.3995, pruned_loss=0.1412, over 28854.00 frames. ], tot_loss[loss=0.3661, simple_loss=0.4144, pruned_loss=0.1589, over 5705399.33 frames. ], libri_tot_loss[loss=0.447, simple_loss=0.4628, pruned_loss=0.2156, over 5671983.97 frames. ], giga_tot_loss[loss=0.3607, simple_loss=0.4114, pruned_loss=0.155, over 5699723.00 frames. ], batch size: 199, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:15:11,455 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 21:15:21,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.761e+02 1.125e+03 1.442e+03 1.968e+03 7.602e+03, threshold=2.884e+03, percent-clipped=10.0
+2023-02-28 21:15:26,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7615, 2.1661, 1.9184, 1.6608], device='cuda:0'), covar=tensor([0.1181, 0.1302, 0.0948, 0.0646], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0843, 0.0721, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:15:28,154 INFO [train.py:968] (0/2) Epoch 1, batch 38600, libri_loss[loss=0.5019, simple_loss=0.4886, pruned_loss=0.2576, over 19631.00 frames. ], tot_loss[loss=0.3686, simple_loss=0.4163, pruned_loss=0.1604, over 5696090.58 frames. ], libri_tot_loss[loss=0.4473, simple_loss=0.4631, pruned_loss=0.2158, over 5663431.22 frames. ], giga_tot_loss[loss=0.3609, simple_loss=0.4119, pruned_loss=0.155, over 5701424.75 frames. ], batch size: 186, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:15:40,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1529, 1.1491, 0.9966, 0.9434], device='cuda:0'), covar=tensor([0.0677, 0.0640, 0.1143, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0564, 0.0597, 0.0528], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 21:16:07,502 INFO [train.py:968] (0/2) Epoch 1, batch 38650, giga_loss[loss=0.3221, simple_loss=0.3885, pruned_loss=0.1278, over 28999.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.4157, pruned_loss=0.1586, over 5706295.23 frames. ], libri_tot_loss[loss=0.4468, simple_loss=0.4627, pruned_loss=0.2154, over 5667192.19 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4119, pruned_loss=0.1539, over 5707635.62 frames. ], batch size: 136, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:16:42,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.561e+02 9.557e+02 1.211e+03 1.580e+03 1.367e+04, threshold=2.422e+03, percent-clipped=10.0
+2023-02-28 21:16:47,516 INFO [train.py:968] (0/2) Epoch 1, batch 38700, giga_loss[loss=0.3447, simple_loss=0.4073, pruned_loss=0.1411, over 28406.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.413, pruned_loss=0.155, over 5709922.41 frames. ], libri_tot_loss[loss=0.447, simple_loss=0.4629, pruned_loss=0.2155, over 5666655.23 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4096, pruned_loss=0.151, over 5711577.37 frames. ], batch size: 78, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:17:11,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6795, 2.3082, 1.8621, 1.7464], device='cuda:0'), covar=tensor([0.1365, 0.1313, 0.1083, 0.0644], device='cuda:0'), in_proj_covar=tensor([0.0817, 0.0851, 0.0725, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:17:26,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38747.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:17:28,638 INFO [train.py:968] (0/2) Epoch 1, batch 38750, giga_loss[loss=0.3502, simple_loss=0.4093, pruned_loss=0.1455, over 28546.00 frames. ], tot_loss[loss=0.3662, simple_loss=0.4156, pruned_loss=0.1583, over 5700381.71 frames. ], libri_tot_loss[loss=0.4477, simple_loss=0.4636, pruned_loss=0.2159, over 5667934.51 frames. ], giga_tot_loss[loss=0.3571, simple_loss=0.41, pruned_loss=0.152, over 5702600.24 frames. ], batch size: 60, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:17:31,797 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=38753.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:18:02,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.819e+02 1.197e+03 1.547e+03 2.056e+03 7.177e+03, threshold=3.093e+03, percent-clipped=17.0
+2023-02-28 21:18:07,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5842, 1.5718, 3.3694, 2.7860], device='cuda:0'), covar=tensor([0.1479, 0.1298, 0.0324, 0.0529], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0453, 0.0577, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:0')
+2023-02-28 21:18:09,167 INFO [train.py:968] (0/2) Epoch 1, batch 38800, giga_loss[loss=0.4166, simple_loss=0.4439, pruned_loss=0.1947, over 27631.00 frames. ], tot_loss[loss=0.365, simple_loss=0.4142, pruned_loss=0.1579, over 5693660.19 frames. ], libri_tot_loss[loss=0.4474, simple_loss=0.4635, pruned_loss=0.2157, over 5661959.44 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4079, pruned_loss=0.1508, over 5701975.60 frames. ], batch size: 472, lr: 1.79e-02, grad_scale: 8.0
+2023-02-28 21:18:49,237 INFO [train.py:968] (0/2) Epoch 1, batch 38850, giga_loss[loss=0.3255, simple_loss=0.3943, pruned_loss=0.1284, over 28913.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.4114, pruned_loss=0.1569, over 5697123.40 frames. ], libri_tot_loss[loss=0.4477, simple_loss=0.4637, pruned_loss=0.2159, over 5663500.00 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4054, pruned_loss=0.15, over 5702965.66 frames. ], batch size: 145, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:19:03,013 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-02-28 21:19:03,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5841, 1.5014, 1.1405, 1.2828], device='cuda:0'), covar=tensor([0.0772, 0.0861, 0.1228, 0.1091], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0559, 0.0596, 0.0526], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 21:19:06,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5906, 1.9928, 1.6986, 1.5448], device='cuda:0'), covar=tensor([0.1183, 0.1247, 0.1004, 0.0636], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0853, 0.0716, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:19:20,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38890.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:19:21,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.251e+02 1.247e+03 1.643e+03 2.394e+03 4.919e+03, threshold=3.286e+03, percent-clipped=11.0
+2023-02-28 21:19:21,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38893.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:19:26,633 INFO [train.py:968] (0/2) Epoch 1, batch 38900, giga_loss[loss=0.3897, simple_loss=0.4391, pruned_loss=0.1702, over 28304.00 frames. ], tot_loss[loss=0.361, simple_loss=0.4097, pruned_loss=0.1562, over 5705792.32 frames. ], libri_tot_loss[loss=0.4473, simple_loss=0.4635, pruned_loss=0.2155, over 5670460.21 frames. ], giga_tot_loss[loss=0.351, simple_loss=0.4034, pruned_loss=0.1493, over 5704958.40 frames. ], batch size: 368, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:19:37,351 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7712, 1.5483, 1.2805, 1.4081], device='cuda:0'), covar=tensor([0.0679, 0.0787, 0.1103, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0561, 0.0593, 0.0528], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 21:19:44,223 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38922.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:20:08,016 INFO [train.py:968] (0/2) Epoch 1, batch 38950, giga_loss[loss=0.3645, simple_loss=0.4132, pruned_loss=0.1579, over 29073.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.412, pruned_loss=0.1592, over 5707090.07 frames. ], libri_tot_loss[loss=0.4461, simple_loss=0.4628, pruned_loss=0.2147, over 5676813.92 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4053, pruned_loss=0.1519, over 5701954.69 frames. ], batch size: 128, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:20:29,334 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 21:20:43,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.973e+02 1.226e+03 1.547e+03 2.116e+03 4.851e+03, threshold=3.094e+03, percent-clipped=4.0
+2023-02-28 21:20:50,377 INFO [train.py:968] (0/2) Epoch 1, batch 39000, libri_loss[loss=0.4355, simple_loss=0.4599, pruned_loss=0.2055, over 29538.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.4093, pruned_loss=0.1577, over 5703781.84 frames. ], libri_tot_loss[loss=0.4457, simple_loss=0.4625, pruned_loss=0.2145, over 5670122.71 frames. ], giga_tot_loss[loss=0.3532, simple_loss=0.4036, pruned_loss=0.1514, over 5705427.70 frames. ], batch size: 82, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:20:50,382 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 21:20:56,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2072, 1.2648, 1.1824, 1.2019], device='cuda:0'), covar=tensor([0.2007, 0.2186, 0.1703, 0.2037], device='cuda:0'), in_proj_covar=tensor([0.0903, 0.0787, 0.0858, 0.0909], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 21:20:59,056 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.3048, simple_loss=0.3942, pruned_loss=0.1077, over 944034.00 frames. 
+2023-02-28 21:20:59,057 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 21:21:39,871 INFO [train.py:968] (0/2) Epoch 1, batch 39050, giga_loss[loss=0.3981, simple_loss=0.4342, pruned_loss=0.181, over 28376.00 frames. ], tot_loss[loss=0.3613, simple_loss=0.4077, pruned_loss=0.1574, over 5696617.12 frames. ], libri_tot_loss[loss=0.4456, simple_loss=0.4626, pruned_loss=0.2143, over 5666040.67 frames. ], giga_tot_loss[loss=0.3511, simple_loss=0.4012, pruned_loss=0.1505, over 5702834.43 frames. ], batch size: 368, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:22:13,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.826e+02 1.182e+03 1.516e+03 1.951e+03 8.999e+03, threshold=3.032e+03, percent-clipped=10.0
+2023-02-28 21:22:18,462 INFO [train.py:968] (0/2) Epoch 1, batch 39100, giga_loss[loss=0.3851, simple_loss=0.4209, pruned_loss=0.1746, over 28269.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4042, pruned_loss=0.1557, over 5691654.52 frames. ], libri_tot_loss[loss=0.4451, simple_loss=0.4622, pruned_loss=0.214, over 5660397.15 frames. ], giga_tot_loss[loss=0.3482, simple_loss=0.3981, pruned_loss=0.1492, over 5702414.69 frames. ], batch size: 368, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:22:41,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39128.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:22:59,855 INFO [train.py:968] (0/2) Epoch 1, batch 39150, giga_loss[loss=0.2965, simple_loss=0.3629, pruned_loss=0.115, over 28856.00 frames. ], tot_loss[loss=0.354, simple_loss=0.4009, pruned_loss=0.1536, over 5700696.91 frames. ], libri_tot_loss[loss=0.4452, simple_loss=0.4623, pruned_loss=0.214, over 5661195.97 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3946, pruned_loss=0.147, over 5709373.82 frames. ], batch size: 174, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:23:35,817 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.701e+02 1.176e+03 1.577e+03 2.480e+03 7.431e+03, threshold=3.155e+03, percent-clipped=17.0
+2023-02-28 21:23:40,394 INFO [train.py:968] (0/2) Epoch 1, batch 39200, giga_loss[loss=0.4419, simple_loss=0.4574, pruned_loss=0.2132, over 28053.00 frames. ], tot_loss[loss=0.3571, simple_loss=0.4026, pruned_loss=0.1558, over 5706137.25 frames. ], libri_tot_loss[loss=0.4446, simple_loss=0.4619, pruned_loss=0.2137, over 5670719.42 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3956, pruned_loss=0.1486, over 5705806.34 frames. ], batch size: 412, lr: 1.78e-02, grad_scale: 8.0
+2023-02-28 21:23:42,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8931, 0.9918, 1.0626, 0.4884], device='cuda:0'), covar=tensor([0.0370, 0.0345, 0.0274, 0.0419], device='cuda:0'), in_proj_covar=tensor([0.0855, 0.0594, 0.0664, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 21:23:55,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=39216.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:24:25,881 INFO [train.py:968] (0/2) Epoch 1, batch 39250, giga_loss[loss=0.3673, simple_loss=0.406, pruned_loss=0.1643, over 28766.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4042, pruned_loss=0.156, over 5712105.56 frames. ], libri_tot_loss[loss=0.4436, simple_loss=0.4611, pruned_loss=0.2131, over 5676979.52 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.3977, pruned_loss=0.1491, over 5707154.14 frames. ], batch size: 99, lr: 1.78e-02, grad_scale: 8.0
+2023-02-28 21:24:43,187 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39271.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:24:45,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39274.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:25:02,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.924e+02 1.103e+03 1.389e+03 1.934e+03 4.021e+03, threshold=2.778e+03, percent-clipped=3.0
+2023-02-28 21:25:08,029 INFO [train.py:968] (0/2) Epoch 1, batch 39300, giga_loss[loss=0.3167, simple_loss=0.386, pruned_loss=0.1237, over 28985.00 frames. ], tot_loss[loss=0.3575, simple_loss=0.405, pruned_loss=0.155, over 5716789.24 frames. ], libri_tot_loss[loss=0.4428, simple_loss=0.4605, pruned_loss=0.2125, over 5682850.65 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.3989, pruned_loss=0.1486, over 5708197.05 frames. ], batch size: 155, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:25:11,205 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39303.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:25:25,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4181, 2.1208, 1.9188, 1.9781], device='cuda:0'), covar=tensor([0.0415, 0.0492, 0.0706, 0.0741], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0568, 0.0597, 0.0536], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 21:25:49,585 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-02-28 21:25:51,134 INFO [train.py:968] (0/2) Epoch 1, batch 39350, giga_loss[loss=0.3154, simple_loss=0.379, pruned_loss=0.1259, over 28944.00 frames. ], tot_loss[loss=0.3595, simple_loss=0.4072, pruned_loss=0.1559, over 5702961.97 frames. ], libri_tot_loss[loss=0.4429, simple_loss=0.4606, pruned_loss=0.2126, over 5677934.08 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.4008, pruned_loss=0.1492, over 5700842.62 frames. ], batch size: 136, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:25:58,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7719, 2.0635, 1.8304, 1.2190], device='cuda:0'), covar=tensor([0.0339, 0.0331, 0.0285, 0.0459], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0595, 0.0669, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 21:26:26,902 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.763e+02 1.079e+03 1.485e+03 2.105e+03 1.366e+04, threshold=2.969e+03, percent-clipped=16.0
+2023-02-28 21:26:31,824 INFO [train.py:968] (0/2) Epoch 1, batch 39400, giga_loss[loss=0.3444, simple_loss=0.4037, pruned_loss=0.1425, over 28665.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.4081, pruned_loss=0.1559, over 5699048.22 frames. ], libri_tot_loss[loss=0.4414, simple_loss=0.4594, pruned_loss=0.2117, over 5675849.37 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.4019, pruned_loss=0.149, over 5699511.06 frames. ], batch size: 242, lr: 1.77e-02, grad_scale: 4.0
+2023-02-28 21:26:34,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6225, 1.4979, 1.5280, 1.4516], device='cuda:0'), covar=tensor([0.0686, 0.1183, 0.0863, 0.0916], device='cuda:0'), in_proj_covar=tensor([0.0514, 0.0850, 0.0646, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 21:27:16,994 INFO [train.py:968] (0/2) Epoch 1, batch 39450, giga_loss[loss=0.3583, simple_loss=0.4143, pruned_loss=0.1511, over 28699.00 frames. ], tot_loss[loss=0.3574, simple_loss=0.4068, pruned_loss=0.154, over 5691016.06 frames. ], libri_tot_loss[loss=0.4413, simple_loss=0.4594, pruned_loss=0.2116, over 5678125.20 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4015, pruned_loss=0.1481, over 5689568.14 frames. ], batch size: 284, lr: 1.77e-02, grad_scale: 4.0
+2023-02-28 21:27:53,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9294, 1.6983, 1.6838, 1.6400], device='cuda:0'), covar=tensor([0.0732, 0.1619, 0.1138, 0.1308], device='cuda:0'), in_proj_covar=tensor([0.0520, 0.0858, 0.0655, 0.0710], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 21:27:54,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.352e+02 1.218e+03 1.586e+03 1.970e+03 7.709e+03, threshold=3.172e+03, percent-clipped=11.0
+2023-02-28 21:28:00,545 INFO [train.py:968] (0/2) Epoch 1, batch 39500, giga_loss[loss=0.3278, simple_loss=0.3827, pruned_loss=0.1364, over 28129.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.4083, pruned_loss=0.1552, over 5700510.86 frames. ], libri_tot_loss[loss=0.4411, simple_loss=0.4595, pruned_loss=0.2114, over 5682396.95 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4031, pruned_loss=0.1497, over 5695918.58 frames. ], batch size: 77, lr: 1.77e-02, grad_scale: 4.0
+2023-02-28 21:28:27,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-02-28 21:28:37,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9744, 1.4100, 1.0399, 1.0228], device='cuda:0'), covar=tensor([0.1436, 0.0888, 0.1202, 0.2017], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0433, 0.0355, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0013], device='cuda:0')
+2023-02-28 21:28:38,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6139, 1.8622, 1.4905, 1.5077], device='cuda:0'), covar=tensor([0.1072, 0.0466, 0.0599, 0.1395], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0267, 0.0268, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0018, 0.0016, 0.0026], device='cuda:0')
+2023-02-28 21:28:41,299 INFO [train.py:968] (0/2) Epoch 1, batch 39550, giga_loss[loss=0.3484, simple_loss=0.3961, pruned_loss=0.1503, over 29007.00 frames. ], tot_loss[loss=0.3621, simple_loss=0.41, pruned_loss=0.1571, over 5698283.26 frames. ], libri_tot_loss[loss=0.4409, simple_loss=0.4594, pruned_loss=0.2112, over 5681247.39 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.4047, pruned_loss=0.1516, over 5696220.09 frames. ], batch size: 128, lr: 1.77e-02, grad_scale: 4.0
+2023-02-28 21:29:08,855 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=39578.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:29:18,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3755, 1.2324, 1.2057, 1.3599], device='cuda:0'), covar=tensor([0.1661, 0.1869, 0.1419, 0.2004], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0775, 0.0831, 0.0899], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:0')
+2023-02-28 21:29:21,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39591.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:29:23,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.487e+02 1.347e+03 1.641e+03 1.954e+03 4.732e+03, threshold=3.283e+03, percent-clipped=5.0
+2023-02-28 21:29:29,380 INFO [train.py:968] (0/2) Epoch 1, batch 39600, giga_loss[loss=0.3666, simple_loss=0.4183, pruned_loss=0.1574, over 28870.00 frames. ], tot_loss[loss=0.364, simple_loss=0.4115, pruned_loss=0.1582, over 5695630.43 frames. ], libri_tot_loss[loss=0.4405, simple_loss=0.4591, pruned_loss=0.211, over 5683877.11 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4067, pruned_loss=0.1531, over 5691869.91 frames. ], batch size: 186, lr: 1.77e-02, grad_scale: 8.0
+2023-02-28 21:29:30,609 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-02-28 21:29:54,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6175, 3.4765, 4.3365, 1.8002], device='cuda:0'), covar=tensor([0.0438, 0.0465, 0.0857, 0.1833], device='cuda:0'), in_proj_covar=tensor([0.0712, 0.0481, 0.0764, 0.0525], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:30:09,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9069, 2.3109, 1.7566, 1.3768], device='cuda:0'), covar=tensor([0.0329, 0.0320, 0.0320, 0.0462], device='cuda:0'), in_proj_covar=tensor([0.0840, 0.0580, 0.0659, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 21:30:10,803 INFO [train.py:968] (0/2) Epoch 1, batch 39650, libri_loss[loss=0.4134, simple_loss=0.4508, pruned_loss=0.1879, over 29547.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.4142, pruned_loss=0.16, over 5704886.83 frames. ], libri_tot_loss[loss=0.4399, simple_loss=0.4589, pruned_loss=0.2105, over 5689501.00 frames. ], giga_tot_loss[loss=0.3599, simple_loss=0.4095, pruned_loss=0.1551, over 5697088.05 frames. ], batch size: 83, lr: 1.77e-02, grad_scale: 8.0
+2023-02-28 21:30:19,988 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.94 vs. limit=5.0
+2023-02-28 21:30:34,680 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=39677.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:30:48,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.850e+02 1.288e+03 1.595e+03 2.136e+03 5.947e+03, threshold=3.191e+03, percent-clipped=6.0
+2023-02-28 21:30:52,728 INFO [train.py:968] (0/2) Epoch 1, batch 39700, giga_loss[loss=0.388, simple_loss=0.4302, pruned_loss=0.1729, over 28998.00 frames. ], tot_loss[loss=0.3676, simple_loss=0.4159, pruned_loss=0.1596, over 5717076.79 frames. ], libri_tot_loss[loss=0.4406, simple_loss=0.4595, pruned_loss=0.2108, over 5693212.95 frames. ], giga_tot_loss[loss=0.3601, simple_loss=0.411, pruned_loss=0.1547, over 5707633.92 frames. ], batch size: 136, lr: 1.77e-02, grad_scale: 8.0
+2023-02-28 21:31:22,504 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39734.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:31:24,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39737.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:31:35,541 INFO [train.py:968] (0/2) Epoch 1, batch 39750, giga_loss[loss=0.335, simple_loss=0.3905, pruned_loss=0.1398, over 28767.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.4153, pruned_loss=0.1589, over 5718357.45 frames. ], libri_tot_loss[loss=0.4399, simple_loss=0.4591, pruned_loss=0.2103, over 5696645.64 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4111, pruned_loss=0.1547, over 5708065.66 frames. ], batch size: 99, lr: 1.77e-02, grad_scale: 8.0
+2023-02-28 21:31:49,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39766.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:32:13,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.839e+02 1.389e+03 1.731e+03 2.436e+03 5.277e+03, threshold=3.463e+03, percent-clipped=12.0
+2023-02-28 21:32:18,032 INFO [train.py:968] (0/2) Epoch 1, batch 39800, giga_loss[loss=0.4005, simple_loss=0.4394, pruned_loss=0.1808, over 28852.00 frames. ], tot_loss[loss=0.3673, simple_loss=0.4158, pruned_loss=0.1594, over 5718899.20 frames. ], libri_tot_loss[loss=0.4393, simple_loss=0.4587, pruned_loss=0.2099, over 5699980.97 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4114, pruned_loss=0.1548, over 5708559.44 frames. ], batch size: 174, lr: 1.77e-02, grad_scale: 8.0
+2023-02-28 21:32:59,405 INFO [train.py:968] (0/2) Epoch 1, batch 39850, giga_loss[loss=0.3393, simple_loss=0.4034, pruned_loss=0.1376, over 28881.00 frames. ], tot_loss[loss=0.364, simple_loss=0.4136, pruned_loss=0.1572, over 5722214.95 frames. ], libri_tot_loss[loss=0.4394, simple_loss=0.4589, pruned_loss=0.21, over 5702053.98 frames. ], giga_tot_loss[loss=0.3578, simple_loss=0.4096, pruned_loss=0.153, over 5712309.91 frames. ], batch size: 145, lr: 1.76e-02, grad_scale: 8.0
+2023-02-28 21:33:04,610 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=39856.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:33:29,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6065, 2.0168, 1.5894, 1.6242], device='cuda:0'), covar=tensor([0.1351, 0.1374, 0.1141, 0.0678], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0824, 0.0705, 0.0600], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:33:34,305 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.610e+02 1.095e+03 1.525e+03 2.067e+03 6.921e+03, threshold=3.050e+03, percent-clipped=4.0
+2023-02-28 21:33:38,564 INFO [train.py:968] (0/2) Epoch 1, batch 39900, giga_loss[loss=0.3751, simple_loss=0.4238, pruned_loss=0.1632, over 28286.00 frames. ], tot_loss[loss=0.3622, simple_loss=0.4119, pruned_loss=0.1563, over 5715806.41 frames. ], libri_tot_loss[loss=0.4402, simple_loss=0.4595, pruned_loss=0.2104, over 5695056.17 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.4072, pruned_loss=0.1514, over 5713771.29 frames. ], batch size: 368, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:34:21,098 INFO [train.py:968] (0/2) Epoch 1, batch 39950, giga_loss[loss=0.3125, simple_loss=0.3784, pruned_loss=0.1232, over 28606.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.407, pruned_loss=0.153, over 5707926.99 frames. ], libri_tot_loss[loss=0.4404, simple_loss=0.4597, pruned_loss=0.2105, over 5688331.54 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.4025, pruned_loss=0.1484, over 5711944.63 frames. ], batch size: 336, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:34:23,591 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39953.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:34:58,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.008e+02 1.180e+03 1.471e+03 2.027e+03 5.947e+03, threshold=2.943e+03, percent-clipped=3.0
+2023-02-28 21:35:02,684 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-40000.pt
+2023-02-28 21:35:02,991 INFO [train.py:968] (0/2) Epoch 1, batch 40000, giga_loss[loss=0.3236, simple_loss=0.391, pruned_loss=0.1281, over 28946.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.4033, pruned_loss=0.1502, over 5716039.89 frames. ], libri_tot_loss[loss=0.4394, simple_loss=0.459, pruned_loss=0.2099, over 5694760.92 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.3988, pruned_loss=0.1456, over 5714157.01 frames. ], batch size: 213, lr: 1.76e-02, grad_scale: 8.0
+2023-02-28 21:35:29,537 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40032.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:35:37,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0765, 2.8088, 2.2083, 1.9300], device='cuda:0'), covar=tensor([0.1091, 0.1004, 0.0868, 0.0599], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0818, 0.0708, 0.0600], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:35:43,939 INFO [train.py:968] (0/2) Epoch 1, batch 40050, giga_loss[loss=0.4177, simple_loss=0.4603, pruned_loss=0.1876, over 27718.00 frames. ], tot_loss[loss=0.354, simple_loss=0.4059, pruned_loss=0.1511, over 5710204.06 frames. ], libri_tot_loss[loss=0.4399, simple_loss=0.4594, pruned_loss=0.2102, over 5693173.26 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.4008, pruned_loss=0.1458, over 5710866.21 frames. ], batch size: 472, lr: 1.76e-02, grad_scale: 8.0
+2023-02-28 21:35:45,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40052.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:36:25,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.431e+02 1.231e+03 1.525e+03 2.138e+03 6.925e+03, threshold=3.050e+03, percent-clipped=8.0
+2023-02-28 21:36:25,477 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40096.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:36:27,916 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40099.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:36:28,217 INFO [train.py:968] (0/2) Epoch 1, batch 40100, giga_loss[loss=0.3426, simple_loss=0.4122, pruned_loss=0.1365, over 28693.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.4085, pruned_loss=0.1523, over 5699651.05 frames. ], libri_tot_loss[loss=0.4408, simple_loss=0.46, pruned_loss=0.2108, over 5686606.06 frames. ], giga_tot_loss[loss=0.3482, simple_loss=0.403, pruned_loss=0.1467, over 5705544.33 frames. ], batch size: 262, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:36:52,976 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40128.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:37:08,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8696, 2.0921, 2.3424, 1.9452], device='cuda:0'), covar=tensor([0.0666, 0.1800, 0.0956, 0.1361], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0861, 0.0657, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 21:37:10,993 INFO [train.py:968] (0/2) Epoch 1, batch 40150, giga_loss[loss=0.3979, simple_loss=0.4322, pruned_loss=0.1818, over 28963.00 frames. ], tot_loss[loss=0.3588, simple_loss=0.4095, pruned_loss=0.154, over 5702378.04 frames. ], libri_tot_loss[loss=0.4412, simple_loss=0.4604, pruned_loss=0.211, over 5685086.06 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4041, pruned_loss=0.1485, over 5708413.45 frames. ], batch size: 213, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:37:14,150 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-02-28 21:37:18,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40160.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:37:45,758 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40195.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:37:46,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.287e+02 1.202e+03 1.594e+03 2.291e+03 5.341e+03, threshold=3.189e+03, percent-clipped=14.0
+2023-02-28 21:37:47,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40198.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:37:48,524 INFO [train.py:968] (0/2) Epoch 1, batch 40200, giga_loss[loss=0.3373, simple_loss=0.3914, pruned_loss=0.1416, over 29021.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.4099, pruned_loss=0.1566, over 5703045.50 frames. ], libri_tot_loss[loss=0.4405, simple_loss=0.4598, pruned_loss=0.2106, over 5683610.54 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4038, pruned_loss=0.15, over 5710308.10 frames. ], batch size: 213, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:37:53,948 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40208.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:38:07,935 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40227.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:38:11,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40231.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:38:28,923 INFO [train.py:968] (0/2) Epoch 1, batch 40250, giga_loss[loss=0.3084, simple_loss=0.3602, pruned_loss=0.1283, over 28795.00 frames. ], tot_loss[loss=0.3619, simple_loss=0.4088, pruned_loss=0.1575, over 5707728.04 frames. ], libri_tot_loss[loss=0.4397, simple_loss=0.4593, pruned_loss=0.21, over 5681617.41 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4021, pruned_loss=0.1503, over 5716350.93 frames. ], batch size: 119, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:39:00,489 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-02-28 21:39:07,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.456e+02 1.194e+03 1.503e+03 2.366e+03 6.662e+03, threshold=3.007e+03, percent-clipped=9.0
+2023-02-28 21:39:10,682 INFO [train.py:968] (0/2) Epoch 1, batch 40300, giga_loss[loss=0.3574, simple_loss=0.4038, pruned_loss=0.1555, over 28997.00 frames. ], tot_loss[loss=0.3611, simple_loss=0.4068, pruned_loss=0.1577, over 5706254.66 frames. ], libri_tot_loss[loss=0.4389, simple_loss=0.4588, pruned_loss=0.2095, over 5686777.44 frames. ], giga_tot_loss[loss=0.3515, simple_loss=0.4006, pruned_loss=0.1511, over 5709076.55 frames. ], batch size: 128, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:39:47,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5203, 2.0015, 1.6807, 0.4751], device='cuda:0'), covar=tensor([0.1499, 0.1088, 0.1320, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.1054, 0.1046, 0.1067, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 21:39:50,276 INFO [train.py:968] (0/2) Epoch 1, batch 40350, giga_loss[loss=0.3262, simple_loss=0.3859, pruned_loss=0.1333, over 28876.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.4073, pruned_loss=0.1589, over 5706331.54 frames. ], libri_tot_loss[loss=0.4391, simple_loss=0.4592, pruned_loss=0.2095, over 5688632.90 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.4003, pruned_loss=0.152, over 5707315.35 frames. ], batch size: 186, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:40:10,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40374.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:40:13,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40377.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:40:28,549 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.285e+02 1.138e+03 1.503e+03 2.011e+03 6.876e+03, threshold=3.007e+03, percent-clipped=9.0
+2023-02-28 21:40:31,332 INFO [train.py:968] (0/2) Epoch 1, batch 40400, giga_loss[loss=0.2887, simple_loss=0.3487, pruned_loss=0.1143, over 28932.00 frames. ], tot_loss[loss=0.3597, simple_loss=0.4045, pruned_loss=0.1574, over 5695397.72 frames. ], libri_tot_loss[loss=0.4383, simple_loss=0.4585, pruned_loss=0.209, over 5681755.59 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.3981, pruned_loss=0.151, over 5702052.97 frames. ], batch size: 106, lr: 1.75e-02, grad_scale: 8.0
+2023-02-28 21:40:36,984 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40406.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:40:37,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40407.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:41:12,992 INFO [train.py:968] (0/2) Epoch 1, batch 40450, giga_loss[loss=0.2906, simple_loss=0.3511, pruned_loss=0.1151, over 28879.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.3997, pruned_loss=0.1545, over 5695724.32 frames. ], libri_tot_loss[loss=0.4387, simple_loss=0.4588, pruned_loss=0.2093, over 5676405.64 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.3932, pruned_loss=0.1479, over 5706542.40 frames. ], batch size: 145, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:41:49,825 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.430e+02 1.215e+03 1.553e+03 2.209e+03 5.537e+03, threshold=3.106e+03, percent-clipped=12.0
+2023-02-28 21:41:52,523 INFO [train.py:968] (0/2) Epoch 1, batch 40500, giga_loss[loss=0.3482, simple_loss=0.3989, pruned_loss=0.1488, over 28805.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.3955, pruned_loss=0.152, over 5693821.96 frames. ], libri_tot_loss[loss=0.4375, simple_loss=0.458, pruned_loss=0.2086, over 5675628.86 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3881, pruned_loss=0.1448, over 5704707.11 frames. ], batch size: 285, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:42:19,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40535.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:42:29,478 INFO [train.py:968] (0/2) Epoch 1, batch 40550, giga_loss[loss=0.3784, simple_loss=0.4264, pruned_loss=0.1652, over 28646.00 frames. ], tot_loss[loss=0.348, simple_loss=0.3944, pruned_loss=0.1508, over 5704754.08 frames. ], libri_tot_loss[loss=0.4374, simple_loss=0.4579, pruned_loss=0.2084, over 5677535.33 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3868, pruned_loss=0.1435, over 5712547.29 frames. ], batch size: 242, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:42:29,795 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40550.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:42:32,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40553.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:42:47,913 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-02-28 21:42:54,820 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40582.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:42:55,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40583.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:42:59,901 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40588.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:43:06,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.202e+02 1.390e+03 1.771e+03 2.461e+03 9.690e+03, threshold=3.541e+03, percent-clipped=13.0
+2023-02-28 21:43:09,847 INFO [train.py:968] (0/2) Epoch 1, batch 40600, giga_loss[loss=0.4078, simple_loss=0.4248, pruned_loss=0.1954, over 23873.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.3988, pruned_loss=0.1534, over 5700680.70 frames. ], libri_tot_loss[loss=0.4369, simple_loss=0.4575, pruned_loss=0.2082, over 5679064.14 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3907, pruned_loss=0.1458, over 5706956.81 frames. ], batch size: 705, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:43:14,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7514, 2.0974, 1.7144, 1.7453], device='cuda:0'), covar=tensor([0.1365, 0.1508, 0.1165, 0.0717], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0834, 0.0713, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:43:18,769 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40611.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:43:51,027 INFO [train.py:968] (0/2) Epoch 1, batch 40650, libri_loss[loss=0.4488, simple_loss=0.4697, pruned_loss=0.2139, over 29757.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4038, pruned_loss=0.1561, over 5700939.58 frames. ], libri_tot_loss[loss=0.4373, simple_loss=0.4579, pruned_loss=0.2084, over 5677670.57 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.3952, pruned_loss=0.148, over 5707095.85 frames. ], batch size: 87, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:44:02,764 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40665.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:06,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1922, 1.2400, 1.0731, 1.0115], device='cuda:0'), covar=tensor([0.0603, 0.0539, 0.0938, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0563, 0.0575, 0.0524], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 21:44:12,268 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40678.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:14,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40681.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:27,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.154e+02 1.363e+03 1.640e+03 2.145e+03 5.522e+03, threshold=3.281e+03, percent-clipped=5.0
+2023-02-28 21:44:28,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7303, 2.1885, 1.8012, 0.7422], device='cuda:0'), covar=tensor([0.0862, 0.0663, 0.0838, 0.1102], device='cuda:0'), in_proj_covar=tensor([0.1049, 0.1056, 0.1078, 0.0924], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 21:44:29,983 INFO [train.py:968] (0/2) Epoch 1, batch 40700, giga_loss[loss=0.3454, simple_loss=0.3989, pruned_loss=0.146, over 28374.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.4073, pruned_loss=0.1578, over 5703777.57 frames. ], libri_tot_loss[loss=0.4368, simple_loss=0.4574, pruned_loss=0.2081, over 5679170.53 frames. ], giga_tot_loss[loss=0.3503, simple_loss=0.3997, pruned_loss=0.1505, over 5707812.64 frames. ], batch size: 71, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:44:37,985 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40710.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:49,300 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40723.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:49,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3878, 1.6518, 1.3563, 1.4853], device='cuda:0'), covar=tensor([0.1235, 0.0495, 0.0582, 0.1459], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0263, 0.0263, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0018, 0.0016, 0.0026], device='cuda:0')
+2023-02-28 21:44:52,657 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40726.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:54,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40729.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:45:11,804 INFO [train.py:968] (0/2) Epoch 1, batch 40750, giga_loss[loss=0.3708, simple_loss=0.4234, pruned_loss=0.1591, over 28253.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.412, pruned_loss=0.1605, over 5692613.67 frames. ], libri_tot_loss[loss=0.4362, simple_loss=0.4571, pruned_loss=0.2076, over 5675920.10 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4046, pruned_loss=0.1535, over 5699330.58 frames. ], batch size: 368, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:45:19,713 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40758.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:45:39,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9911, 2.8284, 3.7530, 1.4789], device='cuda:0'), covar=tensor([0.0553, 0.0602, 0.0797, 0.1944], device='cuda:0'), in_proj_covar=tensor([0.0756, 0.0508, 0.0810, 0.0542], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-02-28 21:45:51,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.959e+02 1.295e+03 1.643e+03 2.380e+03 1.570e+04, threshold=3.286e+03, percent-clipped=5.0
+2023-02-28 21:45:54,088 INFO [train.py:968] (0/2) Epoch 1, batch 40800, giga_loss[loss=0.3364, simple_loss=0.4006, pruned_loss=0.1362, over 28939.00 frames. ], tot_loss[loss=0.3677, simple_loss=0.4134, pruned_loss=0.161, over 5701636.10 frames. ], libri_tot_loss[loss=0.4361, simple_loss=0.457, pruned_loss=0.2076, over 5677870.67 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.4066, pruned_loss=0.1543, over 5705529.05 frames. ], batch size: 174, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:46:45,121 INFO [train.py:968] (0/2) Epoch 1, batch 40850, giga_loss[loss=0.3804, simple_loss=0.4233, pruned_loss=0.1688, over 28794.00 frames. ], tot_loss[loss=0.3798, simple_loss=0.4211, pruned_loss=0.1693, over 5698760.18 frames. ], libri_tot_loss[loss=0.4359, simple_loss=0.4567, pruned_loss=0.2075, over 5683688.84 frames. ], giga_tot_loss[loss=0.3705, simple_loss=0.415, pruned_loss=0.163, over 5697372.63 frames. ], batch size: 199, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:47:20,873 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40885.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:47:31,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6593, 2.0838, 2.1817, 1.7410], device='cuda:0'), covar=tensor([0.0693, 0.1620, 0.0951, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0509, 0.0844, 0.0649, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 21:47:33,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.205e+03 1.827e+03 2.208e+03 3.037e+03 1.875e+04, threshold=4.416e+03, percent-clipped=22.0
+2023-02-28 21:47:35,463 INFO [train.py:968] (0/2) Epoch 1, batch 40900, giga_loss[loss=0.4318, simple_loss=0.4597, pruned_loss=0.2019, over 27892.00 frames. ], tot_loss[loss=0.3859, simple_loss=0.4257, pruned_loss=0.1731, over 5700023.71 frames. ], libri_tot_loss[loss=0.4356, simple_loss=0.4563, pruned_loss=0.2074, over 5683558.16 frames. ], giga_tot_loss[loss=0.3781, simple_loss=0.4207, pruned_loss=0.1677, over 5699258.44 frames. ], batch size: 412, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:47:48,185 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5766, 2.0830, 1.7711, 1.7395], device='cuda:0'), covar=tensor([0.1143, 0.1426, 0.1075, 0.0639], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0839, 0.0711, 0.0616], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 21:48:11,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1406, 1.2113, 1.1503, 0.9849], device='cuda:0'), covar=tensor([0.1210, 0.1213, 0.1024, 0.1113], device='cuda:0'), in_proj_covar=tensor([0.0911, 0.0782, 0.0851, 0.0912], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:0')
+2023-02-28 21:48:21,450 INFO [train.py:968] (0/2) Epoch 1, batch 40950, giga_loss[loss=0.3855, simple_loss=0.4332, pruned_loss=0.169, over 28857.00 frames. ], tot_loss[loss=0.3964, simple_loss=0.4336, pruned_loss=0.1797, over 5698491.42 frames. ], libri_tot_loss[loss=0.4351, simple_loss=0.456, pruned_loss=0.2071, over 5690100.45 frames. ], giga_tot_loss[loss=0.3892, simple_loss=0.429, pruned_loss=0.1747, over 5692630.05 frames. ], batch size: 199, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:48:34,610 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40963.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:48:56,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40986.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:49:04,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.392e+02 1.567e+03 1.962e+03 2.682e+03 4.382e+03, threshold=3.924e+03, percent-clipped=0.0
+2023-02-28 21:49:06,369 INFO [train.py:968] (0/2) Epoch 1, batch 41000, giga_loss[loss=0.434, simple_loss=0.4651, pruned_loss=0.2014, over 28903.00 frames. ], tot_loss[loss=0.4028, simple_loss=0.4381, pruned_loss=0.1837, over 5703769.97 frames. ], libri_tot_loss[loss=0.4344, simple_loss=0.4557, pruned_loss=0.2066, over 5696324.15 frames. ], giga_tot_loss[loss=0.3965, simple_loss=0.4341, pruned_loss=0.1795, over 5693622.01 frames. ], batch size: 186, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:49:44,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41040.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:49:49,040 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41044.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:49:53,499 INFO [train.py:968] (0/2) Epoch 1, batch 41050, giga_loss[loss=0.4487, simple_loss=0.4784, pruned_loss=0.2095, over 28642.00 frames. ], tot_loss[loss=0.4139, simple_loss=0.4459, pruned_loss=0.1909, over 5700182.05 frames. ], libri_tot_loss[loss=0.4346, simple_loss=0.456, pruned_loss=0.2066, over 5696969.70 frames. ], giga_tot_loss[loss=0.4079, simple_loss=0.442, pruned_loss=0.1869, over 5692111.01 frames. ], batch size: 262, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:50:39,530 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.836e+03 2.388e+03 2.803e+03 7.108e+03, threshold=4.776e+03, percent-clipped=11.0
+2023-02-28 21:50:40,115 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41098.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:50:41,433 INFO [train.py:968] (0/2) Epoch 1, batch 41100, giga_loss[loss=0.5867, simple_loss=0.5423, pruned_loss=0.3156, over 26536.00 frames. ], tot_loss[loss=0.4227, simple_loss=0.4515, pruned_loss=0.1969, over 5676178.05 frames. ], libri_tot_loss[loss=0.4346, simple_loss=0.4561, pruned_loss=0.2066, over 5693683.33 frames. ], giga_tot_loss[loss=0.4173, simple_loss=0.4481, pruned_loss=0.1933, over 5672254.78 frames. ], batch size: 555, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:50:47,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41106.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:50:51,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41109.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:51:02,565 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-02-28 21:51:02,587 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-02-28 21:51:15,055 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41129.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:51:16,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41132.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:51:24,855 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41138.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:51:35,878 INFO [train.py:968] (0/2) Epoch 1, batch 41150, giga_loss[loss=0.5398, simple_loss=0.5228, pruned_loss=0.2784, over 28061.00 frames. ], tot_loss[loss=0.4264, simple_loss=0.4535, pruned_loss=0.1997, over 5661062.50 frames. ], libri_tot_loss[loss=0.4345, simple_loss=0.4561, pruned_loss=0.2065, over 5688053.18 frames. ], giga_tot_loss[loss=0.4221, simple_loss=0.4507, pruned_loss=0.1967, over 5662512.73 frames. ], batch size: 412, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:51:47,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41161.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:52:15,063 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41183.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:52:19,129 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41186.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:52:31,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.099e+03 1.996e+03 2.735e+03 3.672e+03 1.034e+04, threshold=5.470e+03, percent-clipped=9.0
+2023-02-28 21:52:33,189 INFO [train.py:968] (0/2) Epoch 1, batch 41200, giga_loss[loss=0.4582, simple_loss=0.4701, pruned_loss=0.2231, over 28654.00 frames. ], tot_loss[loss=0.4341, simple_loss=0.4575, pruned_loss=0.2053, over 5656319.47 frames. ], libri_tot_loss[loss=0.4342, simple_loss=0.4557, pruned_loss=0.2064, over 5692446.11 frames. ], giga_tot_loss[loss=0.4308, simple_loss=0.4556, pruned_loss=0.203, over 5653019.60 frames. ], batch size: 242, lr: 1.74e-02, grad_scale: 8.0
+2023-02-28 21:52:46,256 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41215.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:52:58,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-02-28 21:53:11,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41241.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:53:15,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41244.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:53:20,094 INFO [train.py:968] (0/2) Epoch 1, batch 41250, giga_loss[loss=0.4767, simple_loss=0.483, pruned_loss=0.2352, over 28962.00 frames. ], tot_loss[loss=0.4393, simple_loss=0.4603, pruned_loss=0.2092, over 5642327.42 frames. ], libri_tot_loss[loss=0.4329, simple_loss=0.4548, pruned_loss=0.2055, over 5693226.58 frames. ], giga_tot_loss[loss=0.4379, simple_loss=0.4597, pruned_loss=0.2081, over 5636486.33 frames. ], batch size: 106, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:53:29,223 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41258.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:53:30,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41260.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:53:30,737 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 21:53:36,142 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-02-28 21:53:45,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41273.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:54:08,831 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41293.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:54:08,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1609, 2.2972, 2.9219, 1.2707], device='cuda:0'), covar=tensor([0.0759, 0.0700, 0.1065, 0.1917], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0512, 0.0830, 0.0547], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-02-28 21:54:13,067 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.197e+03 1.848e+03 2.243e+03 3.071e+03 5.537e+03, threshold=4.486e+03, percent-clipped=2.0
+2023-02-28 21:54:13,785 INFO [train.py:968] (0/2) Epoch 1, batch 41300, giga_loss[loss=0.4134, simple_loss=0.4616, pruned_loss=0.1826, over 28720.00 frames. ], tot_loss[loss=0.4431, simple_loss=0.4629, pruned_loss=0.2116, over 5642966.22 frames. ], libri_tot_loss[loss=0.4319, simple_loss=0.454, pruned_loss=0.2049, over 5700333.90 frames. ], giga_tot_loss[loss=0.4432, simple_loss=0.4635, pruned_loss=0.2115, over 5629974.55 frames. ], batch size: 119, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:54:59,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=2.05 vs. limit=2.0
+2023-02-28 21:55:05,725 INFO [train.py:968] (0/2) Epoch 1, batch 41350, giga_loss[loss=0.4339, simple_loss=0.4546, pruned_loss=0.2066, over 28518.00 frames. ], tot_loss[loss=0.444, simple_loss=0.4632, pruned_loss=0.2124, over 5638080.36 frames. ], libri_tot_loss[loss=0.431, simple_loss=0.4534, pruned_loss=0.2043, over 5704499.55 frames. ], giga_tot_loss[loss=0.4454, simple_loss=0.4646, pruned_loss=0.2132, over 5621483.01 frames. ], batch size: 336, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:55:44,452 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41390.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:55:52,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.860e+02 1.736e+03 2.244e+03 2.896e+03 1.041e+04, threshold=4.488e+03, percent-clipped=9.0
+2023-02-28 21:55:53,533 INFO [train.py:968] (0/2) Epoch 1, batch 41400, libri_loss[loss=0.5662, simple_loss=0.5419, pruned_loss=0.2953, over 29025.00 frames. ], tot_loss[loss=0.4441, simple_loss=0.4626, pruned_loss=0.2128, over 5635033.25 frames. ], libri_tot_loss[loss=0.4315, simple_loss=0.4537, pruned_loss=0.2046, over 5700417.02 frames. ], giga_tot_loss[loss=0.4451, simple_loss=0.4636, pruned_loss=0.2132, over 5623609.95 frames. ], batch size: 101, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:55:56,628 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41403.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:55:57,479 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.42 vs. limit=5.0
+2023-02-28 21:56:00,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41406.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:56:03,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1425, 1.3039, 1.0031, 0.9887], device='cuda:0'), covar=tensor([0.0600, 0.0491, 0.0915, 0.0761], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0572, 0.0586, 0.0534], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 21:56:11,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41419.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:56:21,721 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-02-28 21:56:29,893 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41435.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:56:34,082 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41438.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:56:47,333 INFO [train.py:968] (0/2) Epoch 1, batch 41450, giga_loss[loss=0.4378, simple_loss=0.4685, pruned_loss=0.2035, over 28997.00 frames. ], tot_loss[loss=0.4382, simple_loss=0.459, pruned_loss=0.2087, over 5651831.61 frames. ], libri_tot_loss[loss=0.4308, simple_loss=0.4532, pruned_loss=0.2043, over 5703707.31 frames. ], giga_tot_loss[loss=0.4397, simple_loss=0.4604, pruned_loss=0.2095, over 5639156.51 frames. ], batch size: 164, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:57:36,321 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41495.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:57:38,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.183e+02 1.591e+03 2.185e+03 3.068e+03 7.317e+03, threshold=4.369e+03, percent-clipped=3.0
+2023-02-28 21:57:40,727 INFO [train.py:968] (0/2) Epoch 1, batch 41500, giga_loss[loss=0.4016, simple_loss=0.4444, pruned_loss=0.1794, over 28615.00 frames. ], tot_loss[loss=0.4362, simple_loss=0.4589, pruned_loss=0.2067, over 5656526.55 frames. ], libri_tot_loss[loss=0.431, simple_loss=0.4531, pruned_loss=0.2045, over 5706763.84 frames. ], giga_tot_loss[loss=0.4374, simple_loss=0.4602, pruned_loss=0.2073, over 5643164.42 frames. ], batch size: 242, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:58:32,976 INFO [train.py:968] (0/2) Epoch 1, batch 41550, libri_loss[loss=0.4454, simple_loss=0.4713, pruned_loss=0.2097, over 29187.00 frames. ], tot_loss[loss=0.4361, simple_loss=0.4599, pruned_loss=0.2061, over 5672313.98 frames. ], libri_tot_loss[loss=0.431, simple_loss=0.4532, pruned_loss=0.2044, over 5710776.39 frames. ], giga_tot_loss[loss=0.4371, simple_loss=0.461, pruned_loss=0.2066, over 5656876.77 frames. ], batch size: 101, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:58:46,988 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41562.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:58:52,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41565.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:59:22,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41594.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:59:28,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.799e+03 2.386e+03 3.130e+03 6.260e+03, threshold=4.772e+03, percent-clipped=4.0
+2023-02-28 21:59:29,104 INFO [train.py:968] (0/2) Epoch 1, batch 41600, giga_loss[loss=0.3823, simple_loss=0.4337, pruned_loss=0.1654, over 28879.00 frames. ], tot_loss[loss=0.4359, simple_loss=0.4598, pruned_loss=0.206, over 5655947.42 frames. ], libri_tot_loss[loss=0.4308, simple_loss=0.453, pruned_loss=0.2043, over 5712636.35 frames. ], giga_tot_loss[loss=0.437, simple_loss=0.4609, pruned_loss=0.2066, over 5640875.43 frames. ], batch size: 186, lr: 1.73e-02, grad_scale: 8.0
+2023-02-28 22:00:03,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41633.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:00:18,654 INFO [train.py:968] (0/2) Epoch 1, batch 41650, giga_loss[loss=0.5266, simple_loss=0.5056, pruned_loss=0.2738, over 26647.00 frames. ], tot_loss[loss=0.432, simple_loss=0.4578, pruned_loss=0.2031, over 5658578.54 frames. ], libri_tot_loss[loss=0.4311, simple_loss=0.4533, pruned_loss=0.2044, over 5718384.18 frames. ], giga_tot_loss[loss=0.4328, simple_loss=0.4587, pruned_loss=0.2034, over 5639406.27 frames. ], batch size: 555, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 22:00:35,914 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41668.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:01:09,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.751e+02 1.769e+03 2.499e+03 3.114e+03 6.356e+03, threshold=4.999e+03, percent-clipped=2.0
+2023-02-28 22:01:09,902 INFO [train.py:968] (0/2) Epoch 1, batch 41700, giga_loss[loss=0.4087, simple_loss=0.449, pruned_loss=0.1842, over 28763.00 frames. ], tot_loss[loss=0.4265, simple_loss=0.4547, pruned_loss=0.1992, over 5658964.89 frames. ], libri_tot_loss[loss=0.4305, simple_loss=0.4528, pruned_loss=0.2041, over 5711383.23 frames. ], giga_tot_loss[loss=0.4277, simple_loss=0.4559, pruned_loss=0.1997, over 5648240.30 frames. ], batch size: 243, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 22:02:01,558 INFO [train.py:968] (0/2) Epoch 1, batch 41750, giga_loss[loss=0.4076, simple_loss=0.4231, pruned_loss=0.196, over 23656.00 frames. ], tot_loss[loss=0.4198, simple_loss=0.4503, pruned_loss=0.1947, over 5653316.39 frames. ], libri_tot_loss[loss=0.4301, simple_loss=0.4526, pruned_loss=0.2039, over 5703341.40 frames. ], giga_tot_loss[loss=0.4209, simple_loss=0.4515, pruned_loss=0.1952, over 5650157.58 frames. ], batch size: 705, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:02:09,674 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4066, 1.9845, 1.4216, 1.2141], device='cuda:0'), covar=tensor([0.0967, 0.0747, 0.0993, 0.1610], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0463, 0.0367, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0010, 0.0014], device='cuda:0')
+2023-02-28 22:02:18,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41765.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:02:30,382 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41776.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:02:32,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41779.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:02:53,801 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.380e+02 1.635e+03 2.217e+03 3.027e+03 7.605e+03, threshold=4.434e+03, percent-clipped=5.0
+2023-02-28 22:02:53,872 INFO [train.py:968] (0/2) Epoch 1, batch 41800, giga_loss[loss=0.4985, simple_loss=0.487, pruned_loss=0.255, over 26629.00 frames. ], tot_loss[loss=0.4152, simple_loss=0.447, pruned_loss=0.1917, over 5657501.58 frames. ], libri_tot_loss[loss=0.4304, simple_loss=0.4528, pruned_loss=0.204, over 5707222.30 frames. ], giga_tot_loss[loss=0.4156, simple_loss=0.4476, pruned_loss=0.1918, over 5650695.28 frames. ], batch size: 555, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:02:58,085 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 22:03:04,128 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41808.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:03:07,925 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41811.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:03:09,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41813.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:03:11,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41814.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:03:17,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2230, 1.1790, 1.1770, 1.3590], device='cuda:0'), covar=tensor([0.1612, 0.1624, 0.1310, 0.1600], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0809, 0.0869, 0.0928], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 22:03:40,845 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41843.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:03:47,540 INFO [train.py:968] (0/2) Epoch 1, batch 41850, giga_loss[loss=0.4251, simple_loss=0.4568, pruned_loss=0.1966, over 28844.00 frames. ], tot_loss[loss=0.4147, simple_loss=0.4465, pruned_loss=0.1914, over 5647010.52 frames. ], libri_tot_loss[loss=0.4301, simple_loss=0.4527, pruned_loss=0.2038, over 5699712.83 frames. ], giga_tot_loss[loss=0.4151, simple_loss=0.4471, pruned_loss=0.1915, over 5647465.69 frames. ], batch size: 199, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:04:05,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9563, 1.9231, 2.0663, 1.8444], device='cuda:0'), covar=tensor([0.0618, 0.1793, 0.1117, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0864, 0.0647, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 22:04:06,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41870.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:04:38,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.538e+03 1.894e+03 2.540e+03 5.101e+03, threshold=3.788e+03, percent-clipped=2.0
+2023-02-28 22:04:38,153 INFO [train.py:968] (0/2) Epoch 1, batch 41900, giga_loss[loss=0.3563, simple_loss=0.4126, pruned_loss=0.15, over 29022.00 frames. ], tot_loss[loss=0.4129, simple_loss=0.4456, pruned_loss=0.1902, over 5659550.37 frames. ], libri_tot_loss[loss=0.4301, simple_loss=0.4525, pruned_loss=0.2038, over 5701924.93 frames. ], giga_tot_loss[loss=0.4132, simple_loss=0.4461, pruned_loss=0.1901, over 5657660.44 frames. ], batch size: 155, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:04:49,445 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41908.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:04:51,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41911.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:05:23,528 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41940.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:05:34,809 INFO [train.py:968] (0/2) Epoch 1, batch 41950, giga_loss[loss=0.4547, simple_loss=0.4826, pruned_loss=0.2134, over 27947.00 frames. ], tot_loss[loss=0.4075, simple_loss=0.4423, pruned_loss=0.1863, over 5669042.80 frames. ], libri_tot_loss[loss=0.4297, simple_loss=0.4525, pruned_loss=0.2034, over 5704718.98 frames. ], giga_tot_loss[loss=0.4075, simple_loss=0.4425, pruned_loss=0.1862, over 5664083.69 frames. ], batch size: 412, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:05:39,587 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41956.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:05:41,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41959.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:06:11,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41988.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:06:21,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.114e+02 1.638e+03 2.184e+03 3.039e+03 6.467e+03, threshold=4.368e+03, percent-clipped=15.0
+2023-02-28 22:06:21,413 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-42000.pt
+2023-02-28 22:06:21,710 INFO [train.py:968] (0/2) Epoch 1, batch 42000, giga_loss[loss=0.3968, simple_loss=0.4506, pruned_loss=0.1715, over 28833.00 frames. ], tot_loss[loss=0.4056, simple_loss=0.4426, pruned_loss=0.1843, over 5680475.43 frames. ], libri_tot_loss[loss=0.4289, simple_loss=0.4519, pruned_loss=0.2029, over 5712525.89 frames. ], giga_tot_loss[loss=0.4052, simple_loss=0.4428, pruned_loss=0.1838, over 5667856.08 frames. ], batch size: 284, lr: 1.72e-02, grad_scale: 8.0
+2023-02-28 22:06:21,714 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 22:06:30,178 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.2815, simple_loss=0.3712, pruned_loss=0.09584, over 944034.00 frames. 
+2023-02-28 22:06:30,179 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19340MB
+2023-02-28 22:06:31,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7934, 1.6766, 3.7876, 3.0341], device='cuda:0'), covar=tensor([0.1482, 0.1327, 0.0315, 0.0530], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0464, 0.0604, 0.0471], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 22:06:40,911 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42013.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:06:43,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42016.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:07:13,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42045.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:07:16,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=42047.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:07:19,475 INFO [train.py:968] (0/2) Epoch 1, batch 42050, giga_loss[loss=0.4387, simple_loss=0.4669, pruned_loss=0.2052, over 28017.00 frames. ], tot_loss[loss=0.4065, simple_loss=0.4441, pruned_loss=0.1844, over 5675814.69 frames. ], libri_tot_loss[loss=0.4282, simple_loss=0.4514, pruned_loss=0.2025, over 5709858.13 frames. ], giga_tot_loss[loss=0.4061, simple_loss=0.4446, pruned_loss=0.1838, over 5666772.45 frames. ], batch size: 412, lr: 1.72e-02, grad_scale: 8.0
+2023-02-28 22:07:26,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8021, 1.5502, 3.2792, 2.7231], device='cuda:0'), covar=tensor([0.1332, 0.1253, 0.0371, 0.1221], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0467, 0.0600, 0.0470], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 22:07:31,097 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-02-28 22:08:08,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.781e+02 1.677e+03 2.104e+03 2.612e+03 5.221e+03, threshold=4.207e+03, percent-clipped=3.0
+2023-02-28 22:08:08,455 INFO [train.py:968] (0/2) Epoch 1, batch 42100, giga_loss[loss=0.5731, simple_loss=0.5342, pruned_loss=0.306, over 26573.00 frames. ], tot_loss[loss=0.4101, simple_loss=0.4461, pruned_loss=0.1871, over 5675884.27 frames. ], libri_tot_loss[loss=0.4284, simple_loss=0.4516, pruned_loss=0.2027, over 5712639.94 frames. ], giga_tot_loss[loss=0.4091, simple_loss=0.4461, pruned_loss=0.1861, over 5665475.33 frames. ], batch size: 555, lr: 1.72e-02, grad_scale: 8.0
+2023-02-28 22:08:54,208 INFO [train.py:968] (0/2) Epoch 1, batch 42150, giga_loss[loss=0.3702, simple_loss=0.4161, pruned_loss=0.1622, over 28847.00 frames. ], tot_loss[loss=0.4107, simple_loss=0.4461, pruned_loss=0.1877, over 5680252.38 frames. ], libri_tot_loss[loss=0.4278, simple_loss=0.4511, pruned_loss=0.2022, over 5714920.35 frames. ], giga_tot_loss[loss=0.41, simple_loss=0.4463, pruned_loss=0.1868, over 5669299.86 frames. ], batch size: 199, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:09:43,054 INFO [train.py:968] (0/2) Epoch 1, batch 42200, giga_loss[loss=0.39, simple_loss=0.4065, pruned_loss=0.1868, over 23391.00 frames. ], tot_loss[loss=0.4085, simple_loss=0.4432, pruned_loss=0.1869, over 5677649.77 frames. ], libri_tot_loss[loss=0.4266, simple_loss=0.4503, pruned_loss=0.2015, over 5717847.23 frames. ], giga_tot_loss[loss=0.4087, simple_loss=0.4441, pruned_loss=0.1867, over 5665566.26 frames. ], batch size: 705, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:09:43,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.802e+03 2.323e+03 2.835e+03 1.118e+04, threshold=4.646e+03, percent-clipped=6.0
+2023-02-28 22:10:06,302 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-02-28 22:10:21,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5613, 3.6053, 4.2383, 1.8042], device='cuda:0'), covar=tensor([0.0509, 0.0516, 0.0989, 0.1738], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0510, 0.0831, 0.0545], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-02-28 22:10:31,741 INFO [train.py:968] (0/2) Epoch 1, batch 42250, giga_loss[loss=0.3479, simple_loss=0.4009, pruned_loss=0.1474, over 28503.00 frames. ], tot_loss[loss=0.4083, simple_loss=0.4419, pruned_loss=0.1873, over 5672469.45 frames. ], libri_tot_loss[loss=0.4271, simple_loss=0.4507, pruned_loss=0.2018, over 5720290.90 frames. ], giga_tot_loss[loss=0.4073, simple_loss=0.442, pruned_loss=0.1864, over 5659171.91 frames. ], batch size: 60, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:11:23,499 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=42299.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:11:23,925 INFO [train.py:968] (0/2) Epoch 1, batch 42300, giga_loss[loss=0.3632, simple_loss=0.4291, pruned_loss=0.1486, over 28943.00 frames. ], tot_loss[loss=0.4045, simple_loss=0.4397, pruned_loss=0.1847, over 5679222.57 frames. ], libri_tot_loss[loss=0.4266, simple_loss=0.4503, pruned_loss=0.2015, over 5722711.04 frames. ], giga_tot_loss[loss=0.4039, simple_loss=0.4399, pruned_loss=0.184, over 5665948.88 frames. ], batch size: 145, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:11:27,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.550e+02 1.728e+03 2.331e+03 3.001e+03 6.768e+03, threshold=4.662e+03, percent-clipped=6.0
+2023-02-28 22:12:04,195 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=42341.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 22:12:11,150 INFO [train.py:968] (0/2) Epoch 1, batch 42350, giga_loss[loss=0.401, simple_loss=0.4377, pruned_loss=0.1821, over 28062.00 frames. ], tot_loss[loss=0.4002, simple_loss=0.4379, pruned_loss=0.1813, over 5689626.64 frames. ], libri_tot_loss[loss=0.4269, simple_loss=0.4506, pruned_loss=0.2016, over 5725111.41 frames. ], giga_tot_loss[loss=0.399, simple_loss=0.4376, pruned_loss=0.1802, over 5676203.72 frames. ], batch size: 412, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:12:43,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8455, 1.5235, 1.2547, 1.3344], device='cuda:0'), covar=tensor([0.0565, 0.0759, 0.1006, 0.0954], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0562, 0.0573, 0.0509], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-02-28 22:12:46,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3038, 1.5867, 1.2494, 1.3069], device='cuda:0'), covar=tensor([0.1201, 0.0526, 0.0596, 0.1510], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0263, 0.0263, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0019, 0.0017, 0.0027], device='cuda:0')
+2023-02-28 22:13:03,581 INFO [train.py:968] (0/2) Epoch 1, batch 42400, giga_loss[loss=0.4122, simple_loss=0.4507, pruned_loss=0.1869, over 28810.00 frames. ], tot_loss[loss=0.4001, simple_loss=0.4384, pruned_loss=0.1809, over 5690785.13 frames. ], libri_tot_loss[loss=0.4268, simple_loss=0.4506, pruned_loss=0.2015, over 5728796.11 frames. ], giga_tot_loss[loss=0.3987, simple_loss=0.4379, pruned_loss=0.1797, over 5676307.70 frames. ], batch size: 174, lr: 1.71e-02, grad_scale: 8.0
+2023-02-28 22:13:04,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.956e+02 1.406e+03 1.876e+03 2.240e+03 5.760e+03, threshold=3.752e+03, percent-clipped=1.0
+2023-02-28 22:13:25,442 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42422.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:13:52,118 INFO [train.py:968] (0/2) Epoch 1, batch 42450, libri_loss[loss=0.395, simple_loss=0.4286, pruned_loss=0.1807, over 29673.00 frames. ], tot_loss[loss=0.3979, simple_loss=0.4365, pruned_loss=0.1797, over 5683740.47 frames. ], libri_tot_loss[loss=0.4268, simple_loss=0.4505, pruned_loss=0.2015, over 5720125.58 frames. ], giga_tot_loss[loss=0.3962, simple_loss=0.4359, pruned_loss=0.1783, over 5677935.59 frames. ], batch size: 73, lr: 1.71e-02, grad_scale: 8.0
+2023-02-28 22:14:38,788 INFO [train.py:968] (0/2) Epoch 1, batch 42500, giga_loss[loss=0.4251, simple_loss=0.4533, pruned_loss=0.1984, over 28650.00 frames. ], tot_loss[loss=0.3977, simple_loss=0.4357, pruned_loss=0.1799, over 5681736.61 frames. ], libri_tot_loss[loss=0.4267, simple_loss=0.4505, pruned_loss=0.2015, over 5719563.85 frames. ], giga_tot_loss[loss=0.3961, simple_loss=0.4351, pruned_loss=0.1785, over 5677069.89 frames. ], batch size: 307, lr: 1.71e-02, grad_scale: 8.0
+2023-02-28 22:14:39,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.425e+03 1.876e+03 2.393e+03 5.973e+03, threshold=3.753e+03, percent-clipped=4.0
+2023-02-28 22:15:29,242 INFO [train.py:968] (0/2) Epoch 1, batch 42550, libri_loss[loss=0.4118, simple_loss=0.4469, pruned_loss=0.1884, over 29488.00 frames. ], tot_loss[loss=0.3999, simple_loss=0.4363, pruned_loss=0.1817, over 5671676.88 frames. ], libri_tot_loss[loss=0.4272, simple_loss=0.451, pruned_loss=0.2017, over 5719090.07 frames. ], giga_tot_loss[loss=0.3975, simple_loss=0.4351, pruned_loss=0.1799, over 5667407.06 frames. ], batch size: 85, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:15:41,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42565.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:15:44,411 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42568.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:16:12,153 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42597.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:16:15,580 INFO [train.py:968] (0/2) Epoch 1, batch 42600, giga_loss[loss=0.444, simple_loss=0.4627, pruned_loss=0.2127, over 28721.00 frames. ], tot_loss[loss=0.4013, simple_loss=0.4366, pruned_loss=0.1831, over 5670728.13 frames. ], libri_tot_loss[loss=0.4271, simple_loss=0.451, pruned_loss=0.2016, over 5722250.34 frames. ], giga_tot_loss[loss=0.3988, simple_loss=0.4351, pruned_loss=0.1812, over 5663244.56 frames. ], batch size: 262, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:16:18,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.957e+03 2.495e+03 3.394e+03 8.344e+03, threshold=4.991e+03, percent-clipped=21.0
+2023-02-28 22:17:04,759 INFO [train.py:968] (0/2) Epoch 1, batch 42650, libri_loss[loss=0.4052, simple_loss=0.4488, pruned_loss=0.1808, over 29176.00 frames. ], tot_loss[loss=0.3987, simple_loss=0.4343, pruned_loss=0.1815, over 5672030.04 frames. ], libri_tot_loss[loss=0.427, simple_loss=0.4511, pruned_loss=0.2015, over 5716410.31 frames. ], giga_tot_loss[loss=0.396, simple_loss=0.4327, pruned_loss=0.1796, over 5670261.49 frames. ], batch size: 97, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:17:29,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42674.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:17:37,944 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-02-28 22:17:57,323 INFO [train.py:968] (0/2) Epoch 1, batch 42700, libri_loss[loss=0.4531, simple_loss=0.4775, pruned_loss=0.2144, over 29145.00 frames. ], tot_loss[loss=0.3982, simple_loss=0.4337, pruned_loss=0.1813, over 5679587.10 frames. ], libri_tot_loss[loss=0.4273, simple_loss=0.4513, pruned_loss=0.2016, over 5719578.15 frames. ], giga_tot_loss[loss=0.3953, simple_loss=0.432, pruned_loss=0.1793, over 5674628.72 frames. ], batch size: 101, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:17:58,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.524e+02 1.534e+03 2.076e+03 2.649e+03 7.130e+03, threshold=4.152e+03, percent-clipped=2.0
+2023-02-28 22:18:13,453 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42716.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 22:18:14,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3063, 1.3077, 1.1481, 1.6076], device='cuda:0'), covar=tensor([0.1826, 0.1932, 0.1533, 0.2122], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0810, 0.0871, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 22:18:46,839 INFO [train.py:968] (0/2) Epoch 1, batch 42750, giga_loss[loss=0.3767, simple_loss=0.4308, pruned_loss=0.1613, over 29001.00 frames. ], tot_loss[loss=0.4021, simple_loss=0.4364, pruned_loss=0.1839, over 5683648.18 frames. ], libri_tot_loss[loss=0.4279, simple_loss=0.4516, pruned_loss=0.202, over 5722378.80 frames. ], giga_tot_loss[loss=0.3985, simple_loss=0.4342, pruned_loss=0.1814, over 5676245.64 frames. ], batch size: 155, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:19:05,335 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=42772.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:19:32,853 INFO [train.py:968] (0/2) Epoch 1, batch 42800, libri_loss[loss=0.4441, simple_loss=0.4658, pruned_loss=0.2113, over 19251.00 frames. ], tot_loss[loss=0.4006, simple_loss=0.4361, pruned_loss=0.1826, over 5671130.02 frames. ], libri_tot_loss[loss=0.4275, simple_loss=0.4513, pruned_loss=0.2019, over 5708876.17 frames. ], giga_tot_loss[loss=0.3973, simple_loss=0.4342, pruned_loss=0.1802, over 5676498.63 frames. ], batch size: 187, lr: 1.70e-02, grad_scale: 8.0
+2023-02-28 22:19:35,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.218e+02 1.710e+03 2.275e+03 3.157e+03 6.632e+03, threshold=4.551e+03, percent-clipped=10.0
+2023-02-28 22:19:49,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42817.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:19:53,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42820.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:20:20,566 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42849.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:20:20,967 INFO [train.py:968] (0/2) Epoch 1, batch 42850, giga_loss[loss=0.3627, simple_loss=0.4179, pruned_loss=0.1537, over 28971.00 frames. ], tot_loss[loss=0.4015, simple_loss=0.4372, pruned_loss=0.183, over 5674539.13 frames. ], libri_tot_loss[loss=0.4272, simple_loss=0.451, pruned_loss=0.2017, over 5711957.91 frames. ], giga_tot_loss[loss=0.3989, simple_loss=0.4358, pruned_loss=0.181, over 5675571.78 frames. ], batch size: 213, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:20:33,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42859.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 22:20:35,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42862.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 22:21:00,104 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42891.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 22:21:09,703 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=42899.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:21:10,248 INFO [train.py:968] (0/2) Epoch 1, batch 42900, libri_loss[loss=0.3657, simple_loss=0.397, pruned_loss=0.1672, over 29381.00 frames. ], tot_loss[loss=0.4031, simple_loss=0.4385, pruned_loss=0.1839, over 5668029.71 frames. ], libri_tot_loss[loss=0.4274, simple_loss=0.4511, pruned_loss=0.2018, over 5707415.52 frames. ], giga_tot_loss[loss=0.4002, simple_loss=0.437, pruned_loss=0.1817, over 5672440.78 frames. ], batch size: 67, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:21:10,841 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-02-28 22:21:12,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.789e+02 1.577e+03 2.208e+03 3.025e+03 7.152e+03, threshold=4.416e+03, percent-clipped=7.0
+2023-02-28 22:22:04,288 INFO [train.py:968] (0/2) Epoch 1, batch 42950, libri_loss[loss=0.4826, simple_loss=0.4918, pruned_loss=0.2367, over 29521.00 frames. ], tot_loss[loss=0.4059, simple_loss=0.44, pruned_loss=0.1859, over 5651502.30 frames. ], libri_tot_loss[loss=0.4281, simple_loss=0.4518, pruned_loss=0.2022, over 5697898.31 frames. ], giga_tot_loss[loss=0.4024, simple_loss=0.4379, pruned_loss=0.1834, over 5662726.98 frames. ], batch size: 89, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:22:51,472 INFO [train.py:968] (0/2) Epoch 1, batch 43000, libri_loss[loss=0.3863, simple_loss=0.4136, pruned_loss=0.1795, over 29361.00 frames. ], tot_loss[loss=0.4124, simple_loss=0.4437, pruned_loss=0.1906, over 5659602.48 frames. ], libri_tot_loss[loss=0.4275, simple_loss=0.4514, pruned_loss=0.2017, over 5704587.40 frames. ], giga_tot_loss[loss=0.4096, simple_loss=0.442, pruned_loss=0.1886, over 5661110.55 frames. ], batch size: 71, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:22:55,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.685e+02 1.649e+03 2.103e+03 3.146e+03 1.011e+04, threshold=4.206e+03, percent-clipped=13.0
+2023-02-28 22:23:37,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=43042.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:23:46,450 INFO [train.py:968] (0/2) Epoch 1, batch 43050, giga_loss[loss=0.3719, simple_loss=0.4145, pruned_loss=0.1646, over 28903.00 frames. ], tot_loss[loss=0.4141, simple_loss=0.4437, pruned_loss=0.1923, over 5654310.30 frames. ], libri_tot_loss[loss=0.4273, simple_loss=0.4513, pruned_loss=0.2017, over 5707768.33 frames. ], giga_tot_loss[loss=0.4117, simple_loss=0.4423, pruned_loss=0.1906, over 5651741.58 frames. ], batch size: 112, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:23:53,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7126, 1.7512, 3.4426, 2.7821], device='cuda:0'), covar=tensor([0.1368, 0.1224, 0.0338, 0.0699], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0482, 0.0633, 0.0482], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 22:23:53,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5658, 2.8837, 1.9605, 1.3929], device='cuda:0'), covar=tensor([0.0495, 0.0247, 0.0277, 0.0493], device='cuda:0'), in_proj_covar=tensor([0.0880, 0.0610, 0.0699, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 22:24:19,363 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-02-28 22:24:42,042 INFO [train.py:968] (0/2) Epoch 1, batch 43100, giga_loss[loss=0.6004, simple_loss=0.5689, pruned_loss=0.3159, over 24353.00 frames. ], tot_loss[loss=0.4156, simple_loss=0.4443, pruned_loss=0.1934, over 5654049.90 frames. ], libri_tot_loss[loss=0.4271, simple_loss=0.4512, pruned_loss=0.2015, over 5707849.98 frames. ], giga_tot_loss[loss=0.4137, simple_loss=0.4432, pruned_loss=0.1921, over 5650909.25 frames. ], batch size: 705, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:24:46,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.458e+02 1.738e+03 2.159e+03 3.019e+03 8.287e+03, threshold=4.317e+03, percent-clipped=10.0
+2023-02-28 22:25:26,917 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43147.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:25:28,680 INFO [train.py:968] (0/2) Epoch 1, batch 43150, giga_loss[loss=0.4357, simple_loss=0.4588, pruned_loss=0.2063, over 28769.00 frames. ], tot_loss[loss=0.4148, simple_loss=0.444, pruned_loss=0.1928, over 5667525.42 frames. ], libri_tot_loss[loss=0.4273, simple_loss=0.4514, pruned_loss=0.2016, over 5711187.08 frames. ], giga_tot_loss[loss=0.4127, simple_loss=0.4427, pruned_loss=0.1914, over 5660352.49 frames. ], batch size: 285, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:26:18,542 INFO [train.py:968] (0/2) Epoch 1, batch 43200, giga_loss[loss=0.3699, simple_loss=0.423, pruned_loss=0.1584, over 28657.00 frames. ], tot_loss[loss=0.4105, simple_loss=0.4409, pruned_loss=0.19, over 5667855.48 frames. ], libri_tot_loss[loss=0.4275, simple_loss=0.4516, pruned_loss=0.2017, over 5704112.86 frames. ], giga_tot_loss[loss=0.4084, simple_loss=0.4396, pruned_loss=0.1886, over 5668190.78 frames. ], batch size: 92, lr: 1.70e-02, grad_scale: 8.0
+2023-02-28 22:26:21,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.186e+02 1.619e+03 2.141e+03 2.845e+03 7.198e+03, threshold=4.281e+03, percent-clipped=5.0
+2023-02-28 22:27:06,257 INFO [train.py:968] (0/2) Epoch 1, batch 43250, giga_loss[loss=0.3384, simple_loss=0.3986, pruned_loss=0.1391, over 28939.00 frames. ], tot_loss[loss=0.4088, simple_loss=0.4409, pruned_loss=0.1883, over 5656870.65 frames. ], libri_tot_loss[loss=0.4292, simple_loss=0.4526, pruned_loss=0.2029, over 5686431.67 frames. ], giga_tot_loss[loss=0.405, simple_loss=0.4386, pruned_loss=0.1857, over 5673422.81 frames. ], batch size: 227, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:27:07,933 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=43252.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:27:25,000 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=43272.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:27:27,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43274.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:27:41,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 22:27:44,790 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43290.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:27:47,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43293.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:27:53,284 INFO [train.py:968] (0/2) Epoch 1, batch 43300, libri_loss[loss=0.3616, simple_loss=0.3976, pruned_loss=0.1628, over 29368.00 frames. ], tot_loss[loss=0.4013, simple_loss=0.435, pruned_loss=0.1838, over 5657497.74 frames. ], libri_tot_loss[loss=0.429, simple_loss=0.4523, pruned_loss=0.2028, over 5688870.66 frames. ], giga_tot_loss[loss=0.3982, simple_loss=0.4332, pruned_loss=0.1816, over 5667920.45 frames. ], batch size: 67, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:27:58,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.650e+03 2.384e+03 3.254e+03 9.080e+03, threshold=4.768e+03, percent-clipped=11.0
+2023-02-28 22:27:58,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4604, 1.2322, 1.2344, 1.6005], device='cuda:0'), covar=tensor([0.1767, 0.1990, 0.1558, 0.2093], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0810, 0.0868, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 22:28:04,352 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8009, 1.6532, 4.2602, 3.1723], device='cuda:0'), covar=tensor([0.1423, 0.1325, 0.0273, 0.0435], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0466, 0.0610, 0.0477], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 22:28:05,935 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-02-28 22:28:12,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43322.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:28:21,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9956, 1.8053, 1.4020, 1.3789], device='cuda:0'), covar=tensor([0.0497, 0.0601, 0.0864, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0565, 0.0580, 0.0523], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 22:28:37,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2233, 0.9850, 0.8320, 1.1658], device='cuda:0'), covar=tensor([0.1129, 0.0468, 0.0626, 0.1480], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0257, 0.0257, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0019, 0.0017, 0.0027], device='cuda:0')
+2023-02-28 22:28:37,827 INFO [train.py:968] (0/2) Epoch 1, batch 43350, libri_loss[loss=0.4419, simple_loss=0.4628, pruned_loss=0.2105, over 29532.00 frames. ], tot_loss[loss=0.3995, simple_loss=0.4329, pruned_loss=0.1831, over 5665769.22 frames. ], libri_tot_loss[loss=0.4265, simple_loss=0.4504, pruned_loss=0.2013, over 5701786.62 frames. ], giga_tot_loss[loss=0.3977, simple_loss=0.4322, pruned_loss=0.1815, over 5660672.57 frames. ], batch size: 80, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:29:21,883 INFO [train.py:968] (0/2) Epoch 1, batch 43400, giga_loss[loss=0.3909, simple_loss=0.4, pruned_loss=0.1909, over 23581.00 frames. ], tot_loss[loss=0.3991, simple_loss=0.4321, pruned_loss=0.1831, over 5672317.06 frames. ], libri_tot_loss[loss=0.4259, simple_loss=0.45, pruned_loss=0.2009, over 5704848.99 frames. ], giga_tot_loss[loss=0.3971, simple_loss=0.4312, pruned_loss=0.1815, over 5663972.18 frames. ], batch size: 705, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:29:26,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.813e+03 2.265e+03 2.871e+03 1.007e+04, threshold=4.529e+03, percent-clipped=8.0
+2023-02-28 22:29:40,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43417.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:29:40,836 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43417.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:29:42,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43420.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:30:08,297 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43449.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:30:09,371 INFO [train.py:968] (0/2) Epoch 1, batch 43450, giga_loss[loss=0.4202, simple_loss=0.4436, pruned_loss=0.1984, over 27935.00 frames. ], tot_loss[loss=0.4019, simple_loss=0.4337, pruned_loss=0.1851, over 5657015.54 frames. ], libri_tot_loss[loss=0.4253, simple_loss=0.4496, pruned_loss=0.2005, over 5692539.59 frames. ], giga_tot_loss[loss=0.4001, simple_loss=0.4329, pruned_loss=0.1837, over 5660519.38 frames. ], batch size: 412, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:30:48,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-02-28 22:30:59,444 INFO [train.py:968] (0/2) Epoch 1, batch 43500, giga_loss[loss=0.4124, simple_loss=0.4514, pruned_loss=0.1867, over 28728.00 frames. ], tot_loss[loss=0.4051, simple_loss=0.4372, pruned_loss=0.1866, over 5652493.28 frames. ], libri_tot_loss[loss=0.4249, simple_loss=0.4492, pruned_loss=0.2003, over 5685914.17 frames. ], giga_tot_loss[loss=0.4037, simple_loss=0.4367, pruned_loss=0.1854, over 5661261.32 frames. ], batch size: 262, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:31:02,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.682e+03 2.184e+03 3.116e+03 8.550e+03, threshold=4.368e+03, percent-clipped=7.0
+2023-02-28 22:31:23,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2142, 3.2308, 3.9398, 1.8496], device='cuda:0'), covar=tensor([0.0555, 0.0570, 0.0893, 0.1841], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0521, 0.0837, 0.0551], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-02-28 22:31:48,131 INFO [train.py:968] (0/2) Epoch 1, batch 43550, giga_loss[loss=0.3778, simple_loss=0.4339, pruned_loss=0.1608, over 29037.00 frames. ], tot_loss[loss=0.4056, simple_loss=0.4404, pruned_loss=0.1854, over 5653504.60 frames. ], libri_tot_loss[loss=0.425, simple_loss=0.4492, pruned_loss=0.2005, over 5689195.77 frames. ], giga_tot_loss[loss=0.4037, simple_loss=0.4396, pruned_loss=0.1839, over 5656676.73 frames. ], batch size: 213, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:31:57,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43560.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:32:01,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43563.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:32:31,738 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43592.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:32:38,214 INFO [train.py:968] (0/2) Epoch 1, batch 43600, giga_loss[loss=0.4294, simple_loss=0.4592, pruned_loss=0.1998, over 28747.00 frames. ], tot_loss[loss=0.4069, simple_loss=0.4425, pruned_loss=0.1856, over 5654187.36 frames. ], libri_tot_loss[loss=0.4245, simple_loss=0.4488, pruned_loss=0.2001, over 5685161.33 frames. ], giga_tot_loss[loss=0.4053, simple_loss=0.4421, pruned_loss=0.1843, over 5659029.41 frames. ], batch size: 262, lr: 1.69e-02, grad_scale: 8.0
+2023-02-28 22:32:43,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.258e+02 1.513e+03 1.969e+03 2.803e+03 1.254e+04, threshold=3.938e+03, percent-clipped=9.0
+2023-02-28 22:33:05,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43627.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:33:24,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43647.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:33:27,396 INFO [train.py:968] (0/2) Epoch 1, batch 43650, libri_loss[loss=0.3874, simple_loss=0.4321, pruned_loss=0.1713, over 27951.00 frames. ], tot_loss[loss=0.4078, simple_loss=0.4435, pruned_loss=0.1861, over 5666074.26 frames. ], libri_tot_loss[loss=0.4235, simple_loss=0.448, pruned_loss=0.1995, over 5688345.47 frames. ], giga_tot_loss[loss=0.4071, simple_loss=0.4437, pruned_loss=0.1852, over 5666689.80 frames. ], batch size: 116, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:34:16,781 INFO [train.py:968] (0/2) Epoch 1, batch 43700, giga_loss[loss=0.5425, simple_loss=0.5201, pruned_loss=0.2824, over 26767.00 frames. ], tot_loss[loss=0.4121, simple_loss=0.4461, pruned_loss=0.189, over 5667225.93 frames. ], libri_tot_loss[loss=0.4231, simple_loss=0.4479, pruned_loss=0.1991, over 5690972.97 frames. ], giga_tot_loss[loss=0.4114, simple_loss=0.4462, pruned_loss=0.1883, over 5665084.21 frames. ], batch size: 555, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:34:21,815 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.981e+02 1.736e+03 2.126e+03 2.642e+03 4.401e+03, threshold=4.253e+03, percent-clipped=5.0
+2023-02-28 22:35:00,833 INFO [train.py:968] (0/2) Epoch 1, batch 43750, giga_loss[loss=0.4069, simple_loss=0.4494, pruned_loss=0.1822, over 28854.00 frames. ], tot_loss[loss=0.4116, simple_loss=0.4452, pruned_loss=0.189, over 5669329.36 frames. ], libri_tot_loss[loss=0.4225, simple_loss=0.4475, pruned_loss=0.1988, over 5686647.42 frames. ], giga_tot_loss[loss=0.4112, simple_loss=0.4457, pruned_loss=0.1884, over 5670359.78 frames. ], batch size: 174, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:35:21,257 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43770.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:35:25,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43773.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:35:43,593 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43790.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:35:45,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43793.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:35:51,743 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7117, 1.6034, 1.5135, 0.9647], device='cuda:0'), covar=tensor([0.0541, 0.0382, 0.0294, 0.0492], device='cuda:0'), in_proj_covar=tensor([0.0907, 0.0593, 0.0697, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 22:35:52,886 INFO [train.py:968] (0/2) Epoch 1, batch 43800, giga_loss[loss=0.3858, simple_loss=0.4186, pruned_loss=0.1765, over 28576.00 frames. ], tot_loss[loss=0.4111, simple_loss=0.4438, pruned_loss=0.1892, over 5644267.30 frames. ], libri_tot_loss[loss=0.4227, simple_loss=0.4476, pruned_loss=0.1989, over 5671141.14 frames. ], giga_tot_loss[loss=0.4105, simple_loss=0.4441, pruned_loss=0.1885, over 5658505.16 frames. ], batch size: 85, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:35:55,366 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43802.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:35:57,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.751e+03 2.349e+03 3.307e+03 8.558e+03, threshold=4.698e+03, percent-clipped=12.0
+2023-02-28 22:36:16,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43822.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:36:42,958 INFO [train.py:968] (0/2) Epoch 1, batch 43850, giga_loss[loss=0.4469, simple_loss=0.4418, pruned_loss=0.226, over 23520.00 frames. ], tot_loss[loss=0.41, simple_loss=0.4424, pruned_loss=0.1888, over 5651300.05 frames. ], libri_tot_loss[loss=0.4227, simple_loss=0.4477, pruned_loss=0.1988, over 5674896.43 frames. ], giga_tot_loss[loss=0.4093, simple_loss=0.4423, pruned_loss=0.1881, over 5658587.08 frames. ], batch size: 705, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:37:31,566 INFO [train.py:968] (0/2) Epoch 1, batch 43900, giga_loss[loss=0.4184, simple_loss=0.4241, pruned_loss=0.2064, over 23681.00 frames. ], tot_loss[loss=0.4088, simple_loss=0.4409, pruned_loss=0.1883, over 5656518.56 frames. ], libri_tot_loss[loss=0.4229, simple_loss=0.4479, pruned_loss=0.199, over 5680180.70 frames. ], giga_tot_loss[loss=0.4077, simple_loss=0.4405, pruned_loss=0.1874, over 5657028.19 frames. ], batch size: 705, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:37:37,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.142e+02 1.863e+03 2.296e+03 2.866e+03 5.779e+03, threshold=4.593e+03, percent-clipped=3.0
+2023-02-28 22:37:43,353 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=43907.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:38:20,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8783, 3.0662, 3.5977, 1.6141], device='cuda:0'), covar=tensor([0.0598, 0.0568, 0.0944, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0528, 0.0870, 0.0551], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0006, 0.0010, 0.0007], device='cuda:0')
+2023-02-28 22:38:24,117 INFO [train.py:968] (0/2) Epoch 1, batch 43950, giga_loss[loss=0.3935, simple_loss=0.436, pruned_loss=0.1755, over 28820.00 frames. ], tot_loss[loss=0.4116, simple_loss=0.4423, pruned_loss=0.1904, over 5653906.05 frames. ], libri_tot_loss[loss=0.4227, simple_loss=0.4477, pruned_loss=0.1989, over 5687950.13 frames. ], giga_tot_loss[loss=0.4104, simple_loss=0.4421, pruned_loss=0.1894, over 5646392.77 frames. ], batch size: 186, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:39:16,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6114, 1.4577, 3.5201, 2.7927], device='cuda:0'), covar=tensor([0.1394, 0.1322, 0.0326, 0.0680], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0485, 0.0640, 0.0496], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 22:39:17,578 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-44000.pt
+2023-02-28 22:39:18,876 INFO [train.py:968] (0/2) Epoch 1, batch 44000, giga_loss[loss=0.3646, simple_loss=0.4152, pruned_loss=0.157, over 28881.00 frames. ], tot_loss[loss=0.4096, simple_loss=0.4404, pruned_loss=0.1893, over 5657743.27 frames. ], libri_tot_loss[loss=0.4227, simple_loss=0.4476, pruned_loss=0.1989, over 5690176.06 frames. ], giga_tot_loss[loss=0.4086, simple_loss=0.4402, pruned_loss=0.1885, over 5649426.67 frames. ], batch size: 199, lr: 1.68e-02, grad_scale: 8.0
+2023-02-28 22:39:22,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.833e+03 2.446e+03 3.089e+03 7.012e+03, threshold=4.892e+03, percent-clipped=9.0
+2023-02-28 22:40:06,145 INFO [train.py:968] (0/2) Epoch 1, batch 44050, giga_loss[loss=0.4714, simple_loss=0.471, pruned_loss=0.2359, over 26622.00 frames. ], tot_loss[loss=0.4055, simple_loss=0.4375, pruned_loss=0.1868, over 5658924.34 frames. ], libri_tot_loss[loss=0.4226, simple_loss=0.4475, pruned_loss=0.1988, over 5683306.94 frames. ], giga_tot_loss[loss=0.4047, simple_loss=0.4372, pruned_loss=0.186, over 5658247.93 frames. ], batch size: 555, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:40:29,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-02-28 22:40:50,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3882, 1.2524, 1.2058, 1.4152], device='cuda:0'), covar=tensor([0.1776, 0.1960, 0.1574, 0.1921], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0804, 0.0873, 0.0933], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 22:40:53,722 INFO [train.py:968] (0/2) Epoch 1, batch 44100, giga_loss[loss=0.3865, simple_loss=0.4313, pruned_loss=0.1708, over 28241.00 frames. ], tot_loss[loss=0.402, simple_loss=0.4351, pruned_loss=0.1844, over 5661053.38 frames. ], libri_tot_loss[loss=0.4216, simple_loss=0.4468, pruned_loss=0.1982, over 5689184.55 frames. ], giga_tot_loss[loss=0.4018, simple_loss=0.4354, pruned_loss=0.1841, over 5654708.22 frames. ], batch size: 368, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:41:00,003 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.441e+03 2.021e+03 2.636e+03 6.413e+03, threshold=4.043e+03, percent-clipped=2.0
+2023-02-28 22:41:33,876 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.37 vs. limit=5.0
+2023-02-28 22:41:39,691 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-02-28 22:41:50,176 INFO [train.py:968] (0/2) Epoch 1, batch 44150, giga_loss[loss=0.3798, simple_loss=0.433, pruned_loss=0.1633, over 28857.00 frames. ], tot_loss[loss=0.4053, simple_loss=0.4383, pruned_loss=0.1862, over 5651817.02 frames. ], libri_tot_loss[loss=0.4215, simple_loss=0.4468, pruned_loss=0.1981, over 5690269.62 frames. ], giga_tot_loss[loss=0.4052, simple_loss=0.4385, pruned_loss=0.186, over 5645751.80 frames. ], batch size: 186, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:42:01,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2725, 1.6970, 1.3028, 0.3686], device='cuda:0'), covar=tensor([0.0852, 0.0687, 0.1018, 0.1282], device='cuda:0'), in_proj_covar=tensor([0.1085, 0.1077, 0.1094, 0.0933], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 22:42:18,242 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=44180.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:42:40,181 INFO [train.py:968] (0/2) Epoch 1, batch 44200, libri_loss[loss=0.4234, simple_loss=0.4568, pruned_loss=0.195, over 28633.00 frames. ], tot_loss[loss=0.4082, simple_loss=0.4403, pruned_loss=0.188, over 5655590.78 frames. ], libri_tot_loss[loss=0.4219, simple_loss=0.4472, pruned_loss=0.1983, over 5693346.28 frames. ], giga_tot_loss[loss=0.4074, simple_loss=0.44, pruned_loss=0.1874, over 5647524.79 frames. ], batch size: 106, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:42:43,910 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-02-28 22:42:45,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.439e+02 1.719e+03 2.280e+03 2.978e+03 9.588e+03, threshold=4.560e+03, percent-clipped=11.0
+2023-02-28 22:43:04,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9710, 2.1116, 2.4252, 1.7927], device='cuda:0'), covar=tensor([0.0678, 0.1361, 0.0888, 0.1194], device='cuda:0'), in_proj_covar=tensor([0.0513, 0.0841, 0.0654, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 22:43:30,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1923, 1.6317, 1.2386, 1.1804], device='cuda:0'), covar=tensor([0.0802, 0.0659, 0.0694, 0.1155], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0472, 0.0364, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:0')
+2023-02-28 22:43:30,460 INFO [train.py:968] (0/2) Epoch 1, batch 44250, giga_loss[loss=0.5668, simple_loss=0.5142, pruned_loss=0.3097, over 23597.00 frames. ], tot_loss[loss=0.4087, simple_loss=0.4412, pruned_loss=0.1881, over 5647647.51 frames. ], libri_tot_loss[loss=0.422, simple_loss=0.4472, pruned_loss=0.1984, over 5677594.48 frames. ], giga_tot_loss[loss=0.4077, simple_loss=0.4408, pruned_loss=0.1874, over 5656199.40 frames. ], batch size: 705, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:44:00,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44282.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:44:03,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3337, 1.7967, 5.2086, 3.7533], device='cuda:0'), covar=tensor([0.1309, 0.1346, 0.0227, 0.0388], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0472, 0.0615, 0.0477], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 22:44:16,174 INFO [train.py:968] (0/2) Epoch 1, batch 44300, giga_loss[loss=0.3453, simple_loss=0.4271, pruned_loss=0.1317, over 28973.00 frames. ], tot_loss[loss=0.406, simple_loss=0.4419, pruned_loss=0.185, over 5655400.01 frames. ], libri_tot_loss[loss=0.4228, simple_loss=0.4479, pruned_loss=0.1989, over 5675390.05 frames. ], giga_tot_loss[loss=0.404, simple_loss=0.4408, pruned_loss=0.1836, over 5663719.79 frames. ], batch size: 155, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:44:18,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=44302.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:44:23,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.435e+02 1.591e+03 1.922e+03 2.493e+03 5.762e+03, threshold=3.844e+03, percent-clipped=5.0
+2023-02-28 22:45:03,112 INFO [train.py:968] (0/2) Epoch 1, batch 44350, giga_loss[loss=0.4077, simple_loss=0.454, pruned_loss=0.1807, over 28812.00 frames. ], tot_loss[loss=0.4055, simple_loss=0.4433, pruned_loss=0.1838, over 5658943.34 frames. ], libri_tot_loss[loss=0.4232, simple_loss=0.4482, pruned_loss=0.1991, over 5676325.91 frames. ], giga_tot_loss[loss=0.4032, simple_loss=0.442, pruned_loss=0.1822, over 5664182.43 frames. ], batch size: 199, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:45:05,515 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=44352.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:45:55,212 INFO [train.py:968] (0/2) Epoch 1, batch 44400, giga_loss[loss=0.4249, simple_loss=0.4615, pruned_loss=0.1942, over 28176.00 frames. ], tot_loss[loss=0.41, simple_loss=0.4462, pruned_loss=0.1869, over 5643994.36 frames. ], libri_tot_loss[loss=0.423, simple_loss=0.448, pruned_loss=0.1989, over 5672137.69 frames. ], giga_tot_loss[loss=0.408, simple_loss=0.4452, pruned_loss=0.1854, over 5651187.59 frames. ], batch size: 77, lr: 1.67e-02, grad_scale: 8.0
+2023-02-28 22:46:00,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.334e+02 1.550e+03 1.977e+03 2.855e+03 9.348e+03, threshold=3.955e+03, percent-clipped=14.0
+2023-02-28 22:46:14,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6863, 1.4971, 1.1782, 1.2925], device='cuda:0'), covar=tensor([0.0574, 0.0710, 0.0862, 0.1113], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0554, 0.0571, 0.0528], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 22:46:18,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44425.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:46:22,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44428.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:46:41,338 INFO [train.py:968] (0/2) Epoch 1, batch 44450, giga_loss[loss=0.4596, simple_loss=0.4724, pruned_loss=0.2234, over 29007.00 frames. ], tot_loss[loss=0.4158, simple_loss=0.4493, pruned_loss=0.1911, over 5653413.01 frames. ], libri_tot_loss[loss=0.4224, simple_loss=0.4477, pruned_loss=0.1986, over 5674885.17 frames. ], giga_tot_loss[loss=0.4142, simple_loss=0.4488, pruned_loss=0.1898, over 5655843.81 frames. ], batch size: 120, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:46:51,008 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44457.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:47:31,075 INFO [train.py:968] (0/2) Epoch 1, batch 44500, giga_loss[loss=0.3767, simple_loss=0.4274, pruned_loss=0.163, over 29084.00 frames. ], tot_loss[loss=0.4184, simple_loss=0.4506, pruned_loss=0.1931, over 5656353.80 frames. ], libri_tot_loss[loss=0.4221, simple_loss=0.4475, pruned_loss=0.1984, over 5672632.32 frames. ], giga_tot_loss[loss=0.4172, simple_loss=0.4504, pruned_loss=0.192, over 5659071.96 frames. ], batch size: 128, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:47:37,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.849e+02 1.673e+03 2.367e+03 3.028e+03 7.365e+03, threshold=4.733e+03, percent-clipped=11.0
+2023-02-28 22:48:16,768 INFO [train.py:968] (0/2) Epoch 1, batch 44550, giga_loss[loss=0.4612, simple_loss=0.4796, pruned_loss=0.2214, over 28689.00 frames. ], tot_loss[loss=0.4164, simple_loss=0.4491, pruned_loss=0.1919, over 5661658.72 frames. ], libri_tot_loss[loss=0.4219, simple_loss=0.4474, pruned_loss=0.1981, over 5666017.17 frames. ], giga_tot_loss[loss=0.4156, simple_loss=0.4491, pruned_loss=0.1911, over 5669538.18 frames. ], batch size: 262, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:48:22,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44555.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:48:36,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5280, 2.1165, 1.5361, 1.3184], device='cuda:0'), covar=tensor([0.0934, 0.0688, 0.0887, 0.1469], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0485, 0.0370, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:0')
+2023-02-28 22:49:02,918 INFO [train.py:968] (0/2) Epoch 1, batch 44600, giga_loss[loss=0.4373, simple_loss=0.4643, pruned_loss=0.2051, over 28540.00 frames. ], tot_loss[loss=0.4153, simple_loss=0.4476, pruned_loss=0.1915, over 5659962.33 frames. ], libri_tot_loss[loss=0.421, simple_loss=0.4466, pruned_loss=0.1977, over 5673151.63 frames. ], giga_tot_loss[loss=0.4152, simple_loss=0.4484, pruned_loss=0.191, over 5659661.77 frames. ], batch size: 336, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:49:09,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.912e+03 2.442e+03 3.282e+03 1.005e+04, threshold=4.884e+03, percent-clipped=10.0
+2023-02-28 22:49:49,572 INFO [train.py:968] (0/2) Epoch 1, batch 44650, giga_loss[loss=0.372, simple_loss=0.4298, pruned_loss=0.1571, over 28819.00 frames. ], tot_loss[loss=0.411, simple_loss=0.4467, pruned_loss=0.1877, over 5656133.29 frames. ], libri_tot_loss[loss=0.421, simple_loss=0.4466, pruned_loss=0.1977, over 5656274.36 frames. ], giga_tot_loss[loss=0.4108, simple_loss=0.4473, pruned_loss=0.1871, over 5671516.26 frames. ], batch size: 285, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:50:18,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44677.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:50:36,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44698.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:50:37,827 INFO [train.py:968] (0/2) Epoch 1, batch 44700, giga_loss[loss=0.4332, simple_loss=0.4649, pruned_loss=0.2007, over 28966.00 frames. ], tot_loss[loss=0.4098, simple_loss=0.4469, pruned_loss=0.1863, over 5669593.95 frames. ], libri_tot_loss[loss=0.4214, simple_loss=0.447, pruned_loss=0.1979, over 5659566.17 frames. ], giga_tot_loss[loss=0.4091, simple_loss=0.447, pruned_loss=0.1856, over 5678841.86 frames. ], batch size: 112, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:50:39,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44701.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:50:44,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.792e+02 1.381e+03 1.849e+03 2.508e+03 5.721e+03, threshold=3.698e+03, percent-clipped=1.0
+2023-02-28 22:51:09,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44727.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:51:11,880 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44730.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:51:32,326 INFO [train.py:968] (0/2) Epoch 1, batch 44750, giga_loss[loss=0.4281, simple_loss=0.458, pruned_loss=0.1991, over 28822.00 frames. ], tot_loss[loss=0.4135, simple_loss=0.4487, pruned_loss=0.1891, over 5654304.90 frames. ], libri_tot_loss[loss=0.4213, simple_loss=0.4469, pruned_loss=0.1979, over 5660861.34 frames. ], giga_tot_loss[loss=0.413, simple_loss=0.4489, pruned_loss=0.1885, over 5660554.52 frames. ], batch size: 284, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:51:52,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3349, 1.5016, 1.3148, 1.7574], device='cuda:0'), covar=tensor([0.1584, 0.1536, 0.1294, 0.1663], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0787, 0.0864, 0.0929], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 22:52:22,839 INFO [train.py:968] (0/2) Epoch 1, batch 44800, giga_loss[loss=0.3876, simple_loss=0.4289, pruned_loss=0.1731, over 28518.00 frames. ], tot_loss[loss=0.4118, simple_loss=0.447, pruned_loss=0.1883, over 5653508.61 frames. ], libri_tot_loss[loss=0.4221, simple_loss=0.4476, pruned_loss=0.1983, over 5664515.50 frames. ], giga_tot_loss[loss=0.4106, simple_loss=0.4466, pruned_loss=0.1873, over 5655071.77 frames. ], batch size: 71, lr: 1.67e-02, grad_scale: 8.0
+2023-02-28 22:52:28,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.603e+02 1.536e+03 2.043e+03 2.623e+03 4.590e+03, threshold=4.086e+03, percent-clipped=9.0
+2023-02-28 22:52:45,550 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44820.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:52:47,563 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44823.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:52:50,695 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7626, 1.5248, 1.4447, 1.4655], device='cuda:0'), covar=tensor([0.0616, 0.1136, 0.1099, 0.0889], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0844, 0.0642, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 22:52:57,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4702, 1.5848, 2.9264, 2.6521], device='cuda:0'), covar=tensor([0.1289, 0.1104, 0.0419, 0.0509], device='cuda:0'), in_proj_covar=tensor([0.0493, 0.0478, 0.0627, 0.0483], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 22:53:04,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3285, 1.3324, 1.1650, 1.2786], device='cuda:0'), covar=tensor([0.0515, 0.0714, 0.0946, 0.0666], device='cuda:0'), in_proj_covar=tensor([0.0498, 0.0839, 0.0640, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 22:53:12,676 INFO [train.py:968] (0/2) Epoch 1, batch 44850, giga_loss[loss=0.4546, simple_loss=0.4652, pruned_loss=0.222, over 27916.00 frames. ], tot_loss[loss=0.4128, simple_loss=0.4464, pruned_loss=0.1896, over 5653102.06 frames. ], libri_tot_loss[loss=0.4226, simple_loss=0.448, pruned_loss=0.1986, over 5655497.12 frames. ], giga_tot_loss[loss=0.4112, simple_loss=0.4457, pruned_loss=0.1883, over 5662877.94 frames. ], batch size: 412, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:53:13,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7807, 1.5953, 1.1529, 1.4432], device='cuda:0'), covar=tensor([0.0625, 0.0726, 0.1092, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0544, 0.0572, 0.0515], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 22:53:14,436 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44852.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:53:36,240 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44870.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:53:38,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44873.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:54:01,555 INFO [train.py:968] (0/2) Epoch 1, batch 44900, giga_loss[loss=0.359, simple_loss=0.4121, pruned_loss=0.1529, over 28845.00 frames. ], tot_loss[loss=0.4122, simple_loss=0.445, pruned_loss=0.1897, over 5642834.65 frames. ], libri_tot_loss[loss=0.4232, simple_loss=0.4486, pruned_loss=0.1989, over 5650631.61 frames. ], giga_tot_loss[loss=0.4101, simple_loss=0.4438, pruned_loss=0.1882, over 5655016.60 frames. ], batch size: 199, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:54:04,084 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44902.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:54:09,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.626e+03 2.076e+03 2.603e+03 9.336e+03, threshold=4.152e+03, percent-clipped=9.0
+2023-02-28 22:54:21,305 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=44920.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:54:21,503 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-02-28 22:54:24,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2805, 1.6708, 1.3181, 0.2942], device='cuda:0'), covar=tensor([0.1054, 0.0796, 0.1094, 0.1584], device='cuda:0'), in_proj_covar=tensor([0.1106, 0.1092, 0.1110, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 22:54:44,255 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=44943.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:54:50,249 INFO [train.py:968] (0/2) Epoch 1, batch 44950, giga_loss[loss=0.3875, simple_loss=0.4401, pruned_loss=0.1674, over 28629.00 frames. ], tot_loss[loss=0.4089, simple_loss=0.4422, pruned_loss=0.1878, over 5649656.08 frames. ], libri_tot_loss[loss=0.4233, simple_loss=0.4487, pruned_loss=0.1989, over 5656971.93 frames. ], giga_tot_loss[loss=0.4069, simple_loss=0.441, pruned_loss=0.1864, over 5653491.81 frames. ], batch size: 60, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:55:36,634 INFO [train.py:968] (0/2) Epoch 1, batch 45000, giga_loss[loss=0.4719, simple_loss=0.4849, pruned_loss=0.2295, over 28848.00 frames. ], tot_loss[loss=0.4097, simple_loss=0.4417, pruned_loss=0.1888, over 5650030.27 frames. ], libri_tot_loss[loss=0.4227, simple_loss=0.4483, pruned_loss=0.1986, over 5659108.21 frames. ], giga_tot_loss[loss=0.4082, simple_loss=0.4409, pruned_loss=0.1877, over 5651281.88 frames. ], batch size: 174, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:55:36,639 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 22:55:44,896 INFO [train.py:1012] (0/2) Epoch 1, validation: loss=0.2964, simple_loss=0.3891, pruned_loss=0.1019, over 944034.00 frames. 
+2023-02-28 22:55:44,896 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19478MB
+2023-02-28 22:55:51,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.571e+02 1.681e+03 2.237e+03 3.213e+03 8.606e+03, threshold=4.475e+03, percent-clipped=10.0
+2023-02-28 22:56:26,740 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-02-28 22:56:27,533 INFO [train.py:968] (0/2) Epoch 1, batch 45050, giga_loss[loss=0.4407, simple_loss=0.465, pruned_loss=0.2082, over 28739.00 frames. ], tot_loss[loss=0.4062, simple_loss=0.4389, pruned_loss=0.1867, over 5638982.97 frames. ], libri_tot_loss[loss=0.4209, simple_loss=0.4468, pruned_loss=0.1975, over 5657286.13 frames. ], giga_tot_loss[loss=0.4061, simple_loss=0.4393, pruned_loss=0.1864, over 5641507.07 frames. ], batch size: 284, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:57:13,898 INFO [train.py:968] (0/2) Epoch 1, batch 45100, giga_loss[loss=0.3662, simple_loss=0.4114, pruned_loss=0.1605, over 28713.00 frames. ], tot_loss[loss=0.4, simple_loss=0.4358, pruned_loss=0.1821, over 5644181.73 frames. ], libri_tot_loss[loss=0.4211, simple_loss=0.447, pruned_loss=0.1976, over 5652126.04 frames. ], giga_tot_loss[loss=0.3993, simple_loss=0.4359, pruned_loss=0.1814, over 5650777.34 frames. ], batch size: 262, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:57:20,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.068e+03 1.509e+03 1.904e+03 2.606e+03 6.369e+03, threshold=3.807e+03, percent-clipped=3.0
+2023-02-28 22:57:57,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1530, 1.2807, 1.1684, 0.7186], device='cuda:0'), covar=tensor([0.0489, 0.0368, 0.0282, 0.0385], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0630, 0.0745, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 22:57:58,355 INFO [train.py:968] (0/2) Epoch 1, batch 45150, giga_loss[loss=0.3836, simple_loss=0.4332, pruned_loss=0.167, over 28842.00 frames. ], tot_loss[loss=0.3953, simple_loss=0.4327, pruned_loss=0.1789, over 5644758.79 frames. ], libri_tot_loss[loss=0.4211, simple_loss=0.447, pruned_loss=0.1976, over 5657712.46 frames. ], giga_tot_loss[loss=0.3943, simple_loss=0.4325, pruned_loss=0.1781, over 5644715.07 frames. ], batch size: 186, lr: 1.66e-02, grad_scale: 2.0
+2023-02-28 22:58:05,705 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3290, 1.2816, 1.0677, 0.8919], device='cuda:0'), covar=tensor([0.0400, 0.0319, 0.0353, 0.0393], device='cuda:0'), in_proj_covar=tensor([0.0929, 0.0635, 0.0752, 0.0758], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 22:58:50,790 INFO [train.py:968] (0/2) Epoch 1, batch 45200, giga_loss[loss=0.3859, simple_loss=0.4248, pruned_loss=0.1735, over 28779.00 frames. ], tot_loss[loss=0.3964, simple_loss=0.4332, pruned_loss=0.1798, over 5654056.60 frames. ], libri_tot_loss[loss=0.4213, simple_loss=0.4471, pruned_loss=0.1978, over 5661358.40 frames. ], giga_tot_loss[loss=0.3949, simple_loss=0.4327, pruned_loss=0.1786, over 5650300.53 frames. ], batch size: 284, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:59:00,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.705e+03 2.207e+03 3.028e+03 7.932e+03, threshold=4.414e+03, percent-clipped=14.0
+2023-02-28 22:59:39,409 INFO [train.py:968] (0/2) Epoch 1, batch 45250, giga_loss[loss=0.3544, simple_loss=0.4023, pruned_loss=0.1532, over 28885.00 frames. ], tot_loss[loss=0.3961, simple_loss=0.4317, pruned_loss=0.1802, over 5660484.67 frames. ], libri_tot_loss[loss=0.4215, simple_loss=0.4472, pruned_loss=0.198, over 5656369.56 frames. ], giga_tot_loss[loss=0.3941, simple_loss=0.4309, pruned_loss=0.1787, over 5661634.17 frames. ], batch size: 145, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 23:00:16,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4114, 3.4069, 4.1553, 1.8184], device='cuda:0'), covar=tensor([0.0474, 0.0472, 0.0850, 0.1733], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0523, 0.0838, 0.0544], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-02-28 23:00:21,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45295.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:00:26,498 INFO [train.py:968] (0/2) Epoch 1, batch 45300, giga_loss[loss=0.4202, simple_loss=0.4516, pruned_loss=0.1944, over 28336.00 frames. ], tot_loss[loss=0.3936, simple_loss=0.4299, pruned_loss=0.1787, over 5672283.40 frames. ], libri_tot_loss[loss=0.4212, simple_loss=0.447, pruned_loss=0.1977, over 5662793.20 frames. ], giga_tot_loss[loss=0.3917, simple_loss=0.4289, pruned_loss=0.1772, over 5667930.95 frames. ], batch size: 368, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 23:00:35,054 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.697e+03 2.230e+03 2.943e+03 9.651e+03, threshold=4.460e+03, percent-clipped=9.0
+2023-02-28 23:00:43,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45318.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:00:50,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3190, 3.1145, 2.4113, 1.8558], device='cuda:0'), covar=tensor([0.0735, 0.0476, 0.0687, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0473, 0.0362, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:0')
+2023-02-28 23:01:11,184 INFO [train.py:968] (0/2) Epoch 1, batch 45350, giga_loss[loss=0.3751, simple_loss=0.4313, pruned_loss=0.1594, over 29108.00 frames. ], tot_loss[loss=0.393, simple_loss=0.4305, pruned_loss=0.1777, over 5685611.83 frames. ], libri_tot_loss[loss=0.4212, simple_loss=0.4471, pruned_loss=0.1976, over 5665756.50 frames. ], giga_tot_loss[loss=0.3909, simple_loss=0.4293, pruned_loss=0.1762, over 5679887.33 frames. ], batch size: 155, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 23:02:00,904 INFO [train.py:968] (0/2) Epoch 1, batch 45400, giga_loss[loss=0.4725, simple_loss=0.4707, pruned_loss=0.2372, over 26578.00 frames. ], tot_loss[loss=0.3928, simple_loss=0.4308, pruned_loss=0.1774, over 5668469.73 frames. ], libri_tot_loss[loss=0.4213, simple_loss=0.4473, pruned_loss=0.1977, over 5664526.90 frames. ], giga_tot_loss[loss=0.3907, simple_loss=0.4295, pruned_loss=0.176, over 5665451.05 frames. ], batch size: 555, lr: 1.65e-02, grad_scale: 4.0
+2023-02-28 23:02:09,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.523e+03 1.915e+03 2.728e+03 5.765e+03, threshold=3.830e+03, percent-clipped=4.0
+2023-02-28 23:02:24,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1832, 1.1702, 1.0382, 1.6068], device='cuda:0'), covar=tensor([0.1997, 0.2033, 0.1661, 0.1933], device='cuda:0'), in_proj_covar=tensor([0.0950, 0.0817, 0.0889, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 23:02:34,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45438.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:02:38,423 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45441.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:02:44,731 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=45449.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:02:45,189 INFO [train.py:968] (0/2) Epoch 1, batch 45450, libri_loss[loss=0.4263, simple_loss=0.4604, pruned_loss=0.1961, over 29294.00 frames. ], tot_loss[loss=0.394, simple_loss=0.4313, pruned_loss=0.1783, over 5665152.54 frames. ], libri_tot_loss[loss=0.4219, simple_loss=0.4478, pruned_loss=0.198, over 5660832.81 frames. ], giga_tot_loss[loss=0.3906, simple_loss=0.4291, pruned_loss=0.1761, over 5665988.27 frames. ], batch size: 94, lr: 1.65e-02, grad_scale: 4.0
+2023-02-28 23:02:56,135 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45461.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:02:58,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45464.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:03:03,950 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45470.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:03:04,716 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2823, 1.2345, 1.2493, 1.4345], device='cuda:0'), covar=tensor([0.1710, 0.1854, 0.1409, 0.1832], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0814, 0.0873, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 23:03:07,827 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=45474.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:03:24,538 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45493.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:03:30,671 INFO [train.py:968] (0/2) Epoch 1, batch 45500, giga_loss[loss=0.3593, simple_loss=0.4081, pruned_loss=0.1553, over 28590.00 frames. ], tot_loss[loss=0.3967, simple_loss=0.4328, pruned_loss=0.1803, over 5645457.96 frames. ], libri_tot_loss[loss=0.4226, simple_loss=0.4483, pruned_loss=0.1984, over 5644943.79 frames. ], giga_tot_loss[loss=0.3931, simple_loss=0.4304, pruned_loss=0.1779, over 5661020.38 frames. ], batch size: 307, lr: 1.65e-02, grad_scale: 4.0
+2023-02-28 23:03:41,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.722e+03 2.310e+03 2.749e+03 5.309e+03, threshold=4.619e+03, percent-clipped=8.0
+2023-02-28 23:03:46,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7339, 1.5904, 3.8787, 3.0403], device='cuda:0'), covar=tensor([0.1470, 0.1337, 0.0287, 0.0529], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0488, 0.0633, 0.0492], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 23:04:20,447 INFO [train.py:968] (0/2) Epoch 1, batch 45550, giga_loss[loss=0.4216, simple_loss=0.4621, pruned_loss=0.1906, over 28674.00 frames. ], tot_loss[loss=0.3996, simple_loss=0.435, pruned_loss=0.1821, over 5628274.97 frames. ], libri_tot_loss[loss=0.423, simple_loss=0.4486, pruned_loss=0.1987, over 5637059.07 frames. ], giga_tot_loss[loss=0.3964, simple_loss=0.4329, pruned_loss=0.18, over 5647717.73 frames. ], batch size: 85, lr: 1.65e-02, grad_scale: 4.0
+2023-02-28 23:05:09,265 INFO [train.py:968] (0/2) Epoch 1, batch 45600, giga_loss[loss=0.488, simple_loss=0.4979, pruned_loss=0.2391, over 28551.00 frames. ], tot_loss[loss=0.4041, simple_loss=0.4385, pruned_loss=0.1849, over 5589422.21 frames. ], libri_tot_loss[loss=0.4247, simple_loss=0.4497, pruned_loss=0.1998, over 5593205.55 frames. ], giga_tot_loss[loss=0.3996, simple_loss=0.4355, pruned_loss=0.1818, over 5645397.64 frames. ], batch size: 307, lr: 1.65e-02, grad_scale: 8.0
+2023-02-28 23:05:16,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.469e+02 1.687e+03 2.239e+03 2.865e+03 7.220e+03, threshold=4.478e+03, percent-clipped=8.0
+2023-02-28 23:05:51,788 INFO [train.py:968] (0/2) Epoch 1, batch 45650, giga_loss[loss=0.4178, simple_loss=0.4464, pruned_loss=0.1946, over 28352.00 frames. ], tot_loss[loss=0.4059, simple_loss=0.4397, pruned_loss=0.1861, over 5538161.33 frames. ], libri_tot_loss[loss=0.4264, simple_loss=0.4507, pruned_loss=0.201, over 5519298.40 frames. ], giga_tot_loss[loss=0.4001, simple_loss=0.436, pruned_loss=0.1821, over 5651246.62 frames. ], batch size: 369, lr: 1.65e-02, grad_scale: 4.0
+2023-02-28 23:06:15,234 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-02-28 23:06:19,256 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-1.pt
+2023-02-28 23:07:27,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.882e+02 1.648e+03 2.048e+03 2.765e+03 7.883e+03, threshold=4.096e+03, percent-clipped=6.0
+2023-02-28 23:07:34,250 INFO [train.py:968] (0/2) Epoch 2, batch 50, giga_loss[loss=0.384, simple_loss=0.44, pruned_loss=0.164, over 29018.00 frames. ], tot_loss[loss=0.3936, simple_loss=0.4428, pruned_loss=0.1722, over 1267745.75 frames. ], libri_tot_loss[loss=0.3923, simple_loss=0.4327, pruned_loss=0.1759, over 231978.81 frames. ], giga_tot_loss[loss=0.3946, simple_loss=0.4452, pruned_loss=0.172, over 1079198.70 frames. ], batch size: 136, lr: 1.62e-02, grad_scale: 4.0
+2023-02-28 23:07:55,634 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=45741.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:08:24,238 INFO [train.py:968] (0/2) Epoch 2, batch 100, libri_loss[loss=0.3514, simple_loss=0.4038, pruned_loss=0.1495, over 29517.00 frames. ], tot_loss[loss=0.3794, simple_loss=0.4309, pruned_loss=0.164, over 2235612.00 frames. ], libri_tot_loss[loss=0.3791, simple_loss=0.4235, pruned_loss=0.1673, over 407676.51 frames. ], giga_tot_loss[loss=0.3805, simple_loss=0.4329, pruned_loss=0.164, over 1971744.12 frames. ], batch size: 83, lr: 1.62e-02, grad_scale: 4.0
+2023-02-28 23:08:44,917 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8253, 1.5456, 1.4766, 1.4190], device='cuda:0'), covar=tensor([0.0712, 0.1339, 0.1185, 0.1147], device='cuda:0'), in_proj_covar=tensor([0.0490, 0.0825, 0.0641, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 23:09:00,595 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.771e+02 1.150e+03 1.498e+03 2.105e+03 7.312e+03, threshold=2.996e+03, percent-clipped=3.0
+2023-02-28 23:09:10,266 INFO [train.py:968] (0/2) Epoch 2, batch 150, giga_loss[loss=0.3069, simple_loss=0.367, pruned_loss=0.1234, over 28651.00 frames. ], tot_loss[loss=0.36, simple_loss=0.413, pruned_loss=0.1534, over 3001784.17 frames. ], libri_tot_loss[loss=0.3763, simple_loss=0.4228, pruned_loss=0.1649, over 488670.75 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4127, pruned_loss=0.1527, over 2750619.62 frames. ], batch size: 307, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:09:15,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45824.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:09:37,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45849.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:09:39,233 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-02-28 23:09:52,622 INFO [train.py:968] (0/2) Epoch 2, batch 200, giga_loss[loss=0.3056, simple_loss=0.3536, pruned_loss=0.1288, over 28666.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3983, pruned_loss=0.1451, over 3591521.25 frames. ], libri_tot_loss[loss=0.376, simple_loss=0.4246, pruned_loss=0.1637, over 630287.36 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.396, pruned_loss=0.1436, over 3337467.72 frames. ], batch size: 85, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:10:27,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.466e+02 1.045e+03 1.287e+03 1.996e+03 6.648e+03, threshold=2.575e+03, percent-clipped=6.0
+2023-02-28 23:10:36,025 INFO [train.py:968] (0/2) Epoch 2, batch 250, giga_loss[loss=0.2716, simple_loss=0.3398, pruned_loss=0.1017, over 28664.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.382, pruned_loss=0.1355, over 4060712.79 frames. ], libri_tot_loss[loss=0.376, simple_loss=0.4246, pruned_loss=0.1637, over 630287.36 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3796, pruned_loss=0.134, over 3862979.73 frames. ], batch size: 262, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:11:16,734 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45967.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:11:17,642 INFO [train.py:968] (0/2) Epoch 2, batch 300, giga_loss[loss=0.3305, simple_loss=0.3792, pruned_loss=0.1409, over 29085.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.374, pruned_loss=0.1313, over 4429238.32 frames. ], libri_tot_loss[loss=0.3754, simple_loss=0.4243, pruned_loss=0.1632, over 811873.68 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3698, pruned_loss=0.1287, over 4218121.06 frames. ], batch size: 155, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:11:19,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45970.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:11:39,471 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45992.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:11:41,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45995.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:11:48,468 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45999.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:11:48,965 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-46000.pt
+2023-02-28 23:11:59,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.983e+02 1.108e+03 1.463e+03 1.960e+03 4.605e+03, threshold=2.926e+03, percent-clipped=9.0
+2023-02-28 23:12:06,627 INFO [train.py:968] (0/2) Epoch 2, batch 350, giga_loss[loss=0.2982, simple_loss=0.3544, pruned_loss=0.1211, over 28894.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3639, pruned_loss=0.1255, over 4702997.91 frames. ], libri_tot_loss[loss=0.3754, simple_loss=0.4247, pruned_loss=0.1631, over 852326.57 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3598, pruned_loss=0.1231, over 4534398.92 frames. ], batch size: 145, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:12:11,898 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46024.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:12:46,721 INFO [train.py:968] (0/2) Epoch 2, batch 400, libri_loss[loss=0.3097, simple_loss=0.3733, pruned_loss=0.1231, over 28043.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.359, pruned_loss=0.123, over 4933172.57 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4167, pruned_loss=0.1574, over 1072134.78 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3543, pruned_loss=0.1203, over 4754906.11 frames. ], batch size: 62, lr: 1.61e-02, grad_scale: 8.0
+2023-02-28 23:12:53,441 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 23:13:09,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46095.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:13:13,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0741, 1.3645, 1.0614, 1.0386], device='cuda:0'), covar=tensor([0.1455, 0.0987, 0.1410, 0.2211], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0472, 0.0363, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:0')
+2023-02-28 23:13:20,299 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.154e+02 1.161e+03 1.571e+03 2.020e+03 6.177e+03, threshold=3.141e+03, percent-clipped=11.0
+2023-02-28 23:13:23,296 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-02-28 23:13:25,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46116.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:13:26,888 INFO [train.py:968] (0/2) Epoch 2, batch 450, giga_loss[loss=0.2619, simple_loss=0.3331, pruned_loss=0.09538, over 28931.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3563, pruned_loss=0.1213, over 5110193.70 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.4151, pruned_loss=0.1569, over 1213798.28 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3511, pruned_loss=0.1182, over 4944871.04 frames. ], batch size: 174, lr: 1.61e-02, grad_scale: 8.0
+2023-02-28 23:14:11,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46167.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:14:12,487 INFO [train.py:968] (0/2) Epoch 2, batch 500, giga_loss[loss=0.2827, simple_loss=0.3454, pruned_loss=0.11, over 28913.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.353, pruned_loss=0.1191, over 5248887.05 frames. ], libri_tot_loss[loss=0.3633, simple_loss=0.4145, pruned_loss=0.156, over 1329505.49 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3474, pruned_loss=0.1159, over 5101104.40 frames. ], batch size: 213, lr: 1.61e-02, grad_scale: 8.0
+2023-02-28 23:14:34,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5454, 2.3135, 1.5951, 1.2272], device='cuda:0'), covar=tensor([0.0962, 0.0678, 0.0917, 0.1601], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0465, 0.0356, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:0')
+2023-02-28 23:14:48,991 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.976e+02 1.021e+03 1.324e+03 1.824e+03 5.616e+03, threshold=2.647e+03, percent-clipped=6.0
+2023-02-28 23:14:56,633 INFO [train.py:968] (0/2) Epoch 2, batch 550, giga_loss[loss=0.2816, simple_loss=0.3444, pruned_loss=0.1094, over 28672.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3509, pruned_loss=0.118, over 5349673.21 frames. ], libri_tot_loss[loss=0.3584, simple_loss=0.4105, pruned_loss=0.1531, over 1485841.79 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3453, pruned_loss=0.1149, over 5213024.16 frames. ], batch size: 242, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:15:25,653 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46252.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:15:32,432 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46259.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:15:34,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46262.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 23:15:39,752 INFO [train.py:968] (0/2) Epoch 2, batch 600, giga_loss[loss=0.2766, simple_loss=0.3357, pruned_loss=0.1087, over 28725.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.35, pruned_loss=0.1175, over 5435199.42 frames. ], libri_tot_loss[loss=0.3577, simple_loss=0.4106, pruned_loss=0.1524, over 1637078.37 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3434, pruned_loss=0.114, over 5310045.17 frames. ], batch size: 119, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:16:03,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46291.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:16:04,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5209, 2.0948, 1.6683, 1.2974], device='cuda:0'), covar=tensor([0.0989, 0.0746, 0.0912, 0.1551], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0453, 0.0350, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0010, 0.0014], device='cuda:0')
+2023-02-28 23:16:22,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.461e+02 1.045e+03 1.343e+03 1.903e+03 5.305e+03, threshold=2.686e+03, percent-clipped=8.0
+2023-02-28 23:16:30,297 INFO [train.py:968] (0/2) Epoch 2, batch 650, giga_loss[loss=0.3044, simple_loss=0.3496, pruned_loss=0.1295, over 27569.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3466, pruned_loss=0.1158, over 5484137.89 frames. ], libri_tot_loss[loss=0.3579, simple_loss=0.4105, pruned_loss=0.1526, over 1721704.00 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3402, pruned_loss=0.1122, over 5376945.12 frames. ], batch size: 472, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:16:39,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8541, 1.7128, 1.2819, 1.3645], device='cuda:0'), covar=tensor([0.0684, 0.0792, 0.1163, 0.1174], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0541, 0.0572, 0.0510], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 23:17:20,664 INFO [train.py:968] (0/2) Epoch 2, batch 700, giga_loss[loss=0.3025, simple_loss=0.3567, pruned_loss=0.1241, over 28238.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.342, pruned_loss=0.1131, over 5528246.08 frames. ], libri_tot_loss[loss=0.3582, simple_loss=0.4108, pruned_loss=0.1527, over 1742463.48 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3366, pruned_loss=0.1101, over 5442794.24 frames. ], batch size: 368, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:17:22,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4725, 1.9921, 1.6519, 1.5044], device='cuda:0'), covar=tensor([0.1212, 0.0461, 0.0551, 0.1418], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0248, 0.0250, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0020, 0.0018, 0.0029], device='cuda:0')
+2023-02-28 23:17:58,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.025e+02 1.005e+03 1.339e+03 1.878e+03 6.019e+03, threshold=2.679e+03, percent-clipped=9.0
+2023-02-28 23:18:05,569 INFO [train.py:968] (0/2) Epoch 2, batch 750, giga_loss[loss=0.2673, simple_loss=0.3213, pruned_loss=0.1066, over 28471.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3396, pruned_loss=0.1117, over 5576237.45 frames. ], libri_tot_loss[loss=0.3587, simple_loss=0.4113, pruned_loss=0.1531, over 1825236.40 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3337, pruned_loss=0.1084, over 5501945.41 frames. ], batch size: 65, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:18:15,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-02-28 23:18:49,614 INFO [train.py:968] (0/2) Epoch 2, batch 800, giga_loss[loss=0.308, simple_loss=0.3642, pruned_loss=0.1259, over 28925.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3375, pruned_loss=0.111, over 5607790.56 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4126, pruned_loss=0.154, over 1906529.96 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3313, pruned_loss=0.1074, over 5543239.87 frames. ], batch size: 145, lr: 1.60e-02, grad_scale: 8.0
+2023-02-28 23:18:51,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46470.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:19:12,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1896, 2.3940, 1.9832, 1.9744], device='cuda:0'), covar=tensor([0.0957, 0.0391, 0.0506, 0.1214], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0246, 0.0247, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0019, 0.0017, 0.0028], device='cuda:0')
+2023-02-28 23:19:35,063 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.774e+02 1.199e+03 1.451e+03 2.045e+03 3.913e+03, threshold=2.902e+03, percent-clipped=13.0
+2023-02-28 23:19:41,753 INFO [train.py:968] (0/2) Epoch 2, batch 850, giga_loss[loss=0.4512, simple_loss=0.4734, pruned_loss=0.2145, over 28827.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3482, pruned_loss=0.1189, over 5608847.25 frames. ], libri_tot_loss[loss=0.3612, simple_loss=0.413, pruned_loss=0.1547, over 1955273.52 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3423, pruned_loss=0.1154, over 5563941.43 frames. ], batch size: 119, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:19:44,374 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-02-28 23:20:05,534 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1482, 3.1997, 2.0420, 1.9110], device='cuda:0'), covar=tensor([0.0872, 0.0438, 0.0827, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0453, 0.0350, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0010, 0.0014], device='cuda:0')
+2023-02-28 23:20:05,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46542.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:20:10,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3467, 1.2374, 1.2394, 1.4866], device='cuda:0'), covar=tensor([0.1720, 0.1638, 0.1263, 0.1832], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0793, 0.0857, 0.0918], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 23:20:31,320 INFO [train.py:968] (0/2) Epoch 2, batch 900, giga_loss[loss=0.3684, simple_loss=0.4258, pruned_loss=0.1555, over 28893.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3654, pruned_loss=0.1291, over 5630077.50 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4126, pruned_loss=0.154, over 2072420.33 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3593, pruned_loss=0.1257, over 5586402.39 frames. ], batch size: 174, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:20:52,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0251, 1.7169, 1.3295, 1.3767], device='cuda:0'), covar=tensor([0.0628, 0.0793, 0.0959, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0524, 0.0549, 0.0493], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-02-28 23:21:05,062 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.130e+02 1.571e+03 2.087e+03 2.493e+03 6.422e+03, threshold=4.173e+03, percent-clipped=8.0
+2023-02-28 23:21:06,046 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46613.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:21:09,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46616.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:21:12,641 INFO [train.py:968] (0/2) Epoch 2, batch 950, giga_loss[loss=0.3525, simple_loss=0.4107, pruned_loss=0.1471, over 28789.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3795, pruned_loss=0.1374, over 5656039.69 frames. ], libri_tot_loss[loss=0.3617, simple_loss=0.4136, pruned_loss=0.1549, over 2223446.85 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3729, pruned_loss=0.1337, over 5611058.20 frames. ], batch size: 92, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:21:19,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46627.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:21:36,029 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46645.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:21:54,918 INFO [train.py:968] (0/2) Epoch 2, batch 1000, libri_loss[loss=0.4447, simple_loss=0.4746, pruned_loss=0.2075, over 27877.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3893, pruned_loss=0.1424, over 5666691.80 frames. ], libri_tot_loss[loss=0.3624, simple_loss=0.4142, pruned_loss=0.1553, over 2294439.10 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3834, pruned_loss=0.139, over 5628429.62 frames. ], batch size: 116, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:21:55,130 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46669.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:22:07,901 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46685.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:22:09,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46688.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:22:29,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.765e+02 1.182e+03 1.485e+03 1.943e+03 4.227e+03, threshold=2.969e+03, percent-clipped=1.0
+2023-02-28 23:22:32,696 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46717.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:22:33,862 INFO [train.py:968] (0/2) Epoch 2, batch 1050, libri_loss[loss=0.3012, simple_loss=0.3612, pruned_loss=0.1206, over 29681.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3927, pruned_loss=0.1422, over 5681905.29 frames. ], libri_tot_loss[loss=0.3613, simple_loss=0.4135, pruned_loss=0.1546, over 2382542.89 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3878, pruned_loss=0.1395, over 5647422.01 frames. ], batch size: 69, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:23:26,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-02-28 23:23:27,794 INFO [train.py:968] (0/2) Epoch 2, batch 1100, libri_loss[loss=0.3558, simple_loss=0.4142, pruned_loss=0.1487, over 29544.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3937, pruned_loss=0.1415, over 5675484.16 frames. ], libri_tot_loss[loss=0.3608, simple_loss=0.4131, pruned_loss=0.1542, over 2435102.11 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3897, pruned_loss=0.1393, over 5644902.41 frames. ], batch size: 89, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:23:28,735 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46770.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:23:32,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46773.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:23:58,022 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46802.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:23:58,172 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7848, 2.1873, 1.9409, 1.8747], device='cuda:0'), covar=tensor([0.1309, 0.1387, 0.1032, 0.0634], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0836, 0.0715, 0.0608], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 23:24:07,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.829e+02 1.122e+03 1.381e+03 1.846e+03 4.504e+03, threshold=2.762e+03, percent-clipped=4.0
+2023-02-28 23:24:13,330 INFO [train.py:968] (0/2) Epoch 2, batch 1150, giga_loss[loss=0.3582, simple_loss=0.4113, pruned_loss=0.1526, over 28825.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3959, pruned_loss=0.1428, over 5692660.74 frames. ], libri_tot_loss[loss=0.3614, simple_loss=0.4136, pruned_loss=0.1546, over 2486685.34 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3923, pruned_loss=0.1408, over 5665826.76 frames. ], batch size: 186, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:24:17,684 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-02-28 23:24:57,109 INFO [train.py:968] (0/2) Epoch 2, batch 1200, giga_loss[loss=0.3599, simple_loss=0.4089, pruned_loss=0.1554, over 28510.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3992, pruned_loss=0.1454, over 5684656.05 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4134, pruned_loss=0.1536, over 2723393.30 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3952, pruned_loss=0.1434, over 5657526.08 frames. ], batch size: 60, lr: 1.60e-02, grad_scale: 8.0
+2023-02-28 23:25:35,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.037e+02 1.282e+03 1.647e+03 2.067e+03 3.785e+03, threshold=3.295e+03, percent-clipped=9.0
+2023-02-28 23:25:43,297 INFO [train.py:968] (0/2) Epoch 2, batch 1250, libri_loss[loss=0.3342, simple_loss=0.3856, pruned_loss=0.1413, over 29673.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.402, pruned_loss=0.1476, over 5677749.50 frames. ], libri_tot_loss[loss=0.3612, simple_loss=0.4142, pruned_loss=0.1541, over 2804047.81 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3982, pruned_loss=0.1456, over 5661896.00 frames. ], batch size: 73, lr: 1.60e-02, grad_scale: 8.0
+2023-02-28 23:25:53,137 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-02-28 23:26:26,381 INFO [train.py:968] (0/2) Epoch 2, batch 1300, giga_loss[loss=0.3396, simple_loss=0.3999, pruned_loss=0.1396, over 29040.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.4054, pruned_loss=0.1496, over 5680158.34 frames. ], libri_tot_loss[loss=0.3599, simple_loss=0.4134, pruned_loss=0.1532, over 2911241.84 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4023, pruned_loss=0.1483, over 5659842.22 frames. ], batch size: 136, lr: 1.60e-02, grad_scale: 8.0
+2023-02-28 23:26:43,252 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46988.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:26:49,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4728, 2.0023, 1.6284, 1.2819], device='cuda:0'), covar=tensor([0.0937, 0.0633, 0.0824, 0.1585], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0446, 0.0347, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0010, 0.0014], device='cuda:0')
+2023-02-28 23:26:50,452 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46995.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:27:04,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.621e+02 1.292e+03 1.604e+03 2.267e+03 6.577e+03, threshold=3.208e+03, percent-clipped=9.0
+2023-02-28 23:27:09,814 INFO [train.py:968] (0/2) Epoch 2, batch 1350, giga_loss[loss=0.3537, simple_loss=0.4243, pruned_loss=0.1415, over 28695.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.4072, pruned_loss=0.1494, over 5689268.15 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4136, pruned_loss=0.1535, over 2931218.42 frames. ], giga_tot_loss[loss=0.3505, simple_loss=0.4046, pruned_loss=0.1482, over 5679683.88 frames. ], batch size: 60, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:27:33,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47044.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:27:56,389 INFO [train.py:968] (0/2) Epoch 2, batch 1400, giga_loss[loss=0.3558, simple_loss=0.4239, pruned_loss=0.1439, over 29076.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.4079, pruned_loss=0.1491, over 5686380.17 frames. ], libri_tot_loss[loss=0.3599, simple_loss=0.4132, pruned_loss=0.1533, over 2946198.33 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4061, pruned_loss=0.1483, over 5677695.84 frames. ], batch size: 155, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:28:32,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.856e+02 1.104e+03 1.388e+03 1.912e+03 4.805e+03, threshold=2.775e+03, percent-clipped=2.0
+2023-02-28 23:28:37,726 INFO [train.py:968] (0/2) Epoch 2, batch 1450, giga_loss[loss=0.3343, simple_loss=0.4022, pruned_loss=0.1332, over 28941.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4079, pruned_loss=0.1481, over 5692844.80 frames. ], libri_tot_loss[loss=0.3592, simple_loss=0.4129, pruned_loss=0.1527, over 3086592.98 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4064, pruned_loss=0.1475, over 5679716.01 frames. ], batch size: 213, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:29:21,297 INFO [train.py:968] (0/2) Epoch 2, batch 1500, giga_loss[loss=0.3166, simple_loss=0.3874, pruned_loss=0.1229, over 28517.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.4058, pruned_loss=0.1448, over 5690337.25 frames. ], libri_tot_loss[loss=0.3599, simple_loss=0.4136, pruned_loss=0.1531, over 3115152.13 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.4042, pruned_loss=0.1441, over 5689237.55 frames. ], batch size: 60, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:29:37,133 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.12 vs. limit=5.0
+2023-02-28 23:29:38,864 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47187.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:29:40,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47190.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:29:49,509 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=47203.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:29:56,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.024e+02 1.106e+03 1.385e+03 1.810e+03 4.504e+03, threshold=2.770e+03, percent-clipped=5.0
+2023-02-28 23:30:02,101 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-02-28 23:30:03,137 INFO [train.py:968] (0/2) Epoch 2, batch 1550, giga_loss[loss=0.3539, simple_loss=0.409, pruned_loss=0.1494, over 28694.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.403, pruned_loss=0.1422, over 5693408.07 frames. ], libri_tot_loss[loss=0.3597, simple_loss=0.4135, pruned_loss=0.153, over 3155681.06 frames. ], giga_tot_loss[loss=0.3423, simple_loss=0.4016, pruned_loss=0.1415, over 5690621.85 frames. ], batch size: 242, lr: 1.59e-02, grad_scale: 4.0
+2023-02-28 23:30:03,381 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47219.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:30:04,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3102, 1.2991, 1.0783, 1.6741], device='cuda:0'), covar=tensor([0.1781, 0.1719, 0.1506, 0.1976], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0774, 0.0861, 0.0926], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 23:30:52,159 INFO [train.py:968] (0/2) Epoch 2, batch 1600, libri_loss[loss=0.4122, simple_loss=0.4566, pruned_loss=0.1839, over 25887.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.4038, pruned_loss=0.1446, over 5701480.06 frames. ], libri_tot_loss[loss=0.3592, simple_loss=0.413, pruned_loss=0.1527, over 3193075.18 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.4028, pruned_loss=0.144, over 5700059.99 frames. ], batch size: 136, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:31:32,908 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.948e+02 1.325e+03 1.641e+03 2.084e+03 4.877e+03, threshold=3.283e+03, percent-clipped=15.0
+2023-02-28 23:31:37,954 INFO [train.py:968] (0/2) Epoch 2, batch 1650, giga_loss[loss=0.3838, simple_loss=0.4174, pruned_loss=0.1751, over 28817.00 frames. ], tot_loss[loss=0.3548, simple_loss=0.4076, pruned_loss=0.151, over 5706304.50 frames. ], libri_tot_loss[loss=0.3588, simple_loss=0.4127, pruned_loss=0.1525, over 3244744.35 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4069, pruned_loss=0.1506, over 5703605.38 frames. ], batch size: 119, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:32:15,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47363.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:32:18,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2984, 1.8820, 1.7220, 1.8703], device='cuda:0'), covar=tensor([0.0687, 0.1344, 0.1040, 0.0954], device='cuda:0'), in_proj_covar=tensor([0.0495, 0.0820, 0.0626, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-02-28 23:32:21,375 INFO [train.py:968] (0/2) Epoch 2, batch 1700, giga_loss[loss=0.4018, simple_loss=0.4207, pruned_loss=0.1914, over 26535.00 frames. ], tot_loss[loss=0.3604, simple_loss=0.4103, pruned_loss=0.1553, over 5691366.20 frames. ], libri_tot_loss[loss=0.3596, simple_loss=0.4132, pruned_loss=0.153, over 3340096.07 frames. ], giga_tot_loss[loss=0.3593, simple_loss=0.4093, pruned_loss=0.1547, over 5690914.51 frames. ], batch size: 555, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:32:22,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47370.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:32:38,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7291, 1.8295, 1.2819, 1.0089], device='cuda:0'), covar=tensor([0.0453, 0.0349, 0.0423, 0.0479], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0648, 0.0746, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-02-28 23:32:59,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.596e+02 1.125e+03 1.532e+03 2.158e+03 5.376e+03, threshold=3.064e+03, percent-clipped=8.0
+2023-02-28 23:33:05,211 INFO [train.py:968] (0/2) Epoch 2, batch 1750, giga_loss[loss=0.3381, simple_loss=0.3919, pruned_loss=0.1422, over 28888.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.4082, pruned_loss=0.1543, over 5694965.25 frames. ], libri_tot_loss[loss=0.3592, simple_loss=0.413, pruned_loss=0.1527, over 3403236.76 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4074, pruned_loss=0.1541, over 5690035.17 frames. ], batch size: 145, lr: 1.59e-02, grad_scale: 4.0
+2023-02-28 23:33:07,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9556, 1.8599, 4.1402, 3.3345], device='cuda:0'), covar=tensor([0.1404, 0.1287, 0.0279, 0.0396], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0476, 0.0617, 0.0492], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 23:33:48,394 INFO [train.py:968] (0/2) Epoch 2, batch 1800, giga_loss[loss=0.3834, simple_loss=0.4322, pruned_loss=0.1673, over 28624.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4068, pruned_loss=0.1535, over 5704886.42 frames. ], libri_tot_loss[loss=0.3588, simple_loss=0.413, pruned_loss=0.1524, over 3495562.38 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.406, pruned_loss=0.1535, over 5699335.53 frames. ], batch size: 307, lr: 1.59e-02, grad_scale: 4.0
+2023-02-28 23:34:20,484 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47506.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:34:23,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47509.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:34:30,373 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47513.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:34:30,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.407e+02 1.301e+03 1.683e+03 2.558e+03 5.742e+03, threshold=3.366e+03, percent-clipped=15.0
+2023-02-28 23:34:32,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47516.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:34:33,344 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-02-28 23:34:34,404 INFO [train.py:968] (0/2) Epoch 2, batch 1850, giga_loss[loss=0.3433, simple_loss=0.398, pruned_loss=0.1443, over 28704.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.408, pruned_loss=0.1543, over 5709094.37 frames. ], libri_tot_loss[loss=0.36, simple_loss=0.4137, pruned_loss=0.1531, over 3542682.22 frames. ], giga_tot_loss[loss=0.3574, simple_loss=0.4069, pruned_loss=0.1539, over 5701746.16 frames. ], batch size: 65, lr: 1.59e-02, grad_scale: 4.0
+2023-02-28 23:34:54,635 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47538.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:35:01,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47545.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:35:27,856 INFO [train.py:968] (0/2) Epoch 2, batch 1900, giga_loss[loss=0.3781, simple_loss=0.413, pruned_loss=0.1716, over 26558.00 frames. ], tot_loss[loss=0.3551, simple_loss=0.4062, pruned_loss=0.152, over 5710567.64 frames. ], libri_tot_loss[loss=0.3592, simple_loss=0.4134, pruned_loss=0.1525, over 3600876.82 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4053, pruned_loss=0.152, over 5700532.32 frames. ], batch size: 555, lr: 1.59e-02, grad_scale: 4.0
+2023-02-28 23:35:40,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47578.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:36:23,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.107e+02 1.134e+03 1.367e+03 1.702e+03 5.060e+03, threshold=2.734e+03, percent-clipped=3.0
+2023-02-28 23:36:28,936 INFO [train.py:968] (0/2) Epoch 2, batch 1950, giga_loss[loss=0.3055, simple_loss=0.3691, pruned_loss=0.121, over 28874.00 frames. ], tot_loss[loss=0.3484, simple_loss=0.401, pruned_loss=0.1479, over 5699143.12 frames. ], libri_tot_loss[loss=0.3587, simple_loss=0.4129, pruned_loss=0.1522, over 3634855.80 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.4004, pruned_loss=0.148, over 5689048.72 frames. ], batch size: 186, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:36:31,723 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-02-28 23:36:43,715 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=47634.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:37:17,635 INFO [train.py:968] (0/2) Epoch 2, batch 2000, giga_loss[loss=0.2952, simple_loss=0.3545, pruned_loss=0.1179, over 28778.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3937, pruned_loss=0.1427, over 5694477.55 frames. ], libri_tot_loss[loss=0.3586, simple_loss=0.4128, pruned_loss=0.1522, over 3707934.79 frames. ], giga_tot_loss[loss=0.3391, simple_loss=0.3928, pruned_loss=0.1427, over 5684552.29 frames. ], batch size: 119, lr: 1.58e-02, grad_scale: 8.0
+2023-02-28 23:37:58,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.026e+02 1.095e+03 1.402e+03 1.739e+03 7.438e+03, threshold=2.803e+03, percent-clipped=12.0
+2023-02-28 23:38:02,277 INFO [train.py:968] (0/2) Epoch 2, batch 2050, giga_loss[loss=0.272, simple_loss=0.3381, pruned_loss=0.103, over 28994.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.388, pruned_loss=0.1394, over 5692277.13 frames. ], libri_tot_loss[loss=0.3595, simple_loss=0.4134, pruned_loss=0.1528, over 3792934.96 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3861, pruned_loss=0.1386, over 5677768.01 frames. ], batch size: 136, lr: 1.58e-02, grad_scale: 8.0
+2023-02-28 23:38:04,329 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47721.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:38:07,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47724.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:38:36,574 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47753.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:38:54,278 INFO [train.py:968] (0/2) Epoch 2, batch 2100, giga_loss[loss=0.3315, simple_loss=0.3914, pruned_loss=0.1358, over 28674.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3835, pruned_loss=0.1366, over 5689914.40 frames. ], libri_tot_loss[loss=0.3602, simple_loss=0.4139, pruned_loss=0.1532, over 3813036.84 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3814, pruned_loss=0.1355, over 5677407.79 frames. ], batch size: 284, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:38:57,155 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=47772.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:39:31,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.075e+02 1.072e+03 1.382e+03 1.705e+03 4.571e+03, threshold=2.764e+03, percent-clipped=10.0
+2023-02-28 23:39:32,343 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 23:39:34,687 INFO [train.py:968] (0/2) Epoch 2, batch 2150, giga_loss[loss=0.335, simple_loss=0.3825, pruned_loss=0.1437, over 28696.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3842, pruned_loss=0.1365, over 5685608.72 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4142, pruned_loss=0.1532, over 3860978.90 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3816, pruned_loss=0.1352, over 5682410.44 frames. ], batch size: 78, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:39:59,662 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-02-28 23:40:14,066 INFO [train.py:968] (0/2) Epoch 2, batch 2200, libri_loss[loss=0.4184, simple_loss=0.4452, pruned_loss=0.1958, over 28119.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3848, pruned_loss=0.1366, over 5693792.92 frames. ], libri_tot_loss[loss=0.3609, simple_loss=0.4148, pruned_loss=0.1535, over 3929327.15 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3813, pruned_loss=0.1349, over 5686705.46 frames. ], batch size: 62, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:40:51,832 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-02-28 23:40:52,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.067e+02 1.056e+03 1.335e+03 1.821e+03 6.823e+03, threshold=2.671e+03, percent-clipped=7.0
+2023-02-28 23:40:55,849 INFO [train.py:968] (0/2) Epoch 2, batch 2250, giga_loss[loss=0.2878, simple_loss=0.3548, pruned_loss=0.1104, over 29048.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3825, pruned_loss=0.1351, over 5701452.41 frames. ], libri_tot_loss[loss=0.3614, simple_loss=0.4154, pruned_loss=0.1537, over 3978691.35 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3787, pruned_loss=0.1332, over 5690780.97 frames. ], batch size: 155, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:41:39,414 INFO [train.py:968] (0/2) Epoch 2, batch 2300, giga_loss[loss=0.3231, simple_loss=0.374, pruned_loss=0.1361, over 28630.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3792, pruned_loss=0.1335, over 5695240.56 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4165, pruned_loss=0.1544, over 3997202.42 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.375, pruned_loss=0.1312, over 5694085.72 frames. ], batch size: 262, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:42:07,101 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-48000.pt
+2023-02-28 23:42:15,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48009.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:42:20,205 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.327e+02 1.006e+03 1.227e+03 1.573e+03 4.014e+03, threshold=2.454e+03, percent-clipped=6.0
+2023-02-28 23:42:23,020 INFO [train.py:968] (0/2) Epoch 2, batch 2350, giga_loss[loss=0.2903, simple_loss=0.3506, pruned_loss=0.115, over 28934.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3756, pruned_loss=0.1313, over 5709114.32 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.4168, pruned_loss=0.1545, over 4026005.34 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3715, pruned_loss=0.1291, over 5705290.53 frames. ], batch size: 227, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:42:58,415 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 23:43:02,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2429, 1.5722, 1.3096, 0.2854], device='cuda:0'), covar=tensor([0.1076, 0.0868, 0.1439, 0.1633], device='cuda:0'), in_proj_covar=tensor([0.1089, 0.1035, 0.1074, 0.0930], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 23:43:06,477 INFO [train.py:968] (0/2) Epoch 2, batch 2400, giga_loss[loss=0.3077, simple_loss=0.3614, pruned_loss=0.127, over 28723.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3747, pruned_loss=0.1311, over 5715819.09 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4187, pruned_loss=0.1557, over 4069898.29 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3693, pruned_loss=0.1281, over 5710901.94 frames. ], batch size: 262, lr: 1.58e-02, grad_scale: 8.0
+2023-02-28 23:43:42,995 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.580e+02 1.076e+03 1.362e+03 1.923e+03 4.633e+03, threshold=2.724e+03, percent-clipped=11.0
+2023-02-28 23:43:46,337 INFO [train.py:968] (0/2) Epoch 2, batch 2450, giga_loss[loss=0.3071, simple_loss=0.3649, pruned_loss=0.1247, over 28733.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3718, pruned_loss=0.1294, over 5723350.39 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4195, pruned_loss=0.156, over 4106595.11 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3662, pruned_loss=0.1264, over 5715918.98 frames. ], batch size: 262, lr: 1.58e-02, grad_scale: 8.0
+2023-02-28 23:43:55,190 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48131.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:02,025 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-02-28 23:44:06,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48147.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:10,686 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48152.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:13,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48155.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:25,030 INFO [train.py:968] (0/2) Epoch 2, batch 2500, giga_loss[loss=0.2764, simple_loss=0.339, pruned_loss=0.1069, over 28994.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3686, pruned_loss=0.1276, over 5726682.09 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4207, pruned_loss=0.1566, over 4133319.71 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3628, pruned_loss=0.1244, over 5718590.88 frames. ], batch size: 136, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:44:35,160 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48181.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:37,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48184.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:51,593 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48202.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:45:03,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.325e+02 1.092e+03 1.392e+03 1.810e+03 7.407e+03, threshold=2.784e+03, percent-clipped=9.0
+2023-02-28 23:45:06,124 INFO [train.py:968] (0/2) Epoch 2, batch 2550, giga_loss[loss=0.3068, simple_loss=0.3687, pruned_loss=0.1225, over 28301.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3681, pruned_loss=0.1272, over 5723576.74 frames. ], libri_tot_loss[loss=0.368, simple_loss=0.4218, pruned_loss=0.1571, over 4192829.18 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.361, pruned_loss=0.1234, over 5712792.49 frames. ], batch size: 368, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:45:17,662 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-02-28 23:45:42,603 INFO [train.py:968] (0/2) Epoch 2, batch 2600, libri_loss[loss=0.3099, simple_loss=0.3778, pruned_loss=0.121, over 29652.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3678, pruned_loss=0.1265, over 5733804.61 frames. ], libri_tot_loss[loss=0.3683, simple_loss=0.4223, pruned_loss=0.1571, over 4260374.14 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3598, pruned_loss=0.1223, over 5718589.00 frames. ], batch size: 73, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:46:00,344 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48290.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:46:02,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48293.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:46:18,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.904e+02 1.404e+03 1.710e+03 2.450e+03 9.596e+03, threshold=3.421e+03, percent-clipped=16.0
+2023-02-28 23:46:22,439 INFO [train.py:968] (0/2) Epoch 2, batch 2650, giga_loss[loss=0.2578, simple_loss=0.3311, pruned_loss=0.09219, over 28549.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3675, pruned_loss=0.1265, over 5722409.14 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.423, pruned_loss=0.1576, over 4312268.43 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3587, pruned_loss=0.1217, over 5715958.46 frames. ], batch size: 65, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:46:25,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5482, 2.1511, 2.0796, 1.6012], device='cuda:0'), covar=tensor([0.1209, 0.0486, 0.0471, 0.1370], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0241, 0.0241, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0020, 0.0018, 0.0029], device='cuda:0')
+2023-02-28 23:46:25,669 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48322.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:46:27,091 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4013, 2.0644, 1.6670, 0.5517], device='cuda:0'), covar=tensor([0.1768, 0.1065, 0.1535, 0.2256], device='cuda:0'), in_proj_covar=tensor([0.1111, 0.1062, 0.1101, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-02-28 23:47:07,072 INFO [train.py:968] (0/2) Epoch 2, batch 2700, giga_loss[loss=0.2969, simple_loss=0.3549, pruned_loss=0.1195, over 28474.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3684, pruned_loss=0.1272, over 5712836.41 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4238, pruned_loss=0.1584, over 4334509.91 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3598, pruned_loss=0.1223, over 5712145.10 frames. ], batch size: 78, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:47:50,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.406e+02 1.067e+03 1.350e+03 1.837e+03 5.196e+03, threshold=2.701e+03, percent-clipped=4.0
+2023-02-28 23:47:54,478 INFO [train.py:968] (0/2) Epoch 2, batch 2750, giga_loss[loss=0.4534, simple_loss=0.4719, pruned_loss=0.2175, over 27661.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3742, pruned_loss=0.1315, over 5712783.59 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4238, pruned_loss=0.1583, over 4372965.78 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3661, pruned_loss=0.1269, over 5708214.24 frames. ], batch size: 472, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:48:39,037 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48465.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:48:41,464 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3952, 1.7415, 1.5047, 1.5677], device='cuda:0'), covar=tensor([0.0969, 0.1226, 0.0914, 0.0572], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0819, 0.0710, 0.0602], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 23:48:41,710 INFO [train.py:968] (0/2) Epoch 2, batch 2800, giga_loss[loss=0.4346, simple_loss=0.4613, pruned_loss=0.204, over 28286.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3824, pruned_loss=0.1378, over 5703886.84 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4237, pruned_loss=0.1582, over 4409629.72 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3749, pruned_loss=0.1336, over 5697113.12 frames. ], batch size: 368, lr: 1.57e-02, grad_scale: 8.0
+2023-02-28 23:49:21,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48506.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:49:31,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.599e+02 1.195e+03 1.537e+03 2.107e+03 3.899e+03, threshold=3.074e+03, percent-clipped=6.0
+2023-02-28 23:49:32,410 INFO [train.py:968] (0/2) Epoch 2, batch 2850, giga_loss[loss=0.3778, simple_loss=0.4303, pruned_loss=0.1626, over 28659.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3932, pruned_loss=0.1465, over 5696686.91 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4237, pruned_loss=0.1581, over 4417307.57 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3872, pruned_loss=0.1433, over 5690207.91 frames. ], batch size: 242, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:50:12,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48556.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:50:25,350 INFO [train.py:968] (0/2) Epoch 2, batch 2900, giga_loss[loss=0.4122, simple_loss=0.4337, pruned_loss=0.1954, over 23819.00 frames. ], tot_loss[loss=0.35, simple_loss=0.3999, pruned_loss=0.15, over 5681385.77 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4243, pruned_loss=0.1585, over 4424610.03 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3947, pruned_loss=0.1472, over 5675536.66 frames. ], batch size: 710, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:50:35,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48577.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:50:58,960 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48605.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:51:08,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.516e+02 1.228e+03 1.553e+03 2.164e+03 6.393e+03, threshold=3.106e+03, percent-clipped=9.0
+2023-02-28 23:51:10,212 INFO [train.py:968] (0/2) Epoch 2, batch 2950, giga_loss[loss=0.3481, simple_loss=0.403, pruned_loss=0.1466, over 28770.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.4055, pruned_loss=0.1531, over 5680114.80 frames. ], libri_tot_loss[loss=0.3704, simple_loss=0.4238, pruned_loss=0.1585, over 4471794.35 frames. ], giga_tot_loss[loss=0.3511, simple_loss=0.401, pruned_loss=0.1506, over 5671294.26 frames. ], batch size: 119, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:51:23,726 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-02-28 23:51:43,082 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48649.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:51:46,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48652.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:51:51,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0589, 0.9891, 0.7479, 1.2216], device='cuda:0'), covar=tensor([0.1065, 0.0554, 0.0564, 0.1292], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0233, 0.0237, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0019, 0.0017, 0.0029], device='cuda:0')
+2023-02-28 23:51:57,804 INFO [train.py:968] (0/2) Epoch 2, batch 3000, libri_loss[loss=0.3522, simple_loss=0.3979, pruned_loss=0.1533, over 29533.00 frames. ], tot_loss[loss=0.3624, simple_loss=0.4112, pruned_loss=0.1568, over 5692236.40 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.423, pruned_loss=0.1578, over 4519418.15 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4078, pruned_loss=0.1552, over 5679668.48 frames. ], batch size: 79, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:51:57,815 INFO [train.py:1003] (0/2) Computing validation loss
+2023-02-28 23:52:07,258 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.294, simple_loss=0.3813, pruned_loss=0.1034, over 944034.00 frames. 
+2023-02-28 23:52:07,258 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19478MB
+2023-02-28 23:52:13,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2242, 1.2161, 1.0887, 1.2877], device='cuda:0'), covar=tensor([0.1802, 0.1806, 0.1489, 0.1809], device='cuda:0'), in_proj_covar=tensor([0.0929, 0.0774, 0.0857, 0.0910], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 23:52:17,623 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48681.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:52:35,462 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48699.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:52:37,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5105, 2.4320, 1.7782, 1.9102], device='cuda:0'), covar=tensor([0.0635, 0.0641, 0.0845, 0.0951], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0542, 0.0575, 0.0502], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-02-28 23:52:38,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48702.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:52:52,701 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.898e+02 1.356e+03 1.909e+03 2.404e+03 8.049e+03, threshold=3.817e+03, percent-clipped=11.0
+2023-02-28 23:52:52,933 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48717.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:52:54,057 INFO [train.py:968] (0/2) Epoch 2, batch 3050, libri_loss[loss=0.3987, simple_loss=0.4404, pruned_loss=0.1785, over 19692.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4073, pruned_loss=0.1545, over 5668492.58 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4221, pruned_loss=0.1576, over 4554464.55 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4047, pruned_loss=0.1533, over 5667902.56 frames. ], batch size: 187, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:52:55,198 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48720.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 23:52:57,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48723.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:53:06,293 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48731.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:53:18,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6987, 1.3453, 1.4617, 1.9527], device='cuda:0'), covar=tensor([0.1724, 0.1912, 0.1445, 0.1960], device='cuda:0'), in_proj_covar=tensor([0.0925, 0.0774, 0.0856, 0.0903], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-02-28 23:53:28,308 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48752.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 23:53:46,196 INFO [train.py:968] (0/2) Epoch 2, batch 3100, giga_loss[loss=0.3025, simple_loss=0.3677, pruned_loss=0.1186, over 28532.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.3986, pruned_loss=0.1471, over 5678270.81 frames. ], libri_tot_loss[loss=0.368, simple_loss=0.4215, pruned_loss=0.1573, over 4574789.08 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3967, pruned_loss=0.1463, over 5674593.00 frames. ], batch size: 336, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:54:06,948 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48790.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:54:26,654 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-02-28 23:54:37,728 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48816.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:54:38,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.150e+02 9.638e+02 1.146e+03 1.588e+03 3.034e+03, threshold=2.292e+03, percent-clipped=0.0
+2023-02-28 23:54:39,508 INFO [train.py:968] (0/2) Epoch 2, batch 3150, giga_loss[loss=0.3182, simple_loss=0.3849, pruned_loss=0.1258, over 28961.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3968, pruned_loss=0.1447, over 5677438.92 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4213, pruned_loss=0.1571, over 4592920.11 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3951, pruned_loss=0.1441, over 5672941.09 frames. ], batch size: 213, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:54:54,235 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-02-28 23:54:59,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48840.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:55:23,487 INFO [train.py:968] (0/2) Epoch 2, batch 3200, giga_loss[loss=0.3969, simple_loss=0.442, pruned_loss=0.1759, over 28265.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3985, pruned_loss=0.1455, over 5671352.18 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4211, pruned_loss=0.157, over 4628364.76 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3966, pruned_loss=0.1447, over 5669393.91 frames. ], batch size: 368, lr: 1.56e-02, grad_scale: 8.0
+2023-02-28 23:56:05,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.700e+02 1.168e+03 1.579e+03 2.230e+03 6.467e+03, threshold=3.158e+03, percent-clipped=23.0
+2023-02-28 23:56:06,408 INFO [train.py:968] (0/2) Epoch 2, batch 3250, giga_loss[loss=0.3894, simple_loss=0.4308, pruned_loss=0.174, over 28670.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.4022, pruned_loss=0.1478, over 5673662.59 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.421, pruned_loss=0.1568, over 4647280.79 frames. ], giga_tot_loss[loss=0.3474, simple_loss=0.4004, pruned_loss=0.1471, over 5669123.93 frames. ], batch size: 92, lr: 1.56e-02, grad_scale: 8.0
+2023-02-28 23:56:06,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6985, 2.1699, 1.8075, 1.6898], device='cuda:0'), covar=tensor([0.1445, 0.1414, 0.1076, 0.0710], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0824, 0.0703, 0.0609], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:0')
+2023-02-28 23:56:51,399 INFO [train.py:968] (0/2) Epoch 2, batch 3300, giga_loss[loss=0.3146, simple_loss=0.3755, pruned_loss=0.1269, over 28649.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.4042, pruned_loss=0.1491, over 5691876.88 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4206, pruned_loss=0.1567, over 4684338.41 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.4027, pruned_loss=0.1484, over 5682411.97 frames. ], batch size: 66, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:57:00,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48980.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:57:02,955 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48983.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:57:06,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48986.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:57:31,379 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49015.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:57:33,801 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.867e+02 1.161e+03 1.440e+03 1.810e+03 3.366e+03, threshold=2.879e+03, percent-clipped=2.0
+2023-02-28 23:57:34,516 INFO [train.py:968] (0/2) Epoch 2, batch 3350, giga_loss[loss=0.3449, simple_loss=0.4016, pruned_loss=0.1441, over 28905.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.4062, pruned_loss=0.1506, over 5690588.20 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4205, pruned_loss=0.1565, over 4723282.20 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.4045, pruned_loss=0.15, over 5678754.40 frames. ], batch size: 112, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:58:18,136 INFO [train.py:968] (0/2) Epoch 2, batch 3400, giga_loss[loss=0.3807, simple_loss=0.4178, pruned_loss=0.1718, over 28536.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4073, pruned_loss=0.1519, over 5692340.21 frames. ], libri_tot_loss[loss=0.3664, simple_loss=0.4203, pruned_loss=0.1562, over 4744924.00 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4059, pruned_loss=0.1516, over 5680933.08 frames. ], batch size: 71, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:58:39,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49092.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:59:01,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.929e+02 1.287e+03 1.625e+03 2.115e+03 8.356e+03, threshold=3.250e+03, percent-clipped=11.0
+2023-02-28 23:59:02,341 INFO [train.py:968] (0/2) Epoch 2, batch 3450, giga_loss[loss=0.3944, simple_loss=0.439, pruned_loss=0.1749, over 28960.00 frames. ], tot_loss[loss=0.3592, simple_loss=0.4096, pruned_loss=0.1545, over 5689142.67 frames. ], libri_tot_loss[loss=0.3664, simple_loss=0.4205, pruned_loss=0.1562, over 4773278.70 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.408, pruned_loss=0.1541, over 5675562.15 frames. ], batch size: 174, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:59:06,351 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49123.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:59:08,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49126.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 23:59:33,373 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49155.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:59:41,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49165.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:59:43,584 INFO [train.py:968] (0/2) Epoch 2, batch 3500, giga_loss[loss=0.3434, simple_loss=0.4045, pruned_loss=0.1411, over 28864.00 frames. ], tot_loss[loss=0.3586, simple_loss=0.41, pruned_loss=0.1536, over 5694688.59 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4199, pruned_loss=0.1557, over 4803111.86 frames. ], giga_tot_loss[loss=0.3582, simple_loss=0.4089, pruned_loss=0.1537, over 5681039.05 frames. ], batch size: 92, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:59:55,145 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-01 00:00:02,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:00:22,795 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=49214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:00:25,654 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7586, 2.0281, 1.7051, 1.7077], device='cuda:0'), covar=tensor([0.1365, 0.1293, 0.1016, 0.0619], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0831, 0.0713, 0.0608], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 00:00:26,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.499e+02 1.094e+03 1.438e+03 1.971e+03 3.989e+03, threshold=2.876e+03, percent-clipped=4.0
+2023-03-01 00:00:26,710 INFO [train.py:968] (0/2) Epoch 2, batch 3550, giga_loss[loss=0.3695, simple_loss=0.4317, pruned_loss=0.1537, over 28898.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4091, pruned_loss=0.1513, over 5694918.77 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4199, pruned_loss=0.1558, over 4814720.82 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4081, pruned_loss=0.1513, over 5685450.01 frames. ], batch size: 227, lr: 1.56e-02, grad_scale: 4.0
+2023-03-01 00:00:35,889 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=49231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:00:38,592 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:00:41,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:01:07,679 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2339, 1.2041, 1.1944, 1.2752], device='cuda:0'), covar=tensor([0.1746, 0.1622, 0.1323, 0.1760], device='cuda:0'), in_proj_covar=tensor([0.0914, 0.0773, 0.0853, 0.0904], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 00:01:08,336 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:01:12,443 INFO [train.py:968] (0/2) Epoch 2, batch 3600, giga_loss[loss=0.4104, simple_loss=0.4203, pruned_loss=0.2002, over 23764.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.409, pruned_loss=0.1507, over 5694297.52 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4197, pruned_loss=0.1557, over 4841427.45 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4081, pruned_loss=0.1506, over 5682634.18 frames. ], batch size: 705, lr: 1.56e-02, grad_scale: 8.0
+2023-03-01 00:01:40,526 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-01 00:01:49,553 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:01:51,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:01:57,686 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.850e+02 1.157e+03 1.387e+03 1.941e+03 6.378e+03, threshold=2.774e+03, percent-clipped=5.0
+2023-03-01 00:01:58,475 INFO [train.py:968] (0/2) Epoch 2, batch 3650, giga_loss[loss=0.3384, simple_loss=0.3934, pruned_loss=0.1417, over 28584.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4069, pruned_loss=0.1486, over 5706647.09 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4198, pruned_loss=0.1557, over 4867462.90 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4058, pruned_loss=0.1484, over 5693028.66 frames. ], batch size: 71, lr: 1.56e-02, grad_scale: 8.0
+2023-03-01 00:02:15,596 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:02:17,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:02:19,864 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:02:45,319 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:02:47,198 INFO [train.py:968] (0/2) Epoch 2, batch 3700, giga_loss[loss=0.3513, simple_loss=0.396, pruned_loss=0.1533, over 28973.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4039, pruned_loss=0.1475, over 5703335.70 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.4195, pruned_loss=0.1555, over 4903048.63 frames. ], giga_tot_loss[loss=0.3486, simple_loss=0.4028, pruned_loss=0.1472, over 5685961.95 frames. ], batch size: 106, lr: 1.56e-02, grad_scale: 4.0
+2023-03-01 00:02:50,419 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.81 vs. limit=5.0
+2023-03-01 00:03:20,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6616, 2.2629, 1.7571, 1.3505], device='cuda:0'), covar=tensor([0.0868, 0.0611, 0.0775, 0.1494], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0439, 0.0335, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0010, 0.0014], device='cuda:0')
+2023-03-01 00:03:30,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.454e+02 1.184e+03 1.426e+03 1.797e+03 5.283e+03, threshold=2.851e+03, percent-clipped=7.0
+2023-03-01 00:03:30,766 INFO [train.py:968] (0/2) Epoch 2, batch 3750, giga_loss[loss=0.365, simple_loss=0.4143, pruned_loss=0.1578, over 28995.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.4003, pruned_loss=0.1449, over 5710014.21 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4194, pruned_loss=0.1554, over 4912906.90 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3993, pruned_loss=0.1446, over 5694675.54 frames. ], batch size: 128, lr: 1.56e-02, grad_scale: 4.0
+2023-03-01 00:03:50,883 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-01 00:04:13,829 INFO [train.py:968] (0/2) Epoch 2, batch 3800, giga_loss[loss=0.4015, simple_loss=0.4392, pruned_loss=0.1819, over 28853.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.4025, pruned_loss=0.1473, over 5701100.78 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4198, pruned_loss=0.1558, over 4938053.31 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.4008, pruned_loss=0.1465, over 5690270.24 frames. ], batch size: 227, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:04:26,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 00:04:55,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.471e+02 1.172e+03 1.420e+03 1.995e+03 5.001e+03, threshold=2.839e+03, percent-clipped=7.0
+2023-03-01 00:04:55,287 INFO [train.py:968] (0/2) Epoch 2, batch 3850, giga_loss[loss=0.3361, simple_loss=0.3972, pruned_loss=0.1375, over 28454.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.4033, pruned_loss=0.1477, over 5705813.45 frames. ], libri_tot_loss[loss=0.3658, simple_loss=0.4197, pruned_loss=0.1559, over 4956870.73 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.4018, pruned_loss=0.1469, over 5693931.99 frames. ], batch size: 65, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:05:34,280 INFO [train.py:968] (0/2) Epoch 2, batch 3900, giga_loss[loss=0.3489, simple_loss=0.4075, pruned_loss=0.1451, over 28911.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.4022, pruned_loss=0.1456, over 5704909.99 frames. ], libri_tot_loss[loss=0.3659, simple_loss=0.4198, pruned_loss=0.156, over 4971462.11 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.4006, pruned_loss=0.1446, over 5699332.81 frames. ], batch size: 136, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:05:51,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:06:07,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:06:17,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.767e+02 9.997e+02 1.334e+03 1.816e+03 5.524e+03, threshold=2.669e+03, percent-clipped=8.0
+2023-03-01 00:06:17,427 INFO [train.py:968] (0/2) Epoch 2, batch 3950, giga_loss[loss=0.3298, simple_loss=0.3979, pruned_loss=0.1309, over 28811.00 frames. ], tot_loss[loss=0.345, simple_loss=0.4013, pruned_loss=0.1444, over 5713798.22 frames. ], libri_tot_loss[loss=0.3649, simple_loss=0.4187, pruned_loss=0.1555, over 5003139.46 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.4002, pruned_loss=0.1436, over 5703297.26 frames. ], batch size: 119, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:06:26,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=49629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:06:58,544 INFO [train.py:968] (0/2) Epoch 2, batch 4000, giga_loss[loss=0.3027, simple_loss=0.3644, pruned_loss=0.1205, over 28831.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.4006, pruned_loss=0.1439, over 5707680.46 frames. ], libri_tot_loss[loss=0.3648, simple_loss=0.4187, pruned_loss=0.1555, over 5007643.76 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3997, pruned_loss=0.1433, over 5698601.36 frames. ], batch size: 119, lr: 1.55e-02, grad_scale: 8.0
+2023-03-01 00:07:39,266 INFO [train.py:968] (0/2) Epoch 2, batch 4050, libri_loss[loss=0.3748, simple_loss=0.431, pruned_loss=0.1593, over 29519.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.4004, pruned_loss=0.1449, over 5708766.11 frames. ], libri_tot_loss[loss=0.3648, simple_loss=0.4187, pruned_loss=0.1554, over 5025039.01 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3992, pruned_loss=0.1441, over 5704195.29 frames. ], batch size: 81, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:07:39,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.594e+02 1.102e+03 1.351e+03 1.804e+03 2.808e+03, threshold=2.701e+03, percent-clipped=2.0
+2023-03-01 00:07:49,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:07:51,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:08:03,443 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:08:05,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:08:05,507 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 00:08:16,358 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:08:19,897 INFO [train.py:968] (0/2) Epoch 2, batch 4100, giga_loss[loss=0.3075, simple_loss=0.3728, pruned_loss=0.1211, over 28917.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3971, pruned_loss=0.1427, over 5712331.87 frames. ], libri_tot_loss[loss=0.3649, simple_loss=0.4187, pruned_loss=0.1555, over 5038166.12 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3958, pruned_loss=0.1419, over 5705986.43 frames. ], batch size: 227, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:08:29,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:08:58,621 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=49815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:09:01,118 INFO [train.py:968] (0/2) Epoch 2, batch 4150, giga_loss[loss=0.3375, simple_loss=0.3925, pruned_loss=0.1412, over 28480.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3943, pruned_loss=0.1412, over 5716850.53 frames. ], libri_tot_loss[loss=0.3649, simple_loss=0.4185, pruned_loss=0.1557, over 5064053.11 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3927, pruned_loss=0.1401, over 5706524.37 frames. ], batch size: 65, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:09:01,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.965e+02 1.159e+03 1.584e+03 2.136e+03 5.562e+03, threshold=3.168e+03, percent-clipped=13.0
+2023-03-01 00:09:05,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=49825.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:09:40,674 INFO [train.py:968] (0/2) Epoch 2, batch 4200, giga_loss[loss=0.3545, simple_loss=0.4056, pruned_loss=0.1517, over 28880.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3953, pruned_loss=0.1428, over 5720020.77 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4185, pruned_loss=0.1559, over 5085161.68 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3936, pruned_loss=0.1414, over 5707197.96 frames. ], batch size: 227, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:09:41,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3041, 1.9600, 1.4995, 0.4894], device='cuda:0'), covar=tensor([0.1294, 0.0728, 0.1200, 0.1499], device='cuda:0'), in_proj_covar=tensor([0.1097, 0.1062, 0.1085, 0.0927], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 00:10:14,637 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-01 00:10:21,175 INFO [train.py:968] (0/2) Epoch 2, batch 4250, giga_loss[loss=0.3275, simple_loss=0.3746, pruned_loss=0.1402, over 28766.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3966, pruned_loss=0.1448, over 5713989.37 frames. ], libri_tot_loss[loss=0.3663, simple_loss=0.4192, pruned_loss=0.1567, over 5107653.96 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3937, pruned_loss=0.1426, over 5706252.34 frames. ], batch size: 119, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:10:21,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.480e+02 1.362e+03 1.702e+03 2.174e+03 8.707e+03, threshold=3.404e+03, percent-clipped=9.0
+2023-03-01 00:10:33,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5878, 1.4686, 1.3749, 1.3905], device='cuda:0'), covar=tensor([0.0735, 0.1248, 0.1243, 0.1028], device='cuda:0'), in_proj_covar=tensor([0.0505, 0.0825, 0.0633, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 00:10:50,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-01 00:11:06,015 INFO [train.py:968] (0/2) Epoch 2, batch 4300, giga_loss[loss=0.3309, simple_loss=0.3912, pruned_loss=0.1353, over 28670.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3958, pruned_loss=0.1452, over 5710709.82 frames. ], libri_tot_loss[loss=0.3658, simple_loss=0.4189, pruned_loss=0.1564, over 5129544.55 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3931, pruned_loss=0.1433, over 5700843.97 frames. ], batch size: 262, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:11:22,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5619, 1.4385, 4.1164, 3.0240], device='cuda:0'), covar=tensor([0.1598, 0.1506, 0.0254, 0.0464], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0476, 0.0601, 0.0495], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 00:11:28,469 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-50000.pt
+2023-03-01 00:11:34,892 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:11:48,357 INFO [train.py:968] (0/2) Epoch 2, batch 4350, giga_loss[loss=0.3531, simple_loss=0.3966, pruned_loss=0.1549, over 27972.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3914, pruned_loss=0.1427, over 5713727.55 frames. ], libri_tot_loss[loss=0.3655, simple_loss=0.4187, pruned_loss=0.1562, over 5137349.50 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3892, pruned_loss=0.1413, over 5704371.17 frames. ], batch size: 412, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:11:48,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8990, 1.7041, 1.3650, 1.4294], device='cuda:0'), covar=tensor([0.0621, 0.0825, 0.1035, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0532, 0.0553, 0.0492], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 00:11:48,978 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.871e+02 1.104e+03 1.302e+03 1.822e+03 7.282e+03, threshold=2.605e+03, percent-clipped=5.0
+2023-03-01 00:12:01,896 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2986, 1.3963, 1.0383, 1.4029], device='cuda:0'), covar=tensor([0.1154, 0.0515, 0.0605, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0231, 0.0233, 0.0400], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0020, 0.0018, 0.0029], device='cuda:0')
+2023-03-01 00:12:38,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-01 00:12:39,188 INFO [train.py:968] (0/2) Epoch 2, batch 4400, giga_loss[loss=0.4838, simple_loss=0.4756, pruned_loss=0.246, over 26719.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3882, pruned_loss=0.1412, over 5707853.85 frames. ], libri_tot_loss[loss=0.3649, simple_loss=0.4181, pruned_loss=0.1558, over 5147409.33 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3866, pruned_loss=0.1402, over 5699221.16 frames. ], batch size: 555, lr: 1.55e-02, grad_scale: 8.0
+2023-03-01 00:13:25,804 INFO [train.py:968] (0/2) Epoch 2, batch 4450, giga_loss[loss=0.3271, simple_loss=0.3753, pruned_loss=0.1394, over 28665.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3896, pruned_loss=0.1415, over 5714763.24 frames. ], libri_tot_loss[loss=0.3654, simple_loss=0.4186, pruned_loss=0.1561, over 5166911.06 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3871, pruned_loss=0.14, over 5705370.12 frames. ], batch size: 85, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:13:27,243 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.629e+02 1.184e+03 1.432e+03 2.014e+03 4.289e+03, threshold=2.864e+03, percent-clipped=7.0
+2023-03-01 00:13:53,564 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:13:55,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:14:03,709 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=50159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:14:14,522 INFO [train.py:968] (0/2) Epoch 2, batch 4500, libri_loss[loss=0.3384, simple_loss=0.3958, pruned_loss=0.1404, over 29538.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3903, pruned_loss=0.141, over 5715501.21 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4189, pruned_loss=0.1562, over 5184684.20 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3875, pruned_loss=0.1395, over 5704190.06 frames. ], batch size: 76, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:14:22,311 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:14:31,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:14:39,871 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50200.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:14:55,093 INFO [train.py:968] (0/2) Epoch 2, batch 4550, giga_loss[loss=0.3733, simple_loss=0.423, pruned_loss=0.1619, over 28992.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3938, pruned_loss=0.1425, over 5722831.25 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4192, pruned_loss=0.1564, over 5196471.57 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3908, pruned_loss=0.1409, over 5712905.26 frames. ], batch size: 155, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:14:57,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.144e+02 1.063e+03 1.473e+03 2.005e+03 7.142e+03, threshold=2.946e+03, percent-clipped=14.0
+2023-03-01 00:15:38,003 INFO [train.py:968] (0/2) Epoch 2, batch 4600, libri_loss[loss=0.3145, simple_loss=0.3707, pruned_loss=0.1292, over 29507.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3961, pruned_loss=0.1438, over 5713835.99 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.419, pruned_loss=0.1563, over 5218649.33 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3932, pruned_loss=0.1422, over 5701769.28 frames. ], batch size: 70, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:15:42,614 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2168, 1.6154, 1.1326, 1.2800], device='cuda:0'), covar=tensor([0.0902, 0.0597, 0.0953, 0.1382], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0440, 0.0346, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:0')
+2023-03-01 00:15:53,695 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1650, 1.7728, 1.6316, 1.5618], device='cuda:0'), covar=tensor([0.0647, 0.0886, 0.0970, 0.1242], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0533, 0.0570, 0.0497], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 00:16:23,457 INFO [train.py:968] (0/2) Epoch 2, batch 4650, giga_loss[loss=0.3124, simple_loss=0.3754, pruned_loss=0.1247, over 28849.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3962, pruned_loss=0.1429, over 5707019.76 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4196, pruned_loss=0.1568, over 5233308.20 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3929, pruned_loss=0.1409, over 5695494.61 frames. ], batch size: 199, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:16:25,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.605e+02 1.083e+03 1.308e+03 1.929e+03 5.433e+03, threshold=2.616e+03, percent-clipped=9.0
+2023-03-01 00:16:35,360 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5190, 2.0349, 1.5160, 0.6521], device='cuda:0'), covar=tensor([0.1124, 0.0755, 0.1216, 0.1757], device='cuda:0'), in_proj_covar=tensor([0.1073, 0.1056, 0.1079, 0.0943], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 00:16:36,668 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:16:39,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:16:44,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50343.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:16:44,883 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 00:16:46,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50346.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:16:54,554 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5625, 2.3572, 1.4384, 1.2145], device='cuda:0'), covar=tensor([0.0846, 0.0581, 0.0931, 0.1477], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0437, 0.0338, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0010, 0.0014], device='cuda:0')
+2023-03-01 00:17:03,756 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:17:07,613 INFO [train.py:968] (0/2) Epoch 2, batch 4700, giga_loss[loss=0.3241, simple_loss=0.3883, pruned_loss=0.13, over 28713.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3951, pruned_loss=0.1415, over 5702799.22 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.4192, pruned_loss=0.1565, over 5237699.97 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3926, pruned_loss=0.1401, over 5693983.21 frames. ], batch size: 262, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:17:12,334 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50375.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:17:49,866 INFO [train.py:968] (0/2) Epoch 2, batch 4750, giga_loss[loss=0.3576, simple_loss=0.4085, pruned_loss=0.1533, over 28959.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3977, pruned_loss=0.1438, over 5706777.63 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4196, pruned_loss=0.1569, over 5244048.20 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.395, pruned_loss=0.1421, over 5700283.32 frames. ], batch size: 136, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:17:51,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.489e+02 1.259e+03 1.559e+03 1.920e+03 4.634e+03, threshold=3.118e+03, percent-clipped=6.0
+2023-03-01 00:17:53,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6346, 2.0073, 1.8533, 1.7680], device='cuda:0'), covar=tensor([0.1177, 0.1353, 0.0993, 0.0650], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0818, 0.0719, 0.0607], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 00:18:15,319 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 00:18:27,843 INFO [train.py:968] (0/2) Epoch 2, batch 4800, giga_loss[loss=0.3224, simple_loss=0.3826, pruned_loss=0.1311, over 28825.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.4004, pruned_loss=0.1466, over 5710407.03 frames. ], libri_tot_loss[loss=0.368, simple_loss=0.42, pruned_loss=0.158, over 5278716.74 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3965, pruned_loss=0.1435, over 5702449.98 frames. ], batch size: 186, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:19:12,389 INFO [train.py:968] (0/2) Epoch 2, batch 4850, giga_loss[loss=0.3434, simple_loss=0.4059, pruned_loss=0.1405, over 28827.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4022, pruned_loss=0.1484, over 5710860.48 frames. ], libri_tot_loss[loss=0.3681, simple_loss=0.4199, pruned_loss=0.1582, over 5292607.60 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.3988, pruned_loss=0.1456, over 5701700.16 frames. ], batch size: 119, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:19:15,004 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=50521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:19:15,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.542e+02 1.407e+03 2.006e+03 3.089e+03 9.896e+03, threshold=4.011e+03, percent-clipped=24.0
+2023-03-01 00:19:24,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:19:53,599 INFO [train.py:968] (0/2) Epoch 2, batch 4900, giga_loss[loss=0.3337, simple_loss=0.3936, pruned_loss=0.1369, over 28544.00 frames. ], tot_loss[loss=0.353, simple_loss=0.4056, pruned_loss=0.1502, over 5717436.09 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.4209, pruned_loss=0.1587, over 5317362.33 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.4013, pruned_loss=0.147, over 5704360.76 frames. ], batch size: 60, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:20:04,617 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-01 00:20:13,070 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=50590.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:20:13,292 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-01 00:20:35,358 INFO [train.py:968] (0/2) Epoch 2, batch 4950, giga_loss[loss=0.3991, simple_loss=0.4417, pruned_loss=0.1783, over 28737.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4075, pruned_loss=0.1511, over 5713409.22 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4212, pruned_loss=0.1588, over 5321944.60 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.4036, pruned_loss=0.1483, over 5706140.67 frames. ], batch size: 262, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:20:37,869 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.383e+02 1.292e+03 1.659e+03 2.070e+03 4.553e+03, threshold=3.317e+03, percent-clipped=1.0
+2023-03-01 00:20:45,322 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 00:20:54,682 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-01 00:21:15,442 INFO [train.py:968] (0/2) Epoch 2, batch 5000, giga_loss[loss=0.3417, simple_loss=0.4038, pruned_loss=0.1398, over 28908.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.4071, pruned_loss=0.1503, over 5718180.91 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.421, pruned_loss=0.1585, over 5340605.98 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.4036, pruned_loss=0.1481, over 5707722.41 frames. ], batch size: 186, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:21:20,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5052, 1.8772, 1.6563, 1.7006], device='cuda:0'), covar=tensor([0.1186, 0.1400, 0.1080, 0.0640], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0799, 0.0710, 0.0598], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 00:21:23,359 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:21:25,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:21:25,537 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-01 00:21:32,554 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0308, 0.8529, 0.7690, 1.2306], device='cuda:0'), covar=tensor([0.1100, 0.0529, 0.0596, 0.1274], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0230, 0.0232, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0020, 0.0018, 0.0029], device='cuda:0')
+2023-03-01 00:21:51,674 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:21:58,967 INFO [train.py:968] (0/2) Epoch 2, batch 5050, giga_loss[loss=0.3124, simple_loss=0.374, pruned_loss=0.1254, over 28982.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4062, pruned_loss=0.1495, over 5710848.49 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4209, pruned_loss=0.1584, over 5343485.95 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4034, pruned_loss=0.1478, over 5701664.27 frames. ], batch size: 112, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:22:00,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.125e+02 1.137e+03 1.443e+03 1.848e+03 6.243e+03, threshold=2.886e+03, percent-clipped=6.0
+2023-03-01 00:22:31,191 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-01 00:22:35,975 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0777, 1.1591, 1.0074, 0.9759], device='cuda:0'), covar=tensor([0.1593, 0.1706, 0.1444, 0.1552], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0778, 0.0855, 0.0924], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 00:22:38,258 INFO [train.py:968] (0/2) Epoch 2, batch 5100, giga_loss[loss=0.3085, simple_loss=0.3763, pruned_loss=0.1203, over 28791.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4061, pruned_loss=0.1491, over 5716498.52 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.421, pruned_loss=0.1584, over 5364890.48 frames. ], giga_tot_loss[loss=0.3489, simple_loss=0.4032, pruned_loss=0.1474, over 5702701.34 frames. ], batch size: 119, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:22:54,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2614, 1.4333, 1.1836, 0.8134], device='cuda:0'), covar=tensor([0.0668, 0.0441, 0.0345, 0.0574], device='cuda:0'), in_proj_covar=tensor([0.0956, 0.0669, 0.0780, 0.0796], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 00:23:19,567 INFO [train.py:968] (0/2) Epoch 2, batch 5150, giga_loss[loss=0.3658, simple_loss=0.4067, pruned_loss=0.1624, over 28747.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.4054, pruned_loss=0.1488, over 5711644.17 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4209, pruned_loss=0.1582, over 5371883.52 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4028, pruned_loss=0.1473, over 5704378.40 frames. ], batch size: 99, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:23:21,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.306e+02 1.305e+03 1.605e+03 2.107e+03 7.549e+03, threshold=3.210e+03, percent-clipped=14.0
+2023-03-01 00:23:22,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.94 vs. limit=5.0
+2023-03-01 00:24:01,860 INFO [train.py:968] (0/2) Epoch 2, batch 5200, giga_loss[loss=0.2954, simple_loss=0.3639, pruned_loss=0.1135, over 28912.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.4006, pruned_loss=0.1461, over 5698206.49 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4209, pruned_loss=0.1584, over 5369267.06 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3982, pruned_loss=0.1446, over 5700082.41 frames. ], batch size: 186, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:24:24,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:24:43,816 INFO [train.py:968] (0/2) Epoch 2, batch 5250, giga_loss[loss=0.3064, simple_loss=0.3658, pruned_loss=0.1235, over 28480.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3968, pruned_loss=0.1437, over 5705370.46 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4215, pruned_loss=0.1589, over 5376891.81 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3941, pruned_loss=0.1419, over 5704504.97 frames. ], batch size: 71, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:24:48,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.326e+02 1.045e+03 1.307e+03 1.686e+03 4.307e+03, threshold=2.615e+03, percent-clipped=1.0
+2023-03-01 00:25:00,626 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=50939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:25:22,301 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50965.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:25:25,785 INFO [train.py:968] (0/2) Epoch 2, batch 5300, giga_loss[loss=0.3399, simple_loss=0.4093, pruned_loss=0.1353, over 29002.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.3995, pruned_loss=0.1443, over 5700438.45 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4224, pruned_loss=0.1594, over 5380120.65 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3962, pruned_loss=0.1422, over 5702591.35 frames. ], batch size: 164, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:26:06,435 INFO [train.py:968] (0/2) Epoch 2, batch 5350, giga_loss[loss=0.3393, simple_loss=0.4013, pruned_loss=0.1387, over 28899.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.4018, pruned_loss=0.1445, over 5701300.49 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4222, pruned_loss=0.1595, over 5388893.49 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3986, pruned_loss=0.1422, over 5706221.67 frames. ], batch size: 164, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:26:10,044 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.093e+02 1.457e+03 1.989e+03 2.645e+03 7.720e+03, threshold=3.978e+03, percent-clipped=25.0
+2023-03-01 00:26:16,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4683, 3.2338, 2.5480, 2.0858], device='cuda:0'), covar=tensor([0.0737, 0.0538, 0.0684, 0.1223], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0442, 0.0346, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:0')
+2023-03-01 00:26:22,737 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:26:25,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:26:48,266 INFO [train.py:968] (0/2) Epoch 2, batch 5400, giga_loss[loss=0.3492, simple_loss=0.3853, pruned_loss=0.1565, over 28541.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.4025, pruned_loss=0.1454, over 5709085.37 frames. ], libri_tot_loss[loss=0.3705, simple_loss=0.4222, pruned_loss=0.1594, over 5399273.94 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3996, pruned_loss=0.1434, over 5709290.65 frames. ], batch size: 71, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:26:49,933 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:27:18,313 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51108.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:27:21,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51111.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:27:26,054 INFO [train.py:968] (0/2) Epoch 2, batch 5450, giga_loss[loss=0.3512, simple_loss=0.3867, pruned_loss=0.1578, over 28658.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.4005, pruned_loss=0.146, over 5718233.27 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4216, pruned_loss=0.1591, over 5416896.57 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3979, pruned_loss=0.144, over 5714979.63 frames. ], batch size: 92, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:27:27,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9849, 1.6797, 1.7061, 1.5408], device='cuda:0'), covar=tensor([0.0796, 0.1927, 0.1200, 0.1569], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0826, 0.0643, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 00:27:28,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.356e+02 1.267e+03 1.549e+03 2.076e+03 5.984e+03, threshold=3.097e+03, percent-clipped=2.0
+2023-03-01 00:27:45,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51140.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:28:10,851 INFO [train.py:968] (0/2) Epoch 2, batch 5500, giga_loss[loss=0.3389, simple_loss=0.3815, pruned_loss=0.1481, over 28924.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.399, pruned_loss=0.1466, over 5723574.33 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4216, pruned_loss=0.1591, over 5422551.94 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.3965, pruned_loss=0.1448, over 5720051.12 frames. ], batch size: 106, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:28:17,342 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:28:39,305 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51201.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:28:54,510 INFO [train.py:968] (0/2) Epoch 2, batch 5550, giga_loss[loss=0.3277, simple_loss=0.3858, pruned_loss=0.1348, over 28697.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3972, pruned_loss=0.1472, over 5725626.54 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4223, pruned_loss=0.1595, over 5427819.42 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3944, pruned_loss=0.1453, over 5722209.87 frames. ], batch size: 307, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:28:57,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.446e+02 1.139e+03 1.453e+03 2.051e+03 4.297e+03, threshold=2.906e+03, percent-clipped=7.0
+2023-03-01 00:29:26,026 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 00:29:39,254 INFO [train.py:968] (0/2) Epoch 2, batch 5600, giga_loss[loss=0.3354, simple_loss=0.3915, pruned_loss=0.1397, over 28980.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3952, pruned_loss=0.1462, over 5724154.10 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.4225, pruned_loss=0.1596, over 5432063.94 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3926, pruned_loss=0.1445, over 5720314.56 frames. ], batch size: 213, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:30:17,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:30:22,905 INFO [train.py:968] (0/2) Epoch 2, batch 5650, giga_loss[loss=0.3147, simple_loss=0.3703, pruned_loss=0.1296, over 29126.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3924, pruned_loss=0.1445, over 5714246.82 frames. ], libri_tot_loss[loss=0.3704, simple_loss=0.4222, pruned_loss=0.1593, over 5439601.78 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3898, pruned_loss=0.143, over 5710428.42 frames. ], batch size: 155, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:30:25,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.181e+02 1.122e+03 1.382e+03 1.721e+03 5.210e+03, threshold=2.765e+03, percent-clipped=2.0
+2023-03-01 00:30:38,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:30:38,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1171, 1.8121, 1.4725, 1.6603], device='cuda:0'), covar=tensor([0.0572, 0.0756, 0.0914, 0.0996], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0529, 0.0557, 0.0494], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 00:31:00,012 INFO [train.py:968] (0/2) Epoch 2, batch 5700, giga_loss[loss=0.3082, simple_loss=0.3673, pruned_loss=0.1245, over 29051.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3884, pruned_loss=0.1417, over 5717698.20 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.4225, pruned_loss=0.1596, over 5451472.14 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.385, pruned_loss=0.1398, over 5712075.26 frames. ], batch size: 164, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:31:43,424 INFO [train.py:968] (0/2) Epoch 2, batch 5750, giga_loss[loss=0.3129, simple_loss=0.3668, pruned_loss=0.1295, over 28732.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3834, pruned_loss=0.1388, over 5713485.17 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4228, pruned_loss=0.1598, over 5456286.25 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3802, pruned_loss=0.137, over 5707068.38 frames. ], batch size: 99, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:31:47,181 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.049e+02 1.203e+03 1.556e+03 2.430e+03 7.261e+03, threshold=3.112e+03, percent-clipped=18.0
+2023-03-01 00:32:12,247 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:32:13,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:32:15,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:32:21,483 INFO [train.py:968] (0/2) Epoch 2, batch 5800, giga_loss[loss=0.2927, simple_loss=0.3643, pruned_loss=0.1105, over 29105.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3843, pruned_loss=0.1391, over 5711577.01 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4235, pruned_loss=0.1603, over 5461848.27 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.38, pruned_loss=0.1366, over 5708173.04 frames. ], batch size: 113, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:32:38,553 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:33:03,675 INFO [train.py:968] (0/2) Epoch 2, batch 5850, giga_loss[loss=0.3127, simple_loss=0.3706, pruned_loss=0.1274, over 28592.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3883, pruned_loss=0.1413, over 5710456.50 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.4238, pruned_loss=0.1605, over 5465314.73 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3843, pruned_loss=0.139, over 5706991.81 frames. ], batch size: 60, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:33:07,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.731e+02 1.223e+03 1.459e+03 1.854e+03 6.841e+03, threshold=2.917e+03, percent-clipped=4.0
+2023-03-01 00:33:28,191 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:33:39,013 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51565.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:33:41,420 INFO [train.py:968] (0/2) Epoch 2, batch 5900, giga_loss[loss=0.4125, simple_loss=0.4494, pruned_loss=0.1878, over 28257.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3934, pruned_loss=0.1438, over 5713258.81 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.4239, pruned_loss=0.1605, over 5483298.58 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3887, pruned_loss=0.1412, over 5703901.51 frames. ], batch size: 368, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:33:46,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4127, 1.3303, 1.1820, 1.7482], device='cuda:0'), covar=tensor([0.1633, 0.1673, 0.1460, 0.1996], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0775, 0.0843, 0.0924], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:0')
+2023-03-01 00:33:47,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51576.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:34:23,614 INFO [train.py:968] (0/2) Epoch 2, batch 5950, giga_loss[loss=0.3575, simple_loss=0.4086, pruned_loss=0.1532, over 28404.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3969, pruned_loss=0.1453, over 5718461.87 frames. ], libri_tot_loss[loss=0.3718, simple_loss=0.4232, pruned_loss=0.1602, over 5492612.87 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.393, pruned_loss=0.143, over 5707857.39 frames. ], batch size: 71, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:34:28,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.846e+02 1.233e+03 1.609e+03 2.153e+03 5.454e+03, threshold=3.218e+03, percent-clipped=6.0
+2023-03-01 00:34:30,874 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 00:35:05,342 INFO [train.py:968] (0/2) Epoch 2, batch 6000, libri_loss[loss=0.4035, simple_loss=0.4524, pruned_loss=0.1773, over 25630.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.4018, pruned_loss=0.1483, over 5715213.07 frames. ], libri_tot_loss[loss=0.373, simple_loss=0.4242, pruned_loss=0.1609, over 5501778.84 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.397, pruned_loss=0.1454, over 5705642.61 frames. ], batch size: 137, lr: 1.52e-02, grad_scale: 8.0
+2023-03-01 00:35:05,346 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 00:35:14,524 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.288, simple_loss=0.3799, pruned_loss=0.09807, over 944034.00 frames. 
+2023-03-01 00:35:14,525 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19478MB
+2023-03-01 00:35:35,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:35:39,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:35:55,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:35:59,652 INFO [train.py:968] (0/2) Epoch 2, batch 6050, giga_loss[loss=0.3794, simple_loss=0.4266, pruned_loss=0.1661, over 28721.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4044, pruned_loss=0.1503, over 5709458.82 frames. ], libri_tot_loss[loss=0.3729, simple_loss=0.4241, pruned_loss=0.1609, over 5506443.89 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4005, pruned_loss=0.1478, over 5700519.83 frames. ], batch size: 284, lr: 1.52e-02, grad_scale: 8.0
+2023-03-01 00:35:59,953 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51719.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:36:03,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51722.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:36:05,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.531e+02 1.331e+03 1.634e+03 2.068e+03 3.807e+03, threshold=3.269e+03, percent-clipped=2.0
+2023-03-01 00:36:06,713 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:36:29,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51751.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:36:47,722 INFO [train.py:968] (0/2) Epoch 2, batch 6100, giga_loss[loss=0.5259, simple_loss=0.5031, pruned_loss=0.2743, over 26821.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.4134, pruned_loss=0.1597, over 5696678.84 frames. ], libri_tot_loss[loss=0.373, simple_loss=0.4242, pruned_loss=0.1609, over 5505462.30 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.4098, pruned_loss=0.1575, over 5695512.40 frames. ], batch size: 555, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:36:52,690 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:37:03,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:37:34,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1212, 1.1741, 1.1509, 0.6699], device='cuda:0'), covar=tensor([0.0427, 0.0374, 0.0296, 0.0409], device='cuda:0'), in_proj_covar=tensor([0.0978, 0.0690, 0.0783, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 00:37:36,940 INFO [train.py:968] (0/2) Epoch 2, batch 6150, libri_loss[loss=0.3167, simple_loss=0.3779, pruned_loss=0.1278, over 29573.00 frames. ], tot_loss[loss=0.3761, simple_loss=0.4205, pruned_loss=0.1659, over 5697273.51 frames. ], libri_tot_loss[loss=0.3726, simple_loss=0.4239, pruned_loss=0.1607, over 5513956.91 frames. ], giga_tot_loss[loss=0.3734, simple_loss=0.4177, pruned_loss=0.1645, over 5692927.26 frames. ], batch size: 75, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:37:41,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.753e+03 2.232e+03 2.818e+03 8.705e+03, threshold=4.463e+03, percent-clipped=17.0
+2023-03-01 00:37:47,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:38:13,388 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:38:15,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:38:24,920 INFO [train.py:968] (0/2) Epoch 2, batch 6200, giga_loss[loss=0.4924, simple_loss=0.4672, pruned_loss=0.2588, over 23411.00 frames. ], tot_loss[loss=0.3848, simple_loss=0.4267, pruned_loss=0.1714, over 5691167.15 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4235, pruned_loss=0.1602, over 5521475.74 frames. ], giga_tot_loss[loss=0.3833, simple_loss=0.425, pruned_loss=0.1708, over 5684304.48 frames. ], batch size: 705, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:38:32,007 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-01 00:38:43,084 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:38:45,166 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:39:10,835 INFO [train.py:968] (0/2) Epoch 2, batch 6250, giga_loss[loss=0.4226, simple_loss=0.4476, pruned_loss=0.1988, over 28554.00 frames. ], tot_loss[loss=0.3897, simple_loss=0.4302, pruned_loss=0.1746, over 5696401.19 frames. ], libri_tot_loss[loss=0.3716, simple_loss=0.4232, pruned_loss=0.16, over 5531472.87 frames. ], giga_tot_loss[loss=0.3895, simple_loss=0.4292, pruned_loss=0.175, over 5690094.29 frames. ], batch size: 336, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:39:14,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.616e+03 2.128e+03 2.502e+03 4.620e+03, threshold=4.256e+03, percent-clipped=1.0
+2023-03-01 00:39:30,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51940.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:39:57,885 INFO [train.py:968] (0/2) Epoch 2, batch 6300, giga_loss[loss=0.4818, simple_loss=0.4965, pruned_loss=0.2335, over 28247.00 frames. ], tot_loss[loss=0.3992, simple_loss=0.4368, pruned_loss=0.1807, over 5692923.41 frames. ], libri_tot_loss[loss=0.3717, simple_loss=0.4234, pruned_loss=0.16, over 5539892.94 frames. ], giga_tot_loss[loss=0.3995, simple_loss=0.4361, pruned_loss=0.1814, over 5684109.55 frames. ], batch size: 368, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:40:01,125 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:40:04,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:40:27,851 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-52000.pt
+2023-03-01 00:40:32,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:40:48,378 INFO [train.py:968] (0/2) Epoch 2, batch 6350, giga_loss[loss=0.3747, simple_loss=0.4186, pruned_loss=0.1654, over 28812.00 frames. ], tot_loss[loss=0.4037, simple_loss=0.4398, pruned_loss=0.1838, over 5685176.22 frames. ], libri_tot_loss[loss=0.3715, simple_loss=0.4231, pruned_loss=0.1599, over 5548151.43 frames. ], giga_tot_loss[loss=0.4049, simple_loss=0.4399, pruned_loss=0.185, over 5674728.13 frames. ], batch size: 99, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:40:53,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.802e+03 2.384e+03 3.216e+03 7.094e+03, threshold=4.767e+03, percent-clipped=10.0
+2023-03-01 00:41:39,555 INFO [train.py:968] (0/2) Epoch 2, batch 6400, giga_loss[loss=0.4047, simple_loss=0.4358, pruned_loss=0.1868, over 28896.00 frames. ], tot_loss[loss=0.4081, simple_loss=0.4422, pruned_loss=0.187, over 5666686.89 frames. ], libri_tot_loss[loss=0.3718, simple_loss=0.4233, pruned_loss=0.1601, over 5545468.21 frames. ], giga_tot_loss[loss=0.4095, simple_loss=0.4425, pruned_loss=0.1883, over 5663595.03 frames. ], batch size: 112, lr: 1.52e-02, grad_scale: 8.0
+2023-03-01 00:41:53,827 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52083.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:41:57,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52086.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:42:28,473 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52115.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:42:31,880 INFO [train.py:968] (0/2) Epoch 2, batch 6450, giga_loss[loss=0.4092, simple_loss=0.4378, pruned_loss=0.1903, over 28607.00 frames. ], tot_loss[loss=0.4133, simple_loss=0.4452, pruned_loss=0.1907, over 5668692.78 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.4227, pruned_loss=0.1596, over 5555649.30 frames. ], giga_tot_loss[loss=0.4163, simple_loss=0.4466, pruned_loss=0.1931, over 5660124.75 frames. ], batch size: 92, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:42:37,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.273e+02 1.759e+03 2.300e+03 3.383e+03 1.321e+04, threshold=4.600e+03, percent-clipped=9.0
+2023-03-01 00:42:53,313 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=52140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:43:03,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:43:14,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:43:26,181 INFO [train.py:968] (0/2) Epoch 2, batch 6500, giga_loss[loss=0.5274, simple_loss=0.4952, pruned_loss=0.2798, over 23440.00 frames. ], tot_loss[loss=0.4189, simple_loss=0.4481, pruned_loss=0.1948, over 5656150.47 frames. ], libri_tot_loss[loss=0.3698, simple_loss=0.4218, pruned_loss=0.1589, over 5563153.15 frames. ], giga_tot_loss[loss=0.4233, simple_loss=0.4505, pruned_loss=0.198, over 5645110.28 frames. ], batch size: 705, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:43:45,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5303, 1.5025, 1.3668, 1.4214], device='cuda:0'), covar=tensor([0.0756, 0.1145, 0.1212, 0.0984], device='cuda:0'), in_proj_covar=tensor([0.0491, 0.0841, 0.0648, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 00:44:17,370 INFO [train.py:968] (0/2) Epoch 2, batch 6550, giga_loss[loss=0.4547, simple_loss=0.4663, pruned_loss=0.2215, over 28010.00 frames. ], tot_loss[loss=0.4215, simple_loss=0.4501, pruned_loss=0.1965, over 5649271.10 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.4216, pruned_loss=0.1589, over 5568050.75 frames. ], giga_tot_loss[loss=0.426, simple_loss=0.4526, pruned_loss=0.1997, over 5637794.00 frames. ], batch size: 412, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:44:23,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.743e+03 2.113e+03 2.748e+03 5.795e+03, threshold=4.225e+03, percent-clipped=3.0
+2023-03-01 00:44:47,109 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-01 00:45:00,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:45:06,597 INFO [train.py:968] (0/2) Epoch 2, batch 6600, giga_loss[loss=0.3969, simple_loss=0.4331, pruned_loss=0.1803, over 28835.00 frames. ], tot_loss[loss=0.422, simple_loss=0.4498, pruned_loss=0.1971, over 5646118.49 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4214, pruned_loss=0.1587, over 5568774.82 frames. ], giga_tot_loss[loss=0.4271, simple_loss=0.4527, pruned_loss=0.2008, over 5638734.44 frames. ], batch size: 186, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:45:09,697 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 00:45:29,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:45:34,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:45:39,261 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:45:41,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:45:55,460 INFO [train.py:968] (0/2) Epoch 2, batch 6650, giga_loss[loss=0.4635, simple_loss=0.4798, pruned_loss=0.2236, over 28670.00 frames. ], tot_loss[loss=0.4206, simple_loss=0.4483, pruned_loss=0.1964, over 5637883.49 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4215, pruned_loss=0.1588, over 5567093.99 frames. ], giga_tot_loss[loss=0.4257, simple_loss=0.4512, pruned_loss=0.2001, over 5634410.64 frames. ], batch size: 307, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:46:02,873 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:46:03,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.902e+03 2.372e+03 2.990e+03 5.113e+03, threshold=4.744e+03, percent-clipped=6.0
+2023-03-01 00:46:13,486 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:46:47,416 INFO [train.py:968] (0/2) Epoch 2, batch 6700, giga_loss[loss=0.4708, simple_loss=0.4796, pruned_loss=0.231, over 28253.00 frames. ], tot_loss[loss=0.4197, simple_loss=0.4486, pruned_loss=0.1954, over 5638174.63 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4215, pruned_loss=0.1588, over 5572342.76 frames. ], giga_tot_loss[loss=0.4245, simple_loss=0.4513, pruned_loss=0.1989, over 5631544.68 frames. ], batch size: 368, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:47:20,076 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:47:24,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:47:33,157 INFO [train.py:968] (0/2) Epoch 2, batch 6750, giga_loss[loss=0.5773, simple_loss=0.5363, pruned_loss=0.3092, over 26527.00 frames. ], tot_loss[loss=0.4159, simple_loss=0.4469, pruned_loss=0.1925, over 5652975.88 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4207, pruned_loss=0.1582, over 5586363.37 frames. ], giga_tot_loss[loss=0.4229, simple_loss=0.451, pruned_loss=0.1974, over 5638182.38 frames. ], batch size: 555, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:47:40,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.702e+03 2.276e+03 2.900e+03 5.252e+03, threshold=4.552e+03, percent-clipped=1.0
+2023-03-01 00:47:47,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=52433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:47:54,589 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:48:24,682 INFO [train.py:968] (0/2) Epoch 2, batch 6800, giga_loss[loss=0.4851, simple_loss=0.4703, pruned_loss=0.25, over 24015.00 frames. ], tot_loss[loss=0.4176, simple_loss=0.4482, pruned_loss=0.1935, over 5628255.30 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4205, pruned_loss=0.1581, over 5582901.09 frames. ], giga_tot_loss[loss=0.4242, simple_loss=0.4521, pruned_loss=0.1981, over 5621197.10 frames. ], batch size: 705, lr: 1.51e-02, grad_scale: 8.0
+2023-03-01 00:49:10,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:49:14,583 INFO [train.py:968] (0/2) Epoch 2, batch 6850, giga_loss[loss=0.3628, simple_loss=0.4189, pruned_loss=0.1533, over 28924.00 frames. ], tot_loss[loss=0.4134, simple_loss=0.4456, pruned_loss=0.1906, over 5634492.46 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.421, pruned_loss=0.1586, over 5589258.19 frames. ], giga_tot_loss[loss=0.4192, simple_loss=0.4489, pruned_loss=0.1947, over 5624025.64 frames. ], batch size: 145, lr: 1.51e-02, grad_scale: 8.0
+2023-03-01 00:49:26,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.743e+02 1.599e+03 2.160e+03 2.964e+03 1.009e+04, threshold=4.320e+03, percent-clipped=8.0
+2023-03-01 00:50:08,607 INFO [train.py:968] (0/2) Epoch 2, batch 6900, giga_loss[loss=0.3754, simple_loss=0.4213, pruned_loss=0.1647, over 28545.00 frames. ], tot_loss[loss=0.4101, simple_loss=0.4441, pruned_loss=0.1881, over 5643939.83 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4209, pruned_loss=0.1584, over 5595168.49 frames. ], giga_tot_loss[loss=0.4157, simple_loss=0.4473, pruned_loss=0.1921, over 5631157.77 frames. ], batch size: 336, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:50:27,153 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=52586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:51:01,311 INFO [train.py:968] (0/2) Epoch 2, batch 6950, giga_loss[loss=0.3777, simple_loss=0.4197, pruned_loss=0.1678, over 28941.00 frames. ], tot_loss[loss=0.4069, simple_loss=0.4418, pruned_loss=0.186, over 5642039.66 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.4211, pruned_loss=0.1586, over 5591212.25 frames. ], giga_tot_loss[loss=0.4117, simple_loss=0.4445, pruned_loss=0.1895, over 5635442.95 frames. ], batch size: 227, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:51:08,212 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.095e+02 1.730e+03 2.262e+03 3.129e+03 7.640e+03, threshold=4.525e+03, percent-clipped=11.0
+2023-03-01 00:51:39,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:51:43,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:51:51,938 INFO [train.py:968] (0/2) Epoch 2, batch 7000, libri_loss[loss=0.3295, simple_loss=0.3894, pruned_loss=0.1348, over 29521.00 frames. ], tot_loss[loss=0.4017, simple_loss=0.4382, pruned_loss=0.1825, over 5650780.95 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4208, pruned_loss=0.1582, over 5595853.75 frames. ], giga_tot_loss[loss=0.4064, simple_loss=0.4409, pruned_loss=0.186, over 5642126.44 frames. ], batch size: 80, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:52:11,263 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:52:27,344 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=52705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:52:40,744 INFO [train.py:968] (0/2) Epoch 2, batch 7050, giga_loss[loss=0.3974, simple_loss=0.4405, pruned_loss=0.1772, over 28992.00 frames. ], tot_loss[loss=0.3969, simple_loss=0.4344, pruned_loss=0.1797, over 5647875.38 frames. ], libri_tot_loss[loss=0.3681, simple_loss=0.4202, pruned_loss=0.1579, over 5602232.24 frames. ], giga_tot_loss[loss=0.4018, simple_loss=0.4373, pruned_loss=0.1831, over 5636182.18 frames. ], batch size: 186, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:52:48,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.871e+02 1.548e+03 1.931e+03 2.712e+03 5.561e+03, threshold=3.862e+03, percent-clipped=4.0
+2023-03-01 00:53:22,564 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-01 00:53:30,131 INFO [train.py:968] (0/2) Epoch 2, batch 7100, giga_loss[loss=0.3737, simple_loss=0.4236, pruned_loss=0.1619, over 28618.00 frames. ], tot_loss[loss=0.3966, simple_loss=0.4343, pruned_loss=0.1794, over 5642860.25 frames. ], libri_tot_loss[loss=0.3682, simple_loss=0.4204, pruned_loss=0.158, over 5597958.02 frames. ], giga_tot_loss[loss=0.4008, simple_loss=0.4367, pruned_loss=0.1825, over 5638724.64 frames. ], batch size: 262, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:53:38,218 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 00:54:09,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:54:22,982 INFO [train.py:968] (0/2) Epoch 2, batch 7150, giga_loss[loss=0.3439, simple_loss=0.4031, pruned_loss=0.1424, over 28933.00 frames. ], tot_loss[loss=0.3936, simple_loss=0.4325, pruned_loss=0.1773, over 5651642.50 frames. ], libri_tot_loss[loss=0.3685, simple_loss=0.4205, pruned_loss=0.1582, over 5610017.97 frames. ], giga_tot_loss[loss=0.398, simple_loss=0.435, pruned_loss=0.1805, over 5639518.32 frames. ], batch size: 164, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:54:31,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.674e+02 1.558e+03 1.995e+03 2.665e+03 6.377e+03, threshold=3.991e+03, percent-clipped=6.0
+2023-03-01 00:55:15,677 INFO [train.py:968] (0/2) Epoch 2, batch 7200, giga_loss[loss=0.4565, simple_loss=0.4973, pruned_loss=0.2079, over 28649.00 frames. ], tot_loss[loss=0.3882, simple_loss=0.4296, pruned_loss=0.1734, over 5658983.73 frames. ], libri_tot_loss[loss=0.3679, simple_loss=0.4198, pruned_loss=0.158, over 5614232.00 frames. ], giga_tot_loss[loss=0.3925, simple_loss=0.4323, pruned_loss=0.1763, over 5646352.25 frames. ], batch size: 262, lr: 1.50e-02, grad_scale: 8.0
+2023-03-01 00:55:47,482 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8499, 1.5403, 1.5402, 1.5186], device='cuda:0'), covar=tensor([0.0703, 0.1443, 0.0972, 0.1097], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0802, 0.0625, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 00:56:12,211 INFO [train.py:968] (0/2) Epoch 2, batch 7250, libri_loss[loss=0.3412, simple_loss=0.3849, pruned_loss=0.1488, over 29351.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4292, pruned_loss=0.1704, over 5666093.23 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4187, pruned_loss=0.1572, over 5621886.20 frames. ], giga_tot_loss[loss=0.3901, simple_loss=0.4326, pruned_loss=0.1738, over 5650582.84 frames. ], batch size: 71, lr: 1.50e-02, grad_scale: 8.0
+2023-03-01 00:56:18,050 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.475e+02 1.526e+03 1.967e+03 2.511e+03 4.253e+03, threshold=3.934e+03, percent-clipped=1.0
+2023-03-01 00:56:35,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8615, 2.4706, 1.8885, 1.7511], device='cuda:0'), covar=tensor([0.1051, 0.0345, 0.0463, 0.1235], device='cuda:0'), in_proj_covar=tensor([0.0331, 0.0228, 0.0230, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0021, 0.0018, 0.0030], device='cuda:0')
+2023-03-01 00:56:37,926 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=52948.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:56:40,491 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:56:43,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:56:51,919 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:57:01,899 INFO [train.py:968] (0/2) Epoch 2, batch 7300, giga_loss[loss=0.3852, simple_loss=0.4265, pruned_loss=0.1719, over 28631.00 frames. ], tot_loss[loss=0.3868, simple_loss=0.4312, pruned_loss=0.1712, over 5679982.02 frames. ], libri_tot_loss[loss=0.3663, simple_loss=0.4185, pruned_loss=0.1571, over 5630627.86 frames. ], giga_tot_loss[loss=0.3918, simple_loss=0.4345, pruned_loss=0.1746, over 5661259.05 frames. ], batch size: 307, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 00:57:18,067 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:57:22,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2843, 1.6223, 1.1591, 1.3385], device='cuda:0'), covar=tensor([0.1027, 0.0504, 0.0510, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0329, 0.0228, 0.0230, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0021, 0.0018, 0.0030], device='cuda:0')
+2023-03-01 00:57:37,078 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 00:57:53,244 INFO [train.py:968] (0/2) Epoch 2, batch 7350, libri_loss[loss=0.3688, simple_loss=0.4127, pruned_loss=0.1625, over 29585.00 frames. ], tot_loss[loss=0.3888, simple_loss=0.432, pruned_loss=0.1728, over 5671481.33 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4182, pruned_loss=0.1569, over 5637050.11 frames. ], giga_tot_loss[loss=0.3936, simple_loss=0.4351, pruned_loss=0.176, over 5651740.47 frames. ], batch size: 74, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 00:58:03,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.757e+03 2.282e+03 3.594e+03 8.306e+03, threshold=4.564e+03, percent-clipped=15.0
+2023-03-01 00:58:39,319 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8375, 3.8012, 4.5622, 1.8401], device='cuda:0'), covar=tensor([0.0387, 0.0405, 0.0680, 0.1746], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0520, 0.0827, 0.0539], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 00:58:41,695 INFO [train.py:968] (0/2) Epoch 2, batch 7400, giga_loss[loss=0.3238, simple_loss=0.3846, pruned_loss=0.1315, over 28753.00 frames. ], tot_loss[loss=0.3878, simple_loss=0.4307, pruned_loss=0.1724, over 5671648.48 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4179, pruned_loss=0.1567, over 5632811.01 frames. ], giga_tot_loss[loss=0.3923, simple_loss=0.4337, pruned_loss=0.1754, over 5660381.48 frames. ], batch size: 119, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 00:58:56,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:59:21,851 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:59:23,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:59:24,029 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-01 00:59:34,541 INFO [train.py:968] (0/2) Epoch 2, batch 7450, giga_loss[loss=0.3892, simple_loss=0.4253, pruned_loss=0.1765, over 28754.00 frames. ], tot_loss[loss=0.3912, simple_loss=0.4316, pruned_loss=0.1754, over 5674283.41 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4183, pruned_loss=0.1568, over 5641478.63 frames. ], giga_tot_loss[loss=0.3951, simple_loss=0.434, pruned_loss=0.1781, over 5658353.43 frames. ], batch size: 119, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 00:59:43,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.246e+02 1.685e+03 2.142e+03 3.027e+03 1.065e+04, threshold=4.283e+03, percent-clipped=7.0
+2023-03-01 00:59:53,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:00:00,243 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:00:21,287 INFO [train.py:968] (0/2) Epoch 2, batch 7500, giga_loss[loss=0.3353, simple_loss=0.399, pruned_loss=0.1358, over 28918.00 frames. ], tot_loss[loss=0.3903, simple_loss=0.4306, pruned_loss=0.175, over 5680565.09 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4181, pruned_loss=0.1565, over 5645383.45 frames. ], giga_tot_loss[loss=0.3945, simple_loss=0.433, pruned_loss=0.178, over 5665198.01 frames. ], batch size: 174, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:01:18,989 INFO [train.py:968] (0/2) Epoch 2, batch 7550, giga_loss[loss=0.387, simple_loss=0.4068, pruned_loss=0.1836, over 23819.00 frames. ], tot_loss[loss=0.3886, simple_loss=0.4301, pruned_loss=0.1735, over 5666273.08 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4183, pruned_loss=0.1565, over 5649071.48 frames. ], giga_tot_loss[loss=0.3922, simple_loss=0.4321, pruned_loss=0.1762, over 5651164.33 frames. ], batch size: 705, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:01:22,241 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:01:24,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:01:26,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.931e+02 1.421e+03 1.789e+03 2.387e+03 4.590e+03, threshold=3.578e+03, percent-clipped=2.0
+2023-03-01 01:01:41,431 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0479, 1.8243, 1.6240, 1.6308], device='cuda:0'), covar=tensor([0.0743, 0.1442, 0.1227, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0833, 0.0638, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 01:01:55,033 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:02:06,858 INFO [train.py:968] (0/2) Epoch 2, batch 7600, giga_loss[loss=0.3675, simple_loss=0.4109, pruned_loss=0.162, over 28060.00 frames. ], tot_loss[loss=0.3876, simple_loss=0.4302, pruned_loss=0.1725, over 5677094.64 frames. ], libri_tot_loss[loss=0.3662, simple_loss=0.4186, pruned_loss=0.1569, over 5653980.24 frames. ], giga_tot_loss[loss=0.3905, simple_loss=0.4318, pruned_loss=0.1746, over 5661043.98 frames. ], batch size: 412, lr: 1.50e-02, grad_scale: 8.0
+2023-03-01 01:02:53,286 INFO [train.py:968] (0/2) Epoch 2, batch 7650, giga_loss[loss=0.3943, simple_loss=0.4352, pruned_loss=0.1767, over 29035.00 frames. ], tot_loss[loss=0.3873, simple_loss=0.43, pruned_loss=0.1723, over 5687867.00 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4185, pruned_loss=0.1567, over 5663276.37 frames. ], giga_tot_loss[loss=0.3907, simple_loss=0.4318, pruned_loss=0.1747, over 5667209.33 frames. ], batch size: 155, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:02:58,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53323.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 01:03:02,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.331e+02 1.615e+03 2.190e+03 3.048e+03 1.090e+04, threshold=4.380e+03, percent-clipped=17.0
+2023-03-01 01:03:45,274 INFO [train.py:968] (0/2) Epoch 2, batch 7700, giga_loss[loss=0.3428, simple_loss=0.4069, pruned_loss=0.1394, over 29005.00 frames. ], tot_loss[loss=0.3867, simple_loss=0.4291, pruned_loss=0.1721, over 5688331.51 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4185, pruned_loss=0.1567, over 5664529.99 frames. ], giga_tot_loss[loss=0.3894, simple_loss=0.4306, pruned_loss=0.1741, over 5671152.62 frames. ], batch size: 164, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:04:38,082 INFO [train.py:968] (0/2) Epoch 2, batch 7750, giga_loss[loss=0.3913, simple_loss=0.4285, pruned_loss=0.177, over 28841.00 frames. ], tot_loss[loss=0.3846, simple_loss=0.4268, pruned_loss=0.1712, over 5676242.15 frames. ], libri_tot_loss[loss=0.3662, simple_loss=0.4189, pruned_loss=0.1568, over 5667539.00 frames. ], giga_tot_loss[loss=0.3869, simple_loss=0.4278, pruned_loss=0.173, over 5660002.45 frames. ], batch size: 112, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:04:47,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.524e+03 1.966e+03 2.668e+03 8.369e+03, threshold=3.933e+03, percent-clipped=8.0
+2023-03-01 01:05:05,587 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:05:24,310 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53466.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 01:05:25,873 INFO [train.py:968] (0/2) Epoch 2, batch 7800, giga_loss[loss=0.3738, simple_loss=0.4156, pruned_loss=0.166, over 28857.00 frames. ], tot_loss[loss=0.3828, simple_loss=0.4248, pruned_loss=0.1704, over 5672899.42 frames. ], libri_tot_loss[loss=0.3649, simple_loss=0.4179, pruned_loss=0.156, over 5674770.99 frames. ], giga_tot_loss[loss=0.3864, simple_loss=0.4266, pruned_loss=0.173, over 5653407.74 frames. ], batch size: 186, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:05:26,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53469.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:05:33,188 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 01:05:54,304 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53498.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:06:14,864 INFO [train.py:968] (0/2) Epoch 2, batch 7850, giga_loss[loss=0.4072, simple_loss=0.4137, pruned_loss=0.2003, over 23615.00 frames. ], tot_loss[loss=0.3838, simple_loss=0.4249, pruned_loss=0.1713, over 5662268.72 frames. ], libri_tot_loss[loss=0.3647, simple_loss=0.4178, pruned_loss=0.1558, over 5669497.03 frames. ], giga_tot_loss[loss=0.3873, simple_loss=0.4267, pruned_loss=0.174, over 5650786.15 frames. ], batch size: 705, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:06:15,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:06:25,905 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.338e+02 1.769e+03 2.108e+03 2.865e+03 9.967e+03, threshold=4.216e+03, percent-clipped=12.0
+2023-03-01 01:06:42,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5540, 1.4411, 1.3205, 1.3555], device='cuda:0'), covar=tensor([0.0619, 0.0939, 0.1074, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0823, 0.0631, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 01:07:00,255 INFO [train.py:968] (0/2) Epoch 2, batch 7900, giga_loss[loss=0.3556, simple_loss=0.4028, pruned_loss=0.1542, over 29005.00 frames. ], tot_loss[loss=0.3812, simple_loss=0.4228, pruned_loss=0.1698, over 5658301.23 frames. ], libri_tot_loss[loss=0.3641, simple_loss=0.4174, pruned_loss=0.1554, over 5669106.17 frames. ], giga_tot_loss[loss=0.3853, simple_loss=0.4248, pruned_loss=0.1729, over 5648919.76 frames. ], batch size: 106, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:07:48,895 INFO [train.py:968] (0/2) Epoch 2, batch 7950, giga_loss[loss=0.3525, simple_loss=0.4031, pruned_loss=0.151, over 29038.00 frames. ], tot_loss[loss=0.3829, simple_loss=0.4236, pruned_loss=0.1711, over 5661219.36 frames. ], libri_tot_loss[loss=0.3647, simple_loss=0.4178, pruned_loss=0.1558, over 5671405.61 frames. ], giga_tot_loss[loss=0.3858, simple_loss=0.4249, pruned_loss=0.1734, over 5651506.80 frames. ], batch size: 136, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:08:00,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.437e+02 1.589e+03 2.030e+03 2.858e+03 8.257e+03, threshold=4.061e+03, percent-clipped=7.0
+2023-03-01 01:08:04,676 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3504, 2.0791, 1.5608, 0.4169], device='cuda:0'), covar=tensor([0.1376, 0.0803, 0.1130, 0.1855], device='cuda:0'), in_proj_covar=tensor([0.1143, 0.1139, 0.1143, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 01:08:05,299 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:08:09,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2139, 1.2700, 1.1756, 1.6044], device='cuda:0'), covar=tensor([0.1917, 0.1979, 0.1588, 0.1793], device='cuda:0'), in_proj_covar=tensor([0.0947, 0.0813, 0.0879, 0.0926], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 01:08:24,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1133, 3.3448, 3.8352, 1.8408], device='cuda:0'), covar=tensor([0.0436, 0.0482, 0.0749, 0.1643], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0529, 0.0821, 0.0544], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 01:08:37,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:08:40,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:08:43,803 INFO [train.py:968] (0/2) Epoch 2, batch 8000, giga_loss[loss=0.4398, simple_loss=0.4348, pruned_loss=0.2224, over 23616.00 frames. ], tot_loss[loss=0.3861, simple_loss=0.4263, pruned_loss=0.173, over 5658167.65 frames. ], libri_tot_loss[loss=0.3647, simple_loss=0.4178, pruned_loss=0.1558, over 5671405.61 frames. ], giga_tot_loss[loss=0.3884, simple_loss=0.4273, pruned_loss=0.1747, over 5650608.24 frames. ], batch size: 705, lr: 1.49e-02, grad_scale: 8.0
+2023-03-01 01:09:04,956 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:09:06,140 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:09:17,647 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:09:27,622 INFO [train.py:968] (0/2) Epoch 2, batch 8050, giga_loss[loss=0.3754, simple_loss=0.4249, pruned_loss=0.163, over 28930.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4264, pruned_loss=0.1718, over 5671285.94 frames. ], libri_tot_loss[loss=0.3644, simple_loss=0.4176, pruned_loss=0.1556, over 5677833.26 frames. ], giga_tot_loss[loss=0.3878, simple_loss=0.4277, pruned_loss=0.1739, over 5659122.32 frames. ], batch size: 213, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:09:27,874 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2249, 1.6850, 1.5605, 1.6617], device='cuda:0'), covar=tensor([0.0736, 0.1480, 0.1146, 0.1185], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0828, 0.0633, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 01:09:37,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.565e+02 1.696e+03 2.313e+03 3.010e+03 5.218e+03, threshold=4.625e+03, percent-clipped=9.0
+2023-03-01 01:10:03,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:10:11,728 INFO [train.py:968] (0/2) Epoch 2, batch 8100, giga_loss[loss=0.3347, simple_loss=0.392, pruned_loss=0.1387, over 28727.00 frames. ], tot_loss[loss=0.3822, simple_loss=0.4251, pruned_loss=0.1697, over 5688799.19 frames. ], libri_tot_loss[loss=0.3633, simple_loss=0.4166, pruned_loss=0.155, over 5685682.08 frames. ], giga_tot_loss[loss=0.3863, simple_loss=0.4274, pruned_loss=0.1726, over 5671590.91 frames. ], batch size: 99, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:11:00,439 INFO [train.py:968] (0/2) Epoch 2, batch 8150, giga_loss[loss=0.3765, simple_loss=0.4175, pruned_loss=0.1678, over 28295.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.4256, pruned_loss=0.1704, over 5687514.95 frames. ], libri_tot_loss[loss=0.3628, simple_loss=0.4163, pruned_loss=0.1547, over 5691021.07 frames. ], giga_tot_loss[loss=0.3873, simple_loss=0.4279, pruned_loss=0.1733, over 5669193.34 frames. ], batch size: 368, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:11:09,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:11:17,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.200e+02 1.819e+03 2.134e+03 2.999e+03 9.346e+03, threshold=4.268e+03, percent-clipped=6.0
+2023-03-01 01:11:51,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5619, 2.1641, 1.5457, 0.6192], device='cuda:0'), covar=tensor([0.1588, 0.1058, 0.0884, 0.1672], device='cuda:0'), in_proj_covar=tensor([0.1145, 0.1129, 0.1127, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 01:11:53,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4323, 1.6394, 1.4074, 0.8961], device='cuda:0'), covar=tensor([0.0722, 0.0425, 0.0360, 0.0565], device='cuda:0'), in_proj_covar=tensor([0.0973, 0.0698, 0.0781, 0.0791], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 01:11:58,817 INFO [train.py:968] (0/2) Epoch 2, batch 8200, giga_loss[loss=0.4221, simple_loss=0.4515, pruned_loss=0.1964, over 28941.00 frames. ], tot_loss[loss=0.3884, simple_loss=0.4289, pruned_loss=0.174, over 5682149.31 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4163, pruned_loss=0.1546, over 5693344.49 frames. ], giga_tot_loss[loss=0.3921, simple_loss=0.4308, pruned_loss=0.1767, over 5665417.27 frames. ], batch size: 285, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:12:02,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2697, 1.2621, 1.1650, 1.2090], device='cuda:0'), covar=tensor([0.0695, 0.0936, 0.1243, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0832, 0.0641, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 01:12:56,911 INFO [train.py:968] (0/2) Epoch 2, batch 8250, giga_loss[loss=0.3768, simple_loss=0.4195, pruned_loss=0.1671, over 28940.00 frames. ], tot_loss[loss=0.3927, simple_loss=0.4305, pruned_loss=0.1775, over 5667624.39 frames. ], libri_tot_loss[loss=0.3624, simple_loss=0.4161, pruned_loss=0.1543, over 5696822.39 frames. ], giga_tot_loss[loss=0.3963, simple_loss=0.4324, pruned_loss=0.1801, over 5650821.39 frames. ], batch size: 213, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:13:08,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.726e+03 2.316e+03 3.169e+03 7.412e+03, threshold=4.632e+03, percent-clipped=12.0
+2023-03-01 01:13:49,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:13:50,191 INFO [train.py:968] (0/2) Epoch 2, batch 8300, giga_loss[loss=0.3678, simple_loss=0.4073, pruned_loss=0.1642, over 28878.00 frames. ], tot_loss[loss=0.3941, simple_loss=0.4309, pruned_loss=0.1787, over 5674763.12 frames. ], libri_tot_loss[loss=0.3625, simple_loss=0.4162, pruned_loss=0.1544, over 5700664.89 frames. ], giga_tot_loss[loss=0.3978, simple_loss=0.4328, pruned_loss=0.1814, over 5657090.78 frames. ], batch size: 199, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:13:52,133 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:14:21,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:14:22,097 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-54000.pt
+2023-03-01 01:14:32,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:14:43,627 INFO [train.py:968] (0/2) Epoch 2, batch 8350, giga_loss[loss=0.369, simple_loss=0.411, pruned_loss=0.1634, over 28581.00 frames. ], tot_loss[loss=0.3978, simple_loss=0.4328, pruned_loss=0.1814, over 5668056.61 frames. ], libri_tot_loss[loss=0.3622, simple_loss=0.416, pruned_loss=0.1542, over 5703423.62 frames. ], giga_tot_loss[loss=0.4015, simple_loss=0.4348, pruned_loss=0.1841, over 5651325.06 frames. ], batch size: 92, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:14:53,991 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.723e+02 1.667e+03 2.309e+03 3.370e+03 6.072e+03, threshold=4.618e+03, percent-clipped=6.0
+2023-03-01 01:14:54,429 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 01:15:31,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-01 01:15:32,134 INFO [train.py:968] (0/2) Epoch 2, batch 8400, giga_loss[loss=0.3593, simple_loss=0.4142, pruned_loss=0.1523, over 28927.00 frames. ], tot_loss[loss=0.3934, simple_loss=0.4298, pruned_loss=0.1785, over 5675398.20 frames. ], libri_tot_loss[loss=0.3616, simple_loss=0.4154, pruned_loss=0.1539, over 5704722.23 frames. ], giga_tot_loss[loss=0.3973, simple_loss=0.4321, pruned_loss=0.1813, over 5660578.31 frames. ], batch size: 213, lr: 1.49e-02, grad_scale: 8.0
+2023-03-01 01:15:36,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:15:48,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:15:57,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9275, 1.6860, 1.6071, 1.4805], device='cuda:0'), covar=tensor([0.0812, 0.1489, 0.1065, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0842, 0.0641, 0.0710], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 01:16:04,933 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-01 01:16:20,683 INFO [train.py:968] (0/2) Epoch 2, batch 8450, giga_loss[loss=0.3659, simple_loss=0.4271, pruned_loss=0.1524, over 28852.00 frames. ], tot_loss[loss=0.3896, simple_loss=0.4285, pruned_loss=0.1753, over 5683344.10 frames. ], libri_tot_loss[loss=0.3618, simple_loss=0.4158, pruned_loss=0.154, over 5708544.41 frames. ], giga_tot_loss[loss=0.393, simple_loss=0.4302, pruned_loss=0.1779, over 5667544.32 frames. ], batch size: 145, lr: 1.49e-02, grad_scale: 8.0
+2023-03-01 01:16:30,475 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.818e+02 1.616e+03 2.049e+03 2.838e+03 5.996e+03, threshold=4.098e+03, percent-clipped=2.0
+2023-03-01 01:16:37,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:16:52,448 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:16:56,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:17:08,975 INFO [train.py:968] (0/2) Epoch 2, batch 8500, libri_loss[loss=0.3338, simple_loss=0.3959, pruned_loss=0.1359, over 29543.00 frames. ], tot_loss[loss=0.3848, simple_loss=0.4257, pruned_loss=0.172, over 5693093.77 frames. ], libri_tot_loss[loss=0.3623, simple_loss=0.4161, pruned_loss=0.1542, over 5714807.55 frames. ], giga_tot_loss[loss=0.388, simple_loss=0.4272, pruned_loss=0.1744, over 5673836.54 frames. ], batch size: 79, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:17:22,304 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:17:48,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:17:51,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:17:51,926 INFO [train.py:968] (0/2) Epoch 2, batch 8550, giga_loss[loss=0.4241, simple_loss=0.4493, pruned_loss=0.1995, over 29105.00 frames. ], tot_loss[loss=0.3818, simple_loss=0.423, pruned_loss=0.1703, over 5689742.99 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4163, pruned_loss=0.1545, over 5715782.84 frames. ], giga_tot_loss[loss=0.3844, simple_loss=0.4241, pruned_loss=0.1723, over 5673023.93 frames. ], batch size: 128, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:17:55,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 01:17:57,287 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6012, 1.7535, 1.7494, 1.6532], device='cuda:0'), covar=tensor([0.0583, 0.0871, 0.0913, 0.1139], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0537, 0.0558, 0.0504], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 01:18:00,307 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:18:04,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:18:05,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.103e+02 1.792e+03 2.366e+03 3.162e+03 6.450e+03, threshold=4.731e+03, percent-clipped=6.0
+2023-03-01 01:18:20,641 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:18:32,155 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:18:42,845 INFO [train.py:968] (0/2) Epoch 2, batch 8600, giga_loss[loss=0.3883, simple_loss=0.4282, pruned_loss=0.1742, over 28721.00 frames. ], tot_loss[loss=0.3798, simple_loss=0.4208, pruned_loss=0.1694, over 5683497.80 frames. ], libri_tot_loss[loss=0.3624, simple_loss=0.416, pruned_loss=0.1544, over 5715569.67 frames. ], giga_tot_loss[loss=0.3824, simple_loss=0.4221, pruned_loss=0.1714, over 5669974.47 frames. ], batch size: 262, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:18:51,745 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:18:54,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:19:26,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:19:35,302 INFO [train.py:968] (0/2) Epoch 2, batch 8650, giga_loss[loss=0.403, simple_loss=0.4433, pruned_loss=0.1814, over 28963.00 frames. ], tot_loss[loss=0.3822, simple_loss=0.4224, pruned_loss=0.171, over 5678338.61 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4164, pruned_loss=0.1545, over 5718635.40 frames. ], giga_tot_loss[loss=0.3842, simple_loss=0.4231, pruned_loss=0.1727, over 5664391.42 frames. ], batch size: 213, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:19:48,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.580e+03 2.024e+03 2.595e+03 6.862e+03, threshold=4.049e+03, percent-clipped=2.0
+2023-03-01 01:20:02,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8425, 2.4051, 1.9227, 1.8793], device='cuda:0'), covar=tensor([0.1461, 0.1380, 0.1104, 0.0694], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0836, 0.0727, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0011, 0.0009, 0.0006], device='cuda:0')
+2023-03-01 01:20:25,241 INFO [train.py:968] (0/2) Epoch 2, batch 8700, giga_loss[loss=0.4295, simple_loss=0.4752, pruned_loss=0.1919, over 28868.00 frames. ], tot_loss[loss=0.3871, simple_loss=0.4268, pruned_loss=0.1737, over 5683274.44 frames. ], libri_tot_loss[loss=0.3628, simple_loss=0.4164, pruned_loss=0.1546, over 5721431.42 frames. ], giga_tot_loss[loss=0.3894, simple_loss=0.4276, pruned_loss=0.1756, over 5668714.24 frames. ], batch size: 174, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:20:50,654 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-01 01:21:09,843 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-01 01:21:16,800 INFO [train.py:968] (0/2) Epoch 2, batch 8750, libri_loss[loss=0.3673, simple_loss=0.4202, pruned_loss=0.1572, over 29756.00 frames. ], tot_loss[loss=0.3882, simple_loss=0.4301, pruned_loss=0.1731, over 5685209.49 frames. ], libri_tot_loss[loss=0.362, simple_loss=0.4158, pruned_loss=0.154, over 5729319.26 frames. ], giga_tot_loss[loss=0.3917, simple_loss=0.4318, pruned_loss=0.1759, over 5664401.29 frames. ], batch size: 87, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:21:28,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.643e+02 1.578e+03 1.963e+03 2.736e+03 6.213e+03, threshold=3.927e+03, percent-clipped=9.0
+2023-03-01 01:22:09,613 INFO [train.py:968] (0/2) Epoch 2, batch 8800, libri_loss[loss=0.3646, simple_loss=0.4198, pruned_loss=0.1548, over 29553.00 frames. ], tot_loss[loss=0.3908, simple_loss=0.4331, pruned_loss=0.1743, over 5688556.37 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.4159, pruned_loss=0.1541, over 5732765.97 frames. ], giga_tot_loss[loss=0.3941, simple_loss=0.4347, pruned_loss=0.1768, over 5667999.58 frames. ], batch size: 84, lr: 1.48e-02, grad_scale: 8.0
+2023-03-01 01:22:13,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6308, 1.9551, 1.7868, 1.6602], device='cuda:0'), covar=tensor([0.1179, 0.1421, 0.0960, 0.0672], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0821, 0.0712, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 01:22:56,580 INFO [train.py:968] (0/2) Epoch 2, batch 8850, giga_loss[loss=0.4044, simple_loss=0.4498, pruned_loss=0.1795, over 29059.00 frames. ], tot_loss[loss=0.3948, simple_loss=0.4359, pruned_loss=0.1768, over 5688463.58 frames. ], libri_tot_loss[loss=0.3616, simple_loss=0.4156, pruned_loss=0.1538, over 5733960.55 frames. ], giga_tot_loss[loss=0.3986, simple_loss=0.4379, pruned_loss=0.1796, over 5669966.07 frames. ], batch size: 155, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:23:06,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.400e+02 1.677e+03 2.031e+03 2.581e+03 5.451e+03, threshold=4.061e+03, percent-clipped=4.0
+2023-03-01 01:23:12,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-01 01:23:42,197 INFO [train.py:968] (0/2) Epoch 2, batch 8900, giga_loss[loss=0.3854, simple_loss=0.4279, pruned_loss=0.1715, over 28598.00 frames. ], tot_loss[loss=0.3954, simple_loss=0.4364, pruned_loss=0.1772, over 5693352.15 frames. ], libri_tot_loss[loss=0.3624, simple_loss=0.4162, pruned_loss=0.1543, over 5732556.22 frames. ], giga_tot_loss[loss=0.3991, simple_loss=0.4382, pruned_loss=0.18, over 5677310.28 frames. ], batch size: 78, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:24:32,846 INFO [train.py:968] (0/2) Epoch 2, batch 8950, giga_loss[loss=0.4051, simple_loss=0.4343, pruned_loss=0.1879, over 28585.00 frames. ], tot_loss[loss=0.396, simple_loss=0.4359, pruned_loss=0.1781, over 5692819.86 frames. ], libri_tot_loss[loss=0.3617, simple_loss=0.4157, pruned_loss=0.1538, over 5734196.11 frames. ], giga_tot_loss[loss=0.4, simple_loss=0.4381, pruned_loss=0.181, over 5678215.81 frames. ], batch size: 242, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:24:46,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.140e+02 1.618e+03 2.111e+03 3.010e+03 1.070e+04, threshold=4.223e+03, percent-clipped=11.0
+2023-03-01 01:25:26,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5546, 2.3088, 1.7178, 0.6215], device='cuda:0'), covar=tensor([0.1505, 0.0910, 0.1177, 0.1721], device='cuda:0'), in_proj_covar=tensor([0.1131, 0.1122, 0.1129, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 01:25:27,107 INFO [train.py:968] (0/2) Epoch 2, batch 9000, libri_loss[loss=0.386, simple_loss=0.4404, pruned_loss=0.1658, over 29243.00 frames. ], tot_loss[loss=0.3953, simple_loss=0.4345, pruned_loss=0.178, over 5691346.17 frames. ], libri_tot_loss[loss=0.3617, simple_loss=0.4157, pruned_loss=0.1538, over 5734990.01 frames. ], giga_tot_loss[loss=0.3989, simple_loss=0.4365, pruned_loss=0.1806, over 5678375.44 frames. ], batch size: 94, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:25:27,112 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 01:25:36,185 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.2828, simple_loss=0.3785, pruned_loss=0.09356, over 944034.00 frames. 
+2023-03-01 01:25:36,186 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19478MB
+2023-03-01 01:26:22,806 INFO [train.py:968] (0/2) Epoch 2, batch 9050, giga_loss[loss=0.3728, simple_loss=0.4045, pruned_loss=0.1706, over 29016.00 frames. ], tot_loss[loss=0.3946, simple_loss=0.4329, pruned_loss=0.1782, over 5687165.25 frames. ], libri_tot_loss[loss=0.3618, simple_loss=0.4158, pruned_loss=0.1539, over 5738510.00 frames. ], giga_tot_loss[loss=0.3983, simple_loss=0.4349, pruned_loss=0.1808, over 5672614.09 frames. ], batch size: 106, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:26:33,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.757e+02 1.842e+03 2.388e+03 3.233e+03 8.496e+03, threshold=4.776e+03, percent-clipped=9.0
+2023-03-01 01:27:13,559 INFO [train.py:968] (0/2) Epoch 2, batch 9100, giga_loss[loss=0.4352, simple_loss=0.4578, pruned_loss=0.2063, over 28752.00 frames. ], tot_loss[loss=0.3948, simple_loss=0.4325, pruned_loss=0.1785, over 5674847.96 frames. ], libri_tot_loss[loss=0.3624, simple_loss=0.4162, pruned_loss=0.1543, over 5723922.55 frames. ], giga_tot_loss[loss=0.398, simple_loss=0.4342, pruned_loss=0.1809, over 5674398.19 frames. ], batch size: 284, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:28:02,107 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.91 vs. limit=5.0
+2023-03-01 01:28:04,569 INFO [train.py:968] (0/2) Epoch 2, batch 9150, giga_loss[loss=0.3917, simple_loss=0.4259, pruned_loss=0.1787, over 28616.00 frames. ], tot_loss[loss=0.3926, simple_loss=0.4309, pruned_loss=0.1772, over 5677223.56 frames. ], libri_tot_loss[loss=0.3624, simple_loss=0.4162, pruned_loss=0.1543, over 5728179.49 frames. ], giga_tot_loss[loss=0.3959, simple_loss=0.4325, pruned_loss=0.1797, over 5671788.35 frames. ], batch size: 85, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:28:08,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-01 01:28:18,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.796e+03 2.603e+03 3.392e+03 9.663e+03, threshold=5.206e+03, percent-clipped=9.0
+2023-03-01 01:28:34,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3267, 1.8304, 1.3632, 1.4181], device='cuda:0'), covar=tensor([0.1083, 0.0447, 0.0488, 0.1245], device='cuda:0'), in_proj_covar=tensor([0.0328, 0.0226, 0.0227, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0021, 0.0019, 0.0031], device='cuda:0')
+2023-03-01 01:28:57,793 INFO [train.py:968] (0/2) Epoch 2, batch 9200, giga_loss[loss=0.3789, simple_loss=0.4179, pruned_loss=0.17, over 28698.00 frames. ], tot_loss[loss=0.3938, simple_loss=0.4313, pruned_loss=0.1781, over 5666726.19 frames. ], libri_tot_loss[loss=0.3628, simple_loss=0.4166, pruned_loss=0.1545, over 5722626.82 frames. ], giga_tot_loss[loss=0.3966, simple_loss=0.4326, pruned_loss=0.1803, over 5666142.56 frames. ], batch size: 243, lr: 1.48e-02, grad_scale: 8.0
+2023-03-01 01:29:46,787 INFO [train.py:968] (0/2) Epoch 2, batch 9250, giga_loss[loss=0.3041, simple_loss=0.3667, pruned_loss=0.1207, over 28445.00 frames. ], tot_loss[loss=0.3896, simple_loss=0.428, pruned_loss=0.1756, over 5677869.44 frames. ], libri_tot_loss[loss=0.3618, simple_loss=0.4159, pruned_loss=0.1539, over 5726828.60 frames. ], giga_tot_loss[loss=0.3935, simple_loss=0.43, pruned_loss=0.1785, over 5672511.93 frames. ], batch size: 60, lr: 1.48e-02, grad_scale: 8.0
+2023-03-01 01:30:01,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.638e+03 2.049e+03 2.757e+03 5.110e+03, threshold=4.097e+03, percent-clipped=0.0
+2023-03-01 01:30:18,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2267, 2.5653, 2.9208, 1.4140], device='cuda:0'), covar=tensor([0.0769, 0.0609, 0.1188, 0.1799], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0537, 0.0848, 0.0537], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0006, 0.0010, 0.0007], device='cuda:0')
+2023-03-01 01:30:37,552 INFO [train.py:968] (0/2) Epoch 2, batch 9300, giga_loss[loss=0.3793, simple_loss=0.4319, pruned_loss=0.1633, over 28864.00 frames. ], tot_loss[loss=0.3893, simple_loss=0.4275, pruned_loss=0.1755, over 5681654.27 frames. ], libri_tot_loss[loss=0.3612, simple_loss=0.4153, pruned_loss=0.1535, over 5729423.23 frames. ], giga_tot_loss[loss=0.3934, simple_loss=0.4298, pruned_loss=0.1785, over 5674334.79 frames. ], batch size: 106, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:31:30,787 INFO [train.py:968] (0/2) Epoch 2, batch 9350, giga_loss[loss=0.411, simple_loss=0.4445, pruned_loss=0.1887, over 28935.00 frames. ], tot_loss[loss=0.3927, simple_loss=0.4307, pruned_loss=0.1774, over 5669379.10 frames. ], libri_tot_loss[loss=0.3614, simple_loss=0.4155, pruned_loss=0.1537, over 5724726.98 frames. ], giga_tot_loss[loss=0.3964, simple_loss=0.4326, pruned_loss=0.1801, over 5666505.90 frames. ], batch size: 213, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:31:44,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5449, 1.9397, 1.4245, 0.7742], device='cuda:0'), covar=tensor([0.1492, 0.1036, 0.0815, 0.1669], device='cuda:0'), in_proj_covar=tensor([0.1140, 0.1136, 0.1139, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 01:31:44,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.275e+02 1.825e+03 2.315e+03 3.318e+03 6.936e+03, threshold=4.631e+03, percent-clipped=4.0
+2023-03-01 01:32:09,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=55057.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 01:32:20,011 INFO [train.py:968] (0/2) Epoch 2, batch 9400, giga_loss[loss=0.3766, simple_loss=0.4187, pruned_loss=0.1672, over 28772.00 frames. ], tot_loss[loss=0.3961, simple_loss=0.4334, pruned_loss=0.1794, over 5673795.53 frames. ], libri_tot_loss[loss=0.3613, simple_loss=0.4155, pruned_loss=0.1536, over 5727231.65 frames. ], giga_tot_loss[loss=0.3996, simple_loss=0.4352, pruned_loss=0.182, over 5668532.53 frames. ], batch size: 119, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:33:08,895 INFO [train.py:968] (0/2) Epoch 2, batch 9450, libri_loss[loss=0.3715, simple_loss=0.4305, pruned_loss=0.1563, over 29662.00 frames. ], tot_loss[loss=0.3939, simple_loss=0.4318, pruned_loss=0.178, over 5676801.78 frames. ], libri_tot_loss[loss=0.3615, simple_loss=0.4157, pruned_loss=0.1536, over 5731502.12 frames. ], giga_tot_loss[loss=0.3973, simple_loss=0.4334, pruned_loss=0.1806, over 5667422.60 frames. ], batch size: 88, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:33:17,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6535, 3.7041, 4.3017, 1.9096], device='cuda:0'), covar=tensor([0.0448, 0.0447, 0.0891, 0.1783], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0533, 0.0853, 0.0546], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0006, 0.0010, 0.0007], device='cuda:0')
+2023-03-01 01:33:22,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.659e+03 1.890e+03 2.561e+03 7.280e+03, threshold=3.781e+03, percent-clipped=4.0
+2023-03-01 01:33:56,026 INFO [train.py:968] (0/2) Epoch 2, batch 9500, giga_loss[loss=0.3259, simple_loss=0.3997, pruned_loss=0.126, over 29008.00 frames. ], tot_loss[loss=0.391, simple_loss=0.4324, pruned_loss=0.1748, over 5686990.86 frames. ], libri_tot_loss[loss=0.3615, simple_loss=0.4157, pruned_loss=0.1536, over 5735707.88 frames. ], giga_tot_loss[loss=0.3947, simple_loss=0.4341, pruned_loss=0.1776, over 5674065.30 frames. ], batch size: 136, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:34:21,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0588, 0.9835, 0.8329, 1.2685], device='cuda:0'), covar=tensor([0.1135, 0.0502, 0.0593, 0.1283], device='cuda:0'), in_proj_covar=tensor([0.0324, 0.0220, 0.0225, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0021, 0.0019, 0.0031], device='cuda:0')
+2023-03-01 01:34:44,921 INFO [train.py:968] (0/2) Epoch 2, batch 9550, libri_loss[loss=0.3808, simple_loss=0.4375, pruned_loss=0.1621, over 29065.00 frames. ], tot_loss[loss=0.3919, simple_loss=0.4348, pruned_loss=0.1745, over 5684648.63 frames. ], libri_tot_loss[loss=0.3616, simple_loss=0.4159, pruned_loss=0.1536, over 5738903.71 frames. ], giga_tot_loss[loss=0.3953, simple_loss=0.4364, pruned_loss=0.1771, over 5670634.77 frames. ], batch size: 101, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:34:56,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.778e+02 1.525e+03 2.101e+03 2.925e+03 6.479e+03, threshold=4.203e+03, percent-clipped=13.0
+2023-03-01 01:35:08,545 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-01 01:35:20,859 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=55255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:35:31,797 INFO [train.py:968] (0/2) Epoch 2, batch 9600, giga_loss[loss=0.4181, simple_loss=0.4579, pruned_loss=0.1892, over 28895.00 frames. ], tot_loss[loss=0.3963, simple_loss=0.4379, pruned_loss=0.1773, over 5684119.18 frames. ], libri_tot_loss[loss=0.3608, simple_loss=0.4152, pruned_loss=0.1532, over 5743012.00 frames. ], giga_tot_loss[loss=0.4006, simple_loss=0.4404, pruned_loss=0.1804, over 5667607.28 frames. ], batch size: 174, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:36:00,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7631, 1.4630, 1.1703, 1.2378], device='cuda:0'), covar=tensor([0.0579, 0.0708, 0.0981, 0.0941], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0536, 0.0549, 0.0490], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 01:36:00,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-01 01:36:22,042 INFO [train.py:968] (0/2) Epoch 2, batch 9650, giga_loss[loss=0.4496, simple_loss=0.4788, pruned_loss=0.2102, over 28982.00 frames. ], tot_loss[loss=0.3999, simple_loss=0.4401, pruned_loss=0.1799, over 5676929.66 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.4151, pruned_loss=0.1531, over 5736259.06 frames. ], giga_tot_loss[loss=0.4045, simple_loss=0.4428, pruned_loss=0.183, over 5668504.27 frames. ], batch size: 128, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:36:34,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.081e+02 1.502e+03 2.145e+03 2.994e+03 1.092e+04, threshold=4.291e+03, percent-clipped=11.0
+2023-03-01 01:36:49,453 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-01 01:37:00,562 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-01 01:37:11,487 INFO [train.py:968] (0/2) Epoch 2, batch 9700, libri_loss[loss=0.3636, simple_loss=0.4251, pruned_loss=0.1511, over 29237.00 frames. ], tot_loss[loss=0.4022, simple_loss=0.4408, pruned_loss=0.1818, over 5675733.47 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4148, pruned_loss=0.1529, over 5738286.72 frames. ], giga_tot_loss[loss=0.407, simple_loss=0.4438, pruned_loss=0.1851, over 5666101.02 frames. ], batch size: 94, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:38:04,143 INFO [train.py:968] (0/2) Epoch 2, batch 9750, giga_loss[loss=0.4263, simple_loss=0.4344, pruned_loss=0.2091, over 23505.00 frames. ], tot_loss[loss=0.403, simple_loss=0.4404, pruned_loss=0.1828, over 5665516.59 frames. ], libri_tot_loss[loss=0.36, simple_loss=0.4146, pruned_loss=0.1527, over 5740422.07 frames. ], giga_tot_loss[loss=0.4076, simple_loss=0.4432, pruned_loss=0.186, over 5655000.42 frames. ], batch size: 705, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:38:15,332 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55432.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:38:16,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.787e+03 2.216e+03 3.048e+03 6.828e+03, threshold=4.431e+03, percent-clipped=4.0
+2023-03-01 01:38:49,152 INFO [train.py:968] (0/2) Epoch 2, batch 9800, giga_loss[loss=0.3336, simple_loss=0.403, pruned_loss=0.1321, over 28892.00 frames. ], tot_loss[loss=0.4004, simple_loss=0.439, pruned_loss=0.1809, over 5668661.70 frames. ], libri_tot_loss[loss=0.3606, simple_loss=0.415, pruned_loss=0.1531, over 5745360.73 frames. ], giga_tot_loss[loss=0.4047, simple_loss=0.4416, pruned_loss=0.1839, over 5653919.00 frames. ], batch size: 174, lr: 1.47e-02, grad_scale: 2.0
+2023-03-01 01:39:37,706 INFO [train.py:968] (0/2) Epoch 2, batch 9850, giga_loss[loss=0.4256, simple_loss=0.4722, pruned_loss=0.1895, over 28917.00 frames. ], tot_loss[loss=0.3964, simple_loss=0.4375, pruned_loss=0.1777, over 5672262.87 frames. ], libri_tot_loss[loss=0.3606, simple_loss=0.415, pruned_loss=0.1531, over 5747707.01 frames. ], giga_tot_loss[loss=0.4008, simple_loss=0.4401, pruned_loss=0.1807, over 5656599.37 frames. ], batch size: 227, lr: 1.47e-02, grad_scale: 2.0
+2023-03-01 01:39:53,007 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.340e+02 1.548e+03 2.173e+03 2.881e+03 7.015e+03, threshold=4.347e+03, percent-clipped=6.0
+2023-03-01 01:40:00,887 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=55541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:40:24,695 INFO [train.py:968] (0/2) Epoch 2, batch 9900, giga_loss[loss=0.4623, simple_loss=0.4839, pruned_loss=0.2204, over 28657.00 frames. ], tot_loss[loss=0.3955, simple_loss=0.4377, pruned_loss=0.1767, over 5676131.81 frames. ], libri_tot_loss[loss=0.3604, simple_loss=0.4148, pruned_loss=0.153, over 5746587.64 frames. ], giga_tot_loss[loss=0.3996, simple_loss=0.4403, pruned_loss=0.1795, over 5663905.16 frames. ], batch size: 336, lr: 1.47e-02, grad_scale: 2.0
+2023-03-01 01:40:31,783 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55575.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 01:40:33,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55578.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:41:01,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3056, 3.5969, 3.9634, 1.7545], device='cuda:0'), covar=tensor([0.0606, 0.0531, 0.1144, 0.2081], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0541, 0.0854, 0.0542], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0006, 0.0010, 0.0007], device='cuda:0')
+2023-03-01 01:41:02,250 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=55604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:41:06,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55607.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:41:16,560 INFO [train.py:968] (0/2) Epoch 2, batch 9950, giga_loss[loss=0.4213, simple_loss=0.4313, pruned_loss=0.2057, over 23223.00 frames. ], tot_loss[loss=0.3959, simple_loss=0.4378, pruned_loss=0.177, over 5669363.48 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4149, pruned_loss=0.1529, over 5750149.13 frames. ], giga_tot_loss[loss=0.4002, simple_loss=0.4405, pruned_loss=0.18, over 5654294.88 frames. ], batch size: 705, lr: 1.47e-02, grad_scale: 2.0
+2023-03-01 01:41:29,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:41:34,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.715e+03 2.193e+03 2.719e+03 6.943e+03, threshold=4.387e+03, percent-clipped=6.0
+2023-03-01 01:41:50,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7409, 1.5987, 1.5011, 1.5667], device='cuda:0'), covar=tensor([0.0801, 0.1420, 0.1117, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0821, 0.0627, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 01:41:58,974 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 01:42:07,094 INFO [train.py:968] (0/2) Epoch 2, batch 10000, libri_loss[loss=0.4485, simple_loss=0.4785, pruned_loss=0.2092, over 20029.00 frames. ], tot_loss[loss=0.3948, simple_loss=0.4367, pruned_loss=0.1764, over 5667525.50 frames. ], libri_tot_loss[loss=0.3605, simple_loss=0.4149, pruned_loss=0.153, over 5744485.05 frames. ], giga_tot_loss[loss=0.399, simple_loss=0.4393, pruned_loss=0.1793, over 5659376.41 frames. ], batch size: 187, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:42:37,566 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 01:42:53,865 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-03-01 01:42:55,251 INFO [train.py:968] (0/2) Epoch 2, batch 10050, giga_loss[loss=0.4078, simple_loss=0.4367, pruned_loss=0.1895, over 28782.00 frames. ], tot_loss[loss=0.3944, simple_loss=0.4352, pruned_loss=0.1768, over 5669460.10 frames. ], libri_tot_loss[loss=0.3604, simple_loss=0.4148, pruned_loss=0.153, over 5747612.52 frames. ], giga_tot_loss[loss=0.3986, simple_loss=0.4379, pruned_loss=0.1797, over 5658606.73 frames. ], batch size: 284, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:42:55,614 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5109, 2.8543, 1.6068, 1.1973], device='cuda:0'), covar=tensor([0.0962, 0.0602, 0.0917, 0.1616], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0464, 0.0351, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 01:43:12,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.778e+03 2.323e+03 3.150e+03 1.080e+04, threshold=4.646e+03, percent-clipped=6.0
+2023-03-01 01:43:46,769 INFO [train.py:968] (0/2) Epoch 2, batch 10100, giga_loss[loss=0.4046, simple_loss=0.4387, pruned_loss=0.1852, over 28232.00 frames. ], tot_loss[loss=0.3941, simple_loss=0.4339, pruned_loss=0.1772, over 5667161.00 frames. ], libri_tot_loss[loss=0.3601, simple_loss=0.4145, pruned_loss=0.1529, over 5749469.02 frames. ], giga_tot_loss[loss=0.3983, simple_loss=0.4367, pruned_loss=0.18, over 5655505.73 frames. ], batch size: 368, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:43:50,529 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:43:55,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:44:24,217 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:44:38,330 INFO [train.py:968] (0/2) Epoch 2, batch 10150, giga_loss[loss=0.4374, simple_loss=0.4562, pruned_loss=0.2093, over 27562.00 frames. ], tot_loss[loss=0.3926, simple_loss=0.4321, pruned_loss=0.1766, over 5675953.37 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4146, pruned_loss=0.153, over 5752103.65 frames. ], giga_tot_loss[loss=0.3964, simple_loss=0.4346, pruned_loss=0.1791, over 5662990.11 frames. ], batch size: 472, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:44:56,653 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.864e+02 1.598e+03 2.266e+03 2.985e+03 7.416e+03, threshold=4.533e+03, percent-clipped=4.0
+2023-03-01 01:45:03,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6053, 1.5015, 1.4900, 1.8377], device='cuda:0'), covar=tensor([0.1909, 0.1843, 0.1513, 0.2036], device='cuda:0'), in_proj_covar=tensor([0.0947, 0.0819, 0.0885, 0.0938], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 01:45:28,882 INFO [train.py:968] (0/2) Epoch 2, batch 10200, giga_loss[loss=0.4009, simple_loss=0.4409, pruned_loss=0.1805, over 29047.00 frames. ], tot_loss[loss=0.3915, simple_loss=0.4306, pruned_loss=0.1763, over 5675618.83 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4147, pruned_loss=0.153, over 5755230.69 frames. ], giga_tot_loss[loss=0.395, simple_loss=0.4327, pruned_loss=0.1787, over 5661255.41 frames. ], batch size: 155, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:45:59,755 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-01 01:46:03,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3243, 3.5729, 4.0349, 1.8107], device='cuda:0'), covar=tensor([0.0472, 0.0438, 0.0812, 0.1663], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0534, 0.0839, 0.0534], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0006, 0.0010, 0.0007], device='cuda:0')
+2023-03-01 01:46:14,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:46:16,406 INFO [train.py:968] (0/2) Epoch 2, batch 10250, giga_loss[loss=0.3609, simple_loss=0.4058, pruned_loss=0.158, over 28389.00 frames. ], tot_loss[loss=0.3913, simple_loss=0.4303, pruned_loss=0.1761, over 5675939.83 frames. ], libri_tot_loss[loss=0.361, simple_loss=0.4153, pruned_loss=0.1534, over 5755291.14 frames. ], giga_tot_loss[loss=0.3941, simple_loss=0.4318, pruned_loss=0.1783, over 5662478.33 frames. ], batch size: 71, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:46:31,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.941e+02 1.645e+03 2.146e+03 3.002e+03 7.981e+03, threshold=4.291e+03, percent-clipped=6.0
+2023-03-01 01:46:40,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9474, 1.8176, 1.6620, 1.6596], device='cuda:0'), covar=tensor([0.0837, 0.1594, 0.1219, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0830, 0.0636, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 01:47:04,093 INFO [train.py:968] (0/2) Epoch 2, batch 10300, giga_loss[loss=0.3508, simple_loss=0.4085, pruned_loss=0.1466, over 28805.00 frames. ], tot_loss[loss=0.3859, simple_loss=0.4274, pruned_loss=0.1722, over 5665082.23 frames. ], libri_tot_loss[loss=0.3608, simple_loss=0.4152, pruned_loss=0.1531, over 5751760.04 frames. ], giga_tot_loss[loss=0.3893, simple_loss=0.429, pruned_loss=0.1748, over 5655300.32 frames. ], batch size: 243, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:47:12,354 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:47:30,861 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-56000.pt
+2023-03-01 01:47:52,707 INFO [train.py:968] (0/2) Epoch 2, batch 10350, giga_loss[loss=0.3311, simple_loss=0.3941, pruned_loss=0.134, over 28572.00 frames. ], tot_loss[loss=0.3807, simple_loss=0.4239, pruned_loss=0.1687, over 5660691.74 frames. ], libri_tot_loss[loss=0.3606, simple_loss=0.4151, pruned_loss=0.1531, over 5750739.91 frames. ], giga_tot_loss[loss=0.3841, simple_loss=0.4256, pruned_loss=0.1713, over 5651582.15 frames. ], batch size: 85, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:47:54,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9818, 1.0789, 0.9556, 0.5486], device='cuda:0'), covar=tensor([0.0435, 0.0374, 0.0325, 0.0447], device='cuda:0'), in_proj_covar=tensor([0.0965, 0.0701, 0.0775, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 01:47:57,482 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-01 01:48:01,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7548, 1.5675, 1.4699, 1.5949], device='cuda:0'), covar=tensor([0.0825, 0.1496, 0.1351, 0.1143], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0820, 0.0641, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 01:48:09,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.300e+02 1.426e+03 1.803e+03 2.240e+03 3.954e+03, threshold=3.606e+03, percent-clipped=0.0
+2023-03-01 01:48:30,721 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:48:34,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:48:41,504 INFO [train.py:968] (0/2) Epoch 2, batch 10400, giga_loss[loss=0.3443, simple_loss=0.3957, pruned_loss=0.1465, over 28889.00 frames. ], tot_loss[loss=0.3796, simple_loss=0.4231, pruned_loss=0.168, over 5665626.84 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.4151, pruned_loss=0.1531, over 5750972.13 frames. ], giga_tot_loss[loss=0.3827, simple_loss=0.4246, pruned_loss=0.1704, over 5656316.30 frames. ], batch size: 199, lr: 1.46e-02, grad_scale: 8.0
+2023-03-01 01:48:50,144 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-01 01:49:05,703 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:49:37,193 INFO [train.py:968] (0/2) Epoch 2, batch 10450, giga_loss[loss=0.3546, simple_loss=0.3985, pruned_loss=0.1554, over 28435.00 frames. ], tot_loss[loss=0.3779, simple_loss=0.4203, pruned_loss=0.1677, over 5663807.02 frames. ], libri_tot_loss[loss=0.361, simple_loss=0.4153, pruned_loss=0.1534, over 5752550.34 frames. ], giga_tot_loss[loss=0.3802, simple_loss=0.4214, pruned_loss=0.1695, over 5654057.26 frames. ], batch size: 336, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:49:39,430 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:49:41,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:49:52,679 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.871e+03 2.409e+03 3.538e+03 1.035e+04, threshold=4.817e+03, percent-clipped=22.0
+2023-03-01 01:49:55,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3427, 1.2899, 1.2152, 1.6030], device='cuda:0'), covar=tensor([0.1778, 0.1883, 0.1501, 0.1817], device='cuda:0'), in_proj_covar=tensor([0.0964, 0.0824, 0.0890, 0.0943], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 01:49:57,489 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=56141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:50:13,673 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:50:30,669 INFO [train.py:968] (0/2) Epoch 2, batch 10500, giga_loss[loss=0.3164, simple_loss=0.374, pruned_loss=0.1294, over 28890.00 frames. ], tot_loss[loss=0.3767, simple_loss=0.419, pruned_loss=0.1672, over 5671802.39 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.4151, pruned_loss=0.1532, over 5753526.93 frames. ], giga_tot_loss[loss=0.3789, simple_loss=0.4201, pruned_loss=0.1689, over 5662527.69 frames. ], batch size: 227, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:51:20,197 INFO [train.py:968] (0/2) Epoch 2, batch 10550, giga_loss[loss=0.4057, simple_loss=0.4409, pruned_loss=0.1853, over 28271.00 frames. ], tot_loss[loss=0.3785, simple_loss=0.4212, pruned_loss=0.1679, over 5676066.80 frames. ], libri_tot_loss[loss=0.3605, simple_loss=0.4149, pruned_loss=0.1531, over 5755472.83 frames. ], giga_tot_loss[loss=0.3808, simple_loss=0.4223, pruned_loss=0.1696, over 5665473.29 frames. ], batch size: 368, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:51:37,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.012e+02 1.575e+03 2.036e+03 2.630e+03 7.219e+03, threshold=4.071e+03, percent-clipped=3.0
+2023-03-01 01:51:42,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=56241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:52:06,255 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5104, 2.4198, 1.7386, 0.7408], device='cuda:0'), covar=tensor([0.2306, 0.1058, 0.1424, 0.2582], device='cuda:0'), in_proj_covar=tensor([0.1152, 0.1153, 0.1146, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 01:52:08,406 INFO [train.py:968] (0/2) Epoch 2, batch 10600, giga_loss[loss=0.3178, simple_loss=0.3859, pruned_loss=0.1248, over 28454.00 frames. ], tot_loss[loss=0.3759, simple_loss=0.42, pruned_loss=0.1659, over 5656119.94 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4149, pruned_loss=0.1529, over 5746810.59 frames. ], giga_tot_loss[loss=0.3785, simple_loss=0.4211, pruned_loss=0.1679, over 5651703.78 frames. ], batch size: 65, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:52:57,691 INFO [train.py:968] (0/2) Epoch 2, batch 10650, libri_loss[loss=0.3787, simple_loss=0.4373, pruned_loss=0.16, over 29502.00 frames. ], tot_loss[loss=0.3748, simple_loss=0.4194, pruned_loss=0.1651, over 5635786.58 frames. ], libri_tot_loss[loss=0.3611, simple_loss=0.4157, pruned_loss=0.1533, over 5741288.27 frames. ], giga_tot_loss[loss=0.3766, simple_loss=0.4196, pruned_loss=0.1667, over 5634210.52 frames. ], batch size: 85, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:53:15,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.021e+02 1.283e+03 1.629e+03 2.043e+03 3.981e+03, threshold=3.257e+03, percent-clipped=0.0
+2023-03-01 01:53:17,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4619, 1.2363, 1.3310, 1.6277], device='cuda:0'), covar=tensor([0.1825, 0.1967, 0.1496, 0.1902], device='cuda:0'), in_proj_covar=tensor([0.0963, 0.0818, 0.0886, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 01:53:43,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2632, 1.5385, 1.1467, 1.3725], device='cuda:0'), covar=tensor([0.1149, 0.0443, 0.0526, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0326, 0.0221, 0.0225, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0022, 0.0019, 0.0032], device='cuda:0')
+2023-03-01 01:53:48,695 INFO [train.py:968] (0/2) Epoch 2, batch 10700, giga_loss[loss=0.4114, simple_loss=0.4223, pruned_loss=0.2003, over 23665.00 frames. ], tot_loss[loss=0.3778, simple_loss=0.4207, pruned_loss=0.1675, over 5636880.51 frames. ], libri_tot_loss[loss=0.3617, simple_loss=0.416, pruned_loss=0.1537, over 5744371.15 frames. ], giga_tot_loss[loss=0.379, simple_loss=0.4207, pruned_loss=0.1686, over 5630926.94 frames. ], batch size: 705, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:54:24,123 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-01 01:54:37,380 INFO [train.py:968] (0/2) Epoch 2, batch 10750, giga_loss[loss=0.4559, simple_loss=0.4785, pruned_loss=0.2167, over 27900.00 frames. ], tot_loss[loss=0.3824, simple_loss=0.4241, pruned_loss=0.1704, over 5639142.60 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.4165, pruned_loss=0.1539, over 5746314.72 frames. ], giga_tot_loss[loss=0.3835, simple_loss=0.4238, pruned_loss=0.1716, over 5629577.34 frames. ], batch size: 412, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:54:56,918 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.934e+02 1.771e+03 2.327e+03 3.104e+03 6.166e+03, threshold=4.654e+03, percent-clipped=22.0
+2023-03-01 01:55:26,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1463, 1.1233, 0.9232, 1.2943], device='cuda:0'), covar=tensor([0.0903, 0.0371, 0.0467, 0.0933], device='cuda:0'), in_proj_covar=tensor([0.0321, 0.0219, 0.0225, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0021, 0.0019, 0.0031], device='cuda:0')
+2023-03-01 01:55:27,469 INFO [train.py:968] (0/2) Epoch 2, batch 10800, giga_loss[loss=0.4299, simple_loss=0.4517, pruned_loss=0.204, over 28872.00 frames. ], tot_loss[loss=0.3855, simple_loss=0.4269, pruned_loss=0.172, over 5649672.25 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.4171, pruned_loss=0.1543, over 5750927.47 frames. ], giga_tot_loss[loss=0.3863, simple_loss=0.4264, pruned_loss=0.1732, over 5634638.12 frames. ], batch size: 112, lr: 1.46e-02, grad_scale: 8.0
+2023-03-01 01:55:46,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4923, 1.6640, 1.2950, 1.4592], device='cuda:0'), covar=tensor([0.1081, 0.0407, 0.0498, 0.1236], device='cuda:0'), in_proj_covar=tensor([0.0323, 0.0219, 0.0225, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0022, 0.0019, 0.0031], device='cuda:0')
+2023-03-01 01:55:51,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5131, 2.3641, 1.6748, 0.8396], device='cuda:0'), covar=tensor([0.1956, 0.1064, 0.1177, 0.2053], device='cuda:0'), in_proj_covar=tensor([0.1154, 0.1157, 0.1141, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 01:56:01,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5326, 2.0787, 1.6929, 1.4574], device='cuda:0'), covar=tensor([0.1066, 0.0399, 0.0482, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0324, 0.0220, 0.0226, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0022, 0.0019, 0.0032], device='cuda:0')
+2023-03-01 01:56:02,015 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-01 01:56:10,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:56:12,195 INFO [train.py:968] (0/2) Epoch 2, batch 10850, giga_loss[loss=0.4476, simple_loss=0.4444, pruned_loss=0.2253, over 23236.00 frames. ], tot_loss[loss=0.3853, simple_loss=0.427, pruned_loss=0.1718, over 5655982.63 frames. ], libri_tot_loss[loss=0.3625, simple_loss=0.4168, pruned_loss=0.1541, over 5749790.23 frames. ], giga_tot_loss[loss=0.3873, simple_loss=0.4273, pruned_loss=0.1737, over 5640878.35 frames. ], batch size: 705, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:56:29,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.574e+02 1.545e+03 2.132e+03 2.967e+03 1.014e+04, threshold=4.265e+03, percent-clipped=7.0
+2023-03-01 01:56:31,027 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=56538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:56:59,922 INFO [train.py:968] (0/2) Epoch 2, batch 10900, libri_loss[loss=0.3674, simple_loss=0.4331, pruned_loss=0.1508, over 29248.00 frames. ], tot_loss[loss=0.3873, simple_loss=0.4282, pruned_loss=0.1732, over 5658661.62 frames. ], libri_tot_loss[loss=0.3617, simple_loss=0.4161, pruned_loss=0.1537, over 5745504.04 frames. ], giga_tot_loss[loss=0.3905, simple_loss=0.4295, pruned_loss=0.1757, over 5646823.02 frames. ], batch size: 94, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 01:57:51,119 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:57:53,645 INFO [train.py:968] (0/2) Epoch 2, batch 10950, giga_loss[loss=0.3998, simple_loss=0.4469, pruned_loss=0.1763, over 28915.00 frames. ], tot_loss[loss=0.3898, simple_loss=0.43, pruned_loss=0.1748, over 5656078.08 frames. ], libri_tot_loss[loss=0.3614, simple_loss=0.4158, pruned_loss=0.1535, over 5746620.62 frames. ], giga_tot_loss[loss=0.3931, simple_loss=0.4315, pruned_loss=0.1774, over 5643930.13 frames. ], batch size: 227, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 01:58:13,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.671e+03 2.092e+03 2.754e+03 6.957e+03, threshold=4.184e+03, percent-clipped=4.0
+2023-03-01 01:58:27,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7356, 1.7164, 1.5184, 1.2046], device='cuda:0'), covar=tensor([0.0350, 0.0307, 0.0285, 0.0403], device='cuda:0'), in_proj_covar=tensor([0.1005, 0.0738, 0.0805, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 01:58:38,326 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:58:41,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:58:47,699 INFO [train.py:968] (0/2) Epoch 2, batch 11000, giga_loss[loss=0.3525, simple_loss=0.4105, pruned_loss=0.1473, over 29102.00 frames. ], tot_loss[loss=0.3918, simple_loss=0.432, pruned_loss=0.1758, over 5644732.29 frames. ], libri_tot_loss[loss=0.3613, simple_loss=0.4157, pruned_loss=0.1534, over 5737313.49 frames. ], giga_tot_loss[loss=0.3951, simple_loss=0.4336, pruned_loss=0.1783, over 5641886.97 frames. ], batch size: 155, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 01:59:11,657 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:59:22,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0593, 0.9218, 0.7673, 1.2555], device='cuda:0'), covar=tensor([0.1057, 0.0468, 0.0553, 0.1235], device='cuda:0'), in_proj_covar=tensor([0.0322, 0.0216, 0.0224, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0021, 0.0019, 0.0032], device='cuda:0')
+2023-03-01 01:59:44,534 INFO [train.py:968] (0/2) Epoch 2, batch 11050, giga_loss[loss=0.3949, simple_loss=0.4407, pruned_loss=0.1746, over 28473.00 frames. ], tot_loss[loss=0.3938, simple_loss=0.4326, pruned_loss=0.1775, over 5646639.00 frames. ], libri_tot_loss[loss=0.3615, simple_loss=0.4159, pruned_loss=0.1536, over 5735426.42 frames. ], giga_tot_loss[loss=0.3965, simple_loss=0.4338, pruned_loss=0.1796, over 5645276.28 frames. ], batch size: 336, lr: 1.45e-02, grad_scale: 2.0
+2023-03-01 02:00:03,048 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.909e+02 1.762e+03 2.339e+03 3.286e+03 1.386e+04, threshold=4.679e+03, percent-clipped=14.0
+2023-03-01 02:00:17,265 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-01 02:00:29,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:00:32,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:00:41,987 INFO [train.py:968] (0/2) Epoch 2, batch 11100, giga_loss[loss=0.5031, simple_loss=0.4798, pruned_loss=0.2633, over 23601.00 frames. ], tot_loss[loss=0.392, simple_loss=0.4309, pruned_loss=0.1766, over 5652701.23 frames. ], libri_tot_loss[loss=0.3616, simple_loss=0.4161, pruned_loss=0.1535, over 5735030.73 frames. ], giga_tot_loss[loss=0.3943, simple_loss=0.4318, pruned_loss=0.1784, over 5651240.47 frames. ], batch size: 705, lr: 1.45e-02, grad_scale: 2.0
+2023-03-01 02:01:09,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:01:25,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1524, 1.2777, 1.0638, 1.1556], device='cuda:0'), covar=tensor([0.1882, 0.1945, 0.1671, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.0951, 0.0807, 0.0879, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 02:01:35,514 INFO [train.py:968] (0/2) Epoch 2, batch 11150, giga_loss[loss=0.3744, simple_loss=0.4237, pruned_loss=0.1625, over 29040.00 frames. ], tot_loss[loss=0.3901, simple_loss=0.4295, pruned_loss=0.1753, over 5658233.00 frames. ], libri_tot_loss[loss=0.3618, simple_loss=0.4164, pruned_loss=0.1536, over 5740360.31 frames. ], giga_tot_loss[loss=0.3929, simple_loss=0.4305, pruned_loss=0.1777, over 5649096.18 frames. ], batch size: 128, lr: 1.45e-02, grad_scale: 2.0
+2023-03-01 02:01:54,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.591e+02 1.545e+03 1.912e+03 2.567e+03 5.071e+03, threshold=3.825e+03, percent-clipped=1.0
+2023-03-01 02:02:02,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2109, 3.2835, 1.9095, 1.9215], device='cuda:0'), covar=tensor([0.0737, 0.0492, 0.0754, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0461, 0.0345, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 02:02:27,423 INFO [train.py:968] (0/2) Epoch 2, batch 11200, giga_loss[loss=0.4384, simple_loss=0.4594, pruned_loss=0.2087, over 28086.00 frames. ], tot_loss[loss=0.3898, simple_loss=0.4286, pruned_loss=0.1755, over 5664707.72 frames. ], libri_tot_loss[loss=0.3619, simple_loss=0.4164, pruned_loss=0.1537, over 5743033.33 frames. ], giga_tot_loss[loss=0.3923, simple_loss=0.4295, pruned_loss=0.1775, over 5653974.25 frames. ], batch size: 412, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:02:34,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7595, 1.8246, 3.7352, 3.0045], device='cuda:0'), covar=tensor([0.1520, 0.1315, 0.0319, 0.0453], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0487, 0.0627, 0.0514], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 02:03:06,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:03:08,663 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=56916.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 02:03:10,481 INFO [train.py:968] (0/2) Epoch 2, batch 11250, giga_loss[loss=0.4393, simple_loss=0.4666, pruned_loss=0.206, over 29006.00 frames. ], tot_loss[loss=0.3878, simple_loss=0.4269, pruned_loss=0.1743, over 5674031.60 frames. ], libri_tot_loss[loss=0.3608, simple_loss=0.4154, pruned_loss=0.1531, over 5748624.19 frames. ], giga_tot_loss[loss=0.3919, simple_loss=0.4291, pruned_loss=0.1774, over 5657317.64 frames. ], batch size: 164, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:03:31,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.709e+02 1.728e+03 2.189e+03 3.092e+03 7.879e+03, threshold=4.378e+03, percent-clipped=11.0
+2023-03-01 02:04:03,199 INFO [train.py:968] (0/2) Epoch 2, batch 11300, giga_loss[loss=0.4443, simple_loss=0.4403, pruned_loss=0.2241, over 23590.00 frames. ], tot_loss[loss=0.388, simple_loss=0.4268, pruned_loss=0.1746, over 5669658.88 frames. ], libri_tot_loss[loss=0.3606, simple_loss=0.4153, pruned_loss=0.1529, over 5751079.12 frames. ], giga_tot_loss[loss=0.3918, simple_loss=0.4288, pruned_loss=0.1774, over 5653186.64 frames. ], batch size: 705, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:04:33,550 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-01 02:04:51,250 INFO [train.py:968] (0/2) Epoch 2, batch 11350, libri_loss[loss=0.3005, simple_loss=0.3595, pruned_loss=0.1208, over 29490.00 frames. ], tot_loss[loss=0.39, simple_loss=0.4282, pruned_loss=0.176, over 5684385.26 frames. ], libri_tot_loss[loss=0.3596, simple_loss=0.4143, pruned_loss=0.1524, over 5756844.28 frames. ], giga_tot_loss[loss=0.3953, simple_loss=0.4312, pruned_loss=0.1797, over 5662495.03 frames. ], batch size: 70, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:04:59,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1424, 2.0831, 1.1587, 1.0462], device='cuda:0'), covar=tensor([0.1295, 0.0825, 0.1233, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0450, 0.0345, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 02:05:15,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 2.001e+03 2.665e+03 3.709e+03 7.783e+03, threshold=5.331e+03, percent-clipped=19.0
+2023-03-01 02:05:31,864 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:05:33,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:05:43,719 INFO [train.py:968] (0/2) Epoch 2, batch 11400, giga_loss[loss=0.4118, simple_loss=0.4463, pruned_loss=0.1886, over 28568.00 frames. ], tot_loss[loss=0.3931, simple_loss=0.4305, pruned_loss=0.1778, over 5678558.54 frames. ], libri_tot_loss[loss=0.3593, simple_loss=0.4141, pruned_loss=0.1523, over 5757428.77 frames. ], giga_tot_loss[loss=0.3978, simple_loss=0.4332, pruned_loss=0.1812, over 5660064.24 frames. ], batch size: 307, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:06:00,687 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:06:11,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-01 02:06:30,746 INFO [train.py:968] (0/2) Epoch 2, batch 11450, giga_loss[loss=0.3982, simple_loss=0.4417, pruned_loss=0.1774, over 29080.00 frames. ], tot_loss[loss=0.3911, simple_loss=0.4295, pruned_loss=0.1764, over 5675603.90 frames. ], libri_tot_loss[loss=0.3589, simple_loss=0.4139, pruned_loss=0.152, over 5758308.96 frames. ], giga_tot_loss[loss=0.3962, simple_loss=0.4323, pruned_loss=0.18, over 5657707.74 frames. ], batch size: 136, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:06:51,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.556e+02 1.597e+03 2.069e+03 2.574e+03 6.533e+03, threshold=4.138e+03, percent-clipped=1.0
+2023-03-01 02:07:23,701 INFO [train.py:968] (0/2) Epoch 2, batch 11500, giga_loss[loss=0.4207, simple_loss=0.4499, pruned_loss=0.1958, over 28694.00 frames. ], tot_loss[loss=0.393, simple_loss=0.4299, pruned_loss=0.178, over 5667706.32 frames. ], libri_tot_loss[loss=0.3591, simple_loss=0.4139, pruned_loss=0.1521, over 5762330.15 frames. ], giga_tot_loss[loss=0.3979, simple_loss=0.4327, pruned_loss=0.1815, over 5647170.17 frames. ], batch size: 262, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:07:49,520 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:08:11,603 INFO [train.py:968] (0/2) Epoch 2, batch 11550, giga_loss[loss=0.3724, simple_loss=0.4177, pruned_loss=0.1635, over 28895.00 frames. ], tot_loss[loss=0.3927, simple_loss=0.4301, pruned_loss=0.1776, over 5663590.52 frames. ], libri_tot_loss[loss=0.3592, simple_loss=0.4141, pruned_loss=0.1521, over 5755963.81 frames. ], giga_tot_loss[loss=0.3972, simple_loss=0.4324, pruned_loss=0.181, over 5650271.89 frames. ], batch size: 136, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:08:20,313 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-01 02:08:29,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.131e+03 1.653e+03 2.173e+03 3.078e+03 9.496e+03, threshold=4.346e+03, percent-clipped=9.0
+2023-03-01 02:08:47,713 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-01 02:09:00,069 INFO [train.py:968] (0/2) Epoch 2, batch 11600, giga_loss[loss=0.402, simple_loss=0.4434, pruned_loss=0.1803, over 28649.00 frames. ], tot_loss[loss=0.3934, simple_loss=0.431, pruned_loss=0.1779, over 5672641.46 frames. ], libri_tot_loss[loss=0.3591, simple_loss=0.4141, pruned_loss=0.1521, over 5757303.77 frames. ], giga_tot_loss[loss=0.3982, simple_loss=0.4334, pruned_loss=0.1815, over 5657905.16 frames. ], batch size: 336, lr: 1.45e-02, grad_scale: 8.0
+2023-03-01 02:09:19,619 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-01 02:09:22,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57291.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 02:09:32,420 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:09:47,521 INFO [train.py:968] (0/2) Epoch 2, batch 11650, giga_loss[loss=0.3816, simple_loss=0.4267, pruned_loss=0.1682, over 28775.00 frames. ], tot_loss[loss=0.392, simple_loss=0.4305, pruned_loss=0.1767, over 5678744.02 frames. ], libri_tot_loss[loss=0.3585, simple_loss=0.4136, pruned_loss=0.1517, over 5761856.44 frames. ], giga_tot_loss[loss=0.3973, simple_loss=0.4334, pruned_loss=0.1806, over 5660466.37 frames. ], batch size: 243, lr: 1.45e-02, grad_scale: 8.0
+2023-03-01 02:10:05,459 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-01 02:10:10,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.638e+03 1.994e+03 2.475e+03 4.977e+03, threshold=3.988e+03, percent-clipped=3.0
+2023-03-01 02:10:17,106 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8415, 1.5970, 1.5277, 1.4671], device='cuda:0'), covar=tensor([0.0790, 0.1511, 0.1299, 0.1213], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0826, 0.0640, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 02:10:43,003 INFO [train.py:968] (0/2) Epoch 2, batch 11700, giga_loss[loss=0.4187, simple_loss=0.4517, pruned_loss=0.1929, over 28628.00 frames. ], tot_loss[loss=0.3919, simple_loss=0.4308, pruned_loss=0.1764, over 5688506.07 frames. ], libri_tot_loss[loss=0.358, simple_loss=0.4134, pruned_loss=0.1513, over 5764295.07 frames. ], giga_tot_loss[loss=0.3971, simple_loss=0.4336, pruned_loss=0.1803, over 5670345.03 frames. ], batch size: 307, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:11:01,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:11:03,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:11:04,534 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4671, 2.4467, 1.4820, 1.2639], device='cuda:0'), covar=tensor([0.0899, 0.0562, 0.0839, 0.1386], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0458, 0.0344, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 02:11:36,393 INFO [train.py:968] (0/2) Epoch 2, batch 11750, libri_loss[loss=0.3072, simple_loss=0.3728, pruned_loss=0.1208, over 29574.00 frames. ], tot_loss[loss=0.3954, simple_loss=0.4331, pruned_loss=0.1788, over 5682070.20 frames. ], libri_tot_loss[loss=0.3577, simple_loss=0.4133, pruned_loss=0.1511, over 5764725.36 frames. ], giga_tot_loss[loss=0.401, simple_loss=0.436, pruned_loss=0.183, over 5664652.94 frames. ], batch size: 74, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:11:48,418 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57434.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 02:11:51,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57437.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 02:11:56,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.970e+02 1.612e+03 2.030e+03 2.877e+03 1.059e+04, threshold=4.060e+03, percent-clipped=10.0
+2023-03-01 02:12:21,514 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57466.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 02:12:24,436 INFO [train.py:968] (0/2) Epoch 2, batch 11800, giga_loss[loss=0.441, simple_loss=0.4669, pruned_loss=0.2075, over 28594.00 frames. ], tot_loss[loss=0.3945, simple_loss=0.4326, pruned_loss=0.1782, over 5689776.29 frames. ], libri_tot_loss[loss=0.3568, simple_loss=0.4125, pruned_loss=0.1505, over 5765617.28 frames. ], giga_tot_loss[loss=0.4003, simple_loss=0.4359, pruned_loss=0.1824, over 5673674.85 frames. ], batch size: 336, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:13:14,335 INFO [train.py:968] (0/2) Epoch 2, batch 11850, giga_loss[loss=0.4331, simple_loss=0.4408, pruned_loss=0.2127, over 23552.00 frames. ], tot_loss[loss=0.3925, simple_loss=0.4322, pruned_loss=0.1764, over 5688349.13 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4124, pruned_loss=0.1504, over 5768869.45 frames. ], giga_tot_loss[loss=0.3984, simple_loss=0.4355, pruned_loss=0.1807, over 5670374.50 frames. ], batch size: 705, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:13:35,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.975e+02 1.467e+03 1.963e+03 2.516e+03 5.538e+03, threshold=3.926e+03, percent-clipped=3.0
+2023-03-01 02:14:06,459 INFO [train.py:968] (0/2) Epoch 2, batch 11900, giga_loss[loss=0.4456, simple_loss=0.4654, pruned_loss=0.2129, over 27453.00 frames. ], tot_loss[loss=0.3917, simple_loss=0.4319, pruned_loss=0.1757, over 5681313.44 frames. ], libri_tot_loss[loss=0.3572, simple_loss=0.4127, pruned_loss=0.1509, over 5770802.06 frames. ], giga_tot_loss[loss=0.3965, simple_loss=0.4346, pruned_loss=0.1792, over 5663692.83 frames. ], batch size: 472, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:14:09,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:14:15,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:14:26,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-01 02:14:26,939 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:14:44,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5333, 1.9337, 1.6556, 1.6895], device='cuda:0'), covar=tensor([0.1124, 0.1428, 0.1052, 0.0692], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0839, 0.0729, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0006], device='cuda:0')
+2023-03-01 02:14:47,452 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.90 vs. limit=5.0
+2023-03-01 02:14:54,817 INFO [train.py:968] (0/2) Epoch 2, batch 11950, giga_loss[loss=0.3962, simple_loss=0.4318, pruned_loss=0.1803, over 28924.00 frames. ], tot_loss[loss=0.3894, simple_loss=0.43, pruned_loss=0.1743, over 5676213.75 frames. ], libri_tot_loss[loss=0.3572, simple_loss=0.4127, pruned_loss=0.1509, over 5762940.35 frames. ], giga_tot_loss[loss=0.3937, simple_loss=0.4325, pruned_loss=0.1775, over 5667950.51 frames. ], batch size: 145, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:14:59,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4320, 1.9951, 1.6073, 1.6038], device='cuda:0'), covar=tensor([0.1433, 0.1564, 0.1220, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0826, 0.0841, 0.0731, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0006], device='cuda:0')
+2023-03-01 02:15:10,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:15:12,679 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.563e+03 1.979e+03 2.659e+03 6.524e+03, threshold=3.958e+03, percent-clipped=5.0
+2023-03-01 02:15:29,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-01 02:15:39,719 INFO [train.py:968] (0/2) Epoch 2, batch 12000, giga_loss[loss=0.4394, simple_loss=0.4414, pruned_loss=0.2187, over 23533.00 frames. ], tot_loss[loss=0.388, simple_loss=0.4288, pruned_loss=0.1736, over 5681287.83 frames. ], libri_tot_loss[loss=0.3564, simple_loss=0.4122, pruned_loss=0.1504, over 5766289.17 frames. ], giga_tot_loss[loss=0.393, simple_loss=0.4317, pruned_loss=0.1772, over 5669491.08 frames. ], batch size: 705, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:15:39,724 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 02:15:46,129 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2928, 1.4174, 1.0586, 1.3565], device='cuda:0'), covar=tensor([0.1324, 0.0467, 0.0626, 0.1556], device='cuda:0'), in_proj_covar=tensor([0.0320, 0.0218, 0.0222, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0022, 0.0020, 0.0032], device='cuda:0')
+2023-03-01 02:15:47,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6689, 1.4604, 3.2716, 2.9239], device='cuda:0'), covar=tensor([0.1587, 0.1521, 0.0406, 0.0431], device='cuda:0'), in_proj_covar=tensor([0.0512, 0.0495, 0.0653, 0.0527], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 02:15:48,565 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.28, simple_loss=0.3748, pruned_loss=0.09263, over 944034.00 frames. 
+2023-03-01 02:15:48,566 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19478MB
+2023-03-01 02:15:54,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:16:31,046 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:16:33,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:16:34,844 INFO [train.py:968] (0/2) Epoch 2, batch 12050, libri_loss[loss=0.3586, simple_loss=0.4004, pruned_loss=0.1584, over 29613.00 frames. ], tot_loss[loss=0.3907, simple_loss=0.4306, pruned_loss=0.1754, over 5671741.63 frames. ], libri_tot_loss[loss=0.3568, simple_loss=0.4125, pruned_loss=0.1505, over 5766726.26 frames. ], giga_tot_loss[loss=0.3957, simple_loss=0.4333, pruned_loss=0.179, over 5658380.01 frames. ], batch size: 74, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:16:54,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.324e+02 1.477e+03 2.029e+03 2.616e+03 5.769e+03, threshold=4.057e+03, percent-clipped=9.0
+2023-03-01 02:17:00,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:17:14,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:17:17,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:17:22,075 INFO [train.py:968] (0/2) Epoch 2, batch 12100, giga_loss[loss=0.3832, simple_loss=0.4244, pruned_loss=0.171, over 28658.00 frames. ], tot_loss[loss=0.3909, simple_loss=0.4309, pruned_loss=0.1754, over 5665624.46 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4124, pruned_loss=0.1504, over 5759214.18 frames. ], giga_tot_loss[loss=0.3958, simple_loss=0.4336, pruned_loss=0.179, over 5659720.62 frames. ], batch size: 262, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:17:28,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3729, 1.3311, 1.2507, 1.6531], device='cuda:0'), covar=tensor([0.1610, 0.1578, 0.1273, 0.1814], device='cuda:0'), in_proj_covar=tensor([0.0954, 0.0810, 0.0884, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 02:17:44,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:18:13,923 INFO [train.py:968] (0/2) Epoch 2, batch 12150, libri_loss[loss=0.3798, simple_loss=0.4369, pruned_loss=0.1614, over 29380.00 frames. ], tot_loss[loss=0.3901, simple_loss=0.4298, pruned_loss=0.1752, over 5671531.45 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.4126, pruned_loss=0.1504, over 5765273.28 frames. ], giga_tot_loss[loss=0.3953, simple_loss=0.4324, pruned_loss=0.1791, over 5657818.35 frames. ], batch size: 92, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:18:14,256 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:18:17,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:18:22,747 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-01 02:18:33,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.608e+03 1.970e+03 2.902e+03 1.211e+04, threshold=3.940e+03, percent-clipped=6.0
+2023-03-01 02:18:44,252 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:18:47,000 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:18:49,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3202, 1.2603, 1.1915, 1.3468], device='cuda:0'), covar=tensor([0.1707, 0.1787, 0.1485, 0.1757], device='cuda:0'), in_proj_covar=tensor([0.0949, 0.0796, 0.0879, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 02:19:00,854 INFO [train.py:968] (0/2) Epoch 2, batch 12200, libri_loss[loss=0.3024, simple_loss=0.3737, pruned_loss=0.1156, over 29583.00 frames. ], tot_loss[loss=0.3915, simple_loss=0.4304, pruned_loss=0.1763, over 5664199.96 frames. ], libri_tot_loss[loss=0.3565, simple_loss=0.4125, pruned_loss=0.1502, over 5757629.18 frames. ], giga_tot_loss[loss=0.3974, simple_loss=0.4333, pruned_loss=0.1807, over 5656260.57 frames. ], batch size: 76, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:19:19,937 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:33,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:36,879 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:37,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:39,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:48,305 INFO [train.py:968] (0/2) Epoch 2, batch 12250, giga_loss[loss=0.4661, simple_loss=0.4771, pruned_loss=0.2275, over 27624.00 frames. ], tot_loss[loss=0.3931, simple_loss=0.4314, pruned_loss=0.1774, over 5670094.72 frames. ], libri_tot_loss[loss=0.3562, simple_loss=0.4124, pruned_loss=0.15, over 5760072.95 frames. ], giga_tot_loss[loss=0.3989, simple_loss=0.4343, pruned_loss=0.1817, over 5659544.66 frames. ], batch size: 472, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:20:05,144 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3874, 1.7175, 1.5732, 1.6127], device='cuda:0'), covar=tensor([0.1199, 0.1673, 0.1074, 0.0668], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0834, 0.0723, 0.0601], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0006], device='cuda:0')
+2023-03-01 02:20:09,537 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:20:11,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:20:13,181 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.704e+02 1.603e+03 2.091e+03 3.008e+03 1.017e+04, threshold=4.183e+03, percent-clipped=11.0
+2023-03-01 02:20:27,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:20:37,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:20:39,837 INFO [train.py:968] (0/2) Epoch 2, batch 12300, libri_loss[loss=0.4394, simple_loss=0.4797, pruned_loss=0.1995, over 29371.00 frames. ], tot_loss[loss=0.3935, simple_loss=0.4319, pruned_loss=0.1775, over 5664841.99 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.4128, pruned_loss=0.1503, over 5761038.89 frames. ], giga_tot_loss[loss=0.3984, simple_loss=0.4343, pruned_loss=0.1813, over 5653597.20 frames. ], batch size: 92, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:21:08,740 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-58000.pt
+2023-03-01 02:21:21,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:21:27,321 INFO [train.py:968] (0/2) Epoch 2, batch 12350, giga_loss[loss=0.4245, simple_loss=0.4279, pruned_loss=0.2106, over 23588.00 frames. ], tot_loss[loss=0.3927, simple_loss=0.4309, pruned_loss=0.1772, over 5653641.81 frames. ], libri_tot_loss[loss=0.3564, simple_loss=0.4126, pruned_loss=0.1501, over 5765340.78 frames. ], giga_tot_loss[loss=0.3979, simple_loss=0.4335, pruned_loss=0.1811, over 5638128.97 frames. ], batch size: 705, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:21:49,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.561e+03 1.959e+03 2.678e+03 6.737e+03, threshold=3.918e+03, percent-clipped=5.0
+2023-03-01 02:22:15,612 INFO [train.py:968] (0/2) Epoch 2, batch 12400, giga_loss[loss=0.4683, simple_loss=0.4641, pruned_loss=0.2363, over 26661.00 frames. ], tot_loss[loss=0.3924, simple_loss=0.4308, pruned_loss=0.177, over 5655321.84 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4124, pruned_loss=0.1501, over 5765134.68 frames. ], giga_tot_loss[loss=0.398, simple_loss=0.4337, pruned_loss=0.1812, over 5638550.76 frames. ], batch size: 555, lr: 1.44e-02, grad_scale: 8.0
+2023-03-01 02:22:45,867 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:22:48,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:22:53,315 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:22:56,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:02,683 INFO [train.py:968] (0/2) Epoch 2, batch 12450, libri_loss[loss=0.3157, simple_loss=0.3697, pruned_loss=0.1308, over 29658.00 frames. ], tot_loss[loss=0.3924, simple_loss=0.4313, pruned_loss=0.1768, over 5658277.31 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.4124, pruned_loss=0.1499, over 5766062.66 frames. ], giga_tot_loss[loss=0.3982, simple_loss=0.4342, pruned_loss=0.1811, over 5640702.96 frames. ], batch size: 69, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:23:09,026 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=58126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:13,772 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:23,639 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:25,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.988e+02 1.685e+03 2.196e+03 3.176e+03 7.265e+03, threshold=4.392e+03, percent-clipped=10.0
+2023-03-01 02:23:39,102 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:40,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5967, 1.5136, 1.1884, 1.1951], device='cuda:0'), covar=tensor([0.0583, 0.0575, 0.0923, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0530, 0.0562, 0.0496], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 02:23:42,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:51,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:53,121 INFO [train.py:968] (0/2) Epoch 2, batch 12500, giga_loss[loss=0.4277, simple_loss=0.4503, pruned_loss=0.2025, over 28461.00 frames. ], tot_loss[loss=0.3921, simple_loss=0.4309, pruned_loss=0.1766, over 5661413.89 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.4124, pruned_loss=0.1499, over 5769304.72 frames. ], giga_tot_loss[loss=0.3976, simple_loss=0.4338, pruned_loss=0.1807, over 5641802.03 frames. ], batch size: 65, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:24:11,851 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:24:14,363 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-01 02:24:40,914 INFO [train.py:968] (0/2) Epoch 2, batch 12550, giga_loss[loss=0.3726, simple_loss=0.4208, pruned_loss=0.1622, over 28841.00 frames. ], tot_loss[loss=0.3903, simple_loss=0.4295, pruned_loss=0.1755, over 5669561.24 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4122, pruned_loss=0.1497, over 5773445.94 frames. ], giga_tot_loss[loss=0.396, simple_loss=0.4324, pruned_loss=0.1798, over 5647320.80 frames. ], batch size: 243, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:24:51,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:25:06,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.840e+02 1.583e+03 2.269e+03 3.071e+03 7.549e+03, threshold=4.538e+03, percent-clipped=9.0
+2023-03-01 02:25:06,831 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-01 02:25:28,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:25:33,552 INFO [train.py:968] (0/2) Epoch 2, batch 12600, giga_loss[loss=0.3873, simple_loss=0.4247, pruned_loss=0.175, over 28635.00 frames. ], tot_loss[loss=0.3888, simple_loss=0.4278, pruned_loss=0.1749, over 5679632.10 frames. ], libri_tot_loss[loss=0.3556, simple_loss=0.4122, pruned_loss=0.1496, over 5774123.71 frames. ], giga_tot_loss[loss=0.3935, simple_loss=0.4302, pruned_loss=0.1784, over 5661295.29 frames. ], batch size: 242, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:25:33,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0625, 1.9844, 4.9407, 3.5355], device='cuda:0'), covar=tensor([0.1476, 0.1326, 0.0278, 0.0382], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0487, 0.0636, 0.0511], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 02:25:54,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2923, 1.7090, 1.2952, 1.2386], device='cuda:0'), covar=tensor([0.0850, 0.0636, 0.0787, 0.1299], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0466, 0.0348, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 02:26:14,220 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:26:16,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:26:23,218 INFO [train.py:968] (0/2) Epoch 2, batch 12650, giga_loss[loss=0.3153, simple_loss=0.3669, pruned_loss=0.1318, over 28758.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.4226, pruned_loss=0.1719, over 5671951.72 frames. ], libri_tot_loss[loss=0.3556, simple_loss=0.4121, pruned_loss=0.1495, over 5776786.27 frames. ], giga_tot_loss[loss=0.3876, simple_loss=0.4248, pruned_loss=0.1752, over 5653550.36 frames. ], batch size: 71, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:26:28,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7453, 3.8245, 4.4761, 1.9818], device='cuda:0'), covar=tensor([0.0420, 0.0434, 0.0717, 0.1664], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0539, 0.0829, 0.0552], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0006, 0.0010, 0.0007], device='cuda:0')
+2023-03-01 02:26:47,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.580e+03 2.072e+03 3.001e+03 5.283e+03, threshold=4.143e+03, percent-clipped=3.0
+2023-03-01 02:26:49,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:27:15,577 INFO [train.py:968] (0/2) Epoch 2, batch 12700, giga_loss[loss=0.415, simple_loss=0.4433, pruned_loss=0.1933, over 28582.00 frames. ], tot_loss[loss=0.384, simple_loss=0.4223, pruned_loss=0.1728, over 5662510.37 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4124, pruned_loss=0.1496, over 5778305.17 frames. ], giga_tot_loss[loss=0.3875, simple_loss=0.4238, pruned_loss=0.1755, over 5645685.58 frames. ], batch size: 336, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:27:23,392 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:27:25,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:27:54,921 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:27:57,875 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:28:01,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:28:09,294 INFO [train.py:968] (0/2) Epoch 2, batch 12750, giga_loss[loss=0.3518, simple_loss=0.4001, pruned_loss=0.1518, over 28936.00 frames. ], tot_loss[loss=0.3844, simple_loss=0.4221, pruned_loss=0.1734, over 5657711.23 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.4127, pruned_loss=0.1497, over 5779521.41 frames. ], giga_tot_loss[loss=0.3874, simple_loss=0.4232, pruned_loss=0.1758, over 5641493.44 frames. ], batch size: 199, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:28:32,099 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:28:33,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.308e+02 1.683e+03 2.153e+03 3.035e+03 6.123e+03, threshold=4.307e+03, percent-clipped=6.0
+2023-03-01 02:28:52,173 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-01 02:29:00,038 INFO [train.py:968] (0/2) Epoch 2, batch 12800, giga_loss[loss=0.3446, simple_loss=0.404, pruned_loss=0.1426, over 27990.00 frames. ], tot_loss[loss=0.3818, simple_loss=0.4207, pruned_loss=0.1714, over 5656961.96 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4125, pruned_loss=0.1496, over 5781114.81 frames. ], giga_tot_loss[loss=0.3847, simple_loss=0.4219, pruned_loss=0.1737, over 5641033.02 frames. ], batch size: 412, lr: 1.43e-02, grad_scale: 8.0
+2023-03-01 02:29:09,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8533, 1.5882, 3.8955, 3.2501], device='cuda:0'), covar=tensor([0.1485, 0.1481, 0.0292, 0.0512], device='cuda:0'), in_proj_covar=tensor([0.0496, 0.0492, 0.0647, 0.0513], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 02:29:37,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:29:55,476 INFO [train.py:968] (0/2) Epoch 2, batch 12850, giga_loss[loss=0.4464, simple_loss=0.4667, pruned_loss=0.2131, over 28246.00 frames. ], tot_loss[loss=0.3737, simple_loss=0.4164, pruned_loss=0.1655, over 5661033.00 frames. ], libri_tot_loss[loss=0.3554, simple_loss=0.4121, pruned_loss=0.1493, over 5783690.75 frames. ], giga_tot_loss[loss=0.3772, simple_loss=0.4179, pruned_loss=0.1682, over 5642336.32 frames. ], batch size: 368, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:30:13,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=58534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:30:16,935 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=58537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:30:20,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.898e+02 1.401e+03 1.851e+03 2.396e+03 7.770e+03, threshold=3.702e+03, percent-clipped=6.0
+2023-03-01 02:30:43,650 INFO [train.py:968] (0/2) Epoch 2, batch 12900, giga_loss[loss=0.336, simple_loss=0.3932, pruned_loss=0.1394, over 28311.00 frames. ], tot_loss[loss=0.3662, simple_loss=0.4115, pruned_loss=0.1605, over 5661633.95 frames. ], libri_tot_loss[loss=0.3542, simple_loss=0.4108, pruned_loss=0.1488, over 5784747.08 frames. ], giga_tot_loss[loss=0.3709, simple_loss=0.4141, pruned_loss=0.1638, over 5639554.60 frames. ], batch size: 368, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:31:38,091 INFO [train.py:968] (0/2) Epoch 2, batch 12950, giga_loss[loss=0.314, simple_loss=0.3774, pruned_loss=0.1253, over 28534.00 frames. ], tot_loss[loss=0.3585, simple_loss=0.4064, pruned_loss=0.1553, over 5660795.70 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4101, pruned_loss=0.1486, over 5786152.60 frames. ], giga_tot_loss[loss=0.3629, simple_loss=0.409, pruned_loss=0.1583, over 5639894.94 frames. ], batch size: 336, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:32:05,635 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 02:32:07,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.754e+02 1.309e+03 1.693e+03 2.511e+03 6.294e+03, threshold=3.387e+03, percent-clipped=11.0
+2023-03-01 02:32:12,202 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:32:12,318 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-01 02:32:14,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:32:35,013 INFO [train.py:968] (0/2) Epoch 2, batch 13000, giga_loss[loss=0.3094, simple_loss=0.3725, pruned_loss=0.1232, over 28652.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.4011, pruned_loss=0.1507, over 5649124.27 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.4095, pruned_loss=0.1485, over 5781526.78 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.4036, pruned_loss=0.1533, over 5633048.96 frames. ], batch size: 66, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:32:37,694 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3260, 1.4215, 1.3574, 0.7031], device='cuda:0'), covar=tensor([0.0434, 0.0313, 0.0231, 0.0403], device='cuda:0'), in_proj_covar=tensor([0.1005, 0.0708, 0.0775, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 02:32:41,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:33:22,774 INFO [train.py:968] (0/2) Epoch 2, batch 13050, giga_loss[loss=0.3574, simple_loss=0.4199, pruned_loss=0.1475, over 28550.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3999, pruned_loss=0.147, over 5663573.21 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.4093, pruned_loss=0.1485, over 5784613.14 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.4018, pruned_loss=0.149, over 5643103.96 frames. ], batch size: 307, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:33:37,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1407, 3.4489, 3.8624, 1.8010], device='cuda:0'), covar=tensor([0.0443, 0.0441, 0.0847, 0.1752], device='cuda:0'), in_proj_covar=tensor([0.0767, 0.0541, 0.0802, 0.0544], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 02:33:48,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.942e+02 1.376e+03 1.804e+03 2.434e+03 1.127e+04, threshold=3.609e+03, percent-clipped=13.0
+2023-03-01 02:34:19,208 INFO [train.py:968] (0/2) Epoch 2, batch 13100, giga_loss[loss=0.3326, simple_loss=0.4028, pruned_loss=0.1312, over 29054.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3994, pruned_loss=0.1456, over 5662846.46 frames. ], libri_tot_loss[loss=0.3528, simple_loss=0.4088, pruned_loss=0.1484, over 5786644.02 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.401, pruned_loss=0.1473, over 5641571.39 frames. ], batch size: 155, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:34:30,075 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 02:35:12,644 INFO [train.py:968] (0/2) Epoch 2, batch 13150, giga_loss[loss=0.3968, simple_loss=0.422, pruned_loss=0.1858, over 26786.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3982, pruned_loss=0.1447, over 5662113.44 frames. ], libri_tot_loss[loss=0.3523, simple_loss=0.4083, pruned_loss=0.1481, over 5789021.98 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.3998, pruned_loss=0.1462, over 5640588.00 frames. ], batch size: 555, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:35:36,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.374e+02 1.377e+03 1.765e+03 2.130e+03 7.296e+03, threshold=3.530e+03, percent-clipped=3.0
+2023-03-01 02:35:53,302 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-01 02:36:07,859 INFO [train.py:968] (0/2) Epoch 2, batch 13200, giga_loss[loss=0.2958, simple_loss=0.3616, pruned_loss=0.115, over 28884.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3942, pruned_loss=0.1414, over 5660452.34 frames. ], libri_tot_loss[loss=0.3522, simple_loss=0.4082, pruned_loss=0.1481, over 5788713.75 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3954, pruned_loss=0.1425, over 5642154.15 frames. ], batch size: 199, lr: 1.43e-02, grad_scale: 8.0
+2023-03-01 02:36:50,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:36:53,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:37:00,106 INFO [train.py:968] (0/2) Epoch 2, batch 13250, libri_loss[loss=0.3091, simple_loss=0.3681, pruned_loss=0.125, over 29533.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3926, pruned_loss=0.1407, over 5657091.30 frames. ], libri_tot_loss[loss=0.3517, simple_loss=0.4078, pruned_loss=0.1479, over 5791604.83 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3936, pruned_loss=0.1417, over 5636564.85 frames. ], batch size: 80, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:37:11,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3525, 1.8738, 1.2676, 1.4833], device='cuda:0'), covar=tensor([0.0956, 0.0344, 0.0482, 0.1072], device='cuda:0'), in_proj_covar=tensor([0.0322, 0.0214, 0.0220, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0022, 0.0020, 0.0033], device='cuda:0')
+2023-03-01 02:37:23,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.630e+02 1.515e+03 2.015e+03 2.866e+03 7.774e+03, threshold=4.031e+03, percent-clipped=15.0
+2023-03-01 02:37:47,025 INFO [train.py:968] (0/2) Epoch 2, batch 13300, libri_loss[loss=0.317, simple_loss=0.362, pruned_loss=0.136, over 29363.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3927, pruned_loss=0.1408, over 5655941.57 frames. ], libri_tot_loss[loss=0.3503, simple_loss=0.4064, pruned_loss=0.1471, over 5791075.56 frames. ], giga_tot_loss[loss=0.3391, simple_loss=0.3942, pruned_loss=0.142, over 5632700.84 frames. ], batch size: 67, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:38:37,214 INFO [train.py:968] (0/2) Epoch 2, batch 13350, giga_loss[loss=0.3103, simple_loss=0.3654, pruned_loss=0.1276, over 28765.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3902, pruned_loss=0.1386, over 5651552.37 frames. ], libri_tot_loss[loss=0.3499, simple_loss=0.4059, pruned_loss=0.1469, over 5780396.45 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3915, pruned_loss=0.1396, over 5638933.14 frames. ], batch size: 99, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:38:59,825 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.220e+02 1.300e+03 1.640e+03 2.399e+03 4.589e+03, threshold=3.281e+03, percent-clipped=1.0
+2023-03-01 02:39:08,036 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:10,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:10,173 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:11,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4472, 1.9800, 1.4955, 0.4650], device='cuda:0'), covar=tensor([0.1339, 0.0847, 0.1212, 0.1866], device='cuda:0'), in_proj_covar=tensor([0.1130, 0.1124, 0.1156, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 02:39:15,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:26,725 INFO [train.py:968] (0/2) Epoch 2, batch 13400, giga_loss[loss=0.3422, simple_loss=0.4069, pruned_loss=0.1387, over 28560.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.387, pruned_loss=0.1357, over 5655543.59 frames. ], libri_tot_loss[loss=0.349, simple_loss=0.405, pruned_loss=0.1465, over 5784460.82 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3881, pruned_loss=0.1364, over 5636084.27 frames. ], batch size: 307, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:39:32,225 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=59074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:42,390 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:45,668 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:40:21,346 INFO [train.py:968] (0/2) Epoch 2, batch 13450, giga_loss[loss=0.2881, simple_loss=0.3546, pruned_loss=0.1108, over 28647.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3828, pruned_loss=0.1322, over 5645265.90 frames. ], libri_tot_loss[loss=0.3489, simple_loss=0.4048, pruned_loss=0.1465, over 5775766.95 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3834, pruned_loss=0.1326, over 5635888.92 frames. ], batch size: 242, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:40:47,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.668e+02 1.490e+03 1.793e+03 2.442e+03 6.737e+03, threshold=3.586e+03, percent-clipped=11.0
+2023-03-01 02:41:00,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4721, 1.5220, 1.3609, 0.9271], device='cuda:0'), covar=tensor([0.0491, 0.0339, 0.0233, 0.0424], device='cuda:0'), in_proj_covar=tensor([0.1013, 0.0705, 0.0770, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 02:41:12,600 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=59168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:41:13,943 INFO [train.py:968] (0/2) Epoch 2, batch 13500, giga_loss[loss=0.3311, simple_loss=0.3805, pruned_loss=0.1408, over 28596.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3796, pruned_loss=0.1305, over 5647964.43 frames. ], libri_tot_loss[loss=0.3487, simple_loss=0.4046, pruned_loss=0.1464, over 5769192.34 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3796, pruned_loss=0.1304, over 5642540.24 frames. ], batch size: 92, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:41:30,348 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-01 02:42:10,075 INFO [train.py:968] (0/2) Epoch 2, batch 13550, giga_loss[loss=0.323, simple_loss=0.3807, pruned_loss=0.1327, over 28728.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3789, pruned_loss=0.1307, over 5653293.54 frames. ], libri_tot_loss[loss=0.3483, simple_loss=0.4043, pruned_loss=0.1462, over 5770297.94 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3789, pruned_loss=0.1306, over 5646962.75 frames. ], batch size: 284, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:42:38,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.171e+02 1.327e+03 1.756e+03 2.446e+03 6.556e+03, threshold=3.512e+03, percent-clipped=12.0
+2023-03-01 02:43:13,736 INFO [train.py:968] (0/2) Epoch 2, batch 13600, giga_loss[loss=0.365, simple_loss=0.4198, pruned_loss=0.1551, over 28278.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3795, pruned_loss=0.1314, over 5645782.39 frames. ], libri_tot_loss[loss=0.3479, simple_loss=0.4039, pruned_loss=0.146, over 5771825.46 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3797, pruned_loss=0.1314, over 5638268.63 frames. ], batch size: 368, lr: 1.42e-02, grad_scale: 8.0
+2023-03-01 02:43:42,573 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=59299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:43:46,372 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0188, 3.3336, 1.6992, 1.5922], device='cuda:0'), covar=tensor([0.0805, 0.0480, 0.0922, 0.1429], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0442, 0.0344, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 02:44:09,489 INFO [train.py:968] (0/2) Epoch 2, batch 13650, giga_loss[loss=0.284, simple_loss=0.338, pruned_loss=0.1149, over 24351.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3816, pruned_loss=0.1317, over 5645617.57 frames. ], libri_tot_loss[loss=0.346, simple_loss=0.4022, pruned_loss=0.1449, over 5774980.29 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3824, pruned_loss=0.1322, over 5633140.39 frames. ], batch size: 705, lr: 1.42e-02, grad_scale: 8.0
+2023-03-01 02:44:42,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.232e+02 1.533e+03 1.920e+03 2.803e+03 6.021e+03, threshold=3.839e+03, percent-clipped=10.0
+2023-03-01 02:45:11,385 INFO [train.py:968] (0/2) Epoch 2, batch 13700, libri_loss[loss=0.323, simple_loss=0.3875, pruned_loss=0.1292, over 25667.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3821, pruned_loss=0.1307, over 5658563.37 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.4015, pruned_loss=0.1445, over 5771154.83 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.383, pruned_loss=0.1311, over 5649636.15 frames. ], batch size: 136, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:45:23,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3791, 2.5950, 5.0425, 3.8437], device='cuda:0'), covar=tensor([0.1371, 0.1122, 0.0262, 0.0368], device='cuda:0'), in_proj_covar=tensor([0.0495, 0.0490, 0.0626, 0.0493], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 02:46:12,030 INFO [train.py:968] (0/2) Epoch 2, batch 13750, giga_loss[loss=0.3032, simple_loss=0.3675, pruned_loss=0.1195, over 28114.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3819, pruned_loss=0.1302, over 5672723.75 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.4003, pruned_loss=0.1436, over 5773985.18 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3832, pruned_loss=0.131, over 5659447.21 frames. ], batch size: 412, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:46:48,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.816e+02 1.397e+03 1.659e+03 2.184e+03 4.069e+03, threshold=3.318e+03, percent-clipped=1.0
+2023-03-01 02:46:54,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:47:07,764 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-01 02:47:22,864 INFO [train.py:968] (0/2) Epoch 2, batch 13800, giga_loss[loss=0.2807, simple_loss=0.3578, pruned_loss=0.1018, over 28971.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3788, pruned_loss=0.1277, over 5669940.18 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.4003, pruned_loss=0.1436, over 5773985.18 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3798, pruned_loss=0.1283, over 5659606.89 frames. ], batch size: 155, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:47:32,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5660, 2.0136, 1.5765, 1.6894], device='cuda:0'), covar=tensor([0.1501, 0.1508, 0.1301, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0786, 0.0699, 0.0600], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 02:48:23,437 INFO [train.py:968] (0/2) Epoch 2, batch 13850, giga_loss[loss=0.2837, simple_loss=0.3533, pruned_loss=0.1071, over 28458.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3769, pruned_loss=0.1246, over 5672248.03 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.4001, pruned_loss=0.1436, over 5777117.73 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3774, pruned_loss=0.1247, over 5658934.17 frames. ], batch size: 336, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:48:49,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7729, 1.7555, 1.5542, 1.6306], device='cuda:0'), covar=tensor([0.0864, 0.1661, 0.1322, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0809, 0.0630, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 02:48:53,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:48:53,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.649e+02 1.247e+03 1.671e+03 2.267e+03 4.452e+03, threshold=3.341e+03, percent-clipped=7.0
+2023-03-01 02:49:14,875 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=59557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:49:29,821 INFO [train.py:968] (0/2) Epoch 2, batch 13900, giga_loss[loss=0.3548, simple_loss=0.3791, pruned_loss=0.1652, over 26684.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3756, pruned_loss=0.125, over 5664169.37 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.4, pruned_loss=0.1436, over 5777678.27 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3754, pruned_loss=0.1246, over 5650108.94 frames. ], batch size: 555, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:49:46,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8921, 3.1932, 3.6315, 1.5551], device='cuda:0'), covar=tensor([0.0583, 0.0501, 0.0996, 0.1904], device='cuda:0'), in_proj_covar=tensor([0.0736, 0.0528, 0.0766, 0.0523], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 02:50:00,276 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:50:04,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:50:34,716 INFO [train.py:968] (0/2) Epoch 2, batch 13950, giga_loss[loss=0.297, simple_loss=0.3616, pruned_loss=0.1162, over 28930.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3741, pruned_loss=0.1249, over 5673255.49 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.3998, pruned_loss=0.1434, over 5777542.47 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3738, pruned_loss=0.1245, over 5660771.68 frames. ], batch size: 145, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:50:40,633 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:51:05,452 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.353e+02 1.471e+03 1.859e+03 2.516e+03 9.628e+03, threshold=3.719e+03, percent-clipped=11.0
+2023-03-01 02:51:31,365 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-01 02:51:32,233 INFO [train.py:968] (0/2) Epoch 2, batch 14000, giga_loss[loss=0.3567, simple_loss=0.4051, pruned_loss=0.1542, over 27691.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3719, pruned_loss=0.1239, over 5667572.09 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.399, pruned_loss=0.143, over 5767915.34 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3717, pruned_loss=0.1233, over 5662816.03 frames. ], batch size: 472, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:51:40,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:51:52,600 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:51:55,598 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:52:13,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9086, 1.6942, 3.9361, 3.2417], device='cuda:0'), covar=tensor([0.1397, 0.1401, 0.0278, 0.0882], device='cuda:0'), in_proj_covar=tensor([0.0492, 0.0488, 0.0606, 0.0489], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 02:52:30,997 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:52:31,368 INFO [train.py:968] (0/2) Epoch 2, batch 14050, giga_loss[loss=0.3197, simple_loss=0.3841, pruned_loss=0.1276, over 28863.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3729, pruned_loss=0.1241, over 5662191.39 frames. ], libri_tot_loss[loss=0.3414, simple_loss=0.3981, pruned_loss=0.1424, over 5772107.13 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3727, pruned_loss=0.1236, over 5650965.22 frames. ], batch size: 227, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:52:57,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7162, 1.5454, 1.0880, 1.2730], device='cuda:0'), covar=tensor([0.0685, 0.0685, 0.1083, 0.1039], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0527, 0.0566, 0.0493], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 02:53:05,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.978e+02 1.384e+03 1.859e+03 2.724e+03 1.142e+04, threshold=3.719e+03, percent-clipped=9.0
+2023-03-01 02:53:36,571 INFO [train.py:968] (0/2) Epoch 2, batch 14100, giga_loss[loss=0.2902, simple_loss=0.3598, pruned_loss=0.1102, over 28912.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3755, pruned_loss=0.1254, over 5666746.63 frames. ], libri_tot_loss[loss=0.3411, simple_loss=0.3975, pruned_loss=0.1423, over 5775488.20 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3752, pruned_loss=0.1247, over 5652199.15 frames. ], batch size: 186, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:54:04,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8816, 1.6122, 1.3402, 1.3164], device='cuda:0'), covar=tensor([0.0707, 0.0681, 0.0976, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0519, 0.0560, 0.0486], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 02:54:41,986 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:54:42,911 INFO [train.py:968] (0/2) Epoch 2, batch 14150, giga_loss[loss=0.3171, simple_loss=0.3791, pruned_loss=0.1275, over 28729.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3714, pruned_loss=0.1226, over 5675328.93 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3965, pruned_loss=0.1417, over 5779455.41 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3713, pruned_loss=0.122, over 5656684.62 frames. ], batch size: 307, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:54:44,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:55:20,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.707e+02 1.608e+03 2.069e+03 2.842e+03 6.905e+03, threshold=4.138e+03, percent-clipped=11.0
+2023-03-01 02:55:26,195 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:55:50,629 INFO [train.py:968] (0/2) Epoch 2, batch 14200, giga_loss[loss=0.2606, simple_loss=0.3358, pruned_loss=0.09265, over 28499.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3726, pruned_loss=0.1237, over 5687769.60 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3958, pruned_loss=0.1412, over 5782299.54 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3725, pruned_loss=0.1231, over 5667661.65 frames. ], batch size: 71, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 02:56:25,946 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=59895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:56:45,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2564, 3.5749, 4.0299, 1.7602], device='cuda:0'), covar=tensor([0.0390, 0.0381, 0.0593, 0.1746], device='cuda:0'), in_proj_covar=tensor([0.0739, 0.0528, 0.0763, 0.0530], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 02:57:01,565 INFO [train.py:968] (0/2) Epoch 2, batch 14250, giga_loss[loss=0.3494, simple_loss=0.4175, pruned_loss=0.1407, over 28687.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3761, pruned_loss=0.1251, over 5688367.67 frames. ], libri_tot_loss[loss=0.3389, simple_loss=0.3956, pruned_loss=0.1411, over 5784274.74 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3757, pruned_loss=0.1244, over 5668352.42 frames. ], batch size: 262, lr: 1.41e-02, grad_scale: 2.0
+2023-03-01 02:57:17,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:57:36,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.064e+02 1.469e+03 2.031e+03 3.108e+03 1.445e+04, threshold=4.062e+03, percent-clipped=13.0
+2023-03-01 02:58:06,370 INFO [train.py:968] (0/2) Epoch 2, batch 14300, giga_loss[loss=0.2892, simple_loss=0.3723, pruned_loss=0.1031, over 28938.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3786, pruned_loss=0.1237, over 5682240.09 frames. ], libri_tot_loss[loss=0.3383, simple_loss=0.3952, pruned_loss=0.1407, over 5783301.17 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3783, pruned_loss=0.1231, over 5664563.84 frames. ], batch size: 120, lr: 1.41e-02, grad_scale: 2.0
+2023-03-01 02:58:44,410 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-60000.pt
+2023-03-01 02:59:08,821 INFO [train.py:968] (0/2) Epoch 2, batch 14350, giga_loss[loss=0.2975, simple_loss=0.3844, pruned_loss=0.1053, over 28824.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3787, pruned_loss=0.1221, over 5678813.42 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3946, pruned_loss=0.1403, over 5785689.53 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3787, pruned_loss=0.1217, over 5660901.54 frames. ], batch size: 174, lr: 1.41e-02, grad_scale: 2.0
+2023-03-01 02:59:10,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4063, 1.3608, 1.2634, 1.6232], device='cuda:0'), covar=tensor([0.1786, 0.1622, 0.1413, 0.2031], device='cuda:0'), in_proj_covar=tensor([0.0954, 0.0784, 0.0890, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 02:59:42,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.187e+02 1.383e+03 1.680e+03 2.089e+03 4.688e+03, threshold=3.361e+03, percent-clipped=2.0
+2023-03-01 03:00:12,258 INFO [train.py:968] (0/2) Epoch 2, batch 14400, giga_loss[loss=0.3007, simple_loss=0.3711, pruned_loss=0.1151, over 29017.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3778, pruned_loss=0.1209, over 5683184.03 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3943, pruned_loss=0.1401, over 5787063.04 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.378, pruned_loss=0.1205, over 5667063.57 frames. ], batch size: 199, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:00:21,168 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:00:25,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:01:02,767 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:01:17,500 INFO [train.py:968] (0/2) Epoch 2, batch 14450, giga_loss[loss=0.3053, simple_loss=0.3728, pruned_loss=0.1189, over 28766.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3791, pruned_loss=0.1227, over 5680131.43 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3942, pruned_loss=0.1401, over 5787535.75 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.379, pruned_loss=0.1221, over 5665154.91 frames. ], batch size: 174, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:01:53,550 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.932e+02 1.306e+03 1.624e+03 2.068e+03 5.190e+03, threshold=3.248e+03, percent-clipped=6.0
+2023-03-01 03:02:25,680 INFO [train.py:968] (0/2) Epoch 2, batch 14500, libri_loss[loss=0.3163, simple_loss=0.381, pruned_loss=0.1258, over 29271.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3786, pruned_loss=0.1237, over 5689118.92 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3939, pruned_loss=0.14, over 5789909.50 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3785, pruned_loss=0.123, over 5673237.17 frames. ], batch size: 94, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:03:36,480 INFO [train.py:968] (0/2) Epoch 2, batch 14550, libri_loss[loss=0.3091, simple_loss=0.3627, pruned_loss=0.1278, over 29578.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3798, pruned_loss=0.1256, over 5699629.15 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3926, pruned_loss=0.1393, over 5794509.04 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3804, pruned_loss=0.1252, over 5678961.98 frames. ], batch size: 78, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:04:16,517 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:04:26,054 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.391e+02 1.390e+03 1.917e+03 3.257e+03 9.488e+03, threshold=3.834e+03, percent-clipped=25.0
+2023-03-01 03:04:30,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 03:05:02,461 INFO [train.py:968] (0/2) Epoch 2, batch 14600, giga_loss[loss=0.3055, simple_loss=0.375, pruned_loss=0.118, over 29058.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.375, pruned_loss=0.123, over 5690570.92 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.392, pruned_loss=0.1389, over 5794973.61 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3757, pruned_loss=0.1228, over 5672103.19 frames. ], batch size: 155, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:05:06,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:05:23,434 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:06:09,437 INFO [train.py:968] (0/2) Epoch 2, batch 14650, giga_loss[loss=0.2946, simple_loss=0.372, pruned_loss=0.1086, over 28663.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3734, pruned_loss=0.1216, over 5693604.56 frames. ], libri_tot_loss[loss=0.3342, simple_loss=0.3914, pruned_loss=0.1385, over 5797893.43 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3742, pruned_loss=0.1214, over 5673940.75 frames. ], batch size: 307, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:06:51,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.812e+02 1.210e+03 1.594e+03 2.170e+03 4.330e+03, threshold=3.187e+03, percent-clipped=2.0
+2023-03-01 03:07:21,453 INFO [train.py:968] (0/2) Epoch 2, batch 14700, giga_loss[loss=0.3612, simple_loss=0.4139, pruned_loss=0.1542, over 28893.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3724, pruned_loss=0.1219, over 5691023.45 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3912, pruned_loss=0.1383, over 5800092.24 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3729, pruned_loss=0.1215, over 5671427.58 frames. ], batch size: 186, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:08:15,179 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:08:20,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:08:22,193 INFO [train.py:968] (0/2) Epoch 2, batch 14750, libri_loss[loss=0.2866, simple_loss=0.3566, pruned_loss=0.1083, over 29581.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3769, pruned_loss=0.125, over 5689607.87 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3901, pruned_loss=0.1377, over 5804207.96 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3777, pruned_loss=0.1248, over 5665906.62 frames. ], batch size: 76, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:08:39,817 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:09:02,214 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:09:02,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:09:02,550 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.482e+02 1.576e+03 1.963e+03 2.726e+03 5.722e+03, threshold=3.926e+03, percent-clipped=12.0
+2023-03-01 03:09:18,439 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:09:33,326 INFO [train.py:968] (0/2) Epoch 2, batch 14800, giga_loss[loss=0.4138, simple_loss=0.4285, pruned_loss=0.1996, over 26962.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3789, pruned_loss=0.1268, over 5695244.90 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3898, pruned_loss=0.1377, over 5806292.85 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3795, pruned_loss=0.1264, over 5671941.30 frames. ], batch size: 555, lr: 1.41e-02, grad_scale: 8.0
+2023-03-01 03:09:38,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:09:42,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4967, 2.0216, 1.5118, 1.3576], device='cuda:0'), covar=tensor([0.1134, 0.0410, 0.0508, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0317, 0.0211, 0.0220, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0022, 0.0020, 0.0033], device='cuda:0')
+2023-03-01 03:10:24,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3928, 2.1862, 1.3844, 1.2143], device='cuda:0'), covar=tensor([0.0857, 0.0577, 0.0934, 0.1458], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0438, 0.0343, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 03:10:36,063 INFO [train.py:968] (0/2) Epoch 2, batch 14850, giga_loss[loss=0.3236, simple_loss=0.3882, pruned_loss=0.1295, over 28883.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3778, pruned_loss=0.1273, over 5684468.97 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3897, pruned_loss=0.1376, over 5795911.68 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3782, pruned_loss=0.1268, over 5672037.01 frames. ], batch size: 284, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:10:49,816 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 03:11:06,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1985, 1.2791, 1.1730, 1.0197], device='cuda:0'), covar=tensor([0.1472, 0.1278, 0.1096, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0760, 0.0865, 0.0909], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 03:11:14,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.480e+03 1.865e+03 2.569e+03 8.058e+03, threshold=3.731e+03, percent-clipped=7.0
+2023-03-01 03:11:29,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7237, 4.0502, 4.4250, 1.8104], device='cuda:0'), covar=tensor([0.0422, 0.0402, 0.0878, 0.1799], device='cuda:0'), in_proj_covar=tensor([0.0718, 0.0524, 0.0749, 0.0528], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 03:11:38,687 INFO [train.py:968] (0/2) Epoch 2, batch 14900, giga_loss[loss=0.3288, simple_loss=0.3833, pruned_loss=0.1372, over 28905.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3787, pruned_loss=0.1283, over 5692338.52 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3894, pruned_loss=0.1373, over 5798742.00 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.379, pruned_loss=0.1278, over 5676853.31 frames. ], batch size: 186, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:11:54,061 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 03:11:54,095 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-01 03:12:12,508 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:12:49,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:12:52,101 INFO [train.py:968] (0/2) Epoch 2, batch 14950, giga_loss[loss=0.3156, simple_loss=0.3817, pruned_loss=0.1247, over 28639.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3799, pruned_loss=0.1283, over 5686730.04 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3892, pruned_loss=0.1372, over 5798719.08 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3802, pruned_loss=0.1279, over 5672977.21 frames. ], batch size: 242, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:13:05,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4007, 1.4479, 2.8875, 2.6189], device='cuda:0'), covar=tensor([0.1432, 0.1361, 0.0449, 0.0545], device='cuda:0'), in_proj_covar=tensor([0.0494, 0.0488, 0.0620, 0.0482], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 03:13:28,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.757e+02 1.493e+03 1.845e+03 2.266e+03 5.641e+03, threshold=3.689e+03, percent-clipped=6.0
+2023-03-01 03:13:45,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:14:08,119 INFO [train.py:968] (0/2) Epoch 2, batch 15000, giga_loss[loss=0.3438, simple_loss=0.3894, pruned_loss=0.1491, over 26864.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3811, pruned_loss=0.1281, over 5681215.28 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.389, pruned_loss=0.1372, over 5800681.05 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3812, pruned_loss=0.1276, over 5666231.81 frames. ], batch size: 555, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:14:08,124 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 03:14:15,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2525, 1.3724, 1.2272, 1.1834], device='cuda:0'), covar=tensor([0.2150, 0.1917, 0.1747, 0.1836], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0759, 0.0867, 0.0917], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 03:14:17,216 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.2578, simple_loss=0.3473, pruned_loss=0.08416, over 944034.00 frames. 
+2023-03-01 03:14:17,217 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19478MB
+2023-03-01 03:14:49,635 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.31 vs. limit=5.0
+2023-03-01 03:15:38,804 INFO [train.py:968] (0/2) Epoch 2, batch 15050, giga_loss[loss=0.2725, simple_loss=0.349, pruned_loss=0.09801, over 28508.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3794, pruned_loss=0.127, over 5675422.27 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3888, pruned_loss=0.1371, over 5802200.79 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3797, pruned_loss=0.1266, over 5660137.29 frames. ], batch size: 85, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:16:20,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.675e+02 1.528e+03 1.951e+03 2.663e+03 1.023e+04, threshold=3.901e+03, percent-clipped=8.0
+2023-03-01 03:16:37,542 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:16:40,903 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:16:49,212 INFO [train.py:968] (0/2) Epoch 2, batch 15100, giga_loss[loss=0.2529, simple_loss=0.3259, pruned_loss=0.08997, over 28908.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3754, pruned_loss=0.1259, over 5679495.93 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3889, pruned_loss=0.1371, over 5806052.89 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3752, pruned_loss=0.1252, over 5659559.57 frames. ], batch size: 213, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:17:19,627 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:17:31,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:17:34,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:17:42,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:17:54,096 INFO [train.py:968] (0/2) Epoch 2, batch 15150, giga_loss[loss=0.3072, simple_loss=0.364, pruned_loss=0.1252, over 28855.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3685, pruned_loss=0.1223, over 5671947.04 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3887, pruned_loss=0.1371, over 5798462.44 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3679, pruned_loss=0.1214, over 5658704.01 frames. ], batch size: 164, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:17:55,568 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:18:10,736 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:18:11,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:18:27,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:18:28,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.240e+02 1.511e+03 2.106e+03 2.952e+03 2.049e+04, threshold=4.212e+03, percent-clipped=15.0
+2023-03-01 03:18:53,598 INFO [train.py:968] (0/2) Epoch 2, batch 15200, giga_loss[loss=0.2909, simple_loss=0.3584, pruned_loss=0.1117, over 29019.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3698, pruned_loss=0.1237, over 5672432.10 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3883, pruned_loss=0.137, over 5793629.88 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3688, pruned_loss=0.1225, over 5661012.19 frames. ], batch size: 213, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:19:51,927 INFO [train.py:968] (0/2) Epoch 2, batch 15250, giga_loss[loss=0.3289, simple_loss=0.3805, pruned_loss=0.1386, over 28014.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3726, pruned_loss=0.1264, over 5655394.29 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3886, pruned_loss=0.1373, over 5780580.45 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3712, pruned_loss=0.1249, over 5654163.64 frames. ], batch size: 412, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:20:31,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.238e+02 1.530e+03 1.962e+03 2.616e+03 1.323e+04, threshold=3.925e+03, percent-clipped=6.0
+2023-03-01 03:20:36,723 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:20:41,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:20:52,396 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:20:55,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:20:55,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:20:57,573 INFO [train.py:968] (0/2) Epoch 2, batch 15300, giga_loss[loss=0.3073, simple_loss=0.3711, pruned_loss=0.1218, over 28843.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3697, pruned_loss=0.1237, over 5660160.21 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3887, pruned_loss=0.1373, over 5781161.54 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3685, pruned_loss=0.1224, over 5658060.09 frames. ], batch size: 174, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:21:09,303 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:13,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:19,970 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:25,280 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-01 03:21:26,911 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:30,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:34,584 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:52,214 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:22:06,014 INFO [train.py:968] (0/2) Epoch 2, batch 15350, giga_loss[loss=0.332, simple_loss=0.3923, pruned_loss=0.1359, over 28927.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.368, pruned_loss=0.1213, over 5659340.31 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3883, pruned_loss=0.1371, over 5782495.68 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3671, pruned_loss=0.1204, over 5655537.30 frames. ], batch size: 199, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:22:09,097 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:22:45,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.697e+02 1.459e+03 1.818e+03 2.343e+03 3.632e+03, threshold=3.636e+03, percent-clipped=0.0
+2023-03-01 03:23:18,330 INFO [train.py:968] (0/2) Epoch 2, batch 15400, giga_loss[loss=0.2877, simple_loss=0.3572, pruned_loss=0.1091, over 28446.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3665, pruned_loss=0.1215, over 5659095.75 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3876, pruned_loss=0.1368, over 5786366.37 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3659, pruned_loss=0.1206, over 5649511.54 frames. ], batch size: 336, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:23:28,401 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-01 03:23:34,884 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-01 03:24:14,739 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:24:16,747 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=61111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:24:17,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:24:27,293 INFO [train.py:968] (0/2) Epoch 2, batch 15450, giga_loss[loss=0.405, simple_loss=0.4358, pruned_loss=0.1871, over 28065.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3666, pruned_loss=0.1209, over 5661613.12 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.3873, pruned_loss=0.1366, over 5789461.88 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3659, pruned_loss=0.12, over 5648404.98 frames. ], batch size: 412, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:24:28,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6131, 2.1558, 1.8599, 1.7399], device='cuda:0'), covar=tensor([0.1534, 0.1528, 0.1148, 0.0791], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0787, 0.0704, 0.0600], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 03:24:45,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.47 vs. limit=5.0
+2023-03-01 03:24:59,825 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:25:08,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.384e+02 1.307e+03 1.580e+03 2.329e+03 8.074e+03, threshold=3.159e+03, percent-clipped=9.0
+2023-03-01 03:25:36,217 INFO [train.py:968] (0/2) Epoch 2, batch 15500, giga_loss[loss=0.3613, simple_loss=0.4028, pruned_loss=0.1599, over 28020.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3667, pruned_loss=0.121, over 5654362.31 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3874, pruned_loss=0.1367, over 5781636.36 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3656, pruned_loss=0.1198, over 5648649.23 frames. ], batch size: 412, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:25:52,028 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7007, 1.7148, 3.7363, 2.9944], device='cuda:0'), covar=tensor([0.1448, 0.1316, 0.0279, 0.0482], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0483, 0.0613, 0.0486], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 03:25:55,072 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 03:26:01,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8378, 3.2208, 3.5373, 1.5893], device='cuda:0'), covar=tensor([0.0542, 0.0473, 0.0876, 0.1896], device='cuda:0'), in_proj_covar=tensor([0.0721, 0.0513, 0.0733, 0.0526], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 03:26:26,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6174, 1.4732, 3.8422, 2.9330], device='cuda:0'), covar=tensor([0.1588, 0.1588, 0.0305, 0.0482], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0485, 0.0612, 0.0485], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 03:26:44,801 INFO [train.py:968] (0/2) Epoch 2, batch 15550, giga_loss[loss=0.2631, simple_loss=0.3428, pruned_loss=0.09171, over 28934.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3673, pruned_loss=0.1219, over 5656647.19 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3868, pruned_loss=0.1364, over 5782186.92 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3663, pruned_loss=0.1208, over 5648096.12 frames. ], batch size: 164, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:27:14,239 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.134e+02 1.296e+03 1.722e+03 2.415e+03 8.797e+03, threshold=3.444e+03, percent-clipped=13.0
+2023-03-01 03:27:40,815 INFO [train.py:968] (0/2) Epoch 2, batch 15600, libri_loss[loss=0.3243, simple_loss=0.3873, pruned_loss=0.1307, over 27915.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3669, pruned_loss=0.121, over 5644470.29 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3868, pruned_loss=0.1363, over 5761157.72 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3654, pruned_loss=0.1196, over 5651724.09 frames. ], batch size: 116, lr: 1.40e-02, grad_scale: 8.0
+2023-03-01 03:28:42,164 INFO [train.py:968] (0/2) Epoch 2, batch 15650, giga_loss[loss=0.3196, simple_loss=0.3937, pruned_loss=0.1227, over 28914.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3687, pruned_loss=0.1198, over 5660662.31 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3866, pruned_loss=0.1363, over 5764043.97 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3673, pruned_loss=0.1184, over 5661861.05 frames. ], batch size: 186, lr: 1.40e-02, grad_scale: 8.0
+2023-03-01 03:29:19,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.381e+02 1.286e+03 1.646e+03 2.121e+03 5.073e+03, threshold=3.293e+03, percent-clipped=6.0
+2023-03-01 03:29:46,749 INFO [train.py:968] (0/2) Epoch 2, batch 15700, libri_loss[loss=0.3356, simple_loss=0.4053, pruned_loss=0.1329, over 29514.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.372, pruned_loss=0.1217, over 5659851.19 frames. ], libri_tot_loss[loss=0.3293, simple_loss=0.3864, pruned_loss=0.1361, over 5767607.46 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3707, pruned_loss=0.1203, over 5655042.02 frames. ], batch size: 81, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:30:07,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7937, 1.7647, 1.6238, 1.0228], device='cuda:0'), covar=tensor([0.0477, 0.0378, 0.0280, 0.0547], device='cuda:0'), in_proj_covar=tensor([0.0997, 0.0701, 0.0767, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 03:30:42,353 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-01 03:30:47,481 INFO [train.py:968] (0/2) Epoch 2, batch 15750, giga_loss[loss=0.3214, simple_loss=0.3792, pruned_loss=0.1318, over 27564.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3743, pruned_loss=0.1231, over 5663422.94 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3865, pruned_loss=0.1362, over 5768482.17 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3727, pruned_loss=0.1216, over 5656118.06 frames. ], batch size: 472, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:31:22,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.521e+02 1.431e+03 1.839e+03 2.718e+03 5.426e+03, threshold=3.677e+03, percent-clipped=17.0
+2023-03-01 03:31:28,492 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=61454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:31:46,485 INFO [train.py:968] (0/2) Epoch 2, batch 15800, giga_loss[loss=0.3035, simple_loss=0.3771, pruned_loss=0.115, over 28514.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3741, pruned_loss=0.1227, over 5668186.14 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3865, pruned_loss=0.1361, over 5758194.69 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3726, pruned_loss=0.1212, over 5668872.96 frames. ], batch size: 336, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:31:53,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8069, 1.7040, 1.4741, 1.2246], device='cuda:0'), covar=tensor([0.0472, 0.0433, 0.0329, 0.0435], device='cuda:0'), in_proj_covar=tensor([0.1002, 0.0707, 0.0773, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 03:32:04,666 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-01 03:32:09,583 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:32:48,751 INFO [train.py:968] (0/2) Epoch 2, batch 15850, giga_loss[loss=0.3456, simple_loss=0.4097, pruned_loss=0.1408, over 28682.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3716, pruned_loss=0.1206, over 5680734.43 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3865, pruned_loss=0.1363, over 5761210.97 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3702, pruned_loss=0.1191, over 5677051.29 frames. ], batch size: 307, lr: 1.40e-02, grad_scale: 2.0
+2023-03-01 03:32:54,429 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9705, 2.8158, 2.0320, 0.8270], device='cuda:0'), covar=tensor([0.2122, 0.1047, 0.1392, 0.2285], device='cuda:0'), in_proj_covar=tensor([0.1127, 0.1152, 0.1130, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 03:33:29,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.984e+02 1.319e+03 1.783e+03 2.334e+03 5.532e+03, threshold=3.566e+03, percent-clipped=5.0
+2023-03-01 03:33:54,010 INFO [train.py:968] (0/2) Epoch 2, batch 15900, giga_loss[loss=0.2836, simple_loss=0.3599, pruned_loss=0.1037, over 29043.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3694, pruned_loss=0.1192, over 5675403.99 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3859, pruned_loss=0.136, over 5754617.40 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3684, pruned_loss=0.1178, over 5675616.23 frames. ], batch size: 155, lr: 1.39e-02, grad_scale: 2.0
+2023-03-01 03:34:49,105 INFO [train.py:968] (0/2) Epoch 2, batch 15950, giga_loss[loss=0.2637, simple_loss=0.3347, pruned_loss=0.09638, over 28933.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3679, pruned_loss=0.1196, over 5679763.49 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3855, pruned_loss=0.1357, over 5758846.46 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3668, pruned_loss=0.118, over 5673120.87 frames. ], batch size: 145, lr: 1.39e-02, grad_scale: 2.0
+2023-03-01 03:35:02,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:35:04,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:35:26,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.327e+02 1.708e+03 2.292e+03 3.119e+03 7.018e+03, threshold=4.584e+03, percent-clipped=16.0
+2023-03-01 03:35:43,764 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:35:52,820 INFO [train.py:968] (0/2) Epoch 2, batch 16000, giga_loss[loss=0.3357, simple_loss=0.3874, pruned_loss=0.142, over 29060.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3699, pruned_loss=0.1209, over 5672332.60 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3857, pruned_loss=0.1357, over 5751873.01 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3686, pruned_loss=0.1193, over 5670982.26 frames. ], batch size: 128, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:37:02,556 INFO [train.py:968] (0/2) Epoch 2, batch 16050, giga_loss[loss=0.2974, simple_loss=0.366, pruned_loss=0.1144, over 28998.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3727, pruned_loss=0.1227, over 5675941.32 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3859, pruned_loss=0.136, over 5753067.15 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3712, pruned_loss=0.1212, over 5673045.95 frames. ], batch size: 120, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:37:44,497 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.817e+02 1.503e+03 1.995e+03 2.648e+03 8.384e+03, threshold=3.989e+03, percent-clipped=4.0
+2023-03-01 03:38:09,293 INFO [train.py:968] (0/2) Epoch 2, batch 16100, giga_loss[loss=0.3391, simple_loss=0.3956, pruned_loss=0.1413, over 28897.00 frames. ], tot_loss[loss=0.31, simple_loss=0.373, pruned_loss=0.1235, over 5674744.19 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3852, pruned_loss=0.1355, over 5757491.86 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3721, pruned_loss=0.1222, over 5666665.26 frames. ], batch size: 213, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:38:10,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3172, 1.6962, 1.2488, 1.4160], device='cuda:0'), covar=tensor([0.1063, 0.0455, 0.0531, 0.1261], device='cuda:0'), in_proj_covar=tensor([0.0307, 0.0202, 0.0213, 0.0369], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0022, 0.0020, 0.0033], device='cuda:0')
+2023-03-01 03:39:07,443 INFO [train.py:968] (0/2) Epoch 2, batch 16150, giga_loss[loss=0.3018, simple_loss=0.3606, pruned_loss=0.1215, over 28423.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3762, pruned_loss=0.125, over 5684303.16 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3847, pruned_loss=0.1352, over 5760341.31 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3756, pruned_loss=0.1241, over 5673096.55 frames. ], batch size: 71, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:39:20,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:39:41,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.144e+02 1.471e+03 1.949e+03 2.644e+03 5.811e+03, threshold=3.897e+03, percent-clipped=6.0
+2023-03-01 03:40:02,125 INFO [train.py:968] (0/2) Epoch 2, batch 16200, giga_loss[loss=0.2979, simple_loss=0.3718, pruned_loss=0.112, over 28973.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3775, pruned_loss=0.1251, over 5691760.64 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3834, pruned_loss=0.1345, over 5764711.12 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.378, pruned_loss=0.1246, over 5675889.25 frames. ], batch size: 186, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:41:06,889 INFO [train.py:968] (0/2) Epoch 2, batch 16250, giga_loss[loss=0.292, simple_loss=0.367, pruned_loss=0.1085, over 28977.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3788, pruned_loss=0.1256, over 5687588.12 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3833, pruned_loss=0.1344, over 5762409.33 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3791, pruned_loss=0.1252, over 5675732.89 frames. ], batch size: 155, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:41:50,908 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2720, 1.7648, 1.4814, 1.5394], device='cuda:0'), covar=tensor([0.1190, 0.1436, 0.1039, 0.0684], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0770, 0.0682, 0.0579], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 03:41:56,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.122e+02 1.404e+03 1.850e+03 2.385e+03 5.727e+03, threshold=3.700e+03, percent-clipped=9.0
+2023-03-01 03:42:24,053 INFO [train.py:968] (0/2) Epoch 2, batch 16300, giga_loss[loss=0.3024, simple_loss=0.3677, pruned_loss=0.1185, over 29062.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3764, pruned_loss=0.1241, over 5691732.56 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3834, pruned_loss=0.1345, over 5763553.23 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3766, pruned_loss=0.1236, over 5680769.48 frames. ], batch size: 200, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:42:28,880 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:42:33,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:43:05,326 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-62000.pt
+2023-03-01 03:43:14,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:43:25,658 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-01 03:43:34,061 INFO [train.py:968] (0/2) Epoch 2, batch 16350, giga_loss[loss=0.2936, simple_loss=0.3658, pruned_loss=0.1106, over 28852.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3758, pruned_loss=0.1241, over 5688229.56 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3834, pruned_loss=0.1345, over 5765042.12 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3759, pruned_loss=0.1236, over 5677652.22 frames. ], batch size: 243, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:43:57,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8728, 3.2482, 3.5644, 1.5292], device='cuda:0'), covar=tensor([0.0498, 0.0487, 0.0858, 0.2077], device='cuda:0'), in_proj_covar=tensor([0.0727, 0.0522, 0.0745, 0.0532], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 03:44:07,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9779, 1.7181, 3.2822, 2.9982], device='cuda:0'), covar=tensor([0.1170, 0.1250, 0.0369, 0.0520], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0490, 0.0611, 0.0479], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 03:44:14,912 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.253e+02 1.522e+03 2.141e+03 3.017e+03 7.456e+03, threshold=4.283e+03, percent-clipped=11.0
+2023-03-01 03:44:17,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5363, 2.2985, 1.4929, 1.3495], device='cuda:0'), covar=tensor([0.0925, 0.0542, 0.0924, 0.1433], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0436, 0.0338, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 03:44:38,106 INFO [train.py:968] (0/2) Epoch 2, batch 16400, giga_loss[loss=0.2756, simple_loss=0.3487, pruned_loss=0.1013, over 29030.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3742, pruned_loss=0.1235, over 5680299.04 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3837, pruned_loss=0.1347, over 5767943.27 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3737, pruned_loss=0.1226, over 5666855.71 frames. ], batch size: 136, lr: 1.39e-02, grad_scale: 8.0
+2023-03-01 03:45:41,092 INFO [train.py:968] (0/2) Epoch 2, batch 16450, giga_loss[loss=0.2985, simple_loss=0.3442, pruned_loss=0.1264, over 28419.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3728, pruned_loss=0.1241, over 5670245.13 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3839, pruned_loss=0.1348, over 5760205.86 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3721, pruned_loss=0.1231, over 5665094.23 frames. ], batch size: 71, lr: 1.39e-02, grad_scale: 8.0
+2023-03-01 03:45:47,037 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:46:02,036 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:46:16,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6704, 1.9683, 2.2124, 1.9898], device='cuda:0'), covar=tensor([0.0609, 0.1661, 0.0983, 0.1150], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0799, 0.0614, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 03:46:18,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.543e+02 1.349e+03 1.846e+03 2.536e+03 4.990e+03, threshold=3.691e+03, percent-clipped=2.0
+2023-03-01 03:46:36,709 INFO [train.py:968] (0/2) Epoch 2, batch 16500, giga_loss[loss=0.3274, simple_loss=0.3738, pruned_loss=0.1404, over 26661.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3735, pruned_loss=0.1249, over 5680566.05 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3835, pruned_loss=0.1346, over 5765323.00 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3729, pruned_loss=0.1239, over 5668430.97 frames. ], batch size: 555, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:47:20,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-01 03:47:35,715 INFO [train.py:968] (0/2) Epoch 2, batch 16550, giga_loss[loss=0.2952, simple_loss=0.3698, pruned_loss=0.1103, over 28981.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3727, pruned_loss=0.1233, over 5681514.06 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3828, pruned_loss=0.1341, over 5768735.38 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3726, pruned_loss=0.1226, over 5666166.42 frames. ], batch size: 136, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:48:16,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.456e+02 1.388e+03 1.799e+03 2.433e+03 5.702e+03, threshold=3.598e+03, percent-clipped=10.0
+2023-03-01 03:48:45,400 INFO [train.py:968] (0/2) Epoch 2, batch 16600, giga_loss[loss=0.2852, simple_loss=0.3692, pruned_loss=0.1006, over 28178.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3743, pruned_loss=0.1232, over 5684157.31 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3832, pruned_loss=0.1344, over 5771418.46 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3736, pruned_loss=0.1222, over 5668243.86 frames. ], batch size: 412, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:49:33,471 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:49:48,489 INFO [train.py:968] (0/2) Epoch 2, batch 16650, libri_loss[loss=0.3389, simple_loss=0.3736, pruned_loss=0.1521, over 29652.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3747, pruned_loss=0.1214, over 5678972.60 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.383, pruned_loss=0.1344, over 5773594.73 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3743, pruned_loss=0.1205, over 5663149.09 frames. ], batch size: 69, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:50:27,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.254e+02 1.364e+03 1.782e+03 2.414e+03 6.455e+03, threshold=3.564e+03, percent-clipped=9.0
+2023-03-01 03:50:52,370 INFO [train.py:968] (0/2) Epoch 2, batch 16700, giga_loss[loss=0.2894, simple_loss=0.3658, pruned_loss=0.1065, over 28931.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3752, pruned_loss=0.1204, over 5690344.04 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.383, pruned_loss=0.1344, over 5773594.73 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3748, pruned_loss=0.1197, over 5678028.41 frames. ], batch size: 145, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:51:33,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3614, 1.5894, 1.1540, 1.3839], device='cuda:0'), covar=tensor([0.1035, 0.0423, 0.0495, 0.1208], device='cuda:0'), in_proj_covar=tensor([0.0308, 0.0204, 0.0211, 0.0368], device='cuda:0'), out_proj_covar=tensor([0.0030, 0.0022, 0.0020, 0.0033], device='cuda:0')
+2023-03-01 03:51:39,014 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4452, 1.6518, 1.2632, 0.7417], device='cuda:0'), covar=tensor([0.0520, 0.0348, 0.0298, 0.0495], device='cuda:0'), in_proj_covar=tensor([0.0992, 0.0697, 0.0753, 0.0821], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 03:52:03,109 INFO [train.py:968] (0/2) Epoch 2, batch 16750, giga_loss[loss=0.3137, simple_loss=0.3795, pruned_loss=0.124, over 28938.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3767, pruned_loss=0.122, over 5687063.60 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3833, pruned_loss=0.1345, over 5774936.82 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.376, pruned_loss=0.121, over 5675081.07 frames. ], batch size: 186, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:52:42,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.013e+02 1.597e+03 1.964e+03 2.767e+03 5.677e+03, threshold=3.929e+03, percent-clipped=11.0
+2023-03-01 03:53:08,523 INFO [train.py:968] (0/2) Epoch 2, batch 16800, giga_loss[loss=0.2839, simple_loss=0.3344, pruned_loss=0.1167, over 24953.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3757, pruned_loss=0.1214, over 5688370.24 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3827, pruned_loss=0.1342, over 5780349.97 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3754, pruned_loss=0.1204, over 5670226.85 frames. ], batch size: 705, lr: 1.38e-02, grad_scale: 8.0
+2023-03-01 03:53:55,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:54:06,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:54:11,084 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:54:14,963 INFO [train.py:968] (0/2) Epoch 2, batch 16850, libri_loss[loss=0.2399, simple_loss=0.3095, pruned_loss=0.08515, over 29372.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3752, pruned_loss=0.1208, over 5690821.68 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.3818, pruned_loss=0.1336, over 5784941.39 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3755, pruned_loss=0.12, over 5667804.76 frames. ], batch size: 67, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 03:54:53,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4497, 3.8257, 4.2108, 1.7755], device='cuda:0'), covar=tensor([0.0330, 0.0307, 0.0656, 0.1704], device='cuda:0'), in_proj_covar=tensor([0.0712, 0.0514, 0.0740, 0.0523], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 03:55:02,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.903e+02 1.411e+03 1.731e+03 2.518e+03 5.543e+03, threshold=3.462e+03, percent-clipped=11.0
+2023-03-01 03:55:27,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1668, 1.4596, 1.2047, 0.4229], device='cuda:0'), covar=tensor([0.0787, 0.0625, 0.0798, 0.1126], device='cuda:0'), in_proj_covar=tensor([0.1149, 0.1164, 0.1166, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 03:55:27,713 INFO [train.py:968] (0/2) Epoch 2, batch 16900, giga_loss[loss=0.3303, simple_loss=0.3579, pruned_loss=0.1514, over 24092.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3754, pruned_loss=0.1205, over 5691427.29 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.382, pruned_loss=0.1338, over 5787527.87 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3754, pruned_loss=0.1195, over 5668607.37 frames. ], batch size: 705, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 03:56:39,299 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 03:56:41,716 INFO [train.py:968] (0/2) Epoch 2, batch 16950, giga_loss[loss=0.2953, simple_loss=0.372, pruned_loss=0.1093, over 28970.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3799, pruned_loss=0.1233, over 5686955.98 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3821, pruned_loss=0.1338, over 5780044.12 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3797, pruned_loss=0.1221, over 5673113.28 frames. ], batch size: 136, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 03:57:15,390 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:57:18,493 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:57:26,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.519e+02 1.550e+03 1.855e+03 2.633e+03 8.725e+03, threshold=3.710e+03, percent-clipped=11.0
+2023-03-01 03:57:29,487 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:57:32,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:57:47,488 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:57:50,248 INFO [train.py:968] (0/2) Epoch 2, batch 17000, giga_loss[loss=0.3051, simple_loss=0.3725, pruned_loss=0.1188, over 29060.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.379, pruned_loss=0.1224, over 5684793.98 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3816, pruned_loss=0.1335, over 5774893.03 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3793, pruned_loss=0.1214, over 5675105.61 frames. ], batch size: 199, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 03:58:02,991 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:58:12,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:58:17,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:59:07,070 INFO [train.py:968] (0/2) Epoch 2, batch 17050, libri_loss[loss=0.3432, simple_loss=0.4056, pruned_loss=0.1405, over 27518.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3775, pruned_loss=0.1224, over 5690876.01 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.382, pruned_loss=0.1338, over 5774819.99 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3773, pruned_loss=0.1212, over 5682146.38 frames. ], batch size: 115, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:00:01,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.036e+02 1.393e+03 1.790e+03 2.478e+03 6.787e+03, threshold=3.580e+03, percent-clipped=3.0
+2023-03-01 04:00:24,369 INFO [train.py:968] (0/2) Epoch 2, batch 17100, giga_loss[loss=0.2931, simple_loss=0.3693, pruned_loss=0.1084, over 28203.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3743, pruned_loss=0.12, over 5692505.15 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.382, pruned_loss=0.1338, over 5775184.39 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.374, pruned_loss=0.119, over 5684522.95 frames. ], batch size: 412, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:00:47,609 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:01:34,475 INFO [train.py:968] (0/2) Epoch 2, batch 17150, giga_loss[loss=0.2763, simple_loss=0.3465, pruned_loss=0.1031, over 28420.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3726, pruned_loss=0.118, over 5700472.64 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3824, pruned_loss=0.134, over 5773915.46 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3717, pruned_loss=0.1166, over 5692526.25 frames. ], batch size: 65, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:01:38,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7222, 2.5968, 1.9623, 1.9080], device='cuda:0'), covar=tensor([0.1582, 0.1239, 0.1116, 0.0746], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0770, 0.0697, 0.0591], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 04:01:43,285 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:01:46,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:02:18,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.372e+02 1.380e+03 1.727e+03 2.411e+03 6.906e+03, threshold=3.453e+03, percent-clipped=12.0
+2023-03-01 04:02:27,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:02:43,301 INFO [train.py:968] (0/2) Epoch 2, batch 17200, giga_loss[loss=0.3741, simple_loss=0.4271, pruned_loss=0.1606, over 28661.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3725, pruned_loss=0.1183, over 5691086.12 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3822, pruned_loss=0.1338, over 5775321.63 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.372, pruned_loss=0.1172, over 5683101.91 frames. ], batch size: 307, lr: 1.38e-02, grad_scale: 8.0
+2023-03-01 04:03:06,693 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 04:03:08,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:03:40,546 INFO [train.py:968] (0/2) Epoch 2, batch 17250, giga_loss[loss=0.3649, simple_loss=0.4189, pruned_loss=0.1554, over 28645.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3751, pruned_loss=0.12, over 5695444.74 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3817, pruned_loss=0.1334, over 5779338.26 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3748, pruned_loss=0.119, over 5683264.55 frames. ], batch size: 242, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:04:02,339 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:04:17,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.800e+02 1.323e+03 1.880e+03 2.431e+03 7.106e+03, threshold=3.760e+03, percent-clipped=12.0
+2023-03-01 04:04:35,510 INFO [train.py:968] (0/2) Epoch 2, batch 17300, giga_loss[loss=0.3276, simple_loss=0.3919, pruned_loss=0.1317, over 28452.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3772, pruned_loss=0.1222, over 5689170.91 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3818, pruned_loss=0.1333, over 5774425.59 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3768, pruned_loss=0.1211, over 5680030.22 frames. ], batch size: 78, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:04:49,405 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:05:27,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4054, 3.7368, 4.1859, 1.7812], device='cuda:0'), covar=tensor([0.0440, 0.0422, 0.0748, 0.1944], device='cuda:0'), in_proj_covar=tensor([0.0731, 0.0536, 0.0757, 0.0535], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 04:05:37,839 INFO [train.py:968] (0/2) Epoch 2, batch 17350, giga_loss[loss=0.3284, simple_loss=0.3863, pruned_loss=0.1353, over 28943.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3753, pruned_loss=0.1224, over 5687979.67 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3821, pruned_loss=0.1336, over 5775625.31 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3745, pruned_loss=0.121, over 5677898.80 frames. ], batch size: 227, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:05:58,447 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:06:02,133 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:06:06,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:06:13,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3386, 1.8938, 1.2937, 1.1550], device='cuda:0'), covar=tensor([0.0880, 0.0509, 0.0867, 0.1446], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0438, 0.0333, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 04:06:20,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.385e+02 1.466e+03 1.978e+03 2.630e+03 5.732e+03, threshold=3.955e+03, percent-clipped=9.0
+2023-03-01 04:06:26,732 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:06:33,739 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:06:35,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3465, 1.2556, 1.1986, 1.3880], device='cuda:0'), covar=tensor([0.1952, 0.1942, 0.1717, 0.1970], device='cuda:0'), in_proj_covar=tensor([0.0919, 0.0766, 0.0854, 0.0904], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 04:06:36,358 INFO [train.py:968] (0/2) Epoch 2, batch 17400, giga_loss[loss=0.3016, simple_loss=0.3642, pruned_loss=0.1195, over 28922.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3748, pruned_loss=0.1228, over 5686708.37 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3822, pruned_loss=0.1336, over 5775966.08 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3739, pruned_loss=0.1214, over 5675989.20 frames. ], batch size: 213, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:07:32,302 INFO [train.py:968] (0/2) Epoch 2, batch 17450, giga_loss[loss=0.4757, simple_loss=0.495, pruned_loss=0.2282, over 28908.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3804, pruned_loss=0.1277, over 5686673.84 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3824, pruned_loss=0.1337, over 5769296.80 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3793, pruned_loss=0.1262, over 5682000.52 frames. ], batch size: 186, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:08:09,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0691, 1.2917, 1.0018, 0.1462], device='cuda:0'), covar=tensor([0.0918, 0.0796, 0.1279, 0.1494], device='cuda:0'), in_proj_covar=tensor([0.1143, 0.1161, 0.1164, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 04:08:15,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.812e+02 1.574e+03 2.006e+03 2.827e+03 7.034e+03, threshold=4.012e+03, percent-clipped=10.0
+2023-03-01 04:08:20,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:08:28,470 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:08:31,515 INFO [train.py:968] (0/2) Epoch 2, batch 17500, giga_loss[loss=0.4069, simple_loss=0.4461, pruned_loss=0.1839, over 28017.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.392, pruned_loss=0.1361, over 5677069.46 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3827, pruned_loss=0.1338, over 5760264.49 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.391, pruned_loss=0.1347, over 5680274.92 frames. ], batch size: 412, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:08:45,501 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:08:47,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:09:07,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5922, 2.0592, 1.5298, 1.7466], device='cuda:0'), covar=tensor([0.0518, 0.0738, 0.0953, 0.1008], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0510, 0.0550, 0.0471], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 04:09:14,944 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:09:15,841 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-01 04:09:17,115 INFO [train.py:968] (0/2) Epoch 2, batch 17550, giga_loss[loss=0.3416, simple_loss=0.3975, pruned_loss=0.1428, over 29063.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3999, pruned_loss=0.1418, over 5688261.92 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3824, pruned_loss=0.1336, over 5762997.57 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3995, pruned_loss=0.141, over 5686973.66 frames. ], batch size: 128, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:09:32,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8260, 1.5379, 1.5335, 1.4947], device='cuda:0'), covar=tensor([0.0728, 0.1377, 0.1186, 0.1044], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0802, 0.0616, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 04:09:47,321 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.233e+02 1.200e+03 1.460e+03 1.787e+03 3.657e+03, threshold=2.921e+03, percent-clipped=0.0
+2023-03-01 04:10:03,752 INFO [train.py:968] (0/2) Epoch 2, batch 17600, giga_loss[loss=0.3827, simple_loss=0.4171, pruned_loss=0.1741, over 28037.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3973, pruned_loss=0.1414, over 5678657.59 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.3827, pruned_loss=0.1337, over 5753280.53 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3972, pruned_loss=0.141, over 5683643.64 frames. ], batch size: 412, lr: 1.38e-02, grad_scale: 8.0
+2023-03-01 04:10:33,487 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:10:35,591 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:10:43,201 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:10:48,546 INFO [train.py:968] (0/2) Epoch 2, batch 17650, giga_loss[loss=0.2792, simple_loss=0.3385, pruned_loss=0.11, over 28583.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3901, pruned_loss=0.1382, over 5681857.95 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3836, pruned_loss=0.1343, over 5755312.42 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3895, pruned_loss=0.1374, over 5682420.93 frames. ], batch size: 85, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:11:02,476 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:11:25,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:11:25,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.728e+02 1.236e+03 1.525e+03 1.932e+03 4.317e+03, threshold=3.050e+03, percent-clipped=7.0
+2023-03-01 04:11:38,516 INFO [train.py:968] (0/2) Epoch 2, batch 17700, giga_loss[loss=0.303, simple_loss=0.3575, pruned_loss=0.1243, over 29043.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3819, pruned_loss=0.1344, over 5670752.42 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3838, pruned_loss=0.1345, over 5747672.70 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3813, pruned_loss=0.1337, over 5677821.03 frames. ], batch size: 136, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:11:58,483 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.16 vs. limit=2.0
+2023-03-01 04:12:21,568 INFO [train.py:968] (0/2) Epoch 2, batch 17750, giga_loss[loss=0.3276, simple_loss=0.3721, pruned_loss=0.1416, over 27836.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3731, pruned_loss=0.1296, over 5661521.23 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3847, pruned_loss=0.135, over 5731511.43 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3716, pruned_loss=0.1285, over 5680571.49 frames. ], batch size: 412, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:12:36,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:12:54,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.707e+02 1.014e+03 1.305e+03 1.747e+03 9.794e+03, threshold=2.610e+03, percent-clipped=5.0
+2023-03-01 04:12:55,825 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:12:58,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:13:08,063 INFO [train.py:968] (0/2) Epoch 2, batch 17800, giga_loss[loss=0.2824, simple_loss=0.3306, pruned_loss=0.1171, over 28756.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3659, pruned_loss=0.1261, over 5671140.51 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3855, pruned_loss=0.1354, over 5732901.36 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3637, pruned_loss=0.1247, over 5683742.41 frames. ], batch size: 99, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:13:23,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:13:30,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:13:32,824 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:13:49,823 INFO [train.py:968] (0/2) Epoch 2, batch 17850, giga_loss[loss=0.2873, simple_loss=0.3382, pruned_loss=0.1182, over 27607.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3601, pruned_loss=0.1227, over 5671936.15 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3861, pruned_loss=0.1357, over 5726787.06 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3576, pruned_loss=0.1213, over 5685590.26 frames. ], batch size: 472, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:13:57,266 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:14:06,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:14:16,761 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.528e+02 1.227e+03 1.759e+03 2.525e+03 7.614e+03, threshold=3.517e+03, percent-clipped=22.0
+2023-03-01 04:14:28,560 INFO [train.py:968] (0/2) Epoch 2, batch 17900, libri_loss[loss=0.3392, simple_loss=0.4007, pruned_loss=0.1388, over 29546.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3596, pruned_loss=0.1226, over 5677834.28 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3872, pruned_loss=0.1364, over 5722697.88 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3553, pruned_loss=0.1201, over 5690370.92 frames. ], batch size: 83, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:14:35,295 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:14:37,165 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:15:04,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:15:09,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:15:14,089 INFO [train.py:968] (0/2) Epoch 2, batch 17950, giga_loss[loss=0.2908, simple_loss=0.3448, pruned_loss=0.1184, over 28883.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.355, pruned_loss=0.12, over 5675688.13 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3876, pruned_loss=0.1366, over 5723409.78 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3504, pruned_loss=0.1174, over 5683900.32 frames. ], batch size: 112, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:15:46,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.930e+02 9.522e+02 1.329e+03 1.864e+03 8.081e+03, threshold=2.658e+03, percent-clipped=3.0
+2023-03-01 04:15:57,781 INFO [train.py:968] (0/2) Epoch 2, batch 18000, libri_loss[loss=0.4616, simple_loss=0.4987, pruned_loss=0.2122, over 25685.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3509, pruned_loss=0.1172, over 5689953.44 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3884, pruned_loss=0.1371, over 5722487.97 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3461, pruned_loss=0.1145, over 5697088.61 frames. ], batch size: 136, lr: 1.37e-02, grad_scale: 8.0
+2023-03-01 04:15:57,785 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 04:16:06,576 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.2793, simple_loss=0.3737, pruned_loss=0.09248, over 944034.00 frames. 
+2023-03-01 04:16:06,576 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19478MB
+2023-03-01 04:16:20,958 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:16:23,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:16:51,454 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:16:53,434 INFO [train.py:968] (0/2) Epoch 2, batch 18050, giga_loss[loss=0.2395, simple_loss=0.3081, pruned_loss=0.08548, over 28612.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3478, pruned_loss=0.1155, over 5672481.72 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3891, pruned_loss=0.1375, over 5715972.00 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3425, pruned_loss=0.1126, over 5683001.42 frames. ], batch size: 307, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:17:23,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.068e+02 1.151e+03 1.581e+03 2.198e+03 5.510e+03, threshold=3.163e+03, percent-clipped=13.0
+2023-03-01 04:17:36,332 INFO [train.py:968] (0/2) Epoch 2, batch 18100, libri_loss[loss=0.3525, simple_loss=0.418, pruned_loss=0.1435, over 26054.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3474, pruned_loss=0.1155, over 5679850.83 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3902, pruned_loss=0.1381, over 5714966.52 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3404, pruned_loss=0.1117, over 5688123.80 frames. ], batch size: 136, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:18:16,331 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:18:17,638 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:18:20,205 INFO [train.py:968] (0/2) Epoch 2, batch 18150, giga_loss[loss=0.2451, simple_loss=0.3093, pruned_loss=0.09045, over 28876.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3428, pruned_loss=0.1131, over 5683294.68 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3898, pruned_loss=0.1377, over 5718005.35 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3366, pruned_loss=0.1099, over 5686492.33 frames. ], batch size: 112, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:18:52,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.896e+02 1.023e+03 1.208e+03 1.502e+03 3.991e+03, threshold=2.417e+03, percent-clipped=3.0
+2023-03-01 04:19:08,662 INFO [train.py:968] (0/2) Epoch 2, batch 18200, giga_loss[loss=0.3125, simple_loss=0.3558, pruned_loss=0.1346, over 27975.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3398, pruned_loss=0.1116, over 5680355.34 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3904, pruned_loss=0.138, over 5720682.44 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3336, pruned_loss=0.1084, over 5679995.32 frames. ], batch size: 412, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:19:54,912 INFO [train.py:968] (0/2) Epoch 2, batch 18250, giga_loss[loss=0.2249, simple_loss=0.2974, pruned_loss=0.0762, over 28911.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3368, pruned_loss=0.1098, over 5683672.71 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3912, pruned_loss=0.1384, over 5725448.44 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3295, pruned_loss=0.1061, over 5677577.10 frames. ], batch size: 145, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:20:27,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.980e+02 1.087e+03 1.539e+03 2.466e+03 5.877e+03, threshold=3.078e+03, percent-clipped=25.0
+2023-03-01 04:20:37,257 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:20:42,743 INFO [train.py:968] (0/2) Epoch 2, batch 18300, giga_loss[loss=0.3256, simple_loss=0.3867, pruned_loss=0.1323, over 28935.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3438, pruned_loss=0.1148, over 5666535.58 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3913, pruned_loss=0.1383, over 5720169.12 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3366, pruned_loss=0.1111, over 5665462.33 frames. ], batch size: 213, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:20:53,603 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:21:03,967 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:21:14,041 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-64000.pt
+2023-03-01 04:21:30,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7393, 2.3524, 2.0733, 1.8916], device='cuda:0'), covar=tensor([0.0512, 0.0650, 0.0839, 0.0939], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0507, 0.0538, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 04:21:32,422 INFO [train.py:968] (0/2) Epoch 2, batch 18350, giga_loss[loss=0.3635, simple_loss=0.4185, pruned_loss=0.1543, over 28904.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3603, pruned_loss=0.1252, over 5670911.07 frames. ], libri_tot_loss[loss=0.3342, simple_loss=0.3914, pruned_loss=0.1384, over 5718798.82 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3539, pruned_loss=0.1219, over 5670530.45 frames. ], batch size: 112, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:22:04,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.563e+02 1.373e+03 1.760e+03 2.137e+03 6.031e+03, threshold=3.520e+03, percent-clipped=6.0
+2023-03-01 04:22:13,679 INFO [train.py:968] (0/2) Epoch 2, batch 18400, giga_loss[loss=0.3704, simple_loss=0.4118, pruned_loss=0.1645, over 28991.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3745, pruned_loss=0.1339, over 5688314.33 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.392, pruned_loss=0.139, over 5723174.15 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.368, pruned_loss=0.1305, over 5682459.17 frames. ], batch size: 106, lr: 1.37e-02, grad_scale: 8.0
+2023-03-01 04:22:56,530 INFO [train.py:968] (0/2) Epoch 2, batch 18450, libri_loss[loss=0.3968, simple_loss=0.4403, pruned_loss=0.1766, over 29527.00 frames. ], tot_loss[loss=0.331, simple_loss=0.384, pruned_loss=0.139, over 5683287.11 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3922, pruned_loss=0.1393, over 5717246.45 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3783, pruned_loss=0.1359, over 5682825.66 frames. ], batch size: 84, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:23:08,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1268, 3.4364, 2.1181, 1.8707], device='cuda:0'), covar=tensor([0.0755, 0.0345, 0.0747, 0.1289], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0423, 0.0330, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 04:23:08,123 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:23:11,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:23:11,340 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-01 04:23:23,644 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-01 04:23:30,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.615e+02 1.261e+03 1.586e+03 2.127e+03 6.784e+03, threshold=3.172e+03, percent-clipped=5.0
+2023-03-01 04:23:39,066 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:23:43,132 INFO [train.py:968] (0/2) Epoch 2, batch 18500, giga_loss[loss=0.3728, simple_loss=0.4214, pruned_loss=0.1621, over 29011.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3889, pruned_loss=0.1403, over 5675329.41 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3922, pruned_loss=0.1393, over 5709741.08 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3843, pruned_loss=0.1379, over 5680465.41 frames. ], batch size: 136, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:23:57,745 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:24:00,056 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:24:11,400 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:24:26,017 INFO [train.py:968] (0/2) Epoch 2, batch 18550, giga_loss[loss=0.3166, simple_loss=0.3903, pruned_loss=0.1215, over 28735.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.389, pruned_loss=0.1382, over 5674770.54 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3923, pruned_loss=0.139, over 5712845.67 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3851, pruned_loss=0.1365, over 5675304.71 frames. ], batch size: 242, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:24:42,424 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:25:00,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.811e+02 1.121e+03 1.538e+03 1.918e+03 6.279e+03, threshold=3.076e+03, percent-clipped=6.0
+2023-03-01 04:25:12,034 INFO [train.py:968] (0/2) Epoch 2, batch 18600, giga_loss[loss=0.3286, simple_loss=0.3824, pruned_loss=0.1374, over 28496.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3912, pruned_loss=0.1397, over 5656524.96 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3926, pruned_loss=0.1392, over 5702843.80 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3878, pruned_loss=0.1382, over 5665127.02 frames. ], batch size: 85, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:25:18,355 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-01 04:25:46,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:25:54,275 INFO [train.py:968] (0/2) Epoch 2, batch 18650, giga_loss[loss=0.3266, simple_loss=0.3861, pruned_loss=0.1335, over 28607.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3941, pruned_loss=0.1421, over 5663496.64 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3932, pruned_loss=0.1394, over 5700248.19 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3908, pruned_loss=0.1407, over 5671813.52 frames. ], batch size: 71, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:26:05,025 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:06,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:06,935 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:09,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:10,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:26,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:27,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.162e+02 1.150e+03 1.431e+03 2.080e+03 5.407e+03, threshold=2.861e+03, percent-clipped=9.0
+2023-03-01 04:26:33,143 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:34,363 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:36,906 INFO [train.py:968] (0/2) Epoch 2, batch 18700, giga_loss[loss=0.3628, simple_loss=0.4172, pruned_loss=0.1542, over 28662.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3974, pruned_loss=0.1442, over 5671139.35 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3945, pruned_loss=0.1401, over 5703401.99 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3935, pruned_loss=0.1426, over 5673365.05 frames. ], batch size: 242, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:26:37,941 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9027, 1.7940, 3.1939, 2.9662], device='cuda:0'), covar=tensor([0.1184, 0.1273, 0.0312, 0.0493], device='cuda:0'), in_proj_covar=tensor([0.0476, 0.0473, 0.0591, 0.0479], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:0')
+2023-03-01 04:27:18,418 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:27:20,076 INFO [train.py:968] (0/2) Epoch 2, batch 18750, giga_loss[loss=0.3479, simple_loss=0.4055, pruned_loss=0.1452, over 28694.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.4006, pruned_loss=0.1458, over 5670757.18 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3946, pruned_loss=0.1403, over 5697280.96 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3976, pruned_loss=0.1444, over 5677847.07 frames. ], batch size: 243, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:27:28,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3599, 1.3165, 1.3716, 1.6712], device='cuda:0'), covar=tensor([0.1754, 0.1402, 0.1196, 0.1479], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0787, 0.0869, 0.0908], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 04:27:50,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.152e+02 1.284e+03 1.688e+03 2.201e+03 6.916e+03, threshold=3.377e+03, percent-clipped=7.0
+2023-03-01 04:28:00,874 INFO [train.py:968] (0/2) Epoch 2, batch 18800, giga_loss[loss=0.3567, simple_loss=0.4196, pruned_loss=0.1469, over 29074.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.4025, pruned_loss=0.146, over 5667977.16 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3947, pruned_loss=0.1402, over 5698937.80 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.4, pruned_loss=0.1451, over 5671586.24 frames. ], batch size: 128, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:28:11,630 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:28:15,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:28:22,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6527, 1.4651, 3.5745, 2.9715], device='cuda:0'), covar=tensor([0.1454, 0.1503, 0.0276, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0478, 0.0595, 0.0486], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:0')
+2023-03-01 04:28:31,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:28:32,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:28:41,262 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:28:46,894 INFO [train.py:968] (0/2) Epoch 2, batch 18850, giga_loss[loss=0.3255, simple_loss=0.3969, pruned_loss=0.1271, over 28569.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.4034, pruned_loss=0.1454, over 5683132.57 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3947, pruned_loss=0.1399, over 5704253.02 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.4017, pruned_loss=0.1451, over 5680694.69 frames. ], batch size: 60, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:28:55,509 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:29:20,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.455e+02 1.224e+03 1.530e+03 2.300e+03 7.718e+03, threshold=3.061e+03, percent-clipped=11.0
+2023-03-01 04:29:31,539 INFO [train.py:968] (0/2) Epoch 2, batch 18900, giga_loss[loss=0.4065, simple_loss=0.4296, pruned_loss=0.1917, over 26698.00 frames. ], tot_loss[loss=0.3474, simple_loss=0.4045, pruned_loss=0.1452, over 5689900.29 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.3953, pruned_loss=0.1403, over 5708511.08 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.4028, pruned_loss=0.1447, over 5683875.40 frames. ], batch size: 555, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:29:39,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:30:07,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:30:12,142 INFO [train.py:968] (0/2) Epoch 2, batch 18950, giga_loss[loss=0.2916, simple_loss=0.3713, pruned_loss=0.106, over 28836.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.4021, pruned_loss=0.1416, over 5707871.86 frames. ], libri_tot_loss[loss=0.3381, simple_loss=0.3954, pruned_loss=0.1404, over 5713250.88 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.4008, pruned_loss=0.1412, over 5698499.83 frames. ], batch size: 112, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:30:17,584 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:30:42,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.482e+02 1.099e+03 1.366e+03 1.768e+03 4.492e+03, threshold=2.732e+03, percent-clipped=5.0
+2023-03-01 04:30:54,939 INFO [train.py:968] (0/2) Epoch 2, batch 19000, giga_loss[loss=0.3138, simple_loss=0.3565, pruned_loss=0.1355, over 23650.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.4005, pruned_loss=0.1402, over 5705554.55 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3962, pruned_loss=0.141, over 5717153.46 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3989, pruned_loss=0.1394, over 5694399.07 frames. ], batch size: 705, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:31:06,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:31:36,171 INFO [train.py:968] (0/2) Epoch 2, batch 19050, giga_loss[loss=0.3833, simple_loss=0.4224, pruned_loss=0.1721, over 28621.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.4032, pruned_loss=0.1433, over 5706068.61 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3966, pruned_loss=0.1411, over 5710778.70 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.4017, pruned_loss=0.1426, over 5703140.10 frames. ], batch size: 307, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:31:37,860 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:31:39,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:31:54,385 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:06,705 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:10,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.041e+02 1.281e+03 1.743e+03 2.302e+03 5.153e+03, threshold=3.486e+03, percent-clipped=14.0
+2023-03-01 04:32:11,157 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:14,030 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:24,054 INFO [train.py:968] (0/2) Epoch 2, batch 19100, giga_loss[loss=0.4436, simple_loss=0.4545, pruned_loss=0.2163, over 27651.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4053, pruned_loss=0.1479, over 5706416.40 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3969, pruned_loss=0.1412, over 5709669.30 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.404, pruned_loss=0.1473, over 5705171.78 frames. ], batch size: 472, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:32:41,890 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:45,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:49,694 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4389, 2.1918, 1.3936, 1.3806], device='cuda:0'), covar=tensor([0.0972, 0.0588, 0.0953, 0.1548], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0438, 0.0332, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 04:33:09,655 INFO [train.py:968] (0/2) Epoch 2, batch 19150, giga_loss[loss=0.3397, simple_loss=0.3885, pruned_loss=0.1454, over 28722.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.4059, pruned_loss=0.1502, over 5699741.02 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3971, pruned_loss=0.1413, over 5707002.02 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.4048, pruned_loss=0.1498, over 5701573.34 frames. ], batch size: 85, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:33:15,178 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:33:18,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:33:43,044 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.441e+02 1.252e+03 1.705e+03 2.080e+03 6.092e+03, threshold=3.409e+03, percent-clipped=6.0
+2023-03-01 04:33:43,879 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:33:45,240 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:33:51,558 INFO [train.py:968] (0/2) Epoch 2, batch 19200, giga_loss[loss=0.3518, simple_loss=0.3976, pruned_loss=0.153, over 28741.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.4053, pruned_loss=0.1511, over 5701834.33 frames. ], libri_tot_loss[loss=0.3403, simple_loss=0.3976, pruned_loss=0.1415, over 5711781.04 frames. ], giga_tot_loss[loss=0.3529, simple_loss=0.4042, pruned_loss=0.1508, over 5698776.01 frames. ], batch size: 119, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:34:11,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3958, 1.9709, 1.3601, 1.2362], device='cuda:0'), covar=tensor([0.0951, 0.0646, 0.0900, 0.1579], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0438, 0.0330, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 04:34:35,861 INFO [train.py:968] (0/2) Epoch 2, batch 19250, giga_loss[loss=0.3624, simple_loss=0.4086, pruned_loss=0.1581, over 28848.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4043, pruned_loss=0.1512, over 5702097.89 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3981, pruned_loss=0.1416, over 5717514.14 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.4032, pruned_loss=0.1512, over 5693933.76 frames. ], batch size: 284, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:34:46,868 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:34:48,965 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:35:11,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.935e+02 1.321e+03 1.832e+03 2.599e+03 5.312e+03, threshold=3.664e+03, percent-clipped=6.0
+2023-03-01 04:35:19,267 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:35:20,958 INFO [train.py:968] (0/2) Epoch 2, batch 19300, giga_loss[loss=0.3763, simple_loss=0.4207, pruned_loss=0.166, over 28683.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.4029, pruned_loss=0.1493, over 5716015.49 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3986, pruned_loss=0.142, over 5722331.39 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4017, pruned_loss=0.1492, over 5704954.25 frames. ], batch size: 262, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:35:33,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:35:47,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:36:03,997 INFO [train.py:968] (0/2) Epoch 2, batch 19350, giga_loss[loss=0.3131, simple_loss=0.3765, pruned_loss=0.1249, over 28532.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.4006, pruned_loss=0.1464, over 5719377.18 frames. ], libri_tot_loss[loss=0.3406, simple_loss=0.398, pruned_loss=0.1416, over 5724801.08 frames. ], giga_tot_loss[loss=0.3467, simple_loss=0.4001, pruned_loss=0.1467, over 5708354.45 frames. ], batch size: 336, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:36:17,602 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-01 04:36:27,527 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:36:27,632 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8145, 2.1736, 1.8587, 1.8198], device='cuda:0'), covar=tensor([0.0528, 0.0717, 0.0826, 0.0876], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0509, 0.0545, 0.0471], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 04:36:41,280 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.221e+02 1.177e+03 1.499e+03 1.908e+03 6.457e+03, threshold=2.998e+03, percent-clipped=3.0
+2023-03-01 04:36:50,519 INFO [train.py:968] (0/2) Epoch 2, batch 19400, giga_loss[loss=0.3202, simple_loss=0.3735, pruned_loss=0.1335, over 27720.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3967, pruned_loss=0.1436, over 5704721.56 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.3982, pruned_loss=0.1417, over 5730744.10 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3962, pruned_loss=0.1439, over 5689964.66 frames. ], batch size: 472, lr: 1.36e-02, grad_scale: 2.0
+2023-03-01 04:37:16,022 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:37:32,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:37:39,690 INFO [train.py:968] (0/2) Epoch 2, batch 19450, giga_loss[loss=0.3506, simple_loss=0.3855, pruned_loss=0.1579, over 26650.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3912, pruned_loss=0.1405, over 5693390.73 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3986, pruned_loss=0.142, over 5731253.96 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3903, pruned_loss=0.1405, over 5680116.95 frames. ], batch size: 555, lr: 1.36e-02, grad_scale: 2.0
+2023-03-01 04:37:58,598 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:38:02,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7328, 2.1946, 1.8294, 1.8125], device='cuda:0'), covar=tensor([0.0515, 0.0743, 0.0883, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0505, 0.0544, 0.0470], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 04:38:02,534 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:38:04,026 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:38:04,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:38:15,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.289e+02 1.011e+03 1.364e+03 2.077e+03 9.899e+03, threshold=2.727e+03, percent-clipped=6.0
+2023-03-01 04:38:23,526 INFO [train.py:968] (0/2) Epoch 2, batch 19500, giga_loss[loss=0.2815, simple_loss=0.346, pruned_loss=0.1085, over 28578.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3846, pruned_loss=0.136, over 5694766.28 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3986, pruned_loss=0.1419, over 5735440.67 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3835, pruned_loss=0.1359, over 5678948.37 frames. ], batch size: 307, lr: 1.36e-02, grad_scale: 2.0
+2023-03-01 04:38:30,005 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:39:14,812 INFO [train.py:968] (0/2) Epoch 2, batch 19550, giga_loss[loss=0.3821, simple_loss=0.4053, pruned_loss=0.1795, over 26341.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3791, pruned_loss=0.133, over 5671939.09 frames. ], libri_tot_loss[loss=0.3416, simple_loss=0.399, pruned_loss=0.1421, over 5739888.65 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3774, pruned_loss=0.1326, over 5653958.15 frames. ], batch size: 555, lr: 1.36e-02, grad_scale: 2.0
+2023-03-01 04:39:29,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:39:49,733 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:39:51,815 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.447e+02 9.886e+02 1.340e+03 1.800e+03 4.687e+03, threshold=2.680e+03, percent-clipped=8.0
+2023-03-01 04:39:52,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:40:01,189 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-01 04:40:02,344 INFO [train.py:968] (0/2) Epoch 2, batch 19600, giga_loss[loss=0.3551, simple_loss=0.3994, pruned_loss=0.1554, over 27640.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3795, pruned_loss=0.1329, over 5671151.70 frames. ], libri_tot_loss[loss=0.3422, simple_loss=0.3996, pruned_loss=0.1424, over 5742530.71 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3772, pruned_loss=0.1321, over 5653607.73 frames. ], batch size: 472, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:40:21,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:40:38,180 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9941, 1.5868, 1.6548, 1.5744], device='cuda:0'), covar=tensor([0.0722, 0.1650, 0.1210, 0.1303], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0814, 0.0637, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 04:40:48,077 INFO [train.py:968] (0/2) Epoch 2, batch 19650, giga_loss[loss=0.3159, simple_loss=0.3706, pruned_loss=0.1306, over 28737.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3794, pruned_loss=0.1328, over 5673385.04 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.4, pruned_loss=0.1426, over 5746209.19 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3766, pruned_loss=0.1317, over 5653979.75 frames. ], batch size: 92, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:41:14,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5075, 2.0247, 1.5622, 1.7273], device='cuda:0'), covar=tensor([0.0512, 0.0664, 0.0884, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0495, 0.0534, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 04:41:17,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:41:18,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.693e+02 9.958e+02 1.370e+03 1.797e+03 4.742e+03, threshold=2.739e+03, percent-clipped=9.0
+2023-03-01 04:41:29,400 INFO [train.py:968] (0/2) Epoch 2, batch 19700, giga_loss[loss=0.2592, simple_loss=0.3305, pruned_loss=0.09392, over 28356.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3782, pruned_loss=0.1316, over 5680957.07 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.4008, pruned_loss=0.1428, over 5743330.86 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3745, pruned_loss=0.1301, over 5665425.57 frames. ], batch size: 65, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:41:29,650 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:41:35,472 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:41:37,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:42:02,951 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:42:09,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:42:12,416 INFO [train.py:968] (0/2) Epoch 2, batch 19750, giga_loss[loss=0.3302, simple_loss=0.3704, pruned_loss=0.145, over 24102.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3749, pruned_loss=0.1299, over 5685624.91 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.4009, pruned_loss=0.1428, over 5744142.20 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3718, pruned_loss=0.1287, over 5672582.29 frames. ], batch size: 705, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:42:35,412 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:42:43,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.656e+02 1.026e+03 1.227e+03 1.559e+03 5.813e+03, threshold=2.453e+03, percent-clipped=9.0
+2023-03-01 04:42:51,757 INFO [train.py:968] (0/2) Epoch 2, batch 19800, giga_loss[loss=0.2597, simple_loss=0.325, pruned_loss=0.09719, over 28494.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.373, pruned_loss=0.1284, over 5690317.21 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.4022, pruned_loss=0.1435, over 5740641.53 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3685, pruned_loss=0.1264, over 5680937.95 frames. ], batch size: 65, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:42:53,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:43:16,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4645, 1.2563, 1.3402, 1.2293], device='cuda:0'), covar=tensor([0.0788, 0.1444, 0.1460, 0.1208], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0823, 0.0643, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 04:43:18,783 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:43:21,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:43:32,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:43:35,961 INFO [train.py:968] (0/2) Epoch 2, batch 19850, giga_loss[loss=0.3006, simple_loss=0.3639, pruned_loss=0.1187, over 28559.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3685, pruned_loss=0.1255, over 5698734.93 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.4024, pruned_loss=0.1435, over 5741492.32 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3647, pruned_loss=0.1238, over 5690360.13 frames. ], batch size: 307, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:43:38,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:43:48,554 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:08,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.420e+02 9.402e+02 1.161e+03 1.543e+03 3.664e+03, threshold=2.321e+03, percent-clipped=10.0
+2023-03-01 04:44:08,943 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:11,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:16,625 INFO [train.py:968] (0/2) Epoch 2, batch 19900, giga_loss[loss=0.2701, simple_loss=0.3394, pruned_loss=0.1004, over 28958.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3666, pruned_loss=0.1247, over 5709254.98 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.4025, pruned_loss=0.1435, over 5747811.37 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.362, pruned_loss=0.1227, over 5695082.11 frames. ], batch size: 213, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:44:32,673 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:52,259 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:54,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:55,250 INFO [train.py:968] (0/2) Epoch 2, batch 19950, giga_loss[loss=0.3122, simple_loss=0.3711, pruned_loss=0.1267, over 29007.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.364, pruned_loss=0.1231, over 5719420.78 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.4032, pruned_loss=0.1437, over 5752113.35 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3587, pruned_loss=0.1207, over 5703113.12 frames. ], batch size: 164, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:45:12,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6101, 1.3310, 4.2867, 3.1082], device='cuda:0'), covar=tensor([0.1532, 0.1558, 0.0268, 0.0409], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0478, 0.0609, 0.0495], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 04:45:17,713 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:24,953 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:27,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.507e+02 1.098e+03 1.414e+03 2.114e+03 6.836e+03, threshold=2.828e+03, percent-clipped=18.0
+2023-03-01 04:45:27,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:30,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:33,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:35,029 INFO [train.py:968] (0/2) Epoch 2, batch 20000, giga_loss[loss=0.2959, simple_loss=0.3521, pruned_loss=0.1198, over 28840.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3643, pruned_loss=0.123, over 5730997.18 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.4046, pruned_loss=0.1444, over 5758477.64 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3572, pruned_loss=0.1196, over 5711025.36 frames. ], batch size: 112, lr: 1.35e-02, grad_scale: 8.0
+2023-03-01 04:45:44,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:51,177 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:56,111 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:46:17,412 INFO [train.py:968] (0/2) Epoch 2, batch 20050, giga_loss[loss=0.2745, simple_loss=0.3421, pruned_loss=0.1034, over 29043.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3624, pruned_loss=0.1223, over 5724515.41 frames. ], libri_tot_loss[loss=0.348, simple_loss=0.4057, pruned_loss=0.1451, over 5759995.00 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3553, pruned_loss=0.1188, over 5706953.10 frames. ], batch size: 164, lr: 1.35e-02, grad_scale: 8.0
+2023-03-01 04:46:36,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:46:50,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.109e+02 9.791e+02 1.343e+03 1.786e+03 3.563e+03, threshold=2.686e+03, percent-clipped=4.0
+2023-03-01 04:46:56,598 INFO [train.py:968] (0/2) Epoch 2, batch 20100, giga_loss[loss=0.2682, simple_loss=0.3351, pruned_loss=0.1007, over 28622.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3628, pruned_loss=0.1225, over 5725423.57 frames. ], libri_tot_loss[loss=0.3498, simple_loss=0.4075, pruned_loss=0.1461, over 5760111.67 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3543, pruned_loss=0.1181, over 5710062.34 frames. ], batch size: 242, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:47:07,016 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-01 04:47:22,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3275, 1.3423, 1.2542, 1.4463], device='cuda:0'), covar=tensor([0.1824, 0.1937, 0.1613, 0.1992], device='cuda:0'), in_proj_covar=tensor([0.0955, 0.0794, 0.0867, 0.0918], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 04:47:36,137 INFO [train.py:968] (0/2) Epoch 2, batch 20150, giga_loss[loss=0.3694, simple_loss=0.4127, pruned_loss=0.1631, over 27928.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3636, pruned_loss=0.1232, over 5727764.25 frames. ], libri_tot_loss[loss=0.351, simple_loss=0.4085, pruned_loss=0.1467, over 5763172.57 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3549, pruned_loss=0.1185, over 5712062.04 frames. ], batch size: 412, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:47:40,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:48:05,250 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-01 04:48:05,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3395, 1.4649, 1.1157, 1.3705], device='cuda:0'), covar=tensor([0.1011, 0.0448, 0.0509, 0.1230], device='cuda:0'), in_proj_covar=tensor([0.0293, 0.0197, 0.0201, 0.0352], device='cuda:0'), out_proj_covar=tensor([0.0030, 0.0023, 0.0020, 0.0034], device='cuda:0')
+2023-03-01 04:48:15,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.975e+02 1.072e+03 1.454e+03 1.986e+03 6.776e+03, threshold=2.908e+03, percent-clipped=13.0
+2023-03-01 04:48:21,929 INFO [train.py:968] (0/2) Epoch 2, batch 20200, giga_loss[loss=0.3245, simple_loss=0.386, pruned_loss=0.1315, over 29001.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3701, pruned_loss=0.1281, over 5725510.60 frames. ], libri_tot_loss[loss=0.3516, simple_loss=0.4089, pruned_loss=0.1472, over 5765838.42 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.362, pruned_loss=0.1237, over 5709852.31 frames. ], batch size: 164, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:48:39,027 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:48:42,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:49:12,201 INFO [train.py:968] (0/2) Epoch 2, batch 20250, giga_loss[loss=0.3941, simple_loss=0.4192, pruned_loss=0.1845, over 23692.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3769, pruned_loss=0.1334, over 5698618.56 frames. ], libri_tot_loss[loss=0.3522, simple_loss=0.4093, pruned_loss=0.1475, over 5758176.58 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3695, pruned_loss=0.1293, over 5692821.19 frames. ], batch size: 705, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:49:12,480 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:49:17,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65925.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 04:49:56,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.427e+02 1.285e+03 1.651e+03 2.264e+03 5.886e+03, threshold=3.302e+03, percent-clipped=16.0
+2023-03-01 04:50:00,877 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:50:02,500 INFO [train.py:968] (0/2) Epoch 2, batch 20300, giga_loss[loss=0.3188, simple_loss=0.3837, pruned_loss=0.127, over 28798.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3874, pruned_loss=0.1416, over 5689404.57 frames. ], libri_tot_loss[loss=0.3529, simple_loss=0.4099, pruned_loss=0.1479, over 5750825.01 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3804, pruned_loss=0.1377, over 5690202.19 frames. ], batch size: 186, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:50:03,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:50:32,388 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:50:33,531 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-66000.pt
+2023-03-01 04:50:49,106 INFO [train.py:968] (0/2) Epoch 2, batch 20350, libri_loss[loss=0.3721, simple_loss=0.4225, pruned_loss=0.1609, over 19888.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3924, pruned_loss=0.1438, over 5685639.25 frames. ], libri_tot_loss[loss=0.3528, simple_loss=0.4097, pruned_loss=0.148, over 5744917.80 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3864, pruned_loss=0.1405, over 5690543.75 frames. ], batch size: 187, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:50:51,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8295, 1.4829, 1.4595, 1.5507], device='cuda:0'), covar=tensor([0.0851, 0.1658, 0.1418, 0.1325], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0816, 0.0646, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 04:51:26,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:51:32,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.668e+02 1.169e+03 1.457e+03 1.896e+03 4.301e+03, threshold=2.914e+03, percent-clipped=4.0
+2023-03-01 04:51:39,023 INFO [train.py:968] (0/2) Epoch 2, batch 20400, giga_loss[loss=0.3338, simple_loss=0.3998, pruned_loss=0.1338, over 28988.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3986, pruned_loss=0.1468, over 5683885.01 frames. ], libri_tot_loss[loss=0.3526, simple_loss=0.4096, pruned_loss=0.1478, over 5746552.52 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3939, pruned_loss=0.1444, over 5685687.14 frames. ], batch size: 164, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:52:28,467 INFO [train.py:968] (0/2) Epoch 2, batch 20450, giga_loss[loss=0.3738, simple_loss=0.4249, pruned_loss=0.1614, over 28894.00 frames. ], tot_loss[loss=0.3532, simple_loss=0.4046, pruned_loss=0.1509, over 5680002.08 frames. ], libri_tot_loss[loss=0.3533, simple_loss=0.4098, pruned_loss=0.1484, over 5740612.44 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4003, pruned_loss=0.1484, over 5684916.99 frames. ], batch size: 112, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:52:46,925 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 04:53:04,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.515e+02 1.230e+03 1.522e+03 2.406e+03 1.193e+04, threshold=3.043e+03, percent-clipped=17.0
+2023-03-01 04:53:10,701 INFO [train.py:968] (0/2) Epoch 2, batch 20500, giga_loss[loss=0.3017, simple_loss=0.3613, pruned_loss=0.1211, over 28827.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4053, pruned_loss=0.1517, over 5691004.60 frames. ], libri_tot_loss[loss=0.3538, simple_loss=0.4099, pruned_loss=0.1488, over 5744683.04 frames. ], giga_tot_loss[loss=0.3503, simple_loss=0.4017, pruned_loss=0.1494, over 5689894.81 frames. ], batch size: 99, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:53:26,029 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.33 vs. limit=5.0
+2023-03-01 04:53:26,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8924, 2.5743, 1.9832, 1.9387], device='cuda:0'), covar=tensor([0.1561, 0.1324, 0.1107, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0798, 0.0705, 0.0598], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 04:53:34,397 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:53:36,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:53:54,736 INFO [train.py:968] (0/2) Epoch 2, batch 20550, giga_loss[loss=0.3016, simple_loss=0.3706, pruned_loss=0.1163, over 28883.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3968, pruned_loss=0.1449, over 5695566.12 frames. ], libri_tot_loss[loss=0.354, simple_loss=0.4101, pruned_loss=0.149, over 5750263.87 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3935, pruned_loss=0.1429, over 5687982.25 frames. ], batch size: 174, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:54:02,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:54:05,991 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=66231.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 04:54:10,817 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 04:54:13,487 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-01 04:54:33,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.991e+02 1.108e+03 1.411e+03 1.952e+03 3.582e+03, threshold=2.821e+03, percent-clipped=5.0
+2023-03-01 04:54:39,747 INFO [train.py:968] (0/2) Epoch 2, batch 20600, giga_loss[loss=0.3794, simple_loss=0.4307, pruned_loss=0.164, over 28612.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3957, pruned_loss=0.1437, over 5695420.89 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.4108, pruned_loss=0.1498, over 5751316.99 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3921, pruned_loss=0.1412, over 5687700.56 frames. ], batch size: 307, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:55:09,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66300.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 04:55:25,632 INFO [train.py:968] (0/2) Epoch 2, batch 20650, giga_loss[loss=0.3287, simple_loss=0.3898, pruned_loss=0.1338, over 28916.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3957, pruned_loss=0.1429, over 5697140.21 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.4113, pruned_loss=0.1504, over 5753763.32 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3921, pruned_loss=0.1402, over 5687505.93 frames. ], batch size: 199, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:56:00,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.966e+02 1.371e+03 1.634e+03 2.230e+03 6.184e+03, threshold=3.267e+03, percent-clipped=9.0
+2023-03-01 04:56:03,014 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4574, 2.1839, 1.5275, 1.4235], device='cuda:0'), covar=tensor([0.0728, 0.0594, 0.0657, 0.1120], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0436, 0.0328, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:0')
+2023-03-01 04:56:07,023 INFO [train.py:968] (0/2) Epoch 2, batch 20700, giga_loss[loss=0.3531, simple_loss=0.4121, pruned_loss=0.1471, over 29098.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3991, pruned_loss=0.1447, over 5688451.31 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.4119, pruned_loss=0.1508, over 5747287.45 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3954, pruned_loss=0.142, over 5685874.85 frames. ], batch size: 128, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:56:57,232 INFO [train.py:968] (0/2) Epoch 2, batch 20750, giga_loss[loss=0.3276, simple_loss=0.384, pruned_loss=0.1356, over 28558.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.4028, pruned_loss=0.1479, over 5685707.17 frames. ], libri_tot_loss[loss=0.3572, simple_loss=0.4121, pruned_loss=0.1511, over 5747910.36 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3996, pruned_loss=0.1454, over 5682602.09 frames. ], batch size: 71, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:57:18,704 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66443.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 04:57:21,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66446.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 04:57:28,247 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-01 04:57:32,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.852e+02 1.288e+03 1.631e+03 2.216e+03 9.404e+03, threshold=3.263e+03, percent-clipped=11.0
+2023-03-01 04:57:35,129 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1983, 1.7808, 1.3246, 1.2542], device='cuda:0'), covar=tensor([0.1107, 0.0398, 0.0462, 0.1428], device='cuda:0'), in_proj_covar=tensor([0.0290, 0.0195, 0.0197, 0.0350], device='cuda:0'), out_proj_covar=tensor([0.0030, 0.0023, 0.0020, 0.0034], device='cuda:0')
+2023-03-01 04:57:43,208 INFO [train.py:968] (0/2) Epoch 2, batch 20800, giga_loss[loss=0.3347, simple_loss=0.389, pruned_loss=0.1402, over 28654.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.4053, pruned_loss=0.1502, over 5700606.14 frames. ], libri_tot_loss[loss=0.3585, simple_loss=0.4131, pruned_loss=0.1519, over 5751146.69 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.4017, pruned_loss=0.1475, over 5694189.29 frames. ], batch size: 262, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:57:47,531 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66475.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 04:58:29,834 INFO [train.py:968] (0/2) Epoch 2, batch 20850, giga_loss[loss=0.3525, simple_loss=0.4081, pruned_loss=0.1484, over 28873.00 frames. ], tot_loss[loss=0.3547, simple_loss=0.4066, pruned_loss=0.1514, over 5711511.26 frames. ], libri_tot_loss[loss=0.3594, simple_loss=0.4136, pruned_loss=0.1526, over 5753270.96 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4031, pruned_loss=0.1487, over 5703820.22 frames. ], batch size: 106, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:58:49,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=66537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:59:00,344 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=66553.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 04:59:07,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.633e+02 1.287e+03 1.739e+03 2.231e+03 1.468e+04, threshold=3.477e+03, percent-clipped=12.0
+2023-03-01 04:59:14,488 INFO [train.py:968] (0/2) Epoch 2, batch 20900, giga_loss[loss=0.3486, simple_loss=0.4012, pruned_loss=0.148, over 28824.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.4065, pruned_loss=0.1519, over 5710485.20 frames. ], libri_tot_loss[loss=0.3595, simple_loss=0.4134, pruned_loss=0.1528, over 5757927.43 frames. ], giga_tot_loss[loss=0.3514, simple_loss=0.4037, pruned_loss=0.1495, over 5698780.72 frames. ], batch size: 119, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:59:14,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2308, 1.5316, 1.1051, 1.3955], device='cuda:0'), covar=tensor([0.1068, 0.0428, 0.0514, 0.1201], device='cuda:0'), in_proj_covar=tensor([0.0290, 0.0194, 0.0197, 0.0350], device='cuda:0'), out_proj_covar=tensor([0.0030, 0.0023, 0.0020, 0.0034], device='cuda:0')
+2023-03-01 04:59:44,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66606.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 04:59:52,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6706, 1.7462, 1.2764, 1.4160], device='cuda:0'), covar=tensor([0.0683, 0.0718, 0.1050, 0.0948], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0506, 0.0547, 0.0472], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 04:59:53,861 INFO [train.py:968] (0/2) Epoch 2, batch 20950, giga_loss[loss=0.3185, simple_loss=0.3764, pruned_loss=0.1303, over 28707.00 frames. ], tot_loss[loss=0.3532, simple_loss=0.406, pruned_loss=0.1502, over 5713990.86 frames. ], libri_tot_loss[loss=0.3597, simple_loss=0.4135, pruned_loss=0.1529, over 5759811.83 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4036, pruned_loss=0.1482, over 5702449.39 frames. ], batch size: 92, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:00:19,016 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=66651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:00:20,763 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 05:00:27,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.785e+02 1.117e+03 1.447e+03 1.886e+03 3.735e+03, threshold=2.893e+03, percent-clipped=1.0
+2023-03-01 05:00:32,856 INFO [train.py:968] (0/2) Epoch 2, batch 21000, giga_loss[loss=0.3694, simple_loss=0.4242, pruned_loss=0.1573, over 28841.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.4069, pruned_loss=0.1493, over 5707565.16 frames. ], libri_tot_loss[loss=0.3609, simple_loss=0.4143, pruned_loss=0.1537, over 5753779.66 frames. ], giga_tot_loss[loss=0.3489, simple_loss=0.4041, pruned_loss=0.1469, over 5703284.83 frames. ], batch size: 199, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:00:32,861 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 05:00:38,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2387, 1.7175, 1.3469, 0.2735], device='cuda:0'), covar=tensor([0.1127, 0.0848, 0.1362, 0.1480], device='cuda:0'), in_proj_covar=tensor([0.1134, 0.1114, 0.1154, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 05:00:41,509 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.2782, simple_loss=0.3748, pruned_loss=0.09082, over 944034.00 frames. 
+2023-03-01 05:00:41,510 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19478MB
+2023-03-01 05:01:05,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7614, 1.4951, 3.6026, 3.0447], device='cuda:0'), covar=tensor([0.1412, 0.1535, 0.0308, 0.0687], device='cuda:0'), in_proj_covar=tensor([0.0491, 0.0482, 0.0616, 0.0509], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 05:01:22,402 INFO [train.py:968] (0/2) Epoch 2, batch 21050, giga_loss[loss=0.3207, simple_loss=0.3873, pruned_loss=0.127, over 28714.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4057, pruned_loss=0.1472, over 5718635.87 frames. ], libri_tot_loss[loss=0.3605, simple_loss=0.4137, pruned_loss=0.1536, over 5758274.42 frames. ], giga_tot_loss[loss=0.347, simple_loss=0.4038, pruned_loss=0.1451, over 5709998.52 frames. ], batch size: 262, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:01:46,745 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66749.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:01:49,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66752.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:01:56,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.834e+02 1.159e+03 1.432e+03 1.947e+03 6.418e+03, threshold=2.864e+03, percent-clipped=4.0
+2023-03-01 05:02:02,926 INFO [train.py:968] (0/2) Epoch 2, batch 21100, giga_loss[loss=0.4361, simple_loss=0.4378, pruned_loss=0.2172, over 26633.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.4026, pruned_loss=0.1449, over 5716157.57 frames. ], libri_tot_loss[loss=0.3605, simple_loss=0.4138, pruned_loss=0.1537, over 5759006.05 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.4011, pruned_loss=0.1433, over 5708658.95 frames. ], batch size: 555, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:02:13,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66781.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:02:39,355 INFO [train.py:968] (0/2) Epoch 2, batch 21150, giga_loss[loss=0.3233, simple_loss=0.3845, pruned_loss=0.1311, over 28653.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.4009, pruned_loss=0.1449, over 5716960.47 frames. ], libri_tot_loss[loss=0.362, simple_loss=0.4143, pruned_loss=0.1548, over 5760364.74 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3987, pruned_loss=0.1422, over 5708475.65 frames. ], batch size: 78, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:03:15,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.335e+02 1.190e+03 1.500e+03 2.287e+03 6.551e+03, threshold=2.999e+03, percent-clipped=16.0
+2023-03-01 05:03:20,229 INFO [train.py:968] (0/2) Epoch 2, batch 21200, giga_loss[loss=0.3229, simple_loss=0.3813, pruned_loss=0.1323, over 28944.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3987, pruned_loss=0.143, over 5716797.85 frames. ], libri_tot_loss[loss=0.3626, simple_loss=0.4147, pruned_loss=0.1553, over 5764407.05 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3963, pruned_loss=0.1402, over 5705493.91 frames. ], batch size: 186, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:03:57,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:03:58,505 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=66914.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:04:04,213 INFO [train.py:968] (0/2) Epoch 2, batch 21250, giga_loss[loss=0.3595, simple_loss=0.4116, pruned_loss=0.1537, over 28769.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3981, pruned_loss=0.1427, over 5721414.24 frames. ], libri_tot_loss[loss=0.3633, simple_loss=0.4151, pruned_loss=0.1557, over 5763383.08 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3955, pruned_loss=0.1399, over 5712647.23 frames. ], batch size: 284, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:04:12,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66928.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:04:40,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.292e+02 1.060e+03 1.282e+03 1.856e+03 9.473e+03, threshold=2.565e+03, percent-clipped=4.0
+2023-03-01 05:04:45,514 INFO [train.py:968] (0/2) Epoch 2, batch 21300, giga_loss[loss=0.3021, simple_loss=0.3658, pruned_loss=0.1192, over 28712.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.4001, pruned_loss=0.1448, over 5713676.34 frames. ], libri_tot_loss[loss=0.3636, simple_loss=0.4153, pruned_loss=0.1559, over 5762982.97 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3974, pruned_loss=0.142, over 5705666.18 frames. ], batch size: 92, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:05:19,657 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 05:05:25,577 INFO [train.py:968] (0/2) Epoch 2, batch 21350, libri_loss[loss=0.3396, simple_loss=0.3788, pruned_loss=0.1501, over 29516.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.4, pruned_loss=0.1439, over 5720184.29 frames. ], libri_tot_loss[loss=0.3643, simple_loss=0.4155, pruned_loss=0.1565, over 5766016.56 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3971, pruned_loss=0.1407, over 5709355.33 frames. ], batch size: 70, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:05:31,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:05:55,824 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:05:57,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:06:01,295 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.296e+02 1.031e+03 1.404e+03 1.868e+03 7.220e+03, threshold=2.808e+03, percent-clipped=13.0
+2023-03-01 05:06:06,587 INFO [train.py:968] (0/2) Epoch 2, batch 21400, libri_loss[loss=0.4172, simple_loss=0.4566, pruned_loss=0.1889, over 29416.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3987, pruned_loss=0.142, over 5710953.12 frames. ], libri_tot_loss[loss=0.3647, simple_loss=0.4159, pruned_loss=0.1568, over 5758546.10 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3958, pruned_loss=0.139, over 5707814.81 frames. ], batch size: 92, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:06:08,890 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67071.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:06:10,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67074.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:06:21,781 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:06:35,124 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67103.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:06:47,354 INFO [train.py:968] (0/2) Epoch 2, batch 21450, giga_loss[loss=0.3294, simple_loss=0.3871, pruned_loss=0.1358, over 28957.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3981, pruned_loss=0.1419, over 5704195.52 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.416, pruned_loss=0.1571, over 5757075.60 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3955, pruned_loss=0.139, over 5702006.86 frames. ], batch size: 227, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:06:52,233 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=67124.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:07:18,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4140, 2.0038, 1.6223, 0.5033], device='cuda:0'), covar=tensor([0.1670, 0.0943, 0.1308, 0.2048], device='cuda:0'), in_proj_covar=tensor([0.1130, 0.1104, 0.1139, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 05:07:22,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.551e+02 1.051e+03 1.424e+03 1.846e+03 9.481e+03, threshold=2.849e+03, percent-clipped=9.0
+2023-03-01 05:07:26,183 INFO [train.py:968] (0/2) Epoch 2, batch 21500, giga_loss[loss=0.3123, simple_loss=0.3682, pruned_loss=0.1282, over 28606.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3963, pruned_loss=0.1419, over 5703434.93 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4156, pruned_loss=0.1572, over 5761646.00 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3941, pruned_loss=0.139, over 5696207.20 frames. ], batch size: 78, lr: 1.34e-02, grad_scale: 2.0
+2023-03-01 05:07:26,459 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:07:28,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:07:53,949 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:08:09,071 INFO [train.py:968] (0/2) Epoch 2, batch 21550, giga_loss[loss=0.3197, simple_loss=0.3853, pruned_loss=0.127, over 28761.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3964, pruned_loss=0.143, over 5698308.28 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4167, pruned_loss=0.1588, over 5751001.33 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3925, pruned_loss=0.1384, over 5699122.71 frames. ], batch size: 66, lr: 1.34e-02, grad_scale: 2.0
+2023-03-01 05:08:17,308 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-01 05:08:46,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.238e+02 1.140e+03 1.377e+03 2.058e+03 1.463e+04, threshold=2.753e+03, percent-clipped=14.0
+2023-03-01 05:08:51,473 INFO [train.py:968] (0/2) Epoch 2, batch 21600, giga_loss[loss=0.433, simple_loss=0.4455, pruned_loss=0.2103, over 26588.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3922, pruned_loss=0.1404, over 5694889.47 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4167, pruned_loss=0.1589, over 5752135.45 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3888, pruned_loss=0.1365, over 5693874.02 frames. ], batch size: 555, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:08:56,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4793, 1.4485, 1.0684, 1.2092], device='cuda:0'), covar=tensor([0.0542, 0.0524, 0.0990, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0501, 0.0543, 0.0470], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 05:09:07,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67289.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:09:11,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1616, 1.5659, 1.1266, 1.3109], device='cuda:0'), covar=tensor([0.1222, 0.0429, 0.0523, 0.1554], device='cuda:0'), in_proj_covar=tensor([0.0288, 0.0195, 0.0198, 0.0354], device='cuda:0'), out_proj_covar=tensor([0.0030, 0.0023, 0.0020, 0.0034], device='cuda:0')
+2023-03-01 05:09:30,372 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-01 05:09:30,489 INFO [train.py:968] (0/2) Epoch 2, batch 21650, libri_loss[loss=0.334, simple_loss=0.381, pruned_loss=0.1435, over 28087.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3928, pruned_loss=0.1413, over 5694727.94 frames. ], libri_tot_loss[loss=0.3679, simple_loss=0.417, pruned_loss=0.1594, over 5751567.21 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3892, pruned_loss=0.1372, over 5692951.11 frames. ], batch size: 62, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:09:57,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=67350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:10:08,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.935e+02 1.203e+03 1.392e+03 1.744e+03 5.634e+03, threshold=2.784e+03, percent-clipped=9.0
+2023-03-01 05:10:11,733 INFO [train.py:968] (0/2) Epoch 2, batch 21700, giga_loss[loss=0.3315, simple_loss=0.3845, pruned_loss=0.1392, over 28725.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3929, pruned_loss=0.142, over 5695855.93 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4179, pruned_loss=0.1602, over 5753015.75 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3887, pruned_loss=0.1375, over 5691892.30 frames. ], batch size: 92, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:10:52,423 INFO [train.py:968] (0/2) Epoch 2, batch 21750, giga_loss[loss=0.2825, simple_loss=0.3441, pruned_loss=0.1104, over 29046.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3905, pruned_loss=0.1408, over 5678357.22 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.4177, pruned_loss=0.1603, over 5728887.02 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3862, pruned_loss=0.1364, over 5695271.38 frames. ], batch size: 128, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:11:03,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67432.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:11:05,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67435.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:11:13,071 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-01 05:11:27,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.230e+02 1.288e+03 1.691e+03 2.196e+03 5.586e+03, threshold=3.383e+03, percent-clipped=17.0
+2023-03-01 05:11:28,047 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67464.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:11:32,071 INFO [train.py:968] (0/2) Epoch 2, batch 21800, giga_loss[loss=0.298, simple_loss=0.3573, pruned_loss=0.1193, over 28982.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3887, pruned_loss=0.1406, over 5688329.79 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4184, pruned_loss=0.161, over 5731105.88 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3842, pruned_loss=0.1361, over 5699034.80 frames. ], batch size: 106, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:11:37,782 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=67476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:11:47,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1952, 1.2973, 1.1841, 0.8791], device='cuda:0'), covar=tensor([0.0530, 0.0362, 0.0328, 0.0487], device='cuda:0'), in_proj_covar=tensor([0.1064, 0.0730, 0.0846, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 05:11:56,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67499.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:12:14,573 INFO [train.py:968] (0/2) Epoch 2, batch 21850, giga_loss[loss=0.3302, simple_loss=0.3858, pruned_loss=0.1373, over 28629.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3866, pruned_loss=0.1397, over 5702263.11 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4187, pruned_loss=0.1614, over 5733765.57 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.382, pruned_loss=0.1353, over 5707400.89 frames. ], batch size: 242, lr: 1.33e-02, grad_scale: 2.0
+2023-03-01 05:12:50,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.659e+02 1.185e+03 1.498e+03 2.133e+03 1.083e+04, threshold=2.996e+03, percent-clipped=8.0
+2023-03-01 05:12:55,954 INFO [train.py:968] (0/2) Epoch 2, batch 21900, libri_loss[loss=0.4211, simple_loss=0.4491, pruned_loss=0.1966, over 29526.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3863, pruned_loss=0.1397, over 5707409.52 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.4198, pruned_loss=0.1624, over 5737567.13 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3809, pruned_loss=0.1348, over 5707270.90 frames. ], batch size: 84, lr: 1.33e-02, grad_scale: 2.0
+2023-03-01 05:13:16,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0908, 0.8968, 0.7478, 1.2037], device='cuda:0'), covar=tensor([0.1051, 0.0464, 0.0556, 0.1252], device='cuda:0'), in_proj_covar=tensor([0.0289, 0.0197, 0.0198, 0.0352], device='cuda:0'), out_proj_covar=tensor([0.0030, 0.0023, 0.0020, 0.0034], device='cuda:0')
+2023-03-01 05:13:39,967 INFO [train.py:968] (0/2) Epoch 2, batch 21950, giga_loss[loss=0.3232, simple_loss=0.3865, pruned_loss=0.1299, over 28797.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3889, pruned_loss=0.1406, over 5702264.04 frames. ], libri_tot_loss[loss=0.3731, simple_loss=0.4204, pruned_loss=0.1629, over 5730617.40 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3835, pruned_loss=0.1359, over 5707895.29 frames. ], batch size: 285, lr: 1.33e-02, grad_scale: 2.0
+2023-03-01 05:14:00,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67642.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:14:03,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67645.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:14:13,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0034, 1.2769, 1.0181, 0.1002], device='cuda:0'), covar=tensor([0.0954, 0.0934, 0.1452, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.1120, 0.1106, 0.1149, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 05:14:20,106 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.942e+02 1.121e+03 1.446e+03 1.956e+03 6.706e+03, threshold=2.892e+03, percent-clipped=6.0
+2023-03-01 05:14:23,034 INFO [train.py:968] (0/2) Epoch 2, batch 22000, libri_loss[loss=0.3643, simple_loss=0.4012, pruned_loss=0.1637, over 29547.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3931, pruned_loss=0.1434, over 5694616.76 frames. ], libri_tot_loss[loss=0.3751, simple_loss=0.4216, pruned_loss=0.1642, over 5732731.31 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3867, pruned_loss=0.1376, over 5695940.67 frames. ], batch size: 76, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:14:29,464 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67674.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:14:49,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0889, 1.2656, 0.9546, 0.3309], device='cuda:0'), covar=tensor([0.0914, 0.0866, 0.1485, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.1129, 0.1110, 0.1157, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 05:15:06,204 INFO [train.py:968] (0/2) Epoch 2, batch 22050, giga_loss[loss=0.3521, simple_loss=0.4007, pruned_loss=0.1518, over 28889.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3953, pruned_loss=0.144, over 5695506.92 frames. ], libri_tot_loss[loss=0.376, simple_loss=0.4219, pruned_loss=0.165, over 5733169.06 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3891, pruned_loss=0.138, over 5694772.12 frames. ], batch size: 106, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:15:12,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:15:48,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.560e+02 1.054e+03 1.378e+03 2.069e+03 8.291e+03, threshold=2.756e+03, percent-clipped=8.0
+2023-03-01 05:15:51,385 INFO [train.py:968] (0/2) Epoch 2, batch 22100, giga_loss[loss=0.2921, simple_loss=0.3625, pruned_loss=0.1109, over 28738.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3948, pruned_loss=0.1424, over 5699412.19 frames. ], libri_tot_loss[loss=0.3769, simple_loss=0.4225, pruned_loss=0.1656, over 5734472.05 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3891, pruned_loss=0.137, over 5697334.92 frames. ], batch size: 119, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:16:15,923 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-01 05:16:37,737 INFO [train.py:968] (0/2) Epoch 2, batch 22150, giga_loss[loss=0.3197, simple_loss=0.3745, pruned_loss=0.1325, over 28845.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3929, pruned_loss=0.1403, over 5692457.15 frames. ], libri_tot_loss[loss=0.3778, simple_loss=0.4231, pruned_loss=0.1662, over 5727978.61 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3874, pruned_loss=0.135, over 5695702.46 frames. ], batch size: 112, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:17:04,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:17:14,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.009e+02 1.137e+03 1.426e+03 2.034e+03 1.354e+04, threshold=2.851e+03, percent-clipped=15.0
+2023-03-01 05:17:17,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:17:18,012 INFO [train.py:968] (0/2) Epoch 2, batch 22200, giga_loss[loss=0.3221, simple_loss=0.381, pruned_loss=0.1316, over 28859.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3942, pruned_loss=0.1421, over 5697457.20 frames. ], libri_tot_loss[loss=0.3781, simple_loss=0.4231, pruned_loss=0.1666, over 5734493.60 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3886, pruned_loss=0.1365, over 5693072.61 frames. ], batch size: 145, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:17:19,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:17:44,320 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:17:51,347 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=67906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:18:02,000 INFO [train.py:968] (0/2) Epoch 2, batch 22250, giga_loss[loss=0.2914, simple_loss=0.3608, pruned_loss=0.111, over 29020.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3935, pruned_loss=0.1416, over 5703153.48 frames. ], libri_tot_loss[loss=0.3784, simple_loss=0.4233, pruned_loss=0.1667, over 5735855.12 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3887, pruned_loss=0.1369, over 5698208.09 frames. ], batch size: 155, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:18:40,815 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.831e+02 1.262e+03 1.520e+03 1.926e+03 5.137e+03, threshold=3.041e+03, percent-clipped=7.0
+2023-03-01 05:18:45,289 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-01 05:18:46,456 INFO [train.py:968] (0/2) Epoch 2, batch 22300, giga_loss[loss=0.3372, simple_loss=0.4003, pruned_loss=0.1371, over 29039.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3964, pruned_loss=0.144, over 5702822.13 frames. ], libri_tot_loss[loss=0.379, simple_loss=0.4236, pruned_loss=0.1672, over 5740372.73 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3916, pruned_loss=0.1393, over 5694235.01 frames. ], batch size: 128, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:19:07,335 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:19:09,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:19:11,648 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-68000.pt
+2023-03-01 05:19:27,317 INFO [train.py:968] (0/2) Epoch 2, batch 22350, giga_loss[loss=0.3247, simple_loss=0.3805, pruned_loss=0.1344, over 28059.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.4007, pruned_loss=0.1465, over 5712789.65 frames. ], libri_tot_loss[loss=0.3803, simple_loss=0.4246, pruned_loss=0.168, over 5743213.59 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3953, pruned_loss=0.1414, over 5702460.99 frames. ], batch size: 77, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:19:32,706 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:20:05,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.920e+02 1.369e+03 1.731e+03 2.748e+03 7.498e+03, threshold=3.462e+03, percent-clipped=16.0
+2023-03-01 05:20:09,018 INFO [train.py:968] (0/2) Epoch 2, batch 22400, giga_loss[loss=0.4311, simple_loss=0.4545, pruned_loss=0.2039, over 26499.00 frames. ], tot_loss[loss=0.3502, simple_loss=0.4032, pruned_loss=0.1485, over 5712440.73 frames. ], libri_tot_loss[loss=0.3819, simple_loss=0.4255, pruned_loss=0.1691, over 5743395.67 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3975, pruned_loss=0.1429, over 5703362.84 frames. ], batch size: 555, lr: 1.33e-02, grad_scale: 8.0
+2023-03-01 05:20:11,776 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:20:21,779 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:20:41,160 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-01 05:20:46,537 INFO [train.py:968] (0/2) Epoch 2, batch 22450, giga_loss[loss=0.3888, simple_loss=0.4345, pruned_loss=0.1715, over 29025.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4046, pruned_loss=0.1488, over 5713421.65 frames. ], libri_tot_loss[loss=0.3831, simple_loss=0.4265, pruned_loss=0.1699, over 5737381.64 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3981, pruned_loss=0.1426, over 5709718.33 frames. ], batch size: 155, lr: 1.33e-02, grad_scale: 8.0
+2023-03-01 05:20:50,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2683, 3.6554, 2.1899, 1.9284], device='cuda:0'), covar=tensor([0.0705, 0.0350, 0.0764, 0.1108], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0439, 0.0335, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0012, 0.0015], device='cuda:0')
+2023-03-01 05:21:02,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0755, 2.5204, 1.0876, 0.9824], device='cuda:0'), covar=tensor([0.0703, 0.0378, 0.0575, 0.0695], device='cuda:0'), in_proj_covar=tensor([0.1043, 0.0730, 0.0852, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 05:21:24,549 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.133e+02 1.332e+03 1.687e+03 2.060e+03 4.988e+03, threshold=3.374e+03, percent-clipped=3.0
+2023-03-01 05:21:26,682 INFO [train.py:968] (0/2) Epoch 2, batch 22500, giga_loss[loss=0.3237, simple_loss=0.3739, pruned_loss=0.1367, over 28689.00 frames. ], tot_loss[loss=0.348, simple_loss=0.4024, pruned_loss=0.1468, over 5723943.56 frames. ], libri_tot_loss[loss=0.3838, simple_loss=0.4268, pruned_loss=0.1704, over 5742128.02 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3963, pruned_loss=0.1408, over 5716316.61 frames. ], batch size: 85, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:22:10,797 INFO [train.py:968] (0/2) Epoch 2, batch 22550, giga_loss[loss=0.293, simple_loss=0.3625, pruned_loss=0.1118, over 28844.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4031, pruned_loss=0.1471, over 5719230.78 frames. ], libri_tot_loss[loss=0.3849, simple_loss=0.4276, pruned_loss=0.1711, over 5743072.71 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3968, pruned_loss=0.1408, over 5711281.81 frames. ], batch size: 119, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:22:50,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.866e+02 1.251e+03 1.696e+03 2.277e+03 6.177e+03, threshold=3.392e+03, percent-clipped=9.0
+2023-03-01 05:22:53,650 INFO [train.py:968] (0/2) Epoch 2, batch 22600, giga_loss[loss=0.3214, simple_loss=0.3827, pruned_loss=0.13, over 28674.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.4025, pruned_loss=0.1472, over 5719254.94 frames. ], libri_tot_loss[loss=0.3852, simple_loss=0.4277, pruned_loss=0.1714, over 5746856.75 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3967, pruned_loss=0.1414, over 5708924.66 frames. ], batch size: 262, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:23:02,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5014, 1.6746, 1.2903, 0.9950], device='cuda:0'), covar=tensor([0.0708, 0.0418, 0.0405, 0.0577], device='cuda:0'), in_proj_covar=tensor([0.1051, 0.0735, 0.0858, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 05:23:03,446 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:23:37,308 INFO [train.py:968] (0/2) Epoch 2, batch 22650, giga_loss[loss=0.3268, simple_loss=0.3872, pruned_loss=0.1332, over 28592.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.3995, pruned_loss=0.1454, over 5722853.34 frames. ], libri_tot_loss[loss=0.3862, simple_loss=0.4281, pruned_loss=0.1721, over 5747951.70 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.394, pruned_loss=0.1396, over 5713374.52 frames. ], batch size: 336, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:23:39,206 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4255, 2.4358, 1.3886, 1.3874], device='cuda:0'), covar=tensor([0.0890, 0.0507, 0.0925, 0.1434], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0444, 0.0339, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0016, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 05:24:15,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.892e+02 1.128e+03 1.465e+03 1.915e+03 7.026e+03, threshold=2.930e+03, percent-clipped=6.0
+2023-03-01 05:24:16,853 INFO [train.py:968] (0/2) Epoch 2, batch 22700, giga_loss[loss=0.3215, simple_loss=0.3837, pruned_loss=0.1296, over 28568.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3955, pruned_loss=0.1429, over 5725229.43 frames. ], libri_tot_loss[loss=0.3864, simple_loss=0.4281, pruned_loss=0.1723, over 5753107.52 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.39, pruned_loss=0.1371, over 5712034.61 frames. ], batch size: 336, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:24:20,901 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:24:46,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0121, 1.8747, 1.2605, 1.4287], device='cuda:0'), covar=tensor([0.0523, 0.0558, 0.0934, 0.0816], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0507, 0.0547, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 05:24:55,644 INFO [train.py:968] (0/2) Epoch 2, batch 22750, giga_loss[loss=0.3539, simple_loss=0.4094, pruned_loss=0.1492, over 28905.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3955, pruned_loss=0.1435, over 5729724.12 frames. ], libri_tot_loss[loss=0.3871, simple_loss=0.428, pruned_loss=0.1731, over 5757839.15 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3894, pruned_loss=0.1365, over 5713023.90 frames. ], batch size: 174, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:25:00,660 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:25:03,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:25:23,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:25:30,965 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:25:33,559 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:25:38,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.849e+02 1.105e+03 1.463e+03 2.021e+03 9.817e+03, threshold=2.925e+03, percent-clipped=12.0
+2023-03-01 05:25:40,361 INFO [train.py:968] (0/2) Epoch 2, batch 22800, giga_loss[loss=0.3817, simple_loss=0.4387, pruned_loss=0.1624, over 28584.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3969, pruned_loss=0.1425, over 5719744.59 frames. ], libri_tot_loss[loss=0.388, simple_loss=0.4286, pruned_loss=0.1736, over 5754953.30 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3908, pruned_loss=0.1359, over 5708552.44 frames. ], batch size: 307, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:25:51,823 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68480.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:26:13,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4014, 1.9848, 1.5108, 0.5467], device='cuda:0'), covar=tensor([0.1210, 0.0778, 0.1187, 0.1549], device='cuda:0'), in_proj_covar=tensor([0.1144, 0.1105, 0.1160, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 05:26:22,121 INFO [train.py:968] (0/2) Epoch 2, batch 22850, giga_loss[loss=0.3218, simple_loss=0.3912, pruned_loss=0.1262, over 28686.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3986, pruned_loss=0.1428, over 5716545.86 frames. ], libri_tot_loss[loss=0.388, simple_loss=0.4286, pruned_loss=0.1737, over 5750058.39 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3928, pruned_loss=0.1364, over 5711553.07 frames. ], batch size: 262, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:27:02,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.746e+02 1.028e+03 1.356e+03 1.965e+03 7.291e+03, threshold=2.713e+03, percent-clipped=7.0
+2023-03-01 05:27:04,902 INFO [train.py:968] (0/2) Epoch 2, batch 22900, giga_loss[loss=0.3312, simple_loss=0.3924, pruned_loss=0.135, over 28677.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3961, pruned_loss=0.1417, over 5716001.03 frames. ], libri_tot_loss[loss=0.3885, simple_loss=0.4289, pruned_loss=0.1741, over 5743035.97 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3909, pruned_loss=0.136, over 5717734.94 frames. ], batch size: 307, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:27:07,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2356, 1.4397, 1.1652, 1.3346], device='cuda:0'), covar=tensor([0.1027, 0.0450, 0.0482, 0.1166], device='cuda:0'), in_proj_covar=tensor([0.0289, 0.0198, 0.0198, 0.0349], device='cuda:0'), out_proj_covar=tensor([0.0031, 0.0024, 0.0021, 0.0035], device='cuda:0')
+2023-03-01 05:27:25,823 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:27:29,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:27:36,305 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:27:38,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:27:49,654 INFO [train.py:968] (0/2) Epoch 2, batch 22950, libri_loss[loss=0.3897, simple_loss=0.4207, pruned_loss=0.1794, over 29329.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3949, pruned_loss=0.1423, over 5712402.22 frames. ], libri_tot_loss[loss=0.389, simple_loss=0.4293, pruned_loss=0.1744, over 5738998.89 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3895, pruned_loss=0.1366, over 5716763.50 frames. ], batch size: 71, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:27:52,455 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:28:01,608 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:28:29,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.380e+02 1.129e+03 1.340e+03 1.952e+03 5.375e+03, threshold=2.679e+03, percent-clipped=8.0
+2023-03-01 05:28:29,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-01 05:28:30,653 INFO [train.py:968] (0/2) Epoch 2, batch 23000, giga_loss[loss=0.356, simple_loss=0.3964, pruned_loss=0.1578, over 29024.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3922, pruned_loss=0.1423, over 5714483.22 frames. ], libri_tot_loss[loss=0.3884, simple_loss=0.4287, pruned_loss=0.174, over 5740794.46 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.388, pruned_loss=0.1377, over 5715984.01 frames. ], batch size: 164, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:29:02,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7879, 3.1699, 3.4655, 1.5373], device='cuda:0'), covar=tensor([0.0695, 0.0561, 0.1223, 0.2273], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0545, 0.0801, 0.0566], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 05:29:14,351 INFO [train.py:968] (0/2) Epoch 2, batch 23050, giga_loss[loss=0.2995, simple_loss=0.3646, pruned_loss=0.1172, over 28999.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.392, pruned_loss=0.1438, over 5710065.46 frames. ], libri_tot_loss[loss=0.3901, simple_loss=0.4297, pruned_loss=0.1752, over 5731647.37 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3867, pruned_loss=0.1382, over 5718888.91 frames. ], batch size: 164, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:29:40,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:29:54,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.352e+02 1.085e+03 1.400e+03 1.858e+03 7.950e+03, threshold=2.800e+03, percent-clipped=13.0
+2023-03-01 05:29:55,579 INFO [train.py:968] (0/2) Epoch 2, batch 23100, giga_loss[loss=0.331, simple_loss=0.3908, pruned_loss=0.1356, over 28321.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3888, pruned_loss=0.1419, over 5709333.46 frames. ], libri_tot_loss[loss=0.3895, simple_loss=0.4293, pruned_loss=0.1749, over 5733976.08 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3846, pruned_loss=0.1373, over 5714040.92 frames. ], batch size: 368, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:30:08,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8522, 3.2886, 2.3117, 0.6861], device='cuda:0'), covar=tensor([0.2151, 0.0698, 0.1096, 0.1891], device='cuda:0'), in_proj_covar=tensor([0.1133, 0.1111, 0.1166, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 05:30:15,004 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:30:36,306 INFO [train.py:968] (0/2) Epoch 2, batch 23150, giga_loss[loss=0.2975, simple_loss=0.3558, pruned_loss=0.1196, over 28917.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3847, pruned_loss=0.1392, over 5717155.53 frames. ], libri_tot_loss[loss=0.39, simple_loss=0.4295, pruned_loss=0.1753, over 5737131.19 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3803, pruned_loss=0.1346, over 5717602.62 frames. ], batch size: 174, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:31:09,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68855.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:31:18,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.819e+02 1.137e+03 1.459e+03 1.992e+03 5.763e+03, threshold=2.918e+03, percent-clipped=6.0
+2023-03-01 05:31:21,804 INFO [train.py:968] (0/2) Epoch 2, batch 23200, giga_loss[loss=0.2685, simple_loss=0.3249, pruned_loss=0.1061, over 28621.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3787, pruned_loss=0.1354, over 5714151.45 frames. ], libri_tot_loss[loss=0.3904, simple_loss=0.4298, pruned_loss=0.1755, over 5737529.54 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3747, pruned_loss=0.1314, over 5714000.40 frames. ], batch size: 92, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:31:34,436 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-01 05:31:39,481 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:31:41,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:31:59,948 INFO [train.py:968] (0/2) Epoch 2, batch 23250, giga_loss[loss=0.3198, simple_loss=0.3812, pruned_loss=0.1293, over 28869.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3752, pruned_loss=0.133, over 5722094.91 frames. ], libri_tot_loss[loss=0.3903, simple_loss=0.4297, pruned_loss=0.1754, over 5739876.13 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.371, pruned_loss=0.1289, over 5719354.79 frames. ], batch size: 186, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:32:05,334 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:32:44,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.103e+02 1.158e+03 1.397e+03 1.689e+03 3.863e+03, threshold=2.794e+03, percent-clipped=3.0
+2023-03-01 05:32:45,726 INFO [train.py:968] (0/2) Epoch 2, batch 23300, giga_loss[loss=0.35, simple_loss=0.3995, pruned_loss=0.1502, over 29059.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3797, pruned_loss=0.1354, over 5716533.35 frames. ], libri_tot_loss[loss=0.3904, simple_loss=0.4296, pruned_loss=0.1756, over 5742292.33 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3756, pruned_loss=0.1315, over 5711960.17 frames. ], batch size: 128, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:32:47,021 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:33:11,155 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68998.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:33:13,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69001.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:33:28,447 INFO [train.py:968] (0/2) Epoch 2, batch 23350, giga_loss[loss=0.3645, simple_loss=0.4162, pruned_loss=0.1564, over 28557.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3839, pruned_loss=0.1377, over 5703561.77 frames. ], libri_tot_loss[loss=0.3905, simple_loss=0.4296, pruned_loss=0.1757, over 5733370.08 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3796, pruned_loss=0.1336, over 5708077.19 frames. ], batch size: 336, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:33:38,394 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69030.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:33:56,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2061, 1.7688, 1.2644, 1.2097], device='cuda:0'), covar=tensor([0.0929, 0.0603, 0.0877, 0.1407], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0444, 0.0331, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0016, 0.0012, 0.0015], device='cuda:0')
+2023-03-01 05:34:10,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.133e+02 1.277e+03 1.582e+03 2.177e+03 4.171e+03, threshold=3.164e+03, percent-clipped=11.0
+2023-03-01 05:34:10,707 INFO [train.py:968] (0/2) Epoch 2, batch 23400, libri_loss[loss=0.3977, simple_loss=0.4334, pruned_loss=0.181, over 29518.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3888, pruned_loss=0.14, over 5711220.09 frames. ], libri_tot_loss[loss=0.3907, simple_loss=0.4298, pruned_loss=0.1758, over 5736155.17 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3838, pruned_loss=0.1353, over 5711556.31 frames. ], batch size: 81, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:34:18,811 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:34:41,954 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 05:34:53,268 INFO [train.py:968] (0/2) Epoch 2, batch 23450, libri_loss[loss=0.3384, simple_loss=0.3823, pruned_loss=0.1472, over 29395.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3919, pruned_loss=0.1412, over 5715558.09 frames. ], libri_tot_loss[loss=0.3904, simple_loss=0.4295, pruned_loss=0.1756, over 5740361.95 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3872, pruned_loss=0.1369, over 5711309.25 frames. ], batch size: 67, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:35:41,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:35:41,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.693e+02 1.137e+03 1.465e+03 2.013e+03 4.379e+03, threshold=2.929e+03, percent-clipped=7.0
+2023-03-01 05:35:42,743 INFO [train.py:968] (0/2) Epoch 2, batch 23500, giga_loss[loss=0.3352, simple_loss=0.3886, pruned_loss=0.1409, over 28919.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.393, pruned_loss=0.1414, over 5721149.84 frames. ], libri_tot_loss[loss=0.3902, simple_loss=0.4294, pruned_loss=0.1755, over 5741969.04 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3891, pruned_loss=0.1378, over 5716066.95 frames. ], batch size: 66, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:36:28,079 INFO [train.py:968] (0/2) Epoch 2, batch 23550, giga_loss[loss=0.38, simple_loss=0.4258, pruned_loss=0.1671, over 28866.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.4003, pruned_loss=0.1483, over 5709834.85 frames. ], libri_tot_loss[loss=0.3902, simple_loss=0.4294, pruned_loss=0.1755, over 5737089.45 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3963, pruned_loss=0.1446, over 5709411.44 frames. ], batch size: 186, lr: 1.32e-02, grad_scale: 2.0
+2023-03-01 05:37:18,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.981e+03 2.520e+03 3.468e+03 6.962e+03, threshold=5.040e+03, percent-clipped=35.0
+2023-03-01 05:37:18,351 INFO [train.py:968] (0/2) Epoch 2, batch 23600, giga_loss[loss=0.3847, simple_loss=0.4267, pruned_loss=0.1714, over 28744.00 frames. ], tot_loss[loss=0.3602, simple_loss=0.4086, pruned_loss=0.1559, over 5685273.61 frames. ], libri_tot_loss[loss=0.3905, simple_loss=0.4295, pruned_loss=0.1758, over 5720475.57 frames. ], giga_tot_loss[loss=0.3548, simple_loss=0.4049, pruned_loss=0.1523, over 5698331.85 frames. ], batch size: 92, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:38:00,222 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:38:02,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:38:10,049 INFO [train.py:968] (0/2) Epoch 2, batch 23650, giga_loss[loss=0.3951, simple_loss=0.4392, pruned_loss=0.1755, over 28928.00 frames. ], tot_loss[loss=0.372, simple_loss=0.4171, pruned_loss=0.1634, over 5674734.01 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.4296, pruned_loss=0.176, over 5722905.55 frames. ], giga_tot_loss[loss=0.3668, simple_loss=0.4137, pruned_loss=0.16, over 5682129.62 frames. ], batch size: 106, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:38:33,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:38:36,434 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:38:56,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.628e+02 1.761e+03 2.255e+03 2.822e+03 6.309e+03, threshold=4.511e+03, percent-clipped=3.0
+2023-03-01 05:38:56,804 INFO [train.py:968] (0/2) Epoch 2, batch 23700, giga_loss[loss=0.418, simple_loss=0.4546, pruned_loss=0.1907, over 28948.00 frames. ], tot_loss[loss=0.3815, simple_loss=0.4237, pruned_loss=0.1697, over 5681663.11 frames. ], libri_tot_loss[loss=0.3918, simple_loss=0.4302, pruned_loss=0.1766, over 5729253.19 frames. ], giga_tot_loss[loss=0.376, simple_loss=0.4201, pruned_loss=0.166, over 5680414.18 frames. ], batch size: 164, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:39:42,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8619, 2.2536, 1.9219, 1.8490], device='cuda:0'), covar=tensor([0.1389, 0.1465, 0.1022, 0.0730], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0817, 0.0706, 0.0603], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 05:39:48,974 INFO [train.py:968] (0/2) Epoch 2, batch 23750, libri_loss[loss=0.4189, simple_loss=0.4374, pruned_loss=0.2002, over 29625.00 frames. ], tot_loss[loss=0.3931, simple_loss=0.4316, pruned_loss=0.1773, over 5675590.21 frames. ], libri_tot_loss[loss=0.3922, simple_loss=0.4304, pruned_loss=0.177, over 5722424.64 frames. ], giga_tot_loss[loss=0.3882, simple_loss=0.4284, pruned_loss=0.174, over 5680691.49 frames. ], batch size: 69, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:40:05,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:40:23,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:40:39,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.977e+03 2.601e+03 3.605e+03 8.331e+03, threshold=5.203e+03, percent-clipped=8.0
+2023-03-01 05:40:39,481 INFO [train.py:968] (0/2) Epoch 2, batch 23800, giga_loss[loss=0.4591, simple_loss=0.4809, pruned_loss=0.2187, over 28791.00 frames. ], tot_loss[loss=0.3997, simple_loss=0.436, pruned_loss=0.1817, over 5671133.59 frames. ], libri_tot_loss[loss=0.3915, simple_loss=0.4298, pruned_loss=0.1766, over 5726428.18 frames. ], giga_tot_loss[loss=0.3965, simple_loss=0.4341, pruned_loss=0.1794, over 5670442.31 frames. ], batch size: 186, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:40:55,490 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:40:59,843 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:41:24,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7266, 1.4519, 4.1210, 3.2872], device='cuda:0'), covar=tensor([0.1520, 0.1577, 0.0277, 0.0629], device='cuda:0'), in_proj_covar=tensor([0.0492, 0.0484, 0.0637, 0.0516], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 05:41:26,039 INFO [train.py:968] (0/2) Epoch 2, batch 23850, giga_loss[loss=0.3472, simple_loss=0.3987, pruned_loss=0.1479, over 28734.00 frames. ], tot_loss[loss=0.4024, simple_loss=0.4374, pruned_loss=0.1837, over 5666399.88 frames. ], libri_tot_loss[loss=0.3918, simple_loss=0.4301, pruned_loss=0.1768, over 5730850.52 frames. ], giga_tot_loss[loss=0.3998, simple_loss=0.4358, pruned_loss=0.1819, over 5660719.63 frames. ], batch size: 99, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:41:27,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:42:02,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:42:07,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:42:17,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.224e+02 1.592e+03 1.983e+03 2.859e+03 1.272e+04, threshold=3.965e+03, percent-clipped=6.0
+2023-03-01 05:42:17,685 INFO [train.py:968] (0/2) Epoch 2, batch 23900, libri_loss[loss=0.3854, simple_loss=0.4283, pruned_loss=0.1713, over 29557.00 frames. ], tot_loss[loss=0.4092, simple_loss=0.4414, pruned_loss=0.1885, over 5659046.45 frames. ], libri_tot_loss[loss=0.3921, simple_loss=0.4302, pruned_loss=0.177, over 5726830.71 frames. ], giga_tot_loss[loss=0.4072, simple_loss=0.4402, pruned_loss=0.1871, over 5656144.86 frames. ], batch size: 79, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:42:38,852 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-01 05:42:44,488 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:42:49,469 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:43:08,260 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:43:08,663 INFO [train.py:968] (0/2) Epoch 2, batch 23950, giga_loss[loss=0.5001, simple_loss=0.4977, pruned_loss=0.2512, over 27509.00 frames. ], tot_loss[loss=0.4164, simple_loss=0.4458, pruned_loss=0.1936, over 5646993.95 frames. ], libri_tot_loss[loss=0.3924, simple_loss=0.4305, pruned_loss=0.1772, over 5722007.73 frames. ], giga_tot_loss[loss=0.415, simple_loss=0.4449, pruned_loss=0.1925, over 5647217.61 frames. ], batch size: 472, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:43:21,136 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:43:29,950 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69638.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:43:55,179 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 05:44:03,179 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:44:06,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.314e+02 1.774e+03 2.292e+03 3.211e+03 6.742e+03, threshold=4.584e+03, percent-clipped=11.0
+2023-03-01 05:44:06,509 INFO [train.py:968] (0/2) Epoch 2, batch 24000, giga_loss[loss=0.5732, simple_loss=0.545, pruned_loss=0.3007, over 26509.00 frames. ], tot_loss[loss=0.4228, simple_loss=0.4507, pruned_loss=0.1975, over 5633605.01 frames. ], libri_tot_loss[loss=0.3927, simple_loss=0.4305, pruned_loss=0.1774, over 5708605.48 frames. ], giga_tot_loss[loss=0.4221, simple_loss=0.4504, pruned_loss=0.1969, over 5643106.89 frames. ], batch size: 555, lr: 1.31e-02, grad_scale: 8.0
+2023-03-01 05:44:06,514 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 05:44:14,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1865, 1.3387, 1.1539, 1.0692], device='cuda:0'), covar=tensor([0.1959, 0.2035, 0.1786, 0.1857], device='cuda:0'), in_proj_covar=tensor([0.0961, 0.0815, 0.0882, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 05:44:15,957 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.275, simple_loss=0.37, pruned_loss=0.08995, over 944034.00 frames. 
+2023-03-01 05:44:15,958 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19478MB
+2023-03-01 05:45:06,297 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:45:11,156 INFO [train.py:968] (0/2) Epoch 2, batch 24050, giga_loss[loss=0.368, simple_loss=0.4137, pruned_loss=0.1611, over 28491.00 frames. ], tot_loss[loss=0.4217, simple_loss=0.4496, pruned_loss=0.1968, over 5640008.67 frames. ], libri_tot_loss[loss=0.3931, simple_loss=0.4309, pruned_loss=0.1776, over 5712741.59 frames. ], giga_tot_loss[loss=0.4213, simple_loss=0.4494, pruned_loss=0.1966, over 5642356.66 frames. ], batch size: 60, lr: 1.31e-02, grad_scale: 8.0
+2023-03-01 05:45:31,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2918, 1.4127, 1.3839, 1.2576], device='cuda:0'), covar=tensor([0.0690, 0.0852, 0.1063, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0472, 0.0832, 0.0643, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 05:45:34,382 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 05:45:52,200 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:45:59,360 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 05:46:00,285 INFO [train.py:968] (0/2) Epoch 2, batch 24100, giga_loss[loss=0.3843, simple_loss=0.424, pruned_loss=0.1723, over 28896.00 frames. ], tot_loss[loss=0.4187, simple_loss=0.4471, pruned_loss=0.1951, over 5644897.35 frames. ], libri_tot_loss[loss=0.3931, simple_loss=0.4309, pruned_loss=0.1776, over 5715505.72 frames. ], giga_tot_loss[loss=0.4193, simple_loss=0.4475, pruned_loss=0.1955, over 5641772.42 frames. ], batch size: 199, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:46:01,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.750e+03 2.115e+03 3.081e+03 8.287e+03, threshold=4.229e+03, percent-clipped=7.0
+2023-03-01 05:46:20,356 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:46:45,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:46:54,163 INFO [train.py:968] (0/2) Epoch 2, batch 24150, giga_loss[loss=0.4716, simple_loss=0.4775, pruned_loss=0.2329, over 26674.00 frames. ], tot_loss[loss=0.4169, simple_loss=0.4461, pruned_loss=0.1939, over 5643590.20 frames. ], libri_tot_loss[loss=0.3924, simple_loss=0.4304, pruned_loss=0.1772, over 5718102.24 frames. ], giga_tot_loss[loss=0.4183, simple_loss=0.4471, pruned_loss=0.1948, over 5637975.82 frames. ], batch size: 555, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:47:51,188 INFO [train.py:968] (0/2) Epoch 2, batch 24200, giga_loss[loss=0.4107, simple_loss=0.4502, pruned_loss=0.1857, over 29000.00 frames. ], tot_loss[loss=0.415, simple_loss=0.4456, pruned_loss=0.1922, over 5643906.31 frames. ], libri_tot_loss[loss=0.3925, simple_loss=0.4304, pruned_loss=0.1773, over 5719836.27 frames. ], giga_tot_loss[loss=0.4163, simple_loss=0.4465, pruned_loss=0.193, over 5637254.91 frames. ], batch size: 213, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:47:52,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.511e+03 2.004e+03 2.668e+03 6.289e+03, threshold=4.008e+03, percent-clipped=5.0
+2023-03-01 05:48:43,296 INFO [train.py:968] (0/2) Epoch 2, batch 24250, giga_loss[loss=0.4013, simple_loss=0.4453, pruned_loss=0.1787, over 28866.00 frames. ], tot_loss[loss=0.4163, simple_loss=0.4467, pruned_loss=0.1929, over 5632871.24 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4313, pruned_loss=0.1782, over 5723505.77 frames. ], giga_tot_loss[loss=0.4165, simple_loss=0.447, pruned_loss=0.193, over 5622177.08 frames. ], batch size: 199, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:48:56,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:49:02,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:49:14,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5019, 1.5929, 1.4343, 0.8794], device='cuda:0'), covar=tensor([0.0579, 0.0368, 0.0380, 0.0544], device='cuda:0'), in_proj_covar=tensor([0.1056, 0.0760, 0.0871, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 05:49:17,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2070, 1.6103, 1.2362, 1.3032], device='cuda:0'), covar=tensor([0.1013, 0.0385, 0.0468, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.0292, 0.0195, 0.0199, 0.0353], device='cuda:0'), out_proj_covar=tensor([0.0032, 0.0024, 0.0021, 0.0036], device='cuda:0')
+2023-03-01 05:49:21,076 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:49:24,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:49:33,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3779, 2.8552, 1.7519, 1.3127], device='cuda:0'), covar=tensor([0.0918, 0.0329, 0.0436, 0.0771], device='cuda:0'), in_proj_covar=tensor([0.1053, 0.0754, 0.0862, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 05:49:35,982 INFO [train.py:968] (0/2) Epoch 2, batch 24300, giga_loss[loss=0.492, simple_loss=0.4817, pruned_loss=0.2511, over 23797.00 frames. ], tot_loss[loss=0.4118, simple_loss=0.4436, pruned_loss=0.19, over 5639950.16 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4311, pruned_loss=0.1784, over 5727222.24 frames. ], giga_tot_loss[loss=0.4123, simple_loss=0.4442, pruned_loss=0.1902, over 5626648.68 frames. ], batch size: 705, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:49:37,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.938e+02 1.711e+03 2.286e+03 3.407e+03 1.263e+04, threshold=4.573e+03, percent-clipped=14.0
+2023-03-01 05:50:00,461 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:50:06,378 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:50:12,265 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-70000.pt
+2023-03-01 05:50:27,488 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70013.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:50:31,009 INFO [train.py:968] (0/2) Epoch 2, batch 24350, giga_loss[loss=0.3677, simple_loss=0.4238, pruned_loss=0.1558, over 28891.00 frames. ], tot_loss[loss=0.4055, simple_loss=0.4403, pruned_loss=0.1853, over 5634021.58 frames. ], libri_tot_loss[loss=0.3941, simple_loss=0.4312, pruned_loss=0.1785, over 5720889.94 frames. ], giga_tot_loss[loss=0.4061, simple_loss=0.441, pruned_loss=0.1856, over 5626753.52 frames. ], batch size: 164, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:50:49,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-01 05:50:55,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:09,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6598, 3.9937, 4.3427, 1.7399], device='cuda:0'), covar=tensor([0.0531, 0.0511, 0.1202, 0.1879], device='cuda:0'), in_proj_covar=tensor([0.0764, 0.0547, 0.0814, 0.0562], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0007], device='cuda:0')
+2023-03-01 05:51:20,072 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=70067.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:51:21,652 INFO [train.py:968] (0/2) Epoch 2, batch 24400, giga_loss[loss=0.3967, simple_loss=0.4452, pruned_loss=0.1741, over 28917.00 frames. ], tot_loss[loss=0.3998, simple_loss=0.4367, pruned_loss=0.1814, over 5643009.58 frames. ], libri_tot_loss[loss=0.3938, simple_loss=0.4308, pruned_loss=0.1784, over 5715503.93 frames. ], giga_tot_loss[loss=0.4008, simple_loss=0.4378, pruned_loss=0.1819, over 5639648.92 frames. ], batch size: 186, lr: 1.31e-02, grad_scale: 8.0
+2023-03-01 05:51:22,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.826e+02 1.547e+03 2.017e+03 2.711e+03 6.200e+03, threshold=4.034e+03, percent-clipped=4.0
+2023-03-01 05:51:25,816 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:25,943 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-01 05:51:30,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:32,264 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:34,923 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:35,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4241, 1.6857, 1.3359, 1.4983], device='cuda:0'), covar=tensor([0.0943, 0.0343, 0.0476, 0.1156], device='cuda:0'), in_proj_covar=tensor([0.0291, 0.0196, 0.0199, 0.0352], device='cuda:0'), out_proj_covar=tensor([0.0032, 0.0024, 0.0021, 0.0036], device='cuda:0')
+2023-03-01 05:51:40,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:40,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4214, 1.5440, 1.3689, 1.1455], device='cuda:0'), covar=tensor([0.0614, 0.0412, 0.0306, 0.0476], device='cuda:0'), in_proj_covar=tensor([0.1041, 0.0748, 0.0844, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 05:51:56,953 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:03,455 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:11,723 INFO [train.py:968] (0/2) Epoch 2, batch 24450, giga_loss[loss=0.4003, simple_loss=0.4334, pruned_loss=0.1836, over 27661.00 frames. ], tot_loss[loss=0.3942, simple_loss=0.433, pruned_loss=0.1777, over 5658348.96 frames. ], libri_tot_loss[loss=0.3937, simple_loss=0.4305, pruned_loss=0.1785, over 5716306.44 frames. ], giga_tot_loss[loss=0.3952, simple_loss=0.4344, pruned_loss=0.1781, over 5653251.12 frames. ], batch size: 472, lr: 1.31e-02, grad_scale: 8.0
+2023-03-01 05:52:28,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:28,077 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:31,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:47,902 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70156.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:52:50,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70159.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:52:51,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1021, 1.0202, 0.7600, 1.1663], device='cuda:0'), covar=tensor([0.1037, 0.0412, 0.0501, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0294, 0.0196, 0.0199, 0.0354], device='cuda:0'), out_proj_covar=tensor([0.0032, 0.0024, 0.0021, 0.0036], device='cuda:0')
+2023-03-01 05:52:52,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:59,884 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:53:00,263 INFO [train.py:968] (0/2) Epoch 2, batch 24500, giga_loss[loss=0.3982, simple_loss=0.4126, pruned_loss=0.1919, over 23589.00 frames. ], tot_loss[loss=0.3907, simple_loss=0.4302, pruned_loss=0.1756, over 5654527.00 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4306, pruned_loss=0.1786, over 5718642.40 frames. ], giga_tot_loss[loss=0.3913, simple_loss=0.4312, pruned_loss=0.1757, over 5646624.14 frames. ], batch size: 705, lr: 1.31e-02, grad_scale: 8.0
+2023-03-01 05:53:00,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.129e+02 1.665e+03 2.101e+03 2.613e+03 5.089e+03, threshold=4.203e+03, percent-clipped=5.0
+2023-03-01 05:53:10,106 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9738, 3.4000, 3.6947, 1.8112], device='cuda:0'), covar=tensor([0.0507, 0.0449, 0.0861, 0.1766], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0551, 0.0830, 0.0568], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 05:53:15,587 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:53:18,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:53:19,324 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70188.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:53:46,937 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:53:50,214 INFO [train.py:968] (0/2) Epoch 2, batch 24550, giga_loss[loss=0.3621, simple_loss=0.4131, pruned_loss=0.1555, over 28577.00 frames. ], tot_loss[loss=0.3877, simple_loss=0.4284, pruned_loss=0.1735, over 5671656.96 frames. ], libri_tot_loss[loss=0.3932, simple_loss=0.43, pruned_loss=0.1782, over 5720479.04 frames. ], giga_tot_loss[loss=0.3886, simple_loss=0.4297, pruned_loss=0.1738, over 5662765.88 frames. ], batch size: 336, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:54:02,185 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:54:05,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:54:11,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6091, 2.1517, 1.6225, 0.8407], device='cuda:0'), covar=tensor([0.1865, 0.1118, 0.1147, 0.2046], device='cuda:0'), in_proj_covar=tensor([0.1194, 0.1163, 0.1204, 0.1034], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 05:54:39,660 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:54:46,151 INFO [train.py:968] (0/2) Epoch 2, batch 24600, giga_loss[loss=0.3335, simple_loss=0.3952, pruned_loss=0.1359, over 28561.00 frames. ], tot_loss[loss=0.3883, simple_loss=0.4287, pruned_loss=0.1739, over 5670960.82 frames. ], libri_tot_loss[loss=0.3933, simple_loss=0.4299, pruned_loss=0.1784, over 5723438.72 frames. ], giga_tot_loss[loss=0.3889, simple_loss=0.4299, pruned_loss=0.1739, over 5660087.55 frames. ], batch size: 71, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:54:48,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.194e+02 1.538e+03 1.875e+03 2.463e+03 5.531e+03, threshold=3.750e+03, percent-clipped=2.0
+2023-03-01 05:54:55,164 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:54:59,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:55:23,667 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:55:27,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:55:29,217 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:55:39,463 INFO [train.py:968] (0/2) Epoch 2, batch 24650, giga_loss[loss=0.3532, simple_loss=0.4246, pruned_loss=0.1409, over 28865.00 frames. ], tot_loss[loss=0.3846, simple_loss=0.4267, pruned_loss=0.1712, over 5660971.21 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4307, pruned_loss=0.179, over 5714112.85 frames. ], giga_tot_loss[loss=0.3839, simple_loss=0.4267, pruned_loss=0.1705, over 5659624.74 frames. ], batch size: 174, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:55:46,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-01 05:55:58,377 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:56:29,875 INFO [train.py:968] (0/2) Epoch 2, batch 24700, giga_loss[loss=0.3528, simple_loss=0.4244, pruned_loss=0.1406, over 29000.00 frames. ], tot_loss[loss=0.3815, simple_loss=0.4273, pruned_loss=0.1679, over 5681172.38 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4306, pruned_loss=0.1789, over 5716433.70 frames. ], giga_tot_loss[loss=0.3808, simple_loss=0.4273, pruned_loss=0.1672, over 5677410.74 frames. ], batch size: 164, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:56:33,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.855e+02 1.521e+03 2.013e+03 2.564e+03 4.927e+03, threshold=4.025e+03, percent-clipped=8.0
+2023-03-01 05:56:36,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2128, 1.3570, 1.1689, 0.8760], device='cuda:0'), covar=tensor([0.0566, 0.0441, 0.0340, 0.0464], device='cuda:0'), in_proj_covar=tensor([0.1063, 0.0765, 0.0876, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 05:57:27,215 INFO [train.py:968] (0/2) Epoch 2, batch 24750, giga_loss[loss=0.4495, simple_loss=0.4672, pruned_loss=0.2159, over 27918.00 frames. ], tot_loss[loss=0.3836, simple_loss=0.4284, pruned_loss=0.1694, over 5663459.47 frames. ], libri_tot_loss[loss=0.3932, simple_loss=0.4296, pruned_loss=0.1784, over 5720873.32 frames. ], giga_tot_loss[loss=0.3837, simple_loss=0.4293, pruned_loss=0.1691, over 5655047.84 frames. ], batch size: 412, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 05:57:50,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70442.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:58:12,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4052, 2.8859, 3.1683, 1.6132], device='cuda:0'), covar=tensor([0.0634, 0.0561, 0.0860, 0.1866], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0548, 0.0813, 0.0558], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0007], device='cuda:0')
+2023-03-01 05:58:18,408 INFO [train.py:968] (0/2) Epoch 2, batch 24800, giga_loss[loss=0.4237, simple_loss=0.4593, pruned_loss=0.194, over 28928.00 frames. ], tot_loss[loss=0.3872, simple_loss=0.4304, pruned_loss=0.172, over 5659059.95 frames. ], libri_tot_loss[loss=0.3933, simple_loss=0.4296, pruned_loss=0.1785, over 5715787.33 frames. ], giga_tot_loss[loss=0.387, simple_loss=0.4312, pruned_loss=0.1714, over 5656428.22 frames. ], batch size: 136, lr: 1.30e-02, grad_scale: 8.0
+2023-03-01 05:58:19,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.199e+03 1.811e+03 2.399e+03 3.523e+03 9.489e+03, threshold=4.799e+03, percent-clipped=19.0
+2023-03-01 05:58:51,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=70503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:59:05,396 INFO [train.py:968] (0/2) Epoch 2, batch 24850, giga_loss[loss=0.3763, simple_loss=0.4098, pruned_loss=0.1714, over 28761.00 frames. ], tot_loss[loss=0.3874, simple_loss=0.4298, pruned_loss=0.1726, over 5660713.23 frames. ], libri_tot_loss[loss=0.3925, simple_loss=0.4288, pruned_loss=0.1781, over 5721304.87 frames. ], giga_tot_loss[loss=0.3878, simple_loss=0.431, pruned_loss=0.1723, over 5651627.02 frames. ], batch size: 99, lr: 1.30e-02, grad_scale: 8.0
+2023-03-01 05:59:28,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8116, 2.8275, 1.9931, 0.7942], device='cuda:0'), covar=tensor([0.1859, 0.0855, 0.1215, 0.2138], device='cuda:0'), in_proj_covar=tensor([0.1177, 0.1142, 0.1185, 0.1025], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 05:59:52,698 INFO [train.py:968] (0/2) Epoch 2, batch 24900, giga_loss[loss=0.3612, simple_loss=0.413, pruned_loss=0.1547, over 28371.00 frames. ], tot_loss[loss=0.3871, simple_loss=0.4284, pruned_loss=0.1729, over 5662847.80 frames. ], libri_tot_loss[loss=0.3921, simple_loss=0.4285, pruned_loss=0.1779, over 5727214.66 frames. ], giga_tot_loss[loss=0.3875, simple_loss=0.4297, pruned_loss=0.1726, over 5647867.15 frames. ], batch size: 368, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 05:59:57,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.411e+02 1.738e+03 2.325e+03 2.957e+03 7.896e+03, threshold=4.650e+03, percent-clipped=3.0
+2023-03-01 06:00:07,458 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70585.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 06:00:10,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70588.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 06:00:36,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70617.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 06:00:38,316 INFO [train.py:968] (0/2) Epoch 2, batch 24950, giga_loss[loss=0.3895, simple_loss=0.4292, pruned_loss=0.1749, over 28822.00 frames. ], tot_loss[loss=0.3855, simple_loss=0.4269, pruned_loss=0.172, over 5678564.86 frames. ], libri_tot_loss[loss=0.3919, simple_loss=0.4283, pruned_loss=0.1777, over 5731008.46 frames. ], giga_tot_loss[loss=0.3859, simple_loss=0.4281, pruned_loss=0.1718, over 5662114.57 frames. ], batch size: 186, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:01:11,888 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=70653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:01:19,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7427, 2.1701, 1.8480, 1.7740], device='cuda:0'), covar=tensor([0.1393, 0.1538, 0.1167, 0.0789], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0816, 0.0706, 0.0594], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 06:01:26,041 INFO [train.py:968] (0/2) Epoch 2, batch 25000, giga_loss[loss=0.3845, simple_loss=0.4408, pruned_loss=0.1641, over 29000.00 frames. ], tot_loss[loss=0.3847, simple_loss=0.4264, pruned_loss=0.1714, over 5681746.68 frames. ], libri_tot_loss[loss=0.3923, simple_loss=0.4286, pruned_loss=0.178, over 5735473.54 frames. ], giga_tot_loss[loss=0.3844, simple_loss=0.4271, pruned_loss=0.1709, over 5662922.59 frames. ], batch size: 128, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:01:27,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.382e+02 1.716e+03 2.120e+03 2.646e+03 4.806e+03, threshold=4.240e+03, percent-clipped=2.0
+2023-03-01 06:02:14,567 INFO [train.py:968] (0/2) Epoch 2, batch 25050, giga_loss[loss=0.354, simple_loss=0.4166, pruned_loss=0.1457, over 29038.00 frames. ], tot_loss[loss=0.3821, simple_loss=0.4254, pruned_loss=0.1694, over 5681256.77 frames. ], libri_tot_loss[loss=0.393, simple_loss=0.429, pruned_loss=0.1785, over 5734894.03 frames. ], giga_tot_loss[loss=0.3812, simple_loss=0.4256, pruned_loss=0.1685, over 5666149.27 frames. ], batch size: 155, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:03:03,358 INFO [train.py:968] (0/2) Epoch 2, batch 25100, giga_loss[loss=0.398, simple_loss=0.4405, pruned_loss=0.1778, over 28904.00 frames. ], tot_loss[loss=0.3847, simple_loss=0.4273, pruned_loss=0.1711, over 5677549.63 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4296, pruned_loss=0.1791, over 5734993.72 frames. ], giga_tot_loss[loss=0.3829, simple_loss=0.4268, pruned_loss=0.1695, over 5663396.23 frames. ], batch size: 174, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:03:05,695 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.998e+02 1.475e+03 2.158e+03 2.861e+03 6.258e+03, threshold=4.317e+03, percent-clipped=9.0
+2023-03-01 06:03:09,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6567, 1.6167, 1.5859, 1.6170], device='cuda:0'), covar=tensor([0.0808, 0.1213, 0.0933, 0.1013], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0819, 0.0641, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 06:03:55,267 INFO [train.py:968] (0/2) Epoch 2, batch 25150, giga_loss[loss=0.3257, simple_loss=0.3776, pruned_loss=0.1369, over 28592.00 frames. ], tot_loss[loss=0.3816, simple_loss=0.4249, pruned_loss=0.1691, over 5679718.28 frames. ], libri_tot_loss[loss=0.3933, simple_loss=0.4292, pruned_loss=0.1787, over 5737019.77 frames. ], giga_tot_loss[loss=0.3805, simple_loss=0.4248, pruned_loss=0.1681, over 5665501.68 frames. ], batch size: 78, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:03:58,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5370, 2.1058, 1.5259, 0.4391], device='cuda:0'), covar=tensor([0.1413, 0.0822, 0.1223, 0.1885], device='cuda:0'), in_proj_covar=tensor([0.1205, 0.1166, 0.1194, 0.1036], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 06:04:46,278 INFO [train.py:968] (0/2) Epoch 2, batch 25200, giga_loss[loss=0.4135, simple_loss=0.4263, pruned_loss=0.2004, over 23639.00 frames. ], tot_loss[loss=0.3796, simple_loss=0.4229, pruned_loss=0.1681, over 5688035.71 frames. ], libri_tot_loss[loss=0.3932, simple_loss=0.429, pruned_loss=0.1786, over 5739028.49 frames. ], giga_tot_loss[loss=0.3785, simple_loss=0.4229, pruned_loss=0.1671, over 5673822.83 frames. ], batch size: 705, lr: 1.30e-02, grad_scale: 8.0
+2023-03-01 06:04:48,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.437e+02 1.621e+03 2.014e+03 2.727e+03 5.103e+03, threshold=4.028e+03, percent-clipped=5.0
+2023-03-01 06:04:54,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:05:33,976 INFO [train.py:968] (0/2) Epoch 2, batch 25250, giga_loss[loss=0.433, simple_loss=0.4567, pruned_loss=0.2047, over 28600.00 frames. ], tot_loss[loss=0.3792, simple_loss=0.4223, pruned_loss=0.1681, over 5688689.11 frames. ], libri_tot_loss[loss=0.393, simple_loss=0.4291, pruned_loss=0.1784, over 5741250.80 frames. ], giga_tot_loss[loss=0.378, simple_loss=0.4219, pruned_loss=0.167, over 5673712.47 frames. ], batch size: 336, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:05:55,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9350, 1.1500, 0.9804, 0.5359], device='cuda:0'), covar=tensor([0.0445, 0.0402, 0.0320, 0.0487], device='cuda:0'), in_proj_covar=tensor([0.1041, 0.0773, 0.0859, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 06:06:25,308 INFO [train.py:968] (0/2) Epoch 2, batch 25300, giga_loss[loss=0.3421, simple_loss=0.392, pruned_loss=0.1461, over 28807.00 frames. ], tot_loss[loss=0.377, simple_loss=0.4203, pruned_loss=0.1669, over 5696359.48 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.429, pruned_loss=0.1783, over 5741923.55 frames. ], giga_tot_loss[loss=0.3762, simple_loss=0.4202, pruned_loss=0.1661, over 5683858.98 frames. ], batch size: 92, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:06:28,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.926e+03 2.345e+03 3.277e+03 1.155e+04, threshold=4.690e+03, percent-clipped=13.0
+2023-03-01 06:07:17,490 INFO [train.py:968] (0/2) Epoch 2, batch 25350, giga_loss[loss=0.4416, simple_loss=0.4629, pruned_loss=0.2102, over 28001.00 frames. ], tot_loss[loss=0.3761, simple_loss=0.4191, pruned_loss=0.1666, over 5693780.21 frames. ], libri_tot_loss[loss=0.3927, simple_loss=0.4289, pruned_loss=0.1783, over 5744292.81 frames. ], giga_tot_loss[loss=0.3753, simple_loss=0.4189, pruned_loss=0.1658, over 5681110.42 frames. ], batch size: 412, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:07:19,326 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:07:22,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:07:25,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:07:38,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3720, 1.1645, 1.1213, 1.6301], device='cuda:0'), covar=tensor([0.1780, 0.1875, 0.1563, 0.1943], device='cuda:0'), in_proj_covar=tensor([0.0965, 0.0823, 0.0877, 0.0936], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 06:07:48,489 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:08:05,383 INFO [train.py:968] (0/2) Epoch 2, batch 25400, giga_loss[loss=0.4308, simple_loss=0.4363, pruned_loss=0.2126, over 23580.00 frames. ], tot_loss[loss=0.3768, simple_loss=0.419, pruned_loss=0.1673, over 5681960.97 frames. ], libri_tot_loss[loss=0.3936, simple_loss=0.4295, pruned_loss=0.1788, over 5734705.02 frames. ], giga_tot_loss[loss=0.3748, simple_loss=0.418, pruned_loss=0.1658, over 5678185.16 frames. ], batch size: 705, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:08:09,894 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.003e+02 1.693e+03 2.078e+03 2.855e+03 5.459e+03, threshold=4.157e+03, percent-clipped=2.0
+2023-03-01 06:08:52,585 INFO [train.py:968] (0/2) Epoch 2, batch 25450, giga_loss[loss=0.3528, simple_loss=0.4082, pruned_loss=0.1487, over 28936.00 frames. ], tot_loss[loss=0.3782, simple_loss=0.4199, pruned_loss=0.1683, over 5679545.33 frames. ], libri_tot_loss[loss=0.3923, simple_loss=0.4284, pruned_loss=0.1781, over 5730776.82 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4195, pruned_loss=0.1672, over 5677322.74 frames. ], batch size: 145, lr: 1.30e-02, grad_scale: 2.0
+2023-03-01 06:09:43,000 INFO [train.py:968] (0/2) Epoch 2, batch 25500, giga_loss[loss=0.3105, simple_loss=0.3876, pruned_loss=0.1167, over 29045.00 frames. ], tot_loss[loss=0.3763, simple_loss=0.4195, pruned_loss=0.1665, over 5683117.28 frames. ], libri_tot_loss[loss=0.3918, simple_loss=0.4281, pruned_loss=0.1778, over 5730770.24 frames. ], giga_tot_loss[loss=0.3755, simple_loss=0.4194, pruned_loss=0.1658, over 5680427.59 frames. ], batch size: 155, lr: 1.30e-02, grad_scale: 2.0
+2023-03-01 06:09:44,593 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:09:47,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.256e+02 1.553e+03 2.045e+03 3.113e+03 1.356e+04, threshold=4.089e+03, percent-clipped=10.0
+2023-03-01 06:09:47,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:10:11,662 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:10:20,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7511, 2.4889, 1.4649, 1.6806], device='cuda:0'), covar=tensor([0.0933, 0.0293, 0.0457, 0.1100], device='cuda:0'), in_proj_covar=tensor([0.0289, 0.0197, 0.0195, 0.0345], device='cuda:0'), out_proj_covar=tensor([0.0032, 0.0025, 0.0021, 0.0036], device='cuda:0')
+2023-03-01 06:10:25,008 INFO [train.py:968] (0/2) Epoch 2, batch 25550, giga_loss[loss=0.3688, simple_loss=0.4196, pruned_loss=0.159, over 28804.00 frames. ], tot_loss[loss=0.3758, simple_loss=0.4193, pruned_loss=0.1662, over 5680732.47 frames. ], libri_tot_loss[loss=0.3915, simple_loss=0.4277, pruned_loss=0.1777, over 5729802.59 frames. ], giga_tot_loss[loss=0.3746, simple_loss=0.4191, pruned_loss=0.1651, over 5676133.91 frames. ], batch size: 119, lr: 1.30e-02, grad_scale: 2.0
+2023-03-01 06:11:17,076 INFO [train.py:968] (0/2) Epoch 2, batch 25600, giga_loss[loss=0.4005, simple_loss=0.438, pruned_loss=0.1815, over 28641.00 frames. ], tot_loss[loss=0.3766, simple_loss=0.4207, pruned_loss=0.1663, over 5679726.22 frames. ], libri_tot_loss[loss=0.3915, simple_loss=0.4277, pruned_loss=0.1776, over 5727295.57 frames. ], giga_tot_loss[loss=0.3756, simple_loss=0.4205, pruned_loss=0.1654, over 5678064.80 frames. ], batch size: 92, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:11:22,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.914e+02 1.500e+03 1.822e+03 2.426e+03 4.839e+03, threshold=3.644e+03, percent-clipped=4.0
+2023-03-01 06:11:36,688 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-01 06:12:06,905 INFO [train.py:968] (0/2) Epoch 2, batch 25650, giga_loss[loss=0.3901, simple_loss=0.4342, pruned_loss=0.173, over 28558.00 frames. ], tot_loss[loss=0.3807, simple_loss=0.4232, pruned_loss=0.1691, over 5680951.99 frames. ], libri_tot_loss[loss=0.3916, simple_loss=0.4278, pruned_loss=0.1777, over 5728774.97 frames. ], giga_tot_loss[loss=0.3796, simple_loss=0.4229, pruned_loss=0.1682, over 5677954.19 frames. ], batch size: 307, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:12:56,102 INFO [train.py:968] (0/2) Epoch 2, batch 25700, giga_loss[loss=0.3635, simple_loss=0.4069, pruned_loss=0.16, over 28602.00 frames. ], tot_loss[loss=0.3858, simple_loss=0.4263, pruned_loss=0.1727, over 5678319.17 frames. ], libri_tot_loss[loss=0.3924, simple_loss=0.4283, pruned_loss=0.1783, over 5722453.73 frames. ], giga_tot_loss[loss=0.384, simple_loss=0.4254, pruned_loss=0.1713, over 5680537.68 frames. ], batch size: 92, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:13:01,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.813e+03 2.320e+03 3.127e+03 9.971e+03, threshold=4.639e+03, percent-clipped=14.0
+2023-03-01 06:13:15,025 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=71385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:13:31,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9870, 2.6490, 1.5377, 2.0779], device='cuda:0'), covar=tensor([0.0850, 0.0266, 0.0456, 0.1043], device='cuda:0'), in_proj_covar=tensor([0.0286, 0.0195, 0.0196, 0.0345], device='cuda:0'), out_proj_covar=tensor([0.0032, 0.0024, 0.0021, 0.0036], device='cuda:0')
+2023-03-01 06:13:50,619 INFO [train.py:968] (0/2) Epoch 2, batch 25750, giga_loss[loss=0.489, simple_loss=0.4648, pruned_loss=0.2566, over 23658.00 frames. ], tot_loss[loss=0.3893, simple_loss=0.4274, pruned_loss=0.1756, over 5665861.09 frames. ], libri_tot_loss[loss=0.3916, simple_loss=0.4277, pruned_loss=0.1778, over 5715509.59 frames. ], giga_tot_loss[loss=0.3885, simple_loss=0.4273, pruned_loss=0.1749, over 5672150.09 frames. ], batch size: 705, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:13:53,806 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=71423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:14:44,806 INFO [train.py:968] (0/2) Epoch 2, batch 25800, giga_loss[loss=0.5269, simple_loss=0.5173, pruned_loss=0.2682, over 27557.00 frames. ], tot_loss[loss=0.3912, simple_loss=0.4279, pruned_loss=0.1772, over 5674600.23 frames. ], libri_tot_loss[loss=0.3914, simple_loss=0.4274, pruned_loss=0.1777, over 5719348.59 frames. ], giga_tot_loss[loss=0.3907, simple_loss=0.4281, pruned_loss=0.1767, over 5675158.69 frames. ], batch size: 472, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:14:49,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.063e+02 1.668e+03 2.051e+03 2.871e+03 7.123e+03, threshold=4.101e+03, percent-clipped=8.0
+2023-03-01 06:15:30,750 INFO [train.py:968] (0/2) Epoch 2, batch 25850, libri_loss[loss=0.3506, simple_loss=0.3933, pruned_loss=0.154, over 29569.00 frames. ], tot_loss[loss=0.3899, simple_loss=0.4272, pruned_loss=0.1763, over 5686959.59 frames. ], libri_tot_loss[loss=0.3907, simple_loss=0.4269, pruned_loss=0.1772, over 5725053.34 frames. ], giga_tot_loss[loss=0.3902, simple_loss=0.4278, pruned_loss=0.1763, over 5680966.18 frames. ], batch size: 78, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:15:57,456 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-01 06:16:18,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7739, 1.5056, 3.6553, 2.9563], device='cuda:0'), covar=tensor([0.1406, 0.1549, 0.0341, 0.0648], device='cuda:0'), in_proj_covar=tensor([0.0506, 0.0495, 0.0658, 0.0521], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 06:16:19,298 INFO [train.py:968] (0/2) Epoch 2, batch 25900, giga_loss[loss=0.3562, simple_loss=0.404, pruned_loss=0.1542, over 28570.00 frames. ], tot_loss[loss=0.3891, simple_loss=0.4262, pruned_loss=0.176, over 5667275.32 frames. ], libri_tot_loss[loss=0.3916, simple_loss=0.4276, pruned_loss=0.1778, over 5718182.24 frames. ], giga_tot_loss[loss=0.3885, simple_loss=0.4261, pruned_loss=0.1755, over 5667929.24 frames. ], batch size: 336, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:16:23,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.007e+02 1.719e+03 2.208e+03 3.138e+03 8.206e+03, threshold=4.416e+03, percent-clipped=8.0
+2023-03-01 06:17:02,982 INFO [train.py:968] (0/2) Epoch 2, batch 25950, giga_loss[loss=0.34, simple_loss=0.4047, pruned_loss=0.1377, over 29066.00 frames. ], tot_loss[loss=0.3844, simple_loss=0.4246, pruned_loss=0.172, over 5678271.47 frames. ], libri_tot_loss[loss=0.392, simple_loss=0.4279, pruned_loss=0.178, over 5722033.17 frames. ], giga_tot_loss[loss=0.3835, simple_loss=0.4242, pruned_loss=0.1714, over 5674433.56 frames. ], batch size: 155, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:17:49,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9954, 4.2707, 4.7225, 1.7956], device='cuda:0'), covar=tensor([0.0351, 0.0326, 0.0613, 0.2053], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0559, 0.0830, 0.0566], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 06:17:52,813 INFO [train.py:968] (0/2) Epoch 2, batch 26000, giga_loss[loss=0.3672, simple_loss=0.4167, pruned_loss=0.1588, over 28484.00 frames. ], tot_loss[loss=0.3801, simple_loss=0.4209, pruned_loss=0.1696, over 5661293.31 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.4286, pruned_loss=0.1786, over 5724489.16 frames. ], giga_tot_loss[loss=0.3783, simple_loss=0.4198, pruned_loss=0.1684, over 5654786.98 frames. ], batch size: 71, lr: 1.29e-02, grad_scale: 8.0
+2023-03-01 06:17:56,495 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 06:17:57,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.212e+02 1.671e+03 2.081e+03 3.346e+03 7.305e+03, threshold=4.161e+03, percent-clipped=10.0
+2023-03-01 06:18:41,627 INFO [train.py:968] (0/2) Epoch 2, batch 26050, giga_loss[loss=0.3773, simple_loss=0.4157, pruned_loss=0.1695, over 28813.00 frames. ], tot_loss[loss=0.3765, simple_loss=0.4179, pruned_loss=0.1675, over 5670336.48 frames. ], libri_tot_loss[loss=0.3923, simple_loss=0.4281, pruned_loss=0.1782, over 5729023.01 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.4173, pruned_loss=0.1666, over 5659877.67 frames. ], batch size: 227, lr: 1.29e-02, grad_scale: 8.0
+2023-03-01 06:19:21,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:19:29,900 INFO [train.py:968] (0/2) Epoch 2, batch 26100, giga_loss[loss=0.3703, simple_loss=0.4081, pruned_loss=0.1663, over 28679.00 frames. ], tot_loss[loss=0.377, simple_loss=0.4174, pruned_loss=0.1683, over 5664395.24 frames. ], libri_tot_loss[loss=0.3924, simple_loss=0.4281, pruned_loss=0.1783, over 5730521.18 frames. ], giga_tot_loss[loss=0.3757, simple_loss=0.4168, pruned_loss=0.1674, over 5654189.21 frames. ], batch size: 92, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:19:36,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.057e+02 1.657e+03 2.175e+03 2.722e+03 9.100e+03, threshold=4.349e+03, percent-clipped=8.0
+2023-03-01 06:19:57,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4072, 2.2122, 1.4736, 1.2462], device='cuda:0'), covar=tensor([0.0749, 0.0556, 0.0735, 0.1244], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0445, 0.0328, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0016, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 06:20:01,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:20:22,696 INFO [train.py:968] (0/2) Epoch 2, batch 26150, giga_loss[loss=0.5137, simple_loss=0.4961, pruned_loss=0.2656, over 26699.00 frames. ], tot_loss[loss=0.3813, simple_loss=0.4206, pruned_loss=0.171, over 5662563.87 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4285, pruned_loss=0.1786, over 5729960.71 frames. ], giga_tot_loss[loss=0.3799, simple_loss=0.4198, pruned_loss=0.17, over 5654374.04 frames. ], batch size: 555, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:21:03,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5005, 3.8674, 4.1269, 2.0955], device='cuda:0'), covar=tensor([0.0536, 0.0520, 0.0928, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.0787, 0.0566, 0.0830, 0.0570], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 06:21:10,384 INFO [train.py:968] (0/2) Epoch 2, batch 26200, giga_loss[loss=0.4018, simple_loss=0.4607, pruned_loss=0.1714, over 28951.00 frames. ], tot_loss[loss=0.3849, simple_loss=0.4251, pruned_loss=0.1723, over 5670930.17 frames. ], libri_tot_loss[loss=0.3925, simple_loss=0.4283, pruned_loss=0.1783, over 5733365.48 frames. ], giga_tot_loss[loss=0.3837, simple_loss=0.4244, pruned_loss=0.1715, over 5659458.48 frames. ], batch size: 164, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:21:17,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.671e+03 2.116e+03 2.782e+03 5.963e+03, threshold=4.231e+03, percent-clipped=3.0
+2023-03-01 06:21:44,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:21:49,165 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:22:01,124 INFO [train.py:968] (0/2) Epoch 2, batch 26250, giga_loss[loss=0.4182, simple_loss=0.4601, pruned_loss=0.1881, over 27993.00 frames. ], tot_loss[loss=0.3873, simple_loss=0.429, pruned_loss=0.1728, over 5669696.75 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.4285, pruned_loss=0.1786, over 5736741.68 frames. ], giga_tot_loss[loss=0.3859, simple_loss=0.4282, pruned_loss=0.1718, over 5656337.63 frames. ], batch size: 412, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:22:14,746 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:22:21,008 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:22:23,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:22:27,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2327, 1.1448, 1.2250, 1.0776], device='cuda:0'), covar=tensor([0.0706, 0.0973, 0.1203, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0807, 0.0631, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 06:22:49,233 INFO [train.py:968] (0/2) Epoch 2, batch 26300, giga_loss[loss=0.4842, simple_loss=0.4712, pruned_loss=0.2486, over 23690.00 frames. ], tot_loss[loss=0.3888, simple_loss=0.4302, pruned_loss=0.1737, over 5668552.62 frames. ], libri_tot_loss[loss=0.3922, simple_loss=0.4279, pruned_loss=0.1783, over 5740577.55 frames. ], giga_tot_loss[loss=0.3881, simple_loss=0.4301, pruned_loss=0.1731, over 5653309.27 frames. ], batch size: 705, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:22:54,255 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:22:55,858 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.248e+02 1.610e+03 2.348e+03 3.031e+03 4.714e+03, threshold=4.696e+03, percent-clipped=2.0
+2023-03-01 06:23:19,926 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-72000.pt
+2023-03-01 06:23:27,412 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=72009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:23:36,333 INFO [train.py:968] (0/2) Epoch 2, batch 26350, giga_loss[loss=0.4901, simple_loss=0.4837, pruned_loss=0.2482, over 26674.00 frames. ], tot_loss[loss=0.3926, simple_loss=0.4325, pruned_loss=0.1763, over 5661076.84 frames. ], libri_tot_loss[loss=0.3926, simple_loss=0.428, pruned_loss=0.1786, over 5735470.95 frames. ], giga_tot_loss[loss=0.3917, simple_loss=0.4325, pruned_loss=0.1755, over 5650363.14 frames. ], batch size: 555, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:23:41,800 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-01 06:24:21,742 INFO [train.py:968] (0/2) Epoch 2, batch 26400, giga_loss[loss=0.4117, simple_loss=0.4427, pruned_loss=0.1904, over 28265.00 frames. ], tot_loss[loss=0.3928, simple_loss=0.4321, pruned_loss=0.1767, over 5659500.30 frames. ], libri_tot_loss[loss=0.392, simple_loss=0.4276, pruned_loss=0.1783, over 5738450.33 frames. ], giga_tot_loss[loss=0.3926, simple_loss=0.4327, pruned_loss=0.1763, over 5646217.01 frames. ], batch size: 368, lr: 1.29e-02, grad_scale: 8.0
+2023-03-01 06:24:27,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.753e+02 1.646e+03 2.075e+03 2.843e+03 7.413e+03, threshold=4.151e+03, percent-clipped=7.0
+2023-03-01 06:25:11,172 INFO [train.py:968] (0/2) Epoch 2, batch 26450, giga_loss[loss=0.3585, simple_loss=0.3999, pruned_loss=0.1586, over 28909.00 frames. ], tot_loss[loss=0.3903, simple_loss=0.4299, pruned_loss=0.1754, over 5646044.32 frames. ], libri_tot_loss[loss=0.3922, simple_loss=0.4277, pruned_loss=0.1783, over 5726513.27 frames. ], giga_tot_loss[loss=0.3901, simple_loss=0.4303, pruned_loss=0.1749, over 5644580.33 frames. ], batch size: 119, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:25:41,730 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-01 06:26:00,604 INFO [train.py:968] (0/2) Epoch 2, batch 26500, giga_loss[loss=0.3703, simple_loss=0.4162, pruned_loss=0.1622, over 28891.00 frames. ], tot_loss[loss=0.3895, simple_loss=0.4288, pruned_loss=0.1751, over 5648580.11 frames. ], libri_tot_loss[loss=0.393, simple_loss=0.4283, pruned_loss=0.1789, over 5728166.81 frames. ], giga_tot_loss[loss=0.3884, simple_loss=0.4285, pruned_loss=0.1742, over 5643656.93 frames. ], batch size: 174, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:26:07,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.968e+02 1.495e+03 1.815e+03 2.588e+03 7.332e+03, threshold=3.630e+03, percent-clipped=5.0
+2023-03-01 06:26:49,564 INFO [train.py:968] (0/2) Epoch 2, batch 26550, giga_loss[loss=0.3956, simple_loss=0.4266, pruned_loss=0.1823, over 28805.00 frames. ], tot_loss[loss=0.3876, simple_loss=0.4269, pruned_loss=0.1742, over 5662658.64 frames. ], libri_tot_loss[loss=0.3937, simple_loss=0.429, pruned_loss=0.1792, over 5734029.78 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4261, pruned_loss=0.1729, over 5650419.11 frames. ], batch size: 99, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:27:41,769 INFO [train.py:968] (0/2) Epoch 2, batch 26600, giga_loss[loss=0.3867, simple_loss=0.4332, pruned_loss=0.1701, over 28843.00 frames. ], tot_loss[loss=0.3871, simple_loss=0.426, pruned_loss=0.1741, over 5651051.58 frames. ], libri_tot_loss[loss=0.3941, simple_loss=0.4293, pruned_loss=0.1795, over 5733711.07 frames. ], giga_tot_loss[loss=0.3853, simple_loss=0.425, pruned_loss=0.1728, over 5640206.29 frames. ], batch size: 285, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:27:45,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2979, 2.5370, 1.2537, 1.2127], device='cuda:0'), covar=tensor([0.1163, 0.0599, 0.1068, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0456, 0.0330, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0016, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 06:27:47,546 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-01 06:27:49,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.705e+03 2.146e+03 2.763e+03 7.381e+03, threshold=4.292e+03, percent-clipped=12.0
+2023-03-01 06:28:23,671 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=72312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:28:31,988 INFO [train.py:968] (0/2) Epoch 2, batch 26650, giga_loss[loss=0.3847, simple_loss=0.4034, pruned_loss=0.183, over 23580.00 frames. ], tot_loss[loss=0.3894, simple_loss=0.4273, pruned_loss=0.1757, over 5651124.51 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4292, pruned_loss=0.1793, over 5736554.27 frames. ], giga_tot_loss[loss=0.3881, simple_loss=0.4265, pruned_loss=0.1748, over 5638945.90 frames. ], batch size: 705, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:28:58,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7790, 1.3895, 3.8618, 3.3374], device='cuda:0'), covar=tensor([0.1555, 0.1695, 0.0334, 0.0444], device='cuda:0'), in_proj_covar=tensor([0.0511, 0.0500, 0.0672, 0.0532], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 06:29:18,662 INFO [train.py:968] (0/2) Epoch 2, batch 26700, giga_loss[loss=0.3634, simple_loss=0.4036, pruned_loss=0.1616, over 28833.00 frames. ], tot_loss[loss=0.3857, simple_loss=0.4243, pruned_loss=0.1735, over 5667519.09 frames. ], libri_tot_loss[loss=0.394, simple_loss=0.4292, pruned_loss=0.1793, over 5738856.76 frames. ], giga_tot_loss[loss=0.3844, simple_loss=0.4236, pruned_loss=0.1726, over 5653953.23 frames. ], batch size: 119, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:29:26,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.791e+03 2.290e+03 3.195e+03 1.019e+04, threshold=4.580e+03, percent-clipped=10.0
+2023-03-01 06:29:28,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5924, 1.4893, 1.4032, 1.4220], device='cuda:0'), covar=tensor([0.0733, 0.1445, 0.1327, 0.1184], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0838, 0.0649, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 06:29:32,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:29:46,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7204, 1.8521, 1.4474, 0.9339], device='cuda:0'), covar=tensor([0.0463, 0.0334, 0.0271, 0.0490], device='cuda:0'), in_proj_covar=tensor([0.1036, 0.0771, 0.0863, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 06:30:06,397 INFO [train.py:968] (0/2) Epoch 2, batch 26750, giga_loss[loss=0.3325, simple_loss=0.3936, pruned_loss=0.1357, over 28903.00 frames. ], tot_loss[loss=0.3844, simple_loss=0.4235, pruned_loss=0.1727, over 5681145.20 frames. ], libri_tot_loss[loss=0.3937, simple_loss=0.4291, pruned_loss=0.1792, over 5743970.05 frames. ], giga_tot_loss[loss=0.3832, simple_loss=0.4227, pruned_loss=0.1718, over 5663422.19 frames. ], batch size: 174, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:30:10,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2444, 2.7312, 3.0162, 1.2706], device='cuda:0'), covar=tensor([0.0718, 0.0591, 0.1120, 0.1924], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0584, 0.0842, 0.0572], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 06:30:54,589 INFO [train.py:968] (0/2) Epoch 2, batch 26800, giga_loss[loss=0.4518, simple_loss=0.4705, pruned_loss=0.2166, over 27921.00 frames. ], tot_loss[loss=0.3873, simple_loss=0.4259, pruned_loss=0.1744, over 5674535.65 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.4298, pruned_loss=0.1797, over 5738321.50 frames. ], giga_tot_loss[loss=0.3853, simple_loss=0.4245, pruned_loss=0.173, over 5662389.39 frames. ], batch size: 412, lr: 1.29e-02, grad_scale: 8.0
+2023-03-01 06:30:59,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.135e+02 1.620e+03 2.000e+03 2.396e+03 1.166e+04, threshold=4.000e+03, percent-clipped=3.0
+2023-03-01 06:31:46,412 INFO [train.py:968] (0/2) Epoch 2, batch 26850, giga_loss[loss=0.46, simple_loss=0.4703, pruned_loss=0.2249, over 27720.00 frames. ], tot_loss[loss=0.39, simple_loss=0.4286, pruned_loss=0.1757, over 5672528.30 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.4298, pruned_loss=0.1797, over 5742295.87 frames. ], giga_tot_loss[loss=0.3883, simple_loss=0.4275, pruned_loss=0.1745, over 5658241.54 frames. ], batch size: 474, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:31:48,485 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=72522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:31:52,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:31:55,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:32:24,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1917, 1.5418, 1.2097, 0.6774], device='cuda:0'), covar=tensor([0.0815, 0.0689, 0.0778, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.1180, 0.1155, 0.1167, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 06:32:28,135 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:32:36,977 INFO [train.py:968] (0/2) Epoch 2, batch 26900, giga_loss[loss=0.326, simple_loss=0.3811, pruned_loss=0.1354, over 28959.00 frames. ], tot_loss[loss=0.3883, simple_loss=0.4272, pruned_loss=0.1747, over 5668793.78 frames. ], libri_tot_loss[loss=0.3944, simple_loss=0.4297, pruned_loss=0.1795, over 5744353.98 frames. ], giga_tot_loss[loss=0.3871, simple_loss=0.4264, pruned_loss=0.1739, over 5654535.57 frames. ], batch size: 213, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:32:45,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.695e+03 2.280e+03 3.136e+03 8.299e+03, threshold=4.560e+03, percent-clipped=11.0
+2023-03-01 06:33:03,566 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-01 06:33:25,518 INFO [train.py:968] (0/2) Epoch 2, batch 26950, giga_loss[loss=0.3417, simple_loss=0.4156, pruned_loss=0.1339, over 28980.00 frames. ], tot_loss[loss=0.388, simple_loss=0.428, pruned_loss=0.174, over 5668799.71 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4293, pruned_loss=0.1793, over 5737976.65 frames. ], giga_tot_loss[loss=0.3873, simple_loss=0.4276, pruned_loss=0.1734, over 5661516.01 frames. ], batch size: 106, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:33:39,726 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 06:34:12,132 INFO [train.py:968] (0/2) Epoch 2, batch 27000, giga_loss[loss=0.4242, simple_loss=0.4502, pruned_loss=0.1991, over 27562.00 frames. ], tot_loss[loss=0.387, simple_loss=0.429, pruned_loss=0.1725, over 5679190.86 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.4296, pruned_loss=0.1798, over 5741617.42 frames. ], giga_tot_loss[loss=0.3856, simple_loss=0.4284, pruned_loss=0.1714, over 5668100.00 frames. ], batch size: 472, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:34:12,136 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 06:34:20,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2110, 1.3581, 1.1884, 1.1651], device='cuda:0'), covar=tensor([0.2117, 0.2027, 0.1862, 0.1837], device='cuda:0'), in_proj_covar=tensor([0.0974, 0.0816, 0.0886, 0.0940], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 06:34:20,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4888, 1.3947, 1.4167, 1.3955], device='cuda:0'), covar=tensor([0.0681, 0.1090, 0.1146, 0.0944], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0809, 0.0629, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 06:34:21,353 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.2666, simple_loss=0.3619, pruned_loss=0.08569, over 944034.00 frames. 
+2023-03-01 06:34:21,353 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19537MB
+2023-03-01 06:34:26,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.430e+02 1.526e+03 1.979e+03 2.685e+03 7.630e+03, threshold=3.958e+03, percent-clipped=6.0
+2023-03-01 06:34:38,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:35:07,032 INFO [train.py:968] (0/2) Epoch 2, batch 27050, giga_loss[loss=0.3201, simple_loss=0.399, pruned_loss=0.1206, over 28856.00 frames. ], tot_loss[loss=0.3854, simple_loss=0.4296, pruned_loss=0.1706, over 5671941.18 frames. ], libri_tot_loss[loss=0.3947, simple_loss=0.4296, pruned_loss=0.1799, over 5732929.95 frames. ], giga_tot_loss[loss=0.3838, simple_loss=0.4291, pruned_loss=0.1693, over 5669826.26 frames. ], batch size: 66, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:35:07,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=72719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:35:50,300 INFO [train.py:968] (0/2) Epoch 2, batch 27100, giga_loss[loss=0.3595, simple_loss=0.4103, pruned_loss=0.1544, over 28947.00 frames. ], tot_loss[loss=0.3869, simple_loss=0.4306, pruned_loss=0.1716, over 5665022.14 frames. ], libri_tot_loss[loss=0.3941, simple_loss=0.4291, pruned_loss=0.1796, over 5721885.04 frames. ], giga_tot_loss[loss=0.3859, simple_loss=0.4307, pruned_loss=0.1705, over 5670835.55 frames. ], batch size: 106, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:35:57,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.658e+03 2.138e+03 2.911e+03 6.651e+03, threshold=4.276e+03, percent-clipped=9.0
+2023-03-01 06:36:45,862 INFO [train.py:968] (0/2) Epoch 2, batch 27150, giga_loss[loss=0.382, simple_loss=0.4247, pruned_loss=0.1697, over 29034.00 frames. ], tot_loss[loss=0.3927, simple_loss=0.434, pruned_loss=0.1757, over 5667147.20 frames. ], libri_tot_loss[loss=0.3942, simple_loss=0.4292, pruned_loss=0.1796, over 5722997.51 frames. ], giga_tot_loss[loss=0.3918, simple_loss=0.434, pruned_loss=0.1748, over 5670374.76 frames. ], batch size: 136, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:36:55,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:36:58,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:37:29,608 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:37:36,507 INFO [train.py:968] (0/2) Epoch 2, batch 27200, giga_loss[loss=0.3759, simple_loss=0.4211, pruned_loss=0.1653, over 29050.00 frames. ], tot_loss[loss=0.3946, simple_loss=0.4347, pruned_loss=0.1773, over 5671750.67 frames. ], libri_tot_loss[loss=0.3944, simple_loss=0.4291, pruned_loss=0.1798, over 5719633.84 frames. ], giga_tot_loss[loss=0.3938, simple_loss=0.4348, pruned_loss=0.1763, over 5675180.58 frames. ], batch size: 136, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:37:44,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.774e+03 2.172e+03 3.022e+03 9.082e+03, threshold=4.344e+03, percent-clipped=6.0
+2023-03-01 06:38:03,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:38:24,990 INFO [train.py:968] (0/2) Epoch 2, batch 27250, giga_loss[loss=0.4252, simple_loss=0.4608, pruned_loss=0.1948, over 28864.00 frames. ], tot_loss[loss=0.3951, simple_loss=0.4346, pruned_loss=0.1778, over 5661768.18 frames. ], libri_tot_loss[loss=0.3947, simple_loss=0.4293, pruned_loss=0.18, over 5713003.31 frames. ], giga_tot_loss[loss=0.3942, simple_loss=0.4347, pruned_loss=0.1768, over 5668750.90 frames. ], batch size: 227, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:39:15,394 INFO [train.py:968] (0/2) Epoch 2, batch 27300, giga_loss[loss=0.4534, simple_loss=0.4439, pruned_loss=0.2314, over 23489.00 frames. ], tot_loss[loss=0.3894, simple_loss=0.431, pruned_loss=0.1739, over 5661451.58 frames. ], libri_tot_loss[loss=0.3942, simple_loss=0.429, pruned_loss=0.1797, over 5707645.38 frames. ], giga_tot_loss[loss=0.389, simple_loss=0.4315, pruned_loss=0.1732, over 5669995.65 frames. ], batch size: 705, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:39:21,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.396e+02 1.560e+03 2.147e+03 2.560e+03 5.450e+03, threshold=4.293e+03, percent-clipped=5.0
+2023-03-01 06:40:02,651 INFO [train.py:968] (0/2) Epoch 2, batch 27350, giga_loss[loss=0.3916, simple_loss=0.4442, pruned_loss=0.1695, over 28914.00 frames. ], tot_loss[loss=0.3871, simple_loss=0.4305, pruned_loss=0.1719, over 5660907.39 frames. ], libri_tot_loss[loss=0.394, simple_loss=0.4288, pruned_loss=0.1796, over 5711992.10 frames. ], giga_tot_loss[loss=0.3868, simple_loss=0.4311, pruned_loss=0.1713, over 5662493.14 frames. ], batch size: 112, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:40:23,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:40:26,011 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:40:50,332 INFO [train.py:968] (0/2) Epoch 2, batch 27400, giga_loss[loss=0.3399, simple_loss=0.4024, pruned_loss=0.1387, over 28814.00 frames. ], tot_loss[loss=0.3877, simple_loss=0.4313, pruned_loss=0.172, over 5652528.39 frames. ], libri_tot_loss[loss=0.3936, simple_loss=0.4284, pruned_loss=0.1794, over 5699617.05 frames. ], giga_tot_loss[loss=0.3876, simple_loss=0.4322, pruned_loss=0.1715, over 5662496.32 frames. ], batch size: 119, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:40:53,711 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:40:57,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.800e+02 1.565e+03 2.081e+03 2.785e+03 6.203e+03, threshold=4.161e+03, percent-clipped=5.0
+2023-03-01 06:41:18,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:41:43,431 INFO [train.py:968] (0/2) Epoch 2, batch 27450, giga_loss[loss=0.3924, simple_loss=0.4369, pruned_loss=0.1739, over 28626.00 frames. ], tot_loss[loss=0.3894, simple_loss=0.4322, pruned_loss=0.1733, over 5647484.21 frames. ], libri_tot_loss[loss=0.3926, simple_loss=0.4277, pruned_loss=0.1788, over 5702839.58 frames. ], giga_tot_loss[loss=0.39, simple_loss=0.4336, pruned_loss=0.1732, over 5651562.98 frames. ], batch size: 85, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:42:30,577 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-01 06:42:30,679 INFO [train.py:968] (0/2) Epoch 2, batch 27500, giga_loss[loss=0.3484, simple_loss=0.4069, pruned_loss=0.145, over 28694.00 frames. ], tot_loss[loss=0.3892, simple_loss=0.4317, pruned_loss=0.1733, over 5653294.62 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.4279, pruned_loss=0.1789, over 5699647.91 frames. ], giga_tot_loss[loss=0.3894, simple_loss=0.4328, pruned_loss=0.173, over 5657888.53 frames. ], batch size: 242, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:42:40,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.670e+02 1.650e+03 2.287e+03 3.229e+03 1.151e+04, threshold=4.575e+03, percent-clipped=14.0
+2023-03-01 06:42:59,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2524, 3.6509, 3.9259, 1.8519], device='cuda:0'), covar=tensor([0.0442, 0.0441, 0.0759, 0.1730], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0578, 0.0837, 0.0561], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 06:43:25,314 INFO [train.py:968] (0/2) Epoch 2, batch 27550, giga_loss[loss=0.361, simple_loss=0.4085, pruned_loss=0.1568, over 28911.00 frames. ], tot_loss[loss=0.3896, simple_loss=0.431, pruned_loss=0.1742, over 5659460.29 frames. ], libri_tot_loss[loss=0.3926, simple_loss=0.4277, pruned_loss=0.1787, over 5702940.62 frames. ], giga_tot_loss[loss=0.39, simple_loss=0.432, pruned_loss=0.174, over 5659508.87 frames. ], batch size: 213, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:43:44,907 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:43:47,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:44:15,762 INFO [train.py:968] (0/2) Epoch 2, batch 27600, giga_loss[loss=0.379, simple_loss=0.4083, pruned_loss=0.1749, over 28737.00 frames. ], tot_loss[loss=0.3873, simple_loss=0.4288, pruned_loss=0.1729, over 5671010.65 frames. ], libri_tot_loss[loss=0.3922, simple_loss=0.4276, pruned_loss=0.1784, over 5707790.54 frames. ], giga_tot_loss[loss=0.3879, simple_loss=0.4299, pruned_loss=0.1729, over 5665553.94 frames. ], batch size: 92, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:44:15,998 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:44:27,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.990e+02 1.694e+03 2.101e+03 3.086e+03 8.221e+03, threshold=4.203e+03, percent-clipped=9.0
+2023-03-01 06:44:28,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3207, 3.8252, 4.0675, 1.6179], device='cuda:0'), covar=tensor([0.0544, 0.0421, 0.0882, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0585, 0.0840, 0.0570], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 06:44:37,224 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=73287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:45:09,731 INFO [train.py:968] (0/2) Epoch 2, batch 27650, giga_loss[loss=0.374, simple_loss=0.4255, pruned_loss=0.1612, over 28450.00 frames. ], tot_loss[loss=0.3836, simple_loss=0.4258, pruned_loss=0.1707, over 5668280.93 frames. ], libri_tot_loss[loss=0.3926, simple_loss=0.428, pruned_loss=0.1786, over 5709106.82 frames. ], giga_tot_loss[loss=0.3835, simple_loss=0.4262, pruned_loss=0.1704, over 5661953.24 frames. ], batch size: 65, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:45:48,771 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=73357.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 06:45:56,949 INFO [train.py:968] (0/2) Epoch 2, batch 27700, giga_loss[loss=0.3996, simple_loss=0.442, pruned_loss=0.1786, over 28932.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.4249, pruned_loss=0.1708, over 5666159.50 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4282, pruned_loss=0.1787, over 5706628.16 frames. ], giga_tot_loss[loss=0.3826, simple_loss=0.4248, pruned_loss=0.1702, over 5660903.14 frames. ], batch size: 186, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:46:05,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.814e+02 1.592e+03 2.051e+03 2.745e+03 4.300e+03, threshold=4.102e+03, percent-clipped=1.0
+2023-03-01 06:46:10,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-01 06:46:44,075 INFO [train.py:968] (0/2) Epoch 2, batch 27750, giga_loss[loss=0.3494, simple_loss=0.4031, pruned_loss=0.1478, over 28705.00 frames. ], tot_loss[loss=0.3824, simple_loss=0.4237, pruned_loss=0.1705, over 5666595.95 frames. ], libri_tot_loss[loss=0.3925, simple_loss=0.4281, pruned_loss=0.1785, over 5711546.26 frames. ], giga_tot_loss[loss=0.3819, simple_loss=0.4237, pruned_loss=0.1701, over 5656784.65 frames. ], batch size: 284, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:47:31,653 INFO [train.py:968] (0/2) Epoch 2, batch 27800, giga_loss[loss=0.3144, simple_loss=0.3868, pruned_loss=0.121, over 28920.00 frames. ], tot_loss[loss=0.3789, simple_loss=0.422, pruned_loss=0.1679, over 5660527.38 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.4284, pruned_loss=0.1787, over 5704586.07 frames. ], giga_tot_loss[loss=0.378, simple_loss=0.4217, pruned_loss=0.1671, over 5657600.01 frames. ], batch size: 174, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:47:41,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.978e+02 1.370e+03 2.116e+03 2.987e+03 8.053e+03, threshold=4.232e+03, percent-clipped=10.0
+2023-03-01 06:47:52,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1953, 1.2649, 1.0787, 0.7721], device='cuda:0'), covar=tensor([0.0377, 0.0383, 0.0312, 0.0451], device='cuda:0'), in_proj_covar=tensor([0.1061, 0.0799, 0.0892, 0.0897], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 06:48:23,354 INFO [train.py:968] (0/2) Epoch 2, batch 27850, libri_loss[loss=0.4031, simple_loss=0.4386, pruned_loss=0.1838, over 26172.00 frames. ], tot_loss[loss=0.3749, simple_loss=0.4195, pruned_loss=0.1651, over 5662827.04 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4283, pruned_loss=0.1786, over 5704302.04 frames. ], giga_tot_loss[loss=0.374, simple_loss=0.4191, pruned_loss=0.1644, over 5660234.96 frames. ], batch size: 137, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:49:16,143 INFO [train.py:968] (0/2) Epoch 2, batch 27900, giga_loss[loss=0.4554, simple_loss=0.462, pruned_loss=0.2244, over 27559.00 frames. ], tot_loss[loss=0.3729, simple_loss=0.4178, pruned_loss=0.164, over 5658129.23 frames. ], libri_tot_loss[loss=0.3924, simple_loss=0.428, pruned_loss=0.1784, over 5708233.79 frames. ], giga_tot_loss[loss=0.3721, simple_loss=0.4176, pruned_loss=0.1634, over 5651911.33 frames. ], batch size: 472, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:49:25,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.563e+02 1.457e+03 2.003e+03 2.654e+03 8.100e+03, threshold=4.006e+03, percent-clipped=2.0
+2023-03-01 06:50:14,596 INFO [train.py:968] (0/2) Epoch 2, batch 27950, giga_loss[loss=0.3524, simple_loss=0.3785, pruned_loss=0.1631, over 23396.00 frames. ], tot_loss[loss=0.3703, simple_loss=0.4143, pruned_loss=0.1631, over 5653196.52 frames. ], libri_tot_loss[loss=0.393, simple_loss=0.4286, pruned_loss=0.1787, over 5710845.96 frames. ], giga_tot_loss[loss=0.3688, simple_loss=0.4135, pruned_loss=0.1621, over 5645366.75 frames. ], batch size: 705, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:50:22,249 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-01 06:51:03,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:51:08,715 INFO [train.py:968] (0/2) Epoch 2, batch 28000, giga_loss[loss=0.4567, simple_loss=0.471, pruned_loss=0.2212, over 28338.00 frames. ], tot_loss[loss=0.3706, simple_loss=0.4143, pruned_loss=0.1634, over 5648911.00 frames. ], libri_tot_loss[loss=0.3932, simple_loss=0.4287, pruned_loss=0.1788, over 5704830.26 frames. ], giga_tot_loss[loss=0.3689, simple_loss=0.4133, pruned_loss=0.1623, over 5647772.04 frames. ], batch size: 368, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:51:18,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.599e+02 1.651e+03 2.219e+03 2.785e+03 7.727e+03, threshold=4.439e+03, percent-clipped=9.0
+2023-03-01 06:51:23,860 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=73686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:51:48,529 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-01 06:51:59,039 INFO [train.py:968] (0/2) Epoch 2, batch 28050, libri_loss[loss=0.4309, simple_loss=0.4572, pruned_loss=0.2023, over 29533.00 frames. ], tot_loss[loss=0.3711, simple_loss=0.4157, pruned_loss=0.1633, over 5660395.62 frames. ], libri_tot_loss[loss=0.393, simple_loss=0.4286, pruned_loss=0.1787, over 5709054.36 frames. ], giga_tot_loss[loss=0.3692, simple_loss=0.4145, pruned_loss=0.1619, over 5653955.23 frames. ], batch size: 84, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:52:11,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73732.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 06:52:49,899 INFO [train.py:968] (0/2) Epoch 2, batch 28100, giga_loss[loss=0.3299, simple_loss=0.3874, pruned_loss=0.1362, over 28435.00 frames. ], tot_loss[loss=0.3719, simple_loss=0.4165, pruned_loss=0.1637, over 5651169.50 frames. ], libri_tot_loss[loss=0.3932, simple_loss=0.4287, pruned_loss=0.1789, over 5703145.23 frames. ], giga_tot_loss[loss=0.3698, simple_loss=0.4153, pruned_loss=0.1622, over 5649842.09 frames. ], batch size: 78, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:52:55,812 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-01 06:53:01,299 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.764e+02 1.488e+03 1.868e+03 2.645e+03 6.670e+03, threshold=3.737e+03, percent-clipped=6.0
+2023-03-01 06:53:26,074 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:53:28,172 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:53:39,966 INFO [train.py:968] (0/2) Epoch 2, batch 28150, giga_loss[loss=0.4031, simple_loss=0.4377, pruned_loss=0.1843, over 28884.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.4159, pruned_loss=0.1637, over 5645770.63 frames. ], libri_tot_loss[loss=0.3937, simple_loss=0.429, pruned_loss=0.1792, over 5705167.78 frames. ], giga_tot_loss[loss=0.3694, simple_loss=0.4146, pruned_loss=0.1621, over 5642573.06 frames. ], batch size: 145, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:53:54,994 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:53:57,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4587, 3.8636, 4.1577, 1.7532], device='cuda:0'), covar=tensor([0.0445, 0.0403, 0.0769, 0.1931], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0561, 0.0813, 0.0563], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 06:53:59,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3582, 1.6872, 1.2654, 0.5118], device='cuda:0'), covar=tensor([0.0903, 0.0659, 0.1106, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.1180, 0.1146, 0.1181, 0.1016], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 06:54:25,993 INFO [train.py:968] (0/2) Epoch 2, batch 28200, giga_loss[loss=0.3852, simple_loss=0.4262, pruned_loss=0.1721, over 28777.00 frames. ], tot_loss[loss=0.3738, simple_loss=0.4172, pruned_loss=0.1653, over 5655756.93 frames. ], libri_tot_loss[loss=0.3944, simple_loss=0.4296, pruned_loss=0.1796, over 5713886.77 frames. ], giga_tot_loss[loss=0.3705, simple_loss=0.415, pruned_loss=0.163, over 5643049.23 frames. ], batch size: 199, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:54:26,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 06:54:30,403 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73875.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 06:54:32,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73878.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 06:54:34,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.104e+02 1.606e+03 2.156e+03 2.649e+03 5.718e+03, threshold=4.311e+03, percent-clipped=7.0
+2023-03-01 06:55:01,745 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73907.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 06:55:12,045 INFO [train.py:968] (0/2) Epoch 2, batch 28250, giga_loss[loss=0.3766, simple_loss=0.4226, pruned_loss=0.1653, over 28993.00 frames. ], tot_loss[loss=0.3768, simple_loss=0.4194, pruned_loss=0.1671, over 5645094.29 frames. ], libri_tot_loss[loss=0.395, simple_loss=0.4301, pruned_loss=0.1799, over 5707414.74 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.417, pruned_loss=0.1647, over 5638788.75 frames. ], batch size: 128, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:55:59,199 INFO [train.py:968] (0/2) Epoch 2, batch 28300, giga_loss[loss=0.3712, simple_loss=0.4262, pruned_loss=0.1581, over 28875.00 frames. ], tot_loss[loss=0.3785, simple_loss=0.4213, pruned_loss=0.1678, over 5653380.87 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4296, pruned_loss=0.1794, over 5706990.77 frames. ], giga_tot_loss[loss=0.3758, simple_loss=0.4195, pruned_loss=0.166, over 5647180.92 frames. ], batch size: 199, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:56:09,370 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 06:56:09,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.407e+02 1.570e+03 1.972e+03 2.747e+03 7.838e+03, threshold=3.944e+03, percent-clipped=7.0
+2023-03-01 06:56:13,775 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=73983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:56:31,424 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-74000.pt
+2023-03-01 06:56:43,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0854, 2.4438, 2.3926, 2.1270], device='cuda:0'), covar=tensor([0.0736, 0.1427, 0.0973, 0.1067], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0819, 0.0635, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 06:56:53,080 INFO [train.py:968] (0/2) Epoch 2, batch 28350, giga_loss[loss=0.3947, simple_loss=0.4329, pruned_loss=0.1783, over 28837.00 frames. ], tot_loss[loss=0.382, simple_loss=0.4235, pruned_loss=0.1702, over 5651335.33 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4298, pruned_loss=0.1795, over 5708822.65 frames. ], giga_tot_loss[loss=0.3797, simple_loss=0.422, pruned_loss=0.1687, over 5644339.58 frames. ], batch size: 186, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:57:11,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8960, 1.5811, 1.5925, 1.4868], device='cuda:0'), covar=tensor([0.0872, 0.1618, 0.1251, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0809, 0.0629, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 06:57:30,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:57:38,297 INFO [train.py:968] (0/2) Epoch 2, batch 28400, giga_loss[loss=0.4102, simple_loss=0.4181, pruned_loss=0.2011, over 23602.00 frames. ], tot_loss[loss=0.3849, simple_loss=0.4251, pruned_loss=0.1724, over 5655697.91 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.4298, pruned_loss=0.1797, over 5715376.96 frames. ], giga_tot_loss[loss=0.3823, simple_loss=0.4236, pruned_loss=0.1705, over 5641869.84 frames. ], batch size: 705, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 06:57:50,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.668e+03 2.112e+03 2.894e+03 6.133e+03, threshold=4.223e+03, percent-clipped=8.0
+2023-03-01 06:58:02,534 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-01 06:58:29,645 INFO [train.py:968] (0/2) Epoch 2, batch 28450, giga_loss[loss=0.3817, simple_loss=0.4364, pruned_loss=0.1635, over 28662.00 frames. ], tot_loss[loss=0.387, simple_loss=0.4272, pruned_loss=0.1734, over 5653583.31 frames. ], libri_tot_loss[loss=0.3952, simple_loss=0.4302, pruned_loss=0.1801, over 5710705.11 frames. ], giga_tot_loss[loss=0.3841, simple_loss=0.4255, pruned_loss=0.1714, over 5644314.67 frames. ], batch size: 262, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 06:58:39,760 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=74125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:59:09,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3918, 1.5740, 1.3012, 0.7530], device='cuda:0'), covar=tensor([0.0534, 0.0372, 0.0309, 0.0488], device='cuda:0'), in_proj_covar=tensor([0.1025, 0.0777, 0.0877, 0.0876], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 06:59:22,270 INFO [train.py:968] (0/2) Epoch 2, batch 28500, giga_loss[loss=0.3905, simple_loss=0.4353, pruned_loss=0.1728, over 28917.00 frames. ], tot_loss[loss=0.3875, simple_loss=0.4278, pruned_loss=0.1736, over 5659097.70 frames. ], libri_tot_loss[loss=0.3948, simple_loss=0.4298, pruned_loss=0.1799, over 5716981.57 frames. ], giga_tot_loss[loss=0.3853, simple_loss=0.4267, pruned_loss=0.1719, over 5644003.66 frames. ], batch size: 227, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 06:59:31,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.666e+03 2.383e+03 3.071e+03 6.357e+03, threshold=4.765e+03, percent-clipped=6.0
+2023-03-01 06:59:38,481 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-01 06:59:57,778 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:59:59,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:00:11,323 INFO [train.py:968] (0/2) Epoch 2, batch 28550, giga_loss[loss=0.3814, simple_loss=0.4182, pruned_loss=0.1723, over 28013.00 frames. ], tot_loss[loss=0.3878, simple_loss=0.4276, pruned_loss=0.174, over 5645258.71 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.4297, pruned_loss=0.1797, over 5709132.28 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4267, pruned_loss=0.1726, over 5638611.61 frames. ], batch size: 412, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:00:30,501 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:01:05,386 INFO [train.py:968] (0/2) Epoch 2, batch 28600, giga_loss[loss=0.3846, simple_loss=0.4272, pruned_loss=0.171, over 28323.00 frames. ], tot_loss[loss=0.3881, simple_loss=0.4273, pruned_loss=0.1745, over 5625902.80 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.4297, pruned_loss=0.1797, over 5704117.13 frames. ], giga_tot_loss[loss=0.3866, simple_loss=0.4266, pruned_loss=0.1733, over 5623322.81 frames. ], batch size: 71, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:01:23,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.625e+03 2.098e+03 2.947e+03 8.549e+03, threshold=4.196e+03, percent-clipped=7.0
+2023-03-01 07:01:55,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3898, 1.4421, 1.1038, 1.0651], device='cuda:0'), covar=tensor([0.0507, 0.0461, 0.0368, 0.0489], device='cuda:0'), in_proj_covar=tensor([0.1048, 0.0777, 0.0882, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 07:02:05,301 INFO [train.py:968] (0/2) Epoch 2, batch 28650, giga_loss[loss=0.3963, simple_loss=0.4313, pruned_loss=0.1806, over 28691.00 frames. ], tot_loss[loss=0.3868, simple_loss=0.4259, pruned_loss=0.1738, over 5630897.48 frames. ], libri_tot_loss[loss=0.3948, simple_loss=0.43, pruned_loss=0.1798, over 5705671.10 frames. ], giga_tot_loss[loss=0.3852, simple_loss=0.425, pruned_loss=0.1727, over 5626062.55 frames. ], batch size: 242, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:02:48,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:02:51,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3911, 1.9655, 1.3787, 0.7449], device='cuda:0'), covar=tensor([0.1711, 0.0898, 0.1253, 0.1911], device='cuda:0'), in_proj_covar=tensor([0.1192, 0.1172, 0.1198, 0.1046], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 07:02:57,486 INFO [train.py:968] (0/2) Epoch 2, batch 28700, giga_loss[loss=0.3656, simple_loss=0.4109, pruned_loss=0.1602, over 28902.00 frames. ], tot_loss[loss=0.3846, simple_loss=0.4239, pruned_loss=0.1727, over 5629174.06 frames. ], libri_tot_loss[loss=0.3944, simple_loss=0.4296, pruned_loss=0.1796, over 5697712.26 frames. ], giga_tot_loss[loss=0.3836, simple_loss=0.4235, pruned_loss=0.1719, over 5630813.45 frames. ], batch size: 227, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:03:07,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.088e+02 1.626e+03 2.267e+03 3.108e+03 8.283e+03, threshold=4.534e+03, percent-clipped=12.0
+2023-03-01 07:03:18,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0560, 3.5478, 3.7643, 1.6689], device='cuda:0'), covar=tensor([0.0504, 0.0408, 0.0790, 0.1773], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0565, 0.0823, 0.0564], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 07:03:43,022 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-01 07:03:47,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2558, 1.6353, 1.2469, 1.3778], device='cuda:0'), covar=tensor([0.0953, 0.0468, 0.0457, 0.1082], device='cuda:0'), in_proj_covar=tensor([0.0287, 0.0193, 0.0195, 0.0346], device='cuda:0'), out_proj_covar=tensor([0.0033, 0.0025, 0.0022, 0.0038], device='cuda:0')
+2023-03-01 07:03:48,452 INFO [train.py:968] (0/2) Epoch 2, batch 28750, giga_loss[loss=0.3489, simple_loss=0.4063, pruned_loss=0.1458, over 28367.00 frames. ], tot_loss[loss=0.3838, simple_loss=0.423, pruned_loss=0.1723, over 5642277.25 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.4298, pruned_loss=0.1797, over 5698955.13 frames. ], giga_tot_loss[loss=0.3826, simple_loss=0.4224, pruned_loss=0.1714, over 5641694.45 frames. ], batch size: 71, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:04:27,071 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=74457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:04:39,179 INFO [train.py:968] (0/2) Epoch 2, batch 28800, giga_loss[loss=0.443, simple_loss=0.4504, pruned_loss=0.2178, over 26544.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.4224, pruned_loss=0.172, over 5642296.12 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4296, pruned_loss=0.1795, over 5699001.20 frames. ], giga_tot_loss[loss=0.3824, simple_loss=0.422, pruned_loss=0.1714, over 5641040.50 frames. ], batch size: 555, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 07:04:50,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.892e+02 1.525e+03 1.845e+03 2.451e+03 6.314e+03, threshold=3.689e+03, percent-clipped=4.0
+2023-03-01 07:04:52,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1369, 1.2553, 1.1439, 1.4643], device='cuda:0'), covar=tensor([0.1940, 0.1918, 0.1601, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0990, 0.0829, 0.0894, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 07:05:07,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2930, 2.8870, 1.7886, 1.1516], device='cuda:0'), covar=tensor([0.0476, 0.0242, 0.0313, 0.0522], device='cuda:0'), in_proj_covar=tensor([0.1044, 0.0786, 0.0879, 0.0891], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 07:05:11,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:05:11,886 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:05:14,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:05:15,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4622, 1.9026, 1.3704, 1.6052], device='cuda:0'), covar=tensor([0.0987, 0.0382, 0.0448, 0.1102], device='cuda:0'), in_proj_covar=tensor([0.0285, 0.0194, 0.0193, 0.0342], device='cuda:0'), out_proj_covar=tensor([0.0033, 0.0025, 0.0022, 0.0037], device='cuda:0')
+2023-03-01 07:05:29,645 INFO [train.py:968] (0/2) Epoch 2, batch 28850, giga_loss[loss=0.4184, simple_loss=0.4253, pruned_loss=0.2057, over 23619.00 frames. ], tot_loss[loss=0.3837, simple_loss=0.423, pruned_loss=0.1722, over 5650197.87 frames. ], libri_tot_loss[loss=0.3941, simple_loss=0.4296, pruned_loss=0.1794, over 5703535.90 frames. ], giga_tot_loss[loss=0.3829, simple_loss=0.4225, pruned_loss=0.1716, over 5643974.96 frames. ], batch size: 705, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 07:05:34,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7073, 2.0730, 1.7457, 1.7097], device='cuda:0'), covar=tensor([0.1262, 0.1515, 0.1053, 0.0721], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0829, 0.0718, 0.0605], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 07:05:41,595 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:06:17,976 INFO [train.py:968] (0/2) Epoch 2, batch 28900, giga_loss[loss=0.378, simple_loss=0.4228, pruned_loss=0.1666, over 28853.00 frames. ], tot_loss[loss=0.3873, simple_loss=0.4258, pruned_loss=0.1745, over 5658489.96 frames. ], libri_tot_loss[loss=0.395, simple_loss=0.4302, pruned_loss=0.1799, over 5698520.84 frames. ], giga_tot_loss[loss=0.3856, simple_loss=0.4246, pruned_loss=0.1733, over 5655602.80 frames. ], batch size: 99, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:06:35,206 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.094e+02 1.582e+03 2.037e+03 3.037e+03 1.294e+04, threshold=4.074e+03, percent-clipped=16.0
+2023-03-01 07:07:15,092 INFO [train.py:968] (0/2) Epoch 2, batch 28950, giga_loss[loss=0.4046, simple_loss=0.4411, pruned_loss=0.1841, over 28714.00 frames. ], tot_loss[loss=0.3872, simple_loss=0.4257, pruned_loss=0.1743, over 5661651.26 frames. ], libri_tot_loss[loss=0.3956, simple_loss=0.4306, pruned_loss=0.1803, over 5701654.27 frames. ], giga_tot_loss[loss=0.3851, simple_loss=0.4244, pruned_loss=0.1729, over 5655816.82 frames. ], batch size: 284, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:07:28,636 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=74633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:07:37,521 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:07:39,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:08:02,488 INFO [train.py:968] (0/2) Epoch 2, batch 29000, giga_loss[loss=0.4815, simple_loss=0.4805, pruned_loss=0.2412, over 26626.00 frames. ], tot_loss[loss=0.3905, simple_loss=0.4277, pruned_loss=0.1766, over 5675168.85 frames. ], libri_tot_loss[loss=0.3962, simple_loss=0.4309, pruned_loss=0.1808, over 5706308.87 frames. ], giga_tot_loss[loss=0.3881, simple_loss=0.4262, pruned_loss=0.175, over 5665467.46 frames. ], batch size: 555, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:08:06,532 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:08:12,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.862e+02 1.634e+03 2.071e+03 3.209e+03 7.113e+03, threshold=4.143e+03, percent-clipped=11.0
+2023-03-01 07:08:20,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6730, 2.2267, 1.8526, 1.7824], device='cuda:0'), covar=tensor([0.1427, 0.1672, 0.1164, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0836, 0.0726, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 07:08:47,270 INFO [train.py:968] (0/2) Epoch 2, batch 29050, giga_loss[loss=0.3834, simple_loss=0.4251, pruned_loss=0.1708, over 28934.00 frames. ], tot_loss[loss=0.3906, simple_loss=0.4279, pruned_loss=0.1766, over 5680010.00 frames. ], libri_tot_loss[loss=0.3969, simple_loss=0.4315, pruned_loss=0.1812, over 5711256.92 frames. ], giga_tot_loss[loss=0.3878, simple_loss=0.4261, pruned_loss=0.1748, over 5666629.10 frames. ], batch size: 199, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:09:27,946 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-01 07:09:40,857 INFO [train.py:968] (0/2) Epoch 2, batch 29100, giga_loss[loss=0.3436, simple_loss=0.4119, pruned_loss=0.1376, over 29021.00 frames. ], tot_loss[loss=0.3902, simple_loss=0.4281, pruned_loss=0.1762, over 5671915.02 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.4316, pruned_loss=0.1813, over 5711979.64 frames. ], giga_tot_loss[loss=0.3879, simple_loss=0.4266, pruned_loss=0.1746, over 5660724.40 frames. ], batch size: 145, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:09:53,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.420e+02 1.718e+03 2.225e+03 3.459e+03 8.089e+03, threshold=4.451e+03, percent-clipped=12.0
+2023-03-01 07:10:28,272 INFO [train.py:968] (0/2) Epoch 2, batch 29150, giga_loss[loss=0.4911, simple_loss=0.491, pruned_loss=0.2456, over 28316.00 frames. ], tot_loss[loss=0.3894, simple_loss=0.4279, pruned_loss=0.1755, over 5682527.23 frames. ], libri_tot_loss[loss=0.3967, simple_loss=0.4313, pruned_loss=0.1811, over 5716896.46 frames. ], giga_tot_loss[loss=0.3877, simple_loss=0.4268, pruned_loss=0.1743, over 5668447.80 frames. ], batch size: 368, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:10:41,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:11:07,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6937, 2.1744, 1.9036, 1.8251], device='cuda:0'), covar=tensor([0.1327, 0.1456, 0.1072, 0.0708], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0821, 0.0721, 0.0605], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 07:11:15,242 INFO [train.py:968] (0/2) Epoch 2, batch 29200, giga_loss[loss=0.401, simple_loss=0.436, pruned_loss=0.183, over 29007.00 frames. ], tot_loss[loss=0.3902, simple_loss=0.4283, pruned_loss=0.176, over 5684530.46 frames. ], libri_tot_loss[loss=0.3967, simple_loss=0.4314, pruned_loss=0.181, over 5725528.96 frames. ], giga_tot_loss[loss=0.3885, simple_loss=0.4272, pruned_loss=0.1749, over 5663472.88 frames. ], batch size: 213, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 07:11:25,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.570e+03 1.972e+03 2.698e+03 6.060e+03, threshold=3.944e+03, percent-clipped=6.0
+2023-03-01 07:11:59,682 INFO [train.py:968] (0/2) Epoch 2, batch 29250, giga_loss[loss=0.3533, simple_loss=0.4085, pruned_loss=0.149, over 29001.00 frames. ], tot_loss[loss=0.3925, simple_loss=0.4301, pruned_loss=0.1774, over 5669261.43 frames. ], libri_tot_loss[loss=0.3973, simple_loss=0.4319, pruned_loss=0.1813, over 5712169.20 frames. ], giga_tot_loss[loss=0.3905, simple_loss=0.4287, pruned_loss=0.1761, over 5662505.12 frames. ], batch size: 106, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:12:48,299 INFO [train.py:968] (0/2) Epoch 2, batch 29300, giga_loss[loss=0.38, simple_loss=0.4239, pruned_loss=0.168, over 28896.00 frames. ], tot_loss[loss=0.39, simple_loss=0.4285, pruned_loss=0.1757, over 5657216.75 frames. ], libri_tot_loss[loss=0.3969, simple_loss=0.4315, pruned_loss=0.1811, over 5703957.30 frames. ], giga_tot_loss[loss=0.3886, simple_loss=0.4277, pruned_loss=0.1748, over 5658927.47 frames. ], batch size: 136, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:12:53,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:12:57,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:12:58,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9049, 2.9090, 1.9393, 0.8218], device='cuda:0'), covar=tensor([0.1975, 0.1012, 0.1333, 0.2291], device='cuda:0'), in_proj_covar=tensor([0.1201, 0.1179, 0.1206, 0.1046], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 07:13:01,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.875e+02 1.654e+03 2.031e+03 2.676e+03 5.892e+03, threshold=4.062e+03, percent-clipped=6.0
+2023-03-01 07:13:08,084 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=74988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:13:22,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-01 07:13:27,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:13:28,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:13:38,794 INFO [train.py:968] (0/2) Epoch 2, batch 29350, libri_loss[loss=0.3781, simple_loss=0.4306, pruned_loss=0.1628, over 29752.00 frames. ], tot_loss[loss=0.3892, simple_loss=0.4285, pruned_loss=0.1749, over 5638758.41 frames. ], libri_tot_loss[loss=0.3968, simple_loss=0.4316, pruned_loss=0.181, over 5699645.98 frames. ], giga_tot_loss[loss=0.3879, simple_loss=0.4277, pruned_loss=0.1741, over 5641731.98 frames. ], batch size: 87, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:14:28,844 INFO [train.py:968] (0/2) Epoch 2, batch 29400, giga_loss[loss=0.3909, simple_loss=0.4356, pruned_loss=0.1731, over 28725.00 frames. ], tot_loss[loss=0.39, simple_loss=0.4295, pruned_loss=0.1753, over 5636457.70 frames. ], libri_tot_loss[loss=0.3972, simple_loss=0.4319, pruned_loss=0.1812, over 5696308.15 frames. ], giga_tot_loss[loss=0.3884, simple_loss=0.4285, pruned_loss=0.1741, over 5640555.45 frames. ], batch size: 242, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:14:36,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 07:14:40,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.420e+03 1.747e+03 2.387e+03 6.171e+03, threshold=3.495e+03, percent-clipped=6.0
+2023-03-01 07:15:15,863 INFO [train.py:968] (0/2) Epoch 2, batch 29450, giga_loss[loss=0.3804, simple_loss=0.425, pruned_loss=0.1679, over 28193.00 frames. ], tot_loss[loss=0.3886, simple_loss=0.4286, pruned_loss=0.1743, over 5650753.66 frames. ], libri_tot_loss[loss=0.3968, simple_loss=0.4316, pruned_loss=0.181, over 5702743.98 frames. ], giga_tot_loss[loss=0.3874, simple_loss=0.4279, pruned_loss=0.1735, over 5646992.23 frames. ], batch size: 368, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:15:42,517 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7046, 4.2184, 4.4471, 2.7413], device='cuda:0'), covar=tensor([0.0415, 0.0398, 0.0893, 0.1333], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0581, 0.0844, 0.0577], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 07:15:46,513 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:15:48,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:15:58,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5756, 2.6993, 1.5080, 1.4709], device='cuda:0'), covar=tensor([0.0898, 0.0541, 0.0916, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0463, 0.0335, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 07:16:01,780 INFO [train.py:968] (0/2) Epoch 2, batch 29500, giga_loss[loss=0.4243, simple_loss=0.4533, pruned_loss=0.1976, over 28735.00 frames. ], tot_loss[loss=0.3869, simple_loss=0.4272, pruned_loss=0.1733, over 5649475.22 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4314, pruned_loss=0.1807, over 5694345.44 frames. ], giga_tot_loss[loss=0.3861, simple_loss=0.4268, pruned_loss=0.1727, over 5653193.34 frames. ], batch size: 284, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:16:14,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.680e+03 2.086e+03 3.083e+03 6.954e+03, threshold=4.172e+03, percent-clipped=19.0
+2023-03-01 07:16:14,905 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:16:49,554 INFO [train.py:968] (0/2) Epoch 2, batch 29550, giga_loss[loss=0.3665, simple_loss=0.4185, pruned_loss=0.1572, over 28590.00 frames. ], tot_loss[loss=0.3868, simple_loss=0.4271, pruned_loss=0.1732, over 5649266.10 frames. ], libri_tot_loss[loss=0.3961, simple_loss=0.4313, pruned_loss=0.1805, over 5698089.68 frames. ], giga_tot_loss[loss=0.3862, simple_loss=0.4268, pruned_loss=0.1728, over 5647594.15 frames. ], batch size: 78, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:16:57,871 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-01 07:17:40,154 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-01 07:17:42,658 INFO [train.py:968] (0/2) Epoch 2, batch 29600, giga_loss[loss=0.446, simple_loss=0.4616, pruned_loss=0.2152, over 27567.00 frames. ], tot_loss[loss=0.3882, simple_loss=0.4281, pruned_loss=0.1742, over 5654327.44 frames. ], libri_tot_loss[loss=0.3963, simple_loss=0.4314, pruned_loss=0.1806, over 5696780.44 frames. ], giga_tot_loss[loss=0.3875, simple_loss=0.4277, pruned_loss=0.1736, over 5653758.71 frames. ], batch size: 472, lr: 1.26e-02, grad_scale: 8.0
+2023-03-01 07:17:58,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.064e+02 1.753e+03 2.344e+03 3.138e+03 6.365e+03, threshold=4.688e+03, percent-clipped=7.0
+2023-03-01 07:18:16,914 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=75302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:18:34,693 INFO [train.py:968] (0/2) Epoch 2, batch 29650, giga_loss[loss=0.3861, simple_loss=0.43, pruned_loss=0.1711, over 28795.00 frames. ], tot_loss[loss=0.3867, simple_loss=0.426, pruned_loss=0.1737, over 5652200.83 frames. ], libri_tot_loss[loss=0.3964, simple_loss=0.4315, pruned_loss=0.1807, over 5698644.77 frames. ], giga_tot_loss[loss=0.3859, simple_loss=0.4256, pruned_loss=0.1731, over 5649676.31 frames. ], batch size: 284, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:19:07,015 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=75357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:19:12,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:19:18,389 INFO [train.py:968] (0/2) Epoch 2, batch 29700, giga_loss[loss=0.4151, simple_loss=0.445, pruned_loss=0.1926, over 28889.00 frames. ], tot_loss[loss=0.3878, simple_loss=0.4269, pruned_loss=0.1743, over 5661610.49 frames. ], libri_tot_loss[loss=0.3976, simple_loss=0.4323, pruned_loss=0.1814, over 5693101.39 frames. ], giga_tot_loss[loss=0.3857, simple_loss=0.4256, pruned_loss=0.1729, over 5663161.25 frames. ], batch size: 227, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:19:34,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.123e+02 1.511e+03 1.869e+03 2.507e+03 9.600e+03, threshold=3.738e+03, percent-clipped=4.0
+2023-03-01 07:20:09,072 INFO [train.py:968] (0/2) Epoch 2, batch 29750, libri_loss[loss=0.3828, simple_loss=0.4263, pruned_loss=0.1696, over 29515.00 frames. ], tot_loss[loss=0.3919, simple_loss=0.4298, pruned_loss=0.1771, over 5659855.29 frames. ], libri_tot_loss[loss=0.3976, simple_loss=0.4324, pruned_loss=0.1814, over 5696055.66 frames. ], giga_tot_loss[loss=0.3902, simple_loss=0.4286, pruned_loss=0.1759, over 5657720.31 frames. ], batch size: 80, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:21:00,432 INFO [train.py:968] (0/2) Epoch 2, batch 29800, giga_loss[loss=0.3496, simple_loss=0.4068, pruned_loss=0.1462, over 28651.00 frames. ], tot_loss[loss=0.3902, simple_loss=0.4283, pruned_loss=0.176, over 5640423.64 frames. ], libri_tot_loss[loss=0.3971, simple_loss=0.4321, pruned_loss=0.1811, over 5690211.95 frames. ], giga_tot_loss[loss=0.389, simple_loss=0.4275, pruned_loss=0.1752, over 5643550.65 frames. ], batch size: 262, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:21:13,957 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.582e+03 2.035e+03 2.893e+03 6.829e+03, threshold=4.069e+03, percent-clipped=8.0
+2023-03-01 07:21:35,290 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:21:39,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:21:47,146 INFO [train.py:968] (0/2) Epoch 2, batch 29850, giga_loss[loss=0.3993, simple_loss=0.4125, pruned_loss=0.193, over 23679.00 frames. ], tot_loss[loss=0.3906, simple_loss=0.4287, pruned_loss=0.1762, over 5646641.48 frames. ], libri_tot_loss[loss=0.3978, simple_loss=0.4326, pruned_loss=0.1815, over 5694691.10 frames. ], giga_tot_loss[loss=0.3889, simple_loss=0.4275, pruned_loss=0.1752, over 5644116.12 frames. ], batch size: 705, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:21:49,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5083, 1.3323, 1.3692, 1.3139], device='cuda:0'), covar=tensor([0.0848, 0.1184, 0.1189, 0.0982], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0806, 0.0651, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 07:22:07,950 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:22:11,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4121, 1.4574, 1.2358, 0.9173], device='cuda:0'), covar=tensor([0.0408, 0.0358, 0.0290, 0.0456], device='cuda:0'), in_proj_covar=tensor([0.1069, 0.0809, 0.0880, 0.0928], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 07:22:21,193 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 07:22:34,321 INFO [train.py:968] (0/2) Epoch 2, batch 29900, giga_loss[loss=0.3544, simple_loss=0.4134, pruned_loss=0.1477, over 28736.00 frames. ], tot_loss[loss=0.3901, simple_loss=0.429, pruned_loss=0.1755, over 5645599.34 frames. ], libri_tot_loss[loss=0.3978, simple_loss=0.4328, pruned_loss=0.1814, over 5683902.91 frames. ], giga_tot_loss[loss=0.3886, simple_loss=0.4279, pruned_loss=0.1746, over 5651717.22 frames. ], batch size: 262, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:22:51,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.522e+02 1.578e+03 2.026e+03 2.623e+03 6.110e+03, threshold=4.052e+03, percent-clipped=4.0
+2023-03-01 07:23:20,944 INFO [train.py:968] (0/2) Epoch 2, batch 29950, giga_loss[loss=0.3763, simple_loss=0.4224, pruned_loss=0.1651, over 28679.00 frames. ], tot_loss[loss=0.3892, simple_loss=0.4286, pruned_loss=0.1749, over 5651804.64 frames. ], libri_tot_loss[loss=0.3978, simple_loss=0.4325, pruned_loss=0.1815, over 5689118.26 frames. ], giga_tot_loss[loss=0.3878, simple_loss=0.4278, pruned_loss=0.1739, over 5650676.61 frames. ], batch size: 242, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:23:24,414 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=75621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:24:07,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=75663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:24:13,758 INFO [train.py:968] (0/2) Epoch 2, batch 30000, giga_loss[loss=0.3626, simple_loss=0.4153, pruned_loss=0.1549, over 28988.00 frames. ], tot_loss[loss=0.3871, simple_loss=0.4269, pruned_loss=0.1736, over 5656643.17 frames. ], libri_tot_loss[loss=0.3971, simple_loss=0.432, pruned_loss=0.1811, over 5691111.00 frames. ], giga_tot_loss[loss=0.3865, simple_loss=0.4267, pruned_loss=0.1731, over 5653201.69 frames. ], batch size: 227, lr: 1.26e-02, grad_scale: 8.0
+2023-03-01 07:24:13,763 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 07:24:21,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4345, 2.4890, 1.4007, 1.3203], device='cuda:0'), covar=tensor([0.0968, 0.0518, 0.1047, 0.1542], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0461, 0.0330, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 07:24:22,530 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.2678, simple_loss=0.3662, pruned_loss=0.08473, over 944034.00 frames. 
+2023-03-01 07:24:22,530 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19537MB
+2023-03-01 07:24:31,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:24:36,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.773e+03 2.176e+03 2.741e+03 8.041e+03, threshold=4.352e+03, percent-clipped=14.0
+2023-03-01 07:24:41,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3969, 1.4135, 1.1222, 1.2414], device='cuda:0'), covar=tensor([0.0672, 0.0633, 0.1108, 0.0889], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0521, 0.0545, 0.0469], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 07:25:07,937 INFO [train.py:968] (0/2) Epoch 2, batch 30050, giga_loss[loss=0.4532, simple_loss=0.4594, pruned_loss=0.2235, over 26653.00 frames. ], tot_loss[loss=0.3874, simple_loss=0.427, pruned_loss=0.1739, over 5656962.72 frames. ], libri_tot_loss[loss=0.3967, simple_loss=0.4318, pruned_loss=0.1807, over 5687555.08 frames. ], giga_tot_loss[loss=0.3869, simple_loss=0.4268, pruned_loss=0.1735, over 5656951.36 frames. ], batch size: 555, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:25:17,914 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:25:51,370 INFO [train.py:968] (0/2) Epoch 2, batch 30100, giga_loss[loss=0.3198, simple_loss=0.3741, pruned_loss=0.1327, over 28557.00 frames. ], tot_loss[loss=0.3841, simple_loss=0.4244, pruned_loss=0.1719, over 5657805.93 frames. ], libri_tot_loss[loss=0.3972, simple_loss=0.4325, pruned_loss=0.181, over 5682967.31 frames. ], giga_tot_loss[loss=0.3827, simple_loss=0.4233, pruned_loss=0.171, over 5660836.46 frames. ], batch size: 60, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:26:09,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.777e+03 2.253e+03 2.927e+03 5.628e+03, threshold=4.506e+03, percent-clipped=2.0
+2023-03-01 07:26:42,841 INFO [train.py:968] (0/2) Epoch 2, batch 30150, giga_loss[loss=0.3389, simple_loss=0.3904, pruned_loss=0.1436, over 28651.00 frames. ], tot_loss[loss=0.3802, simple_loss=0.4203, pruned_loss=0.17, over 5651863.08 frames. ], libri_tot_loss[loss=0.3976, simple_loss=0.4327, pruned_loss=0.1812, over 5687062.46 frames. ], giga_tot_loss[loss=0.3784, simple_loss=0.419, pruned_loss=0.1689, over 5649858.43 frames. ], batch size: 262, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:26:43,694 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:26:47,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:27:14,852 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:27:33,093 INFO [train.py:968] (0/2) Epoch 2, batch 30200, giga_loss[loss=0.3894, simple_loss=0.4121, pruned_loss=0.1834, over 28590.00 frames. ], tot_loss[loss=0.3786, simple_loss=0.4183, pruned_loss=0.1694, over 5659913.19 frames. ], libri_tot_loss[loss=0.3976, simple_loss=0.4326, pruned_loss=0.1813, over 5690622.56 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4173, pruned_loss=0.1683, over 5654901.64 frames. ], batch size: 85, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:27:38,398 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:27:41,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:27:47,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.579e+02 1.998e+03 2.233e+03 2.898e+03 5.943e+03, threshold=4.466e+03, percent-clipped=6.0
+2023-03-01 07:27:50,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6682, 1.6166, 1.6306, 1.5829], device='cuda:0'), covar=tensor([0.0766, 0.1155, 0.0926, 0.1057], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0808, 0.0646, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 07:28:12,863 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:28:25,117 INFO [train.py:968] (0/2) Epoch 2, batch 30250, giga_loss[loss=0.3467, simple_loss=0.3926, pruned_loss=0.1504, over 28914.00 frames. ], tot_loss[loss=0.3779, simple_loss=0.4172, pruned_loss=0.1693, over 5647486.27 frames. ], libri_tot_loss[loss=0.3974, simple_loss=0.4324, pruned_loss=0.1812, over 5693952.44 frames. ], giga_tot_loss[loss=0.3765, simple_loss=0.4163, pruned_loss=0.1684, over 5640122.91 frames. ], batch size: 106, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:29:15,376 INFO [train.py:968] (0/2) Epoch 2, batch 30300, giga_loss[loss=0.3498, simple_loss=0.4055, pruned_loss=0.1471, over 28596.00 frames. ], tot_loss[loss=0.3747, simple_loss=0.4161, pruned_loss=0.1666, over 5645401.69 frames. ], libri_tot_loss[loss=0.3973, simple_loss=0.4322, pruned_loss=0.1813, over 5686769.62 frames. ], giga_tot_loss[loss=0.3733, simple_loss=0.4154, pruned_loss=0.1656, over 5644480.86 frames. ], batch size: 336, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:29:30,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 1.592e+03 2.087e+03 2.895e+03 6.773e+03, threshold=4.173e+03, percent-clipped=7.0
+2023-03-01 07:29:45,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:29:48,637 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-76000.pt
+2023-03-01 07:30:12,145 INFO [train.py:968] (0/2) Epoch 2, batch 30350, giga_loss[loss=0.3514, simple_loss=0.3955, pruned_loss=0.1536, over 27584.00 frames. ], tot_loss[loss=0.3662, simple_loss=0.4109, pruned_loss=0.1607, over 5636477.25 frames. ], libri_tot_loss[loss=0.3973, simple_loss=0.432, pruned_loss=0.1813, over 5688873.88 frames. ], giga_tot_loss[loss=0.3647, simple_loss=0.4102, pruned_loss=0.1596, over 5633094.76 frames. ], batch size: 472, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:30:31,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:31:03,439 INFO [train.py:968] (0/2) Epoch 2, batch 30400, giga_loss[loss=0.3174, simple_loss=0.385, pruned_loss=0.1249, over 28796.00 frames. ], tot_loss[loss=0.3577, simple_loss=0.406, pruned_loss=0.1547, over 5643896.05 frames. ], libri_tot_loss[loss=0.3971, simple_loss=0.4318, pruned_loss=0.1812, over 5681002.18 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.4054, pruned_loss=0.1538, over 5647272.21 frames. ], batch size: 284, lr: 1.26e-02, grad_scale: 8.0
+2023-03-01 07:31:18,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.794e+02 1.482e+03 1.873e+03 2.328e+03 7.692e+03, threshold=3.746e+03, percent-clipped=7.0
+2023-03-01 07:31:51,883 INFO [train.py:968] (0/2) Epoch 2, batch 30450, libri_loss[loss=0.3786, simple_loss=0.4228, pruned_loss=0.1672, over 29272.00 frames. ], tot_loss[loss=0.354, simple_loss=0.4033, pruned_loss=0.1524, over 5632954.81 frames. ], libri_tot_loss[loss=0.3964, simple_loss=0.4311, pruned_loss=0.1809, over 5667669.13 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4023, pruned_loss=0.1507, over 5644734.55 frames. ], batch size: 94, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:32:11,814 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:32:14,567 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:32:21,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6547, 1.8007, 1.1606, 1.4278], device='cuda:0'), covar=tensor([0.0574, 0.0457, 0.0973, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0510, 0.0548, 0.0476], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 07:32:26,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3145, 2.3308, 1.8879, 1.5803], device='cuda:0'), covar=tensor([0.1105, 0.0308, 0.0417, 0.1242], device='cuda:0'), in_proj_covar=tensor([0.0287, 0.0191, 0.0195, 0.0345], device='cuda:0'), out_proj_covar=tensor([0.0034, 0.0025, 0.0023, 0.0038], device='cuda:0')
+2023-03-01 07:32:42,412 INFO [train.py:968] (0/2) Epoch 2, batch 30500, giga_loss[loss=0.3332, simple_loss=0.3995, pruned_loss=0.1334, over 28568.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.398, pruned_loss=0.1472, over 5641387.20 frames. ], libri_tot_loss[loss=0.396, simple_loss=0.4307, pruned_loss=0.1806, over 5671799.94 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.3971, pruned_loss=0.1455, over 5646536.82 frames. ], batch size: 307, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:32:44,303 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:32:54,388 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:32:56,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:32:58,492 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.471e+02 1.537e+03 1.935e+03 2.557e+03 5.329e+03, threshold=3.870e+03, percent-clipped=6.0
+2023-03-01 07:33:12,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6675, 2.2487, 1.8887, 1.8297], device='cuda:0'), covar=tensor([0.1808, 0.1629, 0.1280, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0807, 0.0723, 0.0618], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 07:33:29,764 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:33:31,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5221, 1.3687, 1.2761, 1.6174], device='cuda:0'), covar=tensor([0.1968, 0.1900, 0.1721, 0.1918], device='cuda:0'), in_proj_covar=tensor([0.0997, 0.0814, 0.0908, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 07:33:36,238 INFO [train.py:968] (0/2) Epoch 2, batch 30550, giga_loss[loss=0.3028, simple_loss=0.3777, pruned_loss=0.114, over 28315.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3962, pruned_loss=0.1425, over 5658003.52 frames. ], libri_tot_loss[loss=0.3955, simple_loss=0.4301, pruned_loss=0.1804, over 5674200.98 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3956, pruned_loss=0.1411, over 5659752.65 frames. ], batch size: 368, lr: 1.25e-02, grad_scale: 2.0
+2023-03-01 07:34:32,363 INFO [train.py:968] (0/2) Epoch 2, batch 30600, giga_loss[loss=0.2754, simple_loss=0.3508, pruned_loss=0.09995, over 28507.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3954, pruned_loss=0.1419, over 5660498.86 frames. ], libri_tot_loss[loss=0.3956, simple_loss=0.4301, pruned_loss=0.1805, over 5675724.66 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3943, pruned_loss=0.1399, over 5660030.13 frames. ], batch size: 71, lr: 1.25e-02, grad_scale: 2.0
+2023-03-01 07:34:50,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.399e+02 1.517e+03 1.955e+03 2.959e+03 1.472e+04, threshold=3.911e+03, percent-clipped=14.0
+2023-03-01 07:35:21,956 INFO [train.py:968] (0/2) Epoch 2, batch 30650, giga_loss[loss=0.3308, simple_loss=0.3968, pruned_loss=0.1324, over 28754.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.392, pruned_loss=0.1389, over 5667063.45 frames. ], libri_tot_loss[loss=0.3942, simple_loss=0.429, pruned_loss=0.1797, over 5678801.54 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3911, pruned_loss=0.137, over 5663347.40 frames. ], batch size: 262, lr: 1.25e-02, grad_scale: 2.0
+2023-03-01 07:35:41,703 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-01 07:36:10,450 INFO [train.py:968] (0/2) Epoch 2, batch 30700, giga_loss[loss=0.2984, simple_loss=0.372, pruned_loss=0.1124, over 28864.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3885, pruned_loss=0.1365, over 5669626.34 frames. ], libri_tot_loss[loss=0.3925, simple_loss=0.4277, pruned_loss=0.1787, over 5684310.97 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3878, pruned_loss=0.1344, over 5661075.49 frames. ], batch size: 174, lr: 1.25e-02, grad_scale: 2.0
+2023-03-01 07:36:27,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.448e+02 1.251e+03 1.545e+03 2.094e+03 4.765e+03, threshold=3.090e+03, percent-clipped=1.0
+2023-03-01 07:37:00,725 INFO [train.py:968] (0/2) Epoch 2, batch 30750, giga_loss[loss=0.3514, simple_loss=0.4098, pruned_loss=0.1465, over 28730.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3866, pruned_loss=0.1356, over 5663294.06 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.4262, pruned_loss=0.1777, over 5686565.89 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3863, pruned_loss=0.1337, over 5653933.92 frames. ], batch size: 284, lr: 1.25e-02, grad_scale: 2.0
+2023-03-01 07:37:34,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3679, 2.2245, 1.3685, 1.2633], device='cuda:0'), covar=tensor([0.0802, 0.0559, 0.0827, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0467, 0.0338, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 07:37:49,161 INFO [train.py:968] (0/2) Epoch 2, batch 30800, giga_loss[loss=0.3292, simple_loss=0.386, pruned_loss=0.1362, over 28024.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3862, pruned_loss=0.1347, over 5670208.60 frames. ], libri_tot_loss[loss=0.39, simple_loss=0.4256, pruned_loss=0.1773, over 5690335.22 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3858, pruned_loss=0.1328, over 5659060.12 frames. ], batch size: 412, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:38:09,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.817e+02 1.554e+03 2.078e+03 2.921e+03 9.378e+03, threshold=4.156e+03, percent-clipped=23.0
+2023-03-01 07:38:42,101 INFO [train.py:968] (0/2) Epoch 2, batch 30850, giga_loss[loss=0.3131, simple_loss=0.3816, pruned_loss=0.1222, over 28276.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3831, pruned_loss=0.1323, over 5667691.89 frames. ], libri_tot_loss[loss=0.389, simple_loss=0.4245, pruned_loss=0.1767, over 5694820.89 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3829, pruned_loss=0.1304, over 5654208.53 frames. ], batch size: 368, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:39:32,265 INFO [train.py:968] (0/2) Epoch 2, batch 30900, giga_loss[loss=0.2747, simple_loss=0.3494, pruned_loss=0.09999, over 28666.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3794, pruned_loss=0.1293, over 5664990.40 frames. ], libri_tot_loss[loss=0.3879, simple_loss=0.4235, pruned_loss=0.1761, over 5691037.48 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3791, pruned_loss=0.1272, over 5656951.65 frames. ], batch size: 307, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:39:50,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.966e+02 1.197e+03 1.560e+03 2.057e+03 6.157e+03, threshold=3.119e+03, percent-clipped=2.0
+2023-03-01 07:40:23,772 INFO [train.py:968] (0/2) Epoch 2, batch 30950, giga_loss[loss=0.3212, simple_loss=0.3649, pruned_loss=0.1387, over 26604.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3756, pruned_loss=0.1267, over 5675154.08 frames. ], libri_tot_loss[loss=0.3873, simple_loss=0.423, pruned_loss=0.1758, over 5695521.07 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3747, pruned_loss=0.1242, over 5664233.23 frames. ], batch size: 555, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:40:32,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7085, 2.1429, 1.6737, 2.0019], device='cuda:0'), covar=tensor([0.0516, 0.0598, 0.0784, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0501, 0.0540, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 07:40:42,374 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-01 07:41:04,106 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=76662.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 07:41:09,835 INFO [train.py:968] (0/2) Epoch 2, batch 31000, giga_loss[loss=0.36, simple_loss=0.4041, pruned_loss=0.158, over 28534.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3749, pruned_loss=0.1273, over 5674443.88 frames. ], libri_tot_loss[loss=0.3862, simple_loss=0.422, pruned_loss=0.1752, over 5694531.58 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3732, pruned_loss=0.1237, over 5665780.30 frames. ], batch size: 336, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:41:29,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.349e+02 1.391e+03 1.992e+03 2.739e+03 4.322e+03, threshold=3.985e+03, percent-clipped=12.0
+2023-03-01 07:41:29,400 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=76687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:42:02,476 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=76716.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 07:42:06,155 INFO [train.py:968] (0/2) Epoch 2, batch 31050, giga_loss[loss=0.2912, simple_loss=0.3595, pruned_loss=0.1114, over 29007.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3751, pruned_loss=0.128, over 5652135.21 frames. ], libri_tot_loss[loss=0.3863, simple_loss=0.422, pruned_loss=0.1753, over 5685433.55 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3733, pruned_loss=0.1248, over 5652348.36 frames. ], batch size: 136, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:42:21,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1463, 4.6257, 4.7959, 2.1562], device='cuda:0'), covar=tensor([0.0380, 0.0320, 0.0685, 0.1802], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0562, 0.0795, 0.0556], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 07:42:59,999 INFO [train.py:968] (0/2) Epoch 2, batch 31100, giga_loss[loss=0.3134, simple_loss=0.3855, pruned_loss=0.1206, over 28696.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3775, pruned_loss=0.1295, over 5648039.39 frames. ], libri_tot_loss[loss=0.3861, simple_loss=0.4217, pruned_loss=0.1752, over 5686550.45 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3756, pruned_loss=0.1262, over 5646701.77 frames. ], batch size: 243, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:43:04,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7834, 2.1565, 1.8591, 1.8424], device='cuda:0'), covar=tensor([0.1467, 0.1413, 0.1104, 0.0749], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0795, 0.0705, 0.0602], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 07:43:21,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.217e+02 1.490e+03 2.073e+03 2.938e+03 7.333e+03, threshold=4.146e+03, percent-clipped=11.0
+2023-03-01 07:43:58,930 INFO [train.py:968] (0/2) Epoch 2, batch 31150, giga_loss[loss=0.2903, simple_loss=0.3682, pruned_loss=0.1062, over 28327.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3788, pruned_loss=0.129, over 5649598.11 frames. ], libri_tot_loss[loss=0.3849, simple_loss=0.4206, pruned_loss=0.1746, over 5692370.65 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3773, pruned_loss=0.1259, over 5642624.89 frames. ], batch size: 368, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:45:03,954 INFO [train.py:968] (0/2) Epoch 2, batch 31200, giga_loss[loss=0.3763, simple_loss=0.4082, pruned_loss=0.1722, over 26880.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3787, pruned_loss=0.129, over 5637712.59 frames. ], libri_tot_loss[loss=0.3844, simple_loss=0.4203, pruned_loss=0.1743, over 5695567.11 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3772, pruned_loss=0.1261, over 5628484.78 frames. ], batch size: 555, lr: 1.25e-02, grad_scale: 8.0
+2023-03-01 07:45:22,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1977, 1.2991, 1.1742, 1.1802], device='cuda:0'), covar=tensor([0.1887, 0.1759, 0.1533, 0.1786], device='cuda:0'), in_proj_covar=tensor([0.0973, 0.0789, 0.0878, 0.0938], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 07:45:26,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1443, 2.0390, 1.5670, 1.5781], device='cuda:0'), covar=tensor([0.0685, 0.0695, 0.0859, 0.1092], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0495, 0.0537, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 07:45:33,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.184e+02 1.540e+03 2.014e+03 2.889e+03 5.625e+03, threshold=4.029e+03, percent-clipped=4.0
+2023-03-01 07:46:19,549 INFO [train.py:968] (0/2) Epoch 2, batch 31250, giga_loss[loss=0.2926, simple_loss=0.3609, pruned_loss=0.1122, over 29038.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3762, pruned_loss=0.1273, over 5637567.99 frames. ], libri_tot_loss[loss=0.3834, simple_loss=0.4193, pruned_loss=0.1737, over 5689533.16 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3752, pruned_loss=0.1247, over 5634632.72 frames. ], batch size: 186, lr: 1.25e-02, grad_scale: 8.0
+2023-03-01 07:47:26,147 INFO [train.py:968] (0/2) Epoch 2, batch 31300, libri_loss[loss=0.3664, simple_loss=0.4113, pruned_loss=0.1607, over 27816.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3742, pruned_loss=0.1245, over 5633292.20 frames. ], libri_tot_loss[loss=0.3832, simple_loss=0.4192, pruned_loss=0.1736, over 5689980.41 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3731, pruned_loss=0.1221, over 5630089.88 frames. ], batch size: 116, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:47:52,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.467e+02 1.175e+03 1.564e+03 1.910e+03 5.943e+03, threshold=3.128e+03, percent-clipped=3.0
+2023-03-01 07:48:30,787 INFO [train.py:968] (0/2) Epoch 2, batch 31350, giga_loss[loss=0.3065, simple_loss=0.374, pruned_loss=0.1195, over 28957.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3717, pruned_loss=0.1214, over 5639074.01 frames. ], libri_tot_loss[loss=0.3831, simple_loss=0.4192, pruned_loss=0.1735, over 5692223.96 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3703, pruned_loss=0.1191, over 5633902.86 frames. ], batch size: 106, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:48:45,522 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-01 07:48:50,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77037.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 07:48:54,348 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-01 07:49:18,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:49:21,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8629, 3.6292, 1.6820, 1.7896], device='cuda:0'), covar=tensor([0.0857, 0.0631, 0.0935, 0.1310], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0453, 0.0332, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 07:49:26,816 INFO [train.py:968] (0/2) Epoch 2, batch 31400, giga_loss[loss=0.2887, simple_loss=0.3569, pruned_loss=0.1103, over 28981.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3703, pruned_loss=0.1229, over 5661453.94 frames. ], libri_tot_loss[loss=0.3818, simple_loss=0.4178, pruned_loss=0.1728, over 5701494.20 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3682, pruned_loss=0.1193, over 5646916.27 frames. ], batch size: 155, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:49:34,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1745, 1.8867, 1.1147, 1.1899], device='cuda:0'), covar=tensor([0.0961, 0.0667, 0.0969, 0.1441], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0451, 0.0332, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 07:49:51,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.234e+02 1.466e+03 2.040e+03 2.862e+03 5.785e+03, threshold=4.081e+03, percent-clipped=20.0
+2023-03-01 07:49:54,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77091.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 07:50:29,857 INFO [train.py:968] (0/2) Epoch 2, batch 31450, giga_loss[loss=0.3395, simple_loss=0.3915, pruned_loss=0.1437, over 28089.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3709, pruned_loss=0.1237, over 5664476.77 frames. ], libri_tot_loss[loss=0.3816, simple_loss=0.4177, pruned_loss=0.1727, over 5695403.44 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1197, over 5657012.93 frames. ], batch size: 412, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:51:18,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4601, 2.0826, 1.9064, 2.0671], device='cuda:0'), covar=tensor([0.0730, 0.1669, 0.1145, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0791, 0.0629, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 07:51:30,590 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 07:51:32,570 INFO [train.py:968] (0/2) Epoch 2, batch 31500, giga_loss[loss=0.2951, simple_loss=0.3661, pruned_loss=0.1121, over 29001.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3707, pruned_loss=0.1228, over 5669735.40 frames. ], libri_tot_loss[loss=0.3816, simple_loss=0.4178, pruned_loss=0.1728, over 5696464.42 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3682, pruned_loss=0.1193, over 5662752.85 frames. ], batch size: 199, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:51:36,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2913, 1.2706, 1.2262, 1.6420], device='cuda:0'), covar=tensor([0.2085, 0.2014, 0.1692, 0.2004], device='cuda:0'), in_proj_covar=tensor([0.0984, 0.0795, 0.0891, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 07:51:47,142 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77180.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 07:51:49,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77183.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 07:51:55,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.757e+02 1.403e+03 1.854e+03 2.590e+03 6.059e+03, threshold=3.709e+03, percent-clipped=1.0
+2023-03-01 07:52:07,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0903, 1.2427, 1.1123, 0.9289], device='cuda:0'), covar=tensor([0.1728, 0.1496, 0.1356, 0.1583], device='cuda:0'), in_proj_covar=tensor([0.0983, 0.0791, 0.0886, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 07:52:17,294 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:52:21,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:52:27,085 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 07:52:33,733 INFO [train.py:968] (0/2) Epoch 2, batch 31550, giga_loss[loss=0.3332, simple_loss=0.3716, pruned_loss=0.1474, over 24252.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3731, pruned_loss=0.1244, over 5661495.90 frames. ], libri_tot_loss[loss=0.3809, simple_loss=0.4169, pruned_loss=0.1724, over 5700403.95 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3704, pruned_loss=0.1204, over 5650986.30 frames. ], batch size: 705, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:52:47,640 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 07:52:50,580 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=77232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:52:52,792 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77234.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 07:52:56,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:52:56,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77237.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 07:53:37,986 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77266.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 07:53:42,175 INFO [train.py:968] (0/2) Epoch 2, batch 31600, giga_loss[loss=0.2648, simple_loss=0.3377, pruned_loss=0.09599, over 28514.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3724, pruned_loss=0.1237, over 5666560.71 frames. ], libri_tot_loss[loss=0.381, simple_loss=0.4168, pruned_loss=0.1726, over 5703838.44 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3695, pruned_loss=0.1193, over 5654360.45 frames. ], batch size: 336, lr: 1.25e-02, grad_scale: 8.0
+2023-03-01 07:54:04,211 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.932e+02 1.444e+03 1.856e+03 2.657e+03 6.961e+03, threshold=3.712e+03, percent-clipped=10.0
+2023-03-01 07:54:48,120 INFO [train.py:968] (0/2) Epoch 2, batch 31650, giga_loss[loss=0.3223, simple_loss=0.383, pruned_loss=0.1308, over 28426.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3708, pruned_loss=0.1231, over 5670780.41 frames. ], libri_tot_loss[loss=0.3804, simple_loss=0.4162, pruned_loss=0.1723, over 5695998.35 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3675, pruned_loss=0.1182, over 5667646.76 frames. ], batch size: 336, lr: 1.25e-02, grad_scale: 8.0
+2023-03-01 07:55:56,964 INFO [train.py:968] (0/2) Epoch 2, batch 31700, giga_loss[loss=0.3104, simple_loss=0.3784, pruned_loss=0.1212, over 28952.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3728, pruned_loss=0.1248, over 5662237.45 frames. ], libri_tot_loss[loss=0.38, simple_loss=0.4157, pruned_loss=0.1721, over 5691832.38 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3695, pruned_loss=0.1198, over 5663741.73 frames. ], batch size: 213, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:56:21,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.616e+02 1.398e+03 1.709e+03 2.705e+03 8.705e+03, threshold=3.418e+03, percent-clipped=14.0
+2023-03-01 07:57:02,122 INFO [train.py:968] (0/2) Epoch 2, batch 31750, giga_loss[loss=0.2762, simple_loss=0.3629, pruned_loss=0.09473, over 28392.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3777, pruned_loss=0.1256, over 5670654.93 frames. ], libri_tot_loss[loss=0.379, simple_loss=0.4148, pruned_loss=0.1716, over 5698187.12 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3746, pruned_loss=0.1208, over 5665382.11 frames. ], batch size: 369, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 07:58:08,436 INFO [train.py:968] (0/2) Epoch 2, batch 31800, giga_loss[loss=0.2857, simple_loss=0.3669, pruned_loss=0.1023, over 28052.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3775, pruned_loss=0.1237, over 5661726.67 frames. ], libri_tot_loss[loss=0.3787, simple_loss=0.4146, pruned_loss=0.1714, over 5701803.08 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3745, pruned_loss=0.1189, over 5653665.90 frames. ], batch size: 412, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 07:58:33,300 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=77488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:58:34,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.850e+02 1.373e+03 1.704e+03 2.321e+03 5.052e+03, threshold=3.409e+03, percent-clipped=7.0
+2023-03-01 07:58:41,917 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4466, 2.4920, 1.3541, 1.3119], device='cuda:0'), covar=tensor([0.0827, 0.0496, 0.0879, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0449, 0.0335, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 07:59:12,677 INFO [train.py:968] (0/2) Epoch 2, batch 31850, giga_loss[loss=0.3285, simple_loss=0.3825, pruned_loss=0.1373, over 26888.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3768, pruned_loss=0.121, over 5664587.49 frames. ], libri_tot_loss[loss=0.3778, simple_loss=0.4139, pruned_loss=0.1709, over 5703752.68 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3747, pruned_loss=0.1172, over 5656158.40 frames. ], batch size: 555, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 07:59:46,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7659, 2.8814, 1.8454, 0.8297], device='cuda:0'), covar=tensor([0.2226, 0.0828, 0.1543, 0.2391], device='cuda:0'), in_proj_covar=tensor([0.1195, 0.1161, 0.1221, 0.1023], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 07:59:46,886 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 08:00:14,274 INFO [train.py:968] (0/2) Epoch 2, batch 31900, giga_loss[loss=0.3292, simple_loss=0.3975, pruned_loss=0.1304, over 28967.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3766, pruned_loss=0.1207, over 5669525.78 frames. ], libri_tot_loss[loss=0.3772, simple_loss=0.4133, pruned_loss=0.1705, over 5696788.82 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3748, pruned_loss=0.1172, over 5668395.45 frames. ], batch size: 136, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:00:38,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.631e+02 1.320e+03 1.748e+03 2.196e+03 5.321e+03, threshold=3.496e+03, percent-clipped=3.0
+2023-03-01 08:00:57,837 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=77604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:01:01,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:01:08,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1691, 1.2026, 1.2045, 1.0923], device='cuda:0'), covar=tensor([0.0702, 0.0982, 0.1175, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0774, 0.0602, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 08:01:18,227 INFO [train.py:968] (0/2) Epoch 2, batch 31950, giga_loss[loss=0.2623, simple_loss=0.3382, pruned_loss=0.09317, over 29121.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3754, pruned_loss=0.1213, over 5685558.12 frames. ], libri_tot_loss[loss=0.3761, simple_loss=0.4124, pruned_loss=0.1698, over 5703817.73 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3737, pruned_loss=0.1175, over 5677768.15 frames. ], batch size: 200, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:01:48,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4240, 1.4547, 1.0125, 1.3294], device='cuda:0'), covar=tensor([0.0762, 0.0624, 0.1145, 0.0949], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0491, 0.0544, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 08:02:03,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6155, 4.1325, 4.3018, 1.8344], device='cuda:0'), covar=tensor([0.0392, 0.0332, 0.0749, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0556, 0.0788, 0.0555], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 08:02:28,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=77666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:02:31,174 INFO [train.py:968] (0/2) Epoch 2, batch 32000, libri_loss[loss=0.304, simple_loss=0.354, pruned_loss=0.127, over 29573.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3769, pruned_loss=0.1239, over 5670036.88 frames. ], libri_tot_loss[loss=0.3756, simple_loss=0.412, pruned_loss=0.1696, over 5698020.98 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3746, pruned_loss=0.1194, over 5668123.45 frames. ], batch size: 78, lr: 1.24e-02, grad_scale: 8.0
+2023-03-01 08:02:33,244 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=77670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:03:00,187 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.280e+02 1.391e+03 1.956e+03 2.390e+03 4.410e+03, threshold=3.912e+03, percent-clipped=6.0
+2023-03-01 08:03:00,917 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8414, 1.4087, 3.8337, 3.1849], device='cuda:0'), covar=tensor([0.1465, 0.1811, 0.0338, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0487, 0.0642, 0.0502], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 08:03:46,074 INFO [train.py:968] (0/2) Epoch 2, batch 32050, giga_loss[loss=0.276, simple_loss=0.3466, pruned_loss=0.1027, over 28814.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3765, pruned_loss=0.1241, over 5669453.65 frames. ], libri_tot_loss[loss=0.3756, simple_loss=0.4119, pruned_loss=0.1697, over 5689196.63 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3741, pruned_loss=0.1196, over 5675099.86 frames. ], batch size: 227, lr: 1.24e-02, grad_scale: 8.0
+2023-03-01 08:04:33,912 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:04:37,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:04:59,032 INFO [train.py:968] (0/2) Epoch 2, batch 32100, giga_loss[loss=0.2744, simple_loss=0.3428, pruned_loss=0.103, over 27735.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3721, pruned_loss=0.1215, over 5661753.98 frames. ], libri_tot_loss[loss=0.3759, simple_loss=0.4119, pruned_loss=0.1699, over 5685280.98 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3694, pruned_loss=0.1166, over 5669547.02 frames. ], batch size: 474, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:05:14,600 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:05:25,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.233e+02 1.341e+03 1.778e+03 2.738e+03 6.006e+03, threshold=3.556e+03, percent-clipped=7.0
+2023-03-01 08:06:08,147 INFO [train.py:968] (0/2) Epoch 2, batch 32150, giga_loss[loss=0.3158, simple_loss=0.3727, pruned_loss=0.1294, over 26791.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.37, pruned_loss=0.1203, over 5670737.43 frames. ], libri_tot_loss[loss=0.3753, simple_loss=0.4112, pruned_loss=0.1697, over 5690713.30 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3675, pruned_loss=0.1155, over 5671762.86 frames. ], batch size: 555, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:07:07,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:07:15,717 INFO [train.py:968] (0/2) Epoch 2, batch 32200, giga_loss[loss=0.2976, simple_loss=0.3682, pruned_loss=0.1135, over 28490.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.371, pruned_loss=0.1212, over 5677944.91 frames. ], libri_tot_loss[loss=0.3751, simple_loss=0.411, pruned_loss=0.1696, over 5691007.54 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3689, pruned_loss=0.1171, over 5678377.34 frames. ], batch size: 336, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:07:45,380 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.096e+02 1.461e+03 1.951e+03 2.484e+03 4.861e+03, threshold=3.901e+03, percent-clipped=9.0
+2023-03-01 08:08:19,695 INFO [train.py:968] (0/2) Epoch 2, batch 32250, giga_loss[loss=0.3524, simple_loss=0.4068, pruned_loss=0.149, over 27624.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3756, pruned_loss=0.1234, over 5684052.76 frames. ], libri_tot_loss[loss=0.3747, simple_loss=0.4108, pruned_loss=0.1694, over 5695226.48 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3734, pruned_loss=0.1195, over 5680491.14 frames. ], batch size: 472, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:08:28,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3446, 1.9822, 1.3921, 0.6827], device='cuda:0'), covar=tensor([0.1762, 0.0896, 0.1429, 0.1847], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.1177, 0.1224, 0.1020], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 08:09:25,328 INFO [train.py:968] (0/2) Epoch 2, batch 32300, giga_loss[loss=0.3643, simple_loss=0.4023, pruned_loss=0.1631, over 26845.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3746, pruned_loss=0.1244, over 5685823.87 frames. ], libri_tot_loss[loss=0.3746, simple_loss=0.4105, pruned_loss=0.1694, over 5697436.43 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3721, pruned_loss=0.1201, over 5680668.19 frames. ], batch size: 555, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:09:37,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:09:53,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.540e+02 1.368e+03 1.747e+03 2.757e+03 7.342e+03, threshold=3.494e+03, percent-clipped=6.0
+2023-03-01 08:10:05,163 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-78000.pt
+2023-03-01 08:10:13,914 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:10:16,492 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=78009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:10:16,573 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:10:27,322 INFO [train.py:968] (0/2) Epoch 2, batch 32350, libri_loss[loss=0.3279, simple_loss=0.3711, pruned_loss=0.1423, over 29565.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3751, pruned_loss=0.1257, over 5675244.63 frames. ], libri_tot_loss[loss=0.3743, simple_loss=0.4101, pruned_loss=0.1692, over 5691451.83 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3726, pruned_loss=0.1214, over 5676307.58 frames. ], batch size: 75, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:10:52,715 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:10:56,392 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:11:01,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:11:11,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6761, 1.3497, 3.3773, 2.8377], device='cuda:0'), covar=tensor([0.1439, 0.1660, 0.0432, 0.0557], device='cuda:0'), in_proj_covar=tensor([0.0509, 0.0496, 0.0657, 0.0502], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 08:11:14,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5771, 3.0693, 1.6393, 1.3904], device='cuda:0'), covar=tensor([0.0672, 0.0261, 0.0405, 0.0656], device='cuda:0'), in_proj_covar=tensor([0.1094, 0.0754, 0.0861, 0.0916], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 08:11:33,285 INFO [train.py:968] (0/2) Epoch 2, batch 32400, giga_loss[loss=0.3485, simple_loss=0.4005, pruned_loss=0.1482, over 27676.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3756, pruned_loss=0.1259, over 5672079.09 frames. ], libri_tot_loss[loss=0.3746, simple_loss=0.4103, pruned_loss=0.1695, over 5684196.22 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.373, pruned_loss=0.1217, over 5680010.10 frames. ], batch size: 472, lr: 1.24e-02, grad_scale: 8.0
+2023-03-01 08:11:34,420 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=78070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:12:08,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7625, 2.6309, 1.8399, 0.7790], device='cuda:0'), covar=tensor([0.2339, 0.1190, 0.1465, 0.2443], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.1174, 0.1218, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 08:12:08,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.613e+02 1.488e+03 2.013e+03 2.697e+03 6.905e+03, threshold=4.025e+03, percent-clipped=14.0
+2023-03-01 08:12:47,653 INFO [train.py:968] (0/2) Epoch 2, batch 32450, giga_loss[loss=0.3057, simple_loss=0.3824, pruned_loss=0.1145, over 28599.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3785, pruned_loss=0.1269, over 5671886.52 frames. ], libri_tot_loss[loss=0.3751, simple_loss=0.4105, pruned_loss=0.1698, over 5686085.11 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3755, pruned_loss=0.1222, over 5675936.13 frames. ], batch size: 307, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:12:51,823 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:12:56,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:13:36,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5935, 1.9479, 1.7083, 1.7317], device='cuda:0'), covar=tensor([0.1078, 0.1229, 0.0914, 0.0659], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0795, 0.0704, 0.0607], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 08:13:44,224 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:14:09,299 INFO [train.py:968] (0/2) Epoch 2, batch 32500, giga_loss[loss=0.3019, simple_loss=0.3769, pruned_loss=0.1134, over 28713.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3787, pruned_loss=0.1269, over 5655766.01 frames. ], libri_tot_loss[loss=0.3752, simple_loss=0.4107, pruned_loss=0.1699, over 5678698.34 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3758, pruned_loss=0.1226, over 5665291.18 frames. ], batch size: 262, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:14:33,509 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:14:37,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:14:38,283 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:14:41,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.944e+02 1.257e+03 1.723e+03 2.399e+03 6.764e+03, threshold=3.446e+03, percent-clipped=5.0
+2023-03-01 08:14:42,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:14:47,863 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 08:15:16,156 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:15:18,192 INFO [train.py:968] (0/2) Epoch 2, batch 32550, giga_loss[loss=0.2945, simple_loss=0.3537, pruned_loss=0.1177, over 28877.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3735, pruned_loss=0.1244, over 5662171.70 frames. ], libri_tot_loss[loss=0.3745, simple_loss=0.4101, pruned_loss=0.1695, over 5679634.07 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.371, pruned_loss=0.1203, over 5668440.28 frames. ], batch size: 284, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:15:19,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:15:40,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9210, 1.3620, 3.3858, 2.8962], device='cuda:0'), covar=tensor([0.1301, 0.1631, 0.0337, 0.1041], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0491, 0.0641, 0.0498], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 08:16:20,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8628, 1.4211, 4.1199, 3.2900], device='cuda:0'), covar=tensor([0.1456, 0.1750, 0.0277, 0.0765], device='cuda:0'), in_proj_covar=tensor([0.0507, 0.0493, 0.0647, 0.0497], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 08:16:26,963 INFO [train.py:968] (0/2) Epoch 2, batch 32600, giga_loss[loss=0.2707, simple_loss=0.3407, pruned_loss=0.1003, over 28903.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3658, pruned_loss=0.1202, over 5672434.97 frames. ], libri_tot_loss[loss=0.3744, simple_loss=0.4099, pruned_loss=0.1694, over 5681634.38 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3636, pruned_loss=0.1166, over 5675481.98 frames. ], batch size: 227, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:17:00,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.663e+02 1.516e+03 1.851e+03 2.392e+03 7.367e+03, threshold=3.703e+03, percent-clipped=12.0
+2023-03-01 08:17:34,595 INFO [train.py:968] (0/2) Epoch 2, batch 32650, giga_loss[loss=0.2902, simple_loss=0.3557, pruned_loss=0.1123, over 28963.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3642, pruned_loss=0.1196, over 5666664.86 frames. ], libri_tot_loss[loss=0.3742, simple_loss=0.4097, pruned_loss=0.1693, over 5683970.13 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3621, pruned_loss=0.1162, over 5666737.50 frames. ], batch size: 213, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:18:33,256 INFO [train.py:968] (0/2) Epoch 2, batch 32700, giga_loss[loss=0.3292, simple_loss=0.3928, pruned_loss=0.1328, over 28891.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3664, pruned_loss=0.1214, over 5663812.92 frames. ], libri_tot_loss[loss=0.3737, simple_loss=0.4091, pruned_loss=0.1691, over 5678876.56 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3642, pruned_loss=0.1178, over 5668836.36 frames. ], batch size: 164, lr: 1.24e-02, grad_scale: 2.0
+2023-03-01 08:18:53,829 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:19:01,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.457e+02 1.597e+03 2.007e+03 2.813e+03 6.220e+03, threshold=4.014e+03, percent-clipped=9.0
+2023-03-01 08:19:31,170 INFO [train.py:968] (0/2) Epoch 2, batch 32750, giga_loss[loss=0.287, simple_loss=0.3593, pruned_loss=0.1073, over 29076.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3653, pruned_loss=0.1204, over 5661555.71 frames. ], libri_tot_loss[loss=0.3733, simple_loss=0.4087, pruned_loss=0.1689, over 5674976.33 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3626, pruned_loss=0.1162, over 5669352.17 frames. ], batch size: 200, lr: 1.24e-02, grad_scale: 2.0
+2023-03-01 08:20:05,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:20:34,758 INFO [train.py:968] (0/2) Epoch 2, batch 32800, giga_loss[loss=0.3121, simple_loss=0.3692, pruned_loss=0.1275, over 28460.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3631, pruned_loss=0.1182, over 5650582.64 frames. ], libri_tot_loss[loss=0.3729, simple_loss=0.4083, pruned_loss=0.1688, over 5670992.69 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3603, pruned_loss=0.1138, over 5660322.03 frames. ], batch size: 336, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:21:03,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.586e+02 1.380e+03 1.858e+03 2.685e+03 5.167e+03, threshold=3.715e+03, percent-clipped=9.0
+2023-03-01 08:21:04,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2441, 2.8397, 2.9643, 1.4489], device='cuda:0'), covar=tensor([0.0695, 0.0571, 0.1092, 0.1905], device='cuda:0'), in_proj_covar=tensor([0.0753, 0.0547, 0.0780, 0.0537], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 08:21:40,509 INFO [train.py:968] (0/2) Epoch 2, batch 32850, giga_loss[loss=0.3255, simple_loss=0.3796, pruned_loss=0.1357, over 28111.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3614, pruned_loss=0.1174, over 5656482.69 frames. ], libri_tot_loss[loss=0.3726, simple_loss=0.408, pruned_loss=0.1687, over 5673211.77 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3591, pruned_loss=0.1136, over 5662207.39 frames. ], batch size: 412, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:21:54,041 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:21:57,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:22:38,761 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:22:51,693 INFO [train.py:968] (0/2) Epoch 2, batch 32900, giga_loss[loss=0.3278, simple_loss=0.3922, pruned_loss=0.1317, over 28813.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.362, pruned_loss=0.1173, over 5658218.07 frames. ], libri_tot_loss[loss=0.3729, simple_loss=0.4081, pruned_loss=0.1689, over 5667844.46 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.359, pruned_loss=0.113, over 5667539.73 frames. ], batch size: 243, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:23:13,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:23:19,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:23:19,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.137e+02 1.409e+03 1.877e+03 2.482e+03 5.281e+03, threshold=3.755e+03, percent-clipped=7.0
+2023-03-01 08:23:35,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3059, 3.8448, 3.9562, 1.7382], device='cuda:0'), covar=tensor([0.0410, 0.0347, 0.0724, 0.1768], device='cuda:0'), in_proj_covar=tensor([0.0764, 0.0553, 0.0781, 0.0538], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 08:23:57,093 INFO [train.py:968] (0/2) Epoch 2, batch 32950, giga_loss[loss=0.31, simple_loss=0.3628, pruned_loss=0.1286, over 26982.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3641, pruned_loss=0.1185, over 5671009.15 frames. ], libri_tot_loss[loss=0.3732, simple_loss=0.4084, pruned_loss=0.169, over 5673851.93 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3605, pruned_loss=0.1138, over 5673180.64 frames. ], batch size: 555, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:23:58,409 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:24:05,318 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=78625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:25:01,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5932, 1.3778, 1.4097, 1.3873], device='cuda:0'), covar=tensor([0.0673, 0.1176, 0.1274, 0.1012], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0792, 0.0615, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 08:25:05,329 INFO [train.py:968] (0/2) Epoch 2, batch 33000, giga_loss[loss=0.3136, simple_loss=0.3749, pruned_loss=0.1262, over 28928.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3656, pruned_loss=0.1201, over 5681760.75 frames. ], libri_tot_loss[loss=0.3727, simple_loss=0.408, pruned_loss=0.1687, over 5679341.90 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.362, pruned_loss=0.1153, over 5678386.72 frames. ], batch size: 227, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:25:05,334 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 08:25:14,381 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.2412, simple_loss=0.3357, pruned_loss=0.07337, over 944034.00 frames. 
+2023-03-01 08:25:14,382 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19537MB
+2023-03-01 08:25:22,520 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-01 08:25:41,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.910e+02 1.230e+03 1.707e+03 2.503e+03 7.822e+03, threshold=3.414e+03, percent-clipped=14.0
+2023-03-01 08:26:18,582 INFO [train.py:968] (0/2) Epoch 2, batch 33050, giga_loss[loss=0.2652, simple_loss=0.3134, pruned_loss=0.1084, over 24515.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3639, pruned_loss=0.1189, over 5670019.34 frames. ], libri_tot_loss[loss=0.3731, simple_loss=0.4082, pruned_loss=0.169, over 5678513.46 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3603, pruned_loss=0.1144, over 5667814.24 frames. ], batch size: 705, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:26:29,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8351, 1.3917, 3.9060, 3.1974], device='cuda:0'), covar=tensor([0.1351, 0.1787, 0.0286, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0496, 0.0641, 0.0500], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 08:26:58,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3506, 1.3189, 1.2740, 1.5704], device='cuda:0'), covar=tensor([0.1965, 0.1683, 0.1405, 0.1776], device='cuda:0'), in_proj_covar=tensor([0.0971, 0.0796, 0.0886, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 08:27:10,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8854, 2.5966, 2.1341, 1.8496], device='cuda:0'), covar=tensor([0.1565, 0.1390, 0.1129, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0788, 0.0710, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 08:27:20,179 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-01 08:27:20,444 INFO [train.py:968] (0/2) Epoch 2, batch 33100, giga_loss[loss=0.2769, simple_loss=0.3535, pruned_loss=0.1002, over 28022.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3664, pruned_loss=0.1188, over 5665298.20 frames. ], libri_tot_loss[loss=0.3736, simple_loss=0.4087, pruned_loss=0.1693, over 5681325.00 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3625, pruned_loss=0.1142, over 5660692.02 frames. ], batch size: 412, lr: 1.23e-02, grad_scale: 2.0
+2023-03-01 08:27:50,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.592e+02 1.357e+03 1.914e+03 2.653e+03 1.100e+04, threshold=3.827e+03, percent-clipped=15.0
+2023-03-01 08:28:23,249 INFO [train.py:968] (0/2) Epoch 2, batch 33150, giga_loss[loss=0.3178, simple_loss=0.3882, pruned_loss=0.1237, over 28974.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3697, pruned_loss=0.1201, over 5664979.85 frames. ], libri_tot_loss[loss=0.3735, simple_loss=0.4086, pruned_loss=0.1693, over 5683709.57 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.366, pruned_loss=0.1155, over 5658810.19 frames. ], batch size: 155, lr: 1.23e-02, grad_scale: 2.0
+2023-03-01 08:29:06,040 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=78853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:29:23,606 INFO [train.py:968] (0/2) Epoch 2, batch 33200, giga_loss[loss=0.3236, simple_loss=0.3847, pruned_loss=0.1312, over 28900.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.371, pruned_loss=0.1211, over 5664275.35 frames. ], libri_tot_loss[loss=0.3728, simple_loss=0.4078, pruned_loss=0.1689, over 5680146.14 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3675, pruned_loss=0.1162, over 5662064.07 frames. ], batch size: 164, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:30:00,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.310e+02 1.433e+03 1.847e+03 2.865e+03 6.672e+03, threshold=3.694e+03, percent-clipped=11.0
+2023-03-01 08:30:35,244 INFO [train.py:968] (0/2) Epoch 2, batch 33250, giga_loss[loss=0.3394, simple_loss=0.3986, pruned_loss=0.1401, over 28818.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3709, pruned_loss=0.1212, over 5664370.69 frames. ], libri_tot_loss[loss=0.3727, simple_loss=0.4077, pruned_loss=0.1688, over 5682262.46 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3678, pruned_loss=0.117, over 5660540.53 frames. ], batch size: 243, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:31:10,166 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=78949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:31:34,922 INFO [train.py:968] (0/2) Epoch 2, batch 33300, giga_loss[loss=0.2346, simple_loss=0.2948, pruned_loss=0.08718, over 24760.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3675, pruned_loss=0.1187, over 5672927.37 frames. ], libri_tot_loss[loss=0.3723, simple_loss=0.4074, pruned_loss=0.1686, over 5685734.47 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3645, pruned_loss=0.1144, over 5666364.49 frames. ], batch size: 705, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:31:44,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7013, 2.7179, 1.5727, 1.5166], device='cuda:0'), covar=tensor([0.0890, 0.0419, 0.0967, 0.1421], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0441, 0.0335, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 08:32:01,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.301e+02 1.383e+03 1.812e+03 2.342e+03 4.410e+03, threshold=3.624e+03, percent-clipped=4.0
+2023-03-01 08:32:10,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:32:32,923 INFO [train.py:968] (0/2) Epoch 2, batch 33350, libri_loss[loss=0.3884, simple_loss=0.4204, pruned_loss=0.1782, over 25886.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3669, pruned_loss=0.1187, over 5681232.52 frames. ], libri_tot_loss[loss=0.3716, simple_loss=0.4067, pruned_loss=0.1682, over 5690125.75 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3636, pruned_loss=0.1137, over 5671976.53 frames. ], batch size: 136, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:32:54,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4412, 2.3573, 1.4090, 1.3251], device='cuda:0'), covar=tensor([0.0872, 0.0494, 0.0929, 0.1326], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0442, 0.0334, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 08:33:01,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7696, 2.7609, 1.6368, 1.4367], device='cuda:0'), covar=tensor([0.0853, 0.0440, 0.0928, 0.1418], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0442, 0.0334, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 08:33:32,798 INFO [train.py:968] (0/2) Epoch 2, batch 33400, giga_loss[loss=0.2809, simple_loss=0.3522, pruned_loss=0.1048, over 29019.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3649, pruned_loss=0.118, over 5686572.38 frames. ], libri_tot_loss[loss=0.3713, simple_loss=0.4065, pruned_loss=0.168, over 5693981.01 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3617, pruned_loss=0.1134, over 5675762.26 frames. ], batch size: 136, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:33:42,961 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-01 08:34:05,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.730e+02 1.529e+03 1.794e+03 2.554e+03 7.566e+03, threshold=3.589e+03, percent-clipped=8.0
+2023-03-01 08:34:38,801 INFO [train.py:968] (0/2) Epoch 2, batch 33450, giga_loss[loss=0.3468, simple_loss=0.4011, pruned_loss=0.1463, over 29047.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3651, pruned_loss=0.118, over 5681548.48 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4057, pruned_loss=0.1674, over 5698310.10 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3626, pruned_loss=0.1139, over 5668720.11 frames. ], batch size: 165, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:35:15,763 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:35:18,200 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:35:47,877 INFO [train.py:968] (0/2) Epoch 2, batch 33500, giga_loss[loss=0.2462, simple_loss=0.3051, pruned_loss=0.0936, over 24602.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3665, pruned_loss=0.1186, over 5678440.46 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4056, pruned_loss=0.1673, over 5697795.69 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3641, pruned_loss=0.1147, over 5668526.75 frames. ], batch size: 705, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:35:54,226 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:36:18,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.792e+02 1.305e+03 1.905e+03 2.558e+03 5.289e+03, threshold=3.811e+03, percent-clipped=7.0
+2023-03-01 08:36:53,526 INFO [train.py:968] (0/2) Epoch 2, batch 33550, giga_loss[loss=0.308, simple_loss=0.3686, pruned_loss=0.1237, over 27596.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1214, over 5673908.90 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4054, pruned_loss=0.1671, over 5697569.45 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3663, pruned_loss=0.1172, over 5666064.14 frames. ], batch size: 472, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:37:09,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:38:05,436 INFO [train.py:968] (0/2) Epoch 2, batch 33600, giga_loss[loss=0.3447, simple_loss=0.4049, pruned_loss=0.1423, over 28332.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3733, pruned_loss=0.1244, over 5657762.67 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4054, pruned_loss=0.1672, over 5696931.70 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3707, pruned_loss=0.1205, over 5651662.20 frames. ], batch size: 368, lr: 1.23e-02, grad_scale: 8.0
+2023-03-01 08:38:32,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.943e+02 1.280e+03 1.602e+03 2.330e+03 1.298e+04, threshold=3.204e+03, percent-clipped=7.0
+2023-03-01 08:39:03,868 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-01 08:39:04,038 INFO [train.py:968] (0/2) Epoch 2, batch 33650, giga_loss[loss=0.3035, simple_loss=0.378, pruned_loss=0.1145, over 28943.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3761, pruned_loss=0.1247, over 5667315.22 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.4053, pruned_loss=0.167, over 5700224.13 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3737, pruned_loss=0.1211, over 5659102.38 frames. ], batch size: 155, lr: 1.23e-02, grad_scale: 8.0
+2023-03-01 08:39:10,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:39:57,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:40:09,335 INFO [train.py:968] (0/2) Epoch 2, batch 33700, giga_loss[loss=0.2658, simple_loss=0.3413, pruned_loss=0.09512, over 29027.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3765, pruned_loss=0.1254, over 5660630.26 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4045, pruned_loss=0.1664, over 5694138.24 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3744, pruned_loss=0.1218, over 5658665.90 frames. ], batch size: 100, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:40:11,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:40:16,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:40:46,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.749e+02 1.331e+03 1.834e+03 2.823e+03 7.035e+03, threshold=3.668e+03, percent-clipped=16.0
+2023-03-01 08:40:58,507 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:41:23,039 INFO [train.py:968] (0/2) Epoch 2, batch 33750, giga_loss[loss=0.3072, simple_loss=0.3657, pruned_loss=0.1243, over 28137.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3737, pruned_loss=0.1234, over 5671346.75 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4044, pruned_loss=0.1663, over 5700641.44 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3713, pruned_loss=0.1195, over 5663218.38 frames. ], batch size: 412, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:41:26,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2029, 1.7746, 1.3386, 0.5754], device='cuda:0'), covar=tensor([0.1467, 0.1011, 0.1741, 0.1493], device='cuda:0'), in_proj_covar=tensor([0.1222, 0.1206, 0.1220, 0.1035], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 08:42:02,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3275, 1.8131, 1.3634, 0.4290], device='cuda:0'), covar=tensor([0.1171, 0.0899, 0.1467, 0.1615], device='cuda:0'), in_proj_covar=tensor([0.1216, 0.1193, 0.1205, 0.1033], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 08:42:10,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2058, 1.5752, 1.1194, 0.4636], device='cuda:0'), covar=tensor([0.0948, 0.0700, 0.1132, 0.1275], device='cuda:0'), in_proj_covar=tensor([0.1211, 0.1189, 0.1205, 0.1030], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 08:42:23,935 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:42:25,274 INFO [train.py:968] (0/2) Epoch 2, batch 33800, giga_loss[loss=0.3494, simple_loss=0.4007, pruned_loss=0.149, over 27994.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3734, pruned_loss=0.1235, over 5673887.29 frames. ], libri_tot_loss[loss=0.3683, simple_loss=0.4044, pruned_loss=0.1662, over 5694028.46 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3705, pruned_loss=0.1191, over 5672699.49 frames. ], batch size: 412, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:42:27,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:43:01,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.975e+02 1.533e+03 2.058e+03 2.760e+03 5.808e+03, threshold=4.117e+03, percent-clipped=9.0
+2023-03-01 08:43:06,835 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:43:29,438 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-01 08:43:31,570 INFO [train.py:968] (0/2) Epoch 2, batch 33850, giga_loss[loss=0.2973, simple_loss=0.3569, pruned_loss=0.1189, over 28866.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3724, pruned_loss=0.1235, over 5671583.59 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4038, pruned_loss=0.1657, over 5695389.89 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3696, pruned_loss=0.119, over 5669022.67 frames. ], batch size: 112, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:44:38,686 INFO [train.py:968] (0/2) Epoch 2, batch 33900, giga_loss[loss=0.3259, simple_loss=0.3832, pruned_loss=0.1343, over 28619.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3703, pruned_loss=0.1233, over 5670574.45 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4038, pruned_loss=0.1657, over 5696069.48 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3672, pruned_loss=0.1185, over 5667479.42 frames. ], batch size: 307, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:45:00,944 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-01 08:45:13,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.461e+02 1.452e+03 1.973e+03 2.690e+03 9.374e+03, threshold=3.946e+03, percent-clipped=6.0
+2023-03-01 08:45:47,391 INFO [train.py:968] (0/2) Epoch 2, batch 33950, giga_loss[loss=0.3248, simple_loss=0.3928, pruned_loss=0.1284, over 28932.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3685, pruned_loss=0.1217, over 5680934.46 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4036, pruned_loss=0.1656, over 5699279.25 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3656, pruned_loss=0.1174, over 5675276.97 frames. ], batch size: 227, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:46:05,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.7495, 0.8204, 0.6745, 0.4661], device='cuda:0'), covar=tensor([0.0298, 0.0281, 0.0256, 0.0359], device='cuda:0'), in_proj_covar=tensor([0.1070, 0.0742, 0.0821, 0.0907], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 08:46:40,027 INFO [train.py:968] (0/2) Epoch 2, batch 34000, giga_loss[loss=0.2606, simple_loss=0.3489, pruned_loss=0.08609, over 28858.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3687, pruned_loss=0.1208, over 5658576.35 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4034, pruned_loss=0.1654, over 5675448.97 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3651, pruned_loss=0.1156, over 5674836.35 frames. ], batch size: 174, lr: 1.23e-02, grad_scale: 8.0
+2023-03-01 08:46:52,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9392, 1.7260, 1.5927, 1.5406], device='cuda:0'), covar=tensor([0.0982, 0.1945, 0.1384, 0.1427], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0797, 0.0616, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 08:46:56,149 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79678.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 08:47:17,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.758e+02 1.330e+03 1.743e+03 2.298e+03 4.136e+03, threshold=3.485e+03, percent-clipped=5.0
+2023-03-01 08:47:42,319 INFO [train.py:968] (0/2) Epoch 2, batch 34050, giga_loss[loss=0.295, simple_loss=0.386, pruned_loss=0.102, over 28900.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3677, pruned_loss=0.1186, over 5663966.01 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4032, pruned_loss=0.1652, over 5678746.41 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3642, pruned_loss=0.1136, over 5673430.17 frames. ], batch size: 174, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:47:52,094 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:48:02,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:48:31,285 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:48:42,270 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79768.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 08:48:42,581 INFO [train.py:968] (0/2) Epoch 2, batch 34100, giga_loss[loss=0.3314, simple_loss=0.4017, pruned_loss=0.1305, over 28868.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3702, pruned_loss=0.1181, over 5661274.17 frames. ], libri_tot_loss[loss=0.3672, simple_loss=0.4035, pruned_loss=0.1655, over 5669686.94 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3665, pruned_loss=0.113, over 5676552.71 frames. ], batch size: 284, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:49:17,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.447e+02 1.479e+03 1.851e+03 2.535e+03 5.234e+03, threshold=3.702e+03, percent-clipped=8.0
+2023-03-01 08:49:35,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6121, 1.5613, 1.4323, 1.5229], device='cuda:0'), covar=tensor([0.0867, 0.1481, 0.1413, 0.1171], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0776, 0.0609, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 08:49:44,285 INFO [train.py:968] (0/2) Epoch 2, batch 34150, giga_loss[loss=0.3274, simple_loss=0.3901, pruned_loss=0.1323, over 28945.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3722, pruned_loss=0.1191, over 5668710.14 frames. ], libri_tot_loss[loss=0.3672, simple_loss=0.4035, pruned_loss=0.1655, over 5671387.83 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3689, pruned_loss=0.1144, over 5679150.92 frames. ], batch size: 213, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:50:02,410 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-01 08:50:51,256 INFO [train.py:968] (0/2) Epoch 2, batch 34200, giga_loss[loss=0.3183, simple_loss=0.385, pruned_loss=0.1258, over 28642.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.373, pruned_loss=0.1204, over 5673895.05 frames. ], libri_tot_loss[loss=0.3655, simple_loss=0.4025, pruned_loss=0.1642, over 5681986.98 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3694, pruned_loss=0.1151, over 5672830.99 frames. ], batch size: 262, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:51:01,615 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:51:04,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:51:27,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.574e+02 1.617e+03 1.896e+03 2.871e+03 6.506e+03, threshold=3.792e+03, percent-clipped=9.0
+2023-03-01 08:51:49,509 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:52:02,670 INFO [train.py:968] (0/2) Epoch 2, batch 34250, libri_loss[loss=0.3341, simple_loss=0.3671, pruned_loss=0.1506, over 29609.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3717, pruned_loss=0.1192, over 5669506.11 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4023, pruned_loss=0.1641, over 5684906.93 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3686, pruned_loss=0.1145, over 5665761.26 frames. ], batch size: 69, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:53:14,403 INFO [train.py:968] (0/2) Epoch 2, batch 34300, giga_loss[loss=0.3117, simple_loss=0.3857, pruned_loss=0.1188, over 28747.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3709, pruned_loss=0.1181, over 5671453.58 frames. ], libri_tot_loss[loss=0.3643, simple_loss=0.4015, pruned_loss=0.1635, over 5687873.49 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3686, pruned_loss=0.114, over 5665434.53 frames. ], batch size: 262, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 08:53:27,848 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:53:29,788 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-01 08:53:53,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.637e+02 1.567e+03 1.898e+03 2.308e+03 4.712e+03, threshold=3.796e+03, percent-clipped=3.0
+2023-03-01 08:53:59,622 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-80000.pt
+2023-03-01 08:54:25,013 INFO [train.py:968] (0/2) Epoch 2, batch 34350, giga_loss[loss=0.3202, simple_loss=0.3806, pruned_loss=0.1299, over 27555.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3716, pruned_loss=0.1184, over 5675014.73 frames. ], libri_tot_loss[loss=0.3635, simple_loss=0.401, pruned_loss=0.163, over 5694230.85 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3693, pruned_loss=0.1143, over 5663845.11 frames. ], batch size: 472, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 08:55:07,400 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:55:11,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80053.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 08:55:30,502 INFO [train.py:968] (0/2) Epoch 2, batch 34400, giga_loss[loss=0.3141, simple_loss=0.3909, pruned_loss=0.1187, over 28723.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3764, pruned_loss=0.1215, over 5669316.64 frames. ], libri_tot_loss[loss=0.364, simple_loss=0.4013, pruned_loss=0.1634, over 5686531.13 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3739, pruned_loss=0.1172, over 5666770.34 frames. ], batch size: 243, lr: 1.22e-02, grad_scale: 8.0
+2023-03-01 08:56:00,847 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.857e+02 1.484e+03 1.883e+03 2.542e+03 5.227e+03, threshold=3.766e+03, percent-clipped=2.0
+2023-03-01 08:56:09,168 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:56:35,033 INFO [train.py:968] (0/2) Epoch 2, batch 34450, giga_loss[loss=0.2808, simple_loss=0.3529, pruned_loss=0.1044, over 29079.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3763, pruned_loss=0.1216, over 5668590.14 frames. ], libri_tot_loss[loss=0.3634, simple_loss=0.4007, pruned_loss=0.163, over 5683888.88 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.374, pruned_loss=0.117, over 5669142.96 frames. ], batch size: 120, lr: 1.22e-02, grad_scale: 8.0
+2023-03-01 08:56:54,954 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:57:11,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80143.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 08:57:29,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:57:42,167 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.87 vs. limit=5.0
+2023-03-01 08:57:45,135 INFO [train.py:968] (0/2) Epoch 2, batch 34500, giga_loss[loss=0.3194, simple_loss=0.3837, pruned_loss=0.1275, over 28563.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3745, pruned_loss=0.1214, over 5667942.73 frames. ], libri_tot_loss[loss=0.3633, simple_loss=0.4007, pruned_loss=0.163, over 5684686.98 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3721, pruned_loss=0.117, over 5667394.79 frames. ], batch size: 370, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 08:58:04,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9850, 1.7666, 1.6533, 1.6722], device='cuda:0'), covar=tensor([0.0699, 0.1406, 0.1182, 0.1172], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0772, 0.0609, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 08:58:14,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0436, 1.3738, 1.0545, 0.2404], device='cuda:0'), covar=tensor([0.0940, 0.0929, 0.1561, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.1195, 0.1184, 0.1192, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 08:58:24,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.297e+02 1.560e+03 2.124e+03 2.846e+03 9.769e+03, threshold=4.247e+03, percent-clipped=15.0
+2023-03-01 08:58:25,139 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80196.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 08:58:28,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80199.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 08:58:32,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1317, 1.2820, 1.1415, 1.0033], device='cuda:0'), covar=tensor([0.1832, 0.1692, 0.1502, 0.1663], device='cuda:0'), in_proj_covar=tensor([0.0973, 0.0792, 0.0884, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 08:58:55,051 INFO [train.py:968] (0/2) Epoch 2, batch 34550, giga_loss[loss=0.2762, simple_loss=0.3367, pruned_loss=0.1079, over 24610.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3717, pruned_loss=0.1197, over 5673608.50 frames. ], libri_tot_loss[loss=0.3637, simple_loss=0.4011, pruned_loss=0.1632, over 5679491.62 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3687, pruned_loss=0.1148, over 5676719.86 frames. ], batch size: 705, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 08:59:07,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80228.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 08:59:32,879 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:59:40,035 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:00:06,929 INFO [train.py:968] (0/2) Epoch 2, batch 34600, giga_loss[loss=0.3025, simple_loss=0.3693, pruned_loss=0.1178, over 28922.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3689, pruned_loss=0.1168, over 5664050.53 frames. ], libri_tot_loss[loss=0.3637, simple_loss=0.401, pruned_loss=0.1632, over 5665280.13 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.366, pruned_loss=0.112, over 5678105.50 frames. ], batch size: 213, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:00:15,786 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:00:16,500 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:00:19,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:00:28,483 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80286.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 09:00:32,489 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80289.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 09:00:39,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3659, 1.3781, 1.4521, 1.3918], device='cuda:0'), covar=tensor([0.0824, 0.1009, 0.0990, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0784, 0.0616, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 09:00:42,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.915e+02 1.143e+03 1.558e+03 2.250e+03 5.080e+03, threshold=3.117e+03, percent-clipped=1.0
+2023-03-01 09:00:59,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7301, 3.3241, 3.4838, 1.5532], device='cuda:0'), covar=tensor([0.0532, 0.0467, 0.0832, 0.2173], device='cuda:0'), in_proj_covar=tensor([0.0747, 0.0540, 0.0748, 0.0535], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:00:59,721 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:01:11,200 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80318.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 09:01:11,625 INFO [train.py:968] (0/2) Epoch 2, batch 34650, giga_loss[loss=0.2657, simple_loss=0.3369, pruned_loss=0.09724, over 28730.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.37, pruned_loss=0.1178, over 5660683.71 frames. ], libri_tot_loss[loss=0.3639, simple_loss=0.4012, pruned_loss=0.1633, over 5669216.62 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.367, pruned_loss=0.1131, over 5668379.13 frames. ], batch size: 92, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:01:39,054 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4087, 2.0223, 1.6882, 1.6890], device='cuda:0'), covar=tensor([0.1284, 0.1486, 0.1091, 0.0791], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0775, 0.0703, 0.0603], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:01:55,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:01:57,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5712, 2.0313, 1.6703, 1.8498], device='cuda:0'), covar=tensor([0.1194, 0.1157, 0.1005, 0.0677], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0768, 0.0697, 0.0598], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:02:13,838 INFO [train.py:968] (0/2) Epoch 2, batch 34700, giga_loss[loss=0.3852, simple_loss=0.4356, pruned_loss=0.1674, over 27561.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3733, pruned_loss=0.1201, over 5662595.67 frames. ], libri_tot_loss[loss=0.3638, simple_loss=0.401, pruned_loss=0.1633, over 5672772.66 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3705, pruned_loss=0.1156, over 5665393.33 frames. ], batch size: 472, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:02:27,973 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-01 09:02:46,168 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.804e+02 1.406e+03 1.811e+03 2.404e+03 6.947e+03, threshold=3.621e+03, percent-clipped=13.0
+2023-03-01 09:02:52,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4007, 1.4547, 1.1282, 1.2781], device='cuda:0'), covar=tensor([0.0751, 0.0604, 0.1031, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0502, 0.0543, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 09:03:07,971 INFO [train.py:968] (0/2) Epoch 2, batch 34750, giga_loss[loss=0.2904, simple_loss=0.3674, pruned_loss=0.1067, over 28965.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3745, pruned_loss=0.1209, over 5667990.80 frames. ], libri_tot_loss[loss=0.3632, simple_loss=0.4006, pruned_loss=0.163, over 5669742.81 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3716, pruned_loss=0.116, over 5673794.83 frames. ], batch size: 164, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:03:16,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:03:28,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5134, 2.0726, 1.7003, 1.7552], device='cuda:0'), covar=tensor([0.1443, 0.1613, 0.1207, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0764, 0.0693, 0.0596], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:04:10,262 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-01 09:04:10,477 INFO [train.py:968] (0/2) Epoch 2, batch 34800, libri_loss[loss=0.354, simple_loss=0.3988, pruned_loss=0.1546, over 29510.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3708, pruned_loss=0.12, over 5663836.32 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.4003, pruned_loss=0.1627, over 5673889.81 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3682, pruned_loss=0.1156, over 5664754.50 frames. ], batch size: 82, lr: 1.22e-02, grad_scale: 8.0
+2023-03-01 09:04:44,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.748e+02 1.588e+03 2.022e+03 2.813e+03 5.976e+03, threshold=4.045e+03, percent-clipped=11.0
+2023-03-01 09:04:45,979 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:04:47,710 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-01 09:04:49,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:05:14,444 INFO [train.py:968] (0/2) Epoch 2, batch 34850, giga_loss[loss=0.2743, simple_loss=0.3496, pruned_loss=0.09954, over 28786.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3705, pruned_loss=0.1208, over 5662233.40 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.4003, pruned_loss=0.1627, over 5675952.94 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3679, pruned_loss=0.1164, over 5660833.72 frames. ], batch size: 119, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:05:22,612 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:05:25,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:06:07,005 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:06:07,434 INFO [train.py:968] (0/2) Epoch 2, batch 34900, giga_loss[loss=0.3707, simple_loss=0.4259, pruned_loss=0.1577, over 28888.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.374, pruned_loss=0.1244, over 5665725.33 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.3997, pruned_loss=0.1623, over 5680080.74 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3713, pruned_loss=0.1195, over 5660439.48 frames. ], batch size: 174, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:06:09,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:06:36,295 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.574e+02 1.602e+03 2.017e+03 2.897e+03 6.914e+03, threshold=4.034e+03, percent-clipped=11.0
+2023-03-01 09:06:40,007 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:06:56,330 INFO [train.py:968] (0/2) Epoch 2, batch 34950, giga_loss[loss=0.3703, simple_loss=0.4246, pruned_loss=0.158, over 28904.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3861, pruned_loss=0.1327, over 5673804.47 frames. ], libri_tot_loss[loss=0.3622, simple_loss=0.3997, pruned_loss=0.1623, over 5678631.17 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3838, pruned_loss=0.1286, over 5670781.45 frames. ], batch size: 199, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:07:20,317 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.85 vs. limit=5.0
+2023-03-01 09:07:43,280 INFO [train.py:968] (0/2) Epoch 2, batch 35000, giga_loss[loss=0.3506, simple_loss=0.408, pruned_loss=0.1465, over 28801.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3956, pruned_loss=0.1394, over 5670845.28 frames. ], libri_tot_loss[loss=0.3626, simple_loss=0.4003, pruned_loss=0.1624, over 5680713.20 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3929, pruned_loss=0.1352, over 5666279.87 frames. ], batch size: 284, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:07:46,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:07:48,299 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:07:48,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:08:07,902 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.687e+02 1.285e+03 1.568e+03 2.063e+03 4.305e+03, threshold=3.136e+03, percent-clipped=1.0
+2023-03-01 09:08:14,879 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:08:26,313 INFO [train.py:968] (0/2) Epoch 2, batch 35050, giga_loss[loss=0.3126, simple_loss=0.373, pruned_loss=0.1262, over 28712.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3953, pruned_loss=0.1405, over 5677460.64 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4004, pruned_loss=0.1625, over 5684711.30 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.393, pruned_loss=0.1366, over 5670237.59 frames. ], batch size: 262, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:09:10,205 INFO [train.py:968] (0/2) Epoch 2, batch 35100, giga_loss[loss=0.307, simple_loss=0.366, pruned_loss=0.124, over 28913.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3896, pruned_loss=0.1384, over 5685468.00 frames. ], libri_tot_loss[loss=0.3639, simple_loss=0.4015, pruned_loss=0.1631, over 5684727.65 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3863, pruned_loss=0.1336, over 5679715.20 frames. ], batch size: 145, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:09:25,166 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:09:34,957 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.053e+02 1.158e+03 1.607e+03 2.360e+03 8.759e+03, threshold=3.213e+03, percent-clipped=12.0
+2023-03-01 09:09:56,675 INFO [train.py:968] (0/2) Epoch 2, batch 35150, giga_loss[loss=0.2886, simple_loss=0.3379, pruned_loss=0.1196, over 28689.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3807, pruned_loss=0.1337, over 5685388.69 frames. ], libri_tot_loss[loss=0.3638, simple_loss=0.4014, pruned_loss=0.1631, over 5687138.57 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3781, pruned_loss=0.1297, over 5678932.19 frames. ], batch size: 85, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:10:30,429 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:10:40,760 INFO [train.py:968] (0/2) Epoch 2, batch 35200, giga_loss[loss=0.2766, simple_loss=0.3399, pruned_loss=0.1067, over 28824.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3732, pruned_loss=0.1303, over 5679859.38 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.402, pruned_loss=0.1635, over 5682403.50 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3699, pruned_loss=0.126, over 5679650.17 frames. ], batch size: 186, lr: 1.22e-02, grad_scale: 8.0
+2023-03-01 09:10:54,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8537, 1.4863, 3.7564, 2.9503], device='cuda:0'), covar=tensor([0.1436, 0.1722, 0.0320, 0.0622], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0504, 0.0659, 0.0519], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 09:10:56,157 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:11:03,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0884, 4.4298, 4.7697, 2.0753], device='cuda:0'), covar=tensor([0.0307, 0.0284, 0.0632, 0.1842], device='cuda:0'), in_proj_covar=tensor([0.0764, 0.0553, 0.0769, 0.0547], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:11:03,549 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.449e+02 1.065e+03 1.458e+03 2.222e+03 7.854e+03, threshold=2.916e+03, percent-clipped=11.0
+2023-03-01 09:11:22,211 INFO [train.py:968] (0/2) Epoch 2, batch 35250, giga_loss[loss=0.3144, simple_loss=0.3513, pruned_loss=0.1387, over 26490.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3659, pruned_loss=0.1269, over 5675268.41 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4028, pruned_loss=0.1643, over 5676009.32 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3617, pruned_loss=0.1219, over 5681638.06 frames. ], batch size: 555, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:12:02,494 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-01 09:12:04,736 INFO [train.py:968] (0/2) Epoch 2, batch 35300, giga_loss[loss=0.263, simple_loss=0.3262, pruned_loss=0.09996, over 28781.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3592, pruned_loss=0.1233, over 5685182.44 frames. ], libri_tot_loss[loss=0.3658, simple_loss=0.4031, pruned_loss=0.1642, over 5683215.46 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3544, pruned_loss=0.1182, over 5683804.97 frames. ], batch size: 99, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:12:12,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7276, 2.5281, 1.6881, 0.8690], device='cuda:0'), covar=tensor([0.2360, 0.1234, 0.1740, 0.2319], device='cuda:0'), in_proj_covar=tensor([0.1170, 0.1148, 0.1187, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 09:12:29,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.127e+02 1.057e+03 1.474e+03 1.982e+03 4.516e+03, threshold=2.949e+03, percent-clipped=9.0
+2023-03-01 09:12:47,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6740, 1.4112, 1.2532, 1.3052], device='cuda:0'), covar=tensor([0.0539, 0.0535, 0.0777, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0493, 0.0539, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 09:12:48,742 INFO [train.py:968] (0/2) Epoch 2, batch 35350, giga_loss[loss=0.284, simple_loss=0.3458, pruned_loss=0.1111, over 28802.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3556, pruned_loss=0.1212, over 5685112.74 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.4034, pruned_loss=0.1644, over 5686820.48 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3506, pruned_loss=0.1162, over 5680825.25 frames. ], batch size: 243, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:13:00,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4451, 1.4383, 1.2462, 1.4446], device='cuda:0'), covar=tensor([0.1988, 0.1999, 0.1742, 0.2128], device='cuda:0'), in_proj_covar=tensor([0.0988, 0.0801, 0.0896, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 09:13:19,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:13:35,486 INFO [train.py:968] (0/2) Epoch 2, batch 35400, giga_loss[loss=0.2921, simple_loss=0.3228, pruned_loss=0.1307, over 23811.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3502, pruned_loss=0.1175, over 5682412.70 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4037, pruned_loss=0.1646, over 5677631.38 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3459, pruned_loss=0.1132, over 5687739.19 frames. ], batch size: 705, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:13:41,706 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-01 09:13:59,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.453e+02 1.140e+03 1.511e+03 1.950e+03 9.570e+03, threshold=3.022e+03, percent-clipped=11.0
+2023-03-01 09:14:04,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7543, 2.1899, 1.9765, 1.8883], device='cuda:0'), covar=tensor([0.1368, 0.1530, 0.1097, 0.0751], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0817, 0.0726, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:14:19,853 INFO [train.py:968] (0/2) Epoch 2, batch 35450, giga_loss[loss=0.3174, simple_loss=0.3627, pruned_loss=0.136, over 26593.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3463, pruned_loss=0.1147, over 5695641.65 frames. ], libri_tot_loss[loss=0.3664, simple_loss=0.4037, pruned_loss=0.1645, over 5678838.73 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3422, pruned_loss=0.1108, over 5698973.83 frames. ], batch size: 555, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:14:30,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=81129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:14:46,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0699, 1.1401, 0.7982, 0.5726], device='cuda:0'), covar=tensor([0.0508, 0.0475, 0.0410, 0.0492], device='cuda:0'), in_proj_covar=tensor([0.1075, 0.0767, 0.0851, 0.0932], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 09:14:59,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:15:06,109 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=81166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:15:07,681 INFO [train.py:968] (0/2) Epoch 2, batch 35500, giga_loss[loss=0.2358, simple_loss=0.3012, pruned_loss=0.08515, over 28991.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3434, pruned_loss=0.1136, over 5685918.96 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4039, pruned_loss=0.1647, over 5671578.28 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3395, pruned_loss=0.11, over 5695525.82 frames. ], batch size: 136, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:15:22,115 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5977, 3.9254, 4.2779, 1.6749], device='cuda:0'), covar=tensor([0.0500, 0.0497, 0.1013, 0.1942], device='cuda:0'), in_proj_covar=tensor([0.0738, 0.0543, 0.0757, 0.0543], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:15:27,593 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:15:30,135 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:15:30,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.277e+02 9.941e+02 1.341e+03 1.761e+03 5.413e+03, threshold=2.683e+03, percent-clipped=5.0
+2023-03-01 09:15:48,976 INFO [train.py:968] (0/2) Epoch 2, batch 35550, giga_loss[loss=0.2522, simple_loss=0.3175, pruned_loss=0.09349, over 28936.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3406, pruned_loss=0.1119, over 5680690.49 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4048, pruned_loss=0.1652, over 5658078.88 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3353, pruned_loss=0.1073, over 5699877.36 frames. ], batch size: 106, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:15:55,785 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:16:00,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1586, 4.0697, 2.0132, 2.1905], device='cuda:0'), covar=tensor([0.0731, 0.0400, 0.0831, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0450, 0.0330, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0017], device='cuda:0')
+2023-03-01 09:16:00,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:16:25,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:16:31,392 INFO [train.py:968] (0/2) Epoch 2, batch 35600, giga_loss[loss=0.2318, simple_loss=0.3021, pruned_loss=0.08075, over 28908.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3372, pruned_loss=0.1097, over 5686639.14 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4047, pruned_loss=0.165, over 5662211.42 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3316, pruned_loss=0.105, over 5698665.65 frames. ], batch size: 186, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:16:35,783 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4504, 3.7975, 4.1468, 1.7002], device='cuda:0'), covar=tensor([0.0422, 0.0427, 0.0701, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.0726, 0.0534, 0.0749, 0.0540], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:16:43,609 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-01 09:16:57,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.267e+02 1.052e+03 1.258e+03 1.609e+03 3.971e+03, threshold=2.515e+03, percent-clipped=6.0
+2023-03-01 09:17:02,740 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:17:03,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7688, 2.6698, 1.7895, 0.7934], device='cuda:0'), covar=tensor([0.2363, 0.1077, 0.1499, 0.2346], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.1144, 0.1189, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 09:17:04,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:17:15,700 INFO [train.py:968] (0/2) Epoch 2, batch 35650, giga_loss[loss=0.3318, simple_loss=0.3623, pruned_loss=0.1507, over 26668.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3353, pruned_loss=0.1089, over 5674884.28 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4061, pruned_loss=0.1659, over 5659565.20 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3276, pruned_loss=0.1028, over 5687101.54 frames. ], batch size: 555, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:17:31,960 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:04,548 INFO [train.py:968] (0/2) Epoch 2, batch 35700, giga_loss[loss=0.2855, simple_loss=0.3395, pruned_loss=0.1157, over 28676.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3317, pruned_loss=0.1071, over 5681393.40 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4065, pruned_loss=0.1662, over 5659550.34 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3243, pruned_loss=0.1012, over 5691104.07 frames. ], batch size: 262, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:18:08,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3326, 2.1658, 1.2684, 1.1987], device='cuda:0'), covar=tensor([0.1133, 0.0600, 0.1166, 0.1822], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0448, 0.0323, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 09:18:08,915 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:11,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:30,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.421e+02 1.161e+03 1.521e+03 2.194e+03 5.537e+03, threshold=3.041e+03, percent-clipped=15.0
+2023-03-01 09:18:38,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:40,448 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:42,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:52,123 INFO [train.py:968] (0/2) Epoch 2, batch 35750, giga_loss[loss=0.3463, simple_loss=0.4053, pruned_loss=0.1437, over 28933.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3412, pruned_loss=0.1134, over 5679921.74 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4071, pruned_loss=0.1666, over 5661655.69 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3342, pruned_loss=0.1079, over 5685873.10 frames. ], batch size: 213, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:19:08,696 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:19:39,022 INFO [train.py:968] (0/2) Epoch 2, batch 35800, giga_loss[loss=0.3639, simple_loss=0.4171, pruned_loss=0.1553, over 28906.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3561, pruned_loss=0.1225, over 5685551.73 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4072, pruned_loss=0.1667, over 5665459.50 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3497, pruned_loss=0.1173, over 5687149.04 frames. ], batch size: 213, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:20:07,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.097e+02 1.345e+03 1.729e+03 2.245e+03 5.559e+03, threshold=3.457e+03, percent-clipped=6.0
+2023-03-01 09:20:14,395 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:20:18,667 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=81509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:20:25,823 INFO [train.py:968] (0/2) Epoch 2, batch 35850, giga_loss[loss=0.369, simple_loss=0.4264, pruned_loss=0.1558, over 28783.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3706, pruned_loss=0.1309, over 5694312.06 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4075, pruned_loss=0.1669, over 5671634.62 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3642, pruned_loss=0.1257, over 5690507.30 frames. ], batch size: 284, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:20:34,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5995, 2.6601, 1.5452, 1.3332], device='cuda:0'), covar=tensor([0.0895, 0.0399, 0.0862, 0.1485], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0442, 0.0322, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 09:20:48,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:21:13,154 INFO [train.py:968] (0/2) Epoch 2, batch 35900, libri_loss[loss=0.4583, simple_loss=0.4806, pruned_loss=0.218, over 26140.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3796, pruned_loss=0.1353, over 5678272.64 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4086, pruned_loss=0.1677, over 5659920.10 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3727, pruned_loss=0.1298, over 5686330.62 frames. ], batch size: 136, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:21:22,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0287, 1.7399, 1.6674, 1.6979], device='cuda:0'), covar=tensor([0.0899, 0.1758, 0.1360, 0.1308], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0813, 0.0642, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 09:21:36,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.891e+02 1.321e+03 1.606e+03 2.231e+03 6.048e+03, threshold=3.212e+03, percent-clipped=7.0
+2023-03-01 09:21:53,102 INFO [train.py:968] (0/2) Epoch 2, batch 35950, giga_loss[loss=0.3377, simple_loss=0.4032, pruned_loss=0.1361, over 28865.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.384, pruned_loss=0.1363, over 5671609.94 frames. ], libri_tot_loss[loss=0.373, simple_loss=0.4097, pruned_loss=0.1682, over 5658034.39 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3765, pruned_loss=0.1303, over 5681507.42 frames. ], batch size: 174, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:22:19,167 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:22:21,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:22:35,739 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-01 09:22:39,056 INFO [train.py:968] (0/2) Epoch 2, batch 36000, giga_loss[loss=0.3369, simple_loss=0.3963, pruned_loss=0.1387, over 28951.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3862, pruned_loss=0.1362, over 5682356.56 frames. ], libri_tot_loss[loss=0.373, simple_loss=0.41, pruned_loss=0.168, over 5662539.10 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3793, pruned_loss=0.1306, over 5686783.54 frames. ], batch size: 164, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:22:39,061 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 09:22:47,850 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.2604, simple_loss=0.3566, pruned_loss=0.08209, over 944034.00 frames. 
+2023-03-01 09:22:47,851 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19537MB
+2023-03-01 09:22:57,719 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:23:05,558 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:23:08,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:23:12,763 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=81692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:23:14,070 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=81694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:23:16,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.309e+02 1.091e+03 1.402e+03 2.066e+03 4.078e+03, threshold=2.805e+03, percent-clipped=3.0
+2023-03-01 09:23:29,223 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5411, 3.9010, 4.2517, 1.7508], device='cuda:0'), covar=tensor([0.0416, 0.0411, 0.0864, 0.2027], device='cuda:0'), in_proj_covar=tensor([0.0746, 0.0541, 0.0771, 0.0551], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:23:31,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:23:33,783 INFO [train.py:968] (0/2) Epoch 2, batch 36050, giga_loss[loss=0.3267, simple_loss=0.3926, pruned_loss=0.1304, over 28962.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3878, pruned_loss=0.1369, over 5688228.95 frames. ], libri_tot_loss[loss=0.3728, simple_loss=0.41, pruned_loss=0.1678, over 5665865.18 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3814, pruned_loss=0.1315, over 5689488.45 frames. ], batch size: 227, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:24:16,331 INFO [train.py:968] (0/2) Epoch 2, batch 36100, giga_loss[loss=0.3335, simple_loss=0.3925, pruned_loss=0.1372, over 28824.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3916, pruned_loss=0.1404, over 5684231.06 frames. ], libri_tot_loss[loss=0.3732, simple_loss=0.4103, pruned_loss=0.168, over 5670527.55 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3857, pruned_loss=0.1352, over 5681458.24 frames. ], batch size: 186, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:24:35,756 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2993, 1.3035, 1.0081, 0.7568], device='cuda:0'), covar=tensor([0.0532, 0.0433, 0.0458, 0.0541], device='cuda:0'), in_proj_covar=tensor([0.1071, 0.0759, 0.0844, 0.0916], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 09:24:40,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2292, 1.3186, 1.2333, 1.2331], device='cuda:0'), covar=tensor([0.0936, 0.0963, 0.1433, 0.1061], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0820, 0.0642, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 09:24:41,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.044e+02 1.224e+03 1.567e+03 2.074e+03 5.143e+03, threshold=3.134e+03, percent-clipped=10.0
+2023-03-01 09:24:59,703 INFO [train.py:968] (0/2) Epoch 2, batch 36150, giga_loss[loss=0.3113, simple_loss=0.3708, pruned_loss=0.1258, over 28935.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3944, pruned_loss=0.1425, over 5684256.16 frames. ], libri_tot_loss[loss=0.3728, simple_loss=0.41, pruned_loss=0.1678, over 5672961.65 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3896, pruned_loss=0.1381, over 5680205.85 frames. ], batch size: 106, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:25:38,523 INFO [train.py:968] (0/2) Epoch 2, batch 36200, giga_loss[loss=0.3714, simple_loss=0.4252, pruned_loss=0.1588, over 28824.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.398, pruned_loss=0.1444, over 5684430.91 frames. ], libri_tot_loss[loss=0.3739, simple_loss=0.411, pruned_loss=0.1684, over 5666619.72 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.393, pruned_loss=0.1397, over 5686468.68 frames. ], batch size: 262, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:25:51,423 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:26:05,383 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.112e+02 1.279e+03 1.699e+03 2.215e+03 4.759e+03, threshold=3.399e+03, percent-clipped=9.0
+2023-03-01 09:26:23,143 INFO [train.py:968] (0/2) Epoch 2, batch 36250, giga_loss[loss=0.3161, simple_loss=0.3896, pruned_loss=0.1213, over 28882.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.4, pruned_loss=0.1448, over 5683711.80 frames. ], libri_tot_loss[loss=0.3738, simple_loss=0.4111, pruned_loss=0.1683, over 5669740.22 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3958, pruned_loss=0.1408, over 5682819.67 frames. ], batch size: 174, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:26:26,930 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6647, 1.4254, 3.7840, 3.0542], device='cuda:0'), covar=tensor([0.1630, 0.1864, 0.0312, 0.0456], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0487, 0.0631, 0.0511], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 09:26:43,671 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4574, 1.3385, 1.3000, 1.5088], device='cuda:0'), covar=tensor([0.1879, 0.1866, 0.1571, 0.1839], device='cuda:0'), in_proj_covar=tensor([0.0991, 0.0809, 0.0898, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 09:26:55,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6406, 1.4393, 1.2054, 1.2216], device='cuda:0'), covar=tensor([0.0609, 0.0593, 0.0898, 0.0947], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0498, 0.0539, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 09:27:07,422 INFO [train.py:968] (0/2) Epoch 2, batch 36300, giga_loss[loss=0.3397, simple_loss=0.4038, pruned_loss=0.1377, over 28890.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.4014, pruned_loss=0.1446, over 5684150.49 frames. ], libri_tot_loss[loss=0.3742, simple_loss=0.4115, pruned_loss=0.1685, over 5672816.36 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3975, pruned_loss=0.1408, over 5680763.29 frames. ], batch size: 213, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:27:32,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.392e+02 1.131e+03 1.461e+03 1.879e+03 4.191e+03, threshold=2.922e+03, percent-clipped=3.0
+2023-03-01 09:27:32,955 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-82000.pt
+2023-03-01 09:27:50,739 INFO [train.py:968] (0/2) Epoch 2, batch 36350, giga_loss[loss=0.3013, simple_loss=0.3738, pruned_loss=0.1144, over 28581.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.4007, pruned_loss=0.1424, over 5694233.51 frames. ], libri_tot_loss[loss=0.3748, simple_loss=0.412, pruned_loss=0.1688, over 5676166.60 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.397, pruned_loss=0.1388, over 5688826.68 frames. ], batch size: 85, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:27:58,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:00,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:25,514 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:29,303 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:32,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:33,413 INFO [train.py:968] (0/2) Epoch 2, batch 36400, giga_loss[loss=0.2943, simple_loss=0.3667, pruned_loss=0.111, over 28888.00 frames. ], tot_loss[loss=0.338, simple_loss=0.398, pruned_loss=0.139, over 5689778.97 frames. ], libri_tot_loss[loss=0.3754, simple_loss=0.4125, pruned_loss=0.1691, over 5668677.48 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3945, pruned_loss=0.1357, over 5692969.31 frames. ], batch size: 99, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:28:33,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:54,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 09:29:01,653 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.282e+02 9.852e+02 1.276e+03 1.631e+03 4.304e+03, threshold=2.552e+03, percent-clipped=4.0
+2023-03-01 09:29:16,385 INFO [train.py:968] (0/2) Epoch 2, batch 36450, giga_loss[loss=0.3591, simple_loss=0.409, pruned_loss=0.1546, over 28284.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3961, pruned_loss=0.1372, over 5699909.04 frames. ], libri_tot_loss[loss=0.3759, simple_loss=0.4129, pruned_loss=0.1694, over 5670563.86 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3925, pruned_loss=0.1334, over 5701156.71 frames. ], batch size: 77, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:30:03,816 INFO [train.py:968] (0/2) Epoch 2, batch 36500, giga_loss[loss=0.3517, simple_loss=0.4091, pruned_loss=0.1471, over 28882.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3977, pruned_loss=0.1401, over 5698884.91 frames. ], libri_tot_loss[loss=0.377, simple_loss=0.4138, pruned_loss=0.1701, over 5673359.93 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3937, pruned_loss=0.1357, over 5697971.32 frames. ], batch size: 199, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:30:31,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.483e+02 1.274e+03 1.731e+03 2.537e+03 6.776e+03, threshold=3.462e+03, percent-clipped=23.0
+2023-03-01 09:30:39,946 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:30:42,308 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:30:44,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:30:45,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:30:49,052 INFO [train.py:968] (0/2) Epoch 2, batch 36550, giga_loss[loss=0.3077, simple_loss=0.3763, pruned_loss=0.1195, over 28267.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.4019, pruned_loss=0.1468, over 5688889.00 frames. ], libri_tot_loss[loss=0.3784, simple_loss=0.4149, pruned_loss=0.1709, over 5668457.94 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3972, pruned_loss=0.1417, over 5693079.51 frames. ], batch size: 77, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:30:49,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9328, 2.4161, 2.0905, 2.0170], device='cuda:0'), covar=tensor([0.1488, 0.1390, 0.1007, 0.0723], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0784, 0.0699, 0.0599], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:31:05,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3622, 1.3825, 5.0582, 3.6329], device='cuda:0'), covar=tensor([0.1970, 0.2274, 0.0432, 0.0485], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0492, 0.0642, 0.0514], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 09:31:07,796 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:31:09,118 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:31:29,311 INFO [train.py:968] (0/2) Epoch 2, batch 36600, giga_loss[loss=0.3631, simple_loss=0.4058, pruned_loss=0.1602, over 27920.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4027, pruned_loss=0.1491, over 5697780.62 frames. ], libri_tot_loss[loss=0.3788, simple_loss=0.4152, pruned_loss=0.1712, over 5674545.37 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3984, pruned_loss=0.1442, over 5696021.75 frames. ], batch size: 412, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:32:00,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.080e+02 1.318e+03 1.766e+03 2.150e+03 5.157e+03, threshold=3.532e+03, percent-clipped=5.0
+2023-03-01 09:32:16,275 INFO [train.py:968] (0/2) Epoch 2, batch 36650, giga_loss[loss=0.2932, simple_loss=0.3572, pruned_loss=0.1146, over 28372.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4001, pruned_loss=0.1475, over 5705357.39 frames. ], libri_tot_loss[loss=0.3789, simple_loss=0.4155, pruned_loss=0.1712, over 5677662.55 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3963, pruned_loss=0.1433, over 5701553.09 frames. ], batch size: 65, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:32:57,737 INFO [train.py:968] (0/2) Epoch 2, batch 36700, giga_loss[loss=0.355, simple_loss=0.4031, pruned_loss=0.1534, over 28791.00 frames. ], tot_loss[loss=0.347, simple_loss=0.3991, pruned_loss=0.1475, over 5704584.81 frames. ], libri_tot_loss[loss=0.3797, simple_loss=0.4161, pruned_loss=0.1716, over 5681977.46 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3951, pruned_loss=0.1433, over 5698130.35 frames. ], batch size: 284, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:33:00,121 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-01 09:33:08,927 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:33:20,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6036, 1.8788, 1.6902, 1.7364], device='cuda:0'), covar=tensor([0.1104, 0.1338, 0.0937, 0.0657], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0797, 0.0708, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:33:28,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.784e+02 1.305e+03 1.646e+03 2.359e+03 4.836e+03, threshold=3.291e+03, percent-clipped=8.0
+2023-03-01 09:33:43,386 INFO [train.py:968] (0/2) Epoch 2, batch 36750, giga_loss[loss=0.2983, simple_loss=0.3612, pruned_loss=0.1177, over 29045.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3983, pruned_loss=0.1461, over 5699499.49 frames. ], libri_tot_loss[loss=0.3804, simple_loss=0.4167, pruned_loss=0.172, over 5677398.32 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3943, pruned_loss=0.142, over 5699599.47 frames. ], batch size: 128, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:34:02,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:34:11,429 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:34:29,569 INFO [train.py:968] (0/2) Epoch 2, batch 36800, giga_loss[loss=0.3555, simple_loss=0.4123, pruned_loss=0.1493, over 28895.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.397, pruned_loss=0.1444, over 5689517.85 frames. ], libri_tot_loss[loss=0.3808, simple_loss=0.4172, pruned_loss=0.1721, over 5682652.92 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3928, pruned_loss=0.1403, over 5685009.79 frames. ], batch size: 174, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:35:01,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.324e+02 1.011e+03 1.390e+03 1.692e+03 4.210e+03, threshold=2.781e+03, percent-clipped=2.0
+2023-03-01 09:35:09,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7783, 1.5773, 1.4636, 1.4782], device='cuda:0'), covar=tensor([0.0895, 0.1499, 0.1385, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0465, 0.0796, 0.0628, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 09:35:16,182 INFO [train.py:968] (0/2) Epoch 2, batch 36850, giga_loss[loss=0.3182, simple_loss=0.3805, pruned_loss=0.1279, over 28976.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3921, pruned_loss=0.1405, over 5691919.29 frames. ], libri_tot_loss[loss=0.3812, simple_loss=0.4177, pruned_loss=0.1724, over 5685950.58 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3881, pruned_loss=0.1367, over 5685435.80 frames. ], batch size: 213, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:36:02,842 INFO [train.py:968] (0/2) Epoch 2, batch 36900, giga_loss[loss=0.2906, simple_loss=0.3439, pruned_loss=0.1187, over 27731.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3843, pruned_loss=0.1359, over 5679332.55 frames. ], libri_tot_loss[loss=0.3817, simple_loss=0.4182, pruned_loss=0.1727, over 5688656.95 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3801, pruned_loss=0.1319, over 5671547.07 frames. ], batch size: 474, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:36:16,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:36:20,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:36:38,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.419e+02 1.027e+03 1.330e+03 1.827e+03 4.889e+03, threshold=2.660e+03, percent-clipped=9.0
+2023-03-01 09:36:50,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:36:55,381 INFO [train.py:968] (0/2) Epoch 2, batch 36950, giga_loss[loss=0.267, simple_loss=0.3315, pruned_loss=0.1013, over 28606.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3785, pruned_loss=0.1331, over 5665182.46 frames. ], libri_tot_loss[loss=0.382, simple_loss=0.4185, pruned_loss=0.1728, over 5689472.94 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3741, pruned_loss=0.1291, over 5657872.49 frames. ], batch size: 336, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:37:43,811 INFO [train.py:968] (0/2) Epoch 2, batch 37000, giga_loss[loss=0.326, simple_loss=0.386, pruned_loss=0.133, over 29056.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.375, pruned_loss=0.1306, over 5661851.79 frames. ], libri_tot_loss[loss=0.3821, simple_loss=0.4185, pruned_loss=0.1728, over 5691407.10 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3712, pruned_loss=0.1271, over 5654292.46 frames. ], batch size: 155, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:38:08,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5929, 1.9102, 1.7406, 1.7360], device='cuda:0'), covar=tensor([0.1587, 0.1900, 0.1245, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0810, 0.0718, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:38:13,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.297e+02 1.048e+03 1.274e+03 1.962e+03 9.389e+03, threshold=2.548e+03, percent-clipped=11.0
+2023-03-01 09:38:27,871 INFO [train.py:968] (0/2) Epoch 2, batch 37050, giga_loss[loss=0.3198, simple_loss=0.3649, pruned_loss=0.1373, over 24150.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3762, pruned_loss=0.1308, over 5672441.10 frames. ], libri_tot_loss[loss=0.3832, simple_loss=0.4195, pruned_loss=0.1734, over 5695801.66 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3712, pruned_loss=0.1264, over 5661805.48 frames. ], batch size: 705, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:38:31,132 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-01 09:38:44,348 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:38:46,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9143, 1.0951, 1.0013, 0.4963], device='cuda:0'), covar=tensor([0.0525, 0.0304, 0.0322, 0.0546], device='cuda:0'), in_proj_covar=tensor([0.1100, 0.0767, 0.0871, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 09:38:57,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:39:10,536 INFO [train.py:968] (0/2) Epoch 2, batch 37100, giga_loss[loss=0.353, simple_loss=0.3938, pruned_loss=0.1561, over 28911.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3753, pruned_loss=0.1303, over 5687087.05 frames. ], libri_tot_loss[loss=0.384, simple_loss=0.4202, pruned_loss=0.1739, over 5698808.29 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3702, pruned_loss=0.1258, over 5675774.34 frames. ], batch size: 213, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:39:36,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.668e+02 1.010e+03 1.371e+03 1.895e+03 8.773e+03, threshold=2.743e+03, percent-clipped=13.0
+2023-03-01 09:39:51,217 INFO [train.py:968] (0/2) Epoch 2, batch 37150, giga_loss[loss=0.2909, simple_loss=0.3469, pruned_loss=0.1175, over 28855.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.372, pruned_loss=0.1283, over 5680996.85 frames. ], libri_tot_loss[loss=0.385, simple_loss=0.421, pruned_loss=0.1745, over 5678339.29 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3664, pruned_loss=0.1234, over 5689261.62 frames. ], batch size: 99, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:39:55,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:40:25,028 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 09:40:30,373 INFO [train.py:968] (0/2) Epoch 2, batch 37200, giga_loss[loss=0.2736, simple_loss=0.3348, pruned_loss=0.1062, over 29132.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.37, pruned_loss=0.1268, over 5700357.02 frames. ], libri_tot_loss[loss=0.3858, simple_loss=0.4219, pruned_loss=0.1748, over 5683083.26 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3634, pruned_loss=0.1215, over 5702841.15 frames. ], batch size: 128, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:40:45,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 09:40:54,178 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:40:57,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:40:57,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.852e+02 1.034e+03 1.295e+03 1.966e+03 6.723e+03, threshold=2.590e+03, percent-clipped=12.0
+2023-03-01 09:41:10,437 INFO [train.py:968] (0/2) Epoch 2, batch 37250, giga_loss[loss=0.3092, simple_loss=0.3603, pruned_loss=0.129, over 28378.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3677, pruned_loss=0.1258, over 5698422.25 frames. ], libri_tot_loss[loss=0.3867, simple_loss=0.4228, pruned_loss=0.1753, over 5686141.08 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.361, pruned_loss=0.1205, over 5697853.61 frames. ], batch size: 71, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:41:18,105 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:41:43,275 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:41:46,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:41:48,582 INFO [train.py:968] (0/2) Epoch 2, batch 37300, giga_loss[loss=0.3076, simple_loss=0.3627, pruned_loss=0.1263, over 28670.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3668, pruned_loss=0.1254, over 5703479.88 frames. ], libri_tot_loss[loss=0.3884, simple_loss=0.4245, pruned_loss=0.1762, over 5686482.58 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3577, pruned_loss=0.1184, over 5703270.58 frames. ], batch size: 242, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:41:49,011 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:42:12,411 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:42:15,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.330e+02 1.017e+03 1.298e+03 1.771e+03 7.088e+03, threshold=2.596e+03, percent-clipped=10.0
+2023-03-01 09:42:30,195 INFO [train.py:968] (0/2) Epoch 2, batch 37350, libri_loss[loss=0.4568, simple_loss=0.4908, pruned_loss=0.2114, over 29526.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3645, pruned_loss=0.1238, over 5716716.09 frames. ], libri_tot_loss[loss=0.3893, simple_loss=0.4254, pruned_loss=0.1766, over 5690842.78 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3551, pruned_loss=0.1166, over 5713173.20 frames. ], batch size: 89, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:42:50,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1485, 2.7063, 2.8792, 1.2212], device='cuda:0'), covar=tensor([0.0791, 0.0607, 0.1056, 0.2343], device='cuda:0'), in_proj_covar=tensor([0.0751, 0.0546, 0.0780, 0.0554], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 09:43:10,868 INFO [train.py:968] (0/2) Epoch 2, batch 37400, giga_loss[loss=0.2735, simple_loss=0.3409, pruned_loss=0.1031, over 28986.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3612, pruned_loss=0.1217, over 5717571.08 frames. ], libri_tot_loss[loss=0.3899, simple_loss=0.426, pruned_loss=0.1769, over 5692661.58 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3529, pruned_loss=0.1154, over 5713451.25 frames. ], batch size: 213, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:43:39,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.202e+02 9.876e+02 1.247e+03 1.611e+03 4.397e+03, threshold=2.493e+03, percent-clipped=3.0
+2023-03-01 09:43:47,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:43:50,261 INFO [train.py:968] (0/2) Epoch 2, batch 37450, giga_loss[loss=0.2641, simple_loss=0.33, pruned_loss=0.09911, over 28421.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3569, pruned_loss=0.119, over 5709779.72 frames. ], libri_tot_loss[loss=0.3902, simple_loss=0.4262, pruned_loss=0.1771, over 5685162.99 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3496, pruned_loss=0.1134, over 5713841.96 frames. ], batch size: 65, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:44:14,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7464, 1.3296, 3.6211, 3.2092], device='cuda:0'), covar=tensor([0.1762, 0.1934, 0.0383, 0.0574], device='cuda:0'), in_proj_covar=tensor([0.0506, 0.0490, 0.0642, 0.0516], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 09:44:23,937 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-01 09:44:37,084 INFO [train.py:968] (0/2) Epoch 2, batch 37500, libri_loss[loss=0.4376, simple_loss=0.4749, pruned_loss=0.2002, over 29489.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3584, pruned_loss=0.1202, over 5708866.87 frames. ], libri_tot_loss[loss=0.3915, simple_loss=0.4275, pruned_loss=0.1778, over 5687903.87 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3491, pruned_loss=0.1132, over 5710567.84 frames. ], batch size: 85, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:44:43,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2467, 1.9808, 1.3932, 1.4531], device='cuda:0'), covar=tensor([0.0975, 0.0376, 0.0422, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.0264, 0.0177, 0.0180, 0.0316], device='cuda:0'), out_proj_covar=tensor([0.0034, 0.0026, 0.0023, 0.0039], device='cuda:0')
+2023-03-01 09:45:03,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.655e+02 1.116e+03 1.486e+03 2.005e+03 6.330e+03, threshold=2.971e+03, percent-clipped=19.0
+2023-03-01 09:45:17,850 INFO [train.py:968] (0/2) Epoch 2, batch 37550, giga_loss[loss=0.327, simple_loss=0.3791, pruned_loss=0.1375, over 28710.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3598, pruned_loss=0.1212, over 5704182.63 frames. ], libri_tot_loss[loss=0.3925, simple_loss=0.4285, pruned_loss=0.1782, over 5683793.59 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3493, pruned_loss=0.1134, over 5709457.81 frames. ], batch size: 242, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:45:52,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:45:55,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:45:56,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83261.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 09:46:02,966 INFO [train.py:968] (0/2) Epoch 2, batch 37600, giga_loss[loss=0.3364, simple_loss=0.3865, pruned_loss=0.1432, over 28513.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3616, pruned_loss=0.1221, over 5715066.45 frames. ], libri_tot_loss[loss=0.3927, simple_loss=0.4287, pruned_loss=0.1784, over 5684612.02 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3525, pruned_loss=0.1153, over 5718752.66 frames. ], batch size: 71, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:46:20,244 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:46:33,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.997e+02 1.031e+03 1.355e+03 1.778e+03 3.241e+03, threshold=2.710e+03, percent-clipped=1.0
+2023-03-01 09:46:48,141 INFO [train.py:968] (0/2) Epoch 2, batch 37650, libri_loss[loss=0.3643, simple_loss=0.4054, pruned_loss=0.1616, over 29590.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3692, pruned_loss=0.1277, over 5707300.50 frames. ], libri_tot_loss[loss=0.393, simple_loss=0.429, pruned_loss=0.1785, over 5681949.52 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3596, pruned_loss=0.1204, over 5713169.13 frames. ], batch size: 74, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:47:06,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:47:26,762 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=83356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:47:39,555 INFO [train.py:968] (0/2) Epoch 2, batch 37700, giga_loss[loss=0.3783, simple_loss=0.4201, pruned_loss=0.1683, over 28911.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3809, pruned_loss=0.1365, over 5695532.95 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4298, pruned_loss=0.179, over 5683373.85 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3715, pruned_loss=0.1294, over 5699101.02 frames. ], batch size: 145, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:47:55,676 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.89 vs. limit=5.0
+2023-03-01 09:48:09,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.438e+02 1.304e+03 1.754e+03 2.474e+03 6.484e+03, threshold=3.507e+03, percent-clipped=15.0
+2023-03-01 09:48:09,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83404.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 09:48:13,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83407.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 09:48:27,331 INFO [train.py:968] (0/2) Epoch 2, batch 37750, giga_loss[loss=0.3619, simple_loss=0.3948, pruned_loss=0.1645, over 23288.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3893, pruned_loss=0.1431, over 5663512.96 frames. ], libri_tot_loss[loss=0.3948, simple_loss=0.4304, pruned_loss=0.1797, over 5661228.34 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.38, pruned_loss=0.1356, over 5686902.47 frames. ], batch size: 705, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:48:46,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83436.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 09:49:14,746 INFO [train.py:968] (0/2) Epoch 2, batch 37800, giga_loss[loss=0.314, simple_loss=0.3809, pruned_loss=0.1236, over 28927.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3932, pruned_loss=0.1439, over 5667121.71 frames. ], libri_tot_loss[loss=0.3949, simple_loss=0.4305, pruned_loss=0.1797, over 5660453.20 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3857, pruned_loss=0.1379, over 5686143.71 frames. ], batch size: 112, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:49:27,599 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:49:31,164 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:49:49,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-01 09:49:50,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.005e+02 1.189e+03 1.478e+03 1.967e+03 4.290e+03, threshold=2.955e+03, percent-clipped=3.0
+2023-03-01 09:49:58,234 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:50:05,679 INFO [train.py:968] (0/2) Epoch 2, batch 37850, giga_loss[loss=0.5615, simple_loss=0.5306, pruned_loss=0.2962, over 26581.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.3994, pruned_loss=0.1471, over 5667639.67 frames. ], libri_tot_loss[loss=0.3951, simple_loss=0.4307, pruned_loss=0.1798, over 5660489.53 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3927, pruned_loss=0.1418, over 5682866.19 frames. ], batch size: 555, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:50:50,453 INFO [train.py:968] (0/2) Epoch 2, batch 37900, libri_loss[loss=0.3689, simple_loss=0.4068, pruned_loss=0.1655, over 29566.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4026, pruned_loss=0.1498, over 5671314.70 frames. ], libri_tot_loss[loss=0.3953, simple_loss=0.4306, pruned_loss=0.18, over 5657355.78 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3965, pruned_loss=0.1444, over 5685907.50 frames. ], batch size: 76, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:51:08,189 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=83589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:51:21,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.302e+02 1.128e+03 1.593e+03 2.048e+03 5.653e+03, threshold=3.186e+03, percent-clipped=11.0
+2023-03-01 09:51:32,085 INFO [train.py:968] (0/2) Epoch 2, batch 37950, giga_loss[loss=0.284, simple_loss=0.3526, pruned_loss=0.1077, over 28638.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3961, pruned_loss=0.1451, over 5677964.14 frames. ], libri_tot_loss[loss=0.3952, simple_loss=0.4304, pruned_loss=0.1799, over 5662594.22 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3906, pruned_loss=0.14, over 5685116.47 frames. ], batch size: 242, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:52:16,883 INFO [train.py:968] (0/2) Epoch 2, batch 38000, giga_loss[loss=0.2458, simple_loss=0.336, pruned_loss=0.07784, over 28963.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3926, pruned_loss=0.1415, over 5667669.35 frames. ], libri_tot_loss[loss=0.3959, simple_loss=0.4309, pruned_loss=0.1804, over 5644502.91 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3873, pruned_loss=0.1365, over 5689894.57 frames. ], batch size: 164, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:52:23,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=83677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:52:33,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3316, 1.4672, 1.0833, 0.9525], device='cuda:0'), covar=tensor([0.0613, 0.0420, 0.0416, 0.0627], device='cuda:0'), in_proj_covar=tensor([0.1084, 0.0763, 0.0869, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 09:52:48,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.046e+02 1.269e+03 1.570e+03 1.961e+03 6.896e+03, threshold=3.140e+03, percent-clipped=9.0
+2023-03-01 09:52:59,452 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-01 09:53:01,261 INFO [train.py:968] (0/2) Epoch 2, batch 38050, giga_loss[loss=0.336, simple_loss=0.399, pruned_loss=0.1365, over 28957.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3922, pruned_loss=0.1403, over 5677533.29 frames. ], libri_tot_loss[loss=0.3961, simple_loss=0.4312, pruned_loss=0.1805, over 5646312.47 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.387, pruned_loss=0.1355, over 5694239.44 frames. ], batch size: 145, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:53:12,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:53:45,692 INFO [train.py:968] (0/2) Epoch 2, batch 38100, giga_loss[loss=0.3309, simple_loss=0.3965, pruned_loss=0.1326, over 29031.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3952, pruned_loss=0.1422, over 5683654.59 frames. ], libri_tot_loss[loss=0.3962, simple_loss=0.4312, pruned_loss=0.1806, over 5650248.85 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3906, pruned_loss=0.1379, over 5693883.15 frames. ], batch size: 128, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:54:18,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.463e+02 1.200e+03 1.470e+03 2.321e+03 5.187e+03, threshold=2.940e+03, percent-clipped=10.0
+2023-03-01 09:54:30,974 INFO [train.py:968] (0/2) Epoch 2, batch 38150, giga_loss[loss=0.346, simple_loss=0.4027, pruned_loss=0.1446, over 28876.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3984, pruned_loss=0.1443, over 5693599.49 frames. ], libri_tot_loss[loss=0.3961, simple_loss=0.4312, pruned_loss=0.1805, over 5657135.49 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3937, pruned_loss=0.1398, over 5696536.49 frames. ], batch size: 119, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:55:10,055 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.45 vs. limit=5.0
+2023-03-01 09:55:17,368 INFO [train.py:968] (0/2) Epoch 2, batch 38200, giga_loss[loss=0.3905, simple_loss=0.4353, pruned_loss=0.1728, over 28868.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.4013, pruned_loss=0.1472, over 5685606.60 frames. ], libri_tot_loss[loss=0.3963, simple_loss=0.4313, pruned_loss=0.1806, over 5652132.72 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3963, pruned_loss=0.1422, over 5693927.56 frames. ], batch size: 174, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:55:21,834 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:55:23,932 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:55:50,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.215e+02 1.223e+03 1.651e+03 2.668e+03 8.655e+03, threshold=3.303e+03, percent-clipped=18.0
+2023-03-01 09:55:52,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:56:03,953 INFO [train.py:968] (0/2) Epoch 2, batch 38250, giga_loss[loss=0.3175, simple_loss=0.3756, pruned_loss=0.1297, over 28602.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.4018, pruned_loss=0.1476, over 5687812.59 frames. ], libri_tot_loss[loss=0.3968, simple_loss=0.4317, pruned_loss=0.1809, over 5656559.39 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3969, pruned_loss=0.1428, over 5691236.38 frames. ], batch size: 78, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:56:28,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9125, 1.7298, 1.2566, 1.5015], device='cuda:0'), covar=tensor([0.0556, 0.0639, 0.0955, 0.0912], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0488, 0.0532, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 09:56:43,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:56:48,743 INFO [train.py:968] (0/2) Epoch 2, batch 38300, giga_loss[loss=0.3437, simple_loss=0.404, pruned_loss=0.1417, over 28868.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.4026, pruned_loss=0.1483, over 5696671.90 frames. ], libri_tot_loss[loss=0.3971, simple_loss=0.4322, pruned_loss=0.181, over 5662214.62 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3977, pruned_loss=0.1436, over 5695253.23 frames. ], batch size: 145, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:57:11,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6387, 2.1395, 1.7929, 1.9569], device='cuda:0'), covar=tensor([0.0479, 0.0667, 0.0834, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0488, 0.0533, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 09:57:14,903 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-84000.pt
+2023-03-01 09:57:19,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.135e+02 1.247e+03 1.451e+03 2.005e+03 3.678e+03, threshold=2.902e+03, percent-clipped=2.0
+2023-03-01 09:57:33,821 INFO [train.py:968] (0/2) Epoch 2, batch 38350, giga_loss[loss=0.3121, simple_loss=0.3842, pruned_loss=0.12, over 28761.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4028, pruned_loss=0.1484, over 5696432.86 frames. ], libri_tot_loss[loss=0.3968, simple_loss=0.432, pruned_loss=0.1808, over 5668319.39 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3984, pruned_loss=0.1441, over 5690375.52 frames. ], batch size: 99, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 09:58:03,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:58:10,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8584, 2.1386, 1.9059, 1.8899], device='cuda:0'), covar=tensor([0.1342, 0.1611, 0.1105, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0802, 0.0710, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 09:58:14,698 INFO [train.py:968] (0/2) Epoch 2, batch 38400, giga_loss[loss=0.3278, simple_loss=0.398, pruned_loss=0.1288, over 28991.00 frames. ], tot_loss[loss=0.3474, simple_loss=0.402, pruned_loss=0.1464, over 5704287.02 frames. ], libri_tot_loss[loss=0.3964, simple_loss=0.4317, pruned_loss=0.1805, over 5671724.64 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3983, pruned_loss=0.1428, over 5696728.55 frames. ], batch size: 186, lr: 1.19e-02, grad_scale: 8.0
+2023-03-01 09:58:42,773 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-01 09:58:47,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.065e+02 1.159e+03 1.499e+03 1.915e+03 7.869e+03, threshold=2.997e+03, percent-clipped=15.0
+2023-03-01 09:58:47,896 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:58:50,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:58:56,814 INFO [train.py:968] (0/2) Epoch 2, batch 38450, giga_loss[loss=0.3091, simple_loss=0.3754, pruned_loss=0.1214, over 28535.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.4024, pruned_loss=0.1456, over 5685431.45 frames. ], libri_tot_loss[loss=0.3971, simple_loss=0.4321, pruned_loss=0.1811, over 5651564.52 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.398, pruned_loss=0.1409, over 5698960.75 frames. ], batch size: 85, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 09:59:11,810 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:59:21,333 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:59:22,418 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-01 09:59:35,722 INFO [train.py:968] (0/2) Epoch 2, batch 38500, giga_loss[loss=0.3361, simple_loss=0.3916, pruned_loss=0.1403, over 28920.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.4026, pruned_loss=0.1455, over 5694809.48 frames. ], libri_tot_loss[loss=0.3961, simple_loss=0.4314, pruned_loss=0.1804, over 5659432.63 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3987, pruned_loss=0.1411, over 5700168.75 frames. ], batch size: 174, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 09:59:59,668 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:00:01,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:00:08,669 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.593e+02 1.291e+03 1.626e+03 2.057e+03 6.733e+03, threshold=3.252e+03, percent-clipped=9.0
+2023-03-01 10:00:18,509 INFO [train.py:968] (0/2) Epoch 2, batch 38550, giga_loss[loss=0.3398, simple_loss=0.3923, pruned_loss=0.1437, over 28834.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.4, pruned_loss=0.1443, over 5698084.31 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4316, pruned_loss=0.1807, over 5663925.04 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.396, pruned_loss=0.1397, over 5699174.07 frames. ], batch size: 99, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 10:00:26,355 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:00:37,737 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-01 10:00:59,558 INFO [train.py:968] (0/2) Epoch 2, batch 38600, giga_loss[loss=0.3173, simple_loss=0.3812, pruned_loss=0.1267, over 28462.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3977, pruned_loss=0.1428, over 5682864.15 frames. ], libri_tot_loss[loss=0.3971, simple_loss=0.4321, pruned_loss=0.181, over 5648835.48 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3932, pruned_loss=0.1379, over 5698261.82 frames. ], batch size: 85, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 10:01:29,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.897e+02 1.218e+03 1.611e+03 2.288e+03 7.758e+03, threshold=3.222e+03, percent-clipped=6.0
+2023-03-01 10:01:38,845 INFO [train.py:968] (0/2) Epoch 2, batch 38650, giga_loss[loss=0.3777, simple_loss=0.4207, pruned_loss=0.1673, over 28683.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3968, pruned_loss=0.1425, over 5697655.05 frames. ], libri_tot_loss[loss=0.3961, simple_loss=0.4314, pruned_loss=0.1803, over 5658374.18 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3924, pruned_loss=0.1376, over 5702747.82 frames. ], batch size: 99, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 10:01:59,846 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:02:08,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3313, 2.0459, 1.2944, 1.2157], device='cuda:0'), covar=tensor([0.0729, 0.0499, 0.0763, 0.1168], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0439, 0.0318, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 10:02:24,363 INFO [train.py:968] (0/2) Epoch 2, batch 38700, giga_loss[loss=0.3346, simple_loss=0.3923, pruned_loss=0.1384, over 28854.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3976, pruned_loss=0.1432, over 5698410.37 frames. ], libri_tot_loss[loss=0.3964, simple_loss=0.4318, pruned_loss=0.1805, over 5662334.06 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.393, pruned_loss=0.1382, over 5699723.00 frames. ], batch size: 199, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 10:02:31,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:02:49,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8071, 2.2252, 1.9195, 1.8859], device='cuda:0'), covar=tensor([0.1427, 0.1462, 0.1118, 0.0772], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0806, 0.0719, 0.0613], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 10:02:51,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5023, 2.1732, 1.5910, 0.5761], device='cuda:0'), covar=tensor([0.1576, 0.0799, 0.1376, 0.1854], device='cuda:0'), in_proj_covar=tensor([0.1201, 0.1126, 0.1199, 0.1023], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 10:02:58,878 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.346e+02 1.056e+03 1.272e+03 1.872e+03 7.218e+03, threshold=2.543e+03, percent-clipped=9.0
+2023-03-01 10:03:07,236 INFO [train.py:968] (0/2) Epoch 2, batch 38750, giga_loss[loss=0.3064, simple_loss=0.3736, pruned_loss=0.1196, over 28601.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3967, pruned_loss=0.1417, over 5694999.76 frames. ], libri_tot_loss[loss=0.3968, simple_loss=0.432, pruned_loss=0.1808, over 5653398.39 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3928, pruned_loss=0.1376, over 5703230.47 frames. ], batch size: 92, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 10:03:24,555 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7504, 1.6786, 1.5128, 1.0749], device='cuda:0'), covar=tensor([0.0509, 0.0377, 0.0305, 0.0476], device='cuda:0'), in_proj_covar=tensor([0.1083, 0.0765, 0.0852, 0.0924], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 10:03:48,831 INFO [train.py:968] (0/2) Epoch 2, batch 38800, giga_loss[loss=0.3493, simple_loss=0.4041, pruned_loss=0.1473, over 28684.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3965, pruned_loss=0.1411, over 5702221.81 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4318, pruned_loss=0.1807, over 5659997.40 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3928, pruned_loss=0.137, over 5703866.73 frames. ], batch size: 99, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:04:05,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:04:19,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.238e+02 1.061e+03 1.254e+03 1.670e+03 3.829e+03, threshold=2.508e+03, percent-clipped=10.0
+2023-03-01 10:04:20,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2436, 2.0570, 1.1720, 1.0316], device='cuda:0'), covar=tensor([0.1259, 0.0739, 0.1274, 0.2048], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0439, 0.0322, 0.0400], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 10:04:30,273 INFO [train.py:968] (0/2) Epoch 2, batch 38850, giga_loss[loss=0.4034, simple_loss=0.4274, pruned_loss=0.1897, over 26593.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3961, pruned_loss=0.14, over 5710271.05 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.4322, pruned_loss=0.1809, over 5663444.92 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3922, pruned_loss=0.1359, over 5709004.01 frames. ], batch size: 555, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:04:33,960 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-01 10:04:37,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:05:08,311 INFO [train.py:968] (0/2) Epoch 2, batch 38900, giga_loss[loss=0.3475, simple_loss=0.4046, pruned_loss=0.1452, over 28693.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3963, pruned_loss=0.1407, over 5699230.08 frames. ], libri_tot_loss[loss=0.396, simple_loss=0.4314, pruned_loss=0.1803, over 5661586.87 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3926, pruned_loss=0.1365, over 5701585.22 frames. ], batch size: 242, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:05:30,108 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:05:40,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.857e+02 1.174e+03 1.535e+03 2.178e+03 9.012e+03, threshold=3.069e+03, percent-clipped=16.0
+2023-03-01 10:05:42,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3517, 1.3747, 1.2373, 1.4249], device='cuda:0'), covar=tensor([0.1780, 0.1879, 0.1539, 0.1879], device='cuda:0'), in_proj_covar=tensor([0.0978, 0.0809, 0.0885, 0.0923], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 10:05:43,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9508, 1.6787, 1.2981, 1.4394], device='cuda:0'), covar=tensor([0.0658, 0.0777, 0.1060, 0.1016], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0493, 0.0539, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 10:05:48,954 INFO [train.py:968] (0/2) Epoch 2, batch 38950, giga_loss[loss=0.3036, simple_loss=0.3712, pruned_loss=0.1179, over 28952.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3944, pruned_loss=0.1402, over 5700768.27 frames. ], libri_tot_loss[loss=0.3952, simple_loss=0.4306, pruned_loss=0.1799, over 5661647.47 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3908, pruned_loss=0.1357, over 5703777.48 frames. ], batch size: 227, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:05:58,965 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:06:04,540 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-01 10:06:32,406 INFO [train.py:968] (0/2) Epoch 2, batch 39000, giga_loss[loss=0.2782, simple_loss=0.3508, pruned_loss=0.1028, over 28960.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3905, pruned_loss=0.1379, over 5705413.84 frames. ], libri_tot_loss[loss=0.3949, simple_loss=0.4304, pruned_loss=0.1797, over 5666358.73 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3869, pruned_loss=0.1335, over 5704449.41 frames. ], batch size: 213, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:06:32,410 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 10:06:41,495 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.2687, simple_loss=0.3654, pruned_loss=0.08597, over 944034.00 frames. 
+2023-03-01 10:06:41,496 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19537MB
+2023-03-01 10:06:43,069 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:06:44,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:06:53,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0000, 1.8364, 1.3795, 1.4800], device='cuda:0'), covar=tensor([0.0571, 0.0646, 0.1007, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0494, 0.0537, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 10:06:57,598 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-01 10:07:08,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:07:11,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.887e+02 1.165e+03 1.591e+03 2.198e+03 8.014e+03, threshold=3.182e+03, percent-clipped=8.0
+2023-03-01 10:07:18,446 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:07:19,610 INFO [train.py:968] (0/2) Epoch 2, batch 39050, giga_loss[loss=0.3073, simple_loss=0.3768, pruned_loss=0.1189, over 28896.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3886, pruned_loss=0.1368, over 5712481.57 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.4302, pruned_loss=0.1794, over 5672952.50 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3847, pruned_loss=0.1324, over 5706834.09 frames. ], batch size: 174, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:07:22,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8845, 2.1181, 0.9779, 1.6809], device='cuda:0'), covar=tensor([0.0807, 0.0697, 0.1812, 0.1156], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0492, 0.0537, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 10:07:28,297 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:07:48,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:08:01,994 INFO [train.py:968] (0/2) Epoch 2, batch 39100, giga_loss[loss=0.3181, simple_loss=0.3796, pruned_loss=0.1283, over 28698.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3896, pruned_loss=0.1389, over 5713153.47 frames. ], libri_tot_loss[loss=0.3947, simple_loss=0.4303, pruned_loss=0.1795, over 5679861.39 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3853, pruned_loss=0.1341, over 5703420.83 frames. ], batch size: 119, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:08:23,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6330, 2.2018, 2.3334, 2.1883], device='cuda:0'), covar=tensor([0.0451, 0.0903, 0.0654, 0.0673], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0801, 0.0637, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 10:08:31,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.265e+02 1.054e+03 1.374e+03 1.952e+03 1.166e+04, threshold=2.749e+03, percent-clipped=9.0
+2023-03-01 10:08:41,615 INFO [train.py:968] (0/2) Epoch 2, batch 39150, giga_loss[loss=0.3226, simple_loss=0.3756, pruned_loss=0.1348, over 29004.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3883, pruned_loss=0.1384, over 5706649.88 frames. ], libri_tot_loss[loss=0.3952, simple_loss=0.4308, pruned_loss=0.1798, over 5671476.84 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3835, pruned_loss=0.1334, over 5707557.26 frames. ], batch size: 136, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:09:15,907 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:18,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:20,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:23,326 INFO [train.py:968] (0/2) Epoch 2, batch 39200, libri_loss[loss=0.4108, simple_loss=0.4366, pruned_loss=0.1925, over 29548.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3855, pruned_loss=0.1372, over 5707955.94 frames. ], libri_tot_loss[loss=0.394, simple_loss=0.4298, pruned_loss=0.1791, over 5678115.37 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3806, pruned_loss=0.132, over 5704018.71 frames. ], batch size: 79, lr: 1.19e-02, grad_scale: 8.0
+2023-03-01 10:09:40,647 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:43,510 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:45,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:53,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.163e+02 1.087e+03 1.495e+03 1.957e+03 1.192e+04, threshold=2.990e+03, percent-clipped=12.0
+2023-03-01 10:10:00,927 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:10:01,419 INFO [train.py:968] (0/2) Epoch 2, batch 39250, giga_loss[loss=0.267, simple_loss=0.3367, pruned_loss=0.09869, over 28316.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3832, pruned_loss=0.1361, over 5703325.57 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4299, pruned_loss=0.179, over 5675158.59 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3781, pruned_loss=0.1309, over 5703041.25 frames. ], batch size: 77, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:10:08,765 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:10:42,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:10:43,578 INFO [train.py:968] (0/2) Epoch 2, batch 39300, libri_loss[loss=0.3869, simple_loss=0.4375, pruned_loss=0.1682, over 29226.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3811, pruned_loss=0.1348, over 5710482.71 frames. ], libri_tot_loss[loss=0.3936, simple_loss=0.4298, pruned_loss=0.1787, over 5679083.39 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3756, pruned_loss=0.1295, over 5707874.94 frames. ], batch size: 97, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:11:14,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:11:17,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.532e+02 9.949e+02 1.252e+03 1.580e+03 3.984e+03, threshold=2.504e+03, percent-clipped=1.0
+2023-03-01 10:11:17,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:11:19,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:11:26,651 INFO [train.py:968] (0/2) Epoch 2, batch 39350, giga_loss[loss=0.3353, simple_loss=0.3812, pruned_loss=0.1447, over 28485.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3808, pruned_loss=0.1349, over 5712628.93 frames. ], libri_tot_loss[loss=0.393, simple_loss=0.4294, pruned_loss=0.1783, over 5684558.15 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3756, pruned_loss=0.13, over 5706338.04 frames. ], batch size: 85, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:11:32,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:11:45,033 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:12:13,333 INFO [train.py:968] (0/2) Epoch 2, batch 39400, giga_loss[loss=0.3325, simple_loss=0.3958, pruned_loss=0.1346, over 28841.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3834, pruned_loss=0.136, over 5716172.02 frames. ], libri_tot_loss[loss=0.3931, simple_loss=0.4295, pruned_loss=0.1784, over 5688747.51 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.378, pruned_loss=0.1311, over 5707986.23 frames. ], batch size: 186, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:12:14,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2990, 1.3512, 1.2142, 0.7812], device='cuda:0'), covar=tensor([0.0692, 0.0460, 0.0340, 0.0563], device='cuda:0'), in_proj_covar=tensor([0.1146, 0.0812, 0.0890, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 10:12:43,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:12:45,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.893e+02 1.106e+03 1.353e+03 1.880e+03 8.265e+03, threshold=2.707e+03, percent-clipped=12.0
+2023-03-01 10:12:46,079 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-01 10:12:47,211 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:12:50,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:12:56,743 INFO [train.py:968] (0/2) Epoch 2, batch 39450, giga_loss[loss=0.4196, simple_loss=0.4389, pruned_loss=0.2002, over 26723.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3861, pruned_loss=0.1372, over 5713983.74 frames. ], libri_tot_loss[loss=0.3922, simple_loss=0.4286, pruned_loss=0.1779, over 5690255.06 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.381, pruned_loss=0.1321, over 5707004.93 frames. ], batch size: 555, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:13:17,233 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:13:23,299 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:13:25,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:13:41,714 INFO [train.py:968] (0/2) Epoch 2, batch 39500, giga_loss[loss=0.3522, simple_loss=0.4112, pruned_loss=0.1466, over 28632.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3868, pruned_loss=0.1364, over 5710732.55 frames. ], libri_tot_loss[loss=0.3918, simple_loss=0.4283, pruned_loss=0.1777, over 5694330.45 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3822, pruned_loss=0.1318, over 5701988.53 frames. ], batch size: 307, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:13:52,901 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:14:03,780 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:14:15,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.750e+02 9.533e+02 1.202e+03 1.583e+03 5.373e+03, threshold=2.403e+03, percent-clipped=9.0
+2023-03-01 10:14:18,330 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-01 10:14:22,990 INFO [train.py:968] (0/2) Epoch 2, batch 39550, giga_loss[loss=0.2976, simple_loss=0.3721, pruned_loss=0.1115, over 28865.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3866, pruned_loss=0.1352, over 5707397.62 frames. ], libri_tot_loss[loss=0.3906, simple_loss=0.4274, pruned_loss=0.1769, over 5699268.07 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3817, pruned_loss=0.1302, over 5696176.45 frames. ], batch size: 227, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:14:48,818 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:14:51,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:15:06,372 INFO [train.py:968] (0/2) Epoch 2, batch 39600, giga_loss[loss=0.2998, simple_loss=0.3712, pruned_loss=0.1142, over 28900.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3856, pruned_loss=0.1338, over 5700282.44 frames. ], libri_tot_loss[loss=0.3909, simple_loss=0.4277, pruned_loss=0.1771, over 5700203.94 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3807, pruned_loss=0.1288, over 5690467.21 frames. ], batch size: 199, lr: 1.19e-02, grad_scale: 8.0
+2023-03-01 10:15:14,027 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:15:27,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:15:39,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.539e+02 1.126e+03 1.399e+03 1.794e+03 6.572e+03, threshold=2.799e+03, percent-clipped=12.0
+2023-03-01 10:15:51,951 INFO [train.py:968] (0/2) Epoch 2, batch 39650, giga_loss[loss=0.3296, simple_loss=0.3806, pruned_loss=0.1393, over 23966.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3866, pruned_loss=0.1349, over 5702129.43 frames. ], libri_tot_loss[loss=0.3913, simple_loss=0.4279, pruned_loss=0.1773, over 5704556.00 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3817, pruned_loss=0.13, over 5690631.28 frames. ], batch size: 705, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:16:06,918 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4436, 1.3731, 1.2323, 1.6097], device='cuda:0'), covar=tensor([0.1848, 0.1986, 0.1758, 0.2027], device='cuda:0'), in_proj_covar=tensor([0.0985, 0.0807, 0.0887, 0.0931], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 10:16:12,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:16:21,732 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-01 10:16:31,556 INFO [train.py:968] (0/2) Epoch 2, batch 39700, giga_loss[loss=0.3622, simple_loss=0.4101, pruned_loss=0.1571, over 28843.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3874, pruned_loss=0.1357, over 5710369.21 frames. ], libri_tot_loss[loss=0.3913, simple_loss=0.428, pruned_loss=0.1773, over 5709045.37 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3825, pruned_loss=0.1308, over 5697020.69 frames. ], batch size: 145, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:16:42,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3023, 3.9683, 2.2099, 1.8737], device='cuda:0'), covar=tensor([0.0714, 0.0357, 0.0759, 0.1244], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0442, 0.0324, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 10:16:55,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3777, 1.7898, 1.3735, 0.5501], device='cuda:0'), covar=tensor([0.1178, 0.0823, 0.1522, 0.1590], device='cuda:0'), in_proj_covar=tensor([0.1223, 0.1147, 0.1216, 0.1042], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 10:17:01,133 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:17:07,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.968e+02 1.257e+03 1.667e+03 2.337e+03 7.114e+03, threshold=3.334e+03, percent-clipped=14.0
+2023-03-01 10:17:16,546 INFO [train.py:968] (0/2) Epoch 2, batch 39750, giga_loss[loss=0.3241, simple_loss=0.3927, pruned_loss=0.1278, over 28870.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.39, pruned_loss=0.1377, over 5697701.26 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.4277, pruned_loss=0.177, over 5703655.74 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3853, pruned_loss=0.133, over 5691869.91 frames. ], batch size: 186, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:17:21,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9474, 1.5220, 4.1215, 3.1166], device='cuda:0'), covar=tensor([0.1473, 0.1648, 0.0288, 0.0491], device='cuda:0'), in_proj_covar=tensor([0.0505, 0.0489, 0.0657, 0.0523], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 10:17:30,567 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:17:33,921 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:17:41,448 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 10:17:59,893 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:18:00,272 INFO [train.py:968] (0/2) Epoch 2, batch 39800, giga_loss[loss=0.3622, simple_loss=0.4164, pruned_loss=0.154, over 28699.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.392, pruned_loss=0.1389, over 5702854.34 frames. ], libri_tot_loss[loss=0.3909, simple_loss=0.4278, pruned_loss=0.1771, over 5704893.91 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3879, pruned_loss=0.1348, over 5697246.57 frames. ], batch size: 262, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:18:25,006 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:18:31,409 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:18:31,915 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.649e+02 1.179e+03 1.539e+03 2.340e+03 8.889e+03, threshold=3.078e+03, percent-clipped=10.0
+2023-03-01 10:18:38,930 INFO [train.py:968] (0/2) Epoch 2, batch 39850, giga_loss[loss=0.3269, simple_loss=0.3901, pruned_loss=0.1318, over 28885.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.394, pruned_loss=0.1398, over 5713507.51 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.4276, pruned_loss=0.177, over 5706824.10 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3897, pruned_loss=0.1353, over 5707171.77 frames. ], batch size: 199, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:18:39,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4031, 1.5106, 1.1386, 1.3831], device='cuda:0'), covar=tensor([0.0868, 0.0382, 0.0435, 0.1082], device='cuda:0'), in_proj_covar=tensor([0.0258, 0.0174, 0.0178, 0.0317], device='cuda:0'), out_proj_covar=tensor([0.0035, 0.0026, 0.0024, 0.0039], device='cuda:0')
+2023-03-01 10:18:59,047 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:19:02,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:19:19,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:19:19,781 INFO [train.py:968] (0/2) Epoch 2, batch 39900, giga_loss[loss=0.2926, simple_loss=0.3594, pruned_loss=0.1129, over 28806.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.395, pruned_loss=0.14, over 5718273.51 frames. ], libri_tot_loss[loss=0.3915, simple_loss=0.4283, pruned_loss=0.1774, over 5711860.58 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.39, pruned_loss=0.1349, over 5709061.32 frames. ], batch size: 99, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:19:26,636 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:19:53,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.091e+02 1.305e+03 1.730e+03 2.146e+03 5.578e+03, threshold=3.461e+03, percent-clipped=7.0
+2023-03-01 10:20:02,646 INFO [train.py:968] (0/2) Epoch 2, batch 39950, giga_loss[loss=0.3685, simple_loss=0.4205, pruned_loss=0.1583, over 28267.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3955, pruned_loss=0.1405, over 5713756.51 frames. ], libri_tot_loss[loss=0.3914, simple_loss=0.4284, pruned_loss=0.1773, over 5710429.49 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3909, pruned_loss=0.136, over 5707971.54 frames. ], batch size: 368, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:20:21,797 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-01 10:20:28,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7417, 1.5899, 1.5447, 1.5969], device='cuda:0'), covar=tensor([0.1009, 0.1879, 0.1483, 0.1406], device='cuda:0'), in_proj_covar=tensor([0.0463, 0.0810, 0.0648, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 10:20:40,516 INFO [train.py:968] (0/2) Epoch 2, batch 40000, giga_loss[loss=0.334, simple_loss=0.3868, pruned_loss=0.1406, over 28973.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3959, pruned_loss=0.141, over 5719560.50 frames. ], libri_tot_loss[loss=0.3919, simple_loss=0.4287, pruned_loss=0.1776, over 5714866.90 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3912, pruned_loss=0.1362, over 5711214.40 frames. ], batch size: 106, lr: 1.18e-02, grad_scale: 8.0
+2023-03-01 10:21:04,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5629, 2.0233, 1.6887, 1.7369], device='cuda:0'), covar=tensor([0.1409, 0.1720, 0.1238, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0799, 0.0713, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 10:21:14,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.744e+02 1.101e+03 1.419e+03 1.850e+03 4.506e+03, threshold=2.839e+03, percent-clipped=5.0
+2023-03-01 10:21:15,712 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:21:17,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:21:20,482 INFO [train.py:968] (0/2) Epoch 2, batch 40050, giga_loss[loss=0.374, simple_loss=0.4041, pruned_loss=0.1719, over 28707.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3935, pruned_loss=0.1394, over 5721469.30 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.4294, pruned_loss=0.1782, over 5716558.04 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3884, pruned_loss=0.1343, over 5713306.18 frames. ], batch size: 99, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:21:20,794 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:21:41,709 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:22:03,044 INFO [train.py:968] (0/2) Epoch 2, batch 40100, giga_loss[loss=0.3105, simple_loss=0.3732, pruned_loss=0.1239, over 28287.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3901, pruned_loss=0.1376, over 5710695.72 frames. ], libri_tot_loss[loss=0.3926, simple_loss=0.4292, pruned_loss=0.178, over 5711162.24 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3852, pruned_loss=0.1327, over 5709859.99 frames. ], batch size: 368, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:22:39,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.049e+02 1.065e+03 1.416e+03 1.955e+03 6.496e+03, threshold=2.832e+03, percent-clipped=7.0
+2023-03-01 10:22:44,791 INFO [train.py:968] (0/2) Epoch 2, batch 40150, libri_loss[loss=0.3747, simple_loss=0.4236, pruned_loss=0.1629, over 29349.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3868, pruned_loss=0.1353, over 5705601.80 frames. ], libri_tot_loss[loss=0.3925, simple_loss=0.4292, pruned_loss=0.1779, over 5702364.34 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.382, pruned_loss=0.1305, over 5712549.54 frames. ], batch size: 92, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:22:46,307 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:19,302 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:22,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:24,414 INFO [train.py:968] (0/2) Epoch 2, batch 40200, giga_loss[loss=0.2981, simple_loss=0.3759, pruned_loss=0.1102, over 28998.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3871, pruned_loss=0.1346, over 5700934.34 frames. ], libri_tot_loss[loss=0.3921, simple_loss=0.4287, pruned_loss=0.1778, over 5696388.75 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3828, pruned_loss=0.13, over 5711430.74 frames. ], batch size: 136, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:23:30,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:31,502 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:34,759 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-01 10:23:37,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:49,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:24:05,942 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.132e+02 1.161e+03 1.445e+03 1.818e+03 5.782e+03, threshold=2.889e+03, percent-clipped=4.0
+2023-03-01 10:24:12,422 INFO [train.py:968] (0/2) Epoch 2, batch 40250, giga_loss[loss=0.3081, simple_loss=0.3816, pruned_loss=0.1172, over 28819.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3897, pruned_loss=0.1355, over 5697434.72 frames. ], libri_tot_loss[loss=0.3927, simple_loss=0.4291, pruned_loss=0.1781, over 5698165.03 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3853, pruned_loss=0.1308, over 5704248.33 frames. ], batch size: 199, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:24:15,775 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:24:55,690 INFO [train.py:968] (0/2) Epoch 2, batch 40300, giga_loss[loss=0.3181, simple_loss=0.3815, pruned_loss=0.1274, over 29009.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3913, pruned_loss=0.1372, over 5695017.90 frames. ], libri_tot_loss[loss=0.3932, simple_loss=0.4295, pruned_loss=0.1784, over 5692494.41 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3867, pruned_loss=0.1325, over 5706034.15 frames. ], batch size: 164, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:25:04,343 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-01 10:25:22,019 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-86000.pt
+2023-03-01 10:25:31,027 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.898e+02 1.183e+03 1.435e+03 1.991e+03 4.818e+03, threshold=2.871e+03, percent-clipped=6.0
+2023-03-01 10:25:36,977 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:25:37,055 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:25:37,441 INFO [train.py:968] (0/2) Epoch 2, batch 40350, giga_loss[loss=0.2909, simple_loss=0.3522, pruned_loss=0.1148, over 28779.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3893, pruned_loss=0.1373, over 5693748.61 frames. ], libri_tot_loss[loss=0.3934, simple_loss=0.4296, pruned_loss=0.1786, over 5685482.70 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3846, pruned_loss=0.1325, over 5709509.83 frames. ], batch size: 119, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:25:38,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:25:43,086 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:25:46,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:26:02,582 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:26:12,039 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:26:21,704 INFO [train.py:968] (0/2) Epoch 2, batch 40400, giga_loss[loss=0.2895, simple_loss=0.3473, pruned_loss=0.1158, over 28425.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3861, pruned_loss=0.1367, over 5702492.53 frames. ], libri_tot_loss[loss=0.3935, simple_loss=0.4297, pruned_loss=0.1787, over 5684375.22 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3821, pruned_loss=0.1325, over 5715876.26 frames. ], batch size: 65, lr: 1.18e-02, grad_scale: 8.0
+2023-03-01 10:26:54,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.517e+02 1.096e+03 1.476e+03 2.025e+03 5.800e+03, threshold=2.952e+03, percent-clipped=12.0
+2023-03-01 10:27:00,873 INFO [train.py:968] (0/2) Epoch 2, batch 40450, giga_loss[loss=0.2764, simple_loss=0.3403, pruned_loss=0.1063, over 28387.00 frames. ], tot_loss[loss=0.33, simple_loss=0.385, pruned_loss=0.1375, over 5695038.64 frames. ], libri_tot_loss[loss=0.3944, simple_loss=0.4303, pruned_loss=0.1793, over 5680988.97 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3797, pruned_loss=0.1322, over 5709503.57 frames. ], batch size: 78, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:27:44,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5644, 2.8327, 1.4896, 1.4067], device='cuda:0'), covar=tensor([0.0828, 0.0503, 0.0902, 0.1357], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0450, 0.0323, 0.0400], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 10:27:48,451 INFO [train.py:968] (0/2) Epoch 2, batch 40500, giga_loss[loss=0.2844, simple_loss=0.347, pruned_loss=0.1109, over 28859.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.384, pruned_loss=0.1373, over 5691081.35 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.4304, pruned_loss=0.1793, over 5682135.02 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3795, pruned_loss=0.1328, over 5701381.74 frames. ], batch size: 112, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:27:53,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8143, 2.5106, 1.7816, 0.8715], device='cuda:0'), covar=tensor([0.2050, 0.1075, 0.1358, 0.2373], device='cuda:0'), in_proj_covar=tensor([0.1241, 0.1148, 0.1237, 0.1053], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 10:28:11,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:28:12,552 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 10:28:22,388 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.552e+02 1.139e+03 1.449e+03 1.893e+03 3.678e+03, threshold=2.897e+03, percent-clipped=3.0
+2023-03-01 10:28:28,764 INFO [train.py:968] (0/2) Epoch 2, batch 40550, giga_loss[loss=0.3478, simple_loss=0.4003, pruned_loss=0.1477, over 28030.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3814, pruned_loss=0.1358, over 5699708.87 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4301, pruned_loss=0.1789, over 5686422.48 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.377, pruned_loss=0.1317, over 5704289.67 frames. ], batch size: 412, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:28:57,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:29:08,490 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 10:29:12,222 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:29:12,620 INFO [train.py:968] (0/2) Epoch 2, batch 40600, giga_loss[loss=0.266, simple_loss=0.3351, pruned_loss=0.09851, over 28842.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3759, pruned_loss=0.1324, over 5695505.89 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4305, pruned_loss=0.1791, over 5680641.31 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3713, pruned_loss=0.1282, over 5703867.06 frames. ], batch size: 199, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:29:41,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:29:49,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.941e+02 1.140e+03 1.495e+03 1.854e+03 5.688e+03, threshold=2.990e+03, percent-clipped=9.0
+2023-03-01 10:29:56,047 INFO [train.py:968] (0/2) Epoch 2, batch 40650, giga_loss[loss=0.3407, simple_loss=0.3952, pruned_loss=0.1432, over 27873.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3718, pruned_loss=0.1301, over 5701181.81 frames. ], libri_tot_loss[loss=0.395, simple_loss=0.4309, pruned_loss=0.1795, over 5682173.24 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3668, pruned_loss=0.1257, over 5706586.55 frames. ], batch size: 412, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:29:59,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86323.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:30:11,979 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:30:13,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:30:20,080 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:30:37,249 INFO [train.py:968] (0/2) Epoch 2, batch 40700, giga_loss[loss=0.2888, simple_loss=0.3464, pruned_loss=0.1156, over 28476.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3734, pruned_loss=0.131, over 5712094.25 frames. ], libri_tot_loss[loss=0.3947, simple_loss=0.4306, pruned_loss=0.1793, over 5685475.20 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.368, pruned_loss=0.1261, over 5714057.74 frames. ], batch size: 78, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:30:38,653 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:30:58,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:30:58,763 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:00,798 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:04,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:04,460 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 10:31:05,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9336, 1.0894, 0.9773, 0.6770], device='cuda:0'), covar=tensor([0.0446, 0.0459, 0.0336, 0.0460], device='cuda:0'), in_proj_covar=tensor([0.1146, 0.0821, 0.0894, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 10:31:14,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.259e+02 1.254e+03 1.700e+03 2.512e+03 1.308e+04, threshold=3.401e+03, percent-clipped=12.0
+2023-03-01 10:31:20,653 INFO [train.py:968] (0/2) Epoch 2, batch 40750, giga_loss[loss=0.2804, simple_loss=0.3515, pruned_loss=0.1046, over 28775.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3782, pruned_loss=0.1333, over 5692318.38 frames. ], libri_tot_loss[loss=0.3949, simple_loss=0.4307, pruned_loss=0.1796, over 5670500.17 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3729, pruned_loss=0.1285, over 5707332.90 frames. ], batch size: 119, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:31:27,369 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:41,027 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:44,402 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:46,014 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:59,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:32:04,735 INFO [train.py:968] (0/2) Epoch 2, batch 40800, giga_loss[loss=0.3204, simple_loss=0.3858, pruned_loss=0.1275, over 28730.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3825, pruned_loss=0.1356, over 5696383.79 frames. ], libri_tot_loss[loss=0.3949, simple_loss=0.4306, pruned_loss=0.1795, over 5667599.66 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.377, pruned_loss=0.1306, over 5710936.49 frames. ], batch size: 119, lr: 1.18e-02, grad_scale: 8.0
+2023-03-01 10:32:08,118 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:32:08,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2866, 1.6228, 1.3247, 1.5271], device='cuda:0'), covar=tensor([0.0902, 0.0362, 0.0429, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.0257, 0.0172, 0.0176, 0.0309], device='cuda:0'), out_proj_covar=tensor([0.0035, 0.0027, 0.0024, 0.0039], device='cuda:0')
+2023-03-01 10:32:21,676 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9354, 1.0794, 0.8897, 0.5204], device='cuda:0'), covar=tensor([0.0424, 0.0402, 0.0367, 0.0501], device='cuda:0'), in_proj_covar=tensor([0.1139, 0.0816, 0.0892, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 10:32:38,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.302e+02 1.199e+03 1.551e+03 2.159e+03 6.014e+03, threshold=3.102e+03, percent-clipped=8.0
+2023-03-01 10:32:45,014 INFO [train.py:968] (0/2) Epoch 2, batch 40850, giga_loss[loss=0.3419, simple_loss=0.3993, pruned_loss=0.1423, over 28268.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3858, pruned_loss=0.1375, over 5691171.46 frames. ], libri_tot_loss[loss=0.3937, simple_loss=0.4297, pruned_loss=0.1789, over 5671344.73 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3808, pruned_loss=0.1326, over 5700384.28 frames. ], batch size: 77, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:33:01,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9230, 2.1969, 1.4443, 1.1746], device='cuda:0'), covar=tensor([0.0670, 0.0505, 0.0494, 0.0675], device='cuda:0'), in_proj_covar=tensor([0.1134, 0.0819, 0.0889, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 10:33:01,615 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:33:03,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:33:26,974 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:33:27,386 INFO [train.py:968] (0/2) Epoch 2, batch 40900, giga_loss[loss=0.3151, simple_loss=0.3787, pruned_loss=0.1257, over 28935.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3886, pruned_loss=0.1383, over 5701966.32 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4291, pruned_loss=0.1783, over 5676901.21 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3838, pruned_loss=0.1339, over 5704979.14 frames. ], batch size: 186, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:34:00,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5060, 1.4379, 1.1717, 1.2950], device='cuda:0'), covar=tensor([0.0612, 0.0578, 0.0994, 0.0797], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0493, 0.0527, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 10:34:06,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6994, 3.2196, 3.3864, 1.5282], device='cuda:0'), covar=tensor([0.0570, 0.0474, 0.0893, 0.1963], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0566, 0.0792, 0.0554], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 10:34:06,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.666e+02 1.087e+03 1.389e+03 1.864e+03 5.705e+03, threshold=2.777e+03, percent-clipped=6.0
+2023-03-01 10:34:14,635 INFO [train.py:968] (0/2) Epoch 2, batch 40950, libri_loss[loss=0.3501, simple_loss=0.395, pruned_loss=0.1526, over 29595.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3924, pruned_loss=0.141, over 5705464.91 frames. ], libri_tot_loss[loss=0.3921, simple_loss=0.4286, pruned_loss=0.1778, over 5681245.24 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3884, pruned_loss=0.1371, over 5704195.56 frames. ], batch size: 74, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:34:37,657 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:34:41,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:35:09,422 INFO [train.py:968] (0/2) Epoch 2, batch 41000, giga_loss[loss=0.3981, simple_loss=0.4397, pruned_loss=0.1783, over 28650.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.3997, pruned_loss=0.1481, over 5703258.61 frames. ], libri_tot_loss[loss=0.3919, simple_loss=0.4285, pruned_loss=0.1776, over 5684380.87 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3961, pruned_loss=0.1447, over 5699676.05 frames. ], batch size: 307, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:35:39,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86698.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 10:35:54,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.134e+02 1.565e+03 2.130e+03 3.139e+03 6.807e+03, threshold=4.259e+03, percent-clipped=28.0
+2023-03-01 10:35:59,908 INFO [train.py:968] (0/2) Epoch 2, batch 41050, giga_loss[loss=0.3911, simple_loss=0.437, pruned_loss=0.1726, over 28662.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4064, pruned_loss=0.1534, over 5701960.36 frames. ], libri_tot_loss[loss=0.3916, simple_loss=0.4283, pruned_loss=0.1774, over 5687789.17 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.4035, pruned_loss=0.1506, over 5696346.24 frames. ], batch size: 85, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:36:07,523 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:36:11,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 10:36:47,575 INFO [train.py:968] (0/2) Epoch 2, batch 41100, giga_loss[loss=0.3592, simple_loss=0.4079, pruned_loss=0.1553, over 28554.00 frames. ], tot_loss[loss=0.3688, simple_loss=0.4149, pruned_loss=0.1613, over 5697617.98 frames. ], libri_tot_loss[loss=0.392, simple_loss=0.4286, pruned_loss=0.1777, over 5693094.82 frames. ], giga_tot_loss[loss=0.364, simple_loss=0.4117, pruned_loss=0.1582, over 5688704.07 frames. ], batch size: 60, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:36:48,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:02,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6420, 3.5670, 1.7155, 1.4730], device='cuda:0'), covar=tensor([0.0872, 0.0420, 0.0826, 0.1350], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0450, 0.0324, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:0')
+2023-03-01 10:37:03,823 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:06,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:29,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.395e+02 1.649e+03 2.250e+03 3.216e+03 6.447e+03, threshold=4.500e+03, percent-clipped=13.0
+2023-03-01 10:37:35,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:35,505 INFO [train.py:968] (0/2) Epoch 2, batch 41150, giga_loss[loss=0.4077, simple_loss=0.4406, pruned_loss=0.1874, over 28766.00 frames. ], tot_loss[loss=0.3791, simple_loss=0.4222, pruned_loss=0.168, over 5692253.85 frames. ], libri_tot_loss[loss=0.3927, simple_loss=0.4291, pruned_loss=0.1781, over 5685210.17 frames. ], giga_tot_loss[loss=0.3745, simple_loss=0.4191, pruned_loss=0.165, over 5691857.00 frames. ], batch size: 284, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:37:37,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:47,055 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:49,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:50,514 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:52,658 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86841.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:37:57,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86844.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:37:58,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8365, 1.7009, 1.3670, 1.5225], device='cuda:0'), covar=tensor([0.0615, 0.0689, 0.0882, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0503, 0.0532, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 10:38:19,021 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:38:20,196 INFO [train.py:968] (0/2) Epoch 2, batch 41200, giga_loss[loss=0.3692, simple_loss=0.4187, pruned_loss=0.1599, over 28719.00 frames. ], tot_loss[loss=0.3857, simple_loss=0.4271, pruned_loss=0.1721, over 5692289.35 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4293, pruned_loss=0.1781, over 5688234.51 frames. ], giga_tot_loss[loss=0.3813, simple_loss=0.4243, pruned_loss=0.1692, over 5689198.34 frames. ], batch size: 119, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:38:22,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:38:24,366 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86873.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:38:40,886 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 10:38:56,427 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:39:07,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.549e+02 1.778e+03 2.267e+03 3.159e+03 5.228e+03, threshold=4.534e+03, percent-clipped=5.0
+2023-03-01 10:39:09,299 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:39:11,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:39:16,647 INFO [train.py:968] (0/2) Epoch 2, batch 41250, giga_loss[loss=0.3695, simple_loss=0.4202, pruned_loss=0.1594, over 28913.00 frames. ], tot_loss[loss=0.3946, simple_loss=0.4326, pruned_loss=0.1782, over 5670322.31 frames. ], libri_tot_loss[loss=0.393, simple_loss=0.4295, pruned_loss=0.1782, over 5693360.07 frames. ], giga_tot_loss[loss=0.3909, simple_loss=0.4302, pruned_loss=0.1758, over 5662854.31 frames. ], batch size: 119, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:39:41,537 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-01 10:39:46,178 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:04,879 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:07,133 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:10,598 INFO [train.py:968] (0/2) Epoch 2, batch 41300, giga_loss[loss=0.3957, simple_loss=0.4245, pruned_loss=0.1834, over 28752.00 frames. ], tot_loss[loss=0.3971, simple_loss=0.4337, pruned_loss=0.1802, over 5655159.05 frames. ], libri_tot_loss[loss=0.3927, simple_loss=0.4293, pruned_loss=0.1781, over 5679364.36 frames. ], giga_tot_loss[loss=0.3945, simple_loss=0.432, pruned_loss=0.1785, over 5661960.95 frames. ], batch size: 92, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:40:15,049 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:24,360 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:28,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:41,951 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:41:00,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:41:00,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.440e+03 1.906e+03 2.505e+03 5.135e+03, threshold=3.813e+03, percent-clipped=2.0
+2023-03-01 10:41:02,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:41:08,597 INFO [train.py:968] (0/2) Epoch 2, batch 41350, giga_loss[loss=0.5585, simple_loss=0.5294, pruned_loss=0.2938, over 26646.00 frames. ], tot_loss[loss=0.4035, simple_loss=0.4372, pruned_loss=0.1849, over 5639199.18 frames. ], libri_tot_loss[loss=0.3923, simple_loss=0.4291, pruned_loss=0.1778, over 5682989.86 frames. ], giga_tot_loss[loss=0.4019, simple_loss=0.4362, pruned_loss=0.1838, over 5640637.70 frames. ], batch size: 555, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:41:41,893 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-01 10:42:00,968 INFO [train.py:968] (0/2) Epoch 2, batch 41400, giga_loss[loss=0.4217, simple_loss=0.4494, pruned_loss=0.197, over 28829.00 frames. ], tot_loss[loss=0.4093, simple_loss=0.4411, pruned_loss=0.1887, over 5632114.84 frames. ], libri_tot_loss[loss=0.3924, simple_loss=0.4291, pruned_loss=0.1778, over 5675624.19 frames. ], giga_tot_loss[loss=0.4083, simple_loss=0.4405, pruned_loss=0.188, over 5638334.45 frames. ], batch size: 227, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:42:22,990 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6412, 1.1610, 3.7385, 2.9969], device='cuda:0'), covar=tensor([0.1560, 0.1735, 0.0334, 0.0501], device='cuda:0'), in_proj_covar=tensor([0.0524, 0.0495, 0.0669, 0.0531], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 10:42:49,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.708e+02 1.821e+03 2.197e+03 3.030e+03 7.030e+03, threshold=4.394e+03, percent-clipped=15.0
+2023-03-01 10:42:55,091 INFO [train.py:968] (0/2) Epoch 2, batch 41450, giga_loss[loss=0.3853, simple_loss=0.4198, pruned_loss=0.1755, over 28640.00 frames. ], tot_loss[loss=0.411, simple_loss=0.442, pruned_loss=0.19, over 5625322.07 frames. ], libri_tot_loss[loss=0.3925, simple_loss=0.4291, pruned_loss=0.1779, over 5679579.44 frames. ], giga_tot_loss[loss=0.4106, simple_loss=0.4418, pruned_loss=0.1897, over 5625584.08 frames. ], batch size: 242, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:42:59,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-01 10:43:28,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1721, 1.3404, 1.0746, 1.2996], device='cuda:0'), covar=tensor([0.0966, 0.0417, 0.0425, 0.1079], device='cuda:0'), in_proj_covar=tensor([0.0260, 0.0173, 0.0175, 0.0312], device='cuda:0'), out_proj_covar=tensor([0.0036, 0.0027, 0.0024, 0.0040], device='cuda:0')
+2023-03-01 10:43:38,135 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:43:41,678 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:43:47,607 INFO [train.py:968] (0/2) Epoch 2, batch 41500, giga_loss[loss=0.3756, simple_loss=0.4211, pruned_loss=0.165, over 28906.00 frames. ], tot_loss[loss=0.4099, simple_loss=0.4405, pruned_loss=0.1897, over 5631528.20 frames. ], libri_tot_loss[loss=0.3914, simple_loss=0.4282, pruned_loss=0.1773, over 5688211.92 frames. ], giga_tot_loss[loss=0.4115, simple_loss=0.4417, pruned_loss=0.1906, over 5621501.37 frames. ], batch size: 145, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:43:56,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6133, 2.2934, 1.6378, 0.8478], device='cuda:0'), covar=tensor([0.1403, 0.0821, 0.1372, 0.1604], device='cuda:0'), in_proj_covar=tensor([0.1235, 0.1164, 0.1232, 0.1054], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 10:44:06,327 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:44:24,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:44:26,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:44:30,033 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.883e+02 1.796e+03 2.284e+03 3.099e+03 5.692e+03, threshold=4.568e+03, percent-clipped=5.0
+2023-03-01 10:44:30,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:44:34,260 INFO [train.py:968] (0/2) Epoch 2, batch 41550, libri_loss[loss=0.3847, simple_loss=0.4332, pruned_loss=0.168, over 26141.00 frames. ], tot_loss[loss=0.406, simple_loss=0.4378, pruned_loss=0.1871, over 5639075.44 frames. ], libri_tot_loss[loss=0.3907, simple_loss=0.4279, pruned_loss=0.1768, over 5681425.84 frames. ], giga_tot_loss[loss=0.4085, simple_loss=0.4396, pruned_loss=0.1887, over 5635076.08 frames. ], batch size: 136, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:45:24,200 INFO [train.py:968] (0/2) Epoch 2, batch 41600, giga_loss[loss=0.3934, simple_loss=0.4447, pruned_loss=0.1711, over 28925.00 frames. ], tot_loss[loss=0.4022, simple_loss=0.4366, pruned_loss=0.1839, over 5655022.54 frames. ], libri_tot_loss[loss=0.3904, simple_loss=0.4277, pruned_loss=0.1766, over 5686809.84 frames. ], giga_tot_loss[loss=0.4048, simple_loss=0.4384, pruned_loss=0.1856, over 5646073.95 frames. ], batch size: 145, lr: 1.17e-02, grad_scale: 8.0
+2023-03-01 10:45:28,012 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:46:14,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.071e+02 1.800e+03 2.382e+03 3.299e+03 7.048e+03, threshold=4.764e+03, percent-clipped=9.0
+2023-03-01 10:46:18,907 INFO [train.py:968] (0/2) Epoch 2, batch 41650, giga_loss[loss=0.4293, simple_loss=0.4596, pruned_loss=0.1995, over 28441.00 frames. ], tot_loss[loss=0.403, simple_loss=0.4379, pruned_loss=0.1841, over 5659802.27 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.4279, pruned_loss=0.1768, over 5688050.25 frames. ], giga_tot_loss[loss=0.405, simple_loss=0.4394, pruned_loss=0.1853, over 5650844.56 frames. ], batch size: 65, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:46:42,571 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-01 10:46:50,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:46:54,890 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:46:57,296 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:47:00,982 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:47:04,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:47:13,858 INFO [train.py:968] (0/2) Epoch 2, batch 41700, giga_loss[loss=0.3973, simple_loss=0.4371, pruned_loss=0.1788, over 28741.00 frames. ], tot_loss[loss=0.4039, simple_loss=0.4388, pruned_loss=0.1845, over 5645889.16 frames. ], libri_tot_loss[loss=0.3907, simple_loss=0.4279, pruned_loss=0.1768, over 5682958.47 frames. ], giga_tot_loss[loss=0.4058, simple_loss=0.4401, pruned_loss=0.1857, over 5642918.58 frames. ], batch size: 119, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:47:30,950 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:47:34,073 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:47:43,346 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87395.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:48:00,062 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.804e+02 1.570e+03 1.952e+03 2.633e+03 8.655e+03, threshold=3.904e+03, percent-clipped=4.0
+2023-03-01 10:48:06,581 INFO [train.py:968] (0/2) Epoch 2, batch 41750, giga_loss[loss=0.3548, simple_loss=0.4098, pruned_loss=0.1499, over 28551.00 frames. ], tot_loss[loss=0.3994, simple_loss=0.4356, pruned_loss=0.1815, over 5646220.98 frames. ], libri_tot_loss[loss=0.3903, simple_loss=0.4275, pruned_loss=0.1765, over 5684022.74 frames. ], giga_tot_loss[loss=0.4016, simple_loss=0.4373, pruned_loss=0.1829, over 5641851.37 frames. ], batch size: 307, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:48:22,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0385, 3.6049, 3.7532, 1.6953], device='cuda:0'), covar=tensor([0.0475, 0.0460, 0.0749, 0.1981], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0580, 0.0823, 0.0571], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 10:48:57,109 INFO [train.py:968] (0/2) Epoch 2, batch 41800, giga_loss[loss=0.3692, simple_loss=0.4205, pruned_loss=0.1589, over 28371.00 frames. ], tot_loss[loss=0.3937, simple_loss=0.4327, pruned_loss=0.1773, over 5642706.61 frames. ], libri_tot_loss[loss=0.3902, simple_loss=0.4273, pruned_loss=0.1765, over 5676541.89 frames. ], giga_tot_loss[loss=0.3956, simple_loss=0.4343, pruned_loss=0.1785, over 5644697.78 frames. ], batch size: 368, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:49:20,514 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:49:25,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:49:38,075 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-01 10:49:46,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.175e+02 1.616e+03 1.998e+03 2.694e+03 6.113e+03, threshold=3.996e+03, percent-clipped=7.0
+2023-03-01 10:49:50,078 INFO [train.py:968] (0/2) Epoch 2, batch 41850, giga_loss[loss=0.3977, simple_loss=0.4361, pruned_loss=0.1796, over 28559.00 frames. ], tot_loss[loss=0.3864, simple_loss=0.4281, pruned_loss=0.1723, over 5646945.64 frames. ], libri_tot_loss[loss=0.3898, simple_loss=0.427, pruned_loss=0.1763, over 5670884.62 frames. ], giga_tot_loss[loss=0.3882, simple_loss=0.4297, pruned_loss=0.1734, over 5653367.83 frames. ], batch size: 336, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:49:56,131 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:49:59,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2722, 1.7920, 1.3148, 0.4219], device='cuda:0'), covar=tensor([0.1343, 0.0853, 0.1564, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.1233, 0.1169, 0.1236, 0.1052], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 10:50:16,244 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-01 10:50:40,789 INFO [train.py:968] (0/2) Epoch 2, batch 41900, giga_loss[loss=0.3539, simple_loss=0.4112, pruned_loss=0.1483, over 28315.00 frames. ], tot_loss[loss=0.3801, simple_loss=0.4237, pruned_loss=0.1683, over 5644600.72 frames. ], libri_tot_loss[loss=0.3888, simple_loss=0.4263, pruned_loss=0.1757, over 5665595.63 frames. ], giga_tot_loss[loss=0.3823, simple_loss=0.4255, pruned_loss=0.1695, over 5653523.90 frames. ], batch size: 368, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:50:50,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7986, 3.0712, 2.1758, 0.8104], device='cuda:0'), covar=tensor([0.2136, 0.0742, 0.1190, 0.2130], device='cuda:0'), in_proj_covar=tensor([0.1247, 0.1177, 0.1246, 0.1062], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 10:50:53,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:51:24,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.225e+02 1.679e+03 2.112e+03 2.818e+03 5.154e+03, threshold=4.223e+03, percent-clipped=7.0
+2023-03-01 10:51:30,567 INFO [train.py:968] (0/2) Epoch 2, batch 41950, giga_loss[loss=0.3664, simple_loss=0.4185, pruned_loss=0.1571, over 28878.00 frames. ], tot_loss[loss=0.379, simple_loss=0.4219, pruned_loss=0.168, over 5638691.88 frames. ], libri_tot_loss[loss=0.3885, simple_loss=0.426, pruned_loss=0.1755, over 5673676.81 frames. ], giga_tot_loss[loss=0.3807, simple_loss=0.4236, pruned_loss=0.1689, over 5637993.07 frames. ], batch size: 112, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:51:34,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:51:46,843 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 10:51:58,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:52:15,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5594, 1.5053, 1.1935, 1.2747], device='cuda:0'), covar=tensor([0.0560, 0.0541, 0.0918, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0496, 0.0528, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 10:52:19,519 INFO [train.py:968] (0/2) Epoch 2, batch 42000, giga_loss[loss=0.3514, simple_loss=0.4101, pruned_loss=0.1463, over 28853.00 frames. ], tot_loss[loss=0.3782, simple_loss=0.4219, pruned_loss=0.1673, over 5661225.18 frames. ], libri_tot_loss[loss=0.3881, simple_loss=0.4256, pruned_loss=0.1753, over 5677036.77 frames. ], giga_tot_loss[loss=0.3799, simple_loss=0.4235, pruned_loss=0.1681, over 5657363.32 frames. ], batch size: 119, lr: 1.17e-02, grad_scale: 8.0
+2023-03-01 10:52:19,524 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 10:52:25,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8364, 1.3375, 3.4141, 3.0107], device='cuda:0'), covar=tensor([0.1829, 0.2072, 0.0444, 0.0536], device='cuda:0'), in_proj_covar=tensor([0.0519, 0.0497, 0.0670, 0.0528], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 10:52:28,083 INFO [train.py:1012] (0/2) Epoch 2, validation: loss=0.2594, simple_loss=0.3583, pruned_loss=0.08022, over 944034.00 frames. 
+2023-03-01 10:52:28,084 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19537MB
+2023-03-01 10:52:32,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8491, 3.4779, 3.6287, 1.5996], device='cuda:0'), covar=tensor([0.0555, 0.0477, 0.0776, 0.2010], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0586, 0.0831, 0.0578], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 10:53:14,285 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.900e+02 1.478e+03 2.003e+03 3.147e+03 6.401e+03, threshold=4.006e+03, percent-clipped=14.0
+2023-03-01 10:53:18,130 INFO [train.py:968] (0/2) Epoch 2, batch 42050, libri_loss[loss=0.3924, simple_loss=0.438, pruned_loss=0.1733, over 29214.00 frames. ], tot_loss[loss=0.3784, simple_loss=0.4222, pruned_loss=0.1673, over 5657263.37 frames. ], libri_tot_loss[loss=0.3881, simple_loss=0.4257, pruned_loss=0.1752, over 5669857.54 frames. ], giga_tot_loss[loss=0.3794, simple_loss=0.4233, pruned_loss=0.1678, over 5659862.99 frames. ], batch size: 97, lr: 1.17e-02, grad_scale: 8.0
+2023-03-01 10:53:23,297 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:53:27,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:53:47,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9631, 1.0292, 0.9277, 0.6378], device='cuda:0'), covar=tensor([0.0477, 0.0474, 0.0344, 0.0445], device='cuda:0'), in_proj_covar=tensor([0.1128, 0.0812, 0.0887, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 10:53:57,578 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:53:58,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:54:09,538 INFO [train.py:968] (0/2) Epoch 2, batch 42100, giga_loss[loss=0.339, simple_loss=0.4006, pruned_loss=0.1387, over 29094.00 frames. ], tot_loss[loss=0.3725, simple_loss=0.4188, pruned_loss=0.1631, over 5672349.24 frames. ], libri_tot_loss[loss=0.3876, simple_loss=0.4251, pruned_loss=0.175, over 5675029.54 frames. ], giga_tot_loss[loss=0.3735, simple_loss=0.4201, pruned_loss=0.1635, over 5669540.99 frames. ], batch size: 128, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:54:12,347 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87770.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:54:35,703 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:54:38,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:55:08,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.698e+02 1.545e+03 2.010e+03 2.704e+03 8.330e+03, threshold=4.020e+03, percent-clipped=6.0
+2023-03-01 10:55:10,466 INFO [train.py:968] (0/2) Epoch 2, batch 42150, giga_loss[loss=0.3441, simple_loss=0.4041, pruned_loss=0.1421, over 28617.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.42, pruned_loss=0.1616, over 5670577.42 frames. ], libri_tot_loss[loss=0.3871, simple_loss=0.4247, pruned_loss=0.1747, over 5676051.39 frames. ], giga_tot_loss[loss=0.3728, simple_loss=0.4214, pruned_loss=0.1621, over 5667415.01 frames. ], batch size: 71, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:55:14,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:55:46,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:55:50,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7559, 1.3227, 3.4051, 2.9378], device='cuda:0'), covar=tensor([0.1452, 0.1760, 0.0395, 0.0493], device='cuda:0'), in_proj_covar=tensor([0.0514, 0.0490, 0.0662, 0.0522], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 10:55:57,811 INFO [train.py:968] (0/2) Epoch 2, batch 42200, giga_loss[loss=0.4721, simple_loss=0.471, pruned_loss=0.2366, over 26747.00 frames. ], tot_loss[loss=0.3747, simple_loss=0.422, pruned_loss=0.1637, over 5674129.46 frames. ], libri_tot_loss[loss=0.3874, simple_loss=0.4249, pruned_loss=0.1749, over 5680390.11 frames. ], giga_tot_loss[loss=0.375, simple_loss=0.4228, pruned_loss=0.1636, over 5667636.57 frames. ], batch size: 555, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:56:02,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9546, 2.7012, 2.2124, 2.0276], device='cuda:0'), covar=tensor([0.1368, 0.1320, 0.1019, 0.0783], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0811, 0.0723, 0.0616], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 10:56:24,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:56:31,795 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-01 10:56:39,553 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87913.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:56:42,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.686e+02 1.631e+03 2.109e+03 3.215e+03 7.358e+03, threshold=4.218e+03, percent-clipped=14.0
+2023-03-01 10:56:42,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87916.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 10:56:44,871 INFO [train.py:968] (0/2) Epoch 2, batch 42250, giga_loss[loss=0.3123, simple_loss=0.3782, pruned_loss=0.1232, over 28660.00 frames. ], tot_loss[loss=0.3761, simple_loss=0.4224, pruned_loss=0.1649, over 5678911.74 frames. ], libri_tot_loss[loss=0.3864, simple_loss=0.4241, pruned_loss=0.1743, over 5686353.21 frames. ], giga_tot_loss[loss=0.3768, simple_loss=0.4236, pruned_loss=0.165, over 5668268.93 frames. ], batch size: 92, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:57:09,508 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87945.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 10:57:31,785 INFO [train.py:968] (0/2) Epoch 2, batch 42300, giga_loss[loss=0.3978, simple_loss=0.4261, pruned_loss=0.1848, over 27525.00 frames. ], tot_loss[loss=0.3747, simple_loss=0.421, pruned_loss=0.1642, over 5683512.14 frames. ], libri_tot_loss[loss=0.3869, simple_loss=0.4246, pruned_loss=0.1746, over 5690296.46 frames. ], giga_tot_loss[loss=0.3745, simple_loss=0.4214, pruned_loss=0.1638, over 5671267.85 frames. ], batch size: 472, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:57:39,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:57:42,888 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:57:59,428 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:57:59,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:58:00,548 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-88000.pt
+2023-03-01 10:58:19,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.293e+02 1.640e+03 2.000e+03 2.660e+03 1.162e+04, threshold=4.001e+03, percent-clipped=8.0
+2023-03-01 10:58:23,021 INFO [train.py:968] (0/2) Epoch 2, batch 42350, giga_loss[loss=0.4042, simple_loss=0.4138, pruned_loss=0.1974, over 23738.00 frames. ], tot_loss[loss=0.3749, simple_loss=0.4198, pruned_loss=0.165, over 5675749.43 frames. ], libri_tot_loss[loss=0.3871, simple_loss=0.4248, pruned_loss=0.1747, over 5691371.98 frames. ], giga_tot_loss[loss=0.3745, simple_loss=0.42, pruned_loss=0.1645, over 5665267.15 frames. ], batch size: 705, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:58:23,600 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-01 10:58:44,892 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-01 10:59:06,464 INFO [train.py:968] (0/2) Epoch 2, batch 42400, giga_loss[loss=0.4246, simple_loss=0.4642, pruned_loss=0.1925, over 28907.00 frames. ], tot_loss[loss=0.3733, simple_loss=0.4182, pruned_loss=0.1642, over 5666195.02 frames. ], libri_tot_loss[loss=0.3866, simple_loss=0.4245, pruned_loss=0.1744, over 5684697.16 frames. ], giga_tot_loss[loss=0.3729, simple_loss=0.4184, pruned_loss=0.1637, over 5663149.25 frames. ], batch size: 145, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:59:57,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.661e+02 1.445e+03 1.749e+03 2.185e+03 1.126e+04, threshold=3.498e+03, percent-clipped=3.0
+2023-03-01 11:00:00,936 INFO [train.py:968] (0/2) Epoch 2, batch 42450, giga_loss[loss=0.329, simple_loss=0.3964, pruned_loss=0.1308, over 28539.00 frames. ], tot_loss[loss=0.3707, simple_loss=0.4177, pruned_loss=0.1619, over 5666798.85 frames. ], libri_tot_loss[loss=0.3868, simple_loss=0.4247, pruned_loss=0.1745, over 5675951.67 frames. ], giga_tot_loss[loss=0.3702, simple_loss=0.4176, pruned_loss=0.1614, over 5672504.41 frames. ], batch size: 60, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:00:04,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8815, 3.3317, 2.2486, 0.8503], device='cuda:0'), covar=tensor([0.2261, 0.0875, 0.1244, 0.2438], device='cuda:0'), in_proj_covar=tensor([0.1233, 0.1170, 0.1230, 0.1056], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 11:00:10,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:00:21,637 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:00:25,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:00:47,076 INFO [train.py:968] (0/2) Epoch 2, batch 42500, libri_loss[loss=0.3376, simple_loss=0.3881, pruned_loss=0.1435, over 29580.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4168, pruned_loss=0.1605, over 5663991.42 frames. ], libri_tot_loss[loss=0.3864, simple_loss=0.4242, pruned_loss=0.1743, over 5671844.72 frames. ], giga_tot_loss[loss=0.3683, simple_loss=0.4169, pruned_loss=0.1598, over 5672238.79 frames. ], batch size: 78, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:00:50,040 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:01:03,648 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 11:01:32,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.093e+02 1.673e+03 2.282e+03 3.246e+03 8.465e+03, threshold=4.564e+03, percent-clipped=22.0
+2023-03-01 11:01:35,074 INFO [train.py:968] (0/2) Epoch 2, batch 42550, giga_loss[loss=0.3281, simple_loss=0.3935, pruned_loss=0.1314, over 28787.00 frames. ], tot_loss[loss=0.3692, simple_loss=0.4172, pruned_loss=0.1606, over 5669836.87 frames. ], libri_tot_loss[loss=0.3868, simple_loss=0.4245, pruned_loss=0.1745, over 5664355.15 frames. ], giga_tot_loss[loss=0.3679, simple_loss=0.4168, pruned_loss=0.1595, over 5683132.34 frames. ], batch size: 186, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:01:44,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:01:59,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5521, 2.5817, 1.5334, 2.1206], device='cuda:0'), covar=tensor([0.0712, 0.0739, 0.1534, 0.1106], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0502, 0.0529, 0.0473], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 11:02:24,005 INFO [train.py:968] (0/2) Epoch 2, batch 42600, giga_loss[loss=0.359, simple_loss=0.386, pruned_loss=0.1659, over 23698.00 frames. ], tot_loss[loss=0.3679, simple_loss=0.4154, pruned_loss=0.1602, over 5665794.13 frames. ], libri_tot_loss[loss=0.3868, simple_loss=0.4245, pruned_loss=0.1746, over 5665607.38 frames. ], giga_tot_loss[loss=0.3668, simple_loss=0.4152, pruned_loss=0.1592, over 5675146.04 frames. ], batch size: 705, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:02:28,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:02:28,039 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:02:32,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:03:05,617 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:03:12,732 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.771e+02 1.537e+03 1.912e+03 2.806e+03 5.007e+03, threshold=3.824e+03, percent-clipped=3.0
+2023-03-01 11:03:16,722 INFO [train.py:968] (0/2) Epoch 2, batch 42650, giga_loss[loss=0.3844, simple_loss=0.4015, pruned_loss=0.1837, over 23473.00 frames. ], tot_loss[loss=0.3679, simple_loss=0.4149, pruned_loss=0.1604, over 5660694.04 frames. ], libri_tot_loss[loss=0.387, simple_loss=0.4247, pruned_loss=0.1746, over 5659416.98 frames. ], giga_tot_loss[loss=0.3664, simple_loss=0.4143, pruned_loss=0.1592, over 5673846.74 frames. ], batch size: 705, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:03:28,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3268, 1.2904, 1.1469, 1.4764], device='cuda:0'), covar=tensor([0.1886, 0.1952, 0.1652, 0.2272], device='cuda:0'), in_proj_covar=tensor([0.0987, 0.0811, 0.0891, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 11:03:50,046 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:03:56,571 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:07,554 INFO [train.py:968] (0/2) Epoch 2, batch 42700, giga_loss[loss=0.3402, simple_loss=0.3907, pruned_loss=0.1449, over 29035.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.4138, pruned_loss=0.1606, over 5660826.83 frames. ], libri_tot_loss[loss=0.3866, simple_loss=0.4244, pruned_loss=0.1744, over 5663712.44 frames. ], giga_tot_loss[loss=0.3663, simple_loss=0.4134, pruned_loss=0.1596, over 5667458.42 frames. ], batch size: 128, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:04:12,513 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:13,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:14,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:29,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3801, 4.8756, 5.1409, 2.1385], device='cuda:0'), covar=tensor([0.0294, 0.0268, 0.0611, 0.1597], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0591, 0.0837, 0.0577], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 11:04:44,138 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:54,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.234e+02 1.501e+03 1.930e+03 2.436e+03 6.258e+03, threshold=3.860e+03, percent-clipped=7.0
+2023-03-01 11:04:55,213 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:57,203 INFO [train.py:968] (0/2) Epoch 2, batch 42750, giga_loss[loss=0.3227, simple_loss=0.3816, pruned_loss=0.1319, over 28731.00 frames. ], tot_loss[loss=0.3685, simple_loss=0.4137, pruned_loss=0.1616, over 5667513.43 frames. ], libri_tot_loss[loss=0.3861, simple_loss=0.424, pruned_loss=0.1741, over 5669923.83 frames. ], giga_tot_loss[loss=0.3675, simple_loss=0.4135, pruned_loss=0.1608, over 5666975.43 frames. ], batch size: 262, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:04:57,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:05:27,604 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:05:34,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2936, 1.9173, 1.4114, 0.5177], device='cuda:0'), covar=tensor([0.1504, 0.0932, 0.1551, 0.1824], device='cuda:0'), in_proj_covar=tensor([0.1239, 0.1180, 0.1237, 0.1056], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 11:05:34,758 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6563, 2.0414, 1.7910, 1.8140], device='cuda:0'), covar=tensor([0.1311, 0.1523, 0.1052, 0.0768], device='cuda:0'), in_proj_covar=tensor([0.0787, 0.0808, 0.0716, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 11:05:47,967 INFO [train.py:968] (0/2) Epoch 2, batch 42800, libri_loss[loss=0.2902, simple_loss=0.3426, pruned_loss=0.1189, over 28173.00 frames. ], tot_loss[loss=0.367, simple_loss=0.4123, pruned_loss=0.1609, over 5675584.08 frames. ], libri_tot_loss[loss=0.3859, simple_loss=0.4238, pruned_loss=0.174, over 5671654.15 frames. ], giga_tot_loss[loss=0.3662, simple_loss=0.4121, pruned_loss=0.1601, over 5673758.08 frames. ], batch size: 62, lr: 1.16e-02, grad_scale: 8.0
+2023-03-01 11:06:15,475 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=88493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:17,884 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:19,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:22,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:23,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:35,989 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:36,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.357e+02 1.716e+03 2.166e+03 3.485e+03 6.207e+03, threshold=4.331e+03, percent-clipped=19.0
+2023-03-01 11:06:39,729 INFO [train.py:968] (0/2) Epoch 2, batch 42850, giga_loss[loss=0.377, simple_loss=0.425, pruned_loss=0.1645, over 28489.00 frames. ], tot_loss[loss=0.367, simple_loss=0.4123, pruned_loss=0.1609, over 5676050.33 frames. ], libri_tot_loss[loss=0.3851, simple_loss=0.4233, pruned_loss=0.1734, over 5667633.88 frames. ], giga_tot_loss[loss=0.3667, simple_loss=0.4124, pruned_loss=0.1605, over 5678037.69 frames. ], batch size: 60, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:06:40,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:47,529 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:52,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:52,789 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.00 vs. limit=2.0
+2023-03-01 11:07:06,134 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:07:06,812 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5243, 2.1348, 1.6652, 1.5651], device='cuda:0'), covar=tensor([0.0861, 0.0289, 0.0374, 0.0992], device='cuda:0'), in_proj_covar=tensor([0.0259, 0.0172, 0.0175, 0.0311], device='cuda:0'), out_proj_covar=tensor([0.0036, 0.0027, 0.0024, 0.0040], device='cuda:0')
+2023-03-01 11:07:25,136 INFO [train.py:968] (0/2) Epoch 2, batch 42900, giga_loss[loss=0.3626, simple_loss=0.421, pruned_loss=0.152, over 28360.00 frames. ], tot_loss[loss=0.3682, simple_loss=0.4141, pruned_loss=0.1612, over 5682444.73 frames. ], libri_tot_loss[loss=0.3846, simple_loss=0.4231, pruned_loss=0.1731, over 5673054.23 frames. ], giga_tot_loss[loss=0.3679, simple_loss=0.4141, pruned_loss=0.1609, over 5679466.03 frames. ], batch size: 71, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:08:14,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.608e+02 1.686e+03 2.062e+03 2.807e+03 7.852e+03, threshold=4.124e+03, percent-clipped=7.0
+2023-03-01 11:08:14,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5952, 2.9599, 1.8990, 1.3408], device='cuda:0'), covar=tensor([0.0675, 0.0438, 0.0487, 0.0740], device='cuda:0'), in_proj_covar=tensor([0.1163, 0.0842, 0.0946, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 11:08:15,421 INFO [train.py:968] (0/2) Epoch 2, batch 42950, giga_loss[loss=0.3049, simple_loss=0.3745, pruned_loss=0.1177, over 28301.00 frames. ], tot_loss[loss=0.3703, simple_loss=0.416, pruned_loss=0.1622, over 5681206.34 frames. ], libri_tot_loss[loss=0.3851, simple_loss=0.4234, pruned_loss=0.1733, over 5671963.09 frames. ], giga_tot_loss[loss=0.3695, simple_loss=0.4157, pruned_loss=0.1617, over 5679865.07 frames. ], batch size: 65, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:09:04,755 INFO [train.py:968] (0/2) Epoch 2, batch 43000, giga_loss[loss=0.3422, simple_loss=0.4028, pruned_loss=0.1408, over 28392.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4155, pruned_loss=0.1611, over 5679050.66 frames. ], libri_tot_loss[loss=0.3843, simple_loss=0.4229, pruned_loss=0.1729, over 5676819.38 frames. ], giga_tot_loss[loss=0.3687, simple_loss=0.4156, pruned_loss=0.1608, over 5673439.47 frames. ], batch size: 65, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:09:22,797 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=88684.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 11:09:57,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.366e+02 1.631e+03 2.092e+03 2.617e+03 6.401e+03, threshold=4.184e+03, percent-clipped=3.0
+2023-03-01 11:09:59,748 INFO [train.py:968] (0/2) Epoch 2, batch 43050, giga_loss[loss=0.4366, simple_loss=0.4633, pruned_loss=0.2049, over 28703.00 frames. ], tot_loss[loss=0.3703, simple_loss=0.416, pruned_loss=0.1623, over 5673068.58 frames. ], libri_tot_loss[loss=0.3837, simple_loss=0.4225, pruned_loss=0.1725, over 5682809.48 frames. ], giga_tot_loss[loss=0.3703, simple_loss=0.4162, pruned_loss=0.1622, over 5663378.81 frames. ], batch size: 242, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:10:42,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7652, 2.0151, 1.8189, 1.7800], device='cuda:0'), covar=tensor([0.1153, 0.1386, 0.0925, 0.0736], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0805, 0.0715, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 11:10:48,387 INFO [train.py:968] (0/2) Epoch 2, batch 43100, giga_loss[loss=0.4715, simple_loss=0.4743, pruned_loss=0.2343, over 28651.00 frames. ], tot_loss[loss=0.376, simple_loss=0.4196, pruned_loss=0.1662, over 5674980.43 frames. ], libri_tot_loss[loss=0.3839, simple_loss=0.4228, pruned_loss=0.1725, over 5689405.36 frames. ], giga_tot_loss[loss=0.3754, simple_loss=0.4193, pruned_loss=0.1658, over 5660530.67 frames. ], batch size: 262, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:11:40,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.960e+02 2.008e+03 2.548e+03 3.736e+03 7.610e+03, threshold=5.097e+03, percent-clipped=17.0
+2023-03-01 11:11:42,039 INFO [train.py:968] (0/2) Epoch 2, batch 43150, giga_loss[loss=0.3609, simple_loss=0.409, pruned_loss=0.1564, over 29043.00 frames. ], tot_loss[loss=0.3796, simple_loss=0.4209, pruned_loss=0.1692, over 5665296.32 frames. ], libri_tot_loss[loss=0.383, simple_loss=0.4221, pruned_loss=0.172, over 5689617.50 frames. ], giga_tot_loss[loss=0.3798, simple_loss=0.4212, pruned_loss=0.1692, over 5653575.85 frames. ], batch size: 128, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:11:50,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4556, 1.4259, 1.3149, 1.7433], device='cuda:0'), covar=tensor([0.1557, 0.1311, 0.1170, 0.1389], device='cuda:0'), in_proj_covar=tensor([0.0999, 0.0815, 0.0901, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 11:12:33,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:12:34,062 INFO [train.py:968] (0/2) Epoch 2, batch 43200, giga_loss[loss=0.3276, simple_loss=0.3782, pruned_loss=0.1385, over 28972.00 frames. ], tot_loss[loss=0.382, simple_loss=0.4218, pruned_loss=0.1711, over 5667903.94 frames. ], libri_tot_loss[loss=0.3826, simple_loss=0.4217, pruned_loss=0.1717, over 5693909.93 frames. ], giga_tot_loss[loss=0.3825, simple_loss=0.4223, pruned_loss=0.1714, over 5653758.01 frames. ], batch size: 106, lr: 1.16e-02, grad_scale: 8.0
+2023-03-01 11:12:36,869 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 11:13:17,931 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4363, 2.4610, 1.5070, 1.3641], device='cuda:0'), covar=tensor([0.0823, 0.0477, 0.0800, 0.1295], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0455, 0.0325, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 11:13:27,321 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.388e+02 1.771e+03 2.422e+03 3.177e+03 6.953e+03, threshold=4.845e+03, percent-clipped=6.0
+2023-03-01 11:13:29,766 INFO [train.py:968] (0/2) Epoch 2, batch 43250, giga_loss[loss=0.4927, simple_loss=0.4869, pruned_loss=0.2492, over 26510.00 frames. ], tot_loss[loss=0.383, simple_loss=0.4224, pruned_loss=0.1718, over 5668414.62 frames. ], libri_tot_loss[loss=0.3824, simple_loss=0.4216, pruned_loss=0.1716, over 5695089.38 frames. ], giga_tot_loss[loss=0.3835, simple_loss=0.4229, pruned_loss=0.1721, over 5656325.03 frames. ], batch size: 555, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:13:31,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7170, 2.0464, 1.8105, 1.7949], device='cuda:0'), covar=tensor([0.1213, 0.1741, 0.1077, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0817, 0.0717, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 11:14:13,985 INFO [train.py:968] (0/2) Epoch 2, batch 43300, giga_loss[loss=0.3483, simple_loss=0.3925, pruned_loss=0.1521, over 28731.00 frames. ], tot_loss[loss=0.3803, simple_loss=0.4204, pruned_loss=0.1701, over 5670329.10 frames. ], libri_tot_loss[loss=0.3821, simple_loss=0.4214, pruned_loss=0.1714, over 5691473.40 frames. ], giga_tot_loss[loss=0.3809, simple_loss=0.421, pruned_loss=0.1705, over 5662940.81 frames. ], batch size: 99, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:14:37,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5043, 2.2607, 1.5796, 0.6123], device='cuda:0'), covar=tensor([0.1696, 0.0985, 0.1371, 0.2008], device='cuda:0'), in_proj_covar=tensor([0.1238, 0.1194, 0.1234, 0.1061], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 11:14:55,468 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:14:57,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:15:01,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.220e+02 1.710e+03 2.104e+03 2.946e+03 6.874e+03, threshold=4.208e+03, percent-clipped=4.0
+2023-03-01 11:15:02,403 INFO [train.py:968] (0/2) Epoch 2, batch 43350, giga_loss[loss=0.3752, simple_loss=0.4288, pruned_loss=0.1608, over 28924.00 frames. ], tot_loss[loss=0.3762, simple_loss=0.4189, pruned_loss=0.1668, over 5681511.58 frames. ], libri_tot_loss[loss=0.3824, simple_loss=0.4218, pruned_loss=0.1715, over 5696727.87 frames. ], giga_tot_loss[loss=0.3764, simple_loss=0.4189, pruned_loss=0.1669, over 5670296.07 frames. ], batch size: 145, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:15:24,656 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:15:28,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-01 11:15:30,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2920, 2.4425, 1.4462, 1.2776], device='cuda:0'), covar=tensor([0.0955, 0.0478, 0.0955, 0.1471], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0458, 0.0325, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 11:15:38,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89059.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 11:15:48,466 INFO [train.py:968] (0/2) Epoch 2, batch 43400, giga_loss[loss=0.3571, simple_loss=0.4007, pruned_loss=0.1568, over 28828.00 frames. ], tot_loss[loss=0.3708, simple_loss=0.415, pruned_loss=0.1633, over 5678586.70 frames. ], libri_tot_loss[loss=0.3832, simple_loss=0.4223, pruned_loss=0.172, over 5691679.33 frames. ], giga_tot_loss[loss=0.3699, simple_loss=0.4144, pruned_loss=0.1627, over 5672952.73 frames. ], batch size: 99, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:16:36,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.927e+03 2.373e+03 3.082e+03 5.756e+03, threshold=4.745e+03, percent-clipped=4.0
+2023-03-01 11:16:38,180 INFO [train.py:968] (0/2) Epoch 2, batch 43450, giga_loss[loss=0.3519, simple_loss=0.376, pruned_loss=0.1639, over 23905.00 frames. ], tot_loss[loss=0.3694, simple_loss=0.4129, pruned_loss=0.1629, over 5667255.56 frames. ], libri_tot_loss[loss=0.3828, simple_loss=0.422, pruned_loss=0.1718, over 5695055.67 frames. ], giga_tot_loss[loss=0.3688, simple_loss=0.4126, pruned_loss=0.1625, over 5659490.19 frames. ], batch size: 705, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:17:16,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7011, 2.4653, 2.0100, 1.9916], device='cuda:0'), covar=tensor([0.0893, 0.0271, 0.0354, 0.0939], device='cuda:0'), in_proj_covar=tensor([0.0258, 0.0171, 0.0175, 0.0309], device='cuda:0'), out_proj_covar=tensor([0.0036, 0.0027, 0.0024, 0.0041], device='cuda:0')
+2023-03-01 11:17:25,751 INFO [train.py:968] (0/2) Epoch 2, batch 43500, giga_loss[loss=0.3558, simple_loss=0.375, pruned_loss=0.1683, over 23581.00 frames. ], tot_loss[loss=0.3688, simple_loss=0.4119, pruned_loss=0.1628, over 5662921.04 frames. ], libri_tot_loss[loss=0.3828, simple_loss=0.422, pruned_loss=0.1718, over 5685626.03 frames. ], giga_tot_loss[loss=0.3681, simple_loss=0.4115, pruned_loss=0.1624, over 5663972.18 frames. ], batch size: 705, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:17:55,295 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89202.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 11:17:57,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89205.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 11:18:09,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.238e+02 1.390e+03 1.806e+03 2.382e+03 7.145e+03, threshold=3.613e+03, percent-clipped=2.0
+2023-03-01 11:18:10,199 INFO [train.py:968] (0/2) Epoch 2, batch 43550, giga_loss[loss=0.3584, simple_loss=0.4165, pruned_loss=0.1501, over 28877.00 frames. ], tot_loss[loss=0.3713, simple_loss=0.4136, pruned_loss=0.1645, over 5655639.73 frames. ], libri_tot_loss[loss=0.3836, simple_loss=0.4227, pruned_loss=0.1723, over 5680324.35 frames. ], giga_tot_loss[loss=0.3695, simple_loss=0.4123, pruned_loss=0.1634, over 5660216.91 frames. ], batch size: 174, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:18:23,869 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89234.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 11:18:49,487 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 11:18:59,039 INFO [train.py:968] (0/2) Epoch 2, batch 43600, giga_loss[loss=0.3771, simple_loss=0.4252, pruned_loss=0.1644, over 28701.00 frames. ], tot_loss[loss=0.3746, simple_loss=0.4171, pruned_loss=0.166, over 5658000.22 frames. ], libri_tot_loss[loss=0.3831, simple_loss=0.4224, pruned_loss=0.1719, over 5680976.52 frames. ], giga_tot_loss[loss=0.3734, simple_loss=0.4162, pruned_loss=0.1653, over 5660304.02 frames. ], batch size: 262, lr: 1.16e-02, grad_scale: 8.0
+2023-03-01 11:19:04,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1986, 1.2652, 1.1439, 1.3340], device='cuda:0'), covar=tensor([0.1985, 0.1925, 0.1744, 0.1991], device='cuda:0'), in_proj_covar=tensor([0.0987, 0.0808, 0.0893, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 11:19:45,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.528e+03 2.059e+03 2.715e+03 6.477e+03, threshold=4.117e+03, percent-clipped=12.0
+2023-03-01 11:19:45,507 INFO [train.py:968] (0/2) Epoch 2, batch 43650, giga_loss[loss=0.3697, simple_loss=0.4306, pruned_loss=0.1544, over 28981.00 frames. ], tot_loss[loss=0.3757, simple_loss=0.4205, pruned_loss=0.1655, over 5662327.90 frames. ], libri_tot_loss[loss=0.3824, simple_loss=0.4219, pruned_loss=0.1715, over 5684798.50 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.4201, pruned_loss=0.1651, over 5659971.12 frames. ], batch size: 213, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:19:51,087 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-01 11:20:42,173 INFO [train.py:968] (0/2) Epoch 2, batch 43700, giga_loss[loss=0.4199, simple_loss=0.464, pruned_loss=0.1879, over 28980.00 frames. ], tot_loss[loss=0.3767, simple_loss=0.4225, pruned_loss=0.1655, over 5664195.56 frames. ], libri_tot_loss[loss=0.3821, simple_loss=0.4217, pruned_loss=0.1712, over 5686722.40 frames. ], giga_tot_loss[loss=0.3765, simple_loss=0.4224, pruned_loss=0.1654, over 5660431.52 frames. ], batch size: 227, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:21:10,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4705, 1.4466, 1.4041, 1.6550], device='cuda:0'), covar=tensor([0.1551, 0.1357, 0.1134, 0.1453], device='cuda:0'), in_proj_covar=tensor([0.0984, 0.0809, 0.0893, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 11:21:27,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.744e+02 1.905e+03 2.193e+03 2.740e+03 7.436e+03, threshold=4.385e+03, percent-clipped=7.0
+2023-03-01 11:21:27,296 INFO [train.py:968] (0/2) Epoch 2, batch 43750, giga_loss[loss=0.4347, simple_loss=0.4654, pruned_loss=0.202, over 28310.00 frames. ], tot_loss[loss=0.3825, simple_loss=0.4265, pruned_loss=0.1692, over 5656757.93 frames. ], libri_tot_loss[loss=0.3825, simple_loss=0.422, pruned_loss=0.1715, over 5672600.02 frames. ], giga_tot_loss[loss=0.3818, simple_loss=0.4262, pruned_loss=0.1687, over 5665919.87 frames. ], batch size: 368, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:21:28,487 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:22:09,058 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2475, 1.7863, 1.3577, 0.3905], device='cuda:0'), covar=tensor([0.1010, 0.0652, 0.0989, 0.1199], device='cuda:0'), in_proj_covar=tensor([0.1227, 0.1178, 0.1218, 0.1049], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 11:22:16,186 INFO [train.py:968] (0/2) Epoch 2, batch 43800, giga_loss[loss=0.3036, simple_loss=0.3622, pruned_loss=0.1225, over 28383.00 frames. ], tot_loss[loss=0.3843, simple_loss=0.4273, pruned_loss=0.1706, over 5662654.79 frames. ], libri_tot_loss[loss=0.3818, simple_loss=0.4214, pruned_loss=0.1711, over 5677607.88 frames. ], giga_tot_loss[loss=0.3844, simple_loss=0.4278, pruned_loss=0.1705, over 5665075.37 frames. ], batch size: 77, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:23:02,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.978e+02 1.464e+03 1.839e+03 2.600e+03 6.313e+03, threshold=3.678e+03, percent-clipped=5.0
+2023-03-01 11:23:02,421 INFO [train.py:968] (0/2) Epoch 2, batch 43850, libri_loss[loss=0.4562, simple_loss=0.4756, pruned_loss=0.2184, over 19586.00 frames. ], tot_loss[loss=0.3846, simple_loss=0.4269, pruned_loss=0.1712, over 5663223.86 frames. ], libri_tot_loss[loss=0.3829, simple_loss=0.4222, pruned_loss=0.1718, over 5673295.39 frames. ], giga_tot_loss[loss=0.3838, simple_loss=0.4268, pruned_loss=0.1704, over 5669625.18 frames. ], batch size: 187, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:23:47,497 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:23:54,149 INFO [train.py:968] (0/2) Epoch 2, batch 43900, libri_loss[loss=0.4015, simple_loss=0.4391, pruned_loss=0.1819, over 27512.00 frames. ], tot_loss[loss=0.3848, simple_loss=0.4262, pruned_loss=0.1717, over 5655915.88 frames. ], libri_tot_loss[loss=0.3825, simple_loss=0.4219, pruned_loss=0.1715, over 5676280.97 frames. ], giga_tot_loss[loss=0.3845, simple_loss=0.4263, pruned_loss=0.1714, over 5658032.43 frames. ], batch size: 116, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:24:13,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6558, 1.4476, 1.5414, 1.4476], device='cuda:0'), covar=tensor([0.0891, 0.1641, 0.1410, 0.1273], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0817, 0.0641, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 11:24:39,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.374e+02 1.668e+03 2.043e+03 2.929e+03 7.035e+03, threshold=4.087e+03, percent-clipped=17.0
+2023-03-01 11:24:39,098 INFO [train.py:968] (0/2) Epoch 2, batch 43950, giga_loss[loss=0.4719, simple_loss=0.4723, pruned_loss=0.2357, over 26497.00 frames. ], tot_loss[loss=0.3821, simple_loss=0.4233, pruned_loss=0.1704, over 5656401.36 frames. ], libri_tot_loss[loss=0.3825, simple_loss=0.4217, pruned_loss=0.1716, over 5671343.86 frames. ], giga_tot_loss[loss=0.3818, simple_loss=0.4237, pruned_loss=0.17, over 5661805.68 frames. ], batch size: 555, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:25:26,632 INFO [train.py:968] (0/2) Epoch 2, batch 44000, giga_loss[loss=0.3584, simple_loss=0.4132, pruned_loss=0.1518, over 28898.00 frames. ], tot_loss[loss=0.3821, simple_loss=0.4226, pruned_loss=0.1708, over 5665958.41 frames. ], libri_tot_loss[loss=0.382, simple_loss=0.4213, pruned_loss=0.1714, over 5676289.66 frames. ], giga_tot_loss[loss=0.3823, simple_loss=0.4233, pruned_loss=0.1707, over 5665569.36 frames. ], batch size: 174, lr: 1.16e-02, grad_scale: 8.0
+2023-03-01 11:26:01,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2345, 1.3494, 1.0380, 1.2126], device='cuda:0'), covar=tensor([0.0521, 0.0393, 0.0857, 0.0637], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0497, 0.0530, 0.0473], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 11:26:20,715 INFO [train.py:968] (0/2) Epoch 2, batch 44050, giga_loss[loss=0.3993, simple_loss=0.4331, pruned_loss=0.1827, over 28286.00 frames. ], tot_loss[loss=0.3817, simple_loss=0.422, pruned_loss=0.1707, over 5633011.89 frames. ], libri_tot_loss[loss=0.3823, simple_loss=0.4216, pruned_loss=0.1715, over 5660949.54 frames. ], giga_tot_loss[loss=0.3816, simple_loss=0.4223, pruned_loss=0.1704, over 5645636.58 frames. ], batch size: 368, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:26:21,311 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.243e+02 1.699e+03 2.213e+03 3.127e+03 7.502e+03, threshold=4.426e+03, percent-clipped=10.0
+2023-03-01 11:26:27,980 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 11:26:44,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1352, 1.2893, 1.1869, 1.0108], device='cuda:0'), covar=tensor([0.1765, 0.1718, 0.1570, 0.1583], device='cuda:0'), in_proj_covar=tensor([0.1007, 0.0816, 0.0906, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 11:27:09,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:27:12,820 INFO [train.py:968] (0/2) Epoch 2, batch 44100, giga_loss[loss=0.3521, simple_loss=0.3885, pruned_loss=0.1579, over 28180.00 frames. ], tot_loss[loss=0.3797, simple_loss=0.42, pruned_loss=0.1696, over 5633000.16 frames. ], libri_tot_loss[loss=0.3822, simple_loss=0.4215, pruned_loss=0.1715, over 5656677.54 frames. ], giga_tot_loss[loss=0.3796, simple_loss=0.4204, pruned_loss=0.1694, over 5646885.68 frames. ], batch size: 77, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:27:32,410 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-01 11:27:34,317 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:28:00,114 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:28:00,571 INFO [train.py:968] (0/2) Epoch 2, batch 44150, giga_loss[loss=0.3571, simple_loss=0.4104, pruned_loss=0.1519, over 28798.00 frames. ], tot_loss[loss=0.3767, simple_loss=0.4179, pruned_loss=0.1678, over 5651074.36 frames. ], libri_tot_loss[loss=0.3818, simple_loss=0.4211, pruned_loss=0.1713, over 5659688.18 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4185, pruned_loss=0.1677, over 5659091.56 frames. ], batch size: 285, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:28:01,125 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.687e+03 2.198e+03 2.760e+03 6.810e+03, threshold=4.396e+03, percent-clipped=6.0
+2023-03-01 11:28:13,420 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89832.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 11:28:21,058 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9959, 3.5755, 3.7508, 1.6717], device='cuda:0'), covar=tensor([0.0522, 0.0432, 0.0804, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0591, 0.0822, 0.0567], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 11:28:43,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5451, 5.0039, 5.1933, 2.2122], device='cuda:0'), covar=tensor([0.0382, 0.0446, 0.1031, 0.1872], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0589, 0.0819, 0.0567], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 11:28:45,307 INFO [train.py:968] (0/2) Epoch 2, batch 44200, giga_loss[loss=0.3839, simple_loss=0.4358, pruned_loss=0.1661, over 28999.00 frames. ], tot_loss[loss=0.3764, simple_loss=0.4177, pruned_loss=0.1675, over 5652464.19 frames. ], libri_tot_loss[loss=0.3824, simple_loss=0.4215, pruned_loss=0.1716, over 5662992.74 frames. ], giga_tot_loss[loss=0.376, simple_loss=0.4177, pruned_loss=0.1672, over 5655508.89 frames. ], batch size: 145, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:29:07,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6233, 2.9736, 1.5667, 1.4100], device='cuda:0'), covar=tensor([0.0791, 0.0449, 0.0823, 0.1286], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0457, 0.0323, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 11:29:16,514 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:29:36,267 INFO [train.py:968] (0/2) Epoch 2, batch 44250, giga_loss[loss=0.3602, simple_loss=0.426, pruned_loss=0.1472, over 29013.00 frames. ], tot_loss[loss=0.3779, simple_loss=0.4196, pruned_loss=0.1681, over 5643072.16 frames. ], libri_tot_loss[loss=0.3823, simple_loss=0.4215, pruned_loss=0.1716, over 5656188.74 frames. ], giga_tot_loss[loss=0.3774, simple_loss=0.4196, pruned_loss=0.1677, over 5651465.75 frames. ], batch size: 164, lr: 1.16e-02, grad_scale: 2.0
+2023-03-01 11:29:37,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.389e+02 1.779e+03 2.334e+03 3.143e+03 7.744e+03, threshold=4.667e+03, percent-clipped=13.0
+2023-03-01 11:29:38,641 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 11:29:53,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:29:54,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:29:59,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:30:22,170 INFO [train.py:968] (0/2) Epoch 2, batch 44300, giga_loss[loss=0.3547, simple_loss=0.4123, pruned_loss=0.1486, over 28583.00 frames. ], tot_loss[loss=0.3816, simple_loss=0.4225, pruned_loss=0.1704, over 5636999.45 frames. ], libri_tot_loss[loss=0.3819, simple_loss=0.421, pruned_loss=0.1714, over 5646199.18 frames. ], giga_tot_loss[loss=0.3816, simple_loss=0.4228, pruned_loss=0.1702, over 5653047.19 frames. ], batch size: 78, lr: 1.15e-02, grad_scale: 2.0
+2023-03-01 11:30:25,267 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:30:55,454 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-90000.pt
+2023-03-01 11:31:16,014 INFO [train.py:968] (0/2) Epoch 2, batch 44350, giga_loss[loss=0.3992, simple_loss=0.4341, pruned_loss=0.1821, over 28242.00 frames. ], tot_loss[loss=0.3815, simple_loss=0.422, pruned_loss=0.1705, over 5650598.55 frames. ], libri_tot_loss[loss=0.3814, simple_loss=0.4206, pruned_loss=0.1711, over 5650088.79 frames. ], giga_tot_loss[loss=0.3819, simple_loss=0.4227, pruned_loss=0.1705, over 5659718.52 frames. ], batch size: 368, lr: 1.15e-02, grad_scale: 2.0
+2023-03-01 11:31:17,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.111e+02 1.574e+03 1.976e+03 2.638e+03 7.227e+03, threshold=3.952e+03, percent-clipped=4.0
+2023-03-01 11:31:33,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1408, 1.1685, 0.9661, 0.9077], device='cuda:0'), covar=tensor([0.0308, 0.0321, 0.0253, 0.0359], device='cuda:0'), in_proj_covar=tensor([0.1130, 0.0831, 0.0925, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 11:31:45,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5306, 3.1951, 3.2635, 1.8109], device='cuda:0'), covar=tensor([0.0600, 0.0546, 0.0838, 0.1870], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0598, 0.0834, 0.0573], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 11:32:01,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7194, 1.1477, 3.4924, 2.9926], device='cuda:0'), covar=tensor([0.1352, 0.1544, 0.0364, 0.0529], device='cuda:0'), in_proj_covar=tensor([0.0525, 0.0498, 0.0677, 0.0535], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 11:32:03,421 INFO [train.py:968] (0/2) Epoch 2, batch 44400, giga_loss[loss=0.3247, simple_loss=0.4026, pruned_loss=0.1234, over 28747.00 frames. ], tot_loss[loss=0.3787, simple_loss=0.4228, pruned_loss=0.1673, over 5662456.76 frames. ], libri_tot_loss[loss=0.3816, simple_loss=0.4208, pruned_loss=0.1712, over 5654641.26 frames. ], giga_tot_loss[loss=0.3788, simple_loss=0.4232, pruned_loss=0.1672, over 5665656.40 frames. ], batch size: 119, lr: 1.15e-02, grad_scale: 4.0
+2023-03-01 11:32:09,062 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:32:14,064 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:32:15,921 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:32:41,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:32:48,222 INFO [train.py:968] (0/2) Epoch 2, batch 44450, giga_loss[loss=0.4254, simple_loss=0.4641, pruned_loss=0.1933, over 28948.00 frames. ], tot_loss[loss=0.3771, simple_loss=0.4237, pruned_loss=0.1652, over 5664538.97 frames. ], libri_tot_loss[loss=0.3824, simple_loss=0.4213, pruned_loss=0.1718, over 5658683.45 frames. ], giga_tot_loss[loss=0.3764, simple_loss=0.4235, pruned_loss=0.1646, over 5663573.14 frames. ], batch size: 227, lr: 1.15e-02, grad_scale: 4.0
+2023-03-01 11:32:50,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.925e+02 1.439e+03 1.809e+03 2.634e+03 5.343e+03, threshold=3.618e+03, percent-clipped=11.0
+2023-03-01 11:33:11,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:33:37,781 INFO [train.py:968] (0/2) Epoch 2, batch 44500, giga_loss[loss=0.316, simple_loss=0.3928, pruned_loss=0.1196, over 28737.00 frames. ], tot_loss[loss=0.3802, simple_loss=0.4262, pruned_loss=0.1672, over 5650145.83 frames. ], libri_tot_loss[loss=0.3827, simple_loss=0.4215, pruned_loss=0.1719, over 5652222.04 frames. ], giga_tot_loss[loss=0.3795, simple_loss=0.426, pruned_loss=0.1665, over 5654604.06 frames. ], batch size: 66, lr: 1.15e-02, grad_scale: 4.0
+2023-03-01 11:33:55,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9768, 1.0336, 0.8247, 0.6796], device='cuda:0'), covar=tensor([0.0309, 0.0308, 0.0273, 0.0378], device='cuda:0'), in_proj_covar=tensor([0.1137, 0.0826, 0.0915, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 11:34:01,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:34:15,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90207.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 11:34:24,668 INFO [train.py:968] (0/2) Epoch 2, batch 44550, giga_loss[loss=0.3592, simple_loss=0.4132, pruned_loss=0.1526, over 28893.00 frames. ], tot_loss[loss=0.3885, simple_loss=0.4312, pruned_loss=0.1729, over 5605826.51 frames. ], libri_tot_loss[loss=0.3839, simple_loss=0.4223, pruned_loss=0.1727, over 5601301.83 frames. ], giga_tot_loss[loss=0.3868, simple_loss=0.4305, pruned_loss=0.1716, over 5656536.09 frames. ], batch size: 186, lr: 1.15e-02, grad_scale: 2.0
+2023-03-01 11:34:26,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.582e+02 1.757e+03 2.365e+03 3.154e+03 7.286e+03, threshold=4.731e+03, percent-clipped=11.0
+2023-03-01 11:34:51,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4370, 1.3943, 1.2139, 1.4926], device='cuda:0'), covar=tensor([0.1902, 0.1984, 0.1779, 0.2029], device='cuda:0'), in_proj_covar=tensor([0.0994, 0.0820, 0.0902, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 11:35:03,140 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:35:15,556 INFO [train.py:968] (0/2) Epoch 2, batch 44600, libri_loss[loss=0.4437, simple_loss=0.4757, pruned_loss=0.2059, over 19855.00 frames. ], tot_loss[loss=0.3913, simple_loss=0.4326, pruned_loss=0.175, over 5561687.29 frames. ], libri_tot_loss[loss=0.3851, simple_loss=0.4232, pruned_loss=0.1735, over 5542501.34 frames. ], giga_tot_loss[loss=0.3889, simple_loss=0.4314, pruned_loss=0.1732, over 5654972.47 frames. ], batch size: 187, lr: 1.15e-02, grad_scale: 2.0
+2023-03-01 11:35:21,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:35:29,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:35:31,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:35:35,644 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-01 11:35:36,958 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-2.pt
+2023-03-01 11:36:14,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7141, 2.1079, 1.8706, 1.7631], device='cuda:0'), covar=tensor([0.1539, 0.1754, 0.1225, 0.0949], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0805, 0.0720, 0.0617], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 11:36:28,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-01 11:36:32,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:36:37,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.785e+02 1.458e+03 1.987e+03 2.745e+03 7.013e+03, threshold=3.975e+03, percent-clipped=5.0
+2023-03-01 11:36:52,194 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:36:54,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:36:55,100 INFO [train.py:968] (0/2) Epoch 3, batch 50, giga_loss[loss=0.337, simple_loss=0.4075, pruned_loss=0.1333, over 28695.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.4177, pruned_loss=0.1496, over 1262644.29 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3844, pruned_loss=0.1339, over 170041.47 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.4224, pruned_loss=0.1519, over 1125835.44 frames. ], batch size: 242, lr: 1.10e-02, grad_scale: 2.0
+2023-03-01 11:37:06,134 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90350.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 11:37:09,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90353.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 11:37:22,696 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:37:38,088 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90382.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 11:37:45,756 INFO [train.py:968] (0/2) Epoch 3, batch 100, giga_loss[loss=0.3703, simple_loss=0.4167, pruned_loss=0.1619, over 28710.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.4068, pruned_loss=0.1427, over 2245953.57 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3821, pruned_loss=0.1302, over 255709.37 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4094, pruned_loss=0.1441, over 2082669.07 frames. ], batch size: 92, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:38:13,443 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:38:14,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.291e+02 1.195e+03 1.480e+03 1.904e+03 4.567e+03, threshold=2.959e+03, percent-clipped=3.0
+2023-03-01 11:38:15,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:38:34,139 INFO [train.py:968] (0/2) Epoch 3, batch 150, giga_loss[loss=0.3242, simple_loss=0.3698, pruned_loss=0.1393, over 27651.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3925, pruned_loss=0.1361, over 3008663.39 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3803, pruned_loss=0.1282, over 367362.49 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3941, pruned_loss=0.1372, over 2821309.81 frames. ], batch size: 472, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:38:38,486 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-01 11:38:38,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:38:42,943 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:38:59,578 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2344, 2.0892, 1.2276, 1.1586], device='cuda:0'), covar=tensor([0.0896, 0.0440, 0.0822, 0.1337], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0451, 0.0322, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 11:39:13,492 INFO [train.py:968] (0/2) Epoch 3, batch 200, giga_loss[loss=0.345, simple_loss=0.3843, pruned_loss=0.1528, over 28294.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3777, pruned_loss=0.1281, over 3616228.41 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3819, pruned_loss=0.128, over 477771.37 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.378, pruned_loss=0.1285, over 3421120.30 frames. ], batch size: 368, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:39:41,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.657e+02 1.146e+03 1.446e+03 1.952e+03 4.391e+03, threshold=2.892e+03, percent-clipped=4.0
+2023-03-01 11:39:42,375 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:39:59,808 INFO [train.py:968] (0/2) Epoch 3, batch 250, giga_loss[loss=0.3034, simple_loss=0.3635, pruned_loss=0.1217, over 28830.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.364, pruned_loss=0.1202, over 4066219.45 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3841, pruned_loss=0.1289, over 518315.73 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3632, pruned_loss=0.12, over 3907320.81 frames. ], batch size: 174, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:40:38,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4795, 2.1501, 1.6724, 0.8957], device='cuda:0'), covar=tensor([0.1862, 0.1011, 0.1649, 0.1907], device='cuda:0'), in_proj_covar=tensor([0.1220, 0.1171, 0.1207, 0.1056], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 11:40:43,995 INFO [train.py:968] (0/2) Epoch 3, batch 300, giga_loss[loss=0.2689, simple_loss=0.3301, pruned_loss=0.1038, over 28979.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3536, pruned_loss=0.1147, over 4431594.71 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3868, pruned_loss=0.1308, over 597717.74 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3519, pruned_loss=0.1139, over 4285090.16 frames. ], batch size: 106, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:40:44,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:40:47,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:41:15,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.632e+02 1.038e+03 1.402e+03 1.870e+03 6.623e+03, threshold=2.804e+03, percent-clipped=8.0
+2023-03-01 11:41:17,472 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:41:18,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:41:25,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:41:31,014 INFO [train.py:968] (0/2) Epoch 3, batch 350, giga_loss[loss=0.2849, simple_loss=0.3374, pruned_loss=0.1162, over 28699.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3468, pruned_loss=0.111, over 4706665.32 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3904, pruned_loss=0.1329, over 719809.12 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3438, pruned_loss=0.1096, over 4568870.22 frames. ], batch size: 92, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:41:44,739 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-01 11:42:02,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7654, 3.3167, 3.4479, 1.5400], device='cuda:0'), covar=tensor([0.0578, 0.0480, 0.0862, 0.2006], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0578, 0.0799, 0.0563], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 11:42:16,463 INFO [train.py:968] (0/2) Epoch 3, batch 400, giga_loss[loss=0.2345, simple_loss=0.3031, pruned_loss=0.08294, over 28429.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3433, pruned_loss=0.1093, over 4932738.91 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3921, pruned_loss=0.134, over 898217.68 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3389, pruned_loss=0.1071, over 4788203.75 frames. ], batch size: 71, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:42:43,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.265e+02 1.053e+03 1.295e+03 1.700e+03 5.667e+03, threshold=2.589e+03, percent-clipped=8.0
+2023-03-01 11:43:01,408 INFO [train.py:968] (0/2) Epoch 3, batch 450, giga_loss[loss=0.2458, simple_loss=0.3136, pruned_loss=0.08894, over 28996.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.338, pruned_loss=0.1054, over 5110597.59 frames. ], libri_tot_loss[loss=0.3303, simple_loss=0.3931, pruned_loss=0.1337, over 972898.62 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3336, pruned_loss=0.1032, over 4984605.73 frames. ], batch size: 136, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:43:05,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:43:24,629 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 11:43:33,521 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:43:36,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:43:48,259 INFO [train.py:968] (0/2) Epoch 3, batch 500, giga_loss[loss=0.2532, simple_loss=0.3185, pruned_loss=0.0939, over 28386.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.335, pruned_loss=0.1038, over 5244168.13 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3953, pruned_loss=0.135, over 1021825.72 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3305, pruned_loss=0.1015, over 5138861.13 frames. ], batch size: 65, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:44:04,545 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:44:16,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.450e+02 9.750e+02 1.278e+03 1.581e+03 4.571e+03, threshold=2.557e+03, percent-clipped=5.0
+2023-03-01 11:44:35,020 INFO [train.py:968] (0/2) Epoch 3, batch 550, giga_loss[loss=0.3096, simple_loss=0.3664, pruned_loss=0.1264, over 27913.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3331, pruned_loss=0.1029, over 5350581.59 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.394, pruned_loss=0.1342, over 1119167.52 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3286, pruned_loss=0.1006, over 5255816.89 frames. ], batch size: 412, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:44:36,947 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 11:44:45,988 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:45:22,569 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:45:23,973 INFO [train.py:968] (0/2) Epoch 3, batch 600, giga_loss[loss=0.2911, simple_loss=0.3368, pruned_loss=0.1227, over 26635.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3306, pruned_loss=0.1017, over 5419233.08 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3945, pruned_loss=0.1339, over 1214184.40 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3257, pruned_loss=0.0993, over 5334707.54 frames. ], batch size: 555, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:45:32,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:45:32,672 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2634, 1.3513, 1.2230, 1.2211], device='cuda:0'), covar=tensor([0.1996, 0.1929, 0.1701, 0.1906], device='cuda:0'), in_proj_covar=tensor([0.1031, 0.0834, 0.0920, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 11:45:56,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.763e+02 9.272e+02 1.183e+03 1.662e+03 4.578e+03, threshold=2.366e+03, percent-clipped=8.0
+2023-03-01 11:46:11,867 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-01 11:46:15,517 INFO [train.py:968] (0/2) Epoch 3, batch 650, giga_loss[loss=0.2454, simple_loss=0.3079, pruned_loss=0.09148, over 28759.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3271, pruned_loss=0.0996, over 5475586.94 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3922, pruned_loss=0.1324, over 1259036.25 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3231, pruned_loss=0.09771, over 5406765.89 frames. ], batch size: 199, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:46:24,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4610, 2.1133, 1.7077, 0.6660], device='cuda:0'), covar=tensor([0.1728, 0.0927, 0.1292, 0.1963], device='cuda:0'), in_proj_covar=tensor([0.1225, 0.1165, 0.1227, 0.1057], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 11:46:59,452 INFO [train.py:968] (0/2) Epoch 3, batch 700, giga_loss[loss=0.2267, simple_loss=0.2961, pruned_loss=0.07867, over 28798.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3243, pruned_loss=0.09816, over 5532127.54 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3921, pruned_loss=0.1326, over 1351237.93 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3199, pruned_loss=0.09597, over 5470294.90 frames. ], batch size: 119, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:47:07,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:47:29,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.812e+02 1.082e+03 1.403e+03 2.090e+03 5.174e+03, threshold=2.806e+03, percent-clipped=16.0
+2023-03-01 11:47:45,960 INFO [train.py:968] (0/2) Epoch 3, batch 750, libri_loss[loss=0.3564, simple_loss=0.4166, pruned_loss=0.1481, over 29523.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3228, pruned_loss=0.09742, over 5577491.33 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3911, pruned_loss=0.1325, over 1507219.19 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3172, pruned_loss=0.0945, over 5517441.02 frames. ], batch size: 84, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:47:46,299 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:47:48,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:48:00,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6799, 1.9780, 1.8563, 1.8433], device='cuda:0'), covar=tensor([0.1569, 0.1788, 0.1216, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0809, 0.0741, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 11:48:13,450 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:48:31,775 INFO [train.py:968] (0/2) Epoch 3, batch 800, giga_loss[loss=0.299, simple_loss=0.3625, pruned_loss=0.1178, over 28528.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3228, pruned_loss=0.09805, over 5602997.31 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3911, pruned_loss=0.1327, over 1573225.26 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3173, pruned_loss=0.09519, over 5551034.60 frames. ], batch size: 336, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:48:58,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:49:00,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.977e+02 9.853e+02 1.354e+03 2.068e+03 7.165e+03, threshold=2.707e+03, percent-clipped=9.0
+2023-03-01 11:49:19,963 INFO [train.py:968] (0/2) Epoch 3, batch 850, giga_loss[loss=0.3517, simple_loss=0.4056, pruned_loss=0.1489, over 28620.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3363, pruned_loss=0.1064, over 5610429.34 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3891, pruned_loss=0.1314, over 1670433.90 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3312, pruned_loss=0.1039, over 5571235.11 frames. ], batch size: 242, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:49:22,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:49:25,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:49:52,717 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:50:09,323 INFO [train.py:968] (0/2) Epoch 3, batch 900, libri_loss[loss=0.3377, simple_loss=0.392, pruned_loss=0.1417, over 29545.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3524, pruned_loss=0.1158, over 5628484.03 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3895, pruned_loss=0.1315, over 1791632.50 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3471, pruned_loss=0.1132, over 5593113.52 frames. ], batch size: 80, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:50:36,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.157e+02 1.292e+03 1.647e+03 2.203e+03 3.824e+03, threshold=3.294e+03, percent-clipped=12.0
+2023-03-01 11:50:40,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:50:55,269 INFO [train.py:968] (0/2) Epoch 3, batch 950, giga_loss[loss=0.3615, simple_loss=0.4187, pruned_loss=0.1522, over 28651.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3641, pruned_loss=0.122, over 5647292.22 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3901, pruned_loss=0.1317, over 1873174.07 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3592, pruned_loss=0.1196, over 5614676.55 frames. ], batch size: 262, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:50:56,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=91242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:51:11,116 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:51:14,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:51:15,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:51:17,425 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6236, 2.1216, 1.2052, 1.1831], device='cuda:0'), covar=tensor([0.0839, 0.0519, 0.0587, 0.0691], device='cuda:0'), in_proj_covar=tensor([0.1136, 0.0838, 0.0911, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 11:51:24,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7505, 3.2817, 3.4706, 1.7637], device='cuda:0'), covar=tensor([0.0507, 0.0410, 0.0736, 0.1842], device='cuda:0'), in_proj_covar=tensor([0.0766, 0.0566, 0.0800, 0.0567], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 11:51:37,173 INFO [train.py:968] (0/2) Epoch 3, batch 1000, giga_loss[loss=0.3055, simple_loss=0.3742, pruned_loss=0.1184, over 28494.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3711, pruned_loss=0.1247, over 5652018.82 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3917, pruned_loss=0.1327, over 1939078.22 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3664, pruned_loss=0.1224, over 5634631.62 frames. ], batch size: 71, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:51:40,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:52:03,402 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.462e+02 1.167e+03 1.485e+03 2.194e+03 4.091e+03, threshold=2.970e+03, percent-clipped=8.0
+2023-03-01 11:52:24,920 INFO [train.py:968] (0/2) Epoch 3, batch 1050, giga_loss[loss=0.2809, simple_loss=0.3577, pruned_loss=0.102, over 28625.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3747, pruned_loss=0.1252, over 5646828.33 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3922, pruned_loss=0.1329, over 1987595.46 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3707, pruned_loss=0.1233, over 5638988.27 frames. ], batch size: 71, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:52:50,434 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:52:53,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:53:07,987 INFO [train.py:968] (0/2) Epoch 3, batch 1100, giga_loss[loss=0.305, simple_loss=0.3689, pruned_loss=0.1205, over 28964.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3757, pruned_loss=0.1249, over 5648329.96 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3917, pruned_loss=0.1328, over 2084169.91 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3723, pruned_loss=0.1232, over 5652882.29 frames. ], batch size: 136, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:53:15,671 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:53:22,407 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:53:24,340 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:53:36,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.236e+02 1.033e+03 1.360e+03 1.835e+03 3.321e+03, threshold=2.719e+03, percent-clipped=3.0
+2023-03-01 11:53:51,187 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:53:52,248 INFO [train.py:968] (0/2) Epoch 3, batch 1150, giga_loss[loss=0.3382, simple_loss=0.3896, pruned_loss=0.1434, over 28716.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3777, pruned_loss=0.1268, over 5651281.37 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3916, pruned_loss=0.1329, over 2149667.41 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3747, pruned_loss=0.1253, over 5660748.38 frames. ], batch size: 85, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:54:38,027 INFO [train.py:968] (0/2) Epoch 3, batch 1200, libri_loss[loss=0.3765, simple_loss=0.4311, pruned_loss=0.1609, over 29514.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3818, pruned_loss=0.1299, over 5648886.81 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3923, pruned_loss=0.1333, over 2249722.64 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3787, pruned_loss=0.1284, over 5660294.25 frames. ], batch size: 82, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:55:04,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.089e+02 1.177e+03 1.413e+03 2.039e+03 3.934e+03, threshold=2.825e+03, percent-clipped=11.0
+2023-03-01 11:55:17,377 INFO [train.py:968] (0/2) Epoch 3, batch 1250, giga_loss[loss=0.326, simple_loss=0.3903, pruned_loss=0.1308, over 28906.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3836, pruned_loss=0.1306, over 5663921.43 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3919, pruned_loss=0.1328, over 2393288.61 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3809, pruned_loss=0.1295, over 5664348.36 frames. ], batch size: 186, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:55:25,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1149, 1.2472, 1.1420, 1.1959], device='cuda:0'), covar=tensor([0.0867, 0.0924, 0.1363, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0792, 0.0633, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 11:55:50,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2738, 1.6254, 1.3291, 1.3481], device='cuda:0'), covar=tensor([0.0942, 0.0432, 0.0412, 0.1060], device='cuda:0'), in_proj_covar=tensor([0.0259, 0.0168, 0.0171, 0.0309], device='cuda:0'), out_proj_covar=tensor([0.0037, 0.0028, 0.0024, 0.0042], device='cuda:0')
+2023-03-01 11:56:03,287 INFO [train.py:968] (0/2) Epoch 3, batch 1300, giga_loss[loss=0.334, simple_loss=0.4001, pruned_loss=0.134, over 28935.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3865, pruned_loss=0.1314, over 5674994.97 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3916, pruned_loss=0.1322, over 2478521.88 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3844, pruned_loss=0.1307, over 5671540.00 frames. ], batch size: 145, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:56:23,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:56:28,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.159e+02 1.158e+03 1.377e+03 1.795e+03 3.979e+03, threshold=2.755e+03, percent-clipped=4.0
+2023-03-01 11:56:45,086 INFO [train.py:968] (0/2) Epoch 3, batch 1350, giga_loss[loss=0.3003, simple_loss=0.3728, pruned_loss=0.1139, over 28757.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3887, pruned_loss=0.1317, over 5687257.25 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3917, pruned_loss=0.1322, over 2563901.79 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3868, pruned_loss=0.1311, over 5679112.36 frames. ], batch size: 99, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:57:15,803 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=91678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:57:26,420 INFO [train.py:968] (0/2) Epoch 3, batch 1400, giga_loss[loss=0.332, simple_loss=0.4002, pruned_loss=0.1319, over 28939.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3901, pruned_loss=0.1319, over 5687159.62 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.392, pruned_loss=0.1323, over 2637755.95 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3884, pruned_loss=0.1314, over 5683280.76 frames. ], batch size: 164, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:57:56,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.438e+02 1.160e+03 1.413e+03 1.908e+03 4.502e+03, threshold=2.826e+03, percent-clipped=7.0
+2023-03-01 11:58:10,074 INFO [train.py:968] (0/2) Epoch 3, batch 1450, libri_loss[loss=0.4023, simple_loss=0.4555, pruned_loss=0.1746, over 29535.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.39, pruned_loss=0.1307, over 5696700.00 frames. ], libri_tot_loss[loss=0.3299, simple_loss=0.3936, pruned_loss=0.1331, over 2702187.49 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3879, pruned_loss=0.1299, over 5689962.95 frames. ], batch size: 89, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:58:25,387 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:58:28,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:58:50,448 INFO [train.py:968] (0/2) Epoch 3, batch 1500, giga_loss[loss=0.3327, simple_loss=0.4006, pruned_loss=0.1324, over 28938.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3877, pruned_loss=0.1283, over 5698743.88 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3931, pruned_loss=0.1328, over 2764961.75 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3862, pruned_loss=0.1277, over 5690313.35 frames. ], batch size: 227, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:58:51,380 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:59:17,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.489e+02 1.086e+03 1.369e+03 1.861e+03 5.023e+03, threshold=2.739e+03, percent-clipped=9.0
+2023-03-01 11:59:34,916 INFO [train.py:968] (0/2) Epoch 3, batch 1550, giga_loss[loss=0.3302, simple_loss=0.3937, pruned_loss=0.1334, over 28800.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3859, pruned_loss=0.127, over 5702961.19 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3932, pruned_loss=0.1328, over 2810807.27 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3845, pruned_loss=0.1265, over 5694533.54 frames. ], batch size: 186, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 12:00:08,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5120, 2.4947, 1.5406, 1.2515], device='cuda:0'), covar=tensor([0.0980, 0.0517, 0.0808, 0.1536], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0447, 0.0317, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 12:00:19,722 INFO [train.py:968] (0/2) Epoch 3, batch 1600, giga_loss[loss=0.332, simple_loss=0.3838, pruned_loss=0.1401, over 28972.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3891, pruned_loss=0.1323, over 5707849.39 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3934, pruned_loss=0.133, over 2841696.66 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3879, pruned_loss=0.1318, over 5699520.75 frames. ], batch size: 106, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 12:00:54,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.970e+02 1.253e+03 1.533e+03 1.992e+03 6.330e+03, threshold=3.067e+03, percent-clipped=12.0
+2023-03-01 12:01:03,750 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=91935.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:01:08,427 INFO [train.py:968] (0/2) Epoch 3, batch 1650, giga_loss[loss=0.3567, simple_loss=0.4079, pruned_loss=0.1528, over 28976.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3925, pruned_loss=0.1375, over 5703403.68 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3945, pruned_loss=0.1338, over 2917088.55 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.391, pruned_loss=0.1368, over 5692492.54 frames. ], batch size: 136, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 12:01:49,311 INFO [train.py:968] (0/2) Epoch 3, batch 1700, giga_loss[loss=0.2847, simple_loss=0.3464, pruned_loss=0.1115, over 28636.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3936, pruned_loss=0.1399, over 5686256.10 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.395, pruned_loss=0.1345, over 3042128.81 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.392, pruned_loss=0.1391, over 5687266.75 frames. ], batch size: 85, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 12:01:57,500 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-92000.pt
+2023-03-01 12:02:21,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.640e+02 1.310e+03 1.634e+03 2.107e+03 5.724e+03, threshold=3.268e+03, percent-clipped=10.0
+2023-03-01 12:02:35,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1533, 1.6482, 1.2484, 1.3827], device='cuda:0'), covar=tensor([0.0961, 0.0424, 0.0431, 0.1032], device='cuda:0'), in_proj_covar=tensor([0.0256, 0.0167, 0.0170, 0.0304], device='cuda:0'), out_proj_covar=tensor([0.0037, 0.0027, 0.0024, 0.0041], device='cuda:0')
+2023-03-01 12:02:36,738 INFO [train.py:968] (0/2) Epoch 3, batch 1750, libri_loss[loss=0.4492, simple_loss=0.4756, pruned_loss=0.2114, over 29256.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3916, pruned_loss=0.139, over 5699093.04 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3954, pruned_loss=0.135, over 3098795.27 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3901, pruned_loss=0.1382, over 5696222.88 frames. ], batch size: 94, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 12:02:46,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:03:18,072 INFO [train.py:968] (0/2) Epoch 3, batch 1800, giga_loss[loss=0.3365, simple_loss=0.3986, pruned_loss=0.1372, over 28635.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3905, pruned_loss=0.1385, over 5697547.11 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3949, pruned_loss=0.1348, over 3198328.96 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3894, pruned_loss=0.1381, over 5697288.61 frames. ], batch size: 307, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:03:40,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-01 12:03:44,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.141e+02 1.160e+03 1.525e+03 1.927e+03 4.179e+03, threshold=3.051e+03, percent-clipped=6.0
+2023-03-01 12:03:55,687 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 12:04:01,575 INFO [train.py:968] (0/2) Epoch 3, batch 1850, giga_loss[loss=0.2816, simple_loss=0.3553, pruned_loss=0.1039, over 28846.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3892, pruned_loss=0.1359, over 5710254.13 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.395, pruned_loss=0.1345, over 3252320.41 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3882, pruned_loss=0.1358, over 5706167.12 frames. ], batch size: 145, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:04:37,690 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 12:04:51,127 INFO [train.py:968] (0/2) Epoch 3, batch 1900, giga_loss[loss=0.33, simple_loss=0.3849, pruned_loss=0.1376, over 28548.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3861, pruned_loss=0.1335, over 5699137.67 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3944, pruned_loss=0.1342, over 3289616.85 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3855, pruned_loss=0.1336, over 5695335.91 frames. ], batch size: 336, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:04:57,454 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:04:59,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:05:22,742 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.906e+02 1.139e+03 1.525e+03 1.888e+03 3.683e+03, threshold=3.051e+03, percent-clipped=4.0
+2023-03-01 12:05:26,998 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:05:38,287 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3452, 3.0756, 3.0544, 1.8010], device='cuda:0'), covar=tensor([0.0515, 0.0437, 0.0766, 0.1631], device='cuda:0'), in_proj_covar=tensor([0.0749, 0.0560, 0.0779, 0.0560], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 12:05:38,664 INFO [train.py:968] (0/2) Epoch 3, batch 1950, giga_loss[loss=0.281, simple_loss=0.3261, pruned_loss=0.1179, over 23583.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3803, pruned_loss=0.1297, over 5690855.80 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3942, pruned_loss=0.1341, over 3352283.93 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3797, pruned_loss=0.1298, over 5684956.77 frames. ], batch size: 705, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:05:51,506 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0190, 2.6932, 2.2349, 2.1984], device='cuda:0'), covar=tensor([0.1449, 0.1387, 0.1071, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0802, 0.0734, 0.0627], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 12:05:54,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8358, 2.2864, 2.3315, 2.2705], device='cuda:0'), covar=tensor([0.0894, 0.1592, 0.1204, 0.1150], device='cuda:0'), in_proj_covar=tensor([0.0454, 0.0778, 0.0626, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 12:06:20,104 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=92284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:06:25,836 INFO [train.py:968] (0/2) Epoch 3, batch 2000, giga_loss[loss=0.2969, simple_loss=0.3579, pruned_loss=0.118, over 28558.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3729, pruned_loss=0.125, over 5686692.44 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3935, pruned_loss=0.1336, over 3399279.88 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3724, pruned_loss=0.1251, over 5681457.76 frames. ], batch size: 307, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:06:43,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92310.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:06:57,428 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 12:06:58,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.416e+02 9.757e+02 1.261e+03 1.842e+03 3.603e+03, threshold=2.521e+03, percent-clipped=2.0
+2023-03-01 12:07:03,234 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-01 12:07:13,120 INFO [train.py:968] (0/2) Epoch 3, batch 2050, giga_loss[loss=0.3191, simple_loss=0.3703, pruned_loss=0.1339, over 28725.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3674, pruned_loss=0.122, over 5662493.12 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3932, pruned_loss=0.1334, over 3440786.57 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3666, pruned_loss=0.122, over 5672728.46 frames. ], batch size: 92, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:07:19,580 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=92348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:07:43,346 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=92374.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:07:55,978 INFO [train.py:968] (0/2) Epoch 3, batch 2100, giga_loss[loss=0.3044, simple_loss=0.3607, pruned_loss=0.124, over 27643.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3675, pruned_loss=0.1215, over 5675541.55 frames. ], libri_tot_loss[loss=0.3303, simple_loss=0.3935, pruned_loss=0.1335, over 3487578.48 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3663, pruned_loss=0.1213, over 5681256.45 frames. ], batch size: 472, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:08:24,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.535e+02 9.972e+02 1.467e+03 2.009e+03 5.743e+03, threshold=2.934e+03, percent-clipped=15.0
+2023-03-01 12:08:36,033 INFO [train.py:968] (0/2) Epoch 3, batch 2150, giga_loss[loss=0.3287, simple_loss=0.3801, pruned_loss=0.1386, over 28901.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3697, pruned_loss=0.1229, over 5688872.35 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3951, pruned_loss=0.1345, over 3558662.45 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.367, pruned_loss=0.1218, over 5687844.41 frames. ], batch size: 186, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:08:38,477 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-01 12:08:43,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3361, 2.2707, 1.2153, 1.2306], device='cuda:0'), covar=tensor([0.1052, 0.0598, 0.1078, 0.1620], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0444, 0.0321, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 12:08:45,061 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92453.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:08:47,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92456.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:09:12,059 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92485.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:09:16,596 INFO [train.py:968] (0/2) Epoch 3, batch 2200, giga_loss[loss=0.2792, simple_loss=0.3448, pruned_loss=0.1068, over 28663.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3689, pruned_loss=0.1221, over 5691227.89 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3954, pruned_loss=0.1345, over 3664121.66 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3655, pruned_loss=0.1207, over 5689361.56 frames. ], batch size: 85, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:09:38,506 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9601, 1.2933, 0.9935, 0.2772], device='cuda:0'), covar=tensor([0.1256, 0.1054, 0.1919, 0.1973], device='cuda:0'), in_proj_covar=tensor([0.1236, 0.1155, 0.1218, 0.1050], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 12:09:47,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.454e+02 1.092e+03 1.356e+03 2.273e+03 7.390e+03, threshold=2.712e+03, percent-clipped=15.0
+2023-03-01 12:09:59,519 INFO [train.py:968] (0/2) Epoch 3, batch 2250, giga_loss[loss=0.2977, simple_loss=0.3637, pruned_loss=0.1158, over 28668.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3656, pruned_loss=0.1204, over 5702145.07 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3961, pruned_loss=0.1348, over 3697110.23 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3621, pruned_loss=0.1189, over 5698199.93 frames. ], batch size: 307, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:10:28,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4440, 1.3667, 1.4400, 1.3762], device='cuda:0'), covar=tensor([0.0977, 0.1383, 0.1397, 0.1208], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0792, 0.0642, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 12:10:32,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6482, 4.1364, 2.2163, 2.2558], device='cuda:0'), covar=tensor([0.0643, 0.0261, 0.0709, 0.1123], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0438, 0.0317, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0018, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 12:10:41,522 INFO [train.py:968] (0/2) Epoch 3, batch 2300, giga_loss[loss=0.332, simple_loss=0.3867, pruned_loss=0.1386, over 28590.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3622, pruned_loss=0.1184, over 5706598.36 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3955, pruned_loss=0.1342, over 3751606.73 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.359, pruned_loss=0.1172, over 5699214.79 frames. ], batch size: 307, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:11:05,815 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-01 12:11:12,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.721e+02 1.006e+03 1.261e+03 1.719e+03 4.591e+03, threshold=2.522e+03, percent-clipped=4.0
+2023-03-01 12:11:23,356 INFO [train.py:968] (0/2) Epoch 3, batch 2350, giga_loss[loss=0.2426, simple_loss=0.3123, pruned_loss=0.08638, over 28709.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3609, pruned_loss=0.1179, over 5718442.31 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3961, pruned_loss=0.1346, over 3825738.47 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3567, pruned_loss=0.1161, over 5706428.26 frames. ], batch size: 66, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:11:39,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:12:05,025 INFO [train.py:968] (0/2) Epoch 3, batch 2400, giga_loss[loss=0.2395, simple_loss=0.3058, pruned_loss=0.08663, over 28921.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3568, pruned_loss=0.1156, over 5725178.82 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3962, pruned_loss=0.1347, over 3855163.25 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.353, pruned_loss=0.1139, over 5714686.52 frames. ], batch size: 106, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:12:28,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:12:30,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.195e+02 1.020e+03 1.220e+03 1.727e+03 1.156e+04, threshold=2.439e+03, percent-clipped=14.0
+2023-03-01 12:12:41,260 INFO [train.py:968] (0/2) Epoch 3, batch 2450, giga_loss[loss=0.2509, simple_loss=0.319, pruned_loss=0.09141, over 29158.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3541, pruned_loss=0.1136, over 5728901.47 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3954, pruned_loss=0.1338, over 3929208.66 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3501, pruned_loss=0.1121, over 5720061.35 frames. ], batch size: 113, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:12:47,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92749.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:13:19,588 INFO [train.py:968] (0/2) Epoch 3, batch 2500, libri_loss[loss=0.3253, simple_loss=0.4028, pruned_loss=0.1239, over 29657.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3517, pruned_loss=0.112, over 5727400.50 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.396, pruned_loss=0.1338, over 3978429.92 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3471, pruned_loss=0.1103, over 5716081.51 frames. ], batch size: 88, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:13:28,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:13:32,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:13:49,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.735e+02 9.743e+02 1.258e+03 1.515e+03 4.470e+03, threshold=2.516e+03, percent-clipped=10.0
+2023-03-01 12:13:56,782 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:14:02,893 INFO [train.py:968] (0/2) Epoch 3, batch 2550, giga_loss[loss=0.274, simple_loss=0.3308, pruned_loss=0.1086, over 28477.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.35, pruned_loss=0.1112, over 5723809.13 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3967, pruned_loss=0.1343, over 4005848.15 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3452, pruned_loss=0.1092, over 5713627.49 frames. ], batch size: 71, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:14:22,795 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:14:24,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:14:33,034 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-01 12:14:41,783 INFO [train.py:968] (0/2) Epoch 3, batch 2600, giga_loss[loss=0.2795, simple_loss=0.3376, pruned_loss=0.1107, over 28885.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3494, pruned_loss=0.1105, over 5731476.15 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3976, pruned_loss=0.1346, over 4062046.76 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3437, pruned_loss=0.1081, over 5718589.48 frames. ], batch size: 112, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:14:42,648 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92892.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:14:44,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92895.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:14:47,974 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:15:08,740 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92924.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:15:09,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.882e+02 9.839e+02 1.217e+03 1.680e+03 4.864e+03, threshold=2.434e+03, percent-clipped=14.0
+2023-03-01 12:15:21,598 INFO [train.py:968] (0/2) Epoch 3, batch 2650, giga_loss[loss=0.3799, simple_loss=0.4185, pruned_loss=0.1706, over 28564.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.349, pruned_loss=0.1108, over 5729396.67 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3971, pruned_loss=0.1342, over 4106563.46 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3437, pruned_loss=0.1085, over 5716041.39 frames. ], batch size: 336, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:16:09,099 INFO [train.py:968] (0/2) Epoch 3, batch 2700, giga_loss[loss=0.3235, simple_loss=0.3813, pruned_loss=0.1329, over 28779.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3528, pruned_loss=0.1137, over 5724271.34 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3972, pruned_loss=0.1344, over 4124605.45 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3481, pruned_loss=0.1116, over 5712127.24 frames. ], batch size: 284, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:16:32,571 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:16:38,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.980e+02 1.081e+03 1.337e+03 1.890e+03 4.873e+03, threshold=2.674e+03, percent-clipped=13.0
+2023-03-01 12:16:53,897 INFO [train.py:968] (0/2) Epoch 3, batch 2750, giga_loss[loss=0.2857, simple_loss=0.3519, pruned_loss=0.1098, over 28373.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3607, pruned_loss=0.1194, over 5710276.44 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3979, pruned_loss=0.1347, over 4147457.54 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.356, pruned_loss=0.1173, over 5701263.83 frames. ], batch size: 78, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:17:40,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4163, 2.4411, 1.4463, 1.2565], device='cuda:0'), covar=tensor([0.0689, 0.0409, 0.0640, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0429, 0.0309, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0017], device='cuda:0')
+2023-03-01 12:17:45,501 INFO [train.py:968] (0/2) Epoch 3, batch 2800, giga_loss[loss=0.3401, simple_loss=0.3968, pruned_loss=0.1417, over 28801.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3693, pruned_loss=0.1259, over 5701028.35 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3973, pruned_loss=0.1342, over 4173651.82 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3656, pruned_loss=0.1244, over 5691235.18 frames. ], batch size: 186, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:18:05,963 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:18:14,439 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 12:18:16,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.880e+02 1.163e+03 1.366e+03 1.834e+03 4.889e+03, threshold=2.733e+03, percent-clipped=9.0
+2023-03-01 12:18:31,079 INFO [train.py:968] (0/2) Epoch 3, batch 2850, giga_loss[loss=0.3183, simple_loss=0.3901, pruned_loss=0.1233, over 28737.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3755, pruned_loss=0.129, over 5693067.57 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3968, pruned_loss=0.1339, over 4199706.52 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3726, pruned_loss=0.1279, over 5682246.67 frames. ], batch size: 262, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:19:18,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7803, 3.2400, 1.6699, 1.3877], device='cuda:0'), covar=tensor([0.0822, 0.0392, 0.0836, 0.1484], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0433, 0.0309, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0017], device='cuda:0')
+2023-03-01 12:19:20,921 INFO [train.py:968] (0/2) Epoch 3, batch 2900, giga_loss[loss=0.3394, simple_loss=0.399, pruned_loss=0.1399, over 28729.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3828, pruned_loss=0.1333, over 5675243.43 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3967, pruned_loss=0.134, over 4223496.57 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3803, pruned_loss=0.1324, over 5664965.83 frames. ], batch size: 242, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:19:50,198 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:19:52,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.110e+02 1.113e+03 1.613e+03 2.131e+03 5.006e+03, threshold=3.226e+03, percent-clipped=12.0
+2023-03-01 12:20:07,497 INFO [train.py:968] (0/2) Epoch 3, batch 2950, giga_loss[loss=0.3329, simple_loss=0.3931, pruned_loss=0.1363, over 28794.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3884, pruned_loss=0.1359, over 5694755.87 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3967, pruned_loss=0.1342, over 4265149.96 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3861, pruned_loss=0.1351, over 5681679.09 frames. ], batch size: 119, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:20:51,040 INFO [train.py:968] (0/2) Epoch 3, batch 3000, giga_loss[loss=0.2625, simple_loss=0.3377, pruned_loss=0.09367, over 28919.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3902, pruned_loss=0.1378, over 5674084.34 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3972, pruned_loss=0.1346, over 4279300.99 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.388, pruned_loss=0.1369, over 5670087.09 frames. ], batch size: 145, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:20:51,047 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 12:20:55,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3517, 1.3774, 1.4864, 1.3768], device='cuda:0'), covar=tensor([0.1084, 0.1329, 0.1484, 0.1223], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0791, 0.0639, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 12:20:59,912 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2748, simple_loss=0.3677, pruned_loss=0.0909, over 944034.00 frames. 
+2023-03-01 12:20:59,912 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19537MB
+2023-03-01 12:21:00,843 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:21:30,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:21:34,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.518e+02 1.088e+03 1.492e+03 1.932e+03 5.672e+03, threshold=2.983e+03, percent-clipped=2.0
+2023-03-01 12:21:51,222 INFO [train.py:968] (0/2) Epoch 3, batch 3050, giga_loss[loss=0.2426, simple_loss=0.326, pruned_loss=0.07957, over 28976.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3823, pruned_loss=0.1314, over 5681747.44 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3972, pruned_loss=0.1346, over 4295507.13 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3804, pruned_loss=0.1307, over 5676685.63 frames. ], batch size: 136, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:22:12,597 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:22:32,772 INFO [train.py:968] (0/2) Epoch 3, batch 3100, giga_loss[loss=0.312, simple_loss=0.3721, pruned_loss=0.1259, over 28528.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3802, pruned_loss=0.1294, over 5675665.42 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3969, pruned_loss=0.1346, over 4348862.20 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3783, pruned_loss=0.1287, over 5673283.41 frames. ], batch size: 71, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:22:36,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:22:46,349 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:22:57,127 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 12:23:06,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.073e+02 1.195e+03 1.465e+03 1.931e+03 4.054e+03, threshold=2.929e+03, percent-clipped=5.0
+2023-03-01 12:23:08,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:23:21,476 INFO [train.py:968] (0/2) Epoch 3, batch 3150, giga_loss[loss=0.2908, simple_loss=0.3623, pruned_loss=0.1096, over 28958.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3817, pruned_loss=0.1303, over 5672662.87 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3971, pruned_loss=0.1349, over 4364491.10 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.38, pruned_loss=0.1295, over 5668571.50 frames. ], batch size: 227, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:24:02,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8341, 1.7709, 1.7477, 1.7759], device='cuda:0'), covar=tensor([0.1025, 0.1497, 0.1258, 0.1168], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0784, 0.0634, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 12:24:03,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:24:05,003 INFO [train.py:968] (0/2) Epoch 3, batch 3200, libri_loss[loss=0.2888, simple_loss=0.3531, pruned_loss=0.1122, over 29510.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3844, pruned_loss=0.1318, over 5680223.43 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3965, pruned_loss=0.1345, over 4394939.91 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3832, pruned_loss=0.1314, over 5672785.61 frames. ], batch size: 70, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:24:18,706 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:24:36,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.875e+02 1.293e+03 1.630e+03 2.364e+03 5.224e+03, threshold=3.261e+03, percent-clipped=15.0
+2023-03-01 12:24:43,992 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:24:46,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:24:48,533 INFO [train.py:968] (0/2) Epoch 3, batch 3250, giga_loss[loss=0.3873, simple_loss=0.4302, pruned_loss=0.1722, over 27646.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3875, pruned_loss=0.1339, over 5692214.13 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3959, pruned_loss=0.1339, over 4454720.00 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3866, pruned_loss=0.1339, over 5677837.78 frames. ], batch size: 472, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:25:13,732 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:25:34,999 INFO [train.py:968] (0/2) Epoch 3, batch 3300, giga_loss[loss=0.3174, simple_loss=0.378, pruned_loss=0.1284, over 28173.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3901, pruned_loss=0.1364, over 5688521.97 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3956, pruned_loss=0.1336, over 4473036.11 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3894, pruned_loss=0.1366, over 5677402.26 frames. ], batch size: 77, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:25:41,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:25:48,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2802, 2.5360, 1.5079, 1.2390], device='cuda:0'), covar=tensor([0.0536, 0.0345, 0.0439, 0.0693], device='cuda:0'), in_proj_covar=tensor([0.1075, 0.0794, 0.0859, 0.0928], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 12:25:58,106 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-01 12:26:03,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.450e+02 1.239e+03 1.650e+03 2.184e+03 4.288e+03, threshold=3.301e+03, percent-clipped=4.0
+2023-03-01 12:26:06,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:26:09,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:26:16,021 INFO [train.py:968] (0/2) Epoch 3, batch 3350, giga_loss[loss=0.3112, simple_loss=0.3757, pruned_loss=0.1234, over 28512.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3896, pruned_loss=0.1357, over 5694612.20 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3956, pruned_loss=0.1335, over 4513204.96 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3889, pruned_loss=0.1361, over 5681954.58 frames. ], batch size: 60, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:26:36,187 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:26:39,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:27:01,860 INFO [train.py:968] (0/2) Epoch 3, batch 3400, giga_loss[loss=0.3485, simple_loss=0.404, pruned_loss=0.1465, over 28690.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3901, pruned_loss=0.1369, over 5687761.86 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3958, pruned_loss=0.1336, over 4540186.93 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3893, pruned_loss=0.1371, over 5674449.59 frames. ], batch size: 262, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:27:08,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:27:15,924 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 12:27:32,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.552e+02 1.162e+03 1.426e+03 1.924e+03 4.487e+03, threshold=2.853e+03, percent-clipped=1.0
+2023-03-01 12:27:42,590 INFO [train.py:968] (0/2) Epoch 3, batch 3450, giga_loss[loss=0.3237, simple_loss=0.3884, pruned_loss=0.1295, over 28439.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3904, pruned_loss=0.1359, over 5692364.45 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.396, pruned_loss=0.1336, over 4560611.70 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3895, pruned_loss=0.1361, over 5679024.78 frames. ], batch size: 78, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:27:43,509 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:27:44,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:27:45,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:27:56,233 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-01 12:28:09,594 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:28:12,591 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:28:23,129 INFO [train.py:968] (0/2) Epoch 3, batch 3500, giga_loss[loss=0.2758, simple_loss=0.3573, pruned_loss=0.09717, over 28821.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3902, pruned_loss=0.1343, over 5702941.62 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.396, pruned_loss=0.1336, over 4606826.47 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3893, pruned_loss=0.1345, over 5685991.02 frames. ], batch size: 66, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:28:32,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5131, 2.5102, 1.5076, 1.2615], device='cuda:0'), covar=tensor([0.0842, 0.0411, 0.0810, 0.1397], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0441, 0.0313, 0.0402], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0018, 0.0012, 0.0017], device='cuda:0')
+2023-03-01 12:28:34,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:28:40,153 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:28:42,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:28:57,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.292e+02 1.142e+03 1.388e+03 1.841e+03 9.698e+03, threshold=2.775e+03, percent-clipped=6.0
+2023-03-01 12:29:08,026 INFO [train.py:968] (0/2) Epoch 3, batch 3550, giga_loss[loss=0.3388, simple_loss=0.3973, pruned_loss=0.1402, over 29006.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3911, pruned_loss=0.134, over 5705222.52 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3964, pruned_loss=0.134, over 4645812.36 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3901, pruned_loss=0.1339, over 5685839.35 frames. ], batch size: 136, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:29:08,830 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:08,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:11,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:21,265 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 12:29:21,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9412, 2.2927, 2.5335, 2.3879], device='cuda:0'), covar=tensor([0.0745, 0.1602, 0.0999, 0.1167], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0788, 0.0637, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 12:29:37,031 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:43,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:44,961 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:47,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:48,916 INFO [train.py:968] (0/2) Epoch 3, batch 3600, giga_loss[loss=0.3056, simple_loss=0.3697, pruned_loss=0.1208, over 28707.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3891, pruned_loss=0.1321, over 5709042.42 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3962, pruned_loss=0.1338, over 4662804.11 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3883, pruned_loss=0.1321, over 5692633.10 frames. ], batch size: 242, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:30:12,564 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:14,735 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:16,891 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:20,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.059e+02 9.752e+02 1.260e+03 1.533e+03 2.414e+03, threshold=2.521e+03, percent-clipped=0.0
+2023-03-01 12:30:32,297 INFO [train.py:968] (0/2) Epoch 3, batch 3650, giga_loss[loss=0.32, simple_loss=0.3808, pruned_loss=0.1296, over 28597.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3864, pruned_loss=0.1311, over 5700279.96 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3961, pruned_loss=0.134, over 4685854.61 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3857, pruned_loss=0.131, over 5685378.49 frames. ], batch size: 336, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:30:37,099 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:38,919 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:41,072 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:43,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0354, 1.1069, 0.8897, 0.5761], device='cuda:0'), covar=tensor([0.0467, 0.0460, 0.0377, 0.0507], device='cuda:0'), in_proj_covar=tensor([0.1102, 0.0817, 0.0879, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 12:31:03,733 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:31:12,692 INFO [train.py:968] (0/2) Epoch 3, batch 3700, giga_loss[loss=0.263, simple_loss=0.3441, pruned_loss=0.09091, over 28493.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3838, pruned_loss=0.1288, over 5706885.58 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3968, pruned_loss=0.1342, over 4711877.85 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3825, pruned_loss=0.1285, over 5694097.94 frames. ], batch size: 71, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:31:19,126 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-94000.pt
+2023-03-01 12:31:28,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1681, 1.9174, 1.1732, 1.1153], device='cuda:0'), covar=tensor([0.0955, 0.0499, 0.0870, 0.1474], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0435, 0.0312, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0018, 0.0012, 0.0017], device='cuda:0')
+2023-03-01 12:31:41,886 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:31:42,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.631e+02 1.007e+03 1.297e+03 1.846e+03 6.365e+03, threshold=2.594e+03, percent-clipped=8.0
+2023-03-01 12:31:45,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:31:57,761 INFO [train.py:968] (0/2) Epoch 3, batch 3750, libri_loss[loss=0.3532, simple_loss=0.4168, pruned_loss=0.1447, over 29463.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3835, pruned_loss=0.1292, over 5703505.05 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3968, pruned_loss=0.1341, over 4735385.03 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3822, pruned_loss=0.1289, over 5689897.58 frames. ], batch size: 85, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:32:11,592 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:32:38,979 INFO [train.py:968] (0/2) Epoch 3, batch 3800, giga_loss[loss=0.3246, simple_loss=0.3892, pruned_loss=0.13, over 28509.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3857, pruned_loss=0.1308, over 5706977.29 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.397, pruned_loss=0.1342, over 4753093.26 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3843, pruned_loss=0.1305, over 5693159.32 frames. ], batch size: 65, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:32:43,055 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3391, 1.3168, 1.3023, 1.3608], device='cuda:0'), covar=tensor([0.1949, 0.1851, 0.1553, 0.1849], device='cuda:0'), in_proj_covar=tensor([0.1017, 0.0823, 0.0892, 0.0936], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 12:33:09,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.684e+02 1.024e+03 1.508e+03 2.045e+03 9.526e+03, threshold=3.016e+03, percent-clipped=16.0
+2023-03-01 12:33:20,184 INFO [train.py:968] (0/2) Epoch 3, batch 3850, giga_loss[loss=0.2793, simple_loss=0.3571, pruned_loss=0.1008, over 29145.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3858, pruned_loss=0.1302, over 5712936.53 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3969, pruned_loss=0.1341, over 4775543.30 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3846, pruned_loss=0.1299, over 5698949.39 frames. ], batch size: 128, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:33:39,038 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=94163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:33:45,289 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-01 12:34:04,099 INFO [train.py:968] (0/2) Epoch 3, batch 3900, giga_loss[loss=0.2983, simple_loss=0.3809, pruned_loss=0.1079, over 28811.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3839, pruned_loss=0.128, over 5717523.01 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3968, pruned_loss=0.134, over 4797450.39 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3828, pruned_loss=0.1278, over 5703297.26 frames. ], batch size: 119, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:34:36,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.432e+02 9.530e+02 1.197e+03 1.577e+03 6.295e+03, threshold=2.394e+03, percent-clipped=5.0
+2023-03-01 12:34:44,445 INFO [train.py:968] (0/2) Epoch 3, batch 3950, giga_loss[loss=0.3439, simple_loss=0.3939, pruned_loss=0.147, over 28760.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3835, pruned_loss=0.128, over 5712934.26 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3969, pruned_loss=0.1339, over 4819701.45 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3823, pruned_loss=0.1277, over 5698061.17 frames. ], batch size: 99, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:35:24,330 INFO [train.py:968] (0/2) Epoch 3, batch 4000, libri_loss[loss=0.3567, simple_loss=0.4158, pruned_loss=0.1488, over 29538.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3819, pruned_loss=0.1275, over 5720838.57 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3962, pruned_loss=0.1337, over 4856937.65 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3809, pruned_loss=0.1272, over 5702941.06 frames. ], batch size: 82, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:35:53,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.060e+02 1.000e+03 1.301e+03 1.986e+03 3.797e+03, threshold=2.602e+03, percent-clipped=13.0
+2023-03-01 12:36:01,878 INFO [train.py:968] (0/2) Epoch 3, batch 4050, giga_loss[loss=0.2932, simple_loss=0.3571, pruned_loss=0.1147, over 28443.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3793, pruned_loss=0.1267, over 5722233.82 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3957, pruned_loss=0.1335, over 4872773.96 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3787, pruned_loss=0.1266, over 5705592.48 frames. ], batch size: 85, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:36:28,432 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=94374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:36:41,611 INFO [train.py:968] (0/2) Epoch 3, batch 4100, libri_loss[loss=0.3423, simple_loss=0.4115, pruned_loss=0.1366, over 28727.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3778, pruned_loss=0.1258, over 5723625.51 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3962, pruned_loss=0.1336, over 4914088.34 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3761, pruned_loss=0.1252, over 5706466.68 frames. ], batch size: 106, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:37:14,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.078e+02 1.132e+03 1.434e+03 1.918e+03 4.755e+03, threshold=2.868e+03, percent-clipped=8.0
+2023-03-01 12:37:21,688 INFO [train.py:968] (0/2) Epoch 3, batch 4150, giga_loss[loss=0.329, simple_loss=0.3935, pruned_loss=0.1323, over 27864.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3788, pruned_loss=0.1267, over 5722324.14 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3964, pruned_loss=0.1336, over 4941167.77 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3767, pruned_loss=0.126, over 5705486.39 frames. ], batch size: 412, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:37:27,693 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2132, 1.1332, 0.9995, 1.0112], device='cuda:0'), covar=tensor([0.0565, 0.0502, 0.0975, 0.0694], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0489, 0.0537, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 12:38:01,991 INFO [train.py:968] (0/2) Epoch 3, batch 4200, giga_loss[loss=0.2954, simple_loss=0.3589, pruned_loss=0.116, over 28181.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3784, pruned_loss=0.1271, over 5721997.23 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3962, pruned_loss=0.1333, over 4968400.63 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3764, pruned_loss=0.1267, over 5704754.83 frames. ], batch size: 77, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:38:34,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.636e+02 1.163e+03 1.491e+03 1.979e+03 7.973e+03, threshold=2.983e+03, percent-clipped=8.0
+2023-03-01 12:38:42,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:38:45,307 INFO [train.py:968] (0/2) Epoch 3, batch 4250, giga_loss[loss=0.3185, simple_loss=0.3758, pruned_loss=0.1306, over 27982.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.377, pruned_loss=0.1271, over 5723028.34 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3963, pruned_loss=0.1332, over 4991455.02 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3749, pruned_loss=0.1266, over 5705466.46 frames. ], batch size: 412, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:39:25,417 INFO [train.py:968] (0/2) Epoch 3, batch 4300, giga_loss[loss=0.2685, simple_loss=0.3346, pruned_loss=0.1011, over 28384.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3734, pruned_loss=0.1253, over 5720603.81 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3964, pruned_loss=0.1333, over 5005137.03 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3713, pruned_loss=0.1247, over 5704139.86 frames. ], batch size: 78, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:39:45,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=94616.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:39:55,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.512e+02 9.993e+02 1.221e+03 1.789e+03 6.312e+03, threshold=2.442e+03, percent-clipped=8.0
+2023-03-01 12:40:06,321 INFO [train.py:968] (0/2) Epoch 3, batch 4350, giga_loss[loss=0.2817, simple_loss=0.345, pruned_loss=0.1092, over 28632.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3711, pruned_loss=0.1245, over 5719117.46 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3959, pruned_loss=0.1331, over 5029023.16 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3692, pruned_loss=0.124, over 5703535.02 frames. ], batch size: 262, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:40:21,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3031, 1.2351, 1.1697, 1.5288], device='cuda:0'), covar=tensor([0.1983, 0.2010, 0.1791, 0.1998], device='cuda:0'), in_proj_covar=tensor([0.1027, 0.0820, 0.0898, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 12:40:24,634 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.01 vs. limit=2.0
+2023-03-01 12:40:37,183 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:40:39,959 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:40:44,723 INFO [train.py:968] (0/2) Epoch 3, batch 4400, giga_loss[loss=0.2984, simple_loss=0.3652, pruned_loss=0.1159, over 28977.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.371, pruned_loss=0.1241, over 5720725.15 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3962, pruned_loss=0.1333, over 5046400.86 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3687, pruned_loss=0.1233, over 5705044.13 frames. ], batch size: 136, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:41:07,022 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:41:20,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.692e+02 1.056e+03 1.304e+03 1.677e+03 5.279e+03, threshold=2.609e+03, percent-clipped=10.0
+2023-03-01 12:41:33,926 INFO [train.py:968] (0/2) Epoch 3, batch 4450, giga_loss[loss=0.3242, simple_loss=0.3901, pruned_loss=0.1292, over 28691.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3735, pruned_loss=0.1257, over 5716761.65 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3961, pruned_loss=0.1334, over 5058547.08 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3714, pruned_loss=0.1249, over 5702402.73 frames. ], batch size: 242, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:41:34,157 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=94741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:41:40,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:42:15,371 INFO [train.py:968] (0/2) Epoch 3, batch 4500, giga_loss[loss=0.2919, simple_loss=0.364, pruned_loss=0.1099, over 28977.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3757, pruned_loss=0.1263, over 5726596.77 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3962, pruned_loss=0.1334, over 5078086.11 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3735, pruned_loss=0.1255, over 5712248.06 frames. ], batch size: 164, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:42:50,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.211e+02 1.023e+03 1.324e+03 1.773e+03 3.410e+03, threshold=2.649e+03, percent-clipped=3.0
+2023-03-01 12:43:01,870 INFO [train.py:968] (0/2) Epoch 3, batch 4550, giga_loss[loss=0.3667, simple_loss=0.412, pruned_loss=0.1607, over 28995.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3774, pruned_loss=0.1271, over 5719986.43 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3962, pruned_loss=0.1336, over 5090553.01 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3755, pruned_loss=0.1263, over 5706234.55 frames. ], batch size: 106, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:43:46,599 INFO [train.py:968] (0/2) Epoch 3, batch 4600, libri_loss[loss=0.3357, simple_loss=0.3895, pruned_loss=0.1409, over 29545.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.378, pruned_loss=0.1269, over 5707600.70 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3965, pruned_loss=0.1339, over 5107465.65 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3758, pruned_loss=0.1258, over 5695372.08 frames. ], batch size: 79, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:43:47,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:43:49,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:43:59,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0450, 4.3639, 4.6431, 1.8761], device='cuda:0'), covar=tensor([0.0314, 0.0324, 0.0678, 0.2014], device='cuda:0'), in_proj_covar=tensor([0.0755, 0.0562, 0.0781, 0.0557], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 12:44:14,349 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:44:19,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.209e+02 1.230e+03 1.501e+03 2.058e+03 6.915e+03, threshold=3.001e+03, percent-clipped=9.0
+2023-03-01 12:44:28,467 INFO [train.py:968] (0/2) Epoch 3, batch 4650, giga_loss[loss=0.2865, simple_loss=0.3572, pruned_loss=0.1079, over 28946.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3778, pruned_loss=0.1263, over 5707174.83 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3967, pruned_loss=0.1343, over 5130507.17 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3753, pruned_loss=0.1249, over 5692630.03 frames. ], batch size: 136, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:45:09,625 INFO [train.py:968] (0/2) Epoch 3, batch 4700, giga_loss[loss=0.3215, simple_loss=0.3894, pruned_loss=0.1268, over 28901.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3791, pruned_loss=0.1271, over 5704553.02 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3971, pruned_loss=0.1346, over 5139897.90 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3764, pruned_loss=0.1256, over 5697613.05 frames. ], batch size: 213, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:45:09,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94991.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:45:34,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4366, 3.9851, 4.1156, 1.9001], device='cuda:0'), covar=tensor([0.0436, 0.0329, 0.0729, 0.1738], device='cuda:0'), in_proj_covar=tensor([0.0767, 0.0567, 0.0785, 0.0563], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 12:45:43,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.391e+02 1.126e+03 1.463e+03 1.938e+03 3.479e+03, threshold=2.925e+03, percent-clipped=4.0
+2023-03-01 12:45:51,897 INFO [train.py:968] (0/2) Epoch 3, batch 4750, giga_loss[loss=0.3099, simple_loss=0.3666, pruned_loss=0.1266, over 28922.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3796, pruned_loss=0.1274, over 5708759.55 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3975, pruned_loss=0.1348, over 5154137.75 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3767, pruned_loss=0.1258, over 5702859.73 frames. ], batch size: 106, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:46:34,812 INFO [train.py:968] (0/2) Epoch 3, batch 4800, giga_loss[loss=0.3154, simple_loss=0.3727, pruned_loss=0.129, over 28983.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3821, pruned_loss=0.1294, over 5700769.84 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.3978, pruned_loss=0.1349, over 5170883.04 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3789, pruned_loss=0.1278, over 5701173.93 frames. ], batch size: 106, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:46:54,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:47:04,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.570e+02 1.131e+03 1.436e+03 1.980e+03 3.992e+03, threshold=2.872e+03, percent-clipped=9.0
+2023-03-01 12:47:09,941 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95134.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:47:12,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95137.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:47:14,808 INFO [train.py:968] (0/2) Epoch 3, batch 4850, libri_loss[loss=0.3389, simple_loss=0.4062, pruned_loss=0.1358, over 25720.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3851, pruned_loss=0.1312, over 5698069.71 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3981, pruned_loss=0.1352, over 5182877.31 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3817, pruned_loss=0.1295, over 5704360.76 frames. ], batch size: 136, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:47:34,586 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95166.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:47:55,665 INFO [train.py:968] (0/2) Epoch 3, batch 4900, giga_loss[loss=0.3608, simple_loss=0.4177, pruned_loss=0.152, over 28034.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3868, pruned_loss=0.1319, over 5705714.34 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.3978, pruned_loss=0.1349, over 5200998.58 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3841, pruned_loss=0.1308, over 5705933.34 frames. ], batch size: 412, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:48:28,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.698e+02 1.218e+03 1.562e+03 2.150e+03 5.798e+03, threshold=3.125e+03, percent-clipped=16.0
+2023-03-01 12:48:36,483 INFO [train.py:968] (0/2) Epoch 3, batch 4950, giga_loss[loss=0.2944, simple_loss=0.3693, pruned_loss=0.1098, over 28922.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3887, pruned_loss=0.1333, over 5710835.40 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3985, pruned_loss=0.1355, over 5214406.83 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3859, pruned_loss=0.1318, over 5708201.70 frames. ], batch size: 227, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:48:52,184 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:48:54,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:49:02,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8204, 1.6100, 1.2451, 1.4571], device='cuda:0'), covar=tensor([0.0509, 0.0651, 0.0836, 0.0929], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0488, 0.0528, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 12:49:20,317 INFO [train.py:968] (0/2) Epoch 3, batch 5000, giga_loss[loss=0.2882, simple_loss=0.3535, pruned_loss=0.1114, over 28507.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3901, pruned_loss=0.1343, over 5708191.60 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3996, pruned_loss=0.1363, over 5235186.21 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3866, pruned_loss=0.1325, over 5700571.18 frames. ], batch size: 78, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:49:20,538 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:49:25,392 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-01 12:49:54,497 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.307e+02 1.257e+03 1.522e+03 2.075e+03 5.779e+03, threshold=3.044e+03, percent-clipped=6.0
+2023-03-01 12:50:00,668 INFO [train.py:968] (0/2) Epoch 3, batch 5050, giga_loss[loss=0.2991, simple_loss=0.3631, pruned_loss=0.1175, over 28980.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3898, pruned_loss=0.134, over 5712814.82 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3999, pruned_loss=0.1365, over 5248612.01 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3866, pruned_loss=0.1323, over 5703167.83 frames. ], batch size: 128, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:50:26,558 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-01 12:50:41,761 INFO [train.py:968] (0/2) Epoch 3, batch 5100, giga_loss[loss=0.29, simple_loss=0.36, pruned_loss=0.11, over 28867.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3888, pruned_loss=0.1336, over 5708901.50 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3999, pruned_loss=0.1368, over 5260916.20 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3859, pruned_loss=0.1319, over 5704723.51 frames. ], batch size: 199, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:51:18,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.985e+02 1.206e+03 1.634e+03 2.392e+03 8.615e+03, threshold=3.268e+03, percent-clipped=15.0
+2023-03-01 12:51:24,559 INFO [train.py:968] (0/2) Epoch 3, batch 5150, giga_loss[loss=0.3614, simple_loss=0.4056, pruned_loss=0.1586, over 27657.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3844, pruned_loss=0.1314, over 5708409.39 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3994, pruned_loss=0.1366, over 5279992.21 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3822, pruned_loss=0.13, over 5699668.75 frames. ], batch size: 472, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:51:33,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7355, 1.6022, 1.5672, 1.9161], device='cuda:0'), covar=tensor([0.1924, 0.1775, 0.1566, 0.1938], device='cuda:0'), in_proj_covar=tensor([0.1009, 0.0810, 0.0898, 0.0936], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 12:52:03,986 INFO [train.py:968] (0/2) Epoch 3, batch 5200, giga_loss[loss=0.2876, simple_loss=0.3564, pruned_loss=0.1094, over 28814.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3812, pruned_loss=0.1293, over 5714966.88 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3991, pruned_loss=0.1365, over 5301311.76 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.379, pruned_loss=0.1281, over 5703794.22 frames. ], batch size: 186, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:52:09,954 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=95498.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:52:18,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6187, 1.3824, 1.3493, 1.8842], device='cuda:0'), covar=tensor([0.2048, 0.2001, 0.1838, 0.1984], device='cuda:0'), in_proj_covar=tensor([0.1014, 0.0810, 0.0901, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 12:52:35,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.042e+02 1.107e+03 1.596e+03 2.025e+03 4.061e+03, threshold=3.192e+03, percent-clipped=5.0
+2023-03-01 12:52:44,191 INFO [train.py:968] (0/2) Epoch 3, batch 5250, giga_loss[loss=0.3211, simple_loss=0.3939, pruned_loss=0.1242, over 28696.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3813, pruned_loss=0.1283, over 5716660.81 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3995, pruned_loss=0.1367, over 5313255.57 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3789, pruned_loss=0.1269, over 5704431.56 frames. ], batch size: 262, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:53:30,406 INFO [train.py:968] (0/2) Epoch 3, batch 5300, giga_loss[loss=0.2971, simple_loss=0.3726, pruned_loss=0.1108, over 28899.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3829, pruned_loss=0.1277, over 5717277.88 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3994, pruned_loss=0.1369, over 5319002.90 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3809, pruned_loss=0.1265, over 5706221.67 frames. ], batch size: 164, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 12:54:03,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.125e+02 1.104e+03 1.302e+03 1.688e+03 4.901e+03, threshold=2.604e+03, percent-clipped=5.0
+2023-03-01 12:54:12,391 INFO [train.py:968] (0/2) Epoch 3, batch 5350, giga_loss[loss=0.2735, simple_loss=0.3353, pruned_loss=0.1058, over 28541.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3839, pruned_loss=0.1288, over 5718344.56 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3997, pruned_loss=0.1371, over 5327304.62 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3819, pruned_loss=0.1275, over 5709290.65 frames. ], batch size: 71, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 12:54:14,695 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 12:54:21,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=95653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:54:27,711 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=95661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:54:52,325 INFO [train.py:968] (0/2) Epoch 3, batch 5400, giga_loss[loss=0.3227, simple_loss=0.3791, pruned_loss=0.1332, over 28609.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3817, pruned_loss=0.129, over 5725692.87 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3998, pruned_loss=0.1372, over 5338784.32 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3796, pruned_loss=0.1277, over 5715245.73 frames. ], batch size: 262, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:54:56,734 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-01 12:54:56,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-01 12:55:29,818 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=95732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:55:30,291 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.641e+02 1.077e+03 1.350e+03 1.806e+03 4.100e+03, threshold=2.699e+03, percent-clipped=6.0
+2023-03-01 12:55:36,829 INFO [train.py:968] (0/2) Epoch 3, batch 5450, giga_loss[loss=0.3201, simple_loss=0.376, pruned_loss=0.1321, over 28911.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3803, pruned_loss=0.1293, over 5731962.98 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.4, pruned_loss=0.1372, over 5349778.40 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3781, pruned_loss=0.1281, over 5720618.33 frames. ], batch size: 227, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:56:18,687 INFO [train.py:968] (0/2) Epoch 3, batch 5500, giga_loss[loss=0.2868, simple_loss=0.3498, pruned_loss=0.1119, over 28850.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3776, pruned_loss=0.1295, over 5733370.99 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3995, pruned_loss=0.1371, over 5356538.75 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.376, pruned_loss=0.1285, over 5723501.31 frames. ], batch size: 112, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:56:43,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5925, 3.1962, 3.2628, 1.5369], device='cuda:0'), covar=tensor([0.0646, 0.0528, 0.0971, 0.2244], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0562, 0.0782, 0.0559], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 12:56:58,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.271e+02 1.063e+03 1.325e+03 1.850e+03 3.947e+03, threshold=2.651e+03, percent-clipped=6.0
+2023-03-01 12:57:05,110 INFO [train.py:968] (0/2) Epoch 3, batch 5550, giga_loss[loss=0.2948, simple_loss=0.3636, pruned_loss=0.113, over 28777.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3772, pruned_loss=0.1299, over 5728515.86 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3998, pruned_loss=0.1373, over 5362022.62 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3754, pruned_loss=0.1289, over 5719082.73 frames. ], batch size: 243, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:57:29,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95873.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:57:35,267 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-01 12:57:45,517 INFO [train.py:968] (0/2) Epoch 3, batch 5600, giga_loss[loss=0.2896, simple_loss=0.3534, pruned_loss=0.1129, over 29126.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.376, pruned_loss=0.1293, over 5722860.62 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3999, pruned_loss=0.1376, over 5383372.65 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3735, pruned_loss=0.1279, over 5710428.42 frames. ], batch size: 155, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 12:57:46,979 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7044, 3.1580, 1.6547, 1.6197], device='cuda:0'), covar=tensor([0.0802, 0.0401, 0.0836, 0.1202], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0450, 0.0322, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 12:58:18,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.048e+02 1.278e+03 1.802e+03 2.596e+03 9.286e+03, threshold=3.605e+03, percent-clipped=22.0
+2023-03-01 12:58:23,998 INFO [train.py:968] (0/2) Epoch 3, batch 5650, giga_loss[loss=0.3196, simple_loss=0.3773, pruned_loss=0.131, over 27663.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3732, pruned_loss=0.1282, over 5725946.50 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.4, pruned_loss=0.1379, over 5398680.37 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3705, pruned_loss=0.1266, over 5711177.89 frames. ], batch size: 472, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:58:36,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2179, 4.7623, 4.8567, 1.9568], device='cuda:0'), covar=tensor([0.0296, 0.0233, 0.0641, 0.1831], device='cuda:0'), in_proj_covar=tensor([0.0756, 0.0555, 0.0769, 0.0549], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 12:58:38,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8208, 2.8277, 2.0308, 0.7236], device='cuda:0'), covar=tensor([0.1866, 0.0716, 0.1041, 0.1857], device='cuda:0'), in_proj_covar=tensor([0.1216, 0.1133, 0.1238, 0.1054], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 12:58:39,949 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-01 12:58:46,319 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4463, 1.9775, 2.0878, 1.9655], device='cuda:0'), covar=tensor([0.0903, 0.1905, 0.1293, 0.1439], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0767, 0.0631, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 12:59:03,517 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1257, 1.3358, 1.0500, 0.3991], device='cuda:0'), covar=tensor([0.1065, 0.0993, 0.1708, 0.1781], device='cuda:0'), in_proj_covar=tensor([0.1221, 0.1138, 0.1248, 0.1065], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 12:59:04,491 INFO [train.py:968] (0/2) Epoch 3, batch 5700, giga_loss[loss=0.258, simple_loss=0.3331, pruned_loss=0.09148, over 29041.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.368, pruned_loss=0.1248, over 5711092.04 frames. ], libri_tot_loss[loss=0.3378, simple_loss=0.3999, pruned_loss=0.1378, over 5397599.96 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3653, pruned_loss=0.1233, over 5706975.61 frames. ], batch size: 164, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:59:10,362 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-96000.pt
+2023-03-01 12:59:21,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:59:25,790 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96016.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:59:28,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96019.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:59:34,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:59:38,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.778e+02 1.202e+03 1.512e+03 1.946e+03 4.602e+03, threshold=3.024e+03, percent-clipped=2.0
+2023-03-01 12:59:40,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:59:43,109 INFO [train.py:968] (0/2) Epoch 3, batch 5750, giga_loss[loss=0.3289, simple_loss=0.3883, pruned_loss=0.1348, over 28729.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3683, pruned_loss=0.1251, over 5714346.39 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3996, pruned_loss=0.1378, over 5411290.91 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3655, pruned_loss=0.1236, over 5707286.50 frames. ], batch size: 262, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:59:49,188 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96048.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 13:00:23,383 INFO [train.py:968] (0/2) Epoch 3, batch 5800, giga_loss[loss=0.2969, simple_loss=0.371, pruned_loss=0.1113, over 28852.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.372, pruned_loss=0.1272, over 5717941.91 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3989, pruned_loss=0.1375, over 5431110.13 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3693, pruned_loss=0.1258, over 5705181.80 frames. ], batch size: 119, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:00:39,408 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:00:50,832 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-01 13:01:00,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.396e+02 1.292e+03 1.556e+03 1.949e+03 3.577e+03, threshold=3.113e+03, percent-clipped=3.0
+2023-03-01 13:01:05,124 INFO [train.py:968] (0/2) Epoch 3, batch 5850, giga_loss[loss=0.3915, simple_loss=0.4361, pruned_loss=0.1735, over 28973.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3766, pruned_loss=0.1292, over 5717717.25 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.399, pruned_loss=0.1377, over 5448821.00 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3732, pruned_loss=0.1275, over 5702940.53 frames. ], batch size: 227, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:01:31,256 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:01:33,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:01:37,925 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:01:41,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:01:49,027 INFO [train.py:968] (0/2) Epoch 3, batch 5900, giga_loss[loss=0.3469, simple_loss=0.4108, pruned_loss=0.1415, over 28893.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3793, pruned_loss=0.1296, over 5723079.87 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3986, pruned_loss=0.1374, over 5455529.79 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3766, pruned_loss=0.1283, over 5709102.69 frames. ], batch size: 199, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:02:00,584 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:02:01,854 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96205.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:02:07,226 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:02:07,310 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7227, 2.2726, 1.9864, 2.0689], device='cuda:0'), covar=tensor([0.0486, 0.0580, 0.0696, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0486, 0.0524, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 13:02:28,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.999e+02 1.223e+03 1.507e+03 1.837e+03 6.871e+03, threshold=3.014e+03, percent-clipped=7.0
+2023-03-01 13:02:32,255 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:02:32,753 INFO [train.py:968] (0/2) Epoch 3, batch 5950, giga_loss[loss=0.3317, simple_loss=0.3845, pruned_loss=0.1395, over 28831.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3827, pruned_loss=0.1313, over 5722209.76 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3988, pruned_loss=0.1375, over 5469509.64 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3798, pruned_loss=0.13, over 5706490.41 frames. ], batch size: 99, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:02:40,626 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:02:42,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:03:07,878 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:03:17,412 INFO [train.py:968] (0/2) Epoch 3, batch 6000, giga_loss[loss=0.3134, simple_loss=0.3803, pruned_loss=0.1232, over 28909.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3847, pruned_loss=0.1323, over 5718416.48 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3991, pruned_loss=0.1378, over 5477026.92 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3818, pruned_loss=0.131, over 5704545.82 frames. ], batch size: 145, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 13:03:17,419 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 13:03:26,065 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2665, simple_loss=0.3636, pruned_loss=0.08465, over 944034.00 frames. 
+2023-03-01 13:03:26,065 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19537MB
+2023-03-01 13:03:37,293 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-01 13:04:09,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.727e+02 1.321e+03 1.637e+03 2.273e+03 5.510e+03, threshold=3.275e+03, percent-clipped=13.0
+2023-03-01 13:04:15,888 INFO [train.py:968] (0/2) Epoch 3, batch 6050, libri_loss[loss=0.328, simple_loss=0.3895, pruned_loss=0.1332, over 29518.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3916, pruned_loss=0.1392, over 5708249.30 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.399, pruned_loss=0.1377, over 5479159.78 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3893, pruned_loss=0.1381, over 5696417.69 frames. ], batch size: 80, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:04:44,934 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-01 13:04:47,205 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:04:59,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:05:03,604 INFO [train.py:968] (0/2) Epoch 3, batch 6100, giga_loss[loss=0.4233, simple_loss=0.4592, pruned_loss=0.1937, over 28690.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.4001, pruned_loss=0.1468, over 5695201.13 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3991, pruned_loss=0.1377, over 5479501.99 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3982, pruned_loss=0.1461, over 5691668.15 frames. ], batch size: 307, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:05:08,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2466, 1.6380, 1.4025, 1.4262], device='cuda:0'), covar=tensor([0.0908, 0.0337, 0.0369, 0.0980], device='cuda:0'), in_proj_covar=tensor([0.0247, 0.0165, 0.0165, 0.0296], device='cuda:0'), out_proj_covar=tensor([0.0037, 0.0028, 0.0025, 0.0042], device='cuda:0')
+2023-03-01 13:05:46,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.761e+03 2.358e+03 3.323e+03 7.261e+03, threshold=4.717e+03, percent-clipped=26.0
+2023-03-01 13:05:53,001 INFO [train.py:968] (0/2) Epoch 3, batch 6150, giga_loss[loss=0.3587, simple_loss=0.408, pruned_loss=0.1547, over 28863.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.4083, pruned_loss=0.1534, over 5694515.92 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3987, pruned_loss=0.1375, over 5489796.35 frames. ], giga_tot_loss[loss=0.3569, simple_loss=0.4072, pruned_loss=0.1533, over 5687450.34 frames. ], batch size: 119, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:05:58,820 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 13:06:27,121 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:06:44,313 INFO [train.py:968] (0/2) Epoch 3, batch 6200, giga_loss[loss=0.3549, simple_loss=0.4046, pruned_loss=0.1527, over 28554.00 frames. ], tot_loss[loss=0.3652, simple_loss=0.4134, pruned_loss=0.1585, over 5696658.87 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3989, pruned_loss=0.1378, over 5494906.81 frames. ], giga_tot_loss[loss=0.3648, simple_loss=0.4125, pruned_loss=0.1586, over 5690094.29 frames. ], batch size: 336, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:07:17,664 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:07:21,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:07:24,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.645e+03 2.116e+03 2.978e+03 6.310e+03, threshold=4.231e+03, percent-clipped=3.0
+2023-03-01 13:07:29,773 INFO [train.py:968] (0/2) Epoch 3, batch 6250, giga_loss[loss=0.3852, simple_loss=0.4308, pruned_loss=0.1698, over 28945.00 frames. ], tot_loss[loss=0.3756, simple_loss=0.4207, pruned_loss=0.1653, over 5695548.90 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.3993, pruned_loss=0.138, over 5509060.05 frames. ], giga_tot_loss[loss=0.376, simple_loss=0.4203, pruned_loss=0.1658, over 5683838.16 frames. ], batch size: 227, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:07:35,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9319, 1.6541, 1.6917, 1.6787], device='cuda:0'), covar=tensor([0.0907, 0.1769, 0.1307, 0.1306], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0783, 0.0639, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 13:07:45,310 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:07:46,659 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:07:56,589 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-01 13:08:05,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96580.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 13:08:17,441 INFO [train.py:968] (0/2) Epoch 3, batch 6300, libri_loss[loss=0.3699, simple_loss=0.4246, pruned_loss=0.1575, over 19452.00 frames. ], tot_loss[loss=0.3809, simple_loss=0.4246, pruned_loss=0.1686, over 5684768.09 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.3994, pruned_loss=0.138, over 5513907.74 frames. ], giga_tot_loss[loss=0.3827, simple_loss=0.4251, pruned_loss=0.1701, over 5678383.92 frames. ], batch size: 186, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:08:44,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:08:52,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4126, 1.3787, 1.0602, 1.1246], device='cuda:0'), covar=tensor([0.0492, 0.0524, 0.0377, 0.0462], device='cuda:0'), in_proj_covar=tensor([0.1174, 0.0864, 0.0921, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 13:09:06,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.647e+03 2.092e+03 2.959e+03 8.993e+03, threshold=4.184e+03, percent-clipped=6.0
+2023-03-01 13:09:11,705 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.95 vs. limit=5.0
+2023-03-01 13:09:11,890 INFO [train.py:968] (0/2) Epoch 3, batch 6350, giga_loss[loss=0.4305, simple_loss=0.4562, pruned_loss=0.2024, over 28648.00 frames. ], tot_loss[loss=0.3843, simple_loss=0.4263, pruned_loss=0.1711, over 5669146.00 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3992, pruned_loss=0.138, over 5517717.08 frames. ], giga_tot_loss[loss=0.3861, simple_loss=0.427, pruned_loss=0.1726, over 5662299.60 frames. ], batch size: 307, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:10:05,819 INFO [train.py:968] (0/2) Epoch 3, batch 6400, giga_loss[loss=0.502, simple_loss=0.4993, pruned_loss=0.2523, over 26543.00 frames. ], tot_loss[loss=0.3896, simple_loss=0.4294, pruned_loss=0.1749, over 5665740.60 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.399, pruned_loss=0.1378, over 5522061.44 frames. ], giga_tot_loss[loss=0.3924, simple_loss=0.4307, pruned_loss=0.177, over 5659816.84 frames. ], batch size: 555, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 13:10:19,895 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6135, 2.5822, 1.8270, 2.0893], device='cuda:0'), covar=tensor([0.0587, 0.0534, 0.0846, 0.0950], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0496, 0.0534, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 13:10:34,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:10:40,761 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96723.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 13:10:45,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96726.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:10:55,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.760e+03 2.245e+03 3.065e+03 6.456e+03, threshold=4.491e+03, percent-clipped=10.0
+2023-03-01 13:11:00,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1986, 1.6825, 1.2838, 0.4193], device='cuda:0'), covar=tensor([0.1132, 0.0845, 0.1140, 0.1424], device='cuda:0'), in_proj_covar=tensor([0.1259, 0.1187, 0.1252, 0.1077], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 13:11:02,778 INFO [train.py:968] (0/2) Epoch 3, batch 6450, libri_loss[loss=0.371, simple_loss=0.4334, pruned_loss=0.1543, over 26311.00 frames. ], tot_loss[loss=0.3944, simple_loss=0.432, pruned_loss=0.1784, over 5656774.30 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3988, pruned_loss=0.1375, over 5528327.63 frames. ], giga_tot_loss[loss=0.3981, simple_loss=0.4339, pruned_loss=0.1812, over 5649919.88 frames. ], batch size: 136, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:11:05,503 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-01 13:11:12,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:11:20,138 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96755.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:11:23,900 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:11:26,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:11:50,903 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:11:57,753 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:11:58,158 INFO [train.py:968] (0/2) Epoch 3, batch 6500, giga_loss[loss=0.4063, simple_loss=0.434, pruned_loss=0.1893, over 28010.00 frames. ], tot_loss[loss=0.3992, simple_loss=0.4353, pruned_loss=0.1815, over 5637652.33 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3987, pruned_loss=0.1373, over 5524674.46 frames. ], giga_tot_loss[loss=0.4031, simple_loss=0.4373, pruned_loss=0.1844, over 5637794.00 frames. ], batch size: 412, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:12:43,305 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.837e+02 1.698e+03 2.281e+03 3.352e+03 8.536e+03, threshold=4.563e+03, percent-clipped=13.0
+2023-03-01 13:12:47,855 INFO [train.py:968] (0/2) Epoch 3, batch 6550, giga_loss[loss=0.4765, simple_loss=0.4634, pruned_loss=0.2449, over 23840.00 frames. ], tot_loss[loss=0.3991, simple_loss=0.4345, pruned_loss=0.1819, over 5636084.64 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3992, pruned_loss=0.1375, over 5529653.42 frames. ], giga_tot_loss[loss=0.4033, simple_loss=0.4365, pruned_loss=0.185, over 5634380.77 frames. ], batch size: 705, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:12:54,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:13:11,354 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-01 13:13:27,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5124, 2.1671, 1.6160, 0.8181], device='cuda:0'), covar=tensor([0.1539, 0.0897, 0.1300, 0.1740], device='cuda:0'), in_proj_covar=tensor([0.1275, 0.1210, 0.1274, 0.1092], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 13:13:39,558 INFO [train.py:968] (0/2) Epoch 3, batch 6600, libri_loss[loss=0.3175, simple_loss=0.3859, pruned_loss=0.1245, over 29545.00 frames. ], tot_loss[loss=0.3973, simple_loss=0.4329, pruned_loss=0.1808, over 5628691.72 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3998, pruned_loss=0.1377, over 5529544.16 frames. ], giga_tot_loss[loss=0.4016, simple_loss=0.4348, pruned_loss=0.1841, over 5629601.59 frames. ], batch size: 80, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:13:39,855 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:13:42,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:13:52,115 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-01 13:14:12,233 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:14:22,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:14:23,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.710e+03 2.234e+03 2.718e+03 9.454e+03, threshold=4.468e+03, percent-clipped=9.0
+2023-03-01 13:14:26,956 INFO [train.py:968] (0/2) Epoch 3, batch 6650, giga_loss[loss=0.3876, simple_loss=0.4404, pruned_loss=0.1674, over 28938.00 frames. ], tot_loss[loss=0.3955, simple_loss=0.4326, pruned_loss=0.1792, over 5636884.09 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.4001, pruned_loss=0.138, over 5542632.81 frames. ], giga_tot_loss[loss=0.4008, simple_loss=0.4351, pruned_loss=0.1833, over 5629201.06 frames. ], batch size: 213, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:15:19,505 INFO [train.py:968] (0/2) Epoch 3, batch 6700, giga_loss[loss=0.3891, simple_loss=0.4381, pruned_loss=0.1701, over 28992.00 frames. ], tot_loss[loss=0.3945, simple_loss=0.4328, pruned_loss=0.1781, over 5640396.59 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.4005, pruned_loss=0.1382, over 5544449.48 frames. ], giga_tot_loss[loss=0.3991, simple_loss=0.4349, pruned_loss=0.1817, over 5633843.93 frames. ], batch size: 136, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:15:20,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:15:22,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:15:54,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:16:07,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.538e+02 1.674e+03 2.085e+03 2.801e+03 4.617e+03, threshold=4.170e+03, percent-clipped=2.0
+2023-03-01 13:16:13,126 INFO [train.py:968] (0/2) Epoch 3, batch 6750, giga_loss[loss=0.3462, simple_loss=0.4074, pruned_loss=0.1425, over 28956.00 frames. ], tot_loss[loss=0.3956, simple_loss=0.4337, pruned_loss=0.1787, over 5632646.27 frames. ], libri_tot_loss[loss=0.3383, simple_loss=0.4003, pruned_loss=0.1381, over 5551468.04 frames. ], giga_tot_loss[loss=0.4004, simple_loss=0.4361, pruned_loss=0.1823, over 5623098.66 frames. ], batch size: 164, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:16:37,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2440, 1.3568, 0.9573, 0.5604], device='cuda:0'), covar=tensor([0.0768, 0.0539, 0.0474, 0.0741], device='cuda:0'), in_proj_covar=tensor([0.1157, 0.0854, 0.0900, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 13:16:48,878 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:16:53,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:17:08,841 INFO [train.py:968] (0/2) Epoch 3, batch 6800, giga_loss[loss=0.4585, simple_loss=0.4621, pruned_loss=0.2275, over 26475.00 frames. ], tot_loss[loss=0.3915, simple_loss=0.4311, pruned_loss=0.1759, over 5629321.04 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.4008, pruned_loss=0.1383, over 5556211.20 frames. ], giga_tot_loss[loss=0.3956, simple_loss=0.433, pruned_loss=0.1791, over 5618697.70 frames. ], batch size: 555, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 13:17:11,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:17:27,292 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:17:56,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.481e+02 1.527e+03 1.866e+03 2.366e+03 7.818e+03, threshold=3.732e+03, percent-clipped=6.0
+2023-03-01 13:18:00,162 INFO [train.py:968] (0/2) Epoch 3, batch 6850, giga_loss[loss=0.3626, simple_loss=0.4121, pruned_loss=0.1566, over 28672.00 frames. ], tot_loss[loss=0.3853, simple_loss=0.4277, pruned_loss=0.1714, over 5647059.52 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.4005, pruned_loss=0.1383, over 5561677.97 frames. ], giga_tot_loss[loss=0.3894, simple_loss=0.4298, pruned_loss=0.1745, over 5634911.68 frames. ], batch size: 262, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:18:18,135 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:18:26,919 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:18:46,230 INFO [train.py:968] (0/2) Epoch 3, batch 6900, giga_loss[loss=0.382, simple_loss=0.421, pruned_loss=0.1715, over 28643.00 frames. ], tot_loss[loss=0.3775, simple_loss=0.4219, pruned_loss=0.1665, over 5649289.36 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.399, pruned_loss=0.1376, over 5571457.91 frames. ], giga_tot_loss[loss=0.3844, simple_loss=0.4263, pruned_loss=0.1712, over 5634675.32 frames. ], batch size: 307, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:19:21,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9302, 1.3785, 4.6089, 3.4820], device='cuda:0'), covar=tensor([0.1559, 0.1816, 0.0270, 0.0547], device='cuda:0'), in_proj_covar=tensor([0.0522, 0.0503, 0.0678, 0.0540], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 13:19:30,213 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:19:31,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.578e+02 1.510e+03 2.063e+03 2.868e+03 6.412e+03, threshold=4.126e+03, percent-clipped=9.0
+2023-03-01 13:19:33,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:19:35,136 INFO [train.py:968] (0/2) Epoch 3, batch 6950, giga_loss[loss=0.4356, simple_loss=0.4479, pruned_loss=0.2116, over 26618.00 frames. ], tot_loss[loss=0.3741, simple_loss=0.4197, pruned_loss=0.1643, over 5656667.25 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.399, pruned_loss=0.1377, over 5575267.40 frames. ], giga_tot_loss[loss=0.38, simple_loss=0.4234, pruned_loss=0.1683, over 5642805.46 frames. ], batch size: 555, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:19:57,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:20:23,416 INFO [train.py:968] (0/2) Epoch 3, batch 7000, giga_loss[loss=0.3931, simple_loss=0.4407, pruned_loss=0.1728, over 28853.00 frames. ], tot_loss[loss=0.372, simple_loss=0.4176, pruned_loss=0.1632, over 5656967.80 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3986, pruned_loss=0.1376, over 5585179.31 frames. ], giga_tot_loss[loss=0.3779, simple_loss=0.4215, pruned_loss=0.1672, over 5639313.34 frames. ], batch size: 284, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:20:36,008 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:20:37,903 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:20:41,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-01 13:21:06,713 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:21:09,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.650e+03 2.181e+03 3.204e+03 7.863e+03, threshold=4.363e+03, percent-clipped=8.0
+2023-03-01 13:21:14,988 INFO [train.py:968] (0/2) Epoch 3, batch 7050, giga_loss[loss=0.4087, simple_loss=0.4279, pruned_loss=0.1947, over 26629.00 frames. ], tot_loss[loss=0.3706, simple_loss=0.4167, pruned_loss=0.1623, over 5655795.92 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.398, pruned_loss=0.1373, over 5592517.66 frames. ], giga_tot_loss[loss=0.3769, simple_loss=0.4208, pruned_loss=0.1665, over 5636836.11 frames. ], batch size: 555, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:21:37,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3317, 1.9935, 1.3619, 1.5312], device='cuda:0'), covar=tensor([0.0773, 0.0283, 0.0340, 0.0837], device='cuda:0'), in_proj_covar=tensor([0.0249, 0.0166, 0.0165, 0.0300], device='cuda:0'), out_proj_covar=tensor([0.0038, 0.0029, 0.0025, 0.0043], device='cuda:0')
+2023-03-01 13:22:13,471 INFO [train.py:968] (0/2) Epoch 3, batch 7100, giga_loss[loss=0.3361, simple_loss=0.3882, pruned_loss=0.142, over 29002.00 frames. ], tot_loss[loss=0.3687, simple_loss=0.4154, pruned_loss=0.161, over 5657852.81 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3977, pruned_loss=0.137, over 5596208.33 frames. ], giga_tot_loss[loss=0.3745, simple_loss=0.4192, pruned_loss=0.1649, over 5640322.73 frames. ], batch size: 213, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:22:15,213 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:23:00,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.287e+02 1.355e+03 1.825e+03 2.652e+03 4.591e+03, threshold=3.650e+03, percent-clipped=1.0
+2023-03-01 13:23:04,646 INFO [train.py:968] (0/2) Epoch 3, batch 7150, giga_loss[loss=0.4483, simple_loss=0.481, pruned_loss=0.2078, over 28085.00 frames. ], tot_loss[loss=0.3637, simple_loss=0.4128, pruned_loss=0.1573, over 5664143.96 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3974, pruned_loss=0.1367, over 5602365.08 frames. ], giga_tot_loss[loss=0.3693, simple_loss=0.4165, pruned_loss=0.1611, over 5645932.92 frames. ], batch size: 412, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:24:01,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2657, 1.6833, 1.4031, 1.3303], device='cuda:0'), covar=tensor([0.0884, 0.0353, 0.0384, 0.1036], device='cuda:0'), in_proj_covar=tensor([0.0251, 0.0164, 0.0165, 0.0299], device='cuda:0'), out_proj_covar=tensor([0.0038, 0.0028, 0.0025, 0.0043], device='cuda:0')
+2023-03-01 13:24:07,404 INFO [train.py:968] (0/2) Epoch 3, batch 7200, giga_loss[loss=0.3717, simple_loss=0.429, pruned_loss=0.1572, over 28787.00 frames. ], tot_loss[loss=0.3632, simple_loss=0.4143, pruned_loss=0.1561, over 5668943.23 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3971, pruned_loss=0.1368, over 5607025.90 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.4176, pruned_loss=0.1594, over 5651076.10 frames. ], batch size: 119, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:24:19,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 13:24:21,019 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 13:24:55,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.635e+03 2.133e+03 3.522e+03 1.145e+04, threshold=4.266e+03, percent-clipped=25.0
+2023-03-01 13:24:57,852 INFO [train.py:968] (0/2) Epoch 3, batch 7250, giga_loss[loss=0.3732, simple_loss=0.4193, pruned_loss=0.1635, over 28913.00 frames. ], tot_loss[loss=0.3655, simple_loss=0.4164, pruned_loss=0.1573, over 5679276.11 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3968, pruned_loss=0.1366, over 5610023.04 frames. ], giga_tot_loss[loss=0.3699, simple_loss=0.4194, pruned_loss=0.1602, over 5663216.81 frames. ], batch size: 174, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:25:05,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:25:51,192 INFO [train.py:968] (0/2) Epoch 3, batch 7300, giga_loss[loss=0.4537, simple_loss=0.4666, pruned_loss=0.2204, over 26868.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4174, pruned_loss=0.1594, over 5655679.40 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3968, pruned_loss=0.1365, over 5606126.18 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4203, pruned_loss=0.1623, over 5647228.92 frames. ], batch size: 555, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:26:29,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9443, 4.1267, 1.7201, 1.5681], device='cuda:0'), covar=tensor([0.0972, 0.0345, 0.1003, 0.1594], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0459, 0.0321, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 13:26:35,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.643e+02 1.587e+03 1.966e+03 2.557e+03 7.945e+03, threshold=3.933e+03, percent-clipped=9.0
+2023-03-01 13:26:39,027 INFO [train.py:968] (0/2) Epoch 3, batch 7350, giga_loss[loss=0.3681, simple_loss=0.4105, pruned_loss=0.1629, over 28598.00 frames. ], tot_loss[loss=0.367, simple_loss=0.416, pruned_loss=0.159, over 5659230.38 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3969, pruned_loss=0.1365, over 5600430.42 frames. ], giga_tot_loss[loss=0.3707, simple_loss=0.4184, pruned_loss=0.1615, over 5658217.56 frames. ], batch size: 307, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:26:52,843 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:27:25,015 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:27:28,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:27:28,431 INFO [train.py:968] (0/2) Epoch 3, batch 7400, libri_loss[loss=0.3755, simple_loss=0.4366, pruned_loss=0.1571, over 29555.00 frames. ], tot_loss[loss=0.3654, simple_loss=0.4138, pruned_loss=0.1585, over 5664526.80 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3972, pruned_loss=0.1365, over 5607200.55 frames. ], giga_tot_loss[loss=0.3692, simple_loss=0.416, pruned_loss=0.1612, over 5659091.24 frames. ], batch size: 83, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:27:53,742 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:28:10,272 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:28:12,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.644e+02 1.485e+03 2.020e+03 3.259e+03 1.012e+04, threshold=4.040e+03, percent-clipped=18.0
+2023-03-01 13:28:16,620 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:28:17,054 INFO [train.py:968] (0/2) Epoch 3, batch 7450, giga_loss[loss=0.3815, simple_loss=0.4197, pruned_loss=0.1716, over 28010.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4128, pruned_loss=0.1578, over 5652842.22 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3976, pruned_loss=0.1367, over 5595633.29 frames. ], giga_tot_loss[loss=0.3674, simple_loss=0.4145, pruned_loss=0.1602, over 5660315.26 frames. ], batch size: 412, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:28:46,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:28:47,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6017, 1.9524, 5.7219, 3.9696], device='cuda:0'), covar=tensor([0.1529, 0.1704, 0.0260, 0.0342], device='cuda:0'), in_proj_covar=tensor([0.0525, 0.0500, 0.0689, 0.0547], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 13:28:47,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:29:05,674 INFO [train.py:968] (0/2) Epoch 3, batch 7500, giga_loss[loss=0.4317, simple_loss=0.4536, pruned_loss=0.2049, over 27864.00 frames. ], tot_loss[loss=0.3618, simple_loss=0.4119, pruned_loss=0.1559, over 5653743.72 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.397, pruned_loss=0.1362, over 5604615.73 frames. ], giga_tot_loss[loss=0.3663, simple_loss=0.4145, pruned_loss=0.1591, over 5653096.68 frames. ], batch size: 412, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:29:18,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6611, 2.6470, 1.5238, 1.4143], device='cuda:0'), covar=tensor([0.0766, 0.0481, 0.0831, 0.1280], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0458, 0.0320, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 13:29:46,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.374e+03 1.758e+03 2.273e+03 5.084e+03, threshold=3.517e+03, percent-clipped=3.0
+2023-03-01 13:29:49,454 INFO [train.py:968] (0/2) Epoch 3, batch 7550, giga_loss[loss=0.4524, simple_loss=0.4597, pruned_loss=0.2226, over 26613.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.4119, pruned_loss=0.1549, over 5666639.45 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.397, pruned_loss=0.1362, over 5613143.34 frames. ], giga_tot_loss[loss=0.3652, simple_loss=0.4144, pruned_loss=0.158, over 5660200.91 frames. ], batch size: 555, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:30:37,249 INFO [train.py:968] (0/2) Epoch 3, batch 7600, giga_loss[loss=0.3745, simple_loss=0.4234, pruned_loss=0.1629, over 28903.00 frames. ], tot_loss[loss=0.36, simple_loss=0.4113, pruned_loss=0.1544, over 5679436.04 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3969, pruned_loss=0.1363, over 5620493.10 frames. ], giga_tot_loss[loss=0.3641, simple_loss=0.4137, pruned_loss=0.1572, over 5668998.65 frames. ], batch size: 174, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:30:53,242 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:30:56,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:31:16,746 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-01 13:31:24,443 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.323e+02 1.613e+03 1.862e+03 2.453e+03 1.006e+04, threshold=3.724e+03, percent-clipped=9.0
+2023-03-01 13:31:26,102 INFO [train.py:968] (0/2) Epoch 3, batch 7650, giga_loss[loss=0.4238, simple_loss=0.4395, pruned_loss=0.2041, over 26783.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4095, pruned_loss=0.1533, over 5683574.22 frames. ], libri_tot_loss[loss=0.3342, simple_loss=0.3966, pruned_loss=0.1359, over 5628033.02 frames. ], giga_tot_loss[loss=0.3623, simple_loss=0.412, pruned_loss=0.1563, over 5670111.42 frames. ], batch size: 555, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:31:29,285 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:31:33,882 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:32:12,052 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-01 13:32:12,894 INFO [train.py:968] (0/2) Epoch 3, batch 7700, libri_loss[loss=0.3563, simple_loss=0.4149, pruned_loss=0.1488, over 29673.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4074, pruned_loss=0.1525, over 5668571.80 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3966, pruned_loss=0.1358, over 5629179.05 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4098, pruned_loss=0.1556, over 5658828.89 frames. ], batch size: 91, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:32:22,317 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-98000.pt
+2023-03-01 13:32:48,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:33:02,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.840e+02 1.607e+03 2.000e+03 2.517e+03 5.833e+03, threshold=4.000e+03, percent-clipped=12.0
+2023-03-01 13:33:04,058 INFO [train.py:968] (0/2) Epoch 3, batch 7750, giga_loss[loss=0.3082, simple_loss=0.3725, pruned_loss=0.1219, over 28590.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.4064, pruned_loss=0.1529, over 5656371.81 frames. ], libri_tot_loss[loss=0.3342, simple_loss=0.3967, pruned_loss=0.1359, over 5623741.38 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4085, pruned_loss=0.1556, over 5653730.70 frames. ], batch size: 60, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:33:29,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=98067.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 13:33:30,361 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-01 13:33:53,131 INFO [train.py:968] (0/2) Epoch 3, batch 7800, giga_loss[loss=0.354, simple_loss=0.4004, pruned_loss=0.1538, over 29049.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4069, pruned_loss=0.1543, over 5649814.60 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.397, pruned_loss=0.1361, over 5619663.89 frames. ], giga_tot_loss[loss=0.3608, simple_loss=0.4085, pruned_loss=0.1566, over 5651809.14 frames. ], batch size: 136, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:34:12,908 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:34:20,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:34:39,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.851e+02 1.593e+03 2.133e+03 2.966e+03 7.431e+03, threshold=4.265e+03, percent-clipped=10.0
+2023-03-01 13:34:40,794 INFO [train.py:968] (0/2) Epoch 3, batch 7850, libri_loss[loss=0.3339, simple_loss=0.4054, pruned_loss=0.1312, over 27878.00 frames. ], tot_loss[loss=0.3563, simple_loss=0.4057, pruned_loss=0.1535, over 5653808.63 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3975, pruned_loss=0.1365, over 5626667.68 frames. ], giga_tot_loss[loss=0.3589, simple_loss=0.4067, pruned_loss=0.1555, over 5649705.58 frames. ], batch size: 116, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:34:42,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:35:05,855 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:35:08,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:35:27,760 INFO [train.py:968] (0/2) Epoch 3, batch 7900, giga_loss[loss=0.408, simple_loss=0.416, pruned_loss=0.2, over 23693.00 frames. ], tot_loss[loss=0.356, simple_loss=0.4051, pruned_loss=0.1535, over 5655532.42 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3968, pruned_loss=0.136, over 5633351.74 frames. ], giga_tot_loss[loss=0.3593, simple_loss=0.4068, pruned_loss=0.1559, over 5646942.27 frames. ], batch size: 705, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:35:31,096 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-01 13:35:38,481 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:14,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.753e+02 1.595e+03 2.043e+03 2.742e+03 6.869e+03, threshold=4.086e+03, percent-clipped=11.0
+2023-03-01 13:36:16,107 INFO [train.py:968] (0/2) Epoch 3, batch 7950, giga_loss[loss=0.3209, simple_loss=0.3877, pruned_loss=0.127, over 28618.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4071, pruned_loss=0.1543, over 5664362.66 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.397, pruned_loss=0.1361, over 5637182.01 frames. ], giga_tot_loss[loss=0.3607, simple_loss=0.4084, pruned_loss=0.1565, over 5654593.55 frames. ], batch size: 85, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:36:27,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:29,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:30,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5976, 2.9248, 1.8691, 0.6312], device='cuda:0'), covar=tensor([0.2286, 0.0856, 0.1368, 0.2044], device='cuda:0'), in_proj_covar=tensor([0.1251, 0.1195, 0.1270, 0.1085], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 13:36:30,929 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:33,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:53,949 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=98281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:58,115 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:58,204 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:37:00,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:37:02,281 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:37:02,711 INFO [train.py:968] (0/2) Epoch 3, batch 8000, giga_loss[loss=0.3727, simple_loss=0.4233, pruned_loss=0.161, over 28930.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4081, pruned_loss=0.1538, over 5671689.60 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3974, pruned_loss=0.1363, over 5643340.30 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.409, pruned_loss=0.1558, over 5659122.32 frames. ], batch size: 213, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:37:03,992 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-01 13:37:25,254 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:37:30,264 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:37:35,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8372, 2.0424, 1.8952, 1.8232], device='cuda:0'), covar=tensor([0.1312, 0.1685, 0.1078, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0801, 0.0732, 0.0633], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 13:37:38,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2646, 1.3932, 1.2575, 0.9306], device='cuda:0'), covar=tensor([0.0684, 0.0513, 0.0342, 0.0522], device='cuda:0'), in_proj_covar=tensor([0.1157, 0.0855, 0.0918, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 13:37:48,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.963e+02 1.422e+03 1.915e+03 2.792e+03 7.933e+03, threshold=3.829e+03, percent-clipped=7.0
+2023-03-01 13:37:52,039 INFO [train.py:968] (0/2) Epoch 3, batch 8050, giga_loss[loss=0.3379, simple_loss=0.3966, pruned_loss=0.1397, over 28865.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.407, pruned_loss=0.1524, over 5685705.40 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3968, pruned_loss=0.136, over 5647400.32 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4084, pruned_loss=0.1544, over 5672805.05 frames. ], batch size: 199, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:38:23,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1881, 1.2810, 1.1549, 1.5552], device='cuda:0'), covar=tensor([0.1976, 0.1786, 0.1618, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.1013, 0.0814, 0.0903, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 13:38:42,505 INFO [train.py:968] (0/2) Epoch 3, batch 8100, giga_loss[loss=0.38, simple_loss=0.4276, pruned_loss=0.1661, over 28581.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4082, pruned_loss=0.1538, over 5675167.89 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3972, pruned_loss=0.1364, over 5643266.14 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4092, pruned_loss=0.1555, over 5669235.06 frames. ], batch size: 307, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:39:00,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6392, 3.3039, 3.3454, 1.6654], device='cuda:0'), covar=tensor([0.0608, 0.0479, 0.0966, 0.1838], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0598, 0.0827, 0.0568], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 13:39:33,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.806e+02 1.645e+03 2.304e+03 3.002e+03 1.041e+04, threshold=4.608e+03, percent-clipped=13.0
+2023-03-01 13:39:35,896 INFO [train.py:968] (0/2) Epoch 3, batch 8150, giga_loss[loss=0.4307, simple_loss=0.4493, pruned_loss=0.206, over 28571.00 frames. ], tot_loss[loss=0.3629, simple_loss=0.4112, pruned_loss=0.1573, over 5665765.69 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3972, pruned_loss=0.1365, over 5639324.10 frames. ], giga_tot_loss[loss=0.3651, simple_loss=0.4123, pruned_loss=0.159, over 5665661.19 frames. ], batch size: 307, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:39:37,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98442.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 13:40:01,173 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:40:04,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:40:29,036 INFO [train.py:968] (0/2) Epoch 3, batch 8200, giga_loss[loss=0.3361, simple_loss=0.3958, pruned_loss=0.1383, over 28934.00 frames. ], tot_loss[loss=0.3662, simple_loss=0.4125, pruned_loss=0.16, over 5651227.13 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3973, pruned_loss=0.1364, over 5642608.25 frames. ], giga_tot_loss[loss=0.369, simple_loss=0.4138, pruned_loss=0.1621, over 5648808.57 frames. ], batch size: 145, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:40:36,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:40:37,014 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=98499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:41:14,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8404, 1.6697, 1.6562, 1.6114], device='cuda:0'), covar=tensor([0.0922, 0.1592, 0.1403, 0.1227], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0785, 0.0638, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 13:41:21,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.115e+02 1.724e+03 2.282e+03 3.672e+03 9.075e+03, threshold=4.563e+03, percent-clipped=11.0
+2023-03-01 13:41:22,827 INFO [train.py:968] (0/2) Epoch 3, batch 8250, giga_loss[loss=0.3581, simple_loss=0.401, pruned_loss=0.1576, over 28878.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.4134, pruned_loss=0.1617, over 5661648.60 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3973, pruned_loss=0.1364, over 5645831.25 frames. ], giga_tot_loss[loss=0.371, simple_loss=0.4146, pruned_loss=0.1637, over 5657090.78 frames. ], batch size: 199, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:41:40,070 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-01 13:42:08,676 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98585.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:42:10,935 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98588.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:42:13,024 INFO [train.py:968] (0/2) Epoch 3, batch 8300, giga_loss[loss=0.3802, simple_loss=0.4257, pruned_loss=0.1674, over 28850.00 frames. ], tot_loss[loss=0.3715, simple_loss=0.4153, pruned_loss=0.1639, over 5657139.66 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3972, pruned_loss=0.1362, over 5648679.89 frames. ], giga_tot_loss[loss=0.3747, simple_loss=0.4167, pruned_loss=0.1664, over 5650999.05 frames. ], batch size: 186, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:42:37,159 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98617.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:42:57,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.653e+02 1.632e+03 2.113e+03 2.792e+03 6.779e+03, threshold=4.227e+03, percent-clipped=5.0
+2023-03-01 13:42:58,077 INFO [train.py:968] (0/2) Epoch 3, batch 8350, giga_loss[loss=0.3951, simple_loss=0.4268, pruned_loss=0.1817, over 28805.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4135, pruned_loss=0.1622, over 5667879.30 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3973, pruned_loss=0.1362, over 5652465.29 frames. ], giga_tot_loss[loss=0.3719, simple_loss=0.4148, pruned_loss=0.1645, over 5659951.07 frames. ], batch size: 243, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:43:12,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:43:43,494 INFO [train.py:968] (0/2) Epoch 3, batch 8400, giga_loss[loss=0.3464, simple_loss=0.4148, pruned_loss=0.139, over 28852.00 frames. ], tot_loss[loss=0.3658, simple_loss=0.4121, pruned_loss=0.1597, over 5673103.28 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.397, pruned_loss=0.136, over 5652177.44 frames. ], giga_tot_loss[loss=0.3689, simple_loss=0.4136, pruned_loss=0.1621, over 5667544.32 frames. ], batch size: 145, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:44:00,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1646, 1.3767, 1.0826, 0.2952], device='cuda:0'), covar=tensor([0.0874, 0.0969, 0.1535, 0.1751], device='cuda:0'), in_proj_covar=tensor([0.1192, 0.1152, 0.1216, 0.1041], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 13:44:24,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.376e+02 1.441e+03 1.769e+03 2.384e+03 6.740e+03, threshold=3.539e+03, percent-clipped=3.0
+2023-03-01 13:44:24,981 INFO [train.py:968] (0/2) Epoch 3, batch 8450, giga_loss[loss=0.3432, simple_loss=0.3864, pruned_loss=0.1501, over 27974.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.408, pruned_loss=0.1553, over 5686505.86 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3962, pruned_loss=0.1356, over 5662839.31 frames. ], giga_tot_loss[loss=0.3637, simple_loss=0.4105, pruned_loss=0.1585, over 5673549.46 frames. ], batch size: 412, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:44:49,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4506, 3.1187, 1.4199, 1.3501], device='cuda:0'), covar=tensor([0.1085, 0.0482, 0.1063, 0.1579], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0459, 0.0317, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 13:45:08,461 INFO [train.py:968] (0/2) Epoch 3, batch 8500, giga_loss[loss=0.3337, simple_loss=0.3896, pruned_loss=0.1389, over 28806.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.4056, pruned_loss=0.1539, over 5685056.81 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3962, pruned_loss=0.1356, over 5663981.11 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.4076, pruned_loss=0.1565, over 5673988.06 frames. ], batch size: 243, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:45:18,571 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:45:22,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:45:23,796 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=98803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:45:47,955 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:45:56,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.569e+02 1.500e+03 1.926e+03 3.010e+03 9.371e+03, threshold=3.853e+03, percent-clipped=17.0
+2023-03-01 13:45:56,074 INFO [train.py:968] (0/2) Epoch 3, batch 8550, giga_loss[loss=0.3294, simple_loss=0.3888, pruned_loss=0.135, over 28894.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4036, pruned_loss=0.1531, over 5682993.23 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.396, pruned_loss=0.1354, over 5667991.23 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.4055, pruned_loss=0.1556, over 5670698.82 frames. ], batch size: 112, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:46:30,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:46:47,923 INFO [train.py:968] (0/2) Epoch 3, batch 8600, libri_loss[loss=0.3577, simple_loss=0.427, pruned_loss=0.1442, over 27759.00 frames. ], tot_loss[loss=0.3592, simple_loss=0.4062, pruned_loss=0.156, over 5673014.75 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.396, pruned_loss=0.1353, over 5668633.17 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.4078, pruned_loss=0.1583, over 5662983.05 frames. ], batch size: 116, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:47:21,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-01 13:47:33,118 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=98936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:47:36,243 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.451e+02 1.642e+03 2.149e+03 2.993e+03 8.224e+03, threshold=4.298e+03, percent-clipped=14.0
+2023-03-01 13:47:36,256 INFO [train.py:968] (0/2) Epoch 3, batch 8650, giga_loss[loss=0.391, simple_loss=0.4361, pruned_loss=0.173, over 28533.00 frames. ], tot_loss[loss=0.3638, simple_loss=0.4106, pruned_loss=0.1586, over 5679181.75 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3962, pruned_loss=0.1355, over 5674736.72 frames. ], giga_tot_loss[loss=0.3668, simple_loss=0.412, pruned_loss=0.1608, over 5665699.36 frames. ], batch size: 71, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:47:56,302 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-01 13:48:14,539 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-01 13:48:25,181 INFO [train.py:968] (0/2) Epoch 3, batch 8700, giga_loss[loss=0.3737, simple_loss=0.4325, pruned_loss=0.1575, over 28809.00 frames. ], tot_loss[loss=0.3649, simple_loss=0.4143, pruned_loss=0.1578, over 5681889.05 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.3964, pruned_loss=0.1356, over 5678623.86 frames. ], giga_tot_loss[loss=0.3676, simple_loss=0.4155, pruned_loss=0.1599, over 5667542.85 frames. ], batch size: 284, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:48:49,208 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:48:50,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:49:07,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.451e+02 1.550e+03 1.824e+03 2.325e+03 5.134e+03, threshold=3.648e+03, percent-clipped=1.0
+2023-03-01 13:49:07,852 INFO [train.py:968] (0/2) Epoch 3, batch 8750, giga_loss[loss=0.3962, simple_loss=0.442, pruned_loss=0.1752, over 28748.00 frames. ], tot_loss[loss=0.3659, simple_loss=0.416, pruned_loss=0.1579, over 5689928.84 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3961, pruned_loss=0.1354, over 5688889.79 frames. ], giga_tot_loss[loss=0.37, simple_loss=0.4182, pruned_loss=0.1609, over 5668736.94 frames. ], batch size: 262, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:49:15,664 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:49:31,740 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-01 13:49:53,279 INFO [train.py:968] (0/2) Epoch 3, batch 8800, giga_loss[loss=0.3753, simple_loss=0.4308, pruned_loss=0.1599, over 28936.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4184, pruned_loss=0.1599, over 5684044.64 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3959, pruned_loss=0.1352, over 5682971.20 frames. ], giga_tot_loss[loss=0.373, simple_loss=0.4206, pruned_loss=0.1627, over 5672427.43 frames. ], batch size: 112, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:50:40,621 INFO [train.py:968] (0/2) Epoch 3, batch 8850, giga_loss[loss=0.3543, simple_loss=0.4031, pruned_loss=0.1528, over 28246.00 frames. ], tot_loss[loss=0.3702, simple_loss=0.4189, pruned_loss=0.1608, over 5691538.72 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3963, pruned_loss=0.1355, over 5684955.17 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.4205, pruned_loss=0.1629, over 5680647.07 frames. ], batch size: 77, lr: 1.05e-02, grad_scale: 2.0
+2023-03-01 13:50:41,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.593e+02 1.625e+03 2.199e+03 2.940e+03 5.944e+03, threshold=4.398e+03, percent-clipped=13.0
+2023-03-01 13:50:46,597 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 13:51:13,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:51:17,464 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9183, 1.3023, 4.1943, 3.2779], device='cuda:0'), covar=tensor([0.1745, 0.1922, 0.0335, 0.0474], device='cuda:0'), in_proj_covar=tensor([0.0535, 0.0508, 0.0685, 0.0545], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 13:51:27,160 INFO [train.py:968] (0/2) Epoch 3, batch 8900, libri_loss[loss=0.3663, simple_loss=0.428, pruned_loss=0.1523, over 29468.00 frames. ], tot_loss[loss=0.3735, simple_loss=0.4202, pruned_loss=0.1634, over 5679004.25 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3966, pruned_loss=0.1356, over 5679811.19 frames. ], giga_tot_loss[loss=0.3764, simple_loss=0.4216, pruned_loss=0.1656, over 5674254.09 frames. ], batch size: 85, lr: 1.05e-02, grad_scale: 2.0
+2023-03-01 13:51:28,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2091, 2.1249, 1.1500, 1.1752], device='cuda:0'), covar=tensor([0.0886, 0.0468, 0.0858, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0466, 0.0319, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 13:51:38,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6251, 1.6055, 1.2059, 0.9642], device='cuda:0'), covar=tensor([0.0711, 0.0590, 0.0552, 0.0713], device='cuda:0'), in_proj_covar=tensor([0.1170, 0.0881, 0.0937, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 13:51:57,849 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4565, 1.9282, 2.0357, 1.8714], device='cuda:0'), covar=tensor([0.0988, 0.2077, 0.1318, 0.1494], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0789, 0.0642, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 13:52:16,047 INFO [train.py:968] (0/2) Epoch 3, batch 8950, giga_loss[loss=0.3704, simple_loss=0.4143, pruned_loss=0.1632, over 29096.00 frames. ], tot_loss[loss=0.3714, simple_loss=0.4179, pruned_loss=0.1624, over 5682103.92 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3965, pruned_loss=0.1355, over 5684342.17 frames. ], giga_tot_loss[loss=0.3746, simple_loss=0.4196, pruned_loss=0.1648, over 5674467.43 frames. ], batch size: 128, lr: 1.05e-02, grad_scale: 2.0
+2023-03-01 13:52:17,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.762e+02 1.724e+03 2.235e+03 3.371e+03 6.476e+03, threshold=4.470e+03, percent-clipped=5.0
+2023-03-01 13:52:28,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5571, 2.2506, 1.8372, 1.7603], device='cuda:0'), covar=tensor([0.1634, 0.1670, 0.1285, 0.0981], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0803, 0.0739, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 13:52:49,291 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:52:55,905 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:52:57,906 INFO [train.py:968] (0/2) Epoch 3, batch 9000, giga_loss[loss=0.3142, simple_loss=0.3753, pruned_loss=0.1266, over 28556.00 frames. ], tot_loss[loss=0.3672, simple_loss=0.4141, pruned_loss=0.1602, over 5688362.43 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3954, pruned_loss=0.135, over 5693512.53 frames. ], giga_tot_loss[loss=0.3723, simple_loss=0.4173, pruned_loss=0.1636, over 5673930.65 frames. ], batch size: 85, lr: 1.04e-02, grad_scale: 2.0
+2023-03-01 13:52:57,911 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 13:53:07,098 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2581, simple_loss=0.3577, pruned_loss=0.07929, over 944034.00 frames. 
+2023-03-01 13:53:07,099 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19537MB
+2023-03-01 13:53:19,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-01 13:53:26,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:53:34,870 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:53:36,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:53:48,947 INFO [train.py:968] (0/2) Epoch 3, batch 9050, giga_loss[loss=0.3717, simple_loss=0.4245, pruned_loss=0.1594, over 28910.00 frames. ], tot_loss[loss=0.3641, simple_loss=0.4116, pruned_loss=0.1583, over 5696325.82 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3951, pruned_loss=0.1349, over 5702790.53 frames. ], giga_tot_loss[loss=0.3699, simple_loss=0.4153, pruned_loss=0.1623, over 5675459.58 frames. ], batch size: 213, lr: 1.04e-02, grad_scale: 2.0
+2023-03-01 13:53:51,168 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.576e+02 1.582e+03 1.937e+03 2.669e+03 8.616e+03, threshold=3.873e+03, percent-clipped=6.0
+2023-03-01 13:54:01,067 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:54:12,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:54:21,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1792, 2.0811, 1.1975, 1.1295], device='cuda:0'), covar=tensor([0.0853, 0.0474, 0.0788, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0465, 0.0318, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 13:54:37,363 INFO [train.py:968] (0/2) Epoch 3, batch 9100, giga_loss[loss=0.4384, simple_loss=0.4434, pruned_loss=0.2167, over 23629.00 frames. ], tot_loss[loss=0.3646, simple_loss=0.412, pruned_loss=0.1586, over 5695287.24 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.395, pruned_loss=0.1347, over 5709303.17 frames. ], giga_tot_loss[loss=0.371, simple_loss=0.4158, pruned_loss=0.1631, over 5671660.30 frames. ], batch size: 705, lr: 1.04e-02, grad_scale: 2.0
+2023-03-01 13:55:23,552 INFO [train.py:968] (0/2) Epoch 3, batch 9150, giga_loss[loss=0.3113, simple_loss=0.3676, pruned_loss=0.1275, over 28425.00 frames. ], tot_loss[loss=0.366, simple_loss=0.4128, pruned_loss=0.1596, over 5690080.00 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.395, pruned_loss=0.1346, over 5710392.21 frames. ], giga_tot_loss[loss=0.372, simple_loss=0.4163, pruned_loss=0.1639, over 5669878.95 frames. ], batch size: 78, lr: 1.04e-02, grad_scale: 2.0
+2023-03-01 13:55:26,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.573e+02 1.597e+03 2.019e+03 2.923e+03 1.007e+04, threshold=4.038e+03, percent-clipped=10.0
+2023-03-01 13:55:27,982 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:55:36,952 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:55:39,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:56:06,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:56:11,232 INFO [train.py:968] (0/2) Epoch 3, batch 9200, libri_loss[loss=0.3158, simple_loss=0.3818, pruned_loss=0.1249, over 29588.00 frames. ], tot_loss[loss=0.3647, simple_loss=0.4112, pruned_loss=0.1591, over 5691684.77 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3948, pruned_loss=0.1345, over 5713399.85 frames. ], giga_tot_loss[loss=0.3703, simple_loss=0.4145, pruned_loss=0.1631, over 5672429.08 frames. ], batch size: 75, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 13:56:37,284 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-01 13:56:37,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7881, 3.4186, 3.5056, 1.5892], device='cuda:0'), covar=tensor([0.0554, 0.0422, 0.0877, 0.1934], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0611, 0.0834, 0.0567], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 13:56:56,934 INFO [train.py:968] (0/2) Epoch 3, batch 9250, giga_loss[loss=0.4643, simple_loss=0.4692, pruned_loss=0.2297, over 26552.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.4094, pruned_loss=0.1578, over 5695806.82 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3942, pruned_loss=0.1342, over 5718717.04 frames. ], giga_tot_loss[loss=0.3689, simple_loss=0.4132, pruned_loss=0.1623, over 5674185.33 frames. ], batch size: 555, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 13:56:58,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.000e+02 1.524e+03 1.969e+03 3.116e+03 8.495e+03, threshold=3.938e+03, percent-clipped=11.0
+2023-03-01 13:57:48,286 INFO [train.py:968] (0/2) Epoch 3, batch 9300, giga_loss[loss=0.3983, simple_loss=0.4476, pruned_loss=0.1745, over 28621.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.4128, pruned_loss=0.16, over 5682124.10 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3943, pruned_loss=0.1342, over 5716905.72 frames. ], giga_tot_loss[loss=0.3717, simple_loss=0.4159, pruned_loss=0.1638, over 5666264.47 frames. ], batch size: 262, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 13:58:18,584 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:58:22,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2871, 1.5295, 1.1401, 1.4908], device='cuda:0'), covar=tensor([0.0800, 0.0480, 0.0403, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0249, 0.0163, 0.0163, 0.0297], device='cuda:0'), out_proj_covar=tensor([0.0039, 0.0029, 0.0025, 0.0044], device='cuda:0')
+2023-03-01 13:58:34,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4558, 1.3182, 1.2034, 1.4007], device='cuda:0'), covar=tensor([0.1890, 0.1983, 0.1709, 0.2066], device='cuda:0'), in_proj_covar=tensor([0.1026, 0.0839, 0.0920, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 13:58:34,926 INFO [train.py:968] (0/2) Epoch 3, batch 9350, giga_loss[loss=0.3504, simple_loss=0.4108, pruned_loss=0.145, over 28992.00 frames. ], tot_loss[loss=0.3699, simple_loss=0.4158, pruned_loss=0.162, over 5688061.13 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3942, pruned_loss=0.1341, over 5720957.19 frames. ], giga_tot_loss[loss=0.375, simple_loss=0.4187, pruned_loss=0.1656, over 5671155.35 frames. ], batch size: 128, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 13:58:36,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.261e+02 1.499e+03 2.064e+03 2.809e+03 6.487e+03, threshold=4.128e+03, percent-clipped=8.0
+2023-03-01 13:58:52,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:58:55,473 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:58:57,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:59:23,569 INFO [train.py:968] (0/2) Epoch 3, batch 9400, giga_loss[loss=0.3812, simple_loss=0.421, pruned_loss=0.1707, over 28560.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4147, pruned_loss=0.1616, over 5682680.16 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3944, pruned_loss=0.1341, over 5724484.54 frames. ], giga_tot_loss[loss=0.3736, simple_loss=0.4172, pruned_loss=0.165, over 5665493.62 frames. ], batch size: 336, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 13:59:38,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-01 13:59:46,258 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:00:07,823 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:00:08,827 INFO [train.py:968] (0/2) Epoch 3, batch 9450, giga_loss[loss=0.3278, simple_loss=0.4169, pruned_loss=0.1194, over 28958.00 frames. ], tot_loss[loss=0.3666, simple_loss=0.4153, pruned_loss=0.1589, over 5687703.12 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3948, pruned_loss=0.1344, over 5725904.47 frames. ], giga_tot_loss[loss=0.3708, simple_loss=0.4175, pruned_loss=0.1621, over 5671718.05 frames. ], batch size: 145, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:00:10,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.519e+02 1.531e+03 2.110e+03 2.790e+03 6.038e+03, threshold=4.220e+03, percent-clipped=4.0
+2023-03-01 14:00:18,641 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 14:00:34,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0709, 1.8036, 1.3945, 1.5456], device='cuda:0'), covar=tensor([0.0640, 0.0805, 0.0969, 0.1086], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0484, 0.0522, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 14:00:51,335 INFO [train.py:968] (0/2) Epoch 3, batch 9500, giga_loss[loss=0.3877, simple_loss=0.4389, pruned_loss=0.1683, over 28977.00 frames. ], tot_loss[loss=0.3639, simple_loss=0.4153, pruned_loss=0.1562, over 5679381.59 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3947, pruned_loss=0.1343, over 5717925.90 frames. ], giga_tot_loss[loss=0.368, simple_loss=0.4175, pruned_loss=0.1592, over 5673450.44 frames. ], batch size: 106, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:01:01,810 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:04,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:06,108 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:08,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:11,531 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:18,165 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:27,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4446, 3.3599, 1.5863, 1.3321], device='cuda:0'), covar=tensor([0.0933, 0.0420, 0.0840, 0.1391], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0467, 0.0322, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 14:01:32,052 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:37,754 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:40,307 INFO [train.py:968] (0/2) Epoch 3, batch 9550, giga_loss[loss=0.3975, simple_loss=0.4426, pruned_loss=0.1762, over 28742.00 frames. ], tot_loss[loss=0.3678, simple_loss=0.4184, pruned_loss=0.1586, over 5674954.44 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3943, pruned_loss=0.134, over 5720842.29 frames. ], giga_tot_loss[loss=0.3719, simple_loss=0.4208, pruned_loss=0.1615, over 5667047.52 frames. ], batch size: 119, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:01:41,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.844e+02 1.396e+03 1.839e+03 2.363e+03 5.602e+03, threshold=3.678e+03, percent-clipped=5.0
+2023-03-01 14:01:58,453 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99861.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:02:21,334 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:02:25,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:02:28,856 INFO [train.py:968] (0/2) Epoch 3, batch 9600, giga_loss[loss=0.4164, simple_loss=0.4497, pruned_loss=0.1915, over 28555.00 frames. ], tot_loss[loss=0.3722, simple_loss=0.421, pruned_loss=0.1617, over 5677859.20 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3942, pruned_loss=0.134, over 5723184.53 frames. ], giga_tot_loss[loss=0.3762, simple_loss=0.4234, pruned_loss=0.1645, over 5668716.75 frames. ], batch size: 336, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:02:32,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6816, 3.6270, 1.7388, 1.5416], device='cuda:0'), covar=tensor([0.0800, 0.0396, 0.0754, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0467, 0.0321, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 14:02:50,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:03:11,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1587, 3.8216, 3.9127, 1.5961], device='cuda:0'), covar=tensor([0.0502, 0.0419, 0.0847, 0.1997], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0615, 0.0829, 0.0569], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 14:03:16,302 INFO [train.py:968] (0/2) Epoch 3, batch 9650, giga_loss[loss=0.3999, simple_loss=0.443, pruned_loss=0.1784, over 28735.00 frames. ], tot_loss[loss=0.3762, simple_loss=0.4229, pruned_loss=0.1647, over 5670259.46 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3942, pruned_loss=0.134, over 5717639.57 frames. ], giga_tot_loss[loss=0.3803, simple_loss=0.4255, pruned_loss=0.1676, over 5666101.02 frames. ], batch size: 85, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:03:18,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.895e+02 1.566e+03 2.050e+03 3.473e+03 9.881e+03, threshold=4.100e+03, percent-clipped=24.0
+2023-03-01 14:03:34,414 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:03:35,224 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:03:37,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:03:56,340 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:04:00,552 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-01 14:04:02,040 INFO [train.py:968] (0/2) Epoch 3, batch 9700, giga_loss[loss=0.3649, simple_loss=0.4147, pruned_loss=0.1576, over 28997.00 frames. ], tot_loss[loss=0.377, simple_loss=0.4228, pruned_loss=0.1656, over 5658065.03 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3944, pruned_loss=0.1341, over 5711909.71 frames. ], giga_tot_loss[loss=0.3812, simple_loss=0.4253, pruned_loss=0.1685, over 5659342.10 frames. ], batch size: 164, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:04:05,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:04:09,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4204, 4.0257, 4.1857, 1.8879], device='cuda:0'), covar=tensor([0.0347, 0.0352, 0.0653, 0.1831], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0613, 0.0830, 0.0569], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 14:04:09,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:04:11,905 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-100000.pt
+2023-03-01 14:04:44,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:04:47,679 INFO [train.py:968] (0/2) Epoch 3, batch 9750, libri_loss[loss=0.3782, simple_loss=0.4347, pruned_loss=0.1609, over 29393.00 frames. ], tot_loss[loss=0.3724, simple_loss=0.4198, pruned_loss=0.1625, over 5657987.37 frames. ], libri_tot_loss[loss=0.3309, simple_loss=0.3942, pruned_loss=0.1338, over 5716684.98 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4226, pruned_loss=0.1657, over 5653293.47 frames. ], batch size: 92, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:04:49,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.691e+02 1.733e+03 2.259e+03 3.440e+03 7.559e+03, threshold=4.518e+03, percent-clipped=15.0
+2023-03-01 14:05:17,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:05:17,405 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-03-01 14:05:31,960 INFO [train.py:968] (0/2) Epoch 3, batch 9800, giga_loss[loss=0.3536, simple_loss=0.4185, pruned_loss=0.1443, over 29006.00 frames. ], tot_loss[loss=0.3695, simple_loss=0.4191, pruned_loss=0.16, over 5661337.01 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3943, pruned_loss=0.1339, over 5717392.87 frames. ], giga_tot_loss[loss=0.374, simple_loss=0.4217, pruned_loss=0.1631, over 5655962.18 frames. ], batch size: 145, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:05:32,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:06:14,505 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:06:14,915 INFO [train.py:968] (0/2) Epoch 3, batch 9850, giga_loss[loss=0.3407, simple_loss=0.4077, pruned_loss=0.1368, over 28951.00 frames. ], tot_loss[loss=0.3694, simple_loss=0.42, pruned_loss=0.1595, over 5662922.85 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3946, pruned_loss=0.1341, over 5712919.28 frames. ], giga_tot_loss[loss=0.3739, simple_loss=0.4226, pruned_loss=0.1626, over 5661277.17 frames. ], batch size: 164, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:06:17,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:06:17,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.093e+02 1.433e+03 1.904e+03 2.496e+03 5.986e+03, threshold=3.808e+03, percent-clipped=4.0
+2023-03-01 14:06:36,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6905, 2.1929, 1.7534, 1.9116], device='cuda:0'), covar=tensor([0.0518, 0.0657, 0.0825, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0488, 0.0532, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 14:06:38,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3624, 2.0016, 1.4096, 0.5441], device='cuda:0'), covar=tensor([0.1569, 0.0959, 0.1391, 0.1928], device='cuda:0'), in_proj_covar=tensor([0.1234, 0.1178, 0.1246, 0.1048], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 14:06:47,940 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:06:54,216 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:06:54,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1956, 1.2715, 1.1093, 1.3597], device='cuda:0'), covar=tensor([0.2236, 0.2117, 0.2033, 0.1994], device='cuda:0'), in_proj_covar=tensor([0.1017, 0.0820, 0.0904, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 14:06:57,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:07:01,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:07:02,693 INFO [train.py:968] (0/2) Epoch 3, batch 9900, giga_loss[loss=0.3428, simple_loss=0.4062, pruned_loss=0.1397, over 28941.00 frames. ], tot_loss[loss=0.3703, simple_loss=0.4206, pruned_loss=0.16, over 5665536.43 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3956, pruned_loss=0.1347, over 5717460.03 frames. ], giga_tot_loss[loss=0.3743, simple_loss=0.4228, pruned_loss=0.1629, over 5658406.39 frames. ], batch size: 136, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:07:23,128 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:07:30,868 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 14:07:47,627 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:07:48,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100236.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:07:51,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:07:53,337 INFO [train.py:968] (0/2) Epoch 3, batch 9950, giga_loss[loss=0.3584, simple_loss=0.41, pruned_loss=0.1534, over 28868.00 frames. ], tot_loss[loss=0.3706, simple_loss=0.4205, pruned_loss=0.1603, over 5666708.36 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3958, pruned_loss=0.1346, over 5720995.15 frames. ], giga_tot_loss[loss=0.3746, simple_loss=0.4227, pruned_loss=0.1633, over 5656774.31 frames. ], batch size: 174, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:07:56,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.611e+03 2.004e+03 2.783e+03 4.588e+03, threshold=4.009e+03, percent-clipped=8.0
+2023-03-01 14:08:16,971 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:08:40,234 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-01 14:08:42,703 INFO [train.py:968] (0/2) Epoch 3, batch 10000, giga_loss[loss=0.3917, simple_loss=0.4316, pruned_loss=0.1759, over 28576.00 frames. ], tot_loss[loss=0.3709, simple_loss=0.4194, pruned_loss=0.1612, over 5666100.83 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3961, pruned_loss=0.1348, over 5720036.18 frames. ], giga_tot_loss[loss=0.3739, simple_loss=0.4209, pruned_loss=0.1635, over 5658891.35 frames. ], batch size: 85, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:08:51,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7073, 2.0518, 1.9295, 1.8484], device='cuda:0'), covar=tensor([0.1486, 0.1708, 0.1109, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0815, 0.0745, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 14:09:19,287 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:09:21,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:09:24,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:09:30,589 INFO [train.py:968] (0/2) Epoch 3, batch 10050, giga_loss[loss=0.4223, simple_loss=0.4297, pruned_loss=0.2074, over 23309.00 frames. ], tot_loss[loss=0.3697, simple_loss=0.4177, pruned_loss=0.1609, over 5661703.99 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3965, pruned_loss=0.1352, over 5721181.93 frames. ], giga_tot_loss[loss=0.3724, simple_loss=0.419, pruned_loss=0.1629, over 5653943.07 frames. ], batch size: 705, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:09:33,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.549e+02 1.589e+03 2.094e+03 2.903e+03 6.315e+03, threshold=4.187e+03, percent-clipped=8.0
+2023-03-01 14:09:45,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:09:51,773 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:10:08,452 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100379.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:10:09,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7462, 5.3511, 5.3545, 2.7302], device='cuda:0'), covar=tensor([0.0476, 0.0572, 0.1162, 0.1716], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0605, 0.0819, 0.0561], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 14:10:10,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100382.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:10:19,823 INFO [train.py:968] (0/2) Epoch 3, batch 10100, giga_loss[loss=0.3148, simple_loss=0.3796, pruned_loss=0.125, over 28740.00 frames. ], tot_loss[loss=0.367, simple_loss=0.4151, pruned_loss=0.1594, over 5673781.33 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3966, pruned_loss=0.1352, over 5726127.74 frames. ], giga_tot_loss[loss=0.3699, simple_loss=0.4165, pruned_loss=0.1616, over 5661787.02 frames. ], batch size: 119, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:10:29,617 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100400.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:10:40,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100411.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:11:06,387 INFO [train.py:968] (0/2) Epoch 3, batch 10150, giga_loss[loss=0.3577, simple_loss=0.4041, pruned_loss=0.1557, over 29014.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.414, pruned_loss=0.1595, over 5676828.75 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3965, pruned_loss=0.1349, over 5730479.59 frames. ], giga_tot_loss[loss=0.3701, simple_loss=0.4157, pruned_loss=0.1622, over 5662073.89 frames. ], batch size: 128, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:11:09,351 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.633e+03 2.101e+03 2.841e+03 5.392e+03, threshold=4.202e+03, percent-clipped=5.0
+2023-03-01 14:11:12,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:11:34,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9130, 1.7087, 1.6497, 1.6806], device='cuda:0'), covar=tensor([0.0994, 0.1861, 0.1530, 0.1420], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0786, 0.0638, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 14:11:42,508 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:11:45,675 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:11:53,392 INFO [train.py:968] (0/2) Epoch 3, batch 10200, giga_loss[loss=0.3306, simple_loss=0.3647, pruned_loss=0.1482, over 23765.00 frames. ], tot_loss[loss=0.3654, simple_loss=0.4128, pruned_loss=0.159, over 5677825.30 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3961, pruned_loss=0.1346, over 5734693.37 frames. ], giga_tot_loss[loss=0.3694, simple_loss=0.4149, pruned_loss=0.162, over 5661029.99 frames. ], batch size: 705, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:12:01,746 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:12:03,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2967, 1.5702, 1.1224, 1.3841], device='cuda:0'), covar=tensor([0.0880, 0.0418, 0.0407, 0.0998], device='cuda:0'), in_proj_covar=tensor([0.0252, 0.0166, 0.0165, 0.0299], device='cuda:0'), out_proj_covar=tensor([0.0040, 0.0030, 0.0026, 0.0044], device='cuda:0')
+2023-03-01 14:12:03,970 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:12:13,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:12:32,664 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:12:39,124 INFO [train.py:968] (0/2) Epoch 3, batch 10250, giga_loss[loss=0.3663, simple_loss=0.4176, pruned_loss=0.1575, over 27644.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.4101, pruned_loss=0.1558, over 5673115.15 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3965, pruned_loss=0.1348, over 5737583.06 frames. ], giga_tot_loss[loss=0.3643, simple_loss=0.4118, pruned_loss=0.1584, over 5655786.23 frames. ], batch size: 472, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:12:42,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.603e+02 1.522e+03 1.919e+03 2.949e+03 6.959e+03, threshold=3.837e+03, percent-clipped=7.0
+2023-03-01 14:13:24,297 INFO [train.py:968] (0/2) Epoch 3, batch 10300, giga_loss[loss=0.3391, simple_loss=0.4015, pruned_loss=0.1384, over 28822.00 frames. ], tot_loss[loss=0.3568, simple_loss=0.4075, pruned_loss=0.1531, over 5675399.79 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3966, pruned_loss=0.1349, over 5742855.98 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.4091, pruned_loss=0.1557, over 5654763.60 frames. ], batch size: 145, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:13:24,571 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:13:28,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:13:52,777 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:13:59,289 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:13:59,557 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-01 14:14:08,503 INFO [train.py:968] (0/2) Epoch 3, batch 10350, giga_loss[loss=0.4084, simple_loss=0.4394, pruned_loss=0.1887, over 27657.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.4058, pruned_loss=0.1514, over 5675942.48 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3964, pruned_loss=0.1347, over 5744225.21 frames. ], giga_tot_loss[loss=0.358, simple_loss=0.4076, pruned_loss=0.1542, over 5655753.93 frames. ], batch size: 472, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:14:11,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.043e+02 1.378e+03 1.724e+03 2.281e+03 5.679e+03, threshold=3.447e+03, percent-clipped=6.0
+2023-03-01 14:14:29,747 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-01 14:14:47,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5504, 1.5204, 0.9421, 1.2362], device='cuda:0'), covar=tensor([0.0823, 0.0893, 0.1680, 0.1326], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0495, 0.0536, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 14:14:56,643 INFO [train.py:968] (0/2) Epoch 3, batch 10400, giga_loss[loss=0.3911, simple_loss=0.4123, pruned_loss=0.1849, over 26718.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4023, pruned_loss=0.1499, over 5667801.28 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3961, pruned_loss=0.1346, over 5737599.85 frames. ], giga_tot_loss[loss=0.3548, simple_loss=0.4042, pruned_loss=0.1527, over 5654758.09 frames. ], batch size: 555, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:15:43,130 INFO [train.py:968] (0/2) Epoch 3, batch 10450, giga_loss[loss=0.3247, simple_loss=0.3874, pruned_loss=0.131, over 28979.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.4004, pruned_loss=0.1494, over 5666240.80 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3961, pruned_loss=0.1345, over 5731034.59 frames. ], giga_tot_loss[loss=0.3529, simple_loss=0.402, pruned_loss=0.1519, over 5659779.70 frames. ], batch size: 164, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:15:48,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.142e+02 1.673e+03 2.344e+03 3.124e+03 5.816e+03, threshold=4.687e+03, percent-clipped=17.0
+2023-03-01 14:16:11,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100775.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:16:25,519 INFO [train.py:968] (0/2) Epoch 3, batch 10500, giga_loss[loss=0.3806, simple_loss=0.4217, pruned_loss=0.1697, over 27921.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4034, pruned_loss=0.1508, over 5670153.46 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3962, pruned_loss=0.1345, over 5733533.59 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4047, pruned_loss=0.1534, over 5660881.69 frames. ], batch size: 412, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:16:25,938 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 14:16:28,348 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:16:43,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4982, 1.5410, 1.3083, 0.7955], device='cuda:0'), covar=tensor([0.0701, 0.0563, 0.0456, 0.0700], device='cuda:0'), in_proj_covar=tensor([0.1146, 0.0910, 0.0943, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 14:16:59,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4199, 2.0784, 1.4921, 0.7579], device='cuda:0'), covar=tensor([0.2059, 0.0998, 0.1670, 0.1975], device='cuda:0'), in_proj_covar=tensor([0.1248, 0.1194, 0.1251, 0.1067], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 14:17:10,576 INFO [train.py:968] (0/2) Epoch 3, batch 10550, giga_loss[loss=0.4638, simple_loss=0.4674, pruned_loss=0.2302, over 26514.00 frames. ], tot_loss[loss=0.3535, simple_loss=0.4046, pruned_loss=0.1512, over 5664574.37 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3963, pruned_loss=0.1346, over 5725855.53 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.4057, pruned_loss=0.1534, over 5663383.25 frames. ], batch size: 555, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:17:15,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.015e+02 1.400e+03 1.977e+03 3.003e+03 6.026e+03, threshold=3.954e+03, percent-clipped=7.0
+2023-03-01 14:18:00,480 INFO [train.py:968] (0/2) Epoch 3, batch 10600, giga_loss[loss=0.3302, simple_loss=0.389, pruned_loss=0.1357, over 29031.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.4035, pruned_loss=0.1509, over 5631410.24 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3966, pruned_loss=0.1347, over 5718604.55 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4043, pruned_loss=0.1529, over 5635957.87 frames. ], batch size: 155, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:18:08,609 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-01 14:18:25,189 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100918.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:18:27,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100921.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:18:29,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2213, 1.8197, 1.5868, 1.5935], device='cuda:0'), covar=tensor([0.1188, 0.1806, 0.1151, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0821, 0.0744, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 14:18:45,864 INFO [train.py:968] (0/2) Epoch 3, batch 10650, giga_loss[loss=0.3255, simple_loss=0.3829, pruned_loss=0.1341, over 28905.00 frames. ], tot_loss[loss=0.3545, simple_loss=0.4046, pruned_loss=0.1522, over 5633382.99 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.397, pruned_loss=0.135, over 5721370.36 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.405, pruned_loss=0.1539, over 5632229.55 frames. ], batch size: 213, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:18:49,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.321e+02 1.450e+03 1.777e+03 2.258e+03 4.627e+03, threshold=3.554e+03, percent-clipped=1.0
+2023-03-01 14:18:52,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100950.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:19:21,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5156, 1.3855, 1.3171, 1.8086], device='cuda:0'), covar=tensor([0.1867, 0.1877, 0.1621, 0.2055], device='cuda:0'), in_proj_covar=tensor([0.1017, 0.0822, 0.0907, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 14:19:27,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3617, 1.4822, 1.5116, 1.5398], device='cuda:0'), covar=tensor([0.0934, 0.1019, 0.1009, 0.0923], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0782, 0.0635, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 14:19:30,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7517, 3.4043, 3.4932, 1.5325], device='cuda:0'), covar=tensor([0.0506, 0.0446, 0.0720, 0.2137], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0613, 0.0824, 0.0571], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 14:19:36,327 INFO [train.py:968] (0/2) Epoch 3, batch 10700, giga_loss[loss=0.3406, simple_loss=0.3996, pruned_loss=0.1408, over 28810.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.4069, pruned_loss=0.1542, over 5643032.19 frames. ], libri_tot_loss[loss=0.3342, simple_loss=0.3976, pruned_loss=0.1354, over 5726897.43 frames. ], giga_tot_loss[loss=0.3592, simple_loss=0.4069, pruned_loss=0.1558, over 5634931.45 frames. ], batch size: 199, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:19:52,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:20:14,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0769, 0.8230, 0.6972, 1.2886], device='cuda:0'), covar=tensor([0.0909, 0.0404, 0.0457, 0.1037], device='cuda:0'), in_proj_covar=tensor([0.0248, 0.0165, 0.0164, 0.0296], device='cuda:0'), out_proj_covar=tensor([0.0039, 0.0030, 0.0026, 0.0044], device='cuda:0')
+2023-03-01 14:20:25,297 INFO [train.py:968] (0/2) Epoch 3, batch 10750, giga_loss[loss=0.36, simple_loss=0.4122, pruned_loss=0.1538, over 29021.00 frames. ], tot_loss[loss=0.3607, simple_loss=0.4097, pruned_loss=0.1559, over 5653957.29 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3979, pruned_loss=0.1355, over 5730311.22 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.4096, pruned_loss=0.1574, over 5642983.90 frames. ], batch size: 155, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:20:28,259 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-01 14:20:29,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.675e+03 2.143e+03 2.722e+03 5.226e+03, threshold=4.287e+03, percent-clipped=8.0
+2023-03-01 14:20:32,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4342, 1.4243, 1.5813, 1.5804], device='cuda:0'), covar=tensor([0.0741, 0.0990, 0.0872, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0452, 0.0795, 0.0638, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 14:21:12,201 INFO [train.py:968] (0/2) Epoch 3, batch 10800, giga_loss[loss=0.3599, simple_loss=0.4101, pruned_loss=0.1549, over 28701.00 frames. ], tot_loss[loss=0.3627, simple_loss=0.4113, pruned_loss=0.1571, over 5653038.04 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3981, pruned_loss=0.1355, over 5727679.78 frames. ], giga_tot_loss[loss=0.3647, simple_loss=0.4115, pruned_loss=0.1589, over 5644725.75 frames. ], batch size: 307, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:21:20,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6991, 2.9417, 1.6680, 1.3663], device='cuda:0'), covar=tensor([0.0815, 0.0430, 0.0773, 0.1402], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0470, 0.0321, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0018], device='cuda:0')
+2023-03-01 14:21:42,408 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=101123.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:22:00,141 INFO [train.py:968] (0/2) Epoch 3, batch 10850, giga_loss[loss=0.3457, simple_loss=0.4001, pruned_loss=0.1457, over 28969.00 frames. ], tot_loss[loss=0.3654, simple_loss=0.4128, pruned_loss=0.159, over 5656935.44 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3974, pruned_loss=0.135, over 5731247.49 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.4139, pruned_loss=0.1613, over 5645694.56 frames. ], batch size: 164, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:22:04,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.722e+02 1.563e+03 2.112e+03 2.811e+03 8.034e+03, threshold=4.224e+03, percent-clipped=9.0
+2023-03-01 14:22:05,890 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:22:07,954 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:22:26,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:22:35,268 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:22:44,249 INFO [train.py:968] (0/2) Epoch 3, batch 10900, giga_loss[loss=0.3497, simple_loss=0.4118, pruned_loss=0.1438, over 28742.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.4135, pruned_loss=0.1597, over 5657687.26 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.397, pruned_loss=0.135, over 5735575.96 frames. ], giga_tot_loss[loss=0.3701, simple_loss=0.4152, pruned_loss=0.1624, over 5642269.67 frames. ], batch size: 243, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:23:36,979 INFO [train.py:968] (0/2) Epoch 3, batch 10950, giga_loss[loss=0.4082, simple_loss=0.4425, pruned_loss=0.1869, over 28246.00 frames. ], tot_loss[loss=0.3662, simple_loss=0.4143, pruned_loss=0.1591, over 5648152.74 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3971, pruned_loss=0.1351, over 5728171.38 frames. ], giga_tot_loss[loss=0.3694, simple_loss=0.4158, pruned_loss=0.1615, over 5641415.36 frames. ], batch size: 368, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:23:43,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.712e+02 1.642e+03 2.166e+03 3.030e+03 7.135e+03, threshold=4.331e+03, percent-clipped=10.0
+2023-03-01 14:24:25,762 INFO [train.py:968] (0/2) Epoch 3, batch 11000, giga_loss[loss=0.3586, simple_loss=0.4077, pruned_loss=0.1548, over 28897.00 frames. ], tot_loss[loss=0.3657, simple_loss=0.4133, pruned_loss=0.1591, over 5649894.77 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3966, pruned_loss=0.1348, over 5729730.25 frames. ], giga_tot_loss[loss=0.3697, simple_loss=0.4154, pruned_loss=0.162, over 5640775.24 frames. ], batch size: 227, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:24:47,205 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:24:50,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:25:17,792 INFO [train.py:968] (0/2) Epoch 3, batch 11050, giga_loss[loss=0.3799, simple_loss=0.4223, pruned_loss=0.1687, over 28727.00 frames. ], tot_loss[loss=0.3639, simple_loss=0.4115, pruned_loss=0.1582, over 5661460.39 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3967, pruned_loss=0.1348, over 5729067.86 frames. ], giga_tot_loss[loss=0.3674, simple_loss=0.4134, pruned_loss=0.1607, over 5653604.85 frames. ], batch size: 119, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:25:21,311 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:25:22,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.658e+03 2.276e+03 3.081e+03 6.352e+03, threshold=4.553e+03, percent-clipped=11.0
+2023-03-01 14:25:40,378 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1608, 1.4010, 1.0856, 0.3834], device='cuda:0'), covar=tensor([0.0933, 0.1001, 0.1635, 0.1624], device='cuda:0'), in_proj_covar=tensor([0.1240, 0.1191, 0.1243, 0.1054], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 14:25:47,873 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-01 14:26:09,826 INFO [train.py:968] (0/2) Epoch 3, batch 11100, libri_loss[loss=0.3294, simple_loss=0.3974, pruned_loss=0.1307, over 29292.00 frames. ], tot_loss[loss=0.3634, simple_loss=0.411, pruned_loss=0.1579, over 5649080.69 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3967, pruned_loss=0.1347, over 5724541.94 frames. ], giga_tot_loss[loss=0.3677, simple_loss=0.4131, pruned_loss=0.1611, over 5643997.90 frames. ], batch size: 94, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:26:15,105 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 14:26:51,142 INFO [train.py:968] (0/2) Epoch 3, batch 11150, giga_loss[loss=0.5064, simple_loss=0.4919, pruned_loss=0.2605, over 26580.00 frames. ], tot_loss[loss=0.3617, simple_loss=0.4098, pruned_loss=0.1568, over 5666713.74 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3958, pruned_loss=0.1341, over 5727865.99 frames. ], giga_tot_loss[loss=0.3674, simple_loss=0.413, pruned_loss=0.1609, over 5656961.83 frames. ], batch size: 555, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:26:56,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.030e+02 1.465e+03 1.840e+03 2.387e+03 5.746e+03, threshold=3.681e+03, percent-clipped=4.0
+2023-03-01 14:26:59,580 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-01 14:27:20,432 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-03-01 14:27:34,863 INFO [train.py:968] (0/2) Epoch 3, batch 11200, giga_loss[loss=0.3464, simple_loss=0.4019, pruned_loss=0.1455, over 28830.00 frames. ], tot_loss[loss=0.3633, simple_loss=0.4102, pruned_loss=0.1582, over 5672111.26 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.396, pruned_loss=0.1343, over 5732672.27 frames. ], giga_tot_loss[loss=0.3684, simple_loss=0.4129, pruned_loss=0.1619, over 5658532.14 frames. ], batch size: 199, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:27:37,758 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2457, 1.6619, 1.2089, 1.3564], device='cuda:0'), covar=tensor([0.0909, 0.0365, 0.0410, 0.1031], device='cuda:0'), in_proj_covar=tensor([0.0246, 0.0163, 0.0163, 0.0293], device='cuda:0'), out_proj_covar=tensor([0.0039, 0.0029, 0.0026, 0.0044], device='cuda:0')
+2023-03-01 14:27:40,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101498.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:27:42,389 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=101500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:28:11,189 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-01 14:28:21,429 INFO [train.py:968] (0/2) Epoch 3, batch 11250, giga_loss[loss=0.3547, simple_loss=0.4117, pruned_loss=0.1489, over 28980.00 frames. ], tot_loss[loss=0.365, simple_loss=0.4114, pruned_loss=0.1594, over 5668395.82 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3963, pruned_loss=0.1346, over 5733174.50 frames. ], giga_tot_loss[loss=0.3695, simple_loss=0.4136, pruned_loss=0.1627, over 5655723.89 frames. ], batch size: 164, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:28:28,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.766e+02 1.774e+03 2.353e+03 3.378e+03 1.057e+04, threshold=4.706e+03, percent-clipped=17.0
+2023-03-01 14:29:12,750 INFO [train.py:968] (0/2) Epoch 3, batch 11300, libri_loss[loss=0.3394, simple_loss=0.4053, pruned_loss=0.1367, over 29740.00 frames. ], tot_loss[loss=0.3661, simple_loss=0.4121, pruned_loss=0.16, over 5668791.43 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3961, pruned_loss=0.1344, over 5734062.69 frames. ], giga_tot_loss[loss=0.3701, simple_loss=0.4142, pruned_loss=0.163, over 5657293.45 frames. ], batch size: 87, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:29:44,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5766, 1.8551, 1.0913, 1.1237], device='cuda:0'), covar=tensor([0.0744, 0.0633, 0.0614, 0.0676], device='cuda:0'), in_proj_covar=tensor([0.1175, 0.0927, 0.0955, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 14:29:59,744 INFO [train.py:968] (0/2) Epoch 3, batch 11350, giga_loss[loss=0.3236, simple_loss=0.3897, pruned_loss=0.1287, over 28967.00 frames. ], tot_loss[loss=0.3694, simple_loss=0.4144, pruned_loss=0.1622, over 5672479.82 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.396, pruned_loss=0.1343, over 5736676.65 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.4164, pruned_loss=0.165, over 5660085.99 frames. ], batch size: 174, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:30:00,069 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101641.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:30:01,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101644.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:30:05,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.549e+02 1.737e+03 2.274e+03 2.764e+03 6.861e+03, threshold=4.549e+03, percent-clipped=3.0
+2023-03-01 14:30:31,314 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101673.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:30:48,578 INFO [train.py:968] (0/2) Epoch 3, batch 11400, giga_loss[loss=0.4338, simple_loss=0.459, pruned_loss=0.2043, over 27634.00 frames. ], tot_loss[loss=0.3709, simple_loss=0.4159, pruned_loss=0.1629, over 5675319.91 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3964, pruned_loss=0.1347, over 5739696.06 frames. ], giga_tot_loss[loss=0.3741, simple_loss=0.4174, pruned_loss=0.1654, over 5661534.08 frames. ], batch size: 472, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:30:54,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3346, 1.3414, 1.1604, 1.7377], device='cuda:0'), covar=tensor([0.1944, 0.1811, 0.1630, 0.1882], device='cuda:0'), in_proj_covar=tensor([0.1023, 0.0828, 0.0915, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 14:31:42,035 INFO [train.py:968] (0/2) Epoch 3, batch 11450, giga_loss[loss=0.3839, simple_loss=0.4062, pruned_loss=0.1807, over 23612.00 frames. ], tot_loss[loss=0.3732, simple_loss=0.4164, pruned_loss=0.1651, over 5660361.41 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3963, pruned_loss=0.1345, over 5741732.81 frames. ], giga_tot_loss[loss=0.3765, simple_loss=0.418, pruned_loss=0.1675, over 5647049.52 frames. ], batch size: 705, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:31:48,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.815e+02 1.708e+03 2.070e+03 3.041e+03 6.663e+03, threshold=4.139e+03, percent-clipped=10.0
+2023-03-01 14:32:26,992 INFO [train.py:968] (0/2) Epoch 3, batch 11500, libri_loss[loss=0.3561, simple_loss=0.4218, pruned_loss=0.1452, over 29553.00 frames. ], tot_loss[loss=0.3714, simple_loss=0.4153, pruned_loss=0.1638, over 5657412.58 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3966, pruned_loss=0.1347, over 5733712.76 frames. ], giga_tot_loss[loss=0.3749, simple_loss=0.4169, pruned_loss=0.1664, over 5652332.29 frames. ], batch size: 83, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:33:16,917 INFO [train.py:968] (0/2) Epoch 3, batch 11550, giga_loss[loss=0.4144, simple_loss=0.441, pruned_loss=0.1939, over 27629.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4143, pruned_loss=0.1619, over 5667111.85 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3967, pruned_loss=0.1347, over 5732422.25 frames. ], giga_tot_loss[loss=0.3722, simple_loss=0.4157, pruned_loss=0.1644, over 5663120.35 frames. ], batch size: 472, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:33:23,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.340e+02 1.426e+03 1.953e+03 2.469e+03 4.297e+03, threshold=3.906e+03, percent-clipped=1.0
+2023-03-01 14:33:49,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:34:03,656 INFO [train.py:968] (0/2) Epoch 3, batch 11600, giga_loss[loss=0.3622, simple_loss=0.4131, pruned_loss=0.1557, over 28406.00 frames. ], tot_loss[loss=0.3686, simple_loss=0.4146, pruned_loss=0.1614, over 5666248.95 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.397, pruned_loss=0.135, over 5734483.13 frames. ], giga_tot_loss[loss=0.3717, simple_loss=0.4159, pruned_loss=0.1638, over 5659094.63 frames. ], batch size: 71, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:34:07,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4616, 1.6763, 1.4538, 0.8518], device='cuda:0'), covar=tensor([0.0738, 0.0520, 0.0397, 0.0714], device='cuda:0'), in_proj_covar=tensor([0.1177, 0.0919, 0.0973, 0.1033], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:0')
+2023-03-01 14:34:27,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5901, 1.8226, 1.7051, 1.6954], device='cuda:0'), covar=tensor([0.1253, 0.1690, 0.1082, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0801, 0.0736, 0.0633], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 14:34:57,218 INFO [train.py:968] (0/2) Epoch 3, batch 11650, giga_loss[loss=0.3434, simple_loss=0.3921, pruned_loss=0.1473, over 28726.00 frames. ], tot_loss[loss=0.3676, simple_loss=0.414, pruned_loss=0.1606, over 5678027.57 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.397, pruned_loss=0.1349, over 5736487.22 frames. ], giga_tot_loss[loss=0.3704, simple_loss=0.4153, pruned_loss=0.1628, over 5670067.36 frames. ], batch size: 85, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:35:03,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.941e+02 1.536e+03 2.075e+03 3.161e+03 9.523e+03, threshold=4.150e+03, percent-clipped=13.0
+2023-03-01 14:35:42,807 INFO [train.py:968] (0/2) Epoch 3, batch 11700, giga_loss[loss=0.436, simple_loss=0.4712, pruned_loss=0.2004, over 29031.00 frames. ], tot_loss[loss=0.3697, simple_loss=0.4155, pruned_loss=0.1619, over 5676720.04 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3963, pruned_loss=0.1343, over 5741838.76 frames. ], giga_tot_loss[loss=0.3741, simple_loss=0.4177, pruned_loss=0.1652, over 5663251.36 frames. ], batch size: 155, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:35:50,844 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-102000.pt
+2023-03-01 14:36:06,419 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:36:08,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:36:26,124 INFO [train.py:968] (0/2) Epoch 3, batch 11750, giga_loss[loss=0.3763, simple_loss=0.4215, pruned_loss=0.1656, over 28711.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.415, pruned_loss=0.1608, over 5674994.15 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3966, pruned_loss=0.1346, over 5728286.49 frames. ], giga_tot_loss[loss=0.3727, simple_loss=0.4172, pruned_loss=0.1641, over 5675006.18 frames. ], batch size: 242, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:36:27,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8043, 1.5740, 1.2372, 1.3714], device='cuda:0'), covar=tensor([0.0572, 0.0650, 0.0826, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0489, 0.0535, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 14:36:32,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.406e+02 1.549e+03 2.044e+03 2.793e+03 6.480e+03, threshold=4.087e+03, percent-clipped=4.0
+2023-03-01 14:36:32,798 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2361, 1.6350, 1.1096, 1.2844], device='cuda:0'), covar=tensor([0.0952, 0.0366, 0.0424, 0.1070], device='cuda:0'), in_proj_covar=tensor([0.0247, 0.0162, 0.0163, 0.0294], device='cuda:0'), out_proj_covar=tensor([0.0040, 0.0029, 0.0026, 0.0044], device='cuda:0')
+2023-03-01 14:36:33,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:36:56,076 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=102077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:37:10,051 INFO [train.py:968] (0/2) Epoch 3, batch 11800, giga_loss[loss=0.367, simple_loss=0.4211, pruned_loss=0.1564, over 28864.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4166, pruned_loss=0.1613, over 5678739.38 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3968, pruned_loss=0.1347, over 5730500.02 frames. ], giga_tot_loss[loss=0.3743, simple_loss=0.4188, pruned_loss=0.1648, over 5674814.51 frames. ], batch size: 199, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:37:10,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0140, 1.2637, 4.1693, 3.3091], device='cuda:0'), covar=tensor([0.1515, 0.1937, 0.0305, 0.0479], device='cuda:0'), in_proj_covar=tensor([0.0537, 0.0511, 0.0693, 0.0560], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 14:37:55,847 INFO [train.py:968] (0/2) Epoch 3, batch 11850, giga_loss[loss=0.3623, simple_loss=0.409, pruned_loss=0.1578, over 28769.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4167, pruned_loss=0.1608, over 5680400.71 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3965, pruned_loss=0.1347, over 5734386.34 frames. ], giga_tot_loss[loss=0.3738, simple_loss=0.4191, pruned_loss=0.1643, over 5672697.42 frames. ], batch size: 99, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:38:05,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.966e+02 1.518e+03 1.908e+03 2.375e+03 5.846e+03, threshold=3.816e+03, percent-clipped=4.0
+2023-03-01 14:38:46,514 INFO [train.py:968] (0/2) Epoch 3, batch 11900, giga_loss[loss=0.3831, simple_loss=0.4247, pruned_loss=0.1708, over 28594.00 frames. ], tot_loss[loss=0.3668, simple_loss=0.4146, pruned_loss=0.1596, over 5676410.80 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3964, pruned_loss=0.1347, over 5737063.30 frames. ], giga_tot_loss[loss=0.3711, simple_loss=0.4169, pruned_loss=0.1627, over 5667195.30 frames. ], batch size: 336, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:38:48,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3976, 1.7026, 1.1400, 1.5561], device='cuda:0'), covar=tensor([0.0890, 0.0335, 0.0434, 0.1024], device='cuda:0'), in_proj_covar=tensor([0.0247, 0.0162, 0.0163, 0.0294], device='cuda:0'), out_proj_covar=tensor([0.0040, 0.0029, 0.0026, 0.0044], device='cuda:0')
+2023-03-01 14:39:04,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2248, 2.0162, 1.1985, 1.2863], device='cuda:0'), covar=tensor([0.0857, 0.0429, 0.0811, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0464, 0.0320, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 14:39:22,448 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4359, 2.5048, 1.4329, 1.3110], device='cuda:0'), covar=tensor([0.0779, 0.0460, 0.0765, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0465, 0.0320, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 14:39:30,013 INFO [train.py:968] (0/2) Epoch 3, batch 11950, giga_loss[loss=0.3262, simple_loss=0.3861, pruned_loss=0.1331, over 28899.00 frames. ], tot_loss[loss=0.3657, simple_loss=0.4138, pruned_loss=0.1588, over 5687578.76 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3967, pruned_loss=0.1347, over 5738968.37 frames. ], giga_tot_loss[loss=0.37, simple_loss=0.4159, pruned_loss=0.162, over 5676583.86 frames. ], batch size: 227, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:39:35,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.498e+02 1.610e+03 2.138e+03 3.048e+03 6.684e+03, threshold=4.277e+03, percent-clipped=10.0
+2023-03-01 14:39:46,803 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-01 14:40:09,714 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-01 14:40:17,295 INFO [train.py:968] (0/2) Epoch 3, batch 12000, giga_loss[loss=0.3859, simple_loss=0.426, pruned_loss=0.173, over 27881.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.4154, pruned_loss=0.1608, over 5661470.97 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.397, pruned_loss=0.1349, over 5732640.30 frames. ], giga_tot_loss[loss=0.3724, simple_loss=0.4173, pruned_loss=0.1638, over 5657167.24 frames. ], batch size: 412, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:40:17,300 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 14:40:25,813 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2578, simple_loss=0.3571, pruned_loss=0.07925, over 944034.00 frames. 
+2023-03-01 14:40:25,814 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 14:40:44,671 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=102313.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:41:10,803 INFO [train.py:968] (0/2) Epoch 3, batch 12050, giga_loss[loss=0.3222, simple_loss=0.3803, pruned_loss=0.1321, over 28237.00 frames. ], tot_loss[loss=0.3676, simple_loss=0.4152, pruned_loss=0.16, over 5669730.91 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3966, pruned_loss=0.1347, over 5729729.57 frames. ], giga_tot_loss[loss=0.3722, simple_loss=0.4177, pruned_loss=0.1634, over 5666359.55 frames. ], batch size: 77, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:41:18,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.795e+02 1.398e+03 1.972e+03 2.798e+03 1.097e+04, threshold=3.943e+03, percent-clipped=13.0
+2023-03-01 14:41:48,285 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=102380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:42:00,586 INFO [train.py:968] (0/2) Epoch 3, batch 12100, giga_loss[loss=0.3651, simple_loss=0.4077, pruned_loss=0.1612, over 28801.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4144, pruned_loss=0.1609, over 5667408.68 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3963, pruned_loss=0.1346, over 5733071.13 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4169, pruned_loss=0.1641, over 5660821.11 frames. ], batch size: 243, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:42:48,257 INFO [train.py:968] (0/2) Epoch 3, batch 12150, giga_loss[loss=0.4611, simple_loss=0.4603, pruned_loss=0.231, over 26474.00 frames. ], tot_loss[loss=0.3687, simple_loss=0.4145, pruned_loss=0.1614, over 5665834.32 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3969, pruned_loss=0.1348, over 5736527.83 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4164, pruned_loss=0.1643, over 5656260.57 frames. ], batch size: 555, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:42:55,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.651e+02 1.486e+03 2.003e+03 2.940e+03 6.893e+03, threshold=4.006e+03, percent-clipped=11.0
+2023-03-01 14:42:57,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:43:38,052 INFO [train.py:968] (0/2) Epoch 3, batch 12200, giga_loss[loss=0.4255, simple_loss=0.4285, pruned_loss=0.2113, over 23504.00 frames. ], tot_loss[loss=0.3698, simple_loss=0.415, pruned_loss=0.1623, over 5660293.73 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.397, pruned_loss=0.1348, over 5738561.66 frames. ], giga_tot_loss[loss=0.3731, simple_loss=0.4165, pruned_loss=0.1649, over 5650470.79 frames. ], batch size: 705, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:43:55,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-01 14:44:10,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2836, 1.6758, 1.3946, 1.4974], device='cuda:0'), covar=tensor([0.1303, 0.1806, 0.1165, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0809, 0.0740, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 14:44:25,331 INFO [train.py:968] (0/2) Epoch 3, batch 12250, giga_loss[loss=0.3279, simple_loss=0.3826, pruned_loss=0.1366, over 28336.00 frames. ], tot_loss[loss=0.3701, simple_loss=0.4157, pruned_loss=0.1623, over 5666541.44 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3972, pruned_loss=0.1347, over 5741689.72 frames. ], giga_tot_loss[loss=0.3735, simple_loss=0.4171, pruned_loss=0.1649, over 5654670.35 frames. ], batch size: 65, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:44:28,322 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1449, 1.2978, 1.0950, 1.5140], device='cuda:0'), covar=tensor([0.2082, 0.1940, 0.1956, 0.1996], device='cuda:0'), in_proj_covar=tensor([0.1021, 0.0820, 0.0921, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 14:44:31,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3485, 2.1795, 1.6590, 0.6286], device='cuda:0'), covar=tensor([0.1706, 0.0842, 0.1456, 0.1992], device='cuda:0'), in_proj_covar=tensor([0.1243, 0.1194, 0.1252, 0.1065], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 14:44:34,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.622e+02 1.475e+03 1.966e+03 2.558e+03 7.004e+03, threshold=3.931e+03, percent-clipped=5.0
+2023-03-01 14:45:11,470 INFO [train.py:968] (0/2) Epoch 3, batch 12300, giga_loss[loss=0.3605, simple_loss=0.4145, pruned_loss=0.1533, over 28759.00 frames. ], tot_loss[loss=0.37, simple_loss=0.4151, pruned_loss=0.1624, over 5641206.64 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3974, pruned_loss=0.135, over 5736148.76 frames. ], giga_tot_loss[loss=0.3734, simple_loss=0.4165, pruned_loss=0.1651, over 5634013.22 frames. ], batch size: 92, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:45:15,491 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:45:19,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:45:44,980 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:45:45,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1936, 1.3227, 0.9631, 0.7007], device='cuda:0'), covar=tensor([0.0492, 0.0432, 0.0381, 0.0539], device='cuda:0'), in_proj_covar=tensor([0.1157, 0.0894, 0.0935, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 14:45:57,879 INFO [train.py:968] (0/2) Epoch 3, batch 12350, giga_loss[loss=0.4346, simple_loss=0.4501, pruned_loss=0.2096, over 26521.00 frames. ], tot_loss[loss=0.3707, simple_loss=0.4161, pruned_loss=0.1626, over 5650478.82 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3981, pruned_loss=0.1353, over 5740232.25 frames. ], giga_tot_loss[loss=0.3741, simple_loss=0.4173, pruned_loss=0.1655, over 5638073.43 frames. ], batch size: 555, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:46:03,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.639e+02 1.560e+03 1.999e+03 2.448e+03 4.536e+03, threshold=3.999e+03, percent-clipped=6.0
+2023-03-01 14:46:37,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102688.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:46:40,351 INFO [train.py:968] (0/2) Epoch 3, batch 12400, giga_loss[loss=0.3639, simple_loss=0.4133, pruned_loss=0.1572, over 28644.00 frames. ], tot_loss[loss=0.3686, simple_loss=0.4152, pruned_loss=0.161, over 5659326.41 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3972, pruned_loss=0.1346, over 5746102.27 frames. ], giga_tot_loss[loss=0.3735, simple_loss=0.4175, pruned_loss=0.1648, over 5641659.34 frames. ], batch size: 262, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:47:28,552 INFO [train.py:968] (0/2) Epoch 3, batch 12450, giga_loss[loss=0.3503, simple_loss=0.4102, pruned_loss=0.1452, over 28651.00 frames. ], tot_loss[loss=0.368, simple_loss=0.4146, pruned_loss=0.1607, over 5661580.88 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.397, pruned_loss=0.1344, over 5748986.53 frames. ], giga_tot_loss[loss=0.3727, simple_loss=0.4169, pruned_loss=0.1642, over 5643498.52 frames. ], batch size: 242, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:47:40,990 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.749e+02 1.641e+03 2.170e+03 2.985e+03 5.435e+03, threshold=4.341e+03, percent-clipped=7.0
+2023-03-01 14:47:45,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:47:49,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8149, 1.5745, 1.3242, 1.4772], device='cuda:0'), covar=tensor([0.0621, 0.0661, 0.0953, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0481, 0.0522, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 14:47:55,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7787, 1.2484, 3.7523, 3.0835], device='cuda:0'), covar=tensor([0.1590, 0.1916, 0.0329, 0.0471], device='cuda:0'), in_proj_covar=tensor([0.0536, 0.0509, 0.0689, 0.0550], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 14:48:15,535 INFO [train.py:968] (0/2) Epoch 3, batch 12500, giga_loss[loss=0.3881, simple_loss=0.416, pruned_loss=0.1801, over 28736.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.413, pruned_loss=0.1599, over 5669199.41 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3972, pruned_loss=0.1345, over 5752348.38 frames. ], giga_tot_loss[loss=0.3707, simple_loss=0.415, pruned_loss=0.1632, over 5650046.26 frames. ], batch size: 92, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:48:53,087 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102831.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:48:55,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102834.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:49:00,842 INFO [train.py:968] (0/2) Epoch 3, batch 12550, giga_loss[loss=0.3201, simple_loss=0.3735, pruned_loss=0.1333, over 29027.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.41, pruned_loss=0.1576, over 5680381.41 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3971, pruned_loss=0.1343, over 5753785.67 frames. ], giga_tot_loss[loss=0.367, simple_loss=0.412, pruned_loss=0.161, over 5662015.81 frames. ], batch size: 128, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:49:09,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.495e+02 1.496e+03 1.943e+03 2.635e+03 5.919e+03, threshold=3.886e+03, percent-clipped=5.0
+2023-03-01 14:49:21,833 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102863.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:49:47,317 INFO [train.py:968] (0/2) Epoch 3, batch 12600, giga_loss[loss=0.3465, simple_loss=0.3945, pruned_loss=0.1492, over 28802.00 frames. ], tot_loss[loss=0.3585, simple_loss=0.4061, pruned_loss=0.1554, over 5671331.60 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3972, pruned_loss=0.1344, over 5755390.76 frames. ], giga_tot_loss[loss=0.3626, simple_loss=0.4079, pruned_loss=0.1586, over 5653057.65 frames. ], batch size: 186, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:49:55,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:49:57,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:50:17,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0828, 4.1896, 2.0921, 1.8128], device='cuda:0'), covar=tensor([0.0745, 0.0371, 0.0703, 0.1210], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0469, 0.0321, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0020, 0.0013, 0.0018], device='cuda:0')
+2023-03-01 14:50:24,575 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:50:33,533 INFO [train.py:968] (0/2) Epoch 3, batch 12650, giga_loss[loss=0.4464, simple_loss=0.4549, pruned_loss=0.2189, over 27902.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4048, pruned_loss=0.1555, over 5666116.36 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3972, pruned_loss=0.1342, over 5756317.97 frames. ], giga_tot_loss[loss=0.3617, simple_loss=0.4063, pruned_loss=0.1586, over 5649218.02 frames. ], batch size: 412, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:50:46,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.290e+02 1.592e+03 2.079e+03 3.117e+03 1.174e+04, threshold=4.158e+03, percent-clipped=16.0
+2023-03-01 14:51:22,726 INFO [train.py:968] (0/2) Epoch 3, batch 12700, giga_loss[loss=0.3471, simple_loss=0.4016, pruned_loss=0.1463, over 28921.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.4044, pruned_loss=0.1562, over 5655610.74 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3967, pruned_loss=0.134, over 5757758.10 frames. ], giga_tot_loss[loss=0.3624, simple_loss=0.4062, pruned_loss=0.1593, over 5639194.18 frames. ], batch size: 164, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:51:27,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8056, 1.5823, 1.5653, 1.5669], device='cuda:0'), covar=tensor([0.0808, 0.1564, 0.1199, 0.1234], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0777, 0.0641, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 14:52:11,044 INFO [train.py:968] (0/2) Epoch 3, batch 12750, giga_loss[loss=0.3979, simple_loss=0.4131, pruned_loss=0.1914, over 24198.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4044, pruned_loss=0.156, over 5657743.14 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.397, pruned_loss=0.1342, over 5761115.54 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.4057, pruned_loss=0.1587, over 5639422.31 frames. ], batch size: 705, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:52:20,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.594e+03 2.225e+03 3.309e+03 9.497e+03, threshold=4.450e+03, percent-clipped=14.0
+2023-03-01 14:52:57,367 INFO [train.py:968] (0/2) Epoch 3, batch 12800, giga_loss[loss=0.2922, simple_loss=0.3683, pruned_loss=0.108, over 28553.00 frames. ], tot_loss[loss=0.353, simple_loss=0.4022, pruned_loss=0.1519, over 5665040.15 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.397, pruned_loss=0.1343, over 5764875.85 frames. ], giga_tot_loss[loss=0.3562, simple_loss=0.4035, pruned_loss=0.1545, over 5644365.57 frames. ], batch size: 336, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:53:48,552 INFO [train.py:968] (0/2) Epoch 3, batch 12850, giga_loss[loss=0.2904, simple_loss=0.3595, pruned_loss=0.1106, over 28895.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.398, pruned_loss=0.1469, over 5657921.75 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3969, pruned_loss=0.1343, over 5764257.03 frames. ], giga_tot_loss[loss=0.3489, simple_loss=0.3991, pruned_loss=0.1493, over 5640251.83 frames. ], batch size: 213, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:53:57,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.962e+02 1.371e+03 1.921e+03 3.064e+03 4.890e+03, threshold=3.842e+03, percent-clipped=4.0
+2023-03-01 14:54:39,012 INFO [train.py:968] (0/2) Epoch 3, batch 12900, giga_loss[loss=0.2856, simple_loss=0.3575, pruned_loss=0.1068, over 28917.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3926, pruned_loss=0.1417, over 5659076.22 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3959, pruned_loss=0.1338, over 5767246.65 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3943, pruned_loss=0.1442, over 5640612.46 frames. ], batch size: 227, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:54:44,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6570, 1.5841, 1.2065, 1.3130], device='cuda:0'), covar=tensor([0.0643, 0.0638, 0.0861, 0.1202], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0481, 0.0530, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 14:55:19,085 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.26 vs. limit=5.0
+2023-03-01 14:55:32,581 INFO [train.py:968] (0/2) Epoch 3, batch 12950, libri_loss[loss=0.3528, simple_loss=0.4135, pruned_loss=0.1461, over 29119.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3873, pruned_loss=0.1369, over 5652136.19 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3957, pruned_loss=0.1336, over 5769065.56 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3888, pruned_loss=0.139, over 5634399.11 frames. ], batch size: 101, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:55:35,504 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:55:41,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.104e+02 1.344e+03 1.738e+03 2.427e+03 4.220e+03, threshold=3.477e+03, percent-clipped=6.0
+2023-03-01 14:56:10,343 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:56:19,239 INFO [train.py:968] (0/2) Epoch 3, batch 13000, libri_loss[loss=0.2899, simple_loss=0.3437, pruned_loss=0.118, over 29496.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3862, pruned_loss=0.1333, over 5665152.65 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.395, pruned_loss=0.1335, over 5770490.09 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3878, pruned_loss=0.1352, over 5647007.59 frames. ], batch size: 70, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:57:10,453 INFO [train.py:968] (0/2) Epoch 3, batch 13050, giga_loss[loss=0.305, simple_loss=0.3803, pruned_loss=0.1149, over 28939.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3858, pruned_loss=0.1326, over 5652896.45 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.3942, pruned_loss=0.1331, over 5762454.05 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3876, pruned_loss=0.1345, over 5642880.24 frames. ], batch size: 145, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:57:21,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.592e+02 1.316e+03 1.760e+03 2.378e+03 4.257e+03, threshold=3.521e+03, percent-clipped=3.0
+2023-03-01 14:58:02,127 INFO [train.py:968] (0/2) Epoch 3, batch 13100, libri_loss[loss=0.3045, simple_loss=0.3741, pruned_loss=0.1175, over 29492.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3836, pruned_loss=0.1307, over 5655618.95 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3936, pruned_loss=0.1326, over 5765519.73 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3854, pruned_loss=0.1326, over 5642076.85 frames. ], batch size: 85, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:58:07,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4564, 1.4474, 1.2744, 1.7178], device='cuda:0'), covar=tensor([0.2192, 0.1953, 0.1801, 0.2196], device='cuda:0'), in_proj_covar=tensor([0.1018, 0.0808, 0.0912, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 14:58:22,963 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5434, 1.7014, 1.5635, 1.8037], device='cuda:0'), covar=tensor([0.1890, 0.1605, 0.1486, 0.1544], device='cuda:0'), in_proj_covar=tensor([0.1019, 0.0811, 0.0914, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 14:58:51,128 INFO [train.py:968] (0/2) Epoch 3, batch 13150, giga_loss[loss=0.2616, simple_loss=0.3408, pruned_loss=0.09121, over 28351.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3787, pruned_loss=0.1273, over 5647788.27 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3921, pruned_loss=0.132, over 5759137.62 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3812, pruned_loss=0.1293, over 5638905.43 frames. ], batch size: 71, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:58:59,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.901e+02 1.397e+03 1.926e+03 2.492e+03 5.719e+03, threshold=3.853e+03, percent-clipped=5.0
+2023-03-01 14:59:04,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5519, 2.0957, 1.7803, 1.7833], device='cuda:0'), covar=tensor([0.1550, 0.1551, 0.1203, 0.0912], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0783, 0.0740, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 14:59:39,434 INFO [train.py:968] (0/2) Epoch 3, batch 13200, giga_loss[loss=0.3586, simple_loss=0.4127, pruned_loss=0.1522, over 28721.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3779, pruned_loss=0.1272, over 5648368.75 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3916, pruned_loss=0.1317, over 5761627.23 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3801, pruned_loss=0.1289, over 5637344.20 frames. ], batch size: 284, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 14:59:40,616 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:00:15,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:00:26,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0940, 4.6935, 4.7968, 1.9749], device='cuda:0'), covar=tensor([0.0312, 0.0270, 0.0650, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.0759, 0.0597, 0.0796, 0.0560], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 15:00:30,120 INFO [train.py:968] (0/2) Epoch 3, batch 13250, giga_loss[loss=0.3132, simple_loss=0.3749, pruned_loss=0.1257, over 28605.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3795, pruned_loss=0.1283, over 5644472.62 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3918, pruned_loss=0.1319, over 5758486.74 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3807, pruned_loss=0.1295, over 5635549.25 frames. ], batch size: 307, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:00:42,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.657e+02 1.452e+03 1.965e+03 2.799e+03 1.412e+04, threshold=3.930e+03, percent-clipped=8.0
+2023-03-01 15:01:04,002 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-01 15:01:21,511 INFO [train.py:968] (0/2) Epoch 3, batch 13300, giga_loss[loss=0.2754, simple_loss=0.352, pruned_loss=0.09937, over 29036.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3765, pruned_loss=0.1254, over 5654511.07 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3914, pruned_loss=0.1317, over 5759195.31 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3777, pruned_loss=0.1264, over 5646429.18 frames. ], batch size: 155, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:01:53,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:02:06,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4201, 1.9848, 1.6692, 1.7042], device='cuda:0'), covar=tensor([0.1665, 0.1877, 0.1335, 0.1041], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0775, 0.0730, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 15:02:11,931 INFO [train.py:968] (0/2) Epoch 3, batch 13350, giga_loss[loss=0.2862, simple_loss=0.3545, pruned_loss=0.109, over 28935.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3725, pruned_loss=0.1223, over 5652970.90 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3905, pruned_loss=0.1313, over 5762706.52 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3737, pruned_loss=0.1231, over 5639241.95 frames. ], batch size: 227, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:02:23,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.006e+02 1.280e+03 1.504e+03 2.208e+03 5.260e+03, threshold=3.009e+03, percent-clipped=1.0
+2023-03-01 15:02:27,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:03:01,537 INFO [train.py:968] (0/2) Epoch 3, batch 13400, giga_loss[loss=0.3173, simple_loss=0.3685, pruned_loss=0.1331, over 28273.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3679, pruned_loss=0.1193, over 5654253.15 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3897, pruned_loss=0.131, over 5766179.15 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3689, pruned_loss=0.1199, over 5636365.14 frames. ], batch size: 368, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:03:45,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1074, 1.3072, 0.9885, 0.2522], device='cuda:0'), covar=tensor([0.0909, 0.0984, 0.1568, 0.1692], device='cuda:0'), in_proj_covar=tensor([0.1243, 0.1203, 0.1267, 0.1067], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 15:03:55,904 INFO [train.py:968] (0/2) Epoch 3, batch 13450, giga_loss[loss=0.2445, simple_loss=0.3212, pruned_loss=0.08386, over 28880.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3652, pruned_loss=0.1178, over 5658829.68 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3893, pruned_loss=0.1309, over 5759808.70 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.366, pruned_loss=0.1181, over 5647470.00 frames. ], batch size: 145, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:04:06,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.613e+02 1.471e+03 1.938e+03 2.446e+03 4.905e+03, threshold=3.876e+03, percent-clipped=16.0
+2023-03-01 15:04:06,600 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-01 15:04:12,501 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103759.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 15:04:15,787 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:04:18,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:04:47,424 INFO [train.py:968] (0/2) Epoch 3, batch 13500, giga_loss[loss=0.2803, simple_loss=0.3527, pruned_loss=0.104, over 29016.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3649, pruned_loss=0.1185, over 5647833.03 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3891, pruned_loss=0.1307, over 5762043.14 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3653, pruned_loss=0.1188, over 5635839.94 frames. ], batch size: 155, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:04:50,807 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:04:55,969 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:04:58,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:05:34,664 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:05:36,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0777, 1.1555, 1.2143, 1.1244], device='cuda:0'), covar=tensor([0.0781, 0.0939, 0.1257, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0757, 0.0618, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 15:05:48,182 INFO [train.py:968] (0/2) Epoch 3, batch 13550, giga_loss[loss=0.3123, simple_loss=0.3876, pruned_loss=0.1185, over 28869.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.368, pruned_loss=0.1205, over 5647859.13 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3891, pruned_loss=0.1306, over 5762425.93 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3683, pruned_loss=0.1208, over 5637627.09 frames. ], batch size: 174, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:06:00,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.054e+02 1.468e+03 1.895e+03 3.026e+03 8.205e+03, threshold=3.789e+03, percent-clipped=15.0
+2023-03-01 15:06:14,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:06:41,773 INFO [train.py:968] (0/2) Epoch 3, batch 13600, libri_loss[loss=0.2871, simple_loss=0.349, pruned_loss=0.1126, over 29582.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3705, pruned_loss=0.1207, over 5646903.10 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3886, pruned_loss=0.1305, over 5758173.75 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3705, pruned_loss=0.1207, over 5638550.67 frames. ], batch size: 74, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:06:54,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:07:41,717 INFO [train.py:968] (0/2) Epoch 3, batch 13650, giga_loss[loss=0.3067, simple_loss=0.3717, pruned_loss=0.1208, over 28983.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3706, pruned_loss=0.1202, over 5659633.93 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3883, pruned_loss=0.1304, over 5759726.21 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3706, pruned_loss=0.1201, over 5649624.96 frames. ], batch size: 199, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:07:54,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.329e+02 1.421e+03 2.085e+03 2.706e+03 6.204e+03, threshold=4.170e+03, percent-clipped=13.0
+2023-03-01 15:08:39,654 INFO [train.py:968] (0/2) Epoch 3, batch 13700, giga_loss[loss=0.278, simple_loss=0.3506, pruned_loss=0.1028, over 28623.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3703, pruned_loss=0.1203, over 5662891.56 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3881, pruned_loss=0.1304, over 5759387.34 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.37, pruned_loss=0.12, over 5652265.11 frames. ], batch size: 307, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:08:49,627 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-104000.pt
+2023-03-01 15:08:59,750 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:09:02,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:09:40,145 INFO [train.py:968] (0/2) Epoch 3, batch 13750, giga_loss[loss=0.2869, simple_loss=0.3637, pruned_loss=0.1051, over 28663.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 5665934.87 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3879, pruned_loss=0.1303, over 5762002.16 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3678, pruned_loss=0.1178, over 5654093.39 frames. ], batch size: 262, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:09:40,344 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:09:42,628 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:09:46,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:09:51,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.128e+02 1.302e+03 1.718e+03 2.417e+03 4.639e+03, threshold=3.436e+03, percent-clipped=2.0
+2023-03-01 15:10:18,790 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:10:36,210 INFO [train.py:968] (0/2) Epoch 3, batch 13800, giga_loss[loss=0.2911, simple_loss=0.3633, pruned_loss=0.1094, over 28505.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3666, pruned_loss=0.1156, over 5673330.13 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3867, pruned_loss=0.1295, over 5763473.50 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3667, pruned_loss=0.1156, over 5659854.33 frames. ], batch size: 336, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:11:04,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4351, 1.9053, 1.5079, 1.5501], device='cuda:0'), covar=tensor([0.0854, 0.0294, 0.0374, 0.0998], device='cuda:0'), in_proj_covar=tensor([0.0252, 0.0160, 0.0165, 0.0298], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:0')
+2023-03-01 15:11:28,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:11:29,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104134.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 15:11:37,986 INFO [train.py:968] (0/2) Epoch 3, batch 13850, giga_loss[loss=0.3177, simple_loss=0.3692, pruned_loss=0.1331, over 28895.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.364, pruned_loss=0.1151, over 5664669.06 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3861, pruned_loss=0.1291, over 5759846.52 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3639, pruned_loss=0.115, over 5653793.47 frames. ], batch size: 227, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:11:52,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.590e+02 1.223e+03 1.711e+03 2.341e+03 4.502e+03, threshold=3.422e+03, percent-clipped=4.0
+2023-03-01 15:11:58,688 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:12:36,762 INFO [train.py:968] (0/2) Epoch 3, batch 13900, giga_loss[loss=0.2833, simple_loss=0.3576, pruned_loss=0.1046, over 28589.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3629, pruned_loss=0.115, over 5673440.62 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3861, pruned_loss=0.1293, over 5761842.14 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3624, pruned_loss=0.1145, over 5661445.41 frames. ], batch size: 307, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:13:14,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7082, 1.7981, 1.5573, 1.6634], device='cuda:0'), covar=tensor([0.2014, 0.2537, 0.1663, 0.1275], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0762, 0.0720, 0.0629], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 15:13:34,765 INFO [train.py:968] (0/2) Epoch 3, batch 13950, giga_loss[loss=0.336, simple_loss=0.3911, pruned_loss=0.1404, over 28898.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3607, pruned_loss=0.1142, over 5663750.48 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3855, pruned_loss=0.129, over 5760922.31 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3603, pruned_loss=0.1137, over 5652962.55 frames. ], batch size: 145, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:13:47,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.695e+02 1.486e+03 1.993e+03 2.647e+03 9.288e+03, threshold=3.986e+03, percent-clipped=12.0
+2023-03-01 15:14:05,356 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:14:14,508 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104277.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 15:14:17,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104280.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 15:14:22,633 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 15:14:29,418 INFO [train.py:968] (0/2) Epoch 3, batch 14000, giga_loss[loss=0.3121, simple_loss=0.3817, pruned_loss=0.1212, over 28072.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3632, pruned_loss=0.1151, over 5667894.61 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3854, pruned_loss=0.1292, over 5766559.74 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3621, pruned_loss=0.114, over 5650656.83 frames. ], batch size: 412, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:14:51,855 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104309.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 15:15:26,385 INFO [train.py:968] (0/2) Epoch 3, batch 14050, giga_loss[loss=0.2709, simple_loss=0.344, pruned_loss=0.09886, over 28340.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3641, pruned_loss=0.115, over 5659559.03 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3842, pruned_loss=0.1287, over 5757535.65 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3636, pruned_loss=0.1141, over 5650427.89 frames. ], batch size: 368, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:15:46,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.981e+02 1.356e+03 1.730e+03 2.336e+03 4.675e+03, threshold=3.459e+03, percent-clipped=3.0
+2023-03-01 15:16:08,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2855, 1.5277, 1.1811, 1.4642], device='cuda:0'), covar=tensor([0.0933, 0.0370, 0.0423, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0250, 0.0159, 0.0167, 0.0298], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:0')
+2023-03-01 15:16:23,510 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:16:32,844 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:16:33,202 INFO [train.py:968] (0/2) Epoch 3, batch 14100, giga_loss[loss=0.3073, simple_loss=0.3745, pruned_loss=0.1201, over 28722.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3593, pruned_loss=0.1113, over 5672045.88 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.384, pruned_loss=0.1284, over 5760003.91 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3585, pruned_loss=0.1104, over 5660564.49 frames. ], batch size: 262, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:17:40,524 INFO [train.py:968] (0/2) Epoch 3, batch 14150, giga_loss[loss=0.3288, simple_loss=0.3923, pruned_loss=0.1327, over 28929.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3615, pruned_loss=0.1132, over 5679927.01 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3833, pruned_loss=0.128, over 5762065.49 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3611, pruned_loss=0.1126, over 5668025.46 frames. ], batch size: 186, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:17:41,607 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:17:56,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.490e+02 1.283e+03 1.647e+03 2.327e+03 3.834e+03, threshold=3.294e+03, percent-clipped=3.0
+2023-03-01 15:18:45,836 INFO [train.py:968] (0/2) Epoch 3, batch 14200, giga_loss[loss=0.3009, simple_loss=0.3529, pruned_loss=0.1244, over 24485.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3659, pruned_loss=0.1146, over 5681622.43 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3828, pruned_loss=0.1278, over 5766654.97 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3654, pruned_loss=0.1139, over 5665715.78 frames. ], batch size: 705, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:19:06,446 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:19:37,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:19:46,328 INFO [train.py:968] (0/2) Epoch 3, batch 14250, giga_loss[loss=0.2909, simple_loss=0.369, pruned_loss=0.1064, over 28320.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3686, pruned_loss=0.114, over 5669832.01 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3829, pruned_loss=0.1278, over 5755315.36 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3678, pruned_loss=0.1131, over 5665113.53 frames. ], batch size: 368, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:20:00,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.171e+02 1.567e+03 2.119e+03 2.973e+03 6.483e+03, threshold=4.238e+03, percent-clipped=20.0
+2023-03-01 15:20:10,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8328, 3.5318, 3.5017, 1.5368], device='cuda:0'), covar=tensor([0.0540, 0.0466, 0.0970, 0.2106], device='cuda:0'), in_proj_covar=tensor([0.0739, 0.0582, 0.0758, 0.0551], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 15:20:23,907 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-01 15:20:42,700 INFO [train.py:968] (0/2) Epoch 3, batch 14300, giga_loss[loss=0.2706, simple_loss=0.3598, pruned_loss=0.09067, over 28696.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3689, pruned_loss=0.113, over 5663748.15 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.382, pruned_loss=0.1273, over 5749135.09 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3687, pruned_loss=0.1124, over 5662133.93 frames. ], batch size: 262, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:21:37,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9668, 1.3805, 3.8595, 3.0679], device='cuda:0'), covar=tensor([0.1520, 0.1840, 0.0329, 0.0486], device='cuda:0'), in_proj_covar=tensor([0.0521, 0.0503, 0.0653, 0.0522], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 15:21:39,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6185, 3.0185, 1.5608, 1.4982], device='cuda:0'), covar=tensor([0.0763, 0.0367, 0.0834, 0.1250], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0448, 0.0322, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0018], device='cuda:0')
+2023-03-01 15:21:40,137 INFO [train.py:968] (0/2) Epoch 3, batch 14350, libri_loss[loss=0.3354, simple_loss=0.3989, pruned_loss=0.136, over 29528.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3678, pruned_loss=0.112, over 5663342.68 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3816, pruned_loss=0.1272, over 5740360.82 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3676, pruned_loss=0.1111, over 5667063.57 frames. ], batch size: 84, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:21:44,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:21:53,529 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:21:54,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4332, 2.0867, 1.3889, 0.5358], device='cuda:0'), covar=tensor([0.1491, 0.0854, 0.1715, 0.1873], device='cuda:0'), in_proj_covar=tensor([0.1238, 0.1187, 0.1278, 0.1058], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 15:21:56,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:21:56,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.986e+02 1.364e+03 1.734e+03 2.505e+03 1.136e+04, threshold=3.468e+03, percent-clipped=6.0
+2023-03-01 15:22:10,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6381, 1.5199, 1.5442, 1.5144], device='cuda:0'), covar=tensor([0.0769, 0.1877, 0.1316, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0762, 0.0612, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 15:22:10,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-01 15:22:22,217 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-01 15:22:23,049 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:22:27,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:22:32,088 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:22:39,729 INFO [train.py:968] (0/2) Epoch 3, batch 14400, giga_loss[loss=0.2763, simple_loss=0.3501, pruned_loss=0.1012, over 28763.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.368, pruned_loss=0.1127, over 5664588.16 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3815, pruned_loss=0.1272, over 5741781.89 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3676, pruned_loss=0.1115, over 5664030.62 frames. ], batch size: 119, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:22:59,769 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:23:33,310 INFO [train.py:968] (0/2) Epoch 3, batch 14450, libri_loss[loss=0.2954, simple_loss=0.364, pruned_loss=0.1134, over 29564.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3672, pruned_loss=0.1131, over 5672518.74 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3812, pruned_loss=0.127, over 5738490.78 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3662, pruned_loss=0.1116, over 5671367.17 frames. ], batch size: 77, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:23:54,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.884e+02 1.325e+03 1.721e+03 2.408e+03 5.667e+03, threshold=3.443e+03, percent-clipped=5.0
+2023-03-01 15:23:58,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:24:08,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:24:44,232 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:24:48,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:24:48,467 INFO [train.py:968] (0/2) Epoch 3, batch 14500, libri_loss[loss=0.3653, simple_loss=0.4142, pruned_loss=0.1582, over 29547.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3692, pruned_loss=0.1155, over 5682246.06 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3809, pruned_loss=0.1269, over 5742477.64 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3684, pruned_loss=0.1141, over 5676405.30 frames. ], batch size: 89, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:25:29,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:25:36,319 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:25:53,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3908, 1.3265, 1.2323, 1.5268], device='cuda:0'), covar=tensor([0.2098, 0.1943, 0.1762, 0.2119], device='cuda:0'), in_proj_covar=tensor([0.1012, 0.0807, 0.0915, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 15:26:05,893 INFO [train.py:968] (0/2) Epoch 3, batch 14550, giga_loss[loss=0.3303, simple_loss=0.3723, pruned_loss=0.1442, over 26841.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3652, pruned_loss=0.1138, over 5671195.32 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3805, pruned_loss=0.1268, over 5736041.39 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3646, pruned_loss=0.1125, over 5670954.45 frames. ], batch size: 555, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:26:27,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.897e+02 1.289e+03 1.801e+03 2.529e+03 6.041e+03, threshold=3.603e+03, percent-clipped=8.0
+2023-03-01 15:27:09,585 INFO [train.py:968] (0/2) Epoch 3, batch 14600, giga_loss[loss=0.233, simple_loss=0.3147, pruned_loss=0.07561, over 28462.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3633, pruned_loss=0.1124, over 5675955.90 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3803, pruned_loss=0.1266, over 5740050.30 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3625, pruned_loss=0.1109, over 5669890.81 frames. ], batch size: 71, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:27:19,906 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:27:23,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:27:29,840 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:27:32,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:28:02,819 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:28:10,068 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:28:10,439 INFO [train.py:968] (0/2) Epoch 3, batch 14650, giga_loss[loss=0.2932, simple_loss=0.3573, pruned_loss=0.1146, over 28718.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3604, pruned_loss=0.1113, over 5681262.64 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3798, pruned_loss=0.1265, over 5743247.62 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3598, pruned_loss=0.1099, over 5672200.28 frames. ], batch size: 262, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:28:31,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.371e+02 1.354e+03 1.857e+03 2.249e+03 8.083e+03, threshold=3.714e+03, percent-clipped=12.0
+2023-03-01 15:28:38,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:28:41,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:29:09,462 INFO [train.py:968] (0/2) Epoch 3, batch 14700, giga_loss[loss=0.3144, simple_loss=0.3777, pruned_loss=0.1256, over 27682.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3645, pruned_loss=0.1141, over 5669750.44 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3799, pruned_loss=0.1266, over 5734344.07 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3635, pruned_loss=0.1126, over 5668581.36 frames. ], batch size: 472, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:29:10,883 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:30:05,655 INFO [train.py:968] (0/2) Epoch 3, batch 14750, giga_loss[loss=0.284, simple_loss=0.3543, pruned_loss=0.1068, over 29077.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3673, pruned_loss=0.116, over 5675509.24 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3789, pruned_loss=0.1263, over 5731836.51 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3668, pruned_loss=0.1145, over 5673606.29 frames. ], batch size: 155, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:30:19,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-01 15:30:26,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.112e+02 1.502e+03 1.898e+03 2.758e+03 9.466e+03, threshold=3.796e+03, percent-clipped=7.0
+2023-03-01 15:30:37,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3926, 1.3959, 1.4177, 1.3107], device='cuda:0'), covar=tensor([0.0843, 0.1292, 0.1460, 0.1195], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0757, 0.0613, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 15:31:06,009 INFO [train.py:968] (0/2) Epoch 3, batch 14800, giga_loss[loss=0.3343, simple_loss=0.3847, pruned_loss=0.142, over 27702.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3661, pruned_loss=0.1164, over 5674261.34 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3788, pruned_loss=0.1261, over 5731665.05 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3656, pruned_loss=0.1152, over 5671893.18 frames. ], batch size: 472, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:32:07,098 INFO [train.py:968] (0/2) Epoch 3, batch 14850, libri_loss[loss=0.3025, simple_loss=0.3618, pruned_loss=0.1216, over 29514.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3659, pruned_loss=0.1167, over 5677564.24 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3784, pruned_loss=0.1259, over 5732633.29 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3656, pruned_loss=0.1157, over 5673359.38 frames. ], batch size: 80, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:32:15,810 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0714, 1.8330, 1.8438, 1.6986], device='cuda:0'), covar=tensor([0.0761, 0.1747, 0.1252, 0.1378], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0756, 0.0609, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 15:32:26,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.158e+02 1.219e+03 1.642e+03 2.373e+03 1.259e+04, threshold=3.285e+03, percent-clipped=9.0
+2023-03-01 15:33:09,465 INFO [train.py:968] (0/2) Epoch 3, batch 14900, giga_loss[loss=0.359, simple_loss=0.401, pruned_loss=0.1585, over 27658.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3666, pruned_loss=0.1167, over 5675846.07 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3778, pruned_loss=0.1256, over 5727074.85 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3665, pruned_loss=0.1161, over 5675646.11 frames. ], batch size: 472, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:34:14,126 INFO [train.py:968] (0/2) Epoch 3, batch 14950, libri_loss[loss=0.3259, simple_loss=0.3853, pruned_loss=0.1332, over 19576.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3679, pruned_loss=0.1163, over 5666899.06 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3777, pruned_loss=0.1255, over 5718569.13 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3676, pruned_loss=0.1155, over 5673391.37 frames. ], batch size: 187, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:34:32,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.483e+02 1.592e+03 1.895e+03 2.532e+03 6.918e+03, threshold=3.790e+03, percent-clipped=10.0
+2023-03-01 15:35:20,249 INFO [train.py:968] (0/2) Epoch 3, batch 15000, giga_loss[loss=0.3423, simple_loss=0.4027, pruned_loss=0.1409, over 28686.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3667, pruned_loss=0.1153, over 5668909.02 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.377, pruned_loss=0.1252, over 5726254.59 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3667, pruned_loss=0.1145, over 5665555.67 frames. ], batch size: 262, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:35:20,253 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 15:35:29,188 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2357, simple_loss=0.3304, pruned_loss=0.07053, over 944034.00 frames. 
+2023-03-01 15:35:29,188 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 15:36:35,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5010, 1.4389, 1.2948, 1.2850], device='cuda:0'), covar=tensor([0.0589, 0.0522, 0.0855, 0.0678], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0483, 0.0538, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 15:36:39,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3101, 1.7448, 1.2009, 1.4593], device='cuda:0'), covar=tensor([0.0864, 0.0327, 0.0399, 0.0973], device='cuda:0'), in_proj_covar=tensor([0.0247, 0.0160, 0.0166, 0.0295], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:0')
+2023-03-01 15:36:43,679 INFO [train.py:968] (0/2) Epoch 3, batch 15050, giga_loss[loss=0.2659, simple_loss=0.3327, pruned_loss=0.09953, over 28834.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3643, pruned_loss=0.1151, over 5664640.80 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.377, pruned_loss=0.1252, over 5728157.51 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3642, pruned_loss=0.1143, over 5659502.59 frames. ], batch size: 174, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:36:55,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2400, 1.0053, 0.6986, 1.3811], device='cuda:0'), covar=tensor([0.0920, 0.0367, 0.0457, 0.1013], device='cuda:0'), in_proj_covar=tensor([0.0249, 0.0161, 0.0167, 0.0297], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:0')
+2023-03-01 15:36:58,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.646e+02 1.528e+03 1.965e+03 2.514e+03 4.961e+03, threshold=3.929e+03, percent-clipped=8.0
+2023-03-01 15:37:52,558 INFO [train.py:968] (0/2) Epoch 3, batch 15100, giga_loss[loss=0.237, simple_loss=0.3174, pruned_loss=0.07829, over 29065.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3576, pruned_loss=0.1119, over 5665627.23 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3768, pruned_loss=0.1252, over 5731833.12 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3574, pruned_loss=0.1111, over 5657185.50 frames. ], batch size: 285, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:38:04,778 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-01 15:38:49,592 INFO [train.py:968] (0/2) Epoch 3, batch 15150, libri_loss[loss=0.3052, simple_loss=0.3543, pruned_loss=0.128, over 29475.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3552, pruned_loss=0.1106, over 5675445.31 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.376, pruned_loss=0.1248, over 5735750.91 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3551, pruned_loss=0.1099, over 5662702.23 frames. ], batch size: 70, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:39:04,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.736e+02 1.504e+03 1.898e+03 2.603e+03 5.112e+03, threshold=3.796e+03, percent-clipped=6.0
+2023-03-01 15:39:42,743 INFO [train.py:968] (0/2) Epoch 3, batch 15200, giga_loss[loss=0.2871, simple_loss=0.3587, pruned_loss=0.1078, over 29103.00 frames. ], tot_loss[loss=0.291, simple_loss=0.357, pruned_loss=0.1125, over 5673565.94 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.375, pruned_loss=0.1243, over 5741699.28 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3571, pruned_loss=0.1119, over 5655820.97 frames. ], batch size: 200, lr: 1.01e-02, grad_scale: 8.0
+2023-03-01 15:40:37,989 INFO [train.py:968] (0/2) Epoch 3, batch 15250, giga_loss[loss=0.2811, simple_loss=0.3546, pruned_loss=0.1038, over 28460.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3568, pruned_loss=0.1118, over 5675915.60 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3745, pruned_loss=0.1239, over 5740786.47 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3569, pruned_loss=0.1114, over 5661098.55 frames. ], batch size: 78, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:40:57,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6404, 1.8201, 1.2498, 1.1973], device='cuda:0'), covar=tensor([0.0782, 0.0548, 0.0497, 0.0614], device='cuda:0'), in_proj_covar=tensor([0.1194, 0.0884, 0.0903, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 15:40:58,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.494e+02 1.513e+03 1.800e+03 2.535e+03 5.192e+03, threshold=3.600e+03, percent-clipped=7.0
+2023-03-01 15:41:40,683 INFO [train.py:968] (0/2) Epoch 3, batch 15300, giga_loss[loss=0.2485, simple_loss=0.3281, pruned_loss=0.08446, over 29111.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3545, pruned_loss=0.1095, over 5663175.13 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3743, pruned_loss=0.1239, over 5742603.29 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3546, pruned_loss=0.1091, over 5649153.83 frames. ], batch size: 113, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:41:41,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3893, 4.0909, 4.0885, 1.8397], device='cuda:0'), covar=tensor([0.0439, 0.0371, 0.0787, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.0753, 0.0591, 0.0758, 0.0551], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 15:42:34,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-01 15:42:43,043 INFO [train.py:968] (0/2) Epoch 3, batch 15350, libri_loss[loss=0.2844, simple_loss=0.3547, pruned_loss=0.1071, over 29551.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3539, pruned_loss=0.1093, over 5676188.50 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3741, pruned_loss=0.1237, over 5746686.53 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3537, pruned_loss=0.1088, over 5659613.90 frames. ], batch size: 84, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:43:06,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.836e+02 1.415e+03 1.905e+03 2.851e+03 1.011e+04, threshold=3.811e+03, percent-clipped=14.0
+2023-03-01 15:43:51,962 INFO [train.py:968] (0/2) Epoch 3, batch 15400, libri_loss[loss=0.2609, simple_loss=0.3293, pruned_loss=0.09624, over 29559.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3534, pruned_loss=0.1091, over 5672101.47 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3734, pruned_loss=0.1232, over 5751416.96 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3532, pruned_loss=0.1087, over 5652192.60 frames. ], batch size: 77, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:44:27,644 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 15:44:55,775 INFO [train.py:968] (0/2) Epoch 3, batch 15450, giga_loss[loss=0.3075, simple_loss=0.3732, pruned_loss=0.1209, over 28741.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3531, pruned_loss=0.1084, over 5665662.90 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3731, pruned_loss=0.123, over 5753253.28 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3531, pruned_loss=0.1081, over 5647267.10 frames. ], batch size: 243, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:45:19,062 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.367e+02 1.211e+03 1.618e+03 2.514e+03 8.357e+03, threshold=3.236e+03, percent-clipped=9.0
+2023-03-01 15:45:48,769 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=105781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:45:58,583 INFO [train.py:968] (0/2) Epoch 3, batch 15500, giga_loss[loss=0.2968, simple_loss=0.3621, pruned_loss=0.1158, over 28982.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3549, pruned_loss=0.1101, over 5672129.12 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3731, pruned_loss=0.1231, over 5755688.92 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3543, pruned_loss=0.1095, over 5653592.05 frames. ], batch size: 145, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:47:05,522 INFO [train.py:968] (0/2) Epoch 3, batch 15550, giga_loss[loss=0.2918, simple_loss=0.3605, pruned_loss=0.1115, over 28955.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3531, pruned_loss=0.109, over 5666058.06 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3729, pruned_loss=0.123, over 5757184.98 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3527, pruned_loss=0.1084, over 5649318.69 frames. ], batch size: 213, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:47:25,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.373e+02 1.240e+03 1.505e+03 1.998e+03 3.826e+03, threshold=3.010e+03, percent-clipped=2.0
+2023-03-01 15:47:25,915 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 15:47:46,765 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=105879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:48:00,305 INFO [train.py:968] (0/2) Epoch 3, batch 15600, giga_loss[loss=0.3171, simple_loss=0.3789, pruned_loss=0.1277, over 28108.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3554, pruned_loss=0.1085, over 5665033.17 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3728, pruned_loss=0.123, over 5745274.17 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3544, pruned_loss=0.1075, over 5660073.06 frames. ], batch size: 412, lr: 1.01e-02, grad_scale: 8.0
+2023-03-01 15:48:04,728 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=105895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:48:57,454 INFO [train.py:968] (0/2) Epoch 3, batch 15650, giga_loss[loss=0.2718, simple_loss=0.3501, pruned_loss=0.09674, over 29001.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3566, pruned_loss=0.1084, over 5668968.73 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3715, pruned_loss=0.1222, over 5749830.92 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3565, pruned_loss=0.1079, over 5658462.11 frames. ], batch size: 128, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:49:18,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.366e+02 1.260e+03 1.841e+03 2.290e+03 4.916e+03, threshold=3.682e+03, percent-clipped=7.0
+2023-03-01 15:49:58,094 INFO [train.py:968] (0/2) Epoch 3, batch 15700, giga_loss[loss=0.3921, simple_loss=0.438, pruned_loss=0.1731, over 28918.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3594, pruned_loss=0.1104, over 5666383.67 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3712, pruned_loss=0.1219, over 5752782.11 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3594, pruned_loss=0.11, over 5653949.34 frames. ], batch size: 284, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:50:08,168 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-106000.pt
+2023-03-01 15:50:20,649 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 15:50:40,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2958, 1.3651, 1.2154, 1.4584], device='cuda:0'), covar=tensor([0.2062, 0.1941, 0.1807, 0.1946], device='cuda:0'), in_proj_covar=tensor([0.0998, 0.0794, 0.0911, 0.0928], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 15:50:58,306 INFO [train.py:968] (0/2) Epoch 3, batch 15750, giga_loss[loss=0.2967, simple_loss=0.364, pruned_loss=0.1147, over 28977.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3599, pruned_loss=0.1101, over 5678394.17 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3714, pruned_loss=0.1218, over 5755110.20 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3595, pruned_loss=0.1097, over 5665020.84 frames. ], batch size: 199, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:51:18,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.635e+02 1.437e+03 1.976e+03 2.557e+03 7.042e+03, threshold=3.952e+03, percent-clipped=11.0
+2023-03-01 15:51:55,325 INFO [train.py:968] (0/2) Epoch 3, batch 15800, giga_loss[loss=0.2425, simple_loss=0.3241, pruned_loss=0.08048, over 28825.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3585, pruned_loss=0.1092, over 5680158.92 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3706, pruned_loss=0.1214, over 5749885.51 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3584, pruned_loss=0.1088, over 5671029.24 frames. ], batch size: 174, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:52:12,852 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-01 15:52:57,917 INFO [train.py:968] (0/2) Epoch 3, batch 15850, libri_loss[loss=0.2801, simple_loss=0.349, pruned_loss=0.1056, over 29556.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3554, pruned_loss=0.1067, over 5685708.76 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3699, pruned_loss=0.1209, over 5752166.17 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3556, pruned_loss=0.1064, over 5674879.69 frames. ], batch size: 77, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:53:16,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:53:18,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.706e+02 1.309e+03 1.691e+03 2.560e+03 1.433e+04, threshold=3.381e+03, percent-clipped=12.0
+2023-03-01 15:53:48,471 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:53:55,101 INFO [train.py:968] (0/2) Epoch 3, batch 15900, giga_loss[loss=0.2653, simple_loss=0.3337, pruned_loss=0.0985, over 29019.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3539, pruned_loss=0.1068, over 5666443.02 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3695, pruned_loss=0.1208, over 5737497.44 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3541, pruned_loss=0.1064, over 5668736.01 frames. ], batch size: 285, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:53:56,711 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:53:59,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3812, 1.5309, 1.2331, 1.3970], device='cuda:0'), covar=tensor([0.0919, 0.0329, 0.0407, 0.1046], device='cuda:0'), in_proj_covar=tensor([0.0248, 0.0158, 0.0163, 0.0293], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:0')
+2023-03-01 15:54:58,721 INFO [train.py:968] (0/2) Epoch 3, batch 15950, giga_loss[loss=0.2752, simple_loss=0.3554, pruned_loss=0.09745, over 28950.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3546, pruned_loss=0.1079, over 5672407.19 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3693, pruned_loss=0.1205, over 5740502.11 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3546, pruned_loss=0.1075, over 5670204.71 frames. ], batch size: 199, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:55:17,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:55:22,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.196e+02 1.187e+03 1.565e+03 2.268e+03 4.546e+03, threshold=3.130e+03, percent-clipped=5.0
+2023-03-01 15:55:39,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:55:58,627 INFO [train.py:968] (0/2) Epoch 3, batch 16000, giga_loss[loss=0.296, simple_loss=0.367, pruned_loss=0.1125, over 28698.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.358, pruned_loss=0.1099, over 5676262.21 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3689, pruned_loss=0.1202, over 5745245.67 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.358, pruned_loss=0.1096, over 5667950.00 frames. ], batch size: 307, lr: 1.01e-02, grad_scale: 8.0
+2023-03-01 15:56:10,354 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:56:14,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:56:16,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2268, 1.4750, 1.2238, 0.7702], device='cuda:0'), covar=tensor([0.0622, 0.0527, 0.0357, 0.0578], device='cuda:0'), in_proj_covar=tensor([0.1193, 0.0872, 0.0895, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 15:56:23,086 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-01 15:56:47,642 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:57:02,600 INFO [train.py:968] (0/2) Epoch 3, batch 16050, giga_loss[loss=0.2624, simple_loss=0.3164, pruned_loss=0.1042, over 24503.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3594, pruned_loss=0.1112, over 5682029.65 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3685, pruned_loss=0.1199, over 5749097.59 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3594, pruned_loss=0.1109, over 5669535.58 frames. ], batch size: 705, lr: 1.01e-02, grad_scale: 8.0
+2023-03-01 15:57:21,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.584e+02 1.434e+03 1.800e+03 2.604e+03 6.719e+03, threshold=3.600e+03, percent-clipped=13.0
+2023-03-01 15:58:01,193 INFO [train.py:968] (0/2) Epoch 3, batch 16100, giga_loss[loss=0.4015, simple_loss=0.4457, pruned_loss=0.1787, over 27672.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3615, pruned_loss=0.1129, over 5675870.20 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1196, over 5743758.53 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3616, pruned_loss=0.1125, over 5667660.40 frames. ], batch size: 472, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:58:07,999 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:58:10,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:58:25,168 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:58:29,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:58:45,641 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:58:59,823 INFO [train.py:968] (0/2) Epoch 3, batch 16150, giga_loss[loss=0.2653, simple_loss=0.3497, pruned_loss=0.09039, over 28724.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3646, pruned_loss=0.1135, over 5685209.60 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1197, over 5745964.21 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3646, pruned_loss=0.1131, over 5676121.78 frames. ], batch size: 119, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:59:06,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:59:10,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6868, 3.4273, 3.3948, 1.5730], device='cuda:0'), covar=tensor([0.0573, 0.0452, 0.0886, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.0751, 0.0597, 0.0759, 0.0551], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 15:59:20,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.874e+02 1.400e+03 1.769e+03 2.733e+03 6.165e+03, threshold=3.538e+03, percent-clipped=11.0
+2023-03-01 15:59:32,735 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:59:57,448 INFO [train.py:968] (0/2) Epoch 3, batch 16200, giga_loss[loss=0.3085, simple_loss=0.3785, pruned_loss=0.1192, over 28652.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3658, pruned_loss=0.1137, over 5675422.61 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.368, pruned_loss=0.1196, over 5737546.32 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3657, pruned_loss=0.1133, over 5673713.82 frames. ], batch size: 307, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 16:00:17,424 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 16:01:03,433 INFO [train.py:968] (0/2) Epoch 3, batch 16250, giga_loss[loss=0.3113, simple_loss=0.3755, pruned_loss=0.1236, over 29008.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3658, pruned_loss=0.114, over 5687311.69 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.368, pruned_loss=0.1196, over 5744198.64 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3657, pruned_loss=0.1134, over 5677663.43 frames. ], batch size: 199, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 16:01:27,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.922e+02 1.513e+03 1.913e+03 2.449e+03 4.908e+03, threshold=3.826e+03, percent-clipped=4.0
+2023-03-01 16:01:27,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:01:37,004 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:02:07,014 INFO [train.py:968] (0/2) Epoch 3, batch 16300, giga_loss[loss=0.2858, simple_loss=0.3623, pruned_loss=0.1046, over 28871.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3632, pruned_loss=0.1128, over 5696265.18 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3679, pruned_loss=0.1196, over 5746989.44 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3631, pruned_loss=0.1123, over 5685277.11 frames. ], batch size: 227, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 16:03:12,713 INFO [train.py:968] (0/2) Epoch 3, batch 16350, giga_loss[loss=0.2804, simple_loss=0.3567, pruned_loss=0.102, over 28638.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3634, pruned_loss=0.1132, over 5665364.83 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3682, pruned_loss=0.1198, over 5726031.67 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3629, pruned_loss=0.1124, over 5674164.66 frames. ], batch size: 307, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 16:03:36,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.600e+02 1.396e+03 1.861e+03 2.860e+03 7.431e+03, threshold=3.722e+03, percent-clipped=13.0
+2023-03-01 16:03:39,628 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:04:19,496 INFO [train.py:968] (0/2) Epoch 3, batch 16400, giga_loss[loss=0.2716, simple_loss=0.3433, pruned_loss=0.09991, over 28909.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3612, pruned_loss=0.1125, over 5663837.99 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3683, pruned_loss=0.1199, over 5727254.11 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3608, pruned_loss=0.1118, over 5669198.46 frames. ], batch size: 186, lr: 1.01e-02, grad_scale: 8.0
+2023-03-01 16:04:33,880 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:04:38,549 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:04:44,451 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:04:47,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:05:05,343 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:05:15,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:05:21,775 INFO [train.py:968] (0/2) Epoch 3, batch 16450, giga_loss[loss=0.2987, simple_loss=0.3693, pruned_loss=0.114, over 28978.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3598, pruned_loss=0.1127, over 5654638.94 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3682, pruned_loss=0.1198, over 5719942.36 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3594, pruned_loss=0.112, over 5663645.13 frames. ], batch size: 284, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 16:05:23,210 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:05:46,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.947e+02 1.275e+03 1.726e+03 2.119e+03 7.156e+03, threshold=3.452e+03, percent-clipped=7.0
+2023-03-01 16:06:19,186 INFO [train.py:968] (0/2) Epoch 3, batch 16500, giga_loss[loss=0.3055, simple_loss=0.3738, pruned_loss=0.1186, over 27651.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3609, pruned_loss=0.1124, over 5667827.88 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3679, pruned_loss=0.1197, over 5724778.38 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3606, pruned_loss=0.1118, over 5669127.97 frames. ], batch size: 472, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:06:58,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2677, 2.0066, 1.2137, 1.4401], device='cuda:0'), covar=tensor([0.0987, 0.0323, 0.0422, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0245, 0.0158, 0.0163, 0.0291], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:0')
+2023-03-01 16:07:21,327 INFO [train.py:968] (0/2) Epoch 3, batch 16550, giga_loss[loss=0.3227, simple_loss=0.3789, pruned_loss=0.1333, over 26803.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3594, pruned_loss=0.1103, over 5667139.39 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.368, pruned_loss=0.1196, over 5726807.95 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.359, pruned_loss=0.1098, over 5666026.06 frames. ], batch size: 555, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:07:27,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:07:29,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:07:36,348 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-01 16:07:44,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.544e+02 1.500e+03 1.948e+03 2.582e+03 6.590e+03, threshold=3.895e+03, percent-clipped=12.0
+2023-03-01 16:08:17,570 INFO [train.py:968] (0/2) Epoch 3, batch 16600, giga_loss[loss=0.3087, simple_loss=0.3701, pruned_loss=0.1236, over 27524.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3609, pruned_loss=0.1089, over 5672982.09 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3679, pruned_loss=0.1196, over 5728916.46 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3604, pruned_loss=0.1083, over 5669031.73 frames. ], batch size: 472, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:09:12,025 INFO [train.py:968] (0/2) Epoch 3, batch 16650, giga_loss[loss=0.2841, simple_loss=0.3606, pruned_loss=0.1038, over 28894.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3627, pruned_loss=0.1098, over 5677146.93 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3679, pruned_loss=0.1197, over 5731076.96 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3621, pruned_loss=0.1089, over 5670470.65 frames. ], batch size: 164, lr: 1.01e-02, grad_scale: 1.0
+2023-03-01 16:09:33,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.323e+02 1.226e+03 1.528e+03 2.329e+03 1.729e+04, threshold=3.056e+03, percent-clipped=10.0
+2023-03-01 16:10:05,130 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:10:08,043 INFO [train.py:968] (0/2) Epoch 3, batch 16700, giga_loss[loss=0.2856, simple_loss=0.3571, pruned_loss=0.107, over 27994.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3625, pruned_loss=0.1097, over 5677842.99 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3679, pruned_loss=0.1195, over 5725819.23 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3619, pruned_loss=0.1087, over 5674824.10 frames. ], batch size: 412, lr: 1.01e-02, grad_scale: 1.0
+2023-03-01 16:10:08,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:10:53,381 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:11:08,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5569, 3.7784, 1.4711, 1.6420], device='cuda:0'), covar=tensor([0.1084, 0.0436, 0.1087, 0.1599], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0438, 0.0311, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 16:11:14,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:11:20,642 INFO [train.py:968] (0/2) Epoch 3, batch 16750, giga_loss[loss=0.2926, simple_loss=0.37, pruned_loss=0.1076, over 28828.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3619, pruned_loss=0.1091, over 5671169.56 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3677, pruned_loss=0.1195, over 5726803.14 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3615, pruned_loss=0.1083, over 5667643.97 frames. ], batch size: 284, lr: 1.01e-02, grad_scale: 1.0
+2023-03-01 16:11:46,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.676e+02 1.325e+03 1.861e+03 2.681e+03 6.020e+03, threshold=3.723e+03, percent-clipped=16.0
+2023-03-01 16:12:26,307 INFO [train.py:968] (0/2) Epoch 3, batch 16800, giga_loss[loss=0.2802, simple_loss=0.3618, pruned_loss=0.09927, over 28671.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3622, pruned_loss=0.1095, over 5674017.55 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3673, pruned_loss=0.1194, over 5730574.28 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.362, pruned_loss=0.1086, over 5666414.47 frames. ], batch size: 307, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:12:37,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 16:12:43,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:13:11,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6170, 2.9688, 1.5332, 1.2957], device='cuda:0'), covar=tensor([0.0817, 0.0321, 0.0866, 0.1365], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0438, 0.0314, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0018], device='cuda:0')
+2023-03-01 16:13:11,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3190, 1.5305, 1.2835, 1.3865], device='cuda:0'), covar=tensor([0.0935, 0.0376, 0.0417, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0246, 0.0158, 0.0165, 0.0293], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:0')
+2023-03-01 16:13:35,264 INFO [train.py:968] (0/2) Epoch 3, batch 16850, giga_loss[loss=0.2899, simple_loss=0.3512, pruned_loss=0.1143, over 26793.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3633, pruned_loss=0.1096, over 5679229.08 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3674, pruned_loss=0.1194, over 5731792.53 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3631, pruned_loss=0.1088, over 5671261.34 frames. ], batch size: 555, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:14:05,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.805e+02 1.460e+03 2.025e+03 2.619e+03 8.326e+03, threshold=4.050e+03, percent-clipped=10.0
+2023-03-01 16:14:05,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9793, 1.0592, 0.9018, 0.7381], device='cuda:0'), covar=tensor([0.0450, 0.0494, 0.0364, 0.0426], device='cuda:0'), in_proj_covar=tensor([0.1181, 0.0875, 0.0892, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 16:14:23,939 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:14:28,184 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:14:40,893 INFO [train.py:968] (0/2) Epoch 3, batch 16900, giga_loss[loss=0.3677, simple_loss=0.4115, pruned_loss=0.162, over 26877.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3661, pruned_loss=0.1115, over 5681786.57 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3678, pruned_loss=0.1196, over 5737038.81 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3655, pruned_loss=0.1103, over 5668611.38 frames. ], batch size: 555, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:14:57,310 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-01 16:15:01,970 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:15:11,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9547, 2.0672, 1.7332, 1.8002], device='cuda:0'), covar=tensor([0.1641, 0.2070, 0.1408, 0.1109], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0755, 0.0720, 0.0628], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 16:15:14,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:15:42,415 INFO [train.py:968] (0/2) Epoch 3, batch 16950, giga_loss[loss=0.2554, simple_loss=0.3361, pruned_loss=0.08734, over 28795.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3678, pruned_loss=0.1121, over 5693487.32 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3673, pruned_loss=0.1194, over 5742165.24 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3676, pruned_loss=0.1111, over 5676888.66 frames. ], batch size: 119, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:15:49,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:15:52,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:16:08,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.964e+02 1.372e+03 1.714e+03 2.217e+03 6.785e+03, threshold=3.429e+03, percent-clipped=5.0
+2023-03-01 16:16:34,497 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:16:49,628 INFO [train.py:968] (0/2) Epoch 3, batch 17000, libri_loss[loss=0.2694, simple_loss=0.345, pruned_loss=0.09685, over 29583.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3655, pruned_loss=0.1112, over 5693430.54 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3667, pruned_loss=0.1191, over 5744182.79 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3659, pruned_loss=0.1104, over 5676871.14 frames. ], batch size: 74, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:17:57,731 INFO [train.py:968] (0/2) Epoch 3, batch 17050, giga_loss[loss=0.3245, simple_loss=0.3668, pruned_loss=0.1411, over 24833.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3636, pruned_loss=0.1109, over 5684700.08 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3675, pruned_loss=0.1197, over 5731647.90 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3632, pruned_loss=0.1095, over 5680748.29 frames. ], batch size: 705, lr: 1.00e-02, grad_scale: 2.0
+2023-03-01 16:18:27,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.373e+02 1.462e+03 1.974e+03 2.548e+03 1.280e+04, threshold=3.948e+03, percent-clipped=17.0
+2023-03-01 16:18:30,698 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:18:33,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:19:09,322 INFO [train.py:968] (0/2) Epoch 3, batch 17100, giga_loss[loss=0.2864, simple_loss=0.3632, pruned_loss=0.1047, over 28726.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3603, pruned_loss=0.1076, over 5690046.57 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3674, pruned_loss=0.1197, over 5731071.30 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.36, pruned_loss=0.1064, over 5686888.15 frames. ], batch size: 243, lr: 1.00e-02, grad_scale: 2.0
+2023-03-01 16:19:20,114 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:20:14,758 INFO [train.py:968] (0/2) Epoch 3, batch 17150, libri_loss[loss=0.295, simple_loss=0.3631, pruned_loss=0.1134, over 29532.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3597, pruned_loss=0.1073, over 5695897.73 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3674, pruned_loss=0.1197, over 5735502.52 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3592, pruned_loss=0.106, over 5688447.42 frames. ], batch size: 89, lr: 1.00e-02, grad_scale: 2.0
+2023-03-01 16:20:39,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.634e+02 1.251e+03 1.594e+03 2.190e+03 6.910e+03, threshold=3.187e+03, percent-clipped=3.0
+2023-03-01 16:21:12,540 INFO [train.py:968] (0/2) Epoch 3, batch 17200, giga_loss[loss=0.3166, simple_loss=0.3828, pruned_loss=0.1252, over 28920.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3632, pruned_loss=0.1099, over 5691597.25 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.368, pruned_loss=0.1201, over 5737678.17 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3622, pruned_loss=0.1083, over 5683116.97 frames. ], batch size: 145, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:22:02,815 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=107533.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 16:22:11,431 INFO [train.py:968] (0/2) Epoch 3, batch 17250, giga_loss[loss=0.2823, simple_loss=0.3647, pruned_loss=0.09999, over 28889.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3648, pruned_loss=0.1109, over 5686043.09 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.1199, over 5739429.92 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3644, pruned_loss=0.1098, over 5677282.04 frames. ], batch size: 213, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:22:17,492 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=107545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:22:35,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.019e+02 1.437e+03 2.103e+03 2.793e+03 5.812e+03, threshold=4.206e+03, percent-clipped=14.0
+2023-03-01 16:23:03,509 INFO [train.py:968] (0/2) Epoch 3, batch 17300, giga_loss[loss=0.2651, simple_loss=0.3375, pruned_loss=0.09636, over 28829.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3625, pruned_loss=0.1103, over 5687756.01 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3671, pruned_loss=0.1194, over 5742773.73 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3626, pruned_loss=0.1096, over 5676031.72 frames. ], batch size: 227, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:24:02,951 INFO [train.py:968] (0/2) Epoch 3, batch 17350, giga_loss[loss=0.2771, simple_loss=0.3556, pruned_loss=0.09933, over 28937.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3617, pruned_loss=0.111, over 5686887.90 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3666, pruned_loss=0.1191, over 5746385.67 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3621, pruned_loss=0.1105, over 5673282.28 frames. ], batch size: 164, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:24:27,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.366e+02 1.527e+03 2.028e+03 2.606e+03 4.846e+03, threshold=4.056e+03, percent-clipped=2.0
+2023-03-01 16:24:30,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3563, 1.4827, 1.1049, 0.8833], device='cuda:0'), covar=tensor([0.0742, 0.0582, 0.0459, 0.0615], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.0888, 0.0906, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 16:24:39,141 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1559, 1.3447, 1.1721, 1.0004], device='cuda:0'), covar=tensor([0.1953, 0.1718, 0.1619, 0.1681], device='cuda:0'), in_proj_covar=tensor([0.0992, 0.0786, 0.0898, 0.0931], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 16:24:57,761 INFO [train.py:968] (0/2) Epoch 3, batch 17400, giga_loss[loss=0.3084, simple_loss=0.3817, pruned_loss=0.1176, over 28923.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3616, pruned_loss=0.1119, over 5696518.76 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3658, pruned_loss=0.1188, over 5752525.58 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3625, pruned_loss=0.1115, over 5677878.95 frames. ], batch size: 199, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:25:49,723 INFO [train.py:968] (0/2) Epoch 3, batch 17450, giga_loss[loss=0.3824, simple_loss=0.4362, pruned_loss=0.1643, over 28888.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3711, pruned_loss=0.1192, over 5683380.43 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3655, pruned_loss=0.1186, over 5745369.11 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3721, pruned_loss=0.119, over 5672306.38 frames. ], batch size: 227, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:26:07,663 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.049e+02 1.268e+03 1.623e+03 2.291e+03 5.444e+03, threshold=3.246e+03, percent-clipped=6.0
+2023-03-01 16:26:34,095 INFO [train.py:968] (0/2) Epoch 3, batch 17500, giga_loss[loss=0.351, simple_loss=0.4131, pruned_loss=0.1444, over 28330.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3812, pruned_loss=0.1255, over 5690436.79 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3652, pruned_loss=0.1183, over 5746720.15 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3824, pruned_loss=0.1256, over 5679546.71 frames. ], batch size: 368, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:27:18,697 INFO [train.py:968] (0/2) Epoch 3, batch 17550, giga_loss[loss=0.3001, simple_loss=0.3678, pruned_loss=0.1162, over 28557.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.384, pruned_loss=0.128, over 5694816.16 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3652, pruned_loss=0.1183, over 5749796.55 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3853, pruned_loss=0.1284, over 5682216.69 frames. ], batch size: 60, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:27:37,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.391e+02 1.290e+03 1.814e+03 2.277e+03 5.403e+03, threshold=3.628e+03, percent-clipped=6.0
+2023-03-01 16:27:51,005 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1622, 1.2064, 0.8297, 0.8775], device='cuda:0'), covar=tensor([0.0437, 0.0411, 0.0315, 0.0413], device='cuda:0'), in_proj_covar=tensor([0.1164, 0.0872, 0.0901, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 16:28:02,684 INFO [train.py:968] (0/2) Epoch 3, batch 17600, giga_loss[loss=0.2689, simple_loss=0.3151, pruned_loss=0.1113, over 23906.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3792, pruned_loss=0.1263, over 5695398.55 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3657, pruned_loss=0.1184, over 5754414.57 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3802, pruned_loss=0.1267, over 5679683.78 frames. ], batch size: 705, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:28:10,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4523, 1.9751, 1.5621, 1.4941], device='cuda:0'), covar=tensor([0.0541, 0.0699, 0.0998, 0.1103], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0475, 0.0530, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 16:28:11,279 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=107903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:28:15,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107908.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 16:28:17,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-01 16:28:22,353 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 16:28:24,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:28:46,079 INFO [train.py:968] (0/2) Epoch 3, batch 17650, giga_loss[loss=0.313, simple_loss=0.3456, pruned_loss=0.1403, over 24028.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3711, pruned_loss=0.1225, over 5691775.35 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3659, pruned_loss=0.1184, over 5756509.48 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.372, pruned_loss=0.123, over 5676090.02 frames. ], batch size: 710, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:29:00,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2159, 2.8830, 2.9366, 1.4605], device='cuda:0'), covar=tensor([0.0738, 0.0557, 0.1055, 0.2026], device='cuda:0'), in_proj_covar=tensor([0.0750, 0.0603, 0.0771, 0.0557], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 16:29:05,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.184e+02 1.073e+03 1.326e+03 1.671e+03 3.280e+03, threshold=2.652e+03, percent-clipped=0.0
+2023-03-01 16:29:28,779 INFO [train.py:968] (0/2) Epoch 3, batch 17700, giga_loss[loss=0.2544, simple_loss=0.3239, pruned_loss=0.0924, over 28953.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3628, pruned_loss=0.1183, over 5698423.22 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3658, pruned_loss=0.1182, over 5758457.08 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3636, pruned_loss=0.1189, over 5682987.98 frames. ], batch size: 164, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:29:37,974 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-108000.pt
+2023-03-01 16:30:03,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4663, 4.9272, 5.0901, 2.6376], device='cuda:0'), covar=tensor([0.0287, 0.0269, 0.0568, 0.1514], device='cuda:0'), in_proj_covar=tensor([0.0749, 0.0597, 0.0765, 0.0560], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 16:30:12,520 INFO [train.py:968] (0/2) Epoch 3, batch 17750, giga_loss[loss=0.2639, simple_loss=0.3197, pruned_loss=0.104, over 28741.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3546, pruned_loss=0.1146, over 5698087.26 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3657, pruned_loss=0.1182, over 5762014.67 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3551, pruned_loss=0.115, over 5681003.77 frames. ], batch size: 92, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:30:20,034 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108051.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 16:30:20,515 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:30:21,894 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108054.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 16:30:30,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:30:30,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.900e+02 1.106e+03 1.343e+03 1.909e+03 4.261e+03, threshold=2.686e+03, percent-clipped=10.0
+2023-03-01 16:30:32,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:30:34,563 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3725, 1.4105, 1.1842, 1.5043], device='cuda:0'), covar=tensor([0.0874, 0.0360, 0.0397, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.0240, 0.0157, 0.0160, 0.0286], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0045], device='cuda:0')
+2023-03-01 16:30:45,206 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0142, 1.1703, 0.8296, 0.6131], device='cuda:0'), covar=tensor([0.0527, 0.0545, 0.0479, 0.0490], device='cuda:0'), in_proj_covar=tensor([0.1180, 0.0881, 0.0916, 0.0991], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 16:30:45,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1274, 0.9513, 0.7901, 1.2876], device='cuda:0'), covar=tensor([0.0892, 0.0429, 0.0441, 0.1015], device='cuda:0'), in_proj_covar=tensor([0.0241, 0.0157, 0.0161, 0.0287], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:0')
+2023-03-01 16:30:47,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108083.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 16:30:52,450 INFO [train.py:968] (0/2) Epoch 3, batch 17800, libri_loss[loss=0.3287, simple_loss=0.3934, pruned_loss=0.132, over 29761.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3499, pruned_loss=0.1127, over 5692811.11 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3661, pruned_loss=0.1186, over 5754457.40 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3494, pruned_loss=0.1125, over 5684215.91 frames. ], batch size: 87, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:30:57,091 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:31:27,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:31:32,672 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9741, 1.8507, 1.6314, 1.7535], device='cuda:0'), covar=tensor([0.1882, 0.2458, 0.1590, 0.1210], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0778, 0.0735, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 16:31:34,704 INFO [train.py:968] (0/2) Epoch 3, batch 17850, giga_loss[loss=0.2765, simple_loss=0.3392, pruned_loss=0.1069, over 28570.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3464, pruned_loss=0.1109, over 5694597.93 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3666, pruned_loss=0.1187, over 5754079.94 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3453, pruned_loss=0.1105, over 5687137.59 frames. ], batch size: 336, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:31:52,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.611e+02 1.089e+03 1.480e+03 2.074e+03 8.524e+03, threshold=2.961e+03, percent-clipped=14.0
+2023-03-01 16:32:15,338 INFO [train.py:968] (0/2) Epoch 3, batch 17900, giga_loss[loss=0.2443, simple_loss=0.313, pruned_loss=0.08781, over 28985.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3443, pruned_loss=0.1099, over 5690759.50 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3675, pruned_loss=0.1191, over 5748977.77 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3418, pruned_loss=0.1088, over 5687273.94 frames. ], batch size: 164, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:32:55,870 INFO [train.py:968] (0/2) Epoch 3, batch 17950, giga_loss[loss=0.2761, simple_loss=0.3335, pruned_loss=0.1093, over 28900.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3412, pruned_loss=0.1083, over 5694135.95 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3682, pruned_loss=0.1195, over 5749863.60 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.338, pruned_loss=0.1069, over 5689506.97 frames. ], batch size: 227, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:33:03,511 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-01 16:33:16,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.026e+02 1.107e+03 1.433e+03 2.118e+03 6.492e+03, threshold=2.865e+03, percent-clipped=10.0
+2023-03-01 16:33:26,472 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-01 16:33:28,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:33:37,493 INFO [train.py:968] (0/2) Epoch 3, batch 18000, giga_loss[loss=0.2326, simple_loss=0.3012, pruned_loss=0.08196, over 28600.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3377, pruned_loss=0.106, over 5696753.41 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3689, pruned_loss=0.1198, over 5750945.74 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3336, pruned_loss=0.1042, over 5690473.70 frames. ], batch size: 307, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:33:37,503 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 16:33:43,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2090, 1.3327, 1.0210, 1.3644], device='cuda:0'), covar=tensor([0.1054, 0.0390, 0.0497, 0.1167], device='cuda:0'), in_proj_covar=tensor([0.0241, 0.0156, 0.0161, 0.0284], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0045], device='cuda:0')
+2023-03-01 16:33:44,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4470, 1.4190, 1.4145, 1.4497], device='cuda:0'), covar=tensor([0.0846, 0.1249, 0.1380, 0.0971], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0772, 0.0622, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 16:33:45,891 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2527, simple_loss=0.35, pruned_loss=0.07767, over 944034.00 frames. 
+2023-03-01 16:33:45,891 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 16:34:26,710 INFO [train.py:968] (0/2) Epoch 3, batch 18050, giga_loss[loss=0.2701, simple_loss=0.3262, pruned_loss=0.107, over 28846.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3353, pruned_loss=0.1045, over 5693449.59 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3697, pruned_loss=0.1202, over 5751376.52 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3303, pruned_loss=0.1024, over 5686057.26 frames. ], batch size: 199, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:34:45,736 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.181e+02 9.420e+02 1.245e+03 1.808e+03 5.452e+03, threshold=2.490e+03, percent-clipped=4.0
+2023-03-01 16:35:07,622 INFO [train.py:968] (0/2) Epoch 3, batch 18100, giga_loss[loss=0.2441, simple_loss=0.3084, pruned_loss=0.08989, over 29056.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3345, pruned_loss=0.1043, over 5702089.36 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3707, pruned_loss=0.1208, over 5758158.21 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3276, pruned_loss=0.1011, over 5686987.90 frames. ], batch size: 136, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:35:10,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4376, 3.1195, 3.1910, 1.5988], device='cuda:0'), covar=tensor([0.0619, 0.0520, 0.0825, 0.2110], device='cuda:0'), in_proj_covar=tensor([0.0745, 0.0599, 0.0754, 0.0554], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 16:35:10,334 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-01 16:35:27,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2469, 1.4837, 1.2143, 1.3763], device='cuda:0'), covar=tensor([0.1932, 0.1820, 0.1721, 0.1759], device='cuda:0'), in_proj_covar=tensor([0.1002, 0.0793, 0.0903, 0.0926], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 16:35:34,327 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:35:37,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:35:39,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:35:50,723 INFO [train.py:968] (0/2) Epoch 3, batch 18150, giga_loss[loss=0.2568, simple_loss=0.326, pruned_loss=0.09377, over 29088.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3299, pruned_loss=0.1018, over 5698933.46 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3711, pruned_loss=0.1209, over 5757383.34 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3239, pruned_loss=0.09899, over 5687171.11 frames. ], batch size: 155, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:36:03,950 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:36:12,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.117e+02 1.052e+03 1.354e+03 1.916e+03 4.436e+03, threshold=2.707e+03, percent-clipped=12.0
+2023-03-01 16:36:22,650 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-01 16:36:39,407 INFO [train.py:968] (0/2) Epoch 3, batch 18200, giga_loss[loss=0.2281, simple_loss=0.2942, pruned_loss=0.08103, over 28850.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3258, pruned_loss=0.09992, over 5676396.25 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3714, pruned_loss=0.1211, over 5749429.08 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3202, pruned_loss=0.09729, over 5673802.85 frames. ], batch size: 243, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:36:50,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:36:57,041 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 16:37:22,361 INFO [train.py:968] (0/2) Epoch 3, batch 18250, giga_loss[loss=0.3298, simple_loss=0.374, pruned_loss=0.1428, over 27679.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3255, pruned_loss=0.1001, over 5677018.02 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3718, pruned_loss=0.1212, over 5751322.38 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3201, pruned_loss=0.09764, over 5672361.10 frames. ], batch size: 472, lr: 9.99e-03, grad_scale: 8.0
+2023-03-01 16:37:48,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.245e+02 9.580e+02 1.227e+03 1.940e+03 4.173e+03, threshold=2.454e+03, percent-clipped=11.0
+2023-03-01 16:37:53,707 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:37:55,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:38:12,174 INFO [train.py:968] (0/2) Epoch 3, batch 18300, giga_loss[loss=0.3389, simple_loss=0.3977, pruned_loss=0.1401, over 28736.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.34, pruned_loss=0.1087, over 5676921.64 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3725, pruned_loss=0.1215, over 5751655.17 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3345, pruned_loss=0.1062, over 5671967.28 frames. ], batch size: 92, lr: 9.99e-03, grad_scale: 4.0
+2023-03-01 16:38:23,310 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:38:56,972 INFO [train.py:968] (0/2) Epoch 3, batch 18350, giga_loss[loss=0.3527, simple_loss=0.4123, pruned_loss=0.1466, over 28788.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3567, pruned_loss=0.119, over 5686568.16 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3729, pruned_loss=0.1217, over 5755238.49 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3516, pruned_loss=0.1168, over 5678126.10 frames. ], batch size: 92, lr: 9.99e-03, grad_scale: 4.0
+2023-03-01 16:39:03,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:39:05,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:39:17,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.607e+02 1.432e+03 1.727e+03 2.223e+03 4.959e+03, threshold=3.453e+03, percent-clipped=21.0
+2023-03-01 16:39:21,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3778, 1.2848, 1.2672, 1.7792], device='cuda:0'), covar=tensor([0.1852, 0.1816, 0.1533, 0.1918], device='cuda:0'), in_proj_covar=tensor([0.1009, 0.0806, 0.0907, 0.0924], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 16:39:29,650 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:39:34,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0326, 2.1262, 1.5350, 1.7928], device='cuda:0'), covar=tensor([0.0725, 0.0686, 0.1032, 0.0928], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0481, 0.0524, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 16:39:39,891 INFO [train.py:968] (0/2) Epoch 3, batch 18400, giga_loss[loss=0.2912, simple_loss=0.3649, pruned_loss=0.1088, over 29011.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3672, pruned_loss=0.1248, over 5690204.04 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.373, pruned_loss=0.1217, over 5758750.12 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3628, pruned_loss=0.123, over 5678995.50 frames. ], batch size: 155, lr: 9.99e-03, grad_scale: 8.0
+2023-03-01 16:40:02,833 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:40:22,883 INFO [train.py:968] (0/2) Epoch 3, batch 18450, giga_loss[loss=0.3018, simple_loss=0.3717, pruned_loss=0.116, over 29019.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3736, pruned_loss=0.1268, over 5693728.95 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3738, pruned_loss=0.1221, over 5763545.55 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3692, pruned_loss=0.1252, over 5678032.32 frames. ], batch size: 128, lr: 9.99e-03, grad_scale: 8.0
+2023-03-01 16:40:39,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7404, 1.5603, 1.4873, 1.8171], device='cuda:0'), covar=tensor([0.1779, 0.1895, 0.1597, 0.2056], device='cuda:0'), in_proj_covar=tensor([0.1001, 0.0803, 0.0902, 0.0924], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 16:40:43,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.005e+02 1.171e+03 1.315e+03 1.763e+03 3.208e+03, threshold=2.629e+03, percent-clipped=0.0
+2023-03-01 16:40:50,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3027, 1.4556, 1.0353, 1.1293], device='cuda:0'), covar=tensor([0.0677, 0.0470, 0.1007, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0478, 0.0524, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 16:41:03,867 INFO [train.py:968] (0/2) Epoch 3, batch 18500, giga_loss[loss=0.3811, simple_loss=0.4282, pruned_loss=0.167, over 27683.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3753, pruned_loss=0.1261, over 5699860.81 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3739, pruned_loss=0.1221, over 5766685.42 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3718, pruned_loss=0.1248, over 5683381.62 frames. ], batch size: 472, lr: 9.98e-03, grad_scale: 4.0
+2023-03-01 16:41:36,151 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:41:44,886 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 16:41:49,621 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2606, 1.2936, 1.1015, 1.1623], device='cuda:0'), covar=tensor([0.0597, 0.0486, 0.0972, 0.0690], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0472, 0.0518, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 16:41:50,815 INFO [train.py:968] (0/2) Epoch 3, batch 18550, giga_loss[loss=0.312, simple_loss=0.374, pruned_loss=0.125, over 28555.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3769, pruned_loss=0.1263, over 5682662.72 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3743, pruned_loss=0.1222, over 5767440.86 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3737, pruned_loss=0.1253, over 5666472.09 frames. ], batch size: 85, lr: 9.98e-03, grad_scale: 4.0
+2023-03-01 16:42:13,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.395e+02 1.155e+03 1.398e+03 1.749e+03 4.741e+03, threshold=2.795e+03, percent-clipped=7.0
+2023-03-01 16:42:33,394 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 16:42:34,357 INFO [train.py:968] (0/2) Epoch 3, batch 18600, giga_loss[loss=0.3186, simple_loss=0.3825, pruned_loss=0.1274, over 28377.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3798, pruned_loss=0.1289, over 5686859.74 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3745, pruned_loss=0.1223, over 5770327.20 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3771, pruned_loss=0.1281, over 5669254.82 frames. ], batch size: 65, lr: 9.98e-03, grad_scale: 4.0
+2023-03-01 16:42:52,746 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6687, 1.5201, 1.3471, 1.3285], device='cuda:0'), covar=tensor([0.0509, 0.0407, 0.0702, 0.0744], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0482, 0.0530, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 16:43:17,865 INFO [train.py:968] (0/2) Epoch 3, batch 18650, giga_loss[loss=0.3196, simple_loss=0.3875, pruned_loss=0.1258, over 29057.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3829, pruned_loss=0.1313, over 5674366.91 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3751, pruned_loss=0.1225, over 5760019.08 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3804, pruned_loss=0.1307, over 5668468.28 frames. ], batch size: 155, lr: 9.98e-03, grad_scale: 4.0
+2023-03-01 16:43:21,694 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3690, 1.0809, 2.8604, 2.6075], device='cuda:0'), covar=tensor([0.1604, 0.1859, 0.0464, 0.0651], device='cuda:0'), in_proj_covar=tensor([0.0518, 0.0495, 0.0661, 0.0529], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 16:43:25,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3225, 1.9992, 1.4262, 0.4844], device='cuda:0'), covar=tensor([0.2097, 0.0970, 0.1296, 0.1859], device='cuda:0'), in_proj_covar=tensor([0.1266, 0.1213, 0.1281, 0.1088], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 16:43:39,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.804e+02 1.152e+03 1.367e+03 1.933e+03 4.115e+03, threshold=2.734e+03, percent-clipped=6.0
+2023-03-01 16:43:44,117 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:44:01,854 INFO [train.py:968] (0/2) Epoch 3, batch 18700, giga_loss[loss=0.3125, simple_loss=0.378, pruned_loss=0.1235, over 28537.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3856, pruned_loss=0.1323, over 5680164.88 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3754, pruned_loss=0.1226, over 5762775.95 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3835, pruned_loss=0.1319, over 5671506.73 frames. ], batch size: 78, lr: 9.97e-03, grad_scale: 4.0
+2023-03-01 16:44:07,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 16:44:40,811 INFO [train.py:968] (0/2) Epoch 3, batch 18750, giga_loss[loss=0.3041, simple_loss=0.3752, pruned_loss=0.1165, over 28770.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3881, pruned_loss=0.1331, over 5690175.49 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3758, pruned_loss=0.1229, over 5767016.85 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3864, pruned_loss=0.1327, over 5677336.70 frames. ], batch size: 119, lr: 9.97e-03, grad_scale: 4.0
+2023-03-01 16:44:58,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7225, 1.0768, 3.4868, 2.8378], device='cuda:0'), covar=tensor([0.1531, 0.2036, 0.0332, 0.0508], device='cuda:0'), in_proj_covar=tensor([0.0518, 0.0495, 0.0654, 0.0526], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 16:45:00,012 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-01 16:45:03,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.557e+02 1.094e+03 1.417e+03 1.975e+03 9.111e+03, threshold=2.834e+03, percent-clipped=10.0
+2023-03-01 16:45:23,534 INFO [train.py:968] (0/2) Epoch 3, batch 18800, giga_loss[loss=0.3205, simple_loss=0.3907, pruned_loss=0.1251, over 28605.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.39, pruned_loss=0.1336, over 5689504.10 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3761, pruned_loss=0.1231, over 5768002.99 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3885, pruned_loss=0.1333, over 5677844.23 frames. ], batch size: 336, lr: 9.97e-03, grad_scale: 8.0
+2023-03-01 16:45:25,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:46:05,547 INFO [train.py:968] (0/2) Epoch 3, batch 18850, giga_loss[loss=0.3234, simple_loss=0.3915, pruned_loss=0.1277, over 28660.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3903, pruned_loss=0.1325, over 5688796.02 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3764, pruned_loss=0.1232, over 5766096.36 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3891, pruned_loss=0.1322, over 5680405.53 frames. ], batch size: 92, lr: 9.97e-03, grad_scale: 8.0
+2023-03-01 16:46:24,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.782e+02 1.211e+03 1.494e+03 2.116e+03 4.546e+03, threshold=2.987e+03, percent-clipped=13.0
+2023-03-01 16:46:40,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:46:47,293 INFO [train.py:968] (0/2) Epoch 3, batch 18900, giga_loss[loss=0.2966, simple_loss=0.3741, pruned_loss=0.1096, over 28535.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3888, pruned_loss=0.1294, over 5703642.21 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3765, pruned_loss=0.1233, over 5766776.88 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3877, pruned_loss=0.1292, over 5696241.76 frames. ], batch size: 65, lr: 9.96e-03, grad_scale: 8.0
+2023-03-01 16:46:48,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8637, 1.7154, 1.6646, 1.7895], device='cuda:0'), covar=tensor([0.1064, 0.1857, 0.1523, 0.1466], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0777, 0.0628, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 16:46:54,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:47:22,745 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:47:24,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:47:25,577 INFO [train.py:968] (0/2) Epoch 3, batch 18950, giga_loss[loss=0.4138, simple_loss=0.4301, pruned_loss=0.1987, over 26654.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3869, pruned_loss=0.1278, over 5698149.70 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3774, pruned_loss=0.124, over 5757977.60 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3856, pruned_loss=0.1272, over 5697845.86 frames. ], batch size: 555, lr: 9.96e-03, grad_scale: 4.0
+2023-03-01 16:47:33,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7402, 2.7997, 1.7373, 0.7599], device='cuda:0'), covar=tensor([0.2263, 0.0785, 0.1659, 0.2308], device='cuda:0'), in_proj_covar=tensor([0.1249, 0.1170, 0.1259, 0.1060], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 16:47:46,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.835e+02 9.856e+02 1.246e+03 1.919e+03 4.432e+03, threshold=2.493e+03, percent-clipped=5.0
+2023-03-01 16:47:47,436 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:48:01,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6922, 1.1971, 3.4575, 2.9692], device='cuda:0'), covar=tensor([0.1606, 0.1928, 0.0362, 0.0494], device='cuda:0'), in_proj_covar=tensor([0.0524, 0.0500, 0.0665, 0.0531], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 16:48:04,173 INFO [train.py:968] (0/2) Epoch 3, batch 19000, giga_loss[loss=0.4143, simple_loss=0.4412, pruned_loss=0.1937, over 27642.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3886, pruned_loss=0.1296, over 5705818.61 frames. ], libri_tot_loss[loss=0.3133, simple_loss=0.3781, pruned_loss=0.1242, over 5761865.68 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3872, pruned_loss=0.129, over 5699930.66 frames. ], batch size: 472, lr: 9.96e-03, grad_scale: 4.0
+2023-03-01 16:48:08,632 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-01 16:48:27,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8052, 4.0776, 4.5241, 1.9793], device='cuda:0'), covar=tensor([0.0370, 0.0420, 0.0678, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0577, 0.0758, 0.0552], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 16:48:48,512 INFO [train.py:968] (0/2) Epoch 3, batch 19050, giga_loss[loss=0.3685, simple_loss=0.4111, pruned_loss=0.163, over 28866.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3904, pruned_loss=0.1335, over 5702264.37 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3787, pruned_loss=0.1246, over 5751681.40 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3891, pruned_loss=0.1328, over 5705233.86 frames. ], batch size: 199, lr: 9.96e-03, grad_scale: 4.0
+2023-03-01 16:48:50,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:48:51,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:48:52,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:49:10,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2539, 2.1211, 1.6055, 0.6860], device='cuda:0'), covar=tensor([0.1410, 0.0840, 0.1454, 0.1374], device='cuda:0'), in_proj_covar=tensor([0.1257, 0.1190, 0.1274, 0.1072], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 16:49:11,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.484e+02 1.293e+03 1.740e+03 2.198e+03 4.645e+03, threshold=3.480e+03, percent-clipped=16.0
+2023-03-01 16:49:19,026 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:49:31,550 INFO [train.py:968] (0/2) Epoch 3, batch 19100, giga_loss[loss=0.3787, simple_loss=0.4133, pruned_loss=0.172, over 28636.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.393, pruned_loss=0.1383, over 5704931.65 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.379, pruned_loss=0.125, over 5752986.98 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3917, pruned_loss=0.1375, over 5705756.13 frames. ], batch size: 60, lr: 9.96e-03, grad_scale: 4.0
+2023-03-01 16:49:32,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1190, 1.1592, 0.8119, 0.9002], device='cuda:0'), covar=tensor([0.0505, 0.0423, 0.0326, 0.0470], device='cuda:0'), in_proj_covar=tensor([0.1182, 0.0894, 0.0948, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 16:49:37,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8143, 2.2456, 1.9339, 1.8202], device='cuda:0'), covar=tensor([0.1390, 0.1617, 0.1159, 0.0910], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0779, 0.0734, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 16:50:01,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109430.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 16:50:10,179 INFO [train.py:968] (0/2) Epoch 3, batch 19150, giga_loss[loss=0.3783, simple_loss=0.397, pruned_loss=0.1797, over 23837.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3917, pruned_loss=0.1384, over 5702444.89 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3795, pruned_loss=0.1252, over 5759353.66 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3908, pruned_loss=0.1381, over 5695530.76 frames. ], batch size: 705, lr: 9.95e-03, grad_scale: 4.0
+2023-03-01 16:50:30,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.936e+02 1.232e+03 1.657e+03 2.474e+03 8.398e+03, threshold=3.313e+03, percent-clipped=11.0
+2023-03-01 16:50:47,937 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:50:50,869 INFO [train.py:968] (0/2) Epoch 3, batch 19200, libri_loss[loss=0.2779, simple_loss=0.3532, pruned_loss=0.1013, over 29578.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3892, pruned_loss=0.1373, over 5699055.66 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3802, pruned_loss=0.1259, over 5754750.26 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3882, pruned_loss=0.137, over 5695125.99 frames. ], batch size: 75, lr: 9.95e-03, grad_scale: 8.0
+2023-03-01 16:50:51,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:51:12,976 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:51:31,533 INFO [train.py:968] (0/2) Epoch 3, batch 19250, libri_loss[loss=0.3188, simple_loss=0.3789, pruned_loss=0.1294, over 29586.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3872, pruned_loss=0.1353, over 5708686.13 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3804, pruned_loss=0.126, over 5756639.20 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3865, pruned_loss=0.1353, over 5702241.97 frames. ], batch size: 76, lr: 9.95e-03, grad_scale: 4.0
+2023-03-01 16:51:32,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2453, 2.6320, 1.3613, 1.1665], device='cuda:0'), covar=tensor([0.0911, 0.0403, 0.0828, 0.1425], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0439, 0.0309, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:0')
+2023-03-01 16:51:47,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:51:54,383 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.928e+02 1.271e+03 1.500e+03 2.009e+03 7.669e+03, threshold=3.001e+03, percent-clipped=3.0
+2023-03-01 16:52:10,236 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:52:12,818 INFO [train.py:968] (0/2) Epoch 3, batch 19300, libri_loss[loss=0.3837, simple_loss=0.4352, pruned_loss=0.1661, over 27644.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3859, pruned_loss=0.1335, over 5713003.20 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3805, pruned_loss=0.1261, over 5756059.00 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3853, pruned_loss=0.1334, over 5708158.99 frames. ], batch size: 115, lr: 9.95e-03, grad_scale: 4.0
+2023-03-01 16:52:32,781 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 16:52:57,589 INFO [train.py:968] (0/2) Epoch 3, batch 19350, libri_loss[loss=0.3626, simple_loss=0.4293, pruned_loss=0.1479, over 29257.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3842, pruned_loss=0.1321, over 5695549.69 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3812, pruned_loss=0.1263, over 5758094.54 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3831, pruned_loss=0.1319, over 5689028.74 frames. ], batch size: 97, lr: 9.94e-03, grad_scale: 4.0
+2023-03-01 16:53:20,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.723e+02 1.052e+03 1.418e+03 1.936e+03 3.938e+03, threshold=2.836e+03, percent-clipped=9.0
+2023-03-01 16:53:37,021 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:53:39,620 INFO [train.py:968] (0/2) Epoch 3, batch 19400, libri_loss[loss=0.342, simple_loss=0.4152, pruned_loss=0.1344, over 28709.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3773, pruned_loss=0.1273, over 5697564.62 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3813, pruned_loss=0.1262, over 5758026.20 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3763, pruned_loss=0.1273, over 5690979.63 frames. ], batch size: 106, lr: 9.94e-03, grad_scale: 4.0
+2023-03-01 16:53:52,696 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:53:54,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:54:02,116 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:54:16,711 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3850, 1.6527, 1.2871, 1.5001], device='cuda:0'), covar=tensor([0.0919, 0.0362, 0.0401, 0.1025], device='cuda:0'), in_proj_covar=tensor([0.0238, 0.0154, 0.0161, 0.0283], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:0')
+2023-03-01 16:54:21,494 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:54:27,399 INFO [train.py:968] (0/2) Epoch 3, batch 19450, giga_loss[loss=0.2806, simple_loss=0.3505, pruned_loss=0.1054, over 28653.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3702, pruned_loss=0.1231, over 5685796.64 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3814, pruned_loss=0.1262, over 5760499.41 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3691, pruned_loss=0.1231, over 5677358.31 frames. ], batch size: 78, lr: 9.94e-03, grad_scale: 4.0
+2023-03-01 16:54:48,847 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-01 16:54:52,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.006e+02 9.953e+02 1.273e+03 1.560e+03 3.704e+03, threshold=2.546e+03, percent-clipped=4.0
+2023-03-01 16:55:11,613 INFO [train.py:968] (0/2) Epoch 3, batch 19500, giga_loss[loss=0.3008, simple_loss=0.3617, pruned_loss=0.1199, over 28908.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3656, pruned_loss=0.1204, over 5661498.36 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3826, pruned_loss=0.1269, over 5750476.66 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3629, pruned_loss=0.1194, over 5660932.41 frames. ], batch size: 112, lr: 9.94e-03, grad_scale: 4.0
+2023-03-01 16:55:30,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109805.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 16:55:32,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5754, 2.9831, 1.6103, 1.4791], device='cuda:0'), covar=tensor([0.0875, 0.0390, 0.0783, 0.1367], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0436, 0.0307, 0.0396], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0018], device='cuda:0')
+2023-03-01 16:56:01,584 INFO [train.py:968] (0/2) Epoch 3, batch 19550, giga_loss[loss=0.2865, simple_loss=0.3499, pruned_loss=0.1115, over 28597.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3661, pruned_loss=0.121, over 5658185.58 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.383, pruned_loss=0.1271, over 5753857.69 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3632, pruned_loss=0.12, over 5652729.33 frames. ], batch size: 85, lr: 9.94e-03, grad_scale: 2.0
+2023-03-01 16:56:13,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4814, 2.3554, 1.6683, 1.6691], device='cuda:0'), covar=tensor([0.0592, 0.0651, 0.0931, 0.0972], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0471, 0.0527, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 16:56:26,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.889e+02 1.092e+03 1.401e+03 2.279e+03 1.097e+04, threshold=2.803e+03, percent-clipped=19.0
+2023-03-01 16:56:45,384 INFO [train.py:968] (0/2) Epoch 3, batch 19600, giga_loss[loss=0.3148, simple_loss=0.3776, pruned_loss=0.1261, over 29020.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3667, pruned_loss=0.1214, over 5658418.87 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3836, pruned_loss=0.1273, over 5745937.29 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3635, pruned_loss=0.1203, over 5660325.45 frames. ], batch size: 213, lr: 9.93e-03, grad_scale: 4.0
+2023-03-01 16:57:06,732 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-01 16:57:22,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4043, 1.5046, 1.4991, 1.5706], device='cuda:0'), covar=tensor([0.0928, 0.1129, 0.1037, 0.0942], device='cuda:0'), in_proj_covar=tensor([0.0450, 0.0773, 0.0617, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 16:57:29,520 INFO [train.py:968] (0/2) Epoch 3, batch 19650, giga_loss[loss=0.281, simple_loss=0.3496, pruned_loss=0.1061, over 28556.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3652, pruned_loss=0.1209, over 5666228.79 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3841, pruned_loss=0.1276, over 5747656.28 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3621, pruned_loss=0.1196, over 5665396.55 frames. ], batch size: 336, lr: 9.93e-03, grad_scale: 4.0
+2023-03-01 16:57:35,399 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109948.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 16:57:37,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109951.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 16:57:47,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:57:52,908 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.678e+02 1.009e+03 1.293e+03 1.698e+03 5.106e+03, threshold=2.586e+03, percent-clipped=5.0
+2023-03-01 16:57:55,854 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:58:01,013 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109980.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 16:58:09,733 INFO [train.py:968] (0/2) Epoch 3, batch 19700, giga_loss[loss=0.3152, simple_loss=0.3673, pruned_loss=0.1315, over 28202.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3626, pruned_loss=0.1193, over 5679469.20 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3848, pruned_loss=0.1281, over 5750071.31 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3592, pruned_loss=0.1179, over 5675673.98 frames. ], batch size: 77, lr: 9.93e-03, grad_scale: 4.0
+2023-03-01 16:58:16,443 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-110000.pt
+2023-03-01 16:58:45,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7782, 2.0607, 1.9492, 1.8636], device='cuda:0'), covar=tensor([0.1437, 0.1575, 0.1048, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0778, 0.0734, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 16:58:47,600 INFO [train.py:968] (0/2) Epoch 3, batch 19750, giga_loss[loss=0.2495, simple_loss=0.3238, pruned_loss=0.08754, over 28681.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3604, pruned_loss=0.1179, over 5687887.25 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3849, pruned_loss=0.1279, over 5752151.42 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3569, pruned_loss=0.1165, over 5680745.64 frames. ], batch size: 262, lr: 9.93e-03, grad_scale: 4.0
+2023-03-01 16:58:57,417 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:59:05,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:59:11,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.112e+02 1.112e+03 1.489e+03 2.103e+03 5.417e+03, threshold=2.979e+03, percent-clipped=13.0
+2023-03-01 16:59:15,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2284, 1.2649, 1.1551, 1.3273], device='cuda:0'), covar=tensor([0.2071, 0.1962, 0.1832, 0.2014], device='cuda:0'), in_proj_covar=tensor([0.1014, 0.0815, 0.0912, 0.0938], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 16:59:25,347 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:59:28,525 INFO [train.py:968] (0/2) Epoch 3, batch 19800, giga_loss[loss=0.2634, simple_loss=0.3217, pruned_loss=0.1025, over 28646.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3568, pruned_loss=0.1155, over 5687938.93 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3856, pruned_loss=0.1283, over 5743463.10 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.353, pruned_loss=0.114, over 5688366.26 frames. ], batch size: 92, lr: 9.92e-03, grad_scale: 2.0
+2023-03-01 16:59:40,291 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:59:44,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:59:49,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4994, 3.2083, 3.1840, 1.6317], device='cuda:0'), covar=tensor([0.0604, 0.0481, 0.0810, 0.2265], device='cuda:0'), in_proj_covar=tensor([0.0751, 0.0594, 0.0776, 0.0571], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 17:00:06,656 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:00:10,038 INFO [train.py:968] (0/2) Epoch 3, batch 19850, giga_loss[loss=0.3066, simple_loss=0.3643, pruned_loss=0.1245, over 28900.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3551, pruned_loss=0.1146, over 5697464.67 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.387, pruned_loss=0.129, over 5746182.42 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3502, pruned_loss=0.1125, over 5694318.69 frames. ], batch size: 186, lr: 9.92e-03, grad_scale: 2.0
+2023-03-01 17:00:32,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.498e+02 1.105e+03 1.664e+03 2.458e+03 9.372e+03, threshold=3.328e+03, percent-clipped=15.0
+2023-03-01 17:00:48,837 INFO [train.py:968] (0/2) Epoch 3, batch 19900, libri_loss[loss=0.3666, simple_loss=0.4407, pruned_loss=0.1463, over 29288.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3538, pruned_loss=0.1141, over 5708999.50 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3882, pruned_loss=0.1295, over 5749651.68 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3475, pruned_loss=0.1114, over 5701832.91 frames. ], batch size: 94, lr: 9.92e-03, grad_scale: 2.0
+2023-03-01 17:00:59,778 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:02,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:18,456 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:20,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:24,519 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:27,146 INFO [train.py:968] (0/2) Epoch 3, batch 19950, giga_loss[loss=0.2788, simple_loss=0.342, pruned_loss=0.1079, over 28677.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3526, pruned_loss=0.1133, over 5721305.91 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3891, pruned_loss=0.13, over 5753880.89 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3458, pruned_loss=0.1102, over 5710739.06 frames. ], batch size: 242, lr: 9.92e-03, grad_scale: 2.0
+2023-03-01 17:01:44,682 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:50,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.474e+02 8.930e+02 1.143e+03 1.544e+03 4.415e+03, threshold=2.286e+03, percent-clipped=2.0
+2023-03-01 17:02:08,859 INFO [train.py:968] (0/2) Epoch 3, batch 20000, giga_loss[loss=0.2548, simple_loss=0.3229, pruned_loss=0.09331, over 28989.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3521, pruned_loss=0.1123, over 5720061.19 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3902, pruned_loss=0.1302, over 5758510.98 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3439, pruned_loss=0.1089, over 5705872.42 frames. ], batch size: 128, lr: 9.91e-03, grad_scale: 4.0
+2023-03-01 17:02:18,006 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-01 17:02:47,231 INFO [train.py:968] (0/2) Epoch 3, batch 20050, giga_loss[loss=0.2641, simple_loss=0.3243, pruned_loss=0.1019, over 28851.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3502, pruned_loss=0.1109, over 5723353.49 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3905, pruned_loss=0.13, over 5761064.25 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3425, pruned_loss=0.1078, over 5708703.25 frames. ], batch size: 92, lr: 9.91e-03, grad_scale: 4.0
+2023-03-01 17:02:51,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:03:01,241 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-01 17:03:07,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.940e+02 1.026e+03 1.276e+03 1.774e+03 4.540e+03, threshold=2.552e+03, percent-clipped=9.0
+2023-03-01 17:03:11,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6485, 4.1773, 4.3179, 1.5590], device='cuda:0'), covar=tensor([0.0423, 0.0444, 0.0953, 0.2174], device='cuda:0'), in_proj_covar=tensor([0.0766, 0.0592, 0.0776, 0.0567], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 17:03:22,515 INFO [train.py:968] (0/2) Epoch 3, batch 20100, giga_loss[loss=0.2816, simple_loss=0.3326, pruned_loss=0.1153, over 28616.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3524, pruned_loss=0.1119, over 5733398.29 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3915, pruned_loss=0.1304, over 5766650.75 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.343, pruned_loss=0.108, over 5714490.83 frames. ], batch size: 85, lr: 9.91e-03, grad_scale: 4.0
+2023-03-01 17:03:42,129 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:03:53,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:04:01,465 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:04:03,916 INFO [train.py:968] (0/2) Epoch 3, batch 20150, libri_loss[loss=0.3668, simple_loss=0.4204, pruned_loss=0.1566, over 19938.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3578, pruned_loss=0.116, over 5721151.72 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3919, pruned_loss=0.1306, over 5760733.07 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3488, pruned_loss=0.1123, over 5710957.39 frames. ], batch size: 187, lr: 9.91e-03, grad_scale: 4.0
+2023-03-01 17:04:29,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.070e+02 1.114e+03 1.364e+03 1.839e+03 6.169e+03, threshold=2.727e+03, percent-clipped=14.0
+2023-03-01 17:04:51,945 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:04:52,303 INFO [train.py:968] (0/2) Epoch 3, batch 20200, giga_loss[loss=0.3068, simple_loss=0.3735, pruned_loss=0.1201, over 28919.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3633, pruned_loss=0.1204, over 5707748.54 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3914, pruned_loss=0.1303, over 5762715.25 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3563, pruned_loss=0.1176, over 5697402.79 frames. ], batch size: 186, lr: 9.91e-03, grad_scale: 4.0
+2023-03-01 17:04:53,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:05:02,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:05:02,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5237, 2.0718, 1.6255, 1.6231], device='cuda:0'), covar=tensor([0.0515, 0.0665, 0.0880, 0.0947], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0470, 0.0527, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 17:05:07,574 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:05:20,647 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:05:35,246 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-01 17:05:38,065 INFO [train.py:968] (0/2) Epoch 3, batch 20250, giga_loss[loss=0.3269, simple_loss=0.3875, pruned_loss=0.1331, over 28839.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3713, pruned_loss=0.1258, over 5709315.49 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3917, pruned_loss=0.1302, over 5765801.08 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3645, pruned_loss=0.1233, over 5696093.82 frames. ], batch size: 186, lr: 9.90e-03, grad_scale: 4.0
+2023-03-01 17:06:06,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.227e+02 1.323e+03 1.619e+03 2.148e+03 5.561e+03, threshold=3.238e+03, percent-clipped=13.0
+2023-03-01 17:06:06,784 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:06:08,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:06:25,239 INFO [train.py:968] (0/2) Epoch 3, batch 20300, giga_loss[loss=0.3705, simple_loss=0.4235, pruned_loss=0.1588, over 28650.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3788, pruned_loss=0.1309, over 5696943.97 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3923, pruned_loss=0.1306, over 5761977.98 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3724, pruned_loss=0.1286, over 5688208.09 frames. ], batch size: 307, lr: 9.90e-03, grad_scale: 4.0
+2023-03-01 17:06:33,320 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:07:10,677 INFO [train.py:968] (0/2) Epoch 3, batch 20350, giga_loss[loss=0.3281, simple_loss=0.3921, pruned_loss=0.132, over 28646.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3847, pruned_loss=0.134, over 5690888.64 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3923, pruned_loss=0.1308, over 5755237.32 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3795, pruned_loss=0.1319, over 5687930.96 frames. ], batch size: 242, lr: 9.90e-03, grad_scale: 4.0
+2023-03-01 17:07:36,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.613e+02 1.256e+03 1.530e+03 2.189e+03 7.355e+03, threshold=3.059e+03, percent-clipped=8.0
+2023-03-01 17:07:44,225 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:07:57,195 INFO [train.py:968] (0/2) Epoch 3, batch 20400, giga_loss[loss=0.3801, simple_loss=0.4091, pruned_loss=0.1756, over 23439.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3896, pruned_loss=0.1368, over 5685910.62 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.392, pruned_loss=0.1308, over 5757999.58 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3856, pruned_loss=0.1353, over 5679751.81 frames. ], batch size: 705, lr: 9.90e-03, grad_scale: 8.0
+2023-03-01 17:08:00,615 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:08:15,922 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:08:40,139 INFO [train.py:968] (0/2) Epoch 3, batch 20450, giga_loss[loss=0.3511, simple_loss=0.391, pruned_loss=0.1556, over 27522.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3935, pruned_loss=0.1393, over 5694681.93 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3926, pruned_loss=0.1313, over 5760231.62 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3898, pruned_loss=0.1377, over 5686936.85 frames. ], batch size: 472, lr: 9.89e-03, grad_scale: 8.0
+2023-03-01 17:08:48,637 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2491, 3.8494, 3.9321, 1.7673], device='cuda:0'), covar=tensor([0.0511, 0.0389, 0.0746, 0.1982], device='cuda:0'), in_proj_covar=tensor([0.0756, 0.0599, 0.0775, 0.0565], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 17:09:05,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.380e+02 1.252e+03 1.534e+03 2.149e+03 4.631e+03, threshold=3.068e+03, percent-clipped=6.0
+2023-03-01 17:09:20,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:09:24,991 INFO [train.py:968] (0/2) Epoch 3, batch 20500, giga_loss[loss=0.2359, simple_loss=0.3135, pruned_loss=0.07917, over 28686.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3845, pruned_loss=0.1328, over 5694432.06 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3925, pruned_loss=0.1314, over 5762451.24 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3816, pruned_loss=0.1315, over 5685312.82 frames. ], batch size: 284, lr: 9.89e-03, grad_scale: 8.0
+2023-03-01 17:09:37,702 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.08 vs. limit=2.0
+2023-03-01 17:09:42,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:10:04,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4327, 2.1967, 1.4950, 0.5732], device='cuda:0'), covar=tensor([0.2104, 0.0908, 0.1782, 0.2307], device='cuda:0'), in_proj_covar=tensor([0.1226, 0.1154, 0.1265, 0.1057], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 17:10:05,174 INFO [train.py:968] (0/2) Epoch 3, batch 20550, giga_loss[loss=0.2872, simple_loss=0.36, pruned_loss=0.1073, over 29001.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3812, pruned_loss=0.1296, over 5695726.64 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3925, pruned_loss=0.1315, over 5756462.03 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3786, pruned_loss=0.1284, over 5691844.39 frames. ], batch size: 128, lr: 9.89e-03, grad_scale: 8.0
+2023-03-01 17:10:11,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5785, 1.4821, 1.1648, 1.2835], device='cuda:0'), covar=tensor([0.0642, 0.0562, 0.0967, 0.0770], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0471, 0.0525, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 17:10:31,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.741e+02 1.091e+03 1.286e+03 1.774e+03 3.690e+03, threshold=2.573e+03, percent-clipped=2.0
+2023-03-01 17:10:38,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:10:42,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:10:48,829 INFO [train.py:968] (0/2) Epoch 3, batch 20600, giga_loss[loss=0.3203, simple_loss=0.3803, pruned_loss=0.1302, over 28826.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3821, pruned_loss=0.13, over 5674361.08 frames. ], libri_tot_loss[loss=0.3291, simple_loss=0.3934, pruned_loss=0.1324, over 5741481.01 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3789, pruned_loss=0.1282, over 5684323.95 frames. ], batch size: 199, lr: 9.89e-03, grad_scale: 2.0
+2023-03-01 17:11:19,025 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:11:21,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:11:22,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0115, 1.1029, 0.6770, 0.4878], device='cuda:0'), covar=tensor([0.0526, 0.0504, 0.0569, 0.0630], device='cuda:0'), in_proj_covar=tensor([0.1176, 0.0912, 0.0947, 0.1024], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 17:11:28,660 INFO [train.py:968] (0/2) Epoch 3, batch 20650, giga_loss[loss=0.4498, simple_loss=0.4533, pruned_loss=0.2231, over 26594.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3842, pruned_loss=0.1311, over 5677862.18 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3937, pruned_loss=0.1329, over 5737665.80 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.381, pruned_loss=0.1291, over 5687237.02 frames. ], batch size: 555, lr: 9.89e-03, grad_scale: 2.0
+2023-03-01 17:11:39,968 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:11:42,011 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:11:43,340 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:11:48,234 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-01 17:11:55,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.090e+02 1.205e+03 1.733e+03 2.845e+03 1.635e+04, threshold=3.467e+03, percent-clipped=27.0
+2023-03-01 17:12:09,308 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:12:11,957 INFO [train.py:968] (0/2) Epoch 3, batch 20700, giga_loss[loss=0.3364, simple_loss=0.3942, pruned_loss=0.1392, over 28892.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3865, pruned_loss=0.1328, over 5683052.12 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3934, pruned_loss=0.1328, over 5740344.91 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3841, pruned_loss=0.1313, over 5687369.24 frames. ], batch size: 227, lr: 9.88e-03, grad_scale: 2.0
+2023-03-01 17:12:22,761 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:12:31,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4476, 1.4453, 1.4302, 1.4372], device='cuda:0'), covar=tensor([0.0915, 0.1200, 0.1458, 0.1031], device='cuda:0'), in_proj_covar=tensor([0.0452, 0.0774, 0.0633, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 17:12:38,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:12:41,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:12:43,300 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:12:45,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:12:53,392 INFO [train.py:968] (0/2) Epoch 3, batch 20750, libri_loss[loss=0.3904, simple_loss=0.4346, pruned_loss=0.1731, over 29539.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3884, pruned_loss=0.1344, over 5684237.59 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.394, pruned_loss=0.1332, over 5735767.64 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3857, pruned_loss=0.1328, over 5689140.13 frames. ], batch size: 80, lr: 9.88e-03, grad_scale: 2.0
+2023-03-01 17:13:02,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:13:03,849 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:13:09,087 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:13:19,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:13:20,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.045e+02 1.246e+03 1.571e+03 2.115e+03 4.811e+03, threshold=3.142e+03, percent-clipped=3.0
+2023-03-01 17:13:22,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9197, 2.2125, 2.0860, 1.9318], device='cuda:0'), covar=tensor([0.1526, 0.1597, 0.1073, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0791, 0.0741, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 17:13:29,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1722, 2.8980, 2.9402, 1.4074], device='cuda:0'), covar=tensor([0.0711, 0.0576, 0.0933, 0.2185], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0588, 0.0760, 0.0559], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 17:13:33,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:13:37,742 INFO [train.py:968] (0/2) Epoch 3, batch 20800, giga_loss[loss=0.3714, simple_loss=0.4266, pruned_loss=0.1582, over 28602.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3892, pruned_loss=0.1347, over 5700397.70 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3941, pruned_loss=0.1333, over 5739295.66 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3868, pruned_loss=0.1335, over 5700005.71 frames. ], batch size: 71, lr: 9.88e-03, grad_scale: 4.0
+2023-03-01 17:14:09,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:14:19,314 INFO [train.py:968] (0/2) Epoch 3, batch 20850, giga_loss[loss=0.354, simple_loss=0.4044, pruned_loss=0.1517, over 28673.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3902, pruned_loss=0.1357, over 5703089.86 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3943, pruned_loss=0.1334, over 5744267.47 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3879, pruned_loss=0.1346, over 5696680.95 frames. ], batch size: 242, lr: 9.88e-03, grad_scale: 4.0
+2023-03-01 17:14:34,250 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-01 17:14:43,829 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.775e+02 1.211e+03 1.404e+03 1.942e+03 3.796e+03, threshold=2.807e+03, percent-clipped=3.0
+2023-03-01 17:14:59,012 INFO [train.py:968] (0/2) Epoch 3, batch 20900, giga_loss[loss=0.3202, simple_loss=0.3843, pruned_loss=0.1281, over 28792.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3899, pruned_loss=0.1351, over 5709560.38 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.3941, pruned_loss=0.1334, over 5745860.13 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3882, pruned_loss=0.1343, over 5702822.45 frames. ], batch size: 145, lr: 9.87e-03, grad_scale: 4.0
+2023-03-01 17:15:01,310 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:03,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:16,753 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:18,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:26,411 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:28,178 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:30,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:38,104 INFO [train.py:968] (0/2) Epoch 3, batch 20950, giga_loss[loss=0.2618, simple_loss=0.3463, pruned_loss=0.08859, over 28397.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.39, pruned_loss=0.1342, over 5709674.47 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3946, pruned_loss=0.134, over 5749077.97 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3881, pruned_loss=0.133, over 5700809.58 frames. ], batch size: 60, lr: 9.87e-03, grad_scale: 4.0
+2023-03-01 17:15:42,123 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:56,364 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:56,381 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:16:04,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.054e+02 1.040e+03 1.305e+03 1.728e+03 5.069e+03, threshold=2.610e+03, percent-clipped=8.0
+2023-03-01 17:16:15,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7604, 2.1586, 1.2449, 1.1555], device='cuda:0'), covar=tensor([0.0831, 0.0587, 0.0643, 0.0776], device='cuda:0'), in_proj_covar=tensor([0.1164, 0.0893, 0.0947, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 17:16:18,818 INFO [train.py:968] (0/2) Epoch 3, batch 21000, giga_loss[loss=0.348, simple_loss=0.4076, pruned_loss=0.1442, over 28623.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3909, pruned_loss=0.1336, over 5716866.38 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3948, pruned_loss=0.1344, over 5751469.01 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3892, pruned_loss=0.1323, over 5707164.05 frames. ], batch size: 336, lr: 9.87e-03, grad_scale: 4.0
+2023-03-01 17:16:18,823 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 17:16:23,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6600, 1.1546, 3.4967, 3.0302], device='cuda:0'), covar=tensor([0.1805, 0.2215, 0.0392, 0.0582], device='cuda:0'), in_proj_covar=tensor([0.0527, 0.0496, 0.0675, 0.0545], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 17:16:26,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2042, 1.2541, 0.8965, 1.2576], device='cuda:0'), covar=tensor([0.0898, 0.0348, 0.0434, 0.1025], device='cuda:0'), in_proj_covar=tensor([0.0231, 0.0151, 0.0155, 0.0278], device='cuda:0'), out_proj_covar=tensor([0.0040, 0.0030, 0.0027, 0.0045], device='cuda:0')
+2023-03-01 17:16:27,499 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2638, simple_loss=0.3621, pruned_loss=0.0828, over 944034.00 frames. 
+2023-03-01 17:16:27,500 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 17:17:06,266 INFO [train.py:968] (0/2) Epoch 3, batch 21050, giga_loss[loss=0.3107, simple_loss=0.3758, pruned_loss=0.1228, over 29000.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3898, pruned_loss=0.1323, over 5722997.81 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3954, pruned_loss=0.1348, over 5754662.61 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3878, pruned_loss=0.1309, over 5711828.22 frames. ], batch size: 164, lr: 9.87e-03, grad_scale: 4.0
+2023-03-01 17:17:13,921 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:17:30,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.356e+02 1.062e+03 1.363e+03 1.863e+03 1.199e+04, threshold=2.725e+03, percent-clipped=12.0
+2023-03-01 17:17:36,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:17:44,657 INFO [train.py:968] (0/2) Epoch 3, batch 21100, giga_loss[loss=0.2973, simple_loss=0.3642, pruned_loss=0.1151, over 28327.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3869, pruned_loss=0.1309, over 5710750.48 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3956, pruned_loss=0.1352, over 5748253.06 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3848, pruned_loss=0.1293, over 5706821.61 frames. ], batch size: 65, lr: 9.87e-03, grad_scale: 4.0
+2023-03-01 17:17:50,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7703, 3.4375, 3.5031, 1.6535], device='cuda:0'), covar=tensor([0.0566, 0.0424, 0.0896, 0.2061], device='cuda:0'), in_proj_covar=tensor([0.0752, 0.0597, 0.0771, 0.0561], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 17:18:23,443 INFO [train.py:968] (0/2) Epoch 3, batch 21150, giga_loss[loss=0.2929, simple_loss=0.3683, pruned_loss=0.1088, over 29037.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3854, pruned_loss=0.1303, over 5711294.70 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.396, pruned_loss=0.1356, over 5746983.42 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3834, pruned_loss=0.1286, over 5708749.90 frames. ], batch size: 155, lr: 9.86e-03, grad_scale: 4.0
+2023-03-01 17:18:48,613 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.287e+02 1.033e+03 1.296e+03 1.683e+03 3.697e+03, threshold=2.593e+03, percent-clipped=2.0
+2023-03-01 17:18:57,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4179, 1.4918, 1.2394, 1.6544], device='cuda:0'), covar=tensor([0.2166, 0.1907, 0.1803, 0.2323], device='cuda:0'), in_proj_covar=tensor([0.1014, 0.0819, 0.0920, 0.0940], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 17:19:03,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1742, 1.4196, 1.2218, 1.3428], device='cuda:0'), covar=tensor([0.2200, 0.1855, 0.1828, 0.1937], device='cuda:0'), in_proj_covar=tensor([0.1013, 0.0817, 0.0919, 0.0939], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 17:19:04,088 INFO [train.py:968] (0/2) Epoch 3, batch 21200, giga_loss[loss=0.3239, simple_loss=0.3887, pruned_loss=0.1295, over 28848.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3842, pruned_loss=0.1297, over 5716674.95 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3963, pruned_loss=0.1358, over 5747998.89 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3822, pruned_loss=0.1281, over 5713252.91 frames. ], batch size: 99, lr: 9.86e-03, grad_scale: 8.0
+2023-03-01 17:19:16,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:19:32,613 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:19:34,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:19:46,309 INFO [train.py:968] (0/2) Epoch 3, batch 21250, giga_loss[loss=0.2794, simple_loss=0.3469, pruned_loss=0.106, over 28425.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3844, pruned_loss=0.13, over 5706558.63 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3964, pruned_loss=0.1361, over 5740515.01 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3825, pruned_loss=0.1284, over 5710732.82 frames. ], batch size: 65, lr: 9.86e-03, grad_scale: 8.0
+2023-03-01 17:19:57,633 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:20:03,169 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 17:20:05,681 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6930, 1.9915, 1.8644, 1.7180], device='cuda:0'), covar=tensor([0.1448, 0.1656, 0.1128, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0795, 0.0741, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 17:20:10,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.465e+02 1.011e+03 1.246e+03 1.602e+03 3.381e+03, threshold=2.491e+03, percent-clipped=3.0
+2023-03-01 17:20:26,106 INFO [train.py:968] (0/2) Epoch 3, batch 21300, libri_loss[loss=0.3879, simple_loss=0.4372, pruned_loss=0.1693, over 29755.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3858, pruned_loss=0.1308, over 5700099.34 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3969, pruned_loss=0.1366, over 5736750.71 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3834, pruned_loss=0.1288, over 5705779.35 frames. ], batch size: 87, lr: 9.86e-03, grad_scale: 8.0
+2023-03-01 17:20:47,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:21:02,913 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:21:05,835 INFO [train.py:968] (0/2) Epoch 3, batch 21350, giga_loss[loss=0.2997, simple_loss=0.3707, pruned_loss=0.1144, over 28565.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.386, pruned_loss=0.1307, over 5710693.80 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3973, pruned_loss=0.1375, over 5741360.99 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3833, pruned_loss=0.1281, over 5709839.09 frames. ], batch size: 336, lr: 9.85e-03, grad_scale: 8.0
+2023-03-01 17:21:09,935 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:21:11,986 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:21:30,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.028e+02 1.097e+03 1.371e+03 1.853e+03 6.754e+03, threshold=2.741e+03, percent-clipped=13.0
+2023-03-01 17:21:36,967 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:21:40,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1838, 1.2635, 1.0515, 1.1563], device='cuda:0'), covar=tensor([0.0618, 0.0499, 0.1012, 0.0662], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0466, 0.0521, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 17:21:46,752 INFO [train.py:968] (0/2) Epoch 3, batch 21400, giga_loss[loss=0.3319, simple_loss=0.3645, pruned_loss=0.1496, over 23776.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3847, pruned_loss=0.1297, over 5692848.79 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3971, pruned_loss=0.1375, over 5733649.91 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3825, pruned_loss=0.1274, over 5699150.31 frames. ], batch size: 705, lr: 9.85e-03, grad_scale: 8.0
+2023-03-01 17:22:11,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:22:22,280 INFO [train.py:968] (0/2) Epoch 3, batch 21450, giga_loss[loss=0.325, simple_loss=0.3824, pruned_loss=0.1339, over 28414.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3841, pruned_loss=0.1298, over 5699963.70 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3975, pruned_loss=0.138, over 5736252.23 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3815, pruned_loss=0.1272, over 5701775.47 frames. ], batch size: 71, lr: 9.85e-03, grad_scale: 4.0
+2023-03-01 17:22:45,467 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-01 17:22:48,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.422e+02 1.144e+03 1.369e+03 2.033e+03 6.792e+03, threshold=2.738e+03, percent-clipped=8.0
+2023-03-01 17:22:54,173 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:22:55,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:23:01,601 INFO [train.py:968] (0/2) Epoch 3, batch 21500, giga_loss[loss=0.326, simple_loss=0.3809, pruned_loss=0.1356, over 28618.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3826, pruned_loss=0.1294, over 5695388.36 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.3981, pruned_loss=0.1388, over 5733579.01 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3794, pruned_loss=0.1263, over 5697786.32 frames. ], batch size: 85, lr: 9.85e-03, grad_scale: 4.0
+2023-03-01 17:23:18,437 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:23:18,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3887, 1.7012, 1.0159, 0.8758], device='cuda:0'), covar=tensor([0.0859, 0.0730, 0.0660, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.1175, 0.0906, 0.0956, 0.1021], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 17:23:40,844 INFO [train.py:968] (0/2) Epoch 3, batch 21550, giga_loss[loss=0.2821, simple_loss=0.3491, pruned_loss=0.1076, over 28655.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3793, pruned_loss=0.1278, over 5698952.26 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.3985, pruned_loss=0.1393, over 5737825.96 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3761, pruned_loss=0.1246, over 5696320.01 frames. ], batch size: 71, lr: 9.85e-03, grad_scale: 4.0
+2023-03-01 17:23:56,652 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:24:02,955 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:24:05,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:24:05,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.395e+02 1.163e+03 1.547e+03 2.274e+03 7.813e+03, threshold=3.094e+03, percent-clipped=16.0
+2023-03-01 17:24:21,112 INFO [train.py:968] (0/2) Epoch 3, batch 21600, giga_loss[loss=0.4047, simple_loss=0.4312, pruned_loss=0.1891, over 28620.00 frames. ], tot_loss[loss=0.318, simple_loss=0.379, pruned_loss=0.1285, over 5694451.26 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3992, pruned_loss=0.1401, over 5738823.07 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3754, pruned_loss=0.1249, over 5690835.35 frames. ], batch size: 85, lr: 9.84e-03, grad_scale: 8.0
+2023-03-01 17:24:29,249 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:25:00,668 INFO [train.py:968] (0/2) Epoch 3, batch 21650, giga_loss[loss=0.3171, simple_loss=0.3788, pruned_loss=0.1277, over 28827.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3795, pruned_loss=0.1296, over 5701059.65 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3993, pruned_loss=0.1403, over 5742816.08 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.376, pruned_loss=0.1263, over 5693464.26 frames. ], batch size: 199, lr: 9.84e-03, grad_scale: 8.0
+2023-03-01 17:25:25,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.112e+02 1.289e+03 1.551e+03 1.944e+03 6.047e+03, threshold=3.102e+03, percent-clipped=14.0
+2023-03-01 17:25:37,943 INFO [train.py:968] (0/2) Epoch 3, batch 21700, giga_loss[loss=0.3046, simple_loss=0.369, pruned_loss=0.1201, over 28887.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.378, pruned_loss=0.1294, over 5698272.18 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.3999, pruned_loss=0.1409, over 5741420.44 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3739, pruned_loss=0.1258, over 5691059.57 frames. ], batch size: 174, lr: 9.84e-03, grad_scale: 8.0
+2023-03-01 17:25:39,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:25:44,792 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-112000.pt
+2023-03-01 17:26:13,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0850, 1.1961, 1.0872, 0.9026], device='cuda:0'), covar=tensor([0.1836, 0.1945, 0.1766, 0.1811], device='cuda:0'), in_proj_covar=tensor([0.1025, 0.0823, 0.0921, 0.0937], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 17:26:16,215 INFO [train.py:968] (0/2) Epoch 3, batch 21750, giga_loss[loss=0.3086, simple_loss=0.3621, pruned_loss=0.1275, over 28556.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3758, pruned_loss=0.1286, over 5691445.87 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3999, pruned_loss=0.1414, over 5728160.61 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3717, pruned_loss=0.1249, over 5696282.71 frames. ], batch size: 71, lr: 9.84e-03, grad_scale: 8.0
+2023-03-01 17:26:41,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.871e+02 1.071e+03 1.352e+03 1.948e+03 4.725e+03, threshold=2.705e+03, percent-clipped=8.0
+2023-03-01 17:26:51,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2401, 1.3286, 1.2153, 1.1121], device='cuda:0'), covar=tensor([0.1903, 0.1909, 0.1727, 0.1988], device='cuda:0'), in_proj_covar=tensor([0.1021, 0.0812, 0.0915, 0.0932], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 17:26:56,059 INFO [train.py:968] (0/2) Epoch 3, batch 21800, giga_loss[loss=0.3085, simple_loss=0.3581, pruned_loss=0.1295, over 28929.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3725, pruned_loss=0.1267, over 5700338.96 frames. ], libri_tot_loss[loss=0.3414, simple_loss=0.3999, pruned_loss=0.1415, over 5728822.08 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.369, pruned_loss=0.1236, over 5703213.82 frames. ], batch size: 106, lr: 9.84e-03, grad_scale: 8.0
+2023-03-01 17:27:31,099 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:27:33,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:27:35,268 INFO [train.py:968] (0/2) Epoch 3, batch 21850, giga_loss[loss=0.2695, simple_loss=0.3402, pruned_loss=0.09937, over 28855.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3702, pruned_loss=0.1253, over 5711386.25 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.4005, pruned_loss=0.1422, over 5732389.76 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3659, pruned_loss=0.1216, over 5709683.72 frames. ], batch size: 199, lr: 9.83e-03, grad_scale: 8.0
+2023-03-01 17:27:58,681 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:28:02,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.867e+02 1.005e+03 1.194e+03 1.650e+03 3.847e+03, threshold=2.387e+03, percent-clipped=5.0
+2023-03-01 17:28:11,969 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112184.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 17:28:16,369 INFO [train.py:968] (0/2) Epoch 3, batch 21900, giga_loss[loss=0.3312, simple_loss=0.3869, pruned_loss=0.1378, over 29044.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3728, pruned_loss=0.1267, over 5709127.78 frames. ], libri_tot_loss[loss=0.3435, simple_loss=0.4013, pruned_loss=0.1429, over 5733767.96 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3683, pruned_loss=0.1229, over 5706088.11 frames. ], batch size: 155, lr: 9.83e-03, grad_scale: 8.0
+2023-03-01 17:28:44,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6990, 1.5552, 1.5546, 1.5701], device='cuda:0'), covar=tensor([0.1605, 0.2529, 0.1482, 0.1162], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0788, 0.0732, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 17:28:56,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:29:01,916 INFO [train.py:968] (0/2) Epoch 3, batch 21950, giga_loss[loss=0.3435, simple_loss=0.3967, pruned_loss=0.1451, over 27605.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3763, pruned_loss=0.1285, over 5701860.42 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.4015, pruned_loss=0.1431, over 5735279.33 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3723, pruned_loss=0.1251, over 5697846.82 frames. ], batch size: 472, lr: 9.83e-03, grad_scale: 8.0
+2023-03-01 17:29:28,728 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.973e+02 1.153e+03 1.423e+03 1.992e+03 7.783e+03, threshold=2.845e+03, percent-clipped=15.0
+2023-03-01 17:29:42,325 INFO [train.py:968] (0/2) Epoch 3, batch 22000, giga_loss[loss=0.2817, simple_loss=0.3587, pruned_loss=0.1023, over 28942.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3799, pruned_loss=0.1303, over 5698350.95 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.402, pruned_loss=0.144, over 5738272.16 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3754, pruned_loss=0.1262, over 5691258.98 frames. ], batch size: 145, lr: 9.83e-03, grad_scale: 8.0
+2023-03-01 17:29:57,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0398, 3.6458, 3.7522, 1.6734], device='cuda:0'), covar=tensor([0.0417, 0.0349, 0.0681, 0.2182], device='cuda:0'), in_proj_covar=tensor([0.0753, 0.0591, 0.0766, 0.0554], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 17:30:20,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2865, 1.5927, 1.1913, 1.4298], device='cuda:0'), covar=tensor([0.0771, 0.0479, 0.0409, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0228, 0.0151, 0.0156, 0.0276], device='cuda:0'), out_proj_covar=tensor([0.0040, 0.0030, 0.0027, 0.0045], device='cuda:0')
+2023-03-01 17:30:23,663 INFO [train.py:968] (0/2) Epoch 3, batch 22050, giga_loss[loss=0.2663, simple_loss=0.3414, pruned_loss=0.09566, over 29021.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3801, pruned_loss=0.1294, over 5698171.39 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.4025, pruned_loss=0.1446, over 5731717.03 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3756, pruned_loss=0.1253, over 5698539.11 frames. ], batch size: 128, lr: 9.82e-03, grad_scale: 8.0
+2023-03-01 17:30:50,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.014e+02 1.062e+03 1.312e+03 1.779e+03 7.812e+03, threshold=2.625e+03, percent-clipped=11.0
+2023-03-01 17:30:54,223 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:30:56,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:31:03,652 INFO [train.py:968] (0/2) Epoch 3, batch 22100, libri_loss[loss=0.3649, simple_loss=0.4244, pruned_loss=0.1527, over 29291.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3804, pruned_loss=0.1291, over 5707185.82 frames. ], libri_tot_loss[loss=0.3474, simple_loss=0.4034, pruned_loss=0.1457, over 5737990.54 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3751, pruned_loss=0.1243, over 5700672.36 frames. ], batch size: 94, lr: 9.82e-03, grad_scale: 4.0
+2023-03-01 17:31:20,872 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:31:29,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5694, 1.5589, 1.5517, 1.5146], device='cuda:0'), covar=tensor([0.0861, 0.1368, 0.1365, 0.1172], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0758, 0.0617, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 17:31:31,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5130, 1.3950, 1.1447, 1.2065], device='cuda:0'), covar=tensor([0.0521, 0.0524, 0.0827, 0.0975], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0474, 0.0519, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 17:31:33,170 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 17:31:43,291 INFO [train.py:968] (0/2) Epoch 3, batch 22150, giga_loss[loss=0.3557, simple_loss=0.4048, pruned_loss=0.1533, over 28955.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3798, pruned_loss=0.129, over 5703606.25 frames. ], libri_tot_loss[loss=0.3479, simple_loss=0.4035, pruned_loss=0.1462, over 5739047.77 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3748, pruned_loss=0.1242, over 5696680.99 frames. ], batch size: 136, lr: 9.82e-03, grad_scale: 4.0
+2023-03-01 17:31:46,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4795, 1.4694, 1.5040, 1.5016], device='cuda:0'), covar=tensor([0.0770, 0.1447, 0.1184, 0.1132], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0753, 0.0613, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 17:31:57,127 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:32:09,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.788e+02 1.194e+03 1.531e+03 2.249e+03 6.593e+03, threshold=3.061e+03, percent-clipped=11.0
+2023-03-01 17:32:23,631 INFO [train.py:968] (0/2) Epoch 3, batch 22200, giga_loss[loss=0.2748, simple_loss=0.3481, pruned_loss=0.1007, over 29066.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3814, pruned_loss=0.1305, over 5697256.16 frames. ], libri_tot_loss[loss=0.3487, simple_loss=0.4041, pruned_loss=0.1466, over 5732757.38 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3765, pruned_loss=0.1261, over 5696107.83 frames. ], batch size: 136, lr: 9.82e-03, grad_scale: 4.0
+2023-03-01 17:32:40,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4026, 1.3299, 1.2560, 1.5743], device='cuda:0'), covar=tensor([0.1967, 0.2053, 0.1884, 0.2138], device='cuda:0'), in_proj_covar=tensor([0.1017, 0.0812, 0.0914, 0.0926], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 17:32:58,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2941, 1.3823, 1.0057, 0.7849], device='cuda:0'), covar=tensor([0.0764, 0.0579, 0.0474, 0.0633], device='cuda:0'), in_proj_covar=tensor([0.1187, 0.0915, 0.0974, 0.1024], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 17:33:04,363 INFO [train.py:968] (0/2) Epoch 3, batch 22250, giga_loss[loss=0.3487, simple_loss=0.4023, pruned_loss=0.1475, over 28914.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3816, pruned_loss=0.1306, over 5694219.88 frames. ], libri_tot_loss[loss=0.3486, simple_loss=0.404, pruned_loss=0.1466, over 5732155.43 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3776, pruned_loss=0.1269, over 5693462.98 frames. ], batch size: 174, lr: 9.82e-03, grad_scale: 4.0
+2023-03-01 17:33:19,115 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112559.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 17:33:30,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.931e+02 1.109e+03 1.415e+03 1.844e+03 4.822e+03, threshold=2.830e+03, percent-clipped=2.0
+2023-03-01 17:33:30,722 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-01 17:33:44,765 INFO [train.py:968] (0/2) Epoch 3, batch 22300, giga_loss[loss=0.2981, simple_loss=0.3733, pruned_loss=0.1114, over 28923.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3844, pruned_loss=0.1319, over 5702359.71 frames. ], libri_tot_loss[loss=0.349, simple_loss=0.4043, pruned_loss=0.1468, over 5732048.55 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3808, pruned_loss=0.1287, over 5701652.40 frames. ], batch size: 145, lr: 9.81e-03, grad_scale: 4.0
+2023-03-01 17:34:00,484 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 17:34:24,219 INFO [train.py:968] (0/2) Epoch 3, batch 22350, giga_loss[loss=0.3519, simple_loss=0.4087, pruned_loss=0.1475, over 29032.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.387, pruned_loss=0.1332, over 5706264.58 frames. ], libri_tot_loss[loss=0.3497, simple_loss=0.4048, pruned_loss=0.1473, over 5734473.70 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3834, pruned_loss=0.13, over 5703038.93 frames. ], batch size: 128, lr: 9.81e-03, grad_scale: 4.0
+2023-03-01 17:34:40,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3361, 1.3366, 1.2477, 1.5360], device='cuda:0'), covar=tensor([0.1938, 0.1961, 0.1786, 0.2047], device='cuda:0'), in_proj_covar=tensor([0.1006, 0.0804, 0.0899, 0.0916], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 17:34:52,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.594e+02 1.283e+03 1.711e+03 2.241e+03 9.336e+03, threshold=3.423e+03, percent-clipped=15.0
+2023-03-01 17:35:07,893 INFO [train.py:968] (0/2) Epoch 3, batch 22400, giga_loss[loss=0.2823, simple_loss=0.3575, pruned_loss=0.1035, over 28957.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3872, pruned_loss=0.1329, over 5712619.68 frames. ], libri_tot_loss[loss=0.3501, simple_loss=0.4052, pruned_loss=0.1475, over 5735329.33 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.384, pruned_loss=0.1301, over 5709239.53 frames. ], batch size: 136, lr: 9.81e-03, grad_scale: 8.0
+2023-03-01 17:35:18,820 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112702.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 17:35:20,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112705.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 17:35:37,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8351, 1.5575, 1.3284, 1.2898], device='cuda:0'), covar=tensor([0.0577, 0.0625, 0.0902, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0479, 0.0530, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 17:35:43,974 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112734.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 17:35:49,594 INFO [train.py:968] (0/2) Epoch 3, batch 22450, giga_loss[loss=0.3388, simple_loss=0.4017, pruned_loss=0.138, over 28817.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.386, pruned_loss=0.1316, over 5721206.84 frames. ], libri_tot_loss[loss=0.3511, simple_loss=0.4059, pruned_loss=0.1481, over 5737400.81 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3826, pruned_loss=0.1287, over 5716511.65 frames. ], batch size: 199, lr: 9.81e-03, grad_scale: 4.0
+2023-03-01 17:36:10,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3417, 1.3587, 1.2859, 1.4901], device='cuda:0'), covar=tensor([0.1868, 0.1818, 0.1691, 0.2023], device='cuda:0'), in_proj_covar=tensor([0.1006, 0.0797, 0.0903, 0.0924], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 17:36:21,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.226e+02 1.068e+03 1.322e+03 1.796e+03 6.337e+03, threshold=2.644e+03, percent-clipped=3.0
+2023-03-01 17:36:26,222 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:36:33,556 INFO [train.py:968] (0/2) Epoch 3, batch 22500, libri_loss[loss=0.3769, simple_loss=0.4107, pruned_loss=0.1715, over 29505.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3872, pruned_loss=0.1327, over 5715972.07 frames. ], libri_tot_loss[loss=0.3518, simple_loss=0.4065, pruned_loss=0.1486, over 5740038.68 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3837, pruned_loss=0.1297, over 5709370.08 frames. ], batch size: 70, lr: 9.80e-03, grad_scale: 4.0
+2023-03-01 17:37:01,080 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-03-01 17:37:08,978 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:37:14,576 INFO [train.py:968] (0/2) Epoch 3, batch 22550, giga_loss[loss=0.3, simple_loss=0.3639, pruned_loss=0.118, over 28774.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3867, pruned_loss=0.1325, over 5713528.10 frames. ], libri_tot_loss[loss=0.3529, simple_loss=0.4072, pruned_loss=0.1493, over 5737694.76 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3827, pruned_loss=0.1291, over 5709783.94 frames. ], batch size: 92, lr: 9.80e-03, grad_scale: 4.0
+2023-03-01 17:37:43,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.861e+02 1.218e+03 1.498e+03 2.056e+03 6.436e+03, threshold=2.995e+03, percent-clipped=12.0
+2023-03-01 17:37:55,290 INFO [train.py:968] (0/2) Epoch 3, batch 22600, giga_loss[loss=0.3173, simple_loss=0.3765, pruned_loss=0.1291, over 28631.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3842, pruned_loss=0.1312, over 5720662.05 frames. ], libri_tot_loss[loss=0.3539, simple_loss=0.4079, pruned_loss=0.15, over 5740358.77 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.38, pruned_loss=0.1275, over 5714872.53 frames. ], batch size: 242, lr: 9.80e-03, grad_scale: 4.0
+2023-03-01 17:38:34,338 INFO [train.py:968] (0/2) Epoch 3, batch 22650, giga_loss[loss=0.2837, simple_loss=0.3557, pruned_loss=0.1059, over 28997.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3817, pruned_loss=0.13, over 5715989.20 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4086, pruned_loss=0.1508, over 5739948.48 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3772, pruned_loss=0.126, over 5711675.88 frames. ], batch size: 164, lr: 9.80e-03, grad_scale: 4.0
+2023-03-01 17:38:38,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4218, 1.7948, 1.6036, 1.6176], device='cuda:0'), covar=tensor([0.1414, 0.1714, 0.1176, 0.0942], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0771, 0.0726, 0.0624], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 17:38:44,939 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:38:54,997 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:38:59,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.712e+02 1.116e+03 1.391e+03 1.768e+03 7.819e+03, threshold=2.783e+03, percent-clipped=6.0
+2023-03-01 17:39:00,667 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:39:03,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:39:12,056 INFO [train.py:968] (0/2) Epoch 3, batch 22700, giga_loss[loss=0.338, simple_loss=0.4056, pruned_loss=0.1352, over 28619.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3805, pruned_loss=0.1284, over 5709276.32 frames. ], libri_tot_loss[loss=0.3554, simple_loss=0.4088, pruned_loss=0.151, over 5732574.08 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3765, pruned_loss=0.1247, over 5711888.33 frames. ], batch size: 336, lr: 9.80e-03, grad_scale: 4.0
+2023-03-01 17:39:12,449 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 17:39:29,658 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:39:30,262 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:39:55,307 INFO [train.py:968] (0/2) Epoch 3, batch 22750, giga_loss[loss=0.2671, simple_loss=0.3329, pruned_loss=0.1007, over 28636.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3814, pruned_loss=0.1272, over 5704538.26 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.4092, pruned_loss=0.1515, over 5729583.36 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3775, pruned_loss=0.1237, over 5708963.92 frames. ], batch size: 85, lr: 9.79e-03, grad_scale: 4.0
+2023-03-01 17:40:08,108 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-01 17:40:24,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.941e+02 1.131e+03 1.494e+03 1.902e+03 5.238e+03, threshold=2.989e+03, percent-clipped=9.0
+2023-03-01 17:40:31,777 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.94 vs. limit=5.0
+2023-03-01 17:40:34,523 INFO [train.py:968] (0/2) Epoch 3, batch 22800, giga_loss[loss=0.3201, simple_loss=0.3827, pruned_loss=0.1288, over 28882.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3831, pruned_loss=0.1291, over 5712291.14 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.4096, pruned_loss=0.1521, over 5729722.00 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3789, pruned_loss=0.125, over 5715373.66 frames. ], batch size: 213, lr: 9.79e-03, grad_scale: 8.0
+2023-03-01 17:41:16,442 INFO [train.py:968] (0/2) Epoch 3, batch 22850, giga_loss[loss=0.2725, simple_loss=0.343, pruned_loss=0.101, over 28989.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3824, pruned_loss=0.1298, over 5715903.23 frames. ], libri_tot_loss[loss=0.3578, simple_loss=0.4101, pruned_loss=0.1527, over 5731737.26 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3781, pruned_loss=0.1256, over 5716453.51 frames. ], batch size: 136, lr: 9.79e-03, grad_scale: 8.0
+2023-03-01 17:41:27,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:41:41,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.197e+02 1.251e+03 1.618e+03 2.304e+03 1.357e+04, threshold=3.236e+03, percent-clipped=14.0
+2023-03-01 17:41:44,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2841, 3.1888, 1.3259, 1.2714], device='cuda:0'), covar=tensor([0.0828, 0.0370, 0.0883, 0.1273], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0447, 0.0312, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0018], device='cuda:0')
+2023-03-01 17:41:51,348 INFO [train.py:968] (0/2) Epoch 3, batch 22900, giga_loss[loss=0.3034, simple_loss=0.3627, pruned_loss=0.1221, over 28499.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3819, pruned_loss=0.1312, over 5722043.41 frames. ], libri_tot_loss[loss=0.3587, simple_loss=0.4108, pruned_loss=0.1533, over 5736427.11 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3765, pruned_loss=0.126, over 5717868.39 frames. ], batch size: 60, lr: 9.79e-03, grad_scale: 8.0
+2023-03-01 17:42:34,633 INFO [train.py:968] (0/2) Epoch 3, batch 22950, giga_loss[loss=0.28, simple_loss=0.3386, pruned_loss=0.1108, over 28638.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3793, pruned_loss=0.1306, over 5702280.55 frames. ], libri_tot_loss[loss=0.3591, simple_loss=0.411, pruned_loss=0.1536, over 5719129.23 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3745, pruned_loss=0.1261, over 5714274.01 frames. ], batch size: 92, lr: 9.79e-03, grad_scale: 4.0
+2023-03-01 17:42:55,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8304, 5.5017, 5.4971, 2.3556], device='cuda:0'), covar=tensor([0.0295, 0.0363, 0.0775, 0.1536], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0605, 0.0797, 0.0573], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 17:43:02,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.730e+02 1.001e+03 1.292e+03 2.078e+03 6.344e+03, threshold=2.584e+03, percent-clipped=6.0
+2023-03-01 17:43:10,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4860, 2.3055, 1.6924, 0.6867], device='cuda:0'), covar=tensor([0.2026, 0.0904, 0.1821, 0.2431], device='cuda:0'), in_proj_covar=tensor([0.1248, 0.1155, 0.1275, 0.1080], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 17:43:12,962 INFO [train.py:968] (0/2) Epoch 3, batch 23000, giga_loss[loss=0.3491, simple_loss=0.3867, pruned_loss=0.1557, over 28621.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3777, pruned_loss=0.1305, over 5712037.24 frames. ], libri_tot_loss[loss=0.3597, simple_loss=0.4113, pruned_loss=0.154, over 5722193.98 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3729, pruned_loss=0.1261, over 5718608.33 frames. ], batch size: 92, lr: 9.78e-03, grad_scale: 4.0
+2023-03-01 17:43:21,578 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:43:24,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:43:28,533 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-01 17:43:32,342 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:43:43,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:43:46,382 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:43:53,299 INFO [train.py:968] (0/2) Epoch 3, batch 23050, giga_loss[loss=0.2843, simple_loss=0.3528, pruned_loss=0.1079, over 29027.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3755, pruned_loss=0.1295, over 5712051.78 frames. ], libri_tot_loss[loss=0.3598, simple_loss=0.4113, pruned_loss=0.1541, over 5726728.33 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3709, pruned_loss=0.1253, over 5712976.95 frames. ], batch size: 164, lr: 9.78e-03, grad_scale: 4.0
+2023-03-01 17:43:56,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:44:11,406 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-01 17:44:20,925 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.071e+02 1.012e+03 1.394e+03 1.908e+03 8.352e+03, threshold=2.787e+03, percent-clipped=15.0
+2023-03-01 17:44:26,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:44:30,941 INFO [train.py:968] (0/2) Epoch 3, batch 23100, giga_loss[loss=0.2568, simple_loss=0.3205, pruned_loss=0.09656, over 28433.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3707, pruned_loss=0.1268, over 5716355.53 frames. ], libri_tot_loss[loss=0.3601, simple_loss=0.4115, pruned_loss=0.1544, over 5725987.58 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3662, pruned_loss=0.1229, over 5717379.57 frames. ], batch size: 71, lr: 9.78e-03, grad_scale: 4.0
+2023-03-01 17:45:12,563 INFO [train.py:968] (0/2) Epoch 3, batch 23150, giga_loss[loss=0.2969, simple_loss=0.3528, pruned_loss=0.1205, over 28935.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3666, pruned_loss=0.1248, over 5715118.26 frames. ], libri_tot_loss[loss=0.3602, simple_loss=0.4115, pruned_loss=0.1545, over 5725749.11 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3622, pruned_loss=0.121, over 5716172.77 frames. ], batch size: 227, lr: 9.78e-03, grad_scale: 4.0
+2023-03-01 17:45:38,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:45:40,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:45:40,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.764e+02 1.332e+03 1.926e+03 2.448e+03 4.543e+03, threshold=3.852e+03, percent-clipped=17.0
+2023-03-01 17:45:48,486 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:45:50,369 INFO [train.py:968] (0/2) Epoch 3, batch 23200, giga_loss[loss=0.2846, simple_loss=0.3598, pruned_loss=0.1047, over 28941.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3648, pruned_loss=0.1232, over 5715075.07 frames. ], libri_tot_loss[loss=0.3605, simple_loss=0.4115, pruned_loss=0.1547, over 5724512.57 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3604, pruned_loss=0.1194, over 5716974.05 frames. ], batch size: 164, lr: 9.77e-03, grad_scale: 8.0
+2023-03-01 17:45:50,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:46:02,717 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:46:15,016 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:46:20,325 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:46:22,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:46:30,792 INFO [train.py:968] (0/2) Epoch 3, batch 23250, giga_loss[loss=0.3043, simple_loss=0.3678, pruned_loss=0.1204, over 28545.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3687, pruned_loss=0.125, over 5714130.57 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4113, pruned_loss=0.1547, over 5726641.38 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3631, pruned_loss=0.1204, over 5712907.27 frames. ], batch size: 60, lr: 9.77e-03, grad_scale: 8.0
+2023-03-01 17:46:45,986 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:47:01,818 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.785e+02 1.079e+03 1.303e+03 1.715e+03 4.183e+03, threshold=2.606e+03, percent-clipped=1.0
+2023-03-01 17:47:12,985 INFO [train.py:968] (0/2) Epoch 3, batch 23300, giga_loss[loss=0.3672, simple_loss=0.4231, pruned_loss=0.1557, over 27913.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.373, pruned_loss=0.1273, over 5699836.03 frames. ], libri_tot_loss[loss=0.3602, simple_loss=0.4111, pruned_loss=0.1546, over 5717900.07 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3676, pruned_loss=0.1229, over 5706809.52 frames. ], batch size: 412, lr: 9.77e-03, grad_scale: 4.0
+2023-03-01 17:47:18,294 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-01 17:47:42,169 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:47:54,512 INFO [train.py:968] (0/2) Epoch 3, batch 23350, giga_loss[loss=0.3406, simple_loss=0.3984, pruned_loss=0.1415, over 28629.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.377, pruned_loss=0.129, over 5697277.28 frames. ], libri_tot_loss[loss=0.361, simple_loss=0.4115, pruned_loss=0.1552, over 5713655.95 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3713, pruned_loss=0.1241, over 5707048.27 frames. ], batch size: 92, lr: 9.77e-03, grad_scale: 4.0
+2023-03-01 17:48:25,309 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.092e+02 1.219e+03 1.577e+03 2.614e+03 1.585e+04, threshold=3.154e+03, percent-clipped=24.0
+2023-03-01 17:48:35,215 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:48:36,236 INFO [train.py:968] (0/2) Epoch 3, batch 23400, giga_loss[loss=0.3337, simple_loss=0.3886, pruned_loss=0.1394, over 28728.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3799, pruned_loss=0.1301, over 5698822.26 frames. ], libri_tot_loss[loss=0.3609, simple_loss=0.4115, pruned_loss=0.1552, over 5707862.23 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3746, pruned_loss=0.1257, over 5711257.77 frames. ], batch size: 92, lr: 9.77e-03, grad_scale: 4.0
+2023-03-01 17:49:18,947 INFO [train.py:968] (0/2) Epoch 3, batch 23450, giga_loss[loss=0.2997, simple_loss=0.3645, pruned_loss=0.1175, over 28919.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3819, pruned_loss=0.131, over 5701747.75 frames. ], libri_tot_loss[loss=0.3614, simple_loss=0.4117, pruned_loss=0.1555, over 5703448.82 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3768, pruned_loss=0.1267, over 5716066.95 frames. ], batch size: 66, lr: 9.76e-03, grad_scale: 4.0
+2023-03-01 17:49:46,464 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9511, 1.3053, 1.0495, 0.1842], device='cuda:0'), covar=tensor([0.1038, 0.0779, 0.1237, 0.1585], device='cuda:0'), in_proj_covar=tensor([0.1257, 0.1170, 0.1282, 0.1078], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 17:49:51,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.325e+02 1.302e+03 1.782e+03 2.460e+03 1.072e+04, threshold=3.563e+03, percent-clipped=13.0
+2023-03-01 17:50:01,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2002, 2.9131, 2.9130, 1.4148], device='cuda:0'), covar=tensor([0.0780, 0.0578, 0.1073, 0.2109], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0598, 0.0776, 0.0566], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 17:50:02,874 INFO [train.py:968] (0/2) Epoch 3, batch 23500, giga_loss[loss=0.3097, simple_loss=0.3722, pruned_loss=0.1236, over 28872.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3874, pruned_loss=0.1363, over 5703397.42 frames. ], libri_tot_loss[loss=0.3612, simple_loss=0.4114, pruned_loss=0.1555, over 5710766.12 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3825, pruned_loss=0.1319, over 5708549.65 frames. ], batch size: 112, lr: 9.76e-03, grad_scale: 4.0
+2023-03-01 17:50:15,705 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-01 17:50:47,802 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:50:50,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:51:00,617 INFO [train.py:968] (0/2) Epoch 3, batch 23550, giga_loss[loss=0.4425, simple_loss=0.4471, pruned_loss=0.2189, over 23773.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3953, pruned_loss=0.1434, over 5690380.89 frames. ], libri_tot_loss[loss=0.3612, simple_loss=0.4115, pruned_loss=0.1555, over 5711697.29 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3913, pruned_loss=0.1399, over 5693613.19 frames. ], batch size: 705, lr: 9.76e-03, grad_scale: 4.0
+2023-03-01 17:51:25,204 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:51:39,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.716e+03 2.186e+03 2.926e+03 8.542e+03, threshold=4.373e+03, percent-clipped=13.0
+2023-03-01 17:51:55,448 INFO [train.py:968] (0/2) Epoch 3, batch 23600, giga_loss[loss=0.4288, simple_loss=0.4599, pruned_loss=0.1988, over 28836.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.4032, pruned_loss=0.15, over 5675687.38 frames. ], libri_tot_loss[loss=0.3617, simple_loss=0.4117, pruned_loss=0.1559, over 5707197.56 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.3995, pruned_loss=0.1467, over 5682554.97 frames. ], batch size: 199, lr: 9.76e-03, grad_scale: 8.0
+2023-03-01 17:52:02,360 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7374, 1.8749, 1.7530, 1.7681], device='cuda:0'), covar=tensor([0.1178, 0.1486, 0.0970, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0780, 0.0728, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 17:52:37,653 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 17:52:47,682 INFO [train.py:968] (0/2) Epoch 3, batch 23650, giga_loss[loss=0.3589, simple_loss=0.4132, pruned_loss=0.1523, over 28639.00 frames. ], tot_loss[loss=0.3604, simple_loss=0.4096, pruned_loss=0.1556, over 5671445.89 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.412, pruned_loss=0.1561, over 5700004.63 frames. ], giga_tot_loss[loss=0.3559, simple_loss=0.4064, pruned_loss=0.1527, over 5682414.33 frames. ], batch size: 60, lr: 9.75e-03, grad_scale: 8.0
+2023-03-01 17:53:27,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.024e+02 1.725e+03 2.209e+03 2.786e+03 7.712e+03, threshold=4.417e+03, percent-clipped=5.0
+2023-03-01 17:53:38,914 INFO [train.py:968] (0/2) Epoch 3, batch 23700, giga_loss[loss=0.5214, simple_loss=0.5049, pruned_loss=0.269, over 26655.00 frames. ], tot_loss[loss=0.3717, simple_loss=0.4176, pruned_loss=0.1629, over 5673732.03 frames. ], libri_tot_loss[loss=0.3641, simple_loss=0.4134, pruned_loss=0.1574, over 5702707.96 frames. ], giga_tot_loss[loss=0.3663, simple_loss=0.4137, pruned_loss=0.1595, over 5679209.24 frames. ], batch size: 555, lr: 9.75e-03, grad_scale: 4.0
+2023-03-01 17:53:48,530 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-114000.pt
+2023-03-01 17:53:50,480 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:54:00,080 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=114010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:54:08,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9577, 1.1365, 1.0495, 1.0803], device='cuda:0'), covar=tensor([0.0981, 0.1024, 0.1430, 0.1084], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0783, 0.0633, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 17:54:30,327 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=114039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:54:31,434 INFO [train.py:968] (0/2) Epoch 3, batch 23750, giga_loss[loss=0.4819, simple_loss=0.476, pruned_loss=0.2439, over 26597.00 frames. ], tot_loss[loss=0.3797, simple_loss=0.423, pruned_loss=0.1682, over 5665840.70 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.4136, pruned_loss=0.1576, over 5703237.47 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.4197, pruned_loss=0.1654, over 5669057.17 frames. ], batch size: 555, lr: 9.75e-03, grad_scale: 4.0
+2023-03-01 17:55:07,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.664e+03 2.331e+03 3.299e+03 8.938e+03, threshold=4.662e+03, percent-clipped=11.0
+2023-03-01 17:55:17,631 INFO [train.py:968] (0/2) Epoch 3, batch 23800, giga_loss[loss=0.35, simple_loss=0.402, pruned_loss=0.149, over 28734.00 frames. ], tot_loss[loss=0.3823, simple_loss=0.424, pruned_loss=0.1703, over 5658380.23 frames. ], libri_tot_loss[loss=0.3646, simple_loss=0.4136, pruned_loss=0.1578, over 5701912.97 frames. ], giga_tot_loss[loss=0.3793, simple_loss=0.4218, pruned_loss=0.1684, over 5660719.63 frames. ], batch size: 99, lr: 9.75e-03, grad_scale: 4.0
+2023-03-01 17:56:09,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1883, 1.3815, 1.1354, 1.2014], device='cuda:0'), covar=tensor([0.2051, 0.1932, 0.1840, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.1021, 0.0815, 0.0915, 0.0936], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 17:56:10,259 INFO [train.py:968] (0/2) Epoch 3, batch 23850, libri_loss[loss=0.3786, simple_loss=0.4148, pruned_loss=0.1712, over 29576.00 frames. ], tot_loss[loss=0.3884, simple_loss=0.4273, pruned_loss=0.1748, over 5654835.11 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.4133, pruned_loss=0.1578, over 5702778.48 frames. ], giga_tot_loss[loss=0.3865, simple_loss=0.4261, pruned_loss=0.1735, over 5654872.97 frames. ], batch size: 78, lr: 9.75e-03, grad_scale: 4.0
+2023-03-01 17:56:10,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2440, 1.9441, 1.2983, 1.4369], device='cuda:0'), covar=tensor([0.0830, 0.0340, 0.0389, 0.0922], device='cuda:0'), in_proj_covar=tensor([0.0231, 0.0153, 0.0156, 0.0278], device='cuda:0'), out_proj_covar=tensor([0.0041, 0.0031, 0.0027, 0.0046], device='cuda:0')
+2023-03-01 17:56:15,601 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:56:18,421 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:56:49,281 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:56:51,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.377e+02 1.648e+03 2.140e+03 2.604e+03 6.367e+03, threshold=4.279e+03, percent-clipped=4.0
+2023-03-01 17:57:05,584 INFO [train.py:968] (0/2) Epoch 3, batch 23900, giga_loss[loss=0.5415, simple_loss=0.5157, pruned_loss=0.2836, over 26457.00 frames. ], tot_loss[loss=0.3953, simple_loss=0.4313, pruned_loss=0.1796, over 5637752.37 frames. ], libri_tot_loss[loss=0.3648, simple_loss=0.4135, pruned_loss=0.1581, over 5695388.69 frames. ], giga_tot_loss[loss=0.3938, simple_loss=0.4304, pruned_loss=0.1786, over 5644267.46 frames. ], batch size: 555, lr: 9.74e-03, grad_scale: 4.0
+2023-03-01 17:58:02,897 INFO [train.py:968] (0/2) Epoch 3, batch 23950, giga_loss[loss=0.394, simple_loss=0.4378, pruned_loss=0.1751, over 28934.00 frames. ], tot_loss[loss=0.3981, simple_loss=0.4345, pruned_loss=0.1808, over 5634323.29 frames. ], libri_tot_loss[loss=0.3654, simple_loss=0.4139, pruned_loss=0.1585, over 5685748.79 frames. ], giga_tot_loss[loss=0.397, simple_loss=0.4338, pruned_loss=0.1801, over 5646399.70 frames. ], batch size: 106, lr: 9.74e-03, grad_scale: 4.0
+2023-03-01 17:58:44,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.277e+03 2.200e+03 2.783e+03 3.343e+03 1.006e+04, threshold=5.566e+03, percent-clipped=10.0
+2023-03-01 17:58:57,238 INFO [train.py:968] (0/2) Epoch 3, batch 24000, giga_loss[loss=0.5026, simple_loss=0.4938, pruned_loss=0.2557, over 26565.00 frames. ], tot_loss[loss=0.4012, simple_loss=0.4357, pruned_loss=0.1833, over 5619193.63 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.414, pruned_loss=0.1587, over 5670573.83 frames. ], giga_tot_loss[loss=0.4013, simple_loss=0.4358, pruned_loss=0.1834, over 5640709.88 frames. ], batch size: 555, lr: 9.74e-03, grad_scale: 4.0
+2023-03-01 17:58:57,243 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 17:59:05,879 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2531, simple_loss=0.3533, pruned_loss=0.07639, over 944034.00 frames. 
+2023-03-01 17:59:05,880 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 17:59:52,691 INFO [train.py:968] (0/2) Epoch 3, batch 24050, giga_loss[loss=0.3967, simple_loss=0.4286, pruned_loss=0.1824, over 28298.00 frames. ], tot_loss[loss=0.3991, simple_loss=0.4339, pruned_loss=0.1822, over 5630303.21 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4147, pruned_loss=0.1593, over 5677073.07 frames. ], giga_tot_loss[loss=0.3994, simple_loss=0.4339, pruned_loss=0.1824, over 5640379.14 frames. ], batch size: 368, lr: 9.74e-03, grad_scale: 4.0
+2023-03-01 18:00:34,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.088e+03 1.935e+03 2.616e+03 3.412e+03 7.585e+03, threshold=5.233e+03, percent-clipped=4.0
+2023-03-01 18:00:38,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:00:44,210 INFO [train.py:968] (0/2) Epoch 3, batch 24100, giga_loss[loss=0.3988, simple_loss=0.4393, pruned_loss=0.1792, over 28616.00 frames. ], tot_loss[loss=0.3962, simple_loss=0.4323, pruned_loss=0.1801, over 5632778.92 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.4142, pruned_loss=0.159, over 5679212.61 frames. ], giga_tot_loss[loss=0.3971, simple_loss=0.4328, pruned_loss=0.1807, over 5638316.95 frames. ], batch size: 307, lr: 9.74e-03, grad_scale: 2.0
+2023-03-01 18:01:05,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:01:36,273 INFO [train.py:968] (0/2) Epoch 3, batch 24150, giga_loss[loss=0.4188, simple_loss=0.4295, pruned_loss=0.204, over 23102.00 frames. ], tot_loss[loss=0.3942, simple_loss=0.4314, pruned_loss=0.1784, over 5629335.36 frames. ], libri_tot_loss[loss=0.3659, simple_loss=0.414, pruned_loss=0.1589, over 5680350.86 frames. ], giga_tot_loss[loss=0.3956, simple_loss=0.4324, pruned_loss=0.1794, over 5632170.64 frames. ], batch size: 705, lr: 9.73e-03, grad_scale: 2.0
+2023-03-01 18:01:45,392 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 18:02:18,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.880e+02 1.412e+03 1.724e+03 2.346e+03 7.574e+03, threshold=3.448e+03, percent-clipped=3.0
+2023-03-01 18:02:28,382 INFO [train.py:968] (0/2) Epoch 3, batch 24200, giga_loss[loss=0.3689, simple_loss=0.419, pruned_loss=0.1594, over 28250.00 frames. ], tot_loss[loss=0.3927, simple_loss=0.4308, pruned_loss=0.1773, over 5622171.10 frames. ], libri_tot_loss[loss=0.3654, simple_loss=0.4136, pruned_loss=0.1586, over 5681467.78 frames. ], giga_tot_loss[loss=0.3948, simple_loss=0.4323, pruned_loss=0.1787, over 5622316.20 frames. ], batch size: 77, lr: 9.73e-03, grad_scale: 2.0
+2023-03-01 18:03:01,169 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=114520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:03:08,039 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:03:10,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:03:22,105 INFO [train.py:968] (0/2) Epoch 3, batch 24250, giga_loss[loss=0.3918, simple_loss=0.4166, pruned_loss=0.1835, over 24035.00 frames. ], tot_loss[loss=0.3866, simple_loss=0.4267, pruned_loss=0.1733, over 5624501.29 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4132, pruned_loss=0.1585, over 5682141.51 frames. ], giga_tot_loss[loss=0.3889, simple_loss=0.4284, pruned_loss=0.1747, over 5623476.20 frames. ], batch size: 705, lr: 9.73e-03, grad_scale: 2.0
+2023-03-01 18:03:40,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:03:42,980 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:03:43,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:04:02,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.737e+02 1.598e+03 2.047e+03 2.498e+03 4.970e+03, threshold=4.095e+03, percent-clipped=8.0
+2023-03-01 18:04:10,564 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:04:12,845 INFO [train.py:968] (0/2) Epoch 3, batch 24300, giga_loss[loss=0.3635, simple_loss=0.4159, pruned_loss=0.1555, over 28843.00 frames. ], tot_loss[loss=0.3825, simple_loss=0.4248, pruned_loss=0.1701, over 5635723.72 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4134, pruned_loss=0.1589, over 5686764.64 frames. ], giga_tot_loss[loss=0.3841, simple_loss=0.4262, pruned_loss=0.171, over 5629909.30 frames. ], batch size: 145, lr: 9.73e-03, grad_scale: 2.0
+2023-03-01 18:05:01,575 INFO [train.py:968] (0/2) Epoch 3, batch 24350, giga_loss[loss=0.3139, simple_loss=0.3795, pruned_loss=0.1241, over 29030.00 frames. ], tot_loss[loss=0.3787, simple_loss=0.4224, pruned_loss=0.1675, over 5651482.22 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.4128, pruned_loss=0.1587, over 5693168.18 frames. ], giga_tot_loss[loss=0.381, simple_loss=0.4243, pruned_loss=0.1688, over 5639777.80 frames. ], batch size: 155, lr: 9.73e-03, grad_scale: 2.0
+2023-03-01 18:05:36,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.866e+03 2.318e+03 3.087e+03 7.318e+03, threshold=4.636e+03, percent-clipped=4.0
+2023-03-01 18:05:40,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6049, 1.5438, 1.1188, 1.2154], device='cuda:0'), covar=tensor([0.0653, 0.0658, 0.1026, 0.0982], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0486, 0.0525, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 18:05:47,589 INFO [train.py:968] (0/2) Epoch 3, batch 24400, giga_loss[loss=0.3915, simple_loss=0.4354, pruned_loss=0.1738, over 27891.00 frames. ], tot_loss[loss=0.374, simple_loss=0.4194, pruned_loss=0.1643, over 5670557.08 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4127, pruned_loss=0.1588, over 5699195.33 frames. ], giga_tot_loss[loss=0.376, simple_loss=0.4213, pruned_loss=0.1654, over 5655027.28 frames. ], batch size: 412, lr: 9.72e-03, grad_scale: 4.0
+2023-03-01 18:06:33,915 INFO [train.py:968] (0/2) Epoch 3, batch 24450, giga_loss[loss=0.3982, simple_loss=0.4242, pruned_loss=0.1861, over 28595.00 frames. ], tot_loss[loss=0.3702, simple_loss=0.4161, pruned_loss=0.1622, over 5646461.29 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4128, pruned_loss=0.1591, over 5683143.75 frames. ], giga_tot_loss[loss=0.3716, simple_loss=0.4175, pruned_loss=0.1628, over 5647724.29 frames. ], batch size: 307, lr: 9.72e-03, grad_scale: 4.0
+2023-03-01 18:07:11,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0777, 1.2450, 0.9287, 0.3485], device='cuda:0'), covar=tensor([0.0792, 0.0724, 0.1187, 0.1469], device='cuda:0'), in_proj_covar=tensor([0.1268, 0.1187, 0.1266, 0.1080], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 18:07:14,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.110e+02 1.526e+03 1.964e+03 2.666e+03 1.116e+04, threshold=3.929e+03, percent-clipped=7.0
+2023-03-01 18:07:24,233 INFO [train.py:968] (0/2) Epoch 3, batch 24500, giga_loss[loss=0.3698, simple_loss=0.416, pruned_loss=0.1618, over 28641.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.416, pruned_loss=0.1616, over 5667984.36 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.4126, pruned_loss=0.159, over 5687446.87 frames. ], giga_tot_loss[loss=0.3709, simple_loss=0.4174, pruned_loss=0.1622, over 5664667.72 frames. ], batch size: 336, lr: 9.72e-03, grad_scale: 4.0
+2023-03-01 18:08:13,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3765, 2.0217, 1.5586, 0.6650], device='cuda:0'), covar=tensor([0.1690, 0.1020, 0.1705, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.1258, 0.1182, 0.1268, 0.1077], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 18:08:21,263 INFO [train.py:968] (0/2) Epoch 3, batch 24550, giga_loss[loss=0.3487, simple_loss=0.4118, pruned_loss=0.1428, over 28879.00 frames. ], tot_loss[loss=0.3701, simple_loss=0.4165, pruned_loss=0.1618, over 5668172.92 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.4125, pruned_loss=0.159, over 5689024.11 frames. ], giga_tot_loss[loss=0.3712, simple_loss=0.4177, pruned_loss=0.1623, over 5663857.87 frames. ], batch size: 112, lr: 9.72e-03, grad_scale: 4.0
+2023-03-01 18:08:31,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6857, 1.9621, 1.8547, 1.7611], device='cuda:0'), covar=tensor([0.1462, 0.1756, 0.1143, 0.0941], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0788, 0.0731, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:09:05,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.892e+02 1.315e+03 1.781e+03 2.549e+03 9.548e+03, threshold=3.562e+03, percent-clipped=3.0
+2023-03-01 18:09:13,816 INFO [train.py:968] (0/2) Epoch 3, batch 24600, giga_loss[loss=0.3101, simple_loss=0.3832, pruned_loss=0.1185, over 28677.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4127, pruned_loss=0.1573, over 5668089.35 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4122, pruned_loss=0.1589, over 5690338.88 frames. ], giga_tot_loss[loss=0.3648, simple_loss=0.414, pruned_loss=0.1579, over 5663475.79 frames. ], batch size: 92, lr: 9.71e-03, grad_scale: 4.0
+2023-03-01 18:09:18,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:09:39,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5593, 1.5492, 1.1166, 0.9331], device='cuda:0'), covar=tensor([0.0736, 0.0600, 0.0544, 0.0699], device='cuda:0'), in_proj_covar=tensor([0.1230, 0.0950, 0.0983, 0.1049], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 18:10:07,027 INFO [train.py:968] (0/2) Epoch 3, batch 24650, giga_loss[loss=0.3487, simple_loss=0.4182, pruned_loss=0.1396, over 28915.00 frames. ], tot_loss[loss=0.3611, simple_loss=0.4131, pruned_loss=0.1546, over 5672006.17 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.4122, pruned_loss=0.159, over 5693364.34 frames. ], giga_tot_loss[loss=0.3619, simple_loss=0.4141, pruned_loss=0.1549, over 5665441.98 frames. ], batch size: 213, lr: 9.71e-03, grad_scale: 4.0
+2023-03-01 18:10:51,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.753e+02 1.407e+03 1.816e+03 2.597e+03 6.959e+03, threshold=3.631e+03, percent-clipped=11.0
+2023-03-01 18:10:59,154 INFO [train.py:968] (0/2) Epoch 3, batch 24700, giga_loss[loss=0.3807, simple_loss=0.4338, pruned_loss=0.1638, over 28864.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4141, pruned_loss=0.1555, over 5663933.56 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.4121, pruned_loss=0.1591, over 5692066.83 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.415, pruned_loss=0.1556, over 5659309.14 frames. ], batch size: 66, lr: 9.71e-03, grad_scale: 4.0
+2023-03-01 18:11:49,188 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:11:50,968 INFO [train.py:968] (0/2) Epoch 3, batch 24750, giga_loss[loss=0.369, simple_loss=0.4169, pruned_loss=0.1605, over 27903.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.415, pruned_loss=0.1567, over 5662953.89 frames. ], libri_tot_loss[loss=0.3649, simple_loss=0.4121, pruned_loss=0.1588, over 5696045.86 frames. ], giga_tot_loss[loss=0.3649, simple_loss=0.4158, pruned_loss=0.157, over 5655098.01 frames. ], batch size: 412, lr: 9.71e-03, grad_scale: 4.0
+2023-03-01 18:11:51,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:11:53,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8075, 1.4265, 1.1702, 1.2575], device='cuda:0'), covar=tensor([0.0492, 0.0568, 0.0875, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0487, 0.0524, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 18:12:16,856 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:12:24,086 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.51 vs. limit=5.0
+2023-03-01 18:12:29,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.185e+02 1.907e+03 2.280e+03 2.807e+03 6.471e+03, threshold=4.560e+03, percent-clipped=8.0
+2023-03-01 18:12:39,231 INFO [train.py:968] (0/2) Epoch 3, batch 24800, libri_loss[loss=0.3693, simple_loss=0.4116, pruned_loss=0.1634, over 29533.00 frames. ], tot_loss[loss=0.3661, simple_loss=0.415, pruned_loss=0.1585, over 5655265.02 frames. ], libri_tot_loss[loss=0.3655, simple_loss=0.4125, pruned_loss=0.1592, over 5702461.24 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.4154, pruned_loss=0.1583, over 5641879.20 frames. ], batch size: 81, lr: 9.71e-03, grad_scale: 8.0
+2023-03-01 18:12:56,258 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:13:25,261 INFO [train.py:968] (0/2) Epoch 3, batch 24850, giga_loss[loss=0.4191, simple_loss=0.4425, pruned_loss=0.1978, over 27538.00 frames. ], tot_loss[loss=0.367, simple_loss=0.4146, pruned_loss=0.1597, over 5665277.18 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.4129, pruned_loss=0.1596, over 5701570.87 frames. ], giga_tot_loss[loss=0.3665, simple_loss=0.4146, pruned_loss=0.1592, over 5654149.04 frames. ], batch size: 472, lr: 9.70e-03, grad_scale: 4.0
+2023-03-01 18:13:38,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2574, 1.3490, 0.9962, 0.5904], device='cuda:0'), covar=tensor([0.0676, 0.0638, 0.0487, 0.0735], device='cuda:0'), in_proj_covar=tensor([0.1218, 0.0941, 0.0978, 0.1039], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 18:14:02,177 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.626e+03 2.261e+03 3.526e+03 1.198e+04, threshold=4.521e+03, percent-clipped=10.0
+2023-03-01 18:14:08,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8926, 1.1951, 3.9619, 3.1442], device='cuda:0'), covar=tensor([0.1648, 0.2080, 0.0364, 0.0550], device='cuda:0'), in_proj_covar=tensor([0.0537, 0.0512, 0.0694, 0.0555], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 18:14:11,186 INFO [train.py:968] (0/2) Epoch 3, batch 24900, giga_loss[loss=0.3556, simple_loss=0.4088, pruned_loss=0.1512, over 28688.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.4143, pruned_loss=0.1603, over 5657939.60 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4132, pruned_loss=0.1601, over 5692574.43 frames. ], giga_tot_loss[loss=0.3666, simple_loss=0.4141, pruned_loss=0.1595, over 5656064.58 frames. ], batch size: 92, lr: 9.70e-03, grad_scale: 4.0
+2023-03-01 18:14:36,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-01 18:14:54,333 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2971, 2.0164, 1.4511, 1.4904], device='cuda:0'), covar=tensor([0.0923, 0.0319, 0.0393, 0.0992], device='cuda:0'), in_proj_covar=tensor([0.0232, 0.0154, 0.0159, 0.0279], device='cuda:0'), out_proj_covar=tensor([0.0042, 0.0031, 0.0028, 0.0047], device='cuda:0')
+2023-03-01 18:14:57,913 INFO [train.py:968] (0/2) Epoch 3, batch 24950, giga_loss[loss=0.3778, simple_loss=0.4258, pruned_loss=0.1649, over 28786.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4132, pruned_loss=0.1577, over 5678939.70 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4135, pruned_loss=0.1602, over 5699401.83 frames. ], giga_tot_loss[loss=0.3633, simple_loss=0.4128, pruned_loss=0.157, over 5670476.69 frames. ], batch size: 284, lr: 9.70e-03, grad_scale: 4.0
+2023-03-01 18:15:19,337 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:15:39,646 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.941e+02 1.633e+03 2.158e+03 2.906e+03 6.039e+03, threshold=4.317e+03, percent-clipped=6.0
+2023-03-01 18:15:50,399 INFO [train.py:968] (0/2) Epoch 3, batch 25000, giga_loss[loss=0.3226, simple_loss=0.3856, pruned_loss=0.1298, over 28736.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.411, pruned_loss=0.1559, over 5668456.91 frames. ], libri_tot_loss[loss=0.3662, simple_loss=0.4129, pruned_loss=0.1598, over 5702526.64 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.4112, pruned_loss=0.1556, over 5658791.35 frames. ], batch size: 60, lr: 9.70e-03, grad_scale: 4.0
+2023-03-01 18:16:38,438 INFO [train.py:968] (0/2) Epoch 3, batch 25050, giga_loss[loss=0.3641, simple_loss=0.4161, pruned_loss=0.1561, over 28969.00 frames. ], tot_loss[loss=0.361, simple_loss=0.4113, pruned_loss=0.1554, over 5678519.00 frames. ], libri_tot_loss[loss=0.3662, simple_loss=0.4128, pruned_loss=0.1598, over 5705791.34 frames. ], giga_tot_loss[loss=0.3608, simple_loss=0.4114, pruned_loss=0.1551, over 5667411.33 frames. ], batch size: 213, lr: 9.70e-03, grad_scale: 4.0
+2023-03-01 18:16:41,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3524, 1.4242, 1.2657, 1.4533], device='cuda:0'), covar=tensor([0.1800, 0.1723, 0.1508, 0.1675], device='cuda:0'), in_proj_covar=tensor([0.1030, 0.0836, 0.0925, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 18:17:21,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.030e+02 1.535e+03 1.991e+03 2.598e+03 7.400e+03, threshold=3.982e+03, percent-clipped=7.0
+2023-03-01 18:17:30,379 INFO [train.py:968] (0/2) Epoch 3, batch 25100, giga_loss[loss=0.3439, simple_loss=0.3948, pruned_loss=0.1465, over 28378.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.4097, pruned_loss=0.1546, over 5674082.35 frames. ], libri_tot_loss[loss=0.3663, simple_loss=0.4129, pruned_loss=0.1599, over 5694372.35 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4097, pruned_loss=0.1542, over 5674815.11 frames. ], batch size: 65, lr: 9.69e-03, grad_scale: 4.0
+2023-03-01 18:18:19,680 INFO [train.py:968] (0/2) Epoch 3, batch 25150, giga_loss[loss=0.3205, simple_loss=0.3772, pruned_loss=0.132, over 28562.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.4077, pruned_loss=0.1538, over 5680387.45 frames. ], libri_tot_loss[loss=0.3658, simple_loss=0.4124, pruned_loss=0.1596, over 5700695.63 frames. ], giga_tot_loss[loss=0.3576, simple_loss=0.408, pruned_loss=0.1536, over 5674726.40 frames. ], batch size: 78, lr: 9.69e-03, grad_scale: 4.0
+2023-03-01 18:18:58,396 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:19:01,347 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.882e+02 1.675e+03 2.026e+03 3.054e+03 7.760e+03, threshold=4.052e+03, percent-clipped=12.0
+2023-03-01 18:19:03,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:19:10,244 INFO [train.py:968] (0/2) Epoch 3, batch 25200, giga_loss[loss=0.3898, simple_loss=0.4268, pruned_loss=0.1764, over 28522.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4063, pruned_loss=0.1534, over 5689340.67 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4126, pruned_loss=0.1597, over 5701769.15 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.4064, pruned_loss=0.1531, over 5683840.00 frames. ], batch size: 336, lr: 9.69e-03, grad_scale: 8.0
+2023-03-01 18:19:11,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5684, 1.9167, 1.6701, 1.6755], device='cuda:0'), covar=tensor([0.1264, 0.1720, 0.1145, 0.0965], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0795, 0.0735, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:19:59,892 INFO [train.py:968] (0/2) Epoch 3, batch 25250, libri_loss[loss=0.4045, simple_loss=0.4453, pruned_loss=0.1818, over 29245.00 frames. ], tot_loss[loss=0.3554, simple_loss=0.405, pruned_loss=0.1529, over 5691433.39 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4126, pruned_loss=0.1597, over 5705574.60 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4049, pruned_loss=0.1525, over 5683438.25 frames. ], batch size: 94, lr: 9.69e-03, grad_scale: 4.0
+2023-03-01 18:20:36,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.687e+02 1.598e+03 2.107e+03 2.978e+03 6.010e+03, threshold=4.215e+03, percent-clipped=9.0
+2023-03-01 18:20:46,307 INFO [train.py:968] (0/2) Epoch 3, batch 25300, giga_loss[loss=0.385, simple_loss=0.4229, pruned_loss=0.1736, over 28193.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4035, pruned_loss=0.1525, over 5686589.66 frames. ], libri_tot_loss[loss=0.3659, simple_loss=0.4125, pruned_loss=0.1597, over 5702779.78 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4032, pruned_loss=0.152, over 5682505.55 frames. ], batch size: 368, lr: 9.69e-03, grad_scale: 4.0
+2023-03-01 18:20:51,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8635, 1.5597, 1.5601, 1.5446], device='cuda:0'), covar=tensor([0.0810, 0.1556, 0.1283, 0.1242], device='cuda:0'), in_proj_covar=tensor([0.0452, 0.0775, 0.0627, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 18:21:21,836 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:21:24,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:21:31,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:21:32,692 INFO [train.py:968] (0/2) Epoch 3, batch 25350, giga_loss[loss=0.3526, simple_loss=0.4047, pruned_loss=0.1502, over 28800.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4033, pruned_loss=0.1528, over 5673620.59 frames. ], libri_tot_loss[loss=0.3658, simple_loss=0.4124, pruned_loss=0.1597, over 5691065.04 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.4028, pruned_loss=0.1521, over 5679781.78 frames. ], batch size: 119, lr: 9.68e-03, grad_scale: 4.0
+2023-03-01 18:21:50,656 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:22:10,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.535e+02 1.501e+03 1.941e+03 2.646e+03 8.157e+03, threshold=3.883e+03, percent-clipped=5.0
+2023-03-01 18:22:19,879 INFO [train.py:968] (0/2) Epoch 3, batch 25400, giga_loss[loss=0.3725, simple_loss=0.423, pruned_loss=0.161, over 28887.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.4055, pruned_loss=0.1539, over 5674648.32 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.4126, pruned_loss=0.1599, over 5692800.06 frames. ], giga_tot_loss[loss=0.3554, simple_loss=0.4048, pruned_loss=0.153, over 5677507.63 frames. ], batch size: 186, lr: 9.68e-03, grad_scale: 4.0
+2023-03-01 18:23:02,399 INFO [train.py:968] (0/2) Epoch 3, batch 25450, giga_loss[loss=0.3764, simple_loss=0.4257, pruned_loss=0.1635, over 28667.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.4057, pruned_loss=0.1533, over 5675056.98 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4119, pruned_loss=0.1596, over 5686284.67 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4053, pruned_loss=0.1525, over 5683169.18 frames. ], batch size: 78, lr: 9.68e-03, grad_scale: 4.0
+2023-03-01 18:23:23,996 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:23:24,696 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:23:42,328 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:23:43,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.203e+02 1.629e+03 1.973e+03 2.659e+03 6.373e+03, threshold=3.946e+03, percent-clipped=7.0
+2023-03-01 18:23:46,478 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:23:50,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6910, 2.0191, 1.7887, 1.7261], device='cuda:0'), covar=tensor([0.1494, 0.1805, 0.1213, 0.1034], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0790, 0.0733, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:23:50,530 INFO [train.py:968] (0/2) Epoch 3, batch 25500, giga_loss[loss=0.3391, simple_loss=0.4028, pruned_loss=0.1377, over 28961.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.4055, pruned_loss=0.1525, over 5683177.67 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.4115, pruned_loss=0.1593, over 5689843.94 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4055, pruned_loss=0.152, over 5686335.78 frames. ], batch size: 213, lr: 9.68e-03, grad_scale: 4.0
+2023-03-01 18:24:15,861 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:24:39,753 INFO [train.py:968] (0/2) Epoch 3, batch 25550, giga_loss[loss=0.3581, simple_loss=0.3985, pruned_loss=0.1589, over 28789.00 frames. ], tot_loss[loss=0.3577, simple_loss=0.4071, pruned_loss=0.1542, over 5673942.73 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4116, pruned_loss=0.1594, over 5682932.82 frames. ], giga_tot_loss[loss=0.357, simple_loss=0.4069, pruned_loss=0.1536, over 5682484.58 frames. ], batch size: 99, lr: 9.67e-03, grad_scale: 4.0
+2023-03-01 18:24:40,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5436, 1.7992, 1.6632, 1.7264], device='cuda:0'), covar=tensor([0.0901, 0.1170, 0.0811, 0.0686], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0790, 0.0732, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:24:44,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2563, 3.9516, 3.9166, 1.6621], device='cuda:0'), covar=tensor([0.0491, 0.0430, 0.0879, 0.2090], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0630, 0.0822, 0.0586], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:24:50,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:24:51,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5594, 1.3200, 1.2088, 1.1327], device='cuda:0'), covar=tensor([0.0419, 0.0365, 0.0700, 0.0683], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0486, 0.0525, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 18:25:16,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.652e+02 1.635e+03 2.062e+03 3.075e+03 6.858e+03, threshold=4.125e+03, percent-clipped=10.0
+2023-03-01 18:25:25,449 INFO [train.py:968] (0/2) Epoch 3, batch 25600, giga_loss[loss=0.4201, simple_loss=0.4513, pruned_loss=0.1945, over 28289.00 frames. ], tot_loss[loss=0.3612, simple_loss=0.4096, pruned_loss=0.1564, over 5682053.57 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4115, pruned_loss=0.1594, over 5689698.28 frames. ], giga_tot_loss[loss=0.3604, simple_loss=0.4093, pruned_loss=0.1557, over 5682552.38 frames. ], batch size: 368, lr: 9.67e-03, grad_scale: 8.0
+2023-03-01 18:26:13,634 INFO [train.py:968] (0/2) Epoch 3, batch 25650, giga_loss[loss=0.3254, simple_loss=0.3777, pruned_loss=0.1366, over 28970.00 frames. ], tot_loss[loss=0.3645, simple_loss=0.411, pruned_loss=0.159, over 5672876.37 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.4115, pruned_loss=0.1595, over 5685093.59 frames. ], giga_tot_loss[loss=0.3636, simple_loss=0.4107, pruned_loss=0.1583, over 5676497.05 frames. ], batch size: 106, lr: 9.67e-03, grad_scale: 4.0
+2023-03-01 18:26:24,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2409, 1.3519, 1.2428, 1.4053], device='cuda:0'), covar=tensor([0.2053, 0.2028, 0.1835, 0.1940], device='cuda:0'), in_proj_covar=tensor([0.1035, 0.0832, 0.0926, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 18:26:31,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4022, 1.3940, 5.2167, 3.6483], device='cuda:0'), covar=tensor([0.1497, 0.1942, 0.0243, 0.0444], device='cuda:0'), in_proj_covar=tensor([0.0538, 0.0504, 0.0693, 0.0555], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 18:27:00,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.104e+03 1.840e+03 2.366e+03 3.065e+03 7.656e+03, threshold=4.732e+03, percent-clipped=11.0
+2023-03-01 18:27:09,362 INFO [train.py:968] (0/2) Epoch 3, batch 25700, giga_loss[loss=0.3219, simple_loss=0.3774, pruned_loss=0.1332, over 28739.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4134, pruned_loss=0.1628, over 5670442.29 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4118, pruned_loss=0.1598, over 5688441.55 frames. ], giga_tot_loss[loss=0.3686, simple_loss=0.413, pruned_loss=0.1621, over 5670406.31 frames. ], batch size: 60, lr: 9.67e-03, grad_scale: 4.0
+2023-03-01 18:27:16,090 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:27:18,964 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-116000.pt
+2023-03-01 18:27:19,643 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:27:29,199 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7650, 4.4657, 4.4449, 1.9913], device='cuda:0'), covar=tensor([0.0388, 0.0342, 0.0731, 0.1886], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0622, 0.0815, 0.0574], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:27:47,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:27:58,082 INFO [train.py:968] (0/2) Epoch 3, batch 25750, giga_loss[loss=0.3129, simple_loss=0.3754, pruned_loss=0.1252, over 29018.00 frames. ], tot_loss[loss=0.3712, simple_loss=0.4141, pruned_loss=0.1641, over 5682710.13 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4113, pruned_loss=0.1595, over 5694833.53 frames. ], giga_tot_loss[loss=0.371, simple_loss=0.4143, pruned_loss=0.1638, over 5676389.39 frames. ], batch size: 155, lr: 9.67e-03, grad_scale: 4.0
+2023-03-01 18:28:35,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.961e+02 1.633e+03 2.104e+03 2.776e+03 6.343e+03, threshold=4.209e+03, percent-clipped=6.0
+2023-03-01 18:28:43,446 INFO [train.py:968] (0/2) Epoch 3, batch 25800, giga_loss[loss=0.3632, simple_loss=0.4116, pruned_loss=0.1574, over 28863.00 frames. ], tot_loss[loss=0.3673, simple_loss=0.4114, pruned_loss=0.1617, over 5677856.44 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.4114, pruned_loss=0.1596, over 5698375.14 frames. ], giga_tot_loss[loss=0.3671, simple_loss=0.4114, pruned_loss=0.1614, over 5669335.02 frames. ], batch size: 145, lr: 9.66e-03, grad_scale: 4.0
+2023-03-01 18:29:01,712 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-01 18:29:14,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1357, 1.2888, 4.4522, 3.4467], device='cuda:0'), covar=tensor([0.1621, 0.2111, 0.0319, 0.0604], device='cuda:0'), in_proj_covar=tensor([0.0532, 0.0502, 0.0685, 0.0550], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 18:29:28,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:29:29,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:29:31,529 INFO [train.py:968] (0/2) Epoch 3, batch 25850, giga_loss[loss=0.3652, simple_loss=0.4192, pruned_loss=0.1556, over 28716.00 frames. ], tot_loss[loss=0.3652, simple_loss=0.4103, pruned_loss=0.1601, over 5683365.68 frames. ], libri_tot_loss[loss=0.3649, simple_loss=0.411, pruned_loss=0.1594, over 5705758.05 frames. ], giga_tot_loss[loss=0.3655, simple_loss=0.4106, pruned_loss=0.1602, over 5668908.69 frames. ], batch size: 242, lr: 9.66e-03, grad_scale: 4.0
+2023-03-01 18:30:02,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6759, 1.8463, 1.6763, 1.7471], device='cuda:0'), covar=tensor([0.1136, 0.1485, 0.1005, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0799, 0.0735, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:30:12,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.722e+02 1.569e+03 1.883e+03 2.414e+03 9.277e+03, threshold=3.766e+03, percent-clipped=4.0
+2023-03-01 18:30:18,004 INFO [train.py:968] (0/2) Epoch 3, batch 25900, libri_loss[loss=0.331, simple_loss=0.3858, pruned_loss=0.1381, over 29620.00 frames. ], tot_loss[loss=0.3601, simple_loss=0.4076, pruned_loss=0.1563, over 5682466.44 frames. ], libri_tot_loss[loss=0.3648, simple_loss=0.411, pruned_loss=0.1593, over 5708960.42 frames. ], giga_tot_loss[loss=0.3604, simple_loss=0.4079, pruned_loss=0.1564, over 5667783.16 frames. ], batch size: 74, lr: 9.66e-03, grad_scale: 4.0
+2023-03-01 18:30:58,821 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-01 18:31:06,448 INFO [train.py:968] (0/2) Epoch 3, batch 25950, giga_loss[loss=0.3562, simple_loss=0.3988, pruned_loss=0.1568, over 29050.00 frames. ], tot_loss[loss=0.3577, simple_loss=0.4055, pruned_loss=0.1549, over 5665963.76 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4115, pruned_loss=0.1598, over 5698851.39 frames. ], giga_tot_loss[loss=0.357, simple_loss=0.4052, pruned_loss=0.1544, over 5663210.48 frames. ], batch size: 106, lr: 9.66e-03, grad_scale: 4.0
+2023-03-01 18:31:43,894 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:31:45,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:31:45,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3821, 1.3429, 1.2217, 1.5991], device='cuda:0'), covar=tensor([0.1915, 0.1808, 0.1684, 0.1758], device='cuda:0'), in_proj_covar=tensor([0.1044, 0.0830, 0.0928, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 18:31:46,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.530e+02 1.621e+03 2.013e+03 3.097e+03 7.795e+03, threshold=4.026e+03, percent-clipped=9.0
+2023-03-01 18:31:46,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:31:47,440 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116285.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 18:31:47,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:31:53,781 INFO [train.py:968] (0/2) Epoch 3, batch 26000, giga_loss[loss=0.3801, simple_loss=0.4129, pruned_loss=0.1736, over 28638.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.4041, pruned_loss=0.1543, over 5667801.11 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4122, pruned_loss=0.1605, over 5697520.16 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.403, pruned_loss=0.1532, over 5665650.25 frames. ], batch size: 307, lr: 9.66e-03, grad_scale: 8.0
+2023-03-01 18:32:06,374 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 18:32:14,142 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:32:14,856 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:32:44,010 INFO [train.py:968] (0/2) Epoch 3, batch 26050, giga_loss[loss=0.39, simple_loss=0.425, pruned_loss=0.1775, over 27894.00 frames. ], tot_loss[loss=0.36, simple_loss=0.4057, pruned_loss=0.1571, over 5649895.97 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4122, pruned_loss=0.1606, over 5697276.02 frames. ], giga_tot_loss[loss=0.3582, simple_loss=0.4046, pruned_loss=0.1559, over 5647542.05 frames. ], batch size: 412, lr: 9.65e-03, grad_scale: 4.0
+2023-03-01 18:33:26,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.129e+02 1.732e+03 2.326e+03 3.189e+03 6.977e+03, threshold=4.653e+03, percent-clipped=11.0
+2023-03-01 18:33:31,531 INFO [train.py:968] (0/2) Epoch 3, batch 26100, giga_loss[loss=0.3691, simple_loss=0.4179, pruned_loss=0.1602, over 28777.00 frames. ], tot_loss[loss=0.3639, simple_loss=0.4087, pruned_loss=0.1595, over 5655844.36 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4122, pruned_loss=0.1608, over 5701501.65 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.4076, pruned_loss=0.1583, over 5648962.65 frames. ], batch size: 262, lr: 9.65e-03, grad_scale: 4.0
+2023-03-01 18:33:45,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:34:18,675 INFO [train.py:968] (0/2) Epoch 3, batch 26150, giga_loss[loss=0.3471, simple_loss=0.4039, pruned_loss=0.1451, over 28764.00 frames. ], tot_loss[loss=0.3674, simple_loss=0.4134, pruned_loss=0.1607, over 5666738.25 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.412, pruned_loss=0.1608, over 5703985.91 frames. ], giga_tot_loss[loss=0.3661, simple_loss=0.4127, pruned_loss=0.1597, over 5658082.72 frames. ], batch size: 284, lr: 9.65e-03, grad_scale: 4.0
+2023-03-01 18:35:01,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.052e+02 1.344e+03 1.780e+03 2.519e+03 8.717e+03, threshold=3.559e+03, percent-clipped=3.0
+2023-03-01 18:35:07,727 INFO [train.py:968] (0/2) Epoch 3, batch 26200, giga_loss[loss=0.3441, simple_loss=0.4001, pruned_loss=0.144, over 28893.00 frames. ], tot_loss[loss=0.3673, simple_loss=0.4156, pruned_loss=0.1596, over 5670242.97 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4122, pruned_loss=0.161, over 5706949.74 frames. ], giga_tot_loss[loss=0.3661, simple_loss=0.4149, pruned_loss=0.1586, over 5660159.27 frames. ], batch size: 106, lr: 9.65e-03, grad_scale: 4.0
+2023-03-01 18:35:15,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:35:27,527 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:35:59,928 INFO [train.py:968] (0/2) Epoch 3, batch 26250, giga_loss[loss=0.3634, simple_loss=0.4222, pruned_loss=0.1523, over 28730.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4174, pruned_loss=0.1609, over 5666632.13 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4122, pruned_loss=0.161, over 5708684.80 frames. ], giga_tot_loss[loss=0.3686, simple_loss=0.4169, pruned_loss=0.1601, over 5656464.49 frames. ], batch size: 119, lr: 9.65e-03, grad_scale: 4.0
+2023-03-01 18:36:40,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.668e+02 1.646e+03 2.276e+03 2.777e+03 7.765e+03, threshold=4.551e+03, percent-clipped=15.0
+2023-03-01 18:36:41,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4094, 3.0999, 3.1718, 1.8738], device='cuda:0'), covar=tensor([0.0574, 0.0524, 0.0887, 0.1644], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0636, 0.0815, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:36:45,549 INFO [train.py:968] (0/2) Epoch 3, batch 26300, giga_loss[loss=0.3843, simple_loss=0.4289, pruned_loss=0.1698, over 28261.00 frames. ], tot_loss[loss=0.3718, simple_loss=0.4186, pruned_loss=0.1625, over 5661048.75 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4121, pruned_loss=0.1609, over 5710933.64 frames. ], giga_tot_loss[loss=0.3713, simple_loss=0.4185, pruned_loss=0.1621, over 5649527.48 frames. ], batch size: 368, lr: 9.64e-03, grad_scale: 4.0
+2023-03-01 18:36:59,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4181, 1.6257, 1.1285, 0.8813], device='cuda:0'), covar=tensor([0.0840, 0.0610, 0.0559, 0.0735], device='cuda:0'), in_proj_covar=tensor([0.1221, 0.0964, 0.0987, 0.1052], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 18:37:36,928 INFO [train.py:968] (0/2) Epoch 3, batch 26350, giga_loss[loss=0.3497, simple_loss=0.4072, pruned_loss=0.1461, over 28894.00 frames. ], tot_loss[loss=0.372, simple_loss=0.418, pruned_loss=0.163, over 5654859.67 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4121, pruned_loss=0.1608, over 5713788.36 frames. ], giga_tot_loss[loss=0.3717, simple_loss=0.418, pruned_loss=0.1627, over 5642309.33 frames. ], batch size: 174, lr: 9.64e-03, grad_scale: 2.0
+2023-03-01 18:37:56,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116660.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 18:38:22,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2322, 1.7487, 1.3234, 0.5563], device='cuda:0'), covar=tensor([0.1180, 0.0764, 0.1315, 0.1562], device='cuda:0'), in_proj_covar=tensor([0.1280, 0.1219, 0.1305, 0.1111], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 18:38:25,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.896e+02 1.549e+03 1.905e+03 2.900e+03 1.217e+04, threshold=3.810e+03, percent-clipped=10.0
+2023-03-01 18:38:29,608 INFO [train.py:968] (0/2) Epoch 3, batch 26400, giga_loss[loss=0.4447, simple_loss=0.4714, pruned_loss=0.209, over 28587.00 frames. ], tot_loss[loss=0.3723, simple_loss=0.4175, pruned_loss=0.1635, over 5649978.43 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4121, pruned_loss=0.1608, over 5712827.42 frames. ], giga_tot_loss[loss=0.3721, simple_loss=0.4175, pruned_loss=0.1633, over 5640785.89 frames. ], batch size: 307, lr: 9.64e-03, grad_scale: 4.0
+2023-03-01 18:38:35,410 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-01 18:38:53,597 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:38:54,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4738, 1.5049, 1.3438, 1.7960], device='cuda:0'), covar=tensor([0.2191, 0.2034, 0.1896, 0.1913], device='cuda:0'), in_proj_covar=tensor([0.1044, 0.0839, 0.0934, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 18:39:18,027 INFO [train.py:968] (0/2) Epoch 3, batch 26450, libri_loss[loss=0.3964, simple_loss=0.4408, pruned_loss=0.176, over 29129.00 frames. ], tot_loss[loss=0.3706, simple_loss=0.4157, pruned_loss=0.1628, over 5648549.07 frames. ], libri_tot_loss[loss=0.3679, simple_loss=0.413, pruned_loss=0.1614, over 5703381.16 frames. ], giga_tot_loss[loss=0.3697, simple_loss=0.415, pruned_loss=0.1621, over 5646972.29 frames. ], batch size: 101, lr: 9.64e-03, grad_scale: 4.0
+2023-03-01 18:39:46,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3049, 2.5463, 1.4305, 1.3113], device='cuda:0'), covar=tensor([0.0893, 0.0511, 0.0786, 0.1403], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0463, 0.0316, 0.0402], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 18:40:01,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:40:01,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2168, 1.8205, 1.8103, 1.7638], device='cuda:0'), covar=tensor([0.0966, 0.1825, 0.1449, 0.1472], device='cuda:0'), in_proj_covar=tensor([0.0450, 0.0778, 0.0635, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 18:40:03,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1349, 1.7165, 1.3370, 1.4054], device='cuda:0'), covar=tensor([0.0604, 0.0772, 0.1068, 0.1042], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0488, 0.0528, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 18:40:06,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.582e+03 2.240e+03 3.080e+03 9.364e+03, threshold=4.480e+03, percent-clipped=14.0
+2023-03-01 18:40:10,181 INFO [train.py:968] (0/2) Epoch 3, batch 26500, libri_loss[loss=0.3402, simple_loss=0.3871, pruned_loss=0.1467, over 29400.00 frames. ], tot_loss[loss=0.3686, simple_loss=0.4135, pruned_loss=0.1618, over 5646694.04 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4128, pruned_loss=0.1612, over 5706639.59 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.4132, pruned_loss=0.1616, over 5641362.42 frames. ], batch size: 67, lr: 9.64e-03, grad_scale: 4.0
+2023-03-01 18:40:21,336 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116803.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 18:40:23,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116806.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 18:40:52,347 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116835.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 18:40:56,470 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:40:58,498 INFO [train.py:968] (0/2) Epoch 3, batch 26550, giga_loss[loss=0.4415, simple_loss=0.4634, pruned_loss=0.2098, over 27743.00 frames. ], tot_loss[loss=0.3693, simple_loss=0.4137, pruned_loss=0.1624, over 5641825.10 frames. ], libri_tot_loss[loss=0.3672, simple_loss=0.4123, pruned_loss=0.161, over 5702414.70 frames. ], giga_tot_loss[loss=0.3693, simple_loss=0.4139, pruned_loss=0.1623, over 5639503.98 frames. ], batch size: 472, lr: 9.63e-03, grad_scale: 4.0
+2023-03-01 18:41:28,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:41:38,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:41:40,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.786e+03 2.334e+03 3.492e+03 8.302e+03, threshold=4.668e+03, percent-clipped=11.0
+2023-03-01 18:41:44,596 INFO [train.py:968] (0/2) Epoch 3, batch 26600, giga_loss[loss=0.3443, simple_loss=0.3944, pruned_loss=0.1471, over 28869.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4128, pruned_loss=0.1617, over 5655573.43 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4125, pruned_loss=0.1611, over 5705258.60 frames. ], giga_tot_loss[loss=0.368, simple_loss=0.4128, pruned_loss=0.1616, over 5650342.94 frames. ], batch size: 285, lr: 9.63e-03, grad_scale: 4.0
+2023-03-01 18:41:49,716 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-01 18:41:51,738 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:42:18,031 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:42:20,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:42:32,457 INFO [train.py:968] (0/2) Epoch 3, batch 26650, giga_loss[loss=0.4064, simple_loss=0.4376, pruned_loss=0.1876, over 27672.00 frames. ], tot_loss[loss=0.366, simple_loss=0.411, pruned_loss=0.1605, over 5669933.23 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4128, pruned_loss=0.1613, over 5708936.31 frames. ], giga_tot_loss[loss=0.3656, simple_loss=0.4106, pruned_loss=0.1603, over 5661642.60 frames. ], batch size: 472, lr: 9.63e-03, grad_scale: 4.0
+2023-03-01 18:42:39,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6087, 1.8670, 1.6991, 1.6845], device='cuda:0'), covar=tensor([0.1511, 0.1953, 0.1248, 0.1075], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0811, 0.0750, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:42:49,507 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:43:17,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.120e+02 1.585e+03 2.022e+03 2.549e+03 9.271e+03, threshold=4.044e+03, percent-clipped=5.0
+2023-03-01 18:43:22,048 INFO [train.py:968] (0/2) Epoch 3, batch 26700, giga_loss[loss=0.3416, simple_loss=0.4034, pruned_loss=0.1399, over 28724.00 frames. ], tot_loss[loss=0.3649, simple_loss=0.4098, pruned_loss=0.16, over 5671843.14 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4122, pruned_loss=0.1609, over 5712747.93 frames. ], giga_tot_loss[loss=0.3651, simple_loss=0.4101, pruned_loss=0.16, over 5660921.75 frames. ], batch size: 284, lr: 9.63e-03, grad_scale: 4.0
+2023-03-01 18:43:40,868 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:43:44,240 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:43:45,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5418, 4.1996, 4.2048, 1.9410], device='cuda:0'), covar=tensor([0.0419, 0.0406, 0.0779, 0.1817], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0636, 0.0817, 0.0578], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:43:46,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:43:56,728 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:43:58,585 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:44:08,113 INFO [train.py:968] (0/2) Epoch 3, batch 26750, giga_loss[loss=0.3514, simple_loss=0.4108, pruned_loss=0.146, over 28847.00 frames. ], tot_loss[loss=0.3654, simple_loss=0.4113, pruned_loss=0.1597, over 5673410.37 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4121, pruned_loss=0.1609, over 5715927.46 frames. ], giga_tot_loss[loss=0.3656, simple_loss=0.4116, pruned_loss=0.1598, over 5661235.45 frames. ], batch size: 174, lr: 9.63e-03, grad_scale: 4.0
+2023-03-01 18:44:14,957 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:44:25,518 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:44:35,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3415, 1.3454, 1.2536, 1.4230], device='cuda:0'), covar=tensor([0.2094, 0.2075, 0.1836, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.1053, 0.0849, 0.0947, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 18:44:50,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.275e+02 1.606e+03 1.967e+03 2.736e+03 7.474e+03, threshold=3.933e+03, percent-clipped=6.0
+2023-03-01 18:44:57,415 INFO [train.py:968] (0/2) Epoch 3, batch 26800, giga_loss[loss=0.2848, simple_loss=0.3543, pruned_loss=0.1076, over 28780.00 frames. ], tot_loss[loss=0.3667, simple_loss=0.4126, pruned_loss=0.1604, over 5663174.13 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4122, pruned_loss=0.1608, over 5711290.94 frames. ], giga_tot_loss[loss=0.3669, simple_loss=0.4127, pruned_loss=0.1605, over 5656285.21 frames. ], batch size: 99, lr: 9.62e-03, grad_scale: 8.0
+2023-03-01 18:44:57,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:45:05,475 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:45:12,238 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-01 18:45:44,165 INFO [train.py:968] (0/2) Epoch 3, batch 26850, giga_loss[loss=0.3695, simple_loss=0.4406, pruned_loss=0.1492, over 29004.00 frames. ], tot_loss[loss=0.3683, simple_loss=0.4133, pruned_loss=0.1617, over 5655888.68 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4126, pruned_loss=0.1612, over 5703359.70 frames. ], giga_tot_loss[loss=0.3679, simple_loss=0.4131, pruned_loss=0.1614, over 5656661.79 frames. ], batch size: 106, lr: 9.62e-03, grad_scale: 8.0
+2023-03-01 18:46:16,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2778, 1.9350, 1.3857, 0.5800], device='cuda:0'), covar=tensor([0.1501, 0.0830, 0.1322, 0.1740], device='cuda:0'), in_proj_covar=tensor([0.1281, 0.1221, 0.1284, 0.1109], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 18:46:27,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.402e+02 1.595e+03 1.972e+03 2.605e+03 4.250e+03, threshold=3.944e+03, percent-clipped=2.0
+2023-03-01 18:46:32,181 INFO [train.py:968] (0/2) Epoch 3, batch 26900, giga_loss[loss=0.3967, simple_loss=0.4194, pruned_loss=0.187, over 26706.00 frames. ], tot_loss[loss=0.3649, simple_loss=0.4136, pruned_loss=0.1581, over 5671589.09 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4128, pruned_loss=0.1613, over 5706452.66 frames. ], giga_tot_loss[loss=0.3644, simple_loss=0.4132, pruned_loss=0.1578, over 5668788.31 frames. ], batch size: 555, lr: 9.62e-03, grad_scale: 8.0
+2023-03-01 18:46:36,617 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:46:55,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:47:02,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3477, 1.8948, 1.3711, 0.7015], device='cuda:0'), covar=tensor([0.1655, 0.0885, 0.1255, 0.1874], device='cuda:0'), in_proj_covar=tensor([0.1263, 0.1195, 0.1271, 0.1096], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 18:47:13,806 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:47:16,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:47:18,399 INFO [train.py:968] (0/2) Epoch 3, batch 26950, giga_loss[loss=0.3705, simple_loss=0.4338, pruned_loss=0.1536, over 28732.00 frames. ], tot_loss[loss=0.3659, simple_loss=0.4151, pruned_loss=0.1583, over 5670364.16 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4126, pruned_loss=0.1613, over 5713070.78 frames. ], giga_tot_loss[loss=0.3654, simple_loss=0.415, pruned_loss=0.1579, over 5660523.64 frames. ], batch size: 119, lr: 9.62e-03, grad_scale: 4.0
+2023-03-01 18:47:43,025 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:47:49,580 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:48:00,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.086e+02 1.643e+03 2.029e+03 2.870e+03 5.652e+03, threshold=4.058e+03, percent-clipped=12.0
+2023-03-01 18:48:03,431 INFO [train.py:968] (0/2) Epoch 3, batch 27000, giga_loss[loss=0.3217, simple_loss=0.3904, pruned_loss=0.1265, over 28609.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.4166, pruned_loss=0.1581, over 5676332.62 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4124, pruned_loss=0.1613, over 5714622.10 frames. ], giga_tot_loss[loss=0.3661, simple_loss=0.4168, pruned_loss=0.1577, over 5666549.76 frames. ], batch size: 85, lr: 9.61e-03, grad_scale: 4.0
+2023-03-01 18:48:03,436 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 18:48:10,979 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3226, 1.9483, 1.6571, 1.6278], device='cuda:0'), covar=tensor([0.1536, 0.1907, 0.1362, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0794, 0.0737, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:48:12,287 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.252, simple_loss=0.3514, pruned_loss=0.07629, over 944034.00 frames. 
+2023-03-01 18:48:12,288 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 18:48:16,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6887, 2.0852, 1.8739, 1.7733], device='cuda:0'), covar=tensor([0.1316, 0.1637, 0.1082, 0.0939], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0797, 0.0738, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:48:58,240 INFO [train.py:968] (0/2) Epoch 3, batch 27050, giga_loss[loss=0.5149, simple_loss=0.5003, pruned_loss=0.2647, over 26548.00 frames. ], tot_loss[loss=0.3717, simple_loss=0.4194, pruned_loss=0.162, over 5675105.09 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4118, pruned_loss=0.161, over 5719354.21 frames. ], giga_tot_loss[loss=0.3721, simple_loss=0.4204, pruned_loss=0.1619, over 5661922.47 frames. ], batch size: 555, lr: 9.61e-03, grad_scale: 4.0
+2023-03-01 18:49:13,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:49:15,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:49:31,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3832, 4.0226, 4.1107, 1.9113], device='cuda:0'), covar=tensor([0.0480, 0.0434, 0.0862, 0.1882], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0645, 0.0830, 0.0580], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:49:42,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:49:43,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.998e+02 1.653e+03 2.201e+03 3.255e+03 9.208e+03, threshold=4.403e+03, percent-clipped=17.0
+2023-03-01 18:49:45,439 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:49:46,973 INFO [train.py:968] (0/2) Epoch 3, batch 27100, libri_loss[loss=0.3445, simple_loss=0.3993, pruned_loss=0.1449, over 29465.00 frames. ], tot_loss[loss=0.3725, simple_loss=0.4199, pruned_loss=0.1626, over 5685029.32 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4115, pruned_loss=0.1608, over 5721018.81 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.421, pruned_loss=0.1628, over 5672136.26 frames. ], batch size: 85, lr: 9.61e-03, grad_scale: 2.0
+2023-03-01 18:50:07,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2270, 1.8483, 1.3628, 1.7073], device='cuda:0'), covar=tensor([0.0531, 0.0617, 0.0829, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0478, 0.0518, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 18:50:14,327 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:50:16,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:50:38,099 INFO [train.py:968] (0/2) Epoch 3, batch 27150, giga_loss[loss=0.3362, simple_loss=0.3961, pruned_loss=0.1381, over 28701.00 frames. ], tot_loss[loss=0.3753, simple_loss=0.4211, pruned_loss=0.1648, over 5679878.67 frames. ], libri_tot_loss[loss=0.3663, simple_loss=0.4113, pruned_loss=0.1606, over 5721427.64 frames. ], giga_tot_loss[loss=0.3763, simple_loss=0.4223, pruned_loss=0.1652, over 5668506.90 frames. ], batch size: 242, lr: 9.61e-03, grad_scale: 2.0
+2023-03-01 18:50:46,437 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:51:07,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:51:23,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.528e+03 1.981e+03 2.608e+03 7.250e+03, threshold=3.962e+03, percent-clipped=5.0
+2023-03-01 18:51:25,731 INFO [train.py:968] (0/2) Epoch 3, batch 27200, giga_loss[loss=0.3406, simple_loss=0.4014, pruned_loss=0.1399, over 28661.00 frames. ], tot_loss[loss=0.372, simple_loss=0.4187, pruned_loss=0.1627, over 5673510.73 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4116, pruned_loss=0.1611, over 5713214.50 frames. ], giga_tot_loss[loss=0.3724, simple_loss=0.4197, pruned_loss=0.1625, over 5670892.84 frames. ], batch size: 262, lr: 9.61e-03, grad_scale: 4.0
+2023-03-01 18:51:38,116 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-01 18:51:45,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0825, 3.7935, 3.7833, 1.7626], device='cuda:0'), covar=tensor([0.0515, 0.0498, 0.0872, 0.2041], device='cuda:0'), in_proj_covar=tensor([0.0817, 0.0647, 0.0833, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:52:03,157 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:52:07,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:52:15,429 INFO [train.py:968] (0/2) Epoch 3, batch 27250, giga_loss[loss=0.3455, simple_loss=0.4171, pruned_loss=0.137, over 29010.00 frames. ], tot_loss[loss=0.3679, simple_loss=0.4173, pruned_loss=0.1592, over 5665915.24 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4115, pruned_loss=0.161, over 5711619.83 frames. ], giga_tot_loss[loss=0.3683, simple_loss=0.4183, pruned_loss=0.1592, over 5664972.49 frames. ], batch size: 136, lr: 9.60e-03, grad_scale: 4.0
+2023-03-01 18:52:33,447 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:52:43,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:52:57,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.934e+02 1.424e+03 1.816e+03 2.503e+03 4.228e+03, threshold=3.633e+03, percent-clipped=2.0
+2023-03-01 18:53:00,462 INFO [train.py:968] (0/2) Epoch 3, batch 27300, giga_loss[loss=0.3389, simple_loss=0.4062, pruned_loss=0.1358, over 28918.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4185, pruned_loss=0.1598, over 5656050.58 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.412, pruned_loss=0.1614, over 5707773.21 frames. ], giga_tot_loss[loss=0.3689, simple_loss=0.419, pruned_loss=0.1594, over 5657843.95 frames. ], batch size: 174, lr: 9.60e-03, grad_scale: 4.0
+2023-03-01 18:53:27,832 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:53:30,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:53:34,415 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:53:59,746 INFO [train.py:968] (0/2) Epoch 3, batch 27350, libri_loss[loss=0.3847, simple_loss=0.4285, pruned_loss=0.1705, over 25858.00 frames. ], tot_loss[loss=0.3709, simple_loss=0.4198, pruned_loss=0.161, over 5650933.98 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4117, pruned_loss=0.1612, over 5705240.96 frames. ], giga_tot_loss[loss=0.3712, simple_loss=0.4206, pruned_loss=0.1608, over 5653471.70 frames. ], batch size: 136, lr: 9.60e-03, grad_scale: 4.0
+2023-03-01 18:54:06,952 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:54:45,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.771e+02 1.656e+03 2.199e+03 3.155e+03 1.030e+04, threshold=4.398e+03, percent-clipped=13.0
+2023-03-01 18:54:47,808 INFO [train.py:968] (0/2) Epoch 3, batch 27400, giga_loss[loss=0.3253, simple_loss=0.3848, pruned_loss=0.133, over 28747.00 frames. ], tot_loss[loss=0.3711, simple_loss=0.4193, pruned_loss=0.1615, over 5647855.22 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4117, pruned_loss=0.1611, over 5699908.82 frames. ], giga_tot_loss[loss=0.3715, simple_loss=0.4202, pruned_loss=0.1615, over 5652992.86 frames. ], batch size: 262, lr: 9.60e-03, grad_scale: 4.0
+2023-03-01 18:55:08,718 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:55:12,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:55:17,594 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:55:31,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-01 18:55:36,333 INFO [train.py:968] (0/2) Epoch 3, batch 27450, giga_loss[loss=0.3867, simple_loss=0.4301, pruned_loss=0.1717, over 27831.00 frames. ], tot_loss[loss=0.3701, simple_loss=0.418, pruned_loss=0.1611, over 5656866.76 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4116, pruned_loss=0.1608, over 5705857.42 frames. ], giga_tot_loss[loss=0.371, simple_loss=0.4191, pruned_loss=0.1614, over 5653991.44 frames. ], batch size: 412, lr: 9.60e-03, grad_scale: 4.0
+2023-03-01 18:55:40,383 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:56:20,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.634e+02 1.633e+03 2.176e+03 2.658e+03 6.717e+03, threshold=4.351e+03, percent-clipped=6.0
+2023-03-01 18:56:23,163 INFO [train.py:968] (0/2) Epoch 3, batch 27500, giga_loss[loss=0.4404, simple_loss=0.4563, pruned_loss=0.2123, over 27855.00 frames. ], tot_loss[loss=0.3686, simple_loss=0.4163, pruned_loss=0.1604, over 5668944.84 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4117, pruned_loss=0.1608, over 5707424.81 frames. ], giga_tot_loss[loss=0.3693, simple_loss=0.4173, pruned_loss=0.1607, over 5663976.46 frames. ], batch size: 412, lr: 9.59e-03, grad_scale: 4.0
+2023-03-01 18:57:17,435 INFO [train.py:968] (0/2) Epoch 3, batch 27550, giga_loss[loss=0.3379, simple_loss=0.3959, pruned_loss=0.14, over 28715.00 frames. ], tot_loss[loss=0.3669, simple_loss=0.4143, pruned_loss=0.1598, over 5669160.78 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4122, pruned_loss=0.1614, over 5710781.93 frames. ], giga_tot_loss[loss=0.3667, simple_loss=0.4146, pruned_loss=0.1594, over 5661312.13 frames. ], batch size: 262, lr: 9.59e-03, grad_scale: 4.0
+2023-03-01 18:57:44,699 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 18:58:00,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.672e+02 1.520e+03 1.994e+03 2.861e+03 1.648e+04, threshold=3.988e+03, percent-clipped=10.0
+2023-03-01 18:58:03,745 INFO [train.py:968] (0/2) Epoch 3, batch 27600, giga_loss[loss=0.3489, simple_loss=0.4063, pruned_loss=0.1457, over 28694.00 frames. ], tot_loss[loss=0.3658, simple_loss=0.4128, pruned_loss=0.1594, over 5672155.99 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4123, pruned_loss=0.1613, over 5716307.57 frames. ], giga_tot_loss[loss=0.3656, simple_loss=0.4131, pruned_loss=0.1591, over 5659615.87 frames. ], batch size: 242, lr: 9.59e-03, grad_scale: 8.0
+2023-03-01 18:58:10,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6528, 1.7798, 1.6862, 1.6587], device='cuda:0'), covar=tensor([0.1212, 0.1780, 0.1059, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0801, 0.0739, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 18:58:48,614 INFO [train.py:968] (0/2) Epoch 3, batch 27650, giga_loss[loss=0.4009, simple_loss=0.4389, pruned_loss=0.1815, over 28979.00 frames. ], tot_loss[loss=0.3668, simple_loss=0.4131, pruned_loss=0.1603, over 5671833.51 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4119, pruned_loss=0.161, over 5712014.01 frames. ], giga_tot_loss[loss=0.3671, simple_loss=0.4136, pruned_loss=0.1603, over 5664026.17 frames. ], batch size: 106, lr: 9.59e-03, grad_scale: 8.0
+2023-03-01 18:59:34,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.434e+02 1.450e+03 2.004e+03 2.831e+03 6.105e+03, threshold=4.008e+03, percent-clipped=7.0
+2023-03-01 18:59:34,885 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:59:36,382 INFO [train.py:968] (0/2) Epoch 3, batch 27700, giga_loss[loss=0.3595, simple_loss=0.4033, pruned_loss=0.1579, over 27555.00 frames. ], tot_loss[loss=0.3629, simple_loss=0.4105, pruned_loss=0.1577, over 5667811.46 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4116, pruned_loss=0.1607, over 5714082.45 frames. ], giga_tot_loss[loss=0.3635, simple_loss=0.4112, pruned_loss=0.1579, over 5658632.47 frames. ], batch size: 472, lr: 9.59e-03, grad_scale: 4.0
+2023-03-01 18:59:44,659 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-118000.pt
+2023-03-01 19:00:20,324 INFO [train.py:968] (0/2) Epoch 3, batch 27750, giga_loss[loss=0.359, simple_loss=0.404, pruned_loss=0.157, over 27627.00 frames. ], tot_loss[loss=0.3582, simple_loss=0.4074, pruned_loss=0.1545, over 5676837.33 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4113, pruned_loss=0.1608, over 5720258.14 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.4082, pruned_loss=0.1544, over 5662103.29 frames. ], batch size: 472, lr: 9.58e-03, grad_scale: 4.0
+2023-03-01 19:00:31,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1638, 1.2294, 1.0002, 1.3940], device='cuda:0'), covar=tensor([0.0840, 0.0397, 0.0397, 0.0907], device='cuda:0'), in_proj_covar=tensor([0.0230, 0.0151, 0.0154, 0.0276], device='cuda:0'), out_proj_covar=tensor([0.0042, 0.0031, 0.0028, 0.0048], device='cuda:0')
+2023-03-01 19:01:10,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.767e+02 1.300e+03 1.620e+03 2.339e+03 1.135e+04, threshold=3.241e+03, percent-clipped=9.0
+2023-03-01 19:01:11,421 INFO [train.py:968] (0/2) Epoch 3, batch 27800, giga_loss[loss=0.3259, simple_loss=0.3871, pruned_loss=0.1324, over 28563.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4062, pruned_loss=0.1535, over 5667288.53 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4119, pruned_loss=0.1613, over 5722179.50 frames. ], giga_tot_loss[loss=0.3559, simple_loss=0.4061, pruned_loss=0.1528, over 5652120.15 frames. ], batch size: 71, lr: 9.58e-03, grad_scale: 4.0
+2023-03-01 19:01:20,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:01:33,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3908, 1.8610, 1.3443, 0.6747], device='cuda:0'), covar=tensor([0.1925, 0.0930, 0.1365, 0.2120], device='cuda:0'), in_proj_covar=tensor([0.1268, 0.1207, 0.1285, 0.1103], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 19:01:56,751 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:01:59,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:02:05,224 INFO [train.py:968] (0/2) Epoch 3, batch 27850, libri_loss[loss=0.3902, simple_loss=0.4348, pruned_loss=0.1727, over 29665.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.4048, pruned_loss=0.1528, over 5667438.80 frames. ], libri_tot_loss[loss=0.368, simple_loss=0.4125, pruned_loss=0.1617, over 5723709.26 frames. ], giga_tot_loss[loss=0.3537, simple_loss=0.404, pruned_loss=0.1517, over 5653133.16 frames. ], batch size: 88, lr: 9.58e-03, grad_scale: 2.0
+2023-03-01 19:02:12,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1335, 1.9009, 1.3571, 1.5018], device='cuda:0'), covar=tensor([0.0614, 0.0703, 0.0959, 0.1045], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0489, 0.0526, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 19:02:28,644 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:02:40,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-01 19:02:56,095 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.716e+03 2.356e+03 3.596e+03 1.050e+04, threshold=4.712e+03, percent-clipped=31.0
+2023-03-01 19:02:56,689 INFO [train.py:968] (0/2) Epoch 3, batch 27900, giga_loss[loss=0.3065, simple_loss=0.3667, pruned_loss=0.1231, over 28922.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4012, pruned_loss=0.1514, over 5667385.68 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4123, pruned_loss=0.1614, over 5729206.51 frames. ], giga_tot_loss[loss=0.3507, simple_loss=0.4005, pruned_loss=0.1505, over 5649085.90 frames. ], batch size: 213, lr: 9.58e-03, grad_scale: 2.0
+2023-03-01 19:03:48,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5872, 2.9942, 1.5323, 1.4118], device='cuda:0'), covar=tensor([0.0894, 0.0329, 0.0846, 0.1373], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0465, 0.0316, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 19:03:48,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:03:49,379 INFO [train.py:968] (0/2) Epoch 3, batch 27950, giga_loss[loss=0.3436, simple_loss=0.4072, pruned_loss=0.14, over 28534.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4031, pruned_loss=0.1526, over 5665952.69 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4124, pruned_loss=0.1615, over 5730066.88 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.4022, pruned_loss=0.1517, over 5649771.91 frames. ], batch size: 60, lr: 9.58e-03, grad_scale: 2.0
+2023-03-01 19:03:51,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:04:21,808 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:04:39,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.812e+02 1.657e+03 2.034e+03 2.686e+03 7.681e+03, threshold=4.068e+03, percent-clipped=3.0
+2023-03-01 19:04:40,614 INFO [train.py:968] (0/2) Epoch 3, batch 28000, giga_loss[loss=0.3186, simple_loss=0.3799, pruned_loss=0.1287, over 28786.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4035, pruned_loss=0.1524, over 5652381.55 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4121, pruned_loss=0.1612, over 5729435.64 frames. ], giga_tot_loss[loss=0.3532, simple_loss=0.403, pruned_loss=0.1517, over 5639213.99 frames. ], batch size: 119, lr: 9.57e-03, grad_scale: 4.0
+2023-03-01 19:05:29,464 INFO [train.py:968] (0/2) Epoch 3, batch 28050, giga_loss[loss=0.3734, simple_loss=0.4167, pruned_loss=0.165, over 29003.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4028, pruned_loss=0.151, over 5661536.34 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4126, pruned_loss=0.1614, over 5731339.24 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4019, pruned_loss=0.1502, over 5648427.59 frames. ], batch size: 213, lr: 9.57e-03, grad_scale: 4.0
+2023-03-01 19:05:52,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4913, 1.4382, 1.3799, 1.4355], device='cuda:0'), covar=tensor([0.0937, 0.1384, 0.1567, 0.1137], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0789, 0.0637, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 19:06:10,934 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-01 19:06:19,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.019e+02 1.295e+03 1.641e+03 2.277e+03 7.324e+03, threshold=3.283e+03, percent-clipped=7.0
+2023-03-01 19:06:20,325 INFO [train.py:968] (0/2) Epoch 3, batch 28100, giga_loss[loss=0.3763, simple_loss=0.3984, pruned_loss=0.1771, over 23488.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4047, pruned_loss=0.1533, over 5654820.51 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4136, pruned_loss=0.162, over 5734613.43 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4029, pruned_loss=0.1519, over 5640196.36 frames. ], batch size: 705, lr: 9.57e-03, grad_scale: 4.0
+2023-03-01 19:06:58,572 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:07:04,996 INFO [train.py:968] (0/2) Epoch 3, batch 28150, giga_loss[loss=0.3552, simple_loss=0.4069, pruned_loss=0.1517, over 29116.00 frames. ], tot_loss[loss=0.356, simple_loss=0.4048, pruned_loss=0.1536, over 5659413.51 frames. ], libri_tot_loss[loss=0.3683, simple_loss=0.4131, pruned_loss=0.1617, over 5738095.16 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4036, pruned_loss=0.1527, over 5643246.71 frames. ], batch size: 128, lr: 9.57e-03, grad_scale: 4.0
+2023-03-01 19:07:52,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.734e+02 1.533e+03 1.961e+03 2.603e+03 1.109e+04, threshold=3.923e+03, percent-clipped=13.0
+2023-03-01 19:07:54,224 INFO [train.py:968] (0/2) Epoch 3, batch 28200, giga_loss[loss=0.3792, simple_loss=0.3984, pruned_loss=0.18, over 23561.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.4082, pruned_loss=0.1558, over 5641958.49 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4137, pruned_loss=0.1621, over 5721572.50 frames. ], giga_tot_loss[loss=0.3578, simple_loss=0.4066, pruned_loss=0.1545, over 5643681.18 frames. ], batch size: 705, lr: 9.57e-03, grad_scale: 4.0
+2023-03-01 19:08:42,968 INFO [train.py:968] (0/2) Epoch 3, batch 28250, giga_loss[loss=0.3814, simple_loss=0.4212, pruned_loss=0.1708, over 27908.00 frames. ], tot_loss[loss=0.3629, simple_loss=0.4106, pruned_loss=0.1576, over 5638930.74 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4144, pruned_loss=0.1628, over 5715400.32 frames. ], giga_tot_loss[loss=0.3601, simple_loss=0.4085, pruned_loss=0.1559, over 5643515.32 frames. ], batch size: 412, lr: 9.56e-03, grad_scale: 4.0
+2023-03-01 19:09:04,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:09:28,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.107e+02 1.516e+03 2.130e+03 2.871e+03 5.780e+03, threshold=4.261e+03, percent-clipped=8.0
+2023-03-01 19:09:30,003 INFO [train.py:968] (0/2) Epoch 3, batch 28300, giga_loss[loss=0.388, simple_loss=0.4279, pruned_loss=0.1741, over 27836.00 frames. ], tot_loss[loss=0.3648, simple_loss=0.4117, pruned_loss=0.159, over 5644954.31 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4143, pruned_loss=0.1628, over 5719707.52 frames. ], giga_tot_loss[loss=0.3624, simple_loss=0.4101, pruned_loss=0.1574, over 5641073.53 frames. ], batch size: 412, lr: 9.56e-03, grad_scale: 4.0
+2023-03-01 19:10:19,475 INFO [train.py:968] (0/2) Epoch 3, batch 28350, giga_loss[loss=0.3254, simple_loss=0.3835, pruned_loss=0.1336, over 28581.00 frames. ], tot_loss[loss=0.3666, simple_loss=0.4129, pruned_loss=0.1602, over 5648808.57 frames. ], libri_tot_loss[loss=0.3711, simple_loss=0.4152, pruned_loss=0.1635, over 5720570.89 frames. ], giga_tot_loss[loss=0.3635, simple_loss=0.4107, pruned_loss=0.1581, over 5642973.91 frames. ], batch size: 85, lr: 9.56e-03, grad_scale: 4.0
+2023-03-01 19:10:43,075 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:11:07,835 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:11:16,044 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.627e+03 2.504e+03 4.025e+03 8.832e+03, threshold=5.007e+03, percent-clipped=21.0
+2023-03-01 19:11:16,752 INFO [train.py:968] (0/2) Epoch 3, batch 28400, giga_loss[loss=0.3313, simple_loss=0.393, pruned_loss=0.1348, over 28768.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.412, pruned_loss=0.1576, over 5650733.50 frames. ], libri_tot_loss[loss=0.3715, simple_loss=0.4153, pruned_loss=0.1638, over 5720518.27 frames. ], giga_tot_loss[loss=0.3608, simple_loss=0.41, pruned_loss=0.1557, over 5645278.85 frames. ], batch size: 92, lr: 9.56e-03, grad_scale: 8.0
+2023-03-01 19:11:30,296 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-01 19:11:49,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:12:06,748 INFO [train.py:968] (0/2) Epoch 3, batch 28450, giga_loss[loss=0.3491, simple_loss=0.4015, pruned_loss=0.1484, over 29060.00 frames. ], tot_loss[loss=0.365, simple_loss=0.4124, pruned_loss=0.1588, over 5643809.90 frames. ], libri_tot_loss[loss=0.371, simple_loss=0.4149, pruned_loss=0.1636, over 5725545.06 frames. ], giga_tot_loss[loss=0.3629, simple_loss=0.4111, pruned_loss=0.1573, over 5632621.71 frames. ], batch size: 128, lr: 9.56e-03, grad_scale: 4.0
+2023-03-01 19:12:58,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.022e+02 1.752e+03 2.164e+03 3.078e+03 5.536e+03, threshold=4.328e+03, percent-clipped=2.0
+2023-03-01 19:12:58,182 INFO [train.py:968] (0/2) Epoch 3, batch 28500, giga_loss[loss=0.3622, simple_loss=0.4151, pruned_loss=0.1547, over 27989.00 frames. ], tot_loss[loss=0.3653, simple_loss=0.4122, pruned_loss=0.1592, over 5638431.36 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.4147, pruned_loss=0.1635, over 5723563.90 frames. ], giga_tot_loss[loss=0.3637, simple_loss=0.4114, pruned_loss=0.158, over 5629276.66 frames. ], batch size: 412, lr: 9.55e-03, grad_scale: 4.0
+2023-03-01 19:13:17,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:13:55,679 INFO [train.py:968] (0/2) Epoch 3, batch 28550, giga_loss[loss=0.4109, simple_loss=0.4272, pruned_loss=0.1974, over 23753.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.4122, pruned_loss=0.1604, over 5630107.12 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4145, pruned_loss=0.1635, over 5717301.36 frames. ], giga_tot_loss[loss=0.3652, simple_loss=0.4116, pruned_loss=0.1594, over 5624589.32 frames. ], batch size: 705, lr: 9.55e-03, grad_scale: 4.0
+2023-03-01 19:14:50,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.787e+02 1.554e+03 2.079e+03 2.811e+03 6.956e+03, threshold=4.158e+03, percent-clipped=6.0
+2023-03-01 19:14:50,975 INFO [train.py:968] (0/2) Epoch 3, batch 28600, giga_loss[loss=0.3622, simple_loss=0.4118, pruned_loss=0.1563, over 29085.00 frames. ], tot_loss[loss=0.3649, simple_loss=0.4103, pruned_loss=0.1597, over 5622324.12 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4145, pruned_loss=0.1634, over 5709114.54 frames. ], giga_tot_loss[loss=0.3638, simple_loss=0.4099, pruned_loss=0.1589, over 5623842.35 frames. ], batch size: 155, lr: 9.55e-03, grad_scale: 4.0
+2023-03-01 19:15:33,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:15:37,145 INFO [train.py:968] (0/2) Epoch 3, batch 28650, giga_loss[loss=0.4033, simple_loss=0.44, pruned_loss=0.1833, over 28732.00 frames. ], tot_loss[loss=0.3646, simple_loss=0.41, pruned_loss=0.1597, over 5636959.46 frames. ], libri_tot_loss[loss=0.3708, simple_loss=0.4144, pruned_loss=0.1636, over 5703007.85 frames. ], giga_tot_loss[loss=0.3636, simple_loss=0.4096, pruned_loss=0.1588, over 5642425.54 frames. ], batch size: 284, lr: 9.55e-03, grad_scale: 4.0
+2023-03-01 19:15:49,168 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:15:54,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:16:23,357 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:16:28,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.521e+02 1.475e+03 2.009e+03 2.424e+03 7.920e+03, threshold=4.019e+03, percent-clipped=7.0
+2023-03-01 19:16:28,366 INFO [train.py:968] (0/2) Epoch 3, batch 28700, libri_loss[loss=0.3564, simple_loss=0.4023, pruned_loss=0.1553, over 29562.00 frames. ], tot_loss[loss=0.3629, simple_loss=0.4087, pruned_loss=0.1586, over 5631883.86 frames. ], libri_tot_loss[loss=0.371, simple_loss=0.4147, pruned_loss=0.1636, over 5697236.62 frames. ], giga_tot_loss[loss=0.3617, simple_loss=0.408, pruned_loss=0.1577, over 5640148.25 frames. ], batch size: 79, lr: 9.55e-03, grad_scale: 4.0
+2023-03-01 19:16:59,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:17:07,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3210, 1.3893, 1.2146, 1.4403], device='cuda:0'), covar=tensor([0.2223, 0.2034, 0.1908, 0.2108], device='cuda:0'), in_proj_covar=tensor([0.1040, 0.0838, 0.0938, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 19:17:11,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:17:15,010 INFO [train.py:968] (0/2) Epoch 3, batch 28750, giga_loss[loss=0.3555, simple_loss=0.4068, pruned_loss=0.1522, over 28589.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.4089, pruned_loss=0.1582, over 5646326.51 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.4147, pruned_loss=0.1636, over 5701419.89 frames. ], giga_tot_loss[loss=0.3614, simple_loss=0.4082, pruned_loss=0.1573, over 5647251.54 frames. ], batch size: 307, lr: 9.54e-03, grad_scale: 2.0
+2023-03-01 19:17:30,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:17:53,241 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:17:55,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:18:02,422 INFO [train.py:968] (0/2) Epoch 3, batch 28800, giga_loss[loss=0.3664, simple_loss=0.4131, pruned_loss=0.1598, over 28897.00 frames. ], tot_loss[loss=0.3638, simple_loss=0.4098, pruned_loss=0.1589, over 5653117.08 frames. ], libri_tot_loss[loss=0.3714, simple_loss=0.4151, pruned_loss=0.1639, over 5702363.85 frames. ], giga_tot_loss[loss=0.3623, simple_loss=0.4088, pruned_loss=0.1579, over 5652297.86 frames. ], batch size: 106, lr: 9.54e-03, grad_scale: 4.0
+2023-03-01 19:18:03,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.660e+03 2.136e+03 3.262e+03 8.079e+03, threshold=4.272e+03, percent-clipped=11.0
+2023-03-01 19:18:08,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:18:22,166 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:18:27,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7374, 3.4128, 3.4320, 2.1215], device='cuda:0'), covar=tensor([0.0448, 0.0496, 0.0849, 0.1527], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0652, 0.0828, 0.0585], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 19:18:53,829 INFO [train.py:968] (0/2) Epoch 3, batch 28850, giga_loss[loss=0.3661, simple_loss=0.4087, pruned_loss=0.1618, over 28513.00 frames. ], tot_loss[loss=0.366, simple_loss=0.4114, pruned_loss=0.1603, over 5656697.98 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4149, pruned_loss=0.1637, over 5702946.93 frames. ], giga_tot_loss[loss=0.3648, simple_loss=0.4106, pruned_loss=0.1595, over 5654833.79 frames. ], batch size: 336, lr: 9.54e-03, grad_scale: 4.0
+2023-03-01 19:19:03,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4892, 2.2155, 1.6868, 0.6067], device='cuda:0'), covar=tensor([0.2098, 0.1056, 0.1420, 0.2294], device='cuda:0'), in_proj_covar=tensor([0.1301, 0.1249, 0.1288, 0.1121], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 19:19:16,058 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-03-01 19:19:33,652 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:19:36,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:19:43,416 INFO [train.py:968] (0/2) Epoch 3, batch 28900, giga_loss[loss=0.3263, simple_loss=0.3847, pruned_loss=0.134, over 28924.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.412, pruned_loss=0.1611, over 5668253.76 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4149, pruned_loss=0.1637, over 5705818.75 frames. ], giga_tot_loss[loss=0.3662, simple_loss=0.4114, pruned_loss=0.1605, over 5663615.96 frames. ], batch size: 227, lr: 9.54e-03, grad_scale: 4.0
+2023-03-01 19:19:44,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.311e+02 1.655e+03 2.311e+03 2.842e+03 5.089e+03, threshold=4.622e+03, percent-clipped=4.0
+2023-03-01 19:19:52,976 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:19:56,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:20:06,473 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:20:26,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:20:32,093 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-01 19:20:32,664 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:20:33,035 INFO [train.py:968] (0/2) Epoch 3, batch 28950, giga_loss[loss=0.3688, simple_loss=0.4109, pruned_loss=0.1633, over 28641.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.4114, pruned_loss=0.1608, over 5671877.90 frames. ], libri_tot_loss[loss=0.3711, simple_loss=0.4148, pruned_loss=0.1637, over 5709814.71 frames. ], giga_tot_loss[loss=0.3657, simple_loss=0.4109, pruned_loss=0.1603, over 5663791.15 frames. ], batch size: 307, lr: 9.54e-03, grad_scale: 4.0
+2023-03-01 19:20:35,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:21:04,630 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:21:27,254 INFO [train.py:968] (0/2) Epoch 3, batch 29000, giga_loss[loss=0.3634, simple_loss=0.4138, pruned_loss=0.1566, over 28656.00 frames. ], tot_loss[loss=0.3692, simple_loss=0.4137, pruned_loss=0.1623, over 5663144.28 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4149, pruned_loss=0.1637, over 5702640.49 frames. ], giga_tot_loss[loss=0.3684, simple_loss=0.4132, pruned_loss=0.1618, over 5662787.48 frames. ], batch size: 307, lr: 9.53e-03, grad_scale: 4.0
+2023-03-01 19:21:27,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.664e+03 2.259e+03 3.406e+03 7.493e+03, threshold=4.517e+03, percent-clipped=9.0
+2023-03-01 19:21:40,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-01 19:21:48,965 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4749, 2.6719, 1.3207, 1.3246], device='cuda:0'), covar=tensor([0.0894, 0.0443, 0.0879, 0.1405], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0468, 0.0315, 0.0396], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 19:22:12,332 INFO [train.py:968] (0/2) Epoch 3, batch 29050, giga_loss[loss=0.3296, simple_loss=0.4019, pruned_loss=0.1287, over 28901.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.414, pruned_loss=0.1621, over 5672577.26 frames. ], libri_tot_loss[loss=0.3713, simple_loss=0.4151, pruned_loss=0.1638, over 5707611.75 frames. ], giga_tot_loss[loss=0.3684, simple_loss=0.4134, pruned_loss=0.1617, over 5666772.71 frames. ], batch size: 174, lr: 9.53e-03, grad_scale: 4.0
+2023-03-01 19:22:34,940 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-01 19:23:01,592 INFO [train.py:968] (0/2) Epoch 3, batch 29100, giga_loss[loss=0.3888, simple_loss=0.4349, pruned_loss=0.1714, over 28941.00 frames. ], tot_loss[loss=0.3711, simple_loss=0.4153, pruned_loss=0.1635, over 5664133.40 frames. ], libri_tot_loss[loss=0.3718, simple_loss=0.4154, pruned_loss=0.1641, over 5698355.84 frames. ], giga_tot_loss[loss=0.3701, simple_loss=0.4145, pruned_loss=0.1628, over 5666718.64 frames. ], batch size: 186, lr: 9.53e-03, grad_scale: 4.0
+2023-03-01 19:23:02,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.810e+03 2.364e+03 3.067e+03 6.158e+03, threshold=4.729e+03, percent-clipped=9.0
+2023-03-01 19:23:12,567 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:23:44,735 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7027, 1.6768, 1.5795, 1.7195], device='cuda:0'), covar=tensor([0.1098, 0.1443, 0.1296, 0.1236], device='cuda:0'), in_proj_covar=tensor([0.0450, 0.0779, 0.0640, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 19:23:46,977 INFO [train.py:968] (0/2) Epoch 3, batch 29150, giga_loss[loss=0.3827, simple_loss=0.4266, pruned_loss=0.1694, over 28837.00 frames. ], tot_loss[loss=0.3744, simple_loss=0.4177, pruned_loss=0.1656, over 5662535.75 frames. ], libri_tot_loss[loss=0.3722, simple_loss=0.4159, pruned_loss=0.1643, over 5700040.58 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.4167, pruned_loss=0.1649, over 5661442.57 frames. ], batch size: 186, lr: 9.53e-03, grad_scale: 4.0
+2023-03-01 19:24:12,136 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:24:30,233 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:24:33,302 INFO [train.py:968] (0/2) Epoch 3, batch 29200, libri_loss[loss=0.3537, simple_loss=0.4156, pruned_loss=0.1459, over 29530.00 frames. ], tot_loss[loss=0.374, simple_loss=0.4173, pruned_loss=0.1654, over 5659032.55 frames. ], libri_tot_loss[loss=0.3723, simple_loss=0.416, pruned_loss=0.1643, over 5700993.21 frames. ], giga_tot_loss[loss=0.3731, simple_loss=0.4164, pruned_loss=0.1649, over 5656759.40 frames. ], batch size: 82, lr: 9.53e-03, grad_scale: 8.0
+2023-03-01 19:24:34,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.376e+02 1.602e+03 2.039e+03 2.830e+03 5.461e+03, threshold=4.077e+03, percent-clipped=1.0
+2023-03-01 19:24:42,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6014, 1.0959, 2.8654, 2.6988], device='cuda:0'), covar=tensor([0.1488, 0.1808, 0.0509, 0.0656], device='cuda:0'), in_proj_covar=tensor([0.0545, 0.0507, 0.0711, 0.0567], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 19:25:25,052 INFO [train.py:968] (0/2) Epoch 3, batch 29250, giga_loss[loss=0.3497, simple_loss=0.4068, pruned_loss=0.1463, over 28269.00 frames. ], tot_loss[loss=0.3712, simple_loss=0.4161, pruned_loss=0.1632, over 5651372.74 frames. ], libri_tot_loss[loss=0.3719, simple_loss=0.4157, pruned_loss=0.164, over 5703497.96 frames. ], giga_tot_loss[loss=0.3709, simple_loss=0.4157, pruned_loss=0.1631, over 5646472.37 frames. ], batch size: 368, lr: 9.52e-03, grad_scale: 4.0
+2023-03-01 19:25:26,357 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:25:30,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:25:57,679 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:26:14,916 INFO [train.py:968] (0/2) Epoch 3, batch 29300, libri_loss[loss=0.3318, simple_loss=0.3769, pruned_loss=0.1434, over 29559.00 frames. ], tot_loss[loss=0.3708, simple_loss=0.4161, pruned_loss=0.1628, over 5647841.32 frames. ], libri_tot_loss[loss=0.3721, simple_loss=0.4157, pruned_loss=0.1643, over 5707153.95 frames. ], giga_tot_loss[loss=0.3703, simple_loss=0.4158, pruned_loss=0.1624, over 5638781.30 frames. ], batch size: 78, lr: 9.52e-03, grad_scale: 4.0
+2023-03-01 19:26:16,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.735e+02 1.356e+03 1.908e+03 2.571e+03 6.593e+03, threshold=3.816e+03, percent-clipped=6.0
+2023-03-01 19:26:56,192 INFO [train.py:968] (0/2) Epoch 3, batch 29350, giga_loss[loss=0.3591, simple_loss=0.4083, pruned_loss=0.155, over 28682.00 frames. ], tot_loss[loss=0.3668, simple_loss=0.4136, pruned_loss=0.16, over 5660863.23 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4151, pruned_loss=0.1637, over 5711972.80 frames. ], giga_tot_loss[loss=0.3671, simple_loss=0.4139, pruned_loss=0.1601, over 5647955.98 frames. ], batch size: 262, lr: 9.52e-03, grad_scale: 4.0
+2023-03-01 19:27:43,782 INFO [train.py:968] (0/2) Epoch 3, batch 29400, giga_loss[loss=0.4813, simple_loss=0.4785, pruned_loss=0.2421, over 26578.00 frames. ], tot_loss[loss=0.3672, simple_loss=0.4136, pruned_loss=0.1604, over 5666780.08 frames. ], libri_tot_loss[loss=0.3716, simple_loss=0.4155, pruned_loss=0.1638, over 5715541.93 frames. ], giga_tot_loss[loss=0.367, simple_loss=0.4134, pruned_loss=0.1603, over 5652666.28 frames. ], batch size: 555, lr: 9.52e-03, grad_scale: 4.0
+2023-03-01 19:27:45,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.485e+03 1.840e+03 2.476e+03 5.681e+03, threshold=3.680e+03, percent-clipped=5.0
+2023-03-01 19:28:28,603 INFO [train.py:968] (0/2) Epoch 3, batch 29450, giga_loss[loss=0.3385, simple_loss=0.4023, pruned_loss=0.1374, over 29001.00 frames. ], tot_loss[loss=0.3673, simple_loss=0.4136, pruned_loss=0.1605, over 5663146.99 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.4149, pruned_loss=0.1634, over 5720324.37 frames. ], giga_tot_loss[loss=0.3677, simple_loss=0.4139, pruned_loss=0.1607, over 5646550.10 frames. ], batch size: 164, lr: 9.52e-03, grad_scale: 4.0
+2023-03-01 19:28:34,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9910, 1.3580, 3.8541, 3.1281], device='cuda:0'), covar=tensor([0.1593, 0.1971, 0.0375, 0.0585], device='cuda:0'), in_proj_covar=tensor([0.0554, 0.0513, 0.0715, 0.0572], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 19:29:21,276 INFO [train.py:968] (0/2) Epoch 3, batch 29500, libri_loss[loss=0.3167, simple_loss=0.371, pruned_loss=0.1312, over 29562.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4144, pruned_loss=0.161, over 5667244.73 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4151, pruned_loss=0.1636, over 5720149.20 frames. ], giga_tot_loss[loss=0.3681, simple_loss=0.4145, pruned_loss=0.1609, over 5653132.81 frames. ], batch size: 76, lr: 9.51e-03, grad_scale: 4.0
+2023-03-01 19:29:22,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.173e+02 1.384e+03 2.211e+03 3.441e+03 8.790e+03, threshold=4.423e+03, percent-clipped=21.0
+2023-03-01 19:29:48,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0855, 2.3824, 1.4775, 1.5063], device='cuda:0'), covar=tensor([0.0612, 0.0394, 0.0508, 0.0587], device='cuda:0'), in_proj_covar=tensor([0.1230, 0.0971, 0.0977, 0.1057], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 19:30:11,599 INFO [train.py:968] (0/2) Epoch 3, batch 29550, giga_loss[loss=0.3809, simple_loss=0.4276, pruned_loss=0.167, over 28839.00 frames. ], tot_loss[loss=0.367, simple_loss=0.4125, pruned_loss=0.1607, over 5655094.19 frames. ], libri_tot_loss[loss=0.3711, simple_loss=0.4149, pruned_loss=0.1636, over 5713411.23 frames. ], giga_tot_loss[loss=0.367, simple_loss=0.4127, pruned_loss=0.1606, over 5649096.23 frames. ], batch size: 145, lr: 9.51e-03, grad_scale: 4.0
+2023-03-01 19:30:14,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:30:26,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0041, 1.2167, 3.8945, 3.0319], device='cuda:0'), covar=tensor([0.1546, 0.1973, 0.0351, 0.0598], device='cuda:0'), in_proj_covar=tensor([0.0551, 0.0512, 0.0714, 0.0573], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 19:30:29,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:30:54,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8541, 1.7203, 1.2609, 1.5184], device='cuda:0'), covar=tensor([0.0606, 0.0630, 0.0960, 0.0898], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0480, 0.0530, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 19:30:58,435 INFO [train.py:968] (0/2) Epoch 3, batch 29600, giga_loss[loss=0.3805, simple_loss=0.4093, pruned_loss=0.1759, over 28785.00 frames. ], tot_loss[loss=0.3658, simple_loss=0.4117, pruned_loss=0.1599, over 5671576.48 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.415, pruned_loss=0.1637, over 5715990.77 frames. ], giga_tot_loss[loss=0.3656, simple_loss=0.4118, pruned_loss=0.1597, over 5663864.46 frames. ], batch size: 99, lr: 9.51e-03, grad_scale: 8.0
+2023-03-01 19:31:00,211 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.853e+02 1.462e+03 2.036e+03 2.776e+03 8.465e+03, threshold=4.072e+03, percent-clipped=8.0
+2023-03-01 19:31:14,299 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:31:46,146 INFO [train.py:968] (0/2) Epoch 3, batch 29650, giga_loss[loss=0.3337, simple_loss=0.3892, pruned_loss=0.1391, over 28626.00 frames. ], tot_loss[loss=0.3697, simple_loss=0.4146, pruned_loss=0.1624, over 5667608.86 frames. ], libri_tot_loss[loss=0.371, simple_loss=0.4149, pruned_loss=0.1635, over 5718028.06 frames. ], giga_tot_loss[loss=0.3697, simple_loss=0.4147, pruned_loss=0.1624, over 5659291.34 frames. ], batch size: 307, lr: 9.51e-03, grad_scale: 4.0
+2023-03-01 19:32:27,065 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.06 vs. limit=2.0
+2023-03-01 19:32:35,045 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:32:37,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:32:38,723 INFO [train.py:968] (0/2) Epoch 3, batch 29700, giga_loss[loss=0.4649, simple_loss=0.4668, pruned_loss=0.2315, over 26629.00 frames. ], tot_loss[loss=0.3712, simple_loss=0.4154, pruned_loss=0.1635, over 5640660.51 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.4148, pruned_loss=0.1635, over 5707170.45 frames. ], giga_tot_loss[loss=0.3713, simple_loss=0.4156, pruned_loss=0.1635, over 5641961.90 frames. ], batch size: 555, lr: 9.51e-03, grad_scale: 4.0
+2023-03-01 19:32:40,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.262e+02 1.567e+03 2.064e+03 2.835e+03 1.141e+04, threshold=4.128e+03, percent-clipped=11.0
+2023-03-01 19:32:44,665 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-120000.pt
+2023-03-01 19:32:48,351 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1028, 0.9274, 0.7551, 1.3141], device='cuda:0'), covar=tensor([0.0842, 0.0431, 0.0418, 0.0914], device='cuda:0'), in_proj_covar=tensor([0.0232, 0.0153, 0.0156, 0.0279], device='cuda:0'), out_proj_covar=tensor([0.0043, 0.0032, 0.0029, 0.0049], device='cuda:0')
+2023-03-01 19:32:49,989 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:32:52,876 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:33:02,403 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:33:18,023 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:33:22,116 INFO [train.py:968] (0/2) Epoch 3, batch 29750, giga_loss[loss=0.3518, simple_loss=0.408, pruned_loss=0.1478, over 28910.00 frames. ], tot_loss[loss=0.3708, simple_loss=0.4153, pruned_loss=0.1631, over 5651065.62 frames. ], libri_tot_loss[loss=0.371, simple_loss=0.4149, pruned_loss=0.1635, over 5709676.66 frames. ], giga_tot_loss[loss=0.3708, simple_loss=0.4155, pruned_loss=0.1631, over 5647995.50 frames. ], batch size: 186, lr: 9.50e-03, grad_scale: 4.0
+2023-03-01 19:33:43,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=120058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:34:04,615 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4506, 2.6321, 1.4870, 1.3770], device='cuda:0'), covar=tensor([0.0899, 0.0518, 0.0818, 0.1370], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0468, 0.0317, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 19:34:13,609 INFO [train.py:968] (0/2) Epoch 3, batch 29800, giga_loss[loss=0.3539, simple_loss=0.4137, pruned_loss=0.1471, over 28922.00 frames. ], tot_loss[loss=0.3693, simple_loss=0.4149, pruned_loss=0.1619, over 5657079.27 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.415, pruned_loss=0.1634, over 5711560.04 frames. ], giga_tot_loss[loss=0.3694, simple_loss=0.4149, pruned_loss=0.162, over 5652380.64 frames. ], batch size: 145, lr: 9.50e-03, grad_scale: 4.0
+2023-03-01 19:34:16,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.867e+02 1.578e+03 1.962e+03 2.478e+03 4.846e+03, threshold=3.924e+03, percent-clipped=4.0
+2023-03-01 19:34:22,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3910, 4.0976, 4.0683, 1.8256], device='cuda:0'), covar=tensor([0.0394, 0.0360, 0.0697, 0.2012], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0655, 0.0834, 0.0589], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 19:34:45,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8561, 2.3584, 2.1754, 2.0571], device='cuda:0'), covar=tensor([0.1636, 0.1741, 0.1135, 0.0958], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0821, 0.0756, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 19:35:03,660 INFO [train.py:968] (0/2) Epoch 3, batch 29850, giga_loss[loss=0.3245, simple_loss=0.3815, pruned_loss=0.1337, over 28910.00 frames. ], tot_loss[loss=0.3678, simple_loss=0.4141, pruned_loss=0.1607, over 5660332.27 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4153, pruned_loss=0.1635, over 5717131.59 frames. ], giga_tot_loss[loss=0.3675, simple_loss=0.4138, pruned_loss=0.1606, over 5649651.30 frames. ], batch size: 112, lr: 9.50e-03, grad_scale: 2.0
+2023-03-01 19:35:04,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-01 19:35:15,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3964, 1.5349, 1.0116, 1.1311], device='cuda:0'), covar=tensor([0.0609, 0.0648, 0.0518, 0.0607], device='cuda:0'), in_proj_covar=tensor([0.1244, 0.0995, 0.0987, 0.1078], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 19:35:50,037 INFO [train.py:968] (0/2) Epoch 3, batch 29900, giga_loss[loss=0.3779, simple_loss=0.4305, pruned_loss=0.1627, over 28921.00 frames. ], tot_loss[loss=0.367, simple_loss=0.4134, pruned_loss=0.1603, over 5667359.51 frames. ], libri_tot_loss[loss=0.3708, simple_loss=0.415, pruned_loss=0.1633, over 5721104.99 frames. ], giga_tot_loss[loss=0.3671, simple_loss=0.4134, pruned_loss=0.1604, over 5654505.42 frames. ], batch size: 213, lr: 9.50e-03, grad_scale: 2.0
+2023-03-01 19:35:54,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.449e+02 1.633e+03 2.189e+03 3.258e+03 1.018e+04, threshold=4.377e+03, percent-clipped=16.0
+2023-03-01 19:36:38,572 INFO [train.py:968] (0/2) Epoch 3, batch 29950, giga_loss[loss=0.3718, simple_loss=0.4216, pruned_loss=0.161, over 28609.00 frames. ], tot_loss[loss=0.3668, simple_loss=0.4129, pruned_loss=0.1604, over 5671322.74 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4149, pruned_loss=0.1632, over 5722502.07 frames. ], giga_tot_loss[loss=0.3669, simple_loss=0.4129, pruned_loss=0.1605, over 5658804.14 frames. ], batch size: 262, lr: 9.50e-03, grad_scale: 2.0
+2023-03-01 19:37:14,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:37:25,412 INFO [train.py:968] (0/2) Epoch 3, batch 30000, giga_loss[loss=0.3404, simple_loss=0.3932, pruned_loss=0.1438, over 28726.00 frames. ], tot_loss[loss=0.3607, simple_loss=0.4077, pruned_loss=0.1568, over 5678416.96 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4143, pruned_loss=0.1629, over 5725177.33 frames. ], giga_tot_loss[loss=0.3612, simple_loss=0.4081, pruned_loss=0.1571, over 5665369.09 frames. ], batch size: 284, lr: 9.49e-03, grad_scale: 4.0
+2023-03-01 19:37:25,417 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 19:37:34,312 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2555, simple_loss=0.3575, pruned_loss=0.07679, over 944034.00 frames. 
+2023-03-01 19:37:34,313 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 19:37:37,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.770e+03 2.449e+03 3.187e+03 9.176e+03, threshold=4.898e+03, percent-clipped=12.0
+2023-03-01 19:38:04,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.22 vs. limit=5.0
+2023-03-01 19:38:24,384 INFO [train.py:968] (0/2) Epoch 3, batch 30050, giga_loss[loss=0.2953, simple_loss=0.3563, pruned_loss=0.1172, over 28992.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.4033, pruned_loss=0.155, over 5656523.94 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4143, pruned_loss=0.1629, over 5720584.46 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4035, pruned_loss=0.155, over 5648888.48 frames. ], batch size: 128, lr: 9.49e-03, grad_scale: 4.0
+2023-03-01 19:39:08,614 INFO [train.py:968] (0/2) Epoch 3, batch 30100, giga_loss[loss=0.3386, simple_loss=0.3962, pruned_loss=0.1405, over 28675.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4011, pruned_loss=0.1535, over 5656530.89 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4142, pruned_loss=0.1629, over 5711429.01 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.4011, pruned_loss=0.1534, over 5658043.35 frames. ], batch size: 242, lr: 9.49e-03, grad_scale: 4.0
+2023-03-01 19:39:12,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.460e+03 1.790e+03 2.355e+03 5.546e+03, threshold=3.581e+03, percent-clipped=1.0
+2023-03-01 19:39:41,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:39:44,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:39:50,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:39:57,886 INFO [train.py:968] (0/2) Epoch 3, batch 30150, giga_loss[loss=0.4269, simple_loss=0.4392, pruned_loss=0.2073, over 26686.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.3997, pruned_loss=0.1531, over 5637355.42 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4137, pruned_loss=0.1627, over 5708085.67 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.3997, pruned_loss=0.1529, over 5638970.86 frames. ], batch size: 555, lr: 9.49e-03, grad_scale: 4.0
+2023-03-01 19:40:15,931 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:40:37,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9221, 1.1359, 3.9281, 3.2051], device='cuda:0'), covar=tensor([0.1647, 0.2135, 0.0387, 0.0564], device='cuda:0'), in_proj_covar=tensor([0.0553, 0.0520, 0.0736, 0.0580], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 19:40:46,616 INFO [train.py:968] (0/2) Epoch 3, batch 30200, libri_loss[loss=0.3881, simple_loss=0.4388, pruned_loss=0.1687, over 29530.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.3984, pruned_loss=0.1502, over 5631329.16 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4136, pruned_loss=0.1627, over 5693426.94 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.3981, pruned_loss=0.1497, over 5644480.86 frames. ], batch size: 89, lr: 9.49e-03, grad_scale: 4.0
+2023-03-01 19:40:52,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.832e+02 1.562e+03 1.878e+03 2.504e+03 5.754e+03, threshold=3.756e+03, percent-clipped=10.0
+2023-03-01 19:41:14,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3848, 2.8392, 1.3405, 1.2807], device='cuda:0'), covar=tensor([0.0821, 0.0406, 0.0881, 0.1312], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0464, 0.0313, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 19:41:40,854 INFO [train.py:968] (0/2) Epoch 3, batch 30250, giga_loss[loss=0.3417, simple_loss=0.3836, pruned_loss=0.1499, over 26693.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3945, pruned_loss=0.146, over 5630055.35 frames. ], libri_tot_loss[loss=0.3685, simple_loss=0.4126, pruned_loss=0.1622, over 5699953.82 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3948, pruned_loss=0.1457, over 5632959.95 frames. ], batch size: 555, lr: 9.48e-03, grad_scale: 4.0
+2023-03-01 19:42:04,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3262, 1.6804, 1.3277, 1.3337], device='cuda:0'), covar=tensor([0.0814, 0.0373, 0.0392, 0.0941], device='cuda:0'), in_proj_covar=tensor([0.0230, 0.0151, 0.0154, 0.0275], device='cuda:0'), out_proj_covar=tensor([0.0043, 0.0032, 0.0029, 0.0048], device='cuda:0')
+2023-03-01 19:42:18,533 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:42:20,551 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:42:22,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3382, 1.3512, 1.1426, 1.6991], device='cuda:0'), covar=tensor([0.2233, 0.1918, 0.1879, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.1054, 0.0831, 0.0950, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 19:42:30,440 INFO [train.py:968] (0/2) Epoch 3, batch 30300, giga_loss[loss=0.3565, simple_loss=0.4075, pruned_loss=0.1528, over 28563.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3916, pruned_loss=0.1423, over 5646977.56 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4118, pruned_loss=0.1617, over 5703597.64 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3919, pruned_loss=0.1419, over 5644152.15 frames. ], batch size: 307, lr: 9.48e-03, grad_scale: 4.0
+2023-03-01 19:42:33,519 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-01 19:42:34,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.535e+02 1.439e+03 1.956e+03 2.769e+03 7.284e+03, threshold=3.913e+03, percent-clipped=11.0
+2023-03-01 19:42:48,965 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:43:18,031 INFO [train.py:968] (0/2) Epoch 3, batch 30350, giga_loss[loss=0.3129, simple_loss=0.3759, pruned_loss=0.1249, over 27994.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3885, pruned_loss=0.1398, over 5633543.00 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4105, pruned_loss=0.1615, over 5689021.15 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3888, pruned_loss=0.1386, over 5642929.03 frames. ], batch size: 412, lr: 9.48e-03, grad_scale: 2.0
+2023-03-01 19:44:06,861 INFO [train.py:968] (0/2) Epoch 3, batch 30400, giga_loss[loss=0.3056, simple_loss=0.3649, pruned_loss=0.1232, over 27635.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3844, pruned_loss=0.1357, over 5626234.12 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4103, pruned_loss=0.1616, over 5674524.91 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3844, pruned_loss=0.1342, over 5646148.23 frames. ], batch size: 472, lr: 9.48e-03, grad_scale: 4.0
+2023-03-01 19:44:11,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.761e+02 1.334e+03 1.795e+03 2.601e+03 1.675e+04, threshold=3.590e+03, percent-clipped=14.0
+2023-03-01 19:44:44,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6705, 1.6114, 1.4997, 1.5592], device='cuda:0'), covar=tensor([0.1843, 0.2743, 0.1614, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0799, 0.0747, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 19:44:58,389 INFO [train.py:968] (0/2) Epoch 3, batch 30450, giga_loss[loss=0.2689, simple_loss=0.3529, pruned_loss=0.09242, over 28895.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.384, pruned_loss=0.1326, over 5649357.55 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4102, pruned_loss=0.1618, over 5676877.02 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3835, pruned_loss=0.1306, over 5662232.68 frames. ], batch size: 145, lr: 9.48e-03, grad_scale: 4.0
+2023-03-01 19:45:49,223 INFO [train.py:968] (0/2) Epoch 3, batch 30500, giga_loss[loss=0.3187, simple_loss=0.3803, pruned_loss=0.1286, over 28703.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3832, pruned_loss=0.1314, over 5657835.88 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4091, pruned_loss=0.1611, over 5683675.22 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.383, pruned_loss=0.1295, over 5661431.86 frames. ], batch size: 307, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:45:54,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.826e+02 1.392e+03 1.795e+03 2.375e+03 4.293e+03, threshold=3.590e+03, percent-clipped=5.0
+2023-03-01 19:46:38,516 INFO [train.py:968] (0/2) Epoch 3, batch 30550, giga_loss[loss=0.2794, simple_loss=0.3535, pruned_loss=0.1026, over 28819.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3808, pruned_loss=0.1295, over 5656241.33 frames. ], libri_tot_loss[loss=0.3644, simple_loss=0.4079, pruned_loss=0.1604, over 5678331.14 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3808, pruned_loss=0.1276, over 5662212.76 frames. ], batch size: 199, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:46:53,651 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 19:47:34,902 INFO [train.py:968] (0/2) Epoch 3, batch 30600, giga_loss[loss=0.2731, simple_loss=0.3526, pruned_loss=0.09677, over 28833.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3764, pruned_loss=0.1258, over 5658697.28 frames. ], libri_tot_loss[loss=0.3643, simple_loss=0.4077, pruned_loss=0.1604, over 5679507.48 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3765, pruned_loss=0.1241, over 5662182.25 frames. ], batch size: 145, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:47:39,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.582e+02 1.487e+03 1.873e+03 2.503e+03 5.816e+03, threshold=3.746e+03, percent-clipped=7.0
+2023-03-01 19:48:28,046 INFO [train.py:968] (0/2) Epoch 3, batch 30650, giga_loss[loss=0.3255, simple_loss=0.3896, pruned_loss=0.1307, over 28904.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.374, pruned_loss=0.1245, over 5649544.20 frames. ], libri_tot_loss[loss=0.3636, simple_loss=0.4071, pruned_loss=0.16, over 5680431.58 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3742, pruned_loss=0.123, over 5650980.67 frames. ], batch size: 227, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:49:15,383 INFO [train.py:968] (0/2) Epoch 3, batch 30700, libri_loss[loss=0.3103, simple_loss=0.3593, pruned_loss=0.1307, over 29538.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3744, pruned_loss=0.1243, over 5659682.53 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4063, pruned_loss=0.1595, over 5681001.39 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3746, pruned_loss=0.1229, over 5659566.78 frames. ], batch size: 79, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:49:20,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.661e+02 1.324e+03 1.791e+03 2.452e+03 5.653e+03, threshold=3.583e+03, percent-clipped=8.0
+2023-03-01 19:49:56,660 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 19:50:04,201 INFO [train.py:968] (0/2) Epoch 3, batch 30750, giga_loss[loss=0.251, simple_loss=0.3369, pruned_loss=0.08256, over 29009.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3738, pruned_loss=0.1239, over 5647837.89 frames. ], libri_tot_loss[loss=0.3631, simple_loss=0.4065, pruned_loss=0.1599, over 5671124.88 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3728, pruned_loss=0.1214, over 5656890.95 frames. ], batch size: 155, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:50:34,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5571, 2.1546, 1.7824, 1.6788], device='cuda:0'), covar=tensor([0.1457, 0.1650, 0.1172, 0.0996], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0787, 0.0737, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 19:50:34,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:50:55,648 INFO [train.py:968] (0/2) Epoch 3, batch 30800, giga_loss[loss=0.2553, simple_loss=0.326, pruned_loss=0.09232, over 27627.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3705, pruned_loss=0.1213, over 5645387.01 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.406, pruned_loss=0.1597, over 5667579.73 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3694, pruned_loss=0.1186, over 5654940.38 frames. ], batch size: 472, lr: 9.46e-03, grad_scale: 8.0
+2023-03-01 19:51:00,280 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.065e+02 1.374e+03 2.199e+03 2.878e+03 6.190e+03, threshold=4.397e+03, percent-clipped=12.0
+2023-03-01 19:51:47,108 INFO [train.py:968] (0/2) Epoch 3, batch 30850, giga_loss[loss=0.2813, simple_loss=0.3473, pruned_loss=0.1076, over 28714.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3671, pruned_loss=0.119, over 5654093.73 frames. ], libri_tot_loss[loss=0.3626, simple_loss=0.4059, pruned_loss=0.1597, over 5662676.59 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3657, pruned_loss=0.1162, over 5665846.81 frames. ], batch size: 242, lr: 9.46e-03, grad_scale: 4.0
+2023-03-01 19:51:51,994 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.91 vs. limit=5.0
+2023-03-01 19:52:10,990 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.89 vs. limit=5.0
+2023-03-01 19:52:35,867 INFO [train.py:968] (0/2) Epoch 3, batch 30900, giga_loss[loss=0.3026, simple_loss=0.3662, pruned_loss=0.1195, over 29018.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3656, pruned_loss=0.1187, over 5657601.05 frames. ], libri_tot_loss[loss=0.3623, simple_loss=0.4056, pruned_loss=0.1595, over 5663980.31 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3641, pruned_loss=0.116, over 5665574.18 frames. ], batch size: 155, lr: 9.46e-03, grad_scale: 4.0
+2023-03-01 19:52:44,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.026e+02 1.217e+03 1.517e+03 2.240e+03 5.287e+03, threshold=3.034e+03, percent-clipped=3.0
+2023-03-01 19:52:56,024 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-01 19:53:08,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5923, 2.8427, 1.5629, 1.6652], device='cuda:0'), covar=tensor([0.0685, 0.0418, 0.0703, 0.1023], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0457, 0.0320, 0.0400], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 19:53:28,610 INFO [train.py:968] (0/2) Epoch 3, batch 30950, giga_loss[loss=0.2698, simple_loss=0.351, pruned_loss=0.09425, over 28832.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3655, pruned_loss=0.1189, over 5649362.64 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.4054, pruned_loss=0.1594, over 5664440.89 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3637, pruned_loss=0.1161, over 5655206.48 frames. ], batch size: 174, lr: 9.46e-03, grad_scale: 4.0
+2023-03-01 19:53:48,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-01 19:54:18,612 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0152, 1.2235, 1.2707, 1.1754], device='cuda:0'), covar=tensor([0.0901, 0.0928, 0.1281, 0.0922], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0754, 0.0616, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 19:54:18,963 INFO [train.py:968] (0/2) Epoch 3, batch 31000, giga_loss[loss=0.3516, simple_loss=0.4123, pruned_loss=0.1455, over 28315.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3679, pruned_loss=0.1214, over 5649356.29 frames. ], libri_tot_loss[loss=0.3608, simple_loss=0.4042, pruned_loss=0.1587, over 5674447.17 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.366, pruned_loss=0.1181, over 5644222.18 frames. ], batch size: 368, lr: 9.46e-03, grad_scale: 4.0
+2023-03-01 19:54:26,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.292e+02 1.452e+03 1.947e+03 2.867e+03 5.883e+03, threshold=3.894e+03, percent-clipped=18.0
+2023-03-01 19:55:08,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5099, 2.1907, 1.7855, 1.7274], device='cuda:0'), covar=tensor([0.1570, 0.1642, 0.1243, 0.1042], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0774, 0.0733, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 19:55:20,588 INFO [train.py:968] (0/2) Epoch 3, batch 31050, giga_loss[loss=0.3385, simple_loss=0.4078, pruned_loss=0.1346, over 28894.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.37, pruned_loss=0.1211, over 5648386.41 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4037, pruned_loss=0.1584, over 5676868.17 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3685, pruned_loss=0.1184, over 5641984.23 frames. ], batch size: 186, lr: 9.45e-03, grad_scale: 4.0
+2023-03-01 19:55:23,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2982, 1.9273, 1.3578, 0.5870], device='cuda:0'), covar=tensor([0.1850, 0.0935, 0.1245, 0.1788], device='cuda:0'), in_proj_covar=tensor([0.1254, 0.1220, 0.1268, 0.1088], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 19:55:32,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2110, 3.1837, 2.6406, 2.4464], device='cuda:0'), covar=tensor([0.1180, 0.1124, 0.0839, 0.0724], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0773, 0.0733, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 19:55:42,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1656, 1.1396, 1.0936, 0.9942], device='cuda:0'), covar=tensor([0.0536, 0.0393, 0.0725, 0.0760], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0473, 0.0523, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 19:55:51,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2926, 1.3833, 1.2096, 1.5152], device='cuda:0'), covar=tensor([0.2176, 0.1899, 0.1871, 0.1904], device='cuda:0'), in_proj_covar=tensor([0.1026, 0.0805, 0.0927, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 19:56:20,827 INFO [train.py:968] (0/2) Epoch 3, batch 31100, giga_loss[loss=0.3223, simple_loss=0.3892, pruned_loss=0.1277, over 28878.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3696, pruned_loss=0.121, over 5643424.88 frames. ], libri_tot_loss[loss=0.3594, simple_loss=0.4029, pruned_loss=0.158, over 5683633.75 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3681, pruned_loss=0.118, over 5630794.43 frames. ], batch size: 227, lr: 9.45e-03, grad_scale: 4.0
+2023-03-01 19:56:27,688 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:56:28,106 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.152e+02 1.479e+03 1.941e+03 2.796e+03 9.577e+03, threshold=3.881e+03, percent-clipped=13.0
+2023-03-01 19:57:23,439 INFO [train.py:968] (0/2) Epoch 3, batch 31150, giga_loss[loss=0.2865, simple_loss=0.3524, pruned_loss=0.1104, over 28837.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3696, pruned_loss=0.1213, over 5653472.26 frames. ], libri_tot_loss[loss=0.358, simple_loss=0.4017, pruned_loss=0.1572, over 5691525.25 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3681, pruned_loss=0.1181, over 5634681.24 frames. ], batch size: 112, lr: 9.45e-03, grad_scale: 4.0
+2023-03-01 19:57:27,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:57:49,661 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 19:58:26,187 INFO [train.py:968] (0/2) Epoch 3, batch 31200, giga_loss[loss=0.304, simple_loss=0.3762, pruned_loss=0.1159, over 28992.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3674, pruned_loss=0.1191, over 5656692.22 frames. ], libri_tot_loss[loss=0.3579, simple_loss=0.4015, pruned_loss=0.1572, over 5695470.03 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3658, pruned_loss=0.1159, over 5637750.69 frames. ], batch size: 227, lr: 9.45e-03, grad_scale: 8.0
+2023-03-01 19:58:32,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.739e+02 1.204e+03 1.576e+03 2.144e+03 4.247e+03, threshold=3.151e+03, percent-clipped=3.0
+2023-03-01 19:59:29,815 INFO [train.py:968] (0/2) Epoch 3, batch 31250, giga_loss[loss=0.2583, simple_loss=0.3361, pruned_loss=0.09028, over 29029.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3649, pruned_loss=0.1162, over 5651889.86 frames. ], libri_tot_loss[loss=0.3578, simple_loss=0.4012, pruned_loss=0.1572, over 5699525.39 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3629, pruned_loss=0.1126, over 5632136.67 frames. ], batch size: 213, lr: 9.45e-03, grad_scale: 4.0
+2023-03-01 19:59:50,372 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:59:58,974 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-01 20:00:12,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3264, 1.9127, 1.3448, 0.5296], device='cuda:0'), covar=tensor([0.1937, 0.1043, 0.1558, 0.2297], device='cuda:0'), in_proj_covar=tensor([0.1257, 0.1207, 0.1257, 0.1082], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 20:00:26,351 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:00:29,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:00:31,559 INFO [train.py:968] (0/2) Epoch 3, batch 31300, giga_loss[loss=0.2677, simple_loss=0.331, pruned_loss=0.1022, over 29266.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3635, pruned_loss=0.1162, over 5657928.21 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4001, pruned_loss=0.1565, over 5699952.30 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3616, pruned_loss=0.1125, over 5640001.99 frames. ], batch size: 107, lr: 9.44e-03, grad_scale: 4.0
+2023-03-01 20:00:38,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.809e+02 1.557e+03 1.993e+03 3.229e+03 7.992e+03, threshold=3.987e+03, percent-clipped=26.0
+2023-03-01 20:01:08,228 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:01:32,523 INFO [train.py:968] (0/2) Epoch 3, batch 31350, giga_loss[loss=0.2726, simple_loss=0.3385, pruned_loss=0.1033, over 28849.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.362, pruned_loss=0.1158, over 5672687.69 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.3993, pruned_loss=0.1561, over 5705226.40 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3603, pruned_loss=0.1122, over 5652731.24 frames. ], batch size: 106, lr: 9.44e-03, grad_scale: 4.0
+2023-03-01 20:02:19,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5494, 1.0698, 2.9390, 2.7377], device='cuda:0'), covar=tensor([0.1505, 0.1896, 0.0426, 0.0541], device='cuda:0'), in_proj_covar=tensor([0.0524, 0.0502, 0.0688, 0.0541], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 20:02:27,416 INFO [train.py:968] (0/2) Epoch 3, batch 31400, libri_loss[loss=0.3307, simple_loss=0.3627, pruned_loss=0.1494, over 29498.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3616, pruned_loss=0.1158, over 5677753.99 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.3992, pruned_loss=0.1562, over 5707017.36 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3594, pruned_loss=0.1118, over 5659093.87 frames. ], batch size: 70, lr: 9.44e-03, grad_scale: 4.0
+2023-03-01 20:02:31,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8674, 1.2680, 3.9238, 3.2216], device='cuda:0'), covar=tensor([0.1604, 0.2005, 0.0313, 0.0546], device='cuda:0'), in_proj_covar=tensor([0.0524, 0.0502, 0.0687, 0.0541], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 20:02:33,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.295e+02 1.290e+03 1.840e+03 2.502e+03 6.739e+03, threshold=3.680e+03, percent-clipped=5.0
+2023-03-01 20:02:42,467 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121704.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 20:03:00,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8893, 1.6788, 1.4118, 1.4685], device='cuda:0'), covar=tensor([0.0667, 0.0682, 0.0879, 0.1084], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0477, 0.0532, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 20:03:04,995 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-01 20:03:24,152 INFO [train.py:968] (0/2) Epoch 3, batch 31450, giga_loss[loss=0.2914, simple_loss=0.3627, pruned_loss=0.1101, over 28754.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3624, pruned_loss=0.1154, over 5659785.41 frames. ], libri_tot_loss[loss=0.3559, simple_loss=0.3991, pruned_loss=0.1563, over 5690082.99 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.36, pruned_loss=0.1112, over 5659080.39 frames. ], batch size: 99, lr: 9.44e-03, grad_scale: 4.0
+2023-03-01 20:03:27,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4861, 3.1060, 1.4744, 1.4131], device='cuda:0'), covar=tensor([0.0844, 0.0301, 0.0886, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0451, 0.0315, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 20:04:02,116 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:04:31,162 INFO [train.py:968] (0/2) Epoch 3, batch 31500, giga_loss[loss=0.2979, simple_loss=0.3662, pruned_loss=0.1148, over 28953.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3629, pruned_loss=0.1148, over 5651736.53 frames. ], libri_tot_loss[loss=0.3556, simple_loss=0.3988, pruned_loss=0.1562, over 5692420.57 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3606, pruned_loss=0.1109, over 5648697.41 frames. ], batch size: 199, lr: 9.44e-03, grad_scale: 4.0
+2023-03-01 20:04:41,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.024e+02 1.383e+03 1.758e+03 2.419e+03 5.878e+03, threshold=3.517e+03, percent-clipped=6.0
+2023-03-01 20:04:44,522 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:05:37,339 INFO [train.py:968] (0/2) Epoch 3, batch 31550, giga_loss[loss=0.2803, simple_loss=0.3504, pruned_loss=0.1051, over 28768.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3594, pruned_loss=0.1126, over 5662554.74 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.3987, pruned_loss=0.1564, over 5686764.99 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3571, pruned_loss=0.1086, over 5664805.99 frames. ], batch size: 243, lr: 9.43e-03, grad_scale: 4.0
+2023-03-01 20:06:40,328 INFO [train.py:968] (0/2) Epoch 3, batch 31600, giga_loss[loss=0.3216, simple_loss=0.3901, pruned_loss=0.1266, over 28771.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3632, pruned_loss=0.1159, over 5657736.79 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.3987, pruned_loss=0.1565, over 5682975.46 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3603, pruned_loss=0.1116, over 5662476.86 frames. ], batch size: 243, lr: 9.43e-03, grad_scale: 8.0
+2023-03-01 20:06:48,808 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.287e+02 1.538e+03 2.232e+03 3.035e+03 7.247e+03, threshold=4.464e+03, percent-clipped=15.0
+2023-03-01 20:07:06,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6123, 2.2494, 1.5559, 0.8215], device='cuda:0'), covar=tensor([0.1782, 0.1033, 0.1698, 0.2090], device='cuda:0'), in_proj_covar=tensor([0.1259, 0.1211, 0.1273, 0.1091], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 20:07:11,656 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:07:13,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:07:31,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:07:32,558 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 20:07:41,265 INFO [train.py:968] (0/2) Epoch 3, batch 31650, giga_loss[loss=0.316, simple_loss=0.3973, pruned_loss=0.1174, over 28701.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3661, pruned_loss=0.1159, over 5664051.06 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.3986, pruned_loss=0.1565, over 5685005.55 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3631, pruned_loss=0.1114, over 5665623.11 frames. ], batch size: 262, lr: 9.43e-03, grad_scale: 4.0
+2023-03-01 20:07:47,820 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:08:41,931 INFO [train.py:968] (0/2) Epoch 3, batch 31700, giga_loss[loss=0.2824, simple_loss=0.3678, pruned_loss=0.09851, over 28422.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3674, pruned_loss=0.1144, over 5652962.57 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.3985, pruned_loss=0.1565, over 5679095.47 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3641, pruned_loss=0.1095, over 5659010.47 frames. ], batch size: 336, lr: 9.43e-03, grad_scale: 4.0
+2023-03-01 20:08:53,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.706e+02 1.355e+03 1.827e+03 2.373e+03 1.476e+04, threshold=3.655e+03, percent-clipped=8.0
+2023-03-01 20:08:54,144 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-122000.pt
+2023-03-01 20:09:39,815 INFO [train.py:968] (0/2) Epoch 3, batch 31750, giga_loss[loss=0.2863, simple_loss=0.3652, pruned_loss=0.1037, over 27520.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3685, pruned_loss=0.1142, over 5648044.61 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.3985, pruned_loss=0.1564, over 5676400.52 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3645, pruned_loss=0.1086, over 5654678.29 frames. ], batch size: 472, lr: 9.43e-03, grad_scale: 4.0
+2023-03-01 20:09:40,511 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6918, 1.6535, 1.6459, 1.5474], device='cuda:0'), covar=tensor([0.0916, 0.1848, 0.1424, 0.1505], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0769, 0.0621, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 20:10:19,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:10:25,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:10:27,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 20:10:40,784 INFO [train.py:968] (0/2) Epoch 3, batch 31800, giga_loss[loss=0.2602, simple_loss=0.3497, pruned_loss=0.0853, over 29066.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.367, pruned_loss=0.1118, over 5655554.99 frames. ], libri_tot_loss[loss=0.3554, simple_loss=0.3983, pruned_loss=0.1563, over 5671795.01 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3634, pruned_loss=0.1065, over 5664381.22 frames. ], batch size: 128, lr: 9.42e-03, grad_scale: 4.0
+2023-03-01 20:10:50,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.620e+02 1.357e+03 1.695e+03 2.297e+03 5.823e+03, threshold=3.390e+03, percent-clipped=4.0
+2023-03-01 20:11:01,344 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:11:50,616 INFO [train.py:968] (0/2) Epoch 3, batch 31850, giga_loss[loss=0.2879, simple_loss=0.3592, pruned_loss=0.1083, over 28875.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.368, pruned_loss=0.1133, over 5659954.36 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.3979, pruned_loss=0.156, over 5665632.12 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3652, pruned_loss=0.1088, over 5671646.77 frames. ], batch size: 174, lr: 9.42e-03, grad_scale: 4.0
+2023-03-01 20:12:41,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:13:06,842 INFO [train.py:968] (0/2) Epoch 3, batch 31900, giga_loss[loss=0.2942, simple_loss=0.3642, pruned_loss=0.112, over 28420.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3668, pruned_loss=0.1138, over 5665670.54 frames. ], libri_tot_loss[loss=0.3554, simple_loss=0.3982, pruned_loss=0.1563, over 5667733.24 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.364, pruned_loss=0.1096, over 5672978.59 frames. ], batch size: 336, lr: 9.42e-03, grad_scale: 4.0
+2023-03-01 20:13:20,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.582e+02 1.355e+03 1.773e+03 2.657e+03 6.270e+03, threshold=3.546e+03, percent-clipped=13.0
+2023-03-01 20:13:56,177 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122222.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 20:14:00,437 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122225.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 20:14:25,595 INFO [train.py:968] (0/2) Epoch 3, batch 31950, giga_loss[loss=0.2848, simple_loss=0.3563, pruned_loss=0.1067, over 28973.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3681, pruned_loss=0.1162, over 5673640.59 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.3966, pruned_loss=0.1554, over 5678067.61 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3659, pruned_loss=0.1119, over 5669846.39 frames. ], batch size: 136, lr: 9.42e-03, grad_scale: 4.0
+2023-03-01 20:14:44,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122254.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 20:15:39,000 INFO [train.py:968] (0/2) Epoch 3, batch 32000, giga_loss[loss=0.2404, simple_loss=0.3257, pruned_loss=0.07758, over 28737.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3618, pruned_loss=0.1115, over 5677829.76 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.3965, pruned_loss=0.1552, over 5678757.27 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3601, pruned_loss=0.108, over 5674303.38 frames. ], batch size: 243, lr: 9.42e-03, grad_scale: 8.0
+2023-03-01 20:15:50,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.708e+02 1.226e+03 1.634e+03 2.102e+03 4.918e+03, threshold=3.269e+03, percent-clipped=2.0
+2023-03-01 20:16:17,589 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:16:22,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:16:45,415 INFO [train.py:968] (0/2) Epoch 3, batch 32050, giga_loss[loss=0.2913, simple_loss=0.3584, pruned_loss=0.1121, over 27705.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3603, pruned_loss=0.1107, over 5674236.31 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.3964, pruned_loss=0.1552, over 5681443.88 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3583, pruned_loss=0.1072, over 5668848.16 frames. ], batch size: 472, lr: 9.41e-03, grad_scale: 8.0
+2023-03-01 20:16:48,221 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:16:59,136 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:17:51,951 INFO [train.py:968] (0/2) Epoch 3, batch 32100, giga_loss[loss=0.2629, simple_loss=0.3265, pruned_loss=0.09969, over 24435.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3595, pruned_loss=0.1112, over 5677346.20 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.3963, pruned_loss=0.1551, over 5680188.45 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3574, pruned_loss=0.1076, over 5674472.26 frames. ], batch size: 705, lr: 9.41e-03, grad_scale: 8.0
+2023-03-01 20:18:02,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.652e+02 1.531e+03 2.009e+03 2.722e+03 5.328e+03, threshold=4.018e+03, percent-clipped=16.0
+2023-03-01 20:18:49,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3002, 1.4949, 1.1024, 1.5183], device='cuda:0'), covar=tensor([0.0781, 0.0298, 0.0371, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0233, 0.0149, 0.0155, 0.0276], device='cuda:0'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:0')
+2023-03-01 20:18:52,462 INFO [train.py:968] (0/2) Epoch 3, batch 32150, giga_loss[loss=0.2773, simple_loss=0.3545, pruned_loss=0.1001, over 28744.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3639, pruned_loss=0.1133, over 5678158.75 frames. ], libri_tot_loss[loss=0.3527, simple_loss=0.396, pruned_loss=0.1547, over 5681058.35 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3617, pruned_loss=0.1099, over 5675014.25 frames. ], batch size: 99, lr: 9.41e-03, grad_scale: 8.0
+2023-03-01 20:19:23,081 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-01 20:19:50,743 INFO [train.py:968] (0/2) Epoch 3, batch 32200, giga_loss[loss=0.2589, simple_loss=0.3363, pruned_loss=0.09072, over 28891.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3639, pruned_loss=0.1143, over 5691615.51 frames. ], libri_tot_loss[loss=0.3516, simple_loss=0.3951, pruned_loss=0.1541, over 5685835.47 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3621, pruned_loss=0.111, over 5684873.44 frames. ], batch size: 186, lr: 9.41e-03, grad_scale: 4.0
+2023-03-01 20:20:03,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.685e+02 1.410e+03 1.884e+03 2.418e+03 5.765e+03, threshold=3.767e+03, percent-clipped=6.0
+2023-03-01 20:20:06,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3622, 1.5773, 0.9514, 1.0768], device='cuda:0'), covar=tensor([0.0793, 0.0570, 0.0551, 0.0658], device='cuda:0'), in_proj_covar=tensor([0.1211, 0.0915, 0.0927, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 20:20:17,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:20:53,641 INFO [train.py:968] (0/2) Epoch 3, batch 32250, giga_loss[loss=0.3087, simple_loss=0.3753, pruned_loss=0.121, over 28926.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.363, pruned_loss=0.1147, over 5689215.36 frames. ], libri_tot_loss[loss=0.3509, simple_loss=0.3945, pruned_loss=0.1537, over 5689543.89 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3614, pruned_loss=0.1115, over 5680528.26 frames. ], batch size: 284, lr: 9.41e-03, grad_scale: 4.0
+2023-03-01 20:21:18,336 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-01 20:21:20,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3507, 1.6250, 0.9783, 0.9963], device='cuda:0'), covar=tensor([0.0894, 0.0622, 0.0567, 0.0680], device='cuda:0'), in_proj_covar=tensor([0.1211, 0.0914, 0.0925, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001], device='cuda:0')
+2023-03-01 20:21:48,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:21:51,527 INFO [train.py:968] (0/2) Epoch 3, batch 32300, giga_loss[loss=0.2548, simple_loss=0.3161, pruned_loss=0.09674, over 24810.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3626, pruned_loss=0.1153, over 5688399.83 frames. ], libri_tot_loss[loss=0.3496, simple_loss=0.3933, pruned_loss=0.1529, over 5693165.31 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3614, pruned_loss=0.1122, over 5678318.10 frames. ], batch size: 705, lr: 9.41e-03, grad_scale: 4.0
+2023-03-01 20:22:02,695 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.377e+02 1.436e+03 1.823e+03 2.425e+03 5.114e+03, threshold=3.646e+03, percent-clipped=8.0
+2023-03-01 20:23:01,276 INFO [train.py:968] (0/2) Epoch 3, batch 32350, giga_loss[loss=0.3151, simple_loss=0.3704, pruned_loss=0.1298, over 26969.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3638, pruned_loss=0.1147, over 5680890.79 frames. ], libri_tot_loss[loss=0.3495, simple_loss=0.3932, pruned_loss=0.1529, over 5692116.14 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3625, pruned_loss=0.1118, over 5674004.74 frames. ], batch size: 555, lr: 9.40e-03, grad_scale: 2.0
+2023-03-01 20:24:06,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-01 20:24:13,608 INFO [train.py:968] (0/2) Epoch 3, batch 32400, giga_loss[loss=0.2758, simple_loss=0.3492, pruned_loss=0.1012, over 27460.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3645, pruned_loss=0.1145, over 5658207.05 frames. ], libri_tot_loss[loss=0.3497, simple_loss=0.3932, pruned_loss=0.1531, over 5676331.89 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3629, pruned_loss=0.1112, over 5666530.58 frames. ], batch size: 472, lr: 9.40e-03, grad_scale: 4.0
+2023-03-01 20:24:29,380 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.830e+02 1.517e+03 1.845e+03 2.264e+03 6.423e+03, threshold=3.690e+03, percent-clipped=9.0
+2023-03-01 20:24:45,567 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:24:45,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6215, 1.9473, 1.7541, 1.7783], device='cuda:0'), covar=tensor([0.1565, 0.1831, 0.1234, 0.1012], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0774, 0.0741, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 20:24:50,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4905, 2.2292, 1.4967, 0.5606], device='cuda:0'), covar=tensor([0.2039, 0.0958, 0.1863, 0.2347], device='cuda:0'), in_proj_covar=tensor([0.1281, 0.1217, 0.1282, 0.1107], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 20:24:53,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:25:26,502 INFO [train.py:968] (0/2) Epoch 3, batch 32450, giga_loss[loss=0.26, simple_loss=0.337, pruned_loss=0.09149, over 29037.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3634, pruned_loss=0.1142, over 5652305.46 frames. ], libri_tot_loss[loss=0.3495, simple_loss=0.3931, pruned_loss=0.153, over 5667947.99 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3613, pruned_loss=0.1105, over 5665150.01 frames. ], batch size: 285, lr: 9.40e-03, grad_scale: 4.0
+2023-03-01 20:26:29,069 INFO [train.py:968] (0/2) Epoch 3, batch 32500, giga_loss[loss=0.253, simple_loss=0.3263, pruned_loss=0.08981, over 28939.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3586, pruned_loss=0.1126, over 5668105.08 frames. ], libri_tot_loss[loss=0.3486, simple_loss=0.3924, pruned_loss=0.1524, over 5676279.77 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3565, pruned_loss=0.1088, over 5670358.39 frames. ], batch size: 145, lr: 9.40e-03, grad_scale: 4.0
+2023-03-01 20:26:38,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.385e+02 1.248e+03 1.741e+03 2.654e+03 6.862e+03, threshold=3.482e+03, percent-clipped=6.0
+2023-03-01 20:26:41,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2519, 1.4215, 0.9134, 0.8224], device='cuda:0'), covar=tensor([0.0606, 0.0517, 0.0452, 0.0581], device='cuda:0'), in_proj_covar=tensor([0.1225, 0.0928, 0.0951, 0.1047], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 20:27:32,379 INFO [train.py:968] (0/2) Epoch 3, batch 32550, giga_loss[loss=0.2628, simple_loss=0.3382, pruned_loss=0.09369, over 28679.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3526, pruned_loss=0.1097, over 5670118.92 frames. ], libri_tot_loss[loss=0.3486, simple_loss=0.3922, pruned_loss=0.1525, over 5679372.60 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3506, pruned_loss=0.1061, over 5669214.44 frames. ], batch size: 262, lr: 9.40e-03, grad_scale: 4.0
+2023-03-01 20:27:58,508 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:27:58,606 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:28:02,978 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:28:19,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0903, 3.8754, 3.8301, 2.4031], device='cuda:0'), covar=tensor([0.0518, 0.0581, 0.0898, 0.1633], device='cuda:0'), in_proj_covar=tensor([0.0749, 0.0608, 0.0753, 0.0561], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 20:28:26,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:28:33,627 INFO [train.py:968] (0/2) Epoch 3, batch 32600, giga_loss[loss=0.3119, simple_loss=0.3719, pruned_loss=0.126, over 28318.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3536, pruned_loss=0.1107, over 5646820.60 frames. ], libri_tot_loss[loss=0.3492, simple_loss=0.3925, pruned_loss=0.1529, over 5657584.69 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.351, pruned_loss=0.1066, over 5665285.05 frames. ], batch size: 368, lr: 9.39e-03, grad_scale: 4.0
+2023-03-01 20:28:36,249 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:28:45,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.958e+02 1.496e+03 2.148e+03 3.005e+03 1.097e+04, threshold=4.296e+03, percent-clipped=17.0
+2023-03-01 20:29:03,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8087, 1.7613, 1.6032, 1.6275], device='cuda:0'), covar=tensor([0.0902, 0.1743, 0.1450, 0.1361], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0766, 0.0612, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 20:29:33,046 INFO [train.py:968] (0/2) Epoch 3, batch 32650, giga_loss[loss=0.2663, simple_loss=0.3423, pruned_loss=0.09512, over 28328.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.355, pruned_loss=0.1112, over 5659847.97 frames. ], libri_tot_loss[loss=0.3485, simple_loss=0.3919, pruned_loss=0.1526, over 5661256.91 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3529, pruned_loss=0.1077, over 5671049.46 frames. ], batch size: 368, lr: 9.39e-03, grad_scale: 4.0
+2023-03-01 20:29:59,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:30:34,829 INFO [train.py:968] (0/2) Epoch 3, batch 32700, giga_loss[loss=0.2877, simple_loss=0.3481, pruned_loss=0.1137, over 26754.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.352, pruned_loss=0.1085, over 5657880.49 frames. ], libri_tot_loss[loss=0.3478, simple_loss=0.3913, pruned_loss=0.1522, over 5665055.46 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3502, pruned_loss=0.1053, over 5663365.18 frames. ], batch size: 555, lr: 9.39e-03, grad_scale: 4.0
+2023-03-01 20:30:47,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.954e+02 1.345e+03 1.770e+03 2.687e+03 5.766e+03, threshold=3.540e+03, percent-clipped=2.0
+2023-03-01 20:30:52,971 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 20:31:21,425 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:31:24,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:31:35,050 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 20:31:36,842 INFO [train.py:968] (0/2) Epoch 3, batch 32750, libri_loss[loss=0.3147, simple_loss=0.3532, pruned_loss=0.1381, over 29660.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3509, pruned_loss=0.1079, over 5661316.47 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3903, pruned_loss=0.1517, over 5670690.81 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3492, pruned_loss=0.1045, over 5660343.24 frames. ], batch size: 73, lr: 9.39e-03, grad_scale: 4.0
+2023-03-01 20:32:01,734 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:32:36,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:32:41,679 INFO [train.py:968] (0/2) Epoch 3, batch 32800, giga_loss[loss=0.2702, simple_loss=0.348, pruned_loss=0.09617, over 28994.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3506, pruned_loss=0.1087, over 5668866.00 frames. ], libri_tot_loss[loss=0.3464, simple_loss=0.39, pruned_loss=0.1515, over 5675681.36 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3483, pruned_loss=0.1048, over 5663332.43 frames. ], batch size: 199, lr: 9.39e-03, grad_scale: 8.0
+2023-03-01 20:32:55,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.659e+02 1.201e+03 1.698e+03 2.448e+03 5.328e+03, threshold=3.397e+03, percent-clipped=5.0
+2023-03-01 20:33:01,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:33:07,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:33:28,528 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-01 20:33:44,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:33:50,253 INFO [train.py:968] (0/2) Epoch 3, batch 32850, giga_loss[loss=0.3233, simple_loss=0.374, pruned_loss=0.1363, over 26758.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3512, pruned_loss=0.1077, over 5669737.32 frames. ], libri_tot_loss[loss=0.3466, simple_loss=0.39, pruned_loss=0.1516, over 5669056.44 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3488, pruned_loss=0.104, over 5671822.23 frames. ], batch size: 555, lr: 9.38e-03, grad_scale: 8.0
+2023-03-01 20:34:53,174 INFO [train.py:968] (0/2) Epoch 3, batch 32900, giga_loss[loss=0.2649, simple_loss=0.3384, pruned_loss=0.09573, over 28950.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3515, pruned_loss=0.108, over 5670062.03 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.3897, pruned_loss=0.1514, over 5665071.24 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.349, pruned_loss=0.1042, over 5674703.30 frames. ], batch size: 199, lr: 9.38e-03, grad_scale: 8.0
+2023-03-01 20:35:06,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.517e+02 1.197e+03 1.536e+03 2.158e+03 6.094e+03, threshold=3.072e+03, percent-clipped=7.0
+2023-03-01 20:35:45,183 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:35:47,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:35:53,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:35:59,433 INFO [train.py:968] (0/2) Epoch 3, batch 32950, giga_loss[loss=0.2504, simple_loss=0.3328, pruned_loss=0.08401, over 28587.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.353, pruned_loss=0.1097, over 5664097.04 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3901, pruned_loss=0.1518, over 5656422.17 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3501, pruned_loss=0.1056, over 5676004.92 frames. ], batch size: 307, lr: 9.38e-03, grad_scale: 4.0
+2023-03-01 20:36:27,447 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:36:29,631 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=123265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:37:02,899 INFO [train.py:968] (0/2) Epoch 3, batch 33000, giga_loss[loss=0.2403, simple_loss=0.2979, pruned_loss=0.09131, over 24513.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3513, pruned_loss=0.1071, over 5658811.70 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3901, pruned_loss=0.1518, over 5657332.06 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3489, pruned_loss=0.1037, over 5667341.94 frames. ], batch size: 705, lr: 9.38e-03, grad_scale: 4.0
+2023-03-01 20:37:02,903 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 20:37:11,509 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2357, simple_loss=0.3303, pruned_loss=0.07052, over 944034.00 frames. 
+2023-03-01 20:37:11,509 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 20:37:24,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.435e+02 1.205e+03 1.615e+03 2.134e+03 4.776e+03, threshold=3.229e+03, percent-clipped=6.0
+2023-03-01 20:37:46,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2697, 2.6274, 1.2292, 1.2798], device='cuda:0'), covar=tensor([0.1116, 0.0573, 0.1071, 0.1635], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0457, 0.0318, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:0')
+2023-03-01 20:37:55,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3260, 1.5953, 1.1937, 1.4574], device='cuda:0'), covar=tensor([0.0891, 0.0364, 0.0415, 0.0947], device='cuda:0'), in_proj_covar=tensor([0.0234, 0.0150, 0.0156, 0.0277], device='cuda:0'), out_proj_covar=tensor([0.0045, 0.0032, 0.0030, 0.0050], device='cuda:0')
+2023-03-01 20:38:10,634 INFO [train.py:968] (0/2) Epoch 3, batch 33050, giga_loss[loss=0.3163, simple_loss=0.3715, pruned_loss=0.1305, over 26916.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3545, pruned_loss=0.1077, over 5655198.24 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3901, pruned_loss=0.1518, over 5657999.75 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3522, pruned_loss=0.1046, over 5661191.35 frames. ], batch size: 555, lr: 9.38e-03, grad_scale: 4.0
+2023-03-01 20:38:54,261 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:38:57,165 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:39:07,742 INFO [train.py:968] (0/2) Epoch 3, batch 33100, giga_loss[loss=0.2594, simple_loss=0.3418, pruned_loss=0.08857, over 28954.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3581, pruned_loss=0.1101, over 5661876.07 frames. ], libri_tot_loss[loss=0.3465, simple_loss=0.3898, pruned_loss=0.1516, over 5662945.53 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3554, pruned_loss=0.1064, over 5662243.63 frames. ], batch size: 128, lr: 9.37e-03, grad_scale: 4.0
+2023-03-01 20:39:21,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.519e+02 1.598e+03 2.119e+03 2.750e+03 7.317e+03, threshold=4.238e+03, percent-clipped=16.0
+2023-03-01 20:39:36,176 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:40:14,985 INFO [train.py:968] (0/2) Epoch 3, batch 33150, giga_loss[loss=0.2504, simple_loss=0.3357, pruned_loss=0.08254, over 29052.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3581, pruned_loss=0.1099, over 5662413.46 frames. ], libri_tot_loss[loss=0.346, simple_loss=0.3894, pruned_loss=0.1513, over 5665230.75 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3561, pruned_loss=0.1068, over 5660658.17 frames. ], batch size: 155, lr: 9.37e-03, grad_scale: 4.0
+2023-03-01 20:41:12,618 INFO [train.py:968] (0/2) Epoch 3, batch 33200, giga_loss[loss=0.3107, simple_loss=0.3764, pruned_loss=0.1225, over 28482.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3591, pruned_loss=0.1111, over 5672053.07 frames. ], libri_tot_loss[loss=0.3462, simple_loss=0.3896, pruned_loss=0.1515, over 5671872.07 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3565, pruned_loss=0.1074, over 5664646.81 frames. ], batch size: 336, lr: 9.37e-03, grad_scale: 8.0
+2023-03-01 20:41:27,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.372e+02 1.402e+03 1.969e+03 2.589e+03 5.583e+03, threshold=3.938e+03, percent-clipped=3.0
+2023-03-01 20:42:20,108 INFO [train.py:968] (0/2) Epoch 3, batch 33250, giga_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1144, over 27873.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3547, pruned_loss=0.1073, over 5675217.30 frames. ], libri_tot_loss[loss=0.3462, simple_loss=0.3896, pruned_loss=0.1515, over 5671872.07 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3527, pruned_loss=0.1044, over 5669452.91 frames. ], batch size: 474, lr: 9.37e-03, grad_scale: 8.0
+2023-03-01 20:43:22,500 INFO [train.py:968] (0/2) Epoch 3, batch 33300, giga_loss[loss=0.2516, simple_loss=0.3254, pruned_loss=0.08884, over 28092.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3536, pruned_loss=0.1072, over 5680244.95 frames. ], libri_tot_loss[loss=0.346, simple_loss=0.3893, pruned_loss=0.1514, over 5673151.71 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.352, pruned_loss=0.1049, over 5674549.62 frames. ], batch size: 412, lr: 9.37e-03, grad_scale: 8.0
+2023-03-01 20:43:36,873 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.908e+02 1.371e+03 1.855e+03 2.433e+03 4.729e+03, threshold=3.710e+03, percent-clipped=5.0
+2023-03-01 20:44:24,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:44:24,623 INFO [train.py:968] (0/2) Epoch 3, batch 33350, giga_loss[loss=0.2723, simple_loss=0.3526, pruned_loss=0.09594, over 28721.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3532, pruned_loss=0.1072, over 5676946.27 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.3895, pruned_loss=0.1516, over 5675232.79 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3516, pruned_loss=0.1048, over 5670776.84 frames. ], batch size: 243, lr: 9.37e-03, grad_scale: 4.0
+2023-03-01 20:45:32,780 INFO [train.py:968] (0/2) Epoch 3, batch 33400, giga_loss[loss=0.2626, simple_loss=0.338, pruned_loss=0.09359, over 28960.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3564, pruned_loss=0.109, over 5673809.62 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.3895, pruned_loss=0.1515, over 5675014.54 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3545, pruned_loss=0.1065, over 5669032.36 frames. ], batch size: 106, lr: 9.36e-03, grad_scale: 4.0
+2023-03-01 20:45:47,199 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.827e+02 1.364e+03 1.869e+03 2.567e+03 9.179e+03, threshold=3.738e+03, percent-clipped=9.0
+2023-03-01 20:45:51,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=123707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:45:57,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2542, 1.9182, 1.3724, 0.5476], device='cuda:0'), covar=tensor([0.1769, 0.1085, 0.1786, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.1251, 0.1232, 0.1271, 0.1088], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 20:46:32,154 INFO [train.py:968] (0/2) Epoch 3, batch 33450, giga_loss[loss=0.3418, simple_loss=0.3873, pruned_loss=0.1482, over 27633.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.358, pruned_loss=0.1105, over 5670887.00 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.3896, pruned_loss=0.1515, over 5671654.49 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3556, pruned_loss=0.1074, over 5670654.32 frames. ], batch size: 472, lr: 9.36e-03, grad_scale: 4.0
+2023-03-01 20:47:31,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:47:36,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:47:41,667 INFO [train.py:968] (0/2) Epoch 3, batch 33500, giga_loss[loss=0.2588, simple_loss=0.3414, pruned_loss=0.0881, over 28971.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3617, pruned_loss=0.1139, over 5661571.57 frames. ], libri_tot_loss[loss=0.3457, simple_loss=0.3891, pruned_loss=0.1511, over 5675018.32 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3597, pruned_loss=0.1111, over 5658199.52 frames. ], batch size: 128, lr: 9.36e-03, grad_scale: 4.0
+2023-03-01 20:47:52,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.656e+02 1.523e+03 2.177e+03 2.762e+03 5.037e+03, threshold=4.355e+03, percent-clipped=5.0
+2023-03-01 20:48:01,246 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 20:48:07,688 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:48:12,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=123821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:48:33,021 INFO [train.py:968] (0/2) Epoch 3, batch 33550, giga_loss[loss=0.2711, simple_loss=0.3509, pruned_loss=0.09569, over 28353.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3636, pruned_loss=0.1145, over 5666639.02 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3877, pruned_loss=0.1501, over 5683995.26 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3622, pruned_loss=0.1117, over 5655498.33 frames. ], batch size: 368, lr: 9.36e-03, grad_scale: 4.0
+2023-03-01 20:49:37,557 INFO [train.py:968] (0/2) Epoch 3, batch 33600, giga_loss[loss=0.2944, simple_loss=0.3654, pruned_loss=0.1117, over 28072.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3653, pruned_loss=0.1153, over 5669478.07 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.387, pruned_loss=0.1498, over 5690539.16 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3642, pruned_loss=0.1124, over 5654384.43 frames. ], batch size: 412, lr: 9.36e-03, grad_scale: 8.0
+2023-03-01 20:49:56,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.791e+02 1.295e+03 1.715e+03 2.489e+03 7.014e+03, threshold=3.429e+03, percent-clipped=8.0
+2023-03-01 20:50:44,194 INFO [train.py:968] (0/2) Epoch 3, batch 33650, giga_loss[loss=0.287, simple_loss=0.3537, pruned_loss=0.1102, over 28993.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3655, pruned_loss=0.1158, over 5683517.60 frames. ], libri_tot_loss[loss=0.3435, simple_loss=0.3871, pruned_loss=0.1499, over 5694253.07 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3638, pruned_loss=0.1123, over 5667308.54 frames. ], batch size: 213, lr: 9.35e-03, grad_scale: 4.0
+2023-03-01 20:51:09,547 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 20:51:46,135 INFO [train.py:968] (0/2) Epoch 3, batch 33700, giga_loss[loss=0.2899, simple_loss=0.3698, pruned_loss=0.1049, over 28887.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3641, pruned_loss=0.1159, over 5691586.50 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.3871, pruned_loss=0.1501, over 5700031.86 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3618, pruned_loss=0.1115, over 5672696.91 frames. ], batch size: 174, lr: 9.35e-03, grad_scale: 4.0
+2023-03-01 20:51:58,246 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-124000.pt
+2023-03-01 20:52:04,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.787e+02 1.596e+03 2.250e+03 3.141e+03 8.373e+03, threshold=4.499e+03, percent-clipped=22.0
+2023-03-01 20:52:07,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3326, 1.6530, 1.2454, 1.5390], device='cuda:0'), covar=tensor([0.0879, 0.0364, 0.0384, 0.0940], device='cuda:0'), in_proj_covar=tensor([0.0230, 0.0148, 0.0153, 0.0273], device='cuda:0'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:0')
+2023-03-01 20:52:47,673 INFO [train.py:968] (0/2) Epoch 3, batch 33750, giga_loss[loss=0.293, simple_loss=0.3555, pruned_loss=0.1152, over 28024.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3631, pruned_loss=0.1152, over 5684228.32 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3865, pruned_loss=0.1496, over 5696820.77 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.361, pruned_loss=0.1109, over 5671953.71 frames. ], batch size: 412, lr: 9.35e-03, grad_scale: 4.0
+2023-03-01 20:53:44,843 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7384, 3.4721, 3.4805, 1.7340], device='cuda:0'), covar=tensor([0.0505, 0.0476, 0.0806, 0.2058], device='cuda:0'), in_proj_covar=tensor([0.0754, 0.0616, 0.0755, 0.0556], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 20:53:44,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:53:53,605 INFO [train.py:968] (0/2) Epoch 3, batch 33800, giga_loss[loss=0.2292, simple_loss=0.3068, pruned_loss=0.07582, over 28944.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3616, pruned_loss=0.1149, over 5681482.99 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3865, pruned_loss=0.1497, over 5697325.11 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3595, pruned_loss=0.1108, over 5671078.99 frames. ], batch size: 106, lr: 9.35e-03, grad_scale: 4.0
+2023-03-01 20:53:54,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6172, 1.0728, 2.8793, 2.7031], device='cuda:0'), covar=tensor([0.1455, 0.1844, 0.0487, 0.0657], device='cuda:0'), in_proj_covar=tensor([0.0530, 0.0503, 0.0679, 0.0529], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 20:54:09,991 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.474e+02 1.525e+03 2.021e+03 2.856e+03 8.594e+03, threshold=4.042e+03, percent-clipped=9.0
+2023-03-01 20:54:54,033 INFO [train.py:968] (0/2) Epoch 3, batch 33850, giga_loss[loss=0.2751, simple_loss=0.3491, pruned_loss=0.1006, over 28738.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3599, pruned_loss=0.1146, over 5687384.35 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3861, pruned_loss=0.1494, over 5699456.17 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3577, pruned_loss=0.1105, over 5676732.66 frames. ], batch size: 243, lr: 9.35e-03, grad_scale: 4.0
+2023-03-01 20:55:34,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-01 20:55:40,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3356, 2.1421, 2.0295, 1.9119], device='cuda:0'), covar=tensor([0.0851, 0.1883, 0.1349, 0.1453], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0761, 0.0611, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 20:55:54,509 INFO [train.py:968] (0/2) Epoch 3, batch 33900, giga_loss[loss=0.2792, simple_loss=0.358, pruned_loss=0.1002, over 28499.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3597, pruned_loss=0.1132, over 5692718.27 frames. ], libri_tot_loss[loss=0.342, simple_loss=0.3858, pruned_loss=0.1491, over 5703169.66 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3576, pruned_loss=0.1093, over 5680641.15 frames. ], batch size: 336, lr: 9.34e-03, grad_scale: 4.0
+2023-03-01 20:55:59,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:56:06,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.152e+02 1.370e+03 1.804e+03 2.420e+03 6.763e+03, threshold=3.608e+03, percent-clipped=7.0
+2023-03-01 20:56:32,693 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:56:36,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:56:51,499 INFO [train.py:968] (0/2) Epoch 3, batch 33950, libri_loss[loss=0.2853, simple_loss=0.3468, pruned_loss=0.1119, over 29532.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3572, pruned_loss=0.1118, over 5665513.75 frames. ], libri_tot_loss[loss=0.3416, simple_loss=0.3853, pruned_loss=0.1489, over 5690549.48 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3548, pruned_loss=0.1072, over 5666461.82 frames. ], batch size: 80, lr: 9.34e-03, grad_scale: 4.0
+2023-03-01 20:57:10,363 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:57:18,872 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:57:45,908 INFO [train.py:968] (0/2) Epoch 3, batch 34000, giga_loss[loss=0.2948, simple_loss=0.3788, pruned_loss=0.1054, over 28673.00 frames. ], tot_loss[loss=0.29, simple_loss=0.359, pruned_loss=0.1105, over 5674669.96 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3851, pruned_loss=0.1487, over 5695421.35 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3565, pruned_loss=0.1059, over 5670262.80 frames. ], batch size: 242, lr: 9.34e-03, grad_scale: 8.0
+2023-03-01 20:57:59,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.510e+02 1.441e+03 1.790e+03 2.497e+03 5.924e+03, threshold=3.580e+03, percent-clipped=7.0
+2023-03-01 20:58:39,109 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:58:39,228 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 20:58:41,627 INFO [train.py:968] (0/2) Epoch 3, batch 34050, giga_loss[loss=0.3371, simple_loss=0.3838, pruned_loss=0.1452, over 26815.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3615, pruned_loss=0.111, over 5668079.31 frames. ], libri_tot_loss[loss=0.3416, simple_loss=0.3853, pruned_loss=0.1489, over 5684177.29 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3585, pruned_loss=0.106, over 5674355.58 frames. ], batch size: 555, lr: 9.34e-03, grad_scale: 8.0
+2023-03-01 20:58:42,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:59:16,003 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:59:44,388 INFO [train.py:968] (0/2) Epoch 3, batch 34100, giga_loss[loss=0.2464, simple_loss=0.3078, pruned_loss=0.09246, over 24218.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3612, pruned_loss=0.1104, over 5666763.89 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.385, pruned_loss=0.1487, over 5685975.51 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3588, pruned_loss=0.1061, over 5669967.19 frames. ], batch size: 705, lr: 9.34e-03, grad_scale: 8.0
+2023-03-01 21:00:02,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.891e+02 1.345e+03 1.821e+03 2.700e+03 6.163e+03, threshold=3.642e+03, percent-clipped=11.0
+2023-03-01 21:00:37,274 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:00:53,157 INFO [train.py:968] (0/2) Epoch 3, batch 34150, giga_loss[loss=0.2607, simple_loss=0.3424, pruned_loss=0.08947, over 28131.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3614, pruned_loss=0.1107, over 5666879.98 frames. ], libri_tot_loss[loss=0.3411, simple_loss=0.3849, pruned_loss=0.1487, over 5690228.76 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3591, pruned_loss=0.1066, over 5665266.98 frames. ], batch size: 412, lr: 9.34e-03, grad_scale: 2.0
+2023-03-01 21:01:09,610 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:02:03,133 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-01 21:02:03,311 INFO [train.py:968] (0/2) Epoch 3, batch 34200, giga_loss[loss=0.3323, simple_loss=0.3968, pruned_loss=0.1339, over 27596.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3623, pruned_loss=0.111, over 5668295.98 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3851, pruned_loss=0.1488, over 5694090.38 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3598, pruned_loss=0.107, over 5663214.31 frames. ], batch size: 472, lr: 9.33e-03, grad_scale: 2.0
+2023-03-01 21:02:22,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5389, 1.0804, 2.9109, 2.5782], device='cuda:0'), covar=tensor([0.1469, 0.1871, 0.0450, 0.0550], device='cuda:0'), in_proj_covar=tensor([0.0521, 0.0495, 0.0659, 0.0520], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:0')
+2023-03-01 21:02:22,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.490e+02 1.466e+03 1.905e+03 2.621e+03 6.224e+03, threshold=3.810e+03, percent-clipped=9.0
+2023-03-01 21:03:03,532 INFO [train.py:968] (0/2) Epoch 3, batch 34250, giga_loss[loss=0.2889, simple_loss=0.364, pruned_loss=0.1069, over 28461.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3624, pruned_loss=0.1107, over 5668349.63 frames. ], libri_tot_loss[loss=0.3409, simple_loss=0.3847, pruned_loss=0.1486, over 5689326.85 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3596, pruned_loss=0.1057, over 5667217.25 frames. ], batch size: 369, lr: 9.33e-03, grad_scale: 2.0
+2023-03-01 21:03:41,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5295, 1.5606, 1.4935, 1.4264], device='cuda:0'), covar=tensor([0.0885, 0.1596, 0.1281, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0749, 0.0603, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 21:04:14,296 INFO [train.py:968] (0/2) Epoch 3, batch 34300, giga_loss[loss=0.3063, simple_loss=0.3844, pruned_loss=0.1141, over 28751.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3639, pruned_loss=0.1116, over 5661848.38 frames. ], libri_tot_loss[loss=0.3409, simple_loss=0.3847, pruned_loss=0.1485, over 5682778.32 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3612, pruned_loss=0.1068, over 5666211.45 frames. ], batch size: 243, lr: 9.33e-03, grad_scale: 2.0
+2023-03-01 21:04:32,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.596e+03 2.227e+03 3.218e+03 8.488e+03, threshold=4.454e+03, percent-clipped=19.0
+2023-03-01 21:05:17,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:05:17,540 INFO [train.py:968] (0/2) Epoch 3, batch 34350, giga_loss[loss=0.2538, simple_loss=0.3361, pruned_loss=0.08575, over 28630.00 frames. ], tot_loss[loss=0.295, simple_loss=0.366, pruned_loss=0.112, over 5670602.02 frames. ], libri_tot_loss[loss=0.3405, simple_loss=0.3845, pruned_loss=0.1483, over 5686539.33 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3636, pruned_loss=0.1076, over 5670477.80 frames. ], batch size: 85, lr: 9.33e-03, grad_scale: 2.0
+2023-03-01 21:05:52,793 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:06:29,842 INFO [train.py:968] (0/2) Epoch 3, batch 34400, giga_loss[loss=0.2419, simple_loss=0.3252, pruned_loss=0.07931, over 29084.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3646, pruned_loss=0.1112, over 5671324.46 frames. ], libri_tot_loss[loss=0.3411, simple_loss=0.385, pruned_loss=0.1486, over 5686488.40 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3621, pruned_loss=0.1071, over 5671018.88 frames. ], batch size: 128, lr: 9.33e-03, grad_scale: 4.0
+2023-03-01 21:06:55,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.600e+02 1.305e+03 1.592e+03 2.356e+03 5.959e+03, threshold=3.184e+03, percent-clipped=2.0
+2023-03-01 21:07:41,365 INFO [train.py:968] (0/2) Epoch 3, batch 34450, libri_loss[loss=0.3238, simple_loss=0.3818, pruned_loss=0.1329, over 29366.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3625, pruned_loss=0.1103, over 5681276.09 frames. ], libri_tot_loss[loss=0.3411, simple_loss=0.3852, pruned_loss=0.1485, over 5690008.61 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.36, pruned_loss=0.1064, over 5677426.32 frames. ], batch size: 92, lr: 9.32e-03, grad_scale: 4.0
+2023-03-01 21:07:42,384 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124741.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 21:08:37,004 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:08:40,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:08:49,739 INFO [train.py:968] (0/2) Epoch 3, batch 34500, giga_loss[loss=0.2805, simple_loss=0.3536, pruned_loss=0.1037, over 28960.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3606, pruned_loss=0.109, over 5669050.35 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3852, pruned_loss=0.1486, over 5682847.51 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3577, pruned_loss=0.1047, over 5672719.76 frames. ], batch size: 186, lr: 9.32e-03, grad_scale: 4.0
+2023-03-01 21:09:06,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:09:07,653 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.490e+02 1.352e+03 1.773e+03 2.333e+03 1.043e+04, threshold=3.546e+03, percent-clipped=14.0
+2023-03-01 21:09:16,121 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:09:29,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:09:36,121 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:09:41,038 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-01 21:09:53,649 INFO [train.py:968] (0/2) Epoch 3, batch 34550, giga_loss[loss=0.2479, simple_loss=0.3137, pruned_loss=0.09107, over 24245.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3585, pruned_loss=0.1075, over 5669642.40 frames. ], libri_tot_loss[loss=0.3416, simple_loss=0.3855, pruned_loss=0.1489, over 5685594.00 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3556, pruned_loss=0.103, over 5670072.48 frames. ], batch size: 705, lr: 9.32e-03, grad_scale: 4.0
+2023-03-01 21:10:52,519 INFO [train.py:968] (0/2) Epoch 3, batch 34600, giga_loss[loss=0.3257, simple_loss=0.3965, pruned_loss=0.1274, over 28370.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3615, pruned_loss=0.1101, over 5671470.65 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3851, pruned_loss=0.1486, over 5689161.77 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3589, pruned_loss=0.1059, over 5668320.36 frames. ], batch size: 368, lr: 9.32e-03, grad_scale: 4.0
+2023-03-01 21:11:10,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.349e+02 1.247e+03 1.664e+03 2.228e+03 1.216e+04, threshold=3.328e+03, percent-clipped=8.0
+2023-03-01 21:11:55,771 INFO [train.py:968] (0/2) Epoch 3, batch 34650, giga_loss[loss=0.2999, simple_loss=0.3694, pruned_loss=0.1152, over 28854.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3645, pruned_loss=0.1113, over 5674026.85 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3851, pruned_loss=0.1487, over 5690391.47 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3623, pruned_loss=0.1077, over 5670331.50 frames. ], batch size: 186, lr: 9.32e-03, grad_scale: 4.0
+2023-03-01 21:11:59,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4122, 2.9792, 1.6695, 1.5514], device='cuda:0'), covar=tensor([0.0769, 0.0435, 0.0533, 0.0701], device='cuda:0'), in_proj_covar=tensor([0.1230, 0.0925, 0.0954, 0.1046], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 21:12:01,153 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:12:03,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:12:31,038 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:12:34,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:12:37,232 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:12:49,581 INFO [train.py:968] (0/2) Epoch 3, batch 34700, giga_loss[loss=0.2722, simple_loss=0.3492, pruned_loss=0.09762, over 28948.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3633, pruned_loss=0.1118, over 5672103.01 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.3853, pruned_loss=0.1489, over 5682143.68 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3609, pruned_loss=0.1077, over 5675419.44 frames. ], batch size: 284, lr: 9.31e-03, grad_scale: 4.0
+2023-03-01 21:13:09,377 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:13:10,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.337e+02 1.302e+03 1.897e+03 2.660e+03 6.557e+03, threshold=3.794e+03, percent-clipped=12.0
+2023-03-01 21:13:37,247 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:13:48,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:13:48,396 INFO [train.py:968] (0/2) Epoch 3, batch 34750, giga_loss[loss=0.2713, simple_loss=0.3399, pruned_loss=0.1013, over 28134.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3614, pruned_loss=0.1121, over 5662405.40 frames. ], libri_tot_loss[loss=0.3417, simple_loss=0.3854, pruned_loss=0.149, over 5682843.29 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.359, pruned_loss=0.1081, over 5664060.86 frames. ], batch size: 412, lr: 9.31e-03, grad_scale: 4.0
+2023-03-01 21:14:08,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:14:42,487 INFO [train.py:968] (0/2) Epoch 3, batch 34800, giga_loss[loss=0.2775, simple_loss=0.3525, pruned_loss=0.1013, over 28949.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3614, pruned_loss=0.1129, over 5669741.44 frames. ], libri_tot_loss[loss=0.342, simple_loss=0.3855, pruned_loss=0.1492, over 5686304.65 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3585, pruned_loss=0.1082, over 5667337.96 frames. ], batch size: 284, lr: 9.31e-03, grad_scale: 8.0
+2023-03-01 21:15:00,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.606e+02 1.421e+03 1.938e+03 3.049e+03 8.234e+03, threshold=3.877e+03, percent-clipped=16.0
+2023-03-01 21:15:10,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125116.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 21:15:33,408 INFO [train.py:968] (0/2) Epoch 3, batch 34850, giga_loss[loss=0.312, simple_loss=0.39, pruned_loss=0.1169, over 28652.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3713, pruned_loss=0.1194, over 5668899.41 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3855, pruned_loss=0.1492, over 5686954.13 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.369, pruned_loss=0.1156, over 5666351.40 frames. ], batch size: 242, lr: 9.31e-03, grad_scale: 4.0
+2023-03-01 21:16:11,477 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:16:14,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:16:20,351 INFO [train.py:968] (0/2) Epoch 3, batch 34900, giga_loss[loss=0.4145, simple_loss=0.4519, pruned_loss=0.1885, over 28879.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.379, pruned_loss=0.1247, over 5671743.01 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.385, pruned_loss=0.1488, over 5690831.88 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3773, pruned_loss=0.1213, over 5665873.21 frames. ], batch size: 145, lr: 9.31e-03, grad_scale: 4.0
+2023-03-01 21:16:28,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:16:34,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.780e+02 1.228e+03 1.493e+03 1.898e+03 6.028e+03, threshold=2.986e+03, percent-clipped=1.0
+2023-03-01 21:16:41,923 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:16:43,872 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:17:04,466 INFO [train.py:968] (0/2) Epoch 3, batch 34950, giga_loss[loss=0.2957, simple_loss=0.3616, pruned_loss=0.1149, over 28715.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3817, pruned_loss=0.1271, over 5674968.16 frames. ], libri_tot_loss[loss=0.3414, simple_loss=0.3852, pruned_loss=0.1489, over 5692902.64 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3802, pruned_loss=0.1242, over 5668479.99 frames. ], batch size: 242, lr: 9.31e-03, grad_scale: 4.0
+2023-03-01 21:17:21,960 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125259.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 21:17:24,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125262.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 21:17:48,519 INFO [train.py:968] (0/2) Epoch 3, batch 35000, giga_loss[loss=0.2889, simple_loss=0.35, pruned_loss=0.1139, over 28597.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.376, pruned_loss=0.1251, over 5678670.77 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3856, pruned_loss=0.1491, over 5688304.54 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3742, pruned_loss=0.1221, over 5677835.21 frames. ], batch size: 60, lr: 9.30e-03, grad_scale: 4.0
+2023-03-01 21:17:48,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125291.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 21:18:01,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.376e+02 1.118e+03 1.594e+03 2.246e+03 5.546e+03, threshold=3.189e+03, percent-clipped=10.0
+2023-03-01 21:18:20,072 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-01 21:18:28,449 INFO [train.py:968] (0/2) Epoch 3, batch 35050, libri_loss[loss=0.4073, simple_loss=0.4302, pruned_loss=0.1922, over 19867.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3717, pruned_loss=0.1241, over 5681120.90 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3864, pruned_loss=0.1494, over 5687211.49 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3689, pruned_loss=0.1201, over 5681784.51 frames. ], batch size: 187, lr: 9.30e-03, grad_scale: 4.0
+2023-03-01 21:18:30,259 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:18:33,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:18:56,547 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:19:09,005 INFO [train.py:968] (0/2) Epoch 3, batch 35100, giga_loss[loss=0.2753, simple_loss=0.3337, pruned_loss=0.1084, over 27900.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3642, pruned_loss=0.1209, over 5682996.40 frames. ], libri_tot_loss[loss=0.3422, simple_loss=0.3861, pruned_loss=0.1491, over 5691588.52 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3619, pruned_loss=0.1175, over 5679459.96 frames. ], batch size: 412, lr: 9.30e-03, grad_scale: 4.0
+2023-03-01 21:19:21,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:19:21,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.807e+02 1.143e+03 1.367e+03 2.045e+03 6.077e+03, threshold=2.735e+03, percent-clipped=8.0
+2023-03-01 21:19:42,426 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:19:51,137 INFO [train.py:968] (0/2) Epoch 3, batch 35150, giga_loss[loss=0.2402, simple_loss=0.301, pruned_loss=0.0897, over 28600.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3565, pruned_loss=0.1174, over 5691100.61 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3864, pruned_loss=0.1492, over 5697471.02 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3538, pruned_loss=0.1138, over 5683162.19 frames. ], batch size: 71, lr: 9.30e-03, grad_scale: 4.0
+2023-03-01 21:20:35,737 INFO [train.py:968] (0/2) Epoch 3, batch 35200, giga_loss[loss=0.2479, simple_loss=0.3131, pruned_loss=0.09135, over 28863.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3501, pruned_loss=0.1142, over 5692979.54 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.387, pruned_loss=0.1494, over 5701226.44 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3467, pruned_loss=0.1105, over 5682899.61 frames. ], batch size: 243, lr: 9.30e-03, grad_scale: 8.0
+2023-03-01 21:20:36,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3521, 1.5939, 1.2462, 1.4348], device='cuda:0'), covar=tensor([0.0858, 0.0350, 0.0377, 0.0933], device='cuda:0'), in_proj_covar=tensor([0.0226, 0.0145, 0.0150, 0.0269], device='cuda:0'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:0')
+2023-03-01 21:20:48,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.774e+02 1.112e+03 1.519e+03 1.910e+03 4.537e+03, threshold=3.038e+03, percent-clipped=10.0
+2023-03-01 21:21:02,729 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:16,968 INFO [train.py:968] (0/2) Epoch 3, batch 35250, giga_loss[loss=0.2504, simple_loss=0.3192, pruned_loss=0.09078, over 29053.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3456, pruned_loss=0.1119, over 5693034.90 frames. ], libri_tot_loss[loss=0.3434, simple_loss=0.3875, pruned_loss=0.1496, over 5703421.73 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3419, pruned_loss=0.1083, over 5682950.31 frames. ], batch size: 136, lr: 9.29e-03, grad_scale: 8.0
+2023-03-01 21:21:25,739 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:28,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:46,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:48,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:51,451 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:58,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:22:00,419 INFO [train.py:968] (0/2) Epoch 3, batch 35300, giga_loss[loss=0.2507, simple_loss=0.3161, pruned_loss=0.09261, over 28596.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3438, pruned_loss=0.1112, over 5693992.73 frames. ], libri_tot_loss[loss=0.3443, simple_loss=0.3883, pruned_loss=0.1501, over 5701028.21 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3385, pruned_loss=0.1064, over 5687912.70 frames. ], batch size: 336, lr: 9.29e-03, grad_scale: 4.0
+2023-03-01 21:22:12,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4048, 4.0474, 4.1117, 1.6311], device='cuda:0'), covar=tensor([0.0376, 0.0352, 0.0654, 0.2136], device='cuda:0'), in_proj_covar=tensor([0.0747, 0.0612, 0.0748, 0.0551], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 21:22:12,778 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:22:13,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.023e+02 9.703e+02 1.299e+03 1.987e+03 6.600e+03, threshold=2.598e+03, percent-clipped=11.0
+2023-03-01 21:22:41,838 INFO [train.py:968] (0/2) Epoch 3, batch 35350, giga_loss[loss=0.2715, simple_loss=0.3301, pruned_loss=0.1065, over 29019.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3399, pruned_loss=0.1089, over 5700838.78 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3887, pruned_loss=0.1503, over 5701972.44 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3345, pruned_loss=0.1042, over 5695194.32 frames. ], batch size: 136, lr: 9.29e-03, grad_scale: 4.0
+2023-03-01 21:23:05,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5135, 1.8259, 1.6525, 1.6231], device='cuda:0'), covar=tensor([0.1446, 0.1814, 0.1168, 0.0995], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0777, 0.0743, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 21:23:26,966 INFO [train.py:968] (0/2) Epoch 3, batch 35400, libri_loss[loss=0.3122, simple_loss=0.3601, pruned_loss=0.1322, over 29652.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3378, pruned_loss=0.1078, over 5702881.61 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3895, pruned_loss=0.1507, over 5698842.50 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3314, pruned_loss=0.1025, over 5700986.03 frames. ], batch size: 69, lr: 9.29e-03, grad_scale: 4.0
+2023-03-01 21:23:38,239 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 21:23:41,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.579e+02 1.106e+03 1.484e+03 2.150e+03 9.386e+03, threshold=2.969e+03, percent-clipped=11.0
+2023-03-01 21:24:00,632 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:24:03,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:24:07,834 INFO [train.py:968] (0/2) Epoch 3, batch 35450, giga_loss[loss=0.3321, simple_loss=0.3541, pruned_loss=0.1551, over 23787.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3361, pruned_loss=0.1072, over 5701558.90 frames. ], libri_tot_loss[loss=0.3457, simple_loss=0.3899, pruned_loss=0.1508, over 5706044.50 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3283, pruned_loss=0.1011, over 5693330.56 frames. ], batch size: 705, lr: 9.29e-03, grad_scale: 4.0
+2023-03-01 21:24:13,237 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:24:26,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:24:30,939 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:24:37,266 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0593, 1.0119, 0.7516, 1.2451], device='cuda:0'), covar=tensor([0.0926, 0.0433, 0.0451, 0.1007], device='cuda:0'), in_proj_covar=tensor([0.0227, 0.0146, 0.0150, 0.0270], device='cuda:0'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:0')
+2023-03-01 21:24:47,866 INFO [train.py:968] (0/2) Epoch 3, batch 35500, giga_loss[loss=0.2885, simple_loss=0.354, pruned_loss=0.1115, over 28916.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3345, pruned_loss=0.1064, over 5690464.36 frames. ], libri_tot_loss[loss=0.346, simple_loss=0.3903, pruned_loss=0.1509, over 5695744.58 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3256, pruned_loss=0.09964, over 5693617.26 frames. ], batch size: 174, lr: 9.28e-03, grad_scale: 4.0
+2023-03-01 21:24:53,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5132, 1.7788, 1.6167, 1.6566], device='cuda:0'), covar=tensor([0.1309, 0.1660, 0.1049, 0.0934], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0782, 0.0746, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 21:25:01,922 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.772e+02 1.043e+03 1.250e+03 1.830e+03 5.939e+03, threshold=2.500e+03, percent-clipped=11.0
+2023-03-01 21:25:06,823 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:25:30,487 INFO [train.py:968] (0/2) Epoch 3, batch 35550, giga_loss[loss=0.2434, simple_loss=0.3089, pruned_loss=0.08894, over 29108.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3288, pruned_loss=0.1026, over 5687883.96 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3898, pruned_loss=0.1505, over 5696916.76 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3218, pruned_loss=0.09716, over 5689283.69 frames. ], batch size: 128, lr: 9.28e-03, grad_scale: 4.0
+2023-03-01 21:25:36,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4003, 2.0606, 1.5202, 0.7625], device='cuda:0'), covar=tensor([0.1687, 0.1042, 0.1920, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.1290, 0.1213, 0.1285, 0.1096], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 21:26:16,392 INFO [train.py:968] (0/2) Epoch 3, batch 35600, giga_loss[loss=0.2282, simple_loss=0.2953, pruned_loss=0.08055, over 28567.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3231, pruned_loss=0.09876, over 5695060.01 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.39, pruned_loss=0.1505, over 5697732.18 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3172, pruned_loss=0.09435, over 5695481.75 frames. ], batch size: 336, lr: 9.28e-03, grad_scale: 8.0
+2023-03-01 21:26:22,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:26:24,116 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-01 21:26:32,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.151e+02 1.022e+03 1.253e+03 1.630e+03 3.558e+03, threshold=2.505e+03, percent-clipped=7.0
+2023-03-01 21:26:54,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1856, 1.1852, 1.0707, 1.1148], device='cuda:0'), covar=tensor([0.0676, 0.0547, 0.1064, 0.0735], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0469, 0.0515, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 21:27:00,737 INFO [train.py:968] (0/2) Epoch 3, batch 35650, giga_loss[loss=0.3206, simple_loss=0.3742, pruned_loss=0.1335, over 28944.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3248, pruned_loss=0.1008, over 5688434.45 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.3907, pruned_loss=0.151, over 5700967.53 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3182, pruned_loss=0.09592, over 5685658.53 frames. ], batch size: 213, lr: 9.28e-03, grad_scale: 8.0
+2023-03-01 21:27:41,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5641, 1.5265, 1.1427, 0.9267], device='cuda:0'), covar=tensor([0.0708, 0.0634, 0.0528, 0.0673], device='cuda:0'), in_proj_covar=tensor([0.1214, 0.0913, 0.0947, 0.1045], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 21:27:44,568 INFO [train.py:968] (0/2) Epoch 3, batch 35700, libri_loss[loss=0.3608, simple_loss=0.4015, pruned_loss=0.16, over 29542.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3391, pruned_loss=0.1096, over 5694827.68 frames. ], libri_tot_loss[loss=0.3471, simple_loss=0.3913, pruned_loss=0.1514, over 5708267.73 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3309, pruned_loss=0.1035, over 5685303.73 frames. ], batch size: 80, lr: 9.28e-03, grad_scale: 4.0
+2023-03-01 21:27:49,465 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:27:53,016 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-126000.pt
+2023-03-01 21:28:02,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.086e+02 1.378e+03 1.792e+03 2.659e+03 6.722e+03, threshold=3.585e+03, percent-clipped=29.0
+2023-03-01 21:28:18,403 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-01 21:28:29,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:28:31,683 INFO [train.py:968] (0/2) Epoch 3, batch 35750, libri_loss[loss=0.3921, simple_loss=0.4272, pruned_loss=0.1785, over 29521.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3538, pruned_loss=0.1184, over 5687597.02 frames. ], libri_tot_loss[loss=0.3472, simple_loss=0.3914, pruned_loss=0.1515, over 5701991.92 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.346, pruned_loss=0.1124, over 5684986.36 frames. ], batch size: 81, lr: 9.28e-03, grad_scale: 4.0
+2023-03-01 21:28:32,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:28:56,439 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:29:11,502 INFO [train.py:968] (0/2) Epoch 3, batch 35800, giga_loss[loss=0.3411, simple_loss=0.4039, pruned_loss=0.1392, over 28498.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3665, pruned_loss=0.1253, over 5690839.07 frames. ], libri_tot_loss[loss=0.3478, simple_loss=0.3922, pruned_loss=0.1517, over 5698419.98 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3588, pruned_loss=0.1195, over 5691109.05 frames. ], batch size: 336, lr: 9.27e-03, grad_scale: 4.0
+2023-03-01 21:29:28,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.741e+02 1.336e+03 1.675e+03 2.446e+03 5.924e+03, threshold=3.350e+03, percent-clipped=9.0
+2023-03-01 21:29:35,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:29:40,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:29:54,646 INFO [train.py:968] (0/2) Epoch 3, batch 35850, giga_loss[loss=0.2797, simple_loss=0.3532, pruned_loss=0.1031, over 28507.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3727, pruned_loss=0.1274, over 5683876.94 frames. ], libri_tot_loss[loss=0.3482, simple_loss=0.3926, pruned_loss=0.152, over 5693472.89 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3658, pruned_loss=0.1221, over 5688733.65 frames. ], batch size: 71, lr: 9.27e-03, grad_scale: 4.0
+2023-03-01 21:29:56,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:30:30,591 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 21:30:40,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:30:41,726 INFO [train.py:968] (0/2) Epoch 3, batch 35900, giga_loss[loss=0.2982, simple_loss=0.3754, pruned_loss=0.1105, over 28677.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3741, pruned_loss=0.1258, over 5676871.00 frames. ], libri_tot_loss[loss=0.3486, simple_loss=0.393, pruned_loss=0.1521, over 5685535.82 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3682, pruned_loss=0.1214, over 5688214.64 frames. ], batch size: 242, lr: 9.27e-03, grad_scale: 4.0
+2023-03-01 21:30:58,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.208e+02 1.061e+03 1.327e+03 1.831e+03 4.738e+03, threshold=2.655e+03, percent-clipped=3.0
+2023-03-01 21:31:28,565 INFO [train.py:968] (0/2) Epoch 3, batch 35950, giga_loss[loss=0.2918, simple_loss=0.363, pruned_loss=0.1103, over 29070.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3759, pruned_loss=0.1259, over 5678834.79 frames. ], libri_tot_loss[loss=0.3488, simple_loss=0.3931, pruned_loss=0.1522, over 5687677.32 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.371, pruned_loss=0.1221, over 5685841.37 frames. ], batch size: 128, lr: 9.27e-03, grad_scale: 4.0
+2023-03-01 21:31:49,885 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:31:52,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:32:05,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4408, 1.5070, 1.3304, 1.5520], device='cuda:0'), covar=tensor([0.2114, 0.1886, 0.1699, 0.2009], device='cuda:0'), in_proj_covar=tensor([0.1039, 0.0829, 0.0927, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 21:32:07,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:32:11,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:32:12,164 INFO [train.py:968] (0/2) Epoch 3, batch 36000, giga_loss[loss=0.3335, simple_loss=0.3684, pruned_loss=0.1493, over 23382.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3788, pruned_loss=0.1292, over 5661107.76 frames. ], libri_tot_loss[loss=0.3493, simple_loss=0.3934, pruned_loss=0.1526, over 5674001.76 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3743, pruned_loss=0.1253, over 5678671.80 frames. ], batch size: 705, lr: 9.27e-03, grad_scale: 8.0
+2023-03-01 21:32:12,170 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 21:32:21,066 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2535, simple_loss=0.354, pruned_loss=0.07647, over 944034.00 frames. 
+2023-03-01 21:32:21,067 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 21:32:27,788 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:32:34,847 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.002e+02 1.083e+03 1.340e+03 2.082e+03 7.159e+03, threshold=2.681e+03, percent-clipped=12.0
+2023-03-01 21:32:43,950 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:32:55,771 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:32:57,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:33:02,482 INFO [train.py:968] (0/2) Epoch 3, batch 36050, giga_loss[loss=0.3314, simple_loss=0.4004, pruned_loss=0.1312, over 28869.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3822, pruned_loss=0.1316, over 5668900.65 frames. ], libri_tot_loss[loss=0.3493, simple_loss=0.3935, pruned_loss=0.1525, over 5677640.75 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3783, pruned_loss=0.1282, over 5679369.72 frames. ], batch size: 174, lr: 9.26e-03, grad_scale: 8.0
+2023-03-01 21:33:23,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:33:29,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:33:48,480 INFO [train.py:968] (0/2) Epoch 3, batch 36100, giga_loss[loss=0.318, simple_loss=0.3877, pruned_loss=0.1242, over 29081.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3844, pruned_loss=0.1324, over 5675008.34 frames. ], libri_tot_loss[loss=0.3496, simple_loss=0.3938, pruned_loss=0.1526, over 5678413.55 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.381, pruned_loss=0.1295, over 5682490.80 frames. ], batch size: 128, lr: 9.26e-03, grad_scale: 4.0
+2023-03-01 21:33:48,620 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:33:55,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4466, 1.4996, 1.1729, 0.8231], device='cuda:0'), covar=tensor([0.0713, 0.0620, 0.0531, 0.0734], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.0911, 0.0950, 0.1038], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 21:34:01,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.293e+02 1.157e+03 1.449e+03 1.871e+03 4.266e+03, threshold=2.899e+03, percent-clipped=5.0
+2023-03-01 21:34:26,185 INFO [train.py:968] (0/2) Epoch 3, batch 36150, giga_loss[loss=0.3151, simple_loss=0.3798, pruned_loss=0.1252, over 28576.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3876, pruned_loss=0.1337, over 5687228.21 frames. ], libri_tot_loss[loss=0.3501, simple_loss=0.3944, pruned_loss=0.1529, over 5682655.44 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3842, pruned_loss=0.1307, over 5689520.33 frames. ], batch size: 336, lr: 9.26e-03, grad_scale: 4.0
+2023-03-01 21:35:11,947 INFO [train.py:968] (0/2) Epoch 3, batch 36200, giga_loss[loss=0.3303, simple_loss=0.396, pruned_loss=0.1323, over 28881.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3892, pruned_loss=0.1342, over 5674164.22 frames. ], libri_tot_loss[loss=0.3503, simple_loss=0.3946, pruned_loss=0.153, over 5682766.16 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3863, pruned_loss=0.1316, over 5675806.24 frames. ], batch size: 199, lr: 9.26e-03, grad_scale: 4.0
+2023-03-01 21:35:14,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:35:25,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.158e+02 1.175e+03 1.461e+03 2.071e+03 6.840e+03, threshold=2.922e+03, percent-clipped=10.0
+2023-03-01 21:35:28,166 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:35:30,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:35:49,581 INFO [train.py:968] (0/2) Epoch 3, batch 36250, giga_loss[loss=0.2915, simple_loss=0.3666, pruned_loss=0.1082, over 28666.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3891, pruned_loss=0.1326, over 5677354.29 frames. ], libri_tot_loss[loss=0.351, simple_loss=0.3951, pruned_loss=0.1535, over 5672659.44 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3861, pruned_loss=0.1297, over 5688257.22 frames. ], batch size: 78, lr: 9.26e-03, grad_scale: 4.0
+2023-03-01 21:35:54,237 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:36:30,057 INFO [train.py:968] (0/2) Epoch 3, batch 36300, giga_loss[loss=0.328, simple_loss=0.3743, pruned_loss=0.1409, over 23530.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.387, pruned_loss=0.1299, over 5678157.93 frames. ], libri_tot_loss[loss=0.351, simple_loss=0.395, pruned_loss=0.1535, over 5666274.49 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3847, pruned_loss=0.1274, over 5692218.14 frames. ], batch size: 705, lr: 9.26e-03, grad_scale: 4.0
+2023-03-01 21:36:45,998 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.787e+02 1.028e+03 1.429e+03 2.160e+03 1.011e+04, threshold=2.858e+03, percent-clipped=17.0
+2023-03-01 21:36:52,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4533, 2.8212, 1.7294, 1.1676], device='cuda:0'), covar=tensor([0.0658, 0.0377, 0.0488, 0.0728], device='cuda:0'), in_proj_covar=tensor([0.1191, 0.0901, 0.0944, 0.1032], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 21:37:06,098 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:37:08,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:37:09,092 INFO [train.py:968] (0/2) Epoch 3, batch 36350, giga_loss[loss=0.3056, simple_loss=0.3768, pruned_loss=0.1172, over 28747.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3849, pruned_loss=0.1271, over 5694933.77 frames. ], libri_tot_loss[loss=0.3516, simple_loss=0.3958, pruned_loss=0.1537, over 5671576.96 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3821, pruned_loss=0.1242, over 5701815.06 frames. ], batch size: 92, lr: 9.25e-03, grad_scale: 4.0
+2023-03-01 21:37:31,054 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:37:31,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5123, 1.3193, 1.5331, 1.4058], device='cuda:0'), covar=tensor([0.0821, 0.1288, 0.1261, 0.1058], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0758, 0.0618, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 21:37:51,104 INFO [train.py:968] (0/2) Epoch 3, batch 36400, giga_loss[loss=0.3864, simple_loss=0.425, pruned_loss=0.1739, over 28132.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3849, pruned_loss=0.128, over 5705739.06 frames. ], libri_tot_loss[loss=0.3521, simple_loss=0.3961, pruned_loss=0.1541, over 5679727.74 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.382, pruned_loss=0.1243, over 5704792.59 frames. ], batch size: 368, lr: 9.25e-03, grad_scale: 8.0
+2023-03-01 21:38:08,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.865e+02 1.105e+03 1.562e+03 2.169e+03 8.490e+03, threshold=3.123e+03, percent-clipped=13.0
+2023-03-01 21:38:35,896 INFO [train.py:968] (0/2) Epoch 3, batch 36450, giga_loss[loss=0.3337, simple_loss=0.3884, pruned_loss=0.1395, over 28880.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3875, pruned_loss=0.1326, over 5707841.02 frames. ], libri_tot_loss[loss=0.3527, simple_loss=0.3966, pruned_loss=0.1544, over 5685241.67 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3844, pruned_loss=0.1288, over 5702647.52 frames. ], batch size: 119, lr: 9.25e-03, grad_scale: 4.0
+2023-03-01 21:38:42,225 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 21:38:58,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:39:05,236 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:39:20,312 INFO [train.py:968] (0/2) Epoch 3, batch 36500, giga_loss[loss=0.355, simple_loss=0.4048, pruned_loss=0.1526, over 28172.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3907, pruned_loss=0.1375, over 5695379.11 frames. ], libri_tot_loss[loss=0.3533, simple_loss=0.3971, pruned_loss=0.1547, over 5677224.22 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3877, pruned_loss=0.1338, over 5698816.88 frames. ], batch size: 368, lr: 9.25e-03, grad_scale: 4.0
+2023-03-01 21:39:35,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.156e+02 1.342e+03 1.696e+03 2.307e+03 3.971e+03, threshold=3.392e+03, percent-clipped=9.0
+2023-03-01 21:39:53,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5877, 3.0085, 1.5489, 1.3692], device='cuda:0'), covar=tensor([0.0824, 0.0351, 0.0801, 0.1359], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0450, 0.0305, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 21:40:05,143 INFO [train.py:968] (0/2) Epoch 3, batch 36550, giga_loss[loss=0.3241, simple_loss=0.3824, pruned_loss=0.1329, over 28856.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3898, pruned_loss=0.1376, over 5698899.82 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.3971, pruned_loss=0.1546, over 5679382.06 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3875, pruned_loss=0.1346, over 5699975.48 frames. ], batch size: 86, lr: 9.25e-03, grad_scale: 4.0
+2023-03-01 21:40:06,998 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:40:11,938 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-01 21:40:41,479 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-01 21:40:47,567 INFO [train.py:968] (0/2) Epoch 3, batch 36600, giga_loss[loss=0.2981, simple_loss=0.3614, pruned_loss=0.1174, over 28547.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3877, pruned_loss=0.137, over 5688828.83 frames. ], libri_tot_loss[loss=0.354, simple_loss=0.3977, pruned_loss=0.1552, over 5672752.55 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3851, pruned_loss=0.1338, over 5696236.80 frames. ], batch size: 65, lr: 9.24e-03, grad_scale: 4.0
+2023-03-01 21:41:04,584 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:41:06,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.446e+02 1.211e+03 1.511e+03 2.054e+03 5.401e+03, threshold=3.022e+03, percent-clipped=8.0
+2023-03-01 21:41:07,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:41:29,456 INFO [train.py:968] (0/2) Epoch 3, batch 36650, giga_loss[loss=0.3409, simple_loss=0.3945, pruned_loss=0.1437, over 28467.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3876, pruned_loss=0.137, over 5696288.30 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.3981, pruned_loss=0.1554, over 5678637.20 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3848, pruned_loss=0.1337, over 5697268.66 frames. ], batch size: 71, lr: 9.24e-03, grad_scale: 4.0
+2023-03-01 21:41:29,686 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:42:11,703 INFO [train.py:968] (0/2) Epoch 3, batch 36700, giga_loss[loss=0.3126, simple_loss=0.3797, pruned_loss=0.1227, over 28678.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3868, pruned_loss=0.1354, over 5689055.45 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.3986, pruned_loss=0.1557, over 5672066.89 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3841, pruned_loss=0.1323, over 5695240.61 frames. ], batch size: 284, lr: 9.24e-03, grad_scale: 4.0
+2023-03-01 21:42:30,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.312e+02 1.127e+03 1.568e+03 2.112e+03 7.706e+03, threshold=3.137e+03, percent-clipped=9.0
+2023-03-01 21:42:56,114 INFO [train.py:968] (0/2) Epoch 3, batch 36750, giga_loss[loss=0.2841, simple_loss=0.3569, pruned_loss=0.1057, over 28949.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3839, pruned_loss=0.1332, over 5688401.23 frames. ], libri_tot_loss[loss=0.3546, simple_loss=0.3983, pruned_loss=0.1555, over 5680216.26 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3816, pruned_loss=0.1302, over 5686182.98 frames. ], batch size: 213, lr: 9.24e-03, grad_scale: 4.0
+2023-03-01 21:43:39,151 INFO [train.py:968] (0/2) Epoch 3, batch 36800, giga_loss[loss=0.2827, simple_loss=0.3428, pruned_loss=0.1113, over 27859.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3781, pruned_loss=0.1297, over 5680469.62 frames. ], libri_tot_loss[loss=0.3555, simple_loss=0.3992, pruned_loss=0.1559, over 5687879.47 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3748, pruned_loss=0.126, over 5671939.10 frames. ], batch size: 412, lr: 9.24e-03, grad_scale: 8.0
+2023-03-01 21:43:58,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.787e+02 9.972e+02 1.317e+03 1.740e+03 5.462e+03, threshold=2.635e+03, percent-clipped=5.0
+2023-03-01 21:43:58,517 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4175, 1.4916, 1.1044, 0.9339], device='cuda:0'), covar=tensor([0.0663, 0.0555, 0.0537, 0.0720], device='cuda:0'), in_proj_covar=tensor([0.1209, 0.0915, 0.0975, 0.1031], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 21:44:23,681 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7778, 1.5719, 1.2013, 1.3141], device='cuda:0'), covar=tensor([0.0602, 0.0641, 0.0858, 0.1037], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0472, 0.0509, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 21:44:31,238 INFO [train.py:968] (0/2) Epoch 3, batch 36850, giga_loss[loss=0.2727, simple_loss=0.3402, pruned_loss=0.1026, over 28897.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3706, pruned_loss=0.1253, over 5667895.03 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.3994, pruned_loss=0.156, over 5688010.70 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3675, pruned_loss=0.122, over 5660860.83 frames. ], batch size: 213, lr: 9.24e-03, grad_scale: 8.0
+2023-03-01 21:44:39,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2182, 1.3224, 1.1932, 1.2101], device='cuda:0'), covar=tensor([0.1928, 0.1788, 0.1641, 0.1785], device='cuda:0'), in_proj_covar=tensor([0.1036, 0.0819, 0.0921, 0.0932], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 21:44:39,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:45:24,044 INFO [train.py:968] (0/2) Epoch 3, batch 36900, giga_loss[loss=0.2798, simple_loss=0.3503, pruned_loss=0.1047, over 28952.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3657, pruned_loss=0.1226, over 5656019.91 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.3998, pruned_loss=0.1562, over 5688277.15 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3622, pruned_loss=0.1192, over 5650062.91 frames. ], batch size: 186, lr: 9.23e-03, grad_scale: 4.0
+2023-03-01 21:45:38,724 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:45:40,831 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.922e+02 8.801e+02 1.081e+03 1.547e+03 5.677e+03, threshold=2.162e+03, percent-clipped=5.0
+2023-03-01 21:45:46,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:46:06,306 INFO [train.py:968] (0/2) Epoch 3, batch 36950, giga_loss[loss=0.292, simple_loss=0.3597, pruned_loss=0.1121, over 28740.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3658, pruned_loss=0.122, over 5675039.80 frames. ], libri_tot_loss[loss=0.3564, simple_loss=0.4001, pruned_loss=0.1563, over 5694809.14 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3615, pruned_loss=0.1179, over 5663103.04 frames. ], batch size: 284, lr: 9.23e-03, grad_scale: 4.0
+2023-03-01 21:46:23,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6330, 2.1743, 1.8433, 1.7603], device='cuda:0'), covar=tensor([0.1315, 0.1569, 0.1122, 0.0951], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0774, 0.0737, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 21:46:43,911 INFO [train.py:968] (0/2) Epoch 3, batch 37000, giga_loss[loss=0.3087, simple_loss=0.3657, pruned_loss=0.1259, over 28743.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3662, pruned_loss=0.1218, over 5679030.71 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.4006, pruned_loss=0.1564, over 5699549.20 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3613, pruned_loss=0.1175, over 5664769.93 frames. ], batch size: 92, lr: 9.23e-03, grad_scale: 4.0
+2023-03-01 21:46:45,439 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:46:47,415 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:47:01,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.898e+02 1.005e+03 1.247e+03 1.836e+03 9.425e+03, threshold=2.494e+03, percent-clipped=18.0
+2023-03-01 21:47:09,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:47:25,899 INFO [train.py:968] (0/2) Epoch 3, batch 37050, giga_loss[loss=0.3106, simple_loss=0.372, pruned_loss=0.1246, over 28737.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3653, pruned_loss=0.1212, over 5688747.85 frames. ], libri_tot_loss[loss=0.3572, simple_loss=0.4011, pruned_loss=0.1567, over 5697757.54 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3596, pruned_loss=0.1162, over 5678943.93 frames. ], batch size: 262, lr: 9.23e-03, grad_scale: 4.0
+2023-03-01 21:47:40,563 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:47:42,972 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:47:46,043 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 21:48:05,036 INFO [train.py:968] (0/2) Epoch 3, batch 37100, giga_loss[loss=0.2781, simple_loss=0.3446, pruned_loss=0.1058, over 28728.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3612, pruned_loss=0.1186, over 5699876.67 frames. ], libri_tot_loss[loss=0.358, simple_loss=0.4019, pruned_loss=0.157, over 5696816.97 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3554, pruned_loss=0.1137, over 5692739.89 frames. ], batch size: 284, lr: 9.23e-03, grad_scale: 4.0
+2023-03-01 21:48:07,025 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:48:19,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4914, 4.2129, 4.1919, 1.8532], device='cuda:0'), covar=tensor([0.0416, 0.0318, 0.0667, 0.2163], device='cuda:0'), in_proj_covar=tensor([0.0745, 0.0603, 0.0755, 0.0570], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 21:48:22,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.922e+02 1.159e+03 1.459e+03 2.056e+03 4.570e+03, threshold=2.919e+03, percent-clipped=14.0
+2023-03-01 21:48:46,490 INFO [train.py:968] (0/2) Epoch 3, batch 37150, giga_loss[loss=0.2538, simple_loss=0.3288, pruned_loss=0.08944, over 28894.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3596, pruned_loss=0.1178, over 5701223.48 frames. ], libri_tot_loss[loss=0.3588, simple_loss=0.4029, pruned_loss=0.1574, over 5697568.03 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3533, pruned_loss=0.1129, over 5695104.50 frames. ], batch size: 174, lr: 9.22e-03, grad_scale: 4.0
+2023-03-01 21:49:28,141 INFO [train.py:968] (0/2) Epoch 3, batch 37200, giga_loss[loss=0.2636, simple_loss=0.3276, pruned_loss=0.09979, over 28862.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3565, pruned_loss=0.1161, over 5707293.99 frames. ], libri_tot_loss[loss=0.3593, simple_loss=0.4033, pruned_loss=0.1577, over 5698749.51 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3505, pruned_loss=0.1114, over 5701566.14 frames. ], batch size: 99, lr: 9.22e-03, grad_scale: 8.0
+2023-03-01 21:49:44,469 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.110e+02 1.020e+03 1.250e+03 1.690e+03 9.402e+03, threshold=2.499e+03, percent-clipped=5.0
+2023-03-01 21:50:09,424 INFO [train.py:968] (0/2) Epoch 3, batch 37250, giga_loss[loss=0.2768, simple_loss=0.3382, pruned_loss=0.1077, over 28674.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3537, pruned_loss=0.1143, over 5713317.54 frames. ], libri_tot_loss[loss=0.3599, simple_loss=0.4041, pruned_loss=0.1579, over 5701347.05 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3477, pruned_loss=0.1098, over 5706721.50 frames. ], batch size: 92, lr: 9.22e-03, grad_scale: 8.0
+2023-03-01 21:50:24,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3032, 1.8916, 1.4033, 0.6359], device='cuda:0'), covar=tensor([0.1996, 0.1087, 0.1479, 0.2297], device='cuda:0'), in_proj_covar=tensor([0.1269, 0.1194, 0.1272, 0.1093], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 21:50:29,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6145, 3.3749, 3.2838, 1.5485], device='cuda:0'), covar=tensor([0.0649, 0.0460, 0.0921, 0.2550], device='cuda:0'), in_proj_covar=tensor([0.0758, 0.0609, 0.0766, 0.0571], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 21:50:43,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:50:46,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7087, 1.6005, 1.4533, 1.5128], device='cuda:0'), covar=tensor([0.1530, 0.1961, 0.1331, 0.1071], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0778, 0.0749, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 21:50:48,527 INFO [train.py:968] (0/2) Epoch 3, batch 37300, giga_loss[loss=0.2671, simple_loss=0.3286, pruned_loss=0.1028, over 28818.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3523, pruned_loss=0.1138, over 5719933.93 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.405, pruned_loss=0.1582, over 5705800.58 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3454, pruned_loss=0.1087, over 5711143.28 frames. ], batch size: 92, lr: 9.22e-03, grad_scale: 8.0
+2023-03-01 21:51:04,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.216e+02 1.048e+03 1.329e+03 1.974e+03 6.070e+03, threshold=2.659e+03, percent-clipped=17.0
+2023-03-01 21:51:27,272 INFO [train.py:968] (0/2) Epoch 3, batch 37350, libri_loss[loss=0.3381, simple_loss=0.3889, pruned_loss=0.1436, over 29471.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3519, pruned_loss=0.1139, over 5712660.51 frames. ], libri_tot_loss[loss=0.3624, simple_loss=0.4064, pruned_loss=0.1592, over 5698362.08 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3431, pruned_loss=0.1073, over 5712051.48 frames. ], batch size: 70, lr: 9.22e-03, grad_scale: 8.0
+2023-03-01 21:51:28,240 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:51:33,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4539, 1.9584, 1.6454, 1.6095], device='cuda:0'), covar=tensor([0.1203, 0.1677, 0.1148, 0.0956], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0777, 0.0750, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 21:52:06,012 INFO [train.py:968] (0/2) Epoch 3, batch 37400, giga_loss[loss=0.3642, simple_loss=0.3991, pruned_loss=0.1646, over 26576.00 frames. ], tot_loss[loss=0.289, simple_loss=0.351, pruned_loss=0.1135, over 5713101.29 frames. ], libri_tot_loss[loss=0.3644, simple_loss=0.408, pruned_loss=0.1603, over 5698738.20 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.341, pruned_loss=0.1061, over 5712553.22 frames. ], batch size: 555, lr: 9.22e-03, grad_scale: 2.0
+2023-03-01 21:52:26,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.077e+02 9.331e+02 1.213e+03 1.772e+03 1.145e+04, threshold=2.426e+03, percent-clipped=12.0
+2023-03-01 21:52:36,833 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:52:39,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:52:47,321 INFO [train.py:968] (0/2) Epoch 3, batch 37450, giga_loss[loss=0.316, simple_loss=0.3686, pruned_loss=0.1317, over 26809.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3485, pruned_loss=0.1116, over 5715011.58 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.4088, pruned_loss=0.1609, over 5702908.77 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3389, pruned_loss=0.1043, over 5711111.86 frames. ], batch size: 555, lr: 9.21e-03, grad_scale: 2.0
+2023-03-01 21:52:58,673 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-01 21:53:03,256 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:53:29,968 INFO [train.py:968] (0/2) Epoch 3, batch 37500, giga_loss[loss=0.2849, simple_loss=0.3501, pruned_loss=0.1098, over 28969.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3493, pruned_loss=0.1122, over 5715800.66 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4097, pruned_loss=0.1611, over 5704136.16 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3395, pruned_loss=0.1051, over 5712018.14 frames. ], batch size: 106, lr: 9.21e-03, grad_scale: 2.0
+2023-03-01 21:53:46,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.654e+02 1.044e+03 1.309e+03 1.980e+03 1.770e+04, threshold=2.619e+03, percent-clipped=17.0
+2023-03-01 21:53:52,588 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127820.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 21:54:11,699 INFO [train.py:968] (0/2) Epoch 3, batch 37550, libri_loss[loss=0.4636, simple_loss=0.491, pruned_loss=0.218, over 29536.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3548, pruned_loss=0.1156, over 5720534.55 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4106, pruned_loss=0.1616, over 5709319.75 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3441, pruned_loss=0.1079, over 5713346.76 frames. ], batch size: 89, lr: 9.21e-03, grad_scale: 2.0
+2023-03-01 21:54:53,758 INFO [train.py:968] (0/2) Epoch 3, batch 37600, giga_loss[loss=0.3549, simple_loss=0.4156, pruned_loss=0.1471, over 28929.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3625, pruned_loss=0.1214, over 5718886.67 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.4113, pruned_loss=0.1621, over 5709678.05 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3519, pruned_loss=0.1136, over 5713450.83 frames. ], batch size: 164, lr: 9.21e-03, grad_scale: 4.0
+2023-03-01 21:55:15,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.763e+02 1.138e+03 1.380e+03 1.909e+03 7.029e+03, threshold=2.759e+03, percent-clipped=13.0
+2023-03-01 21:55:26,040 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:55:41,692 INFO [train.py:968] (0/2) Epoch 3, batch 37650, giga_loss[loss=0.3882, simple_loss=0.4117, pruned_loss=0.1823, over 23685.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.372, pruned_loss=0.1285, over 5696352.19 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4111, pruned_loss=0.1621, over 5705108.40 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3628, pruned_loss=0.1215, over 5696877.62 frames. ], batch size: 710, lr: 9.21e-03, grad_scale: 4.0
+2023-03-01 21:56:26,111 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 21:56:34,587 INFO [train.py:968] (0/2) Epoch 3, batch 37700, giga_loss[loss=0.3128, simple_loss=0.3764, pruned_loss=0.1246, over 28913.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3771, pruned_loss=0.1311, over 5686509.42 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4112, pruned_loss=0.1621, over 5706107.85 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3696, pruned_loss=0.1255, over 5685844.27 frames. ], batch size: 106, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 21:56:43,695 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-128000.pt
+2023-03-01 21:56:56,048 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.743e+02 1.174e+03 1.532e+03 2.178e+03 5.709e+03, threshold=3.064e+03, percent-clipped=11.0
+2023-03-01 21:56:59,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:57:08,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1802, 1.3056, 1.1595, 0.9438], device='cuda:0'), covar=tensor([0.2046, 0.1885, 0.1798, 0.1789], device='cuda:0'), in_proj_covar=tensor([0.1047, 0.0823, 0.0926, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 21:57:12,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4954, 2.8133, 1.5350, 1.3249], device='cuda:0'), covar=tensor([0.0820, 0.0370, 0.0815, 0.1375], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0450, 0.0301, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 21:57:21,653 INFO [train.py:968] (0/2) Epoch 3, batch 37750, giga_loss[loss=0.3463, simple_loss=0.405, pruned_loss=0.1438, over 28717.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3821, pruned_loss=0.1327, over 5687829.98 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4112, pruned_loss=0.162, over 5707040.31 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3761, pruned_loss=0.1282, over 5686327.07 frames. ], batch size: 262, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 21:57:35,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:58:08,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4218, 1.4159, 1.2166, 1.5516], device='cuda:0'), covar=tensor([0.1866, 0.1736, 0.1572, 0.1907], device='cuda:0'), in_proj_covar=tensor([0.1046, 0.0831, 0.0929, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 21:58:08,488 INFO [train.py:968] (0/2) Epoch 3, batch 37800, giga_loss[loss=0.4273, simple_loss=0.459, pruned_loss=0.1978, over 27883.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3889, pruned_loss=0.1371, over 5684336.74 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4111, pruned_loss=0.1619, over 5709619.20 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3839, pruned_loss=0.1332, over 5680639.34 frames. ], batch size: 412, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 21:58:28,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.069e+02 1.101e+03 1.413e+03 2.006e+03 5.097e+03, threshold=2.826e+03, percent-clipped=8.0
+2023-03-01 21:58:51,516 INFO [train.py:968] (0/2) Epoch 3, batch 37850, giga_loss[loss=0.3009, simple_loss=0.3554, pruned_loss=0.1232, over 27675.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3875, pruned_loss=0.1358, over 5687286.93 frames. ], libri_tot_loss[loss=0.3682, simple_loss=0.4117, pruned_loss=0.1623, over 5711073.24 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3827, pruned_loss=0.132, over 5682792.95 frames. ], batch size: 472, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 21:59:07,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5382, 2.0323, 1.3826, 1.5837], device='cuda:0'), covar=tensor([0.0824, 0.0263, 0.0353, 0.0925], device='cuda:0'), in_proj_covar=tensor([0.0220, 0.0141, 0.0146, 0.0263], device='cuda:0'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:0')
+2023-03-01 21:59:07,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9174, 2.9395, 2.0169, 1.1860], device='cuda:0'), covar=tensor([0.2304, 0.0988, 0.1557, 0.2177], device='cuda:0'), in_proj_covar=tensor([0.1271, 0.1210, 0.1299, 0.1097], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 21:59:07,668 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:59:10,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:59:16,639 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:59:33,182 INFO [train.py:968] (0/2) Epoch 3, batch 37900, giga_loss[loss=0.2977, simple_loss=0.3696, pruned_loss=0.1129, over 28838.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3827, pruned_loss=0.1317, over 5696475.13 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4118, pruned_loss=0.1624, over 5713703.09 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3784, pruned_loss=0.1283, over 5690369.53 frames. ], batch size: 119, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 21:59:34,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:59:36,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128195.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 21:59:55,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.935e+02 1.089e+03 1.354e+03 1.659e+03 5.169e+03, threshold=2.707e+03, percent-clipped=5.0
+2023-03-01 22:00:16,034 INFO [train.py:968] (0/2) Epoch 3, batch 37950, giga_loss[loss=0.2984, simple_loss=0.3695, pruned_loss=0.1137, over 28680.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3805, pruned_loss=0.1297, over 5697438.91 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4126, pruned_loss=0.1631, over 5715103.07 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3757, pruned_loss=0.1257, over 5690919.32 frames. ], batch size: 92, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 22:00:19,246 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:00:37,244 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:00:58,693 INFO [train.py:968] (0/2) Epoch 3, batch 38000, giga_loss[loss=0.311, simple_loss=0.3787, pruned_loss=0.1217, over 28956.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3803, pruned_loss=0.129, over 5704432.68 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.4123, pruned_loss=0.163, over 5717662.07 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3762, pruned_loss=0.1253, over 5696723.07 frames. ], batch size: 213, lr: 9.19e-03, grad_scale: 8.0
+2023-03-01 22:01:03,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:01:14,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6570, 1.8584, 1.8058, 1.7321], device='cuda:0'), covar=tensor([0.1381, 0.1612, 0.1089, 0.0938], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0773, 0.0748, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 22:01:19,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.323e+02 1.156e+03 1.485e+03 2.122e+03 9.097e+03, threshold=2.969e+03, percent-clipped=11.0
+2023-03-01 22:01:41,525 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128338.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:01:43,120 INFO [train.py:968] (0/2) Epoch 3, batch 38050, libri_loss[loss=0.3801, simple_loss=0.4261, pruned_loss=0.1671, over 29536.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3855, pruned_loss=0.133, over 5704723.89 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4122, pruned_loss=0.1629, over 5722503.24 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3816, pruned_loss=0.1294, over 5693931.38 frames. ], batch size: 81, lr: 9.19e-03, grad_scale: 4.0
+2023-03-01 22:01:43,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5363, 3.0971, 1.4533, 1.3928], device='cuda:0'), covar=tensor([0.0781, 0.0294, 0.0777, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0441, 0.0302, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 22:01:43,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128341.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:02:07,347 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128370.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:02:23,631 INFO [train.py:968] (0/2) Epoch 3, batch 38100, giga_loss[loss=0.3288, simple_loss=0.3866, pruned_loss=0.1355, over 28761.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3876, pruned_loss=0.1344, over 5698186.44 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4123, pruned_loss=0.1629, over 5714310.13 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3837, pruned_loss=0.1307, over 5695498.12 frames. ], batch size: 99, lr: 9.19e-03, grad_scale: 4.0
+2023-03-01 22:02:44,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.439e+02 1.182e+03 1.529e+03 2.123e+03 7.415e+03, threshold=3.057e+03, percent-clipped=11.0
+2023-03-01 22:02:56,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:03:02,364 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-01 22:03:06,983 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:03:07,290 INFO [train.py:968] (0/2) Epoch 3, batch 38150, giga_loss[loss=0.3217, simple_loss=0.386, pruned_loss=0.1287, over 28968.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3903, pruned_loss=0.1365, over 5695848.12 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4129, pruned_loss=0.163, over 5714748.59 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3861, pruned_loss=0.1328, over 5693048.76 frames. ], batch size: 145, lr: 9.19e-03, grad_scale: 4.0
+2023-03-01 22:03:10,592 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:03:35,042 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:03:51,985 INFO [train.py:968] (0/2) Epoch 3, batch 38200, giga_loss[loss=0.3341, simple_loss=0.3881, pruned_loss=0.1401, over 28837.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3903, pruned_loss=0.1369, over 5700692.52 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.4131, pruned_loss=0.1632, over 5717262.93 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3866, pruned_loss=0.1334, over 5695842.80 frames. ], batch size: 112, lr: 9.19e-03, grad_scale: 4.0
+2023-03-01 22:04:13,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.483e+02 1.282e+03 1.629e+03 2.251e+03 4.964e+03, threshold=3.258e+03, percent-clipped=10.0
+2023-03-01 22:04:33,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9182, 1.1588, 3.9580, 3.1170], device='cuda:0'), covar=tensor([0.1614, 0.2054, 0.0313, 0.0541], device='cuda:0'), in_proj_covar=tensor([0.0532, 0.0496, 0.0685, 0.0550], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 22:04:34,839 INFO [train.py:968] (0/2) Epoch 3, batch 38250, giga_loss[loss=0.3245, simple_loss=0.3861, pruned_loss=0.1315, over 28749.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3909, pruned_loss=0.1377, over 5685190.37 frames. ], libri_tot_loss[loss=0.3698, simple_loss=0.4133, pruned_loss=0.1632, over 5709827.49 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3875, pruned_loss=0.1347, over 5688279.54 frames. ], batch size: 99, lr: 9.19e-03, grad_scale: 4.0
+2023-03-01 22:04:40,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:04:53,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7444, 1.7075, 1.2083, 1.4120], device='cuda:0'), covar=tensor([0.0613, 0.0593, 0.0962, 0.0890], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0463, 0.0517, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 22:05:06,004 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:05:08,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:05:19,162 INFO [train.py:968] (0/2) Epoch 3, batch 38300, giga_loss[loss=0.282, simple_loss=0.3585, pruned_loss=0.1027, over 28851.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3909, pruned_loss=0.1368, over 5683962.46 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4135, pruned_loss=0.1634, over 5700613.35 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.388, pruned_loss=0.1342, over 5694408.93 frames. ], batch size: 92, lr: 9.18e-03, grad_scale: 4.0
+2023-03-01 22:05:32,546 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:05:40,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.053e+02 1.048e+03 1.349e+03 1.781e+03 7.648e+03, threshold=2.698e+03, percent-clipped=5.0
+2023-03-01 22:05:42,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:06:01,835 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:06:02,220 INFO [train.py:968] (0/2) Epoch 3, batch 38350, giga_loss[loss=0.3927, simple_loss=0.4419, pruned_loss=0.1718, over 28664.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3904, pruned_loss=0.1349, over 5686733.34 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4135, pruned_loss=0.1634, over 5699705.50 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3878, pruned_loss=0.1325, over 5695560.01 frames. ], batch size: 92, lr: 9.18e-03, grad_scale: 4.0
+2023-03-01 22:06:11,170 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2176, 3.8833, 3.9314, 1.6458], device='cuda:0'), covar=tensor([0.0472, 0.0417, 0.0708, 0.2219], device='cuda:0'), in_proj_covar=tensor([0.0769, 0.0627, 0.0773, 0.0571], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-01 22:06:42,989 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:06:43,380 INFO [train.py:968] (0/2) Epoch 3, batch 38400, libri_loss[loss=0.3207, simple_loss=0.366, pruned_loss=0.1377, over 29641.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3903, pruned_loss=0.1339, over 5693421.62 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4138, pruned_loss=0.1637, over 5694789.65 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3874, pruned_loss=0.1311, over 5704328.53 frames. ], batch size: 69, lr: 9.18e-03, grad_scale: 8.0
+2023-03-01 22:06:45,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:06:48,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5153, 3.3598, 1.4749, 1.4273], device='cuda:0'), covar=tensor([0.0839, 0.0285, 0.0859, 0.1328], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0440, 0.0300, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 22:07:01,611 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:07:04,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.324e+02 1.153e+03 1.475e+03 1.966e+03 9.432e+03, threshold=2.950e+03, percent-clipped=15.0
+2023-03-01 22:07:10,342 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:07:24,225 INFO [train.py:968] (0/2) Epoch 3, batch 38450, giga_loss[loss=0.2799, simple_loss=0.358, pruned_loss=0.1009, over 28950.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3888, pruned_loss=0.1335, over 5698937.31 frames. ], libri_tot_loss[loss=0.3711, simple_loss=0.4141, pruned_loss=0.164, over 5700612.50 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3853, pruned_loss=0.1298, over 5702290.69 frames. ], batch size: 145, lr: 9.18e-03, grad_scale: 4.0
+2023-03-01 22:07:40,912 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:07:43,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:07:56,036 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 22:08:00,834 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:03,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:06,240 INFO [train.py:968] (0/2) Epoch 3, batch 38500, giga_loss[loss=0.3033, simple_loss=0.3634, pruned_loss=0.1216, over 28790.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3861, pruned_loss=0.1321, over 5702477.69 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4134, pruned_loss=0.1636, over 5705432.49 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.383, pruned_loss=0.1284, over 5700632.83 frames. ], batch size: 99, lr: 9.18e-03, grad_scale: 4.0
+2023-03-01 22:08:07,741 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:18,799 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:25,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:25,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.954e+02 9.744e+02 1.174e+03 1.677e+03 3.965e+03, threshold=2.348e+03, percent-clipped=5.0
+2023-03-01 22:08:44,789 INFO [train.py:968] (0/2) Epoch 3, batch 38550, giga_loss[loss=0.2956, simple_loss=0.3638, pruned_loss=0.1137, over 28882.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3844, pruned_loss=0.1312, over 5696046.66 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4133, pruned_loss=0.1635, over 5696860.16 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3811, pruned_loss=0.1273, over 5702459.68 frames. ], batch size: 199, lr: 9.17e-03, grad_scale: 4.0
+2023-03-01 22:09:17,170 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3806, 4.0216, 4.0710, 1.8555], device='cuda:0'), covar=tensor([0.0375, 0.0390, 0.0631, 0.2109], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0627, 0.0780, 0.0569], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 22:09:26,662 INFO [train.py:968] (0/2) Epoch 3, batch 38600, giga_loss[loss=0.2912, simple_loss=0.368, pruned_loss=0.1072, over 29004.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3842, pruned_loss=0.1312, over 5690169.76 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4135, pruned_loss=0.1635, over 5690151.86 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3811, pruned_loss=0.1277, over 5701539.13 frames. ], batch size: 155, lr: 9.17e-03, grad_scale: 4.0
+2023-03-01 22:09:42,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7053, 1.6370, 1.5760, 1.6206], device='cuda:0'), covar=tensor([0.1029, 0.1612, 0.1408, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0765, 0.0627, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 22:09:46,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.998e+02 1.034e+03 1.278e+03 1.918e+03 7.792e+03, threshold=2.557e+03, percent-clipped=17.0
+2023-03-01 22:10:07,916 INFO [train.py:968] (0/2) Epoch 3, batch 38650, giga_loss[loss=0.3171, simple_loss=0.3813, pruned_loss=0.1265, over 28830.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3836, pruned_loss=0.1306, over 5694505.60 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4133, pruned_loss=0.1636, over 5691935.44 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3805, pruned_loss=0.1269, over 5702143.81 frames. ], batch size: 119, lr: 9.17e-03, grad_scale: 4.0
+2023-03-01 22:10:33,956 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-01 22:10:46,032 INFO [train.py:968] (0/2) Epoch 3, batch 38700, giga_loss[loss=0.3103, simple_loss=0.3773, pruned_loss=0.1217, over 28770.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3834, pruned_loss=0.1295, over 5702095.82 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4134, pruned_loss=0.1635, over 5695368.10 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3805, pruned_loss=0.1262, over 5705191.77 frames. ], batch size: 99, lr: 9.17e-03, grad_scale: 4.0
+2023-03-01 22:10:54,239 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 22:11:07,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.428e+02 8.534e+02 1.099e+03 1.461e+03 4.195e+03, threshold=2.198e+03, percent-clipped=6.0
+2023-03-01 22:11:25,732 INFO [train.py:968] (0/2) Epoch 3, batch 38750, giga_loss[loss=0.3078, simple_loss=0.3785, pruned_loss=0.1185, over 28450.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.382, pruned_loss=0.1273, over 5708900.89 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4135, pruned_loss=0.1635, over 5696531.92 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3793, pruned_loss=0.1243, over 5710378.18 frames. ], batch size: 65, lr: 9.17e-03, grad_scale: 4.0
+2023-03-01 22:12:04,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:12:06,875 INFO [train.py:968] (0/2) Epoch 3, batch 38800, giga_loss[loss=0.3086, simple_loss=0.3528, pruned_loss=0.1322, over 23603.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3824, pruned_loss=0.1281, over 5693408.53 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4137, pruned_loss=0.1638, over 5689450.84 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3797, pruned_loss=0.1251, over 5701480.07 frames. ], batch size: 705, lr: 9.17e-03, grad_scale: 8.0
+2023-03-01 22:12:26,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.855e+02 1.020e+03 1.211e+03 1.575e+03 7.099e+03, threshold=2.421e+03, percent-clipped=12.0
+2023-03-01 22:12:43,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3798, 1.4601, 1.2734, 1.5677], device='cuda:0'), covar=tensor([0.2205, 0.1933, 0.1846, 0.1919], device='cuda:0'), in_proj_covar=tensor([0.1055, 0.0834, 0.0929, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 22:12:48,562 INFO [train.py:968] (0/2) Epoch 3, batch 38850, giga_loss[loss=0.2706, simple_loss=0.342, pruned_loss=0.09964, over 28460.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3793, pruned_loss=0.1267, over 5697635.78 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4136, pruned_loss=0.1638, over 5690660.59 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3771, pruned_loss=0.1241, over 5702831.35 frames. ], batch size: 77, lr: 9.16e-03, grad_scale: 8.0
+2023-03-01 22:12:58,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8241, 2.8779, 1.8011, 0.8109], device='cuda:0'), covar=tensor([0.2454, 0.0849, 0.1616, 0.2492], device='cuda:0'), in_proj_covar=tensor([0.1274, 0.1183, 0.1281, 0.1074], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 22:13:15,606 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=129174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:13:21,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:13:27,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-01 22:13:28,206 INFO [train.py:968] (0/2) Epoch 3, batch 38900, giga_loss[loss=0.2998, simple_loss=0.3652, pruned_loss=0.1172, over 28575.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3781, pruned_loss=0.1265, over 5688369.99 frames. ], libri_tot_loss[loss=0.3715, simple_loss=0.4144, pruned_loss=0.1643, over 5680348.39 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.374, pruned_loss=0.1223, over 5703198.48 frames. ], batch size: 307, lr: 9.16e-03, grad_scale: 4.0
+2023-03-01 22:13:48,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.277e+02 1.109e+03 1.425e+03 1.961e+03 5.239e+03, threshold=2.850e+03, percent-clipped=12.0
+2023-03-01 22:13:59,823 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:14:01,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:14:06,972 INFO [train.py:968] (0/2) Epoch 3, batch 38950, giga_loss[loss=0.2744, simple_loss=0.3463, pruned_loss=0.1013, over 28544.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3756, pruned_loss=0.1255, over 5699574.48 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4134, pruned_loss=0.1637, over 5688247.52 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3718, pruned_loss=0.1212, over 5704455.64 frames. ], batch size: 71, lr: 9.16e-03, grad_scale: 4.0
+2023-03-01 22:14:24,170 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:14:39,613 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=129283.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:14:45,756 INFO [train.py:968] (0/2) Epoch 3, batch 39000, giga_loss[loss=0.3294, simple_loss=0.3862, pruned_loss=0.1363, over 28849.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3765, pruned_loss=0.1268, over 5690915.19 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4132, pruned_loss=0.1635, over 5683205.73 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3721, pruned_loss=0.122, over 5700696.42 frames. ], batch size: 145, lr: 9.16e-03, grad_scale: 4.0
+2023-03-01 22:14:45,761 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 22:14:54,202 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2507, simple_loss=0.3503, pruned_loss=0.07558, over 944034.00 frames. 
+2023-03-01 22:14:54,203 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 22:15:16,483 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.754e+02 1.152e+03 1.537e+03 2.124e+03 6.506e+03, threshold=3.073e+03, percent-clipped=14.0
+2023-03-01 22:15:21,260 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:15:23,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:15:34,437 INFO [train.py:968] (0/2) Epoch 3, batch 39050, giga_loss[loss=0.3554, simple_loss=0.396, pruned_loss=0.1574, over 26650.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3749, pruned_loss=0.1263, over 5690248.94 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4132, pruned_loss=0.1634, over 5676675.32 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.371, pruned_loss=0.1221, over 5704219.58 frames. ], batch size: 555, lr: 9.16e-03, grad_scale: 4.0
+2023-03-01 22:15:47,787 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:15:54,970 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-01 22:15:59,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0432, 1.3595, 1.1245, 0.9736], device='cuda:0'), covar=tensor([0.2175, 0.2071, 0.1978, 0.1908], device='cuda:0'), in_proj_covar=tensor([0.1055, 0.0844, 0.0934, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 22:16:16,892 INFO [train.py:968] (0/2) Epoch 3, batch 39100, giga_loss[loss=0.316, simple_loss=0.3772, pruned_loss=0.1274, over 28376.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3729, pruned_loss=0.1256, over 5685167.59 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4135, pruned_loss=0.1635, over 5669968.37 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3688, pruned_loss=0.1215, over 5702834.43 frames. ], batch size: 368, lr: 9.16e-03, grad_scale: 4.0
+2023-03-01 22:16:33,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9989, 5.0561, 1.9271, 2.2634], device='cuda:0'), covar=tensor([0.0703, 0.0248, 0.0749, 0.1024], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0446, 0.0302, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 22:16:35,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.053e+02 9.770e+02 1.153e+03 1.537e+03 4.930e+03, threshold=2.306e+03, percent-clipped=7.0
+2023-03-01 22:16:54,061 INFO [train.py:968] (0/2) Epoch 3, batch 39150, giga_loss[loss=0.2677, simple_loss=0.3358, pruned_loss=0.09976, over 28000.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.371, pruned_loss=0.1252, over 5694820.89 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4136, pruned_loss=0.1635, over 5677939.98 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3662, pruned_loss=0.1206, over 5702421.09 frames. ], batch size: 77, lr: 9.15e-03, grad_scale: 4.0
+2023-03-01 22:17:35,857 INFO [train.py:968] (0/2) Epoch 3, batch 39200, libri_loss[loss=0.3345, simple_loss=0.3803, pruned_loss=0.1443, over 29571.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3683, pruned_loss=0.1235, over 5704297.12 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4134, pruned_loss=0.1633, over 5680029.39 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3637, pruned_loss=0.1192, over 5708956.92 frames. ], batch size: 74, lr: 9.15e-03, grad_scale: 8.0
+2023-03-01 22:17:37,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 22:17:57,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.395e+02 1.124e+03 1.602e+03 2.194e+03 5.793e+03, threshold=3.205e+03, percent-clipped=18.0
+2023-03-01 22:18:18,060 INFO [train.py:968] (0/2) Epoch 3, batch 39250, giga_loss[loss=0.2732, simple_loss=0.3366, pruned_loss=0.1049, over 28485.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.368, pruned_loss=0.1238, over 5703567.96 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4135, pruned_loss=0.1635, over 5681149.24 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.364, pruned_loss=0.1201, over 5706338.04 frames. ], batch size: 85, lr: 9.15e-03, grad_scale: 8.0
+2023-03-01 22:18:24,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:18:38,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3829, 1.6394, 1.2513, 1.4299], device='cuda:0'), covar=tensor([0.0826, 0.0331, 0.0390, 0.0966], device='cuda:0'), in_proj_covar=tensor([0.0219, 0.0142, 0.0147, 0.0262], device='cuda:0'), out_proj_covar=tensor([0.0044, 0.0033, 0.0029, 0.0050], device='cuda:0')
+2023-03-01 22:18:59,515 INFO [train.py:968] (0/2) Epoch 3, batch 39300, giga_loss[loss=0.2885, simple_loss=0.3625, pruned_loss=0.1072, over 28889.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3702, pruned_loss=0.1245, over 5710926.11 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4134, pruned_loss=0.1635, over 5687679.69 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3657, pruned_loss=0.1203, over 5708036.83 frames. ], batch size: 145, lr: 9.15e-03, grad_scale: 4.0
+2023-03-01 22:18:59,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0791, 3.2935, 2.1815, 0.8152], device='cuda:0'), covar=tensor([0.2410, 0.0901, 0.1655, 0.2981], device='cuda:0'), in_proj_covar=tensor([0.1293, 0.1210, 0.1306, 0.1093], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 22:19:19,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.994e+02 1.007e+03 1.227e+03 1.979e+03 9.902e+03, threshold=2.454e+03, percent-clipped=10.0
+2023-03-01 22:19:40,133 INFO [train.py:968] (0/2) Epoch 3, batch 39350, giga_loss[loss=0.2994, simple_loss=0.3788, pruned_loss=0.11, over 28985.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3734, pruned_loss=0.1261, over 5717169.94 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4126, pruned_loss=0.163, over 5695776.58 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3687, pruned_loss=0.1214, over 5708197.05 frames. ], batch size: 155, lr: 9.15e-03, grad_scale: 2.0
+2023-03-01 22:19:55,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129658.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:19:56,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3768, 1.3921, 1.1701, 1.7928], device='cuda:0'), covar=tensor([0.2027, 0.2036, 0.1935, 0.1857], device='cuda:0'), in_proj_covar=tensor([0.1043, 0.0831, 0.0921, 0.0939], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-01 22:20:23,857 INFO [train.py:968] (0/2) Epoch 3, batch 39400, giga_loss[loss=0.2961, simple_loss=0.3729, pruned_loss=0.1097, over 28623.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3744, pruned_loss=0.1255, over 5709842.16 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.4125, pruned_loss=0.1629, over 5697110.49 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3703, pruned_loss=0.1214, over 5701714.43 frames. ], batch size: 307, lr: 9.14e-03, grad_scale: 2.0
+2023-03-01 22:20:24,897 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:20:28,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:20:49,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.012e+02 9.868e+02 1.223e+03 1.542e+03 3.570e+03, threshold=2.446e+03, percent-clipped=4.0
+2023-03-01 22:20:53,634 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:21:09,815 INFO [train.py:968] (0/2) Epoch 3, batch 39450, giga_loss[loss=0.2761, simple_loss=0.3473, pruned_loss=0.1025, over 29035.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3734, pruned_loss=0.1236, over 5698137.36 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4127, pruned_loss=0.1629, over 5697176.94 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3699, pruned_loss=0.1202, over 5691760.85 frames. ], batch size: 106, lr: 9.14e-03, grad_scale: 2.0
+2023-03-01 22:21:27,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7054, 1.6233, 1.6128, 1.5240], device='cuda:0'), covar=tensor([0.0875, 0.1609, 0.1186, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.0439, 0.0738, 0.0612, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0008, 0.0007, 0.0008], device='cuda:0')
+2023-03-01 22:21:50,878 INFO [train.py:968] (0/2) Epoch 3, batch 39500, giga_loss[loss=0.2822, simple_loss=0.3639, pruned_loss=0.1003, over 28862.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3733, pruned_loss=0.1231, over 5702670.60 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4122, pruned_loss=0.1626, over 5703123.15 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3696, pruned_loss=0.1194, over 5692042.07 frames. ], batch size: 174, lr: 9.14e-03, grad_scale: 2.0
+2023-03-01 22:21:57,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129801.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:22:01,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129804.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:22:02,278 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=129806.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:22:12,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.945e+02 1.066e+03 1.425e+03 2.189e+03 5.339e+03, threshold=2.851e+03, percent-clipped=17.0
+2023-03-01 22:22:25,723 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129833.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:22:33,343 INFO [train.py:968] (0/2) Epoch 3, batch 39550, giga_loss[loss=0.3215, simple_loss=0.3887, pruned_loss=0.1272, over 28736.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3767, pruned_loss=0.1264, over 5699493.57 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4134, pruned_loss=0.1635, over 5703029.15 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3717, pruned_loss=0.1216, over 5690824.42 frames. ], batch size: 284, lr: 9.14e-03, grad_scale: 2.0
+2023-03-01 22:23:06,158 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-01 22:23:13,788 INFO [train.py:968] (0/2) Epoch 3, batch 39600, giga_loss[loss=0.2881, simple_loss=0.349, pruned_loss=0.1136, over 23733.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.377, pruned_loss=0.1266, over 5693649.84 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4138, pruned_loss=0.1637, over 5696431.78 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.372, pruned_loss=0.122, over 5692755.91 frames. ], batch size: 705, lr: 9.14e-03, grad_scale: 4.0
+2023-03-01 22:23:23,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-01 22:23:24,281 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 22:23:38,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.188e+02 1.125e+03 1.623e+03 2.050e+03 9.063e+03, threshold=3.247e+03, percent-clipped=6.0
+2023-03-01 22:23:58,131 INFO [train.py:968] (0/2) Epoch 3, batch 39650, libri_loss[loss=0.3955, simple_loss=0.4415, pruned_loss=0.1747, over 28626.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3796, pruned_loss=0.1283, over 5687488.27 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4139, pruned_loss=0.1636, over 5689724.01 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3749, pruned_loss=0.1241, over 5692439.11 frames. ], batch size: 106, lr: 9.14e-03, grad_scale: 4.0
+2023-03-01 22:24:02,782 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-01 22:24:26,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8096, 2.0480, 1.9393, 1.8406], device='cuda:0'), covar=tensor([0.1452, 0.1606, 0.1113, 0.0993], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0769, 0.0750, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 22:24:38,120 INFO [train.py:968] (0/2) Epoch 3, batch 39700, giga_loss[loss=0.3288, simple_loss=0.3939, pruned_loss=0.1319, over 28940.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3822, pruned_loss=0.1302, over 5697210.60 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4137, pruned_loss=0.1633, over 5694352.16 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3775, pruned_loss=0.1258, over 5697032.73 frames. ], batch size: 213, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:24:41,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1857, 1.1937, 0.9427, 1.2962], device='cuda:0'), covar=tensor([0.0831, 0.0388, 0.0403, 0.0930], device='cuda:0'), in_proj_covar=tensor([0.0220, 0.0143, 0.0147, 0.0262], device='cuda:0'), out_proj_covar=tensor([0.0044, 0.0033, 0.0029, 0.0050], device='cuda:0')
+2023-03-01 22:24:44,240 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-130000.pt
+2023-03-01 22:24:48,715 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:25:00,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.952e+02 1.172e+03 1.474e+03 2.085e+03 6.260e+03, threshold=2.948e+03, percent-clipped=4.0
+2023-03-01 22:25:13,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4647, 2.7996, 1.4857, 1.3550], device='cuda:0'), covar=tensor([0.0811, 0.0368, 0.0854, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0451, 0.0303, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 22:25:16,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8954, 1.1015, 3.8696, 3.1441], device='cuda:0'), covar=tensor([0.1567, 0.2131, 0.0340, 0.0606], device='cuda:0'), in_proj_covar=tensor([0.0530, 0.0497, 0.0685, 0.0552], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 22:25:18,584 INFO [train.py:968] (0/2) Epoch 3, batch 39750, giga_loss[loss=0.3402, simple_loss=0.3955, pruned_loss=0.1424, over 28890.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3845, pruned_loss=0.1308, over 5706949.37 frames. ], libri_tot_loss[loss=0.371, simple_loss=0.4145, pruned_loss=0.1638, over 5693255.25 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3796, pruned_loss=0.1263, over 5707985.75 frames. ], batch size: 186, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:26:01,284 INFO [train.py:968] (0/2) Epoch 3, batch 39800, libri_loss[loss=0.3293, simple_loss=0.3781, pruned_loss=0.1403, over 29440.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3856, pruned_loss=0.1316, over 5693093.66 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4152, pruned_loss=0.1644, over 5678454.23 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3804, pruned_loss=0.1269, over 5707837.85 frames. ], batch size: 67, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:26:23,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.830e+02 1.291e+03 1.804e+03 2.855e+03 8.301e+03, threshold=3.608e+03, percent-clipped=22.0
+2023-03-01 22:26:41,511 INFO [train.py:968] (0/2) Epoch 3, batch 39850, giga_loss[loss=0.3892, simple_loss=0.4354, pruned_loss=0.1714, over 28852.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3847, pruned_loss=0.1309, over 5700713.37 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4152, pruned_loss=0.1644, over 5683507.11 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3798, pruned_loss=0.1263, over 5708559.44 frames. ], batch size: 174, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:27:15,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130181.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:27:17,294 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:27:22,024 INFO [train.py:968] (0/2) Epoch 3, batch 39900, giga_loss[loss=0.2942, simple_loss=0.3551, pruned_loss=0.1166, over 28862.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3842, pruned_loss=0.1305, over 5705415.60 frames. ], libri_tot_loss[loss=0.3721, simple_loss=0.4153, pruned_loss=0.1645, over 5683400.61 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3799, pruned_loss=0.1264, over 5711988.86 frames. ], batch size: 86, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:27:43,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.933e+02 1.086e+03 1.325e+03 1.932e+03 5.716e+03, threshold=2.649e+03, percent-clipped=3.0
+2023-03-01 22:28:01,223 INFO [train.py:968] (0/2) Epoch 3, batch 39950, giga_loss[loss=0.2919, simple_loss=0.3611, pruned_loss=0.1113, over 29042.00 frames. ], tot_loss[loss=0.32, simple_loss=0.382, pruned_loss=0.129, over 5711819.27 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.4156, pruned_loss=0.1646, over 5687154.62 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3775, pruned_loss=0.1247, over 5714055.56 frames. ], batch size: 155, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:28:42,900 INFO [train.py:968] (0/2) Epoch 3, batch 40000, giga_loss[loss=0.2898, simple_loss=0.3605, pruned_loss=0.1095, over 28559.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3791, pruned_loss=0.1277, over 5703464.13 frames. ], libri_tot_loss[loss=0.3728, simple_loss=0.4159, pruned_loss=0.1648, over 5680500.40 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3748, pruned_loss=0.1238, over 5711898.12 frames. ], batch size: 307, lr: 9.12e-03, grad_scale: 8.0
+2023-03-01 22:29:07,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.498e+02 1.033e+03 1.282e+03 1.781e+03 4.079e+03, threshold=2.565e+03, percent-clipped=5.0
+2023-03-01 22:29:10,399 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130324.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:29:12,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130327.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:29:21,313 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7830, 3.2656, 1.5107, 1.5970], device='cuda:0'), covar=tensor([0.0717, 0.0300, 0.0825, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0450, 0.0304, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 22:29:22,921 INFO [train.py:968] (0/2) Epoch 3, batch 40050, giga_loss[loss=0.3199, simple_loss=0.3888, pruned_loss=0.1254, over 28946.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3761, pruned_loss=0.1256, over 5711550.96 frames. ], libri_tot_loss[loss=0.3731, simple_loss=0.4163, pruned_loss=0.1649, over 5685848.37 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3715, pruned_loss=0.1214, over 5714157.01 frames. ], batch size: 213, lr: 9.12e-03, grad_scale: 4.0
+2023-03-01 22:29:36,876 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130356.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:29:56,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:30:01,098 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-01 22:30:04,216 INFO [train.py:968] (0/2) Epoch 3, batch 40100, giga_loss[loss=0.2915, simple_loss=0.3724, pruned_loss=0.1053, over 28979.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3781, pruned_loss=0.1259, over 5703168.88 frames. ], libri_tot_loss[loss=0.3733, simple_loss=0.4164, pruned_loss=0.165, over 5679675.00 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3736, pruned_loss=0.1219, over 5711290.87 frames. ], batch size: 164, lr: 9.12e-03, grad_scale: 4.0
+2023-03-01 22:30:06,073 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:30:07,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:30:11,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5651, 1.6653, 1.2483, 1.4990], device='cuda:0'), covar=tensor([0.0718, 0.0292, 0.0377, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0219, 0.0142, 0.0146, 0.0262], device='cuda:0'), out_proj_covar=tensor([0.0044, 0.0033, 0.0029, 0.0050], device='cuda:0')
+2023-03-01 22:30:30,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.948e+02 1.151e+03 1.618e+03 2.315e+03 6.622e+03, threshold=3.236e+03, percent-clipped=19.0
+2023-03-01 22:30:47,962 INFO [train.py:968] (0/2) Epoch 3, batch 40150, giga_loss[loss=0.3339, simple_loss=0.3895, pruned_loss=0.1392, over 28943.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3788, pruned_loss=0.1254, over 5696138.67 frames. ], libri_tot_loss[loss=0.3733, simple_loss=0.4165, pruned_loss=0.1651, over 5680811.34 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.375, pruned_loss=0.1218, over 5701790.22 frames. ], batch size: 213, lr: 9.12e-03, grad_scale: 4.0
+2023-03-01 22:31:28,098 INFO [train.py:968] (0/2) Epoch 3, batch 40200, giga_loss[loss=0.2742, simple_loss=0.3527, pruned_loss=0.09789, over 28931.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3784, pruned_loss=0.1259, over 5707257.93 frames. ], libri_tot_loss[loss=0.3734, simple_loss=0.4165, pruned_loss=0.1652, over 5684228.03 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3748, pruned_loss=0.1225, over 5708888.36 frames. ], batch size: 155, lr: 9.12e-03, grad_scale: 4.0
+2023-03-01 22:31:51,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.268e+02 1.080e+03 1.368e+03 1.736e+03 4.806e+03, threshold=2.736e+03, percent-clipped=4.0
+2023-03-01 22:31:54,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:31:57,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:32:07,652 INFO [train.py:968] (0/2) Epoch 3, batch 40250, giga_loss[loss=0.3149, simple_loss=0.3717, pruned_loss=0.129, over 29060.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3759, pruned_loss=0.1256, over 5714564.92 frames. ], libri_tot_loss[loss=0.3734, simple_loss=0.4165, pruned_loss=0.1651, over 5687168.98 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3725, pruned_loss=0.1224, over 5713492.86 frames. ], batch size: 155, lr: 9.11e-03, grad_scale: 4.0
+2023-03-01 22:32:18,903 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:32:22,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:32:50,436 INFO [train.py:968] (0/2) Epoch 3, batch 40300, giga_loss[loss=0.2981, simple_loss=0.3535, pruned_loss=0.1213, over 28763.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3747, pruned_loss=0.1267, over 5714316.27 frames. ], libri_tot_loss[loss=0.3734, simple_loss=0.4166, pruned_loss=0.1651, over 5687234.50 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3712, pruned_loss=0.1235, over 5713687.05 frames. ], batch size: 119, lr: 9.11e-03, grad_scale: 4.0
+2023-03-01 22:33:14,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.948e+02 1.056e+03 1.397e+03 1.908e+03 4.808e+03, threshold=2.794e+03, percent-clipped=9.0
+2023-03-01 22:33:32,457 INFO [train.py:968] (0/2) Epoch 3, batch 40350, giga_loss[loss=0.3366, simple_loss=0.3799, pruned_loss=0.1466, over 24100.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.372, pruned_loss=0.1261, over 5704784.88 frames. ], libri_tot_loss[loss=0.3737, simple_loss=0.4169, pruned_loss=0.1652, over 5688707.79 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3685, pruned_loss=0.1231, over 5703054.55 frames. ], batch size: 705, lr: 9.11e-03, grad_scale: 4.0
+2023-03-01 22:33:47,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6289, 4.1668, 1.7305, 1.7270], device='cuda:0'), covar=tensor([0.1104, 0.0360, 0.1017, 0.1535], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0454, 0.0307, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 22:34:13,996 INFO [train.py:968] (0/2) Epoch 3, batch 40400, giga_loss[loss=0.319, simple_loss=0.3778, pruned_loss=0.1301, over 28977.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3721, pruned_loss=0.1267, over 5691155.15 frames. ], libri_tot_loss[loss=0.3745, simple_loss=0.4174, pruned_loss=0.1658, over 5671806.60 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3683, pruned_loss=0.1232, over 5705201.22 frames. ], batch size: 227, lr: 9.11e-03, grad_scale: 8.0
+2023-03-01 22:34:24,201 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:34:26,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:34:37,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.047e+02 1.124e+03 1.444e+03 1.862e+03 5.905e+03, threshold=2.888e+03, percent-clipped=10.0
+2023-03-01 22:34:47,748 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:34:51,868 INFO [train.py:968] (0/2) Epoch 3, batch 40450, giga_loss[loss=0.2888, simple_loss=0.3553, pruned_loss=0.1111, over 28835.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3707, pruned_loss=0.1264, over 5698678.16 frames. ], libri_tot_loss[loss=0.3748, simple_loss=0.4177, pruned_loss=0.1659, over 5677187.25 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3655, pruned_loss=0.1219, over 5706003.10 frames. ], batch size: 199, lr: 9.11e-03, grad_scale: 4.0
+2023-03-01 22:35:14,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:35:16,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:35:20,551 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:35:31,785 INFO [train.py:968] (0/2) Epoch 3, batch 40500, giga_loss[loss=0.3715, simple_loss=0.4063, pruned_loss=0.1684, over 26682.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3657, pruned_loss=0.1236, over 5700223.27 frames. ], libri_tot_loss[loss=0.3743, simple_loss=0.4174, pruned_loss=0.1656, over 5679793.38 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.361, pruned_loss=0.1195, over 5704211.06 frames. ], batch size: 555, lr: 9.11e-03, grad_scale: 4.0
+2023-03-01 22:35:55,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.883e+02 1.354e+03 1.635e+03 2.078e+03 5.755e+03, threshold=3.270e+03, percent-clipped=10.0
+2023-03-01 22:36:00,076 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:36:05,266 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:36:06,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6773, 3.3972, 1.6955, 1.5641], device='cuda:0'), covar=tensor([0.0824, 0.0402, 0.0839, 0.1312], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0452, 0.0308, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 22:36:11,891 INFO [train.py:968] (0/2) Epoch 3, batch 40550, giga_loss[loss=0.2537, simple_loss=0.3302, pruned_loss=0.08853, over 28809.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3622, pruned_loss=0.1214, over 5709848.15 frames. ], libri_tot_loss[loss=0.3738, simple_loss=0.4169, pruned_loss=0.1653, over 5685571.94 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3575, pruned_loss=0.1174, over 5708227.07 frames. ], batch size: 174, lr: 9.10e-03, grad_scale: 4.0
+2023-03-01 22:36:54,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2025, 1.3421, 1.1942, 1.4448], device='cuda:0'), covar=tensor([0.0836, 0.0346, 0.0375, 0.0954], device='cuda:0'), in_proj_covar=tensor([0.0219, 0.0143, 0.0146, 0.0263], device='cuda:0'), out_proj_covar=tensor([0.0044, 0.0033, 0.0029, 0.0050], device='cuda:0')
+2023-03-01 22:36:55,178 INFO [train.py:968] (0/2) Epoch 3, batch 40600, giga_loss[loss=0.3258, simple_loss=0.3891, pruned_loss=0.1312, over 29074.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3634, pruned_loss=0.1213, over 5715644.12 frames. ], libri_tot_loss[loss=0.3741, simple_loss=0.4173, pruned_loss=0.1655, over 5688549.46 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3587, pruned_loss=0.1174, over 5712197.41 frames. ], batch size: 136, lr: 9.10e-03, grad_scale: 4.0
+2023-03-01 22:37:10,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5002, 1.0217, 2.7945, 2.5301], device='cuda:0'), covar=tensor([0.1587, 0.1980, 0.0496, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0533, 0.0503, 0.0698, 0.0562], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 22:37:12,269 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:13,945 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:14,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:15,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:17,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6237, 1.8483, 1.6606, 1.6421], device='cuda:0'), covar=tensor([0.1063, 0.1530, 0.1040, 0.0913], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0773, 0.0751, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 22:37:19,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.191e+02 1.147e+03 1.487e+03 2.363e+03 1.073e+04, threshold=2.973e+03, percent-clipped=8.0
+2023-03-01 22:37:34,823 INFO [train.py:968] (0/2) Epoch 3, batch 40650, giga_loss[loss=0.2884, simple_loss=0.3613, pruned_loss=0.1078, over 28944.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3686, pruned_loss=0.1242, over 5714956.65 frames. ], libri_tot_loss[loss=0.3733, simple_loss=0.4165, pruned_loss=0.165, over 5694083.15 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3639, pruned_loss=0.1201, over 5708074.05 frames. ], batch size: 174, lr: 9.10e-03, grad_scale: 4.0
+2023-03-01 22:37:35,004 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:37,358 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:38,979 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:49,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:38:12,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1784, 1.3917, 1.0493, 0.7410], device='cuda:0'), covar=tensor([0.0790, 0.0594, 0.0542, 0.0730], device='cuda:0'), in_proj_covar=tensor([0.1237, 0.0951, 0.1001, 0.1076], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 22:38:14,861 INFO [train.py:968] (0/2) Epoch 3, batch 40700, giga_loss[loss=0.3266, simple_loss=0.3883, pruned_loss=0.1325, over 29018.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3726, pruned_loss=0.1264, over 5713967.97 frames. ], libri_tot_loss[loss=0.3732, simple_loss=0.4164, pruned_loss=0.165, over 5691137.39 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3677, pruned_loss=0.1221, over 5711164.08 frames. ], batch size: 128, lr: 9.10e-03, grad_scale: 4.0
+2023-03-01 22:38:40,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.210e+02 1.145e+03 1.449e+03 1.952e+03 5.156e+03, threshold=2.897e+03, percent-clipped=7.0
+2023-03-01 22:38:57,643 INFO [train.py:968] (0/2) Epoch 3, batch 40750, giga_loss[loss=0.279, simple_loss=0.357, pruned_loss=0.1005, over 28946.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3768, pruned_loss=0.1284, over 5701278.05 frames. ], libri_tot_loss[loss=0.3734, simple_loss=0.4165, pruned_loss=0.1652, over 5690015.57 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3723, pruned_loss=0.1243, over 5700477.45 frames. ], batch size: 145, lr: 9.10e-03, grad_scale: 4.0
+2023-03-01 22:39:40,337 INFO [train.py:968] (0/2) Epoch 3, batch 40800, giga_loss[loss=0.3211, simple_loss=0.3688, pruned_loss=0.1367, over 23938.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3801, pruned_loss=0.1303, over 5706995.21 frames. ], libri_tot_loss[loss=0.3735, simple_loss=0.4164, pruned_loss=0.1652, over 5694306.32 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3759, pruned_loss=0.1263, over 5702744.58 frames. ], batch size: 705, lr: 9.10e-03, grad_scale: 8.0
+2023-03-01 22:40:05,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.137e+02 1.271e+03 1.624e+03 2.265e+03 6.269e+03, threshold=3.248e+03, percent-clipped=14.0
+2023-03-01 22:40:12,777 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-01 22:40:26,023 INFO [train.py:968] (0/2) Epoch 3, batch 40850, giga_loss[loss=0.3648, simple_loss=0.4032, pruned_loss=0.1632, over 28522.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3849, pruned_loss=0.1348, over 5703337.67 frames. ], libri_tot_loss[loss=0.3736, simple_loss=0.4165, pruned_loss=0.1654, over 5693560.72 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3811, pruned_loss=0.1312, over 5700597.72 frames. ], batch size: 71, lr: 9.09e-03, grad_scale: 8.0
+2023-03-01 22:40:40,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:41:18,790 INFO [train.py:968] (0/2) Epoch 3, batch 40900, giga_loss[loss=0.3999, simple_loss=0.4431, pruned_loss=0.1783, over 28578.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3929, pruned_loss=0.1421, over 5698255.13 frames. ], libri_tot_loss[loss=0.3738, simple_loss=0.4167, pruned_loss=0.1655, over 5692071.89 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3894, pruned_loss=0.1388, over 5697734.13 frames. ], batch size: 336, lr: 9.09e-03, grad_scale: 4.0
+2023-03-01 22:41:27,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:41:33,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:41:47,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2961, 1.8298, 1.2934, 0.4977], device='cuda:0'), covar=tensor([0.1495, 0.0823, 0.1369, 0.1812], device='cuda:0'), in_proj_covar=tensor([0.1300, 0.1229, 0.1311, 0.1093], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 22:41:47,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.924e+02 1.564e+03 2.090e+03 2.814e+03 7.872e+03, threshold=4.181e+03, percent-clipped=18.0
+2023-03-01 22:42:05,249 INFO [train.py:968] (0/2) Epoch 3, batch 40950, giga_loss[loss=0.3406, simple_loss=0.4013, pruned_loss=0.1399, over 28920.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4009, pruned_loss=0.1482, over 5695057.97 frames. ], libri_tot_loss[loss=0.3736, simple_loss=0.4165, pruned_loss=0.1653, over 5695243.60 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.398, pruned_loss=0.1456, over 5691698.31 frames. ], batch size: 164, lr: 9.09e-03, grad_scale: 4.0
+2023-03-01 22:42:13,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9860, 1.3202, 4.0431, 3.1665], device='cuda:0'), covar=tensor([0.1537, 0.1819, 0.0321, 0.0500], device='cuda:0'), in_proj_covar=tensor([0.0537, 0.0502, 0.0702, 0.0561], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 22:42:21,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7012, 1.9002, 1.7570, 1.7289], device='cuda:0'), covar=tensor([0.1168, 0.1481, 0.0962, 0.0878], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0764, 0.0744, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 22:42:50,656 INFO [train.py:968] (0/2) Epoch 3, batch 41000, giga_loss[loss=0.3899, simple_loss=0.4293, pruned_loss=0.1752, over 28903.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4066, pruned_loss=0.1533, over 5697934.78 frames. ], libri_tot_loss[loss=0.374, simple_loss=0.4168, pruned_loss=0.1656, over 5697045.73 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.404, pruned_loss=0.1507, over 5693622.01 frames. ], batch size: 186, lr: 9.09e-03, grad_scale: 4.0
+2023-03-01 22:42:53,795 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:42:55,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:12,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:16,340 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.443e+02 1.683e+03 2.231e+03 2.844e+03 7.306e+03, threshold=4.461e+03, percent-clipped=8.0
+2023-03-01 22:43:20,465 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:24,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:26,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1784, 1.1744, 1.0308, 1.4382], device='cuda:0'), covar=tensor([0.0847, 0.0386, 0.0383, 0.0913], device='cuda:0'), in_proj_covar=tensor([0.0220, 0.0143, 0.0146, 0.0263], device='cuda:0'), out_proj_covar=tensor([0.0045, 0.0033, 0.0029, 0.0050], device='cuda:0')
+2023-03-01 22:43:32,604 INFO [train.py:968] (0/2) Epoch 3, batch 41050, giga_loss[loss=0.3458, simple_loss=0.3984, pruned_loss=0.1466, over 28403.00 frames. ], tot_loss[loss=0.3656, simple_loss=0.4129, pruned_loss=0.1592, over 5702068.32 frames. ], libri_tot_loss[loss=0.3736, simple_loss=0.4166, pruned_loss=0.1654, over 5704762.75 frames. ], giga_tot_loss[loss=0.3623, simple_loss=0.4107, pruned_loss=0.1569, over 5691782.57 frames. ], batch size: 71, lr: 9.09e-03, grad_scale: 4.0
+2023-03-01 22:43:34,606 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:38,692 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:40,583 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:43,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:47,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2142, 1.4605, 1.1992, 1.4096], device='cuda:0'), covar=tensor([0.0864, 0.0352, 0.0374, 0.0956], device='cuda:0'), in_proj_covar=tensor([0.0219, 0.0143, 0.0145, 0.0262], device='cuda:0'), out_proj_covar=tensor([0.0044, 0.0033, 0.0029, 0.0050], device='cuda:0')
+2023-03-01 22:44:03,554 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:44:08,713 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:44:18,199 INFO [train.py:968] (0/2) Epoch 3, batch 41100, giga_loss[loss=0.5186, simple_loss=0.5005, pruned_loss=0.2683, over 26536.00 frames. ], tot_loss[loss=0.3727, simple_loss=0.4176, pruned_loss=0.1639, over 5676040.06 frames. ], libri_tot_loss[loss=0.3729, simple_loss=0.4159, pruned_loss=0.165, over 5698829.41 frames. ], giga_tot_loss[loss=0.3706, simple_loss=0.4164, pruned_loss=0.1624, over 5672254.78 frames. ], batch size: 555, lr: 9.09e-03, grad_scale: 4.0
+2023-03-01 22:44:52,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.109e+03 1.716e+03 2.416e+03 3.056e+03 7.869e+03, threshold=4.832e+03, percent-clipped=6.0
+2023-03-01 22:45:09,565 INFO [train.py:968] (0/2) Epoch 3, batch 41150, giga_loss[loss=0.381, simple_loss=0.4205, pruned_loss=0.1707, over 28912.00 frames. ], tot_loss[loss=0.3772, simple_loss=0.4202, pruned_loss=0.1671, over 5658729.62 frames. ], libri_tot_loss[loss=0.3734, simple_loss=0.4163, pruned_loss=0.1652, over 5690647.53 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.4189, pruned_loss=0.1657, over 5662765.56 frames. ], batch size: 213, lr: 9.08e-03, grad_scale: 4.0
+2023-03-01 22:45:25,875 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:45:30,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:45:43,872 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:45:46,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:45:58,096 INFO [train.py:968] (0/2) Epoch 3, batch 41200, giga_loss[loss=0.371, simple_loss=0.4215, pruned_loss=0.1603, over 28725.00 frames. ], tot_loss[loss=0.3819, simple_loss=0.4227, pruned_loss=0.1706, over 5649995.94 frames. ], libri_tot_loss[loss=0.3728, simple_loss=0.4159, pruned_loss=0.1649, over 5681400.54 frames. ], giga_tot_loss[loss=0.381, simple_loss=0.4223, pruned_loss=0.1698, over 5661113.14 frames. ], batch size: 112, lr: 9.08e-03, grad_scale: 8.0
+2023-03-01 22:45:59,149 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:46:16,332 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:46:31,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.268e+02 1.752e+03 2.159e+03 2.992e+03 6.289e+03, threshold=4.317e+03, percent-clipped=6.0
+2023-03-01 22:46:41,363 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3371, 1.6007, 1.1496, 0.7757], device='cuda:0'), covar=tensor([0.0812, 0.0587, 0.0514, 0.0676], device='cuda:0'), in_proj_covar=tensor([0.1274, 0.0973, 0.1009, 0.1096], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 22:46:46,574 INFO [train.py:968] (0/2) Epoch 3, batch 41250, giga_loss[loss=0.3332, simple_loss=0.3899, pruned_loss=0.1383, over 28879.00 frames. ], tot_loss[loss=0.3867, simple_loss=0.4255, pruned_loss=0.1739, over 5610548.65 frames. ], libri_tot_loss[loss=0.3731, simple_loss=0.4161, pruned_loss=0.1651, over 5660254.87 frames. ], giga_tot_loss[loss=0.3862, simple_loss=0.4253, pruned_loss=0.1735, over 5638459.57 frames. ], batch size: 119, lr: 9.08e-03, grad_scale: 4.0
+2023-03-01 22:47:11,873 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:47:39,914 INFO [train.py:968] (0/2) Epoch 3, batch 41300, giga_loss[loss=0.3491, simple_loss=0.3997, pruned_loss=0.1493, over 28810.00 frames. ], tot_loss[loss=0.3918, simple_loss=0.4291, pruned_loss=0.1773, over 5610414.84 frames. ], libri_tot_loss[loss=0.3728, simple_loss=0.4158, pruned_loss=0.1649, over 5664712.90 frames. ], giga_tot_loss[loss=0.392, simple_loss=0.4293, pruned_loss=0.1773, over 5627834.79 frames. ], batch size: 284, lr: 9.08e-03, grad_scale: 4.0
+2023-03-01 22:48:11,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.671e+03 2.325e+03 3.268e+03 8.056e+03, threshold=4.649e+03, percent-clipped=10.0
+2023-03-01 22:48:33,765 INFO [train.py:968] (0/2) Epoch 3, batch 41350, giga_loss[loss=0.4196, simple_loss=0.4386, pruned_loss=0.2003, over 27610.00 frames. ], tot_loss[loss=0.3942, simple_loss=0.4304, pruned_loss=0.179, over 5600218.41 frames. ], libri_tot_loss[loss=0.373, simple_loss=0.4159, pruned_loss=0.165, over 5655402.52 frames. ], giga_tot_loss[loss=0.3944, simple_loss=0.4307, pruned_loss=0.179, over 5621126.47 frames. ], batch size: 472, lr: 9.08e-03, grad_scale: 4.0
+2023-03-01 22:49:24,625 INFO [train.py:968] (0/2) Epoch 3, batch 41400, giga_loss[loss=0.529, simple_loss=0.5108, pruned_loss=0.2736, over 26714.00 frames. ], tot_loss[loss=0.3941, simple_loss=0.4297, pruned_loss=0.1792, over 5611246.66 frames. ], libri_tot_loss[loss=0.373, simple_loss=0.416, pruned_loss=0.165, over 5656263.59 frames. ], giga_tot_loss[loss=0.3944, simple_loss=0.4301, pruned_loss=0.1793, over 5626311.51 frames. ], batch size: 555, lr: 9.07e-03, grad_scale: 4.0
+2023-03-01 22:49:55,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.024e+02 1.776e+03 2.526e+03 3.306e+03 1.099e+04, threshold=5.052e+03, percent-clipped=15.0
+2023-03-01 22:50:13,490 INFO [train.py:968] (0/2) Epoch 3, batch 41450, giga_loss[loss=0.4219, simple_loss=0.4504, pruned_loss=0.1967, over 27566.00 frames. ], tot_loss[loss=0.3908, simple_loss=0.4281, pruned_loss=0.1767, over 5632111.34 frames. ], libri_tot_loss[loss=0.3732, simple_loss=0.4162, pruned_loss=0.1651, over 5659847.57 frames. ], giga_tot_loss[loss=0.3914, simple_loss=0.4286, pruned_loss=0.1771, over 5639953.67 frames. ], batch size: 472, lr: 9.07e-03, grad_scale: 4.0
+2023-03-01 22:50:54,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1310, 1.7730, 1.5445, 1.6243], device='cuda:0'), covar=tensor([0.0575, 0.0690, 0.0877, 0.0997], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0478, 0.0513, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-01 22:51:02,972 INFO [train.py:968] (0/2) Epoch 3, batch 41500, giga_loss[loss=0.3876, simple_loss=0.4249, pruned_loss=0.1751, over 28274.00 frames. ], tot_loss[loss=0.3872, simple_loss=0.4269, pruned_loss=0.1737, over 5638301.02 frames. ], libri_tot_loss[loss=0.3728, simple_loss=0.4159, pruned_loss=0.1649, over 5659775.72 frames. ], giga_tot_loss[loss=0.3882, simple_loss=0.4276, pruned_loss=0.1744, over 5644344.10 frames. ], batch size: 77, lr: 9.07e-03, grad_scale: 4.0
+2023-03-01 22:51:17,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:51:26,346 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:51:38,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.084e+02 1.571e+03 1.900e+03 2.466e+03 7.128e+03, threshold=3.800e+03, percent-clipped=2.0
+2023-03-01 22:51:46,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3221, 1.6000, 1.3697, 1.4217], device='cuda:0'), covar=tensor([0.0807, 0.0354, 0.0351, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0220, 0.0142, 0.0145, 0.0262], device='cuda:0'), out_proj_covar=tensor([0.0045, 0.0033, 0.0029, 0.0050], device='cuda:0')
+2023-03-01 22:52:01,375 INFO [train.py:968] (0/2) Epoch 3, batch 41550, giga_loss[loss=0.3598, simple_loss=0.4181, pruned_loss=0.1507, over 28576.00 frames. ], tot_loss[loss=0.3879, simple_loss=0.4283, pruned_loss=0.1737, over 5652429.49 frames. ], libri_tot_loss[loss=0.3728, simple_loss=0.4159, pruned_loss=0.1649, over 5659775.72 frames. ], giga_tot_loss[loss=0.3887, simple_loss=0.4289, pruned_loss=0.1742, over 5657132.89 frames. ], batch size: 307, lr: 9.07e-03, grad_scale: 4.0
+2023-03-01 22:52:49,300 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:52:53,290 INFO [train.py:968] (0/2) Epoch 3, batch 41600, giga_loss[loss=0.3945, simple_loss=0.4413, pruned_loss=0.1739, over 28824.00 frames. ], tot_loss[loss=0.3869, simple_loss=0.4273, pruned_loss=0.1732, over 5643637.08 frames. ], libri_tot_loss[loss=0.3727, simple_loss=0.4158, pruned_loss=0.1648, over 5664655.06 frames. ], giga_tot_loss[loss=0.388, simple_loss=0.4281, pruned_loss=0.1739, over 5642714.54 frames. ], batch size: 119, lr: 9.07e-03, grad_scale: 8.0
+2023-03-01 22:53:24,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.259e+02 1.692e+03 2.224e+03 3.021e+03 8.404e+03, threshold=4.448e+03, percent-clipped=14.0
+2023-03-01 22:53:42,486 INFO [train.py:968] (0/2) Epoch 3, batch 41650, giga_loss[loss=0.3572, simple_loss=0.4146, pruned_loss=0.1499, over 28804.00 frames. ], tot_loss[loss=0.3809, simple_loss=0.4237, pruned_loss=0.169, over 5645262.78 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4152, pruned_loss=0.1644, over 5669190.13 frames. ], giga_tot_loss[loss=0.3826, simple_loss=0.425, pruned_loss=0.1701, over 5640226.48 frames. ], batch size: 284, lr: 9.07e-03, grad_scale: 4.0
+2023-03-01 22:53:43,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:54:30,883 INFO [train.py:968] (0/2) Epoch 3, batch 41700, giga_loss[loss=0.3732, simple_loss=0.4204, pruned_loss=0.163, over 28583.00 frames. ], tot_loss[loss=0.376, simple_loss=0.4209, pruned_loss=0.1655, over 5651932.92 frames. ], libri_tot_loss[loss=0.3719, simple_loss=0.4152, pruned_loss=0.1643, over 5663597.10 frames. ], giga_tot_loss[loss=0.3775, simple_loss=0.4221, pruned_loss=0.1665, over 5652137.18 frames. ], batch size: 307, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:54:40,449 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-132000.pt
+2023-03-01 22:55:04,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.014e+02 1.659e+03 2.045e+03 2.809e+03 6.499e+03, threshold=4.090e+03, percent-clipped=9.0
+2023-03-01 22:55:21,905 INFO [train.py:968] (0/2) Epoch 3, batch 41750, giga_loss[loss=0.3573, simple_loss=0.4094, pruned_loss=0.1526, over 28906.00 frames. ], tot_loss[loss=0.3718, simple_loss=0.4177, pruned_loss=0.1629, over 5651644.10 frames. ], libri_tot_loss[loss=0.3713, simple_loss=0.4148, pruned_loss=0.1639, over 5666517.98 frames. ], giga_tot_loss[loss=0.3735, simple_loss=0.4191, pruned_loss=0.164, over 5648822.88 frames. ], batch size: 186, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:55:22,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6477, 3.9309, 1.6744, 1.5838], device='cuda:0'), covar=tensor([0.0819, 0.0345, 0.0815, 0.1275], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0453, 0.0308, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:0')
+2023-03-01 22:55:57,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9079, 1.8752, 1.4045, 1.1875], device='cuda:0'), covar=tensor([0.0695, 0.0543, 0.0493, 0.0632], device='cuda:0'), in_proj_covar=tensor([0.1265, 0.0971, 0.1000, 0.1087], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 22:55:59,473 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:56:02,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:56:05,743 INFO [train.py:968] (0/2) Epoch 3, batch 41800, giga_loss[loss=0.3641, simple_loss=0.392, pruned_loss=0.1681, over 23752.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.4148, pruned_loss=0.161, over 5653050.30 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4147, pruned_loss=0.1639, over 5674868.77 frames. ], giga_tot_loss[loss=0.3699, simple_loss=0.4161, pruned_loss=0.1618, over 5642324.81 frames. ], batch size: 705, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:56:32,408 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:56:38,523 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.189e+02 1.697e+03 2.231e+03 2.941e+03 7.080e+03, threshold=4.461e+03, percent-clipped=10.0
+2023-03-01 22:56:48,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7382, 5.4553, 5.3202, 2.4851], device='cuda:0'), covar=tensor([0.0409, 0.0463, 0.1015, 0.1797], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0644, 0.0818, 0.0585], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 22:56:53,812 INFO [train.py:968] (0/2) Epoch 3, batch 41850, giga_loss[loss=0.3168, simple_loss=0.3788, pruned_loss=0.1274, over 28945.00 frames. ], tot_loss[loss=0.3672, simple_loss=0.4141, pruned_loss=0.1602, over 5662595.38 frames. ], libri_tot_loss[loss=0.3716, simple_loss=0.415, pruned_loss=0.1641, over 5680541.50 frames. ], giga_tot_loss[loss=0.368, simple_loss=0.4149, pruned_loss=0.1606, over 5648173.36 frames. ], batch size: 136, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:57:26,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:57:34,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:57:40,281 INFO [train.py:968] (0/2) Epoch 3, batch 41900, giga_loss[loss=0.3411, simple_loss=0.4017, pruned_loss=0.1402, over 29022.00 frames. ], tot_loss[loss=0.3652, simple_loss=0.413, pruned_loss=0.1587, over 5671063.26 frames. ], libri_tot_loss[loss=0.371, simple_loss=0.4144, pruned_loss=0.1638, over 5682631.67 frames. ], giga_tot_loss[loss=0.3663, simple_loss=0.4141, pruned_loss=0.1592, over 5657660.44 frames. ], batch size: 155, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:58:10,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.222e+02 1.644e+03 2.340e+03 3.247e+03 6.596e+03, threshold=4.681e+03, percent-clipped=6.0
+2023-03-01 22:58:30,766 INFO [train.py:968] (0/2) Epoch 3, batch 41950, giga_loss[loss=0.3301, simple_loss=0.3864, pruned_loss=0.1369, over 28874.00 frames. ], tot_loss[loss=0.3607, simple_loss=0.4098, pruned_loss=0.1558, over 5682901.22 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4135, pruned_loss=0.1631, over 5692071.91 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.4114, pruned_loss=0.1565, over 5662976.18 frames. ], batch size: 227, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:58:38,121 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:58:44,315 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-01 22:58:48,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:59:20,299 INFO [train.py:968] (0/2) Epoch 3, batch 42000, giga_loss[loss=0.3458, simple_loss=0.4163, pruned_loss=0.1377, over 28833.00 frames. ], tot_loss[loss=0.3582, simple_loss=0.4098, pruned_loss=0.1533, over 5689674.00 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4129, pruned_loss=0.1626, over 5698423.46 frames. ], giga_tot_loss[loss=0.3599, simple_loss=0.4116, pruned_loss=0.1541, over 5667856.08 frames. ], batch size: 284, lr: 9.05e-03, grad_scale: 8.0
+2023-03-01 22:59:20,305 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 22:59:30,613 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2463, simple_loss=0.3444, pruned_loss=0.07409, over 944034.00 frames. 
+2023-03-01 22:59:30,614 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 23:00:02,194 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:00:04,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:00:04,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.979e+02 1.581e+03 1.947e+03 2.610e+03 6.979e+03, threshold=3.894e+03, percent-clipped=4.0
+2023-03-01 23:00:06,670 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:00:08,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:00:19,555 INFO [train.py:968] (0/2) Epoch 3, batch 42050, giga_loss[loss=0.3698, simple_loss=0.4176, pruned_loss=0.1609, over 28588.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.4113, pruned_loss=0.1528, over 5681743.59 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4129, pruned_loss=0.1626, over 5692975.60 frames. ], giga_tot_loss[loss=0.3596, simple_loss=0.4127, pruned_loss=0.1533, over 5668220.81 frames. ], batch size: 78, lr: 9.05e-03, grad_scale: 8.0
+2023-03-01 23:00:31,902 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:00:37,318 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:01:06,846 INFO [train.py:968] (0/2) Epoch 3, batch 42100, libri_loss[loss=0.3925, simple_loss=0.4364, pruned_loss=0.1743, over 25553.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.4127, pruned_loss=0.1551, over 5665517.39 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4132, pruned_loss=0.1627, over 5683579.64 frames. ], giga_tot_loss[loss=0.3619, simple_loss=0.4136, pruned_loss=0.1551, over 5663767.95 frames. ], batch size: 136, lr: 9.05e-03, grad_scale: 8.0
+2023-03-01 23:01:19,252 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:01:22,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:01:31,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7374, 1.8149, 1.7937, 1.6986], device='cuda:0'), covar=tensor([0.1213, 0.1721, 0.1048, 0.0964], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0787, 0.0749, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 23:01:35,666 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-01 23:01:38,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.739e+03 2.362e+03 3.110e+03 5.168e+03, threshold=4.724e+03, percent-clipped=10.0
+2023-03-01 23:01:38,876 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:01:51,342 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:01:56,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-01 23:01:56,444 INFO [train.py:968] (0/2) Epoch 3, batch 42150, giga_loss[loss=0.4338, simple_loss=0.4376, pruned_loss=0.215, over 23499.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.4129, pruned_loss=0.1558, over 5669594.39 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.4129, pruned_loss=0.1627, over 5685937.71 frames. ], giga_tot_loss[loss=0.3627, simple_loss=0.4138, pruned_loss=0.1558, over 5665917.73 frames. ], batch size: 705, lr: 9.05e-03, grad_scale: 4.0
+2023-03-01 23:02:01,695 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:02:07,654 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4869, 1.5138, 1.1976, 0.9346], device='cuda:0'), covar=tensor([0.0895, 0.0695, 0.0557, 0.0766], device='cuda:0'), in_proj_covar=tensor([0.1268, 0.0969, 0.0998, 0.1086], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 23:02:36,361 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 23:02:40,470 INFO [train.py:968] (0/2) Epoch 3, batch 42200, giga_loss[loss=0.3873, simple_loss=0.4324, pruned_loss=0.1711, over 29059.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4107, pruned_loss=0.1551, over 5666606.87 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.413, pruned_loss=0.1627, over 5680328.78 frames. ], giga_tot_loss[loss=0.3606, simple_loss=0.4113, pruned_loss=0.155, over 5668080.87 frames. ], batch size: 128, lr: 9.05e-03, grad_scale: 4.0
+2023-03-01 23:03:12,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.754e+03 2.202e+03 3.115e+03 7.970e+03, threshold=4.403e+03, percent-clipped=6.0
+2023-03-01 23:03:27,071 INFO [train.py:968] (0/2) Epoch 3, batch 42250, giga_loss[loss=0.369, simple_loss=0.4156, pruned_loss=0.1612, over 27945.00 frames. ], tot_loss[loss=0.3612, simple_loss=0.4101, pruned_loss=0.1561, over 5655067.25 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.413, pruned_loss=0.1627, over 5673610.64 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.4105, pruned_loss=0.1558, over 5661366.25 frames. ], batch size: 412, lr: 9.05e-03, grad_scale: 2.0
+2023-03-01 23:04:16,096 INFO [train.py:968] (0/2) Epoch 3, batch 42300, giga_loss[loss=0.3251, simple_loss=0.3955, pruned_loss=0.1274, over 28508.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.4094, pruned_loss=0.1552, over 5660354.94 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4132, pruned_loss=0.1628, over 5676921.52 frames. ], giga_tot_loss[loss=0.3594, simple_loss=0.4095, pruned_loss=0.1546, over 5662049.42 frames. ], batch size: 71, lr: 9.04e-03, grad_scale: 2.0
+2023-03-01 23:04:47,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:04:47,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9494, 1.8916, 1.3542, 1.0851], device='cuda:0'), covar=tensor([0.0668, 0.0568, 0.0522, 0.0704], device='cuda:0'), in_proj_covar=tensor([0.1269, 0.0974, 0.1014, 0.1099], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 23:04:48,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.201e+02 1.560e+03 2.080e+03 2.815e+03 6.325e+03, threshold=4.159e+03, percent-clipped=6.0
+2023-03-01 23:04:59,863 INFO [train.py:968] (0/2) Epoch 3, batch 42350, giga_loss[loss=0.3746, simple_loss=0.418, pruned_loss=0.1656, over 27575.00 frames. ], tot_loss[loss=0.3573, simple_loss=0.4085, pruned_loss=0.153, over 5681593.10 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4131, pruned_loss=0.1627, over 5683380.61 frames. ], giga_tot_loss[loss=0.3566, simple_loss=0.4085, pruned_loss=0.1523, over 5676524.34 frames. ], batch size: 472, lr: 9.04e-03, grad_scale: 2.0
+2023-03-01 23:05:47,093 INFO [train.py:968] (0/2) Epoch 3, batch 42400, giga_loss[loss=0.3589, simple_loss=0.4093, pruned_loss=0.1543, over 28300.00 frames. ], tot_loss[loss=0.356, simple_loss=0.4079, pruned_loss=0.152, over 5684657.53 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4126, pruned_loss=0.1624, over 5688921.50 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4082, pruned_loss=0.1515, over 5675500.59 frames. ], batch size: 368, lr: 9.04e-03, grad_scale: 4.0
+2023-03-01 23:06:18,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.892e+02 1.623e+03 2.217e+03 2.930e+03 8.100e+03, threshold=4.433e+03, percent-clipped=9.0
+2023-03-01 23:06:32,540 INFO [train.py:968] (0/2) Epoch 3, batch 42450, giga_loss[loss=0.3309, simple_loss=0.3789, pruned_loss=0.1414, over 28509.00 frames. ], tot_loss[loss=0.3551, simple_loss=0.4069, pruned_loss=0.1517, over 5680982.40 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4126, pruned_loss=0.1624, over 5684414.48 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.407, pruned_loss=0.1509, over 5676894.62 frames. ], batch size: 85, lr: 9.04e-03, grad_scale: 4.0
+2023-03-01 23:06:42,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0755, 1.1494, 0.8358, 0.6628], device='cuda:0'), covar=tensor([0.0603, 0.0536, 0.0477, 0.0581], device='cuda:0'), in_proj_covar=tensor([0.1273, 0.0989, 0.1021, 0.1095], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 23:06:46,896 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-01 23:06:53,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:06:57,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:07:02,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1956, 3.3053, 2.2840, 1.0830], device='cuda:0'), covar=tensor([0.2187, 0.0873, 0.1363, 0.2175], device='cuda:0'), in_proj_covar=tensor([0.1290, 0.1242, 0.1302, 0.1111], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-01 23:07:16,319 INFO [train.py:968] (0/2) Epoch 3, batch 42500, giga_loss[loss=0.3346, simple_loss=0.4012, pruned_loss=0.134, over 28842.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4068, pruned_loss=0.1525, over 5667938.44 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4131, pruned_loss=0.1628, over 5672043.13 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4064, pruned_loss=0.1513, over 5674853.52 frames. ], batch size: 174, lr: 9.04e-03, grad_scale: 4.0
+2023-03-01 23:07:23,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:07:24,805 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:07:44,474 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:07:47,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.404e+02 1.692e+03 2.044e+03 3.033e+03 1.121e+04, threshold=4.089e+03, percent-clipped=9.0
+2023-03-01 23:08:01,342 INFO [train.py:968] (0/2) Epoch 3, batch 42550, giga_loss[loss=0.3408, simple_loss=0.3954, pruned_loss=0.1431, over 28722.00 frames. ], tot_loss[loss=0.357, simple_loss=0.407, pruned_loss=0.1535, over 5669498.19 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4137, pruned_loss=0.1634, over 5674659.86 frames. ], giga_tot_loss[loss=0.3549, simple_loss=0.406, pruned_loss=0.1519, over 5672659.90 frames. ], batch size: 284, lr: 9.04e-03, grad_scale: 4.0
+2023-03-01 23:08:49,810 INFO [train.py:968] (0/2) Epoch 3, batch 42600, giga_loss[loss=0.4142, simple_loss=0.4511, pruned_loss=0.1887, over 28629.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4053, pruned_loss=0.1532, over 5667275.86 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4129, pruned_loss=0.1628, over 5677343.79 frames. ], giga_tot_loss[loss=0.3548, simple_loss=0.4051, pruned_loss=0.1523, over 5667082.72 frames. ], batch size: 336, lr: 9.03e-03, grad_scale: 4.0
+2023-03-01 23:09:03,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-01 23:09:23,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.608e+03 2.092e+03 2.775e+03 7.719e+03, threshold=4.184e+03, percent-clipped=9.0
+2023-03-01 23:09:39,358 INFO [train.py:968] (0/2) Epoch 3, batch 42650, giga_loss[loss=0.4085, simple_loss=0.4353, pruned_loss=0.1909, over 26674.00 frames. ], tot_loss[loss=0.354, simple_loss=0.4034, pruned_loss=0.1522, over 5674740.51 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4128, pruned_loss=0.1626, over 5682685.36 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4031, pruned_loss=0.1515, over 5669630.30 frames. ], batch size: 555, lr: 9.03e-03, grad_scale: 4.0
+2023-03-01 23:09:40,278 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:09:44,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:10:03,483 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:10:06,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:10:10,942 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:10:26,384 INFO [train.py:968] (0/2) Epoch 3, batch 42700, giga_loss[loss=0.4295, simple_loss=0.4448, pruned_loss=0.2071, over 26579.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4028, pruned_loss=0.1519, over 5669174.17 frames. ], libri_tot_loss[loss=0.3685, simple_loss=0.4126, pruned_loss=0.1623, over 5674057.76 frames. ], giga_tot_loss[loss=0.3524, simple_loss=0.4025, pruned_loss=0.1512, over 5673700.79 frames. ], batch size: 555, lr: 9.03e-03, grad_scale: 4.0
+2023-03-01 23:10:34,882 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:10:54,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3871, 1.6131, 1.3408, 1.6138], device='cuda:0'), covar=tensor([0.0797, 0.0313, 0.0371, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0219, 0.0142, 0.0147, 0.0261], device='cuda:0'), out_proj_covar=tensor([0.0045, 0.0034, 0.0030, 0.0050], device='cuda:0')
+2023-03-01 23:10:54,939 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-01 23:11:00,225 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.963e+02 1.649e+03 2.196e+03 3.110e+03 6.233e+03, threshold=4.392e+03, percent-clipped=6.0
+2023-03-01 23:11:14,635 INFO [train.py:968] (0/2) Epoch 3, batch 42750, giga_loss[loss=0.3496, simple_loss=0.4061, pruned_loss=0.1466, over 28785.00 frames. ], tot_loss[loss=0.353, simple_loss=0.4027, pruned_loss=0.1516, over 5678128.81 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4119, pruned_loss=0.1615, over 5678627.88 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4028, pruned_loss=0.1514, over 5677935.41 frames. ], batch size: 186, lr: 9.03e-03, grad_scale: 4.0
+2023-03-01 23:11:20,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2671, 4.9764, 4.9341, 2.2222], device='cuda:0'), covar=tensor([0.0359, 0.0292, 0.0683, 0.1752], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0648, 0.0826, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 23:11:26,644 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:11:26,844 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 23:11:33,800 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-01 23:11:47,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4883, 1.3893, 1.1454, 1.1292], device='cuda:0'), covar=tensor([0.0489, 0.0377, 0.0786, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0484, 0.0517, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 23:12:04,184 INFO [train.py:968] (0/2) Epoch 3, batch 42800, giga_loss[loss=0.3381, simple_loss=0.3974, pruned_loss=0.1395, over 28553.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.4032, pruned_loss=0.1511, over 5676654.44 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4116, pruned_loss=0.1613, over 5679712.74 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4035, pruned_loss=0.151, over 5675437.28 frames. ], batch size: 85, lr: 9.03e-03, grad_scale: 8.0
+2023-03-01 23:12:36,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.815e+02 1.647e+03 2.290e+03 3.017e+03 7.223e+03, threshold=4.580e+03, percent-clipped=6.0
+2023-03-01 23:12:47,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:12:48,957 INFO [train.py:968] (0/2) Epoch 3, batch 42850, giga_loss[loss=0.4846, simple_loss=0.4799, pruned_loss=0.2446, over 26608.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.4035, pruned_loss=0.1502, over 5677218.95 frames. ], libri_tot_loss[loss=0.3672, simple_loss=0.4116, pruned_loss=0.1614, over 5674981.71 frames. ], giga_tot_loss[loss=0.3515, simple_loss=0.4035, pruned_loss=0.1498, over 5680067.51 frames. ], batch size: 555, lr: 9.03e-03, grad_scale: 4.0
+2023-03-01 23:12:56,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3276, 2.7652, 1.2970, 1.4645], device='cuda:0'), covar=tensor([0.0846, 0.0415, 0.0918, 0.1262], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0462, 0.0308, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:0')
+2023-03-01 23:13:32,579 INFO [train.py:968] (0/2) Epoch 3, batch 42900, giga_loss[loss=0.3247, simple_loss=0.3845, pruned_loss=0.1324, over 28930.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.404, pruned_loss=0.1501, over 5668493.47 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.412, pruned_loss=0.1617, over 5671523.14 frames. ], giga_tot_loss[loss=0.3509, simple_loss=0.4034, pruned_loss=0.1492, over 5674791.49 frames. ], batch size: 136, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:14:11,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.561e+02 1.404e+03 1.824e+03 2.372e+03 4.872e+03, threshold=3.648e+03, percent-clipped=1.0
+2023-03-01 23:14:26,430 INFO [train.py:968] (0/2) Epoch 3, batch 42950, giga_loss[loss=0.3656, simple_loss=0.4135, pruned_loss=0.1589, over 28835.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.405, pruned_loss=0.1516, over 5653407.83 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.4121, pruned_loss=0.1617, over 5664355.86 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.4042, pruned_loss=0.1507, over 5665570.92 frames. ], batch size: 199, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:15:15,519 INFO [train.py:968] (0/2) Epoch 3, batch 43000, giga_loss[loss=0.5179, simple_loss=0.5135, pruned_loss=0.2611, over 26584.00 frames. ], tot_loss[loss=0.3604, simple_loss=0.4088, pruned_loss=0.156, over 5637757.28 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4121, pruned_loss=0.1617, over 5649896.32 frames. ], giga_tot_loss[loss=0.3593, simple_loss=0.4082, pruned_loss=0.1552, over 5660593.52 frames. ], batch size: 555, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:15:54,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.905e+03 2.381e+03 3.202e+03 5.533e+03, threshold=4.762e+03, percent-clipped=11.0
+2023-03-01 23:16:08,920 INFO [train.py:968] (0/2) Epoch 3, batch 43050, giga_loss[loss=0.3249, simple_loss=0.3832, pruned_loss=0.1333, over 28903.00 frames. ], tot_loss[loss=0.3656, simple_loss=0.411, pruned_loss=0.1601, over 5637211.53 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4119, pruned_loss=0.1614, over 5653520.12 frames. ], giga_tot_loss[loss=0.365, simple_loss=0.4107, pruned_loss=0.1596, over 5651741.58 frames. ], batch size: 112, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:16:59,847 INFO [train.py:968] (0/2) Epoch 3, batch 43100, libri_loss[loss=0.3672, simple_loss=0.4143, pruned_loss=0.16, over 29514.00 frames. ], tot_loss[loss=0.3688, simple_loss=0.4127, pruned_loss=0.1625, over 5626261.53 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.412, pruned_loss=0.1615, over 5638573.40 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.4123, pruned_loss=0.162, over 5650909.25 frames. ], batch size: 82, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:17:31,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.666e+03 2.089e+03 2.720e+03 9.278e+03, threshold=4.177e+03, percent-clipped=10.0
+2023-03-01 23:17:33,014 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:17:43,916 INFO [train.py:968] (0/2) Epoch 3, batch 43150, giga_loss[loss=0.3628, simple_loss=0.4044, pruned_loss=0.1606, over 28765.00 frames. ], tot_loss[loss=0.3667, simple_loss=0.4114, pruned_loss=0.161, over 5639055.89 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4121, pruned_loss=0.1616, over 5634872.79 frames. ], giga_tot_loss[loss=0.3661, simple_loss=0.411, pruned_loss=0.1606, over 5661413.96 frames. ], batch size: 284, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:18:29,029 INFO [train.py:968] (0/2) Epoch 3, batch 43200, giga_loss[loss=0.3613, simple_loss=0.4129, pruned_loss=0.1548, over 28657.00 frames. ], tot_loss[loss=0.3635, simple_loss=0.4088, pruned_loss=0.1591, over 5643075.47 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4119, pruned_loss=0.1615, over 5627592.31 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.4087, pruned_loss=0.1588, over 5668190.78 frames. ], batch size: 92, lr: 9.01e-03, grad_scale: 8.0
+2023-03-01 23:18:40,792 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8540, 1.6106, 1.2965, 1.4718], device='cuda:0'), covar=tensor([0.0636, 0.0641, 0.0980, 0.0921], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0490, 0.0526, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 23:18:53,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:18:57,731 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-01 23:19:01,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.193e+02 1.490e+03 1.873e+03 2.550e+03 9.471e+03, threshold=3.746e+03, percent-clipped=5.0
+2023-03-01 23:19:17,992 INFO [train.py:968] (0/2) Epoch 3, batch 43250, libri_loss[loss=0.3205, simple_loss=0.3709, pruned_loss=0.135, over 29620.00 frames. ], tot_loss[loss=0.3589, simple_loss=0.4068, pruned_loss=0.1555, over 5656208.55 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4117, pruned_loss=0.1614, over 5630143.02 frames. ], giga_tot_loss[loss=0.3587, simple_loss=0.4068, pruned_loss=0.1553, over 5673716.35 frames. ], batch size: 69, lr: 9.01e-03, grad_scale: 8.0
+2023-03-01 23:19:42,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:19:45,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:20:00,979 INFO [train.py:968] (0/2) Epoch 3, batch 43300, libri_loss[loss=0.381, simple_loss=0.4327, pruned_loss=0.1646, over 26257.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.404, pruned_loss=0.1535, over 5658152.58 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.412, pruned_loss=0.1617, over 5635755.89 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4035, pruned_loss=0.1529, over 5668201.85 frames. ], batch size: 136, lr: 9.01e-03, grad_scale: 4.0
+2023-03-01 23:20:12,077 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:20:25,527 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-01 23:20:26,140 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-01 23:20:27,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5781, 1.8251, 1.7017, 1.6434], device='cuda:0'), covar=tensor([0.1164, 0.1629, 0.1008, 0.0891], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0789, 0.0747, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 23:20:34,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.947e+02 1.640e+03 2.098e+03 2.789e+03 5.344e+03, threshold=4.195e+03, percent-clipped=9.0
+2023-03-01 23:20:45,134 INFO [train.py:968] (0/2) Epoch 3, batch 43350, giga_loss[loss=0.4103, simple_loss=0.4327, pruned_loss=0.1939, over 28552.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4024, pruned_loss=0.153, over 5663464.22 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4115, pruned_loss=0.161, over 5646912.53 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.4021, pruned_loss=0.1528, over 5662076.23 frames. ], batch size: 336, lr: 9.01e-03, grad_scale: 4.0
+2023-03-01 23:20:49,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 23:20:58,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:21:01,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:21:30,247 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:21:31,514 INFO [train.py:968] (0/2) Epoch 3, batch 43400, giga_loss[loss=0.3378, simple_loss=0.3965, pruned_loss=0.1396, over 28628.00 frames. ], tot_loss[loss=0.3545, simple_loss=0.4019, pruned_loss=0.1535, over 5671595.54 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4115, pruned_loss=0.1611, over 5652969.35 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.4014, pruned_loss=0.1531, over 5665722.09 frames. ], batch size: 60, lr: 9.01e-03, grad_scale: 4.0
+2023-03-01 23:22:00,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5848, 3.3069, 3.3155, 1.5919], device='cuda:0'), covar=tensor([0.0580, 0.0519, 0.0857, 0.2259], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0652, 0.0816, 0.0585], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 23:22:05,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.613e+02 1.704e+03 2.333e+03 3.092e+03 8.307e+03, threshold=4.667e+03, percent-clipped=12.0
+2023-03-01 23:22:09,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5997, 3.0128, 1.5124, 1.5237], device='cuda:0'), covar=tensor([0.0777, 0.0315, 0.0832, 0.1198], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0459, 0.0306, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:0')
+2023-03-01 23:22:16,392 INFO [train.py:968] (0/2) Epoch 3, batch 43450, giga_loss[loss=0.3715, simple_loss=0.4216, pruned_loss=0.1607, over 29021.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4031, pruned_loss=0.1546, over 5676498.92 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4113, pruned_loss=0.1612, over 5661992.45 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4025, pruned_loss=0.1539, over 5664004.88 frames. ], batch size: 155, lr: 9.01e-03, grad_scale: 4.0
+2023-03-01 23:22:40,548 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133767.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 23:22:44,512 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:23:03,291 INFO [train.py:968] (0/2) Epoch 3, batch 43500, giga_loss[loss=0.3609, simple_loss=0.4255, pruned_loss=0.1482, over 28592.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4072, pruned_loss=0.1562, over 5677345.36 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4114, pruned_loss=0.1611, over 5665283.23 frames. ], giga_tot_loss[loss=0.3589, simple_loss=0.4066, pruned_loss=0.1556, over 5664695.86 frames. ], batch size: 336, lr: 9.00e-03, grad_scale: 4.0
+2023-03-01 23:23:27,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2130, 1.3669, 1.1480, 1.2304], device='cuda:0'), covar=tensor([0.2261, 0.2179, 0.2136, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.1044, 0.0846, 0.0940, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 23:23:40,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.018e+02 1.465e+03 1.953e+03 3.144e+03 9.472e+03, threshold=3.906e+03, percent-clipped=6.0
+2023-03-01 23:23:50,421 INFO [train.py:968] (0/2) Epoch 3, batch 43550, giga_loss[loss=0.3534, simple_loss=0.417, pruned_loss=0.1449, over 28953.00 frames. ], tot_loss[loss=0.3608, simple_loss=0.4107, pruned_loss=0.1555, over 5669997.59 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4116, pruned_loss=0.1613, over 5665534.65 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4101, pruned_loss=0.1548, over 5659950.92 frames. ], batch size: 145, lr: 9.00e-03, grad_scale: 4.0
+2023-03-01 23:23:53,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5216, 1.4438, 1.4038, 1.4378], device='cuda:0'), covar=tensor([0.0946, 0.1463, 0.1407, 0.1213], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0765, 0.0623, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 23:24:20,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:24:43,271 INFO [train.py:968] (0/2) Epoch 3, batch 43600, giga_loss[loss=0.3752, simple_loss=0.425, pruned_loss=0.1627, over 28898.00 frames. ], tot_loss[loss=0.365, simple_loss=0.414, pruned_loss=0.158, over 5672295.96 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4117, pruned_loss=0.1615, over 5668948.47 frames. ], giga_tot_loss[loss=0.3638, simple_loss=0.4133, pruned_loss=0.1572, over 5661608.11 frames. ], batch size: 227, lr: 9.00e-03, grad_scale: 8.0
+2023-03-01 23:25:20,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.759e+02 1.730e+03 2.430e+03 3.632e+03 7.923e+03, threshold=4.861e+03, percent-clipped=18.0
+2023-03-01 23:25:33,108 INFO [train.py:968] (0/2) Epoch 3, batch 43650, giga_loss[loss=0.4796, simple_loss=0.4853, pruned_loss=0.237, over 27522.00 frames. ], tot_loss[loss=0.3672, simple_loss=0.4158, pruned_loss=0.1593, over 5678504.99 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4117, pruned_loss=0.1614, over 5671163.86 frames. ], giga_tot_loss[loss=0.3664, simple_loss=0.4154, pruned_loss=0.1587, over 5668247.83 frames. ], batch size: 472, lr: 9.00e-03, grad_scale: 4.0
+2023-03-01 23:25:46,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1824, 1.2733, 1.1406, 1.3639], device='cuda:0'), covar=tensor([0.2313, 0.2137, 0.2036, 0.1785], device='cuda:0'), in_proj_covar=tensor([0.1034, 0.0833, 0.0932, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 23:26:22,759 INFO [train.py:968] (0/2) Epoch 3, batch 43700, giga_loss[loss=0.3829, simple_loss=0.4265, pruned_loss=0.1697, over 28682.00 frames. ], tot_loss[loss=0.369, simple_loss=0.4165, pruned_loss=0.1607, over 5671399.69 frames. ], libri_tot_loss[loss=0.3672, simple_loss=0.4117, pruned_loss=0.1613, over 5665999.87 frames. ], giga_tot_loss[loss=0.3685, simple_loss=0.4163, pruned_loss=0.1603, over 5667337.35 frames. ], batch size: 307, lr: 9.00e-03, grad_scale: 4.0
+2023-03-01 23:26:30,168 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-134000.pt
+2023-03-01 23:26:32,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3314, 1.3377, 1.1836, 1.4597], device='cuda:0'), covar=tensor([0.0799, 0.0331, 0.0349, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0218, 0.0141, 0.0146, 0.0259], device='cuda:0'), out_proj_covar=tensor([0.0045, 0.0034, 0.0030, 0.0050], device='cuda:0')
+2023-03-01 23:26:54,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.754e+02 1.503e+03 1.819e+03 2.409e+03 7.447e+03, threshold=3.638e+03, percent-clipped=3.0
+2023-03-01 23:27:05,492 INFO [train.py:968] (0/2) Epoch 3, batch 43750, giga_loss[loss=0.3615, simple_loss=0.4117, pruned_loss=0.1556, over 28309.00 frames. ], tot_loss[loss=0.3663, simple_loss=0.4139, pruned_loss=0.1594, over 5677047.59 frames. ], libri_tot_loss[loss=0.3664, simple_loss=0.411, pruned_loss=0.1609, over 5669528.94 frames. ], giga_tot_loss[loss=0.3666, simple_loss=0.4144, pruned_loss=0.1594, over 5670589.62 frames. ], batch size: 368, lr: 9.00e-03, grad_scale: 4.0
+2023-03-01 23:27:12,252 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:27:54,233 INFO [train.py:968] (0/2) Epoch 3, batch 43800, giga_loss[loss=0.3423, simple_loss=0.3993, pruned_loss=0.1426, over 28614.00 frames. ], tot_loss[loss=0.3661, simple_loss=0.4129, pruned_loss=0.1596, over 5670634.66 frames. ], libri_tot_loss[loss=0.3662, simple_loss=0.4107, pruned_loss=0.1608, over 5675265.57 frames. ], giga_tot_loss[loss=0.3666, simple_loss=0.4137, pruned_loss=0.1598, over 5660392.95 frames. ], batch size: 307, lr: 8.99e-03, grad_scale: 4.0
+2023-03-01 23:28:32,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.400e+02 1.824e+03 2.439e+03 3.154e+03 6.488e+03, threshold=4.877e+03, percent-clipped=17.0
+2023-03-01 23:28:44,218 INFO [train.py:968] (0/2) Epoch 3, batch 43850, giga_loss[loss=0.3648, simple_loss=0.4124, pruned_loss=0.1587, over 28322.00 frames. ], tot_loss[loss=0.3659, simple_loss=0.4117, pruned_loss=0.16, over 5671696.80 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.4107, pruned_loss=0.1608, over 5679781.27 frames. ], giga_tot_loss[loss=0.3663, simple_loss=0.4124, pruned_loss=0.1601, over 5659576.99 frames. ], batch size: 368, lr: 8.99e-03, grad_scale: 4.0
+2023-03-01 23:28:46,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134142.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 23:28:47,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:28:57,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2529, 1.3139, 1.1552, 1.4350], device='cuda:0'), covar=tensor([0.1801, 0.1729, 0.1611, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.1046, 0.0848, 0.0946, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 23:29:01,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4389, 2.0126, 1.5069, 1.6913], device='cuda:0'), covar=tensor([0.0601, 0.0727, 0.0975, 0.1024], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0490, 0.0522, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 23:29:07,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1828, 1.3114, 1.1631, 1.0546], device='cuda:0'), covar=tensor([0.1670, 0.1463, 0.1313, 0.1309], device='cuda:0'), in_proj_covar=tensor([0.1045, 0.0848, 0.0945, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 23:29:39,255 INFO [train.py:968] (0/2) Epoch 3, batch 43900, giga_loss[loss=0.3207, simple_loss=0.3818, pruned_loss=0.1298, over 29010.00 frames. ], tot_loss[loss=0.3652, simple_loss=0.4107, pruned_loss=0.1598, over 5649353.28 frames. ], libri_tot_loss[loss=0.3663, simple_loss=0.4108, pruned_loss=0.1609, over 5670377.36 frames. ], giga_tot_loss[loss=0.3654, simple_loss=0.4112, pruned_loss=0.1598, over 5648198.49 frames. ], batch size: 128, lr: 8.99e-03, grad_scale: 4.0
+2023-03-01 23:29:59,270 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-03-01 23:30:20,930 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.517e+02 1.525e+03 2.389e+03 3.896e+03 1.023e+04, threshold=4.778e+03, percent-clipped=17.0
+2023-03-01 23:30:31,205 INFO [train.py:968] (0/2) Epoch 3, batch 43950, giga_loss[loss=0.3564, simple_loss=0.4011, pruned_loss=0.1558, over 28974.00 frames. ], tot_loss[loss=0.3646, simple_loss=0.4102, pruned_loss=0.1595, over 5648487.73 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4113, pruned_loss=0.1611, over 5670578.53 frames. ], giga_tot_loss[loss=0.3643, simple_loss=0.4101, pruned_loss=0.1593, over 5646740.60 frames. ], batch size: 213, lr: 8.99e-03, grad_scale: 4.0
+2023-03-01 23:30:32,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:30:32,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-01 23:31:07,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 23:31:11,515 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134285.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 23:31:13,383 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:31:13,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134288.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 23:31:16,916 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:31:17,322 INFO [train.py:968] (0/2) Epoch 3, batch 44000, giga_loss[loss=0.3468, simple_loss=0.3957, pruned_loss=0.1489, over 28835.00 frames. ], tot_loss[loss=0.3613, simple_loss=0.4074, pruned_loss=0.1576, over 5664953.53 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4112, pruned_loss=0.161, over 5675395.10 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.4075, pruned_loss=0.1574, over 5659159.11 frames. ], batch size: 186, lr: 8.99e-03, grad_scale: 8.0
+2023-03-01 23:31:43,361 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134317.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 23:31:44,771 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:31:54,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.372e+02 1.526e+03 1.878e+03 2.461e+03 4.828e+03, threshold=3.755e+03, percent-clipped=1.0
+2023-03-01 23:32:06,344 INFO [train.py:968] (0/2) Epoch 3, batch 44050, giga_loss[loss=0.2853, simple_loss=0.3577, pruned_loss=0.1065, over 28911.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.4063, pruned_loss=0.1567, over 5657783.67 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4111, pruned_loss=0.1609, over 5677589.65 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4063, pruned_loss=0.1567, over 5651287.51 frames. ], batch size: 174, lr: 8.99e-03, grad_scale: 4.0
+2023-03-01 23:32:38,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7854, 2.0091, 1.8170, 1.7936], device='cuda:0'), covar=tensor([0.1466, 0.1604, 0.1159, 0.0984], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0787, 0.0752, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 23:32:48,651 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:32:55,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:32:57,292 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:32:57,684 INFO [train.py:968] (0/2) Epoch 3, batch 44100, giga_loss[loss=0.3299, simple_loss=0.3921, pruned_loss=0.1338, over 28828.00 frames. ], tot_loss[loss=0.3604, simple_loss=0.4074, pruned_loss=0.1566, over 5657821.66 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4113, pruned_loss=0.161, over 5680866.86 frames. ], giga_tot_loss[loss=0.36, simple_loss=0.4072, pruned_loss=0.1564, over 5649540.15 frames. ], batch size: 227, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:33:24,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:33:26,818 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:33:34,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.111e+03 1.605e+03 2.073e+03 2.636e+03 7.597e+03, threshold=4.147e+03, percent-clipped=12.0
+2023-03-01 23:33:43,373 INFO [train.py:968] (0/2) Epoch 3, batch 44150, libri_loss[loss=0.445, simple_loss=0.4675, pruned_loss=0.2113, over 20295.00 frames. ], tot_loss[loss=0.3653, simple_loss=0.4114, pruned_loss=0.1596, over 5654879.29 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.412, pruned_loss=0.1616, over 5676799.55 frames. ], giga_tot_loss[loss=0.3641, simple_loss=0.4106, pruned_loss=0.1588, over 5651752.27 frames. ], batch size: 187, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:34:08,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3704, 1.6402, 1.2128, 1.4714], device='cuda:0'), covar=tensor([0.0825, 0.0320, 0.0365, 0.0929], device='cuda:0'), in_proj_covar=tensor([0.0219, 0.0142, 0.0145, 0.0260], device='cuda:0'), out_proj_covar=tensor([0.0045, 0.0034, 0.0030, 0.0051], device='cuda:0')
+2023-03-01 23:34:12,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7356, 1.9648, 1.3285, 1.0909], device='cuda:0'), covar=tensor([0.0709, 0.0521, 0.0470, 0.0638], device='cuda:0'), in_proj_covar=tensor([0.1274, 0.1014, 0.1049, 0.1100], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 23:34:32,451 INFO [train.py:968] (0/2) Epoch 3, batch 44200, giga_loss[loss=0.3487, simple_loss=0.4009, pruned_loss=0.1482, over 28930.00 frames. ], tot_loss[loss=0.3648, simple_loss=0.4108, pruned_loss=0.1594, over 5654599.22 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.412, pruned_loss=0.1618, over 5672490.58 frames. ], giga_tot_loss[loss=0.3636, simple_loss=0.41, pruned_loss=0.1586, over 5656227.85 frames. ], batch size: 186, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:35:10,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.101e+02 1.504e+03 2.166e+03 3.046e+03 1.079e+04, threshold=4.331e+03, percent-clipped=9.0
+2023-03-01 23:35:22,273 INFO [train.py:968] (0/2) Epoch 3, batch 44250, giga_loss[loss=0.3218, simple_loss=0.3882, pruned_loss=0.1277, over 28681.00 frames. ], tot_loss[loss=0.3622, simple_loss=0.4111, pruned_loss=0.1566, over 5664664.19 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4121, pruned_loss=0.1617, over 5674810.00 frames. ], giga_tot_loss[loss=0.3612, simple_loss=0.4104, pruned_loss=0.156, over 5663842.01 frames. ], batch size: 242, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:35:42,672 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:35:46,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:35:50,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2898, 1.2554, 1.1586, 1.1402], device='cuda:0'), covar=tensor([0.2123, 0.2151, 0.2130, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.1043, 0.0838, 0.0941, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 23:36:05,456 INFO [train.py:968] (0/2) Epoch 3, batch 44300, giga_loss[loss=0.3656, simple_loss=0.4319, pruned_loss=0.1497, over 29117.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4117, pruned_loss=0.1547, over 5653596.36 frames. ], libri_tot_loss[loss=0.3683, simple_loss=0.4125, pruned_loss=0.162, over 5666178.73 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4108, pruned_loss=0.1537, over 5661194.76 frames. ], batch size: 155, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:36:08,380 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:36:41,003 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.141e+02 1.499e+03 2.051e+03 2.656e+03 7.962e+03, threshold=4.103e+03, percent-clipped=7.0
+2023-03-01 23:36:54,821 INFO [train.py:968] (0/2) Epoch 3, batch 44350, giga_loss[loss=0.2954, simple_loss=0.3666, pruned_loss=0.1121, over 28393.00 frames. ], tot_loss[loss=0.3624, simple_loss=0.4134, pruned_loss=0.1557, over 5643103.49 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4119, pruned_loss=0.1616, over 5660631.68 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.4131, pruned_loss=0.155, over 5653544.74 frames. ], batch size: 71, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:37:24,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3561, 1.5529, 1.2275, 0.8082], device='cuda:0'), covar=tensor([0.0905, 0.0596, 0.0502, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.1253, 0.0994, 0.1021, 0.1081], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 23:37:42,254 INFO [train.py:968] (0/2) Epoch 3, batch 44400, giga_loss[loss=0.4265, simple_loss=0.4586, pruned_loss=0.1972, over 28635.00 frames. ], tot_loss[loss=0.3685, simple_loss=0.4174, pruned_loss=0.1598, over 5652101.62 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.412, pruned_loss=0.1616, over 5666159.63 frames. ], giga_tot_loss[loss=0.3678, simple_loss=0.4172, pruned_loss=0.1592, over 5654851.46 frames. ], batch size: 307, lr: 8.97e-03, grad_scale: 8.0
+2023-03-01 23:37:47,579 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:37:51,768 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:38:02,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6242, 1.9081, 1.3477, 0.9859], device='cuda:0'), covar=tensor([0.0992, 0.0663, 0.0575, 0.0914], device='cuda:0'), in_proj_covar=tensor([0.1262, 0.1005, 0.1023, 0.1088], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-01 23:38:18,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.577e+03 2.200e+03 2.758e+03 5.895e+03, threshold=4.401e+03, percent-clipped=8.0
+2023-03-01 23:38:29,218 INFO [train.py:968] (0/2) Epoch 3, batch 44450, giga_loss[loss=0.349, simple_loss=0.4068, pruned_loss=0.1456, over 28631.00 frames. ], tot_loss[loss=0.3704, simple_loss=0.418, pruned_loss=0.1614, over 5654143.52 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4118, pruned_loss=0.1613, over 5671917.69 frames. ], giga_tot_loss[loss=0.3703, simple_loss=0.4182, pruned_loss=0.1612, over 5650978.97 frames. ], batch size: 65, lr: 8.97e-03, grad_scale: 8.0
+2023-03-01 23:38:38,441 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-01 23:38:54,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:39:19,496 INFO [train.py:968] (0/2) Epoch 3, batch 44500, giga_loss[loss=0.3883, simple_loss=0.4322, pruned_loss=0.1721, over 28874.00 frames. ], tot_loss[loss=0.3692, simple_loss=0.4169, pruned_loss=0.1608, over 5670840.87 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4116, pruned_loss=0.1613, over 5674233.84 frames. ], giga_tot_loss[loss=0.3693, simple_loss=0.4173, pruned_loss=0.1606, over 5666095.47 frames. ], batch size: 186, lr: 8.97e-03, grad_scale: 4.0
+2023-03-01 23:39:54,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.376e+02 1.590e+03 2.023e+03 3.110e+03 9.228e+03, threshold=4.047e+03, percent-clipped=13.0
+2023-03-01 23:40:02,641 INFO [train.py:968] (0/2) Epoch 3, batch 44550, giga_loss[loss=0.3135, simple_loss=0.3838, pruned_loss=0.1216, over 28857.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.4161, pruned_loss=0.1604, over 5657640.23 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4117, pruned_loss=0.1616, over 5666267.56 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.4165, pruned_loss=0.16, over 5659982.60 frames. ], batch size: 145, lr: 8.97e-03, grad_scale: 4.0
+2023-03-01 23:40:04,511 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5882, 1.1190, 2.8718, 2.6305], device='cuda:0'), covar=tensor([0.1514, 0.1820, 0.0470, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0545, 0.0513, 0.0719, 0.0577], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 23:40:09,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5877, 1.5147, 1.5641, 1.5855], device='cuda:0'), covar=tensor([0.0863, 0.1346, 0.1199, 0.1038], device='cuda:0'), in_proj_covar=tensor([0.0439, 0.0750, 0.0622, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 23:40:13,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9719, 1.2527, 0.9722, 0.2973], device='cuda:0'), covar=tensor([0.1185, 0.1060, 0.1807, 0.1911], device='cuda:0'), in_proj_covar=tensor([0.1265, 0.1225, 0.1287, 0.1100], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-01 23:40:47,533 INFO [train.py:968] (0/2) Epoch 3, batch 44600, giga_loss[loss=0.3089, simple_loss=0.3926, pruned_loss=0.1126, over 28838.00 frames. ], tot_loss[loss=0.3647, simple_loss=0.4147, pruned_loss=0.1574, over 5667602.68 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4115, pruned_loss=0.1615, over 5662803.24 frames. ], giga_tot_loss[loss=0.3647, simple_loss=0.4153, pruned_loss=0.1571, over 5672491.25 frames. ], batch size: 174, lr: 8.97e-03, grad_scale: 4.0
+2023-03-01 23:41:04,099 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:41:06,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:41:25,912 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.894e+02 1.519e+03 1.781e+03 2.330e+03 5.055e+03, threshold=3.562e+03, percent-clipped=3.0
+2023-03-01 23:41:30,294 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-01 23:41:33,968 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:41:34,498 INFO [train.py:968] (0/2) Epoch 3, batch 44650, giga_loss[loss=0.3235, simple_loss=0.3899, pruned_loss=0.1286, over 28651.00 frames. ], tot_loss[loss=0.3619, simple_loss=0.4139, pruned_loss=0.1549, over 5672694.12 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4114, pruned_loss=0.1613, over 5663537.92 frames. ], giga_tot_loss[loss=0.3621, simple_loss=0.4145, pruned_loss=0.1548, over 5675939.13 frames. ], batch size: 242, lr: 8.97e-03, grad_scale: 4.0
+2023-03-01 23:41:57,493 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 23:42:19,767 INFO [train.py:968] (0/2) Epoch 3, batch 44700, giga_loss[loss=0.4039, simple_loss=0.4346, pruned_loss=0.1866, over 27568.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4153, pruned_loss=0.1567, over 5664019.36 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4114, pruned_loss=0.1612, over 5669862.20 frames. ], giga_tot_loss[loss=0.3644, simple_loss=0.4159, pruned_loss=0.1564, over 5661035.35 frames. ], batch size: 472, lr: 8.96e-03, grad_scale: 4.0
+2023-03-01 23:42:22,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0234, 1.4639, 1.0975, 1.2026], device='cuda:0'), covar=tensor([0.0912, 0.0349, 0.0361, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0219, 0.0142, 0.0145, 0.0260], device='cuda:0'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:0')
+2023-03-01 23:42:35,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2786, 4.0398, 3.9877, 1.9224], device='cuda:0'), covar=tensor([0.0487, 0.0438, 0.0790, 0.1968], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0666, 0.0833, 0.0589], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-01 23:43:01,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.592e+03 2.139e+03 2.869e+03 7.112e+03, threshold=4.278e+03, percent-clipped=8.0
+2023-03-01 23:43:09,739 INFO [train.py:968] (0/2) Epoch 3, batch 44750, giga_loss[loss=0.3238, simple_loss=0.3848, pruned_loss=0.1313, over 28850.00 frames. ], tot_loss[loss=0.3647, simple_loss=0.415, pruned_loss=0.1572, over 5658951.97 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4115, pruned_loss=0.1613, over 5669778.73 frames. ], giga_tot_loss[loss=0.3647, simple_loss=0.4155, pruned_loss=0.1569, over 5656552.98 frames. ], batch size: 186, lr: 8.96e-03, grad_scale: 4.0
+2023-03-01 23:43:32,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4983, 1.5007, 1.3445, 1.4985], device='cuda:0'), covar=tensor([0.1183, 0.1656, 0.1479, 0.1429], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0759, 0.0628, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 23:43:40,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:43:44,324 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:43:57,862 INFO [train.py:968] (0/2) Epoch 3, batch 44800, giga_loss[loss=0.3233, simple_loss=0.3824, pruned_loss=0.1321, over 28860.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.4124, pruned_loss=0.1561, over 5665567.30 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4115, pruned_loss=0.1612, over 5673595.90 frames. ], giga_tot_loss[loss=0.3623, simple_loss=0.4128, pruned_loss=0.1559, over 5660203.13 frames. ], batch size: 112, lr: 8.96e-03, grad_scale: 8.0
+2023-03-01 23:44:39,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.771e+02 1.674e+03 2.337e+03 3.115e+03 8.459e+03, threshold=4.675e+03, percent-clipped=8.0
+2023-03-01 23:44:47,407 INFO [train.py:968] (0/2) Epoch 3, batch 44850, giga_loss[loss=0.3488, simple_loss=0.4057, pruned_loss=0.1459, over 28927.00 frames. ], tot_loss[loss=0.3639, simple_loss=0.4124, pruned_loss=0.1577, over 5651091.70 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.412, pruned_loss=0.1618, over 5663972.43 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.4122, pruned_loss=0.1569, over 5655825.10 frames. ], batch size: 136, lr: 8.96e-03, grad_scale: 4.0
+2023-03-01 23:45:00,735 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-01 23:45:34,096 INFO [train.py:968] (0/2) Epoch 3, batch 44900, libri_loss[loss=0.3496, simple_loss=0.409, pruned_loss=0.1451, over 29552.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.4102, pruned_loss=0.1563, over 5659553.35 frames. ], libri_tot_loss[loss=0.3682, simple_loss=0.4125, pruned_loss=0.162, over 5667612.84 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4096, pruned_loss=0.1555, over 5659744.77 frames. ], batch size: 83, lr: 8.96e-03, grad_scale: 4.0
+2023-03-01 23:45:55,706 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:45:59,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:46:01,534 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:46:04,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:46:12,181 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.814e+02 1.824e+03 2.534e+03 3.458e+03 7.313e+03, threshold=5.068e+03, percent-clipped=8.0
+2023-03-01 23:46:19,813 INFO [train.py:968] (0/2) Epoch 3, batch 44950, libri_loss[loss=0.3669, simple_loss=0.4141, pruned_loss=0.1598, over 29522.00 frames. ], tot_loss[loss=0.3637, simple_loss=0.4107, pruned_loss=0.1584, over 5656756.45 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4127, pruned_loss=0.162, over 5671563.74 frames. ], giga_tot_loss[loss=0.3624, simple_loss=0.4099, pruned_loss=0.1574, over 5653245.14 frames. ], batch size: 82, lr: 8.96e-03, grad_scale: 4.0
+2023-03-01 23:46:24,194 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:46:25,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-01 23:46:29,557 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:46:31,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=135255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:47:03,564 INFO [train.py:968] (0/2) Epoch 3, batch 45000, libri_loss[loss=0.425, simple_loss=0.4504, pruned_loss=0.1998, over 19024.00 frames. ], tot_loss[loss=0.3621, simple_loss=0.409, pruned_loss=0.1576, over 5634934.75 frames. ], libri_tot_loss[loss=0.3681, simple_loss=0.4125, pruned_loss=0.1619, over 5660770.68 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.4085, pruned_loss=0.1568, over 5642481.84 frames. ], batch size: 187, lr: 8.95e-03, grad_scale: 4.0
+2023-03-01 23:47:03,568 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-01 23:47:11,676 INFO [train.py:1012] (0/2) Epoch 3, validation: loss=0.2493, simple_loss=0.3509, pruned_loss=0.07382, over 944034.00 frames. 
+2023-03-01 23:47:11,677 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-01 23:47:12,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1168, 1.2236, 1.1837, 1.2130], device='cuda:0'), covar=tensor([0.0866, 0.0952, 0.1385, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0769, 0.0638, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 23:47:21,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9484, 1.1255, 3.8737, 3.3391], device='cuda:0'), covar=tensor([0.1596, 0.2166, 0.0354, 0.1062], device='cuda:0'), in_proj_covar=tensor([0.0551, 0.0517, 0.0727, 0.0585], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-01 23:47:47,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.929e+02 1.450e+03 1.997e+03 2.543e+03 6.396e+03, threshold=3.993e+03, percent-clipped=3.0
+2023-03-01 23:47:55,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3994, 1.4684, 1.3282, 1.3886], device='cuda:0'), covar=tensor([0.1372, 0.1825, 0.1713, 0.1591], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0763, 0.0635, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 23:47:56,014 INFO [train.py:968] (0/2) Epoch 3, batch 45050, giga_loss[loss=0.3169, simple_loss=0.3947, pruned_loss=0.1195, over 28991.00 frames. ], tot_loss[loss=0.3577, simple_loss=0.4063, pruned_loss=0.1545, over 5641667.07 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4119, pruned_loss=0.1614, over 5665303.40 frames. ], giga_tot_loss[loss=0.3574, simple_loss=0.4063, pruned_loss=0.1543, over 5643450.85 frames. ], batch size: 164, lr: 8.95e-03, grad_scale: 4.0
+2023-03-01 23:48:34,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7752, 1.9718, 1.8517, 1.7603], device='cuda:0'), covar=tensor([0.1504, 0.1721, 0.1150, 0.1042], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0797, 0.0753, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 23:48:41,840 INFO [train.py:968] (0/2) Epoch 3, batch 45100, giga_loss[loss=0.2878, simple_loss=0.3407, pruned_loss=0.1175, over 23775.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.4013, pruned_loss=0.149, over 5652249.17 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4112, pruned_loss=0.1609, over 5672777.36 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4017, pruned_loss=0.149, over 5646544.83 frames. ], batch size: 705, lr: 8.95e-03, grad_scale: 4.0
+2023-03-01 23:48:58,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3611, 1.4000, 1.3699, 1.3840], device='cuda:0'), covar=tensor([0.0849, 0.1208, 0.1298, 0.1095], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0755, 0.0631, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-01 23:49:21,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.045e+02 1.472e+03 1.787e+03 2.474e+03 6.615e+03, threshold=3.574e+03, percent-clipped=6.0
+2023-03-01 23:49:29,177 INFO [train.py:968] (0/2) Epoch 3, batch 45150, giga_loss[loss=0.3489, simple_loss=0.4027, pruned_loss=0.1475, over 28662.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.4012, pruned_loss=0.1491, over 5644353.40 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4112, pruned_loss=0.161, over 5669298.07 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.4012, pruned_loss=0.1486, over 5641824.84 frames. ], batch size: 92, lr: 8.95e-03, grad_scale: 4.0
+2023-03-01 23:49:45,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-01 23:50:12,231 INFO [train.py:968] (0/2) Epoch 3, batch 45200, giga_loss[loss=0.3659, simple_loss=0.4073, pruned_loss=0.1622, over 28368.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.3998, pruned_loss=0.1483, over 5666667.93 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.411, pruned_loss=0.1606, over 5675704.01 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.3997, pruned_loss=0.1477, over 5658044.97 frames. ], batch size: 368, lr: 8.95e-03, grad_scale: 8.0
+2023-03-01 23:50:58,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.505e+02 1.538e+03 2.202e+03 3.505e+03 1.072e+04, threshold=4.405e+03, percent-clipped=22.0
+2023-03-01 23:51:06,980 INFO [train.py:968] (0/2) Epoch 3, batch 45250, giga_loss[loss=0.4281, simple_loss=0.441, pruned_loss=0.2076, over 26734.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3971, pruned_loss=0.1474, over 5671032.76 frames. ], libri_tot_loss[loss=0.3659, simple_loss=0.4108, pruned_loss=0.1605, over 5677385.56 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.3969, pruned_loss=0.1469, over 5662584.80 frames. ], batch size: 555, lr: 8.95e-03, grad_scale: 2.0
+2023-03-01 23:51:25,679 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=135564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:51:49,340 INFO [train.py:968] (0/2) Epoch 3, batch 45300, giga_loss[loss=0.3429, simple_loss=0.3956, pruned_loss=0.1451, over 28238.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.3972, pruned_loss=0.1468, over 5686915.49 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.4102, pruned_loss=0.16, over 5683285.24 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3973, pruned_loss=0.1466, over 5674842.79 frames. ], batch size: 77, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:52:08,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8101, 1.6392, 1.2034, 1.4311], device='cuda:0'), covar=tensor([0.0561, 0.0580, 0.0929, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0481, 0.0524, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-01 23:52:25,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:52:31,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.934e+02 1.486e+03 1.976e+03 2.602e+03 8.868e+03, threshold=3.952e+03, percent-clipped=6.0
+2023-03-01 23:52:37,789 INFO [train.py:968] (0/2) Epoch 3, batch 45350, giga_loss[loss=0.3172, simple_loss=0.3838, pruned_loss=0.1253, over 28850.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.3997, pruned_loss=0.1483, over 5679705.27 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.4103, pruned_loss=0.16, over 5683470.18 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.3995, pruned_loss=0.148, over 5670281.38 frames. ], batch size: 112, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:53:24,546 INFO [train.py:968] (0/2) Epoch 3, batch 45400, giga_loss[loss=0.4004, simple_loss=0.4331, pruned_loss=0.1839, over 27631.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.4019, pruned_loss=0.1499, over 5681400.64 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4117, pruned_loss=0.1609, over 5690105.65 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4002, pruned_loss=0.1484, over 5667373.86 frames. ], batch size: 472, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:54:03,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.598e+03 2.148e+03 2.956e+03 5.677e+03, threshold=4.296e+03, percent-clipped=9.0
+2023-03-01 23:54:10,931 INFO [train.py:968] (0/2) Epoch 3, batch 45450, giga_loss[loss=0.3513, simple_loss=0.3999, pruned_loss=0.1513, over 28923.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4033, pruned_loss=0.1518, over 5679539.68 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4123, pruned_loss=0.1614, over 5692577.39 frames. ], giga_tot_loss[loss=0.3505, simple_loss=0.4012, pruned_loss=0.1499, over 5666080.44 frames. ], batch size: 213, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:54:39,412 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:54:43,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:54:54,582 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-01 23:54:54,839 INFO [train.py:968] (0/2) Epoch 3, batch 45500, giga_loss[loss=0.3092, simple_loss=0.3715, pruned_loss=0.1235, over 28395.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.4039, pruned_loss=0.1527, over 5669543.70 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.412, pruned_loss=0.1613, over 5695329.10 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.4022, pruned_loss=0.1511, over 5655360.95 frames. ], batch size: 65, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:55:09,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:55:41,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.548e+02 1.678e+03 2.112e+03 2.852e+03 6.100e+03, threshold=4.224e+03, percent-clipped=5.0
+2023-03-01 23:55:46,144 INFO [train.py:968] (0/2) Epoch 3, batch 45550, giga_loss[loss=0.3599, simple_loss=0.4192, pruned_loss=0.1504, over 28565.00 frames. ], tot_loss[loss=0.3591, simple_loss=0.4072, pruned_loss=0.1555, over 5650723.37 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.4123, pruned_loss=0.1616, over 5697452.28 frames. ], giga_tot_loss[loss=0.3566, simple_loss=0.4055, pruned_loss=0.1539, over 5637106.64 frames. ], batch size: 85, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:56:33,147 INFO [train.py:968] (0/2) Epoch 3, batch 45600, giga_loss[loss=0.3323, simple_loss=0.3931, pruned_loss=0.1358, over 28848.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4088, pruned_loss=0.1562, over 5657243.89 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4122, pruned_loss=0.1616, over 5691635.44 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.4075, pruned_loss=0.1548, over 5651644.10 frames. ], batch size: 174, lr: 8.93e-03, grad_scale: 4.0
+2023-03-01 23:56:53,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3361, 1.3900, 1.2189, 1.4925], device='cuda:0'), covar=tensor([0.2011, 0.2013, 0.1898, 0.2058], device='cuda:0'), in_proj_covar=tensor([0.1051, 0.0855, 0.0946, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 23:57:12,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.866e+02 1.669e+03 2.219e+03 3.601e+03 1.171e+04, threshold=4.437e+03, percent-clipped=12.0
+2023-03-01 23:57:17,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:57:19,598 INFO [train.py:968] (0/2) Epoch 3, batch 45650, giga_loss[loss=0.368, simple_loss=0.4076, pruned_loss=0.1641, over 27645.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4107, pruned_loss=0.1583, over 5646925.37 frames. ], libri_tot_loss[loss=0.3681, simple_loss=0.4125, pruned_loss=0.1618, over 5685469.36 frames. ], giga_tot_loss[loss=0.3614, simple_loss=0.4093, pruned_loss=0.1567, over 5646209.44 frames. ], batch size: 472, lr: 8.93e-03, grad_scale: 4.0
+2023-03-01 23:58:10,717 INFO [train.py:968] (0/2) Epoch 3, batch 45700, giga_loss[loss=0.4175, simple_loss=0.4454, pruned_loss=0.1948, over 26575.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4113, pruned_loss=0.1586, over 5642809.48 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4122, pruned_loss=0.1615, over 5678054.73 frames. ], giga_tot_loss[loss=0.3628, simple_loss=0.4104, pruned_loss=0.1576, over 5648795.98 frames. ], batch size: 555, lr: 8.93e-03, grad_scale: 4.0
+2023-03-01 23:58:19,528 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2847, 1.3779, 1.1728, 1.3544], device='cuda:0'), covar=tensor([0.1979, 0.1911, 0.1842, 0.1928], device='cuda:0'), in_proj_covar=tensor([0.1039, 0.0844, 0.0937, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-01 23:58:21,014 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-136000.pt
+2023-03-01 23:58:44,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:58:54,383 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.212e+02 1.533e+03 2.038e+03 2.849e+03 1.479e+04, threshold=4.076e+03, percent-clipped=5.0
+2023-03-01 23:59:02,051 INFO [train.py:968] (0/2) Epoch 3, batch 45750, giga_loss[loss=0.3472, simple_loss=0.4025, pruned_loss=0.1459, over 28903.00 frames. ], tot_loss[loss=0.3634, simple_loss=0.4115, pruned_loss=0.1576, over 5611092.47 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4126, pruned_loss=0.1621, over 5643857.82 frames. ], giga_tot_loss[loss=0.3615, simple_loss=0.4105, pruned_loss=0.1562, over 5646702.44 frames. ], batch size: 145, lr: 8.93e-03, grad_scale: 4.0
+2023-03-01 23:59:43,429 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:59:46,145 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:59:51,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5274, 1.8316, 1.6347, 1.6200], device='cuda:0'), covar=tensor([0.0827, 0.0307, 0.0315, 0.0876], device='cuda:0'), in_proj_covar=tensor([0.0220, 0.0144, 0.0146, 0.0262], device='cuda:0'), out_proj_covar=tensor([0.0046, 0.0035, 0.0030, 0.0052], device='cuda:0')
+2023-03-01 23:59:51,624 INFO [train.py:968] (0/2) Epoch 3, batch 45800, giga_loss[loss=0.3158, simple_loss=0.3773, pruned_loss=0.1272, over 28541.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4116, pruned_loss=0.1578, over 5586534.32 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4129, pruned_loss=0.1624, over 5602972.33 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.4105, pruned_loss=0.1564, over 5649930.30 frames. ], batch size: 71, lr: 8.93e-03, grad_scale: 4.0
+2023-03-02 00:00:11,945 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:00:24,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4800, 4.1710, 4.1874, 1.9438], device='cuda:0'), covar=tensor([0.0420, 0.0427, 0.0825, 0.1809], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0662, 0.0822, 0.0581], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 00:00:29,728 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.323e+02 1.437e+03 1.994e+03 2.853e+03 6.518e+03, threshold=3.988e+03, percent-clipped=8.0
+2023-03-02 00:00:38,114 INFO [train.py:968] (0/2) Epoch 3, batch 45850, giga_loss[loss=0.3372, simple_loss=0.3862, pruned_loss=0.1441, over 28805.00 frames. ], tot_loss[loss=0.3644, simple_loss=0.4114, pruned_loss=0.1587, over 5562074.39 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4136, pruned_loss=0.1631, over 5553422.71 frames. ], giga_tot_loss[loss=0.3617, simple_loss=0.4099, pruned_loss=0.1568, over 5655398.81 frames. ], batch size: 119, lr: 8.93e-03, grad_scale: 4.0
+2023-03-02 00:00:42,033 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-02 00:00:44,095 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-3.pt
+2023-03-02 00:01:45,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5264, 1.8730, 1.7517, 1.6507], device='cuda:0'), covar=tensor([0.1466, 0.1779, 0.1189, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0788, 0.0745, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 00:02:04,849 INFO [train.py:968] (0/2) Epoch 4, batch 50, giga_loss[loss=0.3651, simple_loss=0.4187, pruned_loss=0.1557, over 27623.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.4141, pruned_loss=0.1438, over 1262064.51 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.386, pruned_loss=0.1242, over 88220.30 frames. ], giga_tot_loss[loss=0.3532, simple_loss=0.416, pruned_loss=0.1451, over 1191893.65 frames. ], batch size: 472, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:02:44,071 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.380e+02 1.114e+03 1.447e+03 1.782e+03 7.584e+03, threshold=2.895e+03, percent-clipped=3.0
+2023-03-02 00:02:51,336 INFO [train.py:968] (0/2) Epoch 4, batch 100, giga_loss[loss=0.3047, simple_loss=0.3789, pruned_loss=0.1152, over 28857.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3996, pruned_loss=0.1345, over 2240891.11 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3857, pruned_loss=0.1257, over 277781.61 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.4014, pruned_loss=0.1356, over 2064280.47 frames. ], batch size: 174, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:02:54,825 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 00:03:35,825 INFO [train.py:968] (0/2) Epoch 4, batch 150, libri_loss[loss=0.3327, simple_loss=0.4005, pruned_loss=0.1325, over 29149.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.384, pruned_loss=0.1267, over 3009055.57 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3857, pruned_loss=0.1259, over 469838.60 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3843, pruned_loss=0.1271, over 2765758.52 frames. ], batch size: 101, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:04:08,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.369e+02 1.078e+03 1.416e+03 1.963e+03 8.183e+03, threshold=2.832e+03, percent-clipped=8.0
+2023-03-02 00:04:16,438 INFO [train.py:968] (0/2) Epoch 4, batch 200, giga_loss[loss=0.2572, simple_loss=0.3235, pruned_loss=0.09548, over 28959.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3697, pruned_loss=0.1194, over 3611471.26 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3843, pruned_loss=0.1254, over 575521.83 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3689, pruned_loss=0.1191, over 3374213.48 frames. ], batch size: 106, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:04:43,935 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:04:53,047 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:04:58,423 INFO [train.py:968] (0/2) Epoch 4, batch 250, giga_loss[loss=0.2414, simple_loss=0.3127, pruned_loss=0.08503, over 28610.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3571, pruned_loss=0.1125, over 4076091.13 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3801, pruned_loss=0.1225, over 703660.88 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3558, pruned_loss=0.1121, over 3844097.14 frames. ], batch size: 336, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:05:04,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:05:34,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.589e+02 9.699e+02 1.399e+03 1.644e+03 4.267e+03, threshold=2.798e+03, percent-clipped=4.0
+2023-03-02 00:05:42,123 INFO [train.py:968] (0/2) Epoch 4, batch 300, giga_loss[loss=0.2743, simple_loss=0.3337, pruned_loss=0.1074, over 28893.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3462, pruned_loss=0.107, over 4438422.48 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3804, pruned_loss=0.1223, over 755554.99 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3443, pruned_loss=0.1063, over 4241305.70 frames. ], batch size: 106, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:06:26,566 INFO [train.py:968] (0/2) Epoch 4, batch 350, giga_loss[loss=0.2379, simple_loss=0.3022, pruned_loss=0.08681, over 28761.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3383, pruned_loss=0.1033, over 4720585.78 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3792, pruned_loss=0.1211, over 857973.03 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.336, pruned_loss=0.1025, over 4540487.93 frames. ], batch size: 99, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:06:56,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9940, 1.0328, 4.4506, 3.3776], device='cuda:0'), covar=tensor([0.1638, 0.2307, 0.0298, 0.0691], device='cuda:0'), in_proj_covar=tensor([0.0537, 0.0507, 0.0707, 0.0577], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:06:59,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 4.981e+02 1.079e+03 1.303e+03 1.648e+03 4.045e+03, threshold=2.606e+03, percent-clipped=4.0
+2023-03-02 00:07:00,619 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:07:02,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4671, 1.9067, 1.7383, 1.6242], device='cuda:0'), covar=tensor([0.1543, 0.1862, 0.1258, 0.1099], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0792, 0.0762, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 00:07:06,450 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:07:07,421 INFO [train.py:968] (0/2) Epoch 4, batch 400, giga_loss[loss=0.2717, simple_loss=0.3304, pruned_loss=0.1065, over 28901.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.335, pruned_loss=0.1016, over 4945252.32 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3811, pruned_loss=0.1217, over 1029101.18 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3311, pruned_loss=0.1, over 4768999.64 frames. ], batch size: 227, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:07:10,405 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:07:32,787 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:07:39,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3368, 4.0793, 4.0170, 1.7784], device='cuda:0'), covar=tensor([0.0446, 0.0391, 0.0750, 0.2138], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0635, 0.0786, 0.0567], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 00:07:46,622 INFO [train.py:968] (0/2) Epoch 4, batch 450, giga_loss[loss=0.2621, simple_loss=0.3249, pruned_loss=0.09962, over 28614.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3336, pruned_loss=0.1007, over 5116034.20 frames. ], libri_tot_loss[loss=0.3133, simple_loss=0.3811, pruned_loss=0.1227, over 1258767.41 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3281, pruned_loss=0.09812, over 4940643.26 frames. ], batch size: 307, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:08:25,732 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.451e+02 1.091e+03 1.407e+03 1.740e+03 3.767e+03, threshold=2.815e+03, percent-clipped=7.0
+2023-03-02 00:08:32,020 INFO [train.py:968] (0/2) Epoch 4, batch 500, giga_loss[loss=0.2334, simple_loss=0.3098, pruned_loss=0.07853, over 28913.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3298, pruned_loss=0.09849, over 5253188.92 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3815, pruned_loss=0.1232, over 1350891.53 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3242, pruned_loss=0.09584, over 5101104.40 frames. ], batch size: 213, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:08:35,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-02 00:09:05,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2911, 1.5241, 1.2049, 0.7072], device='cuda:0'), covar=tensor([0.0757, 0.0598, 0.0449, 0.0783], device='cuda:0'), in_proj_covar=tensor([0.1263, 0.1005, 0.1019, 0.1085], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 00:09:16,708 INFO [train.py:968] (0/2) Epoch 4, batch 550, giga_loss[loss=0.2317, simple_loss=0.3058, pruned_loss=0.07874, over 28833.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3272, pruned_loss=0.09701, over 5351902.30 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3822, pruned_loss=0.1238, over 1462045.92 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3213, pruned_loss=0.09409, over 5218384.89 frames. ], batch size: 199, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:09:47,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2793, 1.3711, 1.1988, 1.4789], device='cuda:0'), covar=tensor([0.2065, 0.1920, 0.1926, 0.2002], device='cuda:0'), in_proj_covar=tensor([0.1066, 0.0852, 0.0950, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:09:50,535 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:09:54,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.082e+02 9.468e+02 1.303e+03 1.742e+03 5.692e+03, threshold=2.607e+03, percent-clipped=3.0
+2023-03-02 00:10:03,017 INFO [train.py:968] (0/2) Epoch 4, batch 600, giga_loss[loss=0.2803, simple_loss=0.3346, pruned_loss=0.113, over 28045.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3246, pruned_loss=0.09578, over 5418750.05 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3819, pruned_loss=0.1235, over 1538201.62 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.319, pruned_loss=0.09308, over 5315152.53 frames. ], batch size: 412, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:10:11,116 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:10:20,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3540, 3.9834, 4.0608, 2.0873], device='cuda:0'), covar=tensor([0.0453, 0.0451, 0.0821, 0.1857], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0635, 0.0784, 0.0572], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 00:10:24,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:10:52,051 INFO [train.py:968] (0/2) Epoch 4, batch 650, giga_loss[loss=0.2884, simple_loss=0.3449, pruned_loss=0.1159, over 28285.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3223, pruned_loss=0.09446, over 5467980.35 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3818, pruned_loss=0.1231, over 1639405.63 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3166, pruned_loss=0.09179, over 5381250.19 frames. ], batch size: 369, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:11:29,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.938e+02 9.233e+02 1.160e+03 1.591e+03 3.022e+03, threshold=2.320e+03, percent-clipped=2.0
+2023-03-02 00:11:29,948 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-02 00:11:37,724 INFO [train.py:968] (0/2) Epoch 4, batch 700, giga_loss[loss=0.24, simple_loss=0.3084, pruned_loss=0.08582, over 28701.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.319, pruned_loss=0.09293, over 5520556.98 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3821, pruned_loss=0.1233, over 1703465.31 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3134, pruned_loss=0.09029, over 5445984.73 frames. ], batch size: 284, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:12:27,495 INFO [train.py:968] (0/2) Epoch 4, batch 750, giga_loss[loss=0.2141, simple_loss=0.2911, pruned_loss=0.06857, over 28333.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3152, pruned_loss=0.09067, over 5565904.55 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3821, pruned_loss=0.123, over 1724321.98 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3105, pruned_loss=0.08852, over 5506141.25 frames. ], batch size: 65, lr: 8.32e-03, grad_scale: 8.0
+2023-03-02 00:12:27,821 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:30,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:39,565 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:41,463 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:43,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:55,174 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:59,200 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-02 00:13:02,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.197e+02 1.023e+03 1.233e+03 1.608e+03 2.856e+03, threshold=2.466e+03, percent-clipped=6.0
+2023-03-02 00:13:05,312 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:13:11,462 INFO [train.py:968] (0/2) Epoch 4, batch 800, giga_loss[loss=0.2266, simple_loss=0.2943, pruned_loss=0.07948, over 28496.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3144, pruned_loss=0.09091, over 5588136.08 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3844, pruned_loss=0.1248, over 1801497.36 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3088, pruned_loss=0.08809, over 5540306.34 frames. ], batch size: 71, lr: 8.32e-03, grad_scale: 8.0
+2023-03-02 00:14:00,994 INFO [train.py:968] (0/2) Epoch 4, batch 850, giga_loss[loss=0.4009, simple_loss=0.4375, pruned_loss=0.1821, over 27508.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3256, pruned_loss=0.09816, over 5603450.03 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3845, pruned_loss=0.1247, over 1862450.40 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3203, pruned_loss=0.09559, over 5561929.77 frames. ], batch size: 472, lr: 8.32e-03, grad_scale: 8.0
+2023-03-02 00:14:08,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5408, 3.8826, 1.5057, 1.6611], device='cuda:0'), covar=tensor([0.1107, 0.0304, 0.1017, 0.1534], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0451, 0.0308, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:0')
+2023-03-02 00:14:42,233 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.189e+02 1.311e+03 1.860e+03 2.743e+03 6.105e+03, threshold=3.721e+03, percent-clipped=28.0
+2023-03-02 00:14:42,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=137036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:14:49,466 INFO [train.py:968] (0/2) Epoch 4, batch 900, giga_loss[loss=0.4125, simple_loss=0.4548, pruned_loss=0.1851, over 27456.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3427, pruned_loss=0.1082, over 5623011.07 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3852, pruned_loss=0.1251, over 1919680.32 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3378, pruned_loss=0.1058, over 5588940.91 frames. ], batch size: 472, lr: 8.32e-03, grad_scale: 4.0
+2023-03-02 00:15:00,783 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:15:03,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:15:26,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:15:32,068 INFO [train.py:968] (0/2) Epoch 4, batch 950, giga_loss[loss=0.3291, simple_loss=0.396, pruned_loss=0.1311, over 28503.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3547, pruned_loss=0.1144, over 5648050.10 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3842, pruned_loss=0.1247, over 1997360.44 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3505, pruned_loss=0.1124, over 5617363.21 frames. ], batch size: 85, lr: 8.32e-03, grad_scale: 4.0
+2023-03-02 00:15:41,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:15:42,853 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-02 00:15:54,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1829, 4.3811, 2.2292, 1.9097], device='cuda:0'), covar=tensor([0.0797, 0.0221, 0.0765, 0.1293], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0455, 0.0308, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:0')
+2023-03-02 00:16:08,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.634e+02 1.157e+03 1.558e+03 2.031e+03 7.810e+03, threshold=3.116e+03, percent-clipped=8.0
+2023-03-02 00:16:15,513 INFO [train.py:968] (0/2) Epoch 4, batch 1000, giga_loss[loss=0.3005, simple_loss=0.3805, pruned_loss=0.1102, over 28645.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3613, pruned_loss=0.1166, over 5659610.61 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3831, pruned_loss=0.1238, over 2074424.93 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3578, pruned_loss=0.1152, over 5631170.44 frames. ], batch size: 336, lr: 8.32e-03, grad_scale: 4.0
+2023-03-02 00:16:56,519 INFO [train.py:968] (0/2) Epoch 4, batch 1050, giga_loss[loss=0.359, simple_loss=0.4084, pruned_loss=0.1548, over 26562.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3643, pruned_loss=0.1165, over 5665194.23 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3823, pruned_loss=0.1231, over 2169645.67 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3614, pruned_loss=0.1155, over 5637050.64 frames. ], batch size: 555, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:17:03,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=137199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:17:35,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.850e+02 9.889e+02 1.203e+03 1.644e+03 4.070e+03, threshold=2.406e+03, percent-clipped=3.0
+2023-03-02 00:17:40,511 INFO [train.py:968] (0/2) Epoch 4, batch 1100, giga_loss[loss=0.2876, simple_loss=0.361, pruned_loss=0.1071, over 28930.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3654, pruned_loss=0.1162, over 5668102.56 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3825, pruned_loss=0.1234, over 2233491.15 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3628, pruned_loss=0.1151, over 5651154.01 frames. ], batch size: 174, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:17:42,962 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:17:45,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:18:07,650 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:18:19,422 INFO [train.py:968] (0/2) Epoch 4, batch 1150, giga_loss[loss=0.3673, simple_loss=0.4209, pruned_loss=0.1569, over 29079.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3687, pruned_loss=0.1188, over 5686280.18 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3826, pruned_loss=0.124, over 2374494.29 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.366, pruned_loss=0.1175, over 5667350.80 frames. ], batch size: 120, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:18:48,919 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5861, 2.2292, 1.5689, 0.7494], device='cuda:0'), covar=tensor([0.1610, 0.0882, 0.1694, 0.1676], device='cuda:0'), in_proj_covar=tensor([0.1289, 0.1230, 0.1311, 0.1102], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 00:18:58,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.978e+02 1.168e+03 1.544e+03 1.978e+03 5.942e+03, threshold=3.089e+03, percent-clipped=10.0
+2023-03-02 00:19:03,682 INFO [train.py:968] (0/2) Epoch 4, batch 1200, giga_loss[loss=0.3034, simple_loss=0.373, pruned_loss=0.1169, over 29015.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3711, pruned_loss=0.1208, over 5679244.60 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3826, pruned_loss=0.1243, over 2462217.62 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3686, pruned_loss=0.1195, over 5658930.94 frames. ], batch size: 155, lr: 8.31e-03, grad_scale: 8.0
+2023-03-02 00:19:48,855 INFO [train.py:968] (0/2) Epoch 4, batch 1250, giga_loss[loss=0.3323, simple_loss=0.4032, pruned_loss=0.1307, over 28846.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3754, pruned_loss=0.1237, over 5681210.39 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3832, pruned_loss=0.1246, over 2513494.97 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3731, pruned_loss=0.1226, over 5662237.41 frames. ], batch size: 199, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:20:06,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:20:14,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4210, 1.8284, 1.6241, 1.6076], device='cuda:0'), covar=tensor([0.1386, 0.1713, 0.1117, 0.1026], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0776, 0.0751, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 00:20:26,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.964e+02 1.154e+03 1.417e+03 1.884e+03 7.270e+03, threshold=2.834e+03, percent-clipped=6.0
+2023-03-02 00:20:32,086 INFO [train.py:968] (0/2) Epoch 4, batch 1300, giga_loss[loss=0.2981, simple_loss=0.3708, pruned_loss=0.1127, over 28882.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3774, pruned_loss=0.1239, over 5681198.98 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3826, pruned_loss=0.1242, over 2636077.50 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3756, pruned_loss=0.1231, over 5667899.03 frames. ], batch size: 60, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:21:10,097 INFO [train.py:968] (0/2) Epoch 4, batch 1350, giga_loss[loss=0.3319, simple_loss=0.3982, pruned_loss=0.1328, over 28836.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.379, pruned_loss=0.1237, over 5697798.44 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3833, pruned_loss=0.1247, over 2716931.28 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3771, pruned_loss=0.1229, over 5682581.15 frames. ], batch size: 199, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:21:44,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.878e+02 1.130e+03 1.527e+03 1.995e+03 4.727e+03, threshold=3.055e+03, percent-clipped=8.0
+2023-03-02 00:21:48,986 INFO [train.py:968] (0/2) Epoch 4, batch 1400, giga_loss[loss=0.3123, simple_loss=0.3802, pruned_loss=0.1223, over 28856.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3806, pruned_loss=0.1242, over 5696236.59 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3828, pruned_loss=0.1244, over 2838326.20 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3792, pruned_loss=0.1236, over 5680063.83 frames. ], batch size: 106, lr: 8.30e-03, grad_scale: 4.0
+2023-03-02 00:21:56,830 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:22:00,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:22:17,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:22:25,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6486, 2.6699, 1.6896, 0.6474], device='cuda:0'), covar=tensor([0.3198, 0.1346, 0.2024, 0.3062], device='cuda:0'), in_proj_covar=tensor([0.1288, 0.1219, 0.1303, 0.1098], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 00:22:25,785 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:22:34,936 INFO [train.py:968] (0/2) Epoch 4, batch 1450, giga_loss[loss=0.3619, simple_loss=0.4139, pruned_loss=0.155, over 28991.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3813, pruned_loss=0.1238, over 5701799.99 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3831, pruned_loss=0.1247, over 2927969.61 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3799, pruned_loss=0.1232, over 5684689.00 frames. ], batch size: 128, lr: 8.30e-03, grad_scale: 4.0
+2023-03-02 00:23:08,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.257e+02 1.053e+03 1.362e+03 1.661e+03 4.260e+03, threshold=2.724e+03, percent-clipped=3.0
+2023-03-02 00:23:12,550 INFO [train.py:968] (0/2) Epoch 4, batch 1500, giga_loss[loss=0.3986, simple_loss=0.4373, pruned_loss=0.18, over 27954.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3795, pruned_loss=0.1212, over 5708400.70 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3831, pruned_loss=0.1244, over 3014409.76 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3784, pruned_loss=0.1209, over 5691075.60 frames. ], batch size: 412, lr: 8.30e-03, grad_scale: 4.0
+2023-03-02 00:23:16,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1888, 1.3991, 1.1623, 1.2155], device='cuda:0'), covar=tensor([0.2319, 0.2098, 0.2046, 0.2107], device='cuda:0'), in_proj_covar=tensor([0.1056, 0.0843, 0.0937, 0.0936], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:23:35,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2041, 2.9836, 2.9176, 1.2404], device='cuda:0'), covar=tensor([0.0699, 0.0590, 0.1113, 0.2380], device='cuda:0'), in_proj_covar=tensor([0.0767, 0.0622, 0.0777, 0.0576], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 00:23:53,328 INFO [train.py:968] (0/2) Epoch 4, batch 1550, giga_loss[loss=0.2961, simple_loss=0.3641, pruned_loss=0.114, over 28738.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.377, pruned_loss=0.1191, over 5710168.79 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3823, pruned_loss=0.1237, over 3085799.09 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3764, pruned_loss=0.119, over 5692162.08 frames. ], batch size: 92, lr: 8.30e-03, grad_scale: 4.0
+2023-03-02 00:24:13,261 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:24:16,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:24:33,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.556e+02 1.124e+03 1.415e+03 2.000e+03 5.669e+03, threshold=2.830e+03, percent-clipped=8.0
+2023-03-02 00:24:37,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3407, 1.3751, 1.1596, 1.3228], device='cuda:0'), covar=tensor([0.1166, 0.1358, 0.1448, 0.1280], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0760, 0.0631, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 00:24:37,494 INFO [train.py:968] (0/2) Epoch 4, batch 1600, giga_loss[loss=0.3555, simple_loss=0.407, pruned_loss=0.152, over 28638.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3778, pruned_loss=0.1211, over 5707729.51 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3821, pruned_loss=0.1235, over 3141754.89 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3773, pruned_loss=0.1211, over 5701977.62 frames. ], batch size: 242, lr: 8.30e-03, grad_scale: 8.0
+2023-03-02 00:24:43,507 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:25:24,029 INFO [train.py:968] (0/2) Epoch 4, batch 1650, giga_loss[loss=0.4106, simple_loss=0.4348, pruned_loss=0.1932, over 27602.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3818, pruned_loss=0.1274, over 5703936.54 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3817, pruned_loss=0.1234, over 3206351.26 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3816, pruned_loss=0.1275, over 5698512.70 frames. ], batch size: 472, lr: 8.30e-03, grad_scale: 4.0
+2023-03-02 00:25:44,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5739, 0.8579, 3.5846, 2.9578], device='cuda:0'), covar=tensor([0.2032, 0.2411, 0.0525, 0.0630], device='cuda:0'), in_proj_covar=tensor([0.0531, 0.0505, 0.0697, 0.0570], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:25:59,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.247e+02 1.271e+03 1.659e+03 2.083e+03 4.710e+03, threshold=3.318e+03, percent-clipped=12.0
+2023-03-02 00:26:03,560 INFO [train.py:968] (0/2) Epoch 4, batch 1700, giga_loss[loss=0.3772, simple_loss=0.4089, pruned_loss=0.1727, over 26535.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3835, pruned_loss=0.13, over 5692048.22 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.382, pruned_loss=0.1237, over 3348874.83 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3832, pruned_loss=0.1302, over 5690914.51 frames. ], batch size: 555, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:26:48,612 INFO [train.py:968] (0/2) Epoch 4, batch 1750, libri_loss[loss=0.3109, simple_loss=0.3805, pruned_loss=0.1206, over 29510.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3826, pruned_loss=0.1302, over 5695763.14 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3815, pruned_loss=0.1231, over 3411869.03 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3827, pruned_loss=0.1309, over 5690035.17 frames. ], batch size: 81, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:27:12,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1459, 1.2503, 1.1498, 0.9834], device='cuda:0'), covar=tensor([0.2106, 0.2035, 0.1850, 0.1756], device='cuda:0'), in_proj_covar=tensor([0.1059, 0.0848, 0.0939, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:27:15,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4613, 3.0754, 1.6620, 1.3699], device='cuda:0'), covar=tensor([0.0849, 0.0346, 0.0749, 0.1359], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0453, 0.0307, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0021, 0.0014, 0.0019], device='cuda:0')
+2023-03-02 00:27:23,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2157, 1.4279, 1.2399, 1.4588], device='cuda:0'), covar=tensor([0.0861, 0.0382, 0.0367, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0217, 0.0140, 0.0143, 0.0257], device='cuda:0'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:0')
+2023-03-02 00:27:26,265 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.817e+02 1.271e+03 1.798e+03 2.483e+03 8.333e+03, threshold=3.595e+03, percent-clipped=8.0
+2023-03-02 00:27:29,803 INFO [train.py:968] (0/2) Epoch 4, batch 1800, giga_loss[loss=0.2992, simple_loss=0.3688, pruned_loss=0.1148, over 28997.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3802, pruned_loss=0.129, over 5709897.99 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3812, pruned_loss=0.1231, over 3473723.77 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3805, pruned_loss=0.1297, over 5700642.25 frames. ], batch size: 136, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:28:13,796 INFO [train.py:968] (0/2) Epoch 4, batch 1850, giga_loss[loss=0.3567, simple_loss=0.4111, pruned_loss=0.1511, over 28330.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3795, pruned_loss=0.1277, over 5711318.32 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3812, pruned_loss=0.123, over 3497789.93 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3796, pruned_loss=0.1283, over 5702451.43 frames. ], batch size: 368, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:28:19,455 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-138000.pt
+2023-03-02 00:28:46,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0653, 1.6458, 1.3466, 1.5406], device='cuda:0'), covar=tensor([0.0591, 0.0696, 0.0873, 0.0955], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0475, 0.0511, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-02 00:28:50,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.857e+02 1.095e+03 1.425e+03 2.063e+03 4.383e+03, threshold=2.849e+03, percent-clipped=3.0
+2023-03-02 00:28:56,495 INFO [train.py:968] (0/2) Epoch 4, batch 1900, giga_loss[loss=0.3758, simple_loss=0.4127, pruned_loss=0.1695, over 26558.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3779, pruned_loss=0.1257, over 5707927.54 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3814, pruned_loss=0.1231, over 3605911.71 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3778, pruned_loss=0.1263, over 5700532.32 frames. ], batch size: 555, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:29:00,627 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5046, 1.5313, 1.2846, 1.7716], device='cuda:0'), covar=tensor([0.2041, 0.2011, 0.1882, 0.2172], device='cuda:0'), in_proj_covar=tensor([0.1058, 0.0845, 0.0936, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:29:44,243 INFO [train.py:968] (0/2) Epoch 4, batch 1950, libri_loss[loss=0.3355, simple_loss=0.4089, pruned_loss=0.1311, over 26368.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3735, pruned_loss=0.1225, over 5689068.41 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3808, pruned_loss=0.1228, over 3649867.17 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3737, pruned_loss=0.1232, over 5690490.26 frames. ], batch size: 136, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:30:26,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.818e+02 9.854e+02 1.197e+03 1.630e+03 7.989e+03, threshold=2.394e+03, percent-clipped=6.0
+2023-03-02 00:30:30,458 INFO [train.py:968] (0/2) Epoch 4, batch 2000, giga_loss[loss=0.2532, simple_loss=0.3209, pruned_loss=0.09273, over 28778.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3673, pruned_loss=0.1185, over 5685849.34 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3805, pruned_loss=0.1225, over 3703045.75 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3673, pruned_loss=0.1192, over 5684552.29 frames. ], batch size: 119, lr: 8.29e-03, grad_scale: 8.0
+2023-03-02 00:31:12,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0520, 1.1949, 1.2667, 1.2552], device='cuda:0'), covar=tensor([0.1188, 0.1155, 0.1619, 0.1092], device='cuda:0'), in_proj_covar=tensor([0.0454, 0.0769, 0.0643, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 00:31:12,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4226, 1.0916, 2.9127, 2.7521], device='cuda:0'), covar=tensor([0.1571, 0.1903, 0.0438, 0.0597], device='cuda:0'), in_proj_covar=tensor([0.0535, 0.0506, 0.0694, 0.0569], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:31:17,919 INFO [train.py:968] (0/2) Epoch 4, batch 2050, giga_loss[loss=0.2547, simple_loss=0.3193, pruned_loss=0.09499, over 28337.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3605, pruned_loss=0.1148, over 5680644.84 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3808, pruned_loss=0.1226, over 3736095.15 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3602, pruned_loss=0.1152, over 5676556.21 frames. ], batch size: 65, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:31:47,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3351, 1.3792, 1.2343, 1.4086], device='cuda:0'), covar=tensor([0.2088, 0.1990, 0.1804, 0.2093], device='cuda:0'), in_proj_covar=tensor([0.1071, 0.0855, 0.0952, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:31:50,990 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3126, 1.3714, 1.4348, 1.4090], device='cuda:0'), covar=tensor([0.1198, 0.1406, 0.1571, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0769, 0.0643, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 00:32:03,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.297e+02 9.795e+02 1.184e+03 1.438e+03 4.411e+03, threshold=2.367e+03, percent-clipped=8.0
+2023-03-02 00:32:08,002 INFO [train.py:968] (0/2) Epoch 4, batch 2100, giga_loss[loss=0.2652, simple_loss=0.3471, pruned_loss=0.09165, over 28873.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3576, pruned_loss=0.1128, over 5685268.22 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.381, pruned_loss=0.1226, over 3778965.12 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3568, pruned_loss=0.113, over 5678205.11 frames. ], batch size: 136, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:32:37,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4075, 1.7933, 1.2515, 0.5887], device='cuda:0'), covar=tensor([0.2289, 0.1166, 0.1401, 0.2350], device='cuda:0'), in_proj_covar=tensor([0.1283, 0.1212, 0.1292, 0.1100], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 00:32:40,538 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=138285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:32:42,753 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=138288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:32:47,286 INFO [train.py:968] (0/2) Epoch 4, batch 2150, giga_loss[loss=0.2927, simple_loss=0.3577, pruned_loss=0.1139, over 28768.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3584, pruned_loss=0.1131, over 5683343.34 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3811, pruned_loss=0.1226, over 3810586.69 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3573, pruned_loss=0.1131, over 5683274.71 frames. ], batch size: 284, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:33:23,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.029e+02 1.042e+03 1.190e+03 1.795e+03 6.289e+03, threshold=2.380e+03, percent-clipped=14.0
+2023-03-02 00:33:28,971 INFO [train.py:968] (0/2) Epoch 4, batch 2200, giga_loss[loss=0.2495, simple_loss=0.3223, pruned_loss=0.08839, over 28421.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3576, pruned_loss=0.1124, over 5689682.72 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3811, pruned_loss=0.1223, over 3858839.10 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3563, pruned_loss=0.1124, over 5688001.46 frames. ], batch size: 60, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:33:38,109 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-02 00:34:09,640 INFO [train.py:968] (0/2) Epoch 4, batch 2250, libri_loss[loss=0.3003, simple_loss=0.3779, pruned_loss=0.1113, over 29525.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3571, pruned_loss=0.1122, over 5699031.77 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.382, pruned_loss=0.1227, over 3946050.16 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3544, pruned_loss=0.1115, over 5691676.49 frames. ], batch size: 80, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:34:18,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5724, 1.8858, 1.5093, 1.5763], device='cuda:0'), covar=tensor([0.0749, 0.0300, 0.0324, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0215, 0.0138, 0.0142, 0.0255], device='cuda:0'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:0')
+2023-03-02 00:34:48,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.995e+02 1.108e+03 1.554e+03 1.919e+03 3.989e+03, threshold=3.109e+03, percent-clipped=16.0
+2023-03-02 00:34:52,323 INFO [train.py:968] (0/2) Epoch 4, batch 2300, giga_loss[loss=0.263, simple_loss=0.3301, pruned_loss=0.09799, over 28770.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3548, pruned_loss=0.1118, over 5703262.67 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3824, pruned_loss=0.1229, over 3975206.80 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.352, pruned_loss=0.1109, over 5694848.12 frames. ], batch size: 119, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:35:30,091 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9617, 2.1615, 1.0702, 1.0206], device='cuda:0'), covar=tensor([0.1261, 0.0519, 0.1081, 0.1796], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0449, 0.0304, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:0')
+2023-03-02 00:35:32,493 INFO [train.py:968] (0/2) Epoch 4, batch 2350, giga_loss[loss=0.2528, simple_loss=0.3173, pruned_loss=0.09418, over 28345.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3534, pruned_loss=0.1108, over 5717304.20 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3842, pruned_loss=0.1237, over 4040998.72 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3491, pruned_loss=0.1092, over 5704880.94 frames. ], batch size: 65, lr: 8.28e-03, grad_scale: 4.0
+2023-03-02 00:36:11,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 4.835e+02 9.008e+02 1.177e+03 1.500e+03 3.608e+03, threshold=2.355e+03, percent-clipped=3.0
+2023-03-02 00:36:15,921 INFO [train.py:968] (0/2) Epoch 4, batch 2400, giga_loss[loss=0.3029, simple_loss=0.3617, pruned_loss=0.1221, over 28714.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3505, pruned_loss=0.1094, over 5723063.88 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3843, pruned_loss=0.1237, over 4068498.70 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3465, pruned_loss=0.108, over 5711061.43 frames. ], batch size: 284, lr: 8.27e-03, grad_scale: 8.0
+2023-03-02 00:36:52,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2109, 1.3766, 1.0719, 1.3846], device='cuda:0'), covar=tensor([0.0819, 0.0439, 0.0391, 0.0903], device='cuda:0'), in_proj_covar=tensor([0.0215, 0.0139, 0.0143, 0.0255], device='cuda:0'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:0')
+2023-03-02 00:36:54,670 INFO [train.py:968] (0/2) Epoch 4, batch 2450, giga_loss[loss=0.2657, simple_loss=0.3331, pruned_loss=0.09913, over 28733.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3473, pruned_loss=0.1077, over 5728578.82 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3847, pruned_loss=0.1236, over 4112907.32 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.343, pruned_loss=0.1062, over 5715918.98 frames. ], batch size: 262, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:37:29,918 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.047e+02 9.731e+02 1.225e+03 1.766e+03 5.880e+03, threshold=2.450e+03, percent-clipped=10.0
+2023-03-02 00:37:33,952 INFO [train.py:968] (0/2) Epoch 4, batch 2500, giga_loss[loss=0.2419, simple_loss=0.3135, pruned_loss=0.08513, over 28939.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3444, pruned_loss=0.1061, over 5731300.71 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3852, pruned_loss=0.1237, over 4148562.82 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.34, pruned_loss=0.1045, over 5718187.82 frames. ], batch size: 112, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:37:47,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:37:49,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=138663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:37:49,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:38:13,717 INFO [train.py:968] (0/2) Epoch 4, batch 2550, giga_loss[loss=0.2585, simple_loss=0.3294, pruned_loss=0.0938, over 29065.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3425, pruned_loss=0.1048, over 5717154.40 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3858, pruned_loss=0.1238, over 4180670.07 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3377, pruned_loss=0.1031, over 5713293.53 frames. ], batch size: 164, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:38:33,315 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=138715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:38:53,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.768e+02 8.838e+02 1.121e+03 1.694e+03 1.176e+04, threshold=2.243e+03, percent-clipped=12.0
+2023-03-02 00:38:55,998 INFO [train.py:968] (0/2) Epoch 4, batch 2600, giga_loss[loss=0.2614, simple_loss=0.3268, pruned_loss=0.09798, over 28715.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3402, pruned_loss=0.1036, over 5718367.60 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3862, pruned_loss=0.1239, over 4188970.72 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.336, pruned_loss=0.1021, over 5714809.07 frames. ], batch size: 119, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:38:59,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1487, 3.9502, 3.8760, 1.8448], device='cuda:0'), covar=tensor([0.0483, 0.0376, 0.0689, 0.2031], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0635, 0.0778, 0.0577], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 00:39:34,002 INFO [train.py:968] (0/2) Epoch 4, batch 2650, giga_loss[loss=0.2607, simple_loss=0.3268, pruned_loss=0.09733, over 28737.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3396, pruned_loss=0.1027, over 5723608.61 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3871, pruned_loss=0.1241, over 4244899.35 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3343, pruned_loss=0.1008, over 5718115.31 frames. ], batch size: 99, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:39:43,067 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:39:44,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:39:45,019 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:39:47,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:40:11,997 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:40:14,082 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:40:15,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.278e+02 9.446e+02 1.290e+03 1.787e+03 4.841e+03, threshold=2.579e+03, percent-clipped=12.0
+2023-03-02 00:40:18,083 INFO [train.py:968] (0/2) Epoch 4, batch 2700, giga_loss[loss=0.3619, simple_loss=0.4109, pruned_loss=0.1565, over 27556.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3424, pruned_loss=0.1047, over 5710925.43 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3877, pruned_loss=0.1244, over 4265380.28 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3371, pruned_loss=0.1027, over 5708204.10 frames. ], batch size: 472, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:40:37,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1469, 1.3330, 1.1736, 1.0170], device='cuda:0'), covar=tensor([0.1933, 0.1794, 0.1676, 0.1521], device='cuda:0'), in_proj_covar=tensor([0.1063, 0.0851, 0.0947, 0.0943], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:41:01,494 INFO [train.py:968] (0/2) Epoch 4, batch 2750, giga_loss[loss=0.2972, simple_loss=0.3596, pruned_loss=0.1174, over 28986.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3494, pruned_loss=0.1093, over 5717673.71 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3877, pruned_loss=0.1242, over 4329814.69 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3435, pruned_loss=0.1072, over 5708670.69 frames. ], batch size: 136, lr: 8.26e-03, grad_scale: 4.0
+2023-03-02 00:41:05,372 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-02 00:41:09,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3433, 1.6422, 1.2999, 1.4967], device='cuda:0'), covar=tensor([0.0831, 0.0336, 0.0352, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0214, 0.0138, 0.0142, 0.0253], device='cuda:0'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:0')
+2023-03-02 00:41:44,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.881e+02 1.191e+03 1.720e+03 2.392e+03 5.275e+03, threshold=3.439e+03, percent-clipped=21.0
+2023-03-02 00:41:46,427 INFO [train.py:968] (0/2) Epoch 4, batch 2800, giga_loss[loss=0.315, simple_loss=0.3828, pruned_loss=0.1237, over 28972.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3571, pruned_loss=0.1148, over 5707656.38 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3876, pruned_loss=0.1242, over 4374486.19 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3515, pruned_loss=0.1128, over 5696844.04 frames. ], batch size: 164, lr: 8.26e-03, grad_scale: 8.0
+2023-03-02 00:42:32,735 INFO [train.py:968] (0/2) Epoch 4, batch 2850, giga_loss[loss=0.3064, simple_loss=0.381, pruned_loss=0.1159, over 28520.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3661, pruned_loss=0.1212, over 5697469.42 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.387, pruned_loss=0.124, over 4397372.89 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3617, pruned_loss=0.1196, over 5686421.60 frames. ], batch size: 336, lr: 8.26e-03, grad_scale: 8.0
+2023-03-02 00:43:18,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:43:19,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.190e+02 1.152e+03 1.538e+03 1.904e+03 5.476e+03, threshold=3.076e+03, percent-clipped=3.0
+2023-03-02 00:43:24,979 INFO [train.py:968] (0/2) Epoch 4, batch 2900, giga_loss[loss=0.3429, simple_loss=0.384, pruned_loss=0.1509, over 23454.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3724, pruned_loss=0.1248, over 5673721.53 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3869, pruned_loss=0.1239, over 4404943.02 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.369, pruned_loss=0.1236, over 5663995.09 frames. ], batch size: 705, lr: 8.26e-03, grad_scale: 8.0
+2023-03-02 00:43:51,239 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:43:55,028 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:44:03,851 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:44:03,891 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:44:05,544 INFO [train.py:968] (0/2) Epoch 4, batch 2950, giga_loss[loss=0.3151, simple_loss=0.389, pruned_loss=0.1206, over 28913.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3772, pruned_loss=0.1264, over 5682737.42 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3867, pruned_loss=0.1239, over 4432127.21 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3745, pruned_loss=0.1256, over 5673363.06 frames. ], batch size: 213, lr: 8.26e-03, grad_scale: 8.0
+2023-03-02 00:44:51,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.836e+02 1.057e+03 1.329e+03 1.943e+03 5.854e+03, threshold=2.658e+03, percent-clipped=11.0
+2023-03-02 00:44:53,492 INFO [train.py:968] (0/2) Epoch 4, batch 3000, giga_loss[loss=0.3029, simple_loss=0.3695, pruned_loss=0.1181, over 28441.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3824, pruned_loss=0.1299, over 5686652.58 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3859, pruned_loss=0.1235, over 4465364.46 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3807, pruned_loss=0.1296, over 5677188.29 frames. ], batch size: 60, lr: 8.26e-03, grad_scale: 4.0
+2023-03-02 00:44:53,497 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 00:45:01,935 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2596, simple_loss=0.3561, pruned_loss=0.08156, over 944034.00 frames. 
+2023-03-02 00:45:01,936 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 00:45:05,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7795, 1.0752, 3.2154, 2.9528], device='cuda:0'), covar=tensor([0.1540, 0.2069, 0.0402, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0538, 0.0502, 0.0697, 0.0564], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:45:32,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:45:34,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:45:41,628 INFO [train.py:968] (0/2) Epoch 4, batch 3050, libri_loss[loss=0.3416, simple_loss=0.4027, pruned_loss=0.1402, over 29693.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3778, pruned_loss=0.1267, over 5687158.91 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3861, pruned_loss=0.1237, over 4515196.08 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3761, pruned_loss=0.1266, over 5672442.44 frames. ], batch size: 91, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:45:55,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2056, 1.2609, 1.1117, 0.9679], device='cuda:0'), covar=tensor([0.0704, 0.0526, 0.1006, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0473, 0.0521, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 00:45:59,951 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:46:15,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:46:20,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:46:23,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.077e+02 1.067e+03 1.370e+03 1.916e+03 7.661e+03, threshold=2.741e+03, percent-clipped=9.0
+2023-03-02 00:46:25,334 INFO [train.py:968] (0/2) Epoch 4, batch 3100, giga_loss[loss=0.2762, simple_loss=0.352, pruned_loss=0.1002, over 28385.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.374, pruned_loss=0.1235, over 5694715.48 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3868, pruned_loss=0.1244, over 4563063.13 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3717, pruned_loss=0.1228, over 5675700.66 frames. ], batch size: 77, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:46:39,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7926, 0.9849, 3.8530, 3.2526], device='cuda:0'), covar=tensor([0.1721, 0.2291, 0.0341, 0.0521], device='cuda:0'), in_proj_covar=tensor([0.0528, 0.0494, 0.0687, 0.0555], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 00:46:44,550 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:47:09,889 INFO [train.py:968] (0/2) Epoch 4, batch 3150, giga_loss[loss=0.319, simple_loss=0.3881, pruned_loss=0.125, over 28643.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3731, pruned_loss=0.1222, over 5685730.95 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3868, pruned_loss=0.1244, over 4576467.85 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3711, pruned_loss=0.1216, over 5668909.33 frames. ], batch size: 262, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:47:17,880 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:47:51,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.836e+02 1.127e+03 1.412e+03 1.736e+03 2.953e+03, threshold=2.825e+03, percent-clipped=1.0
+2023-03-02 00:47:53,298 INFO [train.py:968] (0/2) Epoch 4, batch 3200, giga_loss[loss=0.3275, simple_loss=0.3922, pruned_loss=0.1314, over 28950.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.374, pruned_loss=0.1226, over 5684909.74 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3865, pruned_loss=0.1243, over 4596153.03 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3725, pruned_loss=0.1222, over 5668792.68 frames. ], batch size: 213, lr: 8.25e-03, grad_scale: 8.0
+2023-03-02 00:47:55,032 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-02 00:48:33,762 INFO [train.py:968] (0/2) Epoch 4, batch 3250, giga_loss[loss=0.3349, simple_loss=0.3875, pruned_loss=0.1411, over 28925.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3765, pruned_loss=0.124, over 5678784.21 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3869, pruned_loss=0.1246, over 4614257.03 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3749, pruned_loss=0.1234, over 5673140.27 frames. ], batch size: 112, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:49:18,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.087e+02 1.218e+03 1.534e+03 2.124e+03 5.783e+03, threshold=3.069e+03, percent-clipped=13.0
+2023-03-02 00:49:19,052 INFO [train.py:968] (0/2) Epoch 4, batch 3300, giga_loss[loss=0.3068, simple_loss=0.3725, pruned_loss=0.1205, over 28928.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3797, pruned_loss=0.1266, over 5688285.73 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3865, pruned_loss=0.1244, over 4631055.20 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3785, pruned_loss=0.1263, over 5683118.43 frames. ], batch size: 227, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:49:25,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:49:29,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:49:39,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:50:03,676 INFO [train.py:968] (0/2) Epoch 4, batch 3350, giga_loss[loss=0.3256, simple_loss=0.3925, pruned_loss=0.1293, over 28946.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3808, pruned_loss=0.1277, over 5688495.23 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3864, pruned_loss=0.1242, over 4649703.42 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3798, pruned_loss=0.1276, over 5682038.16 frames. ], batch size: 145, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:50:27,091 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3379, 1.4424, 1.3556, 1.4544], device='cuda:0'), covar=tensor([0.0861, 0.1040, 0.1210, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0757, 0.0627, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 00:50:45,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.210e+02 1.083e+03 1.449e+03 2.082e+03 6.742e+03, threshold=2.899e+03, percent-clipped=12.0
+2023-03-02 00:50:46,690 INFO [train.py:968] (0/2) Epoch 4, batch 3400, giga_loss[loss=0.3491, simple_loss=0.3982, pruned_loss=0.15, over 28621.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3819, pruned_loss=0.1292, over 5688219.68 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3864, pruned_loss=0.1243, over 4668440.32 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3811, pruned_loss=0.1291, over 5680201.58 frames. ], batch size: 85, lr: 8.24e-03, grad_scale: 4.0
+2023-03-02 00:51:24,914 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:25,247 INFO [train.py:968] (0/2) Epoch 4, batch 3450, giga_loss[loss=0.3205, simple_loss=0.3873, pruned_loss=0.1269, over 28803.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3828, pruned_loss=0.1297, over 5687134.80 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3869, pruned_loss=0.1245, over 4724841.33 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3816, pruned_loss=0.1298, over 5674497.49 frames. ], batch size: 242, lr: 8.24e-03, grad_scale: 4.0
+2023-03-02 00:51:27,876 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:29,130 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:32,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:39,730 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:42,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:51,453 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:54,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:57,497 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:52:04,835 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:52:05,829 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.305e+02 1.127e+03 1.479e+03 1.757e+03 8.310e+03, threshold=2.958e+03, percent-clipped=5.0
+2023-03-02 00:52:06,457 INFO [train.py:968] (0/2) Epoch 4, batch 3500, giga_loss[loss=0.2986, simple_loss=0.3741, pruned_loss=0.1115, over 29086.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3835, pruned_loss=0.1289, over 5688434.22 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3874, pruned_loss=0.1249, over 4750869.62 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.382, pruned_loss=0.1288, over 5680018.41 frames. ], batch size: 155, lr: 8.24e-03, grad_scale: 4.0
+2023-03-02 00:52:15,694 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-02 00:52:21,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2792, 1.4159, 1.2320, 1.2986], device='cuda:0'), covar=tensor([0.0864, 0.0376, 0.0358, 0.0948], device='cuda:0'), in_proj_covar=tensor([0.0216, 0.0140, 0.0142, 0.0257], device='cuda:0'), out_proj_covar=tensor([0.0046, 0.0035, 0.0030, 0.0052], device='cuda:0')
+2023-03-02 00:52:29,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:52:36,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:52:39,732 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:52:48,547 INFO [train.py:968] (0/2) Epoch 4, batch 3550, giga_loss[loss=0.2882, simple_loss=0.3651, pruned_loss=0.1057, over 28801.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.383, pruned_loss=0.1274, over 5693959.91 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3872, pruned_loss=0.1247, over 4772444.99 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3819, pruned_loss=0.1275, over 5684553.66 frames. ], batch size: 119, lr: 8.24e-03, grad_scale: 4.0
+2023-03-02 00:53:30,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.051e+02 1.067e+03 1.307e+03 1.890e+03 9.933e+03, threshold=2.614e+03, percent-clipped=8.0
+2023-03-02 00:53:31,585 INFO [train.py:968] (0/2) Epoch 4, batch 3600, giga_loss[loss=0.2856, simple_loss=0.3603, pruned_loss=0.1055, over 28436.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3839, pruned_loss=0.1277, over 5697912.66 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3875, pruned_loss=0.125, over 4804522.29 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3828, pruned_loss=0.1276, over 5685874.99 frames. ], batch size: 60, lr: 8.24e-03, grad_scale: 8.0
+2023-03-02 00:54:09,026 INFO [train.py:968] (0/2) Epoch 4, batch 3650, giga_loss[loss=0.2829, simple_loss=0.3512, pruned_loss=0.1074, over 28692.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3811, pruned_loss=0.126, over 5705604.73 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3865, pruned_loss=0.1245, over 4837520.65 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3808, pruned_loss=0.1264, over 5690051.35 frames. ], batch size: 92, lr: 8.24e-03, grad_scale: 8.0
+2023-03-02 00:54:33,611 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:54:37,308 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:54:52,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.452e+02 1.066e+03 1.391e+03 1.730e+03 3.155e+03, threshold=2.782e+03, percent-clipped=8.0
+2023-03-02 00:54:54,355 INFO [train.py:968] (0/2) Epoch 4, batch 3700, giga_loss[loss=0.2537, simple_loss=0.3343, pruned_loss=0.08654, over 29070.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3778, pruned_loss=0.1242, over 5706558.45 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3865, pruned_loss=0.1246, over 4858661.39 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3774, pruned_loss=0.1245, over 5690599.34 frames. ], batch size: 155, lr: 8.24e-03, grad_scale: 8.0
+2023-03-02 00:55:02,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:55:32,324 INFO [train.py:968] (0/2) Epoch 4, batch 3750, libri_loss[loss=0.2708, simple_loss=0.3423, pruned_loss=0.09962, over 29496.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3745, pruned_loss=0.1217, over 5708396.48 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.386, pruned_loss=0.1242, over 4881201.64 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3743, pruned_loss=0.1222, over 5694238.56 frames. ], batch size: 70, lr: 8.23e-03, grad_scale: 8.0
+2023-03-02 00:56:15,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.998e+02 9.987e+02 1.211e+03 1.471e+03 3.707e+03, threshold=2.421e+03, percent-clipped=1.0
+2023-03-02 00:56:15,808 INFO [train.py:968] (0/2) Epoch 4, batch 3800, giga_loss[loss=0.3343, simple_loss=0.3885, pruned_loss=0.1401, over 28399.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3759, pruned_loss=0.1231, over 5707083.15 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3856, pruned_loss=0.1241, over 4891664.48 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3759, pruned_loss=0.1236, over 5693920.35 frames. ], batch size: 65, lr: 8.23e-03, grad_scale: 8.0
+2023-03-02 00:56:22,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3136, 1.9489, 1.5107, 0.5746], device='cuda:0'), covar=tensor([0.1717, 0.0930, 0.1670, 0.2097], device='cuda:0'), in_proj_covar=tensor([0.1261, 0.1184, 0.1276, 0.1094], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 00:56:25,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5238, 1.4672, 1.4509, 1.4963], device='cuda:0'), covar=tensor([0.0888, 0.1226, 0.1303, 0.1101], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0761, 0.0637, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 00:56:34,155 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:56:41,381 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:56:54,355 INFO [train.py:968] (0/2) Epoch 4, batch 3850, giga_loss[loss=0.2881, simple_loss=0.3639, pruned_loss=0.1061, over 28528.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3759, pruned_loss=0.1228, over 5714633.18 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3849, pruned_loss=0.1236, over 4935061.95 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3762, pruned_loss=0.1235, over 5697240.91 frames. ], batch size: 60, lr: 8.23e-03, grad_scale: 4.0
+2023-03-02 00:57:00,318 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-140000.pt
+2023-03-02 00:57:02,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 00:57:05,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:57:33,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2859, 1.7657, 1.3201, 1.5623], device='cuda:0'), covar=tensor([0.0843, 0.0316, 0.0348, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0214, 0.0137, 0.0141, 0.0255], device='cuda:0'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:0')
+2023-03-02 00:57:34,688 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.494e+02 1.146e+03 1.573e+03 2.285e+03 1.051e+04, threshold=3.146e+03, percent-clipped=21.0
+2023-03-02 00:57:34,700 INFO [train.py:968] (0/2) Epoch 4, batch 3900, giga_loss[loss=0.2827, simple_loss=0.3572, pruned_loss=0.1042, over 28915.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3762, pruned_loss=0.1219, over 5715871.02 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.385, pruned_loss=0.1238, over 4961761.33 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.376, pruned_loss=0.1223, over 5698777.31 frames. ], batch size: 112, lr: 8.23e-03, grad_scale: 4.0
+2023-03-02 00:57:35,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:57:45,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:58:16,595 INFO [train.py:968] (0/2) Epoch 4, batch 3950, giga_loss[loss=0.2788, simple_loss=0.3567, pruned_loss=0.1005, over 28934.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3761, pruned_loss=0.1214, over 5720878.95 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3846, pruned_loss=0.1235, over 4975951.54 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3762, pruned_loss=0.1219, over 5704924.68 frames. ], batch size: 164, lr: 8.23e-03, grad_scale: 4.0
+2023-03-02 00:58:20,811 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-02 00:58:55,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.259e+02 1.069e+03 1.322e+03 1.699e+03 3.851e+03, threshold=2.643e+03, percent-clipped=1.0
+2023-03-02 00:58:55,421 INFO [train.py:968] (0/2) Epoch 4, batch 4000, giga_loss[loss=0.337, simple_loss=0.3928, pruned_loss=0.1407, over 28243.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3761, pruned_loss=0.1217, over 5711655.19 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3848, pruned_loss=0.1236, over 5010007.15 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3756, pruned_loss=0.1219, over 5697139.17 frames. ], batch size: 368, lr: 8.23e-03, grad_scale: 8.0
+2023-03-02 00:59:02,800 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:04,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:13,114 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:27,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:28,619 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:33,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:35,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:37,464 INFO [train.py:968] (0/2) Epoch 4, batch 4050, giga_loss[loss=0.3094, simple_loss=0.3776, pruned_loss=0.1206, over 28317.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3748, pruned_loss=0.1213, over 5719175.26 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3843, pruned_loss=0.1233, over 5023465.24 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3746, pruned_loss=0.1216, over 5705218.97 frames. ], batch size: 60, lr: 8.23e-03, grad_scale: 8.0
+2023-03-02 00:59:37,898 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.16 vs. limit=2.0
+2023-03-02 00:59:40,667 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:43,048 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:43,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:45,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:58,771 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:00:08,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:00:18,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.237e+02 1.095e+03 1.465e+03 2.075e+03 5.213e+03, threshold=2.930e+03, percent-clipped=12.0
+2023-03-02 01:00:18,177 INFO [train.py:968] (0/2) Epoch 4, batch 4100, giga_loss[loss=0.2657, simple_loss=0.3367, pruned_loss=0.0973, over 29099.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3724, pruned_loss=0.1203, over 5720966.69 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3841, pruned_loss=0.1232, over 5048787.07 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3722, pruned_loss=0.1206, over 5705599.92 frames. ], batch size: 128, lr: 8.22e-03, grad_scale: 8.0
+2023-03-02 01:00:18,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:00:20,961 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 01:00:32,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5257, 2.8934, 1.5512, 1.3721], device='cuda:0'), covar=tensor([0.0793, 0.0292, 0.0765, 0.1292], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0441, 0.0303, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:0')
+2023-03-02 01:00:56,789 INFO [train.py:968] (0/2) Epoch 4, batch 4150, giga_loss[loss=0.2586, simple_loss=0.3362, pruned_loss=0.09055, over 28627.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3692, pruned_loss=0.1184, over 5724310.13 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3835, pruned_loss=0.1228, over 5074396.47 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.369, pruned_loss=0.1188, over 5706939.89 frames. ], batch size: 60, lr: 8.22e-03, grad_scale: 4.0
+2023-03-02 01:01:04,861 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 01:01:35,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:01:37,972 INFO [train.py:968] (0/2) Epoch 4, batch 4200, giga_loss[loss=0.3098, simple_loss=0.3703, pruned_loss=0.1247, over 29089.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3699, pruned_loss=0.1196, over 5723084.30 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3837, pruned_loss=0.123, over 5082721.81 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3695, pruned_loss=0.1198, over 5707873.66 frames. ], batch size: 155, lr: 8.22e-03, grad_scale: 4.0
+2023-03-02 01:01:38,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.688e+02 1.068e+03 1.364e+03 1.810e+03 9.254e+03, threshold=2.728e+03, percent-clipped=9.0
+2023-03-02 01:01:43,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:02:17,965 INFO [train.py:968] (0/2) Epoch 4, batch 4250, giga_loss[loss=0.2709, simple_loss=0.3387, pruned_loss=0.1015, over 28490.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3695, pruned_loss=0.12, over 5717903.38 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3838, pruned_loss=0.123, over 5106230.44 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3686, pruned_loss=0.12, over 5707262.23 frames. ], batch size: 71, lr: 8.22e-03, grad_scale: 4.0
+2023-03-02 01:02:59,426 INFO [train.py:968] (0/2) Epoch 4, batch 4300, giga_loss[loss=0.2882, simple_loss=0.3472, pruned_loss=0.1146, over 28733.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3684, pruned_loss=0.1203, over 5716787.81 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3832, pruned_loss=0.1229, over 5133078.73 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3676, pruned_loss=0.1203, over 5702899.73 frames. ], batch size: 92, lr: 8.22e-03, grad_scale: 4.0
+2023-03-02 01:03:00,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.195e+02 1.101e+03 1.399e+03 1.985e+03 5.161e+03, threshold=2.798e+03, percent-clipped=13.0
+2023-03-02 01:03:27,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7662, 1.1592, 5.3745, 3.6454], device='cuda:0'), covar=tensor([0.1872, 0.2683, 0.0398, 0.0645], device='cuda:0'), in_proj_covar=tensor([0.0533, 0.0502, 0.0698, 0.0574], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 01:03:30,254 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:32,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:38,795 INFO [train.py:968] (0/2) Epoch 4, batch 4350, giga_loss[loss=0.2992, simple_loss=0.3633, pruned_loss=0.1176, over 28738.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3658, pruned_loss=0.1186, over 5719984.92 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3831, pruned_loss=0.1227, over 5166159.19 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3645, pruned_loss=0.1186, over 5704588.74 frames. ], batch size: 262, lr: 8.22e-03, grad_scale: 4.0
+2023-03-02 01:03:39,108 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:40,955 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:41,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:46,012 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:54,704 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:03,359 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:12,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:16,246 INFO [train.py:968] (0/2) Epoch 4, batch 4400, giga_loss[loss=0.3225, simple_loss=0.3756, pruned_loss=0.1347, over 28739.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3631, pruned_loss=0.1174, over 5712605.52 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3828, pruned_loss=0.1225, over 5174886.58 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3619, pruned_loss=0.1175, over 5704854.13 frames. ], batch size: 99, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:04:17,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.789e+02 1.054e+03 1.239e+03 1.816e+03 4.507e+03, threshold=2.478e+03, percent-clipped=5.0
+2023-03-02 01:04:29,323 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:40,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:42,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:58,530 INFO [train.py:968] (0/2) Epoch 4, batch 4450, libri_loss[loss=0.3446, simple_loss=0.4045, pruned_loss=0.1423, over 29544.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.364, pruned_loss=0.1179, over 5707362.11 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.383, pruned_loss=0.1227, over 5179820.14 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3626, pruned_loss=0.1177, over 5705996.26 frames. ], batch size: 83, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:05:22,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:05:26,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5431, 1.5019, 1.2252, 1.3809], device='cuda:0'), covar=tensor([0.0622, 0.0559, 0.0972, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0469, 0.0515, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 01:05:44,898 INFO [train.py:968] (0/2) Epoch 4, batch 4500, giga_loss[loss=0.2756, simple_loss=0.3421, pruned_loss=0.1045, over 28633.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3648, pruned_loss=0.1181, over 5705331.63 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3832, pruned_loss=0.1229, over 5185060.06 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3634, pruned_loss=0.1177, over 5704302.10 frames. ], batch size: 92, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:05:46,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.440e+02 9.628e+02 1.253e+03 1.515e+03 3.980e+03, threshold=2.506e+03, percent-clipped=5.0
+2023-03-02 01:05:53,315 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:16,642 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:18,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:26,144 INFO [train.py:968] (0/2) Epoch 4, batch 4550, giga_loss[loss=0.2832, simple_loss=0.3597, pruned_loss=0.1033, over 28992.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3674, pruned_loss=0.1185, over 5712319.00 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3838, pruned_loss=0.1232, over 5196083.33 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3653, pruned_loss=0.1179, over 5712905.26 frames. ], batch size: 155, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:06:32,605 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:35,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:41,705 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:46,012 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:47,177 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:47,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:49,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:59,915 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:10,552 INFO [train.py:968] (0/2) Epoch 4, batch 4600, giga_loss[loss=0.2882, simple_loss=0.3626, pruned_loss=0.1069, over 28869.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3691, pruned_loss=0.1191, over 5704238.58 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3837, pruned_loss=0.1232, over 5202974.64 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3674, pruned_loss=0.1186, over 5703136.85 frames. ], batch size: 145, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:07:11,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.115e+02 9.680e+02 1.145e+03 1.468e+03 3.564e+03, threshold=2.291e+03, percent-clipped=3.0
+2023-03-02 01:07:13,733 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:15,305 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:25,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:27,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:44,399 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:54,226 INFO [train.py:968] (0/2) Epoch 4, batch 4650, libri_loss[loss=0.3081, simple_loss=0.3796, pruned_loss=0.1183, over 29515.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3689, pruned_loss=0.1183, over 5696831.99 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3838, pruned_loss=0.1234, over 5213977.51 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3672, pruned_loss=0.1177, over 5696714.58 frames. ], batch size: 84, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:07:54,487 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:08:30,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3345, 3.4953, 1.3236, 1.4139], device='cuda:0'), covar=tensor([0.1054, 0.0408, 0.1077, 0.1554], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0447, 0.0308, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:0')
+2023-03-02 01:08:30,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1336, 1.3237, 1.1156, 1.1864], device='cuda:0'), covar=tensor([0.2027, 0.1914, 0.1907, 0.1946], device='cuda:0'), in_proj_covar=tensor([0.1046, 0.0833, 0.0932, 0.0925], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:0')
+2023-03-02 01:08:34,479 INFO [train.py:968] (0/2) Epoch 4, batch 4700, giga_loss[loss=0.277, simple_loss=0.3547, pruned_loss=0.09962, over 28954.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3687, pruned_loss=0.1178, over 5695660.12 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3841, pruned_loss=0.1237, over 5224568.72 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3668, pruned_loss=0.1169, over 5694628.17 frames. ], batch size: 174, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:08:35,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.919e+02 1.092e+03 1.477e+03 1.910e+03 5.485e+03, threshold=2.953e+03, percent-clipped=15.0
+2023-03-02 01:08:57,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:09:02,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:09:06,109 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:09:14,476 INFO [train.py:968] (0/2) Epoch 4, batch 4750, giga_loss[loss=0.3228, simple_loss=0.3786, pruned_loss=0.1335, over 28873.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3715, pruned_loss=0.1201, over 5701380.80 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3845, pruned_loss=0.124, over 5245999.83 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3691, pruned_loss=0.1189, over 5699584.94 frames. ], batch size: 145, lr: 8.20e-03, grad_scale: 4.0
+2023-03-02 01:09:21,420 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:09:55,943 INFO [train.py:968] (0/2) Epoch 4, batch 4800, giga_loss[loss=0.3239, simple_loss=0.3867, pruned_loss=0.1306, over 27985.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3734, pruned_loss=0.1217, over 5700000.51 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3853, pruned_loss=0.1248, over 5254420.40 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3705, pruned_loss=0.12, over 5703449.80 frames. ], batch size: 412, lr: 8.20e-03, grad_scale: 8.0
+2023-03-02 01:09:57,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.670e+02 1.165e+03 1.465e+03 2.151e+03 6.896e+03, threshold=2.930e+03, percent-clipped=14.0
+2023-03-02 01:10:39,075 INFO [train.py:968] (0/2) Epoch 4, batch 4850, giga_loss[loss=0.3633, simple_loss=0.4182, pruned_loss=0.1542, over 27960.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3755, pruned_loss=0.1235, over 5694460.46 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3855, pruned_loss=0.1248, over 5257263.49 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3729, pruned_loss=0.1221, over 5702593.72 frames. ], batch size: 412, lr: 8.20e-03, grad_scale: 8.0
+2023-03-02 01:10:48,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1979, 2.3153, 1.1560, 1.2611], device='cuda:0'), covar=tensor([0.0879, 0.0392, 0.0860, 0.1300], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0452, 0.0309, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:0')
+2023-03-02 01:10:55,812 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:10:58,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:02,164 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:06,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:09,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:21,656 INFO [train.py:968] (0/2) Epoch 4, batch 4900, giga_loss[loss=0.2984, simple_loss=0.3652, pruned_loss=0.1158, over 28474.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3774, pruned_loss=0.1244, over 5700801.47 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3857, pruned_loss=0.1249, over 5266680.60 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.375, pruned_loss=0.1232, over 5704669.67 frames. ], batch size: 65, lr: 8.20e-03, grad_scale: 8.0
+2023-03-02 01:11:23,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.401e+02 1.131e+03 1.490e+03 1.990e+03 6.419e+03, threshold=2.980e+03, percent-clipped=5.0
+2023-03-02 01:11:24,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:29,601 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:48,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8443, 1.1990, 3.6099, 3.0953], device='cuda:0'), covar=tensor([0.1588, 0.2122, 0.0350, 0.0920], device='cuda:0'), in_proj_covar=tensor([0.0537, 0.0505, 0.0700, 0.0571], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 01:12:01,384 INFO [train.py:968] (0/2) Epoch 4, batch 4950, giga_loss[loss=0.3265, simple_loss=0.4017, pruned_loss=0.1256, over 28555.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.379, pruned_loss=0.125, over 5709303.81 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3862, pruned_loss=0.1252, over 5285628.28 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3764, pruned_loss=0.1238, over 5707146.79 frames. ], batch size: 336, lr: 8.20e-03, grad_scale: 4.0
+2023-03-02 01:12:21,465 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:12:40,915 INFO [train.py:968] (0/2) Epoch 4, batch 5000, giga_loss[loss=0.2803, simple_loss=0.3468, pruned_loss=0.1068, over 28476.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3797, pruned_loss=0.1249, over 5707997.72 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.387, pruned_loss=0.1255, over 5302193.24 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3767, pruned_loss=0.1236, over 5706080.06 frames. ], batch size: 60, lr: 8.20e-03, grad_scale: 4.0
+2023-03-02 01:12:42,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.915e+02 1.124e+03 1.523e+03 1.940e+03 7.837e+03, threshold=3.047e+03, percent-clipped=12.0
+2023-03-02 01:12:49,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:13:02,374 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:13:04,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:13:18,768 INFO [train.py:968] (0/2) Epoch 4, batch 5050, giga_loss[loss=0.2836, simple_loss=0.3574, pruned_loss=0.1049, over 28754.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3807, pruned_loss=0.1256, over 5705402.79 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3878, pruned_loss=0.126, over 5317586.80 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3773, pruned_loss=0.1242, over 5702532.45 frames. ], batch size: 119, lr: 8.20e-03, grad_scale: 4.0
+2023-03-02 01:13:25,894 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:13:25,909 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:14:01,606 INFO [train.py:968] (0/2) Epoch 4, batch 5100, giga_loss[loss=0.3593, simple_loss=0.406, pruned_loss=0.1563, over 27635.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.381, pruned_loss=0.126, over 5705927.08 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3878, pruned_loss=0.126, over 5323527.11 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3783, pruned_loss=0.1249, over 5702020.46 frames. ], batch size: 472, lr: 8.19e-03, grad_scale: 4.0
+2023-03-02 01:14:04,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.850e+02 1.231e+03 1.393e+03 1.745e+03 5.676e+03, threshold=2.786e+03, percent-clipped=5.0
+2023-03-02 01:14:12,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:14:29,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:14:39,779 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 01:14:39,967 INFO [train.py:968] (0/2) Epoch 4, batch 5150, giga_loss[loss=0.3078, simple_loss=0.3703, pruned_loss=0.1227, over 28675.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3798, pruned_loss=0.1254, over 5705427.20 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3886, pruned_loss=0.1266, over 5334675.92 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3767, pruned_loss=0.124, over 5705296.04 frames. ], batch size: 262, lr: 8.19e-03, grad_scale: 4.0
+2023-03-02 01:14:44,704 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:14:49,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:14:56,861 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:15:13,552 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:15:20,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:15:23,513 INFO [train.py:968] (0/2) Epoch 4, batch 5200, giga_loss[loss=0.2812, simple_loss=0.3522, pruned_loss=0.1051, over 28906.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3759, pruned_loss=0.1233, over 5698307.51 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3887, pruned_loss=0.1266, over 5333698.48 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3732, pruned_loss=0.122, over 5703776.84 frames. ], batch size: 227, lr: 8.19e-03, grad_scale: 8.0
+2023-03-02 01:15:24,435 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5310, 1.0687, 2.9037, 2.6841], device='cuda:0'), covar=tensor([0.1997, 0.2355, 0.0737, 0.0685], device='cuda:0'), in_proj_covar=tensor([0.0534, 0.0502, 0.0696, 0.0565], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 01:15:25,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.307e+02 1.027e+03 1.396e+03 1.751e+03 4.093e+03, threshold=2.791e+03, percent-clipped=7.0
+2023-03-02 01:16:05,244 INFO [train.py:968] (0/2) Epoch 4, batch 5250, giga_loss[loss=0.319, simple_loss=0.3787, pruned_loss=0.1296, over 28777.00 frames. ], tot_loss[loss=0.308, simple_loss=0.373, pruned_loss=0.1215, over 5703541.41 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3887, pruned_loss=0.1266, over 5339373.28 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3708, pruned_loss=0.1205, over 5705916.82 frames. ], batch size: 99, lr: 8.19e-03, grad_scale: 8.0
+2023-03-02 01:16:12,428 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:16:14,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:16:18,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9237, 1.3162, 3.3468, 2.8831], device='cuda:0'), covar=tensor([0.1378, 0.1808, 0.0397, 0.0612], device='cuda:0'), in_proj_covar=tensor([0.0532, 0.0500, 0.0692, 0.0565], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 01:16:27,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:16:30,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:16:38,474 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:16:46,194 INFO [train.py:968] (0/2) Epoch 4, batch 5300, giga_loss[loss=0.2954, simple_loss=0.3745, pruned_loss=0.1082, over 28947.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3754, pruned_loss=0.122, over 5704507.69 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3889, pruned_loss=0.1268, over 5352889.24 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3731, pruned_loss=0.1209, over 5704312.36 frames. ], batch size: 213, lr: 8.19e-03, grad_scale: 4.0
+2023-03-02 01:16:49,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.781e+02 1.132e+03 1.390e+03 2.000e+03 4.942e+03, threshold=2.781e+03, percent-clipped=9.0
+2023-03-02 01:16:55,374 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:17:29,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:17:30,166 INFO [train.py:968] (0/2) Epoch 4, batch 5350, giga_loss[loss=0.3159, simple_loss=0.38, pruned_loss=0.1259, over 28856.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3772, pruned_loss=0.122, over 5713078.08 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3893, pruned_loss=0.1272, over 5369529.49 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3747, pruned_loss=0.1207, over 5707231.39 frames. ], batch size: 199, lr: 8.19e-03, grad_scale: 4.0
+2023-03-02 01:17:44,117 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-02 01:18:06,961 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:18:11,489 INFO [train.py:968] (0/2) Epoch 4, batch 5400, giga_loss[loss=0.271, simple_loss=0.3396, pruned_loss=0.1012, over 28859.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3763, pruned_loss=0.1228, over 5717774.21 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3888, pruned_loss=0.127, over 5374835.16 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3745, pruned_loss=0.1218, over 5711573.51 frames. ], batch size: 99, lr: 8.19e-03, grad_scale: 4.0
+2023-03-02 01:18:15,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.321e+02 1.139e+03 1.408e+03 1.879e+03 5.147e+03, threshold=2.815e+03, percent-clipped=5.0
+2023-03-02 01:18:40,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:18:54,326 INFO [train.py:968] (0/2) Epoch 4, batch 5450, giga_loss[loss=0.3548, simple_loss=0.3901, pruned_loss=0.1598, over 28435.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3747, pruned_loss=0.1232, over 5724271.79 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3891, pruned_loss=0.1271, over 5385396.60 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3728, pruned_loss=0.1223, over 5715952.30 frames. ], batch size: 71, lr: 8.18e-03, grad_scale: 4.0
+2023-03-02 01:19:27,225 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:19:29,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:19:33,677 INFO [train.py:968] (0/2) Epoch 4, batch 5500, giga_loss[loss=0.3262, simple_loss=0.3821, pruned_loss=0.1352, over 28818.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3724, pruned_loss=0.1232, over 5733017.48 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.389, pruned_loss=0.1271, over 5398438.56 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1224, over 5722308.60 frames. ], batch size: 199, lr: 8.18e-03, grad_scale: 4.0
+2023-03-02 01:19:37,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.943e+02 1.113e+03 1.390e+03 1.984e+03 5.666e+03, threshold=2.781e+03, percent-clipped=10.0
+2023-03-02 01:19:52,392 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 01:19:54,642 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:04,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4331, 1.9329, 1.7316, 1.8167], device='cuda:0'), covar=tensor([0.0576, 0.0748, 0.0827, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0469, 0.0512, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-02 01:20:08,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:16,489 INFO [train.py:968] (0/2) Epoch 4, batch 5550, giga_loss[loss=0.2645, simple_loss=0.3249, pruned_loss=0.102, over 28672.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3685, pruned_loss=0.1219, over 5730438.81 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.389, pruned_loss=0.1272, over 5402446.51 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3668, pruned_loss=0.1212, over 5722924.06 frames. ], batch size: 60, lr: 8.18e-03, grad_scale: 4.0
+2023-03-02 01:20:28,650 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1715, 1.2792, 1.1020, 1.3581], device='cuda:0'), covar=tensor([0.0816, 0.0379, 0.0391, 0.0922], device='cuda:0'), in_proj_covar=tensor([0.0213, 0.0139, 0.0141, 0.0255], device='cuda:0'), out_proj_covar=tensor([0.0046, 0.0035, 0.0031, 0.0052], device='cuda:0')
+2023-03-02 01:20:36,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:41,145 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:44,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:44,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:59,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7779, 1.6763, 1.6717, 1.6069], device='cuda:0'), covar=tensor([0.0899, 0.1544, 0.1372, 0.1336], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0756, 0.0632, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 01:21:01,699 INFO [train.py:968] (0/2) Epoch 4, batch 5600, giga_loss[loss=0.2679, simple_loss=0.3434, pruned_loss=0.09621, over 28664.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3685, pruned_loss=0.1222, over 5721555.79 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3894, pruned_loss=0.1276, over 5407352.24 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3667, pruned_loss=0.1213, over 5714238.36 frames. ], batch size: 242, lr: 8.18e-03, grad_scale: 8.0
+2023-03-02 01:21:05,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.107e+02 1.087e+03 1.421e+03 1.829e+03 5.647e+03, threshold=2.842e+03, percent-clipped=11.0
+2023-03-02 01:21:11,087 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:21:43,098 INFO [train.py:968] (0/2) Epoch 4, batch 5650, giga_loss[loss=0.2795, simple_loss=0.3487, pruned_loss=0.1051, over 28916.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.365, pruned_loss=0.1202, over 5723815.77 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.389, pruned_loss=0.1274, over 5421854.64 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3634, pruned_loss=0.1195, over 5713210.17 frames. ], batch size: 174, lr: 8.18e-03, grad_scale: 8.0
+2023-03-02 01:21:48,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2764, 1.4630, 1.2713, 1.3970], device='cuda:0'), covar=tensor([0.1526, 0.1500, 0.1386, 0.1420], device='cuda:0'), in_proj_covar=tensor([0.1049, 0.0840, 0.0936, 0.0937], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 01:21:56,786 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141810.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 01:22:11,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:22:12,177 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-02 01:22:13,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:22:15,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:22:23,673 INFO [train.py:968] (0/2) Epoch 4, batch 5700, giga_loss[loss=0.2423, simple_loss=0.3069, pruned_loss=0.08884, over 28591.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3603, pruned_loss=0.1174, over 5710189.75 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3894, pruned_loss=0.1276, over 5423880.84 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3581, pruned_loss=0.1165, over 5706105.00 frames. ], batch size: 85, lr: 8.18e-03, grad_scale: 8.0
+2023-03-02 01:22:27,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.335e+02 1.233e+03 1.462e+03 1.952e+03 5.069e+03, threshold=2.925e+03, percent-clipped=11.0
+2023-03-02 01:22:36,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:22:37,523 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:22:38,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:23:01,064 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:23:02,755 INFO [train.py:968] (0/2) Epoch 4, batch 5750, giga_loss[loss=0.2909, simple_loss=0.3579, pruned_loss=0.1119, over 28897.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3585, pruned_loss=0.1162, over 5714837.95 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3898, pruned_loss=0.1279, over 5434951.69 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3555, pruned_loss=0.1149, over 5709268.49 frames. ], batch size: 145, lr: 8.18e-03, grad_scale: 4.0
+2023-03-02 01:23:03,234 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 01:23:20,408 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:23:34,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0788, 2.8696, 2.8117, 1.4441], device='cuda:0'), covar=tensor([0.0746, 0.0554, 0.0917, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0640, 0.0804, 0.0584], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 01:23:38,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2305, 3.9738, 3.9118, 1.9209], device='cuda:0'), covar=tensor([0.0394, 0.0366, 0.0670, 0.1863], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0637, 0.0799, 0.0581], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 01:23:43,967 INFO [train.py:968] (0/2) Epoch 4, batch 5800, giga_loss[loss=0.3587, simple_loss=0.3989, pruned_loss=0.1593, over 23943.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3596, pruned_loss=0.1169, over 5705316.44 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3901, pruned_loss=0.1281, over 5430551.68 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3568, pruned_loss=0.1156, over 5707012.14 frames. ], batch size: 705, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:23:48,687 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.603e+02 1.171e+03 1.588e+03 2.086e+03 4.875e+03, threshold=3.176e+03, percent-clipped=9.0
+2023-03-02 01:24:24,723 INFO [train.py:968] (0/2) Epoch 4, batch 5850, giga_loss[loss=0.2926, simple_loss=0.3601, pruned_loss=0.1125, over 29092.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3633, pruned_loss=0.1182, over 5714248.11 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3899, pruned_loss=0.1279, over 5444673.85 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3604, pruned_loss=0.117, over 5710343.05 frames. ], batch size: 128, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:24:31,524 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-142000.pt
+2023-03-02 01:25:03,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9132, 1.3199, 3.8195, 3.2345], device='cuda:0'), covar=tensor([0.1626, 0.1989, 0.0370, 0.0624], device='cuda:0'), in_proj_covar=tensor([0.0544, 0.0508, 0.0707, 0.0574], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 01:25:08,842 INFO [train.py:968] (0/2) Epoch 4, batch 5900, giga_loss[loss=0.2963, simple_loss=0.3603, pruned_loss=0.1161, over 28901.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3676, pruned_loss=0.1203, over 5712972.97 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3901, pruned_loss=0.128, over 5446698.48 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3651, pruned_loss=0.1193, over 5709296.35 frames. ], batch size: 106, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:25:13,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.247e+02 1.087e+03 1.374e+03 1.969e+03 7.197e+03, threshold=2.747e+03, percent-clipped=6.0
+2023-03-02 01:25:16,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5264, 2.2907, 1.5335, 0.8531], device='cuda:0'), covar=tensor([0.2983, 0.1461, 0.1808, 0.2588], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.1243, 0.1333, 0.1131], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 01:25:20,134 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:25:22,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:25:49,692 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:25:57,252 INFO [train.py:968] (0/2) Epoch 4, batch 5950, giga_loss[loss=0.3273, simple_loss=0.3762, pruned_loss=0.1392, over 23983.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3718, pruned_loss=0.1224, over 5706136.44 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3901, pruned_loss=0.128, over 5446698.48 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3698, pruned_loss=0.1216, over 5703274.87 frames. ], batch size: 705, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:26:00,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:26:27,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1684, 1.5569, 1.1557, 0.3762], device='cuda:0'), covar=tensor([0.1161, 0.0718, 0.1232, 0.1810], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.1238, 0.1331, 0.1120], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 01:26:41,861 INFO [train.py:968] (0/2) Epoch 4, batch 6000, giga_loss[loss=0.3449, simple_loss=0.402, pruned_loss=0.1439, over 27860.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3754, pruned_loss=0.1244, over 5698879.98 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3906, pruned_loss=0.1283, over 5446528.81 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3729, pruned_loss=0.1234, over 5703367.87 frames. ], batch size: 412, lr: 8.17e-03, grad_scale: 8.0
+2023-03-02 01:26:41,866 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 01:26:50,247 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2551, simple_loss=0.3554, pruned_loss=0.07741, over 944034.00 frames. 
+2023-03-02 01:26:50,248 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 01:26:50,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7389, 2.1966, 1.9216, 1.8138], device='cuda:0'), covar=tensor([0.1457, 0.1580, 0.1159, 0.1052], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0765, 0.0754, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 01:26:53,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.662e+02 1.205e+03 1.499e+03 2.007e+03 6.495e+03, threshold=2.997e+03, percent-clipped=9.0
+2023-03-02 01:26:55,690 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-02 01:27:26,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142185.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 01:27:34,113 INFO [train.py:968] (0/2) Epoch 4, batch 6050, giga_loss[loss=0.3323, simple_loss=0.3954, pruned_loss=0.1346, over 28901.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3817, pruned_loss=0.1307, over 5695527.33 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3904, pruned_loss=0.1281, over 5453464.49 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3798, pruned_loss=0.13, over 5696375.05 frames. ], batch size: 174, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:27:50,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:28:20,755 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:28:22,536 INFO [train.py:968] (0/2) Epoch 4, batch 6100, giga_loss[loss=0.3857, simple_loss=0.428, pruned_loss=0.1717, over 28710.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.387, pruned_loss=0.1358, over 5692363.95 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3894, pruned_loss=0.1277, over 5462940.24 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3862, pruned_loss=0.1357, over 5689896.45 frames. ], batch size: 242, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:28:22,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:28:28,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.138e+02 1.603e+03 1.949e+03 2.435e+03 5.502e+03, threshold=3.898e+03, percent-clipped=11.0
+2023-03-02 01:28:50,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:29:09,207 INFO [train.py:968] (0/2) Epoch 4, batch 6150, giga_loss[loss=0.3534, simple_loss=0.4113, pruned_loss=0.1478, over 28910.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3934, pruned_loss=0.1398, over 5698863.09 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3896, pruned_loss=0.1278, over 5478364.13 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3926, pruned_loss=0.14, over 5691382.18 frames. ], batch size: 136, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:29:24,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:29:37,433 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:29:44,308 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142328.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 01:29:48,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142331.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 01:30:01,380 INFO [train.py:968] (0/2) Epoch 4, batch 6200, giga_loss[loss=0.4033, simple_loss=0.4388, pruned_loss=0.1839, over 27977.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3991, pruned_loss=0.1453, over 5700318.96 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3892, pruned_loss=0.1276, over 5484945.99 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3989, pruned_loss=0.1458, over 5691528.95 frames. ], batch size: 412, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:30:05,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.450e+02 1.523e+03 2.105e+03 3.098e+03 7.336e+03, threshold=4.210e+03, percent-clipped=9.0
+2023-03-02 01:30:07,964 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:30:09,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:30:14,767 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142360.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 01:30:37,048 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 01:30:38,108 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:30:45,834 INFO [train.py:968] (0/2) Epoch 4, batch 6250, giga_loss[loss=0.376, simple_loss=0.4265, pruned_loss=0.1627, over 28842.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.4062, pruned_loss=0.1515, over 5696409.07 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3895, pruned_loss=0.1276, over 5492487.56 frames. ], giga_tot_loss[loss=0.3555, simple_loss=0.4062, pruned_loss=0.1524, over 5688267.59 frames. ], batch size: 119, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:31:33,372 INFO [train.py:968] (0/2) Epoch 4, batch 6300, libri_loss[loss=0.2474, simple_loss=0.3211, pruned_loss=0.08684, over 29635.00 frames. ], tot_loss[loss=0.3617, simple_loss=0.4113, pruned_loss=0.156, over 5693753.68 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3887, pruned_loss=0.1271, over 5505132.02 frames. ], giga_tot_loss[loss=0.3644, simple_loss=0.4126, pruned_loss=0.1581, over 5680659.86 frames. ], batch size: 69, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:31:39,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.767e+03 2.297e+03 2.973e+03 5.007e+03, threshold=4.594e+03, percent-clipped=5.0
+2023-03-02 01:32:13,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3180, 1.3213, 1.3183, 1.3828], device='cuda:0'), covar=tensor([0.0976, 0.1320, 0.1560, 0.1173], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0767, 0.0640, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 01:32:26,063 INFO [train.py:968] (0/2) Epoch 4, batch 6350, giga_loss[loss=0.3748, simple_loss=0.419, pruned_loss=0.1653, over 28994.00 frames. ], tot_loss[loss=0.365, simple_loss=0.4131, pruned_loss=0.1585, over 5677687.18 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3886, pruned_loss=0.1271, over 5509171.61 frames. ], giga_tot_loss[loss=0.3676, simple_loss=0.4144, pruned_loss=0.1604, over 5665612.70 frames. ], batch size: 213, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:33:04,021 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-02 01:33:20,195 INFO [train.py:968] (0/2) Epoch 4, batch 6400, giga_loss[loss=0.5597, simple_loss=0.5358, pruned_loss=0.2918, over 26516.00 frames. ], tot_loss[loss=0.3702, simple_loss=0.4162, pruned_loss=0.1621, over 5675763.21 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3885, pruned_loss=0.1272, over 5515127.97 frames. ], giga_tot_loss[loss=0.3733, simple_loss=0.4179, pruned_loss=0.1643, over 5664437.41 frames. ], batch size: 555, lr: 8.16e-03, grad_scale: 8.0
+2023-03-02 01:33:27,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.703e+02 1.769e+03 2.383e+03 3.311e+03 1.064e+04, threshold=4.766e+03, percent-clipped=11.0
+2023-03-02 01:33:50,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6062, 1.4695, 1.0990, 1.2595], device='cuda:0'), covar=tensor([0.0569, 0.0557, 0.0968, 0.0893], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0478, 0.0519, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 01:34:14,589 INFO [train.py:968] (0/2) Epoch 4, batch 6450, giga_loss[loss=0.4169, simple_loss=0.4361, pruned_loss=0.1988, over 28623.00 frames. ], tot_loss[loss=0.3782, simple_loss=0.4209, pruned_loss=0.1677, over 5655574.37 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3891, pruned_loss=0.1277, over 5509851.81 frames. ], giga_tot_loss[loss=0.3806, simple_loss=0.4223, pruned_loss=0.1695, over 5652755.50 frames. ], batch size: 307, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:35:05,616 INFO [train.py:968] (0/2) Epoch 4, batch 6500, giga_loss[loss=0.3348, simple_loss=0.3953, pruned_loss=0.1372, over 28951.00 frames. ], tot_loss[loss=0.3816, simple_loss=0.4235, pruned_loss=0.1698, over 5645817.04 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3888, pruned_loss=0.1275, over 5518523.88 frames. ], giga_tot_loss[loss=0.3854, simple_loss=0.4257, pruned_loss=0.1726, over 5639416.17 frames. ], batch size: 106, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:35:11,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.159e+02 1.770e+03 2.475e+03 3.452e+03 6.999e+03, threshold=4.951e+03, percent-clipped=8.0
+2023-03-02 01:35:24,422 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5289, 1.4719, 1.5000, 1.5358], device='cuda:0'), covar=tensor([0.0903, 0.1408, 0.1199, 0.1097], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0758, 0.0631, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 01:35:25,270 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-02 01:35:44,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:35:55,791 INFO [train.py:968] (0/2) Epoch 4, batch 6550, giga_loss[loss=0.4449, simple_loss=0.4438, pruned_loss=0.223, over 23426.00 frames. ], tot_loss[loss=0.3814, simple_loss=0.4229, pruned_loss=0.1699, over 5647278.09 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3889, pruned_loss=0.1276, over 5525515.71 frames. ], giga_tot_loss[loss=0.3854, simple_loss=0.4252, pruned_loss=0.1728, over 5638397.22 frames. ], batch size: 705, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:35:59,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:36:15,969 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-02 01:36:31,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-02 01:36:43,658 INFO [train.py:968] (0/2) Epoch 4, batch 6600, giga_loss[loss=0.3076, simple_loss=0.3653, pruned_loss=0.125, over 28841.00 frames. ], tot_loss[loss=0.3801, simple_loss=0.4216, pruned_loss=0.1693, over 5651389.51 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3893, pruned_loss=0.1279, over 5534316.42 frames. ], giga_tot_loss[loss=0.3846, simple_loss=0.4241, pruned_loss=0.1726, over 5639322.18 frames. ], batch size: 119, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:36:49,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.015e+02 1.664e+03 2.184e+03 2.946e+03 4.772e+03, threshold=4.368e+03, percent-clipped=0.0
+2023-03-02 01:37:35,114 INFO [train.py:968] (0/2) Epoch 4, batch 6650, giga_loss[loss=0.499, simple_loss=0.4974, pruned_loss=0.2503, over 26526.00 frames. ], tot_loss[loss=0.3801, simple_loss=0.4217, pruned_loss=0.1693, over 5641323.53 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3894, pruned_loss=0.1279, over 5543704.54 frames. ], giga_tot_loss[loss=0.385, simple_loss=0.4243, pruned_loss=0.1729, over 5625661.19 frames. ], batch size: 555, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:38:07,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:38:09,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:38:17,405 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:38:19,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:38:20,733 INFO [train.py:968] (0/2) Epoch 4, batch 6700, giga_loss[loss=0.3852, simple_loss=0.4205, pruned_loss=0.1749, over 28709.00 frames. ], tot_loss[loss=0.3764, simple_loss=0.4203, pruned_loss=0.1662, over 5644539.90 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3894, pruned_loss=0.1279, over 5539781.77 frames. ], giga_tot_loss[loss=0.3825, simple_loss=0.4237, pruned_loss=0.1707, over 5638632.02 frames. ], batch size: 92, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:38:27,717 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.523e+02 1.708e+03 2.178e+03 3.503e+03 1.230e+04, threshold=4.356e+03, percent-clipped=16.0
+2023-03-02 01:38:37,093 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:38:48,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:39:11,628 INFO [train.py:968] (0/2) Epoch 4, batch 6750, giga_loss[loss=0.3587, simple_loss=0.4103, pruned_loss=0.1535, over 28631.00 frames. ], tot_loss[loss=0.3789, simple_loss=0.4224, pruned_loss=0.1677, over 5627830.80 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3897, pruned_loss=0.1281, over 5533518.06 frames. ], giga_tot_loss[loss=0.3839, simple_loss=0.4251, pruned_loss=0.1713, over 5630446.50 frames. ], batch size: 336, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:39:56,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4862, 1.1015, 2.8871, 2.7399], device='cuda:0'), covar=tensor([0.1582, 0.1910, 0.0473, 0.0583], device='cuda:0'), in_proj_covar=tensor([0.0547, 0.0511, 0.0702, 0.0575], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 01:40:02,325 INFO [train.py:968] (0/2) Epoch 4, batch 6800, giga_loss[loss=0.3568, simple_loss=0.4081, pruned_loss=0.1528, over 28331.00 frames. ], tot_loss[loss=0.3753, simple_loss=0.4196, pruned_loss=0.1655, over 5622217.51 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3897, pruned_loss=0.1283, over 5537326.20 frames. ], giga_tot_loss[loss=0.3798, simple_loss=0.4221, pruned_loss=0.1687, over 5621976.79 frames. ], batch size: 60, lr: 8.15e-03, grad_scale: 8.0
+2023-03-02 01:40:09,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.717e+02 1.461e+03 1.873e+03 2.331e+03 5.161e+03, threshold=3.745e+03, percent-clipped=3.0
+2023-03-02 01:40:28,758 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:40:53,475 INFO [train.py:968] (0/2) Epoch 4, batch 6850, giga_loss[loss=0.3471, simple_loss=0.403, pruned_loss=0.1455, over 28909.00 frames. ], tot_loss[loss=0.3707, simple_loss=0.4173, pruned_loss=0.1621, over 5637753.98 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3894, pruned_loss=0.1281, over 5547493.62 frames. ], giga_tot_loss[loss=0.376, simple_loss=0.4204, pruned_loss=0.1658, over 5630854.42 frames. ], batch size: 186, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:41:18,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8082, 1.6851, 1.7843, 1.6751], device='cuda:0'), covar=tensor([0.0989, 0.1683, 0.1308, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0750, 0.0631, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 01:41:32,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6446, 1.7045, 1.5564, 1.6463], device='cuda:0'), covar=tensor([0.0854, 0.1316, 0.1065, 0.1101], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0751, 0.0632, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 01:41:41,424 INFO [train.py:968] (0/2) Epoch 4, batch 6900, libri_loss[loss=0.3502, simple_loss=0.4024, pruned_loss=0.149, over 19220.00 frames. ], tot_loss[loss=0.365, simple_loss=0.4134, pruned_loss=0.1582, over 5639116.98 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3889, pruned_loss=0.1279, over 5547945.73 frames. ], giga_tot_loss[loss=0.3708, simple_loss=0.4171, pruned_loss=0.1623, over 5636432.42 frames. ], batch size: 187, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:41:48,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.453e+02 1.537e+03 2.179e+03 3.016e+03 7.303e+03, threshold=4.357e+03, percent-clipped=16.0
+2023-03-02 01:42:02,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8532, 1.1652, 3.5142, 3.0837], device='cuda:0'), covar=tensor([0.1547, 0.2026, 0.0391, 0.1032], device='cuda:0'), in_proj_covar=tensor([0.0549, 0.0515, 0.0708, 0.0578], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 01:42:10,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5599, 2.1181, 2.0633, 2.0111], device='cuda:0'), covar=tensor([0.0889, 0.1804, 0.1403, 0.1503], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0752, 0.0634, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 01:42:31,280 INFO [train.py:968] (0/2) Epoch 4, batch 6950, giga_loss[loss=0.3261, simple_loss=0.3939, pruned_loss=0.1292, over 28613.00 frames. ], tot_loss[loss=0.36, simple_loss=0.4099, pruned_loss=0.1551, over 5646364.13 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3887, pruned_loss=0.1278, over 5547535.59 frames. ], giga_tot_loss[loss=0.3652, simple_loss=0.4131, pruned_loss=0.1586, over 5645405.52 frames. ], batch size: 307, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:42:38,930 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-02 01:42:57,624 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=143117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:43:00,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9855, 1.7522, 1.2363, 1.4810], device='cuda:0'), covar=tensor([0.0610, 0.0660, 0.1027, 0.0939], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0471, 0.0512, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-02 01:43:22,908 INFO [train.py:968] (0/2) Epoch 4, batch 7000, giga_loss[loss=0.3428, simple_loss=0.4038, pruned_loss=0.1409, over 28615.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.408, pruned_loss=0.154, over 5644755.01 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3886, pruned_loss=0.1277, over 5550995.56 frames. ], giga_tot_loss[loss=0.3626, simple_loss=0.4109, pruned_loss=0.1572, over 5641894.27 frames. ], batch size: 336, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:43:30,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.541e+02 1.507e+03 2.049e+03 2.797e+03 9.393e+03, threshold=4.099e+03, percent-clipped=7.0
+2023-03-02 01:44:10,855 INFO [train.py:968] (0/2) Epoch 4, batch 7050, libri_loss[loss=0.3517, simple_loss=0.4149, pruned_loss=0.1442, over 27881.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.4068, pruned_loss=0.153, over 5637289.90 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3886, pruned_loss=0.1276, over 5548321.40 frames. ], giga_tot_loss[loss=0.3608, simple_loss=0.4095, pruned_loss=0.1561, over 5639923.65 frames. ], batch size: 116, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:44:36,047 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-02 01:45:06,731 INFO [train.py:968] (0/2) Epoch 4, batch 7100, giga_loss[loss=0.3249, simple_loss=0.3929, pruned_loss=0.1284, over 29073.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.4056, pruned_loss=0.1518, over 5641345.67 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3879, pruned_loss=0.1272, over 5553378.07 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4087, pruned_loss=0.1552, over 5640772.74 frames. ], batch size: 155, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:45:16,873 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.607e+02 1.505e+03 1.941e+03 2.417e+03 6.620e+03, threshold=3.882e+03, percent-clipped=6.0
+2023-03-02 01:46:00,747 INFO [train.py:968] (0/2) Epoch 4, batch 7150, giga_loss[loss=0.3515, simple_loss=0.4217, pruned_loss=0.1406, over 28854.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4025, pruned_loss=0.1482, over 5647922.49 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3878, pruned_loss=0.1271, over 5559369.77 frames. ], giga_tot_loss[loss=0.3542, simple_loss=0.4054, pruned_loss=0.1515, over 5643676.67 frames. ], batch size: 227, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:46:47,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:46:48,938 INFO [train.py:968] (0/2) Epoch 4, batch 7200, libri_loss[loss=0.3243, simple_loss=0.3855, pruned_loss=0.1315, over 29550.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.4031, pruned_loss=0.1461, over 5662968.10 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3874, pruned_loss=0.127, over 5575755.87 frames. ], giga_tot_loss[loss=0.3533, simple_loss=0.4066, pruned_loss=0.1499, over 5648978.59 frames. ], batch size: 78, lr: 8.13e-03, grad_scale: 8.0
+2023-03-02 01:46:55,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.419e+02 1.414e+03 1.867e+03 2.393e+03 6.398e+03, threshold=3.734e+03, percent-clipped=3.0
+2023-03-02 01:47:34,282 INFO [train.py:968] (0/2) Epoch 4, batch 7250, giga_loss[loss=0.3619, simple_loss=0.4112, pruned_loss=0.1563, over 27945.00 frames. ], tot_loss[loss=0.3483, simple_loss=0.4046, pruned_loss=0.146, over 5678072.89 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3865, pruned_loss=0.1266, over 5587368.27 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4087, pruned_loss=0.1501, over 5659811.08 frames. ], batch size: 412, lr: 8.13e-03, grad_scale: 4.0
+2023-03-02 01:48:01,042 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-02 01:48:16,019 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 01:48:28,288 INFO [train.py:968] (0/2) Epoch 4, batch 7300, giga_loss[loss=0.307, simple_loss=0.3758, pruned_loss=0.1191, over 29002.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.405, pruned_loss=0.1471, over 5670476.06 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3861, pruned_loss=0.1264, over 5591309.47 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4089, pruned_loss=0.1508, over 5653672.06 frames. ], batch size: 136, lr: 8.13e-03, grad_scale: 2.0
+2023-03-02 01:48:37,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.896e+02 1.507e+03 1.937e+03 2.815e+03 6.873e+03, threshold=3.873e+03, percent-clipped=7.0
+2023-03-02 01:48:48,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3334, 1.8100, 1.4340, 1.4919], device='cuda:0'), covar=tensor([0.0709, 0.0269, 0.0300, 0.0720], device='cuda:0'), in_proj_covar=tensor([0.0217, 0.0140, 0.0142, 0.0257], device='cuda:0'), out_proj_covar=tensor([0.0047, 0.0035, 0.0031, 0.0053], device='cuda:0')
+2023-03-02 01:48:56,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6902, 1.1204, 3.3903, 2.9136], device='cuda:0'), covar=tensor([0.1610, 0.1986, 0.0406, 0.0632], device='cuda:0'), in_proj_covar=tensor([0.0557, 0.0519, 0.0717, 0.0584], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 01:49:06,976 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:49:09,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:49:14,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:49:14,594 INFO [train.py:968] (0/2) Epoch 4, batch 7350, giga_loss[loss=0.3713, simple_loss=0.4254, pruned_loss=0.1586, over 28878.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4049, pruned_loss=0.1476, over 5676894.80 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3862, pruned_loss=0.1264, over 5599578.39 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4085, pruned_loss=0.1511, over 5658124.97 frames. ], batch size: 186, lr: 8.13e-03, grad_scale: 2.0
+2023-03-02 01:49:35,616 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3444, 2.9486, 1.4741, 1.2790], device='cuda:0'), covar=tensor([0.0873, 0.0380, 0.0783, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0468, 0.0313, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 01:49:39,182 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:49:50,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6029, 3.3325, 3.3260, 1.7460], device='cuda:0'), covar=tensor([0.0532, 0.0512, 0.0779, 0.2078], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0672, 0.0822, 0.0593], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 01:50:07,029 INFO [train.py:968] (0/2) Epoch 4, batch 7400, libri_loss[loss=0.3407, simple_loss=0.4037, pruned_loss=0.1389, over 29517.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4036, pruned_loss=0.1481, over 5678002.68 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3861, pruned_loss=0.1264, over 5602625.92 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4066, pruned_loss=0.1511, over 5660997.47 frames. ], batch size: 84, lr: 8.13e-03, grad_scale: 2.0
+2023-03-02 01:50:15,695 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.542e+02 1.736e+03 2.328e+03 3.223e+03 8.622e+03, threshold=4.657e+03, percent-clipped=15.0
+2023-03-02 01:50:18,446 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3687, 1.5464, 1.1319, 1.5280], device='cuda:0'), covar=tensor([0.0824, 0.0319, 0.0361, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0215, 0.0139, 0.0141, 0.0254], device='cuda:0'), out_proj_covar=tensor([0.0047, 0.0035, 0.0031, 0.0052], device='cuda:0')
+2023-03-02 01:50:19,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1422, 1.3628, 1.0212, 0.4286], device='cuda:0'), covar=tensor([0.1193, 0.1020, 0.1828, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.1275, 0.1230, 0.1312, 0.1107], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 01:50:53,234 INFO [train.py:968] (0/2) Epoch 4, batch 7450, giga_loss[loss=0.3219, simple_loss=0.3864, pruned_loss=0.1288, over 28684.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4031, pruned_loss=0.1489, over 5674692.21 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3864, pruned_loss=0.1264, over 5603697.48 frames. ], giga_tot_loss[loss=0.3543, simple_loss=0.4055, pruned_loss=0.1515, over 5661143.32 frames. ], batch size: 92, lr: 8.13e-03, grad_scale: 2.0
+2023-03-02 01:51:39,863 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:51:43,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:51:43,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4851, 1.7185, 1.2623, 0.9199], device='cuda:0'), covar=tensor([0.1210, 0.0818, 0.0736, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.1311, 0.1033, 0.1084, 0.1134], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 01:51:47,630 INFO [train.py:968] (0/2) Epoch 4, batch 7500, giga_loss[loss=0.3233, simple_loss=0.3893, pruned_loss=0.1287, over 28609.00 frames. ], tot_loss[loss=0.3493, simple_loss=0.4031, pruned_loss=0.1478, over 5673326.23 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3866, pruned_loss=0.1265, over 5609527.53 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.4052, pruned_loss=0.1502, over 5658608.94 frames. ], batch size: 85, lr: 8.13e-03, grad_scale: 2.0
+2023-03-02 01:51:58,741 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.301e+02 1.426e+03 1.887e+03 2.297e+03 3.851e+03, threshold=3.775e+03, percent-clipped=0.0
+2023-03-02 01:52:12,918 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:52:37,103 INFO [train.py:968] (0/2) Epoch 4, batch 7550, giga_loss[loss=0.3701, simple_loss=0.4208, pruned_loss=0.1597, over 28828.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.4024, pruned_loss=0.1464, over 5671790.24 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3862, pruned_loss=0.1262, over 5613653.43 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4047, pruned_loss=0.149, over 5657481.61 frames. ], batch size: 284, lr: 8.12e-03, grad_scale: 2.0
+2023-03-02 01:52:50,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4609, 2.8804, 1.4768, 1.4742], device='cuda:0'), covar=tensor([0.0812, 0.0378, 0.0800, 0.1220], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0466, 0.0313, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 01:53:25,152 INFO [train.py:968] (0/2) Epoch 4, batch 7600, giga_loss[loss=0.3597, simple_loss=0.4199, pruned_loss=0.1498, over 28721.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4026, pruned_loss=0.1462, over 5680379.34 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3864, pruned_loss=0.1263, over 5617235.89 frames. ], giga_tot_loss[loss=0.3507, simple_loss=0.4045, pruned_loss=0.1485, over 5666860.45 frames. ], batch size: 262, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:53:34,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.493e+02 1.400e+03 1.926e+03 2.628e+03 4.993e+03, threshold=3.852e+03, percent-clipped=2.0
+2023-03-02 01:54:12,464 INFO [train.py:968] (0/2) Epoch 4, batch 7650, giga_loss[loss=0.3387, simple_loss=0.3907, pruned_loss=0.1433, over 28656.00 frames. ], tot_loss[loss=0.3456, simple_loss=0.4008, pruned_loss=0.1451, over 5684302.78 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3869, pruned_loss=0.1267, over 5620707.33 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4022, pruned_loss=0.147, over 5672137.40 frames. ], batch size: 262, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:54:13,036 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=143793.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 01:55:04,293 INFO [train.py:968] (0/2) Epoch 4, batch 7700, giga_loss[loss=0.287, simple_loss=0.3508, pruned_loss=0.1116, over 28120.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3995, pruned_loss=0.1455, over 5669481.02 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3872, pruned_loss=0.1268, over 5625649.58 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.4006, pruned_loss=0.1472, over 5656462.36 frames. ], batch size: 77, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:55:14,475 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.656e+03 2.044e+03 2.633e+03 5.000e+03, threshold=4.089e+03, percent-clipped=6.0
+2023-03-02 01:55:28,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=143866.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 01:55:54,574 INFO [train.py:968] (0/2) Epoch 4, batch 7750, giga_loss[loss=0.4303, simple_loss=0.4494, pruned_loss=0.2056, over 27557.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.3995, pruned_loss=0.1465, over 5669815.99 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3874, pruned_loss=0.1269, over 5628441.99 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.4002, pruned_loss=0.1479, over 5657505.34 frames. ], batch size: 472, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:56:28,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0247, 1.3278, 1.0641, 0.3305], device='cuda:0'), covar=tensor([0.1392, 0.1127, 0.1963, 0.2035], device='cuda:0'), in_proj_covar=tensor([0.1297, 0.1244, 0.1330, 0.1119], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 01:56:44,040 INFO [train.py:968] (0/2) Epoch 4, batch 7800, giga_loss[loss=0.2963, simple_loss=0.3515, pruned_loss=0.1205, over 28906.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.4002, pruned_loss=0.1478, over 5663680.85 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3877, pruned_loss=0.1271, over 5627968.97 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4009, pruned_loss=0.1494, over 5654318.04 frames. ], batch size: 99, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:56:54,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.062e+02 1.739e+03 2.159e+03 2.808e+03 8.235e+03, threshold=4.318e+03, percent-clipped=7.0
+2023-03-02 01:57:17,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6000, 1.8680, 1.8669, 1.7477], device='cuda:0'), covar=tensor([0.1408, 0.1753, 0.1070, 0.1032], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0777, 0.0753, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 01:57:35,583 INFO [train.py:968] (0/2) Epoch 4, batch 7850, giga_loss[loss=0.3274, simple_loss=0.3873, pruned_loss=0.1338, over 28526.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3984, pruned_loss=0.1471, over 5660303.62 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3876, pruned_loss=0.1271, over 5632321.36 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.3992, pruned_loss=0.1488, over 5649201.16 frames. ], batch size: 60, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:57:40,483 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-144000.pt
+2023-03-02 01:58:19,922 INFO [train.py:968] (0/2) Epoch 4, batch 7900, giga_loss[loss=0.3956, simple_loss=0.4258, pruned_loss=0.1827, over 27719.00 frames. ], tot_loss[loss=0.3457, simple_loss=0.3979, pruned_loss=0.1467, over 5665893.69 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3881, pruned_loss=0.1272, over 5640150.58 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.3984, pruned_loss=0.1484, over 5650899.92 frames. ], batch size: 472, lr: 8.11e-03, grad_scale: 2.0
+2023-03-02 01:58:30,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.039e+02 1.857e+03 2.322e+03 3.703e+03 8.116e+03, threshold=4.643e+03, percent-clipped=13.0
+2023-03-02 01:59:05,632 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 01:59:07,674 INFO [train.py:968] (0/2) Epoch 4, batch 7950, giga_loss[loss=0.3775, simple_loss=0.4265, pruned_loss=0.1643, over 28836.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.3999, pruned_loss=0.1474, over 5659253.07 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3886, pruned_loss=0.1274, over 5635312.88 frames. ], giga_tot_loss[loss=0.3489, simple_loss=0.4, pruned_loss=0.1489, over 5652341.35 frames. ], batch size: 284, lr: 8.11e-03, grad_scale: 2.0
+2023-03-02 01:59:53,166 INFO [train.py:968] (0/2) Epoch 4, batch 8000, giga_loss[loss=0.3599, simple_loss=0.4111, pruned_loss=0.1544, over 28198.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3994, pruned_loss=0.1462, over 5659478.41 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3884, pruned_loss=0.1274, over 5633347.97 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.4, pruned_loss=0.1479, over 5656733.26 frames. ], batch size: 368, lr: 8.11e-03, grad_scale: 4.0
+2023-03-02 02:00:03,613 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.633e+02 1.410e+03 1.827e+03 2.741e+03 8.582e+03, threshold=3.655e+03, percent-clipped=7.0
+2023-03-02 02:00:17,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144168.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:00:20,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5375, 1.6131, 1.7484, 1.5719], device='cuda:0'), covar=tensor([0.1017, 0.1273, 0.1271, 0.1111], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0762, 0.0637, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 02:00:38,703 INFO [train.py:968] (0/2) Epoch 4, batch 8050, libri_loss[loss=0.2769, simple_loss=0.3493, pruned_loss=0.1022, over 29658.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3981, pruned_loss=0.1441, over 5678729.90 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3877, pruned_loss=0.127, over 5640129.76 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3995, pruned_loss=0.1463, over 5671045.24 frames. ], batch size: 73, lr: 8.11e-03, grad_scale: 4.0
+2023-03-02 02:00:55,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7995, 1.6567, 1.7047, 1.5771], device='cuda:0'), covar=tensor([0.1056, 0.1840, 0.1427, 0.1493], device='cuda:0'), in_proj_covar=tensor([0.0439, 0.0771, 0.0645, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 02:01:21,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144241.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 02:01:22,840 INFO [train.py:968] (0/2) Epoch 4, batch 8100, libri_loss[loss=0.2489, simple_loss=0.3255, pruned_loss=0.08615, over 29389.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3981, pruned_loss=0.1441, over 5679823.81 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3878, pruned_loss=0.1271, over 5646178.62 frames. ], giga_tot_loss[loss=0.3463, simple_loss=0.3996, pruned_loss=0.1465, over 5669736.35 frames. ], batch size: 67, lr: 8.11e-03, grad_scale: 4.0
+2023-03-02 02:01:32,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.276e+02 1.515e+03 1.866e+03 2.941e+03 7.242e+03, threshold=3.732e+03, percent-clipped=12.0
+2023-03-02 02:02:01,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2726, 1.4985, 1.1423, 0.9971], device='cuda:0'), covar=tensor([0.0984, 0.0747, 0.0610, 0.0751], device='cuda:0'), in_proj_covar=tensor([0.1297, 0.1047, 0.1087, 0.1141], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 02:02:11,761 INFO [train.py:968] (0/2) Epoch 4, batch 8150, giga_loss[loss=0.3224, simple_loss=0.3721, pruned_loss=0.1364, over 28143.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.3999, pruned_loss=0.1466, over 5667854.92 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.387, pruned_loss=0.1265, over 5645688.99 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4022, pruned_loss=0.1497, over 5660812.32 frames. ], batch size: 77, lr: 8.11e-03, grad_scale: 4.0
+2023-03-02 02:02:28,539 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144311.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:02:32,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144314.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 02:02:46,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2731, 1.7768, 1.2260, 1.3927], device='cuda:0'), covar=tensor([0.0830, 0.0307, 0.0355, 0.0928], device='cuda:0'), in_proj_covar=tensor([0.0216, 0.0139, 0.0142, 0.0256], device='cuda:0'), out_proj_covar=tensor([0.0047, 0.0035, 0.0031, 0.0053], device='cuda:0')
+2023-03-02 02:03:02,596 INFO [train.py:968] (0/2) Epoch 4, batch 8200, giga_loss[loss=0.3909, simple_loss=0.4262, pruned_loss=0.1778, over 28658.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4019, pruned_loss=0.1496, over 5664017.66 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3869, pruned_loss=0.1265, over 5651837.22 frames. ], giga_tot_loss[loss=0.3546, simple_loss=0.4042, pruned_loss=0.1525, over 5653445.72 frames. ], batch size: 307, lr: 8.11e-03, grad_scale: 4.0
+2023-03-02 02:03:02,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=144343.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:03:13,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.127e+02 1.461e+03 1.979e+03 2.721e+03 5.909e+03, threshold=3.958e+03, percent-clipped=7.0
+2023-03-02 02:03:16,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4796, 1.6064, 1.4684, 1.4319], device='cuda:0'), covar=tensor([0.1057, 0.1457, 0.1410, 0.1412], device='cuda:0'), in_proj_covar=tensor([0.0437, 0.0765, 0.0640, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 02:03:32,942 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2823, 1.7917, 1.3675, 1.3833], device='cuda:0'), covar=tensor([0.0699, 0.0447, 0.0339, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0215, 0.0139, 0.0142, 0.0256], device='cuda:0'), out_proj_covar=tensor([0.0047, 0.0035, 0.0031, 0.0053], device='cuda:0')
+2023-03-02 02:03:45,165 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144384.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:03:47,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144387.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:03:52,621 INFO [train.py:968] (0/2) Epoch 4, batch 8250, giga_loss[loss=0.4332, simple_loss=0.4487, pruned_loss=0.2088, over 27975.00 frames. ], tot_loss[loss=0.3554, simple_loss=0.4041, pruned_loss=0.1533, over 5664480.94 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3861, pruned_loss=0.1262, over 5655958.65 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4068, pruned_loss=0.1564, over 5652560.33 frames. ], batch size: 412, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:04:15,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=144416.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:04:42,705 INFO [train.py:968] (0/2) Epoch 4, batch 8300, giga_loss[loss=0.5106, simple_loss=0.5006, pruned_loss=0.2603, over 26266.00 frames. ], tot_loss[loss=0.3587, simple_loss=0.406, pruned_loss=0.1557, over 5665831.43 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3864, pruned_loss=0.1263, over 5660247.02 frames. ], giga_tot_loss[loss=0.3626, simple_loss=0.4083, pruned_loss=0.1585, over 5652955.31 frames. ], batch size: 555, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:04:54,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.996e+02 1.778e+03 2.418e+03 3.209e+03 6.323e+03, threshold=4.836e+03, percent-clipped=17.0
+2023-03-02 02:05:18,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3273, 3.0255, 1.4596, 1.2794], device='cuda:0'), covar=tensor([0.0881, 0.0359, 0.0834, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0473, 0.0313, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 02:05:27,640 INFO [train.py:968] (0/2) Epoch 4, batch 8350, giga_loss[loss=0.3138, simple_loss=0.3837, pruned_loss=0.122, over 28973.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.4046, pruned_loss=0.1538, over 5674359.84 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3862, pruned_loss=0.126, over 5670903.20 frames. ], giga_tot_loss[loss=0.3613, simple_loss=0.4074, pruned_loss=0.1576, over 5654304.60 frames. ], batch size: 164, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:05:58,174 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-02 02:06:12,366 INFO [train.py:968] (0/2) Epoch 4, batch 8400, giga_loss[loss=0.322, simple_loss=0.377, pruned_loss=0.1335, over 28661.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.4042, pruned_loss=0.1531, over 5682294.25 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3863, pruned_loss=0.1259, over 5672601.44 frames. ], giga_tot_loss[loss=0.3601, simple_loss=0.4067, pruned_loss=0.1567, over 5664807.94 frames. ], batch size: 92, lr: 8.10e-03, grad_scale: 8.0
+2023-03-02 02:06:21,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.452e+02 1.657e+03 2.181e+03 3.075e+03 7.712e+03, threshold=4.363e+03, percent-clipped=3.0
+2023-03-02 02:06:37,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4859, 2.0778, 1.4051, 0.6076], device='cuda:0'), covar=tensor([0.1910, 0.1281, 0.1626, 0.2438], device='cuda:0'), in_proj_covar=tensor([0.1292, 0.1253, 0.1313, 0.1127], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 02:06:56,514 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-02 02:06:58,528 INFO [train.py:968] (0/2) Epoch 4, batch 8450, giga_loss[loss=0.3274, simple_loss=0.3951, pruned_loss=0.1298, over 28751.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4011, pruned_loss=0.1484, over 5689192.21 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3866, pruned_loss=0.126, over 5675015.07 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.403, pruned_loss=0.1513, over 5673226.52 frames. ], batch size: 92, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:07:18,692 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 02:07:41,882 INFO [train.py:968] (0/2) Epoch 4, batch 8500, giga_loss[loss=0.337, simple_loss=0.3945, pruned_loss=0.1397, over 28617.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3975, pruned_loss=0.1457, over 5679946.56 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3868, pruned_loss=0.1262, over 5674013.77 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.399, pruned_loss=0.1483, over 5668254.19 frames. ], batch size: 92, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:07:52,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.651e+02 1.354e+03 1.828e+03 2.533e+03 7.600e+03, threshold=3.656e+03, percent-clipped=10.0
+2023-03-02 02:08:02,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.90 vs. limit=5.0
+2023-03-02 02:08:27,861 INFO [train.py:968] (0/2) Epoch 4, batch 8550, giga_loss[loss=0.3151, simple_loss=0.3673, pruned_loss=0.1314, over 28877.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3962, pruned_loss=0.1453, over 5674484.05 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3871, pruned_loss=0.1264, over 5672419.50 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.3975, pruned_loss=0.1478, over 5666780.41 frames. ], batch size: 112, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:09:17,528 INFO [train.py:968] (0/2) Epoch 4, batch 8600, giga_loss[loss=0.36, simple_loss=0.4032, pruned_loss=0.1584, over 28655.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3958, pruned_loss=0.1456, over 5673917.53 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3876, pruned_loss=0.1266, over 5674484.60 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3965, pruned_loss=0.1477, over 5665983.00 frames. ], batch size: 85, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:09:30,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.330e+02 1.567e+03 1.971e+03 2.692e+03 9.642e+03, threshold=3.942e+03, percent-clipped=12.0
+2023-03-02 02:10:04,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3079, 1.5495, 1.3046, 1.5327], device='cuda:0'), covar=tensor([0.2088, 0.1913, 0.1915, 0.1814], device='cuda:0'), in_proj_covar=tensor([0.1074, 0.0856, 0.0957, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 02:10:09,516 INFO [train.py:968] (0/2) Epoch 4, batch 8650, libri_loss[loss=0.3286, simple_loss=0.3917, pruned_loss=0.1328, over 29526.00 frames. ], tot_loss[loss=0.3481, simple_loss=0.3996, pruned_loss=0.1484, over 5677568.48 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3877, pruned_loss=0.1267, over 5679533.21 frames. ], giga_tot_loss[loss=0.3505, simple_loss=0.4002, pruned_loss=0.1504, over 5666578.40 frames. ], batch size: 82, lr: 8.09e-03, grad_scale: 4.0
+2023-03-02 02:10:18,454 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=144804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:10:57,935 INFO [train.py:968] (0/2) Epoch 4, batch 8700, giga_loss[loss=0.3269, simple_loss=0.3968, pruned_loss=0.1284, over 28520.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4037, pruned_loss=0.1486, over 5673016.86 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3879, pruned_loss=0.1269, over 5673338.31 frames. ], giga_tot_loss[loss=0.3524, simple_loss=0.4042, pruned_loss=0.1503, over 5669825.61 frames. ], batch size: 71, lr: 8.09e-03, grad_scale: 4.0
+2023-03-02 02:11:10,930 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.068e+03 1.474e+03 1.762e+03 2.628e+03 9.129e+03, threshold=3.525e+03, percent-clipped=6.0
+2023-03-02 02:11:39,900 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 02:11:44,880 INFO [train.py:968] (0/2) Epoch 4, batch 8750, giga_loss[loss=0.4043, simple_loss=0.4547, pruned_loss=0.177, over 28942.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4051, pruned_loss=0.1478, over 5675008.26 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3869, pruned_loss=0.1264, over 5679314.78 frames. ], giga_tot_loss[loss=0.3537, simple_loss=0.4069, pruned_loss=0.1503, over 5666948.77 frames. ], batch size: 227, lr: 8.09e-03, grad_scale: 4.0
+2023-03-02 02:12:06,133 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-02 02:12:34,759 INFO [train.py:968] (0/2) Epoch 4, batch 8800, giga_loss[loss=0.3316, simple_loss=0.4002, pruned_loss=0.1315, over 28938.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4068, pruned_loss=0.1492, over 5679839.00 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3863, pruned_loss=0.1259, over 5683773.48 frames. ], giga_tot_loss[loss=0.3566, simple_loss=0.4091, pruned_loss=0.152, over 5669493.38 frames. ], batch size: 164, lr: 8.09e-03, grad_scale: 8.0
+2023-03-02 02:12:39,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7966, 2.8739, 1.8285, 0.6114], device='cuda:0'), covar=tensor([0.2449, 0.1044, 0.1439, 0.2245], device='cuda:0'), in_proj_covar=tensor([0.1316, 0.1269, 0.1342, 0.1127], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 02:12:40,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2193, 1.6061, 1.4687, 1.4342], device='cuda:0'), covar=tensor([0.1250, 0.1768, 0.1028, 0.0981], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0769, 0.0750, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 02:12:42,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.620e+02 1.533e+03 1.950e+03 2.790e+03 8.433e+03, threshold=3.901e+03, percent-clipped=18.0
+2023-03-02 02:13:19,055 INFO [train.py:968] (0/2) Epoch 4, batch 8850, giga_loss[loss=0.3325, simple_loss=0.3949, pruned_loss=0.135, over 28766.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4075, pruned_loss=0.1503, over 5679766.74 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3863, pruned_loss=0.1259, over 5679023.98 frames. ], giga_tot_loss[loss=0.3579, simple_loss=0.4098, pruned_loss=0.153, over 5676260.36 frames. ], batch size: 243, lr: 8.09e-03, grad_scale: 8.0
+2023-03-02 02:13:28,947 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-02 02:13:55,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3592, 1.4825, 1.4084, 1.4220], device='cuda:0'), covar=tensor([0.1408, 0.1623, 0.1626, 0.1481], device='cuda:0'), in_proj_covar=tensor([0.0437, 0.0764, 0.0645, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 02:14:05,219 INFO [train.py:968] (0/2) Epoch 4, batch 8900, giga_loss[loss=0.29, simple_loss=0.3611, pruned_loss=0.1095, over 28944.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.4085, pruned_loss=0.1521, over 5683068.13 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3865, pruned_loss=0.126, over 5682278.77 frames. ], giga_tot_loss[loss=0.3597, simple_loss=0.4105, pruned_loss=0.1545, over 5677434.56 frames. ], batch size: 145, lr: 8.09e-03, grad_scale: 4.0
+2023-03-02 02:14:17,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.550e+02 1.760e+03 2.210e+03 3.196e+03 6.187e+03, threshold=4.421e+03, percent-clipped=13.0
+2023-03-02 02:14:24,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6648, 3.4579, 3.3875, 1.5681], device='cuda:0'), covar=tensor([0.0618, 0.0515, 0.0915, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.0821, 0.0695, 0.0839, 0.0597], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 02:14:53,559 INFO [train.py:968] (0/2) Epoch 4, batch 8950, giga_loss[loss=0.2964, simple_loss=0.3587, pruned_loss=0.117, over 28471.00 frames. ], tot_loss[loss=0.3545, simple_loss=0.4065, pruned_loss=0.1513, over 5691920.31 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.386, pruned_loss=0.1257, over 5688195.80 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.409, pruned_loss=0.1541, over 5682135.00 frames. ], batch size: 85, lr: 8.09e-03, grad_scale: 4.0
+2023-03-02 02:15:39,826 INFO [train.py:968] (0/2) Epoch 4, batch 9000, giga_loss[loss=0.3386, simple_loss=0.3855, pruned_loss=0.1458, over 28635.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.4042, pruned_loss=0.1506, over 5682569.55 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3856, pruned_loss=0.1257, over 5684218.00 frames. ], giga_tot_loss[loss=0.3573, simple_loss=0.4072, pruned_loss=0.1537, over 5677848.44 frames. ], batch size: 119, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:15:39,830 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 02:15:43,727 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2537, 1.4738, 1.0995, 1.4332], device='cuda:0'), covar=tensor([0.0905, 0.0329, 0.0418, 0.1041], device='cuda:0'), in_proj_covar=tensor([0.0216, 0.0139, 0.0142, 0.0256], device='cuda:0'), out_proj_covar=tensor([0.0048, 0.0035, 0.0031, 0.0053], device='cuda:0')
+2023-03-02 02:15:48,058 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.249, simple_loss=0.3507, pruned_loss=0.07367, over 944034.00 frames. 
+2023-03-02 02:15:48,058 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 02:15:58,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3454, 1.9276, 1.5320, 0.4639], device='cuda:0'), covar=tensor([0.1473, 0.1045, 0.1660, 0.1850], device='cuda:0'), in_proj_covar=tensor([0.1305, 0.1258, 0.1333, 0.1123], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 02:16:00,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.966e+02 1.645e+03 2.210e+03 3.424e+03 9.815e+03, threshold=4.419e+03, percent-clipped=13.0
+2023-03-02 02:16:07,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4919, 3.3215, 1.6230, 1.2695], device='cuda:0'), covar=tensor([0.0870, 0.0327, 0.0620, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.1293, 0.1019, 0.1078, 0.1131], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 02:16:22,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:16:35,313 INFO [train.py:968] (0/2) Epoch 4, batch 9050, giga_loss[loss=0.377, simple_loss=0.4184, pruned_loss=0.1678, over 29053.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4019, pruned_loss=0.1495, over 5678793.49 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3851, pruned_loss=0.1256, over 5686569.02 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.405, pruned_loss=0.1526, over 5672839.83 frames. ], batch size: 128, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:17:01,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0449, 1.6889, 1.3422, 1.4877], device='cuda:0'), covar=tensor([0.0572, 0.0715, 0.0954, 0.1013], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0479, 0.0522, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 02:17:10,482 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4747, 1.9750, 1.6138, 1.7043], device='cuda:0'), covar=tensor([0.0561, 0.0696, 0.0943, 0.1000], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0479, 0.0522, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 02:17:25,485 INFO [train.py:968] (0/2) Epoch 4, batch 9100, libri_loss[loss=0.3048, simple_loss=0.3832, pruned_loss=0.1132, over 29250.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4014, pruned_loss=0.1493, over 5677257.19 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.385, pruned_loss=0.1254, over 5682514.02 frames. ], giga_tot_loss[loss=0.3546, simple_loss=0.4044, pruned_loss=0.1524, over 5675787.11 frames. ], batch size: 97, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:17:31,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1001, 1.8604, 1.4130, 1.3355], device='cuda:0'), covar=tensor([0.0908, 0.0354, 0.0328, 0.1083], device='cuda:0'), in_proj_covar=tensor([0.0214, 0.0137, 0.0141, 0.0254], device='cuda:0'), out_proj_covar=tensor([0.0047, 0.0035, 0.0031, 0.0053], device='cuda:0')
+2023-03-02 02:17:38,492 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.247e+02 1.679e+03 2.183e+03 2.965e+03 5.220e+03, threshold=4.366e+03, percent-clipped=4.0
+2023-03-02 02:18:15,874 INFO [train.py:968] (0/2) Epoch 4, batch 9150, giga_loss[loss=0.3524, simple_loss=0.4026, pruned_loss=0.1511, over 28679.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4025, pruned_loss=0.1508, over 5673056.04 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.385, pruned_loss=0.1254, over 5684781.13 frames. ], giga_tot_loss[loss=0.356, simple_loss=0.405, pruned_loss=0.1535, over 5669802.96 frames. ], batch size: 307, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:18:42,873 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:18:46,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:19:02,737 INFO [train.py:968] (0/2) Epoch 4, batch 9200, libri_loss[loss=0.2886, simple_loss=0.3703, pruned_loss=0.1034, over 29571.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.4013, pruned_loss=0.15, over 5679615.84 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3855, pruned_loss=0.1257, over 5691182.10 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4035, pruned_loss=0.1527, over 5670895.14 frames. ], batch size: 75, lr: 8.08e-03, grad_scale: 8.0
+2023-03-02 02:19:07,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7214, 1.1031, 3.3979, 2.8342], device='cuda:0'), covar=tensor([0.1633, 0.2015, 0.0442, 0.0640], device='cuda:0'), in_proj_covar=tensor([0.0556, 0.0521, 0.0729, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 02:19:12,095 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:19:15,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.545e+02 1.666e+03 2.211e+03 3.242e+03 6.763e+03, threshold=4.422e+03, percent-clipped=7.0
+2023-03-02 02:19:45,817 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:19:49,759 INFO [train.py:968] (0/2) Epoch 4, batch 9250, giga_loss[loss=0.3418, simple_loss=0.39, pruned_loss=0.1468, over 28594.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.4003, pruned_loss=0.1493, over 5685788.37 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3855, pruned_loss=0.1258, over 5694448.53 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4025, pruned_loss=0.1522, over 5675529.04 frames. ], batch size: 85, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:20:11,135 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:20:36,417 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9514, 1.1055, 3.6959, 3.0796], device='cuda:0'), covar=tensor([0.1555, 0.2095, 0.0383, 0.0770], device='cuda:0'), in_proj_covar=tensor([0.0557, 0.0517, 0.0732, 0.0581], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 02:20:37,575 INFO [train.py:968] (0/2) Epoch 4, batch 9300, giga_loss[loss=0.3373, simple_loss=0.3969, pruned_loss=0.1389, over 28608.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4016, pruned_loss=0.1496, over 5680961.31 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3856, pruned_loss=0.1258, over 5697341.30 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.4035, pruned_loss=0.1521, over 5670147.78 frames. ], batch size: 60, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:20:52,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.080e+02 1.418e+03 1.957e+03 2.974e+03 1.160e+04, threshold=3.914e+03, percent-clipped=7.0
+2023-03-02 02:20:54,152 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:21:03,264 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-02 02:21:24,361 INFO [train.py:968] (0/2) Epoch 4, batch 9350, libri_loss[loss=0.3382, simple_loss=0.4068, pruned_loss=0.1348, over 29641.00 frames. ], tot_loss[loss=0.3547, simple_loss=0.4052, pruned_loss=0.1521, over 5684276.48 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3857, pruned_loss=0.1258, over 5702861.19 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.4071, pruned_loss=0.1548, over 5670117.52 frames. ], batch size: 88, lr: 8.07e-03, grad_scale: 4.0
+2023-03-02 02:22:13,538 INFO [train.py:968] (0/2) Epoch 4, batch 9400, giga_loss[loss=0.3231, simple_loss=0.3993, pruned_loss=0.1235, over 29041.00 frames. ], tot_loss[loss=0.3551, simple_loss=0.4049, pruned_loss=0.1527, over 5678356.38 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3855, pruned_loss=0.1258, over 5704749.13 frames. ], giga_tot_loss[loss=0.3584, simple_loss=0.4067, pruned_loss=0.1551, over 5665168.76 frames. ], batch size: 164, lr: 8.07e-03, grad_scale: 4.0
+2023-03-02 02:22:28,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.937e+02 1.413e+03 2.084e+03 2.758e+03 4.999e+03, threshold=4.167e+03, percent-clipped=7.0
+2023-03-02 02:23:01,120 INFO [train.py:968] (0/2) Epoch 4, batch 9450, giga_loss[loss=0.3894, simple_loss=0.4293, pruned_loss=0.1748, over 27945.00 frames. ], tot_loss[loss=0.3532, simple_loss=0.4057, pruned_loss=0.1504, over 5684499.96 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3859, pruned_loss=0.1258, over 5707280.43 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4074, pruned_loss=0.1531, over 5670951.39 frames. ], batch size: 412, lr: 8.07e-03, grad_scale: 4.0
+2023-03-02 02:23:46,187 INFO [train.py:968] (0/2) Epoch 4, batch 9500, giga_loss[loss=0.3428, simple_loss=0.4083, pruned_loss=0.1387, over 28886.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4064, pruned_loss=0.1485, over 5686465.63 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3857, pruned_loss=0.1257, over 5710638.71 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4083, pruned_loss=0.1512, over 5672230.78 frames. ], batch size: 186, lr: 8.07e-03, grad_scale: 4.0
+2023-03-02 02:24:01,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.326e+02 1.371e+03 1.650e+03 2.122e+03 4.368e+03, threshold=3.301e+03, percent-clipped=1.0
+2023-03-02 02:24:14,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2614, 2.0245, 1.5053, 1.8122], device='cuda:0'), covar=tensor([0.0648, 0.0687, 0.1042, 0.1031], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0468, 0.0512, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-02 02:24:38,084 INFO [train.py:968] (0/2) Epoch 4, batch 9550, giga_loss[loss=0.4793, simple_loss=0.4792, pruned_loss=0.2397, over 26643.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.4095, pruned_loss=0.1506, over 5680530.66 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3857, pruned_loss=0.1256, over 5713941.13 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4114, pruned_loss=0.1531, over 5666016.80 frames. ], batch size: 555, lr: 8.07e-03, grad_scale: 4.0
+2023-03-02 02:25:27,323 INFO [train.py:968] (0/2) Epoch 4, batch 9600, giga_loss[loss=0.3803, simple_loss=0.4272, pruned_loss=0.1667, over 28555.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.4112, pruned_loss=0.1527, over 5681978.58 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3855, pruned_loss=0.1256, over 5715883.08 frames. ], giga_tot_loss[loss=0.3615, simple_loss=0.413, pruned_loss=0.155, over 5668716.75 frames. ], batch size: 336, lr: 8.07e-03, grad_scale: 8.0
+2023-03-02 02:25:39,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.094e+02 1.512e+03 2.025e+03 2.749e+03 9.118e+03, threshold=4.051e+03, percent-clipped=14.0
+2023-03-02 02:25:45,985 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:25:46,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:26:15,127 INFO [train.py:968] (0/2) Epoch 4, batch 9650, giga_loss[loss=0.3519, simple_loss=0.4049, pruned_loss=0.1494, over 28615.00 frames. ], tot_loss[loss=0.3608, simple_loss=0.4121, pruned_loss=0.1547, over 5672198.37 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.385, pruned_loss=0.1252, over 5710747.47 frames. ], giga_tot_loss[loss=0.365, simple_loss=0.4148, pruned_loss=0.1576, over 5665694.50 frames. ], batch size: 307, lr: 8.07e-03, grad_scale: 8.0
+2023-03-02 02:26:17,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:26:37,189 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 02:26:55,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:27:02,537 INFO [train.py:968] (0/2) Epoch 4, batch 9700, giga_loss[loss=0.3391, simple_loss=0.394, pruned_loss=0.1421, over 28997.00 frames. ], tot_loss[loss=0.3607, simple_loss=0.4113, pruned_loss=0.155, over 5668518.25 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3848, pruned_loss=0.1251, over 5714284.01 frames. ], giga_tot_loss[loss=0.3651, simple_loss=0.4141, pruned_loss=0.158, over 5659342.10 frames. ], batch size: 164, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:27:15,695 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.911e+02 1.436e+03 2.173e+03 3.043e+03 8.541e+03, threshold=4.346e+03, percent-clipped=10.0
+2023-03-02 02:27:20,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1301, 0.9338, 0.7182, 1.4129], device='cuda:0'), covar=tensor([0.0838, 0.0369, 0.0404, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0216, 0.0137, 0.0141, 0.0254], device='cuda:0'), out_proj_covar=tensor([0.0048, 0.0035, 0.0031, 0.0053], device='cuda:0')
+2023-03-02 02:27:47,767 INFO [train.py:968] (0/2) Epoch 4, batch 9750, giga_loss[loss=0.3463, simple_loss=0.4127, pruned_loss=0.1399, over 28917.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.4097, pruned_loss=0.1535, over 5665129.57 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3843, pruned_loss=0.1248, over 5717340.18 frames. ], giga_tot_loss[loss=0.3627, simple_loss=0.4126, pruned_loss=0.1564, over 5654566.41 frames. ], batch size: 186, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:27:59,483 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:28:02,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:28:29,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:28:30,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:28:31,793 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:28:32,835 INFO [train.py:968] (0/2) Epoch 4, batch 9800, libri_loss[loss=0.307, simple_loss=0.3799, pruned_loss=0.1171, over 29177.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4082, pruned_loss=0.1503, over 5670474.30 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3843, pruned_loss=0.1247, over 5721930.71 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4113, pruned_loss=0.1535, over 5656599.37 frames. ], batch size: 101, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:28:44,535 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.469e+02 1.228e+03 1.768e+03 2.701e+03 5.997e+03, threshold=3.536e+03, percent-clipped=7.0
+2023-03-02 02:28:56,501 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:29:01,599 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:29:04,172 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:29:10,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:29:14,707 INFO [train.py:968] (0/2) Epoch 4, batch 9850, giga_loss[loss=0.3468, simple_loss=0.4061, pruned_loss=0.1438, over 28838.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.4074, pruned_loss=0.1486, over 5680220.43 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3835, pruned_loss=0.1243, over 5728334.72 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.4114, pruned_loss=0.1524, over 5661808.78 frames. ], batch size: 112, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:29:20,898 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-146000.pt
+2023-03-02 02:29:32,004 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:29:43,616 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4970, 1.7816, 1.4378, 1.4353], device='cuda:0'), covar=tensor([0.0783, 0.0308, 0.0335, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0214, 0.0138, 0.0140, 0.0254], device='cuda:0'), out_proj_covar=tensor([0.0048, 0.0035, 0.0031, 0.0053], device='cuda:0')
+2023-03-02 02:29:58,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:30:03,464 INFO [train.py:968] (0/2) Epoch 4, batch 9900, giga_loss[loss=0.5065, simple_loss=0.5018, pruned_loss=0.2556, over 26701.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4088, pruned_loss=0.1501, over 5679190.11 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3832, pruned_loss=0.1243, over 5733605.92 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.4129, pruned_loss=0.1538, over 5657990.53 frames. ], batch size: 555, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:30:17,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.521e+02 1.734e+03 2.194e+03 3.128e+03 9.667e+03, threshold=4.388e+03, percent-clipped=21.0
+2023-03-02 02:30:29,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1497, 1.0931, 0.9529, 1.2927], device='cuda:0'), covar=tensor([0.0814, 0.0333, 0.0359, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0214, 0.0138, 0.0141, 0.0255], device='cuda:0'), out_proj_covar=tensor([0.0048, 0.0035, 0.0031, 0.0053], device='cuda:0')
+2023-03-02 02:30:53,549 INFO [train.py:968] (0/2) Epoch 4, batch 9950, giga_loss[loss=0.3269, simple_loss=0.3917, pruned_loss=0.131, over 28987.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.4092, pruned_loss=0.1512, over 5672123.76 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.383, pruned_loss=0.1243, over 5728130.99 frames. ], giga_tot_loss[loss=0.3617, simple_loss=0.4134, pruned_loss=0.1549, over 5657813.88 frames. ], batch size: 164, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:31:37,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:31:39,787 INFO [train.py:968] (0/2) Epoch 4, batch 10000, libri_loss[loss=0.3241, simple_loss=0.3968, pruned_loss=0.1257, over 29745.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.408, pruned_loss=0.1512, over 5680249.22 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3832, pruned_loss=0.1245, over 5732676.62 frames. ], giga_tot_loss[loss=0.3608, simple_loss=0.412, pruned_loss=0.1548, over 5663231.06 frames. ], batch size: 87, lr: 8.06e-03, grad_scale: 8.0
+2023-03-02 02:31:56,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.580e+03 2.202e+03 3.316e+03 9.960e+03, threshold=4.405e+03, percent-clipped=13.0
+2023-03-02 02:32:31,269 INFO [train.py:968] (0/2) Epoch 4, batch 10050, giga_loss[loss=0.3224, simple_loss=0.3801, pruned_loss=0.1324, over 28974.00 frames. ], tot_loss[loss=0.3539, simple_loss=0.4062, pruned_loss=0.1508, over 5674770.11 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3835, pruned_loss=0.1246, over 5734682.69 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4094, pruned_loss=0.1539, over 5658928.72 frames. ], batch size: 164, lr: 8.05e-03, grad_scale: 8.0
+2023-03-02 02:32:53,224 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146214.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 02:33:25,877 INFO [train.py:968] (0/2) Epoch 4, batch 10100, giga_loss[loss=0.3513, simple_loss=0.4135, pruned_loss=0.1445, over 28986.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.4035, pruned_loss=0.1492, over 5677741.09 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3835, pruned_loss=0.1245, over 5736536.82 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4061, pruned_loss=0.1519, over 5663072.10 frames. ], batch size: 155, lr: 8.05e-03, grad_scale: 8.0
+2023-03-02 02:33:43,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.966e+02 1.549e+03 2.002e+03 2.842e+03 5.684e+03, threshold=4.004e+03, percent-clipped=5.0
+2023-03-02 02:34:07,162 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:34:09,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:34:17,046 INFO [train.py:968] (0/2) Epoch 4, batch 10150, giga_loss[loss=0.317, simple_loss=0.3742, pruned_loss=0.1299, over 28840.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.4028, pruned_loss=0.1501, over 5673070.83 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3837, pruned_loss=0.1245, over 5736154.74 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4051, pruned_loss=0.1526, over 5660550.15 frames. ], batch size: 112, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:34:34,969 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:34:58,907 INFO [train.py:968] (0/2) Epoch 4, batch 10200, giga_loss[loss=0.3286, simple_loss=0.3863, pruned_loss=0.1354, over 28905.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.4021, pruned_loss=0.149, over 5683700.63 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3844, pruned_loss=0.1249, over 5742190.06 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.4042, pruned_loss=0.1519, over 5665592.95 frames. ], batch size: 227, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:35:14,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.102e+02 1.507e+03 2.083e+03 2.664e+03 6.043e+03, threshold=4.166e+03, percent-clipped=5.0
+2023-03-02 02:35:17,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:35:47,478 INFO [train.py:968] (0/2) Epoch 4, batch 10250, giga_loss[loss=0.3216, simple_loss=0.387, pruned_loss=0.1281, over 28529.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.4, pruned_loss=0.1463, over 5674237.27 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3848, pruned_loss=0.1251, over 5744669.45 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.4016, pruned_loss=0.1489, over 5656424.36 frames. ], batch size: 336, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:36:00,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9999, 1.1223, 0.9530, 1.1864], device='cuda:0'), covar=tensor([0.0944, 0.0362, 0.0370, 0.1179], device='cuda:0'), in_proj_covar=tensor([0.0215, 0.0138, 0.0141, 0.0255], device='cuda:0'), out_proj_covar=tensor([0.0048, 0.0035, 0.0031, 0.0053], device='cuda:0')
+2023-03-02 02:36:03,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:36:13,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2207, 1.8780, 1.3755, 1.4096], device='cuda:0'), covar=tensor([0.0780, 0.0422, 0.0364, 0.0829], device='cuda:0'), in_proj_covar=tensor([0.0216, 0.0139, 0.0142, 0.0256], device='cuda:0'), out_proj_covar=tensor([0.0048, 0.0035, 0.0032, 0.0053], device='cuda:0')
+2023-03-02 02:36:38,637 INFO [train.py:968] (0/2) Epoch 4, batch 10300, giga_loss[loss=0.317, simple_loss=0.3585, pruned_loss=0.1378, over 23461.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3966, pruned_loss=0.1432, over 5668644.50 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3847, pruned_loss=0.1249, over 5748108.75 frames. ], giga_tot_loss[loss=0.3448, simple_loss=0.3982, pruned_loss=0.1457, over 5650411.65 frames. ], batch size: 705, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:36:54,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.571e+02 1.476e+03 1.993e+03 2.730e+03 7.625e+03, threshold=3.986e+03, percent-clipped=9.0
+2023-03-02 02:37:26,036 INFO [train.py:968] (0/2) Epoch 4, batch 10350, giga_loss[loss=0.4011, simple_loss=0.4332, pruned_loss=0.1844, over 27825.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3968, pruned_loss=0.143, over 5677410.79 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3843, pruned_loss=0.1245, over 5752662.07 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.3989, pruned_loss=0.146, over 5655979.98 frames. ], batch size: 412, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:37:37,692 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:37:39,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:38:07,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2783, 1.5046, 1.2602, 1.2561], device='cuda:0'), covar=tensor([0.1512, 0.1379, 0.1247, 0.1409], device='cuda:0'), in_proj_covar=tensor([0.1082, 0.0868, 0.0968, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 02:38:12,884 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:38:17,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:38:18,703 INFO [train.py:968] (0/2) Epoch 4, batch 10400, giga_loss[loss=0.2943, simple_loss=0.3499, pruned_loss=0.1194, over 28975.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3938, pruned_loss=0.1425, over 5673729.76 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3838, pruned_loss=0.1243, over 5755998.53 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3962, pruned_loss=0.1455, over 5652327.68 frames. ], batch size: 106, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:38:27,985 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:38:30,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:38:33,184 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 02:38:33,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.778e+02 1.488e+03 2.071e+03 3.169e+03 1.245e+04, threshold=4.141e+03, percent-clipped=16.0
+2023-03-02 02:39:00,779 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:39:02,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146589.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:39:06,157 INFO [train.py:968] (0/2) Epoch 4, batch 10450, giga_loss[loss=0.4214, simple_loss=0.4591, pruned_loss=0.1919, over 28762.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3923, pruned_loss=0.1417, over 5679607.74 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3838, pruned_loss=0.1241, over 5755636.51 frames. ], giga_tot_loss[loss=0.3417, simple_loss=0.3943, pruned_loss=0.1445, over 5661353.26 frames. ], batch size: 119, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:39:53,712 INFO [train.py:968] (0/2) Epoch 4, batch 10500, giga_loss[loss=0.3257, simple_loss=0.3924, pruned_loss=0.1295, over 28971.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.395, pruned_loss=0.1431, over 5680249.24 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3838, pruned_loss=0.1241, over 5756326.00 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3967, pruned_loss=0.1457, over 5664017.18 frames. ], batch size: 164, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:39:57,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3415, 1.4057, 1.2819, 1.3999], device='cuda:0'), covar=tensor([0.2105, 0.2105, 0.1907, 0.2155], device='cuda:0'), in_proj_covar=tensor([0.1081, 0.0867, 0.0968, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 02:39:59,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5708, 1.7038, 1.4140, 0.9660], device='cuda:0'), covar=tensor([0.0970, 0.0694, 0.0625, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.1294, 0.1030, 0.1066, 0.1121], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 02:40:10,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.065e+02 1.540e+03 1.957e+03 2.670e+03 5.319e+03, threshold=3.915e+03, percent-clipped=3.0
+2023-03-02 02:40:14,301 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-02 02:40:42,447 INFO [train.py:968] (0/2) Epoch 4, batch 10550, giga_loss[loss=0.3266, simple_loss=0.3944, pruned_loss=0.1294, over 28940.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3957, pruned_loss=0.1433, over 5667889.64 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3836, pruned_loss=0.1239, over 5759329.24 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3975, pruned_loss=0.1459, over 5650878.96 frames. ], batch size: 164, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:40:43,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2442, 1.6624, 1.2812, 0.4553], device='cuda:0'), covar=tensor([0.1216, 0.0658, 0.0961, 0.1507], device='cuda:0'), in_proj_covar=tensor([0.1305, 0.1233, 0.1308, 0.1126], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 02:40:59,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2261, 2.4861, 2.3484, 2.1552], device='cuda:0'), covar=tensor([0.1488, 0.1528, 0.1006, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0778, 0.0760, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 02:41:19,789 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146732.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:41:22,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146735.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:41:28,495 INFO [train.py:968] (0/2) Epoch 4, batch 10600, giga_loss[loss=0.2844, simple_loss=0.3602, pruned_loss=0.1043, over 29031.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3963, pruned_loss=0.1437, over 5648872.19 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3849, pruned_loss=0.1248, over 5753885.09 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3972, pruned_loss=0.146, over 5635957.87 frames. ], batch size: 155, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:41:44,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.770e+02 1.348e+03 1.749e+03 2.337e+03 5.970e+03, threshold=3.497e+03, percent-clipped=5.0
+2023-03-02 02:41:50,577 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146764.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:42:15,338 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-02 02:42:17,748 INFO [train.py:968] (0/2) Epoch 4, batch 10650, libri_loss[loss=0.3496, simple_loss=0.4187, pruned_loss=0.1402, over 29254.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3954, pruned_loss=0.1435, over 5646015.79 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3849, pruned_loss=0.1247, over 5755955.80 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3963, pruned_loss=0.1457, over 5632336.65 frames. ], batch size: 94, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:43:07,678 INFO [train.py:968] (0/2) Epoch 4, batch 10700, giga_loss[loss=0.347, simple_loss=0.3991, pruned_loss=0.1475, over 28951.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3971, pruned_loss=0.1451, over 5641580.63 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3849, pruned_loss=0.1247, over 5756413.41 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.398, pruned_loss=0.1471, over 5628788.95 frames. ], batch size: 145, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:43:24,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.132e+02 1.618e+03 2.337e+03 3.815e+03 1.323e+04, threshold=4.674e+03, percent-clipped=28.0
+2023-03-02 02:43:58,858 INFO [train.py:968] (0/2) Epoch 4, batch 10750, giga_loss[loss=0.4323, simple_loss=0.436, pruned_loss=0.2143, over 23365.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3997, pruned_loss=0.1469, over 5639853.69 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.385, pruned_loss=0.1247, over 5749045.79 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.4005, pruned_loss=0.1487, over 5633935.80 frames. ], batch size: 705, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:44:21,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:44:42,659 INFO [train.py:968] (0/2) Epoch 4, batch 10800, giga_loss[loss=0.2846, simple_loss=0.3523, pruned_loss=0.1085, over 28369.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.4001, pruned_loss=0.1466, over 5651769.15 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3851, pruned_loss=0.1247, over 5745328.23 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4013, pruned_loss=0.1491, over 5645629.86 frames. ], batch size: 65, lr: 8.03e-03, grad_scale: 8.0
+2023-03-02 02:44:58,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.967e+02 1.453e+03 1.787e+03 2.312e+03 6.183e+03, threshold=3.574e+03, percent-clipped=3.0
+2023-03-02 02:45:13,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3519, 1.8444, 1.3737, 0.6582], device='cuda:0'), covar=tensor([0.1749, 0.1018, 0.1425, 0.2081], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.1255, 0.1322, 0.1141], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 02:45:21,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:45:29,672 INFO [train.py:968] (0/2) Epoch 4, batch 10850, giga_loss[loss=0.3461, simple_loss=0.4007, pruned_loss=0.1458, over 29037.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.403, pruned_loss=0.149, over 5640469.74 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3851, pruned_loss=0.1247, over 5732105.18 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4045, pruned_loss=0.1519, over 5644950.32 frames. ], batch size: 155, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:45:57,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:46:12,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2072, 1.2623, 1.0387, 1.0064], device='cuda:0'), covar=tensor([0.0616, 0.0474, 0.0959, 0.0727], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0482, 0.0523, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 02:46:13,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4286, 1.9663, 1.2969, 1.5425], device='cuda:0'), covar=tensor([0.0836, 0.0285, 0.0375, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0212, 0.0137, 0.0140, 0.0252], device='cuda:0'), out_proj_covar=tensor([0.0048, 0.0035, 0.0031, 0.0053], device='cuda:0')
+2023-03-02 02:46:17,742 INFO [train.py:968] (0/2) Epoch 4, batch 10900, libri_loss[loss=0.2746, simple_loss=0.3479, pruned_loss=0.1006, over 29577.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4033, pruned_loss=0.1495, over 5644984.52 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3847, pruned_loss=0.1244, over 5735960.42 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4055, pruned_loss=0.1528, over 5642269.67 frames. ], batch size: 75, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:46:34,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:46:34,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.980e+02 1.515e+03 2.013e+03 2.560e+03 7.024e+03, threshold=4.026e+03, percent-clipped=9.0
+2023-03-02 02:46:38,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:47:09,440 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:47:09,831 INFO [train.py:968] (0/2) Epoch 4, batch 10950, giga_loss[loss=0.3757, simple_loss=0.4187, pruned_loss=0.1664, over 27616.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4031, pruned_loss=0.1479, over 5654012.40 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3843, pruned_loss=0.1242, over 5741369.44 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4059, pruned_loss=0.1516, over 5644177.59 frames. ], batch size: 472, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:48:02,723 INFO [train.py:968] (0/2) Epoch 4, batch 11000, giga_loss[loss=0.3215, simple_loss=0.3794, pruned_loss=0.1318, over 28897.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.4025, pruned_loss=0.1482, over 5649341.25 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3843, pruned_loss=0.1242, over 5741274.61 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.4049, pruned_loss=0.1514, over 5640775.24 frames. ], batch size: 227, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:48:19,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.436e+02 1.776e+03 2.264e+03 3.175e+03 1.320e+04, threshold=4.528e+03, percent-clipped=10.0
+2023-03-02 02:48:28,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3272, 1.3992, 1.2073, 1.3305], device='cuda:0'), covar=tensor([0.2022, 0.1946, 0.1842, 0.1937], device='cuda:0'), in_proj_covar=tensor([0.1076, 0.0863, 0.0964, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 02:48:54,151 INFO [train.py:968] (0/2) Epoch 4, batch 11050, giga_loss[loss=0.3327, simple_loss=0.3852, pruned_loss=0.1401, over 28842.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4014, pruned_loss=0.1479, over 5645393.60 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3851, pruned_loss=0.1248, over 5723688.18 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4029, pruned_loss=0.1503, over 5652409.62 frames. ], batch size: 199, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:49:25,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3264, 1.3962, 1.2028, 1.3066], device='cuda:0'), covar=tensor([0.2059, 0.2070, 0.1913, 0.1969], device='cuda:0'), in_proj_covar=tensor([0.1080, 0.0864, 0.0965, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 02:49:43,779 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:49:49,846 INFO [train.py:968] (0/2) Epoch 4, batch 11100, giga_loss[loss=0.3203, simple_loss=0.3823, pruned_loss=0.1292, over 28734.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4019, pruned_loss=0.1489, over 5635953.58 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3857, pruned_loss=0.1253, over 5716923.74 frames. ], giga_tot_loss[loss=0.3526, simple_loss=0.403, pruned_loss=0.1511, over 5644511.91 frames. ], batch size: 262, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:50:07,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.542e+02 1.631e+03 2.027e+03 2.781e+03 5.795e+03, threshold=4.054e+03, percent-clipped=2.0
+2023-03-02 02:50:39,408 INFO [train.py:968] (0/2) Epoch 4, batch 11150, giga_loss[loss=0.345, simple_loss=0.395, pruned_loss=0.1475, over 28890.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.4012, pruned_loss=0.149, over 5654301.68 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3857, pruned_loss=0.1253, over 5716923.74 frames. ], giga_tot_loss[loss=0.3518, simple_loss=0.4021, pruned_loss=0.1507, over 5660962.74 frames. ], batch size: 227, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:51:17,813 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 02:51:29,942 INFO [train.py:968] (0/2) Epoch 4, batch 11200, giga_loss[loss=0.316, simple_loss=0.3816, pruned_loss=0.1252, over 29006.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.4018, pruned_loss=0.1504, over 5652827.58 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3858, pruned_loss=0.1253, over 5717464.12 frames. ], giga_tot_loss[loss=0.3531, simple_loss=0.4025, pruned_loss=0.1518, over 5657317.64 frames. ], batch size: 164, lr: 8.02e-03, grad_scale: 8.0
+2023-03-02 02:51:46,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:51:48,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.810e+02 1.477e+03 1.913e+03 2.543e+03 8.089e+03, threshold=3.825e+03, percent-clipped=10.0
+2023-03-02 02:52:14,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0324, 1.8223, 1.2834, 1.4830], device='cuda:0'), covar=tensor([0.0569, 0.0622, 0.0943, 0.0961], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0474, 0.0515, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 02:52:18,228 INFO [train.py:968] (0/2) Epoch 4, batch 11250, giga_loss[loss=0.3781, simple_loss=0.3946, pruned_loss=0.1808, over 23576.00 frames. ], tot_loss[loss=0.3514, simple_loss=0.4017, pruned_loss=0.1506, over 5656377.25 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3858, pruned_loss=0.1254, over 5718503.68 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4025, pruned_loss=0.1522, over 5657886.07 frames. ], batch size: 705, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:52:20,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:53:03,751 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 02:53:07,336 INFO [train.py:968] (0/2) Epoch 4, batch 11300, giga_loss[loss=0.34, simple_loss=0.4041, pruned_loss=0.138, over 28921.00 frames. ], tot_loss[loss=0.353, simple_loss=0.4031, pruned_loss=0.1514, over 5664387.53 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.386, pruned_loss=0.1255, over 5718676.93 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4039, pruned_loss=0.1531, over 5664249.27 frames. ], batch size: 174, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:53:10,616 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-02 02:53:26,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.884e+02 1.547e+03 1.980e+03 2.772e+03 1.052e+04, threshold=3.960e+03, percent-clipped=6.0
+2023-03-02 02:53:55,389 INFO [train.py:968] (0/2) Epoch 4, batch 11350, giga_loss[loss=0.3863, simple_loss=0.4315, pruned_loss=0.1706, over 28928.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.4053, pruned_loss=0.1532, over 5665070.29 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3858, pruned_loss=0.1253, over 5721835.41 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4066, pruned_loss=0.1553, over 5660691.92 frames. ], batch size: 145, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:54:03,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:54:05,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:54:06,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6957, 2.4359, 1.7125, 0.7946], device='cuda:0'), covar=tensor([0.2109, 0.1067, 0.1804, 0.2280], device='cuda:0'), in_proj_covar=tensor([0.1319, 0.1271, 0.1345, 0.1155], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 02:54:21,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6189, 1.4110, 1.1844, 1.2423], device='cuda:0'), covar=tensor([0.0554, 0.0566, 0.0903, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0474, 0.0514, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 02:54:26,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3061, 1.3657, 1.0855, 0.7298], device='cuda:0'), covar=tensor([0.0765, 0.0666, 0.0494, 0.0720], device='cuda:0'), in_proj_covar=tensor([0.1318, 0.1058, 0.1074, 0.1134], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 02:54:32,218 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:54:35,792 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:54:38,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:54:44,296 INFO [train.py:968] (0/2) Epoch 4, batch 11400, libri_loss[loss=0.3053, simple_loss=0.3645, pruned_loss=0.1231, over 29491.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4054, pruned_loss=0.1535, over 5658398.58 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3858, pruned_loss=0.1254, over 5714972.11 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4069, pruned_loss=0.1556, over 5658959.34 frames. ], batch size: 70, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:54:59,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.980e+02 1.589e+03 2.040e+03 2.734e+03 6.814e+03, threshold=4.080e+03, percent-clipped=16.0
+2023-03-02 02:55:05,996 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:55:27,594 INFO [train.py:968] (0/2) Epoch 4, batch 11450, giga_loss[loss=0.3235, simple_loss=0.3826, pruned_loss=0.1322, over 28694.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.4045, pruned_loss=0.1535, over 5655823.36 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3853, pruned_loss=0.125, over 5722485.97 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4069, pruned_loss=0.1567, over 5647170.17 frames. ], batch size: 262, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:55:45,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:56:05,589 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:56:07,156 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-02 02:56:12,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9190, 2.3580, 2.1930, 1.9741], device='cuda:0'), covar=tensor([0.1537, 0.1640, 0.1095, 0.1052], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0782, 0.0757, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 02:56:13,282 INFO [train.py:968] (0/2) Epoch 4, batch 11500, giga_loss[loss=0.3145, simple_loss=0.3755, pruned_loss=0.1268, over 28881.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4036, pruned_loss=0.1526, over 5661196.21 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3853, pruned_loss=0.1248, over 5724999.99 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4059, pruned_loss=0.1558, over 5650901.53 frames. ], batch size: 106, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:56:32,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.038e+02 1.585e+03 1.959e+03 2.845e+03 7.283e+03, threshold=3.918e+03, percent-clipped=5.0
+2023-03-02 02:56:58,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3746, 2.8046, 1.5165, 1.3942], device='cuda:0'), covar=tensor([0.0809, 0.0419, 0.0779, 0.1228], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0481, 0.0316, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 02:57:04,261 INFO [train.py:968] (0/2) Epoch 4, batch 11550, giga_loss[loss=0.3577, simple_loss=0.3854, pruned_loss=0.165, over 23375.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4024, pruned_loss=0.151, over 5666411.87 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3847, pruned_loss=0.1244, over 5728245.76 frames. ], giga_tot_loss[loss=0.3569, simple_loss=0.4051, pruned_loss=0.1544, over 5654463.94 frames. ], batch size: 705, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 02:57:08,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4828, 2.2001, 1.5507, 0.6308], device='cuda:0'), covar=tensor([0.2366, 0.1168, 0.1902, 0.2486], device='cuda:0'), in_proj_covar=tensor([0.1337, 0.1278, 0.1348, 0.1150], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 02:57:11,297 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:57:17,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6378, 1.0512, 3.3119, 2.9848], device='cuda:0'), covar=tensor([0.1668, 0.2227, 0.0470, 0.0682], device='cuda:0'), in_proj_covar=tensor([0.0555, 0.0519, 0.0737, 0.0586], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 02:57:22,269 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-02 02:57:26,411 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-02 02:57:47,104 INFO [train.py:968] (0/2) Epoch 4, batch 11600, giga_loss[loss=0.3232, simple_loss=0.3829, pruned_loss=0.1317, over 29065.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.4024, pruned_loss=0.15, over 5673010.00 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3847, pruned_loss=0.1243, over 5733826.34 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4052, pruned_loss=0.1538, over 5656298.48 frames. ], batch size: 106, lr: 8.01e-03, grad_scale: 8.0
+2023-03-02 02:57:58,608 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:57:58,761 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.36 vs. limit=5.0
+2023-03-02 02:58:01,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:58:05,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.628e+02 1.653e+03 2.251e+03 2.966e+03 6.158e+03, threshold=4.503e+03, percent-clipped=9.0
+2023-03-02 02:58:31,731 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:58:37,284 INFO [train.py:968] (0/2) Epoch 4, batch 11650, giga_loss[loss=0.3491, simple_loss=0.4043, pruned_loss=0.1469, over 28704.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4036, pruned_loss=0.1507, over 5689142.21 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3846, pruned_loss=0.1241, over 5738264.29 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4064, pruned_loss=0.1545, over 5670697.30 frames. ], batch size: 242, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 02:58:43,200 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:59:09,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7805, 3.5260, 3.4941, 1.5432], device='cuda:0'), covar=tensor([0.0611, 0.0499, 0.0941, 0.2200], device='cuda:0'), in_proj_covar=tensor([0.0823, 0.0691, 0.0836, 0.0597], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 02:59:12,911 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:59:26,855 INFO [train.py:968] (0/2) Epoch 4, batch 11700, libri_loss[loss=0.2847, simple_loss=0.3526, pruned_loss=0.1084, over 29480.00 frames. ], tot_loss[loss=0.3574, simple_loss=0.4066, pruned_loss=0.1541, over 5683829.11 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3844, pruned_loss=0.124, over 5741382.42 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.4094, pruned_loss=0.1578, over 5665345.77 frames. ], batch size: 70, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 02:59:45,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.424e+02 1.681e+03 2.223e+03 2.942e+03 5.449e+03, threshold=4.446e+03, percent-clipped=6.0
+2023-03-02 03:00:12,689 INFO [train.py:968] (0/2) Epoch 4, batch 11750, giga_loss[loss=0.3606, simple_loss=0.4168, pruned_loss=0.1522, over 28893.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.4059, pruned_loss=0.1532, over 5694085.32 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3836, pruned_loss=0.1236, over 5745535.33 frames. ], giga_tot_loss[loss=0.3619, simple_loss=0.4094, pruned_loss=0.1572, over 5674199.47 frames. ], batch size: 174, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 03:00:50,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3680, 1.4831, 1.1275, 0.8359], device='cuda:0'), covar=tensor([0.0873, 0.0682, 0.0504, 0.0759], device='cuda:0'), in_proj_covar=tensor([0.1311, 0.1056, 0.1077, 0.1128], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 03:00:57,812 INFO [train.py:968] (0/2) Epoch 4, batch 11800, giga_loss[loss=0.3528, simple_loss=0.4133, pruned_loss=0.1462, over 28881.00 frames. ], tot_loss[loss=0.3554, simple_loss=0.4062, pruned_loss=0.1522, over 5688574.72 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3836, pruned_loss=0.1235, over 5745842.59 frames. ], giga_tot_loss[loss=0.3612, simple_loss=0.4097, pruned_loss=0.1564, over 5670903.63 frames. ], batch size: 186, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 03:01:17,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.729e+02 1.495e+03 2.154e+03 2.685e+03 5.822e+03, threshold=4.308e+03, percent-clipped=7.0
+2023-03-02 03:01:48,335 INFO [train.py:968] (0/2) Epoch 4, batch 11850, giga_loss[loss=0.3552, simple_loss=0.4127, pruned_loss=0.1489, over 28343.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4052, pruned_loss=0.1508, over 5675944.40 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3836, pruned_loss=0.1235, over 5747888.72 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4083, pruned_loss=0.1545, over 5658969.76 frames. ], batch size: 368, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 03:01:53,310 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-148000.pt
+2023-03-02 03:02:01,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:02:33,962 INFO [train.py:968] (0/2) Epoch 4, batch 11900, giga_loss[loss=0.3902, simple_loss=0.4266, pruned_loss=0.177, over 28610.00 frames. ], tot_loss[loss=0.3502, simple_loss=0.4029, pruned_loss=0.1487, over 5686422.81 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3837, pruned_loss=0.1235, over 5751617.66 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.4057, pruned_loss=0.1522, over 5668220.76 frames. ], batch size: 307, lr: 8.00e-03, grad_scale: 4.0
+2023-03-02 03:02:42,566 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:02:45,177 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-02 03:02:51,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.620e+03 2.216e+03 2.704e+03 6.191e+03, threshold=4.433e+03, percent-clipped=5.0
+2023-03-02 03:02:59,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6240, 1.3856, 1.1852, 1.2403], device='cuda:0'), covar=tensor([0.0554, 0.0580, 0.0917, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0477, 0.0518, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 03:03:05,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:03:17,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2561, 2.3652, 1.2951, 1.2804], device='cuda:0'), covar=tensor([0.0842, 0.0417, 0.0843, 0.1314], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0477, 0.0314, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 03:03:18,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1061, 1.4892, 3.1224, 2.8548], device='cuda:0'), covar=tensor([0.1130, 0.1661, 0.0399, 0.0694], device='cuda:0'), in_proj_covar=tensor([0.0553, 0.0522, 0.0735, 0.0584], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 03:03:18,932 INFO [train.py:968] (0/2) Epoch 4, batch 11950, giga_loss[loss=0.3904, simple_loss=0.4085, pruned_loss=0.1861, over 23384.00 frames. ], tot_loss[loss=0.351, simple_loss=0.403, pruned_loss=0.1495, over 5676038.13 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.384, pruned_loss=0.1239, over 5746385.69 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4053, pruned_loss=0.1525, over 5664866.46 frames. ], batch size: 705, lr: 8.00e-03, grad_scale: 4.0
+2023-03-02 03:04:05,548 INFO [train.py:968] (0/2) Epoch 4, batch 12000, giga_loss[loss=0.3445, simple_loss=0.4022, pruned_loss=0.1434, over 28693.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.405, pruned_loss=0.1509, over 5674571.25 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3842, pruned_loss=0.1238, over 5749682.10 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4072, pruned_loss=0.1541, over 5660803.53 frames. ], batch size: 119, lr: 8.00e-03, grad_scale: 8.0
+2023-03-02 03:04:05,553 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 03:04:09,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8134, 1.1843, 3.4725, 3.0543], device='cuda:0'), covar=tensor([0.1892, 0.2479, 0.0440, 0.0624], device='cuda:0'), in_proj_covar=tensor([0.0560, 0.0525, 0.0739, 0.0586], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 03:04:14,258 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2481, simple_loss=0.3495, pruned_loss=0.0733, over 944034.00 frames. 
+2023-03-02 03:04:14,259 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 03:04:25,018 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:04:27,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:04:33,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.545e+02 1.422e+03 1.794e+03 2.733e+03 1.166e+04, threshold=3.588e+03, percent-clipped=3.0
+2023-03-02 03:04:44,168 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:04:54,497 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:01,868 INFO [train.py:968] (0/2) Epoch 4, batch 12050, giga_loss[loss=0.3383, simple_loss=0.4035, pruned_loss=0.1365, over 29022.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4055, pruned_loss=0.1515, over 5677780.38 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3837, pruned_loss=0.1235, over 5752665.75 frames. ], giga_tot_loss[loss=0.3592, simple_loss=0.4083, pruned_loss=0.155, over 5662071.72 frames. ], batch size: 164, lr: 8.00e-03, grad_scale: 2.0
+2023-03-02 03:05:10,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:14,965 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:28,027 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:30,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:49,235 INFO [train.py:968] (0/2) Epoch 4, batch 12100, giga_loss[loss=0.328, simple_loss=0.39, pruned_loss=0.133, over 28995.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4051, pruned_loss=0.1519, over 5676153.90 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3841, pruned_loss=0.1235, over 5756434.91 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4077, pruned_loss=0.1556, over 5657859.28 frames. ], batch size: 164, lr: 8.00e-03, grad_scale: 2.0
+2023-03-02 03:05:57,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:59,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:06:10,403 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-02 03:06:11,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.762e+02 1.366e+03 1.672e+03 2.119e+03 1.088e+04, threshold=3.344e+03, percent-clipped=7.0
+2023-03-02 03:06:35,830 INFO [train.py:968] (0/2) Epoch 4, batch 12150, giga_loss[loss=0.3161, simple_loss=0.3801, pruned_loss=0.1261, over 28675.00 frames. ], tot_loss[loss=0.3551, simple_loss=0.405, pruned_loss=0.1526, over 5678505.85 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.384, pruned_loss=0.1236, over 5761043.66 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.4078, pruned_loss=0.1564, over 5657157.30 frames. ], batch size: 262, lr: 8.00e-03, grad_scale: 2.0
+2023-03-02 03:06:47,786 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:00,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:04,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:06,745 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 03:07:19,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7351, 3.5139, 3.4807, 1.5873], device='cuda:0'), covar=tensor([0.0575, 0.0489, 0.0775, 0.2027], device='cuda:0'), in_proj_covar=tensor([0.0834, 0.0702, 0.0845, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 03:07:24,323 INFO [train.py:968] (0/2) Epoch 4, batch 12200, giga_loss[loss=0.4014, simple_loss=0.4455, pruned_loss=0.1787, over 28856.00 frames. ], tot_loss[loss=0.3587, simple_loss=0.4073, pruned_loss=0.155, over 5672346.84 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3838, pruned_loss=0.1234, over 5761675.22 frames. ], giga_tot_loss[loss=0.364, simple_loss=0.4102, pruned_loss=0.1589, over 5652862.57 frames. ], batch size: 199, lr: 8.00e-03, grad_scale: 2.0
+2023-03-02 03:07:26,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:30,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:32,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:46,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.468e+02 1.559e+03 1.986e+03 2.890e+03 6.409e+03, threshold=3.971e+03, percent-clipped=16.0
+2023-03-02 03:07:56,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:08:10,061 INFO [train.py:968] (0/2) Epoch 4, batch 12250, giga_loss[loss=0.4019, simple_loss=0.4169, pruned_loss=0.1935, over 23503.00 frames. ], tot_loss[loss=0.359, simple_loss=0.4079, pruned_loss=0.1551, over 5663427.07 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3841, pruned_loss=0.1235, over 5755441.82 frames. ], giga_tot_loss[loss=0.364, simple_loss=0.4105, pruned_loss=0.1588, over 5651608.79 frames. ], batch size: 705, lr: 8.00e-03, grad_scale: 2.0
+2023-03-02 03:08:15,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6839, 1.7276, 1.5696, 1.7016], device='cuda:0'), covar=tensor([0.1030, 0.1393, 0.1420, 0.1143], device='cuda:0'), in_proj_covar=tensor([0.0442, 0.0763, 0.0639, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 03:08:31,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7421, 3.1141, 2.2089, 0.8499], device='cuda:0'), covar=tensor([0.2906, 0.1024, 0.1536, 0.2929], device='cuda:0'), in_proj_covar=tensor([0.1340, 0.1287, 0.1347, 0.1153], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 03:08:42,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:08:57,239 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-02 03:08:57,313 INFO [train.py:968] (0/2) Epoch 4, batch 12300, giga_loss[loss=0.327, simple_loss=0.3921, pruned_loss=0.131, over 28975.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.4058, pruned_loss=0.1538, over 5636582.37 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3838, pruned_loss=0.1234, over 5747986.42 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.4089, pruned_loss=0.1578, over 5631499.32 frames. ], batch size: 213, lr: 7.99e-03, grad_scale: 2.0
+2023-03-02 03:09:18,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.408e+02 1.599e+03 2.094e+03 2.917e+03 6.699e+03, threshold=4.188e+03, percent-clipped=9.0
+2023-03-02 03:09:44,324 INFO [train.py:968] (0/2) Epoch 4, batch 12350, giga_loss[loss=0.3839, simple_loss=0.4265, pruned_loss=0.1707, over 28962.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4068, pruned_loss=0.1546, over 5637448.23 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3839, pruned_loss=0.1234, over 5742097.12 frames. ], giga_tot_loss[loss=0.3632, simple_loss=0.4096, pruned_loss=0.1584, over 5636363.16 frames. ], batch size: 227, lr: 7.99e-03, grad_scale: 2.0
+2023-03-02 03:10:29,752 INFO [train.py:968] (0/2) Epoch 4, batch 12400, giga_loss[loss=0.3964, simple_loss=0.415, pruned_loss=0.1889, over 23552.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4058, pruned_loss=0.1527, over 5637353.94 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3838, pruned_loss=0.1232, over 5736564.70 frames. ], giga_tot_loss[loss=0.3609, simple_loss=0.4085, pruned_loss=0.1566, over 5640172.95 frames. ], batch size: 705, lr: 7.99e-03, grad_scale: 4.0
+2023-03-02 03:10:48,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.665e+02 1.505e+03 1.895e+03 2.553e+03 7.779e+03, threshold=3.789e+03, percent-clipped=9.0
+2023-03-02 03:10:55,670 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:10:58,411 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:11:06,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:11:17,591 INFO [train.py:968] (0/2) Epoch 4, batch 12450, giga_loss[loss=0.3252, simple_loss=0.3834, pruned_loss=0.1335, over 28749.00 frames. ], tot_loss[loss=0.3539, simple_loss=0.4042, pruned_loss=0.1518, over 5647185.66 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3833, pruned_loss=0.1229, over 5739046.46 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4071, pruned_loss=0.1555, over 5646007.56 frames. ], batch size: 119, lr: 7.99e-03, grad_scale: 4.0
+2023-03-02 03:11:28,032 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:11:39,199 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1852, 1.6847, 1.4001, 1.6008], device='cuda:0'), covar=tensor([0.0665, 0.0875, 0.0992, 0.1085], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0476, 0.0515, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 03:11:48,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:12:04,881 INFO [train.py:968] (0/2) Epoch 4, batch 12500, giga_loss[loss=0.3406, simple_loss=0.3924, pruned_loss=0.1444, over 28964.00 frames. ], tot_loss[loss=0.351, simple_loss=0.402, pruned_loss=0.15, over 5654268.70 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3836, pruned_loss=0.1229, over 5741584.69 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4044, pruned_loss=0.1534, over 5649453.06 frames. ], batch size: 213, lr: 7.99e-03, grad_scale: 4.0
+2023-03-02 03:12:23,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3796, 1.8734, 1.7018, 1.6981], device='cuda:0'), covar=tensor([0.1265, 0.1716, 0.1057, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0778, 0.0752, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 03:12:27,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.173e+02 1.508e+03 1.992e+03 2.769e+03 6.736e+03, threshold=3.985e+03, percent-clipped=8.0
+2023-03-02 03:12:39,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:12:39,800 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:12:52,363 INFO [train.py:968] (0/2) Epoch 4, batch 12550, giga_loss[loss=0.3434, simple_loss=0.3917, pruned_loss=0.1476, over 28946.00 frames. ], tot_loss[loss=0.3487, simple_loss=0.3998, pruned_loss=0.1488, over 5665993.74 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3841, pruned_loss=0.1234, over 5735963.19 frames. ], giga_tot_loss[loss=0.3525, simple_loss=0.4017, pruned_loss=0.1517, over 5665146.20 frames. ], batch size: 136, lr: 7.99e-03, grad_scale: 4.0
+2023-03-02 03:13:23,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:13:26,132 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:13:38,075 INFO [train.py:968] (0/2) Epoch 4, batch 12600, giga_loss[loss=0.2851, simple_loss=0.3487, pruned_loss=0.1107, over 28335.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.3955, pruned_loss=0.1463, over 5644092.14 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3844, pruned_loss=0.1235, over 5730268.45 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3971, pruned_loss=0.1492, over 5646868.67 frames. ], batch size: 77, lr: 7.99e-03, grad_scale: 4.0
+2023-03-02 03:13:45,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6829, 1.8280, 1.8891, 1.8560], device='cuda:0'), covar=tensor([0.1271, 0.1472, 0.0969, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0773, 0.0749, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 03:13:53,149 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-02 03:13:53,578 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:13:58,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.618e+02 1.691e+03 2.094e+03 2.916e+03 6.615e+03, threshold=4.187e+03, percent-clipped=10.0
+2023-03-02 03:14:01,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:14:04,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:14:25,768 INFO [train.py:968] (0/2) Epoch 4, batch 12650, giga_loss[loss=0.5076, simple_loss=0.4857, pruned_loss=0.2647, over 26632.00 frames. ], tot_loss[loss=0.3457, simple_loss=0.3961, pruned_loss=0.1476, over 5646224.71 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3851, pruned_loss=0.1238, over 5732338.54 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.3971, pruned_loss=0.1503, over 5644051.16 frames. ], batch size: 555, lr: 7.98e-03, grad_scale: 4.0
+2023-03-02 03:14:30,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:14:50,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:14:50,884 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:14:53,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:15:10,091 INFO [train.py:968] (0/2) Epoch 4, batch 12700, giga_loss[loss=0.3121, simple_loss=0.3749, pruned_loss=0.1246, over 28936.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3962, pruned_loss=0.148, over 5638146.88 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3855, pruned_loss=0.1243, over 5723742.89 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.3969, pruned_loss=0.1505, over 5641493.44 frames. ], batch size: 199, lr: 7.98e-03, grad_scale: 4.0
+2023-03-02 03:15:23,332 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:15:32,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.678e+02 1.696e+03 2.227e+03 3.406e+03 1.124e+04, threshold=4.455e+03, percent-clipped=15.0
+2023-03-02 03:15:58,364 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:16:00,300 INFO [train.py:968] (0/2) Epoch 4, batch 12750, giga_loss[loss=0.3072, simple_loss=0.3702, pruned_loss=0.1221, over 28586.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3968, pruned_loss=0.1479, over 5643173.75 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3853, pruned_loss=0.1242, over 5727180.99 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.3978, pruned_loss=0.1504, over 5641249.27 frames. ], batch size: 71, lr: 7.98e-03, grad_scale: 4.0
+2023-03-02 03:16:50,477 INFO [train.py:968] (0/2) Epoch 4, batch 12800, giga_loss[loss=0.3013, simple_loss=0.3798, pruned_loss=0.1114, over 28582.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3938, pruned_loss=0.1437, over 5648450.33 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3853, pruned_loss=0.1244, over 5729458.53 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3948, pruned_loss=0.1459, over 5643670.61 frames. ], batch size: 307, lr: 7.98e-03, grad_scale: 8.0
+2023-03-02 03:17:14,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.125e+02 1.420e+03 1.944e+03 2.564e+03 5.007e+03, threshold=3.889e+03, percent-clipped=2.0
+2023-03-02 03:17:40,721 INFO [train.py:968] (0/2) Epoch 4, batch 12850, giga_loss[loss=0.2778, simple_loss=0.3518, pruned_loss=0.1018, over 28878.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3895, pruned_loss=0.1389, over 5655535.46 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3847, pruned_loss=0.1243, over 5736865.85 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.391, pruned_loss=0.1414, over 5642232.33 frames. ], batch size: 186, lr: 7.98e-03, grad_scale: 8.0
+2023-03-02 03:18:28,149 INFO [train.py:968] (0/2) Epoch 4, batch 12900, giga_loss[loss=0.2776, simple_loss=0.35, pruned_loss=0.1026, over 28763.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3851, pruned_loss=0.1344, over 5658697.69 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3842, pruned_loss=0.124, over 5741429.01 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3869, pruned_loss=0.137, over 5641172.40 frames. ], batch size: 284, lr: 7.98e-03, grad_scale: 4.0
+2023-03-02 03:18:30,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:18:43,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:18:53,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.574e+02 1.179e+03 1.507e+03 2.241e+03 1.001e+04, threshold=3.015e+03, percent-clipped=9.0
+2023-03-02 03:19:20,815 INFO [train.py:968] (0/2) Epoch 4, batch 12950, giga_loss[loss=0.3054, simple_loss=0.3698, pruned_loss=0.1205, over 27961.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3798, pruned_loss=0.1298, over 5643760.80 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3835, pruned_loss=0.1237, over 5735162.42 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3817, pruned_loss=0.1324, over 5633322.07 frames. ], batch size: 412, lr: 7.98e-03, grad_scale: 4.0
+2023-03-02 03:19:35,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:20:08,862 INFO [train.py:968] (0/2) Epoch 4, batch 13000, giga_loss[loss=0.3028, simple_loss=0.3834, pruned_loss=0.1111, over 28915.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3777, pruned_loss=0.1255, over 5657766.22 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3824, pruned_loss=0.1232, over 5736945.34 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3801, pruned_loss=0.1281, over 5645630.26 frames. ], batch size: 199, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:20:34,187 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.145e+02 1.193e+03 1.621e+03 2.059e+03 4.431e+03, threshold=3.242e+03, percent-clipped=8.0
+2023-03-02 03:21:04,931 INFO [train.py:968] (0/2) Epoch 4, batch 13050, giga_loss[loss=0.3515, simple_loss=0.3983, pruned_loss=0.1523, over 27699.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3777, pruned_loss=0.1251, over 5653907.51 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3824, pruned_loss=0.1233, over 5738360.49 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3795, pruned_loss=0.1271, over 5641717.59 frames. ], batch size: 474, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:21:08,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:21:09,536 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:21:12,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:21:42,623 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:21:56,201 INFO [train.py:968] (0/2) Epoch 4, batch 13100, giga_loss[loss=0.2894, simple_loss=0.3591, pruned_loss=0.1098, over 28633.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3764, pruned_loss=0.1243, over 5654518.36 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3822, pruned_loss=0.1233, over 5740292.43 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.378, pruned_loss=0.1259, over 5642076.85 frames. ], batch size: 242, lr: 7.97e-03, grad_scale: 2.0
+2023-03-02 03:22:20,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:22:20,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.108e+02 1.370e+03 1.824e+03 2.582e+03 9.079e+03, threshold=3.648e+03, percent-clipped=15.0
+2023-03-02 03:22:47,834 INFO [train.py:968] (0/2) Epoch 4, batch 13150, giga_loss[loss=0.3141, simple_loss=0.3787, pruned_loss=0.1247, over 28849.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3724, pruned_loss=0.1215, over 5637703.24 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3822, pruned_loss=0.1233, over 5733053.64 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3735, pruned_loss=0.1227, over 5632659.57 frames. ], batch size: 112, lr: 7.97e-03, grad_scale: 2.0
+2023-03-02 03:22:56,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4438, 1.5201, 1.2600, 1.9580], device='cuda:0'), covar=tensor([0.2178, 0.1972, 0.1856, 0.2020], device='cuda:0'), in_proj_covar=tensor([0.1071, 0.0845, 0.0956, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 03:23:13,763 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-02 03:23:35,642 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:23:37,268 INFO [train.py:968] (0/2) Epoch 4, batch 13200, giga_loss[loss=0.258, simple_loss=0.3426, pruned_loss=0.08673, over 29022.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3716, pruned_loss=0.1211, over 5648370.54 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3819, pruned_loss=0.1232, over 5735742.07 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3726, pruned_loss=0.1221, over 5640276.65 frames. ], batch size: 155, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:23:38,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:24:01,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.109e+02 1.344e+03 1.694e+03 2.195e+03 6.149e+03, threshold=3.388e+03, percent-clipped=4.0
+2023-03-02 03:24:07,857 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:24:25,572 INFO [train.py:968] (0/2) Epoch 4, batch 13250, giga_loss[loss=0.2622, simple_loss=0.3421, pruned_loss=0.09121, over 28764.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3697, pruned_loss=0.1196, over 5650600.82 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3805, pruned_loss=0.1226, over 5740775.42 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3714, pruned_loss=0.121, over 5636652.16 frames. ], batch size: 99, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:24:40,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:24:42,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:24:48,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:25:09,808 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:25:10,855 INFO [train.py:968] (0/2) Epoch 4, batch 13300, giga_loss[loss=0.2487, simple_loss=0.3347, pruned_loss=0.08135, over 28952.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 5661155.26 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3801, pruned_loss=0.1224, over 5743263.43 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3696, pruned_loss=0.1192, over 5645044.13 frames. ], batch size: 145, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:25:34,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.895e+02 1.384e+03 1.757e+03 2.427e+03 4.334e+03, threshold=3.515e+03, percent-clipped=4.0
+2023-03-02 03:25:53,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:26:01,674 INFO [train.py:968] (0/2) Epoch 4, batch 13350, giga_loss[loss=0.2517, simple_loss=0.3121, pruned_loss=0.09569, over 24118.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3649, pruned_loss=0.1154, over 5646630.90 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3799, pruned_loss=0.1224, over 5735319.40 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3658, pruned_loss=0.1161, over 5637699.95 frames. ], batch size: 705, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:26:48,767 INFO [train.py:968] (0/2) Epoch 4, batch 13400, giga_loss[loss=0.2627, simple_loss=0.3338, pruned_loss=0.09582, over 28890.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3611, pruned_loss=0.1133, over 5652987.43 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3789, pruned_loss=0.1221, over 5739013.29 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3621, pruned_loss=0.1139, over 5638808.31 frames. ], batch size: 213, lr: 7.96e-03, grad_scale: 2.0
+2023-03-02 03:26:51,655 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2363, 1.4191, 1.3157, 0.8486], device='cuda:0'), covar=tensor([0.0710, 0.0618, 0.0335, 0.0626], device='cuda:0'), in_proj_covar=tensor([0.1278, 0.1023, 0.1019, 0.1079], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 03:27:12,305 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:27:15,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:27:17,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.198e+02 1.426e+03 1.691e+03 2.115e+03 5.572e+03, threshold=3.383e+03, percent-clipped=6.0
+2023-03-02 03:27:42,987 INFO [train.py:968] (0/2) Epoch 4, batch 13450, giga_loss[loss=0.2774, simple_loss=0.3505, pruned_loss=0.1021, over 28998.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3584, pruned_loss=0.112, over 5657943.02 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.378, pruned_loss=0.1217, over 5735384.49 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5647524.13 frames. ], batch size: 128, lr: 7.96e-03, grad_scale: 2.0
+2023-03-02 03:27:44,231 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:28:19,832 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:28:21,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:28:36,508 INFO [train.py:968] (0/2) Epoch 4, batch 13500, giga_loss[loss=0.2671, simple_loss=0.3236, pruned_loss=0.1053, over 24054.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3571, pruned_loss=0.1118, over 5645296.09 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3778, pruned_loss=0.1216, over 5736285.13 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3579, pruned_loss=0.1122, over 5635687.45 frames. ], batch size: 705, lr: 7.96e-03, grad_scale: 2.0
+2023-03-02 03:28:53,514 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:29:02,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.087e+02 1.369e+03 1.805e+03 2.629e+03 9.454e+03, threshold=3.611e+03, percent-clipped=13.0
+2023-03-02 03:29:13,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4014, 1.4135, 1.1868, 1.6297], device='cuda:0'), covar=tensor([0.2550, 0.2163, 0.2200, 0.2322], device='cuda:0'), in_proj_covar=tensor([0.1070, 0.0837, 0.0956, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 03:29:33,535 INFO [train.py:968] (0/2) Epoch 4, batch 13550, giga_loss[loss=0.27, simple_loss=0.347, pruned_loss=0.09654, over 29098.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3579, pruned_loss=0.1118, over 5651751.05 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3774, pruned_loss=0.1215, over 5736302.07 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3584, pruned_loss=0.1121, over 5642010.66 frames. ], batch size: 155, lr: 7.96e-03, grad_scale: 2.0
+2023-03-02 03:30:25,525 INFO [train.py:968] (0/2) Epoch 4, batch 13600, giga_loss[loss=0.3296, simple_loss=0.3964, pruned_loss=0.1314, over 28554.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3608, pruned_loss=0.1125, over 5637896.14 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3769, pruned_loss=0.1212, over 5727148.46 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3611, pruned_loss=0.1126, over 5635122.61 frames. ], batch size: 307, lr: 7.96e-03, grad_scale: 4.0
+2023-03-02 03:30:51,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.083e+02 1.496e+03 1.847e+03 2.311e+03 5.891e+03, threshold=3.694e+03, percent-clipped=3.0
+2023-03-02 03:31:20,002 INFO [train.py:968] (0/2) Epoch 4, batch 13650, giga_loss[loss=0.271, simple_loss=0.3484, pruned_loss=0.09678, over 28935.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3612, pruned_loss=0.1117, over 5661040.78 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3762, pruned_loss=0.1208, over 5733333.71 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3615, pruned_loss=0.1118, over 5650322.97 frames. ], batch size: 186, lr: 7.96e-03, grad_scale: 4.0
+2023-03-02 03:32:25,105 INFO [train.py:968] (0/2) Epoch 4, batch 13700, giga_loss[loss=0.2733, simple_loss=0.3457, pruned_loss=0.1005, over 27601.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3614, pruned_loss=0.112, over 5667617.16 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3756, pruned_loss=0.1205, over 5736269.18 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3619, pruned_loss=0.1123, over 5655104.19 frames. ], batch size: 472, lr: 7.96e-03, grad_scale: 4.0
+2023-03-02 03:32:52,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.771e+02 1.351e+03 1.818e+03 2.508e+03 7.231e+03, threshold=3.636e+03, percent-clipped=11.0
+2023-03-02 03:32:58,776 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 03:33:21,468 INFO [train.py:968] (0/2) Epoch 4, batch 13750, libri_loss[loss=0.2962, simple_loss=0.3693, pruned_loss=0.1116, over 29664.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3591, pruned_loss=0.1103, over 5672013.11 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3755, pruned_loss=0.1206, over 5738323.60 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3593, pruned_loss=0.1101, over 5658228.68 frames. ], batch size: 91, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:33:38,692 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:34:18,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8402, 1.9169, 1.8430, 1.8481], device='cuda:0'), covar=tensor([0.0961, 0.1636, 0.1310, 0.1290], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0751, 0.0622, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 03:34:18,945 INFO [train.py:968] (0/2) Epoch 4, batch 13800, giga_loss[loss=0.2791, simple_loss=0.3559, pruned_loss=0.1012, over 28814.00 frames. ], tot_loss[loss=0.287, simple_loss=0.358, pruned_loss=0.108, over 5674151.45 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3748, pruned_loss=0.1202, over 5742027.09 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3583, pruned_loss=0.1079, over 5658053.78 frames. ], batch size: 99, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:34:37,928 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:34:49,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.580e+02 1.193e+03 1.616e+03 2.137e+03 4.653e+03, threshold=3.231e+03, percent-clipped=5.0
+2023-03-02 03:35:17,198 INFO [train.py:968] (0/2) Epoch 4, batch 13850, giga_loss[loss=0.2487, simple_loss=0.3276, pruned_loss=0.08489, over 28892.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3556, pruned_loss=0.1068, over 5673583.11 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.374, pruned_loss=0.1198, over 5746545.84 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3559, pruned_loss=0.1066, over 5653372.17 frames. ], batch size: 164, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:35:25,378 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-150000.pt
+2023-03-02 03:36:12,909 INFO [train.py:968] (0/2) Epoch 4, batch 13900, giga_loss[loss=0.3002, simple_loss=0.369, pruned_loss=0.1157, over 28151.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3541, pruned_loss=0.1072, over 5677543.02 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3731, pruned_loss=0.1193, over 5749545.26 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3541, pruned_loss=0.1069, over 5654878.81 frames. ], batch size: 412, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:36:15,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3104, 1.7706, 1.2847, 0.5229], device='cuda:0'), covar=tensor([0.1459, 0.0901, 0.1547, 0.1892], device='cuda:0'), in_proj_covar=tensor([0.1292, 0.1242, 0.1311, 0.1104], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 03:36:32,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3444, 1.5727, 1.4952, 1.5373], device='cuda:0'), covar=tensor([0.0942, 0.1419, 0.1286, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0751, 0.0621, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 03:36:42,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.033e+02 1.273e+03 1.769e+03 2.318e+03 5.257e+03, threshold=3.539e+03, percent-clipped=8.0
+2023-03-02 03:36:56,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8518, 2.0163, 1.8895, 1.8515], device='cuda:0'), covar=tensor([0.1013, 0.1801, 0.1319, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0757, 0.0625, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 03:37:13,906 INFO [train.py:968] (0/2) Epoch 4, batch 13950, giga_loss[loss=0.2636, simple_loss=0.3457, pruned_loss=0.09077, over 28991.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3533, pruned_loss=0.1072, over 5680337.78 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3732, pruned_loss=0.1193, over 5750856.65 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3531, pruned_loss=0.1068, over 5660718.17 frames. ], batch size: 285, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:37:53,817 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 03:37:58,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-02 03:38:03,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150139.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 03:38:08,447 INFO [train.py:968] (0/2) Epoch 4, batch 14000, giga_loss[loss=0.2613, simple_loss=0.3463, pruned_loss=0.08812, over 28916.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3544, pruned_loss=0.1079, over 5669530.04 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3724, pruned_loss=0.119, over 5748298.07 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3542, pruned_loss=0.1074, over 5653136.32 frames. ], batch size: 145, lr: 7.95e-03, grad_scale: 8.0
+2023-03-02 03:38:22,179 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2045, 1.4009, 0.8817, 0.9792], device='cuda:0'), covar=tensor([0.0825, 0.0602, 0.0601, 0.0637], device='cuda:0'), in_proj_covar=tensor([0.1310, 0.1021, 0.1015, 0.1085], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 03:38:39,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.138e+02 1.392e+03 1.782e+03 2.530e+03 4.763e+03, threshold=3.563e+03, percent-clipped=6.0
+2023-03-02 03:38:57,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4676, 1.7078, 1.5315, 1.5215], device='cuda:0'), covar=tensor([0.1001, 0.1599, 0.1293, 0.1329], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0749, 0.0617, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 03:39:12,853 INFO [train.py:968] (0/2) Epoch 4, batch 14050, giga_loss[loss=0.2511, simple_loss=0.3388, pruned_loss=0.08173, over 28973.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3562, pruned_loss=0.108, over 5665070.56 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3721, pruned_loss=0.1189, over 5749726.09 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3562, pruned_loss=0.1077, over 5650343.31 frames. ], batch size: 136, lr: 7.95e-03, grad_scale: 8.0
+2023-03-02 03:40:15,420 INFO [train.py:968] (0/2) Epoch 4, batch 14100, libri_loss[loss=0.252, simple_loss=0.3227, pruned_loss=0.0906, over 29577.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3525, pruned_loss=0.1051, over 5677682.87 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3714, pruned_loss=0.1185, over 5754867.65 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3525, pruned_loss=0.1047, over 5657406.58 frames. ], batch size: 74, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:40:17,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3693, 1.9979, 1.7412, 1.6969], device='cuda:0'), covar=tensor([0.1745, 0.1951, 0.1263, 0.1227], device='cuda:0'), in_proj_covar=tensor([0.0787, 0.0746, 0.0746, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 03:40:39,227 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 03:40:53,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.293e+02 1.270e+03 1.727e+03 2.463e+03 5.871e+03, threshold=3.453e+03, percent-clipped=9.0
+2023-03-02 03:41:01,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6990, 1.5686, 1.2456, 1.3763], device='cuda:0'), covar=tensor([0.0539, 0.0495, 0.0900, 0.0778], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0454, 0.0513, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 03:41:10,678 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:41:23,219 INFO [train.py:968] (0/2) Epoch 4, batch 14150, giga_loss[loss=0.3016, simple_loss=0.3631, pruned_loss=0.12, over 28684.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3529, pruned_loss=0.1056, over 5674669.51 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3716, pruned_loss=0.1186, over 5746234.77 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3525, pruned_loss=0.105, over 5665901.01 frames. ], batch size: 99, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:42:17,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:42:33,008 INFO [train.py:968] (0/2) Epoch 4, batch 14200, giga_loss[loss=0.3243, simple_loss=0.4027, pruned_loss=0.123, over 28969.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3558, pruned_loss=0.1067, over 5678303.14 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3717, pruned_loss=0.1188, over 5748743.54 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3551, pruned_loss=0.1059, over 5667548.27 frames. ], batch size: 284, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:43:08,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.437e+02 1.350e+03 2.071e+03 2.654e+03 5.397e+03, threshold=4.142e+03, percent-clipped=14.0
+2023-03-02 03:43:37,514 INFO [train.py:968] (0/2) Epoch 4, batch 14250, giga_loss[loss=0.3075, simple_loss=0.3806, pruned_loss=0.1172, over 28598.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3599, pruned_loss=0.1068, over 5668214.65 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3716, pruned_loss=0.1187, over 5742641.34 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3592, pruned_loss=0.106, over 5664030.43 frames. ], batch size: 307, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:44:03,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2590, 2.9407, 1.2678, 1.2940], device='cuda:0'), covar=tensor([0.1142, 0.0450, 0.1082, 0.1638], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0463, 0.0313, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 03:44:16,539 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-02 03:44:17,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:44:22,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:44:40,980 INFO [train.py:968] (0/2) Epoch 4, batch 14300, giga_loss[loss=0.2985, simple_loss=0.3762, pruned_loss=0.1104, over 28502.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3613, pruned_loss=0.1064, over 5662754.98 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3718, pruned_loss=0.1189, over 5741434.26 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3603, pruned_loss=0.1054, over 5659752.20 frames. ], batch size: 85, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:44:57,053 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:45:09,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.882e+02 1.346e+03 1.624e+03 2.191e+03 5.729e+03, threshold=3.248e+03, percent-clipped=5.0
+2023-03-02 03:45:13,113 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:45:17,596 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:45:21,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:45:37,298 INFO [train.py:968] (0/2) Epoch 4, batch 14350, giga_loss[loss=0.2839, simple_loss=0.3631, pruned_loss=0.1024, over 28884.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3606, pruned_loss=0.1054, over 5672923.46 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3712, pruned_loss=0.1185, over 5745138.07 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3601, pruned_loss=0.1046, over 5665555.66 frames. ], batch size: 164, lr: 7.94e-03, grad_scale: 2.0
+2023-03-02 03:45:45,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1909, 1.6604, 1.1774, 0.4508], device='cuda:0'), covar=tensor([0.1482, 0.0971, 0.1841, 0.1906], device='cuda:0'), in_proj_covar=tensor([0.1310, 0.1259, 0.1336, 0.1117], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 03:45:58,906 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:46:05,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150514.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 03:46:42,084 INFO [train.py:968] (0/2) Epoch 4, batch 14400, giga_loss[loss=0.2773, simple_loss=0.3546, pruned_loss=0.09997, over 28654.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3601, pruned_loss=0.1056, over 5671167.76 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3709, pruned_loss=0.1183, over 5745958.45 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3597, pruned_loss=0.1048, over 5662638.70 frames. ], batch size: 262, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:47:13,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.415e+02 1.400e+03 2.014e+03 3.066e+03 1.030e+04, threshold=4.027e+03, percent-clipped=20.0
+2023-03-02 03:47:22,958 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-02 03:47:44,185 INFO [train.py:968] (0/2) Epoch 4, batch 14450, giga_loss[loss=0.2914, simple_loss=0.356, pruned_loss=0.1134, over 27574.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3583, pruned_loss=0.1053, over 5682073.18 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3709, pruned_loss=0.1184, over 5748955.72 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3576, pruned_loss=0.1043, over 5670988.47 frames. ], batch size: 472, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:47:56,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6093, 3.4227, 1.5821, 1.5554], device='cuda:0'), covar=tensor([0.0828, 0.0314, 0.0862, 0.1254], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0463, 0.0311, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 03:48:58,583 INFO [train.py:968] (0/2) Epoch 4, batch 14500, giga_loss[loss=0.2565, simple_loss=0.3396, pruned_loss=0.08671, over 28963.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3597, pruned_loss=0.1071, over 5691103.52 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3706, pruned_loss=0.1182, over 5752780.28 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3592, pruned_loss=0.1061, over 5677599.16 frames. ], batch size: 106, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:49:19,239 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150657.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 03:49:25,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150660.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 03:49:34,903 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:49:41,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.579e+02 1.282e+03 1.675e+03 2.185e+03 5.634e+03, threshold=3.350e+03, percent-clipped=2.0
+2023-03-02 03:50:11,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150689.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 03:50:18,280 INFO [train.py:968] (0/2) Epoch 4, batch 14550, giga_loss[loss=0.2469, simple_loss=0.3313, pruned_loss=0.08121, over 29020.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3556, pruned_loss=0.1054, over 5686596.93 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3698, pruned_loss=0.1178, over 5756569.01 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3557, pruned_loss=0.1048, over 5670956.24 frames. ], batch size: 136, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:50:34,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7523, 2.5204, 1.6994, 0.8105], device='cuda:0'), covar=tensor([0.2941, 0.1468, 0.1993, 0.2769], device='cuda:0'), in_proj_covar=tensor([0.1307, 0.1262, 0.1337, 0.1115], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 03:50:57,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3987, 1.7943, 1.5900, 1.6225], device='cuda:0'), covar=tensor([0.1188, 0.1379, 0.0975, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0755, 0.0749, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 03:51:22,775 INFO [train.py:968] (0/2) Epoch 4, batch 14600, giga_loss[loss=0.2554, simple_loss=0.3391, pruned_loss=0.08583, over 28856.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3526, pruned_loss=0.103, over 5689289.92 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3688, pruned_loss=0.1171, over 5760603.53 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.353, pruned_loss=0.1027, over 5671060.42 frames. ], batch size: 164, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:51:49,113 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-02 03:51:54,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.137e+02 1.271e+03 1.919e+03 2.787e+03 7.282e+03, threshold=3.838e+03, percent-clipped=19.0
+2023-03-02 03:52:22,618 INFO [train.py:968] (0/2) Epoch 4, batch 14650, giga_loss[loss=0.2449, simple_loss=0.3257, pruned_loss=0.08202, over 28718.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3514, pruned_loss=0.1032, over 5677266.43 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3686, pruned_loss=0.1172, over 5746990.47 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3514, pruned_loss=0.1024, over 5672200.28 frames. ], batch size: 262, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:53:24,377 INFO [train.py:968] (0/2) Epoch 4, batch 14700, giga_loss[loss=0.315, simple_loss=0.3841, pruned_loss=0.123, over 28948.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3559, pruned_loss=0.1063, over 5665929.77 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3687, pruned_loss=0.1173, over 5738142.01 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3555, pruned_loss=0.1054, over 5669186.45 frames. ], batch size: 199, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:53:33,753 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:54:00,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.147e+02 1.525e+03 1.898e+03 2.541e+03 5.206e+03, threshold=3.796e+03, percent-clipped=5.0
+2023-03-02 03:54:02,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4133, 3.0595, 1.5216, 1.3632], device='cuda:0'), covar=tensor([0.0882, 0.0329, 0.0884, 0.1380], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0460, 0.0312, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 03:54:30,328 INFO [train.py:968] (0/2) Epoch 4, batch 14750, giga_loss[loss=0.2647, simple_loss=0.3351, pruned_loss=0.09714, over 28491.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3585, pruned_loss=0.1081, over 5665989.77 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3683, pruned_loss=0.1172, over 5731378.27 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3584, pruned_loss=0.1073, over 5673339.21 frames. ], batch size: 336, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:54:34,055 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9170, 3.6841, 3.6262, 1.8689], device='cuda:0'), covar=tensor([0.0463, 0.0396, 0.0849, 0.1936], device='cuda:0'), in_proj_covar=tensor([0.0768, 0.0660, 0.0761, 0.0570], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 03:55:14,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0556, 1.2910, 0.9143, 0.8199], device='cuda:0'), covar=tensor([0.0658, 0.0476, 0.0417, 0.0583], device='cuda:0'), in_proj_covar=tensor([0.1285, 0.0983, 0.1014, 0.1072], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001], device='cuda:0')
+2023-03-02 03:55:17,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:55:32,255 INFO [train.py:968] (0/2) Epoch 4, batch 14800, giga_loss[loss=0.263, simple_loss=0.3194, pruned_loss=0.1033, over 24533.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3576, pruned_loss=0.1089, over 5664867.53 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.368, pruned_loss=0.1171, over 5731373.73 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3576, pruned_loss=0.1082, over 5669217.00 frames. ], batch size: 705, lr: 7.93e-03, grad_scale: 8.0
+2023-03-02 03:56:07,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.740e+02 1.273e+03 1.710e+03 2.361e+03 9.368e+03, threshold=3.421e+03, percent-clipped=7.0
+2023-03-02 03:56:30,316 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:56:32,280 INFO [train.py:968] (0/2) Epoch 4, batch 14850, libri_loss[loss=0.336, simple_loss=0.3922, pruned_loss=0.1399, over 29687.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3586, pruned_loss=0.1102, over 5675408.74 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3674, pruned_loss=0.1167, over 5735202.92 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3588, pruned_loss=0.1096, over 5673399.11 frames. ], batch size: 91, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:56:32,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:56:39,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4454, 3.4439, 1.5102, 1.4899], device='cuda:0'), covar=tensor([0.0874, 0.0315, 0.0889, 0.1340], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0459, 0.0312, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 03:56:51,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8260, 3.6041, 3.5256, 1.6844], device='cuda:0'), covar=tensor([0.0469, 0.0465, 0.0874, 0.2272], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0674, 0.0778, 0.0585], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 03:57:10,464 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:57:36,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:57:36,781 INFO [train.py:968] (0/2) Epoch 4, batch 14900, giga_loss[loss=0.2705, simple_loss=0.3556, pruned_loss=0.09271, over 28996.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3599, pruned_loss=0.1108, over 5682439.41 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3669, pruned_loss=0.1165, over 5738749.12 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3603, pruned_loss=0.1104, over 5676467.67 frames. ], batch size: 136, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 03:58:16,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.893e+02 1.542e+03 2.023e+03 2.836e+03 8.449e+03, threshold=4.047e+03, percent-clipped=16.0
+2023-03-02 03:58:47,232 INFO [train.py:968] (0/2) Epoch 4, batch 14950, giga_loss[loss=0.2759, simple_loss=0.3601, pruned_loss=0.09581, over 28884.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3604, pruned_loss=0.1097, over 5684203.72 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3665, pruned_loss=0.1164, over 5742143.00 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3609, pruned_loss=0.1094, over 5675342.49 frames. ], batch size: 112, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 03:59:41,347 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:00:07,655 INFO [train.py:968] (0/2) Epoch 4, batch 15000, giga_loss[loss=0.2578, simple_loss=0.3343, pruned_loss=0.09069, over 28920.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3594, pruned_loss=0.1092, over 5665412.10 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3664, pruned_loss=0.1164, over 5735611.37 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3599, pruned_loss=0.1088, over 5662163.39 frames. ], batch size: 186, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 04:00:07,661 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 04:00:16,002 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.23, simple_loss=0.3254, pruned_loss=0.06734, over 944034.00 frames. 
+2023-03-02 04:00:16,002 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 04:00:56,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.907e+02 1.440e+03 1.838e+03 2.776e+03 6.663e+03, threshold=3.675e+03, percent-clipped=5.0
+2023-03-02 04:01:13,923 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:01:19,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:01:25,065 INFO [train.py:968] (0/2) Epoch 4, batch 15050, giga_loss[loss=0.2917, simple_loss=0.3418, pruned_loss=0.1208, over 26918.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3566, pruned_loss=0.1094, over 5662784.44 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3663, pruned_loss=0.1164, over 5737171.87 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.357, pruned_loss=0.109, over 5658064.80 frames. ], batch size: 555, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 04:01:58,962 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:02:29,526 INFO [train.py:968] (0/2) Epoch 4, batch 15100, giga_loss[loss=0.2563, simple_loss=0.3298, pruned_loss=0.09136, over 28092.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.349, pruned_loss=0.1052, over 5666959.28 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.366, pruned_loss=0.1163, over 5740669.22 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3492, pruned_loss=0.1048, over 5658139.71 frames. ], batch size: 412, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 04:02:58,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2672, 1.4876, 1.1033, 0.7970], device='cuda:0'), covar=tensor([0.0822, 0.0578, 0.0492, 0.0682], device='cuda:0'), in_proj_covar=tensor([0.1293, 0.0980, 0.1016, 0.1078], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 04:03:00,691 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 04:03:03,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.770e+02 1.517e+03 2.172e+03 2.732e+03 7.235e+03, threshold=4.344e+03, percent-clipped=10.0
+2023-03-02 04:03:22,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9122, 3.7125, 3.6048, 1.8931], device='cuda:0'), covar=tensor([0.0507, 0.0484, 0.0840, 0.2050], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0665, 0.0772, 0.0584], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 04:03:29,266 INFO [train.py:968] (0/2) Epoch 4, batch 15150, giga_loss[loss=0.2784, simple_loss=0.3466, pruned_loss=0.1051, over 28320.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.349, pruned_loss=0.1053, over 5669248.17 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3658, pruned_loss=0.1162, over 5743847.32 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3491, pruned_loss=0.1049, over 5657997.21 frames. ], batch size: 368, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 04:03:29,758 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:03:47,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:04:28,723 INFO [train.py:968] (0/2) Epoch 4, batch 15200, giga_loss[loss=0.297, simple_loss=0.3662, pruned_loss=0.1139, over 28735.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3521, pruned_loss=0.1079, over 5666697.09 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3656, pruned_loss=0.1161, over 5745708.76 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3522, pruned_loss=0.1076, over 5655531.60 frames. ], batch size: 262, lr: 7.92e-03, grad_scale: 8.0
+2023-03-02 04:04:45,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6888, 2.4651, 2.0341, 1.8620], device='cuda:0'), covar=tensor([0.1709, 0.1507, 0.1198, 0.1120], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0740, 0.0747, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 04:05:00,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.300e+02 1.406e+03 1.783e+03 2.448e+03 5.124e+03, threshold=3.565e+03, percent-clipped=2.0
+2023-03-02 04:05:26,687 INFO [train.py:968] (0/2) Epoch 4, batch 15250, giga_loss[loss=0.2675, simple_loss=0.3296, pruned_loss=0.1027, over 26777.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3491, pruned_loss=0.1054, over 5668820.23 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3654, pruned_loss=0.1161, over 5746942.47 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.349, pruned_loss=0.1048, over 5656546.79 frames. ], batch size: 555, lr: 7.92e-03, grad_scale: 8.0
+2023-03-02 04:06:19,280 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3534, 1.5045, 1.1145, 1.0176], device='cuda:0'), covar=tensor([0.0848, 0.0582, 0.0501, 0.0653], device='cuda:0'), in_proj_covar=tensor([0.1291, 0.0982, 0.1022, 0.1078], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001], device='cuda:0')
+2023-03-02 04:06:32,259 INFO [train.py:968] (0/2) Epoch 4, batch 15300, giga_loss[loss=0.2677, simple_loss=0.3457, pruned_loss=0.09483, over 28597.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3478, pruned_loss=0.1037, over 5665979.42 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3653, pruned_loss=0.1161, over 5746315.76 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3477, pruned_loss=0.1032, over 5656564.38 frames. ], batch size: 307, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:06:32,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2568, 1.3515, 1.1167, 1.4908], device='cuda:0'), covar=tensor([0.0803, 0.0389, 0.0379, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0214, 0.0136, 0.0142, 0.0254], device='cuda:0'), out_proj_covar=tensor([0.0049, 0.0036, 0.0033, 0.0054], device='cuda:0')
+2023-03-02 04:06:39,141 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:06:42,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:07:12,970 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.665e+02 1.223e+03 1.661e+03 2.439e+03 6.763e+03, threshold=3.322e+03, percent-clipped=5.0
+2023-03-02 04:07:24,459 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:07:40,744 INFO [train.py:968] (0/2) Epoch 4, batch 15350, libri_loss[loss=0.2782, simple_loss=0.3545, pruned_loss=0.101, over 29507.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3474, pruned_loss=0.104, over 5664991.77 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3653, pruned_loss=0.1161, over 5751418.86 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3467, pruned_loss=0.1032, over 5650157.98 frames. ], batch size: 84, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:07:51,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:08:49,312 INFO [train.py:968] (0/2) Epoch 4, batch 15400, giga_loss[loss=0.2628, simple_loss=0.3189, pruned_loss=0.1034, over 24331.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3478, pruned_loss=0.1041, over 5652483.85 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3649, pruned_loss=0.1159, over 5744702.76 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3473, pruned_loss=0.1035, over 5645910.34 frames. ], batch size: 705, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:09:22,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.911e+02 1.343e+03 1.903e+03 2.511e+03 5.943e+03, threshold=3.806e+03, percent-clipped=12.0
+2023-03-02 04:09:28,676 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:09:52,392 INFO [train.py:968] (0/2) Epoch 4, batch 15450, giga_loss[loss=0.3135, simple_loss=0.3761, pruned_loss=0.1255, over 28396.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3483, pruned_loss=0.1044, over 5648573.84 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3647, pruned_loss=0.1158, over 5739703.89 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3478, pruned_loss=0.1037, over 5645795.32 frames. ], batch size: 368, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:10:37,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3806, 2.4946, 1.4807, 1.5256], device='cuda:0'), covar=tensor([0.0686, 0.0355, 0.0706, 0.1052], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0453, 0.0312, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 04:10:59,343 INFO [train.py:968] (0/2) Epoch 4, batch 15500, giga_loss[loss=0.2905, simple_loss=0.3586, pruned_loss=0.1112, over 28472.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3495, pruned_loss=0.1059, over 5650373.47 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.365, pruned_loss=0.116, over 5740458.60 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3486, pruned_loss=0.105, over 5646358.45 frames. ], batch size: 336, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:11:00,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:11:02,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:11:28,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:11:32,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.198e+02 1.263e+03 1.627e+03 2.135e+03 4.935e+03, threshold=3.254e+03, percent-clipped=4.0
+2023-03-02 04:11:37,738 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:11:37,755 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:11:57,429 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:12:00,020 INFO [train.py:968] (0/2) Epoch 4, batch 15550, giga_loss[loss=0.3118, simple_loss=0.3702, pruned_loss=0.1268, over 28051.00 frames. ], tot_loss[loss=0.279, simple_loss=0.349, pruned_loss=0.1045, over 5661839.42 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3647, pruned_loss=0.1158, over 5743780.10 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3482, pruned_loss=0.1038, over 5654531.29 frames. ], batch size: 412, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:12:56,468 INFO [train.py:968] (0/2) Epoch 4, batch 15600, giga_loss[loss=0.3043, simple_loss=0.383, pruned_loss=0.1128, over 28878.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3532, pruned_loss=0.1057, over 5673440.42 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3645, pruned_loss=0.1157, over 5747570.18 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3524, pruned_loss=0.1049, over 5662177.92 frames. ], batch size: 213, lr: 7.91e-03, grad_scale: 8.0
+2023-03-02 04:13:27,979 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6394, 1.5937, 1.1801, 1.3393], device='cuda:0'), covar=tensor([0.0579, 0.0467, 0.0802, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0455, 0.0519, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 04:13:34,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.028e+02 1.528e+03 2.025e+03 2.646e+03 6.443e+03, threshold=4.049e+03, percent-clipped=13.0
+2023-03-02 04:13:58,773 INFO [train.py:968] (0/2) Epoch 4, batch 15650, giga_loss[loss=0.3181, simple_loss=0.3907, pruned_loss=0.1228, over 28943.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3554, pruned_loss=0.1068, over 5664775.68 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3648, pruned_loss=0.1159, over 5749792.23 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3543, pruned_loss=0.1059, over 5652908.82 frames. ], batch size: 213, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:14:21,069 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:14:23,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:14:32,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-02 04:14:53,616 INFO [train.py:968] (0/2) Epoch 4, batch 15700, libri_loss[loss=0.2575, simple_loss=0.3345, pruned_loss=0.09029, over 29535.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3556, pruned_loss=0.1064, over 5675540.45 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3641, pruned_loss=0.1153, over 5753727.24 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3551, pruned_loss=0.1059, over 5658188.62 frames. ], batch size: 80, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:14:54,064 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:14:59,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6549, 1.5094, 1.3082, 2.0778], device='cuda:0'), covar=tensor([0.2124, 0.2068, 0.1997, 0.1934], device='cuda:0'), in_proj_covar=tensor([0.1059, 0.0845, 0.0962, 0.0937], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 04:15:28,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.242e+02 1.509e+03 2.173e+03 3.372e+03 9.579e+03, threshold=4.346e+03, percent-clipped=17.0
+2023-03-02 04:15:53,850 INFO [train.py:968] (0/2) Epoch 4, batch 15750, libri_loss[loss=0.299, simple_loss=0.3649, pruned_loss=0.1165, over 29542.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3558, pruned_loss=0.1067, over 5688246.79 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.364, pruned_loss=0.1153, over 5756565.51 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3553, pruned_loss=0.1061, over 5670637.83 frames. ], batch size: 81, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:16:54,810 INFO [train.py:968] (0/2) Epoch 4, batch 15800, giga_loss[loss=0.2401, simple_loss=0.3219, pruned_loss=0.07915, over 29083.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3517, pruned_loss=0.1034, over 5691902.29 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3639, pruned_loss=0.1152, over 5757828.98 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3513, pruned_loss=0.1028, over 5676136.55 frames. ], batch size: 128, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:17:04,598 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:17:32,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.211e+02 1.333e+03 1.723e+03 2.164e+03 6.574e+03, threshold=3.446e+03, percent-clipped=4.0
+2023-03-02 04:17:53,360 INFO [train.py:968] (0/2) Epoch 4, batch 15850, giga_loss[loss=0.3068, simple_loss=0.3701, pruned_loss=0.1218, over 28872.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3513, pruned_loss=0.1034, over 5691374.20 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3639, pruned_loss=0.1152, over 5759532.10 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3506, pruned_loss=0.1027, over 5675870.25 frames. ], batch size: 174, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:18:02,476 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-152000.pt
+2023-03-02 04:18:22,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3219, 1.7681, 1.3658, 1.4750], device='cuda:0'), covar=tensor([0.0848, 0.0325, 0.0368, 0.0930], device='cuda:0'), in_proj_covar=tensor([0.0214, 0.0135, 0.0142, 0.0254], device='cuda:0'), out_proj_covar=tensor([0.0049, 0.0036, 0.0033, 0.0055], device='cuda:0')
+2023-03-02 04:18:50,823 INFO [train.py:968] (0/2) Epoch 4, batch 15900, giga_loss[loss=0.3393, simple_loss=0.3961, pruned_loss=0.1413, over 28699.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3492, pruned_loss=0.1031, over 5689144.09 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.363, pruned_loss=0.1146, over 5763172.23 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3491, pruned_loss=0.1027, over 5671056.22 frames. ], batch size: 262, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:19:00,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:19:22,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:19:29,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.034e+02 1.287e+03 1.553e+03 2.032e+03 6.302e+03, threshold=3.105e+03, percent-clipped=8.0
+2023-03-02 04:19:40,569 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:19:56,730 INFO [train.py:968] (0/2) Epoch 4, batch 15950, giga_loss[loss=0.2775, simple_loss=0.3489, pruned_loss=0.103, over 28910.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3514, pruned_loss=0.1045, over 5682618.47 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3629, pruned_loss=0.1145, over 5764196.53 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.351, pruned_loss=0.1041, over 5665771.30 frames. ], batch size: 136, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:19:59,372 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:20:04,264 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:20:23,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0128, 2.9654, 2.0911, 0.9163], device='cuda:0'), covar=tensor([0.2586, 0.1135, 0.1697, 0.2663], device='cuda:0'), in_proj_covar=tensor([0.1303, 0.1265, 0.1333, 0.1120], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 04:20:38,560 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:21:00,163 INFO [train.py:968] (0/2) Epoch 4, batch 16000, giga_loss[loss=0.2645, simple_loss=0.3427, pruned_loss=0.09309, over 28893.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3528, pruned_loss=0.1052, over 5676729.83 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3629, pruned_loss=0.1146, over 5753054.09 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3524, pruned_loss=0.1047, over 5672485.44 frames. ], batch size: 164, lr: 7.90e-03, grad_scale: 8.0
+2023-03-02 04:21:41,639 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.047e+02 1.305e+03 1.644e+03 2.313e+03 9.035e+03, threshold=3.289e+03, percent-clipped=8.0
+2023-03-02 04:21:59,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7020, 3.4963, 3.4175, 1.7067], device='cuda:0'), covar=tensor([0.0565, 0.0492, 0.0826, 0.2046], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0671, 0.0769, 0.0573], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 04:22:05,420 INFO [train.py:968] (0/2) Epoch 4, batch 16050, giga_loss[loss=0.2887, simple_loss=0.3618, pruned_loss=0.1079, over 28890.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3535, pruned_loss=0.1059, over 5671586.91 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3628, pruned_loss=0.1146, over 5755293.92 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3531, pruned_loss=0.1053, over 5664899.51 frames. ], batch size: 186, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:22:05,925 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:06,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:08,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:21,558 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:24,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:40,986 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:58,253 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:59,550 INFO [train.py:968] (0/2) Epoch 4, batch 16100, giga_loss[loss=0.3085, simple_loss=0.3818, pruned_loss=0.1176, over 28512.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3576, pruned_loss=0.1078, over 5680931.83 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3625, pruned_loss=0.1143, over 5756354.45 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3572, pruned_loss=0.1073, over 5672022.01 frames. ], batch size: 370, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:23:31,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.828e+02 1.360e+03 1.676e+03 2.290e+03 4.832e+03, threshold=3.352e+03, percent-clipped=6.0
+2023-03-02 04:23:44,995 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.35 vs. limit=5.0
+2023-03-02 04:23:52,753 INFO [train.py:968] (0/2) Epoch 4, batch 16150, giga_loss[loss=0.2976, simple_loss=0.3754, pruned_loss=0.1099, over 28529.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3598, pruned_loss=0.1083, over 5685949.24 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3626, pruned_loss=0.1145, over 5758121.41 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3593, pruned_loss=0.1075, over 5674262.71 frames. ], batch size: 370, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:25:01,458 INFO [train.py:968] (0/2) Epoch 4, batch 16200, giga_loss[loss=0.2925, simple_loss=0.3646, pruned_loss=0.1102, over 28400.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3584, pruned_loss=0.107, over 5687955.08 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3623, pruned_loss=0.1143, over 5759906.43 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3583, pruned_loss=0.1065, over 5676178.06 frames. ], batch size: 71, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:25:31,098 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-02 04:25:38,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.539e+02 1.343e+03 1.721e+03 2.326e+03 4.820e+03, threshold=3.442e+03, percent-clipped=7.0
+2023-03-02 04:26:05,527 INFO [train.py:968] (0/2) Epoch 4, batch 16250, giga_loss[loss=0.3032, simple_loss=0.3702, pruned_loss=0.1181, over 28692.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3559, pruned_loss=0.106, over 5696960.77 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5763133.26 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3561, pruned_loss=0.1057, over 5682815.76 frames. ], batch size: 262, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:26:30,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8244, 2.1953, 1.6766, 1.8904], device='cuda:0'), covar=tensor([0.0773, 0.0246, 0.0330, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0214, 0.0135, 0.0142, 0.0254], device='cuda:0'), out_proj_covar=tensor([0.0049, 0.0036, 0.0033, 0.0055], device='cuda:0')
+2023-03-02 04:27:08,235 INFO [train.py:968] (0/2) Epoch 4, batch 16300, libri_loss[loss=0.2868, simple_loss=0.3617, pruned_loss=0.1059, over 29384.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3559, pruned_loss=0.1065, over 5693327.38 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3613, pruned_loss=0.1136, over 5765396.32 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3564, pruned_loss=0.1063, over 5677905.85 frames. ], batch size: 92, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:27:12,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9730, 2.6870, 1.6008, 1.2772], device='cuda:0'), covar=tensor([0.0983, 0.0493, 0.0622, 0.0842], device='cuda:0'), in_proj_covar=tensor([0.1294, 0.0963, 0.0994, 0.1081], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 04:27:20,876 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:27:40,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.476e+02 1.216e+03 1.654e+03 2.093e+03 6.065e+03, threshold=3.307e+03, percent-clipped=8.0
+2023-03-02 04:28:06,033 INFO [train.py:968] (0/2) Epoch 4, batch 16350, giga_loss[loss=0.2777, simple_loss=0.3506, pruned_loss=0.1024, over 28523.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3542, pruned_loss=0.1062, over 5674463.99 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3606, pruned_loss=0.1133, over 5760323.03 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3549, pruned_loss=0.1061, over 5663605.22 frames. ], batch size: 336, lr: 7.89e-03, grad_scale: 4.0
+2023-03-02 04:28:15,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9720, 1.1354, 0.9112, 0.6003], device='cuda:0'), covar=tensor([0.0675, 0.0572, 0.0467, 0.0647], device='cuda:0'), in_proj_covar=tensor([0.1306, 0.0976, 0.1006, 0.1094], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 04:29:09,226 INFO [train.py:968] (0/2) Epoch 4, batch 16400, giga_loss[loss=0.2529, simple_loss=0.3289, pruned_loss=0.0885, over 28992.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3524, pruned_loss=0.1063, over 5676630.09 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1133, over 5759980.97 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.353, pruned_loss=0.1061, over 5666702.45 frames. ], batch size: 136, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:29:41,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:29:46,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.902e+02 1.325e+03 1.733e+03 2.484e+03 4.567e+03, threshold=3.467e+03, percent-clipped=10.0
+2023-03-02 04:30:07,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3976, 1.7307, 1.3897, 0.9827], device='cuda:0'), covar=tensor([0.1047, 0.0606, 0.0537, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.1307, 0.0982, 0.1007, 0.1103], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 04:30:08,245 INFO [train.py:968] (0/2) Epoch 4, batch 16450, giga_loss[loss=0.2944, simple_loss=0.3734, pruned_loss=0.1077, over 28898.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3532, pruned_loss=0.1062, over 5674230.62 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3608, pruned_loss=0.1135, over 5751605.96 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3532, pruned_loss=0.1058, over 5671775.50 frames. ], batch size: 227, lr: 7.88e-03, grad_scale: 8.0
+2023-03-02 04:30:16,700 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:30:20,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:30:39,393 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:30:55,529 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:31:08,209 INFO [train.py:968] (0/2) Epoch 4, batch 16500, giga_loss[loss=0.2551, simple_loss=0.3321, pruned_loss=0.08907, over 28511.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3518, pruned_loss=0.1047, over 5672009.82 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3603, pruned_loss=0.1132, over 5756468.39 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.352, pruned_loss=0.1043, over 5663368.68 frames. ], batch size: 78, lr: 7.88e-03, grad_scale: 4.0
+2023-03-02 04:31:20,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5324, 2.6088, 1.5954, 1.4731], device='cuda:0'), covar=tensor([0.0676, 0.0334, 0.0670, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0453, 0.0311, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 04:31:34,936 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:31:46,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.140e+02 1.434e+03 1.822e+03 2.749e+03 6.767e+03, threshold=3.644e+03, percent-clipped=12.0
+2023-03-02 04:32:06,417 INFO [train.py:968] (0/2) Epoch 4, batch 16550, giga_loss[loss=0.258, simple_loss=0.3407, pruned_loss=0.08761, over 28957.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3532, pruned_loss=0.1042, over 5676011.92 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1133, over 5750119.18 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3533, pruned_loss=0.1036, over 5672432.59 frames. ], batch size: 186, lr: 7.88e-03, grad_scale: 4.0
+2023-03-02 04:32:30,413 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:32:30,442 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8207, 2.5111, 2.2327, 1.8929], device='cuda:0'), covar=tensor([0.1591, 0.1466, 0.1062, 0.1105], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0727, 0.0731, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 04:32:32,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:33:03,299 INFO [train.py:968] (0/2) Epoch 4, batch 16600, giga_loss[loss=0.2991, simple_loss=0.3727, pruned_loss=0.1127, over 28442.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3544, pruned_loss=0.1035, over 5654440.09 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3605, pruned_loss=0.1136, over 5732375.20 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3541, pruned_loss=0.1025, over 5665226.53 frames. ], batch size: 369, lr: 7.88e-03, grad_scale: 2.0
+2023-03-02 04:33:05,509 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:33:38,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.258e+02 1.436e+03 2.128e+03 3.019e+03 1.138e+04, threshold=4.256e+03, percent-clipped=17.0
+2023-03-02 04:33:57,410 INFO [train.py:968] (0/2) Epoch 4, batch 16650, giga_loss[loss=0.2663, simple_loss=0.3446, pruned_loss=0.09399, over 28911.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3543, pruned_loss=0.1027, over 5675579.38 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3599, pruned_loss=0.1132, over 5738384.42 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3544, pruned_loss=0.102, over 5676737.74 frames. ], batch size: 199, lr: 7.88e-03, grad_scale: 2.0
+2023-03-02 04:34:53,130 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:35:00,131 INFO [train.py:968] (0/2) Epoch 4, batch 16700, giga_loss[loss=0.2406, simple_loss=0.3067, pruned_loss=0.08724, over 24628.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3546, pruned_loss=0.1033, over 5681208.71 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3595, pruned_loss=0.1129, over 5743203.54 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3549, pruned_loss=0.1027, over 5676335.52 frames. ], batch size: 705, lr: 7.88e-03, grad_scale: 2.0
+2023-03-02 04:35:00,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2201, 1.2825, 1.1445, 1.2244], device='cuda:0'), covar=tensor([0.2019, 0.1868, 0.1836, 0.1791], device='cuda:0'), in_proj_covar=tensor([0.1052, 0.0840, 0.0952, 0.0934], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 04:35:48,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.551e+02 1.249e+03 1.568e+03 2.053e+03 9.205e+03, threshold=3.135e+03, percent-clipped=4.0
+2023-03-02 04:36:11,921 INFO [train.py:968] (0/2) Epoch 4, batch 16750, giga_loss[loss=0.3077, simple_loss=0.378, pruned_loss=0.1187, over 28058.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3541, pruned_loss=0.1027, over 5675438.37 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3593, pruned_loss=0.1128, over 5744057.52 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3545, pruned_loss=0.1022, over 5670506.71 frames. ], batch size: 412, lr: 7.88e-03, grad_scale: 2.0
+2023-03-02 04:37:21,342 INFO [train.py:968] (0/2) Epoch 4, batch 16800, giga_loss[loss=0.249, simple_loss=0.3387, pruned_loss=0.07963, over 28830.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.354, pruned_loss=0.1022, over 5668146.50 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3594, pruned_loss=0.113, over 5737858.01 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3541, pruned_loss=0.1015, over 5668966.29 frames. ], batch size: 99, lr: 7.88e-03, grad_scale: 4.0
+2023-03-02 04:38:05,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:38:05,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.591e+02 1.296e+03 1.716e+03 2.307e+03 4.283e+03, threshold=3.432e+03, percent-clipped=12.0
+2023-03-02 04:38:13,278 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:38:29,007 INFO [train.py:968] (0/2) Epoch 4, batch 16850, giga_loss[loss=0.2765, simple_loss=0.3536, pruned_loss=0.09966, over 28655.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3549, pruned_loss=0.1029, over 5675450.81 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3593, pruned_loss=0.113, over 5743399.88 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3549, pruned_loss=0.102, over 5669103.34 frames. ], batch size: 307, lr: 7.87e-03, grad_scale: 4.0
+2023-03-02 04:38:29,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:38:32,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5562, 1.9948, 1.8294, 1.6957], device='cuda:0'), covar=tensor([0.1517, 0.1667, 0.1170, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0771, 0.0727, 0.0734, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 04:39:32,579 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:39:34,373 INFO [train.py:968] (0/2) Epoch 4, batch 16900, giga_loss[loss=0.274, simple_loss=0.3531, pruned_loss=0.09748, over 29151.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3594, pruned_loss=0.1053, over 5683432.51 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3592, pruned_loss=0.1129, over 5746501.40 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3595, pruned_loss=0.1044, over 5674029.14 frames. ], batch size: 100, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:40:20,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.277e+02 1.406e+03 2.220e+03 2.874e+03 1.151e+04, threshold=4.441e+03, percent-clipped=13.0
+2023-03-02 04:40:30,419 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-02 04:40:43,346 INFO [train.py:968] (0/2) Epoch 4, batch 16950, giga_loss[loss=0.2519, simple_loss=0.3317, pruned_loss=0.08599, over 28784.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3588, pruned_loss=0.1048, over 5683559.72 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3592, pruned_loss=0.1128, over 5749413.87 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3588, pruned_loss=0.104, over 5672331.71 frames. ], batch size: 243, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:40:54,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0381, 1.1828, 1.2757, 1.0277], device='cuda:0'), covar=tensor([0.0799, 0.1072, 0.1421, 0.1221], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0736, 0.0609, 0.0633], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-02 04:41:17,874 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:41:36,845 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:41:41,634 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:41:45,025 INFO [train.py:968] (0/2) Epoch 4, batch 17000, giga_loss[loss=0.2638, simple_loss=0.3506, pruned_loss=0.08848, over 29154.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3561, pruned_loss=0.1037, over 5697216.05 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3591, pruned_loss=0.1127, over 5752641.93 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3561, pruned_loss=0.1029, over 5683331.67 frames. ], batch size: 200, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:41:52,142 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-02 04:42:24,807 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:42:30,406 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0141, 1.3170, 2.9627, 2.8328], device='cuda:0'), covar=tensor([0.1854, 0.2353, 0.0775, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0555, 0.0515, 0.0704, 0.0561], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 04:42:34,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.601e+02 1.217e+03 1.651e+03 2.227e+03 4.378e+03, threshold=3.302e+03, percent-clipped=0.0
+2023-03-02 04:42:44,394 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:42:49,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:42:59,243 INFO [train.py:968] (0/2) Epoch 4, batch 17050, giga_loss[loss=0.2602, simple_loss=0.3448, pruned_loss=0.08783, over 28915.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3533, pruned_loss=0.1018, over 5696152.54 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3588, pruned_loss=0.1125, over 5753703.01 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3536, pruned_loss=0.1013, over 5683551.48 frames. ], batch size: 227, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:43:25,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:43:29,757 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:44:02,960 INFO [train.py:968] (0/2) Epoch 4, batch 17100, giga_loss[loss=0.2482, simple_loss=0.3299, pruned_loss=0.08331, over 28798.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.351, pruned_loss=0.09978, over 5707113.01 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3581, pruned_loss=0.1122, over 5755522.06 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3517, pruned_loss=0.09927, over 5693846.52 frames. ], batch size: 119, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:44:42,155 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.430e+02 1.350e+03 1.923e+03 2.732e+03 1.059e+04, threshold=3.847e+03, percent-clipped=16.0
+2023-03-02 04:44:57,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3920, 2.7915, 1.2786, 1.4810], device='cuda:0'), covar=tensor([0.0915, 0.0420, 0.0930, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0454, 0.0311, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 04:45:00,670 INFO [train.py:968] (0/2) Epoch 4, batch 17150, giga_loss[loss=0.2984, simple_loss=0.3608, pruned_loss=0.118, over 28955.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3524, pruned_loss=0.1015, over 5689129.04 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3578, pruned_loss=0.112, over 5749802.96 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.353, pruned_loss=0.1008, over 5682368.60 frames. ], batch size: 186, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:45:20,559 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 04:45:55,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6533, 1.6346, 1.6258, 1.6109], device='cuda:0'), covar=tensor([0.0792, 0.1502, 0.1170, 0.1119], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0744, 0.0621, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 04:45:58,604 INFO [train.py:968] (0/2) Epoch 4, batch 17200, giga_loss[loss=0.3204, simple_loss=0.3938, pruned_loss=0.1235, over 28645.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3556, pruned_loss=0.1034, over 5691386.22 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3574, pruned_loss=0.1119, over 5752212.70 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3563, pruned_loss=0.1029, over 5683264.55 frames. ], batch size: 242, lr: 7.87e-03, grad_scale: 4.0
+2023-03-02 04:46:09,184 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:46:15,021 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:46:15,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:46:17,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:46:36,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.691e+02 1.536e+03 2.111e+03 2.835e+03 6.355e+03, threshold=4.222e+03, percent-clipped=10.0
+2023-03-02 04:46:47,876 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:46:51,497 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-02 04:46:51,595 INFO [train.py:968] (0/2) Epoch 4, batch 17250, giga_loss[loss=0.2572, simple_loss=0.3366, pruned_loss=0.08892, over 28675.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3573, pruned_loss=0.1052, over 5691353.61 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3575, pruned_loss=0.1119, over 5754927.55 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3579, pruned_loss=0.1045, over 5680305.07 frames. ], batch size: 262, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:47:42,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9887, 1.9248, 1.3516, 1.6313], device='cuda:0'), covar=tensor([0.0595, 0.0594, 0.0948, 0.0928], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0457, 0.0517, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 04:47:51,935 INFO [train.py:968] (0/2) Epoch 4, batch 17300, giga_loss[loss=0.2661, simple_loss=0.3356, pruned_loss=0.09827, over 28628.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3553, pruned_loss=0.1053, over 5676244.89 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3575, pruned_loss=0.1119, over 5746756.41 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3557, pruned_loss=0.1046, over 5674838.62 frames. ], batch size: 307, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:48:32,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.005e+02 1.294e+03 1.663e+03 2.147e+03 4.262e+03, threshold=3.325e+03, percent-clipped=1.0
+2023-03-02 04:48:49,893 INFO [train.py:968] (0/2) Epoch 4, batch 17350, libri_loss[loss=0.2889, simple_loss=0.3578, pruned_loss=0.11, over 29543.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3555, pruned_loss=0.106, over 5682105.73 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3573, pruned_loss=0.1118, over 5749979.88 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3559, pruned_loss=0.1055, over 5675937.37 frames. ], batch size: 80, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:48:50,261 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:48:50,903 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:48:52,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:48:58,328 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:49:00,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:49:08,768 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:49:28,066 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:49:34,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:49:45,583 INFO [train.py:968] (0/2) Epoch 4, batch 17400, giga_loss[loss=0.3481, simple_loss=0.4081, pruned_loss=0.1441, over 27924.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3616, pruned_loss=0.1108, over 5682801.60 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3574, pruned_loss=0.1118, over 5751281.64 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3619, pruned_loss=0.1103, over 5676186.66 frames. ], batch size: 412, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:50:22,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.920e+02 1.350e+03 1.737e+03 2.553e+03 6.249e+03, threshold=3.474e+03, percent-clipped=14.0
+2023-03-02 04:50:38,323 INFO [train.py:968] (0/2) Epoch 4, batch 17450, giga_loss[loss=0.3604, simple_loss=0.4195, pruned_loss=0.1507, over 27964.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3741, pruned_loss=0.1193, over 5686793.58 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3576, pruned_loss=0.1119, over 5752171.23 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3742, pruned_loss=0.1189, over 5680417.46 frames. ], batch size: 412, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:51:14,125 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:51:16,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:51:19,359 INFO [train.py:968] (0/2) Epoch 4, batch 17500, giga_loss[loss=0.3031, simple_loss=0.3713, pruned_loss=0.1174, over 28289.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3794, pruned_loss=0.1228, over 5698820.34 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3576, pruned_loss=0.1118, over 5756219.92 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3801, pruned_loss=0.1228, over 5688348.19 frames. ], batch size: 368, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:51:41,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:51:42,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:51:47,442 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 04:51:49,990 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.571e+02 1.229e+03 1.495e+03 1.972e+03 3.725e+03, threshold=2.989e+03, percent-clipped=3.0
+2023-03-02 04:52:05,427 INFO [train.py:968] (0/2) Epoch 4, batch 17550, giga_loss[loss=0.3881, simple_loss=0.409, pruned_loss=0.1836, over 26812.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3753, pruned_loss=0.1216, over 5695988.73 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3578, pruned_loss=0.1118, over 5758598.50 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.376, pruned_loss=0.1217, over 5684945.28 frames. ], batch size: 555, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:52:49,337 INFO [train.py:968] (0/2) Epoch 4, batch 17600, giga_loss[loss=0.2884, simple_loss=0.3491, pruned_loss=0.1138, over 28788.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3675, pruned_loss=0.1177, over 5687999.60 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3584, pruned_loss=0.1121, over 5755306.44 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3677, pruned_loss=0.1178, over 5680665.21 frames. ], batch size: 243, lr: 7.85e-03, grad_scale: 8.0
+2023-03-02 04:53:01,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0661, 1.4171, 3.1316, 2.8104], device='cuda:0'), covar=tensor([0.1622, 0.2173, 0.0659, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0543, 0.0502, 0.0691, 0.0556], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 04:53:18,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.048e+02 1.058e+03 1.311e+03 1.798e+03 3.507e+03, threshold=2.622e+03, percent-clipped=2.0
+2023-03-02 04:53:29,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8367, 3.5819, 3.5613, 1.5998], device='cuda:0'), covar=tensor([0.0524, 0.0476, 0.0820, 0.2210], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0673, 0.0769, 0.0584], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 04:53:32,598 INFO [train.py:968] (0/2) Epoch 4, batch 17650, giga_loss[loss=0.2698, simple_loss=0.3343, pruned_loss=0.1027, over 28753.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3595, pruned_loss=0.1141, over 5691091.31 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3588, pruned_loss=0.1122, over 5758501.06 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3595, pruned_loss=0.1141, over 5681019.35 frames. ], batch size: 262, lr: 7.85e-03, grad_scale: 8.0
+2023-03-02 04:53:52,848 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:53:59,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 04:54:19,125 INFO [train.py:968] (0/2) Epoch 4, batch 17700, libri_loss[loss=0.3369, simple_loss=0.4036, pruned_loss=0.1351, over 28596.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3519, pruned_loss=0.1105, over 5688851.43 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5761842.48 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3514, pruned_loss=0.1103, over 5675534.50 frames. ], batch size: 106, lr: 7.85e-03, grad_scale: 8.0
+2023-03-02 04:54:46,206 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.080e+02 1.054e+03 1.487e+03 2.135e+03 6.022e+03, threshold=2.975e+03, percent-clipped=17.0
+2023-03-02 04:54:53,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:55:01,707 INFO [train.py:968] (0/2) Epoch 4, batch 17750, giga_loss[loss=0.2426, simple_loss=0.3079, pruned_loss=0.08862, over 28003.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3443, pruned_loss=0.1067, over 5694093.65 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.359, pruned_loss=0.1122, over 5763903.71 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3439, pruned_loss=0.1066, over 5680823.39 frames. ], batch size: 412, lr: 7.85e-03, grad_scale: 8.0
+2023-03-02 04:55:43,014 INFO [train.py:968] (0/2) Epoch 4, batch 17800, giga_loss[loss=0.2314, simple_loss=0.3077, pruned_loss=0.07758, over 28914.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3392, pruned_loss=0.1036, over 5701230.30 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3592, pruned_loss=0.1122, over 5766480.88 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3383, pruned_loss=0.1034, over 5687191.49 frames. ], batch size: 186, lr: 7.85e-03, grad_scale: 4.0
+2023-03-02 04:56:11,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.979e+02 1.035e+03 1.355e+03 1.820e+03 4.835e+03, threshold=2.710e+03, percent-clipped=8.0
+2023-03-02 04:56:17,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1614, 2.7384, 2.3399, 2.1571], device='cuda:0'), covar=tensor([0.1441, 0.1328, 0.1020, 0.0978], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0756, 0.0756, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 04:56:25,867 INFO [train.py:968] (0/2) Epoch 4, batch 17850, libri_loss[loss=0.2532, simple_loss=0.3158, pruned_loss=0.09529, over 29381.00 frames. ], tot_loss[loss=0.27, simple_loss=0.336, pruned_loss=0.102, over 5707879.21 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3597, pruned_loss=0.1124, over 5770024.96 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3344, pruned_loss=0.1014, over 5692020.27 frames. ], batch size: 67, lr: 7.85e-03, grad_scale: 4.0
+2023-03-02 04:56:31,789 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-154000.pt
+2023-03-02 04:56:56,493 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:56:58,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:57:07,696 INFO [train.py:968] (0/2) Epoch 4, batch 17900, libri_loss[loss=0.2787, simple_loss=0.3443, pruned_loss=0.1066, over 29644.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3328, pruned_loss=0.1005, over 5705820.44 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3599, pruned_loss=0.1125, over 5773908.90 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3306, pruned_loss=0.09964, over 5687585.70 frames. ], batch size: 69, lr: 7.85e-03, grad_scale: 4.0
+2023-03-02 04:57:10,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:57:21,077 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:57:34,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.964e+02 9.806e+02 1.349e+03 1.955e+03 4.793e+03, threshold=2.697e+03, percent-clipped=9.0
+2023-03-02 04:57:38,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7741, 4.3436, 1.9640, 1.7250], device='cuda:0'), covar=tensor([0.0848, 0.0264, 0.0806, 0.1305], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0454, 0.0306, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 04:57:47,426 INFO [train.py:968] (0/2) Epoch 4, batch 17950, giga_loss[loss=0.2243, simple_loss=0.2836, pruned_loss=0.08252, over 23779.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3293, pruned_loss=0.09826, over 5703260.46 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1125, over 5765481.80 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3266, pruned_loss=0.09716, over 5693396.45 frames. ], batch size: 705, lr: 7.85e-03, grad_scale: 4.0
+2023-03-02 04:58:30,657 INFO [train.py:968] (0/2) Epoch 4, batch 18000, giga_loss[loss=0.2149, simple_loss=0.2909, pruned_loss=0.06944, over 28969.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3271, pruned_loss=0.09711, over 5693828.39 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3604, pruned_loss=0.1126, over 5767770.33 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3236, pruned_loss=0.09578, over 5682232.67 frames. ], batch size: 213, lr: 7.84e-03, grad_scale: 8.0
+2023-03-02 04:58:30,662 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 04:58:39,397 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2448, simple_loss=0.3452, pruned_loss=0.0722, over 944034.00 frames. 
+2023-03-02 04:58:39,398 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 04:59:08,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.063e+02 9.879e+02 1.277e+03 1.675e+03 6.120e+03, threshold=2.553e+03, percent-clipped=11.0
+2023-03-02 04:59:17,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:59:17,929 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:59:20,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:59:20,985 INFO [train.py:968] (0/2) Epoch 4, batch 18050, giga_loss[loss=0.2272, simple_loss=0.2959, pruned_loss=0.0792, over 28361.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3241, pruned_loss=0.09587, over 5695630.14 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3611, pruned_loss=0.1129, over 5766765.85 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3202, pruned_loss=0.0943, over 5686135.93 frames. ], batch size: 65, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 04:59:25,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4598, 1.3322, 5.5963, 3.8990], device='cuda:0'), covar=tensor([0.1626, 0.2263, 0.0253, 0.0477], device='cuda:0'), in_proj_covar=tensor([0.0547, 0.0509, 0.0707, 0.0563], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 04:59:47,003 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:00:04,884 INFO [train.py:968] (0/2) Epoch 4, batch 18100, giga_loss[loss=0.2128, simple_loss=0.2852, pruned_loss=0.07014, over 28953.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3199, pruned_loss=0.09367, over 5696151.60 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.361, pruned_loss=0.1127, over 5769362.66 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.316, pruned_loss=0.09216, over 5684532.66 frames. ], batch size: 227, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:00:37,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 4.891e+02 9.061e+02 1.188e+03 1.777e+03 5.737e+03, threshold=2.375e+03, percent-clipped=8.0
+2023-03-02 05:00:51,937 INFO [train.py:968] (0/2) Epoch 4, batch 18150, giga_loss[loss=0.2393, simple_loss=0.3094, pruned_loss=0.0846, over 28622.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3172, pruned_loss=0.09245, over 5688354.79 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3612, pruned_loss=0.1128, over 5770789.42 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3132, pruned_loss=0.09081, over 5676689.89 frames. ], batch size: 242, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:01:01,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3864, 1.4171, 1.2946, 1.4976], device='cuda:0'), covar=tensor([0.0799, 0.0350, 0.0360, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0208, 0.0133, 0.0139, 0.0246], device='cuda:0'), out_proj_covar=tensor([0.0049, 0.0036, 0.0032, 0.0054], device='cuda:0')
+2023-03-02 05:01:15,214 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:01:30,803 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:01:32,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:01:37,926 INFO [train.py:968] (0/2) Epoch 4, batch 18200, giga_loss[loss=0.2968, simple_loss=0.3516, pruned_loss=0.121, over 28801.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3155, pruned_loss=0.09196, over 5679790.28 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3616, pruned_loss=0.1132, over 5761746.71 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3113, pruned_loss=0.09004, over 5678321.04 frames. ], batch size: 199, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:02:00,724 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:02:07,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1989, 2.8457, 1.2975, 1.1558], device='cuda:0'), covar=tensor([0.1250, 0.0482, 0.1116, 0.1850], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0458, 0.0309, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 05:02:13,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.865e+02 1.050e+03 1.374e+03 1.767e+03 4.849e+03, threshold=2.748e+03, percent-clipped=13.0
+2023-03-02 05:02:27,743 INFO [train.py:968] (0/2) Epoch 4, batch 18250, giga_loss[loss=0.3101, simple_loss=0.3751, pruned_loss=0.1225, over 28604.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3278, pruned_loss=0.0995, over 5676760.76 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.362, pruned_loss=0.1134, over 5765480.18 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3227, pruned_loss=0.09717, over 5669302.42 frames. ], batch size: 78, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:03:13,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154442.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:03:15,218 INFO [train.py:968] (0/2) Epoch 4, batch 18300, giga_loss[loss=0.3726, simple_loss=0.4244, pruned_loss=0.1604, over 28979.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3435, pruned_loss=0.1086, over 5675768.19 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3624, pruned_loss=0.1136, over 5758396.41 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3388, pruned_loss=0.1064, over 5675036.90 frames. ], batch size: 136, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:03:26,180 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-02 05:03:35,875 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:03:42,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:03:44,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.657e+02 1.273e+03 1.578e+03 1.991e+03 6.903e+03, threshold=3.156e+03, percent-clipped=9.0
+2023-03-02 05:03:54,783 INFO [train.py:968] (0/2) Epoch 4, batch 18350, giga_loss[loss=0.294, simple_loss=0.3716, pruned_loss=0.1082, over 28827.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3555, pruned_loss=0.115, over 5685457.08 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3624, pruned_loss=0.1137, over 5759479.18 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3516, pruned_loss=0.1132, over 5682786.19 frames. ], batch size: 174, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:04:22,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2815, 1.4431, 1.2644, 1.5403], device='cuda:0'), covar=tensor([0.2212, 0.2052, 0.1948, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.1067, 0.0853, 0.0954, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 05:04:40,348 INFO [train.py:968] (0/2) Epoch 4, batch 18400, giga_loss[loss=0.3575, simple_loss=0.4188, pruned_loss=0.1481, over 27966.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3632, pruned_loss=0.1184, over 5688724.45 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3628, pruned_loss=0.1138, over 5762654.44 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3597, pruned_loss=0.117, over 5681950.55 frames. ], batch size: 412, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:04:53,702 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7225, 2.1419, 1.9586, 1.8381], device='cuda:0'), covar=tensor([0.1408, 0.1676, 0.1094, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0749, 0.0749, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 05:05:08,211 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.747e+02 1.209e+03 1.804e+03 2.098e+03 5.132e+03, threshold=3.609e+03, percent-clipped=8.0
+2023-03-02 05:05:19,928 INFO [train.py:968] (0/2) Epoch 4, batch 18450, giga_loss[loss=0.3295, simple_loss=0.3972, pruned_loss=0.1309, over 28672.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3655, pruned_loss=0.1177, over 5687377.54 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.364, pruned_loss=0.1145, over 5759774.98 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3617, pruned_loss=0.1161, over 5682214.57 frames. ], batch size: 262, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:05:41,579 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:06:07,952 INFO [train.py:968] (0/2) Epoch 4, batch 18500, giga_loss[loss=0.3502, simple_loss=0.3797, pruned_loss=0.1604, over 23546.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3665, pruned_loss=0.1173, over 5670617.54 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3638, pruned_loss=0.1142, over 5760071.82 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3636, pruned_loss=0.1163, over 5664954.82 frames. ], batch size: 705, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:06:39,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.127e+02 1.070e+03 1.388e+03 1.749e+03 4.706e+03, threshold=2.776e+03, percent-clipped=2.0
+2023-03-02 05:06:39,313 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:06:42,725 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:06:47,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:06:48,708 INFO [train.py:968] (0/2) Epoch 4, batch 18550, giga_loss[loss=0.3244, simple_loss=0.3881, pruned_loss=0.1303, over 28951.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3697, pruned_loss=0.1198, over 5680910.66 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3647, pruned_loss=0.1146, over 5762146.57 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3668, pruned_loss=0.1188, over 5671257.89 frames. ], batch size: 106, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:07:23,560 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154730.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:07:33,557 INFO [train.py:968] (0/2) Epoch 4, batch 18600, giga_loss[loss=0.4027, simple_loss=0.4262, pruned_loss=0.1896, over 26485.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3744, pruned_loss=0.1239, over 5674727.68 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3656, pruned_loss=0.1151, over 5755948.52 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3714, pruned_loss=0.1228, over 5670102.21 frames. ], batch size: 555, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:08:01,973 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6428, 3.7975, 1.7775, 1.5463], device='cuda:0'), covar=tensor([0.0843, 0.0204, 0.0739, 0.1322], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0453, 0.0304, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 05:08:03,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2397, 1.6535, 1.2706, 1.4256], device='cuda:0'), covar=tensor([0.0849, 0.0359, 0.0365, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0208, 0.0133, 0.0138, 0.0247], device='cuda:0'), out_proj_covar=tensor([0.0049, 0.0036, 0.0032, 0.0054], device='cuda:0')
+2023-03-02 05:08:04,022 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 05:08:04,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.455e+02 1.221e+03 1.567e+03 2.111e+03 5.185e+03, threshold=3.133e+03, percent-clipped=5.0
+2023-03-02 05:08:18,251 INFO [train.py:968] (0/2) Epoch 4, batch 18650, giga_loss[loss=0.2958, simple_loss=0.3727, pruned_loss=0.1094, over 28920.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3766, pruned_loss=0.1251, over 5676663.75 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3652, pruned_loss=0.1148, over 5758895.18 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3747, pruned_loss=0.1247, over 5668590.72 frames. ], batch size: 145, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:08:38,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154817.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:08:54,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:08:57,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:09:00,828 INFO [train.py:968] (0/2) Epoch 4, batch 18700, giga_loss[loss=0.3395, simple_loss=0.4084, pruned_loss=0.1353, over 28938.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.379, pruned_loss=0.1253, over 5679026.55 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3654, pruned_loss=0.1148, over 5756026.70 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3774, pruned_loss=0.125, over 5674824.42 frames. ], batch size: 213, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:09:01,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:09:08,402 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:09:21,795 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:09:22,340 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:09:26,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6720, 1.9398, 1.8049, 1.9356], device='cuda:0'), covar=tensor([0.0783, 0.0276, 0.0301, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0210, 0.0133, 0.0138, 0.0248], device='cuda:0'), out_proj_covar=tensor([0.0049, 0.0036, 0.0032, 0.0054], device='cuda:0')
+2023-03-02 05:09:32,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.169e+02 1.177e+03 1.425e+03 1.821e+03 3.475e+03, threshold=2.850e+03, percent-clipped=4.0
+2023-03-02 05:09:44,929 INFO [train.py:968] (0/2) Epoch 4, batch 18750, libri_loss[loss=0.3158, simple_loss=0.3832, pruned_loss=0.1242, over 29663.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3817, pruned_loss=0.1265, over 5672624.69 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.366, pruned_loss=0.1152, over 5749333.04 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3803, pruned_loss=0.1262, over 5673700.99 frames. ], batch size: 91, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:10:24,823 INFO [train.py:968] (0/2) Epoch 4, batch 18800, giga_loss[loss=0.3045, simple_loss=0.38, pruned_loss=0.1145, over 28279.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3821, pruned_loss=0.1255, over 5680630.08 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3664, pruned_loss=0.1152, over 5750813.67 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3809, pruned_loss=0.1254, over 5678713.57 frames. ], batch size: 368, lr: 7.82e-03, grad_scale: 8.0
+2023-03-02 05:10:38,114 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154960.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:10:40,094 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154963.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:10:52,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.801e+02 1.193e+03 1.439e+03 1.749e+03 3.397e+03, threshold=2.879e+03, percent-clipped=4.0
+2023-03-02 05:10:57,351 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:10:57,531 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 05:10:58,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:10:59,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:00,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154992.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:11:01,432 INFO [train.py:968] (0/2) Epoch 4, batch 18850, giga_loss[loss=0.303, simple_loss=0.3762, pruned_loss=0.1149, over 28610.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3821, pruned_loss=0.1242, over 5682897.51 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3671, pruned_loss=0.1155, over 5739571.16 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3811, pruned_loss=0.1242, over 5689273.54 frames. ], batch size: 85, lr: 7.82e-03, grad_scale: 8.0
+2023-03-02 05:11:03,407 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:05,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:24,527 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:25,880 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:30,007 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:43,149 INFO [train.py:968] (0/2) Epoch 4, batch 18900, giga_loss[loss=0.3148, simple_loss=0.3833, pruned_loss=0.1231, over 28916.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3802, pruned_loss=0.1216, over 5696268.70 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3677, pruned_loss=0.1158, over 5743471.58 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3791, pruned_loss=0.1215, over 5696789.21 frames. ], batch size: 186, lr: 7.82e-03, grad_scale: 8.0
+2023-03-02 05:11:51,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:55,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:12:13,132 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.740e+02 9.774e+02 1.211e+03 1.584e+03 4.786e+03, threshold=2.421e+03, percent-clipped=4.0
+2023-03-02 05:12:22,402 INFO [train.py:968] (0/2) Epoch 4, batch 18950, giga_loss[loss=0.3327, simple_loss=0.3975, pruned_loss=0.134, over 28696.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3795, pruned_loss=0.1212, over 5699365.02 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3681, pruned_loss=0.116, over 5743951.07 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3784, pruned_loss=0.121, over 5698918.79 frames. ], batch size: 262, lr: 7.82e-03, grad_scale: 8.0
+2023-03-02 05:12:31,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155105.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:12:38,472 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:12:55,144 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:12:57,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:13:02,521 INFO [train.py:968] (0/2) Epoch 4, batch 19000, giga_loss[loss=0.3108, simple_loss=0.3798, pruned_loss=0.1209, over 28904.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3821, pruned_loss=0.1248, over 5707360.87 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.369, pruned_loss=0.1164, over 5747233.46 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3811, pruned_loss=0.1246, over 5702202.56 frames. ], batch size: 145, lr: 7.82e-03, grad_scale: 4.0
+2023-03-02 05:13:21,064 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:13:34,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.969e+02 1.344e+03 1.667e+03 2.660e+03 8.553e+03, threshold=3.333e+03, percent-clipped=29.0
+2023-03-02 05:13:44,256 INFO [train.py:968] (0/2) Epoch 4, batch 19050, giga_loss[loss=0.4204, simple_loss=0.4417, pruned_loss=0.1996, over 27900.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3854, pruned_loss=0.1299, over 5712643.05 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3696, pruned_loss=0.1167, over 5753270.71 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3846, pruned_loss=0.13, over 5701424.18 frames. ], batch size: 412, lr: 7.82e-03, grad_scale: 4.0
+2023-03-02 05:13:47,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:13:50,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:13:51,414 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:13:54,065 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:14:13,332 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:14:17,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:14:23,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:14:23,991 INFO [train.py:968] (0/2) Epoch 4, batch 19100, giga_loss[loss=0.2944, simple_loss=0.3534, pruned_loss=0.1177, over 28952.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3857, pruned_loss=0.1316, over 5717663.17 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3704, pruned_loss=0.1172, over 5757704.65 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3848, pruned_loss=0.1316, over 5703823.61 frames. ], batch size: 106, lr: 7.82e-03, grad_scale: 4.0
+2023-03-02 05:14:27,902 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155248.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:14:29,745 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155251.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:14:55,772 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155280.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:14:56,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.628e+02 1.280e+03 1.501e+03 2.061e+03 6.113e+03, threshold=3.001e+03, percent-clipped=4.0
+2023-03-02 05:15:06,395 INFO [train.py:968] (0/2) Epoch 4, batch 19150, giga_loss[loss=0.3297, simple_loss=0.4064, pruned_loss=0.1265, over 28249.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3843, pruned_loss=0.1316, over 5712092.51 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3709, pruned_loss=0.1175, over 5757148.94 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3834, pruned_loss=0.1316, over 5700623.78 frames. ], batch size: 77, lr: 7.82e-03, grad_scale: 4.0
+2023-03-02 05:15:49,368 INFO [train.py:968] (0/2) Epoch 4, batch 19200, giga_loss[loss=0.3222, simple_loss=0.3793, pruned_loss=0.1326, over 28963.00 frames. ], tot_loss[loss=0.32, simple_loss=0.381, pruned_loss=0.1295, over 5709596.09 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3709, pruned_loss=0.1174, over 5759750.61 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3806, pruned_loss=0.13, over 5696811.49 frames. ], batch size: 136, lr: 7.81e-03, grad_scale: 8.0
+2023-03-02 05:16:12,844 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:16:21,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.982e+02 1.167e+03 1.570e+03 1.990e+03 1.129e+04, threshold=3.140e+03, percent-clipped=11.0
+2023-03-02 05:16:25,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8714, 1.2202, 3.3125, 2.8520], device='cuda:0'), covar=tensor([0.1501, 0.1991, 0.0361, 0.1212], device='cuda:0'), in_proj_covar=tensor([0.0547, 0.0505, 0.0701, 0.0567], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 05:16:26,443 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:16:30,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:16:33,314 INFO [train.py:968] (0/2) Epoch 4, batch 19250, giga_loss[loss=0.2972, simple_loss=0.3679, pruned_loss=0.1133, over 28852.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3798, pruned_loss=0.1277, over 5717359.37 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3714, pruned_loss=0.1175, over 5761981.84 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3792, pruned_loss=0.1282, over 5704267.82 frames. ], batch size: 227, lr: 7.81e-03, grad_scale: 8.0
+2023-03-02 05:16:35,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:16:46,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7599, 1.6378, 1.7398, 1.5546], device='cuda:0'), covar=tensor([0.0931, 0.1532, 0.1228, 0.1238], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0754, 0.0631, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 05:16:47,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5188, 2.2397, 1.5802, 0.6284], device='cuda:0'), covar=tensor([0.2649, 0.1253, 0.2125, 0.3035], device='cuda:0'), in_proj_covar=tensor([0.1319, 0.1243, 0.1329, 0.1121], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 05:16:51,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:17:16,159 INFO [train.py:968] (0/2) Epoch 4, batch 19300, giga_loss[loss=0.3505, simple_loss=0.3913, pruned_loss=0.1548, over 26533.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3779, pruned_loss=0.1258, over 5713572.31 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3715, pruned_loss=0.1174, over 5764429.38 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3776, pruned_loss=0.1265, over 5699967.59 frames. ], batch size: 555, lr: 7.81e-03, grad_scale: 8.0
+2023-03-02 05:17:34,490 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:17:50,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.006e+02 9.923e+02 1.228e+03 1.703e+03 3.659e+03, threshold=2.456e+03, percent-clipped=2.0
+2023-03-02 05:17:57,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:18:02,319 INFO [train.py:968] (0/2) Epoch 4, batch 19350, giga_loss[loss=0.2544, simple_loss=0.3246, pruned_loss=0.09205, over 28966.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3734, pruned_loss=0.123, over 5699821.16 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3716, pruned_loss=0.1174, over 5766389.66 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3731, pruned_loss=0.1236, over 5686392.49 frames. ], batch size: 213, lr: 7.81e-03, grad_scale: 8.0
+2023-03-02 05:18:45,883 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:18:47,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:18:48,055 INFO [train.py:968] (0/2) Epoch 4, batch 19400, giga_loss[loss=0.2736, simple_loss=0.3439, pruned_loss=0.1017, over 28942.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3674, pruned_loss=0.1195, over 5686038.37 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3722, pruned_loss=0.1177, over 5761145.08 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3665, pruned_loss=0.1198, over 5677949.63 frames. ], batch size: 136, lr: 7.81e-03, grad_scale: 4.0
+2023-03-02 05:19:12,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:19:21,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3404, 1.9149, 1.4004, 0.6052], device='cuda:0'), covar=tensor([0.1886, 0.0930, 0.1671, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.1246, 0.1318, 0.1113], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 05:19:22,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.585e+02 9.972e+02 1.236e+03 1.657e+03 6.281e+03, threshold=2.473e+03, percent-clipped=12.0
+2023-03-02 05:19:33,540 INFO [train.py:968] (0/2) Epoch 4, batch 19450, giga_loss[loss=0.2565, simple_loss=0.3292, pruned_loss=0.09187, over 28860.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.361, pruned_loss=0.1156, over 5687596.34 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3728, pruned_loss=0.1181, over 5763301.22 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3596, pruned_loss=0.1155, over 5678152.55 frames. ], batch size: 174, lr: 7.81e-03, grad_scale: 4.0
+2023-03-02 05:19:54,873 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-02 05:20:11,785 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:20:14,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:20:24,282 INFO [train.py:968] (0/2) Epoch 4, batch 19500, giga_loss[loss=0.3001, simple_loss=0.3704, pruned_loss=0.1149, over 28830.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3583, pruned_loss=0.1147, over 5646244.56 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3731, pruned_loss=0.1183, over 5754780.06 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3568, pruned_loss=0.1144, over 5645760.45 frames. ], batch size: 199, lr: 7.81e-03, grad_scale: 4.0
+2023-03-02 05:20:34,674 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5955, 1.8380, 1.8050, 1.7274], device='cuda:0'), covar=tensor([0.1418, 0.1613, 0.1037, 0.1001], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0749, 0.0751, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 05:20:39,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:20:40,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3175, 4.0988, 3.9777, 1.9372], device='cuda:0'), covar=tensor([0.0438, 0.0364, 0.0649, 0.2031], device='cuda:0'), in_proj_covar=tensor([0.0766, 0.0662, 0.0760, 0.0580], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 05:20:54,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.761e+02 9.752e+02 1.270e+03 1.801e+03 4.329e+03, threshold=2.541e+03, percent-clipped=7.0
+2023-03-02 05:21:05,874 INFO [train.py:968] (0/2) Epoch 4, batch 19550, giga_loss[loss=0.2877, simple_loss=0.3542, pruned_loss=0.1106, over 28977.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.36, pruned_loss=0.1155, over 5649573.74 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3742, pruned_loss=0.1189, over 5746710.83 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3575, pruned_loss=0.1147, over 5654192.96 frames. ], batch size: 186, lr: 7.81e-03, grad_scale: 4.0
+2023-03-02 05:21:47,283 INFO [train.py:968] (0/2) Epoch 4, batch 19600, giga_loss[loss=0.2612, simple_loss=0.3379, pruned_loss=0.09228, over 29096.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3587, pruned_loss=0.1141, over 5663154.92 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.375, pruned_loss=0.119, over 5748014.61 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3555, pruned_loss=0.1132, over 5663375.05 frames. ], batch size: 128, lr: 7.80e-03, grad_scale: 8.0
+2023-03-02 05:21:48,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:22:00,903 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-02 05:22:21,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.852e+02 9.821e+02 1.193e+03 1.617e+03 2.683e+03, threshold=2.387e+03, percent-clipped=2.0
+2023-03-02 05:22:29,543 INFO [train.py:968] (0/2) Epoch 4, batch 19650, giga_loss[loss=0.296, simple_loss=0.3609, pruned_loss=0.1156, over 28028.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3571, pruned_loss=0.1137, over 5672458.83 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3752, pruned_loss=0.1191, over 5750515.30 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3541, pruned_loss=0.1128, over 5669274.16 frames. ], batch size: 412, lr: 7.80e-03, grad_scale: 8.0
+2023-03-02 05:22:54,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8770, 2.3672, 2.6430, 2.2250], device='cuda:0'), covar=tensor([0.0818, 0.1481, 0.1056, 0.1188], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0749, 0.0629, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 05:23:03,567 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:23:08,546 INFO [train.py:968] (0/2) Epoch 4, batch 19700, giga_loss[loss=0.2448, simple_loss=0.3168, pruned_loss=0.0864, over 28746.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3542, pruned_loss=0.112, over 5684879.65 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3755, pruned_loss=0.1191, over 5754509.01 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3511, pruned_loss=0.1111, over 5677137.10 frames. ], batch size: 92, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:23:15,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:23:41,483 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.852e+02 9.423e+02 1.274e+03 2.099e+03 1.191e+04, threshold=2.548e+03, percent-clipped=19.0
+2023-03-02 05:23:45,788 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:23:47,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:23:49,809 INFO [train.py:968] (0/2) Epoch 4, batch 19750, giga_loss[loss=0.2751, simple_loss=0.3379, pruned_loss=0.1062, over 29021.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3512, pruned_loss=0.1099, over 5693439.03 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3755, pruned_loss=0.119, over 5753363.62 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3479, pruned_loss=0.109, over 5686195.33 frames. ], batch size: 136, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:24:11,048 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:24:31,568 INFO [train.py:968] (0/2) Epoch 4, batch 19800, libri_loss[loss=0.3434, simple_loss=0.4041, pruned_loss=0.1413, over 29549.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3498, pruned_loss=0.1091, over 5696216.67 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3765, pruned_loss=0.1193, over 5755429.32 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3455, pruned_loss=0.1077, over 5686753.91 frames. ], batch size: 77, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:24:59,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:25:00,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3996, 1.5056, 1.2784, 1.5301], device='cuda:0'), covar=tensor([0.0844, 0.0336, 0.0359, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0208, 0.0133, 0.0138, 0.0245], device='cuda:0'), out_proj_covar=tensor([0.0049, 0.0036, 0.0033, 0.0054], device='cuda:0')
+2023-03-02 05:25:00,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.033e+02 9.927e+02 1.258e+03 1.959e+03 6.452e+03, threshold=2.517e+03, percent-clipped=15.0
+2023-03-02 05:25:01,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:25:08,025 INFO [train.py:968] (0/2) Epoch 4, batch 19850, giga_loss[loss=0.2802, simple_loss=0.3409, pruned_loss=0.1097, over 28808.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3466, pruned_loss=0.1073, over 5704763.41 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3772, pruned_loss=0.1196, over 5752934.68 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3419, pruned_loss=0.1058, over 5698451.21 frames. ], batch size: 119, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:25:13,354 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:25:14,752 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-156000.pt
+2023-03-02 05:25:25,449 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:25:42,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2054, 1.3960, 1.0883, 1.3458], device='cuda:0'), covar=tensor([0.0873, 0.0342, 0.0394, 0.0927], device='cuda:0'), in_proj_covar=tensor([0.0208, 0.0133, 0.0139, 0.0245], device='cuda:0'), out_proj_covar=tensor([0.0049, 0.0036, 0.0033, 0.0054], device='cuda:0')
+2023-03-02 05:25:50,395 INFO [train.py:968] (0/2) Epoch 4, batch 19900, giga_loss[loss=0.2428, simple_loss=0.3116, pruned_loss=0.08696, over 28832.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3436, pruned_loss=0.1057, over 5715384.15 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3774, pruned_loss=0.1196, over 5755632.03 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3393, pruned_loss=0.1043, over 5707326.02 frames. ], batch size: 99, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:25:56,559 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:26:09,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4336, 2.1375, 1.5308, 0.4832], device='cuda:0'), covar=tensor([0.1686, 0.0994, 0.1794, 0.2070], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.1220, 0.1320, 0.1110], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 05:26:13,028 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5246, 1.9416, 1.2731, 0.7969], device='cuda:0'), covar=tensor([0.2628, 0.1366, 0.1397, 0.2648], device='cuda:0'), in_proj_covar=tensor([0.1296, 0.1222, 0.1322, 0.1111], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 05:26:21,851 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.125e+02 1.051e+03 1.385e+03 1.909e+03 5.606e+03, threshold=2.770e+03, percent-clipped=13.0
+2023-03-02 05:26:29,446 INFO [train.py:968] (0/2) Epoch 4, batch 19950, giga_loss[loss=0.2381, simple_loss=0.3136, pruned_loss=0.08127, over 28938.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3427, pruned_loss=0.1051, over 5720447.91 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3784, pruned_loss=0.12, over 5758972.44 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3371, pruned_loss=0.1031, over 5709836.03 frames. ], batch size: 186, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:27:07,374 INFO [train.py:968] (0/2) Epoch 4, batch 20000, giga_loss[loss=0.2605, simple_loss=0.3259, pruned_loss=0.0976, over 28739.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3427, pruned_loss=0.1051, over 5724328.86 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3789, pruned_loss=0.1202, over 5764656.01 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3364, pruned_loss=0.1028, over 5709154.72 frames. ], batch size: 92, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:27:21,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2698, 1.3714, 0.9869, 1.2027], device='cuda:0'), covar=tensor([0.0915, 0.0723, 0.1647, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0465, 0.0523, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 05:27:22,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2023, 1.2880, 1.0031, 0.6823], device='cuda:0'), covar=tensor([0.0947, 0.0717, 0.0588, 0.0860], device='cuda:0'), in_proj_covar=tensor([0.1308, 0.1010, 0.1053, 0.1144], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 05:27:31,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3254, 1.8487, 1.3880, 0.4951], device='cuda:0'), covar=tensor([0.2084, 0.1024, 0.1483, 0.2479], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.1225, 0.1312, 0.1102], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 05:27:32,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7455, 1.5790, 1.3168, 1.3643], device='cuda:0'), covar=tensor([0.0635, 0.0587, 0.0932, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0464, 0.0521, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 05:27:34,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:27:34,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9788, 5.6324, 5.5925, 2.4017], device='cuda:0'), covar=tensor([0.0375, 0.0267, 0.0668, 0.1966], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0663, 0.0761, 0.0574], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 05:27:39,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.080e+02 9.654e+02 1.249e+03 1.775e+03 1.713e+04, threshold=2.499e+03, percent-clipped=12.0
+2023-03-02 05:27:46,236 INFO [train.py:968] (0/2) Epoch 4, batch 20050, libri_loss[loss=0.337, simple_loss=0.4165, pruned_loss=0.1287, over 29522.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3421, pruned_loss=0.1045, over 5725564.32 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3796, pruned_loss=0.1202, over 5763888.46 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3356, pruned_loss=0.1022, over 5713133.86 frames. ], batch size: 82, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:28:13,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:28:28,649 INFO [train.py:968] (0/2) Epoch 4, batch 20100, giga_loss[loss=0.2934, simple_loss=0.3566, pruned_loss=0.1151, over 28190.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3467, pruned_loss=0.1081, over 5723162.37 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3799, pruned_loss=0.1204, over 5766650.40 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3406, pruned_loss=0.1059, over 5710195.50 frames. ], batch size: 77, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:29:05,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.958e+02 1.095e+03 1.365e+03 1.850e+03 4.864e+03, threshold=2.730e+03, percent-clipped=9.0
+2023-03-02 05:29:07,853 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 05:29:13,257 INFO [train.py:968] (0/2) Epoch 4, batch 20150, giga_loss[loss=0.3641, simple_loss=0.3912, pruned_loss=0.1684, over 23690.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3539, pruned_loss=0.1133, over 5716546.67 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3805, pruned_loss=0.1209, over 5765255.16 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3479, pruned_loss=0.1109, over 5706161.09 frames. ], batch size: 705, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:30:07,337 INFO [train.py:968] (0/2) Epoch 4, batch 20200, giga_loss[loss=0.3268, simple_loss=0.3795, pruned_loss=0.1371, over 28542.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3618, pruned_loss=0.1188, over 5705049.09 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3806, pruned_loss=0.1208, over 5767287.91 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3566, pruned_loss=0.1169, over 5694202.15 frames. ], batch size: 78, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:30:07,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1703, 1.1666, 4.9460, 3.4982], device='cuda:0'), covar=tensor([0.1653, 0.2273, 0.0257, 0.0529], device='cuda:0'), in_proj_covar=tensor([0.0550, 0.0508, 0.0695, 0.0564], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 05:30:32,753 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:30:35,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:30:37,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:30:48,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.083e+02 1.185e+03 1.477e+03 2.012e+03 7.681e+03, threshold=2.953e+03, percent-clipped=10.0
+2023-03-02 05:30:56,687 INFO [train.py:968] (0/2) Epoch 4, batch 20250, giga_loss[loss=0.3401, simple_loss=0.3768, pruned_loss=0.1517, over 23652.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3688, pruned_loss=0.1236, over 5697982.37 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3807, pruned_loss=0.1208, over 5767452.84 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3645, pruned_loss=0.1221, over 5688677.04 frames. ], batch size: 705, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:31:04,889 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:31:29,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:31:45,324 INFO [train.py:968] (0/2) Epoch 4, batch 20300, giga_loss[loss=0.2991, simple_loss=0.3784, pruned_loss=0.1099, over 29081.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3751, pruned_loss=0.1262, over 5697209.92 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3811, pruned_loss=0.1208, over 5768779.17 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3711, pruned_loss=0.1251, over 5686846.61 frames. ], batch size: 155, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:32:04,582 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:32:27,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.072e+02 1.006e+03 1.234e+03 1.571e+03 2.984e+03, threshold=2.467e+03, percent-clipped=0.0
+2023-03-02 05:32:34,410 INFO [train.py:968] (0/2) Epoch 4, batch 20350, giga_loss[loss=0.3269, simple_loss=0.3929, pruned_loss=0.1304, over 28683.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3788, pruned_loss=0.1273, over 5697640.83 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.381, pruned_loss=0.1206, over 5771011.09 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3757, pruned_loss=0.1267, over 5685917.39 frames. ], batch size: 262, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:32:55,275 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:32:58,611 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:33:16,175 INFO [train.py:968] (0/2) Epoch 4, batch 20400, libri_loss[loss=0.3064, simple_loss=0.3812, pruned_loss=0.1158, over 25984.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3825, pruned_loss=0.1292, over 5703765.31 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3803, pruned_loss=0.1201, over 5771456.75 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3806, pruned_loss=0.1295, over 5691654.80 frames. ], batch size: 136, lr: 7.78e-03, grad_scale: 8.0
+2023-03-02 05:33:20,237 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:33:24,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:33:36,775 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:33:38,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:33:52,815 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.180e+02 1.179e+03 1.502e+03 2.096e+03 4.443e+03, threshold=3.004e+03, percent-clipped=13.0
+2023-03-02 05:34:00,210 INFO [train.py:968] (0/2) Epoch 4, batch 20450, giga_loss[loss=0.2616, simple_loss=0.3373, pruned_loss=0.09296, over 28563.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3771, pruned_loss=0.126, over 5692964.10 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3801, pruned_loss=0.1202, over 5770555.36 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3758, pruned_loss=0.1263, over 5682824.78 frames. ], batch size: 336, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:34:06,521 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:34:13,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4082, 1.5336, 1.2345, 0.9614], device='cuda:0'), covar=tensor([0.1239, 0.0851, 0.0665, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.1318, 0.1018, 0.1066, 0.1143], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 05:34:14,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:34:42,375 INFO [train.py:968] (0/2) Epoch 4, batch 20500, giga_loss[loss=0.2905, simple_loss=0.3571, pruned_loss=0.112, over 28941.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3734, pruned_loss=0.1226, over 5695725.18 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3802, pruned_loss=0.1204, over 5761861.58 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3723, pruned_loss=0.1227, over 5693757.45 frames. ], batch size: 213, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:35:17,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:35:19,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.437e+02 1.088e+03 1.545e+03 2.244e+03 4.655e+03, threshold=3.090e+03, percent-clipped=11.0
+2023-03-02 05:35:27,664 INFO [train.py:968] (0/2) Epoch 4, batch 20550, giga_loss[loss=0.3121, simple_loss=0.3797, pruned_loss=0.1223, over 28624.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3726, pruned_loss=0.1216, over 5686304.78 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3802, pruned_loss=0.1205, over 5763666.98 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3715, pruned_loss=0.1216, over 5682140.65 frames. ], batch size: 92, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:35:29,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:35:31,138 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:35:42,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4699, 3.0941, 1.6062, 1.2432], device='cuda:0'), covar=tensor([0.0843, 0.0288, 0.0779, 0.1417], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0453, 0.0303, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 05:35:57,368 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:36:10,921 INFO [train.py:968] (0/2) Epoch 4, batch 20600, giga_loss[loss=0.2725, simple_loss=0.3476, pruned_loss=0.09873, over 28932.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3741, pruned_loss=0.1217, over 5692091.67 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3806, pruned_loss=0.1208, over 5762793.59 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3728, pruned_loss=0.1215, over 5689088.47 frames. ], batch size: 112, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:36:48,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.946e+02 1.290e+03 1.640e+03 2.144e+03 6.908e+03, threshold=3.280e+03, percent-clipped=11.0
+2023-03-02 05:36:54,593 INFO [train.py:968] (0/2) Epoch 4, batch 20650, giga_loss[loss=0.334, simple_loss=0.3986, pruned_loss=0.1347, over 28953.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3784, pruned_loss=0.1249, over 5692627.04 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.381, pruned_loss=0.1209, over 5765807.25 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3769, pruned_loss=0.1246, over 5686291.25 frames. ], batch size: 213, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:37:21,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8327, 1.6687, 1.6435, 1.4766], device='cuda:0'), covar=tensor([0.1062, 0.1831, 0.1385, 0.1449], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0750, 0.0625, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 05:37:34,688 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:37:38,655 INFO [train.py:968] (0/2) Epoch 4, batch 20700, giga_loss[loss=0.3004, simple_loss=0.3639, pruned_loss=0.1185, over 28887.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3802, pruned_loss=0.1265, over 5699909.01 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3805, pruned_loss=0.1205, over 5769959.30 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3794, pruned_loss=0.1268, over 5689029.27 frames. ], batch size: 106, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:38:19,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.369e+02 1.318e+03 1.643e+03 2.092e+03 5.736e+03, threshold=3.286e+03, percent-clipped=7.0
+2023-03-02 05:38:27,471 INFO [train.py:968] (0/2) Epoch 4, batch 20750, giga_loss[loss=0.3097, simple_loss=0.3762, pruned_loss=0.1216, over 28831.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.381, pruned_loss=0.1272, over 5710409.51 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3806, pruned_loss=0.1205, over 5770782.50 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3803, pruned_loss=0.1275, over 5700953.20 frames. ], batch size: 112, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:39:12,352 INFO [train.py:968] (0/2) Epoch 4, batch 20800, giga_loss[loss=0.3145, simple_loss=0.3763, pruned_loss=0.1264, over 28920.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3823, pruned_loss=0.1286, over 5709981.00 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3811, pruned_loss=0.1207, over 5773435.22 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3814, pruned_loss=0.1288, over 5698896.50 frames. ], batch size: 213, lr: 7.77e-03, grad_scale: 8.0
+2023-03-02 05:39:40,285 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-02 05:39:42,427 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:39:44,442 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:39:44,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:39:44,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.173e+02 1.212e+03 1.680e+03 2.323e+03 1.063e+04, threshold=3.359e+03, percent-clipped=10.0
+2023-03-02 05:39:47,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5535, 1.7177, 1.5869, 1.7792], device='cuda:0'), covar=tensor([0.1822, 0.1653, 0.1548, 0.1537], device='cuda:0'), in_proj_covar=tensor([0.1066, 0.0848, 0.0959, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 05:39:50,462 INFO [train.py:968] (0/2) Epoch 4, batch 20850, giga_loss[loss=0.2844, simple_loss=0.3582, pruned_loss=0.1052, over 28916.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3826, pruned_loss=0.1283, over 5708754.25 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3813, pruned_loss=0.1207, over 5767809.01 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3817, pruned_loss=0.1286, over 5704301.53 frames. ], batch size: 227, lr: 7.77e-03, grad_scale: 8.0
+2023-03-02 05:40:01,215 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4881, 2.0135, 1.4779, 0.7598], device='cuda:0'), covar=tensor([0.2068, 0.1116, 0.1758, 0.2250], device='cuda:0'), in_proj_covar=tensor([0.1316, 0.1224, 0.1338, 0.1110], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 05:40:06,365 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:40:29,621 INFO [train.py:968] (0/2) Epoch 4, batch 20900, giga_loss[loss=0.2568, simple_loss=0.3418, pruned_loss=0.08584, over 28397.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3819, pruned_loss=0.1266, over 5711055.14 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.381, pruned_loss=0.1206, over 5772557.86 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3815, pruned_loss=0.1272, over 5700809.58 frames. ], batch size: 60, lr: 7.77e-03, grad_scale: 8.0
+2023-03-02 05:40:39,005 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-02 05:40:42,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:41:01,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2886, 2.0187, 1.5376, 0.5011], device='cuda:0'), covar=tensor([0.1980, 0.0962, 0.1733, 0.2277], device='cuda:0'), in_proj_covar=tensor([0.1315, 0.1213, 0.1332, 0.1100], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 05:41:02,097 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 4.724e+02 1.057e+03 1.372e+03 1.776e+03 3.986e+03, threshold=2.744e+03, percent-clipped=2.0
+2023-03-02 05:41:04,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 05:41:09,396 INFO [train.py:968] (0/2) Epoch 4, batch 20950, giga_loss[loss=0.2875, simple_loss=0.3734, pruned_loss=0.1008, over 29021.00 frames. ], tot_loss[loss=0.316, simple_loss=0.382, pruned_loss=0.125, over 5707982.61 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3813, pruned_loss=0.1207, over 5762778.83 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3813, pruned_loss=0.1255, over 5706433.16 frames. ], batch size: 164, lr: 7.77e-03, grad_scale: 4.0
+2023-03-02 05:41:38,607 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:41:41,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:41:49,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:41:49,963 INFO [train.py:968] (0/2) Epoch 4, batch 21000, giga_loss[loss=0.2913, simple_loss=0.3655, pruned_loss=0.1085, over 28934.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3815, pruned_loss=0.1246, over 5706072.12 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3811, pruned_loss=0.1208, over 5755110.80 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3811, pruned_loss=0.125, over 5711161.82 frames. ], batch size: 128, lr: 7.77e-03, grad_scale: 2.0
+2023-03-02 05:41:49,971 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 05:41:59,968 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2497, simple_loss=0.3505, pruned_loss=0.07442, over 944034.00 frames. 
+2023-03-02 05:41:59,969 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 05:42:12,297 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:42:34,285 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.237e+02 1.110e+03 1.431e+03 2.054e+03 6.099e+03, threshold=2.863e+03, percent-clipped=12.0
+2023-03-02 05:42:38,531 INFO [train.py:968] (0/2) Epoch 4, batch 21050, giga_loss[loss=0.3148, simple_loss=0.3779, pruned_loss=0.1259, over 28647.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3789, pruned_loss=0.1232, over 5708389.73 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3811, pruned_loss=0.1208, over 5760370.90 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3786, pruned_loss=0.1236, over 5706278.98 frames. ], batch size: 307, lr: 7.77e-03, grad_scale: 2.0
+2023-03-02 05:42:45,860 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:42:47,675 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:42:49,907 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-02 05:43:09,169 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:43:16,871 INFO [train.py:968] (0/2) Epoch 4, batch 21100, libri_loss[loss=0.3348, simple_loss=0.3874, pruned_loss=0.1411, over 29566.00 frames. ], tot_loss[loss=0.311, simple_loss=0.377, pruned_loss=0.1225, over 5702759.45 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3817, pruned_loss=0.1214, over 5751661.88 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3761, pruned_loss=0.1222, over 5707908.30 frames. ], batch size: 76, lr: 7.77e-03, grad_scale: 2.0
+2023-03-02 05:43:29,896 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-02 05:43:35,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 05:43:39,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3957, 1.6079, 1.2880, 1.6086], device='cuda:0'), covar=tensor([0.0836, 0.0333, 0.0359, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0206, 0.0132, 0.0136, 0.0245], device='cuda:0'), out_proj_covar=tensor([0.0049, 0.0036, 0.0032, 0.0054], device='cuda:0')
+2023-03-02 05:43:40,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9661, 2.3426, 2.1561, 1.9008], device='cuda:0'), covar=tensor([0.1493, 0.1570, 0.1040, 0.1005], device='cuda:0'), in_proj_covar=tensor([0.0787, 0.0750, 0.0755, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 05:43:45,277 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:43:51,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.320e+02 1.048e+03 1.297e+03 1.726e+03 5.140e+03, threshold=2.594e+03, percent-clipped=6.0
+2023-03-02 05:43:55,828 INFO [train.py:968] (0/2) Epoch 4, batch 21150, libri_loss[loss=0.3222, simple_loss=0.3932, pruned_loss=0.1256, over 26028.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3758, pruned_loss=0.1218, over 5707953.36 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3818, pruned_loss=0.1215, over 5752641.09 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3749, pruned_loss=0.1215, over 5710304.17 frames. ], batch size: 136, lr: 7.77e-03, grad_scale: 2.0
+2023-03-02 05:44:37,001 INFO [train.py:968] (0/2) Epoch 4, batch 21200, giga_loss[loss=0.3447, simple_loss=0.399, pruned_loss=0.1452, over 28415.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3769, pruned_loss=0.1231, over 5711273.59 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3825, pruned_loss=0.1222, over 5755725.91 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3753, pruned_loss=0.1223, over 5709337.03 frames. ], batch size: 71, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:45:11,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5170, 1.4401, 1.5024, 1.4188], device='cuda:0'), covar=tensor([0.1089, 0.1627, 0.1489, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0739, 0.0613, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:0')
+2023-03-02 05:45:13,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.901e+02 1.054e+03 1.292e+03 1.618e+03 8.057e+03, threshold=2.584e+03, percent-clipped=9.0
+2023-03-02 05:45:16,417 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-02 05:45:17,432 INFO [train.py:968] (0/2) Epoch 4, batch 21250, giga_loss[loss=0.2712, simple_loss=0.3501, pruned_loss=0.0962, over 29104.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.377, pruned_loss=0.123, over 5709678.66 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3828, pruned_loss=0.1223, over 5758943.19 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3754, pruned_loss=0.1222, over 5704773.38 frames. ], batch size: 155, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:45:26,776 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:45:59,293 INFO [train.py:968] (0/2) Epoch 4, batch 21300, giga_loss[loss=0.2927, simple_loss=0.3586, pruned_loss=0.1134, over 28671.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3782, pruned_loss=0.1231, over 5709902.18 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3831, pruned_loss=0.1227, over 5753639.09 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3766, pruned_loss=0.1222, over 5709543.34 frames. ], batch size: 78, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:46:37,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.194e+02 1.052e+03 1.255e+03 1.675e+03 7.856e+03, threshold=2.511e+03, percent-clipped=7.0
+2023-03-02 05:46:38,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4707, 1.4807, 5.1170, 3.6227], device='cuda:0'), covar=tensor([0.2188, 0.2645, 0.0427, 0.0645], device='cuda:0'), in_proj_covar=tensor([0.0544, 0.0501, 0.0690, 0.0562], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 05:46:41,262 INFO [train.py:968] (0/2) Epoch 4, batch 21350, giga_loss[loss=0.3188, simple_loss=0.383, pruned_loss=0.1273, over 28874.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3768, pruned_loss=0.122, over 5708136.54 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3828, pruned_loss=0.1228, over 5756573.73 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3757, pruned_loss=0.1212, over 5704562.71 frames. ], batch size: 119, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:46:48,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2840, 1.3671, 1.0648, 0.8108], device='cuda:0'), covar=tensor([0.0820, 0.0772, 0.0580, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.1308, 0.1035, 0.1064, 0.1154], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 05:46:52,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3325, 2.0112, 1.4266, 1.7464], device='cuda:0'), covar=tensor([0.0597, 0.0661, 0.0966, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0457, 0.0515, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 05:47:01,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:47:15,212 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:47:20,468 INFO [train.py:968] (0/2) Epoch 4, batch 21400, giga_loss[loss=0.2683, simple_loss=0.3422, pruned_loss=0.09719, over 29144.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3765, pruned_loss=0.1225, over 5699454.66 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3834, pruned_loss=0.1233, over 5749119.85 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3748, pruned_loss=0.1213, over 5701537.54 frames. ], batch size: 128, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:47:36,359 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-02 05:47:57,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.187e+02 1.012e+03 1.369e+03 1.894e+03 6.589e+03, threshold=2.738e+03, percent-clipped=12.0
+2023-03-02 05:48:02,325 INFO [train.py:968] (0/2) Epoch 4, batch 21450, giga_loss[loss=0.319, simple_loss=0.3748, pruned_loss=0.1317, over 28659.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3736, pruned_loss=0.1211, over 5690966.69 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3836, pruned_loss=0.1235, over 5742158.68 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.372, pruned_loss=0.12, over 5698403.21 frames. ], batch size: 99, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:48:41,330 INFO [train.py:968] (0/2) Epoch 4, batch 21500, giga_loss[loss=0.2745, simple_loss=0.3422, pruned_loss=0.1034, over 28394.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3705, pruned_loss=0.1195, over 5696252.12 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3841, pruned_loss=0.1239, over 5747325.39 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3685, pruned_loss=0.1182, over 5696232.41 frames. ], batch size: 65, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:48:49,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:48:55,575 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:48:58,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:49:19,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.148e+02 1.151e+03 1.491e+03 1.945e+03 6.536e+03, threshold=2.983e+03, percent-clipped=15.0
+2023-03-02 05:49:23,125 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:49:23,510 INFO [train.py:968] (0/2) Epoch 4, batch 21550, giga_loss[loss=0.3109, simple_loss=0.3768, pruned_loss=0.1225, over 28898.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3701, pruned_loss=0.1202, over 5691717.26 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3843, pruned_loss=0.1242, over 5745875.76 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3683, pruned_loss=0.1189, over 5692620.16 frames. ], batch size: 174, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:50:07,150 INFO [train.py:968] (0/2) Epoch 4, batch 21600, giga_loss[loss=0.3205, simple_loss=0.367, pruned_loss=0.137, over 23545.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3689, pruned_loss=0.12, over 5692232.80 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3845, pruned_loss=0.1245, over 5749136.14 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3671, pruned_loss=0.1187, over 5689098.51 frames. ], batch size: 705, lr: 7.75e-03, grad_scale: 8.0
+2023-03-02 05:50:11,467 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:50:34,874 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:50:36,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:50:42,902 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.359e+02 1.092e+03 1.330e+03 1.701e+03 4.225e+03, threshold=2.660e+03, percent-clipped=2.0
+2023-03-02 05:50:47,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6994, 2.0162, 1.9038, 1.8118], device='cuda:0'), covar=tensor([0.1523, 0.1758, 0.1162, 0.1089], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0758, 0.0755, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 05:50:48,747 INFO [train.py:968] (0/2) Epoch 4, batch 21650, giga_loss[loss=0.2722, simple_loss=0.3365, pruned_loss=0.1039, over 28917.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3665, pruned_loss=0.1192, over 5699718.90 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3848, pruned_loss=0.1248, over 5751188.61 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3645, pruned_loss=0.1178, over 5694582.79 frames. ], batch size: 106, lr: 7.75e-03, grad_scale: 8.0
+2023-03-02 05:50:51,222 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:50:51,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157797.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:50:53,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:51:15,305 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:51:28,033 INFO [train.py:968] (0/2) Epoch 4, batch 21700, giga_loss[loss=0.2895, simple_loss=0.3472, pruned_loss=0.1159, over 28951.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3642, pruned_loss=0.1186, over 5704399.55 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3848, pruned_loss=0.1248, over 5752850.18 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3625, pruned_loss=0.1174, over 5698545.53 frames. ], batch size: 112, lr: 7.75e-03, grad_scale: 8.0
+2023-03-02 05:52:02,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.111e+02 1.109e+03 1.372e+03 1.989e+03 5.133e+03, threshold=2.744e+03, percent-clipped=11.0
+2023-03-02 05:52:06,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2754, 1.4245, 1.1800, 1.2349], device='cuda:0'), covar=tensor([0.2070, 0.2082, 0.2083, 0.2024], device='cuda:0'), in_proj_covar=tensor([0.1071, 0.0848, 0.0958, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 05:52:07,718 INFO [train.py:968] (0/2) Epoch 4, batch 21750, libri_loss[loss=0.3991, simple_loss=0.4435, pruned_loss=0.1774, over 28597.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3632, pruned_loss=0.1186, over 5713008.49 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3852, pruned_loss=0.1254, over 5752843.43 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3609, pruned_loss=0.1169, over 5707308.43 frames. ], batch size: 106, lr: 7.75e-03, grad_scale: 8.0
+2023-03-02 05:52:12,224 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157899.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:52:20,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:52:29,200 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:52:31,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:52:47,610 INFO [train.py:968] (0/2) Epoch 4, batch 21800, giga_loss[loss=0.2985, simple_loss=0.3578, pruned_loss=0.1196, over 29021.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3616, pruned_loss=0.1176, over 5710940.64 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3856, pruned_loss=0.1257, over 5750990.27 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3593, pruned_loss=0.116, over 5707755.55 frames. ], batch size: 136, lr: 7.75e-03, grad_scale: 8.0
+2023-03-02 05:52:58,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:53:04,630 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:53:24,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.051e+02 1.006e+03 1.374e+03 1.751e+03 5.624e+03, threshold=2.747e+03, percent-clipped=6.0
+2023-03-02 05:53:25,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6055, 4.2769, 4.3053, 1.7060], device='cuda:0'), covar=tensor([0.0415, 0.0527, 0.0975, 0.2058], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0671, 0.0778, 0.0591], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 05:53:28,965 INFO [train.py:968] (0/2) Epoch 4, batch 21850, libri_loss[loss=0.3609, simple_loss=0.4192, pruned_loss=0.1513, over 29247.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3648, pruned_loss=0.119, over 5715013.06 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3863, pruned_loss=0.1265, over 5754864.57 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3616, pruned_loss=0.1167, over 5707636.48 frames. ], batch size: 94, lr: 7.75e-03, grad_scale: 4.0
+2023-03-02 05:53:35,168 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-158000.pt
+2023-03-02 05:54:11,267 INFO [train.py:968] (0/2) Epoch 4, batch 21900, giga_loss[loss=0.3042, simple_loss=0.381, pruned_loss=0.1138, over 28927.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3679, pruned_loss=0.1205, over 5706404.29 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3862, pruned_loss=0.1268, over 5758767.49 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3648, pruned_loss=0.1182, over 5695940.67 frames. ], batch size: 227, lr: 7.75e-03, grad_scale: 4.0
+2023-03-02 05:54:20,202 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:54:22,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:54:46,304 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:54:48,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.416e+02 1.139e+03 1.453e+03 2.293e+03 6.871e+03, threshold=2.905e+03, percent-clipped=16.0
+2023-03-02 05:54:51,285 INFO [train.py:968] (0/2) Epoch 4, batch 21950, giga_loss[loss=0.2833, simple_loss=0.3554, pruned_loss=0.1056, over 28881.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3711, pruned_loss=0.122, over 5711395.68 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3871, pruned_loss=0.128, over 5766193.08 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.367, pruned_loss=0.1187, over 5693538.04 frames. ], batch size: 227, lr: 7.75e-03, grad_scale: 4.0
+2023-03-02 05:55:16,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:55:31,832 INFO [train.py:968] (0/2) Epoch 4, batch 22000, giga_loss[loss=0.3443, simple_loss=0.4044, pruned_loss=0.1421, over 28708.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3724, pruned_loss=0.1223, over 5714170.01 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.388, pruned_loss=0.1293, over 5762654.48 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3677, pruned_loss=0.1183, over 5700421.18 frames. ], batch size: 243, lr: 7.74e-03, grad_scale: 8.0
+2023-03-02 05:55:35,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2659, 1.8343, 1.2453, 1.4597], device='cuda:0'), covar=tensor([0.0737, 0.0385, 0.0368, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0204, 0.0131, 0.0136, 0.0242], device='cuda:0'), out_proj_covar=tensor([0.0049, 0.0036, 0.0032, 0.0054], device='cuda:0')
+2023-03-02 05:55:37,781 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:55:55,365 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158172.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:56:10,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.136e+02 1.147e+03 1.475e+03 2.080e+03 6.952e+03, threshold=2.949e+03, percent-clipped=14.0
+2023-03-02 05:56:13,414 INFO [train.py:968] (0/2) Epoch 4, batch 22050, giga_loss[loss=0.2547, simple_loss=0.335, pruned_loss=0.08723, over 28952.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3714, pruned_loss=0.1208, over 5718118.52 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3885, pruned_loss=0.13, over 5766313.22 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3668, pruned_loss=0.1168, over 5702858.85 frames. ], batch size: 136, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:56:56,205 INFO [train.py:968] (0/2) Epoch 4, batch 22100, giga_loss[loss=0.3206, simple_loss=0.3841, pruned_loss=0.1285, over 28575.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3714, pruned_loss=0.1212, over 5708576.54 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3885, pruned_loss=0.1302, over 5770305.70 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3674, pruned_loss=0.1176, over 5691399.82 frames. ], batch size: 336, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:56:59,432 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:15,569 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:17,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:19,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158274.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:57:30,543 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-02 05:57:33,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.742e+02 1.099e+03 1.364e+03 1.818e+03 6.738e+03, threshold=2.729e+03, percent-clipped=10.0
+2023-03-02 05:57:36,165 INFO [train.py:968] (0/2) Epoch 4, batch 22150, libri_loss[loss=0.3655, simple_loss=0.4222, pruned_loss=0.1544, over 28543.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.372, pruned_loss=0.1219, over 5710521.69 frames. ], libri_tot_loss[loss=0.325, simple_loss=0.3889, pruned_loss=0.1306, over 5768931.84 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3681, pruned_loss=0.1185, over 5697052.70 frames. ], batch size: 106, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:57:36,470 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:38,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:40,307 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:54,844 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158315.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:57:56,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158318.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:58:01,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:58:09,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:58:14,084 INFO [train.py:968] (0/2) Epoch 4, batch 22200, giga_loss[loss=0.3325, simple_loss=0.3839, pruned_loss=0.1405, over 28801.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3746, pruned_loss=0.124, over 5711217.76 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3901, pruned_loss=0.1316, over 5773039.88 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3697, pruned_loss=0.12, over 5694246.46 frames. ], batch size: 112, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:58:18,941 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158347.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:58:53,663 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.837e+02 1.154e+03 1.628e+03 2.114e+03 6.147e+03, threshold=3.255e+03, percent-clipped=8.0
+2023-03-02 05:58:56,191 INFO [train.py:968] (0/2) Epoch 4, batch 22250, libri_loss[loss=0.4241, simple_loss=0.4452, pruned_loss=0.2015, over 29571.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3774, pruned_loss=0.1255, over 5717779.42 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3908, pruned_loss=0.1326, over 5774408.49 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3724, pruned_loss=0.1212, over 5701652.40 frames. ], batch size: 75, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:59:06,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6527, 3.2686, 1.5796, 1.6043], device='cuda:0'), covar=tensor([0.0798, 0.0342, 0.0828, 0.1227], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0457, 0.0304, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 05:59:16,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158417.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:59:17,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158420.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:59:35,823 INFO [train.py:968] (0/2) Epoch 4, batch 22300, libri_loss[loss=0.3948, simple_loss=0.4514, pruned_loss=0.1691, over 29554.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3797, pruned_loss=0.1265, over 5709937.04 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3912, pruned_loss=0.1328, over 5767387.84 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3752, pruned_loss=0.1228, over 5702519.52 frames. ], batch size: 89, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:59:41,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158449.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:00:03,252 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:00:05,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:00:10,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.136e+02 1.310e+03 1.684e+03 2.253e+03 8.502e+03, threshold=3.367e+03, percent-clipped=9.0
+2023-03-02 06:00:13,236 INFO [train.py:968] (0/2) Epoch 4, batch 22350, giga_loss[loss=0.3199, simple_loss=0.3859, pruned_loss=0.1269, over 28262.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.381, pruned_loss=0.1273, over 5715493.67 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3923, pruned_loss=0.1337, over 5767487.90 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3761, pruned_loss=0.1233, over 5707959.28 frames. ], batch size: 368, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 06:00:27,155 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-02 06:00:27,556 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:00:30,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:00:53,583 INFO [train.py:968] (0/2) Epoch 4, batch 22400, giga_loss[loss=0.3041, simple_loss=0.369, pruned_loss=0.1196, over 28466.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3807, pruned_loss=0.1266, over 5720055.76 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3928, pruned_loss=0.1341, over 5764057.40 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3761, pruned_loss=0.123, over 5715727.12 frames. ], batch size: 71, lr: 7.74e-03, grad_scale: 8.0
+2023-03-02 06:00:58,240 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 06:01:32,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8986, 0.9502, 3.9235, 3.0111], device='cuda:0'), covar=tensor([0.1637, 0.2341, 0.0351, 0.0678], device='cuda:0'), in_proj_covar=tensor([0.0543, 0.0502, 0.0701, 0.0574], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 06:01:32,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.751e+02 1.222e+03 1.509e+03 1.863e+03 4.835e+03, threshold=3.019e+03, percent-clipped=3.0
+2023-03-02 06:01:35,927 INFO [train.py:968] (0/2) Epoch 4, batch 22450, giga_loss[loss=0.2939, simple_loss=0.3675, pruned_loss=0.1102, over 28935.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.381, pruned_loss=0.1266, over 5708967.38 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3935, pruned_loss=0.1348, over 5757454.14 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3763, pruned_loss=0.1228, over 5709690.50 frames. ], batch size: 186, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:01:37,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2063, 4.7547, 4.8134, 2.4242], device='cuda:0'), covar=tensor([0.0353, 0.0356, 0.0676, 0.1749], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0675, 0.0790, 0.0593], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 06:01:57,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:02:16,722 INFO [train.py:968] (0/2) Epoch 4, batch 22500, giga_loss[loss=0.2972, simple_loss=0.3662, pruned_loss=0.1141, over 28917.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3787, pruned_loss=0.1251, over 5717863.45 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3936, pruned_loss=0.1349, over 5760508.45 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3747, pruned_loss=0.1218, over 5715068.87 frames. ], batch size: 213, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:02:17,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5489, 2.0025, 1.1972, 1.2217], device='cuda:0'), covar=tensor([0.1344, 0.0791, 0.0909, 0.1007], device='cuda:0'), in_proj_covar=tensor([0.1317, 0.1070, 0.1086, 0.1161], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 06:02:39,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0907, 1.2050, 1.0157, 0.6794], device='cuda:0'), covar=tensor([0.0619, 0.0622, 0.0452, 0.0616], device='cuda:0'), in_proj_covar=tensor([0.1312, 0.1067, 0.1086, 0.1160], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 06:02:58,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.430e+02 1.226e+03 1.584e+03 2.316e+03 4.830e+03, threshold=3.167e+03, percent-clipped=11.0
+2023-03-02 06:03:00,040 INFO [train.py:968] (0/2) Epoch 4, batch 22550, giga_loss[loss=0.2824, simple_loss=0.3529, pruned_loss=0.106, over 29033.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3768, pruned_loss=0.1241, over 5722971.48 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3937, pruned_loss=0.1351, over 5765104.76 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3731, pruned_loss=0.121, over 5715530.73 frames. ], batch size: 128, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:03:26,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-02 06:03:43,754 INFO [train.py:968] (0/2) Epoch 4, batch 22600, giga_loss[loss=0.3256, simple_loss=0.381, pruned_loss=0.1351, over 27493.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3734, pruned_loss=0.1226, over 5718714.05 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3941, pruned_loss=0.1355, over 5765597.91 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.37, pruned_loss=0.1197, over 5712026.74 frames. ], batch size: 472, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:03:54,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9163, 1.0530, 0.8817, 0.6445], device='cuda:0'), covar=tensor([0.0851, 0.0747, 0.0521, 0.0779], device='cuda:0'), in_proj_covar=tensor([0.1319, 0.1062, 0.1091, 0.1157], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 06:03:57,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2110, 1.6771, 1.5393, 1.4753], device='cuda:0'), covar=tensor([0.1299, 0.1808, 0.1104, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0747, 0.0756, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 06:04:02,295 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:04:04,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:04:11,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:04:20,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.806e+02 1.076e+03 1.330e+03 1.895e+03 4.634e+03, threshold=2.659e+03, percent-clipped=8.0
+2023-03-02 06:04:22,863 INFO [train.py:968] (0/2) Epoch 4, batch 22650, giga_loss[loss=0.2564, simple_loss=0.3347, pruned_loss=0.0891, over 28774.00 frames. ], tot_loss[loss=0.306, simple_loss=0.371, pruned_loss=0.1205, over 5722192.10 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3938, pruned_loss=0.1357, over 5768012.03 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.368, pruned_loss=0.1177, over 5713530.63 frames. ], batch size: 112, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:04:26,306 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:05:04,770 INFO [train.py:968] (0/2) Epoch 4, batch 22700, giga_loss[loss=0.2867, simple_loss=0.368, pruned_loss=0.1027, over 29013.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3738, pruned_loss=0.1212, over 5721302.78 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3944, pruned_loss=0.1362, over 5771181.31 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3705, pruned_loss=0.1182, over 5710757.94 frames. ], batch size: 128, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:05:22,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4007, 1.3919, 1.1521, 1.3119], device='cuda:0'), covar=tensor([0.0612, 0.0535, 0.0891, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0462, 0.0512, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 06:05:45,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.308e+02 1.141e+03 1.458e+03 1.880e+03 3.563e+03, threshold=2.916e+03, percent-clipped=2.0
+2023-03-02 06:05:45,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:05:46,986 INFO [train.py:968] (0/2) Epoch 4, batch 22750, giga_loss[loss=0.3542, simple_loss=0.4072, pruned_loss=0.1506, over 28581.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3741, pruned_loss=0.1209, over 5723360.73 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3945, pruned_loss=0.1363, over 5771852.40 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3714, pruned_loss=0.1184, over 5714247.45 frames. ], batch size: 336, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:06:26,752 INFO [train.py:968] (0/2) Epoch 4, batch 22800, giga_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1214, over 28998.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3728, pruned_loss=0.1212, over 5732606.65 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.3945, pruned_loss=0.1366, over 5776067.25 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3699, pruned_loss=0.1185, over 5720116.67 frames. ], batch size: 155, lr: 7.73e-03, grad_scale: 8.0
+2023-03-02 06:06:36,199 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7436, 2.0113, 1.8957, 1.7429], device='cuda:0'), covar=tensor([0.1514, 0.1799, 0.1160, 0.1178], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0746, 0.0753, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 06:06:38,715 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-02 06:07:06,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.820e+02 1.104e+03 1.690e+03 2.229e+03 7.934e+03, threshold=3.381e+03, percent-clipped=10.0
+2023-03-02 06:07:08,157 INFO [train.py:968] (0/2) Epoch 4, batch 22850, giga_loss[loss=0.2716, simple_loss=0.3401, pruned_loss=0.1016, over 28902.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3726, pruned_loss=0.1229, over 5729794.34 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.395, pruned_loss=0.137, over 5776631.40 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3694, pruned_loss=0.12, over 5718099.53 frames. ], batch size: 227, lr: 7.72e-03, grad_scale: 8.0
+2023-03-02 06:07:39,707 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:07:41,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:07:41,885 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-02 06:07:48,033 INFO [train.py:968] (0/2) Epoch 4, batch 22900, giga_loss[loss=0.3023, simple_loss=0.3594, pruned_loss=0.1227, over 28823.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3715, pruned_loss=0.1237, over 5710899.20 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3959, pruned_loss=0.1379, over 5758849.26 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3675, pruned_loss=0.1203, over 5715622.65 frames. ], batch size: 99, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:08:06,128 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:08:15,632 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6102, 4.1478, 1.6046, 1.4817], device='cuda:0'), covar=tensor([0.0811, 0.0327, 0.0889, 0.1291], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0459, 0.0303, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 06:08:27,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.769e+02 1.152e+03 1.537e+03 2.137e+03 1.002e+04, threshold=3.073e+03, percent-clipped=12.0
+2023-03-02 06:08:28,347 INFO [train.py:968] (0/2) Epoch 4, batch 22950, giga_loss[loss=0.3293, simple_loss=0.3871, pruned_loss=0.1357, over 28318.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3704, pruned_loss=0.1243, over 5716285.92 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3958, pruned_loss=0.1382, over 5758694.74 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3665, pruned_loss=0.1208, over 5718780.23 frames. ], batch size: 368, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:09:01,944 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-02 06:09:08,224 INFO [train.py:968] (0/2) Epoch 4, batch 23000, giga_loss[loss=0.3064, simple_loss=0.3714, pruned_loss=0.1207, over 28285.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3689, pruned_loss=0.1237, over 5715572.55 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3961, pruned_loss=0.1386, over 5761096.73 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3651, pruned_loss=0.1203, over 5714622.24 frames. ], batch size: 368, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:09:15,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:09:21,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4527, 3.5574, 1.6008, 1.4334], device='cuda:0'), covar=tensor([0.0823, 0.0312, 0.0862, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.0333, 0.0462, 0.0305, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 06:09:23,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4131, 1.6159, 1.3674, 1.5569], device='cuda:0'), covar=tensor([0.2118, 0.2001, 0.2046, 0.2094], device='cuda:0'), in_proj_covar=tensor([0.1081, 0.0853, 0.0959, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 06:09:29,307 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:09:33,545 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:09:37,133 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-02 06:09:46,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.628e+02 1.075e+03 1.320e+03 1.685e+03 3.064e+03, threshold=2.639e+03, percent-clipped=0.0
+2023-03-02 06:09:48,151 INFO [train.py:968] (0/2) Epoch 4, batch 23050, giga_loss[loss=0.2709, simple_loss=0.3416, pruned_loss=0.1001, over 29096.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3636, pruned_loss=0.1203, over 5723443.20 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.396, pruned_loss=0.1387, over 5764602.56 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3598, pruned_loss=0.117, over 5718452.87 frames. ], batch size: 155, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:10:14,601 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159228.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:10:27,872 INFO [train.py:968] (0/2) Epoch 4, batch 23100, giga_loss[loss=0.2614, simple_loss=0.3254, pruned_loss=0.0987, over 28645.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3599, pruned_loss=0.1186, over 5714459.55 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3962, pruned_loss=0.1394, over 5758409.90 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3557, pruned_loss=0.1148, over 5714003.86 frames. ], batch size: 85, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:11:04,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.043e+02 1.174e+03 1.436e+03 1.973e+03 4.902e+03, threshold=2.872e+03, percent-clipped=12.0
+2023-03-02 06:11:06,056 INFO [train.py:968] (0/2) Epoch 4, batch 23150, giga_loss[loss=0.2426, simple_loss=0.3099, pruned_loss=0.08767, over 28598.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3562, pruned_loss=0.1162, over 5720357.87 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3963, pruned_loss=0.1394, over 5758946.14 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3522, pruned_loss=0.1128, over 5718717.39 frames. ], batch size: 85, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:11:07,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:11:11,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:11:36,737 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:11:50,951 INFO [train.py:968] (0/2) Epoch 4, batch 23200, giga_loss[loss=0.312, simple_loss=0.3823, pruned_loss=0.1209, over 28882.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3594, pruned_loss=0.1176, over 5706062.57 frames. ], libri_tot_loss[loss=0.3378, simple_loss=0.3963, pruned_loss=0.1396, over 5751515.51 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3559, pruned_loss=0.1146, over 5711414.53 frames. ], batch size: 174, lr: 7.72e-03, grad_scale: 8.0
+2023-03-02 06:12:31,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.766e+02 1.179e+03 1.585e+03 1.979e+03 6.157e+03, threshold=3.169e+03, percent-clipped=7.0
+2023-03-02 06:12:32,391 INFO [train.py:968] (0/2) Epoch 4, batch 23250, giga_loss[loss=0.3365, simple_loss=0.3881, pruned_loss=0.1424, over 28480.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3644, pruned_loss=0.1204, over 5699784.04 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.3964, pruned_loss=0.1398, over 5745919.32 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3607, pruned_loss=0.1174, over 5707519.09 frames. ], batch size: 85, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:13:10,974 INFO [train.py:968] (0/2) Epoch 4, batch 23300, giga_loss[loss=0.3116, simple_loss=0.3803, pruned_loss=0.1214, over 28337.00 frames. ], tot_loss[loss=0.308, simple_loss=0.37, pruned_loss=0.123, over 5706114.83 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3972, pruned_loss=0.1406, over 5746592.56 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3652, pruned_loss=0.1191, over 5710411.46 frames. ], batch size: 368, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:13:15,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8848, 1.2078, 3.9241, 3.0502], device='cuda:0'), covar=tensor([0.1574, 0.2088, 0.0338, 0.0599], device='cuda:0'), in_proj_covar=tensor([0.0550, 0.0507, 0.0712, 0.0577], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 06:13:25,583 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:13:49,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.113e+02 1.222e+03 1.637e+03 2.201e+03 5.205e+03, threshold=3.274e+03, percent-clipped=8.0
+2023-03-02 06:13:50,545 INFO [train.py:968] (0/2) Epoch 4, batch 23350, giga_loss[loss=0.2823, simple_loss=0.352, pruned_loss=0.1063, over 28878.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3736, pruned_loss=0.1247, over 5703747.56 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3974, pruned_loss=0.1409, over 5741416.61 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3688, pruned_loss=0.1207, over 5711201.90 frames. ], batch size: 86, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:14:33,938 INFO [train.py:968] (0/2) Epoch 4, batch 23400, giga_loss[loss=0.3092, simple_loss=0.369, pruned_loss=0.1247, over 28800.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3749, pruned_loss=0.1249, over 5712356.55 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.3971, pruned_loss=0.141, over 5743653.13 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3709, pruned_loss=0.1214, over 5715740.25 frames. ], batch size: 99, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:14:36,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:14:39,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4943, 2.3623, 1.6995, 0.6283], device='cuda:0'), covar=tensor([0.2738, 0.1107, 0.2101, 0.2859], device='cuda:0'), in_proj_covar=tensor([0.1317, 0.1225, 0.1324, 0.1117], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 06:14:40,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:14:41,232 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:14:44,187 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-02 06:15:20,027 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.617e+02 1.213e+03 1.676e+03 2.037e+03 4.732e+03, threshold=3.353e+03, percent-clipped=5.0
+2023-03-02 06:15:21,494 INFO [train.py:968] (0/2) Epoch 4, batch 23450, giga_loss[loss=0.3116, simple_loss=0.3672, pruned_loss=0.128, over 28563.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3803, pruned_loss=0.1299, over 5707958.46 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3972, pruned_loss=0.1411, over 5746230.48 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3768, pruned_loss=0.1268, over 5707830.89 frames. ], batch size: 85, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:15:33,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159603.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 06:16:08,399 INFO [train.py:968] (0/2) Epoch 4, batch 23500, giga_loss[loss=0.4006, simple_loss=0.444, pruned_loss=0.1786, over 28863.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3876, pruned_loss=0.1364, over 5694817.47 frames. ], libri_tot_loss[loss=0.3398, simple_loss=0.3971, pruned_loss=0.1412, over 5735544.98 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3847, pruned_loss=0.1337, over 5702869.81 frames. ], batch size: 284, lr: 7.71e-03, grad_scale: 4.0
+2023-03-02 06:16:48,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9040, 1.0115, 0.8269, 0.5570], device='cuda:0'), covar=tensor([0.0634, 0.0597, 0.0466, 0.0656], device='cuda:0'), in_proj_covar=tensor([0.1327, 0.1074, 0.1085, 0.1159], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 06:16:48,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:16:52,540 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:16:56,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:16:56,930 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.807e+03 2.320e+03 3.285e+03 7.563e+03, threshold=4.641e+03, percent-clipped=24.0
+2023-03-02 06:16:57,538 INFO [train.py:968] (0/2) Epoch 4, batch 23550, libri_loss[loss=0.4163, simple_loss=0.4423, pruned_loss=0.1951, over 19559.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3953, pruned_loss=0.1425, over 5676507.13 frames. ], libri_tot_loss[loss=0.3402, simple_loss=0.3973, pruned_loss=0.1415, over 5730044.48 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3926, pruned_loss=0.14, over 5687859.15 frames. ], batch size: 186, lr: 7.71e-03, grad_scale: 4.0
+2023-03-02 06:16:59,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:17:02,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:17:24,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:17:31,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:17:43,588 INFO [train.py:968] (0/2) Epoch 4, batch 23600, giga_loss[loss=0.3525, simple_loss=0.4131, pruned_loss=0.1459, over 28952.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.4015, pruned_loss=0.1484, over 5662737.10 frames. ], libri_tot_loss[loss=0.3403, simple_loss=0.3973, pruned_loss=0.1416, over 5722784.99 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.3993, pruned_loss=0.1465, over 5677491.39 frames. ], batch size: 136, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:17:46,541 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159746.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 06:17:49,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159749.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 06:18:17,043 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159778.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 06:18:23,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6917, 5.3463, 5.2410, 2.2756], device='cuda:0'), covar=tensor([0.0451, 0.0563, 0.1043, 0.1946], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0683, 0.0795, 0.0587], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 06:18:31,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.275e+03 1.788e+03 2.438e+03 3.237e+03 8.008e+03, threshold=4.875e+03, percent-clipped=10.0
+2023-03-02 06:18:31,792 INFO [train.py:968] (0/2) Epoch 4, batch 23650, libri_loss[loss=0.3228, simple_loss=0.377, pruned_loss=0.1343, over 29581.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4076, pruned_loss=0.154, over 5674466.82 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3979, pruned_loss=0.1423, over 5729545.07 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4056, pruned_loss=0.1522, over 5678145.79 frames. ], batch size: 75, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:19:08,273 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 06:19:15,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:19:21,365 INFO [train.py:968] (0/2) Epoch 4, batch 23700, giga_loss[loss=0.3629, simple_loss=0.4157, pruned_loss=0.1551, over 28779.00 frames. ], tot_loss[loss=0.364, simple_loss=0.4124, pruned_loss=0.1578, over 5680341.29 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.398, pruned_loss=0.1425, over 5732536.35 frames. ], giga_tot_loss[loss=0.3619, simple_loss=0.411, pruned_loss=0.1564, over 5679521.60 frames. ], batch size: 284, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:19:36,793 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-02 06:20:10,404 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.105e+02 1.680e+03 2.306e+03 2.943e+03 8.242e+03, threshold=4.613e+03, percent-clipped=10.0
+2023-03-02 06:20:10,418 INFO [train.py:968] (0/2) Epoch 4, batch 23750, giga_loss[loss=0.4474, simple_loss=0.4485, pruned_loss=0.2231, over 23548.00 frames. ], tot_loss[loss=0.3674, simple_loss=0.4138, pruned_loss=0.1605, over 5665529.47 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3976, pruned_loss=0.1424, over 5733191.96 frames. ], giga_tot_loss[loss=0.3662, simple_loss=0.4131, pruned_loss=0.1597, over 5663772.58 frames. ], batch size: 705, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:20:42,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:20:58,347 INFO [train.py:968] (0/2) Epoch 4, batch 23800, giga_loss[loss=0.3943, simple_loss=0.4367, pruned_loss=0.176, over 28613.00 frames. ], tot_loss[loss=0.3738, simple_loss=0.4175, pruned_loss=0.1651, over 5654220.99 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.3976, pruned_loss=0.1426, over 5729072.15 frames. ], giga_tot_loss[loss=0.3735, simple_loss=0.4175, pruned_loss=0.1648, over 5654621.07 frames. ], batch size: 262, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:21:39,806 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:21:43,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:21:50,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.664e+02 1.606e+03 2.179e+03 3.127e+03 6.575e+03, threshold=4.359e+03, percent-clipped=5.0
+2023-03-02 06:21:50,890 INFO [train.py:968] (0/2) Epoch 4, batch 23850, libri_loss[loss=0.4097, simple_loss=0.434, pruned_loss=0.1927, over 29538.00 frames. ], tot_loss[loss=0.3822, simple_loss=0.4226, pruned_loss=0.1709, over 5648496.97 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.3978, pruned_loss=0.1429, over 5729964.79 frames. ], giga_tot_loss[loss=0.3819, simple_loss=0.4226, pruned_loss=0.1706, over 5647507.89 frames. ], batch size: 81, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:22:00,098 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-160000.pt
+2023-03-02 06:22:15,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:22:30,641 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=160024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:22:50,938 INFO [train.py:968] (0/2) Epoch 4, batch 23900, giga_loss[loss=0.4153, simple_loss=0.4422, pruned_loss=0.1942, over 28928.00 frames. ], tot_loss[loss=0.3863, simple_loss=0.4259, pruned_loss=0.1733, over 5644597.29 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3981, pruned_loss=0.1434, over 5728736.64 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4259, pruned_loss=0.173, over 5643819.36 frames. ], batch size: 186, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:22:52,614 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7251, 5.3466, 5.3240, 2.5046], device='cuda:0'), covar=tensor([0.0361, 0.0366, 0.0745, 0.1782], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0693, 0.0808, 0.0593], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 06:23:08,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:23:22,495 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:23:25,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:23:43,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4915, 1.3625, 1.4968, 1.3571], device='cuda:0'), covar=tensor([0.0935, 0.1344, 0.1479, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0756, 0.0642, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 06:23:47,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.991e+02 1.731e+03 2.203e+03 2.942e+03 1.002e+04, threshold=4.405e+03, percent-clipped=8.0
+2023-03-02 06:23:47,073 INFO [train.py:968] (0/2) Epoch 4, batch 23950, giga_loss[loss=0.4513, simple_loss=0.4601, pruned_loss=0.2213, over 26565.00 frames. ], tot_loss[loss=0.3864, simple_loss=0.4256, pruned_loss=0.1736, over 5645601.86 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.398, pruned_loss=0.1434, over 5732146.39 frames. ], giga_tot_loss[loss=0.387, simple_loss=0.4262, pruned_loss=0.1739, over 5640709.88 frames. ], batch size: 555, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:23:55,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:24:04,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0823, 1.7053, 1.4495, 1.4042], device='cuda:0'), covar=tensor([0.1219, 0.1791, 0.1104, 0.1051], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0758, 0.0755, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 06:24:12,349 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-02 06:24:20,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3669, 1.3639, 1.2374, 1.3009], device='cuda:0'), covar=tensor([0.1107, 0.1684, 0.1550, 0.1446], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0766, 0.0646, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 06:24:35,140 INFO [train.py:968] (0/2) Epoch 4, batch 24000, giga_loss[loss=0.4092, simple_loss=0.4159, pruned_loss=0.2013, over 23177.00 frames. ], tot_loss[loss=0.3837, simple_loss=0.4232, pruned_loss=0.172, over 5644646.96 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.398, pruned_loss=0.1434, over 5735835.34 frames. ], giga_tot_loss[loss=0.3849, simple_loss=0.4242, pruned_loss=0.1728, over 5636093.64 frames. ], batch size: 705, lr: 7.70e-03, grad_scale: 8.0
+2023-03-02 06:24:35,144 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 06:24:43,740 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2484, simple_loss=0.3515, pruned_loss=0.07269, over 944034.00 frames. 
+2023-03-02 06:24:43,741 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 06:25:29,081 INFO [train.py:968] (0/2) Epoch 4, batch 24050, giga_loss[loss=0.4618, simple_loss=0.4707, pruned_loss=0.2264, over 26674.00 frames. ], tot_loss[loss=0.3824, simple_loss=0.4229, pruned_loss=0.1709, over 5647865.86 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3991, pruned_loss=0.1444, over 5736316.32 frames. ], giga_tot_loss[loss=0.3832, simple_loss=0.4235, pruned_loss=0.1715, over 5637975.82 frames. ], batch size: 555, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:25:29,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.190e+02 1.671e+03 2.102e+03 2.863e+03 7.259e+03, threshold=4.204e+03, percent-clipped=8.0
+2023-03-02 06:25:34,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:25:37,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:26:13,084 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:26:21,714 INFO [train.py:968] (0/2) Epoch 4, batch 24100, giga_loss[loss=0.3906, simple_loss=0.4374, pruned_loss=0.1719, over 29000.00 frames. ], tot_loss[loss=0.3812, simple_loss=0.4228, pruned_loss=0.1698, over 5638153.48 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3991, pruned_loss=0.1445, over 5728611.29 frames. ], giga_tot_loss[loss=0.3822, simple_loss=0.4235, pruned_loss=0.1704, over 5637254.91 frames. ], batch size: 213, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:26:27,428 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=160248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:27:10,656 INFO [train.py:968] (0/2) Epoch 4, batch 24150, giga_loss[loss=0.3882, simple_loss=0.4027, pruned_loss=0.1868, over 23920.00 frames. ], tot_loss[loss=0.3817, simple_loss=0.4231, pruned_loss=0.1701, over 5625452.27 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.3996, pruned_loss=0.1452, over 5729735.07 frames. ], giga_tot_loss[loss=0.3825, simple_loss=0.4238, pruned_loss=0.1706, over 5621418.17 frames. ], batch size: 710, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:27:12,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.710e+03 2.301e+03 3.279e+03 8.681e+03, threshold=4.602e+03, percent-clipped=10.0
+2023-03-02 06:27:16,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9306, 1.0477, 3.9474, 3.2462], device='cuda:0'), covar=tensor([0.2197, 0.2656, 0.0571, 0.0688], device='cuda:0'), in_proj_covar=tensor([0.0564, 0.0521, 0.0732, 0.0598], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 06:27:23,811 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=160303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:28:00,724 INFO [train.py:968] (0/2) Epoch 4, batch 24200, giga_loss[loss=0.4067, simple_loss=0.4279, pruned_loss=0.1928, over 23797.00 frames. ], tot_loss[loss=0.3777, simple_loss=0.4208, pruned_loss=0.1673, over 5629598.51 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.4004, pruned_loss=0.1459, over 5728157.88 frames. ], giga_tot_loss[loss=0.3779, simple_loss=0.4209, pruned_loss=0.1674, over 5626648.68 frames. ], batch size: 705, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:28:42,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4805, 1.1594, 5.0416, 3.6537], device='cuda:0'), covar=tensor([0.1504, 0.2324, 0.0325, 0.0519], device='cuda:0'), in_proj_covar=tensor([0.0557, 0.0517, 0.0721, 0.0589], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 06:28:53,596 INFO [train.py:968] (0/2) Epoch 4, batch 24250, giga_loss[loss=0.3584, simple_loss=0.4164, pruned_loss=0.1502, over 29043.00 frames. ], tot_loss[loss=0.3718, simple_loss=0.4176, pruned_loss=0.163, over 5616480.39 frames. ], libri_tot_loss[loss=0.3466, simple_loss=0.4007, pruned_loss=0.1462, over 5710444.50 frames. ], giga_tot_loss[loss=0.3718, simple_loss=0.4177, pruned_loss=0.163, over 5628591.32 frames. ], batch size: 136, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:28:54,475 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.559e+02 1.617e+03 2.194e+03 2.847e+03 9.720e+03, threshold=4.388e+03, percent-clipped=6.0
+2023-03-02 06:29:00,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:29:42,203 INFO [train.py:968] (0/2) Epoch 4, batch 24300, giga_loss[loss=0.3189, simple_loss=0.3803, pruned_loss=0.1287, over 28339.00 frames. ], tot_loss[loss=0.3662, simple_loss=0.4141, pruned_loss=0.1592, over 5636220.37 frames. ], libri_tot_loss[loss=0.3464, simple_loss=0.4004, pruned_loss=0.1461, over 5716589.38 frames. ], giga_tot_loss[loss=0.367, simple_loss=0.4148, pruned_loss=0.1596, over 5638290.97 frames. ], batch size: 369, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:30:13,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7068, 3.1870, 1.6852, 1.6858], device='cuda:0'), covar=tensor([0.0794, 0.0373, 0.0775, 0.1212], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0474, 0.0313, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 06:30:28,500 INFO [train.py:968] (0/2) Epoch 4, batch 24350, giga_loss[loss=0.371, simple_loss=0.4218, pruned_loss=0.1601, over 29083.00 frames. ], tot_loss[loss=0.3619, simple_loss=0.4114, pruned_loss=0.1562, over 5647273.01 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.4002, pruned_loss=0.1462, over 5708744.35 frames. ], giga_tot_loss[loss=0.363, simple_loss=0.4124, pruned_loss=0.1568, over 5654758.14 frames. ], batch size: 155, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:30:29,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.285e+02 1.578e+03 1.990e+03 2.696e+03 6.205e+03, threshold=3.979e+03, percent-clipped=4.0
+2023-03-02 06:30:52,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3164, 3.0879, 3.0054, 1.8576], device='cuda:0'), covar=tensor([0.0604, 0.0583, 0.0840, 0.1688], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0695, 0.0801, 0.0585], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 06:30:54,491 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 06:31:14,037 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:31:14,335 INFO [train.py:968] (0/2) Epoch 4, batch 24400, giga_loss[loss=0.3837, simple_loss=0.4221, pruned_loss=0.1726, over 27902.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.4082, pruned_loss=0.1542, over 5643094.49 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.4007, pruned_loss=0.147, over 5709914.17 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.409, pruned_loss=0.1544, over 5646293.02 frames. ], batch size: 412, lr: 7.69e-03, grad_scale: 8.0
+2023-03-02 06:31:17,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:31:27,443 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-02 06:31:44,547 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:31:56,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2263, 3.5163, 2.4439, 0.8676], device='cuda:0'), covar=tensor([0.2026, 0.0871, 0.1324, 0.2279], device='cuda:0'), in_proj_covar=tensor([0.1350, 0.1251, 0.1339, 0.1120], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 06:32:02,621 INFO [train.py:968] (0/2) Epoch 4, batch 24450, giga_loss[loss=0.4263, simple_loss=0.4599, pruned_loss=0.1963, over 29035.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.4062, pruned_loss=0.1522, over 5665742.40 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.4, pruned_loss=0.1466, over 5714517.49 frames. ], giga_tot_loss[loss=0.3566, simple_loss=0.4076, pruned_loss=0.1528, over 5662501.39 frames. ], batch size: 164, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:32:03,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.831e+02 1.563e+03 2.045e+03 2.805e+03 5.544e+03, threshold=4.091e+03, percent-clipped=8.0
+2023-03-02 06:32:35,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:32:55,184 INFO [train.py:968] (0/2) Epoch 4, batch 24500, libri_loss[loss=0.4026, simple_loss=0.4297, pruned_loss=0.1878, over 19407.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4064, pruned_loss=0.1524, over 5656106.75 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.4002, pruned_loss=0.1468, over 5707758.79 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4077, pruned_loss=0.1528, over 5658637.42 frames. ], batch size: 187, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:33:28,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:33:41,978 INFO [train.py:968] (0/2) Epoch 4, batch 24550, libri_loss[loss=0.3969, simple_loss=0.4384, pruned_loss=0.1776, over 29396.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.403, pruned_loss=0.149, over 5664467.67 frames. ], libri_tot_loss[loss=0.3477, simple_loss=0.4006, pruned_loss=0.1474, over 5711970.48 frames. ], giga_tot_loss[loss=0.3509, simple_loss=0.4038, pruned_loss=0.149, over 5660669.19 frames. ], batch size: 92, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:33:45,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.568e+02 1.564e+03 2.123e+03 2.854e+03 1.239e+04, threshold=4.246e+03, percent-clipped=12.0
+2023-03-02 06:33:49,520 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=160700.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:34:31,260 INFO [train.py:968] (0/2) Epoch 4, batch 24600, giga_loss[loss=0.3581, simple_loss=0.4297, pruned_loss=0.1433, over 28962.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4041, pruned_loss=0.147, over 5681719.74 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.4004, pruned_loss=0.1473, over 5715933.27 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4049, pruned_loss=0.1471, over 5674094.95 frames. ], batch size: 145, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:34:36,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3581, 1.5314, 1.2014, 0.8289], device='cuda:0'), covar=tensor([0.1093, 0.0820, 0.0646, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.1337, 0.1073, 0.1093, 0.1178], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 06:34:53,075 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:34:54,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:35:17,360 INFO [train.py:968] (0/2) Epoch 4, batch 24650, giga_loss[loss=0.3767, simple_loss=0.4277, pruned_loss=0.1628, over 28000.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4055, pruned_loss=0.1476, over 5660266.27 frames. ], libri_tot_loss[loss=0.3472, simple_loss=0.4, pruned_loss=0.1472, over 5713360.73 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4068, pruned_loss=0.1479, over 5654320.75 frames. ], batch size: 412, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:35:20,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.606e+02 1.514e+03 1.974e+03 2.810e+03 6.250e+03, threshold=3.949e+03, percent-clipped=4.0
+2023-03-02 06:35:23,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:35:23,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9482, 2.0207, 1.8117, 1.7709], device='cuda:0'), covar=tensor([0.0978, 0.1462, 0.1256, 0.1150], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0740, 0.0627, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 06:35:44,980 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:35:47,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:35:57,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3466, 1.5434, 1.2125, 0.9835], device='cuda:0'), covar=tensor([0.0841, 0.0587, 0.0590, 0.0763], device='cuda:0'), in_proj_covar=tensor([0.1334, 0.1072, 0.1097, 0.1173], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 06:36:06,195 INFO [train.py:968] (0/2) Epoch 4, batch 24700, giga_loss[loss=0.3535, simple_loss=0.4078, pruned_loss=0.1496, over 28288.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.4073, pruned_loss=0.1501, over 5667542.98 frames. ], libri_tot_loss[loss=0.3474, simple_loss=0.3998, pruned_loss=0.1474, over 5718107.62 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4087, pruned_loss=0.15, over 5657013.49 frames. ], batch size: 368, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:36:17,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:36:24,017 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-02 06:36:52,895 INFO [train.py:968] (0/2) Epoch 4, batch 24750, giga_loss[loss=0.3326, simple_loss=0.3913, pruned_loss=0.137, over 28675.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4082, pruned_loss=0.1518, over 5667182.00 frames. ], libri_tot_loss[loss=0.3482, simple_loss=0.4003, pruned_loss=0.1481, over 5722529.96 frames. ], giga_tot_loss[loss=0.3559, simple_loss=0.4091, pruned_loss=0.1513, over 5653482.86 frames. ], batch size: 262, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:36:54,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.610e+02 1.723e+03 2.355e+03 3.967e+03 1.119e+04, threshold=4.709e+03, percent-clipped=28.0
+2023-03-02 06:36:56,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2533, 4.0249, 3.9782, 1.7355], device='cuda:0'), covar=tensor([0.0440, 0.0392, 0.0733, 0.2008], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0706, 0.0817, 0.0590], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 06:37:44,055 INFO [train.py:968] (0/2) Epoch 4, batch 24800, libri_loss[loss=0.351, simple_loss=0.4062, pruned_loss=0.1478, over 29512.00 frames. ], tot_loss[loss=0.3545, simple_loss=0.4059, pruned_loss=0.1515, over 5661426.99 frames. ], libri_tot_loss[loss=0.3481, simple_loss=0.4002, pruned_loss=0.148, over 5725303.96 frames. ], giga_tot_loss[loss=0.3546, simple_loss=0.4068, pruned_loss=0.1512, over 5646966.54 frames. ], batch size: 81, lr: 7.68e-03, grad_scale: 8.0
+2023-03-02 06:37:59,511 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-02 06:38:27,601 INFO [train.py:968] (0/2) Epoch 4, batch 24850, giga_loss[loss=0.3208, simple_loss=0.3792, pruned_loss=0.1312, over 28985.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.4042, pruned_loss=0.1506, over 5673968.39 frames. ], libri_tot_loss[loss=0.348, simple_loss=0.4002, pruned_loss=0.1479, over 5727669.29 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4049, pruned_loss=0.1505, over 5659631.00 frames. ], batch size: 128, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:38:30,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.155e+03 1.626e+03 1.938e+03 2.682e+03 7.445e+03, threshold=3.876e+03, percent-clipped=4.0
+2023-03-02 06:38:44,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3608, 2.7543, 1.4790, 1.3297], device='cuda:0'), covar=tensor([0.0748, 0.0354, 0.0741, 0.1179], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0471, 0.0307, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 06:39:07,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-02 06:39:10,362 INFO [train.py:968] (0/2) Epoch 4, batch 24900, giga_loss[loss=0.3479, simple_loss=0.4129, pruned_loss=0.1414, over 28998.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4021, pruned_loss=0.1488, over 5674048.29 frames. ], libri_tot_loss[loss=0.3483, simple_loss=0.4003, pruned_loss=0.1481, over 5727625.71 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4027, pruned_loss=0.1486, over 5659768.01 frames. ], batch size: 213, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:39:37,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161075.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 06:39:56,467 INFO [train.py:968] (0/2) Epoch 4, batch 24950, giga_loss[loss=0.3972, simple_loss=0.4279, pruned_loss=0.1832, over 27589.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.4008, pruned_loss=0.1463, over 5683375.65 frames. ], libri_tot_loss[loss=0.3482, simple_loss=0.4001, pruned_loss=0.1481, over 5730079.54 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.4015, pruned_loss=0.1462, over 5669139.36 frames. ], batch size: 472, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:40:00,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.304e+02 1.524e+03 1.944e+03 2.483e+03 4.725e+03, threshold=3.889e+03, percent-clipped=10.0
+2023-03-02 06:40:26,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7436, 1.9644, 1.9117, 1.8052], device='cuda:0'), covar=tensor([0.1502, 0.1763, 0.1169, 0.1096], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0769, 0.0759, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 06:40:47,225 INFO [train.py:968] (0/2) Epoch 4, batch 25000, giga_loss[loss=0.3875, simple_loss=0.4417, pruned_loss=0.1666, over 28879.00 frames. ], tot_loss[loss=0.348, simple_loss=0.4018, pruned_loss=0.1471, over 5676441.24 frames. ], libri_tot_loss[loss=0.3489, simple_loss=0.4006, pruned_loss=0.1485, over 5733280.93 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.4019, pruned_loss=0.1466, over 5660869.28 frames. ], batch size: 174, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:41:39,068 INFO [train.py:968] (0/2) Epoch 4, batch 25050, giga_loss[loss=0.3115, simple_loss=0.3803, pruned_loss=0.1213, over 28556.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.4015, pruned_loss=0.147, over 5674712.21 frames. ], libri_tot_loss[loss=0.3491, simple_loss=0.4008, pruned_loss=0.1487, over 5730579.53 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.4014, pruned_loss=0.1464, over 5664744.34 frames. ], batch size: 71, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:41:41,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.524e+02 1.561e+03 2.051e+03 2.676e+03 8.363e+03, threshold=4.102e+03, percent-clipped=13.0
+2023-03-02 06:42:03,434 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161218.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:42:05,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161221.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:42:29,815 INFO [train.py:968] (0/2) Epoch 4, batch 25100, libri_loss[loss=0.3629, simple_loss=0.4196, pruned_loss=0.1531, over 29537.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.3994, pruned_loss=0.1457, over 5683723.01 frames. ], libri_tot_loss[loss=0.3493, simple_loss=0.401, pruned_loss=0.1488, over 5732349.28 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3991, pruned_loss=0.1451, over 5673822.83 frames. ], batch size: 89, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:42:36,767 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161250.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 06:43:16,532 INFO [train.py:968] (0/2) Epoch 4, batch 25150, giga_loss[loss=0.3751, simple_loss=0.4129, pruned_loss=0.1687, over 27604.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4003, pruned_loss=0.1473, over 5683204.21 frames. ], libri_tot_loss[loss=0.3498, simple_loss=0.4013, pruned_loss=0.1492, over 5733167.21 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3999, pruned_loss=0.1465, over 5672947.90 frames. ], batch size: 472, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:43:17,661 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:43:18,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.165e+02 1.658e+03 2.199e+03 3.262e+03 9.573e+03, threshold=4.399e+03, percent-clipped=19.0
+2023-03-02 06:44:00,980 INFO [train.py:968] (0/2) Epoch 4, batch 25200, giga_loss[loss=0.3735, simple_loss=0.4123, pruned_loss=0.1674, over 29017.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.3992, pruned_loss=0.147, over 5691341.86 frames. ], libri_tot_loss[loss=0.3503, simple_loss=0.4017, pruned_loss=0.1495, over 5732559.89 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3984, pruned_loss=0.146, over 5682925.95 frames. ], batch size: 106, lr: 7.67e-03, grad_scale: 8.0
+2023-03-02 06:44:04,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2104, 4.0148, 3.9197, 1.6255], device='cuda:0'), covar=tensor([0.0439, 0.0401, 0.0730, 0.2035], device='cuda:0'), in_proj_covar=tensor([0.0829, 0.0716, 0.0829, 0.0601], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 06:44:14,246 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:44:48,107 INFO [train.py:968] (0/2) Epoch 4, batch 25250, giga_loss[loss=0.3722, simple_loss=0.4119, pruned_loss=0.1663, over 28734.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3968, pruned_loss=0.1456, over 5682455.80 frames. ], libri_tot_loss[loss=0.3499, simple_loss=0.4013, pruned_loss=0.1493, over 5726176.09 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.3964, pruned_loss=0.1449, over 5680027.50 frames. ], batch size: 242, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:44:51,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.109e+03 1.689e+03 2.116e+03 3.442e+03 5.209e+03, threshold=4.232e+03, percent-clipped=7.0
+2023-03-02 06:45:35,091 INFO [train.py:968] (0/2) Epoch 4, batch 25300, giga_loss[loss=0.343, simple_loss=0.3957, pruned_loss=0.1451, over 28658.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.396, pruned_loss=0.1458, over 5689546.48 frames. ], libri_tot_loss[loss=0.3497, simple_loss=0.4011, pruned_loss=0.1492, over 5730399.24 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3958, pruned_loss=0.1453, over 5683020.26 frames. ], batch size: 262, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:45:55,725 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-02 06:46:21,382 INFO [train.py:968] (0/2) Epoch 4, batch 25350, giga_loss[loss=0.3427, simple_loss=0.3952, pruned_loss=0.1451, over 28497.00 frames. ], tot_loss[loss=0.3456, simple_loss=0.3972, pruned_loss=0.147, over 5682821.76 frames. ], libri_tot_loss[loss=0.3498, simple_loss=0.4011, pruned_loss=0.1493, over 5727313.52 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.3968, pruned_loss=0.1465, over 5678197.43 frames. ], batch size: 71, lr: 7.66e-03, grad_scale: 2.0
+2023-03-02 06:46:24,998 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.808e+02 1.553e+03 2.279e+03 2.840e+03 1.038e+04, threshold=4.558e+03, percent-clipped=11.0
+2023-03-02 06:47:06,407 INFO [train.py:968] (0/2) Epoch 4, batch 25400, giga_loss[loss=0.3137, simple_loss=0.3875, pruned_loss=0.12, over 29045.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3976, pruned_loss=0.1462, over 5673054.32 frames. ], libri_tot_loss[loss=0.3506, simple_loss=0.4015, pruned_loss=0.1499, over 5713867.17 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3968, pruned_loss=0.1451, over 5680427.59 frames. ], batch size: 155, lr: 7.66e-03, grad_scale: 2.0
+2023-03-02 06:47:50,864 INFO [train.py:968] (0/2) Epoch 4, batch 25450, giga_loss[loss=0.363, simple_loss=0.4086, pruned_loss=0.1587, over 27997.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3972, pruned_loss=0.1452, over 5675897.22 frames. ], libri_tot_loss[loss=0.3504, simple_loss=0.4014, pruned_loss=0.1497, over 5716251.76 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3967, pruned_loss=0.1444, over 5678980.15 frames. ], batch size: 412, lr: 7.66e-03, grad_scale: 2.0
+2023-03-02 06:47:57,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.297e+02 1.520e+03 2.064e+03 2.889e+03 1.559e+04, threshold=4.128e+03, percent-clipped=7.0
+2023-03-02 06:48:06,559 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:48:18,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7191, 4.3900, 1.8439, 1.7140], device='cuda:0'), covar=tensor([0.0806, 0.0292, 0.0759, 0.1212], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0467, 0.0305, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 06:48:21,308 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:48:39,664 INFO [train.py:968] (0/2) Epoch 4, batch 25500, giga_loss[loss=0.3559, simple_loss=0.4112, pruned_loss=0.1503, over 29048.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.398, pruned_loss=0.1451, over 5678855.49 frames. ], libri_tot_loss[loss=0.3508, simple_loss=0.4017, pruned_loss=0.15, over 5718185.39 frames. ], giga_tot_loss[loss=0.3429, simple_loss=0.3973, pruned_loss=0.1442, over 5679145.26 frames. ], batch size: 106, lr: 7.66e-03, grad_scale: 2.0
+2023-03-02 06:49:02,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:49:24,550 INFO [train.py:968] (0/2) Epoch 4, batch 25550, giga_loss[loss=0.3272, simple_loss=0.39, pruned_loss=0.1322, over 28823.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.4007, pruned_loss=0.1479, over 5683238.93 frames. ], libri_tot_loss[loss=0.3504, simple_loss=0.4012, pruned_loss=0.1498, over 5721804.74 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.4004, pruned_loss=0.1473, over 5679527.98 frames. ], batch size: 285, lr: 7.66e-03, grad_scale: 2.0
+2023-03-02 06:49:31,285 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.370e+02 1.519e+03 1.859e+03 2.455e+03 5.186e+03, threshold=3.719e+03, percent-clipped=1.0
+2023-03-02 06:49:46,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5054, 3.5018, 1.5701, 1.5715], device='cuda:0'), covar=tensor([0.0878, 0.0308, 0.0862, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0469, 0.0309, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 06:50:04,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:50:14,149 INFO [train.py:968] (0/2) Epoch 4, batch 25600, giga_loss[loss=0.2829, simple_loss=0.3531, pruned_loss=0.1064, over 28625.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4031, pruned_loss=0.1509, over 5684688.08 frames. ], libri_tot_loss[loss=0.3502, simple_loss=0.4011, pruned_loss=0.1497, over 5724679.60 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4031, pruned_loss=0.1505, over 5678611.41 frames. ], batch size: 60, lr: 7.66e-03, grad_scale: 4.0
+2023-03-02 06:51:05,431 INFO [train.py:968] (0/2) Epoch 4, batch 25650, giga_loss[loss=0.3189, simple_loss=0.375, pruned_loss=0.1314, over 28147.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4039, pruned_loss=0.1529, over 5679042.39 frames. ], libri_tot_loss[loss=0.3505, simple_loss=0.4014, pruned_loss=0.1499, over 5724697.27 frames. ], giga_tot_loss[loss=0.3543, simple_loss=0.4036, pruned_loss=0.1525, over 5673934.89 frames. ], batch size: 77, lr: 7.66e-03, grad_scale: 4.0
+2023-03-02 06:51:07,988 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:51:12,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.982e+02 1.536e+03 2.206e+03 3.021e+03 1.366e+04, threshold=4.412e+03, percent-clipped=11.0
+2023-03-02 06:51:24,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:51:26,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:51:56,085 INFO [train.py:968] (0/2) Epoch 4, batch 25700, libri_loss[loss=0.4266, simple_loss=0.4527, pruned_loss=0.2003, over 19590.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4056, pruned_loss=0.1553, over 5673521.78 frames. ], libri_tot_loss[loss=0.3512, simple_loss=0.4018, pruned_loss=0.1503, over 5716769.89 frames. ], giga_tot_loss[loss=0.3572, simple_loss=0.4051, pruned_loss=0.1546, over 5676272.78 frames. ], batch size: 186, lr: 7.66e-03, grad_scale: 4.0
+2023-03-02 06:51:56,929 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:52:04,975 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:52:25,377 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:52:27,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:52:40,882 INFO [train.py:968] (0/2) Epoch 4, batch 25750, giga_loss[loss=0.3846, simple_loss=0.4161, pruned_loss=0.1766, over 28855.00 frames. ], tot_loss[loss=0.3568, simple_loss=0.4044, pruned_loss=0.1546, over 5664905.65 frames. ], libri_tot_loss[loss=0.3513, simple_loss=0.4018, pruned_loss=0.1503, over 5710964.15 frames. ], giga_tot_loss[loss=0.3561, simple_loss=0.404, pruned_loss=0.1541, over 5671079.72 frames. ], batch size: 186, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:52:45,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.668e+03 2.277e+03 3.244e+03 8.891e+03, threshold=4.554e+03, percent-clipped=12.0
+2023-03-02 06:52:54,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:53:26,911 INFO [train.py:968] (0/2) Epoch 4, batch 25800, giga_loss[loss=0.3619, simple_loss=0.4202, pruned_loss=0.1518, over 28500.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4032, pruned_loss=0.1532, over 5660192.22 frames. ], libri_tot_loss[loss=0.3516, simple_loss=0.4021, pruned_loss=0.1505, over 5707671.63 frames. ], giga_tot_loss[loss=0.3542, simple_loss=0.4027, pruned_loss=0.1528, over 5666887.03 frames. ], batch size: 336, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:53:32,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7467, 4.2928, 1.7308, 1.7829], device='cuda:0'), covar=tensor([0.0866, 0.0280, 0.0842, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0469, 0.0308, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0019], device='cuda:0')
+2023-03-02 06:54:02,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:54:09,601 INFO [train.py:968] (0/2) Epoch 4, batch 25850, giga_loss[loss=0.2948, simple_loss=0.3674, pruned_loss=0.1111, over 28849.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4013, pruned_loss=0.1499, over 5662262.35 frames. ], libri_tot_loss[loss=0.3515, simple_loss=0.402, pruned_loss=0.1506, over 5700855.16 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.4011, pruned_loss=0.1496, over 5671992.50 frames. ], batch size: 99, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:54:13,379 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.327e+02 1.511e+03 1.974e+03 2.509e+03 4.640e+03, threshold=3.949e+03, percent-clipped=1.0
+2023-03-02 06:54:15,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:54:16,425 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-162000.pt
+2023-03-02 06:54:57,790 INFO [train.py:968] (0/2) Epoch 4, batch 25900, giga_loss[loss=0.3589, simple_loss=0.4056, pruned_loss=0.1561, over 27907.00 frames. ], tot_loss[loss=0.347, simple_loss=0.3983, pruned_loss=0.1478, over 5645802.13 frames. ], libri_tot_loss[loss=0.3519, simple_loss=0.4022, pruned_loss=0.1508, over 5693318.76 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.3979, pruned_loss=0.1473, over 5660177.01 frames. ], batch size: 412, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:55:43,821 INFO [train.py:968] (0/2) Epoch 4, batch 25950, giga_loss[loss=0.2857, simple_loss=0.3594, pruned_loss=0.106, over 29009.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3953, pruned_loss=0.1461, over 5655568.70 frames. ], libri_tot_loss[loss=0.3508, simple_loss=0.4013, pruned_loss=0.1502, over 5696130.51 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3956, pruned_loss=0.1462, over 5663590.58 frames. ], batch size: 164, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:55:50,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.753e+02 1.606e+03 2.075e+03 3.319e+03 9.436e+03, threshold=4.150e+03, percent-clipped=16.0
+2023-03-02 06:56:18,780 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:56:22,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:56:34,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2911, 1.3816, 1.2985, 1.1377], device='cuda:0'), covar=tensor([0.1847, 0.1848, 0.1695, 0.1749], device='cuda:0'), in_proj_covar=tensor([0.1089, 0.0865, 0.0970, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 06:56:35,911 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:56:36,917 INFO [train.py:968] (0/2) Epoch 4, batch 26000, giga_loss[loss=0.3157, simple_loss=0.3761, pruned_loss=0.1276, over 28651.00 frames. ], tot_loss[loss=0.348, simple_loss=0.3976, pruned_loss=0.1492, over 5642283.76 frames. ], libri_tot_loss[loss=0.351, simple_loss=0.4015, pruned_loss=0.1503, over 5696532.15 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.3977, pruned_loss=0.1492, over 5647955.34 frames. ], batch size: 307, lr: 7.65e-03, grad_scale: 8.0
+2023-03-02 06:56:38,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:56:43,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8426, 2.4655, 2.1320, 1.9882], device='cuda:0'), covar=tensor([0.1337, 0.1530, 0.1051, 0.1028], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0767, 0.0766, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 06:56:50,748 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:56:53,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0067, 1.9245, 1.8060, 1.8152], device='cuda:0'), covar=tensor([0.0995, 0.1680, 0.1514, 0.1378], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0746, 0.0629, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 06:57:00,586 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:57:04,165 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:57:20,376 INFO [train.py:968] (0/2) Epoch 4, batch 26050, giga_loss[loss=0.3687, simple_loss=0.4151, pruned_loss=0.1611, over 28630.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4006, pruned_loss=0.1514, over 5649551.43 frames. ], libri_tot_loss[loss=0.3521, simple_loss=0.4022, pruned_loss=0.151, over 5700582.80 frames. ], giga_tot_loss[loss=0.3505, simple_loss=0.3998, pruned_loss=0.1506, over 5648427.78 frames. ], batch size: 242, lr: 7.65e-03, grad_scale: 8.0
+2023-03-02 06:57:24,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4601, 1.5097, 4.9425, 3.5993], device='cuda:0'), covar=tensor([0.1497, 0.2022, 0.0306, 0.0537], device='cuda:0'), in_proj_covar=tensor([0.0569, 0.0520, 0.0736, 0.0598], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 06:57:24,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.740e+03 2.241e+03 2.999e+03 5.286e+03, threshold=4.481e+03, percent-clipped=5.0
+2023-03-02 06:57:32,743 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:57:50,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:58:02,532 INFO [train.py:968] (0/2) Epoch 4, batch 26100, giga_loss[loss=0.3801, simple_loss=0.4322, pruned_loss=0.164, over 28585.00 frames. ], tot_loss[loss=0.3545, simple_loss=0.4048, pruned_loss=0.1521, over 5646267.10 frames. ], libri_tot_loss[loss=0.3522, simple_loss=0.4022, pruned_loss=0.1511, over 5685418.76 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.4042, pruned_loss=0.1514, over 5657024.73 frames. ], batch size: 60, lr: 7.65e-03, grad_scale: 8.0
+2023-03-02 06:58:47,663 INFO [train.py:968] (0/2) Epoch 4, batch 26150, giga_loss[loss=0.3131, simple_loss=0.3861, pruned_loss=0.12, over 28936.00 frames. ], tot_loss[loss=0.3535, simple_loss=0.4057, pruned_loss=0.1506, over 5637410.90 frames. ], libri_tot_loss[loss=0.3524, simple_loss=0.402, pruned_loss=0.1514, over 5671790.43 frames. ], giga_tot_loss[loss=0.3525, simple_loss=0.4055, pruned_loss=0.1498, over 5658029.25 frames. ], batch size: 164, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:58:52,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.368e+02 1.433e+03 1.940e+03 2.677e+03 9.565e+03, threshold=3.880e+03, percent-clipped=6.0
+2023-03-02 06:59:05,935 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:59:07,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:59:23,400 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-02 06:59:29,176 INFO [train.py:968] (0/2) Epoch 4, batch 26200, giga_loss[loss=0.3413, simple_loss=0.4061, pruned_loss=0.1382, over 28886.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.4062, pruned_loss=0.1507, over 5644291.47 frames. ], libri_tot_loss[loss=0.3519, simple_loss=0.4015, pruned_loss=0.1511, over 5674539.34 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.4066, pruned_loss=0.1502, over 5657444.04 frames. ], batch size: 174, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 06:59:31,741 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:59:36,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5976, 1.9239, 1.8430, 1.7751], device='cuda:0'), covar=tensor([0.1424, 0.1699, 0.1061, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0768, 0.0764, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 06:59:44,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0024, 3.7938, 3.6765, 1.6699], device='cuda:0'), covar=tensor([0.0437, 0.0460, 0.0761, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.0829, 0.0717, 0.0824, 0.0589], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 06:59:58,445 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:00:00,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:00:18,973 INFO [train.py:968] (0/2) Epoch 4, batch 26250, giga_loss[loss=0.3282, simple_loss=0.3915, pruned_loss=0.1324, over 28966.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.409, pruned_loss=0.1535, over 5642235.17 frames. ], libri_tot_loss[loss=0.352, simple_loss=0.4016, pruned_loss=0.1511, over 5677716.02 frames. ], giga_tot_loss[loss=0.3578, simple_loss=0.4093, pruned_loss=0.1531, over 5649458.65 frames. ], batch size: 145, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:00:24,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.553e+03 1.976e+03 2.872e+03 1.191e+04, threshold=3.952e+03, percent-clipped=11.0
+2023-03-02 07:00:28,197 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:01:05,269 INFO [train.py:968] (0/2) Epoch 4, batch 26300, giga_loss[loss=0.4225, simple_loss=0.4435, pruned_loss=0.2008, over 27901.00 frames. ], tot_loss[loss=0.3606, simple_loss=0.4103, pruned_loss=0.1554, over 5639896.70 frames. ], libri_tot_loss[loss=0.3523, simple_loss=0.4019, pruned_loss=0.1513, over 5681112.80 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4104, pruned_loss=0.155, over 5642073.03 frames. ], batch size: 412, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:01:32,488 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:01:56,519 INFO [train.py:968] (0/2) Epoch 4, batch 26350, giga_loss[loss=0.4507, simple_loss=0.4658, pruned_loss=0.2178, over 27525.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.4088, pruned_loss=0.155, over 5639709.17 frames. ], libri_tot_loss[loss=0.3521, simple_loss=0.4017, pruned_loss=0.1512, over 5681491.07 frames. ], giga_tot_loss[loss=0.3593, simple_loss=0.4091, pruned_loss=0.1548, over 5640714.30 frames. ], batch size: 472, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:02:04,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.478e+02 1.533e+03 1.993e+03 2.559e+03 6.272e+03, threshold=3.987e+03, percent-clipped=7.0
+2023-03-02 07:02:44,812 INFO [train.py:968] (0/2) Epoch 4, batch 26400, giga_loss[loss=0.3632, simple_loss=0.4116, pruned_loss=0.1574, over 28865.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.4053, pruned_loss=0.1528, over 5645740.77 frames. ], libri_tot_loss[loss=0.3515, simple_loss=0.4013, pruned_loss=0.1509, over 5683610.47 frames. ], giga_tot_loss[loss=0.356, simple_loss=0.406, pruned_loss=0.153, over 5644488.66 frames. ], batch size: 174, lr: 7.64e-03, grad_scale: 8.0
+2023-03-02 07:03:24,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:03:38,193 INFO [train.py:968] (0/2) Epoch 4, batch 26450, giga_loss[loss=0.4495, simple_loss=0.4508, pruned_loss=0.2241, over 26557.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.4037, pruned_loss=0.1528, over 5646196.17 frames. ], libri_tot_loss[loss=0.3516, simple_loss=0.4013, pruned_loss=0.1509, over 5685846.84 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4042, pruned_loss=0.1529, over 5643019.11 frames. ], batch size: 555, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:03:43,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.712e+02 1.652e+03 2.465e+03 3.753e+03 9.254e+03, threshold=4.931e+03, percent-clipped=22.0
+2023-03-02 07:04:02,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5936, 1.8036, 1.8587, 1.8344], device='cuda:0'), covar=tensor([0.1116, 0.1331, 0.0871, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0771, 0.0763, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 07:04:06,660 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:04:24,499 INFO [train.py:968] (0/2) Epoch 4, batch 26500, giga_loss[loss=0.3248, simple_loss=0.3846, pruned_loss=0.1325, over 28954.00 frames. ], tot_loss[loss=0.3554, simple_loss=0.404, pruned_loss=0.1534, over 5636823.62 frames. ], libri_tot_loss[loss=0.3517, simple_loss=0.4014, pruned_loss=0.151, over 5678150.92 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4043, pruned_loss=0.1534, over 5640030.49 frames. ], batch size: 213, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:04:50,434 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4273, 2.1568, 1.6146, 0.5697], device='cuda:0'), covar=tensor([0.1367, 0.0906, 0.1411, 0.1821], device='cuda:0'), in_proj_covar=tensor([0.1317, 0.1243, 0.1312, 0.1113], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 07:05:06,631 INFO [train.py:968] (0/2) Epoch 4, batch 26550, giga_loss[loss=0.3436, simple_loss=0.3987, pruned_loss=0.1443, over 28625.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.4046, pruned_loss=0.1542, over 5650227.83 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4025, pruned_loss=0.1521, over 5685811.20 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.404, pruned_loss=0.1533, over 5644887.12 frames. ], batch size: 242, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:05:14,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.997e+02 1.781e+03 2.404e+03 2.976e+03 9.105e+03, threshold=4.808e+03, percent-clipped=4.0
+2023-03-02 07:05:36,690 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:05:40,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:05:52,007 INFO [train.py:968] (0/2) Epoch 4, batch 26600, giga_loss[loss=0.3517, simple_loss=0.3988, pruned_loss=0.1522, over 28186.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4017, pruned_loss=0.1526, over 5667642.45 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.4023, pruned_loss=0.152, over 5690101.04 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.4015, pruned_loss=0.152, over 5659179.04 frames. ], batch size: 368, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:06:06,576 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:06:07,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7232, 2.0272, 2.0004, 1.7914], device='cuda:0'), covar=tensor([0.1303, 0.1637, 0.0998, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0768, 0.0760, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 07:06:30,048 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9997, 1.2199, 1.1347, 1.1064], device='cuda:0'), covar=tensor([0.1011, 0.0983, 0.1390, 0.1040], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0760, 0.0648, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 07:06:38,438 INFO [train.py:968] (0/2) Epoch 4, batch 26650, giga_loss[loss=0.3858, simple_loss=0.4329, pruned_loss=0.1693, over 29045.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.4022, pruned_loss=0.1532, over 5663114.10 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4026, pruned_loss=0.1523, over 5683855.55 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4017, pruned_loss=0.1525, over 5661086.35 frames. ], batch size: 155, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:06:44,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.513e+02 1.635e+03 2.209e+03 3.035e+03 7.778e+03, threshold=4.419e+03, percent-clipped=6.0
+2023-03-02 07:06:56,810 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-02 07:07:15,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3582, 1.5670, 1.1872, 1.4002], device='cuda:0'), covar=tensor([0.0817, 0.0341, 0.0353, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0206, 0.0134, 0.0135, 0.0244], device='cuda:0'), out_proj_covar=tensor([0.0050, 0.0037, 0.0033, 0.0056], device='cuda:0')
+2023-03-02 07:07:20,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3416, 1.9098, 1.4939, 1.5591], device='cuda:0'), covar=tensor([0.0827, 0.0305, 0.0328, 0.0893], device='cuda:0'), in_proj_covar=tensor([0.0206, 0.0134, 0.0135, 0.0244], device='cuda:0'), out_proj_covar=tensor([0.0050, 0.0037, 0.0033, 0.0056], device='cuda:0')
+2023-03-02 07:07:22,080 INFO [train.py:968] (0/2) Epoch 4, batch 26700, giga_loss[loss=0.3315, simple_loss=0.3978, pruned_loss=0.1326, over 28929.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.4029, pruned_loss=0.1523, over 5668036.84 frames. ], libri_tot_loss[loss=0.3544, simple_loss=0.4032, pruned_loss=0.1528, over 5686985.18 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.4019, pruned_loss=0.1514, over 5663154.14 frames. ], batch size: 186, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:07:27,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:08:07,397 INFO [train.py:968] (0/2) Epoch 4, batch 26750, giga_loss[loss=0.338, simple_loss=0.3932, pruned_loss=0.1414, over 28528.00 frames. ], tot_loss[loss=0.3554, simple_loss=0.4047, pruned_loss=0.1531, over 5660136.65 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4035, pruned_loss=0.153, over 5683099.91 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.4037, pruned_loss=0.152, over 5660020.99 frames. ], batch size: 78, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:08:13,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.716e+02 1.535e+03 1.925e+03 2.444e+03 1.243e+04, threshold=3.849e+03, percent-clipped=9.0
+2023-03-02 07:08:19,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:08:55,784 INFO [train.py:968] (0/2) Epoch 4, batch 26800, giga_loss[loss=0.2875, simple_loss=0.3484, pruned_loss=0.1133, over 28212.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.4045, pruned_loss=0.1536, over 5651490.97 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4035, pruned_loss=0.153, over 5681629.23 frames. ], giga_tot_loss[loss=0.3546, simple_loss=0.4037, pruned_loss=0.1527, over 5652063.35 frames. ], batch size: 77, lr: 7.63e-03, grad_scale: 8.0
+2023-03-02 07:09:20,193 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-02 07:09:34,120 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:09:39,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:09:40,290 INFO [train.py:968] (0/2) Epoch 4, batch 26850, giga_loss[loss=0.3344, simple_loss=0.4063, pruned_loss=0.1313, over 29000.00 frames. ], tot_loss[loss=0.3539, simple_loss=0.4052, pruned_loss=0.1513, over 5670537.13 frames. ], libri_tot_loss[loss=0.3541, simple_loss=0.403, pruned_loss=0.1526, over 5687665.67 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.405, pruned_loss=0.151, over 5665047.17 frames. ], batch size: 155, lr: 7.63e-03, grad_scale: 8.0
+2023-03-02 07:09:45,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:09:46,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.743e+02 1.588e+03 2.150e+03 2.898e+03 6.637e+03, threshold=4.299e+03, percent-clipped=9.0
+2023-03-02 07:10:03,567 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:10:09,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9460, 2.8167, 1.7027, 1.4479], device='cuda:0'), covar=tensor([0.1431, 0.0603, 0.0734, 0.1000], device='cuda:0'), in_proj_covar=tensor([0.1328, 0.1074, 0.1066, 0.1165], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 07:10:25,304 INFO [train.py:968] (0/2) Epoch 4, batch 26900, giga_loss[loss=0.3255, simple_loss=0.3983, pruned_loss=0.1263, over 28883.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.4052, pruned_loss=0.1497, over 5675671.70 frames. ], libri_tot_loss[loss=0.354, simple_loss=0.4028, pruned_loss=0.1526, over 5693078.05 frames. ], giga_tot_loss[loss=0.352, simple_loss=0.4053, pruned_loss=0.1494, over 5666059.62 frames. ], batch size: 112, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:11:04,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1194, 1.2756, 4.0707, 3.2513], device='cuda:0'), covar=tensor([0.1546, 0.2212, 0.0381, 0.0568], device='cuda:0'), in_proj_covar=tensor([0.0557, 0.0512, 0.0721, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 07:11:08,591 INFO [train.py:968] (0/2) Epoch 4, batch 26950, giga_loss[loss=0.394, simple_loss=0.4447, pruned_loss=0.1717, over 28900.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.4071, pruned_loss=0.1494, over 5679117.35 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4036, pruned_loss=0.1533, over 5697049.62 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4065, pruned_loss=0.1484, over 5667271.51 frames. ], batch size: 227, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:11:15,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.333e+02 1.570e+03 2.415e+03 3.636e+03 7.890e+03, threshold=4.830e+03, percent-clipped=20.0
+2023-03-02 07:11:53,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4594, 1.3347, 1.4221, 1.3415], device='cuda:0'), covar=tensor([0.0964, 0.1443, 0.1567, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0748, 0.0637, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 07:11:54,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:11:54,297 INFO [train.py:968] (0/2) Epoch 4, batch 27000, giga_loss[loss=0.4876, simple_loss=0.4877, pruned_loss=0.2437, over 27578.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.4101, pruned_loss=0.1525, over 5682683.05 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4035, pruned_loss=0.1533, over 5699619.78 frames. ], giga_tot_loss[loss=0.3566, simple_loss=0.4099, pruned_loss=0.1517, over 5670458.70 frames. ], batch size: 472, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:11:54,302 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 07:12:03,199 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2409, simple_loss=0.3417, pruned_loss=0.07006, over 944034.00 frames. 
+2023-03-02 07:12:03,199 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 07:12:05,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:12:31,936 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:12:52,753 INFO [train.py:968] (0/2) Epoch 4, batch 27050, giga_loss[loss=0.373, simple_loss=0.4321, pruned_loss=0.157, over 29118.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.4126, pruned_loss=0.1553, over 5673910.40 frames. ], libri_tot_loss[loss=0.3554, simple_loss=0.4037, pruned_loss=0.1536, over 5692537.98 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4123, pruned_loss=0.1544, over 5671212.37 frames. ], batch size: 155, lr: 7.62e-03, grad_scale: 4.0
+2023-03-02 07:12:59,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.637e+03 2.063e+03 2.742e+03 7.347e+03, threshold=4.126e+03, percent-clipped=3.0
+2023-03-02 07:13:21,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=163220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:13:40,895 INFO [train.py:968] (0/2) Epoch 4, batch 27100, giga_loss[loss=0.3253, simple_loss=0.3873, pruned_loss=0.1317, over 28834.00 frames. ], tot_loss[loss=0.3619, simple_loss=0.4121, pruned_loss=0.1559, over 5682560.72 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4039, pruned_loss=0.1538, over 5697078.52 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.4119, pruned_loss=0.155, over 5675745.50 frames. ], batch size: 186, lr: 7.62e-03, grad_scale: 4.0
+2023-03-02 07:14:17,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:14:33,637 INFO [train.py:968] (0/2) Epoch 4, batch 27150, giga_loss[loss=0.3385, simple_loss=0.3949, pruned_loss=0.1411, over 28542.00 frames. ], tot_loss[loss=0.3616, simple_loss=0.4116, pruned_loss=0.1558, over 5675465.05 frames. ], libri_tot_loss[loss=0.3554, simple_loss=0.4036, pruned_loss=0.1536, over 5701323.56 frames. ], giga_tot_loss[loss=0.3613, simple_loss=0.4119, pruned_loss=0.1553, over 5665673.39 frames. ], batch size: 85, lr: 7.62e-03, grad_scale: 4.0
+2023-03-02 07:14:41,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.568e+03 2.055e+03 2.665e+03 8.968e+03, threshold=4.109e+03, percent-clipped=3.0
+2023-03-02 07:14:52,250 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-02 07:15:15,783 INFO [train.py:968] (0/2) Epoch 4, batch 27200, giga_loss[loss=0.3764, simple_loss=0.4366, pruned_loss=0.1581, over 29037.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4098, pruned_loss=0.1532, over 5674720.06 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4033, pruned_loss=0.1534, over 5698272.44 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.4106, pruned_loss=0.153, over 5669011.57 frames. ], batch size: 145, lr: 7.62e-03, grad_scale: 8.0
+2023-03-02 07:15:34,799 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4689, 3.2929, 1.4885, 1.3891], device='cuda:0'), covar=tensor([0.0825, 0.0342, 0.0823, 0.1268], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0472, 0.0309, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 07:16:02,373 INFO [train.py:968] (0/2) Epoch 4, batch 27250, giga_loss[loss=0.3976, simple_loss=0.4172, pruned_loss=0.1889, over 23445.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4085, pruned_loss=0.1514, over 5665131.99 frames. ], libri_tot_loss[loss=0.354, simple_loss=0.4023, pruned_loss=0.1528, over 5703733.91 frames. ], giga_tot_loss[loss=0.3568, simple_loss=0.4102, pruned_loss=0.1517, over 5654578.78 frames. ], batch size: 705, lr: 7.62e-03, grad_scale: 8.0
+2023-03-02 07:16:07,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.983e+02 1.489e+03 1.960e+03 2.509e+03 7.984e+03, threshold=3.921e+03, percent-clipped=5.0
+2023-03-02 07:16:27,428 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:16:29,740 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:16:36,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4628, 1.8459, 1.7269, 1.6247], device='cuda:0'), covar=tensor([0.1365, 0.1748, 0.1067, 0.1090], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0771, 0.0764, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 07:16:47,936 INFO [train.py:968] (0/2) Epoch 4, batch 27300, giga_loss[loss=0.3892, simple_loss=0.4073, pruned_loss=0.1856, over 23525.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4094, pruned_loss=0.1515, over 5674697.79 frames. ], libri_tot_loss[loss=0.3547, simple_loss=0.4029, pruned_loss=0.1532, over 5708970.18 frames. ], giga_tot_loss[loss=0.3566, simple_loss=0.4105, pruned_loss=0.1514, over 5660731.16 frames. ], batch size: 705, lr: 7.62e-03, grad_scale: 8.0
+2023-03-02 07:16:58,411 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:17:27,471 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=163483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:17:37,733 INFO [train.py:968] (0/2) Epoch 4, batch 27350, giga_loss[loss=0.3163, simple_loss=0.3784, pruned_loss=0.1271, over 28618.00 frames. ], tot_loss[loss=0.3589, simple_loss=0.4108, pruned_loss=0.1535, over 5660293.25 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.403, pruned_loss=0.1533, over 5707337.71 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4116, pruned_loss=0.1533, over 5649958.00 frames. ], batch size: 92, lr: 7.62e-03, grad_scale: 4.0
+2023-03-02 07:17:46,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.379e+02 1.485e+03 2.085e+03 2.988e+03 1.111e+04, threshold=4.169e+03, percent-clipped=15.0
+2023-03-02 07:18:21,170 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-02 07:18:25,462 INFO [train.py:968] (0/2) Epoch 4, batch 27400, giga_loss[loss=0.3475, simple_loss=0.3973, pruned_loss=0.1489, over 28728.00 frames. ], tot_loss[loss=0.3573, simple_loss=0.4091, pruned_loss=0.1528, over 5667472.64 frames. ], libri_tot_loss[loss=0.3544, simple_loss=0.4026, pruned_loss=0.153, over 5710977.58 frames. ], giga_tot_loss[loss=0.358, simple_loss=0.4102, pruned_loss=0.1529, over 5655128.18 frames. ], batch size: 262, lr: 7.62e-03, grad_scale: 4.0
+2023-03-02 07:19:12,165 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8231, 4.5569, 4.4970, 1.8196], device='cuda:0'), covar=tensor([0.0370, 0.0396, 0.0745, 0.2196], device='cuda:0'), in_proj_covar=tensor([0.0838, 0.0729, 0.0840, 0.0590], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 07:19:13,323 INFO [train.py:968] (0/2) Epoch 4, batch 27450, giga_loss[loss=0.3541, simple_loss=0.4059, pruned_loss=0.1511, over 29042.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4072, pruned_loss=0.1523, over 5673217.14 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.4027, pruned_loss=0.1531, over 5705744.16 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4081, pruned_loss=0.1523, over 5666485.09 frames. ], batch size: 106, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:19:15,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:19:20,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.105e+02 1.345e+03 1.726e+03 2.386e+03 6.424e+03, threshold=3.452e+03, percent-clipped=7.0
+2023-03-02 07:19:42,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4315, 2.8497, 1.3541, 1.3646], device='cuda:0'), covar=tensor([0.0879, 0.0364, 0.0880, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0474, 0.0308, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 07:19:43,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0368, 2.2693, 2.3322, 2.1232], device='cuda:0'), covar=tensor([0.1516, 0.1642, 0.0989, 0.1014], device='cuda:0'), in_proj_covar=tensor([0.0787, 0.0770, 0.0765, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 07:19:54,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=163633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:20:04,004 INFO [train.py:968] (0/2) Epoch 4, batch 27500, giga_loss[loss=0.3295, simple_loss=0.3637, pruned_loss=0.1476, over 23783.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4047, pruned_loss=0.1511, over 5672077.91 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.403, pruned_loss=0.1533, over 5710423.76 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.4053, pruned_loss=0.1509, over 5661801.43 frames. ], batch size: 705, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:20:50,547 INFO [train.py:968] (0/2) Epoch 4, batch 27550, libri_loss[loss=0.3508, simple_loss=0.399, pruned_loss=0.1513, over 29672.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4024, pruned_loss=0.1499, over 5672935.78 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4032, pruned_loss=0.1535, over 5712886.56 frames. ], giga_tot_loss[loss=0.3507, simple_loss=0.4026, pruned_loss=0.1494, over 5661641.03 frames. ], batch size: 73, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:20:58,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.879e+03 2.371e+03 3.145e+03 9.967e+03, threshold=4.741e+03, percent-clipped=17.0
+2023-03-02 07:21:30,436 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:21:33,194 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:21:35,670 INFO [train.py:968] (0/2) Epoch 4, batch 27600, giga_loss[loss=0.3565, simple_loss=0.4135, pruned_loss=0.1497, over 28183.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4026, pruned_loss=0.1508, over 5674952.70 frames. ], libri_tot_loss[loss=0.3546, simple_loss=0.4028, pruned_loss=0.1532, over 5717185.74 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.4031, pruned_loss=0.1506, over 5661149.19 frames. ], batch size: 368, lr: 7.61e-03, grad_scale: 8.0
+2023-03-02 07:22:00,043 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:22:10,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9615, 1.6781, 1.2019, 1.4998], device='cuda:0'), covar=tensor([0.0634, 0.0684, 0.1040, 0.0958], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0464, 0.0513, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 07:22:19,566 INFO [train.py:968] (0/2) Epoch 4, batch 27650, giga_loss[loss=0.3256, simple_loss=0.3845, pruned_loss=0.1334, over 28737.00 frames. ], tot_loss[loss=0.3506, simple_loss=0.4016, pruned_loss=0.1498, over 5676653.57 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4032, pruned_loss=0.1535, over 5720843.10 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.4016, pruned_loss=0.1493, over 5661281.50 frames. ], batch size: 262, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:22:27,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.099e+02 1.456e+03 1.824e+03 2.491e+03 1.223e+04, threshold=3.648e+03, percent-clipped=9.0
+2023-03-02 07:23:04,088 INFO [train.py:968] (0/2) Epoch 4, batch 27700, giga_loss[loss=0.3302, simple_loss=0.3921, pruned_loss=0.1342, over 27929.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3974, pruned_loss=0.1448, over 5675291.76 frames. ], libri_tot_loss[loss=0.3544, simple_loss=0.4028, pruned_loss=0.153, over 5722734.03 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3978, pruned_loss=0.1448, over 5660003.76 frames. ], batch size: 412, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:23:16,779 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 07:23:20,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:23:58,195 INFO [train.py:968] (0/2) Epoch 4, batch 27750, giga_loss[loss=0.469, simple_loss=0.4618, pruned_loss=0.2382, over 23834.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3969, pruned_loss=0.1444, over 5664976.20 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.4029, pruned_loss=0.153, over 5723803.36 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.397, pruned_loss=0.1443, over 5651816.23 frames. ], batch size: 705, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:24:08,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.603e+02 1.322e+03 1.769e+03 2.455e+03 5.401e+03, threshold=3.539e+03, percent-clipped=6.0
+2023-03-02 07:24:45,826 INFO [train.py:968] (0/2) Epoch 4, batch 27800, giga_loss[loss=0.3619, simple_loss=0.4023, pruned_loss=0.1608, over 27939.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3964, pruned_loss=0.1446, over 5668292.80 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.403, pruned_loss=0.153, over 5723491.21 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3962, pruned_loss=0.1443, over 5656052.76 frames. ], batch size: 412, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:25:12,498 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-02 07:25:38,598 INFO [train.py:968] (0/2) Epoch 4, batch 27850, giga_loss[loss=0.3281, simple_loss=0.3744, pruned_loss=0.1409, over 28931.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3927, pruned_loss=0.1434, over 5663658.59 frames. ], libri_tot_loss[loss=0.3542, simple_loss=0.4028, pruned_loss=0.1528, over 5728505.74 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3925, pruned_loss=0.143, over 5647375.63 frames. ], batch size: 213, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:25:45,018 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-164000.pt
+2023-03-02 07:25:46,362 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:25:48,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.710e+03 2.335e+03 3.023e+03 5.992e+03, threshold=4.670e+03, percent-clipped=14.0
+2023-03-02 07:25:51,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:25:56,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:26:17,990 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:26:26,523 INFO [train.py:968] (0/2) Epoch 4, batch 27900, giga_loss[loss=0.3575, simple_loss=0.4125, pruned_loss=0.1512, over 28924.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3934, pruned_loss=0.1439, over 5665478.14 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.4029, pruned_loss=0.153, over 5731654.12 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3929, pruned_loss=0.1432, over 5648223.41 frames. ], batch size: 164, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:27:09,898 INFO [train.py:968] (0/2) Epoch 4, batch 27950, libri_loss[loss=0.3535, simple_loss=0.4124, pruned_loss=0.1474, over 29096.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3957, pruned_loss=0.145, over 5655565.22 frames. ], libri_tot_loss[loss=0.3543, simple_loss=0.4027, pruned_loss=0.1529, over 5717405.15 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3952, pruned_loss=0.1442, over 5650771.47 frames. ], batch size: 101, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:27:20,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.580e+02 1.531e+03 2.113e+03 2.715e+03 7.220e+03, threshold=4.227e+03, percent-clipped=5.0
+2023-03-02 07:27:59,167 INFO [train.py:968] (0/2) Epoch 4, batch 28000, giga_loss[loss=0.308, simple_loss=0.3702, pruned_loss=0.1229, over 28750.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3952, pruned_loss=0.1442, over 5652065.31 frames. ], libri_tot_loss[loss=0.3539, simple_loss=0.4024, pruned_loss=0.1527, over 5718388.23 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3949, pruned_loss=0.1436, over 5646514.78 frames. ], batch size: 92, lr: 7.60e-03, grad_scale: 8.0
+2023-03-02 07:28:04,475 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:28:07,923 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:28:34,868 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:28:42,046 INFO [train.py:968] (0/2) Epoch 4, batch 28050, giga_loss[loss=0.3114, simple_loss=0.3744, pruned_loss=0.1241, over 28737.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3945, pruned_loss=0.1438, over 5653086.50 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4023, pruned_loss=0.1524, over 5720748.58 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3942, pruned_loss=0.1433, over 5644753.13 frames. ], batch size: 99, lr: 7.60e-03, grad_scale: 8.0
+2023-03-02 07:28:51,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.088e+02 1.408e+03 1.799e+03 2.360e+03 6.406e+03, threshold=3.598e+03, percent-clipped=3.0
+2023-03-02 07:29:27,871 INFO [train.py:968] (0/2) Epoch 4, batch 28100, giga_loss[loss=0.3297, simple_loss=0.3844, pruned_loss=0.1375, over 28428.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3956, pruned_loss=0.1454, over 5643020.93 frames. ], libri_tot_loss[loss=0.3542, simple_loss=0.4027, pruned_loss=0.1529, over 5711912.63 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3948, pruned_loss=0.1445, over 5643640.81 frames. ], batch size: 60, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:29:48,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1652, 1.3106, 1.0905, 1.3081], device='cuda:0'), covar=tensor([0.0718, 0.0436, 0.0353, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0205, 0.0131, 0.0135, 0.0244], device='cuda:0'), out_proj_covar=tensor([0.0051, 0.0037, 0.0033, 0.0056], device='cuda:0')
+2023-03-02 07:30:12,033 INFO [train.py:968] (0/2) Epoch 4, batch 28150, giga_loss[loss=0.3188, simple_loss=0.3931, pruned_loss=0.1222, over 29031.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3986, pruned_loss=0.1476, over 5650642.56 frames. ], libri_tot_loss[loss=0.3546, simple_loss=0.403, pruned_loss=0.1531, over 5716747.92 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3976, pruned_loss=0.1465, over 5644853.26 frames. ], batch size: 155, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:30:22,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.221e+02 1.499e+03 2.001e+03 2.745e+03 8.098e+03, threshold=4.002e+03, percent-clipped=12.0
+2023-03-02 07:30:58,872 INFO [train.py:968] (0/2) Epoch 4, batch 28200, giga_loss[loss=0.3777, simple_loss=0.4032, pruned_loss=0.1761, over 23506.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.3992, pruned_loss=0.1477, over 5653311.21 frames. ], libri_tot_loss[loss=0.3542, simple_loss=0.4026, pruned_loss=0.1528, over 5718917.00 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3987, pruned_loss=0.147, over 5646249.31 frames. ], batch size: 705, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:31:46,563 INFO [train.py:968] (0/2) Epoch 4, batch 28250, giga_loss[loss=0.4191, simple_loss=0.4296, pruned_loss=0.2043, over 23165.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4024, pruned_loss=0.1509, over 5642262.51 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4032, pruned_loss=0.1532, over 5713801.45 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4014, pruned_loss=0.1499, over 5639452.85 frames. ], batch size: 705, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:31:56,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.548e+02 1.662e+03 2.123e+03 2.764e+03 9.119e+03, threshold=4.245e+03, percent-clipped=12.0
+2023-03-02 07:32:27,925 INFO [train.py:968] (0/2) Epoch 4, batch 28300, giga_loss[loss=0.3633, simple_loss=0.4122, pruned_loss=0.1572, over 29054.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4036, pruned_loss=0.1526, over 5642773.80 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.403, pruned_loss=0.153, over 5708599.18 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.4031, pruned_loss=0.152, over 5641529.36 frames. ], batch size: 128, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:33:04,854 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:33:16,919 INFO [train.py:968] (0/2) Epoch 4, batch 28350, giga_loss[loss=0.3264, simple_loss=0.3971, pruned_loss=0.1279, over 28831.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.4043, pruned_loss=0.1517, over 5644471.24 frames. ], libri_tot_loss[loss=0.3542, simple_loss=0.4027, pruned_loss=0.1528, over 5709687.38 frames. ], giga_tot_loss[loss=0.3533, simple_loss=0.4041, pruned_loss=0.1513, over 5641200.43 frames. ], batch size: 284, lr: 7.59e-03, grad_scale: 4.0
+2023-03-02 07:33:27,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.185e+02 1.698e+03 2.228e+03 3.102e+03 9.575e+03, threshold=4.455e+03, percent-clipped=13.0
+2023-03-02 07:33:35,298 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 07:34:08,155 INFO [train.py:968] (0/2) Epoch 4, batch 28400, giga_loss[loss=0.4101, simple_loss=0.4407, pruned_loss=0.1898, over 26722.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4033, pruned_loss=0.1505, over 5651006.86 frames. ], libri_tot_loss[loss=0.3541, simple_loss=0.4025, pruned_loss=0.1528, over 5713016.82 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4034, pruned_loss=0.1502, over 5644886.22 frames. ], batch size: 555, lr: 7.59e-03, grad_scale: 8.0
+2023-03-02 07:34:14,749 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:34:57,771 INFO [train.py:968] (0/2) Epoch 4, batch 28450, libri_loss[loss=0.3994, simple_loss=0.4429, pruned_loss=0.1779, over 29129.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4042, pruned_loss=0.152, over 5639010.20 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.4029, pruned_loss=0.153, over 5713765.82 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4038, pruned_loss=0.1515, over 5632487.74 frames. ], batch size: 101, lr: 7.59e-03, grad_scale: 4.0
+2023-03-02 07:35:10,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.728e+02 1.659e+03 2.163e+03 2.856e+03 9.474e+03, threshold=4.326e+03, percent-clipped=7.0
+2023-03-02 07:35:54,157 INFO [train.py:968] (0/2) Epoch 4, batch 28500, giga_loss[loss=0.3053, simple_loss=0.3662, pruned_loss=0.1222, over 28787.00 frames. ], tot_loss[loss=0.3539, simple_loss=0.4034, pruned_loss=0.1521, over 5637164.89 frames. ], libri_tot_loss[loss=0.3543, simple_loss=0.4027, pruned_loss=0.1529, over 5713212.72 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4033, pruned_loss=0.1519, over 5631281.19 frames. ], batch size: 186, lr: 7.59e-03, grad_scale: 2.0
+2023-03-02 07:36:17,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2626, 1.9521, 1.4160, 1.7335], device='cuda:0'), covar=tensor([0.0568, 0.0654, 0.0920, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0470, 0.0517, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 07:36:49,775 INFO [train.py:968] (0/2) Epoch 4, batch 28550, giga_loss[loss=0.3053, simple_loss=0.3653, pruned_loss=0.1226, over 28878.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.4014, pruned_loss=0.1516, over 5632833.87 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4022, pruned_loss=0.1525, over 5719175.31 frames. ], giga_tot_loss[loss=0.3526, simple_loss=0.4018, pruned_loss=0.1517, over 5620363.29 frames. ], batch size: 227, lr: 7.59e-03, grad_scale: 2.0
+2023-03-02 07:37:01,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.290e+02 1.653e+03 2.053e+03 2.564e+03 6.109e+03, threshold=4.106e+03, percent-clipped=5.0
+2023-03-02 07:37:32,970 INFO [train.py:968] (0/2) Epoch 4, batch 28600, giga_loss[loss=0.3653, simple_loss=0.4147, pruned_loss=0.1579, over 28613.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4014, pruned_loss=0.1515, over 5650465.09 frames. ], libri_tot_loss[loss=0.3544, simple_loss=0.4028, pruned_loss=0.153, over 5721291.83 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4012, pruned_loss=0.1511, over 5636573.38 frames. ], batch size: 307, lr: 7.59e-03, grad_scale: 2.0
+2023-03-02 07:37:43,555 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:37:44,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9786, 1.0921, 0.8606, 0.5984], device='cuda:0'), covar=tensor([0.0705, 0.0753, 0.0591, 0.0771], device='cuda:0'), in_proj_covar=tensor([0.1353, 0.1107, 0.1090, 0.1201], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 07:38:18,052 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-02 07:38:21,787 INFO [train.py:968] (0/2) Epoch 4, batch 28650, giga_loss[loss=0.3397, simple_loss=0.3933, pruned_loss=0.1431, over 28937.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.4022, pruned_loss=0.1526, over 5651059.81 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4021, pruned_loss=0.1526, over 5715374.41 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.4025, pruned_loss=0.1526, over 5644194.26 frames. ], batch size: 186, lr: 7.59e-03, grad_scale: 2.0
+2023-03-02 07:38:35,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.428e+02 1.718e+03 2.016e+03 2.653e+03 8.074e+03, threshold=4.032e+03, percent-clipped=4.0
+2023-03-02 07:39:08,601 INFO [train.py:968] (0/2) Epoch 4, batch 28700, giga_loss[loss=0.3798, simple_loss=0.4258, pruned_loss=0.1669, over 28956.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.4018, pruned_loss=0.1523, over 5655491.43 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.4021, pruned_loss=0.1525, over 5717701.84 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4021, pruned_loss=0.1523, over 5645372.44 frames. ], batch size: 145, lr: 7.59e-03, grad_scale: 2.0
+2023-03-02 07:39:25,365 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:39:54,953 INFO [train.py:968] (0/2) Epoch 4, batch 28750, giga_loss[loss=0.2963, simple_loss=0.3653, pruned_loss=0.1137, over 28804.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4012, pruned_loss=0.1516, over 5662494.36 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.4019, pruned_loss=0.1522, over 5720750.15 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4016, pruned_loss=0.1519, over 5650014.13 frames. ], batch size: 99, lr: 7.58e-03, grad_scale: 2.0
+2023-03-02 07:40:07,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.677e+03 2.261e+03 3.298e+03 1.211e+04, threshold=4.522e+03, percent-clipped=17.0
+2023-03-02 07:40:23,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:40:44,545 INFO [train.py:968] (0/2) Epoch 4, batch 28800, giga_loss[loss=0.4175, simple_loss=0.4437, pruned_loss=0.1956, over 27606.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.404, pruned_loss=0.1538, over 5662878.06 frames. ], libri_tot_loss[loss=0.3531, simple_loss=0.4018, pruned_loss=0.1522, over 5723666.59 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.4045, pruned_loss=0.1541, over 5649221.68 frames. ], batch size: 472, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:41:03,306 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:41:14,338 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:41:32,413 INFO [train.py:968] (0/2) Epoch 4, batch 28850, giga_loss[loss=0.3891, simple_loss=0.4177, pruned_loss=0.1802, over 26625.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.4044, pruned_loss=0.1542, over 5669237.91 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.402, pruned_loss=0.1524, over 5724655.84 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.4046, pruned_loss=0.1542, over 5656647.42 frames. ], batch size: 555, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:41:39,180 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:41:42,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:41:44,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.562e+02 1.514e+03 2.091e+03 2.894e+03 6.307e+03, threshold=4.182e+03, percent-clipped=8.0
+2023-03-02 07:42:04,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3601, 1.5813, 1.3921, 1.4852], device='cuda:0'), covar=tensor([0.1584, 0.1387, 0.1257, 0.1267], device='cuda:0'), in_proj_covar=tensor([0.1097, 0.0866, 0.0981, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 07:42:09,342 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:42:17,556 INFO [train.py:968] (0/2) Epoch 4, batch 28900, giga_loss[loss=0.3247, simple_loss=0.3888, pruned_loss=0.1303, over 28980.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4046, pruned_loss=0.1546, over 5675135.26 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.4021, pruned_loss=0.1525, over 5726723.45 frames. ], giga_tot_loss[loss=0.3569, simple_loss=0.4047, pruned_loss=0.1546, over 5662563.76 frames. ], batch size: 128, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:42:25,822 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:42:29,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2361, 1.4945, 1.2307, 1.5363], device='cuda:0'), covar=tensor([0.2055, 0.1902, 0.1867, 0.1822], device='cuda:0'), in_proj_covar=tensor([0.1093, 0.0863, 0.0979, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 07:42:40,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:42:42,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:43:05,939 INFO [train.py:968] (0/2) Epoch 4, batch 28950, giga_loss[loss=0.3304, simple_loss=0.3925, pruned_loss=0.1341, over 28719.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4047, pruned_loss=0.1538, over 5675296.48 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4022, pruned_loss=0.1525, over 5718705.70 frames. ], giga_tot_loss[loss=0.3561, simple_loss=0.4047, pruned_loss=0.1538, over 5671802.58 frames. ], batch size: 99, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:43:11,210 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:43:17,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.354e+02 1.602e+03 2.042e+03 2.790e+03 1.014e+04, threshold=4.083e+03, percent-clipped=7.0
+2023-03-02 07:43:40,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:43:50,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:43:55,282 INFO [train.py:968] (0/2) Epoch 4, batch 29000, giga_loss[loss=0.4681, simple_loss=0.4639, pruned_loss=0.2361, over 26620.00 frames. ], tot_loss[loss=0.3582, simple_loss=0.4063, pruned_loss=0.1551, over 5659125.62 frames. ], libri_tot_loss[loss=0.3543, simple_loss=0.4028, pruned_loss=0.1529, over 5712730.05 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4058, pruned_loss=0.1547, over 5661403.43 frames. ], batch size: 555, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:44:21,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3596, 1.4091, 1.2009, 1.6643], device='cuda:0'), covar=tensor([0.1992, 0.1967, 0.1870, 0.2033], device='cuda:0'), in_proj_covar=tensor([0.1095, 0.0863, 0.0979, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 07:44:41,080 INFO [train.py:968] (0/2) Epoch 4, batch 29050, giga_loss[loss=0.3134, simple_loss=0.3704, pruned_loss=0.1283, over 28551.00 frames. ], tot_loss[loss=0.3597, simple_loss=0.4074, pruned_loss=0.1559, over 5670631.19 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4032, pruned_loss=0.1532, over 5717112.40 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4068, pruned_loss=0.1554, over 5667406.99 frames. ], batch size: 78, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:44:54,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.256e+02 1.666e+03 2.190e+03 2.758e+03 1.276e+04, threshold=4.380e+03, percent-clipped=11.0
+2023-03-02 07:45:29,503 INFO [train.py:968] (0/2) Epoch 4, batch 29100, giga_loss[loss=0.4552, simple_loss=0.4532, pruned_loss=0.2286, over 23726.00 frames. ], tot_loss[loss=0.362, simple_loss=0.4092, pruned_loss=0.1574, over 5666489.23 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4035, pruned_loss=0.1534, over 5719035.91 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.4084, pruned_loss=0.1569, over 5661533.01 frames. ], batch size: 705, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:45:54,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:45:58,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:46:07,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0052, 1.2318, 4.0158, 3.2355], device='cuda:0'), covar=tensor([0.1605, 0.2047, 0.0378, 0.0598], device='cuda:0'), in_proj_covar=tensor([0.0559, 0.0509, 0.0731, 0.0591], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 07:46:13,190 INFO [train.py:968] (0/2) Epoch 4, batch 29150, giga_loss[loss=0.3395, simple_loss=0.3933, pruned_loss=0.1428, over 28702.00 frames. ], tot_loss[loss=0.3601, simple_loss=0.4079, pruned_loss=0.1561, over 5670771.59 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4033, pruned_loss=0.1532, over 5722337.49 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4075, pruned_loss=0.156, over 5662639.76 frames. ], batch size: 262, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:46:21,956 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:46:23,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2795, 3.0861, 3.0385, 1.9935], device='cuda:0'), covar=tensor([0.0660, 0.0671, 0.0965, 0.1560], device='cuda:0'), in_proj_covar=tensor([0.0846, 0.0738, 0.0847, 0.0596], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 07:46:23,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.300e+02 1.496e+03 2.139e+03 2.738e+03 6.496e+03, threshold=4.278e+03, percent-clipped=5.0
+2023-03-02 07:46:26,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2445, 1.4423, 1.2082, 1.3236], device='cuda:0'), covar=tensor([0.2197, 0.2051, 0.2064, 0.1959], device='cuda:0'), in_proj_covar=tensor([0.1085, 0.0861, 0.0980, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 07:46:55,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:47:01,881 INFO [train.py:968] (0/2) Epoch 4, batch 29200, giga_loss[loss=0.3452, simple_loss=0.404, pruned_loss=0.1432, over 29009.00 frames. ], tot_loss[loss=0.3586, simple_loss=0.4073, pruned_loss=0.1549, over 5635805.38 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4035, pruned_loss=0.1533, over 5702615.34 frames. ], giga_tot_loss[loss=0.3582, simple_loss=0.4069, pruned_loss=0.1547, over 5644819.32 frames. ], batch size: 213, lr: 7.57e-03, grad_scale: 8.0
+2023-03-02 07:47:05,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:47:09,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-02 07:47:50,432 INFO [train.py:968] (0/2) Epoch 4, batch 29250, giga_loss[loss=0.3893, simple_loss=0.4454, pruned_loss=0.1666, over 29044.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.4082, pruned_loss=0.1553, over 5640105.53 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4032, pruned_loss=0.1532, over 5705992.78 frames. ], giga_tot_loss[loss=0.3594, simple_loss=0.4083, pruned_loss=0.1553, over 5642219.67 frames. ], batch size: 155, lr: 7.57e-03, grad_scale: 4.0
+2023-03-02 07:48:04,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.360e+02 1.588e+03 2.183e+03 3.258e+03 1.076e+04, threshold=4.366e+03, percent-clipped=10.0
+2023-03-02 07:48:18,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:48:33,275 INFO [train.py:968] (0/2) Epoch 4, batch 29300, giga_loss[loss=0.3383, simple_loss=0.3993, pruned_loss=0.1386, over 28919.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4074, pruned_loss=0.1542, over 5651547.79 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4032, pruned_loss=0.1532, over 5710370.37 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.4077, pruned_loss=0.1542, over 5647314.42 frames. ], batch size: 145, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:49:10,000 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:12,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:18,746 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:19,542 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.47 vs. limit=5.0
+2023-03-02 07:49:21,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:21,964 INFO [train.py:968] (0/2) Epoch 4, batch 29350, giga_loss[loss=0.3138, simple_loss=0.3771, pruned_loss=0.1253, over 28930.00 frames. ], tot_loss[loss=0.3545, simple_loss=0.4047, pruned_loss=0.1522, over 5656613.85 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4034, pruned_loss=0.1534, over 5709170.53 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4048, pruned_loss=0.152, over 5653955.23 frames. ], batch size: 136, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:49:34,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.685e+02 1.460e+03 1.880e+03 2.638e+03 5.724e+03, threshold=3.761e+03, percent-clipped=9.0
+2023-03-02 07:49:38,499 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:40,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:42,259 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:46,809 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:50:03,579 INFO [train.py:968] (0/2) Epoch 4, batch 29400, libri_loss[loss=0.275, simple_loss=0.3344, pruned_loss=0.1078, over 28638.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.4033, pruned_loss=0.1511, over 5655938.82 frames. ], libri_tot_loss[loss=0.3543, simple_loss=0.4026, pruned_loss=0.153, over 5705245.62 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4041, pruned_loss=0.1513, over 5654302.71 frames. ], batch size: 63, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:50:30,779 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:50:33,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:50:53,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3104, 1.4116, 1.1851, 1.6396], device='cuda:0'), covar=tensor([0.2104, 0.2026, 0.1901, 0.2118], device='cuda:0'), in_proj_covar=tensor([0.1099, 0.0876, 0.0993, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 07:50:57,172 INFO [train.py:968] (0/2) Epoch 4, batch 29450, giga_loss[loss=0.3761, simple_loss=0.3981, pruned_loss=0.1771, over 23430.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.406, pruned_loss=0.1532, over 5653649.05 frames. ], libri_tot_loss[loss=0.3554, simple_loss=0.4036, pruned_loss=0.1536, over 5707018.73 frames. ], giga_tot_loss[loss=0.3557, simple_loss=0.4058, pruned_loss=0.1528, over 5649476.68 frames. ], batch size: 705, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:51:05,567 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:51:11,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.680e+03 2.093e+03 2.754e+03 5.743e+03, threshold=4.186e+03, percent-clipped=8.0
+2023-03-02 07:51:45,976 INFO [train.py:968] (0/2) Epoch 4, batch 29500, libri_loss[loss=0.3472, simple_loss=0.4112, pruned_loss=0.1416, over 29530.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4048, pruned_loss=0.1532, over 5650710.64 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.4035, pruned_loss=0.1535, over 5706241.63 frames. ], giga_tot_loss[loss=0.3554, simple_loss=0.4048, pruned_loss=0.153, over 5647222.30 frames. ], batch size: 84, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:51:59,802 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:52:03,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:52:27,704 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:52:32,900 INFO [train.py:968] (0/2) Epoch 4, batch 29550, giga_loss[loss=0.3339, simple_loss=0.3954, pruned_loss=0.1362, over 29023.00 frames. ], tot_loss[loss=0.355, simple_loss=0.4042, pruned_loss=0.1529, over 5657499.12 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4038, pruned_loss=0.1538, over 5701291.27 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4039, pruned_loss=0.1524, over 5659084.03 frames. ], batch size: 164, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:52:47,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.217e+02 1.593e+03 2.113e+03 2.751e+03 7.763e+03, threshold=4.225e+03, percent-clipped=6.0
+2023-03-02 07:53:21,339 INFO [train.py:968] (0/2) Epoch 4, batch 29600, giga_loss[loss=0.3645, simple_loss=0.4086, pruned_loss=0.1602, over 28742.00 frames. ], tot_loss[loss=0.3591, simple_loss=0.4069, pruned_loss=0.1557, over 5656115.94 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4042, pruned_loss=0.1542, over 5702616.02 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.4064, pruned_loss=0.155, over 5655487.00 frames. ], batch size: 99, lr: 7.57e-03, grad_scale: 4.0
+2023-03-02 07:54:09,415 INFO [train.py:968] (0/2) Epoch 4, batch 29650, giga_loss[loss=0.3079, simple_loss=0.3775, pruned_loss=0.1192, over 28792.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.4071, pruned_loss=0.1557, over 5649844.42 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4042, pruned_loss=0.1542, over 5702100.46 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4068, pruned_loss=0.1552, over 5649045.73 frames. ], batch size: 242, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:54:27,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.363e+02 1.353e+03 1.742e+03 2.140e+03 5.058e+03, threshold=3.484e+03, percent-clipped=2.0
+2023-03-02 07:54:58,557 INFO [train.py:968] (0/2) Epoch 4, batch 29700, giga_loss[loss=0.3278, simple_loss=0.3859, pruned_loss=0.1349, over 28889.00 frames. ], tot_loss[loss=0.3595, simple_loss=0.4073, pruned_loss=0.1559, over 5646290.78 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.4045, pruned_loss=0.1545, over 5701306.22 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4067, pruned_loss=0.1553, over 5645467.03 frames. ], batch size: 227, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:55:01,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2587, 3.0515, 3.0038, 1.5068], device='cuda:0'), covar=tensor([0.0657, 0.0567, 0.0896, 0.1913], device='cuda:0'), in_proj_covar=tensor([0.0825, 0.0718, 0.0828, 0.0592], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 07:55:30,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8707, 1.6527, 1.1882, 1.5136], device='cuda:0'), covar=tensor([0.0585, 0.0631, 0.0989, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0466, 0.0514, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 07:55:46,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:55:47,799 INFO [train.py:968] (0/2) Epoch 4, batch 29750, giga_loss[loss=0.3215, simple_loss=0.3844, pruned_loss=0.1293, over 28445.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4067, pruned_loss=0.1547, over 5653728.08 frames. ], libri_tot_loss[loss=0.3571, simple_loss=0.4048, pruned_loss=0.1547, over 5705335.04 frames. ], giga_tot_loss[loss=0.3571, simple_loss=0.406, pruned_loss=0.1541, over 5648353.38 frames. ], batch size: 60, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:56:00,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.840e+02 1.690e+03 2.166e+03 2.953e+03 6.232e+03, threshold=4.332e+03, percent-clipped=16.0
+2023-03-02 07:56:27,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0041, 1.1151, 1.2611, 1.0924], device='cuda:0'), covar=tensor([0.1206, 0.1216, 0.1711, 0.1220], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0742, 0.0642, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 07:56:30,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5660, 1.4670, 1.1902, 1.2909], device='cuda:0'), covar=tensor([0.0633, 0.0578, 0.1010, 0.0860], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0468, 0.0519, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 07:56:35,761 INFO [train.py:968] (0/2) Epoch 4, batch 29800, giga_loss[loss=0.319, simple_loss=0.382, pruned_loss=0.128, over 28242.00 frames. ], tot_loss[loss=0.3585, simple_loss=0.4074, pruned_loss=0.1548, over 5654758.26 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4045, pruned_loss=0.1544, over 5706806.93 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.4071, pruned_loss=0.1545, over 5648470.15 frames. ], batch size: 77, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:56:38,341 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:57:24,671 INFO [train.py:968] (0/2) Epoch 4, batch 29850, giga_loss[loss=0.3447, simple_loss=0.3996, pruned_loss=0.1449, over 28901.00 frames. ], tot_loss[loss=0.3582, simple_loss=0.4071, pruned_loss=0.1547, over 5660647.82 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.405, pruned_loss=0.1545, over 5710700.27 frames. ], giga_tot_loss[loss=0.3576, simple_loss=0.4066, pruned_loss=0.1543, over 5650539.91 frames. ], batch size: 174, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:57:33,409 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-166000.pt
+2023-03-02 07:57:41,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.318e+02 1.638e+03 2.152e+03 3.323e+03 1.089e+04, threshold=4.303e+03, percent-clipped=16.0
+2023-03-02 07:58:04,454 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:58:07,453 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:58:12,511 INFO [train.py:968] (0/2) Epoch 4, batch 29900, giga_loss[loss=0.3643, simple_loss=0.4087, pruned_loss=0.16, over 28843.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4057, pruned_loss=0.1537, over 5668347.10 frames. ], libri_tot_loss[loss=0.3564, simple_loss=0.4045, pruned_loss=0.1542, over 5712897.90 frames. ], giga_tot_loss[loss=0.3566, simple_loss=0.4058, pruned_loss=0.1537, over 5657832.99 frames. ], batch size: 186, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:58:30,380 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:58:54,087 INFO [train.py:968] (0/2) Epoch 4, batch 29950, giga_loss[loss=0.4052, simple_loss=0.4127, pruned_loss=0.1988, over 23375.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.4032, pruned_loss=0.1522, over 5669378.45 frames. ], libri_tot_loss[loss=0.3555, simple_loss=0.4039, pruned_loss=0.1536, over 5712313.20 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4038, pruned_loss=0.1527, over 5659445.86 frames. ], batch size: 705, lr: 7.56e-03, grad_scale: 2.0
+2023-03-02 07:59:08,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.845e+02 1.776e+03 2.459e+03 3.246e+03 6.876e+03, threshold=4.917e+03, percent-clipped=8.0
+2023-03-02 07:59:28,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1665, 1.6637, 1.5137, 1.3750], device='cuda:0'), covar=tensor([0.0556, 0.0729, 0.0923, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0333, 0.0460, 0.0507, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 07:59:43,083 INFO [train.py:968] (0/2) Epoch 4, batch 30000, giga_loss[loss=0.3208, simple_loss=0.3695, pruned_loss=0.1361, over 28810.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.3991, pruned_loss=0.1504, over 5657276.49 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.4037, pruned_loss=0.1534, over 5713160.29 frames. ], giga_tot_loss[loss=0.3509, simple_loss=0.3998, pruned_loss=0.1509, over 5647792.58 frames. ], batch size: 99, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:59:43,087 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 07:59:51,386 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.245, simple_loss=0.3484, pruned_loss=0.07078, over 944034.00 frames. 
+2023-03-02 07:59:51,387 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 07:59:51,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:59:53,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8836, 1.7090, 1.7615, 1.7163], device='cuda:0'), covar=tensor([0.1010, 0.1769, 0.1388, 0.1352], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0744, 0.0639, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 08:00:32,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6463, 2.2061, 1.4627, 0.7147], device='cuda:0'), covar=tensor([0.2472, 0.1451, 0.1526, 0.2584], device='cuda:0'), in_proj_covar=tensor([0.1343, 0.1274, 0.1345, 0.1126], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 08:00:34,479 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-02 08:00:34,697 INFO [train.py:968] (0/2) Epoch 4, batch 30050, giga_loss[loss=0.316, simple_loss=0.3765, pruned_loss=0.1278, over 28882.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.3969, pruned_loss=0.1493, over 5650078.01 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4042, pruned_loss=0.1536, over 5698460.74 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3968, pruned_loss=0.1494, over 5653392.65 frames. ], batch size: 199, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 08:00:36,834 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:00:50,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.550e+02 1.554e+03 1.902e+03 2.641e+03 6.361e+03, threshold=3.805e+03, percent-clipped=1.0
+2023-03-02 08:00:53,111 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2140, 1.6810, 1.3414, 1.3933], device='cuda:0'), covar=tensor([0.0767, 0.0297, 0.0322, 0.0837], device='cuda:0'), in_proj_covar=tensor([0.0206, 0.0133, 0.0136, 0.0243], device='cuda:0'), out_proj_covar=tensor([0.0051, 0.0038, 0.0034, 0.0057], device='cuda:0')
+2023-03-02 08:01:24,462 INFO [train.py:968] (0/2) Epoch 4, batch 30100, giga_loss[loss=0.3834, simple_loss=0.4026, pruned_loss=0.1821, over 23602.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3955, pruned_loss=0.149, over 5639696.67 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.4038, pruned_loss=0.1533, over 5701198.63 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.3956, pruned_loss=0.1494, over 5639396.43 frames. ], batch size: 705, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:01:43,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2486, 1.5871, 1.2314, 1.4321], device='cuda:0'), covar=tensor([0.0805, 0.0365, 0.0372, 0.0907], device='cuda:0'), in_proj_covar=tensor([0.0206, 0.0133, 0.0136, 0.0243], device='cuda:0'), out_proj_covar=tensor([0.0051, 0.0038, 0.0034, 0.0057], device='cuda:0')
+2023-03-02 08:02:11,202 INFO [train.py:968] (0/2) Epoch 4, batch 30150, giga_loss[loss=0.3953, simple_loss=0.444, pruned_loss=0.1733, over 27976.00 frames. ], tot_loss[loss=0.3452, simple_loss=0.3952, pruned_loss=0.1476, over 5641226.41 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4043, pruned_loss=0.1536, over 5697953.19 frames. ], giga_tot_loss[loss=0.3448, simple_loss=0.3946, pruned_loss=0.1475, over 5642259.19 frames. ], batch size: 412, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:02:29,421 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.785e+02 1.540e+03 2.136e+03 3.017e+03 1.012e+04, threshold=4.272e+03, percent-clipped=16.0
+2023-03-02 08:02:38,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:02:52,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2657, 1.7290, 1.5834, 1.4852], device='cuda:0'), covar=tensor([0.1527, 0.2040, 0.1265, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0778, 0.0782, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 08:03:01,505 INFO [train.py:968] (0/2) Epoch 4, batch 30200, giga_loss[loss=0.3876, simple_loss=0.4248, pruned_loss=0.1752, over 28591.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3911, pruned_loss=0.1428, over 5642805.65 frames. ], libri_tot_loss[loss=0.3543, simple_loss=0.4029, pruned_loss=0.1528, over 5703412.76 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3916, pruned_loss=0.1431, over 5637015.79 frames. ], batch size: 307, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:03:38,181 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-02 08:03:51,521 INFO [train.py:968] (0/2) Epoch 4, batch 30250, giga_loss[loss=0.2929, simple_loss=0.3648, pruned_loss=0.1105, over 28058.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3887, pruned_loss=0.1398, over 5645358.87 frames. ], libri_tot_loss[loss=0.3539, simple_loss=0.4024, pruned_loss=0.1527, over 5701000.60 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3892, pruned_loss=0.1398, over 5640833.12 frames. ], batch size: 412, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:04:04,998 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.688e+02 1.464e+03 1.961e+03 2.933e+03 9.513e+03, threshold=3.923e+03, percent-clipped=10.0
+2023-03-02 08:04:36,215 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-02 08:04:44,321 INFO [train.py:968] (0/2) Epoch 4, batch 30300, giga_loss[loss=0.305, simple_loss=0.3709, pruned_loss=0.1195, over 28562.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3848, pruned_loss=0.1354, over 5645268.70 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.4018, pruned_loss=0.1523, over 5702604.37 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3855, pruned_loss=0.1356, over 5639729.61 frames. ], batch size: 336, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:04:47,241 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166447.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 08:05:00,412 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:05:02,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:05:31,825 INFO [train.py:968] (0/2) Epoch 4, batch 30350, giga_loss[loss=0.292, simple_loss=0.3752, pruned_loss=0.1043, over 28893.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3799, pruned_loss=0.1307, over 5652291.38 frames. ], libri_tot_loss[loss=0.3529, simple_loss=0.4012, pruned_loss=0.1523, over 5707262.55 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3805, pruned_loss=0.1304, over 5642583.88 frames. ], batch size: 186, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:05:32,874 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:05:47,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.753e+02 1.463e+03 1.887e+03 2.534e+03 7.459e+03, threshold=3.774e+03, percent-clipped=5.0
+2023-03-02 08:05:55,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:05:58,679 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5217, 1.6647, 1.5330, 1.6034], device='cuda:0'), covar=tensor([0.1164, 0.1632, 0.1478, 0.1367], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0725, 0.0624, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 08:06:07,406 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 08:06:18,084 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:06:21,315 INFO [train.py:968] (0/2) Epoch 4, batch 30400, giga_loss[loss=0.3075, simple_loss=0.3786, pruned_loss=0.1183, over 28709.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3774, pruned_loss=0.1261, over 5670386.04 frames. ], libri_tot_loss[loss=0.352, simple_loss=0.4004, pruned_loss=0.1518, over 5709213.27 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3783, pruned_loss=0.1259, over 5660078.38 frames. ], batch size: 242, lr: 7.55e-03, grad_scale: 8.0
+2023-03-02 08:06:50,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3437, 1.7584, 1.0978, 1.0835], device='cuda:0'), covar=tensor([0.0965, 0.0574, 0.0728, 0.0737], device='cuda:0'), in_proj_covar=tensor([0.1301, 0.1090, 0.1059, 0.1149], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 08:06:50,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:07:02,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3747, 1.8444, 1.6825, 1.6492], device='cuda:0'), covar=tensor([0.1371, 0.1814, 0.1170, 0.1096], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0762, 0.0767, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 08:07:13,088 INFO [train.py:968] (0/2) Epoch 4, batch 30450, giga_loss[loss=0.3189, simple_loss=0.3946, pruned_loss=0.1216, over 28970.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3777, pruned_loss=0.1256, over 5664677.22 frames. ], libri_tot_loss[loss=0.3516, simple_loss=0.3999, pruned_loss=0.1516, over 5704815.79 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3782, pruned_loss=0.1249, over 5660576.09 frames. ], batch size: 164, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:07:30,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.937e+02 1.294e+03 1.766e+03 2.363e+03 5.747e+03, threshold=3.531e+03, percent-clipped=4.0
+2023-03-02 08:08:02,223 INFO [train.py:968] (0/2) Epoch 4, batch 30500, giga_loss[loss=0.3125, simple_loss=0.3752, pruned_loss=0.1249, over 28928.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3764, pruned_loss=0.1247, over 5668986.89 frames. ], libri_tot_loss[loss=0.3503, simple_loss=0.3987, pruned_loss=0.151, over 5709301.25 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3774, pruned_loss=0.1242, over 5660919.09 frames. ], batch size: 227, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:08:21,058 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:08:21,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3864, 5.1912, 5.0224, 2.0970], device='cuda:0'), covar=tensor([0.0299, 0.0306, 0.0641, 0.1881], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0706, 0.0804, 0.0581], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 08:08:23,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:08:53,824 INFO [train.py:968] (0/2) Epoch 4, batch 30550, libri_loss[loss=0.3098, simple_loss=0.3766, pruned_loss=0.1215, over 29517.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3718, pruned_loss=0.121, over 5669777.22 frames. ], libri_tot_loss[loss=0.3502, simple_loss=0.3985, pruned_loss=0.151, over 5711131.29 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3723, pruned_loss=0.1201, over 5660697.23 frames. ], batch size: 84, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:08:54,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:08:59,835 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:09:10,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.824e+02 1.398e+03 1.831e+03 2.264e+03 4.990e+03, threshold=3.661e+03, percent-clipped=5.0
+2023-03-02 08:09:13,160 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:09:14,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:09:41,185 INFO [train.py:968] (0/2) Epoch 4, batch 30600, giga_loss[loss=0.2778, simple_loss=0.3571, pruned_loss=0.09925, over 28880.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.369, pruned_loss=0.1198, over 5665394.40 frames. ], libri_tot_loss[loss=0.3483, simple_loss=0.3967, pruned_loss=0.1499, over 5716213.41 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3702, pruned_loss=0.119, over 5652070.77 frames. ], batch size: 145, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:09:43,910 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:10:26,453 INFO [train.py:968] (0/2) Epoch 4, batch 30650, giga_loss[loss=0.3777, simple_loss=0.4205, pruned_loss=0.1675, over 27592.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3705, pruned_loss=0.1206, over 5675229.39 frames. ], libri_tot_loss[loss=0.3483, simple_loss=0.3968, pruned_loss=0.15, over 5721548.58 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3707, pruned_loss=0.1191, over 5658522.62 frames. ], batch size: 472, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:10:42,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.565e+02 1.360e+03 1.756e+03 2.553e+03 5.787e+03, threshold=3.512e+03, percent-clipped=7.0
+2023-03-02 08:10:54,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166822.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 08:11:12,284 INFO [train.py:968] (0/2) Epoch 4, batch 30700, giga_loss[loss=0.2962, simple_loss=0.3698, pruned_loss=0.1113, over 28832.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3691, pruned_loss=0.1198, over 5677145.85 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3953, pruned_loss=0.1492, over 5726436.96 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3696, pruned_loss=0.1182, over 5657502.41 frames. ], batch size: 106, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:11:20,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4970, 1.9487, 1.8931, 1.7909], device='cuda:0'), covar=tensor([0.1583, 0.1789, 0.1146, 0.1126], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0756, 0.0760, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 08:11:43,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:12:05,104 INFO [train.py:968] (0/2) Epoch 4, batch 30750, giga_loss[loss=0.2929, simple_loss=0.3608, pruned_loss=0.1125, over 28716.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3654, pruned_loss=0.1165, over 5664829.46 frames. ], libri_tot_loss[loss=0.3472, simple_loss=0.3954, pruned_loss=0.1495, over 5719194.90 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3654, pruned_loss=0.1146, over 5655294.57 frames. ], batch size: 262, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:12:14,896 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:12:20,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.303e+02 1.434e+03 1.980e+03 2.892e+03 8.377e+03, threshold=3.960e+03, percent-clipped=15.0
+2023-03-02 08:12:24,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:12:26,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6087, 2.1363, 1.3774, 0.9078], device='cuda:0'), covar=tensor([0.3085, 0.1821, 0.1842, 0.2539], device='cuda:0'), in_proj_covar=tensor([0.1335, 0.1266, 0.1332, 0.1107], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 08:12:33,870 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:12:33,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6564, 1.5927, 1.1505, 1.3530], device='cuda:0'), covar=tensor([0.0644, 0.0509, 0.0964, 0.0869], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0457, 0.0509, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 08:12:51,219 INFO [train.py:968] (0/2) Epoch 4, batch 30800, giga_loss[loss=0.256, simple_loss=0.3317, pruned_loss=0.09016, over 28517.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3623, pruned_loss=0.1144, over 5670284.76 frames. ], libri_tot_loss[loss=0.3465, simple_loss=0.3947, pruned_loss=0.1491, over 5716885.44 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3616, pruned_loss=0.1118, over 5662374.34 frames. ], batch size: 71, lr: 7.54e-03, grad_scale: 8.0
+2023-03-02 08:13:16,115 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166965.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 08:13:19,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166968.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 08:13:43,730 INFO [train.py:968] (0/2) Epoch 4, batch 30850, libri_loss[loss=0.3033, simple_loss=0.3532, pruned_loss=0.1267, over 29544.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3596, pruned_loss=0.1135, over 5666837.65 frames. ], libri_tot_loss[loss=0.3458, simple_loss=0.3941, pruned_loss=0.1488, over 5716826.54 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3589, pruned_loss=0.111, over 5659384.39 frames. ], batch size: 78, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:13:46,669 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166997.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 08:14:01,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.642e+02 1.337e+03 1.732e+03 2.632e+03 7.631e+03, threshold=3.463e+03, percent-clipped=10.0
+2023-03-02 08:14:06,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7952, 2.6206, 2.2045, 2.0057], device='cuda:0'), covar=tensor([0.1716, 0.1561, 0.1223, 0.1209], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0747, 0.0762, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 08:14:31,866 INFO [train.py:968] (0/2) Epoch 4, batch 30900, giga_loss[loss=0.3047, simple_loss=0.3546, pruned_loss=0.1274, over 26519.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3576, pruned_loss=0.1127, over 5664513.09 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3934, pruned_loss=0.1483, over 5720422.34 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3569, pruned_loss=0.1103, over 5654251.13 frames. ], batch size: 555, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:14:47,188 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:14:49,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:15:05,155 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3635, 1.9348, 1.4394, 0.7056], device='cuda:0'), covar=tensor([0.2355, 0.1296, 0.1679, 0.2314], device='cuda:0'), in_proj_covar=tensor([0.1342, 0.1273, 0.1347, 0.1117], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 08:15:07,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:15:12,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2271, 4.0184, 3.8925, 1.7426], device='cuda:0'), covar=tensor([0.0515, 0.0512, 0.0956, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0700, 0.0788, 0.0575], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 08:15:18,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4418, 1.5891, 1.0084, 1.4147], device='cuda:0'), covar=tensor([0.0869, 0.0612, 0.1508, 0.1052], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0452, 0.0501, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-02 08:15:22,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:15:26,969 INFO [train.py:968] (0/2) Epoch 4, batch 30950, giga_loss[loss=0.2878, simple_loss=0.3595, pruned_loss=0.1081, over 28972.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3583, pruned_loss=0.1129, over 5660243.06 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.393, pruned_loss=0.1482, over 5724447.05 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3575, pruned_loss=0.1106, over 5647695.15 frames. ], batch size: 136, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:15:43,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.503e+02 1.225e+03 1.537e+03 2.229e+03 3.943e+03, threshold=3.074e+03, percent-clipped=6.0
+2023-03-02 08:16:06,114 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:16:21,811 INFO [train.py:968] (0/2) Epoch 4, batch 31000, giga_loss[loss=0.2776, simple_loss=0.361, pruned_loss=0.09709, over 28553.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1153, over 5657104.60 frames. ], libri_tot_loss[loss=0.3445, simple_loss=0.3927, pruned_loss=0.1481, over 5730524.32 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.361, pruned_loss=0.1123, over 5639278.32 frames. ], batch size: 336, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:16:34,795 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:17:20,618 INFO [train.py:968] (0/2) Epoch 4, batch 31050, libri_loss[loss=0.3803, simple_loss=0.4159, pruned_loss=0.1723, over 29173.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3625, pruned_loss=0.1144, over 5643433.23 frames. ], libri_tot_loss[loss=0.3444, simple_loss=0.3925, pruned_loss=0.1481, over 5725214.28 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3608, pruned_loss=0.1113, over 5631949.64 frames. ], batch size: 97, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:17:21,098 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-02 08:17:22,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6106, 3.4105, 1.6707, 1.4997], device='cuda:0'), covar=tensor([0.0789, 0.0274, 0.0818, 0.1207], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0469, 0.0312, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 08:17:29,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3578, 1.4608, 1.2873, 1.5161], device='cuda:0'), covar=tensor([0.2230, 0.2050, 0.1990, 0.2021], device='cuda:0'), in_proj_covar=tensor([0.1087, 0.0847, 0.0983, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 08:17:40,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.902e+02 1.364e+03 1.842e+03 2.771e+03 1.187e+04, threshold=3.683e+03, percent-clipped=19.0
+2023-03-02 08:17:49,862 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:17:57,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:18:17,977 INFO [train.py:968] (0/2) Epoch 4, batch 31100, giga_loss[loss=0.2626, simple_loss=0.3392, pruned_loss=0.09297, over 28669.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3629, pruned_loss=0.115, over 5635864.96 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3919, pruned_loss=0.1478, over 5717014.56 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.361, pruned_loss=0.1114, over 5630903.23 frames. ], batch size: 262, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:18:25,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:18:26,878 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:19:01,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:19:18,569 INFO [train.py:968] (0/2) Epoch 4, batch 31150, giga_loss[loss=0.2497, simple_loss=0.3221, pruned_loss=0.08865, over 24233.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3602, pruned_loss=0.1128, over 5643435.00 frames. ], libri_tot_loss[loss=0.343, simple_loss=0.3914, pruned_loss=0.1473, over 5720121.09 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3584, pruned_loss=0.1096, over 5634900.90 frames. ], batch size: 705, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:19:25,388 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:19:39,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.740e+02 1.332e+03 1.757e+03 2.508e+03 8.801e+03, threshold=3.513e+03, percent-clipped=3.0
+2023-03-02 08:19:54,076 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:20:07,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8313, 3.6853, 3.5355, 1.8239], device='cuda:0'), covar=tensor([0.0423, 0.0445, 0.0714, 0.2141], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0686, 0.0779, 0.0565], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 08:20:21,261 INFO [train.py:968] (0/2) Epoch 4, batch 31200, giga_loss[loss=0.2582, simple_loss=0.345, pruned_loss=0.08571, over 28469.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3585, pruned_loss=0.1103, over 5642229.96 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3913, pruned_loss=0.1472, over 5723027.74 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3566, pruned_loss=0.1071, over 5631423.77 frames. ], batch size: 336, lr: 7.53e-03, grad_scale: 8.0
+2023-03-02 08:21:18,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:21:19,024 INFO [train.py:968] (0/2) Epoch 4, batch 31250, giga_loss[loss=0.2568, simple_loss=0.3304, pruned_loss=0.09162, over 28889.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.357, pruned_loss=0.1092, over 5643679.63 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3909, pruned_loss=0.1472, over 5718763.98 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3547, pruned_loss=0.1056, over 5636801.57 frames. ], batch size: 120, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:21:22,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:21:37,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 08:21:42,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.177e+02 1.419e+03 1.868e+03 2.542e+03 1.273e+04, threshold=3.736e+03, percent-clipped=7.0
+2023-03-02 08:21:52,344 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:21:55,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:21:56,340 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:22:24,718 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:22:25,029 INFO [train.py:968] (0/2) Epoch 4, batch 31300, giga_loss[loss=0.3241, simple_loss=0.3791, pruned_loss=0.1345, over 27002.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3535, pruned_loss=0.1078, over 5656064.91 frames. ], libri_tot_loss[loss=0.3422, simple_loss=0.3904, pruned_loss=0.1469, over 5718908.39 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3518, pruned_loss=0.1048, over 5650001.69 frames. ], batch size: 555, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:22:26,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:22:33,253 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:22:59,002 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:23:20,758 INFO [train.py:968] (0/2) Epoch 4, batch 31350, giga_loss[loss=0.2903, simple_loss=0.3594, pruned_loss=0.1106, over 28328.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3535, pruned_loss=0.1083, over 5670483.32 frames. ], libri_tot_loss[loss=0.3414, simple_loss=0.3898, pruned_loss=0.1465, over 5723974.03 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3515, pruned_loss=0.105, over 5659429.17 frames. ], batch size: 368, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:23:21,009 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:23:24,680 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:23:32,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:23:40,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.778e+02 1.279e+03 1.738e+03 2.323e+03 6.484e+03, threshold=3.477e+03, percent-clipped=5.0
+2023-03-02 08:24:06,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1702, 2.4182, 1.1412, 1.2405], device='cuda:0'), covar=tensor([0.0914, 0.0337, 0.0967, 0.1397], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0473, 0.0319, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0023, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 08:24:06,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:24:19,583 INFO [train.py:968] (0/2) Epoch 4, batch 31400, giga_loss[loss=0.2179, simple_loss=0.2864, pruned_loss=0.07467, over 24568.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3545, pruned_loss=0.1084, over 5666870.35 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3897, pruned_loss=0.1465, over 5726818.40 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3525, pruned_loss=0.1053, over 5655034.74 frames. ], batch size: 705, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:25:23,782 INFO [train.py:968] (0/2) Epoch 4, batch 31450, giga_loss[loss=0.2554, simple_loss=0.343, pruned_loss=0.08387, over 29067.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3567, pruned_loss=0.1091, over 5647914.41 frames. ], libri_tot_loss[loss=0.3417, simple_loss=0.39, pruned_loss=0.1467, over 5716008.92 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3546, pruned_loss=0.1061, over 5648497.76 frames. ], batch size: 128, lr: 7.52e-03, grad_scale: 2.0
+2023-03-02 08:25:50,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.032e+02 1.284e+03 1.646e+03 2.367e+03 7.933e+03, threshold=3.291e+03, percent-clipped=10.0
+2023-03-02 08:26:28,566 INFO [train.py:968] (0/2) Epoch 4, batch 31500, libri_loss[loss=0.2814, simple_loss=0.3357, pruned_loss=0.1135, over 29659.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3518, pruned_loss=0.1054, over 5665424.81 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.3892, pruned_loss=0.1462, over 5719087.49 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3504, pruned_loss=0.103, over 5662225.88 frames. ], batch size: 73, lr: 7.52e-03, grad_scale: 2.0
+2023-03-02 08:26:34,376 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:26:38,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:27:11,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:27:17,197 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:27:17,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:27:37,515 INFO [train.py:968] (0/2) Epoch 4, batch 31550, libri_loss[loss=0.3087, simple_loss=0.3582, pruned_loss=0.1296, over 29591.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3546, pruned_loss=0.1079, over 5673611.03 frames. ], libri_tot_loss[loss=0.3409, simple_loss=0.3891, pruned_loss=0.1463, over 5720600.36 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3527, pruned_loss=0.105, over 5668399.76 frames. ], batch size: 75, lr: 7.52e-03, grad_scale: 2.0
+2023-03-02 08:27:43,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:27:55,885 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:28:01,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.199e+02 1.371e+03 1.784e+03 2.702e+03 4.816e+03, threshold=3.568e+03, percent-clipped=9.0
+2023-03-02 08:28:34,657 INFO [train.py:968] (0/2) Epoch 4, batch 31600, giga_loss[loss=0.2726, simple_loss=0.3647, pruned_loss=0.09028, over 28976.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3569, pruned_loss=0.1083, over 5672098.26 frames. ], libri_tot_loss[loss=0.3403, simple_loss=0.3887, pruned_loss=0.146, over 5721762.51 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3547, pruned_loss=0.105, over 5665042.03 frames. ], batch size: 199, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:29:41,445 INFO [train.py:968] (0/2) Epoch 4, batch 31650, giga_loss[loss=0.2588, simple_loss=0.3521, pruned_loss=0.08276, over 28876.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3591, pruned_loss=0.1071, over 5656304.66 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3888, pruned_loss=0.1463, over 5712790.51 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3566, pruned_loss=0.1034, over 5656795.26 frames. ], batch size: 174, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:30:07,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.292e+02 1.321e+03 1.986e+03 2.637e+03 5.430e+03, threshold=3.972e+03, percent-clipped=8.0
+2023-03-02 08:30:19,514 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-02 08:30:22,583 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:30:40,782 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:30:44,917 INFO [train.py:968] (0/2) Epoch 4, batch 31700, giga_loss[loss=0.2742, simple_loss=0.3515, pruned_loss=0.09842, over 27520.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3589, pruned_loss=0.1054, over 5654255.02 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3881, pruned_loss=0.1458, over 5711836.20 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3572, pruned_loss=0.1024, over 5654678.29 frames. ], batch size: 472, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:30:45,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:31:14,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:31:18,213 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:31:18,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:31:37,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9582, 1.8029, 1.1993, 1.5895], device='cuda:0'), covar=tensor([0.0588, 0.0616, 0.1023, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0459, 0.0512, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 08:31:46,178 INFO [train.py:968] (0/2) Epoch 4, batch 31750, giga_loss[loss=0.2459, simple_loss=0.3324, pruned_loss=0.07965, over 28911.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3573, pruned_loss=0.1029, over 5668461.70 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3875, pruned_loss=0.1454, over 5714898.37 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3561, pruned_loss=0.1004, over 5665557.46 frames. ], batch size: 227, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:32:10,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.373e+02 1.320e+03 1.706e+03 2.270e+03 8.027e+03, threshold=3.412e+03, percent-clipped=6.0
+2023-03-02 08:32:45,714 INFO [train.py:968] (0/2) Epoch 4, batch 31800, giga_loss[loss=0.2658, simple_loss=0.3438, pruned_loss=0.09393, over 28875.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3596, pruned_loss=0.1057, over 5677478.29 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3874, pruned_loss=0.1455, over 5717713.71 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3581, pruned_loss=0.1027, over 5671646.77 frames. ], batch size: 174, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:33:55,690 INFO [train.py:968] (0/2) Epoch 4, batch 31850, giga_loss[loss=0.2932, simple_loss=0.3632, pruned_loss=0.1115, over 28893.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3583, pruned_loss=0.1066, over 5680015.68 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.3865, pruned_loss=0.145, over 5720303.87 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3573, pruned_loss=0.1039, over 5672263.52 frames. ], batch size: 186, lr: 7.51e-03, grad_scale: 4.0
+2023-03-02 08:34:07,335 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-168000.pt
+2023-03-02 08:34:28,558 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:34:29,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.060e+02 1.464e+03 1.963e+03 2.694e+03 5.445e+03, threshold=3.927e+03, percent-clipped=10.0
+2023-03-02 08:34:30,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4824, 1.7067, 1.1778, 0.9688], device='cuda:0'), covar=tensor([0.1063, 0.0665, 0.0599, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.1311, 0.1048, 0.1036, 0.1131], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 08:34:31,446 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:34:34,014 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:34:36,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:35:15,588 INFO [train.py:968] (0/2) Epoch 4, batch 31900, giga_loss[loss=0.3163, simple_loss=0.3819, pruned_loss=0.1253, over 28679.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.359, pruned_loss=0.1079, over 5681311.75 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3859, pruned_loss=0.1446, over 5723212.80 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3585, pruned_loss=0.1056, over 5671802.18 frames. ], batch size: 307, lr: 7.51e-03, grad_scale: 4.0
+2023-03-02 08:35:16,234 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:35:21,916 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:36:29,305 INFO [train.py:968] (0/2) Epoch 4, batch 31950, giga_loss[loss=0.2628, simple_loss=0.3423, pruned_loss=0.09163, over 28901.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3543, pruned_loss=0.105, over 5678605.38 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.386, pruned_loss=0.1447, over 5723120.00 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3536, pruned_loss=0.1028, over 5670899.93 frames. ], batch size: 213, lr: 7.51e-03, grad_scale: 4.0
+2023-03-02 08:36:52,212 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:36:52,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.036e+02 1.318e+03 1.830e+03 2.693e+03 1.106e+04, threshold=3.660e+03, percent-clipped=16.0
+2023-03-02 08:37:33,520 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-02 08:37:33,654 INFO [train.py:968] (0/2) Epoch 4, batch 32000, giga_loss[loss=0.2629, simple_loss=0.3383, pruned_loss=0.09375, over 28089.00 frames. ], tot_loss[loss=0.281, simple_loss=0.353, pruned_loss=0.1045, over 5675697.67 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3857, pruned_loss=0.1445, over 5723113.64 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3519, pruned_loss=0.1019, over 5668592.92 frames. ], batch size: 412, lr: 7.51e-03, grad_scale: 8.0
+2023-03-02 08:38:00,777 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:38:43,781 INFO [train.py:968] (0/2) Epoch 4, batch 32050, giga_loss[loss=0.257, simple_loss=0.3246, pruned_loss=0.09469, over 28584.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3515, pruned_loss=0.1043, over 5683105.15 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3854, pruned_loss=0.1445, over 5724779.64 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3505, pruned_loss=0.1018, over 5675524.63 frames. ], batch size: 78, lr: 7.51e-03, grad_scale: 8.0
+2023-03-02 08:38:55,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:39:08,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.054e+02 1.364e+03 1.788e+03 2.338e+03 7.602e+03, threshold=3.576e+03, percent-clipped=10.0
+2023-03-02 08:39:44,820 INFO [train.py:968] (0/2) Epoch 4, batch 32100, giga_loss[loss=0.3029, simple_loss=0.3721, pruned_loss=0.1168, over 28141.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3566, pruned_loss=0.107, over 5687107.16 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3843, pruned_loss=0.1437, over 5729688.89 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3561, pruned_loss=0.1048, over 5675781.62 frames. ], batch size: 412, lr: 7.51e-03, grad_scale: 2.0
+2023-03-02 08:40:42,204 INFO [train.py:968] (0/2) Epoch 4, batch 32150, giga_loss[loss=0.279, simple_loss=0.3441, pruned_loss=0.107, over 28650.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3575, pruned_loss=0.1089, over 5691673.86 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3846, pruned_loss=0.144, over 5725162.44 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.356, pruned_loss=0.1057, over 5685686.75 frames. ], batch size: 242, lr: 7.51e-03, grad_scale: 2.0
+2023-03-02 08:41:12,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.264e+02 1.428e+03 1.759e+03 2.671e+03 7.487e+03, threshold=3.517e+03, percent-clipped=8.0
+2023-03-02 08:41:27,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8147, 0.9604, 3.7077, 3.0141], device='cuda:0'), covar=tensor([0.1734, 0.2299, 0.0427, 0.0617], device='cuda:0'), in_proj_covar=tensor([0.0556, 0.0514, 0.0713, 0.0564], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 08:41:45,921 INFO [train.py:968] (0/2) Epoch 4, batch 32200, giga_loss[loss=0.2495, simple_loss=0.3307, pruned_loss=0.0842, over 28935.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3568, pruned_loss=0.1093, over 5671303.43 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3845, pruned_loss=0.144, over 5708490.91 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3553, pruned_loss=0.1063, over 5680746.34 frames. ], batch size: 145, lr: 7.51e-03, grad_scale: 2.0
+2023-03-02 08:41:46,322 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:41:51,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:42:08,872 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-02 08:42:26,118 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:42:48,112 INFO [train.py:968] (0/2) Epoch 4, batch 32250, giga_loss[loss=0.2863, simple_loss=0.3542, pruned_loss=0.1092, over 28645.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3565, pruned_loss=0.1094, over 5674334.13 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.384, pruned_loss=0.1436, over 5711537.84 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3553, pruned_loss=0.1067, over 5678569.77 frames. ], batch size: 242, lr: 7.51e-03, grad_scale: 2.0
+2023-03-02 08:43:20,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.455e+02 1.425e+03 1.777e+03 2.275e+03 5.095e+03, threshold=3.553e+03, percent-clipped=6.0
+2023-03-02 08:43:38,851 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:43:57,197 INFO [train.py:968] (0/2) Epoch 4, batch 32300, giga_loss[loss=0.2783, simple_loss=0.3557, pruned_loss=0.1004, over 28886.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3579, pruned_loss=0.1091, over 5676295.97 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3837, pruned_loss=0.1435, over 5715523.29 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3568, pruned_loss=0.1065, over 5675373.28 frames. ], batch size: 99, lr: 7.50e-03, grad_scale: 2.0
+2023-03-02 08:44:59,840 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:45:07,579 INFO [train.py:968] (0/2) Epoch 4, batch 32350, giga_loss[loss=0.2746, simple_loss=0.3546, pruned_loss=0.0973, over 28606.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3588, pruned_loss=0.1093, over 5659688.35 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3835, pruned_loss=0.1434, over 5707497.97 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3572, pruned_loss=0.1062, over 5665572.66 frames. ], batch size: 242, lr: 7.50e-03, grad_scale: 2.0
+2023-03-02 08:45:43,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.647e+02 1.460e+03 1.836e+03 2.484e+03 4.436e+03, threshold=3.671e+03, percent-clipped=5.0
+2023-03-02 08:45:59,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3470, 1.8682, 1.6324, 1.5234], device='cuda:0'), covar=tensor([0.1254, 0.1621, 0.1039, 0.1059], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0747, 0.0764, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 08:46:11,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:46:18,386 INFO [train.py:968] (0/2) Epoch 4, batch 32400, giga_loss[loss=0.2694, simple_loss=0.3359, pruned_loss=0.1014, over 29120.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3572, pruned_loss=0.1087, over 5660496.49 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3837, pruned_loss=0.1437, over 5703387.84 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3552, pruned_loss=0.1052, over 5667834.76 frames. ], batch size: 113, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:46:30,228 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-02 08:47:19,810 INFO [train.py:968] (0/2) Epoch 4, batch 32450, giga_loss[loss=0.2168, simple_loss=0.2986, pruned_loss=0.06746, over 28840.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3527, pruned_loss=0.1077, over 5671580.56 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3834, pruned_loss=0.1437, over 5708978.30 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3505, pruned_loss=0.1038, over 5671398.04 frames. ], batch size: 174, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:47:45,719 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1303, 1.8924, 1.6185, 1.7081], device='cuda:0'), covar=tensor([0.0529, 0.0513, 0.0796, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0456, 0.0515, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 08:47:48,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.087e+02 1.343e+03 1.861e+03 2.569e+03 8.393e+03, threshold=3.721e+03, percent-clipped=9.0
+2023-03-02 08:48:03,598 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:48:07,616 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:48:09,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:48:15,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9743, 3.7525, 3.6663, 1.6105], device='cuda:0'), covar=tensor([0.0525, 0.0444, 0.0804, 0.2096], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0689, 0.0775, 0.0570], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 08:48:24,816 INFO [train.py:968] (0/2) Epoch 4, batch 32500, giga_loss[loss=0.2825, simple_loss=0.3436, pruned_loss=0.1107, over 28725.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3479, pruned_loss=0.1055, over 5666479.93 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3831, pruned_loss=0.1434, over 5708799.49 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3454, pruned_loss=0.1016, over 5665431.21 frames. ], batch size: 307, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:48:49,319 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:49:04,814 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:49:07,916 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:49:20,504 INFO [train.py:968] (0/2) Epoch 4, batch 32550, giga_loss[loss=0.2733, simple_loss=0.3453, pruned_loss=0.1006, over 28980.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3483, pruned_loss=0.1064, over 5664255.75 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3818, pruned_loss=0.1427, over 5705953.26 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3463, pruned_loss=0.1024, over 5665293.52 frames. ], batch size: 128, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:49:38,775 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:49:44,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.071e+02 1.695e+03 2.363e+03 3.559e+03 7.843e+03, threshold=4.727e+03, percent-clipped=20.0
+2023-03-02 08:50:19,272 INFO [train.py:968] (0/2) Epoch 4, batch 32600, giga_loss[loss=0.2556, simple_loss=0.3384, pruned_loss=0.08644, over 28763.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3505, pruned_loss=0.1075, over 5667445.80 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3818, pruned_loss=0.1428, over 5701265.30 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3481, pruned_loss=0.1034, over 5670792.56 frames. ], batch size: 119, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:51:21,095 INFO [train.py:968] (0/2) Epoch 4, batch 32650, giga_loss[loss=0.2483, simple_loss=0.3288, pruned_loss=0.08391, over 28760.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3478, pruned_loss=0.1048, over 5664309.10 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.3819, pruned_loss=0.143, over 5701422.71 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3455, pruned_loss=0.1011, over 5666411.92 frames. ], batch size: 243, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:51:32,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:51:50,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.919e+02 1.202e+03 1.569e+03 2.166e+03 8.186e+03, threshold=3.139e+03, percent-clipped=3.0
+2023-03-02 08:52:19,527 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-02 08:52:23,103 INFO [train.py:968] (0/2) Epoch 4, batch 32700, giga_loss[loss=0.2284, simple_loss=0.2927, pruned_loss=0.08203, over 24365.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3463, pruned_loss=0.1034, over 5650594.85 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3816, pruned_loss=0.1427, over 5691833.24 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3438, pruned_loss=0.09963, over 5659736.37 frames. ], batch size: 705, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:52:25,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2627, 2.1061, 2.1070, 1.9765], device='cuda:0'), covar=tensor([0.0942, 0.1870, 0.1230, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0726, 0.0623, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 08:53:35,148 INFO [train.py:968] (0/2) Epoch 4, batch 32750, giga_loss[loss=0.2696, simple_loss=0.3402, pruned_loss=0.09952, over 28647.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3437, pruned_loss=0.1017, over 5658491.54 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3817, pruned_loss=0.1427, over 5693034.07 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3415, pruned_loss=0.09853, over 5664289.83 frames. ], batch size: 307, lr: 7.49e-03, grad_scale: 4.0
+2023-03-02 08:54:01,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.244e+02 1.428e+03 1.953e+03 2.714e+03 8.416e+03, threshold=3.905e+03, percent-clipped=16.0
+2023-03-02 08:54:32,913 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-02 08:54:38,123 INFO [train.py:968] (0/2) Epoch 4, batch 32800, giga_loss[loss=0.268, simple_loss=0.3431, pruned_loss=0.09646, over 28989.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3458, pruned_loss=0.1023, over 5675803.54 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3817, pruned_loss=0.1428, over 5699765.43 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3426, pruned_loss=0.09827, over 5673431.39 frames. ], batch size: 199, lr: 7.49e-03, grad_scale: 8.0
+2023-03-02 08:54:44,805 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:54:47,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:55:00,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0722, 1.1717, 4.2509, 3.3134], device='cuda:0'), covar=tensor([0.1579, 0.2258, 0.0285, 0.0706], device='cuda:0'), in_proj_covar=tensor([0.0548, 0.0505, 0.0701, 0.0553], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 08:55:26,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:55:40,224 INFO [train.py:968] (0/2) Epoch 4, batch 32850, giga_loss[loss=0.3245, simple_loss=0.3818, pruned_loss=0.1336, over 28037.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3464, pruned_loss=0.1033, over 5669864.90 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3815, pruned_loss=0.1428, over 5692962.13 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3428, pruned_loss=0.09858, over 5673130.84 frames. ], batch size: 412, lr: 7.49e-03, grad_scale: 8.0
+2023-03-02 08:55:40,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2243, 1.4653, 1.2029, 1.3379], device='cuda:0'), covar=tensor([0.2065, 0.1855, 0.1870, 0.1757], device='cuda:0'), in_proj_covar=tensor([0.1064, 0.0830, 0.0963, 0.0933], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 08:55:48,163 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:56:06,004 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.928e+02 1.284e+03 1.657e+03 2.178e+03 5.013e+03, threshold=3.314e+03, percent-clipped=3.0
+2023-03-02 08:56:38,030 INFO [train.py:968] (0/2) Epoch 4, batch 32900, giga_loss[loss=0.26, simple_loss=0.3363, pruned_loss=0.09189, over 28918.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3489, pruned_loss=0.1059, over 5663113.50 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3823, pruned_loss=0.1435, over 5679486.72 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3443, pruned_loss=0.1004, over 5676528.23 frames. ], batch size: 227, lr: 7.49e-03, grad_scale: 4.0
+2023-03-02 08:56:54,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2555, 1.3273, 1.1302, 1.4772], device='cuda:0'), covar=tensor([0.0831, 0.0345, 0.0363, 0.0898], device='cuda:0'), in_proj_covar=tensor([0.0208, 0.0133, 0.0137, 0.0248], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:0')
+2023-03-02 08:56:56,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5742, 2.1183, 1.9443, 1.8372], device='cuda:0'), covar=tensor([0.1714, 0.1921, 0.1250, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0740, 0.0759, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 08:57:40,017 INFO [train.py:968] (0/2) Epoch 4, batch 32950, giga_loss[loss=0.2851, simple_loss=0.3647, pruned_loss=0.1027, over 28678.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3469, pruned_loss=0.1035, over 5653809.04 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3822, pruned_loss=0.1434, over 5672418.94 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.343, pruned_loss=0.09875, over 5670918.03 frames. ], batch size: 242, lr: 7.49e-03, grad_scale: 4.0
+2023-03-02 08:58:07,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.749e+02 1.220e+03 1.667e+03 2.467e+03 6.749e+03, threshold=3.333e+03, percent-clipped=12.0
+2023-03-02 08:58:14,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0656, 1.8412, 1.3241, 1.5292], device='cuda:0'), covar=tensor([0.0526, 0.0539, 0.0866, 0.0860], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0450, 0.0505, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 08:58:35,838 INFO [train.py:968] (0/2) Epoch 4, batch 33000, giga_loss[loss=0.2999, simple_loss=0.374, pruned_loss=0.1129, over 28544.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3495, pruned_loss=0.1039, over 5645823.36 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3823, pruned_loss=0.1434, over 5669252.16 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3452, pruned_loss=0.09881, over 5661361.92 frames. ], batch size: 85, lr: 7.49e-03, grad_scale: 4.0
+2023-03-02 08:58:35,843 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 08:58:44,544 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2225, simple_loss=0.3201, pruned_loss=0.06249, over 944034.00 frames. 
+2023-03-02 08:58:44,545 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 08:58:45,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:58:50,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:59:22,464 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:59:42,442 INFO [train.py:968] (0/2) Epoch 4, batch 33050, giga_loss[loss=0.3059, simple_loss=0.3754, pruned_loss=0.1182, over 28909.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3533, pruned_loss=0.1057, over 5654207.84 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.382, pruned_loss=0.1431, over 5675502.21 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3492, pruned_loss=0.1009, over 5660347.99 frames. ], batch size: 227, lr: 7.49e-03, grad_scale: 4.0
+2023-03-02 09:00:12,199 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.537e+02 1.393e+03 1.708e+03 2.568e+03 1.024e+04, threshold=3.416e+03, percent-clipped=15.0
+2023-03-02 09:00:46,187 INFO [train.py:968] (0/2) Epoch 4, batch 33100, giga_loss[loss=0.3231, simple_loss=0.3756, pruned_loss=0.1353, over 26979.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3551, pruned_loss=0.107, over 5662375.58 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3816, pruned_loss=0.1429, over 5679175.17 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3514, pruned_loss=0.1024, over 5663785.33 frames. ], batch size: 555, lr: 7.49e-03, grad_scale: 2.0
+2023-03-02 09:01:03,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5363, 3.5825, 1.5325, 1.5422], device='cuda:0'), covar=tensor([0.0823, 0.0275, 0.0836, 0.1216], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0468, 0.0311, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 09:01:49,027 INFO [train.py:968] (0/2) Epoch 4, batch 33150, giga_loss[loss=0.2839, simple_loss=0.3584, pruned_loss=0.1047, over 28703.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.355, pruned_loss=0.1071, over 5658441.99 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.3817, pruned_loss=0.143, over 5673869.19 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3516, pruned_loss=0.1028, over 5663379.21 frames. ], batch size: 307, lr: 7.49e-03, grad_scale: 2.0
+2023-03-02 09:02:17,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.984e+02 1.432e+03 2.017e+03 2.669e+03 7.591e+03, threshold=4.034e+03, percent-clipped=14.0
+2023-03-02 09:02:23,844 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-02 09:02:47,790 INFO [train.py:968] (0/2) Epoch 4, batch 33200, giga_loss[loss=0.226, simple_loss=0.3045, pruned_loss=0.07372, over 28626.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3508, pruned_loss=0.1041, over 5667294.79 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3817, pruned_loss=0.1431, over 5675642.15 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3477, pruned_loss=0.1, over 5669566.17 frames. ], batch size: 85, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:03:08,239 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:03:09,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7307, 1.8340, 1.8057, 1.6643], device='cuda:0'), covar=tensor([0.1007, 0.1601, 0.1329, 0.1370], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0729, 0.0619, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 09:03:11,076 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169362.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 09:03:49,166 INFO [train.py:968] (0/2) Epoch 4, batch 33250, giga_loss[loss=0.2649, simple_loss=0.3356, pruned_loss=0.09711, over 28978.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3497, pruned_loss=0.1033, over 5670249.06 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3818, pruned_loss=0.1432, over 5678503.77 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3467, pruned_loss=0.09956, over 5669569.48 frames. ], batch size: 155, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:04:20,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.311e+02 1.385e+03 1.789e+03 2.739e+03 5.038e+03, threshold=3.579e+03, percent-clipped=4.0
+2023-03-02 09:04:50,415 INFO [train.py:968] (0/2) Epoch 4, batch 33300, giga_loss[loss=0.3037, simple_loss=0.3731, pruned_loss=0.1172, over 28717.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3481, pruned_loss=0.1031, over 5674732.55 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3814, pruned_loss=0.143, over 5681582.75 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3457, pruned_loss=0.09979, over 5671241.46 frames. ], batch size: 307, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:05:27,688 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:05:52,311 INFO [train.py:968] (0/2) Epoch 4, batch 33350, giga_loss[loss=0.3124, simple_loss=0.3825, pruned_loss=0.1212, over 28537.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3492, pruned_loss=0.1031, over 5677635.66 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3811, pruned_loss=0.1426, over 5687385.09 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3468, pruned_loss=0.09988, over 5669763.02 frames. ], batch size: 370, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:06:15,976 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:06:24,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.858e+02 1.298e+03 1.799e+03 2.601e+03 6.049e+03, threshold=3.599e+03, percent-clipped=9.0
+2023-03-02 09:06:46,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5713, 1.7179, 1.6237, 1.6349], device='cuda:0'), covar=tensor([0.0930, 0.1478, 0.1373, 0.1203], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0726, 0.0613, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 09:06:57,332 INFO [train.py:968] (0/2) Epoch 4, batch 33400, giga_loss[loss=0.2228, simple_loss=0.2897, pruned_loss=0.07797, over 24504.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3507, pruned_loss=0.1042, over 5676928.10 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.381, pruned_loss=0.1425, over 5690680.52 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3485, pruned_loss=0.1012, over 5667687.16 frames. ], batch size: 705, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:07:06,727 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:07:58,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5198, 1.9392, 1.2957, 0.9099], device='cuda:0'), covar=tensor([0.2861, 0.1739, 0.1724, 0.2413], device='cuda:0'), in_proj_covar=tensor([0.1322, 0.1274, 0.1334, 0.1113], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 09:08:00,196 INFO [train.py:968] (0/2) Epoch 4, batch 33450, giga_loss[loss=0.3473, simple_loss=0.4072, pruned_loss=0.1437, over 28468.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3532, pruned_loss=0.1066, over 5672904.77 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3813, pruned_loss=0.1426, over 5695253.41 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3505, pruned_loss=0.1033, over 5661336.44 frames. ], batch size: 336, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:08:34,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.971e+02 1.400e+03 1.878e+03 2.885e+03 8.289e+03, threshold=3.756e+03, percent-clipped=16.0
+2023-03-02 09:09:02,454 INFO [train.py:968] (0/2) Epoch 4, batch 33500, giga_loss[loss=0.2809, simple_loss=0.36, pruned_loss=0.1009, over 28896.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3577, pruned_loss=0.1091, over 5667597.83 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3804, pruned_loss=0.142, over 5698515.54 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3558, pruned_loss=0.1063, over 5655056.37 frames. ], batch size: 213, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:09:55,845 INFO [train.py:968] (0/2) Epoch 4, batch 33550, giga_loss[loss=0.3113, simple_loss=0.38, pruned_loss=0.1213, over 28686.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3594, pruned_loss=0.1095, over 5669460.35 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3803, pruned_loss=0.1418, over 5697729.57 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3572, pruned_loss=0.1063, over 5658876.63 frames. ], batch size: 242, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:10:28,732 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.848e+02 1.316e+03 1.633e+03 2.235e+03 4.542e+03, threshold=3.267e+03, percent-clipped=2.0
+2023-03-02 09:10:49,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:10:50,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169737.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 09:10:54,725 INFO [train.py:968] (0/2) Epoch 4, batch 33600, giga_loss[loss=0.2822, simple_loss=0.3634, pruned_loss=0.1005, over 28794.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3609, pruned_loss=0.1116, over 5652138.37 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3796, pruned_loss=0.1417, over 5678155.22 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3588, pruned_loss=0.1076, over 5660445.26 frames. ], batch size: 307, lr: 7.48e-03, grad_scale: 8.0
+2023-03-02 09:11:49,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4550, 2.1590, 1.3918, 0.6297], device='cuda:0'), covar=tensor([0.2781, 0.1303, 0.2418, 0.2985], device='cuda:0'), in_proj_covar=tensor([0.1327, 0.1277, 0.1336, 0.1118], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 09:12:05,453 INFO [train.py:968] (0/2) Epoch 4, batch 33650, giga_loss[loss=0.2736, simple_loss=0.3536, pruned_loss=0.09676, over 28920.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3576, pruned_loss=0.1095, over 5654513.63 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3791, pruned_loss=0.1412, over 5673324.18 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.356, pruned_loss=0.106, over 5665030.32 frames. ], batch size: 145, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:12:32,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.800e+02 1.468e+03 1.899e+03 3.149e+03 9.282e+03, threshold=3.798e+03, percent-clipped=24.0
+2023-03-02 09:12:52,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0550, 1.4054, 4.0660, 3.1524], device='cuda:0'), covar=tensor([0.1658, 0.2111, 0.0369, 0.0591], device='cuda:0'), in_proj_covar=tensor([0.0557, 0.0514, 0.0709, 0.0558], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 09:13:01,518 INFO [train.py:968] (0/2) Epoch 4, batch 33700, giga_loss[loss=0.3371, simple_loss=0.3892, pruned_loss=0.1425, over 27043.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3565, pruned_loss=0.1088, over 5666107.41 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3785, pruned_loss=0.1406, over 5675079.77 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3548, pruned_loss=0.1052, over 5672331.19 frames. ], batch size: 555, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:13:10,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:13:31,632 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9131, 1.0297, 0.9128, 0.7244], device='cuda:0'), covar=tensor([0.0730, 0.0683, 0.0463, 0.0634], device='cuda:0'), in_proj_covar=tensor([0.1318, 0.1055, 0.1056, 0.1133], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 09:13:48,668 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:13:52,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169880.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 09:13:52,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:13:54,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169883.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 09:13:54,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:14:04,638 INFO [train.py:968] (0/2) Epoch 4, batch 33750, libri_loss[loss=0.2583, simple_loss=0.3164, pruned_loss=0.1001, over 28574.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3555, pruned_loss=0.1087, over 5669654.16 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3784, pruned_loss=0.1406, over 5680247.06 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3536, pruned_loss=0.105, over 5669692.59 frames. ], batch size: 63, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:14:29,795 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:14:31,908 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169912.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 09:14:38,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.467e+02 1.489e+03 1.918e+03 2.472e+03 7.270e+03, threshold=3.837e+03, percent-clipped=6.0
+2023-03-02 09:14:53,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:15:13,378 INFO [train.py:968] (0/2) Epoch 4, batch 33800, giga_loss[loss=0.236, simple_loss=0.3239, pruned_loss=0.07403, over 28906.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.353, pruned_loss=0.1077, over 5668608.24 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3785, pruned_loss=0.1407, over 5678158.49 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3511, pruned_loss=0.1044, over 5670407.08 frames. ], batch size: 174, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:15:22,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3339, 1.8043, 1.2177, 1.5256], device='cuda:0'), covar=tensor([0.0781, 0.0370, 0.0362, 0.0840], device='cuda:0'), in_proj_covar=tensor([0.0208, 0.0132, 0.0136, 0.0246], device='cuda:0'), out_proj_covar=tensor([0.0053, 0.0038, 0.0034, 0.0058], device='cuda:0')
+2023-03-02 09:16:10,233 INFO [train.py:968] (0/2) Epoch 4, batch 33850, giga_loss[loss=0.27, simple_loss=0.3435, pruned_loss=0.09819, over 28915.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3522, pruned_loss=0.1074, over 5681199.04 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3783, pruned_loss=0.1405, over 5685545.92 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3501, pruned_loss=0.1038, over 5675702.34 frames. ], batch size: 106, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:16:11,019 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:16:13,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:16:16,440 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-170000.pt
+2023-03-02 09:16:34,028 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:16:37,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.784e+02 1.528e+03 1.912e+03 2.462e+03 5.235e+03, threshold=3.824e+03, percent-clipped=5.0
+2023-03-02 09:16:46,233 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:16:48,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:16:54,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:17:10,039 INFO [train.py:968] (0/2) Epoch 4, batch 33900, giga_loss[loss=0.2821, simple_loss=0.365, pruned_loss=0.09955, over 28952.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3516, pruned_loss=0.106, over 5679902.25 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3783, pruned_loss=0.1405, over 5690745.34 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3493, pruned_loss=0.1024, over 5670863.63 frames. ], batch size: 155, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:17:31,289 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:17:41,336 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:17:47,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4155, 1.8980, 1.3585, 1.5181], device='cuda:0'), covar=tensor([0.0837, 0.0281, 0.0339, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0207, 0.0133, 0.0136, 0.0245], device='cuda:0'), out_proj_covar=tensor([0.0053, 0.0038, 0.0034, 0.0058], device='cuda:0')
+2023-03-02 09:17:47,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:18:05,071 INFO [train.py:968] (0/2) Epoch 4, batch 33950, giga_loss[loss=0.2496, simple_loss=0.3162, pruned_loss=0.09151, over 24539.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3508, pruned_loss=0.1038, over 5673037.36 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3782, pruned_loss=0.1405, over 5683802.64 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3481, pruned_loss=0.09977, over 5672591.75 frames. ], batch size: 705, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:18:14,863 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:18:30,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.431e+02 1.267e+03 1.684e+03 2.288e+03 6.824e+03, threshold=3.367e+03, percent-clipped=2.0
+2023-03-02 09:18:36,762 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:18:59,212 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:19:02,494 INFO [train.py:968] (0/2) Epoch 4, batch 34000, giga_loss[loss=0.2704, simple_loss=0.3518, pruned_loss=0.09454, over 28855.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3526, pruned_loss=0.1028, over 5678383.90 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3781, pruned_loss=0.1403, over 5687779.42 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3501, pruned_loss=0.09913, over 5674494.31 frames. ], batch size: 186, lr: 7.47e-03, grad_scale: 8.0
+2023-03-02 09:19:11,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5938, 4.4328, 4.2267, 1.9244], device='cuda:0'), covar=tensor([0.0345, 0.0357, 0.0654, 0.2029], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0681, 0.0756, 0.0555], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 09:19:41,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:19:42,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8418, 4.6925, 4.4957, 2.2554], device='cuda:0'), covar=tensor([0.0345, 0.0354, 0.0694, 0.1772], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0685, 0.0762, 0.0558], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 09:19:57,721 INFO [train.py:968] (0/2) Epoch 4, batch 34050, giga_loss[loss=0.2615, simple_loss=0.3433, pruned_loss=0.08988, over 28658.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3533, pruned_loss=0.1034, over 5687840.78 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3774, pruned_loss=0.1401, over 5692695.91 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.351, pruned_loss=0.09918, over 5679975.59 frames. ], batch size: 307, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:20:32,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.706e+02 1.362e+03 1.760e+03 2.602e+03 8.116e+03, threshold=3.520e+03, percent-clipped=14.0
+2023-03-02 09:21:06,829 INFO [train.py:968] (0/2) Epoch 4, batch 34100, giga_loss[loss=0.2514, simple_loss=0.3348, pruned_loss=0.08398, over 28865.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3536, pruned_loss=0.1042, over 5682783.93 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3771, pruned_loss=0.1399, over 5700304.06 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3513, pruned_loss=0.0999, over 5669435.83 frames. ], batch size: 174, lr: 7.46e-03, grad_scale: 4.0
+2023-03-02 09:21:26,170 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6875, 1.1767, 2.9000, 2.6010], device='cuda:0'), covar=tensor([0.1438, 0.1869, 0.0482, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0547, 0.0505, 0.0702, 0.0550], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 09:22:09,563 INFO [train.py:968] (0/2) Epoch 4, batch 34150, giga_loss[loss=0.2327, simple_loss=0.3217, pruned_loss=0.07191, over 28859.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3544, pruned_loss=0.1048, over 5677130.34 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3771, pruned_loss=0.1398, over 5698748.79 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3519, pruned_loss=0.1004, over 5667161.73 frames. ], batch size: 174, lr: 7.46e-03, grad_scale: 2.0
+2023-03-02 09:22:47,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.881e+02 1.446e+03 1.962e+03 2.905e+03 1.294e+04, threshold=3.925e+03, percent-clipped=17.0
+2023-03-02 09:23:19,977 INFO [train.py:968] (0/2) Epoch 4, batch 34200, giga_loss[loss=0.2634, simple_loss=0.3395, pruned_loss=0.09365, over 28105.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3546, pruned_loss=0.1047, over 5677487.93 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3774, pruned_loss=0.1401, over 5703117.44 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3518, pruned_loss=0.09993, over 5664975.37 frames. ], batch size: 412, lr: 7.46e-03, grad_scale: 2.0
+2023-03-02 09:23:22,203 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-02 09:23:46,258 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 09:24:09,342 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-02 09:24:10,851 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 09:24:22,577 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:24:28,985 INFO [train.py:968] (0/2) Epoch 4, batch 34250, giga_loss[loss=0.3086, simple_loss=0.375, pruned_loss=0.1211, over 27717.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3544, pruned_loss=0.104, over 5679158.34 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3767, pruned_loss=0.1396, over 5708949.53 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.352, pruned_loss=0.09963, over 5663276.49 frames. ], batch size: 474, lr: 7.46e-03, grad_scale: 2.0
+2023-03-02 09:24:59,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.235e+02 1.305e+03 1.967e+03 2.971e+03 7.807e+03, threshold=3.934e+03, percent-clipped=13.0
+2023-03-02 09:25:25,528 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:25:30,283 INFO [train.py:968] (0/2) Epoch 4, batch 34300, giga_loss[loss=0.3515, simple_loss=0.4112, pruned_loss=0.146, over 29104.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3585, pruned_loss=0.107, over 5679597.86 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3765, pruned_loss=0.1396, over 5708522.31 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.356, pruned_loss=0.1022, over 5665534.81 frames. ], batch size: 200, lr: 7.46e-03, grad_scale: 2.0
+2023-03-02 09:26:08,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4796, 1.3739, 1.2507, 1.8003], device='cuda:0'), covar=tensor([0.2151, 0.2188, 0.2183, 0.2377], device='cuda:0'), in_proj_covar=tensor([0.1081, 0.0840, 0.0970, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 09:26:36,209 INFO [train.py:968] (0/2) Epoch 4, batch 34350, giga_loss[loss=0.2895, simple_loss=0.3676, pruned_loss=0.1057, over 28643.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.358, pruned_loss=0.1054, over 5683002.64 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3762, pruned_loss=0.1393, over 5710780.13 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3559, pruned_loss=0.1012, over 5669440.25 frames. ], batch size: 307, lr: 7.46e-03, grad_scale: 2.0
+2023-03-02 09:26:43,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:27:08,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:27:15,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.562e+02 1.336e+03 1.778e+03 2.271e+03 6.235e+03, threshold=3.556e+03, percent-clipped=4.0
+2023-03-02 09:27:32,471 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:27:36,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:27:45,666 INFO [train.py:968] (0/2) Epoch 4, batch 34400, giga_loss[loss=0.3141, simple_loss=0.3881, pruned_loss=0.1201, over 28367.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3578, pruned_loss=0.1065, over 5682612.92 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3766, pruned_loss=0.1397, over 5714871.51 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3554, pruned_loss=0.1022, over 5667424.81 frames. ], batch size: 368, lr: 7.46e-03, grad_scale: 4.0
+2023-03-02 09:27:59,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:28:10,988 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:28:46,797 INFO [train.py:968] (0/2) Epoch 4, batch 34450, giga_loss[loss=0.2565, simple_loss=0.3346, pruned_loss=0.08918, over 29147.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3558, pruned_loss=0.1058, over 5696153.69 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3769, pruned_loss=0.1397, over 5717891.99 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3528, pruned_loss=0.1009, over 5679937.11 frames. ], batch size: 200, lr: 7.46e-03, grad_scale: 4.0
+2023-03-02 09:29:23,687 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.743e+02 1.201e+03 1.632e+03 2.512e+03 7.925e+03, threshold=3.265e+03, percent-clipped=12.0
+2023-03-02 09:29:46,453 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:29:50,411 INFO [train.py:968] (0/2) Epoch 4, batch 34500, giga_loss[loss=0.2524, simple_loss=0.328, pruned_loss=0.08843, over 28604.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3534, pruned_loss=0.1038, over 5685833.62 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3767, pruned_loss=0.1396, over 5712400.06 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3501, pruned_loss=0.09828, over 5676831.01 frames. ], batch size: 78, lr: 7.46e-03, grad_scale: 4.0
+2023-03-02 09:29:50,794 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:30:08,167 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:30:11,011 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:30:26,417 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:30:49,929 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:30:52,434 INFO [train.py:968] (0/2) Epoch 4, batch 34550, giga_loss[loss=0.328, simple_loss=0.3931, pruned_loss=0.1314, over 27037.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3531, pruned_loss=0.1036, over 5679420.72 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3765, pruned_loss=0.1392, over 5717604.74 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3497, pruned_loss=0.09814, over 5666246.44 frames. ], batch size: 555, lr: 7.46e-03, grad_scale: 4.0
+2023-03-02 09:30:56,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:31:01,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:31:25,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.802e+02 1.318e+03 1.667e+03 2.205e+03 6.830e+03, threshold=3.333e+03, percent-clipped=7.0
+2023-03-02 09:31:35,618 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:31:52,020 INFO [train.py:968] (0/2) Epoch 4, batch 34600, giga_loss[loss=0.2877, simple_loss=0.3711, pruned_loss=0.1021, over 28509.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3556, pruned_loss=0.1057, over 5679060.20 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3757, pruned_loss=0.1387, over 5722143.60 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3532, pruned_loss=0.1009, over 5663451.56 frames. ], batch size: 336, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:32:06,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9287, 1.1708, 3.8993, 2.9213], device='cuda:0'), covar=tensor([0.1606, 0.2201, 0.0332, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0548, 0.0505, 0.0706, 0.0548], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 09:32:53,957 INFO [train.py:968] (0/2) Epoch 4, batch 34650, giga_loss[loss=0.2931, simple_loss=0.3671, pruned_loss=0.1096, over 27594.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3578, pruned_loss=0.1062, over 5686382.25 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3755, pruned_loss=0.1387, over 5723140.89 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3559, pruned_loss=0.1024, over 5673136.15 frames. ], batch size: 472, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:33:16,615 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7403, 1.1687, 2.9524, 2.6619], device='cuda:0'), covar=tensor([0.1420, 0.1920, 0.0476, 0.0671], device='cuda:0'), in_proj_covar=tensor([0.0545, 0.0502, 0.0703, 0.0549], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 09:33:21,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:33:27,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.073e+02 1.408e+03 1.885e+03 2.605e+03 4.886e+03, threshold=3.770e+03, percent-clipped=11.0
+2023-03-02 09:33:54,486 INFO [train.py:968] (0/2) Epoch 4, batch 34700, giga_loss[loss=0.2561, simple_loss=0.3272, pruned_loss=0.0925, over 28927.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3569, pruned_loss=0.108, over 5671820.44 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3759, pruned_loss=0.1391, over 5721049.05 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3547, pruned_loss=0.1039, over 5662429.41 frames. ], batch size: 199, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:34:08,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:34:11,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6792, 1.5993, 1.2421, 1.3862], device='cuda:0'), covar=tensor([0.0693, 0.0602, 0.0850, 0.0998], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0442, 0.0510, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 09:34:41,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3698, 1.4889, 1.2442, 1.6582], device='cuda:0'), covar=tensor([0.2039, 0.1753, 0.1754, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.1065, 0.0821, 0.0957, 0.0931], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 09:34:52,270 INFO [train.py:968] (0/2) Epoch 4, batch 34750, libri_loss[loss=0.3242, simple_loss=0.3694, pruned_loss=0.1395, over 29549.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3551, pruned_loss=0.107, over 5672860.20 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.376, pruned_loss=0.1393, over 5723076.64 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.353, pruned_loss=0.1033, over 5662975.92 frames. ], batch size: 75, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:35:22,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.581e+02 1.530e+03 1.970e+03 2.664e+03 9.225e+03, threshold=3.940e+03, percent-clipped=10.0
+2023-03-02 09:35:49,579 INFO [train.py:968] (0/2) Epoch 4, batch 34800, giga_loss[loss=0.2821, simple_loss=0.3641, pruned_loss=0.1, over 28888.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3572, pruned_loss=0.1092, over 5664709.36 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3759, pruned_loss=0.1392, over 5717253.50 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3551, pruned_loss=0.1056, over 5660439.48 frames. ], batch size: 174, lr: 7.45e-03, grad_scale: 8.0
+2023-03-02 09:36:00,812 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:36:04,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:36:07,617 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:36:29,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:36:34,036 INFO [train.py:968] (0/2) Epoch 4, batch 34850, giga_loss[loss=0.3383, simple_loss=0.4101, pruned_loss=0.1333, over 29067.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3679, pruned_loss=0.1161, over 5678035.56 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3762, pruned_loss=0.1395, over 5721036.90 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3657, pruned_loss=0.1125, over 5670529.23 frames. ], batch size: 155, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:37:00,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.272e+02 1.131e+03 1.655e+03 2.220e+03 7.274e+03, threshold=3.309e+03, percent-clipped=5.0
+2023-03-02 09:37:20,347 INFO [train.py:968] (0/2) Epoch 4, batch 34900, giga_loss[loss=0.3202, simple_loss=0.3924, pruned_loss=0.124, over 28823.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3754, pruned_loss=0.1216, over 5675462.07 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.376, pruned_loss=0.1394, over 5723821.34 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3737, pruned_loss=0.1181, over 5665280.31 frames. ], batch size: 119, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:37:52,519 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-02 09:38:03,277 INFO [train.py:968] (0/2) Epoch 4, batch 34950, giga_loss[loss=0.2872, simple_loss=0.3531, pruned_loss=0.1106, over 28963.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3759, pruned_loss=0.123, over 5673160.96 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3762, pruned_loss=0.1395, over 5718520.41 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3745, pruned_loss=0.1198, over 5669691.42 frames. ], batch size: 136, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:38:25,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.074e+02 1.209e+03 1.587e+03 2.147e+03 5.047e+03, threshold=3.174e+03, percent-clipped=7.0
+2023-03-02 09:38:43,436 INFO [train.py:968] (0/2) Epoch 4, batch 35000, giga_loss[loss=0.2821, simple_loss=0.3437, pruned_loss=0.1103, over 28886.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3692, pruned_loss=0.1199, over 5679982.81 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3763, pruned_loss=0.1394, over 5713591.04 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3679, pruned_loss=0.117, over 5680202.62 frames. ], batch size: 186, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:39:29,475 INFO [train.py:968] (0/2) Epoch 4, batch 35050, giga_loss[loss=0.2479, simple_loss=0.3119, pruned_loss=0.09195, over 28545.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3623, pruned_loss=0.1169, over 5682381.58 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3767, pruned_loss=0.1396, over 5717766.24 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3606, pruned_loss=0.1139, over 5678136.05 frames. ], batch size: 65, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:39:52,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.373e+02 1.064e+03 1.330e+03 1.876e+03 9.353e+03, threshold=2.660e+03, percent-clipped=10.0
+2023-03-02 09:40:00,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:40:13,183 INFO [train.py:968] (0/2) Epoch 4, batch 35100, giga_loss[loss=0.23, simple_loss=0.2963, pruned_loss=0.08186, over 28960.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3528, pruned_loss=0.1119, over 5685132.82 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3769, pruned_loss=0.1397, over 5718814.41 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3512, pruned_loss=0.1094, over 5680828.57 frames. ], batch size: 213, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:40:41,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1754, 0.8889, 0.8414, 1.3266], device='cuda:0'), covar=tensor([0.0793, 0.0374, 0.0371, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0204, 0.0130, 0.0132, 0.0242], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0057], device='cuda:0')
+2023-03-02 09:40:54,634 INFO [train.py:968] (0/2) Epoch 4, batch 35150, giga_loss[loss=0.253, simple_loss=0.3216, pruned_loss=0.09218, over 28796.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3467, pruned_loss=0.1095, over 5686364.70 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3774, pruned_loss=0.14, over 5718502.34 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3447, pruned_loss=0.1068, over 5682601.21 frames. ], batch size: 99, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:41:20,269 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.058e+02 9.268e+02 1.230e+03 1.729e+03 1.007e+04, threshold=2.461e+03, percent-clipped=11.0
+2023-03-02 09:41:20,473 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:41:29,559 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-02 09:41:31,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:41:35,744 INFO [train.py:968] (0/2) Epoch 4, batch 35200, giga_loss[loss=0.2525, simple_loss=0.3184, pruned_loss=0.09335, over 28866.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3437, pruned_loss=0.1089, over 5685032.96 frames. ], libri_tot_loss[loss=0.3299, simple_loss=0.3786, pruned_loss=0.1406, over 5714942.33 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3395, pruned_loss=0.1048, over 5684251.94 frames. ], batch size: 119, lr: 7.44e-03, grad_scale: 8.0
+2023-03-02 09:41:37,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1677, 2.1605, 1.4806, 1.3610], device='cuda:0'), covar=tensor([0.0968, 0.0296, 0.0348, 0.1159], device='cuda:0'), in_proj_covar=tensor([0.0203, 0.0129, 0.0132, 0.0242], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0057], device='cuda:0')
+2023-03-02 09:41:59,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:42:01,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:42:16,005 INFO [train.py:968] (0/2) Epoch 4, batch 35250, giga_loss[loss=0.3147, simple_loss=0.3621, pruned_loss=0.1336, over 27857.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3412, pruned_loss=0.1079, over 5685870.78 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3789, pruned_loss=0.1407, over 5714645.21 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3363, pruned_loss=0.1034, over 5684628.84 frames. ], batch size: 412, lr: 7.44e-03, grad_scale: 8.0
+2023-03-02 09:42:27,769 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:42:31,230 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-02 09:42:42,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.646e+02 1.287e+03 1.793e+03 2.320e+03 6.217e+03, threshold=3.587e+03, percent-clipped=23.0
+2023-03-02 09:42:56,695 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-02 09:42:58,752 INFO [train.py:968] (0/2) Epoch 4, batch 35300, libri_loss[loss=0.3348, simple_loss=0.3935, pruned_loss=0.138, over 29491.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3385, pruned_loss=0.1063, over 5687748.55 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.38, pruned_loss=0.1416, over 5709672.57 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3324, pruned_loss=0.1009, over 5690430.25 frames. ], batch size: 85, lr: 7.44e-03, grad_scale: 8.0
+2023-03-02 09:43:30,113 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:43:32,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:43:33,154 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:43:39,255 INFO [train.py:968] (0/2) Epoch 4, batch 35350, giga_loss[loss=0.2946, simple_loss=0.3468, pruned_loss=0.1212, over 28193.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3356, pruned_loss=0.1047, over 5699963.47 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.381, pruned_loss=0.1421, over 5714938.33 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3286, pruned_loss=0.09896, over 5696823.16 frames. ], batch size: 368, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:43:57,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:43:59,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9291, 3.6484, 3.6284, 1.6876], device='cuda:0'), covar=tensor([0.0500, 0.0476, 0.0714, 0.2079], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0697, 0.0766, 0.0573], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 09:44:04,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4083, 1.8318, 1.2705, 1.6268], device='cuda:0'), covar=tensor([0.0794, 0.0302, 0.0345, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0204, 0.0130, 0.0134, 0.0242], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:0')
+2023-03-02 09:44:05,576 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.110e+02 1.003e+03 1.361e+03 2.155e+03 5.586e+03, threshold=2.723e+03, percent-clipped=5.0
+2023-03-02 09:44:21,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-02 09:44:23,212 INFO [train.py:968] (0/2) Epoch 4, batch 35400, giga_loss[loss=0.2363, simple_loss=0.3073, pruned_loss=0.08269, over 28790.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3318, pruned_loss=0.1023, over 5694641.32 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3813, pruned_loss=0.1421, over 5708042.31 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3252, pruned_loss=0.09701, over 5697869.91 frames. ], batch size: 285, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:45:02,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:45:05,273 INFO [train.py:968] (0/2) Epoch 4, batch 35450, giga_loss[loss=0.2353, simple_loss=0.2991, pruned_loss=0.0857, over 28535.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3273, pruned_loss=0.09964, over 5697356.39 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3813, pruned_loss=0.142, over 5712164.67 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3212, pruned_loss=0.09482, over 5696067.10 frames. ], batch size: 85, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:45:32,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 4.261e+02 9.514e+02 1.323e+03 1.675e+03 8.166e+03, threshold=2.645e+03, percent-clipped=8.0
+2023-03-02 09:45:48,261 INFO [train.py:968] (0/2) Epoch 4, batch 35500, giga_loss[loss=0.23, simple_loss=0.2994, pruned_loss=0.08034, over 28877.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.325, pruned_loss=0.09875, over 5700531.89 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3812, pruned_loss=0.142, over 5717098.34 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3182, pruned_loss=0.09341, over 5694295.22 frames. ], batch size: 174, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:46:32,288 INFO [train.py:968] (0/2) Epoch 4, batch 35550, giga_loss[loss=0.2225, simple_loss=0.2904, pruned_loss=0.07733, over 28884.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3216, pruned_loss=0.09675, over 5690601.77 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3821, pruned_loss=0.1426, over 5710131.46 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3145, pruned_loss=0.09123, over 5691734.15 frames. ], batch size: 112, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:46:32,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1988, 1.3649, 1.2431, 1.4701], device='cuda:0'), covar=tensor([0.0856, 0.0345, 0.0351, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0203, 0.0130, 0.0133, 0.0241], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0057], device='cuda:0')
+2023-03-02 09:46:34,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:46:51,812 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3583, 2.0375, 1.4970, 0.6444], device='cuda:0'), covar=tensor([0.1838, 0.1106, 0.1911, 0.2252], device='cuda:0'), in_proj_covar=tensor([0.1328, 0.1251, 0.1335, 0.1102], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 09:46:55,802 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.874e+02 9.827e+02 1.256e+03 1.919e+03 7.485e+03, threshold=2.511e+03, percent-clipped=11.0
+2023-03-02 09:47:08,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:47:14,405 INFO [train.py:968] (0/2) Epoch 4, batch 35600, giga_loss[loss=0.1936, simple_loss=0.2683, pruned_loss=0.0595, over 28965.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3199, pruned_loss=0.09632, over 5676751.74 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3832, pruned_loss=0.1436, over 5688768.10 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3111, pruned_loss=0.08949, over 5695755.58 frames. ], batch size: 213, lr: 7.43e-03, grad_scale: 8.0
+2023-03-02 09:47:56,317 INFO [train.py:968] (0/2) Epoch 4, batch 35650, giga_loss[loss=0.3436, simple_loss=0.3917, pruned_loss=0.1478, over 28604.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3227, pruned_loss=0.09865, over 5674458.93 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3839, pruned_loss=0.1438, over 5691349.91 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3133, pruned_loss=0.09169, over 5687059.57 frames. ], batch size: 336, lr: 7.43e-03, grad_scale: 8.0
+2023-03-02 09:48:04,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8088, 1.6511, 1.3096, 1.3884], device='cuda:0'), covar=tensor([0.0679, 0.0649, 0.0991, 0.0957], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0448, 0.0508, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 09:48:05,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5505, 3.1125, 1.5570, 1.3640], device='cuda:0'), covar=tensor([0.0884, 0.0346, 0.0881, 0.1379], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0462, 0.0308, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 09:48:24,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.380e+02 1.081e+03 1.460e+03 2.055e+03 4.197e+03, threshold=2.920e+03, percent-clipped=15.0
+2023-03-02 09:48:27,357 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 09:48:39,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:48:42,932 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:48:43,361 INFO [train.py:968] (0/2) Epoch 4, batch 35700, libri_loss[loss=0.3598, simple_loss=0.4123, pruned_loss=0.1537, over 29159.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3353, pruned_loss=0.1054, over 5679272.43 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.384, pruned_loss=0.1438, over 5693558.53 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3269, pruned_loss=0.09933, over 5687040.04 frames. ], batch size: 97, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:48:59,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:49:09,659 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:49:15,109 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4686, 1.6030, 1.3768, 1.8055], device='cuda:0'), covar=tensor([0.1806, 0.1498, 0.1378, 0.1479], device='cuda:0'), in_proj_covar=tensor([0.1072, 0.0842, 0.0964, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 09:49:16,048 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-02 09:49:24,724 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:49:25,929 INFO [train.py:968] (0/2) Epoch 4, batch 35750, giga_loss[loss=0.3423, simple_loss=0.4026, pruned_loss=0.141, over 28553.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3496, pruned_loss=0.1135, over 5685221.19 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3843, pruned_loss=0.1436, over 5699431.82 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.341, pruned_loss=0.1075, over 5685621.98 frames. ], batch size: 336, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:49:49,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.994e+02 1.486e+03 1.823e+03 2.449e+03 3.863e+03, threshold=3.645e+03, percent-clipped=9.0
+2023-03-02 09:49:53,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6219, 1.7785, 1.7737, 1.7263], device='cuda:0'), covar=tensor([0.1220, 0.1494, 0.0957, 0.0938], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0742, 0.0759, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 09:50:06,764 INFO [train.py:968] (0/2) Epoch 4, batch 35800, giga_loss[loss=0.3714, simple_loss=0.4299, pruned_loss=0.1564, over 28906.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3605, pruned_loss=0.1193, over 5691877.51 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3843, pruned_loss=0.1434, over 5704843.76 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3526, pruned_loss=0.1137, over 5686972.05 frames. ], batch size: 136, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:50:24,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:50:40,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5227, 3.1364, 1.5218, 1.4432], device='cuda:0'), covar=tensor([0.0852, 0.0277, 0.0831, 0.1309], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0463, 0.0305, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 09:50:49,055 INFO [train.py:968] (0/2) Epoch 4, batch 35850, giga_loss[loss=0.3048, simple_loss=0.3701, pruned_loss=0.1197, over 28821.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3659, pruned_loss=0.1207, over 5695755.66 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3847, pruned_loss=0.1435, over 5706578.29 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3588, pruned_loss=0.1156, over 5690161.26 frames. ], batch size: 112, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:50:56,604 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-172000.pt
+2023-03-02 09:50:59,287 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:51:02,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:51:07,011 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-02 09:51:15,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.905e+02 1.116e+03 1.405e+03 1.967e+03 5.168e+03, threshold=2.810e+03, percent-clipped=5.0
+2023-03-02 09:51:30,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:51:36,075 INFO [train.py:968] (0/2) Epoch 4, batch 35900, giga_loss[loss=0.3669, simple_loss=0.4231, pruned_loss=0.1554, over 27898.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3664, pruned_loss=0.1189, over 5693333.80 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3844, pruned_loss=0.1433, over 5707272.54 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3608, pruned_loss=0.1148, over 5688311.80 frames. ], batch size: 412, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:51:42,496 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-02 09:52:13,215 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0697, 3.2828, 2.0993, 1.1444], device='cuda:0'), covar=tensor([0.2552, 0.1178, 0.1736, 0.2061], device='cuda:0'), in_proj_covar=tensor([0.1355, 0.1258, 0.1348, 0.1107], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 09:52:19,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3221, 1.3965, 1.1355, 0.8353], device='cuda:0'), covar=tensor([0.0678, 0.0629, 0.0485, 0.0695], device='cuda:0'), in_proj_covar=tensor([0.1322, 0.1062, 0.1076, 0.1145], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 09:52:19,923 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-02 09:52:20,682 INFO [train.py:968] (0/2) Epoch 4, batch 35950, giga_loss[loss=0.2996, simple_loss=0.3759, pruned_loss=0.1116, over 28913.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3696, pruned_loss=0.1204, over 5696672.23 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3847, pruned_loss=0.1433, over 5715272.12 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3642, pruned_loss=0.1162, over 5684862.67 frames. ], batch size: 213, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:52:32,914 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:52:34,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8133, 1.8741, 1.2930, 1.5333], device='cuda:0'), covar=tensor([0.0627, 0.0553, 0.0968, 0.0797], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0456, 0.0508, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 09:52:35,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:52:37,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:52:45,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.952e+02 1.204e+03 1.594e+03 2.142e+03 9.773e+03, threshold=3.187e+03, percent-clipped=14.0
+2023-03-02 09:53:01,738 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:53:02,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0481, 2.6830, 2.5065, 2.3461], device='cuda:0'), covar=tensor([0.0851, 0.1369, 0.1081, 0.1032], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0758, 0.0634, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 09:53:04,180 INFO [train.py:968] (0/2) Epoch 4, batch 36000, libri_loss[loss=0.3124, simple_loss=0.3627, pruned_loss=0.1311, over 29581.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1217, over 5699654.72 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3842, pruned_loss=0.1428, over 5719985.51 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3668, pruned_loss=0.1183, over 5685564.16 frames. ], batch size: 75, lr: 7.42e-03, grad_scale: 8.0
+2023-03-02 09:53:04,185 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 09:53:12,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2792, 1.4486, 1.4612, 1.4409], device='cuda:0'), covar=tensor([0.0932, 0.1282, 0.1432, 0.1097], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0757, 0.0633, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 09:53:12,630 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2385, simple_loss=0.3417, pruned_loss=0.06769, over 944034.00 frames. 
+2023-03-02 09:53:12,631 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 09:53:53,546 INFO [train.py:968] (0/2) Epoch 4, batch 36050, giga_loss[loss=0.2737, simple_loss=0.3479, pruned_loss=0.09971, over 28453.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3743, pruned_loss=0.1246, over 5694319.06 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3844, pruned_loss=0.1429, over 5724282.54 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3706, pruned_loss=0.1213, over 5678475.07 frames. ], batch size: 71, lr: 7.42e-03, grad_scale: 8.0
+2023-03-02 09:54:19,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.95 vs. limit=5.0
+2023-03-02 09:54:19,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.771e+02 1.115e+03 1.391e+03 1.796e+03 3.472e+03, threshold=2.781e+03, percent-clipped=2.0
+2023-03-02 09:54:37,934 INFO [train.py:968] (0/2) Epoch 4, batch 36100, giga_loss[loss=0.3896, simple_loss=0.4312, pruned_loss=0.174, over 27580.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3769, pruned_loss=0.1252, over 5695681.25 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3848, pruned_loss=0.143, over 5724216.39 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3735, pruned_loss=0.1224, over 5683096.13 frames. ], batch size: 472, lr: 7.42e-03, grad_scale: 8.0
+2023-03-02 09:54:49,481 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:54:51,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:54:56,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:55:15,508 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:55:19,522 INFO [train.py:968] (0/2) Epoch 4, batch 36150, giga_loss[loss=0.316, simple_loss=0.3853, pruned_loss=0.1234, over 28625.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3796, pruned_loss=0.1262, over 5694206.13 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3851, pruned_loss=0.1432, over 5719422.47 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3765, pruned_loss=0.1234, over 5688722.16 frames. ], batch size: 307, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:55:47,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.298e+02 1.122e+03 1.342e+03 1.967e+03 4.629e+03, threshold=2.683e+03, percent-clipped=8.0
+2023-03-02 09:55:52,621 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=172330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:56:02,663 INFO [train.py:968] (0/2) Epoch 4, batch 36200, giga_loss[loss=0.3165, simple_loss=0.3884, pruned_loss=0.1223, over 28524.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.382, pruned_loss=0.1276, over 5692092.77 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3862, pruned_loss=0.144, over 5723728.15 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3784, pruned_loss=0.1242, over 5682800.26 frames. ], batch size: 336, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:56:03,422 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 09:56:43,364 INFO [train.py:968] (0/2) Epoch 4, batch 36250, giga_loss[loss=0.2828, simple_loss=0.361, pruned_loss=0.1023, over 28625.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3831, pruned_loss=0.1273, over 5680084.52 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3866, pruned_loss=0.1443, over 5708061.91 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3797, pruned_loss=0.124, over 5686383.03 frames. ], batch size: 85, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:56:57,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:57:00,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:57:10,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.584e+02 1.203e+03 1.556e+03 1.865e+03 8.047e+03, threshold=3.111e+03, percent-clipped=10.0
+2023-03-02 09:57:23,508 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:57:24,589 INFO [train.py:968] (0/2) Epoch 4, batch 36300, giga_loss[loss=0.3012, simple_loss=0.3795, pruned_loss=0.1114, over 28964.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3825, pruned_loss=0.1254, over 5691338.75 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.3873, pruned_loss=0.1446, over 5708652.07 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3791, pruned_loss=0.1222, over 5695423.82 frames. ], batch size: 213, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:57:44,128 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2802, 1.3596, 1.1756, 1.4120], device='cuda:0'), covar=tensor([0.2084, 0.1913, 0.1899, 0.1946], device='cuda:0'), in_proj_covar=tensor([0.1082, 0.0840, 0.0974, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 09:58:08,465 INFO [train.py:968] (0/2) Epoch 4, batch 36350, giga_loss[loss=0.2812, simple_loss=0.3553, pruned_loss=0.1035, over 28895.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3787, pruned_loss=0.1213, over 5698458.91 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.3871, pruned_loss=0.1445, over 5709494.81 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3762, pruned_loss=0.1188, over 5700832.73 frames. ], batch size: 112, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:58:23,457 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=172511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:58:34,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.927e+02 1.030e+03 1.254e+03 1.649e+03 5.292e+03, threshold=2.508e+03, percent-clipped=5.0
+2023-03-02 09:58:49,006 INFO [train.py:968] (0/2) Epoch 4, batch 36400, giga_loss[loss=0.3089, simple_loss=0.3757, pruned_loss=0.1211, over 28849.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3797, pruned_loss=0.1226, over 5698356.51 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3885, pruned_loss=0.1454, over 5703391.49 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3762, pruned_loss=0.1191, over 5704897.00 frames. ], batch size: 112, lr: 7.41e-03, grad_scale: 8.0
+2023-03-02 09:59:33,551 INFO [train.py:968] (0/2) Epoch 4, batch 36450, giga_loss[loss=0.3176, simple_loss=0.3814, pruned_loss=0.1269, over 28919.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3823, pruned_loss=0.1276, over 5700639.35 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3882, pruned_loss=0.1451, over 5707334.48 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3796, pruned_loss=0.1246, over 5702169.77 frames. ], batch size: 227, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:00:00,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.037e+02 1.345e+03 2.021e+03 2.914e+03 9.317e+03, threshold=4.042e+03, percent-clipped=32.0
+2023-03-02 10:00:14,120 INFO [train.py:968] (0/2) Epoch 4, batch 36500, libri_loss[loss=0.2875, simple_loss=0.354, pruned_loss=0.1105, over 29354.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3848, pruned_loss=0.1317, over 5694485.12 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3884, pruned_loss=0.1452, over 5703031.69 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3823, pruned_loss=0.1289, over 5698490.90 frames. ], batch size: 71, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:00:25,472 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 10:00:59,227 INFO [train.py:968] (0/2) Epoch 4, batch 36550, giga_loss[loss=0.3468, simple_loss=0.4018, pruned_loss=0.146, over 28813.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3842, pruned_loss=0.1321, over 5698256.44 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3892, pruned_loss=0.1454, over 5706033.55 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3814, pruned_loss=0.1293, over 5698181.07 frames. ], batch size: 145, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:01:07,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5336, 3.8249, 1.6759, 1.4462], device='cuda:0'), covar=tensor([0.0832, 0.0265, 0.0757, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0466, 0.0306, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 10:01:08,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:01:21,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.016e+02 1.102e+03 1.520e+03 2.284e+03 1.548e+04, threshold=3.040e+03, percent-clipped=9.0
+2023-03-02 10:01:24,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5053, 4.1725, 4.1842, 1.5129], device='cuda:0'), covar=tensor([0.0445, 0.0453, 0.0815, 0.1982], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0698, 0.0780, 0.0578], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 10:01:37,287 INFO [train.py:968] (0/2) Epoch 4, batch 36600, giga_loss[loss=0.2971, simple_loss=0.3649, pruned_loss=0.1147, over 29058.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3828, pruned_loss=0.1323, over 5698488.35 frames. ], libri_tot_loss[loss=0.3402, simple_loss=0.3893, pruned_loss=0.1456, over 5705117.09 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3801, pruned_loss=0.1292, over 5698960.85 frames. ], batch size: 155, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:02:23,162 INFO [train.py:968] (0/2) Epoch 4, batch 36650, giga_loss[loss=0.3561, simple_loss=0.4019, pruned_loss=0.1552, over 26836.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3813, pruned_loss=0.1311, over 5697037.78 frames. ], libri_tot_loss[loss=0.3404, simple_loss=0.3895, pruned_loss=0.1456, over 5707894.53 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3789, pruned_loss=0.1285, over 5694847.38 frames. ], batch size: 555, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:02:51,265 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.018e+02 1.107e+03 1.338e+03 2.015e+03 8.241e+03, threshold=2.677e+03, percent-clipped=8.0
+2023-03-02 10:03:08,716 INFO [train.py:968] (0/2) Epoch 4, batch 36700, giga_loss[loss=0.3312, simple_loss=0.3998, pruned_loss=0.1313, over 29015.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3794, pruned_loss=0.1285, over 5695268.84 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3898, pruned_loss=0.1458, over 5707949.05 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3773, pruned_loss=0.1262, over 5693474.45 frames. ], batch size: 164, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:03:12,284 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:03:16,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:03:43,061 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:03:48,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:03:55,179 INFO [train.py:968] (0/2) Epoch 4, batch 36750, giga_loss[loss=0.331, simple_loss=0.378, pruned_loss=0.142, over 26437.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3765, pruned_loss=0.126, over 5689452.40 frames. ], libri_tot_loss[loss=0.3409, simple_loss=0.3901, pruned_loss=0.1459, over 5710729.68 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3744, pruned_loss=0.1238, over 5685302.80 frames. ], batch size: 555, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:04:26,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.350e+02 1.047e+03 1.309e+03 1.681e+03 3.720e+03, threshold=2.619e+03, percent-clipped=3.0
+2023-03-02 10:04:43,049 INFO [train.py:968] (0/2) Epoch 4, batch 36800, giga_loss[loss=0.28, simple_loss=0.3457, pruned_loss=0.1072, over 28225.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3708, pruned_loss=0.1227, over 5677720.96 frames. ], libri_tot_loss[loss=0.341, simple_loss=0.3902, pruned_loss=0.1459, over 5713022.12 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3687, pruned_loss=0.1205, over 5671804.40 frames. ], batch size: 368, lr: 7.41e-03, grad_scale: 8.0
+2023-03-02 10:04:53,545 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 10:04:59,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6957, 2.1878, 2.0025, 1.8180], device='cuda:0'), covar=tensor([0.1637, 0.1594, 0.1118, 0.1105], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0755, 0.0768, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 10:05:31,450 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=172991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:05:32,395 INFO [train.py:968] (0/2) Epoch 4, batch 36850, giga_loss[loss=0.2732, simple_loss=0.3443, pruned_loss=0.101, over 28791.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3649, pruned_loss=0.1198, over 5660740.63 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3913, pruned_loss=0.1469, over 5707630.60 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3616, pruned_loss=0.1166, over 5660265.15 frames. ], batch size: 284, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:06:00,665 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:06:09,080 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.846e+02 1.091e+03 1.363e+03 2.025e+03 8.423e+03, threshold=2.727e+03, percent-clipped=13.0
+2023-03-02 10:06:12,100 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:06:15,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:06:27,255 INFO [train.py:968] (0/2) Epoch 4, batch 36900, giga_loss[loss=0.2966, simple_loss=0.3464, pruned_loss=0.1234, over 26495.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3602, pruned_loss=0.1175, over 5644744.37 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3923, pruned_loss=0.1476, over 5700626.40 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3562, pruned_loss=0.1138, over 5649357.70 frames. ], batch size: 555, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:06:40,592 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:07:03,807 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-02 10:07:10,327 INFO [train.py:968] (0/2) Epoch 4, batch 36950, giga_loss[loss=0.3395, simple_loss=0.3984, pruned_loss=0.1403, over 28319.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3614, pruned_loss=0.1175, over 5663949.43 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.3934, pruned_loss=0.1483, over 5705185.62 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3565, pruned_loss=0.1133, over 5662564.73 frames. ], batch size: 369, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:07:38,044 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.584e+02 9.918e+02 1.220e+03 1.576e+03 3.832e+03, threshold=2.440e+03, percent-clipped=4.0
+2023-03-02 10:07:38,878 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:07:39,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2838, 1.5530, 1.0833, 0.6748], device='cuda:0'), covar=tensor([0.1043, 0.0715, 0.0658, 0.0954], device='cuda:0'), in_proj_covar=tensor([0.1344, 0.1081, 0.1106, 0.1189], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 10:07:46,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9128, 1.0330, 4.1359, 3.1475], device='cuda:0'), covar=tensor([0.1553, 0.2123, 0.0336, 0.0623], device='cuda:0'), in_proj_covar=tensor([0.0553, 0.0514, 0.0711, 0.0569], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 10:07:50,983 INFO [train.py:968] (0/2) Epoch 4, batch 37000, giga_loss[loss=0.2517, simple_loss=0.3175, pruned_loss=0.09295, over 28538.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3612, pruned_loss=0.1172, over 5669807.81 frames. ], libri_tot_loss[loss=0.3458, simple_loss=0.3942, pruned_loss=0.1486, over 5708820.32 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.356, pruned_loss=0.1129, over 5664614.71 frames. ], batch size: 71, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:08:17,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4908, 2.1623, 2.1318, 1.9063], device='cuda:0'), covar=tensor([0.1042, 0.1926, 0.1572, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0751, 0.0641, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 10:08:29,228 INFO [train.py:968] (0/2) Epoch 4, batch 37050, giga_loss[loss=0.2842, simple_loss=0.354, pruned_loss=0.1072, over 28737.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3633, pruned_loss=0.1187, over 5680837.11 frames. ], libri_tot_loss[loss=0.3481, simple_loss=0.3964, pruned_loss=0.1499, over 5703786.54 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3553, pruned_loss=0.1124, over 5678943.93 frames. ], batch size: 262, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:08:31,637 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5788, 1.4930, 1.3599, 1.6588], device='cuda:0'), covar=tensor([0.2131, 0.2184, 0.2022, 0.2482], device='cuda:0'), in_proj_covar=tensor([0.1081, 0.0838, 0.0960, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 10:08:42,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4375, 2.8452, 1.4211, 1.4205], device='cuda:0'), covar=tensor([0.0874, 0.0366, 0.0821, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0459, 0.0304, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 10:08:55,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.783e+02 1.067e+03 1.432e+03 2.215e+03 1.235e+04, threshold=2.864e+03, percent-clipped=19.0
+2023-03-02 10:09:04,211 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:09:07,743 INFO [train.py:968] (0/2) Epoch 4, batch 37100, giga_loss[loss=0.2863, simple_loss=0.3412, pruned_loss=0.1157, over 28769.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3597, pruned_loss=0.1169, over 5678813.78 frames. ], libri_tot_loss[loss=0.3487, simple_loss=0.3969, pruned_loss=0.1502, over 5687247.74 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3519, pruned_loss=0.1107, over 5691870.91 frames. ], batch size: 92, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:09:28,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5534, 4.3627, 4.2506, 1.7290], device='cuda:0'), covar=tensor([0.0440, 0.0376, 0.0639, 0.2180], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0698, 0.0772, 0.0577], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 10:09:47,590 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-02 10:09:49,809 INFO [train.py:968] (0/2) Epoch 4, batch 37150, giga_loss[loss=0.2937, simple_loss=0.3473, pruned_loss=0.12, over 23974.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3562, pruned_loss=0.115, over 5686010.85 frames. ], libri_tot_loss[loss=0.3486, simple_loss=0.3968, pruned_loss=0.1502, over 5689073.70 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3497, pruned_loss=0.1098, over 5694683.92 frames. ], batch size: 705, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:10:14,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.361e+02 9.675e+02 1.168e+03 1.787e+03 7.335e+03, threshold=2.336e+03, percent-clipped=10.0
+2023-03-02 10:10:27,422 INFO [train.py:968] (0/2) Epoch 4, batch 37200, giga_loss[loss=0.2373, simple_loss=0.3041, pruned_loss=0.08518, over 28616.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3545, pruned_loss=0.1141, over 5687267.30 frames. ], libri_tot_loss[loss=0.349, simple_loss=0.3974, pruned_loss=0.1503, over 5682306.00 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3473, pruned_loss=0.1085, over 5699794.61 frames. ], batch size: 85, lr: 7.40e-03, grad_scale: 8.0
+2023-03-02 10:10:43,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4869, 1.3379, 1.2706, 1.6631], device='cuda:0'), covar=tensor([0.2088, 0.2133, 0.2020, 0.2373], device='cuda:0'), in_proj_covar=tensor([0.1086, 0.0840, 0.0964, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 10:10:44,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:11:00,648 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-02 10:11:06,938 INFO [train.py:968] (0/2) Epoch 4, batch 37250, giga_loss[loss=0.4281, simple_loss=0.45, pruned_loss=0.2031, over 27986.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3514, pruned_loss=0.1121, over 5689082.87 frames. ], libri_tot_loss[loss=0.35, simple_loss=0.3984, pruned_loss=0.1508, over 5675928.48 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3439, pruned_loss=0.1065, over 5704933.61 frames. ], batch size: 412, lr: 7.40e-03, grad_scale: 8.0
+2023-03-02 10:11:07,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:11:33,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.472e+02 8.967e+02 1.104e+03 1.610e+03 4.135e+03, threshold=2.208e+03, percent-clipped=9.0
+2023-03-02 10:11:48,450 INFO [train.py:968] (0/2) Epoch 4, batch 37300, giga_loss[loss=0.2525, simple_loss=0.3162, pruned_loss=0.09439, over 28500.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3482, pruned_loss=0.1098, over 5692015.85 frames. ], libri_tot_loss[loss=0.3507, simple_loss=0.3991, pruned_loss=0.1511, over 5665010.37 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3411, pruned_loss=0.1047, over 5715119.57 frames. ], batch size: 71, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:12:27,962 INFO [train.py:968] (0/2) Epoch 4, batch 37350, giga_loss[loss=0.2594, simple_loss=0.3235, pruned_loss=0.09766, over 28089.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3456, pruned_loss=0.1083, over 5694854.94 frames. ], libri_tot_loss[loss=0.3514, simple_loss=0.3998, pruned_loss=0.1514, over 5667089.98 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3386, pruned_loss=0.1034, over 5711580.22 frames. ], batch size: 77, lr: 7.39e-03, grad_scale: 4.0
+2023-03-02 10:12:34,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:12:39,703 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:12:42,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:12:54,921 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.688e+02 9.518e+02 1.136e+03 1.501e+03 7.411e+03, threshold=2.273e+03, percent-clipped=8.0
+2023-03-02 10:12:55,223 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2986, 1.3880, 1.2017, 1.5004], device='cuda:0'), covar=tensor([0.2127, 0.2032, 0.2058, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.1091, 0.0846, 0.0976, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 10:13:01,992 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:13:03,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:13:05,407 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:13:06,359 INFO [train.py:968] (0/2) Epoch 4, batch 37400, giga_loss[loss=0.2506, simple_loss=0.3221, pruned_loss=0.08955, over 28951.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3445, pruned_loss=0.1078, over 5696509.57 frames. ], libri_tot_loss[loss=0.3522, simple_loss=0.4007, pruned_loss=0.1519, over 5663866.31 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3367, pruned_loss=0.1023, over 5712941.46 frames. ], batch size: 174, lr: 7.39e-03, grad_scale: 4.0
+2023-03-02 10:13:06,628 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:13:27,150 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:13:33,434 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-02 10:13:46,308 INFO [train.py:968] (0/2) Epoch 4, batch 37450, giga_loss[loss=0.233, simple_loss=0.3106, pruned_loss=0.07768, over 28936.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3438, pruned_loss=0.107, over 5694358.23 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4018, pruned_loss=0.1524, over 5659176.94 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.335, pruned_loss=0.1009, over 5712867.19 frames. ], batch size: 155, lr: 7.39e-03, grad_scale: 4.0
+2023-03-02 10:14:05,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:14:15,927 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-02 10:14:18,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.706e+02 1.016e+03 1.212e+03 1.597e+03 3.794e+03, threshold=2.424e+03, percent-clipped=9.0
+2023-03-02 10:14:30,636 INFO [train.py:968] (0/2) Epoch 4, batch 37500, giga_loss[loss=0.3545, simple_loss=0.4042, pruned_loss=0.1524, over 28800.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3435, pruned_loss=0.1071, over 5698492.85 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.4021, pruned_loss=0.1525, over 5660439.05 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3362, pruned_loss=0.1021, over 5712017.87 frames. ], batch size: 242, lr: 7.39e-03, grad_scale: 4.0
+2023-03-02 10:14:33,241 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:14:35,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:14:40,633 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-02 10:15:01,895 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:15:14,248 INFO [train.py:968] (0/2) Epoch 4, batch 37550, giga_loss[loss=0.3482, simple_loss=0.4016, pruned_loss=0.1474, over 28601.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3493, pruned_loss=0.1107, over 5687891.00 frames. ], libri_tot_loss[loss=0.3543, simple_loss=0.403, pruned_loss=0.1528, over 5644399.70 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3412, pruned_loss=0.1053, over 5714196.02 frames. ], batch size: 336, lr: 7.39e-03, grad_scale: 4.0
+2023-03-02 10:15:22,873 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:15:41,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.538e+02 1.164e+03 1.483e+03 2.281e+03 5.282e+03, threshold=2.966e+03, percent-clipped=22.0
+2023-03-02 10:15:57,428 INFO [train.py:968] (0/2) Epoch 4, batch 37600, giga_loss[loss=0.3471, simple_loss=0.4, pruned_loss=0.1471, over 27659.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3559, pruned_loss=0.1156, over 5692142.29 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4023, pruned_loss=0.1523, over 5651931.26 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3484, pruned_loss=0.1104, over 5708063.66 frames. ], batch size: 472, lr: 7.39e-03, grad_scale: 8.0
+2023-03-02 10:16:11,651 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:16:13,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:16:44,583 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:16:45,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5148, 2.9473, 1.6433, 1.7367], device='cuda:0'), covar=tensor([0.0940, 0.0535, 0.0710, 0.0784], device='cuda:0'), in_proj_covar=tensor([0.1342, 0.1093, 0.1099, 0.1182], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 10:16:47,990 INFO [train.py:968] (0/2) Epoch 4, batch 37650, giga_loss[loss=0.3289, simple_loss=0.3921, pruned_loss=0.1328, over 28906.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3646, pruned_loss=0.1213, over 5685083.08 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4025, pruned_loss=0.1522, over 5652570.06 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3579, pruned_loss=0.1168, over 5697593.78 frames. ], batch size: 106, lr: 7.39e-03, grad_scale: 8.0
+2023-03-02 10:17:21,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.885e+02 1.198e+03 1.537e+03 1.897e+03 3.500e+03, threshold=3.074e+03, percent-clipped=5.0
+2023-03-02 10:17:37,384 INFO [train.py:968] (0/2) Epoch 4, batch 37700, giga_loss[loss=0.3153, simple_loss=0.3866, pruned_loss=0.122, over 29020.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3703, pruned_loss=0.1249, over 5663743.77 frames. ], libri_tot_loss[loss=0.3542, simple_loss=0.4028, pruned_loss=0.1528, over 5641558.13 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3633, pruned_loss=0.1197, over 5684864.38 frames. ], batch size: 164, lr: 7.39e-03, grad_scale: 2.0
+2023-03-02 10:18:07,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5986, 3.1774, 1.5884, 1.6706], device='cuda:0'), covar=tensor([0.0870, 0.0302, 0.0840, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0331, 0.0455, 0.0302, 0.0375], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 10:18:12,589 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173884.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 10:18:20,286 INFO [train.py:968] (0/2) Epoch 4, batch 37750, giga_loss[loss=0.3656, simple_loss=0.4195, pruned_loss=0.1558, over 28918.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.375, pruned_loss=0.1267, over 5674043.18 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4023, pruned_loss=0.1524, over 5644627.31 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3692, pruned_loss=0.1223, over 5688996.86 frames. ], batch size: 199, lr: 7.39e-03, grad_scale: 2.0
+2023-03-02 10:18:27,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-02 10:18:30,555 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 10:18:43,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:18:52,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.255e+02 1.134e+03 1.446e+03 2.409e+03 2.080e+04, threshold=2.891e+03, percent-clipped=15.0
+2023-03-02 10:19:01,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1737, 3.8978, 3.8602, 1.6489], device='cuda:0'), covar=tensor([0.0486, 0.0459, 0.0738, 0.2292], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0704, 0.0787, 0.0583], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 10:19:06,412 INFO [train.py:968] (0/2) Epoch 4, batch 37800, giga_loss[loss=0.461, simple_loss=0.4678, pruned_loss=0.2271, over 23384.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3815, pruned_loss=0.1307, over 5666487.36 frames. ], libri_tot_loss[loss=0.3539, simple_loss=0.4026, pruned_loss=0.1526, over 5644797.82 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3757, pruned_loss=0.1261, over 5679337.70 frames. ], batch size: 705, lr: 7.39e-03, grad_scale: 2.0
+2023-03-02 10:19:08,865 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-02 10:19:48,884 INFO [train.py:968] (0/2) Epoch 4, batch 37850, giga_loss[loss=0.3262, simple_loss=0.3793, pruned_loss=0.1365, over 28748.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3827, pruned_loss=0.1312, over 5678234.76 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4023, pruned_loss=0.1524, over 5647541.78 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3781, pruned_loss=0.1275, over 5686138.53 frames. ], batch size: 92, lr: 7.38e-03, grad_scale: 2.0
+2023-03-02 10:19:55,445 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-174000.pt
+2023-03-02 10:19:58,791 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-02 10:20:00,568 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:20:17,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.908e+02 1.161e+03 1.475e+03 2.720e+03 8.461e+03, threshold=2.949e+03, percent-clipped=18.0
+2023-03-02 10:20:29,969 INFO [train.py:968] (0/2) Epoch 4, batch 37900, giga_loss[loss=0.2885, simple_loss=0.3588, pruned_loss=0.1091, over 28715.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3772, pruned_loss=0.1268, over 5683332.90 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4023, pruned_loss=0.1524, over 5651312.50 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3729, pruned_loss=0.1232, over 5687032.84 frames. ], batch size: 284, lr: 7.38e-03, grad_scale: 2.0
+2023-03-02 10:20:44,949 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:20:46,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:21:01,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:21:13,424 INFO [train.py:968] (0/2) Epoch 4, batch 37950, giga_loss[loss=0.3653, simple_loss=0.4123, pruned_loss=0.1592, over 28894.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3738, pruned_loss=0.1235, over 5691608.83 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4024, pruned_loss=0.1525, over 5653700.18 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3699, pruned_loss=0.1201, over 5692879.48 frames. ], batch size: 213, lr: 7.38e-03, grad_scale: 2.0
+2023-03-02 10:21:14,320 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:21:15,032 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-02 10:21:44,291 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.768e+02 1.078e+03 1.435e+03 1.953e+03 5.230e+03, threshold=2.870e+03, percent-clipped=13.0
+2023-03-02 10:21:55,437 INFO [train.py:968] (0/2) Epoch 4, batch 38000, giga_loss[loss=0.3328, simple_loss=0.3877, pruned_loss=0.1389, over 28945.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3747, pruned_loss=0.1241, over 5697243.15 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4024, pruned_loss=0.1525, over 5659783.46 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3708, pruned_loss=0.1205, over 5693600.97 frames. ], batch size: 106, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:22:08,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3042, 2.0265, 1.4548, 0.5329], device='cuda:0'), covar=tensor([0.2407, 0.1171, 0.1908, 0.2558], device='cuda:0'), in_proj_covar=tensor([0.1350, 0.1259, 0.1363, 0.1118], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 10:22:29,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6703, 1.6164, 1.5085, 1.5546], device='cuda:0'), covar=tensor([0.1153, 0.1890, 0.1703, 0.1487], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0748, 0.0634, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 10:22:40,355 INFO [train.py:968] (0/2) Epoch 4, batch 38050, giga_loss[loss=0.2882, simple_loss=0.3575, pruned_loss=0.1094, over 28833.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3766, pruned_loss=0.1248, over 5697769.19 frames. ], libri_tot_loss[loss=0.3538, simple_loss=0.4024, pruned_loss=0.1526, over 5661028.55 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3734, pruned_loss=0.1218, over 5694064.47 frames. ], batch size: 112, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:23:01,729 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:23:04,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:23:09,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.639e+02 1.140e+03 1.697e+03 2.420e+03 6.931e+03, threshold=3.394e+03, percent-clipped=19.0
+2023-03-02 10:23:21,065 INFO [train.py:968] (0/2) Epoch 4, batch 38100, giga_loss[loss=0.3157, simple_loss=0.383, pruned_loss=0.1242, over 28713.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3803, pruned_loss=0.1278, over 5706261.92 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.402, pruned_loss=0.1524, over 5666952.57 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3773, pruned_loss=0.1248, over 5699151.84 frames. ], batch size: 242, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:23:30,094 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:23:35,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174259.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 10:24:07,145 INFO [train.py:968] (0/2) Epoch 4, batch 38150, giga_loss[loss=0.3818, simple_loss=0.4144, pruned_loss=0.1745, over 26617.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3826, pruned_loss=0.1298, over 5690348.73 frames. ], libri_tot_loss[loss=0.3538, simple_loss=0.4024, pruned_loss=0.1526, over 5656927.54 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3798, pruned_loss=0.1271, over 5694789.99 frames. ], batch size: 555, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:24:09,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3221, 1.3886, 1.1254, 0.7224], device='cuda:0'), covar=tensor([0.0786, 0.0702, 0.0586, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.1333, 0.1089, 0.1103, 0.1183], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 10:24:15,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1225, 1.2712, 1.1151, 1.3826], device='cuda:0'), covar=tensor([0.0822, 0.0345, 0.0338, 0.0869], device='cuda:0'), in_proj_covar=tensor([0.0201, 0.0129, 0.0132, 0.0239], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:0')
+2023-03-02 10:24:40,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.049e+02 1.168e+03 1.467e+03 1.939e+03 4.451e+03, threshold=2.934e+03, percent-clipped=3.0
+2023-03-02 10:24:43,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1898, 1.2695, 1.0002, 0.9787], device='cuda:0'), covar=tensor([0.0570, 0.0423, 0.0914, 0.0751], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0447, 0.0510, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 10:24:52,959 INFO [train.py:968] (0/2) Epoch 4, batch 38200, giga_loss[loss=0.3042, simple_loss=0.3753, pruned_loss=0.1166, over 29040.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3823, pruned_loss=0.1299, over 5695794.48 frames. ], libri_tot_loss[loss=0.3538, simple_loss=0.4022, pruned_loss=0.1527, over 5663361.37 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3797, pruned_loss=0.127, over 5694307.58 frames. ], batch size: 164, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:24:54,073 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-02 10:25:28,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:25:35,904 INFO [train.py:968] (0/2) Epoch 4, batch 38250, giga_loss[loss=0.279, simple_loss=0.355, pruned_loss=0.1015, over 28259.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3837, pruned_loss=0.1314, over 5701555.16 frames. ], libri_tot_loss[loss=0.354, simple_loss=0.4023, pruned_loss=0.1528, over 5668189.31 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3812, pruned_loss=0.1287, over 5696469.26 frames. ], batch size: 77, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:25:44,674 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174402.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 10:25:47,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174405.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 10:25:49,299 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:25:51,342 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:25:59,152 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 10:26:08,159 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.162e+02 1.212e+03 1.482e+03 1.945e+03 5.528e+03, threshold=2.965e+03, percent-clipped=8.0
+2023-03-02 10:26:11,933 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174434.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 10:26:14,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0264, 1.1070, 0.9817, 0.7286], device='cuda:0'), covar=tensor([0.0684, 0.0715, 0.0566, 0.0697], device='cuda:0'), in_proj_covar=tensor([0.1347, 0.1101, 0.1116, 0.1198], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 10:26:19,922 INFO [train.py:968] (0/2) Epoch 4, batch 38300, giga_loss[loss=0.3449, simple_loss=0.4019, pruned_loss=0.1439, over 27801.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3835, pruned_loss=0.1307, over 5693837.80 frames. ], libri_tot_loss[loss=0.3543, simple_loss=0.4026, pruned_loss=0.153, over 5666833.27 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3811, pruned_loss=0.1282, over 5691095.91 frames. ], batch size: 412, lr: 7.37e-03, grad_scale: 4.0
+2023-03-02 10:26:59,159 INFO [train.py:968] (0/2) Epoch 4, batch 38350, giga_loss[loss=0.3263, simple_loss=0.4005, pruned_loss=0.126, over 28959.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3834, pruned_loss=0.129, over 5705307.89 frames. ], libri_tot_loss[loss=0.3546, simple_loss=0.4028, pruned_loss=0.1532, over 5670982.02 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3809, pruned_loss=0.1264, over 5699986.15 frames. ], batch size: 213, lr: 7.37e-03, grad_scale: 4.0
+2023-03-02 10:27:27,686 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:27:29,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:27:29,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.165e+02 1.015e+03 1.240e+03 1.751e+03 5.365e+03, threshold=2.481e+03, percent-clipped=9.0
+2023-03-02 10:27:32,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:27:39,674 INFO [train.py:968] (0/2) Epoch 4, batch 38400, giga_loss[loss=0.3419, simple_loss=0.4029, pruned_loss=0.1405, over 28744.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3841, pruned_loss=0.1289, over 5709952.40 frames. ], libri_tot_loss[loss=0.3547, simple_loss=0.4028, pruned_loss=0.1533, over 5676080.90 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3815, pruned_loss=0.1259, over 5701900.91 frames. ], batch size: 284, lr: 7.37e-03, grad_scale: 8.0
+2023-03-02 10:27:50,446 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:28:00,133 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-02 10:28:19,593 INFO [train.py:968] (0/2) Epoch 4, batch 38450, giga_loss[loss=0.3158, simple_loss=0.3834, pruned_loss=0.1241, over 29055.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3836, pruned_loss=0.1293, over 5697663.46 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4031, pruned_loss=0.1538, over 5670332.59 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3805, pruned_loss=0.1255, over 5698133.85 frames. ], batch size: 128, lr: 7.37e-03, grad_scale: 4.0
+2023-03-02 10:28:53,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.157e+02 1.023e+03 1.280e+03 1.767e+03 5.909e+03, threshold=2.560e+03, percent-clipped=15.0
+2023-03-02 10:29:03,497 INFO [train.py:968] (0/2) Epoch 4, batch 38500, giga_loss[loss=0.2931, simple_loss=0.3629, pruned_loss=0.1117, over 28938.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3813, pruned_loss=0.1279, over 5688900.70 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.4036, pruned_loss=0.1542, over 5658840.06 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3784, pruned_loss=0.1244, over 5699058.63 frames. ], batch size: 136, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:29:30,568 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:29:38,234 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:29:44,331 INFO [train.py:968] (0/2) Epoch 4, batch 38550, giga_loss[loss=0.3034, simple_loss=0.3748, pruned_loss=0.116, over 29041.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3784, pruned_loss=0.1256, over 5690414.08 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.4036, pruned_loss=0.1542, over 5657364.19 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3753, pruned_loss=0.122, over 5701064.85 frames. ], batch size: 136, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:30:04,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2466, 1.4055, 1.0824, 0.7406], device='cuda:0'), covar=tensor([0.0904, 0.0798, 0.0642, 0.0929], device='cuda:0'), in_proj_covar=tensor([0.1357, 0.1116, 0.1118, 0.1202], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 10:30:14,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.953e+02 1.040e+03 1.282e+03 1.892e+03 6.083e+03, threshold=2.564e+03, percent-clipped=13.0
+2023-03-02 10:30:24,431 INFO [train.py:968] (0/2) Epoch 4, batch 38600, giga_loss[loss=0.3415, simple_loss=0.3999, pruned_loss=0.1416, over 28892.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3793, pruned_loss=0.1265, over 5690099.95 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.4044, pruned_loss=0.1548, over 5651885.75 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3753, pruned_loss=0.1221, over 5704367.79 frames. ], batch size: 186, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:30:56,871 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:30:59,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:31:00,591 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:31:06,999 INFO [train.py:968] (0/2) Epoch 4, batch 38650, giga_loss[loss=0.2639, simple_loss=0.3364, pruned_loss=0.09569, over 28578.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3795, pruned_loss=0.1265, over 5694589.47 frames. ], libri_tot_loss[loss=0.3576, simple_loss=0.4049, pruned_loss=0.1552, over 5653870.83 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3755, pruned_loss=0.1223, over 5704566.16 frames. ], batch size: 85, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:31:37,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.564e+02 1.201e+03 1.486e+03 2.348e+03 1.923e+04, threshold=2.972e+03, percent-clipped=20.0
+2023-03-02 10:31:39,263 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 10:31:45,773 INFO [train.py:968] (0/2) Epoch 4, batch 38700, giga_loss[loss=0.2976, simple_loss=0.3683, pruned_loss=0.1134, over 28589.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3783, pruned_loss=0.1254, over 5701459.90 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.404, pruned_loss=0.1546, over 5660830.08 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3752, pruned_loss=0.1216, over 5704064.04 frames. ], batch size: 92, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:31:46,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2384, 1.6914, 1.1583, 1.4297], device='cuda:0'), covar=tensor([0.0724, 0.0269, 0.0343, 0.0818], device='cuda:0'), in_proj_covar=tensor([0.0198, 0.0127, 0.0131, 0.0236], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0057], device='cuda:0')
+2023-03-02 10:31:51,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.09 vs. limit=2.0
+2023-03-02 10:32:23,787 INFO [train.py:968] (0/2) Epoch 4, batch 38750, giga_loss[loss=0.2823, simple_loss=0.36, pruned_loss=0.1023, over 28700.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3777, pruned_loss=0.1241, over 5710235.44 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4033, pruned_loss=0.154, over 5669525.82 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3748, pruned_loss=0.1206, over 5706103.63 frames. ], batch size: 85, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:32:34,248 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:32:48,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:32:50,661 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:32:51,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:32:53,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.017e+02 9.723e+02 1.182e+03 1.739e+03 5.153e+03, threshold=2.364e+03, percent-clipped=6.0
+2023-03-02 10:32:54,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:32:59,004 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-02 10:33:02,852 INFO [train.py:968] (0/2) Epoch 4, batch 38800, giga_loss[loss=0.2762, simple_loss=0.3608, pruned_loss=0.09577, over 28933.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3771, pruned_loss=0.1235, over 5716305.59 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.403, pruned_loss=0.1538, over 5674187.04 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3744, pruned_loss=0.12, over 5710021.67 frames. ], batch size: 174, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:33:15,612 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:33:16,894 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:33:43,237 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4943, 1.4402, 1.3175, 1.7413], device='cuda:0'), covar=tensor([0.2050, 0.1987, 0.1836, 0.2123], device='cuda:0'), in_proj_covar=tensor([0.1084, 0.0847, 0.0970, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 10:33:44,219 INFO [train.py:968] (0/2) Epoch 4, batch 38850, giga_loss[loss=0.3067, simple_loss=0.371, pruned_loss=0.1213, over 28892.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3765, pruned_loss=0.1236, over 5704213.90 frames. ], libri_tot_loss[loss=0.3555, simple_loss=0.4031, pruned_loss=0.1539, over 5668689.25 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3735, pruned_loss=0.12, over 5704890.74 frames. ], batch size: 227, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:33:45,470 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 10:34:15,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.583e+02 1.019e+03 1.365e+03 1.968e+03 7.529e+03, threshold=2.731e+03, percent-clipped=19.0
+2023-03-02 10:34:25,742 INFO [train.py:968] (0/2) Epoch 4, batch 38900, giga_loss[loss=0.2611, simple_loss=0.3396, pruned_loss=0.09124, over 28345.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3731, pruned_loss=0.1217, over 5693065.68 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.4029, pruned_loss=0.1537, over 5663726.32 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3704, pruned_loss=0.1184, over 5698462.63 frames. ], batch size: 71, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:34:26,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8048, 1.5638, 1.2268, 1.3087], device='cuda:0'), covar=tensor([0.0580, 0.0583, 0.0910, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0447, 0.0505, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 10:34:31,447 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:34:32,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:34:33,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:34:40,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:34:51,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2787, 4.0010, 3.9147, 1.6419], device='cuda:0'), covar=tensor([0.0480, 0.0447, 0.0756, 0.2259], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0709, 0.0779, 0.0578], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 10:34:58,455 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:35:05,632 INFO [train.py:968] (0/2) Epoch 4, batch 38950, giga_loss[loss=0.2629, simple_loss=0.3355, pruned_loss=0.09521, over 28854.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3692, pruned_loss=0.1197, over 5700235.72 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.4029, pruned_loss=0.1537, over 5668594.25 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3666, pruned_loss=0.1165, over 5700824.63 frames. ], batch size: 186, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:35:09,952 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:35:21,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3338, 1.3258, 1.4479, 1.3485], device='cuda:0'), covar=tensor([0.1204, 0.1549, 0.1757, 0.1499], device='cuda:0'), in_proj_covar=tensor([0.0437, 0.0751, 0.0634, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 10:35:33,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0415, 2.7609, 1.4993, 1.2593], device='cuda:0'), covar=tensor([0.1128, 0.0506, 0.0816, 0.0978], device='cuda:0'), in_proj_covar=tensor([0.1347, 0.1104, 0.1119, 0.1200], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 10:35:35,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.148e+02 1.040e+03 1.406e+03 1.914e+03 6.498e+03, threshold=2.811e+03, percent-clipped=6.0
+2023-03-02 10:35:45,262 INFO [train.py:968] (0/2) Epoch 4, batch 39000, giga_loss[loss=0.298, simple_loss=0.3694, pruned_loss=0.1133, over 28749.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3691, pruned_loss=0.12, over 5707945.00 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4033, pruned_loss=0.1541, over 5670914.41 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3663, pruned_loss=0.1168, over 5706927.61 frames. ], batch size: 71, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:35:45,267 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 10:35:53,163 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2519, 1.3677, 1.0386, 1.4410], device='cuda:0'), covar=tensor([0.0828, 0.0296, 0.0387, 0.0920], device='cuda:0'), in_proj_covar=tensor([0.0199, 0.0128, 0.0131, 0.0237], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:0')
+2023-03-02 10:35:53,792 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2511, simple_loss=0.352, pruned_loss=0.0751, over 944034.00 frames. 
+2023-03-02 10:35:53,792 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 10:36:03,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:36:34,417 INFO [train.py:968] (0/2) Epoch 4, batch 39050, libri_loss[loss=0.2889, simple_loss=0.3471, pruned_loss=0.1154, over 29368.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3701, pruned_loss=0.1218, over 5703343.07 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.4036, pruned_loss=0.1543, over 5667165.20 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3665, pruned_loss=0.1181, over 5707484.47 frames. ], batch size: 67, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:36:35,431 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:36:37,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:36:43,111 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:36:45,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:37:00,878 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:37:04,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.146e+02 1.059e+03 1.443e+03 1.857e+03 6.671e+03, threshold=2.885e+03, percent-clipped=8.0
+2023-03-02 10:37:07,466 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:37:14,676 INFO [train.py:968] (0/2) Epoch 4, batch 39100, giga_loss[loss=0.2153, simple_loss=0.2908, pruned_loss=0.06984, over 28385.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3662, pruned_loss=0.1195, over 5711208.61 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.4036, pruned_loss=0.1543, over 5671877.12 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3626, pruned_loss=0.1158, over 5710983.49 frames. ], batch size: 65, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:37:45,482 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2000, 3.9869, 3.8717, 2.0398], device='cuda:0'), covar=tensor([0.0446, 0.0422, 0.0736, 0.1931], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0708, 0.0780, 0.0577], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 10:37:53,931 INFO [train.py:968] (0/2) Epoch 4, batch 39150, giga_loss[loss=0.278, simple_loss=0.3452, pruned_loss=0.1054, over 28890.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3639, pruned_loss=0.1185, over 5708112.36 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4038, pruned_loss=0.1544, over 5672657.72 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3605, pruned_loss=0.1152, over 5707660.13 frames. ], batch size: 213, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:37:59,187 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:38:02,135 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:38:24,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.744e+02 9.801e+02 1.280e+03 2.070e+03 9.271e+03, threshold=2.560e+03, percent-clipped=14.0
+2023-03-02 10:38:24,442 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:38:35,049 INFO [train.py:968] (0/2) Epoch 4, batch 39200, giga_loss[loss=0.2866, simple_loss=0.3429, pruned_loss=0.1151, over 29028.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.361, pruned_loss=0.1174, over 5712583.77 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4034, pruned_loss=0.1541, over 5677715.51 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3579, pruned_loss=0.1143, over 5708319.24 frames. ], batch size: 128, lr: 7.36e-03, grad_scale: 8.0
+2023-03-02 10:38:41,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0981, 1.8863, 1.4088, 1.6958], device='cuda:0'), covar=tensor([0.0635, 0.0712, 0.0966, 0.1024], device='cuda:0'), in_proj_covar=tensor([0.0333, 0.0445, 0.0503, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 10:39:18,135 INFO [train.py:968] (0/2) Epoch 4, batch 39250, giga_loss[loss=0.2663, simple_loss=0.3322, pruned_loss=0.1002, over 29032.00 frames. ], tot_loss[loss=0.297, simple_loss=0.36, pruned_loss=0.117, over 5711706.51 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.4037, pruned_loss=0.1542, over 5682678.78 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3563, pruned_loss=0.1136, over 5704539.37 frames. ], batch size: 136, lr: 7.35e-03, grad_scale: 8.0
+2023-03-02 10:39:50,194 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.324e+02 1.123e+03 1.333e+03 1.811e+03 4.881e+03, threshold=2.666e+03, percent-clipped=8.0
+2023-03-02 10:40:02,986 INFO [train.py:968] (0/2) Epoch 4, batch 39300, libri_loss[loss=0.3938, simple_loss=0.4354, pruned_loss=0.1761, over 27644.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3635, pruned_loss=0.1189, over 5704015.88 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.4044, pruned_loss=0.1547, over 5676549.99 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.359, pruned_loss=0.1149, over 5705185.68 frames. ], batch size: 116, lr: 7.35e-03, grad_scale: 8.0
+2023-03-02 10:40:27,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:40:39,129 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-02 10:40:46,779 INFO [train.py:968] (0/2) Epoch 4, batch 39350, giga_loss[loss=0.2797, simple_loss=0.3489, pruned_loss=0.1052, over 28729.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3667, pruned_loss=0.1201, over 5711062.64 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.4044, pruned_loss=0.1547, over 5681486.25 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.362, pruned_loss=0.116, over 5708563.50 frames. ], batch size: 119, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:41:24,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.665e+02 9.911e+02 1.236e+03 1.691e+03 5.746e+03, threshold=2.472e+03, percent-clipped=8.0
+2023-03-02 10:41:24,634 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:41:34,124 INFO [train.py:968] (0/2) Epoch 4, batch 39400, giga_loss[loss=0.3031, simple_loss=0.3759, pruned_loss=0.1152, over 28680.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.368, pruned_loss=0.1199, over 5704002.43 frames. ], libri_tot_loss[loss=0.3568, simple_loss=0.4043, pruned_loss=0.1547, over 5682736.82 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3643, pruned_loss=0.1165, over 5701013.73 frames. ], batch size: 284, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:41:38,885 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 10:41:59,420 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8182, 1.5935, 1.3258, 1.4785], device='cuda:0'), covar=tensor([0.0474, 0.0463, 0.0812, 0.0697], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0445, 0.0506, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 10:42:06,066 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2956, 1.7059, 1.5199, 1.4846], device='cuda:0'), covar=tensor([0.1342, 0.1682, 0.1115, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0746, 0.0758, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 10:42:13,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3180, 1.4955, 1.1197, 0.8157], device='cuda:0'), covar=tensor([0.1042, 0.0784, 0.0561, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.1330, 0.1084, 0.1097, 0.1166], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 10:42:16,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0888, 1.1416, 4.9267, 3.4080], device='cuda:0'), covar=tensor([0.1661, 0.2263, 0.0262, 0.0585], device='cuda:0'), in_proj_covar=tensor([0.0545, 0.0513, 0.0712, 0.0573], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 10:42:19,033 INFO [train.py:968] (0/2) Epoch 4, batch 39450, giga_loss[loss=0.257, simple_loss=0.3408, pruned_loss=0.08661, over 29027.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3691, pruned_loss=0.1194, over 5704096.54 frames. ], libri_tot_loss[loss=0.3577, simple_loss=0.405, pruned_loss=0.1552, over 5685809.12 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3646, pruned_loss=0.1154, over 5699342.77 frames. ], batch size: 128, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:42:20,703 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-02 10:42:24,146 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:42:37,096 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:42:39,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:42:50,779 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-02 10:42:50,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.331e+02 1.043e+03 1.198e+03 1.603e+03 5.283e+03, threshold=2.396e+03, percent-clipped=4.0
+2023-03-02 10:43:01,169 INFO [train.py:968] (0/2) Epoch 4, batch 39500, giga_loss[loss=0.2799, simple_loss=0.3623, pruned_loss=0.09872, over 28669.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3685, pruned_loss=0.1187, over 5696131.60 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.4043, pruned_loss=0.1548, over 5690959.88 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3647, pruned_loss=0.115, over 5688055.45 frames. ], batch size: 262, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:43:04,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:43:42,579 INFO [train.py:968] (0/2) Epoch 4, batch 39550, giga_loss[loss=0.3177, simple_loss=0.3736, pruned_loss=0.1309, over 28734.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3676, pruned_loss=0.1176, over 5704628.15 frames. ], libri_tot_loss[loss=0.3572, simple_loss=0.4045, pruned_loss=0.1549, over 5694155.29 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3639, pruned_loss=0.1141, over 5695532.57 frames. ], batch size: 99, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:43:58,130 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-02 10:44:15,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.357e+02 1.165e+03 1.475e+03 1.868e+03 3.837e+03, threshold=2.950e+03, percent-clipped=14.0
+2023-03-02 10:44:23,829 INFO [train.py:968] (0/2) Epoch 4, batch 39600, giga_loss[loss=0.2928, simple_loss=0.3575, pruned_loss=0.114, over 29007.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3681, pruned_loss=0.1182, over 5702178.07 frames. ], libri_tot_loss[loss=0.3571, simple_loss=0.4045, pruned_loss=0.1549, over 5692673.63 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3649, pruned_loss=0.1152, over 5696220.09 frames. ], batch size: 128, lr: 7.35e-03, grad_scale: 8.0
+2023-03-02 10:44:37,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5973, 4.1608, 1.7129, 1.5886], device='cuda:0'), covar=tensor([0.0818, 0.0324, 0.0812, 0.1221], device='cuda:0'), in_proj_covar=tensor([0.0331, 0.0460, 0.0302, 0.0373], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 10:44:43,875 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:45:03,013 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:45:07,479 INFO [train.py:968] (0/2) Epoch 4, batch 39650, giga_loss[loss=0.298, simple_loss=0.3761, pruned_loss=0.11, over 28985.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3718, pruned_loss=0.121, over 5701127.59 frames. ], libri_tot_loss[loss=0.3574, simple_loss=0.4048, pruned_loss=0.155, over 5698138.45 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3681, pruned_loss=0.1176, over 5691457.19 frames. ], batch size: 164, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:45:17,673 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-02 10:45:41,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.378e+02 1.132e+03 1.407e+03 2.003e+03 5.589e+03, threshold=2.813e+03, percent-clipped=11.0
+2023-03-02 10:45:49,919 INFO [train.py:968] (0/2) Epoch 4, batch 39700, libri_loss[loss=0.3373, simple_loss=0.3831, pruned_loss=0.1457, over 27671.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3753, pruned_loss=0.1239, over 5707533.14 frames. ], libri_tot_loss[loss=0.3582, simple_loss=0.4054, pruned_loss=0.1555, over 5701188.01 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.1199, over 5697113.62 frames. ], batch size: 61, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:46:15,502 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:46:20,484 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 10:46:30,783 INFO [train.py:968] (0/2) Epoch 4, batch 39750, giga_loss[loss=0.3326, simple_loss=0.3927, pruned_loss=0.1362, over 28885.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3766, pruned_loss=0.1237, over 5717974.48 frames. ], libri_tot_loss[loss=0.3584, simple_loss=0.4057, pruned_loss=0.1555, over 5704050.16 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3726, pruned_loss=0.1201, over 5707171.77 frames. ], batch size: 199, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:46:35,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0980, 1.8238, 1.4236, 1.5521], device='cuda:0'), covar=tensor([0.0578, 0.0649, 0.0984, 0.0939], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0449, 0.0508, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 10:46:41,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:47:03,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.344e+02 1.307e+03 1.617e+03 2.244e+03 5.509e+03, threshold=3.235e+03, percent-clipped=10.0
+2023-03-02 10:47:13,324 INFO [train.py:968] (0/2) Epoch 4, batch 39800, giga_loss[loss=0.2971, simple_loss=0.3602, pruned_loss=0.117, over 28767.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3771, pruned_loss=0.124, over 5718136.99 frames. ], libri_tot_loss[loss=0.3584, simple_loss=0.4058, pruned_loss=0.1556, over 5706119.38 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3736, pruned_loss=0.1209, over 5708065.66 frames. ], batch size: 99, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:47:41,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:47:53,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3859, 1.6504, 1.1996, 1.0034], device='cuda:0'), covar=tensor([0.1366, 0.0883, 0.0725, 0.1073], device='cuda:0'), in_proj_covar=tensor([0.1372, 0.1136, 0.1143, 0.1206], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 10:47:55,589 INFO [train.py:968] (0/2) Epoch 4, batch 39850, giga_loss[loss=0.3062, simple_loss=0.3709, pruned_loss=0.1207, over 28424.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.378, pruned_loss=0.1248, over 5715777.55 frames. ], libri_tot_loss[loss=0.3594, simple_loss=0.4067, pruned_loss=0.156, over 5705695.94 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3739, pruned_loss=0.1213, over 5708608.50 frames. ], batch size: 71, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:48:00,163 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-176000.pt
+2023-03-02 10:48:22,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-02 10:48:27,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.752e+02 1.209e+03 1.588e+03 2.265e+03 8.179e+03, threshold=3.175e+03, percent-clipped=6.0
+2023-03-02 10:48:33,706 INFO [train.py:968] (0/2) Epoch 4, batch 39900, giga_loss[loss=0.2588, simple_loss=0.3318, pruned_loss=0.09289, over 28862.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3789, pruned_loss=0.1255, over 5719717.98 frames. ], libri_tot_loss[loss=0.361, simple_loss=0.408, pruned_loss=0.157, over 5708115.21 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3737, pruned_loss=0.121, over 5711988.86 frames. ], batch size: 86, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:48:39,901 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:48:42,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:03,750 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:11,654 INFO [train.py:968] (0/2) Epoch 4, batch 39950, giga_loss[loss=0.2831, simple_loss=0.3555, pruned_loss=0.1054, over 28913.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3769, pruned_loss=0.1246, over 5720919.03 frames. ], libri_tot_loss[loss=0.3609, simple_loss=0.4079, pruned_loss=0.157, over 5709566.91 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.372, pruned_loss=0.1202, over 5713581.47 frames. ], batch size: 199, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:49:20,731 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:33,742 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:35,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:45,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.087e+02 1.096e+03 1.389e+03 2.146e+03 1.050e+04, threshold=2.778e+03, percent-clipped=13.0
+2023-03-02 10:49:48,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:51,555 INFO [train.py:968] (0/2) Epoch 4, batch 40000, giga_loss[loss=0.2979, simple_loss=0.3631, pruned_loss=0.1164, over 29013.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.373, pruned_loss=0.1221, over 5721627.12 frames. ], libri_tot_loss[loss=0.3614, simple_loss=0.4083, pruned_loss=0.1572, over 5714585.57 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3679, pruned_loss=0.1176, over 5711305.70 frames. ], batch size: 213, lr: 7.34e-03, grad_scale: 8.0
+2023-03-02 10:49:57,102 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:50:06,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:50:30,953 INFO [train.py:968] (0/2) Epoch 4, batch 40050, giga_loss[loss=0.3032, simple_loss=0.3798, pruned_loss=0.1133, over 28824.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3695, pruned_loss=0.12, over 5725143.29 frames. ], libri_tot_loss[loss=0.3612, simple_loss=0.4082, pruned_loss=0.1571, over 5718959.83 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3646, pruned_loss=0.1155, over 5713182.18 frames. ], batch size: 186, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:51:03,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.433e+02 1.083e+03 1.483e+03 2.148e+03 6.335e+03, threshold=2.966e+03, percent-clipped=14.0
+2023-03-02 10:51:10,402 INFO [train.py:968] (0/2) Epoch 4, batch 40100, giga_loss[loss=0.2905, simple_loss=0.38, pruned_loss=0.1005, over 28587.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3702, pruned_loss=0.1191, over 5720448.10 frames. ], libri_tot_loss[loss=0.3613, simple_loss=0.4084, pruned_loss=0.1571, over 5719128.02 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3651, pruned_loss=0.1146, over 5710876.36 frames. ], batch size: 307, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:51:15,610 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:51:40,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:51:46,395 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:51:49,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:51:55,622 INFO [train.py:968] (0/2) Epoch 4, batch 40150, giga_loss[loss=0.2788, simple_loss=0.3517, pruned_loss=0.103, over 28839.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3712, pruned_loss=0.1187, over 5714691.04 frames. ], libri_tot_loss[loss=0.361, simple_loss=0.408, pruned_loss=0.157, over 5721097.99 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3673, pruned_loss=0.115, over 5705418.31 frames. ], batch size: 199, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:52:00,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3620, 1.8610, 1.3542, 1.5437], device='cuda:0'), covar=tensor([0.0676, 0.0262, 0.0315, 0.0698], device='cuda:0'), in_proj_covar=tensor([0.0198, 0.0129, 0.0131, 0.0236], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:0')
+2023-03-02 10:52:07,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:52:09,906 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:52:15,348 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:52:29,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.146e+02 1.072e+03 1.242e+03 1.693e+03 3.669e+03, threshold=2.484e+03, percent-clipped=7.0
+2023-03-02 10:52:32,617 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:52:33,194 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:52:36,591 INFO [train.py:968] (0/2) Epoch 4, batch 40200, giga_loss[loss=0.2696, simple_loss=0.3337, pruned_loss=0.1027, over 28615.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3706, pruned_loss=0.1188, over 5718014.68 frames. ], libri_tot_loss[loss=0.3606, simple_loss=0.4077, pruned_loss=0.1567, over 5724647.45 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.367, pruned_loss=0.1153, over 5707266.92 frames. ], batch size: 92, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:53:16,795 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:53:17,179 INFO [train.py:968] (0/2) Epoch 4, batch 40250, giga_loss[loss=0.3066, simple_loss=0.3687, pruned_loss=0.1223, over 28610.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3697, pruned_loss=0.1197, over 5722748.60 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4076, pruned_loss=0.1565, over 5728152.95 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3661, pruned_loss=0.1162, over 5710818.75 frames. ], batch size: 242, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:53:18,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:53:30,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-02 10:53:41,202 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:53:50,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.351e+02 1.133e+03 1.463e+03 1.830e+03 9.785e+03, threshold=2.926e+03, percent-clipped=13.0
+2023-03-02 10:53:58,840 INFO [train.py:968] (0/2) Epoch 4, batch 40300, giga_loss[loss=0.3388, simple_loss=0.3864, pruned_loss=0.1456, over 27690.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3679, pruned_loss=0.1206, over 5727566.37 frames. ], libri_tot_loss[loss=0.3599, simple_loss=0.4072, pruned_loss=0.1563, over 5731586.72 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3647, pruned_loss=0.1174, over 5715090.97 frames. ], batch size: 472, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:54:32,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:54:43,309 INFO [train.py:968] (0/2) Epoch 4, batch 40350, giga_loss[loss=0.2891, simple_loss=0.3661, pruned_loss=0.1061, over 28925.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3664, pruned_loss=0.121, over 5722009.41 frames. ], libri_tot_loss[loss=0.36, simple_loss=0.4074, pruned_loss=0.1563, over 5734032.29 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3631, pruned_loss=0.118, over 5709620.28 frames. ], batch size: 174, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:55:17,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.148e+02 9.771e+02 1.150e+03 1.581e+03 2.857e+03, threshold=2.301e+03, percent-clipped=0.0
+2023-03-02 10:55:24,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1667, 3.9780, 3.8386, 1.8378], device='cuda:0'), covar=tensor([0.0439, 0.0409, 0.0796, 0.2058], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0702, 0.0789, 0.0581], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 10:55:27,446 INFO [train.py:968] (0/2) Epoch 4, batch 40400, giga_loss[loss=0.2425, simple_loss=0.3149, pruned_loss=0.08502, over 28589.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.366, pruned_loss=0.1211, over 5705843.72 frames. ], libri_tot_loss[loss=0.3601, simple_loss=0.4075, pruned_loss=0.1564, over 5726519.03 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.363, pruned_loss=0.1183, over 5702548.93 frames. ], batch size: 60, lr: 7.33e-03, grad_scale: 8.0
+2023-03-02 10:56:08,824 INFO [train.py:968] (0/2) Epoch 4, batch 40450, libri_loss[loss=0.3527, simple_loss=0.3943, pruned_loss=0.1556, over 29348.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3642, pruned_loss=0.1207, over 5708338.40 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.4079, pruned_loss=0.1568, over 5726150.67 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3606, pruned_loss=0.1175, over 5705354.36 frames. ], batch size: 71, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:56:32,878 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:56:35,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:56:42,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.525e+02 1.122e+03 1.476e+03 2.090e+03 4.957e+03, threshold=2.952e+03, percent-clipped=17.0
+2023-03-02 10:56:42,762 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:56:50,172 INFO [train.py:968] (0/2) Epoch 4, batch 40500, giga_loss[loss=0.3576, simple_loss=0.3956, pruned_loss=0.1598, over 26682.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3585, pruned_loss=0.1175, over 5708391.81 frames. ], libri_tot_loss[loss=0.361, simple_loss=0.4082, pruned_loss=0.1569, over 5727931.71 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.355, pruned_loss=0.1144, over 5704211.06 frames. ], batch size: 555, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:56:54,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:56:55,067 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:56:59,115 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:57:16,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9217, 1.7473, 1.3944, 1.5237], device='cuda:0'), covar=tensor([0.0602, 0.0633, 0.0950, 0.0900], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0456, 0.0507, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 10:57:31,864 INFO [train.py:968] (0/2) Epoch 4, batch 40550, giga_loss[loss=0.3179, simple_loss=0.3799, pruned_loss=0.1279, over 28680.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3551, pruned_loss=0.1153, over 5714068.23 frames. ], libri_tot_loss[loss=0.3615, simple_loss=0.4085, pruned_loss=0.1573, over 5730274.56 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3513, pruned_loss=0.112, over 5708413.56 frames. ], batch size: 307, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:57:48,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:58:07,003 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.354e+02 1.194e+03 1.534e+03 2.146e+03 7.861e+03, threshold=3.067e+03, percent-clipped=15.0
+2023-03-02 10:58:14,178 INFO [train.py:968] (0/2) Epoch 4, batch 40600, giga_loss[loss=0.2867, simple_loss=0.3608, pruned_loss=0.1062, over 28639.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3562, pruned_loss=0.1153, over 5721624.89 frames. ], libri_tot_loss[loss=0.3606, simple_loss=0.4078, pruned_loss=0.1567, over 5736690.73 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.352, pruned_loss=0.1118, over 5711175.70 frames. ], batch size: 262, lr: 7.33e-03, grad_scale: 2.0
+2023-03-02 10:58:50,746 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-02 10:58:54,073 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:58:54,361 INFO [train.py:968] (0/2) Epoch 4, batch 40650, giga_loss[loss=0.325, simple_loss=0.3791, pruned_loss=0.1355, over 28900.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3616, pruned_loss=0.118, over 5707961.82 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.408, pruned_loss=0.1567, over 5726433.61 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3569, pruned_loss=0.1142, over 5707668.39 frames. ], batch size: 112, lr: 7.33e-03, grad_scale: 2.0
+2023-03-02 10:58:56,051 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:59:19,981 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:59:28,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.615e+02 1.127e+03 1.396e+03 2.170e+03 5.185e+03, threshold=2.792e+03, percent-clipped=9.0
+2023-03-02 10:59:33,734 INFO [train.py:968] (0/2) Epoch 4, batch 40700, giga_loss[loss=0.2883, simple_loss=0.3653, pruned_loss=0.1056, over 28747.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3645, pruned_loss=0.119, over 5708857.75 frames. ], libri_tot_loss[loss=0.3601, simple_loss=0.4076, pruned_loss=0.1563, over 5723593.41 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3597, pruned_loss=0.1152, over 5710442.79 frames. ], batch size: 284, lr: 7.32e-03, grad_scale: 2.0
+2023-03-02 10:59:45,607 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:59:47,985 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:59:48,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:59:53,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4219, 3.2733, 1.5090, 1.4445], device='cuda:0'), covar=tensor([0.0820, 0.0301, 0.0839, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0472, 0.0310, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 11:00:09,752 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2519, 1.7499, 1.3524, 1.4556], device='cuda:0'), covar=tensor([0.0803, 0.0310, 0.0341, 0.0867], device='cuda:0'), in_proj_covar=tensor([0.0199, 0.0130, 0.0133, 0.0238], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0039, 0.0035, 0.0058], device='cuda:0')
+2023-03-02 11:00:13,534 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:00:15,447 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-02 11:00:17,546 INFO [train.py:968] (0/2) Epoch 4, batch 40750, giga_loss[loss=0.3022, simple_loss=0.3744, pruned_loss=0.115, over 28515.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.369, pruned_loss=0.1211, over 5702108.19 frames. ], libri_tot_loss[loss=0.3602, simple_loss=0.4078, pruned_loss=0.1563, over 5726960.96 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3642, pruned_loss=0.1173, over 5700038.70 frames. ], batch size: 71, lr: 7.32e-03, grad_scale: 2.0
+2023-03-02 11:00:38,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9955, 1.3396, 1.0773, 0.2567], device='cuda:0'), covar=tensor([0.1519, 0.1299, 0.2386, 0.2516], device='cuda:0'), in_proj_covar=tensor([0.1356, 0.1255, 0.1356, 0.1124], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 11:00:56,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.083e+02 1.088e+03 1.352e+03 1.838e+03 4.386e+03, threshold=2.704e+03, percent-clipped=5.0
+2023-03-02 11:01:01,506 INFO [train.py:968] (0/2) Epoch 4, batch 40800, giga_loss[loss=0.3087, simple_loss=0.3744, pruned_loss=0.1215, over 28894.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3723, pruned_loss=0.1227, over 5711076.58 frames. ], libri_tot_loss[loss=0.3602, simple_loss=0.4078, pruned_loss=0.1563, over 5728817.75 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.368, pruned_loss=0.1192, over 5707325.93 frames. ], batch size: 186, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:01:48,723 INFO [train.py:968] (0/2) Epoch 4, batch 40850, giga_loss[loss=0.3149, simple_loss=0.383, pruned_loss=0.1234, over 28845.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3777, pruned_loss=0.1275, over 5701992.77 frames. ], libri_tot_loss[loss=0.3601, simple_loss=0.4077, pruned_loss=0.1562, over 5726793.41 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3739, pruned_loss=0.1244, over 5700285.04 frames. ], batch size: 145, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:02:08,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:02:16,945 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:02:22,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:02:33,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.258e+02 1.335e+03 1.978e+03 2.543e+03 6.677e+03, threshold=3.955e+03, percent-clipped=18.0
+2023-03-02 11:02:40,016 INFO [train.py:968] (0/2) Epoch 4, batch 40900, giga_loss[loss=0.341, simple_loss=0.3972, pruned_loss=0.1424, over 28929.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3843, pruned_loss=0.1336, over 5706435.93 frames. ], libri_tot_loss[loss=0.3596, simple_loss=0.4074, pruned_loss=0.1559, over 5729737.51 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3812, pruned_loss=0.131, over 5702045.25 frames. ], batch size: 164, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:02:56,067 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:03:27,494 INFO [train.py:968] (0/2) Epoch 4, batch 40950, giga_loss[loss=0.3885, simple_loss=0.444, pruned_loss=0.1665, over 29019.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3923, pruned_loss=0.14, over 5695207.47 frames. ], libri_tot_loss[loss=0.3589, simple_loss=0.4068, pruned_loss=0.1555, over 5730509.59 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3899, pruned_loss=0.1378, over 5690290.40 frames. ], batch size: 155, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:03:28,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3251, 1.8464, 1.5132, 1.5090], device='cuda:0'), covar=tensor([0.0777, 0.0293, 0.0302, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0197, 0.0128, 0.0131, 0.0237], device='cuda:0'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:0')
+2023-03-02 11:03:37,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5513, 4.3676, 4.2454, 1.6044], device='cuda:0'), covar=tensor([0.0459, 0.0476, 0.0838, 0.2229], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0710, 0.0794, 0.0588], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 11:04:07,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.210e+02 1.760e+03 2.174e+03 3.023e+03 5.658e+03, threshold=4.347e+03, percent-clipped=9.0
+2023-03-02 11:04:12,802 INFO [train.py:968] (0/2) Epoch 4, batch 41000, giga_loss[loss=0.3824, simple_loss=0.435, pruned_loss=0.1649, over 28656.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3987, pruned_loss=0.1455, over 5696223.06 frames. ], libri_tot_loss[loss=0.3594, simple_loss=0.4073, pruned_loss=0.1558, over 5731313.93 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3961, pruned_loss=0.1433, over 5690681.39 frames. ], batch size: 307, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:04:20,226 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:04:22,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:04:35,303 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:04:37,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:04:48,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:04:55,204 INFO [train.py:968] (0/2) Epoch 4, batch 41050, giga_loss[loss=0.4321, simple_loss=0.458, pruned_loss=0.2031, over 28916.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4062, pruned_loss=0.1528, over 5686549.03 frames. ], libri_tot_loss[loss=0.3602, simple_loss=0.4077, pruned_loss=0.1563, over 5720169.59 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4035, pruned_loss=0.1503, over 5691445.58 frames. ], batch size: 112, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:05:03,546 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:05:03,741 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 11:05:14,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2570, 2.3988, 1.2777, 1.2713], device='cuda:0'), covar=tensor([0.0867, 0.0385, 0.0834, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0471, 0.0310, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 11:05:36,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:05:39,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.649e+03 2.280e+03 3.423e+03 6.406e+03, threshold=4.560e+03, percent-clipped=10.0
+2023-03-02 11:05:44,891 INFO [train.py:968] (0/2) Epoch 4, batch 41100, giga_loss[loss=0.3823, simple_loss=0.4247, pruned_loss=0.1699, over 28921.00 frames. ], tot_loss[loss=0.3639, simple_loss=0.4118, pruned_loss=0.158, over 5669528.80 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.408, pruned_loss=0.1567, over 5713065.21 frames. ], giga_tot_loss[loss=0.3604, simple_loss=0.4095, pruned_loss=0.1557, over 5678816.11 frames. ], batch size: 227, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:05:50,572 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:05:53,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0085, 1.1264, 4.0168, 3.1519], device='cuda:0'), covar=tensor([0.1593, 0.2183, 0.0330, 0.0992], device='cuda:0'), in_proj_covar=tensor([0.0554, 0.0513, 0.0725, 0.0583], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 11:05:55,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0417, 1.1798, 1.2651, 1.1013], device='cuda:0'), covar=tensor([0.1128, 0.1147, 0.1659, 0.1201], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0749, 0.0645, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 11:06:34,885 INFO [train.py:968] (0/2) Epoch 4, batch 41150, giga_loss[loss=0.2759, simple_loss=0.3433, pruned_loss=0.1042, over 28427.00 frames. ], tot_loss[loss=0.3685, simple_loss=0.4144, pruned_loss=0.1613, over 5658276.47 frames. ], libri_tot_loss[loss=0.3605, simple_loss=0.4079, pruned_loss=0.1566, over 5713696.69 frames. ], giga_tot_loss[loss=0.3661, simple_loss=0.4128, pruned_loss=0.1597, over 5663453.51 frames. ], batch size: 71, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:07:23,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.434e+02 1.924e+03 2.448e+03 3.422e+03 8.703e+03, threshold=4.896e+03, percent-clipped=11.0
+2023-03-02 11:07:32,467 INFO [train.py:968] (0/2) Epoch 4, batch 41200, giga_loss[loss=0.4809, simple_loss=0.4882, pruned_loss=0.2368, over 27858.00 frames. ], tot_loss[loss=0.3717, simple_loss=0.4159, pruned_loss=0.1638, over 5658983.97 frames. ], libri_tot_loss[loss=0.3601, simple_loss=0.4075, pruned_loss=0.1564, over 5715432.13 frames. ], giga_tot_loss[loss=0.3703, simple_loss=0.4151, pruned_loss=0.1628, over 5660665.57 frames. ], batch size: 412, lr: 7.31e-03, grad_scale: 8.0
+2023-03-02 11:08:06,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:08:08,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:08:10,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9030, 1.6065, 1.3707, 1.3386], device='cuda:0'), covar=tensor([0.0541, 0.0630, 0.0915, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0464, 0.0509, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 11:08:21,771 INFO [train.py:968] (0/2) Epoch 4, batch 41250, giga_loss[loss=0.4602, simple_loss=0.4749, pruned_loss=0.2228, over 27978.00 frames. ], tot_loss[loss=0.376, simple_loss=0.4183, pruned_loss=0.1668, over 5632055.36 frames. ], libri_tot_loss[loss=0.3599, simple_loss=0.4075, pruned_loss=0.1561, over 5710614.20 frames. ], giga_tot_loss[loss=0.3755, simple_loss=0.418, pruned_loss=0.1665, over 5634923.46 frames. ], batch size: 412, lr: 7.31e-03, grad_scale: 8.0
+2023-03-02 11:08:22,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:08:36,907 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:08:52,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0522, 3.8359, 3.7221, 1.6284], device='cuda:0'), covar=tensor([0.0514, 0.0551, 0.0843, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.0830, 0.0727, 0.0809, 0.0598], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 11:09:02,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:09:08,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.797e+02 1.574e+03 2.006e+03 2.540e+03 6.856e+03, threshold=4.012e+03, percent-clipped=2.0
+2023-03-02 11:09:08,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1960, 2.9120, 1.2615, 1.2406], device='cuda:0'), covar=tensor([0.1110, 0.0425, 0.0959, 0.1558], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0473, 0.0306, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0024, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 11:09:14,477 INFO [train.py:968] (0/2) Epoch 4, batch 41300, giga_loss[loss=0.3738, simple_loss=0.4185, pruned_loss=0.1645, over 28640.00 frames. ], tot_loss[loss=0.3811, simple_loss=0.4222, pruned_loss=0.17, over 5632598.80 frames. ], libri_tot_loss[loss=0.3601, simple_loss=0.4076, pruned_loss=0.1563, over 5714139.34 frames. ], giga_tot_loss[loss=0.3808, simple_loss=0.422, pruned_loss=0.1698, over 5630464.68 frames. ], batch size: 307, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:09:39,953 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=2.01 vs. limit=2.0
+2023-03-02 11:09:51,117 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-02 11:09:59,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3205, 1.4059, 1.2205, 1.5813], device='cuda:0'), covar=tensor([0.2180, 0.2106, 0.2001, 0.2018], device='cuda:0'), in_proj_covar=tensor([0.1093, 0.0859, 0.0973, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 11:10:10,329 INFO [train.py:968] (0/2) Epoch 4, batch 41350, giga_loss[loss=0.4898, simple_loss=0.4861, pruned_loss=0.2467, over 26448.00 frames. ], tot_loss[loss=0.3847, simple_loss=0.4242, pruned_loss=0.1726, over 5625206.03 frames. ], libri_tot_loss[loss=0.361, simple_loss=0.4084, pruned_loss=0.1568, over 5715682.69 frames. ], giga_tot_loss[loss=0.3839, simple_loss=0.4236, pruned_loss=0.1721, over 5620961.91 frames. ], batch size: 555, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:10:21,064 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:10:22,671 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:10:53,984 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:10:54,282 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.796e+02 1.878e+03 2.399e+03 3.386e+03 9.700e+03, threshold=4.799e+03, percent-clipped=11.0
+2023-03-02 11:10:57,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:11:02,006 INFO [train.py:968] (0/2) Epoch 4, batch 41400, giga_loss[loss=0.3849, simple_loss=0.4217, pruned_loss=0.174, over 28914.00 frames. ], tot_loss[loss=0.3844, simple_loss=0.4233, pruned_loss=0.1727, over 5634113.24 frames. ], libri_tot_loss[loss=0.3612, simple_loss=0.4086, pruned_loss=0.157, over 5718344.38 frames. ], giga_tot_loss[loss=0.3839, simple_loss=0.4228, pruned_loss=0.1725, over 5627085.48 frames. ], batch size: 106, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:11:02,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5419, 1.4747, 1.4788, 1.4856], device='cuda:0'), covar=tensor([0.1023, 0.1587, 0.1534, 0.1355], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0748, 0.0646, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 11:11:25,256 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:11:33,381 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:11:36,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:11:52,534 INFO [train.py:968] (0/2) Epoch 4, batch 41450, giga_loss[loss=0.3876, simple_loss=0.4252, pruned_loss=0.175, over 27541.00 frames. ], tot_loss[loss=0.3788, simple_loss=0.42, pruned_loss=0.1688, over 5651872.34 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4077, pruned_loss=0.1564, over 5721133.16 frames. ], giga_tot_loss[loss=0.3796, simple_loss=0.4206, pruned_loss=0.1693, over 5642907.64 frames. ], batch size: 472, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:12:08,945 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:12:20,935 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:12:36,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.974e+02 1.405e+03 1.783e+03 2.233e+03 5.672e+03, threshold=3.566e+03, percent-clipped=1.0
+2023-03-02 11:12:41,544 INFO [train.py:968] (0/2) Epoch 4, batch 41500, giga_loss[loss=0.3557, simple_loss=0.4007, pruned_loss=0.1553, over 28579.00 frames. ], tot_loss[loss=0.3752, simple_loss=0.4186, pruned_loss=0.1659, over 5654648.32 frames. ], libri_tot_loss[loss=0.3599, simple_loss=0.4076, pruned_loss=0.1561, over 5721569.58 frames. ], giga_tot_loss[loss=0.3767, simple_loss=0.4196, pruned_loss=0.1669, over 5645178.17 frames. ], batch size: 85, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:13:31,296 INFO [train.py:968] (0/2) Epoch 4, batch 41550, giga_loss[loss=0.3431, simple_loss=0.4039, pruned_loss=0.1411, over 28634.00 frames. ], tot_loss[loss=0.3756, simple_loss=0.4196, pruned_loss=0.1658, over 5668600.63 frames. ], libri_tot_loss[loss=0.3595, simple_loss=0.4073, pruned_loss=0.1559, over 5724804.26 frames. ], giga_tot_loss[loss=0.3775, simple_loss=0.421, pruned_loss=0.167, over 5656841.10 frames. ], batch size: 242, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:14:15,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.631e+03 2.170e+03 3.060e+03 8.499e+03, threshold=4.339e+03, percent-clipped=20.0
+2023-03-02 11:14:21,767 INFO [train.py:968] (0/2) Epoch 4, batch 41600, giga_loss[loss=0.4701, simple_loss=0.4739, pruned_loss=0.2332, over 26521.00 frames. ], tot_loss[loss=0.3746, simple_loss=0.4187, pruned_loss=0.1652, over 5657504.20 frames. ], libri_tot_loss[loss=0.3593, simple_loss=0.407, pruned_loss=0.1557, over 5730917.90 frames. ], giga_tot_loss[loss=0.3769, simple_loss=0.4204, pruned_loss=0.1667, over 5640374.49 frames. ], batch size: 555, lr: 7.31e-03, grad_scale: 8.0
+2023-03-02 11:14:40,549 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:14:42,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:14:57,488 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0606, 4.7986, 4.7061, 2.0770], device='cuda:0'), covar=tensor([0.0320, 0.0382, 0.0632, 0.1894], device='cuda:0'), in_proj_covar=tensor([0.0826, 0.0726, 0.0802, 0.0587], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 11:15:07,409 INFO [train.py:968] (0/2) Epoch 4, batch 41650, libri_loss[loss=0.3178, simple_loss=0.37, pruned_loss=0.1328, over 29590.00 frames. ], tot_loss[loss=0.3701, simple_loss=0.4165, pruned_loss=0.1619, over 5655144.78 frames. ], libri_tot_loss[loss=0.3597, simple_loss=0.4074, pruned_loss=0.156, over 5728267.95 frames. ], giga_tot_loss[loss=0.3721, simple_loss=0.4179, pruned_loss=0.1631, over 5640641.57 frames. ], batch size: 75, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:15:11,775 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:15:46,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5344, 2.9359, 1.6552, 1.5984], device='cuda:0'), covar=tensor([0.0687, 0.0343, 0.0628, 0.0977], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0470, 0.0308, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 11:15:55,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.666e+02 1.663e+03 2.271e+03 3.468e+03 7.715e+03, threshold=4.542e+03, percent-clipped=10.0
+2023-03-02 11:16:01,826 INFO [train.py:968] (0/2) Epoch 4, batch 41700, giga_loss[loss=0.3285, simple_loss=0.3853, pruned_loss=0.1358, over 28878.00 frames. ], tot_loss[loss=0.3641, simple_loss=0.4129, pruned_loss=0.1577, over 5659864.19 frames. ], libri_tot_loss[loss=0.3598, simple_loss=0.4074, pruned_loss=0.156, over 5725187.61 frames. ], giga_tot_loss[loss=0.3656, simple_loss=0.414, pruned_loss=0.1586, over 5650800.27 frames. ], batch size: 112, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:16:29,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3958, 1.6828, 1.1747, 0.9727], device='cuda:0'), covar=tensor([0.1254, 0.0857, 0.0746, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.1378, 0.1123, 0.1135, 0.1197], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 11:16:37,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:16:39,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:16:53,993 INFO [train.py:968] (0/2) Epoch 4, batch 41750, giga_loss[loss=0.353, simple_loss=0.3826, pruned_loss=0.1617, over 23727.00 frames. ], tot_loss[loss=0.3572, simple_loss=0.4078, pruned_loss=0.1533, over 5657638.93 frames. ], libri_tot_loss[loss=0.3587, simple_loss=0.4066, pruned_loss=0.1555, over 5728501.50 frames. ], giga_tot_loss[loss=0.3592, simple_loss=0.4094, pruned_loss=0.1545, over 5646589.41 frames. ], batch size: 705, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:17:40,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.175e+02 1.542e+03 2.008e+03 2.677e+03 6.280e+03, threshold=4.016e+03, percent-clipped=6.0
+2023-03-02 11:17:44,755 INFO [train.py:968] (0/2) Epoch 4, batch 41800, libri_loss[loss=0.3156, simple_loss=0.3774, pruned_loss=0.1269, over 29548.00 frames. ], tot_loss[loss=0.3524, simple_loss=0.4039, pruned_loss=0.1505, over 5648557.51 frames. ], libri_tot_loss[loss=0.3582, simple_loss=0.4061, pruned_loss=0.1551, over 5722017.80 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4056, pruned_loss=0.1517, over 5644010.40 frames. ], batch size: 78, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:17:48,816 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:18:13,190 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-02 11:18:32,747 INFO [train.py:968] (0/2) Epoch 4, batch 41850, giga_loss[loss=0.3593, simple_loss=0.4119, pruned_loss=0.1533, over 28760.00 frames. ], tot_loss[loss=0.3536, simple_loss=0.4047, pruned_loss=0.1513, over 5655861.66 frames. ], libri_tot_loss[loss=0.3581, simple_loss=0.4061, pruned_loss=0.1551, over 5723727.73 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.406, pruned_loss=0.1522, over 5648377.21 frames. ], batch size: 119, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:18:39,869 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-178000.pt
+2023-03-02 11:18:56,711 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:18:58,300 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:18:59,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:19:02,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:19:13,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.921e+02 1.545e+03 2.211e+03 3.127e+03 6.405e+03, threshold=4.422e+03, percent-clipped=11.0
+2023-03-02 11:19:18,544 INFO [train.py:968] (0/2) Epoch 4, batch 41900, giga_loss[loss=0.3153, simple_loss=0.3821, pruned_loss=0.1243, over 29022.00 frames. ], tot_loss[loss=0.3547, simple_loss=0.4056, pruned_loss=0.1519, over 5665720.15 frames. ], libri_tot_loss[loss=0.3582, simple_loss=0.4062, pruned_loss=0.1551, over 5724581.58 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4066, pruned_loss=0.1525, over 5657660.44 frames. ], batch size: 155, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:19:31,815 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:19:33,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:19:35,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:20:06,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7750, 2.5748, 1.7901, 2.3920], device='cuda:0'), covar=tensor([0.0508, 0.0534, 0.0854, 0.0785], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0459, 0.0509, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 11:20:09,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4281, 2.1532, 1.5553, 0.6114], device='cuda:0'), covar=tensor([0.2808, 0.1157, 0.1917, 0.2935], device='cuda:0'), in_proj_covar=tensor([0.1354, 0.1252, 0.1338, 0.1118], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 11:20:14,416 INFO [train.py:968] (0/2) Epoch 4, batch 41950, giga_loss[loss=0.4245, simple_loss=0.4651, pruned_loss=0.1919, over 27947.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4025, pruned_loss=0.1488, over 5667243.51 frames. ], libri_tot_loss[loss=0.3585, simple_loss=0.4063, pruned_loss=0.1553, over 5719616.13 frames. ], giga_tot_loss[loss=0.3505, simple_loss=0.4031, pruned_loss=0.149, over 5664083.69 frames. ], batch size: 412, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:20:19,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8552, 3.1891, 2.0241, 0.8593], device='cuda:0'), covar=tensor([0.3574, 0.1175, 0.1921, 0.3261], device='cuda:0'), in_proj_covar=tensor([0.1352, 0.1253, 0.1339, 0.1119], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 11:20:58,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.446e+02 1.402e+03 2.017e+03 3.105e+03 1.372e+04, threshold=4.033e+03, percent-clipped=12.0
+2023-03-02 11:21:04,632 INFO [train.py:968] (0/2) Epoch 4, batch 42000, giga_loss[loss=0.3774, simple_loss=0.4399, pruned_loss=0.1575, over 28919.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.4024, pruned_loss=0.1461, over 5666952.15 frames. ], libri_tot_loss[loss=0.3578, simple_loss=0.4056, pruned_loss=0.155, over 5712588.14 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4034, pruned_loss=0.1464, over 5670127.35 frames. ], batch size: 174, lr: 7.30e-03, grad_scale: 8.0
+2023-03-02 11:21:04,637 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 11:21:13,475 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2346, simple_loss=0.3352, pruned_loss=0.06698, over 944034.00 frames. 
+2023-03-02 11:21:13,476 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 11:21:45,729 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:22:01,306 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:22:03,604 INFO [train.py:968] (0/2) Epoch 4, batch 42050, giga_loss[loss=0.4827, simple_loss=0.4842, pruned_loss=0.2406, over 26531.00 frames. ], tot_loss[loss=0.3483, simple_loss=0.4041, pruned_loss=0.1462, over 5669062.84 frames. ], libri_tot_loss[loss=0.3575, simple_loss=0.4055, pruned_loss=0.1547, over 5715319.34 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.405, pruned_loss=0.1465, over 5668527.40 frames. ], batch size: 555, lr: 7.30e-03, grad_scale: 8.0
+2023-03-02 11:22:46,934 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-02 11:22:47,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.412e+02 1.617e+03 2.040e+03 2.940e+03 4.651e+03, threshold=4.080e+03, percent-clipped=4.0
+2023-03-02 11:22:50,235 INFO [train.py:968] (0/2) Epoch 4, batch 42100, giga_loss[loss=0.3393, simple_loss=0.3994, pruned_loss=0.1396, over 28898.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4058, pruned_loss=0.1493, over 5672543.50 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.4047, pruned_loss=0.1543, over 5721853.71 frames. ], giga_tot_loss[loss=0.3533, simple_loss=0.4071, pruned_loss=0.1497, over 5664842.26 frames. ], batch size: 174, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:23:34,128 INFO [train.py:968] (0/2) Epoch 4, batch 42150, giga_loss[loss=0.3717, simple_loss=0.3978, pruned_loss=0.1728, over 23499.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4057, pruned_loss=0.1496, over 5673668.18 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.4048, pruned_loss=0.1543, over 5721125.16 frames. ], giga_tot_loss[loss=0.3532, simple_loss=0.4068, pruned_loss=0.1498, over 5665917.73 frames. ], batch size: 705, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:23:58,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:24:04,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3804, 3.1352, 1.4471, 1.3781], device='cuda:0'), covar=tensor([0.0903, 0.0312, 0.0862, 0.1348], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0471, 0.0308, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0024, 0.0015, 0.0020], device='cuda:0')
+2023-03-02 11:24:14,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.949e+02 1.572e+03 2.047e+03 2.990e+03 8.214e+03, threshold=4.094e+03, percent-clipped=13.0
+2023-03-02 11:24:18,018 INFO [train.py:968] (0/2) Epoch 4, batch 42200, giga_loss[loss=0.3322, simple_loss=0.391, pruned_loss=0.1367, over 29014.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.4041, pruned_loss=0.1495, over 5659479.47 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.4049, pruned_loss=0.1545, over 5705011.03 frames. ], giga_tot_loss[loss=0.3518, simple_loss=0.4048, pruned_loss=0.1494, over 5667358.67 frames. ], batch size: 164, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:25:02,706 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5913, 3.3147, 1.6255, 1.6172], device='cuda:0'), covar=tensor([0.0819, 0.0294, 0.0786, 0.1177], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0474, 0.0309, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 11:25:04,617 INFO [train.py:968] (0/2) Epoch 4, batch 42250, giga_loss[loss=0.3315, simple_loss=0.3886, pruned_loss=0.1372, over 27945.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4034, pruned_loss=0.1504, over 5657139.30 frames. ], libri_tot_loss[loss=0.3571, simple_loss=0.405, pruned_loss=0.1545, over 5706584.59 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4039, pruned_loss=0.1502, over 5661366.25 frames. ], batch size: 412, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:25:26,033 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-02 11:25:36,213 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6038, 1.5391, 1.2536, 1.3366], device='cuda:0'), covar=tensor([0.0562, 0.0483, 0.0905, 0.0756], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0458, 0.0508, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 11:25:40,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:25:47,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.816e+02 1.680e+03 2.155e+03 3.197e+03 9.637e+03, threshold=4.310e+03, percent-clipped=11.0
+2023-03-02 11:25:53,435 INFO [train.py:968] (0/2) Epoch 4, batch 42300, giga_loss[loss=0.3304, simple_loss=0.3661, pruned_loss=0.1473, over 23674.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.4024, pruned_loss=0.1491, over 5659948.55 frames. ], libri_tot_loss[loss=0.3568, simple_loss=0.4049, pruned_loss=0.1543, over 5707109.55 frames. ], giga_tot_loss[loss=0.3503, simple_loss=0.4028, pruned_loss=0.1489, over 5661293.14 frames. ], batch size: 705, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:26:00,422 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1975, 1.4375, 1.1398, 1.4397], device='cuda:0'), covar=tensor([0.0822, 0.0335, 0.0358, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0199, 0.0129, 0.0132, 0.0238], device='cuda:0'), out_proj_covar=tensor([0.0053, 0.0039, 0.0035, 0.0058], device='cuda:0')
+2023-03-02 11:26:13,314 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:26:16,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:26:17,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5690, 1.8859, 1.8031, 1.7209], device='cuda:0'), covar=tensor([0.1578, 0.1849, 0.1207, 0.1231], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0757, 0.0759, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 11:26:39,515 INFO [train.py:968] (0/2) Epoch 4, batch 42350, giga_loss[loss=0.3843, simple_loss=0.4276, pruned_loss=0.1706, over 28567.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.4005, pruned_loss=0.1457, over 5660138.76 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4047, pruned_loss=0.1543, over 5690412.33 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.4009, pruned_loss=0.1454, over 5676389.70 frames. ], batch size: 336, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:26:42,601 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:27:26,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.707e+02 1.451e+03 2.096e+03 2.649e+03 4.876e+03, threshold=4.192e+03, percent-clipped=1.0
+2023-03-02 11:27:30,396 INFO [train.py:968] (0/2) Epoch 4, batch 42400, libri_loss[loss=0.3758, simple_loss=0.4109, pruned_loss=0.1704, over 29554.00 frames. ], tot_loss[loss=0.346, simple_loss=0.4011, pruned_loss=0.1454, over 5668987.76 frames. ], libri_tot_loss[loss=0.3576, simple_loss=0.4054, pruned_loss=0.1549, over 5695589.56 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.4006, pruned_loss=0.1444, over 5676307.70 frames. ], batch size: 78, lr: 7.29e-03, grad_scale: 8.0
+2023-03-02 11:27:35,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:27:50,835 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:28:00,216 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:28:03,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:28:19,762 INFO [train.py:968] (0/2) Epoch 4, batch 42450, giga_loss[loss=0.3521, simple_loss=0.4069, pruned_loss=0.1487, over 28716.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.4, pruned_loss=0.1446, over 5674708.48 frames. ], libri_tot_loss[loss=0.3575, simple_loss=0.4054, pruned_loss=0.1548, over 5697374.02 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3996, pruned_loss=0.1439, over 5678768.21 frames. ], batch size: 262, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:28:33,710 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:29:03,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.680e+02 1.701e+03 2.175e+03 2.796e+03 7.112e+03, threshold=4.349e+03, percent-clipped=9.0
+2023-03-02 11:29:06,681 INFO [train.py:968] (0/2) Epoch 4, batch 42500, giga_loss[loss=0.3222, simple_loss=0.3827, pruned_loss=0.1308, over 28774.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3985, pruned_loss=0.1443, over 5674608.99 frames. ], libri_tot_loss[loss=0.3575, simple_loss=0.4054, pruned_loss=0.1547, over 5696458.09 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3981, pruned_loss=0.1436, over 5678271.95 frames. ], batch size: 92, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:29:54,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1586, 1.6675, 1.5314, 1.4598], device='cuda:0'), covar=tensor([0.1359, 0.1743, 0.1075, 0.1075], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0753, 0.0759, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 11:29:56,911 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:29:58,305 INFO [train.py:968] (0/2) Epoch 4, batch 42550, giga_loss[loss=0.3256, simple_loss=0.3838, pruned_loss=0.1337, over 28840.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3978, pruned_loss=0.1449, over 5669145.16 frames. ], libri_tot_loss[loss=0.3573, simple_loss=0.4052, pruned_loss=0.1546, over 5699610.61 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3975, pruned_loss=0.1442, over 5668928.91 frames. ], batch size: 199, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:29:58,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7133, 2.1314, 2.0250, 1.9239], device='cuda:0'), covar=tensor([0.1544, 0.1798, 0.1141, 0.1175], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0754, 0.0760, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 11:29:59,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:30:11,963 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:30:14,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:30:30,961 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:30:46,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.962e+02 1.689e+03 2.249e+03 3.422e+03 7.651e+03, threshold=4.499e+03, percent-clipped=15.0
+2023-03-02 11:30:47,114 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:30:50,021 INFO [train.py:968] (0/2) Epoch 4, batch 42600, giga_loss[loss=0.3434, simple_loss=0.3951, pruned_loss=0.1459, over 27967.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3981, pruned_loss=0.1463, over 5656936.99 frames. ], libri_tot_loss[loss=0.3572, simple_loss=0.4052, pruned_loss=0.1546, over 5691442.79 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.3977, pruned_loss=0.1456, over 5663206.22 frames. ], batch size: 412, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:31:04,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6042, 1.9466, 1.8830, 1.7616], device='cuda:0'), covar=tensor([0.1487, 0.1754, 0.1135, 0.1150], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0754, 0.0760, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 11:31:07,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5793, 1.5658, 0.9414, 1.2418], device='cuda:0'), covar=tensor([0.0871, 0.0829, 0.1724, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0458, 0.0510, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 11:31:37,459 INFO [train.py:968] (0/2) Epoch 4, batch 42650, giga_loss[loss=0.3022, simple_loss=0.365, pruned_loss=0.1197, over 28573.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3964, pruned_loss=0.1455, over 5646748.69 frames. ], libri_tot_loss[loss=0.3575, simple_loss=0.4054, pruned_loss=0.1548, over 5666043.96 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3958, pruned_loss=0.1447, over 5672982.02 frames. ], batch size: 60, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:32:08,910 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 11:32:24,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.580e+02 1.584e+03 2.134e+03 2.589e+03 1.150e+04, threshold=4.267e+03, percent-clipped=3.0
+2023-03-02 11:32:27,823 INFO [train.py:968] (0/2) Epoch 4, batch 42700, giga_loss[loss=0.3694, simple_loss=0.4137, pruned_loss=0.1626, over 28308.00 frames. ], tot_loss[loss=0.345, simple_loss=0.397, pruned_loss=0.1465, over 5657838.82 frames. ], libri_tot_loss[loss=0.3578, simple_loss=0.4056, pruned_loss=0.1549, over 5671161.14 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3962, pruned_loss=0.1455, over 5674039.40 frames. ], batch size: 368, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:32:30,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:32:42,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9232, 1.9047, 1.6092, 1.7370], device='cuda:0'), covar=tensor([0.0936, 0.1394, 0.1422, 0.1259], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0761, 0.0644, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 11:33:03,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5415, 1.4575, 1.4287, 1.3866], device='cuda:0'), covar=tensor([0.1058, 0.1884, 0.1635, 0.1616], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0761, 0.0644, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 11:33:13,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6496, 2.0159, 1.9508, 1.7516], device='cuda:0'), covar=tensor([0.1515, 0.1731, 0.1132, 0.1200], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0756, 0.0758, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 11:33:16,146 INFO [train.py:968] (0/2) Epoch 4, batch 42750, giga_loss[loss=0.3374, simple_loss=0.3936, pruned_loss=0.1406, over 28580.00 frames. ], tot_loss[loss=0.3447, simple_loss=0.3972, pruned_loss=0.1461, over 5670250.40 frames. ], libri_tot_loss[loss=0.3576, simple_loss=0.4055, pruned_loss=0.1549, over 5676918.59 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3964, pruned_loss=0.1452, over 5678053.61 frames. ], batch size: 71, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:33:49,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3857, 3.1884, 1.4177, 1.3502], device='cuda:0'), covar=tensor([0.0862, 0.0336, 0.0857, 0.1284], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0480, 0.0309, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 11:34:01,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.878e+02 1.658e+03 2.077e+03 3.397e+03 1.075e+04, threshold=4.153e+03, percent-clipped=15.0
+2023-03-02 11:34:05,571 INFO [train.py:968] (0/2) Epoch 4, batch 42800, giga_loss[loss=0.3143, simple_loss=0.3872, pruned_loss=0.1207, over 29016.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3975, pruned_loss=0.1455, over 5672309.02 frames. ], libri_tot_loss[loss=0.3572, simple_loss=0.4052, pruned_loss=0.1546, over 5683962.22 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3969, pruned_loss=0.1448, over 5672050.23 frames. ], batch size: 136, lr: 7.28e-03, grad_scale: 8.0
+2023-03-02 11:34:49,461 INFO [train.py:968] (0/2) Epoch 4, batch 42850, giga_loss[loss=0.3316, simple_loss=0.3927, pruned_loss=0.1352, over 28851.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3968, pruned_loss=0.1442, over 5678406.91 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4048, pruned_loss=0.1543, over 5690200.28 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3966, pruned_loss=0.1436, over 5672411.02 frames. ], batch size: 119, lr: 7.28e-03, grad_scale: 8.0
+2023-03-02 11:35:07,480 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:35:32,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.886e+02 1.439e+03 1.860e+03 2.433e+03 4.993e+03, threshold=3.720e+03, percent-clipped=1.0
+2023-03-02 11:35:34,693 INFO [train.py:968] (0/2) Epoch 4, batch 42900, giga_loss[loss=0.3395, simple_loss=0.3984, pruned_loss=0.1403, over 28753.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3976, pruned_loss=0.1444, over 5672424.18 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.405, pruned_loss=0.1544, over 5683625.35 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3969, pruned_loss=0.1435, over 5673583.96 frames. ], batch size: 284, lr: 7.28e-03, grad_scale: 8.0
+2023-03-02 11:36:20,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:36:21,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5755, 3.5747, 1.5954, 1.5584], device='cuda:0'), covar=tensor([0.0878, 0.0338, 0.0815, 0.1295], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0477, 0.0309, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 11:36:23,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179090.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:36:25,584 INFO [train.py:968] (0/2) Epoch 4, batch 42950, giga_loss[loss=0.363, simple_loss=0.4034, pruned_loss=0.1613, over 28554.00 frames. ], tot_loss[loss=0.347, simple_loss=0.3995, pruned_loss=0.1473, over 5651356.15 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.4048, pruned_loss=0.1543, over 5670030.37 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.399, pruned_loss=0.1465, over 5662948.64 frames. ], batch size: 92, lr: 7.28e-03, grad_scale: 8.0
+2023-03-02 11:36:48,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4231, 1.8896, 2.0007, 1.7840], device='cuda:0'), covar=tensor([0.0934, 0.2158, 0.1386, 0.1662], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0750, 0.0637, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 11:37:09,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.773e+03 2.260e+03 2.999e+03 9.016e+03, threshold=4.521e+03, percent-clipped=13.0
+2023-03-02 11:37:10,446 INFO [train.py:968] (0/2) Epoch 4, batch 43000, giga_loss[loss=0.4939, simple_loss=0.4949, pruned_loss=0.2465, over 26584.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.4023, pruned_loss=0.1503, over 5654866.68 frames. ], libri_tot_loss[loss=0.3565, simple_loss=0.4046, pruned_loss=0.1542, over 5672756.28 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.4018, pruned_loss=0.1494, over 5660593.52 frames. ], batch size: 555, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:37:22,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1921, 1.5770, 1.2515, 1.4215], device='cuda:0'), covar=tensor([0.0777, 0.0332, 0.0331, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0198, 0.0128, 0.0132, 0.0237], device='cuda:0'), out_proj_covar=tensor([0.0053, 0.0039, 0.0035, 0.0058], device='cuda:0')
+2023-03-02 11:37:32,183 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:38:05,471 INFO [train.py:968] (0/2) Epoch 4, batch 43050, giga_loss[loss=0.3228, simple_loss=0.3814, pruned_loss=0.1321, over 28903.00 frames. ], tot_loss[loss=0.3535, simple_loss=0.4026, pruned_loss=0.1522, over 5650733.07 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.4043, pruned_loss=0.1539, over 5676180.09 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4025, pruned_loss=0.1517, over 5651741.58 frames. ], batch size: 112, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:38:32,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:38:54,211 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.774e+03 2.374e+03 2.953e+03 7.665e+03, threshold=4.749e+03, percent-clipped=11.0
+2023-03-02 11:38:56,673 INFO [train.py:968] (0/2) Epoch 4, batch 43100, giga_loss[loss=0.3678, simple_loss=0.4112, pruned_loss=0.1622, over 28906.00 frames. ], tot_loss[loss=0.3554, simple_loss=0.4034, pruned_loss=0.1537, over 5654465.15 frames. ], libri_tot_loss[loss=0.3556, simple_loss=0.404, pruned_loss=0.1536, over 5676491.59 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4035, pruned_loss=0.1536, over 5654494.88 frames. ], batch size: 186, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:39:35,704 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179287.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:39:40,155 INFO [train.py:968] (0/2) Epoch 4, batch 43150, giga_loss[loss=0.3299, simple_loss=0.3911, pruned_loss=0.1343, over 28892.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4037, pruned_loss=0.1541, over 5667860.74 frames. ], libri_tot_loss[loss=0.3554, simple_loss=0.4037, pruned_loss=0.1535, over 5685657.87 frames. ], giga_tot_loss[loss=0.3562, simple_loss=0.4041, pruned_loss=0.1541, over 5658861.89 frames. ], batch size: 199, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:40:06,608 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 11:40:27,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.906e+02 1.614e+03 2.021e+03 2.657e+03 1.004e+04, threshold=4.041e+03, percent-clipped=2.0
+2023-03-02 11:40:28,460 INFO [train.py:968] (0/2) Epoch 4, batch 43200, libri_loss[loss=0.388, simple_loss=0.4299, pruned_loss=0.173, over 20245.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.4014, pruned_loss=0.1521, over 5666761.52 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4037, pruned_loss=0.1535, over 5678728.10 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4017, pruned_loss=0.1522, over 5666539.41 frames. ], batch size: 187, lr: 7.27e-03, grad_scale: 8.0
+2023-03-02 11:40:44,782 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:40:46,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:41:04,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:41:08,743 INFO [train.py:968] (0/2) Epoch 4, batch 43250, giga_loss[loss=0.3057, simple_loss=0.3756, pruned_loss=0.1179, over 28829.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.4004, pruned_loss=0.1492, over 5674635.21 frames. ], libri_tot_loss[loss=0.3555, simple_loss=0.404, pruned_loss=0.1535, over 5681326.17 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.4003, pruned_loss=0.1491, over 5671679.97 frames. ], batch size: 99, lr: 7.27e-03, grad_scale: 8.0
+2023-03-02 11:41:12,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:41:57,991 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.487e+02 1.804e+03 2.290e+03 3.117e+03 5.777e+03, threshold=4.579e+03, percent-clipped=9.0
+2023-03-02 11:42:00,260 INFO [train.py:968] (0/2) Epoch 4, batch 43300, giga_loss[loss=0.2943, simple_loss=0.3701, pruned_loss=0.1092, over 28920.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.3964, pruned_loss=0.1459, over 5672343.49 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4043, pruned_loss=0.1537, over 5684636.90 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.396, pruned_loss=0.1456, over 5666982.99 frames. ], batch size: 145, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:42:16,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:42:18,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179465.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:42:44,435 INFO [train.py:968] (0/2) Epoch 4, batch 43350, giga_loss[loss=0.3451, simple_loss=0.3911, pruned_loss=0.1495, over 28268.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3956, pruned_loss=0.146, over 5651672.44 frames. ], libri_tot_loss[loss=0.3568, simple_loss=0.4051, pruned_loss=0.1542, over 5669649.96 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3943, pruned_loss=0.145, over 5661237.82 frames. ], batch size: 368, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:42:44,810 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7998, 1.3491, 3.7174, 2.9980], device='cuda:0'), covar=tensor([0.1751, 0.2071, 0.0408, 0.0632], device='cuda:0'), in_proj_covar=tensor([0.0561, 0.0519, 0.0733, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 11:43:20,250 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:43:23,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:43:27,493 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:43:30,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.493e+03 1.956e+03 2.701e+03 6.233e+03, threshold=3.912e+03, percent-clipped=4.0
+2023-03-02 11:43:31,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4329, 1.7622, 1.2057, 0.8498], device='cuda:0'), covar=tensor([0.1344, 0.0906, 0.0746, 0.1169], device='cuda:0'), in_proj_covar=tensor([0.1367, 0.1133, 0.1148, 0.1211], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 11:43:31,749 INFO [train.py:968] (0/2) Epoch 4, batch 43400, giga_loss[loss=0.3072, simple_loss=0.3752, pruned_loss=0.1196, over 28971.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.394, pruned_loss=0.1454, over 5650994.29 frames. ], libri_tot_loss[loss=0.3568, simple_loss=0.4052, pruned_loss=0.1542, over 5662264.85 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3928, pruned_loss=0.1445, over 5665421.20 frames. ], batch size: 164, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:43:52,476 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:44:02,767 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-02 11:44:19,607 INFO [train.py:968] (0/2) Epoch 4, batch 43450, giga_loss[loss=0.4682, simple_loss=0.4705, pruned_loss=0.2329, over 26647.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3967, pruned_loss=0.148, over 5651641.29 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4051, pruned_loss=0.1541, over 5668376.23 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.3955, pruned_loss=0.1472, over 5657307.08 frames. ], batch size: 555, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:44:28,805 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:44:30,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:44:30,683 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179608.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:44:33,278 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179611.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:44:51,360 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4036, 1.4627, 1.5461, 1.4430], device='cuda:0'), covar=tensor([0.0861, 0.1166, 0.1033, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0751, 0.0640, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 11:44:58,649 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:45:00,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179640.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:45:02,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.600e+03 2.340e+03 3.265e+03 7.190e+03, threshold=4.680e+03, percent-clipped=13.0
+2023-03-02 11:45:03,037 INFO [train.py:968] (0/2) Epoch 4, batch 43500, giga_loss[loss=0.3478, simple_loss=0.4217, pruned_loss=0.137, over 29036.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.4004, pruned_loss=0.1494, over 5665027.12 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.4053, pruned_loss=0.1543, over 5674441.88 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.3991, pruned_loss=0.1484, over 5663975.87 frames. ], batch size: 155, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:45:21,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179662.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:45:43,201 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:45:43,223 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4898, 1.6845, 1.7469, 1.6990], device='cuda:0'), covar=tensor([0.1290, 0.1534, 0.1062, 0.1007], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0753, 0.0764, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 11:45:45,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:45:52,007 INFO [train.py:968] (0/2) Epoch 4, batch 43550, libri_loss[loss=0.3251, simple_loss=0.3874, pruned_loss=0.1314, over 29527.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.403, pruned_loss=0.1486, over 5662838.32 frames. ], libri_tot_loss[loss=0.3564, simple_loss=0.4048, pruned_loss=0.154, over 5676593.78 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4024, pruned_loss=0.1479, over 5659419.11 frames. ], batch size: 82, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:46:10,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:46:41,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.653e+02 1.568e+03 2.007e+03 2.802e+03 8.019e+03, threshold=4.013e+03, percent-clipped=5.0
+2023-03-02 11:46:41,711 INFO [train.py:968] (0/2) Epoch 4, batch 43600, giga_loss[loss=0.3906, simple_loss=0.4364, pruned_loss=0.1724, over 28799.00 frames. ], tot_loss[loss=0.3514, simple_loss=0.4046, pruned_loss=0.1492, over 5657303.59 frames. ], libri_tot_loss[loss=0.3562, simple_loss=0.4045, pruned_loss=0.1539, over 5668867.30 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4043, pruned_loss=0.1486, over 5660684.02 frames. ], batch size: 119, lr: 7.27e-03, grad_scale: 8.0
+2023-03-02 11:46:47,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3526, 1.3723, 1.1942, 1.4932], device='cuda:0'), covar=tensor([0.2012, 0.1978, 0.1888, 0.1875], device='cuda:0'), in_proj_covar=tensor([0.1096, 0.0866, 0.0969, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 11:46:52,483 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:47:10,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-02 11:47:33,070 INFO [train.py:968] (0/2) Epoch 4, batch 43650, giga_loss[loss=0.3402, simple_loss=0.3964, pruned_loss=0.142, over 28352.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.4083, pruned_loss=0.1524, over 5668172.30 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.405, pruned_loss=0.1544, over 5671263.61 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4076, pruned_loss=0.1515, over 5668634.78 frames. ], batch size: 65, lr: 7.26e-03, grad_scale: 8.0
+2023-03-02 11:47:43,712 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179805.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:47:46,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179808.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:48:14,163 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179837.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:48:14,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179837.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:48:20,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.978e+02 1.505e+03 1.825e+03 2.512e+03 5.291e+03, threshold=3.650e+03, percent-clipped=3.0
+2023-03-02 11:48:20,190 INFO [train.py:968] (0/2) Epoch 4, batch 43700, giga_loss[loss=0.3329, simple_loss=0.3882, pruned_loss=0.1388, over 28855.00 frames. ], tot_loss[loss=0.3591, simple_loss=0.4097, pruned_loss=0.1543, over 5660036.72 frames. ], libri_tot_loss[loss=0.3581, simple_loss=0.4059, pruned_loss=0.1551, over 5668041.75 frames. ], giga_tot_loss[loss=0.3571, simple_loss=0.4085, pruned_loss=0.1528, over 5662878.08 frames. ], batch size: 199, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:49:02,620 INFO [train.py:968] (0/2) Epoch 4, batch 43750, giga_loss[loss=0.4302, simple_loss=0.4517, pruned_loss=0.2044, over 27560.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4068, pruned_loss=0.1525, over 5676119.40 frames. ], libri_tot_loss[loss=0.3573, simple_loss=0.4053, pruned_loss=0.1547, over 5674145.85 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4065, pruned_loss=0.1518, over 5672907.61 frames. ], batch size: 472, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:49:54,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.195e+02 1.419e+03 1.939e+03 2.731e+03 5.498e+03, threshold=3.877e+03, percent-clipped=7.0
+2023-03-02 11:49:54,087 INFO [train.py:968] (0/2) Epoch 4, batch 43800, giga_loss[loss=0.3093, simple_loss=0.3677, pruned_loss=0.1255, over 28600.00 frames. ], tot_loss[loss=0.3563, simple_loss=0.4061, pruned_loss=0.1533, over 5663464.22 frames. ], libri_tot_loss[loss=0.3573, simple_loss=0.4053, pruned_loss=0.1546, over 5678299.52 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4059, pruned_loss=0.1527, over 5657151.82 frames. ], batch size: 92, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:50:43,047 INFO [train.py:968] (0/2) Epoch 4, batch 43850, giga_loss[loss=0.3069, simple_loss=0.3703, pruned_loss=0.1217, over 28938.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.4043, pruned_loss=0.1531, over 5663849.58 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.405, pruned_loss=0.1545, over 5680205.31 frames. ], giga_tot_loss[loss=0.3549, simple_loss=0.4044, pruned_loss=0.1527, over 5656858.12 frames. ], batch size: 145, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:50:49,082 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-180000.pt
+2023-03-02 11:51:16,805 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:51:26,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4181, 4.1974, 4.0154, 1.8821], device='cuda:0'), covar=tensor([0.0478, 0.0555, 0.1024, 0.1972], device='cuda:0'), in_proj_covar=tensor([0.0834, 0.0740, 0.0816, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 11:51:33,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.510e+02 1.780e+03 2.283e+03 3.150e+03 7.721e+03, threshold=4.566e+03, percent-clipped=15.0
+2023-03-02 11:51:33,436 INFO [train.py:968] (0/2) Epoch 4, batch 43900, giga_loss[loss=0.3406, simple_loss=0.3932, pruned_loss=0.144, over 28943.00 frames. ], tot_loss[loss=0.3548, simple_loss=0.4034, pruned_loss=0.1531, over 5653436.62 frames. ], libri_tot_loss[loss=0.3568, simple_loss=0.4049, pruned_loss=0.1544, over 5677175.78 frames. ], giga_tot_loss[loss=0.3546, simple_loss=0.4035, pruned_loss=0.1529, over 5649843.97 frames. ], batch size: 112, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:51:50,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1710, 1.8512, 1.4234, 0.3904], device='cuda:0'), covar=tensor([0.1889, 0.1279, 0.2024, 0.2350], device='cuda:0'), in_proj_covar=tensor([0.1351, 0.1279, 0.1346, 0.1128], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 11:52:19,185 INFO [train.py:968] (0/2) Epoch 4, batch 43950, giga_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1222, over 28554.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4029, pruned_loss=0.1526, over 5648477.77 frames. ], libri_tot_loss[loss=0.3565, simple_loss=0.4047, pruned_loss=0.1542, over 5674176.91 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4032, pruned_loss=0.1525, over 5647451.96 frames. ], batch size: 60, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:52:57,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:53:09,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.235e+02 1.697e+03 2.015e+03 2.720e+03 9.159e+03, threshold=4.031e+03, percent-clipped=3.0
+2023-03-02 11:53:09,616 INFO [train.py:968] (0/2) Epoch 4, batch 44000, giga_loss[loss=0.4244, simple_loss=0.4517, pruned_loss=0.1985, over 27754.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4011, pruned_loss=0.1515, over 5656590.57 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4048, pruned_loss=0.1542, over 5676124.39 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4012, pruned_loss=0.1515, over 5653920.73 frames. ], batch size: 472, lr: 7.26e-03, grad_scale: 8.0
+2023-03-02 11:53:34,030 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.22 vs. limit=5.0
+2023-03-02 11:53:50,758 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:53:53,221 INFO [train.py:968] (0/2) Epoch 4, batch 44050, giga_loss[loss=0.338, simple_loss=0.3978, pruned_loss=0.1391, over 28618.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.4002, pruned_loss=0.1507, over 5654744.29 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.4052, pruned_loss=0.1545, over 5668604.75 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.3997, pruned_loss=0.1502, over 5658623.13 frames. ], batch size: 336, lr: 7.26e-03, grad_scale: 8.0
+2023-03-02 11:54:13,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:54:40,191 INFO [train.py:968] (0/2) Epoch 4, batch 44100, giga_loss[loss=0.2895, simple_loss=0.3619, pruned_loss=0.1085, over 28781.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.3992, pruned_loss=0.1496, over 5658683.01 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4046, pruned_loss=0.154, over 5672933.78 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.3993, pruned_loss=0.1496, over 5657262.99 frames. ], batch size: 119, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:54:40,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.291e+02 1.416e+03 1.911e+03 2.955e+03 5.839e+03, threshold=3.822e+03, percent-clipped=9.0
+2023-03-02 11:55:08,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:55:12,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:55:26,535 INFO [train.py:968] (0/2) Epoch 4, batch 44150, giga_loss[loss=0.337, simple_loss=0.4046, pruned_loss=0.1347, over 29024.00 frames. ], tot_loss[loss=0.3524, simple_loss=0.4022, pruned_loss=0.1513, over 5643568.52 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.4047, pruned_loss=0.1538, over 5671015.13 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.402, pruned_loss=0.1512, over 5643909.92 frames. ], batch size: 136, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:55:34,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180299.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:55:38,977 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:55:49,434 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:56:15,745 INFO [train.py:968] (0/2) Epoch 4, batch 44200, giga_loss[loss=0.3384, simple_loss=0.3958, pruned_loss=0.1405, over 28821.00 frames. ], tot_loss[loss=0.3548, simple_loss=0.4043, pruned_loss=0.1527, over 5640910.75 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.4051, pruned_loss=0.1543, over 5664108.59 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.4037, pruned_loss=0.1522, over 5647521.41 frames. ], batch size: 199, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:56:16,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.310e+02 1.540e+03 2.283e+03 3.473e+03 7.667e+03, threshold=4.566e+03, percent-clipped=15.0
+2023-03-02 11:56:29,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180355.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:56:33,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180358.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:56:58,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180387.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:57:03,220 INFO [train.py:968] (0/2) Epoch 4, batch 44250, giga_loss[loss=0.3796, simple_loss=0.4202, pruned_loss=0.1695, over 28596.00 frames. ], tot_loss[loss=0.3524, simple_loss=0.403, pruned_loss=0.1509, over 5651986.56 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4047, pruned_loss=0.154, over 5661014.19 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.4029, pruned_loss=0.1508, over 5659941.60 frames. ], batch size: 307, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:57:14,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:57:28,154 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:57:49,467 INFO [train.py:968] (0/2) Epoch 4, batch 44300, giga_loss[loss=0.4455, simple_loss=0.4746, pruned_loss=0.2082, over 28683.00 frames. ], tot_loss[loss=0.3508, simple_loss=0.4043, pruned_loss=0.1486, over 5657386.67 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.405, pruned_loss=0.1542, over 5660312.89 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4039, pruned_loss=0.1482, over 5664084.19 frames. ], batch size: 262, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:57:50,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.751e+02 1.269e+03 1.810e+03 2.655e+03 5.283e+03, threshold=3.620e+03, percent-clipped=3.0
+2023-03-02 11:58:32,526 INFO [train.py:968] (0/2) Epoch 4, batch 44350, giga_loss[loss=0.3337, simple_loss=0.4002, pruned_loss=0.1336, over 28812.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4061, pruned_loss=0.148, over 5662570.69 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.405, pruned_loss=0.1541, over 5664244.10 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4058, pruned_loss=0.1477, over 5664182.43 frames. ], batch size: 199, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:59:11,826 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-02 11:59:20,078 INFO [train.py:968] (0/2) Epoch 4, batch 44400, giga_loss[loss=0.3218, simple_loss=0.3913, pruned_loss=0.1261, over 28956.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4092, pruned_loss=0.1516, over 5655996.87 frames. ], libri_tot_loss[loss=0.3565, simple_loss=0.4047, pruned_loss=0.1542, over 5670667.00 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4092, pruned_loss=0.1512, over 5651267.93 frames. ], batch size: 106, lr: 7.25e-03, grad_scale: 8.0
+2023-03-02 11:59:20,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.352e+02 1.758e+03 2.404e+03 3.567e+03 7.692e+03, threshold=4.807e+03, percent-clipped=22.0
+2023-03-02 11:59:23,373 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:59:25,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:59:40,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:59:53,292 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:00:06,388 INFO [train.py:968] (0/2) Epoch 4, batch 44450, giga_loss[loss=0.4148, simple_loss=0.451, pruned_loss=0.1893, over 28068.00 frames. ], tot_loss[loss=0.3603, simple_loss=0.4114, pruned_loss=0.1546, over 5656029.44 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.4043, pruned_loss=0.1538, over 5666228.66 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4121, pruned_loss=0.1545, over 5655632.59 frames. ], batch size: 412, lr: 7.25e-03, grad_scale: 8.0
+2023-03-02 12:00:12,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:00:54,136 INFO [train.py:968] (0/2) Epoch 4, batch 44500, giga_loss[loss=0.3141, simple_loss=0.3826, pruned_loss=0.1228, over 28985.00 frames. ], tot_loss[loss=0.3602, simple_loss=0.4108, pruned_loss=0.1548, over 5664089.01 frames. ], libri_tot_loss[loss=0.3559, simple_loss=0.4041, pruned_loss=0.1538, over 5671050.77 frames. ], giga_tot_loss[loss=0.3606, simple_loss=0.4117, pruned_loss=0.1547, over 5659279.87 frames. ], batch size: 174, lr: 7.25e-03, grad_scale: 8.0
+2023-03-02 12:00:55,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.304e+02 1.560e+03 2.058e+03 2.769e+03 5.993e+03, threshold=4.116e+03, percent-clipped=3.0
+2023-03-02 12:01:23,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180674.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 12:01:38,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:01:39,024 INFO [train.py:968] (0/2) Epoch 4, batch 44550, giga_loss[loss=0.3514, simple_loss=0.4067, pruned_loss=0.148, over 28657.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.4087, pruned_loss=0.1533, over 5674088.16 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4037, pruned_loss=0.1534, over 5678070.02 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4099, pruned_loss=0.1536, over 5663672.53 frames. ], batch size: 262, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 12:01:47,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5931, 2.9328, 1.5853, 1.4722], device='cuda:0'), covar=tensor([0.0751, 0.0322, 0.0768, 0.1216], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0475, 0.0310, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 12:01:52,082 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:01:54,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:01:58,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1844, 1.5804, 1.2191, 0.7535], device='cuda:0'), covar=tensor([0.1474, 0.0979, 0.1109, 0.1831], device='cuda:0'), in_proj_covar=tensor([0.1366, 0.1270, 0.1350, 0.1139], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 12:02:13,286 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:02:21,691 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:02:24,265 INFO [train.py:968] (0/2) Epoch 4, batch 44600, giga_loss[loss=0.2917, simple_loss=0.37, pruned_loss=0.1067, over 28574.00 frames. ], tot_loss[loss=0.3557, simple_loss=0.4077, pruned_loss=0.1519, over 5672871.13 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4038, pruned_loss=0.1534, over 5679462.07 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4085, pruned_loss=0.1521, over 5663277.76 frames. ], batch size: 60, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:02:25,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.593e+02 1.419e+03 1.943e+03 2.481e+03 6.244e+03, threshold=3.885e+03, percent-clipped=4.0
+2023-03-02 12:03:07,388 INFO [train.py:968] (0/2) Epoch 4, batch 44650, giga_loss[loss=0.31, simple_loss=0.3891, pruned_loss=0.1154, over 29057.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.4076, pruned_loss=0.15, over 5679595.87 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4036, pruned_loss=0.1533, over 5681024.69 frames. ], giga_tot_loss[loss=0.3546, simple_loss=0.4087, pruned_loss=0.1503, over 5670180.48 frames. ], batch size: 155, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:03:08,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:03:26,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8044, 1.6384, 1.2962, 1.2712], device='cuda:0'), covar=tensor([0.0648, 0.0691, 0.1024, 0.1060], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0459, 0.0506, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 12:03:29,989 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180817.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 12:03:32,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180820.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 12:03:45,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:03:47,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:03:50,691 INFO [train.py:968] (0/2) Epoch 4, batch 44700, giga_loss[loss=0.2835, simple_loss=0.3548, pruned_loss=0.1061, over 28493.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.4057, pruned_loss=0.1474, over 5691978.97 frames. ], libri_tot_loss[loss=0.354, simple_loss=0.4027, pruned_loss=0.1526, over 5687477.53 frames. ], giga_tot_loss[loss=0.3518, simple_loss=0.4074, pruned_loss=0.1481, over 5678694.25 frames. ], batch size: 85, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:03:52,130 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.600e+02 1.412e+03 2.075e+03 2.767e+03 6.252e+03, threshold=4.149e+03, percent-clipped=6.0
+2023-03-02 12:03:57,592 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180849.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 12:04:18,246 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:04:42,258 INFO [train.py:968] (0/2) Epoch 4, batch 44750, giga_loss[loss=0.3421, simple_loss=0.4015, pruned_loss=0.1414, over 28624.00 frames. ], tot_loss[loss=0.3551, simple_loss=0.4082, pruned_loss=0.151, over 5674660.33 frames. ], libri_tot_loss[loss=0.3538, simple_loss=0.4026, pruned_loss=0.1525, over 5690926.53 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.4098, pruned_loss=0.1516, over 5660875.75 frames. ], batch size: 307, lr: 7.24e-03, grad_scale: 2.0
+2023-03-02 12:05:22,392 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:05:25,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:05:27,431 INFO [train.py:968] (0/2) Epoch 4, batch 44800, giga_loss[loss=0.4426, simple_loss=0.4641, pruned_loss=0.2106, over 26597.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4083, pruned_loss=0.1517, over 5651018.43 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4034, pruned_loss=0.1531, over 5674671.27 frames. ], giga_tot_loss[loss=0.3562, simple_loss=0.4091, pruned_loss=0.1516, over 5653938.10 frames. ], batch size: 555, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:05:30,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.742e+02 1.883e+03 2.673e+03 4.019e+03 8.800e+03, threshold=5.346e+03, percent-clipped=22.0
+2023-03-02 12:05:38,971 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6339, 1.5163, 1.4093, 1.3727], device='cuda:0'), covar=tensor([0.0867, 0.1462, 0.1457, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0748, 0.0638, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 12:05:44,782 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:05:51,436 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:05:52,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2318, 1.2452, 1.0429, 1.3293], device='cuda:0'), covar=tensor([0.0766, 0.0338, 0.0344, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0199, 0.0128, 0.0131, 0.0236], device='cuda:0'), out_proj_covar=tensor([0.0053, 0.0039, 0.0035, 0.0059], device='cuda:0')
+2023-03-02 12:05:55,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:06:03,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2773, 1.8261, 1.3630, 1.3394], device='cuda:0'), covar=tensor([0.0797, 0.0285, 0.0327, 0.0874], device='cuda:0'), in_proj_covar=tensor([0.0199, 0.0128, 0.0132, 0.0236], device='cuda:0'), out_proj_covar=tensor([0.0053, 0.0039, 0.0035, 0.0059], device='cuda:0')
+2023-03-02 12:06:08,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6763, 1.4280, 1.1542, 1.1729], device='cuda:0'), covar=tensor([0.0558, 0.0574, 0.0874, 0.0955], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0458, 0.0509, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 12:06:15,088 INFO [train.py:968] (0/2) Epoch 4, batch 44850, giga_loss[loss=0.4485, simple_loss=0.4676, pruned_loss=0.2148, over 28627.00 frames. ], tot_loss[loss=0.3539, simple_loss=0.4061, pruned_loss=0.1509, over 5660083.32 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4036, pruned_loss=0.1533, over 5673957.85 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.4065, pruned_loss=0.1506, over 5662997.31 frames. ], batch size: 85, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:06:19,676 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-02 12:06:44,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.04 vs. limit=2.0
+2023-03-02 12:07:00,642 INFO [train.py:968] (0/2) Epoch 4, batch 44900, giga_loss[loss=0.3497, simple_loss=0.3984, pruned_loss=0.1505, over 28342.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.4041, pruned_loss=0.1508, over 5652272.58 frames. ], libri_tot_loss[loss=0.3549, simple_loss=0.4033, pruned_loss=0.1533, over 5668992.11 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.4047, pruned_loss=0.1505, over 5658351.17 frames. ], batch size: 65, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:07:02,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.921e+02 1.642e+03 2.195e+03 3.145e+03 6.957e+03, threshold=4.391e+03, percent-clipped=5.0
+2023-03-02 12:07:11,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2845, 1.2253, 1.0752, 0.9365], device='cuda:0'), covar=tensor([0.0604, 0.0493, 0.0948, 0.0747], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0456, 0.0507, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 12:07:41,685 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 12:07:50,970 INFO [train.py:968] (0/2) Epoch 4, batch 44950, giga_loss[loss=0.36, simple_loss=0.4245, pruned_loss=0.1477, over 28629.00 frames. ], tot_loss[loss=0.3506, simple_loss=0.402, pruned_loss=0.1496, over 5649779.83 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.4035, pruned_loss=0.1534, over 5670199.15 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.4023, pruned_loss=0.1493, over 5653491.81 frames. ], batch size: 60, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:07:58,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:08:11,315 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:08:14,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:08:19,003 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:08:30,555 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-02 12:08:33,472 INFO [train.py:968] (0/2) Epoch 4, batch 45000, giga_loss[loss=0.3403, simple_loss=0.3911, pruned_loss=0.1448, over 28731.00 frames. ], tot_loss[loss=0.352, simple_loss=0.402, pruned_loss=0.151, over 5651240.36 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.4036, pruned_loss=0.1534, over 5668333.14 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4022, pruned_loss=0.1506, over 5655502.83 frames. ], batch size: 284, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:08:33,477 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 12:08:41,615 INFO [train.py:1012] (0/2) Epoch 4, validation: loss=0.2436, simple_loss=0.3481, pruned_loss=0.06962, over 944034.00 frames. 
+2023-03-02 12:08:41,616 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 12:08:44,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.519e+02 1.580e+03 1.995e+03 2.718e+03 1.036e+04, threshold=3.989e+03, percent-clipped=4.0
+2023-03-02 12:08:44,975 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8681, 2.1183, 2.0463, 1.8519], device='cuda:0'), covar=tensor([0.1420, 0.1707, 0.1028, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0761, 0.0756, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 12:08:46,155 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:09:25,338 INFO [train.py:968] (0/2) Epoch 4, batch 45050, giga_loss[loss=0.367, simple_loss=0.4207, pruned_loss=0.1566, over 27983.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4019, pruned_loss=0.1513, over 5642226.56 frames. ], libri_tot_loss[loss=0.3556, simple_loss=0.4039, pruned_loss=0.1537, over 5672348.57 frames. ], giga_tot_loss[loss=0.3515, simple_loss=0.4017, pruned_loss=0.1507, over 5640972.94 frames. ], batch size: 412, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:09:36,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4313, 2.2232, 1.6985, 0.6202], device='cuda:0'), covar=tensor([0.1860, 0.1087, 0.1739, 0.2011], device='cuda:0'), in_proj_covar=tensor([0.1363, 0.1273, 0.1352, 0.1135], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 12:09:47,260 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:09:55,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3961, 4.2042, 4.0857, 2.1790], device='cuda:0'), covar=tensor([0.0389, 0.0414, 0.0676, 0.1746], device='cuda:0'), in_proj_covar=tensor([0.0833, 0.0740, 0.0809, 0.0586], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 12:09:57,944 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1533, 1.3963, 1.1115, 0.8439], device='cuda:0'), covar=tensor([0.1053, 0.0811, 0.0639, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.1365, 0.1122, 0.1120, 0.1206], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 12:10:11,105 INFO [train.py:968] (0/2) Epoch 4, batch 45100, giga_loss[loss=0.3317, simple_loss=0.3886, pruned_loss=0.1374, over 28713.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3977, pruned_loss=0.1459, over 5656453.42 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4031, pruned_loss=0.1532, over 5676863.42 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.3982, pruned_loss=0.1458, over 5650777.34 frames. ], batch size: 262, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:10:12,106 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.10 vs. limit=2.0
+2023-03-02 12:10:12,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.943e+02 1.350e+03 1.662e+03 2.276e+03 4.740e+03, threshold=3.324e+03, percent-clipped=3.0
+2023-03-02 12:10:13,834 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:10:16,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:10:40,671 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:10:53,372 INFO [train.py:968] (0/2) Epoch 4, batch 45150, giga_loss[loss=0.2966, simple_loss=0.362, pruned_loss=0.1156, over 28691.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3944, pruned_loss=0.143, over 5654818.03 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4032, pruned_loss=0.1535, over 5684833.46 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3945, pruned_loss=0.1424, over 5642864.82 frames. ], batch size: 92, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:11:43,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:11:47,387 INFO [train.py:968] (0/2) Epoch 4, batch 45200, giga_loss[loss=0.3134, simple_loss=0.3708, pruned_loss=0.128, over 28886.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3943, pruned_loss=0.1432, over 5660238.36 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4032, pruned_loss=0.1535, over 5684833.46 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3944, pruned_loss=0.1427, over 5650935.03 frames. ], batch size: 199, lr: 7.23e-03, grad_scale: 8.0
+2023-03-02 12:11:49,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.751e+02 1.609e+03 1.956e+03 3.035e+03 8.944e+03, threshold=3.912e+03, percent-clipped=16.0
+2023-03-02 12:12:34,511 INFO [train.py:968] (0/2) Epoch 4, batch 45250, giga_loss[loss=0.3196, simple_loss=0.3771, pruned_loss=0.1311, over 28885.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3932, pruned_loss=0.1434, over 5664970.70 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.4034, pruned_loss=0.1535, over 5680231.05 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3927, pruned_loss=0.1426, over 5661634.17 frames. ], batch size: 145, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:12:44,314 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-02 12:13:24,341 INFO [train.py:968] (0/2) Epoch 4, batch 45300, giga_loss[loss=0.3109, simple_loss=0.3738, pruned_loss=0.124, over 28642.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3922, pruned_loss=0.1429, over 5675243.50 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.4032, pruned_loss=0.1533, over 5685064.99 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3917, pruned_loss=0.1423, over 5668233.29 frames. ], batch size: 242, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:13:28,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.172e+02 1.705e+03 2.150e+03 3.262e+03 9.687e+03, threshold=4.299e+03, percent-clipped=20.0
+2023-03-02 12:13:58,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:14:01,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:14:08,300 INFO [train.py:968] (0/2) Epoch 4, batch 45350, giga_loss[loss=0.3136, simple_loss=0.3858, pruned_loss=0.1207, over 29108.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3933, pruned_loss=0.1425, over 5678669.52 frames. ], libri_tot_loss[loss=0.3547, simple_loss=0.4032, pruned_loss=0.1531, over 5678115.28 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3927, pruned_loss=0.142, over 5679887.33 frames. ], batch size: 155, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:14:15,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:14:27,462 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:14:36,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-02 12:14:55,416 INFO [train.py:968] (0/2) Epoch 4, batch 45400, giga_loss[loss=0.352, simple_loss=0.4124, pruned_loss=0.1458, over 28700.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3935, pruned_loss=0.1422, over 5668072.38 frames. ], libri_tot_loss[loss=0.3539, simple_loss=0.4026, pruned_loss=0.1526, over 5680060.02 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3934, pruned_loss=0.142, over 5667209.10 frames. ], batch size: 284, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:15:01,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.692e+02 1.297e+03 1.688e+03 2.300e+03 8.561e+03, threshold=3.376e+03, percent-clipped=5.0
+2023-03-02 12:15:44,962 INFO [train.py:968] (0/2) Epoch 4, batch 45450, giga_loss[loss=0.3764, simple_loss=0.4352, pruned_loss=0.1587, over 28654.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3944, pruned_loss=0.143, over 5670527.28 frames. ], libri_tot_loss[loss=0.3546, simple_loss=0.4032, pruned_loss=0.1531, over 5681606.69 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3937, pruned_loss=0.1423, over 5668298.53 frames. ], batch size: 284, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:15:45,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:16:33,250 INFO [train.py:968] (0/2) Epoch 4, batch 45500, libri_loss[loss=0.4167, simple_loss=0.4552, pruned_loss=0.1891, over 29660.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3959, pruned_loss=0.1452, over 5660936.87 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4033, pruned_loss=0.1532, over 5683929.24 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3952, pruned_loss=0.1445, over 5656933.17 frames. ], batch size: 91, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:16:33,589 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:16:37,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:16:37,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.512e+03 1.960e+03 2.680e+03 7.610e+03, threshold=3.920e+03, percent-clipped=9.0
+2023-03-02 12:16:44,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5398, 3.5028, 1.6311, 1.3817], device='cuda:0'), covar=tensor([0.0851, 0.0289, 0.0784, 0.1311], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0480, 0.0312, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 12:17:02,750 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:17:16,010 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:17:17,886 INFO [train.py:968] (0/2) Epoch 4, batch 45550, giga_loss[loss=0.366, simple_loss=0.4159, pruned_loss=0.158, over 28925.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3984, pruned_loss=0.1473, over 5628949.66 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4036, pruned_loss=0.1535, over 5657992.80 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3975, pruned_loss=0.1463, over 5649651.75 frames. ], batch size: 213, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:17:58,413 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:18:02,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:18:06,616 INFO [train.py:968] (0/2) Epoch 4, batch 45600, giga_loss[loss=0.3343, simple_loss=0.3885, pruned_loss=0.1401, over 28792.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.4003, pruned_loss=0.1483, over 5593269.67 frames. ], libri_tot_loss[loss=0.3562, simple_loss=0.4043, pruned_loss=0.1541, over 5613855.96 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3989, pruned_loss=0.1469, over 5646768.37 frames. ], batch size: 99, lr: 7.23e-03, grad_scale: 8.0
+2023-03-02 12:18:09,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.355e+02 1.458e+03 2.077e+03 2.837e+03 9.065e+03, threshold=4.154e+03, percent-clipped=9.0
+2023-03-02 12:18:29,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:18:51,099 INFO [train.py:968] (0/2) Epoch 4, batch 45650, giga_loss[loss=0.3841, simple_loss=0.4323, pruned_loss=0.168, over 28352.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4022, pruned_loss=0.15, over 5552410.08 frames. ], libri_tot_loss[loss=0.3575, simple_loss=0.4052, pruned_loss=0.1549, over 5562278.15 frames. ], giga_tot_loss[loss=0.3482, simple_loss=0.4002, pruned_loss=0.1481, over 5643327.93 frames. ], batch size: 368, lr: 7.22e-03, grad_scale: 8.0
+2023-03-02 12:19:41,133 INFO [train.py:968] (0/2) Epoch 4, batch 45700, giga_loss[loss=0.3899, simple_loss=0.4122, pruned_loss=0.1838, over 23618.00 frames. ], tot_loss[loss=0.3535, simple_loss=0.4037, pruned_loss=0.1516, over 5546270.77 frames. ], libri_tot_loss[loss=0.3584, simple_loss=0.4058, pruned_loss=0.1555, over 5528831.30 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4014, pruned_loss=0.1494, over 5648908.49 frames. ], batch size: 705, lr: 7.22e-03, grad_scale: 2.0
+2023-03-02 12:19:46,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.241e+02 1.782e+03 2.369e+03 3.053e+03 8.413e+03, threshold=4.737e+03, percent-clipped=9.0
+2023-03-02 12:20:34,109 INFO [train.py:968] (0/2) Epoch 4, batch 45750, giga_loss[loss=0.3137, simple_loss=0.3873, pruned_loss=0.12, over 28888.00 frames. ], tot_loss[loss=0.3551, simple_loss=0.4058, pruned_loss=0.1522, over 5551292.80 frames. ], libri_tot_loss[loss=0.3588, simple_loss=0.4061, pruned_loss=0.1558, over 5513802.42 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4038, pruned_loss=0.1502, over 5645861.17 frames. ], batch size: 284, lr: 7.22e-03, grad_scale: 2.0
+2023-03-02 12:20:43,554 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-02 12:20:46,505 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-4.pt
+2023-03-02 12:21:37,989 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:21:53,203 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-02 12:22:02,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.591e+02 1.276e+03 1.689e+03 2.579e+03 1.495e+04, threshold=3.379e+03, percent-clipped=13.0
+2023-03-02 12:22:03,018 INFO [train.py:968] (0/2) Epoch 5, batch 50, giga_loss[loss=0.2912, simple_loss=0.3726, pruned_loss=0.105, over 28855.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.4079, pruned_loss=0.1383, over 1262437.77 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3969, pruned_loss=0.1296, over 115742.17 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.409, pruned_loss=0.1392, over 1170121.25 frames. ], batch size: 199, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:22:49,428 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-182000.pt
+2023-03-02 12:22:49,731 INFO [train.py:968] (0/2) Epoch 5, batch 100, giga_loss[loss=0.2682, simple_loss=0.3477, pruned_loss=0.09437, over 28422.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3918, pruned_loss=0.128, over 2244846.39 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3683, pruned_loss=0.1133, over 337366.81 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3958, pruned_loss=0.1305, over 2026738.42 frames. ], batch size: 71, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:23:33,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.105e+02 1.023e+03 1.352e+03 1.980e+03 9.590e+03, threshold=2.705e+03, percent-clipped=5.0
+2023-03-02 12:23:34,097 INFO [train.py:968] (0/2) Epoch 5, batch 150, giga_loss[loss=0.3088, simple_loss=0.3679, pruned_loss=0.1249, over 28892.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3762, pruned_loss=0.1208, over 3011921.06 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.363, pruned_loss=0.1112, over 474030.26 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3789, pruned_loss=0.1225, over 2765758.52 frames. ], batch size: 145, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:23:47,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:24:15,568 INFO [train.py:968] (0/2) Epoch 5, batch 200, giga_loss[loss=0.2761, simple_loss=0.347, pruned_loss=0.1026, over 29098.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3624, pruned_loss=0.1138, over 3615951.37 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3656, pruned_loss=0.1117, over 608212.75 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3631, pruned_loss=0.1146, over 3362064.81 frames. ], batch size: 155, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:24:20,711 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.91 vs. limit=5.0
+2023-03-02 12:24:56,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.466e+02 9.994e+02 1.166e+03 1.524e+03 3.906e+03, threshold=2.331e+03, percent-clipped=4.0
+2023-03-02 12:24:57,220 INFO [train.py:968] (0/2) Epoch 5, batch 250, giga_loss[loss=0.2483, simple_loss=0.316, pruned_loss=0.0903, over 28707.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3492, pruned_loss=0.1064, over 4081993.17 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3639, pruned_loss=0.1104, over 687765.54 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3491, pruned_loss=0.1068, over 3853583.65 frames. ], batch size: 92, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:25:38,163 INFO [train.py:968] (0/2) Epoch 5, batch 300, giga_loss[loss=0.2303, simple_loss=0.3044, pruned_loss=0.07807, over 29085.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.341, pruned_loss=0.1022, over 4445389.07 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.369, pruned_loss=0.1131, over 842448.27 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3388, pruned_loss=0.1016, over 4218121.06 frames. ], batch size: 155, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:25:45,526 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:25:47,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:26:05,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5135, 2.2270, 1.7855, 1.9791], device='cuda:0'), covar=tensor([0.0600, 0.0707, 0.0890, 0.0966], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0458, 0.0511, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 12:26:19,667 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:26:26,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.469e+02 9.869e+02 1.372e+03 1.874e+03 8.501e+03, threshold=2.745e+03, percent-clipped=18.0
+2023-03-02 12:26:27,792 INFO [train.py:968] (0/2) Epoch 5, batch 350, giga_loss[loss=0.2452, simple_loss=0.3142, pruned_loss=0.08815, over 27961.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.332, pruned_loss=0.09759, over 4725529.26 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3703, pruned_loss=0.1137, over 918280.70 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3292, pruned_loss=0.09655, over 4528145.65 frames. ], batch size: 412, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:27:07,421 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:27:08,432 INFO [train.py:968] (0/2) Epoch 5, batch 400, giga_loss[loss=0.2325, simple_loss=0.3025, pruned_loss=0.08125, over 28901.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3274, pruned_loss=0.09551, over 4949191.72 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3687, pruned_loss=0.1132, over 1040347.87 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3242, pruned_loss=0.09424, over 4768999.64 frames. ], batch size: 227, lr: 6.72e-03, grad_scale: 4.0
+2023-03-02 12:27:38,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5798, 1.1116, 2.9185, 2.7449], device='cuda:0'), covar=tensor([0.1545, 0.2026, 0.0478, 0.0633], device='cuda:0'), in_proj_covar=tensor([0.0561, 0.0519, 0.0740, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 12:27:49,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.827e+02 9.502e+02 1.269e+03 1.675e+03 2.792e+03, threshold=2.538e+03, percent-clipped=1.0
+2023-03-02 12:27:50,468 INFO [train.py:968] (0/2) Epoch 5, batch 450, giga_loss[loss=0.2426, simple_loss=0.3029, pruned_loss=0.09114, over 28577.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3246, pruned_loss=0.09419, over 5118028.40 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3692, pruned_loss=0.1135, over 1135770.71 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3212, pruned_loss=0.09277, over 4960603.60 frames. ], batch size: 85, lr: 6.72e-03, grad_scale: 4.0
+2023-03-02 12:28:09,460 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=182371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:28:09,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 12:28:09,771 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-02 12:28:33,439 INFO [train.py:968] (0/2) Epoch 5, batch 500, giga_loss[loss=0.253, simple_loss=0.3187, pruned_loss=0.09371, over 27964.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3227, pruned_loss=0.09281, over 5253212.86 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3701, pruned_loss=0.1141, over 1293105.49 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3182, pruned_loss=0.09083, over 5109988.79 frames. ], batch size: 412, lr: 6.72e-03, grad_scale: 4.0
+2023-03-02 12:29:10,873 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:29:12,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:29:15,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.306e+02 9.761e+02 1.219e+03 1.592e+03 5.352e+03, threshold=2.438e+03, percent-clipped=8.0
+2023-03-02 12:29:16,322 INFO [train.py:968] (0/2) Epoch 5, batch 550, giga_loss[loss=0.2303, simple_loss=0.3, pruned_loss=0.08028, over 28922.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3213, pruned_loss=0.09217, over 5354147.65 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3722, pruned_loss=0.1153, over 1406776.08 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.316, pruned_loss=0.08975, over 5226610.17 frames. ], batch size: 186, lr: 6.72e-03, grad_scale: 4.0
+2023-03-02 12:29:34,322 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:29:58,202 INFO [train.py:968] (0/2) Epoch 5, batch 600, giga_loss[loss=0.2415, simple_loss=0.3086, pruned_loss=0.08721, over 28045.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3186, pruned_loss=0.09063, over 5433722.76 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.37, pruned_loss=0.1141, over 1562030.74 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3132, pruned_loss=0.08821, over 5315152.53 frames. ], batch size: 412, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:30:40,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.505e+02 9.137e+02 1.160e+03 1.471e+03 3.459e+03, threshold=2.319e+03, percent-clipped=7.0
+2023-03-02 12:30:42,838 INFO [train.py:968] (0/2) Epoch 5, batch 650, giga_loss[loss=0.2867, simple_loss=0.3336, pruned_loss=0.1199, over 26611.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3175, pruned_loss=0.08998, over 5478984.31 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3712, pruned_loss=0.1148, over 1701867.17 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3111, pruned_loss=0.08708, over 5378845.74 frames. ], batch size: 555, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:31:03,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4615, 2.2019, 1.5359, 0.6922], device='cuda:0'), covar=tensor([0.2742, 0.1326, 0.2085, 0.2874], device='cuda:0'), in_proj_covar=tensor([0.1366, 0.1263, 0.1332, 0.1131], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 12:31:28,206 INFO [train.py:968] (0/2) Epoch 5, batch 700, giga_loss[loss=0.2215, simple_loss=0.2907, pruned_loss=0.07617, over 29008.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3149, pruned_loss=0.08894, over 5527070.08 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3727, pruned_loss=0.1158, over 1783676.26 frames. ], giga_tot_loss[loss=0.2401, simple_loss=0.3084, pruned_loss=0.08587, over 5441765.06 frames. ], batch size: 106, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:32:10,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.372e+02 9.711e+02 1.236e+03 2.054e+03 5.301e+03, threshold=2.472e+03, percent-clipped=14.0
+2023-03-02 12:32:10,806 INFO [train.py:968] (0/2) Epoch 5, batch 750, giga_loss[loss=0.2251, simple_loss=0.291, pruned_loss=0.07957, over 27623.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3131, pruned_loss=0.08761, over 5578414.31 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3731, pruned_loss=0.1161, over 1925042.67 frames. ], giga_tot_loss[loss=0.2369, simple_loss=0.3057, pruned_loss=0.0841, over 5499236.25 frames. ], batch size: 472, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:32:32,668 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=182674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:32:53,370 INFO [train.py:968] (0/2) Epoch 5, batch 800, giga_loss[loss=0.2187, simple_loss=0.2836, pruned_loss=0.07689, over 28612.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3096, pruned_loss=0.08615, over 5597776.36 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3732, pruned_loss=0.1159, over 1994228.03 frames. ], giga_tot_loss[loss=0.2343, simple_loss=0.3027, pruned_loss=0.08294, over 5538902.59 frames. ], batch size: 85, lr: 6.71e-03, grad_scale: 8.0
+2023-03-02 12:33:21,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5804, 1.9218, 1.9603, 1.7451], device='cuda:0'), covar=tensor([0.1509, 0.1834, 0.1138, 0.1174], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0768, 0.0782, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 12:33:39,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:33:41,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.821e+02 1.086e+03 1.284e+03 1.872e+03 3.257e+03, threshold=2.568e+03, percent-clipped=9.0
+2023-03-02 12:33:42,314 INFO [train.py:968] (0/2) Epoch 5, batch 850, giga_loss[loss=0.283, simple_loss=0.3425, pruned_loss=0.1117, over 28890.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3174, pruned_loss=0.09122, over 5606219.52 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3733, pruned_loss=0.116, over 2009846.89 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3117, pruned_loss=0.08863, over 5562329.72 frames. ], batch size: 112, lr: 6.71e-03, grad_scale: 8.0
+2023-03-02 12:34:28,618 INFO [train.py:968] (0/2) Epoch 5, batch 900, giga_loss[loss=0.3144, simple_loss=0.3841, pruned_loss=0.1224, over 28827.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3325, pruned_loss=0.09961, over 5624127.65 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3731, pruned_loss=0.1156, over 2158130.14 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3263, pruned_loss=0.09688, over 5583068.73 frames. ], batch size: 174, lr: 6.71e-03, grad_scale: 8.0
+2023-03-02 12:35:11,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.602e+02 1.313e+03 1.673e+03 2.338e+03 3.996e+03, threshold=3.346e+03, percent-clipped=15.0
+2023-03-02 12:35:11,960 INFO [train.py:968] (0/2) Epoch 5, batch 950, giga_loss[loss=0.3218, simple_loss=0.3877, pruned_loss=0.1279, over 28789.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3462, pruned_loss=0.1074, over 5644847.70 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3728, pruned_loss=0.1156, over 2228703.94 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3409, pruned_loss=0.1051, over 5611058.20 frames. ], batch size: 92, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:35:45,780 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:35:48,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:35:55,600 INFO [train.py:968] (0/2) Epoch 5, batch 1000, giga_loss[loss=0.2884, simple_loss=0.362, pruned_loss=0.1074, over 28829.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3545, pruned_loss=0.1111, over 5658656.80 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3728, pruned_loss=0.1154, over 2265545.80 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3501, pruned_loss=0.1093, over 5629764.07 frames. ], batch size: 119, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:36:01,946 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=182907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:36:13,152 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:36:34,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.906e+02 1.149e+03 1.414e+03 1.838e+03 3.697e+03, threshold=2.828e+03, percent-clipped=1.0
+2023-03-02 12:36:34,170 INFO [train.py:968] (0/2) Epoch 5, batch 1050, giga_loss[loss=0.2996, simple_loss=0.3703, pruned_loss=0.1145, over 28736.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3576, pruned_loss=0.1109, over 5669081.08 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3722, pruned_loss=0.115, over 2352258.77 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3539, pruned_loss=0.1096, over 5643478.12 frames. ], batch size: 78, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:37:21,870 INFO [train.py:968] (0/2) Epoch 5, batch 1100, giga_loss[loss=0.2665, simple_loss=0.3454, pruned_loss=0.09379, over 28881.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3604, pruned_loss=0.1115, over 5667132.54 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3725, pruned_loss=0.1149, over 2419421.02 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3571, pruned_loss=0.1104, over 5645558.90 frames. ], batch size: 186, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:37:22,263 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-02 12:38:05,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:38:05,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.934e+02 1.026e+03 1.311e+03 1.724e+03 9.763e+03, threshold=2.622e+03, percent-clipped=7.0
+2023-03-02 12:38:05,719 INFO [train.py:968] (0/2) Epoch 5, batch 1150, giga_loss[loss=0.3447, simple_loss=0.3999, pruned_loss=0.1448, over 28877.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.362, pruned_loss=0.1125, over 5686117.48 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3724, pruned_loss=0.1148, over 2503908.77 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3592, pruned_loss=0.1116, over 5665328.40 frames. ], batch size: 186, lr: 6.70e-03, grad_scale: 4.0
+2023-03-02 12:38:49,616 INFO [train.py:968] (0/2) Epoch 5, batch 1200, giga_loss[loss=0.3074, simple_loss=0.3728, pruned_loss=0.121, over 28617.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3644, pruned_loss=0.1145, over 5673890.21 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3723, pruned_loss=0.1144, over 2584562.90 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.362, pruned_loss=0.1138, over 5655791.79 frames. ], batch size: 85, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:39:13,802 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 12:39:33,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.688e+02 1.181e+03 1.344e+03 1.569e+03 5.889e+03, threshold=2.689e+03, percent-clipped=2.0
+2023-03-02 12:39:33,824 INFO [train.py:968] (0/2) Epoch 5, batch 1250, giga_loss[loss=0.288, simple_loss=0.358, pruned_loss=0.109, over 29065.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3656, pruned_loss=0.1152, over 5678268.91 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3711, pruned_loss=0.1137, over 2665139.85 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3641, pruned_loss=0.115, over 5660228.96 frames. ], batch size: 128, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:40:09,689 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:40:12,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:40:16,404 INFO [train.py:968] (0/2) Epoch 5, batch 1300, libri_loss[loss=0.337, simple_loss=0.3969, pruned_loss=0.1386, over 29523.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3684, pruned_loss=0.1162, over 5685889.21 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3702, pruned_loss=0.1133, over 2776555.99 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3674, pruned_loss=0.1164, over 5664631.57 frames. ], batch size: 80, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:40:35,598 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:40:55,583 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=183249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:40:56,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.303e+02 1.045e+03 1.328e+03 1.693e+03 4.013e+03, threshold=2.655e+03, percent-clipped=8.0
+2023-03-02 12:40:56,126 INFO [train.py:968] (0/2) Epoch 5, batch 1350, giga_loss[loss=0.2867, simple_loss=0.3633, pruned_loss=0.105, over 28432.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3697, pruned_loss=0.1156, over 5703875.64 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3693, pruned_loss=0.1126, over 2882932.72 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3693, pruned_loss=0.1161, over 5681407.53 frames. ], batch size: 60, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:41:00,937 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=183255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:41:24,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:41:38,201 INFO [train.py:968] (0/2) Epoch 5, batch 1400, giga_loss[loss=0.3301, simple_loss=0.3999, pruned_loss=0.1302, over 29036.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3709, pruned_loss=0.1158, over 5696314.65 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.369, pruned_loss=0.1125, over 2924048.67 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3708, pruned_loss=0.1163, over 5679253.03 frames. ], batch size: 136, lr: 6.70e-03, grad_scale: 4.0
+2023-03-02 12:42:20,663 INFO [train.py:968] (0/2) Epoch 5, batch 1450, giga_loss[loss=0.297, simple_loss=0.3704, pruned_loss=0.1118, over 28988.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3706, pruned_loss=0.1145, over 5693141.47 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3684, pruned_loss=0.1122, over 3016623.36 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3708, pruned_loss=0.1152, over 5682641.76 frames. ], batch size: 106, lr: 6.70e-03, grad_scale: 4.0
+2023-03-02 12:42:21,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.036e+02 1.099e+03 1.372e+03 1.932e+03 6.527e+03, threshold=2.744e+03, percent-clipped=10.0
+2023-03-02 12:43:01,226 INFO [train.py:968] (0/2) Epoch 5, batch 1500, giga_loss[loss=0.3012, simple_loss=0.3735, pruned_loss=0.1144, over 28565.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3694, pruned_loss=0.1126, over 5699612.39 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3679, pruned_loss=0.1118, over 3071307.94 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3698, pruned_loss=0.1133, over 5690111.64 frames. ], batch size: 307, lr: 6.70e-03, grad_scale: 4.0
+2023-03-02 12:43:21,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:43:23,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:43:38,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2774, 1.5168, 1.4424, 1.4444], device='cuda:0'), covar=tensor([0.1143, 0.1188, 0.1510, 0.1233], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0746, 0.0634, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 12:43:42,982 INFO [train.py:968] (0/2) Epoch 5, batch 1550, giga_loss[loss=0.3676, simple_loss=0.4085, pruned_loss=0.1634, over 27894.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3681, pruned_loss=0.1115, over 5698997.32 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3685, pruned_loss=0.1122, over 3124226.93 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3682, pruned_loss=0.1119, over 5690908.41 frames. ], batch size: 412, lr: 6.70e-03, grad_scale: 4.0
+2023-03-02 12:43:43,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.017e+02 1.063e+03 1.338e+03 1.808e+03 3.117e+03, threshold=2.676e+03, percent-clipped=4.0
+2023-03-02 12:43:48,321 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:44:05,111 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8964, 1.0766, 4.0565, 3.1652], device='cuda:0'), covar=tensor([0.1559, 0.2161, 0.0299, 0.0574], device='cuda:0'), in_proj_covar=tensor([0.0560, 0.0519, 0.0729, 0.0599], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 12:44:27,755 INFO [train.py:968] (0/2) Epoch 5, batch 1600, libri_loss[loss=0.3186, simple_loss=0.3942, pruned_loss=0.1215, over 29217.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3693, pruned_loss=0.1135, over 5701806.09 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3686, pruned_loss=0.1122, over 3220684.57 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3694, pruned_loss=0.1138, over 5699417.08 frames. ], batch size: 97, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:44:38,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2407, 1.9031, 1.4397, 0.6103], device='cuda:0'), covar=tensor([0.2277, 0.1357, 0.2010, 0.2540], device='cuda:0'), in_proj_covar=tensor([0.1363, 0.1263, 0.1353, 0.1129], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 12:45:12,586 INFO [train.py:968] (0/2) Epoch 5, batch 1650, giga_loss[loss=0.4405, simple_loss=0.4212, pruned_loss=0.2299, over 23562.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3731, pruned_loss=0.1197, over 5701962.17 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3685, pruned_loss=0.1121, over 3234107.15 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3731, pruned_loss=0.12, over 5699181.65 frames. ], batch size: 705, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:45:13,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.830e+02 1.290e+03 1.459e+03 2.157e+03 4.009e+03, threshold=2.919e+03, percent-clipped=6.0
+2023-03-02 12:45:44,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4114, 1.4354, 1.4947, 1.4133], device='cuda:0'), covar=tensor([0.1149, 0.1434, 0.1704, 0.1351], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0750, 0.0636, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 12:45:58,466 INFO [train.py:968] (0/2) Epoch 5, batch 1700, giga_loss[loss=0.3527, simple_loss=0.3829, pruned_loss=0.1612, over 23859.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3742, pruned_loss=0.1226, over 5682714.40 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3676, pruned_loss=0.1116, over 3286754.46 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3749, pruned_loss=0.1232, over 5687470.92 frames. ], batch size: 705, lr: 6.69e-03, grad_scale: 8.0
+2023-03-02 12:46:16,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:46:21,695 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:46:40,428 INFO [train.py:968] (0/2) Epoch 5, batch 1750, giga_loss[loss=0.3064, simple_loss=0.3696, pruned_loss=0.1216, over 29118.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3729, pruned_loss=0.1221, over 5692545.79 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3681, pruned_loss=0.1116, over 3375084.91 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3734, pruned_loss=0.123, over 5691489.73 frames. ], batch size: 113, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:46:41,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.720e+02 1.246e+03 1.653e+03 2.330e+03 5.403e+03, threshold=3.306e+03, percent-clipped=15.0
+2023-03-02 12:47:09,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4659, 1.5697, 1.5076, 1.4257], device='cuda:0'), covar=tensor([0.1156, 0.1512, 0.1612, 0.1504], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0749, 0.0637, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 12:47:22,688 INFO [train.py:968] (0/2) Epoch 5, batch 1800, giga_loss[loss=0.3052, simple_loss=0.3537, pruned_loss=0.1283, over 23778.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3714, pruned_loss=0.1212, over 5701525.18 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3684, pruned_loss=0.1119, over 3425154.46 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3716, pruned_loss=0.122, over 5697014.03 frames. ], batch size: 705, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:48:05,659 INFO [train.py:968] (0/2) Epoch 5, batch 1850, giga_loss[loss=0.3132, simple_loss=0.3774, pruned_loss=0.1245, over 28870.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3711, pruned_loss=0.1205, over 5708767.00 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3679, pruned_loss=0.1115, over 3449815.42 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3716, pruned_loss=0.1214, over 5703446.94 frames. ], batch size: 227, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:48:08,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.995e+02 1.102e+03 1.390e+03 1.941e+03 7.701e+03, threshold=2.780e+03, percent-clipped=3.0
+2023-03-02 12:48:21,175 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:23,245 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:26,145 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:28,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:30,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3466, 1.4173, 1.2592, 1.6577], device='cuda:0'), covar=tensor([0.1984, 0.1771, 0.1695, 0.1852], device='cuda:0'), in_proj_covar=tensor([0.1112, 0.0862, 0.0980, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 12:48:51,784 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:52,297 INFO [train.py:968] (0/2) Epoch 5, batch 1900, giga_loss[loss=0.2802, simple_loss=0.3617, pruned_loss=0.09934, over 28406.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3689, pruned_loss=0.1181, over 5706594.84 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3677, pruned_loss=0.1111, over 3521944.11 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3694, pruned_loss=0.1192, over 5697746.42 frames. ], batch size: 65, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:48:53,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=183801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:55,872 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:49:35,289 INFO [train.py:968] (0/2) Epoch 5, batch 1950, giga_loss[loss=0.2728, simple_loss=0.3555, pruned_loss=0.09508, over 28946.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3664, pruned_loss=0.1164, over 5695216.29 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3679, pruned_loss=0.1112, over 3604740.49 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3668, pruned_loss=0.1175, over 5690476.80 frames. ], batch size: 174, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:49:36,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.737e+02 1.061e+03 1.456e+03 1.877e+03 5.762e+03, threshold=2.911e+03, percent-clipped=9.0
+2023-03-02 12:50:19,166 INFO [train.py:968] (0/2) Epoch 5, batch 2000, giga_loss[loss=0.2689, simple_loss=0.3366, pruned_loss=0.1006, over 28926.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.359, pruned_loss=0.1119, over 5691483.86 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3667, pruned_loss=0.1105, over 3684079.52 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3598, pruned_loss=0.1133, over 5680749.86 frames. ], batch size: 213, lr: 6.69e-03, grad_scale: 8.0
+2023-03-02 12:50:44,848 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-02 12:51:05,768 INFO [train.py:968] (0/2) Epoch 5, batch 2050, giga_loss[loss=0.2818, simple_loss=0.3455, pruned_loss=0.1091, over 28930.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3534, pruned_loss=0.1094, over 5683208.99 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3666, pruned_loss=0.1106, over 3706418.59 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.354, pruned_loss=0.1104, over 5672684.42 frames. ], batch size: 213, lr: 6.69e-03, grad_scale: 8.0
+2023-03-02 12:51:09,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.207e+02 9.506e+02 1.145e+03 1.691e+03 7.305e+03, threshold=2.291e+03, percent-clipped=9.0
+2023-03-02 12:51:57,094 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-184000.pt
+2023-03-02 12:51:57,392 INFO [train.py:968] (0/2) Epoch 5, batch 2100, giga_loss[loss=0.2625, simple_loss=0.3349, pruned_loss=0.09507, over 28775.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3512, pruned_loss=0.108, over 5687723.29 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3666, pruned_loss=0.1107, over 3728409.25 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3515, pruned_loss=0.1088, over 5677435.65 frames. ], batch size: 186, lr: 6.69e-03, grad_scale: 8.0
+2023-03-02 12:52:03,194 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:52:33,026 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-02 12:52:36,011 INFO [train.py:968] (0/2) Epoch 5, batch 2150, giga_loss[loss=0.2706, simple_loss=0.3467, pruned_loss=0.09721, over 28674.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3526, pruned_loss=0.1082, over 5699073.57 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3672, pruned_loss=0.1112, over 3813590.56 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3519, pruned_loss=0.1084, over 5683709.08 frames. ], batch size: 262, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:52:37,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.461e+02 1.004e+03 1.286e+03 1.672e+03 4.375e+03, threshold=2.573e+03, percent-clipped=12.0
+2023-03-02 12:52:44,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5177, 3.2451, 1.6728, 1.4596], device='cuda:0'), covar=tensor([0.0918, 0.0289, 0.0784, 0.1337], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0463, 0.0307, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0023, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 12:52:53,836 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-02 12:52:59,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2354, 4.8914, 2.3245, 2.3119], device='cuda:0'), covar=tensor([0.0788, 0.0183, 0.0692, 0.1063], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0464, 0.0307, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0023, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 12:53:17,244 INFO [train.py:968] (0/2) Epoch 5, batch 2200, giga_loss[loss=0.2777, simple_loss=0.3559, pruned_loss=0.09979, over 28968.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3514, pruned_loss=0.1069, over 5706246.05 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3675, pruned_loss=0.1111, over 3865569.66 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3504, pruned_loss=0.1071, over 5689148.33 frames. ], batch size: 155, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:54:02,508 INFO [train.py:968] (0/2) Epoch 5, batch 2250, giga_loss[loss=0.3064, simple_loss=0.366, pruned_loss=0.1234, over 28645.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.349, pruned_loss=0.1059, over 5710126.40 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3685, pruned_loss=0.1118, over 3895588.89 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3473, pruned_loss=0.1055, over 5694300.58 frames. ], batch size: 336, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:54:04,639 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.832e+02 9.552e+02 1.167e+03 1.615e+03 4.604e+03, threshold=2.333e+03, percent-clipped=7.0
+2023-03-02 12:54:23,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8417, 1.0825, 3.8649, 2.9251], device='cuda:0'), covar=tensor([0.1791, 0.2376, 0.0365, 0.0807], device='cuda:0'), in_proj_covar=tensor([0.0553, 0.0516, 0.0718, 0.0590], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 12:54:23,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:54:43,341 INFO [train.py:968] (0/2) Epoch 5, batch 2300, giga_loss[loss=0.2243, simple_loss=0.3058, pruned_loss=0.07141, over 29016.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3475, pruned_loss=0.1054, over 5710111.73 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3693, pruned_loss=0.1119, over 3953588.15 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3451, pruned_loss=0.1047, over 5693510.34 frames. ], batch size: 164, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:55:12,972 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:55:25,089 INFO [train.py:968] (0/2) Epoch 5, batch 2350, giga_loss[loss=0.261, simple_loss=0.3351, pruned_loss=0.09345, over 28308.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.346, pruned_loss=0.1047, over 5705335.02 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3696, pruned_loss=0.1121, over 3972415.36 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3435, pruned_loss=0.104, over 5706330.71 frames. ], batch size: 368, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:55:27,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.377e+02 9.914e+02 1.133e+03 1.644e+03 8.499e+03, threshold=2.267e+03, percent-clipped=8.0
+2023-03-02 12:56:09,560 INFO [train.py:968] (0/2) Epoch 5, batch 2400, giga_loss[loss=0.2458, simple_loss=0.3092, pruned_loss=0.09121, over 28601.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3436, pruned_loss=0.1039, over 5712314.35 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3709, pruned_loss=0.1129, over 3999228.04 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3406, pruned_loss=0.1028, over 5712283.55 frames. ], batch size: 85, lr: 6.68e-03, grad_scale: 8.0
+2023-03-02 12:56:22,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-02 12:56:24,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:56:26,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:56:49,817 INFO [train.py:968] (0/2) Epoch 5, batch 2450, giga_loss[loss=0.2485, simple_loss=0.3201, pruned_loss=0.08846, over 28991.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3412, pruned_loss=0.1024, over 5722286.52 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3716, pruned_loss=0.1134, over 4054547.06 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3375, pruned_loss=0.1009, over 5718168.75 frames. ], batch size: 213, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:56:50,795 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:56:52,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.749e+02 8.967e+02 1.180e+03 1.711e+03 6.827e+03, threshold=2.360e+03, percent-clipped=11.0
+2023-03-02 12:57:13,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:57:28,022 INFO [train.py:968] (0/2) Epoch 5, batch 2500, libri_loss[loss=0.3672, simple_loss=0.4287, pruned_loss=0.1528, over 29689.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3393, pruned_loss=0.101, over 5725940.09 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3726, pruned_loss=0.1136, over 4116166.80 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3347, pruned_loss=0.09925, over 5718893.88 frames. ], batch size: 88, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:57:31,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8539, 1.1563, 3.8548, 3.0819], device='cuda:0'), covar=tensor([0.1855, 0.2351, 0.0365, 0.0711], device='cuda:0'), in_proj_covar=tensor([0.0563, 0.0523, 0.0737, 0.0600], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 12:58:02,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:58:08,803 INFO [train.py:968] (0/2) Epoch 5, batch 2550, giga_loss[loss=0.2936, simple_loss=0.3597, pruned_loss=0.1138, over 28554.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3381, pruned_loss=0.1004, over 5724491.15 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3737, pruned_loss=0.1141, over 4169842.18 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3327, pruned_loss=0.09835, over 5713651.02 frames. ], batch size: 307, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:58:09,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8213, 1.7868, 1.2794, 1.4444], device='cuda:0'), covar=tensor([0.0613, 0.0554, 0.0938, 0.0923], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0447, 0.0508, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 12:58:11,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.047e+02 1.121e+03 1.403e+03 1.842e+03 6.540e+03, threshold=2.806e+03, percent-clipped=13.0
+2023-03-02 12:58:16,828 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-02 12:58:52,070 INFO [train.py:968] (0/2) Epoch 5, batch 2600, giga_loss[loss=0.3143, simple_loss=0.3696, pruned_loss=0.1295, over 27625.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3362, pruned_loss=0.09884, over 5727156.67 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3734, pruned_loss=0.1136, over 4212955.51 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3314, pruned_loss=0.09718, over 5714667.41 frames. ], batch size: 472, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:59:10,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6154, 1.4759, 1.2424, 1.3072], device='cuda:0'), covar=tensor([0.0596, 0.0524, 0.0946, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0448, 0.0506, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 12:59:11,455 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:59:14,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:59:25,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1962, 1.4531, 1.0392, 0.7388], device='cuda:0'), covar=tensor([0.1069, 0.0740, 0.0663, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.1349, 0.1120, 0.1151, 0.1215], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 12:59:31,969 INFO [train.py:968] (0/2) Epoch 5, batch 2650, giga_loss[loss=0.2597, simple_loss=0.3256, pruned_loss=0.09691, over 29026.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3352, pruned_loss=0.09839, over 5730400.87 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3739, pruned_loss=0.1138, over 4252777.79 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3301, pruned_loss=0.09662, over 5718550.73 frames. ], batch size: 128, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:59:35,059 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.048e+02 8.900e+02 1.080e+03 1.410e+03 3.682e+03, threshold=2.160e+03, percent-clipped=2.0
+2023-03-02 12:59:40,148 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:59:59,446 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-02 13:00:16,011 INFO [train.py:968] (0/2) Epoch 5, batch 2700, giga_loss[loss=0.3453, simple_loss=0.3968, pruned_loss=0.1469, over 27556.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3379, pruned_loss=0.1006, over 5719289.22 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3739, pruned_loss=0.1136, over 4283878.24 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3332, pruned_loss=0.09907, over 5708204.10 frames. ], batch size: 472, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 13:00:26,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:00:53,166 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 13:01:01,744 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 13:01:02,728 INFO [train.py:968] (0/2) Epoch 5, batch 2750, libri_loss[loss=0.3303, simple_loss=0.4023, pruned_loss=0.1291, over 29262.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.344, pruned_loss=0.1044, over 5720861.89 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3741, pruned_loss=0.1137, over 4322677.61 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3393, pruned_loss=0.1028, over 5709194.48 frames. ], batch size: 97, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 13:01:06,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.355e+02 1.112e+03 1.406e+03 1.811e+03 3.517e+03, threshold=2.811e+03, percent-clipped=12.0
+2023-03-02 13:01:50,304 INFO [train.py:968] (0/2) Epoch 5, batch 2800, giga_loss[loss=0.3191, simple_loss=0.3781, pruned_loss=0.1301, over 28858.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3519, pruned_loss=0.1103, over 5709077.27 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3742, pruned_loss=0.1138, over 4346227.47 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3478, pruned_loss=0.1089, over 5697359.72 frames. ], batch size: 199, lr: 6.67e-03, grad_scale: 8.0
+2023-03-02 13:02:34,490 INFO [train.py:968] (0/2) Epoch 5, batch 2850, giga_loss[loss=0.3299, simple_loss=0.3919, pruned_loss=0.134, over 28782.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.361, pruned_loss=0.1166, over 5696492.10 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3746, pruned_loss=0.1139, over 4393326.38 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3569, pruned_loss=0.1155, over 5686886.80 frames. ], batch size: 199, lr: 6.67e-03, grad_scale: 8.0
+2023-03-02 13:02:36,051 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:02:36,987 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:02:37,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.284e+02 1.201e+03 1.637e+03 2.401e+03 6.651e+03, threshold=3.274e+03, percent-clipped=14.0
+2023-03-02 13:02:41,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:03:08,036 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:03:23,551 INFO [train.py:968] (0/2) Epoch 5, batch 2900, libri_loss[loss=0.2955, simple_loss=0.3576, pruned_loss=0.1167, over 28145.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3665, pruned_loss=0.1192, over 5680269.62 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3739, pruned_loss=0.1137, over 4429441.71 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3635, pruned_loss=0.1186, over 5668873.88 frames. ], batch size: 62, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:03:41,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:04:06,563 INFO [train.py:968] (0/2) Epoch 5, batch 2950, giga_loss[loss=0.2997, simple_loss=0.3687, pruned_loss=0.1154, over 28852.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3711, pruned_loss=0.1211, over 5684540.20 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3732, pruned_loss=0.1134, over 4463847.75 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3691, pruned_loss=0.1209, over 5672258.11 frames. ], batch size: 199, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:04:10,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.120e+02 1.145e+03 1.447e+03 1.897e+03 5.166e+03, threshold=2.894e+03, percent-clipped=7.0
+2023-03-02 13:04:37,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:04:52,478 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3977, 3.7503, 1.5797, 1.4850], device='cuda:0'), covar=tensor([0.0841, 0.0256, 0.0796, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0464, 0.0306, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 13:04:54,683 INFO [train.py:968] (0/2) Epoch 5, batch 3000, giga_loss[loss=0.3017, simple_loss=0.3742, pruned_loss=0.1146, over 28864.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3765, pruned_loss=0.1244, over 5690890.13 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3732, pruned_loss=0.1135, over 4499322.25 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3749, pruned_loss=0.1244, over 5676366.82 frames. ], batch size: 186, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:04:54,687 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 13:05:03,544 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2496, simple_loss=0.3486, pruned_loss=0.0753, over 944034.00 frames. 
+2023-03-02 13:05:03,545 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 13:05:45,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5520, 2.1857, 1.8850, 1.7280], device='cuda:0'), covar=tensor([0.1840, 0.1833, 0.1276, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0739, 0.0765, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 13:05:45,976 INFO [train.py:968] (0/2) Epoch 5, batch 3050, giga_loss[loss=0.2558, simple_loss=0.3355, pruned_loss=0.08804, over 28896.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3718, pruned_loss=0.121, over 5687655.89 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3732, pruned_loss=0.1136, over 4532915.16 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5672086.87 frames. ], batch size: 145, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:05:50,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.979e+02 1.010e+03 1.259e+03 1.752e+03 4.093e+03, threshold=2.517e+03, percent-clipped=5.0
+2023-03-02 13:05:57,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:05:59,563 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:06:23,285 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:06:28,291 INFO [train.py:968] (0/2) Epoch 5, batch 3100, giga_loss[loss=0.267, simple_loss=0.3445, pruned_loss=0.09473, over 28385.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3669, pruned_loss=0.1172, over 5681708.99 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3734, pruned_loss=0.114, over 4559962.12 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3657, pruned_loss=0.1172, over 5675700.66 frames. ], batch size: 77, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:06:52,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0072, 1.3198, 1.0480, 0.2189], device='cuda:0'), covar=tensor([0.1508, 0.1267, 0.2479, 0.2638], device='cuda:0'), in_proj_covar=tensor([0.1358, 0.1260, 0.1356, 0.1122], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 13:06:58,965 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4988, 2.0099, 1.7708, 1.6529], device='cuda:0'), covar=tensor([0.1496, 0.1657, 0.1140, 0.1178], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0742, 0.0766, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 13:07:11,464 INFO [train.py:968] (0/2) Epoch 5, batch 3150, giga_loss[loss=0.3116, simple_loss=0.3735, pruned_loss=0.1249, over 28216.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3658, pruned_loss=0.1158, over 5677852.84 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3733, pruned_loss=0.114, over 4603989.44 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3646, pruned_loss=0.1158, over 5667855.49 frames. ], batch size: 368, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:07:14,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.325e+02 1.002e+03 1.454e+03 2.331e+03 1.324e+04, threshold=2.908e+03, percent-clipped=20.0
+2023-03-02 13:07:49,555 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7201, 2.2974, 1.9459, 1.8769], device='cuda:0'), covar=tensor([0.0553, 0.0694, 0.0869, 0.0961], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0451, 0.0509, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:07:55,400 INFO [train.py:968] (0/2) Epoch 5, batch 3200, giga_loss[loss=0.3117, simple_loss=0.3796, pruned_loss=0.1219, over 28905.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3668, pruned_loss=0.1163, over 5679499.52 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3727, pruned_loss=0.1137, over 4636479.28 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3661, pruned_loss=0.1166, over 5666095.50 frames. ], batch size: 164, lr: 6.67e-03, grad_scale: 8.0
+2023-03-02 13:08:04,681 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-02 13:08:16,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:08:21,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4338, 3.2097, 1.4981, 1.3197], device='cuda:0'), covar=tensor([0.0891, 0.0314, 0.0827, 0.1368], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0466, 0.0308, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 13:08:32,184 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-02 13:08:36,100 INFO [train.py:968] (0/2) Epoch 5, batch 3250, giga_loss[loss=0.3099, simple_loss=0.3861, pruned_loss=0.1169, over 28848.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3701, pruned_loss=0.1183, over 5686230.36 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3727, pruned_loss=0.1137, over 4674158.53 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3695, pruned_loss=0.1187, over 5669626.31 frames. ], batch size: 174, lr: 6.67e-03, grad_scale: 8.0
+2023-03-02 13:08:40,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.955e+02 1.182e+03 1.370e+03 1.786e+03 3.402e+03, threshold=2.741e+03, percent-clipped=1.0
+2023-03-02 13:09:10,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2355, 2.4050, 1.3398, 1.2875], device='cuda:0'), covar=tensor([0.0879, 0.0340, 0.0763, 0.1309], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0466, 0.0308, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0023, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 13:09:10,754 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-02 13:09:19,764 INFO [train.py:968] (0/2) Epoch 5, batch 3300, giga_loss[loss=0.2692, simple_loss=0.3415, pruned_loss=0.09844, over 28649.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3724, pruned_loss=0.1197, over 5699369.19 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3728, pruned_loss=0.1138, over 4712989.18 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3717, pruned_loss=0.1201, over 5682411.97 frames. ], batch size: 66, lr: 6.67e-03, grad_scale: 8.0
+2023-03-02 13:09:20,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2834, 1.4720, 1.2663, 1.4250], device='cuda:0'), covar=tensor([0.0785, 0.0336, 0.0321, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0200, 0.0126, 0.0131, 0.0236], device='cuda:0'), out_proj_covar=tensor([0.0054, 0.0039, 0.0035, 0.0060], device='cuda:0')
+2023-03-02 13:09:46,041 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-02 13:10:03,737 INFO [train.py:968] (0/2) Epoch 5, batch 3350, giga_loss[loss=0.2979, simple_loss=0.3652, pruned_loss=0.1153, over 29033.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3743, pruned_loss=0.1213, over 5696661.82 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3733, pruned_loss=0.114, over 4745995.78 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3734, pruned_loss=0.1216, over 5679393.63 frames. ], batch size: 136, lr: 6.66e-03, grad_scale: 8.0
+2023-03-02 13:10:06,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:10:07,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.384e+02 1.276e+03 1.558e+03 2.198e+03 4.282e+03, threshold=3.116e+03, percent-clipped=12.0
+2023-03-02 13:10:12,212 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:10:20,584 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:10:22,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:10:43,684 INFO [train.py:968] (0/2) Epoch 5, batch 3400, giga_loss[loss=0.3168, simple_loss=0.3747, pruned_loss=0.1295, over 29031.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3758, pruned_loss=0.123, over 5695251.45 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.374, pruned_loss=0.1145, over 4774765.97 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3746, pruned_loss=0.1231, over 5680801.08 frames. ], batch size: 106, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:10:45,389 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:11:13,453 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:11:29,913 INFO [train.py:968] (0/2) Epoch 5, batch 3450, giga_loss[loss=0.3028, simple_loss=0.3711, pruned_loss=0.1172, over 28493.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3755, pruned_loss=0.1233, over 5686971.97 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3736, pruned_loss=0.1143, over 4784804.45 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3749, pruned_loss=0.1236, over 5674881.53 frames. ], batch size: 60, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:11:35,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.685e+02 1.158e+03 1.529e+03 2.238e+03 8.954e+03, threshold=3.058e+03, percent-clipped=6.0
+2023-03-02 13:11:44,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6105, 1.5732, 1.6422, 1.5484], device='cuda:0'), covar=tensor([0.1048, 0.1653, 0.1334, 0.1401], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0742, 0.0642, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 13:12:09,920 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:12:13,206 INFO [train.py:968] (0/2) Epoch 5, batch 3500, giga_loss[loss=0.2899, simple_loss=0.3629, pruned_loss=0.1084, over 28840.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3752, pruned_loss=0.1219, over 5694095.77 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3735, pruned_loss=0.1142, over 4795633.36 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3748, pruned_loss=0.1223, over 5683107.11 frames. ], batch size: 99, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:12:13,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:12:34,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5528, 3.3933, 1.5838, 1.5353], device='cuda:0'), covar=tensor([0.0783, 0.0226, 0.0803, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0466, 0.0306, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0023, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 13:12:36,940 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:12:39,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1184, 2.9176, 2.8407, 1.2987], device='cuda:0'), covar=tensor([0.0722, 0.0629, 0.0859, 0.2433], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0733, 0.0790, 0.0591], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:12:55,145 INFO [train.py:968] (0/2) Epoch 5, batch 3550, giga_loss[loss=0.2699, simple_loss=0.3503, pruned_loss=0.09473, over 28958.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3747, pruned_loss=0.1204, over 5693005.78 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3736, pruned_loss=0.1141, over 4811501.42 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3743, pruned_loss=0.1208, over 5682173.03 frames. ], batch size: 145, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:12:57,417 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:12:59,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.482e+02 9.947e+02 1.206e+03 1.695e+03 7.268e+03, threshold=2.413e+03, percent-clipped=4.0
+2023-03-02 13:13:39,054 INFO [train.py:968] (0/2) Epoch 5, batch 3600, giga_loss[loss=0.298, simple_loss=0.3709, pruned_loss=0.1125, over 28990.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3748, pruned_loss=0.1198, over 5690455.74 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3734, pruned_loss=0.1142, over 4833789.57 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3747, pruned_loss=0.1203, over 5684823.08 frames. ], batch size: 136, lr: 6.66e-03, grad_scale: 8.0
+2023-03-02 13:13:56,855 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:14:18,717 INFO [train.py:968] (0/2) Epoch 5, batch 3650, giga_loss[loss=0.2911, simple_loss=0.3385, pruned_loss=0.1218, over 23521.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3721, pruned_loss=0.1177, over 5697397.08 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3734, pruned_loss=0.1141, over 4854967.85 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.372, pruned_loss=0.1183, over 5689420.78 frames. ], batch size: 705, lr: 6.66e-03, grad_scale: 8.0
+2023-03-02 13:14:22,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.727e+02 1.060e+03 1.323e+03 1.654e+03 3.863e+03, threshold=2.647e+03, percent-clipped=8.0
+2023-03-02 13:15:00,849 INFO [train.py:968] (0/2) Epoch 5, batch 3700, giga_loss[loss=0.2906, simple_loss=0.3626, pruned_loss=0.1093, over 28915.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3691, pruned_loss=0.1164, over 5700412.65 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3727, pruned_loss=0.1136, over 4885891.34 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3695, pruned_loss=0.1172, over 5688769.73 frames. ], batch size: 145, lr: 6.66e-03, grad_scale: 8.0
+2023-03-02 13:15:28,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:15:40,748 INFO [train.py:968] (0/2) Epoch 5, batch 3750, giga_loss[loss=0.3308, simple_loss=0.3913, pruned_loss=0.1351, over 28995.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3684, pruned_loss=0.1161, over 5702291.94 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3733, pruned_loss=0.1142, over 4919769.08 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3681, pruned_loss=0.1164, over 5694675.54 frames. ], batch size: 128, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:15:47,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.585e+02 1.100e+03 1.475e+03 2.148e+03 5.981e+03, threshold=2.949e+03, percent-clipped=15.0
+2023-03-02 13:16:26,010 INFO [train.py:968] (0/2) Epoch 5, batch 3800, giga_loss[loss=0.3368, simple_loss=0.3894, pruned_loss=0.1421, over 27936.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.37, pruned_loss=0.1177, over 5692394.05 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3737, pruned_loss=0.1144, over 4935067.41 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3694, pruned_loss=0.1178, over 5690026.17 frames. ], batch size: 412, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:16:30,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3263, 1.9309, 1.5346, 0.5358], device='cuda:0'), covar=tensor([0.2057, 0.1066, 0.1841, 0.2367], device='cuda:0'), in_proj_covar=tensor([0.1340, 0.1241, 0.1328, 0.1103], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 13:16:31,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:17:06,192 INFO [train.py:968] (0/2) Epoch 5, batch 3850, giga_loss[loss=0.3037, simple_loss=0.3725, pruned_loss=0.1174, over 28588.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3697, pruned_loss=0.1172, over 5701321.74 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3736, pruned_loss=0.1145, over 4967480.38 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3692, pruned_loss=0.1173, over 5693947.73 frames. ], batch size: 307, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:17:11,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.484e+02 1.046e+03 1.280e+03 1.702e+03 3.361e+03, threshold=2.560e+03, percent-clipped=3.0
+2023-03-02 13:17:12,610 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-02 13:17:27,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:17:29,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:17:46,150 INFO [train.py:968] (0/2) Epoch 5, batch 3900, giga_loss[loss=0.3228, simple_loss=0.3798, pruned_loss=0.1329, over 27994.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3702, pruned_loss=0.1169, over 5700239.63 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.373, pruned_loss=0.1144, over 4984309.24 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3701, pruned_loss=0.1172, over 5698448.64 frames. ], batch size: 77, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:17:53,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2968, 1.6319, 1.5275, 1.4738], device='cuda:0'), covar=tensor([0.1248, 0.1598, 0.1018, 0.1062], device='cuda:0'), in_proj_covar=tensor([0.0787, 0.0749, 0.0776, 0.0711], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 13:17:55,499 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:18:11,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:18:31,886 INFO [train.py:968] (0/2) Epoch 5, batch 3950, giga_loss[loss=0.2944, simple_loss=0.3641, pruned_loss=0.1123, over 28701.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.37, pruned_loss=0.1163, over 5708270.14 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3734, pruned_loss=0.1147, over 5002453.02 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3696, pruned_loss=0.1163, over 5703481.78 frames. ], batch size: 85, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:18:32,102 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:18:34,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:18:37,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.772e+02 1.119e+03 1.503e+03 2.456e+03 6.817e+03, threshold=3.006e+03, percent-clipped=21.0
+2023-03-02 13:18:41,191 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:18:58,351 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4101, 1.9812, 1.5890, 0.6810], device='cuda:0'), covar=tensor([0.2104, 0.1105, 0.1665, 0.2135], device='cuda:0'), in_proj_covar=tensor([0.1357, 0.1257, 0.1339, 0.1109], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-02 13:19:00,911 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:19:10,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:19:12,958 INFO [train.py:968] (0/2) Epoch 5, batch 4000, giga_loss[loss=0.314, simple_loss=0.382, pruned_loss=0.1231, over 28514.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3696, pruned_loss=0.1165, over 5704395.68 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3735, pruned_loss=0.1148, over 5015345.84 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3692, pruned_loss=0.1165, over 5698261.67 frames. ], batch size: 85, lr: 6.65e-03, grad_scale: 8.0
+2023-03-02 13:19:54,195 INFO [train.py:968] (0/2) Epoch 5, batch 4050, giga_loss[loss=0.3243, simple_loss=0.3784, pruned_loss=0.1351, over 28605.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3676, pruned_loss=0.1158, over 5703939.07 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3732, pruned_loss=0.1146, over 5022261.21 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3674, pruned_loss=0.1159, over 5704279.32 frames. ], batch size: 85, lr: 6.65e-03, grad_scale: 8.0
+2023-03-02 13:19:59,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.984e+02 8.979e+02 1.094e+03 1.452e+03 2.813e+03, threshold=2.189e+03, percent-clipped=0.0
+2023-03-02 13:20:11,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:20:13,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:20:33,344 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-186000.pt
+2023-03-02 13:20:33,657 INFO [train.py:968] (0/2) Epoch 5, batch 4100, giga_loss[loss=0.2979, simple_loss=0.3638, pruned_loss=0.116, over 28443.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3657, pruned_loss=0.1149, over 5705757.27 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3737, pruned_loss=0.1153, over 5049958.58 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3648, pruned_loss=0.1144, over 5705592.48 frames. ], batch size: 85, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:20:37,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:20:37,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7741, 1.6841, 1.7041, 1.6230], device='cuda:0'), covar=tensor([0.1100, 0.1739, 0.1526, 0.1390], device='cuda:0'), in_proj_covar=tensor([0.0437, 0.0734, 0.0635, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 13:21:00,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7598, 4.0659, 1.8890, 1.6263], device='cuda:0'), covar=tensor([0.0781, 0.0222, 0.0732, 0.1232], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0464, 0.0307, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0023, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 13:21:07,535 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:21:10,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:21:15,671 INFO [train.py:968] (0/2) Epoch 5, batch 4150, giga_loss[loss=0.2863, simple_loss=0.3526, pruned_loss=0.11, over 28480.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3632, pruned_loss=0.1137, over 5705333.78 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3741, pruned_loss=0.1156, over 5054833.99 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3621, pruned_loss=0.113, over 5706524.37 frames. ], batch size: 65, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:21:17,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2319, 1.2168, 1.0650, 1.0528], device='cuda:0'), covar=tensor([0.0575, 0.0491, 0.0989, 0.0763], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0446, 0.0507, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:21:18,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4084, 1.7353, 1.0831, 1.3977], device='cuda:0'), covar=tensor([0.0715, 0.0282, 0.0357, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0197, 0.0125, 0.0129, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0054, 0.0039, 0.0035, 0.0059], device='cuda:0')
+2023-03-02 13:21:20,248 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-02 13:21:21,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.176e+02 1.226e+03 1.506e+03 2.199e+03 4.784e+03, threshold=3.011e+03, percent-clipped=25.0
+2023-03-02 13:21:34,695 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:21:57,761 INFO [train.py:968] (0/2) Epoch 5, batch 4200, giga_loss[loss=0.3223, simple_loss=0.3844, pruned_loss=0.1301, over 28304.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3624, pruned_loss=0.1131, over 5711371.02 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.374, pruned_loss=0.1156, over 5080305.48 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3612, pruned_loss=0.1126, over 5706852.22 frames. ], batch size: 368, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:22:40,252 INFO [train.py:968] (0/2) Epoch 5, batch 4250, giga_loss[loss=0.3045, simple_loss=0.3651, pruned_loss=0.122, over 28867.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3618, pruned_loss=0.1132, over 5710931.58 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3744, pruned_loss=0.1157, over 5096796.21 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3603, pruned_loss=0.1126, over 5706831.60 frames. ], batch size: 186, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:22:41,975 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:22:48,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.031e+02 1.074e+03 1.301e+03 1.732e+03 3.581e+03, threshold=2.602e+03, percent-clipped=4.0
+2023-03-02 13:23:05,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4831, 1.9596, 1.8636, 1.7185], device='cuda:0'), covar=tensor([0.1420, 0.1723, 0.1114, 0.1099], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0747, 0.0771, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 13:23:20,249 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186192.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 13:23:25,313 INFO [train.py:968] (0/2) Epoch 5, batch 4300, giga_loss[loss=0.2866, simple_loss=0.3507, pruned_loss=0.1112, over 28967.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3611, pruned_loss=0.1139, over 5708784.09 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3744, pruned_loss=0.1157, over 5108933.51 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3597, pruned_loss=0.1135, over 5702680.13 frames. ], batch size: 164, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:23:33,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1951, 3.9467, 3.8802, 1.9303], device='cuda:0'), covar=tensor([0.0496, 0.0473, 0.0774, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.0827, 0.0736, 0.0797, 0.0596], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:23:54,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:24:06,598 INFO [train.py:968] (0/2) Epoch 5, batch 4350, giga_loss[loss=0.2665, simple_loss=0.3386, pruned_loss=0.09719, over 28927.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3577, pruned_loss=0.112, over 5714283.96 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3745, pruned_loss=0.1157, over 5128740.77 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3562, pruned_loss=0.1115, over 5705108.79 frames. ], batch size: 213, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:24:12,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.331e+02 1.126e+03 1.514e+03 2.127e+03 9.615e+03, threshold=3.027e+03, percent-clipped=15.0
+2023-03-02 13:24:38,803 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-02 13:24:48,342 INFO [train.py:968] (0/2) Epoch 5, batch 4400, giga_loss[loss=0.2791, simple_loss=0.3473, pruned_loss=0.1055, over 28859.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3551, pruned_loss=0.1111, over 5708550.62 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3745, pruned_loss=0.1157, over 5132631.07 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3538, pruned_loss=0.1106, over 5700553.26 frames. ], batch size: 112, lr: 6.65e-03, grad_scale: 8.0
+2023-03-02 13:25:26,598 INFO [train.py:968] (0/2) Epoch 5, batch 4450, giga_loss[loss=0.33, simple_loss=0.3947, pruned_loss=0.1326, over 28278.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3573, pruned_loss=0.112, over 5709755.78 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.375, pruned_loss=0.116, over 5149259.97 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3553, pruned_loss=0.1114, over 5705734.17 frames. ], batch size: 368, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:25:34,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.948e+02 9.184e+02 1.163e+03 1.562e+03 4.279e+03, threshold=2.327e+03, percent-clipped=3.0
+2023-03-02 13:25:49,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4677, 1.7388, 1.7262, 1.6107], device='cuda:0'), covar=tensor([0.1485, 0.1733, 0.1192, 0.1202], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0748, 0.0769, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 13:25:53,134 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:25:56,968 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:26:00,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2395, 1.1448, 1.1517, 0.9907], device='cuda:0'), covar=tensor([0.0569, 0.0507, 0.0927, 0.0770], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0444, 0.0506, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:26:09,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3653, 2.8222, 1.4462, 1.3696], device='cuda:0'), covar=tensor([0.0789, 0.0275, 0.0816, 0.1245], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0475, 0.0311, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 13:26:12,992 INFO [train.py:968] (0/2) Epoch 5, batch 4500, giga_loss[loss=0.2734, simple_loss=0.3373, pruned_loss=0.1047, over 28752.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3603, pruned_loss=0.1139, over 5710173.33 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3755, pruned_loss=0.1163, over 5160360.77 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.358, pruned_loss=0.113, over 5704468.68 frames. ], batch size: 78, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:26:24,994 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:26:38,874 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 13:26:59,517 INFO [train.py:968] (0/2) Epoch 5, batch 4550, giga_loss[loss=0.3028, simple_loss=0.3633, pruned_loss=0.1211, over 29025.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3626, pruned_loss=0.1145, over 5706967.48 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3758, pruned_loss=0.1166, over 5164769.73 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3603, pruned_loss=0.1135, over 5708924.05 frames. ], batch size: 113, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:27:05,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.854e+02 1.027e+03 1.272e+03 1.665e+03 5.311e+03, threshold=2.545e+03, percent-clipped=12.0
+2023-03-02 13:27:23,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1523, 4.8336, 4.7949, 2.0517], device='cuda:0'), covar=tensor([0.0285, 0.0318, 0.0549, 0.2079], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0727, 0.0788, 0.0583], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:27:37,481 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:27:44,573 INFO [train.py:968] (0/2) Epoch 5, batch 4600, giga_loss[loss=0.3941, simple_loss=0.4239, pruned_loss=0.1822, over 26659.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3645, pruned_loss=0.1151, over 5703453.95 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3763, pruned_loss=0.117, over 5178728.75 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3621, pruned_loss=0.1139, over 5702242.64 frames. ], batch size: 555, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:28:08,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:28:31,126 INFO [train.py:968] (0/2) Epoch 5, batch 4650, giga_loss[loss=0.304, simple_loss=0.3711, pruned_loss=0.1185, over 28800.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3633, pruned_loss=0.1137, over 5694202.83 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3763, pruned_loss=0.117, over 5182385.10 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3613, pruned_loss=0.1128, over 5692395.14 frames. ], batch size: 186, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:28:31,931 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:28:36,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.662e+02 9.978e+02 1.221e+03 1.636e+03 3.568e+03, threshold=2.441e+03, percent-clipped=4.0
+2023-03-02 13:28:44,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186567.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 13:29:10,103 INFO [train.py:968] (0/2) Epoch 5, batch 4700, giga_loss[loss=0.2654, simple_loss=0.349, pruned_loss=0.09095, over 28632.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3638, pruned_loss=0.1135, over 5684579.51 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3768, pruned_loss=0.1173, over 5190000.84 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3613, pruned_loss=0.1123, over 5695704.97 frames. ], batch size: 307, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:29:44,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5260, 1.5387, 1.3435, 2.0572], device='cuda:0'), covar=tensor([0.1902, 0.1952, 0.1907, 0.2014], device='cuda:0'), in_proj_covar=tensor([0.1102, 0.0864, 0.0975, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 13:29:55,014 INFO [train.py:968] (0/2) Epoch 5, batch 4750, giga_loss[loss=0.3077, simple_loss=0.3698, pruned_loss=0.1228, over 29123.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3637, pruned_loss=0.1134, over 5693827.41 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3766, pruned_loss=0.1172, over 5195837.83 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3618, pruned_loss=0.1126, over 5701776.15 frames. ], batch size: 113, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:29:57,537 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:30:03,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.606e+02 1.222e+03 1.477e+03 2.022e+03 4.175e+03, threshold=2.954e+03, percent-clipped=12.0
+2023-03-02 13:30:13,985 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:30:15,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6954, 1.6791, 1.7007, 1.5971], device='cuda:0'), covar=tensor([0.0969, 0.1589, 0.1340, 0.1359], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0736, 0.0634, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 13:30:16,004 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:30:37,272 INFO [train.py:968] (0/2) Epoch 5, batch 4800, giga_loss[loss=0.2889, simple_loss=0.3615, pruned_loss=0.1081, over 28965.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3652, pruned_loss=0.1146, over 5699387.59 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3776, pruned_loss=0.1178, over 5209433.65 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3625, pruned_loss=0.1134, over 5704461.82 frames. ], batch size: 155, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:30:38,814 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:30:47,625 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186710.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 13:30:49,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186713.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 13:31:18,105 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186742.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 13:31:23,437 INFO [train.py:968] (0/2) Epoch 5, batch 4850, libri_loss[loss=0.3392, simple_loss=0.401, pruned_loss=0.1388, over 29659.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3683, pruned_loss=0.1168, over 5707791.95 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3779, pruned_loss=0.1179, over 5233397.13 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3656, pruned_loss=0.1156, over 5705304.89 frames. ], batch size: 88, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:31:28,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.358e+02 1.270e+03 1.598e+03 2.149e+03 4.590e+03, threshold=3.196e+03, percent-clipped=12.0
+2023-03-02 13:32:05,896 INFO [train.py:968] (0/2) Epoch 5, batch 4900, giga_loss[loss=0.3149, simple_loss=0.3837, pruned_loss=0.123, over 28567.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3703, pruned_loss=0.1179, over 5707070.19 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3787, pruned_loss=0.1183, over 5239251.95 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3675, pruned_loss=0.1167, over 5706220.25 frames. ], batch size: 60, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:32:18,344 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:32:50,076 INFO [train.py:968] (0/2) Epoch 5, batch 4950, giga_loss[loss=0.3131, simple_loss=0.3759, pruned_loss=0.1252, over 27611.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.371, pruned_loss=0.1179, over 5706744.97 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3785, pruned_loss=0.1181, over 5248371.52 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3688, pruned_loss=0.1171, over 5704340.73 frames. ], batch size: 472, lr: 6.64e-03, grad_scale: 4.0
+2023-03-02 13:32:56,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.146e+02 1.180e+03 1.512e+03 2.141e+03 5.091e+03, threshold=3.025e+03, percent-clipped=5.0
+2023-03-02 13:33:02,634 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:33:02,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:33:26,736 INFO [train.py:968] (0/2) Epoch 5, batch 5000, giga_loss[loss=0.2973, simple_loss=0.368, pruned_loss=0.1133, over 29039.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.371, pruned_loss=0.1173, over 5712068.61 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3786, pruned_loss=0.1181, over 5268740.76 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3689, pruned_loss=0.1167, over 5708597.04 frames. ], batch size: 128, lr: 6.64e-03, grad_scale: 4.0
+2023-03-02 13:33:34,418 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:33:48,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7449, 1.0281, 3.6511, 2.9737], device='cuda:0'), covar=tensor([0.1743, 0.2298, 0.0376, 0.0738], device='cuda:0'), in_proj_covar=tensor([0.0561, 0.0515, 0.0725, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 13:33:49,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:33:53,647 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:34:08,808 INFO [train.py:968] (0/2) Epoch 5, batch 5050, giga_loss[loss=0.3558, simple_loss=0.4135, pruned_loss=0.1491, over 28354.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3727, pruned_loss=0.1188, over 5705451.00 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3788, pruned_loss=0.1181, over 5280830.12 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3707, pruned_loss=0.1182, over 5699849.44 frames. ], batch size: 368, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:34:16,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.200e+02 1.145e+03 1.319e+03 1.832e+03 4.266e+03, threshold=2.638e+03, percent-clipped=2.0
+2023-03-02 13:34:17,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:34:30,021 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:34:48,671 INFO [train.py:968] (0/2) Epoch 5, batch 5100, giga_loss[loss=0.2875, simple_loss=0.3611, pruned_loss=0.107, over 28691.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3718, pruned_loss=0.118, over 5706103.99 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3787, pruned_loss=0.1179, over 5291031.52 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3701, pruned_loss=0.1177, over 5704026.18 frames. ], batch size: 262, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:34:58,564 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:35:01,634 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:35:08,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3024, 1.4562, 1.2127, 1.5329], device='cuda:0'), covar=tensor([0.2051, 0.2004, 0.2034, 0.1995], device='cuda:0'), in_proj_covar=tensor([0.1102, 0.0862, 0.0975, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 13:35:12,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:35:20,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0953, 3.8351, 3.7107, 1.6316], device='cuda:0'), covar=tensor([0.0480, 0.0431, 0.0753, 0.2221], device='cuda:0'), in_proj_covar=tensor([0.0828, 0.0727, 0.0794, 0.0587], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:35:24,594 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:35:32,399 INFO [train.py:968] (0/2) Epoch 5, batch 5150, giga_loss[loss=0.2345, simple_loss=0.3139, pruned_loss=0.07758, over 28807.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.369, pruned_loss=0.1165, over 5705831.86 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3785, pruned_loss=0.1178, over 5303135.85 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3677, pruned_loss=0.1164, over 5700820.08 frames. ], batch size: 112, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:35:39,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.337e+02 1.043e+03 1.323e+03 1.769e+03 3.488e+03, threshold=2.647e+03, percent-clipped=7.0
+2023-03-02 13:35:48,522 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:35:51,167 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:36:15,165 INFO [train.py:968] (0/2) Epoch 5, batch 5200, giga_loss[loss=0.2652, simple_loss=0.3434, pruned_loss=0.09351, over 28702.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3654, pruned_loss=0.115, over 5710063.69 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3786, pruned_loss=0.1178, over 5306128.17 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3642, pruned_loss=0.1149, over 5705232.71 frames. ], batch size: 242, lr: 6.63e-03, grad_scale: 8.0
+2023-03-02 13:36:16,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:36:16,949 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:36:30,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 13:36:57,083 INFO [train.py:968] (0/2) Epoch 5, batch 5250, giga_loss[loss=0.2859, simple_loss=0.3642, pruned_loss=0.1038, over 29039.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3642, pruned_loss=0.1137, over 5713833.73 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3789, pruned_loss=0.118, over 5316798.82 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3627, pruned_loss=0.1133, over 5707661.40 frames. ], batch size: 164, lr: 6.63e-03, grad_scale: 8.0
+2023-03-02 13:37:01,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3440, 1.3362, 1.2031, 1.6197], device='cuda:0'), covar=tensor([0.2407, 0.2281, 0.2259, 0.2151], device='cuda:0'), in_proj_covar=tensor([0.1111, 0.0866, 0.0982, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 13:37:03,818 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.308e+02 1.028e+03 1.273e+03 1.670e+03 3.520e+03, threshold=2.545e+03, percent-clipped=6.0
+2023-03-02 13:37:11,641 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:37:13,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:37:25,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:37:37,226 INFO [train.py:968] (0/2) Epoch 5, batch 5300, giga_loss[loss=0.2785, simple_loss=0.3522, pruned_loss=0.1024, over 29050.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3658, pruned_loss=0.1137, over 5709827.73 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3787, pruned_loss=0.1182, over 5331729.60 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3642, pruned_loss=0.1131, over 5706663.55 frames. ], batch size: 136, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:37:39,710 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:38:04,904 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:38:13,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:38:18,726 INFO [train.py:968] (0/2) Epoch 5, batch 5350, giga_loss[loss=0.3639, simple_loss=0.4133, pruned_loss=0.1573, over 28977.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3679, pruned_loss=0.114, over 5714681.29 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3786, pruned_loss=0.1182, over 5341167.12 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3665, pruned_loss=0.1134, over 5710429.77 frames. ], batch size: 213, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:38:28,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.376e+02 1.062e+03 1.253e+03 1.709e+03 7.222e+03, threshold=2.506e+03, percent-clipped=6.0
+2023-03-02 13:38:49,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:01,266 INFO [train.py:968] (0/2) Epoch 5, batch 5400, libri_loss[loss=0.2608, simple_loss=0.3325, pruned_loss=0.09456, over 29642.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3665, pruned_loss=0.1146, over 5717289.45 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3783, pruned_loss=0.1181, over 5352246.36 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3654, pruned_loss=0.1142, over 5710388.30 frames. ], batch size: 69, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:39:05,238 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:07,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:27,043 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:29,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:29,574 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:42,392 INFO [train.py:968] (0/2) Epoch 5, batch 5450, giga_loss[loss=0.3144, simple_loss=0.376, pruned_loss=0.1264, over 28571.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3657, pruned_loss=0.1161, over 5714736.69 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3785, pruned_loss=0.1184, over 5353347.14 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3645, pruned_loss=0.1154, over 5715438.00 frames. ], batch size: 336, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:39:45,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:50,129 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5117, 4.2741, 4.1270, 1.7738], device='cuda:0'), covar=tensor([0.0421, 0.0436, 0.0776, 0.2213], device='cuda:0'), in_proj_covar=tensor([0.0821, 0.0723, 0.0793, 0.0584], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:39:51,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.164e+02 1.114e+03 1.528e+03 2.110e+03 8.509e+03, threshold=3.055e+03, percent-clipped=11.0
+2023-03-02 13:39:54,166 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:12,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:15,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:19,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4675, 1.6102, 1.4004, 1.6107], device='cuda:0'), covar=tensor([0.2010, 0.2037, 0.2010, 0.2052], device='cuda:0'), in_proj_covar=tensor([0.1110, 0.0864, 0.0977, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 13:40:23,899 INFO [train.py:968] (0/2) Epoch 5, batch 5500, libri_loss[loss=0.3101, simple_loss=0.3763, pruned_loss=0.122, over 29573.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3631, pruned_loss=0.1161, over 5722062.61 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3783, pruned_loss=0.1184, over 5362452.36 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.362, pruned_loss=0.1156, over 5720742.10 frames. ], batch size: 74, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:40:30,021 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6644, 1.5545, 1.5535, 1.4230], device='cuda:0'), covar=tensor([0.0869, 0.1612, 0.1460, 0.1490], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0737, 0.0635, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 13:40:38,977 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:38,992 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:44,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2501, 1.1737, 4.9976, 3.4841], device='cuda:0'), covar=tensor([0.1657, 0.2323, 0.0258, 0.0581], device='cuda:0'), in_proj_covar=tensor([0.0559, 0.0516, 0.0727, 0.0603], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 13:40:45,534 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:48,495 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:48,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:49,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4442, 1.5566, 1.3140, 1.6631], device='cuda:0'), covar=tensor([0.2029, 0.2006, 0.2042, 0.2413], device='cuda:0'), in_proj_covar=tensor([0.1109, 0.0860, 0.0974, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 13:41:05,320 INFO [train.py:968] (0/2) Epoch 5, batch 5550, giga_loss[loss=0.2843, simple_loss=0.352, pruned_loss=0.1083, over 28797.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3606, pruned_loss=0.1151, over 5727371.38 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3787, pruned_loss=0.1185, over 5375330.35 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3591, pruned_loss=0.1145, over 5722911.02 frames. ], batch size: 174, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:41:05,650 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:07,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:13,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.877e+02 1.032e+03 1.378e+03 1.908e+03 6.525e+03, threshold=2.755e+03, percent-clipped=3.0
+2023-03-02 13:41:13,666 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:28,236 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-02 13:41:28,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:29,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:32,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:33,124 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:45,557 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4175, 1.4383, 1.0636, 1.1965], device='cuda:0'), covar=tensor([0.0671, 0.0658, 0.1261, 0.1294], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0455, 0.0508, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:41:45,575 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:47,302 INFO [train.py:968] (0/2) Epoch 5, batch 5600, libri_loss[loss=0.3454, simple_loss=0.4173, pruned_loss=0.1367, over 29478.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3615, pruned_loss=0.1161, over 5720657.62 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3786, pruned_loss=0.1188, over 5396513.28 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3597, pruned_loss=0.1153, over 5710307.40 frames. ], batch size: 85, lr: 6.62e-03, grad_scale: 8.0
+2023-03-02 13:41:47,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:55,943 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:42:11,608 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:42:29,034 INFO [train.py:968] (0/2) Epoch 5, batch 5650, giga_loss[loss=0.3081, simple_loss=0.3572, pruned_loss=0.1295, over 28910.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3575, pruned_loss=0.114, over 5722354.38 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3787, pruned_loss=0.119, over 5409097.27 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3556, pruned_loss=0.113, over 5710089.33 frames. ], batch size: 186, lr: 6.62e-03, grad_scale: 8.0
+2023-03-02 13:42:36,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.071e+02 1.298e+03 1.648e+03 2.470e+03 7.261e+03, threshold=3.297e+03, percent-clipped=17.0
+2023-03-02 13:42:40,483 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-02 13:43:00,838 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-02 13:43:08,009 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-02 13:43:10,296 INFO [train.py:968] (0/2) Epoch 5, batch 5700, giga_loss[loss=0.2453, simple_loss=0.3148, pruned_loss=0.08789, over 28892.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3532, pruned_loss=0.1117, over 5719050.88 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3791, pruned_loss=0.1193, over 5415650.36 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.351, pruned_loss=0.1107, over 5707777.17 frames. ], batch size: 112, lr: 6.62e-03, grad_scale: 4.0
+2023-03-02 13:43:17,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:43:27,089 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:43:29,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:43:50,892 INFO [train.py:968] (0/2) Epoch 5, batch 5750, giga_loss[loss=0.2676, simple_loss=0.3338, pruned_loss=0.1007, over 28497.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.351, pruned_loss=0.1103, over 5711129.21 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3795, pruned_loss=0.1196, over 5413854.83 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3483, pruned_loss=0.109, over 5710641.75 frames. ], batch size: 71, lr: 6.62e-03, grad_scale: 2.0
+2023-03-02 13:43:52,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:44:00,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.397e+02 1.030e+03 1.409e+03 2.244e+03 9.721e+03, threshold=2.818e+03, percent-clipped=8.0
+2023-03-02 13:44:05,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-02 13:44:14,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:44:19,896 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-02 13:44:27,023 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9918, 1.3510, 3.0654, 2.9408], device='cuda:0'), covar=tensor([0.1315, 0.1873, 0.0361, 0.0669], device='cuda:0'), in_proj_covar=tensor([0.0564, 0.0518, 0.0733, 0.0608], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 13:44:32,111 INFO [train.py:968] (0/2) Epoch 5, batch 5800, giga_loss[loss=0.3111, simple_loss=0.3794, pruned_loss=0.1214, over 28411.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.353, pruned_loss=0.1114, over 5700721.28 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3794, pruned_loss=0.1197, over 5414648.57 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3504, pruned_loss=0.1101, over 5705657.67 frames. ], batch size: 60, lr: 6.62e-03, grad_scale: 2.0
+2023-03-02 13:45:04,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1582, 1.6340, 1.4972, 1.4290], device='cuda:0'), covar=tensor([0.1307, 0.1822, 0.1091, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0739, 0.0771, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 13:45:13,978 INFO [train.py:968] (0/2) Epoch 5, batch 5850, libri_loss[loss=0.3043, simple_loss=0.3679, pruned_loss=0.1203, over 28529.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3569, pruned_loss=0.1132, over 5705494.38 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3791, pruned_loss=0.1195, over 5421119.25 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3547, pruned_loss=0.1123, over 5707456.57 frames. ], batch size: 63, lr: 6.62e-03, grad_scale: 2.0
+2023-03-02 13:45:14,959 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:45:17,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:45:23,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.221e+02 1.142e+03 1.432e+03 1.839e+03 6.317e+03, threshold=2.863e+03, percent-clipped=7.0
+2023-03-02 13:45:26,004 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:45:28,714 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.37 vs. limit=5.0
+2023-03-02 13:45:38,361 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-02 13:45:40,531 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:45:50,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:45:55,345 INFO [train.py:968] (0/2) Epoch 5, batch 5900, libri_loss[loss=0.3191, simple_loss=0.3846, pruned_loss=0.1268, over 29531.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3614, pruned_loss=0.115, over 5710123.18 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3794, pruned_loss=0.1196, over 5432791.08 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3587, pruned_loss=0.114, over 5709800.58 frames. ], batch size: 79, lr: 6.62e-03, grad_scale: 2.0
+2023-03-02 13:46:02,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:46:07,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3240, 1.3836, 1.3847, 1.3591], device='cuda:0'), covar=tensor([0.1281, 0.1965, 0.1183, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0739, 0.0772, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 13:46:15,411 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:46:20,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:46:23,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6377, 4.4004, 4.3363, 1.9080], device='cuda:0'), covar=tensor([0.0374, 0.0397, 0.0692, 0.1977], device='cuda:0'), in_proj_covar=tensor([0.0824, 0.0722, 0.0790, 0.0587], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:46:42,895 INFO [train.py:968] (0/2) Epoch 5, batch 5950, giga_loss[loss=0.3059, simple_loss=0.362, pruned_loss=0.1249, over 23983.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3653, pruned_loss=0.1169, over 5702298.99 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3796, pruned_loss=0.1197, over 5434209.65 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3629, pruned_loss=0.116, over 5703274.87 frames. ], batch size: 705, lr: 6.62e-03, grad_scale: 2.0
+2023-03-02 13:46:46,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:46:52,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.503e+02 1.047e+03 1.335e+03 1.789e+03 3.899e+03, threshold=2.670e+03, percent-clipped=7.0
+2023-03-02 13:46:59,931 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2104, 1.2066, 1.0950, 1.1121], device='cuda:0'), covar=tensor([0.0566, 0.0435, 0.0923, 0.0688], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0455, 0.0507, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:47:26,054 INFO [train.py:968] (0/2) Epoch 5, batch 6000, libri_loss[loss=0.3241, simple_loss=0.3959, pruned_loss=0.1261, over 29154.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3683, pruned_loss=0.1182, over 5707001.29 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3799, pruned_loss=0.1197, over 5445223.13 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3658, pruned_loss=0.1174, over 5703725.03 frames. ], batch size: 101, lr: 6.62e-03, grad_scale: 4.0
+2023-03-02 13:47:26,058 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 13:47:35,025 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2474, simple_loss=0.3477, pruned_loss=0.07358, over 944034.00 frames. 
+2023-03-02 13:47:35,026 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 13:47:39,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:47:45,209 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:47:49,436 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:08,920 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:12,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:21,177 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:21,568 INFO [train.py:968] (0/2) Epoch 5, batch 6050, giga_loss[loss=0.2865, simple_loss=0.3632, pruned_loss=0.1048, over 28901.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3743, pruned_loss=0.1242, over 5702416.61 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3791, pruned_loss=0.1193, over 5454283.16 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3728, pruned_loss=0.1239, over 5696375.05 frames. ], batch size: 174, lr: 6.62e-03, grad_scale: 4.0
+2023-03-02 13:48:24,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:32,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.867e+02 1.349e+03 1.857e+03 2.432e+03 7.683e+03, threshold=3.713e+03, percent-clipped=16.0
+2023-03-02 13:48:37,820 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:50,738 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:49:02,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9015, 1.1377, 4.0095, 3.0522], device='cuda:0'), covar=tensor([0.1723, 0.2218, 0.0361, 0.0708], device='cuda:0'), in_proj_covar=tensor([0.0564, 0.0520, 0.0729, 0.0607], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 13:49:08,251 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-188000.pt
+2023-03-02 13:49:08,554 INFO [train.py:968] (0/2) Epoch 5, batch 6100, giga_loss[loss=0.3537, simple_loss=0.405, pruned_loss=0.1512, over 28849.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3806, pruned_loss=0.1295, over 5695832.86 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3796, pruned_loss=0.1196, over 5463106.65 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3789, pruned_loss=0.1293, over 5689634.62 frames. ], batch size: 145, lr: 6.62e-03, grad_scale: 4.0
+2023-03-02 13:49:23,424 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3694, 2.9461, 1.4274, 1.3722], device='cuda:0'), covar=tensor([0.0873, 0.0361, 0.0835, 0.1264], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0478, 0.0310, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 13:49:57,050 INFO [train.py:968] (0/2) Epoch 5, batch 6150, giga_loss[loss=0.4126, simple_loss=0.4271, pruned_loss=0.1991, over 23736.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3868, pruned_loss=0.1337, over 5702445.46 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3791, pruned_loss=0.1192, over 5479565.08 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3862, pruned_loss=0.1343, over 5690926.81 frames. ], batch size: 705, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:50:07,708 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.233e+02 1.594e+03 1.915e+03 2.578e+03 5.780e+03, threshold=3.830e+03, percent-clipped=10.0
+2023-03-02 13:50:28,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5554, 2.2212, 1.6462, 0.7122], device='cuda:0'), covar=tensor([0.2185, 0.1127, 0.1772, 0.2525], device='cuda:0'), in_proj_covar=tensor([0.1373, 0.1263, 0.1365, 0.1145], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 13:50:48,046 INFO [train.py:968] (0/2) Epoch 5, batch 6200, giga_loss[loss=0.3856, simple_loss=0.426, pruned_loss=0.1726, over 28594.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3926, pruned_loss=0.1385, over 5701678.67 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3793, pruned_loss=0.1192, over 5485158.58 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3922, pruned_loss=0.1393, over 5691428.15 frames. ], batch size: 307, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:50:56,454 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-02 13:51:06,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5378, 1.8483, 1.8095, 1.6884], device='cuda:0'), covar=tensor([0.1296, 0.1635, 0.1037, 0.1081], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0735, 0.0766, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 13:51:24,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:51:33,443 INFO [train.py:968] (0/2) Epoch 5, batch 6250, giga_loss[loss=0.3951, simple_loss=0.4385, pruned_loss=0.1759, over 28685.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.4004, pruned_loss=0.1459, over 5700329.89 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.379, pruned_loss=0.119, over 5496728.72 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.4008, pruned_loss=0.1474, over 5687401.93 frames. ], batch size: 99, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:51:44,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.415e+02 1.598e+03 2.049e+03 2.972e+03 8.223e+03, threshold=4.098e+03, percent-clipped=16.0
+2023-03-02 13:51:52,939 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-02 13:52:18,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2941, 1.9453, 1.0998, 1.1009], device='cuda:0'), covar=tensor([0.1238, 0.0771, 0.0985, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.1418, 0.1183, 0.1186, 0.1228], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 13:52:20,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0686, 1.2806, 0.9246, 0.3245], device='cuda:0'), covar=tensor([0.1084, 0.1029, 0.1571, 0.1969], device='cuda:0'), in_proj_covar=tensor([0.1364, 0.1261, 0.1354, 0.1141], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 13:52:22,638 INFO [train.py:968] (0/2) Epoch 5, batch 6300, giga_loss[loss=0.3541, simple_loss=0.3988, pruned_loss=0.1547, over 28776.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4064, pruned_loss=0.151, over 5693860.00 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3792, pruned_loss=0.119, over 5504226.24 frames. ], giga_tot_loss[loss=0.3561, simple_loss=0.4069, pruned_loss=0.1527, over 5680659.86 frames. ], batch size: 99, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:53:16,128 INFO [train.py:968] (0/2) Epoch 5, batch 6350, giga_loss[loss=0.3391, simple_loss=0.4007, pruned_loss=0.1388, over 28988.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.4065, pruned_loss=0.152, over 5680538.01 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3788, pruned_loss=0.1189, over 5516307.15 frames. ], giga_tot_loss[loss=0.3584, simple_loss=0.4081, pruned_loss=0.1544, over 5663987.13 frames. ], batch size: 155, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:53:27,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.639e+03 2.124e+03 3.121e+03 6.154e+03, threshold=4.247e+03, percent-clipped=9.0
+2023-03-02 13:53:45,686 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:53:49,241 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:53:52,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:53:53,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:53:57,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:54:05,922 INFO [train.py:968] (0/2) Epoch 5, batch 6400, libri_loss[loss=0.2618, simple_loss=0.3329, pruned_loss=0.0953, over 27152.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.4081, pruned_loss=0.1542, over 5680215.12 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3781, pruned_loss=0.1185, over 5527503.22 frames. ], giga_tot_loss[loss=0.3634, simple_loss=0.4112, pruned_loss=0.1578, over 5662636.48 frames. ], batch size: 60, lr: 6.61e-03, grad_scale: 8.0
+2023-03-02 13:54:20,834 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:54:59,295 INFO [train.py:968] (0/2) Epoch 5, batch 6450, giga_loss[loss=0.3539, simple_loss=0.3972, pruned_loss=0.1553, over 28765.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4116, pruned_loss=0.1584, over 5673607.43 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3778, pruned_loss=0.1185, over 5536826.75 frames. ], giga_tot_loss[loss=0.3696, simple_loss=0.4149, pruned_loss=0.1621, over 5654981.01 frames. ], batch size: 99, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:55:02,291 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4191, 2.1606, 1.6721, 0.6760], device='cuda:0'), covar=tensor([0.2217, 0.1189, 0.1938, 0.2517], device='cuda:0'), in_proj_covar=tensor([0.1367, 0.1258, 0.1352, 0.1138], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 13:55:14,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.771e+03 2.212e+03 2.913e+03 8.856e+03, threshold=4.425e+03, percent-clipped=10.0
+2023-03-02 13:55:49,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2716, 1.7160, 1.2088, 1.5025], device='cuda:0'), covar=tensor([0.0747, 0.0323, 0.0332, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0197, 0.0126, 0.0130, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0054, 0.0039, 0.0036, 0.0060], device='cuda:0')
+2023-03-02 13:55:55,444 INFO [train.py:968] (0/2) Epoch 5, batch 6500, giga_loss[loss=0.4394, simple_loss=0.4587, pruned_loss=0.2101, over 28295.00 frames. ], tot_loss[loss=0.3709, simple_loss=0.4161, pruned_loss=0.1629, over 5661424.47 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3779, pruned_loss=0.1187, over 5540363.63 frames. ], giga_tot_loss[loss=0.3756, simple_loss=0.419, pruned_loss=0.1661, over 5644498.20 frames. ], batch size: 368, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:56:18,125 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:22,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:26,287 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:28,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:29,275 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:33,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:39,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8422, 5.6181, 5.4349, 2.8127], device='cuda:0'), covar=tensor([0.0357, 0.0396, 0.0752, 0.1579], device='cuda:0'), in_proj_covar=tensor([0.0844, 0.0748, 0.0817, 0.0600], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 13:56:46,095 INFO [train.py:968] (0/2) Epoch 5, batch 6550, giga_loss[loss=0.34, simple_loss=0.3948, pruned_loss=0.1426, over 28922.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4147, pruned_loss=0.1623, over 5649620.39 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3779, pruned_loss=0.1187, over 5537599.81 frames. ], giga_tot_loss[loss=0.3743, simple_loss=0.4177, pruned_loss=0.1655, over 5641483.20 frames. ], batch size: 227, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:56:51,911 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:59,268 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:59,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.367e+02 1.648e+03 2.153e+03 3.442e+03 1.059e+04, threshold=4.305e+03, percent-clipped=17.0
+2023-03-02 13:57:02,260 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:57:13,424 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-02 13:57:36,289 INFO [train.py:968] (0/2) Epoch 5, batch 6600, libri_loss[loss=0.2883, simple_loss=0.3519, pruned_loss=0.1123, over 29478.00 frames. ], tot_loss[loss=0.3705, simple_loss=0.4146, pruned_loss=0.1632, over 5648046.89 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.378, pruned_loss=0.1188, over 5547284.93 frames. ], giga_tot_loss[loss=0.3759, simple_loss=0.4179, pruned_loss=0.167, over 5635546.75 frames. ], batch size: 70, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:58:26,002 INFO [train.py:968] (0/2) Epoch 5, batch 6650, giga_loss[loss=0.3445, simple_loss=0.3989, pruned_loss=0.1451, over 28968.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.412, pruned_loss=0.1603, over 5642386.69 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.378, pruned_loss=0.1186, over 5551018.22 frames. ], giga_tot_loss[loss=0.3724, simple_loss=0.4156, pruned_loss=0.1646, over 5631393.61 frames. ], batch size: 145, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:58:40,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.731e+02 1.663e+03 2.022e+03 2.870e+03 8.480e+03, threshold=4.043e+03, percent-clipped=8.0
+2023-03-02 13:59:19,585 INFO [train.py:968] (0/2) Epoch 5, batch 6700, giga_loss[loss=0.4012, simple_loss=0.4438, pruned_loss=0.1793, over 28834.00 frames. ], tot_loss[loss=0.3655, simple_loss=0.4122, pruned_loss=0.1594, over 5649830.02 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3778, pruned_loss=0.1185, over 5552830.13 frames. ], giga_tot_loss[loss=0.3707, simple_loss=0.4153, pruned_loss=0.163, over 5640010.00 frames. ], batch size: 199, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:59:30,927 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8755, 1.1038, 3.7809, 3.0889], device='cuda:0'), covar=tensor([0.1682, 0.2212, 0.0360, 0.0732], device='cuda:0'), in_proj_covar=tensor([0.0571, 0.0527, 0.0742, 0.0613], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 13:59:37,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4286, 1.9495, 1.3422, 1.5687], device='cuda:0'), covar=tensor([0.0790, 0.0279, 0.0324, 0.0830], device='cuda:0'), in_proj_covar=tensor([0.0197, 0.0127, 0.0130, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0054, 0.0040, 0.0036, 0.0060], device='cuda:0')
+2023-03-02 13:59:57,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4471, 1.6443, 1.3412, 1.5354], device='cuda:0'), covar=tensor([0.0793, 0.0310, 0.0333, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0127, 0.0130, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0054, 0.0040, 0.0036, 0.0060], device='cuda:0')
+2023-03-02 14:00:11,632 INFO [train.py:968] (0/2) Epoch 5, batch 6750, giga_loss[loss=0.4524, simple_loss=0.4554, pruned_loss=0.2247, over 26498.00 frames. ], tot_loss[loss=0.3697, simple_loss=0.4152, pruned_loss=0.1621, over 5631276.57 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.378, pruned_loss=0.1187, over 5548103.91 frames. ], giga_tot_loss[loss=0.3743, simple_loss=0.4179, pruned_loss=0.1653, over 5628792.26 frames. ], batch size: 555, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:00:22,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.646e+03 2.188e+03 3.062e+03 8.589e+03, threshold=4.376e+03, percent-clipped=13.0
+2023-03-02 14:00:36,160 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=188676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:01:00,378 INFO [train.py:968] (0/2) Epoch 5, batch 6800, libri_loss[loss=0.3111, simple_loss=0.382, pruned_loss=0.1201, over 29493.00 frames. ], tot_loss[loss=0.3657, simple_loss=0.4125, pruned_loss=0.1594, over 5624443.94 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3782, pruned_loss=0.1188, over 5551778.89 frames. ], giga_tot_loss[loss=0.3715, simple_loss=0.4159, pruned_loss=0.1635, over 5620940.98 frames. ], batch size: 85, lr: 6.60e-03, grad_scale: 8.0
+2023-03-02 14:01:17,057 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-02 14:01:25,898 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 14:01:48,514 INFO [train.py:968] (0/2) Epoch 5, batch 6850, giga_loss[loss=0.3017, simple_loss=0.3815, pruned_loss=0.1109, over 28963.00 frames. ], tot_loss[loss=0.3607, simple_loss=0.41, pruned_loss=0.1557, over 5633903.67 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3785, pruned_loss=0.1191, over 5557188.29 frames. ], giga_tot_loss[loss=0.3665, simple_loss=0.4134, pruned_loss=0.1598, over 5628417.80 frames. ], batch size: 164, lr: 6.60e-03, grad_scale: 8.0
+2023-03-02 14:02:02,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8167, 1.7519, 1.2571, 1.5595], device='cuda:0'), covar=tensor([0.0663, 0.0657, 0.0964, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0456, 0.0504, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 14:02:02,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.815e+02 1.365e+03 1.903e+03 2.270e+03 4.901e+03, threshold=3.805e+03, percent-clipped=5.0
+2023-03-02 14:02:37,759 INFO [train.py:968] (0/2) Epoch 5, batch 6900, giga_loss[loss=0.3913, simple_loss=0.4204, pruned_loss=0.1811, over 26699.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.4073, pruned_loss=0.1527, over 5641235.49 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3784, pruned_loss=0.119, over 5559647.70 frames. ], giga_tot_loss[loss=0.3617, simple_loss=0.4105, pruned_loss=0.1565, over 5635821.52 frames. ], batch size: 555, lr: 6.60e-03, grad_scale: 8.0
+2023-03-02 14:02:54,490 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 14:03:08,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2112, 1.4151, 1.1161, 0.7828], device='cuda:0'), covar=tensor([0.0849, 0.0824, 0.0557, 0.0882], device='cuda:0'), in_proj_covar=tensor([0.1416, 0.1194, 0.1175, 0.1246], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 14:03:29,140 INFO [train.py:968] (0/2) Epoch 5, batch 6950, giga_loss[loss=0.3186, simple_loss=0.3873, pruned_loss=0.125, over 28613.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.403, pruned_loss=0.1487, over 5649945.60 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3782, pruned_loss=0.119, over 5560410.46 frames. ], giga_tot_loss[loss=0.3548, simple_loss=0.4057, pruned_loss=0.1519, over 5645405.52 frames. ], batch size: 307, lr: 6.60e-03, grad_scale: 8.0
+2023-03-02 14:03:41,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.594e+02 1.545e+03 2.011e+03 2.912e+03 6.036e+03, threshold=4.021e+03, percent-clipped=13.0
+2023-03-02 14:03:54,202 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=188874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:04:18,885 INFO [train.py:968] (0/2) Epoch 5, batch 7000, giga_loss[loss=0.3248, simple_loss=0.384, pruned_loss=0.1327, over 28369.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.4002, pruned_loss=0.147, over 5648374.63 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3783, pruned_loss=0.1192, over 5565402.98 frames. ], giga_tot_loss[loss=0.3511, simple_loss=0.4027, pruned_loss=0.1497, over 5641486.70 frames. ], batch size: 368, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:04:20,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8928, 4.5324, 1.9101, 1.7498], device='cuda:0'), covar=tensor([0.0749, 0.0257, 0.0741, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0480, 0.0310, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 14:05:07,255 INFO [train.py:968] (0/2) Epoch 5, batch 7050, giga_loss[loss=0.3825, simple_loss=0.4361, pruned_loss=0.1645, over 28860.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.4003, pruned_loss=0.1472, over 5649454.65 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3789, pruned_loss=0.1196, over 5573830.95 frames. ], giga_tot_loss[loss=0.351, simple_loss=0.4024, pruned_loss=0.1498, over 5638349.54 frames. ], batch size: 145, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:05:22,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.884e+02 1.564e+03 1.966e+03 2.610e+03 4.462e+03, threshold=3.931e+03, percent-clipped=2.0
+2023-03-02 14:06:03,984 INFO [train.py:968] (0/2) Epoch 5, batch 7100, giga_loss[loss=0.3602, simple_loss=0.4079, pruned_loss=0.1563, over 28740.00 frames. ], tot_loss[loss=0.346, simple_loss=0.3997, pruned_loss=0.1462, over 5652125.44 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3789, pruned_loss=0.1196, over 5576910.57 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.4015, pruned_loss=0.1485, over 5641308.88 frames. ], batch size: 262, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:06:59,025 INFO [train.py:968] (0/2) Epoch 5, batch 7150, libri_loss[loss=0.311, simple_loss=0.3799, pruned_loss=0.121, over 29561.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.396, pruned_loss=0.1421, over 5657090.12 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.379, pruned_loss=0.1197, over 5581864.15 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3976, pruned_loss=0.1442, over 5644856.16 frames. ], batch size: 77, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:07:00,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:07:12,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.513e+02 1.274e+03 1.681e+03 2.339e+03 7.725e+03, threshold=3.363e+03, percent-clipped=10.0
+2023-03-02 14:07:13,002 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-02 14:07:53,255 INFO [train.py:968] (0/2) Epoch 5, batch 7200, giga_loss[loss=0.3498, simple_loss=0.4185, pruned_loss=0.1406, over 28863.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3963, pruned_loss=0.1396, over 5664545.45 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3787, pruned_loss=0.1194, over 5594002.51 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3984, pruned_loss=0.1423, over 5646435.89 frames. ], batch size: 186, lr: 6.60e-03, grad_scale: 8.0
+2023-03-02 14:08:37,238 INFO [train.py:968] (0/2) Epoch 5, batch 7250, giga_loss[loss=0.3752, simple_loss=0.4218, pruned_loss=0.1643, over 28605.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3992, pruned_loss=0.1406, over 5675877.16 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3789, pruned_loss=0.1197, over 5598417.63 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.4017, pruned_loss=0.1434, over 5661152.31 frames. ], batch size: 78, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:08:46,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3186, 1.4886, 1.3012, 1.2618], device='cuda:0'), covar=tensor([0.2042, 0.2006, 0.1985, 0.1953], device='cuda:0'), in_proj_covar=tensor([0.1115, 0.0871, 0.0983, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 14:08:54,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.615e+02 1.664e+03 2.182e+03 3.178e+03 8.019e+03, threshold=4.364e+03, percent-clipped=24.0
+2023-03-02 14:09:08,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1911, 3.0408, 1.5791, 1.4228], device='cuda:0'), covar=tensor([0.1314, 0.0648, 0.0998, 0.1120], device='cuda:0'), in_proj_covar=tensor([0.1400, 0.1172, 0.1171, 0.1236], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 14:09:24,683 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:09:26,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:09:28,548 INFO [train.py:968] (0/2) Epoch 5, batch 7300, giga_loss[loss=0.3663, simple_loss=0.4071, pruned_loss=0.1627, over 28464.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.4003, pruned_loss=0.1426, over 5662102.30 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3786, pruned_loss=0.1198, over 5596668.69 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.403, pruned_loss=0.1453, over 5653867.70 frames. ], batch size: 71, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:09:47,105 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-02 14:09:56,090 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:10:15,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:10:15,642 INFO [train.py:968] (0/2) Epoch 5, batch 7350, giga_loss[loss=0.3678, simple_loss=0.4113, pruned_loss=0.1622, over 28753.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3998, pruned_loss=0.1425, over 5661285.65 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.379, pruned_loss=0.1199, over 5594937.87 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.4022, pruned_loss=0.1451, over 5658587.35 frames. ], batch size: 243, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:10:27,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.950e+02 1.613e+03 2.350e+03 2.931e+03 8.180e+03, threshold=4.700e+03, percent-clipped=5.0
+2023-03-02 14:10:52,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3518, 2.0635, 1.5707, 1.5548], device='cuda:0'), covar=tensor([0.0789, 0.0260, 0.0304, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0199, 0.0128, 0.0132, 0.0235], device='cuda:0'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0061], device='cuda:0')
+2023-03-02 14:11:03,663 INFO [train.py:968] (0/2) Epoch 5, batch 7400, giga_loss[loss=0.3154, simple_loss=0.3782, pruned_loss=0.1264, over 28649.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3976, pruned_loss=0.142, over 5672433.07 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3788, pruned_loss=0.1199, over 5603787.28 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.4003, pruned_loss=0.1449, over 5664235.98 frames. ], batch size: 92, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:11:14,185 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-02 14:11:24,754 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:11:26,150 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:11:48,315 INFO [train.py:968] (0/2) Epoch 5, batch 7450, giga_loss[loss=0.2755, simple_loss=0.3481, pruned_loss=0.1014, over 28345.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3965, pruned_loss=0.1426, over 5672967.17 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3789, pruned_loss=0.12, over 5614071.41 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3992, pruned_loss=0.1455, over 5659013.88 frames. ], batch size: 60, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:12:02,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.830e+02 1.666e+03 2.221e+03 3.314e+03 8.915e+03, threshold=4.442e+03, percent-clipped=8.0
+2023-03-02 14:12:28,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:12:30,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:12:34,915 INFO [train.py:968] (0/2) Epoch 5, batch 7500, giga_loss[loss=0.3552, simple_loss=0.3989, pruned_loss=0.1558, over 27546.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3953, pruned_loss=0.141, over 5677929.28 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3788, pruned_loss=0.1199, over 5625394.32 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3982, pruned_loss=0.1443, over 5658352.32 frames. ], batch size: 472, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:12:50,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4840, 2.4805, 1.5949, 1.5775], device='cuda:0'), covar=tensor([0.0624, 0.0313, 0.0585, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0484, 0.0312, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 14:12:58,821 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:12:59,558 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:13:22,295 INFO [train.py:968] (0/2) Epoch 5, batch 7550, giga_loss[loss=0.3182, simple_loss=0.3833, pruned_loss=0.1265, over 28452.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3949, pruned_loss=0.14, over 5675560.58 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3783, pruned_loss=0.1197, over 5631135.23 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3982, pruned_loss=0.1434, over 5655747.24 frames. ], batch size: 78, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:13:28,739 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:13:33,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.899e+02 1.322e+03 1.708e+03 2.453e+03 9.480e+03, threshold=3.416e+03, percent-clipped=9.0
+2023-03-02 14:13:37,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3472, 3.3221, 1.3883, 1.3330], device='cuda:0'), covar=tensor([0.0882, 0.0317, 0.0841, 0.1286], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0483, 0.0311, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 14:13:39,019 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:13:59,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1366, 1.6691, 1.4777, 1.5417], device='cuda:0'), covar=tensor([0.0591, 0.0757, 0.0845, 0.0979], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0458, 0.0508, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 14:14:05,253 INFO [train.py:968] (0/2) Epoch 5, batch 7600, giga_loss[loss=0.3459, simple_loss=0.4054, pruned_loss=0.1432, over 28782.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.396, pruned_loss=0.1401, over 5674437.44 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3784, pruned_loss=0.1196, over 5632571.15 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3994, pruned_loss=0.1439, over 5658866.54 frames. ], batch size: 284, lr: 6.59e-03, grad_scale: 8.0
+2023-03-02 14:14:38,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2141, 1.2087, 1.0685, 1.0176], device='cuda:0'), covar=tensor([0.0662, 0.0525, 0.0970, 0.0788], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0458, 0.0509, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 14:14:46,466 INFO [train.py:968] (0/2) Epoch 5, batch 7650, giga_loss[loss=0.3567, simple_loss=0.4082, pruned_loss=0.1526, over 28934.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3936, pruned_loss=0.1379, over 5690551.03 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3781, pruned_loss=0.1194, over 5637595.99 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3969, pruned_loss=0.1415, over 5674566.99 frames. ], batch size: 145, lr: 6.59e-03, grad_scale: 8.0
+2023-03-02 14:15:03,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.106e+02 1.357e+03 1.776e+03 2.283e+03 4.034e+03, threshold=3.553e+03, percent-clipped=7.0
+2023-03-02 14:15:34,612 INFO [train.py:968] (0/2) Epoch 5, batch 7700, giga_loss[loss=0.2964, simple_loss=0.3599, pruned_loss=0.1164, over 28353.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3917, pruned_loss=0.138, over 5682235.17 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3775, pruned_loss=0.119, over 5642600.25 frames. ], giga_tot_loss[loss=0.3391, simple_loss=0.3952, pruned_loss=0.1415, over 5666304.12 frames. ], batch size: 78, lr: 6.59e-03, grad_scale: 8.0
+2023-03-02 14:15:58,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-02 14:16:15,741 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:16:22,321 INFO [train.py:968] (0/2) Epoch 5, batch 7750, giga_loss[loss=0.3104, simple_loss=0.3692, pruned_loss=0.1258, over 29187.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3916, pruned_loss=0.1388, over 5674796.83 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3779, pruned_loss=0.1193, over 5645922.40 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3946, pruned_loss=0.1422, over 5659472.16 frames. ], batch size: 113, lr: 6.59e-03, grad_scale: 8.0
+2023-03-02 14:16:34,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.942e+02 1.388e+03 1.969e+03 2.671e+03 5.532e+03, threshold=3.937e+03, percent-clipped=11.0
+2023-03-02 14:17:03,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:17:05,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:17:06,158 INFO [train.py:968] (0/2) Epoch 5, batch 7800, giga_loss[loss=0.3443, simple_loss=0.4007, pruned_loss=0.1439, over 28578.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.392, pruned_loss=0.1402, over 5675224.84 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3776, pruned_loss=0.1192, over 5656136.07 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3953, pruned_loss=0.1439, over 5654509.59 frames. ], batch size: 336, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:17:48,594 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:17:56,270 INFO [train.py:968] (0/2) Epoch 5, batch 7850, giga_loss[loss=0.3395, simple_loss=0.3902, pruned_loss=0.1445, over 28919.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.391, pruned_loss=0.1403, over 5673609.54 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3775, pruned_loss=0.1191, over 5660333.19 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.394, pruned_loss=0.1437, over 5653680.22 frames. ], batch size: 284, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:18:14,187 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.461e+02 1.679e+03 2.246e+03 3.422e+03 8.833e+03, threshold=4.491e+03, percent-clipped=14.0
+2023-03-02 14:18:39,887 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 14:18:44,421 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:18:45,585 INFO [train.py:968] (0/2) Epoch 5, batch 7900, giga_loss[loss=0.3152, simple_loss=0.3742, pruned_loss=0.1281, over 29059.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3902, pruned_loss=0.1403, over 5663381.19 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3777, pruned_loss=0.1192, over 5661204.52 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3925, pruned_loss=0.143, over 5647071.22 frames. ], batch size: 136, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:18:45,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5827, 1.6172, 1.6170, 1.3938], device='cuda:0'), covar=tensor([0.0966, 0.1400, 0.1376, 0.1305], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0752, 0.0647, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 14:19:14,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:22,201 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:23,819 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:24,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:24,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5978, 2.0131, 1.9564, 1.7735], device='cuda:0'), covar=tensor([0.1559, 0.1728, 0.1138, 0.1163], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0746, 0.0773, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 14:19:27,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:28,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:30,310 INFO [train.py:968] (0/2) Epoch 5, batch 7950, giga_loss[loss=0.307, simple_loss=0.3797, pruned_loss=0.1171, over 28658.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3911, pruned_loss=0.1402, over 5673769.52 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3776, pruned_loss=0.1191, over 5666819.95 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3933, pruned_loss=0.1431, over 5655720.05 frames. ], batch size: 78, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:19:37,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2645, 1.6555, 1.2612, 0.6937], device='cuda:0'), covar=tensor([0.1792, 0.1115, 0.1282, 0.2169], device='cuda:0'), in_proj_covar=tensor([0.1373, 0.1272, 0.1370, 0.1137], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 14:19:47,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.641e+03 2.271e+03 3.176e+03 7.036e+03, threshold=4.542e+03, percent-clipped=9.0
+2023-03-02 14:19:54,033 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:56,014 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:20:14,163 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:20:18,051 INFO [train.py:968] (0/2) Epoch 5, batch 8000, giga_loss[loss=0.3047, simple_loss=0.3817, pruned_loss=0.1139, over 28946.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3923, pruned_loss=0.1401, over 5676165.18 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3779, pruned_loss=0.1192, over 5671749.99 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3941, pruned_loss=0.1428, over 5657254.51 frames. ], batch size: 136, lr: 6.58e-03, grad_scale: 8.0
+2023-03-02 14:20:38,953 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-02 14:20:57,428 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:20:59,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:06,504 INFO [train.py:968] (0/2) Epoch 5, batch 8050, giga_loss[loss=0.3146, simple_loss=0.3822, pruned_loss=0.1235, over 28963.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3932, pruned_loss=0.14, over 5681219.86 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.378, pruned_loss=0.1193, over 5673314.28 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3948, pruned_loss=0.1423, over 5664954.82 frames. ], batch size: 227, lr: 6.58e-03, grad_scale: 8.0
+2023-03-02 14:21:18,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.760e+02 1.821e+03 2.331e+03 3.347e+03 8.564e+03, threshold=4.662e+03, percent-clipped=11.0
+2023-03-02 14:21:27,117 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:28,767 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:30,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:42,420 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:44,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:51,649 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-190000.pt
+2023-03-02 14:21:52,875 INFO [train.py:968] (0/2) Epoch 5, batch 8100, giga_loss[loss=0.3337, simple_loss=0.3905, pruned_loss=0.1385, over 28596.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3926, pruned_loss=0.1388, over 5690457.62 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3778, pruned_loss=0.1191, over 5678455.29 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3945, pruned_loss=0.1416, over 5673203.25 frames. ], batch size: 307, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:21:56,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5631, 2.1596, 1.5048, 0.7026], device='cuda:0'), covar=tensor([0.3234, 0.1688, 0.1660, 0.2928], device='cuda:0'), in_proj_covar=tensor([0.1377, 0.1279, 0.1381, 0.1142], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 14:21:57,777 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:22:07,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:22:14,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:22:25,262 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:22:39,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7897, 4.4256, 1.8778, 1.7849], device='cuda:0'), covar=tensor([0.0827, 0.0244, 0.0754, 0.1201], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0480, 0.0310, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 14:22:42,096 INFO [train.py:968] (0/2) Epoch 5, batch 8150, giga_loss[loss=0.3295, simple_loss=0.3918, pruned_loss=0.1336, over 28878.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3923, pruned_loss=0.1391, over 5693242.27 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3776, pruned_loss=0.119, over 5683480.44 frames. ], giga_tot_loss[loss=0.3391, simple_loss=0.3944, pruned_loss=0.1419, over 5675470.40 frames. ], batch size: 227, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:23:01,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.157e+02 1.507e+03 2.031e+03 3.070e+03 8.645e+03, threshold=4.061e+03, percent-clipped=6.0
+2023-03-02 14:23:30,647 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:23:32,422 INFO [train.py:968] (0/2) Epoch 5, batch 8200, libri_loss[loss=0.3033, simple_loss=0.3781, pruned_loss=0.1142, over 29570.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3964, pruned_loss=0.1438, over 5680060.97 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3775, pruned_loss=0.1189, over 5688725.69 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.3985, pruned_loss=0.1466, over 5660924.52 frames. ], batch size: 77, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:23:49,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:23:55,121 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:24:05,107 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-02 14:24:22,323 INFO [train.py:968] (0/2) Epoch 5, batch 8250, giga_loss[loss=0.3271, simple_loss=0.3835, pruned_loss=0.1354, over 28938.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3963, pruned_loss=0.1452, over 5671149.74 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3774, pruned_loss=0.1187, over 5692996.86 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.3987, pruned_loss=0.1483, over 5651686.53 frames. ], batch size: 213, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:24:32,707 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:24:34,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:24:40,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.590e+02 1.617e+03 1.981e+03 2.619e+03 5.820e+03, threshold=3.963e+03, percent-clipped=6.0
+2023-03-02 14:25:06,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:25:15,587 INFO [train.py:968] (0/2) Epoch 5, batch 8300, giga_loss[loss=0.3748, simple_loss=0.4171, pruned_loss=0.1662, over 28566.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3977, pruned_loss=0.1473, over 5679674.07 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.377, pruned_loss=0.1185, over 5696326.55 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.4002, pruned_loss=0.1503, over 5661228.71 frames. ], batch size: 85, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:25:48,548 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-02 14:26:08,032 INFO [train.py:968] (0/2) Epoch 5, batch 8350, giga_loss[loss=0.3542, simple_loss=0.4033, pruned_loss=0.1526, over 28605.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.3998, pruned_loss=0.1493, over 5670237.18 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.377, pruned_loss=0.1184, over 5699525.81 frames. ], giga_tot_loss[loss=0.3533, simple_loss=0.4021, pruned_loss=0.1523, over 5652593.37 frames. ], batch size: 336, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:26:15,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:26:18,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:26:22,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.616e+03 2.096e+03 3.153e+03 7.651e+03, threshold=4.192e+03, percent-clipped=11.0
+2023-03-02 14:26:26,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:26:46,551 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:26:55,024 INFO [train.py:968] (0/2) Epoch 5, batch 8400, giga_loss[loss=0.3419, simple_loss=0.3985, pruned_loss=0.1427, over 28733.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.3987, pruned_loss=0.1484, over 5675387.73 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3768, pruned_loss=0.1182, over 5697800.52 frames. ], giga_tot_loss[loss=0.3515, simple_loss=0.4008, pruned_loss=0.1511, over 5662492.41 frames. ], batch size: 284, lr: 6.58e-03, grad_scale: 8.0
+2023-03-02 14:27:40,001 INFO [train.py:968] (0/2) Epoch 5, batch 8450, giga_loss[loss=0.2989, simple_loss=0.3728, pruned_loss=0.1125, over 28410.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3965, pruned_loss=0.1442, over 5688188.32 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3769, pruned_loss=0.1181, over 5703018.26 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3987, pruned_loss=0.1471, over 5672998.80 frames. ], batch size: 65, lr: 6.57e-03, grad_scale: 8.0
+2023-03-02 14:27:55,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.530e+02 1.412e+03 1.732e+03 2.336e+03 9.132e+03, threshold=3.464e+03, percent-clipped=5.0
+2023-03-02 14:28:24,036 INFO [train.py:968] (0/2) Epoch 5, batch 8500, giga_loss[loss=0.2955, simple_loss=0.3683, pruned_loss=0.1113, over 28983.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3929, pruned_loss=0.141, over 5687448.21 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3769, pruned_loss=0.118, over 5709557.47 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3952, pruned_loss=0.1442, over 5668440.89 frames. ], batch size: 164, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:28:24,316 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3040, 1.3132, 1.0888, 1.4037], device='cuda:0'), covar=tensor([0.0799, 0.0342, 0.0371, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0198, 0.0127, 0.0131, 0.0234], device='cuda:0'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0061], device='cuda:0')
+2023-03-02 14:28:30,571 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:28:34,784 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:28:37,924 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:29:07,636 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:29:12,855 INFO [train.py:968] (0/2) Epoch 5, batch 8550, giga_loss[loss=0.363, simple_loss=0.4074, pruned_loss=0.1593, over 28645.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3903, pruned_loss=0.1395, over 5677681.25 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.377, pruned_loss=0.118, over 5709677.96 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3923, pruned_loss=0.1424, over 5662162.34 frames. ], batch size: 262, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:29:17,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6552, 1.5813, 1.2583, 1.3642], device='cuda:0'), covar=tensor([0.0634, 0.0618, 0.0991, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0455, 0.0506, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 14:29:21,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4117, 2.1120, 1.6376, 0.5620], device='cuda:0'), covar=tensor([0.1954, 0.1229, 0.1799, 0.2513], device='cuda:0'), in_proj_covar=tensor([0.1367, 0.1282, 0.1384, 0.1154], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 14:29:28,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.381e+02 1.568e+03 2.105e+03 3.099e+03 8.214e+03, threshold=4.209e+03, percent-clipped=17.0
+2023-03-02 14:29:33,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:29:54,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:29:57,636 INFO [train.py:968] (0/2) Epoch 5, batch 8600, giga_loss[loss=0.3376, simple_loss=0.3917, pruned_loss=0.1418, over 28828.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3896, pruned_loss=0.1396, over 5685821.14 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3771, pruned_loss=0.1182, over 5713359.65 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3912, pruned_loss=0.1421, over 5669725.80 frames. ], batch size: 112, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:30:31,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190532.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 14:30:44,709 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:30:51,584 INFO [train.py:968] (0/2) Epoch 5, batch 8650, giga_loss[loss=0.3218, simple_loss=0.3769, pruned_loss=0.1334, over 28566.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3908, pruned_loss=0.1404, over 5684613.67 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3771, pruned_loss=0.118, over 5717256.63 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3926, pruned_loss=0.1432, over 5667344.23 frames. ], batch size: 85, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:30:52,699 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:30:56,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:31:12,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.731e+02 1.530e+03 2.043e+03 2.869e+03 6.644e+03, threshold=4.086e+03, percent-clipped=9.0
+2023-03-02 14:31:24,479 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:31:36,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8700, 3.6855, 3.5561, 1.6460], device='cuda:0'), covar=tensor([0.0584, 0.0615, 0.0924, 0.1983], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0766, 0.0823, 0.0597], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 14:31:42,394 INFO [train.py:968] (0/2) Epoch 5, batch 8700, giga_loss[loss=0.376, simple_loss=0.428, pruned_loss=0.162, over 28785.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3952, pruned_loss=0.1423, over 5685181.04 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3765, pruned_loss=0.1177, over 5718786.10 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3972, pruned_loss=0.145, over 5669940.17 frames. ], batch size: 186, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:31:57,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:32:02,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:32:20,828 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:32:22,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:32:27,638 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:32:31,100 INFO [train.py:968] (0/2) Epoch 5, batch 8750, giga_loss[loss=0.3396, simple_loss=0.3832, pruned_loss=0.148, over 23572.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3985, pruned_loss=0.1421, over 5681250.06 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3771, pruned_loss=0.1183, over 5723733.21 frames. ], giga_tot_loss[loss=0.3444, simple_loss=0.4001, pruned_loss=0.1444, over 5663481.96 frames. ], batch size: 705, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:32:48,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.868e+02 1.427e+03 1.919e+03 2.526e+03 9.220e+03, threshold=3.837e+03, percent-clipped=4.0
+2023-03-02 14:32:51,386 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:33:14,592 INFO [train.py:968] (0/2) Epoch 5, batch 8800, giga_loss[loss=0.3353, simple_loss=0.3979, pruned_loss=0.1364, over 28685.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3995, pruned_loss=0.1423, over 5688548.64 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3776, pruned_loss=0.1187, over 5721799.48 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.4014, pruned_loss=0.1449, over 5673996.02 frames. ], batch size: 242, lr: 6.57e-03, grad_scale: 8.0
+2023-03-02 14:33:24,624 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:33:58,527 INFO [train.py:968] (0/2) Epoch 5, batch 8850, giga_loss[loss=0.3095, simple_loss=0.3713, pruned_loss=0.1239, over 28809.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.4004, pruned_loss=0.1432, over 5694706.82 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3773, pruned_loss=0.1185, over 5727397.19 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.4029, pruned_loss=0.1461, over 5676990.40 frames. ], batch size: 99, lr: 6.57e-03, grad_scale: 8.0
+2023-03-02 14:34:14,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.575e+02 1.490e+03 2.057e+03 2.739e+03 7.928e+03, threshold=4.114e+03, percent-clipped=9.0
+2023-03-02 14:34:34,853 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:34:39,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3256, 1.7381, 1.3335, 1.4630], device='cuda:0'), covar=tensor([0.0705, 0.0398, 0.0332, 0.0744], device='cuda:0'), in_proj_covar=tensor([0.0197, 0.0126, 0.0130, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0060], device='cuda:0')
+2023-03-02 14:34:42,700 INFO [train.py:968] (0/2) Epoch 5, batch 8900, giga_loss[loss=0.2929, simple_loss=0.3613, pruned_loss=0.1122, over 28827.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.4015, pruned_loss=0.1446, over 5693308.69 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3778, pruned_loss=0.119, over 5728105.90 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.4038, pruned_loss=0.1474, over 5677376.47 frames. ], batch size: 112, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:35:25,716 INFO [train.py:968] (0/2) Epoch 5, batch 8950, giga_loss[loss=0.4002, simple_loss=0.4373, pruned_loss=0.1815, over 27996.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.4019, pruned_loss=0.1463, over 5681972.39 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3779, pruned_loss=0.1191, over 5721440.87 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.4044, pruned_loss=0.1494, over 5674308.39 frames. ], batch size: 412, lr: 6.57e-03, grad_scale: 1.0
+2023-03-02 14:35:47,556 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.663e+03 2.404e+03 4.353e+03 1.587e+04, threshold=4.809e+03, percent-clipped=27.0
+2023-03-02 14:36:17,688 INFO [train.py:968] (0/2) Epoch 5, batch 9000, giga_loss[loss=0.2985, simple_loss=0.3675, pruned_loss=0.1148, over 28931.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3995, pruned_loss=0.1452, over 5676278.57 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3778, pruned_loss=0.1191, over 5714423.27 frames. ], giga_tot_loss[loss=0.3489, simple_loss=0.4018, pruned_loss=0.1479, over 5675593.19 frames. ], batch size: 145, lr: 6.57e-03, grad_scale: 1.0
+2023-03-02 14:36:17,692 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 14:36:27,677 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2397, simple_loss=0.3441, pruned_loss=0.0677, over 944034.00 frames. 
+2023-03-02 14:36:27,678 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 14:36:32,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190907.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:36:34,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1762, 0.9657, 1.0090, 1.3726], device='cuda:0'), covar=tensor([0.0791, 0.0339, 0.0344, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0126, 0.0130, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0054, 0.0040, 0.0036, 0.0060], device='cuda:0')
+2023-03-02 14:36:41,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:36:59,316 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5818, 1.7044, 1.4201, 1.8934], device='cuda:0'), covar=tensor([0.2123, 0.2060, 0.2053, 0.2322], device='cuda:0'), in_proj_covar=tensor([0.1127, 0.0871, 0.0993, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 14:37:12,372 INFO [train.py:968] (0/2) Epoch 5, batch 9050, giga_loss[loss=0.3059, simple_loss=0.3719, pruned_loss=0.1199, over 28858.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.397, pruned_loss=0.1447, over 5674260.92 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.378, pruned_loss=0.1194, over 5718110.58 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.399, pruned_loss=0.1471, over 5669937.98 frames. ], batch size: 227, lr: 6.56e-03, grad_scale: 1.0
+2023-03-02 14:37:31,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.655e+02 1.440e+03 1.706e+03 2.231e+03 9.466e+03, threshold=3.413e+03, percent-clipped=3.0
+2023-03-02 14:37:45,639 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190984.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:37:59,970 INFO [train.py:968] (0/2) Epoch 5, batch 9100, giga_loss[loss=0.2968, simple_loss=0.357, pruned_loss=0.1183, over 28344.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3943, pruned_loss=0.143, over 5675454.72 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3779, pruned_loss=0.1192, over 5720385.44 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3964, pruned_loss=0.1455, over 5669153.21 frames. ], batch size: 65, lr: 6.56e-03, grad_scale: 1.0
+2023-03-02 14:38:03,012 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:38:51,644 INFO [train.py:968] (0/2) Epoch 5, batch 9150, giga_loss[loss=0.313, simple_loss=0.3811, pruned_loss=0.1224, over 28799.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3954, pruned_loss=0.1439, over 5675493.74 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3781, pruned_loss=0.1193, over 5721297.48 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3973, pruned_loss=0.1464, over 5668514.26 frames. ], batch size: 284, lr: 6.56e-03, grad_scale: 1.0
+2023-03-02 14:38:51,924 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191050.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:38:54,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191053.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:39:03,694 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:39:06,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:39:08,984 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:39:12,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.589e+03 2.076e+03 2.887e+03 1.042e+04, threshold=4.153e+03, percent-clipped=16.0
+2023-03-02 14:39:22,427 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191082.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:39:24,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:39:26,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9679, 1.1572, 3.5042, 2.9538], device='cuda:0'), covar=tensor([0.1626, 0.2266, 0.0418, 0.1150], device='cuda:0'), in_proj_covar=tensor([0.0573, 0.0523, 0.0740, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 14:39:32,224 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:39:35,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3619, 1.5297, 1.3077, 1.7345], device='cuda:0'), covar=tensor([0.1918, 0.1868, 0.1861, 0.1996], device='cuda:0'), in_proj_covar=tensor([0.1127, 0.0874, 0.0995, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 14:39:37,197 INFO [train.py:968] (0/2) Epoch 5, batch 9200, giga_loss[loss=0.2882, simple_loss=0.3571, pruned_loss=0.1097, over 28988.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.394, pruned_loss=0.1434, over 5680408.03 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3779, pruned_loss=0.1191, over 5726252.99 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.3961, pruned_loss=0.1461, over 5669371.17 frames. ], batch size: 136, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:39:57,978 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-02 14:40:24,269 INFO [train.py:968] (0/2) Epoch 5, batch 9250, giga_loss[loss=0.3069, simple_loss=0.3672, pruned_loss=0.1234, over 28293.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3931, pruned_loss=0.143, over 5684232.90 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3779, pruned_loss=0.1191, over 5730524.26 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3951, pruned_loss=0.1458, over 5670344.27 frames. ], batch size: 77, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:40:26,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1578, 1.6657, 1.5474, 1.4447], device='cuda:0'), covar=tensor([0.1334, 0.1909, 0.1123, 0.1173], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0757, 0.0779, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 14:40:40,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:40:43,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.748e+02 1.308e+03 2.062e+03 2.911e+03 9.122e+03, threshold=4.124e+03, percent-clipped=7.0
+2023-03-02 14:41:07,350 INFO [train.py:968] (0/2) Epoch 5, batch 9300, giga_loss[loss=0.2844, simple_loss=0.353, pruned_loss=0.1079, over 29030.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3938, pruned_loss=0.1425, over 5692723.40 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3783, pruned_loss=0.1191, over 5732931.18 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3956, pruned_loss=0.1456, over 5677769.33 frames. ], batch size: 128, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:41:09,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2728, 1.4233, 1.2235, 1.3032], device='cuda:0'), covar=tensor([0.2007, 0.1789, 0.1795, 0.1810], device='cuda:0'), in_proj_covar=tensor([0.1126, 0.0880, 0.0998, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 14:41:34,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:41:36,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:41:49,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6194, 1.0713, 2.8741, 2.7363], device='cuda:0'), covar=tensor([0.1588, 0.2095, 0.0505, 0.0760], device='cuda:0'), in_proj_covar=tensor([0.0570, 0.0522, 0.0740, 0.0603], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 14:41:54,932 INFO [train.py:968] (0/2) Epoch 5, batch 9350, giga_loss[loss=0.3812, simple_loss=0.42, pruned_loss=0.1712, over 29069.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3956, pruned_loss=0.1435, over 5688020.78 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3781, pruned_loss=0.1191, over 5737369.12 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.3979, pruned_loss=0.1468, over 5670512.38 frames. ], batch size: 113, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:42:02,481 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:42:12,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.656e+02 1.494e+03 1.815e+03 2.651e+03 5.895e+03, threshold=3.630e+03, percent-clipped=8.0
+2023-03-02 14:42:25,911 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 14:42:42,550 INFO [train.py:968] (0/2) Epoch 5, batch 9400, libri_loss[loss=0.3133, simple_loss=0.3895, pruned_loss=0.1185, over 29386.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3971, pruned_loss=0.1445, over 5683006.04 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3781, pruned_loss=0.1189, over 5742080.18 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3995, pruned_loss=0.1481, over 5663240.66 frames. ], batch size: 92, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:42:50,870 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:42:54,147 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:43:21,409 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:43:28,184 INFO [train.py:968] (0/2) Epoch 5, batch 9450, giga_loss[loss=0.3637, simple_loss=0.4087, pruned_loss=0.1593, over 27984.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3967, pruned_loss=0.1445, over 5687759.07 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3782, pruned_loss=0.119, over 5747515.12 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.399, pruned_loss=0.148, over 5665121.94 frames. ], batch size: 412, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:43:35,592 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191359.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:43:47,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.364e+02 1.617e+03 2.162e+03 3.261e+03 9.639e+03, threshold=4.325e+03, percent-clipped=20.0
+2023-03-02 14:43:54,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:44:15,966 INFO [train.py:968] (0/2) Epoch 5, batch 9500, giga_loss[loss=0.3346, simple_loss=0.4095, pruned_loss=0.1298, over 28977.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.3987, pruned_loss=0.1433, over 5687332.92 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3782, pruned_loss=0.1191, over 5741574.94 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.401, pruned_loss=0.1466, over 5672665.20 frames. ], batch size: 186, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:44:52,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:45:00,233 INFO [train.py:968] (0/2) Epoch 5, batch 9550, giga_loss[loss=0.3416, simple_loss=0.4126, pruned_loss=0.1353, over 28874.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3994, pruned_loss=0.1416, over 5678758.76 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3785, pruned_loss=0.1193, over 5734645.88 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.4016, pruned_loss=0.1448, over 5671155.60 frames. ], batch size: 112, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:45:10,015 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:45:18,811 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.071e+02 1.255e+03 1.596e+03 2.238e+03 6.930e+03, threshold=3.193e+03, percent-clipped=3.0
+2023-03-02 14:45:44,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0608, 1.0935, 4.1457, 3.4023], device='cuda:0'), covar=tensor([0.1627, 0.2265, 0.0359, 0.0523], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0525, 0.0748, 0.0607], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 14:45:47,323 INFO [train.py:968] (0/2) Epoch 5, batch 9600, giga_loss[loss=0.3689, simple_loss=0.4184, pruned_loss=0.1597, over 28825.00 frames. ], tot_loss[loss=0.345, simple_loss=0.4025, pruned_loss=0.1437, over 5679678.10 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.379, pruned_loss=0.1196, over 5739070.84 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4043, pruned_loss=0.1465, over 5668448.43 frames. ], batch size: 243, lr: 6.56e-03, grad_scale: 4.0
+2023-03-02 14:45:50,370 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191502.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:45:54,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191505.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 14:46:08,237 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:46:10,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:46:12,362 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:46:22,141 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191534.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 14:46:35,854 INFO [train.py:968] (0/2) Epoch 5, batch 9650, giga_loss[loss=0.4223, simple_loss=0.4462, pruned_loss=0.1992, over 27871.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4059, pruned_loss=0.1476, over 5683059.57 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3787, pruned_loss=0.1194, over 5743196.97 frames. ], giga_tot_loss[loss=0.3548, simple_loss=0.4083, pruned_loss=0.1507, over 5668939.09 frames. ], batch size: 412, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:46:37,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3079, 1.4667, 1.1203, 1.5967], device='cuda:0'), covar=tensor([0.2083, 0.2015, 0.2106, 0.2018], device='cuda:0'), in_proj_covar=tensor([0.1124, 0.0870, 0.0995, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 14:46:38,515 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:46:54,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.353e+02 1.597e+03 2.247e+03 3.009e+03 6.373e+03, threshold=4.494e+03, percent-clipped=21.0
+2023-03-02 14:47:01,026 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:47:09,535 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:47:12,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:47:22,716 INFO [train.py:968] (0/2) Epoch 5, batch 9700, giga_loss[loss=0.4069, simple_loss=0.4342, pruned_loss=0.1898, over 27551.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4052, pruned_loss=0.1479, over 5681640.71 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3786, pruned_loss=0.1192, over 5746977.16 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4079, pruned_loss=0.1513, over 5665321.52 frames. ], batch size: 472, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:47:29,705 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 14:47:39,785 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:48:09,828 INFO [train.py:968] (0/2) Epoch 5, batch 9750, giga_loss[loss=0.3845, simple_loss=0.4077, pruned_loss=0.1806, over 23505.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.4042, pruned_loss=0.1481, over 5661057.94 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.379, pruned_loss=0.1195, over 5738388.35 frames. ], giga_tot_loss[loss=0.3542, simple_loss=0.4064, pruned_loss=0.151, over 5655000.42 frames. ], batch size: 705, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:48:29,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.714e+02 1.656e+03 2.097e+03 3.000e+03 8.793e+03, threshold=4.194e+03, percent-clipped=10.0
+2023-03-02 14:48:51,471 INFO [train.py:968] (0/2) Epoch 5, batch 9800, giga_loss[loss=0.3027, simple_loss=0.3715, pruned_loss=0.1169, over 28892.00 frames. ], tot_loss[loss=0.347, simple_loss=0.4021, pruned_loss=0.146, over 5663529.90 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3784, pruned_loss=0.1191, over 5741625.88 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4051, pruned_loss=0.1494, over 5653919.00 frames. ], batch size: 174, lr: 6.55e-03, grad_scale: 2.0
+2023-03-02 14:48:59,778 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-02 14:49:39,397 INFO [train.py:968] (0/2) Epoch 5, batch 9850, giga_loss[loss=0.2975, simple_loss=0.3772, pruned_loss=0.1089, over 28336.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.4013, pruned_loss=0.1434, over 5666893.11 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3785, pruned_loss=0.1192, over 5742433.75 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.4036, pruned_loss=0.1462, over 5658193.70 frames. ], batch size: 65, lr: 6.55e-03, grad_scale: 2.0
+2023-03-02 14:49:46,640 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:50:01,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.498e+02 1.519e+03 1.903e+03 2.669e+03 6.411e+03, threshold=3.807e+03, percent-clipped=7.0
+2023-03-02 14:50:13,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9162, 3.6836, 3.5466, 1.6518], device='cuda:0'), covar=tensor([0.0608, 0.0732, 0.1112, 0.2121], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0781, 0.0835, 0.0602], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 14:50:16,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0562, 1.3839, 1.1443, 0.2583], device='cuda:0'), covar=tensor([0.1162, 0.1030, 0.1550, 0.1704], device='cuda:0'), in_proj_covar=tensor([0.1355, 0.1264, 0.1352, 0.1128], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 14:50:24,653 INFO [train.py:968] (0/2) Epoch 5, batch 9900, giga_loss[loss=0.3767, simple_loss=0.4259, pruned_loss=0.1638, over 28676.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.4012, pruned_loss=0.1425, over 5674764.78 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3784, pruned_loss=0.1191, over 5744381.10 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.4035, pruned_loss=0.1451, over 5665066.21 frames. ], batch size: 92, lr: 6.55e-03, grad_scale: 2.0
+2023-03-02 14:51:00,332 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:51:13,119 INFO [train.py:968] (0/2) Epoch 5, batch 9950, giga_loss[loss=0.3487, simple_loss=0.4033, pruned_loss=0.147, over 28895.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.4018, pruned_loss=0.1433, over 5669045.23 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3783, pruned_loss=0.119, over 5747903.36 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.4044, pruned_loss=0.1461, over 5655794.79 frames. ], batch size: 227, lr: 6.55e-03, grad_scale: 2.0
+2023-03-02 14:51:16,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4930, 2.5670, 1.9006, 1.7584], device='cuda:0'), covar=tensor([0.0734, 0.0218, 0.0291, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0198, 0.0126, 0.0131, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0061], device='cuda:0')
+2023-03-02 14:51:33,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.005e+02 1.513e+03 1.932e+03 2.677e+03 5.827e+03, threshold=3.863e+03, percent-clipped=9.0
+2023-03-02 14:51:35,638 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-02 14:51:43,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9246, 3.7490, 3.6417, 1.6453], device='cuda:0'), covar=tensor([0.0500, 0.0505, 0.0724, 0.2075], device='cuda:0'), in_proj_covar=tensor([0.0857, 0.0774, 0.0830, 0.0599], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 14:52:02,039 INFO [train.py:968] (0/2) Epoch 5, batch 10000, giga_loss[loss=0.323, simple_loss=0.3902, pruned_loss=0.1278, over 28541.00 frames. ], tot_loss[loss=0.3452, simple_loss=0.4021, pruned_loss=0.1441, over 5672187.64 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3782, pruned_loss=0.119, over 5749117.03 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.4048, pruned_loss=0.147, over 5658760.67 frames. ], batch size: 78, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:52:02,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:52:24,925 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-02 14:52:30,686 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-02 14:52:50,611 INFO [train.py:968] (0/2) Epoch 5, batch 10050, libri_loss[loss=0.2689, simple_loss=0.3399, pruned_loss=0.09896, over 29491.00 frames. ], tot_loss[loss=0.345, simple_loss=0.4008, pruned_loss=0.1446, over 5670509.86 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3781, pruned_loss=0.119, over 5752192.36 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.4035, pruned_loss=0.1474, over 5655626.61 frames. ], batch size: 70, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:52:52,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:53:10,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.757e+03 2.162e+03 2.980e+03 6.597e+03, threshold=4.324e+03, percent-clipped=9.0
+2023-03-02 14:53:17,009 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:53:19,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:53:35,562 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-192000.pt
+2023-03-02 14:53:35,854 INFO [train.py:968] (0/2) Epoch 5, batch 10100, giga_loss[loss=0.3225, simple_loss=0.3792, pruned_loss=0.133, over 29109.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.4, pruned_loss=0.1448, over 5675265.31 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3793, pruned_loss=0.1197, over 5756217.21 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4018, pruned_loss=0.1472, over 5656778.46 frames. ], batch size: 113, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:53:44,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:53:48,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5640, 2.2908, 1.6494, 0.7235], device='cuda:0'), covar=tensor([0.2465, 0.1145, 0.1999, 0.2569], device='cuda:0'), in_proj_covar=tensor([0.1379, 0.1271, 0.1363, 0.1141], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 14:54:16,098 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:54:18,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:54:21,962 INFO [train.py:968] (0/2) Epoch 5, batch 10150, giga_loss[loss=0.3775, simple_loss=0.4144, pruned_loss=0.1703, over 27562.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3981, pruned_loss=0.1437, over 5676104.23 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3795, pruned_loss=0.1198, over 5751244.32 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.3998, pruned_loss=0.1462, over 5662990.11 frames. ], batch size: 472, lr: 6.55e-03, grad_scale: 2.0
+2023-03-02 14:54:46,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.447e+02 1.669e+03 2.045e+03 3.243e+03 7.374e+03, threshold=4.091e+03, percent-clipped=9.0
+2023-03-02 14:54:47,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2353, 1.4392, 1.1943, 1.5952], device='cuda:0'), covar=tensor([0.2149, 0.1988, 0.2050, 0.1865], device='cuda:0'), in_proj_covar=tensor([0.1123, 0.0878, 0.0996, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 14:54:48,446 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:55:07,175 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:55:09,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:55:10,352 INFO [train.py:968] (0/2) Epoch 5, batch 10200, giga_loss[loss=0.3801, simple_loss=0.4188, pruned_loss=0.1707, over 27928.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3978, pruned_loss=0.145, over 5664687.54 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3797, pruned_loss=0.1198, over 5741527.10 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.3995, pruned_loss=0.1475, over 5660510.96 frames. ], batch size: 412, lr: 6.54e-03, grad_scale: 2.0
+2023-03-02 14:55:38,676 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:55:43,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:55:52,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5744, 2.4709, 1.6908, 0.9083], device='cuda:0'), covar=tensor([0.2602, 0.1267, 0.2307, 0.2345], device='cuda:0'), in_proj_covar=tensor([0.1369, 0.1259, 0.1355, 0.1135], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 14:55:57,450 INFO [train.py:968] (0/2) Epoch 5, batch 10250, giga_loss[loss=0.2827, simple_loss=0.3538, pruned_loss=0.1059, over 28961.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3963, pruned_loss=0.1441, over 5669821.81 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3795, pruned_loss=0.1197, over 5743992.06 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.398, pruned_loss=0.1465, over 5663126.94 frames. ], batch size: 213, lr: 6.54e-03, grad_scale: 2.0
+2023-03-02 14:56:04,685 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-02 14:56:18,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.775e+02 1.473e+03 2.020e+03 3.061e+03 5.749e+03, threshold=4.039e+03, percent-clipped=13.0
+2023-03-02 14:56:42,613 INFO [train.py:968] (0/2) Epoch 5, batch 10300, libri_loss[loss=0.2782, simple_loss=0.3504, pruned_loss=0.103, over 29587.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3937, pruned_loss=0.1408, over 5662856.36 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3796, pruned_loss=0.1198, over 5748110.76 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3953, pruned_loss=0.1432, over 5652004.15 frames. ], batch size: 74, lr: 6.54e-03, grad_scale: 2.0
+2023-03-02 14:57:28,429 INFO [train.py:968] (0/2) Epoch 5, batch 10350, giga_loss[loss=0.2981, simple_loss=0.3681, pruned_loss=0.1141, over 28540.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3907, pruned_loss=0.1377, over 5664309.67 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3794, pruned_loss=0.1196, over 5748766.71 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3926, pruned_loss=0.1404, over 5653237.51 frames. ], batch size: 71, lr: 6.54e-03, grad_scale: 2.0
+2023-03-02 14:57:49,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.556e+02 1.324e+03 1.537e+03 2.110e+03 4.796e+03, threshold=3.075e+03, percent-clipped=2.0
+2023-03-02 14:57:51,106 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2941, 1.9667, 1.6683, 1.5182], device='cuda:0'), covar=tensor([0.1817, 0.2221, 0.1454, 0.1503], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0760, 0.0776, 0.0710], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 14:57:51,802 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:57:54,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:57:59,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1606, 1.3356, 1.1561, 1.1968], device='cuda:0'), covar=tensor([0.1170, 0.1353, 0.1673, 0.1409], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0741, 0.0631, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 14:58:14,126 INFO [train.py:968] (0/2) Epoch 5, batch 10400, giga_loss[loss=0.2986, simple_loss=0.3674, pruned_loss=0.1149, over 28883.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3897, pruned_loss=0.137, over 5670183.18 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.379, pruned_loss=0.1193, over 5750630.73 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3919, pruned_loss=0.14, over 5656917.72 frames. ], batch size: 227, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 14:58:22,148 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:59:03,229 INFO [train.py:968] (0/2) Epoch 5, batch 10450, libri_loss[loss=0.3176, simple_loss=0.385, pruned_loss=0.1251, over 29546.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3854, pruned_loss=0.135, over 5672068.86 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3783, pruned_loss=0.119, over 5755685.16 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3881, pruned_loss=0.1383, over 5653910.44 frames. ], batch size: 81, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 14:59:17,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2677, 1.0985, 4.9012, 3.5001], device='cuda:0'), covar=tensor([0.1670, 0.2394, 0.0319, 0.0576], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0533, 0.0746, 0.0605], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 14:59:20,265 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8583, 1.5461, 1.2593, 1.3601], device='cuda:0'), covar=tensor([0.0630, 0.0749, 0.0929, 0.1029], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0455, 0.0506, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 14:59:23,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.276e+02 1.527e+03 2.308e+03 3.107e+03 6.848e+03, threshold=4.616e+03, percent-clipped=26.0
+2023-03-02 14:59:40,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0263, 1.2553, 1.2741, 1.1836], device='cuda:0'), covar=tensor([0.0987, 0.0936, 0.1364, 0.0961], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0757, 0.0641, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 14:59:51,682 INFO [train.py:968] (0/2) Epoch 5, batch 10500, giga_loss[loss=0.3032, simple_loss=0.3724, pruned_loss=0.117, over 28962.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3843, pruned_loss=0.1348, over 5674143.98 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3781, pruned_loss=0.1188, over 5754453.48 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3868, pruned_loss=0.1379, over 5659431.74 frames. ], batch size: 136, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:00:30,382 INFO [train.py:968] (0/2) Epoch 5, batch 10550, giga_loss[loss=0.3477, simple_loss=0.4065, pruned_loss=0.1445, over 28971.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3869, pruned_loss=0.1359, over 5676139.29 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3774, pruned_loss=0.1184, over 5751580.91 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3898, pruned_loss=0.1393, over 5664017.18 frames. ], batch size: 164, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:00:49,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.895e+02 1.493e+03 1.759e+03 2.388e+03 6.110e+03, threshold=3.517e+03, percent-clipped=3.0
+2023-03-02 15:01:10,632 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:01:14,968 INFO [train.py:968] (0/2) Epoch 5, batch 10600, giga_loss[loss=0.3143, simple_loss=0.3842, pruned_loss=0.1222, over 28940.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3887, pruned_loss=0.1367, over 5664088.97 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3778, pruned_loss=0.1184, over 5753899.16 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3908, pruned_loss=0.1396, over 5650878.96 frames. ], batch size: 164, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:01:48,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8068, 2.3951, 1.7200, 2.1149], device='cuda:0'), covar=tensor([0.0519, 0.0604, 0.0874, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0445, 0.0498, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:01:55,504 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:01:58,328 INFO [train.py:968] (0/2) Epoch 5, batch 10650, giga_loss[loss=0.3174, simple_loss=0.3823, pruned_loss=0.1262, over 28624.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3891, pruned_loss=0.1369, over 5647747.44 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.379, pruned_loss=0.1192, over 5750921.54 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3901, pruned_loss=0.1392, over 5636933.41 frames. ], batch size: 78, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:02:00,493 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 15:02:18,922 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.773e+02 1.572e+03 2.198e+03 3.031e+03 1.000e+04, threshold=4.396e+03, percent-clipped=13.0
+2023-03-02 15:02:39,623 INFO [train.py:968] (0/2) Epoch 5, batch 10700, libri_loss[loss=0.3404, simple_loss=0.4132, pruned_loss=0.1338, over 29539.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.39, pruned_loss=0.1378, over 5639630.36 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3798, pruned_loss=0.1195, over 5744330.30 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3904, pruned_loss=0.1401, over 5631709.40 frames. ], batch size: 84, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:02:50,412 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:02:51,710 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:03:21,749 INFO [train.py:968] (0/2) Epoch 5, batch 10750, giga_loss[loss=0.4121, simple_loss=0.4457, pruned_loss=0.1893, over 28810.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3916, pruned_loss=0.1392, over 5643096.70 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3799, pruned_loss=0.1195, over 5742607.38 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3922, pruned_loss=0.1416, over 5634931.45 frames. ], batch size: 199, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:03:42,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.585e+02 1.446e+03 2.115e+03 2.873e+03 6.840e+03, threshold=4.230e+03, percent-clipped=4.0
+2023-03-02 15:04:04,113 INFO [train.py:968] (0/2) Epoch 5, batch 10800, giga_loss[loss=0.3812, simple_loss=0.4152, pruned_loss=0.1736, over 28393.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3916, pruned_loss=0.1385, over 5657278.54 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.379, pruned_loss=0.119, over 5749633.78 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3935, pruned_loss=0.1419, over 5640064.19 frames. ], batch size: 65, lr: 6.53e-03, grad_scale: 8.0
+2023-03-02 15:04:45,153 INFO [train.py:968] (0/2) Epoch 5, batch 10850, libri_loss[loss=0.2656, simple_loss=0.3387, pruned_loss=0.09629, over 29564.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3917, pruned_loss=0.1386, over 5655996.72 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3788, pruned_loss=0.1188, over 5744735.60 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.394, pruned_loss=0.1423, over 5642116.50 frames. ], batch size: 77, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:05:05,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.638e+02 1.462e+03 1.999e+03 2.623e+03 4.954e+03, threshold=3.998e+03, percent-clipped=4.0
+2023-03-02 15:05:13,396 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:05:21,501 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:05:29,064 INFO [train.py:968] (0/2) Epoch 5, batch 10900, giga_loss[loss=0.3694, simple_loss=0.4107, pruned_loss=0.1641, over 27624.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3935, pruned_loss=0.1401, over 5655163.90 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3791, pruned_loss=0.119, over 5742848.13 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3955, pruned_loss=0.1435, over 5642986.02 frames. ], batch size: 472, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:05:50,645 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:06:11,973 INFO [train.py:968] (0/2) Epoch 5, batch 10950, giga_loss[loss=0.3863, simple_loss=0.4421, pruned_loss=0.1652, over 28105.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3946, pruned_loss=0.141, over 5643715.43 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.379, pruned_loss=0.1189, over 5728784.17 frames. ], giga_tot_loss[loss=0.3429, simple_loss=0.3967, pruned_loss=0.1445, over 5644108.51 frames. ], batch size: 412, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:06:31,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:06:36,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.665e+02 1.475e+03 2.000e+03 2.707e+03 1.138e+04, threshold=4.000e+03, percent-clipped=12.0
+2023-03-02 15:06:59,028 INFO [train.py:968] (0/2) Epoch 5, batch 11000, giga_loss[loss=0.3087, simple_loss=0.376, pruned_loss=0.1207, over 28378.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3957, pruned_loss=0.1407, over 5644000.82 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3788, pruned_loss=0.1188, over 5722894.17 frames. ], giga_tot_loss[loss=0.3429, simple_loss=0.3979, pruned_loss=0.1439, over 5648497.13 frames. ], batch size: 71, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:07:23,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:07:50,655 INFO [train.py:968] (0/2) Epoch 5, batch 11050, giga_loss[loss=0.3762, simple_loss=0.4219, pruned_loss=0.1652, over 28655.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3951, pruned_loss=0.1413, over 5635147.18 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3784, pruned_loss=0.1188, over 5723596.69 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3977, pruned_loss=0.1446, over 5635641.42 frames. ], batch size: 262, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:07:58,207 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2955, 2.1556, 1.6380, 1.7442], device='cuda:0'), covar=tensor([0.0636, 0.0707, 0.0944, 0.0993], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0455, 0.0504, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:08:13,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.100e+02 1.406e+03 1.894e+03 2.537e+03 9.834e+03, threshold=3.788e+03, percent-clipped=10.0
+2023-03-02 15:08:24,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:08:27,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:08:39,863 INFO [train.py:968] (0/2) Epoch 5, batch 11100, giga_loss[loss=0.3481, simple_loss=0.4016, pruned_loss=0.1473, over 28931.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.393, pruned_loss=0.1404, over 5653240.69 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3782, pruned_loss=0.1186, over 5728174.96 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3956, pruned_loss=0.1436, over 5647931.49 frames. ], batch size: 199, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:08:50,682 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:08:54,521 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:08:57,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:09:32,394 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:09:37,163 INFO [train.py:968] (0/2) Epoch 5, batch 11150, giga_loss[loss=0.3681, simple_loss=0.4127, pruned_loss=0.1618, over 28993.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3937, pruned_loss=0.1417, over 5652941.71 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3781, pruned_loss=0.1186, over 5731926.54 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3961, pruned_loss=0.1447, over 5643600.38 frames. ], batch size: 213, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:09:51,856 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:09:51,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:09:54,138 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:09:57,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.845e+02 1.474e+03 2.057e+03 2.776e+03 8.293e+03, threshold=4.113e+03, percent-clipped=10.0
+2023-03-02 15:09:59,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4648, 1.1230, 5.0850, 3.4429], device='cuda:0'), covar=tensor([0.1667, 0.2411, 0.0305, 0.0701], device='cuda:0'), in_proj_covar=tensor([0.0580, 0.0530, 0.0748, 0.0609], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 15:10:10,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9069, 1.7211, 1.2461, 1.4727], device='cuda:0'), covar=tensor([0.0611, 0.0699, 0.0968, 0.1052], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0460, 0.0507, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:10:19,574 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:10:20,675 INFO [train.py:968] (0/2) Epoch 5, batch 11200, giga_loss[loss=0.3099, simple_loss=0.3769, pruned_loss=0.1214, over 29053.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3926, pruned_loss=0.1405, over 5663127.46 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3784, pruned_loss=0.1188, over 5728079.51 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3948, pruned_loss=0.1436, over 5656630.45 frames. ], batch size: 155, lr: 6.53e-03, grad_scale: 8.0
+2023-03-02 15:10:45,946 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4573, 1.9579, 1.2223, 1.0811], device='cuda:0'), covar=tensor([0.1174, 0.0833, 0.0897, 0.1021], device='cuda:0'), in_proj_covar=tensor([0.1399, 0.1186, 0.1176, 0.1249], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 15:10:47,204 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:10:48,597 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:10:50,160 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:10:50,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:10:52,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:11:06,050 INFO [train.py:968] (0/2) Epoch 5, batch 11250, giga_loss[loss=0.4186, simple_loss=0.4369, pruned_loss=0.2001, over 26571.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3923, pruned_loss=0.1412, over 5666738.91 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3782, pruned_loss=0.1186, over 5732065.10 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3945, pruned_loss=0.1443, over 5656723.94 frames. ], batch size: 555, lr: 6.53e-03, grad_scale: 8.0
+2023-03-02 15:11:14,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:11:16,742 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:11:19,201 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:11:23,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:11:28,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.918e+02 1.377e+03 1.834e+03 2.682e+03 5.343e+03, threshold=3.669e+03, percent-clipped=9.0
+2023-03-02 15:11:53,502 INFO [train.py:968] (0/2) Epoch 5, batch 11300, giga_loss[loss=0.3595, simple_loss=0.4091, pruned_loss=0.1549, over 28728.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3928, pruned_loss=0.1417, over 5655484.04 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3785, pruned_loss=0.1187, over 5724068.87 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3947, pruned_loss=0.1446, over 5653644.19 frames. ], batch size: 262, lr: 6.53e-03, grad_scale: 8.0
+2023-03-02 15:11:54,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:12:15,669 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193224.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:12:42,251 INFO [train.py:968] (0/2) Epoch 5, batch 11350, giga_loss[loss=0.355, simple_loss=0.4113, pruned_loss=0.1493, over 28721.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3943, pruned_loss=0.1433, over 5657650.18 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3787, pruned_loss=0.1189, over 5725391.48 frames. ], giga_tot_loss[loss=0.3444, simple_loss=0.3962, pruned_loss=0.1463, over 5653036.55 frames. ], batch size: 262, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:13:03,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.476e+02 1.736e+03 2.460e+03 3.834e+03 1.010e+04, threshold=4.919e+03, percent-clipped=24.0
+2023-03-02 15:13:28,598 INFO [train.py:968] (0/2) Epoch 5, batch 11400, giga_loss[loss=0.2814, simple_loss=0.3514, pruned_loss=0.1057, over 28789.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3963, pruned_loss=0.1452, over 5663508.48 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.379, pruned_loss=0.1191, over 5726572.44 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.3978, pruned_loss=0.1479, over 5657248.19 frames. ], batch size: 119, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:13:34,072 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:13:35,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:13:38,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:13:42,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:00,534 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:06,512 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:08,017 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:12,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:15,775 INFO [train.py:968] (0/2) Epoch 5, batch 11450, giga_loss[loss=0.4, simple_loss=0.4179, pruned_loss=0.1911, over 23272.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3965, pruned_loss=0.1448, over 5669400.91 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3785, pruned_loss=0.1187, over 5731253.81 frames. ], giga_tot_loss[loss=0.3473, simple_loss=0.3986, pruned_loss=0.148, over 5658839.54 frames. ], batch size: 705, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:14:25,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5315, 3.8376, 1.7215, 1.4563], device='cuda:0'), covar=tensor([0.0832, 0.0286, 0.0754, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0485, 0.0312, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 15:14:41,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.841e+02 1.723e+03 2.405e+03 3.332e+03 9.114e+03, threshold=4.810e+03, percent-clipped=6.0
+2023-03-02 15:14:43,513 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:52,675 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:59,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3540, 1.4083, 1.1827, 1.5739], device='cuda:0'), covar=tensor([0.2143, 0.2109, 0.2094, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.1119, 0.0872, 0.0992, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 15:15:07,489 INFO [train.py:968] (0/2) Epoch 5, batch 11500, giga_loss[loss=0.3227, simple_loss=0.3868, pruned_loss=0.1293, over 28841.00 frames. ], tot_loss[loss=0.346, simple_loss=0.3978, pruned_loss=0.1471, over 5662117.72 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3785, pruned_loss=0.1187, over 5731253.81 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.3995, pruned_loss=0.1495, over 5653897.67 frames. ], batch size: 199, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:15:12,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5474, 1.4301, 1.5221, 1.3727], device='cuda:0'), covar=tensor([0.0997, 0.1528, 0.1516, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0751, 0.0643, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 15:15:47,202 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:15:54,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:16:00,896 INFO [train.py:968] (0/2) Epoch 5, batch 11550, giga_loss[loss=0.3532, simple_loss=0.4013, pruned_loss=0.1526, over 27899.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.3977, pruned_loss=0.1478, over 5659837.92 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3784, pruned_loss=0.1186, over 5733034.35 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.3994, pruned_loss=0.1501, over 5651158.85 frames. ], batch size: 412, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:16:25,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.344e+02 1.619e+03 2.131e+03 2.938e+03 8.346e+03, threshold=4.263e+03, percent-clipped=6.0
+2023-03-02 15:16:36,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4474, 3.0175, 1.4883, 1.3911], device='cuda:0'), covar=tensor([0.0843, 0.0398, 0.0818, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0488, 0.0312, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 15:16:52,466 INFO [train.py:968] (0/2) Epoch 5, batch 11600, giga_loss[loss=0.3525, simple_loss=0.4058, pruned_loss=0.1496, over 28798.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3969, pruned_loss=0.146, over 5668534.22 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3784, pruned_loss=0.1186, over 5728099.68 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3985, pruned_loss=0.1485, over 5664888.23 frames. ], batch size: 243, lr: 6.52e-03, grad_scale: 8.0
+2023-03-02 15:16:55,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:17:17,074 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:17:20,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:17:38,321 INFO [train.py:968] (0/2) Epoch 5, batch 11650, giga_loss[loss=0.4352, simple_loss=0.4601, pruned_loss=0.2052, over 28727.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3971, pruned_loss=0.146, over 5663894.28 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3783, pruned_loss=0.1186, over 5731507.76 frames. ], giga_tot_loss[loss=0.3482, simple_loss=0.3991, pruned_loss=0.1487, over 5656286.24 frames. ], batch size: 284, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:17:46,923 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:17:55,497 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 15:18:01,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.618e+03 2.198e+03 3.041e+03 6.245e+03, threshold=4.397e+03, percent-clipped=8.0
+2023-03-02 15:18:09,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:18:11,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:18:22,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193599.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:18:22,970 INFO [train.py:968] (0/2) Epoch 5, batch 11700, giga_loss[loss=0.4079, simple_loss=0.4425, pruned_loss=0.1866, over 28960.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.398, pruned_loss=0.1459, over 5675942.83 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3788, pruned_loss=0.1191, over 5733187.67 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4, pruned_loss=0.1488, over 5665533.10 frames. ], batch size: 106, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:18:30,294 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:18:42,284 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:19:11,138 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:19:13,925 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:19:14,324 INFO [train.py:968] (0/2) Epoch 5, batch 11750, giga_loss[loss=0.3793, simple_loss=0.4149, pruned_loss=0.1719, over 27926.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.4013, pruned_loss=0.149, over 5666696.22 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.379, pruned_loss=0.1192, over 5726445.22 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4032, pruned_loss=0.1519, over 5662951.76 frames. ], batch size: 412, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:19:39,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.503e+03 1.990e+03 2.785e+03 8.463e+03, threshold=3.980e+03, percent-clipped=4.0
+2023-03-02 15:19:39,919 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:19:40,604 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:20:01,881 INFO [train.py:968] (0/2) Epoch 5, batch 11800, giga_loss[loss=0.4021, simple_loss=0.431, pruned_loss=0.1866, over 28646.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.3996, pruned_loss=0.1474, over 5672877.51 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3787, pruned_loss=0.1191, over 5721677.82 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4018, pruned_loss=0.1504, over 5672354.91 frames. ], batch size: 242, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:20:39,606 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193742.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:20:42,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193745.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:20:45,663 INFO [train.py:968] (0/2) Epoch 5, batch 11850, giga_loss[loss=0.3455, simple_loss=0.405, pruned_loss=0.143, over 28484.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.4007, pruned_loss=0.1473, over 5668021.18 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3791, pruned_loss=0.1193, over 5711747.57 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4027, pruned_loss=0.1503, over 5675761.25 frames. ], batch size: 85, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:20:52,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9460, 3.7592, 3.5933, 1.8493], device='cuda:0'), covar=tensor([0.0481, 0.0540, 0.0800, 0.2096], device='cuda:0'), in_proj_covar=tensor([0.0891, 0.0794, 0.0846, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-02 15:21:09,654 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193774.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:21:12,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.965e+02 1.506e+03 2.132e+03 3.011e+03 1.170e+04, threshold=4.264e+03, percent-clipped=14.0
+2023-03-02 15:21:15,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9973, 1.9615, 1.3699, 1.5966], device='cuda:0'), covar=tensor([0.0622, 0.0551, 0.0932, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0458, 0.0508, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:21:31,297 INFO [train.py:968] (0/2) Epoch 5, batch 11900, giga_loss[loss=0.2869, simple_loss=0.3609, pruned_loss=0.1064, over 28393.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.4009, pruned_loss=0.1466, over 5662603.07 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3788, pruned_loss=0.1191, over 5708293.60 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4035, pruned_loss=0.15, over 5670986.98 frames. ], batch size: 65, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:21:40,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:21:47,504 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:22:17,352 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193847.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:22:20,053 INFO [train.py:968] (0/2) Epoch 5, batch 11950, giga_loss[loss=0.3048, simple_loss=0.3714, pruned_loss=0.1191, over 28966.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3996, pruned_loss=0.1457, over 5660916.88 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3786, pruned_loss=0.1189, over 5712364.80 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4022, pruned_loss=0.1491, over 5662975.87 frames. ], batch size: 213, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:22:44,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.191e+02 1.378e+03 1.671e+03 2.058e+03 4.187e+03, threshold=3.341e+03, percent-clipped=0.0
+2023-03-02 15:23:06,780 INFO [train.py:968] (0/2) Epoch 5, batch 12000, giga_loss[loss=0.3128, simple_loss=0.3754, pruned_loss=0.1251, over 28912.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3982, pruned_loss=0.1445, over 5674001.64 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3788, pruned_loss=0.119, over 5713069.57 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.4004, pruned_loss=0.1476, over 5674191.31 frames. ], batch size: 186, lr: 6.51e-03, grad_scale: 8.0
+2023-03-02 15:23:06,786 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 15:23:16,110 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2404, simple_loss=0.3437, pruned_loss=0.06857, over 944034.00 frames. 
+2023-03-02 15:23:16,111 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 15:24:04,109 INFO [train.py:968] (0/2) Epoch 5, batch 12050, giga_loss[loss=0.4614, simple_loss=0.4664, pruned_loss=0.2282, over 26604.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3985, pruned_loss=0.1457, over 5660346.95 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3786, pruned_loss=0.119, over 5714859.86 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4007, pruned_loss=0.1484, over 5658351.61 frames. ], batch size: 555, lr: 6.51e-03, grad_scale: 8.0
+2023-03-02 15:24:06,700 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:24:10,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:24:34,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.799e+02 1.509e+03 1.860e+03 2.650e+03 4.989e+03, threshold=3.720e+03, percent-clipped=9.0
+2023-03-02 15:24:36,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:24:39,900 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:24:43,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5768, 1.8364, 1.2257, 1.0561], device='cuda:0'), covar=tensor([0.1165, 0.0882, 0.0834, 0.1061], device='cuda:0'), in_proj_covar=tensor([0.1384, 0.1180, 0.1178, 0.1241], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 15:24:53,402 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-02 15:24:54,385 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-194000.pt
+2023-03-02 15:24:54,670 INFO [train.py:968] (0/2) Epoch 5, batch 12100, giga_loss[loss=0.4137, simple_loss=0.4489, pruned_loss=0.1893, over 28939.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3995, pruned_loss=0.1458, over 5670211.67 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3782, pruned_loss=0.1187, over 5716133.69 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.4019, pruned_loss=0.1486, over 5666454.83 frames. ], batch size: 227, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:24:57,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9635, 1.0358, 4.0069, 3.0577], device='cuda:0'), covar=tensor([0.1768, 0.2477, 0.0375, 0.0753], device='cuda:0'), in_proj_covar=tensor([0.0582, 0.0534, 0.0760, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 15:25:04,370 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:25:44,430 INFO [train.py:968] (0/2) Epoch 5, batch 12150, giga_loss[loss=0.349, simple_loss=0.3992, pruned_loss=0.1494, over 28626.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3993, pruned_loss=0.1471, over 5665507.91 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.378, pruned_loss=0.1186, over 5717084.02 frames. ], giga_tot_loss[loss=0.3503, simple_loss=0.4014, pruned_loss=0.1496, over 5661526.97 frames. ], batch size: 307, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:25:46,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:25:57,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6584, 1.6222, 1.3576, 1.2185], device='cuda:0'), covar=tensor([0.0568, 0.0459, 0.0817, 0.0757], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0457, 0.0505, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:26:00,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-02 15:26:07,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.978e+02 1.577e+03 1.983e+03 3.069e+03 8.133e+03, threshold=3.965e+03, percent-clipped=13.0
+2023-03-02 15:26:30,747 INFO [train.py:968] (0/2) Epoch 5, batch 12200, giga_loss[loss=0.3469, simple_loss=0.4042, pruned_loss=0.1448, over 28893.00 frames. ], tot_loss[loss=0.3488, simple_loss=0.4006, pruned_loss=0.1485, over 5668663.71 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3782, pruned_loss=0.1185, over 5723784.11 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4029, pruned_loss=0.1515, over 5658076.95 frames. ], batch size: 174, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:26:52,427 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:26:54,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:27:15,842 INFO [train.py:968] (0/2) Epoch 5, batch 12250, giga_loss[loss=0.4052, simple_loss=0.4387, pruned_loss=0.1858, over 28586.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.4014, pruned_loss=0.1489, over 5674152.36 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3784, pruned_loss=0.1185, over 5728763.37 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.4036, pruned_loss=0.1521, over 5659650.01 frames. ], batch size: 307, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:27:20,934 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:27:32,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2331, 1.3547, 0.9025, 1.0427], device='cuda:0'), covar=tensor([0.0673, 0.0577, 0.0510, 0.0576], device='cuda:0'), in_proj_covar=tensor([0.1395, 0.1190, 0.1187, 0.1246], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 15:27:42,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.888e+02 1.583e+03 1.888e+03 2.487e+03 5.365e+03, threshold=3.776e+03, percent-clipped=3.0
+2023-03-02 15:27:54,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:28:00,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:28:02,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:28:03,580 INFO [train.py:968] (0/2) Epoch 5, batch 12300, giga_loss[loss=0.3816, simple_loss=0.4047, pruned_loss=0.1792, over 23395.00 frames. ], tot_loss[loss=0.349, simple_loss=0.401, pruned_loss=0.1485, over 5668016.59 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3781, pruned_loss=0.1183, over 5732349.74 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4035, pruned_loss=0.1518, over 5652328.84 frames. ], batch size: 705, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:28:12,532 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-02 15:28:25,583 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194222.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:28:30,194 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:28:45,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4774, 1.7173, 1.7414, 1.6298], device='cuda:0'), covar=tensor([0.1327, 0.1760, 0.1046, 0.1127], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0764, 0.0783, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 15:28:48,905 INFO [train.py:968] (0/2) Epoch 5, batch 12350, giga_loss[loss=0.342, simple_loss=0.4015, pruned_loss=0.1413, over 29009.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.3994, pruned_loss=0.1472, over 5666114.19 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.378, pruned_loss=0.1183, over 5737862.72 frames. ], giga_tot_loss[loss=0.352, simple_loss=0.4023, pruned_loss=0.1509, over 5645881.81 frames. ], batch size: 145, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:29:16,132 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.026e+02 1.666e+03 2.300e+03 3.125e+03 6.120e+03, threshold=4.600e+03, percent-clipped=14.0
+2023-03-02 15:29:16,408 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:29:37,372 INFO [train.py:968] (0/2) Epoch 5, batch 12400, giga_loss[loss=0.3442, simple_loss=0.3974, pruned_loss=0.1455, over 28548.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.3998, pruned_loss=0.1473, over 5661294.91 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.378, pruned_loss=0.1181, over 5741742.44 frames. ], giga_tot_loss[loss=0.3529, simple_loss=0.4029, pruned_loss=0.1515, over 5638707.42 frames. ], batch size: 336, lr: 6.51e-03, grad_scale: 8.0
+2023-03-02 15:29:52,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2826, 1.4857, 1.1974, 1.4783], device='cuda:0'), covar=tensor([0.0755, 0.0374, 0.0351, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0198, 0.0126, 0.0130, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0061], device='cuda:0')
+2023-03-02 15:29:54,167 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-03-02 15:30:06,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:30:08,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:30:22,775 INFO [train.py:968] (0/2) Epoch 5, batch 12450, giga_loss[loss=0.3315, simple_loss=0.3867, pruned_loss=0.1382, over 28722.00 frames. ], tot_loss[loss=0.347, simple_loss=0.3997, pruned_loss=0.1471, over 5660964.42 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3782, pruned_loss=0.1182, over 5743222.74 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4021, pruned_loss=0.1506, over 5641221.44 frames. ], batch size: 92, lr: 6.51e-03, grad_scale: 8.0
+2023-03-02 15:30:36,032 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:30:36,656 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194365.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:30:38,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194368.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:30:47,942 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.290e+02 1.463e+03 1.747e+03 2.307e+03 5.863e+03, threshold=3.495e+03, percent-clipped=5.0
+2023-03-02 15:30:52,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:31:09,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194397.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:31:10,933 INFO [train.py:968] (0/2) Epoch 5, batch 12500, giga_loss[loss=0.3594, simple_loss=0.4056, pruned_loss=0.1566, over 28530.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.3991, pruned_loss=0.1468, over 5661587.37 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.378, pruned_loss=0.1182, over 5745098.79 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4014, pruned_loss=0.1499, over 5643811.03 frames. ], batch size: 336, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:31:41,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3415, 1.4892, 1.5110, 1.5432], device='cuda:0'), covar=tensor([0.0958, 0.1216, 0.1158, 0.1050], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0741, 0.0638, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 15:32:00,392 INFO [train.py:968] (0/2) Epoch 5, batch 12550, libri_loss[loss=0.3272, simple_loss=0.398, pruned_loss=0.1281, over 29560.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3972, pruned_loss=0.146, over 5661669.44 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3781, pruned_loss=0.1182, over 5746777.56 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.399, pruned_loss=0.1486, over 5645368.75 frames. ], batch size: 83, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:32:17,900 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:32:29,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.454e+03 2.006e+03 2.823e+03 4.808e+03, threshold=4.012e+03, percent-clipped=12.0
+2023-03-02 15:32:49,600 INFO [train.py:968] (0/2) Epoch 5, batch 12600, giga_loss[loss=0.3299, simple_loss=0.3906, pruned_loss=0.1346, over 28976.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3938, pruned_loss=0.1435, over 5677467.73 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3782, pruned_loss=0.1182, over 5748160.71 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3953, pruned_loss=0.1459, over 5662744.18 frames. ], batch size: 106, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:33:17,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:33:19,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:33:38,395 INFO [train.py:968] (0/2) Epoch 5, batch 12650, libri_loss[loss=0.3207, simple_loss=0.3945, pruned_loss=0.1235, over 28876.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3907, pruned_loss=0.1422, over 5662588.80 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3784, pruned_loss=0.1182, over 5751505.95 frames. ], giga_tot_loss[loss=0.3406, simple_loss=0.3921, pruned_loss=0.1446, over 5645738.97 frames. ], batch size: 107, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:33:45,359 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:34:04,873 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.165e+02 1.658e+03 2.166e+03 2.819e+03 7.865e+03, threshold=4.332e+03, percent-clipped=10.0
+2023-03-02 15:34:28,233 INFO [train.py:968] (0/2) Epoch 5, batch 12700, giga_loss[loss=0.382, simple_loss=0.4115, pruned_loss=0.1763, over 26632.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3904, pruned_loss=0.1431, over 5660225.00 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3784, pruned_loss=0.1182, over 5753643.63 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3916, pruned_loss=0.1454, over 5644051.16 frames. ], batch size: 555, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:34:42,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2817, 3.1216, 3.0337, 1.4727], device='cuda:0'), covar=tensor([0.0761, 0.0694, 0.1063, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0786, 0.0836, 0.0601], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:35:14,794 INFO [train.py:968] (0/2) Epoch 5, batch 12750, giga_loss[loss=0.3381, simple_loss=0.3921, pruned_loss=0.1421, over 28521.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3889, pruned_loss=0.1421, over 5659758.78 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3779, pruned_loss=0.1179, over 5757340.85 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3907, pruned_loss=0.1448, over 5641167.71 frames. ], batch size: 336, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:35:17,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:35:39,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.198e+02 1.574e+03 2.010e+03 2.652e+03 7.778e+03, threshold=4.020e+03, percent-clipped=7.0
+2023-03-02 15:36:00,488 INFO [train.py:968] (0/2) Epoch 5, batch 12800, giga_loss[loss=0.3529, simple_loss=0.4146, pruned_loss=0.1456, over 28963.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3881, pruned_loss=0.1405, over 5662795.23 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3775, pruned_loss=0.1175, over 5761845.11 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3902, pruned_loss=0.1438, over 5640867.60 frames. ], batch size: 164, lr: 6.50e-03, grad_scale: 8.0
+2023-03-02 15:36:12,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:36:46,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:36:48,926 INFO [train.py:968] (0/2) Epoch 5, batch 12850, giga_loss[loss=0.3088, simple_loss=0.3805, pruned_loss=0.1186, over 28619.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3852, pruned_loss=0.136, over 5662840.61 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3771, pruned_loss=0.1173, over 5762610.40 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3873, pruned_loss=0.1391, over 5643305.13 frames. ], batch size: 307, lr: 6.50e-03, grad_scale: 8.0
+2023-03-02 15:36:51,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7280, 2.0146, 1.4260, 1.1765], device='cuda:0'), covar=tensor([0.0960, 0.0680, 0.0695, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.1399, 0.1182, 0.1190, 0.1250], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 15:36:57,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194755.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:37:09,165 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.90 vs. limit=2.0
+2023-03-02 15:37:20,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.425e+02 1.421e+03 1.935e+03 2.844e+03 6.657e+03, threshold=3.869e+03, percent-clipped=7.0
+2023-03-02 15:37:35,037 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:37:37,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:37:38,068 INFO [train.py:968] (0/2) Epoch 5, batch 12900, giga_loss[loss=0.3245, simple_loss=0.3909, pruned_loss=0.1291, over 28768.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3816, pruned_loss=0.1324, over 5655489.62 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3761, pruned_loss=0.117, over 5756916.16 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3844, pruned_loss=0.1357, over 5641562.14 frames. ], batch size: 284, lr: 6.50e-03, grad_scale: 8.0
+2023-03-02 15:38:06,429 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:38:08,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1880, 1.7539, 1.5222, 1.4379], device='cuda:0'), covar=tensor([0.1543, 0.1917, 0.1236, 0.1276], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0750, 0.0779, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 15:38:17,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:38:25,966 INFO [train.py:968] (0/2) Epoch 5, batch 12950, giga_loss[loss=0.2692, simple_loss=0.3457, pruned_loss=0.09639, over 28917.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3774, pruned_loss=0.1279, over 5655578.09 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3757, pruned_loss=0.1167, over 5757617.92 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3802, pruned_loss=0.1313, over 5640612.46 frames. ], batch size: 227, lr: 6.50e-03, grad_scale: 8.0
+2023-03-02 15:38:54,942 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.740e+02 1.272e+03 1.614e+03 2.153e+03 4.872e+03, threshold=3.228e+03, percent-clipped=5.0
+2023-03-02 15:39:00,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4172, 1.4747, 1.1463, 1.1242], device='cuda:0'), covar=tensor([0.0617, 0.0459, 0.0863, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0444, 0.0498, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:39:01,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8197, 1.7154, 1.2598, 1.3964], device='cuda:0'), covar=tensor([0.0671, 0.0616, 0.0982, 0.0916], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0444, 0.0498, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:39:12,905 INFO [train.py:968] (0/2) Epoch 5, batch 13000, libri_loss[loss=0.2578, simple_loss=0.3366, pruned_loss=0.08951, over 29529.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3729, pruned_loss=0.1242, over 5646372.33 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3742, pruned_loss=0.1159, over 5754474.76 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3767, pruned_loss=0.1282, over 5631628.06 frames. ], batch size: 80, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:39:23,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2329, 1.4948, 1.1423, 0.8858], device='cuda:0'), covar=tensor([0.0781, 0.0648, 0.0560, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.1407, 0.1174, 0.1193, 0.1244], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 15:39:39,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3356, 1.9637, 1.6092, 1.6011], device='cuda:0'), covar=tensor([0.0852, 0.0269, 0.0326, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0198, 0.0126, 0.0131, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0040, 0.0036, 0.0061], device='cuda:0')
+2023-03-02 15:39:58,106 INFO [train.py:968] (0/2) Epoch 5, batch 13050, giga_loss[loss=0.3131, simple_loss=0.379, pruned_loss=0.1236, over 28730.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3717, pruned_loss=0.121, over 5652574.25 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3734, pruned_loss=0.1156, over 5749095.89 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3753, pruned_loss=0.1247, over 5642767.80 frames. ], batch size: 92, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:40:29,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.051e+02 1.273e+03 1.718e+03 2.588e+03 6.297e+03, threshold=3.436e+03, percent-clipped=12.0
+2023-03-02 15:40:33,080 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:40:35,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:40:49,706 INFO [train.py:968] (0/2) Epoch 5, batch 13100, libri_loss[loss=0.4008, simple_loss=0.4446, pruned_loss=0.1785, over 28593.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3715, pruned_loss=0.12, over 5654123.57 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3729, pruned_loss=0.1155, over 5751977.25 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3748, pruned_loss=0.1231, over 5641571.39 frames. ], batch size: 106, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:41:02,414 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:41:15,698 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:41:34,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1721, 1.0701, 0.9119, 1.3236], device='cuda:0'), covar=tensor([0.0775, 0.0342, 0.0360, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0199, 0.0126, 0.0131, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0040, 0.0036, 0.0061], device='cuda:0')
+2023-03-02 15:41:38,432 INFO [train.py:968] (0/2) Epoch 5, batch 13150, giga_loss[loss=0.3034, simple_loss=0.371, pruned_loss=0.1179, over 28010.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3715, pruned_loss=0.12, over 5645696.32 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3731, pruned_loss=0.1157, over 5744509.18 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.374, pruned_loss=0.1223, over 5640395.06 frames. ], batch size: 412, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:42:07,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.523e+02 1.307e+03 1.731e+03 2.592e+03 1.036e+04, threshold=3.462e+03, percent-clipped=8.0
+2023-03-02 15:42:16,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:42:31,999 INFO [train.py:968] (0/2) Epoch 5, batch 13200, giga_loss[loss=0.242, simple_loss=0.3271, pruned_loss=0.07842, over 28698.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3674, pruned_loss=0.1169, over 5645070.11 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3727, pruned_loss=0.1156, over 5746167.86 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3696, pruned_loss=0.1188, over 5638748.17 frames. ], batch size: 262, lr: 6.49e-03, grad_scale: 8.0
+2023-03-02 15:42:43,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3139, 1.5389, 1.3061, 0.9833], device='cuda:0'), covar=tensor([0.0944, 0.0722, 0.0554, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.1401, 0.1164, 0.1175, 0.1233], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 15:42:52,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:43:02,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195130.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:43:04,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4970, 4.3842, 4.1801, 2.0618], device='cuda:0'), covar=tensor([0.0379, 0.0424, 0.0669, 0.1797], device='cuda:0'), in_proj_covar=tensor([0.0827, 0.0750, 0.0784, 0.0581], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:43:21,215 INFO [train.py:968] (0/2) Epoch 5, batch 13250, giga_loss[loss=0.2582, simple_loss=0.3393, pruned_loss=0.0885, over 28659.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3667, pruned_loss=0.1168, over 5637958.34 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3728, pruned_loss=0.1158, over 5739076.62 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3683, pruned_loss=0.1182, over 5636807.24 frames. ], batch size: 119, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:43:31,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6156, 1.7189, 1.6042, 1.5886], device='cuda:0'), covar=tensor([0.1152, 0.1749, 0.1555, 0.1493], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0723, 0.0620, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 15:43:35,951 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-02 15:43:50,062 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.046e+02 1.361e+03 1.669e+03 2.337e+03 1.200e+04, threshold=3.339e+03, percent-clipped=12.0
+2023-03-02 15:44:10,788 INFO [train.py:968] (0/2) Epoch 5, batch 13300, giga_loss[loss=0.2655, simple_loss=0.3429, pruned_loss=0.09406, over 28697.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.366, pruned_loss=0.1159, over 5637478.53 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3726, pruned_loss=0.1157, over 5738045.37 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3674, pruned_loss=0.1171, over 5636068.50 frames. ], batch size: 92, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:44:37,514 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:44:39,968 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:44:47,834 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:44:47,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2772, 2.1446, 1.5041, 1.7410], device='cuda:0'), covar=tensor([0.0639, 0.0645, 0.0958, 0.1032], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0441, 0.0499, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:44:56,774 INFO [train.py:968] (0/2) Epoch 5, batch 13350, giga_loss[loss=0.267, simple_loss=0.3504, pruned_loss=0.09178, over 29184.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3641, pruned_loss=0.114, over 5652849.79 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3717, pruned_loss=0.1153, over 5743456.57 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3658, pruned_loss=0.1153, over 5642880.72 frames. ], batch size: 128, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:45:08,798 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:45:12,147 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:45:14,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:45:22,404 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195273.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:45:25,871 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195276.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:45:29,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.208e+02 1.332e+03 1.767e+03 2.577e+03 5.396e+03, threshold=3.534e+03, percent-clipped=7.0
+2023-03-02 15:45:37,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3866, 1.9812, 1.5656, 1.6107], device='cuda:0'), covar=tensor([0.0776, 0.0278, 0.0311, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0200, 0.0126, 0.0131, 0.0234], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0040, 0.0037, 0.0062], device='cuda:0')
+2023-03-02 15:45:44,517 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:45:49,068 INFO [train.py:968] (0/2) Epoch 5, batch 13400, giga_loss[loss=0.2482, simple_loss=0.3072, pruned_loss=0.0946, over 24118.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3609, pruned_loss=0.1114, over 5640540.17 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3716, pruned_loss=0.1153, over 5736340.03 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3622, pruned_loss=0.1125, over 5637699.95 frames. ], batch size: 705, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:45:54,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195305.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:46:40,765 INFO [train.py:968] (0/2) Epoch 5, batch 13450, libri_loss[loss=0.3469, simple_loss=0.3939, pruned_loss=0.1499, over 29523.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3569, pruned_loss=0.1093, over 5644052.94 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3711, pruned_loss=0.1151, over 5740067.44 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.358, pruned_loss=0.1101, over 5636365.14 frames. ], batch size: 82, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:47:00,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3926, 1.5629, 1.5492, 1.4878], device='cuda:0'), covar=tensor([0.1066, 0.1448, 0.1303, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0720, 0.0618, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 15:47:09,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.021e+02 1.343e+03 1.786e+03 2.323e+03 4.787e+03, threshold=3.571e+03, percent-clipped=10.0
+2023-03-02 15:47:33,476 INFO [train.py:968] (0/2) Epoch 5, batch 13500, giga_loss[loss=0.2477, simple_loss=0.3212, pruned_loss=0.08708, over 28880.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.354, pruned_loss=0.1078, over 5659111.09 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3702, pruned_loss=0.1148, over 5744269.73 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3554, pruned_loss=0.1086, over 5647470.00 frames. ], batch size: 145, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:47:35,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:47:38,454 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:48:09,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0110, 1.9283, 1.4256, 1.5634], device='cuda:0'), covar=tensor([0.0644, 0.0561, 0.0904, 0.0956], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0446, 0.0503, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 15:48:14,584 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195437.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:48:26,846 INFO [train.py:968] (0/2) Epoch 5, batch 13550, giga_loss[loss=0.2905, simple_loss=0.3616, pruned_loss=0.1097, over 28310.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3537, pruned_loss=0.1084, over 5646179.07 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3698, pruned_loss=0.1146, over 5744592.27 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3548, pruned_loss=0.1091, over 5634998.93 frames. ], batch size: 368, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:48:43,279 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:48:59,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.366e+02 1.573e+03 2.077e+03 3.146e+03 8.511e+03, threshold=4.154e+03, percent-clipped=13.0
+2023-03-02 15:49:27,822 INFO [train.py:968] (0/2) Epoch 5, batch 13600, giga_loss[loss=0.3562, simple_loss=0.4036, pruned_loss=0.1544, over 26794.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3558, pruned_loss=0.1098, over 5646085.66 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3694, pruned_loss=0.1144, over 5743563.01 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3569, pruned_loss=0.1104, over 5636801.13 frames. ], batch size: 555, lr: 6.49e-03, grad_scale: 8.0
+2023-03-02 15:50:16,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:50:20,632 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7073, 1.7114, 1.6649, 1.5909], device='cuda:0'), covar=tensor([0.0953, 0.2000, 0.1283, 0.1558], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0726, 0.0622, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 15:50:20,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:50:22,108 INFO [train.py:968] (0/2) Epoch 5, batch 13650, giga_loss[loss=0.3108, simple_loss=0.3899, pruned_loss=0.1159, over 28918.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3577, pruned_loss=0.1092, over 5650911.87 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3687, pruned_loss=0.1141, over 5746945.74 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.359, pruned_loss=0.11, over 5638550.67 frames. ], batch size: 145, lr: 6.49e-03, grad_scale: 8.0
+2023-03-02 15:50:43,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2894, 1.4827, 1.2620, 1.4983], device='cuda:0'), covar=tensor([0.2106, 0.1750, 0.1816, 0.1756], device='cuda:0'), in_proj_covar=tensor([0.1115, 0.0855, 0.0983, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 15:50:52,892 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:50:55,915 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.566e+02 1.382e+03 1.887e+03 2.582e+03 5.182e+03, threshold=3.773e+03, percent-clipped=3.0
+2023-03-02 15:51:16,511 INFO [train.py:968] (0/2) Epoch 5, batch 13700, giga_loss[loss=0.3048, simple_loss=0.3871, pruned_loss=0.1112, over 28933.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3576, pruned_loss=0.1087, over 5661667.64 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3678, pruned_loss=0.1138, over 5745923.54 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.359, pruned_loss=0.1094, over 5648869.65 frames. ], batch size: 164, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:51:33,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:52:18,434 INFO [train.py:968] (0/2) Epoch 5, batch 13750, giga_loss[loss=0.2442, simple_loss=0.3259, pruned_loss=0.08125, over 28709.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3582, pruned_loss=0.1092, over 5664315.04 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3679, pruned_loss=0.1139, over 5746594.45 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.359, pruned_loss=0.1095, over 5652016.22 frames. ], batch size: 262, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:52:50,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.006e+02 1.374e+03 1.765e+03 2.522e+03 6.531e+03, threshold=3.530e+03, percent-clipped=11.0
+2023-03-02 15:53:15,750 INFO [train.py:968] (0/2) Epoch 5, batch 13800, giga_loss[loss=0.3062, simple_loss=0.3715, pruned_loss=0.1205, over 27813.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3549, pruned_loss=0.1066, over 5665183.17 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.367, pruned_loss=0.1135, over 5739086.07 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3561, pruned_loss=0.107, over 5659378.97 frames. ], batch size: 476, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:53:35,896 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-02 15:54:12,277 INFO [train.py:968] (0/2) Epoch 5, batch 13850, giga_loss[loss=0.2612, simple_loss=0.3448, pruned_loss=0.08882, over 28972.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3538, pruned_loss=0.1046, over 5671866.54 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3666, pruned_loss=0.1135, over 5744895.11 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3547, pruned_loss=0.1047, over 5658897.05 frames. ], batch size: 155, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:54:22,066 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:54:24,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:54:24,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3915, 1.5923, 1.5070, 1.4768], device='cuda:0'), covar=tensor([0.1877, 0.2505, 0.1542, 0.1464], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0734, 0.0769, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 15:54:29,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9451, 1.1653, 3.4444, 2.9237], device='cuda:0'), covar=tensor([0.1530, 0.2220, 0.0393, 0.0961], device='cuda:0'), in_proj_covar=tensor([0.0561, 0.0522, 0.0730, 0.0589], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 15:54:46,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:54:47,695 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.666e+02 1.340e+03 1.807e+03 2.476e+03 6.753e+03, threshold=3.614e+03, percent-clipped=12.0
+2023-03-02 15:54:54,394 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:54:59,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:55:13,675 INFO [train.py:968] (0/2) Epoch 5, batch 13900, giga_loss[loss=0.2701, simple_loss=0.3372, pruned_loss=0.1015, over 28233.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3519, pruned_loss=0.1044, over 5657202.34 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3665, pruned_loss=0.1135, over 5737853.35 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3524, pruned_loss=0.1043, over 5651392.56 frames. ], batch size: 412, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:55:25,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195812.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:55:58,935 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:56:10,887 INFO [train.py:968] (0/2) Epoch 5, batch 13950, giga_loss[loss=0.2454, simple_loss=0.3265, pruned_loss=0.08212, over 28897.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3502, pruned_loss=0.1039, over 5666256.90 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3664, pruned_loss=0.1135, over 5738968.46 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3504, pruned_loss=0.1037, over 5659794.76 frames. ], batch size: 174, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:56:48,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.512e+02 1.185e+03 1.794e+03 2.321e+03 4.857e+03, threshold=3.588e+03, percent-clipped=7.0
+2023-03-02 15:57:07,909 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-02 15:57:09,175 INFO [train.py:968] (0/2) Epoch 5, batch 14000, giga_loss[loss=0.2722, simple_loss=0.3258, pruned_loss=0.1094, over 24458.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3489, pruned_loss=0.1034, over 5666181.34 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3664, pruned_loss=0.1135, over 5742305.55 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3487, pruned_loss=0.103, over 5656296.59 frames. ], batch size: 705, lr: 6.48e-03, grad_scale: 8.0
+2023-03-02 15:57:32,663 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:57:36,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:58:07,373 INFO [train.py:968] (0/2) Epoch 5, batch 14050, giga_loss[loss=0.2843, simple_loss=0.359, pruned_loss=0.1048, over 27461.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3504, pruned_loss=0.104, over 5658045.64 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3658, pruned_loss=0.1133, over 5740992.32 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3505, pruned_loss=0.1037, over 5649864.55 frames. ], batch size: 472, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:58:13,889 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:58:15,728 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195955.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:58:20,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195958.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:58:46,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.662e+02 1.372e+03 2.084e+03 3.006e+03 4.660e+03, threshold=4.168e+03, percent-clipped=9.0
+2023-03-02 15:58:47,932 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:58:51,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:58:54,296 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195987.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:59:09,490 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-196000.pt
+2023-03-02 15:59:09,766 INFO [train.py:968] (0/2) Epoch 5, batch 14100, giga_loss[loss=0.2594, simple_loss=0.3369, pruned_loss=0.09097, over 28775.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3522, pruned_loss=0.1045, over 5660678.89 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3655, pruned_loss=0.1131, over 5741140.75 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3523, pruned_loss=0.1042, over 5652389.05 frames. ], batch size: 99, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:59:25,739 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:59:38,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4362, 1.5114, 1.4327, 1.4041], device='cuda:0'), covar=tensor([0.1121, 0.1873, 0.1657, 0.1453], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0724, 0.0623, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 16:00:11,899 INFO [train.py:968] (0/2) Epoch 5, batch 14150, giga_loss[loss=0.2936, simple_loss=0.3634, pruned_loss=0.1119, over 28880.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3482, pruned_loss=0.1019, over 5657513.15 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3648, pruned_loss=0.1128, over 5728287.01 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3484, pruned_loss=0.1017, over 5659862.71 frames. ], batch size: 227, lr: 6.48e-03, grad_scale: 2.0
+2023-03-02 16:00:55,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.784e+02 1.301e+03 1.921e+03 2.377e+03 5.854e+03, threshold=3.842e+03, percent-clipped=2.0
+2023-03-02 16:01:05,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2880, 1.4407, 1.2341, 1.0606], device='cuda:0'), covar=tensor([0.0805, 0.0730, 0.0464, 0.0705], device='cuda:0'), in_proj_covar=tensor([0.1397, 0.1127, 0.1134, 0.1223], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 16:01:08,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6674, 1.0713, 2.8172, 2.5529], device='cuda:0'), covar=tensor([0.1956, 0.2442, 0.0868, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0563, 0.0522, 0.0728, 0.0587], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 16:01:17,626 INFO [train.py:968] (0/2) Epoch 5, batch 14200, giga_loss[loss=0.2802, simple_loss=0.3507, pruned_loss=0.1048, over 28182.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3492, pruned_loss=0.1028, over 5666351.82 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3642, pruned_loss=0.1126, over 5726542.90 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3497, pruned_loss=0.1026, over 5668460.23 frames. ], batch size: 412, lr: 6.48e-03, grad_scale: 2.0
+2023-03-02 16:02:25,669 INFO [train.py:968] (0/2) Epoch 5, batch 14250, giga_loss[loss=0.2939, simple_loss=0.3798, pruned_loss=0.104, over 29068.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3532, pruned_loss=0.1038, over 5669469.37 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3636, pruned_loss=0.1123, over 5731714.09 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3538, pruned_loss=0.1037, over 5664974.06 frames. ], batch size: 155, lr: 6.48e-03, grad_scale: 2.0
+2023-03-02 16:02:37,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=196160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:02:59,220 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 16:03:04,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.005e+02 1.292e+03 1.818e+03 2.613e+03 1.003e+04, threshold=3.636e+03, percent-clipped=13.0
+2023-03-02 16:03:23,894 INFO [train.py:968] (0/2) Epoch 5, batch 14300, giga_loss[loss=0.2682, simple_loss=0.3579, pruned_loss=0.08929, over 28552.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3555, pruned_loss=0.1027, over 5673682.76 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3632, pruned_loss=0.112, over 5735233.84 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3561, pruned_loss=0.1026, over 5665400.09 frames. ], batch size: 307, lr: 6.48e-03, grad_scale: 2.0
+2023-03-02 16:04:21,451 INFO [train.py:968] (0/2) Epoch 5, batch 14350, giga_loss[loss=0.2556, simple_loss=0.348, pruned_loss=0.08162, over 29007.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3567, pruned_loss=0.1021, over 5669887.81 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3634, pruned_loss=0.1121, over 5733352.63 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3568, pruned_loss=0.1018, over 5663845.87 frames. ], batch size: 136, lr: 6.48e-03, grad_scale: 2.0
+2023-03-02 16:04:45,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3700, 1.4546, 1.1833, 1.8258], device='cuda:0'), covar=tensor([0.2334, 0.2183, 0.2162, 0.2263], device='cuda:0'), in_proj_covar=tensor([0.1122, 0.0856, 0.0988, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 16:05:03,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.865e+02 1.243e+03 1.682e+03 2.170e+03 3.854e+03, threshold=3.364e+03, percent-clipped=2.0
+2023-03-02 16:05:23,476 INFO [train.py:968] (0/2) Epoch 5, batch 14400, giga_loss[loss=0.2745, simple_loss=0.355, pruned_loss=0.09701, over 28578.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3557, pruned_loss=0.101, over 5671561.23 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3635, pruned_loss=0.1121, over 5734922.17 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3557, pruned_loss=0.1006, over 5664822.85 frames. ], batch size: 307, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:05:27,675 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:05:30,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:06:08,418 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:06:28,129 INFO [train.py:968] (0/2) Epoch 5, batch 14450, giga_loss[loss=0.2663, simple_loss=0.3431, pruned_loss=0.09472, over 28879.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3557, pruned_loss=0.1018, over 5667637.13 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3634, pruned_loss=0.112, over 5732531.07 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3557, pruned_loss=0.1015, over 5663435.90 frames. ], batch size: 164, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:07:10,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.625e+02 1.242e+03 1.609e+03 2.080e+03 4.103e+03, threshold=3.218e+03, percent-clipped=6.0
+2023-03-02 16:07:33,176 INFO [train.py:968] (0/2) Epoch 5, batch 14500, giga_loss[loss=0.2775, simple_loss=0.3553, pruned_loss=0.09988, over 28927.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3547, pruned_loss=0.102, over 5683861.09 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3627, pruned_loss=0.1116, over 5735851.12 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3552, pruned_loss=0.102, over 5676296.07 frames. ], batch size: 145, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:08:00,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2196, 1.4258, 1.1874, 1.2676], device='cuda:0'), covar=tensor([0.2042, 0.1928, 0.1961, 0.1883], device='cuda:0'), in_proj_covar=tensor([0.1109, 0.0846, 0.0981, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 16:08:50,031 INFO [train.py:968] (0/2) Epoch 5, batch 14550, libri_loss[loss=0.2721, simple_loss=0.3455, pruned_loss=0.09931, over 29174.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3546, pruned_loss=0.103, over 5684241.48 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3626, pruned_loss=0.1115, over 5737204.29 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.355, pruned_loss=0.103, over 5676413.88 frames. ], batch size: 97, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:09:10,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0346, 1.6770, 1.4181, 1.3484], device='cuda:0'), covar=tensor([0.1442, 0.1824, 0.1175, 0.1269], device='cuda:0'), in_proj_covar=tensor([0.0771, 0.0732, 0.0770, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 16:09:40,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.937e+02 1.236e+03 1.580e+03 1.990e+03 4.721e+03, threshold=3.159e+03, percent-clipped=3.0
+2023-03-02 16:09:43,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1678, 1.3743, 1.1556, 1.3927], device='cuda:0'), covar=tensor([0.2476, 0.2096, 0.2259, 0.2043], device='cuda:0'), in_proj_covar=tensor([0.1114, 0.0851, 0.0987, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 16:10:07,133 INFO [train.py:968] (0/2) Epoch 5, batch 14600, giga_loss[loss=0.2707, simple_loss=0.3262, pruned_loss=0.1076, over 23698.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.351, pruned_loss=0.1009, over 5678822.31 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3622, pruned_loss=0.1113, over 5738855.06 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3515, pruned_loss=0.101, over 5670706.21 frames. ], batch size: 705, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:11:07,203 INFO [train.py:968] (0/2) Epoch 5, batch 14650, giga_loss[loss=0.3758, simple_loss=0.405, pruned_loss=0.1733, over 26837.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3497, pruned_loss=0.1006, over 5685698.68 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3611, pruned_loss=0.1105, over 5743885.66 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3507, pruned_loss=0.101, over 5672014.74 frames. ], batch size: 555, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:11:07,582 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-02 16:11:36,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3713, 1.9069, 1.7002, 1.5757], device='cuda:0'), covar=tensor([0.1548, 0.1859, 0.1193, 0.1258], device='cuda:0'), in_proj_covar=tensor([0.0770, 0.0732, 0.0769, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 16:11:43,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.174e+02 1.316e+03 1.708e+03 2.641e+03 6.569e+03, threshold=3.416e+03, percent-clipped=16.0
+2023-03-02 16:12:03,435 INFO [train.py:968] (0/2) Epoch 5, batch 14700, giga_loss[loss=0.2559, simple_loss=0.3199, pruned_loss=0.0959, over 28547.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3493, pruned_loss=0.1014, over 5679461.67 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3607, pruned_loss=0.1104, over 5740140.62 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.35, pruned_loss=0.1014, over 5668430.78 frames. ], batch size: 85, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:12:22,224 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-02 16:12:28,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2913, 1.3633, 4.2963, 3.3357], device='cuda:0'), covar=tensor([0.1523, 0.2016, 0.0322, 0.0632], device='cuda:0'), in_proj_covar=tensor([0.0562, 0.0518, 0.0725, 0.0585], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 16:12:37,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1705, 3.9876, 3.8657, 1.8203], device='cuda:0'), covar=tensor([0.0405, 0.0477, 0.0771, 0.2187], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0735, 0.0776, 0.0585], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 16:12:59,775 INFO [train.py:968] (0/2) Epoch 5, batch 14750, giga_loss[loss=0.2736, simple_loss=0.3577, pruned_loss=0.09472, over 28961.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3542, pruned_loss=0.1041, over 5673323.39 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3601, pruned_loss=0.1101, over 5733295.88 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.355, pruned_loss=0.1041, over 5668314.60 frames. ], batch size: 155, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:13:21,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2525, 3.1064, 2.9810, 1.4686], device='cuda:0'), covar=tensor([0.0656, 0.0669, 0.0968, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0734, 0.0772, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 16:13:38,728 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.977e+02 1.465e+03 1.873e+03 2.611e+03 4.715e+03, threshold=3.747e+03, percent-clipped=7.0
+2023-03-02 16:13:56,584 INFO [train.py:968] (0/2) Epoch 5, batch 14800, giga_loss[loss=0.2684, simple_loss=0.3347, pruned_loss=0.101, over 29134.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3536, pruned_loss=0.1045, over 5671282.25 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3601, pruned_loss=0.1103, over 5726149.34 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.354, pruned_loss=0.1041, over 5672958.28 frames. ], batch size: 200, lr: 6.47e-03, grad_scale: 8.0
+2023-03-02 16:14:13,683 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=196714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:14:37,999 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-02 16:14:55,705 INFO [train.py:968] (0/2) Epoch 5, batch 14850, giga_loss[loss=0.2721, simple_loss=0.3455, pruned_loss=0.09938, over 29188.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3532, pruned_loss=0.1051, over 5673414.46 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3596, pruned_loss=0.11, over 5722715.83 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3538, pruned_loss=0.1049, over 5675758.86 frames. ], batch size: 113, lr: 6.47e-03, grad_scale: 8.0
+2023-03-02 16:15:00,272 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=196753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:15:35,130 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.699e+02 1.289e+03 1.709e+03 2.351e+03 4.862e+03, threshold=3.418e+03, percent-clipped=5.0
+2023-03-02 16:15:54,433 INFO [train.py:968] (0/2) Epoch 5, batch 14900, giga_loss[loss=0.2796, simple_loss=0.3462, pruned_loss=0.1065, over 26993.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3525, pruned_loss=0.105, over 5677039.67 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3592, pruned_loss=0.1099, over 5725895.19 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3533, pruned_loss=0.1049, over 5674816.95 frames. ], batch size: 555, lr: 6.47e-03, grad_scale: 8.0
+2023-03-02 16:17:02,990 INFO [train.py:968] (0/2) Epoch 5, batch 14950, giga_loss[loss=0.267, simple_loss=0.3526, pruned_loss=0.09069, over 28798.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3536, pruned_loss=0.1048, over 5679154.27 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3589, pruned_loss=0.1098, over 5727653.13 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3545, pruned_loss=0.1047, over 5675446.18 frames. ], batch size: 263, lr: 6.47e-03, grad_scale: 8.0
+2023-03-02 16:17:42,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5152, 2.3311, 1.6210, 0.5865], device='cuda:0'), covar=tensor([0.2506, 0.1209, 0.2063, 0.2646], device='cuda:0'), in_proj_covar=tensor([0.1369, 0.1283, 0.1376, 0.1144], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 16:17:50,204 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-03-02 16:17:55,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.430e+02 1.545e+03 1.904e+03 2.573e+03 6.653e+03, threshold=3.809e+03, percent-clipped=13.0
+2023-03-02 16:18:17,900 INFO [train.py:968] (0/2) Epoch 5, batch 15000, giga_loss[loss=0.2963, simple_loss=0.368, pruned_loss=0.1123, over 28431.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3539, pruned_loss=0.1042, over 5676230.77 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3586, pruned_loss=0.1097, over 5730570.79 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3548, pruned_loss=0.1042, over 5669552.59 frames. ], batch size: 369, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:18:17,905 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 16:18:25,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5647, 1.4662, 1.2399, 1.3151], device='cuda:0'), covar=tensor([0.0599, 0.0505, 0.0945, 0.0770], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0448, 0.0508, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 16:18:26,278 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2221, simple_loss=0.3187, pruned_loss=0.06272, over 944034.00 frames. 
+2023-03-02 16:18:26,278 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 16:18:39,372 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6219, 1.5471, 1.2252, 1.3554], device='cuda:0'), covar=tensor([0.0688, 0.0544, 0.0914, 0.0923], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0448, 0.0508, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 16:18:50,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5668, 1.7005, 1.6212, 1.5494], device='cuda:0'), covar=tensor([0.1040, 0.1702, 0.1487, 0.1453], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0737, 0.0632, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 16:19:37,502 INFO [train.py:968] (0/2) Epoch 5, batch 15050, giga_loss[loss=0.271, simple_loss=0.3356, pruned_loss=0.1032, over 28929.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3522, pruned_loss=0.1037, over 5657544.95 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3584, pruned_loss=0.1097, over 5715941.08 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.353, pruned_loss=0.1035, over 5663531.40 frames. ], batch size: 213, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:20:09,573 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=196971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:20:22,654 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2793, 1.5039, 1.2564, 1.3138], device='cuda:0'), covar=tensor([0.1873, 0.1666, 0.1609, 0.1695], device='cuda:0'), in_proj_covar=tensor([0.1114, 0.0855, 0.0986, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 16:20:24,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.595e+02 1.315e+03 1.675e+03 2.552e+03 5.838e+03, threshold=3.350e+03, percent-clipped=7.0
+2023-03-02 16:20:47,721 INFO [train.py:968] (0/2) Epoch 5, batch 15100, giga_loss[loss=0.229, simple_loss=0.3075, pruned_loss=0.07528, over 28969.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3465, pruned_loss=0.1015, over 5649798.28 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3586, pruned_loss=0.1098, over 5713119.38 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3469, pruned_loss=0.1013, over 5656496.46 frames. ], batch size: 155, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:21:33,913 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 16:21:52,060 INFO [train.py:968] (0/2) Epoch 5, batch 15150, giga_loss[loss=0.2368, simple_loss=0.2984, pruned_loss=0.08759, over 24176.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3423, pruned_loss=0.09972, over 5655830.12 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3583, pruned_loss=0.1096, over 5716641.23 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3426, pruned_loss=0.09945, over 5656813.37 frames. ], batch size: 705, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:21:57,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8914, 1.1351, 3.6452, 2.9981], device='cuda:0'), covar=tensor([0.1624, 0.2149, 0.0372, 0.0640], device='cuda:0'), in_proj_covar=tensor([0.0561, 0.0525, 0.0726, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 16:22:23,643 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-02 16:22:34,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.504e+02 1.506e+03 2.100e+03 2.778e+03 6.183e+03, threshold=4.200e+03, percent-clipped=10.0
+2023-03-02 16:22:40,480 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:22:53,434 INFO [train.py:968] (0/2) Epoch 5, batch 15200, giga_loss[loss=0.3485, simple_loss=0.4116, pruned_loss=0.1427, over 28879.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3452, pruned_loss=0.1019, over 5653581.91 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3583, pruned_loss=0.1096, over 5716641.23 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3454, pruned_loss=0.1017, over 5654347.18 frames. ], batch size: 284, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:23:03,064 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-02 16:23:21,576 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:23:44,593 INFO [train.py:968] (0/2) Epoch 5, batch 15250, giga_loss[loss=0.2661, simple_loss=0.3472, pruned_loss=0.09252, over 28995.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3462, pruned_loss=0.1024, over 5663671.04 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3577, pruned_loss=0.1093, over 5714730.79 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3463, pruned_loss=0.1022, over 5663299.35 frames. ], batch size: 285, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:23:58,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0719, 5.9188, 5.6498, 2.7103], device='cuda:0'), covar=tensor([0.0300, 0.0401, 0.0693, 0.1680], device='cuda:0'), in_proj_covar=tensor([0.0817, 0.0741, 0.0774, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 16:24:16,835 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:24:28,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.031e+02 1.370e+03 2.100e+03 3.238e+03 1.251e+04, threshold=4.201e+03, percent-clipped=7.0
+2023-03-02 16:24:47,859 INFO [train.py:968] (0/2) Epoch 5, batch 15300, giga_loss[loss=0.2814, simple_loss=0.3562, pruned_loss=0.1033, over 28338.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3427, pruned_loss=0.0998, over 5655174.96 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3568, pruned_loss=0.1089, over 5719438.21 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3433, pruned_loss=0.09984, over 5649130.36 frames. ], batch size: 368, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:25:00,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3797, 2.3132, 2.2290, 2.1748], device='cuda:0'), covar=tensor([0.0928, 0.1801, 0.1338, 0.1403], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0725, 0.0626, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 16:25:25,703 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:25:28,577 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:25:43,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:25:48,130 INFO [train.py:968] (0/2) Epoch 5, batch 15350, giga_loss[loss=0.2945, simple_loss=0.3639, pruned_loss=0.1125, over 28840.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3428, pruned_loss=0.09932, over 5660898.81 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3572, pruned_loss=0.1092, over 5713792.58 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3426, pruned_loss=0.09886, over 5659658.34 frames. ], batch size: 263, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:26:05,113 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:26:14,376 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:26:19,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9842, 1.2591, 3.1335, 3.0002], device='cuda:0'), covar=tensor([0.1394, 0.2047, 0.0390, 0.0669], device='cuda:0'), in_proj_covar=tensor([0.0560, 0.0523, 0.0726, 0.0582], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 16:26:20,310 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:26:20,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:26:36,016 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.157e+02 1.272e+03 1.697e+03 2.546e+03 8.988e+03, threshold=3.393e+03, percent-clipped=4.0
+2023-03-02 16:26:53,971 INFO [train.py:968] (0/2) Epoch 5, batch 15400, libri_loss[loss=0.2573, simple_loss=0.3145, pruned_loss=0.1, over 29364.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.341, pruned_loss=0.0989, over 5658051.13 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3565, pruned_loss=0.109, over 5719418.62 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3411, pruned_loss=0.09841, over 5650433.04 frames. ], batch size: 67, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:26:58,181 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:27:42,081 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2481, 1.6514, 1.2765, 0.6245], device='cuda:0'), covar=tensor([0.1920, 0.1198, 0.1366, 0.2372], device='cuda:0'), in_proj_covar=tensor([0.1359, 0.1287, 0.1366, 0.1134], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 16:27:54,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:27:59,758 INFO [train.py:968] (0/2) Epoch 5, batch 15450, giga_loss[loss=0.2749, simple_loss=0.3515, pruned_loss=0.09913, over 29116.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3417, pruned_loss=0.09865, over 5655846.16 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3565, pruned_loss=0.109, over 5721981.71 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3414, pruned_loss=0.09808, over 5646651.69 frames. ], batch size: 285, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:28:44,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.662e+02 1.403e+03 1.722e+03 2.429e+03 1.151e+04, threshold=3.443e+03, percent-clipped=8.0
+2023-03-02 16:29:02,109 INFO [train.py:968] (0/2) Epoch 5, batch 15500, giga_loss[loss=0.2653, simple_loss=0.3409, pruned_loss=0.09488, over 28839.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3422, pruned_loss=0.09901, over 5653127.92 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3566, pruned_loss=0.109, over 5716277.19 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3417, pruned_loss=0.0984, over 5650782.89 frames. ], batch size: 227, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:30:09,555 INFO [train.py:968] (0/2) Epoch 5, batch 15550, giga_loss[loss=0.2843, simple_loss=0.3578, pruned_loss=0.1054, over 28753.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3419, pruned_loss=0.0996, over 5652434.97 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3566, pruned_loss=0.1091, over 5718482.82 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3412, pruned_loss=0.09889, over 5647607.25 frames. ], batch size: 262, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:30:46,041 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-02 16:30:55,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.658e+02 1.210e+03 1.450e+03 2.082e+03 4.741e+03, threshold=2.901e+03, percent-clipped=7.0
+2023-03-02 16:31:02,758 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:31:06,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:31:14,734 INFO [train.py:968] (0/2) Epoch 5, batch 15600, libri_loss[loss=0.3173, simple_loss=0.3773, pruned_loss=0.1286, over 26135.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.342, pruned_loss=0.09762, over 5661766.08 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3565, pruned_loss=0.109, over 5717061.78 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3415, pruned_loss=0.09699, over 5658963.36 frames. ], batch size: 136, lr: 6.45e-03, grad_scale: 8.0
+2023-03-02 16:31:42,478 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:32:15,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:32:20,945 INFO [train.py:968] (0/2) Epoch 5, batch 15650, giga_loss[loss=0.2532, simple_loss=0.3211, pruned_loss=0.0927, over 24250.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3457, pruned_loss=0.0989, over 5658029.51 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3566, pruned_loss=0.1091, over 5715830.19 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3451, pruned_loss=0.09822, over 5656344.57 frames. ], batch size: 705, lr: 6.45e-03, grad_scale: 8.0
+2023-03-02 16:33:07,351 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.027e+02 1.433e+03 2.030e+03 2.684e+03 4.942e+03, threshold=4.061e+03, percent-clipped=21.0
+2023-03-02 16:33:23,948 INFO [train.py:968] (0/2) Epoch 5, batch 15700, giga_loss[loss=0.3153, simple_loss=0.3688, pruned_loss=0.1309, over 26781.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3482, pruned_loss=0.1003, over 5658203.49 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3567, pruned_loss=0.1091, over 5719110.17 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3475, pruned_loss=0.09963, over 5652923.44 frames. ], batch size: 555, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:33:54,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:34:24,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:34:24,819 INFO [train.py:968] (0/2) Epoch 5, batch 15750, giga_loss[loss=0.254, simple_loss=0.3311, pruned_loss=0.08843, over 29058.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.349, pruned_loss=0.1008, over 5673836.29 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3566, pruned_loss=0.1091, over 5723982.32 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3483, pruned_loss=0.1, over 5663812.83 frames. ], batch size: 128, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:35:10,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.248e+02 1.469e+03 2.004e+03 2.772e+03 7.762e+03, threshold=4.008e+03, percent-clipped=9.0
+2023-03-02 16:35:16,227 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:35:20,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:35:28,773 INFO [train.py:968] (0/2) Epoch 5, batch 15800, libri_loss[loss=0.2475, simple_loss=0.3226, pruned_loss=0.08621, over 29558.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3477, pruned_loss=0.0997, over 5682785.96 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3562, pruned_loss=0.1089, over 5726742.91 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3474, pruned_loss=0.09915, over 5671709.71 frames. ], batch size: 76, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:35:35,316 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6244, 3.5824, 1.5679, 1.5785], device='cuda:0'), covar=tensor([0.0804, 0.0278, 0.0837, 0.1261], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0469, 0.0312, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 16:35:51,378 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-02 16:35:54,607 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:36:03,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0976, 1.2699, 1.2993, 1.1446], device='cuda:0'), covar=tensor([0.0848, 0.0945, 0.1379, 0.1028], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0725, 0.0625, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 16:36:30,085 INFO [train.py:968] (0/2) Epoch 5, batch 15850, giga_loss[loss=0.2723, simple_loss=0.3246, pruned_loss=0.11, over 24540.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3452, pruned_loss=0.09827, over 5691703.70 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3562, pruned_loss=0.1092, over 5733844.59 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3446, pruned_loss=0.09714, over 5674879.69 frames. ], batch size: 705, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:36:48,608 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:36:52,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:37:08,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.724e+02 1.245e+03 1.671e+03 2.788e+03 8.421e+03, threshold=3.342e+03, percent-clipped=9.0
+2023-03-02 16:37:18,937 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:37:21,906 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:37:23,374 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:37:26,547 INFO [train.py:968] (0/2) Epoch 5, batch 15900, giga_loss[loss=0.2466, simple_loss=0.3271, pruned_loss=0.08309, over 28841.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3436, pruned_loss=0.09816, over 5687500.65 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3556, pruned_loss=0.1089, over 5738519.27 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3432, pruned_loss=0.09717, over 5668057.30 frames. ], batch size: 164, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:37:38,347 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197808.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 16:37:56,671 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:38:31,580 INFO [train.py:968] (0/2) Epoch 5, batch 15950, giga_loss[loss=0.2674, simple_loss=0.3286, pruned_loss=0.1031, over 24492.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3432, pruned_loss=0.09809, over 5683166.59 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3555, pruned_loss=0.1087, over 5739785.22 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.343, pruned_loss=0.09732, over 5666345.69 frames. ], batch size: 705, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:39:07,501 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-02 16:39:13,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.360e+02 1.396e+03 1.835e+03 2.756e+03 9.187e+03, threshold=3.669e+03, percent-clipped=16.0
+2023-03-02 16:39:29,468 INFO [train.py:968] (0/2) Epoch 5, batch 16000, giga_loss[loss=0.2981, simple_loss=0.3706, pruned_loss=0.1128, over 28658.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3461, pruned_loss=0.09961, over 5686521.94 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.355, pruned_loss=0.1085, over 5743388.37 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.346, pruned_loss=0.09893, over 5668268.25 frames. ], batch size: 262, lr: 6.45e-03, grad_scale: 8.0
+2023-03-02 16:39:45,111 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:40:17,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5691, 1.0130, 2.8732, 2.6200], device='cuda:0'), covar=tensor([0.1613, 0.2095, 0.0498, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0565, 0.0524, 0.0730, 0.0583], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 16:40:23,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:40:39,616 INFO [train.py:968] (0/2) Epoch 5, batch 16050, giga_loss[loss=0.2778, simple_loss=0.3536, pruned_loss=0.1009, over 28936.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3469, pruned_loss=0.1006, over 5683443.74 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3547, pruned_loss=0.1084, over 5745806.22 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3469, pruned_loss=0.09992, over 5665257.97 frames. ], batch size: 199, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:40:39,946 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3920, 1.8519, 1.5883, 1.7570], device='cuda:0'), covar=tensor([0.0595, 0.0709, 0.0891, 0.1004], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0444, 0.0505, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 16:40:52,580 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-02 16:41:03,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2713, 1.5642, 1.3154, 1.5237], device='cuda:0'), covar=tensor([0.0724, 0.0405, 0.0344, 0.0768], device='cuda:0'), in_proj_covar=tensor([0.0199, 0.0125, 0.0131, 0.0235], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0040, 0.0037, 0.0062], device='cuda:0')
+2023-03-02 16:41:21,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.034e+02 1.271e+03 1.690e+03 2.266e+03 5.268e+03, threshold=3.381e+03, percent-clipped=6.0
+2023-03-02 16:41:36,995 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-198000.pt
+2023-03-02 16:41:37,290 INFO [train.py:968] (0/2) Epoch 5, batch 16100, giga_loss[loss=0.2773, simple_loss=0.3501, pruned_loss=0.1022, over 28775.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3496, pruned_loss=0.1022, over 5683720.59 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3546, pruned_loss=0.1084, over 5744987.63 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3495, pruned_loss=0.1016, over 5668000.42 frames. ], batch size: 99, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:42:39,966 INFO [train.py:968] (0/2) Epoch 5, batch 16150, giga_loss[loss=0.3088, simple_loss=0.3834, pruned_loss=0.1171, over 28498.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3525, pruned_loss=0.1029, over 5688632.88 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3544, pruned_loss=0.1082, over 5747403.41 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3526, pruned_loss=0.1025, over 5673338.33 frames. ], batch size: 336, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:43:20,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.293e+02 1.517e+03 2.223e+03 3.187e+03 7.086e+03, threshold=4.446e+03, percent-clipped=23.0
+2023-03-02 16:43:36,815 INFO [train.py:968] (0/2) Epoch 5, batch 16200, giga_loss[loss=0.2651, simple_loss=0.3437, pruned_loss=0.0933, over 28995.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3541, pruned_loss=0.1041, over 5682053.49 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3538, pruned_loss=0.108, over 5740726.32 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3547, pruned_loss=0.1037, over 5674659.19 frames. ], batch size: 164, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:44:01,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2357, 3.0852, 2.9519, 1.3412], device='cuda:0'), covar=tensor([0.0788, 0.0673, 0.0989, 0.2246], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0730, 0.0772, 0.0579], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 16:44:46,491 INFO [train.py:968] (0/2) Epoch 5, batch 16250, giga_loss[loss=0.2347, simple_loss=0.3178, pruned_loss=0.07581, over 28924.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3522, pruned_loss=0.1028, over 5686968.19 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3535, pruned_loss=0.1078, over 5743007.76 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.353, pruned_loss=0.1026, over 5677910.29 frames. ], batch size: 174, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:44:58,520 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 16:45:01,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9587, 1.9186, 1.8930, 1.8701], device='cuda:0'), covar=tensor([0.0977, 0.1895, 0.1299, 0.1398], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0730, 0.0626, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 16:45:23,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198183.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 16:45:26,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.263e+02 1.274e+03 1.680e+03 2.236e+03 1.259e+04, threshold=3.360e+03, percent-clipped=7.0
+2023-03-02 16:45:42,266 INFO [train.py:968] (0/2) Epoch 5, batch 16300, giga_loss[loss=0.2698, simple_loss=0.3465, pruned_loss=0.09655, over 29041.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3504, pruned_loss=0.1018, over 5688102.59 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3535, pruned_loss=0.1076, over 5735225.85 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3509, pruned_loss=0.1016, over 5685304.95 frames. ], batch size: 199, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:46:22,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1880, 0.9756, 0.8030, 1.3494], device='cuda:0'), covar=tensor([0.0771, 0.0379, 0.0380, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0199, 0.0126, 0.0132, 0.0235], device='cuda:0'), out_proj_covar=tensor([0.0057, 0.0041, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 16:46:48,514 INFO [train.py:968] (0/2) Epoch 5, batch 16350, libri_loss[loss=0.295, simple_loss=0.3738, pruned_loss=0.1081, over 26154.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3502, pruned_loss=0.102, over 5669918.33 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3538, pruned_loss=0.1077, over 5735300.09 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3503, pruned_loss=0.1016, over 5667122.55 frames. ], batch size: 136, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:46:56,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4900, 2.1160, 1.5306, 0.7477], device='cuda:0'), covar=tensor([0.2538, 0.1435, 0.2056, 0.2599], device='cuda:0'), in_proj_covar=tensor([0.1373, 0.1293, 0.1364, 0.1126], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 16:47:29,571 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:47:30,845 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:47:33,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.329e+02 1.264e+03 1.689e+03 2.090e+03 4.165e+03, threshold=3.379e+03, percent-clipped=4.0
+2023-03-02 16:47:50,965 INFO [train.py:968] (0/2) Epoch 5, batch 16400, giga_loss[loss=0.2448, simple_loss=0.3204, pruned_loss=0.08463, over 28840.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3487, pruned_loss=0.1018, over 5676161.80 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3536, pruned_loss=0.1075, over 5738734.27 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3488, pruned_loss=0.1016, over 5669887.49 frames. ], batch size: 284, lr: 6.44e-03, grad_scale: 8.0
+2023-03-02 16:48:08,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:48:22,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198326.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 16:48:26,261 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198329.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 16:48:52,240 INFO [train.py:968] (0/2) Epoch 5, batch 16450, libri_loss[loss=0.2648, simple_loss=0.3405, pruned_loss=0.0945, over 29540.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3477, pruned_loss=0.1024, over 5672165.60 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3534, pruned_loss=0.1074, over 5740323.23 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3479, pruned_loss=0.1022, over 5663645.13 frames. ], batch size: 79, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:49:01,555 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198358.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 16:49:37,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.888e+02 1.394e+03 2.061e+03 2.735e+03 7.048e+03, threshold=4.122e+03, percent-clipped=16.0
+2023-03-02 16:49:53,186 INFO [train.py:968] (0/2) Epoch 5, batch 16500, giga_loss[loss=0.2822, simple_loss=0.3315, pruned_loss=0.1164, over 24484.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3489, pruned_loss=0.102, over 5676071.04 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3535, pruned_loss=0.1073, over 5743211.84 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3488, pruned_loss=0.1018, over 5665266.33 frames. ], batch size: 705, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:50:25,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:50:28,344 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:50:36,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7174, 2.4270, 1.5922, 0.9258], device='cuda:0'), covar=tensor([0.3190, 0.1715, 0.2099, 0.2998], device='cuda:0'), in_proj_covar=tensor([0.1374, 0.1298, 0.1371, 0.1142], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 16:50:53,155 INFO [train.py:968] (0/2) Epoch 5, batch 16550, giga_loss[loss=0.3206, simple_loss=0.3768, pruned_loss=0.1323, over 26803.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3477, pruned_loss=0.1003, over 5677483.73 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3539, pruned_loss=0.1075, over 5745456.63 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3472, pruned_loss=0.09989, over 5666026.06 frames. ], batch size: 555, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:51:01,414 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:51:04,998 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:51:05,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:51:20,061 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:51:25,181 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-02 16:51:34,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.497e+02 1.274e+03 1.762e+03 2.178e+03 6.198e+03, threshold=3.524e+03, percent-clipped=2.0
+2023-03-02 16:51:35,127 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:51:42,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2390, 1.7782, 1.6048, 1.4714], device='cuda:0'), covar=tensor([0.1623, 0.1969, 0.1291, 0.1327], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0718, 0.0766, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 16:51:49,339 INFO [train.py:968] (0/2) Epoch 5, batch 16600, giga_loss[loss=0.3049, simple_loss=0.3789, pruned_loss=0.1155, over 27524.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3488, pruned_loss=0.09886, over 5682336.76 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3535, pruned_loss=0.1073, over 5748584.23 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3486, pruned_loss=0.0985, over 5669031.73 frames. ], batch size: 472, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:52:46,150 INFO [train.py:968] (0/2) Epoch 5, batch 16650, giga_loss[loss=0.2813, simple_loss=0.3592, pruned_loss=0.1016, over 28404.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3486, pruned_loss=0.09766, over 5685605.73 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3532, pruned_loss=0.1072, over 5751253.00 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3486, pruned_loss=0.09731, over 5671014.47 frames. ], batch size: 369, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:53:32,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.539e+02 1.399e+03 1.752e+03 2.432e+03 8.281e+03, threshold=3.505e+03, percent-clipped=14.0
+2023-03-02 16:53:48,559 INFO [train.py:968] (0/2) Epoch 5, batch 16700, giga_loss[loss=0.2542, simple_loss=0.3392, pruned_loss=0.0846, over 28907.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3486, pruned_loss=0.09761, over 5683356.70 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3533, pruned_loss=0.1073, over 5744616.15 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3485, pruned_loss=0.09707, over 5676147.43 frames. ], batch size: 164, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:54:56,086 INFO [train.py:968] (0/2) Epoch 5, batch 16750, giga_loss[loss=0.2627, simple_loss=0.3367, pruned_loss=0.09431, over 28959.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.349, pruned_loss=0.09825, over 5679234.53 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3531, pruned_loss=0.1072, over 5747726.18 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.349, pruned_loss=0.09771, over 5669425.85 frames. ], batch size: 106, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:55:09,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3760, 1.8735, 1.7730, 1.6075], device='cuda:0'), covar=tensor([0.1531, 0.1733, 0.1147, 0.1249], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0715, 0.0767, 0.0711], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 16:55:11,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:55:54,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.362e+02 1.401e+03 1.778e+03 2.350e+03 8.931e+03, threshold=3.556e+03, percent-clipped=9.0
+2023-03-02 16:56:09,048 INFO [train.py:968] (0/2) Epoch 5, batch 16800, giga_loss[loss=0.256, simple_loss=0.3379, pruned_loss=0.08702, over 28519.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3494, pruned_loss=0.09841, over 5672879.43 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3534, pruned_loss=0.1074, over 5747928.25 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.349, pruned_loss=0.09761, over 5663806.23 frames. ], batch size: 85, lr: 6.44e-03, grad_scale: 8.0
+2023-03-02 16:56:17,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4834, 3.2564, 1.4937, 1.3932], device='cuda:0'), covar=tensor([0.0844, 0.0259, 0.0886, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0471, 0.0313, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 16:57:19,045 INFO [train.py:968] (0/2) Epoch 5, batch 16850, giga_loss[loss=0.3266, simple_loss=0.3882, pruned_loss=0.1325, over 28168.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3489, pruned_loss=0.09703, over 5681661.53 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3531, pruned_loss=0.1071, over 5750866.21 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3488, pruned_loss=0.09647, over 5670649.04 frames. ], batch size: 412, lr: 6.43e-03, grad_scale: 8.0
+2023-03-02 16:58:11,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.266e+02 1.382e+03 1.765e+03 2.315e+03 5.297e+03, threshold=3.530e+03, percent-clipped=6.0
+2023-03-02 16:58:28,207 INFO [train.py:968] (0/2) Epoch 5, batch 16900, giga_loss[loss=0.3211, simple_loss=0.3964, pruned_loss=0.1229, over 28698.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.354, pruned_loss=0.1008, over 5673191.81 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3537, pruned_loss=0.1076, over 5743134.78 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3533, pruned_loss=0.09955, over 5668307.60 frames. ], batch size: 307, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 16:58:30,711 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:58:33,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:58:33,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8745, 1.1021, 0.8535, 0.6068], device='cuda:0'), covar=tensor([0.0972, 0.0851, 0.0613, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.1409, 0.1120, 0.1137, 0.1219], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 16:59:14,161 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:59:32,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:59:33,002 INFO [train.py:968] (0/2) Epoch 5, batch 16950, giga_loss[loss=0.2791, simple_loss=0.3575, pruned_loss=0.1004, over 28640.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3548, pruned_loss=0.1007, over 5689398.18 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3536, pruned_loss=0.1076, over 5748140.01 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3543, pruned_loss=0.09949, over 5679519.17 frames. ], batch size: 307, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 17:00:07,812 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-02 17:00:24,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.834e+02 1.359e+03 1.628e+03 2.417e+03 7.508e+03, threshold=3.256e+03, percent-clipped=7.0
+2023-03-02 17:00:29,919 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:00:43,454 INFO [train.py:968] (0/2) Epoch 5, batch 17000, giga_loss[loss=0.2939, simple_loss=0.3622, pruned_loss=0.1128, over 28201.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3532, pruned_loss=0.1005, over 5689094.13 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3535, pruned_loss=0.1075, over 5749397.17 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3529, pruned_loss=0.09957, over 5678771.12 frames. ], batch size: 412, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 17:00:58,203 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-02 17:01:49,428 INFO [train.py:968] (0/2) Epoch 5, batch 17050, giga_loss[loss=0.2514, simple_loss=0.3334, pruned_loss=0.08472, over 28866.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3509, pruned_loss=0.09999, over 5696718.13 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3532, pruned_loss=0.1072, over 5753122.59 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3509, pruned_loss=0.0993, over 5683224.23 frames. ], batch size: 174, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 17:02:43,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.080e+02 1.265e+03 1.914e+03 2.533e+03 5.940e+03, threshold=3.828e+03, percent-clipped=10.0
+2023-03-02 17:02:50,616 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:02:54,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:03:01,561 INFO [train.py:968] (0/2) Epoch 5, batch 17100, giga_loss[loss=0.3198, simple_loss=0.3811, pruned_loss=0.1293, over 28905.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3482, pruned_loss=0.09711, over 5703290.65 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3532, pruned_loss=0.1071, over 5755389.89 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3481, pruned_loss=0.09637, over 5688994.56 frames. ], batch size: 284, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 17:03:11,092 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:03:34,647 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:04:04,021 INFO [train.py:968] (0/2) Epoch 5, batch 17150, giga_loss[loss=0.282, simple_loss=0.3532, pruned_loss=0.1054, over 27559.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3482, pruned_loss=0.09731, over 5700996.96 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3538, pruned_loss=0.1074, over 5754465.90 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3475, pruned_loss=0.0963, over 5689293.72 frames. ], batch size: 472, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 17:04:51,758 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.440e+02 1.269e+03 1.692e+03 2.791e+03 7.109e+03, threshold=3.385e+03, percent-clipped=12.0
+2023-03-02 17:04:54,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6995, 1.9076, 1.4598, 1.8119], device='cuda:0'), covar=tensor([0.0718, 0.0271, 0.0329, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0198, 0.0125, 0.0131, 0.0234], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:0')
+2023-03-02 17:05:05,390 INFO [train.py:968] (0/2) Epoch 5, batch 17200, giga_loss[loss=0.2384, simple_loss=0.3265, pruned_loss=0.07512, over 28570.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3505, pruned_loss=0.099, over 5694437.43 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3537, pruned_loss=0.1074, over 5756742.22 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.35, pruned_loss=0.09805, over 5682342.90 frames. ], batch size: 60, lr: 6.43e-03, grad_scale: 8.0
+2023-03-02 17:06:07,186 INFO [train.py:968] (0/2) Epoch 5, batch 17250, giga_loss[loss=0.2636, simple_loss=0.3419, pruned_loss=0.09267, over 28744.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3533, pruned_loss=0.1011, over 5689638.09 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3536, pruned_loss=0.1074, over 5759066.82 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.353, pruned_loss=0.1002, over 5676755.59 frames. ], batch size: 243, lr: 6.43e-03, grad_scale: 8.0
+2023-03-02 17:06:18,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0525, 2.8857, 1.6278, 1.6042], device='cuda:0'), covar=tensor([0.1202, 0.0552, 0.0771, 0.0926], device='cuda:0'), in_proj_covar=tensor([0.1406, 0.1121, 0.1141, 0.1212], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 17:06:48,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.272e+02 1.497e+03 1.941e+03 2.921e+03 5.841e+03, threshold=3.882e+03, percent-clipped=18.0
+2023-03-02 17:06:51,675 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3499, 1.8342, 1.3669, 1.5715], device='cuda:0'), covar=tensor([0.0751, 0.0287, 0.0347, 0.0860], device='cuda:0'), in_proj_covar=tensor([0.0199, 0.0125, 0.0132, 0.0235], device='cuda:0'), out_proj_covar=tensor([0.0057, 0.0041, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 17:07:05,816 INFO [train.py:968] (0/2) Epoch 5, batch 17300, giga_loss[loss=0.2761, simple_loss=0.3311, pruned_loss=0.1106, over 26779.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3504, pruned_loss=0.1001, over 5689707.47 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3533, pruned_loss=0.1071, over 5762392.91 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3504, pruned_loss=0.09957, over 5675055.52 frames. ], batch size: 555, lr: 6.43e-03, grad_scale: 8.0
+2023-03-02 17:07:57,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9914, 1.2056, 3.5757, 2.8945], device='cuda:0'), covar=tensor([0.1622, 0.2255, 0.0422, 0.0882], device='cuda:0'), in_proj_covar=tensor([0.0569, 0.0527, 0.0738, 0.0588], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 17:07:59,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7514, 1.6398, 1.1805, 1.4036], device='cuda:0'), covar=tensor([0.0620, 0.0536, 0.0961, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0438, 0.0504, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 17:08:02,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:08:06,643 INFO [train.py:968] (0/2) Epoch 5, batch 17350, giga_loss[loss=0.2508, simple_loss=0.3301, pruned_loss=0.08574, over 28021.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3498, pruned_loss=0.1006, over 5687868.80 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3531, pruned_loss=0.107, over 5763251.94 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.35, pruned_loss=0.1002, over 5675183.11 frames. ], batch size: 412, lr: 6.43e-03, grad_scale: 2.0
+2023-03-02 17:08:26,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:08:37,289 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:08:55,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4103, 1.8918, 1.3412, 1.5655], device='cuda:0'), covar=tensor([0.0703, 0.0299, 0.0316, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0200, 0.0126, 0.0132, 0.0235], device='cuda:0'), out_proj_covar=tensor([0.0057, 0.0041, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 17:08:55,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.641e+02 1.350e+03 2.200e+03 3.530e+03 9.912e+03, threshold=4.400e+03, percent-clipped=24.0
+2023-03-02 17:09:05,344 INFO [train.py:968] (0/2) Epoch 5, batch 17400, giga_loss[loss=0.3624, simple_loss=0.4113, pruned_loss=0.1568, over 27479.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3531, pruned_loss=0.1034, over 5694430.32 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3531, pruned_loss=0.107, over 5766326.86 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3532, pruned_loss=0.103, over 5680316.25 frames. ], batch size: 472, lr: 6.43e-03, grad_scale: 2.0
+2023-03-02 17:09:32,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:09:47,408 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 17:09:59,665 INFO [train.py:968] (0/2) Epoch 5, batch 17450, libri_loss[loss=0.2806, simple_loss=0.3537, pruned_loss=0.1037, over 29531.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3628, pruned_loss=0.1097, over 5692415.05 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.353, pruned_loss=0.1069, over 5767706.18 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3631, pruned_loss=0.1096, over 5679254.56 frames. ], batch size: 83, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:10:30,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:10:36,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.176e+02 1.138e+03 1.417e+03 2.027e+03 5.270e+03, threshold=2.834e+03, percent-clipped=2.0
+2023-03-02 17:10:45,136 INFO [train.py:968] (0/2) Epoch 5, batch 17500, giga_loss[loss=0.3551, simple_loss=0.4077, pruned_loss=0.1513, over 28595.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.371, pruned_loss=0.1146, over 5696522.94 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3532, pruned_loss=0.1069, over 5767434.44 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3712, pruned_loss=0.1145, over 5685807.66 frames. ], batch size: 92, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:10:49,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1923, 3.9401, 3.8592, 1.4709], device='cuda:0'), covar=tensor([0.0453, 0.0456, 0.0730, 0.2338], device='cuda:0'), in_proj_covar=tensor([0.0829, 0.0752, 0.0774, 0.0588], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 17:10:53,979 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:10:57,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:11:00,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9656, 1.0780, 3.9501, 3.0261], device='cuda:0'), covar=tensor([0.1721, 0.2368, 0.0364, 0.0705], device='cuda:0'), in_proj_covar=tensor([0.0572, 0.0536, 0.0747, 0.0600], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 17:11:21,320 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:11:30,724 INFO [train.py:968] (0/2) Epoch 5, batch 17550, giga_loss[loss=0.2519, simple_loss=0.317, pruned_loss=0.09344, over 28990.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3697, pruned_loss=0.1151, over 5697357.62 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3534, pruned_loss=0.1068, over 5770569.60 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3702, pruned_loss=0.1153, over 5683986.25 frames. ], batch size: 106, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:11:56,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2488, 1.4744, 1.1369, 0.8148], device='cuda:0'), covar=tensor([0.1115, 0.0830, 0.0591, 0.1013], device='cuda:0'), in_proj_covar=tensor([0.1412, 0.1121, 0.1165, 0.1238], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 17:12:06,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.787e+02 1.125e+03 1.350e+03 1.752e+03 4.059e+03, threshold=2.700e+03, percent-clipped=2.0
+2023-03-02 17:12:12,722 INFO [train.py:968] (0/2) Epoch 5, batch 17600, giga_loss[loss=0.2319, simple_loss=0.3129, pruned_loss=0.07548, over 28944.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3622, pruned_loss=0.112, over 5694735.42 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3537, pruned_loss=0.1071, over 5771508.86 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3623, pruned_loss=0.112, over 5682249.18 frames. ], batch size: 164, lr: 6.42e-03, grad_scale: 4.0
+2023-03-02 17:12:37,819 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:12:39,621 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6064, 1.8341, 1.8183, 1.7387], device='cuda:0'), covar=tensor([0.1208, 0.1384, 0.0922, 0.0978], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0725, 0.0772, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 17:12:40,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:12:57,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8458, 1.0712, 3.6632, 2.9606], device='cuda:0'), covar=tensor([0.1707, 0.2293, 0.0375, 0.0681], device='cuda:0'), in_proj_covar=tensor([0.0563, 0.0527, 0.0737, 0.0596], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 17:12:59,903 INFO [train.py:968] (0/2) Epoch 5, batch 17650, giga_loss[loss=0.2746, simple_loss=0.3302, pruned_loss=0.1095, over 28994.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3547, pruned_loss=0.109, over 5689832.00 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3539, pruned_loss=0.1072, over 5773383.15 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3547, pruned_loss=0.109, over 5677163.85 frames. ], batch size: 100, lr: 6.42e-03, grad_scale: 4.0
+2023-03-02 17:13:05,538 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:13:13,441 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:13:35,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.465e+02 9.730e+02 1.300e+03 2.099e+03 5.526e+03, threshold=2.601e+03, percent-clipped=13.0
+2023-03-02 17:13:43,709 INFO [train.py:968] (0/2) Epoch 5, batch 17700, giga_loss[loss=0.2688, simple_loss=0.3349, pruned_loss=0.1014, over 28316.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3473, pruned_loss=0.1056, over 5695937.75 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3545, pruned_loss=0.1074, over 5776479.43 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3467, pruned_loss=0.1052, over 5680856.48 frames. ], batch size: 368, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:13:58,515 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 17:14:00,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:14:26,543 INFO [train.py:968] (0/2) Epoch 5, batch 17750, giga_loss[loss=0.3249, simple_loss=0.3549, pruned_loss=0.1474, over 26599.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3409, pruned_loss=0.1025, over 5701503.19 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3548, pruned_loss=0.1074, over 5780581.86 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3397, pruned_loss=0.1021, over 5682831.73 frames. ], batch size: 555, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:14:28,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:14:54,505 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-02 17:15:01,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.132e+02 9.673e+02 1.310e+03 1.924e+03 1.003e+04, threshold=2.620e+03, percent-clipped=11.0
+2023-03-02 17:15:04,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:15:07,702 INFO [train.py:968] (0/2) Epoch 5, batch 17800, giga_loss[loss=0.2615, simple_loss=0.3106, pruned_loss=0.1062, over 23941.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3355, pruned_loss=0.09967, over 5702605.46 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.355, pruned_loss=0.1075, over 5781780.97 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3337, pruned_loss=0.09911, over 5684175.23 frames. ], batch size: 705, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:15:49,956 INFO [train.py:968] (0/2) Epoch 5, batch 17850, giga_loss[loss=0.2436, simple_loss=0.3155, pruned_loss=0.08585, over 28927.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.333, pruned_loss=0.09861, over 5708680.40 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3551, pruned_loss=0.1075, over 5783566.35 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.331, pruned_loss=0.098, over 5690513.09 frames. ], batch size: 164, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:16:01,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7221, 3.5084, 3.3850, 1.7191], device='cuda:0'), covar=tensor([0.0535, 0.0552, 0.0771, 0.2215], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0732, 0.0761, 0.0574], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-02 17:16:01,350 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:06,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:29,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.781e+02 9.879e+02 1.215e+03 1.542e+03 4.401e+03, threshold=2.430e+03, percent-clipped=4.0
+2023-03-02 17:16:30,620 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:33,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:34,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:35,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:37,733 INFO [train.py:968] (0/2) Epoch 5, batch 17900, giga_loss[loss=0.2315, simple_loss=0.3021, pruned_loss=0.08043, over 28883.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3285, pruned_loss=0.09631, over 5700780.25 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3555, pruned_loss=0.1076, over 5784921.71 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3261, pruned_loss=0.09554, over 5683900.32 frames. ], batch size: 112, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:16:58,349 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:17:09,108 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:17:11,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:17:17,328 INFO [train.py:968] (0/2) Epoch 5, batch 17950, libri_loss[loss=0.2854, simple_loss=0.3639, pruned_loss=0.1035, over 29529.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3251, pruned_loss=0.09429, over 5710877.72 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3558, pruned_loss=0.1076, over 5783645.27 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3221, pruned_loss=0.09331, over 5696082.61 frames. ], batch size: 81, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:17:18,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:17:34,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:17:45,867 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199887.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:17:50,925 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.997e+02 9.971e+02 1.399e+03 1.941e+03 1.258e+04, threshold=2.797e+03, percent-clipped=16.0
+2023-03-02 17:17:59,099 INFO [train.py:968] (0/2) Epoch 5, batch 18000, giga_loss[loss=0.2534, simple_loss=0.3245, pruned_loss=0.09117, over 28270.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3233, pruned_loss=0.09359, over 5695528.16 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3565, pruned_loss=0.108, over 5775855.90 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3193, pruned_loss=0.09201, over 5688370.43 frames. ], batch size: 368, lr: 6.42e-03, grad_scale: 4.0
+2023-03-02 17:17:59,103 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 17:18:06,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2908, 1.8742, 1.6879, 1.4930], device='cuda:0'), covar=tensor([0.1548, 0.2004, 0.1276, 0.1344], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0733, 0.0779, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 17:18:07,928 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.234, simple_loss=0.3355, pruned_loss=0.06627, over 944034.00 frames. 
+2023-03-02 17:18:07,929 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 17:18:33,240 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199928.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:18:46,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:18:52,665 INFO [train.py:968] (0/2) Epoch 5, batch 18050, giga_loss[loss=0.1976, simple_loss=0.2736, pruned_loss=0.06084, over 28573.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.321, pruned_loss=0.09275, over 5696683.06 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3571, pruned_loss=0.1083, over 5777644.41 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3167, pruned_loss=0.091, over 5688256.18 frames. ], batch size: 60, lr: 6.42e-03, grad_scale: 4.0
+2023-03-02 17:19:21,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2816, 2.4098, 1.2565, 1.2690], device='cuda:0'), covar=tensor([0.0876, 0.0374, 0.0858, 0.1367], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0471, 0.0311, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 17:19:27,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4381, 1.5065, 1.4876, 1.3578], device='cuda:0'), covar=tensor([0.1054, 0.1543, 0.1373, 0.1401], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0741, 0.0630, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 17:19:30,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.851e+02 1.049e+03 1.357e+03 1.845e+03 4.218e+03, threshold=2.715e+03, percent-clipped=6.0
+2023-03-02 17:19:36,529 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-200000.pt
+2023-03-02 17:19:36,831 INFO [train.py:968] (0/2) Epoch 5, batch 18100, giga_loss[loss=0.2206, simple_loss=0.3023, pruned_loss=0.0695, over 29034.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.317, pruned_loss=0.09068, over 5697629.85 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3575, pruned_loss=0.1083, over 5779703.42 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3123, pruned_loss=0.08886, over 5687093.87 frames. ], batch size: 128, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:19:53,601 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.67 vs. limit=5.0
+2023-03-02 17:20:23,396 INFO [train.py:968] (0/2) Epoch 5, batch 18150, giga_loss[loss=0.2097, simple_loss=0.2801, pruned_loss=0.06968, over 28677.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3143, pruned_loss=0.08942, over 5684926.70 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3582, pruned_loss=0.1089, over 5772957.91 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.309, pruned_loss=0.087, over 5680422.43 frames. ], batch size: 92, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:20:57,324 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:21:00,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:21:02,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 4.975e+02 9.583e+02 1.238e+03 1.649e+03 6.918e+03, threshold=2.475e+03, percent-clipped=8.0
+2023-03-02 17:21:11,638 INFO [train.py:968] (0/2) Epoch 5, batch 18200, giga_loss[loss=0.1984, simple_loss=0.2714, pruned_loss=0.0627, over 28903.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.312, pruned_loss=0.08828, over 5687063.63 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3585, pruned_loss=0.109, over 5777199.28 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.3061, pruned_loss=0.08565, over 5677051.35 frames. ], batch size: 174, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:21:28,417 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:21:39,266 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2488, 1.6609, 1.3455, 1.4798], device='cuda:0'), covar=tensor([0.0760, 0.0376, 0.0335, 0.0783], device='cuda:0'), in_proj_covar=tensor([0.0197, 0.0126, 0.0131, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:0')
+2023-03-02 17:21:59,861 INFO [train.py:968] (0/2) Epoch 5, batch 18250, giga_loss[loss=0.2495, simple_loss=0.3134, pruned_loss=0.09279, over 28719.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3179, pruned_loss=0.09249, over 5676684.22 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3585, pruned_loss=0.1089, over 5778852.53 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3127, pruned_loss=0.09023, over 5666160.17 frames. ], batch size: 242, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:22:15,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:22:24,699 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 17:22:39,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.796e+02 1.163e+03 1.543e+03 1.856e+03 4.767e+03, threshold=3.085e+03, percent-clipped=13.0
+2023-03-02 17:22:45,542 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-02 17:22:46,489 INFO [train.py:968] (0/2) Epoch 5, batch 18300, giga_loss[loss=0.3536, simple_loss=0.4131, pruned_loss=0.1471, over 28904.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3333, pruned_loss=0.1011, over 5683504.83 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3591, pruned_loss=0.1094, over 5781751.57 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3278, pruned_loss=0.09863, over 5670530.45 frames. ], batch size: 112, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:23:01,908 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-02 17:23:09,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:23:28,143 INFO [train.py:968] (0/2) Epoch 5, batch 18350, giga_loss[loss=0.3199, simple_loss=0.3833, pruned_loss=0.1283, over 28920.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3464, pruned_loss=0.1081, over 5700290.75 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3595, pruned_loss=0.1096, over 5785976.21 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.341, pruned_loss=0.1058, over 5682966.87 frames. ], batch size: 213, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:23:39,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200262.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:24:05,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.444e+02 1.409e+03 1.830e+03 2.734e+03 8.512e+03, threshold=3.660e+03, percent-clipped=17.0
+2023-03-02 17:24:10,771 INFO [train.py:968] (0/2) Epoch 5, batch 18400, giga_loss[loss=0.2896, simple_loss=0.36, pruned_loss=0.1096, over 28927.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3553, pruned_loss=0.1125, over 5687933.45 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3598, pruned_loss=0.1097, over 5773144.46 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3506, pruned_loss=0.1106, over 5683338.54 frames. ], batch size: 106, lr: 6.41e-03, grad_scale: 8.0
+2023-03-02 17:24:13,160 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:24:13,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200303.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:24:21,457 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:24:23,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:24:47,852 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:24:53,983 INFO [train.py:968] (0/2) Epoch 5, batch 18450, giga_loss[loss=0.2826, simple_loss=0.3618, pruned_loss=0.1017, over 28886.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3602, pruned_loss=0.1139, over 5684903.42 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3602, pruned_loss=0.1098, over 5772247.57 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.356, pruned_loss=0.1124, over 5679319.27 frames. ], batch size: 186, lr: 6.41e-03, grad_scale: 8.0
+2023-03-02 17:25:10,794 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:25:13,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:25:36,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.652e+02 9.430e+02 1.197e+03 1.644e+03 2.955e+03, threshold=2.395e+03, percent-clipped=0.0
+2023-03-02 17:25:42,542 INFO [train.py:968] (0/2) Epoch 5, batch 18500, giga_loss[loss=0.3048, simple_loss=0.3833, pruned_loss=0.1132, over 28727.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3614, pruned_loss=0.113, over 5680555.33 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3602, pruned_loss=0.1098, over 5772247.57 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3581, pruned_loss=0.1118, over 5676209.12 frames. ], batch size: 262, lr: 6.41e-03, grad_scale: 8.0
+2023-03-02 17:25:44,123 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:25:48,535 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200405.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:25:51,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200408.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:26:15,980 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200437.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:26:24,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200446.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:26:26,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200449.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:26:27,000 INFO [train.py:968] (0/2) Epoch 5, batch 18550, libri_loss[loss=0.3134, simple_loss=0.3741, pruned_loss=0.1264, over 29585.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3636, pruned_loss=0.1145, over 5679555.28 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3606, pruned_loss=0.11, over 5777720.01 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3607, pruned_loss=0.1135, over 5666371.11 frames. ], batch size: 74, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:26:51,106 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200478.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:27:04,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.839e+02 1.139e+03 1.473e+03 1.999e+03 6.069e+03, threshold=2.946e+03, percent-clipped=15.0
+2023-03-02 17:27:11,682 INFO [train.py:968] (0/2) Epoch 5, batch 18600, giga_loss[loss=0.2809, simple_loss=0.3568, pruned_loss=0.1025, over 28907.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3663, pruned_loss=0.1168, over 5685518.56 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3605, pruned_loss=0.1098, over 5779385.39 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3642, pruned_loss=0.1162, over 5672361.45 frames. ], batch size: 145, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:27:27,076 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 17:27:33,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5892, 2.2444, 1.7256, 0.8952], device='cuda:0'), covar=tensor([0.2516, 0.1325, 0.1933, 0.2536], device='cuda:0'), in_proj_covar=tensor([0.1355, 0.1278, 0.1347, 0.1114], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 17:27:34,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3012, 1.4744, 1.1901, 1.4242], device='cuda:0'), covar=tensor([0.0809, 0.0356, 0.0338, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0197, 0.0125, 0.0129, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 17:27:57,560 INFO [train.py:968] (0/2) Epoch 5, batch 18650, giga_loss[loss=0.3228, simple_loss=0.3893, pruned_loss=0.1281, over 28573.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.37, pruned_loss=0.1194, over 5682180.94 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3608, pruned_loss=0.1099, over 5779535.48 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3681, pruned_loss=0.119, over 5669858.41 frames. ], batch size: 336, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:27:59,303 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 17:28:14,927 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200570.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:28:31,945 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 17:28:33,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.393e+02 1.109e+03 1.443e+03 2.131e+03 4.678e+03, threshold=2.885e+03, percent-clipped=5.0
+2023-03-02 17:28:40,486 INFO [train.py:968] (0/2) Epoch 5, batch 18700, giga_loss[loss=0.3903, simple_loss=0.4393, pruned_loss=0.1707, over 28966.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3738, pruned_loss=0.1214, over 5684957.81 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3612, pruned_loss=0.11, over 5778365.08 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3722, pruned_loss=0.1212, over 5674186.41 frames. ], batch size: 227, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:29:20,762 INFO [train.py:968] (0/2) Epoch 5, batch 18750, libri_loss[loss=0.2755, simple_loss=0.3382, pruned_loss=0.1064, over 29358.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3754, pruned_loss=0.121, over 5671308.79 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3618, pruned_loss=0.1102, over 5764378.25 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.374, pruned_loss=0.1211, over 5671586.24 frames. ], batch size: 67, lr: 6.40e-03, grad_scale: 2.0
+2023-03-02 17:29:37,984 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:29:42,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:29:55,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.703e+02 1.067e+03 1.538e+03 2.110e+03 5.062e+03, threshold=3.077e+03, percent-clipped=8.0
+2023-03-02 17:30:00,517 INFO [train.py:968] (0/2) Epoch 5, batch 18800, giga_loss[loss=0.3021, simple_loss=0.3753, pruned_loss=0.1144, over 28758.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3781, pruned_loss=0.1221, over 5686548.80 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3624, pruned_loss=0.1104, over 5768764.53 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3769, pruned_loss=0.1223, over 5680528.33 frames. ], batch size: 99, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:30:14,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7129, 1.0389, 3.6884, 2.9453], device='cuda:0'), covar=tensor([0.1825, 0.2404, 0.0350, 0.0789], device='cuda:0'), in_proj_covar=tensor([0.0562, 0.0521, 0.0721, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 17:30:35,718 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 17:30:43,671 INFO [train.py:968] (0/2) Epoch 5, batch 18850, giga_loss[loss=0.3508, simple_loss=0.3928, pruned_loss=0.1543, over 26698.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3776, pruned_loss=0.1204, over 5692411.44 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3624, pruned_loss=0.1103, over 5771689.87 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3768, pruned_loss=0.1209, over 5683875.40 frames. ], batch size: 555, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:31:10,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7385, 2.2728, 1.4997, 1.0874], device='cuda:0'), covar=tensor([0.1262, 0.0711, 0.0853, 0.1156], device='cuda:0'), in_proj_covar=tensor([0.1381, 0.1129, 0.1154, 0.1222], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 17:31:18,840 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.516e+02 1.123e+03 1.343e+03 1.788e+03 9.709e+03, threshold=2.685e+03, percent-clipped=6.0
+2023-03-02 17:31:23,129 INFO [train.py:968] (0/2) Epoch 5, batch 18900, giga_loss[loss=0.2896, simple_loss=0.3639, pruned_loss=0.1076, over 29019.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3758, pruned_loss=0.1173, over 5709357.31 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3632, pruned_loss=0.1106, over 5774280.34 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3749, pruned_loss=0.1176, over 5698154.60 frames. ], batch size: 128, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:31:40,715 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:31:42,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:31:43,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1201, 2.5163, 1.1904, 1.1377], device='cuda:0'), covar=tensor([0.0954, 0.0301, 0.0835, 0.1436], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0467, 0.0305, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 17:31:54,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-02 17:31:59,441 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-02 17:32:05,090 INFO [train.py:968] (0/2) Epoch 5, batch 18950, giga_loss[loss=0.3219, simple_loss=0.3812, pruned_loss=0.1313, over 28901.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3743, pruned_loss=0.1162, over 5710692.03 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3634, pruned_loss=0.1106, over 5776262.70 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3737, pruned_loss=0.1166, over 5698386.37 frames. ], batch size: 199, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:32:07,502 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:32:42,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.763e+02 9.620e+02 1.295e+03 1.846e+03 5.865e+03, threshold=2.590e+03, percent-clipped=8.0
+2023-03-02 17:32:47,023 INFO [train.py:968] (0/2) Epoch 5, batch 19000, giga_loss[loss=0.327, simple_loss=0.3794, pruned_loss=0.1373, over 28859.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3745, pruned_loss=0.117, over 5714269.50 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3637, pruned_loss=0.1106, over 5776447.28 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.374, pruned_loss=0.1174, over 5702589.71 frames. ], batch size: 199, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:32:51,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6017, 2.0262, 1.8832, 1.6625], device='cuda:0'), covar=tensor([0.1494, 0.1714, 0.1115, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0740, 0.0784, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-02 17:33:28,065 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200945.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:33:31,366 INFO [train.py:968] (0/2) Epoch 5, batch 19050, giga_loss[loss=0.3089, simple_loss=0.3735, pruned_loss=0.1221, over 28963.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3772, pruned_loss=0.1217, over 5709956.03 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3642, pruned_loss=0.1109, over 5768923.52 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3766, pruned_loss=0.122, over 5706079.67 frames. ], batch size: 164, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:34:10,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.283e+02 1.314e+03 1.682e+03 2.211e+03 1.037e+04, threshold=3.363e+03, percent-clipped=15.0
+2023-03-02 17:34:14,715 INFO [train.py:968] (0/2) Epoch 5, batch 19100, giga_loss[loss=0.3032, simple_loss=0.3605, pruned_loss=0.123, over 28559.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3785, pruned_loss=0.1248, over 5709767.23 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3643, pruned_loss=0.1109, over 5772848.84 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3784, pruned_loss=0.1255, over 5701210.22 frames. ], batch size: 71, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:34:42,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7197, 2.2956, 1.7826, 1.9909], device='cuda:0'), covar=tensor([0.0536, 0.0556, 0.0795, 0.0840], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0444, 0.0503, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 17:34:46,807 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:34:54,968 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:34:57,053 INFO [train.py:968] (0/2) Epoch 5, batch 19150, libri_loss[loss=0.3465, simple_loss=0.4172, pruned_loss=0.1379, over 25924.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3763, pruned_loss=0.1243, over 5703698.45 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3643, pruned_loss=0.1108, over 5771172.44 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3763, pruned_loss=0.1251, over 5697715.29 frames. ], batch size: 136, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:35:06,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0098, 3.8146, 3.6416, 1.8906], device='cuda:0'), covar=tensor([0.0488, 0.0493, 0.0691, 0.2086], device='cuda:0'), in_proj_covar=tensor([0.0821, 0.0740, 0.0778, 0.0588], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 17:35:13,623 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 17:35:31,246 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201088.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:35:35,480 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201091.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:35:37,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.588e+02 1.213e+03 1.521e+03 2.072e+03 9.052e+03, threshold=3.041e+03, percent-clipped=11.0
+2023-03-02 17:35:38,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4163, 1.5619, 1.1438, 0.9212], device='cuda:0'), covar=tensor([0.1222, 0.0876, 0.0883, 0.1136], device='cuda:0'), in_proj_covar=tensor([0.1419, 0.1166, 0.1200, 0.1274], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 17:35:41,583 INFO [train.py:968] (0/2) Epoch 5, batch 19200, giga_loss[loss=0.3297, simple_loss=0.3883, pruned_loss=0.1356, over 28848.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3741, pruned_loss=0.1237, over 5702424.37 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3646, pruned_loss=0.111, over 5773818.84 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3742, pruned_loss=0.1244, over 5693933.76 frames. ], batch size: 284, lr: 6.40e-03, grad_scale: 8.0
+2023-03-02 17:35:57,865 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201120.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:36:26,232 INFO [train.py:968] (0/2) Epoch 5, batch 19250, giga_loss[loss=0.2807, simple_loss=0.3521, pruned_loss=0.1047, over 28502.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3729, pruned_loss=0.1219, over 5704197.48 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3649, pruned_loss=0.1111, over 5763997.64 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3728, pruned_loss=0.1227, over 5704795.22 frames. ], batch size: 65, lr: 6.40e-03, grad_scale: 8.0
+2023-03-02 17:36:27,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7115, 1.6856, 1.5820, 1.6469], device='cuda:0'), covar=tensor([0.1041, 0.1606, 0.1489, 0.1363], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0749, 0.0635, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 17:36:58,788 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:37:00,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:37:01,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.872e+02 1.118e+03 1.361e+03 1.664e+03 3.873e+03, threshold=2.722e+03, percent-clipped=5.0
+2023-03-02 17:37:04,183 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-03-02 17:37:06,182 INFO [train.py:968] (0/2) Epoch 5, batch 19300, giga_loss[loss=0.2995, simple_loss=0.3654, pruned_loss=0.1168, over 28605.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3711, pruned_loss=0.1198, over 5703020.34 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3652, pruned_loss=0.1112, over 5757670.37 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.371, pruned_loss=0.1206, over 5707171.69 frames. ], batch size: 307, lr: 6.40e-03, grad_scale: 8.0
+2023-03-02 17:37:28,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:37:54,065 INFO [train.py:968] (0/2) Epoch 5, batch 19350, giga_loss[loss=0.2591, simple_loss=0.3365, pruned_loss=0.09087, over 28834.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3675, pruned_loss=0.1173, over 5691411.99 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3649, pruned_loss=0.1111, over 5761571.15 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3678, pruned_loss=0.1183, over 5690137.96 frames. ], batch size: 243, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:38:23,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9679, 1.7509, 1.3957, 1.6165], device='cuda:0'), covar=tensor([0.0623, 0.0684, 0.0905, 0.0900], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0447, 0.0506, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 17:38:27,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9789, 1.7890, 1.4210, 1.6214], device='cuda:0'), covar=tensor([0.0598, 0.0559, 0.0949, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0447, 0.0506, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 17:38:31,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.458e+02 9.499e+02 1.237e+03 1.633e+03 5.475e+03, threshold=2.475e+03, percent-clipped=7.0
+2023-03-02 17:38:37,134 INFO [train.py:968] (0/2) Epoch 5, batch 19400, giga_loss[loss=0.3, simple_loss=0.3522, pruned_loss=0.1239, over 27564.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3612, pruned_loss=0.1135, over 5687821.73 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3645, pruned_loss=0.1109, over 5755698.98 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3618, pruned_loss=0.1146, over 5689169.26 frames. ], batch size: 472, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:39:25,013 INFO [train.py:968] (0/2) Epoch 5, batch 19450, giga_loss[loss=0.2899, simple_loss=0.3489, pruned_loss=0.1155, over 28869.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3549, pruned_loss=0.1101, over 5682115.27 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3643, pruned_loss=0.1107, over 5758804.30 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3555, pruned_loss=0.1113, over 5678764.19 frames. ], batch size: 186, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:40:07,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.975e+02 9.707e+02 1.275e+03 1.747e+03 6.138e+03, threshold=2.550e+03, percent-clipped=12.0
+2023-03-02 17:40:13,925 INFO [train.py:968] (0/2) Epoch 5, batch 19500, giga_loss[loss=0.2783, simple_loss=0.3479, pruned_loss=0.1044, over 28981.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3498, pruned_loss=0.1077, over 5662769.60 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3646, pruned_loss=0.1108, over 5761303.27 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3497, pruned_loss=0.1084, over 5655371.49 frames. ], batch size: 227, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:40:26,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:41:01,741 INFO [train.py:968] (0/2) Epoch 5, batch 19550, giga_loss[loss=0.2942, simple_loss=0.3602, pruned_loss=0.1141, over 28966.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3504, pruned_loss=0.1076, over 5661265.01 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3652, pruned_loss=0.1111, over 5759109.89 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3495, pruned_loss=0.1078, over 5655934.59 frames. ], batch size: 136, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:41:13,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:41:41,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.396e+02 9.710e+02 1.297e+03 1.774e+03 4.595e+03, threshold=2.594e+03, percent-clipped=8.0
+2023-03-02 17:41:46,057 INFO [train.py:968] (0/2) Epoch 5, batch 19600, giga_loss[loss=0.2604, simple_loss=0.3295, pruned_loss=0.09563, over 28737.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3519, pruned_loss=0.1088, over 5659395.12 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.366, pruned_loss=0.1114, over 5758444.26 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3501, pruned_loss=0.1086, over 5653979.75 frames. ], batch size: 92, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:42:30,491 INFO [train.py:968] (0/2) Epoch 5, batch 19650, giga_loss[loss=0.2723, simple_loss=0.3416, pruned_loss=0.1015, over 28987.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3497, pruned_loss=0.1075, over 5674875.82 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3658, pruned_loss=0.1113, over 5760058.01 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3483, pruned_loss=0.1075, over 5668300.64 frames. ], batch size: 128, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:42:36,252 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:42:40,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:43:01,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:43:05,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.166e+02 1.019e+03 1.250e+03 1.638e+03 5.072e+03, threshold=2.501e+03, percent-clipped=8.0
+2023-03-02 17:43:10,913 INFO [train.py:968] (0/2) Epoch 5, batch 19700, giga_loss[loss=0.2541, simple_loss=0.3276, pruned_loss=0.09031, over 28848.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3487, pruned_loss=0.1069, over 5689000.10 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3666, pruned_loss=0.1115, over 5764170.67 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3463, pruned_loss=0.1065, over 5676864.61 frames. ], batch size: 174, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:43:48,300 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:43:52,244 INFO [train.py:968] (0/2) Epoch 5, batch 19750, giga_loss[loss=0.2679, simple_loss=0.3319, pruned_loss=0.1019, over 29007.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3458, pruned_loss=0.1051, over 5695493.29 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3668, pruned_loss=0.1115, over 5767060.88 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3433, pruned_loss=0.1046, over 5681540.27 frames. ], batch size: 155, lr: 6.39e-03, grad_scale: 4.0
+2023-03-02 17:44:05,599 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:44:28,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.178e+02 9.211e+02 1.187e+03 1.734e+03 1.497e+04, threshold=2.375e+03, percent-clipped=13.0
+2023-03-02 17:44:33,371 INFO [train.py:968] (0/2) Epoch 5, batch 19800, giga_loss[loss=0.2575, simple_loss=0.3317, pruned_loss=0.09169, over 28871.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3435, pruned_loss=0.1035, over 5703125.67 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3672, pruned_loss=0.1117, over 5767851.32 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3405, pruned_loss=0.1027, over 5689367.72 frames. ], batch size: 199, lr: 6.39e-03, grad_scale: 4.0
+2023-03-02 17:44:47,022 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-02 17:44:55,874 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 17:45:11,732 INFO [train.py:968] (0/2) Epoch 5, batch 19850, giga_loss[loss=0.209, simple_loss=0.28, pruned_loss=0.06901, over 28491.00 frames. ], tot_loss[loss=0.275, simple_loss=0.343, pruned_loss=0.1034, over 5711919.80 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3689, pruned_loss=0.1125, over 5773622.75 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3381, pruned_loss=0.1018, over 5693050.33 frames. ], batch size: 71, lr: 6.39e-03, grad_scale: 4.0
+2023-03-02 17:45:46,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.827e+02 1.039e+03 1.302e+03 1.727e+03 9.386e+03, threshold=2.604e+03, percent-clipped=11.0
+2023-03-02 17:45:51,029 INFO [train.py:968] (0/2) Epoch 5, batch 19900, giga_loss[loss=0.2534, simple_loss=0.3122, pruned_loss=0.09732, over 28715.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3395, pruned_loss=0.1017, over 5720869.27 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3686, pruned_loss=0.1123, over 5777095.84 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3353, pruned_loss=0.1003, over 5701562.72 frames. ], batch size: 85, lr: 6.39e-03, grad_scale: 4.0
+2023-03-02 17:46:22,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:46:31,722 INFO [train.py:968] (0/2) Epoch 5, batch 19950, giga_loss[loss=0.2526, simple_loss=0.3263, pruned_loss=0.08947, over 28677.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3386, pruned_loss=0.101, over 5725420.73 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3698, pruned_loss=0.1129, over 5774849.13 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3334, pruned_loss=0.09907, over 5710739.06 frames. ], batch size: 242, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:47:11,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.503e+02 9.966e+02 1.370e+03 2.104e+03 8.565e+03, threshold=2.740e+03, percent-clipped=20.0
+2023-03-02 17:47:14,312 INFO [train.py:968] (0/2) Epoch 5, batch 20000, giga_loss[loss=0.2623, simple_loss=0.3291, pruned_loss=0.09772, over 28869.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3382, pruned_loss=0.1013, over 5714963.13 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3707, pruned_loss=0.1134, over 5769761.53 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3322, pruned_loss=0.09887, over 5706104.24 frames. ], batch size: 112, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:47:52,883 INFO [train.py:968] (0/2) Epoch 5, batch 20050, giga_loss[loss=0.2716, simple_loss=0.3397, pruned_loss=0.1018, over 29056.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.337, pruned_loss=0.1001, over 5719707.00 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3718, pruned_loss=0.1139, over 5771975.56 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3304, pruned_loss=0.09757, over 5709621.18 frames. ], batch size: 155, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:48:16,441 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:48:18,323 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:48:25,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.271e+02 1.000e+03 1.377e+03 1.729e+03 8.137e+03, threshold=2.754e+03, percent-clipped=7.0
+2023-03-02 17:48:29,238 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-202000.pt
+2023-03-02 17:48:29,557 INFO [train.py:968] (0/2) Epoch 5, batch 20100, giga_loss[loss=0.2419, simple_loss=0.3113, pruned_loss=0.08626, over 28531.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3376, pruned_loss=0.09996, over 5725861.26 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3725, pruned_loss=0.1138, over 5775345.31 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3306, pruned_loss=0.09747, over 5713156.47 frames. ], batch size: 71, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:48:31,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0593, 2.3972, 2.2482, 2.0881], device='cuda:0'), covar=tensor([0.1505, 0.1613, 0.1122, 0.1107], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0739, 0.0784, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-02 17:48:42,292 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:48:48,184 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:49:02,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:49:07,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:49:14,096 INFO [train.py:968] (0/2) Epoch 5, batch 20150, giga_loss[loss=0.2863, simple_loss=0.3579, pruned_loss=0.1074, over 28997.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3432, pruned_loss=0.1038, over 5722298.31 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3727, pruned_loss=0.1137, over 5777848.85 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3367, pruned_loss=0.1016, over 5708367.15 frames. ], batch size: 164, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:49:17,447 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:49:30,620 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-02 17:49:59,375 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 17:50:00,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.145e+02 1.085e+03 1.425e+03 1.769e+03 4.189e+03, threshold=2.850e+03, percent-clipped=4.0
+2023-03-02 17:50:02,830 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:50:03,236 INFO [train.py:968] (0/2) Epoch 5, batch 20200, giga_loss[loss=0.3752, simple_loss=0.4143, pruned_loss=0.168, over 26666.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3496, pruned_loss=0.1085, over 5710334.09 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3729, pruned_loss=0.1138, over 5779933.16 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3437, pruned_loss=0.1065, over 5696302.91 frames. ], batch size: 555, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:50:11,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4202, 1.6002, 1.3055, 1.6455], device='cuda:0'), covar=tensor([0.2077, 0.1963, 0.2023, 0.2008], device='cuda:0'), in_proj_covar=tensor([0.1115, 0.0859, 0.0985, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 17:50:52,406 INFO [train.py:968] (0/2) Epoch 5, batch 20250, giga_loss[loss=0.4597, simple_loss=0.4709, pruned_loss=0.2243, over 26758.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3583, pruned_loss=0.1147, over 5707567.72 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3733, pruned_loss=0.1141, over 5780187.27 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3527, pruned_loss=0.1128, over 5694060.45 frames. ], batch size: 555, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:51:05,858 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:51:07,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:51:21,883 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:51:25,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:51:32,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.573e+02 1.161e+03 1.550e+03 2.065e+03 5.020e+03, threshold=3.101e+03, percent-clipped=11.0
+2023-03-02 17:51:32,871 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:51:36,584 INFO [train.py:968] (0/2) Epoch 5, batch 20300, giga_loss[loss=0.3024, simple_loss=0.3708, pruned_loss=0.117, over 28870.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3635, pruned_loss=0.1174, over 5701249.96 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3735, pruned_loss=0.1142, over 5778987.71 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3586, pruned_loss=0.1159, over 5690032.34 frames. ], batch size: 99, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:51:55,241 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:52:09,699 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-02 17:52:25,464 INFO [train.py:968] (0/2) Epoch 5, batch 20350, giga_loss[loss=0.2886, simple_loss=0.3644, pruned_loss=0.1064, over 28658.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3697, pruned_loss=0.1206, over 5687930.47 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3734, pruned_loss=0.1142, over 5770519.25 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3659, pruned_loss=0.1194, over 5685363.19 frames. ], batch size: 60, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:52:54,687 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-02 17:52:59,576 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-02 17:53:13,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.284e+02 1.117e+03 1.351e+03 1.659e+03 5.266e+03, threshold=2.701e+03, percent-clipped=7.0
+2023-03-02 17:53:16,246 INFO [train.py:968] (0/2) Epoch 5, batch 20400, giga_loss[loss=0.3424, simple_loss=0.4032, pruned_loss=0.1408, over 28894.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3755, pruned_loss=0.1237, over 5688014.43 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3734, pruned_loss=0.1142, over 5771195.66 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3725, pruned_loss=0.1228, over 5684916.99 frames. ], batch size: 112, lr: 6.38e-03, grad_scale: 8.0
+2023-03-02 17:53:25,592 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-02 17:53:54,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4426, 2.2652, 1.5474, 0.6766], device='cuda:0'), covar=tensor([0.2872, 0.1435, 0.2589, 0.3011], device='cuda:0'), in_proj_covar=tensor([0.1372, 0.1291, 0.1383, 0.1144], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 17:53:56,627 INFO [train.py:968] (0/2) Epoch 5, batch 20450, giga_loss[loss=0.2876, simple_loss=0.3531, pruned_loss=0.111, over 28815.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3776, pruned_loss=0.1252, over 5678727.15 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3739, pruned_loss=0.1146, over 5753885.59 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3748, pruned_loss=0.1243, over 5689515.38 frames. ], batch size: 112, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:54:28,526 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:54:39,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.230e+02 1.280e+03 1.772e+03 2.608e+03 7.184e+03, threshold=3.545e+03, percent-clipped=21.0
+2023-03-02 17:54:41,836 INFO [train.py:968] (0/2) Epoch 5, batch 20500, giga_loss[loss=0.3242, simple_loss=0.3863, pruned_loss=0.131, over 28919.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3706, pruned_loss=0.1199, over 5683844.02 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3741, pruned_loss=0.1148, over 5756135.89 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3681, pruned_loss=0.1193, over 5688461.34 frames. ], batch size: 199, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:54:52,531 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:54:54,743 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-02 17:55:06,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:55:24,638 INFO [train.py:968] (0/2) Epoch 5, batch 20550, giga_loss[loss=0.3282, simple_loss=0.3899, pruned_loss=0.1333, over 28612.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3699, pruned_loss=0.1189, over 5684965.09 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3745, pruned_loss=0.1152, over 5755517.31 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3675, pruned_loss=0.1181, over 5687700.56 frames. ], batch size: 307, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:55:45,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:56:04,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.536e+02 1.248e+03 1.658e+03 2.577e+03 1.127e+04, threshold=3.316e+03, percent-clipped=12.0
+2023-03-02 17:56:06,811 INFO [train.py:968] (0/2) Epoch 5, batch 20600, giga_loss[loss=0.3157, simple_loss=0.3883, pruned_loss=0.1215, over 29019.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.369, pruned_loss=0.1175, over 5680713.66 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3746, pruned_loss=0.1153, over 5748251.54 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3669, pruned_loss=0.1168, over 5687025.06 frames. ], batch size: 136, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 17:56:29,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2500, 3.1229, 3.0005, 1.2709], device='cuda:0'), covar=tensor([0.0762, 0.0684, 0.0928, 0.2396], device='cuda:0'), in_proj_covar=tensor([0.0838, 0.0756, 0.0786, 0.0597], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 17:56:50,765 INFO [train.py:968] (0/2) Epoch 5, batch 20650, giga_loss[loss=0.3139, simple_loss=0.3709, pruned_loss=0.1284, over 28942.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3714, pruned_loss=0.1188, over 5677053.65 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3746, pruned_loss=0.1153, over 5741484.96 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3697, pruned_loss=0.1184, over 5686387.47 frames. ], batch size: 106, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 17:56:56,329 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:56:58,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:03,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2076, 2.4348, 1.1909, 1.3522], device='cuda:0'), covar=tensor([0.0754, 0.0287, 0.0738, 0.1137], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0468, 0.0303, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 17:57:10,260 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:13,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:22,659 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:29,978 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.458e+02 1.094e+03 1.379e+03 1.647e+03 5.588e+03, threshold=2.758e+03, percent-clipped=4.0
+2023-03-02 17:57:33,435 INFO [train.py:968] (0/2) Epoch 5, batch 20700, libri_loss[loss=0.2783, simple_loss=0.3523, pruned_loss=0.1022, over 29568.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3744, pruned_loss=0.1216, over 5684440.44 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3746, pruned_loss=0.1158, over 5745900.08 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.373, pruned_loss=0.1209, over 5685577.77 frames. ], batch size: 77, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 17:57:38,551 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:50,057 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:54,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:58:18,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:58:19,094 INFO [train.py:968] (0/2) Epoch 5, batch 20750, giga_loss[loss=0.3309, simple_loss=0.3887, pruned_loss=0.1366, over 27904.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3756, pruned_loss=0.1227, over 5693624.98 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3751, pruned_loss=0.1161, over 5746880.43 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3741, pruned_loss=0.122, over 5692768.38 frames. ], batch size: 412, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 17:58:44,854 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-02 17:59:03,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.538e+02 1.150e+03 1.464e+03 1.893e+03 5.394e+03, threshold=2.927e+03, percent-clipped=8.0
+2023-03-02 17:59:04,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4105, 1.3653, 1.5053, 1.3771], device='cuda:0'), covar=tensor([0.1059, 0.1378, 0.1442, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0736, 0.0630, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 17:59:06,696 INFO [train.py:968] (0/2) Epoch 5, batch 20800, giga_loss[loss=0.3618, simple_loss=0.4164, pruned_loss=0.1535, over 28762.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3761, pruned_loss=0.1227, over 5700373.60 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3754, pruned_loss=0.1161, over 5741725.76 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3746, pruned_loss=0.1223, over 5702481.40 frames. ], batch size: 99, lr: 6.37e-03, grad_scale: 8.0
+2023-03-02 17:59:09,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3292, 1.9710, 1.3493, 0.5760], device='cuda:0'), covar=tensor([0.2613, 0.1180, 0.1920, 0.2772], device='cuda:0'), in_proj_covar=tensor([0.1369, 0.1292, 0.1375, 0.1141], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 17:59:49,203 INFO [train.py:968] (0/2) Epoch 5, batch 20850, giga_loss[loss=0.3366, simple_loss=0.3878, pruned_loss=0.1427, over 28824.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3777, pruned_loss=0.1245, over 5697162.12 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3758, pruned_loss=0.1165, over 5742557.68 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3762, pruned_loss=0.124, over 5697367.09 frames. ], batch size: 227, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 17:59:56,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:00:28,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.108e+02 1.103e+03 1.476e+03 1.876e+03 4.419e+03, threshold=2.951e+03, percent-clipped=6.0
+2023-03-02 18:00:29,702 INFO [train.py:968] (0/2) Epoch 5, batch 20900, giga_loss[loss=0.3018, simple_loss=0.3664, pruned_loss=0.1186, over 28871.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3766, pruned_loss=0.123, over 5707473.86 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3756, pruned_loss=0.1163, over 5748838.39 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3757, pruned_loss=0.123, over 5700681.06 frames. ], batch size: 86, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:01:08,895 INFO [train.py:968] (0/2) Epoch 5, batch 20950, giga_loss[loss=0.2671, simple_loss=0.3547, pruned_loss=0.08977, over 28901.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3766, pruned_loss=0.1216, over 5714202.91 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.376, pruned_loss=0.1167, over 5753717.97 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3756, pruned_loss=0.1215, over 5703359.17 frames. ], batch size: 145, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:01:50,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.480e+02 1.042e+03 1.320e+03 1.708e+03 6.216e+03, threshold=2.640e+03, percent-clipped=8.0
+2023-03-02 18:01:51,914 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 18:01:52,188 INFO [train.py:968] (0/2) Epoch 5, batch 21000, giga_loss[loss=0.2973, simple_loss=0.372, pruned_loss=0.1113, over 29045.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3771, pruned_loss=0.1211, over 5719138.94 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3756, pruned_loss=0.1165, over 5755750.88 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3767, pruned_loss=0.1213, over 5708175.61 frames. ], batch size: 155, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:01:52,194 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 18:02:00,844 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2422, simple_loss=0.3449, pruned_loss=0.0698, over 944034.00 frames. 
+2023-03-02 18:02:00,844 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 18:02:01,737 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:02:03,830 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:02:04,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5575, 2.4503, 1.8337, 2.1827], device='cuda:0'), covar=tensor([0.0554, 0.0553, 0.0757, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0331, 0.0442, 0.0499, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 18:02:26,458 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:02:40,108 INFO [train.py:968] (0/2) Epoch 5, batch 21050, giga_loss[loss=0.3652, simple_loss=0.4082, pruned_loss=0.1611, over 28947.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3765, pruned_loss=0.1208, over 5726130.88 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3754, pruned_loss=0.1167, over 5759109.22 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3764, pruned_loss=0.1209, over 5713381.39 frames. ], batch size: 112, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:02:51,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4645, 1.4872, 1.4924, 1.4770], device='cuda:0'), covar=tensor([0.1079, 0.1648, 0.1495, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0732, 0.0632, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 18:03:16,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.338e+02 1.231e+03 1.594e+03 2.943e+03 1.023e+04, threshold=3.188e+03, percent-clipped=30.0
+2023-03-02 18:03:18,134 INFO [train.py:968] (0/2) Epoch 5, batch 21100, libri_loss[loss=0.298, simple_loss=0.3712, pruned_loss=0.1124, over 29525.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3754, pruned_loss=0.1212, over 5718212.17 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3763, pruned_loss=0.1176, over 5760413.39 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3745, pruned_loss=0.1206, over 5705585.11 frames. ], batch size: 81, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:03:19,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1037, 1.8255, 1.6436, 1.6447], device='cuda:0'), covar=tensor([0.1274, 0.2024, 0.1648, 0.1719], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0731, 0.0632, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 18:03:37,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6238, 1.5490, 1.5947, 1.5088], device='cuda:0'), covar=tensor([0.1043, 0.1680, 0.1416, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0732, 0.0630, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 18:03:56,907 INFO [train.py:968] (0/2) Epoch 5, batch 21150, giga_loss[loss=0.2819, simple_loss=0.3541, pruned_loss=0.1049, over 28484.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3736, pruned_loss=0.1203, over 5726750.27 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3769, pruned_loss=0.1186, over 5767500.85 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3722, pruned_loss=0.119, over 5708254.17 frames. ], batch size: 78, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:04:37,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.045e+02 9.726e+02 1.252e+03 1.944e+03 6.303e+03, threshold=2.504e+03, percent-clipped=5.0
+2023-03-02 18:04:38,603 INFO [train.py:968] (0/2) Epoch 5, batch 21200, giga_loss[loss=0.288, simple_loss=0.3608, pruned_loss=0.1076, over 28543.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3723, pruned_loss=0.1196, over 5713634.75 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3774, pruned_loss=0.1191, over 5752591.51 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3707, pruned_loss=0.1181, over 5711792.37 frames. ], batch size: 60, lr: 6.37e-03, grad_scale: 8.0
+2023-03-02 18:04:57,046 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 18:05:22,123 INFO [train.py:968] (0/2) Epoch 5, batch 21250, giga_loss[loss=0.2877, simple_loss=0.3547, pruned_loss=0.1104, over 28328.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3736, pruned_loss=0.1206, over 5715676.42 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3776, pruned_loss=0.1194, over 5756278.49 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.372, pruned_loss=0.1192, over 5710000.22 frames. ], batch size: 65, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:05:30,486 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-02 18:05:36,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4517, 1.5048, 1.4850, 1.4946], device='cuda:0'), covar=tensor([0.0932, 0.1268, 0.1318, 0.1094], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0724, 0.0627, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 18:06:02,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.414e+02 1.050e+03 1.400e+03 1.810e+03 5.757e+03, threshold=2.799e+03, percent-clipped=10.0
+2023-03-02 18:06:02,874 INFO [train.py:968] (0/2) Epoch 5, batch 21300, giga_loss[loss=0.3, simple_loss=0.3727, pruned_loss=0.1137, over 28910.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3733, pruned_loss=0.1202, over 5714846.49 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3783, pruned_loss=0.1201, over 5760253.02 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3713, pruned_loss=0.1185, over 5705777.05 frames. ], batch size: 106, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:06:05,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1890, 3.9728, 3.8394, 1.6149], device='cuda:0'), covar=tensor([0.0453, 0.0466, 0.0741, 0.2342], device='cuda:0'), in_proj_covar=tensor([0.0835, 0.0758, 0.0784, 0.0592], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 18:06:20,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=203223.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:06:43,228 INFO [train.py:968] (0/2) Epoch 5, batch 21350, giga_loss[loss=0.3262, simple_loss=0.3881, pruned_loss=0.1321, over 28269.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.373, pruned_loss=0.1192, over 5719869.94 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3792, pruned_loss=0.1211, over 5762238.94 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3704, pruned_loss=0.1169, over 5709823.21 frames. ], batch size: 77, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:07:22,590 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 18:07:23,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.036e+02 9.779e+02 1.246e+03 1.671e+03 7.700e+03, threshold=2.491e+03, percent-clipped=9.0
+2023-03-02 18:07:24,271 INFO [train.py:968] (0/2) Epoch 5, batch 21400, giga_loss[loss=0.3542, simple_loss=0.4039, pruned_loss=0.1523, over 27593.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3719, pruned_loss=0.1183, over 5715330.08 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3792, pruned_loss=0.1213, over 5764895.84 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3696, pruned_loss=0.1161, over 5704145.50 frames. ], batch size: 472, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:08:01,147 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-02 18:08:05,445 INFO [train.py:968] (0/2) Epoch 5, batch 21450, giga_loss[loss=0.3, simple_loss=0.3635, pruned_loss=0.1182, over 28773.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3711, pruned_loss=0.1183, over 5713785.51 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3796, pruned_loss=0.1217, over 5766746.51 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3687, pruned_loss=0.1161, over 5701870.82 frames. ], batch size: 284, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:08:47,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.879e+02 9.291e+02 1.171e+03 1.645e+03 6.574e+03, threshold=2.342e+03, percent-clipped=10.0
+2023-03-02 18:08:48,117 INFO [train.py:968] (0/2) Epoch 5, batch 21500, giga_loss[loss=0.3048, simple_loss=0.3671, pruned_loss=0.1212, over 28854.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3677, pruned_loss=0.1166, over 5707625.25 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3794, pruned_loss=0.1216, over 5767288.77 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.366, pruned_loss=0.1151, over 5697494.78 frames. ], batch size: 99, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:09:14,238 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=203431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:09:29,639 INFO [train.py:968] (0/2) Epoch 5, batch 21550, giga_loss[loss=0.2472, simple_loss=0.3231, pruned_loss=0.08572, over 28775.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3645, pruned_loss=0.1152, over 5705090.12 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3801, pruned_loss=0.1223, over 5770188.65 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3624, pruned_loss=0.1133, over 5693440.02 frames. ], batch size: 119, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:09:55,294 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=203478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:10:10,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.646e+02 1.151e+03 1.429e+03 1.935e+03 5.072e+03, threshold=2.859e+03, percent-clipped=10.0
+2023-03-02 18:10:11,222 INFO [train.py:968] (0/2) Epoch 5, batch 21600, giga_loss[loss=0.3202, simple_loss=0.3792, pruned_loss=0.1306, over 28903.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3648, pruned_loss=0.1161, over 5705371.08 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3804, pruned_loss=0.1227, over 5771533.16 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3626, pruned_loss=0.1141, over 5694269.85 frames. ], batch size: 227, lr: 6.36e-03, grad_scale: 8.0
+2023-03-02 18:10:44,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-02 18:10:56,069 INFO [train.py:968] (0/2) Epoch 5, batch 21650, giga_loss[loss=0.2804, simple_loss=0.3479, pruned_loss=0.1064, over 28725.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.364, pruned_loss=0.1165, over 5701464.73 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3803, pruned_loss=0.1226, over 5772229.49 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3624, pruned_loss=0.1149, over 5691892.30 frames. ], batch size: 92, lr: 6.36e-03, grad_scale: 8.0
+2023-03-02 18:11:07,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3647, 3.1309, 3.1202, 1.2605], device='cuda:0'), covar=tensor([0.0671, 0.0778, 0.0960, 0.2018], device='cuda:0'), in_proj_covar=tensor([0.0825, 0.0752, 0.0775, 0.0590], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 18:11:36,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203598.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:11:37,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.837e+02 1.186e+03 1.499e+03 1.944e+03 5.299e+03, threshold=2.999e+03, percent-clipped=14.0
+2023-03-02 18:11:37,748 INFO [train.py:968] (0/2) Epoch 5, batch 21700, giga_loss[loss=0.2389, simple_loss=0.3121, pruned_loss=0.08286, over 28338.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3617, pruned_loss=0.1156, over 5708150.00 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3807, pruned_loss=0.1233, over 5775304.80 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3595, pruned_loss=0.1136, over 5695991.18 frames. ], batch size: 71, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:12:09,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7180, 1.6336, 1.2604, 1.4245], device='cuda:0'), covar=tensor([0.0617, 0.0580, 0.0964, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0447, 0.0506, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 18:12:16,181 INFO [train.py:968] (0/2) Epoch 5, batch 21750, giga_loss[loss=0.2711, simple_loss=0.3361, pruned_loss=0.1031, over 28982.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3605, pruned_loss=0.1154, over 5712269.39 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3817, pruned_loss=0.1242, over 5777309.08 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3573, pruned_loss=0.1127, over 5699034.80 frames. ], batch size: 106, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:12:47,005 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2966, 1.5271, 1.3071, 1.4371], device='cuda:0'), covar=tensor([0.2049, 0.2031, 0.2074, 0.2061], device='cuda:0'), in_proj_covar=tensor([0.1099, 0.0854, 0.0971, 0.0936], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 18:12:58,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.705e+02 9.351e+02 1.178e+03 1.724e+03 6.735e+03, threshold=2.355e+03, percent-clipped=8.0
+2023-03-02 18:12:58,305 INFO [train.py:968] (0/2) Epoch 5, batch 21800, giga_loss[loss=0.2747, simple_loss=0.3446, pruned_loss=0.1024, over 28629.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3569, pruned_loss=0.1129, over 5717827.66 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.382, pruned_loss=0.1244, over 5775869.06 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3536, pruned_loss=0.1103, over 5707400.89 frames. ], batch size: 242, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:13:30,197 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-02 18:13:31,988 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203741.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 18:13:33,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203744.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 18:13:37,345 INFO [train.py:968] (0/2) Epoch 5, batch 21850, giga_loss[loss=0.3188, simple_loss=0.3729, pruned_loss=0.1324, over 28710.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3572, pruned_loss=0.1136, over 5720038.15 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3827, pruned_loss=0.1251, over 5779164.46 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3531, pruned_loss=0.1106, over 5706891.52 frames. ], batch size: 92, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:13:59,321 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203773.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:14:21,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.135e+02 1.069e+03 1.319e+03 1.767e+03 5.652e+03, threshold=2.637e+03, percent-clipped=13.0
+2023-03-02 18:14:22,014 INFO [train.py:968] (0/2) Epoch 5, batch 21900, giga_loss[loss=0.2976, simple_loss=0.3671, pruned_loss=0.1141, over 28882.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3596, pruned_loss=0.1147, over 5716546.57 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.383, pruned_loss=0.1256, over 5776998.72 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3557, pruned_loss=0.1117, over 5707636.48 frames. ], batch size: 186, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:14:27,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:14:28,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7320, 2.2403, 1.7407, 1.9434], device='cuda:0'), covar=tensor([0.0684, 0.0230, 0.0290, 0.0724], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0123, 0.0126, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0062], device='cuda:0')
+2023-03-02 18:15:05,843 INFO [train.py:968] (0/2) Epoch 5, batch 21950, giga_loss[loss=0.3895, simple_loss=0.4282, pruned_loss=0.1754, over 27626.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3634, pruned_loss=0.1166, over 5705852.72 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3831, pruned_loss=0.1257, over 5779656.46 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3598, pruned_loss=0.1139, over 5695284.27 frames. ], batch size: 472, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:15:08,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:15:14,798 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=203860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:15:49,413 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.115e+02 9.872e+02 1.219e+03 1.609e+03 6.616e+03, threshold=2.439e+03, percent-clipped=8.0
+2023-03-02 18:15:49,425 INFO [train.py:968] (0/2) Epoch 5, batch 22000, giga_loss[loss=0.3199, simple_loss=0.3856, pruned_loss=0.1271, over 28357.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3657, pruned_loss=0.1173, over 5708258.02 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3839, pruned_loss=0.1267, over 5782687.76 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3617, pruned_loss=0.114, over 5695512.13 frames. ], batch size: 368, lr: 6.35e-03, grad_scale: 8.0
+2023-03-02 18:16:32,587 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:16:32,967 INFO [train.py:968] (0/2) Epoch 5, batch 22050, giga_loss[loss=0.2965, simple_loss=0.3667, pruned_loss=0.1132, over 28286.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.367, pruned_loss=0.1174, over 5702082.11 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3844, pruned_loss=0.1272, over 5775599.37 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.363, pruned_loss=0.1141, over 5697134.24 frames. ], batch size: 368, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:16:35,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:16:51,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5643, 4.3639, 1.6377, 1.5011], device='cuda:0'), covar=tensor([0.1019, 0.0256, 0.1001, 0.1530], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0472, 0.0304, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 18:17:00,495 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:17:14,884 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:17:15,018 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-02 18:17:16,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:17:17,160 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-204000.pt
+2023-03-02 18:17:17,477 INFO [train.py:968] (0/2) Epoch 5, batch 22100, giga_loss[loss=0.2831, simple_loss=0.3567, pruned_loss=0.1047, over 28901.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3663, pruned_loss=0.1166, over 5702127.16 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3846, pruned_loss=0.1275, over 5776397.79 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3626, pruned_loss=0.1136, over 5696124.95 frames. ], batch size: 227, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:17:18,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.280e+02 1.023e+03 1.397e+03 1.804e+03 6.065e+03, threshold=2.793e+03, percent-clipped=11.0
+2023-03-02 18:17:32,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4363, 1.7480, 1.6786, 1.5852], device='cuda:0'), covar=tensor([0.1475, 0.1655, 0.1177, 0.1193], device='cuda:0'), in_proj_covar=tensor([0.0771, 0.0729, 0.0773, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 18:17:42,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:17:59,062 INFO [train.py:968] (0/2) Epoch 5, batch 22150, giga_loss[loss=0.2847, simple_loss=0.3586, pruned_loss=0.1054, over 28955.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3665, pruned_loss=0.1171, over 5695232.85 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3851, pruned_loss=0.1278, over 5768789.75 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3629, pruned_loss=0.1143, over 5694981.99 frames. ], batch size: 164, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:18:07,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3078, 1.9212, 1.5275, 0.5825], device='cuda:0'), covar=tensor([0.1684, 0.1274, 0.1892, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.1359, 0.1258, 0.1351, 0.1132], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 18:18:33,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3385, 1.9364, 1.2853, 1.5549], device='cuda:0'), covar=tensor([0.0755, 0.0270, 0.0329, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0123, 0.0126, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0062], device='cuda:0')
+2023-03-02 18:18:41,804 INFO [train.py:968] (0/2) Epoch 5, batch 22200, giga_loss[loss=0.3129, simple_loss=0.3742, pruned_loss=0.1258, over 27687.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3664, pruned_loss=0.1173, over 5701137.98 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3854, pruned_loss=0.1281, over 5770611.58 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3631, pruned_loss=0.1146, over 5698547.35 frames. ], batch size: 472, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:18:43,319 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.994e+02 1.256e+03 1.475e+03 2.005e+03 5.959e+03, threshold=2.950e+03, percent-clipped=13.0
+2023-03-02 18:19:15,060 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:19:23,031 INFO [train.py:968] (0/2) Epoch 5, batch 22250, giga_loss[loss=0.2839, simple_loss=0.3573, pruned_loss=0.1052, over 29011.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3684, pruned_loss=0.1188, over 5696779.88 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3852, pruned_loss=0.1283, over 5767584.33 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3652, pruned_loss=0.116, over 5694774.83 frames. ], batch size: 155, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:19:31,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:20:04,201 INFO [train.py:968] (0/2) Epoch 5, batch 22300, giga_loss[loss=0.2754, simple_loss=0.3484, pruned_loss=0.1012, over 28694.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3713, pruned_loss=0.1201, over 5704945.83 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3857, pruned_loss=0.1288, over 5766084.70 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3681, pruned_loss=0.1174, over 5703406.37 frames. ], batch size: 92, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:20:04,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.108e+02 1.170e+03 1.426e+03 1.937e+03 5.622e+03, threshold=2.853e+03, percent-clipped=9.0
+2023-03-02 18:20:33,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:20:38,873 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:20:45,187 INFO [train.py:968] (0/2) Epoch 5, batch 22350, giga_loss[loss=0.3112, simple_loss=0.3749, pruned_loss=0.1237, over 28865.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3737, pruned_loss=0.1214, over 5697876.00 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3865, pruned_loss=0.1296, over 5756486.43 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3701, pruned_loss=0.1183, over 5704124.80 frames. ], batch size: 112, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:20:50,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9020, 1.1950, 3.3303, 2.9769], device='cuda:0'), covar=tensor([0.1520, 0.2174, 0.0398, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0557, 0.0518, 0.0733, 0.0603], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 18:21:19,572 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204293.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 18:21:25,251 INFO [train.py:968] (0/2) Epoch 5, batch 22400, giga_loss[loss=0.3267, simple_loss=0.3894, pruned_loss=0.132, over 29078.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.375, pruned_loss=0.1222, over 5707274.19 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.387, pruned_loss=0.1301, over 5757882.78 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3714, pruned_loss=0.1191, over 5709491.80 frames. ], batch size: 136, lr: 6.35e-03, grad_scale: 8.0
+2023-03-02 18:21:25,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.734e+02 1.357e+03 1.950e+03 2.776e+03 6.792e+03, threshold=3.900e+03, percent-clipped=23.0
+2023-03-02 18:21:52,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3546, 3.4650, 1.4486, 1.3221], device='cuda:0'), covar=tensor([0.0841, 0.0329, 0.0836, 0.1285], device='cuda:0'), in_proj_covar=tensor([0.0331, 0.0475, 0.0305, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 18:22:05,735 INFO [train.py:968] (0/2) Epoch 5, batch 22450, giga_loss[loss=0.3312, simple_loss=0.3998, pruned_loss=0.1313, over 28994.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3747, pruned_loss=0.1216, over 5710833.00 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.387, pruned_loss=0.1301, over 5760607.79 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3715, pruned_loss=0.1189, over 5709156.41 frames. ], batch size: 213, lr: 6.35e-03, grad_scale: 8.0
+2023-03-02 18:22:08,634 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-02 18:22:30,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:22:33,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:22:48,086 INFO [train.py:968] (0/2) Epoch 5, batch 22500, giga_loss[loss=0.3271, simple_loss=0.3854, pruned_loss=0.1344, over 28616.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3751, pruned_loss=0.1219, over 5717257.76 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3874, pruned_loss=0.1306, over 5764968.12 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3716, pruned_loss=0.1189, over 5709910.10 frames. ], batch size: 85, lr: 6.34e-03, grad_scale: 8.0
+2023-03-02 18:22:48,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.137e+02 1.153e+03 1.515e+03 1.963e+03 4.269e+03, threshold=3.030e+03, percent-clipped=1.0
+2023-03-02 18:22:53,705 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4545, 1.8690, 1.4643, 1.4254], device='cuda:0'), covar=tensor([0.0730, 0.0258, 0.0305, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0124, 0.0127, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0055, 0.0041, 0.0037, 0.0062], device='cuda:0')
+2023-03-02 18:22:55,924 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:22:59,086 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-02 18:23:05,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1650, 1.7107, 1.5682, 1.4342], device='cuda:0'), covar=tensor([0.1379, 0.1761, 0.1153, 0.1184], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0732, 0.0774, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 18:23:26,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4733, 2.1751, 1.7198, 0.6174], device='cuda:0'), covar=tensor([0.3037, 0.1437, 0.2019, 0.3169], device='cuda:0'), in_proj_covar=tensor([0.1374, 0.1277, 0.1369, 0.1149], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 18:23:28,381 INFO [train.py:968] (0/2) Epoch 5, batch 22550, giga_loss[loss=0.2714, simple_loss=0.3374, pruned_loss=0.1027, over 28745.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3733, pruned_loss=0.121, over 5720261.91 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.3877, pruned_loss=0.1312, over 5768508.41 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.37, pruned_loss=0.1179, over 5710413.33 frames. ], batch size: 99, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:23:44,732 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:24:10,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2573, 1.1539, 1.1440, 0.9754], device='cuda:0'), covar=tensor([0.0605, 0.0520, 0.0904, 0.0792], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0446, 0.0503, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 18:24:11,805 INFO [train.py:968] (0/2) Epoch 5, batch 22600, giga_loss[loss=0.2604, simple_loss=0.3339, pruned_loss=0.09347, over 28652.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3709, pruned_loss=0.1197, over 5726884.58 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3882, pruned_loss=0.1318, over 5771432.26 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3674, pruned_loss=0.1165, over 5715293.08 frames. ], batch size: 92, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:24:12,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5931, 1.5044, 1.3207, 1.2985], device='cuda:0'), covar=tensor([0.0610, 0.0567, 0.0919, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0446, 0.0503, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 18:24:13,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.279e+02 1.174e+03 1.514e+03 2.133e+03 5.306e+03, threshold=3.027e+03, percent-clipped=11.0
+2023-03-02 18:24:24,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:24:39,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:24:51,781 INFO [train.py:968] (0/2) Epoch 5, batch 22650, giga_loss[loss=0.2667, simple_loss=0.3378, pruned_loss=0.09778, over 28899.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3675, pruned_loss=0.118, over 5724672.77 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3886, pruned_loss=0.1322, over 5773962.98 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.364, pruned_loss=0.1148, over 5712310.64 frames. ], batch size: 227, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:24:56,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:25:16,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4804, 3.3988, 1.5825, 1.5051], device='cuda:0'), covar=tensor([0.0801, 0.0284, 0.0834, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.0330, 0.0476, 0.0304, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 18:25:17,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5406, 0.9472, 2.8551, 2.5670], device='cuda:0'), covar=tensor([0.1649, 0.2161, 0.0538, 0.0754], device='cuda:0'), in_proj_covar=tensor([0.0564, 0.0520, 0.0744, 0.0609], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 18:25:31,281 INFO [train.py:968] (0/2) Epoch 5, batch 22700, giga_loss[loss=0.2705, simple_loss=0.3582, pruned_loss=0.09138, over 28619.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3676, pruned_loss=0.1175, over 5724145.86 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3887, pruned_loss=0.1326, over 5775255.84 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3641, pruned_loss=0.1142, over 5711888.33 frames. ], batch size: 336, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:25:32,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.770e+02 1.063e+03 1.226e+03 1.521e+03 3.280e+03, threshold=2.451e+03, percent-clipped=2.0
+2023-03-02 18:25:43,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:13,837 INFO [train.py:968] (0/2) Epoch 5, batch 22750, libri_loss[loss=0.3688, simple_loss=0.4117, pruned_loss=0.163, over 29529.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.37, pruned_loss=0.1177, over 5711808.01 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.389, pruned_loss=0.133, over 5765178.71 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3666, pruned_loss=0.1144, over 5708730.93 frames. ], batch size: 84, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:26:20,540 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:23,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:30,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204668.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:26:36,134 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:39,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:48,280 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:55,683 INFO [train.py:968] (0/2) Epoch 5, batch 22800, giga_loss[loss=0.2944, simple_loss=0.3678, pruned_loss=0.1105, over 28764.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3702, pruned_loss=0.1179, over 5720027.08 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3893, pruned_loss=0.1332, over 5766288.03 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3671, pruned_loss=0.115, over 5716112.15 frames. ], batch size: 199, lr: 6.34e-03, grad_scale: 8.0
+2023-03-02 18:26:56,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.817e+02 1.094e+03 1.337e+03 1.903e+03 4.352e+03, threshold=2.673e+03, percent-clipped=13.0
+2023-03-02 18:27:01,981 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:27:39,034 INFO [train.py:968] (0/2) Epoch 5, batch 22850, giga_loss[loss=0.2846, simple_loss=0.3398, pruned_loss=0.1147, over 28058.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3682, pruned_loss=0.1177, over 5721436.53 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3892, pruned_loss=0.1332, over 5767815.69 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3656, pruned_loss=0.1152, over 5716540.80 frames. ], batch size: 77, lr: 6.34e-03, grad_scale: 8.0
+2023-03-02 18:27:45,307 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:27:47,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:28:01,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8934, 4.7170, 1.8715, 2.0447], device='cuda:0'), covar=tensor([0.0775, 0.0315, 0.0818, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0331, 0.0476, 0.0305, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 18:28:08,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:28:17,640 INFO [train.py:968] (0/2) Epoch 5, batch 22900, giga_loss[loss=0.283, simple_loss=0.3446, pruned_loss=0.1107, over 28765.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3667, pruned_loss=0.1184, over 5710084.10 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3898, pruned_loss=0.1336, over 5752119.46 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3638, pruned_loss=0.1158, over 5718922.49 frames. ], batch size: 119, lr: 6.34e-03, grad_scale: 8.0
+2023-03-02 18:28:18,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.769e+02 1.112e+03 1.349e+03 1.922e+03 4.354e+03, threshold=2.698e+03, percent-clipped=5.0
+2023-03-02 18:28:27,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204811.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:28:29,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3197, 4.1477, 3.9405, 1.7800], device='cuda:0'), covar=tensor([0.0389, 0.0448, 0.0717, 0.2272], device='cuda:0'), in_proj_covar=tensor([0.0841, 0.0761, 0.0792, 0.0601], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 18:28:30,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204814.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:28:53,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204843.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:28:55,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:28:58,673 INFO [train.py:968] (0/2) Epoch 5, batch 22950, giga_loss[loss=0.3133, simple_loss=0.3772, pruned_loss=0.1247, over 28934.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.366, pruned_loss=0.1194, over 5704205.16 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3901, pruned_loss=0.1339, over 5745163.31 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.363, pruned_loss=0.1169, over 5716537.08 frames. ], batch size: 145, lr: 6.34e-03, grad_scale: 8.0
+2023-03-02 18:29:16,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6378, 0.9820, 2.8444, 2.7558], device='cuda:0'), covar=tensor([0.1550, 0.2093, 0.0493, 0.0682], device='cuda:0'), in_proj_covar=tensor([0.0563, 0.0519, 0.0742, 0.0608], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 18:29:39,664 INFO [train.py:968] (0/2) Epoch 5, batch 23000, giga_loss[loss=0.2955, simple_loss=0.3477, pruned_loss=0.1216, over 28763.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3642, pruned_loss=0.1188, over 5711479.65 frames. ], libri_tot_loss[loss=0.3292, simple_loss=0.3903, pruned_loss=0.1341, over 5746998.17 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3612, pruned_loss=0.1164, over 5719076.99 frames. ], batch size: 99, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:29:41,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.872e+02 1.130e+03 1.429e+03 1.754e+03 4.879e+03, threshold=2.859e+03, percent-clipped=9.0
+2023-03-02 18:30:05,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:30:20,355 INFO [train.py:968] (0/2) Epoch 5, batch 23050, libri_loss[loss=0.2881, simple_loss=0.3576, pruned_loss=0.1093, over 29578.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3608, pruned_loss=0.1168, over 5701515.45 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3906, pruned_loss=0.1343, over 5739027.19 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3577, pruned_loss=0.1144, over 5714360.91 frames. ], batch size: 75, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:30:21,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0133, 1.2329, 4.2646, 3.4449], device='cuda:0'), covar=tensor([0.1722, 0.2318, 0.0319, 0.0768], device='cuda:0'), in_proj_covar=tensor([0.0565, 0.0519, 0.0742, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 18:30:49,554 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:30:51,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:30:58,040 INFO [train.py:968] (0/2) Epoch 5, batch 23100, giga_loss[loss=0.2747, simple_loss=0.3416, pruned_loss=0.1039, over 28644.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3569, pruned_loss=0.115, over 5709823.04 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3905, pruned_loss=0.1344, over 5740551.94 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.354, pruned_loss=0.1126, over 5718119.44 frames. ], batch size: 262, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:31:00,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.864e+02 1.156e+03 1.463e+03 2.167e+03 3.929e+03, threshold=2.925e+03, percent-clipped=4.0
+2023-03-02 18:31:15,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:31:37,652 INFO [train.py:968] (0/2) Epoch 5, batch 23150, giga_loss[loss=0.2382, simple_loss=0.3131, pruned_loss=0.08162, over 28900.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3524, pruned_loss=0.1122, over 5705551.43 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.3909, pruned_loss=0.1348, over 5734112.58 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3488, pruned_loss=0.1095, over 5717334.93 frames. ], batch size: 106, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:31:57,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:31:59,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:32:03,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3021, 1.8369, 1.7268, 1.5520], device='cuda:0'), covar=tensor([0.1457, 0.1836, 0.1153, 0.1221], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0736, 0.0778, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 18:32:18,628 INFO [train.py:968] (0/2) Epoch 5, batch 23200, giga_loss[loss=0.2982, simple_loss=0.3634, pruned_loss=0.1165, over 27605.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3521, pruned_loss=0.1116, over 5695292.95 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.391, pruned_loss=0.135, over 5724491.26 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3485, pruned_loss=0.1089, over 5712407.81 frames. ], batch size: 472, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:32:22,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.744e+02 1.268e+03 1.673e+03 2.337e+03 7.037e+03, threshold=3.346e+03, percent-clipped=15.0
+2023-03-02 18:32:24,741 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:33:01,873 INFO [train.py:968] (0/2) Epoch 5, batch 23250, giga_loss[loss=0.2802, simple_loss=0.3602, pruned_loss=0.1001, over 28930.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3569, pruned_loss=0.1142, over 5694585.55 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3911, pruned_loss=0.1352, over 5719979.89 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3529, pruned_loss=0.1112, over 5711854.28 frames. ], batch size: 213, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:33:43,093 INFO [train.py:968] (0/2) Epoch 5, batch 23300, libri_loss[loss=0.3114, simple_loss=0.3616, pruned_loss=0.1306, over 29600.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3612, pruned_loss=0.1165, over 5701725.72 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3909, pruned_loss=0.1353, over 5724454.42 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3573, pruned_loss=0.1135, over 5710703.34 frames. ], batch size: 74, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:33:45,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.758e+02 1.254e+03 1.649e+03 2.344e+03 5.097e+03, threshold=3.299e+03, percent-clipped=14.0
+2023-03-02 18:33:54,691 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=205214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:34:22,463 INFO [train.py:968] (0/2) Epoch 5, batch 23350, giga_loss[loss=0.3121, simple_loss=0.3758, pruned_loss=0.1242, over 29150.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3649, pruned_loss=0.1179, over 5697053.07 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3912, pruned_loss=0.1356, over 5718713.29 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3606, pruned_loss=0.1145, over 5709321.95 frames. ], batch size: 113, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:34:23,987 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-02 18:34:46,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4499, 1.7819, 1.7488, 1.6072], device='cuda:0'), covar=tensor([0.1509, 0.1804, 0.1166, 0.1203], device='cuda:0'), in_proj_covar=tensor([0.0768, 0.0734, 0.0775, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 18:35:04,327 INFO [train.py:968] (0/2) Epoch 5, batch 23400, giga_loss[loss=0.2859, simple_loss=0.3643, pruned_loss=0.1037, over 29007.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3671, pruned_loss=0.1185, over 5702103.71 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3913, pruned_loss=0.1359, over 5714590.94 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.363, pruned_loss=0.1152, over 5714865.21 frames. ], batch size: 213, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:35:07,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.447e+02 1.171e+03 1.403e+03 1.742e+03 5.228e+03, threshold=2.806e+03, percent-clipped=4.0
+2023-03-02 18:35:47,722 INFO [train.py:968] (0/2) Epoch 5, batch 23450, giga_loss[loss=0.268, simple_loss=0.3421, pruned_loss=0.09699, over 28550.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3689, pruned_loss=0.1195, over 5710849.59 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3913, pruned_loss=0.1362, over 5718482.79 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3652, pruned_loss=0.1164, over 5717212.81 frames. ], batch size: 78, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:36:36,583 INFO [train.py:968] (0/2) Epoch 5, batch 23500, giga_loss[loss=0.3308, simple_loss=0.385, pruned_loss=0.1383, over 28892.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3766, pruned_loss=0.1267, over 5701850.48 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3916, pruned_loss=0.1364, over 5723340.21 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3728, pruned_loss=0.1235, over 5701966.67 frames. ], batch size: 66, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:36:40,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.591e+02 1.277e+03 1.821e+03 2.526e+03 6.766e+03, threshold=3.641e+03, percent-clipped=14.0
+2023-03-02 18:36:46,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2002, 1.8188, 1.5171, 1.3861], device='cuda:0'), covar=tensor([0.0785, 0.0288, 0.0308, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0124, 0.0128, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:0')
+2023-03-02 18:37:23,259 INFO [train.py:968] (0/2) Epoch 5, batch 23550, libri_loss[loss=0.3413, simple_loss=0.4016, pruned_loss=0.1405, over 28565.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3824, pruned_loss=0.1314, over 5699706.06 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3916, pruned_loss=0.1367, over 5728194.65 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3789, pruned_loss=0.1284, over 5694559.10 frames. ], batch size: 106, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:37:38,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-02 18:38:03,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9960, 1.3477, 1.0895, 0.1541], device='cuda:0'), covar=tensor([0.1509, 0.1198, 0.2224, 0.2492], device='cuda:0'), in_proj_covar=tensor([0.1374, 0.1275, 0.1371, 0.1136], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 18:38:14,874 INFO [train.py:968] (0/2) Epoch 5, batch 23600, giga_loss[loss=0.3337, simple_loss=0.3986, pruned_loss=0.1344, over 28974.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3899, pruned_loss=0.1376, over 5683365.44 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3919, pruned_loss=0.1371, over 5722544.04 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3868, pruned_loss=0.1348, over 5683150.68 frames. ], batch size: 186, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:38:17,173 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=205502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:38:18,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.630e+03 2.221e+03 3.288e+03 1.341e+04, threshold=4.442e+03, percent-clipped=13.0
+2023-03-02 18:39:03,373 INFO [train.py:968] (0/2) Epoch 5, batch 23650, giga_loss[loss=0.4496, simple_loss=0.4461, pruned_loss=0.2266, over 23438.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3959, pruned_loss=0.1426, over 5667286.28 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3926, pruned_loss=0.1377, over 5707437.31 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3927, pruned_loss=0.1399, over 5679624.21 frames. ], batch size: 705, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:39:33,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5657, 1.9477, 1.8374, 1.6856], device='cuda:0'), covar=tensor([0.1313, 0.1623, 0.1042, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0763, 0.0735, 0.0771, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 18:39:44,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:39:45,711 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=205591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:39:54,593 INFO [train.py:968] (0/2) Epoch 5, batch 23700, giga_loss[loss=0.4508, simple_loss=0.453, pruned_loss=0.2243, over 23411.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.4023, pruned_loss=0.1486, over 5669045.84 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3922, pruned_loss=0.1374, over 5710481.37 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.4003, pruned_loss=0.1468, over 5675515.63 frames. ], batch size: 705, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:40:01,269 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.598e+02 1.438e+03 1.792e+03 2.525e+03 1.023e+04, threshold=3.585e+03, percent-clipped=4.0
+2023-03-02 18:40:38,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4918, 1.9924, 1.8865, 1.6949], device='cuda:0'), covar=tensor([0.1388, 0.1684, 0.1034, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0760, 0.0732, 0.0767, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-02 18:40:42,323 INFO [train.py:968] (0/2) Epoch 5, batch 23750, giga_loss[loss=0.4231, simple_loss=0.4514, pruned_loss=0.1974, over 28299.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4053, pruned_loss=0.1515, over 5662414.00 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3921, pruned_loss=0.1376, over 5704665.67 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.404, pruned_loss=0.1502, over 5672066.62 frames. ], batch size: 368, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:41:33,270 INFO [train.py:968] (0/2) Epoch 5, batch 23800, giga_loss[loss=0.326, simple_loss=0.3814, pruned_loss=0.1353, over 28757.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4074, pruned_loss=0.1542, over 5656117.22 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3926, pruned_loss=0.138, over 5701971.09 frames. ], giga_tot_loss[loss=0.3561, simple_loss=0.4062, pruned_loss=0.1531, over 5665434.29 frames. ], batch size: 242, lr: 6.32e-03, grad_scale: 2.0
+2023-03-02 18:41:40,646 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.746e+03 2.092e+03 3.169e+03 7.482e+03, threshold=4.184e+03, percent-clipped=18.0
+2023-03-02 18:41:51,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-02 18:42:11,358 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:42:15,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:42:29,930 INFO [train.py:968] (0/2) Epoch 5, batch 23850, giga_loss[loss=0.4535, simple_loss=0.4701, pruned_loss=0.2184, over 27854.00 frames. ], tot_loss[loss=0.3677, simple_loss=0.4134, pruned_loss=0.1609, over 5644212.31 frames. ], libri_tot_loss[loss=0.3344, simple_loss=0.3928, pruned_loss=0.138, over 5703120.23 frames. ], giga_tot_loss[loss=0.3663, simple_loss=0.4124, pruned_loss=0.1601, over 5650218.22 frames. ], batch size: 412, lr: 6.32e-03, grad_scale: 2.0
+2023-03-02 18:42:44,549 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:43:25,377 INFO [train.py:968] (0/2) Epoch 5, batch 23900, giga_loss[loss=0.3666, simple_loss=0.4191, pruned_loss=0.157, over 28620.00 frames. ], tot_loss[loss=0.3732, simple_loss=0.417, pruned_loss=0.1647, over 5632670.26 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3929, pruned_loss=0.1381, over 5698095.49 frames. ], giga_tot_loss[loss=0.3722, simple_loss=0.4163, pruned_loss=0.1641, over 5640956.49 frames. ], batch size: 336, lr: 6.32e-03, grad_scale: 2.0
+2023-03-02 18:43:35,194 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.011e+02 1.619e+03 2.251e+03 3.010e+03 7.500e+03, threshold=4.503e+03, percent-clipped=10.0
+2023-03-02 18:44:21,351 INFO [train.py:968] (0/2) Epoch 5, batch 23950, giga_loss[loss=0.3671, simple_loss=0.4102, pruned_loss=0.162, over 28811.00 frames. ], tot_loss[loss=0.3745, simple_loss=0.4181, pruned_loss=0.1655, over 5643785.50 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3927, pruned_loss=0.1381, over 5703912.39 frames. ], giga_tot_loss[loss=0.375, simple_loss=0.4185, pruned_loss=0.1658, over 5643515.07 frames. ], batch size: 284, lr: 6.32e-03, grad_scale: 2.0
+2023-03-02 18:44:26,037 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-02 18:44:51,814 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:45:14,067 INFO [train.py:968] (0/2) Epoch 5, batch 24000, giga_loss[loss=0.4001, simple_loss=0.4297, pruned_loss=0.1853, over 28719.00 frames. ], tot_loss[loss=0.3744, simple_loss=0.4174, pruned_loss=0.1657, over 5633621.56 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3929, pruned_loss=0.1383, over 5706344.29 frames. ], giga_tot_loss[loss=0.3751, simple_loss=0.4178, pruned_loss=0.1662, over 5630305.11 frames. ], batch size: 243, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:45:14,072 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 18:45:22,826 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2388, simple_loss=0.3426, pruned_loss=0.06746, over 944034.00 frames. 
+2023-03-02 18:45:22,827 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 18:45:28,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.868e+02 1.770e+03 2.194e+03 2.893e+03 1.018e+04, threshold=4.388e+03, percent-clipped=7.0
+2023-03-02 18:46:07,859 INFO [train.py:968] (0/2) Epoch 5, batch 24050, giga_loss[loss=0.3282, simple_loss=0.391, pruned_loss=0.1327, over 28784.00 frames. ], tot_loss[loss=0.3724, simple_loss=0.4158, pruned_loss=0.1644, over 5634319.77 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3937, pruned_loss=0.139, over 5695256.96 frames. ], giga_tot_loss[loss=0.3729, simple_loss=0.4161, pruned_loss=0.1649, over 5639641.12 frames. ], batch size: 99, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:46:25,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:46:53,649 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-206000.pt
+2023-03-02 18:46:53,953 INFO [train.py:968] (0/2) Epoch 5, batch 24100, giga_loss[loss=0.365, simple_loss=0.4156, pruned_loss=0.1572, over 28786.00 frames. ], tot_loss[loss=0.3694, simple_loss=0.4145, pruned_loss=0.1622, over 5644779.19 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3929, pruned_loss=0.1385, over 5699188.95 frames. ], giga_tot_loss[loss=0.3717, simple_loss=0.4161, pruned_loss=0.1636, over 5643926.82 frames. ], batch size: 284, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:47:00,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.527e+03 1.858e+03 2.290e+03 5.639e+03, threshold=3.715e+03, percent-clipped=5.0
+2023-03-02 18:47:17,111 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:47:20,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:47:21,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2455, 1.5052, 1.2205, 1.4025], device='cuda:0'), covar=tensor([0.0762, 0.0347, 0.0340, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0125, 0.0129, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 18:47:44,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:47:46,010 INFO [train.py:968] (0/2) Epoch 5, batch 24150, giga_loss[loss=0.4272, simple_loss=0.4574, pruned_loss=0.1985, over 28787.00 frames. ], tot_loss[loss=0.3682, simple_loss=0.414, pruned_loss=0.1612, over 5644484.44 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3926, pruned_loss=0.1384, over 5704168.51 frames. ], giga_tot_loss[loss=0.371, simple_loss=0.416, pruned_loss=0.163, over 5637874.82 frames. ], batch size: 99, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:47:48,867 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:47:52,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9984, 1.7675, 1.3704, 1.6051], device='cuda:0'), covar=tensor([0.0538, 0.0567, 0.0935, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0451, 0.0501, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 18:47:52,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3814, 1.4912, 1.2320, 1.4959], device='cuda:0'), covar=tensor([0.2135, 0.2028, 0.2074, 0.1938], device='cuda:0'), in_proj_covar=tensor([0.1122, 0.0871, 0.0995, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 18:48:41,739 INFO [train.py:968] (0/2) Epoch 5, batch 24200, giga_loss[loss=0.3957, simple_loss=0.4113, pruned_loss=0.19, over 24082.00 frames. ], tot_loss[loss=0.3692, simple_loss=0.4149, pruned_loss=0.1617, over 5630346.49 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3925, pruned_loss=0.1384, over 5707263.31 frames. ], giga_tot_loss[loss=0.3719, simple_loss=0.4168, pruned_loss=0.1635, over 5621493.10 frames. ], batch size: 705, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:48:48,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.151e+02 1.591e+03 2.085e+03 3.155e+03 7.037e+03, threshold=4.170e+03, percent-clipped=13.0
+2023-03-02 18:48:51,962 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:48:53,753 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:49:17,851 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:49:17,896 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8912, 1.0126, 3.7516, 3.1880], device='cuda:0'), covar=tensor([0.1734, 0.2345, 0.0393, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0576, 0.0532, 0.0755, 0.0617], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 18:49:17,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3093, 1.6095, 1.2235, 1.4303], device='cuda:0'), covar=tensor([0.0772, 0.0344, 0.0348, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0125, 0.0129, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0042, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 18:49:22,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:49:32,826 INFO [train.py:968] (0/2) Epoch 5, batch 24250, giga_loss[loss=0.3022, simple_loss=0.3759, pruned_loss=0.1143, over 28899.00 frames. ], tot_loss[loss=0.3621, simple_loss=0.4101, pruned_loss=0.157, over 5635105.61 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3929, pruned_loss=0.1389, over 5712396.16 frames. ], giga_tot_loss[loss=0.3646, simple_loss=0.4119, pruned_loss=0.1586, over 5621520.69 frames. ], batch size: 213, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:50:22,102 INFO [train.py:968] (0/2) Epoch 5, batch 24300, giga_loss[loss=0.3504, simple_loss=0.4072, pruned_loss=0.1467, over 28708.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4073, pruned_loss=0.1532, over 5645062.45 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3924, pruned_loss=0.1387, over 5716561.19 frames. ], giga_tot_loss[loss=0.3597, simple_loss=0.4095, pruned_loss=0.155, over 5629240.49 frames. ], batch size: 99, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:50:29,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.576e+02 1.616e+03 1.959e+03 3.145e+03 9.066e+03, threshold=3.917e+03, percent-clipped=15.0
+2023-03-02 18:51:11,813 INFO [train.py:968] (0/2) Epoch 5, batch 24350, giga_loss[loss=0.3386, simple_loss=0.3954, pruned_loss=0.1408, over 28021.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.403, pruned_loss=0.149, over 5665635.02 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3921, pruned_loss=0.1387, over 5721967.96 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4053, pruned_loss=0.1507, over 5646492.60 frames. ], batch size: 412, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:51:44,015 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:51:59,990 INFO [train.py:968] (0/2) Epoch 5, batch 24400, giga_loss[loss=0.3167, simple_loss=0.3799, pruned_loss=0.1267, over 28891.00 frames. ], tot_loss[loss=0.3457, simple_loss=0.3997, pruned_loss=0.1458, over 5673468.21 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.392, pruned_loss=0.1387, over 5722784.87 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.4019, pruned_loss=0.1474, over 5656052.88 frames. ], batch size: 174, lr: 6.32e-03, grad_scale: 8.0
+2023-03-02 18:52:05,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.424e+02 1.641e+03 2.035e+03 2.933e+03 1.051e+04, threshold=4.071e+03, percent-clipped=11.0
+2023-03-02 18:52:29,731 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-02 18:52:48,397 INFO [train.py:968] (0/2) Epoch 5, batch 24450, giga_loss[loss=0.3079, simple_loss=0.3809, pruned_loss=0.1175, over 29025.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3987, pruned_loss=0.1455, over 5668645.37 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3919, pruned_loss=0.1387, over 5723631.62 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.4005, pruned_loss=0.1468, over 5653765.52 frames. ], batch size: 155, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:53:34,415 INFO [train.py:968] (0/2) Epoch 5, batch 24500, giga_loss[loss=0.329, simple_loss=0.3861, pruned_loss=0.136, over 28236.00 frames. ], tot_loss[loss=0.3457, simple_loss=0.3987, pruned_loss=0.1463, over 5672246.19 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3915, pruned_loss=0.1389, over 5730751.38 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.4009, pruned_loss=0.1475, over 5651315.01 frames. ], batch size: 368, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:53:39,820 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 18:53:40,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.090e+02 1.584e+03 2.064e+03 2.865e+03 7.630e+03, threshold=4.128e+03, percent-clipped=11.0
+2023-03-02 18:53:59,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:54:08,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2535, 1.5176, 1.2090, 0.8129], device='cuda:0'), covar=tensor([0.1158, 0.0939, 0.0651, 0.1001], device='cuda:0'), in_proj_covar=tensor([0.1393, 0.1184, 0.1183, 0.1249], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 18:54:24,295 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:54:24,761 INFO [train.py:968] (0/2) Epoch 5, batch 24550, giga_loss[loss=0.3212, simple_loss=0.3844, pruned_loss=0.129, over 28922.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3972, pruned_loss=0.1444, over 5684629.68 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3915, pruned_loss=0.1391, over 5734334.52 frames. ], giga_tot_loss[loss=0.3448, simple_loss=0.399, pruned_loss=0.1453, over 5663888.78 frames. ], batch size: 213, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:55:02,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6500, 2.3272, 1.4515, 0.7584], device='cuda:0'), covar=tensor([0.3848, 0.1983, 0.2293, 0.3537], device='cuda:0'), in_proj_covar=tensor([0.1399, 0.1295, 0.1384, 0.1151], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 18:55:17,165 INFO [train.py:968] (0/2) Epoch 5, batch 24600, giga_loss[loss=0.3182, simple_loss=0.3929, pruned_loss=0.1217, over 28874.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3971, pruned_loss=0.1423, over 5683413.34 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3918, pruned_loss=0.1395, over 5732493.42 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3983, pruned_loss=0.1426, over 5667945.62 frames. ], batch size: 145, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:55:23,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.476e+02 1.435e+03 1.915e+03 2.814e+03 8.747e+03, threshold=3.831e+03, percent-clipped=14.0
+2023-03-02 18:55:31,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:56:12,140 INFO [train.py:968] (0/2) Epoch 5, batch 24650, giga_loss[loss=0.3552, simple_loss=0.418, pruned_loss=0.1462, over 28943.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3982, pruned_loss=0.1412, over 5665755.38 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.392, pruned_loss=0.1396, over 5729825.95 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.399, pruned_loss=0.1414, over 5655562.19 frames. ], batch size: 213, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:56:28,965 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:56:32,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:56:40,883 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-02 18:56:50,978 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:56:59,830 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:57:01,576 INFO [train.py:968] (0/2) Epoch 5, batch 24700, giga_loss[loss=0.3439, simple_loss=0.3979, pruned_loss=0.1449, over 29014.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3999, pruned_loss=0.1431, over 5673104.68 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.392, pruned_loss=0.1396, over 5734370.05 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.4007, pruned_loss=0.1433, over 5658990.20 frames. ], batch size: 227, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:57:08,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.749e+02 1.444e+03 1.830e+03 2.466e+03 4.527e+03, threshold=3.660e+03, percent-clipped=2.0
+2023-03-02 18:57:12,471 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-02 18:57:20,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5351, 4.3853, 4.2130, 1.7079], device='cuda:0'), covar=tensor([0.0443, 0.0496, 0.0772, 0.2110], device='cuda:0'), in_proj_covar=tensor([0.0858, 0.0787, 0.0812, 0.0600], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 18:57:28,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9001, 1.0568, 0.8441, 0.2401], device='cuda:0'), covar=tensor([0.1367, 0.1013, 0.1372, 0.2218], device='cuda:0'), in_proj_covar=tensor([0.1395, 0.1287, 0.1377, 0.1146], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 18:57:52,781 INFO [train.py:968] (0/2) Epoch 5, batch 24750, giga_loss[loss=0.3435, simple_loss=0.4, pruned_loss=0.1435, over 28869.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.401, pruned_loss=0.1447, over 5663877.19 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.392, pruned_loss=0.1396, over 5735249.20 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.4017, pruned_loss=0.1448, over 5651866.10 frames. ], batch size: 186, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:57:57,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:58:00,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:58:00,317 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:58:33,166 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:58:44,954 INFO [train.py:968] (0/2) Epoch 5, batch 24800, giga_loss[loss=0.3069, simple_loss=0.3692, pruned_loss=0.1224, over 28381.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3998, pruned_loss=0.1454, over 5652742.20 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3922, pruned_loss=0.1399, over 5727459.46 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.4003, pruned_loss=0.1454, over 5649245.50 frames. ], batch size: 71, lr: 6.31e-03, grad_scale: 8.0
+2023-03-02 18:58:51,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.637e+02 1.698e+03 2.303e+03 3.141e+03 1.013e+04, threshold=4.606e+03, percent-clipped=18.0
+2023-03-02 18:58:51,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 18:59:31,764 INFO [train.py:968] (0/2) Epoch 5, batch 24850, giga_loss[loss=0.2988, simple_loss=0.3717, pruned_loss=0.113, over 28977.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3974, pruned_loss=0.1443, over 5663104.72 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3923, pruned_loss=0.14, over 5729697.12 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3978, pruned_loss=0.1443, over 5657465.05 frames. ], batch size: 164, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:59:47,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3842, 1.8333, 1.2912, 1.6327], device='cuda:0'), covar=tensor([0.0754, 0.0313, 0.0326, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0125, 0.0129, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0057, 0.0042, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 19:00:16,800 INFO [train.py:968] (0/2) Epoch 5, batch 24900, giga_loss[loss=0.3143, simple_loss=0.3883, pruned_loss=0.1202, over 28933.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3963, pruned_loss=0.1433, over 5669438.68 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3926, pruned_loss=0.1403, over 5732726.80 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3963, pruned_loss=0.143, over 5661484.55 frames. ], batch size: 145, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:00:17,856 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:00:20,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:00:22,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.725e+02 1.603e+03 2.234e+03 2.868e+03 7.395e+03, threshold=4.469e+03, percent-clipped=3.0
+2023-03-02 19:00:28,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0332, 1.3237, 1.0785, 0.2906], device='cuda:0'), covar=tensor([0.1614, 0.1486, 0.2541, 0.2845], device='cuda:0'), in_proj_covar=tensor([0.1390, 0.1276, 0.1378, 0.1140], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 19:00:35,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:00:42,643 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:01:01,366 INFO [train.py:968] (0/2) Epoch 5, batch 24950, giga_loss[loss=0.3117, simple_loss=0.3855, pruned_loss=0.119, over 28762.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3964, pruned_loss=0.1422, over 5681102.98 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.393, pruned_loss=0.1406, over 5736728.76 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3962, pruned_loss=0.1418, over 5670090.28 frames. ], batch size: 119, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:01:50,669 INFO [train.py:968] (0/2) Epoch 5, batch 25000, giga_loss[loss=0.3421, simple_loss=0.4004, pruned_loss=0.1419, over 28725.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3961, pruned_loss=0.1418, over 5671973.93 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3934, pruned_loss=0.1409, over 5736933.46 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3956, pruned_loss=0.1412, over 5662011.96 frames. ], batch size: 262, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:01:57,808 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.289e+02 1.537e+03 1.995e+03 2.802e+03 4.730e+03, threshold=3.991e+03, percent-clipped=2.0
+2023-03-02 19:02:38,780 INFO [train.py:968] (0/2) Epoch 5, batch 25050, giga_loss[loss=0.3131, simple_loss=0.3781, pruned_loss=0.124, over 28624.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3953, pruned_loss=0.1413, over 5678410.36 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3931, pruned_loss=0.1409, over 5740062.86 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3952, pruned_loss=0.1408, over 5665923.61 frames. ], batch size: 307, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:02:52,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:02:54,082 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:02:56,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:03:24,586 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:03:28,249 INFO [train.py:968] (0/2) Epoch 5, batch 25100, giga_loss[loss=0.303, simple_loss=0.3634, pruned_loss=0.1213, over 28506.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3945, pruned_loss=0.1412, over 5691216.70 frames. ], libri_tot_loss[loss=0.3388, simple_loss=0.394, pruned_loss=0.1418, over 5740709.94 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3936, pruned_loss=0.14, over 5679216.85 frames. ], batch size: 85, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:03:31,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5160, 3.5003, 1.6585, 1.4453], device='cuda:0'), covar=tensor([0.0816, 0.0297, 0.0758, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0481, 0.0309, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 19:03:35,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.997e+02 1.708e+03 2.292e+03 3.077e+03 7.312e+03, threshold=4.585e+03, percent-clipped=10.0
+2023-03-02 19:04:13,842 INFO [train.py:968] (0/2) Epoch 5, batch 25150, giga_loss[loss=0.3377, simple_loss=0.3719, pruned_loss=0.1518, over 23507.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3939, pruned_loss=0.1417, over 5666926.92 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3943, pruned_loss=0.142, over 5722239.02 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.393, pruned_loss=0.1406, over 5672060.55 frames. ], batch size: 705, lr: 6.30e-03, grad_scale: 2.0
+2023-03-02 19:04:16,454 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 19:04:55,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4048, 4.2924, 4.1227, 1.8334], device='cuda:0'), covar=tensor([0.0415, 0.0455, 0.0692, 0.1940], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0795, 0.0815, 0.0599], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 19:04:59,734 INFO [train.py:968] (0/2) Epoch 5, batch 25200, giga_loss[loss=0.2833, simple_loss=0.3548, pruned_loss=0.1059, over 28314.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3925, pruned_loss=0.1408, over 5682211.01 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3949, pruned_loss=0.1425, over 5721614.95 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3911, pruned_loss=0.1394, over 5686322.26 frames. ], batch size: 65, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:05:08,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=207108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:05:10,660 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.646e+03 2.259e+03 3.206e+03 6.300e+03, threshold=4.518e+03, percent-clipped=10.0
+2023-03-02 19:05:12,611 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=207111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:05:39,931 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=207140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:05:50,655 INFO [train.py:968] (0/2) Epoch 5, batch 25250, giga_loss[loss=0.2889, simple_loss=0.3579, pruned_loss=0.1099, over 29019.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3904, pruned_loss=0.1399, over 5679038.88 frames. ], libri_tot_loss[loss=0.3398, simple_loss=0.3947, pruned_loss=0.1425, over 5724563.90 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3895, pruned_loss=0.1388, over 5679252.18 frames. ], batch size: 136, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:06:36,698 INFO [train.py:968] (0/2) Epoch 5, batch 25300, giga_loss[loss=0.2924, simple_loss=0.3629, pruned_loss=0.111, over 28837.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.391, pruned_loss=0.1409, over 5675758.15 frames. ], libri_tot_loss[loss=0.3406, simple_loss=0.3953, pruned_loss=0.1429, over 5716880.84 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3897, pruned_loss=0.1396, over 5682157.95 frames. ], batch size: 186, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:06:47,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.759e+02 1.800e+03 2.442e+03 3.465e+03 7.359e+03, threshold=4.884e+03, percent-clipped=8.0
+2023-03-02 19:07:02,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=207227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:07:24,894 INFO [train.py:968] (0/2) Epoch 5, batch 25350, giga_loss[loss=0.3148, simple_loss=0.3767, pruned_loss=0.1265, over 28497.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3918, pruned_loss=0.1416, over 5665563.63 frames. ], libri_tot_loss[loss=0.3414, simple_loss=0.3959, pruned_loss=0.1435, over 5707266.48 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3902, pruned_loss=0.1401, over 5678197.43 frames. ], batch size: 71, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:08:14,149 INFO [train.py:968] (0/2) Epoch 5, batch 25400, giga_loss[loss=0.3539, simple_loss=0.4063, pruned_loss=0.1508, over 28010.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.393, pruned_loss=0.1414, over 5673024.90 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3967, pruned_loss=0.1443, over 5709856.80 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3908, pruned_loss=0.1395, over 5680076.01 frames. ], batch size: 412, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:08:22,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.675e+02 1.627e+03 2.283e+03 3.781e+03 1.153e+04, threshold=4.566e+03, percent-clipped=10.0
+2023-03-02 19:08:26,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0174, 1.9338, 1.9036, 1.7676], device='cuda:0'), covar=tensor([0.0855, 0.1388, 0.1208, 0.1271], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0749, 0.0644, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 19:08:59,468 INFO [train.py:968] (0/2) Epoch 5, batch 25450, giga_loss[loss=0.334, simple_loss=0.3953, pruned_loss=0.1363, over 28886.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3936, pruned_loss=0.1414, over 5678708.58 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3969, pruned_loss=0.1445, over 5713978.50 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3917, pruned_loss=0.1395, over 5680120.09 frames. ], batch size: 186, lr: 6.30e-03, grad_scale: 2.0
+2023-03-02 19:09:45,719 INFO [train.py:968] (0/2) Epoch 5, batch 25500, giga_loss[loss=0.3884, simple_loss=0.4211, pruned_loss=0.1778, over 26475.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3943, pruned_loss=0.1418, over 5672612.06 frames. ], libri_tot_loss[loss=0.3431, simple_loss=0.3967, pruned_loss=0.1447, over 5708571.59 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3927, pruned_loss=0.14, over 5677145.11 frames. ], batch size: 555, lr: 6.30e-03, grad_scale: 2.0
+2023-03-02 19:09:49,201 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0047, 1.3532, 3.8162, 3.1779], device='cuda:0'), covar=tensor([0.1646, 0.2031, 0.0386, 0.0667], device='cuda:0'), in_proj_covar=tensor([0.0572, 0.0526, 0.0754, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 19:09:52,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.459e+02 1.475e+03 1.851e+03 2.783e+03 8.107e+03, threshold=3.702e+03, percent-clipped=7.0
+2023-03-02 19:10:31,684 INFO [train.py:968] (0/2) Epoch 5, batch 25550, giga_loss[loss=0.3104, simple_loss=0.3685, pruned_loss=0.1261, over 28259.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3943, pruned_loss=0.1419, over 5675148.57 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3965, pruned_loss=0.1445, over 5709043.07 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3932, pruned_loss=0.1405, over 5677823.42 frames. ], batch size: 77, lr: 6.30e-03, grad_scale: 2.0
+2023-03-02 19:11:10,750 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 19:11:17,499 INFO [train.py:968] (0/2) Epoch 5, batch 25600, giga_loss[loss=0.4715, simple_loss=0.4668, pruned_loss=0.2381, over 26562.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3965, pruned_loss=0.1441, over 5652986.82 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.3969, pruned_loss=0.1448, over 5682651.40 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3951, pruned_loss=0.1427, over 5679228.63 frames. ], batch size: 555, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:11:26,483 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.532e+02 1.528e+03 1.995e+03 2.777e+03 9.017e+03, threshold=3.991e+03, percent-clipped=9.0
+2023-03-02 19:11:44,681 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1822, 4.0058, 3.8467, 1.9656], device='cuda:0'), covar=tensor([0.0523, 0.0562, 0.0876, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.0881, 0.0805, 0.0831, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 19:12:03,023 INFO [train.py:968] (0/2) Epoch 5, batch 25650, giga_loss[loss=0.3512, simple_loss=0.4013, pruned_loss=0.1505, over 28922.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3976, pruned_loss=0.1467, over 5647300.20 frames. ], libri_tot_loss[loss=0.343, simple_loss=0.3964, pruned_loss=0.1448, over 5672278.43 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.3969, pruned_loss=0.1456, over 5676388.42 frames. ], batch size: 186, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:12:09,897 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 19:12:24,558 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=207569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:12:54,148 INFO [train.py:968] (0/2) Epoch 5, batch 25700, giga_loss[loss=0.3676, simple_loss=0.4126, pruned_loss=0.1613, over 28692.00 frames. ], tot_loss[loss=0.347, simple_loss=0.3981, pruned_loss=0.1479, over 5658145.01 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.3968, pruned_loss=0.1449, over 5678874.64 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.3973, pruned_loss=0.1469, over 5674904.05 frames. ], batch size: 242, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:12:55,329 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 19:12:56,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=207602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:13:04,285 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.798e+03 2.237e+03 3.659e+03 1.070e+04, threshold=4.475e+03, percent-clipped=19.0
+2023-03-02 19:13:38,203 INFO [train.py:968] (0/2) Epoch 5, batch 25750, giga_loss[loss=0.3241, simple_loss=0.3865, pruned_loss=0.1309, over 28972.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.3968, pruned_loss=0.1467, over 5667149.13 frames. ], libri_tot_loss[loss=0.343, simple_loss=0.3966, pruned_loss=0.1447, over 5679783.93 frames. ], giga_tot_loss[loss=0.3444, simple_loss=0.3963, pruned_loss=0.1462, over 5679276.86 frames. ], batch size: 164, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:14:25,221 INFO [train.py:968] (0/2) Epoch 5, batch 25800, giga_loss[loss=0.3098, simple_loss=0.3753, pruned_loss=0.1221, over 28857.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.3965, pruned_loss=0.1471, over 5653563.51 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.397, pruned_loss=0.1451, over 5674753.88 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3958, pruned_loss=0.1464, over 5667125.35 frames. ], batch size: 227, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:14:34,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.703e+02 1.578e+03 2.005e+03 2.477e+03 7.266e+03, threshold=4.011e+03, percent-clipped=3.0
+2023-03-02 19:15:06,236 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=207745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:15:08,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=207748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:15:09,614 INFO [train.py:968] (0/2) Epoch 5, batch 25850, giga_loss[loss=0.2928, simple_loss=0.3734, pruned_loss=0.1061, over 29066.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3962, pruned_loss=0.1448, over 5666963.75 frames. ], libri_tot_loss[loss=0.3434, simple_loss=0.397, pruned_loss=0.1449, over 5678236.17 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3956, pruned_loss=0.1444, over 5674433.56 frames. ], batch size: 155, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:15:33,584 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=207777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:15:56,262 INFO [train.py:968] (0/2) Epoch 5, batch 25900, giga_loss[loss=0.3057, simple_loss=0.3791, pruned_loss=0.1161, over 29043.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3929, pruned_loss=0.1425, over 5651713.38 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3972, pruned_loss=0.1451, over 5676080.12 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3922, pruned_loss=0.142, over 5658963.68 frames. ], batch size: 136, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:15:58,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0452, 1.1714, 0.9939, 0.6026], device='cuda:0'), covar=tensor([0.0832, 0.0868, 0.0605, 0.0947], device='cuda:0'), in_proj_covar=tensor([0.1420, 0.1212, 0.1200, 0.1284], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 19:16:04,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.774e+02 1.527e+03 1.974e+03 2.916e+03 6.882e+03, threshold=3.947e+03, percent-clipped=7.0
+2023-03-02 19:16:10,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0868, 1.2776, 1.2071, 1.2414], device='cuda:0'), covar=tensor([0.0978, 0.1017, 0.1562, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0747, 0.0639, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 19:16:39,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6606, 1.7975, 1.5648, 1.6029], device='cuda:0'), covar=tensor([0.1117, 0.1620, 0.1693, 0.1448], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0752, 0.0643, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 19:16:43,209 INFO [train.py:968] (0/2) Epoch 5, batch 25950, libri_loss[loss=0.3813, simple_loss=0.4002, pruned_loss=0.1812, over 28585.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3919, pruned_loss=0.1426, over 5657692.62 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.3971, pruned_loss=0.145, over 5680504.95 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3914, pruned_loss=0.1422, over 5658902.99 frames. ], batch size: 63, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:17:26,835 INFO [train.py:968] (0/2) Epoch 5, batch 26000, giga_loss[loss=0.4826, simple_loss=0.4744, pruned_loss=0.2454, over 26623.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3919, pruned_loss=0.1436, over 5660670.42 frames. ], libri_tot_loss[loss=0.3435, simple_loss=0.3969, pruned_loss=0.1451, over 5685462.86 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3914, pruned_loss=0.1432, over 5656444.79 frames. ], batch size: 555, lr: 6.29e-03, grad_scale: 8.0
+2023-03-02 19:17:37,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.836e+03 2.496e+03 4.346e+03 8.732e+03, threshold=4.992e+03, percent-clipped=26.0
+2023-03-02 19:17:53,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0563, 5.1140, 2.1012, 2.0195], device='cuda:0'), covar=tensor([0.0706, 0.0248, 0.0757, 0.1115], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0478, 0.0309, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 19:18:11,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=207944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:18:15,049 INFO [train.py:968] (0/2) Epoch 5, batch 26050, giga_loss[loss=0.3628, simple_loss=0.4168, pruned_loss=0.1544, over 28843.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3936, pruned_loss=0.1454, over 5650287.46 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3972, pruned_loss=0.1454, over 5678232.03 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3929, pruned_loss=0.1447, over 5653126.65 frames. ], batch size: 186, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:19:01,767 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-208000.pt
+2023-03-02 19:19:02,088 INFO [train.py:968] (0/2) Epoch 5, batch 26100, giga_loss[loss=0.3764, simple_loss=0.4225, pruned_loss=0.1652, over 27570.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3956, pruned_loss=0.1452, over 5650522.77 frames. ], libri_tot_loss[loss=0.3442, simple_loss=0.3973, pruned_loss=0.1455, over 5672527.10 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3949, pruned_loss=0.1446, over 5656681.38 frames. ], batch size: 472, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:19:13,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.572e+02 1.433e+03 1.722e+03 2.644e+03 7.587e+03, threshold=3.444e+03, percent-clipped=3.0
+2023-03-02 19:19:18,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2354, 1.7107, 1.2436, 1.5333], device='cuda:0'), covar=tensor([0.0772, 0.0323, 0.0347, 0.0808], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0123, 0.0127, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:0')
+2023-03-02 19:19:51,851 INFO [train.py:968] (0/2) Epoch 5, batch 26150, giga_loss[loss=0.3083, simple_loss=0.3859, pruned_loss=0.1153, over 28927.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3988, pruned_loss=0.1442, over 5658880.99 frames. ], libri_tot_loss[loss=0.3445, simple_loss=0.3976, pruned_loss=0.1457, over 5676777.74 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.398, pruned_loss=0.1435, over 5659817.39 frames. ], batch size: 199, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:20:07,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3473, 1.5710, 1.3048, 1.3878], device='cuda:0'), covar=tensor([0.0765, 0.0318, 0.0331, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0124, 0.0128, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0056, 0.0042, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 19:20:21,792 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6074, 2.2116, 1.5067, 0.8055], device='cuda:0'), covar=tensor([0.2497, 0.1418, 0.2143, 0.2759], device='cuda:0'), in_proj_covar=tensor([0.1369, 0.1284, 0.1365, 0.1140], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 19:20:30,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:20:32,876 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=208090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:20:41,038 INFO [train.py:968] (0/2) Epoch 5, batch 26200, giga_loss[loss=0.3977, simple_loss=0.4388, pruned_loss=0.1783, over 28967.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3997, pruned_loss=0.1444, over 5659586.73 frames. ], libri_tot_loss[loss=0.3444, simple_loss=0.3976, pruned_loss=0.1456, over 5680298.07 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3991, pruned_loss=0.1439, over 5656842.25 frames. ], batch size: 213, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:20:52,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.431e+02 1.397e+03 1.813e+03 2.237e+03 5.329e+03, threshold=3.626e+03, percent-clipped=8.0
+2023-03-02 19:21:00,840 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:21:31,310 INFO [train.py:968] (0/2) Epoch 5, batch 26250, giga_loss[loss=0.436, simple_loss=0.4524, pruned_loss=0.2098, over 26625.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.4019, pruned_loss=0.1467, over 5644288.56 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.3977, pruned_loss=0.1458, over 5673406.67 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.4013, pruned_loss=0.1462, over 5648736.33 frames. ], batch size: 555, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:22:15,031 INFO [train.py:968] (0/2) Epoch 5, batch 26300, libri_loss[loss=0.3608, simple_loss=0.4158, pruned_loss=0.1529, over 28773.00 frames. ], tot_loss[loss=0.3493, simple_loss=0.4027, pruned_loss=0.1479, over 5647252.17 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.3981, pruned_loss=0.146, over 5677732.62 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.4021, pruned_loss=0.1474, over 5646217.01 frames. ], batch size: 106, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:22:26,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.683e+02 1.623e+03 2.283e+03 2.975e+03 7.144e+03, threshold=4.566e+03, percent-clipped=13.0
+2023-03-02 19:23:03,504 INFO [train.py:968] (0/2) Epoch 5, batch 26350, giga_loss[loss=0.3407, simple_loss=0.3928, pruned_loss=0.1443, over 28926.00 frames. ], tot_loss[loss=0.3491, simple_loss=0.4018, pruned_loss=0.1483, over 5648512.94 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3978, pruned_loss=0.1458, over 5680069.86 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4016, pruned_loss=0.148, over 5645283.43 frames. ], batch size: 213, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:23:53,030 INFO [train.py:968] (0/2) Epoch 5, batch 26400, libri_loss[loss=0.3894, simple_loss=0.4279, pruned_loss=0.1755, over 29531.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4, pruned_loss=0.1475, over 5652391.66 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.3982, pruned_loss=0.1461, over 5684562.25 frames. ], giga_tot_loss[loss=0.3467, simple_loss=0.3995, pruned_loss=0.147, over 5644869.00 frames. ], batch size: 80, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:24:04,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.703e+02 1.546e+03 2.294e+03 3.071e+03 1.129e+04, threshold=4.587e+03, percent-clipped=10.0
+2023-03-02 19:24:42,084 INFO [train.py:968] (0/2) Epoch 5, batch 26450, giga_loss[loss=0.3504, simple_loss=0.3809, pruned_loss=0.16, over 23355.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3979, pruned_loss=0.1468, over 5656539.83 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3981, pruned_loss=0.1461, over 5687467.65 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3976, pruned_loss=0.1465, over 5647404.84 frames. ], batch size: 705, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:25:34,629 INFO [train.py:968] (0/2) Epoch 5, batch 26500, giga_loss[loss=0.3616, simple_loss=0.4115, pruned_loss=0.1558, over 29032.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3975, pruned_loss=0.1471, over 5647242.28 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3983, pruned_loss=0.1461, over 5687262.86 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.3971, pruned_loss=0.1468, over 5639511.10 frames. ], batch size: 136, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:25:47,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.088e+03 1.489e+03 1.823e+03 2.533e+03 6.128e+03, threshold=3.645e+03, percent-clipped=4.0
+2023-03-02 19:25:58,230 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=208425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:26:18,677 INFO [train.py:968] (0/2) Epoch 5, batch 26550, giga_loss[loss=0.3699, simple_loss=0.4235, pruned_loss=0.1581, over 28822.00 frames. ], tot_loss[loss=0.3484, simple_loss=0.3988, pruned_loss=0.1489, over 5645148.89 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3977, pruned_loss=0.1458, over 5683684.54 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.399, pruned_loss=0.149, over 5641233.39 frames. ], batch size: 174, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:26:35,557 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4317, 1.8340, 1.2458, 1.6212], device='cuda:0'), covar=tensor([0.0731, 0.0268, 0.0342, 0.0779], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0124, 0.0127, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0057, 0.0042, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 19:27:03,467 INFO [train.py:968] (0/2) Epoch 5, batch 26600, giga_loss[loss=0.3088, simple_loss=0.365, pruned_loss=0.1263, over 28968.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3963, pruned_loss=0.1471, over 5663707.62 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3981, pruned_loss=0.1461, over 5687266.96 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.3962, pruned_loss=0.1469, over 5656697.55 frames. ], batch size: 106, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:27:14,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.826e+02 1.586e+03 2.199e+03 3.112e+03 9.068e+03, threshold=4.398e+03, percent-clipped=18.0
+2023-03-02 19:27:50,905 INFO [train.py:968] (0/2) Epoch 5, batch 26650, giga_loss[loss=0.3005, simple_loss=0.361, pruned_loss=0.12, over 28531.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3949, pruned_loss=0.1463, over 5663791.73 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3982, pruned_loss=0.1464, over 5680954.03 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3946, pruned_loss=0.1459, over 5664078.93 frames. ], batch size: 78, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:28:38,483 INFO [train.py:968] (0/2) Epoch 5, batch 26700, libri_loss[loss=0.371, simple_loss=0.4185, pruned_loss=0.1617, over 29716.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3956, pruned_loss=0.1459, over 5668591.08 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3985, pruned_loss=0.1466, over 5683392.10 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3949, pruned_loss=0.1454, over 5666120.08 frames. ], batch size: 87, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:28:48,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.765e+03 2.391e+03 3.229e+03 8.273e+03, threshold=4.782e+03, percent-clipped=11.0
+2023-03-02 19:29:21,804 INFO [train.py:968] (0/2) Epoch 5, batch 26750, giga_loss[loss=0.3262, simple_loss=0.3921, pruned_loss=0.1301, over 28883.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.398, pruned_loss=0.1468, over 5659457.29 frames. ], libri_tot_loss[loss=0.346, simple_loss=0.3986, pruned_loss=0.1467, over 5676232.65 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.3974, pruned_loss=0.1463, over 5663233.82 frames. ], batch size: 186, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:29:26,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6069, 1.8666, 1.3678, 1.0660], device='cuda:0'), covar=tensor([0.1197, 0.0935, 0.0812, 0.1106], device='cuda:0'), in_proj_covar=tensor([0.1433, 0.1226, 0.1209, 0.1313], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 19:29:30,454 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=208658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:29:59,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6572, 2.0008, 1.9936, 1.8165], device='cuda:0'), covar=tensor([0.1542, 0.1725, 0.1119, 0.1199], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0752, 0.0785, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 19:30:11,921 INFO [train.py:968] (0/2) Epoch 5, batch 26800, giga_loss[loss=0.3507, simple_loss=0.4064, pruned_loss=0.1475, over 28688.00 frames. ], tot_loss[loss=0.3452, simple_loss=0.3969, pruned_loss=0.1467, over 5653967.96 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.398, pruned_loss=0.1464, over 5679576.76 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3969, pruned_loss=0.1467, over 5653686.34 frames. ], batch size: 262, lr: 6.28e-03, grad_scale: 8.0
+2023-03-02 19:30:23,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.761e+02 1.538e+03 1.949e+03 2.457e+03 4.511e+03, threshold=3.897e+03, percent-clipped=0.0
+2023-03-02 19:30:44,485 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3654, 1.9349, 1.3798, 0.5349], device='cuda:0'), covar=tensor([0.1634, 0.0942, 0.1404, 0.1878], device='cuda:0'), in_proj_covar=tensor([0.1403, 0.1311, 0.1385, 0.1159], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 19:30:54,943 INFO [train.py:968] (0/2) Epoch 5, batch 26850, giga_loss[loss=0.3567, simple_loss=0.4264, pruned_loss=0.1435, over 28627.00 frames. ], tot_loss[loss=0.3456, simple_loss=0.3991, pruned_loss=0.146, over 5667221.25 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.398, pruned_loss=0.1463, over 5682552.70 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.3992, pruned_loss=0.146, over 5663675.39 frames. ], batch size: 85, lr: 6.28e-03, grad_scale: 8.0
+2023-03-02 19:31:45,108 INFO [train.py:968] (0/2) Epoch 5, batch 26900, giga_loss[loss=0.3003, simple_loss=0.3818, pruned_loss=0.1094, over 29094.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3997, pruned_loss=0.1444, over 5671308.57 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3981, pruned_loss=0.1464, over 5684433.59 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.3996, pruned_loss=0.1443, over 5666762.33 frames. ], batch size: 128, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:31:46,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=208800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:31:57,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.922e+02 1.329e+03 1.718e+03 2.320e+03 7.846e+03, threshold=3.437e+03, percent-clipped=5.0
+2023-03-02 19:32:09,452 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 19:32:31,114 INFO [train.py:968] (0/2) Epoch 5, batch 26950, giga_loss[loss=0.3526, simple_loss=0.4175, pruned_loss=0.1439, over 28589.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.4006, pruned_loss=0.1433, over 5670645.59 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3975, pruned_loss=0.146, over 5687630.99 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.4012, pruned_loss=0.1435, over 5663974.65 frames. ], batch size: 92, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:32:40,869 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-02 19:33:17,774 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-02 19:33:20,911 INFO [train.py:968] (0/2) Epoch 5, batch 27000, giga_loss[loss=0.39, simple_loss=0.4296, pruned_loss=0.1752, over 28209.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.4053, pruned_loss=0.1481, over 5670189.07 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3977, pruned_loss=0.1462, over 5686165.93 frames. ], giga_tot_loss[loss=0.3509, simple_loss=0.4056, pruned_loss=0.1481, over 5665843.02 frames. ], batch size: 368, lr: 6.28e-03, grad_scale: 2.0
+2023-03-02 19:33:20,916 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 19:33:26,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2370, 1.5345, 1.5528, 1.4607], device='cuda:0'), covar=tensor([0.1184, 0.1250, 0.1570, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0745, 0.0635, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 19:33:28,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2417, 1.8188, 1.3861, 0.4072], device='cuda:0'), covar=tensor([0.1762, 0.1122, 0.2035, 0.2260], device='cuda:0'), in_proj_covar=tensor([0.1405, 0.1312, 0.1378, 0.1158], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 19:33:29,451 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2372, simple_loss=0.3405, pruned_loss=0.06699, over 944034.00 frames. 
+2023-03-02 19:33:29,452 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 19:33:41,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.284e+02 1.511e+03 2.122e+03 2.915e+03 5.995e+03, threshold=4.244e+03, percent-clipped=16.0
+2023-03-02 19:34:04,311 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-02 19:34:07,019 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:34:10,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=208946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:34:14,832 INFO [train.py:968] (0/2) Epoch 5, batch 27050, libri_loss[loss=0.3288, simple_loss=0.3903, pruned_loss=0.1337, over 27818.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.4063, pruned_loss=0.1497, over 5681965.55 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3977, pruned_loss=0.1463, over 5692407.16 frames. ], giga_tot_loss[loss=0.3532, simple_loss=0.4069, pruned_loss=0.1498, over 5672253.74 frames. ], batch size: 116, lr: 6.28e-03, grad_scale: 2.0
+2023-03-02 19:34:34,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=208969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:34:41,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:35:01,763 INFO [train.py:968] (0/2) Epoch 5, batch 27100, giga_loss[loss=0.3632, simple_loss=0.4169, pruned_loss=0.1547, over 28229.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4065, pruned_loss=0.1508, over 5686165.48 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.3975, pruned_loss=0.1462, over 5697001.00 frames. ], giga_tot_loss[loss=0.3548, simple_loss=0.4074, pruned_loss=0.1511, over 5673896.58 frames. ], batch size: 368, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:35:16,305 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.105e+02 1.909e+03 2.819e+03 4.384e+03 8.425e+03, threshold=5.639e+03, percent-clipped=28.0
+2023-03-02 19:35:35,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:35:54,759 INFO [train.py:968] (0/2) Epoch 5, batch 27150, giga_loss[loss=0.3131, simple_loss=0.3793, pruned_loss=0.1234, over 28803.00 frames. ], tot_loss[loss=0.3532, simple_loss=0.4056, pruned_loss=0.1504, over 5680710.45 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3978, pruned_loss=0.1466, over 5701175.65 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.4062, pruned_loss=0.1504, over 5667011.77 frames. ], batch size: 186, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:36:10,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2015, 3.0341, 2.8915, 1.5451], device='cuda:0'), covar=tensor([0.0868, 0.0839, 0.1147, 0.2170], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0810, 0.0821, 0.0605], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 19:36:40,297 INFO [train.py:968] (0/2) Epoch 5, batch 27200, giga_loss[loss=0.3363, simple_loss=0.4041, pruned_loss=0.1342, over 28939.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.4047, pruned_loss=0.1485, over 5672871.44 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.398, pruned_loss=0.1471, over 5695137.57 frames. ], giga_tot_loss[loss=0.3507, simple_loss=0.4052, pruned_loss=0.1481, over 5666641.21 frames. ], batch size: 145, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:36:51,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.040e+02 1.487e+03 1.926e+03 2.468e+03 6.280e+03, threshold=3.852e+03, percent-clipped=1.0
+2023-03-02 19:37:26,818 INFO [train.py:968] (0/2) Epoch 5, batch 27250, giga_loss[loss=0.3321, simple_loss=0.3958, pruned_loss=0.1342, over 28823.00 frames. ], tot_loss[loss=0.351, simple_loss=0.4052, pruned_loss=0.1484, over 5662539.43 frames. ], libri_tot_loss[loss=0.3458, simple_loss=0.3978, pruned_loss=0.1469, over 5697571.83 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.406, pruned_loss=0.1482, over 5654611.35 frames. ], batch size: 119, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:37:51,773 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:37:54,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:38:13,021 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4906, 1.7029, 1.3726, 1.4778], device='cuda:0'), covar=tensor([0.0750, 0.0298, 0.0316, 0.0824], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0125, 0.0128, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0057, 0.0042, 0.0038, 0.0063], device='cuda:0')
+2023-03-02 19:38:13,460 INFO [train.py:968] (0/2) Epoch 5, batch 27300, giga_loss[loss=0.3265, simple_loss=0.3873, pruned_loss=0.1328, over 28832.00 frames. ], tot_loss[loss=0.3487, simple_loss=0.404, pruned_loss=0.1467, over 5667989.09 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3973, pruned_loss=0.1468, over 5693652.15 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.4052, pruned_loss=0.1467, over 5664568.68 frames. ], batch size: 99, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:38:22,861 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:38:28,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.711e+02 1.495e+03 1.926e+03 2.805e+03 8.664e+03, threshold=3.853e+03, percent-clipped=11.0
+2023-03-02 19:39:02,505 INFO [train.py:968] (0/2) Epoch 5, batch 27350, giga_loss[loss=0.347, simple_loss=0.3981, pruned_loss=0.1479, over 28954.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4044, pruned_loss=0.1478, over 5657444.83 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3973, pruned_loss=0.1466, over 5696056.19 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4056, pruned_loss=0.148, over 5651760.90 frames. ], batch size: 186, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:39:47,759 INFO [train.py:968] (0/2) Epoch 5, batch 27400, giga_loss[loss=0.383, simple_loss=0.4094, pruned_loss=0.1783, over 23309.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4034, pruned_loss=0.1478, over 5658528.72 frames. ], libri_tot_loss[loss=0.3458, simple_loss=0.3977, pruned_loss=0.1469, over 5700178.94 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4042, pruned_loss=0.1477, over 5649461.42 frames. ], batch size: 705, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:40:05,313 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.866e+02 1.572e+03 1.974e+03 2.852e+03 9.801e+03, threshold=3.947e+03, percent-clipped=11.0
+2023-03-02 19:40:20,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5394, 2.1043, 2.0621, 1.8618], device='cuda:0'), covar=tensor([0.1005, 0.1956, 0.1527, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0744, 0.0643, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 19:40:31,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:40:36,425 INFO [train.py:968] (0/2) Epoch 5, batch 27450, giga_loss[loss=0.3708, simple_loss=0.4126, pruned_loss=0.1645, over 28256.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.4011, pruned_loss=0.1467, over 5671432.29 frames. ], libri_tot_loss[loss=0.3457, simple_loss=0.3976, pruned_loss=0.1469, over 5695958.59 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.402, pruned_loss=0.1466, over 5666408.67 frames. ], batch size: 368, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:40:56,201 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4974, 1.9490, 1.3292, 0.7979], device='cuda:0'), covar=tensor([0.2894, 0.1713, 0.1566, 0.2860], device='cuda:0'), in_proj_covar=tensor([0.1392, 0.1309, 0.1374, 0.1165], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 19:41:26,113 INFO [train.py:968] (0/2) Epoch 5, batch 27500, libri_loss[loss=0.3727, simple_loss=0.4176, pruned_loss=0.1638, over 29476.00 frames. ], tot_loss[loss=0.3457, simple_loss=0.3993, pruned_loss=0.146, over 5667191.35 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3973, pruned_loss=0.1468, over 5691962.00 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.4003, pruned_loss=0.146, over 5665177.73 frames. ], batch size: 85, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:41:41,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.710e+02 1.732e+03 2.095e+03 2.587e+03 4.896e+03, threshold=4.189e+03, percent-clipped=7.0
+2023-03-02 19:41:52,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5571, 1.4412, 1.0782, 1.1824], device='cuda:0'), covar=tensor([0.0592, 0.0542, 0.0978, 0.0944], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0458, 0.0504, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 19:41:59,093 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 19:42:14,671 INFO [train.py:968] (0/2) Epoch 5, batch 27550, libri_loss[loss=0.3515, simple_loss=0.4086, pruned_loss=0.1472, over 29148.00 frames. ], tot_loss[loss=0.3447, simple_loss=0.3978, pruned_loss=0.1458, over 5666922.35 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3972, pruned_loss=0.1469, over 5697167.35 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3987, pruned_loss=0.1457, over 5660090.56 frames. ], batch size: 101, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:42:35,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5454, 3.1850, 1.5789, 1.5029], device='cuda:0'), covar=tensor([0.0846, 0.0313, 0.0779, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0489, 0.0311, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 19:42:50,194 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:42:53,580 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:43:00,261 INFO [train.py:968] (0/2) Epoch 5, batch 27600, giga_loss[loss=0.3271, simple_loss=0.3852, pruned_loss=0.1345, over 28835.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3975, pruned_loss=0.1462, over 5672127.32 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3972, pruned_loss=0.1468, over 5701528.99 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3982, pruned_loss=0.1461, over 5662056.84 frames. ], batch size: 213, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:43:15,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.956e+02 1.659e+03 2.062e+03 3.175e+03 1.113e+04, threshold=4.124e+03, percent-clipped=17.0
+2023-03-02 19:43:18,501 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:43:37,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2522, 1.5101, 1.3584, 1.4472], device='cuda:0'), covar=tensor([0.0686, 0.0436, 0.0319, 0.0730], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0125, 0.0128, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0057, 0.0042, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 19:43:44,549 INFO [train.py:968] (0/2) Epoch 5, batch 27650, giga_loss[loss=0.3583, simple_loss=0.4126, pruned_loss=0.152, over 27905.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3965, pruned_loss=0.1452, over 5663677.76 frames. ], libri_tot_loss[loss=0.3462, simple_loss=0.3977, pruned_loss=0.1473, over 5694963.58 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3966, pruned_loss=0.1447, over 5660267.08 frames. ], batch size: 412, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:44:25,075 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:44:28,299 INFO [train.py:968] (0/2) Epoch 5, batch 27700, giga_loss[loss=0.2866, simple_loss=0.3629, pruned_loss=0.1052, over 29014.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3931, pruned_loss=0.1412, over 5662001.11 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.398, pruned_loss=0.1477, over 5693337.91 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3928, pruned_loss=0.1402, over 5659161.32 frames. ], batch size: 155, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:44:38,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.190e+02 1.414e+03 1.828e+03 2.599e+03 7.205e+03, threshold=3.655e+03, percent-clipped=10.0
+2023-03-02 19:44:49,411 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9959, 3.8820, 3.6776, 1.9794], device='cuda:0'), covar=tensor([0.0438, 0.0474, 0.0656, 0.1964], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0810, 0.0818, 0.0607], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 19:45:09,941 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:45:11,109 INFO [train.py:968] (0/2) Epoch 5, batch 27750, libri_loss[loss=0.339, simple_loss=0.3973, pruned_loss=0.1404, over 27908.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3917, pruned_loss=0.1399, over 5665897.78 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.3968, pruned_loss=0.1469, over 5699090.66 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3922, pruned_loss=0.1395, over 5657265.89 frames. ], batch size: 116, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:45:54,582 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:46:02,155 INFO [train.py:968] (0/2) Epoch 5, batch 27800, giga_loss[loss=0.3298, simple_loss=0.3859, pruned_loss=0.1369, over 28695.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3908, pruned_loss=0.1392, over 5662221.18 frames. ], libri_tot_loss[loss=0.3456, simple_loss=0.3971, pruned_loss=0.1471, over 5699419.57 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3909, pruned_loss=0.1387, over 5654449.11 frames. ], batch size: 262, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:46:20,523 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.470e+02 1.559e+03 2.059e+03 2.647e+03 9.635e+03, threshold=4.117e+03, percent-clipped=13.0
+2023-03-02 19:46:23,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3874, 1.8553, 1.7289, 1.5579], device='cuda:0'), covar=tensor([0.1388, 0.1791, 0.1054, 0.1140], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0757, 0.0780, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-02 19:46:59,297 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-02 19:47:00,386 INFO [train.py:968] (0/2) Epoch 5, batch 27850, giga_loss[loss=0.3051, simple_loss=0.3703, pruned_loss=0.12, over 28814.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3864, pruned_loss=0.1375, over 5652337.74 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3969, pruned_loss=0.1469, over 5699536.47 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3866, pruned_loss=0.1371, over 5645991.63 frames. ], batch size: 174, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:47:50,428 INFO [train.py:968] (0/2) Epoch 5, batch 27900, giga_loss[loss=0.3435, simple_loss=0.3969, pruned_loss=0.1451, over 27994.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3874, pruned_loss=0.1387, over 5644839.24 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.397, pruned_loss=0.1469, over 5693901.74 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3872, pruned_loss=0.1382, over 5644395.06 frames. ], batch size: 412, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:48:05,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.556e+02 1.691e+03 2.461e+03 3.437e+03 6.570e+03, threshold=4.922e+03, percent-clipped=13.0
+2023-03-02 19:48:06,168 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-02 19:48:37,689 INFO [train.py:968] (0/2) Epoch 5, batch 27950, giga_loss[loss=0.3355, simple_loss=0.3986, pruned_loss=0.1361, over 28940.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3889, pruned_loss=0.1383, over 5662447.87 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3968, pruned_loss=0.1468, over 5698205.32 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3886, pruned_loss=0.1378, over 5656800.80 frames. ], batch size: 136, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:48:40,470 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 19:48:46,488 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:49:14,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3489, 1.4351, 1.2363, 1.5676], device='cuda:0'), covar=tensor([0.2160, 0.2004, 0.2017, 0.2037], device='cuda:0'), in_proj_covar=tensor([0.1122, 0.0875, 0.0994, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 19:49:25,254 INFO [train.py:968] (0/2) Epoch 5, batch 28000, giga_loss[loss=0.3107, simple_loss=0.3826, pruned_loss=0.1195, over 28949.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3889, pruned_loss=0.138, over 5648928.58 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3971, pruned_loss=0.1469, over 5693646.20 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3884, pruned_loss=0.1373, over 5647447.49 frames. ], batch size: 164, lr: 6.26e-03, grad_scale: 8.0
+2023-03-02 19:49:28,157 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:49:37,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.532e+02 1.379e+03 1.609e+03 2.240e+03 7.689e+03, threshold=3.218e+03, percent-clipped=4.0
+2023-03-02 19:50:10,084 INFO [train.py:968] (0/2) Epoch 5, batch 28050, giga_loss[loss=0.3426, simple_loss=0.3759, pruned_loss=0.1546, over 23481.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3902, pruned_loss=0.1394, over 5645269.49 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.3982, pruned_loss=0.1477, over 5694634.33 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3884, pruned_loss=0.1378, over 5641898.56 frames. ], batch size: 705, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:50:27,978 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:50:53,859 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:50:54,386 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-210000.pt
+2023-03-02 19:50:54,688 INFO [train.py:968] (0/2) Epoch 5, batch 28100, giga_loss[loss=0.2649, simple_loss=0.336, pruned_loss=0.09687, over 28611.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3911, pruned_loss=0.1408, over 5650731.87 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.3986, pruned_loss=0.148, over 5697893.66 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3892, pruned_loss=0.1391, over 5644153.20 frames. ], batch size: 85, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:51:07,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.898e+02 1.501e+03 2.010e+03 2.951e+03 1.091e+04, threshold=4.020e+03, percent-clipped=19.0
+2023-03-02 19:51:13,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:51:37,999 INFO [train.py:968] (0/2) Epoch 5, batch 28150, giga_loss[loss=0.345, simple_loss=0.4099, pruned_loss=0.14, over 28850.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.393, pruned_loss=0.1424, over 5642859.34 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.3983, pruned_loss=0.1477, over 5694626.12 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3915, pruned_loss=0.1411, over 5639842.86 frames. ], batch size: 174, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:51:54,842 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:52:07,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=210081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:52:25,965 INFO [train.py:968] (0/2) Epoch 5, batch 28200, giga_loss[loss=0.3483, simple_loss=0.3971, pruned_loss=0.1498, over 28939.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3947, pruned_loss=0.143, over 5653107.64 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.3983, pruned_loss=0.1476, over 5695797.06 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3935, pruned_loss=0.142, over 5649151.95 frames. ], batch size: 199, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:52:40,167 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:52:42,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.270e+02 1.520e+03 1.861e+03 2.763e+03 7.625e+03, threshold=3.721e+03, percent-clipped=11.0
+2023-03-02 19:52:42,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:53:01,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2991, 1.5042, 1.2806, 1.4799], device='cuda:0'), covar=tensor([0.1796, 0.1647, 0.1597, 0.1657], device='cuda:0'), in_proj_covar=tensor([0.1133, 0.0882, 0.1006, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 19:53:15,186 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:53:19,558 INFO [train.py:968] (0/2) Epoch 5, batch 28250, giga_loss[loss=0.3209, simple_loss=0.3913, pruned_loss=0.1252, over 28864.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3972, pruned_loss=0.145, over 5651920.23 frames. ], libri_tot_loss[loss=0.3472, simple_loss=0.3987, pruned_loss=0.1478, over 5696984.68 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3959, pruned_loss=0.144, over 5647373.67 frames. ], batch size: 174, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:53:35,878 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:53:38,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:54:08,602 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:54:12,060 INFO [train.py:968] (0/2) Epoch 5, batch 28300, giga_loss[loss=0.4159, simple_loss=0.4451, pruned_loss=0.1933, over 27542.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3989, pruned_loss=0.1472, over 5647444.51 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.3985, pruned_loss=0.1476, over 5699841.19 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.398, pruned_loss=0.1466, over 5640863.71 frames. ], batch size: 472, lr: 6.26e-03, grad_scale: 2.0
+2023-03-02 19:54:22,687 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:54:25,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:54:31,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.980e+02 1.699e+03 2.253e+03 3.468e+03 1.272e+04, threshold=4.507e+03, percent-clipped=22.0
+2023-03-02 19:54:49,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:54:54,058 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:55:05,837 INFO [train.py:968] (0/2) Epoch 5, batch 28350, giga_loss[loss=0.3553, simple_loss=0.4157, pruned_loss=0.1475, over 28831.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3996, pruned_loss=0.1463, over 5650435.48 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.3985, pruned_loss=0.1475, over 5702978.27 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3989, pruned_loss=0.1459, over 5641200.43 frames. ], batch size: 284, lr: 6.26e-03, grad_scale: 2.0
+2023-03-02 19:55:14,310 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 19:55:28,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:55:48,959 INFO [train.py:968] (0/2) Epoch 5, batch 28400, libri_loss[loss=0.3639, simple_loss=0.4119, pruned_loss=0.158, over 29644.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3997, pruned_loss=0.1463, over 5657498.59 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.3987, pruned_loss=0.1479, over 5703096.01 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.399, pruned_loss=0.1455, over 5647357.63 frames. ], batch size: 88, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:56:09,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.869e+02 1.314e+03 1.673e+03 2.130e+03 1.114e+04, threshold=3.345e+03, percent-clipped=2.0
+2023-03-02 19:56:40,126 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-02 19:56:41,023 INFO [train.py:968] (0/2) Epoch 5, batch 28450, giga_loss[loss=0.401, simple_loss=0.4463, pruned_loss=0.1778, over 28713.00 frames. ], tot_loss[loss=0.3483, simple_loss=0.4005, pruned_loss=0.148, over 5639350.55 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.3988, pruned_loss=0.1481, over 5704773.20 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.3999, pruned_loss=0.1472, over 5628707.09 frames. ], batch size: 284, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 19:57:04,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:57:09,775 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:57:14,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:57:34,015 INFO [train.py:968] (0/2) Epoch 5, batch 28500, libri_loss[loss=0.3046, simple_loss=0.3547, pruned_loss=0.1273, over 27233.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.3996, pruned_loss=0.148, over 5643359.66 frames. ], libri_tot_loss[loss=0.3466, simple_loss=0.398, pruned_loss=0.1476, over 5709570.14 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.3998, pruned_loss=0.1478, over 5628557.84 frames. ], batch size: 60, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 19:57:44,848 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:57:53,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.585e+02 1.880e+03 2.311e+03 3.078e+03 7.338e+03, threshold=4.622e+03, percent-clipped=21.0
+2023-03-02 19:57:59,071 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:58:01,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:58:03,068 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-02 19:58:10,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3013, 1.5178, 1.2813, 1.7191], device='cuda:0'), covar=tensor([0.2291, 0.2149, 0.2196, 0.2030], device='cuda:0'), in_proj_covar=tensor([0.1127, 0.0880, 0.1002, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 19:58:29,034 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-02 19:58:32,668 INFO [train.py:968] (0/2) Epoch 5, batch 28550, giga_loss[loss=0.3143, simple_loss=0.3767, pruned_loss=0.126, over 28740.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3973, pruned_loss=0.1472, over 5630411.18 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3975, pruned_loss=0.1471, over 5712229.47 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.398, pruned_loss=0.1474, over 5614480.67 frames. ], batch size: 284, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 19:58:32,853 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=210450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:58:34,689 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:58:36,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:59:01,523 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-02 19:59:19,240 INFO [train.py:968] (0/2) Epoch 5, batch 28600, giga_loss[loss=0.2995, simple_loss=0.361, pruned_loss=0.119, over 28849.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3948, pruned_loss=0.1456, over 5646907.99 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3971, pruned_loss=0.1468, over 5714374.74 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.3957, pruned_loss=0.1461, over 5632033.30 frames. ], batch size: 66, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 19:59:37,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.493e+02 1.420e+03 1.804e+03 2.280e+03 4.730e+03, threshold=3.608e+03, percent-clipped=1.0
+2023-03-02 19:59:37,737 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:59:40,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:00:10,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:00:11,209 INFO [train.py:968] (0/2) Epoch 5, batch 28650, giga_loss[loss=0.353, simple_loss=0.4036, pruned_loss=0.1512, over 28995.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.395, pruned_loss=0.1458, over 5654668.08 frames. ], libri_tot_loss[loss=0.3458, simple_loss=0.3974, pruned_loss=0.1471, over 5715453.87 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3953, pruned_loss=0.146, over 5641651.46 frames. ], batch size: 164, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:00:26,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5408, 1.7935, 1.7604, 1.6703], device='cuda:0'), covar=tensor([0.1482, 0.1867, 0.1144, 0.1224], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0747, 0.0775, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-02 20:00:50,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6110, 1.9730, 1.9803, 1.8096], device='cuda:0'), covar=tensor([0.1529, 0.1828, 0.1121, 0.1268], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0750, 0.0777, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-02 20:01:01,451 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:01:01,774 INFO [train.py:968] (0/2) Epoch 5, batch 28700, giga_loss[loss=0.3489, simple_loss=0.3971, pruned_loss=0.1504, over 28611.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3949, pruned_loss=0.1459, over 5658667.29 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3973, pruned_loss=0.1469, over 5717344.68 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3953, pruned_loss=0.1462, over 5645756.58 frames. ], batch size: 307, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:01:04,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:01:18,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.428e+02 1.542e+03 2.084e+03 2.805e+03 8.634e+03, threshold=4.167e+03, percent-clipped=17.0
+2023-03-02 20:01:35,437 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:01:49,738 INFO [train.py:968] (0/2) Epoch 5, batch 28750, giga_loss[loss=0.3277, simple_loss=0.3903, pruned_loss=0.1325, over 28918.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3951, pruned_loss=0.1459, over 5647185.71 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.3977, pruned_loss=0.1473, over 5700277.02 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.395, pruned_loss=0.1457, over 5651826.98 frames. ], batch size: 164, lr: 6.25e-03, grad_scale: 2.0
+2023-03-02 20:02:01,138 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4913, 3.3213, 3.1670, 1.7391], device='cuda:0'), covar=tensor([0.0650, 0.0693, 0.0894, 0.2139], device='cuda:0'), in_proj_covar=tensor([0.0881, 0.0813, 0.0827, 0.0603], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 20:02:38,427 INFO [train.py:968] (0/2) Epoch 5, batch 28800, giga_loss[loss=0.3755, simple_loss=0.4293, pruned_loss=0.1608, over 28568.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3976, pruned_loss=0.1479, over 5652638.98 frames. ], libri_tot_loss[loss=0.3462, simple_loss=0.3976, pruned_loss=0.1473, over 5706302.90 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3975, pruned_loss=0.1477, over 5649543.57 frames. ], batch size: 336, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:02:54,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.824e+02 1.716e+03 2.396e+03 3.260e+03 1.011e+04, threshold=4.791e+03, percent-clipped=14.0
+2023-03-02 20:03:24,981 INFO [train.py:968] (0/2) Epoch 5, batch 28850, giga_loss[loss=0.3623, simple_loss=0.4099, pruned_loss=0.1574, over 28621.00 frames. ], tot_loss[loss=0.3474, simple_loss=0.398, pruned_loss=0.1484, over 5661003.25 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.3977, pruned_loss=0.1475, over 5705825.96 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.3979, pruned_loss=0.1481, over 5657048.65 frames. ], batch size: 262, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:03:53,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7328, 3.5797, 1.6485, 1.6374], device='cuda:0'), covar=tensor([0.0792, 0.0302, 0.0797, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0488, 0.0313, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 20:04:09,656 INFO [train.py:968] (0/2) Epoch 5, batch 28900, giga_loss[loss=0.325, simple_loss=0.386, pruned_loss=0.132, over 29019.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.3984, pruned_loss=0.1487, over 5660333.46 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.3977, pruned_loss=0.1474, over 5702529.05 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3983, pruned_loss=0.1486, over 5659476.70 frames. ], batch size: 155, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:04:16,835 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 20:04:20,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-02 20:04:23,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.559e+03 2.064e+03 2.867e+03 6.069e+03, threshold=4.129e+03, percent-clipped=6.0
+2023-03-02 20:04:29,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7523, 4.6099, 4.4071, 2.2531], device='cuda:0'), covar=tensor([0.0508, 0.0567, 0.0876, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.0882, 0.0819, 0.0829, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 20:04:30,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:04:53,561 INFO [train.py:968] (0/2) Epoch 5, batch 28950, giga_loss[loss=0.3552, simple_loss=0.4085, pruned_loss=0.151, over 28582.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.3982, pruned_loss=0.1477, over 5672691.09 frames. ], libri_tot_loss[loss=0.3464, simple_loss=0.3981, pruned_loss=0.1474, over 5704117.39 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.3977, pruned_loss=0.1477, over 5669448.19 frames. ], batch size: 92, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:05:42,277 INFO [train.py:968] (0/2) Epoch 5, batch 29000, giga_loss[loss=0.3731, simple_loss=0.4162, pruned_loss=0.165, over 27958.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4001, pruned_loss=0.1494, over 5669734.91 frames. ], libri_tot_loss[loss=0.3466, simple_loss=0.3981, pruned_loss=0.1475, over 5707972.94 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.3999, pruned_loss=0.1493, over 5663274.31 frames. ], batch size: 412, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:06:00,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.355e+02 1.596e+03 2.315e+03 3.613e+03 9.003e+03, threshold=4.631e+03, percent-clipped=20.0
+2023-03-02 20:06:30,186 INFO [train.py:968] (0/2) Epoch 5, batch 29050, giga_loss[loss=0.3197, simple_loss=0.3793, pruned_loss=0.13, over 28679.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.4001, pruned_loss=0.1489, over 5676947.18 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.3978, pruned_loss=0.1472, over 5710848.94 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.4001, pruned_loss=0.1491, over 5668874.39 frames. ], batch size: 60, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:06:48,337 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:06:50,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:07:14,439 INFO [train.py:968] (0/2) Epoch 5, batch 29100, giga_loss[loss=0.3797, simple_loss=0.4243, pruned_loss=0.1676, over 27938.00 frames. ], tot_loss[loss=0.3506, simple_loss=0.4012, pruned_loss=0.15, over 5674614.12 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.398, pruned_loss=0.1473, over 5713532.20 frames. ], giga_tot_loss[loss=0.3507, simple_loss=0.4012, pruned_loss=0.1501, over 5664925.85 frames. ], batch size: 412, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:07:14,710 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:07:31,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.539e+02 1.603e+03 2.154e+03 3.193e+03 6.304e+03, threshold=4.308e+03, percent-clipped=2.0
+2023-03-02 20:07:36,930 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-02 20:07:37,797 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:07:48,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6910, 2.1374, 1.6753, 2.0830], device='cuda:0'), covar=tensor([0.0491, 0.0581, 0.0835, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0454, 0.0507, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 20:07:56,552 INFO [train.py:968] (0/2) Epoch 5, batch 29150, giga_loss[loss=0.3138, simple_loss=0.3819, pruned_loss=0.1229, over 28757.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4021, pruned_loss=0.1509, over 5667852.56 frames. ], libri_tot_loss[loss=0.3456, simple_loss=0.3975, pruned_loss=0.1468, over 5710931.14 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4026, pruned_loss=0.1517, over 5661123.34 frames. ], batch size: 92, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:08:04,190 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:08:06,453 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-02 20:08:19,105 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:08:43,617 INFO [train.py:968] (0/2) Epoch 5, batch 29200, giga_loss[loss=0.3924, simple_loss=0.4334, pruned_loss=0.1757, over 27519.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4026, pruned_loss=0.1508, over 5665331.65 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3975, pruned_loss=0.1468, over 5713116.96 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4031, pruned_loss=0.1514, over 5657753.45 frames. ], batch size: 472, lr: 6.24e-03, grad_scale: 8.0
+2023-03-02 20:08:52,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1206, 1.3778, 3.3485, 3.1073], device='cuda:0'), covar=tensor([0.1332, 0.1976, 0.0412, 0.0785], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0531, 0.0761, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 20:09:03,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.642e+02 1.595e+03 2.100e+03 2.880e+03 6.491e+03, threshold=4.200e+03, percent-clipped=8.0
+2023-03-02 20:09:33,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0308, 1.4593, 1.3020, 1.2882], device='cuda:0'), covar=tensor([0.0851, 0.0355, 0.0313, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0124, 0.0127, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0057, 0.0042, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 20:09:37,469 INFO [train.py:968] (0/2) Epoch 5, batch 29250, giga_loss[loss=0.3226, simple_loss=0.3913, pruned_loss=0.1269, over 28975.00 frames. ], tot_loss[loss=0.3508, simple_loss=0.4022, pruned_loss=0.1496, over 5653367.73 frames. ], libri_tot_loss[loss=0.3458, simple_loss=0.3978, pruned_loss=0.1469, over 5715193.89 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4024, pruned_loss=0.15, over 5644762.05 frames. ], batch size: 136, lr: 6.24e-03, grad_scale: 8.0
+2023-03-02 20:09:41,279 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-02 20:09:45,166 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:10:22,882 INFO [train.py:968] (0/2) Epoch 5, batch 29300, giga_loss[loss=0.3139, simple_loss=0.3866, pruned_loss=0.1206, over 28947.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.4019, pruned_loss=0.149, over 5647237.87 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.398, pruned_loss=0.1471, over 5709803.21 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.4019, pruned_loss=0.1492, over 5643710.11 frames. ], batch size: 164, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:10:40,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.028e+02 1.501e+03 1.749e+03 2.543e+03 6.034e+03, threshold=3.497e+03, percent-clipped=1.0
+2023-03-02 20:10:51,123 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 20:11:08,605 INFO [train.py:968] (0/2) Epoch 5, batch 29350, giga_loss[loss=0.3975, simple_loss=0.4322, pruned_loss=0.1814, over 27967.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.4002, pruned_loss=0.1477, over 5659200.40 frames. ], libri_tot_loss[loss=0.3456, simple_loss=0.3976, pruned_loss=0.1468, over 5710471.62 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4007, pruned_loss=0.1481, over 5654496.44 frames. ], batch size: 412, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:11:26,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3188, 1.4774, 1.2413, 1.5570], device='cuda:0'), covar=tensor([0.0778, 0.0328, 0.0326, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0124, 0.0126, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0057, 0.0042, 0.0037, 0.0063], device='cuda:0')
+2023-03-02 20:11:32,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2783, 4.0867, 3.9508, 1.7259], device='cuda:0'), covar=tensor([0.0433, 0.0513, 0.0679, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0817, 0.0828, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 20:11:36,804 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-02 20:11:52,655 INFO [train.py:968] (0/2) Epoch 5, batch 29400, giga_loss[loss=0.3531, simple_loss=0.4025, pruned_loss=0.1518, over 27875.00 frames. ], tot_loss[loss=0.346, simple_loss=0.399, pruned_loss=0.1465, over 5658834.82 frames. ], libri_tot_loss[loss=0.3464, simple_loss=0.3983, pruned_loss=0.1473, over 5705541.95 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3988, pruned_loss=0.1465, over 5657562.16 frames. ], batch size: 412, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:12:10,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.488e+03 2.102e+03 3.296e+03 1.069e+04, threshold=4.203e+03, percent-clipped=21.0
+2023-03-02 20:12:37,402 INFO [train.py:968] (0/2) Epoch 5, batch 29450, giga_loss[loss=0.3363, simple_loss=0.3982, pruned_loss=0.1371, over 28885.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3992, pruned_loss=0.1467, over 5644777.37 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3974, pruned_loss=0.1467, over 5694037.99 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.3998, pruned_loss=0.1473, over 5651703.09 frames. ], batch size: 186, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:13:06,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:13:27,125 INFO [train.py:968] (0/2) Epoch 5, batch 29500, giga_loss[loss=0.3629, simple_loss=0.4153, pruned_loss=0.1553, over 28583.00 frames. ], tot_loss[loss=0.347, simple_loss=0.3994, pruned_loss=0.1473, over 5642410.96 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3973, pruned_loss=0.1466, over 5688628.65 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.4001, pruned_loss=0.1478, over 5652198.55 frames. ], batch size: 336, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:13:27,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:13:45,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.464e+02 1.463e+03 1.842e+03 2.501e+03 4.894e+03, threshold=3.685e+03, percent-clipped=3.0
+2023-03-02 20:13:46,219 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-02 20:14:00,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:14:13,333 INFO [train.py:968] (0/2) Epoch 5, batch 29550, libri_loss[loss=0.316, simple_loss=0.3781, pruned_loss=0.127, over 29550.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.3989, pruned_loss=0.1481, over 5653012.99 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3971, pruned_loss=0.1466, over 5694020.78 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.3998, pruned_loss=0.1486, over 5653886.39 frames. ], batch size: 79, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:14:13,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:14:31,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2895, 1.5287, 1.1857, 1.4454], device='cuda:0'), covar=tensor([0.2084, 0.2125, 0.2156, 0.1954], device='cuda:0'), in_proj_covar=tensor([0.1124, 0.0886, 0.0999, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 20:14:58,884 INFO [train.py:968] (0/2) Epoch 5, batch 29600, giga_loss[loss=0.322, simple_loss=0.3802, pruned_loss=0.1319, over 28986.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.3991, pruned_loss=0.1483, over 5655642.88 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.3969, pruned_loss=0.1465, over 5686943.77 frames. ], giga_tot_loss[loss=0.3489, simple_loss=0.4, pruned_loss=0.1489, over 5661701.81 frames. ], batch size: 213, lr: 6.24e-03, grad_scale: 8.0
+2023-03-02 20:15:17,233 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.964e+02 1.512e+03 2.104e+03 2.733e+03 4.759e+03, threshold=4.208e+03, percent-clipped=10.0
+2023-03-02 20:15:29,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:15:38,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:15:41,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:15:44,870 INFO [train.py:968] (0/2) Epoch 5, batch 29650, giga_loss[loss=0.3605, simple_loss=0.4122, pruned_loss=0.1544, over 28445.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4007, pruned_loss=0.1497, over 5640143.19 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3971, pruned_loss=0.1467, over 5672675.99 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4013, pruned_loss=0.1499, over 5657819.48 frames. ], batch size: 336, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:16:09,867 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:11,264 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:13,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1923, 3.0355, 1.3189, 1.2439], device='cuda:0'), covar=tensor([0.0958, 0.0407, 0.0911, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0484, 0.0310, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 20:16:13,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:26,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:30,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:33,276 INFO [train.py:968] (0/2) Epoch 5, batch 29700, giga_loss[loss=0.3643, simple_loss=0.4162, pruned_loss=0.1562, over 27926.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.4004, pruned_loss=0.1496, over 5638984.04 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.3968, pruned_loss=0.1465, over 5678959.38 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4012, pruned_loss=0.15, over 5646766.95 frames. ], batch size: 412, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:16:43,962 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:51,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.680e+03 2.319e+03 3.293e+03 7.130e+03, threshold=4.638e+03, percent-clipped=11.0
+2023-03-02 20:16:57,499 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:17:24,456 INFO [train.py:968] (0/2) Epoch 5, batch 29750, giga_loss[loss=0.3215, simple_loss=0.3855, pruned_loss=0.1288, over 29022.00 frames. ], tot_loss[loss=0.3484, simple_loss=0.3995, pruned_loss=0.1487, over 5638099.42 frames. ], libri_tot_loss[loss=0.3445, simple_loss=0.3966, pruned_loss=0.1463, over 5682308.88 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4004, pruned_loss=0.1493, over 5640791.05 frames. ], batch size: 106, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:17:47,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:17:49,919 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:17:59,716 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4274, 1.4830, 1.5257, 1.4474], device='cuda:0'), covar=tensor([0.0805, 0.1093, 0.1124, 0.1029], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0743, 0.0638, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 20:18:10,492 INFO [train.py:968] (0/2) Epoch 5, batch 29800, giga_loss[loss=0.4012, simple_loss=0.4316, pruned_loss=0.1854, over 27859.00 frames. ], tot_loss[loss=0.3483, simple_loss=0.3999, pruned_loss=0.1484, over 5652080.47 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.3962, pruned_loss=0.146, over 5687913.53 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.4011, pruned_loss=0.1492, over 5648131.78 frames. ], batch size: 412, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:18:17,808 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:18:29,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.077e+02 1.306e+03 1.650e+03 2.383e+03 5.602e+03, threshold=3.300e+03, percent-clipped=2.0
+2023-03-02 20:18:32,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0217, 1.1278, 3.8858, 3.1379], device='cuda:0'), covar=tensor([0.1616, 0.2225, 0.0363, 0.0779], device='cuda:0'), in_proj_covar=tensor([0.0581, 0.0537, 0.0769, 0.0624], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 20:18:55,503 INFO [train.py:968] (0/2) Epoch 5, batch 29850, giga_loss[loss=0.4741, simple_loss=0.4669, pruned_loss=0.2406, over 26482.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.3994, pruned_loss=0.1478, over 5651553.99 frames. ], libri_tot_loss[loss=0.3438, simple_loss=0.396, pruned_loss=0.1458, over 5683902.44 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4006, pruned_loss=0.1487, over 5650637.27 frames. ], batch size: 555, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:18:58,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:19:23,800 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:19:42,151 INFO [train.py:968] (0/2) Epoch 5, batch 29900, giga_loss[loss=0.3112, simple_loss=0.379, pruned_loss=0.1217, over 29058.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3989, pruned_loss=0.1471, over 5661237.85 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.3963, pruned_loss=0.1459, over 5685777.02 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.3997, pruned_loss=0.1477, over 5658175.06 frames. ], batch size: 128, lr: 6.23e-03, grad_scale: 2.0
+2023-03-02 20:19:45,601 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:19:59,206 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.664e+02 1.841e+03 2.743e+03 4.237e+03 1.167e+04, threshold=5.485e+03, percent-clipped=41.0
+2023-03-02 20:20:15,015 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-02 20:20:25,976 INFO [train.py:968] (0/2) Epoch 5, batch 29950, giga_loss[loss=0.3401, simple_loss=0.3903, pruned_loss=0.1449, over 27809.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3965, pruned_loss=0.1454, over 5668428.22 frames. ], libri_tot_loss[loss=0.3443, simple_loss=0.3963, pruned_loss=0.1461, over 5690919.60 frames. ], giga_tot_loss[loss=0.3444, simple_loss=0.3972, pruned_loss=0.1458, over 5660983.63 frames. ], batch size: 412, lr: 6.23e-03, grad_scale: 2.0
+2023-03-02 20:21:06,765 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:21:10,144 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-02 20:21:13,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:21:13,889 INFO [train.py:968] (0/2) Epoch 5, batch 30000, giga_loss[loss=0.381, simple_loss=0.3957, pruned_loss=0.1831, over 23539.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3923, pruned_loss=0.143, over 5654118.34 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3966, pruned_loss=0.1463, over 5684918.27 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3926, pruned_loss=0.1431, over 5653808.16 frames. ], batch size: 705, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:21:13,894 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 20:21:22,279 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2353, simple_loss=0.3403, pruned_loss=0.06511, over 944034.00 frames. 
+2023-03-02 20:21:22,280 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 20:21:39,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.555e+02 1.568e+03 1.935e+03 2.368e+03 4.302e+03, threshold=3.871e+03, percent-clipped=0.0
+2023-03-02 20:21:44,065 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:21:50,434 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0491, 1.9216, 1.2899, 1.6772], device='cuda:0'), covar=tensor([0.0569, 0.0564, 0.0985, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0453, 0.0501, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 20:22:04,578 INFO [train.py:968] (0/2) Epoch 5, batch 30050, giga_loss[loss=0.3664, simple_loss=0.4119, pruned_loss=0.1605, over 28874.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3895, pruned_loss=0.1423, over 5655869.19 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.397, pruned_loss=0.1465, over 5689243.82 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3893, pruned_loss=0.1421, over 5651345.24 frames. ], batch size: 199, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:22:54,903 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-212000.pt
+2023-03-02 20:22:55,238 INFO [train.py:968] (0/2) Epoch 5, batch 30100, giga_loss[loss=0.335, simple_loss=0.387, pruned_loss=0.1415, over 28625.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3888, pruned_loss=0.1427, over 5651109.32 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.3972, pruned_loss=0.1467, over 5692579.67 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3884, pruned_loss=0.1423, over 5644077.72 frames. ], batch size: 262, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:23:12,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.409e+02 1.458e+03 1.824e+03 2.492e+03 6.287e+03, threshold=3.647e+03, percent-clipped=7.0
+2023-03-02 20:23:43,227 INFO [train.py:968] (0/2) Epoch 5, batch 30150, giga_loss[loss=0.3728, simple_loss=0.4193, pruned_loss=0.1632, over 28603.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3878, pruned_loss=0.1417, over 5647252.33 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3973, pruned_loss=0.1469, over 5696306.83 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3872, pruned_loss=0.1412, over 5637976.88 frames. ], batch size: 262, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:24:05,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4069, 3.4054, 1.4211, 1.4604], device='cuda:0'), covar=tensor([0.0891, 0.0297, 0.0891, 0.1322], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0487, 0.0310, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 20:24:30,192 INFO [train.py:968] (0/2) Epoch 5, batch 30200, giga_loss[loss=0.3123, simple_loss=0.3786, pruned_loss=0.123, over 28883.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3859, pruned_loss=0.1383, over 5653697.97 frames. ], libri_tot_loss[loss=0.3456, simple_loss=0.3973, pruned_loss=0.147, over 5699797.13 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3852, pruned_loss=0.1376, over 5641718.38 frames. ], batch size: 213, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:24:52,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.003e+02 1.403e+03 1.832e+03 2.414e+03 5.730e+03, threshold=3.664e+03, percent-clipped=6.0
+2023-03-02 20:25:24,166 INFO [train.py:968] (0/2) Epoch 5, batch 30250, giga_loss[loss=0.2948, simple_loss=0.3645, pruned_loss=0.1125, over 28924.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3826, pruned_loss=0.1343, over 5652812.81 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3964, pruned_loss=0.1465, over 5703336.12 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3825, pruned_loss=0.1339, over 5639084.18 frames. ], batch size: 106, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:25:29,823 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:25:54,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:26:15,119 INFO [train.py:968] (0/2) Epoch 5, batch 30300, libri_loss[loss=0.3548, simple_loss=0.3928, pruned_loss=0.1584, over 29526.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3789, pruned_loss=0.1298, over 5664386.72 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.3962, pruned_loss=0.1465, over 5706336.25 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3788, pruned_loss=0.1293, over 5650158.09 frames. ], batch size: 80, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:26:36,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.651e+02 1.438e+03 1.910e+03 3.156e+03 6.473e+03, threshold=3.820e+03, percent-clipped=17.0
+2023-03-02 20:27:04,350 INFO [train.py:968] (0/2) Epoch 5, batch 30350, giga_loss[loss=0.3149, simple_loss=0.3742, pruned_loss=0.1278, over 27943.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3755, pruned_loss=0.1266, over 5661870.13 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.3956, pruned_loss=0.1461, over 5706628.34 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3755, pruned_loss=0.1261, over 5649168.83 frames. ], batch size: 412, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:27:12,365 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 20:27:29,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7752, 3.6534, 3.4232, 1.6351], device='cuda:0'), covar=tensor([0.0619, 0.0672, 0.0925, 0.2232], device='cuda:0'), in_proj_covar=tensor([0.0868, 0.0804, 0.0816, 0.0601], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 20:27:51,380 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:27:53,682 INFO [train.py:968] (0/2) Epoch 5, batch 30400, giga_loss[loss=0.2946, simple_loss=0.3646, pruned_loss=0.1122, over 27613.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3728, pruned_loss=0.1226, over 5664907.38 frames. ], libri_tot_loss[loss=0.3438, simple_loss=0.3954, pruned_loss=0.1461, over 5705996.55 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3727, pruned_loss=0.1219, over 5654598.10 frames. ], batch size: 472, lr: 6.23e-03, grad_scale: 8.0
+2023-03-02 20:27:54,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:28:02,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2299, 1.3767, 1.1362, 1.3357], device='cuda:0'), covar=tensor([0.2524, 0.2283, 0.2400, 0.2118], device='cuda:0'), in_proj_covar=tensor([0.1141, 0.0887, 0.1015, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 20:28:16,994 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.711e+02 1.347e+03 1.594e+03 2.341e+03 8.152e+03, threshold=3.189e+03, percent-clipped=10.0
+2023-03-02 20:28:19,819 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:28:23,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:28:27,628 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:28:50,036 INFO [train.py:968] (0/2) Epoch 5, batch 30450, giga_loss[loss=0.3357, simple_loss=0.4014, pruned_loss=0.135, over 28354.00 frames. ], tot_loss[loss=0.306, simple_loss=0.372, pruned_loss=0.12, over 5670361.04 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3952, pruned_loss=0.1461, over 5704123.74 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3719, pruned_loss=0.1193, over 5663426.62 frames. ], batch size: 368, lr: 6.23e-03, grad_scale: 8.0
+2023-03-02 20:28:55,563 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:29:42,682 INFO [train.py:968] (0/2) Epoch 5, batch 30500, giga_loss[loss=0.2662, simple_loss=0.3484, pruned_loss=0.09202, over 28701.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3703, pruned_loss=0.1186, over 5669742.92 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3952, pruned_loss=0.1462, over 5707400.95 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3698, pruned_loss=0.1174, over 5660644.43 frames. ], batch size: 242, lr: 6.22e-03, grad_scale: 8.0
+2023-03-02 20:30:06,144 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8710, 1.8133, 1.2362, 1.5368], device='cuda:0'), covar=tensor([0.0673, 0.0587, 0.1008, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0440, 0.0496, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 20:30:06,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.226e+02 1.198e+03 1.609e+03 2.268e+03 1.034e+04, threshold=3.219e+03, percent-clipped=9.0
+2023-03-02 20:30:34,619 INFO [train.py:968] (0/2) Epoch 5, batch 30550, giga_loss[loss=0.2655, simple_loss=0.3431, pruned_loss=0.09392, over 28834.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3664, pruned_loss=0.1154, over 5669600.78 frames. ], libri_tot_loss[loss=0.3434, simple_loss=0.3949, pruned_loss=0.146, over 5709423.01 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.366, pruned_loss=0.1144, over 5660086.75 frames. ], batch size: 186, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:30:43,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3688, 1.5867, 1.5050, 1.4835], device='cuda:0'), covar=tensor([0.0963, 0.1163, 0.1203, 0.1089], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0727, 0.0625, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 20:31:19,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=212497.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 20:31:22,904 INFO [train.py:968] (0/2) Epoch 5, batch 30600, giga_loss[loss=0.3336, simple_loss=0.3953, pruned_loss=0.1359, over 28815.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.365, pruned_loss=0.1153, over 5656608.63 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.3945, pruned_loss=0.146, over 5701109.49 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3639, pruned_loss=0.1134, over 5655870.82 frames. ], batch size: 186, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:31:39,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2797, 1.4538, 1.2099, 1.5017], device='cuda:0'), covar=tensor([0.0783, 0.0331, 0.0351, 0.0869], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0124, 0.0129, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0058, 0.0042, 0.0038, 0.0065], device='cuda:0')
+2023-03-02 20:31:45,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.086e+02 1.264e+03 1.656e+03 2.193e+03 5.877e+03, threshold=3.313e+03, percent-clipped=5.0
+2023-03-02 20:32:08,332 INFO [train.py:968] (0/2) Epoch 5, batch 30650, giga_loss[loss=0.2632, simple_loss=0.345, pruned_loss=0.09075, over 28784.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3645, pruned_loss=0.1149, over 5666699.33 frames. ], libri_tot_loss[loss=0.3421, simple_loss=0.3935, pruned_loss=0.1454, over 5706647.19 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3638, pruned_loss=0.1131, over 5659769.78 frames. ], batch size: 119, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:32:10,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-02 20:32:22,395 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-02 20:32:34,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8009, 1.9919, 1.4803, 1.3026], device='cuda:0'), covar=tensor([0.0847, 0.0647, 0.0506, 0.0677], device='cuda:0'), in_proj_covar=tensor([0.1405, 0.1228, 0.1172, 0.1270], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 20:32:35,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4140, 3.0449, 1.4296, 1.4011], device='cuda:0'), covar=tensor([0.0803, 0.0279, 0.0862, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0477, 0.0310, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 20:32:58,538 INFO [train.py:968] (0/2) Epoch 5, batch 30700, giga_loss[loss=0.3348, simple_loss=0.3914, pruned_loss=0.1391, over 29010.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.365, pruned_loss=0.1154, over 5666191.61 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3937, pruned_loss=0.1457, over 5710636.27 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3637, pruned_loss=0.113, over 5656243.40 frames. ], batch size: 136, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:33:18,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.571e+02 1.424e+03 1.793e+03 2.384e+03 6.083e+03, threshold=3.587e+03, percent-clipped=11.0
+2023-03-02 20:33:41,607 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=212646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:33:45,235 INFO [train.py:968] (0/2) Epoch 5, batch 30750, giga_loss[loss=0.282, simple_loss=0.3609, pruned_loss=0.1015, over 28987.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3624, pruned_loss=0.1136, over 5655962.27 frames. ], libri_tot_loss[loss=0.3428, simple_loss=0.3934, pruned_loss=0.1461, over 5698746.81 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3606, pruned_loss=0.1103, over 5656893.43 frames. ], batch size: 145, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:34:20,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3191, 1.4599, 1.2305, 1.6534], device='cuda:0'), covar=tensor([0.2460, 0.2216, 0.2252, 0.2104], device='cuda:0'), in_proj_covar=tensor([0.1131, 0.0872, 0.1004, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 20:34:33,887 INFO [train.py:968] (0/2) Epoch 5, batch 30800, giga_loss[loss=0.259, simple_loss=0.3364, pruned_loss=0.09083, over 28758.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.358, pruned_loss=0.1107, over 5657349.24 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3924, pruned_loss=0.1457, over 5694796.76 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3566, pruned_loss=0.1075, over 5661053.85 frames. ], batch size: 119, lr: 6.22e-03, grad_scale: 8.0
+2023-03-02 20:34:55,313 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.108e+02 1.374e+03 1.935e+03 2.562e+03 6.214e+03, threshold=3.870e+03, percent-clipped=8.0
+2023-03-02 20:35:14,022 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2282, 1.2455, 3.4911, 3.3175], device='cuda:0'), covar=tensor([0.1247, 0.2018, 0.0340, 0.0752], device='cuda:0'), in_proj_covar=tensor([0.0566, 0.0522, 0.0748, 0.0607], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 20:35:20,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3503, 2.0285, 1.5127, 1.8753], device='cuda:0'), covar=tensor([0.0612, 0.0645, 0.0916, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0446, 0.0497, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 20:35:22,925 INFO [train.py:968] (0/2) Epoch 5, batch 30850, libri_loss[loss=0.2767, simple_loss=0.3365, pruned_loss=0.1084, over 29585.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.355, pruned_loss=0.1094, over 5665026.82 frames. ], libri_tot_loss[loss=0.3402, simple_loss=0.391, pruned_loss=0.1447, over 5698235.40 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3537, pruned_loss=0.1062, over 5663038.29 frames. ], batch size: 77, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:35:51,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6385, 3.0157, 1.6181, 1.6062], device='cuda:0'), covar=tensor([0.0634, 0.0334, 0.0700, 0.1041], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0477, 0.0310, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 20:35:55,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3251, 2.1159, 1.5788, 0.5675], device='cuda:0'), covar=tensor([0.2330, 0.1450, 0.2387, 0.2817], device='cuda:0'), in_proj_covar=tensor([0.1379, 0.1315, 0.1371, 0.1159], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 20:36:12,927 INFO [train.py:968] (0/2) Epoch 5, batch 30900, giga_loss[loss=0.2743, simple_loss=0.3415, pruned_loss=0.1036, over 28900.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.354, pruned_loss=0.1092, over 5661935.61 frames. ], libri_tot_loss[loss=0.3403, simple_loss=0.391, pruned_loss=0.1448, over 5697272.85 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3524, pruned_loss=0.1061, over 5660626.19 frames. ], batch size: 199, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:36:36,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.084e+02 1.333e+03 1.893e+03 2.518e+03 4.561e+03, threshold=3.786e+03, percent-clipped=4.0
+2023-03-02 20:36:46,685 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=212832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:37:01,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4275, 1.5563, 1.3293, 1.9428], device='cuda:0'), covar=tensor([0.2409, 0.2100, 0.2210, 0.2044], device='cuda:0'), in_proj_covar=tensor([0.1132, 0.0865, 0.1003, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 20:37:04,051 INFO [train.py:968] (0/2) Epoch 5, batch 30950, giga_loss[loss=0.2856, simple_loss=0.3432, pruned_loss=0.114, over 26677.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3535, pruned_loss=0.1091, over 5650901.46 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3903, pruned_loss=0.1444, over 5702352.17 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3516, pruned_loss=0.1058, over 5643802.29 frames. ], batch size: 555, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:37:27,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212872.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 20:38:00,462 INFO [train.py:968] (0/2) Epoch 5, batch 31000, giga_loss[loss=0.2961, simple_loss=0.3708, pruned_loss=0.1107, over 28462.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3565, pruned_loss=0.1102, over 5654012.37 frames. ], libri_tot_loss[loss=0.3386, simple_loss=0.3895, pruned_loss=0.1439, over 5707759.77 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3549, pruned_loss=0.1073, over 5642468.53 frames. ], batch size: 336, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:38:24,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.398e+02 1.457e+03 2.063e+03 2.849e+03 1.022e+04, threshold=4.126e+03, percent-clipped=14.0
+2023-03-02 20:38:33,941 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3164, 1.5052, 1.4839, 1.3750], device='cuda:0'), covar=tensor([0.0976, 0.1432, 0.1373, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0721, 0.0618, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 20:38:45,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9012, 1.0458, 3.8308, 3.0463], device='cuda:0'), covar=tensor([0.1765, 0.2466, 0.0358, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0571, 0.0525, 0.0754, 0.0609], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 20:38:59,066 INFO [train.py:968] (0/2) Epoch 5, batch 31050, giga_loss[loss=0.2849, simple_loss=0.3593, pruned_loss=0.1052, over 28897.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3562, pruned_loss=0.109, over 5646386.07 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.3887, pruned_loss=0.1435, over 5711232.91 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3551, pruned_loss=0.1063, over 5633123.73 frames. ], batch size: 213, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:39:58,604 INFO [train.py:968] (0/2) Epoch 5, batch 31100, giga_loss[loss=0.2936, simple_loss=0.3733, pruned_loss=0.1069, over 28916.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3567, pruned_loss=0.1096, over 5642225.22 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3878, pruned_loss=0.1428, over 5713456.92 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3555, pruned_loss=0.1069, over 5626817.17 frames. ], batch size: 174, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:40:15,754 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213015.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 20:40:19,014 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213018.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 20:40:22,643 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:40:23,660 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.179e+02 1.457e+03 1.932e+03 2.387e+03 4.606e+03, threshold=3.864e+03, percent-clipped=2.0
+2023-03-02 20:40:26,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4206, 1.5271, 1.3217, 1.5891], device='cuda:0'), covar=tensor([0.2258, 0.2016, 0.2029, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.1128, 0.0866, 0.1003, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 20:40:53,024 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213047.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 20:40:55,541 INFO [train.py:968] (0/2) Epoch 5, batch 31150, giga_loss[loss=0.2402, simple_loss=0.3212, pruned_loss=0.07957, over 28559.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3561, pruned_loss=0.1096, over 5655055.46 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3875, pruned_loss=0.1429, over 5717483.76 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3542, pruned_loss=0.1061, over 5636729.71 frames. ], batch size: 85, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:41:42,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:41:46,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3512, 1.7745, 1.4407, 1.6233], device='cuda:0'), covar=tensor([0.0727, 0.0269, 0.0300, 0.0704], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0124, 0.0129, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0059, 0.0042, 0.0038, 0.0065], device='cuda:0')
+2023-03-02 20:41:52,400 INFO [train.py:968] (0/2) Epoch 5, batch 31200, giga_loss[loss=0.2338, simple_loss=0.3036, pruned_loss=0.08203, over 24285.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.355, pruned_loss=0.1084, over 5642932.53 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3867, pruned_loss=0.1424, over 5713026.09 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3527, pruned_loss=0.1043, over 5628510.81 frames. ], batch size: 705, lr: 6.21e-03, grad_scale: 8.0
+2023-03-02 20:41:55,994 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:42:12,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3925, 1.8714, 1.7490, 1.5992], device='cuda:0'), covar=tensor([0.1671, 0.1933, 0.1246, 0.1310], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0726, 0.0771, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-02 20:42:20,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.568e+02 1.335e+03 1.661e+03 2.322e+03 6.385e+03, threshold=3.323e+03, percent-clipped=7.0
+2023-03-02 20:42:49,850 INFO [train.py:968] (0/2) Epoch 5, batch 31250, giga_loss[loss=0.2869, simple_loss=0.3566, pruned_loss=0.1086, over 28129.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3521, pruned_loss=0.1054, over 5636756.27 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3863, pruned_loss=0.1422, over 5705968.99 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3499, pruned_loss=0.1016, over 5630218.93 frames. ], batch size: 412, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:43:05,943 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:43:09,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:43:09,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:43:42,732 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:43:48,387 INFO [train.py:968] (0/2) Epoch 5, batch 31300, giga_loss[loss=0.2639, simple_loss=0.3331, pruned_loss=0.09736, over 28981.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3497, pruned_loss=0.1051, over 5643726.80 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3862, pruned_loss=0.1423, over 5695215.30 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3474, pruned_loss=0.1012, over 5646916.27 frames. ], batch size: 155, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:43:58,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:44:20,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.093e+02 1.410e+03 1.955e+03 2.981e+03 1.179e+04, threshold=3.909e+03, percent-clipped=19.0
+2023-03-02 20:44:48,585 INFO [train.py:968] (0/2) Epoch 5, batch 31350, giga_loss[loss=0.2825, simple_loss=0.3537, pruned_loss=0.1057, over 28089.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3503, pruned_loss=0.1059, over 5659818.34 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3856, pruned_loss=0.1421, over 5699867.79 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.348, pruned_loss=0.102, over 5657012.93 frames. ], batch size: 412, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:45:33,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4668, 3.6180, 1.4457, 1.6092], device='cuda:0'), covar=tensor([0.0939, 0.0340, 0.0943, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0475, 0.0312, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 20:45:48,084 INFO [train.py:968] (0/2) Epoch 5, batch 31400, giga_loss[loss=0.2554, simple_loss=0.3392, pruned_loss=0.08581, over 28914.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3511, pruned_loss=0.106, over 5666399.35 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3855, pruned_loss=0.1419, over 5701795.78 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3491, pruned_loss=0.1028, over 5662062.16 frames. ], batch size: 213, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:46:12,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.095e+02 1.343e+03 1.704e+03 2.615e+03 8.222e+03, threshold=3.408e+03, percent-clipped=7.0
+2023-03-02 20:46:45,725 INFO [train.py:968] (0/2) Epoch 5, batch 31450, giga_loss[loss=0.2969, simple_loss=0.3853, pruned_loss=0.1042, over 28905.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3522, pruned_loss=0.1056, over 5662733.56 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.385, pruned_loss=0.1415, over 5704901.78 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.35, pruned_loss=0.1022, over 5655009.34 frames. ], batch size: 119, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:46:45,983 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:46:49,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:46:54,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3231, 1.4862, 1.2007, 1.4616], device='cuda:0'), covar=tensor([0.2143, 0.1999, 0.2060, 0.1945], device='cuda:0'), in_proj_covar=tensor([0.1124, 0.0866, 0.1000, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 20:47:27,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:47:46,915 INFO [train.py:968] (0/2) Epoch 5, batch 31500, giga_loss[loss=0.2411, simple_loss=0.3194, pruned_loss=0.08144, over 28361.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3516, pruned_loss=0.1048, over 5657621.31 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3845, pruned_loss=0.1414, over 5699647.06 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3495, pruned_loss=0.1014, over 5654117.03 frames. ], batch size: 368, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:48:21,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.711e+02 1.211e+03 1.579e+03 1.975e+03 4.703e+03, threshold=3.159e+03, percent-clipped=5.0
+2023-03-02 20:48:55,916 INFO [train.py:968] (0/2) Epoch 5, batch 31550, giga_loss[loss=0.2274, simple_loss=0.3108, pruned_loss=0.07203, over 28897.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3488, pruned_loss=0.1029, over 5674588.00 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3842, pruned_loss=0.1412, over 5701560.09 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3472, pruned_loss=0.1, over 5669852.35 frames. ], batch size: 106, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:49:20,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:49:38,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:50:05,811 INFO [train.py:968] (0/2) Epoch 5, batch 31600, giga_loss[loss=0.3884, simple_loss=0.4225, pruned_loss=0.1772, over 26783.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3518, pruned_loss=0.1051, over 5669175.12 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3838, pruned_loss=0.1412, over 5702031.38 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3503, pruned_loss=0.1022, over 5664138.97 frames. ], batch size: 555, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:50:37,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.587e+02 1.387e+03 2.072e+03 2.887e+03 8.017e+03, threshold=4.143e+03, percent-clipped=22.0
+2023-03-02 20:50:42,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2618, 3.0252, 1.3243, 1.4376], device='cuda:0'), covar=tensor([0.0872, 0.0244, 0.0849, 0.1250], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0472, 0.0311, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 20:50:59,970 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:51:09,839 INFO [train.py:968] (0/2) Epoch 5, batch 31650, giga_loss[loss=0.2455, simple_loss=0.3395, pruned_loss=0.07575, over 28081.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3553, pruned_loss=0.104, over 5665886.19 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3836, pruned_loss=0.1411, over 5700889.22 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.354, pruned_loss=0.1015, over 5662703.49 frames. ], batch size: 412, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:52:11,888 INFO [train.py:968] (0/2) Epoch 5, batch 31700, libri_loss[loss=0.3662, simple_loss=0.414, pruned_loss=0.1592, over 29372.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3563, pruned_loss=0.1035, over 5663823.10 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3834, pruned_loss=0.1408, over 5706971.82 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3547, pruned_loss=0.1006, over 5654497.10 frames. ], batch size: 92, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:52:18,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3282, 1.5563, 1.3080, 1.4293], device='cuda:0'), covar=tensor([0.1359, 0.1876, 0.1790, 0.1702], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0729, 0.0623, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 20:52:20,865 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:52:25,964 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:52:37,190 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:52:39,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.027e+02 1.449e+03 2.016e+03 2.868e+03 7.119e+03, threshold=4.032e+03, percent-clipped=11.0
+2023-03-02 20:52:40,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:53:00,661 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:53:10,197 INFO [train.py:968] (0/2) Epoch 5, batch 31750, giga_loss[loss=0.2751, simple_loss=0.349, pruned_loss=0.1006, over 27621.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3557, pruned_loss=0.1014, over 5664867.09 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3836, pruned_loss=0.141, over 5706662.23 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3539, pruned_loss=0.09844, over 5657396.03 frames. ], batch size: 472, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:53:16,068 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:53:38,623 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 20:53:51,065 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:53:54,415 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:53:58,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8264, 1.6221, 1.3154, 1.3531], device='cuda:0'), covar=tensor([0.0610, 0.0577, 0.0917, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0441, 0.0500, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 20:54:05,083 INFO [train.py:968] (0/2) Epoch 5, batch 31800, giga_loss[loss=0.2553, simple_loss=0.3322, pruned_loss=0.08925, over 28592.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3568, pruned_loss=0.1028, over 5683246.08 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3832, pruned_loss=0.1408, over 5713104.59 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3546, pruned_loss=0.09909, over 5669790.18 frames. ], batch size: 78, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:54:25,594 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:54:33,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.051e+02 1.326e+03 1.875e+03 2.550e+03 5.135e+03, threshold=3.750e+03, percent-clipped=3.0
+2023-03-02 20:55:04,344 INFO [train.py:968] (0/2) Epoch 5, batch 31850, giga_loss[loss=0.2624, simple_loss=0.3402, pruned_loss=0.0923, over 28876.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3553, pruned_loss=0.1032, over 5692743.36 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3824, pruned_loss=0.1403, over 5715145.45 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3533, pruned_loss=0.09925, over 5678907.40 frames. ], batch size: 199, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 20:55:19,671 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7464, 1.6965, 1.1665, 1.4566], device='cuda:0'), covar=tensor([0.0593, 0.0498, 0.0926, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0441, 0.0501, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 20:56:15,844 INFO [train.py:968] (0/2) Epoch 5, batch 31900, giga_loss[loss=0.2894, simple_loss=0.3565, pruned_loss=0.1112, over 28949.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3563, pruned_loss=0.1051, over 5681927.33 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3819, pruned_loss=0.1401, over 5715737.58 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3544, pruned_loss=0.1011, over 5669492.14 frames. ], batch size: 213, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 20:56:35,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9573, 1.2012, 3.3828, 2.9406], device='cuda:0'), covar=tensor([0.1504, 0.2175, 0.0381, 0.1301], device='cuda:0'), in_proj_covar=tensor([0.0567, 0.0523, 0.0739, 0.0599], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 20:56:58,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.984e+02 1.360e+03 1.766e+03 2.307e+03 6.824e+03, threshold=3.532e+03, percent-clipped=5.0
+2023-03-02 20:57:31,503 INFO [train.py:968] (0/2) Epoch 5, batch 31950, giga_loss[loss=0.2297, simple_loss=0.314, pruned_loss=0.07275, over 28641.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3547, pruned_loss=0.1046, over 5688560.96 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3813, pruned_loss=0.1397, over 5718156.98 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3533, pruned_loss=0.1012, over 5675941.28 frames. ], batch size: 242, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 20:57:40,378 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-02 20:58:38,288 INFO [train.py:968] (0/2) Epoch 5, batch 32000, giga_loss[loss=0.2725, simple_loss=0.3454, pruned_loss=0.09983, over 28344.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3515, pruned_loss=0.1026, over 5684579.47 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.3814, pruned_loss=0.1398, over 5722116.38 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3496, pruned_loss=0.09897, over 5670135.60 frames. ], batch size: 368, lr: 6.20e-03, grad_scale: 8.0
+2023-03-02 20:59:09,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.097e+02 1.508e+03 1.927e+03 2.797e+03 5.587e+03, threshold=3.854e+03, percent-clipped=12.0
+2023-03-02 20:59:41,813 INFO [train.py:968] (0/2) Epoch 5, batch 32050, giga_loss[loss=0.2232, simple_loss=0.3089, pruned_loss=0.0688, over 29057.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3487, pruned_loss=0.1011, over 5689140.43 frames. ], libri_tot_loss[loss=0.3299, simple_loss=0.3808, pruned_loss=0.1395, over 5725417.19 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.347, pruned_loss=0.09768, over 5673745.82 frames. ], batch size: 285, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 20:59:52,941 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:59:53,132 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 21:00:46,973 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-214000.pt
+2023-03-02 21:00:47,265 INFO [train.py:968] (0/2) Epoch 5, batch 32100, giga_loss[loss=0.2936, simple_loss=0.3758, pruned_loss=0.1057, over 28967.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3507, pruned_loss=0.1026, over 5690843.10 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3809, pruned_loss=0.1395, over 5723265.52 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3491, pruned_loss=0.09952, over 5680180.53 frames. ], batch size: 199, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:01:20,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.205e+02 1.423e+03 1.976e+03 2.861e+03 6.099e+03, threshold=3.952e+03, percent-clipped=8.0
+2023-03-02 21:01:48,322 INFO [train.py:968] (0/2) Epoch 5, batch 32150, giga_loss[loss=0.3258, simple_loss=0.3773, pruned_loss=0.1371, over 28148.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 5690325.20 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3808, pruned_loss=0.1397, over 5722343.69 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3527, pruned_loss=0.1016, over 5681670.35 frames. ], batch size: 412, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:02:15,181 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 21:02:51,226 INFO [train.py:968] (0/2) Epoch 5, batch 32200, giga_loss[loss=0.259, simple_loss=0.3331, pruned_loss=0.09242, over 28913.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3521, pruned_loss=0.1052, over 5686203.09 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3802, pruned_loss=0.1396, over 5717853.58 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3503, pruned_loss=0.1015, over 5681347.06 frames. ], batch size: 199, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:03:22,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.606e+02 1.407e+03 1.768e+03 2.648e+03 8.349e+03, threshold=3.536e+03, percent-clipped=5.0
+2023-03-02 21:03:50,714 INFO [train.py:968] (0/2) Epoch 5, batch 32250, giga_loss[loss=0.3089, simple_loss=0.3711, pruned_loss=0.1233, over 28876.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3536, pruned_loss=0.1069, over 5672789.81 frames. ], libri_tot_loss[loss=0.3303, simple_loss=0.3806, pruned_loss=0.14, over 5709954.42 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3514, pruned_loss=0.1031, over 5676024.61 frames. ], batch size: 213, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:04:58,302 INFO [train.py:968] (0/2) Epoch 5, batch 32300, giga_loss[loss=0.3195, simple_loss=0.3721, pruned_loss=0.1335, over 26840.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3523, pruned_loss=0.1056, over 5676837.38 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.38, pruned_loss=0.1397, over 5712757.09 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3508, pruned_loss=0.1024, over 5676522.57 frames. ], batch size: 555, lr: 6.20e-03, grad_scale: 2.0
+2023-03-02 21:05:34,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.288e+02 1.536e+03 2.004e+03 2.490e+03 5.675e+03, threshold=4.009e+03, percent-clipped=8.0
+2023-03-02 21:06:09,187 INFO [train.py:968] (0/2) Epoch 5, batch 32350, giga_loss[loss=0.2262, simple_loss=0.2946, pruned_loss=0.07889, over 24336.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3539, pruned_loss=0.1049, over 5671853.44 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3802, pruned_loss=0.1397, over 5713793.36 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3522, pruned_loss=0.1019, over 5670123.26 frames. ], batch size: 705, lr: 6.20e-03, grad_scale: 2.0
+2023-03-02 21:07:02,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2674, 1.3595, 1.2363, 1.4733], device='cuda:0'), covar=tensor([0.0766, 0.0360, 0.0352, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0197, 0.0124, 0.0129, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0059, 0.0043, 0.0038, 0.0065], device='cuda:0')
+2023-03-02 21:07:20,837 INFO [train.py:968] (0/2) Epoch 5, batch 32400, giga_loss[loss=0.3456, simple_loss=0.3815, pruned_loss=0.1549, over 26803.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3541, pruned_loss=0.1053, over 5658367.64 frames. ], libri_tot_loss[loss=0.3293, simple_loss=0.3797, pruned_loss=0.1395, over 5705465.03 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3524, pruned_loss=0.1022, over 5663344.14 frames. ], batch size: 555, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:07:53,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 21:07:58,265 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.926e+02 1.337e+03 1.611e+03 2.121e+03 7.949e+03, threshold=3.222e+03, percent-clipped=5.0
+2023-03-02 21:08:08,583 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:08:27,731 INFO [train.py:968] (0/2) Epoch 5, batch 32450, giga_loss[loss=0.2543, simple_loss=0.3282, pruned_loss=0.09017, over 28642.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3491, pruned_loss=0.1031, over 5671165.78 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3788, pruned_loss=0.1389, over 5709482.65 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3481, pruned_loss=0.1005, over 5670693.10 frames. ], batch size: 307, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:09:34,008 INFO [train.py:968] (0/2) Epoch 5, batch 32500, giga_loss[loss=0.2374, simple_loss=0.3174, pruned_loss=0.07872, over 28911.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3429, pruned_loss=0.1004, over 5675505.02 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3787, pruned_loss=0.1388, over 5711465.77 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3417, pruned_loss=0.09774, over 5672610.47 frames. ], batch size: 155, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:10:11,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.160e+02 1.482e+03 1.945e+03 2.758e+03 1.064e+04, threshold=3.889e+03, percent-clipped=16.0
+2023-03-02 21:10:39,146 INFO [train.py:968] (0/2) Epoch 5, batch 32550, giga_loss[loss=0.2567, simple_loss=0.3355, pruned_loss=0.08893, over 28961.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3425, pruned_loss=0.1007, over 5669782.24 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3782, pruned_loss=0.1386, over 5711025.56 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3414, pruned_loss=0.09829, over 5667297.39 frames. ], batch size: 213, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:11:07,760 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=214477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:11:12,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=214480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:11:31,405 INFO [train.py:968] (0/2) Epoch 5, batch 32600, libri_loss[loss=0.3155, simple_loss=0.3681, pruned_loss=0.1314, over 25895.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.347, pruned_loss=0.1043, over 5676499.53 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3785, pruned_loss=0.1387, over 5716234.68 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3441, pruned_loss=0.1003, over 5667765.63 frames. ], batch size: 136, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:11:40,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=214509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:11:46,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7581, 1.8073, 1.2183, 1.4710], device='cuda:0'), covar=tensor([0.0633, 0.0497, 0.0924, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0437, 0.0498, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 21:11:56,629 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=214522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:12:01,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.863e+02 1.468e+03 1.881e+03 2.772e+03 8.785e+03, threshold=3.762e+03, percent-clipped=10.0
+2023-03-02 21:12:32,338 INFO [train.py:968] (0/2) Epoch 5, batch 32650, giga_loss[loss=0.2433, simple_loss=0.3261, pruned_loss=0.08027, over 28709.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3445, pruned_loss=0.1016, over 5675060.31 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3778, pruned_loss=0.1382, over 5718720.61 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3424, pruned_loss=0.09849, over 5665639.13 frames. ], batch size: 243, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:13:33,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7356, 1.6476, 1.2376, 1.3249], device='cuda:0'), covar=tensor([0.0629, 0.0489, 0.0861, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0331, 0.0434, 0.0495, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 21:13:33,829 INFO [train.py:968] (0/2) Epoch 5, batch 32700, giga_loss[loss=0.2587, simple_loss=0.333, pruned_loss=0.09214, over 28136.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.343, pruned_loss=0.1003, over 5662495.59 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3775, pruned_loss=0.1382, over 5712017.69 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3407, pruned_loss=0.09677, over 5660156.42 frames. ], batch size: 412, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:13:47,365 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=214611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:13:50,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2523, 1.4700, 1.2477, 0.9297], device='cuda:0'), covar=tensor([0.1052, 0.0941, 0.0616, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.1414, 0.1192, 0.1156, 0.1256], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 21:14:09,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.555e+02 1.327e+03 1.729e+03 2.382e+03 4.590e+03, threshold=3.458e+03, percent-clipped=4.0
+2023-03-02 21:14:35,795 INFO [train.py:968] (0/2) Epoch 5, batch 32750, giga_loss[loss=0.2664, simple_loss=0.3401, pruned_loss=0.09633, over 28733.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3416, pruned_loss=0.09993, over 5663672.40 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3775, pruned_loss=0.1382, over 5710466.51 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.339, pruned_loss=0.09634, over 5662408.43 frames. ], batch size: 307, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:14:42,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2375, 1.8169, 1.4063, 0.4552], device='cuda:0'), covar=tensor([0.1875, 0.1415, 0.2239, 0.2376], device='cuda:0'), in_proj_covar=tensor([0.1381, 0.1308, 0.1359, 0.1146], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 21:15:28,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-02 21:15:36,869 INFO [train.py:968] (0/2) Epoch 5, batch 32800, giga_loss[loss=0.3052, simple_loss=0.3758, pruned_loss=0.1173, over 28698.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3414, pruned_loss=0.09977, over 5675029.09 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3769, pruned_loss=0.1379, over 5717709.25 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3382, pruned_loss=0.09535, over 5665308.52 frames. ], batch size: 242, lr: 6.19e-03, grad_scale: 8.0
+2023-03-02 21:15:38,246 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 21:16:11,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.563e+02 1.304e+03 1.603e+03 2.483e+03 6.813e+03, threshold=3.206e+03, percent-clipped=12.0
+2023-03-02 21:16:41,276 INFO [train.py:968] (0/2) Epoch 5, batch 32850, giga_loss[loss=0.226, simple_loss=0.2936, pruned_loss=0.07916, over 24263.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3424, pruned_loss=0.09991, over 5684040.04 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3768, pruned_loss=0.1377, over 5721088.97 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.339, pruned_loss=0.09544, over 5672028.71 frames. ], batch size: 705, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:17:41,171 INFO [train.py:968] (0/2) Epoch 5, batch 32900, giga_loss[loss=0.2857, simple_loss=0.3516, pruned_loss=0.1099, over 28911.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3445, pruned_loss=0.1021, over 5679281.87 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3768, pruned_loss=0.1378, over 5713506.39 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3406, pruned_loss=0.09703, over 5674800.10 frames. ], batch size: 227, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:18:12,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.224e+02 1.264e+03 1.819e+03 2.564e+03 1.417e+04, threshold=3.638e+03, percent-clipped=17.0
+2023-03-02 21:18:34,362 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-02 21:18:42,512 INFO [train.py:968] (0/2) Epoch 5, batch 32950, giga_loss[loss=0.249, simple_loss=0.3166, pruned_loss=0.09071, over 27501.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3439, pruned_loss=0.102, over 5667833.18 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3767, pruned_loss=0.138, over 5696922.50 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3404, pruned_loss=0.09734, over 5677683.42 frames. ], batch size: 472, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:19:03,523 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 21:19:39,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:19:40,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2519, 1.6588, 1.3251, 1.5062], device='cuda:0'), covar=tensor([0.0797, 0.0315, 0.0330, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0128, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0059, 0.0043, 0.0038, 0.0065], device='cuda:0')
+2023-03-02 21:19:42,141 INFO [train.py:968] (0/2) Epoch 5, batch 33000, giga_loss[loss=0.25, simple_loss=0.3367, pruned_loss=0.08165, over 28766.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3436, pruned_loss=0.1005, over 5646909.64 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3769, pruned_loss=0.1382, over 5689300.85 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3402, pruned_loss=0.09615, over 5660921.12 frames. ], batch size: 263, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:19:42,146 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 21:19:50,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8234, 3.6824, 3.5341, 1.6453], device='cuda:0'), covar=tensor([0.0723, 0.0647, 0.0833, 0.2410], device='cuda:0'), in_proj_covar=tensor([0.0832, 0.0771, 0.0771, 0.0583], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 21:19:50,442 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2195, simple_loss=0.317, pruned_loss=0.06099, over 944034.00 frames. 
+2023-03-02 21:19:50,443 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 21:20:21,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.487e+02 1.238e+03 1.782e+03 2.428e+03 7.402e+03, threshold=3.564e+03, percent-clipped=9.0
+2023-03-02 21:20:47,509 INFO [train.py:968] (0/2) Epoch 5, batch 33050, giga_loss[loss=0.2322, simple_loss=0.2982, pruned_loss=0.08307, over 24072.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3473, pruned_loss=0.1014, over 5645146.84 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.377, pruned_loss=0.1382, over 5689657.64 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3437, pruned_loss=0.09714, over 5655384.18 frames. ], batch size: 705, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:21:20,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2794, 1.4271, 1.3773, 1.2926], device='cuda:0'), covar=tensor([0.0801, 0.1071, 0.1247, 0.1031], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0723, 0.0618, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 21:21:28,598 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:21:33,376 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 21:21:46,679 INFO [train.py:968] (0/2) Epoch 5, batch 33100, giga_loss[loss=0.3048, simple_loss=0.3696, pruned_loss=0.12, over 28944.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3496, pruned_loss=0.1025, over 5659585.47 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3776, pruned_loss=0.1387, over 5695342.72 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3456, pruned_loss=0.09768, over 5661807.07 frames. ], batch size: 186, lr: 6.19e-03, grad_scale: 2.0
+2023-03-02 21:22:27,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.935e+02 1.477e+03 1.935e+03 2.861e+03 7.394e+03, threshold=3.870e+03, percent-clipped=10.0
+2023-03-02 21:22:41,343 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:22:45,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:22:52,493 INFO [train.py:968] (0/2) Epoch 5, batch 33150, giga_loss[loss=0.2547, simple_loss=0.3446, pruned_loss=0.08237, over 28790.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3498, pruned_loss=0.1024, over 5665614.99 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3771, pruned_loss=0.1385, over 5698753.35 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3466, pruned_loss=0.09827, over 5663560.77 frames. ], batch size: 174, lr: 6.19e-03, grad_scale: 2.0
+2023-03-02 21:23:15,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:23:29,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0994, 1.6011, 1.3938, 1.3085], device='cuda:0'), covar=tensor([0.1336, 0.1733, 0.1087, 0.1172], device='cuda:0'), in_proj_covar=tensor([0.0767, 0.0716, 0.0774, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-02 21:23:45,111 INFO [train.py:968] (0/2) Epoch 5, batch 33200, giga_loss[loss=0.2805, simple_loss=0.3506, pruned_loss=0.1052, over 28897.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3486, pruned_loss=0.102, over 5673244.98 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3766, pruned_loss=0.138, over 5704533.43 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3455, pruned_loss=0.09775, over 5665229.72 frames. ], batch size: 284, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:24:17,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.594e+02 1.419e+03 2.072e+03 2.933e+03 7.246e+03, threshold=4.145e+03, percent-clipped=10.0
+2023-03-02 21:24:19,384 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:24:22,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:24:40,198 INFO [train.py:968] (0/2) Epoch 5, batch 33250, giga_loss[loss=0.2698, simple_loss=0.3403, pruned_loss=0.09963, over 27649.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3447, pruned_loss=0.09928, over 5681404.57 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3758, pruned_loss=0.1375, over 5710001.30 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3419, pruned_loss=0.09508, over 5669055.06 frames. ], batch size: 472, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:24:57,134 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:25:42,211 INFO [train.py:968] (0/2) Epoch 5, batch 33300, giga_loss[loss=0.2332, simple_loss=0.3129, pruned_loss=0.0768, over 28092.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3435, pruned_loss=0.09902, over 5688430.46 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.376, pruned_loss=0.1376, over 5713861.90 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3405, pruned_loss=0.09486, over 5674549.62 frames. ], batch size: 412, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:26:01,502 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 21:26:16,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.062e+02 1.134e+03 1.630e+03 2.141e+03 5.957e+03, threshold=3.259e+03, percent-clipped=3.0
+2023-03-02 21:26:24,021 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6232, 2.4022, 2.3939, 2.2167], device='cuda:0'), covar=tensor([0.0894, 0.1827, 0.1258, 0.1505], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0716, 0.0611, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 21:26:37,528 INFO [train.py:968] (0/2) Epoch 5, batch 33350, giga_loss[loss=0.3142, simple_loss=0.3867, pruned_loss=0.1208, over 28943.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3443, pruned_loss=0.1003, over 5677017.44 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3756, pruned_loss=0.1374, over 5710305.77 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3411, pruned_loss=0.09571, over 5668170.72 frames. ], batch size: 284, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:27:38,861 INFO [train.py:968] (0/2) Epoch 5, batch 33400, giga_loss[loss=0.2828, simple_loss=0.3555, pruned_loss=0.105, over 28869.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3477, pruned_loss=0.1021, over 5676232.29 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3753, pruned_loss=0.137, over 5708080.29 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3445, pruned_loss=0.09756, over 5670335.10 frames. ], batch size: 199, lr: 6.18e-03, grad_scale: 2.0
+2023-03-02 21:28:15,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.829e+02 1.386e+03 1.812e+03 2.941e+03 1.321e+04, threshold=3.624e+03, percent-clipped=20.0
+2023-03-02 21:28:41,905 INFO [train.py:968] (0/2) Epoch 5, batch 33450, giga_loss[loss=0.2398, simple_loss=0.3151, pruned_loss=0.08226, over 28547.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3484, pruned_loss=0.1025, over 5679159.89 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.375, pruned_loss=0.1369, over 5711802.17 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3454, pruned_loss=0.09822, over 5670453.52 frames. ], batch size: 85, lr: 6.18e-03, grad_scale: 2.0
+2023-03-02 21:29:46,523 INFO [train.py:968] (0/2) Epoch 5, batch 33500, giga_loss[loss=0.2994, simple_loss=0.3774, pruned_loss=0.1107, over 28531.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.352, pruned_loss=0.1056, over 5661279.94 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3751, pruned_loss=0.1372, over 5704568.61 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3489, pruned_loss=0.1011, over 5659705.38 frames. ], batch size: 336, lr: 6.18e-03, grad_scale: 2.0
+2023-03-02 21:30:28,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.451e+02 1.209e+03 1.580e+03 2.143e+03 5.168e+03, threshold=3.160e+03, percent-clipped=3.0
+2023-03-02 21:30:34,331 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-02 21:30:45,714 INFO [train.py:968] (0/2) Epoch 5, batch 33550, giga_loss[loss=0.2845, simple_loss=0.3672, pruned_loss=0.1009, over 28879.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3549, pruned_loss=0.1068, over 5650708.76 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.375, pruned_loss=0.1372, over 5697547.34 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3521, pruned_loss=0.1025, over 5654815.15 frames. ], batch size: 227, lr: 6.18e-03, grad_scale: 2.0
+2023-03-02 21:31:03,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9351, 3.7680, 3.6153, 1.7128], device='cuda:0'), covar=tensor([0.0576, 0.0626, 0.0940, 0.1883], device='cuda:0'), in_proj_covar=tensor([0.0832, 0.0774, 0.0768, 0.0579], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 21:31:20,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3898, 1.6212, 1.3491, 1.5288], device='cuda:0'), covar=tensor([0.1106, 0.1819, 0.1642, 0.1534], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0720, 0.0619, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 21:31:49,830 INFO [train.py:968] (0/2) Epoch 5, batch 33600, giga_loss[loss=0.2818, simple_loss=0.352, pruned_loss=0.1058, over 29181.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3553, pruned_loss=0.1061, over 5650955.97 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3745, pruned_loss=0.1367, over 5697042.54 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3532, pruned_loss=0.1026, over 5653646.98 frames. ], batch size: 107, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:32:27,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.936e+02 1.248e+03 1.647e+03 2.446e+03 8.231e+03, threshold=3.293e+03, percent-clipped=9.0
+2023-03-02 21:32:56,601 INFO [train.py:968] (0/2) Epoch 5, batch 33650, giga_loss[loss=0.2791, simple_loss=0.3561, pruned_loss=0.101, over 28581.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.354, pruned_loss=0.1046, over 5667577.56 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3743, pruned_loss=0.1366, over 5700003.31 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3522, pruned_loss=0.1016, over 5666648.78 frames. ], batch size: 307, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:34:04,744 INFO [train.py:968] (0/2) Epoch 5, batch 33700, giga_loss[loss=0.2743, simple_loss=0.3504, pruned_loss=0.09913, over 28833.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3522, pruned_loss=0.1041, over 5677372.64 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3748, pruned_loss=0.137, over 5701108.89 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.35, pruned_loss=0.1008, over 5674861.42 frames. ], batch size: 174, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:34:11,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3892, 1.8686, 1.7752, 1.7065], device='cuda:0'), covar=tensor([0.1353, 0.1544, 0.1061, 0.1156], device='cuda:0'), in_proj_covar=tensor([0.0768, 0.0721, 0.0772, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-02 21:34:32,545 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215623.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:34:39,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.571e+02 1.398e+03 1.743e+03 2.451e+03 9.094e+03, threshold=3.486e+03, percent-clipped=13.0
+2023-03-02 21:35:02,848 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-02 21:35:06,822 INFO [train.py:968] (0/2) Epoch 5, batch 33750, giga_loss[loss=0.2803, simple_loss=0.3492, pruned_loss=0.1057, over 27636.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3506, pruned_loss=0.1035, over 5679754.79 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3741, pruned_loss=0.1366, over 5706442.15 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3489, pruned_loss=0.1004, over 5672453.57 frames. ], batch size: 472, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:36:10,356 INFO [train.py:968] (0/2) Epoch 5, batch 33800, giga_loss[loss=0.2305, simple_loss=0.3039, pruned_loss=0.07859, over 28899.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3493, pruned_loss=0.1038, over 5676530.11 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.374, pruned_loss=0.1366, over 5705724.39 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3476, pruned_loss=0.1006, over 5670537.46 frames. ], batch size: 119, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:36:35,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215718.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:36:40,557 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 21:36:47,082 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:36:51,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.685e+02 1.309e+03 1.655e+03 2.408e+03 5.309e+03, threshold=3.310e+03, percent-clipped=8.0
+2023-03-02 21:37:16,878 INFO [train.py:968] (0/2) Epoch 5, batch 33850, giga_loss[loss=0.2504, simple_loss=0.3296, pruned_loss=0.08562, over 28890.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3461, pruned_loss=0.102, over 5683292.87 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3739, pruned_loss=0.1365, over 5706602.77 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3447, pruned_loss=0.09946, over 5677804.00 frames. ], batch size: 213, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:38:18,154 INFO [train.py:968] (0/2) Epoch 5, batch 33900, libri_loss[loss=0.3487, simple_loss=0.3928, pruned_loss=0.1522, over 29500.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3457, pruned_loss=0.1002, over 5680239.15 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3739, pruned_loss=0.1365, over 5708570.60 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3441, pruned_loss=0.09748, over 5673558.56 frames. ], batch size: 85, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:38:35,642 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-02 21:38:56,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.890e+02 1.437e+03 1.770e+03 2.335e+03 7.676e+03, threshold=3.539e+03, percent-clipped=13.0
+2023-03-02 21:39:20,464 INFO [train.py:968] (0/2) Epoch 5, batch 33950, giga_loss[loss=0.2482, simple_loss=0.3382, pruned_loss=0.07913, over 28889.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3439, pruned_loss=0.09764, over 5679428.18 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3736, pruned_loss=0.1364, over 5709700.91 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3426, pruned_loss=0.0953, over 5672894.64 frames. ], batch size: 227, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:40:12,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:40:18,459 INFO [train.py:968] (0/2) Epoch 5, batch 34000, giga_loss[loss=0.2614, simple_loss=0.3458, pruned_loss=0.08851, over 28868.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3457, pruned_loss=0.09656, over 5684808.46 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3735, pruned_loss=0.1364, over 5712901.36 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3444, pruned_loss=0.09423, over 5676552.71 frames. ], batch size: 284, lr: 6.17e-03, grad_scale: 8.0
+2023-03-02 21:40:55,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.552e+02 1.314e+03 1.700e+03 2.490e+03 6.487e+03, threshold=3.399e+03, percent-clipped=10.0
+2023-03-02 21:41:07,048 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215941.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:41:17,039 INFO [train.py:968] (0/2) Epoch 5, batch 34050, libri_loss[loss=0.3501, simple_loss=0.3873, pruned_loss=0.1565, over 19148.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3473, pruned_loss=0.09724, over 5678397.21 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3732, pruned_loss=0.1363, over 5705537.56 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3462, pruned_loss=0.09497, over 5679150.92 frames. ], batch size: 187, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:41:28,742 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:42:23,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215998.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:42:24,918 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-216000.pt
+2023-03-02 21:42:25,210 INFO [train.py:968] (0/2) Epoch 5, batch 34100, giga_loss[loss=0.2434, simple_loss=0.305, pruned_loss=0.09088, over 24259.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3487, pruned_loss=0.09929, over 5663164.56 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.373, pruned_loss=0.1364, over 5699476.58 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3473, pruned_loss=0.09634, over 5668915.41 frames. ], batch size: 705, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:43:06,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.554e+02 1.489e+03 2.118e+03 2.684e+03 7.686e+03, threshold=4.235e+03, percent-clipped=12.0
+2023-03-02 21:43:26,968 INFO [train.py:968] (0/2) Epoch 5, batch 34150, giga_loss[loss=0.263, simple_loss=0.3452, pruned_loss=0.09037, over 28859.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3491, pruned_loss=0.09967, over 5666342.65 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3727, pruned_loss=0.1362, over 5702706.75 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3477, pruned_loss=0.09668, over 5667161.73 frames. ], batch size: 174, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:44:27,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216093.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:44:35,031 INFO [train.py:968] (0/2) Epoch 5, batch 34200, giga_loss[loss=0.2953, simple_loss=0.3615, pruned_loss=0.1145, over 27546.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3494, pruned_loss=0.09953, over 5659945.34 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3725, pruned_loss=0.1361, over 5698275.06 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3479, pruned_loss=0.0964, over 5664196.49 frames. ], batch size: 472, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:44:36,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:45:01,560 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 21:45:20,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.366e+02 1.396e+03 1.870e+03 2.666e+03 5.955e+03, threshold=3.740e+03, percent-clipped=4.0
+2023-03-02 21:45:35,692 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216141.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:45:39,933 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216144.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:45:47,566 INFO [train.py:968] (0/2) Epoch 5, batch 34250, giga_loss[loss=0.3007, simple_loss=0.3756, pruned_loss=0.1129, over 28645.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3509, pruned_loss=0.1003, over 5660081.90 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3725, pruned_loss=0.1359, over 5697196.88 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3493, pruned_loss=0.09716, over 5663621.64 frames. ], batch size: 307, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:46:16,050 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216173.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:46:50,688 INFO [train.py:968] (0/2) Epoch 5, batch 34300, libri_loss[loss=0.331, simple_loss=0.3782, pruned_loss=0.1419, over 25859.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3552, pruned_loss=0.103, over 5658399.66 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3725, pruned_loss=0.1359, over 5690489.81 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3533, pruned_loss=0.09941, over 5666845.49 frames. ], batch size: 137, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:47:18,085 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:47:30,930 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.511e+02 1.369e+03 1.833e+03 2.545e+03 5.032e+03, threshold=3.666e+03, percent-clipped=7.0
+2023-03-02 21:47:38,772 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216236.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:47:42,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216239.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:47:47,875 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:47:51,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:47:54,440 INFO [train.py:968] (0/2) Epoch 5, batch 34350, giga_loss[loss=0.2642, simple_loss=0.3404, pruned_loss=0.09403, over 29135.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.355, pruned_loss=0.1024, over 5666024.85 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3727, pruned_loss=0.136, over 5692425.07 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3529, pruned_loss=0.09875, over 5670457.21 frames. ], batch size: 113, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:48:21,527 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216268.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:48:23,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:48:32,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:49:01,512 INFO [train.py:968] (0/2) Epoch 5, batch 34400, giga_loss[loss=0.2604, simple_loss=0.3395, pruned_loss=0.09061, over 28979.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3535, pruned_loss=0.1026, over 5671112.13 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3726, pruned_loss=0.1356, over 5699501.50 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3511, pruned_loss=0.09856, over 5667167.63 frames. ], batch size: 284, lr: 6.17e-03, grad_scale: 8.0
+2023-03-02 21:49:21,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216316.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:49:30,271 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-02 21:49:43,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.917e+02 1.529e+03 1.906e+03 2.587e+03 7.661e+03, threshold=3.812e+03, percent-clipped=14.0
+2023-03-02 21:49:45,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:50:08,589 INFO [train.py:968] (0/2) Epoch 5, batch 34450, libri_loss[loss=0.3713, simple_loss=0.398, pruned_loss=0.1723, over 29561.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3505, pruned_loss=0.1011, over 5685191.01 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3726, pruned_loss=0.1357, over 5704867.52 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3481, pruned_loss=0.09697, over 5676719.86 frames. ], batch size: 76, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:51:20,732 INFO [train.py:968] (0/2) Epoch 5, batch 34500, giga_loss[loss=0.2699, simple_loss=0.3419, pruned_loss=0.09893, over 26937.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3484, pruned_loss=0.09835, over 5684366.33 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3727, pruned_loss=0.1358, over 5704870.18 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3463, pruned_loss=0.09494, over 5677660.96 frames. ], batch size: 555, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:51:29,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1908, 2.5252, 1.1204, 1.2953], device='cuda:0'), covar=tensor([0.0939, 0.0331, 0.0947, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0473, 0.0313, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 21:51:35,205 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-02 21:51:38,797 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:51:41,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:51:50,266 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:52:02,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.317e+02 1.121e+03 1.372e+03 2.006e+03 9.216e+03, threshold=2.744e+03, percent-clipped=5.0
+2023-03-02 21:52:14,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3159, 2.8834, 1.3778, 1.3665], device='cuda:0'), covar=tensor([0.0890, 0.0307, 0.0854, 0.1235], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0473, 0.0311, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 21:52:19,360 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:52:23,458 INFO [train.py:968] (0/2) Epoch 5, batch 34550, giga_loss[loss=0.2871, simple_loss=0.3589, pruned_loss=0.1077, over 27838.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3494, pruned_loss=0.09953, over 5666034.07 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3736, pruned_loss=0.1366, over 5695931.05 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3464, pruned_loss=0.09532, over 5668246.70 frames. ], batch size: 474, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:52:37,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216459.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:52:40,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216462.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:52:55,878 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:53:00,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:53:15,548 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216491.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:53:25,900 INFO [train.py:968] (0/2) Epoch 5, batch 34600, giga_loss[loss=0.3394, simple_loss=0.3999, pruned_loss=0.1394, over 28126.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.353, pruned_loss=0.1017, over 5665960.94 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.3733, pruned_loss=0.1364, over 5697043.94 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3505, pruned_loss=0.09811, over 5666296.17 frames. ], batch size: 412, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:53:35,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4633, 1.9720, 1.7618, 1.5633], device='cuda:0'), covar=tensor([0.1646, 0.1805, 0.1222, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0722, 0.0778, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-02 21:53:36,337 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:54:04,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.782e+02 1.328e+03 1.679e+03 2.601e+03 1.433e+04, threshold=3.358e+03, percent-clipped=21.0
+2023-03-02 21:54:14,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9856, 1.0745, 4.1617, 3.1296], device='cuda:0'), covar=tensor([0.1633, 0.2266, 0.0335, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.0565, 0.0521, 0.0739, 0.0594], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 21:54:24,482 INFO [train.py:968] (0/2) Epoch 5, batch 34650, giga_loss[loss=0.3018, simple_loss=0.3676, pruned_loss=0.118, over 28900.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3543, pruned_loss=0.1028, over 5670509.97 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.3731, pruned_loss=0.1364, over 5691712.80 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3521, pruned_loss=0.09911, over 5675772.63 frames. ], batch size: 186, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:55:17,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:55:21,446 INFO [train.py:968] (0/2) Epoch 5, batch 34700, giga_loss[loss=0.239, simple_loss=0.3175, pruned_loss=0.08024, over 28293.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3515, pruned_loss=0.1032, over 5655569.47 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3731, pruned_loss=0.1365, over 5686749.73 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3491, pruned_loss=0.09903, over 5664024.27 frames. ], batch size: 368, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:55:58,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.695e+03 2.391e+03 3.275e+03 6.899e+03, threshold=4.783e+03, percent-clipped=23.0
+2023-03-02 21:56:16,942 INFO [train.py:968] (0/2) Epoch 5, batch 34750, giga_loss[loss=0.2358, simple_loss=0.3227, pruned_loss=0.07449, over 28989.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3506, pruned_loss=0.1034, over 5656907.86 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.373, pruned_loss=0.1364, over 5688730.45 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3486, pruned_loss=0.09983, over 5661307.11 frames. ], batch size: 155, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:56:24,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1847, 2.4811, 1.2045, 1.2262], device='cuda:0'), covar=tensor([0.0854, 0.0349, 0.0830, 0.1323], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0470, 0.0309, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 21:56:50,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3710, 1.4852, 1.2626, 1.7256], device='cuda:0'), covar=tensor([0.2151, 0.1958, 0.2044, 0.1935], device='cuda:0'), in_proj_covar=tensor([0.1094, 0.0845, 0.0983, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 21:57:06,854 INFO [train.py:968] (0/2) Epoch 5, batch 34800, libri_loss[loss=0.337, simple_loss=0.375, pruned_loss=0.1495, over 29569.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3549, pruned_loss=0.1068, over 5669007.18 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3726, pruned_loss=0.1361, over 5697469.23 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3528, pruned_loss=0.1029, over 5663696.35 frames. ], batch size: 75, lr: 6.16e-03, grad_scale: 8.0
+2023-03-02 21:57:23,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6639, 2.5884, 1.7820, 0.8036], device='cuda:0'), covar=tensor([0.3938, 0.1848, 0.2128, 0.3319], device='cuda:0'), in_proj_covar=tensor([0.1414, 0.1342, 0.1389, 0.1168], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 21:57:36,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.308e+02 1.358e+03 1.645e+03 2.069e+03 4.159e+03, threshold=3.290e+03, percent-clipped=0.0
+2023-03-02 21:57:43,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:57:45,592 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:57:51,033 INFO [train.py:968] (0/2) Epoch 5, batch 34850, libri_loss[loss=0.3813, simple_loss=0.4132, pruned_loss=0.1746, over 19243.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3654, pruned_loss=0.1139, over 5671516.69 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.3727, pruned_loss=0.1363, over 5692808.75 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3632, pruned_loss=0.1098, over 5671509.54 frames. ], batch size: 186, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:58:12,658 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:58:33,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:58:37,348 INFO [train.py:968] (0/2) Epoch 5, batch 34900, giga_loss[loss=0.3874, simple_loss=0.4238, pruned_loss=0.1755, over 27567.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.372, pruned_loss=0.1182, over 5669574.27 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3729, pruned_loss=0.1363, over 5695521.17 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.37, pruned_loss=0.1145, over 5666448.50 frames. ], batch size: 472, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:59:05,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.300e+02 1.159e+03 1.662e+03 2.139e+03 7.337e+03, threshold=3.325e+03, percent-clipped=10.0
+2023-03-02 21:59:06,095 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-02 21:59:16,784 INFO [train.py:968] (0/2) Epoch 5, batch 34950, libri_loss[loss=0.3042, simple_loss=0.3436, pruned_loss=0.1324, over 29656.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3726, pruned_loss=0.12, over 5672614.04 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.373, pruned_loss=0.1364, over 5693374.97 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.371, pruned_loss=0.1163, over 5670237.59 frames. ], batch size: 69, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:59:39,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3710, 2.6349, 1.4564, 1.3301], device='cuda:0'), covar=tensor([0.0780, 0.0319, 0.0761, 0.1167], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0470, 0.0308, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 21:59:48,378 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7372, 1.8281, 1.3774, 1.1525], device='cuda:0'), covar=tensor([0.1285, 0.0958, 0.0823, 0.1151], device='cuda:0'), in_proj_covar=tensor([0.1438, 0.1210, 0.1190, 0.1275], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 22:00:01,318 INFO [train.py:968] (0/2) Epoch 5, batch 35000, giga_loss[loss=0.2438, simple_loss=0.3108, pruned_loss=0.0884, over 28782.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3651, pruned_loss=0.1162, over 5687125.49 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3731, pruned_loss=0.1364, over 5696221.00 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3636, pruned_loss=0.113, over 5682512.86 frames. ], batch size: 92, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:00:29,095 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.576e+02 1.015e+03 1.279e+03 1.812e+03 6.652e+03, threshold=2.557e+03, percent-clipped=3.0
+2023-03-02 22:00:34,968 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:00:36,753 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:00:43,310 INFO [train.py:968] (0/2) Epoch 5, batch 35050, giga_loss[loss=0.2293, simple_loss=0.3013, pruned_loss=0.0786, over 28613.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3584, pruned_loss=0.1135, over 5679644.42 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3732, pruned_loss=0.1362, over 5692167.09 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3569, pruned_loss=0.1105, over 5679446.38 frames. ], batch size: 71, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:01:00,938 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:01:11,121 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:01:25,168 INFO [train.py:968] (0/2) Epoch 5, batch 35100, giga_loss[loss=0.2526, simple_loss=0.3121, pruned_loss=0.09655, over 29018.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3527, pruned_loss=0.1115, over 5685982.04 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3742, pruned_loss=0.1368, over 5697866.29 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.35, pruned_loss=0.1079, over 5680269.92 frames. ], batch size: 136, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:01:52,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.385e+02 1.066e+03 1.353e+03 1.699e+03 6.035e+03, threshold=2.706e+03, percent-clipped=8.0
+2023-03-02 22:02:03,970 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217048.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:02:05,854 INFO [train.py:968] (0/2) Epoch 5, batch 35150, giga_loss[loss=0.302, simple_loss=0.3512, pruned_loss=0.1265, over 27823.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3444, pruned_loss=0.1076, over 5690603.81 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3739, pruned_loss=0.1367, over 5701790.56 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.342, pruned_loss=0.1044, over 5682216.29 frames. ], batch size: 412, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:02:08,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-02 22:02:11,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7884, 1.1323, 3.2991, 2.7833], device='cuda:0'), covar=tensor([0.2283, 0.2835, 0.0791, 0.1017], device='cuda:0'), in_proj_covar=tensor([0.0571, 0.0530, 0.0749, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 22:02:39,051 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-02 22:02:49,040 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:02:50,311 INFO [train.py:968] (0/2) Epoch 5, batch 35200, giga_loss[loss=0.239, simple_loss=0.3055, pruned_loss=0.08628, over 27664.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3377, pruned_loss=0.1041, over 5694235.53 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.374, pruned_loss=0.1365, over 5704899.14 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3353, pruned_loss=0.1012, over 5684658.36 frames. ], batch size: 472, lr: 6.16e-03, grad_scale: 8.0
+2023-03-02 22:03:19,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.128e+02 1.022e+03 1.296e+03 2.030e+03 8.864e+03, threshold=2.593e+03, percent-clipped=13.0
+2023-03-02 22:03:33,124 INFO [train.py:968] (0/2) Epoch 5, batch 35250, giga_loss[loss=0.22, simple_loss=0.2964, pruned_loss=0.07177, over 28786.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3358, pruned_loss=0.1033, over 5675910.22 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3747, pruned_loss=0.1369, over 5692688.09 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3319, pruned_loss=0.09945, over 5679493.61 frames. ], batch size: 66, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:04:15,097 INFO [train.py:968] (0/2) Epoch 5, batch 35300, libri_loss[loss=0.3021, simple_loss=0.3721, pruned_loss=0.1161, over 29545.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3329, pruned_loss=0.1012, over 5679830.17 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3752, pruned_loss=0.137, over 5683533.59 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3279, pruned_loss=0.0968, over 5691595.65 frames. ], batch size: 83, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:04:15,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6737, 2.0149, 1.9934, 1.7269], device='cuda:0'), covar=tensor([0.1413, 0.1755, 0.1093, 0.1169], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0736, 0.0789, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 22:04:39,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5910, 3.4789, 1.6838, 1.4742], device='cuda:0'), covar=tensor([0.0905, 0.0340, 0.0884, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0469, 0.0306, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 22:04:43,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.443e+02 1.052e+03 1.469e+03 2.287e+03 7.796e+03, threshold=2.939e+03, percent-clipped=21.0
+2023-03-02 22:04:57,155 INFO [train.py:968] (0/2) Epoch 5, batch 35350, giga_loss[loss=0.2491, simple_loss=0.3158, pruned_loss=0.09116, over 28954.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3309, pruned_loss=0.1005, over 5685350.09 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3755, pruned_loss=0.1372, over 5680218.77 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3253, pruned_loss=0.09578, over 5697843.09 frames. ], batch size: 213, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:05:40,501 INFO [train.py:968] (0/2) Epoch 5, batch 35400, libri_loss[loss=0.3019, simple_loss=0.3648, pruned_loss=0.1195, over 29525.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3286, pruned_loss=0.09925, over 5692803.94 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.376, pruned_loss=0.1373, over 5685883.10 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3221, pruned_loss=0.09408, over 5697869.91 frames. ], batch size: 80, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:06:07,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.292e+02 1.045e+03 1.228e+03 1.669e+03 4.560e+03, threshold=2.456e+03, percent-clipped=6.0
+2023-03-02 22:06:21,561 INFO [train.py:968] (0/2) Epoch 5, batch 35450, giga_loss[loss=0.2539, simple_loss=0.3213, pruned_loss=0.09327, over 28828.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3235, pruned_loss=0.09601, over 5695311.27 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3755, pruned_loss=0.1369, over 5689316.70 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3181, pruned_loss=0.09174, over 5696414.77 frames. ], batch size: 199, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:06:29,409 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:06:30,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:06:59,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5956, 1.9156, 1.8488, 1.6314], device='cuda:0'), covar=tensor([0.1518, 0.1848, 0.1140, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0739, 0.0793, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 22:07:04,320 INFO [train.py:968] (0/2) Epoch 5, batch 35500, giga_loss[loss=0.2189, simple_loss=0.2932, pruned_loss=0.07225, over 29032.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.32, pruned_loss=0.09411, over 5697872.73 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3756, pruned_loss=0.1368, over 5692854.19 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3148, pruned_loss=0.09003, over 5695833.54 frames. ], batch size: 155, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:07:23,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217423.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:07:30,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.531e+02 9.759e+02 1.363e+03 2.163e+03 9.426e+03, threshold=2.726e+03, percent-clipped=21.0
+2023-03-02 22:07:44,744 INFO [train.py:968] (0/2) Epoch 5, batch 35550, giga_loss[loss=0.2489, simple_loss=0.3119, pruned_loss=0.09294, over 27679.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3177, pruned_loss=0.09296, over 5698238.29 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3761, pruned_loss=0.1371, over 5698898.04 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3115, pruned_loss=0.08832, over 5691518.97 frames. ], batch size: 472, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:08:04,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:08:28,783 INFO [train.py:968] (0/2) Epoch 5, batch 35600, giga_loss[loss=0.2533, simple_loss=0.3159, pruned_loss=0.09536, over 28221.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3149, pruned_loss=0.09169, over 5702730.83 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3773, pruned_loss=0.1379, over 5701420.04 frames. ], giga_tot_loss[loss=0.2405, simple_loss=0.3079, pruned_loss=0.08653, over 5695088.68 frames. ], batch size: 368, lr: 6.15e-03, grad_scale: 8.0
+2023-03-02 22:08:30,672 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:08:34,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:09:01,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.596e+02 9.662e+02 1.250e+03 1.794e+03 6.492e+03, threshold=2.499e+03, percent-clipped=11.0
+2023-03-02 22:09:02,266 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:09:16,140 INFO [train.py:968] (0/2) Epoch 5, batch 35650, giga_loss[loss=0.3362, simple_loss=0.3973, pruned_loss=0.1376, over 28528.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3201, pruned_loss=0.0954, over 5688133.76 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3778, pruned_loss=0.1382, over 5695718.29 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3131, pruned_loss=0.09024, over 5686673.83 frames. ], batch size: 336, lr: 6.15e-03, grad_scale: 8.0
+2023-03-02 22:09:31,620 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217566.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:09:33,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217569.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:09:58,854 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217598.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:10:00,093 INFO [train.py:968] (0/2) Epoch 5, batch 35700, giga_loss[loss=0.3444, simple_loss=0.4062, pruned_loss=0.1413, over 28908.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3343, pruned_loss=0.1034, over 5690768.07 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3785, pruned_loss=0.1386, over 5696252.78 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3271, pruned_loss=0.09822, over 5689420.56 frames. ], batch size: 136, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:10:10,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2464, 4.0480, 3.8735, 1.6595], device='cuda:0'), covar=tensor([0.0411, 0.0463, 0.0663, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0760, 0.0749, 0.0573], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-02 22:10:14,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:10:16,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:10:18,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1793, 1.3306, 1.0910, 1.3152], device='cuda:0'), covar=tensor([0.2242, 0.2124, 0.2226, 0.1773], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.0873, 0.0996, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 22:10:28,121 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.95 vs. limit=5.0
+2023-03-02 22:10:30,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.533e+02 1.312e+03 1.724e+03 2.254e+03 9.436e+03, threshold=3.448e+03, percent-clipped=18.0
+2023-03-02 22:10:41,730 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:10:42,859 INFO [train.py:968] (0/2) Epoch 5, batch 35750, giga_loss[loss=0.297, simple_loss=0.3734, pruned_loss=0.1103, over 28903.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3482, pruned_loss=0.1117, over 5686606.20 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.379, pruned_loss=0.139, over 5692672.57 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3408, pruned_loss=0.1062, over 5688036.78 frames. ], batch size: 174, lr: 6.15e-03, grad_scale: 2.0
+2023-03-02 22:11:25,886 INFO [train.py:968] (0/2) Epoch 5, batch 35800, giga_loss[loss=0.2724, simple_loss=0.3435, pruned_loss=0.1007, over 28973.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3582, pruned_loss=0.1166, over 5688798.47 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3788, pruned_loss=0.1388, over 5697750.42 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3519, pruned_loss=0.1119, over 5685251.33 frames. ], batch size: 106, lr: 6.15e-03, grad_scale: 2.0
+2023-03-02 22:11:52,470 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6854, 2.0384, 2.0442, 1.8025], device='cuda:0'), covar=tensor([0.1357, 0.1854, 0.1069, 0.1259], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0730, 0.0783, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 22:11:52,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:11:54,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.860e+02 1.222e+03 1.515e+03 2.153e+03 5.381e+03, threshold=3.030e+03, percent-clipped=5.0
+2023-03-02 22:12:04,481 INFO [train.py:968] (0/2) Epoch 5, batch 35850, libri_loss[loss=0.3136, simple_loss=0.3831, pruned_loss=0.122, over 29552.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3621, pruned_loss=0.1166, over 5691732.83 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3791, pruned_loss=0.1388, over 5691368.93 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3563, pruned_loss=0.1123, over 5693535.87 frames. ], batch size: 78, lr: 6.15e-03, grad_scale: 2.0
+2023-03-02 22:12:33,590 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:12:49,419 INFO [train.py:968] (0/2) Epoch 5, batch 35900, giga_loss[loss=0.3316, simple_loss=0.3915, pruned_loss=0.1359, over 28815.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3637, pruned_loss=0.1161, over 5689934.99 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3791, pruned_loss=0.1386, over 5692965.54 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3586, pruned_loss=0.1121, over 5689676.51 frames. ], batch size: 186, lr: 6.15e-03, grad_scale: 2.0
+2023-03-02 22:13:06,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4172, 1.9096, 1.3845, 0.6174], device='cuda:0'), covar=tensor([0.2259, 0.1196, 0.1943, 0.2542], device='cuda:0'), in_proj_covar=tensor([0.1374, 0.1309, 0.1372, 0.1154], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 22:13:24,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.493e+02 1.193e+03 1.541e+03 2.077e+03 6.313e+03, threshold=3.082e+03, percent-clipped=5.0
+2023-03-02 22:13:36,471 INFO [train.py:968] (0/2) Epoch 5, batch 35950, giga_loss[loss=0.3058, simple_loss=0.375, pruned_loss=0.1182, over 28560.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3662, pruned_loss=0.1171, over 5689141.92 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3793, pruned_loss=0.1387, over 5693260.76 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3619, pruned_loss=0.1138, over 5688512.95 frames. ], batch size: 336, lr: 6.15e-03, grad_scale: 2.0
+2023-03-02 22:13:57,692 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:13:59,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2725, 4.0661, 3.9013, 1.9955], device='cuda:0'), covar=tensor([0.0471, 0.0536, 0.0731, 0.1858], device='cuda:0'), in_proj_covar=tensor([0.0834, 0.0771, 0.0771, 0.0586], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 22:13:59,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:14:04,984 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:14:09,882 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 22:14:16,689 INFO [train.py:968] (0/2) Epoch 5, batch 36000, giga_loss[loss=0.3152, simple_loss=0.3849, pruned_loss=0.1228, over 28613.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3681, pruned_loss=0.1187, over 5685539.02 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3795, pruned_loss=0.1386, over 5698487.16 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3642, pruned_loss=0.1156, over 5680046.14 frames. ], batch size: 242, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:14:16,694 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 22:14:26,800 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2315, simple_loss=0.3356, pruned_loss=0.0637, over 944034.00 frames. 
+2023-03-02 22:14:26,801 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 22:14:33,264 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:14:56,497 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.327e+02 1.137e+03 1.470e+03 1.972e+03 7.259e+03, threshold=2.940e+03, percent-clipped=4.0
+2023-03-02 22:15:06,799 INFO [train.py:968] (0/2) Epoch 5, batch 36050, giga_loss[loss=0.3281, simple_loss=0.3888, pruned_loss=0.1338, over 28940.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3716, pruned_loss=0.1213, over 5690742.64 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3797, pruned_loss=0.1386, over 5703471.80 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.368, pruned_loss=0.1184, over 5681659.99 frames. ], batch size: 227, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:15:48,588 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-218000.pt
+2023-03-02 22:15:48,982 INFO [train.py:968] (0/2) Epoch 5, batch 36100, giga_loss[loss=0.2985, simple_loss=0.3783, pruned_loss=0.1093, over 28491.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.374, pruned_loss=0.1224, over 5692051.52 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.38, pruned_loss=0.1386, over 5704636.70 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3708, pruned_loss=0.1197, over 5683402.37 frames. ], batch size: 71, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:16:16,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.369e+02 1.180e+03 1.417e+03 1.947e+03 9.280e+03, threshold=2.834e+03, percent-clipped=11.0
+2023-03-02 22:16:26,246 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-02 22:16:27,856 INFO [train.py:968] (0/2) Epoch 5, batch 36150, giga_loss[loss=0.304, simple_loss=0.3715, pruned_loss=0.1182, over 28658.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3765, pruned_loss=0.1231, over 5692341.24 frames. ], libri_tot_loss[loss=0.3292, simple_loss=0.3806, pruned_loss=0.1389, over 5703200.46 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3734, pruned_loss=0.1204, over 5686814.34 frames. ], batch size: 99, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:16:33,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:17:10,175 INFO [train.py:968] (0/2) Epoch 5, batch 36200, giga_loss[loss=0.3294, simple_loss=0.3985, pruned_loss=0.1302, over 28775.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3778, pruned_loss=0.1236, over 5679175.35 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3811, pruned_loss=0.1391, over 5700314.67 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3747, pruned_loss=0.1209, over 5676202.21 frames. ], batch size: 284, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:17:18,944 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-02 22:17:37,562 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.774e+02 1.215e+03 1.501e+03 2.172e+03 9.345e+03, threshold=3.003e+03, percent-clipped=15.0
+2023-03-02 22:17:50,453 INFO [train.py:968] (0/2) Epoch 5, batch 36250, giga_loss[loss=0.3146, simple_loss=0.3869, pruned_loss=0.1211, over 28790.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.378, pruned_loss=0.1222, over 5693988.27 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.381, pruned_loss=0.139, over 5703051.31 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3756, pruned_loss=0.1198, over 5688990.72 frames. ], batch size: 92, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:17:54,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:18:28,203 INFO [train.py:968] (0/2) Epoch 5, batch 36300, giga_loss[loss=0.3035, simple_loss=0.3789, pruned_loss=0.114, over 28754.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3786, pruned_loss=0.122, over 5699298.44 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3823, pruned_loss=0.1399, over 5701315.75 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3753, pruned_loss=0.1186, over 5697170.67 frames. ], batch size: 99, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:18:43,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1921, 1.4447, 1.1915, 1.3659], device='cuda:0'), covar=tensor([0.0881, 0.0347, 0.0354, 0.0910], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0123, 0.0127, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0059, 0.0043, 0.0038, 0.0065], device='cuda:0')
+2023-03-02 22:18:58,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.059e+02 1.113e+03 1.407e+03 1.891e+03 7.844e+03, threshold=2.814e+03, percent-clipped=13.0
+2023-03-02 22:19:09,141 INFO [train.py:968] (0/2) Epoch 5, batch 36350, giga_loss[loss=0.2666, simple_loss=0.3487, pruned_loss=0.0923, over 28736.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3751, pruned_loss=0.1185, over 5701912.10 frames. ], libri_tot_loss[loss=0.3309, simple_loss=0.3822, pruned_loss=0.1397, over 5699773.66 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3725, pruned_loss=0.1155, over 5701576.83 frames. ], batch size: 284, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:19:17,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:19:43,637 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218291.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:19:50,670 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:19:51,134 INFO [train.py:968] (0/2) Epoch 5, batch 36400, giga_loss[loss=0.307, simple_loss=0.3752, pruned_loss=0.1194, over 28845.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3752, pruned_loss=0.1187, over 5698905.81 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3828, pruned_loss=0.1401, over 5691552.75 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3724, pruned_loss=0.1157, over 5705666.88 frames. ], batch size: 99, lr: 6.14e-03, grad_scale: 8.0
+2023-03-02 22:19:52,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:20:18,537 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:20:23,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.250e+02 1.193e+03 1.390e+03 1.904e+03 5.194e+03, threshold=2.781e+03, percent-clipped=5.0
+2023-03-02 22:20:36,341 INFO [train.py:968] (0/2) Epoch 5, batch 36450, giga_loss[loss=0.4573, simple_loss=0.4593, pruned_loss=0.2277, over 28844.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3779, pruned_loss=0.1238, over 5696648.66 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.383, pruned_loss=0.1403, over 5695372.03 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3755, pruned_loss=0.1209, over 5698761.91 frames. ], batch size: 99, lr: 6.14e-03, grad_scale: 8.0
+2023-03-02 22:21:19,957 INFO [train.py:968] (0/2) Epoch 5, batch 36500, giga_loss[loss=0.3263, simple_loss=0.3796, pruned_loss=0.1365, over 28899.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3794, pruned_loss=0.1269, over 5694237.71 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.383, pruned_loss=0.1403, over 5697490.19 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3774, pruned_loss=0.1245, over 5694052.43 frames. ], batch size: 112, lr: 6.14e-03, grad_scale: 8.0
+2023-03-02 22:21:22,082 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:21:24,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:21:47,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:21:50,220 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:21:51,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.179e+02 1.175e+03 1.439e+03 2.263e+03 5.403e+03, threshold=2.878e+03, percent-clipped=19.0
+2023-03-02 22:22:01,651 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-02 22:22:02,489 INFO [train.py:968] (0/2) Epoch 5, batch 36550, giga_loss[loss=0.3718, simple_loss=0.4092, pruned_loss=0.1671, over 28737.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3791, pruned_loss=0.1277, over 5697006.16 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3841, pruned_loss=0.1411, over 5692984.99 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3764, pruned_loss=0.1247, over 5700475.11 frames. ], batch size: 284, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:22:38,914 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:22:43,509 INFO [train.py:968] (0/2) Epoch 5, batch 36600, giga_loss[loss=0.2873, simple_loss=0.3522, pruned_loss=0.1112, over 28547.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3767, pruned_loss=0.1264, over 5696520.25 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3848, pruned_loss=0.1413, over 5696463.32 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3736, pruned_loss=0.1232, over 5696236.80 frames. ], batch size: 65, lr: 6.14e-03, grad_scale: 2.0
+2023-03-02 22:23:16,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.975e+02 1.099e+03 1.367e+03 2.078e+03 1.521e+04, threshold=2.734e+03, percent-clipped=18.0
+2023-03-02 22:23:26,238 INFO [train.py:968] (0/2) Epoch 5, batch 36650, giga_loss[loss=0.3028, simple_loss=0.367, pruned_loss=0.1193, over 28801.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3769, pruned_loss=0.1263, over 5701381.87 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3851, pruned_loss=0.1415, over 5700570.15 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.374, pruned_loss=0.1234, over 5697733.22 frames. ], batch size: 99, lr: 6.14e-03, grad_scale: 2.0
+2023-03-02 22:23:48,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:23:48,302 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:23:52,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:24:04,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-02 22:24:05,375 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-02 22:24:10,652 INFO [train.py:968] (0/2) Epoch 5, batch 36700, giga_loss[loss=0.309, simple_loss=0.3681, pruned_loss=0.125, over 28478.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3755, pruned_loss=0.1244, over 5700026.59 frames. ], libri_tot_loss[loss=0.3344, simple_loss=0.3855, pruned_loss=0.1416, over 5702472.29 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3728, pruned_loss=0.1218, over 5695479.02 frames. ], batch size: 60, lr: 6.14e-03, grad_scale: 2.0
+2023-03-02 22:24:15,555 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:24:18,054 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:24:42,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7253, 1.1607, 3.4396, 2.8331], device='cuda:0'), covar=tensor([0.1708, 0.2205, 0.0405, 0.0732], device='cuda:0'), in_proj_covar=tensor([0.0569, 0.0529, 0.0742, 0.0609], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 22:24:45,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.503e+02 1.064e+03 1.414e+03 1.821e+03 3.605e+03, threshold=2.829e+03, percent-clipped=5.0
+2023-03-02 22:24:49,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2291, 1.6178, 1.1928, 0.5019], device='cuda:0'), covar=tensor([0.1682, 0.0891, 0.1459, 0.2667], device='cuda:0'), in_proj_covar=tensor([0.1396, 0.1326, 0.1371, 0.1171], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 22:24:59,871 INFO [train.py:968] (0/2) Epoch 5, batch 36750, giga_loss[loss=0.2771, simple_loss=0.3502, pruned_loss=0.102, over 28461.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3718, pruned_loss=0.1218, over 5688930.21 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3853, pruned_loss=0.1414, over 5705325.78 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3696, pruned_loss=0.1196, over 5682759.90 frames. ], batch size: 85, lr: 6.13e-03, grad_scale: 2.0
+2023-03-02 22:25:05,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8388, 1.2141, 3.5769, 2.9238], device='cuda:0'), covar=tensor([0.1678, 0.2213, 0.0368, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0566, 0.0528, 0.0737, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 22:25:14,643 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:25:47,689 INFO [train.py:968] (0/2) Epoch 5, batch 36800, giga_loss[loss=0.2793, simple_loss=0.3475, pruned_loss=0.1056, over 28930.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3645, pruned_loss=0.1176, over 5676497.41 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3853, pruned_loss=0.1413, over 5708308.95 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3625, pruned_loss=0.1157, over 5668732.37 frames. ], batch size: 227, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:26:20,148 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-02 22:26:24,646 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.568e+02 9.112e+02 1.217e+03 1.861e+03 8.973e+03, threshold=2.434e+03, percent-clipped=11.0
+2023-03-02 22:26:26,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1509, 1.9058, 1.4669, 1.6929], device='cuda:0'), covar=tensor([0.0621, 0.0718, 0.0946, 0.1043], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0443, 0.0503, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 22:26:37,243 INFO [train.py:968] (0/2) Epoch 5, batch 36850, giga_loss[loss=0.2437, simple_loss=0.3141, pruned_loss=0.08661, over 29018.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3589, pruned_loss=0.115, over 5664918.71 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.386, pruned_loss=0.1419, over 5708477.00 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3563, pruned_loss=0.1125, over 5657939.63 frames. ], batch size: 128, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:27:27,336 INFO [train.py:968] (0/2) Epoch 5, batch 36900, giga_loss[loss=0.3587, simple_loss=0.3917, pruned_loss=0.1628, over 26509.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3552, pruned_loss=0.113, over 5661363.44 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3859, pruned_loss=0.1417, over 5712121.18 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3527, pruned_loss=0.1107, over 5651903.08 frames. ], batch size: 555, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:27:28,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6221, 1.9942, 1.9324, 1.6781], device='cuda:0'), covar=tensor([0.1623, 0.1851, 0.1120, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0736, 0.0783, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 22:27:33,987 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218809.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:27:36,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218812.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:27:59,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.456e+02 9.801e+02 1.296e+03 1.969e+03 6.389e+03, threshold=2.592e+03, percent-clipped=16.0
+2023-03-02 22:28:02,856 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218841.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:28:09,459 INFO [train.py:968] (0/2) Epoch 5, batch 36950, giga_loss[loss=0.2782, simple_loss=0.3468, pruned_loss=0.1048, over 28826.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3554, pruned_loss=0.112, over 5671299.20 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3859, pruned_loss=0.1413, over 5708315.34 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3523, pruned_loss=0.1096, over 5666371.48 frames. ], batch size: 199, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:28:25,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:28:51,124 INFO [train.py:968] (0/2) Epoch 5, batch 37000, giga_loss[loss=0.2915, simple_loss=0.3464, pruned_loss=0.1183, over 28723.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3535, pruned_loss=0.1106, over 5671296.78 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.386, pruned_loss=0.1413, over 5706448.05 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3507, pruned_loss=0.1085, over 5668910.24 frames. ], batch size: 92, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:28:52,521 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:29:24,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.239e+02 9.644e+02 1.227e+03 1.613e+03 4.966e+03, threshold=2.454e+03, percent-clipped=9.0
+2023-03-02 22:29:34,351 INFO [train.py:968] (0/2) Epoch 5, batch 37050, libri_loss[loss=0.4618, simple_loss=0.4889, pruned_loss=0.2174, over 19035.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3508, pruned_loss=0.1087, over 5677654.73 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3865, pruned_loss=0.1416, over 5698165.46 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3478, pruned_loss=0.1065, over 5684069.74 frames. ], batch size: 186, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:29:34,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:29:46,535 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:29:55,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:30:11,050 INFO [train.py:968] (0/2) Epoch 5, batch 37100, giga_loss[loss=0.2752, simple_loss=0.3416, pruned_loss=0.1044, over 28955.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3506, pruned_loss=0.1092, over 5699366.37 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3879, pruned_loss=0.1423, over 5704343.86 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3457, pruned_loss=0.1055, over 5698573.74 frames. ], batch size: 136, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:30:20,882 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:30:22,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:30:41,818 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.024e+02 1.010e+03 1.249e+03 1.755e+03 6.699e+03, threshold=2.499e+03, percent-clipped=12.0
+2023-03-02 22:30:44,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8379, 5.6495, 5.4276, 2.6746], device='cuda:0'), covar=tensor([0.0295, 0.0318, 0.0558, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.0841, 0.0776, 0.0768, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 22:30:46,956 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:30:51,204 INFO [train.py:968] (0/2) Epoch 5, batch 37150, giga_loss[loss=0.2519, simple_loss=0.3291, pruned_loss=0.08741, over 29036.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3469, pruned_loss=0.1075, over 5702943.11 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3879, pruned_loss=0.1422, over 5707514.73 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3425, pruned_loss=0.1042, over 5699438.64 frames. ], batch size: 164, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:31:26,458 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:31:29,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:31:31,269 INFO [train.py:968] (0/2) Epoch 5, batch 37200, libri_loss[loss=0.4309, simple_loss=0.4712, pruned_loss=0.1953, over 29353.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3445, pruned_loss=0.1062, over 5706474.85 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3889, pruned_loss=0.1427, over 5709930.92 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3395, pruned_loss=0.1025, over 5701380.50 frames. ], batch size: 92, lr: 6.13e-03, grad_scale: 8.0
+2023-03-02 22:31:48,320 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:31:51,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:31:53,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:32:01,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.365e+02 9.820e+02 1.333e+03 2.009e+03 9.149e+03, threshold=2.666e+03, percent-clipped=13.0
+2023-03-02 22:32:11,008 INFO [train.py:968] (0/2) Epoch 5, batch 37250, libri_loss[loss=0.3973, simple_loss=0.45, pruned_loss=0.1724, over 28658.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3431, pruned_loss=0.1055, over 5718343.80 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3894, pruned_loss=0.1428, over 5713142.77 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3376, pruned_loss=0.1018, over 5711400.93 frames. ], batch size: 106, lr: 6.13e-03, grad_scale: 8.0
+2023-03-02 22:32:13,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:32:48,985 INFO [train.py:968] (0/2) Epoch 5, batch 37300, giga_loss[loss=0.2568, simple_loss=0.3219, pruned_loss=0.09583, over 28673.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3432, pruned_loss=0.1056, over 5715879.54 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.3903, pruned_loss=0.143, over 5710261.80 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3361, pruned_loss=0.101, over 5712862.17 frames. ], batch size: 60, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:33:21,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.311e+02 1.043e+03 1.397e+03 1.853e+03 5.381e+03, threshold=2.794e+03, percent-clipped=8.0
+2023-03-02 22:33:27,320 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 22:33:31,242 INFO [train.py:968] (0/2) Epoch 5, batch 37350, giga_loss[loss=0.2198, simple_loss=0.3005, pruned_loss=0.06953, over 28981.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3401, pruned_loss=0.1037, over 5709429.21 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3914, pruned_loss=0.1435, over 5702796.77 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3327, pruned_loss=0.0989, over 5713574.07 frames. ], batch size: 136, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:33:50,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:34:09,895 INFO [train.py:968] (0/2) Epoch 5, batch 37400, giga_loss[loss=0.2319, simple_loss=0.3028, pruned_loss=0.08055, over 28387.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3402, pruned_loss=0.1034, over 5702990.09 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.3929, pruned_loss=0.1443, over 5698623.89 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.331, pruned_loss=0.0974, over 5710822.35 frames. ], batch size: 78, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:34:27,758 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-02 22:34:30,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8670, 5.6384, 5.4113, 2.4046], device='cuda:0'), covar=tensor([0.0336, 0.0470, 0.0710, 0.1625], device='cuda:0'), in_proj_covar=tensor([0.0846, 0.0781, 0.0776, 0.0597], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 22:34:39,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.091e+02 9.497e+02 1.204e+03 1.677e+03 4.915e+03, threshold=2.409e+03, percent-clipped=7.0
+2023-03-02 22:34:41,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:34:47,726 INFO [train.py:968] (0/2) Epoch 5, batch 37450, giga_loss[loss=0.2644, simple_loss=0.3284, pruned_loss=0.1002, over 28723.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3388, pruned_loss=0.1025, over 5702969.98 frames. ], libri_tot_loss[loss=0.3409, simple_loss=0.3933, pruned_loss=0.1443, over 5700370.13 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3299, pruned_loss=0.09671, over 5707807.95 frames. ], batch size: 99, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:34:55,622 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:35:09,519 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:35:18,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7998, 1.0338, 3.7074, 2.8828], device='cuda:0'), covar=tensor([0.1856, 0.2502, 0.0390, 0.0837], device='cuda:0'), in_proj_covar=tensor([0.0578, 0.0537, 0.0756, 0.0615], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 22:35:29,127 INFO [train.py:968] (0/2) Epoch 5, batch 37500, giga_loss[loss=0.2969, simple_loss=0.3616, pruned_loss=0.1161, over 28925.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3396, pruned_loss=0.1028, over 5715006.48 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3932, pruned_loss=0.1441, over 5705442.54 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3313, pruned_loss=0.09743, over 5714607.83 frames. ], batch size: 145, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:35:30,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6490, 2.2739, 1.5069, 0.9133], device='cuda:0'), covar=tensor([0.2743, 0.1289, 0.2618, 0.2918], device='cuda:0'), in_proj_covar=tensor([0.1392, 0.1301, 0.1367, 0.1153], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 22:35:30,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3794, 1.9676, 1.3816, 0.5782], device='cuda:0'), covar=tensor([0.2430, 0.1232, 0.2288, 0.2883], device='cuda:0'), in_proj_covar=tensor([0.1392, 0.1302, 0.1367, 0.1153], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 22:35:42,019 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:35:46,336 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:35:51,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:36:04,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.551e+02 1.008e+03 1.388e+03 2.505e+03 6.682e+03, threshold=2.775e+03, percent-clipped=27.0
+2023-03-02 22:36:14,684 INFO [train.py:968] (0/2) Epoch 5, batch 37550, giga_loss[loss=0.3501, simple_loss=0.3875, pruned_loss=0.1564, over 23638.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3451, pruned_loss=0.1068, over 5704984.55 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3931, pruned_loss=0.1441, over 5699681.52 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3377, pruned_loss=0.1019, over 5709136.75 frames. ], batch size: 705, lr: 6.12e-03, grad_scale: 2.0
+2023-03-02 22:36:16,339 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:36:16,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0382, 1.1509, 1.2373, 1.0722], device='cuda:0'), covar=tensor([0.1212, 0.1171, 0.1630, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0747, 0.0643, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 22:36:28,111 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:36:46,999 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:36:49,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:37:01,840 INFO [train.py:968] (0/2) Epoch 5, batch 37600, giga_loss[loss=0.3628, simple_loss=0.4108, pruned_loss=0.1574, over 28873.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3532, pruned_loss=0.1123, over 5701132.44 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3931, pruned_loss=0.1441, over 5700254.87 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3465, pruned_loss=0.1078, over 5704357.81 frames. ], batch size: 227, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:37:20,598 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:37:42,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.312e+02 1.112e+03 1.599e+03 2.242e+03 1.179e+04, threshold=3.197e+03, percent-clipped=17.0
+2023-03-02 22:37:51,477 INFO [train.py:968] (0/2) Epoch 5, batch 37650, giga_loss[loss=0.2936, simple_loss=0.3568, pruned_loss=0.1152, over 28285.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.362, pruned_loss=0.1188, over 5684764.30 frames. ], libri_tot_loss[loss=0.3409, simple_loss=0.3933, pruned_loss=0.1442, over 5690223.91 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3564, pruned_loss=0.1149, over 5695460.95 frames. ], batch size: 65, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:38:12,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4642, 1.8348, 1.7887, 1.6506], device='cuda:0'), covar=tensor([0.1602, 0.1820, 0.1208, 0.1342], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0745, 0.0789, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 22:38:36,172 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:38:40,748 INFO [train.py:968] (0/2) Epoch 5, batch 37700, giga_loss[loss=0.328, simple_loss=0.3958, pruned_loss=0.1301, over 28928.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3676, pruned_loss=0.1216, over 5680194.71 frames. ], libri_tot_loss[loss=0.341, simple_loss=0.3934, pruned_loss=0.1443, over 5697126.28 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3622, pruned_loss=0.1178, over 5682447.19 frames. ], batch size: 164, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:38:49,409 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6904, 2.3306, 1.5239, 1.2559], device='cuda:0'), covar=tensor([0.1574, 0.0816, 0.0991, 0.1262], device='cuda:0'), in_proj_covar=tensor([0.1445, 0.1228, 0.1221, 0.1310], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 22:39:14,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.334e+02 9.851e+02 1.302e+03 1.838e+03 4.720e+03, threshold=2.603e+03, percent-clipped=2.0
+2023-03-02 22:39:23,214 INFO [train.py:968] (0/2) Epoch 5, batch 37750, giga_loss[loss=0.3258, simple_loss=0.3967, pruned_loss=0.1274, over 29050.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.373, pruned_loss=0.1239, over 5681815.82 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.3938, pruned_loss=0.1446, over 5699642.48 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3676, pruned_loss=0.12, over 5680997.42 frames. ], batch size: 155, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:40:09,144 INFO [train.py:968] (0/2) Epoch 5, batch 37800, giga_loss[loss=0.2964, simple_loss=0.3697, pruned_loss=0.1115, over 28892.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3783, pruned_loss=0.127, over 5685021.07 frames. ], libri_tot_loss[loss=0.3411, simple_loss=0.3935, pruned_loss=0.1444, over 5701541.76 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3741, pruned_loss=0.1239, over 5682572.83 frames. ], batch size: 227, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:40:37,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:40:39,332 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219735.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:40:43,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.430e+02 1.063e+03 1.386e+03 2.182e+03 1.438e+04, threshold=2.773e+03, percent-clipped=16.0
+2023-03-02 22:40:53,426 INFO [train.py:968] (0/2) Epoch 5, batch 37850, giga_loss[loss=0.2912, simple_loss=0.3615, pruned_loss=0.1104, over 28471.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1234, over 5683587.26 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3936, pruned_loss=0.1444, over 5702695.05 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3702, pruned_loss=0.1209, over 5680683.28 frames. ], batch size: 71, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:40:54,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:40:56,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6576, 1.9406, 1.4039, 1.5427], device='cuda:0'), covar=tensor([0.0763, 0.0271, 0.0328, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0121, 0.0125, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0058, 0.0042, 0.0038, 0.0064], device='cuda:0')
+2023-03-02 22:41:05,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9496, 1.1203, 0.8173, 0.6595], device='cuda:0'), covar=tensor([0.0984, 0.0959, 0.0742, 0.0931], device='cuda:0'), in_proj_covar=tensor([0.1409, 0.1200, 0.1192, 0.1281], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 22:41:26,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:41:38,514 INFO [train.py:968] (0/2) Epoch 5, batch 37900, giga_loss[loss=0.2746, simple_loss=0.3463, pruned_loss=0.1015, over 28680.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.369, pruned_loss=0.1192, over 5692009.23 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3937, pruned_loss=0.1445, over 5703643.57 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3661, pruned_loss=0.1171, over 5688901.65 frames. ], batch size: 284, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:41:42,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6921, 5.4691, 5.2768, 2.3992], device='cuda:0'), covar=tensor([0.0340, 0.0357, 0.0576, 0.2004], device='cuda:0'), in_proj_covar=tensor([0.0838, 0.0776, 0.0764, 0.0589], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 22:42:10,708 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.782e+02 1.180e+03 1.541e+03 2.084e+03 7.453e+03, threshold=3.082e+03, percent-clipped=11.0
+2023-03-02 22:42:10,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:42:19,722 INFO [train.py:968] (0/2) Epoch 5, batch 37950, giga_loss[loss=0.281, simple_loss=0.352, pruned_loss=0.105, over 28863.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3688, pruned_loss=0.1186, over 5697085.45 frames. ], libri_tot_loss[loss=0.3416, simple_loss=0.394, pruned_loss=0.1446, over 5705549.18 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3654, pruned_loss=0.1159, over 5692542.57 frames. ], batch size: 227, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:42:41,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:42:44,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:42:47,271 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:42:48,963 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1898, 2.4312, 1.2443, 1.3034], device='cuda:0'), covar=tensor([0.0868, 0.0348, 0.0807, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0468, 0.0304, 0.0375], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 22:42:59,000 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:02,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:03,108 INFO [train.py:968] (0/2) Epoch 5, batch 38000, giga_loss[loss=0.3326, simple_loss=0.3952, pruned_loss=0.135, over 28980.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3693, pruned_loss=0.1183, over 5698734.19 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.3939, pruned_loss=0.1445, over 5706621.43 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3666, pruned_loss=0.1162, over 5694236.96 frames. ], batch size: 136, lr: 6.12e-03, grad_scale: 8.0
+2023-03-02 22:43:11,243 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:28,678 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-02 22:43:29,074 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:32,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:35,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:39,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.523e+02 1.083e+03 1.398e+03 1.837e+03 4.284e+03, threshold=2.795e+03, percent-clipped=1.0
+2023-03-02 22:43:46,463 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1099, 1.1890, 4.2404, 3.2257], device='cuda:0'), covar=tensor([0.1618, 0.2327, 0.0345, 0.0623], device='cuda:0'), in_proj_covar=tensor([0.0570, 0.0527, 0.0741, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 22:43:46,931 INFO [train.py:968] (0/2) Epoch 5, batch 38050, giga_loss[loss=0.2914, simple_loss=0.3616, pruned_loss=0.1106, over 28351.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3726, pruned_loss=0.1202, over 5704814.30 frames. ], libri_tot_loss[loss=0.3416, simple_loss=0.3941, pruned_loss=0.1446, over 5708583.23 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3701, pruned_loss=0.1182, over 5699475.70 frames. ], batch size: 65, lr: 6.12e-03, grad_scale: 8.0
+2023-03-02 22:44:00,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:44:06,549 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:44:09,756 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1297, 1.4050, 4.1833, 3.1121], device='cuda:0'), covar=tensor([0.1586, 0.2111, 0.0320, 0.0686], device='cuda:0'), in_proj_covar=tensor([0.0568, 0.0525, 0.0739, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-02 22:44:15,761 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:44:18,119 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:44:33,191 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-220000.pt
+2023-03-02 22:44:33,498 INFO [train.py:968] (0/2) Epoch 5, batch 38100, giga_loss[loss=0.3126, simple_loss=0.3804, pruned_loss=0.1224, over 28607.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3754, pruned_loss=0.1225, over 5698504.56 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3948, pruned_loss=0.1451, over 5709099.70 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3723, pruned_loss=0.1201, over 5693671.55 frames. ], batch size: 336, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:44:46,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:44:53,876 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-02 22:45:07,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.374e+02 1.280e+03 1.708e+03 2.330e+03 5.235e+03, threshold=3.416e+03, percent-clipped=15.0
+2023-03-02 22:45:14,114 INFO [train.py:968] (0/2) Epoch 5, batch 38150, giga_loss[loss=0.3187, simple_loss=0.383, pruned_loss=0.1272, over 28800.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3773, pruned_loss=0.1242, over 5703852.28 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3959, pruned_loss=0.1458, over 5713851.49 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3735, pruned_loss=0.1211, over 5695558.58 frames. ], batch size: 284, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:45:32,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6354, 4.4580, 4.2824, 1.8819], device='cuda:0'), covar=tensor([0.0371, 0.0423, 0.0582, 0.2091], device='cuda:0'), in_proj_covar=tensor([0.0847, 0.0790, 0.0774, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 22:45:58,956 INFO [train.py:968] (0/2) Epoch 5, batch 38200, giga_loss[loss=0.3022, simple_loss=0.3664, pruned_loss=0.119, over 28869.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3785, pruned_loss=0.1259, over 5696928.80 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.3961, pruned_loss=0.1458, over 5709308.53 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3749, pruned_loss=0.1229, over 5693892.44 frames. ], batch size: 186, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:46:04,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5612, 1.8294, 1.4569, 1.2881], device='cuda:0'), covar=tensor([0.1226, 0.0856, 0.0756, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.1444, 0.1233, 0.1224, 0.1311], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 22:46:06,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:46:09,791 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:46:11,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:46:13,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3939, 1.6440, 1.3060, 1.4993], device='cuda:0'), covar=tensor([0.0792, 0.0319, 0.0335, 0.0850], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0122, 0.0125, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0058, 0.0043, 0.0038, 0.0065], device='cuda:0')
+2023-03-02 22:46:13,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3091, 4.1481, 3.9748, 1.8569], device='cuda:0'), covar=tensor([0.0503, 0.0541, 0.0734, 0.2207], device='cuda:0'), in_proj_covar=tensor([0.0852, 0.0796, 0.0779, 0.0599], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 22:46:34,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.680e+02 1.228e+03 1.533e+03 2.332e+03 6.114e+03, threshold=3.066e+03, percent-clipped=9.0
+2023-03-02 22:46:38,343 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:46:40,642 INFO [train.py:968] (0/2) Epoch 5, batch 38250, giga_loss[loss=0.304, simple_loss=0.3746, pruned_loss=0.1166, over 28669.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3789, pruned_loss=0.1263, over 5687186.34 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3962, pruned_loss=0.1459, over 5703157.99 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3755, pruned_loss=0.1234, over 5690442.19 frames. ], batch size: 92, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:46:44,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:47:18,240 INFO [train.py:968] (0/2) Epoch 5, batch 38300, giga_loss[loss=0.2822, simple_loss=0.3682, pruned_loss=0.09813, over 28713.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3789, pruned_loss=0.1255, over 5690821.43 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.396, pruned_loss=0.1457, over 5700618.49 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3757, pruned_loss=0.1226, over 5694974.61 frames. ], batch size: 242, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:47:41,314 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 22:47:54,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.520e+02 1.157e+03 1.519e+03 2.201e+03 9.466e+03, threshold=3.037e+03, percent-clipped=15.0
+2023-03-02 22:47:57,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 22:48:00,843 INFO [train.py:968] (0/2) Epoch 5, batch 38350, giga_loss[loss=0.261, simple_loss=0.3445, pruned_loss=0.08875, over 28514.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3783, pruned_loss=0.1238, over 5694994.23 frames. ], libri_tot_loss[loss=0.3435, simple_loss=0.3959, pruned_loss=0.1456, over 5701064.02 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3756, pruned_loss=0.1212, over 5697507.11 frames. ], batch size: 65, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:48:01,018 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:48:03,493 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220253.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:48:05,362 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220256.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:48:07,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:48:29,252 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220285.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:48:41,340 INFO [train.py:968] (0/2) Epoch 5, batch 38400, giga_loss[loss=0.2752, simple_loss=0.3533, pruned_loss=0.09852, over 28878.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3784, pruned_loss=0.1233, over 5697195.81 frames. ], libri_tot_loss[loss=0.3443, simple_loss=0.3965, pruned_loss=0.1461, over 5699464.83 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3754, pruned_loss=0.1205, over 5700545.91 frames. ], batch size: 145, lr: 6.11e-03, grad_scale: 8.0
+2023-03-02 22:49:15,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.103e+02 1.041e+03 1.268e+03 1.668e+03 8.935e+03, threshold=2.536e+03, percent-clipped=9.0
+2023-03-02 22:49:24,914 INFO [train.py:968] (0/2) Epoch 5, batch 38450, giga_loss[loss=0.2674, simple_loss=0.3434, pruned_loss=0.09569, over 28834.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3758, pruned_loss=0.1221, over 5702297.22 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3968, pruned_loss=0.1462, over 5705720.50 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3726, pruned_loss=0.119, over 5699174.07 frames. ], batch size: 99, lr: 6.11e-03, grad_scale: 8.0
+2023-03-02 22:49:35,640 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 22:50:04,822 INFO [train.py:968] (0/2) Epoch 5, batch 38500, giga_loss[loss=0.2988, simple_loss=0.3699, pruned_loss=0.1138, over 28638.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3725, pruned_loss=0.12, over 5704463.70 frames. ], libri_tot_loss[loss=0.3444, simple_loss=0.3967, pruned_loss=0.1461, over 5709521.65 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3697, pruned_loss=0.1172, over 5698356.46 frames. ], batch size: 336, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:50:05,817 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:50:08,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:50:31,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:50:37,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.991e+02 1.041e+03 1.247e+03 1.748e+03 5.344e+03, threshold=2.494e+03, percent-clipped=9.0
+2023-03-02 22:50:46,370 INFO [train.py:968] (0/2) Epoch 5, batch 38550, giga_loss[loss=0.29, simple_loss=0.3656, pruned_loss=0.1072, over 28777.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3714, pruned_loss=0.1194, over 5707738.34 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.3969, pruned_loss=0.1462, over 5711241.94 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3688, pruned_loss=0.1169, over 5701438.71 frames. ], batch size: 119, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:51:11,116 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:51:19,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:51:27,135 INFO [train.py:968] (0/2) Epoch 5, batch 38600, giga_loss[loss=0.3419, simple_loss=0.3988, pruned_loss=0.1425, over 29010.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3729, pruned_loss=0.1209, over 5710609.08 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3974, pruned_loss=0.1467, over 5715148.52 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3696, pruned_loss=0.1176, over 5701607.03 frames. ], batch size: 136, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:51:47,612 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 22:51:52,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:52:01,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.659e+02 9.861e+02 1.207e+03 1.668e+03 6.949e+03, threshold=2.414e+03, percent-clipped=6.0
+2023-03-02 22:52:08,194 INFO [train.py:968] (0/2) Epoch 5, batch 38650, giga_loss[loss=0.3021, simple_loss=0.3714, pruned_loss=0.1164, over 28925.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3728, pruned_loss=0.1206, over 5705130.30 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3978, pruned_loss=0.147, over 5710976.46 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3694, pruned_loss=0.1174, over 5701988.44 frames. ], batch size: 199, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:52:35,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5263, 1.6089, 1.4022, 1.2610], device='cuda:0'), covar=tensor([0.1064, 0.0839, 0.0702, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.1400, 0.1216, 0.1210, 0.1304], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 22:52:45,702 INFO [train.py:968] (0/2) Epoch 5, batch 38700, libri_loss[loss=0.3492, simple_loss=0.4044, pruned_loss=0.1471, over 29544.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3718, pruned_loss=0.119, over 5708337.98 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3972, pruned_loss=0.1465, over 5712409.51 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3693, pruned_loss=0.1165, over 5704511.70 frames. ], batch size: 81, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:52:47,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4977, 1.8899, 1.8253, 1.6354], device='cuda:0'), covar=tensor([0.1636, 0.1869, 0.1222, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0746, 0.0787, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 22:53:02,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:53:04,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:53:16,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.925e+02 1.011e+03 1.265e+03 1.697e+03 4.187e+03, threshold=2.531e+03, percent-clipped=8.0
+2023-03-02 22:53:23,289 INFO [train.py:968] (0/2) Epoch 5, batch 38750, giga_loss[loss=0.2975, simple_loss=0.3666, pruned_loss=0.1142, over 28744.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3708, pruned_loss=0.1177, over 5717710.65 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.3971, pruned_loss=0.1463, over 5718387.91 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3682, pruned_loss=0.115, over 5709343.44 frames. ], batch size: 284, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:53:24,184 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:53:25,854 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-02 22:53:42,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:53:44,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:54:06,187 INFO [train.py:968] (0/2) Epoch 5, batch 38800, giga_loss[loss=0.2947, simple_loss=0.3605, pruned_loss=0.1144, over 28772.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3702, pruned_loss=0.1177, over 5711024.19 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.397, pruned_loss=0.1464, over 5719747.77 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3678, pruned_loss=0.1152, over 5703238.91 frames. ], batch size: 119, lr: 6.11e-03, grad_scale: 8.0
+2023-03-02 22:54:10,189 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:54:40,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.482e+02 9.618e+02 1.208e+03 1.957e+03 1.086e+04, threshold=2.417e+03, percent-clipped=17.0
+2023-03-02 22:54:46,630 INFO [train.py:968] (0/2) Epoch 5, batch 38850, giga_loss[loss=0.243, simple_loss=0.3243, pruned_loss=0.08086, over 28920.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3671, pruned_loss=0.116, over 5714412.57 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.3971, pruned_loss=0.1463, over 5722504.66 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3647, pruned_loss=0.1137, over 5705729.43 frames. ], batch size: 174, lr: 6.11e-03, grad_scale: 8.0
+2023-03-02 22:55:03,383 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:55:06,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:55:15,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3527, 2.9988, 1.4709, 1.4300], device='cuda:0'), covar=tensor([0.0886, 0.0261, 0.0821, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0329, 0.0459, 0.0302, 0.0370], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:0')
+2023-03-02 22:55:28,198 INFO [train.py:968] (0/2) Epoch 5, batch 38900, giga_loss[loss=0.2484, simple_loss=0.3306, pruned_loss=0.08307, over 28914.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3636, pruned_loss=0.1144, over 5711788.73 frames. ], libri_tot_loss[loss=0.3448, simple_loss=0.397, pruned_loss=0.1463, over 5727694.53 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3608, pruned_loss=0.1116, over 5699562.66 frames. ], batch size: 174, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 22:55:28,456 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:56:00,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.046e+02 1.156e+03 1.325e+03 1.829e+03 5.711e+03, threshold=2.650e+03, percent-clipped=14.0
+2023-03-02 22:56:07,114 INFO [train.py:968] (0/2) Epoch 5, batch 38950, giga_loss[loss=0.3102, simple_loss=0.3673, pruned_loss=0.1266, over 28789.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3619, pruned_loss=0.1135, over 5716997.57 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.3971, pruned_loss=0.1463, over 5730079.63 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3592, pruned_loss=0.1109, over 5704983.56 frames. ], batch size: 86, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 22:56:12,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:56:22,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:56:48,317 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-02 22:56:51,052 INFO [train.py:968] (0/2) Epoch 5, batch 39000, giga_loss[loss=0.2547, simple_loss=0.3302, pruned_loss=0.08963, over 28407.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3613, pruned_loss=0.1136, over 5710352.63 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3977, pruned_loss=0.1467, over 5722725.60 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3585, pruned_loss=0.1109, over 5707484.47 frames. ], batch size: 65, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 22:56:51,057 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 22:57:00,202 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2426, simple_loss=0.3456, pruned_loss=0.0698, over 944034.00 frames. 
+2023-03-02 22:57:00,203 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 22:57:31,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.881e+02 1.024e+03 1.271e+03 1.666e+03 3.947e+03, threshold=2.543e+03, percent-clipped=5.0
+2023-03-02 22:57:36,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:57:38,376 INFO [train.py:968] (0/2) Epoch 5, batch 39050, giga_loss[loss=0.3679, simple_loss=0.401, pruned_loss=0.1675, over 26773.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3609, pruned_loss=0.1144, over 5718844.66 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.397, pruned_loss=0.1462, over 5728606.43 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.358, pruned_loss=0.1116, over 5710758.05 frames. ], batch size: 555, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 22:57:56,021 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-02 22:58:16,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:17,335 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:18,493 INFO [train.py:968] (0/2) Epoch 5, batch 39100, giga_loss[loss=0.3234, simple_loss=0.3819, pruned_loss=0.1325, over 27896.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.358, pruned_loss=0.1132, over 5714060.38 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.3969, pruned_loss=0.1462, over 5729683.34 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3553, pruned_loss=0.1106, over 5706705.54 frames. ], batch size: 412, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 22:58:19,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:25,523 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:27,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:37,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:40,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:44,966 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-02 22:58:49,226 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:51,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.183e+02 1.011e+03 1.243e+03 1.639e+03 5.943e+03, threshold=2.487e+03, percent-clipped=9.0
+2023-03-02 22:58:51,830 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:56,582 INFO [train.py:968] (0/2) Epoch 5, batch 39150, giga_loss[loss=0.3018, simple_loss=0.3622, pruned_loss=0.1207, over 28224.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3559, pruned_loss=0.1121, over 5707956.05 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3974, pruned_loss=0.1464, over 5722706.22 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3524, pruned_loss=0.1092, over 5707080.59 frames. ], batch size: 65, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 22:59:32,399 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-02 22:59:36,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3483, 4.1389, 3.9790, 2.0015], device='cuda:0'), covar=tensor([0.0432, 0.0496, 0.0684, 0.1908], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0793, 0.0781, 0.0592], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 22:59:40,279 INFO [train.py:968] (0/2) Epoch 5, batch 39200, giga_loss[loss=0.2821, simple_loss=0.3426, pruned_loss=0.1108, over 28679.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3552, pruned_loss=0.1122, over 5699498.72 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3975, pruned_loss=0.1464, over 5716388.56 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3517, pruned_loss=0.1093, over 5703471.61 frames. ], batch size: 92, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 22:59:56,506 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-02 23:00:06,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5686, 1.9167, 1.8597, 1.6455], device='cuda:0'), covar=tensor([0.1573, 0.1807, 0.1183, 0.1235], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0736, 0.0783, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 23:00:13,861 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:00:16,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:00:17,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.624e+02 1.053e+03 1.293e+03 1.872e+03 4.425e+03, threshold=2.587e+03, percent-clipped=10.0
+2023-03-02 23:00:25,116 INFO [train.py:968] (0/2) Epoch 5, batch 39250, giga_loss[loss=0.2964, simple_loss=0.3694, pruned_loss=0.1117, over 28815.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3555, pruned_loss=0.1118, over 5703977.63 frames. ], libri_tot_loss[loss=0.3448, simple_loss=0.3973, pruned_loss=0.1462, over 5719010.40 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3524, pruned_loss=0.1092, over 5704756.46 frames. ], batch size: 284, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:00:42,352 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:00:45,027 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:00:45,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:01:09,709 INFO [train.py:968] (0/2) Epoch 5, batch 39300, giga_loss[loss=0.3663, simple_loss=0.4052, pruned_loss=0.1638, over 26723.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3577, pruned_loss=0.1121, over 5706555.50 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.3977, pruned_loss=0.1465, over 5719089.60 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3542, pruned_loss=0.1093, over 5706962.77 frames. ], batch size: 555, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:01:10,777 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:01:22,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:01:22,798 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1312, 2.0000, 1.7835, 1.8932], device='cuda:0'), covar=tensor([0.0418, 0.0341, 0.0567, 0.0598], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0443, 0.0498, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 23:01:22,979 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-02 23:01:28,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3862, 1.8077, 1.7371, 1.5505], device='cuda:0'), covar=tensor([0.1564, 0.1860, 0.1256, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0734, 0.0782, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 23:01:45,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.686e+02 1.017e+03 1.273e+03 1.775e+03 5.648e+03, threshold=2.547e+03, percent-clipped=13.0
+2023-03-02 23:01:51,523 INFO [train.py:968] (0/2) Epoch 5, batch 39350, giga_loss[loss=0.2782, simple_loss=0.3525, pruned_loss=0.102, over 29008.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3619, pruned_loss=0.1145, over 5679068.82 frames. ], libri_tot_loss[loss=0.3462, simple_loss=0.3984, pruned_loss=0.147, over 5694771.15 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3574, pruned_loss=0.1109, over 5699651.18 frames. ], batch size: 128, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:02:32,854 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:02:34,703 INFO [train.py:968] (0/2) Epoch 5, batch 39400, libri_loss[loss=0.3003, simple_loss=0.354, pruned_loss=0.1234, over 29385.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3621, pruned_loss=0.1138, over 5685203.52 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3974, pruned_loss=0.1465, over 5698740.07 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3585, pruned_loss=0.1103, over 5697777.82 frames. ], batch size: 67, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:02:38,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3697, 1.5269, 1.2769, 1.3951], device='cuda:0'), covar=tensor([0.2206, 0.2123, 0.2279, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.1122, 0.0863, 0.0995, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 23:02:54,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:03:11,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.341e+02 1.040e+03 1.302e+03 1.640e+03 5.351e+03, threshold=2.605e+03, percent-clipped=9.0
+2023-03-02 23:03:16,513 INFO [train.py:968] (0/2) Epoch 5, batch 39450, giga_loss[loss=0.2474, simple_loss=0.3296, pruned_loss=0.08258, over 29068.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.362, pruned_loss=0.1135, over 5675985.38 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.3973, pruned_loss=0.1467, over 5693076.07 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.358, pruned_loss=0.1096, over 5691453.40 frames. ], batch size: 155, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:03:56,765 INFO [train.py:968] (0/2) Epoch 5, batch 39500, giga_loss[loss=0.3505, simple_loss=0.3917, pruned_loss=0.1546, over 24244.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3628, pruned_loss=0.1141, over 5687834.88 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3971, pruned_loss=0.1466, over 5698590.05 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3587, pruned_loss=0.11, over 5694939.03 frames. ], batch size: 705, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:04:12,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:04:32,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.509e+02 1.146e+03 1.609e+03 2.120e+03 8.125e+03, threshold=3.219e+03, percent-clipped=16.0
+2023-03-02 23:04:36,136 INFO [train.py:968] (0/2) Epoch 5, batch 39550, giga_loss[loss=0.2863, simple_loss=0.3493, pruned_loss=0.1116, over 28653.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3641, pruned_loss=0.1152, over 5678721.88 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3974, pruned_loss=0.1467, over 5689610.24 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3591, pruned_loss=0.1105, over 5691523.38 frames. ], batch size: 92, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:04:45,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9610, 1.6622, 1.2911, 1.4393], device='cuda:0'), covar=tensor([0.0603, 0.0682, 0.0985, 0.0966], device='cuda:0'), in_proj_covar=tensor([0.0333, 0.0443, 0.0498, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 23:04:48,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:04:51,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:05:08,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4868, 1.7064, 1.7891, 1.6292], device='cuda:0'), covar=tensor([0.1538, 0.1695, 0.1172, 0.1209], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0742, 0.0786, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 23:05:19,292 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:05:20,358 INFO [train.py:968] (0/2) Epoch 5, batch 39600, libri_loss[loss=0.4224, simple_loss=0.4596, pruned_loss=0.1926, over 25923.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3652, pruned_loss=0.1163, over 5670483.51 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.3978, pruned_loss=0.1472, over 5681015.00 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3604, pruned_loss=0.1116, over 5687638.59 frames. ], batch size: 136, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 23:05:53,287 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-02 23:05:56,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.705e+02 1.140e+03 1.577e+03 2.211e+03 4.719e+03, threshold=3.153e+03, percent-clipped=8.0
+2023-03-02 23:06:02,135 INFO [train.py:968] (0/2) Epoch 5, batch 39650, giga_loss[loss=0.3012, simple_loss=0.38, pruned_loss=0.1112, over 28777.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3698, pruned_loss=0.1195, over 5685232.69 frames. ], libri_tot_loss[loss=0.3465, simple_loss=0.3981, pruned_loss=0.1475, over 5685296.86 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3643, pruned_loss=0.1142, over 5695306.29 frames. ], batch size: 243, lr: 6.09e-03, grad_scale: 8.0
+2023-03-02 23:06:11,043 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:06:13,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:06:33,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:06:36,337 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:06:40,418 INFO [train.py:968] (0/2) Epoch 5, batch 39700, giga_loss[loss=0.2849, simple_loss=0.3569, pruned_loss=0.1065, over 28535.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3718, pruned_loss=0.1204, over 5689518.72 frames. ], libri_tot_loss[loss=0.3466, simple_loss=0.3982, pruned_loss=0.1475, over 5679687.89 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3669, pruned_loss=0.1157, over 5701785.62 frames. ], batch size: 78, lr: 6.09e-03, grad_scale: 8.0
+2023-03-02 23:07:14,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.756e+02 1.202e+03 1.572e+03 2.147e+03 5.052e+03, threshold=3.144e+03, percent-clipped=5.0
+2023-03-02 23:07:18,456 INFO [train.py:968] (0/2) Epoch 5, batch 39750, giga_loss[loss=0.2959, simple_loss=0.3585, pruned_loss=0.1166, over 28642.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3724, pruned_loss=0.1201, over 5701268.83 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3986, pruned_loss=0.1476, over 5684205.05 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3674, pruned_loss=0.1156, over 5707396.52 frames. ], batch size: 92, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:07:37,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:08:01,242 INFO [train.py:968] (0/2) Epoch 5, batch 39800, giga_loss[loss=0.2926, simple_loss=0.369, pruned_loss=0.1081, over 28984.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3722, pruned_loss=0.1198, over 5708711.10 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.3986, pruned_loss=0.1475, over 5687104.64 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3679, pruned_loss=0.1158, over 5711351.26 frames. ], batch size: 164, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:08:27,416 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:08:29,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:08:36,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.685e+02 1.150e+03 1.489e+03 2.062e+03 4.630e+03, threshold=2.978e+03, percent-clipped=7.0
+2023-03-02 23:08:39,470 INFO [train.py:968] (0/2) Epoch 5, batch 39850, giga_loss[loss=0.2952, simple_loss=0.3673, pruned_loss=0.1116, over 29014.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3715, pruned_loss=0.1194, over 5712963.61 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.3982, pruned_loss=0.1472, over 5692906.95 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3676, pruned_loss=0.1157, over 5710538.59 frames. ], batch size: 155, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:08:51,595 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:09:20,250 INFO [train.py:968] (0/2) Epoch 5, batch 39900, giga_loss[loss=0.2859, simple_loss=0.3601, pruned_loss=0.1058, over 28801.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3708, pruned_loss=0.1192, over 5713542.14 frames. ], libri_tot_loss[loss=0.3458, simple_loss=0.3979, pruned_loss=0.1469, over 5694888.55 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3678, pruned_loss=0.1163, over 5710133.30 frames. ], batch size: 186, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:09:32,342 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:09:35,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:09:57,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.667e+02 1.046e+03 1.303e+03 1.723e+03 4.920e+03, threshold=2.607e+03, percent-clipped=2.0
+2023-03-02 23:09:59,603 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:10:02,159 INFO [train.py:968] (0/2) Epoch 5, batch 39950, libri_loss[loss=0.4056, simple_loss=0.4456, pruned_loss=0.1828, over 28535.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3681, pruned_loss=0.1178, over 5710391.68 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3979, pruned_loss=0.147, over 5695979.49 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3654, pruned_loss=0.1151, over 5706940.05 frames. ], batch size: 106, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:10:38,116 INFO [train.py:968] (0/2) Epoch 5, batch 40000, giga_loss[loss=0.2341, simple_loss=0.3148, pruned_loss=0.07668, over 29031.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3657, pruned_loss=0.1168, over 5712638.77 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3972, pruned_loss=0.1466, over 5692056.43 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3623, pruned_loss=0.1133, over 5714061.34 frames. ], batch size: 128, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:10:38,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:10:48,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3931, 1.4618, 1.3280, 1.5297], device='cuda:0'), covar=tensor([0.2246, 0.2196, 0.2237, 0.2053], device='cuda:0'), in_proj_covar=tensor([0.1117, 0.0862, 0.0991, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 23:11:15,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.201e+02 1.171e+03 1.789e+03 2.350e+03 5.750e+03, threshold=3.578e+03, percent-clipped=15.0
+2023-03-02 23:11:18,842 INFO [train.py:968] (0/2) Epoch 5, batch 40050, giga_loss[loss=0.2367, simple_loss=0.3174, pruned_loss=0.078, over 28523.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3637, pruned_loss=0.1154, over 5711716.06 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3973, pruned_loss=0.1465, over 5695380.74 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3606, pruned_loss=0.1123, over 5710185.28 frames. ], batch size: 85, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:11:59,653 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-222000.pt
+2023-03-02 23:11:59,947 INFO [train.py:968] (0/2) Epoch 5, batch 40100, giga_loss[loss=0.2996, simple_loss=0.3601, pruned_loss=0.1196, over 23966.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3647, pruned_loss=0.1142, over 5710577.05 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3973, pruned_loss=0.1465, over 5698771.11 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3617, pruned_loss=0.1113, over 5706676.40 frames. ], batch size: 705, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:12:02,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3682, 2.0059, 1.4333, 0.5184], device='cuda:0'), covar=tensor([0.2859, 0.1235, 0.1740, 0.2596], device='cuda:0'), in_proj_covar=tensor([0.1385, 0.1283, 0.1363, 0.1145], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 23:12:35,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.109e+02 1.084e+03 1.335e+03 1.714e+03 2.899e+03, threshold=2.670e+03, percent-clipped=0.0
+2023-03-02 23:12:37,865 INFO [train.py:968] (0/2) Epoch 5, batch 40150, giga_loss[loss=0.2855, simple_loss=0.362, pruned_loss=0.1045, over 29058.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3676, pruned_loss=0.1162, over 5709270.31 frames. ], libri_tot_loss[loss=0.3448, simple_loss=0.3969, pruned_loss=0.1463, over 5702336.17 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3644, pruned_loss=0.1128, over 5703097.93 frames. ], batch size: 155, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:12:38,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7139, 2.4489, 1.8272, 0.6786], device='cuda:0'), covar=tensor([0.3613, 0.1749, 0.1767, 0.3768], device='cuda:0'), in_proj_covar=tensor([0.1382, 0.1281, 0.1360, 0.1146], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 23:13:19,322 INFO [train.py:968] (0/2) Epoch 5, batch 40200, giga_loss[loss=0.2748, simple_loss=0.3493, pruned_loss=0.1002, over 28905.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.366, pruned_loss=0.1161, over 5717115.65 frames. ], libri_tot_loss[loss=0.3442, simple_loss=0.3965, pruned_loss=0.146, over 5704502.15 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3636, pruned_loss=0.1135, over 5710459.96 frames. ], batch size: 174, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:13:53,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.782e+02 1.063e+03 1.297e+03 1.625e+03 5.043e+03, threshold=2.594e+03, percent-clipped=6.0
+2023-03-02 23:13:56,223 INFO [train.py:968] (0/2) Epoch 5, batch 40250, giga_loss[loss=0.3737, simple_loss=0.4153, pruned_loss=0.1661, over 26796.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3657, pruned_loss=0.1172, over 5724990.94 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3965, pruned_loss=0.1458, over 5711862.28 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3624, pruned_loss=0.114, over 5713265.10 frames. ], batch size: 555, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:13:57,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:14:41,672 INFO [train.py:968] (0/2) Epoch 5, batch 40300, giga_loss[loss=0.2856, simple_loss=0.3452, pruned_loss=0.113, over 28841.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3633, pruned_loss=0.1173, over 5720707.44 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.3966, pruned_loss=0.1458, over 5712831.96 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3605, pruned_loss=0.1147, over 5710727.03 frames. ], batch size: 119, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:15:17,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.263e+02 1.158e+03 1.418e+03 2.321e+03 5.047e+03, threshold=2.836e+03, percent-clipped=13.0
+2023-03-02 23:15:21,235 INFO [train.py:968] (0/2) Epoch 5, batch 40350, giga_loss[loss=0.2707, simple_loss=0.3371, pruned_loss=0.1021, over 28996.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3617, pruned_loss=0.1177, over 5715251.27 frames. ], libri_tot_loss[loss=0.3443, simple_loss=0.3967, pruned_loss=0.1459, over 5715784.08 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3588, pruned_loss=0.115, over 5704847.45 frames. ], batch size: 128, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:15:42,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:15:49,126 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:16:04,189 INFO [train.py:968] (0/2) Epoch 5, batch 40400, giga_loss[loss=0.2344, simple_loss=0.3011, pruned_loss=0.08384, over 28568.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3606, pruned_loss=0.1169, over 5703813.70 frames. ], libri_tot_loss[loss=0.3445, simple_loss=0.3969, pruned_loss=0.1461, over 5708625.55 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3576, pruned_loss=0.1143, over 5701789.87 frames. ], batch size: 60, lr: 6.08e-03, grad_scale: 8.0
+2023-03-02 23:16:39,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.495e+02 1.025e+03 1.372e+03 2.138e+03 5.417e+03, threshold=2.745e+03, percent-clipped=13.0
+2023-03-02 23:16:42,675 INFO [train.py:968] (0/2) Epoch 5, batch 40450, giga_loss[loss=0.2509, simple_loss=0.3271, pruned_loss=0.08738, over 28571.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3577, pruned_loss=0.1155, over 5704759.50 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3975, pruned_loss=0.1465, over 5703758.49 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3538, pruned_loss=0.1122, over 5707148.97 frames. ], batch size: 336, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:17:22,631 INFO [train.py:968] (0/2) Epoch 5, batch 40500, giga_loss[loss=0.2891, simple_loss=0.3437, pruned_loss=0.1173, over 28702.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3542, pruned_loss=0.1136, over 5709058.01 frames. ], libri_tot_loss[loss=0.3457, simple_loss=0.398, pruned_loss=0.1467, over 5708895.15 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3489, pruned_loss=0.1094, over 5705997.42 frames. ], batch size: 99, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:17:24,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:17:38,475 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:17:40,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:18:01,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.175e+02 1.107e+03 1.490e+03 1.987e+03 1.127e+04, threshold=2.979e+03, percent-clipped=7.0
+2023-03-02 23:18:02,902 INFO [train.py:968] (0/2) Epoch 5, batch 40550, giga_loss[loss=0.2731, simple_loss=0.3508, pruned_loss=0.09769, over 28882.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3511, pruned_loss=0.1111, over 5704161.42 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3975, pruned_loss=0.1464, over 5699889.24 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3463, pruned_loss=0.1073, over 5709424.49 frames. ], batch size: 174, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:18:03,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:18:40,481 INFO [train.py:968] (0/2) Epoch 5, batch 40600, giga_loss[loss=0.3386, simple_loss=0.3977, pruned_loss=0.1398, over 27502.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3532, pruned_loss=0.1121, over 5696150.37 frames. ], libri_tot_loss[loss=0.346, simple_loss=0.398, pruned_loss=0.147, over 5688336.50 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3473, pruned_loss=0.1073, over 5711363.31 frames. ], batch size: 472, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:18:45,115 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2985, 1.3200, 4.9662, 3.4239], device='cuda:0'), covar=tensor([0.1603, 0.2264, 0.0280, 0.0635], device='cuda:0'), in_proj_covar=tensor([0.0570, 0.0527, 0.0754, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 23:18:45,938 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-02 23:19:00,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:19:18,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.932e+02 1.143e+03 1.390e+03 1.852e+03 6.751e+03, threshold=2.780e+03, percent-clipped=11.0
+2023-03-02 23:19:21,052 INFO [train.py:968] (0/2) Epoch 5, batch 40650, giga_loss[loss=0.3133, simple_loss=0.3792, pruned_loss=0.1237, over 28804.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3568, pruned_loss=0.1135, over 5700400.13 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.397, pruned_loss=0.1464, over 5692557.99 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3522, pruned_loss=0.1095, over 5709049.68 frames. ], batch size: 112, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:19:30,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8124, 4.6094, 4.4285, 1.7868], device='cuda:0'), covar=tensor([0.0320, 0.0393, 0.0560, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.0860, 0.0788, 0.0784, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 23:19:45,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:20:01,159 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:20:02,407 INFO [train.py:968] (0/2) Epoch 5, batch 40700, giga_loss[loss=0.2921, simple_loss=0.356, pruned_loss=0.1141, over 28554.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3603, pruned_loss=0.115, over 5701973.07 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3974, pruned_loss=0.1467, over 5695571.57 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3554, pruned_loss=0.111, over 5706171.39 frames. ], batch size: 85, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:20:22,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0100, 1.0060, 3.6100, 3.0784], device='cuda:0'), covar=tensor([0.2055, 0.2936, 0.0683, 0.1201], device='cuda:0'), in_proj_covar=tensor([0.0576, 0.0532, 0.0759, 0.0625], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 23:20:43,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.731e+02 1.109e+03 1.453e+03 2.342e+03 9.147e+03, threshold=2.906e+03, percent-clipped=17.0
+2023-03-02 23:20:45,395 INFO [train.py:968] (0/2) Epoch 5, batch 40750, libri_loss[loss=0.3017, simple_loss=0.3647, pruned_loss=0.1193, over 29566.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3639, pruned_loss=0.1165, over 5695430.08 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3974, pruned_loss=0.1466, over 5695098.21 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3597, pruned_loss=0.1131, over 5698783.03 frames. ], batch size: 78, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:20:50,957 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:21:00,714 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:21:03,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:21:27,462 INFO [train.py:968] (0/2) Epoch 5, batch 40800, giga_loss[loss=0.3261, simple_loss=0.3924, pruned_loss=0.1299, over 28662.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3671, pruned_loss=0.1181, over 5705451.63 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3976, pruned_loss=0.1467, over 5699120.17 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3629, pruned_loss=0.1147, over 5704704.35 frames. ], batch size: 307, lr: 6.08e-03, grad_scale: 8.0
+2023-03-02 23:21:28,464 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:21:43,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4686, 1.9191, 1.7776, 1.5491], device='cuda:0'), covar=tensor([0.1457, 0.1721, 0.1144, 0.1213], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0732, 0.0789, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 23:21:57,277 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-02 23:22:12,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.934e+02 1.193e+03 1.533e+03 1.849e+03 4.341e+03, threshold=3.065e+03, percent-clipped=6.0
+2023-03-02 23:22:14,971 INFO [train.py:968] (0/2) Epoch 5, batch 40850, giga_loss[loss=0.3854, simple_loss=0.4288, pruned_loss=0.171, over 28666.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.373, pruned_loss=0.1238, over 5700506.61 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3971, pruned_loss=0.1465, over 5700858.42 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3695, pruned_loss=0.1206, over 5698402.51 frames. ], batch size: 336, lr: 6.08e-03, grad_scale: 8.0
+2023-03-02 23:22:43,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:23:04,462 INFO [train.py:968] (0/2) Epoch 5, batch 40900, giga_loss[loss=0.334, simple_loss=0.3917, pruned_loss=0.1382, over 28779.00 frames. ], tot_loss[loss=0.317, simple_loss=0.378, pruned_loss=0.128, over 5701017.40 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.397, pruned_loss=0.1464, over 5702196.79 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3748, pruned_loss=0.1252, over 5698054.64 frames. ], batch size: 284, lr: 6.08e-03, grad_scale: 8.0
+2023-03-02 23:23:05,437 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:23:07,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:23:14,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5079, 3.3674, 1.5038, 1.4870], device='cuda:0'), covar=tensor([0.0777, 0.0380, 0.0761, 0.1119], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0481, 0.0308, 0.0375], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 23:23:38,170 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:23:40,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9764, 4.5241, 1.9322, 1.8939], device='cuda:0'), covar=tensor([0.0744, 0.0310, 0.0726, 0.1101], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0479, 0.0308, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-02 23:23:50,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.719e+03 2.148e+03 3.146e+03 4.814e+03, threshold=4.297e+03, percent-clipped=26.0
+2023-03-02 23:23:54,004 INFO [train.py:968] (0/2) Epoch 5, batch 40950, giga_loss[loss=0.3498, simple_loss=0.403, pruned_loss=0.1483, over 28725.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3874, pruned_loss=0.1351, over 5696229.15 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3973, pruned_loss=0.1466, over 5703218.81 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3847, pruned_loss=0.1327, over 5692891.90 frames. ], batch size: 262, lr: 6.08e-03, grad_scale: 8.0
+2023-03-02 23:24:02,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6321, 1.5412, 1.5815, 1.5195], device='cuda:0'), covar=tensor([0.1067, 0.1779, 0.1698, 0.1424], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0743, 0.0646, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 23:24:32,830 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-02 23:24:39,037 INFO [train.py:968] (0/2) Epoch 5, batch 41000, giga_loss[loss=0.3621, simple_loss=0.4071, pruned_loss=0.1585, over 28801.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3933, pruned_loss=0.1408, over 5700234.34 frames. ], libri_tot_loss[loss=0.3448, simple_loss=0.3969, pruned_loss=0.1464, over 5707271.33 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3913, pruned_loss=0.1388, over 5693954.90 frames. ], batch size: 99, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:24:39,429 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.02 vs. limit=5.0
+2023-03-02 23:24:52,480 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:24:55,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:24:58,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:25:20,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.599e+03 1.960e+03 2.561e+03 5.803e+03, threshold=3.920e+03, percent-clipped=2.0
+2023-03-02 23:25:21,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6600, 2.1285, 2.0216, 1.7922], device='cuda:0'), covar=tensor([0.1408, 0.1611, 0.1062, 0.1199], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0730, 0.0783, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-02 23:25:21,881 INFO [train.py:968] (0/2) Epoch 5, batch 41050, giga_loss[loss=0.3254, simple_loss=0.3841, pruned_loss=0.1333, over 28403.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3987, pruned_loss=0.1453, over 5687638.13 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.3965, pruned_loss=0.1459, over 5694745.71 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3974, pruned_loss=0.144, over 5691782.57 frames. ], batch size: 71, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:25:24,541 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:25:26,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:25:33,850 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:25:41,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:25:48,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-02 23:26:01,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4892, 1.6164, 1.3079, 1.1631], device='cuda:0'), covar=tensor([0.1136, 0.0915, 0.0721, 0.0957], device='cuda:0'), in_proj_covar=tensor([0.1463, 0.1268, 0.1249, 0.1344], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-02 23:26:09,887 INFO [train.py:968] (0/2) Epoch 5, batch 41100, giga_loss[loss=0.4673, simple_loss=0.47, pruned_loss=0.2323, over 26536.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4046, pruned_loss=0.151, over 5671517.53 frames. ], libri_tot_loss[loss=0.3442, simple_loss=0.3966, pruned_loss=0.1459, over 5697202.92 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4035, pruned_loss=0.15, over 5672254.78 frames. ], batch size: 555, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:26:16,462 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 23:26:58,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.689e+03 2.440e+03 3.286e+03 8.867e+03, threshold=4.880e+03, percent-clipped=15.0
+2023-03-02 23:27:01,767 INFO [train.py:968] (0/2) Epoch 5, batch 41150, giga_loss[loss=0.4775, simple_loss=0.4724, pruned_loss=0.2413, over 26657.00 frames. ], tot_loss[loss=0.357, simple_loss=0.4067, pruned_loss=0.1537, over 5664529.10 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.3964, pruned_loss=0.1457, over 5699434.13 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.4062, pruned_loss=0.1532, over 5662164.38 frames. ], batch size: 555, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:27:55,544 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:27:59,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:28:00,825 INFO [train.py:968] (0/2) Epoch 5, batch 41200, giga_loss[loss=0.3573, simple_loss=0.4057, pruned_loss=0.1545, over 28654.00 frames. ], tot_loss[loss=0.3617, simple_loss=0.4092, pruned_loss=0.157, over 5658260.03 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.3963, pruned_loss=0.1455, over 5702660.11 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.4092, pruned_loss=0.157, over 5653019.60 frames. ], batch size: 242, lr: 6.07e-03, grad_scale: 8.0
+2023-03-02 23:28:16,830 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:28:19,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:28:28,305 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:28:31,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-02 23:28:45,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:28:46,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.550e+02 1.610e+03 1.932e+03 3.083e+03 1.230e+04, threshold=3.864e+03, percent-clipped=6.0
+2023-03-02 23:28:47,524 INFO [train.py:968] (0/2) Epoch 5, batch 41250, giga_loss[loss=0.3838, simple_loss=0.4208, pruned_loss=0.1734, over 28962.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.4108, pruned_loss=0.1597, over 5638735.86 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3956, pruned_loss=0.1451, over 5698759.16 frames. ], giga_tot_loss[loss=0.3665, simple_loss=0.412, pruned_loss=0.1605, over 5636486.33 frames. ], batch size: 106, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:29:41,141 INFO [train.py:968] (0/2) Epoch 5, batch 41300, giga_loss[loss=0.4658, simple_loss=0.4821, pruned_loss=0.2248, over 28907.00 frames. ], tot_loss[loss=0.3699, simple_loss=0.4144, pruned_loss=0.1627, over 5635338.54 frames. ], libri_tot_loss[loss=0.3431, simple_loss=0.3958, pruned_loss=0.1452, over 5699259.84 frames. ], giga_tot_loss[loss=0.3714, simple_loss=0.4155, pruned_loss=0.1636, over 5631983.19 frames. ], batch size: 186, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:30:00,071 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-02 23:30:05,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6387, 3.4734, 3.3366, 1.6006], device='cuda:0'), covar=tensor([0.0626, 0.0669, 0.0814, 0.2202], device='cuda:0'), in_proj_covar=tensor([0.0874, 0.0814, 0.0806, 0.0615], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 23:30:35,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.691e+03 2.110e+03 2.594e+03 9.055e+03, threshold=4.220e+03, percent-clipped=12.0
+2023-03-02 23:30:36,840 INFO [train.py:968] (0/2) Epoch 5, batch 41350, giga_loss[loss=0.325, simple_loss=0.386, pruned_loss=0.132, over 28953.00 frames. ], tot_loss[loss=0.373, simple_loss=0.416, pruned_loss=0.165, over 5627597.16 frames. ], libri_tot_loss[loss=0.3431, simple_loss=0.3957, pruned_loss=0.1452, over 5701172.45 frames. ], giga_tot_loss[loss=0.3747, simple_loss=0.4173, pruned_loss=0.166, over 5621810.10 frames. ], batch size: 106, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:30:47,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7490, 1.6590, 1.6386, 1.7058], device='cuda:0'), covar=tensor([0.1079, 0.1793, 0.1590, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0740, 0.0638, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 23:31:13,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:31:20,620 INFO [train.py:968] (0/2) Epoch 5, batch 41400, giga_loss[loss=0.4903, simple_loss=0.4891, pruned_loss=0.2457, over 26714.00 frames. ], tot_loss[loss=0.3715, simple_loss=0.4142, pruned_loss=0.1644, over 5629279.45 frames. ], libri_tot_loss[loss=0.3434, simple_loss=0.3958, pruned_loss=0.1455, over 5697422.82 frames. ], giga_tot_loss[loss=0.3735, simple_loss=0.4158, pruned_loss=0.1656, over 5626311.51 frames. ], batch size: 555, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:31:51,606 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:31:56,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:32:05,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.407e+02 1.676e+03 2.199e+03 2.982e+03 8.219e+03, threshold=4.397e+03, percent-clipped=11.0
+2023-03-02 23:32:06,549 INFO [train.py:968] (0/2) Epoch 5, batch 41450, giga_loss[loss=0.3391, simple_loss=0.3972, pruned_loss=0.1405, over 28861.00 frames. ], tot_loss[loss=0.3678, simple_loss=0.412, pruned_loss=0.1618, over 5645875.25 frames. ], libri_tot_loss[loss=0.343, simple_loss=0.3955, pruned_loss=0.1453, over 5698898.12 frames. ], giga_tot_loss[loss=0.3708, simple_loss=0.4142, pruned_loss=0.1637, over 5639821.73 frames. ], batch size: 199, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:32:16,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7319, 1.5802, 1.2191, 1.3638], device='cuda:0'), covar=tensor([0.0566, 0.0535, 0.0901, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0453, 0.0503, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 23:32:19,994 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:32:34,499 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:32:56,485 INFO [train.py:968] (0/2) Epoch 5, batch 41500, giga_loss[loss=0.329, simple_loss=0.3979, pruned_loss=0.1301, over 28892.00 frames. ], tot_loss[loss=0.3666, simple_loss=0.4123, pruned_loss=0.1604, over 5644443.03 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3951, pruned_loss=0.1451, over 5693717.10 frames. ], giga_tot_loss[loss=0.3699, simple_loss=0.4149, pruned_loss=0.1624, over 5642763.23 frames. ], batch size: 174, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:33:30,213 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:33:33,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:33:44,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.232e+02 1.593e+03 2.076e+03 2.749e+03 7.034e+03, threshold=4.152e+03, percent-clipped=4.0
+2023-03-02 23:33:45,325 INFO [train.py:968] (0/2) Epoch 5, batch 41550, giga_loss[loss=0.3699, simple_loss=0.4148, pruned_loss=0.1625, over 27857.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.4131, pruned_loss=0.1599, over 5661402.54 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.395, pruned_loss=0.1449, over 5697173.57 frames. ], giga_tot_loss[loss=0.3698, simple_loss=0.4156, pruned_loss=0.162, over 5656250.03 frames. ], batch size: 412, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:33:54,818 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:34:05,604 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:34:17,803 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:34:22,362 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:34:41,423 INFO [train.py:968] (0/2) Epoch 5, batch 41600, giga_loss[loss=0.3146, simple_loss=0.3774, pruned_loss=0.1259, over 28742.00 frames. ], tot_loss[loss=0.3645, simple_loss=0.4117, pruned_loss=0.1586, over 5645710.39 frames. ], libri_tot_loss[loss=0.342, simple_loss=0.3947, pruned_loss=0.1446, over 5697934.52 frames. ], giga_tot_loss[loss=0.3679, simple_loss=0.4143, pruned_loss=0.1608, over 5640197.42 frames. ], batch size: 262, lr: 6.07e-03, grad_scale: 8.0
+2023-03-02 23:34:52,883 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:35:30,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.579e+02 1.619e+03 2.198e+03 3.204e+03 1.010e+04, threshold=4.395e+03, percent-clipped=9.0
+2023-03-02 23:35:30,873 INFO [train.py:968] (0/2) Epoch 5, batch 41650, giga_loss[loss=0.3289, simple_loss=0.3852, pruned_loss=0.1363, over 28725.00 frames. ], tot_loss[loss=0.361, simple_loss=0.4101, pruned_loss=0.156, over 5648550.82 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.395, pruned_loss=0.1449, over 5702080.30 frames. ], giga_tot_loss[loss=0.3638, simple_loss=0.4121, pruned_loss=0.1577, over 5639934.24 frames. ], batch size: 284, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:35:47,641 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223566.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 23:36:18,193 INFO [train.py:968] (0/2) Epoch 5, batch 41700, giga_loss[loss=0.3245, simple_loss=0.3832, pruned_loss=0.1329, over 28951.00 frames. ], tot_loss[loss=0.3574, simple_loss=0.4079, pruned_loss=0.1534, over 5644751.20 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.3955, pruned_loss=0.1454, over 5688812.31 frames. ], giga_tot_loss[loss=0.3593, simple_loss=0.4095, pruned_loss=0.1546, over 5647715.88 frames. ], batch size: 164, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:36:41,638 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6335, 1.9249, 1.5326, 1.2077], device='cuda:0'), covar=tensor([0.1331, 0.0910, 0.0817, 0.1167], device='cuda:0'), in_proj_covar=tensor([0.1454, 0.1273, 0.1261, 0.1341], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-02 23:37:05,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.913e+02 1.530e+03 2.009e+03 2.988e+03 9.299e+03, threshold=4.019e+03, percent-clipped=7.0
+2023-03-02 23:37:05,512 INFO [train.py:968] (0/2) Epoch 5, batch 41750, giga_loss[loss=0.3579, simple_loss=0.4124, pruned_loss=0.1517, over 28611.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4038, pruned_loss=0.1497, over 5649482.15 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3952, pruned_loss=0.1453, over 5685532.40 frames. ], giga_tot_loss[loss=0.3537, simple_loss=0.4057, pruned_loss=0.1509, over 5654072.82 frames. ], batch size: 336, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:37:54,215 INFO [train.py:968] (0/2) Epoch 5, batch 41800, giga_loss[loss=0.377, simple_loss=0.4195, pruned_loss=0.1672, over 27505.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.4002, pruned_loss=0.1471, over 5645277.95 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.395, pruned_loss=0.1452, over 5681998.82 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.4021, pruned_loss=0.1482, over 5650951.76 frames. ], batch size: 472, lr: 6.07e-03, grad_scale: 2.0
+2023-03-02 23:38:01,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:38:12,054 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2795, 2.0612, 1.6175, 0.5587], device='cuda:0'), covar=tensor([0.2241, 0.1267, 0.2114, 0.2525], device='cuda:0'), in_proj_covar=tensor([0.1410, 0.1314, 0.1379, 0.1162], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 23:38:31,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:38:44,537 INFO [train.py:968] (0/2) Epoch 5, batch 41850, giga_loss[loss=0.4445, simple_loss=0.4551, pruned_loss=0.2169, over 26708.00 frames. ], tot_loss[loss=0.3481, simple_loss=0.4003, pruned_loss=0.148, over 5632856.99 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3948, pruned_loss=0.1453, over 5673037.38 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.402, pruned_loss=0.1489, over 5643890.24 frames. ], batch size: 555, lr: 6.06e-03, grad_scale: 2.0
+2023-03-02 23:38:45,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.412e+02 1.550e+03 2.021e+03 2.955e+03 8.435e+03, threshold=4.042e+03, percent-clipped=9.0
+2023-03-02 23:38:46,145 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:39:32,470 INFO [train.py:968] (0/2) Epoch 5, batch 41900, libri_loss[loss=0.3119, simple_loss=0.3817, pruned_loss=0.1211, over 29510.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4003, pruned_loss=0.1473, over 5650269.03 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3946, pruned_loss=0.145, over 5675370.03 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.4019, pruned_loss=0.1483, over 5656458.92 frames. ], batch size: 82, lr: 6.06e-03, grad_scale: 2.0
+2023-03-02 23:40:03,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:40:23,076 INFO [train.py:968] (0/2) Epoch 5, batch 41950, libri_loss[loss=0.3074, simple_loss=0.3552, pruned_loss=0.1298, over 29495.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3972, pruned_loss=0.1444, over 5642966.02 frames. ], libri_tot_loss[loss=0.343, simple_loss=0.395, pruned_loss=0.1455, over 5659571.21 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3983, pruned_loss=0.1448, over 5661509.00 frames. ], batch size: 70, lr: 6.06e-03, grad_scale: 2.0
+2023-03-02 23:40:23,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.444e+02 1.537e+03 1.867e+03 2.527e+03 8.444e+03, threshold=3.735e+03, percent-clipped=7.0
+2023-03-02 23:40:24,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:40:27,275 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-02 23:40:28,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:40:53,470 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:40:57,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:40:57,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:41:12,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:41:14,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:41:16,173 INFO [train.py:968] (0/2) Epoch 5, batch 42000, giga_loss[loss=0.3337, simple_loss=0.4042, pruned_loss=0.1316, over 28983.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3973, pruned_loss=0.1421, over 5652529.03 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3956, pruned_loss=0.1459, over 5660382.35 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3977, pruned_loss=0.142, over 5666175.57 frames. ], batch size: 136, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:41:16,177 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-02 23:41:24,563 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.2309, simple_loss=0.3321, pruned_loss=0.06484, over 944034.00 frames. 
+2023-03-02 23:41:24,564 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-02 23:41:34,801 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:41:43,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0252, 1.0811, 4.4123, 3.1948], device='cuda:0'), covar=tensor([0.1768, 0.2447, 0.0372, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0582, 0.0540, 0.0772, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 23:41:55,326 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:42:06,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223941.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 23:42:15,363 INFO [train.py:968] (0/2) Epoch 5, batch 42050, giga_loss[loss=0.392, simple_loss=0.4492, pruned_loss=0.1674, over 28943.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3989, pruned_loss=0.1416, over 5659910.46 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.3955, pruned_loss=0.1458, over 5663941.19 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3994, pruned_loss=0.1416, over 5667381.59 frames. ], batch size: 164, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:42:16,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.426e+02 1.516e+03 1.877e+03 3.111e+03 7.202e+03, threshold=3.753e+03, percent-clipped=10.0
+2023-03-02 23:42:34,667 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:42:38,274 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:42:41,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:43:04,842 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-224000.pt
+2023-03-02 23:43:05,137 INFO [train.py:968] (0/2) Epoch 5, batch 42100, giga_loss[loss=0.3446, simple_loss=0.3966, pruned_loss=0.1463, over 28340.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3999, pruned_loss=0.1436, over 5660257.91 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.3952, pruned_loss=0.1457, over 5667505.63 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.4005, pruned_loss=0.1436, over 5663028.09 frames. ], batch size: 368, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:43:10,539 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:43:48,636 INFO [train.py:968] (0/2) Epoch 5, batch 42150, giga_loss[loss=0.3152, simple_loss=0.3824, pruned_loss=0.124, over 28923.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3987, pruned_loss=0.1431, over 5675823.83 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3944, pruned_loss=0.1452, over 5675840.02 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.4, pruned_loss=0.1434, over 5670563.41 frames. ], batch size: 199, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:43:49,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.293e+02 1.689e+03 2.174e+03 2.845e+03 7.444e+03, threshold=4.348e+03, percent-clipped=11.0
+2023-03-02 23:44:18,868 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224084.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 23:44:20,986 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=224087.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 23:44:33,131 INFO [train.py:968] (0/2) Epoch 5, batch 42200, giga_loss[loss=0.3103, simple_loss=0.3719, pruned_loss=0.1243, over 28934.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3968, pruned_loss=0.1429, over 5665409.47 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3946, pruned_loss=0.1453, over 5669999.08 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3978, pruned_loss=0.143, over 5666678.06 frames. ], batch size: 112, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:44:41,793 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-02 23:44:48,766 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224116.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 23:45:17,344 INFO [train.py:968] (0/2) Epoch 5, batch 42250, giga_loss[loss=0.3379, simple_loss=0.3899, pruned_loss=0.143, over 28841.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3954, pruned_loss=0.1431, over 5659127.36 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3942, pruned_loss=0.1452, over 5669720.04 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3966, pruned_loss=0.1432, over 5659717.05 frames. ], batch size: 112, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:45:17,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.630e+03 2.399e+03 3.228e+03 1.289e+04, threshold=4.799e+03, percent-clipped=11.0
+2023-03-02 23:46:04,395 INFO [train.py:968] (0/2) Epoch 5, batch 42300, giga_loss[loss=0.3085, simple_loss=0.3865, pruned_loss=0.1152, over 28567.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3943, pruned_loss=0.1416, over 5670318.90 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.3938, pruned_loss=0.1449, over 5675828.67 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3957, pruned_loss=0.1419, over 5665343.26 frames. ], batch size: 85, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:46:51,401 INFO [train.py:968] (0/2) Epoch 5, batch 42350, giga_loss[loss=0.3518, simple_loss=0.3998, pruned_loss=0.1519, over 28526.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3932, pruned_loss=0.1391, over 5680312.92 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3943, pruned_loss=0.1452, over 5674969.71 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3938, pruned_loss=0.139, over 5677336.03 frames. ], batch size: 336, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:46:52,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.538e+02 1.373e+03 1.808e+03 2.505e+03 8.458e+03, threshold=3.616e+03, percent-clipped=6.0
+2023-03-02 23:47:40,949 INFO [train.py:968] (0/2) Epoch 5, batch 42400, giga_loss[loss=0.3262, simple_loss=0.3858, pruned_loss=0.1333, over 28300.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3946, pruned_loss=0.1395, over 5680134.40 frames. ], libri_tot_loss[loss=0.3428, simple_loss=0.3948, pruned_loss=0.1454, over 5677643.25 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3946, pruned_loss=0.1391, over 5675500.59 frames. ], batch size: 368, lr: 6.06e-03, grad_scale: 8.0
+2023-03-02 23:48:01,370 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-02 23:48:19,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=224346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:48:24,757 INFO [train.py:968] (0/2) Epoch 5, batch 42450, libri_loss[loss=0.263, simple_loss=0.3244, pruned_loss=0.1008, over 29454.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3934, pruned_loss=0.1388, over 5688314.05 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3941, pruned_loss=0.1448, over 5682511.89 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3942, pruned_loss=0.1388, over 5680384.55 frames. ], batch size: 70, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:48:27,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.349e+02 1.521e+03 1.964e+03 2.480e+03 6.376e+03, threshold=3.928e+03, percent-clipped=6.0
+2023-03-02 23:48:28,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6964, 4.5716, 4.3210, 1.7735], device='cuda:0'), covar=tensor([0.0474, 0.0575, 0.0812, 0.2168], device='cuda:0'), in_proj_covar=tensor([0.0875, 0.0821, 0.0810, 0.0607], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 23:49:07,744 INFO [train.py:968] (0/2) Epoch 5, batch 42500, giga_loss[loss=0.3252, simple_loss=0.3804, pruned_loss=0.135, over 28991.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3921, pruned_loss=0.1388, over 5679789.07 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3947, pruned_loss=0.1455, over 5678538.35 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3921, pruned_loss=0.138, over 5676704.40 frames. ], batch size: 106, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:49:39,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5913, 1.5553, 1.4503, 1.4981], device='cuda:0'), covar=tensor([0.1264, 0.2090, 0.1737, 0.1810], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0741, 0.0642, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-02 23:49:54,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5988, 1.4494, 1.1776, 1.2602], device='cuda:0'), covar=tensor([0.0561, 0.0514, 0.0928, 0.0750], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0458, 0.0508, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-02 23:49:58,014 INFO [train.py:968] (0/2) Epoch 5, batch 42550, giga_loss[loss=0.3222, simple_loss=0.379, pruned_loss=0.1327, over 28934.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3922, pruned_loss=0.1397, over 5667997.20 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3945, pruned_loss=0.1454, over 5671944.67 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3923, pruned_loss=0.1391, over 5671104.85 frames. ], batch size: 199, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:49:59,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.000e+02 1.598e+03 2.194e+03 2.859e+03 5.626e+03, threshold=4.388e+03, percent-clipped=5.0
+2023-03-02 23:50:35,558 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:50:39,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=224492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:50:46,205 INFO [train.py:968] (0/2) Epoch 5, batch 42600, giga_loss[loss=0.3221, simple_loss=0.3771, pruned_loss=0.1335, over 28520.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3926, pruned_loss=0.1415, over 5660434.93 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3943, pruned_loss=0.1452, over 5671509.63 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3928, pruned_loss=0.141, over 5664047.16 frames. ], batch size: 78, lr: 6.05e-03, grad_scale: 4.0
+2023-03-02 23:51:07,977 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:51:34,451 INFO [train.py:968] (0/2) Epoch 5, batch 42650, giga_loss[loss=0.2966, simple_loss=0.3631, pruned_loss=0.1151, over 28999.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3908, pruned_loss=0.1404, over 5662191.74 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.395, pruned_loss=0.1456, over 5666683.79 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3903, pruned_loss=0.1395, over 5669221.07 frames. ], batch size: 106, lr: 6.05e-03, grad_scale: 4.0
+2023-03-02 23:51:36,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.803e+03 2.323e+03 3.678e+03 8.445e+03, threshold=4.646e+03, percent-clipped=15.0
+2023-03-02 23:52:19,722 INFO [train.py:968] (0/2) Epoch 5, batch 42700, giga_loss[loss=0.3603, simple_loss=0.4059, pruned_loss=0.1573, over 28747.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3904, pruned_loss=0.1403, over 5677212.62 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3946, pruned_loss=0.1452, over 5676554.86 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3902, pruned_loss=0.1398, over 5674136.29 frames. ], batch size: 66, lr: 6.05e-03, grad_scale: 4.0
+2023-03-02 23:52:25,000 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=224605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:53:12,862 INFO [train.py:968] (0/2) Epoch 5, batch 42750, giga_loss[loss=0.3326, simple_loss=0.389, pruned_loss=0.1381, over 28835.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3909, pruned_loss=0.1409, over 5673198.15 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3945, pruned_loss=0.1451, over 5670047.34 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3908, pruned_loss=0.1405, over 5677006.13 frames. ], batch size: 145, lr: 6.05e-03, grad_scale: 4.0
+2023-03-02 23:53:14,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.670e+02 1.612e+03 1.925e+03 2.591e+03 4.855e+03, threshold=3.850e+03, percent-clipped=2.0
+2023-03-02 23:53:25,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8846, 1.1283, 3.5134, 3.0337], device='cuda:0'), covar=tensor([0.1708, 0.2354, 0.0425, 0.1249], device='cuda:0'), in_proj_covar=tensor([0.0586, 0.0541, 0.0775, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-02 23:53:58,400 INFO [train.py:968] (0/2) Epoch 5, batch 42800, giga_loss[loss=0.2879, simple_loss=0.3613, pruned_loss=0.1072, over 29005.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3912, pruned_loss=0.14, over 5678781.94 frames. ], libri_tot_loss[loss=0.3417, simple_loss=0.3941, pruned_loss=0.1447, over 5673581.18 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3914, pruned_loss=0.14, over 5678677.52 frames. ], batch size: 136, lr: 6.05e-03, grad_scale: 8.0
+2023-03-02 23:54:44,085 INFO [train.py:968] (0/2) Epoch 5, batch 42850, giga_loss[loss=0.2823, simple_loss=0.3624, pruned_loss=0.1011, over 29023.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3924, pruned_loss=0.14, over 5683756.01 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3941, pruned_loss=0.1448, over 5676857.15 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3925, pruned_loss=0.1398, over 5680620.63 frames. ], batch size: 164, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:54:47,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.451e+02 1.788e+03 2.310e+03 3.163e+03 8.760e+03, threshold=4.620e+03, percent-clipped=15.0
+2023-03-02 23:55:29,135 INFO [train.py:968] (0/2) Epoch 5, batch 42900, giga_loss[loss=0.3693, simple_loss=0.4137, pruned_loss=0.1624, over 28392.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.394, pruned_loss=0.1412, over 5672668.70 frames. ], libri_tot_loss[loss=0.3421, simple_loss=0.3943, pruned_loss=0.1449, over 5673811.59 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3938, pruned_loss=0.1408, over 5673439.47 frames. ], batch size: 65, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:56:26,181 INFO [train.py:968] (0/2) Epoch 5, batch 42950, giga_loss[loss=0.3417, simple_loss=0.4024, pruned_loss=0.1405, over 29004.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.395, pruned_loss=0.1427, over 5664642.12 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.394, pruned_loss=0.1447, over 5673988.54 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3952, pruned_loss=0.1425, over 5665061.23 frames. ], batch size: 213, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:56:29,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.643e+02 1.594e+03 2.066e+03 2.851e+03 5.068e+03, threshold=4.132e+03, percent-clipped=1.0
+2023-03-02 23:57:13,816 INFO [train.py:968] (0/2) Epoch 5, batch 43000, giga_loss[loss=0.3678, simple_loss=0.4072, pruned_loss=0.1642, over 28527.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3972, pruned_loss=0.1452, over 5666124.40 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3937, pruned_loss=0.1444, over 5678227.68 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.3976, pruned_loss=0.1454, over 5662321.12 frames. ], batch size: 336, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:57:18,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4238, 1.5645, 1.3704, 2.1209], device='cuda:0'), covar=tensor([0.2132, 0.2077, 0.2126, 0.1928], device='cuda:0'), in_proj_covar=tensor([0.1135, 0.0880, 0.1005, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-02 23:57:50,763 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 23:58:06,396 INFO [train.py:968] (0/2) Epoch 5, batch 43050, giga_loss[loss=0.2774, simple_loss=0.3458, pruned_loss=0.1045, over 28586.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3981, pruned_loss=0.1475, over 5658422.31 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.3942, pruned_loss=0.1447, over 5682343.82 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.3981, pruned_loss=0.1474, over 5651274.18 frames. ], batch size: 71, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:58:10,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.394e+02 1.708e+03 2.207e+03 3.339e+03 5.966e+03, threshold=4.415e+03, percent-clipped=9.0
+2023-03-02 23:58:11,018 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=224955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:58:36,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=224980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:58:57,017 INFO [train.py:968] (0/2) Epoch 5, batch 43100, giga_loss[loss=0.3494, simple_loss=0.3987, pruned_loss=0.15, over 28906.00 frames. ], tot_loss[loss=0.3493, simple_loss=0.3994, pruned_loss=0.1496, over 5660246.90 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3943, pruned_loss=0.1448, over 5681979.25 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.3994, pruned_loss=0.1496, over 5654494.88 frames. ], batch size: 186, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:59:36,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1595, 1.5682, 1.1310, 0.2866], device='cuda:0'), covar=tensor([0.1611, 0.1065, 0.1430, 0.2656], device='cuda:0'), in_proj_covar=tensor([0.1400, 0.1320, 0.1375, 0.1173], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-02 23:59:41,060 INFO [train.py:968] (0/2) Epoch 5, batch 43150, giga_loss[loss=0.3242, simple_loss=0.3869, pruned_loss=0.1308, over 28892.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.3999, pruned_loss=0.15, over 5656709.76 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3948, pruned_loss=0.145, over 5674631.25 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.3996, pruned_loss=0.1499, over 5658861.89 frames. ], batch size: 199, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:59:43,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.472e+02 1.731e+03 2.379e+03 3.118e+03 7.040e+03, threshold=4.758e+03, percent-clipped=17.0
+2023-03-02 23:59:52,007 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-03 00:00:26,817 INFO [train.py:968] (0/2) Epoch 5, batch 43200, giga_loss[loss=0.3049, simple_loss=0.3691, pruned_loss=0.1204, over 28939.00 frames. ], tot_loss[loss=0.346, simple_loss=0.3968, pruned_loss=0.1476, over 5661801.77 frames. ], libri_tot_loss[loss=0.3422, simple_loss=0.3946, pruned_loss=0.1449, over 5671086.49 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.3968, pruned_loss=0.1477, over 5666539.41 frames. ], batch size: 213, lr: 6.05e-03, grad_scale: 4.0
+2023-03-03 00:00:44,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:00:47,767 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:00:48,101 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 00:00:50,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:00:56,915 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:01:01,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4174, 2.9806, 1.3831, 1.4197], device='cuda:0'), covar=tensor([0.0865, 0.0349, 0.0884, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0486, 0.0308, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0026, 0.0016, 0.0021], device='cuda:0')
+2023-03-03 00:01:11,348 INFO [train.py:968] (0/2) Epoch 5, batch 43250, giga_loss[loss=0.3563, simple_loss=0.4177, pruned_loss=0.1474, over 29037.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3951, pruned_loss=0.144, over 5665274.65 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3948, pruned_loss=0.1449, over 5666572.27 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3949, pruned_loss=0.144, over 5672848.05 frames. ], batch size: 164, lr: 6.05e-03, grad_scale: 4.0
+2023-03-03 00:01:17,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.910e+02 1.456e+03 1.959e+03 2.895e+03 5.338e+03, threshold=3.918e+03, percent-clipped=4.0
+2023-03-03 00:01:18,378 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:01:58,824 INFO [train.py:968] (0/2) Epoch 5, batch 43300, giga_loss[loss=0.325, simple_loss=0.376, pruned_loss=0.137, over 28730.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3911, pruned_loss=0.1409, over 5665319.49 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3952, pruned_loss=0.145, over 5670398.02 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3906, pruned_loss=0.1408, over 5667378.07 frames. ], batch size: 99, lr: 6.05e-03, grad_scale: 4.0
+2023-03-03 00:02:27,807 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 00:02:43,905 INFO [train.py:968] (0/2) Epoch 5, batch 43350, libri_loss[loss=0.3212, simple_loss=0.3836, pruned_loss=0.1293, over 29531.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3896, pruned_loss=0.1404, over 5667099.60 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3951, pruned_loss=0.145, over 5677645.49 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3891, pruned_loss=0.1402, over 5661833.40 frames. ], batch size: 81, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:02:46,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.914e+02 1.650e+03 2.275e+03 3.318e+03 1.145e+04, threshold=4.549e+03, percent-clipped=15.0
+2023-03-03 00:03:25,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7162, 1.5512, 1.1518, 1.3942], device='cuda:0'), covar=tensor([0.0590, 0.0586, 0.0919, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0456, 0.0506, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 00:03:29,803 INFO [train.py:968] (0/2) Epoch 5, batch 43400, giga_loss[loss=0.3414, simple_loss=0.398, pruned_loss=0.1424, over 28929.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3884, pruned_loss=0.1405, over 5671759.23 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.395, pruned_loss=0.145, over 5680560.89 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3878, pruned_loss=0.1402, over 5664774.07 frames. ], batch size: 199, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:03:59,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:04:05,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 00:04:11,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4126, 1.8275, 1.6697, 1.5136], device='cuda:0'), covar=tensor([0.1215, 0.1809, 0.1028, 0.1147], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0739, 0.0785, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 00:04:17,653 INFO [train.py:968] (0/2) Epoch 5, batch 43450, giga_loss[loss=0.3503, simple_loss=0.4013, pruned_loss=0.1497, over 28591.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.39, pruned_loss=0.142, over 5665934.30 frames. ], libri_tot_loss[loss=0.3422, simple_loss=0.3947, pruned_loss=0.1448, over 5682911.36 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3898, pruned_loss=0.1419, over 5658291.60 frames. ], batch size: 71, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:04:20,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.631e+03 2.002e+03 2.658e+03 7.532e+03, threshold=4.004e+03, percent-clipped=3.0
+2023-03-03 00:04:31,654 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 00:04:47,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1102, 1.3671, 3.2630, 3.1308], device='cuda:0'), covar=tensor([0.1390, 0.1982, 0.0428, 0.0779], device='cuda:0'), in_proj_covar=tensor([0.0583, 0.0539, 0.0771, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 00:05:03,100 INFO [train.py:968] (0/2) Epoch 5, batch 43500, giga_loss[loss=0.4286, simple_loss=0.4686, pruned_loss=0.1943, over 28604.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3936, pruned_loss=0.1432, over 5677487.45 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3949, pruned_loss=0.145, over 5689642.46 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3932, pruned_loss=0.1429, over 5664976.38 frames. ], batch size: 307, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:05:47,155 INFO [train.py:968] (0/2) Epoch 5, batch 43550, giga_loss[loss=0.3338, simple_loss=0.4013, pruned_loss=0.1332, over 29057.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3964, pruned_loss=0.1428, over 5669788.85 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3947, pruned_loss=0.145, over 5688472.06 frames. ], giga_tot_loss[loss=0.3406, simple_loss=0.3963, pruned_loss=0.1425, over 5660179.02 frames. ], batch size: 128, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:05:50,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.346e+02 1.479e+03 2.138e+03 2.906e+03 6.259e+03, threshold=4.275e+03, percent-clipped=11.0
+2023-03-03 00:06:14,617 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:06:17,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:06:33,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:06:33,506 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-03 00:06:34,420 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:06:37,431 INFO [train.py:968] (0/2) Epoch 5, batch 43600, giga_loss[loss=0.2918, simple_loss=0.3636, pruned_loss=0.11, over 28559.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3981, pruned_loss=0.1437, over 5672311.79 frames. ], libri_tot_loss[loss=0.3417, simple_loss=0.3942, pruned_loss=0.1446, over 5692436.18 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.3986, pruned_loss=0.1438, over 5660185.95 frames. ], batch size: 65, lr: 6.04e-03, grad_scale: 8.0
+2023-03-03 00:06:43,623 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:06:46,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:07:11,083 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:07:24,348 INFO [train.py:968] (0/2) Epoch 5, batch 43650, giga_loss[loss=0.3546, simple_loss=0.4044, pruned_loss=0.1524, over 28811.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.4012, pruned_loss=0.1463, over 5680317.57 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3938, pruned_loss=0.1443, over 5695659.79 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.4022, pruned_loss=0.1467, over 5667196.03 frames. ], batch size: 186, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:07:28,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.443e+02 1.659e+03 2.113e+03 2.718e+03 8.292e+03, threshold=4.227e+03, percent-clipped=10.0
+2023-03-03 00:08:13,578 INFO [train.py:968] (0/2) Epoch 5, batch 43700, giga_loss[loss=0.4311, simple_loss=0.456, pruned_loss=0.2031, over 27509.00 frames. ], tot_loss[loss=0.3496, simple_loss=0.4028, pruned_loss=0.1482, over 5675340.95 frames. ], libri_tot_loss[loss=0.3414, simple_loss=0.3939, pruned_loss=0.1445, over 5698679.69 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4036, pruned_loss=0.1484, over 5662096.23 frames. ], batch size: 472, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:08:44,480 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:08:47,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:08:55,692 INFO [train.py:968] (0/2) Epoch 5, batch 43750, giga_loss[loss=0.3125, simple_loss=0.3744, pruned_loss=0.1253, over 28974.00 frames. ], tot_loss[loss=0.3487, simple_loss=0.4015, pruned_loss=0.1479, over 5680453.54 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.3942, pruned_loss=0.1447, over 5695818.47 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4022, pruned_loss=0.148, over 5671948.02 frames. ], batch size: 119, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:08:56,556 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:08:58,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:08:58,683 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 00:08:58,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.194e+02 1.719e+03 2.362e+03 3.367e+03 1.017e+04, threshold=4.723e+03, percent-clipped=9.0
+2023-03-03 00:09:15,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:09:28,729 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:09:44,215 INFO [train.py:968] (0/2) Epoch 5, batch 43800, giga_loss[loss=0.3242, simple_loss=0.3818, pruned_loss=0.1333, over 28960.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.4001, pruned_loss=0.1477, over 5671140.83 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3939, pruned_loss=0.1443, over 5699459.01 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4011, pruned_loss=0.1482, over 5660882.22 frames. ], batch size: 227, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:10:30,000 INFO [train.py:968] (0/2) Epoch 5, batch 43850, giga_loss[loss=0.3265, simple_loss=0.3869, pruned_loss=0.1331, over 28999.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3975, pruned_loss=0.1465, over 5667436.56 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3938, pruned_loss=0.1443, over 5697647.73 frames. ], giga_tot_loss[loss=0.3463, simple_loss=0.3986, pruned_loss=0.147, over 5659614.01 frames. ], batch size: 155, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:10:34,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.086e+03 1.576e+03 2.111e+03 2.831e+03 8.645e+03, threshold=4.222e+03, percent-clipped=6.0
+2023-03-03 00:11:22,208 INFO [train.py:968] (0/2) Epoch 5, batch 43900, giga_loss[loss=0.2981, simple_loss=0.3697, pruned_loss=0.1133, over 28874.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3966, pruned_loss=0.1466, over 5658310.65 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3939, pruned_loss=0.1443, over 5700503.15 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.3973, pruned_loss=0.147, over 5649263.04 frames. ], batch size: 174, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:11:41,276 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-03 00:11:54,213 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:12:10,439 INFO [train.py:968] (0/2) Epoch 5, batch 43950, giga_loss[loss=0.3627, simple_loss=0.3873, pruned_loss=0.1691, over 23606.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3975, pruned_loss=0.1475, over 5655485.69 frames. ], libri_tot_loss[loss=0.3414, simple_loss=0.3941, pruned_loss=0.1443, over 5701255.71 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.3979, pruned_loss=0.1478, over 5647133.63 frames. ], batch size: 705, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:12:16,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.834e+02 1.641e+03 2.136e+03 2.879e+03 1.097e+04, threshold=4.271e+03, percent-clipped=13.0
+2023-03-03 00:12:20,274 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:12:33,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:12:33,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6260, 2.6155, 1.6870, 0.6392], device='cuda:0'), covar=tensor([0.4121, 0.1940, 0.2323, 0.3820], device='cuda:0'), in_proj_covar=tensor([0.1408, 0.1328, 0.1381, 0.1168], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 00:12:56,139 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:12:56,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2418, 1.6011, 1.2775, 1.4368], device='cuda:0'), covar=tensor([0.0726, 0.0362, 0.0328, 0.0761], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0121, 0.0125, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0059, 0.0044, 0.0039, 0.0066], device='cuda:0')
+2023-03-03 00:12:57,434 INFO [train.py:968] (0/2) Epoch 5, batch 44000, giga_loss[loss=0.333, simple_loss=0.3874, pruned_loss=0.1393, over 28510.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3952, pruned_loss=0.1463, over 5667418.59 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.3936, pruned_loss=0.144, over 5711408.33 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3962, pruned_loss=0.147, over 5649311.30 frames. ], batch size: 71, lr: 6.04e-03, grad_scale: 8.0
+2023-03-03 00:13:06,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:13:09,603 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:13:16,054 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5747, 4.2000, 1.7593, 1.6382], device='cuda:0'), covar=tensor([0.0838, 0.0238, 0.0790, 0.1264], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0486, 0.0310, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0016, 0.0021], device='cuda:0')
+2023-03-03 00:13:19,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3889, 1.5632, 1.3413, 1.5848], device='cuda:0'), covar=tensor([0.1828, 0.1652, 0.1633, 0.1752], device='cuda:0'), in_proj_covar=tensor([0.1140, 0.0877, 0.1010, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 00:13:42,316 INFO [train.py:968] (0/2) Epoch 5, batch 44050, giga_loss[loss=0.4544, simple_loss=0.4652, pruned_loss=0.2218, over 26622.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3948, pruned_loss=0.1461, over 5672989.07 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.3936, pruned_loss=0.144, over 5710973.68 frames. ], giga_tot_loss[loss=0.3444, simple_loss=0.3955, pruned_loss=0.1467, over 5658247.93 frames. ], batch size: 555, lr: 6.03e-03, grad_scale: 8.0
+2023-03-03 00:13:46,173 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225952.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 00:13:47,808 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.132e+02 1.683e+03 2.047e+03 2.928e+03 1.118e+04, threshold=4.094e+03, percent-clipped=9.0
+2023-03-03 00:14:23,968 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-226000.pt
+2023-03-03 00:14:24,280 INFO [train.py:968] (0/2) Epoch 5, batch 44100, giga_loss[loss=0.3055, simple_loss=0.3702, pruned_loss=0.1204, over 28933.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3927, pruned_loss=0.1444, over 5665918.22 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.3926, pruned_loss=0.1433, over 5709466.36 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3942, pruned_loss=0.1455, over 5653023.07 frames. ], batch size: 213, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:14:29,193 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4265, 1.5657, 1.2974, 1.9742], device='cuda:0'), covar=tensor([0.2118, 0.2086, 0.2065, 0.2064], device='cuda:0'), in_proj_covar=tensor([0.1128, 0.0872, 0.0998, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 00:14:36,389 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:14:39,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:15:15,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:15:20,279 INFO [train.py:968] (0/2) Epoch 5, batch 44150, libri_loss[loss=0.336, simple_loss=0.3938, pruned_loss=0.1391, over 29752.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.396, pruned_loss=0.146, over 5657304.17 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.3927, pruned_loss=0.1433, over 5712259.30 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.3972, pruned_loss=0.147, over 5643970.48 frames. ], batch size: 87, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:15:22,466 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:15:24,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.616e+03 1.998e+03 2.762e+03 6.432e+03, threshold=3.996e+03, percent-clipped=7.0
+2023-03-03 00:15:24,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:15:52,221 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:16:08,704 INFO [train.py:968] (0/2) Epoch 5, batch 44200, giga_loss[loss=0.4198, simple_loss=0.4301, pruned_loss=0.2048, over 23700.00 frames. ], tot_loss[loss=0.346, simple_loss=0.3977, pruned_loss=0.1471, over 5658604.37 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3926, pruned_loss=0.1432, over 5714465.69 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.3988, pruned_loss=0.148, over 5645221.21 frames. ], batch size: 705, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:16:16,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9532, 1.7019, 1.2829, 1.4972], device='cuda:0'), covar=tensor([0.0592, 0.0605, 0.1034, 0.0914], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0454, 0.0503, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 00:16:18,201 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6220, 1.9345, 1.9275, 1.7024], device='cuda:0'), covar=tensor([0.1389, 0.1791, 0.1039, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0747, 0.0786, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 00:16:32,178 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-03 00:16:41,241 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-03 00:16:58,334 INFO [train.py:968] (0/2) Epoch 5, batch 44250, giga_loss[loss=0.3132, simple_loss=0.3899, pruned_loss=0.1182, over 28802.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3977, pruned_loss=0.1467, over 5672187.73 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3922, pruned_loss=0.1429, over 5716082.14 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.399, pruned_loss=0.1477, over 5659478.21 frames. ], batch size: 99, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:17:03,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7435, 2.0667, 1.4290, 1.2254], device='cuda:0'), covar=tensor([0.1075, 0.0752, 0.0735, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.1473, 0.1281, 0.1244, 0.1353], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 00:17:03,847 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.559e+02 1.626e+03 2.140e+03 3.053e+03 5.424e+03, threshold=4.280e+03, percent-clipped=10.0
+2023-03-03 00:17:46,204 INFO [train.py:968] (0/2) Epoch 5, batch 44300, libri_loss[loss=0.341, simple_loss=0.4034, pruned_loss=0.1392, over 29542.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3986, pruned_loss=0.1439, over 5677355.53 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3927, pruned_loss=0.1432, over 5719704.55 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.3993, pruned_loss=0.1446, over 5663062.10 frames. ], batch size: 89, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:17:52,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:17:58,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4473, 1.7059, 1.5841, 1.4820], device='cuda:0'), covar=tensor([0.1204, 0.1494, 0.1634, 0.1664], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0731, 0.0631, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 00:18:12,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:18:31,911 INFO [train.py:968] (0/2) Epoch 5, batch 44350, giga_loss[loss=0.3423, simple_loss=0.4068, pruned_loss=0.1389, over 28812.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.4004, pruned_loss=0.1435, over 5675140.57 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3925, pruned_loss=0.143, over 5718801.58 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.4012, pruned_loss=0.1441, over 5664182.43 frames. ], batch size: 199, lr: 6.03e-03, grad_scale: 2.0
+2023-03-03 00:18:36,456 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:18:37,569 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.651e+02 1.692e+03 2.463e+03 4.178e+03 1.390e+04, threshold=4.926e+03, percent-clipped=23.0
+2023-03-03 00:18:59,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:19:11,046 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:19:12,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:19:22,646 INFO [train.py:968] (0/2) Epoch 5, batch 44400, giga_loss[loss=0.3597, simple_loss=0.4183, pruned_loss=0.1506, over 28176.00 frames. ], tot_loss[loss=0.3487, simple_loss=0.4041, pruned_loss=0.1466, over 5663252.39 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3931, pruned_loss=0.1433, over 5721145.18 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.4044, pruned_loss=0.147, over 5651187.59 frames. ], batch size: 77, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:19:49,324 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226327.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 00:20:08,774 INFO [train.py:968] (0/2) Epoch 5, batch 44450, giga_loss[loss=0.3503, simple_loss=0.4035, pruned_loss=0.1486, over 28278.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.4063, pruned_loss=0.1497, over 5656148.16 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.3928, pruned_loss=0.1432, over 5714578.48 frames. ], giga_tot_loss[loss=0.3537, simple_loss=0.4071, pruned_loss=0.1502, over 5651982.23 frames. ], batch size: 65, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:20:11,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:20:13,906 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:20:17,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.173e+02 1.487e+03 2.003e+03 2.658e+03 5.162e+03, threshold=4.006e+03, percent-clipped=2.0
+2023-03-03 00:20:31,726 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:20:36,592 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:20:44,807 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:20:47,654 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-03 00:20:58,802 INFO [train.py:968] (0/2) Epoch 5, batch 44500, giga_loss[loss=0.3473, simple_loss=0.401, pruned_loss=0.1468, over 28763.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.4059, pruned_loss=0.1502, over 5667311.67 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3925, pruned_loss=0.1429, over 5717046.72 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4072, pruned_loss=0.1509, over 5660498.85 frames. ], batch size: 284, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:21:02,474 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:14,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7196, 2.6776, 1.7858, 1.0101], device='cuda:0'), covar=tensor([0.3625, 0.1722, 0.1962, 0.3111], device='cuda:0'), in_proj_covar=tensor([0.1421, 0.1346, 0.1385, 0.1181], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 00:21:14,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:17,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:29,994 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:30,449 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:32,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:44,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:46,065 INFO [train.py:968] (0/2) Epoch 5, batch 44550, giga_loss[loss=0.3452, simple_loss=0.4016, pruned_loss=0.1444, over 27896.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.4049, pruned_loss=0.1495, over 5666968.11 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3924, pruned_loss=0.1428, over 5717993.49 frames. ], giga_tot_loss[loss=0.3532, simple_loss=0.406, pruned_loss=0.1502, over 5660685.49 frames. ], batch size: 412, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:21:51,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.695e+03 2.209e+03 2.667e+03 4.552e+03, threshold=4.417e+03, percent-clipped=4.0
+2023-03-03 00:22:00,006 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:22:03,236 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226470.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 00:22:05,908 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226473.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 00:22:31,043 INFO [train.py:968] (0/2) Epoch 5, batch 44600, giga_loss[loss=0.3183, simple_loss=0.3925, pruned_loss=0.122, over 28938.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4039, pruned_loss=0.1478, over 5671906.56 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3928, pruned_loss=0.1431, over 5717970.76 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4047, pruned_loss=0.1483, over 5665637.66 frames. ], batch size: 186, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:22:32,417 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226502.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 00:22:54,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-03 00:23:16,731 INFO [train.py:968] (0/2) Epoch 5, batch 44650, giga_loss[loss=0.3101, simple_loss=0.3785, pruned_loss=0.1208, over 28468.00 frames. ], tot_loss[loss=0.347, simple_loss=0.4032, pruned_loss=0.1454, over 5675012.30 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3928, pruned_loss=0.1431, over 5712923.34 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.404, pruned_loss=0.1458, over 5672925.10 frames. ], batch size: 85, lr: 6.03e-03, grad_scale: 2.0
+2023-03-03 00:23:24,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9338, 1.8228, 1.3234, 1.6299], device='cuda:0'), covar=tensor([0.0533, 0.0490, 0.0879, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0458, 0.0507, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 00:23:24,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.648e+02 1.576e+03 2.714e+03 3.905e+03 1.243e+04, threshold=5.429e+03, percent-clipped=18.0
+2023-03-03 00:23:59,634 INFO [train.py:968] (0/2) Epoch 5, batch 44700, giga_loss[loss=0.3843, simple_loss=0.4365, pruned_loss=0.166, over 29044.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.4031, pruned_loss=0.1454, over 5674215.85 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3926, pruned_loss=0.1432, over 5709771.13 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.4043, pruned_loss=0.1457, over 5674529.63 frames. ], batch size: 155, lr: 6.03e-03, grad_scale: 2.0
+2023-03-03 00:24:06,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2983, 1.5334, 1.2010, 1.5458], device='cuda:0'), covar=tensor([0.0737, 0.0304, 0.0331, 0.0771], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0121, 0.0124, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0059, 0.0043, 0.0039, 0.0066], device='cuda:0')
+2023-03-03 00:24:25,144 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:24:28,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:24:47,533 INFO [train.py:968] (0/2) Epoch 5, batch 44750, giga_loss[loss=0.3159, simple_loss=0.3801, pruned_loss=0.1258, over 29017.00 frames. ], tot_loss[loss=0.3488, simple_loss=0.4037, pruned_loss=0.147, over 5659387.10 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.392, pruned_loss=0.1425, over 5710967.04 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4057, pruned_loss=0.1479, over 5657366.57 frames. ], batch size: 213, lr: 6.03e-03, grad_scale: 2.0
+2023-03-03 00:24:57,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.424e+02 1.573e+03 1.954e+03 2.767e+03 5.410e+03, threshold=3.908e+03, percent-clipped=0.0
+2023-03-03 00:25:03,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:25:24,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3151, 1.4257, 1.0963, 0.9643], device='cuda:0'), covar=tensor([0.0999, 0.0937, 0.0786, 0.0989], device='cuda:0'), in_proj_covar=tensor([0.1478, 0.1279, 0.1252, 0.1350], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 00:25:34,170 INFO [train.py:968] (0/2) Epoch 5, batch 44800, giga_loss[loss=0.3311, simple_loss=0.3975, pruned_loss=0.1323, over 29074.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.4017, pruned_loss=0.1456, over 5662073.30 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.3919, pruned_loss=0.1425, over 5714262.54 frames. ], giga_tot_loss[loss=0.3482, simple_loss=0.4034, pruned_loss=0.1465, over 5656876.15 frames. ], batch size: 128, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:26:20,300 INFO [train.py:968] (0/2) Epoch 5, batch 44850, giga_loss[loss=0.3133, simple_loss=0.3812, pruned_loss=0.1227, over 28658.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.4006, pruned_loss=0.1458, over 5664057.82 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3928, pruned_loss=0.1431, over 5708421.19 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.4015, pruned_loss=0.1461, over 5663221.55 frames. ], batch size: 262, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:26:28,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.338e+02 1.512e+03 2.053e+03 2.913e+03 6.059e+03, threshold=4.106e+03, percent-clipped=10.0
+2023-03-03 00:26:45,865 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:26:47,385 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-03 00:26:47,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:27:09,460 INFO [train.py:968] (0/2) Epoch 5, batch 44900, giga_loss[loss=0.3205, simple_loss=0.3775, pruned_loss=0.1318, over 29029.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3985, pruned_loss=0.1455, over 5660172.86 frames. ], libri_tot_loss[loss=0.3388, simple_loss=0.3922, pruned_loss=0.1427, over 5711261.60 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3998, pruned_loss=0.1461, over 5656367.13 frames. ], batch size: 106, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:27:14,314 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:27:15,220 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:27:18,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:27:18,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:27:24,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3151, 1.4903, 1.2408, 1.4202], device='cuda:0'), covar=tensor([0.2031, 0.1932, 0.1904, 0.1854], device='cuda:0'), in_proj_covar=tensor([0.1133, 0.0881, 0.1004, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 00:27:47,133 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:27:58,381 INFO [train.py:968] (0/2) Epoch 5, batch 44950, giga_loss[loss=0.3276, simple_loss=0.3837, pruned_loss=0.1357, over 28733.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.3965, pruned_loss=0.1444, over 5661132.45 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3926, pruned_loss=0.1428, over 5712753.39 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3972, pruned_loss=0.1449, over 5656013.31 frames. ], batch size: 262, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:28:06,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.399e+02 1.481e+03 1.952e+03 2.425e+03 4.419e+03, threshold=3.905e+03, percent-clipped=1.0
+2023-03-03 00:28:21,439 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-03 00:28:47,165 INFO [train.py:968] (0/2) Epoch 5, batch 45000, giga_loss[loss=0.3798, simple_loss=0.4106, pruned_loss=0.1745, over 26575.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3963, pruned_loss=0.1456, over 5654002.98 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3927, pruned_loss=0.1429, over 5715475.43 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3968, pruned_loss=0.1459, over 5646955.75 frames. ], batch size: 555, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:28:47,170 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 00:28:55,867 INFO [train.py:1012] (0/2) Epoch 5, validation: loss=0.239, simple_loss=0.3446, pruned_loss=0.06674, over 944034.00 frames. 
+2023-03-03 00:28:55,867 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 00:29:09,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4045, 2.2772, 2.1113, 2.0365], device='cuda:0'), covar=tensor([0.1190, 0.1909, 0.1514, 0.1581], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0747, 0.0640, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 00:29:42,247 INFO [train.py:968] (0/2) Epoch 5, batch 45050, giga_loss[loss=0.2842, simple_loss=0.335, pruned_loss=0.1167, over 23669.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3947, pruned_loss=0.1443, over 5645778.59 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3927, pruned_loss=0.1428, over 5719443.20 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3951, pruned_loss=0.1446, over 5635685.29 frames. ], batch size: 705, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:29:44,023 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:29:46,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:29:47,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.029e+02 1.395e+03 2.366e+03 3.418e+03 9.843e+03, threshold=4.732e+03, percent-clipped=19.0
+2023-03-03 00:30:14,440 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:30:29,100 INFO [train.py:968] (0/2) Epoch 5, batch 45100, giga_loss[loss=0.2915, simple_loss=0.3411, pruned_loss=0.1209, over 23748.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3907, pruned_loss=0.1391, over 5657909.99 frames. ], libri_tot_loss[loss=0.3389, simple_loss=0.3926, pruned_loss=0.1426, over 5723231.05 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3912, pruned_loss=0.1395, over 5645409.86 frames. ], batch size: 705, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:30:29,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=227001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:31:15,404 INFO [train.py:968] (0/2) Epoch 5, batch 45150, giga_loss[loss=0.3178, simple_loss=0.3821, pruned_loss=0.1268, over 28772.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3886, pruned_loss=0.1374, over 5659806.55 frames. ], libri_tot_loss[loss=0.3389, simple_loss=0.3924, pruned_loss=0.1427, over 5727720.54 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.389, pruned_loss=0.1375, over 5644448.21 frames. ], batch size: 284, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:31:16,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0147, 1.2017, 3.8603, 3.1351], device='cuda:0'), covar=tensor([0.1591, 0.2261, 0.0350, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0587, 0.0540, 0.0783, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 00:31:22,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.887e+02 1.317e+03 1.694e+03 2.309e+03 4.761e+03, threshold=3.387e+03, percent-clipped=1.0
+2023-03-03 00:32:01,958 INFO [train.py:968] (0/2) Epoch 5, batch 45200, giga_loss[loss=0.2964, simple_loss=0.365, pruned_loss=0.1139, over 28918.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3884, pruned_loss=0.1376, over 5672583.77 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3924, pruned_loss=0.1428, over 5733314.35 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3885, pruned_loss=0.1374, over 5652965.65 frames. ], batch size: 164, lr: 6.02e-03, grad_scale: 8.0
+2023-03-03 00:32:43,480 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=227144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:32:49,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=227147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:32:51,943 INFO [train.py:968] (0/2) Epoch 5, batch 45250, giga_loss[loss=0.3698, simple_loss=0.4001, pruned_loss=0.1698, over 28801.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3868, pruned_loss=0.1378, over 5686543.81 frames. ], libri_tot_loss[loss=0.3389, simple_loss=0.3922, pruned_loss=0.1428, over 5738061.10 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.387, pruned_loss=0.1375, over 5665276.08 frames. ], batch size: 99, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:32:59,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.851e+02 1.766e+03 2.258e+03 3.242e+03 1.124e+04, threshold=4.516e+03, percent-clipped=21.0
+2023-03-03 00:33:15,326 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:33:20,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4844, 2.2798, 1.6410, 0.6644], device='cuda:0'), covar=tensor([0.2368, 0.1310, 0.2164, 0.2917], device='cuda:0'), in_proj_covar=tensor([0.1415, 0.1342, 0.1378, 0.1186], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 00:33:36,775 INFO [train.py:968] (0/2) Epoch 5, batch 45300, giga_loss[loss=0.2738, simple_loss=0.3452, pruned_loss=0.1011, over 28595.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3878, pruned_loss=0.1389, over 5684435.13 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3924, pruned_loss=0.1428, over 5732189.73 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3876, pruned_loss=0.1385, over 5670219.73 frames. ], batch size: 78, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:34:19,290 INFO [train.py:968] (0/2) Epoch 5, batch 45350, giga_loss[loss=0.2998, simple_loss=0.3747, pruned_loss=0.1125, over 29065.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3894, pruned_loss=0.1389, over 5679938.35 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3926, pruned_loss=0.143, over 5716083.92 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.389, pruned_loss=0.1384, over 5682166.56 frames. ], batch size: 155, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:34:27,995 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.712e+03 2.275e+03 2.984e+03 6.633e+03, threshold=4.549e+03, percent-clipped=4.0
+2023-03-03 00:34:39,359 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-03 00:35:09,940 INFO [train.py:968] (0/2) Epoch 5, batch 45400, giga_loss[loss=0.3223, simple_loss=0.3845, pruned_loss=0.13, over 27920.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3911, pruned_loss=0.1403, over 5664475.97 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3924, pruned_loss=0.1429, over 5716031.85 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3909, pruned_loss=0.14, over 5665401.74 frames. ], batch size: 412, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:35:57,070 INFO [train.py:968] (0/2) Epoch 5, batch 45450, giga_loss[loss=0.3365, simple_loss=0.3943, pruned_loss=0.1393, over 28609.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3906, pruned_loss=0.1398, over 5671666.76 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3926, pruned_loss=0.1429, over 5718688.11 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3903, pruned_loss=0.1394, over 5668979.46 frames. ], batch size: 307, lr: 6.02e-03, grad_scale: 2.0
+2023-03-03 00:36:04,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.581e+02 1.422e+03 1.827e+03 2.354e+03 5.399e+03, threshold=3.654e+03, percent-clipped=4.0
+2023-03-03 00:36:25,222 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 00:36:26,409 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=227382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:36:44,023 INFO [train.py:968] (0/2) Epoch 5, batch 45500, giga_loss[loss=0.3745, simple_loss=0.4198, pruned_loss=0.1646, over 28572.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3915, pruned_loss=0.1409, over 5660394.24 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.3929, pruned_loss=0.1431, over 5718704.87 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3909, pruned_loss=0.1403, over 5657714.36 frames. ], batch size: 336, lr: 6.02e-03, grad_scale: 2.0
+2023-03-03 00:37:10,206 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4204, 2.1122, 1.5063, 0.5954], device='cuda:0'), covar=tensor([0.2885, 0.1408, 0.2088, 0.3091], device='cuda:0'), in_proj_covar=tensor([0.1415, 0.1342, 0.1368, 0.1185], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 00:37:27,215 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 00:37:27,243 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-03 00:37:32,162 INFO [train.py:968] (0/2) Epoch 5, batch 45550, giga_loss[loss=0.3792, simple_loss=0.4035, pruned_loss=0.1775, over 23529.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3927, pruned_loss=0.1418, over 5637446.94 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.393, pruned_loss=0.1432, over 5711584.82 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3922, pruned_loss=0.1413, over 5641170.87 frames. ], batch size: 705, lr: 6.02e-03, grad_scale: 2.0
+2023-03-03 00:37:41,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.697e+03 2.377e+03 3.274e+03 1.590e+04, threshold=4.754e+03, percent-clipped=23.0
+2023-03-03 00:38:19,275 INFO [train.py:968] (0/2) Epoch 5, batch 45600, giga_loss[loss=0.309, simple_loss=0.3719, pruned_loss=0.1231, over 29038.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3933, pruned_loss=0.1415, over 5640483.11 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3932, pruned_loss=0.1433, over 5703808.54 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3927, pruned_loss=0.141, over 5650515.43 frames. ], batch size: 128, lr: 6.01e-03, grad_scale: 4.0
+2023-03-03 00:39:01,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7339, 1.7200, 1.5623, 1.6134], device='cuda:0'), covar=tensor([0.0992, 0.1791, 0.1548, 0.1505], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0750, 0.0647, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 00:39:02,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9282, 1.6902, 1.6612, 1.6285], device='cuda:0'), covar=tensor([0.1049, 0.1841, 0.1629, 0.1571], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0750, 0.0647, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 00:39:09,197 INFO [train.py:968] (0/2) Epoch 5, batch 45650, libri_loss[loss=0.4303, simple_loss=0.4562, pruned_loss=0.2023, over 19383.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3961, pruned_loss=0.1439, over 5614175.55 frames. ], libri_tot_loss[loss=0.3405, simple_loss=0.3937, pruned_loss=0.1437, over 5676891.72 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3952, pruned_loss=0.1431, over 5646719.44 frames. ], batch size: 187, lr: 6.01e-03, grad_scale: 4.0
+2023-03-03 00:39:19,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.996e+02 1.450e+03 1.819e+03 2.254e+03 5.159e+03, threshold=3.639e+03, percent-clipped=2.0
+2023-03-03 00:39:41,526 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.26 vs. limit=2.0
+2023-03-03 00:39:58,617 INFO [train.py:968] (0/2) Epoch 5, batch 45700, giga_loss[loss=0.3298, simple_loss=0.4028, pruned_loss=0.1283, over 28666.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.397, pruned_loss=0.145, over 5596741.77 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3941, pruned_loss=0.1442, over 5640705.12 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3959, pruned_loss=0.1439, over 5653382.05 frames. ], batch size: 307, lr: 6.01e-03, grad_scale: 4.0
+2023-03-03 00:40:50,089 INFO [train.py:968] (0/2) Epoch 5, batch 45750, giga_loss[loss=0.358, simple_loss=0.4174, pruned_loss=0.1493, over 28806.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3983, pruned_loss=0.1448, over 5554861.22 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3952, pruned_loss=0.1453, over 5589356.68 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3965, pruned_loss=0.143, over 5644251.93 frames. ], batch size: 243, lr: 6.01e-03, grad_scale: 4.0
+2023-03-03 00:41:01,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.577e+03 2.373e+03 3.540e+03 9.002e+03, threshold=4.746e+03, percent-clipped=20.0
+2023-03-03 00:41:39,117 INFO [train.py:968] (0/2) Epoch 5, batch 45800, giga_loss[loss=0.3277, simple_loss=0.3918, pruned_loss=0.1318, over 28781.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3984, pruned_loss=0.1446, over 5547261.52 frames. ], libri_tot_loss[loss=0.3438, simple_loss=0.3958, pruned_loss=0.1459, over 5547378.10 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3966, pruned_loss=0.1426, over 5653973.87 frames. ], batch size: 119, lr: 6.01e-03, grad_scale: 4.0
+2023-03-03 00:41:53,284 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-03 00:41:56,506 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-5.pt
+2023-03-03 00:43:14,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4849, 1.4745, 1.1849, 1.2287], device='cuda:0'), covar=tensor([0.0688, 0.0556, 0.1013, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0456, 0.0514, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 00:43:15,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=227757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:43:18,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.351e+02 1.135e+03 1.536e+03 2.253e+03 1.048e+04, threshold=3.072e+03, percent-clipped=3.0
+2023-03-03 00:43:18,801 INFO [train.py:968] (0/2) Epoch 6, batch 50, giga_loss[loss=0.3053, simple_loss=0.3839, pruned_loss=0.1134, over 28740.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3965, pruned_loss=0.1284, over 1261330.10 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3433, pruned_loss=0.09464, over 59024.11 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3988, pruned_loss=0.1299, over 1214674.18 frames. ], batch size: 284, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:43:59,100 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=227804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:44:05,219 INFO [train.py:968] (0/2) Epoch 6, batch 100, giga_loss[loss=0.2657, simple_loss=0.3411, pruned_loss=0.09508, over 28712.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3826, pruned_loss=0.1208, over 2249408.61 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3439, pruned_loss=0.1002, over 232146.83 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3866, pruned_loss=0.123, over 2100967.72 frames. ], batch size: 242, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:44:10,649 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 00:44:13,734 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-03 00:44:34,763 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-03 00:44:52,145 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-03 00:44:52,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.082e+02 1.059e+03 1.322e+03 1.658e+03 4.176e+03, threshold=2.643e+03, percent-clipped=1.0
+2023-03-03 00:44:52,875 INFO [train.py:968] (0/2) Epoch 6, batch 150, giga_loss[loss=0.2906, simple_loss=0.3533, pruned_loss=0.1139, over 28636.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3675, pruned_loss=0.1142, over 3010400.08 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3437, pruned_loss=0.1004, over 288668.28 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3698, pruned_loss=0.1155, over 2864868.63 frames. ], batch size: 336, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:45:24,268 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=227900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:45:26,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=227903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:45:31,371 INFO [train.py:968] (0/2) Epoch 6, batch 200, giga_loss[loss=0.2345, simple_loss=0.3067, pruned_loss=0.08112, over 28959.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3549, pruned_loss=0.1072, over 3611447.81 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3542, pruned_loss=0.1045, over 577587.28 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3556, pruned_loss=0.108, over 3374213.48 frames. ], batch size: 106, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:45:48,070 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:46:12,576 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.004e+02 9.263e+02 1.264e+03 1.682e+03 5.671e+03, threshold=2.527e+03, percent-clipped=6.0
+2023-03-03 00:46:13,368 INFO [train.py:968] (0/2) Epoch 6, batch 250, giga_loss[loss=0.2287, simple_loss=0.3033, pruned_loss=0.07705, over 28610.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3435, pruned_loss=0.1013, over 4068529.20 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3581, pruned_loss=0.1072, over 698529.00 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3427, pruned_loss=0.1011, over 3844097.14 frames. ], batch size: 336, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:46:14,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3857, 1.5523, 1.2251, 1.1056], device='cuda:0'), covar=tensor([0.1218, 0.0938, 0.0718, 0.0982], device='cuda:0'), in_proj_covar=tensor([0.1452, 0.1265, 0.1239, 0.1327], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 00:46:48,799 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-228000.pt
+2023-03-03 00:47:00,224 INFO [train.py:968] (0/2) Epoch 6, batch 300, giga_loss[loss=0.2346, simple_loss=0.3062, pruned_loss=0.08151, over 28745.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3333, pruned_loss=0.09665, over 4421295.28 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.36, pruned_loss=0.1089, over 713706.35 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3318, pruned_loss=0.09598, over 4248844.17 frames. ], batch size: 242, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:47:37,160 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 00:47:42,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.721e+02 9.945e+02 1.290e+03 1.685e+03 4.310e+03, threshold=2.579e+03, percent-clipped=10.0
+2023-03-03 00:47:43,234 INFO [train.py:968] (0/2) Epoch 6, batch 350, giga_loss[loss=0.2534, simple_loss=0.3218, pruned_loss=0.09249, over 28894.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3272, pruned_loss=0.09397, over 4708109.47 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3621, pruned_loss=0.1092, over 867292.18 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3246, pruned_loss=0.0929, over 4534398.92 frames. ], batch size: 145, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:48:03,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:48:23,085 INFO [train.py:968] (0/2) Epoch 6, batch 400, giga_loss[loss=0.2567, simple_loss=0.3026, pruned_loss=0.1054, over 24135.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3223, pruned_loss=0.09151, over 4922318.59 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3626, pruned_loss=0.1091, over 982135.55 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3191, pruned_loss=0.0902, over 4769289.64 frames. ], batch size: 705, lr: 5.61e-03, grad_scale: 8.0
+2023-03-03 00:49:00,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.655e+02 9.074e+02 1.174e+03 1.718e+03 4.405e+03, threshold=2.349e+03, percent-clipped=11.0
+2023-03-03 00:49:01,623 INFO [train.py:968] (0/2) Epoch 6, batch 450, giga_loss[loss=0.2224, simple_loss=0.2837, pruned_loss=0.08056, over 28577.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3196, pruned_loss=0.08988, over 5091962.96 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3649, pruned_loss=0.11, over 1092930.91 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3156, pruned_loss=0.08819, over 4960603.60 frames. ], batch size: 85, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:49:20,793 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:49:49,772 INFO [train.py:968] (0/2) Epoch 6, batch 500, giga_loss[loss=0.2124, simple_loss=0.2875, pruned_loss=0.06863, over 28933.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3157, pruned_loss=0.088, over 5229692.56 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3647, pruned_loss=0.11, over 1117060.26 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3124, pruned_loss=0.08657, over 5123654.18 frames. ], batch size: 213, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:49:57,154 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:50:11,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-03 00:50:16,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4459, 2.0872, 1.6192, 0.7522], device='cuda:0'), covar=tensor([0.2778, 0.1470, 0.2281, 0.2983], device='cuda:0'), in_proj_covar=tensor([0.1411, 0.1335, 0.1374, 0.1179], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 00:50:26,693 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0588, 1.3236, 1.0288, 0.3674], device='cuda:0'), covar=tensor([0.1664, 0.1549, 0.2554, 0.2843], device='cuda:0'), in_proj_covar=tensor([0.1408, 0.1331, 0.1370, 0.1174], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 00:50:31,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.627e+02 9.861e+02 1.240e+03 1.696e+03 5.385e+03, threshold=2.481e+03, percent-clipped=11.0
+2023-03-03 00:50:32,420 INFO [train.py:968] (0/2) Epoch 6, batch 550, giga_loss[loss=0.2375, simple_loss=0.304, pruned_loss=0.0855, over 28825.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3143, pruned_loss=0.08732, over 5339119.53 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3658, pruned_loss=0.1104, over 1212047.18 frames. ], giga_tot_loss[loss=0.2409, simple_loss=0.3105, pruned_loss=0.08568, over 5243987.32 frames. ], batch size: 285, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:51:19,352 INFO [train.py:968] (0/2) Epoch 6, batch 600, giga_loss[loss=0.2391, simple_loss=0.2916, pruned_loss=0.09324, over 23886.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3119, pruned_loss=0.08626, over 5409575.44 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3659, pruned_loss=0.1106, over 1301878.97 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3078, pruned_loss=0.08443, over 5327944.80 frames. ], batch size: 705, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:51:28,112 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:51:31,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:51:59,184 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:52:04,669 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.979e+02 9.694e+02 1.283e+03 1.781e+03 4.340e+03, threshold=2.567e+03, percent-clipped=12.0
+2023-03-03 00:52:04,682 INFO [train.py:968] (0/2) Epoch 6, batch 650, giga_loss[loss=0.2258, simple_loss=0.2922, pruned_loss=0.07968, over 28629.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3112, pruned_loss=0.08611, over 5462858.31 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.368, pruned_loss=0.1118, over 1406731.51 frames. ], giga_tot_loss[loss=0.237, simple_loss=0.3064, pruned_loss=0.08387, over 5394874.65 frames. ], batch size: 92, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:52:40,970 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:52:48,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4662, 1.6871, 0.9752, 1.3086], device='cuda:0'), covar=tensor([0.0914, 0.0743, 0.1703, 0.1213], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0454, 0.0506, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 00:52:48,613 INFO [train.py:968] (0/2) Epoch 6, batch 700, giga_loss[loss=0.2282, simple_loss=0.2974, pruned_loss=0.07948, over 28830.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3091, pruned_loss=0.08489, over 5508512.44 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3704, pruned_loss=0.113, over 1482145.78 frames. ], giga_tot_loss[loss=0.2342, simple_loss=0.3036, pruned_loss=0.08236, over 5459634.64 frames. ], batch size: 186, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:53:35,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.089e+02 9.515e+02 1.297e+03 1.629e+03 5.146e+03, threshold=2.594e+03, percent-clipped=6.0
+2023-03-03 00:53:35,424 INFO [train.py:968] (0/2) Epoch 6, batch 750, giga_loss[loss=0.2218, simple_loss=0.288, pruned_loss=0.07775, over 28875.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3064, pruned_loss=0.08345, over 5558208.58 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3704, pruned_loss=0.1133, over 1568972.37 frames. ], giga_tot_loss[loss=0.2313, simple_loss=0.3009, pruned_loss=0.08085, over 5513707.66 frames. ], batch size: 119, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:53:35,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:53:44,674 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:54:17,312 INFO [train.py:968] (0/2) Epoch 6, batch 800, giga_loss[loss=0.2873, simple_loss=0.3411, pruned_loss=0.1168, over 28855.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3062, pruned_loss=0.08434, over 5587531.90 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3712, pruned_loss=0.1136, over 1655066.91 frames. ], giga_tot_loss[loss=0.2319, simple_loss=0.3004, pruned_loss=0.08163, over 5546219.87 frames. ], batch size: 199, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:55:08,443 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.875e+02 1.009e+03 1.368e+03 1.892e+03 5.776e+03, threshold=2.735e+03, percent-clipped=11.0
+2023-03-03 00:55:08,456 INFO [train.py:968] (0/2) Epoch 6, batch 850, giga_loss[loss=0.3259, simple_loss=0.3935, pruned_loss=0.1291, over 28988.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3181, pruned_loss=0.0911, over 5600211.75 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3722, pruned_loss=0.1144, over 1734633.47 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3122, pruned_loss=0.08829, over 5566268.26 frames. ], batch size: 128, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:55:09,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4672, 1.6240, 1.2903, 1.3617], device='cuda:0'), covar=tensor([0.1236, 0.1524, 0.1707, 0.1601], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0733, 0.0633, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 00:55:39,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:55:47,185 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:55:49,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:55:53,223 INFO [train.py:968] (0/2) Epoch 6, batch 900, libri_loss[loss=0.3503, simple_loss=0.4141, pruned_loss=0.1432, over 29665.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3337, pruned_loss=0.09955, over 5626073.28 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3734, pruned_loss=0.1154, over 1878630.54 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3272, pruned_loss=0.09641, over 5589208.21 frames. ], batch size: 88, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:55:57,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4069, 1.5193, 1.3686, 1.5745], device='cuda:0'), covar=tensor([0.2305, 0.2196, 0.2187, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1139, 0.0886, 0.1015, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 00:56:12,484 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:56:33,803 INFO [train.py:968] (0/2) Epoch 6, batch 950, giga_loss[loss=0.2961, simple_loss=0.3709, pruned_loss=0.1106, over 28836.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3452, pruned_loss=0.1054, over 5648569.67 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3737, pruned_loss=0.1159, over 2048778.27 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3385, pruned_loss=0.1022, over 5614892.55 frames. ], batch size: 99, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:56:34,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.208e+02 1.326e+03 1.732e+03 2.468e+03 9.096e+03, threshold=3.465e+03, percent-clipped=18.0
+2023-03-03 00:57:17,390 INFO [train.py:968] (0/2) Epoch 6, batch 1000, giga_loss[loss=0.2869, simple_loss=0.3483, pruned_loss=0.1128, over 23597.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3523, pruned_loss=0.1082, over 5659251.59 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3742, pruned_loss=0.1162, over 2106792.50 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3466, pruned_loss=0.1055, over 5628806.78 frames. ], batch size: 705, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:57:25,277 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:57:28,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:57:38,385 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:57:40,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:57:58,533 INFO [train.py:968] (0/2) Epoch 6, batch 1050, giga_loss[loss=0.2833, simple_loss=0.3628, pruned_loss=0.1019, over 28945.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3556, pruned_loss=0.1082, over 5662756.64 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3742, pruned_loss=0.1163, over 2125830.54 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.351, pruned_loss=0.1059, over 5637810.38 frames. ], batch size: 164, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:58:00,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.643e+02 1.133e+03 1.570e+03 2.191e+03 4.877e+03, threshold=3.140e+03, percent-clipped=3.0
+2023-03-03 00:58:07,655 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:58:13,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:58:35,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6428, 1.6173, 1.5770, 1.6481], device='cuda:0'), covar=tensor([0.1132, 0.1576, 0.1656, 0.1221], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0730, 0.0634, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 00:58:39,919 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9279, 1.0314, 3.7116, 3.0989], device='cuda:0'), covar=tensor([0.1658, 0.2459, 0.0375, 0.1015], device='cuda:0'), in_proj_covar=tensor([0.0568, 0.0525, 0.0745, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 00:58:39,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6163, 1.6874, 1.5373, 1.6872], device='cuda:0'), covar=tensor([0.1088, 0.1548, 0.1514, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0731, 0.0635, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 00:58:41,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-03 00:58:42,292 INFO [train.py:968] (0/2) Epoch 6, batch 1100, giga_loss[loss=0.292, simple_loss=0.3652, pruned_loss=0.1094, over 29020.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3561, pruned_loss=0.1074, over 5675880.74 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3732, pruned_loss=0.1156, over 2200697.69 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3525, pruned_loss=0.1057, over 5651918.24 frames. ], batch size: 155, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:58:53,330 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228822.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 00:59:00,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6299, 2.0625, 2.0413, 1.8104], device='cuda:0'), covar=tensor([0.1379, 0.1671, 0.1014, 0.1113], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0733, 0.0796, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 00:59:14,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:59:25,761 INFO [train.py:968] (0/2) Epoch 6, batch 1150, giga_loss[loss=0.2869, simple_loss=0.3608, pruned_loss=0.1065, over 28674.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3589, pruned_loss=0.1097, over 5686889.48 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3737, pruned_loss=0.1161, over 2273806.79 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3556, pruned_loss=0.1081, over 5664180.24 frames. ], batch size: 78, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:59:27,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.520e+02 1.126e+03 1.445e+03 2.193e+03 5.501e+03, threshold=2.891e+03, percent-clipped=9.0
+2023-03-03 00:59:32,290 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 01:00:13,578 INFO [train.py:968] (0/2) Epoch 6, batch 1200, giga_loss[loss=0.3259, simple_loss=0.3848, pruned_loss=0.1335, over 27918.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3626, pruned_loss=0.1125, over 5679169.89 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3744, pruned_loss=0.1164, over 2309969.48 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3597, pruned_loss=0.1111, over 5659301.71 frames. ], batch size: 412, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 01:00:21,872 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:00:23,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:00:43,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9167, 1.8367, 1.3744, 1.5878], device='cuda:0'), covar=tensor([0.0552, 0.0478, 0.0884, 0.0740], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0457, 0.0514, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:00:46,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:00:55,031 INFO [train.py:968] (0/2) Epoch 6, batch 1250, giga_loss[loss=0.3355, simple_loss=0.4056, pruned_loss=0.1327, over 28924.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3649, pruned_loss=0.1138, over 5682318.54 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3755, pruned_loss=0.1171, over 2378032.59 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.362, pruned_loss=0.1123, over 5665306.09 frames. ], batch size: 112, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:00:55,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.968e+02 1.127e+03 1.475e+03 1.898e+03 7.110e+03, threshold=2.950e+03, percent-clipped=13.0
+2023-03-03 01:01:07,650 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-03 01:01:24,506 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:01:26,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:01:39,552 INFO [train.py:968] (0/2) Epoch 6, batch 1300, giga_loss[loss=0.3075, simple_loss=0.3779, pruned_loss=0.1186, over 28584.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3669, pruned_loss=0.1137, over 5688815.91 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3748, pruned_loss=0.1167, over 2430543.62 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3647, pruned_loss=0.1126, over 5672526.79 frames. ], batch size: 336, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:01:51,541 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:02:23,525 INFO [train.py:968] (0/2) Epoch 6, batch 1350, giga_loss[loss=0.2807, simple_loss=0.3623, pruned_loss=0.09956, over 28524.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3693, pruned_loss=0.1144, over 5697828.99 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3748, pruned_loss=0.1161, over 2532959.50 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3674, pruned_loss=0.1137, over 5679607.86 frames. ], batch size: 336, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:02:24,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.807e+02 1.102e+03 1.302e+03 1.715e+03 4.558e+03, threshold=2.605e+03, percent-clipped=7.0
+2023-03-03 01:02:30,989 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 01:02:38,463 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1759, 1.8323, 1.3595, 0.3964], device='cuda:0'), covar=tensor([0.1786, 0.0936, 0.1396, 0.1980], device='cuda:0'), in_proj_covar=tensor([0.1406, 0.1332, 0.1379, 0.1169], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 01:02:54,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:02:56,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:03:04,348 INFO [train.py:968] (0/2) Epoch 6, batch 1400, giga_loss[loss=0.283, simple_loss=0.3603, pruned_loss=0.1028, over 28599.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3695, pruned_loss=0.1137, over 5701207.58 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3744, pruned_loss=0.1161, over 2599853.69 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.368, pruned_loss=0.1132, over 5683102.90 frames. ], batch size: 307, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:03:45,963 INFO [train.py:968] (0/2) Epoch 6, batch 1450, giga_loss[loss=0.2872, simple_loss=0.3644, pruned_loss=0.105, over 28840.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3684, pruned_loss=0.112, over 5705685.91 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3742, pruned_loss=0.1157, over 2646086.57 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3673, pruned_loss=0.1117, over 5691304.57 frames. ], batch size: 199, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:03:47,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.902e+02 1.055e+03 1.275e+03 1.818e+03 6.642e+03, threshold=2.551e+03, percent-clipped=9.0
+2023-03-03 01:04:14,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229197.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:04:25,422 INFO [train.py:968] (0/2) Epoch 6, batch 1500, giga_loss[loss=0.2565, simple_loss=0.3408, pruned_loss=0.08606, over 28538.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3653, pruned_loss=0.1089, over 5705500.38 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3732, pruned_loss=0.1149, over 2754707.24 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3646, pruned_loss=0.1089, over 5690874.83 frames. ], batch size: 71, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:04:52,408 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:04:54,754 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:04:55,314 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:04:58,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:05:02,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6929, 1.8469, 1.5319, 1.2649], device='cuda:0'), covar=tensor([0.1473, 0.1146, 0.0931, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.1458, 0.1278, 0.1248, 0.1336], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 01:05:09,774 INFO [train.py:968] (0/2) Epoch 6, batch 1550, giga_loss[loss=0.31, simple_loss=0.3711, pruned_loss=0.1244, over 28843.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3642, pruned_loss=0.1084, over 5708761.61 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.373, pruned_loss=0.1147, over 2785919.52 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3637, pruned_loss=0.1083, over 5695699.36 frames. ], batch size: 199, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:05:11,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.427e+02 9.091e+02 1.136e+03 1.646e+03 4.573e+03, threshold=2.273e+03, percent-clipped=9.0
+2023-03-03 01:05:19,117 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:05:23,516 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:05:30,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9666, 1.9144, 1.7860, 1.7319], device='cuda:0'), covar=tensor([0.1214, 0.1710, 0.1423, 0.1521], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0732, 0.0635, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 01:05:34,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9410, 1.7459, 1.3871, 1.3619], device='cuda:0'), covar=tensor([0.0594, 0.0584, 0.0885, 0.0950], device='cuda:0'), in_proj_covar=tensor([0.0333, 0.0447, 0.0501, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:05:51,941 INFO [train.py:968] (0/2) Epoch 6, batch 1600, giga_loss[loss=0.2917, simple_loss=0.3585, pruned_loss=0.1125, over 28548.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3677, pruned_loss=0.1136, over 5714542.77 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3734, pruned_loss=0.1153, over 2878476.94 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3669, pruned_loss=0.1132, over 5699043.97 frames. ], batch size: 85, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:06:08,814 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:06:19,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229340.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:06:23,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229343.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:06:36,996 INFO [train.py:968] (0/2) Epoch 6, batch 1650, libri_loss[loss=0.2575, simple_loss=0.347, pruned_loss=0.08397, over 29539.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3703, pruned_loss=0.1179, over 5710225.98 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3735, pruned_loss=0.1151, over 2968596.87 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3695, pruned_loss=0.1177, over 5691931.08 frames. ], batch size: 82, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:06:38,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.425e+02 1.244e+03 1.648e+03 2.147e+03 4.748e+03, threshold=3.297e+03, percent-clipped=22.0
+2023-03-03 01:06:45,098 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 01:07:19,021 INFO [train.py:968] (0/2) Epoch 6, batch 1700, libri_loss[loss=0.3288, simple_loss=0.3969, pruned_loss=0.1304, over 29653.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3707, pruned_loss=0.1192, over 5707241.12 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3743, pruned_loss=0.1154, over 3068913.75 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3695, pruned_loss=0.119, over 5686781.41 frames. ], batch size: 88, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:07:25,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1835, 1.0226, 0.9381, 1.3287], device='cuda:0'), covar=tensor([0.0753, 0.0351, 0.0327, 0.0815], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0120, 0.0124, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0059, 0.0043, 0.0039, 0.0066], device='cuda:0')
+2023-03-03 01:08:01,862 INFO [train.py:968] (0/2) Epoch 6, batch 1750, giga_loss[loss=0.2607, simple_loss=0.339, pruned_loss=0.09117, over 28790.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3686, pruned_loss=0.1181, over 5716653.32 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3734, pruned_loss=0.1148, over 3151323.58 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.368, pruned_loss=0.1184, over 5696369.73 frames. ], batch size: 119, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:08:03,808 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.833e+02 1.121e+03 1.386e+03 1.820e+03 3.569e+03, threshold=2.772e+03, percent-clipped=2.0
+2023-03-03 01:08:43,575 INFO [train.py:968] (0/2) Epoch 6, batch 1800, giga_loss[loss=0.2867, simple_loss=0.3542, pruned_loss=0.1095, over 28873.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3687, pruned_loss=0.1186, over 5708075.72 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3738, pruned_loss=0.1149, over 3209232.41 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3678, pruned_loss=0.1188, over 5697451.85 frames. ], batch size: 186, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:09:26,533 INFO [train.py:968] (0/2) Epoch 6, batch 1850, giga_loss[loss=0.2787, simple_loss=0.3592, pruned_loss=0.09906, over 28864.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3677, pruned_loss=0.1167, over 5714541.61 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3737, pruned_loss=0.1147, over 3258527.93 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.367, pruned_loss=0.117, over 5706744.22 frames. ], batch size: 145, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:09:28,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.172e+02 1.144e+03 1.411e+03 2.094e+03 4.121e+03, threshold=2.821e+03, percent-clipped=10.0
+2023-03-03 01:10:08,867 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-03 01:10:09,760 INFO [train.py:968] (0/2) Epoch 6, batch 1900, giga_loss[loss=0.2963, simple_loss=0.3671, pruned_loss=0.1127, over 27868.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3663, pruned_loss=0.1154, over 5700948.61 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3744, pruned_loss=0.1148, over 3341430.19 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3652, pruned_loss=0.1157, over 5696163.04 frames. ], batch size: 412, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:10:34,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5282, 1.5289, 1.2939, 1.2981], device='cuda:0'), covar=tensor([0.0644, 0.0511, 0.0911, 0.0818], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0446, 0.0502, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:10:35,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3199, 2.0502, 1.5209, 1.8377], device='cuda:0'), covar=tensor([0.0580, 0.0621, 0.0879, 0.0911], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0446, 0.0502, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:10:44,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7463, 4.2687, 1.7905, 1.6540], device='cuda:0'), covar=tensor([0.0803, 0.0180, 0.0806, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0473, 0.0307, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-03 01:10:48,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3630, 1.4946, 1.5554, 1.4176], device='cuda:0'), covar=tensor([0.1288, 0.1420, 0.1632, 0.1326], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0741, 0.0641, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 01:10:53,399 INFO [train.py:968] (0/2) Epoch 6, batch 1950, giga_loss[loss=0.2715, simple_loss=0.3479, pruned_loss=0.0976, over 29085.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.361, pruned_loss=0.1117, over 5698513.20 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3741, pruned_loss=0.1144, over 3417500.25 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.36, pruned_loss=0.1121, over 5689099.15 frames. ], batch size: 128, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:10:56,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.279e+02 1.093e+03 1.537e+03 2.131e+03 6.839e+03, threshold=3.074e+03, percent-clipped=14.0
+2023-03-03 01:11:23,999 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:11:33,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:11:39,647 INFO [train.py:968] (0/2) Epoch 6, batch 2000, libri_loss[loss=0.321, simple_loss=0.3924, pruned_loss=0.1249, over 29668.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.355, pruned_loss=0.1083, over 5689689.23 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3742, pruned_loss=0.1144, over 3510194.25 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3535, pruned_loss=0.1085, over 5678993.75 frames. ], batch size: 88, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:12:25,621 INFO [train.py:968] (0/2) Epoch 6, batch 2050, giga_loss[loss=0.2556, simple_loss=0.326, pruned_loss=0.09255, over 28999.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.35, pruned_loss=0.1059, over 5684663.35 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3743, pruned_loss=0.1144, over 3580464.76 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3481, pruned_loss=0.1058, over 5670773.00 frames. ], batch size: 136, lr: 5.58e-03, grad_scale: 8.0
+2023-03-03 01:12:28,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.992e+02 9.377e+02 1.253e+03 1.682e+03 5.607e+03, threshold=2.506e+03, percent-clipped=5.0
+2023-03-03 01:13:13,547 INFO [train.py:968] (0/2) Epoch 6, batch 2100, giga_loss[loss=0.285, simple_loss=0.3591, pruned_loss=0.1055, over 28965.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3488, pruned_loss=0.1049, over 5696226.89 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3739, pruned_loss=0.1141, over 3637856.39 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.347, pruned_loss=0.1048, over 5680339.76 frames. ], batch size: 213, lr: 5.58e-03, grad_scale: 8.0
+2023-03-03 01:13:17,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4399, 3.5621, 1.6068, 1.5624], device='cuda:0'), covar=tensor([0.0925, 0.0223, 0.0787, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0466, 0.0302, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-03 01:13:34,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:13:43,165 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:13:45,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:13:54,720 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 01:13:54,901 INFO [train.py:968] (0/2) Epoch 6, batch 2150, giga_loss[loss=0.2954, simple_loss=0.3504, pruned_loss=0.1203, over 28718.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.35, pruned_loss=0.1057, over 5691664.84 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3742, pruned_loss=0.1144, over 3660804.49 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.348, pruned_loss=0.1053, over 5685080.95 frames. ], batch size: 92, lr: 5.58e-03, grad_scale: 8.0
+2023-03-03 01:13:57,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.629e+02 9.693e+02 1.182e+03 1.434e+03 3.270e+03, threshold=2.363e+03, percent-clipped=3.0
+2023-03-03 01:14:02,483 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:14:08,897 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:14:36,452 INFO [train.py:968] (0/2) Epoch 6, batch 2200, libri_loss[loss=0.3296, simple_loss=0.4035, pruned_loss=0.1279, over 29674.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3489, pruned_loss=0.1049, over 5699445.22 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3749, pruned_loss=0.1148, over 3716528.72 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3462, pruned_loss=0.1042, over 5689144.28 frames. ], batch size: 91, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:15:14,879 INFO [train.py:968] (0/2) Epoch 6, batch 2250, giga_loss[loss=0.2503, simple_loss=0.3189, pruned_loss=0.0908, over 28794.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3458, pruned_loss=0.1033, over 5710721.46 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3749, pruned_loss=0.1146, over 3771014.06 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3431, pruned_loss=0.1026, over 5697822.20 frames. ], batch size: 119, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:15:17,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.847e+02 1.043e+03 1.341e+03 1.953e+03 6.904e+03, threshold=2.681e+03, percent-clipped=17.0
+2023-03-03 01:15:23,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:15:33,091 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-03 01:15:40,569 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-03 01:15:47,665 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-230000.pt
+2023-03-03 01:15:55,222 INFO [train.py:968] (0/2) Epoch 6, batch 2300, giga_loss[loss=0.2599, simple_loss=0.3333, pruned_loss=0.09322, over 28692.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3443, pruned_loss=0.1025, over 5704525.16 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3761, pruned_loss=0.115, over 3814618.68 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3407, pruned_loss=0.1014, over 5698242.69 frames. ], batch size: 284, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:16:01,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5027, 1.4526, 4.9353, 3.5009], device='cuda:0'), covar=tensor([0.1576, 0.2239, 0.0303, 0.0667], device='cuda:0'), in_proj_covar=tensor([0.0566, 0.0523, 0.0742, 0.0613], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 01:16:27,853 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230052.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 01:16:33,658 INFO [train.py:968] (0/2) Epoch 6, batch 2350, giga_loss[loss=0.2335, simple_loss=0.3118, pruned_loss=0.07754, over 29016.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3425, pruned_loss=0.1016, over 5711539.96 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3765, pruned_loss=0.115, over 3861976.83 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3385, pruned_loss=0.1003, over 5706250.51 frames. ], batch size: 155, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:16:36,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.636e+02 8.935e+02 1.199e+03 1.726e+03 7.355e+03, threshold=2.397e+03, percent-clipped=11.0
+2023-03-03 01:16:39,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:16:41,489 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 01:16:51,399 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 01:17:11,988 INFO [train.py:968] (0/2) Epoch 6, batch 2400, giga_loss[loss=0.236, simple_loss=0.3048, pruned_loss=0.08364, over 28781.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3393, pruned_loss=0.1001, over 5722018.96 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3764, pruned_loss=0.1149, over 3902737.33 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3355, pruned_loss=0.09894, over 5714389.15 frames. ], batch size: 99, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:17:30,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4475, 1.5886, 1.2698, 1.2165], device='cuda:0'), covar=tensor([0.1377, 0.0998, 0.0952, 0.1086], device='cuda:0'), in_proj_covar=tensor([0.1424, 0.1246, 0.1247, 0.1322], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 01:17:38,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3569, 1.7469, 1.6701, 1.5185], device='cuda:0'), covar=tensor([0.1679, 0.1991, 0.1276, 0.1350], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0733, 0.0794, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 01:17:49,955 INFO [train.py:968] (0/2) Epoch 6, batch 2450, giga_loss[loss=0.2475, simple_loss=0.3227, pruned_loss=0.08619, over 28684.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3366, pruned_loss=0.09819, over 5728157.93 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3765, pruned_loss=0.1144, over 3968169.79 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3324, pruned_loss=0.0971, over 5719500.86 frames. ], batch size: 284, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:17:52,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.652e+02 9.022e+02 1.157e+03 1.525e+03 7.252e+03, threshold=2.314e+03, percent-clipped=11.0
+2023-03-03 01:18:11,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3542, 2.7111, 1.3761, 1.3693], device='cuda:0'), covar=tensor([0.0791, 0.0358, 0.0786, 0.1198], device='cuda:0'), in_proj_covar=tensor([0.0333, 0.0474, 0.0305, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-03 01:18:31,057 INFO [train.py:968] (0/2) Epoch 6, batch 2500, giga_loss[loss=0.2556, simple_loss=0.3191, pruned_loss=0.09605, over 28561.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.334, pruned_loss=0.0971, over 5716199.04 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3769, pruned_loss=0.1147, over 3987379.53 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.33, pruned_loss=0.09581, over 5716062.11 frames. ], batch size: 85, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:18:31,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:18:31,337 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:18:34,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:18:49,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3914, 1.5016, 1.2900, 1.5575], device='cuda:0'), covar=tensor([0.2150, 0.2070, 0.2145, 0.1956], device='cuda:0'), in_proj_covar=tensor([0.1139, 0.0877, 0.1010, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 01:18:59,168 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:19:00,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:19:12,351 INFO [train.py:968] (0/2) Epoch 6, batch 2550, giga_loss[loss=0.2369, simple_loss=0.3117, pruned_loss=0.08105, over 28897.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.333, pruned_loss=0.09682, over 5716868.88 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3772, pruned_loss=0.1148, over 4016075.98 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3289, pruned_loss=0.0955, over 5713956.35 frames. ], batch size: 213, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:19:16,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.152e+02 9.523e+02 1.253e+03 1.540e+03 1.068e+04, threshold=2.505e+03, percent-clipped=16.0
+2023-03-03 01:19:26,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7286, 3.5750, 3.3589, 1.5974], device='cuda:0'), covar=tensor([0.0608, 0.0586, 0.0745, 0.2255], device='cuda:0'), in_proj_covar=tensor([0.0857, 0.0792, 0.0772, 0.0593], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:19:34,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:19:50,879 INFO [train.py:968] (0/2) Epoch 6, batch 2600, giga_loss[loss=0.2225, simple_loss=0.2966, pruned_loss=0.07417, over 28558.00 frames. ], tot_loss[loss=0.262, simple_loss=0.332, pruned_loss=0.09604, over 5716417.67 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3775, pruned_loss=0.1149, over 4052947.84 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3275, pruned_loss=0.09448, over 5719008.31 frames. ], batch size: 60, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:20:17,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:24,907 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:26,917 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8730, 1.1703, 3.8442, 3.0740], device='cuda:0'), covar=tensor([0.1797, 0.2354, 0.0397, 0.0724], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0531, 0.0755, 0.0623], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 01:20:26,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:29,300 INFO [train.py:968] (0/2) Epoch 6, batch 2650, giga_loss[loss=0.2435, simple_loss=0.3187, pruned_loss=0.08415, over 28459.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3318, pruned_loss=0.09589, over 5715109.26 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3779, pruned_loss=0.1148, over 4094771.04 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3269, pruned_loss=0.09427, over 5716318.97 frames. ], batch size: 78, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:20:34,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.371e+02 9.416e+02 1.243e+03 1.716e+03 5.995e+03, threshold=2.487e+03, percent-clipped=7.0
+2023-03-03 01:20:44,040 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:50,917 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:52,327 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:54,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:21:11,828 INFO [train.py:968] (0/2) Epoch 6, batch 2700, giga_loss[loss=0.3435, simple_loss=0.4079, pruned_loss=0.1395, over 28333.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3355, pruned_loss=0.09865, over 5703670.32 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.378, pruned_loss=0.1149, over 4094669.18 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3315, pruned_loss=0.09728, over 5711657.30 frames. ], batch size: 368, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:21:20,968 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:21:28,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230427.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:21:55,828 INFO [train.py:968] (0/2) Epoch 6, batch 2750, giga_loss[loss=0.2767, simple_loss=0.3401, pruned_loss=0.1067, over 28744.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3419, pruned_loss=0.1029, over 5701386.03 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3781, pruned_loss=0.1153, over 4156960.63 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3372, pruned_loss=0.1011, over 5701397.82 frames. ], batch size: 71, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:22:00,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.983e+02 1.079e+03 1.361e+03 1.902e+03 6.139e+03, threshold=2.723e+03, percent-clipped=11.0
+2023-03-03 01:22:16,782 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:22:20,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:22:37,410 INFO [train.py:968] (0/2) Epoch 6, batch 2800, giga_loss[loss=0.2895, simple_loss=0.3527, pruned_loss=0.1132, over 28541.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3512, pruned_loss=0.1098, over 5700405.76 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3785, pruned_loss=0.1158, over 4208787.38 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3463, pruned_loss=0.1077, over 5695120.11 frames. ], batch size: 71, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:22:44,933 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:23:04,009 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:23:19,369 INFO [train.py:968] (0/2) Epoch 6, batch 2850, giga_loss[loss=0.2913, simple_loss=0.3755, pruned_loss=0.1036, over 28717.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3573, pruned_loss=0.1126, over 5691570.26 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3779, pruned_loss=0.1152, over 4271428.97 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3529, pruned_loss=0.1111, over 5683628.25 frames. ], batch size: 242, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:23:25,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.217e+02 1.193e+03 1.646e+03 2.091e+03 5.265e+03, threshold=3.291e+03, percent-clipped=13.0
+2023-03-03 01:23:28,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230570.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 01:23:30,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230573.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 01:23:42,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2013, 1.5083, 1.2626, 1.0001], device='cuda:0'), covar=tensor([0.2006, 0.1794, 0.1949, 0.1662], device='cuda:0'), in_proj_covar=tensor([0.1129, 0.0868, 0.1002, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 01:23:59,293 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230602.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 01:24:06,206 INFO [train.py:968] (0/2) Epoch 6, batch 2900, giga_loss[loss=0.3434, simple_loss=0.4029, pruned_loss=0.142, over 28003.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3639, pruned_loss=0.1165, over 5664165.27 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3774, pruned_loss=0.115, over 4318927.90 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3601, pruned_loss=0.1154, over 5664644.79 frames. ], batch size: 412, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:24:09,694 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8298, 1.9597, 1.6833, 1.6764], device='cuda:0'), covar=tensor([0.1183, 0.1846, 0.1550, 0.1558], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0727, 0.0632, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 01:24:25,199 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:24:49,565 INFO [train.py:968] (0/2) Epoch 6, batch 2950, giga_loss[loss=0.3444, simple_loss=0.4009, pruned_loss=0.144, over 28678.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3694, pruned_loss=0.1188, over 5680629.21 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3773, pruned_loss=0.1149, over 4350282.89 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3662, pruned_loss=0.118, over 5677069.65 frames. ], batch size: 242, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:24:50,573 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:24:54,352 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.677e+02 1.108e+03 1.369e+03 2.169e+03 6.469e+03, threshold=2.738e+03, percent-clipped=12.0
+2023-03-03 01:25:05,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4710, 4.2294, 4.0785, 1.8572], device='cuda:0'), covar=tensor([0.0418, 0.0512, 0.0639, 0.2122], device='cuda:0'), in_proj_covar=tensor([0.0852, 0.0792, 0.0766, 0.0594], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:25:07,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1549, 1.9921, 1.4090, 1.8565], device='cuda:0'), covar=tensor([0.0624, 0.0606, 0.0897, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0445, 0.0503, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:25:34,484 INFO [train.py:968] (0/2) Epoch 6, batch 3000, giga_loss[loss=0.284, simple_loss=0.3538, pruned_loss=0.1071, over 27977.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.373, pruned_loss=0.1213, over 5675127.98 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3762, pruned_loss=0.1142, over 4392740.84 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.371, pruned_loss=0.1213, over 5669643.41 frames. ], batch size: 412, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:25:34,488 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 01:25:42,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1908, 1.7621, 1.3797, 0.3775], device='cuda:0'), covar=tensor([0.1849, 0.1143, 0.2526, 0.2423], device='cuda:0'), in_proj_covar=tensor([0.1399, 0.1319, 0.1383, 0.1163], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 01:25:42,835 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2441, simple_loss=0.3436, pruned_loss=0.07229, over 944034.00 frames. 
+2023-03-03 01:25:42,836 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 01:26:13,281 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-03 01:26:19,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:26:24,793 INFO [train.py:968] (0/2) Epoch 6, batch 3050, libri_loss[loss=0.3603, simple_loss=0.416, pruned_loss=0.1523, over 28596.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3668, pruned_loss=0.1165, over 5683485.98 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3762, pruned_loss=0.1144, over 4421755.88 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3651, pruned_loss=0.1164, over 5675905.64 frames. ], batch size: 106, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:26:28,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.316e+02 1.117e+03 1.423e+03 1.941e+03 4.322e+03, threshold=2.846e+03, percent-clipped=5.0
+2023-03-03 01:26:53,530 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-03 01:27:03,148 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:27:05,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:27:07,682 INFO [train.py:968] (0/2) Epoch 6, batch 3100, giga_loss[loss=0.2576, simple_loss=0.3376, pruned_loss=0.08884, over 28029.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3639, pruned_loss=0.1141, over 5685448.85 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3771, pruned_loss=0.1151, over 4451045.84 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3617, pruned_loss=0.1135, over 5675647.20 frames. ], batch size: 412, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:27:32,750 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:27:49,883 INFO [train.py:968] (0/2) Epoch 6, batch 3150, giga_loss[loss=0.2744, simple_loss=0.3444, pruned_loss=0.1022, over 28631.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3631, pruned_loss=0.1131, over 5681137.39 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3765, pruned_loss=0.1147, over 4476591.75 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3615, pruned_loss=0.1129, over 5672636.79 frames. ], batch size: 85, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:27:50,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:27:55,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.956e+02 1.074e+03 1.413e+03 1.746e+03 1.013e+04, threshold=2.826e+03, percent-clipped=7.0
+2023-03-03 01:28:21,922 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:28:24,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4662, 1.8289, 1.8088, 1.5776], device='cuda:0'), covar=tensor([0.1501, 0.1795, 0.1133, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0735, 0.0797, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 01:28:24,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:28:31,360 INFO [train.py:968] (0/2) Epoch 6, batch 3200, giga_loss[loss=0.2943, simple_loss=0.3663, pruned_loss=0.1111, over 28974.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3656, pruned_loss=0.1148, over 5683913.46 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3764, pruned_loss=0.1148, over 4497813.91 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3642, pruned_loss=0.1145, over 5674146.68 frames. ], batch size: 145, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:28:37,889 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:28:37,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:28:38,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4676, 2.0898, 1.5465, 0.5818], device='cuda:0'), covar=tensor([0.1993, 0.1301, 0.2021, 0.2400], device='cuda:0'), in_proj_covar=tensor([0.1384, 0.1296, 0.1370, 0.1155], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 01:28:42,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:28:42,926 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1267, 1.4035, 1.1597, 1.0526], device='cuda:0'), covar=tensor([0.1993, 0.1842, 0.2005, 0.1722], device='cuda:0'), in_proj_covar=tensor([0.1127, 0.0871, 0.1005, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 01:28:46,545 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:29:09,866 INFO [train.py:968] (0/2) Epoch 6, batch 3250, giga_loss[loss=0.2945, simple_loss=0.3657, pruned_loss=0.1116, over 29120.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3676, pruned_loss=0.1161, over 5690741.73 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3756, pruned_loss=0.1144, over 4553203.79 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3667, pruned_loss=0.1162, over 5675048.68 frames. ], batch size: 155, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:29:15,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.658e+02 1.279e+03 1.561e+03 2.073e+03 5.115e+03, threshold=3.122e+03, percent-clipped=10.0
+2023-03-03 01:29:18,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0000, 1.1542, 3.7164, 2.9743], device='cuda:0'), covar=tensor([0.1535, 0.2295, 0.0362, 0.0944], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0531, 0.0750, 0.0625], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 01:29:21,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5650, 1.5551, 1.0750, 1.3326], device='cuda:0'), covar=tensor([0.0667, 0.0615, 0.1078, 0.0970], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0443, 0.0504, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:29:51,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:29:52,787 INFO [train.py:968] (0/2) Epoch 6, batch 3300, giga_loss[loss=0.3093, simple_loss=0.3741, pruned_loss=0.1223, over 28670.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3709, pruned_loss=0.1186, over 5695137.94 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3753, pruned_loss=0.1142, over 4583880.23 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3702, pruned_loss=0.1189, over 5680263.24 frames. ], batch size: 262, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:30:16,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-03 01:30:19,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2574, 1.4597, 1.1864, 0.8956], device='cuda:0'), covar=tensor([0.0997, 0.0929, 0.0727, 0.0992], device='cuda:0'), in_proj_covar=tensor([0.1420, 0.1259, 0.1243, 0.1328], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 01:30:35,209 INFO [train.py:968] (0/2) Epoch 6, batch 3350, giga_loss[loss=0.3227, simple_loss=0.3812, pruned_loss=0.1321, over 28558.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3712, pruned_loss=0.119, over 5694857.52 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3754, pruned_loss=0.1142, over 4605473.43 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3706, pruned_loss=0.1193, over 5683499.52 frames. ], batch size: 336, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:30:35,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:30:37,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:30:40,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.271e+02 1.124e+03 1.398e+03 2.057e+03 4.743e+03, threshold=2.797e+03, percent-clipped=6.0
+2023-03-03 01:30:50,165 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-03 01:31:07,193 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:31:19,820 INFO [train.py:968] (0/2) Epoch 6, batch 3400, libri_loss[loss=0.3099, simple_loss=0.3773, pruned_loss=0.1212, over 29563.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3725, pruned_loss=0.1206, over 5693279.32 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3753, pruned_loss=0.1141, over 4631411.54 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.372, pruned_loss=0.1211, over 5680069.21 frames. ], batch size: 78, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:31:53,609 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:31:55,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:32:02,886 INFO [train.py:968] (0/2) Epoch 6, batch 3450, giga_loss[loss=0.3947, simple_loss=0.4492, pruned_loss=0.1701, over 29044.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3739, pruned_loss=0.1215, over 5685719.33 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3755, pruned_loss=0.1143, over 4649563.92 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3733, pruned_loss=0.1219, over 5673070.83 frames. ], batch size: 136, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:32:06,424 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-03 01:32:07,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.553e+02 1.157e+03 1.422e+03 2.285e+03 6.421e+03, threshold=2.845e+03, percent-clipped=16.0
+2023-03-03 01:32:20,382 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:32:21,009 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:32:24,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4280, 1.8038, 1.8059, 1.6540], device='cuda:0'), covar=tensor([0.1187, 0.1390, 0.0908, 0.1010], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0734, 0.0795, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 01:32:42,628 INFO [train.py:968] (0/2) Epoch 6, batch 3500, libri_loss[loss=0.3615, simple_loss=0.4283, pruned_loss=0.1473, over 29110.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3733, pruned_loss=0.1195, over 5695498.40 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.376, pruned_loss=0.1146, over 4661615.18 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3725, pruned_loss=0.1196, over 5684099.56 frames. ], batch size: 101, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:33:00,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:33:22,539 INFO [train.py:968] (0/2) Epoch 6, batch 3550, giga_loss[loss=0.295, simple_loss=0.3737, pruned_loss=0.1082, over 28673.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3724, pruned_loss=0.1176, over 5698675.07 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3756, pruned_loss=0.1144, over 4708467.74 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3718, pruned_loss=0.118, over 5683824.85 frames. ], batch size: 307, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:33:27,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.008e+02 1.113e+03 1.590e+03 2.223e+03 8.549e+03, threshold=3.179e+03, percent-clipped=16.0
+2023-03-03 01:33:40,023 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:33:49,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:33:55,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:34:03,659 INFO [train.py:968] (0/2) Epoch 6, batch 3600, giga_loss[loss=0.3363, simple_loss=0.3922, pruned_loss=0.1402, over 28877.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3709, pruned_loss=0.1162, over 5705659.13 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3749, pruned_loss=0.1141, over 4743528.60 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3709, pruned_loss=0.1167, over 5688653.99 frames. ], batch size: 186, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:34:40,735 INFO [train.py:968] (0/2) Epoch 6, batch 3650, giga_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.114, over 28702.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3685, pruned_loss=0.1152, over 5699747.56 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3746, pruned_loss=0.1139, over 4767287.73 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3687, pruned_loss=0.1159, over 5686136.67 frames. ], batch size: 262, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:34:43,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.05 vs. limit=2.0
+2023-03-03 01:34:48,483 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.374e+02 1.022e+03 1.200e+03 1.557e+03 5.697e+03, threshold=2.401e+03, percent-clipped=3.0
+2023-03-03 01:34:56,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-03 01:34:57,346 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:34:59,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:35:19,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4837, 1.7687, 1.3988, 1.5678], device='cuda:0'), covar=tensor([0.0780, 0.0290, 0.0306, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0123, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0060, 0.0043, 0.0039, 0.0066], device='cuda:0')
+2023-03-03 01:35:21,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4846, 4.2718, 4.0938, 1.8185], device='cuda:0'), covar=tensor([0.0421, 0.0501, 0.0598, 0.2065], device='cuda:0'), in_proj_covar=tensor([0.0857, 0.0790, 0.0764, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:35:23,008 INFO [train.py:968] (0/2) Epoch 6, batch 3700, giga_loss[loss=0.2675, simple_loss=0.3414, pruned_loss=0.09679, over 29041.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3653, pruned_loss=0.1127, over 5703749.43 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3739, pruned_loss=0.1134, over 4798342.31 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3656, pruned_loss=0.1137, over 5693457.05 frames. ], batch size: 128, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:35:23,258 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:35:25,709 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-03 01:35:42,466 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:35:44,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:35:48,858 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:35:51,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:36:02,094 INFO [train.py:968] (0/2) Epoch 6, batch 3750, giga_loss[loss=0.3523, simple_loss=0.4048, pruned_loss=0.1499, over 28743.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3639, pruned_loss=0.1119, over 5708769.48 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3739, pruned_loss=0.1133, over 4824557.63 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.364, pruned_loss=0.1127, over 5697359.82 frames. ], batch size: 262, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:36:02,512 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 01:36:08,341 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:36:08,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.529e+02 9.791e+02 1.212e+03 1.531e+03 2.973e+03, threshold=2.423e+03, percent-clipped=5.0
+2023-03-03 01:36:13,891 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:36:41,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5717, 3.5657, 1.6426, 1.5418], device='cuda:0'), covar=tensor([0.0867, 0.0258, 0.0779, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.0331, 0.0463, 0.0303, 0.0374], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:0')
+2023-03-03 01:36:42,507 INFO [train.py:968] (0/2) Epoch 6, batch 3800, giga_loss[loss=0.2915, simple_loss=0.3652, pruned_loss=0.1089, over 28612.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3656, pruned_loss=0.1133, over 5708928.32 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3738, pruned_loss=0.1132, over 4860767.32 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3655, pruned_loss=0.114, over 5694471.22 frames. ], batch size: 85, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:37:22,978 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:37:23,983 INFO [train.py:968] (0/2) Epoch 6, batch 3850, giga_loss[loss=0.3375, simple_loss=0.3925, pruned_loss=0.1412, over 27577.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3661, pruned_loss=0.1135, over 5712674.28 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3738, pruned_loss=0.1133, over 4871355.97 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3659, pruned_loss=0.114, over 5699336.42 frames. ], batch size: 472, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:37:28,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1883, 1.3830, 1.1333, 1.4562], device='cuda:0'), covar=tensor([0.2129, 0.1932, 0.1983, 0.1720], device='cuda:0'), in_proj_covar=tensor([0.1136, 0.0884, 0.1013, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 01:37:30,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.865e+02 1.078e+03 1.342e+03 2.223e+03 1.159e+04, threshold=2.684e+03, percent-clipped=19.0
+2023-03-03 01:38:02,277 INFO [train.py:968] (0/2) Epoch 6, batch 3900, giga_loss[loss=0.2935, simple_loss=0.3667, pruned_loss=0.1102, over 28205.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3652, pruned_loss=0.1121, over 5716997.42 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3735, pruned_loss=0.1132, over 4901461.49 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3651, pruned_loss=0.1126, over 5701698.19 frames. ], batch size: 77, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:38:08,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4227, 1.7763, 1.8225, 1.6120], device='cuda:0'), covar=tensor([0.1578, 0.1782, 0.1155, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0735, 0.0794, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 01:38:30,722 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:38:42,278 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:38:44,058 INFO [train.py:968] (0/2) Epoch 6, batch 3950, giga_loss[loss=0.2962, simple_loss=0.366, pruned_loss=0.1132, over 28703.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.365, pruned_loss=0.1119, over 5713623.28 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3735, pruned_loss=0.1131, over 4926087.44 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3646, pruned_loss=0.1123, over 5697189.26 frames. ], batch size: 262, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:38:50,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.452e+02 9.201e+02 1.133e+03 1.479e+03 5.104e+03, threshold=2.266e+03, percent-clipped=5.0
+2023-03-03 01:39:18,389 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:39:20,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:39:24,647 INFO [train.py:968] (0/2) Epoch 6, batch 4000, giga_loss[loss=0.3292, simple_loss=0.386, pruned_loss=0.1362, over 28912.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3651, pruned_loss=0.1125, over 5705739.99 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3736, pruned_loss=0.1131, over 4929937.79 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3646, pruned_loss=0.1128, over 5699091.53 frames. ], batch size: 99, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:39:44,258 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:40:02,081 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6179, 1.4727, 1.2228, 1.3418], device='cuda:0'), covar=tensor([0.0516, 0.0407, 0.0744, 0.0773], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0437, 0.0502, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:40:03,153 INFO [train.py:968] (0/2) Epoch 6, batch 4050, libri_loss[loss=0.3099, simple_loss=0.3845, pruned_loss=0.1177, over 29509.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3625, pruned_loss=0.1112, over 5712380.76 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3738, pruned_loss=0.1132, over 4949416.49 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3618, pruned_loss=0.1113, over 5703344.08 frames. ], batch size: 81, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:40:11,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.901e+02 8.982e+02 1.088e+03 1.355e+03 5.438e+03, threshold=2.175e+03, percent-clipped=8.0
+2023-03-03 01:40:27,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1620, 1.6782, 1.3898, 0.3818], device='cuda:0'), covar=tensor([0.2259, 0.1299, 0.2173, 0.2723], device='cuda:0'), in_proj_covar=tensor([0.1382, 0.1288, 0.1359, 0.1146], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 01:40:35,794 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:40:37,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:40:42,366 INFO [train.py:968] (0/2) Epoch 6, batch 4100, giga_loss[loss=0.27, simple_loss=0.3433, pruned_loss=0.0984, over 28917.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3587, pruned_loss=0.109, over 5717797.15 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3738, pruned_loss=0.1133, over 4966627.28 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3579, pruned_loss=0.109, over 5708836.63 frames. ], batch size: 145, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:40:53,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6661, 4.4444, 4.3523, 1.8697], device='cuda:0'), covar=tensor([0.0445, 0.0492, 0.0663, 0.2218], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0799, 0.0768, 0.0603], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:41:01,034 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:41:21,346 INFO [train.py:968] (0/2) Epoch 6, batch 4150, giga_loss[loss=0.2605, simple_loss=0.3311, pruned_loss=0.0949, over 28623.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3577, pruned_loss=0.1088, over 5716717.60 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3737, pruned_loss=0.1131, over 4989767.15 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3569, pruned_loss=0.1088, over 5705455.11 frames. ], batch size: 85, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:41:29,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.729e+02 1.145e+03 1.471e+03 2.087e+03 1.022e+04, threshold=2.943e+03, percent-clipped=22.0
+2023-03-03 01:41:56,740 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 01:42:01,346 INFO [train.py:968] (0/2) Epoch 6, batch 4200, giga_loss[loss=0.2599, simple_loss=0.3311, pruned_loss=0.09441, over 28820.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3583, pruned_loss=0.1098, over 5721142.49 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3735, pruned_loss=0.113, over 5017709.44 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3573, pruned_loss=0.1098, over 5709598.80 frames. ], batch size: 119, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:42:38,455 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3123, 2.8173, 1.4404, 1.3097], device='cuda:0'), covar=tensor([0.0759, 0.0373, 0.0769, 0.1167], device='cuda:0'), in_proj_covar=tensor([0.0328, 0.0469, 0.0302, 0.0374], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-03 01:42:42,674 INFO [train.py:968] (0/2) Epoch 6, batch 4250, giga_loss[loss=0.2509, simple_loss=0.3261, pruned_loss=0.0879, over 28644.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3564, pruned_loss=0.1093, over 5718760.10 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3733, pruned_loss=0.113, over 5035284.20 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3554, pruned_loss=0.1093, over 5706408.44 frames. ], batch size: 242, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:42:43,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 01:42:50,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.275e+02 1.021e+03 1.224e+03 1.535e+03 2.997e+03, threshold=2.448e+03, percent-clipped=2.0
+2023-03-03 01:43:13,838 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-232000.pt
+2023-03-03 01:43:22,634 INFO [train.py:968] (0/2) Epoch 6, batch 4300, giga_loss[loss=0.3096, simple_loss=0.3699, pruned_loss=0.1246, over 28653.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3527, pruned_loss=0.1075, over 5719695.00 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3729, pruned_loss=0.1127, over 5048409.00 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3521, pruned_loss=0.1076, over 5707585.63 frames. ], batch size: 242, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:43:30,935 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:43:47,447 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:44:02,140 INFO [train.py:968] (0/2) Epoch 6, batch 4350, libri_loss[loss=0.3002, simple_loss=0.38, pruned_loss=0.1102, over 29537.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3504, pruned_loss=0.1067, over 5719885.65 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.373, pruned_loss=0.1128, over 5060465.67 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3494, pruned_loss=0.1066, over 5708286.54 frames. ], batch size: 81, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:44:10,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.186e+02 1.047e+03 1.290e+03 1.820e+03 6.506e+03, threshold=2.581e+03, percent-clipped=9.0
+2023-03-03 01:44:10,802 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-03-03 01:44:41,058 INFO [train.py:968] (0/2) Epoch 6, batch 4400, giga_loss[loss=0.2401, simple_loss=0.3101, pruned_loss=0.08508, over 28423.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3504, pruned_loss=0.1068, over 5715807.72 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3732, pruned_loss=0.113, over 5085916.20 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3488, pruned_loss=0.1064, over 5703346.17 frames. ], batch size: 71, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:45:13,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5438, 3.3100, 1.5744, 1.4438], device='cuda:0'), covar=tensor([0.0771, 0.0316, 0.0769, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0328, 0.0472, 0.0304, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-03 01:45:22,006 INFO [train.py:968] (0/2) Epoch 6, batch 4450, giga_loss[loss=0.3208, simple_loss=0.3859, pruned_loss=0.1279, over 27975.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3541, pruned_loss=0.1089, over 5704354.42 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3739, pruned_loss=0.1133, over 5100616.55 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3516, pruned_loss=0.1081, over 5698530.31 frames. ], batch size: 412, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:45:22,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:45:24,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:45:29,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.337e+02 1.028e+03 1.314e+03 1.759e+03 4.877e+03, threshold=2.629e+03, percent-clipped=8.0
+2023-03-03 01:45:48,106 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:45:50,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7228, 1.6588, 1.0429, 1.3977], device='cuda:0'), covar=tensor([0.0955, 0.0977, 0.1772, 0.1401], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0441, 0.0505, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:45:50,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7558, 2.8620, 1.8633, 0.6988], device='cuda:0'), covar=tensor([0.3480, 0.1516, 0.1992, 0.3524], device='cuda:0'), in_proj_covar=tensor([0.1399, 0.1309, 0.1372, 0.1159], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 01:46:01,711 INFO [train.py:968] (0/2) Epoch 6, batch 4500, giga_loss[loss=0.296, simple_loss=0.3711, pruned_loss=0.1105, over 28846.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.355, pruned_loss=0.1087, over 5719030.93 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3725, pruned_loss=0.1127, over 5132381.34 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3533, pruned_loss=0.1085, over 5707302.51 frames. ], batch size: 227, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:46:42,072 INFO [train.py:968] (0/2) Epoch 6, batch 4550, libri_loss[loss=0.2633, simple_loss=0.3389, pruned_loss=0.09383, over 29509.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3564, pruned_loss=0.1086, over 5724526.48 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3721, pruned_loss=0.1125, over 5151089.10 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3551, pruned_loss=0.1084, over 5711427.06 frames. ], batch size: 70, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:46:50,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.577e+02 9.590e+02 1.234e+03 1.693e+03 5.039e+03, threshold=2.469e+03, percent-clipped=8.0
+2023-03-03 01:47:25,106 INFO [train.py:968] (0/2) Epoch 6, batch 4600, giga_loss[loss=0.2686, simple_loss=0.3507, pruned_loss=0.09325, over 28532.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3579, pruned_loss=0.1092, over 5715327.02 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3721, pruned_loss=0.1126, over 5177130.18 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3564, pruned_loss=0.1088, over 5698650.38 frames. ], batch size: 336, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:47:55,523 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-03 01:48:06,383 INFO [train.py:968] (0/2) Epoch 6, batch 4650, giga_loss[loss=0.2734, simple_loss=0.3493, pruned_loss=0.09873, over 28989.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3577, pruned_loss=0.1087, over 5706487.78 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3723, pruned_loss=0.1128, over 5192001.34 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3559, pruned_loss=0.1082, over 5694122.40 frames. ], batch size: 213, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:48:13,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.370e+02 1.105e+03 1.330e+03 1.831e+03 5.673e+03, threshold=2.659e+03, percent-clipped=12.0
+2023-03-03 01:48:42,642 INFO [train.py:968] (0/2) Epoch 6, batch 4700, giga_loss[loss=0.2935, simple_loss=0.3657, pruned_loss=0.1106, over 28985.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3589, pruned_loss=0.1091, over 5694126.33 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3719, pruned_loss=0.1127, over 5204628.43 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3571, pruned_loss=0.1086, over 5695580.83 frames. ], batch size: 145, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:48:46,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:49:26,376 INFO [train.py:968] (0/2) Epoch 6, batch 4750, giga_loss[loss=0.293, simple_loss=0.3655, pruned_loss=0.1102, over 28866.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.359, pruned_loss=0.1094, over 5704504.50 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3713, pruned_loss=0.1123, over 5215287.82 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3579, pruned_loss=0.1093, over 5702553.27 frames. ], batch size: 174, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:49:34,185 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.604e+02 1.265e+03 1.645e+03 2.774e+03 1.048e+04, threshold=3.290e+03, percent-clipped=26.0
+2023-03-03 01:49:52,971 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-03 01:49:58,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6546, 1.6002, 1.5797, 1.5850], device='cuda:0'), covar=tensor([0.1137, 0.1781, 0.1809, 0.1536], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0737, 0.0647, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 01:50:04,859 INFO [train.py:968] (0/2) Epoch 6, batch 4800, giga_loss[loss=0.2774, simple_loss=0.3536, pruned_loss=0.1006, over 28965.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3602, pruned_loss=0.1102, over 5710199.18 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.372, pruned_loss=0.1127, over 5236853.01 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3584, pruned_loss=0.1097, over 5704461.82 frames. ], batch size: 155, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:50:42,087 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:50:44,921 INFO [train.py:968] (0/2) Epoch 6, batch 4850, giga_loss[loss=0.2863, simple_loss=0.3546, pruned_loss=0.109, over 28749.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3623, pruned_loss=0.1117, over 5703091.40 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3721, pruned_loss=0.1128, over 5246163.18 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3605, pruned_loss=0.1112, over 5705304.89 frames. ], batch size: 99, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:50:45,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:50:45,996 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 01:50:51,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.096e+02 1.085e+03 1.368e+03 1.913e+03 4.245e+03, threshold=2.736e+03, percent-clipped=2.0
+2023-03-03 01:51:08,227 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:51:23,707 INFO [train.py:968] (0/2) Epoch 6, batch 4900, giga_loss[loss=0.3024, simple_loss=0.364, pruned_loss=0.1203, over 28860.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3641, pruned_loss=0.1124, over 5701184.57 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3723, pruned_loss=0.1127, over 5256577.87 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3622, pruned_loss=0.112, over 5705249.18 frames. ], batch size: 106, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:51:45,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:51:53,110 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-03 01:52:00,462 INFO [train.py:968] (0/2) Epoch 6, batch 4950, giga_loss[loss=0.2931, simple_loss=0.3668, pruned_loss=0.1096, over 28970.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3662, pruned_loss=0.1133, over 5711137.94 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3723, pruned_loss=0.1126, over 5280651.72 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3644, pruned_loss=0.1131, over 5708406.41 frames. ], batch size: 213, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:52:09,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.236e+02 1.193e+03 1.566e+03 2.144e+03 4.476e+03, threshold=3.132e+03, percent-clipped=13.0
+2023-03-03 01:52:40,063 INFO [train.py:968] (0/2) Epoch 6, batch 5000, giga_loss[loss=0.284, simple_loss=0.3513, pruned_loss=0.1083, over 28291.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3672, pruned_loss=0.114, over 5716647.96 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3722, pruned_loss=0.1127, over 5308577.33 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3655, pruned_loss=0.1138, over 5706156.67 frames. ], batch size: 65, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:53:20,994 INFO [train.py:968] (0/2) Epoch 6, batch 5050, giga_loss[loss=0.3312, simple_loss=0.3914, pruned_loss=0.1355, over 27722.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3689, pruned_loss=0.1156, over 5706283.98 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3728, pruned_loss=0.1132, over 5310194.34 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.367, pruned_loss=0.115, over 5702800.70 frames. ], batch size: 472, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:53:29,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.828e+02 1.275e+03 1.596e+03 2.084e+03 5.556e+03, threshold=3.192e+03, percent-clipped=6.0
+2023-03-03 01:53:32,441 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:53:38,318 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5902, 1.6169, 1.5324, 1.5829], device='cuda:0'), covar=tensor([0.1080, 0.1742, 0.1607, 0.1510], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0729, 0.0639, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 01:54:00,911 INFO [train.py:968] (0/2) Epoch 6, batch 5100, giga_loss[loss=0.2811, simple_loss=0.3535, pruned_loss=0.1044, over 28859.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3686, pruned_loss=0.1151, over 5707643.06 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3731, pruned_loss=0.1132, over 5322593.73 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3668, pruned_loss=0.1147, over 5702477.71 frames. ], batch size: 186, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:54:24,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3714, 2.5640, 1.5519, 1.5300], device='cuda:0'), covar=tensor([0.0664, 0.0287, 0.0646, 0.1014], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0484, 0.0309, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0026, 0.0017, 0.0021], device='cuda:0')
+2023-03-03 01:54:42,175 INFO [train.py:968] (0/2) Epoch 6, batch 5150, giga_loss[loss=0.2595, simple_loss=0.334, pruned_loss=0.09253, over 28789.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3665, pruned_loss=0.1143, over 5711918.23 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3735, pruned_loss=0.1135, over 5328666.05 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3647, pruned_loss=0.1137, over 5705985.20 frames. ], batch size: 119, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:54:51,472 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:54:52,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.597e+02 1.057e+03 1.390e+03 1.857e+03 4.436e+03, threshold=2.781e+03, percent-clipped=7.0
+2023-03-03 01:55:03,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7680, 4.5900, 1.8013, 1.7522], device='cuda:0'), covar=tensor([0.0777, 0.0214, 0.0807, 0.1179], device='cuda:0'), in_proj_covar=tensor([0.0329, 0.0478, 0.0306, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:0')
+2023-03-03 01:55:04,445 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 01:55:23,152 INFO [train.py:968] (0/2) Epoch 6, batch 5200, giga_loss[loss=0.2872, simple_loss=0.3593, pruned_loss=0.1075, over 28690.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3638, pruned_loss=0.1135, over 5714815.26 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3737, pruned_loss=0.1137, over 5340440.50 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.362, pruned_loss=0.1129, over 5706145.45 frames. ], batch size: 119, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:55:23,460 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:55:40,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9694, 1.9964, 1.7315, 1.8406], device='cuda:0'), covar=tensor([0.1151, 0.1915, 0.1654, 0.1566], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0728, 0.0641, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 01:55:43,792 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6029, 1.5285, 1.2116, 1.3274], device='cuda:0'), covar=tensor([0.0589, 0.0516, 0.0932, 0.0812], device='cuda:0'), in_proj_covar=tensor([0.0333, 0.0441, 0.0505, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:56:01,288 INFO [train.py:968] (0/2) Epoch 6, batch 5250, giga_loss[loss=0.2912, simple_loss=0.369, pruned_loss=0.1067, over 28766.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3608, pruned_loss=0.1113, over 5717510.40 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3736, pruned_loss=0.1136, over 5354129.91 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3592, pruned_loss=0.1108, over 5706532.51 frames. ], batch size: 284, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:56:09,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.085e+02 1.031e+03 1.375e+03 1.984e+03 5.652e+03, threshold=2.749e+03, percent-clipped=13.0
+2023-03-03 01:56:28,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6152, 1.7089, 1.4372, 1.3052], device='cuda:0'), covar=tensor([0.1146, 0.0965, 0.0708, 0.0945], device='cuda:0'), in_proj_covar=tensor([0.1445, 0.1273, 0.1255, 0.1337], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 01:56:42,357 INFO [train.py:968] (0/2) Epoch 6, batch 5300, giga_loss[loss=0.2514, simple_loss=0.3262, pruned_loss=0.08831, over 28573.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3615, pruned_loss=0.1102, over 5717365.74 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3732, pruned_loss=0.1134, over 5369190.75 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3602, pruned_loss=0.1099, over 5704814.25 frames. ], batch size: 78, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:56:46,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:57:24,215 INFO [train.py:968] (0/2) Epoch 6, batch 5350, giga_loss[loss=0.3004, simple_loss=0.3783, pruned_loss=0.1112, over 29004.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3633, pruned_loss=0.11, over 5712129.53 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3734, pruned_loss=0.1135, over 5369131.74 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3619, pruned_loss=0.1096, over 5708411.69 frames. ], batch size: 213, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:57:33,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.420e+02 1.059e+03 1.322e+03 1.700e+03 3.395e+03, threshold=2.643e+03, percent-clipped=3.0
+2023-03-03 01:57:36,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2895, 1.7869, 1.1335, 1.0783], device='cuda:0'), covar=tensor([0.1505, 0.0970, 0.1118, 0.1242], device='cuda:0'), in_proj_covar=tensor([0.1438, 0.1264, 0.1250, 0.1326], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 01:58:02,842 INFO [train.py:968] (0/2) Epoch 6, batch 5400, giga_loss[loss=0.3104, simple_loss=0.3609, pruned_loss=0.1299, over 28859.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3629, pruned_loss=0.1112, over 5713168.48 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3733, pruned_loss=0.1138, over 5380921.23 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3615, pruned_loss=0.1105, over 5711573.51 frames. ], batch size: 99, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 01:58:31,488 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:58:38,828 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:58:40,439 INFO [train.py:968] (0/2) Epoch 6, batch 5450, giga_loss[loss=0.273, simple_loss=0.3451, pruned_loss=0.1005, over 28791.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3611, pruned_loss=0.1114, over 5711410.38 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3728, pruned_loss=0.1137, over 5388161.13 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3601, pruned_loss=0.1109, over 5715641.99 frames. ], batch size: 284, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 01:58:41,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:58:49,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.834e+02 1.104e+03 1.401e+03 1.769e+03 5.063e+03, threshold=2.803e+03, percent-clipped=9.0
+2023-03-03 01:58:58,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9926, 1.0571, 3.9710, 3.1296], device='cuda:0'), covar=tensor([0.1677, 0.2365, 0.0378, 0.0644], device='cuda:0'), in_proj_covar=tensor([0.0573, 0.0531, 0.0756, 0.0628], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 01:59:04,905 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:59:21,234 INFO [train.py:968] (0/2) Epoch 6, batch 5500, giga_loss[loss=0.2924, simple_loss=0.3578, pruned_loss=0.1135, over 28972.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3599, pruned_loss=0.1121, over 5713126.43 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3727, pruned_loss=0.1137, over 5390524.44 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3589, pruned_loss=0.1116, over 5721350.31 frames. ], batch size: 213, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 01:59:22,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1635, 4.9821, 4.7345, 2.2553], device='cuda:0'), covar=tensor([0.0335, 0.0471, 0.0680, 0.1821], device='cuda:0'), in_proj_covar=tensor([0.0861, 0.0797, 0.0775, 0.0596], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 01:59:47,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:59:59,680 INFO [train.py:968] (0/2) Epoch 6, batch 5550, giga_loss[loss=0.2713, simple_loss=0.3375, pruned_loss=0.1026, over 28914.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.358, pruned_loss=0.1126, over 5721192.47 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.373, pruned_loss=0.1142, over 5403357.47 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3566, pruned_loss=0.1118, over 5723569.44 frames. ], batch size: 227, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 02:00:09,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.124e+02 1.064e+03 1.296e+03 1.899e+03 2.958e+03, threshold=2.592e+03, percent-clipped=1.0
+2023-03-03 02:00:20,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:00:24,985 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4787, 1.7721, 1.8045, 1.6125], device='cuda:0'), covar=tensor([0.1310, 0.1602, 0.1048, 0.1109], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0725, 0.0784, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 02:00:25,427 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:00:28,171 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:00:40,831 INFO [train.py:968] (0/2) Epoch 6, batch 5600, giga_loss[loss=0.2931, simple_loss=0.3583, pruned_loss=0.114, over 28777.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3581, pruned_loss=0.1129, over 5718885.15 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3733, pruned_loss=0.1144, over 5415671.59 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3561, pruned_loss=0.112, over 5719082.73 frames. ], batch size: 243, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:00:53,161 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:01:20,632 INFO [train.py:968] (0/2) Epoch 6, batch 5650, giga_loss[loss=0.2481, simple_loss=0.3214, pruned_loss=0.08737, over 28800.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3559, pruned_loss=0.1117, over 5716627.38 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3738, pruned_loss=0.115, over 5433833.87 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3533, pruned_loss=0.1104, over 5710676.28 frames. ], batch size: 145, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:01:28,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1184, 3.4740, 1.4141, 1.2069], device='cuda:0'), covar=tensor([0.1135, 0.0397, 0.1004, 0.1653], device='cuda:0'), in_proj_covar=tensor([0.0329, 0.0483, 0.0309, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0026, 0.0017, 0.0021], device='cuda:0')
+2023-03-03 02:01:28,912 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.817e+02 1.237e+03 1.589e+03 2.131e+03 7.890e+03, threshold=3.178e+03, percent-clipped=15.0
+2023-03-03 02:01:36,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:01:38,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3171, 1.7194, 1.6410, 1.5061], device='cuda:0'), covar=tensor([0.1155, 0.1459, 0.0956, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0731, 0.0787, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 02:01:40,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:01:43,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:01:52,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4387, 1.5930, 1.4431, 1.1638], device='cuda:0'), covar=tensor([0.1184, 0.0998, 0.0648, 0.1030], device='cuda:0'), in_proj_covar=tensor([0.1454, 0.1275, 0.1254, 0.1337], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 02:01:57,623 INFO [train.py:968] (0/2) Epoch 6, batch 5700, giga_loss[loss=0.2286, simple_loss=0.3044, pruned_loss=0.0764, over 28927.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3524, pruned_loss=0.1099, over 5721274.90 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3733, pruned_loss=0.1148, over 5447724.51 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3501, pruned_loss=0.1089, over 5711549.00 frames. ], batch size: 174, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:01:57,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:02:04,269 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:02:12,494 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:02:15,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:02:37,323 INFO [train.py:968] (0/2) Epoch 6, batch 5750, libri_loss[loss=0.3088, simple_loss=0.3894, pruned_loss=0.1141, over 29212.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3493, pruned_loss=0.108, over 5716319.03 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.374, pruned_loss=0.1151, over 5459216.82 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3462, pruned_loss=0.1067, over 5706466.95 frames. ], batch size: 97, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:02:37,520 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:02:45,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.593e+02 1.091e+03 1.389e+03 1.918e+03 3.887e+03, threshold=2.777e+03, percent-clipped=2.0
+2023-03-03 02:03:16,258 INFO [train.py:968] (0/2) Epoch 6, batch 5800, giga_loss[loss=0.2506, simple_loss=0.3277, pruned_loss=0.08681, over 28700.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3484, pruned_loss=0.1075, over 5720187.29 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3736, pruned_loss=0.115, over 5472515.26 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3456, pruned_loss=0.1064, over 5707092.96 frames. ], batch size: 119, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:03:45,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3357, 1.6557, 1.3196, 1.4135], device='cuda:0'), covar=tensor([0.0722, 0.0301, 0.0322, 0.0807], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0121, 0.0125, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0060, 0.0044, 0.0039, 0.0067], device='cuda:0')
+2023-03-03 02:03:56,009 INFO [train.py:968] (0/2) Epoch 6, batch 5850, giga_loss[loss=0.2743, simple_loss=0.3496, pruned_loss=0.09947, over 28905.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3511, pruned_loss=0.1085, over 5719495.59 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3734, pruned_loss=0.1148, over 5481264.67 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3486, pruned_loss=0.1076, over 5705672.47 frames. ], batch size: 106, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:04:03,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.900e+02 1.085e+03 1.337e+03 1.682e+03 5.512e+03, threshold=2.675e+03, percent-clipped=6.0
+2023-03-03 02:04:09,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3320, 4.2180, 1.5498, 1.3828], device='cuda:0'), covar=tensor([0.1128, 0.0341, 0.1011, 0.1536], device='cuda:0'), in_proj_covar=tensor([0.0331, 0.0483, 0.0310, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0026, 0.0017, 0.0021], device='cuda:0')
+2023-03-03 02:04:16,990 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2301, 1.4566, 1.1875, 1.2605], device='cuda:0'), covar=tensor([0.2130, 0.2081, 0.2277, 0.2121], device='cuda:0'), in_proj_covar=tensor([0.1141, 0.0872, 0.1007, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 02:04:36,917 INFO [train.py:968] (0/2) Epoch 6, batch 5900, giga_loss[loss=0.3523, simple_loss=0.4122, pruned_loss=0.1462, over 28257.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3556, pruned_loss=0.1104, over 5717269.17 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3737, pruned_loss=0.115, over 5488472.36 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3529, pruned_loss=0.1094, over 5703901.51 frames. ], batch size: 368, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:05:08,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:05:17,169 INFO [train.py:968] (0/2) Epoch 6, batch 5950, giga_loss[loss=0.3591, simple_loss=0.4067, pruned_loss=0.1557, over 28870.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3591, pruned_loss=0.1117, over 5721549.31 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3734, pruned_loss=0.1148, over 5495008.46 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3571, pruned_loss=0.1111, over 5708374.87 frames. ], batch size: 227, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:05:26,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.450e+02 1.145e+03 1.472e+03 2.011e+03 5.025e+03, threshold=2.944e+03, percent-clipped=8.0
+2023-03-03 02:05:57,743 INFO [train.py:968] (0/2) Epoch 6, batch 6000, giga_loss[loss=0.3176, simple_loss=0.3863, pruned_loss=0.1244, over 28661.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3617, pruned_loss=0.1129, over 5719540.26 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3731, pruned_loss=0.1146, over 5503230.08 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.36, pruned_loss=0.1126, over 5706545.34 frames. ], batch size: 336, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:05:57,748 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 02:06:06,026 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.238, simple_loss=0.3404, pruned_loss=0.06783, over 944034.00 frames. 
+2023-03-03 02:06:06,026 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 02:06:50,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:06:51,721 INFO [train.py:968] (0/2) Epoch 6, batch 6050, giga_loss[loss=0.3097, simple_loss=0.3692, pruned_loss=0.1251, over 28609.00 frames. ], tot_loss[loss=0.298, simple_loss=0.365, pruned_loss=0.1155, over 5706454.90 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3732, pruned_loss=0.1146, over 5503445.93 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3635, pruned_loss=0.1152, over 5697004.09 frames. ], batch size: 85, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:06:59,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.281e+02 1.133e+03 1.459e+03 1.895e+03 4.503e+03, threshold=2.917e+03, percent-clipped=6.0
+2023-03-03 02:07:12,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:07:32,877 INFO [train.py:968] (0/2) Epoch 6, batch 6100, giga_loss[loss=0.3497, simple_loss=0.4079, pruned_loss=0.1457, over 28531.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3714, pruned_loss=0.121, over 5702011.09 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3726, pruned_loss=0.1142, over 5511316.86 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5695598.12 frames. ], batch size: 336, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 02:08:20,236 INFO [train.py:968] (0/2) Epoch 6, batch 6150, giga_loss[loss=0.3566, simple_loss=0.423, pruned_loss=0.1451, over 29060.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3775, pruned_loss=0.1257, over 5702759.80 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3725, pruned_loss=0.1142, over 5518406.47 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3771, pruned_loss=0.1262, over 5694362.02 frames. ], batch size: 155, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 02:08:31,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.478e+03 1.986e+03 2.701e+03 7.243e+03, threshold=3.972e+03, percent-clipped=19.0
+2023-03-03 02:08:53,665 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:08:55,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:08:59,977 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 02:09:02,421 INFO [train.py:968] (0/2) Epoch 6, batch 6200, giga_loss[loss=0.3495, simple_loss=0.4031, pruned_loss=0.148, over 28867.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3849, pruned_loss=0.1314, over 5701742.84 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3728, pruned_loss=0.1144, over 5531912.60 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3846, pruned_loss=0.1321, over 5689340.18 frames. ], batch size: 243, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 02:09:19,775 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:09:24,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:09:25,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:09:36,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1908, 1.4377, 1.1754, 1.3698], device='cuda:0'), covar=tensor([0.2000, 0.1896, 0.1983, 0.1623], device='cuda:0'), in_proj_covar=tensor([0.1138, 0.0874, 0.1010, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 02:09:49,080 INFO [train.py:968] (0/2) Epoch 6, batch 6250, giga_loss[loss=0.3789, simple_loss=0.419, pruned_loss=0.1694, over 28497.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3908, pruned_loss=0.1369, over 5703514.99 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3733, pruned_loss=0.1147, over 5538590.51 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3905, pruned_loss=0.1375, over 5690719.79 frames. ], batch size: 336, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:09:49,305 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:09:55,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4007, 1.9005, 1.3955, 0.6107], device='cuda:0'), covar=tensor([0.2270, 0.1211, 0.1673, 0.2578], device='cuda:0'), in_proj_covar=tensor([0.1399, 0.1317, 0.1379, 0.1168], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 02:10:00,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.056e+02 1.455e+03 1.959e+03 2.686e+03 5.084e+03, threshold=3.917e+03, percent-clipped=3.0
+2023-03-03 02:10:07,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6744, 2.1621, 1.4322, 0.8607], device='cuda:0'), covar=tensor([0.2292, 0.1388, 0.1341, 0.2438], device='cuda:0'), in_proj_covar=tensor([0.1397, 0.1315, 0.1378, 0.1168], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 02:10:24,723 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-234000.pt
+2023-03-03 02:10:32,546 INFO [train.py:968] (0/2) Epoch 6, batch 6300, giga_loss[loss=0.3734, simple_loss=0.4236, pruned_loss=0.1616, over 28920.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3972, pruned_loss=0.1424, over 5692023.67 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3734, pruned_loss=0.1147, over 5537520.23 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3972, pruned_loss=0.1435, over 5686132.31 frames. ], batch size: 145, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:10:45,638 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2701, 1.8029, 1.4174, 1.4121], device='cuda:0'), covar=tensor([0.0718, 0.0300, 0.0308, 0.0767], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0120, 0.0123, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0060, 0.0044, 0.0039, 0.0066], device='cuda:0')
+2023-03-03 02:10:49,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:10:53,697 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:11:21,302 INFO [train.py:968] (0/2) Epoch 6, batch 6350, giga_loss[loss=0.3621, simple_loss=0.4109, pruned_loss=0.1566, over 28691.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.4, pruned_loss=0.1454, over 5677287.11 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3727, pruned_loss=0.1142, over 5536965.77 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.4013, pruned_loss=0.1473, over 5675733.36 frames. ], batch size: 242, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:11:36,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.191e+02 1.775e+03 2.320e+03 3.316e+03 8.880e+03, threshold=4.640e+03, percent-clipped=17.0
+2023-03-03 02:12:03,991 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:12:09,463 INFO [train.py:968] (0/2) Epoch 6, batch 6400, giga_loss[loss=0.3933, simple_loss=0.4377, pruned_loss=0.1744, over 28896.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.4026, pruned_loss=0.1488, over 5659117.34 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3726, pruned_loss=0.1143, over 5537165.69 frames. ], giga_tot_loss[loss=0.3531, simple_loss=0.4043, pruned_loss=0.151, over 5660060.48 frames. ], batch size: 186, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:13:03,031 INFO [train.py:968] (0/2) Epoch 6, batch 6450, giga_loss[loss=0.4019, simple_loss=0.4413, pruned_loss=0.1812, over 28682.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.4051, pruned_loss=0.152, over 5665862.37 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3721, pruned_loss=0.114, over 5544756.39 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.4075, pruned_loss=0.1547, over 5661808.70 frames. ], batch size: 262, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:13:14,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:13:18,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:13:18,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.797e+03 2.288e+03 3.186e+03 6.723e+03, threshold=4.576e+03, percent-clipped=6.0
+2023-03-03 02:13:44,990 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 02:13:49,632 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:13:58,059 INFO [train.py:968] (0/2) Epoch 6, batch 6500, giga_loss[loss=0.4363, simple_loss=0.4701, pruned_loss=0.2012, over 28443.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.4094, pruned_loss=0.1568, over 5644894.29 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.372, pruned_loss=0.114, over 5546343.44 frames. ], giga_tot_loss[loss=0.3655, simple_loss=0.4119, pruned_loss=0.1596, over 5641929.31 frames. ], batch size: 60, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:14:46,901 INFO [train.py:968] (0/2) Epoch 6, batch 6550, giga_loss[loss=0.3552, simple_loss=0.4031, pruned_loss=0.1537, over 28860.00 frames. ], tot_loss[loss=0.3645, simple_loss=0.4112, pruned_loss=0.1589, over 5643462.02 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3718, pruned_loss=0.1138, over 5551901.70 frames. ], giga_tot_loss[loss=0.369, simple_loss=0.414, pruned_loss=0.162, over 5638133.64 frames. ], batch size: 186, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:14:59,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.815e+03 2.345e+03 3.415e+03 1.095e+04, threshold=4.691e+03, percent-clipped=9.0
+2023-03-03 02:15:32,926 INFO [train.py:968] (0/2) Epoch 6, batch 6600, giga_loss[loss=0.3624, simple_loss=0.4079, pruned_loss=0.1584, over 28809.00 frames. ], tot_loss[loss=0.3637, simple_loss=0.4101, pruned_loss=0.1587, over 5647076.24 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3723, pruned_loss=0.1139, over 5559142.59 frames. ], giga_tot_loss[loss=0.3685, simple_loss=0.4128, pruned_loss=0.1621, over 5638529.26 frames. ], batch size: 284, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:16:23,949 INFO [train.py:968] (0/2) Epoch 6, batch 6650, giga_loss[loss=0.3032, simple_loss=0.374, pruned_loss=0.1162, over 28796.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.4078, pruned_loss=0.157, over 5627900.71 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3727, pruned_loss=0.1141, over 5557916.26 frames. ], giga_tot_loss[loss=0.3658, simple_loss=0.4104, pruned_loss=0.1606, over 5624335.84 frames. ], batch size: 99, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:16:28,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4114, 1.4822, 1.2193, 1.7808], device='cuda:0'), covar=tensor([0.2170, 0.2174, 0.2245, 0.1859], device='cuda:0'), in_proj_covar=tensor([0.1144, 0.0882, 0.1011, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 02:16:36,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.473e+02 1.478e+03 2.025e+03 2.721e+03 1.796e+04, threshold=4.050e+03, percent-clipped=9.0
+2023-03-03 02:17:03,365 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:17:06,908 INFO [train.py:968] (0/2) Epoch 6, batch 6700, giga_loss[loss=0.3451, simple_loss=0.406, pruned_loss=0.1421, over 28544.00 frames. ], tot_loss[loss=0.3587, simple_loss=0.4074, pruned_loss=0.155, over 5634852.40 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.373, pruned_loss=0.1144, over 5558235.99 frames. ], giga_tot_loss[loss=0.3644, simple_loss=0.4106, pruned_loss=0.1591, over 5634574.62 frames. ], batch size: 336, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:17:14,292 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-03 02:17:15,160 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:17:46,251 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 02:17:54,067 INFO [train.py:968] (0/2) Epoch 6, batch 6750, giga_loss[loss=0.4401, simple_loss=0.4654, pruned_loss=0.2074, over 28285.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4075, pruned_loss=0.1541, over 5640172.34 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3725, pruned_loss=0.114, over 5563414.08 frames. ], giga_tot_loss[loss=0.3642, simple_loss=0.4112, pruned_loss=0.1586, over 5637040.97 frames. ], batch size: 368, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:18:09,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.347e+02 1.528e+03 2.057e+03 2.843e+03 7.722e+03, threshold=4.114e+03, percent-clipped=7.0
+2023-03-03 02:18:11,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:18:43,014 INFO [train.py:968] (0/2) Epoch 6, batch 6800, giga_loss[loss=0.317, simple_loss=0.3834, pruned_loss=0.1253, over 28794.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4073, pruned_loss=0.1542, over 5617975.44 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3721, pruned_loss=0.1139, over 5558490.45 frames. ], giga_tot_loss[loss=0.3645, simple_loss=0.4114, pruned_loss=0.1588, over 5621696.14 frames. ], batch size: 199, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:19:17,280 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 02:19:19,032 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:19:21,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:19:31,882 INFO [train.py:968] (0/2) Epoch 6, batch 6850, giga_loss[loss=0.3586, simple_loss=0.4119, pruned_loss=0.1526, over 28581.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4039, pruned_loss=0.1506, over 5624585.92 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3719, pruned_loss=0.1136, over 5570184.91 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.4086, pruned_loss=0.156, over 5619185.22 frames. ], batch size: 336, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:19:47,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.447e+02 1.531e+03 2.080e+03 2.846e+03 6.465e+03, threshold=4.161e+03, percent-clipped=11.0
+2023-03-03 02:19:54,241 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:20:06,057 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:20:10,266 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-03 02:20:21,418 INFO [train.py:968] (0/2) Epoch 6, batch 6900, giga_loss[loss=0.335, simple_loss=0.3939, pruned_loss=0.138, over 28932.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.4025, pruned_loss=0.1479, over 5641852.85 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3721, pruned_loss=0.1137, over 5573205.73 frames. ], giga_tot_loss[loss=0.3559, simple_loss=0.4065, pruned_loss=0.1526, over 5635669.12 frames. ], batch size: 106, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:20:32,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:20:35,652 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:20:35,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:21:02,934 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:21:08,333 INFO [train.py:968] (0/2) Epoch 6, batch 6950, giga_loss[loss=0.2981, simple_loss=0.3656, pruned_loss=0.1153, over 28467.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3991, pruned_loss=0.145, over 5648415.10 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3719, pruned_loss=0.1136, over 5582359.22 frames. ], giga_tot_loss[loss=0.3514, simple_loss=0.4033, pruned_loss=0.1498, over 5637049.80 frames. ], batch size: 78, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:21:21,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.473e+02 1.650e+03 2.188e+03 3.138e+03 7.698e+03, threshold=4.377e+03, percent-clipped=8.0
+2023-03-03 02:21:30,154 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4621, 2.0558, 1.4421, 0.6089], device='cuda:0'), covar=tensor([0.2332, 0.1242, 0.2126, 0.2997], device='cuda:0'), in_proj_covar=tensor([0.1410, 0.1334, 0.1380, 0.1179], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 02:21:54,451 INFO [train.py:968] (0/2) Epoch 6, batch 7000, giga_loss[loss=0.3362, simple_loss=0.3752, pruned_loss=0.1486, over 23584.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3961, pruned_loss=0.1423, over 5642619.83 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3721, pruned_loss=0.1137, over 5578907.00 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.3998, pruned_loss=0.1466, over 5638870.64 frames. ], batch size: 705, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:22:23,985 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0307, 1.2374, 1.1866, 1.0964], device='cuda:0'), covar=tensor([0.1051, 0.0944, 0.1584, 0.1140], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0729, 0.0640, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 02:22:43,130 INFO [train.py:968] (0/2) Epoch 6, batch 7050, giga_loss[loss=0.4368, simple_loss=0.4576, pruned_loss=0.208, over 27589.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3949, pruned_loss=0.142, over 5644370.20 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.372, pruned_loss=0.1136, over 5586816.16 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3985, pruned_loss=0.1461, over 5636136.62 frames. ], batch size: 472, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:22:55,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.733e+02 1.550e+03 2.036e+03 2.696e+03 5.838e+03, threshold=4.073e+03, percent-clipped=10.0
+2023-03-03 02:23:10,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:23:14,913 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 02:23:16,672 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0474, 1.2411, 1.2641, 1.2327], device='cuda:0'), covar=tensor([0.1201, 0.1064, 0.1734, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0733, 0.0642, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 02:23:29,920 INFO [train.py:968] (0/2) Epoch 6, batch 7100, libri_loss[loss=0.2826, simple_loss=0.3701, pruned_loss=0.09754, over 29531.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3955, pruned_loss=0.1425, over 5642054.84 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3722, pruned_loss=0.1136, over 5582496.26 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.3985, pruned_loss=0.1461, over 5639817.65 frames. ], batch size: 81, lr: 5.52e-03, grad_scale: 2.0
+2023-03-03 02:24:18,868 INFO [train.py:968] (0/2) Epoch 6, batch 7150, giga_loss[loss=0.3007, simple_loss=0.3682, pruned_loss=0.1166, over 28748.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.393, pruned_loss=0.14, over 5652357.64 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3719, pruned_loss=0.1134, over 5594249.95 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3965, pruned_loss=0.1442, over 5641889.00 frames. ], batch size: 262, lr: 5.52e-03, grad_scale: 2.0
+2023-03-03 02:24:33,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.439e+02 1.410e+03 1.788e+03 2.825e+03 1.067e+04, threshold=3.576e+03, percent-clipped=14.0
+2023-03-03 02:25:04,503 INFO [train.py:968] (0/2) Epoch 6, batch 7200, giga_loss[loss=0.3438, simple_loss=0.414, pruned_loss=0.1368, over 28085.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3898, pruned_loss=0.1364, over 5663687.08 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3712, pruned_loss=0.1132, over 5606828.36 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.394, pruned_loss=0.1409, over 5645932.92 frames. ], batch size: 412, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:25:36,188 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:25:36,284 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-03 02:25:39,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:26:02,234 INFO [train.py:968] (0/2) Epoch 6, batch 7250, giga_loss[loss=0.3236, simple_loss=0.399, pruned_loss=0.1241, over 28892.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3911, pruned_loss=0.1351, over 5660517.38 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3708, pruned_loss=0.113, over 5603373.97 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3953, pruned_loss=0.1394, over 5651221.92 frames. ], batch size: 213, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:26:07,653 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:26:10,534 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:26:12,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2549, 1.4567, 1.1073, 1.4518], device='cuda:0'), covar=tensor([0.0752, 0.0348, 0.0335, 0.0794], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0121, 0.0124, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0061, 0.0044, 0.0039, 0.0067], device='cuda:0')
+2023-03-03 02:26:14,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.497e+02 1.440e+03 1.859e+03 2.399e+03 5.445e+03, threshold=3.719e+03, percent-clipped=11.0
+2023-03-03 02:26:30,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2654, 1.4100, 1.2957, 1.4372], device='cuda:0'), covar=tensor([0.0750, 0.0325, 0.0307, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0121, 0.0123, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0060, 0.0044, 0.0039, 0.0067], device='cuda:0')
+2023-03-03 02:26:37,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:26:46,838 INFO [train.py:968] (0/2) Epoch 6, batch 7300, giga_loss[loss=0.3634, simple_loss=0.3907, pruned_loss=0.1681, over 23835.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3924, pruned_loss=0.1351, over 5669336.26 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3707, pruned_loss=0.1131, over 5607496.68 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3968, pruned_loss=0.1394, over 5660932.72 frames. ], batch size: 705, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:27:31,884 INFO [train.py:968] (0/2) Epoch 6, batch 7350, giga_loss[loss=0.3101, simple_loss=0.3839, pruned_loss=0.1181, over 28970.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3929, pruned_loss=0.1363, over 5667992.87 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3704, pruned_loss=0.1129, over 5614891.78 frames. ], giga_tot_loss[loss=0.3391, simple_loss=0.3973, pruned_loss=0.1405, over 5656085.97 frames. ], batch size: 164, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:27:34,172 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:27:48,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.995e+02 1.509e+03 2.112e+03 2.801e+03 6.145e+03, threshold=4.224e+03, percent-clipped=7.0
+2023-03-03 02:28:12,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5168, 1.5743, 1.4059, 1.8475], device='cuda:0'), covar=tensor([0.2062, 0.2083, 0.2061, 0.1919], device='cuda:0'), in_proj_covar=tensor([0.1145, 0.0883, 0.1012, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 02:28:19,223 INFO [train.py:968] (0/2) Epoch 6, batch 7400, giga_loss[loss=0.3061, simple_loss=0.373, pruned_loss=0.1196, over 29001.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3918, pruned_loss=0.1362, over 5673573.86 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3701, pruned_loss=0.1127, over 5620989.50 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.396, pruned_loss=0.1402, over 5659448.36 frames. ], batch size: 136, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:28:20,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:28:24,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:28:26,633 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:28:50,075 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:28:51,898 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:28:51,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:29:06,585 INFO [train.py:968] (0/2) Epoch 6, batch 7450, giga_loss[loss=0.3658, simple_loss=0.4045, pruned_loss=0.1635, over 28691.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3903, pruned_loss=0.1361, over 5674407.18 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.37, pruned_loss=0.1126, over 5628713.87 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3944, pruned_loss=0.1402, over 5657413.24 frames. ], batch size: 243, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:29:18,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:29:19,456 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:29:19,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.963e+02 1.519e+03 2.036e+03 3.087e+03 9.952e+03, threshold=4.072e+03, percent-clipped=12.0
+2023-03-03 02:29:50,429 INFO [train.py:968] (0/2) Epoch 6, batch 7500, giga_loss[loss=0.2911, simple_loss=0.359, pruned_loss=0.1116, over 28900.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3895, pruned_loss=0.1365, over 5673059.05 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3702, pruned_loss=0.1127, over 5624889.44 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3931, pruned_loss=0.1402, over 5664266.36 frames. ], batch size: 145, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:30:37,795 INFO [train.py:968] (0/2) Epoch 6, batch 7550, giga_loss[loss=0.3119, simple_loss=0.386, pruned_loss=0.1189, over 28992.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3879, pruned_loss=0.1342, over 5669677.89 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3693, pruned_loss=0.112, over 5629099.01 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3922, pruned_loss=0.1386, over 5659928.74 frames. ], batch size: 213, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:30:52,644 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:30:53,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.583e+02 1.406e+03 1.894e+03 2.491e+03 7.191e+03, threshold=3.787e+03, percent-clipped=6.0
+2023-03-03 02:31:24,450 INFO [train.py:968] (0/2) Epoch 6, batch 7600, giga_loss[loss=0.3574, simple_loss=0.412, pruned_loss=0.1514, over 28631.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3889, pruned_loss=0.1338, over 5669474.41 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3694, pruned_loss=0.1119, over 5633284.76 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3926, pruned_loss=0.1378, over 5658488.46 frames. ], batch size: 336, lr: 5.52e-03, grad_scale: 8.0
+2023-03-03 02:31:57,640 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:31:59,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:32:07,096 INFO [train.py:968] (0/2) Epoch 6, batch 7650, giga_loss[loss=0.3075, simple_loss=0.3709, pruned_loss=0.122, over 29070.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.388, pruned_loss=0.1331, over 5683355.57 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3688, pruned_loss=0.1118, over 5645831.32 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3923, pruned_loss=0.1374, over 5664973.98 frames. ], batch size: 128, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:32:21,650 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.563e+03 1.908e+03 2.480e+03 4.959e+03, threshold=3.816e+03, percent-clipped=5.0
+2023-03-03 02:32:56,022 INFO [train.py:968] (0/2) Epoch 6, batch 7700, giga_loss[loss=0.3234, simple_loss=0.3805, pruned_loss=0.1331, over 28656.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3871, pruned_loss=0.1329, over 5687509.21 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3687, pruned_loss=0.1116, over 5647180.17 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3907, pruned_loss=0.1364, over 5672137.40 frames. ], batch size: 262, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:33:11,852 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-03 02:33:20,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:33:46,408 INFO [train.py:968] (0/2) Epoch 6, batch 7750, giga_loss[loss=0.2983, simple_loss=0.3575, pruned_loss=0.1195, over 28120.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3865, pruned_loss=0.1341, over 5660759.46 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3685, pruned_loss=0.1117, over 5638412.12 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3898, pruned_loss=0.1373, over 5656462.36 frames. ], batch size: 77, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:34:01,007 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.265e+02 1.596e+03 2.211e+03 2.834e+03 5.240e+03, threshold=4.423e+03, percent-clipped=13.0
+2023-03-03 02:34:14,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8760, 1.1083, 3.7459, 3.0802], device='cuda:0'), covar=tensor([0.1687, 0.2322, 0.0403, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0582, 0.0541, 0.0772, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 02:34:14,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:34:28,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2309, 4.1009, 3.8696, 1.7239], device='cuda:0'), covar=tensor([0.0479, 0.0558, 0.0705, 0.2116], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0844, 0.0810, 0.0618], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 02:34:32,047 INFO [train.py:968] (0/2) Epoch 6, batch 7800, giga_loss[loss=0.3597, simple_loss=0.3911, pruned_loss=0.1641, over 23505.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3862, pruned_loss=0.1345, over 5664663.63 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3686, pruned_loss=0.1116, over 5640837.24 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3892, pruned_loss=0.1377, over 5659401.66 frames. ], batch size: 705, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:35:08,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:35:14,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9155, 1.2441, 3.5342, 2.9143], device='cuda:0'), covar=tensor([0.1559, 0.2187, 0.0400, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0580, 0.0542, 0.0769, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 02:35:20,128 INFO [train.py:968] (0/2) Epoch 6, batch 7850, giga_loss[loss=0.4442, simple_loss=0.4568, pruned_loss=0.2158, over 26744.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.386, pruned_loss=0.1355, over 5662814.71 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3678, pruned_loss=0.1111, over 5647651.23 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3896, pruned_loss=0.1391, over 5653043.66 frames. ], batch size: 555, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:35:34,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.262e+02 1.751e+03 2.158e+03 2.694e+03 6.222e+03, threshold=4.315e+03, percent-clipped=3.0
+2023-03-03 02:35:37,598 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:35:39,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:04,706 INFO [train.py:968] (0/2) Epoch 6, batch 7900, libri_loss[loss=0.2323, simple_loss=0.3113, pruned_loss=0.07665, over 28102.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3838, pruned_loss=0.1344, over 5661240.83 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3675, pruned_loss=0.111, over 5649199.08 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3873, pruned_loss=0.1379, over 5652256.12 frames. ], batch size: 62, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:36:07,111 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:25,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-03 02:36:30,369 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:32,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:40,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:44,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:53,633 INFO [train.py:968] (0/2) Epoch 6, batch 7950, giga_loss[loss=0.3543, simple_loss=0.4002, pruned_loss=0.1543, over 27719.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3836, pruned_loss=0.1345, over 5660354.09 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3674, pruned_loss=0.1108, over 5651834.01 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3866, pruned_loss=0.1376, over 5650899.92 frames. ], batch size: 472, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:37:01,752 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:08,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.869e+02 1.644e+03 2.200e+03 3.601e+03 9.165e+03, threshold=4.400e+03, percent-clipped=16.0
+2023-03-03 02:37:23,140 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:25,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:40,764 INFO [train.py:968] (0/2) Epoch 6, batch 8000, giga_loss[loss=0.3365, simple_loss=0.3964, pruned_loss=0.1383, over 28836.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3857, pruned_loss=0.1355, over 5656543.03 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3677, pruned_loss=0.1111, over 5648645.92 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3882, pruned_loss=0.1383, over 5652341.35 frames. ], batch size: 284, lr: 5.51e-03, grad_scale: 8.0
+2023-03-03 02:37:50,986 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 02:37:53,839 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:54,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:56,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:56,185 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-03 02:38:20,733 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235753.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 02:38:26,546 INFO [train.py:968] (0/2) Epoch 6, batch 8050, giga_loss[loss=0.3369, simple_loss=0.4034, pruned_loss=0.1352, over 28812.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3856, pruned_loss=0.1344, over 5666439.48 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3676, pruned_loss=0.1111, over 5655192.21 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3882, pruned_loss=0.1372, over 5657261.59 frames. ], batch size: 119, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:38:33,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1884, 1.6416, 1.1730, 0.3873], device='cuda:0'), covar=tensor([0.1487, 0.0961, 0.1578, 0.2565], device='cuda:0'), in_proj_covar=tensor([0.1417, 0.1332, 0.1389, 0.1179], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 02:38:40,978 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.027e+02 1.553e+03 1.902e+03 2.905e+03 6.237e+03, threshold=3.804e+03, percent-clipped=6.0
+2023-03-03 02:38:54,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:38:56,395 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:39:10,689 INFO [train.py:968] (0/2) Epoch 6, batch 8100, giga_loss[loss=0.2875, simple_loss=0.3608, pruned_loss=0.1071, over 28530.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3847, pruned_loss=0.1322, over 5681891.05 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3674, pruned_loss=0.1108, over 5659346.60 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3873, pruned_loss=0.1352, over 5671220.01 frames. ], batch size: 65, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:39:25,242 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:39:27,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5870, 5.4412, 5.1259, 2.4287], device='cuda:0'), covar=tensor([0.0372, 0.0530, 0.0713, 0.1913], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0852, 0.0823, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 02:39:48,255 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 02:40:00,077 INFO [train.py:968] (0/2) Epoch 6, batch 8150, giga_loss[loss=0.3173, simple_loss=0.3873, pruned_loss=0.1237, over 28879.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3853, pruned_loss=0.1331, over 5679225.09 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3673, pruned_loss=0.1107, over 5661532.70 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3877, pruned_loss=0.1357, over 5668999.06 frames. ], batch size: 174, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:40:07,805 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:09,558 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:10,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:12,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:15,340 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.759e+02 1.445e+03 1.880e+03 2.790e+03 8.686e+03, threshold=3.760e+03, percent-clipped=13.0
+2023-03-03 02:40:17,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2958, 1.9121, 1.4830, 0.3675], device='cuda:0'), covar=tensor([0.2031, 0.1351, 0.2343, 0.2646], device='cuda:0'), in_proj_covar=tensor([0.1424, 0.1340, 0.1390, 0.1185], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 02:40:22,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3100, 2.0751, 1.3989, 1.7371], device='cuda:0'), covar=tensor([0.0594, 0.0687, 0.0964, 0.0995], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0459, 0.0510, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 02:40:40,593 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:41,932 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:49,486 INFO [train.py:968] (0/2) Epoch 6, batch 8200, giga_loss[loss=0.3454, simple_loss=0.4079, pruned_loss=0.1414, over 28941.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3884, pruned_loss=0.1358, over 5668918.97 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3675, pruned_loss=0.1107, over 5657428.72 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3905, pruned_loss=0.1384, over 5665417.27 frames. ], batch size: 285, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:41:43,018 INFO [train.py:968] (0/2) Epoch 6, batch 8250, giga_loss[loss=0.3137, simple_loss=0.3721, pruned_loss=0.1276, over 28870.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3915, pruned_loss=0.1405, over 5656223.25 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3675, pruned_loss=0.1107, over 5659930.30 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3934, pruned_loss=0.1429, over 5651437.28 frames. ], batch size: 112, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:41:58,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.804e+02 1.819e+03 2.227e+03 3.221e+03 9.229e+03, threshold=4.454e+03, percent-clipped=15.0
+2023-03-03 02:42:05,893 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-03 02:42:21,488 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-236000.pt
+2023-03-03 02:42:33,192 INFO [train.py:968] (0/2) Epoch 6, batch 8300, giga_loss[loss=0.2992, simple_loss=0.3582, pruned_loss=0.1201, over 28865.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3928, pruned_loss=0.1423, over 5666956.15 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3678, pruned_loss=0.1107, over 5664698.26 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3944, pruned_loss=0.1447, over 5658993.43 frames. ], batch size: 112, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:42:45,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:42:52,671 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6418, 1.4712, 5.6562, 3.8449], device='cuda:0'), covar=tensor([0.1582, 0.2314, 0.0277, 0.0585], device='cuda:0'), in_proj_covar=tensor([0.0582, 0.0542, 0.0769, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 02:43:24,676 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-03 02:43:24,917 INFO [train.py:968] (0/2) Epoch 6, batch 8350, giga_loss[loss=0.3101, simple_loss=0.3734, pruned_loss=0.1234, over 28671.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3955, pruned_loss=0.1455, over 5661590.37 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3678, pruned_loss=0.1108, over 5665940.77 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3969, pruned_loss=0.1475, over 5654240.62 frames. ], batch size: 262, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:43:36,405 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=236072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:43:41,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.308e+02 1.692e+03 2.027e+03 3.223e+03 5.454e+03, threshold=4.055e+03, percent-clipped=13.0
+2023-03-03 02:43:46,051 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-03 02:44:02,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3596, 4.2053, 4.0114, 1.9175], device='cuda:0'), covar=tensor([0.0463, 0.0548, 0.0674, 0.2018], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0843, 0.0815, 0.0613], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 02:44:08,812 INFO [train.py:968] (0/2) Epoch 6, batch 8400, giga_loss[loss=0.364, simple_loss=0.416, pruned_loss=0.156, over 28927.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3923, pruned_loss=0.1425, over 5674582.94 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3678, pruned_loss=0.111, over 5674684.91 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3946, pruned_loss=0.1453, over 5660578.31 frames. ], batch size: 213, lr: 5.51e-03, grad_scale: 8.0
+2023-03-03 02:44:09,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2163, 1.3594, 4.1479, 3.1496], device='cuda:0'), covar=tensor([0.1644, 0.2327, 0.0392, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0583, 0.0538, 0.0768, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 02:44:21,885 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236128.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 02:44:52,177 INFO [train.py:968] (0/2) Epoch 6, batch 8450, giga_loss[loss=0.3235, simple_loss=0.3872, pruned_loss=0.1299, over 28683.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3905, pruned_loss=0.1398, over 5686044.04 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3677, pruned_loss=0.111, over 5679110.80 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.393, pruned_loss=0.1428, over 5671127.09 frames. ], batch size: 78, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:44:56,557 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:44:58,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:45:05,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.006e+02 1.418e+03 1.862e+03 2.635e+03 4.136e+03, threshold=3.724e+03, percent-clipped=1.0
+2023-03-03 02:45:11,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=236184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:45:18,446 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1405, 1.3448, 1.1509, 0.8505], device='cuda:0'), covar=tensor([0.1200, 0.1121, 0.0707, 0.1021], device='cuda:0'), in_proj_covar=tensor([0.1482, 0.1308, 0.1281, 0.1374], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 02:45:23,330 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:45:35,473 INFO [train.py:968] (0/2) Epoch 6, batch 8500, giga_loss[loss=0.3318, simple_loss=0.3839, pruned_loss=0.1398, over 27974.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3883, pruned_loss=0.1367, over 5690369.10 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3681, pruned_loss=0.1113, over 5684843.48 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3907, pruned_loss=0.1397, over 5673549.46 frames. ], batch size: 412, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:45:51,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2230, 1.7193, 1.4530, 1.4198], device='cuda:0'), covar=tensor([0.0679, 0.0371, 0.0296, 0.0685], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0120, 0.0124, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0060, 0.0044, 0.0040, 0.0067], device='cuda:0')
+2023-03-03 02:45:55,437 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-03 02:46:02,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1980, 1.4323, 1.2395, 0.9469], device='cuda:0'), covar=tensor([0.1115, 0.1003, 0.0674, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.1479, 0.1305, 0.1276, 0.1372], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 02:46:18,661 INFO [train.py:968] (0/2) Epoch 6, batch 8550, giga_loss[loss=0.2953, simple_loss=0.3609, pruned_loss=0.1149, over 29105.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3859, pruned_loss=0.1352, over 5689322.86 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3683, pruned_loss=0.1114, over 5688298.94 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3879, pruned_loss=0.1378, over 5673023.93 frames. ], batch size: 128, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:46:29,522 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236271.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 02:46:32,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236274.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 02:46:37,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.310e+02 1.370e+03 1.851e+03 2.458e+03 5.164e+03, threshold=3.702e+03, percent-clipped=6.0
+2023-03-03 02:47:00,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236303.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 02:47:06,298 INFO [train.py:968] (0/2) Epoch 6, batch 8600, giga_loss[loss=0.3698, simple_loss=0.412, pruned_loss=0.1638, over 28966.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3849, pruned_loss=0.1352, over 5685046.38 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3688, pruned_loss=0.1119, over 5690349.97 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3864, pruned_loss=0.1375, over 5669844.70 frames. ], batch size: 145, lr: 5.51e-03, grad_scale: 2.0
+2023-03-03 02:47:53,121 INFO [train.py:968] (0/2) Epoch 6, batch 8650, giga_loss[loss=0.3447, simple_loss=0.3935, pruned_loss=0.1479, over 27495.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3848, pruned_loss=0.1356, over 5679398.90 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3689, pruned_loss=0.1119, over 5694359.74 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3865, pruned_loss=0.1382, over 5663121.57 frames. ], batch size: 472, lr: 5.51e-03, grad_scale: 2.0
+2023-03-03 02:47:56,280 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4359, 1.6517, 1.2391, 1.1473], device='cuda:0'), covar=tensor([0.1265, 0.1079, 0.0916, 0.1151], device='cuda:0'), in_proj_covar=tensor([0.1478, 0.1309, 0.1278, 0.1369], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 02:48:11,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.882e+02 1.760e+03 2.632e+03 3.668e+03 1.530e+04, threshold=5.264e+03, percent-clipped=24.0
+2023-03-03 02:48:41,100 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-03 02:48:42,612 INFO [train.py:968] (0/2) Epoch 6, batch 8700, giga_loss[loss=0.3405, simple_loss=0.3948, pruned_loss=0.1431, over 28790.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3894, pruned_loss=0.1386, over 5685628.18 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3687, pruned_loss=0.1117, over 5699529.41 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3913, pruned_loss=0.1414, over 5667814.85 frames. ], batch size: 284, lr: 5.51e-03, grad_scale: 2.0
+2023-03-03 02:48:42,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=236411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:49:15,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:49:28,879 INFO [train.py:968] (0/2) Epoch 6, batch 8750, giga_loss[loss=0.3156, simple_loss=0.3937, pruned_loss=0.1187, over 29031.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3922, pruned_loss=0.1381, over 5686973.44 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3685, pruned_loss=0.1118, over 5707962.37 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3947, pruned_loss=0.1414, over 5664384.48 frames. ], batch size: 128, lr: 5.51e-03, grad_scale: 2.0
+2023-03-03 02:49:46,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.638e+02 1.372e+03 1.739e+03 2.463e+03 8.858e+03, threshold=3.479e+03, percent-clipped=1.0
+2023-03-03 02:50:15,932 INFO [train.py:968] (0/2) Epoch 6, batch 8800, giga_loss[loss=0.3215, simple_loss=0.3835, pruned_loss=0.1298, over 28999.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3949, pruned_loss=0.1385, over 5689269.47 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3686, pruned_loss=0.1119, over 5712176.26 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3974, pruned_loss=0.1415, over 5667179.68 frames. ], batch size: 155, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:50:59,753 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:51:00,770 INFO [train.py:968] (0/2) Epoch 6, batch 8850, giga_loss[loss=0.3679, simple_loss=0.4202, pruned_loss=0.1578, over 28922.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3963, pruned_loss=0.1395, over 5690952.51 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3685, pruned_loss=0.1119, over 5715483.35 frames. ], giga_tot_loss[loss=0.3429, simple_loss=0.3996, pruned_loss=0.1431, over 5668899.88 frames. ], batch size: 112, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:51:16,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 1.624e+03 1.893e+03 2.568e+03 5.295e+03, threshold=3.787e+03, percent-clipped=10.0
+2023-03-03 02:51:25,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:51:28,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:51:48,615 INFO [train.py:968] (0/2) Epoch 6, batch 8900, libri_loss[loss=0.3289, simple_loss=0.4007, pruned_loss=0.1285, over 29185.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3971, pruned_loss=0.1405, over 5696634.84 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3687, pruned_loss=0.112, over 5717989.72 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3999, pruned_loss=0.1436, over 5676658.06 frames. ], batch size: 101, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:51:55,843 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:52:31,224 INFO [train.py:968] (0/2) Epoch 6, batch 8950, giga_loss[loss=0.2814, simple_loss=0.3485, pruned_loss=0.1072, over 28632.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3976, pruned_loss=0.142, over 5696621.26 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3691, pruned_loss=0.1122, over 5721750.18 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.4002, pruned_loss=0.145, over 5676874.93 frames. ], batch size: 60, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 02:52:50,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.239e+02 1.484e+03 2.158e+03 3.181e+03 1.146e+04, threshold=4.317e+03, percent-clipped=16.0
+2023-03-03 02:53:09,757 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:53:12,277 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:53:18,997 INFO [train.py:968] (0/2) Epoch 6, batch 9000, giga_loss[loss=0.3094, simple_loss=0.3681, pruned_loss=0.1254, over 29037.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3957, pruned_loss=0.1413, over 5702226.01 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3686, pruned_loss=0.112, over 5726067.11 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3989, pruned_loss=0.1446, over 5682074.37 frames. ], batch size: 128, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 02:53:19,001 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 02:53:27,486 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2325, simple_loss=0.3368, pruned_loss=0.0641, over 944034.00 frames. 
+2023-03-03 02:53:27,486 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 02:53:50,952 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:54:15,666 INFO [train.py:968] (0/2) Epoch 6, batch 9050, giga_loss[loss=0.2577, simple_loss=0.3366, pruned_loss=0.08937, over 28961.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3931, pruned_loss=0.1402, over 5698119.23 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3688, pruned_loss=0.1121, over 5729344.07 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3959, pruned_loss=0.1433, over 5678634.88 frames. ], batch size: 174, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 02:54:35,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.859e+02 1.530e+03 1.851e+03 2.547e+03 6.947e+03, threshold=3.702e+03, percent-clipped=8.0
+2023-03-03 02:54:40,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:54:51,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7373, 1.2369, 3.3785, 2.8338], device='cuda:0'), covar=tensor([0.1688, 0.2136, 0.0425, 0.0780], device='cuda:0'), in_proj_covar=tensor([0.0587, 0.0544, 0.0779, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 02:55:03,548 INFO [train.py:968] (0/2) Epoch 6, batch 9100, giga_loss[loss=0.3186, simple_loss=0.3793, pruned_loss=0.129, over 28696.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3905, pruned_loss=0.1391, over 5687131.06 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3683, pruned_loss=0.1119, over 5725378.56 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3939, pruned_loss=0.1424, over 5673171.63 frames. ], batch size: 262, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 02:55:34,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3809, 1.2739, 1.0379, 1.5374], device='cuda:0'), covar=tensor([0.0733, 0.0320, 0.0336, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0121, 0.0125, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0061, 0.0045, 0.0040, 0.0067], device='cuda:0')
+2023-03-03 02:55:47,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9515, 1.9274, 1.3918, 1.5965], device='cuda:0'), covar=tensor([0.0623, 0.0572, 0.0926, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0457, 0.0508, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 02:55:55,826 INFO [train.py:968] (0/2) Epoch 6, batch 9150, giga_loss[loss=0.3434, simple_loss=0.3962, pruned_loss=0.1453, over 28923.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3904, pruned_loss=0.1393, over 5686944.05 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3684, pruned_loss=0.1119, over 5723999.22 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3932, pruned_loss=0.1423, over 5676413.37 frames. ], batch size: 199, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 02:56:16,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.234e+02 1.663e+03 2.236e+03 2.912e+03 8.136e+03, threshold=4.472e+03, percent-clipped=14.0
+2023-03-03 02:56:43,321 INFO [train.py:968] (0/2) Epoch 6, batch 9200, giga_loss[loss=0.3304, simple_loss=0.3826, pruned_loss=0.1391, over 28930.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3916, pruned_loss=0.1408, over 5674817.55 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3688, pruned_loss=0.1121, over 5719445.64 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3942, pruned_loss=0.1439, over 5669471.32 frames. ], batch size: 106, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:56:44,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9270, 1.1252, 1.0405, 0.6832], device='cuda:0'), covar=tensor([0.0898, 0.0915, 0.0614, 0.0960], device='cuda:0'), in_proj_covar=tensor([0.1477, 0.1307, 0.1274, 0.1372], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 02:57:00,460 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:57:03,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:57:20,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7524, 1.8888, 1.7251, 1.7537], device='cuda:0'), covar=tensor([0.1066, 0.1661, 0.1404, 0.1348], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0746, 0.0650, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 02:57:28,338 INFO [train.py:968] (0/2) Epoch 6, batch 9250, giga_loss[loss=0.3391, simple_loss=0.4005, pruned_loss=0.1389, over 28179.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3893, pruned_loss=0.1391, over 5671987.18 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3684, pruned_loss=0.1116, over 5715797.34 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3928, pruned_loss=0.1432, over 5669023.67 frames. ], batch size: 368, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:57:28,499 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:57:45,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.828e+02 1.513e+03 1.935e+03 2.657e+03 9.442e+03, threshold=3.870e+03, percent-clipped=6.0
+2023-03-03 02:58:13,805 INFO [train.py:968] (0/2) Epoch 6, batch 9300, libri_loss[loss=0.2406, simple_loss=0.3159, pruned_loss=0.08269, over 29360.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3881, pruned_loss=0.1384, over 5681577.32 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3679, pruned_loss=0.1113, over 5719675.27 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3918, pruned_loss=0.1428, over 5674564.72 frames. ], batch size: 67, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:59:01,734 INFO [train.py:968] (0/2) Epoch 6, batch 9350, giga_loss[loss=0.4521, simple_loss=0.4627, pruned_loss=0.2208, over 26455.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3902, pruned_loss=0.14, over 5668774.14 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3675, pruned_loss=0.1112, over 5714309.00 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.394, pruned_loss=0.1442, over 5667947.85 frames. ], batch size: 555, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:59:21,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.468e+02 1.416e+03 1.979e+03 3.056e+03 1.072e+04, threshold=3.958e+03, percent-clipped=17.0
+2023-03-03 02:59:22,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5280, 1.7072, 1.4064, 1.0717], device='cuda:0'), covar=tensor([0.1386, 0.1018, 0.0852, 0.1212], device='cuda:0'), in_proj_covar=tensor([0.1458, 0.1295, 0.1255, 0.1349], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 02:59:48,168 INFO [train.py:968] (0/2) Epoch 6, batch 9400, giga_loss[loss=0.3365, simple_loss=0.4057, pruned_loss=0.1337, over 29049.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3914, pruned_loss=0.1401, over 5674774.48 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3674, pruned_loss=0.1111, over 5717896.83 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3952, pruned_loss=0.1443, over 5669802.93 frames. ], batch size: 155, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 03:00:30,337 INFO [train.py:968] (0/2) Epoch 6, batch 9450, giga_loss[loss=0.3287, simple_loss=0.3882, pruned_loss=0.1346, over 28845.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.39, pruned_loss=0.1393, over 5671374.69 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3667, pruned_loss=0.1107, over 5717008.94 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3946, pruned_loss=0.1442, over 5666922.49 frames. ], batch size: 199, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 03:00:50,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.582e+02 1.658e+03 2.186e+03 3.284e+03 1.746e+04, threshold=4.373e+03, percent-clipped=19.0
+2023-03-03 03:00:55,094 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 03:01:18,374 INFO [train.py:968] (0/2) Epoch 6, batch 9500, giga_loss[loss=0.295, simple_loss=0.3655, pruned_loss=0.1123, over 28619.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3908, pruned_loss=0.1377, over 5675540.08 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3669, pruned_loss=0.1107, over 5720185.02 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3946, pruned_loss=0.1422, over 5668417.56 frames. ], batch size: 336, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 03:01:34,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.1474, 5.9334, 5.6673, 3.0971], device='cuda:0'), covar=tensor([0.0361, 0.0522, 0.0799, 0.1481], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0851, 0.0822, 0.0620], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 03:02:01,219 INFO [train.py:968] (0/2) Epoch 6, batch 9550, libri_loss[loss=0.288, simple_loss=0.3634, pruned_loss=0.1063, over 27694.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3914, pruned_loss=0.1359, over 5678820.73 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3662, pruned_loss=0.1104, over 5718827.44 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3964, pruned_loss=0.1411, over 5672065.52 frames. ], batch size: 116, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 03:02:18,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.341e+02 1.303e+03 1.599e+03 2.174e+03 5.291e+03, threshold=3.198e+03, percent-clipped=1.0
+2023-03-03 03:02:21,032 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=237285.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:02:44,099 INFO [train.py:968] (0/2) Epoch 6, batch 9600, giga_loss[loss=0.3334, simple_loss=0.4016, pruned_loss=0.1326, over 28845.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3946, pruned_loss=0.1367, over 5670098.59 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3668, pruned_loss=0.1107, over 5712190.41 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3987, pruned_loss=0.1411, over 5670625.45 frames. ], batch size: 186, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 03:02:46,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4584, 4.3003, 4.0656, 1.8861], device='cuda:0'), covar=tensor([0.0460, 0.0615, 0.0754, 0.2063], device='cuda:0'), in_proj_covar=tensor([0.0909, 0.0849, 0.0815, 0.0616], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 03:03:35,493 INFO [train.py:968] (0/2) Epoch 6, batch 9650, giga_loss[loss=0.3291, simple_loss=0.3868, pruned_loss=0.1357, over 28926.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.398, pruned_loss=0.1399, over 5677307.50 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3669, pruned_loss=0.1108, over 5714099.55 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.4014, pruned_loss=0.1435, over 5675618.14 frames. ], batch size: 213, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:03:58,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.953e+02 1.418e+03 1.816e+03 2.275e+03 5.650e+03, threshold=3.631e+03, percent-clipped=8.0
+2023-03-03 03:04:24,487 INFO [train.py:968] (0/2) Epoch 6, batch 9700, giga_loss[loss=0.4175, simple_loss=0.4547, pruned_loss=0.1901, over 28060.00 frames. ], tot_loss[loss=0.344, simple_loss=0.4008, pruned_loss=0.1435, over 5674005.77 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3673, pruned_loss=0.1112, over 5716025.08 frames. ], giga_tot_loss[loss=0.3482, simple_loss=0.4036, pruned_loss=0.1465, over 5670263.10 frames. ], batch size: 412, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:04:24,783 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3945, 2.8739, 1.4848, 1.2767], device='cuda:0'), covar=tensor([0.0771, 0.0332, 0.0769, 0.1195], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0485, 0.0314, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0021], device='cuda:0')
+2023-03-03 03:05:08,759 INFO [train.py:968] (0/2) Epoch 6, batch 9750, libri_loss[loss=0.2844, simple_loss=0.3622, pruned_loss=0.1033, over 29650.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.4007, pruned_loss=0.1439, over 5669550.42 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3671, pruned_loss=0.1111, over 5713786.84 frames. ], giga_tot_loss[loss=0.3503, simple_loss=0.4046, pruned_loss=0.148, over 5666298.37 frames. ], batch size: 88, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:05:28,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.906e+02 1.788e+03 2.389e+03 3.742e+03 9.143e+03, threshold=4.778e+03, percent-clipped=26.0
+2023-03-03 03:05:52,975 INFO [train.py:968] (0/2) Epoch 6, batch 9800, giga_loss[loss=0.3181, simple_loss=0.3866, pruned_loss=0.1248, over 28638.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3982, pruned_loss=0.1428, over 5657889.70 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3665, pruned_loss=0.1106, over 5719218.27 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.403, pruned_loss=0.1475, over 5648883.62 frames. ], batch size: 307, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:06:37,156 INFO [train.py:968] (0/2) Epoch 6, batch 9850, giga_loss[loss=0.319, simple_loss=0.3939, pruned_loss=0.1221, over 29061.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3975, pruned_loss=0.1414, over 5662373.33 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3664, pruned_loss=0.1107, over 5723554.25 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.4022, pruned_loss=0.1458, over 5650118.20 frames. ], batch size: 155, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:06:56,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.686e+02 1.605e+03 2.110e+03 2.801e+03 1.044e+04, threshold=4.220e+03, percent-clipped=9.0
+2023-03-03 03:07:25,927 INFO [train.py:968] (0/2) Epoch 6, batch 9900, giga_loss[loss=0.3041, simple_loss=0.3857, pruned_loss=0.1112, over 28957.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3985, pruned_loss=0.1405, over 5674866.16 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3661, pruned_loss=0.1106, over 5726225.63 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.4028, pruned_loss=0.1444, over 5662123.79 frames. ], batch size: 164, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:07:41,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4913, 1.5425, 1.3976, 1.8150], device='cuda:0'), covar=tensor([0.1665, 0.1535, 0.1386, 0.1448], device='cuda:0'), in_proj_covar=tensor([0.1149, 0.0889, 0.1018, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 03:08:07,095 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=237660.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:08:07,621 INFO [train.py:968] (0/2) Epoch 6, batch 9950, giga_loss[loss=0.2977, simple_loss=0.3756, pruned_loss=0.1099, over 28497.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3975, pruned_loss=0.1392, over 5664725.04 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3663, pruned_loss=0.1106, over 5712502.67 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.4021, pruned_loss=0.1435, over 5664439.70 frames. ], batch size: 60, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:08:29,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.953e+02 1.423e+03 1.766e+03 2.306e+03 6.448e+03, threshold=3.533e+03, percent-clipped=6.0
+2023-03-03 03:08:35,779 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-03 03:08:58,016 INFO [train.py:968] (0/2) Epoch 6, batch 10000, libri_loss[loss=0.2877, simple_loss=0.3609, pruned_loss=0.1073, over 29538.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3969, pruned_loss=0.1395, over 5659369.17 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.366, pruned_loss=0.1106, over 5716521.13 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.4013, pruned_loss=0.1436, over 5654636.46 frames. ], batch size: 80, lr: 5.49e-03, grad_scale: 8.0
+2023-03-03 03:09:09,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4135, 2.2115, 1.5285, 0.6916], device='cuda:0'), covar=tensor([0.3189, 0.1517, 0.2322, 0.3091], device='cuda:0'), in_proj_covar=tensor([0.1407, 0.1333, 0.1369, 0.1175], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 03:09:28,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3692, 1.4789, 1.4031, 1.4579], device='cuda:0'), covar=tensor([0.0968, 0.1239, 0.1315, 0.1086], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0743, 0.0642, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 03:09:45,751 INFO [train.py:968] (0/2) Epoch 6, batch 10050, giga_loss[loss=0.3287, simple_loss=0.3891, pruned_loss=0.1342, over 28927.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3961, pruned_loss=0.1398, over 5662320.26 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3659, pruned_loss=0.1106, over 5709699.08 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.4001, pruned_loss=0.1433, over 5663938.83 frames. ], batch size: 186, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:10:10,245 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.295e+02 1.480e+03 1.940e+03 3.201e+03 9.573e+03, threshold=3.880e+03, percent-clipped=18.0
+2023-03-03 03:10:30,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=237803.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:10:33,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=237806.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:10:36,954 INFO [train.py:968] (0/2) Epoch 6, batch 10100, giga_loss[loss=0.387, simple_loss=0.4308, pruned_loss=0.1716, over 28858.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3954, pruned_loss=0.1406, over 5656209.25 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3662, pruned_loss=0.1109, over 5710878.21 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3988, pruned_loss=0.1436, over 5655633.01 frames. ], batch size: 199, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:11:01,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237835.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:11:26,937 INFO [train.py:968] (0/2) Epoch 6, batch 10150, giga_loss[loss=0.3085, simple_loss=0.3714, pruned_loss=0.1228, over 28549.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3944, pruned_loss=0.1408, over 5659145.36 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3665, pruned_loss=0.111, over 5711286.97 frames. ], giga_tot_loss[loss=0.3423, simple_loss=0.3974, pruned_loss=0.1436, over 5657383.71 frames. ], batch size: 71, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:11:39,466 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-03-03 03:11:49,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.065e+02 1.807e+03 2.286e+03 2.907e+03 8.303e+03, threshold=4.571e+03, percent-clipped=11.0
+2023-03-03 03:11:49,688 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=237884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:11:53,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2418, 1.4638, 1.2106, 1.3097], device='cuda:0'), covar=tensor([0.2223, 0.2171, 0.2305, 0.1816], device='cuda:0'), in_proj_covar=tensor([0.1154, 0.0891, 0.1021, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 03:12:18,973 INFO [train.py:968] (0/2) Epoch 6, batch 10200, giga_loss[loss=0.3265, simple_loss=0.3734, pruned_loss=0.1398, over 28872.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.392, pruned_loss=0.1397, over 5663490.94 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3666, pruned_loss=0.111, over 5712501.56 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3951, pruned_loss=0.1427, over 5659633.64 frames. ], batch size: 99, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:12:47,941 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 03:13:00,943 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=237956.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:13:03,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3865, 2.8043, 1.5190, 1.4009], device='cuda:0'), covar=tensor([0.0754, 0.0344, 0.0724, 0.1163], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0488, 0.0314, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 03:13:04,034 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 03:13:05,530 INFO [train.py:968] (0/2) Epoch 6, batch 10250, giga_loss[loss=0.3124, simple_loss=0.3758, pruned_loss=0.1245, over 28727.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3925, pruned_loss=0.1408, over 5665131.26 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3666, pruned_loss=0.111, over 5715362.50 frames. ], giga_tot_loss[loss=0.3417, simple_loss=0.3955, pruned_loss=0.144, over 5658372.64 frames. ], batch size: 119, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:13:23,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.510e+02 1.458e+03 1.766e+03 2.417e+03 5.349e+03, threshold=3.531e+03, percent-clipped=1.0
+2023-03-03 03:13:32,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2225, 1.2837, 1.0802, 1.0270], device='cuda:0'), covar=tensor([0.0620, 0.0442, 0.0981, 0.0751], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0453, 0.0503, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 03:13:39,106 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-238000.pt
+2023-03-03 03:13:52,108 INFO [train.py:968] (0/2) Epoch 6, batch 10300, giga_loss[loss=0.3326, simple_loss=0.3968, pruned_loss=0.1342, over 28278.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3902, pruned_loss=0.1389, over 5666087.96 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3669, pruned_loss=0.111, over 5718171.45 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3928, pruned_loss=0.1419, over 5657624.34 frames. ], batch size: 368, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:14:35,818 INFO [train.py:968] (0/2) Epoch 6, batch 10350, libri_loss[loss=0.2834, simple_loss=0.3656, pruned_loss=0.1006, over 29520.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3878, pruned_loss=0.1354, over 5672574.13 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3669, pruned_loss=0.111, over 5724485.15 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3905, pruned_loss=0.1387, over 5658070.19 frames. ], batch size: 81, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:15:02,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.424e+02 1.402e+03 1.742e+03 2.295e+03 6.706e+03, threshold=3.483e+03, percent-clipped=4.0
+2023-03-03 03:15:28,603 INFO [train.py:968] (0/2) Epoch 6, batch 10400, giga_loss[loss=0.3585, simple_loss=0.404, pruned_loss=0.1565, over 28968.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3865, pruned_loss=0.1341, over 5667318.65 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.367, pruned_loss=0.1109, over 5725912.60 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3889, pruned_loss=0.137, over 5654160.94 frames. ], batch size: 213, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:15:30,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3490, 1.8662, 1.4353, 1.4649], device='cuda:0'), covar=tensor([0.0750, 0.0281, 0.0312, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0121, 0.0125, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0061, 0.0045, 0.0040, 0.0068], device='cuda:0')
+2023-03-03 03:16:02,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0226, 1.2450, 3.5955, 3.0683], device='cuda:0'), covar=tensor([0.1583, 0.2259, 0.0402, 0.1002], device='cuda:0'), in_proj_covar=tensor([0.0589, 0.0541, 0.0781, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 03:16:17,385 INFO [train.py:968] (0/2) Epoch 6, batch 10450, giga_loss[loss=0.2675, simple_loss=0.3388, pruned_loss=0.09812, over 28897.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.385, pruned_loss=0.1339, over 5673684.02 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3669, pruned_loss=0.1108, over 5729774.40 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3873, pruned_loss=0.1367, over 5658754.96 frames. ], batch size: 145, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:16:36,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1168, 1.4657, 3.1911, 2.9579], device='cuda:0'), covar=tensor([0.1234, 0.1714, 0.0366, 0.0720], device='cuda:0'), in_proj_covar=tensor([0.0585, 0.0537, 0.0777, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 03:16:44,500 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:16:46,205 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.731e+02 1.674e+03 2.173e+03 3.775e+03 1.181e+04, threshold=4.346e+03, percent-clipped=28.0
+2023-03-03 03:17:09,553 INFO [train.py:968] (0/2) Epoch 6, batch 10500, giga_loss[loss=0.2687, simple_loss=0.3388, pruned_loss=0.09928, over 28916.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3815, pruned_loss=0.1322, over 5678794.77 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3668, pruned_loss=0.1107, over 5732946.09 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3837, pruned_loss=0.135, over 5662945.02 frames. ], batch size: 112, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:17:44,826 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-03 03:17:57,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:17:58,695 INFO [train.py:968] (0/2) Epoch 6, batch 10550, giga_loss[loss=0.3415, simple_loss=0.4031, pruned_loss=0.14, over 28892.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3837, pruned_loss=0.1338, over 5677622.43 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3666, pruned_loss=0.1106, over 5734698.11 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3857, pruned_loss=0.1363, over 5663111.88 frames. ], batch size: 186, lr: 5.48e-03, grad_scale: 2.0
+2023-03-03 03:18:21,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.778e+02 1.549e+03 1.907e+03 3.167e+03 5.491e+03, threshold=3.814e+03, percent-clipped=9.0
+2023-03-03 03:18:40,092 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:18:43,540 INFO [train.py:968] (0/2) Epoch 6, batch 10600, libri_loss[loss=0.3062, simple_loss=0.3825, pruned_loss=0.1149, over 29534.00 frames. ], tot_loss[loss=0.326, simple_loss=0.385, pruned_loss=0.1335, over 5676419.21 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3673, pruned_loss=0.1108, over 5736743.89 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3864, pruned_loss=0.1358, over 5661641.62 frames. ], batch size: 84, lr: 5.48e-03, grad_scale: 2.0
+2023-03-03 03:19:06,051 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238331.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:19:35,237 INFO [train.py:968] (0/2) Epoch 6, batch 10650, giga_loss[loss=0.2902, simple_loss=0.3618, pruned_loss=0.1092, over 28436.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3854, pruned_loss=0.134, over 5644486.30 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3682, pruned_loss=0.1113, over 5729948.79 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3861, pruned_loss=0.136, over 5637596.83 frames. ], batch size: 65, lr: 5.48e-03, grad_scale: 2.0
+2023-03-03 03:19:57,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.923e+02 1.339e+03 1.671e+03 2.103e+03 6.374e+03, threshold=3.342e+03, percent-clipped=7.0
+2023-03-03 03:20:14,727 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:20:18,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:20:22,244 INFO [train.py:968] (0/2) Epoch 6, batch 10700, giga_loss[loss=0.3032, simple_loss=0.3667, pruned_loss=0.1199, over 28562.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3859, pruned_loss=0.1347, over 5633426.35 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3687, pruned_loss=0.1115, over 5725379.30 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3863, pruned_loss=0.1366, over 5629844.45 frames. ], batch size: 78, lr: 5.48e-03, grad_scale: 2.0
+2023-03-03 03:20:45,051 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:21:04,437 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:21:09,073 INFO [train.py:968] (0/2) Epoch 6, batch 10750, giga_loss[loss=0.4008, simple_loss=0.4161, pruned_loss=0.1928, over 23540.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3869, pruned_loss=0.1359, over 5637554.32 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3688, pruned_loss=0.1115, over 5725318.07 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3875, pruned_loss=0.1379, over 5633252.23 frames. ], batch size: 705, lr: 5.48e-03, grad_scale: 2.0
+2023-03-03 03:21:16,557 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:21:25,185 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238474.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:21:27,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238477.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:21:35,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.753e+02 1.577e+03 2.056e+03 2.792e+03 4.719e+03, threshold=4.111e+03, percent-clipped=13.0
+2023-03-03 03:21:55,373 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238506.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:22:00,630 INFO [train.py:968] (0/2) Epoch 6, batch 10800, giga_loss[loss=0.4472, simple_loss=0.4582, pruned_loss=0.2181, over 26615.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3888, pruned_loss=0.1369, over 5648024.57 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3686, pruned_loss=0.1115, over 5729621.39 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3899, pruned_loss=0.1391, over 5638694.50 frames. ], batch size: 555, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:22:08,683 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:22:45,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:22:47,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2756, 1.9313, 1.4192, 0.3261], device='cuda:0'), covar=tensor([0.2074, 0.1384, 0.2170, 0.2739], device='cuda:0'), in_proj_covar=tensor([0.1432, 0.1345, 0.1384, 0.1195], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 03:22:47,388 INFO [train.py:968] (0/2) Epoch 6, batch 10850, libri_loss[loss=0.3431, simple_loss=0.4103, pruned_loss=0.138, over 29558.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3895, pruned_loss=0.137, over 5656146.94 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.369, pruned_loss=0.1116, over 5734141.37 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3907, pruned_loss=0.1393, over 5642116.50 frames. ], batch size: 89, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:22:52,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2099, 1.7385, 1.2303, 0.3580], device='cuda:0'), covar=tensor([0.1686, 0.1065, 0.1703, 0.2512], device='cuda:0'), in_proj_covar=tensor([0.1432, 0.1346, 0.1386, 0.1196], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 03:23:07,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.405e+02 1.469e+03 1.935e+03 3.138e+03 9.913e+03, threshold=3.870e+03, percent-clipped=10.0
+2023-03-03 03:23:29,498 INFO [train.py:968] (0/2) Epoch 6, batch 10900, giga_loss[loss=0.3916, simple_loss=0.4301, pruned_loss=0.1766, over 28974.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3918, pruned_loss=0.1388, over 5663409.25 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.37, pruned_loss=0.1122, over 5740437.91 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3926, pruned_loss=0.1411, over 5643579.91 frames. ], batch size: 136, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:23:49,165 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 03:24:13,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5539, 1.5764, 1.5369, 1.4485], device='cuda:0'), covar=tensor([0.1042, 0.1477, 0.1557, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0742, 0.0652, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 03:24:19,789 INFO [train.py:968] (0/2) Epoch 6, batch 10950, giga_loss[loss=0.3791, simple_loss=0.4245, pruned_loss=0.1669, over 28974.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3941, pruned_loss=0.1411, over 5656961.59 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3704, pruned_loss=0.1124, over 5733338.93 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3948, pruned_loss=0.1433, over 5645476.34 frames. ], batch size: 164, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:24:38,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:24:45,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.449e+02 1.718e+03 2.216e+03 3.203e+03 1.182e+04, threshold=4.432e+03, percent-clipped=17.0
+2023-03-03 03:25:01,039 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:25:04,035 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:25:10,398 INFO [train.py:968] (0/2) Epoch 6, batch 11000, libri_loss[loss=0.2652, simple_loss=0.3319, pruned_loss=0.0992, over 29379.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3949, pruned_loss=0.1406, over 5663321.66 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3701, pruned_loss=0.1122, over 5737346.34 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3965, pruned_loss=0.1433, over 5648497.13 frames. ], batch size: 71, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:25:32,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:26:01,748 INFO [train.py:968] (0/2) Epoch 6, batch 11050, giga_loss[loss=0.4293, simple_loss=0.4332, pruned_loss=0.2127, over 23654.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3936, pruned_loss=0.14, over 5649786.43 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.37, pruned_loss=0.1121, over 5737734.38 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3957, pruned_loss=0.1432, over 5635162.24 frames. ], batch size: 705, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:26:25,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.843e+02 1.580e+03 2.157e+03 3.160e+03 1.286e+04, threshold=4.315e+03, percent-clipped=7.0
+2023-03-03 03:26:44,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2972, 1.6809, 1.3697, 1.4698], device='cuda:0'), covar=tensor([0.0757, 0.0286, 0.0327, 0.0860], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0122, 0.0126, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0062, 0.0045, 0.0041, 0.0068], device='cuda:0')
+2023-03-03 03:26:49,565 INFO [train.py:968] (0/2) Epoch 6, batch 11100, libri_loss[loss=0.3022, simple_loss=0.3757, pruned_loss=0.1144, over 29529.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3919, pruned_loss=0.1396, over 5661256.02 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.37, pruned_loss=0.1121, over 5738437.22 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3941, pruned_loss=0.1428, over 5647236.68 frames. ], batch size: 89, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:26:51,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238813.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:27:03,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:27:07,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:27:12,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:27:28,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:27:44,776 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:27:50,146 INFO [train.py:968] (0/2) Epoch 6, batch 11150, giga_loss[loss=0.3288, simple_loss=0.3844, pruned_loss=0.1366, over 28993.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3913, pruned_loss=0.1398, over 5654910.41 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3702, pruned_loss=0.1122, over 5736960.10 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3933, pruned_loss=0.1427, over 5643600.38 frames. ], batch size: 213, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:28:12,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.607e+02 1.483e+03 1.860e+03 2.463e+03 6.017e+03, threshold=3.720e+03, percent-clipped=5.0
+2023-03-03 03:28:21,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0631, 3.8718, 3.7006, 1.9336], device='cuda:0'), covar=tensor([0.0457, 0.0567, 0.0703, 0.2046], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0864, 0.0830, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-03 03:28:22,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:28:40,095 INFO [train.py:968] (0/2) Epoch 6, batch 11200, giga_loss[loss=0.3102, simple_loss=0.3682, pruned_loss=0.1261, over 28671.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3899, pruned_loss=0.1387, over 5667976.89 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3705, pruned_loss=0.1124, over 5737018.24 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3915, pruned_loss=0.1412, over 5658002.22 frames. ], batch size: 262, lr: 5.48e-03, grad_scale: 8.0
+2023-03-03 03:29:25,218 INFO [train.py:968] (0/2) Epoch 6, batch 11250, giga_loss[loss=0.3508, simple_loss=0.3986, pruned_loss=0.1515, over 28561.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3897, pruned_loss=0.1393, over 5671144.58 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3709, pruned_loss=0.1126, over 5741094.44 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.391, pruned_loss=0.1418, over 5657992.10 frames. ], batch size: 336, lr: 5.48e-03, grad_scale: 8.0
+2023-03-03 03:29:40,177 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:29:42,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:29:51,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.757e+02 1.377e+03 1.956e+03 2.411e+03 4.326e+03, threshold=3.912e+03, percent-clipped=6.0
+2023-03-03 03:29:53,520 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:29:55,301 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:30:14,705 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:30:17,480 INFO [train.py:968] (0/2) Epoch 6, batch 11300, giga_loss[loss=0.3405, simple_loss=0.388, pruned_loss=0.1465, over 28873.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3906, pruned_loss=0.1402, over 5668549.67 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3714, pruned_loss=0.1129, over 5740974.46 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3914, pruned_loss=0.1421, over 5657640.95 frames. ], batch size: 119, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:30:24,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9097, 2.3822, 2.2503, 1.9766], device='cuda:0'), covar=tensor([0.1501, 0.1667, 0.1064, 0.1220], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0742, 0.0796, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 03:30:25,206 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:30:39,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6641, 1.4522, 1.2327, 1.3245], device='cuda:0'), covar=tensor([0.0471, 0.0372, 0.0728, 0.0711], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0455, 0.0504, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 03:30:49,037 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:30:51,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:31:09,232 INFO [train.py:968] (0/2) Epoch 6, batch 11350, giga_loss[loss=0.329, simple_loss=0.3946, pruned_loss=0.1317, over 28963.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3916, pruned_loss=0.1411, over 5671131.44 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3714, pruned_loss=0.113, over 5744076.26 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3926, pruned_loss=0.1431, over 5657969.99 frames. ], batch size: 145, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:31:15,809 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:31:20,326 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=239074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:31:22,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3159, 1.7597, 1.6670, 1.4743], device='cuda:0'), covar=tensor([0.1478, 0.2028, 0.1188, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0742, 0.0799, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 03:31:31,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.700e+03 2.273e+03 3.684e+03 7.690e+03, threshold=4.545e+03, percent-clipped=21.0
+2023-03-03 03:31:56,234 INFO [train.py:968] (0/2) Epoch 6, batch 11400, giga_loss[loss=0.3385, simple_loss=0.4031, pruned_loss=0.1369, over 28614.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3939, pruned_loss=0.1432, over 5664611.27 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3711, pruned_loss=0.1128, over 5737376.41 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3954, pruned_loss=0.1455, over 5659416.07 frames. ], batch size: 307, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:32:42,522 INFO [train.py:968] (0/2) Epoch 6, batch 11450, libri_loss[loss=0.277, simple_loss=0.3386, pruned_loss=0.1078, over 29665.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3939, pruned_loss=0.1426, over 5662865.25 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3706, pruned_loss=0.1126, over 5732644.91 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3963, pruned_loss=0.1456, over 5660806.65 frames. ], batch size: 69, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:33:08,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.775e+03 2.092e+03 2.893e+03 1.057e+04, threshold=4.184e+03, percent-clipped=6.0
+2023-03-03 03:33:10,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239188.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:33:30,300 INFO [train.py:968] (0/2) Epoch 6, batch 11500, giga_loss[loss=0.3665, simple_loss=0.4069, pruned_loss=0.163, over 27563.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.396, pruned_loss=0.1451, over 5662920.18 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.371, pruned_loss=0.1128, over 5737674.49 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.3983, pruned_loss=0.1483, over 5654358.43 frames. ], batch size: 472, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:33:39,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7768, 1.9016, 1.5074, 1.2655], device='cuda:0'), covar=tensor([0.1336, 0.1116, 0.0919, 0.1129], device='cuda:0'), in_proj_covar=tensor([0.1474, 0.1317, 0.1270, 0.1348], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 03:34:19,179 INFO [train.py:968] (0/2) Epoch 6, batch 11550, libri_loss[loss=0.2569, simple_loss=0.3364, pruned_loss=0.08873, over 29581.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3967, pruned_loss=0.1463, over 5651195.31 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3714, pruned_loss=0.1131, over 5730418.57 frames. ], giga_tot_loss[loss=0.3482, simple_loss=0.3985, pruned_loss=0.149, over 5650289.10 frames. ], batch size: 74, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:34:28,717 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-03 03:34:42,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.254e+02 1.589e+03 2.124e+03 2.824e+03 6.322e+03, threshold=4.248e+03, percent-clipped=7.0
+2023-03-03 03:35:06,144 INFO [train.py:968] (0/2) Epoch 6, batch 11600, giga_loss[loss=0.3294, simple_loss=0.386, pruned_loss=0.1364, over 28680.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3957, pruned_loss=0.1445, over 5665472.24 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3714, pruned_loss=0.113, over 5733514.33 frames. ], giga_tot_loss[loss=0.3463, simple_loss=0.3977, pruned_loss=0.1474, over 5660442.47 frames. ], batch size: 92, lr: 5.47e-03, grad_scale: 8.0
+2023-03-03 03:35:24,912 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=239329.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:35:26,184 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239331.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:35:28,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239334.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:35:52,840 INFO [train.py:968] (0/2) Epoch 6, batch 11650, giga_loss[loss=0.3403, simple_loss=0.3974, pruned_loss=0.1416, over 28727.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3952, pruned_loss=0.1436, over 5665354.56 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.371, pruned_loss=0.1128, over 5736759.18 frames. ], giga_tot_loss[loss=0.3463, simple_loss=0.398, pruned_loss=0.1473, over 5656286.24 frames. ], batch size: 284, lr: 5.47e-03, grad_scale: 8.0
+2023-03-03 03:35:55,872 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239363.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:36:20,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.149e+03 1.662e+03 2.244e+03 2.928e+03 4.857e+03, threshold=4.487e+03, percent-clipped=5.0
+2023-03-03 03:36:38,869 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-03 03:36:44,381 INFO [train.py:968] (0/2) Epoch 6, batch 11700, giga_loss[loss=0.3481, simple_loss=0.4066, pruned_loss=0.1448, over 28736.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3958, pruned_loss=0.1436, over 5679822.01 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3709, pruned_loss=0.1126, over 5741338.04 frames. ], giga_tot_loss[loss=0.3467, simple_loss=0.3986, pruned_loss=0.1474, over 5666758.64 frames. ], batch size: 262, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:37:21,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:37:35,100 INFO [train.py:968] (0/2) Epoch 6, batch 11750, libri_loss[loss=0.2691, simple_loss=0.3458, pruned_loss=0.09625, over 29550.00 frames. ], tot_loss[loss=0.346, simple_loss=0.3987, pruned_loss=0.1466, over 5678038.72 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3707, pruned_loss=0.1125, over 5745338.81 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4019, pruned_loss=0.1507, over 5662266.04 frames. ], batch size: 78, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:37:58,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.909e+02 1.745e+03 2.324e+03 3.096e+03 7.019e+03, threshold=4.648e+03, percent-clipped=9.0
+2023-03-03 03:38:22,805 INFO [train.py:968] (0/2) Epoch 6, batch 11800, giga_loss[loss=0.3352, simple_loss=0.3957, pruned_loss=0.1373, over 28816.00 frames. ], tot_loss[loss=0.345, simple_loss=0.398, pruned_loss=0.146, over 5688636.16 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3707, pruned_loss=0.1124, over 5747182.32 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4009, pruned_loss=0.1496, over 5673981.86 frames. ], batch size: 186, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:39:08,624 INFO [train.py:968] (0/2) Epoch 6, batch 11850, giga_loss[loss=0.3766, simple_loss=0.4252, pruned_loss=0.164, over 28713.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3979, pruned_loss=0.1451, over 5690869.56 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3701, pruned_loss=0.1121, over 5751404.57 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.4013, pruned_loss=0.149, over 5674322.13 frames. ], batch size: 262, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:39:32,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.393e+02 1.703e+03 2.015e+03 2.743e+03 9.217e+03, threshold=4.030e+03, percent-clipped=4.0
+2023-03-03 03:39:37,699 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:39:39,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:39:53,198 INFO [train.py:968] (0/2) Epoch 6, batch 11900, giga_loss[loss=0.3511, simple_loss=0.4125, pruned_loss=0.1449, over 28929.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.3976, pruned_loss=0.1439, over 5683074.31 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3701, pruned_loss=0.1121, over 5747536.83 frames. ], giga_tot_loss[loss=0.3486, simple_loss=0.4012, pruned_loss=0.148, over 5671566.70 frames. ], batch size: 227, lr: 5.47e-03, grad_scale: 2.0
+2023-03-03 03:40:04,521 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:40:39,382 INFO [train.py:968] (0/2) Epoch 6, batch 11950, giga_loss[loss=0.3179, simple_loss=0.3801, pruned_loss=0.1278, over 28925.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3949, pruned_loss=0.1416, over 5678479.70 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3703, pruned_loss=0.1121, over 5749897.80 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3983, pruned_loss=0.1456, over 5665969.47 frames. ], batch size: 227, lr: 5.47e-03, grad_scale: 2.0
+2023-03-03 03:41:07,063 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.210e+02 1.466e+03 2.127e+03 3.136e+03 1.079e+04, threshold=4.255e+03, percent-clipped=17.0
+2023-03-03 03:41:20,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2683, 1.3161, 1.3885, 1.3091], device='cuda:0'), covar=tensor([0.0874, 0.1073, 0.1269, 0.0927], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0743, 0.0647, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 03:41:21,474 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:41:26,909 INFO [train.py:968] (0/2) Epoch 6, batch 12000, giga_loss[loss=0.3768, simple_loss=0.418, pruned_loss=0.1678, over 28307.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3944, pruned_loss=0.1413, over 5690581.67 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3706, pruned_loss=0.1124, over 5752853.40 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3972, pruned_loss=0.1447, over 5676833.36 frames. ], batch size: 368, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:41:26,914 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 03:41:35,219 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2359, simple_loss=0.3395, pruned_loss=0.06612, over 944034.00 frames. 
+2023-03-03 03:41:35,219 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 03:42:27,875 INFO [train.py:968] (0/2) Epoch 6, batch 12050, giga_loss[loss=0.3986, simple_loss=0.429, pruned_loss=0.1841, over 26641.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3962, pruned_loss=0.1436, over 5668000.51 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3707, pruned_loss=0.1125, over 5753543.00 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3986, pruned_loss=0.1466, over 5655522.40 frames. ], batch size: 555, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:42:53,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.824e+02 1.667e+03 2.504e+03 3.473e+03 9.873e+03, threshold=5.008e+03, percent-clipped=12.0
+2023-03-03 03:43:12,980 INFO [train.py:968] (0/2) Epoch 6, batch 12100, libri_loss[loss=0.2866, simple_loss=0.3604, pruned_loss=0.1064, over 29544.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3958, pruned_loss=0.1422, over 5678504.85 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.371, pruned_loss=0.1126, over 5752602.46 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.3985, pruned_loss=0.1456, over 5666359.55 frames. ], batch size: 78, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:43:13,919 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 03:43:25,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2777, 1.1552, 1.0784, 1.4139], device='cuda:0'), covar=tensor([0.0741, 0.0349, 0.0337, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0121, 0.0125, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0061, 0.0045, 0.0040, 0.0067], device='cuda:0')
+2023-03-03 03:43:45,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239847.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:43:50,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239850.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:43:59,971 INFO [train.py:968] (0/2) Epoch 6, batch 12150, giga_loss[loss=0.3294, simple_loss=0.3871, pruned_loss=0.1359, over 28983.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.395, pruned_loss=0.1429, over 5674319.42 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3707, pruned_loss=0.1126, over 5755821.46 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.398, pruned_loss=0.1465, over 5659743.43 frames. ], batch size: 213, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:44:17,207 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239879.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:44:17,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9813, 1.1760, 3.1846, 2.9337], device='cuda:0'), covar=tensor([0.1459, 0.2120, 0.0460, 0.1154], device='cuda:0'), in_proj_covar=tensor([0.0592, 0.0541, 0.0783, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 03:44:24,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.562e+02 1.319e+03 1.718e+03 1.953e+03 4.968e+03, threshold=3.436e+03, percent-clipped=0.0
+2023-03-03 03:44:46,648 INFO [train.py:968] (0/2) Epoch 6, batch 12200, giga_loss[loss=0.3447, simple_loss=0.3937, pruned_loss=0.1478, over 28552.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3942, pruned_loss=0.1425, over 5675847.80 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3706, pruned_loss=0.1125, over 5761045.08 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3975, pruned_loss=0.1465, over 5656633.08 frames. ], batch size: 85, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:45:32,555 INFO [train.py:968] (0/2) Epoch 6, batch 12250, giga_loss[loss=0.3929, simple_loss=0.4384, pruned_loss=0.1737, over 28801.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.395, pruned_loss=0.1432, over 5678830.34 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3707, pruned_loss=0.1126, over 5763992.53 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.3984, pruned_loss=0.1474, over 5657968.92 frames. ], batch size: 284, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:45:45,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5351, 1.8469, 1.2649, 1.0167], device='cuda:0'), covar=tensor([0.1329, 0.1058, 0.0945, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.1500, 0.1339, 0.1287, 0.1375], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 03:45:58,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.244e+02 1.555e+03 2.014e+03 3.301e+03 1.028e+04, threshold=4.028e+03, percent-clipped=21.0
+2023-03-03 03:46:00,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1609, 1.8069, 1.3697, 0.3680], device='cuda:0'), covar=tensor([0.2051, 0.1257, 0.2193, 0.2718], device='cuda:0'), in_proj_covar=tensor([0.1431, 0.1351, 0.1394, 0.1184], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 03:46:10,244 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-240000.pt
+2023-03-03 03:46:18,418 INFO [train.py:968] (0/2) Epoch 6, batch 12300, giga_loss[loss=0.3495, simple_loss=0.4066, pruned_loss=0.1462, over 28806.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.396, pruned_loss=0.1438, over 5676128.70 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3709, pruned_loss=0.1127, over 5766668.63 frames. ], giga_tot_loss[loss=0.3474, simple_loss=0.3992, pruned_loss=0.1478, over 5654907.57 frames. ], batch size: 243, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:46:21,811 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=240015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:46:32,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8648, 2.4572, 2.7596, 2.1362], device='cuda:0'), covar=tensor([0.0995, 0.1636, 0.1100, 0.1406], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0740, 0.0643, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 03:47:05,909 INFO [train.py:968] (0/2) Epoch 6, batch 12350, giga_loss[loss=0.2865, simple_loss=0.3543, pruned_loss=0.1094, over 28889.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3949, pruned_loss=0.1434, over 5655269.80 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.371, pruned_loss=0.1127, over 5757328.54 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.398, pruned_loss=0.1474, over 5643835.89 frames. ], batch size: 186, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:47:35,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.040e+02 1.433e+03 2.005e+03 2.772e+03 8.216e+03, threshold=4.010e+03, percent-clipped=12.0
+2023-03-03 03:47:54,196 INFO [train.py:968] (0/2) Epoch 6, batch 12400, giga_loss[loss=0.3496, simple_loss=0.4057, pruned_loss=0.1467, over 28580.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3952, pruned_loss=0.1434, over 5651201.10 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3708, pruned_loss=0.1126, over 5759145.45 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3983, pruned_loss=0.1473, over 5638351.17 frames. ], batch size: 336, lr: 5.46e-03, grad_scale: 8.0
+2023-03-03 03:48:42,160 INFO [train.py:968] (0/2) Epoch 6, batch 12450, giga_loss[loss=0.3197, simple_loss=0.3772, pruned_loss=0.1312, over 28418.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3963, pruned_loss=0.1442, over 5651724.64 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3711, pruned_loss=0.1128, over 5758981.59 frames. ], giga_tot_loss[loss=0.3474, simple_loss=0.3991, pruned_loss=0.1478, over 5638885.86 frames. ], batch size: 71, lr: 5.46e-03, grad_scale: 8.0
+2023-03-03 03:49:03,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1042, 2.8362, 1.2302, 1.2452], device='cuda:0'), covar=tensor([0.1119, 0.0482, 0.0986, 0.1555], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0494, 0.0316, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 03:49:10,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.804e+02 1.483e+03 1.874e+03 2.933e+03 5.148e+03, threshold=3.748e+03, percent-clipped=10.0
+2023-03-03 03:49:34,705 INFO [train.py:968] (0/2) Epoch 6, batch 12500, giga_loss[loss=0.2832, simple_loss=0.3544, pruned_loss=0.106, over 29044.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3954, pruned_loss=0.1436, over 5653911.63 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.371, pruned_loss=0.1127, over 5759867.68 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.3978, pruned_loss=0.1467, over 5642637.02 frames. ], batch size: 128, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:50:20,277 INFO [train.py:968] (0/2) Epoch 6, batch 12550, giga_loss[loss=0.3179, simple_loss=0.3787, pruned_loss=0.1286, over 28841.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3937, pruned_loss=0.1424, over 5664425.80 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3716, pruned_loss=0.1133, over 5765466.72 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3958, pruned_loss=0.1454, over 5647320.80 frames. ], batch size: 243, lr: 5.46e-03, grad_scale: 2.0
+2023-03-03 03:50:52,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.828e+02 1.456e+03 1.928e+03 3.193e+03 2.076e+04, threshold=3.857e+03, percent-clipped=16.0
+2023-03-03 03:51:07,586 INFO [train.py:968] (0/2) Epoch 6, batch 12600, giga_loss[loss=0.3152, simple_loss=0.3763, pruned_loss=0.127, over 28718.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3908, pruned_loss=0.1401, over 5667703.76 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3718, pruned_loss=0.1132, over 5757637.90 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3926, pruned_loss=0.1429, over 5659726.01 frames. ], batch size: 119, lr: 5.46e-03, grad_scale: 2.0
+2023-03-03 03:51:19,981 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-03 03:51:56,104 INFO [train.py:968] (0/2) Epoch 6, batch 12650, giga_loss[loss=0.3792, simple_loss=0.4056, pruned_loss=0.1764, over 26612.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3873, pruned_loss=0.1385, over 5663551.65 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.372, pruned_loss=0.1133, over 5762262.27 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3891, pruned_loss=0.1414, over 5650920.75 frames. ], batch size: 555, lr: 5.46e-03, grad_scale: 2.0
+2023-03-03 03:52:26,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.490e+02 1.452e+03 1.955e+03 2.386e+03 6.545e+03, threshold=3.910e+03, percent-clipped=2.0
+2023-03-03 03:52:26,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=240390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:52:29,168 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-03 03:52:42,958 INFO [train.py:968] (0/2) Epoch 6, batch 12700, giga_loss[loss=0.3097, simple_loss=0.3653, pruned_loss=0.127, over 28902.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3859, pruned_loss=0.1384, over 5661759.02 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3727, pruned_loss=0.1138, over 5764220.00 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3871, pruned_loss=0.1409, over 5647429.15 frames. ], batch size: 106, lr: 5.46e-03, grad_scale: 2.0
+2023-03-03 03:52:50,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7204, 3.5313, 3.3911, 1.6562], device='cuda:0'), covar=tensor([0.0643, 0.0688, 0.0798, 0.2086], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0882, 0.0835, 0.0624], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-03 03:53:30,446 INFO [train.py:968] (0/2) Epoch 6, batch 12750, giga_loss[loss=0.3482, simple_loss=0.3976, pruned_loss=0.1495, over 28549.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3853, pruned_loss=0.1385, over 5659700.88 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3727, pruned_loss=0.1139, over 5767294.82 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3866, pruned_loss=0.1411, over 5642884.54 frames. ], batch size: 336, lr: 5.46e-03, grad_scale: 2.0
+2023-03-03 03:54:00,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.157e+03 1.637e+03 2.113e+03 2.865e+03 8.656e+03, threshold=4.226e+03, percent-clipped=15.0
+2023-03-03 03:54:11,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4512, 2.2041, 1.6625, 0.6539], device='cuda:0'), covar=tensor([0.2646, 0.1616, 0.2020, 0.2973], device='cuda:0'), in_proj_covar=tensor([0.1454, 0.1366, 0.1410, 0.1206], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 03:54:21,866 INFO [train.py:968] (0/2) Epoch 6, batch 12800, libri_loss[loss=0.2977, simple_loss=0.3721, pruned_loss=0.1117, over 29550.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3846, pruned_loss=0.1376, over 5653369.33 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3731, pruned_loss=0.1142, over 5761028.00 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3856, pruned_loss=0.14, over 5641785.63 frames. ], batch size: 83, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:54:43,371 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=240533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:54:45,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=240536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:54:48,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9460, 4.7610, 4.5392, 2.0413], device='cuda:0'), covar=tensor([0.0414, 0.0492, 0.0712, 0.2172], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0875, 0.0822, 0.0623], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 03:55:07,756 INFO [train.py:968] (0/2) Epoch 6, batch 12850, giga_loss[loss=0.3589, simple_loss=0.4, pruned_loss=0.1589, over 26709.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3834, pruned_loss=0.1351, over 5648475.58 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3728, pruned_loss=0.1141, over 5756125.65 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3849, pruned_loss=0.1379, over 5640092.47 frames. ], batch size: 555, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:55:11,896 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=240565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:55:18,635 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-03 03:55:36,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.129e+02 1.399e+03 1.808e+03 2.505e+03 4.390e+03, threshold=3.615e+03, percent-clipped=2.0
+2023-03-03 03:55:49,058 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 03:55:56,273 INFO [train.py:968] (0/2) Epoch 6, batch 12900, giga_loss[loss=0.3078, simple_loss=0.3775, pruned_loss=0.1191, over 28901.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3791, pruned_loss=0.1303, over 5653582.00 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3709, pruned_loss=0.1131, over 5761095.69 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3824, pruned_loss=0.1344, over 5637545.70 frames. ], batch size: 106, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:56:48,857 INFO [train.py:968] (0/2) Epoch 6, batch 12950, giga_loss[loss=0.2885, simple_loss=0.3671, pruned_loss=0.1049, over 28738.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3752, pruned_loss=0.1262, over 5652803.76 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3704, pruned_loss=0.113, over 5764442.08 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3784, pruned_loss=0.1299, over 5634778.80 frames. ], batch size: 262, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:57:19,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.330e+02 1.358e+03 1.687e+03 2.510e+03 6.605e+03, threshold=3.373e+03, percent-clipped=9.0
+2023-03-03 03:57:27,879 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 03:57:43,477 INFO [train.py:968] (0/2) Epoch 6, batch 13000, giga_loss[loss=0.2478, simple_loss=0.3249, pruned_loss=0.08532, over 27590.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.1231, over 5639724.24 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3703, pruned_loss=0.1131, over 5756355.82 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3743, pruned_loss=0.126, over 5631714.27 frames. ], batch size: 472, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:58:06,562 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=240733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:58:35,032 INFO [train.py:968] (0/2) Epoch 6, batch 13050, giga_loss[loss=0.2878, simple_loss=0.3651, pruned_loss=0.1052, over 28892.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3691, pruned_loss=0.119, over 5648219.88 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3697, pruned_loss=0.1128, over 5757225.24 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3717, pruned_loss=0.1216, over 5639586.00 frames. ], batch size: 199, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:58:36,066 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5997, 1.5950, 1.1594, 1.3002], device='cuda:0'), covar=tensor([0.0680, 0.0541, 0.0956, 0.0956], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0451, 0.0506, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 03:59:04,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.063e+02 1.221e+03 1.704e+03 2.448e+03 8.906e+03, threshold=3.408e+03, percent-clipped=12.0
+2023-03-03 03:59:25,746 INFO [train.py:968] (0/2) Epoch 6, batch 13100, giga_loss[loss=0.3115, simple_loss=0.3811, pruned_loss=0.1209, over 28711.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3684, pruned_loss=0.1169, over 5658322.04 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3694, pruned_loss=0.1128, over 5761890.70 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3707, pruned_loss=0.1192, over 5644437.13 frames. ], batch size: 262, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 04:00:11,356 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-03 04:00:18,604 INFO [train.py:968] (0/2) Epoch 6, batch 13150, giga_loss[loss=0.2279, simple_loss=0.3147, pruned_loss=0.07061, over 28970.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.368, pruned_loss=0.1167, over 5653032.86 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3691, pruned_loss=0.1128, over 5763202.10 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3701, pruned_loss=0.1186, over 5640080.52 frames. ], batch size: 155, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:00:48,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.750e+02 1.375e+03 1.794e+03 2.407e+03 5.932e+03, threshold=3.588e+03, percent-clipped=14.0
+2023-03-03 04:01:09,952 INFO [train.py:968] (0/2) Epoch 6, batch 13200, giga_loss[loss=0.2396, simple_loss=0.3264, pruned_loss=0.07637, over 29027.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3654, pruned_loss=0.1149, over 5648954.24 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3689, pruned_loss=0.1128, over 5760031.95 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3672, pruned_loss=0.1164, over 5640181.85 frames. ], batch size: 155, lr: 5.45e-03, grad_scale: 8.0
+2023-03-03 04:01:43,679 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3504, 1.8207, 1.3216, 0.5918], device='cuda:0'), covar=tensor([0.1613, 0.1019, 0.1765, 0.2296], device='cuda:0'), in_proj_covar=tensor([0.1407, 0.1327, 0.1387, 0.1169], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 04:01:53,433 INFO [train.py:968] (0/2) Epoch 6, batch 13250, giga_loss[loss=0.2988, simple_loss=0.3677, pruned_loss=0.115, over 28619.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.363, pruned_loss=0.1137, over 5652688.05 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3677, pruned_loss=0.1122, over 5763963.52 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3654, pruned_loss=0.1156, over 5635274.23 frames. ], batch size: 336, lr: 5.45e-03, grad_scale: 8.0
+2023-03-03 04:02:05,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6719, 1.7182, 1.7478, 1.6274], device='cuda:0'), covar=tensor([0.1191, 0.1760, 0.1407, 0.1447], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0721, 0.0634, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 04:02:25,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.602e+02 1.320e+03 1.748e+03 2.335e+03 5.124e+03, threshold=3.495e+03, percent-clipped=5.0
+2023-03-03 04:02:45,681 INFO [train.py:968] (0/2) Epoch 6, batch 13300, giga_loss[loss=0.2711, simple_loss=0.3458, pruned_loss=0.09817, over 28883.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3632, pruned_loss=0.1136, over 5645432.40 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3677, pruned_loss=0.1122, over 5764410.33 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3649, pruned_loss=0.1151, over 5630300.38 frames. ], batch size: 106, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:03:07,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4216, 1.7359, 1.2980, 1.5648], device='cuda:0'), covar=tensor([0.0752, 0.0263, 0.0332, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0125, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0062, 0.0044, 0.0040, 0.0068], device='cuda:0')
+2023-03-03 04:03:36,363 INFO [train.py:968] (0/2) Epoch 6, batch 13350, giga_loss[loss=0.2801, simple_loss=0.3493, pruned_loss=0.1055, over 27738.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3617, pruned_loss=0.1122, over 5653613.89 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3675, pruned_loss=0.1122, over 5767250.26 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3632, pruned_loss=0.1134, over 5637068.28 frames. ], batch size: 472, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:03:45,693 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-03 04:03:47,693 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2721, 1.3144, 1.1078, 1.4923], device='cuda:0'), covar=tensor([0.0783, 0.0307, 0.0362, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0125, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0062, 0.0044, 0.0040, 0.0068], device='cuda:0')
+2023-03-03 04:04:05,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.656e+02 1.230e+03 1.669e+03 2.271e+03 4.804e+03, threshold=3.338e+03, percent-clipped=8.0
+2023-03-03 04:04:23,892 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:04:25,729 INFO [train.py:968] (0/2) Epoch 6, batch 13400, giga_loss[loss=0.2562, simple_loss=0.3355, pruned_loss=0.08842, over 28877.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3585, pruned_loss=0.1096, over 5658117.68 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3666, pruned_loss=0.1117, over 5772463.39 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3603, pruned_loss=0.111, over 5635702.78 frames. ], batch size: 199, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:04:35,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3952, 2.0787, 1.5967, 0.4990], device='cuda:0'), covar=tensor([0.2036, 0.1252, 0.1963, 0.2523], device='cuda:0'), in_proj_covar=tensor([0.1410, 0.1330, 0.1386, 0.1169], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 04:05:11,218 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.04 vs. limit=2.0
+2023-03-03 04:05:16,133 INFO [train.py:968] (0/2) Epoch 6, batch 13450, giga_loss[loss=0.2436, simple_loss=0.323, pruned_loss=0.08208, over 28302.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3539, pruned_loss=0.1062, over 5657473.21 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3656, pruned_loss=0.1111, over 5775050.24 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3561, pruned_loss=0.1078, over 5635419.02 frames. ], batch size: 368, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:05:46,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.731e+02 1.288e+03 1.740e+03 2.319e+03 9.823e+03, threshold=3.480e+03, percent-clipped=10.0
+2023-03-03 04:06:07,332 INFO [train.py:968] (0/2) Epoch 6, batch 13500, giga_loss[loss=0.2642, simple_loss=0.3415, pruned_loss=0.09348, over 29041.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3507, pruned_loss=0.1045, over 5666042.69 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3649, pruned_loss=0.1107, over 5778038.64 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3526, pruned_loss=0.1059, over 5642155.01 frames. ], batch size: 136, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:06:25,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5959, 1.7069, 1.4405, 2.0910], device='cuda:0'), covar=tensor([0.2217, 0.2064, 0.2108, 0.1932], device='cuda:0'), in_proj_covar=tensor([0.1146, 0.0874, 0.1016, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 04:06:48,566 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:06:50,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:06:56,508 INFO [train.py:968] (0/2) Epoch 6, batch 13550, giga_loss[loss=0.2718, simple_loss=0.3359, pruned_loss=0.1038, over 28793.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3503, pruned_loss=0.1048, over 5667018.42 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3647, pruned_loss=0.1106, over 5778660.33 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3516, pruned_loss=0.1059, over 5643415.95 frames. ], batch size: 119, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:07:21,949 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:07:25,581 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-03 04:07:29,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.364e+02 1.391e+03 1.872e+03 2.384e+03 6.086e+03, threshold=3.743e+03, percent-clipped=10.0
+2023-03-03 04:07:48,266 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3675, 2.0058, 1.5332, 1.5823], device='cuda:0'), covar=tensor([0.0764, 0.0262, 0.0317, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0119, 0.0126, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0062, 0.0044, 0.0041, 0.0069], device='cuda:0')
+2023-03-03 04:07:52,922 INFO [train.py:968] (0/2) Epoch 6, batch 13600, giga_loss[loss=0.2732, simple_loss=0.3463, pruned_loss=0.1, over 28880.00 frames. ], tot_loss[loss=0.281, simple_loss=0.351, pruned_loss=0.1055, over 5654682.59 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3648, pruned_loss=0.1109, over 5776879.96 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3517, pruned_loss=0.1061, over 5635943.95 frames. ], batch size: 112, lr: 5.45e-03, grad_scale: 8.0
+2023-03-03 04:08:08,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=241325.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:08:48,334 INFO [train.py:968] (0/2) Epoch 6, batch 13650, giga_loss[loss=0.303, simple_loss=0.3796, pruned_loss=0.1132, over 28647.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3536, pruned_loss=0.106, over 5657281.92 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3647, pruned_loss=0.1108, over 5777714.38 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3539, pruned_loss=0.1063, over 5637644.13 frames. ], batch size: 262, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:09:10,807 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=241379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:09:28,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.405e+02 1.363e+03 1.889e+03 2.445e+03 5.216e+03, threshold=3.777e+03, percent-clipped=7.0
+2023-03-03 04:09:50,763 INFO [train.py:968] (0/2) Epoch 6, batch 13700, giga_loss[loss=0.2784, simple_loss=0.3506, pruned_loss=0.1031, over 28847.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3553, pruned_loss=0.106, over 5666167.76 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3647, pruned_loss=0.1108, over 5778880.87 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3555, pruned_loss=0.1062, over 5648642.16 frames. ], batch size: 174, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:10:48,885 INFO [train.py:968] (0/2) Epoch 6, batch 13750, giga_loss[loss=0.2524, simple_loss=0.3298, pruned_loss=0.08756, over 28879.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3548, pruned_loss=0.1057, over 5669026.99 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3638, pruned_loss=0.1106, over 5770619.55 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3554, pruned_loss=0.106, over 5658956.65 frames. ], batch size: 227, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:11:13,306 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 04:11:28,177 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.596e+02 1.276e+03 1.625e+03 2.315e+03 4.691e+03, threshold=3.251e+03, percent-clipped=5.0
+2023-03-03 04:11:39,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4208, 1.9310, 1.8045, 1.5856], device='cuda:0'), covar=tensor([0.1658, 0.1892, 0.1233, 0.1365], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0721, 0.0790, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 04:11:49,635 INFO [train.py:968] (0/2) Epoch 6, batch 13800, giga_loss[loss=0.2217, simple_loss=0.3101, pruned_loss=0.0666, over 28897.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3511, pruned_loss=0.1034, over 5659854.55 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3634, pruned_loss=0.1104, over 5763292.63 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3519, pruned_loss=0.1037, over 5657586.93 frames. ], batch size: 164, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:11:49,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7865, 1.1862, 5.4419, 3.7316], device='cuda:0'), covar=tensor([0.1493, 0.2469, 0.0282, 0.0647], device='cuda:0'), in_proj_covar=tensor([0.0574, 0.0536, 0.0762, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 04:12:04,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2650, 1.3687, 4.0137, 3.1777], device='cuda:0'), covar=tensor([0.1498, 0.2412, 0.0335, 0.1101], device='cuda:0'), in_proj_covar=tensor([0.0574, 0.0536, 0.0760, 0.0621], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 04:12:50,216 INFO [train.py:968] (0/2) Epoch 6, batch 13850, giga_loss[loss=0.2581, simple_loss=0.3368, pruned_loss=0.08966, over 28845.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3504, pruned_loss=0.1016, over 5661810.21 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3633, pruned_loss=0.1103, over 5765093.35 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3509, pruned_loss=0.1017, over 5656916.00 frames. ], batch size: 186, lr: 5.45e-03, grad_scale: 2.0
+2023-03-03 04:13:32,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.184e+02 1.202e+03 1.589e+03 2.051e+03 7.691e+03, threshold=3.178e+03, percent-clipped=7.0
+2023-03-03 04:13:34,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 04:13:53,811 INFO [train.py:968] (0/2) Epoch 6, batch 13900, giga_loss[loss=0.3057, simple_loss=0.3614, pruned_loss=0.125, over 28423.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.348, pruned_loss=0.1006, over 5662213.97 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3632, pruned_loss=0.1104, over 5767086.03 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3483, pruned_loss=0.1006, over 5655231.51 frames. ], batch size: 336, lr: 5.45e-03, grad_scale: 2.0
+2023-03-03 04:14:55,821 INFO [train.py:968] (0/2) Epoch 6, batch 13950, giga_loss[loss=0.3022, simple_loss=0.3718, pruned_loss=0.1163, over 29061.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3463, pruned_loss=0.1007, over 5668701.15 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3629, pruned_loss=0.1102, over 5768700.37 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3466, pruned_loss=0.1007, over 5660421.50 frames. ], batch size: 285, lr: 5.45e-03, grad_scale: 2.0
+2023-03-03 04:15:32,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.135e+02 1.267e+03 1.729e+03 2.488e+03 6.326e+03, threshold=3.458e+03, percent-clipped=10.0
+2023-03-03 04:15:42,583 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241700.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:15:54,467 INFO [train.py:968] (0/2) Epoch 6, batch 14000, giga_loss[loss=0.2489, simple_loss=0.3286, pruned_loss=0.0846, over 28560.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3453, pruned_loss=0.1006, over 5673950.12 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3628, pruned_loss=0.1102, over 5771663.60 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3453, pruned_loss=0.1004, over 5663061.95 frames. ], batch size: 336, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:16:47,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:16:54,071 INFO [train.py:968] (0/2) Epoch 6, batch 14050, giga_loss[loss=0.2659, simple_loss=0.3478, pruned_loss=0.09197, over 28163.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3467, pruned_loss=0.101, over 5658763.01 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3625, pruned_loss=0.11, over 5769920.29 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3467, pruned_loss=0.1009, over 5649523.94 frames. ], batch size: 412, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:17:36,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.459e+02 1.335e+03 2.017e+03 2.562e+03 6.884e+03, threshold=4.033e+03, percent-clipped=11.0
+2023-03-03 04:17:45,790 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=241799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:17:58,991 INFO [train.py:968] (0/2) Epoch 6, batch 14100, giga_loss[loss=0.2369, simple_loss=0.3205, pruned_loss=0.07666, over 28340.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3484, pruned_loss=0.1012, over 5659683.47 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3625, pruned_loss=0.1102, over 5770252.21 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3482, pruned_loss=0.1008, over 5650427.89 frames. ], batch size: 368, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:18:41,874 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241843.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 04:18:46,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241846.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 04:19:02,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.2190, 6.0211, 5.7051, 2.6849], device='cuda:0'), covar=tensor([0.0366, 0.0496, 0.0768, 0.1685], device='cuda:0'), in_proj_covar=tensor([0.0875, 0.0814, 0.0777, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 04:19:05,381 INFO [train.py:968] (0/2) Epoch 6, batch 14150, giga_loss[loss=0.287, simple_loss=0.3516, pruned_loss=0.1112, over 28943.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3444, pruned_loss=0.09875, over 5673762.49 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3618, pruned_loss=0.1099, over 5773865.89 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3444, pruned_loss=0.09839, over 5660143.21 frames. ], batch size: 284, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:19:24,396 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241875.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:19:46,282 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.293e+02 1.369e+03 1.755e+03 2.400e+03 8.320e+03, threshold=3.510e+03, percent-clipped=7.0
+2023-03-03 04:19:52,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:19:57,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:20:11,100 INFO [train.py:968] (0/2) Epoch 6, batch 14200, giga_loss[loss=0.2647, simple_loss=0.3467, pruned_loss=0.09139, over 28925.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3464, pruned_loss=0.09992, over 5681157.86 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3615, pruned_loss=0.1097, over 5775992.72 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3463, pruned_loss=0.09965, over 5666721.24 frames. ], batch size: 284, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:20:36,762 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:21:18,682 INFO [train.py:968] (0/2) Epoch 6, batch 14250, giga_loss[loss=0.262, simple_loss=0.3242, pruned_loss=0.09991, over 24485.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3491, pruned_loss=0.0998, over 5680971.62 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3615, pruned_loss=0.1096, over 5778251.37 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3488, pruned_loss=0.09948, over 5665715.78 frames. ], batch size: 705, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:21:49,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4239, 1.5821, 1.2112, 1.2696], device='cuda:0'), covar=tensor([0.1104, 0.0991, 0.0932, 0.0947], device='cuda:0'), in_proj_covar=tensor([0.1456, 0.1265, 0.1232, 0.1313], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 04:21:58,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.899e+02 1.318e+03 1.709e+03 2.302e+03 7.095e+03, threshold=3.418e+03, percent-clipped=7.0
+2023-03-03 04:22:05,510 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-242000.pt
+2023-03-03 04:22:19,401 INFO [train.py:968] (0/2) Epoch 6, batch 14300, giga_loss[loss=0.2544, simple_loss=0.3391, pruned_loss=0.08489, over 28386.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3512, pruned_loss=0.09869, over 5680719.31 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.361, pruned_loss=0.1093, over 5779558.90 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3511, pruned_loss=0.0984, over 5665173.96 frames. ], batch size: 368, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:22:24,700 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:23:08,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0185, 1.3303, 1.1289, 0.1193], device='cuda:0'), covar=tensor([0.1690, 0.1470, 0.2345, 0.2994], device='cuda:0'), in_proj_covar=tensor([0.1422, 0.1335, 0.1401, 0.1190], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 04:23:19,717 INFO [train.py:968] (0/2) Epoch 6, batch 14350, giga_loss[loss=0.2748, simple_loss=0.3582, pruned_loss=0.09569, over 29203.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3526, pruned_loss=0.09872, over 5680486.10 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3605, pruned_loss=0.1093, over 5782742.46 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3528, pruned_loss=0.09831, over 5662935.32 frames. ], batch size: 107, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:23:30,033 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 04:23:56,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.105e+02 1.153e+03 1.579e+03 2.476e+03 1.091e+04, threshold=3.157e+03, percent-clipped=11.0
+2023-03-03 04:24:18,847 INFO [train.py:968] (0/2) Epoch 6, batch 14400, giga_loss[loss=0.2915, simple_loss=0.3656, pruned_loss=0.1087, over 28675.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3513, pruned_loss=0.09771, over 5686957.27 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3598, pruned_loss=0.109, over 5785908.37 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3519, pruned_loss=0.09737, over 5667668.75 frames. ], batch size: 262, lr: 5.44e-03, grad_scale: 8.0
+2023-03-03 04:24:21,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1142, 1.0776, 4.9620, 3.4240], device='cuda:0'), covar=tensor([0.1785, 0.2548, 0.0281, 0.0675], device='cuda:0'), in_proj_covar=tensor([0.0582, 0.0542, 0.0762, 0.0627], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 04:25:23,334 INFO [train.py:968] (0/2) Epoch 6, batch 14450, giga_loss[loss=0.283, simple_loss=0.3543, pruned_loss=0.1058, over 28740.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3519, pruned_loss=0.09893, over 5674724.93 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3599, pruned_loss=0.1091, over 5777148.79 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3522, pruned_loss=0.09845, over 5665804.29 frames. ], batch size: 262, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:25:37,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:26:02,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.267e+02 1.171e+03 1.465e+03 2.030e+03 4.255e+03, threshold=2.929e+03, percent-clipped=4.0
+2023-03-03 04:26:04,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 04:26:05,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:26:08,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1873, 4.0611, 3.8092, 1.8943], device='cuda:0'), covar=tensor([0.0460, 0.0535, 0.0732, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.0874, 0.0813, 0.0777, 0.0599], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 04:26:26,265 INFO [train.py:968] (0/2) Epoch 6, batch 14500, giga_loss[loss=0.2949, simple_loss=0.3659, pruned_loss=0.112, over 28101.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3516, pruned_loss=0.09981, over 5688038.36 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3597, pruned_loss=0.109, over 5780856.59 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3519, pruned_loss=0.09931, over 5674945.56 frames. ], batch size: 412, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:26:44,814 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242224.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 04:27:08,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7251, 1.8731, 1.5870, 1.7295], device='cuda:0'), covar=tensor([0.0679, 0.0252, 0.0296, 0.0762], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0119, 0.0125, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0062, 0.0045, 0.0041, 0.0069], device='cuda:0')
+2023-03-03 04:27:40,184 INFO [train.py:968] (0/2) Epoch 6, batch 14550, giga_loss[loss=0.2433, simple_loss=0.3273, pruned_loss=0.07967, over 28987.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3526, pruned_loss=0.1011, over 5692229.66 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3592, pruned_loss=0.1086, over 5783413.40 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3531, pruned_loss=0.1009, over 5677861.71 frames. ], batch size: 155, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:28:34,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.294e+02 1.349e+03 1.692e+03 2.183e+03 4.042e+03, threshold=3.384e+03, percent-clipped=8.0
+2023-03-03 04:28:59,253 INFO [train.py:968] (0/2) Epoch 6, batch 14600, giga_loss[loss=0.2688, simple_loss=0.3446, pruned_loss=0.09655, over 28437.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3481, pruned_loss=0.09891, over 5689840.08 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3586, pruned_loss=0.1083, over 5786204.14 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3488, pruned_loss=0.09878, over 5673372.37 frames. ], batch size: 368, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:29:08,123 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:29:10,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:29:16,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3316, 1.8467, 1.7871, 1.6003], device='cuda:0'), covar=tensor([0.1477, 0.1760, 0.1110, 0.1249], device='cuda:0'), in_proj_covar=tensor([0.0771, 0.0714, 0.0785, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 04:29:47,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:30:01,567 INFO [train.py:968] (0/2) Epoch 6, batch 14650, giga_loss[loss=0.2924, simple_loss=0.3661, pruned_loss=0.1094, over 27768.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3466, pruned_loss=0.09795, over 5694137.46 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3582, pruned_loss=0.1081, over 5788983.21 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3471, pruned_loss=0.09765, over 5675019.71 frames. ], batch size: 474, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:30:37,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:30:44,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.180e+02 1.292e+03 1.566e+03 2.465e+03 4.934e+03, threshold=3.133e+03, percent-clipped=8.0
+2023-03-03 04:31:08,254 INFO [train.py:968] (0/2) Epoch 6, batch 14700, giga_loss[loss=0.3043, simple_loss=0.3574, pruned_loss=0.1256, over 26763.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3444, pruned_loss=0.0973, over 5691382.76 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3576, pruned_loss=0.1077, over 5792545.74 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3449, pruned_loss=0.09712, over 5670528.81 frames. ], batch size: 555, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:32:12,916 INFO [train.py:968] (0/2) Epoch 6, batch 14750, giga_loss[loss=0.2643, simple_loss=0.3441, pruned_loss=0.09222, over 28131.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3496, pruned_loss=0.1006, over 5684126.31 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3573, pruned_loss=0.1075, over 5792698.40 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3503, pruned_loss=0.1005, over 5666093.57 frames. ], batch size: 412, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:32:30,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4476, 1.5646, 1.5345, 1.4387], device='cuda:0'), covar=tensor([0.1058, 0.1511, 0.1499, 0.1417], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0709, 0.0619, 0.0624], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 04:32:55,155 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.875e+02 1.581e+03 1.973e+03 2.886e+03 5.885e+03, threshold=3.946e+03, percent-clipped=21.0
+2023-03-03 04:33:16,371 INFO [train.py:968] (0/2) Epoch 6, batch 14800, giga_loss[loss=0.2696, simple_loss=0.3387, pruned_loss=0.1002, over 28989.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3491, pruned_loss=0.1007, over 5680619.82 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3571, pruned_loss=0.1074, over 5784626.15 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3497, pruned_loss=0.1007, over 5672185.21 frames. ], batch size: 155, lr: 5.44e-03, grad_scale: 8.0
+2023-03-03 04:33:24,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 04:33:44,705 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:33:48,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:34:02,449 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:34:17,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4725, 1.6931, 1.3562, 2.1911], device='cuda:0'), covar=tensor([0.2132, 0.2072, 0.2214, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.1131, 0.0865, 0.1001, 0.0943], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 04:34:23,940 INFO [train.py:968] (0/2) Epoch 6, batch 14850, giga_loss[loss=0.2596, simple_loss=0.3388, pruned_loss=0.09017, over 28918.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3489, pruned_loss=0.1014, over 5677679.45 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3571, pruned_loss=0.1075, over 5776370.27 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3493, pruned_loss=0.1013, over 5676813.72 frames. ], batch size: 213, lr: 5.44e-03, grad_scale: 8.0
+2023-03-03 04:34:27,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:34:33,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:34:58,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.477e+02 1.341e+03 1.785e+03 2.450e+03 6.242e+03, threshold=3.570e+03, percent-clipped=5.0
+2023-03-03 04:35:03,413 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-03 04:35:04,278 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242599.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 04:35:18,348 INFO [train.py:968] (0/2) Epoch 6, batch 14900, libri_loss[loss=0.2909, simple_loss=0.3595, pruned_loss=0.1111, over 29514.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3496, pruned_loss=0.1025, over 5674704.74 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3567, pruned_loss=0.1074, over 5772095.28 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3499, pruned_loss=0.1022, over 5674816.95 frames. ], batch size: 84, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:35:27,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:36:20,171 INFO [train.py:968] (0/2) Epoch 6, batch 14950, giga_loss[loss=0.2748, simple_loss=0.3589, pruned_loss=0.0954, over 28426.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3519, pruned_loss=0.1031, over 5671794.24 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3568, pruned_loss=0.1075, over 5766466.76 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3519, pruned_loss=0.1027, over 5673843.79 frames. ], batch size: 369, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:36:59,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6299, 5.3313, 2.3257, 2.8176], device='cuda:0'), covar=tensor([0.0586, 0.0123, 0.0720, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0472, 0.0314, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0025, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 04:37:11,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.104e+02 1.403e+03 1.676e+03 2.179e+03 7.299e+03, threshold=3.352e+03, percent-clipped=5.0
+2023-03-03 04:37:23,982 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:37:32,477 INFO [train.py:968] (0/2) Epoch 6, batch 15000, giga_loss[loss=0.2606, simple_loss=0.3407, pruned_loss=0.09022, over 28509.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3523, pruned_loss=0.1028, over 5668721.82 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3567, pruned_loss=0.1076, over 5766499.40 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3524, pruned_loss=0.1022, over 5667380.36 frames. ], batch size: 336, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:37:32,481 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 04:37:36,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2188, 1.8963, 1.6175, 1.3923], device='cuda:0'), covar=tensor([0.1781, 0.2152, 0.1437, 0.1513], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0713, 0.0791, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 04:37:41,145 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2165, simple_loss=0.3142, pruned_loss=0.05942, over 944034.00 frames. 
+2023-03-03 04:37:41,145 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 04:37:43,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:37:47,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:38:27,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242742.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 04:38:35,407 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242745.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:38:37,397 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:38:57,802 INFO [train.py:968] (0/2) Epoch 6, batch 15050, giga_loss[loss=0.2467, simple_loss=0.3255, pruned_loss=0.08392, over 29093.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3504, pruned_loss=0.1016, over 5667778.48 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3563, pruned_loss=0.1074, over 5769043.24 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3507, pruned_loss=0.1012, over 5662487.41 frames. ], batch size: 136, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:39:00,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3588, 1.5015, 1.2489, 1.6640], device='cuda:0'), covar=tensor([0.2201, 0.1949, 0.1967, 0.1881], device='cuda:0'), in_proj_covar=tensor([0.1136, 0.0866, 0.1008, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 04:39:19,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3766, 1.9915, 1.4054, 0.5815], device='cuda:0'), covar=tensor([0.2307, 0.1218, 0.2194, 0.2721], device='cuda:0'), in_proj_covar=tensor([0.1407, 0.1324, 0.1385, 0.1175], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 04:39:22,472 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242774.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:39:48,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.888e+02 1.385e+03 1.795e+03 2.581e+03 8.270e+03, threshold=3.590e+03, percent-clipped=13.0
+2023-03-03 04:40:08,881 INFO [train.py:968] (0/2) Epoch 6, batch 15100, giga_loss[loss=0.238, simple_loss=0.325, pruned_loss=0.07551, over 28833.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3467, pruned_loss=0.101, over 5656716.37 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3563, pruned_loss=0.1075, over 5761088.89 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3468, pruned_loss=0.1006, over 5658144.90 frames. ], batch size: 174, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:41:03,681 INFO [train.py:968] (0/2) Epoch 6, batch 15150, giga_loss[loss=0.229, simple_loss=0.3063, pruned_loss=0.07587, over 29056.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.341, pruned_loss=0.09831, over 5670460.47 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3555, pruned_loss=0.1072, over 5769150.93 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3412, pruned_loss=0.09779, over 5658598.91 frames. ], batch size: 120, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:41:14,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7642, 2.6098, 1.5972, 1.4446], device='cuda:0'), covar=tensor([0.1610, 0.0720, 0.0939, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.1457, 0.1251, 0.1201, 0.1309], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 04:41:25,323 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:41:47,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0502, 1.1374, 3.4954, 2.9965], device='cuda:0'), covar=tensor([0.1546, 0.2454, 0.0403, 0.1226], device='cuda:0'), in_proj_covar=tensor([0.0570, 0.0538, 0.0747, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 04:41:47,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.032e+02 1.445e+03 1.785e+03 2.289e+03 1.035e+04, threshold=3.571e+03, percent-clipped=6.0
+2023-03-03 04:42:04,936 INFO [train.py:968] (0/2) Epoch 6, batch 15200, giga_loss[loss=0.271, simple_loss=0.3411, pruned_loss=0.1004, over 28583.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3409, pruned_loss=0.0982, over 5669461.03 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3553, pruned_loss=0.107, over 5771909.36 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3409, pruned_loss=0.0978, over 5655762.98 frames. ], batch size: 307, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:42:17,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:42:48,715 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:43:03,288 INFO [train.py:968] (0/2) Epoch 6, batch 15250, giga_loss[loss=0.2277, simple_loss=0.3162, pruned_loss=0.06965, over 28900.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3432, pruned_loss=0.09995, over 5670585.60 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3553, pruned_loss=0.1069, over 5773217.73 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.343, pruned_loss=0.09955, over 5657342.35 frames. ], batch size: 136, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:43:45,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:43:46,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:43:47,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.836e+02 1.442e+03 2.014e+03 2.762e+03 7.267e+03, threshold=4.027e+03, percent-clipped=14.0
+2023-03-03 04:44:04,165 INFO [train.py:968] (0/2) Epoch 6, batch 15300, giga_loss[loss=0.2939, simple_loss=0.3547, pruned_loss=0.1165, over 27660.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3408, pruned_loss=0.09812, over 5665112.99 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3551, pruned_loss=0.1069, over 5773543.22 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3405, pruned_loss=0.09766, over 5651214.81 frames. ], batch size: 472, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:45:04,756 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4129, 2.0046, 1.3443, 1.2639], device='cuda:0'), covar=tensor([0.1525, 0.0972, 0.0965, 0.1148], device='cuda:0'), in_proj_covar=tensor([0.1462, 0.1254, 0.1214, 0.1316], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 04:45:08,035 INFO [train.py:968] (0/2) Epoch 6, batch 15350, giga_loss[loss=0.2734, simple_loss=0.3444, pruned_loss=0.1012, over 28664.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3403, pruned_loss=0.09703, over 5670132.28 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3551, pruned_loss=0.107, over 5774711.20 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3399, pruned_loss=0.09648, over 5657231.36 frames. ], batch size: 307, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:45:11,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:45:18,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5867, 1.6413, 1.2933, 1.3511], device='cuda:0'), covar=tensor([0.0595, 0.0413, 0.0809, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0438, 0.0502, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 04:45:18,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:45:34,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:45:57,027 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 04:45:59,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.583e+02 1.380e+03 2.070e+03 3.209e+03 8.912e+03, threshold=4.141e+03, percent-clipped=13.0
+2023-03-03 04:46:00,266 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:46:20,413 INFO [train.py:968] (0/2) Epoch 6, batch 15400, giga_loss[loss=0.2628, simple_loss=0.335, pruned_loss=0.09527, over 28498.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3389, pruned_loss=0.09678, over 5658522.11 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3551, pruned_loss=0.1069, over 5768585.21 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3383, pruned_loss=0.09624, over 5651653.90 frames. ], batch size: 369, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:46:55,596 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-03 04:46:58,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:46:58,809 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-03 04:47:00,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:47:28,668 INFO [train.py:968] (0/2) Epoch 6, batch 15450, giga_loss[loss=0.3063, simple_loss=0.36, pruned_loss=0.1263, over 26897.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3394, pruned_loss=0.09674, over 5653036.74 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3543, pruned_loss=0.1064, over 5772028.74 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3393, pruned_loss=0.09649, over 5641812.89 frames. ], batch size: 555, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:47:33,294 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:47:38,364 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:48:16,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.591e+02 1.410e+03 1.902e+03 2.564e+03 6.909e+03, threshold=3.803e+03, percent-clipped=8.0
+2023-03-03 04:48:31,820 INFO [train.py:968] (0/2) Epoch 6, batch 15500, giga_loss[loss=0.268, simple_loss=0.3406, pruned_loss=0.09772, over 29125.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3387, pruned_loss=0.09628, over 5659316.71 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3539, pruned_loss=0.1063, over 5773942.83 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3388, pruned_loss=0.09609, over 5647118.28 frames. ], batch size: 200, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:48:32,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4410, 1.9619, 1.7648, 1.5702], device='cuda:0'), covar=tensor([0.1643, 0.2013, 0.1314, 0.1398], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0711, 0.0789, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 04:48:41,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:48:46,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:48:47,485 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1592, 2.4221, 1.2841, 1.2520], device='cuda:0'), covar=tensor([0.0818, 0.0455, 0.0810, 0.1247], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0477, 0.0314, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0025, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 04:49:28,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:49:29,637 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:49:37,021 INFO [train.py:968] (0/2) Epoch 6, batch 15550, giga_loss[loss=0.2666, simple_loss=0.3244, pruned_loss=0.1044, over 26997.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3391, pruned_loss=0.09741, over 5660817.71 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3534, pruned_loss=0.106, over 5776220.20 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3393, pruned_loss=0.09728, over 5647047.24 frames. ], batch size: 555, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:50:20,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.814e+02 1.420e+03 1.743e+03 2.372e+03 1.104e+04, threshold=3.486e+03, percent-clipped=7.0
+2023-03-03 04:50:25,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0878, 1.1512, 4.2633, 3.3959], device='cuda:0'), covar=tensor([0.1643, 0.2468, 0.0314, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0569, 0.0533, 0.0746, 0.0609], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 04:50:31,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4274, 1.8953, 1.3021, 0.6495], device='cuda:0'), covar=tensor([0.2425, 0.1368, 0.2324, 0.2873], device='cuda:0'), in_proj_covar=tensor([0.1399, 0.1326, 0.1378, 0.1163], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 04:50:35,926 INFO [train.py:968] (0/2) Epoch 6, batch 15600, giga_loss[loss=0.2501, simple_loss=0.3322, pruned_loss=0.08405, over 28427.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3395, pruned_loss=0.09616, over 5670583.69 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3534, pruned_loss=0.1059, over 5777193.16 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3393, pruned_loss=0.09588, over 5656057.98 frames. ], batch size: 368, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:50:39,177 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:50:50,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:51:02,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2629, 1.6735, 1.1971, 0.4756], device='cuda:0'), covar=tensor([0.1628, 0.1150, 0.2008, 0.2407], device='cuda:0'), in_proj_covar=tensor([0.1407, 0.1336, 0.1386, 0.1170], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 04:51:26,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1879, 1.2223, 3.8091, 3.1374], device='cuda:0'), covar=tensor([0.1526, 0.2300, 0.0365, 0.0716], device='cuda:0'), in_proj_covar=tensor([0.0568, 0.0533, 0.0746, 0.0607], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 04:51:33,066 INFO [train.py:968] (0/2) Epoch 6, batch 15650, giga_loss[loss=0.2457, simple_loss=0.3344, pruned_loss=0.07849, over 28864.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3419, pruned_loss=0.09609, over 5675355.60 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3534, pruned_loss=0.1059, over 5778941.66 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3414, pruned_loss=0.09566, over 5659471.57 frames. ], batch size: 112, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:51:43,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:51:47,960 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-03 04:52:12,045 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:52:14,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.253e+02 1.332e+03 1.644e+03 2.578e+03 8.163e+03, threshold=3.289e+03, percent-clipped=12.0
+2023-03-03 04:52:14,965 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:52:27,387 INFO [train.py:968] (0/2) Epoch 6, batch 15700, giga_loss[loss=0.2567, simple_loss=0.3424, pruned_loss=0.08551, over 28847.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3435, pruned_loss=0.09684, over 5674197.71 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3524, pruned_loss=0.1052, over 5782994.83 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3435, pruned_loss=0.0967, over 5652536.28 frames. ], batch size: 174, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:52:44,487 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:53:23,213 INFO [train.py:968] (0/2) Epoch 6, batch 15750, giga_loss[loss=0.2455, simple_loss=0.3303, pruned_loss=0.08037, over 28711.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3454, pruned_loss=0.09753, over 5681956.47 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3529, pruned_loss=0.1055, over 5785878.29 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3447, pruned_loss=0.09689, over 5658201.63 frames. ], batch size: 119, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:53:28,720 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:53:32,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:54:03,685 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.787e+02 1.323e+03 1.642e+03 2.279e+03 6.969e+03, threshold=3.284e+03, percent-clipped=9.0
+2023-03-03 04:54:04,476 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:54:19,345 INFO [train.py:968] (0/2) Epoch 6, batch 15800, libri_loss[loss=0.3322, simple_loss=0.3797, pruned_loss=0.1423, over 20242.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3453, pruned_loss=0.09762, over 5685301.00 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3526, pruned_loss=0.1054, over 5778869.90 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3447, pruned_loss=0.09685, over 5669613.18 frames. ], batch size: 187, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:54:23,264 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:54:27,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:54:55,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:55:00,473 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:55:22,164 INFO [train.py:968] (0/2) Epoch 6, batch 15850, giga_loss[loss=0.2701, simple_loss=0.3453, pruned_loss=0.09751, over 28887.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3424, pruned_loss=0.09549, over 5688762.76 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3529, pruned_loss=0.1056, over 5779864.43 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3416, pruned_loss=0.09467, over 5674711.82 frames. ], batch size: 227, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 04:55:30,495 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 04:55:57,492 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-03 04:56:05,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.464e+02 1.274e+03 1.628e+03 2.031e+03 4.227e+03, threshold=3.256e+03, percent-clipped=6.0
+2023-03-03 04:56:23,881 INFO [train.py:968] (0/2) Epoch 6, batch 15900, giga_loss[loss=0.2906, simple_loss=0.347, pruned_loss=0.1171, over 26948.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3426, pruned_loss=0.09587, over 5690466.72 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3526, pruned_loss=0.1053, over 5781614.84 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3419, pruned_loss=0.09514, over 5675229.30 frames. ], batch size: 555, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 04:56:24,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5937, 1.3584, 5.3598, 3.5213], device='cuda:0'), covar=tensor([0.1553, 0.2284, 0.0295, 0.0638], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0535, 0.0751, 0.0615], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 04:57:23,531 INFO [train.py:968] (0/2) Epoch 6, batch 15950, giga_loss[loss=0.2737, simple_loss=0.3469, pruned_loss=0.1003, over 28938.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3398, pruned_loss=0.09483, over 5684516.84 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3526, pruned_loss=0.1053, over 5780228.02 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.339, pruned_loss=0.09403, over 5672133.44 frames. ], batch size: 186, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 04:57:51,746 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:57:54,796 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:57:54,842 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:57:58,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:58:04,260 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.37 vs. limit=5.0
+2023-03-03 04:58:05,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.239e+02 1.359e+03 1.825e+03 2.570e+03 4.329e+03, threshold=3.650e+03, percent-clipped=10.0
+2023-03-03 04:58:22,623 INFO [train.py:968] (0/2) Epoch 6, batch 16000, giga_loss[loss=0.2966, simple_loss=0.3673, pruned_loss=0.1129, over 28114.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3413, pruned_loss=0.09582, over 5670340.54 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3523, pruned_loss=0.1053, over 5766694.49 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3405, pruned_loss=0.0949, over 5669155.67 frames. ], batch size: 412, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 04:58:27,298 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:58:29,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3160, 1.4961, 1.5947, 1.3775], device='cuda:0'), covar=tensor([0.0944, 0.1117, 0.1233, 0.1165], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0711, 0.0626, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 04:59:27,953 INFO [train.py:968] (0/2) Epoch 6, batch 16050, giga_loss[loss=0.2719, simple_loss=0.3486, pruned_loss=0.09766, over 28633.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3435, pruned_loss=0.09698, over 5674454.30 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.352, pruned_loss=0.1051, over 5767767.73 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3429, pruned_loss=0.09624, over 5670845.46 frames. ], batch size: 307, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:00:10,442 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5797, 2.3311, 1.5773, 0.6772], device='cuda:0'), covar=tensor([0.2771, 0.1415, 0.2490, 0.3180], device='cuda:0'), in_proj_covar=tensor([0.1425, 0.1360, 0.1392, 0.1189], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 05:00:17,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.138e+02 1.349e+03 1.701e+03 2.440e+03 7.416e+03, threshold=3.401e+03, percent-clipped=14.0
+2023-03-03 05:00:34,801 INFO [train.py:968] (0/2) Epoch 6, batch 16100, giga_loss[loss=0.2781, simple_loss=0.3567, pruned_loss=0.09973, over 28593.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.344, pruned_loss=0.09778, over 5667366.11 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3519, pruned_loss=0.105, over 5768989.10 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3436, pruned_loss=0.09719, over 5662637.99 frames. ], batch size: 307, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:00:43,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3519, 2.0463, 1.4685, 0.5509], device='cuda:0'), covar=tensor([0.3013, 0.1335, 0.2182, 0.3180], device='cuda:0'), in_proj_covar=tensor([0.1430, 0.1366, 0.1400, 0.1193], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 05:00:55,245 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:00:58,392 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:01:01,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4636, 3.3435, 1.4656, 1.4477], device='cuda:0'), covar=tensor([0.0816, 0.0269, 0.0873, 0.1268], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0478, 0.0315, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 05:01:27,001 INFO [train.py:968] (0/2) Epoch 6, batch 16150, giga_loss[loss=0.2619, simple_loss=0.3522, pruned_loss=0.0858, over 28816.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3474, pruned_loss=0.09919, over 5680449.82 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3516, pruned_loss=0.1048, over 5771929.49 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3473, pruned_loss=0.09871, over 5670922.05 frames. ], batch size: 174, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:01:31,488 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:02:08,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.548e+02 1.249e+03 1.691e+03 2.279e+03 4.666e+03, threshold=3.382e+03, percent-clipped=8.0
+2023-03-03 05:02:16,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2843, 1.4958, 1.0400, 1.1756], device='cuda:0'), covar=tensor([0.1066, 0.0844, 0.0750, 0.0816], device='cuda:0'), in_proj_covar=tensor([0.1461, 0.1260, 0.1219, 0.1316], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 05:02:24,715 INFO [train.py:968] (0/2) Epoch 6, batch 16200, giga_loss[loss=0.3047, simple_loss=0.3746, pruned_loss=0.1174, over 28915.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3496, pruned_loss=0.09949, over 5685184.88 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3516, pruned_loss=0.1048, over 5773951.07 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3494, pruned_loss=0.09905, over 5674806.40 frames. ], batch size: 227, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:03:28,996 INFO [train.py:968] (0/2) Epoch 6, batch 16250, giga_loss[loss=0.2459, simple_loss=0.3273, pruned_loss=0.08226, over 29164.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3508, pruned_loss=0.1007, over 5676650.50 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3517, pruned_loss=0.1051, over 5757347.51 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3506, pruned_loss=0.1, over 5679942.57 frames. ], batch size: 200, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:04:18,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.448e+02 1.412e+03 2.157e+03 2.981e+03 6.791e+03, threshold=4.315e+03, percent-clipped=20.0
+2023-03-03 05:04:23,002 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-244000.pt
+2023-03-03 05:04:32,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1849, 4.0268, 3.8326, 1.9044], device='cuda:0'), covar=tensor([0.0456, 0.0546, 0.0719, 0.2177], device='cuda:0'), in_proj_covar=tensor([0.0869, 0.0810, 0.0763, 0.0594], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 05:04:36,953 INFO [train.py:968] (0/2) Epoch 6, batch 16300, giga_loss[loss=0.247, simple_loss=0.326, pruned_loss=0.084, over 29163.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3477, pruned_loss=0.09909, over 5684995.24 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3514, pruned_loss=0.105, over 5761180.14 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3477, pruned_loss=0.09851, over 5682536.44 frames. ], batch size: 113, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:05:14,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-03 05:05:45,040 INFO [train.py:968] (0/2) Epoch 6, batch 16350, giga_loss[loss=0.2707, simple_loss=0.3447, pruned_loss=0.09835, over 28109.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3474, pruned_loss=0.0992, over 5682941.28 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3515, pruned_loss=0.1051, over 5763249.55 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3473, pruned_loss=0.09854, over 5678260.60 frames. ], batch size: 412, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:05:45,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=244061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:05:54,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0838, 1.5822, 1.3966, 1.2755], device='cuda:0'), covar=tensor([0.1160, 0.1607, 0.0944, 0.1091], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0709, 0.0785, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 05:05:59,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5578, 1.9799, 1.7920, 1.7263], device='cuda:0'), covar=tensor([0.1242, 0.1533, 0.1544, 0.1611], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0720, 0.0633, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 05:06:26,401 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 05:06:32,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.267e+02 1.353e+03 1.749e+03 2.508e+03 9.668e+03, threshold=3.498e+03, percent-clipped=6.0
+2023-03-03 05:06:44,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9271, 1.9306, 1.8191, 1.8349], device='cuda:0'), covar=tensor([0.0984, 0.1696, 0.1562, 0.1374], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0718, 0.0630, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 05:06:47,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0168, 1.1642, 1.0354, 0.7639], device='cuda:0'), covar=tensor([0.1019, 0.0997, 0.0604, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.1448, 0.1256, 0.1211, 0.1307], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 05:06:47,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0772, 1.3854, 1.0848, 0.2202], device='cuda:0'), covar=tensor([0.1594, 0.1468, 0.2539, 0.2738], device='cuda:0'), in_proj_covar=tensor([0.1425, 0.1350, 0.1397, 0.1182], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 05:06:48,649 INFO [train.py:968] (0/2) Epoch 6, batch 16400, giga_loss[loss=0.2349, simple_loss=0.3144, pruned_loss=0.07768, over 29096.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3458, pruned_loss=0.09902, over 5672692.71 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3512, pruned_loss=0.105, over 5765627.21 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3459, pruned_loss=0.09855, over 5665515.28 frames. ], batch size: 128, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:07:49,067 INFO [train.py:968] (0/2) Epoch 6, batch 16450, libri_loss[loss=0.2802, simple_loss=0.3545, pruned_loss=0.1029, over 29222.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3432, pruned_loss=0.0982, over 5676584.00 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3512, pruned_loss=0.1049, over 5766209.84 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3431, pruned_loss=0.09774, over 5668271.06 frames. ], batch size: 97, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:07:49,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1725, 1.4072, 1.1774, 0.9057], device='cuda:0'), covar=tensor([0.1873, 0.1807, 0.1829, 0.1780], device='cuda:0'), in_proj_covar=tensor([0.1128, 0.0862, 0.1005, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 05:08:31,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.513e+02 1.321e+03 1.648e+03 2.271e+03 5.947e+03, threshold=3.295e+03, percent-clipped=7.0
+2023-03-03 05:08:37,643 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=244204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:08:40,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=244207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:08:46,769 INFO [train.py:968] (0/2) Epoch 6, batch 16500, giga_loss[loss=0.2345, simple_loss=0.3257, pruned_loss=0.07163, over 28895.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3438, pruned_loss=0.09813, over 5685908.35 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3512, pruned_loss=0.1049, over 5770768.44 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3435, pruned_loss=0.09757, over 5672300.33 frames. ], batch size: 164, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:09:15,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4119, 1.8002, 1.7335, 1.5289], device='cuda:0'), covar=tensor([0.1656, 0.1936, 0.1259, 0.1334], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0707, 0.0786, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 05:09:15,680 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=244236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:09:43,654 INFO [train.py:968] (0/2) Epoch 6, batch 16550, giga_loss[loss=0.2552, simple_loss=0.3277, pruned_loss=0.09135, over 27780.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3422, pruned_loss=0.09671, over 5678276.18 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3505, pruned_loss=0.1044, over 5772285.30 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3423, pruned_loss=0.09638, over 5662831.84 frames. ], batch size: 474, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:09:51,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 05:10:19,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0598, 2.9205, 2.7929, 1.5769], device='cuda:0'), covar=tensor([0.0623, 0.0744, 0.0775, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0808, 0.0758, 0.0592], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 05:10:26,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.546e+03 1.905e+03 2.672e+03 5.088e+03, threshold=3.810e+03, percent-clipped=7.0
+2023-03-03 05:10:39,478 INFO [train.py:968] (0/2) Epoch 6, batch 16600, libri_loss[loss=0.2506, simple_loss=0.3193, pruned_loss=0.09096, over 29623.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3431, pruned_loss=0.0957, over 5681997.99 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.35, pruned_loss=0.1043, over 5766623.71 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3434, pruned_loss=0.09538, over 5672432.59 frames. ], batch size: 69, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:11:37,607 INFO [train.py:968] (0/2) Epoch 6, batch 16650, giga_loss[loss=0.2466, simple_loss=0.3396, pruned_loss=0.07678, over 29010.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3448, pruned_loss=0.09537, over 5677560.06 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3502, pruned_loss=0.1045, over 5768810.98 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3448, pruned_loss=0.09474, over 5666898.43 frames. ], batch size: 128, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:12:22,619 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.576e+02 1.315e+03 1.693e+03 2.528e+03 7.059e+03, threshold=3.385e+03, percent-clipped=8.0
+2023-03-03 05:12:35,647 INFO [train.py:968] (0/2) Epoch 6, batch 16700, giga_loss[loss=0.2486, simple_loss=0.3362, pruned_loss=0.08051, over 28907.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3453, pruned_loss=0.09499, over 5680385.79 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3501, pruned_loss=0.1045, over 5759240.39 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3452, pruned_loss=0.0943, over 5678545.26 frames. ], batch size: 174, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:13:34,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6946, 3.5555, 3.3350, 1.7910], device='cuda:0'), covar=tensor([0.0658, 0.0678, 0.0847, 0.2053], device='cuda:0'), in_proj_covar=tensor([0.0862, 0.0811, 0.0757, 0.0592], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 05:13:45,112 INFO [train.py:968] (0/2) Epoch 6, batch 16750, giga_loss[loss=0.2247, simple_loss=0.2944, pruned_loss=0.07751, over 24467.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3452, pruned_loss=0.09507, over 5675152.51 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3498, pruned_loss=0.1042, over 5760529.30 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3455, pruned_loss=0.0947, over 5671802.10 frames. ], batch size: 705, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:14:36,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.099e+02 1.448e+03 1.905e+03 2.643e+03 5.250e+03, threshold=3.810e+03, percent-clipped=8.0
+2023-03-03 05:14:53,340 INFO [train.py:968] (0/2) Epoch 6, batch 16800, giga_loss[loss=0.2976, simple_loss=0.3727, pruned_loss=0.1113, over 28673.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3451, pruned_loss=0.09476, over 5676550.04 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3497, pruned_loss=0.1041, over 5764132.68 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3454, pruned_loss=0.09437, over 5668497.23 frames. ], batch size: 262, lr: 5.41e-03, grad_scale: 8.0
+2023-03-03 05:15:57,685 INFO [train.py:968] (0/2) Epoch 6, batch 16850, libri_loss[loss=0.2335, simple_loss=0.3039, pruned_loss=0.08153, over 29570.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3451, pruned_loss=0.09453, over 5684384.09 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3489, pruned_loss=0.1036, over 5770093.09 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3458, pruned_loss=0.09422, over 5668966.29 frames. ], batch size: 74, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:16:21,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2947, 1.2003, 5.0755, 3.2004], device='cuda:0'), covar=tensor([0.1659, 0.2486, 0.0278, 0.0737], device='cuda:0'), in_proj_covar=tensor([0.0571, 0.0541, 0.0751, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 05:16:52,686 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.607e+02 1.380e+03 1.869e+03 2.300e+03 5.879e+03, threshold=3.738e+03, percent-clipped=5.0
+2023-03-03 05:17:09,090 INFO [train.py:968] (0/2) Epoch 6, batch 16900, giga_loss[loss=0.3176, simple_loss=0.3939, pruned_loss=0.1207, over 28974.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3466, pruned_loss=0.09523, over 5686485.34 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3493, pruned_loss=0.1039, over 5772244.18 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3466, pruned_loss=0.09456, over 5670617.03 frames. ], batch size: 186, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:18:17,207 INFO [train.py:968] (0/2) Epoch 6, batch 16950, giga_loss[loss=0.2572, simple_loss=0.3227, pruned_loss=0.09589, over 24770.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3502, pruned_loss=0.09703, over 5687397.32 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3492, pruned_loss=0.1038, over 5774216.83 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3503, pruned_loss=0.09641, over 5671163.75 frames. ], batch size: 705, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:19:05,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.922e+02 1.367e+03 1.849e+03 2.344e+03 7.119e+03, threshold=3.699e+03, percent-clipped=5.0
+2023-03-03 05:19:19,715 INFO [train.py:968] (0/2) Epoch 6, batch 17000, giga_loss[loss=0.2317, simple_loss=0.3203, pruned_loss=0.07155, over 28784.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3487, pruned_loss=0.09608, over 5692419.21 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3486, pruned_loss=0.1034, over 5778955.93 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3494, pruned_loss=0.09572, over 5672331.71 frames. ], batch size: 243, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:20:08,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=244746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:20:32,396 INFO [train.py:968] (0/2) Epoch 6, batch 17050, giga_loss[loss=0.2792, simple_loss=0.3511, pruned_loss=0.1036, over 28163.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3473, pruned_loss=0.09625, over 5700582.73 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3488, pruned_loss=0.1035, over 5780151.52 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3476, pruned_loss=0.09582, over 5682882.66 frames. ], batch size: 412, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:20:39,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2344, 1.3023, 1.2447, 1.4336], device='cuda:0'), covar=tensor([0.0765, 0.0325, 0.0336, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0126, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0063, 0.0045, 0.0041, 0.0069], device='cuda:0')
+2023-03-03 05:21:23,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.090e+02 1.261e+03 1.660e+03 2.308e+03 8.692e+03, threshold=3.321e+03, percent-clipped=10.0
+2023-03-03 05:21:40,570 INFO [train.py:968] (0/2) Epoch 6, batch 17100, libri_loss[loss=0.292, simple_loss=0.3665, pruned_loss=0.1088, over 29756.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3464, pruned_loss=0.09536, over 5696886.75 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3495, pruned_loss=0.1038, over 5775406.20 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3459, pruned_loss=0.09446, over 5683814.72 frames. ], batch size: 87, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:22:44,565 INFO [train.py:968] (0/2) Epoch 6, batch 17150, giga_loss[loss=0.2843, simple_loss=0.3615, pruned_loss=0.1035, over 28479.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3446, pruned_loss=0.09385, over 5709333.19 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3494, pruned_loss=0.1037, over 5778553.76 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3442, pruned_loss=0.09288, over 5693883.92 frames. ], batch size: 336, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:22:46,417 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6678, 3.5057, 3.3011, 1.6432], device='cuda:0'), covar=tensor([0.0571, 0.0646, 0.0787, 0.2359], device='cuda:0'), in_proj_covar=tensor([0.0876, 0.0803, 0.0760, 0.0597], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 05:23:32,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.984e+02 1.225e+03 1.622e+03 2.461e+03 6.422e+03, threshold=3.244e+03, percent-clipped=6.0
+2023-03-03 05:23:46,273 INFO [train.py:968] (0/2) Epoch 6, batch 17200, giga_loss[loss=0.2814, simple_loss=0.3626, pruned_loss=0.1001, over 28042.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3451, pruned_loss=0.09456, over 5690721.22 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3494, pruned_loss=0.1036, over 5770706.90 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3447, pruned_loss=0.09366, over 5683402.51 frames. ], batch size: 412, lr: 5.41e-03, grad_scale: 8.0
+2023-03-03 05:24:46,674 INFO [train.py:968] (0/2) Epoch 6, batch 17250, giga_loss[loss=0.2909, simple_loss=0.3579, pruned_loss=0.112, over 26706.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3472, pruned_loss=0.09576, over 5688401.62 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3488, pruned_loss=0.1033, over 5772127.87 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3474, pruned_loss=0.09524, over 5679668.62 frames. ], batch size: 555, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:25:27,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.211e+02 1.436e+03 1.852e+03 2.435e+03 1.191e+04, threshold=3.704e+03, percent-clipped=7.0
+2023-03-03 05:25:37,371 INFO [train.py:968] (0/2) Epoch 6, batch 17300, giga_loss[loss=0.2394, simple_loss=0.3026, pruned_loss=0.08815, over 24264.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3478, pruned_loss=0.09704, over 5680138.80 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3482, pruned_loss=0.1031, over 5765229.93 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3485, pruned_loss=0.09658, over 5675866.53 frames. ], batch size: 705, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:26:35,799 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4795, 4.3365, 4.1134, 1.7092], device='cuda:0'), covar=tensor([0.0441, 0.0532, 0.0654, 0.2088], device='cuda:0'), in_proj_covar=tensor([0.0870, 0.0799, 0.0753, 0.0588], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 05:26:36,143 INFO [train.py:968] (0/2) Epoch 6, batch 17350, giga_loss[loss=0.2542, simple_loss=0.3298, pruned_loss=0.0893, over 28853.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3455, pruned_loss=0.09678, over 5678214.25 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3481, pruned_loss=0.103, over 5763661.94 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3461, pruned_loss=0.09641, over 5675317.43 frames. ], batch size: 199, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:26:50,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4240, 1.8691, 1.3248, 0.6183], device='cuda:0'), covar=tensor([0.2789, 0.1429, 0.2274, 0.3176], device='cuda:0'), in_proj_covar=tensor([0.1419, 0.1346, 0.1377, 0.1178], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 05:27:24,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.041e+02 1.364e+03 1.695e+03 2.275e+03 5.447e+03, threshold=3.390e+03, percent-clipped=5.0
+2023-03-03 05:27:36,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:27:36,839 INFO [train.py:968] (0/2) Epoch 6, batch 17400, giga_loss[loss=0.3153, simple_loss=0.3756, pruned_loss=0.1275, over 28556.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.346, pruned_loss=0.09787, over 5677414.01 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3484, pruned_loss=0.1032, over 5761782.56 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3462, pruned_loss=0.09727, over 5675833.59 frames. ], batch size: 307, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:27:48,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:28:31,733 INFO [train.py:968] (0/2) Epoch 6, batch 17450, giga_loss[loss=0.3133, simple_loss=0.3797, pruned_loss=0.1235, over 28378.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3527, pruned_loss=0.1029, over 5679979.67 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3484, pruned_loss=0.1033, over 5763757.32 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3529, pruned_loss=0.1023, over 5675400.73 frames. ], batch size: 368, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:28:59,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6653, 1.9084, 1.4754, 1.4084], device='cuda:0'), covar=tensor([0.1335, 0.1029, 0.0926, 0.1097], device='cuda:0'), in_proj_covar=tensor([0.1461, 0.1268, 0.1220, 0.1325], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 05:29:05,469 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 05:29:10,174 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.144e+02 1.458e+03 1.914e+03 2.941e+03 7.304e+03, threshold=3.829e+03, percent-clipped=17.0
+2023-03-03 05:29:17,787 INFO [train.py:968] (0/2) Epoch 6, batch 17500, giga_loss[loss=0.3617, simple_loss=0.4104, pruned_loss=0.1565, over 27497.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3619, pruned_loss=0.109, over 5678507.21 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3482, pruned_loss=0.1031, over 5753990.84 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3626, pruned_loss=0.1088, over 5679503.42 frames. ], batch size: 472, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:29:23,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3634, 1.4450, 1.1899, 1.8058], device='cuda:0'), covar=tensor([0.2308, 0.2139, 0.2415, 0.1942], device='cuda:0'), in_proj_covar=tensor([0.1125, 0.0856, 0.1001, 0.0937], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 05:29:58,119 INFO [train.py:968] (0/2) Epoch 6, batch 17550, giga_loss[loss=0.2853, simple_loss=0.3536, pruned_loss=0.1085, over 28562.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3684, pruned_loss=0.1129, over 5683341.93 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3489, pruned_loss=0.1035, over 5748199.49 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3689, pruned_loss=0.1126, over 5687100.79 frames. ], batch size: 60, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:30:01,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:30:02,137 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-03 05:30:03,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:30:16,839 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:30:28,761 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:30:31,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.955e+02 1.228e+03 1.555e+03 2.117e+03 3.632e+03, threshold=3.110e+03, percent-clipped=0.0
+2023-03-03 05:30:43,204 INFO [train.py:968] (0/2) Epoch 6, batch 17600, giga_loss[loss=0.2891, simple_loss=0.3561, pruned_loss=0.111, over 28037.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3654, pruned_loss=0.112, over 5687155.67 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3489, pruned_loss=0.1034, over 5753474.19 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3664, pruned_loss=0.1122, over 5683643.64 frames. ], batch size: 412, lr: 5.41e-03, grad_scale: 8.0
+2023-03-03 05:31:05,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5074, 1.6117, 1.3405, 1.6374], device='cuda:0'), covar=tensor([0.2095, 0.1968, 0.2168, 0.1898], device='cuda:0'), in_proj_covar=tensor([0.1125, 0.0857, 0.1001, 0.0940], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 05:31:14,777 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:31:26,872 INFO [train.py:968] (0/2) Epoch 6, batch 17650, giga_loss[loss=0.2295, simple_loss=0.3057, pruned_loss=0.07669, over 28955.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3576, pruned_loss=0.1088, over 5688663.29 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3487, pruned_loss=0.1032, over 5755733.11 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3587, pruned_loss=0.1092, over 5683066.21 frames. ], batch size: 145, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:32:04,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.062e+02 1.020e+03 1.289e+03 1.668e+03 8.301e+03, threshold=2.577e+03, percent-clipped=7.0
+2023-03-03 05:32:11,833 INFO [train.py:968] (0/2) Epoch 6, batch 17700, giga_loss[loss=0.2307, simple_loss=0.2997, pruned_loss=0.08081, over 28994.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3504, pruned_loss=0.1057, over 5687301.29 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3488, pruned_loss=0.1032, over 5759791.52 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3514, pruned_loss=0.1062, over 5677163.85 frames. ], batch size: 100, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:32:19,889 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-03 05:32:56,077 INFO [train.py:968] (0/2) Epoch 6, batch 17750, giga_loss[loss=0.2269, simple_loss=0.2972, pruned_loss=0.07826, over 28689.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3416, pruned_loss=0.1012, over 5694012.78 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3485, pruned_loss=0.1029, over 5763272.53 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3425, pruned_loss=0.1019, over 5681141.20 frames. ], batch size: 262, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:33:16,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245485.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:33:29,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.161e+02 9.520e+02 1.276e+03 1.687e+03 3.816e+03, threshold=2.552e+03, percent-clipped=10.0
+2023-03-03 05:33:38,283 INFO [train.py:968] (0/2) Epoch 6, batch 17800, giga_loss[loss=0.2329, simple_loss=0.3081, pruned_loss=0.07882, over 28022.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3343, pruned_loss=0.09787, over 5696867.08 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3482, pruned_loss=0.1027, over 5765914.90 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3352, pruned_loss=0.09849, over 5682855.70 frames. ], batch size: 77, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:34:00,646 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2064, 1.8776, 1.4541, 1.7108], device='cuda:0'), covar=tensor([0.0623, 0.0739, 0.0997, 0.1037], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0442, 0.0501, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 05:34:03,827 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-03 05:34:13,167 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4596, 1.6327, 1.3508, 1.6314], device='cuda:0'), covar=tensor([0.0780, 0.0306, 0.0317, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0126, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0063, 0.0045, 0.0041, 0.0069], device='cuda:0')
+2023-03-03 05:34:17,006 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-03 05:34:19,581 INFO [train.py:968] (0/2) Epoch 6, batch 17850, giga_loss[loss=0.2375, simple_loss=0.3099, pruned_loss=0.08252, over 28878.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3292, pruned_loss=0.09506, over 5699655.25 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3486, pruned_loss=0.1028, over 5766339.36 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3291, pruned_loss=0.09529, over 5686113.97 frames. ], batch size: 186, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:34:53,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.223e+02 1.080e+03 1.408e+03 1.825e+03 7.441e+03, threshold=2.817e+03, percent-clipped=9.0
+2023-03-03 05:35:01,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5965, 3.2880, 1.6905, 1.4746], device='cuda:0'), covar=tensor([0.0831, 0.0329, 0.0803, 0.1294], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0476, 0.0310, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0021], device='cuda:0')
+2023-03-03 05:35:02,020 INFO [train.py:968] (0/2) Epoch 6, batch 17900, giga_loss[loss=0.2256, simple_loss=0.2956, pruned_loss=0.07783, over 28996.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3268, pruned_loss=0.09415, over 5703205.87 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3489, pruned_loss=0.1028, over 5767303.55 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3263, pruned_loss=0.09423, over 5691168.51 frames. ], batch size: 106, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:35:19,920 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245628.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:35:21,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245631.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:35:38,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2946, 1.3842, 1.0891, 1.1875], device='cuda:0'), covar=tensor([0.0912, 0.0769, 0.0635, 0.0766], device='cuda:0'), in_proj_covar=tensor([0.1451, 0.1264, 0.1211, 0.1325], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 05:35:42,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:35:45,757 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245660.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:35:46,158 INFO [train.py:968] (0/2) Epoch 6, batch 17950, giga_loss[loss=0.2479, simple_loss=0.3153, pruned_loss=0.09027, over 28713.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3239, pruned_loss=0.09281, over 5700609.34 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3493, pruned_loss=0.103, over 5771462.47 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3223, pruned_loss=0.09244, over 5685147.36 frames. ], batch size: 284, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:36:20,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.807e+02 9.954e+02 1.368e+03 1.969e+03 9.797e+03, threshold=2.735e+03, percent-clipped=14.0
+2023-03-03 05:36:27,672 INFO [train.py:968] (0/2) Epoch 6, batch 18000, giga_loss[loss=0.232, simple_loss=0.3027, pruned_loss=0.08066, over 28821.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3199, pruned_loss=0.09041, over 5704350.52 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3493, pruned_loss=0.103, over 5764607.47 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3182, pruned_loss=0.08991, over 5697014.06 frames. ], batch size: 285, lr: 5.40e-03, grad_scale: 8.0
+2023-03-03 05:36:27,676 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 05:36:36,326 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2317, simple_loss=0.3331, pruned_loss=0.06516, over 944034.00 frames. 
+2023-03-03 05:36:36,327 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 05:36:45,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:37:05,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2281, 1.4425, 1.1291, 0.8992], device='cuda:0'), covar=tensor([0.1230, 0.1089, 0.0761, 0.1044], device='cuda:0'), in_proj_covar=tensor([0.1456, 0.1273, 0.1221, 0.1333], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 05:37:18,150 INFO [train.py:968] (0/2) Epoch 6, batch 18050, giga_loss[loss=0.2313, simple_loss=0.3021, pruned_loss=0.08024, over 28948.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3184, pruned_loss=0.0899, over 5682381.33 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3501, pruned_loss=0.1033, over 5753108.89 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3155, pruned_loss=0.08884, over 5683961.26 frames. ], batch size: 106, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:37:48,287 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:37:50,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:37:54,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.398e+02 8.949e+02 1.191e+03 1.701e+03 4.749e+03, threshold=2.381e+03, percent-clipped=8.0
+2023-03-03 05:37:59,709 INFO [train.py:968] (0/2) Epoch 6, batch 18100, libri_loss[loss=0.328, simple_loss=0.4002, pruned_loss=0.1279, over 26242.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3166, pruned_loss=0.08884, over 5684925.55 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3504, pruned_loss=0.1034, over 5748132.95 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3127, pruned_loss=0.08739, over 5688663.32 frames. ], batch size: 137, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:38:15,174 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:38:34,116 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:38:38,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9653, 1.0156, 3.9015, 2.9905], device='cuda:0'), covar=tensor([0.1627, 0.2343, 0.0381, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0574, 0.0535, 0.0752, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 05:38:39,308 INFO [train.py:968] (0/2) Epoch 6, batch 18150, giga_loss[loss=0.207, simple_loss=0.2837, pruned_loss=0.06517, over 29034.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3135, pruned_loss=0.08711, over 5690505.94 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3507, pruned_loss=0.1033, over 5752840.80 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.3088, pruned_loss=0.08551, over 5687093.87 frames. ], batch size: 128, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:38:42,335 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:38:45,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:39:11,659 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:39:16,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.657e+02 9.659e+02 1.189e+03 1.595e+03 3.753e+03, threshold=2.378e+03, percent-clipped=7.0
+2023-03-03 05:39:25,152 INFO [train.py:968] (0/2) Epoch 6, batch 18200, giga_loss[loss=0.2168, simple_loss=0.2862, pruned_loss=0.0737, over 28973.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3101, pruned_loss=0.08551, over 5686322.27 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3506, pruned_loss=0.1032, over 5755199.45 frames. ], giga_tot_loss[loss=0.237, simple_loss=0.3059, pruned_loss=0.0841, over 5680568.34 frames. ], batch size: 136, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:39:53,042 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:40:01,695 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9857, 3.0649, 2.1102, 1.0270], device='cuda:0'), covar=tensor([0.3851, 0.1744, 0.2272, 0.3667], device='cuda:0'), in_proj_covar=tensor([0.1406, 0.1332, 0.1377, 0.1172], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 05:40:10,615 INFO [train.py:968] (0/2) Epoch 6, batch 18250, giga_loss[loss=0.1969, simple_loss=0.2765, pruned_loss=0.05864, over 28911.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3085, pruned_loss=0.08506, over 5686387.59 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3511, pruned_loss=0.1034, over 5757008.84 frames. ], giga_tot_loss[loss=0.2347, simple_loss=0.3032, pruned_loss=0.08308, over 5677577.10 frames. ], batch size: 145, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:40:43,558 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-246000.pt
+2023-03-03 05:40:49,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.435e+02 1.301e+03 1.644e+03 2.953e+03 8.829e+03, threshold=3.288e+03, percent-clipped=28.0
+2023-03-03 05:40:57,809 INFO [train.py:968] (0/2) Epoch 6, batch 18300, giga_loss[loss=0.2506, simple_loss=0.3247, pruned_loss=0.08824, over 28236.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3162, pruned_loss=0.09043, over 5675253.38 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3515, pruned_loss=0.1036, over 5758362.67 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3102, pruned_loss=0.08815, over 5664851.59 frames. ], batch size: 77, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:41:46,961 INFO [train.py:968] (0/2) Epoch 6, batch 18350, giga_loss[loss=0.3257, simple_loss=0.3871, pruned_loss=0.1322, over 28230.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3309, pruned_loss=0.09883, over 5661730.91 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3519, pruned_loss=0.1038, over 5739428.16 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3255, pruned_loss=0.09671, over 5669976.33 frames. ], batch size: 368, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:42:01,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246079.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:42:20,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.521e+02 1.234e+03 1.535e+03 1.858e+03 3.435e+03, threshold=3.069e+03, percent-clipped=2.0
+2023-03-03 05:42:25,470 INFO [train.py:968] (0/2) Epoch 6, batch 18400, libri_loss[loss=0.3117, simple_loss=0.3859, pruned_loss=0.1187, over 29244.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3446, pruned_loss=0.1062, over 5674765.17 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3528, pruned_loss=0.1043, over 5735349.30 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.339, pruned_loss=0.104, over 5682459.17 frames. ], batch size: 97, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:43:05,885 INFO [train.py:968] (0/2) Epoch 6, batch 18450, giga_loss[loss=0.2715, simple_loss=0.3469, pruned_loss=0.09806, over 28927.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3526, pruned_loss=0.1099, over 5680842.49 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3531, pruned_loss=0.1045, over 5738372.95 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3479, pruned_loss=0.1081, over 5683338.54 frames. ], batch size: 106, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:43:17,336 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-03 05:43:27,876 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:43:43,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.581e+02 1.255e+03 1.576e+03 2.068e+03 1.060e+04, threshold=3.153e+03, percent-clipped=7.0
+2023-03-03 05:43:48,079 INFO [train.py:968] (0/2) Epoch 6, batch 18500, giga_loss[loss=0.3039, simple_loss=0.3705, pruned_loss=0.1187, over 28354.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3569, pruned_loss=0.1107, over 5682551.25 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3535, pruned_loss=0.1047, over 5741698.70 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3528, pruned_loss=0.1092, over 5680417.08 frames. ], batch size: 65, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:44:02,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:44:31,478 INFO [train.py:968] (0/2) Epoch 6, batch 18550, giga_loss[loss=0.2692, simple_loss=0.3345, pruned_loss=0.1019, over 28687.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3579, pruned_loss=0.1098, over 5679529.46 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3539, pruned_loss=0.1048, over 5743006.34 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3543, pruned_loss=0.1085, over 5675861.43 frames. ], batch size: 92, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:44:53,305 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:45:13,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.882e+02 1.102e+03 1.389e+03 1.787e+03 3.698e+03, threshold=2.777e+03, percent-clipped=5.0
+2023-03-03 05:45:18,412 INFO [train.py:968] (0/2) Epoch 6, batch 18600, giga_loss[loss=0.2448, simple_loss=0.3231, pruned_loss=0.08328, over 28473.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3604, pruned_loss=0.1115, over 5671188.43 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3542, pruned_loss=0.1049, over 5743394.05 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3573, pruned_loss=0.1105, over 5667171.07 frames. ], batch size: 60, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:45:20,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:45:59,770 INFO [train.py:968] (0/2) Epoch 6, batch 18650, giga_loss[loss=0.3598, simple_loss=0.3979, pruned_loss=0.1609, over 23611.00 frames. ], tot_loss[loss=0.295, simple_loss=0.363, pruned_loss=0.1135, over 5675610.54 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.355, pruned_loss=0.1052, over 5745403.56 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3601, pruned_loss=0.1127, over 5667938.19 frames. ], batch size: 705, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:46:08,597 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:46:10,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:46:37,977 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:46:38,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.142e+02 1.257e+03 1.539e+03 1.971e+03 5.147e+03, threshold=3.079e+03, percent-clipped=9.0
+2023-03-03 05:46:44,155 INFO [train.py:968] (0/2) Epoch 6, batch 18700, giga_loss[loss=0.2997, simple_loss=0.3694, pruned_loss=0.115, over 28364.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3658, pruned_loss=0.1153, over 5679261.66 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3552, pruned_loss=0.1054, over 5747713.51 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3634, pruned_loss=0.1146, over 5670454.15 frames. ], batch size: 65, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:47:18,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246454.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:47:22,056 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:47:24,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:47:24,668 INFO [train.py:968] (0/2) Epoch 6, batch 18750, giga_loss[loss=0.3311, simple_loss=0.406, pruned_loss=0.1281, over 28620.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.37, pruned_loss=0.1172, over 5674243.20 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3562, pruned_loss=0.1057, over 5737876.77 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3675, pruned_loss=0.1166, over 5674728.07 frames. ], batch size: 71, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:47:47,893 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:47:58,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2361, 1.5491, 1.2392, 1.3210], device='cuda:0'), covar=tensor([0.2216, 0.2038, 0.2174, 0.1966], device='cuda:0'), in_proj_covar=tensor([0.1142, 0.0876, 0.1014, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 05:47:59,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.379e+02 1.143e+03 1.402e+03 1.728e+03 4.235e+03, threshold=2.804e+03, percent-clipped=3.0
+2023-03-03 05:48:00,173 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:48:06,881 INFO [train.py:968] (0/2) Epoch 6, batch 18800, giga_loss[loss=0.3176, simple_loss=0.3901, pruned_loss=0.1225, over 29070.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3721, pruned_loss=0.1173, over 5675658.60 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3564, pruned_loss=0.1057, over 5740324.12 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3701, pruned_loss=0.117, over 5672960.62 frames. ], batch size: 106, lr: 5.39e-03, grad_scale: 8.0
+2023-03-03 05:48:11,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1415, 1.6624, 1.2843, 0.3332], device='cuda:0'), covar=tensor([0.1498, 0.0893, 0.1360, 0.2266], device='cuda:0'), in_proj_covar=tensor([0.1403, 0.1321, 0.1376, 0.1170], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 05:48:34,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3156, 2.9311, 1.3269, 1.3134], device='cuda:0'), covar=tensor([0.0945, 0.0259, 0.0898, 0.1389], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0475, 0.0311, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0025, 0.0017, 0.0021], device='cuda:0')
+2023-03-03 05:48:46,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:48:46,815 INFO [train.py:968] (0/2) Epoch 6, batch 18850, giga_loss[loss=0.325, simple_loss=0.397, pruned_loss=0.1265, over 29056.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3738, pruned_loss=0.1175, over 5683955.71 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3567, pruned_loss=0.1059, over 5739339.41 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3722, pruned_loss=0.1172, over 5681904.94 frames. ], batch size: 164, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:49:10,533 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246589.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:49:15,602 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246597.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:49:17,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246600.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:49:22,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.574e+02 1.149e+03 1.477e+03 1.867e+03 5.552e+03, threshold=2.954e+03, percent-clipped=3.0
+2023-03-03 05:49:26,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6649, 2.3957, 1.5514, 0.8337], device='cuda:0'), covar=tensor([0.4278, 0.2263, 0.2438, 0.3786], device='cuda:0'), in_proj_covar=tensor([0.1421, 0.1336, 0.1391, 0.1180], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 05:49:26,473 INFO [train.py:968] (0/2) Epoch 6, batch 18900, giga_loss[loss=0.2728, simple_loss=0.3543, pruned_loss=0.09569, over 28735.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3728, pruned_loss=0.1159, over 5690504.72 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3568, pruned_loss=0.106, over 5741490.06 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.372, pruned_loss=0.116, over 5685255.76 frames. ], batch size: 284, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:49:29,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:49:41,027 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246629.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:49:41,304 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-03 05:50:06,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:50:07,167 INFO [train.py:968] (0/2) Epoch 6, batch 18950, giga_loss[loss=0.2738, simple_loss=0.3562, pruned_loss=0.09566, over 28882.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3711, pruned_loss=0.1134, over 5707566.65 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.357, pruned_loss=0.106, over 5745856.12 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3706, pruned_loss=0.1136, over 5698437.88 frames. ], batch size: 86, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:50:12,675 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6093, 1.0255, 2.8729, 2.6064], device='cuda:0'), covar=tensor([0.1658, 0.2247, 0.0502, 0.0686], device='cuda:0'), in_proj_covar=tensor([0.0579, 0.0540, 0.0757, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 05:50:42,023 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:50:42,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.511e+02 1.016e+03 1.345e+03 1.662e+03 4.140e+03, threshold=2.690e+03, percent-clipped=4.0
+2023-03-03 05:50:44,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:50:47,503 INFO [train.py:968] (0/2) Epoch 6, batch 19000, giga_loss[loss=0.3184, simple_loss=0.3859, pruned_loss=0.1254, over 29013.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3692, pruned_loss=0.1122, over 5703307.14 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3565, pruned_loss=0.1057, over 5746572.48 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3695, pruned_loss=0.1128, over 5694599.33 frames. ], batch size: 136, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:51:07,597 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:51:29,914 INFO [train.py:968] (0/2) Epoch 6, batch 19050, giga_loss[loss=0.248, simple_loss=0.3351, pruned_loss=0.08049, over 28615.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3706, pruned_loss=0.1142, over 5712432.27 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3571, pruned_loss=0.1059, over 5746207.82 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3706, pruned_loss=0.1146, over 5705136.00 frames. ], batch size: 60, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:52:07,832 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:52:10,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.730e+02 1.194e+03 1.538e+03 2.135e+03 5.804e+03, threshold=3.075e+03, percent-clipped=17.0
+2023-03-03 05:52:11,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:52:18,732 INFO [train.py:968] (0/2) Epoch 6, batch 19100, giga_loss[loss=0.3871, simple_loss=0.4196, pruned_loss=0.1773, over 27900.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3741, pruned_loss=0.1201, over 5709444.16 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3574, pruned_loss=0.1061, over 5747841.91 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3741, pruned_loss=0.1205, over 5701424.18 frames. ], batch size: 412, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:52:36,432 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:52:39,273 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-03 05:52:58,259 INFO [train.py:968] (0/2) Epoch 6, batch 19150, giga_loss[loss=0.2834, simple_loss=0.3532, pruned_loss=0.1068, over 28687.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3745, pruned_loss=0.1217, over 5703516.52 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3581, pruned_loss=0.1063, over 5739997.23 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3742, pruned_loss=0.1221, over 5703991.49 frames. ], batch size: 60, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:53:12,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:53:19,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9940, 1.1744, 3.9981, 3.2589], device='cuda:0'), covar=tensor([0.1522, 0.2175, 0.0327, 0.0592], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0534, 0.0748, 0.0629], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 05:53:31,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.519e+02 1.286e+03 1.559e+03 2.331e+03 4.913e+03, threshold=3.118e+03, percent-clipped=10.0
+2023-03-03 05:53:36,871 INFO [train.py:968] (0/2) Epoch 6, batch 19200, giga_loss[loss=0.2711, simple_loss=0.3601, pruned_loss=0.09109, over 28249.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3722, pruned_loss=0.1208, over 5703566.13 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3584, pruned_loss=0.1064, over 5742669.57 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.372, pruned_loss=0.1215, over 5700623.78 frames. ], batch size: 77, lr: 5.39e-03, grad_scale: 8.0
+2023-03-03 05:53:48,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3487, 2.0483, 1.5916, 0.5396], device='cuda:0'), covar=tensor([0.2456, 0.1387, 0.2355, 0.2820], device='cuda:0'), in_proj_covar=tensor([0.1417, 0.1325, 0.1385, 0.1173], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 05:54:19,753 INFO [train.py:968] (0/2) Epoch 6, batch 19250, giga_loss[loss=0.3142, simple_loss=0.3794, pruned_loss=0.1245, over 28900.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3703, pruned_loss=0.1201, over 5701977.35 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3588, pruned_loss=0.1064, over 5744628.45 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3701, pruned_loss=0.1208, over 5697227.43 frames. ], batch size: 186, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:54:23,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246964.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:54:38,974 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9393, 1.0797, 4.0856, 3.2192], device='cuda:0'), covar=tensor([0.2284, 0.2792, 0.0581, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0578, 0.0539, 0.0751, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 05:54:45,990 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:54:50,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6824, 1.1550, 5.2925, 3.6418], device='cuda:0'), covar=tensor([0.1389, 0.2277, 0.0270, 0.0596], device='cuda:0'), in_proj_covar=tensor([0.0578, 0.0539, 0.0752, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 05:54:58,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.319e+02 1.203e+03 1.666e+03 2.181e+03 4.497e+03, threshold=3.331e+03, percent-clipped=2.0
+2023-03-03 05:55:03,248 INFO [train.py:968] (0/2) Epoch 6, batch 19300, giga_loss[loss=0.2872, simple_loss=0.3605, pruned_loss=0.107, over 28938.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3696, pruned_loss=0.1188, over 5714042.68 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3592, pruned_loss=0.1065, over 5750504.72 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3694, pruned_loss=0.1198, over 5703935.50 frames. ], batch size: 186, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:55:12,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5130, 1.7721, 1.7779, 1.5901], device='cuda:0'), covar=tensor([0.1562, 0.1936, 0.1199, 0.1303], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0727, 0.0798, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 05:55:13,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:55:15,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:55:16,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2084, 3.0291, 2.8632, 1.2697], device='cuda:0'), covar=tensor([0.0888, 0.0953, 0.1039, 0.2529], device='cuda:0'), in_proj_covar=tensor([0.0874, 0.0819, 0.0771, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 05:55:38,240 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:55:44,878 INFO [train.py:968] (0/2) Epoch 6, batch 19350, giga_loss[loss=0.3393, simple_loss=0.3831, pruned_loss=0.1478, over 26533.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3682, pruned_loss=0.1175, over 5707986.95 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3595, pruned_loss=0.1066, over 5749832.48 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3679, pruned_loss=0.1183, over 5699967.59 frames. ], batch size: 555, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:55:51,072 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:56:03,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4207, 1.6008, 1.3029, 1.8720], device='cuda:0'), covar=tensor([0.2204, 0.2056, 0.2181, 0.2237], device='cuda:0'), in_proj_covar=tensor([0.1133, 0.0862, 0.1004, 0.0938], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 05:56:23,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.304e+02 1.083e+03 1.499e+03 2.176e+03 9.991e+03, threshold=2.997e+03, percent-clipped=6.0
+2023-03-03 05:56:25,955 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247107.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:56:28,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:56:29,131 INFO [train.py:968] (0/2) Epoch 6, batch 19400, giga_loss[loss=0.2568, simple_loss=0.3096, pruned_loss=0.102, over 23367.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3637, pruned_loss=0.1143, over 5695611.98 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3601, pruned_loss=0.1069, over 5752825.55 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3631, pruned_loss=0.1149, over 5685591.73 frames. ], batch size: 705, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:56:44,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:56:47,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:56:50,781 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247139.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:57:12,493 INFO [train.py:968] (0/2) Epoch 6, batch 19450, giga_loss[loss=0.2704, simple_loss=0.3391, pruned_loss=0.1009, over 28853.00 frames. ], tot_loss[loss=0.289, simple_loss=0.357, pruned_loss=0.1105, over 5686013.70 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3602, pruned_loss=0.1069, over 5751973.42 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3564, pruned_loss=0.1111, over 5677311.56 frames. ], batch size: 174, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 05:57:16,346 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:57:18,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2554, 1.7660, 1.3665, 0.4833], device='cuda:0'), covar=tensor([0.2063, 0.1314, 0.2125, 0.2843], device='cuda:0'), in_proj_covar=tensor([0.1387, 0.1297, 0.1353, 0.1144], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:0')
+2023-03-03 05:57:51,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.710e+02 9.310e+02 1.235e+03 1.685e+03 6.677e+03, threshold=2.471e+03, percent-clipped=9.0
+2023-03-03 05:57:55,967 INFO [train.py:968] (0/2) Epoch 6, batch 19500, giga_loss[loss=0.2382, simple_loss=0.3112, pruned_loss=0.08262, over 28579.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3517, pruned_loss=0.1072, over 5694136.68 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3604, pruned_loss=0.1069, over 5755706.68 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3509, pruned_loss=0.1077, over 5682415.90 frames. ], batch size: 92, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 05:58:04,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3518, 1.5955, 1.1859, 1.2993], device='cuda:0'), covar=tensor([0.1250, 0.1006, 0.0977, 0.1019], device='cuda:0'), in_proj_covar=tensor([0.1482, 0.1305, 0.1247, 0.1371], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 05:58:25,843 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 05:58:47,820 INFO [train.py:968] (0/2) Epoch 6, batch 19550, giga_loss[loss=0.2596, simple_loss=0.3384, pruned_loss=0.09042, over 28760.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.348, pruned_loss=0.1057, over 5657115.25 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3604, pruned_loss=0.1069, over 5757143.08 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3473, pruned_loss=0.1061, over 5645615.92 frames. ], batch size: 60, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 05:58:59,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6669, 2.4897, 1.8958, 2.0492], device='cuda:0'), covar=tensor([0.0531, 0.0560, 0.0781, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0435, 0.0490, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 05:59:16,821 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-03 05:59:23,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2679, 4.1054, 3.9045, 1.8951], device='cuda:0'), covar=tensor([0.0427, 0.0489, 0.0595, 0.2089], device='cuda:0'), in_proj_covar=tensor([0.0855, 0.0801, 0.0758, 0.0601], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 05:59:29,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.932e+02 9.955e+02 1.248e+03 1.784e+03 7.926e+03, threshold=2.497e+03, percent-clipped=11.0
+2023-03-03 05:59:31,504 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-03 05:59:34,879 INFO [train.py:968] (0/2) Epoch 6, batch 19600, giga_loss[loss=0.2748, simple_loss=0.3451, pruned_loss=0.1022, over 28977.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3488, pruned_loss=0.1062, over 5665544.47 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3605, pruned_loss=0.107, over 5758666.85 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.348, pruned_loss=0.1065, over 5654192.96 frames. ], batch size: 186, lr: 5.38e-03, grad_scale: 8.0
+2023-03-03 05:59:46,714 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:00:15,956 INFO [train.py:968] (0/2) Epoch 6, batch 19650, giga_loss[loss=0.2613, simple_loss=0.3296, pruned_loss=0.09648, over 28464.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3495, pruned_loss=0.1064, over 5659528.47 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3618, pruned_loss=0.1076, over 5741462.92 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3472, pruned_loss=0.106, over 5662280.42 frames. ], batch size: 85, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:00:56,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.689e+02 1.081e+03 1.322e+03 1.919e+03 9.934e+03, threshold=2.645e+03, percent-clipped=14.0
+2023-03-03 06:00:59,884 INFO [train.py:968] (0/2) Epoch 6, batch 19700, giga_loss[loss=0.2719, simple_loss=0.3409, pruned_loss=0.1015, over 28693.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.348, pruned_loss=0.1056, over 5669944.90 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3624, pruned_loss=0.1077, over 5743343.44 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3454, pruned_loss=0.1052, over 5669338.83 frames. ], batch size: 262, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:01:16,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-03 06:01:25,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=247445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:01:38,822 INFO [train.py:968] (0/2) Epoch 6, batch 19750, giga_loss[loss=0.2907, simple_loss=0.3576, pruned_loss=0.1119, over 28552.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3461, pruned_loss=0.1046, over 5681243.90 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3628, pruned_loss=0.1079, over 5746420.48 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3434, pruned_loss=0.104, over 5676774.97 frames. ], batch size: 307, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:02:01,224 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 06:02:17,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.494e+02 9.645e+02 1.148e+03 1.551e+03 4.219e+03, threshold=2.296e+03, percent-clipped=8.0
+2023-03-03 06:02:19,076 INFO [train.py:968] (0/2) Epoch 6, batch 19800, giga_loss[loss=0.2664, simple_loss=0.3334, pruned_loss=0.09971, over 28736.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3432, pruned_loss=0.1031, over 5693083.93 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3632, pruned_loss=0.108, over 5747899.44 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3406, pruned_loss=0.1025, over 5687456.40 frames. ], batch size: 242, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:03:00,167 INFO [train.py:968] (0/2) Epoch 6, batch 19850, giga_loss[loss=0.2488, simple_loss=0.3133, pruned_loss=0.09217, over 29058.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.342, pruned_loss=0.1023, over 5699655.84 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3636, pruned_loss=0.108, over 5754601.14 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3386, pruned_loss=0.1017, over 5687378.15 frames. ], batch size: 128, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:03:19,436 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:03:21,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:03:34,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.306e+02 8.984e+02 1.202e+03 2.041e+03 5.419e+03, threshold=2.404e+03, percent-clipped=19.0
+2023-03-03 06:03:37,172 INFO [train.py:968] (0/2) Epoch 6, batch 19900, giga_loss[loss=0.3222, simple_loss=0.3815, pruned_loss=0.1314, over 28613.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3389, pruned_loss=0.1008, over 5711963.79 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.364, pruned_loss=0.1081, over 5757292.12 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3354, pruned_loss=0.1, over 5699012.48 frames. ], batch size: 307, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:03:45,295 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:04:08,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9926, 1.2324, 0.9700, 0.2460], device='cuda:0'), covar=tensor([0.1438, 0.1111, 0.1845, 0.2756], device='cuda:0'), in_proj_covar=tensor([0.1413, 0.1324, 0.1389, 0.1170], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 06:04:16,680 INFO [train.py:968] (0/2) Epoch 6, batch 19950, giga_loss[loss=0.264, simple_loss=0.329, pruned_loss=0.09952, over 28708.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3377, pruned_loss=0.09979, over 5721079.42 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3649, pruned_loss=0.1083, over 5760316.41 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3331, pruned_loss=0.09871, over 5706401.06 frames. ], batch size: 119, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:04:49,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=247702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:04:52,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.008e+02 1.036e+03 1.310e+03 1.916e+03 9.350e+03, threshold=2.620e+03, percent-clipped=13.0
+2023-03-03 06:04:55,755 INFO [train.py:968] (0/2) Epoch 6, batch 20000, giga_loss[loss=0.2521, simple_loss=0.3213, pruned_loss=0.09148, over 28703.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3349, pruned_loss=0.09838, over 5722109.72 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3654, pruned_loss=0.1084, over 5760297.69 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3305, pruned_loss=0.09728, over 5709989.85 frames. ], batch size: 262, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:05:12,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0172, 1.2608, 1.2836, 1.2570], device='cuda:0'), covar=tensor([0.1272, 0.1125, 0.1757, 0.1192], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0739, 0.0647, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 06:05:34,967 INFO [train.py:968] (0/2) Epoch 6, batch 20050, giga_loss[loss=0.2432, simple_loss=0.3191, pruned_loss=0.08362, over 28719.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3328, pruned_loss=0.09711, over 5714401.68 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3657, pruned_loss=0.1084, over 5752508.48 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3287, pruned_loss=0.09609, over 5710559.56 frames. ], batch size: 66, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:05:40,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4461, 1.0491, 5.2482, 3.6564], device='cuda:0'), covar=tensor([0.1392, 0.2403, 0.0281, 0.0653], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0541, 0.0749, 0.0627], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 06:05:45,738 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:06:11,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.262e+02 9.329e+02 1.240e+03 2.084e+03 1.081e+04, threshold=2.479e+03, percent-clipped=17.0
+2023-03-03 06:06:13,856 INFO [train.py:968] (0/2) Epoch 6, batch 20100, giga_loss[loss=0.2566, simple_loss=0.3219, pruned_loss=0.09567, over 28611.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3328, pruned_loss=0.0966, over 5721387.21 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3667, pruned_loss=0.1089, over 5756562.81 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3277, pruned_loss=0.09505, over 5713662.91 frames. ], batch size: 85, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:06:20,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:06:26,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2255, 1.5110, 1.2298, 1.4992], device='cuda:0'), covar=tensor([0.0780, 0.0342, 0.0326, 0.0788], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0123, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0062, 0.0045, 0.0040, 0.0068], device='cuda:0')
+2023-03-03 06:06:38,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2467, 3.1025, 2.9196, 1.3007], device='cuda:0'), covar=tensor([0.0797, 0.0768, 0.0914, 0.2384], device='cuda:0'), in_proj_covar=tensor([0.0872, 0.0810, 0.0765, 0.0599], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 06:06:40,906 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:06:44,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:06:53,174 INFO [train.py:968] (0/2) Epoch 6, batch 20150, libri_loss[loss=0.3245, simple_loss=0.4024, pruned_loss=0.1233, over 29474.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3385, pruned_loss=0.1001, over 5715677.11 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3672, pruned_loss=0.109, over 5753837.22 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3324, pruned_loss=0.09831, over 5710558.29 frames. ], batch size: 85, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:07:10,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2130, 1.1927, 4.6006, 3.3967], device='cuda:0'), covar=tensor([0.1693, 0.2503, 0.0315, 0.0765], device='cuda:0'), in_proj_covar=tensor([0.0578, 0.0538, 0.0748, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 06:07:10,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:07:23,089 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:07:32,757 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-03 06:07:37,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.408e+02 1.040e+03 1.350e+03 1.923e+03 4.180e+03, threshold=2.700e+03, percent-clipped=8.0
+2023-03-03 06:07:41,431 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 06:07:41,604 INFO [train.py:968] (0/2) Epoch 6, batch 20200, giga_loss[loss=0.2969, simple_loss=0.3625, pruned_loss=0.1156, over 28820.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3443, pruned_loss=0.1045, over 5706900.43 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3672, pruned_loss=0.109, over 5754419.03 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3394, pruned_loss=0.103, over 5702223.11 frames. ], batch size: 199, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:07:50,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:08:17,029 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:08:31,691 INFO [train.py:968] (0/2) Epoch 6, batch 20250, giga_loss[loss=0.3209, simple_loss=0.3807, pruned_loss=0.1305, over 28810.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3526, pruned_loss=0.1103, over 5701504.01 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3673, pruned_loss=0.109, over 5756692.09 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3482, pruned_loss=0.109, over 5694541.14 frames. ], batch size: 119, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:09:10,750 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-248000.pt
+2023-03-03 06:09:18,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.572e+02 1.280e+03 1.680e+03 2.592e+03 7.968e+03, threshold=3.359e+03, percent-clipped=24.0
+2023-03-03 06:09:21,453 INFO [train.py:968] (0/2) Epoch 6, batch 20300, giga_loss[loss=0.4252, simple_loss=0.453, pruned_loss=0.1987, over 26666.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.359, pruned_loss=0.1142, over 5676621.77 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3676, pruned_loss=0.1092, over 5739380.23 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3552, pruned_loss=0.1131, over 5686899.66 frames. ], batch size: 555, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:10:05,699 INFO [train.py:968] (0/2) Epoch 6, batch 20350, giga_loss[loss=0.327, simple_loss=0.3935, pruned_loss=0.1303, over 28954.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3641, pruned_loss=0.1163, over 5680522.52 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3676, pruned_loss=0.1094, over 5740089.64 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3608, pruned_loss=0.1154, over 5686196.59 frames. ], batch size: 145, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:10:48,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.754e+02 1.082e+03 1.440e+03 1.940e+03 5.349e+03, threshold=2.880e+03, percent-clipped=7.0
+2023-03-03 06:10:51,096 INFO [train.py:968] (0/2) Epoch 6, batch 20400, giga_loss[loss=0.2963, simple_loss=0.3685, pruned_loss=0.1121, over 28796.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3681, pruned_loss=0.1176, over 5685197.37 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3675, pruned_loss=0.1093, over 5743310.07 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3656, pruned_loss=0.1171, over 5685662.70 frames. ], batch size: 99, lr: 5.37e-03, grad_scale: 8.0
+2023-03-03 06:11:25,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:11:35,096 INFO [train.py:968] (0/2) Epoch 6, batch 20450, giga_loss[loss=0.3365, simple_loss=0.3964, pruned_loss=0.1383, over 28940.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3736, pruned_loss=0.121, over 5693315.65 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3679, pruned_loss=0.1096, over 5743110.24 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3713, pruned_loss=0.1205, over 5692752.10 frames. ], batch size: 227, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:11:50,187 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:12:06,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:12:19,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.359e+02 1.166e+03 1.406e+03 1.969e+03 4.255e+03, threshold=2.812e+03, percent-clipped=5.0
+2023-03-03 06:12:21,155 INFO [train.py:968] (0/2) Epoch 6, batch 20500, giga_loss[loss=0.221, simple_loss=0.305, pruned_loss=0.06846, over 28802.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3679, pruned_loss=0.1174, over 5686177.37 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3683, pruned_loss=0.11, over 5744738.29 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3657, pruned_loss=0.1168, over 5683670.50 frames. ], batch size: 186, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:13:02,168 INFO [train.py:968] (0/2) Epoch 6, batch 20550, giga_loss[loss=0.2818, simple_loss=0.3534, pruned_loss=0.105, over 28540.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3638, pruned_loss=0.1136, over 5685725.09 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3683, pruned_loss=0.1101, over 5730458.22 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3621, pruned_loss=0.1132, over 5693975.14 frames. ], batch size: 307, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:13:08,009 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:13:16,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3744, 1.6201, 1.4121, 1.6128], device='cuda:0'), covar=tensor([0.0924, 0.0966, 0.1513, 0.0898], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0736, 0.0639, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 06:13:26,781 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:13:29,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:13:29,201 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:13:43,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.479e+02 1.133e+03 1.423e+03 1.805e+03 3.727e+03, threshold=2.845e+03, percent-clipped=5.0
+2023-03-03 06:13:45,146 INFO [train.py:968] (0/2) Epoch 6, batch 20600, giga_loss[loss=0.2577, simple_loss=0.3348, pruned_loss=0.09029, over 28548.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3631, pruned_loss=0.1128, over 5676188.53 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3687, pruned_loss=0.1103, over 5729913.53 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3612, pruned_loss=0.1123, over 5682277.95 frames. ], batch size: 60, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:13:54,619 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:13:54,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:14:08,514 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:14:11,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:14:29,115 INFO [train.py:968] (0/2) Epoch 6, batch 20650, giga_loss[loss=0.3324, simple_loss=0.3835, pruned_loss=0.1406, over 26594.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3635, pruned_loss=0.1122, over 5683022.91 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3686, pruned_loss=0.1103, over 5730038.91 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.362, pruned_loss=0.1118, over 5687237.02 frames. ], batch size: 555, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:14:34,714 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:14:42,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:14:52,271 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:10,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.671e+02 1.098e+03 1.436e+03 1.822e+03 3.329e+03, threshold=2.871e+03, percent-clipped=2.0
+2023-03-03 06:15:12,176 INFO [train.py:968] (0/2) Epoch 6, batch 20700, giga_loss[loss=0.294, simple_loss=0.3632, pruned_loss=0.1124, over 28953.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3667, pruned_loss=0.1145, over 5683879.59 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3688, pruned_loss=0.1104, over 5730166.09 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3653, pruned_loss=0.1142, over 5686291.25 frames. ], batch size: 213, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:15:12,511 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:14,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:36,616 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:39,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:41,792 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:55,891 INFO [train.py:968] (0/2) Epoch 6, batch 20750, giga_loss[loss=0.3381, simple_loss=0.4108, pruned_loss=0.1327, over 28970.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3683, pruned_loss=0.1161, over 5687213.49 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3687, pruned_loss=0.1103, over 5728582.11 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3673, pruned_loss=0.116, over 5689664.43 frames. ], batch size: 174, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:15:58,938 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:16:03,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:16:03,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:16:11,069 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:16:31,742 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:16:42,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.538e+02 1.168e+03 1.511e+03 2.095e+03 6.834e+03, threshold=3.021e+03, percent-clipped=11.0
+2023-03-03 06:16:43,924 INFO [train.py:968] (0/2) Epoch 6, batch 20800, giga_loss[loss=0.3182, simple_loss=0.3843, pruned_loss=0.1261, over 28056.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3691, pruned_loss=0.1169, over 5700074.78 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.369, pruned_loss=0.1106, over 5729513.20 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.368, pruned_loss=0.1165, over 5700920.57 frames. ], batch size: 412, lr: 5.37e-03, grad_scale: 8.0
+2023-03-03 06:16:55,448 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6214, 1.4749, 1.4899, 1.4929], device='cuda:0'), covar=tensor([0.0921, 0.1438, 0.1416, 0.1225], device='cuda:0'), in_proj_covar=tensor([0.0437, 0.0744, 0.0648, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 06:17:19,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6244, 4.3973, 4.2924, 1.9987], device='cuda:0'), covar=tensor([0.0511, 0.0709, 0.0854, 0.1926], device='cuda:0'), in_proj_covar=tensor([0.0875, 0.0825, 0.0771, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 06:17:19,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:17:26,676 INFO [train.py:968] (0/2) Epoch 6, batch 20850, giga_loss[loss=0.3061, simple_loss=0.3699, pruned_loss=0.1211, over 28487.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3704, pruned_loss=0.1183, over 5694455.81 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3695, pruned_loss=0.111, over 5724630.22 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3691, pruned_loss=0.1179, over 5698468.85 frames. ], batch size: 65, lr: 5.37e-03, grad_scale: 8.0
+2023-03-03 06:18:06,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.500e+02 1.184e+03 1.513e+03 1.969e+03 6.035e+03, threshold=3.026e+03, percent-clipped=12.0
+2023-03-03 06:18:08,248 INFO [train.py:968] (0/2) Epoch 6, batch 20900, giga_loss[loss=0.3096, simple_loss=0.3783, pruned_loss=0.1205, over 28703.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3707, pruned_loss=0.118, over 5700762.78 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3697, pruned_loss=0.1111, over 5723701.63 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3695, pruned_loss=0.1177, over 5704483.02 frames. ], batch size: 284, lr: 5.37e-03, grad_scale: 8.0
+2023-03-03 06:18:47,752 INFO [train.py:968] (0/2) Epoch 6, batch 20950, giga_loss[loss=0.2701, simple_loss=0.3557, pruned_loss=0.09229, over 28969.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3702, pruned_loss=0.1166, over 5693518.42 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3699, pruned_loss=0.1114, over 5717627.76 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.369, pruned_loss=0.1163, over 5701770.64 frames. ], batch size: 136, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:19:15,801 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:19:17,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:19:29,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.453e+02 1.008e+03 1.215e+03 1.692e+03 8.234e+03, threshold=2.431e+03, percent-clipped=5.0
+2023-03-03 06:19:31,785 INFO [train.py:968] (0/2) Epoch 6, batch 21000, giga_loss[loss=0.3408, simple_loss=0.3913, pruned_loss=0.1452, over 27640.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3715, pruned_loss=0.1163, over 5701049.83 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.37, pruned_loss=0.1115, over 5718620.51 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3705, pruned_loss=0.116, over 5706549.44 frames. ], batch size: 472, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:19:31,790 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 06:19:39,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3046, 1.3956, 1.0878, 1.4439], device='cuda:0'), covar=tensor([0.0788, 0.0298, 0.0365, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0122, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0062, 0.0045, 0.0040, 0.0068], device='cuda:0')
+2023-03-03 06:19:40,437 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2408, simple_loss=0.3442, pruned_loss=0.06874, over 944034.00 frames. 
+2023-03-03 06:19:40,438 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 06:19:52,028 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:19:53,349 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:20:13,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:20:21,375 INFO [train.py:968] (0/2) Epoch 6, batch 21050, giga_loss[loss=0.2713, simple_loss=0.3608, pruned_loss=0.09092, over 29012.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3711, pruned_loss=0.116, over 5713007.41 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.37, pruned_loss=0.1117, over 5722954.30 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3703, pruned_loss=0.1157, over 5713203.41 frames. ], batch size: 136, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:20:23,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:20:24,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6168, 1.5494, 1.1647, 1.1736], device='cuda:0'), covar=tensor([0.0647, 0.0547, 0.0936, 0.0912], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0438, 0.0497, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 06:20:33,126 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:20:48,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5028, 1.7477, 0.9272, 1.3124], device='cuda:0'), covar=tensor([0.0949, 0.0693, 0.1672, 0.1156], device='cuda:0'), in_proj_covar=tensor([0.0333, 0.0435, 0.0494, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 06:20:58,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.208e+02 1.107e+03 1.351e+03 1.984e+03 8.008e+03, threshold=2.702e+03, percent-clipped=12.0
+2023-03-03 06:20:59,462 INFO [train.py:968] (0/2) Epoch 6, batch 21100, giga_loss[loss=0.2863, simple_loss=0.3527, pruned_loss=0.11, over 28629.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3683, pruned_loss=0.1148, over 5700342.43 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3707, pruned_loss=0.1123, over 5716539.75 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.367, pruned_loss=0.1141, over 5706159.88 frames. ], batch size: 71, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:21:01,609 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-03 06:21:30,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:21:39,735 INFO [train.py:968] (0/2) Epoch 6, batch 21150, giga_loss[loss=0.3105, simple_loss=0.3751, pruned_loss=0.1229, over 29120.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3666, pruned_loss=0.1139, over 5704619.55 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.371, pruned_loss=0.1124, over 5720993.89 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3653, pruned_loss=0.1132, over 5704937.34 frames. ], batch size: 128, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:22:06,421 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:08,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:11,054 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7643, 1.0825, 3.3808, 2.7974], device='cuda:0'), covar=tensor([0.1748, 0.2292, 0.0433, 0.0824], device='cuda:0'), in_proj_covar=tensor([0.0574, 0.0529, 0.0743, 0.0625], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 06:22:14,867 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:17,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.721e+02 1.024e+03 1.298e+03 1.844e+03 4.781e+03, threshold=2.596e+03, percent-clipped=3.0
+2023-03-03 06:22:17,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:18,222 INFO [train.py:968] (0/2) Epoch 6, batch 21200, giga_loss[loss=0.2883, simple_loss=0.3634, pruned_loss=0.1066, over 28605.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3656, pruned_loss=0.1135, over 5713681.71 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3711, pruned_loss=0.1127, over 5723193.11 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3643, pruned_loss=0.1127, over 5711513.32 frames. ], batch size: 71, lr: 5.37e-03, grad_scale: 8.0
+2023-03-03 06:22:32,628 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:39,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6658, 2.2901, 2.2454, 2.1831], device='cuda:0'), covar=tensor([0.0931, 0.1668, 0.1379, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0732, 0.0637, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 06:22:42,383 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:58,428 INFO [train.py:968] (0/2) Epoch 6, batch 21250, giga_loss[loss=0.3011, simple_loss=0.3712, pruned_loss=0.1155, over 28830.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3671, pruned_loss=0.1151, over 5720214.01 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3717, pruned_loss=0.1134, over 5731204.02 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3653, pruned_loss=0.1139, over 5710584.69 frames. ], batch size: 199, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:23:24,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:23:26,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:23:30,660 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-03 06:23:38,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.966e+02 1.029e+03 1.297e+03 2.023e+03 9.701e+03, threshold=2.594e+03, percent-clipped=16.0
+2023-03-03 06:23:38,298 INFO [train.py:968] (0/2) Epoch 6, batch 21300, giga_loss[loss=0.2747, simple_loss=0.3532, pruned_loss=0.09806, over 28910.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3674, pruned_loss=0.1151, over 5718042.81 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3719, pruned_loss=0.1139, over 5735826.43 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3656, pruned_loss=0.1138, over 5705777.05 frames. ], batch size: 106, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:23:49,158 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:24:08,185 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-03 06:24:12,769 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 06:24:16,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6317, 2.2474, 1.7582, 0.7140], device='cuda:0'), covar=tensor([0.2864, 0.1314, 0.2024, 0.3092], device='cuda:0'), in_proj_covar=tensor([0.1411, 0.1299, 0.1367, 0.1157], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 06:24:17,881 INFO [train.py:968] (0/2) Epoch 6, batch 21350, giga_loss[loss=0.2908, simple_loss=0.3701, pruned_loss=0.1058, over 29079.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3671, pruned_loss=0.1141, over 5722564.23 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3722, pruned_loss=0.1142, over 5738241.31 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3653, pruned_loss=0.1127, over 5710368.90 frames. ], batch size: 155, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:24:23,069 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 06:24:30,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8308, 1.7597, 1.3009, 1.4455], device='cuda:0'), covar=tensor([0.0697, 0.0604, 0.0997, 0.0961], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0437, 0.0498, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 06:24:38,481 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=249084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:24:51,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:24:56,765 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=249108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:24:58,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.248e+02 9.987e+02 1.239e+03 1.557e+03 5.380e+03, threshold=2.478e+03, percent-clipped=7.0
+2023-03-03 06:24:58,951 INFO [train.py:968] (0/2) Epoch 6, batch 21400, giga_loss[loss=0.2969, simple_loss=0.3696, pruned_loss=0.1121, over 28269.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3664, pruned_loss=0.1135, over 5717389.92 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3726, pruned_loss=0.1148, over 5741881.57 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3644, pruned_loss=0.1118, over 5703893.78 frames. ], batch size: 368, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:25:32,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:25:39,416 INFO [train.py:968] (0/2) Epoch 6, batch 21450, giga_loss[loss=0.2726, simple_loss=0.3467, pruned_loss=0.09921, over 28483.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3678, pruned_loss=0.1155, over 5717070.29 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3744, pruned_loss=0.1166, over 5746145.32 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3641, pruned_loss=0.1124, over 5700590.87 frames. ], batch size: 71, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:25:49,565 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-03 06:26:18,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.406e+02 1.091e+03 1.310e+03 1.880e+03 4.488e+03, threshold=2.620e+03, percent-clipped=11.0
+2023-03-03 06:26:18,512 INFO [train.py:968] (0/2) Epoch 6, batch 21500, giga_loss[loss=0.3229, simple_loss=0.3878, pruned_loss=0.129, over 27960.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3658, pruned_loss=0.1146, over 5715308.79 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3743, pruned_loss=0.1168, over 5749170.52 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3626, pruned_loss=0.1118, over 5698389.94 frames. ], batch size: 412, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:26:43,341 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=249243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:26:46,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:26:57,343 INFO [train.py:968] (0/2) Epoch 6, batch 21550, giga_loss[loss=0.2703, simple_loss=0.3448, pruned_loss=0.09791, over 28429.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3627, pruned_loss=0.1133, over 5703487.17 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3749, pruned_loss=0.1174, over 5742585.78 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3594, pruned_loss=0.1104, over 5695744.64 frames. ], batch size: 60, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:27:08,067 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:27:21,719 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=249294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:27:24,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:27:36,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.394e+02 1.145e+03 1.533e+03 2.073e+03 7.126e+03, threshold=3.066e+03, percent-clipped=16.0
+2023-03-03 06:27:36,476 INFO [train.py:968] (0/2) Epoch 6, batch 21600, giga_loss[loss=0.2959, simple_loss=0.3715, pruned_loss=0.1101, over 28619.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3622, pruned_loss=0.1135, over 5693481.88 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3758, pruned_loss=0.1183, over 5739976.42 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3581, pruned_loss=0.1102, over 5687989.51 frames. ], batch size: 307, lr: 5.36e-03, grad_scale: 8.0
+2023-03-03 06:27:47,345 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:28:00,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9868, 1.1675, 3.5165, 3.0021], device='cuda:0'), covar=tensor([0.1586, 0.2310, 0.0377, 0.1267], device='cuda:0'), in_proj_covar=tensor([0.0572, 0.0527, 0.0743, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 06:28:15,275 INFO [train.py:968] (0/2) Epoch 6, batch 21650, giga_loss[loss=0.2613, simple_loss=0.3318, pruned_loss=0.09539, over 28474.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3617, pruned_loss=0.1137, over 5701286.50 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3752, pruned_loss=0.1183, over 5744689.36 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3585, pruned_loss=0.1108, over 5690832.45 frames. ], batch size: 78, lr: 5.36e-03, grad_scale: 8.0
+2023-03-03 06:28:54,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.863e+02 1.020e+03 1.258e+03 1.719e+03 4.476e+03, threshold=2.517e+03, percent-clipped=5.0
+2023-03-03 06:28:54,919 INFO [train.py:968] (0/2) Epoch 6, batch 21700, libri_loss[loss=0.3679, simple_loss=0.3981, pruned_loss=0.1689, over 29354.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3595, pruned_loss=0.1129, over 5706522.31 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3762, pruned_loss=0.1194, over 5747829.07 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3557, pruned_loss=0.1095, over 5694439.38 frames. ], batch size: 67, lr: 5.36e-03, grad_scale: 8.0
+2023-03-03 06:28:57,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-03 06:28:57,460 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-03 06:29:31,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:29:32,877 INFO [train.py:968] (0/2) Epoch 6, batch 21750, giga_loss[loss=0.3487, simple_loss=0.3889, pruned_loss=0.1543, over 23708.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3579, pruned_loss=0.1123, over 5703359.08 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.377, pruned_loss=0.1203, over 5745104.50 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3532, pruned_loss=0.1083, over 5694807.15 frames. ], batch size: 705, lr: 5.36e-03, grad_scale: 2.0
+2023-03-03 06:29:48,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:30:10,448 INFO [train.py:968] (0/2) Epoch 6, batch 21800, giga_loss[loss=0.3134, simple_loss=0.3655, pruned_loss=0.1307, over 26701.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3559, pruned_loss=0.1114, over 5702775.94 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3777, pruned_loss=0.1208, over 5741223.24 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3508, pruned_loss=0.1074, over 5698781.33 frames. ], batch size: 555, lr: 5.36e-03, grad_scale: 2.0
+2023-03-03 06:30:11,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.765e+02 1.097e+03 1.534e+03 2.602e+03 1.381e+04, threshold=3.068e+03, percent-clipped=27.0
+2023-03-03 06:30:29,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5584, 3.4955, 1.6256, 1.5726], device='cuda:0'), covar=tensor([0.0781, 0.0309, 0.0815, 0.1239], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0477, 0.0308, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 06:30:48,719 INFO [train.py:968] (0/2) Epoch 6, batch 21850, giga_loss[loss=0.2501, simple_loss=0.3227, pruned_loss=0.08869, over 28892.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3523, pruned_loss=0.1091, over 5713709.22 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3777, pruned_loss=0.1209, over 5743738.88 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3479, pruned_loss=0.1057, over 5707822.36 frames. ], batch size: 186, lr: 5.36e-03, grad_scale: 2.0
+2023-03-03 06:31:24,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=249602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:31:27,923 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:31:30,236 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=249608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:31:32,027 INFO [train.py:968] (0/2) Epoch 6, batch 21900, giga_loss[loss=0.2851, simple_loss=0.3609, pruned_loss=0.1046, over 29009.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3521, pruned_loss=0.1086, over 5710065.83 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3779, pruned_loss=0.121, over 5744680.18 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3484, pruned_loss=0.1058, over 5704590.09 frames. ], batch size: 155, lr: 5.36e-03, grad_scale: 2.0
+2023-03-03 06:31:33,365 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.306e+02 9.146e+02 1.181e+03 1.603e+03 2.970e+03, threshold=2.363e+03, percent-clipped=0.0
+2023-03-03 06:31:45,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=249626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:31:47,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:31:51,396 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:32:11,188 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:32:12,970 INFO [train.py:968] (0/2) Epoch 6, batch 21950, giga_loss[loss=0.2635, simple_loss=0.3338, pruned_loss=0.09656, over 28737.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3546, pruned_loss=0.1099, over 5712518.26 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3777, pruned_loss=0.1215, over 5747978.60 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.351, pruned_loss=0.1069, over 5704033.45 frames. ], batch size: 99, lr: 5.36e-03, grad_scale: 2.0
+2023-03-03 06:32:52,890 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 06:32:59,092 INFO [train.py:968] (0/2) Epoch 6, batch 22000, giga_loss[loss=0.2887, simple_loss=0.3662, pruned_loss=0.1056, over 28549.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3584, pruned_loss=0.1114, over 5697876.82 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3777, pruned_loss=0.1216, over 5750382.76 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3553, pruned_loss=0.1086, over 5688434.03 frames. ], batch size: 336, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:33:00,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.379e+02 9.278e+02 1.142e+03 1.733e+03 5.956e+03, threshold=2.284e+03, percent-clipped=10.0
+2023-03-03 06:33:17,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3938, 2.1456, 1.5458, 0.6660], device='cuda:0'), covar=tensor([0.3233, 0.1525, 0.2354, 0.3662], device='cuda:0'), in_proj_covar=tensor([0.1425, 0.1316, 0.1384, 0.1170], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 06:33:39,061 INFO [train.py:968] (0/2) Epoch 6, batch 22050, giga_loss[loss=0.342, simple_loss=0.3975, pruned_loss=0.1432, over 27546.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3598, pruned_loss=0.1113, over 5709697.34 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3782, pruned_loss=0.1223, over 5754413.75 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3562, pruned_loss=0.1082, over 5696963.59 frames. ], batch size: 472, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:34:13,681 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5249, 2.2129, 1.5354, 0.6236], device='cuda:0'), covar=tensor([0.3318, 0.1467, 0.2360, 0.3660], device='cuda:0'), in_proj_covar=tensor([0.1431, 0.1320, 0.1387, 0.1169], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 06:34:21,830 INFO [train.py:968] (0/2) Epoch 6, batch 22100, giga_loss[loss=0.2604, simple_loss=0.3401, pruned_loss=0.09035, over 28859.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3588, pruned_loss=0.11, over 5713248.96 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3783, pruned_loss=0.1225, over 5755989.68 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3557, pruned_loss=0.1072, over 5701353.81 frames. ], batch size: 199, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:34:23,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.810e+02 9.448e+02 1.199e+03 1.908e+03 7.789e+03, threshold=2.399e+03, percent-clipped=13.0
+2023-03-03 06:35:06,181 INFO [train.py:968] (0/2) Epoch 6, batch 22150, giga_loss[loss=0.3202, simple_loss=0.3912, pruned_loss=0.1246, over 27576.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3592, pruned_loss=0.1105, over 5708289.84 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3787, pruned_loss=0.1229, over 5758459.36 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3562, pruned_loss=0.1078, over 5696207.02 frames. ], batch size: 472, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:35:44,891 INFO [train.py:968] (0/2) Epoch 6, batch 22200, giga_loss[loss=0.2971, simple_loss=0.3558, pruned_loss=0.1192, over 28634.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.361, pruned_loss=0.1123, over 5708186.95 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3785, pruned_loss=0.123, over 5760406.98 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3582, pruned_loss=0.1097, over 5695019.59 frames. ], batch size: 78, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:35:45,165 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0204, 4.7895, 2.0045, 1.8832], device='cuda:0'), covar=tensor([0.0759, 0.0304, 0.0756, 0.1154], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0481, 0.0309, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 06:35:46,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.776e+02 1.094e+03 1.463e+03 2.132e+03 9.892e+03, threshold=2.926e+03, percent-clipped=21.0
+2023-03-03 06:36:25,296 INFO [train.py:968] (0/2) Epoch 6, batch 22250, giga_loss[loss=0.3679, simple_loss=0.4084, pruned_loss=0.1638, over 26667.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.361, pruned_loss=0.1125, over 5702625.26 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3783, pruned_loss=0.123, over 5753820.07 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3584, pruned_loss=0.11, over 5695882.97 frames. ], batch size: 555, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:36:45,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:36:57,773 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-250000.pt
+2023-03-03 06:37:07,217 INFO [train.py:968] (0/2) Epoch 6, batch 22300, giga_loss[loss=0.2629, simple_loss=0.3428, pruned_loss=0.09148, over 28831.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.364, pruned_loss=0.1143, over 5707411.80 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3786, pruned_loss=0.1234, over 5754465.72 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3615, pruned_loss=0.112, over 5700816.49 frames. ], batch size: 186, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:37:08,492 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.838e+02 1.272e+03 1.595e+03 2.101e+03 4.114e+03, threshold=3.189e+03, percent-clipped=10.0
+2023-03-03 06:37:31,055 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-03 06:37:45,602 INFO [train.py:968] (0/2) Epoch 6, batch 22350, libri_loss[loss=0.2918, simple_loss=0.3445, pruned_loss=0.1196, over 29364.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3677, pruned_loss=0.1165, over 5717488.21 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3799, pruned_loss=0.1245, over 5759364.07 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.364, pruned_loss=0.1132, over 5706055.54 frames. ], batch size: 67, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:38:26,702 INFO [train.py:968] (0/2) Epoch 6, batch 22400, giga_loss[loss=0.2526, simple_loss=0.3339, pruned_loss=0.0856, over 28502.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3678, pruned_loss=0.1158, over 5720624.52 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3799, pruned_loss=0.1245, over 5761587.72 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3646, pruned_loss=0.113, over 5708303.10 frames. ], batch size: 71, lr: 5.35e-03, grad_scale: 8.0
+2023-03-03 06:38:28,344 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.689e+02 1.128e+03 1.581e+03 2.181e+03 4.803e+03, threshold=3.163e+03, percent-clipped=7.0
+2023-03-03 06:38:38,175 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:38:40,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:39:02,886 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=250158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:39:05,253 INFO [train.py:968] (0/2) Epoch 6, batch 22450, giga_loss[loss=0.2815, simple_loss=0.3602, pruned_loss=0.1014, over 29045.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3677, pruned_loss=0.1153, over 5719358.79 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3799, pruned_loss=0.1246, over 5754481.50 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3649, pruned_loss=0.1129, over 5714294.20 frames. ], batch size: 155, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:39:10,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4218, 1.4551, 1.2709, 1.7879], device='cuda:0'), covar=tensor([0.1913, 0.1982, 0.2088, 0.1984], device='cuda:0'), in_proj_covar=tensor([0.1132, 0.0868, 0.1002, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 06:39:18,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8451, 0.8773, 3.6293, 2.9986], device='cuda:0'), covar=tensor([0.1749, 0.2615, 0.0418, 0.0807], device='cuda:0'), in_proj_covar=tensor([0.0587, 0.0540, 0.0767, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 06:39:35,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4169, 1.4455, 1.4093, 1.4701], device='cuda:0'), covar=tensor([0.0918, 0.1548, 0.1478, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0746, 0.0649, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 06:39:43,859 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=250207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:39:47,798 INFO [train.py:968] (0/2) Epoch 6, batch 22500, giga_loss[loss=0.3048, simple_loss=0.3746, pruned_loss=0.1175, over 28878.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.369, pruned_loss=0.1166, over 5714153.09 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3805, pruned_loss=0.1251, over 5757434.97 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.366, pruned_loss=0.114, over 5706949.14 frames. ], batch size: 164, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:39:50,321 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.011e+02 1.212e+03 1.614e+03 2.221e+03 5.994e+03, threshold=3.228e+03, percent-clipped=11.0
+2023-03-03 06:40:08,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3754, 1.5946, 1.3950, 1.5565], device='cuda:0'), covar=tensor([0.1638, 0.1548, 0.1611, 0.1419], device='cuda:0'), in_proj_covar=tensor([0.1132, 0.0869, 0.1001, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:0')
+2023-03-03 06:40:28,285 INFO [train.py:968] (0/2) Epoch 6, batch 22550, giga_loss[loss=0.2875, simple_loss=0.3488, pruned_loss=0.1132, over 28716.00 frames. ], tot_loss[loss=0.3, simple_loss=0.368, pruned_loss=0.116, over 5712927.98 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3812, pruned_loss=0.1256, over 5750586.50 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3648, pruned_loss=0.1133, over 5712564.29 frames. ], batch size: 92, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:40:33,278 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-03 06:41:12,642 INFO [train.py:968] (0/2) Epoch 6, batch 22600, giga_loss[loss=0.3188, simple_loss=0.372, pruned_loss=0.1328, over 28898.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3663, pruned_loss=0.1151, over 5717023.57 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3818, pruned_loss=0.1261, over 5750982.29 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.363, pruned_loss=0.1124, over 5715726.61 frames. ], batch size: 199, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:41:15,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.403e+02 1.176e+03 1.334e+03 1.613e+03 3.083e+03, threshold=2.667e+03, percent-clipped=0.0
+2023-03-03 06:41:15,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0055, 1.1516, 3.3517, 2.9223], device='cuda:0'), covar=tensor([0.2007, 0.2726, 0.0830, 0.1685], device='cuda:0'), in_proj_covar=tensor([0.0581, 0.0536, 0.0762, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 06:41:42,963 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=250348.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 06:41:51,734 INFO [train.py:968] (0/2) Epoch 6, batch 22650, giga_loss[loss=0.3063, simple_loss=0.3686, pruned_loss=0.122, over 28892.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.364, pruned_loss=0.1141, over 5713489.99 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.382, pruned_loss=0.1262, over 5743217.33 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3609, pruned_loss=0.1117, over 5718622.87 frames. ], batch size: 136, lr: 5.35e-03, grad_scale: 2.0
+2023-03-03 06:41:58,590 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=250369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:42:18,099 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-03 06:42:31,711 INFO [train.py:968] (0/2) Epoch 6, batch 22700, giga_loss[loss=0.2695, simple_loss=0.3524, pruned_loss=0.09328, over 28890.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3628, pruned_loss=0.1136, over 5714666.37 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.383, pruned_loss=0.1271, over 5745168.65 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3588, pruned_loss=0.1104, over 5715690.20 frames. ], batch size: 186, lr: 5.35e-03, grad_scale: 2.0
+2023-03-03 06:42:34,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.244e+02 1.187e+03 1.574e+03 2.529e+03 7.353e+03, threshold=3.149e+03, percent-clipped=21.0
+2023-03-03 06:43:13,735 INFO [train.py:968] (0/2) Epoch 6, batch 22750, giga_loss[loss=0.2946, simple_loss=0.3594, pruned_loss=0.1149, over 23819.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3633, pruned_loss=0.1128, over 5709597.62 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3832, pruned_loss=0.1275, over 5748539.01 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3591, pruned_loss=0.1093, over 5706509.75 frames. ], batch size: 705, lr: 5.35e-03, grad_scale: 2.0
+2023-03-03 06:43:32,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0463, 3.8654, 3.6604, 1.9585], device='cuda:0'), covar=tensor([0.0502, 0.0539, 0.0720, 0.1876], device='cuda:0'), in_proj_covar=tensor([0.0881, 0.0825, 0.0777, 0.0602], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 06:43:55,755 INFO [train.py:968] (0/2) Epoch 6, batch 22800, giga_loss[loss=0.2666, simple_loss=0.3448, pruned_loss=0.09423, over 28981.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3634, pruned_loss=0.1117, over 5712433.66 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.383, pruned_loss=0.1277, over 5747990.63 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3597, pruned_loss=0.1084, over 5709256.07 frames. ], batch size: 128, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:43:58,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.307e+02 1.073e+03 1.292e+03 1.560e+03 5.223e+03, threshold=2.585e+03, percent-clipped=2.0
+2023-03-03 06:44:17,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-03 06:44:34,002 INFO [train.py:968] (0/2) Epoch 6, batch 22850, giga_loss[loss=0.2689, simple_loss=0.3257, pruned_loss=0.106, over 28327.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3623, pruned_loss=0.1114, over 5722544.73 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3831, pruned_loss=0.1278, over 5747791.81 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3592, pruned_loss=0.1085, over 5719877.45 frames. ], batch size: 77, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:44:34,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4987, 3.0595, 1.4883, 1.4090], device='cuda:0'), covar=tensor([0.0825, 0.0412, 0.0797, 0.1280], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0488, 0.0311, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 06:44:51,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=250582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:45:16,398 INFO [train.py:968] (0/2) Epoch 6, batch 22900, giga_loss[loss=0.2665, simple_loss=0.3371, pruned_loss=0.09794, over 28979.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3616, pruned_loss=0.1124, over 5714466.88 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3834, pruned_loss=0.1282, over 5739960.38 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3586, pruned_loss=0.1096, over 5719474.55 frames. ], batch size: 155, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:45:17,474 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6661, 3.4798, 3.2640, 1.7972], device='cuda:0'), covar=tensor([0.0576, 0.0666, 0.0785, 0.2269], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0829, 0.0780, 0.0603], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 06:45:18,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2834, 1.5620, 1.2756, 1.4391], device='cuda:0'), covar=tensor([0.0698, 0.0334, 0.0331, 0.0776], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0119, 0.0122, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0061, 0.0045, 0.0041, 0.0068], device='cuda:0')
+2023-03-03 06:45:20,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.269e+02 1.083e+03 1.406e+03 1.743e+03 4.242e+03, threshold=2.811e+03, percent-clipped=8.0
+2023-03-03 06:45:23,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.1684, 5.9270, 5.6964, 2.4088], device='cuda:0'), covar=tensor([0.0330, 0.0570, 0.0665, 0.1703], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0830, 0.0780, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 06:45:58,453 INFO [train.py:968] (0/2) Epoch 6, batch 22950, giga_loss[loss=0.2405, simple_loss=0.3147, pruned_loss=0.08315, over 28966.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5713009.68 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3829, pruned_loss=0.1279, over 5742292.56 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3564, pruned_loss=0.1098, over 5714395.23 frames. ], batch size: 213, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:46:36,995 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-03 06:46:38,209 INFO [train.py:968] (0/2) Epoch 6, batch 23000, giga_loss[loss=0.2509, simple_loss=0.3275, pruned_loss=0.08722, over 28943.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.357, pruned_loss=0.1121, over 5719167.00 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3828, pruned_loss=0.128, over 5745177.64 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3547, pruned_loss=0.1099, over 5717205.84 frames. ], batch size: 174, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:46:43,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.935e+02 1.061e+03 1.292e+03 1.676e+03 3.158e+03, threshold=2.583e+03, percent-clipped=3.0
+2023-03-03 06:46:49,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=250723.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 06:46:51,294 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:46:54,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:47:05,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=250744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:47:14,588 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=250757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:47:18,787 INFO [train.py:968] (0/2) Epoch 6, batch 23050, giga_loss[loss=0.2859, simple_loss=0.3449, pruned_loss=0.1134, over 28501.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.359, pruned_loss=0.1143, over 5721504.27 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3842, pruned_loss=0.1293, over 5747408.38 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.355, pruned_loss=0.1108, over 5717058.52 frames. ], batch size: 78, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:47:56,489 INFO [train.py:968] (0/2) Epoch 6, batch 23100, giga_loss[loss=0.2335, simple_loss=0.3064, pruned_loss=0.08033, over 28892.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.355, pruned_loss=0.1124, over 5713446.35 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3841, pruned_loss=0.1295, over 5740104.15 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3514, pruned_loss=0.1091, over 5715194.93 frames. ], batch size: 174, lr: 5.35e-03, grad_scale: 2.0
+2023-03-03 06:48:01,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.732e+02 1.085e+03 1.409e+03 1.970e+03 6.931e+03, threshold=2.818e+03, percent-clipped=11.0
+2023-03-03 06:48:28,699 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1242, 1.4212, 1.1830, 0.9114], device='cuda:0'), covar=tensor([0.1295, 0.1081, 0.0723, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.1482, 0.1302, 0.1265, 0.1375], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 06:48:35,833 INFO [train.py:968] (0/2) Epoch 6, batch 23150, giga_loss[loss=0.3281, simple_loss=0.3749, pruned_loss=0.1407, over 28965.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3521, pruned_loss=0.1111, over 5724382.01 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3843, pruned_loss=0.1299, over 5746073.36 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.348, pruned_loss=0.1075, over 5719683.31 frames. ], batch size: 136, lr: 5.35e-03, grad_scale: 2.0
+2023-03-03 06:48:39,465 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250866.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 06:48:41,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250869.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 06:48:56,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:48:58,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:49:04,510 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=250898.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 06:49:12,537 INFO [train.py:968] (0/2) Epoch 6, batch 23200, giga_loss[loss=0.2674, simple_loss=0.3376, pruned_loss=0.0986, over 28884.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3488, pruned_loss=0.1092, over 5711120.14 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.3848, pruned_loss=0.1304, over 5734215.36 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3441, pruned_loss=0.1054, over 5717580.51 frames. ], batch size: 174, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:49:16,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.026e+02 1.321e+03 1.794e+03 2.305e+03 7.860e+03, threshold=3.588e+03, percent-clipped=15.0
+2023-03-03 06:49:18,135 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=250919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:49:21,918 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.03 vs. limit=2.0
+2023-03-03 06:49:40,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-03 06:49:52,079 INFO [train.py:968] (0/2) Epoch 6, batch 23250, giga_loss[loss=0.2956, simple_loss=0.3596, pruned_loss=0.1158, over 28781.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3508, pruned_loss=0.1104, over 5707304.24 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3853, pruned_loss=0.1309, over 5734007.29 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3457, pruned_loss=0.1063, over 5712315.57 frames. ], batch size: 92, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:49:58,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1620, 1.4527, 1.1864, 1.0021], device='cuda:0'), covar=tensor([0.2012, 0.1955, 0.2134, 0.1811], device='cuda:0'), in_proj_covar=tensor([0.1146, 0.0878, 0.1013, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 06:50:20,159 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-03 06:50:34,705 INFO [train.py:968] (0/2) Epoch 6, batch 23300, giga_loss[loss=0.3612, simple_loss=0.405, pruned_loss=0.1587, over 26693.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3536, pruned_loss=0.1115, over 5709875.62 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.385, pruned_loss=0.1309, over 5737061.59 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3492, pruned_loss=0.108, over 5710444.20 frames. ], batch size: 555, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:50:38,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.110e+02 1.101e+03 1.332e+03 1.665e+03 4.288e+03, threshold=2.663e+03, percent-clipped=1.0
+2023-03-03 06:51:11,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:51:17,076 INFO [train.py:968] (0/2) Epoch 6, batch 23350, giga_loss[loss=0.2933, simple_loss=0.3541, pruned_loss=0.1163, over 23806.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3567, pruned_loss=0.1127, over 5709430.20 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3848, pruned_loss=0.131, over 5739907.18 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3525, pruned_loss=0.1092, over 5707043.01 frames. ], batch size: 705, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:51:53,756 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-03 06:51:56,115 INFO [train.py:968] (0/2) Epoch 6, batch 23400, giga_loss[loss=0.2726, simple_loss=0.3507, pruned_loss=0.09726, over 29000.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3604, pruned_loss=0.1141, over 5706818.59 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3854, pruned_loss=0.1314, over 5734068.94 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.356, pruned_loss=0.1106, over 5708764.43 frames. ], batch size: 145, lr: 5.34e-03, grad_scale: 2.0
+2023-03-03 06:52:01,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.482e+02 1.082e+03 1.452e+03 2.134e+03 1.065e+04, threshold=2.904e+03, percent-clipped=15.0
+2023-03-03 06:52:22,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4860, 4.3542, 1.6412, 1.4392], device='cuda:0'), covar=tensor([0.1075, 0.0348, 0.0929, 0.1517], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0491, 0.0313, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 06:52:38,743 INFO [train.py:968] (0/2) Epoch 6, batch 23450, giga_loss[loss=0.2929, simple_loss=0.3564, pruned_loss=0.1147, over 28716.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3636, pruned_loss=0.1153, over 5715846.32 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3861, pruned_loss=0.1322, over 5735543.44 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3584, pruned_loss=0.1111, over 5715255.53 frames. ], batch size: 99, lr: 5.34e-03, grad_scale: 2.0
+2023-03-03 06:53:21,625 INFO [train.py:968] (0/2) Epoch 6, batch 23500, giga_loss[loss=0.3031, simple_loss=0.3674, pruned_loss=0.1193, over 29038.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3661, pruned_loss=0.1174, over 5715836.26 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3864, pruned_loss=0.1325, over 5737896.36 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3615, pruned_loss=0.1135, over 5713138.04 frames. ], batch size: 155, lr: 5.34e-03, grad_scale: 2.0
+2023-03-03 06:53:30,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.138e+02 1.318e+03 1.665e+03 2.276e+03 7.544e+03, threshold=3.330e+03, percent-clipped=15.0
+2023-03-03 06:53:38,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:53:48,432 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:54:14,103 INFO [train.py:968] (0/2) Epoch 6, batch 23550, giga_loss[loss=0.3195, simple_loss=0.3891, pruned_loss=0.125, over 28944.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3729, pruned_loss=0.1235, over 5703887.29 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3862, pruned_loss=0.1325, over 5736628.65 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3692, pruned_loss=0.1203, over 5702513.71 frames. ], batch size: 174, lr: 5.34e-03, grad_scale: 2.0
+2023-03-03 06:55:08,279 INFO [train.py:968] (0/2) Epoch 6, batch 23600, giga_loss[loss=0.3566, simple_loss=0.4142, pruned_loss=0.1495, over 28738.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3804, pruned_loss=0.1297, over 5684224.32 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3864, pruned_loss=0.1327, over 5735634.50 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3773, pruned_loss=0.127, over 5683875.56 frames. ], batch size: 119, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:55:12,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.371e+02 1.652e+03 2.426e+03 3.361e+03 8.120e+03, threshold=4.852e+03, percent-clipped=26.0
+2023-03-03 06:55:38,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8872, 2.6348, 1.5329, 1.3247], device='cuda:0'), covar=tensor([0.1485, 0.0747, 0.1096, 0.1340], device='cuda:0'), in_proj_covar=tensor([0.1483, 0.1299, 0.1263, 0.1371], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 06:55:40,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1771, 1.3748, 3.7571, 3.1539], device='cuda:0'), covar=tensor([0.1516, 0.2146, 0.0388, 0.0891], device='cuda:0'), in_proj_covar=tensor([0.0584, 0.0537, 0.0767, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 06:55:58,013 INFO [train.py:968] (0/2) Epoch 6, batch 23650, giga_loss[loss=0.3387, simple_loss=0.3921, pruned_loss=0.1427, over 28781.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.387, pruned_loss=0.1349, over 5682579.90 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3867, pruned_loss=0.1329, over 5737858.78 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3842, pruned_loss=0.1326, over 5679715.39 frames. ], batch size: 284, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:56:11,357 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0774, 1.7746, 1.4113, 1.6692], device='cuda:0'), covar=tensor([0.0621, 0.0674, 0.0894, 0.0890], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0453, 0.0502, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 06:56:22,106 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 06:56:46,925 INFO [train.py:968] (0/2) Epoch 6, batch 23700, giga_loss[loss=0.4087, simple_loss=0.4481, pruned_loss=0.1847, over 28896.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3942, pruned_loss=0.1415, over 5681405.84 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.387, pruned_loss=0.1331, over 5739700.72 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3918, pruned_loss=0.1396, over 5676401.11 frames. ], batch size: 213, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:56:53,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.665e+02 1.760e+03 2.283e+03 2.999e+03 7.644e+03, threshold=4.565e+03, percent-clipped=5.0
+2023-03-03 06:57:04,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:57:34,934 INFO [train.py:968] (0/2) Epoch 6, batch 23750, giga_loss[loss=0.3632, simple_loss=0.4068, pruned_loss=0.1598, over 28832.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.4005, pruned_loss=0.1466, over 5675018.64 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3876, pruned_loss=0.1337, over 5731108.90 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3982, pruned_loss=0.1448, over 5678332.67 frames. ], batch size: 99, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:57:38,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1687, 1.3749, 1.0340, 0.9470], device='cuda:0'), covar=tensor([0.0943, 0.0848, 0.0637, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.1497, 0.1314, 0.1278, 0.1392], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 06:58:24,280 INFO [train.py:968] (0/2) Epoch 6, batch 23800, giga_loss[loss=0.4583, simple_loss=0.4565, pruned_loss=0.2301, over 23492.00 frames. ], tot_loss[loss=0.3502, simple_loss=0.4023, pruned_loss=0.1491, over 5661005.25 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3875, pruned_loss=0.1336, over 5723985.65 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.4008, pruned_loss=0.148, over 5669238.87 frames. ], batch size: 705, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:58:29,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.252e+02 1.797e+03 2.462e+03 3.336e+03 1.054e+04, threshold=4.923e+03, percent-clipped=11.0
+2023-03-03 06:59:00,289 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-03 06:59:12,398 INFO [train.py:968] (0/2) Epoch 6, batch 23850, giga_loss[loss=0.4947, simple_loss=0.4771, pruned_loss=0.2562, over 23336.00 frames. ], tot_loss[loss=0.3587, simple_loss=0.4072, pruned_loss=0.1551, over 5650967.73 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3881, pruned_loss=0.1343, over 5727267.29 frames. ], giga_tot_loss[loss=0.357, simple_loss=0.4059, pruned_loss=0.154, over 5652498.33 frames. ], batch size: 705, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:59:25,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:59:28,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:59:43,590 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:59:53,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:59:58,478 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:00:03,064 INFO [train.py:968] (0/2) Epoch 6, batch 23900, giga_loss[loss=0.3312, simple_loss=0.3834, pruned_loss=0.1395, over 28536.00 frames. ], tot_loss[loss=0.3638, simple_loss=0.4103, pruned_loss=0.1587, over 5646488.27 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3884, pruned_loss=0.1346, over 5730947.09 frames. ], giga_tot_loss[loss=0.3627, simple_loss=0.4093, pruned_loss=0.158, over 5643254.71 frames. ], batch size: 85, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 07:00:06,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:00:09,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.674e+03 2.018e+03 2.928e+03 7.277e+03, threshold=4.036e+03, percent-clipped=9.0
+2023-03-03 07:01:01,784 INFO [train.py:968] (0/2) Epoch 6, batch 23950, giga_loss[loss=0.3643, simple_loss=0.4206, pruned_loss=0.154, over 29062.00 frames. ], tot_loss[loss=0.3688, simple_loss=0.4141, pruned_loss=0.1617, over 5647196.94 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3885, pruned_loss=0.1347, over 5733911.64 frames. ], giga_tot_loss[loss=0.3684, simple_loss=0.4138, pruned_loss=0.1615, over 5640549.58 frames. ], batch size: 155, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 07:01:06,444 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-03 07:01:06,528 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-03 07:01:36,796 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:01:56,217 INFO [train.py:968] (0/2) Epoch 6, batch 24000, giga_loss[loss=0.3727, simple_loss=0.4183, pruned_loss=0.1635, over 28507.00 frames. ], tot_loss[loss=0.3687, simple_loss=0.4141, pruned_loss=0.1616, over 5641931.40 frames. ], libri_tot_loss[loss=0.3291, simple_loss=0.3887, pruned_loss=0.1348, over 5727296.88 frames. ], giga_tot_loss[loss=0.3689, simple_loss=0.4141, pruned_loss=0.1619, over 5640867.47 frames. ], batch size: 336, lr: 5.34e-03, grad_scale: 8.0
+2023-03-03 07:01:56,223 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 07:02:04,322 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1314, 1.7982, 1.5115, 1.2701], device='cuda:0'), covar=tensor([0.1667, 0.2191, 0.1364, 0.1459], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0723, 0.0791, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 07:02:04,973 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2315, simple_loss=0.3361, pruned_loss=0.06339, over 944034.00 frames. 
+2023-03-03 07:02:04,974 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 07:02:10,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.283e+02 1.660e+03 2.284e+03 3.081e+03 7.411e+03, threshold=4.569e+03, percent-clipped=9.0
+2023-03-03 07:02:18,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4891, 3.9195, 1.6585, 1.5282], device='cuda:0'), covar=tensor([0.0893, 0.0240, 0.0761, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0490, 0.0314, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 07:02:38,375 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:02:41,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:02:47,368 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:02:50,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:02:51,232 INFO [train.py:968] (0/2) Epoch 6, batch 24050, giga_loss[loss=0.3419, simple_loss=0.3954, pruned_loss=0.1442, over 28919.00 frames. ], tot_loss[loss=0.3673, simple_loss=0.4122, pruned_loss=0.1611, over 5644884.70 frames. ], libri_tot_loss[loss=0.3291, simple_loss=0.3886, pruned_loss=0.1348, over 5732057.43 frames. ], giga_tot_loss[loss=0.3684, simple_loss=0.4129, pruned_loss=0.162, over 5637597.56 frames. ], batch size: 199, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 07:03:06,446 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:03:17,264 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:03:24,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:03:38,390 INFO [train.py:968] (0/2) Epoch 6, batch 24100, giga_loss[loss=0.3786, simple_loss=0.4005, pruned_loss=0.1784, over 23518.00 frames. ], tot_loss[loss=0.3648, simple_loss=0.4106, pruned_loss=0.1595, over 5639736.35 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3888, pruned_loss=0.1352, over 5726956.75 frames. ], giga_tot_loss[loss=0.3661, simple_loss=0.4114, pruned_loss=0.1604, over 5637553.27 frames. ], batch size: 705, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:03:43,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.779e+03 2.158e+03 3.299e+03 5.952e+03, threshold=4.316e+03, percent-clipped=7.0
+2023-03-03 07:04:24,571 INFO [train.py:968] (0/2) Epoch 6, batch 24150, giga_loss[loss=0.3146, simple_loss=0.3845, pruned_loss=0.1223, over 29013.00 frames. ], tot_loss[loss=0.365, simple_loss=0.4114, pruned_loss=0.1593, over 5641992.97 frames. ], libri_tot_loss[loss=0.3299, simple_loss=0.3889, pruned_loss=0.1354, over 5726428.63 frames. ], giga_tot_loss[loss=0.3667, simple_loss=0.4126, pruned_loss=0.1604, over 5638708.15 frames. ], batch size: 128, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:05:07,155 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1378, 0.8695, 0.7523, 1.3282], device='cuda:0'), covar=tensor([0.0774, 0.0363, 0.0365, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0119, 0.0123, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0062, 0.0046, 0.0041, 0.0068], device='cuda:0')
+2023-03-03 07:05:09,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6686, 2.3808, 1.5011, 0.7901], device='cuda:0'), covar=tensor([0.3956, 0.1889, 0.2231, 0.3559], device='cuda:0'), in_proj_covar=tensor([0.1452, 0.1338, 0.1392, 0.1177], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 07:05:14,635 INFO [train.py:968] (0/2) Epoch 6, batch 24200, giga_loss[loss=0.4423, simple_loss=0.4663, pruned_loss=0.2092, over 27908.00 frames. ], tot_loss[loss=0.3658, simple_loss=0.4124, pruned_loss=0.1597, over 5628311.09 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3894, pruned_loss=0.1359, over 5718481.33 frames. ], giga_tot_loss[loss=0.3671, simple_loss=0.4132, pruned_loss=0.1605, over 5631062.59 frames. ], batch size: 412, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:05:19,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.850e+02 1.544e+03 2.038e+03 2.867e+03 1.163e+04, threshold=4.076e+03, percent-clipped=11.0
+2023-03-03 07:06:03,674 INFO [train.py:968] (0/2) Epoch 6, batch 24250, libri_loss[loss=0.3405, simple_loss=0.386, pruned_loss=0.1475, over 29560.00 frames. ], tot_loss[loss=0.3644, simple_loss=0.4113, pruned_loss=0.1587, over 5620159.41 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3895, pruned_loss=0.1364, over 5711950.28 frames. ], giga_tot_loss[loss=0.3662, simple_loss=0.4127, pruned_loss=0.1598, over 5624374.77 frames. ], batch size: 77, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:06:07,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:06:43,986 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-252000.pt
+2023-03-03 07:06:54,117 INFO [train.py:968] (0/2) Epoch 6, batch 24300, giga_loss[loss=0.3583, simple_loss=0.4136, pruned_loss=0.1515, over 28690.00 frames. ], tot_loss[loss=0.359, simple_loss=0.4081, pruned_loss=0.155, over 5626348.39 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.39, pruned_loss=0.137, over 5715348.51 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4091, pruned_loss=0.1556, over 5625174.60 frames. ], batch size: 262, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:07:00,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.296e+02 1.447e+03 2.011e+03 2.708e+03 5.458e+03, threshold=4.023e+03, percent-clipped=4.0
+2023-03-03 07:07:00,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6010, 5.4088, 5.1301, 2.8090], device='cuda:0'), covar=tensor([0.0309, 0.0505, 0.0655, 0.1501], device='cuda:0'), in_proj_covar=tensor([0.0911, 0.0860, 0.0800, 0.0613], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 07:07:16,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:07:41,080 INFO [train.py:968] (0/2) Epoch 6, batch 24350, giga_loss[loss=0.3212, simple_loss=0.3838, pruned_loss=0.1293, over 28335.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4048, pruned_loss=0.151, over 5627381.75 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3897, pruned_loss=0.137, over 5704657.54 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4064, pruned_loss=0.152, over 5633514.16 frames. ], batch size: 368, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:07:48,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=252069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:08:21,122 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:08:28,424 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:08:29,462 INFO [train.py:968] (0/2) Epoch 6, batch 24400, libri_loss[loss=0.3137, simple_loss=0.3756, pruned_loss=0.1259, over 29570.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.4021, pruned_loss=0.1481, over 5649724.00 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.39, pruned_loss=0.1373, over 5707412.91 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4033, pruned_loss=0.1489, over 5650768.56 frames. ], batch size: 76, lr: 5.33e-03, grad_scale: 8.0
+2023-03-03 07:08:30,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:08:35,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.291e+02 1.530e+03 2.270e+03 3.300e+03 7.717e+03, threshold=4.541e+03, percent-clipped=13.0
+2023-03-03 07:08:59,603 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:09:20,008 INFO [train.py:968] (0/2) Epoch 6, batch 24450, giga_loss[loss=0.2801, simple_loss=0.3579, pruned_loss=0.1012, over 28825.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.3983, pruned_loss=0.145, over 5650031.77 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3892, pruned_loss=0.1368, over 5710619.98 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.4001, pruned_loss=0.1461, over 5647342.57 frames. ], batch size: 174, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:09:31,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=252172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:10:07,156 INFO [train.py:968] (0/2) Epoch 6, batch 24500, giga_loss[loss=0.3644, simple_loss=0.407, pruned_loss=0.1609, over 27830.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3958, pruned_loss=0.1431, over 5664716.85 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3885, pruned_loss=0.1364, over 5712536.07 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3981, pruned_loss=0.1446, over 5659186.54 frames. ], batch size: 412, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:10:08,077 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:10:10,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:10:15,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.041e+02 1.660e+03 2.111e+03 2.919e+03 6.932e+03, threshold=4.222e+03, percent-clipped=6.0
+2023-03-03 07:10:19,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3772, 4.2252, 3.9912, 1.9930], device='cuda:0'), covar=tensor([0.0469, 0.0564, 0.0690, 0.1940], device='cuda:0'), in_proj_covar=tensor([0.0919, 0.0860, 0.0802, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 07:10:44,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:11:02,631 INFO [train.py:968] (0/2) Epoch 6, batch 24550, giga_loss[loss=0.3174, simple_loss=0.3849, pruned_loss=0.125, over 28592.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3964, pruned_loss=0.1432, over 5664061.42 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3885, pruned_loss=0.1365, over 5713490.82 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3983, pruned_loss=0.1444, over 5658283.16 frames. ], batch size: 65, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:11:52,948 INFO [train.py:968] (0/2) Epoch 6, batch 24600, giga_loss[loss=0.2918, simple_loss=0.3604, pruned_loss=0.1116, over 28702.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3942, pruned_loss=0.1409, over 5668676.18 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3888, pruned_loss=0.1369, over 5716558.99 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3956, pruned_loss=0.1416, over 5660669.19 frames. ], batch size: 242, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:11:55,154 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-03 07:11:58,691 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:12:03,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:12:04,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.002e+02 1.670e+03 2.212e+03 3.074e+03 8.071e+03, threshold=4.424e+03, percent-clipped=6.0
+2023-03-03 07:12:12,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4362, 4.1687, 1.6133, 1.5404], device='cuda:0'), covar=tensor([0.0934, 0.0309, 0.0862, 0.1333], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0492, 0.0316, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 07:12:15,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1273, 1.4988, 1.3119, 0.8248], device='cuda:0'), covar=tensor([0.1629, 0.1227, 0.0802, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.1485, 0.1319, 0.1279, 0.1393], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 07:12:18,331 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 07:12:28,427 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:12:43,320 INFO [train.py:968] (0/2) Epoch 6, batch 24650, libri_loss[loss=0.3481, simple_loss=0.4023, pruned_loss=0.147, over 29373.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3952, pruned_loss=0.1391, over 5676073.84 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3891, pruned_loss=0.1373, over 5709230.67 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3963, pruned_loss=0.1393, over 5673812.58 frames. ], batch size: 92, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:12:50,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9425, 2.9059, 1.9763, 1.6085], device='cuda:0'), covar=tensor([0.1293, 0.0574, 0.0758, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.1489, 0.1321, 0.1279, 0.1392], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 07:13:06,757 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-03 07:13:32,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=252409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:13:36,655 INFO [train.py:968] (0/2) Epoch 6, batch 24700, giga_loss[loss=0.3767, simple_loss=0.4007, pruned_loss=0.1763, over 23597.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3967, pruned_loss=0.1394, over 5638052.65 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3893, pruned_loss=0.1376, over 5693301.19 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3975, pruned_loss=0.1394, over 5649838.57 frames. ], batch size: 705, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:13:45,050 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.643e+03 2.120e+03 3.219e+03 8.753e+03, threshold=4.240e+03, percent-clipped=11.0
+2023-03-03 07:13:46,192 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-03 07:14:10,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 07:14:30,476 INFO [train.py:968] (0/2) Epoch 6, batch 24750, giga_loss[loss=0.3933, simple_loss=0.4248, pruned_loss=0.1809, over 26637.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3967, pruned_loss=0.14, over 5644187.07 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3893, pruned_loss=0.1376, over 5694370.68 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3973, pruned_loss=0.14, over 5652341.17 frames. ], batch size: 555, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:14:44,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=252477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:15:14,112 INFO [train.py:968] (0/2) Epoch 6, batch 24800, libri_loss[loss=0.3199, simple_loss=0.3787, pruned_loss=0.1305, over 29570.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3967, pruned_loss=0.1407, over 5655394.99 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3881, pruned_loss=0.1368, over 5700528.40 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3985, pruned_loss=0.1416, over 5654935.15 frames. ], batch size: 74, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:15:24,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.856e+03 2.157e+03 3.222e+03 1.163e+04, threshold=4.313e+03, percent-clipped=10.0
+2023-03-03 07:15:49,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5678, 1.5696, 1.0471, 1.2088], device='cuda:0'), covar=tensor([0.0800, 0.0731, 0.1422, 0.1207], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0452, 0.0502, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 07:15:51,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1719, 1.3576, 1.1068, 0.8675], device='cuda:0'), covar=tensor([0.1143, 0.1139, 0.0755, 0.1083], device='cuda:0'), in_proj_covar=tensor([0.1516, 0.1348, 0.1304, 0.1416], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 07:15:54,742 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:15:58,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:16:04,294 INFO [train.py:968] (0/2) Epoch 6, batch 24850, giga_loss[loss=0.2736, simple_loss=0.3374, pruned_loss=0.1049, over 28574.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3939, pruned_loss=0.14, over 5652260.02 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3879, pruned_loss=0.1366, over 5704299.01 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3957, pruned_loss=0.1408, over 5647734.82 frames. ], batch size: 85, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:16:16,777 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 07:16:25,834 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:16:47,605 INFO [train.py:968] (0/2) Epoch 6, batch 24900, giga_loss[loss=0.3842, simple_loss=0.4303, pruned_loss=0.1691, over 28822.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.392, pruned_loss=0.1388, over 5668548.44 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3886, pruned_loss=0.1371, over 5706466.61 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3928, pruned_loss=0.1392, over 5662114.57 frames. ], batch size: 186, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:16:56,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.823e+02 1.495e+03 2.018e+03 2.570e+03 7.205e+03, threshold=4.037e+03, percent-clipped=6.0
+2023-03-03 07:16:57,056 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:17:00,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:17:26,705 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:17:34,361 INFO [train.py:968] (0/2) Epoch 6, batch 24950, giga_loss[loss=0.2991, simple_loss=0.3696, pruned_loss=0.1143, over 28822.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3919, pruned_loss=0.1385, over 5669346.03 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3887, pruned_loss=0.1371, over 5705526.28 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3925, pruned_loss=0.1387, over 5664890.07 frames. ], batch size: 186, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:18:02,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2170, 1.2533, 1.2029, 1.5153], device='cuda:0'), covar=tensor([0.0793, 0.0355, 0.0343, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0120, 0.0123, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0063, 0.0046, 0.0041, 0.0069], device='cuda:0')
+2023-03-03 07:18:24,584 INFO [train.py:968] (0/2) Epoch 6, batch 25000, giga_loss[loss=0.291, simple_loss=0.3652, pruned_loss=0.1084, over 28754.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3898, pruned_loss=0.136, over 5666317.00 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3884, pruned_loss=0.1369, over 5706674.64 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3906, pruned_loss=0.1363, over 5661773.85 frames. ], batch size: 119, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:18:33,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.801e+02 1.366e+03 1.859e+03 2.628e+03 6.072e+03, threshold=3.718e+03, percent-clipped=3.0
+2023-03-03 07:19:08,127 INFO [train.py:968] (0/2) Epoch 6, batch 25050, giga_loss[loss=0.3256, simple_loss=0.3885, pruned_loss=0.1313, over 28674.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3905, pruned_loss=0.1363, over 5672868.91 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3889, pruned_loss=0.1373, over 5713493.62 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3907, pruned_loss=0.1362, over 5661915.35 frames. ], batch size: 242, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:19:35,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1346, 1.2233, 3.7618, 3.1997], device='cuda:0'), covar=tensor([0.1392, 0.2115, 0.0376, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0581, 0.0539, 0.0775, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 07:19:56,972 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:19:57,382 INFO [train.py:968] (0/2) Epoch 6, batch 25100, giga_loss[loss=0.3242, simple_loss=0.3855, pruned_loss=0.1314, over 28928.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3904, pruned_loss=0.1369, over 5681848.59 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3887, pruned_loss=0.1373, over 5717153.16 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3907, pruned_loss=0.1368, over 5669399.42 frames. ], batch size: 227, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:20:05,375 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.251e+02 1.468e+03 1.858e+03 2.553e+03 5.123e+03, threshold=3.717e+03, percent-clipped=9.0
+2023-03-03 07:20:14,060 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:20:44,658 INFO [train.py:968] (0/2) Epoch 6, batch 25150, giga_loss[loss=0.3279, simple_loss=0.387, pruned_loss=0.1344, over 28894.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3905, pruned_loss=0.1374, over 5687347.90 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3894, pruned_loss=0.1376, over 5719270.72 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3902, pruned_loss=0.137, over 5674855.34 frames. ], batch size: 227, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:21:31,689 INFO [train.py:968] (0/2) Epoch 6, batch 25200, giga_loss[loss=0.3052, simple_loss=0.3661, pruned_loss=0.1221, over 28880.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3901, pruned_loss=0.1379, over 5687110.72 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3899, pruned_loss=0.1379, over 5720039.17 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3894, pruned_loss=0.1373, over 5675819.99 frames. ], batch size: 186, lr: 5.32e-03, grad_scale: 8.0
+2023-03-03 07:21:40,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.484e+02 1.598e+03 2.006e+03 2.565e+03 6.227e+03, threshold=4.011e+03, percent-clipped=6.0
+2023-03-03 07:22:19,287 INFO [train.py:968] (0/2) Epoch 6, batch 25250, giga_loss[loss=0.3506, simple_loss=0.3986, pruned_loss=0.1513, over 27967.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.388, pruned_loss=0.1366, over 5697244.47 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3899, pruned_loss=0.1378, over 5722722.98 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3875, pruned_loss=0.1362, over 5685678.16 frames. ], batch size: 412, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:22:43,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:23:02,509 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:23:06,940 INFO [train.py:968] (0/2) Epoch 6, batch 25300, giga_loss[loss=0.292, simple_loss=0.3491, pruned_loss=0.1174, over 28631.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3855, pruned_loss=0.1353, over 5694512.63 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3895, pruned_loss=0.1376, over 5725078.08 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3853, pruned_loss=0.1352, over 5682648.61 frames. ], batch size: 92, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:23:15,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.778e+03 2.200e+03 3.320e+03 9.743e+03, threshold=4.400e+03, percent-clipped=16.0
+2023-03-03 07:23:41,086 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 07:23:56,271 INFO [train.py:968] (0/2) Epoch 6, batch 25350, giga_loss[loss=0.3133, simple_loss=0.3716, pruned_loss=0.1275, over 28264.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3862, pruned_loss=0.1366, over 5690773.55 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.39, pruned_loss=0.1379, over 5726870.20 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3855, pruned_loss=0.1362, over 5678952.31 frames. ], batch size: 77, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:23:56,506 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:24:39,298 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 07:24:45,312 INFO [train.py:968] (0/2) Epoch 6, batch 25400, giga_loss[loss=0.3097, simple_loss=0.3757, pruned_loss=0.1219, over 28492.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3861, pruned_loss=0.1363, over 5689910.66 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.39, pruned_loss=0.138, over 5729321.48 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3855, pruned_loss=0.1358, over 5677877.52 frames. ], batch size: 71, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:24:54,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.282e+02 1.678e+03 2.180e+03 2.973e+03 6.924e+03, threshold=4.360e+03, percent-clipped=10.0
+2023-03-03 07:25:27,476 INFO [train.py:968] (0/2) Epoch 6, batch 25450, giga_loss[loss=0.3196, simple_loss=0.3777, pruned_loss=0.1308, over 28311.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3867, pruned_loss=0.1358, over 5693903.03 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3895, pruned_loss=0.1378, over 5730061.60 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3865, pruned_loss=0.1356, over 5682399.63 frames. ], batch size: 368, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:25:31,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4250, 1.8429, 1.7287, 1.5286], device='cuda:0'), covar=tensor([0.1496, 0.1849, 0.1195, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0736, 0.0803, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 07:25:50,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:26:04,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:26:11,641 INFO [train.py:968] (0/2) Epoch 6, batch 25500, giga_loss[loss=0.2891, simple_loss=0.3656, pruned_loss=0.1062, over 28931.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3858, pruned_loss=0.1342, over 5692957.79 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3896, pruned_loss=0.138, over 5728906.54 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3855, pruned_loss=0.1338, over 5683723.06 frames. ], batch size: 227, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:26:20,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.543e+02 1.460e+03 1.870e+03 2.470e+03 7.582e+03, threshold=3.741e+03, percent-clipped=4.0
+2023-03-03 07:26:48,791 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:26:56,732 INFO [train.py:968] (0/2) Epoch 6, batch 25550, giga_loss[loss=0.3916, simple_loss=0.4283, pruned_loss=0.1774, over 27934.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3866, pruned_loss=0.1345, over 5691383.16 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3894, pruned_loss=0.1378, over 5730685.66 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3864, pruned_loss=0.1343, over 5681238.67 frames. ], batch size: 412, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:27:46,148 INFO [train.py:968] (0/2) Epoch 6, batch 25600, libri_loss[loss=0.3953, simple_loss=0.443, pruned_loss=0.1738, over 28721.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3888, pruned_loss=0.1366, over 5692278.91 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3895, pruned_loss=0.1379, over 5731952.58 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3885, pruned_loss=0.1363, over 5682285.09 frames. ], batch size: 106, lr: 5.32e-03, grad_scale: 8.0
+2023-03-03 07:27:56,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.705e+03 2.180e+03 2.818e+03 7.554e+03, threshold=4.360e+03, percent-clipped=11.0
+2023-03-03 07:28:04,241 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:06,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:18,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:21,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:36,618 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:36,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:37,009 INFO [train.py:968] (0/2) Epoch 6, batch 25650, giga_loss[loss=0.3028, simple_loss=0.3709, pruned_loss=0.1174, over 28961.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3917, pruned_loss=0.1404, over 5687042.29 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3896, pruned_loss=0.1379, over 5733538.63 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3914, pruned_loss=0.1401, over 5677414.37 frames. ], batch size: 164, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:28:52,311 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:55,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:59,534 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:29:22,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2231, 1.5171, 1.2676, 1.1747], device='cuda:0'), covar=tensor([0.1740, 0.1580, 0.1658, 0.1426], device='cuda:0'), in_proj_covar=tensor([0.1149, 0.0882, 0.1018, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 07:29:28,462 INFO [train.py:968] (0/2) Epoch 6, batch 25700, giga_loss[loss=0.3502, simple_loss=0.4036, pruned_loss=0.1484, over 28862.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3933, pruned_loss=0.1435, over 5673824.05 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3899, pruned_loss=0.1383, over 5726853.58 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3929, pruned_loss=0.1431, over 5670017.40 frames. ], batch size: 199, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:29:34,761 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-03 07:29:39,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.816e+03 2.548e+03 3.762e+03 1.414e+04, threshold=5.097e+03, percent-clipped=15.0
+2023-03-03 07:29:54,990 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:30:19,010 INFO [train.py:968] (0/2) Epoch 6, batch 25750, giga_loss[loss=0.3281, simple_loss=0.3877, pruned_loss=0.1342, over 28658.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3939, pruned_loss=0.1446, over 5683650.67 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3897, pruned_loss=0.1382, over 5730216.97 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3938, pruned_loss=0.1444, over 5676665.44 frames. ], batch size: 242, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:30:55,003 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:30:56,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:31:02,696 INFO [train.py:968] (0/2) Epoch 6, batch 25800, giga_loss[loss=0.3286, simple_loss=0.3804, pruned_loss=0.1384, over 28561.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3918, pruned_loss=0.1433, over 5668998.10 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3895, pruned_loss=0.1383, over 5722413.32 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3919, pruned_loss=0.1432, over 5668396.08 frames. ], batch size: 78, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:31:13,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.928e+02 1.649e+03 2.170e+03 2.831e+03 7.684e+03, threshold=4.340e+03, percent-clipped=6.0
+2023-03-03 07:31:15,542 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:31:19,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:31:29,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:31:47,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:31:51,284 INFO [train.py:968] (0/2) Epoch 6, batch 25850, giga_loss[loss=0.3222, simple_loss=0.3874, pruned_loss=0.1285, over 28635.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3913, pruned_loss=0.142, over 5664256.61 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3895, pruned_loss=0.1383, over 5714009.99 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3914, pruned_loss=0.142, over 5671454.74 frames. ], batch size: 307, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:32:05,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7002, 1.6963, 1.2003, 1.4082], device='cuda:0'), covar=tensor([0.0737, 0.0629, 0.1063, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0455, 0.0506, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 07:32:07,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:32:10,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:32:35,440 INFO [train.py:968] (0/2) Epoch 6, batch 25900, giga_loss[loss=0.2703, simple_loss=0.342, pruned_loss=0.09924, over 28647.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3894, pruned_loss=0.1393, over 5662150.65 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3896, pruned_loss=0.1383, over 5714105.21 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3894, pruned_loss=0.1393, over 5667112.41 frames. ], batch size: 307, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:32:36,229 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:32:48,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.340e+02 1.422e+03 1.976e+03 2.606e+03 8.574e+03, threshold=3.951e+03, percent-clipped=4.0
+2023-03-03 07:32:52,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:33:20,865 INFO [train.py:968] (0/2) Epoch 6, batch 25950, giga_loss[loss=0.3016, simple_loss=0.363, pruned_loss=0.1201, over 29050.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3874, pruned_loss=0.1378, over 5664537.60 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3899, pruned_loss=0.1386, over 5717910.04 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3871, pruned_loss=0.1376, over 5663210.48 frames. ], batch size: 106, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:34:08,323 INFO [train.py:968] (0/2) Epoch 6, batch 26000, giga_loss[loss=0.3301, simple_loss=0.3864, pruned_loss=0.1369, over 29035.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3843, pruned_loss=0.1358, over 5668833.58 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3898, pruned_loss=0.1385, over 5717766.80 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.384, pruned_loss=0.1357, over 5666957.73 frames. ], batch size: 155, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:34:19,579 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:34:19,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.668e+02 1.576e+03 1.927e+03 2.937e+03 1.157e+04, threshold=3.854e+03, percent-clipped=10.0
+2023-03-03 07:35:00,926 INFO [train.py:968] (0/2) Epoch 6, batch 26050, giga_loss[loss=0.3763, simple_loss=0.4196, pruned_loss=0.1665, over 28514.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3853, pruned_loss=0.1377, over 5652684.16 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3894, pruned_loss=0.1382, over 5717567.06 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3854, pruned_loss=0.1379, over 5650299.83 frames. ], batch size: 336, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:35:01,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:08,118 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:10,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:13,151 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:39,239 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:46,603 INFO [train.py:968] (0/2) Epoch 6, batch 26100, giga_loss[loss=0.2964, simple_loss=0.3727, pruned_loss=0.1101, over 29001.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3883, pruned_loss=0.1396, over 5651213.63 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.39, pruned_loss=0.1386, over 5711036.13 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3878, pruned_loss=0.1394, over 5653374.19 frames. ], batch size: 145, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:35:47,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:56,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0777, 1.2051, 1.2969, 1.1087], device='cuda:0'), covar=tensor([0.0928, 0.0994, 0.1432, 0.1106], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0743, 0.0647, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 07:35:58,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.295e+02 1.395e+03 2.031e+03 3.014e+03 8.069e+03, threshold=4.062e+03, percent-clipped=15.0
+2023-03-03 07:36:29,231 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.53 vs. limit=5.0
+2023-03-03 07:36:32,655 INFO [train.py:968] (0/2) Epoch 6, batch 26150, giga_loss[loss=0.3199, simple_loss=0.3991, pruned_loss=0.1203, over 28799.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3925, pruned_loss=0.1398, over 5650501.53 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.3902, pruned_loss=0.1388, over 5705170.05 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3919, pruned_loss=0.1394, over 5657040.19 frames. ], batch size: 284, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:37:12,599 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:37:14,324 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:37:18,812 INFO [train.py:968] (0/2) Epoch 6, batch 26200, giga_loss[loss=0.3141, simple_loss=0.3837, pruned_loss=0.1223, over 28865.00 frames. ], tot_loss[loss=0.336, simple_loss=0.394, pruned_loss=0.139, over 5656291.99 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3895, pruned_loss=0.1384, over 5707684.04 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3943, pruned_loss=0.1391, over 5658155.89 frames. ], batch size: 174, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:37:19,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4482, 3.5044, 1.5390, 1.4234], device='cuda:0'), covar=tensor([0.0890, 0.0340, 0.0827, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0494, 0.0317, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 07:37:21,670 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:37:31,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.358e+02 1.394e+03 1.710e+03 2.581e+03 7.878e+03, threshold=3.421e+03, percent-clipped=7.0
+2023-03-03 07:37:42,130 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:38:05,933 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:38:06,313 INFO [train.py:968] (0/2) Epoch 6, batch 26250, giga_loss[loss=0.3427, simple_loss=0.401, pruned_loss=0.1422, over 28837.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3965, pruned_loss=0.1417, over 5661942.95 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3895, pruned_loss=0.1388, over 5713982.91 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3969, pruned_loss=0.1415, over 5656319.36 frames. ], batch size: 199, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:38:45,324 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-254000.pt
+2023-03-03 07:38:53,343 INFO [train.py:968] (0/2) Epoch 6, batch 26300, giga_loss[loss=0.3774, simple_loss=0.4247, pruned_loss=0.165, over 28713.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3976, pruned_loss=0.143, over 5659429.76 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3893, pruned_loss=0.1387, over 5716061.58 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3982, pruned_loss=0.143, over 5652504.47 frames. ], batch size: 262, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:39:04,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.154e+02 1.508e+03 1.943e+03 2.742e+03 7.993e+03, threshold=3.887e+03, percent-clipped=16.0
+2023-03-03 07:39:24,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:39:25,219 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 07:39:42,880 INFO [train.py:968] (0/2) Epoch 6, batch 26350, giga_loss[loss=0.2998, simple_loss=0.3631, pruned_loss=0.1183, over 28734.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3978, pruned_loss=0.144, over 5654644.90 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3896, pruned_loss=0.1388, over 5719754.52 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3981, pruned_loss=0.1439, over 5644668.05 frames. ], batch size: 99, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:40:17,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:40:29,891 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-03 07:40:30,117 INFO [train.py:968] (0/2) Epoch 6, batch 26400, giga_loss[loss=0.2757, simple_loss=0.3414, pruned_loss=0.105, over 28609.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3968, pruned_loss=0.1442, over 5643864.86 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3899, pruned_loss=0.1391, over 5713284.26 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3971, pruned_loss=0.1441, over 5640400.89 frames. ], batch size: 78, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:40:34,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8688, 1.7799, 1.7494, 1.6650], device='cuda:0'), covar=tensor([0.1076, 0.1730, 0.1593, 0.1523], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0742, 0.0646, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 07:40:41,033 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.304e+02 1.601e+03 2.083e+03 2.996e+03 7.215e+03, threshold=4.165e+03, percent-clipped=9.0
+2023-03-03 07:41:03,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:41:15,380 INFO [train.py:968] (0/2) Epoch 6, batch 26450, giga_loss[loss=0.2989, simple_loss=0.3644, pruned_loss=0.1167, over 29068.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3929, pruned_loss=0.1417, over 5645572.53 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3897, pruned_loss=0.1391, over 5707100.51 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3933, pruned_loss=0.1417, over 5647805.43 frames. ], batch size: 155, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:41:37,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:41:57,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4020, 2.0995, 1.6146, 0.6537], device='cuda:0'), covar=tensor([0.3010, 0.1340, 0.1917, 0.3086], device='cuda:0'), in_proj_covar=tensor([0.1437, 0.1341, 0.1376, 0.1168], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 07:42:04,406 INFO [train.py:968] (0/2) Epoch 6, batch 26500, giga_loss[loss=0.3434, simple_loss=0.3964, pruned_loss=0.1451, over 29015.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3925, pruned_loss=0.1426, over 5644811.14 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3902, pruned_loss=0.1395, over 5710763.23 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3925, pruned_loss=0.1423, over 5642170.60 frames. ], batch size: 164, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:42:09,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5497, 2.2915, 1.6463, 0.8479], device='cuda:0'), covar=tensor([0.3210, 0.1489, 0.2208, 0.2931], device='cuda:0'), in_proj_covar=tensor([0.1437, 0.1344, 0.1378, 0.1169], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 07:42:16,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.378e+02 1.677e+03 2.085e+03 2.958e+03 5.002e+03, threshold=4.169e+03, percent-clipped=6.0
+2023-03-03 07:42:32,290 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:42:34,149 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:42:48,697 INFO [train.py:968] (0/2) Epoch 6, batch 26550, giga_loss[loss=0.4214, simple_loss=0.4459, pruned_loss=0.1984, over 26579.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3927, pruned_loss=0.143, over 5637330.40 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3902, pruned_loss=0.1394, over 5703431.24 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3928, pruned_loss=0.1429, over 5639468.12 frames. ], batch size: 555, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:42:55,470 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9777, 1.1889, 0.9199, 0.2044], device='cuda:0'), covar=tensor([0.1348, 0.1121, 0.1598, 0.2486], device='cuda:0'), in_proj_covar=tensor([0.1431, 0.1343, 0.1372, 0.1166], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 07:42:58,481 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:12,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:15,047 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:16,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:26,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3472, 1.9361, 2.0600, 1.7684], device='cuda:0'), covar=tensor([0.1047, 0.2106, 0.1495, 0.1675], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0741, 0.0649, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 07:43:31,038 INFO [train.py:968] (0/2) Epoch 6, batch 26600, giga_loss[loss=0.3116, simple_loss=0.3717, pruned_loss=0.1257, over 28789.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3916, pruned_loss=0.1423, over 5649751.32 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3903, pruned_loss=0.1396, over 5705484.04 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3917, pruned_loss=0.1422, over 5648152.19 frames. ], batch size: 99, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:43:42,212 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:43,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.581e+03 2.237e+03 3.346e+03 1.739e+04, threshold=4.474e+03, percent-clipped=15.0
+2023-03-03 07:43:49,457 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:52,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:54,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:44:18,798 INFO [train.py:968] (0/2) Epoch 6, batch 26650, giga_loss[loss=0.3517, simple_loss=0.4141, pruned_loss=0.1446, over 28873.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3902, pruned_loss=0.1417, over 5665354.18 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3903, pruned_loss=0.1395, over 5708275.71 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3902, pruned_loss=0.1417, over 5660943.06 frames. ], batch size: 174, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:44:19,675 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:45:07,375 INFO [train.py:968] (0/2) Epoch 6, batch 26700, giga_loss[loss=0.3013, simple_loss=0.3625, pruned_loss=0.12, over 28817.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3896, pruned_loss=0.1414, over 5656615.97 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3901, pruned_loss=0.1395, over 5700222.56 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3897, pruned_loss=0.1415, over 5660500.26 frames. ], batch size: 99, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:45:10,902 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-03 07:45:13,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:45:18,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.297e+02 1.729e+03 2.114e+03 2.931e+03 9.497e+03, threshold=4.229e+03, percent-clipped=9.0
+2023-03-03 07:45:26,414 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:45:28,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:45:53,510 INFO [train.py:968] (0/2) Epoch 6, batch 26750, giga_loss[loss=0.3294, simple_loss=0.3961, pruned_loss=0.1314, over 29059.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3917, pruned_loss=0.1417, over 5661435.17 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3906, pruned_loss=0.1398, over 5703959.58 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3914, pruned_loss=0.1415, over 5660691.91 frames. ], batch size: 155, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:45:57,886 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:46:12,826 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:46:15,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:46:45,100 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:46:45,436 INFO [train.py:968] (0/2) Epoch 6, batch 26800, giga_loss[loss=0.3684, simple_loss=0.3916, pruned_loss=0.1726, over 23804.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3935, pruned_loss=0.143, over 5652031.94 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3905, pruned_loss=0.1397, over 5702335.56 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3934, pruned_loss=0.143, over 5652425.26 frames. ], batch size: 705, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:46:58,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.294e+02 1.639e+03 1.994e+03 2.721e+03 1.476e+04, threshold=3.989e+03, percent-clipped=9.0
+2023-03-03 07:47:30,716 INFO [train.py:968] (0/2) Epoch 6, batch 26850, giga_loss[loss=0.3492, simple_loss=0.4109, pruned_loss=0.1437, over 28443.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3932, pruned_loss=0.1431, over 5663029.00 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3905, pruned_loss=0.1396, over 5707174.94 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3932, pruned_loss=0.1432, over 5657552.50 frames. ], batch size: 71, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:47:31,091 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:47:31,222 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 07:47:34,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:47:40,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8563, 1.1200, 3.8483, 3.0494], device='cuda:0'), covar=tensor([0.1978, 0.2682, 0.0442, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0597, 0.0550, 0.0784, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 07:47:58,597 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:48:15,795 INFO [train.py:968] (0/2) Epoch 6, batch 26900, giga_loss[loss=0.3591, simple_loss=0.3958, pruned_loss=0.1612, over 26706.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.393, pruned_loss=0.1396, over 5677177.34 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3907, pruned_loss=0.1399, over 5710592.59 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3929, pruned_loss=0.1395, over 5668788.31 frames. ], batch size: 555, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:48:20,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6330, 2.3810, 1.7386, 0.7898], device='cuda:0'), covar=tensor([0.2051, 0.1211, 0.1903, 0.2317], device='cuda:0'), in_proj_covar=tensor([0.1442, 0.1352, 0.1377, 0.1179], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 07:48:26,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.558e+02 1.523e+03 1.955e+03 2.648e+03 8.024e+03, threshold=3.911e+03, percent-clipped=6.0
+2023-03-03 07:48:26,982 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-03 07:49:01,692 INFO [train.py:968] (0/2) Epoch 6, batch 26950, giga_loss[loss=0.2793, simple_loss=0.3606, pruned_loss=0.09894, over 28462.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3937, pruned_loss=0.1383, over 5671445.32 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3904, pruned_loss=0.1398, over 5711563.44 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3939, pruned_loss=0.1383, over 5663648.11 frames. ], batch size: 71, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:49:28,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1763, 1.4877, 1.1369, 0.4818], device='cuda:0'), covar=tensor([0.1222, 0.0880, 0.1115, 0.1945], device='cuda:0'), in_proj_covar=tensor([0.1442, 0.1353, 0.1382, 0.1181], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 07:49:39,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4191, 1.8721, 1.7992, 1.5153], device='cuda:0'), covar=tensor([0.1593, 0.1886, 0.1236, 0.1384], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0734, 0.0804, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 07:49:49,212 INFO [train.py:968] (0/2) Epoch 6, batch 27000, giga_loss[loss=0.3111, simple_loss=0.3833, pruned_loss=0.1194, over 28879.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3957, pruned_loss=0.1385, over 5676859.47 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3903, pruned_loss=0.1397, over 5712540.62 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3958, pruned_loss=0.1386, over 5669612.50 frames. ], batch size: 119, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:49:49,216 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 07:49:56,874 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6159, 1.5806, 1.2252, 1.2815], device='cuda:0'), covar=tensor([0.0576, 0.0450, 0.0916, 0.0756], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0454, 0.0505, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 07:49:57,828 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2321, simple_loss=0.3361, pruned_loss=0.06409, over 944034.00 frames. 
+2023-03-03 07:49:57,829 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 07:50:09,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.993e+02 1.320e+03 1.690e+03 2.285e+03 6.275e+03, threshold=3.380e+03, percent-clipped=6.0
+2023-03-03 07:50:46,747 INFO [train.py:968] (0/2) Epoch 6, batch 27050, giga_loss[loss=0.3824, simple_loss=0.4225, pruned_loss=0.1711, over 29002.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.4008, pruned_loss=0.1445, over 5674123.06 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3903, pruned_loss=0.1396, over 5713521.92 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.4011, pruned_loss=0.1447, over 5667383.42 frames. ], batch size: 106, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:51:37,263 INFO [train.py:968] (0/2) Epoch 6, batch 27100, giga_loss[loss=0.363, simple_loss=0.4173, pruned_loss=0.1543, over 28869.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.4005, pruned_loss=0.1447, over 5683853.45 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3902, pruned_loss=0.1396, over 5715959.41 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.401, pruned_loss=0.145, over 5675663.17 frames. ], batch size: 174, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:51:49,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.601e+03 2.165e+03 2.755e+03 5.367e+03, threshold=4.331e+03, percent-clipped=12.0
+2023-03-03 07:52:02,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4188, 1.7518, 1.8277, 1.5372], device='cuda:0'), covar=tensor([0.1473, 0.2061, 0.1137, 0.1340], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0738, 0.0803, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 07:52:06,423 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-03 07:52:26,233 INFO [train.py:968] (0/2) Epoch 6, batch 27150, giga_loss[loss=0.3345, simple_loss=0.3964, pruned_loss=0.1363, over 28910.00 frames. ], tot_loss[loss=0.345, simple_loss=0.4, pruned_loss=0.145, over 5675946.48 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.39, pruned_loss=0.1395, over 5717815.85 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.4008, pruned_loss=0.1454, over 5667171.08 frames. ], batch size: 112, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:52:43,752 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3679, 1.6925, 1.3377, 1.1196], device='cuda:0'), covar=tensor([0.1733, 0.1177, 0.1002, 0.1325], device='cuda:0'), in_proj_covar=tensor([0.1515, 0.1338, 0.1287, 0.1403], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 07:53:14,817 INFO [train.py:968] (0/2) Epoch 6, batch 27200, libri_loss[loss=0.3412, simple_loss=0.3983, pruned_loss=0.142, over 29204.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3976, pruned_loss=0.142, over 5667096.80 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3906, pruned_loss=0.1399, over 5699416.40 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3979, pruned_loss=0.142, over 5675406.94 frames. ], batch size: 97, lr: 5.30e-03, grad_scale: 8.0
+2023-03-03 07:53:25,231 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.292e+02 1.603e+03 2.240e+03 3.380e+03 7.216e+03, threshold=4.480e+03, percent-clipped=7.0
+2023-03-03 07:53:29,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:53:59,639 INFO [train.py:968] (0/2) Epoch 6, batch 27250, giga_loss[loss=0.3595, simple_loss=0.3797, pruned_loss=0.1697, over 23572.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3977, pruned_loss=0.1406, over 5658940.74 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3908, pruned_loss=0.1401, over 5702937.39 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3979, pruned_loss=0.1405, over 5661888.58 frames. ], batch size: 705, lr: 5.30e-03, grad_scale: 8.0
+2023-03-03 07:54:06,244 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:54:31,888 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-03 07:54:46,613 INFO [train.py:968] (0/2) Epoch 6, batch 27300, giga_loss[loss=0.3167, simple_loss=0.3873, pruned_loss=0.1231, over 28834.00 frames. ], tot_loss[loss=0.341, simple_loss=0.399, pruned_loss=0.1415, over 5652514.29 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3909, pruned_loss=0.1403, over 5692312.95 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3993, pruned_loss=0.1413, over 5663017.84 frames. ], batch size: 199, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:54:57,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5334, 1.5406, 1.4943, 1.4863], device='cuda:0'), covar=tensor([0.0934, 0.1461, 0.1338, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0747, 0.0653, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 07:55:02,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.462e+03 1.835e+03 2.589e+03 5.452e+03, threshold=3.670e+03, percent-clipped=4.0
+2023-03-03 07:55:37,489 INFO [train.py:968] (0/2) Epoch 6, batch 27350, giga_loss[loss=0.3406, simple_loss=0.4021, pruned_loss=0.1396, over 28954.00 frames. ], tot_loss[loss=0.344, simple_loss=0.4008, pruned_loss=0.1436, over 5647082.97 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3905, pruned_loss=0.14, over 5694193.35 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.4016, pruned_loss=0.1437, over 5653029.48 frames. ], batch size: 227, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:56:23,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4808, 1.6583, 1.2772, 1.1756], device='cuda:0'), covar=tensor([0.1316, 0.1163, 0.0968, 0.1236], device='cuda:0'), in_proj_covar=tensor([0.1516, 0.1350, 0.1298, 0.1408], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 07:56:23,750 INFO [train.py:968] (0/2) Epoch 6, batch 27400, giga_loss[loss=0.3133, simple_loss=0.3776, pruned_loss=0.1245, over 28874.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3991, pruned_loss=0.1427, over 5656139.30 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3907, pruned_loss=0.1401, over 5696463.46 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3995, pruned_loss=0.1427, over 5658331.93 frames. ], batch size: 227, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:56:39,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.065e+02 1.507e+03 1.959e+03 2.652e+03 4.168e+03, threshold=3.919e+03, percent-clipped=4.0
+2023-03-03 07:57:01,124 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:57:13,969 INFO [train.py:968] (0/2) Epoch 6, batch 27450, giga_loss[loss=0.4246, simple_loss=0.4402, pruned_loss=0.2045, over 27586.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3982, pruned_loss=0.1433, over 5664368.64 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3908, pruned_loss=0.1401, over 5701007.00 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3987, pruned_loss=0.1434, over 5660959.04 frames. ], batch size: 472, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:57:43,152 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:58:04,480 INFO [train.py:968] (0/2) Epoch 6, batch 27500, giga_loss[loss=0.3043, simple_loss=0.3706, pruned_loss=0.119, over 28311.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3953, pruned_loss=0.1417, over 5660813.44 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3908, pruned_loss=0.1401, over 5693924.90 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3957, pruned_loss=0.1418, over 5663238.56 frames. ], batch size: 65, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:58:18,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.330e+02 1.671e+03 2.123e+03 3.014e+03 1.075e+04, threshold=4.246e+03, percent-clipped=12.0
+2023-03-03 07:58:38,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:58:55,815 INFO [train.py:968] (0/2) Epoch 6, batch 27550, giga_loss[loss=0.3476, simple_loss=0.3941, pruned_loss=0.1506, over 28912.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3927, pruned_loss=0.1403, over 5658793.41 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3913, pruned_loss=0.1405, over 5696993.73 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3927, pruned_loss=0.1401, over 5657534.32 frames. ], batch size: 213, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:59:17,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4843, 2.1954, 1.6008, 0.7551], device='cuda:0'), covar=tensor([0.2601, 0.1403, 0.2238, 0.2828], device='cuda:0'), in_proj_covar=tensor([0.1448, 0.1373, 0.1394, 0.1186], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 07:59:33,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:59:38,784 INFO [train.py:968] (0/2) Epoch 6, batch 27600, giga_loss[loss=0.4305, simple_loss=0.4548, pruned_loss=0.2031, over 29055.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3937, pruned_loss=0.1422, over 5657654.51 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3918, pruned_loss=0.1407, over 5692597.40 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3932, pruned_loss=0.1418, over 5659985.50 frames. ], batch size: 164, lr: 5.30e-03, grad_scale: 8.0
+2023-03-03 07:59:52,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.145e+02 1.555e+03 1.961e+03 2.375e+03 7.166e+03, threshold=3.921e+03, percent-clipped=6.0
+2023-03-03 08:00:07,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:00:22,888 INFO [train.py:968] (0/2) Epoch 6, batch 27650, giga_loss[loss=0.3776, simple_loss=0.4133, pruned_loss=0.171, over 26570.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3918, pruned_loss=0.1411, over 5651962.07 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3919, pruned_loss=0.1407, over 5688413.39 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3914, pruned_loss=0.1408, over 5657006.86 frames. ], batch size: 555, lr: 5.30e-03, grad_scale: 8.0
+2023-03-03 08:01:10,054 INFO [train.py:968] (0/2) Epoch 6, batch 27700, giga_loss[loss=0.4014, simple_loss=0.4361, pruned_loss=0.1833, over 26712.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.389, pruned_loss=0.1373, over 5649051.47 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3923, pruned_loss=0.1411, over 5682982.21 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3882, pruned_loss=0.1367, over 5658055.80 frames. ], batch size: 555, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 08:01:24,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.858e+02 1.411e+03 1.973e+03 2.800e+03 6.064e+03, threshold=3.945e+03, percent-clipped=6.0
+2023-03-03 08:01:46,711 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:01:49,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:02:01,052 INFO [train.py:968] (0/2) Epoch 6, batch 27750, giga_loss[loss=0.3545, simple_loss=0.4118, pruned_loss=0.1485, over 28610.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3867, pruned_loss=0.1352, over 5649331.52 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3923, pruned_loss=0.1411, over 5682982.21 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3861, pruned_loss=0.1347, over 5656339.70 frames. ], batch size: 307, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 08:02:16,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4465, 1.6505, 1.3830, 1.6923], device='cuda:0'), covar=tensor([0.2118, 0.1980, 0.2113, 0.1879], device='cuda:0'), in_proj_covar=tensor([0.1155, 0.0888, 0.1020, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:02:20,383 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:02:27,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:02:29,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:02:52,186 INFO [train.py:968] (0/2) Epoch 6, batch 27800, libri_loss[loss=0.3089, simple_loss=0.3588, pruned_loss=0.1295, over 29362.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3854, pruned_loss=0.1347, over 5656068.90 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3922, pruned_loss=0.141, over 5688332.72 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.385, pruned_loss=0.1343, over 5656175.92 frames. ], batch size: 67, lr: 5.30e-03, grad_scale: 2.0
+2023-03-03 08:03:01,506 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:03:06,094 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:03:10,925 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.315e+02 1.416e+03 1.910e+03 2.522e+03 8.164e+03, threshold=3.820e+03, percent-clipped=6.0
+2023-03-03 08:03:47,745 INFO [train.py:968] (0/2) Epoch 6, batch 27850, giga_loss[loss=0.3052, simple_loss=0.3668, pruned_loss=0.1218, over 29080.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3829, pruned_loss=0.1349, over 5645468.11 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3922, pruned_loss=0.1411, over 5687100.33 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3824, pruned_loss=0.1344, over 5646188.96 frames. ], batch size: 128, lr: 5.30e-03, grad_scale: 2.0
+2023-03-03 08:03:51,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:04:00,705 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5078, 1.7662, 1.8661, 1.6242], device='cuda:0'), covar=tensor([0.1423, 0.1776, 0.1067, 0.1213], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0740, 0.0812, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 08:04:38,209 INFO [train.py:968] (0/2) Epoch 6, batch 27900, giga_loss[loss=0.2943, simple_loss=0.3703, pruned_loss=0.1092, over 28889.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3826, pruned_loss=0.1343, over 5651558.22 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.392, pruned_loss=0.1409, over 5691032.84 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3822, pruned_loss=0.134, over 5647537.10 frames. ], batch size: 227, lr: 5.30e-03, grad_scale: 2.0
+2023-03-03 08:04:43,434 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:04:50,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.890e+02 1.852e+03 2.353e+03 3.924e+03 1.594e+04, threshold=4.706e+03, percent-clipped=25.0
+2023-03-03 08:04:55,018 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-03 08:05:23,895 INFO [train.py:968] (0/2) Epoch 6, batch 27950, giga_loss[loss=0.3299, simple_loss=0.3947, pruned_loss=0.1325, over 28874.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3848, pruned_loss=0.1347, over 5660980.69 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.392, pruned_loss=0.1409, over 5696361.36 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3843, pruned_loss=0.1344, over 5652648.51 frames. ], batch size: 186, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:05:31,773 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:05:34,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:06:02,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:06:12,475 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:06:14,800 INFO [train.py:968] (0/2) Epoch 6, batch 28000, giga_loss[loss=0.3345, simple_loss=0.3882, pruned_loss=0.1404, over 28334.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3845, pruned_loss=0.1338, over 5648154.35 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3925, pruned_loss=0.1411, over 5690114.79 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3835, pruned_loss=0.1332, over 5646235.55 frames. ], batch size: 368, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:06:15,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:06:29,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.347e+02 1.446e+03 1.885e+03 2.322e+03 5.858e+03, threshold=3.770e+03, percent-clipped=1.0
+2023-03-03 08:06:36,849 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9255, 3.7533, 3.5841, 1.7015], device='cuda:0'), covar=tensor([0.0586, 0.0678, 0.0783, 0.2063], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0874, 0.0814, 0.0605], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 08:06:39,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-03 08:06:40,625 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:06:59,006 INFO [train.py:968] (0/2) Epoch 6, batch 28050, giga_loss[loss=0.2956, simple_loss=0.3663, pruned_loss=0.1125, over 29022.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3848, pruned_loss=0.1346, over 5642989.06 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.3931, pruned_loss=0.1415, over 5684338.26 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3833, pruned_loss=0.1336, over 5646070.00 frames. ], batch size: 136, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:06:59,824 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:07:02,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:07:17,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:07:19,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4589, 3.3566, 1.4794, 1.4218], device='cuda:0'), covar=tensor([0.0792, 0.0316, 0.0837, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0495, 0.0318, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 08:07:24,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6091, 1.5107, 1.1759, 1.2672], device='cuda:0'), covar=tensor([0.0524, 0.0428, 0.0789, 0.0762], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0451, 0.0505, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 08:07:25,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4457, 4.2930, 4.0790, 2.0735], device='cuda:0'), covar=tensor([0.0509, 0.0611, 0.0812, 0.1844], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0882, 0.0824, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-03 08:07:27,978 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:07:41,530 INFO [train.py:968] (0/2) Epoch 6, batch 28100, giga_loss[loss=0.3768, simple_loss=0.3987, pruned_loss=0.1774, over 23583.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3867, pruned_loss=0.1366, over 5636213.66 frames. ], libri_tot_loss[loss=0.3388, simple_loss=0.3938, pruned_loss=0.1419, over 5678555.58 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3846, pruned_loss=0.1352, over 5642606.99 frames. ], batch size: 705, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:07:55,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.767e+02 1.469e+03 2.000e+03 2.986e+03 8.757e+03, threshold=4.001e+03, percent-clipped=14.0
+2023-03-03 08:08:28,547 INFO [train.py:968] (0/2) Epoch 6, batch 28150, giga_loss[loss=0.3102, simple_loss=0.3793, pruned_loss=0.1205, over 29147.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3896, pruned_loss=0.1387, over 5635798.12 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.3943, pruned_loss=0.1424, over 5671301.16 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3874, pruned_loss=0.1371, over 5646483.17 frames. ], batch size: 128, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:09:15,793 INFO [train.py:968] (0/2) Epoch 6, batch 28200, giga_loss[loss=0.3347, simple_loss=0.3903, pruned_loss=0.1396, over 28769.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3915, pruned_loss=0.1402, over 5641114.58 frames. ], libri_tot_loss[loss=0.3401, simple_loss=0.3947, pruned_loss=0.1428, over 5673780.97 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3895, pruned_loss=0.1386, over 5646787.06 frames. ], batch size: 284, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:09:30,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.724e+02 1.606e+03 2.211e+03 2.847e+03 7.103e+03, threshold=4.422e+03, percent-clipped=12.0
+2023-03-03 08:10:06,411 INFO [train.py:968] (0/2) Epoch 6, batch 28250, giga_loss[loss=0.3787, simple_loss=0.4204, pruned_loss=0.1685, over 28428.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3942, pruned_loss=0.1428, over 5637332.92 frames. ], libri_tot_loss[loss=0.3404, simple_loss=0.3949, pruned_loss=0.143, over 5675039.26 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3923, pruned_loss=0.1413, over 5640369.93 frames. ], batch size: 369, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:10:28,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4905, 1.8550, 1.7741, 1.5263], device='cuda:0'), covar=tensor([0.1522, 0.1863, 0.1131, 0.1267], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0743, 0.0812, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 08:10:43,198 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-256000.pt
+2023-03-03 08:10:53,567 INFO [train.py:968] (0/2) Epoch 6, batch 28300, giga_loss[loss=0.3646, simple_loss=0.3967, pruned_loss=0.1663, over 23660.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3934, pruned_loss=0.1427, over 5643707.52 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.394, pruned_loss=0.1426, over 5680632.98 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3927, pruned_loss=0.1419, over 5639672.03 frames. ], batch size: 705, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:11:09,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.666e+03 2.233e+03 3.094e+03 7.925e+03, threshold=4.465e+03, percent-clipped=14.0
+2023-03-03 08:11:18,144 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0652, 1.1889, 3.9584, 3.2208], device='cuda:0'), covar=tensor([0.1664, 0.2485, 0.0369, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.0598, 0.0553, 0.0793, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:11:30,105 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-03 08:11:45,974 INFO [train.py:968] (0/2) Epoch 6, batch 28350, giga_loss[loss=0.3483, simple_loss=0.4099, pruned_loss=0.1433, over 28616.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3926, pruned_loss=0.1402, over 5648355.16 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3937, pruned_loss=0.1424, over 5683675.93 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3923, pruned_loss=0.1397, over 5642195.77 frames. ], batch size: 307, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:12:13,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5159, 1.5077, 1.4951, 1.3941], device='cuda:0'), covar=tensor([0.0988, 0.1452, 0.1380, 0.1380], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0747, 0.0651, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 08:12:33,008 INFO [train.py:968] (0/2) Epoch 6, batch 28400, giga_loss[loss=0.3848, simple_loss=0.4029, pruned_loss=0.1833, over 23557.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3938, pruned_loss=0.1415, over 5641462.25 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3938, pruned_loss=0.1425, over 5679208.31 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3935, pruned_loss=0.1409, over 5640200.85 frames. ], batch size: 705, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:12:34,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9146, 3.7088, 3.5141, 1.7398], device='cuda:0'), covar=tensor([0.0679, 0.0898, 0.1087, 0.2014], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0877, 0.0823, 0.0611], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 08:12:48,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.354e+02 1.575e+03 2.012e+03 2.782e+03 9.755e+03, threshold=4.024e+03, percent-clipped=7.0
+2023-03-03 08:13:13,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:13:17,599 INFO [train.py:968] (0/2) Epoch 6, batch 28450, libri_loss[loss=0.2727, simple_loss=0.33, pruned_loss=0.1077, over 29390.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3928, pruned_loss=0.1416, over 5643805.66 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.3928, pruned_loss=0.142, over 5690445.90 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3934, pruned_loss=0.1415, over 5629940.36 frames. ], batch size: 67, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:13:38,065 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-03 08:14:08,508 INFO [train.py:968] (0/2) Epoch 6, batch 28500, giga_loss[loss=0.3175, simple_loss=0.3725, pruned_loss=0.1312, over 28566.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3926, pruned_loss=0.1421, over 5647990.11 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.3926, pruned_loss=0.1419, over 5697195.77 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3934, pruned_loss=0.1421, over 5628854.28 frames. ], batch size: 85, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:14:28,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.197e+02 1.715e+03 2.787e+03 4.155e+03 1.120e+04, threshold=5.573e+03, percent-clipped=25.0
+2023-03-03 08:14:53,237 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1707, 1.4264, 3.4461, 3.1719], device='cuda:0'), covar=tensor([0.1355, 0.2025, 0.0408, 0.0796], device='cuda:0'), in_proj_covar=tensor([0.0596, 0.0547, 0.0787, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:15:00,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3304, 1.4951, 1.1938, 1.8648], device='cuda:0'), covar=tensor([0.2005, 0.1949, 0.1988, 0.1949], device='cuda:0'), in_proj_covar=tensor([0.1152, 0.0885, 0.1020, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:15:01,071 INFO [train.py:968] (0/2) Epoch 6, batch 28550, giga_loss[loss=0.3449, simple_loss=0.3964, pruned_loss=0.1467, over 28372.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3912, pruned_loss=0.1419, over 5636435.22 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.3927, pruned_loss=0.1421, over 5701573.80 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3917, pruned_loss=0.1417, over 5614780.26 frames. ], batch size: 368, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:15:33,740 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:15:35,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:15:41,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9380, 1.1702, 3.6022, 3.1104], device='cuda:0'), covar=tensor([0.1615, 0.2210, 0.0433, 0.0742], device='cuda:0'), in_proj_covar=tensor([0.0593, 0.0545, 0.0783, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:15:43,701 INFO [train.py:968] (0/2) Epoch 6, batch 28600, giga_loss[loss=0.3487, simple_loss=0.3805, pruned_loss=0.1584, over 23674.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3896, pruned_loss=0.141, over 5645027.64 frames. ], libri_tot_loss[loss=0.3381, simple_loss=0.3922, pruned_loss=0.142, over 5696977.01 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3904, pruned_loss=0.141, over 5630548.75 frames. ], batch size: 705, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:16:00,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.554e+02 1.504e+03 1.928e+03 2.681e+03 8.185e+03, threshold=3.855e+03, percent-clipped=1.0
+2023-03-03 08:16:00,974 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:16:15,757 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:16:26,311 INFO [train.py:968] (0/2) Epoch 6, batch 28650, giga_loss[loss=0.3245, simple_loss=0.3892, pruned_loss=0.1299, over 28367.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3896, pruned_loss=0.1407, over 5655875.27 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3916, pruned_loss=0.1414, over 5696120.17 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3906, pruned_loss=0.1413, over 5641694.45 frames. ], batch size: 71, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:16:28,265 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6582, 2.1363, 1.5001, 1.3061], device='cuda:0'), covar=tensor([0.1401, 0.0960, 0.1122, 0.1291], device='cuda:0'), in_proj_covar=tensor([0.1516, 0.1345, 0.1298, 0.1410], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 08:16:57,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3534, 1.9777, 1.5520, 1.5599], device='cuda:0'), covar=tensor([0.0644, 0.0231, 0.0267, 0.0641], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0119, 0.0122, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0063, 0.0046, 0.0041, 0.0069], device='cuda:0')
+2023-03-03 08:17:14,789 INFO [train.py:968] (0/2) Epoch 6, batch 28700, giga_loss[loss=0.3364, simple_loss=0.3909, pruned_loss=0.141, over 28851.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3896, pruned_loss=0.1408, over 5654213.79 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3918, pruned_loss=0.1414, over 5697234.22 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3902, pruned_loss=0.1412, over 5641686.30 frames. ], batch size: 199, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:17:32,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.999e+02 1.535e+03 1.949e+03 2.680e+03 1.173e+04, threshold=3.897e+03, percent-clipped=12.0
+2023-03-03 08:17:33,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:18:04,595 INFO [train.py:968] (0/2) Epoch 6, batch 28750, libri_loss[loss=0.3253, simple_loss=0.3776, pruned_loss=0.1365, over 29351.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3906, pruned_loss=0.1416, over 5658895.33 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3919, pruned_loss=0.1416, over 5701423.90 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.391, pruned_loss=0.1417, over 5644298.09 frames. ], batch size: 71, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:18:10,330 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 08:18:15,524 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:18:18,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:18:47,542 INFO [train.py:968] (0/2) Epoch 6, batch 28800, libri_loss[loss=0.3066, simple_loss=0.3618, pruned_loss=0.1257, over 29543.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3933, pruned_loss=0.1439, over 5675729.19 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3914, pruned_loss=0.1411, over 5708669.14 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3941, pruned_loss=0.1445, over 5656388.79 frames. ], batch size: 78, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:19:03,338 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:19:07,679 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.473e+02 1.805e+03 2.597e+03 3.208e+03 9.424e+03, threshold=5.193e+03, percent-clipped=19.0
+2023-03-03 08:19:37,635 INFO [train.py:968] (0/2) Epoch 6, batch 28850, giga_loss[loss=0.4409, simple_loss=0.4608, pruned_loss=0.2105, over 28714.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3937, pruned_loss=0.1444, over 5675570.72 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3908, pruned_loss=0.1407, over 5712822.98 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3949, pruned_loss=0.1452, over 5655816.82 frames. ], batch size: 284, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:20:21,308 INFO [train.py:968] (0/2) Epoch 6, batch 28900, giga_loss[loss=0.3549, simple_loss=0.4039, pruned_loss=0.153, over 28906.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3949, pruned_loss=0.1459, over 5689143.73 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3903, pruned_loss=0.1404, over 5718709.28 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3964, pruned_loss=0.147, over 5667177.35 frames. ], batch size: 227, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:20:38,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.730e+02 1.749e+03 2.262e+03 3.072e+03 8.728e+03, threshold=4.525e+03, percent-clipped=8.0
+2023-03-03 08:20:43,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2218, 1.4915, 1.2644, 0.9860], device='cuda:0'), covar=tensor([0.2301, 0.2116, 0.2290, 0.1846], device='cuda:0'), in_proj_covar=tensor([0.1154, 0.0890, 0.1024, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:21:09,416 INFO [train.py:968] (0/2) Epoch 6, batch 28950, giga_loss[loss=0.3303, simple_loss=0.3894, pruned_loss=0.1356, over 28535.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3937, pruned_loss=0.1444, over 5684068.44 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3905, pruned_loss=0.1404, over 5717791.64 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3948, pruned_loss=0.1453, over 5667064.64 frames. ], batch size: 336, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:21:24,473 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:21:30,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7103, 1.5313, 1.2633, 1.2534], device='cuda:0'), covar=tensor([0.0580, 0.0581, 0.0822, 0.0965], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0458, 0.0511, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 08:21:56,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4021, 1.9752, 1.4368, 0.6008], device='cuda:0'), covar=tensor([0.2043, 0.1293, 0.1971, 0.2604], device='cuda:0'), in_proj_covar=tensor([0.1456, 0.1370, 0.1404, 0.1195], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 08:21:57,665 INFO [train.py:968] (0/2) Epoch 6, batch 29000, libri_loss[loss=0.3856, simple_loss=0.4345, pruned_loss=0.1684, over 28552.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3955, pruned_loss=0.1456, over 5677240.59 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3909, pruned_loss=0.1406, over 5721346.37 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3961, pruned_loss=0.1462, over 5659703.94 frames. ], batch size: 106, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:22:09,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:22:12,490 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-03 08:22:12,679 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.377e+02 1.475e+03 1.826e+03 2.777e+03 6.413e+03, threshold=3.652e+03, percent-clipped=3.0
+2023-03-03 08:22:40,029 INFO [train.py:968] (0/2) Epoch 6, batch 29050, giga_loss[loss=0.3085, simple_loss=0.3757, pruned_loss=0.1207, over 28966.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3943, pruned_loss=0.1442, over 5680440.29 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3901, pruned_loss=0.1401, over 5718386.98 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3958, pruned_loss=0.1454, over 5666677.11 frames. ], batch size: 128, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:22:44,326 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:22:45,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-03 08:23:11,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5742, 1.6459, 1.4482, 1.9242], device='cuda:0'), covar=tensor([0.1742, 0.1611, 0.1455, 0.1614], device='cuda:0'), in_proj_covar=tensor([0.1159, 0.0893, 0.1030, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:23:21,567 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:23:26,094 INFO [train.py:968] (0/2) Epoch 6, batch 29100, giga_loss[loss=0.3668, simple_loss=0.4195, pruned_loss=0.1571, over 29007.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3962, pruned_loss=0.1459, over 5669595.40 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3905, pruned_loss=0.1405, over 5711632.79 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3971, pruned_loss=0.1466, over 5663472.88 frames. ], batch size: 213, lr: 5.28e-03, grad_scale: 2.0
+2023-03-03 08:23:42,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.403e+02 1.613e+03 2.346e+03 3.735e+03 1.256e+04, threshold=4.693e+03, percent-clipped=25.0
+2023-03-03 08:23:58,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:23:58,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9726, 1.9986, 1.8791, 1.8268], device='cuda:0'), covar=tensor([0.1023, 0.1496, 0.1352, 0.1292], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0746, 0.0651, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 08:24:00,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:09,756 INFO [train.py:968] (0/2) Epoch 6, batch 29150, giga_loss[loss=0.3298, simple_loss=0.3921, pruned_loss=0.1338, over 28813.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3979, pruned_loss=0.1473, over 5671067.68 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3911, pruned_loss=0.141, over 5714401.03 frames. ], giga_tot_loss[loss=0.3467, simple_loss=0.3982, pruned_loss=0.1476, over 5663015.02 frames. ], batch size: 119, lr: 5.28e-03, grad_scale: 2.0
+2023-03-03 08:24:11,927 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:14,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:18,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:34,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0463, 1.1555, 3.5278, 3.0852], device='cuda:0'), covar=tensor([0.1532, 0.2367, 0.0441, 0.1217], device='cuda:0'), in_proj_covar=tensor([0.0588, 0.0545, 0.0785, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:24:42,937 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:43,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:53,778 INFO [train.py:968] (0/2) Epoch 6, batch 29200, libri_loss[loss=0.3464, simple_loss=0.4035, pruned_loss=0.1447, over 29201.00 frames. ], tot_loss[loss=0.3447, simple_loss=0.3971, pruned_loss=0.1462, over 5659646.06 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3909, pruned_loss=0.1408, over 5707238.70 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.3975, pruned_loss=0.1467, over 5658927.47 frames. ], batch size: 97, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:25:11,786 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-03 08:25:14,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.484e+02 1.522e+03 2.079e+03 2.827e+03 8.332e+03, threshold=4.157e+03, percent-clipped=7.0
+2023-03-03 08:25:32,941 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:25:36,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:25:45,022 INFO [train.py:968] (0/2) Epoch 6, batch 29250, giga_loss[loss=0.3744, simple_loss=0.425, pruned_loss=0.1619, over 28723.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3969, pruned_loss=0.1454, over 5644071.97 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3905, pruned_loss=0.1405, over 5707506.74 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3978, pruned_loss=0.1462, over 5642187.31 frames. ], batch size: 242, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:26:04,247 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:16,753 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:19,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:19,056 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:22,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:32,126 INFO [train.py:968] (0/2) Epoch 6, batch 29300, giga_loss[loss=0.2986, simple_loss=0.3701, pruned_loss=0.1136, over 28706.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3965, pruned_loss=0.1446, over 5647673.83 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3906, pruned_loss=0.1405, over 5710328.53 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.3973, pruned_loss=0.1452, over 5642958.75 frames. ], batch size: 307, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:26:43,162 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:46,512 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:49,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.524e+03 2.007e+03 2.824e+03 7.761e+03, threshold=4.014e+03, percent-clipped=9.0
+2023-03-03 08:26:56,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6376, 2.0038, 1.9582, 1.6797], device='cuda:0'), covar=tensor([0.1618, 0.1788, 0.1180, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0734, 0.0807, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 08:26:59,478 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:27:01,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:27:08,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:27:17,067 INFO [train.py:968] (0/2) Epoch 6, batch 29350, giga_loss[loss=0.3215, simple_loss=0.3769, pruned_loss=0.1331, over 28521.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3949, pruned_loss=0.1431, over 5653282.67 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3905, pruned_loss=0.1405, over 5706225.65 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3957, pruned_loss=0.1436, over 5651839.93 frames. ], batch size: 85, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:27:27,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:27:36,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0373, 1.8191, 1.8875, 1.7059], device='cuda:0'), covar=tensor([0.1143, 0.2126, 0.1496, 0.1656], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0737, 0.0642, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 08:28:01,558 INFO [train.py:968] (0/2) Epoch 6, batch 29400, giga_loss[loss=0.3621, simple_loss=0.4106, pruned_loss=0.1568, over 28711.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3943, pruned_loss=0.1428, over 5656880.37 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3904, pruned_loss=0.1405, over 5708237.47 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3951, pruned_loss=0.1433, over 5653720.28 frames. ], batch size: 242, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:28:21,840 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.603e+03 1.862e+03 2.535e+03 7.191e+03, threshold=3.724e+03, percent-clipped=7.0
+2023-03-03 08:28:30,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:28:52,148 INFO [train.py:968] (0/2) Epoch 6, batch 29450, giga_loss[loss=0.3261, simple_loss=0.3861, pruned_loss=0.133, over 28944.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3948, pruned_loss=0.143, over 5648873.87 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3904, pruned_loss=0.1403, over 5702460.66 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3955, pruned_loss=0.1436, over 5650210.79 frames. ], batch size: 164, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:29:26,348 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:29:28,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:29:39,898 INFO [train.py:968] (0/2) Epoch 6, batch 29500, giga_loss[loss=0.346, simple_loss=0.4025, pruned_loss=0.1448, over 28882.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3944, pruned_loss=0.143, over 5646530.00 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3899, pruned_loss=0.14, over 5697359.83 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3955, pruned_loss=0.1438, over 5651356.13 frames. ], batch size: 186, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:29:48,168 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257220.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:29:57,989 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:30:00,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.824e+02 1.435e+03 1.931e+03 2.847e+03 6.010e+03, threshold=3.862e+03, percent-clipped=7.0
+2023-03-03 08:30:05,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:30:24,480 INFO [train.py:968] (0/2) Epoch 6, batch 29550, giga_loss[loss=0.412, simple_loss=0.4341, pruned_loss=0.1949, over 27583.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3936, pruned_loss=0.1433, over 5654917.41 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.39, pruned_loss=0.1401, over 5703598.26 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3945, pruned_loss=0.144, over 5651513.04 frames. ], batch size: 472, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:30:28,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3426, 1.7124, 1.4139, 1.6074], device='cuda:0'), covar=tensor([0.0716, 0.0282, 0.0300, 0.0763], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0121, 0.0122, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0063, 0.0047, 0.0041, 0.0070], device='cuda:0')
+2023-03-03 08:30:36,942 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257276.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:30:42,284 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:30:44,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:30:51,238 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-03 08:31:05,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6638, 1.9475, 1.9538, 1.6691], device='cuda:0'), covar=tensor([0.1610, 0.1950, 0.1168, 0.1380], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0739, 0.0805, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 08:31:06,240 INFO [train.py:968] (0/2) Epoch 6, batch 29600, giga_loss[loss=0.3256, simple_loss=0.3939, pruned_loss=0.1286, over 28993.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3943, pruned_loss=0.1438, over 5646176.84 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3907, pruned_loss=0.1407, over 5679077.76 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3946, pruned_loss=0.1439, over 5664119.92 frames. ], batch size: 155, lr: 5.28e-03, grad_scale: 8.0
+2023-03-03 08:31:11,102 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:31:24,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.083e+02 1.610e+03 2.177e+03 3.400e+03 1.021e+04, threshold=4.354e+03, percent-clipped=17.0
+2023-03-03 08:31:54,447 INFO [train.py:968] (0/2) Epoch 6, batch 29650, giga_loss[loss=0.3791, simple_loss=0.4143, pruned_loss=0.172, over 28883.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3961, pruned_loss=0.1454, over 5649860.76 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3903, pruned_loss=0.1404, over 5684671.17 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.3967, pruned_loss=0.1459, over 5658314.71 frames. ], batch size: 106, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:32:14,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:32:18,839 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:32:39,679 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:32:44,404 INFO [train.py:968] (0/2) Epoch 6, batch 29700, giga_loss[loss=0.2842, simple_loss=0.3569, pruned_loss=0.1058, over 28376.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3961, pruned_loss=0.1459, over 5637888.12 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3898, pruned_loss=0.1402, over 5686868.65 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3972, pruned_loss=0.1466, over 5642128.09 frames. ], batch size: 71, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:32:46,477 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:33:04,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.548e+03 2.020e+03 2.824e+03 8.080e+03, threshold=4.040e+03, percent-clipped=6.0
+2023-03-03 08:33:31,396 INFO [train.py:968] (0/2) Epoch 6, batch 29750, giga_loss[loss=0.33, simple_loss=0.3889, pruned_loss=0.1356, over 28603.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3958, pruned_loss=0.1453, over 5640798.96 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3902, pruned_loss=0.1404, over 5687590.30 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3963, pruned_loss=0.1456, over 5643229.14 frames. ], batch size: 307, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:33:54,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5016, 4.2893, 4.0579, 1.8757], device='cuda:0'), covar=tensor([0.0558, 0.0774, 0.0997, 0.2015], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0886, 0.0827, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0009], device='cuda:0')
+2023-03-03 08:34:13,679 INFO [train.py:968] (0/2) Epoch 6, batch 29800, giga_loss[loss=0.3146, simple_loss=0.3828, pruned_loss=0.1231, over 28736.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3957, pruned_loss=0.1445, over 5653280.12 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3897, pruned_loss=0.14, over 5688849.33 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3969, pruned_loss=0.1454, over 5651717.22 frames. ], batch size: 262, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:34:34,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.638e+02 1.497e+03 2.071e+03 2.756e+03 1.257e+04, threshold=4.141e+03, percent-clipped=7.0
+2023-03-03 08:34:47,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4845, 2.6438, 1.6559, 1.5871], device='cuda:0'), covar=tensor([0.0661, 0.0307, 0.0588, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0497, 0.0320, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 08:34:57,475 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:35:01,125 INFO [train.py:968] (0/2) Epoch 6, batch 29850, giga_loss[loss=0.3139, simple_loss=0.3742, pruned_loss=0.1268, over 28910.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3966, pruned_loss=0.1449, over 5654151.67 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3901, pruned_loss=0.1402, over 5691700.69 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.3973, pruned_loss=0.1456, over 5649651.30 frames. ], batch size: 112, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:35:12,434 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 08:35:33,561 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257595.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:35:47,585 INFO [train.py:968] (0/2) Epoch 6, batch 29900, giga_loss[loss=0.3526, simple_loss=0.4085, pruned_loss=0.1483, over 28921.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3958, pruned_loss=0.144, over 5654281.92 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3904, pruned_loss=0.1405, over 5685823.60 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3961, pruned_loss=0.1443, over 5654505.42 frames. ], batch size: 213, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:35:51,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:36:05,878 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.979e+02 1.502e+03 1.826e+03 2.745e+03 7.280e+03, threshold=3.652e+03, percent-clipped=9.0
+2023-03-03 08:36:08,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:36:22,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257651.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:36:33,079 INFO [train.py:968] (0/2) Epoch 6, batch 29950, giga_loss[loss=0.2978, simple_loss=0.365, pruned_loss=0.1153, over 28609.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3945, pruned_loss=0.1434, over 5653359.99 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3905, pruned_loss=0.1405, over 5678658.52 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3948, pruned_loss=0.1437, over 5658804.14 frames. ], batch size: 262, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:37:16,946 INFO [train.py:968] (0/2) Epoch 6, batch 30000, giga_loss[loss=0.2898, simple_loss=0.3444, pruned_loss=0.1176, over 28300.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3902, pruned_loss=0.1406, over 5655118.00 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3909, pruned_loss=0.1409, over 5672776.06 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3903, pruned_loss=0.1406, over 5663116.60 frames. ], batch size: 77, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:37:16,950 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 08:37:23,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2984, 1.9207, 1.4327, 0.4073], device='cuda:0'), covar=tensor([0.2339, 0.1716, 0.2633, 0.3039], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1353, 0.1394, 0.1193], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 08:37:26,866 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2298, simple_loss=0.3348, pruned_loss=0.06244, over 944034.00 frames. 
+2023-03-03 08:37:26,866 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 08:37:48,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.654e+02 1.707e+03 2.256e+03 3.803e+03 1.509e+04, threshold=4.511e+03, percent-clipped=28.0
+2023-03-03 08:37:54,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257738.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:37:58,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257741.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:38:14,614 INFO [train.py:968] (0/2) Epoch 6, batch 30050, giga_loss[loss=0.2529, simple_loss=0.3214, pruned_loss=0.09221, over 28611.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3862, pruned_loss=0.1389, over 5646386.67 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3908, pruned_loss=0.1408, over 5678764.51 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3863, pruned_loss=0.139, over 5646962.31 frames. ], batch size: 92, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:38:21,455 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257770.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:38:30,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:38:42,511 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257794.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:38:44,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257797.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:38:57,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6196, 3.3680, 2.0651, 1.9947], device='cuda:0'), covar=tensor([0.0936, 0.0475, 0.0731, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.1492, 0.1347, 0.1291, 0.1389], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 08:38:58,232 INFO [train.py:968] (0/2) Epoch 6, batch 30100, giga_loss[loss=0.2911, simple_loss=0.3475, pruned_loss=0.1174, over 28832.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3866, pruned_loss=0.1399, over 5653009.18 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3915, pruned_loss=0.1413, over 5674337.95 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3859, pruned_loss=0.1395, over 5656021.79 frames. ], batch size: 99, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:39:15,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257826.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:39:20,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.236e+02 1.522e+03 1.937e+03 2.520e+03 5.731e+03, threshold=3.874e+03, percent-clipped=4.0
+2023-03-03 08:39:46,512 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:39:47,076 INFO [train.py:968] (0/2) Epoch 6, batch 30150, giga_loss[loss=0.3739, simple_loss=0.4047, pruned_loss=0.1716, over 26686.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3856, pruned_loss=0.14, over 5639800.88 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3918, pruned_loss=0.1414, over 5677575.04 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3848, pruned_loss=0.1395, over 5638970.86 frames. ], batch size: 555, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:39:49,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7105, 1.1459, 2.8818, 2.6973], device='cuda:0'), covar=tensor([0.1629, 0.2194, 0.0524, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0593, 0.0551, 0.0790, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:40:02,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7752, 2.4656, 2.3250, 2.1100], device='cuda:0'), covar=tensor([0.0973, 0.1713, 0.1288, 0.1500], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0732, 0.0643, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 08:40:06,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3276, 2.0186, 1.4382, 0.5340], device='cuda:0'), covar=tensor([0.2376, 0.1415, 0.2145, 0.2401], device='cuda:0'), in_proj_covar=tensor([0.1443, 0.1349, 0.1391, 0.1193], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 08:40:21,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1353, 1.6700, 1.5216, 1.2672], device='cuda:0'), covar=tensor([0.1572, 0.2241, 0.1348, 0.1524], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0738, 0.0807, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 08:40:36,125 INFO [train.py:968] (0/2) Epoch 6, batch 30200, giga_loss[loss=0.3522, simple_loss=0.4052, pruned_loss=0.1497, over 27631.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3841, pruned_loss=0.1366, over 5649776.38 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3913, pruned_loss=0.1411, over 5682305.52 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3838, pruned_loss=0.1364, over 5644550.73 frames. ], batch size: 472, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:40:48,820 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:40:52,801 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:40:56,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.851e+02 1.568e+03 2.044e+03 3.047e+03 7.374e+03, threshold=4.088e+03, percent-clipped=15.0
+2023-03-03 08:40:56,686 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:41:18,926 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:41:22,804 INFO [train.py:968] (0/2) Epoch 6, batch 30250, giga_loss[loss=0.3185, simple_loss=0.3882, pruned_loss=0.1244, over 28291.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3807, pruned_loss=0.1329, over 5640431.88 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3904, pruned_loss=0.1408, over 5682701.95 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3808, pruned_loss=0.1327, over 5634272.51 frames. ], batch size: 368, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:41:32,479 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257971.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:41:52,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:41:59,974 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-258000.pt
+2023-03-03 08:42:08,191 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:42:09,881 INFO [train.py:968] (0/2) Epoch 6, batch 30300, giga_loss[loss=0.2948, simple_loss=0.3616, pruned_loss=0.1139, over 27620.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3771, pruned_loss=0.129, over 5655609.29 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3892, pruned_loss=0.1404, over 5687117.58 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3779, pruned_loss=0.1289, over 5645309.74 frames. ], batch size: 472, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:42:31,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.962e+02 1.465e+03 1.909e+03 2.515e+03 8.512e+03, threshold=3.818e+03, percent-clipped=9.0
+2023-03-03 08:42:46,352 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-03 08:43:00,008 INFO [train.py:968] (0/2) Epoch 6, batch 30350, giga_loss[loss=0.2774, simple_loss=0.3515, pruned_loss=0.1016, over 28591.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.374, pruned_loss=0.1255, over 5645010.46 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3889, pruned_loss=0.1403, over 5676796.72 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3746, pruned_loss=0.1253, over 5644585.21 frames. ], batch size: 307, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:43:13,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:43:15,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:43:41,158 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:43:43,817 INFO [train.py:968] (0/2) Epoch 6, batch 30400, giga_loss[loss=0.3062, simple_loss=0.3603, pruned_loss=0.1261, over 26744.00 frames. ], tot_loss[loss=0.307, simple_loss=0.37, pruned_loss=0.122, over 5643141.14 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.388, pruned_loss=0.1399, over 5672911.21 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3708, pruned_loss=0.1216, over 5644661.49 frames. ], batch size: 555, lr: 5.27e-03, grad_scale: 8.0
+2023-03-03 08:44:01,387 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:44:01,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.890e+02 1.239e+03 1.663e+03 2.142e+03 6.562e+03, threshold=3.326e+03, percent-clipped=7.0
+2023-03-03 08:44:05,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:44:25,040 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:44:27,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:44:33,369 INFO [train.py:968] (0/2) Epoch 6, batch 30450, giga_loss[loss=0.2827, simple_loss=0.3588, pruned_loss=0.1033, over 28537.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3681, pruned_loss=0.1176, over 5666603.38 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3875, pruned_loss=0.1399, over 5678695.75 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3687, pruned_loss=0.1169, over 5662458.52 frames. ], batch size: 307, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:44:36,922 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:44:58,222 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:45:25,982 INFO [train.py:968] (0/2) Epoch 6, batch 30500, giga_loss[loss=0.2634, simple_loss=0.3499, pruned_loss=0.08842, over 28953.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3682, pruned_loss=0.1173, over 5667794.82 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3872, pruned_loss=0.1396, over 5680127.12 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3687, pruned_loss=0.1165, over 5662927.94 frames. ], batch size: 164, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:45:54,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.877e+02 1.438e+03 1.743e+03 2.704e+03 4.800e+03, threshold=3.485e+03, percent-clipped=15.0
+2023-03-03 08:45:54,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:46:12,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:46:21,108 INFO [train.py:968] (0/2) Epoch 6, batch 30550, giga_loss[loss=0.2631, simple_loss=0.3443, pruned_loss=0.09094, over 28958.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.366, pruned_loss=0.1153, over 5657843.31 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3872, pruned_loss=0.1397, over 5671382.48 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3663, pruned_loss=0.1146, over 5661204.55 frames. ], batch size: 136, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:46:37,760 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:47:11,336 INFO [train.py:968] (0/2) Epoch 6, batch 30600, giga_loss[loss=0.2825, simple_loss=0.3577, pruned_loss=0.1036, over 28864.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3628, pruned_loss=0.1129, over 5661800.03 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3872, pruned_loss=0.1398, over 5674820.86 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3628, pruned_loss=0.1119, over 5661075.49 frames. ], batch size: 174, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:47:34,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.902e+02 1.256e+03 1.862e+03 2.753e+03 5.263e+03, threshold=3.723e+03, percent-clipped=12.0
+2023-03-03 08:47:44,008 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:47:46,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258346.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:47:48,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4747, 4.3298, 4.0587, 1.8646], device='cuda:0'), covar=tensor([0.0463, 0.0648, 0.0847, 0.1998], device='cuda:0'), in_proj_covar=tensor([0.0907, 0.0857, 0.0790, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 08:47:49,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5246, 3.6113, 1.6199, 1.4932], device='cuda:0'), covar=tensor([0.0843, 0.0326, 0.0825, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0491, 0.0318, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 08:48:01,302 INFO [train.py:968] (0/2) Epoch 6, batch 30650, giga_loss[loss=0.2863, simple_loss=0.3612, pruned_loss=0.1056, over 28584.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3613, pruned_loss=0.112, over 5659561.88 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3866, pruned_loss=0.1396, over 5679347.44 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3612, pruned_loss=0.1108, over 5654861.50 frames. ], batch size: 307, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:48:17,895 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:48:20,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:48:48,891 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:48:49,314 INFO [train.py:968] (0/2) Epoch 6, batch 30700, giga_loss[loss=0.2817, simple_loss=0.3586, pruned_loss=0.1024, over 28832.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3611, pruned_loss=0.1117, over 5665579.70 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3862, pruned_loss=0.1394, over 5681452.66 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3611, pruned_loss=0.1106, over 5659835.87 frames. ], batch size: 92, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:49:09,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.015e+02 1.381e+03 1.986e+03 2.907e+03 1.002e+04, threshold=3.972e+03, percent-clipped=7.0
+2023-03-03 08:49:34,108 INFO [train.py:968] (0/2) Epoch 6, batch 30750, giga_loss[loss=0.2684, simple_loss=0.3477, pruned_loss=0.09458, over 28276.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3588, pruned_loss=0.1101, over 5663786.77 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3855, pruned_loss=0.1392, over 5686174.30 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3586, pruned_loss=0.1085, over 5654208.53 frames. ], batch size: 368, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:50:03,972 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258489.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:50:07,862 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258492.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:50:28,312 INFO [train.py:968] (0/2) Epoch 6, batch 30800, giga_loss[loss=0.2539, simple_loss=0.3305, pruned_loss=0.08863, over 28548.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3559, pruned_loss=0.1074, over 5666364.22 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3854, pruned_loss=0.1392, over 5687145.43 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3557, pruned_loss=0.106, over 5658005.87 frames. ], batch size: 336, lr: 5.27e-03, grad_scale: 8.0
+2023-03-03 08:50:38,770 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258521.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:50:51,958 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.660e+02 1.366e+03 1.831e+03 2.612e+03 6.329e+03, threshold=3.662e+03, percent-clipped=6.0
+2023-03-03 08:51:19,886 INFO [train.py:968] (0/2) Epoch 6, batch 30850, giga_loss[loss=0.247, simple_loss=0.3253, pruned_loss=0.08438, over 28455.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3518, pruned_loss=0.1051, over 5672135.04 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3849, pruned_loss=0.1389, over 5686759.11 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3517, pruned_loss=0.1038, over 5665496.64 frames. ], batch size: 60, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:51:56,409 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3731, 1.7378, 1.6203, 1.4396], device='cuda:0'), covar=tensor([0.1447, 0.1895, 0.1195, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0719, 0.0799, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 08:52:05,552 INFO [train.py:968] (0/2) Epoch 6, batch 30900, giga_loss[loss=0.2814, simple_loss=0.3497, pruned_loss=0.1065, over 27964.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3504, pruned_loss=0.1052, over 5662915.25 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.3837, pruned_loss=0.1383, over 5682888.66 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3503, pruned_loss=0.1037, over 5661038.91 frames. ], batch size: 412, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:52:06,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2749, 1.3082, 1.1409, 1.0431], device='cuda:0'), covar=tensor([0.0600, 0.0407, 0.0858, 0.0750], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0446, 0.0503, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 08:52:20,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:52:28,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.543e+02 1.301e+03 1.640e+03 2.410e+03 7.782e+03, threshold=3.279e+03, percent-clipped=6.0
+2023-03-03 08:52:28,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1909, 1.4153, 1.1115, 1.0074], device='cuda:0'), covar=tensor([0.0963, 0.0875, 0.0591, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.1484, 0.1298, 0.1248, 0.1346], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 08:52:43,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7583, 1.7673, 1.4029, 2.0437], device='cuda:0'), covar=tensor([0.2174, 0.2087, 0.2268, 0.2062], device='cuda:0'), in_proj_covar=tensor([0.1145, 0.0872, 0.1021, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 08:52:46,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:52:56,056 INFO [train.py:968] (0/2) Epoch 6, batch 30950, libri_loss[loss=0.3549, simple_loss=0.3921, pruned_loss=0.1588, over 19294.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3499, pruned_loss=0.1055, over 5639221.74 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3832, pruned_loss=0.1382, over 5669987.08 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3495, pruned_loss=0.1036, over 5649160.64 frames. ], batch size: 188, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:53:14,627 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0883, 1.3690, 1.1079, 1.0284], device='cuda:0'), covar=tensor([0.2191, 0.1996, 0.2186, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.1150, 0.0873, 0.1024, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:53:50,068 INFO [train.py:968] (0/2) Epoch 6, batch 31000, giga_loss[loss=0.2716, simple_loss=0.3492, pruned_loss=0.09702, over 27605.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3532, pruned_loss=0.1074, over 5635229.95 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3831, pruned_loss=0.1382, over 5667511.10 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3521, pruned_loss=0.105, over 5645549.78 frames. ], batch size: 472, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:54:00,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:54:18,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.233e+02 1.415e+03 1.834e+03 2.918e+03 8.173e+03, threshold=3.668e+03, percent-clipped=15.0
+2023-03-03 08:54:37,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3501, 1.6528, 1.1421, 1.1441], device='cuda:0'), covar=tensor([0.1268, 0.0875, 0.0869, 0.1085], device='cuda:0'), in_proj_covar=tensor([0.1503, 0.1305, 0.1254, 0.1355], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 08:54:45,444 INFO [train.py:968] (0/2) Epoch 6, batch 31050, libri_loss[loss=0.2781, simple_loss=0.3311, pruned_loss=0.1125, over 29671.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3544, pruned_loss=0.1072, over 5633276.78 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3823, pruned_loss=0.1378, over 5673455.20 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3534, pruned_loss=0.1047, over 5635039.77 frames. ], batch size: 73, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:54:56,485 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:55:00,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:55:25,977 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:55:29,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:55:38,382 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:55:52,229 INFO [train.py:968] (0/2) Epoch 6, batch 31100, giga_loss[loss=0.2929, simple_loss=0.3367, pruned_loss=0.1245, over 24297.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3547, pruned_loss=0.1074, over 5626925.25 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3823, pruned_loss=0.1379, over 5676593.52 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3536, pruned_loss=0.105, over 5625211.06 frames. ], batch size: 705, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:56:07,499 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:56:21,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.483e+02 1.456e+03 1.829e+03 2.448e+03 6.505e+03, threshold=3.657e+03, percent-clipped=12.0
+2023-03-03 08:56:36,837 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:56:51,736 INFO [train.py:968] (0/2) Epoch 6, batch 31150, giga_loss[loss=0.2995, simple_loss=0.3589, pruned_loss=0.1201, over 26762.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3526, pruned_loss=0.1061, over 5628366.21 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3819, pruned_loss=0.1378, over 5667668.47 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3512, pruned_loss=0.1033, over 5633763.54 frames. ], batch size: 555, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:56:52,374 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:56:56,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:57:28,558 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:57:42,606 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:57:52,036 INFO [train.py:968] (0/2) Epoch 6, batch 31200, giga_loss[loss=0.2462, simple_loss=0.3327, pruned_loss=0.07983, over 27597.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3501, pruned_loss=0.1035, over 5633726.10 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3805, pruned_loss=0.1369, over 5673623.17 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3495, pruned_loss=0.1013, over 5632263.06 frames. ], batch size: 472, lr: 5.26e-03, grad_scale: 8.0
+2023-03-03 08:58:12,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2601, 1.6964, 1.2487, 0.4175], device='cuda:0'), covar=tensor([0.2084, 0.1318, 0.2376, 0.3053], device='cuda:0'), in_proj_covar=tensor([0.1430, 0.1345, 0.1393, 0.1190], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 08:58:16,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4067, 1.4939, 1.2599, 1.6246], device='cuda:0'), covar=tensor([0.2370, 0.2112, 0.2287, 0.2118], device='cuda:0'), in_proj_covar=tensor([0.1152, 0.0875, 0.1023, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 08:58:21,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7438, 1.7439, 1.2700, 1.3913], device='cuda:0'), covar=tensor([0.0699, 0.0506, 0.0970, 0.0914], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0438, 0.0494, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 08:58:25,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.106e+02 1.225e+03 1.674e+03 2.150e+03 6.216e+03, threshold=3.348e+03, percent-clipped=7.0
+2023-03-03 08:58:37,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1779, 1.4582, 1.1436, 0.9349], device='cuda:0'), covar=tensor([0.1219, 0.0991, 0.0698, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.1488, 0.1297, 0.1245, 0.1352], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 08:58:50,202 INFO [train.py:968] (0/2) Epoch 6, batch 31250, giga_loss[loss=0.2724, simple_loss=0.3476, pruned_loss=0.09859, over 28859.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3493, pruned_loss=0.1025, over 5645133.55 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3808, pruned_loss=0.1375, over 5679483.31 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3476, pruned_loss=0.09922, over 5637488.32 frames. ], batch size: 227, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:59:47,727 INFO [train.py:968] (0/2) Epoch 6, batch 31300, giga_loss[loss=0.2967, simple_loss=0.3582, pruned_loss=0.1176, over 28793.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3461, pruned_loss=0.1015, over 5646740.92 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3803, pruned_loss=0.1371, over 5672945.13 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3445, pruned_loss=0.0984, over 5645366.59 frames. ], batch size: 243, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:00:20,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.981e+02 1.279e+03 1.809e+03 2.467e+03 1.366e+04, threshold=3.618e+03, percent-clipped=12.0
+2023-03-03 09:00:39,826 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5775, 1.5244, 1.2328, 1.2046], device='cuda:0'), covar=tensor([0.0650, 0.0491, 0.0926, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0440, 0.0497, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 09:00:51,852 INFO [train.py:968] (0/2) Epoch 6, batch 31350, giga_loss[loss=0.2657, simple_loss=0.3364, pruned_loss=0.0975, over 28941.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3457, pruned_loss=0.1013, over 5651967.67 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3801, pruned_loss=0.1369, over 5664580.14 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3441, pruned_loss=0.09853, over 5657209.98 frames. ], batch size: 199, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:01:49,341 INFO [train.py:968] (0/2) Epoch 6, batch 31400, giga_loss[loss=0.3143, simple_loss=0.3672, pruned_loss=0.1307, over 26836.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.346, pruned_loss=0.1013, over 5658459.79 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3802, pruned_loss=0.1372, over 5666106.94 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3442, pruned_loss=0.09832, over 5661120.04 frames. ], batch size: 555, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:02:20,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.270e+02 1.234e+03 1.600e+03 2.352e+03 6.054e+03, threshold=3.200e+03, percent-clipped=6.0
+2023-03-03 09:02:32,434 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3142, 1.4676, 1.4815, 1.3889], device='cuda:0'), covar=tensor([0.1043, 0.1363, 0.1450, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0724, 0.0633, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 09:02:48,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4702, 1.8036, 1.2354, 0.7606], device='cuda:0'), covar=tensor([0.3179, 0.2042, 0.1964, 0.3000], device='cuda:0'), in_proj_covar=tensor([0.1418, 0.1340, 0.1375, 0.1178], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 09:02:50,529 INFO [train.py:968] (0/2) Epoch 6, batch 31450, giga_loss[loss=0.278, simple_loss=0.3343, pruned_loss=0.1109, over 24091.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.347, pruned_loss=0.1011, over 5646081.69 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3802, pruned_loss=0.1372, over 5667308.98 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3451, pruned_loss=0.09827, over 5646822.36 frames. ], batch size: 705, lr: 5.26e-03, grad_scale: 2.0
+2023-03-03 09:03:41,520 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-03 09:03:53,094 INFO [train.py:968] (0/2) Epoch 6, batch 31500, giga_loss[loss=0.2666, simple_loss=0.3427, pruned_loss=0.09528, over 29042.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3462, pruned_loss=0.1001, over 5651141.50 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.38, pruned_loss=0.1372, over 5662522.40 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3443, pruned_loss=0.09726, over 5655616.40 frames. ], batch size: 199, lr: 5.26e-03, grad_scale: 2.0
+2023-03-03 09:04:03,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:04:12,362 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-03 09:04:24,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.681e+02 1.211e+03 1.571e+03 2.320e+03 6.104e+03, threshold=3.141e+03, percent-clipped=9.0
+2023-03-03 09:04:53,495 INFO [train.py:968] (0/2) Epoch 6, batch 31550, giga_loss[loss=0.3169, simple_loss=0.3785, pruned_loss=0.1277, over 28426.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3452, pruned_loss=0.09995, over 5665010.23 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3798, pruned_loss=0.1372, over 5664040.75 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3425, pruned_loss=0.09614, over 5667646.76 frames. ], batch size: 336, lr: 5.26e-03, grad_scale: 2.0
+2023-03-03 09:05:18,610 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:05:56,465 INFO [train.py:968] (0/2) Epoch 6, batch 31600, giga_loss[loss=0.2607, simple_loss=0.3444, pruned_loss=0.08848, over 28637.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3469, pruned_loss=0.1014, over 5655045.01 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3793, pruned_loss=0.1371, over 5658319.20 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3439, pruned_loss=0.09709, over 5662423.37 frames. ], batch size: 307, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:06:29,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.513e+02 1.389e+03 1.856e+03 2.419e+03 6.032e+03, threshold=3.712e+03, percent-clipped=5.0
+2023-03-03 09:06:54,099 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:06:58,232 INFO [train.py:968] (0/2) Epoch 6, batch 31650, giga_loss[loss=0.311, simple_loss=0.3943, pruned_loss=0.1139, over 28883.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3511, pruned_loss=0.1017, over 5664197.65 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3789, pruned_loss=0.1369, over 5663264.36 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3484, pruned_loss=0.09766, over 5665316.69 frames. ], batch size: 227, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:07:03,274 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:07:05,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:07:37,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=259397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:07:55,101 INFO [train.py:968] (0/2) Epoch 6, batch 31700, giga_loss[loss=0.2905, simple_loss=0.367, pruned_loss=0.107, over 28016.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3532, pruned_loss=0.1022, over 5643515.86 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3781, pruned_loss=0.1369, over 5645792.41 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3505, pruned_loss=0.09725, over 5659369.58 frames. ], batch size: 412, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:08:09,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:08:13,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:08:25,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.869e+02 1.405e+03 1.860e+03 2.673e+03 9.447e+03, threshold=3.720e+03, percent-clipped=10.0
+2023-03-03 09:08:47,318 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=259454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:08:52,627 INFO [train.py:968] (0/2) Epoch 6, batch 31750, giga_loss[loss=0.243, simple_loss=0.3388, pruned_loss=0.07354, over 28758.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3543, pruned_loss=0.1012, over 5648595.41 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3779, pruned_loss=0.1366, over 5647829.77 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3517, pruned_loss=0.09686, over 5659333.50 frames. ], batch size: 119, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:09:54,012 INFO [train.py:968] (0/2) Epoch 6, batch 31800, giga_loss[loss=0.2726, simple_loss=0.357, pruned_loss=0.09407, over 28618.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3529, pruned_loss=0.09961, over 5656691.56 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3778, pruned_loss=0.1364, over 5649097.75 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3507, pruned_loss=0.09581, over 5664028.75 frames. ], batch size: 307, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:10:12,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1739, 1.4054, 1.3556, 1.3293], device='cuda:0'), covar=tensor([0.1071, 0.1389, 0.1597, 0.1276], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0725, 0.0636, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 09:10:24,194 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.618e+02 1.288e+03 1.632e+03 2.316e+03 5.760e+03, threshold=3.265e+03, percent-clipped=7.0
+2023-03-03 09:10:55,975 INFO [train.py:968] (0/2) Epoch 6, batch 31850, giga_loss[loss=0.2958, simple_loss=0.3634, pruned_loss=0.1141, over 29006.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3533, pruned_loss=0.101, over 5666124.68 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3778, pruned_loss=0.1366, over 5644470.90 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3512, pruned_loss=0.09729, over 5675714.99 frames. ], batch size: 285, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:11:33,170 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 09:12:12,646 INFO [train.py:968] (0/2) Epoch 6, batch 31900, giga_loss[loss=0.3437, simple_loss=0.3896, pruned_loss=0.1489, over 26871.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3527, pruned_loss=0.102, over 5663904.07 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.377, pruned_loss=0.1361, over 5649643.45 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3512, pruned_loss=0.09884, over 5667376.33 frames. ], batch size: 555, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:12:23,209 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:12:23,427 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 09:12:54,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.072e+02 1.270e+03 1.890e+03 2.995e+03 1.240e+04, threshold=3.779e+03, percent-clipped=20.0
+2023-03-03 09:13:05,288 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 09:13:21,651 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:13:25,535 INFO [train.py:968] (0/2) Epoch 6, batch 31950, giga_loss[loss=0.2588, simple_loss=0.3323, pruned_loss=0.09264, over 28814.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3521, pruned_loss=0.1021, over 5672601.97 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3773, pruned_loss=0.1362, over 5650519.92 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3502, pruned_loss=0.09882, over 5675099.86 frames. ], batch size: 227, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:13:58,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-03 09:14:39,709 INFO [train.py:968] (0/2) Epoch 6, batch 32000, giga_loss[loss=0.2675, simple_loss=0.3493, pruned_loss=0.0928, over 28646.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3477, pruned_loss=0.09905, over 5667535.57 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.3772, pruned_loss=0.1362, over 5648416.32 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3462, pruned_loss=0.09641, over 5671614.38 frames. ], batch size: 242, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:15:00,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2910, 1.7646, 1.6497, 1.4454], device='cuda:0'), covar=tensor([0.1656, 0.2117, 0.1324, 0.1524], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0722, 0.0805, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 09:15:08,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:15:13,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.946e+02 1.220e+03 1.555e+03 2.243e+03 7.101e+03, threshold=3.111e+03, percent-clipped=4.0
+2023-03-03 09:15:43,793 INFO [train.py:968] (0/2) Epoch 6, batch 32050, libri_loss[loss=0.3769, simple_loss=0.4076, pruned_loss=0.1731, over 19596.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3461, pruned_loss=0.09861, over 5663524.41 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3772, pruned_loss=0.1363, over 5642978.37 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3442, pruned_loss=0.09553, over 5672994.06 frames. ], batch size: 186, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:16:44,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:16:44,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4909, 3.7434, 1.6377, 1.5902], device='cuda:0'), covar=tensor([0.0791, 0.0295, 0.0765, 0.1177], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0484, 0.0318, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 09:16:48,071 INFO [train.py:968] (0/2) Epoch 6, batch 32100, giga_loss[loss=0.298, simple_loss=0.375, pruned_loss=0.1105, over 28442.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3474, pruned_loss=0.09981, over 5674525.29 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.3771, pruned_loss=0.1362, over 5646764.19 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3454, pruned_loss=0.09675, over 5679039.49 frames. ], batch size: 336, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:17:18,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.316e+02 1.424e+03 1.984e+03 2.702e+03 6.214e+03, threshold=3.968e+03, percent-clipped=16.0
+2023-03-03 09:17:42,779 INFO [train.py:968] (0/2) Epoch 6, batch 32150, giga_loss[loss=0.3174, simple_loss=0.3866, pruned_loss=0.1241, over 28751.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3516, pruned_loss=0.1023, over 5678962.35 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.3767, pruned_loss=0.1361, over 5649919.50 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3493, pruned_loss=0.09868, over 5680582.42 frames. ], batch size: 243, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:18:01,872 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:18:05,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:18:48,460 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=259908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:18:51,457 INFO [train.py:968] (0/2) Epoch 6, batch 32200, giga_loss[loss=0.3075, simple_loss=0.3681, pruned_loss=0.1235, over 27698.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3499, pruned_loss=0.1025, over 5680609.53 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3764, pruned_loss=0.1359, over 5651303.90 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3483, pruned_loss=0.09972, over 5680838.26 frames. ], batch size: 474, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:19:22,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.318e+02 1.478e+03 1.926e+03 2.959e+03 9.046e+03, threshold=3.852e+03, percent-clipped=9.0
+2023-03-03 09:19:52,208 INFO [train.py:968] (0/2) Epoch 6, batch 32250, giga_loss[loss=0.3092, simple_loss=0.3718, pruned_loss=0.1233, over 28430.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3511, pruned_loss=0.1044, over 5677829.13 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3763, pruned_loss=0.136, over 5654700.89 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3495, pruned_loss=0.1017, over 5675526.25 frames. ], batch size: 369, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:20:11,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 09:20:12,266 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-03 09:20:13,377 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-03 09:20:31,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:20:35,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3278, 1.6340, 1.2567, 1.0007], device='cuda:0'), covar=tensor([0.1307, 0.0934, 0.0716, 0.1060], device='cuda:0'), in_proj_covar=tensor([0.1505, 0.1308, 0.1248, 0.1382], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 09:20:40,833 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-260000.pt
+2023-03-03 09:20:52,330 INFO [train.py:968] (0/2) Epoch 6, batch 32300, giga_loss[loss=0.3032, simple_loss=0.3643, pruned_loss=0.1211, over 28681.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3516, pruned_loss=0.1046, over 5682819.23 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3761, pruned_loss=0.1358, over 5656414.31 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3497, pruned_loss=0.1016, over 5680291.05 frames. ], batch size: 92, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:21:12,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2256, 1.6697, 1.2500, 1.4902], device='cuda:0'), covar=tensor([0.0789, 0.0292, 0.0336, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0120, 0.0124, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 09:21:25,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:21:32,994 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.038e+02 1.350e+03 1.707e+03 2.286e+03 3.999e+03, threshold=3.415e+03, percent-clipped=1.0
+2023-03-03 09:22:07,962 INFO [train.py:968] (0/2) Epoch 6, batch 32350, giga_loss[loss=0.3061, simple_loss=0.3596, pruned_loss=0.1263, over 26933.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3532, pruned_loss=0.1039, over 5676823.49 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3763, pruned_loss=0.136, over 5656723.24 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3515, pruned_loss=0.1012, over 5674675.97 frames. ], batch size: 555, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:22:42,491 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-03 09:23:25,933 INFO [train.py:968] (0/2) Epoch 6, batch 32400, giga_loss[loss=0.3236, simple_loss=0.3671, pruned_loss=0.14, over 26803.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3518, pruned_loss=0.1028, over 5667493.22 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3762, pruned_loss=0.1359, over 5659910.62 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3502, pruned_loss=0.1003, over 5663344.14 frames. ], batch size: 555, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:24:02,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:24:02,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.894e+02 1.409e+03 1.744e+03 2.374e+03 5.773e+03, threshold=3.487e+03, percent-clipped=7.0
+2023-03-03 09:24:06,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:24:08,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260142.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:24:30,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4660, 1.9315, 1.8464, 1.5785], device='cuda:0'), covar=tensor([0.1615, 0.1951, 0.1246, 0.1418], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0718, 0.0804, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 09:24:34,797 INFO [train.py:968] (0/2) Epoch 6, batch 32450, giga_loss[loss=0.2581, simple_loss=0.3411, pruned_loss=0.08758, over 28863.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3472, pruned_loss=0.1009, over 5677259.80 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3763, pruned_loss=0.1362, over 5662422.34 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3456, pruned_loss=0.09843, over 5671887.72 frames. ], batch size: 174, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:24:47,355 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:24:55,988 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:24:56,089 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:24:58,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:25:03,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:25:16,258 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 09:25:20,694 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 09:25:40,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:25:43,249 INFO [train.py:968] (0/2) Epoch 6, batch 32500, giga_loss[loss=0.3149, simple_loss=0.3685, pruned_loss=0.1307, over 26911.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3408, pruned_loss=0.09816, over 5674280.15 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3762, pruned_loss=0.1361, over 5661238.09 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3394, pruned_loss=0.09596, over 5671336.28 frames. ], batch size: 555, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:26:21,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4878, 1.5943, 1.3408, 1.6898], device='cuda:0'), covar=tensor([0.2194, 0.2115, 0.2163, 0.1923], device='cuda:0'), in_proj_covar=tensor([0.1150, 0.0876, 0.1032, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 09:26:23,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.175e+02 1.452e+03 1.905e+03 2.766e+03 7.556e+03, threshold=3.810e+03, percent-clipped=19.0
+2023-03-03 09:26:42,606 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-03 09:26:50,118 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:26:51,732 INFO [train.py:968] (0/2) Epoch 6, batch 32550, giga_loss[loss=0.2577, simple_loss=0.3326, pruned_loss=0.09143, over 28528.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3402, pruned_loss=0.09798, over 5671987.79 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.376, pruned_loss=0.136, over 5662475.90 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3391, pruned_loss=0.09619, over 5668757.75 frames. ], batch size: 307, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:27:51,054 INFO [train.py:968] (0/2) Epoch 6, batch 32600, giga_loss[loss=0.2652, simple_loss=0.3439, pruned_loss=0.09325, over 28884.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3417, pruned_loss=0.09892, over 5675268.28 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3754, pruned_loss=0.1356, over 5666910.23 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3408, pruned_loss=0.09715, over 5668813.71 frames. ], batch size: 186, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:28:08,383 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:28:11,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:28:18,295 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-03 09:28:22,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.696e+02 1.501e+03 2.012e+03 3.162e+03 1.112e+04, threshold=4.024e+03, percent-clipped=12.0
+2023-03-03 09:28:46,285 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:28:49,192 INFO [train.py:968] (0/2) Epoch 6, batch 32650, giga_loss[loss=0.2385, simple_loss=0.312, pruned_loss=0.08252, over 27584.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3405, pruned_loss=0.09811, over 5675788.84 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3754, pruned_loss=0.1356, over 5672090.57 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3387, pruned_loss=0.09562, over 5666220.64 frames. ], batch size: 472, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:29:18,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3313, 1.9626, 1.3728, 1.5201], device='cuda:0'), covar=tensor([0.0778, 0.0292, 0.0327, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0120, 0.0124, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 09:29:51,096 INFO [train.py:968] (0/2) Epoch 6, batch 32700, giga_loss[loss=0.2605, simple_loss=0.3372, pruned_loss=0.09193, over 29087.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3389, pruned_loss=0.09647, over 5674100.23 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.3749, pruned_loss=0.1352, over 5675942.44 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3373, pruned_loss=0.09413, over 5662903.54 frames. ], batch size: 200, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:29:53,683 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 09:30:08,332 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:30:08,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 09:30:13,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1580, 1.3326, 3.6380, 3.0171], device='cuda:0'), covar=tensor([0.1502, 0.2148, 0.0432, 0.0724], device='cuda:0'), in_proj_covar=tensor([0.0570, 0.0532, 0.0755, 0.0621], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 09:30:30,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.239e+02 1.300e+03 1.754e+03 2.307e+03 6.388e+03, threshold=3.508e+03, percent-clipped=4.0
+2023-03-03 09:30:32,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2999, 1.6062, 1.2730, 0.9534], device='cuda:0'), covar=tensor([0.1107, 0.0934, 0.0601, 0.0908], device='cuda:0'), in_proj_covar=tensor([0.1492, 0.1292, 0.1242, 0.1367], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 09:30:59,552 INFO [train.py:968] (0/2) Epoch 6, batch 32750, giga_loss[loss=0.2392, simple_loss=0.3173, pruned_loss=0.08052, over 29009.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3377, pruned_loss=0.09621, over 5669060.03 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3748, pruned_loss=0.1351, over 5675399.50 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.336, pruned_loss=0.09394, over 5660655.44 frames. ], batch size: 199, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:32:08,387 INFO [train.py:968] (0/2) Epoch 6, batch 32800, giga_loss[loss=0.2841, simple_loss=0.3537, pruned_loss=0.1072, over 28107.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3389, pruned_loss=0.09613, over 5676689.07 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3749, pruned_loss=0.1352, over 5673553.14 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3369, pruned_loss=0.09365, over 5671893.86 frames. ], batch size: 412, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:32:18,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260517.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:32:50,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.897e+02 1.379e+03 1.827e+03 2.526e+03 5.614e+03, threshold=3.654e+03, percent-clipped=14.0
+2023-03-03 09:33:02,332 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:33:16,141 INFO [train.py:968] (0/2) Epoch 6, batch 32850, giga_loss[loss=0.3068, simple_loss=0.3655, pruned_loss=0.1241, over 28915.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3404, pruned_loss=0.09716, over 5679685.86 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3749, pruned_loss=0.1352, over 5675580.46 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3385, pruned_loss=0.0949, over 5674100.42 frames. ], batch size: 186, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:34:12,809 INFO [train.py:968] (0/2) Epoch 6, batch 32900, giga_loss[loss=0.2558, simple_loss=0.3324, pruned_loss=0.08961, over 28918.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3413, pruned_loss=0.09875, over 5685198.13 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3741, pruned_loss=0.1347, over 5680470.09 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3391, pruned_loss=0.09579, over 5676528.23 frames. ], batch size: 227, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:34:22,661 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260617.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 09:34:41,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:34:46,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.786e+02 1.321e+03 1.680e+03 2.214e+03 3.592e+03, threshold=3.360e+03, percent-clipped=0.0
+2023-03-03 09:35:14,348 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260660.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 09:35:15,530 INFO [train.py:968] (0/2) Epoch 6, batch 32950, giga_loss[loss=0.2613, simple_loss=0.3418, pruned_loss=0.09045, over 28042.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3399, pruned_loss=0.09725, over 5680819.40 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3741, pruned_loss=0.1347, over 5684407.47 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3376, pruned_loss=0.09425, over 5670592.99 frames. ], batch size: 412, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:35:18,453 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260663.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:35:47,283 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260692.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:35:48,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:35:50,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:36:09,857 INFO [train.py:968] (0/2) Epoch 6, batch 33000, giga_loss[loss=0.2706, simple_loss=0.3529, pruned_loss=0.0942, over 28495.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3415, pruned_loss=0.09685, over 5665024.33 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3736, pruned_loss=0.1344, over 5678834.04 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3391, pruned_loss=0.09357, over 5661123.54 frames. ], batch size: 336, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:36:09,862 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 09:36:18,234 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2141, simple_loss=0.3127, pruned_loss=0.05779, over 944034.00 frames. 
+2023-03-03 09:36:18,235 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 09:36:33,345 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:36:47,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1492, 1.4183, 3.0930, 2.9066], device='cuda:0'), covar=tensor([0.1268, 0.1968, 0.0390, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0577, 0.0543, 0.0762, 0.0625], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 09:36:50,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.139e+02 1.210e+03 1.536e+03 2.102e+03 5.710e+03, threshold=3.073e+03, percent-clipped=5.0
+2023-03-03 09:37:02,659 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 09:37:07,134 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-03 09:37:15,042 INFO [train.py:968] (0/2) Epoch 6, batch 33050, giga_loss[loss=0.2763, simple_loss=0.3616, pruned_loss=0.09555, over 28909.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3451, pruned_loss=0.09834, over 5666316.11 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3737, pruned_loss=0.1344, over 5681557.87 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3424, pruned_loss=0.09495, over 5660347.99 frames. ], batch size: 227, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:37:36,284 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:37:41,347 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:38:04,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:38:18,544 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:38:19,921 INFO [train.py:968] (0/2) Epoch 6, batch 33100, giga_loss[loss=0.264, simple_loss=0.3429, pruned_loss=0.0926, over 28671.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3458, pruned_loss=0.09833, over 5671850.36 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3734, pruned_loss=0.1342, over 5682512.09 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3439, pruned_loss=0.09562, over 5666179.70 frames. ], batch size: 307, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:39:00,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.339e+02 1.380e+03 1.757e+03 2.571e+03 5.639e+03, threshold=3.513e+03, percent-clipped=13.0
+2023-03-03 09:39:13,990 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2105, 1.4354, 1.1727, 1.3705], device='cuda:0'), covar=tensor([0.2055, 0.1964, 0.2136, 0.1611], device='cuda:0'), in_proj_covar=tensor([0.1133, 0.0864, 0.1013, 0.0940], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 09:39:22,511 INFO [train.py:968] (0/2) Epoch 6, batch 33150, libri_loss[loss=0.2469, simple_loss=0.3066, pruned_loss=0.09363, over 29667.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3472, pruned_loss=0.0997, over 5674822.54 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3733, pruned_loss=0.1343, over 5687205.61 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3451, pruned_loss=0.09658, over 5665420.26 frames. ], batch size: 69, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:40:18,353 INFO [train.py:968] (0/2) Epoch 6, batch 33200, giga_loss[loss=0.3024, simple_loss=0.3732, pruned_loss=0.1158, over 28441.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3436, pruned_loss=0.09714, over 5678999.79 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3731, pruned_loss=0.1342, over 5688047.91 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3416, pruned_loss=0.09416, over 5670608.65 frames. ], batch size: 370, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:40:21,163 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1570, 3.0167, 2.8222, 1.4064], device='cuda:0'), covar=tensor([0.0874, 0.0870, 0.0943, 0.2355], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0822, 0.0758, 0.0589], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 09:40:54,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.768e+02 1.337e+03 1.897e+03 3.501e+03 1.185e+04, threshold=3.794e+03, percent-clipped=24.0
+2023-03-03 09:40:58,573 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:41:01,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:41:12,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5461, 3.6455, 1.6393, 1.4873], device='cuda:0'), covar=tensor([0.0802, 0.0251, 0.0809, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0485, 0.0321, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 09:41:19,522 INFO [train.py:968] (0/2) Epoch 6, batch 33250, giga_loss[loss=0.2546, simple_loss=0.3329, pruned_loss=0.08815, over 28613.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3428, pruned_loss=0.09692, over 5670286.10 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3732, pruned_loss=0.1343, over 5680378.43 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3404, pruned_loss=0.09366, over 5669834.63 frames. ], batch size: 307, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:41:34,145 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:41:45,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3706, 2.7176, 1.4248, 1.4447], device='cuda:0'), covar=tensor([0.0810, 0.0312, 0.0795, 0.1199], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0482, 0.0319, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 09:41:55,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260992.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 09:42:06,503 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-03 09:42:17,271 INFO [train.py:968] (0/2) Epoch 6, batch 33300, giga_loss[loss=0.2717, simple_loss=0.3463, pruned_loss=0.09858, over 28752.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3409, pruned_loss=0.09689, over 5658470.04 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3725, pruned_loss=0.134, over 5666169.23 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3389, pruned_loss=0.09363, over 5670884.27 frames. ], batch size: 307, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:42:33,683 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-03 09:42:55,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.977e+02 1.176e+03 1.644e+03 2.352e+03 5.277e+03, threshold=3.289e+03, percent-clipped=2.0
+2023-03-03 09:43:21,876 INFO [train.py:968] (0/2) Epoch 6, batch 33350, giga_loss[loss=0.2309, simple_loss=0.3203, pruned_loss=0.07073, over 29016.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3426, pruned_loss=0.09744, over 5653159.00 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3725, pruned_loss=0.1341, over 5658463.58 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3407, pruned_loss=0.09445, over 5670430.20 frames. ], batch size: 155, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:44:24,800 INFO [train.py:968] (0/2) Epoch 6, batch 33400, giga_loss[loss=0.2633, simple_loss=0.311, pruned_loss=0.1078, over 24504.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3439, pruned_loss=0.09859, over 5657972.27 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3722, pruned_loss=0.1339, over 5663106.00 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3422, pruned_loss=0.09576, over 5667687.16 frames. ], batch size: 705, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:44:39,129 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:44:52,752 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261135.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:44:56,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261138.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:45:00,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.186e+02 1.324e+03 1.748e+03 2.642e+03 7.380e+03, threshold=3.496e+03, percent-clipped=15.0
+2023-03-03 09:45:03,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:45:23,791 INFO [train.py:968] (0/2) Epoch 6, batch 33450, giga_loss[loss=0.2721, simple_loss=0.3521, pruned_loss=0.09604, over 28808.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3448, pruned_loss=0.09994, over 5662823.13 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3712, pruned_loss=0.1332, over 5671985.57 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3433, pruned_loss=0.09713, over 5662633.48 frames. ], batch size: 174, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:45:30,180 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261167.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 09:46:28,542 INFO [train.py:968] (0/2) Epoch 6, batch 33500, giga_loss[loss=0.2657, simple_loss=0.3497, pruned_loss=0.09085, over 28896.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3491, pruned_loss=0.1018, over 5656521.54 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3711, pruned_loss=0.1331, over 5673525.91 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3479, pruned_loss=0.09942, over 5655056.37 frames. ], batch size: 213, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:46:53,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3564, 1.9340, 1.4637, 1.5194], device='cuda:0'), covar=tensor([0.0717, 0.0309, 0.0311, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0119, 0.0122, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0064, 0.0046, 0.0042, 0.0070], device='cuda:0')
+2023-03-03 09:46:59,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.229e+02 1.337e+03 1.691e+03 2.355e+03 5.460e+03, threshold=3.381e+03, percent-clipped=6.0
+2023-03-03 09:47:21,710 INFO [train.py:968] (0/2) Epoch 6, batch 33550, giga_loss[loss=0.2995, simple_loss=0.3693, pruned_loss=0.1149, over 29138.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3508, pruned_loss=0.1021, over 5665734.10 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3705, pruned_loss=0.1329, over 5679039.23 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3498, pruned_loss=0.09972, over 5659415.49 frames. ], batch size: 187, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:47:43,511 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:48:33,323 INFO [train.py:968] (0/2) Epoch 6, batch 33600, giga_loss[loss=0.2561, simple_loss=0.3328, pruned_loss=0.08974, over 28901.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3502, pruned_loss=0.1016, over 5673769.41 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3703, pruned_loss=0.1328, over 5683616.43 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3492, pruned_loss=0.0992, over 5664254.60 frames. ], batch size: 145, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:49:15,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.692e+02 1.456e+03 1.912e+03 2.797e+03 8.228e+03, threshold=3.824e+03, percent-clipped=12.0
+2023-03-03 09:49:40,653 INFO [train.py:968] (0/2) Epoch 6, batch 33650, giga_loss[loss=0.3248, simple_loss=0.3772, pruned_loss=0.1362, over 27664.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3476, pruned_loss=0.1007, over 5678701.55 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3696, pruned_loss=0.1324, over 5686132.56 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3472, pruned_loss=0.09865, over 5668672.94 frames. ], batch size: 472, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:50:34,376 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1707, 4.0111, 3.7494, 1.9813], device='cuda:0'), covar=tensor([0.0447, 0.0572, 0.0700, 0.2190], device='cuda:0'), in_proj_covar=tensor([0.0887, 0.0825, 0.0765, 0.0596], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 09:50:44,029 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 09:50:48,534 INFO [train.py:968] (0/2) Epoch 6, batch 33700, libri_loss[loss=0.3121, simple_loss=0.3707, pruned_loss=0.1267, over 27866.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3472, pruned_loss=0.1, over 5678554.35 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3698, pruned_loss=0.1325, over 5683860.59 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3463, pruned_loss=0.09787, over 5672673.02 frames. ], batch size: 116, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:51:27,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.728e+02 1.409e+03 1.781e+03 2.600e+03 7.631e+03, threshold=3.561e+03, percent-clipped=14.0
+2023-03-03 09:51:53,213 INFO [train.py:968] (0/2) Epoch 6, batch 33750, giga_loss[loss=0.275, simple_loss=0.3464, pruned_loss=0.1019, over 28501.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.346, pruned_loss=0.1002, over 5659657.30 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3698, pruned_loss=0.1327, over 5664919.30 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3449, pruned_loss=0.09784, over 5671722.25 frames. ], batch size: 336, lr: 5.24e-03, grad_scale: 2.0
+2023-03-03 09:52:37,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:52:52,894 INFO [train.py:968] (0/2) Epoch 6, batch 33800, giga_loss[loss=0.2673, simple_loss=0.3405, pruned_loss=0.09709, over 28896.00 frames. ], tot_loss[loss=0.272, simple_loss=0.344, pruned_loss=0.09997, over 5673002.85 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3691, pruned_loss=0.1322, over 5672785.62 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.343, pruned_loss=0.09735, over 5675745.74 frames. ], batch size: 112, lr: 5.24e-03, grad_scale: 2.0
+2023-03-03 09:52:58,207 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-03 09:53:02,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:53:16,645 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261529.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 09:53:33,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.609e+02 1.368e+03 1.880e+03 2.692e+03 1.838e+04, threshold=3.761e+03, percent-clipped=10.0
+2023-03-03 09:53:55,066 INFO [train.py:968] (0/2) Epoch 6, batch 33850, giga_loss[loss=0.2898, simple_loss=0.3617, pruned_loss=0.109, over 27666.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3446, pruned_loss=0.09947, over 5682417.73 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3692, pruned_loss=0.1322, over 5677721.06 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3433, pruned_loss=0.09681, over 5680244.81 frames. ], batch size: 472, lr: 5.23e-03, grad_scale: 2.0
+2023-03-03 09:54:53,732 INFO [train.py:968] (0/2) Epoch 6, batch 33900, giga_loss[loss=0.2459, simple_loss=0.3071, pruned_loss=0.09232, over 24476.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.343, pruned_loss=0.09832, over 5662179.83 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3699, pruned_loss=0.1328, over 5672021.75 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3403, pruned_loss=0.09444, over 5665822.93 frames. ], batch size: 705, lr: 5.23e-03, grad_scale: 2.0
+2023-03-03 09:55:26,227 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:55:28,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.712e+02 1.428e+03 1.955e+03 2.784e+03 1.016e+04, threshold=3.910e+03, percent-clipped=13.0
+2023-03-03 09:55:30,675 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:55:37,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:55:51,169 INFO [train.py:968] (0/2) Epoch 6, batch 33950, giga_loss[loss=0.2697, simple_loss=0.3519, pruned_loss=0.09372, over 28961.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3435, pruned_loss=0.09609, over 5671389.08 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3692, pruned_loss=0.1323, over 5675347.25 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3417, pruned_loss=0.09292, over 5671279.13 frames. ], batch size: 213, lr: 5.23e-03, grad_scale: 2.0
+2023-03-03 09:55:52,157 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:55:54,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:56:05,405 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:56:27,961 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:56:43,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3956, 1.5309, 1.4990, 1.5291], device='cuda:0'), covar=tensor([0.0958, 0.1294, 0.1328, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0720, 0.0636, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 09:56:44,644 INFO [train.py:968] (0/2) Epoch 6, batch 34000, giga_loss[loss=0.2507, simple_loss=0.3398, pruned_loss=0.08079, over 28681.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3453, pruned_loss=0.09638, over 5682426.05 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3689, pruned_loss=0.1319, over 5683769.67 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3432, pruned_loss=0.09278, over 5674664.80 frames. ], batch size: 262, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 09:57:01,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4166, 1.8588, 1.4273, 1.5380], device='cuda:0'), covar=tensor([0.0689, 0.0360, 0.0317, 0.0739], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0123, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0064, 0.0046, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 09:57:24,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.571e+02 1.184e+03 1.612e+03 2.170e+03 7.494e+03, threshold=3.225e+03, percent-clipped=6.0
+2023-03-03 09:57:48,504 INFO [train.py:968] (0/2) Epoch 6, batch 34050, giga_loss[loss=0.2635, simple_loss=0.3245, pruned_loss=0.1013, over 24218.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3451, pruned_loss=0.09611, over 5677067.03 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3685, pruned_loss=0.1318, over 5684586.24 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3433, pruned_loss=0.09275, over 5669967.19 frames. ], batch size: 705, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 09:58:31,449 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:58:35,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:58:57,386 INFO [train.py:968] (0/2) Epoch 6, batch 34100, giga_loss[loss=0.2751, simple_loss=0.3296, pruned_loss=0.1103, over 24485.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3463, pruned_loss=0.0974, over 5675849.44 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3686, pruned_loss=0.1318, over 5689691.37 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3442, pruned_loss=0.09383, over 5665463.30 frames. ], batch size: 705, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 09:59:14,323 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:59:32,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4202, 1.8288, 1.7706, 1.5377], device='cuda:0'), covar=tensor([0.1500, 0.1739, 0.1141, 0.1289], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0705, 0.0796, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 09:59:32,265 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-03-03 09:59:34,296 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-03-03 09:59:34,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.702e+02 1.407e+03 1.782e+03 2.804e+03 6.360e+03, threshold=3.565e+03, percent-clipped=16.0
+2023-03-03 09:59:45,735 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:00:03,764 INFO [train.py:968] (0/2) Epoch 6, batch 34150, giga_loss[loss=0.285, simple_loss=0.3419, pruned_loss=0.1141, over 26853.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3456, pruned_loss=0.09701, over 5668229.07 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3683, pruned_loss=0.1317, over 5684770.01 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3437, pruned_loss=0.09356, over 5663375.88 frames. ], batch size: 555, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:01:01,345 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-03 10:01:01,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261904.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:01:13,114 INFO [train.py:968] (0/2) Epoch 6, batch 34200, giga_loss[loss=0.2316, simple_loss=0.2982, pruned_loss=0.08254, over 24651.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.346, pruned_loss=0.09643, over 5673588.29 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3684, pruned_loss=0.1316, over 5691222.31 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3438, pruned_loss=0.09278, over 5663366.21 frames. ], batch size: 705, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:01:27,158 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 10:01:51,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5762, 1.7363, 1.5995, 1.6501], device='cuda:0'), covar=tensor([0.1188, 0.1956, 0.1627, 0.1490], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0725, 0.0635, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 10:01:58,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.122e+02 1.342e+03 1.767e+03 2.956e+03 8.684e+03, threshold=3.534e+03, percent-clipped=15.0
+2023-03-03 10:02:20,938 INFO [train.py:968] (0/2) Epoch 6, batch 34250, giga_loss[loss=0.2584, simple_loss=0.3454, pruned_loss=0.08566, over 28716.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.348, pruned_loss=0.09754, over 5667371.04 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3684, pruned_loss=0.1317, over 5682847.63 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3459, pruned_loss=0.09398, over 5667204.78 frames. ], batch size: 262, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:02:33,447 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0696, 1.1705, 3.9951, 3.0651], device='cuda:0'), covar=tensor([0.1559, 0.2268, 0.0340, 0.0669], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0535, 0.0752, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 10:02:52,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5719, 3.8572, 1.6146, 1.5564], device='cuda:0'), covar=tensor([0.0826, 0.0270, 0.0852, 0.1252], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0475, 0.0316, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 10:03:11,632 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-262000.pt
+2023-03-03 10:03:22,819 INFO [train.py:968] (0/2) Epoch 6, batch 34300, giga_loss[loss=0.2207, simple_loss=0.3131, pruned_loss=0.06418, over 28943.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3516, pruned_loss=0.09973, over 5668790.36 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3687, pruned_loss=0.1319, over 5680102.03 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3491, pruned_loss=0.09561, over 5670309.83 frames. ], batch size: 136, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:04:02,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7089, 1.4099, 5.4850, 3.7377], device='cuda:0'), covar=tensor([0.1491, 0.2328, 0.0265, 0.0601], device='cuda:0'), in_proj_covar=tensor([0.0570, 0.0532, 0.0746, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 10:04:06,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.969e+02 1.548e+03 2.059e+03 2.768e+03 8.131e+03, threshold=4.117e+03, percent-clipped=12.0
+2023-03-03 10:04:12,357 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262047.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 10:04:19,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=262050.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:04:30,959 INFO [train.py:968] (0/2) Epoch 6, batch 34350, giga_loss[loss=0.2283, simple_loss=0.314, pruned_loss=0.07134, over 29084.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3501, pruned_loss=0.09943, over 5663413.17 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3685, pruned_loss=0.1318, over 5673414.64 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3481, pruned_loss=0.09573, over 5671018.88 frames. ], batch size: 128, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:04:57,921 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:04:59,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2446, 1.7105, 1.1952, 0.4194], device='cuda:0'), covar=tensor([0.1514, 0.1074, 0.1967, 0.2565], device='cuda:0'), in_proj_covar=tensor([0.1414, 0.1343, 0.1388, 0.1185], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 10:05:20,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5828, 1.6051, 1.4311, 1.9259], device='cuda:0'), covar=tensor([0.2065, 0.2220, 0.2241, 0.2214], device='cuda:0'), in_proj_covar=tensor([0.1149, 0.0871, 0.1019, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 10:05:35,249 INFO [train.py:968] (0/2) Epoch 6, batch 34400, giga_loss[loss=0.2253, simple_loss=0.3125, pruned_loss=0.06911, over 28944.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3488, pruned_loss=0.0996, over 5669024.24 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3686, pruned_loss=0.1319, over 5672900.29 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3464, pruned_loss=0.09543, over 5675513.71 frames. ], batch size: 155, lr: 5.23e-03, grad_scale: 8.0
+2023-03-03 10:06:02,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3802, 1.4457, 1.5719, 1.4172], device='cuda:0'), covar=tensor([0.1540, 0.2181, 0.1308, 0.1325], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0709, 0.0798, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 10:06:24,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.874e+02 1.343e+03 1.662e+03 2.497e+03 9.039e+03, threshold=3.324e+03, percent-clipped=7.0
+2023-03-03 10:06:51,464 INFO [train.py:968] (0/2) Epoch 6, batch 34450, giga_loss[loss=0.3013, simple_loss=0.3523, pruned_loss=0.1251, over 25050.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3454, pruned_loss=0.09684, over 5670609.09 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3683, pruned_loss=0.1317, over 5676852.70 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3433, pruned_loss=0.09304, over 5672120.36 frames. ], batch size: 705, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:07:21,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9364, 1.9909, 1.3460, 1.6699], device='cuda:0'), covar=tensor([0.0685, 0.0579, 0.0934, 0.1008], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0432, 0.0496, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 10:07:51,633 INFO [train.py:968] (0/2) Epoch 6, batch 34500, giga_loss[loss=0.2461, simple_loss=0.3289, pruned_loss=0.08166, over 28685.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3444, pruned_loss=0.09596, over 5675063.90 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3684, pruned_loss=0.1316, over 5680084.84 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3421, pruned_loss=0.09201, over 5673327.23 frames. ], batch size: 262, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:07:57,978 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 10:08:00,078 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=262217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:08:10,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:08:12,465 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=262226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:08:33,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.056e+02 1.185e+03 1.593e+03 2.281e+03 6.859e+03, threshold=3.186e+03, percent-clipped=7.0
+2023-03-03 10:08:52,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7924, 1.0490, 3.2940, 2.7681], device='cuda:0'), covar=tensor([0.1692, 0.2422, 0.0428, 0.1107], device='cuda:0'), in_proj_covar=tensor([0.0575, 0.0532, 0.0751, 0.0616], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 10:08:53,986 INFO [train.py:968] (0/2) Epoch 6, batch 34550, giga_loss[loss=0.2449, simple_loss=0.3238, pruned_loss=0.08305, over 28515.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3473, pruned_loss=0.09816, over 5669105.39 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3684, pruned_loss=0.1316, over 5679721.90 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3451, pruned_loss=0.09456, over 5668291.81 frames. ], batch size: 78, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:09:58,119 INFO [train.py:968] (0/2) Epoch 6, batch 34600, giga_loss[loss=0.2618, simple_loss=0.3381, pruned_loss=0.09269, over 28854.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3503, pruned_loss=0.09926, over 5672000.82 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3685, pruned_loss=0.1317, over 5680850.29 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3484, pruned_loss=0.09627, over 5670331.50 frames. ], batch size: 186, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:10:33,741 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.547e+02 1.483e+03 2.025e+03 2.753e+03 1.074e+04, threshold=4.051e+03, percent-clipped=17.0
+2023-03-03 10:10:54,139 INFO [train.py:968] (0/2) Epoch 6, batch 34650, giga_loss[loss=0.3168, simple_loss=0.3559, pruned_loss=0.1389, over 26855.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3503, pruned_loss=0.1007, over 5678196.45 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3686, pruned_loss=0.1319, over 5683904.96 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3483, pruned_loss=0.09735, over 5673897.35 frames. ], batch size: 555, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:11:04,816 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:11:07,280 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=262370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:11:07,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=262370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:11:37,266 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:11:43,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3659, 1.7175, 1.6758, 1.4095], device='cuda:0'), covar=tensor([0.1366, 0.1813, 0.1122, 0.1269], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0706, 0.0798, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 10:11:52,284 INFO [train.py:968] (0/2) Epoch 6, batch 34700, giga_loss[loss=0.2352, simple_loss=0.2975, pruned_loss=0.08642, over 24247.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3477, pruned_loss=0.1, over 5666989.20 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3683, pruned_loss=0.1317, over 5686953.82 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3461, pruned_loss=0.09721, over 5660665.85 frames. ], batch size: 705, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:11:54,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0559, 1.0441, 4.1435, 3.1696], device='cuda:0'), covar=tensor([0.1659, 0.2552, 0.0359, 0.1052], device='cuda:0'), in_proj_covar=tensor([0.0579, 0.0539, 0.0761, 0.0625], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 10:12:27,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.732e+02 1.484e+03 1.910e+03 2.488e+03 6.460e+03, threshold=3.820e+03, percent-clipped=6.0
+2023-03-03 10:12:45,054 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 10:12:45,976 INFO [train.py:968] (0/2) Epoch 6, batch 34750, giga_loss[loss=0.2982, simple_loss=0.3696, pruned_loss=0.1134, over 28587.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3477, pruned_loss=0.1007, over 5675421.94 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3673, pruned_loss=0.1309, over 5689927.48 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3467, pruned_loss=0.09819, over 5667354.22 frames. ], batch size: 92, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:13:13,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4967, 1.8192, 1.7331, 1.5168], device='cuda:0'), covar=tensor([0.1314, 0.1591, 0.1045, 0.1184], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0703, 0.0795, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 10:13:35,208 INFO [train.py:968] (0/2) Epoch 6, batch 34800, giga_loss[loss=0.3095, simple_loss=0.3827, pruned_loss=0.1182, over 28154.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3568, pruned_loss=0.107, over 5660470.12 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3675, pruned_loss=0.1314, over 5675932.23 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3555, pruned_loss=0.1041, over 5666029.55 frames. ], batch size: 77, lr: 5.23e-03, grad_scale: 8.0
+2023-03-03 10:13:52,790 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-03 10:14:06,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.791e+02 1.309e+03 1.630e+03 2.081e+03 4.304e+03, threshold=3.261e+03, percent-clipped=2.0
+2023-03-03 10:14:08,875 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3620, 1.4723, 1.2994, 1.3004], device='cuda:0'), covar=tensor([0.2121, 0.2064, 0.2172, 0.1873], device='cuda:0'), in_proj_covar=tensor([0.1141, 0.0868, 0.1017, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 10:14:21,970 INFO [train.py:968] (0/2) Epoch 6, batch 34850, giga_loss[loss=0.2974, simple_loss=0.3833, pruned_loss=0.1058, over 28658.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3662, pruned_loss=0.1129, over 5665543.68 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3674, pruned_loss=0.1311, over 5679872.77 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3652, pruned_loss=0.1105, over 5666201.84 frames. ], batch size: 262, lr: 5.22e-03, grad_scale: 8.0
+2023-03-03 10:14:24,678 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-03 10:14:50,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:14:59,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:15:06,854 INFO [train.py:968] (0/2) Epoch 6, batch 34900, giga_loss[loss=0.278, simple_loss=0.3415, pruned_loss=0.1073, over 28886.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3682, pruned_loss=0.1149, over 5670785.20 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3673, pruned_loss=0.1311, over 5682259.31 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3676, pruned_loss=0.1129, over 5669023.59 frames. ], batch size: 119, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:15:36,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.836e+02 1.221e+03 1.497e+03 1.881e+03 7.410e+03, threshold=2.993e+03, percent-clipped=6.0
+2023-03-03 10:15:49,882 INFO [train.py:968] (0/2) Epoch 6, batch 34950, giga_loss[loss=0.2754, simple_loss=0.341, pruned_loss=0.1049, over 28943.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3625, pruned_loss=0.1129, over 5683373.21 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3674, pruned_loss=0.131, over 5686114.31 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3619, pruned_loss=0.111, over 5678389.03 frames. ], batch size: 227, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:16:31,951 INFO [train.py:968] (0/2) Epoch 6, batch 35000, giga_loss[loss=0.2639, simple_loss=0.3329, pruned_loss=0.09747, over 28617.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3573, pruned_loss=0.111, over 5690678.36 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3679, pruned_loss=0.131, over 5693122.97 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3561, pruned_loss=0.1088, over 5680498.83 frames. ], batch size: 336, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:16:52,202 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:16:54,801 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=262738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:00,277 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:00,527 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.212e+02 1.284e+03 1.712e+03 2.160e+03 5.147e+03, threshold=3.425e+03, percent-clipped=12.0
+2023-03-03 10:17:00,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:02,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=262747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:14,399 INFO [train.py:968] (0/2) Epoch 6, batch 35050, giga_loss[loss=0.2252, simple_loss=0.2974, pruned_loss=0.07647, over 28886.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3504, pruned_loss=0.1081, over 5693105.47 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3685, pruned_loss=0.1312, over 5697884.70 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3487, pruned_loss=0.1057, over 5680844.00 frames. ], batch size: 112, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:17:19,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:27,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:55,732 INFO [train.py:968] (0/2) Epoch 6, batch 35100, giga_loss[loss=0.2429, simple_loss=0.3048, pruned_loss=0.09056, over 28272.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3423, pruned_loss=0.1045, over 5697571.29 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3685, pruned_loss=0.1312, over 5701625.90 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3406, pruned_loss=0.1022, over 5684653.97 frames. ], batch size: 77, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:18:16,093 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-03 10:18:24,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.712e+02 1.074e+03 1.346e+03 1.879e+03 7.915e+03, threshold=2.691e+03, percent-clipped=6.0
+2023-03-03 10:18:38,004 INFO [train.py:968] (0/2) Epoch 6, batch 35150, giga_loss[loss=0.2087, simple_loss=0.2804, pruned_loss=0.06847, over 28521.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3373, pruned_loss=0.1028, over 5686948.39 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3691, pruned_loss=0.1318, over 5692895.70 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3348, pruned_loss=0.09988, over 5683930.63 frames. ], batch size: 71, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:19:03,572 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:19:06,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=262891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:19:19,709 INFO [train.py:968] (0/2) Epoch 6, batch 35200, giga_loss[loss=0.2345, simple_loss=0.3002, pruned_loss=0.08441, over 28786.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3329, pruned_loss=0.1006, over 5678202.50 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3698, pruned_loss=0.1325, over 5683632.85 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3298, pruned_loss=0.09733, over 5684586.15 frames. ], batch size: 99, lr: 5.22e-03, grad_scale: 8.0
+2023-03-03 10:19:29,343 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:19:51,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.369e+02 1.005e+03 1.317e+03 1.834e+03 5.521e+03, threshold=2.633e+03, percent-clipped=9.0
+2023-03-03 10:20:02,902 INFO [train.py:968] (0/2) Epoch 6, batch 35250, giga_loss[loss=0.2264, simple_loss=0.304, pruned_loss=0.07442, over 28938.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3291, pruned_loss=0.09807, over 5688259.03 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3703, pruned_loss=0.1326, over 5687137.40 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3255, pruned_loss=0.09472, over 5690215.33 frames. ], batch size: 186, lr: 5.22e-03, grad_scale: 8.0
+2023-03-03 10:20:23,485 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4453, 3.1832, 1.6105, 1.4299], device='cuda:0'), covar=tensor([0.0869, 0.0311, 0.0808, 0.1274], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0475, 0.0316, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 10:20:33,367 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 10:20:42,882 INFO [train.py:968] (0/2) Epoch 6, batch 35300, giga_loss[loss=0.2327, simple_loss=0.304, pruned_loss=0.08072, over 28890.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3261, pruned_loss=0.09653, over 5700505.37 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3703, pruned_loss=0.1325, over 5693088.99 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3219, pruned_loss=0.0928, over 5696754.51 frames. ], batch size: 227, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:21:14,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.359e+02 1.175e+03 1.580e+03 2.691e+03 6.768e+03, threshold=3.160e+03, percent-clipped=25.0
+2023-03-03 10:21:22,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3482, 1.5031, 1.3204, 1.5308], device='cuda:0'), covar=tensor([0.0794, 0.0323, 0.0340, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0119, 0.0123, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0064, 0.0046, 0.0042, 0.0070], device='cuda:0')
+2023-03-03 10:21:26,273 INFO [train.py:968] (0/2) Epoch 6, batch 35350, giga_loss[loss=0.2307, simple_loss=0.2978, pruned_loss=0.08183, over 27670.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3242, pruned_loss=0.0954, over 5696140.62 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.371, pruned_loss=0.1329, over 5683464.27 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.319, pruned_loss=0.09112, over 5701446.53 frames. ], batch size: 472, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:21:46,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4445, 1.4626, 1.0582, 1.4962], device='cuda:0'), covar=tensor([0.0698, 0.0290, 0.0352, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0123, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 10:21:51,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5456, 1.5333, 1.2637, 1.2537], device='cuda:0'), covar=tensor([0.0727, 0.0579, 0.1024, 0.0919], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0435, 0.0496, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 10:22:07,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6127, 1.7946, 1.6770, 1.6285], device='cuda:0'), covar=tensor([0.1423, 0.1756, 0.1754, 0.1578], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0731, 0.0640, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 10:22:07,924 INFO [train.py:968] (0/2) Epoch 6, batch 35400, giga_loss[loss=0.2051, simple_loss=0.279, pruned_loss=0.06556, over 28895.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3203, pruned_loss=0.0934, over 5685035.45 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3716, pruned_loss=0.1333, over 5677033.06 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3148, pruned_loss=0.08902, over 5694958.11 frames. ], batch size: 213, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:22:35,985 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:22:38,902 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.981e+02 1.011e+03 1.256e+03 1.751e+03 5.565e+03, threshold=2.512e+03, percent-clipped=3.0
+2023-03-03 10:22:51,111 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4626, 3.4711, 1.6103, 1.5008], device='cuda:0'), covar=tensor([0.0909, 0.0245, 0.0832, 0.1291], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0474, 0.0316, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 10:22:51,486 INFO [train.py:968] (0/2) Epoch 6, batch 35450, giga_loss[loss=0.2214, simple_loss=0.2906, pruned_loss=0.07613, over 28874.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.317, pruned_loss=0.0919, over 5688537.27 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.372, pruned_loss=0.1335, over 5679023.01 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3119, pruned_loss=0.08787, over 5694697.74 frames. ], batch size: 112, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:22:53,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5461, 1.7099, 1.4141, 1.2724], device='cuda:0'), covar=tensor([0.1614, 0.1160, 0.0915, 0.1194], device='cuda:0'), in_proj_covar=tensor([0.1511, 0.1287, 0.1280, 0.1393], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 10:23:34,461 INFO [train.py:968] (0/2) Epoch 6, batch 35500, giga_loss[loss=0.2107, simple_loss=0.2775, pruned_loss=0.07197, over 29026.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3139, pruned_loss=0.0903, over 5691266.15 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3722, pruned_loss=0.1336, over 5683444.79 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3086, pruned_loss=0.08625, over 5692301.48 frames. ], batch size: 136, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:24:07,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.350e+02 9.281e+02 1.266e+03 2.004e+03 4.878e+03, threshold=2.532e+03, percent-clipped=13.0
+2023-03-03 10:24:14,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7321, 3.5369, 3.3541, 1.5647], device='cuda:0'), covar=tensor([0.0621, 0.0759, 0.0728, 0.2220], device='cuda:0'), in_proj_covar=tensor([0.0873, 0.0817, 0.0745, 0.0588], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0009, 0.0009], device='cuda:0')
+2023-03-03 10:24:20,974 INFO [train.py:968] (0/2) Epoch 6, batch 35550, giga_loss[loss=0.2279, simple_loss=0.2933, pruned_loss=0.08124, over 27759.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3104, pruned_loss=0.08844, over 5699007.12 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3724, pruned_loss=0.1335, over 5688794.59 frames. ], giga_tot_loss[loss=0.2367, simple_loss=0.3047, pruned_loss=0.08431, over 5695242.62 frames. ], batch size: 472, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:24:26,946 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.29 vs. limit=5.0
+2023-03-03 10:24:49,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7232, 4.5239, 4.2871, 2.0673], device='cuda:0'), covar=tensor([0.0403, 0.0543, 0.0657, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.0870, 0.0817, 0.0747, 0.0588], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 10:24:59,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2036, 1.2839, 1.1280, 1.0337], device='cuda:0'), covar=tensor([0.0706, 0.0487, 0.1019, 0.0807], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0437, 0.0495, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 10:25:04,811 INFO [train.py:968] (0/2) Epoch 6, batch 35600, giga_loss[loss=0.2808, simple_loss=0.3445, pruned_loss=0.1085, over 28256.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3142, pruned_loss=0.09093, over 5683439.79 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3732, pruned_loss=0.1339, over 5682153.22 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3073, pruned_loss=0.08607, over 5687068.16 frames. ], batch size: 368, lr: 5.22e-03, grad_scale: 8.0
+2023-03-03 10:25:08,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 10:25:36,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.447e+02 1.041e+03 1.382e+03 1.882e+03 6.079e+03, threshold=2.765e+03, percent-clipped=10.0
+2023-03-03 10:25:50,323 INFO [train.py:968] (0/2) Epoch 6, batch 35650, giga_loss[loss=0.331, simple_loss=0.3878, pruned_loss=0.1371, over 28402.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3275, pruned_loss=0.09817, over 5688190.39 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3732, pruned_loss=0.134, over 5685529.84 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3213, pruned_loss=0.09376, over 5688114.31 frames. ], batch size: 85, lr: 5.22e-03, grad_scale: 8.0
+2023-03-03 10:25:52,556 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 10:26:35,292 INFO [train.py:968] (0/2) Epoch 6, batch 35700, giga_loss[loss=0.2922, simple_loss=0.3592, pruned_loss=0.1126, over 28930.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3419, pruned_loss=0.1064, over 5690649.02 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3738, pruned_loss=0.1344, over 5688845.27 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3356, pruned_loss=0.1018, over 5687571.63 frames. ], batch size: 106, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:26:35,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3115, 1.5181, 1.4192, 1.4836], device='cuda:0'), covar=tensor([0.1149, 0.1324, 0.1667, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0735, 0.0645, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 10:26:59,771 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-03 10:27:05,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.019e+02 1.323e+03 1.828e+03 2.626e+03 6.741e+03, threshold=3.656e+03, percent-clipped=24.0
+2023-03-03 10:27:19,842 INFO [train.py:968] (0/2) Epoch 6, batch 35750, giga_loss[loss=0.2911, simple_loss=0.3643, pruned_loss=0.1089, over 28740.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3527, pruned_loss=0.1119, over 5697292.23 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3743, pruned_loss=0.1348, over 5695082.13 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3465, pruned_loss=0.1072, over 5689411.90 frames. ], batch size: 119, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:27:43,421 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:27:54,830 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:28:00,257 INFO [train.py:968] (0/2) Epoch 6, batch 35800, giga_loss[loss=0.2593, simple_loss=0.3287, pruned_loss=0.09499, over 23358.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3573, pruned_loss=0.1129, over 5689377.19 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3744, pruned_loss=0.1349, over 5689447.85 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3521, pruned_loss=0.1088, over 5688426.19 frames. ], batch size: 705, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:28:07,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:28:10,701 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-03 10:28:20,599 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:28:33,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.331e+02 1.292e+03 1.730e+03 2.555e+03 9.342e+03, threshold=3.461e+03, percent-clipped=8.0
+2023-03-03 10:28:35,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3860, 2.0597, 1.5368, 1.5231], device='cuda:0'), covar=tensor([0.0775, 0.0273, 0.0306, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0119, 0.0123, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 10:28:44,187 INFO [train.py:968] (0/2) Epoch 6, batch 35850, giga_loss[loss=0.2678, simple_loss=0.3469, pruned_loss=0.09434, over 28870.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3582, pruned_loss=0.1117, over 5696739.17 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3745, pruned_loss=0.1349, over 5695674.55 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3534, pruned_loss=0.1077, over 5690098.13 frames. ], batch size: 145, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:29:07,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1966, 1.6186, 1.2208, 0.3765], device='cuda:0'), covar=tensor([0.2014, 0.1389, 0.2062, 0.2790], device='cuda:0'), in_proj_covar=tensor([0.1412, 0.1334, 0.1387, 0.1167], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 10:29:29,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:29:32,383 INFO [train.py:968] (0/2) Epoch 6, batch 35900, libri_loss[loss=0.3473, simple_loss=0.4095, pruned_loss=0.1426, over 29480.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3606, pruned_loss=0.1127, over 5696223.71 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.3749, pruned_loss=0.1352, over 5697726.53 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3563, pruned_loss=0.1091, over 5689122.38 frames. ], batch size: 85, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:29:54,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:30:03,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.628e+02 1.202e+03 1.526e+03 1.889e+03 7.155e+03, threshold=3.052e+03, percent-clipped=5.0
+2023-03-03 10:30:05,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4443, 3.0150, 1.6589, 1.5494], device='cuda:0'), covar=tensor([0.0719, 0.0257, 0.0626, 0.1067], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0468, 0.0311, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0025, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 10:30:15,383 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=263660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:30:15,827 INFO [train.py:968] (0/2) Epoch 6, batch 35950, libri_loss[loss=0.2854, simple_loss=0.3516, pruned_loss=0.1096, over 27274.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3641, pruned_loss=0.1156, over 5688719.85 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3751, pruned_loss=0.135, over 5700627.49 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3603, pruned_loss=0.1124, over 5680563.91 frames. ], batch size: 60, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:30:17,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=263663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:30:42,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=263692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:30:59,315 INFO [train.py:968] (0/2) Epoch 6, batch 36000, giga_loss[loss=0.4053, simple_loss=0.4304, pruned_loss=0.1901, over 26564.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.368, pruned_loss=0.1183, over 5688889.99 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3756, pruned_loss=0.1353, over 5703717.13 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3643, pruned_loss=0.1152, over 5679510.86 frames. ], batch size: 555, lr: 5.21e-03, grad_scale: 8.0
+2023-03-03 10:30:59,319 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 10:31:08,351 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2324, simple_loss=0.3375, pruned_loss=0.06366, over 944034.00 frames. 
+2023-03-03 10:31:08,352 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 10:31:17,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8817, 1.3195, 3.3154, 2.7198], device='cuda:0'), covar=tensor([0.1611, 0.2212, 0.0402, 0.1034], device='cuda:0'), in_proj_covar=tensor([0.0572, 0.0534, 0.0752, 0.0633], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 10:31:36,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.230e+02 1.133e+03 1.457e+03 1.857e+03 4.145e+03, threshold=2.914e+03, percent-clipped=6.0
+2023-03-03 10:31:48,797 INFO [train.py:968] (0/2) Epoch 6, batch 36050, giga_loss[loss=0.2822, simple_loss=0.3632, pruned_loss=0.1006, over 29034.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3707, pruned_loss=0.1192, over 5694911.22 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3757, pruned_loss=0.1353, over 5705801.93 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3677, pruned_loss=0.1167, over 5685739.75 frames. ], batch size: 136, lr: 5.21e-03, grad_scale: 8.0
+2023-03-03 10:32:29,204 INFO [train.py:968] (0/2) Epoch 6, batch 36100, giga_loss[loss=0.3506, simple_loss=0.405, pruned_loss=0.1481, over 28704.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3727, pruned_loss=0.12, over 5692588.02 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3761, pruned_loss=0.1356, over 5709003.65 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3699, pruned_loss=0.1173, over 5682186.63 frames. ], batch size: 262, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:33:00,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.928e+02 1.146e+03 1.468e+03 1.941e+03 6.427e+03, threshold=2.935e+03, percent-clipped=11.0
+2023-03-03 10:33:11,071 INFO [train.py:968] (0/2) Epoch 6, batch 36150, giga_loss[loss=0.3343, simple_loss=0.3954, pruned_loss=0.1366, over 28683.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3745, pruned_loss=0.1206, over 5690161.78 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3762, pruned_loss=0.1356, over 5712653.09 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3721, pruned_loss=0.1181, over 5678110.81 frames. ], batch size: 262, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:33:13,577 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:33:24,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:33:50,972 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:33:52,726 INFO [train.py:968] (0/2) Epoch 6, batch 36200, giga_loss[loss=0.2686, simple_loss=0.3624, pruned_loss=0.08738, over 28522.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3749, pruned_loss=0.1195, over 5692733.60 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.377, pruned_loss=0.1361, over 5706129.52 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3723, pruned_loss=0.1167, over 5688002.13 frames. ], batch size: 60, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:34:20,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.553e+02 1.091e+03 1.410e+03 2.006e+03 1.061e+04, threshold=2.821e+03, percent-clipped=12.0
+2023-03-03 10:34:31,419 INFO [train.py:968] (0/2) Epoch 6, batch 36250, giga_loss[loss=0.2758, simple_loss=0.3319, pruned_loss=0.1098, over 23734.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3729, pruned_loss=0.1169, over 5696319.69 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3773, pruned_loss=0.1362, over 5709705.21 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3705, pruned_loss=0.1142, over 5689340.74 frames. ], batch size: 710, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:34:37,791 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-03 10:34:50,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:04,121 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-264000.pt
+2023-03-03 10:35:08,768 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:11,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:12,067 INFO [train.py:968] (0/2) Epoch 6, batch 36300, libri_loss[loss=0.3323, simple_loss=0.3945, pruned_loss=0.135, over 25793.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3717, pruned_loss=0.115, over 5705473.59 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.378, pruned_loss=0.1365, over 5709472.99 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.369, pruned_loss=0.1123, over 5700243.31 frames. ], batch size: 136, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:35:14,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=264014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:21,711 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:23,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:35,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:42,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.780e+02 1.082e+03 1.383e+03 1.811e+03 6.245e+03, threshold=2.766e+03, percent-clipped=9.0
+2023-03-03 10:35:45,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3617, 2.9221, 1.3408, 1.4023], device='cuda:0'), covar=tensor([0.0908, 0.0250, 0.0864, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0474, 0.0312, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 10:35:45,122 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:48,850 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:48,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:53,780 INFO [train.py:968] (0/2) Epoch 6, batch 36350, giga_loss[loss=0.3035, simple_loss=0.3713, pruned_loss=0.1179, over 28852.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3725, pruned_loss=0.1161, over 5714063.93 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3785, pruned_loss=0.1368, over 5711609.27 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3698, pruned_loss=0.1133, over 5708062.71 frames. ], batch size: 174, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:36:04,314 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-03 10:36:14,164 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:36:32,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-03 10:36:38,073 INFO [train.py:968] (0/2) Epoch 6, batch 36400, giga_loss[loss=0.2765, simple_loss=0.3539, pruned_loss=0.09953, over 28742.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3754, pruned_loss=0.1208, over 5707993.34 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3796, pruned_loss=0.1373, over 5714224.82 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3721, pruned_loss=0.1175, over 5700520.93 frames. ], batch size: 71, lr: 5.21e-03, grad_scale: 8.0
+2023-03-03 10:36:50,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 10:36:52,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:36:55,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:37:12,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.149e+02 1.322e+03 1.668e+03 2.204e+03 7.102e+03, threshold=3.336e+03, percent-clipped=15.0
+2023-03-03 10:37:19,125 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:37:19,212 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:37:20,957 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:37:21,326 INFO [train.py:968] (0/2) Epoch 6, batch 36450, giga_loss[loss=0.3053, simple_loss=0.3716, pruned_loss=0.1196, over 29074.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3777, pruned_loss=0.1247, over 5703780.83 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3798, pruned_loss=0.1373, over 5716468.17 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3749, pruned_loss=0.122, over 5695818.93 frames. ], batch size: 128, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:37:44,783 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:38:00,800 INFO [train.py:968] (0/2) Epoch 6, batch 36500, giga_loss[loss=0.3017, simple_loss=0.3655, pruned_loss=0.1189, over 28279.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3769, pruned_loss=0.1255, over 5705778.97 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3806, pruned_loss=0.138, over 5714673.24 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3739, pruned_loss=0.1221, over 5700251.74 frames. ], batch size: 368, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:38:12,724 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264226.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:38:31,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.223e+02 1.202e+03 1.786e+03 2.277e+03 9.229e+03, threshold=3.572e+03, percent-clipped=19.0
+2023-03-03 10:38:41,503 INFO [train.py:968] (0/2) Epoch 6, batch 36550, libri_loss[loss=0.3148, simple_loss=0.3755, pruned_loss=0.1271, over 29512.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3747, pruned_loss=0.1246, over 5683939.12 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3811, pruned_loss=0.1383, over 5696482.74 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3716, pruned_loss=0.1211, over 5696170.66 frames. ], batch size: 81, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:39:22,105 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 10:39:23,585 INFO [train.py:968] (0/2) Epoch 6, batch 36600, giga_loss[loss=0.2934, simple_loss=0.3601, pruned_loss=0.1134, over 28467.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3736, pruned_loss=0.1234, over 5692168.61 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3814, pruned_loss=0.1383, over 5701582.92 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3706, pruned_loss=0.1202, over 5697268.66 frames. ], batch size: 71, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:39:56,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.713e+02 1.217e+03 1.609e+03 2.332e+03 8.558e+03, threshold=3.218e+03, percent-clipped=10.0
+2023-03-03 10:40:00,820 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:40:06,763 INFO [train.py:968] (0/2) Epoch 6, batch 36650, giga_loss[loss=0.3611, simple_loss=0.405, pruned_loss=0.1586, over 28508.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3734, pruned_loss=0.123, over 5687341.24 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3821, pruned_loss=0.1388, over 5697979.18 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3701, pruned_loss=0.1195, over 5693627.14 frames. ], batch size: 85, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:40:07,713 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3196, 1.8203, 1.6219, 1.5394], device='cuda:0'), covar=tensor([0.0779, 0.0286, 0.0279, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0118, 0.0121, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0064, 0.0046, 0.0041, 0.0070], device='cuda:0')
+2023-03-03 10:40:56,304 INFO [train.py:968] (0/2) Epoch 6, batch 36700, giga_loss[loss=0.2789, simple_loss=0.3349, pruned_loss=0.1114, over 23579.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3693, pruned_loss=0.1194, over 5678464.60 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3821, pruned_loss=0.1387, over 5700221.38 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3667, pruned_loss=0.1166, over 5681331.06 frames. ], batch size: 705, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:41:23,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5613, 1.4862, 1.2745, 1.1727], device='cuda:0'), covar=tensor([0.0630, 0.0498, 0.0909, 0.0882], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0440, 0.0500, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 10:41:29,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.576e+02 1.038e+03 1.291e+03 1.895e+03 5.944e+03, threshold=2.582e+03, percent-clipped=6.0
+2023-03-03 10:41:41,857 INFO [train.py:968] (0/2) Epoch 6, batch 36750, giga_loss[loss=0.2539, simple_loss=0.3248, pruned_loss=0.09149, over 28966.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3628, pruned_loss=0.1156, over 5674671.45 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3822, pruned_loss=0.1387, over 5703401.17 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3604, pruned_loss=0.1131, over 5673929.33 frames. ], batch size: 227, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:42:04,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5476, 3.3798, 3.1738, 1.9491], device='cuda:0'), covar=tensor([0.0494, 0.0649, 0.0677, 0.1647], device='cuda:0'), in_proj_covar=tensor([0.0880, 0.0825, 0.0760, 0.0595], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 10:42:34,975 INFO [train.py:968] (0/2) Epoch 6, batch 36800, giga_loss[loss=0.2521, simple_loss=0.3227, pruned_loss=0.09069, over 27956.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3572, pruned_loss=0.1131, over 5649091.30 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3828, pruned_loss=0.1393, over 5693821.00 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3544, pruned_loss=0.1101, over 5656951.43 frames. ], batch size: 412, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:43:17,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.422e+02 8.879e+02 1.171e+03 1.623e+03 6.812e+03, threshold=2.342e+03, percent-clipped=7.0
+2023-03-03 10:43:27,050 INFO [train.py:968] (0/2) Epoch 6, batch 36850, giga_loss[loss=0.2435, simple_loss=0.3231, pruned_loss=0.08199, over 29032.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3527, pruned_loss=0.1105, over 5650113.36 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3833, pruned_loss=0.1395, over 5696992.90 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3496, pruned_loss=0.1076, over 5652950.93 frames. ], batch size: 128, lr: 5.20e-03, grad_scale: 4.0
+2023-03-03 10:44:02,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=264601.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 10:44:10,264 INFO [train.py:968] (0/2) Epoch 6, batch 36900, giga_loss[loss=0.2604, simple_loss=0.333, pruned_loss=0.09391, over 29002.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3532, pruned_loss=0.1101, over 5659749.67 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3841, pruned_loss=0.14, over 5691571.73 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3494, pruned_loss=0.1068, over 5665807.89 frames. ], batch size: 213, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:44:22,762 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:44:40,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.087e+02 1.139e+03 1.491e+03 2.288e+03 9.665e+03, threshold=2.981e+03, percent-clipped=23.0
+2023-03-03 10:44:48,684 INFO [train.py:968] (0/2) Epoch 6, batch 36950, giga_loss[loss=0.2888, simple_loss=0.3515, pruned_loss=0.113, over 28768.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3538, pruned_loss=0.1103, over 5669097.55 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3849, pruned_loss=0.14, over 5700193.69 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3488, pruned_loss=0.1063, over 5665059.64 frames. ], batch size: 119, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:45:27,080 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-03 10:45:30,643 INFO [train.py:968] (0/2) Epoch 6, batch 37000, giga_loss[loss=0.2136, simple_loss=0.2909, pruned_loss=0.06818, over 28218.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3506, pruned_loss=0.1078, over 5678976.80 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.385, pruned_loss=0.14, over 5694189.83 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3459, pruned_loss=0.1042, over 5681225.70 frames. ], batch size: 77, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:45:45,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=264728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:45:55,608 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264744.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:45:58,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264747.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:46:01,805 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.036e+02 1.079e+03 1.313e+03 1.937e+03 5.055e+03, threshold=2.626e+03, percent-clipped=9.0
+2023-03-03 10:46:04,562 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-03 10:46:10,573 INFO [train.py:968] (0/2) Epoch 6, batch 37050, giga_loss[loss=0.2594, simple_loss=0.3368, pruned_loss=0.09102, over 29020.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3487, pruned_loss=0.1068, over 5692273.38 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3858, pruned_loss=0.14, over 5693206.91 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3434, pruned_loss=0.103, over 5694786.05 frames. ], batch size: 136, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:46:24,017 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264776.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 10:46:51,213 INFO [train.py:968] (0/2) Epoch 6, batch 37100, giga_loss[loss=0.2388, simple_loss=0.3151, pruned_loss=0.08128, over 28990.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3462, pruned_loss=0.1057, over 5699286.01 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3862, pruned_loss=0.1399, over 5697625.22 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3406, pruned_loss=0.1019, over 5697172.51 frames. ], batch size: 213, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:47:23,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.719e+02 9.267e+02 1.244e+03 1.598e+03 5.466e+03, threshold=2.488e+03, percent-clipped=6.0
+2023-03-03 10:47:29,967 INFO [train.py:968] (0/2) Epoch 6, batch 37150, giga_loss[loss=0.2369, simple_loss=0.3146, pruned_loss=0.07961, over 28827.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3431, pruned_loss=0.104, over 5708316.24 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3862, pruned_loss=0.1396, over 5702272.77 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3374, pruned_loss=0.1003, over 5702325.89 frames. ], batch size: 112, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:47:38,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:47:39,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:47:40,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:47:55,286 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264893.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:48:05,101 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:48:05,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6329, 1.6111, 1.1954, 1.4078], device='cuda:0'), covar=tensor([0.0648, 0.0534, 0.0967, 0.0896], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0436, 0.0498, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 10:48:06,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4168, 1.5495, 1.3062, 1.0482], device='cuda:0'), covar=tensor([0.1434, 0.1131, 0.0921, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.1499, 0.1309, 0.1316, 0.1424], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 10:48:10,656 INFO [train.py:968] (0/2) Epoch 6, batch 37200, giga_loss[loss=0.2868, simple_loss=0.3399, pruned_loss=0.1168, over 28896.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.341, pruned_loss=0.1031, over 5711328.97 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3868, pruned_loss=0.1399, over 5701616.80 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3355, pruned_loss=0.09944, over 5707522.89 frames. ], batch size: 106, lr: 5.20e-03, grad_scale: 4.0
+2023-03-03 10:48:43,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.257e+02 8.895e+02 1.086e+03 1.847e+03 1.232e+04, threshold=2.172e+03, percent-clipped=15.0
+2023-03-03 10:48:49,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2615, 1.4347, 1.0931, 0.9751], device='cuda:0'), covar=tensor([0.1315, 0.0983, 0.0878, 0.1237], device='cuda:0'), in_proj_covar=tensor([0.1491, 0.1300, 0.1312, 0.1417], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 10:48:51,599 INFO [train.py:968] (0/2) Epoch 6, batch 37250, giga_loss[loss=0.2423, simple_loss=0.3188, pruned_loss=0.0829, over 28577.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3378, pruned_loss=0.1013, over 5713273.52 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3872, pruned_loss=0.14, over 5700035.23 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3326, pruned_loss=0.09795, over 5711908.94 frames. ], batch size: 307, lr: 5.20e-03, grad_scale: 4.0
+2023-03-03 10:49:23,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:49:29,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3593, 2.2136, 2.0390, 2.0380], device='cuda:0'), covar=tensor([0.1140, 0.1981, 0.1735, 0.1666], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0745, 0.0654, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 10:49:30,036 INFO [train.py:968] (0/2) Epoch 6, batch 37300, giga_loss[loss=0.2404, simple_loss=0.3085, pruned_loss=0.08612, over 28797.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3359, pruned_loss=0.1, over 5716877.30 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.3878, pruned_loss=0.14, over 5704038.40 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3302, pruned_loss=0.09655, over 5712252.69 frames. ], batch size: 119, lr: 5.20e-03, grad_scale: 4.0
+2023-03-03 10:50:02,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.374e+02 8.523e+02 1.088e+03 1.519e+03 3.521e+03, threshold=2.175e+03, percent-clipped=8.0
+2023-03-03 10:50:09,917 INFO [train.py:968] (0/2) Epoch 6, batch 37350, giga_loss[loss=0.2549, simple_loss=0.3352, pruned_loss=0.08735, over 28860.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3342, pruned_loss=0.09886, over 5708218.14 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.388, pruned_loss=0.1401, over 5697212.03 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3289, pruned_loss=0.09556, over 5710990.30 frames. ], batch size: 145, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:50:20,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5870, 2.2848, 1.6219, 0.6800], device='cuda:0'), covar=tensor([0.2182, 0.1177, 0.1897, 0.2426], device='cuda:0'), in_proj_covar=tensor([0.1430, 0.1328, 0.1382, 0.1179], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 10:50:42,387 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-03 10:50:49,438 INFO [train.py:968] (0/2) Epoch 6, batch 37400, giga_loss[loss=0.2334, simple_loss=0.3035, pruned_loss=0.08171, over 28606.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3349, pruned_loss=0.09955, over 5700498.82 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3892, pruned_loss=0.1409, over 5692075.62 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3284, pruned_loss=0.09538, over 5707623.07 frames. ], batch size: 60, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:51:18,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:51:20,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:51:23,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.146e+02 1.072e+03 1.383e+03 1.775e+03 8.207e+03, threshold=2.767e+03, percent-clipped=19.0
+2023-03-03 10:51:31,281 INFO [train.py:968] (0/2) Epoch 6, batch 37450, giga_loss[loss=0.2776, simple_loss=0.3451, pruned_loss=0.105, over 28968.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3362, pruned_loss=0.1003, over 5713254.93 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3894, pruned_loss=0.1407, over 5697516.43 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3296, pruned_loss=0.09615, over 5714254.10 frames. ], batch size: 213, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:51:44,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:52:03,331 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=265197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:52:14,027 INFO [train.py:968] (0/2) Epoch 6, batch 37500, giga_loss[loss=0.2574, simple_loss=0.324, pruned_loss=0.09545, over 28131.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.342, pruned_loss=0.1042, over 5705616.11 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3903, pruned_loss=0.1413, over 5689396.16 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3346, pruned_loss=0.09934, over 5714736.07 frames. ], batch size: 77, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:52:45,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:52:50,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.633e+02 1.058e+03 1.513e+03 2.084e+03 6.319e+03, threshold=3.027e+03, percent-clipped=8.0
+2023-03-03 10:52:59,453 INFO [train.py:968] (0/2) Epoch 6, batch 37550, giga_loss[loss=0.2908, simple_loss=0.3659, pruned_loss=0.1079, over 28949.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3498, pruned_loss=0.1094, over 5704783.77 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3903, pruned_loss=0.1411, over 5692535.88 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3426, pruned_loss=0.1047, over 5709636.00 frames. ], batch size: 136, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:53:07,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265268.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 10:53:35,165 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-03 10:53:46,011 INFO [train.py:968] (0/2) Epoch 6, batch 37600, giga_loss[loss=0.3247, simple_loss=0.3888, pruned_loss=0.1304, over 28724.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3576, pruned_loss=0.1146, over 5692770.93 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3901, pruned_loss=0.141, over 5688747.06 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.351, pruned_loss=0.1103, over 5700354.97 frames. ], batch size: 242, lr: 5.20e-03, grad_scale: 4.0
+2023-03-03 10:54:29,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9913, 2.5718, 1.5621, 1.5649], device='cuda:0'), covar=tensor([0.1297, 0.0768, 0.1066, 0.1163], device='cuda:0'), in_proj_covar=tensor([0.1497, 0.1312, 0.1325, 0.1418], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 10:54:30,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.274e+02 1.233e+03 1.632e+03 2.612e+03 9.048e+03, threshold=3.265e+03, percent-clipped=18.0
+2023-03-03 10:54:37,532 INFO [train.py:968] (0/2) Epoch 6, batch 37650, giga_loss[loss=0.3271, simple_loss=0.3842, pruned_loss=0.135, over 28827.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3636, pruned_loss=0.1178, over 5683631.22 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3904, pruned_loss=0.1409, over 5693145.28 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3575, pruned_loss=0.1137, over 5685483.36 frames. ], batch size: 99, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:55:01,957 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:55:04,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:55:22,072 INFO [train.py:968] (0/2) Epoch 6, batch 37700, giga_loss[loss=0.3094, simple_loss=0.3824, pruned_loss=0.1182, over 28900.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3682, pruned_loss=0.1191, over 5684830.19 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3904, pruned_loss=0.1409, over 5692417.71 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1157, over 5686808.19 frames. ], batch size: 186, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:55:22,329 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265411.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 10:55:25,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265414.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:55:33,095 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:55:50,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6333, 1.6897, 1.4343, 2.1163], device='cuda:0'), covar=tensor([0.2121, 0.2058, 0.2113, 0.1933], device='cuda:0'), in_proj_covar=tensor([0.1150, 0.0881, 0.1019, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 10:55:53,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265443.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 10:56:00,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.799e+02 1.110e+03 1.369e+03 1.859e+03 9.838e+03, threshold=2.738e+03, percent-clipped=4.0
+2023-03-03 10:56:07,533 INFO [train.py:968] (0/2) Epoch 6, batch 37750, giga_loss[loss=0.3708, simple_loss=0.424, pruned_loss=0.1588, over 28568.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3747, pruned_loss=0.1232, over 5683332.02 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3905, pruned_loss=0.1407, over 5695843.66 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.37, pruned_loss=0.1202, over 5681615.48 frames. ], batch size: 307, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:56:09,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3852, 1.8359, 1.2898, 0.6296], device='cuda:0'), covar=tensor([0.2510, 0.1274, 0.1893, 0.3008], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1354, 0.1412, 0.1194], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 10:56:48,667 INFO [train.py:968] (0/2) Epoch 6, batch 37800, giga_loss[loss=0.26, simple_loss=0.3347, pruned_loss=0.09265, over 28846.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3724, pruned_loss=0.1217, over 5688358.75 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3902, pruned_loss=0.1407, over 5698410.47 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3687, pruned_loss=0.119, over 5684497.41 frames. ], batch size: 136, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:57:21,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.796e+02 1.221e+03 1.573e+03 2.324e+03 7.955e+03, threshold=3.146e+03, percent-clipped=13.0
+2023-03-03 10:57:28,072 INFO [train.py:968] (0/2) Epoch 6, batch 37850, giga_loss[loss=0.3032, simple_loss=0.3699, pruned_loss=0.1183, over 28838.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.37, pruned_loss=0.1196, over 5691016.92 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3912, pruned_loss=0.1415, over 5694796.20 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3655, pruned_loss=0.1161, over 5690369.53 frames. ], batch size: 119, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:57:37,894 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:58:12,964 INFO [train.py:968] (0/2) Epoch 6, batch 37900, giga_loss[loss=0.2882, simple_loss=0.3612, pruned_loss=0.1076, over 28736.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3677, pruned_loss=0.1174, over 5693560.44 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.3917, pruned_loss=0.142, over 5696541.28 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3634, pruned_loss=0.1139, over 5691455.84 frames. ], batch size: 242, lr: 5.19e-03, grad_scale: 2.0
+2023-03-03 10:58:36,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8411, 1.0245, 4.0166, 3.2113], device='cuda:0'), covar=tensor([0.1974, 0.2705, 0.0373, 0.0675], device='cuda:0'), in_proj_covar=tensor([0.0586, 0.0538, 0.0769, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 10:58:50,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.400e+02 1.194e+03 1.480e+03 2.445e+03 1.159e+04, threshold=2.960e+03, percent-clipped=15.0
+2023-03-03 10:58:55,625 INFO [train.py:968] (0/2) Epoch 6, batch 37950, giga_loss[loss=0.2908, simple_loss=0.3614, pruned_loss=0.1102, over 28274.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3679, pruned_loss=0.1174, over 5692931.61 frames. ], libri_tot_loss[loss=0.3383, simple_loss=0.3921, pruned_loss=0.1423, over 5690747.91 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3632, pruned_loss=0.1133, over 5696180.05 frames. ], batch size: 65, lr: 5.19e-03, grad_scale: 2.0
+2023-03-03 10:59:03,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=265671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:59:19,786 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 10:59:37,557 INFO [train.py:968] (0/2) Epoch 6, batch 38000, giga_loss[loss=0.2982, simple_loss=0.37, pruned_loss=0.1132, over 28714.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3705, pruned_loss=0.1189, over 5696330.83 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.3919, pruned_loss=0.1422, over 5694415.47 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3664, pruned_loss=0.1152, over 5695576.29 frames. ], batch size: 284, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 10:59:42,051 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:59:44,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:00:08,181 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:00:13,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.662e+02 1.110e+03 1.418e+03 1.822e+03 4.912e+03, threshold=2.836e+03, percent-clipped=3.0
+2023-03-03 11:00:19,762 INFO [train.py:968] (0/2) Epoch 6, batch 38050, giga_loss[loss=0.3344, simple_loss=0.3848, pruned_loss=0.142, over 23645.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3724, pruned_loss=0.1203, over 5699189.29 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.3915, pruned_loss=0.1419, over 5698586.12 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3689, pruned_loss=0.1169, over 5694878.90 frames. ], batch size: 705, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:01:03,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8129, 1.4680, 1.8203, 1.5909], device='cuda:0'), covar=tensor([0.1657, 0.2539, 0.1262, 0.1346], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0722, 0.0815, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 11:01:05,639 INFO [train.py:968] (0/2) Epoch 6, batch 38100, giga_loss[loss=0.3216, simple_loss=0.3838, pruned_loss=0.1297, over 28930.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3751, pruned_loss=0.1225, over 5697410.77 frames. ], libri_tot_loss[loss=0.3378, simple_loss=0.3917, pruned_loss=0.142, over 5700870.60 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3718, pruned_loss=0.1194, over 5691984.40 frames. ], batch size: 227, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:01:42,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.615e+02 1.247e+03 1.663e+03 2.126e+03 5.548e+03, threshold=3.325e+03, percent-clipped=13.0
+2023-03-03 11:01:49,802 INFO [train.py:968] (0/2) Epoch 6, batch 38150, giga_loss[loss=0.2759, simple_loss=0.3466, pruned_loss=0.1026, over 28636.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3744, pruned_loss=0.1223, over 5702004.02 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.3918, pruned_loss=0.142, over 5704125.04 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3715, pruned_loss=0.1195, over 5694618.80 frames. ], batch size: 78, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:02:28,808 INFO [train.py:968] (0/2) Epoch 6, batch 38200, giga_loss[loss=0.3157, simple_loss=0.3718, pruned_loss=0.1298, over 27573.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3741, pruned_loss=0.1222, over 5704677.10 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.392, pruned_loss=0.142, over 5706346.12 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3712, pruned_loss=0.1195, over 5696818.62 frames. ], batch size: 472, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:02:39,879 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=265923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:03:05,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.637e+02 1.158e+03 1.403e+03 1.912e+03 5.054e+03, threshold=2.806e+03, percent-clipped=4.0
+2023-03-03 11:03:11,596 INFO [train.py:968] (0/2) Epoch 6, batch 38250, giga_loss[loss=0.2806, simple_loss=0.3595, pruned_loss=0.1008, over 28993.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3741, pruned_loss=0.1219, over 5699266.21 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.3922, pruned_loss=0.1421, over 5706458.16 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3713, pruned_loss=0.1194, over 5692997.30 frames. ], batch size: 145, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:03:24,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5047, 1.4698, 1.2821, 1.7844], device='cuda:0'), covar=tensor([0.2029, 0.2028, 0.2015, 0.2059], device='cuda:0'), in_proj_covar=tensor([0.1147, 0.0880, 0.1019, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 11:03:42,654 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-266000.pt
+2023-03-03 11:03:47,839 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:03:51,062 INFO [train.py:968] (0/2) Epoch 6, batch 38300, giga_loss[loss=0.3201, simple_loss=0.3855, pruned_loss=0.1274, over 28486.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3751, pruned_loss=0.1213, over 5710038.18 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.3925, pruned_loss=0.1423, over 5712002.72 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3721, pruned_loss=0.1185, over 5699972.22 frames. ], batch size: 65, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:04:03,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3003, 1.5697, 1.2084, 1.4970], device='cuda:0'), covar=tensor([0.2103, 0.2083, 0.2141, 0.1825], device='cuda:0'), in_proj_covar=tensor([0.1153, 0.0887, 0.1026, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 11:04:12,716 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 11:04:24,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:04:29,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.113e+02 1.172e+03 1.436e+03 2.115e+03 8.329e+03, threshold=2.872e+03, percent-clipped=13.0
+2023-03-03 11:04:36,315 INFO [train.py:968] (0/2) Epoch 6, batch 38350, giga_loss[loss=0.3019, simple_loss=0.3704, pruned_loss=0.1167, over 28984.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3752, pruned_loss=0.1205, over 5715202.24 frames. ], libri_tot_loss[loss=0.3386, simple_loss=0.3925, pruned_loss=0.1423, over 5716079.58 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3723, pruned_loss=0.1178, over 5703445.03 frames. ], batch size: 136, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:04:47,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4797, 1.7212, 1.3325, 2.0320], device='cuda:0'), covar=tensor([0.2245, 0.2062, 0.2113, 0.2063], device='cuda:0'), in_proj_covar=tensor([0.1154, 0.0885, 0.1026, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 11:04:49,948 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-03 11:05:16,184 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:05:17,994 INFO [train.py:968] (0/2) Epoch 6, batch 38400, libri_loss[loss=0.3906, simple_loss=0.4317, pruned_loss=0.1747, over 29237.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3735, pruned_loss=0.1194, over 5711849.56 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3929, pruned_loss=0.1426, over 5718794.20 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3706, pruned_loss=0.1167, over 5700207.24 frames. ], batch size: 97, lr: 5.19e-03, grad_scale: 8.0
+2023-03-03 11:05:52,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.980e+02 1.061e+03 1.335e+03 1.849e+03 9.907e+03, threshold=2.669e+03, percent-clipped=11.0
+2023-03-03 11:05:58,820 INFO [train.py:968] (0/2) Epoch 6, batch 38450, giga_loss[loss=0.2692, simple_loss=0.3462, pruned_loss=0.09606, over 28970.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3705, pruned_loss=0.1178, over 5710584.29 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3928, pruned_loss=0.1426, over 5719841.52 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.368, pruned_loss=0.1153, over 5700344.55 frames. ], batch size: 136, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:06:23,754 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:06:25,793 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:06:42,001 INFO [train.py:968] (0/2) Epoch 6, batch 38500, giga_loss[loss=0.3397, simple_loss=0.3863, pruned_loss=0.1465, over 26590.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3684, pruned_loss=0.1164, over 5711941.27 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.393, pruned_loss=0.1428, over 5722858.25 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3658, pruned_loss=0.1138, over 5700843.55 frames. ], batch size: 555, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:06:48,824 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:07:15,177 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.585e+02 1.006e+03 1.277e+03 1.732e+03 4.136e+03, threshold=2.555e+03, percent-clipped=9.0
+2023-03-03 11:07:21,393 INFO [train.py:968] (0/2) Epoch 6, batch 38550, giga_loss[loss=0.255, simple_loss=0.3357, pruned_loss=0.08711, over 29118.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3682, pruned_loss=0.117, over 5707239.04 frames. ], libri_tot_loss[loss=0.3389, simple_loss=0.3927, pruned_loss=0.1425, over 5718940.97 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3658, pruned_loss=0.1144, over 5701040.34 frames. ], batch size: 155, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:07:24,316 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:07:50,954 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:07:59,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:08:00,441 INFO [train.py:968] (0/2) Epoch 6, batch 38600, giga_loss[loss=0.2631, simple_loss=0.3474, pruned_loss=0.0894, over 28870.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3688, pruned_loss=0.1171, over 5708924.45 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.3933, pruned_loss=0.1429, over 5720827.20 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3656, pruned_loss=0.114, over 5701766.76 frames. ], batch size: 145, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:08:14,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.07 vs. limit=2.0
+2023-03-03 11:08:32,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.981e+02 9.840e+02 1.294e+03 1.944e+03 1.018e+04, threshold=2.588e+03, percent-clipped=16.0
+2023-03-03 11:08:38,016 INFO [train.py:968] (0/2) Epoch 6, batch 38650, giga_loss[loss=0.2988, simple_loss=0.3696, pruned_loss=0.1141, over 28492.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3695, pruned_loss=0.1173, over 5714821.85 frames. ], libri_tot_loss[loss=0.3403, simple_loss=0.3938, pruned_loss=0.1434, over 5726242.80 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3659, pruned_loss=0.1137, over 5704036.89 frames. ], batch size: 71, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:08:55,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:09:24,363 INFO [train.py:968] (0/2) Epoch 6, batch 38700, libri_loss[loss=0.3504, simple_loss=0.4043, pruned_loss=0.1482, over 27568.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3687, pruned_loss=0.1158, over 5701454.90 frames. ], libri_tot_loss[loss=0.3409, simple_loss=0.3944, pruned_loss=0.1437, over 5708507.19 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3645, pruned_loss=0.1119, over 5709071.25 frames. ], batch size: 115, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:09:25,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4499, 1.7707, 1.3369, 1.0666], device='cuda:0'), covar=tensor([0.1359, 0.1036, 0.0952, 0.1247], device='cuda:0'), in_proj_covar=tensor([0.1487, 0.1311, 0.1305, 0.1389], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 11:09:34,712 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-03 11:09:47,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:09:49,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:09:57,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.295e+02 9.213e+02 1.179e+03 1.562e+03 5.463e+03, threshold=2.359e+03, percent-clipped=11.0
+2023-03-03 11:10:02,177 INFO [train.py:968] (0/2) Epoch 6, batch 38750, giga_loss[loss=0.2778, simple_loss=0.3523, pruned_loss=0.1017, over 28874.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3678, pruned_loss=0.1151, over 5708279.76 frames. ], libri_tot_loss[loss=0.3411, simple_loss=0.3944, pruned_loss=0.1439, over 5710668.10 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.364, pruned_loss=0.1114, over 5712319.43 frames. ], batch size: 199, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:10:13,545 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:10:21,409 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:10:24,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3819, 2.0756, 1.7510, 1.6440], device='cuda:0'), covar=tensor([0.0638, 0.0662, 0.0883, 0.1029], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0438, 0.0503, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 11:10:44,042 INFO [train.py:968] (0/2) Epoch 6, batch 38800, giga_loss[loss=0.291, simple_loss=0.3595, pruned_loss=0.1113, over 29058.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3654, pruned_loss=0.1141, over 5700668.35 frames. ], libri_tot_loss[loss=0.3403, simple_loss=0.3938, pruned_loss=0.1434, over 5713231.75 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3625, pruned_loss=0.111, over 5701484.98 frames. ], batch size: 128, lr: 5.19e-03, grad_scale: 8.0
+2023-03-03 11:10:54,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:10:56,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:11:17,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.684e+02 1.036e+03 1.315e+03 1.823e+03 5.185e+03, threshold=2.630e+03, percent-clipped=16.0
+2023-03-03 11:11:20,339 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:11:22,917 INFO [train.py:968] (0/2) Epoch 6, batch 38850, giga_loss[loss=0.2613, simple_loss=0.3297, pruned_loss=0.09646, over 28557.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3625, pruned_loss=0.1127, over 5707024.66 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3931, pruned_loss=0.1428, over 5719871.36 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3597, pruned_loss=0.1097, over 5701290.11 frames. ], batch size: 85, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:11:28,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4636, 1.5904, 1.3121, 1.5618], device='cuda:0'), covar=tensor([0.0755, 0.0311, 0.0310, 0.0778], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0120, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0064, 0.0046, 0.0041, 0.0070], device='cuda:0')
+2023-03-03 11:11:57,964 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:12:02,834 INFO [train.py:968] (0/2) Epoch 6, batch 38900, giga_loss[loss=0.2677, simple_loss=0.335, pruned_loss=0.1002, over 28367.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3604, pruned_loss=0.1119, over 5705378.68 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3936, pruned_loss=0.1431, over 5716744.20 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3568, pruned_loss=0.1084, over 5702884.74 frames. ], batch size: 77, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:12:05,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9682, 3.7891, 3.5698, 1.7436], device='cuda:0'), covar=tensor([0.0535, 0.0604, 0.0697, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.0896, 0.0840, 0.0764, 0.0600], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 11:12:08,441 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266619.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 11:12:14,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:12:15,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:12:24,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:12:36,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.456e+02 1.120e+03 1.328e+03 1.720e+03 6.640e+03, threshold=2.657e+03, percent-clipped=10.0
+2023-03-03 11:12:39,728 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:12:41,797 INFO [train.py:968] (0/2) Epoch 6, batch 38950, giga_loss[loss=0.2443, simple_loss=0.3233, pruned_loss=0.08263, over 28925.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3589, pruned_loss=0.1114, over 5707200.95 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3931, pruned_loss=0.1427, over 5720902.44 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3559, pruned_loss=0.1084, over 5701394.33 frames. ], batch size: 186, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:13:01,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:13:21,567 INFO [train.py:968] (0/2) Epoch 6, batch 39000, giga_loss[loss=0.2861, simple_loss=0.3539, pruned_loss=0.1091, over 28959.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3588, pruned_loss=0.1118, over 5707598.78 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3934, pruned_loss=0.1427, over 5713968.02 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3547, pruned_loss=0.108, over 5709511.01 frames. ], batch size: 106, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:13:21,573 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 11:13:30,554 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2299, simple_loss=0.3329, pruned_loss=0.06347, over 944034.00 frames. 
+2023-03-03 11:13:30,555 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 11:13:52,288 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:14:07,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.101e+02 1.064e+03 1.340e+03 1.617e+03 6.608e+03, threshold=2.679e+03, percent-clipped=6.0
+2023-03-03 11:14:11,449 INFO [train.py:968] (0/2) Epoch 6, batch 39050, giga_loss[loss=0.238, simple_loss=0.3154, pruned_loss=0.08028, over 28536.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3564, pruned_loss=0.111, over 5707589.33 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3934, pruned_loss=0.1429, over 5717087.88 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3527, pruned_loss=0.1075, over 5706189.20 frames. ], batch size: 60, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:14:20,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1879, 1.2990, 1.3717, 1.3176], device='cuda:0'), covar=tensor([0.1142, 0.1310, 0.1606, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0730, 0.0646, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 11:14:29,069 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:14:31,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:14:50,409 INFO [train.py:968] (0/2) Epoch 6, batch 39100, giga_loss[loss=0.2636, simple_loss=0.336, pruned_loss=0.09564, over 28890.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3546, pruned_loss=0.1101, over 5717005.14 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.393, pruned_loss=0.1425, over 5725587.45 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3501, pruned_loss=0.106, over 5707660.13 frames. ], batch size: 213, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:14:54,234 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:15:03,944 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:15:05,686 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:15:27,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.277e+02 1.082e+03 1.427e+03 2.322e+03 1.431e+04, threshold=2.855e+03, percent-clipped=22.0
+2023-03-03 11:15:28,507 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:15:31,127 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:15:31,626 INFO [train.py:968] (0/2) Epoch 6, batch 39150, giga_loss[loss=0.2939, simple_loss=0.3511, pruned_loss=0.1183, over 28490.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3513, pruned_loss=0.1087, over 5718619.77 frames. ], libri_tot_loss[loss=0.3386, simple_loss=0.3926, pruned_loss=0.1423, over 5728655.67 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3475, pruned_loss=0.1052, over 5708267.65 frames. ], batch size: 71, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:16:16,545 INFO [train.py:968] (0/2) Epoch 6, batch 39200, giga_loss[loss=0.2382, simple_loss=0.3162, pruned_loss=0.08009, over 28973.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3508, pruned_loss=0.1089, over 5705672.38 frames. ], libri_tot_loss[loss=0.3383, simple_loss=0.3924, pruned_loss=0.1421, over 5719863.82 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3473, pruned_loss=0.1057, over 5705327.28 frames. ], batch size: 136, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:16:55,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.982e+02 9.377e+02 1.212e+03 1.514e+03 4.661e+03, threshold=2.425e+03, percent-clipped=3.0
+2023-03-03 11:16:59,062 INFO [train.py:968] (0/2) Epoch 6, batch 39250, giga_loss[loss=0.3112, simple_loss=0.3796, pruned_loss=0.1214, over 28248.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.353, pruned_loss=0.11, over 5705726.66 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.3923, pruned_loss=0.142, over 5720069.50 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3493, pruned_loss=0.1068, over 5705117.77 frames. ], batch size: 368, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:17:20,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:17:28,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266994.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 11:17:45,963 INFO [train.py:968] (0/2) Epoch 6, batch 39300, giga_loss[loss=0.2753, simple_loss=0.3419, pruned_loss=0.1044, over 28595.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3553, pruned_loss=0.1104, over 5710289.71 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.3925, pruned_loss=0.1423, over 5722482.29 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3517, pruned_loss=0.1072, over 5707466.46 frames. ], batch size: 78, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:18:05,925 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-03 11:18:25,190 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.212e+02 1.011e+03 1.230e+03 1.608e+03 3.760e+03, threshold=2.459e+03, percent-clipped=7.0
+2023-03-03 11:18:27,883 INFO [train.py:968] (0/2) Epoch 6, batch 39350, giga_loss[loss=0.3053, simple_loss=0.3818, pruned_loss=0.1144, over 28975.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3583, pruned_loss=0.1117, over 5698330.20 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.3926, pruned_loss=0.1424, over 5716862.98 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3542, pruned_loss=0.1081, over 5700837.92 frames. ], batch size: 145, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:18:37,187 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:18:52,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3062, 1.2706, 4.6397, 3.5275], device='cuda:0'), covar=tensor([0.1696, 0.2465, 0.0297, 0.0669], device='cuda:0'), in_proj_covar=tensor([0.0573, 0.0526, 0.0753, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 11:19:01,990 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:19:13,423 INFO [train.py:968] (0/2) Epoch 6, batch 39400, giga_loss[loss=0.2753, simple_loss=0.3523, pruned_loss=0.09914, over 28829.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3582, pruned_loss=0.1102, over 5690684.71 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3928, pruned_loss=0.1426, over 5707405.66 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3548, pruned_loss=0.1071, over 5700274.77 frames. ], batch size: 199, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:19:16,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:19:25,037 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:19:26,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:19:37,275 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267137.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 11:19:39,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267140.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 11:19:51,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.892e+02 1.147e+03 1.408e+03 2.207e+03 1.263e+04, threshold=2.817e+03, percent-clipped=21.0
+2023-03-03 11:19:51,742 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:19:54,274 INFO [train.py:968] (0/2) Epoch 6, batch 39450, giga_loss[loss=0.2669, simple_loss=0.3496, pruned_loss=0.09212, over 28669.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.359, pruned_loss=0.1106, over 5687878.62 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3933, pruned_loss=0.1428, over 5714672.92 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3544, pruned_loss=0.1066, over 5688055.45 frames. ], batch size: 262, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:19:59,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267169.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 11:20:34,595 INFO [train.py:968] (0/2) Epoch 6, batch 39500, giga_loss[loss=0.3507, simple_loss=0.3996, pruned_loss=0.1509, over 26715.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3601, pruned_loss=0.1113, over 5680472.38 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3937, pruned_loss=0.1432, over 5699027.49 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3556, pruned_loss=0.1073, over 5695274.94 frames. ], batch size: 555, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:20:47,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8829, 1.5587, 1.2046, 1.3577], device='cuda:0'), covar=tensor([0.0578, 0.0672, 0.1037, 0.0984], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0441, 0.0496, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 11:20:50,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:21:12,597 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 11:21:12,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.357e+02 1.116e+03 1.444e+03 1.962e+03 6.061e+03, threshold=2.887e+03, percent-clipped=14.0
+2023-03-03 11:21:13,067 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:21:14,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:21:15,153 INFO [train.py:968] (0/2) Epoch 6, batch 39550, giga_loss[loss=0.3183, simple_loss=0.3861, pruned_loss=0.1252, over 28754.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3616, pruned_loss=0.1128, over 5677694.85 frames. ], libri_tot_loss[loss=0.3405, simple_loss=0.3941, pruned_loss=0.1435, over 5693420.52 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3567, pruned_loss=0.1085, over 5694543.28 frames. ], batch size: 284, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:21:18,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1828, 1.3821, 1.1715, 0.9464], device='cuda:0'), covar=tensor([0.2154, 0.2011, 0.2203, 0.1845], device='cuda:0'), in_proj_covar=tensor([0.1161, 0.0879, 0.1025, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 11:21:39,801 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:21:59,281 INFO [train.py:968] (0/2) Epoch 6, batch 39600, giga_loss[loss=0.3112, simple_loss=0.383, pruned_loss=0.1197, over 28845.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3634, pruned_loss=0.1141, over 5679766.64 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3942, pruned_loss=0.1436, over 5697085.95 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3588, pruned_loss=0.1101, over 5689754.43 frames. ], batch size: 174, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:22:01,053 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:22:37,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.626e+02 1.231e+03 1.644e+03 2.374e+03 1.440e+04, threshold=3.289e+03, percent-clipped=15.0
+2023-03-03 11:22:41,425 INFO [train.py:968] (0/2) Epoch 6, batch 39650, giga_loss[loss=0.2954, simple_loss=0.3682, pruned_loss=0.1113, over 28803.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3662, pruned_loss=0.116, over 5692368.22 frames. ], libri_tot_loss[loss=0.3411, simple_loss=0.3946, pruned_loss=0.1438, over 5698886.83 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3617, pruned_loss=0.112, over 5698345.19 frames. ], batch size: 262, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:22:53,225 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:22:55,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:23:17,849 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:23:20,540 INFO [train.py:968] (0/2) Epoch 6, batch 39700, giga_loss[loss=0.296, simple_loss=0.3688, pruned_loss=0.1116, over 28888.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3683, pruned_loss=0.1164, over 5701927.80 frames. ], libri_tot_loss[loss=0.3406, simple_loss=0.3942, pruned_loss=0.1435, over 5699643.43 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3644, pruned_loss=0.1129, over 5705944.99 frames. ], batch size: 227, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:23:40,343 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3722, 1.7174, 1.2922, 1.4706], device='cuda:0'), covar=tensor([0.2122, 0.1944, 0.2109, 0.1877], device='cuda:0'), in_proj_covar=tensor([0.1155, 0.0874, 0.1018, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 11:23:49,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:23:57,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.146e+02 1.078e+03 1.339e+03 1.750e+03 4.320e+03, threshold=2.679e+03, percent-clipped=4.0
+2023-03-03 11:24:01,376 INFO [train.py:968] (0/2) Epoch 6, batch 39750, giga_loss[loss=0.2906, simple_loss=0.3636, pruned_loss=0.1088, over 28217.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3673, pruned_loss=0.1153, over 5707123.10 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.3944, pruned_loss=0.1436, over 5701945.27 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3635, pruned_loss=0.1119, over 5708374.86 frames. ], batch size: 368, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:24:10,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6231, 1.4568, 5.1271, 3.6050], device='cuda:0'), covar=tensor([0.1456, 0.2190, 0.0283, 0.0592], device='cuda:0'), in_proj_covar=tensor([0.0576, 0.0532, 0.0761, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 11:24:11,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:24:13,591 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:24:27,410 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:24:43,862 INFO [train.py:968] (0/2) Epoch 6, batch 39800, giga_loss[loss=0.2891, simple_loss=0.354, pruned_loss=0.1121, over 28475.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3667, pruned_loss=0.115, over 5710247.06 frames. ], libri_tot_loss[loss=0.3405, simple_loss=0.3942, pruned_loss=0.1434, over 5706230.14 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3634, pruned_loss=0.1119, over 5707504.45 frames. ], batch size: 71, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:25:21,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.842e+02 1.088e+03 1.318e+03 1.884e+03 6.479e+03, threshold=2.635e+03, percent-clipped=8.0
+2023-03-03 11:25:23,971 INFO [train.py:968] (0/2) Epoch 6, batch 39850, giga_loss[loss=0.2823, simple_loss=0.3494, pruned_loss=0.1076, over 28973.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3663, pruned_loss=0.1147, over 5714954.06 frames. ], libri_tot_loss[loss=0.3402, simple_loss=0.394, pruned_loss=0.1432, over 5708068.57 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3637, pruned_loss=0.1121, over 5711214.40 frames. ], batch size: 106, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:25:45,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3049, 1.5552, 1.3153, 1.0968], device='cuda:0'), covar=tensor([0.1407, 0.1018, 0.0748, 0.1088], device='cuda:0'), in_proj_covar=tensor([0.1494, 0.1328, 0.1313, 0.1397], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 11:25:47,277 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:25:49,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:26:03,445 INFO [train.py:968] (0/2) Epoch 6, batch 39900, giga_loss[loss=0.292, simple_loss=0.3579, pruned_loss=0.1131, over 28913.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3638, pruned_loss=0.1131, over 5718242.25 frames. ], libri_tot_loss[loss=0.3402, simple_loss=0.394, pruned_loss=0.1432, over 5710178.41 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3614, pruned_loss=0.1109, over 5713581.47 frames. ], batch size: 199, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:26:04,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5996, 1.9716, 1.8414, 1.5942], device='cuda:0'), covar=tensor([0.1683, 0.1978, 0.1354, 0.1398], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0721, 0.0812, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 11:26:06,308 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:26:09,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:26:11,960 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:26:31,017 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:26:40,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.222e+02 9.952e+02 1.260e+03 1.674e+03 5.273e+03, threshold=2.521e+03, percent-clipped=8.0
+2023-03-03 11:26:43,122 INFO [train.py:968] (0/2) Epoch 6, batch 39950, giga_loss[loss=0.2523, simple_loss=0.3298, pruned_loss=0.0874, over 28752.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3607, pruned_loss=0.1116, over 5711767.61 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3934, pruned_loss=0.1427, over 5707663.55 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3582, pruned_loss=0.1091, over 5710123.61 frames. ], batch size: 284, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:27:02,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:27:04,085 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-03 11:27:20,969 INFO [train.py:968] (0/2) Epoch 6, batch 40000, giga_loss[loss=0.2757, simple_loss=0.348, pruned_loss=0.1017, over 28876.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3581, pruned_loss=0.1099, over 5717730.30 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.394, pruned_loss=0.143, over 5712831.28 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3549, pruned_loss=0.107, over 5712246.77 frames. ], batch size: 186, lr: 5.17e-03, grad_scale: 8.0
+2023-03-03 11:27:56,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.094e+02 1.016e+03 1.315e+03 1.819e+03 8.144e+03, threshold=2.631e+03, percent-clipped=9.0
+2023-03-03 11:27:59,297 INFO [train.py:968] (0/2) Epoch 6, batch 40050, giga_loss[loss=0.2729, simple_loss=0.353, pruned_loss=0.09636, over 29088.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3589, pruned_loss=0.1097, over 5709317.05 frames. ], libri_tot_loss[loss=0.3402, simple_loss=0.3942, pruned_loss=0.1431, over 5706520.39 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3554, pruned_loss=0.1065, over 5710987.67 frames. ], batch size: 128, lr: 5.17e-03, grad_scale: 8.0
+2023-03-03 11:28:08,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-03 11:28:41,931 INFO [train.py:968] (0/2) Epoch 6, batch 40100, giga_loss[loss=0.3327, simple_loss=0.3939, pruned_loss=0.1358, over 28819.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3604, pruned_loss=0.1098, over 5704613.88 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3937, pruned_loss=0.1429, over 5708573.01 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3575, pruned_loss=0.1071, over 5704248.33 frames. ], batch size: 199, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:28:58,371 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:00,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:13,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:18,741 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.103e+02 1.154e+03 1.453e+03 2.184e+03 8.006e+03, threshold=2.906e+03, percent-clipped=18.0
+2023-03-03 11:29:20,974 INFO [train.py:968] (0/2) Epoch 6, batch 40150, giga_loss[loss=0.2976, simple_loss=0.3645, pruned_loss=0.1153, over 29009.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3619, pruned_loss=0.1112, over 5709547.77 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3935, pruned_loss=0.1426, over 5712176.05 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3592, pruned_loss=0.1087, over 5706034.15 frames. ], batch size: 164, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:29:22,659 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:23,510 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-03 11:29:24,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:28,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9685, 2.4044, 1.6557, 1.6453], device='cuda:0'), covar=tensor([0.1775, 0.1052, 0.1292, 0.1366], device='cuda:0'), in_proj_covar=tensor([0.1488, 0.1328, 0.1310, 0.1394], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 11:29:42,026 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:30:02,061 INFO [train.py:968] (0/2) Epoch 6, batch 40200, giga_loss[loss=0.275, simple_loss=0.3483, pruned_loss=0.1008, over 28874.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3609, pruned_loss=0.1121, over 5715413.55 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.394, pruned_loss=0.1429, over 5715362.43 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3577, pruned_loss=0.1092, over 5709836.28 frames. ], batch size: 227, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:30:06,081 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4822, 2.0899, 1.5726, 0.6682], device='cuda:0'), covar=tensor([0.2670, 0.1412, 0.2256, 0.3513], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1346, 0.1403, 0.1192], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 11:30:40,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.496e+02 1.040e+03 1.308e+03 1.724e+03 7.991e+03, threshold=2.616e+03, percent-clipped=2.0
+2023-03-03 11:30:43,196 INFO [train.py:968] (0/2) Epoch 6, batch 40250, giga_loss[loss=0.3374, simple_loss=0.3829, pruned_loss=0.1459, over 27690.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3573, pruned_loss=0.1111, over 5720397.08 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3941, pruned_loss=0.1429, over 5716335.62 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3546, pruned_loss=0.1087, over 5715090.97 frames. ], batch size: 472, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:31:10,742 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:31:12,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:31:15,850 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-268000.pt
+2023-03-03 11:31:23,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=268009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:31:24,710 INFO [train.py:968] (0/2) Epoch 6, batch 40300, giga_loss[loss=0.2674, simple_loss=0.3432, pruned_loss=0.0958, over 28925.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3554, pruned_loss=0.1111, over 5716159.88 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3941, pruned_loss=0.1429, over 5719138.44 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3528, pruned_loss=0.1088, over 5709620.28 frames. ], batch size: 174, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:31:26,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=268012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:31:32,149 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-03 11:31:38,085 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=268025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:31:50,999 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=268041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:32:05,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.237e+02 1.214e+03 1.617e+03 2.371e+03 8.532e+03, threshold=3.235e+03, percent-clipped=23.0
+2023-03-03 11:32:07,793 INFO [train.py:968] (0/2) Epoch 6, batch 40350, giga_loss[loss=0.2514, simple_loss=0.3297, pruned_loss=0.08649, over 28988.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3559, pruned_loss=0.1122, over 5709183.90 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3941, pruned_loss=0.1428, over 5721132.04 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3531, pruned_loss=0.1099, over 5702076.67 frames. ], batch size: 213, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:32:12,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0591, 1.0448, 4.0211, 3.0974], device='cuda:0'), covar=tensor([0.1669, 0.2474, 0.0368, 0.0758], device='cuda:0'), in_proj_covar=tensor([0.0576, 0.0533, 0.0769, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 11:32:36,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3583, 1.3974, 1.5390, 1.3238], device='cuda:0'), covar=tensor([0.1078, 0.1352, 0.1481, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0741, 0.0657, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 11:32:39,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5082, 4.3280, 4.0618, 1.9941], device='cuda:0'), covar=tensor([0.0458, 0.0537, 0.0760, 0.2045], device='cuda:0'), in_proj_covar=tensor([0.0913, 0.0845, 0.0783, 0.0606], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 11:32:47,100 INFO [train.py:968] (0/2) Epoch 6, batch 40400, giga_loss[loss=0.2403, simple_loss=0.3176, pruned_loss=0.08155, over 28461.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3523, pruned_loss=0.1101, over 5712830.87 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.3939, pruned_loss=0.1427, over 5723116.31 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3494, pruned_loss=0.1077, over 5705009.67 frames. ], batch size: 60, lr: 5.17e-03, grad_scale: 8.0
+2023-03-03 11:33:24,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.810e+02 1.141e+03 1.405e+03 2.041e+03 3.808e+03, threshold=2.811e+03, percent-clipped=3.0
+2023-03-03 11:33:26,582 INFO [train.py:968] (0/2) Epoch 6, batch 40450, giga_loss[loss=0.2286, simple_loss=0.3083, pruned_loss=0.07441, over 29033.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3489, pruned_loss=0.1088, over 5702021.06 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.394, pruned_loss=0.1427, over 5713397.04 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3453, pruned_loss=0.1059, over 5704655.01 frames. ], batch size: 155, lr: 5.17e-03, grad_scale: 8.0
+2023-03-03 11:33:46,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0259, 1.3233, 1.0459, 0.2798], device='cuda:0'), covar=tensor([0.1774, 0.1567, 0.2678, 0.3228], device='cuda:0'), in_proj_covar=tensor([0.1456, 0.1360, 0.1412, 0.1203], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 11:34:09,387 INFO [train.py:968] (0/2) Epoch 6, batch 40500, giga_loss[loss=0.2644, simple_loss=0.3322, pruned_loss=0.09832, over 28926.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3452, pruned_loss=0.1066, over 5707703.10 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3938, pruned_loss=0.1426, over 5716114.49 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3419, pruned_loss=0.104, over 5707110.03 frames. ], batch size: 136, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:34:11,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8812, 1.1039, 0.9908, 0.7365], device='cuda:0'), covar=tensor([0.1140, 0.1075, 0.0760, 0.1057], device='cuda:0'), in_proj_covar=tensor([0.1492, 0.1333, 0.1318, 0.1403], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 11:34:12,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9093, 4.4294, 2.1425, 1.8368], device='cuda:0'), covar=tensor([0.0740, 0.0280, 0.0738, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0332, 0.0483, 0.0312, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 11:34:34,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0782, 1.1602, 3.9690, 3.1152], device='cuda:0'), covar=tensor([0.1677, 0.2457, 0.0389, 0.0769], device='cuda:0'), in_proj_covar=tensor([0.0578, 0.0534, 0.0769, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 11:34:45,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.161e+02 1.148e+03 1.547e+03 2.149e+03 8.606e+03, threshold=3.095e+03, percent-clipped=12.0
+2023-03-03 11:34:47,245 INFO [train.py:968] (0/2) Epoch 6, batch 40550, giga_loss[loss=0.2966, simple_loss=0.3636, pruned_loss=0.1148, over 28713.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3466, pruned_loss=0.1069, over 5712094.69 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3936, pruned_loss=0.1424, over 5712998.03 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3424, pruned_loss=0.1037, over 5714200.45 frames. ], batch size: 92, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:34:48,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=268262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:35:01,342 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-03 11:35:29,777 INFO [train.py:968] (0/2) Epoch 6, batch 40600, giga_loss[loss=0.3034, simple_loss=0.3719, pruned_loss=0.1175, over 28682.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3518, pruned_loss=0.1098, over 5710447.66 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3938, pruned_loss=0.1426, over 5718414.06 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.347, pruned_loss=0.106, over 5707093.37 frames. ], batch size: 242, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:36:05,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.300e+02 1.092e+03 1.485e+03 2.253e+03 8.176e+03, threshold=2.969e+03, percent-clipped=14.0
+2023-03-03 11:36:06,986 INFO [train.py:968] (0/2) Epoch 6, batch 40650, giga_loss[loss=0.3298, simple_loss=0.3925, pruned_loss=0.1336, over 27977.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3553, pruned_loss=0.1113, over 5715785.45 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3939, pruned_loss=0.1427, over 5721129.78 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3501, pruned_loss=0.1072, over 5710451.55 frames. ], batch size: 412, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:36:43,470 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=268405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:36:47,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=268408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:36:49,132 INFO [train.py:968] (0/2) Epoch 6, batch 40700, giga_loss[loss=0.3022, simple_loss=0.3771, pruned_loss=0.1136, over 28692.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1136, over 5679722.60 frames. ], libri_tot_loss[loss=0.3401, simple_loss=0.3942, pruned_loss=0.143, over 5693350.62 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3548, pruned_loss=0.1096, over 5700574.36 frames. ], batch size: 284, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:36:59,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2364, 4.0635, 3.8316, 1.8642], device='cuda:0'), covar=tensor([0.0469, 0.0582, 0.0667, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.0903, 0.0843, 0.0779, 0.0602], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 11:37:11,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=268437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:37:30,829 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.002e+02 1.081e+03 1.292e+03 1.688e+03 6.528e+03, threshold=2.585e+03, percent-clipped=7.0
+2023-03-03 11:37:32,452 INFO [train.py:968] (0/2) Epoch 6, batch 40750, giga_loss[loss=0.2799, simple_loss=0.3535, pruned_loss=0.1031, over 28959.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3619, pruned_loss=0.1141, over 5690341.81 frames. ], libri_tot_loss[loss=0.3401, simple_loss=0.3942, pruned_loss=0.143, over 5693350.62 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.358, pruned_loss=0.111, over 5706570.99 frames. ], batch size: 136, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:38:16,116 INFO [train.py:968] (0/2) Epoch 6, batch 40800, giga_loss[loss=0.3129, simple_loss=0.3661, pruned_loss=0.1298, over 28951.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3657, pruned_loss=0.1172, over 5693236.20 frames. ], libri_tot_loss[loss=0.3402, simple_loss=0.3942, pruned_loss=0.1431, over 5694085.39 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.362, pruned_loss=0.1141, over 5705444.23 frames. ], batch size: 106, lr: 5.17e-03, grad_scale: 8.0
+2023-03-03 11:38:52,768 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-03 11:39:05,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.285e+02 1.446e+03 1.855e+03 2.837e+03 1.133e+04, threshold=3.710e+03, percent-clipped=27.0
+2023-03-03 11:39:05,851 INFO [train.py:968] (0/2) Epoch 6, batch 40850, giga_loss[loss=0.3563, simple_loss=0.4004, pruned_loss=0.1561, over 27631.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3732, pruned_loss=0.1241, over 5691637.89 frames. ], libri_tot_loss[loss=0.3401, simple_loss=0.3942, pruned_loss=0.143, over 5696499.50 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3694, pruned_loss=0.1209, over 5699377.43 frames. ], batch size: 472, lr: 5.17e-03, grad_scale: 2.0
+2023-03-03 11:39:44,136 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=268605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:39:48,994 INFO [train.py:968] (0/2) Epoch 6, batch 40900, giga_loss[loss=0.3644, simple_loss=0.4204, pruned_loss=0.1542, over 28281.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3804, pruned_loss=0.1299, over 5684250.26 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3941, pruned_loss=0.143, over 5690135.70 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.377, pruned_loss=0.127, over 5696165.07 frames. ], batch size: 368, lr: 5.17e-03, grad_scale: 2.0
+2023-03-03 11:40:29,789 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6132, 3.4330, 3.2353, 1.8291], device='cuda:0'), covar=tensor([0.0519, 0.0701, 0.0741, 0.1829], device='cuda:0'), in_proj_covar=tensor([0.0907, 0.0852, 0.0780, 0.0608], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 11:40:31,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.735e+02 1.603e+03 1.991e+03 2.715e+03 7.046e+03, threshold=3.982e+03, percent-clipped=9.0
+2023-03-03 11:40:31,999 INFO [train.py:968] (0/2) Epoch 6, batch 40950, giga_loss[loss=0.3926, simple_loss=0.4345, pruned_loss=0.1753, over 28749.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.386, pruned_loss=0.1345, over 5675943.27 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3936, pruned_loss=0.1426, over 5686364.94 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3834, pruned_loss=0.1321, over 5688521.65 frames. ], batch size: 284, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:40:37,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=268668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:41:07,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2751, 3.0936, 2.9465, 1.5143], device='cuda:0'), covar=tensor([0.0775, 0.0889, 0.0854, 0.2146], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0858, 0.0782, 0.0611], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 11:41:15,166 INFO [train.py:968] (0/2) Epoch 6, batch 41000, giga_loss[loss=0.4582, simple_loss=0.4798, pruned_loss=0.2183, over 27662.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3923, pruned_loss=0.1399, over 5682604.93 frames. ], libri_tot_loss[loss=0.3388, simple_loss=0.393, pruned_loss=0.1423, over 5687741.79 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3907, pruned_loss=0.1383, over 5691130.40 frames. ], batch size: 472, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:42:02,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.495e+03 1.820e+03 2.571e+03 1.244e+04, threshold=3.639e+03, percent-clipped=6.0
+2023-03-03 11:42:02,123 INFO [train.py:968] (0/2) Epoch 6, batch 41050, giga_loss[loss=0.3864, simple_loss=0.4335, pruned_loss=0.1696, over 28267.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.3983, pruned_loss=0.1449, over 5684020.59 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.3929, pruned_loss=0.1421, over 5690950.51 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3971, pruned_loss=0.1438, over 5688017.73 frames. ], batch size: 368, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:42:09,455 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 11:42:37,658 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 11:42:48,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3536, 3.2785, 1.3870, 1.4284], device='cuda:0'), covar=tensor([0.0962, 0.0368, 0.0871, 0.1386], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0487, 0.0313, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 11:42:54,206 INFO [train.py:968] (0/2) Epoch 6, batch 41100, giga_loss[loss=0.4044, simple_loss=0.4442, pruned_loss=0.1823, over 28903.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4035, pruned_loss=0.1508, over 5664764.41 frames. ], libri_tot_loss[loss=0.3389, simple_loss=0.3933, pruned_loss=0.1422, over 5696065.90 frames. ], giga_tot_loss[loss=0.3511, simple_loss=0.4024, pruned_loss=0.1499, over 5662794.67 frames. ], batch size: 227, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:43:38,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7155, 4.5562, 4.3598, 1.7239], device='cuda:0'), covar=tensor([0.0440, 0.0566, 0.0738, 0.2255], device='cuda:0'), in_proj_covar=tensor([0.0919, 0.0860, 0.0789, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 11:43:46,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.807e+02 1.679e+03 2.328e+03 3.212e+03 6.035e+03, threshold=4.656e+03, percent-clipped=16.0
+2023-03-03 11:43:46,014 INFO [train.py:968] (0/2) Epoch 6, batch 41150, giga_loss[loss=0.3154, simple_loss=0.3701, pruned_loss=0.1303, over 28817.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.4057, pruned_loss=0.1539, over 5653048.20 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3935, pruned_loss=0.1425, over 5682189.21 frames. ], giga_tot_loss[loss=0.3557, simple_loss=0.405, pruned_loss=0.1532, over 5662523.20 frames. ], batch size: 99, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:44:37,301 INFO [train.py:968] (0/2) Epoch 6, batch 41200, giga_loss[loss=0.4869, simple_loss=0.4877, pruned_loss=0.243, over 26634.00 frames. ], tot_loss[loss=0.3631, simple_loss=0.4092, pruned_loss=0.1585, over 5637693.02 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.3927, pruned_loss=0.1421, over 5687248.41 frames. ], giga_tot_loss[loss=0.3635, simple_loss=0.4097, pruned_loss=0.1587, over 5640103.31 frames. ], batch size: 555, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:44:57,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-03 11:44:59,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1444, 1.3387, 3.9616, 3.1098], device='cuda:0'), covar=tensor([0.1623, 0.2225, 0.0369, 0.0689], device='cuda:0'), in_proj_covar=tensor([0.0585, 0.0539, 0.0773, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 11:45:26,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.795e+03 2.511e+03 3.538e+03 8.454e+03, threshold=5.022e+03, percent-clipped=12.0
+2023-03-03 11:45:26,664 INFO [train.py:968] (0/2) Epoch 6, batch 41250, giga_loss[loss=0.3438, simple_loss=0.4105, pruned_loss=0.1386, over 29030.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4136, pruned_loss=0.1623, over 5622940.73 frames. ], libri_tot_loss[loss=0.3389, simple_loss=0.393, pruned_loss=0.1424, over 5669311.33 frames. ], giga_tot_loss[loss=0.3695, simple_loss=0.414, pruned_loss=0.1625, over 5639634.79 frames. ], batch size: 128, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:45:49,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=268980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:46:03,365 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=268994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:46:16,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4875, 1.6750, 1.6602, 1.6799], device='cuda:0'), covar=tensor([0.0967, 0.1280, 0.1090, 0.1070], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0734, 0.0650, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 11:46:18,151 INFO [train.py:968] (0/2) Epoch 6, batch 41300, giga_loss[loss=0.36, simple_loss=0.3994, pruned_loss=0.1603, over 28614.00 frames. ], tot_loss[loss=0.3697, simple_loss=0.4139, pruned_loss=0.1627, over 5616842.87 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.3926, pruned_loss=0.1419, over 5676574.08 frames. ], giga_tot_loss[loss=0.3716, simple_loss=0.4153, pruned_loss=0.1639, over 5621839.86 frames. ], batch size: 92, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:46:50,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:46:53,350 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 11:47:08,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.721e+03 2.223e+03 2.882e+03 1.173e+04, threshold=4.446e+03, percent-clipped=8.0
+2023-03-03 11:47:08,231 INFO [train.py:968] (0/2) Epoch 6, batch 41350, giga_loss[loss=0.4118, simple_loss=0.4402, pruned_loss=0.1917, over 28284.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4126, pruned_loss=0.1626, over 5616884.90 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3918, pruned_loss=0.1414, over 5672474.89 frames. ], giga_tot_loss[loss=0.3721, simple_loss=0.415, pruned_loss=0.1646, over 5622392.36 frames. ], batch size: 368, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:47:13,790 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-03 11:47:55,889 INFO [train.py:968] (0/2) Epoch 6, batch 41400, giga_loss[loss=0.283, simple_loss=0.362, pruned_loss=0.102, over 28626.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.4101, pruned_loss=0.1601, over 5640060.85 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.3921, pruned_loss=0.1416, over 5677274.25 frames. ], giga_tot_loss[loss=0.3681, simple_loss=0.4123, pruned_loss=0.162, over 5639010.61 frames. ], batch size: 60, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:48:06,631 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:48:10,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:48:11,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-03 11:48:36,929 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:48:43,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.958e+02 1.720e+03 2.078e+03 2.571e+03 6.080e+03, threshold=4.156e+03, percent-clipped=2.0
+2023-03-03 11:48:43,110 INFO [train.py:968] (0/2) Epoch 6, batch 41450, giga_loss[loss=0.4845, simple_loss=0.4794, pruned_loss=0.2448, over 26572.00 frames. ], tot_loss[loss=0.3618, simple_loss=0.4086, pruned_loss=0.1575, over 5638209.79 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3916, pruned_loss=0.1413, over 5670298.20 frames. ], giga_tot_loss[loss=0.3655, simple_loss=0.4114, pruned_loss=0.1598, over 5641969.57 frames. ], batch size: 555, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:49:11,041 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:49:14,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:49:32,922 INFO [train.py:968] (0/2) Epoch 6, batch 41500, giga_loss[loss=0.3678, simple_loss=0.4193, pruned_loss=0.1582, over 28347.00 frames. ], tot_loss[loss=0.3603, simple_loss=0.4087, pruned_loss=0.156, over 5653821.71 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3918, pruned_loss=0.1413, over 5672629.52 frames. ], giga_tot_loss[loss=0.3634, simple_loss=0.4109, pruned_loss=0.1579, over 5654279.02 frames. ], batch size: 368, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:49:39,676 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:50:26,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.497e+03 2.014e+03 2.860e+03 7.478e+03, threshold=4.027e+03, percent-clipped=9.0
+2023-03-03 11:50:26,839 INFO [train.py:968] (0/2) Epoch 6, batch 41550, giga_loss[loss=0.2997, simple_loss=0.3705, pruned_loss=0.1144, over 29019.00 frames. ], tot_loss[loss=0.3612, simple_loss=0.4095, pruned_loss=0.1564, over 5631084.98 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3919, pruned_loss=0.1414, over 5666906.04 frames. ], giga_tot_loss[loss=0.3638, simple_loss=0.4113, pruned_loss=0.1581, over 5636104.22 frames. ], batch size: 106, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:50:50,236 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:50:52,951 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 11:51:13,658 INFO [train.py:968] (0/2) Epoch 6, batch 41600, giga_loss[loss=0.3364, simple_loss=0.3937, pruned_loss=0.1395, over 28741.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.4064, pruned_loss=0.1535, over 5633217.51 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.3919, pruned_loss=0.1418, over 5665132.03 frames. ], giga_tot_loss[loss=0.3592, simple_loss=0.4084, pruned_loss=0.155, over 5637541.03 frames. ], batch size: 92, lr: 5.16e-03, grad_scale: 8.0
+2023-03-03 11:51:58,802 INFO [train.py:968] (0/2) Epoch 6, batch 41650, giga_loss[loss=0.3388, simple_loss=0.3912, pruned_loss=0.1432, over 28929.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4054, pruned_loss=0.1514, over 5646756.82 frames. ], libri_tot_loss[loss=0.3381, simple_loss=0.3921, pruned_loss=0.142, over 5667746.39 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.4073, pruned_loss=0.1527, over 5646876.17 frames. ], batch size: 213, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:51:59,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.205e+02 1.546e+03 1.910e+03 2.821e+03 6.524e+03, threshold=3.819e+03, percent-clipped=8.0
+2023-03-03 11:52:07,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:52:10,586 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 11:52:19,684 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:52:44,021 INFO [train.py:968] (0/2) Epoch 6, batch 41700, giga_loss[loss=0.2889, simple_loss=0.3588, pruned_loss=0.1095, over 28454.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4007, pruned_loss=0.1472, over 5655532.08 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3911, pruned_loss=0.1415, over 5668882.65 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4035, pruned_loss=0.1491, over 5653554.99 frames. ], batch size: 60, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:52:45,290 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-03 11:53:30,419 INFO [train.py:968] (0/2) Epoch 6, batch 41750, giga_loss[loss=0.3206, simple_loss=0.3855, pruned_loss=0.1278, over 28811.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3978, pruned_loss=0.1448, over 5645440.21 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.391, pruned_loss=0.1415, over 5661712.43 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.4004, pruned_loss=0.1464, over 5649615.03 frames. ], batch size: 174, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:53:32,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.607e+03 2.103e+03 3.134e+03 1.145e+04, threshold=4.205e+03, percent-clipped=13.0
+2023-03-03 11:54:17,838 INFO [train.py:968] (0/2) Epoch 6, batch 41800, giga_loss[loss=0.3446, simple_loss=0.3994, pruned_loss=0.1449, over 28926.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3962, pruned_loss=0.1443, over 5640698.67 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3914, pruned_loss=0.1418, over 5666644.26 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.3982, pruned_loss=0.1454, over 5639202.76 frames. ], batch size: 213, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:54:18,691 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:54:20,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:54:48,949 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:54:53,339 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 11:54:54,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7450, 1.6165, 1.2144, 1.3344], device='cuda:0'), covar=tensor([0.0593, 0.0582, 0.0894, 0.0965], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0452, 0.0505, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 11:55:01,069 INFO [train.py:968] (0/2) Epoch 6, batch 41850, giga_loss[loss=0.4149, simple_loss=0.4526, pruned_loss=0.1886, over 28987.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3961, pruned_loss=0.1435, over 5653105.85 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3911, pruned_loss=0.1416, over 5658258.30 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.398, pruned_loss=0.1446, over 5658063.50 frames. ], batch size: 136, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:55:03,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.059e+02 1.612e+03 2.149e+03 2.949e+03 1.368e+04, threshold=4.299e+03, percent-clipped=14.0
+2023-03-03 11:55:25,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3346, 1.4883, 1.1391, 1.1271], device='cuda:0'), covar=tensor([0.1339, 0.1090, 0.0898, 0.1118], device='cuda:0'), in_proj_covar=tensor([0.1519, 0.1355, 0.1333, 0.1421], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 11:55:50,597 INFO [train.py:968] (0/2) Epoch 6, batch 41900, giga_loss[loss=0.3081, simple_loss=0.3777, pruned_loss=0.1192, over 28983.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3948, pruned_loss=0.1421, over 5660724.32 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3913, pruned_loss=0.1418, over 5661907.69 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3962, pruned_loss=0.1428, over 5661178.54 frames. ], batch size: 106, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:56:10,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2925, 1.5041, 1.3407, 1.4387], device='cuda:0'), covar=tensor([0.1322, 0.1612, 0.1770, 0.1573], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0728, 0.0644, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 11:56:17,135 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:56:42,050 INFO [train.py:968] (0/2) Epoch 6, batch 41950, giga_loss[loss=0.3364, simple_loss=0.4024, pruned_loss=0.1352, over 29093.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3927, pruned_loss=0.139, over 5664793.58 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3908, pruned_loss=0.1418, over 5656689.78 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3943, pruned_loss=0.1396, over 5669299.86 frames. ], batch size: 128, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:56:42,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:56:43,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.711e+02 1.455e+03 1.818e+03 2.835e+03 9.299e+03, threshold=3.637e+03, percent-clipped=4.0
+2023-03-03 11:56:48,469 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2609, 1.2126, 0.9726, 1.4482], device='cuda:0'), covar=tensor([0.0773, 0.0370, 0.0369, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0118, 0.0121, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 11:57:39,328 INFO [train.py:968] (0/2) Epoch 6, batch 42000, giga_loss[loss=0.3764, simple_loss=0.4314, pruned_loss=0.1607, over 28617.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3938, pruned_loss=0.1373, over 5666646.37 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3908, pruned_loss=0.1417, over 5659207.52 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3952, pruned_loss=0.1378, over 5668017.24 frames. ], batch size: 92, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 11:57:39,333 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 11:57:47,704 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.2263, simple_loss=0.329, pruned_loss=0.06178, over 944034.00 frames. 
+2023-03-03 11:57:47,705 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 11:58:11,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3251, 1.4692, 1.1555, 1.0837], device='cuda:0'), covar=tensor([0.1129, 0.0965, 0.0847, 0.1054], device='cuda:0'), in_proj_covar=tensor([0.1523, 0.1362, 0.1328, 0.1425], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 11:58:18,789 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:58:33,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:58:34,274 INFO [train.py:968] (0/2) Epoch 6, batch 42050, giga_loss[loss=0.35, simple_loss=0.4037, pruned_loss=0.1481, over 28853.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3942, pruned_loss=0.1381, over 5665839.81 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3902, pruned_loss=0.1413, over 5663332.08 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3958, pruned_loss=0.1388, over 5663591.12 frames. ], batch size: 112, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 11:58:36,614 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.777e+02 1.751e+03 2.282e+03 2.950e+03 7.604e+03, threshold=4.563e+03, percent-clipped=9.0
+2023-03-03 11:58:49,207 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3826, 2.7457, 1.5897, 1.4482], device='cuda:0'), covar=tensor([0.0754, 0.0344, 0.0688, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0485, 0.0313, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 11:59:21,817 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:59:24,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:59:26,513 INFO [train.py:968] (0/2) Epoch 6, batch 42100, giga_loss[loss=0.3841, simple_loss=0.4276, pruned_loss=0.1703, over 28278.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3952, pruned_loss=0.1395, over 5670110.78 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3903, pruned_loss=0.1415, over 5665864.75 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3964, pruned_loss=0.1399, over 5666022.02 frames. ], batch size: 368, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 11:59:29,138 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4503, 3.5884, 1.5516, 1.4910], device='cuda:0'), covar=tensor([0.0893, 0.0276, 0.0810, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0487, 0.0313, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 11:59:49,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:59:51,980 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:00:14,077 INFO [train.py:968] (0/2) Epoch 6, batch 42150, giga_loss[loss=0.403, simple_loss=0.439, pruned_loss=0.1835, over 28997.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3951, pruned_loss=0.1402, over 5681673.40 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3905, pruned_loss=0.1414, over 5670413.90 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.396, pruned_loss=0.1405, over 5674507.73 frames. ], batch size: 155, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:00:16,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.725e+02 1.825e+03 2.260e+03 3.139e+03 5.904e+03, threshold=4.520e+03, percent-clipped=6.0
+2023-03-03 12:00:54,949 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-03 12:00:55,335 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:00:58,885 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:01:03,072 INFO [train.py:968] (0/2) Epoch 6, batch 42200, libri_loss[loss=0.3501, simple_loss=0.3999, pruned_loss=0.1501, over 29550.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3932, pruned_loss=0.1402, over 5670791.13 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3901, pruned_loss=0.1411, over 5667284.71 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3943, pruned_loss=0.1407, over 5667426.12 frames. ], batch size: 79, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:01:25,908 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:01:40,128 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0544, 1.1179, 3.8483, 3.1230], device='cuda:0'), covar=tensor([0.1675, 0.2433, 0.0416, 0.0748], device='cuda:0'), in_proj_covar=tensor([0.0588, 0.0539, 0.0776, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 12:01:52,879 INFO [train.py:968] (0/2) Epoch 6, batch 42250, libri_loss[loss=0.2909, simple_loss=0.358, pruned_loss=0.1119, over 29562.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3913, pruned_loss=0.1387, over 5652416.40 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3902, pruned_loss=0.1412, over 5650179.77 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3921, pruned_loss=0.139, over 5664189.47 frames. ], batch size: 76, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:01:54,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.891e+02 1.567e+03 2.046e+03 2.718e+03 5.413e+03, threshold=4.093e+03, percent-clipped=4.0
+2023-03-03 12:02:12,588 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-03 12:02:31,408 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-270000.pt
+2023-03-03 12:02:37,875 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9745, 1.3564, 1.0757, 0.1584], device='cuda:0'), covar=tensor([0.1693, 0.1561, 0.2378, 0.3132], device='cuda:0'), in_proj_covar=tensor([0.1449, 0.1361, 0.1406, 0.1199], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 12:02:40,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:02:41,522 INFO [train.py:968] (0/2) Epoch 6, batch 42300, giga_loss[loss=0.3645, simple_loss=0.4134, pruned_loss=0.1579, over 28995.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3906, pruned_loss=0.1366, over 5669875.72 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3898, pruned_loss=0.1409, over 5653931.21 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3916, pruned_loss=0.1371, over 5676107.20 frames. ], batch size: 128, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:03:33,946 INFO [train.py:968] (0/2) Epoch 6, batch 42350, giga_loss[loss=0.3562, simple_loss=0.4127, pruned_loss=0.1498, over 29014.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3911, pruned_loss=0.1363, over 5671562.72 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3902, pruned_loss=0.141, over 5655379.01 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3916, pruned_loss=0.1364, over 5675307.56 frames. ], batch size: 106, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:03:35,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2573, 2.9576, 1.4336, 1.3557], device='cuda:0'), covar=tensor([0.0870, 0.0310, 0.0783, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0489, 0.0315, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 12:03:35,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.794e+02 1.454e+03 2.156e+03 2.719e+03 7.084e+03, threshold=4.312e+03, percent-clipped=8.0
+2023-03-03 12:04:23,786 INFO [train.py:968] (0/2) Epoch 6, batch 42400, giga_loss[loss=0.306, simple_loss=0.3754, pruned_loss=0.1183, over 28861.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3901, pruned_loss=0.1359, over 5677341.13 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.39, pruned_loss=0.1409, over 5659192.86 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3906, pruned_loss=0.136, over 5677371.15 frames. ], batch size: 112, lr: 5.15e-03, grad_scale: 8.0
+2023-03-03 12:04:33,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:01,240 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:03,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:10,189 INFO [train.py:968] (0/2) Epoch 6, batch 42450, giga_loss[loss=0.3349, simple_loss=0.3954, pruned_loss=0.1372, over 28293.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.39, pruned_loss=0.1365, over 5668095.43 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3908, pruned_loss=0.1412, over 5653339.54 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3896, pruned_loss=0.1361, over 5674383.43 frames. ], batch size: 368, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:05:12,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.791e+03 2.306e+03 3.164e+03 8.026e+03, threshold=4.612e+03, percent-clipped=8.0
+2023-03-03 12:05:14,600 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:32,425 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:57,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:58,500 INFO [train.py:968] (0/2) Epoch 6, batch 42500, giga_loss[loss=0.2857, simple_loss=0.3497, pruned_loss=0.1108, over 28906.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3887, pruned_loss=0.1363, over 5670382.49 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3905, pruned_loss=0.141, over 5657854.74 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3887, pruned_loss=0.1361, over 5671568.24 frames. ], batch size: 112, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:06:17,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:06:49,613 INFO [train.py:968] (0/2) Epoch 6, batch 42550, libri_loss[loss=0.3425, simple_loss=0.3984, pruned_loss=0.1433, over 29489.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3893, pruned_loss=0.1386, over 5662485.37 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3904, pruned_loss=0.1411, over 5655515.33 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3893, pruned_loss=0.1382, over 5665241.40 frames. ], batch size: 85, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:06:52,089 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:06:52,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.909e+02 1.801e+03 2.217e+03 3.169e+03 1.058e+04, threshold=4.433e+03, percent-clipped=10.0
+2023-03-03 12:06:52,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4428, 2.2426, 1.5951, 0.5944], device='cuda:0'), covar=tensor([0.2529, 0.1422, 0.2147, 0.2787], device='cuda:0'), in_proj_covar=tensor([0.1454, 0.1370, 0.1407, 0.1203], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 12:06:54,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:07:22,450 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:07:38,142 INFO [train.py:968] (0/2) Epoch 6, batch 42600, giga_loss[loss=0.2801, simple_loss=0.3495, pruned_loss=0.1054, over 28962.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3878, pruned_loss=0.1381, over 5671824.08 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3902, pruned_loss=0.1409, over 5660009.06 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3881, pruned_loss=0.138, over 5670223.21 frames. ], batch size: 164, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:08:19,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:08:19,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2976, 1.7274, 1.4303, 1.4917], device='cuda:0'), covar=tensor([0.0742, 0.0305, 0.0303, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0118, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 12:08:21,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:08:27,422 INFO [train.py:968] (0/2) Epoch 6, batch 42650, giga_loss[loss=0.3809, simple_loss=0.4223, pruned_loss=0.1697, over 28332.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3872, pruned_loss=0.1376, over 5681018.24 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3901, pruned_loss=0.1407, over 5665165.03 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3873, pruned_loss=0.1376, over 5675493.75 frames. ], batch size: 369, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:08:33,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.169e+03 1.705e+03 2.293e+03 3.505e+03 1.401e+04, threshold=4.587e+03, percent-clipped=13.0
+2023-03-03 12:08:36,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5903, 4.6183, 1.7622, 1.7611], device='cuda:0'), covar=tensor([0.0865, 0.0325, 0.0814, 0.1249], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0491, 0.0315, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 12:08:53,335 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:09:22,908 INFO [train.py:968] (0/2) Epoch 6, batch 42700, giga_loss[loss=0.3254, simple_loss=0.384, pruned_loss=0.1334, over 29045.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3876, pruned_loss=0.1381, over 5684655.34 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3902, pruned_loss=0.1408, over 5666331.21 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3876, pruned_loss=0.138, over 5679461.34 frames. ], batch size: 136, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:10:07,701 INFO [train.py:968] (0/2) Epoch 6, batch 42750, giga_loss[loss=0.3306, simple_loss=0.3943, pruned_loss=0.1335, over 28832.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3884, pruned_loss=0.1379, over 5675937.17 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3909, pruned_loss=0.1413, over 5662232.01 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3878, pruned_loss=0.1373, over 5675892.09 frames. ], batch size: 243, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:10:11,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.576e+02 1.518e+03 2.192e+03 2.805e+03 5.586e+03, threshold=4.384e+03, percent-clipped=5.0
+2023-03-03 12:10:51,296 INFO [train.py:968] (0/2) Epoch 6, batch 42800, libri_loss[loss=0.2562, simple_loss=0.3253, pruned_loss=0.09354, over 29658.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3889, pruned_loss=0.1372, over 5674749.74 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3906, pruned_loss=0.1412, over 5654731.80 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3886, pruned_loss=0.1367, over 5681868.86 frames. ], batch size: 69, lr: 5.15e-03, grad_scale: 8.0
+2023-03-03 12:11:16,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:11:36,792 INFO [train.py:968] (0/2) Epoch 6, batch 42850, giga_loss[loss=0.3461, simple_loss=0.3944, pruned_loss=0.1489, over 28396.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3893, pruned_loss=0.1371, over 5672314.73 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3906, pruned_loss=0.1412, over 5659773.58 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3891, pruned_loss=0.1366, over 5674075.03 frames. ], batch size: 85, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:11:39,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.553e+02 1.512e+03 2.183e+03 2.944e+03 1.017e+04, threshold=4.365e+03, percent-clipped=13.0
+2023-03-03 12:11:55,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6108, 1.8111, 1.4009, 1.2142], device='cuda:0'), covar=tensor([0.1435, 0.1187, 0.1072, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.1526, 0.1381, 0.1329, 0.1431], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 12:12:22,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:12:28,726 INFO [train.py:968] (0/2) Epoch 6, batch 42900, giga_loss[loss=0.3154, simple_loss=0.3855, pruned_loss=0.1227, over 28956.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3892, pruned_loss=0.1375, over 5666544.94 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3899, pruned_loss=0.1406, over 5663740.70 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3896, pruned_loss=0.1376, over 5664697.61 frames. ], batch size: 164, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:12:47,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8207, 4.6628, 1.9464, 1.7553], device='cuda:0'), covar=tensor([0.0847, 0.0266, 0.0759, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0489, 0.0315, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 12:13:09,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3191, 1.3313, 1.0443, 1.4737], device='cuda:0'), covar=tensor([0.0692, 0.0313, 0.0330, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0118, 0.0120, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 12:13:15,137 INFO [train.py:968] (0/2) Epoch 6, batch 42950, giga_loss[loss=0.3457, simple_loss=0.3991, pruned_loss=0.1461, over 28938.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.393, pruned_loss=0.1413, over 5667420.86 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3904, pruned_loss=0.1406, over 5668607.05 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3929, pruned_loss=0.1413, over 5661406.71 frames. ], batch size: 213, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:13:20,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.564e+02 1.421e+03 1.871e+03 2.600e+03 5.872e+03, threshold=3.742e+03, percent-clipped=7.0
+2023-03-03 12:13:30,206 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270673.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:13:39,805 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:13:42,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:13:48,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5090, 1.8189, 1.3125, 1.6653], device='cuda:0'), covar=tensor([0.0744, 0.0279, 0.0329, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0118, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 12:14:07,398 INFO [train.py:968] (0/2) Epoch 6, batch 43000, giga_loss[loss=0.3437, simple_loss=0.3961, pruned_loss=0.1457, over 28694.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3936, pruned_loss=0.1434, over 5662728.79 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3899, pruned_loss=0.1402, over 5671715.08 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.394, pruned_loss=0.1438, over 5655009.69 frames. ], batch size: 92, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:14:15,609 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:14:45,859 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:14:49,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:14:59,461 INFO [train.py:968] (0/2) Epoch 6, batch 43050, giga_loss[loss=0.3258, simple_loss=0.3782, pruned_loss=0.1367, over 28843.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.3947, pruned_loss=0.1454, over 5663346.85 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3899, pruned_loss=0.1402, over 5676947.79 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3952, pruned_loss=0.1458, over 5652159.06 frames. ], batch size: 112, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:15:02,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.745e+03 2.159e+03 3.155e+03 6.257e+03, threshold=4.319e+03, percent-clipped=15.0
+2023-03-03 12:15:17,804 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:15:28,418 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:15:32,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5750, 1.7622, 1.2755, 1.3063], device='cuda:0'), covar=tensor([0.1303, 0.1112, 0.1080, 0.1202], device='cuda:0'), in_proj_covar=tensor([0.1516, 0.1377, 0.1326, 0.1428], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 12:15:48,371 INFO [train.py:968] (0/2) Epoch 6, batch 43100, giga_loss[loss=0.2879, simple_loss=0.3618, pruned_loss=0.107, over 28824.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3971, pruned_loss=0.1473, over 5671619.19 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3905, pruned_loss=0.1407, over 5681740.14 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.397, pruned_loss=0.1473, over 5658261.20 frames. ], batch size: 112, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:16:07,710 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:16:27,601 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-03 12:16:28,129 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 12:16:34,043 INFO [train.py:968] (0/2) Epoch 6, batch 43150, giga_loss[loss=0.3269, simple_loss=0.3864, pruned_loss=0.1337, over 28305.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3945, pruned_loss=0.1452, over 5660357.51 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.391, pruned_loss=0.1412, over 5664261.10 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3941, pruned_loss=0.1448, over 5665012.98 frames. ], batch size: 368, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:16:39,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.879e+02 1.580e+03 2.162e+03 2.864e+03 6.605e+03, threshold=4.324e+03, percent-clipped=6.0
+2023-03-03 12:17:22,420 INFO [train.py:968] (0/2) Epoch 6, batch 43200, giga_loss[loss=0.3032, simple_loss=0.3802, pruned_loss=0.1131, over 28997.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3922, pruned_loss=0.1412, over 5670957.97 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3909, pruned_loss=0.1411, over 5666815.72 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.392, pruned_loss=0.1411, over 5672171.91 frames. ], batch size: 136, lr: 5.14e-03, grad_scale: 8.0
+2023-03-03 12:18:12,841 INFO [train.py:968] (0/2) Epoch 6, batch 43250, giga_loss[loss=0.2909, simple_loss=0.3646, pruned_loss=0.1086, over 28920.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3886, pruned_loss=0.1387, over 5671002.70 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3908, pruned_loss=0.1411, over 5672622.83 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3885, pruned_loss=0.1384, over 5666982.99 frames. ], batch size: 145, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:18:17,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.088e+02 1.644e+03 2.461e+03 3.904e+03 2.289e+04, threshold=4.921e+03, percent-clipped=16.0
+2023-03-03 12:18:52,428 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:18:57,293 INFO [train.py:968] (0/2) Epoch 6, batch 43300, giga_loss[loss=0.2782, simple_loss=0.3464, pruned_loss=0.105, over 28715.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3875, pruned_loss=0.1387, over 5661777.17 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3912, pruned_loss=0.1414, over 5669424.32 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.387, pruned_loss=0.1381, over 5661646.63 frames. ], batch size: 92, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:19:05,005 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3839, 1.5464, 1.2652, 1.7682], device='cuda:0'), covar=tensor([0.1994, 0.1938, 0.2074, 0.1725], device='cuda:0'), in_proj_covar=tensor([0.1162, 0.0890, 0.1016, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 12:19:30,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271048.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 12:19:41,483 INFO [train.py:968] (0/2) Epoch 6, batch 43350, giga_loss[loss=0.292, simple_loss=0.3555, pruned_loss=0.1143, over 28946.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3863, pruned_loss=0.1385, over 5670977.48 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3915, pruned_loss=0.1418, over 5676140.69 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3854, pruned_loss=0.1377, over 5664514.47 frames. ], batch size: 106, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:19:45,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2234, 1.2507, 1.1044, 1.0174], device='cuda:0'), covar=tensor([0.0685, 0.0508, 0.0973, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0452, 0.0502, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 12:19:47,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.381e+02 1.704e+03 2.234e+03 3.312e+03 6.484e+03, threshold=4.468e+03, percent-clipped=7.0
+2023-03-03 12:20:29,467 INFO [train.py:968] (0/2) Epoch 6, batch 43400, giga_loss[loss=0.4335, simple_loss=0.46, pruned_loss=0.2035, over 29021.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3874, pruned_loss=0.1393, over 5664138.17 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.392, pruned_loss=0.142, over 5669120.68 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3862, pruned_loss=0.1383, over 5664004.88 frames. ], batch size: 155, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:20:37,601 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:21:17,234 INFO [train.py:968] (0/2) Epoch 6, batch 43450, giga_loss[loss=0.3082, simple_loss=0.3917, pruned_loss=0.1123, over 29036.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3921, pruned_loss=0.1419, over 5669201.01 frames. ], libri_tot_loss[loss=0.3383, simple_loss=0.3922, pruned_loss=0.1422, over 5674516.07 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3908, pruned_loss=0.141, over 5663975.87 frames. ], batch size: 155, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:21:20,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:21:20,639 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-03 12:21:22,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.490e+02 1.512e+03 1.935e+03 2.470e+03 6.181e+03, threshold=3.870e+03, percent-clipped=4.0
+2023-03-03 12:21:45,065 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271191.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 12:21:48,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271194.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 12:22:02,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:22:06,257 INFO [train.py:968] (0/2) Epoch 6, batch 43500, giga_loss[loss=0.2848, simple_loss=0.3662, pruned_loss=0.1017, over 28802.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3942, pruned_loss=0.1402, over 5650651.00 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3927, pruned_loss=0.1426, over 5657099.03 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3928, pruned_loss=0.1391, over 5660680.12 frames. ], batch size: 119, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:22:20,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271223.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:22:39,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7639, 4.5922, 4.2986, 2.0345], device='cuda:0'), covar=tensor([0.0442, 0.0655, 0.0793, 0.2008], device='cuda:0'), in_proj_covar=tensor([0.0910, 0.0871, 0.0785, 0.0613], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 12:22:53,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9878, 1.2214, 3.4603, 2.9641], device='cuda:0'), covar=tensor([0.1499, 0.2040, 0.0459, 0.0745], device='cuda:0'), in_proj_covar=tensor([0.0587, 0.0541, 0.0779, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 12:23:00,436 INFO [train.py:968] (0/2) Epoch 6, batch 43550, giga_loss[loss=0.3736, simple_loss=0.4302, pruned_loss=0.1585, over 28329.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.396, pruned_loss=0.1411, over 5654998.23 frames. ], libri_tot_loss[loss=0.3388, simple_loss=0.3926, pruned_loss=0.1426, over 5659658.67 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3951, pruned_loss=0.1403, over 5660384.11 frames. ], batch size: 369, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:23:04,159 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.304e+02 1.619e+03 2.128e+03 2.829e+03 6.177e+03, threshold=4.256e+03, percent-clipped=13.0
+2023-03-03 12:23:43,381 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:23:45,339 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:23:45,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:23:45,916 INFO [train.py:968] (0/2) Epoch 6, batch 43600, giga_loss[loss=0.2764, simple_loss=0.3582, pruned_loss=0.09734, over 28612.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3981, pruned_loss=0.1424, over 5662383.35 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3927, pruned_loss=0.1427, over 5656161.03 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3973, pruned_loss=0.1416, over 5668903.61 frames. ], batch size: 60, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:24:15,867 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:24:26,092 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:24:30,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:24:36,669 INFO [train.py:968] (0/2) Epoch 6, batch 43650, giga_loss[loss=0.3581, simple_loss=0.4056, pruned_loss=0.1553, over 28294.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3994, pruned_loss=0.1445, over 5642493.93 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3929, pruned_loss=0.1429, over 5637403.93 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.3987, pruned_loss=0.1437, over 5664405.19 frames. ], batch size: 368, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:24:41,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.829e+02 1.599e+03 2.168e+03 2.696e+03 5.797e+03, threshold=4.336e+03, percent-clipped=6.0
+2023-03-03 12:24:54,201 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:24:54,939 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:25:04,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1856, 1.1263, 0.9789, 1.2790], device='cuda:0'), covar=tensor([0.0713, 0.0285, 0.0319, 0.0802], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0118, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 12:25:18,555 INFO [train.py:968] (0/2) Epoch 6, batch 43700, giga_loss[loss=0.2949, simple_loss=0.3671, pruned_loss=0.1114, over 28883.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3972, pruned_loss=0.1437, over 5647266.42 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3925, pruned_loss=0.1427, over 5634250.31 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3971, pruned_loss=0.1433, over 5668796.15 frames. ], batch size: 174, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:25:54,838 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-03 12:26:10,622 INFO [train.py:968] (0/2) Epoch 6, batch 43750, giga_loss[loss=0.3525, simple_loss=0.3994, pruned_loss=0.1528, over 28930.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3958, pruned_loss=0.1438, over 5644755.18 frames. ], libri_tot_loss[loss=0.3388, simple_loss=0.3925, pruned_loss=0.1426, over 5636778.28 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3959, pruned_loss=0.1435, over 5659272.71 frames. ], batch size: 213, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:26:18,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.093e+02 1.580e+03 2.453e+03 3.194e+03 8.167e+03, threshold=4.906e+03, percent-clipped=11.0
+2023-03-03 12:26:43,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:26:58,415 INFO [train.py:968] (0/2) Epoch 6, batch 43800, giga_loss[loss=0.3176, simple_loss=0.3744, pruned_loss=0.1304, over 28987.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3947, pruned_loss=0.1437, over 5654302.89 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.393, pruned_loss=0.143, over 5643375.32 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3942, pruned_loss=0.1431, over 5659886.98 frames. ], batch size: 128, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:27:07,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4400, 5.2412, 4.9867, 2.1567], device='cuda:0'), covar=tensor([0.0335, 0.0496, 0.0587, 0.1913], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0884, 0.0797, 0.0620], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 12:27:11,596 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:27:14,478 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:27:45,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:27:50,092 INFO [train.py:968] (0/2) Epoch 6, batch 43850, giga_loss[loss=0.3193, simple_loss=0.3819, pruned_loss=0.1283, over 28855.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3948, pruned_loss=0.1446, over 5646296.54 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3928, pruned_loss=0.1426, over 5650262.23 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3946, pruned_loss=0.1445, over 5644804.67 frames. ], batch size: 285, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:27:55,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.068e+03 1.594e+03 2.236e+03 3.398e+03 1.580e+04, threshold=4.473e+03, percent-clipped=13.0
+2023-03-03 12:28:38,142 INFO [train.py:968] (0/2) Epoch 6, batch 43900, giga_loss[loss=0.342, simple_loss=0.3737, pruned_loss=0.1552, over 23464.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3938, pruned_loss=0.144, over 5639506.46 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3928, pruned_loss=0.1427, over 5649114.99 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3938, pruned_loss=0.144, over 5639609.11 frames. ], batch size: 705, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:29:01,230 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:29:04,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:29:16,110 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:29:23,067 INFO [train.py:968] (0/2) Epoch 6, batch 43950, giga_loss[loss=0.3254, simple_loss=0.3811, pruned_loss=0.1349, over 28952.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3918, pruned_loss=0.1427, over 5658116.75 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.393, pruned_loss=0.1427, over 5652781.96 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3916, pruned_loss=0.1426, over 5654943.86 frames. ], batch size: 164, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:29:28,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.800e+02 1.641e+03 2.160e+03 3.376e+03 8.488e+03, threshold=4.320e+03, percent-clipped=6.0
+2023-03-03 12:29:29,679 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:29:46,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:30:09,199 INFO [train.py:968] (0/2) Epoch 6, batch 44000, giga_loss[loss=0.3308, simple_loss=0.3794, pruned_loss=0.1411, over 28546.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3918, pruned_loss=0.1431, over 5665870.21 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3932, pruned_loss=0.1428, over 5655028.92 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3914, pruned_loss=0.1429, over 5661176.00 frames. ], batch size: 85, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:30:53,096 INFO [train.py:968] (0/2) Epoch 6, batch 44050, giga_loss[loss=0.3413, simple_loss=0.4023, pruned_loss=0.1402, over 27866.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3909, pruned_loss=0.1415, over 5662944.94 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3936, pruned_loss=0.1429, over 5657807.45 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3902, pruned_loss=0.1412, over 5656700.18 frames. ], batch size: 412, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:31:01,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.200e+02 1.459e+03 1.894e+03 2.774e+03 5.610e+03, threshold=3.787e+03, percent-clipped=6.0
+2023-03-03 12:31:01,820 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:31:47,225 INFO [train.py:968] (0/2) Epoch 6, batch 44100, giga_loss[loss=0.3766, simple_loss=0.4277, pruned_loss=0.1628, over 28845.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3934, pruned_loss=0.1427, over 5648149.93 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.3935, pruned_loss=0.1428, over 5650578.46 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3928, pruned_loss=0.1425, over 5649479.38 frames. ], batch size: 186, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:31:55,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:32:03,031 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:32:05,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:32:19,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1352, 1.5938, 1.5179, 1.2447], device='cuda:0'), covar=tensor([0.0860, 0.0282, 0.0276, 0.1058], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0118, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 12:32:33,465 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:32:33,929 INFO [train.py:968] (0/2) Epoch 6, batch 44150, giga_loss[loss=0.3402, simple_loss=0.389, pruned_loss=0.1457, over 28655.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3935, pruned_loss=0.1427, over 5653922.71 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3939, pruned_loss=0.143, over 5655217.44 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3927, pruned_loss=0.1424, over 5650871.19 frames. ], batch size: 85, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:32:44,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.896e+02 1.543e+03 1.924e+03 2.673e+03 7.978e+03, threshold=3.849e+03, percent-clipped=11.0
+2023-03-03 12:33:24,616 INFO [train.py:968] (0/2) Epoch 6, batch 44200, giga_loss[loss=0.3004, simple_loss=0.3847, pruned_loss=0.108, over 29119.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3934, pruned_loss=0.1415, over 5655281.74 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3936, pruned_loss=0.1429, over 5647261.56 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.393, pruned_loss=0.1414, over 5659460.82 frames. ], batch size: 128, lr: 5.13e-03, grad_scale: 2.0
+2023-03-03 12:34:08,703 INFO [train.py:968] (0/2) Epoch 6, batch 44250, giga_loss[loss=0.3239, simple_loss=0.4026, pruned_loss=0.1227, over 28476.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3934, pruned_loss=0.1389, over 5662979.67 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.393, pruned_loss=0.1427, over 5654295.18 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3935, pruned_loss=0.1389, over 5660500.07 frames. ], batch size: 71, lr: 5.13e-03, grad_scale: 2.0
+2023-03-03 12:34:16,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.499e+02 1.427e+03 1.839e+03 2.414e+03 4.668e+03, threshold=3.678e+03, percent-clipped=4.0
+2023-03-03 12:34:42,263 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-272000.pt
+2023-03-03 12:34:54,018 INFO [train.py:968] (0/2) Epoch 6, batch 44300, giga_loss[loss=0.3542, simple_loss=0.4127, pruned_loss=0.1479, over 28764.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3958, pruned_loss=0.1391, over 5673556.93 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.393, pruned_loss=0.1427, over 5661132.88 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.396, pruned_loss=0.1389, over 5665629.28 frames. ], batch size: 99, lr: 5.13e-03, grad_scale: 2.0
+2023-03-03 12:35:15,764 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:35:42,752 INFO [train.py:968] (0/2) Epoch 6, batch 44350, giga_loss[loss=0.4012, simple_loss=0.4422, pruned_loss=0.1801, over 28877.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3985, pruned_loss=0.1422, over 5652393.11 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.393, pruned_loss=0.1428, over 5655748.72 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3989, pruned_loss=0.1419, over 5651227.74 frames. ], batch size: 186, lr: 5.13e-03, grad_scale: 2.0
+2023-03-03 12:35:43,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2626, 1.8942, 1.5521, 1.4182], device='cuda:0'), covar=tensor([0.0736, 0.0290, 0.0289, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0118, 0.0121, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 12:35:52,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.440e+02 1.413e+03 2.046e+03 2.924e+03 1.022e+04, threshold=4.093e+03, percent-clipped=14.0
+2023-03-03 12:36:34,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4709, 1.7537, 1.3336, 1.3144], device='cuda:0'), covar=tensor([0.1191, 0.1045, 0.1033, 0.1033], device='cuda:0'), in_proj_covar=tensor([0.1534, 0.1361, 0.1336, 0.1436], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 12:36:36,465 INFO [train.py:968] (0/2) Epoch 6, batch 44400, giga_loss[loss=0.4455, simple_loss=0.4649, pruned_loss=0.213, over 26573.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.4013, pruned_loss=0.1456, over 5646458.26 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3931, pruned_loss=0.1429, over 5655018.98 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.4015, pruned_loss=0.1454, over 5646173.73 frames. ], batch size: 555, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:37:06,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:37:23,551 INFO [train.py:968] (0/2) Epoch 6, batch 44450, giga_loss[loss=0.3312, simple_loss=0.3968, pruned_loss=0.1329, over 28984.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3996, pruned_loss=0.1447, over 5657865.23 frames. ], libri_tot_loss[loss=0.3388, simple_loss=0.3925, pruned_loss=0.1425, over 5648712.02 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.4004, pruned_loss=0.1448, over 5663380.16 frames. ], batch size: 164, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:37:32,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.642e+02 1.596e+03 2.296e+03 3.316e+03 6.585e+03, threshold=4.592e+03, percent-clipped=19.0
+2023-03-03 12:37:34,935 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:37:38,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:37:54,908 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:38:03,441 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:38:07,662 INFO [train.py:968] (0/2) Epoch 6, batch 44500, giga_loss[loss=0.3265, simple_loss=0.3934, pruned_loss=0.1298, over 28628.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3985, pruned_loss=0.1441, over 5655803.67 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.3919, pruned_loss=0.142, over 5648317.30 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3999, pruned_loss=0.1448, over 5661010.07 frames. ], batch size: 307, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:38:37,832 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:38:49,972 INFO [train.py:968] (0/2) Epoch 6, batch 44550, giga_loss[loss=0.3276, simple_loss=0.3988, pruned_loss=0.1282, over 28938.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3972, pruned_loss=0.1422, over 5657315.39 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3916, pruned_loss=0.1418, over 5643205.79 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3988, pruned_loss=0.143, over 5665637.66 frames. ], batch size: 186, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:38:56,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.180e+02 1.379e+03 1.899e+03 2.827e+03 7.950e+03, threshold=3.798e+03, percent-clipped=4.0
+2023-03-03 12:39:12,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:39:14,823 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:39:35,120 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-03 12:39:35,961 INFO [train.py:968] (0/2) Epoch 6, batch 44600, giga_loss[loss=0.2886, simple_loss=0.3653, pruned_loss=0.1059, over 28468.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3968, pruned_loss=0.1401, over 5670329.49 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3914, pruned_loss=0.1417, over 5648464.95 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3984, pruned_loss=0.1407, over 5672925.10 frames. ], batch size: 85, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:39:43,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:39:52,246 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:40:00,303 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:40:03,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:40:20,934 INFO [train.py:968] (0/2) Epoch 6, batch 44650, giga_loss[loss=0.3449, simple_loss=0.4146, pruned_loss=0.1376, over 29044.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.397, pruned_loss=0.14, over 5676614.21 frames. ], libri_tot_loss[loss=0.3378, simple_loss=0.3916, pruned_loss=0.142, over 5653848.69 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3983, pruned_loss=0.1402, over 5674529.63 frames. ], batch size: 155, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:40:27,266 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-03 12:40:32,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.322e+02 1.543e+03 2.003e+03 2.927e+03 1.110e+04, threshold=4.006e+03, percent-clipped=13.0
+2023-03-03 12:40:33,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:41:08,392 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-03 12:41:14,326 INFO [train.py:968] (0/2) Epoch 6, batch 44700, giga_loss[loss=0.3325, simple_loss=0.3954, pruned_loss=0.1348, over 28710.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3979, pruned_loss=0.1416, over 5652893.52 frames. ], libri_tot_loss[loss=0.3378, simple_loss=0.3916, pruned_loss=0.142, over 5646506.06 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.399, pruned_loss=0.1417, over 5657211.45 frames. ], batch size: 92, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:41:21,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0883, 1.1175, 1.2722, 1.0408], device='cuda:0'), covar=tensor([0.1037, 0.1188, 0.1620, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0735, 0.0650, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 12:42:00,314 INFO [train.py:968] (0/2) Epoch 6, batch 44750, giga_loss[loss=0.2896, simple_loss=0.3576, pruned_loss=0.1108, over 28904.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3952, pruned_loss=0.1395, over 5650620.07 frames. ], libri_tot_loss[loss=0.3381, simple_loss=0.3919, pruned_loss=0.1422, over 5640727.02 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3959, pruned_loss=0.1395, over 5659058.09 frames. ], batch size: 145, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:42:06,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.500e+03 2.090e+03 3.021e+03 8.564e+03, threshold=4.179e+03, percent-clipped=10.0
+2023-03-03 12:42:46,737 INFO [train.py:968] (0/2) Epoch 6, batch 44800, giga_loss[loss=0.3012, simple_loss=0.3709, pruned_loss=0.1158, over 29088.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3931, pruned_loss=0.1394, over 5664870.73 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3912, pruned_loss=0.1416, over 5648405.61 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3944, pruned_loss=0.1398, over 5665094.50 frames. ], batch size: 128, lr: 5.13e-03, grad_scale: 8.0
+2023-03-03 12:42:55,893 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:43:15,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4694, 1.8157, 1.7831, 1.4723], device='cuda:0'), covar=tensor([0.1543, 0.1978, 0.1160, 0.1363], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0737, 0.0810, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 12:43:29,606 INFO [train.py:968] (0/2) Epoch 6, batch 44850, giga_loss[loss=0.3149, simple_loss=0.3774, pruned_loss=0.1262, over 28520.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3915, pruned_loss=0.1391, over 5661933.25 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3906, pruned_loss=0.1412, over 5656425.63 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3931, pruned_loss=0.1397, over 5655261.52 frames. ], batch size: 78, lr: 5.13e-03, grad_scale: 8.0
+2023-03-03 12:43:39,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.150e+02 1.704e+03 2.133e+03 2.962e+03 5.577e+03, threshold=4.266e+03, percent-clipped=6.0
+2023-03-03 12:43:44,497 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272574.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:44:19,440 INFO [train.py:968] (0/2) Epoch 6, batch 44900, giga_loss[loss=0.3419, simple_loss=0.3942, pruned_loss=0.1447, over 28933.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3903, pruned_loss=0.1389, over 5664572.93 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3908, pruned_loss=0.1413, over 5662397.26 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3915, pruned_loss=0.1393, over 5654157.35 frames. ], batch size: 213, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:44:25,104 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:44:31,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:44:40,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9264, 3.7038, 3.4970, 1.6818], device='cuda:0'), covar=tensor([0.0628, 0.0853, 0.0964, 0.2063], device='cuda:0'), in_proj_covar=tensor([0.0945, 0.0896, 0.0812, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 12:44:47,589 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:45:07,507 INFO [train.py:968] (0/2) Epoch 6, batch 44950, giga_loss[loss=0.3191, simple_loss=0.3786, pruned_loss=0.1298, over 28999.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3909, pruned_loss=0.1407, over 5662825.84 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.391, pruned_loss=0.1414, over 5663877.75 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3916, pruned_loss=0.1409, over 5653245.30 frames. ], batch size: 128, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:45:16,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.361e+02 1.486e+03 1.975e+03 2.962e+03 9.995e+03, threshold=3.951e+03, percent-clipped=9.0
+2023-03-03 12:45:45,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:45:53,259 INFO [train.py:968] (0/2) Epoch 6, batch 45000, libri_loss[loss=0.3076, simple_loss=0.3603, pruned_loss=0.1275, over 29340.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.39, pruned_loss=0.14, over 5657476.98 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3908, pruned_loss=0.141, over 5671684.88 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3909, pruned_loss=0.1405, over 5642324.91 frames. ], batch size: 71, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:45:53,266 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 12:46:02,092 INFO [train.py:1012] (0/2) Epoch 6, validation: loss=0.229, simple_loss=0.3343, pruned_loss=0.0619, over 944034.00 frames. 
+2023-03-03 12:46:02,093 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 12:46:44,894 INFO [train.py:968] (0/2) Epoch 6, batch 45050, libri_loss[loss=0.2881, simple_loss=0.3438, pruned_loss=0.1162, over 28645.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3866, pruned_loss=0.1355, over 5658273.13 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3905, pruned_loss=0.1409, over 5665515.50 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3874, pruned_loss=0.136, over 5651191.45 frames. ], batch size: 63, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:46:49,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:46:50,013 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:46:52,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:46:53,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.085e+02 1.327e+03 1.684e+03 2.401e+03 9.275e+03, threshold=3.367e+03, percent-clipped=4.0
+2023-03-03 12:47:19,030 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:47:30,898 INFO [train.py:968] (0/2) Epoch 6, batch 45100, giga_loss[loss=0.2949, simple_loss=0.374, pruned_loss=0.1079, over 28842.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.383, pruned_loss=0.1322, over 5655704.76 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3903, pruned_loss=0.1406, over 5671238.83 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3837, pruned_loss=0.1326, over 5644715.07 frames. ], batch size: 186, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:48:08,355 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:48:13,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:48:23,813 INFO [train.py:968] (0/2) Epoch 6, batch 45150, giga_loss[loss=0.3951, simple_loss=0.4311, pruned_loss=0.1796, over 28714.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3838, pruned_loss=0.1333, over 5661422.41 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3903, pruned_loss=0.1406, over 5672424.64 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3843, pruned_loss=0.1336, over 5651825.51 frames. ], batch size: 262, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:48:26,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2132, 0.9313, 0.8293, 1.2852], device='cuda:0'), covar=tensor([0.0737, 0.0346, 0.0346, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0118, 0.0121, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 12:48:32,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.658e+02 1.408e+03 1.844e+03 2.667e+03 9.755e+03, threshold=3.688e+03, percent-clipped=14.0
+2023-03-03 12:48:38,335 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:48:56,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:49:17,365 INFO [train.py:968] (0/2) Epoch 6, batch 45200, libri_loss[loss=0.3775, simple_loss=0.4265, pruned_loss=0.1642, over 27724.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3819, pruned_loss=0.1332, over 5674440.87 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3903, pruned_loss=0.1406, over 5675325.59 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3822, pruned_loss=0.1333, over 5664315.66 frames. ], batch size: 116, lr: 5.12e-03, grad_scale: 8.0
+2023-03-03 12:49:47,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272949.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:49:56,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3118, 1.4666, 1.2004, 1.5821], device='cuda:0'), covar=tensor([0.2187, 0.2067, 0.2077, 0.1931], device='cuda:0'), in_proj_covar=tensor([0.1169, 0.0893, 0.1023, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 12:49:56,940 INFO [train.py:968] (0/2) Epoch 6, batch 45250, giga_loss[loss=0.3604, simple_loss=0.4122, pruned_loss=0.1543, over 28336.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3814, pruned_loss=0.1331, over 5677069.65 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3901, pruned_loss=0.1404, over 5675990.99 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3813, pruned_loss=0.133, over 5667930.95 frames. ], batch size: 368, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:50:04,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.475e+02 1.831e+03 2.394e+03 3.449e+03 6.314e+03, threshold=4.788e+03, percent-clipped=18.0
+2023-03-03 12:50:10,562 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-03 12:50:22,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:50:33,230 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=273006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:50:37,751 INFO [train.py:968] (0/2) Epoch 6, batch 45300, giga_loss[loss=0.3229, simple_loss=0.3781, pruned_loss=0.1338, over 28770.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.382, pruned_loss=0.1325, over 5686861.79 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.39, pruned_loss=0.1401, over 5677396.61 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3816, pruned_loss=0.1323, over 5678398.39 frames. ], batch size: 99, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:50:41,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=273015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:51:01,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:51:04,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:51:23,186 INFO [train.py:968] (0/2) Epoch 6, batch 45350, giga_loss[loss=0.3203, simple_loss=0.3851, pruned_loss=0.1277, over 28867.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.383, pruned_loss=0.1327, over 5676356.71 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.39, pruned_loss=0.1402, over 5681927.19 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3825, pruned_loss=0.1322, over 5665689.11 frames. ], batch size: 145, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:51:32,532 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:51:35,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.579e+03 2.048e+03 2.873e+03 5.438e+03, threshold=4.095e+03, percent-clipped=1.0
+2023-03-03 12:51:56,389 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273092.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 12:51:58,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273095.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:52:14,563 INFO [train.py:968] (0/2) Epoch 6, batch 45400, giga_loss[loss=0.3238, simple_loss=0.3815, pruned_loss=0.133, over 27938.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3831, pruned_loss=0.1328, over 5676746.80 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3903, pruned_loss=0.1404, over 5683267.77 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3825, pruned_loss=0.1322, over 5667200.15 frames. ], batch size: 412, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:52:26,639 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273124.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:52:35,978 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:52:38,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:52:41,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=273141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:52:58,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:52:59,660 INFO [train.py:968] (0/2) Epoch 6, batch 45450, libri_loss[loss=0.3281, simple_loss=0.3931, pruned_loss=0.1315, over 29291.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3855, pruned_loss=0.1351, over 5661696.96 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3913, pruned_loss=0.1411, over 5671292.18 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3837, pruned_loss=0.1337, over 5664028.42 frames. ], batch size: 94, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:52:59,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:53:03,084 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:53:07,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.411e+02 1.410e+03 1.904e+03 2.652e+03 7.225e+03, threshold=3.809e+03, percent-clipped=9.0
+2023-03-03 12:53:25,998 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:53:32,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0112, 1.0575, 3.8917, 3.1015], device='cuda:0'), covar=tensor([0.1729, 0.2502, 0.0405, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0594, 0.0548, 0.0790, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 12:53:46,105 INFO [train.py:968] (0/2) Epoch 6, batch 45500, giga_loss[loss=0.3051, simple_loss=0.3757, pruned_loss=0.1172, over 28320.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3866, pruned_loss=0.1363, over 5655255.13 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3912, pruned_loss=0.1409, over 5675450.55 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3853, pruned_loss=0.1353, over 5653509.85 frames. ], batch size: 71, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:54:14,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4221, 2.0446, 1.4364, 0.6666], device='cuda:0'), covar=tensor([0.2549, 0.1495, 0.2458, 0.2928], device='cuda:0'), in_proj_covar=tensor([0.1462, 0.1359, 0.1413, 0.1198], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 12:54:34,152 INFO [train.py:968] (0/2) Epoch 6, batch 45550, giga_loss[loss=0.3412, simple_loss=0.3918, pruned_loss=0.1453, over 28907.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3887, pruned_loss=0.1376, over 5639373.84 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3911, pruned_loss=0.1407, over 5671286.07 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3876, pruned_loss=0.1369, over 5640691.77 frames. ], batch size: 186, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:54:44,016 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.631e+02 1.642e+03 2.174e+03 2.683e+03 5.604e+03, threshold=4.348e+03, percent-clipped=9.0
+2023-03-03 12:54:54,854 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:54:57,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:55:21,148 INFO [train.py:968] (0/2) Epoch 6, batch 45600, giga_loss[loss=0.3908, simple_loss=0.4359, pruned_loss=0.1728, over 28742.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3899, pruned_loss=0.1383, over 5650629.57 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.391, pruned_loss=0.1406, over 5676116.19 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.389, pruned_loss=0.1378, over 5646998.55 frames. ], batch size: 262, lr: 5.12e-03, grad_scale: 8.0
+2023-03-03 12:55:24,959 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:56:13,169 INFO [train.py:968] (0/2) Epoch 6, batch 45650, giga_loss[loss=0.393, simple_loss=0.4308, pruned_loss=0.1776, over 27883.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3923, pruned_loss=0.1406, over 5644372.77 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3914, pruned_loss=0.1408, over 5667531.97 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3912, pruned_loss=0.1399, over 5649070.36 frames. ], batch size: 412, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:56:23,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.681e+02 1.752e+03 2.224e+03 2.842e+03 1.119e+04, threshold=4.447e+03, percent-clipped=12.0
+2023-03-03 12:56:32,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=273381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:57:02,048 INFO [train.py:968] (0/2) Epoch 6, batch 45700, giga_loss[loss=0.3375, simple_loss=0.3943, pruned_loss=0.1404, over 29076.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3935, pruned_loss=0.1408, over 5647165.55 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3909, pruned_loss=0.1404, over 5674277.20 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3932, pruned_loss=0.1407, over 5644337.06 frames. ], batch size: 113, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:57:43,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9809, 1.3715, 1.2268, 1.2975], device='cuda:0'), covar=tensor([0.1235, 0.1019, 0.1721, 0.1129], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0735, 0.0644, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 12:57:53,569 INFO [train.py:968] (0/2) Epoch 6, batch 45750, giga_loss[loss=0.3102, simple_loss=0.3743, pruned_loss=0.123, over 28590.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3938, pruned_loss=0.1402, over 5648765.61 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3907, pruned_loss=0.1403, over 5666577.24 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3938, pruned_loss=0.1401, over 5652147.84 frames. ], batch size: 85, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:58:03,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.854e+02 1.569e+03 1.945e+03 2.850e+03 8.251e+03, threshold=3.889e+03, percent-clipped=4.0
+2023-03-03 12:58:41,554 INFO [train.py:968] (0/2) Epoch 6, batch 45800, giga_loss[loss=0.3334, simple_loss=0.3884, pruned_loss=0.1392, over 28961.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3934, pruned_loss=0.1405, over 5635219.05 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3908, pruned_loss=0.1404, over 5651319.29 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3934, pruned_loss=0.1404, over 5651850.45 frames. ], batch size: 145, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:58:41,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6372, 1.8740, 1.3888, 1.3598], device='cuda:0'), covar=tensor([0.1430, 0.1040, 0.0911, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.1514, 0.1354, 0.1333, 0.1431], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 12:58:52,250 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:58:55,576 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:59:01,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9201, 3.0912, 2.0749, 0.8410], device='cuda:0'), covar=tensor([0.4047, 0.1483, 0.2179, 0.4053], device='cuda:0'), in_proj_covar=tensor([0.1454, 0.1356, 0.1409, 0.1195], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 12:59:22,884 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:59:27,242 INFO [train.py:968] (0/2) Epoch 6, batch 45850, giga_loss[loss=0.321, simple_loss=0.3829, pruned_loss=0.1295, over 28564.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3927, pruned_loss=0.1407, over 5598306.64 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3916, pruned_loss=0.1412, over 5598919.55 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.392, pruned_loss=0.1398, over 5660611.76 frames. ], batch size: 307, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:59:40,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.446e+02 1.885e+03 2.525e+03 3.874e+03 1.147e+04, threshold=5.051e+03, percent-clipped=24.0
+2023-03-03 13:00:24,079 INFO [train.py:968] (0/2) Epoch 6, batch 45900, libri_loss[loss=0.3943, simple_loss=0.4235, pruned_loss=0.1825, over 18711.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.392, pruned_loss=0.1411, over 5576592.32 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.392, pruned_loss=0.1417, over 5564630.50 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3911, pruned_loss=0.14, over 5657847.15 frames. ], batch size: 187, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 13:01:03,813 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-03 13:01:05,815 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-6.pt
+2023-03-03 13:01:50,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4398, 1.6011, 1.2999, 1.7847], device='cuda:0'), covar=tensor([0.2237, 0.2110, 0.2139, 0.2134], device='cuda:0'), in_proj_covar=tensor([0.1170, 0.0892, 0.1024, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:01:59,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.005e+02 1.343e+03 1.722e+03 2.452e+03 9.669e+03, threshold=3.444e+03, percent-clipped=4.0
+2023-03-03 13:02:19,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2885, 3.0247, 1.3994, 1.3787], device='cuda:0'), covar=tensor([0.0920, 0.0275, 0.0869, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0488, 0.0315, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 13:02:22,714 INFO [train.py:968] (0/2) Epoch 7, batch 50, giga_loss[loss=0.37, simple_loss=0.4186, pruned_loss=0.1607, over 26795.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3895, pruned_loss=0.1232, over 1254979.87 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3497, pruned_loss=0.1035, over 136439.58 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3939, pruned_loss=0.1255, over 1147001.26 frames. ], batch size: 555, lr: 4.80e-03, grad_scale: 2.0
+2023-03-03 13:03:12,021 INFO [train.py:968] (0/2) Epoch 7, batch 100, giga_loss[loss=0.3184, simple_loss=0.3803, pruned_loss=0.1283, over 28710.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.38, pruned_loss=0.1172, over 2242309.10 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3513, pruned_loss=0.1005, over 251186.31 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.383, pruned_loss=0.119, over 2082669.07 frames. ], batch size: 92, lr: 4.80e-03, grad_scale: 2.0
+2023-03-03 13:03:32,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.542e+02 9.997e+02 1.244e+03 1.615e+03 4.647e+03, threshold=2.488e+03, percent-clipped=4.0
+2023-03-03 13:03:56,743 INFO [train.py:968] (0/2) Epoch 7, batch 150, giga_loss[loss=0.3202, simple_loss=0.3666, pruned_loss=0.1369, over 26635.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3655, pruned_loss=0.1111, over 2999972.07 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3499, pruned_loss=0.1009, over 407151.64 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3678, pruned_loss=0.1125, over 2793708.84 frames. ], batch size: 555, lr: 4.80e-03, grad_scale: 2.0
+2023-03-03 13:03:58,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=273802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:04:21,168 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=273828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:04:38,347 INFO [train.py:968] (0/2) Epoch 7, batch 200, giga_loss[loss=0.236, simple_loss=0.3115, pruned_loss=0.08029, over 28713.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3515, pruned_loss=0.1042, over 3608033.03 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.349, pruned_loss=0.1002, over 542471.31 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3526, pruned_loss=0.1051, over 3386055.42 frames. ], batch size: 284, lr: 4.80e-03, grad_scale: 2.0
+2023-03-03 13:04:56,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.904e+02 1.003e+03 1.226e+03 1.566e+03 4.370e+03, threshold=2.451e+03, percent-clipped=7.0
+2023-03-03 13:05:20,594 INFO [train.py:968] (0/2) Epoch 7, batch 250, giga_loss[loss=0.2108, simple_loss=0.2826, pruned_loss=0.06944, over 28707.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3392, pruned_loss=0.09757, over 4073781.85 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3496, pruned_loss=0.09993, over 671939.12 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3391, pruned_loss=0.0979, over 3853583.65 frames. ], batch size: 92, lr: 4.79e-03, grad_scale: 2.0
+2023-03-03 13:05:21,178 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-03 13:05:40,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8886, 1.1056, 3.6338, 2.8836], device='cuda:0'), covar=tensor([0.2185, 0.2868, 0.0661, 0.1098], device='cuda:0'), in_proj_covar=tensor([0.0594, 0.0547, 0.0787, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:06:02,750 INFO [train.py:968] (0/2) Epoch 7, batch 300, giga_loss[loss=0.2475, simple_loss=0.3147, pruned_loss=0.09009, over 28541.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3305, pruned_loss=0.09386, over 4436840.66 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3515, pruned_loss=0.1005, over 774856.33 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3293, pruned_loss=0.09372, over 4233580.60 frames. ], batch size: 336, lr: 4.79e-03, grad_scale: 2.0
+2023-03-03 13:06:07,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4300, 2.8226, 1.5656, 1.4775], device='cuda:0'), covar=tensor([0.0807, 0.0320, 0.0778, 0.1196], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0488, 0.0316, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 13:06:20,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.496e+02 8.399e+02 1.107e+03 1.611e+03 5.079e+03, threshold=2.213e+03, percent-clipped=11.0
+2023-03-03 13:06:45,896 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-274000.pt
+2023-03-03 13:06:46,188 INFO [train.py:968] (0/2) Epoch 7, batch 350, giga_loss[loss=0.2242, simple_loss=0.2932, pruned_loss=0.07758, over 28510.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3245, pruned_loss=0.09105, over 4719368.76 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3513, pruned_loss=0.09968, over 972459.62 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3222, pruned_loss=0.09059, over 4510549.28 frames. ], batch size: 85, lr: 4.79e-03, grad_scale: 2.0
+2023-03-03 13:07:23,928 INFO [train.py:968] (0/2) Epoch 7, batch 400, giga_loss[loss=0.2216, simple_loss=0.2928, pruned_loss=0.07522, over 28968.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3198, pruned_loss=0.08868, over 4937085.72 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.351, pruned_loss=0.0994, over 1108371.41 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.317, pruned_loss=0.08799, over 4749876.49 frames. ], batch size: 106, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:07:38,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3899, 4.1784, 3.9263, 1.8284], device='cuda:0'), covar=tensor([0.0474, 0.0620, 0.0767, 0.2152], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0871, 0.0797, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 13:07:46,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.392e+02 9.238e+02 1.225e+03 1.702e+03 5.475e+03, threshold=2.451e+03, percent-clipped=11.0
+2023-03-03 13:07:55,976 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-03 13:08:05,953 INFO [train.py:968] (0/2) Epoch 7, batch 450, giga_loss[loss=0.2382, simple_loss=0.3048, pruned_loss=0.08581, over 28805.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3178, pruned_loss=0.08746, over 5113679.13 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3517, pruned_loss=0.09933, over 1294062.83 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.314, pruned_loss=0.08641, over 4932897.31 frames. ], batch size: 186, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:08:27,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8201, 1.8108, 1.3800, 1.3746], device='cuda:0'), covar=tensor([0.0694, 0.0584, 0.0947, 0.0996], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0442, 0.0495, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 13:08:47,768 INFO [train.py:968] (0/2) Epoch 7, batch 500, giga_loss[loss=0.2318, simple_loss=0.299, pruned_loss=0.08231, over 27586.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3156, pruned_loss=0.08594, over 5236323.05 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3537, pruned_loss=0.1007, over 1455061.42 frames. ], giga_tot_loss[loss=0.2395, simple_loss=0.3106, pruned_loss=0.0842, over 5083084.73 frames. ], batch size: 472, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:09:09,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.051e+02 9.823e+02 1.249e+03 1.828e+03 3.809e+03, threshold=2.497e+03, percent-clipped=8.0
+2023-03-03 13:09:12,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:09:33,536 INFO [train.py:968] (0/2) Epoch 7, batch 550, giga_loss[loss=0.1985, simple_loss=0.2693, pruned_loss=0.0638, over 28480.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3133, pruned_loss=0.08515, over 5339230.56 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3541, pruned_loss=0.1012, over 1521645.37 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.3087, pruned_loss=0.0834, over 5209722.80 frames. ], batch size: 71, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:09:35,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-03 13:09:36,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:10:09,499 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 13:10:18,644 INFO [train.py:968] (0/2) Epoch 7, batch 600, giga_loss[loss=0.2015, simple_loss=0.2721, pruned_loss=0.0655, over 28541.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3112, pruned_loss=0.08466, over 5418742.85 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3541, pruned_loss=0.1015, over 1587176.45 frames. ], giga_tot_loss[loss=0.2365, simple_loss=0.307, pruned_loss=0.08297, over 5309325.34 frames. ], batch size: 85, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:10:42,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.940e+02 9.319e+02 1.355e+03 1.693e+03 1.010e+04, threshold=2.710e+03, percent-clipped=12.0
+2023-03-03 13:11:09,271 INFO [train.py:968] (0/2) Epoch 7, batch 650, giga_loss[loss=0.279, simple_loss=0.3297, pruned_loss=0.1142, over 26611.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3087, pruned_loss=0.08333, over 5469922.10 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3551, pruned_loss=0.1021, over 1668203.80 frames. ], giga_tot_loss[loss=0.2336, simple_loss=0.3042, pruned_loss=0.08147, over 5378845.74 frames. ], batch size: 555, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:11:26,827 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:11:31,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:11:50,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:11:55,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:11:55,902 INFO [train.py:968] (0/2) Epoch 7, batch 700, giga_loss[loss=0.1964, simple_loss=0.2685, pruned_loss=0.06218, over 28059.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3058, pruned_loss=0.08199, over 5520173.24 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3546, pruned_loss=0.1018, over 1771327.18 frames. ], giga_tot_loss[loss=0.2307, simple_loss=0.3012, pruned_loss=0.08013, over 5439956.85 frames. ], batch size: 77, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:11:57,849 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:11:59,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0521, 1.2921, 3.5748, 2.9406], device='cuda:0'), covar=tensor([0.1629, 0.2300, 0.0422, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0587, 0.0540, 0.0776, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:12:14,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.340e+02 9.168e+02 1.283e+03 1.837e+03 8.779e+03, threshold=2.566e+03, percent-clipped=10.0
+2023-03-03 13:12:21,185 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:12:36,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:12:40,700 INFO [train.py:968] (0/2) Epoch 7, batch 750, giga_loss[loss=0.2289, simple_loss=0.2908, pruned_loss=0.08345, over 28471.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3031, pruned_loss=0.08088, over 5568031.78 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.355, pruned_loss=0.1023, over 1812469.78 frames. ], giga_tot_loss[loss=0.2286, simple_loss=0.299, pruned_loss=0.07912, over 5501945.41 frames. ], batch size: 65, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:13:26,504 INFO [train.py:968] (0/2) Epoch 7, batch 800, libri_loss[loss=0.2448, simple_loss=0.3176, pruned_loss=0.08604, over 29644.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3014, pruned_loss=0.08019, over 5598955.74 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3557, pruned_loss=0.1027, over 1929455.30 frames. ], giga_tot_loss[loss=0.2263, simple_loss=0.2965, pruned_loss=0.07802, over 5541129.65 frames. ], batch size: 69, lr: 4.79e-03, grad_scale: 8.0
+2023-03-03 13:13:49,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.014e+02 1.069e+03 1.361e+03 1.772e+03 3.070e+03, threshold=2.721e+03, percent-clipped=5.0
+2023-03-03 13:14:15,003 INFO [train.py:968] (0/2) Epoch 7, batch 850, giga_loss[loss=0.2646, simple_loss=0.3442, pruned_loss=0.09249, over 28706.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3108, pruned_loss=0.08617, over 5602926.94 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3569, pruned_loss=0.1037, over 1998130.23 frames. ], giga_tot_loss[loss=0.2366, simple_loss=0.3056, pruned_loss=0.08382, over 5561245.95 frames. ], batch size: 242, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:14:38,563 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:15:00,843 INFO [train.py:968] (0/2) Epoch 7, batch 900, giga_loss[loss=0.2964, simple_loss=0.3749, pruned_loss=0.1089, over 28678.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.325, pruned_loss=0.0936, over 5626041.32 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3568, pruned_loss=0.1035, over 2151150.44 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3195, pruned_loss=0.09136, over 5582152.49 frames. ], batch size: 262, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:15:23,208 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.182e+02 1.289e+03 1.869e+03 2.559e+03 6.591e+03, threshold=3.737e+03, percent-clipped=23.0
+2023-03-03 13:15:44,066 INFO [train.py:968] (0/2) Epoch 7, batch 950, giga_loss[loss=0.2886, simple_loss=0.3593, pruned_loss=0.1089, over 28607.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3373, pruned_loss=0.09974, over 5649655.17 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3576, pruned_loss=0.1039, over 2225318.44 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3323, pruned_loss=0.09774, over 5610320.80 frames. ], batch size: 60, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:15:58,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2717, 1.4382, 1.2905, 1.0026], device='cuda:0'), covar=tensor([0.1138, 0.1100, 0.0668, 0.0977], device='cuda:0'), in_proj_covar=tensor([0.1525, 0.1351, 0.1338, 0.1425], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 13:16:25,439 INFO [train.py:968] (0/2) Epoch 7, batch 1000, giga_loss[loss=0.3252, simple_loss=0.3949, pruned_loss=0.1277, over 28717.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3455, pruned_loss=0.1035, over 5666816.89 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3565, pruned_loss=0.1036, over 2334320.99 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3415, pruned_loss=0.102, over 5628620.79 frames. ], batch size: 262, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:16:38,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:16:46,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.677e+02 1.104e+03 1.331e+03 1.796e+03 4.924e+03, threshold=2.663e+03, percent-clipped=2.0
+2023-03-03 13:16:55,639 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-03 13:17:04,485 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:17:07,103 INFO [train.py:968] (0/2) Epoch 7, batch 1050, giga_loss[loss=0.2843, simple_loss=0.3562, pruned_loss=0.1062, over 28406.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.349, pruned_loss=0.1037, over 5681032.13 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3555, pruned_loss=0.1028, over 2422598.26 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3461, pruned_loss=0.1028, over 5645248.19 frames. ], batch size: 71, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:17:56,785 INFO [train.py:968] (0/2) Epoch 7, batch 1100, giga_loss[loss=0.2892, simple_loss=0.365, pruned_loss=0.1067, over 29039.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3521, pruned_loss=0.1046, over 5672753.50 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3555, pruned_loss=0.1028, over 2422598.26 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3499, pruned_loss=0.1039, over 5644902.41 frames. ], batch size: 164, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:18:12,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:18:16,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.161e+02 1.056e+03 1.302e+03 1.699e+03 4.938e+03, threshold=2.605e+03, percent-clipped=8.0
+2023-03-03 13:18:39,143 INFO [train.py:968] (0/2) Epoch 7, batch 1150, giga_loss[loss=0.2867, simple_loss=0.3565, pruned_loss=0.1085, over 28877.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3544, pruned_loss=0.1062, over 5678766.78 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3559, pruned_loss=0.1033, over 2478852.34 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3524, pruned_loss=0.1056, over 5665328.40 frames. ], batch size: 186, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:19:24,749 INFO [train.py:968] (0/2) Epoch 7, batch 1200, giga_loss[loss=0.2621, simple_loss=0.3419, pruned_loss=0.09111, over 28907.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3566, pruned_loss=0.1082, over 5670750.58 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3558, pruned_loss=0.1031, over 2546128.38 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3551, pruned_loss=0.1078, over 5656419.83 frames. ], batch size: 186, lr: 4.79e-03, grad_scale: 8.0
+2023-03-03 13:19:46,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.577e+02 1.078e+03 1.331e+03 1.753e+03 3.242e+03, threshold=2.661e+03, percent-clipped=4.0
+2023-03-03 13:20:06,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5536, 2.3841, 1.5018, 1.1091], device='cuda:0'), covar=tensor([0.1940, 0.1028, 0.1373, 0.1695], device='cuda:0'), in_proj_covar=tensor([0.1542, 0.1363, 0.1350, 0.1441], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 13:20:09,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:20:11,090 INFO [train.py:968] (0/2) Epoch 7, batch 1250, giga_loss[loss=0.3799, simple_loss=0.4174, pruned_loss=0.1712, over 26433.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3601, pruned_loss=0.1105, over 5673325.01 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.356, pruned_loss=0.1031, over 2579418.99 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3588, pruned_loss=0.1103, over 5660377.43 frames. ], batch size: 555, lr: 4.79e-03, grad_scale: 8.0
+2023-03-03 13:20:20,005 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:20:22,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:20:22,266 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-03 13:20:39,853 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:20:49,677 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:20:53,569 INFO [train.py:968] (0/2) Epoch 7, batch 1300, giga_loss[loss=0.3071, simple_loss=0.3821, pruned_loss=0.1161, over 28987.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3627, pruned_loss=0.111, over 5676407.88 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3566, pruned_loss=0.1035, over 2715070.07 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3617, pruned_loss=0.1109, over 5665801.71 frames. ], batch size: 164, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:20:54,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-03 13:21:12,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.530e+02 1.097e+03 1.499e+03 2.044e+03 5.225e+03, threshold=2.997e+03, percent-clipped=13.0
+2023-03-03 13:21:33,426 INFO [train.py:968] (0/2) Epoch 7, batch 1350, giga_loss[loss=0.2913, simple_loss=0.3715, pruned_loss=0.1055, over 29055.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3639, pruned_loss=0.1102, over 5692820.36 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3564, pruned_loss=0.1032, over 2774969.34 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3632, pruned_loss=0.1104, over 5683647.40 frames. ], batch size: 136, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:22:08,830 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:22:12,169 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:22:14,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:22:17,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3763, 1.5089, 1.2447, 1.8091], device='cuda:0'), covar=tensor([0.2069, 0.2043, 0.2018, 0.2032], device='cuda:0'), in_proj_covar=tensor([0.1167, 0.0898, 0.1026, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:22:18,444 INFO [train.py:968] (0/2) Epoch 7, batch 1400, giga_loss[loss=0.2898, simple_loss=0.3673, pruned_loss=0.1062, over 28726.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3651, pruned_loss=0.1105, over 5687762.73 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3564, pruned_loss=0.1032, over 2833608.90 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3648, pruned_loss=0.1108, over 5679605.86 frames. ], batch size: 284, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:22:32,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6095, 2.4973, 1.5250, 0.6081], device='cuda:0'), covar=tensor([0.4388, 0.2130, 0.2547, 0.4223], device='cuda:0'), in_proj_covar=tensor([0.1471, 0.1371, 0.1423, 0.1195], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 13:22:35,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:22:38,741 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:22:39,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.761e+02 1.084e+03 1.336e+03 1.932e+03 6.698e+03, threshold=2.672e+03, percent-clipped=4.0
+2023-03-03 13:23:02,845 INFO [train.py:968] (0/2) Epoch 7, batch 1450, giga_loss[loss=0.2838, simple_loss=0.3655, pruned_loss=0.101, over 29037.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3643, pruned_loss=0.1087, over 5694667.71 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3566, pruned_loss=0.1033, over 2878972.06 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3641, pruned_loss=0.109, over 5685844.38 frames. ], batch size: 136, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:23:18,512 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-03 13:23:40,782 INFO [train.py:968] (0/2) Epoch 7, batch 1500, giga_loss[loss=0.2502, simple_loss=0.3343, pruned_loss=0.08302, over 28822.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3624, pruned_loss=0.1063, over 5702223.63 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3569, pruned_loss=0.1033, over 2980196.29 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3623, pruned_loss=0.1067, over 5691442.22 frames. ], batch size: 66, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:24:01,155 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.288e+02 9.979e+02 1.293e+03 1.896e+03 1.147e+04, threshold=2.587e+03, percent-clipped=15.0
+2023-03-03 13:24:05,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:24:07,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:24:19,727 INFO [train.py:968] (0/2) Epoch 7, batch 1550, giga_loss[loss=0.2807, simple_loss=0.3634, pruned_loss=0.09895, over 29019.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3607, pruned_loss=0.105, over 5705444.93 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3561, pruned_loss=0.1025, over 3092810.49 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3612, pruned_loss=0.1058, over 5691472.87 frames. ], batch size: 164, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:24:32,577 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:24:33,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4243, 4.1462, 4.0708, 1.8234], device='cuda:0'), covar=tensor([0.0457, 0.0646, 0.0771, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.0908, 0.0851, 0.0780, 0.0609], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 13:24:34,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:24:36,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:24:42,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7084, 1.6376, 1.2310, 1.3143], device='cuda:0'), covar=tensor([0.0633, 0.0557, 0.0880, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0445, 0.0497, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 13:25:02,480 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:25:07,070 INFO [train.py:968] (0/2) Epoch 7, batch 1600, giga_loss[loss=0.2933, simple_loss=0.3648, pruned_loss=0.111, over 29057.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3626, pruned_loss=0.1077, over 5714882.16 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3565, pruned_loss=0.1028, over 3107108.44 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3628, pruned_loss=0.1082, over 5702920.86 frames. ], batch size: 155, lr: 4.78e-03, grad_scale: 8.0
+2023-03-03 13:25:15,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2206, 1.8313, 1.6371, 1.2759], device='cuda:0'), covar=tensor([0.1484, 0.1986, 0.1271, 0.1407], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0722, 0.0814, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 13:25:27,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.236e+02 1.254e+03 1.540e+03 1.940e+03 5.174e+03, threshold=3.081e+03, percent-clipped=10.0
+2023-03-03 13:25:40,617 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:25:51,329 INFO [train.py:968] (0/2) Epoch 7, batch 1650, libri_loss[loss=0.3361, simple_loss=0.3944, pruned_loss=0.1389, over 29537.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3663, pruned_loss=0.1131, over 5711406.72 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3579, pruned_loss=0.1035, over 3174704.94 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.366, pruned_loss=0.1133, over 5698884.26 frames. ], batch size: 84, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:25:58,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:26:34,330 INFO [train.py:968] (0/2) Epoch 7, batch 1700, libri_loss[loss=0.284, simple_loss=0.3662, pruned_loss=0.1009, over 29272.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3684, pruned_loss=0.116, over 5704037.12 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3593, pruned_loss=0.1046, over 3300346.03 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3678, pruned_loss=0.1161, over 5692072.28 frames. ], batch size: 94, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:26:57,059 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.524e+02 1.263e+03 1.535e+03 1.871e+03 5.693e+03, threshold=3.070e+03, percent-clipped=6.0
+2023-03-03 13:27:20,340 INFO [train.py:968] (0/2) Epoch 7, batch 1750, giga_loss[loss=0.2659, simple_loss=0.3386, pruned_loss=0.09661, over 28738.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3669, pruned_loss=0.1161, over 5702851.72 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3594, pruned_loss=0.1046, over 3337248.30 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3665, pruned_loss=0.1164, over 5692420.23 frames. ], batch size: 119, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:27:33,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4387, 4.2447, 1.6672, 1.4690], device='cuda:0'), covar=tensor([0.1131, 0.0276, 0.0942, 0.1632], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0482, 0.0315, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0026, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 13:27:39,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1175, 0.8767, 0.9330, 1.4423], device='cuda:0'), covar=tensor([0.0780, 0.0347, 0.0343, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0117, 0.0121, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 13:28:04,466 INFO [train.py:968] (0/2) Epoch 7, batch 1800, giga_loss[loss=0.3022, simple_loss=0.3807, pruned_loss=0.1118, over 29053.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3649, pruned_loss=0.1152, over 5707315.54 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3595, pruned_loss=0.1044, over 3387354.93 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3647, pruned_loss=0.1157, over 5696288.69 frames. ], batch size: 164, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:28:05,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:28:08,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:28:22,686 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-03 13:28:26,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.649e+02 1.121e+03 1.363e+03 2.059e+03 6.046e+03, threshold=2.726e+03, percent-clipped=6.0
+2023-03-03 13:28:29,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3206, 2.0312, 1.5001, 0.4285], device='cuda:0'), covar=tensor([0.2895, 0.1336, 0.2308, 0.3454], device='cuda:0'), in_proj_covar=tensor([0.1447, 0.1353, 0.1399, 0.1183], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 13:28:33,489 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:28:48,211 INFO [train.py:968] (0/2) Epoch 7, batch 1850, giga_loss[loss=0.2861, simple_loss=0.3576, pruned_loss=0.1073, over 28600.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3638, pruned_loss=0.1136, over 5712644.67 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3596, pruned_loss=0.1043, over 3421746.98 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3637, pruned_loss=0.1143, over 5704014.04 frames. ], batch size: 92, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:29:34,998 INFO [train.py:968] (0/2) Epoch 7, batch 1900, giga_loss[loss=0.2764, simple_loss=0.3472, pruned_loss=0.1028, over 27967.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3618, pruned_loss=0.1118, over 5709205.99 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3595, pruned_loss=0.1042, over 3494760.27 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3618, pruned_loss=0.1126, over 5697477.68 frames. ], batch size: 412, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:29:35,965 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:29:36,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3174, 1.5066, 1.2708, 1.4760], device='cuda:0'), covar=tensor([0.2164, 0.1957, 0.1973, 0.1785], device='cuda:0'), in_proj_covar=tensor([0.1164, 0.0887, 0.1021, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 13:30:01,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.153e+02 1.053e+03 1.243e+03 1.713e+03 3.519e+03, threshold=2.487e+03, percent-clipped=5.0
+2023-03-03 13:30:20,356 INFO [train.py:968] (0/2) Epoch 7, batch 1950, giga_loss[loss=0.2493, simple_loss=0.3279, pruned_loss=0.08537, over 29002.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3584, pruned_loss=0.1095, over 5707896.52 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3602, pruned_loss=0.1049, over 3577607.63 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3579, pruned_loss=0.1099, over 5691992.74 frames. ], batch size: 164, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:30:39,253 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-03 13:31:05,873 INFO [train.py:968] (0/2) Epoch 7, batch 2000, giga_loss[loss=0.251, simple_loss=0.3236, pruned_loss=0.08924, over 27781.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3517, pruned_loss=0.1056, over 5690006.79 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3602, pruned_loss=0.1049, over 3646849.27 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3512, pruned_loss=0.106, over 5680216.13 frames. ], batch size: 474, lr: 4.78e-03, grad_scale: 8.0
+2023-03-03 13:31:18,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:31:31,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 4.900e+02 1.065e+03 1.429e+03 2.075e+03 5.775e+03, threshold=2.857e+03, percent-clipped=16.0
+2023-03-03 13:31:49,579 INFO [train.py:968] (0/2) Epoch 7, batch 2050, giga_loss[loss=0.2433, simple_loss=0.318, pruned_loss=0.08432, over 28640.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3463, pruned_loss=0.1027, over 5685592.96 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3602, pruned_loss=0.1047, over 3723639.47 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3455, pruned_loss=0.1031, over 5671684.42 frames. ], batch size: 92, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:31:49,997 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:32:16,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4871, 2.1199, 1.5252, 0.7304], device='cuda:0'), covar=tensor([0.3161, 0.1463, 0.2286, 0.3347], device='cuda:0'), in_proj_covar=tensor([0.1423, 0.1337, 0.1373, 0.1167], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 13:32:27,244 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275740.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 13:32:28,079 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-03 13:32:36,409 INFO [train.py:968] (0/2) Epoch 7, batch 2100, giga_loss[loss=0.2538, simple_loss=0.3371, pruned_loss=0.08523, over 28873.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3441, pruned_loss=0.1012, over 5688265.53 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3604, pruned_loss=0.1049, over 3780816.60 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3429, pruned_loss=0.1013, over 5678205.11 frames. ], batch size: 136, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:32:55,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1605, 4.9646, 4.7773, 2.1584], device='cuda:0'), covar=tensor([0.0348, 0.0431, 0.0530, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0830, 0.0762, 0.0599], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 13:33:00,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.096e+02 1.078e+03 1.390e+03 2.147e+03 9.433e+03, threshold=2.781e+03, percent-clipped=11.0
+2023-03-03 13:33:00,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:33:18,670 INFO [train.py:968] (0/2) Epoch 7, batch 2150, giga_loss[loss=0.2642, simple_loss=0.3423, pruned_loss=0.09306, over 28727.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3457, pruned_loss=0.1021, over 5692076.73 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3612, pruned_loss=0.1054, over 3848871.29 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3438, pruned_loss=0.1017, over 5681996.00 frames. ], batch size: 119, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:33:22,748 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:33:25,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:33:48,104 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:33:59,644 INFO [train.py:968] (0/2) Epoch 7, batch 2200, giga_loss[loss=0.2406, simple_loss=0.3181, pruned_loss=0.08159, over 28808.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3451, pruned_loss=0.1014, over 5699416.31 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3611, pruned_loss=0.1052, over 3889666.34 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3433, pruned_loss=0.1012, over 5687647.80 frames. ], batch size: 119, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:34:25,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.305e+02 9.394e+02 1.300e+03 1.757e+03 6.820e+03, threshold=2.599e+03, percent-clipped=9.0
+2023-03-03 13:34:37,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2762, 1.7097, 1.6655, 1.3310], device='cuda:0'), covar=tensor([0.1518, 0.1855, 0.1138, 0.1285], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0723, 0.0814, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 13:34:39,826 INFO [train.py:968] (0/2) Epoch 7, batch 2250, giga_loss[loss=0.247, simple_loss=0.3139, pruned_loss=0.09004, over 28802.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3426, pruned_loss=0.09998, over 5703535.19 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.361, pruned_loss=0.1049, over 3945622.26 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3408, pruned_loss=0.09989, over 5692020.11 frames. ], batch size: 186, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:34:49,363 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-03 13:35:01,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:35:25,083 INFO [train.py:968] (0/2) Epoch 7, batch 2300, giga_loss[loss=0.3006, simple_loss=0.3578, pruned_loss=0.1217, over 27698.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3406, pruned_loss=0.09974, over 5699698.96 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3614, pruned_loss=0.1051, over 3951967.14 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3388, pruned_loss=0.09953, over 5692732.80 frames. ], batch size: 472, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:35:47,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.181e+02 9.696e+02 1.232e+03 1.591e+03 3.631e+03, threshold=2.463e+03, percent-clipped=5.0
+2023-03-03 13:35:52,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8241, 1.1481, 3.4154, 2.8577], device='cuda:0'), covar=tensor([0.1686, 0.2306, 0.0422, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0580, 0.0534, 0.0757, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:36:06,970 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-276000.pt
+2023-03-03 13:36:07,257 INFO [train.py:968] (0/2) Epoch 7, batch 2350, giga_loss[loss=0.2457, simple_loss=0.3223, pruned_loss=0.08457, over 28633.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3376, pruned_loss=0.09794, over 5708788.33 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3611, pruned_loss=0.1048, over 3990980.35 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.336, pruned_loss=0.09782, over 5706508.91 frames. ], batch size: 262, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:36:23,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9952, 2.5610, 2.4931, 1.9117], device='cuda:0'), covar=tensor([0.1656, 0.1715, 0.1115, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0727, 0.0819, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 13:36:49,952 INFO [train.py:968] (0/2) Epoch 7, batch 2400, giga_loss[loss=0.266, simple_loss=0.335, pruned_loss=0.09852, over 28307.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3358, pruned_loss=0.09717, over 5717512.68 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3617, pruned_loss=0.105, over 4038102.79 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3335, pruned_loss=0.09681, over 5711486.52 frames. ], batch size: 368, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:37:05,781 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:37:08,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:37:10,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:37:12,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.788e+02 8.759e+02 1.168e+03 1.515e+03 4.696e+03, threshold=2.336e+03, percent-clipped=4.0
+2023-03-03 13:37:30,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2029, 1.4506, 1.2525, 1.5523], device='cuda:0'), covar=tensor([0.0775, 0.0331, 0.0323, 0.0783], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0117, 0.0121, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 13:37:31,372 INFO [train.py:968] (0/2) Epoch 7, batch 2450, giga_loss[loss=0.2373, simple_loss=0.3181, pruned_loss=0.0783, over 28855.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3334, pruned_loss=0.09618, over 5714980.82 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3623, pruned_loss=0.1052, over 4046555.99 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.331, pruned_loss=0.09567, over 5717766.59 frames. ], batch size: 174, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:37:32,184 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:37:42,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276115.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 13:38:09,918 INFO [train.py:968] (0/2) Epoch 7, batch 2500, giga_loss[loss=0.2628, simple_loss=0.3282, pruned_loss=0.09874, over 28949.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3316, pruned_loss=0.0949, over 5717896.86 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3626, pruned_loss=0.105, over 4113785.96 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3285, pruned_loss=0.0943, over 5718946.93 frames. ], batch size: 186, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:38:13,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:38:32,867 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.829e+02 9.795e+02 1.205e+03 1.559e+03 5.432e+03, threshold=2.409e+03, percent-clipped=7.0
+2023-03-03 13:38:50,751 INFO [train.py:968] (0/2) Epoch 7, batch 2550, libri_loss[loss=0.3551, simple_loss=0.4293, pruned_loss=0.1405, over 29658.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3303, pruned_loss=0.09427, over 5715801.86 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3638, pruned_loss=0.1055, over 4148628.45 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3264, pruned_loss=0.09322, over 5713651.02 frames. ], batch size: 88, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:39:05,938 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:39:08,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:39:31,810 INFO [train.py:968] (0/2) Epoch 7, batch 2600, giga_loss[loss=0.2755, simple_loss=0.3427, pruned_loss=0.1042, over 27625.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3291, pruned_loss=0.09341, over 5712867.40 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3646, pruned_loss=0.1059, over 4183266.85 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3247, pruned_loss=0.09207, over 5714667.41 frames. ], batch size: 472, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:39:31,984 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:39:32,231 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 13:39:34,325 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:39:37,706 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276258.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 13:39:40,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276261.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 13:39:53,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.202e+02 9.040e+02 1.147e+03 1.564e+03 4.453e+03, threshold=2.295e+03, percent-clipped=6.0
+2023-03-03 13:40:02,259 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276290.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 13:40:06,576 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:40:08,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:40:09,338 INFO [train.py:968] (0/2) Epoch 7, batch 2650, giga_loss[loss=0.274, simple_loss=0.3354, pruned_loss=0.1063, over 28737.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.329, pruned_loss=0.09305, over 5721397.18 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3661, pruned_loss=0.1065, over 4233898.95 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3234, pruned_loss=0.09123, over 5718115.31 frames. ], batch size: 99, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:40:33,195 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:40:49,192 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:40:49,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2754, 2.0464, 2.1504, 1.8168], device='cuda:0'), covar=tensor([0.1300, 0.2225, 0.1692, 0.1848], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0741, 0.0653, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 13:40:53,778 INFO [train.py:968] (0/2) Epoch 7, batch 2700, giga_loss[loss=0.3126, simple_loss=0.376, pruned_loss=0.1247, over 27556.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3307, pruned_loss=0.0943, over 5713013.43 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3669, pruned_loss=0.1068, over 4258648.12 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3253, pruned_loss=0.09248, over 5708204.10 frames. ], batch size: 472, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:41:17,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.956e+02 1.000e+03 1.324e+03 1.946e+03 3.734e+03, threshold=2.649e+03, percent-clipped=17.0
+2023-03-03 13:41:36,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6008, 3.1246, 1.6153, 1.4968], device='cuda:0'), covar=tensor([0.0787, 0.0321, 0.0750, 0.1242], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0480, 0.0313, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 13:41:36,689 INFO [train.py:968] (0/2) Epoch 7, batch 2750, giga_loss[loss=0.3089, simple_loss=0.3742, pruned_loss=0.1217, over 28709.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3366, pruned_loss=0.09802, over 5714492.34 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3668, pruned_loss=0.1066, over 4303810.93 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3314, pruned_loss=0.09642, over 5709168.32 frames. ], batch size: 262, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:41:36,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4919, 1.8749, 1.8724, 1.5226], device='cuda:0'), covar=tensor([0.1390, 0.1646, 0.1037, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0729, 0.0825, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 13:41:37,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4607, 3.4167, 1.5737, 1.4565], device='cuda:0'), covar=tensor([0.0879, 0.0314, 0.0790, 0.1347], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0480, 0.0313, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 13:41:46,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1507, 1.2679, 0.8805, 1.0672], device='cuda:0'), covar=tensor([0.0894, 0.0816, 0.0714, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.1502, 0.1333, 0.1316, 0.1419], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 13:42:19,716 INFO [train.py:968] (0/2) Epoch 7, batch 2800, giga_loss[loss=0.304, simple_loss=0.3746, pruned_loss=0.1167, over 28972.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3446, pruned_loss=0.1035, over 5696213.77 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3673, pruned_loss=0.1068, over 4353775.47 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.339, pruned_loss=0.1018, over 5696844.04 frames. ], batch size: 164, lr: 4.77e-03, grad_scale: 8.0
+2023-03-03 13:42:37,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7735, 0.9512, 3.6676, 2.9545], device='cuda:0'), covar=tensor([0.1813, 0.2386, 0.0412, 0.0776], device='cuda:0'), in_proj_covar=tensor([0.0588, 0.0542, 0.0770, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:42:42,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.564e+02 1.280e+03 1.683e+03 2.225e+03 5.373e+03, threshold=3.366e+03, percent-clipped=16.0
+2023-03-03 13:43:01,024 INFO [train.py:968] (0/2) Epoch 7, batch 2850, giga_loss[loss=0.3, simple_loss=0.3687, pruned_loss=0.1157, over 28730.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3523, pruned_loss=0.1085, over 5687030.96 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3673, pruned_loss=0.1068, over 4411425.09 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3471, pruned_loss=0.1072, over 5689453.17 frames. ], batch size: 284, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:43:42,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 13:43:48,312 INFO [train.py:968] (0/2) Epoch 7, batch 2900, libri_loss[loss=0.2755, simple_loss=0.3599, pruned_loss=0.09551, over 29653.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3577, pruned_loss=0.1111, over 5677660.65 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3674, pruned_loss=0.107, over 4438908.05 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3532, pruned_loss=0.11, over 5684006.60 frames. ], batch size: 88, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:44:02,735 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-03 13:44:15,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.135e+02 1.061e+03 1.353e+03 1.847e+03 3.656e+03, threshold=2.706e+03, percent-clipped=2.0
+2023-03-03 13:44:21,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8107, 1.0478, 3.2699, 2.8180], device='cuda:0'), covar=tensor([0.1762, 0.2512, 0.0483, 0.0891], device='cuda:0'), in_proj_covar=tensor([0.0586, 0.0542, 0.0766, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:44:34,675 INFO [train.py:968] (0/2) Epoch 7, batch 2950, giga_loss[loss=0.3026, simple_loss=0.3761, pruned_loss=0.1146, over 28899.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3622, pruned_loss=0.113, over 5666918.16 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.367, pruned_loss=0.1068, over 4459174.87 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3588, pruned_loss=0.1122, over 5670204.67 frames. ], batch size: 136, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:44:57,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:45:19,813 INFO [train.py:968] (0/2) Epoch 7, batch 3000, giga_loss[loss=0.3395, simple_loss=0.3974, pruned_loss=0.1408, over 28904.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.367, pruned_loss=0.1157, over 5681941.03 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3663, pruned_loss=0.1063, over 4518914.58 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3647, pruned_loss=0.1157, over 5679084.69 frames. ], batch size: 199, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:45:19,818 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 13:45:25,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3113, 1.4172, 1.1123, 1.4758], device='cuda:0'), covar=tensor([0.0852, 0.0320, 0.0363, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0120, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 13:45:27,428 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3019, 1.5852, 1.3091, 1.3228], device='cuda:0'), covar=tensor([0.2271, 0.2123, 0.2104, 0.1858], device='cuda:0'), in_proj_covar=tensor([0.1170, 0.0894, 0.1029, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:45:28,383 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2341, simple_loss=0.3363, pruned_loss=0.06595, over 944034.00 frames. 
+2023-03-03 13:45:28,384 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 13:45:55,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.901e+02 1.104e+03 1.409e+03 1.916e+03 8.953e+03, threshold=2.818e+03, percent-clipped=13.0
+2023-03-03 13:46:00,443 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1772, 1.3583, 3.3471, 3.1006], device='cuda:0'), covar=tensor([0.1282, 0.2120, 0.0377, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0584, 0.0540, 0.0764, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:46:12,265 INFO [train.py:968] (0/2) Epoch 7, batch 3050, giga_loss[loss=0.2817, simple_loss=0.3448, pruned_loss=0.1093, over 28830.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3661, pruned_loss=0.1153, over 5673995.66 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3666, pruned_loss=0.1069, over 4539793.62 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.364, pruned_loss=0.1151, over 5668393.04 frames. ], batch size: 199, lr: 4.77e-03, grad_scale: 2.0
+2023-03-03 13:46:21,403 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:46:23,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3598, 2.1883, 2.3561, 2.0322], device='cuda:0'), covar=tensor([0.1114, 0.1899, 0.1352, 0.1512], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0726, 0.0641, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 13:46:30,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:46:41,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3767, 1.6388, 1.1942, 1.5131], device='cuda:0'), covar=tensor([0.0720, 0.0277, 0.0317, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0121, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 13:46:54,308 INFO [train.py:968] (0/2) Epoch 7, batch 3100, giga_loss[loss=0.3262, simple_loss=0.3844, pruned_loss=0.134, over 28532.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3602, pruned_loss=0.1109, over 5678945.99 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3663, pruned_loss=0.1068, over 4562252.23 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3587, pruned_loss=0.1109, over 5674593.00 frames. ], batch size: 336, lr: 4.77e-03, grad_scale: 2.0
+2023-03-03 13:47:12,995 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:47:17,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:47:20,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.540e+02 1.016e+03 1.334e+03 1.863e+03 5.934e+03, threshold=2.669e+03, percent-clipped=8.0
+2023-03-03 13:47:40,837 INFO [train.py:968] (0/2) Epoch 7, batch 3150, giga_loss[loss=0.3369, simple_loss=0.3867, pruned_loss=0.1436, over 23403.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3591, pruned_loss=0.1098, over 5673110.47 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3661, pruned_loss=0.1067, over 4575297.99 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.358, pruned_loss=0.1099, over 5667979.39 frames. ], batch size: 705, lr: 4.77e-03, grad_scale: 2.0
+2023-03-03 13:47:42,952 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:48:23,577 INFO [train.py:968] (0/2) Epoch 7, batch 3200, libri_loss[loss=0.2841, simple_loss=0.3641, pruned_loss=0.102, over 29181.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3587, pruned_loss=0.1092, over 5677682.58 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3653, pruned_loss=0.1064, over 4611720.78 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3581, pruned_loss=0.1096, over 5670109.64 frames. ], batch size: 97, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:48:34,087 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:48:35,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:48:41,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3431, 1.5415, 1.2422, 1.3800], device='cuda:0'), covar=tensor([0.2020, 0.1952, 0.2108, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.1171, 0.0891, 0.1034, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:48:49,080 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.210e+02 1.236e+03 1.665e+03 2.395e+03 7.153e+03, threshold=3.330e+03, percent-clipped=15.0
+2023-03-03 13:49:03,177 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:49:06,726 INFO [train.py:968] (0/2) Epoch 7, batch 3250, giga_loss[loss=0.325, simple_loss=0.3932, pruned_loss=0.1284, over 28754.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3618, pruned_loss=0.1111, over 5678749.64 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3656, pruned_loss=0.1065, over 4630897.79 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3612, pruned_loss=0.1114, over 5670032.17 frames. ], batch size: 284, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:49:51,755 INFO [train.py:968] (0/2) Epoch 7, batch 3300, giga_loss[loss=0.3952, simple_loss=0.4223, pruned_loss=0.1841, over 26603.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3653, pruned_loss=0.1136, over 5692010.63 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3656, pruned_loss=0.1065, over 4649867.07 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3647, pruned_loss=0.1139, over 5682213.90 frames. ], batch size: 555, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:50:17,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.244e+02 1.120e+03 1.402e+03 2.037e+03 5.852e+03, threshold=2.804e+03, percent-clipped=8.0
+2023-03-03 13:50:34,183 INFO [train.py:968] (0/2) Epoch 7, batch 3350, giga_loss[loss=0.2573, simple_loss=0.3372, pruned_loss=0.08872, over 28641.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3662, pruned_loss=0.1142, over 5694026.15 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3657, pruned_loss=0.1064, over 4707774.53 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3657, pruned_loss=0.1149, over 5679637.66 frames. ], batch size: 60, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:50:37,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:50:54,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4583, 1.0976, 5.0558, 3.4604], device='cuda:0'), covar=tensor([0.1610, 0.2509, 0.0279, 0.0683], device='cuda:0'), in_proj_covar=tensor([0.0584, 0.0539, 0.0766, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:51:02,150 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.37 vs. limit=5.0
+2023-03-03 13:51:16,991 INFO [train.py:968] (0/2) Epoch 7, batch 3400, giga_loss[loss=0.2655, simple_loss=0.3404, pruned_loss=0.09525, over 29035.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3679, pruned_loss=0.1164, over 5687397.62 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3656, pruned_loss=0.1063, over 4710302.66 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3676, pruned_loss=0.1171, over 5681991.56 frames. ], batch size: 128, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:51:28,042 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:51:43,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.872e+02 1.255e+03 1.635e+03 2.455e+03 7.330e+03, threshold=3.270e+03, percent-clipped=15.0
+2023-03-03 13:51:48,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:52:00,083 INFO [train.py:968] (0/2) Epoch 7, batch 3450, giga_loss[loss=0.3012, simple_loss=0.3664, pruned_loss=0.118, over 28786.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3688, pruned_loss=0.1172, over 5676382.66 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3659, pruned_loss=0.1063, over 4734674.90 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3683, pruned_loss=0.118, over 5675293.12 frames. ], batch size: 99, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:52:20,774 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 13:52:42,963 INFO [train.py:968] (0/2) Epoch 7, batch 3500, giga_loss[loss=0.2658, simple_loss=0.3442, pruned_loss=0.09367, over 28744.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3695, pruned_loss=0.117, over 5677241.66 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3658, pruned_loss=0.1065, over 4751684.83 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3693, pruned_loss=0.1178, over 5682627.65 frames. ], batch size: 99, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:52:54,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0993, 2.0507, 1.6034, 1.3908], device='cuda:0'), covar=tensor([0.0880, 0.0294, 0.0287, 0.1044], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0120, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 13:53:06,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.402e+02 1.163e+03 1.430e+03 2.051e+03 6.544e+03, threshold=2.859e+03, percent-clipped=7.0
+2023-03-03 13:53:10,058 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:53:25,599 INFO [train.py:968] (0/2) Epoch 7, batch 3550, giga_loss[loss=0.3134, simple_loss=0.38, pruned_loss=0.1234, over 28958.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3696, pruned_loss=0.1158, over 5678832.94 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3661, pruned_loss=0.1068, over 4757220.41 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3692, pruned_loss=0.1162, over 5682173.03 frames. ], batch size: 145, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:53:50,973 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3667, 2.2444, 2.1798, 2.0842], device='cuda:0'), covar=tensor([0.1107, 0.1911, 0.1526, 0.1458], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0729, 0.0647, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 13:53:50,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:53:52,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:54:08,001 INFO [train.py:968] (0/2) Epoch 7, batch 3600, giga_loss[loss=0.255, simple_loss=0.3427, pruned_loss=0.08368, over 29003.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3688, pruned_loss=0.1143, over 5677835.48 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3662, pruned_loss=0.1069, over 4765019.30 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3685, pruned_loss=0.1147, over 5685921.94 frames. ], batch size: 164, lr: 4.77e-03, grad_scale: 8.0
+2023-03-03 13:54:14,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:54:17,968 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:54:34,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.401e+02 9.533e+02 1.158e+03 1.532e+03 5.565e+03, threshold=2.316e+03, percent-clipped=6.0
+2023-03-03 13:54:37,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6014, 1.1004, 2.8579, 2.7107], device='cuda:0'), covar=tensor([0.1640, 0.2157, 0.0513, 0.0763], device='cuda:0'), in_proj_covar=tensor([0.0584, 0.0540, 0.0767, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 13:54:50,013 INFO [train.py:968] (0/2) Epoch 7, batch 3650, libri_loss[loss=0.3669, simple_loss=0.4195, pruned_loss=0.1572, over 28602.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3661, pruned_loss=0.1126, over 5685008.12 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3667, pruned_loss=0.1072, over 4779594.29 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3655, pruned_loss=0.1127, over 5690611.09 frames. ], batch size: 106, lr: 4.77e-03, grad_scale: 8.0
+2023-03-03 13:55:10,277 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:55:33,494 INFO [train.py:968] (0/2) Epoch 7, batch 3700, giga_loss[loss=0.2598, simple_loss=0.3307, pruned_loss=0.09444, over 28096.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3626, pruned_loss=0.1107, over 5689045.61 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3665, pruned_loss=0.1071, over 4796579.51 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3622, pruned_loss=0.1109, over 5690490.10 frames. ], batch size: 77, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:55:55,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:55:56,663 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.159e+02 1.089e+03 1.396e+03 1.867e+03 6.213e+03, threshold=2.791e+03, percent-clipped=13.0
+2023-03-03 13:56:10,676 INFO [train.py:968] (0/2) Epoch 7, batch 3750, giga_loss[loss=0.2941, simple_loss=0.3635, pruned_loss=0.1123, over 28978.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3609, pruned_loss=0.1097, over 5697084.57 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3672, pruned_loss=0.1076, over 4813118.68 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.36, pruned_loss=0.1096, over 5695474.41 frames. ], batch size: 106, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 13:56:15,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:56:44,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:56:53,665 INFO [train.py:968] (0/2) Epoch 7, batch 3800, giga_loss[loss=0.2777, simple_loss=0.3496, pruned_loss=0.1029, over 28235.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3615, pruned_loss=0.1101, over 5698944.23 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3668, pruned_loss=0.1072, over 4839546.45 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3609, pruned_loss=0.1103, over 5693685.75 frames. ], batch size: 77, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 13:57:19,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.372e+02 9.736e+02 1.194e+03 1.581e+03 7.750e+03, threshold=2.389e+03, percent-clipped=4.0
+2023-03-03 13:57:35,792 INFO [train.py:968] (0/2) Epoch 7, batch 3850, giga_loss[loss=0.2789, simple_loss=0.3509, pruned_loss=0.1034, over 28672.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3612, pruned_loss=0.1094, over 5706430.02 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3663, pruned_loss=0.1069, over 4860638.07 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3611, pruned_loss=0.1099, over 5698919.88 frames. ], batch size: 60, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 13:57:54,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:57:56,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:58:18,042 INFO [train.py:968] (0/2) Epoch 7, batch 3900, giga_loss[loss=0.256, simple_loss=0.3395, pruned_loss=0.08624, over 28990.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3616, pruned_loss=0.1091, over 5706433.41 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3665, pruned_loss=0.107, over 4867882.88 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3613, pruned_loss=0.1094, over 5701824.32 frames. ], batch size: 136, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 13:58:22,821 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:58:26,121 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:58:33,378 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 13:58:37,715 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:58:44,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:58:45,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.578e+02 1.016e+03 1.286e+03 1.685e+03 4.050e+03, threshold=2.572e+03, percent-clipped=9.0
+2023-03-03 13:58:47,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:59:01,313 INFO [train.py:968] (0/2) Epoch 7, batch 3950, giga_loss[loss=0.2956, simple_loss=0.3639, pruned_loss=0.1136, over 28703.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3616, pruned_loss=0.109, over 5695203.63 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3668, pruned_loss=0.1073, over 4873411.80 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.361, pruned_loss=0.109, over 5697189.26 frames. ], batch size: 262, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 13:59:11,360 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:59:11,953 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:59:26,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:59:42,888 INFO [train.py:968] (0/2) Epoch 7, batch 4000, giga_loss[loss=0.2724, simple_loss=0.3437, pruned_loss=0.1006, over 28953.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3621, pruned_loss=0.1099, over 5702087.56 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3672, pruned_loss=0.1074, over 4899189.49 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3612, pruned_loss=0.1098, over 5698668.24 frames. ], batch size: 213, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 13:59:54,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4815, 3.1742, 1.4333, 1.5137], device='cuda:0'), covar=tensor([0.0824, 0.0226, 0.0813, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0333, 0.0465, 0.0309, 0.0374], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 14:00:09,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.501e+02 9.649e+02 1.207e+03 1.528e+03 5.743e+03, threshold=2.414e+03, percent-clipped=7.0
+2023-03-03 14:00:21,563 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:00:22,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1559, 3.9759, 3.7713, 1.8396], device='cuda:0'), covar=tensor([0.0469, 0.0585, 0.0639, 0.2149], device='cuda:0'), in_proj_covar=tensor([0.0894, 0.0841, 0.0761, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 14:00:22,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:00:22,534 INFO [train.py:968] (0/2) Epoch 7, batch 4050, giga_loss[loss=0.2703, simple_loss=0.3476, pruned_loss=0.09652, over 28566.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3604, pruned_loss=0.1089, over 5705775.70 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3678, pruned_loss=0.1077, over 4913822.22 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3591, pruned_loss=0.1087, over 5707682.23 frames. ], batch size: 307, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:00:23,560 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:00:25,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:00:49,239 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:01:00,692 INFO [train.py:968] (0/2) Epoch 7, batch 4100, giga_loss[loss=0.2626, simple_loss=0.336, pruned_loss=0.09463, over 28924.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3568, pruned_loss=0.1071, over 5710561.34 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3675, pruned_loss=0.1075, over 4941712.73 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3558, pruned_loss=0.1071, over 5708174.85 frames. ], batch size: 227, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:01:22,478 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:01:25,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:01:26,843 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:01:28,049 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.492e+02 1.042e+03 1.341e+03 1.710e+03 2.959e+03, threshold=2.682e+03, percent-clipped=4.0
+2023-03-03 14:01:43,710 INFO [train.py:968] (0/2) Epoch 7, batch 4150, libri_loss[loss=0.2772, simple_loss=0.3554, pruned_loss=0.09947, over 29651.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.354, pruned_loss=0.1058, over 5706992.57 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3672, pruned_loss=0.1074, over 4957996.22 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3531, pruned_loss=0.1058, over 5703761.54 frames. ], batch size: 88, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:01:48,489 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:02:15,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:02:18,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:02:21,873 INFO [train.py:968] (0/2) Epoch 7, batch 4200, giga_loss[loss=0.3373, simple_loss=0.3831, pruned_loss=0.1457, over 28458.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3543, pruned_loss=0.1067, over 5715093.25 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3669, pruned_loss=0.1071, over 4976740.88 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3537, pruned_loss=0.1069, over 5709325.43 frames. ], batch size: 60, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:02:40,948 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:02:46,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.317e+02 1.078e+03 1.291e+03 1.856e+03 5.539e+03, threshold=2.583e+03, percent-clipped=9.0
+2023-03-03 14:03:04,255 INFO [train.py:968] (0/2) Epoch 7, batch 4250, giga_loss[loss=0.2816, simple_loss=0.3478, pruned_loss=0.1077, over 28853.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3532, pruned_loss=0.107, over 5714429.80 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3665, pruned_loss=0.1069, over 4995535.72 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3527, pruned_loss=0.1073, over 5706295.92 frames. ], batch size: 199, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:03:08,320 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-03 14:03:24,941 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:03:26,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:03:29,106 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3125, 2.1823, 1.6487, 1.9298], device='cuda:0'), covar=tensor([0.0629, 0.0638, 0.0905, 0.0950], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0436, 0.0495, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 14:03:42,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:03:43,569 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:03:45,372 INFO [train.py:968] (0/2) Epoch 7, batch 4300, giga_loss[loss=0.2528, simple_loss=0.3239, pruned_loss=0.09083, over 28573.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3504, pruned_loss=0.106, over 5715984.45 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3668, pruned_loss=0.1073, over 5008492.15 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3496, pruned_loss=0.106, over 5707469.98 frames. ], batch size: 78, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:03:48,448 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:03:57,728 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:04:10,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.275e+02 1.020e+03 1.323e+03 2.118e+03 4.322e+03, threshold=2.645e+03, percent-clipped=9.0
+2023-03-03 14:04:14,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:04:25,367 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-278000.pt
+2023-03-03 14:04:25,659 INFO [train.py:968] (0/2) Epoch 7, batch 4350, giga_loss[loss=0.2585, simple_loss=0.3248, pruned_loss=0.09608, over 28497.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3476, pruned_loss=0.1049, over 5717183.13 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3665, pruned_loss=0.1071, over 5021965.13 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3469, pruned_loss=0.105, over 5707742.25 frames. ], batch size: 85, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:05:01,992 INFO [train.py:968] (0/2) Epoch 7, batch 4400, giga_loss[loss=0.2789, simple_loss=0.3413, pruned_loss=0.1082, over 28487.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3473, pruned_loss=0.1044, over 5711341.01 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3665, pruned_loss=0.107, over 5052476.34 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3461, pruned_loss=0.1044, over 5703486.03 frames. ], batch size: 85, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:05:20,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:05:27,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.794e+02 9.932e+02 1.333e+03 1.791e+03 9.186e+03, threshold=2.666e+03, percent-clipped=6.0
+2023-03-03 14:05:30,265 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:05:33,710 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:05:35,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:05:41,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5077, 1.5697, 1.2620, 2.0043], device='cuda:0'), covar=tensor([0.2157, 0.2222, 0.2281, 0.2105], device='cuda:0'), in_proj_covar=tensor([0.1164, 0.0884, 0.1025, 0.0938], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 14:05:41,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0720, 2.8092, 1.6800, 1.3978], device='cuda:0'), covar=tensor([0.1758, 0.0761, 0.1090, 0.1488], device='cuda:0'), in_proj_covar=tensor([0.1498, 0.1334, 0.1323, 0.1403], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 14:05:42,849 INFO [train.py:968] (0/2) Epoch 7, batch 4450, giga_loss[loss=0.2993, simple_loss=0.3735, pruned_loss=0.1126, over 28759.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3494, pruned_loss=0.1055, over 5713053.01 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.366, pruned_loss=0.1069, over 5076090.28 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3482, pruned_loss=0.1056, over 5702972.70 frames. ], batch size: 284, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:05:59,757 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:06:09,289 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:06:13,080 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:06:21,494 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-03 14:06:26,332 INFO [train.py:968] (0/2) Epoch 7, batch 4500, giga_loss[loss=0.3154, simple_loss=0.3941, pruned_loss=0.1184, over 28363.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3519, pruned_loss=0.1063, over 5718137.73 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3658, pruned_loss=0.1069, over 5088596.55 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3509, pruned_loss=0.1063, over 5707604.10 frames. ], batch size: 368, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:06:27,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2447, 2.0397, 1.4382, 0.4017], device='cuda:0'), covar=tensor([0.2544, 0.1290, 0.2297, 0.3175], device='cuda:0'), in_proj_covar=tensor([0.1428, 0.1337, 0.1393, 0.1173], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 14:06:38,781 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:06:53,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.057e+02 1.058e+03 1.348e+03 1.751e+03 4.406e+03, threshold=2.696e+03, percent-clipped=4.0
+2023-03-03 14:07:09,031 INFO [train.py:968] (0/2) Epoch 7, batch 4550, giga_loss[loss=0.3022, simple_loss=0.3738, pruned_loss=0.1153, over 27927.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3544, pruned_loss=0.1069, over 5713172.44 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3661, pruned_loss=0.1071, over 5090475.71 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3532, pruned_loss=0.1068, over 5711984.28 frames. ], batch size: 412, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:07:14,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 14:07:24,550 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:07:27,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:07:53,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:07:55,077 INFO [train.py:968] (0/2) Epoch 7, batch 4600, giga_loss[loss=0.2724, simple_loss=0.3481, pruned_loss=0.09839, over 28788.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3545, pruned_loss=0.1065, over 5704057.87 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3658, pruned_loss=0.1069, over 5098493.29 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3538, pruned_loss=0.1065, over 5701490.45 frames. ], batch size: 119, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:08:24,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.256e+02 9.830e+02 1.268e+03 1.650e+03 3.478e+03, threshold=2.537e+03, percent-clipped=3.0
+2023-03-03 14:08:38,766 INFO [train.py:968] (0/2) Epoch 7, batch 4650, libri_loss[loss=0.3067, simple_loss=0.378, pruned_loss=0.1177, over 29661.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.354, pruned_loss=0.1055, over 5693043.67 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3662, pruned_loss=0.1073, over 5108532.49 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3527, pruned_loss=0.1052, over 5695376.95 frames. ], batch size: 88, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:08:56,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:09:10,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:09:19,446 INFO [train.py:968] (0/2) Epoch 7, batch 4700, libri_loss[loss=0.3063, simple_loss=0.3815, pruned_loss=0.1155, over 29252.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3549, pruned_loss=0.1055, over 5696504.09 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3668, pruned_loss=0.1075, over 5132614.41 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.353, pruned_loss=0.105, over 5694765.09 frames. ], batch size: 94, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:09:28,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:09:43,869 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.549e+02 1.155e+03 1.454e+03 1.856e+03 4.528e+03, threshold=2.908e+03, percent-clipped=8.0
+2023-03-03 14:09:59,209 INFO [train.py:968] (0/2) Epoch 7, batch 4750, libri_loss[loss=0.2963, simple_loss=0.3746, pruned_loss=0.109, over 29554.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3563, pruned_loss=0.1065, over 5702994.49 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3669, pruned_loss=0.1076, over 5149855.27 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3544, pruned_loss=0.1061, over 5703062.24 frames. ], batch size: 81, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:10:39,549 INFO [train.py:968] (0/2) Epoch 7, batch 4800, giga_loss[loss=0.2745, simple_loss=0.3514, pruned_loss=0.09882, over 28463.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3574, pruned_loss=0.1076, over 5710894.88 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3666, pruned_loss=0.1074, over 5165181.53 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3559, pruned_loss=0.1074, over 5707159.29 frames. ], batch size: 78, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:10:44,613 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-03 14:10:49,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:10:55,856 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:10:58,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:11:07,193 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:11:08,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.497e+02 1.218e+03 1.487e+03 2.216e+03 4.332e+03, threshold=2.974e+03, percent-clipped=10.0
+2023-03-03 14:11:09,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:11:20,229 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:11:22,281 INFO [train.py:968] (0/2) Epoch 7, batch 4850, libri_loss[loss=0.3241, simple_loss=0.3884, pruned_loss=0.1299, over 29518.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3607, pruned_loss=0.1102, over 5713334.01 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3668, pruned_loss=0.1075, over 5182661.12 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3593, pruned_loss=0.11, over 5706636.33 frames. ], batch size: 81, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:11:34,340 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:12:03,775 INFO [train.py:968] (0/2) Epoch 7, batch 4900, libri_loss[loss=0.3227, simple_loss=0.3921, pruned_loss=0.1266, over 20172.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3628, pruned_loss=0.1112, over 5703597.52 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3672, pruned_loss=0.1077, over 5192284.62 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3611, pruned_loss=0.1109, over 5704380.31 frames. ], batch size: 187, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:12:19,061 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-03 14:12:30,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.490e+02 1.234e+03 1.515e+03 2.448e+03 4.238e+03, threshold=3.030e+03, percent-clipped=12.0
+2023-03-03 14:12:32,416 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 14:12:45,138 INFO [train.py:968] (0/2) Epoch 7, batch 4950, giga_loss[loss=0.2579, simple_loss=0.3347, pruned_loss=0.0906, over 28831.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3643, pruned_loss=0.1119, over 5705843.18 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3676, pruned_loss=0.108, over 5202313.59 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3626, pruned_loss=0.1115, over 5704525.67 frames. ], batch size: 112, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:12:47,507 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:12:50,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:13:12,271 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:13:25,612 INFO [train.py:968] (0/2) Epoch 7, batch 5000, giga_loss[loss=0.2902, simple_loss=0.362, pruned_loss=0.1092, over 27924.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3658, pruned_loss=0.1129, over 5706861.56 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3684, pruned_loss=0.1085, over 5226580.91 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3636, pruned_loss=0.1123, over 5701590.12 frames. ], batch size: 412, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:13:51,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.498e+02 1.188e+03 1.463e+03 1.926e+03 6.431e+03, threshold=2.926e+03, percent-clipped=9.0
+2023-03-03 14:14:05,547 INFO [train.py:968] (0/2) Epoch 7, batch 5050, giga_loss[loss=0.2502, simple_loss=0.3368, pruned_loss=0.08177, over 29120.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3665, pruned_loss=0.1132, over 5709468.26 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3684, pruned_loss=0.1083, over 5246661.06 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3648, pruned_loss=0.113, over 5701830.07 frames. ], batch size: 155, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:14:18,588 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:14:37,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:14:39,839 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-03 14:14:45,882 INFO [train.py:968] (0/2) Epoch 7, batch 5100, giga_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1218, over 28627.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3654, pruned_loss=0.1128, over 5718182.32 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3684, pruned_loss=0.1084, over 5266126.61 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.364, pruned_loss=0.1128, over 5706854.89 frames. ], batch size: 307, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:14:59,546 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:15:13,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.526e+02 1.072e+03 1.493e+03 2.144e+03 7.890e+03, threshold=2.986e+03, percent-clipped=11.0
+2023-03-03 14:15:28,055 INFO [train.py:968] (0/2) Epoch 7, batch 5150, giga_loss[loss=0.3119, simple_loss=0.3649, pruned_loss=0.1295, over 23896.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3625, pruned_loss=0.1112, over 5697330.48 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3685, pruned_loss=0.1084, over 5269720.92 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3611, pruned_loss=0.1112, over 5697635.79 frames. ], batch size: 705, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:16:08,407 INFO [train.py:968] (0/2) Epoch 7, batch 5200, giga_loss[loss=0.2681, simple_loss=0.3452, pruned_loss=0.09547, over 28938.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3574, pruned_loss=0.1083, over 5706011.93 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3688, pruned_loss=0.1086, over 5275730.96 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3561, pruned_loss=0.1082, over 5704750.94 frames. ], batch size: 213, lr: 4.75e-03, grad_scale: 8.0
+2023-03-03 14:16:32,674 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1885, 4.0168, 3.7875, 1.8518], device='cuda:0'), covar=tensor([0.0456, 0.0548, 0.0632, 0.1972], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0846, 0.0775, 0.0615], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 14:16:34,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:16:35,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.682e+02 1.048e+03 1.229e+03 1.804e+03 4.628e+03, threshold=2.458e+03, percent-clipped=5.0
+2023-03-03 14:16:36,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:16:46,997 INFO [train.py:968] (0/2) Epoch 7, batch 5250, giga_loss[loss=0.2614, simple_loss=0.3375, pruned_loss=0.09266, over 28928.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3567, pruned_loss=0.1074, over 5714341.63 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3685, pruned_loss=0.1085, over 5298542.67 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3555, pruned_loss=0.1073, over 5708051.09 frames. ], batch size: 106, lr: 4.75e-03, grad_scale: 8.0
+2023-03-03 14:16:47,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2240, 1.2381, 1.1406, 1.0164], device='cuda:0'), covar=tensor([0.0649, 0.0498, 0.0972, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0449, 0.0499, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 14:16:50,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:17:00,224 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:17:13,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5429, 2.1762, 1.7686, 1.5936], device='cuda:0'), covar=tensor([0.0703, 0.0231, 0.0284, 0.0760], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0120, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 14:17:23,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1128, 0.9506, 0.8981, 1.2423], device='cuda:0'), covar=tensor([0.0712, 0.0372, 0.0344, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0120, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 14:17:31,031 INFO [train.py:968] (0/2) Epoch 7, batch 5300, libri_loss[loss=0.2974, simple_loss=0.3712, pruned_loss=0.1118, over 29668.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3588, pruned_loss=0.1073, over 5708704.59 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3688, pruned_loss=0.1088, over 5307241.53 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3572, pruned_loss=0.107, over 5707257.28 frames. ], batch size: 88, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:18:01,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.084e+02 1.005e+03 1.302e+03 1.651e+03 4.403e+03, threshold=2.604e+03, percent-clipped=4.0
+2023-03-03 14:18:13,780 INFO [train.py:968] (0/2) Epoch 7, batch 5350, giga_loss[loss=0.2696, simple_loss=0.3412, pruned_loss=0.09902, over 28985.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3597, pruned_loss=0.1069, over 5713730.15 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3689, pruned_loss=0.1089, over 5315587.02 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3581, pruned_loss=0.1065, over 5712035.05 frames. ], batch size: 128, lr: 4.75e-03, grad_scale: 2.0
+2023-03-03 14:18:52,282 INFO [train.py:968] (0/2) Epoch 7, batch 5400, giga_loss[loss=0.2735, simple_loss=0.3498, pruned_loss=0.09857, over 28588.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3595, pruned_loss=0.1082, over 5710391.09 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.369, pruned_loss=0.1092, over 5331415.20 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3577, pruned_loss=0.1076, over 5710248.54 frames. ], batch size: 307, lr: 4.75e-03, grad_scale: 2.0
+2023-03-03 14:19:22,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.348e+02 1.196e+03 1.497e+03 2.414e+03 9.624e+03, threshold=2.995e+03, percent-clipped=18.0
+2023-03-03 14:19:27,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=279091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:19:35,145 INFO [train.py:968] (0/2) Epoch 7, batch 5450, giga_loss[loss=0.329, simple_loss=0.3841, pruned_loss=0.1369, over 28952.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3587, pruned_loss=0.109, over 5718806.12 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3693, pruned_loss=0.1093, over 5342597.27 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3569, pruned_loss=0.1083, over 5715444.23 frames. ], batch size: 174, lr: 4.75e-03, grad_scale: 2.0
+2023-03-03 14:19:44,852 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-03 14:19:55,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6954, 1.6799, 1.6341, 1.4802], device='cuda:0'), covar=tensor([0.0938, 0.1288, 0.1425, 0.1271], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0721, 0.0641, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 14:20:09,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=279142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:20:15,518 INFO [train.py:968] (0/2) Epoch 7, batch 5500, giga_loss[loss=0.2888, simple_loss=0.3425, pruned_loss=0.1175, over 28615.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3562, pruned_loss=0.1093, over 5726841.27 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3687, pruned_loss=0.1093, over 5356544.23 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3549, pruned_loss=0.1088, over 5720259.88 frames. ], batch size: 92, lr: 4.75e-03, grad_scale: 2.0
+2023-03-03 14:20:15,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4518, 1.8820, 1.7084, 1.7618], device='cuda:0'), covar=tensor([0.0607, 0.0759, 0.0920, 0.1026], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0450, 0.0501, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 14:20:26,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7793, 2.4652, 1.5597, 1.1992], device='cuda:0'), covar=tensor([0.2244, 0.1002, 0.1370, 0.1766], device='cuda:0'), in_proj_covar=tensor([0.1528, 0.1352, 0.1348, 0.1446], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 14:20:39,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4799, 2.2678, 1.6269, 0.6274], device='cuda:0'), covar=tensor([0.3509, 0.1442, 0.2419, 0.3987], device='cuda:0'), in_proj_covar=tensor([0.1430, 0.1328, 0.1394, 0.1176], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 14:20:43,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.640e+02 1.119e+03 1.454e+03 1.824e+03 4.765e+03, threshold=2.909e+03, percent-clipped=6.0
+2023-03-03 14:20:55,518 INFO [train.py:968] (0/2) Epoch 7, batch 5550, libri_loss[loss=0.3123, simple_loss=0.3909, pruned_loss=0.1169, over 28671.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3538, pruned_loss=0.109, over 5733030.89 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3686, pruned_loss=0.1095, over 5379182.16 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3524, pruned_loss=0.1085, over 5722264.36 frames. ], batch size: 106, lr: 4.75e-03, grad_scale: 2.0
+2023-03-03 14:21:26,750 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=279234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:21:29,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=279237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:21:39,473 INFO [train.py:968] (0/2) Epoch 7, batch 5600, giga_loss[loss=0.2678, simple_loss=0.3289, pruned_loss=0.1033, over 28779.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3536, pruned_loss=0.1088, over 5726312.81 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3684, pruned_loss=0.1094, over 5391180.40 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3523, pruned_loss=0.1084, over 5714446.17 frames. ], batch size: 119, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:21:53,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5439, 3.3721, 1.5083, 1.6199], device='cuda:0'), covar=tensor([0.0819, 0.0323, 0.0852, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.0335, 0.0483, 0.0313, 0.0375], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:0')
+2023-03-03 14:21:54,125 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=279266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:22:04,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=279279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:22:08,126 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=279285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:22:08,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.061e+02 1.235e+03 1.507e+03 1.857e+03 4.172e+03, threshold=3.014e+03, percent-clipped=11.0
+2023-03-03 14:22:10,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=279288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:22:18,617 INFO [train.py:968] (0/2) Epoch 7, batch 5650, giga_loss[loss=0.2534, simple_loss=0.3271, pruned_loss=0.08989, over 28916.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3506, pruned_loss=0.1074, over 5720022.43 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.368, pruned_loss=0.1092, over 5399604.34 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3493, pruned_loss=0.1072, over 5713210.17 frames. ], batch size: 174, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:22:33,517 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=279317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:22:59,214 INFO [train.py:968] (0/2) Epoch 7, batch 5700, libri_loss[loss=0.2434, simple_loss=0.3155, pruned_loss=0.08559, over 29358.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3468, pruned_loss=0.1054, over 5709260.25 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3673, pruned_loss=0.1088, over 5406570.51 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3458, pruned_loss=0.1056, over 5707001.23 frames. ], batch size: 67, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:23:29,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.160e+02 1.196e+03 1.653e+03 2.537e+03 7.515e+03, threshold=3.306e+03, percent-clipped=11.0
+2023-03-03 14:23:39,468 INFO [train.py:968] (0/2) Epoch 7, batch 5750, giga_loss[loss=0.268, simple_loss=0.3432, pruned_loss=0.09639, over 28897.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3442, pruned_loss=0.104, over 5713297.62 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3673, pruned_loss=0.109, over 5415035.23 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.343, pruned_loss=0.1039, over 5709268.49 frames. ], batch size: 145, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:23:48,689 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-03 14:23:59,414 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=279422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:24:01,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=279425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:24:20,097 INFO [train.py:968] (0/2) Epoch 7, batch 5800, giga_loss[loss=0.2957, simple_loss=0.3515, pruned_loss=0.1199, over 23943.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3445, pruned_loss=0.1039, over 5711978.86 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3675, pruned_loss=0.109, over 5419996.51 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3432, pruned_loss=0.1037, over 5707012.14 frames. ], batch size: 705, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:24:23,854 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=279454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:24:46,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-03 14:24:48,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.505e+02 1.109e+03 1.523e+03 2.122e+03 6.095e+03, threshold=3.046e+03, percent-clipped=5.0
+2023-03-03 14:24:58,387 INFO [train.py:968] (0/2) Epoch 7, batch 5850, giga_loss[loss=0.2688, simple_loss=0.3419, pruned_loss=0.09782, over 29092.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3485, pruned_loss=0.1058, over 5719433.63 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3677, pruned_loss=0.1092, over 5434333.70 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3467, pruned_loss=0.1053, over 5710343.05 frames. ], batch size: 128, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:25:18,563 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=279524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:25:40,618 INFO [train.py:968] (0/2) Epoch 7, batch 5900, giga_loss[loss=0.2846, simple_loss=0.3594, pruned_loss=0.1049, over 28858.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3526, pruned_loss=0.1072, over 5714004.45 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3677, pruned_loss=0.1092, over 5440984.07 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3507, pruned_loss=0.1068, over 5708100.60 frames. ], batch size: 119, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:25:50,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7417, 2.7682, 1.6695, 0.8868], device='cuda:0'), covar=tensor([0.4352, 0.1736, 0.2594, 0.4154], device='cuda:0'), in_proj_covar=tensor([0.1432, 0.1333, 0.1396, 0.1177], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 14:26:12,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.631e+02 1.177e+03 1.523e+03 2.105e+03 5.460e+03, threshold=3.046e+03, percent-clipped=13.0
+2023-03-03 14:26:24,023 INFO [train.py:968] (0/2) Epoch 7, batch 5950, giga_loss[loss=0.3081, simple_loss=0.3767, pruned_loss=0.1197, over 27846.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3561, pruned_loss=0.1087, over 5717162.90 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3677, pruned_loss=0.1092, over 5448280.51 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3544, pruned_loss=0.1084, over 5709723.73 frames. ], batch size: 412, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:26:55,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4551, 1.6558, 1.3995, 1.7338], device='cuda:0'), covar=tensor([0.2277, 0.2135, 0.2284, 0.1782], device='cuda:0'), in_proj_covar=tensor([0.1173, 0.0886, 0.1029, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:0')
+2023-03-03 14:27:07,466 INFO [train.py:968] (0/2) Epoch 7, batch 6000, libri_loss[loss=0.2734, simple_loss=0.34, pruned_loss=0.1035, over 29650.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.359, pruned_loss=0.1104, over 5720324.59 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3671, pruned_loss=0.109, over 5467457.42 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3579, pruned_loss=0.1103, over 5707076.73 frames. ], batch size: 69, lr: 4.75e-03, grad_scale: 8.0
+2023-03-03 14:27:07,470 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 14:27:16,122 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2317, simple_loss=0.3342, pruned_loss=0.0646, over 944034.00 frames. 
+2023-03-03 14:27:16,123 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 14:27:49,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.136e+02 1.154e+03 1.486e+03 2.121e+03 8.799e+03, threshold=2.973e+03, percent-clipped=12.0
+2023-03-03 14:28:02,619 INFO [train.py:968] (0/2) Epoch 7, batch 6050, giga_loss[loss=0.3518, simple_loss=0.4135, pruned_loss=0.1451, over 28881.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3632, pruned_loss=0.1139, over 5700202.04 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3673, pruned_loss=0.1092, over 5459870.14 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3621, pruned_loss=0.1138, over 5697590.14 frames. ], batch size: 136, lr: 4.74e-03, grad_scale: 8.0
+2023-03-03 14:28:11,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3747, 2.0223, 1.4866, 0.6001], device='cuda:0'), covar=tensor([0.2755, 0.1353, 0.1975, 0.3222], device='cuda:0'), in_proj_covar=tensor([0.1433, 0.1335, 0.1391, 0.1179], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 14:28:39,206 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.94 vs. limit=5.0
+2023-03-03 14:28:51,453 INFO [train.py:968] (0/2) Epoch 7, batch 6100, giga_loss[loss=0.376, simple_loss=0.4196, pruned_loss=0.1662, over 28501.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3713, pruned_loss=0.1215, over 5691364.15 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3676, pruned_loss=0.1094, over 5464904.29 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3702, pruned_loss=0.1213, over 5688015.57 frames. ], batch size: 78, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:29:27,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.567e+02 1.392e+03 1.890e+03 2.264e+03 5.278e+03, threshold=3.781e+03, percent-clipped=16.0
+2023-03-03 14:29:38,948 INFO [train.py:968] (0/2) Epoch 7, batch 6150, giga_loss[loss=0.3286, simple_loss=0.3959, pruned_loss=0.1307, over 29025.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.378, pruned_loss=0.1258, over 5688447.18 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3682, pruned_loss=0.1098, over 5463363.42 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3767, pruned_loss=0.1257, over 5693759.17 frames. ], batch size: 155, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:30:21,902 INFO [train.py:968] (0/2) Epoch 7, batch 6200, giga_loss[loss=0.3492, simple_loss=0.4084, pruned_loss=0.145, over 28894.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3822, pruned_loss=0.1293, over 5690420.14 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3682, pruned_loss=0.11, over 5481654.99 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3818, pruned_loss=0.1299, over 5687910.18 frames. ], batch size: 186, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:30:53,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1317, 1.2288, 4.4400, 3.4068], device='cuda:0'), covar=tensor([0.1750, 0.2329, 0.0337, 0.0802], device='cuda:0'), in_proj_covar=tensor([0.0586, 0.0542, 0.0778, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 14:30:59,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.980e+02 1.589e+03 2.209e+03 3.497e+03 1.395e+04, threshold=4.417e+03, percent-clipped=21.0
+2023-03-03 14:31:11,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=279899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:31:11,624 INFO [train.py:968] (0/2) Epoch 7, batch 6250, giga_loss[loss=0.3943, simple_loss=0.4316, pruned_loss=0.1785, over 28807.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3881, pruned_loss=0.1346, over 5696228.45 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3684, pruned_loss=0.11, over 5489703.72 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3879, pruned_loss=0.1354, over 5690879.66 frames. ], batch size: 186, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:31:27,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3607, 2.0850, 1.6519, 1.7922], device='cuda:0'), covar=tensor([0.0639, 0.0696, 0.0890, 0.0979], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0450, 0.0500, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 14:31:30,989 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=279922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:31:31,081 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5170, 1.7303, 1.2564, 1.2153], device='cuda:0'), covar=tensor([0.1232, 0.1044, 0.0981, 0.1194], device='cuda:0'), in_proj_covar=tensor([0.1523, 0.1365, 0.1355, 0.1450], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 14:31:43,494 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=279935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:32:00,030 INFO [train.py:968] (0/2) Epoch 7, batch 6300, giga_loss[loss=0.3065, simple_loss=0.3764, pruned_loss=0.1183, over 28819.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3928, pruned_loss=0.1383, over 5691424.71 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.368, pruned_loss=0.1097, over 5498439.42 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3935, pruned_loss=0.1399, over 5683743.84 frames. ], batch size: 145, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:32:10,364 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1790, 1.4872, 1.3051, 1.4993], device='cuda:0'), covar=tensor([0.0785, 0.0310, 0.0303, 0.0797], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0117, 0.0120, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:0')
+2023-03-03 14:32:35,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.438e+02 1.475e+03 1.898e+03 2.394e+03 5.292e+03, threshold=3.796e+03, percent-clipped=4.0
+2023-03-03 14:32:47,163 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-280000.pt
+2023-03-03 14:32:47,466 INFO [train.py:968] (0/2) Epoch 7, batch 6350, giga_loss[loss=0.3186, simple_loss=0.3767, pruned_loss=0.1303, over 28229.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3942, pruned_loss=0.1398, over 5682685.58 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3679, pruned_loss=0.1096, over 5510165.40 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3957, pruned_loss=0.1422, over 5672500.02 frames. ], batch size: 77, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:32:52,533 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-03 14:33:27,593 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:33:29,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:33:34,827 INFO [train.py:968] (0/2) Epoch 7, batch 6400, giga_loss[loss=0.3782, simple_loss=0.4248, pruned_loss=0.1658, over 28911.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3965, pruned_loss=0.143, over 5673200.86 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3677, pruned_loss=0.1096, over 5524077.31 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3992, pruned_loss=0.1464, over 5658770.76 frames. ], batch size: 227, lr: 4.74e-03, grad_scale: 8.0
+2023-03-03 14:34:00,927 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:34:14,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.367e+02 1.569e+03 1.974e+03 2.709e+03 8.199e+03, threshold=3.947e+03, percent-clipped=6.0
+2023-03-03 14:34:22,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2844, 1.5424, 1.2541, 1.3663], device='cuda:0'), covar=tensor([0.1871, 0.1811, 0.1876, 0.1553], device='cuda:0'), in_proj_covar=tensor([0.1175, 0.0893, 0.1031, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 14:34:26,424 INFO [train.py:968] (0/2) Epoch 7, batch 6450, giga_loss[loss=0.3804, simple_loss=0.4312, pruned_loss=0.1648, over 28755.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3999, pruned_loss=0.1465, over 5681776.97 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3678, pruned_loss=0.1097, over 5537308.24 frames. ], giga_tot_loss[loss=0.3518, simple_loss=0.403, pruned_loss=0.1503, over 5662957.08 frames. ], batch size: 242, lr: 4.74e-03, grad_scale: 8.0
+2023-03-03 14:35:20,899 INFO [train.py:968] (0/2) Epoch 7, batch 6500, giga_loss[loss=0.3287, simple_loss=0.4025, pruned_loss=0.1274, over 28950.00 frames. ], tot_loss[loss=0.3539, simple_loss=0.4045, pruned_loss=0.1516, over 5660492.68 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3679, pruned_loss=0.1099, over 5541802.48 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4074, pruned_loss=0.1551, over 5643132.70 frames. ], batch size: 136, lr: 4.74e-03, grad_scale: 8.0
+2023-03-03 14:36:02,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.726e+02 1.788e+03 2.430e+03 3.093e+03 5.466e+03, threshold=4.860e+03, percent-clipped=9.0
+2023-03-03 14:36:14,205 INFO [train.py:968] (0/2) Epoch 7, batch 6550, giga_loss[loss=0.3535, simple_loss=0.4037, pruned_loss=0.1516, over 28682.00 frames. ], tot_loss[loss=0.3557, simple_loss=0.4054, pruned_loss=0.153, over 5656352.28 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3676, pruned_loss=0.1097, over 5548949.23 frames. ], giga_tot_loss[loss=0.3613, simple_loss=0.4088, pruned_loss=0.1568, over 5638624.97 frames. ], batch size: 307, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:36:22,369 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 14:37:01,914 INFO [train.py:968] (0/2) Epoch 7, batch 6600, giga_loss[loss=0.3524, simple_loss=0.3996, pruned_loss=0.1526, over 28798.00 frames. ], tot_loss[loss=0.3547, simple_loss=0.4041, pruned_loss=0.1527, over 5658096.83 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3675, pruned_loss=0.1097, over 5556075.82 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.4075, pruned_loss=0.1566, over 5639618.54 frames. ], batch size: 199, lr: 4.74e-03, grad_scale: 2.0
+2023-03-03 14:37:42,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.275e+02 1.602e+03 2.144e+03 3.137e+03 7.122e+03, threshold=4.288e+03, percent-clipped=5.0
+2023-03-03 14:37:51,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:37:55,066 INFO [train.py:968] (0/2) Epoch 7, batch 6650, giga_loss[loss=0.3183, simple_loss=0.3767, pruned_loss=0.1299, over 28887.00 frames. ], tot_loss[loss=0.3545, simple_loss=0.4033, pruned_loss=0.1528, over 5642753.34 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3675, pruned_loss=0.1096, over 5563222.78 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4069, pruned_loss=0.1571, over 5623583.11 frames. ], batch size: 112, lr: 4.74e-03, grad_scale: 2.0
+2023-03-03 14:38:07,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:38:19,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3874, 1.4649, 1.3673, 1.5667], device='cuda:0'), covar=tensor([0.0743, 0.0313, 0.0289, 0.0777], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0117, 0.0121, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0065, 0.0047, 0.0043, 0.0072], device='cuda:0')
+2023-03-03 14:38:33,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4296, 1.4215, 1.1197, 1.1118], device='cuda:0'), covar=tensor([0.0726, 0.0600, 0.1049, 0.0933], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0451, 0.0504, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 14:38:41,902 INFO [train.py:968] (0/2) Epoch 7, batch 6700, giga_loss[loss=0.3151, simple_loss=0.3741, pruned_loss=0.1281, over 28779.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4026, pruned_loss=0.1504, over 5647708.93 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3676, pruned_loss=0.1097, over 5561943.52 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.4066, pruned_loss=0.155, over 5636217.10 frames. ], batch size: 119, lr: 4.74e-03, grad_scale: 2.0
+2023-03-03 14:39:20,041 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 14:39:20,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.638e+02 1.525e+03 1.830e+03 2.446e+03 5.783e+03, threshold=3.659e+03, percent-clipped=2.0
+2023-03-03 14:39:22,375 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 14:39:31,119 INFO [train.py:968] (0/2) Epoch 7, batch 6750, giga_loss[loss=0.3482, simple_loss=0.4089, pruned_loss=0.1438, over 28842.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.404, pruned_loss=0.1511, over 5638946.22 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3674, pruned_loss=0.1096, over 5558713.18 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4079, pruned_loss=0.1555, over 5634512.75 frames. ], batch size: 119, lr: 4.74e-03, grad_scale: 2.0
+2023-03-03 14:40:08,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2294, 1.8237, 1.4237, 0.3252], device='cuda:0'), covar=tensor([0.2472, 0.1603, 0.2419, 0.3161], device='cuda:0'), in_proj_covar=tensor([0.1449, 0.1373, 0.1416, 0.1198], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 14:40:12,831 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:15,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:16,047 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:23,501 INFO [train.py:968] (0/2) Epoch 7, batch 6800, giga_loss[loss=0.3265, simple_loss=0.385, pruned_loss=0.134, over 28671.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4048, pruned_loss=0.1517, over 5632381.28 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3677, pruned_loss=0.1098, over 5565235.77 frames. ], giga_tot_loss[loss=0.36, simple_loss=0.4083, pruned_loss=0.1558, over 5624682.43 frames. ], batch size: 262, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:40:26,625 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:30,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:42,164 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:51,864 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:55,485 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:59,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.637e+02 1.586e+03 1.976e+03 2.832e+03 7.137e+03, threshold=3.952e+03, percent-clipped=14.0
+2023-03-03 14:41:13,882 INFO [train.py:968] (0/2) Epoch 7, batch 6850, libri_loss[loss=0.3106, simple_loss=0.3899, pruned_loss=0.1156, over 29320.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.4006, pruned_loss=0.1476, over 5635833.57 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3668, pruned_loss=0.1091, over 5579117.14 frames. ], giga_tot_loss[loss=0.3561, simple_loss=0.4057, pruned_loss=0.1533, over 5619454.13 frames. ], batch size: 94, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:41:54,624 INFO [train.py:968] (0/2) Epoch 7, batch 6900, giga_loss[loss=0.3807, simple_loss=0.4293, pruned_loss=0.166, over 28890.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3985, pruned_loss=0.1443, over 5659457.82 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3665, pruned_loss=0.109, over 5595659.15 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4045, pruned_loss=0.1507, over 5633627.88 frames. ], batch size: 227, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:42:36,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.680e+02 1.708e+03 2.347e+03 3.522e+03 6.811e+03, threshold=4.694e+03, percent-clipped=15.0
+2023-03-03 14:42:47,395 INFO [train.py:968] (0/2) Epoch 7, batch 6950, giga_loss[loss=0.2812, simple_loss=0.3582, pruned_loss=0.1021, over 29005.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3954, pruned_loss=0.1416, over 5659631.02 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3664, pruned_loss=0.109, over 5597938.04 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.4004, pruned_loss=0.147, over 5637869.55 frames. ], batch size: 128, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:43:36,982 INFO [train.py:968] (0/2) Epoch 7, batch 7000, giga_loss[loss=0.3173, simple_loss=0.367, pruned_loss=0.1339, over 23884.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3917, pruned_loss=0.1384, over 5653967.92 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3661, pruned_loss=0.1087, over 5601972.03 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3966, pruned_loss=0.1436, over 5634560.29 frames. ], batch size: 705, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:43:41,464 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-03 14:44:12,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.620e+02 1.563e+03 1.942e+03 2.513e+03 4.943e+03, threshold=3.883e+03, percent-clipped=1.0
+2023-03-03 14:44:23,022 INFO [train.py:968] (0/2) Epoch 7, batch 7050, giga_loss[loss=0.3583, simple_loss=0.4021, pruned_loss=0.1573, over 28730.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.391, pruned_loss=0.138, over 5641910.01 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3666, pruned_loss=0.109, over 5591634.49 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3952, pruned_loss=0.1428, over 5636731.27 frames. ], batch size: 262, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:44:42,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:45:12,612 INFO [train.py:968] (0/2) Epoch 7, batch 7100, giga_loss[loss=0.3527, simple_loss=0.3967, pruned_loss=0.1544, over 28917.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3918, pruned_loss=0.1388, over 5644752.42 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3664, pruned_loss=0.1088, over 5594214.58 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3956, pruned_loss=0.1431, over 5639099.15 frames. ], batch size: 213, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:45:23,875 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-03 14:45:58,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.835e+02 1.438e+03 2.000e+03 2.890e+03 1.127e+04, threshold=4.001e+03, percent-clipped=11.0
+2023-03-03 14:45:58,979 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9184, 3.7387, 3.5509, 1.8679], device='cuda:0'), covar=tensor([0.0555, 0.0673, 0.0673, 0.1938], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0881, 0.0799, 0.0621], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 14:46:09,159 INFO [train.py:968] (0/2) Epoch 7, batch 7150, giga_loss[loss=0.3185, simple_loss=0.373, pruned_loss=0.132, over 27619.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3905, pruned_loss=0.1376, over 5643163.71 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3663, pruned_loss=0.1088, over 5593474.10 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3944, pruned_loss=0.1418, over 5641298.55 frames. ], batch size: 472, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:46:28,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:46:31,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2871, 1.6753, 1.5953, 1.2706], device='cuda:0'), covar=tensor([0.1574, 0.1986, 0.1238, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0722, 0.0802, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 14:47:00,840 INFO [train.py:968] (0/2) Epoch 7, batch 7200, giga_loss[loss=0.3244, simple_loss=0.4086, pruned_loss=0.1201, over 28879.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3887, pruned_loss=0.1344, over 5648719.69 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.366, pruned_loss=0.1087, over 5597931.72 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3923, pruned_loss=0.1381, over 5643958.19 frames. ], batch size: 284, lr: 4.74e-03, grad_scale: 8.0
+2023-03-03 14:47:09,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4189, 1.7319, 1.3117, 1.2371], device='cuda:0'), covar=tensor([0.1602, 0.1049, 0.0903, 0.1190], device='cuda:0'), in_proj_covar=tensor([0.1519, 0.1353, 0.1317, 0.1423], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 14:47:10,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:47:13,254 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-03 14:47:33,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1706, 1.2716, 4.2495, 3.3128], device='cuda:0'), covar=tensor([0.1655, 0.2370, 0.0356, 0.0926], device='cuda:0'), in_proj_covar=tensor([0.0595, 0.0554, 0.0791, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 14:47:46,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.353e+02 1.299e+03 1.676e+03 2.436e+03 5.558e+03, threshold=3.351e+03, percent-clipped=6.0
+2023-03-03 14:47:53,466 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-03 14:47:55,356 INFO [train.py:968] (0/2) Epoch 7, batch 7250, libri_loss[loss=0.2108, simple_loss=0.2931, pruned_loss=0.06418, over 28520.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3892, pruned_loss=0.1327, over 5661397.71 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3653, pruned_loss=0.1084, over 5604953.15 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3932, pruned_loss=0.1366, over 5652456.31 frames. ], batch size: 63, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 14:48:04,808 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:48:19,692 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:48:27,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8099, 1.8055, 1.7254, 1.6011], device='cuda:0'), covar=tensor([0.1092, 0.1782, 0.1659, 0.1648], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0734, 0.0654, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 14:48:32,833 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:48:42,166 INFO [train.py:968] (0/2) Epoch 7, batch 7300, giga_loss[loss=0.3577, simple_loss=0.4113, pruned_loss=0.1521, over 28574.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3911, pruned_loss=0.1338, over 5668421.93 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3646, pruned_loss=0.1082, over 5607119.77 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3959, pruned_loss=0.138, over 5661989.85 frames. ], batch size: 307, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 14:48:57,569 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:01,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:22,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.155e+03 1.768e+03 2.219e+03 3.016e+03 7.061e+03, threshold=4.437e+03, percent-clipped=18.0
+2023-03-03 14:49:29,015 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:34,806 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:35,139 INFO [train.py:968] (0/2) Epoch 7, batch 7350, giga_loss[loss=0.3518, simple_loss=0.3851, pruned_loss=0.1592, over 23506.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3923, pruned_loss=0.136, over 5648565.29 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3647, pruned_loss=0.1083, over 5603346.77 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3967, pruned_loss=0.1399, over 5647993.91 frames. ], batch size: 705, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 14:49:36,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:41,450 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:59,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2042, 4.0209, 3.8335, 1.7329], device='cuda:0'), covar=tensor([0.0503, 0.0619, 0.0673, 0.2076], device='cuda:0'), in_proj_covar=tensor([0.0946, 0.0894, 0.0810, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 14:50:01,418 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:50:18,231 INFO [train.py:968] (0/2) Epoch 7, batch 7400, libri_loss[loss=0.2895, simple_loss=0.3633, pruned_loss=0.1078, over 29537.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3919, pruned_loss=0.1361, over 5671710.22 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3648, pruned_loss=0.1084, over 5617884.33 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3966, pruned_loss=0.1404, over 5660381.48 frames. ], batch size: 89, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:50:57,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.995e+02 1.837e+03 2.781e+03 3.764e+03 7.130e+03, threshold=5.562e+03, percent-clipped=14.0
+2023-03-03 14:51:03,282 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:51:03,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:51:07,792 INFO [train.py:968] (0/2) Epoch 7, batch 7450, giga_loss[loss=0.298, simple_loss=0.365, pruned_loss=0.1155, over 28914.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3906, pruned_loss=0.1365, over 5660523.68 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3651, pruned_loss=0.1086, over 5610877.49 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3946, pruned_loss=0.1404, over 5658975.66 frames. ], batch size: 145, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:51:52,819 INFO [train.py:968] (0/2) Epoch 7, batch 7500, giga_loss[loss=0.2935, simple_loss=0.3631, pruned_loss=0.112, over 28360.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3893, pruned_loss=0.1363, over 5659639.68 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3651, pruned_loss=0.1086, over 5603872.63 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3928, pruned_loss=0.1397, over 5665223.99 frames. ], batch size: 65, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:52:32,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.470e+02 1.384e+03 1.836e+03 2.367e+03 6.532e+03, threshold=3.671e+03, percent-clipped=2.0
+2023-03-03 14:52:44,406 INFO [train.py:968] (0/2) Epoch 7, batch 7550, giga_loss[loss=0.3003, simple_loss=0.3499, pruned_loss=0.1254, over 23819.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3889, pruned_loss=0.135, over 5654035.90 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3654, pruned_loss=0.1087, over 5613828.66 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3923, pruned_loss=0.1386, over 5651164.33 frames. ], batch size: 705, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:53:19,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:53:22,967 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:53:28,534 INFO [train.py:968] (0/2) Epoch 7, batch 7600, libri_loss[loss=0.3295, simple_loss=0.4009, pruned_loss=0.1291, over 29108.00 frames. ], tot_loss[loss=0.327, simple_loss=0.388, pruned_loss=0.133, over 5668025.85 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3655, pruned_loss=0.1087, over 5621599.72 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3914, pruned_loss=0.1367, over 5660012.60 frames. ], batch size: 101, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 14:53:47,837 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:54:01,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:54:05,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.908e+02 1.498e+03 1.819e+03 2.684e+03 7.869e+03, threshold=3.638e+03, percent-clipped=12.0
+2023-03-03 14:54:12,559 INFO [train.py:968] (0/2) Epoch 7, batch 7650, libri_loss[loss=0.2466, simple_loss=0.3258, pruned_loss=0.08376, over 29654.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3869, pruned_loss=0.132, over 5680934.40 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3652, pruned_loss=0.1085, over 5631749.45 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3908, pruned_loss=0.1361, over 5666860.45 frames. ], batch size: 73, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:54:13,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:54:25,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:54:47,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3127, 3.1384, 2.9798, 1.5183], device='cuda:0'), covar=tensor([0.0750, 0.0855, 0.0908, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0884, 0.0801, 0.0625], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 14:55:00,272 INFO [train.py:968] (0/2) Epoch 7, batch 7700, giga_loss[loss=0.3009, simple_loss=0.3673, pruned_loss=0.1172, over 28852.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3852, pruned_loss=0.131, over 5685881.61 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3651, pruned_loss=0.1084, over 5635886.36 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3886, pruned_loss=0.1346, over 5671789.24 frames. ], batch size: 186, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:55:10,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6745, 1.7535, 1.4234, 2.3255], device='cuda:0'), covar=tensor([0.2014, 0.2075, 0.2150, 0.1804], device='cuda:0'), in_proj_covar=tensor([0.1175, 0.0890, 0.1033, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 14:55:29,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:55:38,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.789e+02 1.457e+03 1.894e+03 2.362e+03 1.027e+04, threshold=3.788e+03, percent-clipped=8.0
+2023-03-03 14:55:39,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3997, 1.4859, 1.5198, 1.2341], device='cuda:0'), covar=tensor([0.1740, 0.2738, 0.1373, 0.1434], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0728, 0.0815, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 14:55:52,038 INFO [train.py:968] (0/2) Epoch 7, batch 7750, giga_loss[loss=0.2971, simple_loss=0.3663, pruned_loss=0.1139, over 28312.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3841, pruned_loss=0.1311, over 5671349.65 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3652, pruned_loss=0.1082, over 5641345.31 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3872, pruned_loss=0.1347, over 5656492.05 frames. ], batch size: 65, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:56:21,194 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:23,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:33,092 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281442.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 14:56:35,062 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:38,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:42,071 INFO [train.py:968] (0/2) Epoch 7, batch 7800, giga_loss[loss=0.3557, simple_loss=0.3908, pruned_loss=0.1603, over 26715.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3844, pruned_loss=0.1326, over 5674064.30 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3651, pruned_loss=0.1082, over 5646626.55 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3872, pruned_loss=0.1359, over 5658239.53 frames. ], batch size: 555, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:56:52,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:54,582 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:54,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:57:02,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:57:08,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:57:23,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:57:24,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.936e+02 1.615e+03 2.163e+03 2.881e+03 1.239e+04, threshold=4.326e+03, percent-clipped=9.0
+2023-03-03 14:57:24,603 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 14:57:31,548 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-03 14:57:34,084 INFO [train.py:968] (0/2) Epoch 7, batch 7850, giga_loss[loss=0.3551, simple_loss=0.3791, pruned_loss=0.1655, over 23615.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3846, pruned_loss=0.1342, over 5665251.57 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3651, pruned_loss=0.1081, over 5649254.75 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3871, pruned_loss=0.137, over 5650786.15 frames. ], batch size: 705, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:57:47,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4103, 3.9671, 1.5882, 1.4248], device='cuda:0'), covar=tensor([0.0852, 0.0297, 0.0782, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0488, 0.0314, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 14:57:58,882 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:58:01,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:58:20,453 INFO [train.py:968] (0/2) Epoch 7, batch 7900, giga_loss[loss=0.3135, simple_loss=0.3712, pruned_loss=0.1279, over 27959.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3824, pruned_loss=0.1328, over 5663513.56 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3648, pruned_loss=0.108, over 5652761.55 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3851, pruned_loss=0.1359, over 5648914.15 frames. ], batch size: 412, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:58:29,305 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:58:35,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:58:57,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.989e+03 2.702e+03 4.065e+03 8.644e+03, threshold=5.403e+03, percent-clipped=22.0
+2023-03-03 14:59:06,201 INFO [train.py:968] (0/2) Epoch 7, batch 7950, giga_loss[loss=0.353, simple_loss=0.4004, pruned_loss=0.1528, over 27719.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3815, pruned_loss=0.1318, over 5648490.32 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3653, pruned_loss=0.1084, over 5638326.58 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3837, pruned_loss=0.1347, over 5650899.92 frames. ], batch size: 472, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:59:19,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:59:21,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:59:46,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9964, 1.1699, 1.2326, 1.0510], device='cuda:0'), covar=tensor([0.0949, 0.0929, 0.1340, 0.1016], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0743, 0.0656, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 14:59:51,934 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:59:55,849 INFO [train.py:968] (0/2) Epoch 7, batch 8000, libri_loss[loss=0.302, simple_loss=0.3839, pruned_loss=0.1101, over 29268.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3836, pruned_loss=0.1329, over 5653446.98 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3658, pruned_loss=0.1086, over 5641699.49 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3853, pruned_loss=0.1356, over 5652341.35 frames. ], batch size: 94, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 15:00:35,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.734e+02 1.765e+03 2.224e+03 3.109e+03 1.135e+04, threshold=4.449e+03, percent-clipped=5.0
+2023-03-03 15:00:44,539 INFO [train.py:968] (0/2) Epoch 7, batch 8050, giga_loss[loss=0.3187, simple_loss=0.3839, pruned_loss=0.1268, over 28930.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3837, pruned_loss=0.1318, over 5661043.31 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3657, pruned_loss=0.1086, over 5643052.99 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3851, pruned_loss=0.134, over 5659122.32 frames. ], batch size: 213, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:00:52,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2601, 4.0922, 3.8547, 2.0019], device='cuda:0'), covar=tensor([0.0520, 0.0654, 0.0742, 0.1960], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0890, 0.0804, 0.0625], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 15:00:52,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4064, 2.1028, 1.5286, 0.5340], device='cuda:0'), covar=tensor([0.2780, 0.1422, 0.2105, 0.3457], device='cuda:0'), in_proj_covar=tensor([0.1441, 0.1362, 0.1401, 0.1192], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 15:01:30,202 INFO [train.py:968] (0/2) Epoch 7, batch 8100, giga_loss[loss=0.3321, simple_loss=0.3929, pruned_loss=0.1357, over 28744.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.383, pruned_loss=0.1306, over 5677436.78 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3652, pruned_loss=0.1083, over 5647495.68 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3851, pruned_loss=0.1332, over 5672485.45 frames. ], batch size: 119, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:02:06,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.858e+02 1.417e+03 1.957e+03 2.603e+03 8.058e+03, threshold=3.914e+03, percent-clipped=7.0
+2023-03-03 15:02:13,893 INFO [train.py:968] (0/2) Epoch 7, batch 8150, giga_loss[loss=0.3812, simple_loss=0.4263, pruned_loss=0.168, over 28936.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3827, pruned_loss=0.1305, over 5680533.45 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3646, pruned_loss=0.108, over 5657266.79 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3858, pruned_loss=0.1338, over 5668988.50 frames. ], batch size: 164, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:02:31,823 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281817.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:02:33,251 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:03:04,692 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:03:05,264 INFO [train.py:968] (0/2) Epoch 7, batch 8200, giga_loss[loss=0.34, simple_loss=0.3946, pruned_loss=0.1427, over 28899.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3853, pruned_loss=0.1333, over 5676542.47 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3641, pruned_loss=0.1078, over 5660854.38 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3884, pruned_loss=0.1365, over 5664479.67 frames. ], batch size: 186, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:03:25,783 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7015, 2.4237, 2.5140, 2.0657], device='cuda:0'), covar=tensor([0.0983, 0.1674, 0.1245, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0738, 0.0654, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 15:03:50,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.674e+02 1.589e+03 2.091e+03 2.876e+03 6.015e+03, threshold=4.182e+03, percent-clipped=6.0
+2023-03-03 15:03:58,217 INFO [train.py:968] (0/2) Epoch 7, batch 8250, giga_loss[loss=0.3126, simple_loss=0.3694, pruned_loss=0.1279, over 28760.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3878, pruned_loss=0.1373, over 5659270.45 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3642, pruned_loss=0.1077, over 5663011.50 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3908, pruned_loss=0.1406, over 5648115.15 frames. ], batch size: 60, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:04:35,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:04:50,903 INFO [train.py:968] (0/2) Epoch 7, batch 8300, libri_loss[loss=0.2916, simple_loss=0.3717, pruned_loss=0.1058, over 27680.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3888, pruned_loss=0.1387, over 5668182.31 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3645, pruned_loss=0.1078, over 5666494.28 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3915, pruned_loss=0.142, over 5656085.33 frames. ], batch size: 115, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:04:57,696 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281960.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:04:59,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281963.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:05:25,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1721, 1.2508, 4.2293, 3.3203], device='cuda:0'), covar=tensor([0.1593, 0.2348, 0.0362, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0590, 0.0550, 0.0791, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:05:28,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.798e+03 2.344e+03 3.462e+03 1.075e+04, threshold=4.689e+03, percent-clipped=18.0
+2023-03-03 15:05:28,931 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281992.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:05:36,897 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-282000.pt
+2023-03-03 15:05:37,173 INFO [train.py:968] (0/2) Epoch 7, batch 8350, giga_loss[loss=0.4122, simple_loss=0.444, pruned_loss=0.1901, over 27983.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3914, pruned_loss=0.1415, over 5665286.89 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3648, pruned_loss=0.1078, over 5673784.52 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.394, pruned_loss=0.1452, over 5649116.28 frames. ], batch size: 412, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:05:42,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9675, 1.1335, 3.7782, 3.0838], device='cuda:0'), covar=tensor([0.1728, 0.2429, 0.0438, 0.0724], device='cuda:0'), in_proj_covar=tensor([0.0591, 0.0550, 0.0791, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:06:23,762 INFO [train.py:968] (0/2) Epoch 7, batch 8400, libri_loss[loss=0.2372, simple_loss=0.3121, pruned_loss=0.08111, over 29620.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3888, pruned_loss=0.1393, over 5674364.61 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3644, pruned_loss=0.1075, over 5678378.95 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3918, pruned_loss=0.1431, over 5657151.51 frames. ], batch size: 69, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 15:06:33,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6171, 1.7341, 1.3635, 1.2745], device='cuda:0'), covar=tensor([0.1479, 0.1204, 0.1009, 0.1314], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1399, 0.1361, 0.1477], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 15:06:42,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4020, 1.4802, 1.4835, 1.3988], device='cuda:0'), covar=tensor([0.0985, 0.1292, 0.1424, 0.1271], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0740, 0.0654, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 15:06:48,261 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:06:50,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:06:57,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9053, 1.1087, 0.9269, 0.8041], device='cuda:0'), covar=tensor([0.1343, 0.1324, 0.0826, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.1546, 0.1385, 0.1345, 0.1460], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 15:06:57,610 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.928e+02 1.541e+03 1.987e+03 2.954e+03 7.404e+03, threshold=3.974e+03, percent-clipped=8.0
+2023-03-03 15:07:04,687 INFO [train.py:968] (0/2) Epoch 7, batch 8450, libri_loss[loss=0.2936, simple_loss=0.3692, pruned_loss=0.109, over 29751.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3874, pruned_loss=0.1373, over 5685818.56 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3645, pruned_loss=0.1077, over 5683515.13 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3906, pruned_loss=0.1414, over 5666931.64 frames. ], batch size: 87, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:07:15,014 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:07:50,156 INFO [train.py:968] (0/2) Epoch 7, batch 8500, giga_loss[loss=0.349, simple_loss=0.4171, pruned_loss=0.1405, over 28647.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3848, pruned_loss=0.1334, over 5692236.25 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3645, pruned_loss=0.1077, over 5687101.40 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3877, pruned_loss=0.1371, over 5674242.53 frames. ], batch size: 71, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:08:25,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.749e+02 1.531e+03 2.196e+03 3.400e+03 1.321e+04, threshold=4.392e+03, percent-clipped=16.0
+2023-03-03 15:08:28,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:08:32,222 INFO [train.py:968] (0/2) Epoch 7, batch 8550, giga_loss[loss=0.305, simple_loss=0.3657, pruned_loss=0.1222, over 28617.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3829, pruned_loss=0.1321, over 5679655.72 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3654, pruned_loss=0.1082, over 5683885.77 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3851, pruned_loss=0.1355, over 5668254.19 frames. ], batch size: 92, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:08:37,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0199, 1.3312, 3.9032, 3.1504], device='cuda:0'), covar=tensor([0.1781, 0.2241, 0.0399, 0.0746], device='cuda:0'), in_proj_covar=tensor([0.0599, 0.0552, 0.0794, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:08:56,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:09:14,679 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:09:19,827 INFO [train.py:968] (0/2) Epoch 7, batch 8600, giga_loss[loss=0.3627, simple_loss=0.4103, pruned_loss=0.1576, over 28950.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3806, pruned_loss=0.131, over 5669065.87 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3654, pruned_loss=0.1082, over 5677988.79 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3826, pruned_loss=0.134, over 5666088.75 frames. ], batch size: 213, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:09:37,799 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 15:09:59,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4715, 1.6951, 1.3821, 1.0890], device='cuda:0'), covar=tensor([0.1412, 0.1093, 0.0809, 0.1230], device='cuda:0'), in_proj_covar=tensor([0.1550, 0.1384, 0.1354, 0.1465], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 15:10:04,457 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.210e+02 1.429e+03 1.784e+03 2.473e+03 8.358e+03, threshold=3.568e+03, percent-clipped=4.0
+2023-03-03 15:10:11,646 INFO [train.py:968] (0/2) Epoch 7, batch 8650, giga_loss[loss=0.3151, simple_loss=0.3739, pruned_loss=0.1281, over 28340.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3805, pruned_loss=0.1317, over 5666871.75 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3655, pruned_loss=0.1082, over 5680262.24 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3822, pruned_loss=0.1343, over 5662171.91 frames. ], batch size: 65, lr: 4.72e-03, grad_scale: 2.0
+2023-03-03 15:10:13,970 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282303.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:10:49,960 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:10:50,454 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:10:50,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9239, 4.8708, 2.0598, 1.9567], device='cuda:0'), covar=tensor([0.0797, 0.0264, 0.0748, 0.1155], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0492, 0.0316, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 15:10:53,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:11:03,708 INFO [train.py:968] (0/2) Epoch 7, batch 8700, giga_loss[loss=0.4319, simple_loss=0.4542, pruned_loss=0.2048, over 26578.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3851, pruned_loss=0.1347, over 5666239.11 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3661, pruned_loss=0.1087, over 5676180.90 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3861, pruned_loss=0.1369, over 5665816.83 frames. ], batch size: 555, lr: 4.72e-03, grad_scale: 2.0
+2023-03-03 15:11:05,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6664, 2.2129, 1.9370, 1.5848], device='cuda:0'), covar=tensor([0.1702, 0.2054, 0.1373, 0.1625], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0723, 0.0810, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 15:11:19,699 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:11:23,833 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:11:24,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:11:47,721 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.799e+02 1.483e+03 1.815e+03 2.202e+03 4.749e+03, threshold=3.631e+03, percent-clipped=4.0
+2023-03-03 15:11:52,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3943, 1.5981, 1.2897, 1.5733], device='cuda:0'), covar=tensor([0.2469, 0.2328, 0.2465, 0.2054], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.0906, 0.1040, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:11:53,929 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:11:54,359 INFO [train.py:968] (0/2) Epoch 7, batch 8750, giga_loss[loss=0.3256, simple_loss=0.4008, pruned_loss=0.1252, over 28528.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3892, pruned_loss=0.135, over 5666461.00 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3662, pruned_loss=0.1087, over 5678150.13 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3902, pruned_loss=0.1369, over 5664230.45 frames. ], batch size: 336, lr: 4.72e-03, grad_scale: 2.0
+2023-03-03 15:12:42,050 INFO [train.py:968] (0/2) Epoch 7, batch 8800, giga_loss[loss=0.4093, simple_loss=0.4495, pruned_loss=0.1845, over 28646.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3901, pruned_loss=0.1342, over 5675870.84 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3656, pruned_loss=0.1085, over 5684667.27 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3921, pruned_loss=0.1367, over 5667489.78 frames. ], batch size: 336, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:13:18,230 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:13:21,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2085, 1.4873, 1.1405, 0.9257], device='cuda:0'), covar=tensor([0.1379, 0.1225, 0.0906, 0.1176], device='cuda:0'), in_proj_covar=tensor([0.1550, 0.1387, 0.1353, 0.1453], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 15:13:22,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.482e+02 1.541e+03 2.154e+03 3.258e+03 8.080e+03, threshold=4.307e+03, percent-clipped=22.0
+2023-03-03 15:13:27,068 INFO [train.py:968] (0/2) Epoch 7, batch 8850, giga_loss[loss=0.3108, simple_loss=0.3757, pruned_loss=0.123, over 28687.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3933, pruned_loss=0.1368, over 5678953.77 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3658, pruned_loss=0.1086, over 5688101.95 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3951, pruned_loss=0.139, over 5669253.33 frames. ], batch size: 262, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:14:10,772 INFO [train.py:968] (0/2) Epoch 7, batch 8900, giga_loss[loss=0.3826, simple_loss=0.4289, pruned_loss=0.1682, over 28902.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3931, pruned_loss=0.1369, over 5690220.23 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3659, pruned_loss=0.1087, over 5693456.65 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3952, pruned_loss=0.1394, over 5677720.70 frames. ], batch size: 227, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:14:47,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.360e+02 1.470e+03 2.038e+03 2.575e+03 4.253e+03, threshold=4.076e+03, percent-clipped=1.0
+2023-03-03 15:14:53,138 INFO [train.py:968] (0/2) Epoch 7, batch 8950, giga_loss[loss=0.2472, simple_loss=0.3268, pruned_loss=0.08378, over 28632.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3911, pruned_loss=0.1363, over 5692900.50 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3653, pruned_loss=0.1083, over 5699787.92 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3945, pruned_loss=0.1398, over 5676874.93 frames. ], batch size: 60, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:15:12,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:15:19,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2874, 4.1071, 3.8668, 1.9955], device='cuda:0'), covar=tensor([0.0520, 0.0649, 0.0716, 0.1893], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0890, 0.0797, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 15:15:42,799 INFO [train.py:968] (0/2) Epoch 7, batch 9000, giga_loss[loss=0.3251, simple_loss=0.3864, pruned_loss=0.1319, over 28346.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3893, pruned_loss=0.1358, over 5699132.47 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3646, pruned_loss=0.108, over 5704825.71 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3933, pruned_loss=0.1396, over 5681444.59 frames. ], batch size: 368, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:15:42,803 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 15:15:51,394 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2269, simple_loss=0.3315, pruned_loss=0.06111, over 944034.00 frames. 
+2023-03-03 15:15:51,395 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19587MB
+2023-03-03 15:16:17,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282678.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:16:19,075 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.47 vs. limit=5.0
+2023-03-03 15:16:32,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.519e+02 1.511e+03 1.893e+03 2.594e+03 6.616e+03, threshold=3.786e+03, percent-clipped=11.0
+2023-03-03 15:16:37,294 INFO [train.py:968] (0/2) Epoch 7, batch 9050, giga_loss[loss=0.2967, simple_loss=0.3565, pruned_loss=0.1185, over 28605.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3875, pruned_loss=0.1352, over 5692637.49 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3647, pruned_loss=0.108, over 5705128.41 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.391, pruned_loss=0.1388, over 5678064.20 frames. ], batch size: 85, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:16:41,910 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 15:16:52,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:17:25,969 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-03 15:17:26,177 INFO [train.py:968] (0/2) Epoch 7, batch 9100, giga_loss[loss=0.3268, simple_loss=0.3937, pruned_loss=0.13, over 28696.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3862, pruned_loss=0.1349, over 5689498.68 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3651, pruned_loss=0.1082, over 5709689.15 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3892, pruned_loss=0.1383, over 5673171.63 frames. ], batch size: 262, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:17:39,280 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:17:42,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:18:02,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4248, 1.5527, 1.3140, 1.3928], device='cuda:0'), covar=tensor([0.0745, 0.0308, 0.0312, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0066, 0.0048, 0.0043, 0.0073], device='cuda:0')
+2023-03-03 15:18:09,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.591e+02 1.432e+03 1.858e+03 2.615e+03 6.698e+03, threshold=3.716e+03, percent-clipped=7.0
+2023-03-03 15:18:10,111 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:18:14,659 INFO [train.py:968] (0/2) Epoch 7, batch 9150, giga_loss[loss=0.3144, simple_loss=0.3742, pruned_loss=0.1273, over 28884.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3847, pruned_loss=0.1339, over 5693761.15 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3648, pruned_loss=0.108, over 5714284.70 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3878, pruned_loss=0.1373, over 5676292.18 frames. ], batch size: 145, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:18:37,139 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282821.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:18:41,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282824.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:18:44,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4115, 1.4555, 1.1319, 1.1269], device='cuda:0'), covar=tensor([0.0614, 0.0445, 0.0853, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0448, 0.0502, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 15:19:04,458 INFO [train.py:968] (0/2) Epoch 7, batch 9200, libri_loss[loss=0.2931, simple_loss=0.3706, pruned_loss=0.1078, over 27835.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3853, pruned_loss=0.1346, over 5685052.99 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.365, pruned_loss=0.1081, over 5716847.56 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3882, pruned_loss=0.138, over 5667943.95 frames. ], batch size: 116, lr: 4.72e-03, grad_scale: 8.0
+2023-03-03 15:19:07,195 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282853.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:19:09,218 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:19:12,835 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:19:17,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:19:34,722 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:19:41,084 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:19:45,637 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-03 15:19:45,808 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.499e+02 1.717e+03 2.276e+03 3.078e+03 1.183e+04, threshold=4.551e+03, percent-clipped=18.0
+2023-03-03 15:19:50,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2031, 3.0175, 2.8421, 1.4732], device='cuda:0'), covar=tensor([0.0835, 0.0874, 0.0904, 0.2184], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0892, 0.0803, 0.0624], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 15:19:52,622 INFO [train.py:968] (0/2) Epoch 7, batch 9250, giga_loss[loss=0.3002, simple_loss=0.3573, pruned_loss=0.1216, over 28760.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.384, pruned_loss=0.1341, over 5684422.94 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.365, pruned_loss=0.1081, over 5717016.58 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3867, pruned_loss=0.1374, over 5670226.27 frames. ], batch size: 119, lr: 4.72e-03, grad_scale: 8.0
+2023-03-03 15:20:40,364 INFO [train.py:968] (0/2) Epoch 7, batch 9300, giga_loss[loss=0.3088, simple_loss=0.3743, pruned_loss=0.1217, over 28862.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3828, pruned_loss=0.1334, over 5688535.81 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.365, pruned_loss=0.108, over 5717773.20 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3853, pruned_loss=0.1366, over 5676013.39 frames. ], batch size: 174, lr: 4.72e-03, grad_scale: 8.0
+2023-03-03 15:20:43,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5015, 1.6964, 1.4225, 1.6337], device='cuda:0'), covar=tensor([0.1920, 0.1785, 0.1811, 0.1806], device='cuda:0'), in_proj_covar=tensor([0.1181, 0.0901, 0.1037, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:21:23,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.428e+02 1.573e+03 2.074e+03 2.796e+03 7.825e+03, threshold=4.147e+03, percent-clipped=4.0
+2023-03-03 15:21:30,323 INFO [train.py:968] (0/2) Epoch 7, batch 9350, giga_loss[loss=0.342, simple_loss=0.3978, pruned_loss=0.1431, over 28766.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3861, pruned_loss=0.1356, over 5665823.43 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3652, pruned_loss=0.1082, over 5701104.00 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3884, pruned_loss=0.1387, over 5669529.88 frames. ], batch size: 243, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:21:35,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3047, 1.3834, 3.4791, 3.2288], device='cuda:0'), covar=tensor([0.1221, 0.2042, 0.0406, 0.0780], device='cuda:0'), in_proj_covar=tensor([0.0590, 0.0549, 0.0785, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:21:37,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:21:40,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:22:07,314 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:22:17,083 INFO [train.py:968] (0/2) Epoch 7, batch 9400, giga_loss[loss=0.3875, simple_loss=0.4275, pruned_loss=0.1737, over 28208.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3887, pruned_loss=0.137, over 5669725.06 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3655, pruned_loss=0.1084, over 5703135.22 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3909, pruned_loss=0.1401, over 5669661.91 frames. ], batch size: 368, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:22:17,991 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:22:37,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:22:39,082 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 15:22:55,536 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283091.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:22:59,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.998e+02 1.632e+03 2.101e+03 3.143e+03 7.623e+03, threshold=4.201e+03, percent-clipped=12.0
+2023-03-03 15:23:03,389 INFO [train.py:968] (0/2) Epoch 7, batch 9450, giga_loss[loss=0.3391, simple_loss=0.3863, pruned_loss=0.1459, over 28616.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3887, pruned_loss=0.1377, over 5658238.61 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3659, pruned_loss=0.1088, over 5687854.29 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3904, pruned_loss=0.1403, over 5669826.21 frames. ], batch size: 92, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:23:52,358 INFO [train.py:968] (0/2) Epoch 7, batch 9500, giga_loss[loss=0.2933, simple_loss=0.3891, pruned_loss=0.09872, over 28714.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3893, pruned_loss=0.1362, over 5665576.38 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3656, pruned_loss=0.1088, over 5692985.32 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3915, pruned_loss=0.1389, over 5669765.26 frames. ], batch size: 60, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:24:32,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.405e+02 1.391e+03 1.908e+03 2.739e+03 9.300e+03, threshold=3.815e+03, percent-clipped=8.0
+2023-03-03 15:24:34,952 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:24:36,075 INFO [train.py:968] (0/2) Epoch 7, batch 9550, giga_loss[loss=0.2987, simple_loss=0.3793, pruned_loss=0.109, over 28562.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3906, pruned_loss=0.1345, over 5674995.60 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3662, pruned_loss=0.1091, over 5696187.57 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3923, pruned_loss=0.1369, over 5675113.09 frames. ], batch size: 78, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:25:17,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4306, 1.6129, 1.4922, 1.5316], device='cuda:0'), covar=tensor([0.0746, 0.0298, 0.0299, 0.0820], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0117, 0.0120, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0066, 0.0047, 0.0043, 0.0072], device='cuda:0')
+2023-03-03 15:25:25,985 INFO [train.py:968] (0/2) Epoch 7, batch 9600, giga_loss[loss=0.424, simple_loss=0.4365, pruned_loss=0.2058, over 23496.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3927, pruned_loss=0.1354, over 5672008.15 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3659, pruned_loss=0.1089, over 5700441.62 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3949, pruned_loss=0.1381, over 5667627.06 frames. ], batch size: 705, lr: 4.72e-03, grad_scale: 8.0
+2023-03-03 15:25:31,924 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:25:44,799 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-03 15:26:10,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.774e+02 1.411e+03 1.870e+03 2.544e+03 7.338e+03, threshold=3.741e+03, percent-clipped=8.0
+2023-03-03 15:26:13,079 INFO [train.py:968] (0/2) Epoch 7, batch 9650, giga_loss[loss=0.3169, simple_loss=0.3846, pruned_loss=0.1246, over 28888.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3953, pruned_loss=0.1382, over 5676305.70 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3662, pruned_loss=0.1092, over 5700879.17 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3975, pruned_loss=0.1406, over 5671913.20 frames. ], batch size: 119, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:26:31,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6429, 1.7143, 1.4314, 2.0243], device='cuda:0'), covar=tensor([0.1925, 0.1971, 0.2057, 0.1876], device='cuda:0'), in_proj_covar=tensor([0.1179, 0.0900, 0.1038, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:26:32,771 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-03 15:26:40,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 15:26:58,136 INFO [train.py:968] (0/2) Epoch 7, batch 9700, giga_loss[loss=0.3639, simple_loss=0.4127, pruned_loss=0.1575, over 28698.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.396, pruned_loss=0.1399, over 5679565.75 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.366, pruned_loss=0.1091, over 5706174.98 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3986, pruned_loss=0.1426, over 5670720.32 frames. ], batch size: 242, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:27:44,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.664e+03 2.149e+03 3.062e+03 8.676e+03, threshold=4.297e+03, percent-clipped=14.0
+2023-03-03 15:27:47,711 INFO [train.py:968] (0/2) Epoch 7, batch 9750, giga_loss[loss=0.3688, simple_loss=0.3926, pruned_loss=0.1725, over 23832.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3953, pruned_loss=0.1405, over 5665842.52 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3656, pruned_loss=0.1088, over 5707060.92 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3984, pruned_loss=0.1437, over 5657286.81 frames. ], batch size: 705, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:27:48,949 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:27:51,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:28:07,638 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2765, 1.6515, 1.5952, 1.2515], device='cuda:0'), covar=tensor([0.1437, 0.2131, 0.1206, 0.1371], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0734, 0.0818, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 15:28:12,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:28:19,688 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:28:31,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:28:34,602 INFO [train.py:968] (0/2) Epoch 7, batch 9800, giga_loss[loss=0.3227, simple_loss=0.3919, pruned_loss=0.1267, over 28476.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3935, pruned_loss=0.139, over 5659631.44 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3654, pruned_loss=0.1087, over 5710035.59 frames. ], giga_tot_loss[loss=0.3406, simple_loss=0.3967, pruned_loss=0.1422, over 5649340.02 frames. ], batch size: 60, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:28:49,293 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283466.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:29:14,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.434e+02 1.459e+03 1.852e+03 2.264e+03 8.475e+03, threshold=3.705e+03, percent-clipped=4.0
+2023-03-03 15:29:16,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2545, 1.8497, 1.3057, 0.4820], device='cuda:0'), covar=tensor([0.2302, 0.1264, 0.2161, 0.2984], device='cuda:0'), in_proj_covar=tensor([0.1447, 0.1377, 0.1414, 0.1187], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 15:29:17,435 INFO [train.py:968] (0/2) Epoch 7, batch 9850, libri_loss[loss=0.2962, simple_loss=0.3752, pruned_loss=0.1087, over 25824.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3924, pruned_loss=0.1363, over 5660034.46 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3656, pruned_loss=0.1089, over 5704068.81 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3957, pruned_loss=0.1397, over 5655734.65 frames. ], batch size: 136, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:29:25,046 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:29:49,385 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-03 15:29:49,706 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:29:59,991 INFO [train.py:968] (0/2) Epoch 7, batch 9900, giga_loss[loss=0.2701, simple_loss=0.3502, pruned_loss=0.09506, over 28432.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3925, pruned_loss=0.1347, over 5668182.40 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3659, pruned_loss=0.109, over 5706650.73 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3956, pruned_loss=0.1381, over 5661174.33 frames. ], batch size: 71, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:30:11,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3462, 1.5762, 1.2639, 1.5094], device='cuda:0'), covar=tensor([0.2240, 0.2143, 0.2301, 0.1986], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.0902, 0.1041, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:30:17,474 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:18,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2669, 4.0632, 3.8453, 1.9562], device='cuda:0'), covar=tensor([0.0561, 0.0731, 0.0810, 0.1935], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0897, 0.0806, 0.0621], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 15:30:20,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:21,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:36,994 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:42,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:46,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.125e+02 1.384e+03 1.705e+03 2.502e+03 5.924e+03, threshold=3.411e+03, percent-clipped=10.0
+2023-03-03 15:30:49,718 INFO [train.py:968] (0/2) Epoch 7, batch 9950, giga_loss[loss=0.3175, simple_loss=0.3855, pruned_loss=0.1247, over 28839.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3922, pruned_loss=0.1345, over 5671868.40 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3654, pruned_loss=0.1087, over 5710524.98 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3957, pruned_loss=0.1379, over 5662259.87 frames. ], batch size: 199, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:30:50,629 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:57,262 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283609.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:31:00,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283612.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:31:09,854 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:31:29,536 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283641.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:31:31,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2753, 3.3524, 1.3336, 1.5645], device='cuda:0'), covar=tensor([0.0942, 0.0227, 0.0911, 0.1300], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0493, 0.0317, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 15:31:36,417 INFO [train.py:968] (0/2) Epoch 7, batch 10000, giga_loss[loss=0.3112, simple_loss=0.3765, pruned_loss=0.123, over 28903.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3943, pruned_loss=0.1367, over 5667825.76 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.366, pruned_loss=0.1091, over 5718084.98 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3976, pruned_loss=0.1402, over 5651457.56 frames. ], batch size: 106, lr: 4.71e-03, grad_scale: 8.0
+2023-03-03 15:32:01,251 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:32:11,800 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-03 15:32:20,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.489e+02 1.643e+03 2.226e+03 3.389e+03 8.097e+03, threshold=4.453e+03, percent-clipped=24.0
+2023-03-03 15:32:22,810 INFO [train.py:968] (0/2) Epoch 7, batch 10050, giga_loss[loss=0.318, simple_loss=0.3812, pruned_loss=0.1274, over 28957.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3921, pruned_loss=0.1359, over 5682826.36 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3654, pruned_loss=0.1088, over 5720899.46 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3957, pruned_loss=0.1394, over 5666486.22 frames. ], batch size: 164, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:32:41,427 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:32:43,105 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-03 15:32:44,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:33:13,145 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:33:14,133 INFO [train.py:968] (0/2) Epoch 7, batch 10100, giga_loss[loss=0.3313, simple_loss=0.3892, pruned_loss=0.1367, over 28711.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3918, pruned_loss=0.137, over 5659420.01 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.366, pruned_loss=0.1091, over 5712221.61 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3948, pruned_loss=0.1402, over 5653278.17 frames. ], batch size: 284, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:33:41,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9292, 0.9790, 3.7698, 3.1130], device='cuda:0'), covar=tensor([0.1688, 0.2491, 0.0415, 0.0973], device='cuda:0'), in_proj_covar=tensor([0.0596, 0.0562, 0.0800, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:34:02,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.156e+02 1.559e+03 2.049e+03 3.148e+03 8.057e+03, threshold=4.098e+03, percent-clipped=8.0
+2023-03-03 15:34:04,416 INFO [train.py:968] (0/2) Epoch 7, batch 10150, giga_loss[loss=0.3077, simple_loss=0.3694, pruned_loss=0.123, over 28644.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3896, pruned_loss=0.1361, over 5659288.05 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3659, pruned_loss=0.109, over 5705297.47 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3924, pruned_loss=0.139, over 5659410.53 frames. ], batch size: 262, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:34:57,721 INFO [train.py:968] (0/2) Epoch 7, batch 10200, giga_loss[loss=0.3287, simple_loss=0.3877, pruned_loss=0.1348, over 29048.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.388, pruned_loss=0.1361, over 5665018.00 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3659, pruned_loss=0.109, over 5708266.87 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3905, pruned_loss=0.1388, over 5662143.70 frames. ], batch size: 136, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:35:06,644 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 15:35:26,528 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0047, 1.1358, 1.0021, 1.2225], device='cuda:0'), covar=tensor([0.0833, 0.0298, 0.0309, 0.1021], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0066, 0.0048, 0.0043, 0.0073], device='cuda:0')
+2023-03-03 15:35:27,293 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:35:40,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.909e+02 1.549e+03 1.940e+03 2.872e+03 1.061e+04, threshold=3.879e+03, percent-clipped=9.0
+2023-03-03 15:35:41,522 INFO [train.py:968] (0/2) Epoch 7, batch 10250, giga_loss[loss=0.3286, simple_loss=0.3864, pruned_loss=0.1354, over 28984.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3885, pruned_loss=0.1365, over 5669847.31 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3665, pruned_loss=0.1091, over 5713265.08 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3906, pruned_loss=0.1395, over 5661688.76 frames. ], batch size: 106, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:35:55,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:36:11,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2476, 1.7211, 1.6209, 1.2362], device='cuda:0'), covar=tensor([0.1641, 0.2140, 0.1281, 0.1441], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0733, 0.0818, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 15:36:18,812 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1043, 1.0489, 3.9384, 3.2781], device='cuda:0'), covar=tensor([0.1593, 0.2565, 0.0376, 0.0712], device='cuda:0'), in_proj_covar=tensor([0.0598, 0.0562, 0.0800, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:36:19,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283937.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:36:19,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2970, 1.6366, 1.3879, 1.5349], device='cuda:0'), covar=tensor([0.0721, 0.0292, 0.0295, 0.0742], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0066, 0.0048, 0.0043, 0.0073], device='cuda:0')
+2023-03-03 15:36:30,504 INFO [train.py:968] (0/2) Epoch 7, batch 10300, giga_loss[loss=0.2699, simple_loss=0.3523, pruned_loss=0.09375, over 29077.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3854, pruned_loss=0.1338, over 5659619.71 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3664, pruned_loss=0.1091, over 5706703.34 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3875, pruned_loss=0.1365, over 5658665.80 frames. ], batch size: 155, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:37:19,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.193e+02 1.541e+03 2.036e+03 2.906e+03 9.473e+03, threshold=4.072e+03, percent-clipped=10.0
+2023-03-03 15:37:21,441 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-284000.pt
+2023-03-03 15:37:21,733 INFO [train.py:968] (0/2) Epoch 7, batch 10350, giga_loss[loss=0.3173, simple_loss=0.3811, pruned_loss=0.1267, over 28217.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3822, pruned_loss=0.1305, over 5660019.57 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3665, pruned_loss=0.1091, over 5707700.82 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3838, pruned_loss=0.1327, over 5658132.09 frames. ], batch size: 368, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:37:46,349 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:37:50,835 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:09,119 INFO [train.py:968] (0/2) Epoch 7, batch 10400, giga_loss[loss=0.2721, simple_loss=0.3529, pruned_loss=0.0956, over 28884.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3814, pruned_loss=0.1292, over 5649162.72 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3668, pruned_loss=0.1091, over 5697194.34 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.383, pruned_loss=0.1317, over 5655615.38 frames. ], batch size: 174, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:38:10,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:13,210 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:16,524 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:16,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:44,226 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:54,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.549e+02 1.588e+03 2.065e+03 3.067e+03 9.574e+03, threshold=4.130e+03, percent-clipped=10.0
+2023-03-03 15:38:57,188 INFO [train.py:968] (0/2) Epoch 7, batch 10450, giga_loss[loss=0.3129, simple_loss=0.3739, pruned_loss=0.1259, over 28841.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3793, pruned_loss=0.1285, over 5663700.51 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3667, pruned_loss=0.1089, over 5704007.40 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.381, pruned_loss=0.1313, over 5661540.37 frames. ], batch size: 284, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:39:02,180 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2111, 1.5772, 1.1626, 0.4009], device='cuda:0'), covar=tensor([0.1544, 0.1013, 0.1558, 0.2547], device='cuda:0'), in_proj_covar=tensor([0.1462, 0.1381, 0.1423, 0.1190], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 15:39:44,496 INFO [train.py:968] (0/2) Epoch 7, batch 10500, giga_loss[loss=0.3286, simple_loss=0.3872, pruned_loss=0.135, over 29051.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3757, pruned_loss=0.1263, over 5666607.10 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3666, pruned_loss=0.1087, over 5709261.82 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3776, pruned_loss=0.1294, over 5658459.81 frames. ], batch size: 128, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:40:26,333 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:40:28,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:40:29,326 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.491e+02 1.442e+03 2.005e+03 3.044e+03 1.130e+04, threshold=4.010e+03, percent-clipped=10.0
+2023-03-03 15:40:30,052 INFO [train.py:968] (0/2) Epoch 7, batch 10550, giga_loss[loss=0.3658, simple_loss=0.4185, pruned_loss=0.1565, over 28892.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3794, pruned_loss=0.1288, over 5664791.15 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3672, pruned_loss=0.109, over 5702705.55 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3807, pruned_loss=0.1315, over 5663111.88 frames. ], batch size: 186, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:40:52,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:41:00,367 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=284235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:41:12,858 INFO [train.py:968] (0/2) Epoch 7, batch 10600, giga_loss[loss=0.3123, simple_loss=0.3795, pruned_loss=0.1226, over 28774.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.381, pruned_loss=0.1293, over 5675678.43 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3669, pruned_loss=0.1088, over 5710731.40 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3831, pruned_loss=0.1327, over 5665237.60 frames. ], batch size: 284, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:41:42,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2977, 2.9320, 1.4440, 1.2932], device='cuda:0'), covar=tensor([0.0960, 0.0288, 0.0855, 0.1442], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0496, 0.0319, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 15:41:47,889 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-03 15:41:52,921 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.976e+02 1.525e+03 1.800e+03 2.363e+03 4.947e+03, threshold=3.601e+03, percent-clipped=2.0
+2023-03-03 15:41:53,580 INFO [train.py:968] (0/2) Epoch 7, batch 10650, giga_loss[loss=0.2876, simple_loss=0.3624, pruned_loss=0.1064, over 28955.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.381, pruned_loss=0.129, over 5665508.95 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3668, pruned_loss=0.1087, over 5718046.15 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3834, pruned_loss=0.1329, over 5648226.66 frames. ], batch size: 164, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:42:07,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284312.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:42:36,731 INFO [train.py:968] (0/2) Epoch 7, batch 10700, giga_loss[loss=0.4196, simple_loss=0.436, pruned_loss=0.2016, over 26474.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3809, pruned_loss=0.1294, over 5653104.93 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3672, pruned_loss=0.1092, over 5724445.16 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3833, pruned_loss=0.1332, over 5629534.99 frames. ], batch size: 555, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:42:45,588 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=284359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:42:54,555 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4397, 1.9310, 1.3927, 0.7854], device='cuda:0'), covar=tensor([0.2802, 0.1347, 0.1859, 0.2998], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1358, 0.1399, 0.1177], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 15:43:21,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.157e+02 1.404e+03 2.209e+03 3.419e+03 1.052e+04, threshold=4.417e+03, percent-clipped=20.0
+2023-03-03 15:43:22,966 INFO [train.py:968] (0/2) Epoch 7, batch 10750, giga_loss[loss=0.3979, simple_loss=0.4299, pruned_loss=0.183, over 27931.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3803, pruned_loss=0.1293, over 5658311.54 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3665, pruned_loss=0.1088, over 5727762.35 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3831, pruned_loss=0.1332, over 5635075.94 frames. ], batch size: 412, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:44:11,332 INFO [train.py:968] (0/2) Epoch 7, batch 10800, giga_loss[loss=0.3146, simple_loss=0.3809, pruned_loss=0.1242, over 28705.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3821, pruned_loss=0.1309, over 5639074.53 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3664, pruned_loss=0.1087, over 5714561.31 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3848, pruned_loss=0.1346, over 5629840.79 frames. ], batch size: 262, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:44:16,159 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284455.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:44:20,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284458.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:44:33,643 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 15:44:46,191 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284487.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:44:55,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.411e+02 1.476e+03 1.800e+03 2.287e+03 7.053e+03, threshold=3.600e+03, percent-clipped=2.0
+2023-03-03 15:44:56,777 INFO [train.py:968] (0/2) Epoch 7, batch 10850, giga_loss[loss=0.3824, simple_loss=0.4023, pruned_loss=0.1812, over 23387.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3839, pruned_loss=0.1318, over 5640809.59 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3667, pruned_loss=0.1088, over 5712440.89 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3865, pruned_loss=0.1355, over 5633288.00 frames. ], batch size: 705, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:45:41,581 INFO [train.py:968] (0/2) Epoch 7, batch 10900, giga_loss[loss=0.3918, simple_loss=0.4184, pruned_loss=0.1826, over 26563.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3844, pruned_loss=0.1324, over 5649676.55 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3663, pruned_loss=0.1087, over 5714502.53 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3872, pruned_loss=0.136, over 5640349.77 frames. ], batch size: 555, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:46:32,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.024e+02 1.471e+03 1.827e+03 3.045e+03 1.216e+04, threshold=3.654e+03, percent-clipped=16.0
+2023-03-03 15:46:33,246 INFO [train.py:968] (0/2) Epoch 7, batch 10950, giga_loss[loss=0.3392, simple_loss=0.3987, pruned_loss=0.1398, over 28958.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3858, pruned_loss=0.1341, over 5655698.03 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3663, pruned_loss=0.1086, over 5716544.10 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3881, pruned_loss=0.1373, over 5646062.51 frames. ], batch size: 213, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:46:45,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:46:48,588 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6127, 1.7524, 1.3288, 1.3442], device='cuda:0'), covar=tensor([0.1227, 0.1106, 0.0934, 0.1085], device='cuda:0'), in_proj_covar=tensor([0.1533, 0.1389, 0.1333, 0.1431], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 15:46:58,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4749, 1.8561, 1.7040, 1.3624], device='cuda:0'), covar=tensor([0.1823, 0.2275, 0.1410, 0.1570], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0739, 0.0823, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 15:47:27,207 INFO [train.py:968] (0/2) Epoch 7, batch 11000, giga_loss[loss=0.3571, simple_loss=0.4082, pruned_loss=0.153, over 28318.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3884, pruned_loss=0.1349, over 5654700.05 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3663, pruned_loss=0.1086, over 5716594.97 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3905, pruned_loss=0.1377, over 5646525.13 frames. ], batch size: 368, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:48:18,020 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 15:48:18,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.815e+02 1.672e+03 2.132e+03 3.130e+03 5.981e+03, threshold=4.265e+03, percent-clipped=19.0
+2023-03-03 15:48:20,598 INFO [train.py:968] (0/2) Epoch 7, batch 11050, giga_loss[loss=0.3499, simple_loss=0.3769, pruned_loss=0.1614, over 23415.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3876, pruned_loss=0.1348, over 5646199.75 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3663, pruned_loss=0.1087, over 5718268.05 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3894, pruned_loss=0.137, over 5637807.25 frames. ], batch size: 705, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:48:55,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:49:09,531 INFO [train.py:968] (0/2) Epoch 7, batch 11100, giga_loss[loss=0.2868, simple_loss=0.3554, pruned_loss=0.1091, over 28920.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3855, pruned_loss=0.1337, over 5664305.76 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3658, pruned_loss=0.1084, over 5722788.26 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3879, pruned_loss=0.1364, over 5652059.27 frames. ], batch size: 199, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:49:13,938 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:49:18,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:49:49,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:50:03,848 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 15:50:05,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.750e+02 1.614e+03 2.263e+03 3.148e+03 1.084e+04, threshold=4.525e+03, percent-clipped=13.0
+2023-03-03 15:50:06,793 INFO [train.py:968] (0/2) Epoch 7, batch 11150, giga_loss[loss=0.3279, simple_loss=0.3842, pruned_loss=0.1358, over 28233.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3848, pruned_loss=0.1339, over 5664393.81 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3651, pruned_loss=0.108, over 5726077.85 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3877, pruned_loss=0.137, over 5650511.11 frames. ], batch size: 368, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:50:43,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1642, 0.9384, 0.8088, 1.2853], device='cuda:0'), covar=tensor([0.0732, 0.0333, 0.0351, 0.0781], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0118, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0066, 0.0048, 0.0043, 0.0073], device='cuda:0')
+2023-03-03 15:50:45,885 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=284836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:50:58,312 INFO [train.py:968] (0/2) Epoch 7, batch 11200, giga_loss[loss=0.2998, simple_loss=0.3589, pruned_loss=0.1204, over 28688.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3844, pruned_loss=0.1338, over 5659476.10 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3655, pruned_loss=0.1082, over 5718199.08 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3866, pruned_loss=0.1364, over 5654820.38 frames. ], batch size: 92, lr: 4.70e-03, grad_scale: 8.0
+2023-03-03 15:50:58,758 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1558, 1.5443, 1.4817, 1.1851], device='cuda:0'), covar=tensor([0.1395, 0.2082, 0.1118, 0.1274], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0735, 0.0821, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 15:51:25,609 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:51:27,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:51:44,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.426e+03 1.840e+03 2.647e+03 7.597e+03, threshold=3.679e+03, percent-clipped=3.0
+2023-03-03 15:51:44,303 INFO [train.py:968] (0/2) Epoch 7, batch 11250, giga_loss[loss=0.3076, simple_loss=0.3726, pruned_loss=0.1213, over 28984.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.384, pruned_loss=0.1342, over 5666002.47 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3654, pruned_loss=0.1081, over 5721955.30 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3865, pruned_loss=0.1372, over 5657579.54 frames. ], batch size: 213, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:51:51,471 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:52:26,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8633, 1.7553, 1.4913, 1.4851], device='cuda:0'), covar=tensor([0.0661, 0.0594, 0.0818, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0452, 0.0495, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 15:52:33,085 INFO [train.py:968] (0/2) Epoch 7, batch 11300, giga_loss[loss=0.4623, simple_loss=0.4656, pruned_loss=0.2295, over 27653.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3837, pruned_loss=0.1343, over 5663604.54 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3652, pruned_loss=0.1079, over 5725788.00 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3861, pruned_loss=0.1372, over 5652833.26 frames. ], batch size: 474, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:53:16,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3730, 1.6895, 1.2948, 1.5012], device='cuda:0'), covar=tensor([0.0727, 0.0301, 0.0316, 0.0778], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0066, 0.0048, 0.0043, 0.0073], device='cuda:0')
+2023-03-03 15:53:25,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.913e+02 1.485e+03 1.940e+03 2.653e+03 4.684e+03, threshold=3.881e+03, percent-clipped=8.0
+2023-03-03 15:53:25,216 INFO [train.py:968] (0/2) Epoch 7, batch 11350, giga_loss[loss=0.3732, simple_loss=0.4244, pruned_loss=0.161, over 28905.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3858, pruned_loss=0.1362, over 5666713.07 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3656, pruned_loss=0.1081, over 5729039.01 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3877, pruned_loss=0.139, over 5653676.36 frames. ], batch size: 145, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:53:26,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 15:54:10,245 INFO [train.py:968] (0/2) Epoch 7, batch 11400, giga_loss[loss=0.3338, simple_loss=0.388, pruned_loss=0.1398, over 28789.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3877, pruned_loss=0.1377, over 5673434.91 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3661, pruned_loss=0.1083, over 5733656.55 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3894, pruned_loss=0.1405, over 5657248.19 frames. ], batch size: 119, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:54:25,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7495, 1.7404, 1.4726, 1.9560], device='cuda:0'), covar=tensor([0.2036, 0.2126, 0.2181, 0.2163], device='cuda:0'), in_proj_covar=tensor([0.1197, 0.0906, 0.1046, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:54:52,478 INFO [train.py:968] (0/2) Epoch 7, batch 11450, giga_loss[loss=0.3639, simple_loss=0.4207, pruned_loss=0.1535, over 28973.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3874, pruned_loss=0.1366, over 5675114.94 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3656, pruned_loss=0.108, over 5729761.63 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3901, pruned_loss=0.1403, over 5662426.14 frames. ], batch size: 136, lr: 4.70e-03, grad_scale: 2.0
+2023-03-03 15:54:53,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.557e+03 2.096e+03 3.009e+03 9.827e+03, threshold=4.193e+03, percent-clipped=16.0
+2023-03-03 15:55:36,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3394, 1.5127, 1.2532, 1.5832], device='cuda:0'), covar=tensor([0.2170, 0.1988, 0.2078, 0.1823], device='cuda:0'), in_proj_covar=tensor([0.1194, 0.0903, 0.1043, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:55:40,162 INFO [train.py:968] (0/2) Epoch 7, batch 11500, giga_loss[loss=0.347, simple_loss=0.3876, pruned_loss=0.1532, over 28432.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.389, pruned_loss=0.139, over 5661602.76 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3659, pruned_loss=0.1081, over 5731409.68 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3913, pruned_loss=0.1423, over 5649113.78 frames. ], batch size: 77, lr: 4.70e-03, grad_scale: 2.0
+2023-03-03 15:56:26,220 INFO [train.py:968] (0/2) Epoch 7, batch 11550, giga_loss[loss=0.4119, simple_loss=0.4372, pruned_loss=0.1933, over 27655.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3885, pruned_loss=0.1388, over 5664069.51 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3658, pruned_loss=0.1081, over 5731783.91 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3911, pruned_loss=0.1425, over 5651387.36 frames. ], batch size: 472, lr: 4.70e-03, grad_scale: 2.0
+2023-03-03 15:56:27,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.616e+02 1.483e+03 1.996e+03 2.975e+03 8.292e+03, threshold=3.992e+03, percent-clipped=9.0
+2023-03-03 15:56:35,585 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:56:36,894 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:56:38,376 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-03 15:56:46,872 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-03 15:57:10,912 INFO [train.py:968] (0/2) Epoch 7, batch 11600, giga_loss[loss=0.2813, simple_loss=0.3445, pruned_loss=0.1091, over 28651.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3872, pruned_loss=0.1373, over 5664172.82 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3654, pruned_loss=0.1078, over 5725126.31 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3907, pruned_loss=0.142, over 5656366.48 frames. ], batch size: 71, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:57:25,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9269, 1.8423, 1.3362, 1.4820], device='cuda:0'), covar=tensor([0.0629, 0.0603, 0.0958, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0456, 0.0503, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 15:57:36,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-03 15:58:01,514 INFO [train.py:968] (0/2) Epoch 7, batch 11650, giga_loss[loss=0.3929, simple_loss=0.4055, pruned_loss=0.1901, over 23612.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3868, pruned_loss=0.1364, over 5664003.85 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3654, pruned_loss=0.1078, over 5727473.08 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3901, pruned_loss=0.1407, over 5654657.04 frames. ], batch size: 705, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:58:02,170 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.490e+02 1.601e+03 2.181e+03 2.994e+03 8.120e+03, threshold=4.362e+03, percent-clipped=11.0
+2023-03-03 15:58:30,884 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:58:47,642 INFO [train.py:968] (0/2) Epoch 7, batch 11700, giga_loss[loss=0.341, simple_loss=0.3974, pruned_loss=0.1423, over 28933.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3877, pruned_loss=0.1368, over 5663162.68 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3654, pruned_loss=0.1077, over 5722347.09 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3907, pruned_loss=0.1409, over 5658741.27 frames. ], batch size: 227, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:58:51,700 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:58:53,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:59:00,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2084, 0.9934, 4.2840, 3.2370], device='cuda:0'), covar=tensor([0.1635, 0.2638, 0.0366, 0.0791], device='cuda:0'), in_proj_covar=tensor([0.0598, 0.0559, 0.0798, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 15:59:21,255 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:59:23,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:59:33,932 INFO [train.py:968] (0/2) Epoch 7, batch 11750, giga_loss[loss=0.3953, simple_loss=0.4282, pruned_loss=0.1812, over 27932.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3875, pruned_loss=0.1359, over 5680542.33 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3655, pruned_loss=0.1078, over 5729224.22 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3907, pruned_loss=0.1402, over 5669047.70 frames. ], batch size: 412, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:59:36,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.365e+02 1.598e+03 2.043e+03 2.725e+03 4.986e+03, threshold=4.087e+03, percent-clipped=4.0
+2023-03-03 15:59:38,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4373, 1.7344, 1.3575, 1.2034], device='cuda:0'), covar=tensor([0.1418, 0.1157, 0.1007, 0.1323], device='cuda:0'), in_proj_covar=tensor([0.1526, 0.1393, 0.1341, 0.1441], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 16:00:10,960 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285435.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:00:23,189 INFO [train.py:968] (0/2) Epoch 7, batch 11800, libri_loss[loss=0.3319, simple_loss=0.3966, pruned_loss=0.1336, over 29544.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3899, pruned_loss=0.1385, over 5680749.53 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3659, pruned_loss=0.108, over 5728868.76 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3925, pruned_loss=0.1422, over 5670810.09 frames. ], batch size: 82, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 16:00:51,234 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-03 16:01:10,633 INFO [train.py:968] (0/2) Epoch 7, batch 11850, giga_loss[loss=0.3357, simple_loss=0.3933, pruned_loss=0.1391, over 28849.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3904, pruned_loss=0.1383, over 5689635.98 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3661, pruned_loss=0.1081, over 5733123.60 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3929, pruned_loss=0.1419, over 5676543.21 frames. ], batch size: 199, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 16:01:11,233 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.575e+02 1.595e+03 2.125e+03 3.002e+03 7.115e+03, threshold=4.250e+03, percent-clipped=6.0
+2023-03-03 16:01:58,603 INFO [train.py:968] (0/2) Epoch 7, batch 11900, giga_loss[loss=0.3042, simple_loss=0.3738, pruned_loss=0.1173, over 28735.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3899, pruned_loss=0.1369, over 5679523.77 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3659, pruned_loss=0.108, over 5734977.27 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3924, pruned_loss=0.1404, over 5666558.10 frames. ], batch size: 242, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 16:02:32,095 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:02:45,119 INFO [train.py:968] (0/2) Epoch 7, batch 11950, libri_loss[loss=0.2869, simple_loss=0.357, pruned_loss=0.1084, over 29553.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3893, pruned_loss=0.136, over 5679402.68 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3658, pruned_loss=0.1081, over 5740156.83 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3921, pruned_loss=0.1396, over 5662791.66 frames. ], batch size: 77, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 16:02:45,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.771e+02 1.382e+03 1.738e+03 2.717e+03 7.131e+03, threshold=3.476e+03, percent-clipped=5.0
+2023-03-03 16:02:58,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3558, 1.5429, 1.1921, 1.0796], device='cuda:0'), covar=tensor([0.1469, 0.1291, 0.0984, 0.1295], device='cuda:0'), in_proj_covar=tensor([0.1527, 0.1399, 0.1337, 0.1443], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 16:03:10,846 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-03 16:03:32,043 INFO [train.py:968] (0/2) Epoch 7, batch 12000, giga_loss[loss=0.3762, simple_loss=0.4174, pruned_loss=0.1675, over 29073.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3873, pruned_loss=0.1346, over 5688113.16 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3653, pruned_loss=0.1077, over 5739632.50 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3905, pruned_loss=0.1382, over 5674696.51 frames. ], batch size: 128, lr: 4.70e-03, grad_scale: 8.0
+2023-03-03 16:03:32,047 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 16:03:40,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3188, 2.8842, 1.4036, 1.4131], device='cuda:0'), covar=tensor([0.0878, 0.0292, 0.0887, 0.1294], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0500, 0.0320, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 16:03:41,418 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2307, simple_loss=0.3348, pruned_loss=0.06333, over 944034.00 frames. 
+2023-03-03 16:03:41,418 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-03 16:04:28,425 INFO [train.py:968] (0/2) Epoch 7, batch 12050, giga_loss[loss=0.3484, simple_loss=0.404, pruned_loss=0.1464, over 28664.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3883, pruned_loss=0.1358, over 5677436.54 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3653, pruned_loss=0.1077, over 5743299.52 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3911, pruned_loss=0.1392, over 5662770.36 frames. ], batch size: 307, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:04:29,004 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.556e+02 1.392e+03 1.806e+03 2.441e+03 6.128e+03, threshold=3.611e+03, percent-clipped=8.0
+2023-03-03 16:04:31,031 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 16:04:37,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:04:55,747 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:04:57,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:05:15,616 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5140, 1.6511, 1.3867, 1.8205], device='cuda:0'), covar=tensor([0.2073, 0.2113, 0.2107, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.1194, 0.0908, 0.1049, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 16:05:17,232 INFO [train.py:968] (0/2) Epoch 7, batch 12100, giga_loss[loss=0.3059, simple_loss=0.3762, pruned_loss=0.1178, over 28848.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3905, pruned_loss=0.1374, over 5676515.75 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3658, pruned_loss=0.1079, over 5745025.39 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3926, pruned_loss=0.1402, over 5662814.13 frames. ], batch size: 174, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:05:26,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:05:30,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:06:03,988 INFO [train.py:968] (0/2) Epoch 7, batch 12150, giga_loss[loss=0.383, simple_loss=0.4182, pruned_loss=0.1739, over 27594.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.39, pruned_loss=0.1377, over 5680074.92 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3657, pruned_loss=0.1079, over 5748110.50 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3924, pruned_loss=0.1408, over 5664508.26 frames. ], batch size: 472, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:06:06,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.924e+02 1.401e+03 1.728e+03 2.467e+03 6.119e+03, threshold=3.457e+03, percent-clipped=11.0
+2023-03-03 16:06:08,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 16:06:16,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285810.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:06:54,812 INFO [train.py:968] (0/2) Epoch 7, batch 12200, giga_loss[loss=0.3124, simple_loss=0.377, pruned_loss=0.1239, over 28418.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3902, pruned_loss=0.139, over 5668159.04 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3657, pruned_loss=0.1078, over 5749636.34 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3926, pruned_loss=0.1421, over 5653049.75 frames. ], batch size: 65, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:06:55,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:06:57,172 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:07:22,368 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:07:41,484 INFO [train.py:968] (0/2) Epoch 7, batch 12250, giga_loss[loss=0.344, simple_loss=0.396, pruned_loss=0.146, over 27846.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3911, pruned_loss=0.1396, over 5676684.02 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.366, pruned_loss=0.1078, over 5753973.76 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3933, pruned_loss=0.1427, over 5659072.62 frames. ], batch size: 412, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:07:43,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.275e+02 1.503e+03 1.956e+03 2.654e+03 4.339e+03, threshold=3.912e+03, percent-clipped=6.0
+2023-03-03 16:07:49,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:07:51,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:08:19,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:08:27,651 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 16:08:30,371 INFO [train.py:968] (0/2) Epoch 7, batch 12300, giga_loss[loss=0.3166, simple_loss=0.3779, pruned_loss=0.1276, over 28692.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3911, pruned_loss=0.1395, over 5668267.51 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.366, pruned_loss=0.1078, over 5755706.83 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3934, pruned_loss=0.1428, over 5651151.14 frames. ], batch size: 262, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:08:32,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5302, 1.5775, 1.2538, 1.1895], device='cuda:0'), covar=tensor([0.0688, 0.0526, 0.0960, 0.0893], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0454, 0.0501, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 16:08:32,809 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285953.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:08:35,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285956.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:08:55,090 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-03 16:09:05,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285985.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:09:09,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2455, 1.3148, 3.4740, 3.2005], device='cuda:0'), covar=tensor([0.1317, 0.2262, 0.0402, 0.0766], device='cuda:0'), in_proj_covar=tensor([0.0599, 0.0557, 0.0795, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 16:09:16,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4771, 1.6892, 1.3856, 1.1702], device='cuda:0'), covar=tensor([0.1266, 0.1111, 0.0828, 0.1142], device='cuda:0'), in_proj_covar=tensor([0.1511, 0.1389, 0.1327, 0.1442], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 16:09:18,918 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-286000.pt
+2023-03-03 16:09:20,137 INFO [train.py:968] (0/2) Epoch 7, batch 12350, giga_loss[loss=0.3312, simple_loss=0.3883, pruned_loss=0.137, over 28992.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3907, pruned_loss=0.1393, over 5663732.87 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3661, pruned_loss=0.1077, over 5755189.45 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3928, pruned_loss=0.1424, over 5649590.37 frames. ], batch size: 106, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:09:21,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.831e+02 1.660e+03 2.020e+03 2.911e+03 6.387e+03, threshold=4.041e+03, percent-clipped=7.0
+2023-03-03 16:09:36,882 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:10:09,813 INFO [train.py:968] (0/2) Epoch 7, batch 12400, giga_loss[loss=0.3113, simple_loss=0.3821, pruned_loss=0.1202, over 29082.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3913, pruned_loss=0.1396, over 5652507.76 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3668, pruned_loss=0.1081, over 5756267.24 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3931, pruned_loss=0.1426, over 5636989.85 frames. ], batch size: 128, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:10:30,097 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 16:10:56,338 INFO [train.py:968] (0/2) Epoch 7, batch 12450, giga_loss[loss=0.3325, simple_loss=0.3869, pruned_loss=0.1391, over 28756.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3918, pruned_loss=0.1397, over 5653097.72 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.367, pruned_loss=0.1081, over 5757126.19 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3935, pruned_loss=0.1428, over 5638026.29 frames. ], batch size: 99, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:10:57,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.487e+02 1.517e+03 2.245e+03 3.288e+03 6.683e+03, threshold=4.491e+03, percent-clipped=12.0
+2023-03-03 16:11:42,074 INFO [train.py:968] (0/2) Epoch 7, batch 12500, giga_loss[loss=0.3175, simple_loss=0.3825, pruned_loss=0.1262, over 28917.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3901, pruned_loss=0.1384, over 5661733.97 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3665, pruned_loss=0.1078, over 5760002.10 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3923, pruned_loss=0.1416, over 5645414.12 frames. ], batch size: 213, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:11:52,364 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-03 16:12:01,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8990, 1.8896, 1.3616, 1.4779], device='cuda:0'), covar=tensor([0.0639, 0.0552, 0.0961, 0.0953], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0451, 0.0499, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 16:12:33,038 INFO [train.py:968] (0/2) Epoch 7, batch 12550, giga_loss[loss=0.3074, simple_loss=0.3654, pruned_loss=0.1247, over 28834.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3885, pruned_loss=0.1374, over 5666541.70 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3664, pruned_loss=0.1077, over 5762412.33 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3908, pruned_loss=0.1406, over 5649753.30 frames. ], batch size: 199, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:12:37,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.752e+03 2.443e+03 3.123e+03 9.999e+03, threshold=4.887e+03, percent-clipped=10.0
+2023-03-03 16:12:50,350 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-03 16:13:13,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5354, 1.9477, 1.9220, 1.5004], device='cuda:0'), covar=tensor([0.1565, 0.1923, 0.1143, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0731, 0.0819, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 16:13:20,395 INFO [train.py:968] (0/2) Epoch 7, batch 12600, giga_loss[loss=0.3147, simple_loss=0.3799, pruned_loss=0.1247, over 28902.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3857, pruned_loss=0.1352, over 5677496.05 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3664, pruned_loss=0.1076, over 5767155.00 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3882, pruned_loss=0.1388, over 5656657.81 frames. ], batch size: 227, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:13:27,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2493, 1.6575, 1.2487, 0.3936], device='cuda:0'), covar=tensor([0.1640, 0.1063, 0.1604, 0.2745], device='cuda:0'), in_proj_covar=tensor([0.1463, 0.1388, 0.1424, 0.1199], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 16:14:09,572 INFO [train.py:968] (0/2) Epoch 7, batch 12650, giga_loss[loss=0.2864, simple_loss=0.3545, pruned_loss=0.1091, over 28902.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3818, pruned_loss=0.1331, over 5669725.81 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3662, pruned_loss=0.1075, over 5763953.21 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3843, pruned_loss=0.1366, over 5654324.59 frames. ], batch size: 213, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:14:12,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.670e+02 1.572e+03 1.949e+03 2.885e+03 7.919e+03, threshold=3.898e+03, percent-clipped=5.0
+2023-03-03 16:14:59,899 INFO [train.py:968] (0/2) Epoch 7, batch 12700, giga_loss[loss=0.292, simple_loss=0.3445, pruned_loss=0.1197, over 28101.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3802, pruned_loss=0.1335, over 5659344.04 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3663, pruned_loss=0.1075, over 5765537.90 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3822, pruned_loss=0.1364, over 5644628.03 frames. ], batch size: 77, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:15:06,648 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.28 vs. limit=5.0
+2023-03-03 16:15:12,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4298, 1.6770, 1.8363, 1.4632], device='cuda:0'), covar=tensor([0.1371, 0.1686, 0.1048, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0732, 0.0820, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 16:15:39,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=286390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:15:46,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4054, 1.7232, 1.6871, 1.3889], device='cuda:0'), covar=tensor([0.1267, 0.1774, 0.1025, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0727, 0.0817, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 16:15:49,499 INFO [train.py:968] (0/2) Epoch 7, batch 12750, giga_loss[loss=0.2972, simple_loss=0.3646, pruned_loss=0.1149, over 28833.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3796, pruned_loss=0.1337, over 5655499.11 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3666, pruned_loss=0.1078, over 5766384.51 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3811, pruned_loss=0.1363, over 5641206.55 frames. ], batch size: 199, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:15:51,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.846e+02 1.866e+03 2.455e+03 3.377e+03 8.234e+03, threshold=4.910e+03, percent-clipped=23.0
+2023-03-03 16:16:30,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4413, 1.5661, 1.3086, 1.7814], device='cuda:0'), covar=tensor([0.2520, 0.2243, 0.2309, 0.2126], device='cuda:0'), in_proj_covar=tensor([0.1197, 0.0908, 0.1052, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 16:16:35,932 INFO [train.py:968] (0/2) Epoch 7, batch 12800, giga_loss[loss=0.3581, simple_loss=0.4046, pruned_loss=0.1558, over 27607.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3785, pruned_loss=0.1323, over 5655048.85 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3665, pruned_loss=0.1078, over 5763543.45 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3802, pruned_loss=0.1352, over 5642424.34 frames. ], batch size: 472, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:16:44,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7334, 1.8225, 1.3948, 1.4200], device='cuda:0'), covar=tensor([0.0650, 0.0454, 0.0847, 0.0882], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0452, 0.0501, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 16:17:24,383 INFO [train.py:968] (0/2) Epoch 7, batch 12850, libri_loss[loss=0.2701, simple_loss=0.348, pruned_loss=0.09605, over 29548.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3766, pruned_loss=0.129, over 5654860.48 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3662, pruned_loss=0.1077, over 5764498.01 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3786, pruned_loss=0.1321, over 5640821.53 frames. ], batch size: 89, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:17:27,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.786e+02 1.666e+03 2.191e+03 2.887e+03 8.302e+03, threshold=4.381e+03, percent-clipped=7.0
+2023-03-03 16:17:38,184 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286516.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:17:44,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:17:56,434 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=286533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:18:00,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=286536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:18:14,813 INFO [train.py:968] (0/2) Epoch 7, batch 12900, giga_loss[loss=0.2832, simple_loss=0.3529, pruned_loss=0.1068, over 28872.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3743, pruned_loss=0.1258, over 5650652.44 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3659, pruned_loss=0.1077, over 5761716.74 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3765, pruned_loss=0.1289, over 5638229.97 frames. ], batch size: 186, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:18:28,509 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=286565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:19:03,495 INFO [train.py:968] (0/2) Epoch 7, batch 12950, giga_loss[loss=0.2932, simple_loss=0.3702, pruned_loss=0.1081, over 28738.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3699, pruned_loss=0.1215, over 5651373.72 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3652, pruned_loss=0.1076, over 5765841.10 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3725, pruned_loss=0.1245, over 5634778.80 frames. ], batch size: 262, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:19:05,687 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.098e+02 1.354e+03 1.727e+03 2.326e+03 5.662e+03, threshold=3.455e+03, percent-clipped=5.0
+2023-03-03 16:19:42,945 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-03 16:19:55,887 INFO [train.py:968] (0/2) Epoch 7, batch 13000, giga_loss[loss=0.3024, simple_loss=0.3729, pruned_loss=0.116, over 28922.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3672, pruned_loss=0.1188, over 5652398.02 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3654, pruned_loss=0.1079, over 5765357.64 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3691, pruned_loss=0.1212, over 5636994.97 frames. ], batch size: 213, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:20:00,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9143, 2.7721, 1.7834, 0.8015], device='cuda:0'), covar=tensor([0.4232, 0.2107, 0.2561, 0.3991], device='cuda:0'), in_proj_covar=tensor([0.1446, 0.1374, 0.1413, 0.1187], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 16:20:25,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2425, 0.8405, 0.8056, 1.3262], device='cuda:0'), covar=tensor([0.0719, 0.0329, 0.0348, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0121, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0067, 0.0048, 0.0043, 0.0073], device='cuda:0')
+2023-03-03 16:20:34,757 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:20:47,081 INFO [train.py:968] (0/2) Epoch 7, batch 13050, libri_loss[loss=0.2573, simple_loss=0.3289, pruned_loss=0.09282, over 29566.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3639, pruned_loss=0.1143, over 5654361.34 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3651, pruned_loss=0.1079, over 5767540.13 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3657, pruned_loss=0.1164, over 5637862.30 frames. ], batch size: 79, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:20:50,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.187e+02 1.212e+03 1.493e+03 2.069e+03 3.682e+03, threshold=2.987e+03, percent-clipped=2.0
+2023-03-03 16:21:19,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4898, 1.8331, 1.8730, 1.5172], device='cuda:0'), covar=tensor([0.1615, 0.1849, 0.1222, 0.1434], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0719, 0.0817, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 16:21:23,655 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:21:37,256 INFO [train.py:968] (0/2) Epoch 7, batch 13100, giga_loss[loss=0.2523, simple_loss=0.3388, pruned_loss=0.08294, over 28699.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3629, pruned_loss=0.1119, over 5667355.17 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3645, pruned_loss=0.1077, over 5771217.78 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3649, pruned_loss=0.1139, over 5648332.13 frames. ], batch size: 262, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:22:28,318 INFO [train.py:968] (0/2) Epoch 7, batch 13150, giga_loss[loss=0.2634, simple_loss=0.3406, pruned_loss=0.09308, over 28586.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3629, pruned_loss=0.1121, over 5659678.41 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3641, pruned_loss=0.1076, over 5772787.82 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3648, pruned_loss=0.1138, over 5641803.41 frames. ], batch size: 85, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:22:31,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.033e+02 1.271e+03 1.628e+03 2.450e+03 5.222e+03, threshold=3.255e+03, percent-clipped=14.0
+2023-03-03 16:22:49,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5977, 1.7879, 1.1863, 1.4177], device='cuda:0'), covar=tensor([0.0647, 0.0398, 0.0854, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0447, 0.0499, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 16:23:16,321 INFO [train.py:968] (0/2) Epoch 7, batch 13200, giga_loss[loss=0.2712, simple_loss=0.341, pruned_loss=0.1007, over 27682.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3591, pruned_loss=0.1096, over 5658160.60 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3632, pruned_loss=0.1072, over 5775033.99 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3615, pruned_loss=0.1114, over 5639232.44 frames. ], batch size: 472, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:23:58,523 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=286891.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:24:03,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=286897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:24:05,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3724, 2.8531, 1.3845, 1.4363], device='cuda:0'), covar=tensor([0.0867, 0.0337, 0.0881, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0497, 0.0320, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 16:24:05,544 INFO [train.py:968] (0/2) Epoch 7, batch 13250, giga_loss[loss=0.3294, simple_loss=0.3907, pruned_loss=0.134, over 28619.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3577, pruned_loss=0.109, over 5656398.78 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3629, pruned_loss=0.1071, over 5777950.55 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3597, pruned_loss=0.1106, over 5635274.23 frames. ], batch size: 336, lr: 4.69e-03, grad_scale: 2.0
+2023-03-03 16:24:10,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.125e+02 1.216e+03 1.775e+03 2.482e+03 1.117e+04, threshold=3.549e+03, percent-clipped=14.0
+2023-03-03 16:24:16,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5638, 4.5053, 1.7304, 1.7431], device='cuda:0'), covar=tensor([0.0920, 0.0235, 0.0854, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0498, 0.0320, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 16:24:24,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8768, 1.2053, 3.3901, 2.9558], device='cuda:0'), covar=tensor([0.1598, 0.2281, 0.0445, 0.0716], device='cuda:0'), in_proj_covar=tensor([0.0594, 0.0555, 0.0791, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 16:24:33,448 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4671, 2.0829, 1.5137, 0.5823], device='cuda:0'), covar=tensor([0.2629, 0.1479, 0.2239, 0.3260], device='cuda:0'), in_proj_covar=tensor([0.1436, 0.1368, 0.1408, 0.1182], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 16:24:51,648 INFO [train.py:968] (0/2) Epoch 7, batch 13300, giga_loss[loss=0.3092, simple_loss=0.3776, pruned_loss=0.1204, over 28345.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3585, pruned_loss=0.1098, over 5655956.20 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3621, pruned_loss=0.1068, over 5780655.21 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3607, pruned_loss=0.1114, over 5631996.56 frames. ], batch size: 369, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:25:01,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3350, 1.5674, 1.3208, 1.3265], device='cuda:0'), covar=tensor([0.2319, 0.2139, 0.2281, 0.1884], device='cuda:0'), in_proj_covar=tensor([0.1193, 0.0899, 0.1048, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 16:25:19,643 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:25:39,996 INFO [train.py:968] (0/2) Epoch 7, batch 13350, giga_loss[loss=0.2799, simple_loss=0.3547, pruned_loss=0.1026, over 28526.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3577, pruned_loss=0.1091, over 5655214.84 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3617, pruned_loss=0.1068, over 5778052.99 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3597, pruned_loss=0.1106, over 5634083.11 frames. ], batch size: 336, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:25:45,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.714e+02 1.342e+03 1.737e+03 2.224e+03 5.889e+03, threshold=3.474e+03, percent-clipped=8.0
+2023-03-03 16:25:51,138 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1365, 1.4623, 1.1750, 0.9217], device='cuda:0'), covar=tensor([0.2266, 0.2170, 0.2338, 0.1917], device='cuda:0'), in_proj_covar=tensor([0.1196, 0.0905, 0.1051, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 16:25:55,979 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3553, 1.5583, 1.1960, 1.5556], device='cuda:0'), covar=tensor([0.0772, 0.0315, 0.0344, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0122, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0067, 0.0048, 0.0044, 0.0074], device='cuda:0')
+2023-03-03 16:26:06,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287034.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:26:08,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287037.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:26:10,627 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:26:14,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:26:23,142 INFO [train.py:968] (0/2) Epoch 7, batch 13400, giga_loss[loss=0.2625, simple_loss=0.32, pruned_loss=0.1025, over 24242.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3545, pruned_loss=0.1066, over 5658751.33 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3603, pruned_loss=0.1062, over 5775855.13 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3572, pruned_loss=0.1084, over 5636429.49 frames. ], batch size: 705, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:26:37,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:26:38,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:26:41,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287066.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:26:45,944 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:27:07,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6503, 1.7790, 1.6090, 2.3247], device='cuda:0'), covar=tensor([0.2189, 0.2053, 0.2128, 0.1756], device='cuda:0'), in_proj_covar=tensor([0.1193, 0.0900, 0.1050, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 16:27:15,383 INFO [train.py:968] (0/2) Epoch 7, batch 13450, giga_loss[loss=0.2613, simple_loss=0.3293, pruned_loss=0.09672, over 27647.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3516, pruned_loss=0.104, over 5656627.92 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3605, pruned_loss=0.1064, over 5775089.46 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3534, pruned_loss=0.1052, over 5637436.43 frames. ], batch size: 472, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:27:20,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1250, 3.9326, 3.7237, 1.8940], device='cuda:0'), covar=tensor([0.0532, 0.0735, 0.0784, 0.2163], device='cuda:0'), in_proj_covar=tensor([0.0919, 0.0868, 0.0779, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 16:27:21,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.618e+02 1.251e+03 1.589e+03 2.143e+03 4.491e+03, threshold=3.179e+03, percent-clipped=4.0
+2023-03-03 16:27:31,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:28:05,025 INFO [train.py:968] (0/2) Epoch 7, batch 13500, giga_loss[loss=0.2714, simple_loss=0.3436, pruned_loss=0.09961, over 28588.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3484, pruned_loss=0.1023, over 5666237.90 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3599, pruned_loss=0.1061, over 5778975.97 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.35, pruned_loss=0.1034, over 5643145.95 frames. ], batch size: 307, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:28:22,478 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:28:59,151 INFO [train.py:968] (0/2) Epoch 7, batch 13550, libri_loss[loss=0.2285, simple_loss=0.2978, pruned_loss=0.07958, over 29488.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3463, pruned_loss=0.1015, over 5668111.46 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.359, pruned_loss=0.1057, over 5781444.93 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3482, pruned_loss=0.1027, over 5645362.80 frames. ], batch size: 70, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:29:04,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.200e+02 1.359e+03 1.904e+03 2.794e+03 7.507e+03, threshold=3.808e+03, percent-clipped=18.0
+2023-03-03 16:29:05,276 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:29:09,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:29:25,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4561, 3.5317, 1.6226, 1.4013], device='cuda:0'), covar=tensor([0.0825, 0.0293, 0.0828, 0.1280], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0493, 0.0321, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 16:29:36,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:29:46,761 INFO [train.py:968] (0/2) Epoch 7, batch 13600, libri_loss[loss=0.3404, simple_loss=0.3957, pruned_loss=0.1425, over 29376.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3456, pruned_loss=0.1017, over 5658398.30 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3584, pruned_loss=0.1058, over 5779292.78 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.347, pruned_loss=0.1023, over 5636040.79 frames. ], batch size: 92, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:29:47,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-03 16:29:49,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6825, 3.4890, 3.3233, 1.8820], device='cuda:0'), covar=tensor([0.0664, 0.0843, 0.0870, 0.2314], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0870, 0.0780, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 16:29:51,855 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:29:55,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:30:27,354 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:30:41,092 INFO [train.py:968] (0/2) Epoch 7, batch 13650, giga_loss[loss=0.2738, simple_loss=0.3581, pruned_loss=0.09477, over 28855.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3468, pruned_loss=0.1016, over 5661762.82 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3578, pruned_loss=0.1056, over 5779008.54 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3482, pruned_loss=0.1023, over 5641526.20 frames. ], batch size: 145, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:30:46,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.170e+02 1.203e+03 1.563e+03 2.217e+03 5.179e+03, threshold=3.126e+03, percent-clipped=3.0
+2023-03-03 16:31:35,945 INFO [train.py:968] (0/2) Epoch 7, batch 13700, giga_loss[loss=0.278, simple_loss=0.3522, pruned_loss=0.1019, over 28896.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3496, pruned_loss=0.1021, over 5661519.05 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3573, pruned_loss=0.1052, over 5780675.20 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.351, pruned_loss=0.1028, over 5640780.54 frames. ], batch size: 227, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:31:44,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:32:33,446 INFO [train.py:968] (0/2) Epoch 7, batch 13750, giga_loss[loss=0.2763, simple_loss=0.3533, pruned_loss=0.09963, over 28513.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3503, pruned_loss=0.102, over 5671338.96 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3572, pruned_loss=0.1052, over 5782635.03 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3513, pruned_loss=0.1026, over 5649929.97 frames. ], batch size: 336, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:32:41,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.457e+02 1.381e+03 1.852e+03 2.584e+03 6.658e+03, threshold=3.705e+03, percent-clipped=11.0
+2023-03-03 16:33:21,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:33:31,589 INFO [train.py:968] (0/2) Epoch 7, batch 13800, giga_loss[loss=0.2297, simple_loss=0.3172, pruned_loss=0.07111, over 29005.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3485, pruned_loss=0.1012, over 5681153.91 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3566, pruned_loss=0.1049, over 5787566.37 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3497, pruned_loss=0.1018, over 5655839.96 frames. ], batch size: 136, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:34:30,866 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:34:31,144 INFO [train.py:968] (0/2) Epoch 7, batch 13850, giga_loss[loss=0.2609, simple_loss=0.3491, pruned_loss=0.08631, over 29007.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3463, pruned_loss=0.09886, over 5678527.06 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3561, pruned_loss=0.1045, over 5788831.55 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3474, pruned_loss=0.09954, over 5654198.63 frames. ], batch size: 285, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:34:33,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:34:37,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.189e+02 1.177e+03 1.470e+03 1.941e+03 1.040e+04, threshold=2.941e+03, percent-clipped=6.0
+2023-03-03 16:35:07,904 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:35:16,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:35:31,514 INFO [train.py:968] (0/2) Epoch 7, batch 13900, giga_loss[loss=0.2588, simple_loss=0.3392, pruned_loss=0.08918, over 28150.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3454, pruned_loss=0.09691, over 5677233.98 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3564, pruned_loss=0.1048, over 5784517.65 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3459, pruned_loss=0.09716, over 5660966.20 frames. ], batch size: 412, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:36:10,297 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:36:14,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:36:31,775 INFO [train.py:968] (0/2) Epoch 7, batch 13950, giga_loss[loss=0.237, simple_loss=0.3122, pruned_loss=0.08097, over 29032.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.344, pruned_loss=0.09775, over 5668716.41 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3565, pruned_loss=0.105, over 5779740.70 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3439, pruned_loss=0.09741, over 5655154.12 frames. ], batch size: 128, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:36:39,639 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.379e+02 1.225e+03 1.734e+03 2.528e+03 9.524e+03, threshold=3.469e+03, percent-clipped=13.0
+2023-03-03 16:36:46,141 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1766, 1.3392, 1.3896, 1.2453], device='cuda:0'), covar=tensor([0.1083, 0.1256, 0.1463, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0713, 0.0632, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 16:36:48,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:36:57,020 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-03 16:37:32,062 INFO [train.py:968] (0/2) Epoch 7, batch 14000, giga_loss[loss=0.2577, simple_loss=0.3254, pruned_loss=0.09501, over 29146.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3435, pruned_loss=0.09827, over 5652513.79 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3564, pruned_loss=0.1051, over 5762927.01 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3432, pruned_loss=0.09781, over 5655776.87 frames. ], batch size: 113, lr: 4.68e-03, grad_scale: 8.0
+2023-03-03 16:37:44,070 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 16:38:07,258 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:38:09,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:38:28,354 INFO [train.py:968] (0/2) Epoch 7, batch 14050, libri_loss[loss=0.2982, simple_loss=0.3679, pruned_loss=0.1143, over 29237.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3437, pruned_loss=0.09875, over 5657789.79 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3563, pruned_loss=0.1052, over 5765467.81 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3433, pruned_loss=0.09814, over 5655128.42 frames. ], batch size: 94, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:38:36,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.791e+02 1.251e+03 1.524e+03 1.952e+03 7.709e+03, threshold=3.049e+03, percent-clipped=6.0
+2023-03-03 16:38:44,346 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:39:31,281 INFO [train.py:968] (0/2) Epoch 7, batch 14100, giga_loss[loss=0.2652, simple_loss=0.3486, pruned_loss=0.09087, over 28701.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3443, pruned_loss=0.09814, over 5651837.80 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3561, pruned_loss=0.1052, over 5765801.10 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.344, pruned_loss=0.0976, over 5648454.99 frames. ], batch size: 242, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:40:22,900 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 16:40:35,709 INFO [train.py:968] (0/2) Epoch 7, batch 14150, giga_loss[loss=0.2419, simple_loss=0.3232, pruned_loss=0.08031, over 28946.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3452, pruned_loss=0.0982, over 5658440.71 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.356, pruned_loss=0.1053, over 5768972.82 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3448, pruned_loss=0.09752, over 5650316.29 frames. ], batch size: 164, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:40:40,954 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:40:43,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.549e+02 1.360e+03 1.741e+03 2.472e+03 1.108e+04, threshold=3.482e+03, percent-clipped=15.0
+2023-03-03 16:41:38,368 INFO [train.py:968] (0/2) Epoch 7, batch 14200, giga_loss[loss=0.2325, simple_loss=0.3165, pruned_loss=0.07428, over 28837.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3421, pruned_loss=0.09652, over 5675592.26 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3551, pruned_loss=0.1048, over 5774252.83 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3422, pruned_loss=0.09611, over 5660682.23 frames. ], batch size: 174, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:42:33,953 INFO [train.py:968] (0/2) Epoch 7, batch 14250, giga_loss[loss=0.2547, simple_loss=0.3376, pruned_loss=0.08588, over 28940.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3424, pruned_loss=0.09676, over 5690443.57 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3539, pruned_loss=0.1042, over 5779189.45 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.343, pruned_loss=0.09658, over 5669713.33 frames. ], batch size: 155, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:42:41,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-03 16:42:45,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.429e+02 1.272e+03 1.976e+03 3.018e+03 6.659e+03, threshold=3.952e+03, percent-clipped=12.0
+2023-03-03 16:43:17,358 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:43:30,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5894, 1.7432, 1.5488, 1.5698], device='cuda:0'), covar=tensor([0.1290, 0.2073, 0.1770, 0.1712], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0717, 0.0631, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 16:43:44,083 INFO [train.py:968] (0/2) Epoch 7, batch 14300, giga_loss[loss=0.2969, simple_loss=0.3789, pruned_loss=0.1075, over 28957.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3473, pruned_loss=0.098, over 5676327.94 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3538, pruned_loss=0.1042, over 5771867.60 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3477, pruned_loss=0.09779, over 5665841.52 frames. ], batch size: 199, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:43:50,971 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:44:19,679 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:44:50,471 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-288000.pt
+2023-03-03 16:44:50,764 INFO [train.py:968] (0/2) Epoch 7, batch 14350, giga_loss[loss=0.2754, simple_loss=0.3404, pruned_loss=0.1052, over 26920.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3491, pruned_loss=0.0965, over 5675308.20 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3537, pruned_loss=0.1041, over 5772636.26 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3495, pruned_loss=0.09633, over 5666115.39 frames. ], batch size: 555, lr: 4.68e-03, grad_scale: 1.0
+2023-03-03 16:45:00,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.200e+02 1.353e+03 1.726e+03 2.895e+03 1.770e+04, threshold=3.451e+03, percent-clipped=15.0
+2023-03-03 16:45:20,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1687, 1.4261, 1.4440, 1.3470], device='cuda:0'), covar=tensor([0.1107, 0.1251, 0.1538, 0.1271], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0713, 0.0630, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 16:45:38,002 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8699, 1.9213, 1.2427, 1.6001], device='cuda:0'), covar=tensor([0.0699, 0.0611, 0.1014, 0.0945], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0441, 0.0498, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 16:45:49,504 INFO [train.py:968] (0/2) Epoch 7, batch 14400, giga_loss[loss=0.2436, simple_loss=0.3341, pruned_loss=0.07655, over 29041.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3485, pruned_loss=0.09475, over 5672029.03 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3535, pruned_loss=0.104, over 5772039.95 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3489, pruned_loss=0.09454, over 5663031.72 frames. ], batch size: 285, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:46:49,267 INFO [train.py:968] (0/2) Epoch 7, batch 14450, giga_loss[loss=0.2883, simple_loss=0.3625, pruned_loss=0.1071, over 28966.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3477, pruned_loss=0.09454, over 5675777.38 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.353, pruned_loss=0.1038, over 5774849.93 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3482, pruned_loss=0.09429, over 5663440.50 frames. ], batch size: 213, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:47:02,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.163e+02 1.279e+03 1.844e+03 2.623e+03 1.184e+04, threshold=3.687e+03, percent-clipped=11.0
+2023-03-03 16:47:07,432 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:47:54,874 INFO [train.py:968] (0/2) Epoch 7, batch 14500, giga_loss[loss=0.2823, simple_loss=0.3542, pruned_loss=0.1052, over 28650.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3482, pruned_loss=0.09608, over 5677880.88 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3529, pruned_loss=0.1038, over 5776834.14 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3487, pruned_loss=0.09579, over 5665172.64 frames. ], batch size: 242, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:48:15,197 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:48:33,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:49:05,991 INFO [train.py:968] (0/2) Epoch 7, batch 14550, giga_loss[loss=0.2887, simple_loss=0.3616, pruned_loss=0.1079, over 28800.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3488, pruned_loss=0.09734, over 5690273.32 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3529, pruned_loss=0.1039, over 5778070.97 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3492, pruned_loss=0.09693, over 5678390.90 frames. ], batch size: 243, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:49:19,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.253e+02 1.246e+03 1.533e+03 2.288e+03 5.896e+03, threshold=3.065e+03, percent-clipped=2.0
+2023-03-03 16:50:24,758 INFO [train.py:968] (0/2) Epoch 7, batch 14600, giga_loss[loss=0.2199, simple_loss=0.2864, pruned_loss=0.07667, over 24298.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3472, pruned_loss=0.09729, over 5688643.96 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3523, pruned_loss=0.1035, over 5783223.10 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3479, pruned_loss=0.09709, over 5671509.86 frames. ], batch size: 705, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:51:39,846 INFO [train.py:968] (0/2) Epoch 7, batch 14650, giga_loss[loss=0.2805, simple_loss=0.3296, pruned_loss=0.1157, over 23698.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3442, pruned_loss=0.09555, over 5677146.96 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3528, pruned_loss=0.104, over 5773667.14 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3442, pruned_loss=0.09489, over 5670706.21 frames. ], batch size: 705, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:51:46,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:51:50,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.384e+02 1.234e+03 1.702e+03 2.435e+03 6.614e+03, threshold=3.403e+03, percent-clipped=12.0
+2023-03-03 16:52:08,501 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:52:11,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:52:19,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:52:47,832 INFO [train.py:968] (0/2) Epoch 7, batch 14700, giga_loss[loss=0.2409, simple_loss=0.3181, pruned_loss=0.08188, over 28763.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.342, pruned_loss=0.09443, over 5681265.76 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3527, pruned_loss=0.1039, over 5774753.85 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3419, pruned_loss=0.0938, over 5674059.41 frames. ], batch size: 263, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:52:50,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:52:54,300 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:53:06,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5265, 1.9037, 1.9328, 1.5045], device='cuda:0'), covar=tensor([0.1724, 0.1964, 0.1232, 0.1422], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0703, 0.0805, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 16:53:45,003 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:53:49,868 INFO [train.py:968] (0/2) Epoch 7, batch 14750, giga_loss[loss=0.2587, simple_loss=0.3206, pruned_loss=0.09836, over 24757.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3436, pruned_loss=0.09641, over 5676910.88 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3525, pruned_loss=0.1039, over 5775614.87 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3433, pruned_loss=0.09555, over 5666146.58 frames. ], batch size: 705, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:53:58,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.725e+02 1.311e+03 1.748e+03 2.328e+03 7.155e+03, threshold=3.495e+03, percent-clipped=11.0
+2023-03-03 16:54:48,074 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:54:50,421 INFO [train.py:968] (0/2) Epoch 7, batch 14800, giga_loss[loss=0.2313, simple_loss=0.318, pruned_loss=0.07234, over 28393.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.347, pruned_loss=0.09791, over 5677805.04 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3519, pruned_loss=0.1037, over 5771289.85 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3472, pruned_loss=0.09718, over 5670005.40 frames. ], batch size: 78, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 16:54:50,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:18,004 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:22,407 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:27,192 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:38,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:45,965 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:49,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:53,520 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288499.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:55:53,845 INFO [train.py:968] (0/2) Epoch 7, batch 14850, giga_loss[loss=0.296, simple_loss=0.3426, pruned_loss=0.1247, over 24483.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3454, pruned_loss=0.09803, over 5678778.76 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3514, pruned_loss=0.1035, over 5773164.81 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3458, pruned_loss=0.09755, over 5669736.13 frames. ], batch size: 705, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 16:55:59,976 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:56:05,556 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.783e+02 1.431e+03 1.731e+03 2.347e+03 8.678e+03, threshold=3.461e+03, percent-clipped=11.0
+2023-03-03 16:56:25,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:56:50,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:56:55,547 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288545.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:57:01,433 INFO [train.py:968] (0/2) Epoch 7, batch 14900, giga_loss[loss=0.2531, simple_loss=0.33, pruned_loss=0.08806, over 28963.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3454, pruned_loss=0.09888, over 5680842.89 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3516, pruned_loss=0.1036, over 5770136.99 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3456, pruned_loss=0.09839, over 5675910.05 frames. ], batch size: 145, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 16:57:24,048 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-03 16:58:03,775 INFO [train.py:968] (0/2) Epoch 7, batch 14950, giga_loss[loss=0.2507, simple_loss=0.3387, pruned_loss=0.08137, over 28409.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3465, pruned_loss=0.09927, over 5675145.52 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3518, pruned_loss=0.1038, over 5761735.51 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3464, pruned_loss=0.09868, over 5677869.39 frames. ], batch size: 336, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 16:58:16,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.920e+02 1.363e+03 1.787e+03 2.565e+03 7.673e+03, threshold=3.573e+03, percent-clipped=15.0
+2023-03-03 16:58:46,447 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:58:53,132 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:59:15,302 INFO [train.py:968] (0/2) Epoch 7, batch 15000, giga_loss[loss=0.2589, simple_loss=0.3478, pruned_loss=0.08502, over 28950.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.349, pruned_loss=0.09981, over 5674662.54 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3517, pruned_loss=0.1039, over 5763278.99 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3489, pruned_loss=0.09923, over 5674793.34 frames. ], batch size: 164, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 16:59:15,307 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 16:59:23,626 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.215, simple_loss=0.3127, pruned_loss=0.0586, over 944034.00 frames. 
+2023-03-03 16:59:23,626 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-03 16:59:45,447 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:00:17,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:00:21,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:00:37,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4358, 1.6740, 1.5740, 1.5230], device='cuda:0'), covar=tensor([0.1231, 0.1930, 0.1551, 0.1615], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0712, 0.0631, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 17:00:40,588 INFO [train.py:968] (0/2) Epoch 7, batch 15050, giga_loss[loss=0.2557, simple_loss=0.3458, pruned_loss=0.08279, over 28609.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3492, pruned_loss=0.0997, over 5667786.32 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3515, pruned_loss=0.1037, over 5764677.99 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3494, pruned_loss=0.09936, over 5664953.76 frames. ], batch size: 242, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:00:54,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.144e+02 1.243e+03 1.803e+03 2.502e+03 4.916e+03, threshold=3.606e+03, percent-clipped=5.0
+2023-03-03 17:01:08,838 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:01:14,202 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:01:19,343 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-03 17:01:41,831 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:01:57,695 INFO [train.py:968] (0/2) Epoch 7, batch 15100, giga_loss[loss=0.3079, simple_loss=0.3685, pruned_loss=0.1237, over 28465.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3462, pruned_loss=0.09923, over 5657897.05 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3516, pruned_loss=0.1038, over 5756930.85 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3463, pruned_loss=0.09885, over 5660461.59 frames. ], batch size: 369, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:01:59,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-03 17:02:26,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:02:52,391 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:03:05,638 INFO [train.py:968] (0/2) Epoch 7, batch 15150, giga_loss[loss=0.231, simple_loss=0.3097, pruned_loss=0.07615, over 29129.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3398, pruned_loss=0.09645, over 5658662.73 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3512, pruned_loss=0.1036, over 5759114.08 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3399, pruned_loss=0.09618, over 5657088.30 frames. ], batch size: 200, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:03:14,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.335e+02 1.334e+03 1.743e+03 2.260e+03 5.535e+03, threshold=3.487e+03, percent-clipped=4.0
+2023-03-03 17:04:05,839 INFO [train.py:968] (0/2) Epoch 7, batch 15200, giga_loss[loss=0.2767, simple_loss=0.3502, pruned_loss=0.1016, over 28373.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3379, pruned_loss=0.09525, over 5668146.66 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3509, pruned_loss=0.1034, over 5761700.88 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3382, pruned_loss=0.0951, over 5663437.51 frames. ], batch size: 368, lr: 4.67e-03, grad_scale: 8.0
+2023-03-03 17:04:34,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288874.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:04:50,563 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-03 17:05:00,277 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-03 17:05:00,414 INFO [train.py:968] (0/2) Epoch 7, batch 15250, giga_loss[loss=0.2779, simple_loss=0.3517, pruned_loss=0.102, over 27996.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3395, pruned_loss=0.09658, over 5662176.66 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3503, pruned_loss=0.103, over 5761919.52 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3398, pruned_loss=0.09664, over 5654325.38 frames. ], batch size: 412, lr: 4.67e-03, grad_scale: 8.0
+2023-03-03 17:05:05,085 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-03 17:05:06,054 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0484, 1.9377, 1.3570, 1.6113], device='cuda:0'), covar=tensor([0.0655, 0.0594, 0.0939, 0.0956], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0442, 0.0496, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 17:05:09,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.691e+02 1.424e+03 1.723e+03 2.234e+03 7.186e+03, threshold=3.446e+03, percent-clipped=7.0
+2023-03-03 17:05:15,344 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:05:17,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:05:21,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288920.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:05:55,943 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:05:59,030 INFO [train.py:968] (0/2) Epoch 7, batch 15300, giga_loss[loss=0.2626, simple_loss=0.3374, pruned_loss=0.09391, over 28973.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3388, pruned_loss=0.0958, over 5668387.97 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3506, pruned_loss=0.1032, over 5764260.85 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3386, pruned_loss=0.09548, over 5658489.69 frames. ], batch size: 199, lr: 4.67e-03, grad_scale: 8.0
+2023-03-03 17:07:02,764 INFO [train.py:968] (0/2) Epoch 7, batch 15350, giga_loss[loss=0.2426, simple_loss=0.323, pruned_loss=0.08108, over 28737.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3364, pruned_loss=0.09363, over 5655766.17 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3504, pruned_loss=0.1033, over 5757941.04 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3362, pruned_loss=0.09316, over 5651254.50 frames. ], batch size: 243, lr: 4.67e-03, grad_scale: 8.0
+2023-03-03 17:07:06,971 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:07:15,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.805e+02 1.303e+03 1.650e+03 2.344e+03 5.443e+03, threshold=3.300e+03, percent-clipped=4.0
+2023-03-03 17:07:23,976 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289017.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:07:27,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289020.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:07:29,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2660, 1.2469, 3.9160, 3.1704], device='cuda:0'), covar=tensor([0.1853, 0.2539, 0.0699, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0582, 0.0549, 0.0769, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 17:07:59,494 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 17:08:12,278 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289049.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:08:12,623 INFO [train.py:968] (0/2) Epoch 7, batch 15400, giga_loss[loss=0.2535, simple_loss=0.3056, pruned_loss=0.1007, over 24228.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3357, pruned_loss=0.0936, over 5655023.09 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3507, pruned_loss=0.1035, over 5755689.76 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3351, pruned_loss=0.09299, over 5652492.45 frames. ], batch size: 705, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:08:32,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289063.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:08:38,740 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289066.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:09:04,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2806, 1.2756, 1.1328, 1.0245], device='cuda:0'), covar=tensor([0.0589, 0.0388, 0.0830, 0.0720], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0439, 0.0497, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 17:09:15,218 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289095.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:09:15,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:09:21,099 INFO [train.py:968] (0/2) Epoch 7, batch 15450, giga_loss[loss=0.2432, simple_loss=0.3236, pruned_loss=0.08141, over 28727.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3356, pruned_loss=0.09323, over 5655803.13 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3506, pruned_loss=0.1035, over 5757188.34 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3352, pruned_loss=0.09263, over 5651531.24 frames. ], batch size: 119, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:09:32,867 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.753e+02 1.497e+03 2.055e+03 2.932e+03 9.648e+03, threshold=4.110e+03, percent-clipped=20.0
+2023-03-03 17:09:41,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:10:21,919 INFO [train.py:968] (0/2) Epoch 7, batch 15500, giga_loss[loss=0.2432, simple_loss=0.322, pruned_loss=0.08221, over 28988.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3367, pruned_loss=0.09388, over 5660820.44 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3505, pruned_loss=0.1036, over 5760635.08 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3361, pruned_loss=0.09299, over 5651175.36 frames. ], batch size: 213, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:10:46,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:10:53,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3605, 1.6309, 1.2347, 1.3182], device='cuda:0'), covar=tensor([0.1329, 0.0957, 0.0955, 0.1011], device='cuda:0'), in_proj_covar=tensor([0.1563, 0.1388, 0.1322, 0.1449], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 17:11:17,921 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:11:27,759 INFO [train.py:968] (0/2) Epoch 7, batch 15550, giga_loss[loss=0.2162, simple_loss=0.2805, pruned_loss=0.07592, over 24481.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.338, pruned_loss=0.09548, over 5662145.84 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3504, pruned_loss=0.1035, over 5763945.95 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3372, pruned_loss=0.0947, over 5648949.46 frames. ], batch size: 705, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:11:42,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.688e+02 1.181e+03 1.552e+03 2.205e+03 4.342e+03, threshold=3.104e+03, percent-clipped=1.0
+2023-03-03 17:12:08,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1516, 0.9203, 0.8582, 1.3717], device='cuda:0'), covar=tensor([0.0760, 0.0340, 0.0360, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0117, 0.0123, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0068, 0.0048, 0.0044, 0.0074], device='cuda:0')
+2023-03-03 17:12:20,215 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:12:22,144 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:12:32,332 INFO [train.py:968] (0/2) Epoch 7, batch 15600, giga_loss[loss=0.2458, simple_loss=0.3284, pruned_loss=0.08157, over 29108.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3361, pruned_loss=0.09398, over 5663605.39 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3497, pruned_loss=0.103, over 5763242.81 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3359, pruned_loss=0.09368, over 5651824.21 frames. ], batch size: 200, lr: 4.67e-03, grad_scale: 8.0
+2023-03-03 17:12:38,468 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:12:41,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:12:55,382 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:13:18,840 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:13:31,509 INFO [train.py:968] (0/2) Epoch 7, batch 15650, giga_loss[loss=0.3171, simple_loss=0.3811, pruned_loss=0.1265, over 28111.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3393, pruned_loss=0.09426, over 5674761.33 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3499, pruned_loss=0.1031, over 5766097.01 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3388, pruned_loss=0.09372, over 5661253.32 frames. ], batch size: 412, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:13:41,658 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:13:45,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:13:46,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.901e+02 1.292e+03 1.641e+03 2.249e+03 6.545e+03, threshold=3.282e+03, percent-clipped=12.0
+2023-03-03 17:14:21,890 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:14:33,293 INFO [train.py:968] (0/2) Epoch 7, batch 15700, giga_loss[loss=0.2621, simple_loss=0.3437, pruned_loss=0.09021, over 28472.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3417, pruned_loss=0.09509, over 5660662.47 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3498, pruned_loss=0.1031, over 5759119.06 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3411, pruned_loss=0.09454, over 5654440.22 frames. ], batch size: 369, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:15:05,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9161, 2.0003, 1.3644, 1.6526], device='cuda:0'), covar=tensor([0.0700, 0.0532, 0.0972, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0437, 0.0498, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 17:15:08,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:15:20,187 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-03 17:15:35,205 INFO [train.py:968] (0/2) Epoch 7, batch 15750, giga_loss[loss=0.2493, simple_loss=0.3293, pruned_loss=0.08461, over 28563.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3426, pruned_loss=0.09522, over 5655802.45 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3496, pruned_loss=0.103, over 5751855.14 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3422, pruned_loss=0.09471, over 5656400.47 frames. ], batch size: 85, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:15:46,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.208e+02 1.361e+03 1.681e+03 2.133e+03 4.683e+03, threshold=3.362e+03, percent-clipped=7.0
+2023-03-03 17:16:10,316 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4905, 2.2741, 1.6524, 0.6569], device='cuda:0'), covar=tensor([0.2967, 0.1549, 0.2301, 0.3067], device='cuda:0'), in_proj_covar=tensor([0.1406, 0.1347, 0.1391, 0.1160], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 17:16:13,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9465, 1.0541, 3.4975, 2.9604], device='cuda:0'), covar=tensor([0.1495, 0.2153, 0.0410, 0.0769], device='cuda:0'), in_proj_covar=tensor([0.0580, 0.0544, 0.0761, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 17:16:20,537 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289439.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:16:32,396 INFO [train.py:968] (0/2) Epoch 7, batch 15800, giga_loss[loss=0.2946, simple_loss=0.3645, pruned_loss=0.1124, over 28122.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3416, pruned_loss=0.09461, over 5674466.35 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3488, pruned_loss=0.1026, over 5755896.08 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3419, pruned_loss=0.09439, over 5668650.59 frames. ], batch size: 412, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:17:29,033 INFO [train.py:968] (0/2) Epoch 7, batch 15850, giga_loss[loss=0.3171, simple_loss=0.3835, pruned_loss=0.1254, over 28682.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3387, pruned_loss=0.09244, over 5686549.78 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3479, pruned_loss=0.1021, over 5759601.79 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3395, pruned_loss=0.09252, over 5677051.29 frames. ], batch size: 307, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:17:42,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.278e+02 1.274e+03 1.737e+03 2.414e+03 8.093e+03, threshold=3.474e+03, percent-clipped=14.0
+2023-03-03 17:17:45,261 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 17:17:49,342 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:17:52,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:18:00,345 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:18:25,613 INFO [train.py:968] (0/2) Epoch 7, batch 15900, giga_loss[loss=0.2848, simple_loss=0.356, pruned_loss=0.1068, over 28744.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3372, pruned_loss=0.09148, over 5688539.96 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3472, pruned_loss=0.1017, over 5761452.62 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.338, pruned_loss=0.09143, over 5675820.35 frames. ], batch size: 307, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:18:32,362 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:18:45,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:19:24,272 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-03 17:19:25,113 INFO [train.py:968] (0/2) Epoch 7, batch 15950, giga_loss[loss=0.2626, simple_loss=0.3388, pruned_loss=0.09319, over 28094.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3353, pruned_loss=0.0912, over 5687305.58 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3468, pruned_loss=0.1015, over 5762379.35 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3361, pruned_loss=0.09107, over 5673991.58 frames. ], batch size: 412, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:19:39,049 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.918e+02 1.574e+03 2.071e+03 3.115e+03 8.399e+03, threshold=4.141e+03, percent-clipped=18.0
+2023-03-03 17:20:28,864 INFO [train.py:968] (0/2) Epoch 7, batch 16000, giga_loss[loss=0.2487, simple_loss=0.3256, pruned_loss=0.08589, over 29060.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3363, pruned_loss=0.09203, over 5683166.17 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3468, pruned_loss=0.1015, over 5763138.22 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3368, pruned_loss=0.09179, over 5670982.26 frames. ], batch size: 128, lr: 4.66e-03, grad_scale: 8.0
+2023-03-03 17:21:28,420 INFO [train.py:968] (0/2) Epoch 7, batch 16050, giga_loss[loss=0.243, simple_loss=0.3274, pruned_loss=0.07927, over 28959.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3387, pruned_loss=0.09308, over 5684888.98 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3467, pruned_loss=0.1014, over 5765363.12 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.339, pruned_loss=0.09279, over 5671286.75 frames. ], batch size: 213, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:21:42,489 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:21:44,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:21:45,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.710e+02 1.298e+03 1.610e+03 2.164e+03 8.802e+03, threshold=3.221e+03, percent-clipped=3.0
+2023-03-03 17:22:21,017 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:22:32,008 INFO [train.py:968] (0/2) Epoch 7, batch 16100, giga_loss[loss=0.2936, simple_loss=0.3624, pruned_loss=0.1124, over 29081.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3406, pruned_loss=0.09491, over 5680072.39 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3468, pruned_loss=0.1013, over 5767586.68 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3406, pruned_loss=0.09453, over 5665085.14 frames. ], batch size: 285, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:23:29,126 INFO [train.py:968] (0/2) Epoch 7, batch 16150, giga_loss[loss=0.3123, simple_loss=0.3853, pruned_loss=0.1197, over 28943.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3437, pruned_loss=0.09635, over 5686052.66 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3464, pruned_loss=0.1011, over 5767320.28 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3439, pruned_loss=0.09617, over 5672512.31 frames. ], batch size: 186, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:23:41,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.326e+02 1.316e+03 1.743e+03 2.481e+03 5.188e+03, threshold=3.485e+03, percent-clipped=12.0
+2023-03-03 17:23:43,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289814.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:23:43,972 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-03 17:24:23,881 INFO [train.py:968] (0/2) Epoch 7, batch 16200, libri_loss[loss=0.332, simple_loss=0.3924, pruned_loss=0.1358, over 29267.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3456, pruned_loss=0.09655, over 5692274.99 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3461, pruned_loss=0.101, over 5771107.02 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3459, pruned_loss=0.09631, over 5675889.25 frames. ], batch size: 94, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:24:27,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3990, 1.6688, 1.2920, 1.5000], device='cuda:0'), covar=tensor([0.2388, 0.2146, 0.2450, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.1168, 0.0877, 0.1037, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 17:25:19,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:25:23,189 INFO [train.py:968] (0/2) Epoch 7, batch 16250, giga_loss[loss=0.2534, simple_loss=0.3333, pruned_loss=0.08676, over 29002.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3457, pruned_loss=0.09647, over 5675794.81 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3462, pruned_loss=0.1011, over 5756062.40 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3459, pruned_loss=0.09606, over 5674268.89 frames. ], batch size: 186, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:25:29,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:25:40,388 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.832e+02 1.238e+03 1.568e+03 2.189e+03 8.444e+03, threshold=3.135e+03, percent-clipped=4.0
+2023-03-03 17:26:31,063 INFO [train.py:968] (0/2) Epoch 7, batch 16300, giga_loss[loss=0.2323, simple_loss=0.3174, pruned_loss=0.07358, over 28654.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3433, pruned_loss=0.09517, over 5684214.09 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3456, pruned_loss=0.1008, over 5756497.89 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.344, pruned_loss=0.09501, over 5680678.37 frames. ], batch size: 92, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:26:41,560 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289957.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:26:45,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289960.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:26:46,874 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 17:27:24,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289989.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:27:36,782 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-03 17:27:38,212 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-290000.pt
+2023-03-03 17:27:38,499 INFO [train.py:968] (0/2) Epoch 7, batch 16350, libri_loss[loss=0.2562, simple_loss=0.332, pruned_loss=0.09021, over 29523.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3435, pruned_loss=0.09604, over 5685234.00 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3456, pruned_loss=0.1006, over 5759990.55 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.344, pruned_loss=0.09594, over 5677624.70 frames. ], batch size: 81, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:27:53,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.859e+02 1.463e+03 2.022e+03 2.730e+03 7.692e+03, threshold=4.044e+03, percent-clipped=18.0
+2023-03-03 17:28:37,262 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:28:40,506 INFO [train.py:968] (0/2) Epoch 7, batch 16400, giga_loss[loss=0.2582, simple_loss=0.34, pruned_loss=0.08825, over 28722.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3418, pruned_loss=0.09529, over 5676590.19 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3455, pruned_loss=0.1007, over 5763517.66 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3422, pruned_loss=0.09507, over 5665451.75 frames. ], batch size: 243, lr: 4.66e-03, grad_scale: 8.0
+2023-03-03 17:28:41,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:29:20,009 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:29:41,925 INFO [train.py:968] (0/2) Epoch 7, batch 16450, giga_loss[loss=0.2669, simple_loss=0.3316, pruned_loss=0.1012, over 27701.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3402, pruned_loss=0.09551, over 5684149.81 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3447, pruned_loss=0.1003, over 5767696.58 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3411, pruned_loss=0.09554, over 5668945.80 frames. ], batch size: 474, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:29:59,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.067e+02 1.178e+03 1.730e+03 2.422e+03 7.562e+03, threshold=3.460e+03, percent-clipped=9.0
+2023-03-03 17:30:08,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2865, 1.6514, 1.2970, 1.1602], device='cuda:0'), covar=tensor([0.1579, 0.0982, 0.0972, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.1522, 0.1340, 0.1297, 0.1413], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 17:30:41,528 INFO [train.py:968] (0/2) Epoch 7, batch 16500, giga_loss[loss=0.2289, simple_loss=0.3204, pruned_loss=0.06873, over 28833.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3394, pruned_loss=0.09504, over 5685645.65 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3447, pruned_loss=0.1003, over 5771396.37 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.34, pruned_loss=0.09492, over 5668028.71 frames. ], batch size: 174, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:31:03,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1662, 1.4299, 1.1479, 1.0384], device='cuda:0'), covar=tensor([0.1524, 0.1078, 0.0920, 0.1136], device='cuda:0'), in_proj_covar=tensor([0.1527, 0.1345, 0.1304, 0.1418], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 17:31:05,255 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-03 17:31:36,746 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5461, 1.9235, 1.9554, 1.5645], device='cuda:0'), covar=tensor([0.1686, 0.2024, 0.1238, 0.1403], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0702, 0.0810, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 17:31:41,398 INFO [train.py:968] (0/2) Epoch 7, batch 16550, giga_loss[loss=0.2146, simple_loss=0.2805, pruned_loss=0.07436, over 24396.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3397, pruned_loss=0.09461, over 5683863.92 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3443, pruned_loss=0.1001, over 5774496.94 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3404, pruned_loss=0.09456, over 5664808.26 frames. ], batch size: 705, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:31:58,576 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.126e+02 1.317e+03 1.709e+03 2.405e+03 8.056e+03, threshold=3.417e+03, percent-clipped=10.0
+2023-03-03 17:32:39,564 INFO [train.py:968] (0/2) Epoch 7, batch 16600, giga_loss[loss=0.2382, simple_loss=0.3241, pruned_loss=0.07611, over 28116.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3392, pruned_loss=0.09286, over 5685653.73 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3441, pruned_loss=0.09998, over 5774592.31 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3399, pruned_loss=0.09286, over 5669420.25 frames. ], batch size: 412, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:32:40,554 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3540, 1.5738, 1.2026, 1.7049], device='cuda:0'), covar=tensor([0.2340, 0.2147, 0.2360, 0.2103], device='cuda:0'), in_proj_covar=tensor([0.1179, 0.0885, 0.1044, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 17:33:03,488 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:33:10,508 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 17:33:34,460 INFO [train.py:968] (0/2) Epoch 7, batch 16650, giga_loss[loss=0.2272, simple_loss=0.3171, pruned_loss=0.06867, over 28942.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.34, pruned_loss=0.09179, over 5682736.95 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3434, pruned_loss=0.09957, over 5778301.13 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.341, pruned_loss=0.09187, over 5663057.97 frames. ], batch size: 120, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:33:49,115 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290313.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:33:49,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.987e+02 1.244e+03 1.667e+03 2.078e+03 6.815e+03, threshold=3.333e+03, percent-clipped=10.0
+2023-03-03 17:34:23,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3635, 1.5031, 1.4473, 1.3377], device='cuda:0'), covar=tensor([0.1090, 0.1577, 0.1668, 0.1453], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0714, 0.0631, 0.0629], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 17:34:31,338 INFO [train.py:968] (0/2) Epoch 7, batch 16700, giga_loss[loss=0.2536, simple_loss=0.3394, pruned_loss=0.08388, over 28297.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3419, pruned_loss=0.09229, over 5689724.69 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3436, pruned_loss=0.09967, over 5776905.76 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3425, pruned_loss=0.09219, over 5674086.57 frames. ], batch size: 368, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:34:46,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4354, 2.0863, 1.5113, 0.5101], device='cuda:0'), covar=tensor([0.2692, 0.1699, 0.2653, 0.3243], device='cuda:0'), in_proj_covar=tensor([0.1452, 0.1378, 0.1422, 0.1196], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 17:35:24,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:35:36,160 INFO [train.py:968] (0/2) Epoch 7, batch 16750, giga_loss[loss=0.2742, simple_loss=0.3474, pruned_loss=0.1005, over 29009.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3418, pruned_loss=0.09226, over 5685460.64 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3436, pruned_loss=0.09962, over 5769565.25 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3423, pruned_loss=0.09209, over 5677870.93 frames. ], batch size: 285, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:35:41,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3966, 1.9765, 1.4006, 0.6969], device='cuda:0'), covar=tensor([0.2612, 0.1418, 0.2527, 0.2918], device='cuda:0'), in_proj_covar=tensor([0.1454, 0.1380, 0.1422, 0.1195], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 17:35:41,856 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-03 17:35:59,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.843e+02 1.354e+03 1.827e+03 2.423e+03 5.815e+03, threshold=3.655e+03, percent-clipped=5.0
+2023-03-03 17:35:59,945 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:36:03,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:36:12,545 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-03 17:36:35,960 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:36:41,734 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:36:42,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2153, 2.8839, 1.8722, 1.7631], device='cuda:0'), covar=tensor([0.1619, 0.0721, 0.1034, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.1535, 0.1336, 0.1296, 0.1405], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 17:36:46,083 INFO [train.py:968] (0/2) Epoch 7, batch 16800, giga_loss[loss=0.2322, simple_loss=0.3238, pruned_loss=0.07025, over 28882.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3424, pruned_loss=0.09236, over 5681063.83 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3436, pruned_loss=0.09959, over 5771134.26 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3427, pruned_loss=0.09217, over 5672787.31 frames. ], batch size: 145, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:37:51,966 INFO [train.py:968] (0/2) Epoch 7, batch 16850, giga_loss[loss=0.2372, simple_loss=0.3341, pruned_loss=0.07019, over 28771.00 frames. ], tot_loss[loss=0.264, simple_loss=0.343, pruned_loss=0.09251, over 5671408.03 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3438, pruned_loss=0.09977, over 5765146.63 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3431, pruned_loss=0.09203, over 5667287.67 frames. ], batch size: 174, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:38:11,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.977e+02 1.297e+03 1.700e+03 2.324e+03 7.961e+03, threshold=3.400e+03, percent-clipped=12.0
+2023-03-03 17:38:21,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4828, 1.8190, 1.7925, 1.4506], device='cuda:0'), covar=tensor([0.1589, 0.2041, 0.1217, 0.1467], device='cuda:0'), in_proj_covar=tensor([0.0787, 0.0701, 0.0808, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 17:39:01,854 INFO [train.py:968] (0/2) Epoch 7, batch 16900, giga_loss[loss=0.289, simple_loss=0.3625, pruned_loss=0.1078, over 28936.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3432, pruned_loss=0.09223, over 5677161.33 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3437, pruned_loss=0.09959, over 5763966.55 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3433, pruned_loss=0.09181, over 5672367.33 frames. ], batch size: 186, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:39:41,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3971, 1.4091, 1.4846, 1.2325], device='cuda:0'), covar=tensor([0.1807, 0.2601, 0.1468, 0.1524], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0698, 0.0806, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 17:39:58,012 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-03 17:40:13,083 INFO [train.py:968] (0/2) Epoch 7, batch 16950, giga_loss[loss=0.2957, simple_loss=0.3682, pruned_loss=0.1116, over 27621.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3467, pruned_loss=0.09398, over 5678244.54 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3432, pruned_loss=0.09929, over 5765323.24 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3474, pruned_loss=0.09388, over 5672368.72 frames. ], batch size: 472, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:40:35,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.850e+02 1.334e+03 1.798e+03 2.433e+03 7.414e+03, threshold=3.596e+03, percent-clipped=5.0
+2023-03-03 17:41:26,495 INFO [train.py:968] (0/2) Epoch 7, batch 17000, giga_loss[loss=0.3199, simple_loss=0.3654, pruned_loss=0.1372, over 24507.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3464, pruned_loss=0.09358, over 5669735.74 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3434, pruned_loss=0.0995, over 5756474.06 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3467, pruned_loss=0.09326, over 5672191.20 frames. ], batch size: 705, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:42:17,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290688.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:42:25,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5610, 1.9150, 1.9266, 1.5368], device='cuda:0'), covar=tensor([0.1598, 0.1777, 0.1212, 0.1346], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0695, 0.0801, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 17:42:33,580 INFO [train.py:968] (0/2) Epoch 7, batch 17050, giga_loss[loss=0.2335, simple_loss=0.3182, pruned_loss=0.07443, over 28951.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3452, pruned_loss=0.09405, over 5685122.45 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3434, pruned_loss=0.09945, over 5758428.26 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3455, pruned_loss=0.09376, over 5683866.01 frames. ], batch size: 227, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:42:50,875 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:42:53,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.322e+02 1.387e+03 2.058e+03 3.130e+03 7.809e+03, threshold=4.117e+03, percent-clipped=14.0
+2023-03-03 17:43:09,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4255, 1.5152, 1.5821, 1.3626], device='cuda:0'), covar=tensor([0.1050, 0.1651, 0.1614, 0.1573], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0715, 0.0631, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 17:43:24,824 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-03 17:43:44,670 INFO [train.py:968] (0/2) Epoch 7, batch 17100, libri_loss[loss=0.2478, simple_loss=0.3257, pruned_loss=0.08497, over 29502.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3423, pruned_loss=0.09228, over 5686304.05 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3431, pruned_loss=0.0993, over 5758912.89 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3428, pruned_loss=0.09203, over 5683051.74 frames. ], batch size: 81, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:44:09,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:44:53,257 INFO [train.py:968] (0/2) Epoch 7, batch 17150, giga_loss[loss=0.2808, simple_loss=0.3479, pruned_loss=0.1069, over 26792.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3398, pruned_loss=0.08987, over 5696964.70 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3432, pruned_loss=0.09927, over 5760308.13 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3401, pruned_loss=0.08958, over 5692206.50 frames. ], batch size: 555, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:45:09,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.856e+02 1.141e+03 1.573e+03 2.249e+03 7.770e+03, threshold=3.146e+03, percent-clipped=7.0
+2023-03-03 17:45:16,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:45:32,254 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290831.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:45:35,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290834.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:45:52,860 INFO [train.py:968] (0/2) Epoch 7, batch 17200, libri_loss[loss=0.2685, simple_loss=0.3375, pruned_loss=0.09974, over 29550.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3409, pruned_loss=0.0911, over 5691164.58 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3428, pruned_loss=0.09904, over 5762221.44 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3414, pruned_loss=0.09088, over 5683902.50 frames. ], batch size: 75, lr: 4.65e-03, grad_scale: 8.0
+2023-03-03 17:46:06,814 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290863.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:46:32,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0436, 3.8927, 3.6032, 1.7986], device='cuda:0'), covar=tensor([0.0490, 0.0625, 0.0747, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0844, 0.0758, 0.0604], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 17:46:48,694 INFO [train.py:968] (0/2) Epoch 7, batch 17250, giga_loss[loss=0.3149, simple_loss=0.3782, pruned_loss=0.1258, over 26866.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3439, pruned_loss=0.09302, over 5690694.68 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3427, pruned_loss=0.09898, over 5763769.00 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3444, pruned_loss=0.09272, over 5681369.46 frames. ], batch size: 555, lr: 4.65e-03, grad_scale: 8.0
+2023-03-03 17:46:59,978 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:47:02,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:47:06,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.461e+02 1.378e+03 1.911e+03 2.608e+03 4.860e+03, threshold=3.823e+03, percent-clipped=10.0
+2023-03-03 17:47:09,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8391, 1.7844, 1.2492, 1.4012], device='cuda:0'), covar=tensor([0.0632, 0.0520, 0.0941, 0.0956], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0434, 0.0499, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 17:47:34,126 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:47:46,301 INFO [train.py:968] (0/2) Epoch 7, batch 17300, giga_loss[loss=0.3253, simple_loss=0.3709, pruned_loss=0.1399, over 26936.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3446, pruned_loss=0.09416, over 5685351.26 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3426, pruned_loss=0.09889, over 5765174.56 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3451, pruned_loss=0.09396, over 5676074.90 frames. ], batch size: 555, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:47:57,890 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:48:01,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:48:34,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:48:41,058 INFO [train.py:968] (0/2) Epoch 7, batch 17350, giga_loss[loss=0.274, simple_loss=0.3434, pruned_loss=0.1023, over 29037.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3427, pruned_loss=0.09433, over 5684864.36 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3426, pruned_loss=0.0988, over 5765436.61 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3432, pruned_loss=0.09414, over 5675140.63 frames. ], batch size: 128, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:49:00,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.512e+02 1.522e+03 2.075e+03 3.491e+03 8.018e+03, threshold=4.151e+03, percent-clipped=17.0
+2023-03-03 17:49:38,918 INFO [train.py:968] (0/2) Epoch 7, batch 17400, giga_loss[loss=0.2496, simple_loss=0.3324, pruned_loss=0.0834, over 28712.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.342, pruned_loss=0.09447, over 5686489.92 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3426, pruned_loss=0.09878, over 5766608.20 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3423, pruned_loss=0.09427, over 5676691.08 frames. ], batch size: 119, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:49:40,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6045, 4.5837, 1.6326, 1.6594], device='cuda:0'), covar=tensor([0.0898, 0.0295, 0.0907, 0.1285], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0481, 0.0320, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 17:50:16,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=291086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:50:27,843 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4436, 1.8814, 1.3781, 1.2389], device='cuda:0'), covar=tensor([0.1511, 0.1067, 0.1158, 0.1322], device='cuda:0'), in_proj_covar=tensor([0.1522, 0.1326, 0.1287, 0.1419], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 17:50:30,721 INFO [train.py:968] (0/2) Epoch 7, batch 17450, giga_loss[loss=0.2979, simple_loss=0.371, pruned_loss=0.1124, over 28378.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3472, pruned_loss=0.09819, over 5689864.88 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3419, pruned_loss=0.0984, over 5771198.00 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3483, pruned_loss=0.09831, over 5675400.73 frames. ], batch size: 368, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:50:47,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.957e+02 1.457e+03 1.863e+03 2.534e+03 7.138e+03, threshold=3.725e+03, percent-clipped=6.0
+2023-03-03 17:51:03,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2654, 1.5267, 1.3148, 1.0511], device='cuda:0'), covar=tensor([0.1483, 0.1146, 0.0866, 0.1218], device='cuda:0'), in_proj_covar=tensor([0.1528, 0.1328, 0.1292, 0.1425], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 17:51:17,521 INFO [train.py:968] (0/2) Epoch 7, batch 17500, giga_loss[loss=0.3268, simple_loss=0.3965, pruned_loss=0.1286, over 28512.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3567, pruned_loss=0.1041, over 5694551.60 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3416, pruned_loss=0.09806, over 5772280.37 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3581, pruned_loss=0.1046, over 5679512.92 frames. ], batch size: 336, lr: 4.65e-03, grad_scale: 2.0
+2023-03-03 17:51:23,616 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-03 17:51:57,948 INFO [train.py:968] (0/2) Epoch 7, batch 17550, giga_loss[loss=0.2722, simple_loss=0.3483, pruned_loss=0.09804, over 28562.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3624, pruned_loss=0.1074, over 5706365.41 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3416, pruned_loss=0.09777, over 5777136.01 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3642, pruned_loss=0.1084, over 5687100.79 frames. ], batch size: 60, lr: 4.65e-03, grad_scale: 2.0
+2023-03-03 17:52:14,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.420e+02 1.168e+03 1.477e+03 2.091e+03 1.086e+04, threshold=2.954e+03, percent-clipped=7.0
+2023-03-03 17:52:24,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=291229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:52:26,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=291232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:52:43,837 INFO [train.py:968] (0/2) Epoch 7, batch 17600, giga_loss[loss=0.2612, simple_loss=0.3389, pruned_loss=0.09176, over 28715.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3616, pruned_loss=0.1085, over 5692646.08 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3419, pruned_loss=0.09786, over 5767689.95 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3631, pruned_loss=0.1093, over 5684966.98 frames. ], batch size: 262, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:52:54,427 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=291261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:53:26,387 INFO [train.py:968] (0/2) Epoch 7, batch 17650, libri_loss[loss=0.3006, simple_loss=0.3684, pruned_loss=0.1164, over 29527.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3545, pruned_loss=0.1057, over 5689050.58 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3423, pruned_loss=0.09809, over 5765608.58 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3556, pruned_loss=0.1064, over 5683313.03 frames. ], batch size: 82, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:53:26,606 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0585, 3.8603, 3.6321, 1.7930], device='cuda:0'), covar=tensor([0.0481, 0.0645, 0.0675, 0.2125], device='cuda:0'), in_proj_covar=tensor([0.0913, 0.0855, 0.0769, 0.0611], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 17:53:42,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.571e+02 1.103e+03 1.411e+03 1.824e+03 3.284e+03, threshold=2.823e+03, percent-clipped=4.0
+2023-03-03 17:53:45,537 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 17:54:10,707 INFO [train.py:968] (0/2) Epoch 7, batch 17700, libri_loss[loss=0.2875, simple_loss=0.3559, pruned_loss=0.1096, over 29544.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3475, pruned_loss=0.1026, over 5678918.11 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3427, pruned_loss=0.09827, over 5759131.56 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3481, pruned_loss=0.1031, over 5678545.65 frames. ], batch size: 77, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:54:53,461 INFO [train.py:968] (0/2) Epoch 7, batch 17750, giga_loss[loss=0.2425, simple_loss=0.3047, pruned_loss=0.0902, over 28652.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3396, pruned_loss=0.09912, over 5674863.83 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3429, pruned_loss=0.09846, over 5749410.24 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3399, pruned_loss=0.09939, over 5680820.64 frames. ], batch size: 92, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:55:07,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.821e+02 1.115e+03 1.433e+03 1.701e+03 1.040e+04, threshold=2.866e+03, percent-clipped=6.0
+2023-03-03 17:55:30,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4088, 1.6066, 1.3286, 1.6543], device='cuda:0'), covar=tensor([0.1953, 0.1780, 0.1831, 0.1803], device='cuda:0'), in_proj_covar=tensor([0.1169, 0.0890, 0.1041, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 17:55:38,134 INFO [train.py:968] (0/2) Epoch 7, batch 17800, giga_loss[loss=0.2689, simple_loss=0.3133, pruned_loss=0.1122, over 23973.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3334, pruned_loss=0.09671, over 5677096.54 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3434, pruned_loss=0.09874, over 5750769.79 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3331, pruned_loss=0.09662, over 5679659.16 frames. ], batch size: 705, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:55:43,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4435, 2.9930, 1.5762, 1.4367], device='cuda:0'), covar=tensor([0.0850, 0.0313, 0.0803, 0.1231], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0480, 0.0316, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0026, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 17:55:44,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4133, 1.2924, 4.9983, 3.3762], device='cuda:0'), covar=tensor([0.1635, 0.2519, 0.0285, 0.0732], device='cuda:0'), in_proj_covar=tensor([0.0585, 0.0549, 0.0774, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 17:55:47,682 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=291462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:56:16,499 INFO [train.py:968] (0/2) Epoch 7, batch 17850, giga_loss[loss=0.2765, simple_loss=0.337, pruned_loss=0.108, over 29007.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3281, pruned_loss=0.09395, over 5680029.28 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3437, pruned_loss=0.09885, over 5744840.96 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3274, pruned_loss=0.09371, over 5686169.31 frames. ], batch size: 128, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:56:32,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.893e+02 1.124e+03 1.394e+03 2.190e+03 5.352e+03, threshold=2.789e+03, percent-clipped=15.0
+2023-03-03 17:56:59,366 INFO [train.py:968] (0/2) Epoch 7, batch 17900, giga_loss[loss=0.2179, simple_loss=0.2847, pruned_loss=0.07554, over 28610.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3266, pruned_loss=0.09338, over 5685049.30 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3445, pruned_loss=0.09916, over 5741433.95 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3249, pruned_loss=0.09279, over 5691480.03 frames. ], batch size: 85, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:57:45,959 INFO [train.py:968] (0/2) Epoch 7, batch 17950, giga_loss[loss=0.2449, simple_loss=0.3173, pruned_loss=0.08621, over 28873.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3226, pruned_loss=0.09156, over 5683956.58 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3447, pruned_loss=0.09927, over 5743108.43 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3208, pruned_loss=0.09091, over 5686943.99 frames. ], batch size: 199, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:57:57,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.053e+02 1.032e+03 1.215e+03 1.856e+03 5.555e+03, threshold=2.430e+03, percent-clipped=7.0
+2023-03-03 17:58:15,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0295, 1.3021, 3.6479, 3.0323], device='cuda:0'), covar=tensor([0.1654, 0.2441, 0.0405, 0.0808], device='cuda:0'), in_proj_covar=tensor([0.0584, 0.0546, 0.0771, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 17:58:24,955 INFO [train.py:968] (0/2) Epoch 7, batch 18000, libri_loss[loss=0.2795, simple_loss=0.3434, pruned_loss=0.1078, over 28617.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.32, pruned_loss=0.08982, over 5690580.23 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3455, pruned_loss=0.09967, over 5736820.41 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3169, pruned_loss=0.08857, over 5697703.53 frames. ], batch size: 63, lr: 4.65e-03, grad_scale: 8.0
+2023-03-03 17:58:24,960 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 17:58:31,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8009, 1.1327, 3.4420, 2.9601], device='cuda:0'), covar=tensor([0.2093, 0.2795, 0.0494, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0586, 0.0547, 0.0773, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 17:58:33,310 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.224, simple_loss=0.3277, pruned_loss=0.06019, over 944034.00 frames. 
+2023-03-03 17:58:33,311 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-03 17:59:00,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3873, 1.6521, 1.3665, 1.2940], device='cuda:0'), covar=tensor([0.2005, 0.1845, 0.1938, 0.1880], device='cuda:0'), in_proj_covar=tensor([0.1182, 0.0897, 0.1051, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 17:59:18,794 INFO [train.py:968] (0/2) Epoch 7, batch 18050, giga_loss[loss=0.2312, simple_loss=0.302, pruned_loss=0.08015, over 28969.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3161, pruned_loss=0.08811, over 5678820.21 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3454, pruned_loss=0.09953, over 5738483.03 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3135, pruned_loss=0.08715, over 5682232.67 frames. ], batch size: 213, lr: 4.65e-03, grad_scale: 8.0
+2023-03-03 17:59:29,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.839e+02 9.775e+02 1.181e+03 1.654e+03 4.104e+03, threshold=2.362e+03, percent-clipped=8.0
+2023-03-03 17:59:58,827 INFO [train.py:968] (0/2) Epoch 7, batch 18100, libri_loss[loss=0.2847, simple_loss=0.3647, pruned_loss=0.1023, over 29648.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3139, pruned_loss=0.08705, over 5689563.58 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3452, pruned_loss=0.09934, over 5744155.28 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.311, pruned_loss=0.08607, over 5685624.84 frames. ], batch size: 91, lr: 4.65e-03, grad_scale: 8.0
+2023-03-03 18:00:39,546 INFO [train.py:968] (0/2) Epoch 7, batch 18150, giga_loss[loss=0.2269, simple_loss=0.2957, pruned_loss=0.07906, over 28626.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3113, pruned_loss=0.08585, over 5694253.24 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3454, pruned_loss=0.09927, over 5746604.79 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.3078, pruned_loss=0.08475, over 5687441.40 frames. ], batch size: 336, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 18:00:54,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.857e+02 1.109e+03 1.429e+03 2.439e+03 7.157e+03, threshold=2.859e+03, percent-clipped=24.0
+2023-03-03 18:01:16,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=291837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:01:29,762 INFO [train.py:968] (0/2) Epoch 7, batch 18200, giga_loss[loss=0.2252, simple_loss=0.2956, pruned_loss=0.07743, over 27889.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3075, pruned_loss=0.08419, over 5678545.02 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3456, pruned_loss=0.09937, over 5743340.77 frames. ], giga_tot_loss[loss=0.2354, simple_loss=0.3044, pruned_loss=0.08316, over 5676005.17 frames. ], batch size: 412, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 18:01:48,431 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-03 18:02:13,136 INFO [train.py:968] (0/2) Epoch 7, batch 18250, giga_loss[loss=0.2181, simple_loss=0.284, pruned_loss=0.07612, over 28069.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3056, pruned_loss=0.08331, over 5681081.48 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3457, pruned_loss=0.09945, over 5744482.10 frames. ], giga_tot_loss[loss=0.2334, simple_loss=0.3024, pruned_loss=0.08216, over 5677140.02 frames. ], batch size: 77, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:02:28,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.666e+02 9.855e+02 1.385e+03 1.874e+03 3.948e+03, threshold=2.769e+03, percent-clipped=9.0
+2023-03-03 18:03:03,353 INFO [train.py:968] (0/2) Epoch 7, batch 18300, giga_loss[loss=0.2902, simple_loss=0.3616, pruned_loss=0.1094, over 28604.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3157, pruned_loss=0.0889, over 5675635.48 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3459, pruned_loss=0.09948, over 5746452.41 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3123, pruned_loss=0.08773, over 5669302.42 frames. ], batch size: 78, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:03:31,963 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=291980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:03:36,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=291983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:03:51,632 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-292000.pt
+2023-03-03 18:03:51,930 INFO [train.py:968] (0/2) Epoch 7, batch 18350, giga_loss[loss=0.3102, simple_loss=0.3857, pruned_loss=0.1173, over 28920.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3311, pruned_loss=0.09753, over 5672613.07 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3463, pruned_loss=0.09966, over 5737761.57 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.328, pruned_loss=0.0964, over 5675581.72 frames. ], batch size: 145, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:04:00,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=292012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:04:05,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.037e+02 1.159e+03 1.548e+03 2.017e+03 4.465e+03, threshold=3.095e+03, percent-clipped=8.0
+2023-03-03 18:04:06,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8719, 1.7133, 1.2969, 1.3522], device='cuda:0'), covar=tensor([0.0641, 0.0602, 0.0930, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0435, 0.0496, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 18:04:29,936 INFO [train.py:968] (0/2) Epoch 7, batch 18400, giga_loss[loss=0.2899, simple_loss=0.3673, pruned_loss=0.1063, over 28827.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3425, pruned_loss=0.1033, over 5683377.29 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3464, pruned_loss=0.09962, over 5739857.13 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3398, pruned_loss=0.1024, over 5682786.19 frames. ], batch size: 174, lr: 4.64e-03, grad_scale: 8.0
+2023-03-03 18:05:16,061 INFO [train.py:968] (0/2) Epoch 7, batch 18450, giga_loss[loss=0.2701, simple_loss=0.3528, pruned_loss=0.09372, over 28870.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3506, pruned_loss=0.1069, over 5677988.01 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3464, pruned_loss=0.09956, over 5740757.84 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3485, pruned_loss=0.1064, over 5676391.72 frames. ], batch size: 199, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:05:29,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.859e+02 1.157e+03 1.501e+03 2.026e+03 1.419e+04, threshold=3.002e+03, percent-clipped=7.0
+2023-03-03 18:05:37,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1442, 1.2821, 1.0422, 1.0521], device='cuda:0'), covar=tensor([0.0986, 0.1011, 0.0727, 0.0866], device='cuda:0'), in_proj_covar=tensor([0.1534, 0.1355, 0.1322, 0.1437], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 18:05:50,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1661, 1.7138, 1.2847, 0.3986], device='cuda:0'), covar=tensor([0.2526, 0.1541, 0.2506, 0.2920], device='cuda:0'), in_proj_covar=tensor([0.1440, 0.1363, 0.1408, 0.1186], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 18:05:54,964 INFO [train.py:968] (0/2) Epoch 7, batch 18500, libri_loss[loss=0.2547, simple_loss=0.3443, pruned_loss=0.08255, over 29506.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3522, pruned_loss=0.1058, over 5690051.38 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3472, pruned_loss=0.09998, over 5745682.65 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3499, pruned_loss=0.1052, over 5682925.10 frames. ], batch size: 81, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:06:13,946 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=292169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:06:34,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1947, 1.3519, 1.3785, 1.4227], device='cuda:0'), covar=tensor([0.0733, 0.0366, 0.0283, 0.0777], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0067, 0.0048, 0.0044, 0.0074], device='cuda:0')
+2023-03-03 18:06:41,319 INFO [train.py:968] (0/2) Epoch 7, batch 18550, giga_loss[loss=0.2989, simple_loss=0.3615, pruned_loss=0.1181, over 28747.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3546, pruned_loss=0.1065, over 5678099.94 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3475, pruned_loss=0.09985, over 5750914.19 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3527, pruned_loss=0.1063, over 5665615.08 frames. ], batch size: 92, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:06:57,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.347e+02 1.056e+03 1.239e+03 1.777e+03 4.947e+03, threshold=2.479e+03, percent-clipped=4.0
+2023-03-03 18:07:25,631 INFO [train.py:968] (0/2) Epoch 7, batch 18600, giga_loss[loss=0.3539, simple_loss=0.4027, pruned_loss=0.1525, over 28582.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3579, pruned_loss=0.1093, over 5670109.00 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3483, pruned_loss=0.1004, over 5741734.73 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3558, pruned_loss=0.1087, over 5666991.07 frames. ], batch size: 336, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:08:09,560 INFO [train.py:968] (0/2) Epoch 7, batch 18650, giga_loss[loss=0.2957, simple_loss=0.365, pruned_loss=0.1132, over 28530.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3611, pruned_loss=0.1118, over 5674186.23 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3489, pruned_loss=0.1006, over 5745431.63 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3591, pruned_loss=0.1115, over 5666392.48 frames. ], batch size: 85, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:08:25,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.729e+02 1.116e+03 1.517e+03 1.994e+03 5.307e+03, threshold=3.034e+03, percent-clipped=17.0
+2023-03-03 18:08:54,142 INFO [train.py:968] (0/2) Epoch 7, batch 18700, giga_loss[loss=0.2956, simple_loss=0.3694, pruned_loss=0.1109, over 28786.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3639, pruned_loss=0.1131, over 5676663.36 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3491, pruned_loss=0.1007, over 5746236.47 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3623, pruned_loss=0.1128, over 5669623.43 frames. ], batch size: 119, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:09:34,596 INFO [train.py:968] (0/2) Epoch 7, batch 18750, giga_loss[loss=0.3062, simple_loss=0.376, pruned_loss=0.1182, over 28770.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3671, pruned_loss=0.1141, over 5685003.77 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3496, pruned_loss=0.1009, over 5747967.16 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3656, pruned_loss=0.1138, over 5677336.70 frames. ], batch size: 119, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:09:49,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.190e+02 1.120e+03 1.453e+03 1.969e+03 3.865e+03, threshold=2.907e+03, percent-clipped=5.0
+2023-03-03 18:09:50,840 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-03 18:10:15,152 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=292447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:10:17,153 INFO [train.py:968] (0/2) Epoch 7, batch 18800, giga_loss[loss=0.3086, simple_loss=0.3803, pruned_loss=0.1184, over 28953.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3687, pruned_loss=0.1143, over 5686574.47 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3495, pruned_loss=0.1008, over 5750268.25 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3678, pruned_loss=0.1144, over 5677627.37 frames. ], batch size: 145, lr: 4.64e-03, grad_scale: 8.0
+2023-03-03 18:10:58,495 INFO [train.py:968] (0/2) Epoch 7, batch 18850, giga_loss[loss=0.2929, simple_loss=0.3673, pruned_loss=0.1092, over 28709.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3701, pruned_loss=0.1143, over 5689353.94 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3506, pruned_loss=0.1014, over 5751828.01 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3689, pruned_loss=0.1142, over 5679217.09 frames. ], batch size: 242, lr: 4.64e-03, grad_scale: 8.0
+2023-03-03 18:11:10,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.475e+02 1.228e+03 1.588e+03 2.180e+03 7.692e+03, threshold=3.175e+03, percent-clipped=15.0
+2023-03-03 18:11:30,296 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=292544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:11:34,683 INFO [train.py:968] (0/2) Epoch 7, batch 18900, giga_loss[loss=0.275, simple_loss=0.3513, pruned_loss=0.09938, over 28918.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3692, pruned_loss=0.1122, over 5688725.42 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3509, pruned_loss=0.1015, over 5736964.34 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3684, pruned_loss=0.1123, over 5692572.92 frames. ], batch size: 106, lr: 4.64e-03, grad_scale: 8.0
+2023-03-03 18:11:45,944 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 18:12:16,392 INFO [train.py:968] (0/2) Epoch 7, batch 18950, giga_loss[loss=0.2843, simple_loss=0.3602, pruned_loss=0.1042, over 28869.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3666, pruned_loss=0.1095, over 5697028.70 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3509, pruned_loss=0.1014, over 5738519.24 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3662, pruned_loss=0.1097, over 5698422.24 frames. ], batch size: 186, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:12:33,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.623e+02 1.006e+03 1.233e+03 1.619e+03 4.174e+03, threshold=2.467e+03, percent-clipped=2.0
+2023-03-03 18:12:56,939 INFO [train.py:968] (0/2) Epoch 7, batch 19000, giga_loss[loss=0.2681, simple_loss=0.3505, pruned_loss=0.09282, over 28951.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3672, pruned_loss=0.1102, over 5704712.14 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3519, pruned_loss=0.1019, over 5743470.51 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3665, pruned_loss=0.1102, over 5699722.27 frames. ], batch size: 145, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:13:31,379 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=292687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:13:33,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=292690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:13:40,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2617, 1.9649, 1.5968, 0.5190], device='cuda:0'), covar=tensor([0.2512, 0.1613, 0.2104, 0.2908], device='cuda:0'), in_proj_covar=tensor([0.1441, 0.1355, 0.1406, 0.1189], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 18:13:42,406 INFO [train.py:968] (0/2) Epoch 7, batch 19050, libri_loss[loss=0.309, simple_loss=0.3957, pruned_loss=0.1111, over 29319.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3695, pruned_loss=0.1138, over 5712274.59 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3525, pruned_loss=0.1021, over 5747278.19 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3688, pruned_loss=0.1139, over 5703668.42 frames. ], batch size: 94, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:13:48,143 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-03 18:13:57,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9604, 1.8747, 1.3663, 1.4978], device='cuda:0'), covar=tensor([0.0686, 0.0607, 0.0991, 0.0958], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0432, 0.0494, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 18:13:58,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.035e+02 1.450e+03 1.803e+03 2.343e+03 5.646e+03, threshold=3.606e+03, percent-clipped=21.0
+2023-03-03 18:13:59,240 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=292719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:14:27,268 INFO [train.py:968] (0/2) Epoch 7, batch 19100, giga_loss[loss=0.3022, simple_loss=0.3599, pruned_loss=0.1223, over 28550.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3715, pruned_loss=0.118, over 5713869.03 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3521, pruned_loss=0.1018, over 5748952.30 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3717, pruned_loss=0.1186, over 5704774.07 frames. ], batch size: 71, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:15:05,125 INFO [train.py:968] (0/2) Epoch 7, batch 19150, giga_loss[loss=0.3286, simple_loss=0.385, pruned_loss=0.1361, over 28757.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3703, pruned_loss=0.1182, over 5698685.51 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3522, pruned_loss=0.1019, over 5735111.21 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3709, pruned_loss=0.1192, over 5703068.98 frames. ], batch size: 242, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:15:21,527 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.219e+02 1.182e+03 1.567e+03 2.333e+03 4.434e+03, threshold=3.134e+03, percent-clipped=6.0
+2023-03-03 18:15:24,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=292822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:15:37,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4347, 1.5139, 1.5236, 1.4374], device='cuda:0'), covar=tensor([0.1066, 0.1647, 0.1467, 0.1507], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0725, 0.0642, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 18:15:48,842 INFO [train.py:968] (0/2) Epoch 7, batch 19200, giga_loss[loss=0.2849, simple_loss=0.3576, pruned_loss=0.106, over 28368.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3672, pruned_loss=0.117, over 5696099.00 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3525, pruned_loss=0.102, over 5738732.60 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3679, pruned_loss=0.118, over 5695515.67 frames. ], batch size: 65, lr: 4.64e-03, grad_scale: 8.0
+2023-03-03 18:16:07,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4429, 4.2146, 4.0395, 1.8998], device='cuda:0'), covar=tensor([0.0488, 0.0603, 0.0675, 0.1968], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0854, 0.0767, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 18:16:16,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3070, 3.1021, 2.9442, 1.3734], device='cuda:0'), covar=tensor([0.0815, 0.0858, 0.0858, 0.2311], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0854, 0.0766, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 18:16:32,656 INFO [train.py:968] (0/2) Epoch 7, batch 19250, giga_loss[loss=0.3037, simple_loss=0.3701, pruned_loss=0.1187, over 28868.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3668, pruned_loss=0.1169, over 5705645.59 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3527, pruned_loss=0.1021, over 5742377.82 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3674, pruned_loss=0.118, over 5701172.02 frames. ], batch size: 199, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:16:50,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.517e+02 1.206e+03 1.627e+03 2.404e+03 8.199e+03, threshold=3.253e+03, percent-clipped=17.0
+2023-03-03 18:16:53,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3085, 1.6152, 1.5699, 1.3023], device='cuda:0'), covar=tensor([0.1247, 0.1607, 0.0999, 0.1147], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0705, 0.0811, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 18:17:13,133 INFO [train.py:968] (0/2) Epoch 7, batch 19300, giga_loss[loss=0.2614, simple_loss=0.3407, pruned_loss=0.09107, over 28553.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3658, pruned_loss=0.1154, over 5702492.35 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.353, pruned_loss=0.1023, over 5732656.82 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3665, pruned_loss=0.1166, over 5706171.74 frames. ], batch size: 71, lr: 4.64e-03, grad_scale: 2.0
+2023-03-03 18:17:26,501 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=292965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:17:28,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=292968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:17:34,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6779, 3.4999, 3.2836, 1.7962], device='cuda:0'), covar=tensor([0.0600, 0.0696, 0.0683, 0.2293], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0857, 0.0766, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 18:17:54,373 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=292997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:17:56,442 INFO [train.py:968] (0/2) Epoch 7, batch 19350, giga_loss[loss=0.2859, simple_loss=0.3546, pruned_loss=0.1087, over 28819.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3637, pruned_loss=0.1134, over 5701243.46 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.353, pruned_loss=0.102, over 5738559.06 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3646, pruned_loss=0.1149, over 5698091.74 frames. ], batch size: 99, lr: 4.64e-03, grad_scale: 2.0
+2023-03-03 18:18:01,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8923, 1.1077, 3.2538, 2.8302], device='cuda:0'), covar=tensor([0.1570, 0.2322, 0.0429, 0.1127], device='cuda:0'), in_proj_covar=tensor([0.0586, 0.0546, 0.0775, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:18:13,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.006e+02 9.962e+02 1.248e+03 1.545e+03 3.638e+03, threshold=2.495e+03, percent-clipped=1.0
+2023-03-03 18:18:24,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7048, 1.6717, 1.4590, 2.1548], device='cuda:0'), covar=tensor([0.2078, 0.2092, 0.2111, 0.1891], device='cuda:0'), in_proj_covar=tensor([0.1179, 0.0892, 0.1043, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:18:40,293 INFO [train.py:968] (0/2) Epoch 7, batch 19400, giga_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.113, over 28952.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3592, pruned_loss=0.1106, over 5695653.87 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3535, pruned_loss=0.1022, over 5741894.12 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3598, pruned_loss=0.112, over 5688843.45 frames. ], batch size: 136, lr: 4.64e-03, grad_scale: 2.0
+2023-03-03 18:19:20,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3808, 1.7953, 1.2673, 1.5055], device='cuda:0'), covar=tensor([0.0746, 0.0275, 0.0331, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0121, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0068, 0.0048, 0.0044, 0.0074], device='cuda:0')
+2023-03-03 18:19:28,277 INFO [train.py:968] (0/2) Epoch 7, batch 19450, giga_loss[loss=0.2614, simple_loss=0.3345, pruned_loss=0.0941, over 28969.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.352, pruned_loss=0.1067, over 5688087.66 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3533, pruned_loss=0.1021, over 5743504.09 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3527, pruned_loss=0.108, over 5680828.41 frames. ], batch size: 227, lr: 4.64e-03, grad_scale: 2.0
+2023-03-03 18:19:46,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.937e+02 9.177e+02 1.237e+03 1.736e+03 8.069e+03, threshold=2.474e+03, percent-clipped=9.0
+2023-03-03 18:20:18,629 INFO [train.py:968] (0/2) Epoch 7, batch 19500, giga_loss[loss=0.247, simple_loss=0.3248, pruned_loss=0.08455, over 28837.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3461, pruned_loss=0.1037, over 5671577.11 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3538, pruned_loss=0.1023, over 5745120.03 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3462, pruned_loss=0.1046, over 5664087.90 frames. ], batch size: 145, lr: 4.63e-03, grad_scale: 2.0
+2023-03-03 18:21:05,991 INFO [train.py:968] (0/2) Epoch 7, batch 19550, giga_loss[loss=0.2535, simple_loss=0.3369, pruned_loss=0.08506, over 29044.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3454, pruned_loss=0.1033, over 5660597.68 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3544, pruned_loss=0.1027, over 5747390.02 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3447, pruned_loss=0.1036, over 5650642.88 frames. ], batch size: 155, lr: 4.63e-03, grad_scale: 2.0
+2023-03-03 18:21:15,618 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6379, 1.0623, 2.8647, 2.7496], device='cuda:0'), covar=tensor([0.1760, 0.2424, 0.0528, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0585, 0.0551, 0.0772, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:21:21,300 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.880e+02 1.138e+03 1.503e+03 2.459e+03 1.389e+04, threshold=3.006e+03, percent-clipped=23.0
+2023-03-03 18:21:45,747 INFO [train.py:968] (0/2) Epoch 7, batch 19600, giga_loss[loss=0.2705, simple_loss=0.3417, pruned_loss=0.0997, over 28439.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3472, pruned_loss=0.1045, over 5671372.00 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3551, pruned_loss=0.103, over 5751123.03 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3457, pruned_loss=0.1045, over 5657507.92 frames. ], batch size: 65, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:22:11,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3277, 1.4167, 1.3111, 1.0694], device='cuda:0'), covar=tensor([0.1384, 0.1185, 0.0731, 0.1181], device='cuda:0'), in_proj_covar=tensor([0.1519, 0.1350, 0.1328, 0.1455], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 18:22:28,396 INFO [train.py:968] (0/2) Epoch 7, batch 19650, libri_loss[loss=0.3188, simple_loss=0.3988, pruned_loss=0.1194, over 29482.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3467, pruned_loss=0.1042, over 5680009.57 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3559, pruned_loss=0.1035, over 5753785.55 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3445, pruned_loss=0.1038, over 5664369.46 frames. ], batch size: 85, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:22:44,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.015e+02 1.054e+03 1.329e+03 1.788e+03 7.970e+03, threshold=2.658e+03, percent-clipped=10.0
+2023-03-03 18:23:10,102 INFO [train.py:968] (0/2) Epoch 7, batch 19700, giga_loss[loss=0.2564, simple_loss=0.3203, pruned_loss=0.09631, over 28735.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3452, pruned_loss=0.1033, over 5685637.25 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3562, pruned_loss=0.1034, over 5755780.62 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.343, pruned_loss=0.1031, over 5669831.95 frames. ], batch size: 92, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:23:16,763 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:23:20,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5147, 5.3230, 5.0660, 2.4420], device='cuda:0'), covar=tensor([0.0306, 0.0408, 0.0486, 0.1800], device='cuda:0'), in_proj_covar=tensor([0.0896, 0.0845, 0.0759, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 18:23:23,207 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-03 18:23:51,311 INFO [train.py:968] (0/2) Epoch 7, batch 19750, giga_loss[loss=0.2403, simple_loss=0.3105, pruned_loss=0.08509, over 28897.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.342, pruned_loss=0.1016, over 5680387.42 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3563, pruned_loss=0.1035, over 5746937.73 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3401, pruned_loss=0.1014, over 5676159.66 frames. ], batch size: 99, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:24:07,879 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-03 18:24:08,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.419e+02 9.969e+02 1.237e+03 1.747e+03 8.209e+03, threshold=2.473e+03, percent-clipped=7.0
+2023-03-03 18:24:16,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5670, 2.1014, 1.4269, 0.6778], device='cuda:0'), covar=tensor([0.3684, 0.1966, 0.2064, 0.3885], device='cuda:0'), in_proj_covar=tensor([0.1429, 0.1332, 0.1396, 0.1183], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 18:24:30,052 INFO [train.py:968] (0/2) Epoch 7, batch 19800, giga_loss[loss=0.2373, simple_loss=0.3172, pruned_loss=0.07867, over 28744.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.34, pruned_loss=0.1001, over 5689559.62 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3574, pruned_loss=0.104, over 5742109.13 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3371, pruned_loss=0.09936, over 5689055.09 frames. ], batch size: 284, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:24:43,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7898, 1.6597, 1.7398, 1.6487], device='cuda:0'), covar=tensor([0.1220, 0.2135, 0.1786, 0.1759], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0730, 0.0647, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 18:24:49,974 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:25:08,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9328, 1.0851, 0.9978, 0.6503], device='cuda:0'), covar=tensor([0.1119, 0.1173, 0.0725, 0.1235], device='cuda:0'), in_proj_covar=tensor([0.1527, 0.1354, 0.1335, 0.1462], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 18:25:11,534 INFO [train.py:968] (0/2) Epoch 7, batch 19850, libri_loss[loss=0.2565, simple_loss=0.344, pruned_loss=0.08447, over 29549.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.337, pruned_loss=0.09835, over 5695048.42 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3577, pruned_loss=0.1038, over 5745450.44 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.334, pruned_loss=0.09782, over 5690421.52 frames. ], batch size: 77, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:25:18,230 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:25:28,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.567e+02 1.113e+03 1.449e+03 2.163e+03 8.293e+03, threshold=2.897e+03, percent-clipped=14.0
+2023-03-03 18:25:30,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7341, 4.5865, 1.8592, 1.7958], device='cuda:0'), covar=tensor([0.0876, 0.0229, 0.0798, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0479, 0.0315, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0026, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 18:25:50,324 INFO [train.py:968] (0/2) Epoch 7, batch 19900, giga_loss[loss=0.2206, simple_loss=0.2901, pruned_loss=0.07554, over 28806.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3339, pruned_loss=0.09661, over 5709863.90 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3574, pruned_loss=0.1033, over 5749357.53 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3312, pruned_loss=0.09652, over 5701733.60 frames. ], batch size: 99, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:26:28,543 INFO [train.py:968] (0/2) Epoch 7, batch 19950, giga_loss[loss=0.2664, simple_loss=0.3315, pruned_loss=0.1006, over 28957.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3327, pruned_loss=0.09588, over 5715752.06 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3579, pruned_loss=0.1034, over 5748146.02 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3293, pruned_loss=0.09551, over 5709077.17 frames. ], batch size: 213, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:26:46,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.683e+02 9.868e+02 1.305e+03 1.740e+03 8.216e+03, threshold=2.610e+03, percent-clipped=7.0
+2023-03-03 18:27:12,409 INFO [train.py:968] (0/2) Epoch 7, batch 20000, giga_loss[loss=0.2651, simple_loss=0.3281, pruned_loss=0.101, over 28853.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3301, pruned_loss=0.09483, over 5710892.07 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.358, pruned_loss=0.1034, over 5748930.29 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3273, pruned_loss=0.09452, over 5704894.07 frames. ], batch size: 145, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:27:50,632 INFO [train.py:968] (0/2) Epoch 7, batch 20050, giga_loss[loss=0.2313, simple_loss=0.3109, pruned_loss=0.07586, over 28904.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3286, pruned_loss=0.09358, over 5715881.15 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3586, pruned_loss=0.1035, over 5751810.60 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3253, pruned_loss=0.09304, over 5707910.78 frames. ], batch size: 174, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:27:56,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4890, 1.6476, 1.3256, 1.9281], device='cuda:0'), covar=tensor([0.2127, 0.2171, 0.2241, 0.2169], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.0892, 0.1045, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:28:04,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.175e+02 9.671e+02 1.182e+03 1.611e+03 5.349e+03, threshold=2.364e+03, percent-clipped=4.0
+2023-03-03 18:28:15,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=293735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:28:27,336 INFO [train.py:968] (0/2) Epoch 7, batch 20100, giga_loss[loss=0.3186, simple_loss=0.3669, pruned_loss=0.1351, over 26701.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3289, pruned_loss=0.09363, over 5723509.37 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3588, pruned_loss=0.1034, over 5755543.63 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3253, pruned_loss=0.09302, over 5713094.90 frames. ], batch size: 555, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:28:30,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:28:32,315 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:28:37,667 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 18:28:45,214 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 18:28:49,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3061, 1.7909, 1.5093, 1.5878], device='cuda:0'), covar=tensor([0.0799, 0.0288, 0.0302, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0120, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0067, 0.0048, 0.0043, 0.0073], device='cuda:0')
+2023-03-03 18:29:11,855 INFO [train.py:968] (0/2) Epoch 7, batch 20150, giga_loss[loss=0.2635, simple_loss=0.3387, pruned_loss=0.09418, over 28854.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3339, pruned_loss=0.09722, over 5718498.36 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3589, pruned_loss=0.1035, over 5757122.15 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3308, pruned_loss=0.09658, over 5708679.31 frames. ], batch size: 174, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:29:29,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.534e+02 1.043e+03 1.252e+03 1.712e+03 3.377e+03, threshold=2.504e+03, percent-clipped=8.0
+2023-03-03 18:29:58,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=293849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:29:59,529 INFO [train.py:968] (0/2) Epoch 7, batch 20200, giga_loss[loss=0.2926, simple_loss=0.3539, pruned_loss=0.1157, over 28751.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3416, pruned_loss=0.1027, over 5698247.02 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3598, pruned_loss=0.1041, over 5747232.49 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3378, pruned_loss=0.1016, over 5698127.55 frames. ], batch size: 99, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:30:07,377 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-03 18:30:27,395 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=293878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:30:31,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=293881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:30:36,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=293885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:30:51,862 INFO [train.py:968] (0/2) Epoch 7, batch 20250, giga_loss[loss=0.3445, simple_loss=0.3981, pruned_loss=0.1454, over 27622.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3506, pruned_loss=0.1089, over 5683682.26 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.36, pruned_loss=0.1042, over 5740246.56 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3473, pruned_loss=0.1079, over 5688804.32 frames. ], batch size: 472, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:31:00,831 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=293910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:31:10,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.359e+02 1.282e+03 1.632e+03 2.122e+03 4.863e+03, threshold=3.264e+03, percent-clipped=17.0
+2023-03-03 18:31:32,844 INFO [train.py:968] (0/2) Epoch 7, batch 20300, giga_loss[loss=0.3423, simple_loss=0.3981, pruned_loss=0.1433, over 27787.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3564, pruned_loss=0.1116, over 5686991.47 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3605, pruned_loss=0.1044, over 5740547.00 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3531, pruned_loss=0.1108, over 5689369.18 frames. ], batch size: 412, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:31:42,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6554, 0.9862, 2.8148, 2.6917], device='cuda:0'), covar=tensor([0.1651, 0.2332, 0.0506, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0586, 0.0545, 0.0765, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:31:48,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4311, 1.7207, 1.6995, 1.4128], device='cuda:0'), covar=tensor([0.1249, 0.1519, 0.0981, 0.1095], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0709, 0.0816, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 18:31:54,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4567, 1.5620, 1.2594, 1.1629], device='cuda:0'), covar=tensor([0.1373, 0.1161, 0.1115, 0.1269], device='cuda:0'), in_proj_covar=tensor([0.1511, 0.1349, 0.1340, 0.1444], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 18:32:05,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3605, 1.5490, 1.2798, 1.7450], device='cuda:0'), covar=tensor([0.2378, 0.2252, 0.2379, 0.1890], device='cuda:0'), in_proj_covar=tensor([0.1192, 0.0896, 0.1045, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:32:11,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=293992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:32:15,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=293995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:32:19,938 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-294000.pt
+2023-03-03 18:32:20,228 INFO [train.py:968] (0/2) Epoch 7, batch 20350, giga_loss[loss=0.3004, simple_loss=0.3816, pruned_loss=0.1096, over 28992.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3616, pruned_loss=0.1135, over 5687364.27 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3605, pruned_loss=0.1044, over 5744710.76 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3589, pruned_loss=0.113, over 5684596.83 frames. ], batch size: 213, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:32:34,583 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-03 18:32:36,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.676e+02 1.084e+03 1.287e+03 1.664e+03 2.692e+03, threshold=2.575e+03, percent-clipped=0.0
+2023-03-03 18:32:42,310 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:32:45,106 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:32:47,586 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:33:06,640 INFO [train.py:968] (0/2) Epoch 7, batch 20400, libri_loss[loss=0.3386, simple_loss=0.3996, pruned_loss=0.1388, over 20190.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3669, pruned_loss=0.1158, over 5671768.75 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.361, pruned_loss=0.1047, over 5728845.10 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3643, pruned_loss=0.1154, over 5682685.10 frames. ], batch size: 186, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:33:15,317 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:33:45,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5662, 1.6404, 1.3739, 1.7835], device='cuda:0'), covar=tensor([0.2243, 0.2085, 0.2092, 0.2215], device='cuda:0'), in_proj_covar=tensor([0.1182, 0.0892, 0.1037, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:33:47,079 INFO [train.py:968] (0/2) Epoch 7, batch 20450, libri_loss[loss=0.2745, simple_loss=0.3473, pruned_loss=0.1008, over 29595.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3692, pruned_loss=0.1171, over 5680391.83 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3614, pruned_loss=0.1048, over 5728612.60 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.367, pruned_loss=0.1169, over 5688192.58 frames. ], batch size: 74, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:34:05,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.613e+02 1.120e+03 1.597e+03 2.071e+03 4.326e+03, threshold=3.195e+03, percent-clipped=13.0
+2023-03-03 18:34:12,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:34:14,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:34:28,964 INFO [train.py:968] (0/2) Epoch 7, batch 20500, giga_loss[loss=0.2925, simple_loss=0.3606, pruned_loss=0.1122, over 28221.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3624, pruned_loss=0.1125, over 5676395.77 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3614, pruned_loss=0.105, over 5722683.64 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3607, pruned_loss=0.1123, over 5686954.71 frames. ], batch size: 368, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:35:08,376 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-03 18:35:10,783 INFO [train.py:968] (0/2) Epoch 7, batch 20550, giga_loss[loss=0.3276, simple_loss=0.3713, pruned_loss=0.1419, over 23819.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3611, pruned_loss=0.1109, over 5683797.69 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3618, pruned_loss=0.1053, over 5726263.17 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3594, pruned_loss=0.1106, over 5688193.46 frames. ], batch size: 705, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:35:23,493 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 18:35:29,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.135e+02 1.131e+03 1.358e+03 1.721e+03 5.339e+03, threshold=2.717e+03, percent-clipped=5.0
+2023-03-03 18:35:39,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9481, 5.7491, 5.4519, 3.0366], device='cuda:0'), covar=tensor([0.0338, 0.0542, 0.0675, 0.1388], device='cuda:0'), in_proj_covar=tensor([0.0905, 0.0855, 0.0765, 0.0614], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 18:35:50,913 INFO [train.py:968] (0/2) Epoch 7, batch 20600, giga_loss[loss=0.3028, simple_loss=0.372, pruned_loss=0.1168, over 28663.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3615, pruned_loss=0.1109, over 5674488.18 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3622, pruned_loss=0.1057, over 5717835.64 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3598, pruned_loss=0.1106, over 5684905.17 frames. ], batch size: 262, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:36:04,509 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 18:36:08,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3996, 1.5371, 1.2514, 1.5702], device='cuda:0'), covar=tensor([0.2193, 0.2069, 0.2210, 0.2039], device='cuda:0'), in_proj_covar=tensor([0.1185, 0.0893, 0.1041, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:36:09,456 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:11,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:11,572 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:15,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:29,510 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-03 18:36:30,265 INFO [train.py:968] (0/2) Epoch 7, batch 20650, giga_loss[loss=0.3436, simple_loss=0.4055, pruned_loss=0.1408, over 28863.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3634, pruned_loss=0.1117, over 5686401.02 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3629, pruned_loss=0.1064, over 5723102.61 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3613, pruned_loss=0.1109, over 5688625.67 frames. ], batch size: 145, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:36:33,054 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:34,934 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:50,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.352e+02 1.323e+03 1.705e+03 2.480e+03 9.972e+03, threshold=3.410e+03, percent-clipped=19.0
+2023-03-03 18:37:12,804 INFO [train.py:968] (0/2) Epoch 7, batch 20700, giga_loss[loss=0.288, simple_loss=0.3575, pruned_loss=0.1093, over 28953.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3658, pruned_loss=0.1135, over 5690664.02 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3625, pruned_loss=0.1063, over 5728309.53 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3646, pruned_loss=0.1132, over 5686291.25 frames. ], batch size: 213, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:37:14,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3127, 1.5285, 0.9864, 1.1866], device='cuda:0'), covar=tensor([0.0814, 0.0550, 0.1436, 0.0978], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0435, 0.0498, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 18:37:40,813 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:37:56,378 INFO [train.py:968] (0/2) Epoch 7, batch 20750, giga_loss[loss=0.31, simple_loss=0.3784, pruned_loss=0.1208, over 28681.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.367, pruned_loss=0.1147, over 5687487.57 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3628, pruned_loss=0.1065, over 5720271.60 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3659, pruned_loss=0.1144, over 5690216.18 frames. ], batch size: 307, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:38:18,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.689e+02 1.061e+03 1.319e+03 1.721e+03 3.007e+03, threshold=2.637e+03, percent-clipped=0.0
+2023-03-03 18:38:44,911 INFO [train.py:968] (0/2) Epoch 7, batch 20800, giga_loss[loss=0.2784, simple_loss=0.3594, pruned_loss=0.09866, over 29111.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3672, pruned_loss=0.1149, over 5700586.50 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3627, pruned_loss=0.1065, over 5721070.24 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3664, pruned_loss=0.1147, over 5701932.33 frames. ], batch size: 128, lr: 4.62e-03, grad_scale: 8.0
+2023-03-03 18:39:24,215 INFO [train.py:968] (0/2) Epoch 7, batch 20850, giga_loss[loss=0.3058, simple_loss=0.3729, pruned_loss=0.1193, over 28882.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3703, pruned_loss=0.1176, over 5701562.03 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3644, pruned_loss=0.1077, over 5724163.15 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3683, pruned_loss=0.1168, over 5699284.02 frames. ], batch size: 112, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:39:43,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.476e+02 1.149e+03 1.564e+03 2.106e+03 5.761e+03, threshold=3.129e+03, percent-clipped=12.0
+2023-03-03 18:40:02,700 INFO [train.py:968] (0/2) Epoch 7, batch 20900, giga_loss[loss=0.2594, simple_loss=0.3403, pruned_loss=0.08923, over 28916.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3708, pruned_loss=0.1177, over 5701509.46 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.365, pruned_loss=0.108, over 5717864.02 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3689, pruned_loss=0.117, over 5704301.53 frames. ], batch size: 227, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:40:13,074 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:40:29,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2476, 3.0656, 2.8976, 1.4009], device='cuda:0'), covar=tensor([0.0753, 0.0852, 0.0839, 0.2307], device='cuda:0'), in_proj_covar=tensor([0.0907, 0.0857, 0.0768, 0.0618], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 18:40:41,629 INFO [train.py:968] (0/2) Epoch 7, batch 20950, giga_loss[loss=0.304, simple_loss=0.377, pruned_loss=0.1155, over 28876.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3695, pruned_loss=0.1154, over 5704136.32 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.366, pruned_loss=0.1087, over 5723161.34 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3672, pruned_loss=0.1145, over 5701181.54 frames. ], batch size: 186, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:41:00,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.298e+02 1.052e+03 1.365e+03 1.935e+03 6.576e+03, threshold=2.729e+03, percent-clipped=8.0
+2023-03-03 18:41:21,875 INFO [train.py:968] (0/2) Epoch 7, batch 21000, giga_loss[loss=0.2825, simple_loss=0.3587, pruned_loss=0.1031, over 28811.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3705, pruned_loss=0.1152, over 5713080.16 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3659, pruned_loss=0.1089, over 5727131.20 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3688, pruned_loss=0.1144, over 5706711.99 frames. ], batch size: 92, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:41:21,879 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 18:41:29,431 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3880, 1.3846, 1.0917, 1.5444], device='cuda:0'), covar=tensor([0.0740, 0.0291, 0.0351, 0.0829], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0119, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0067, 0.0048, 0.0043, 0.0073], device='cuda:0')
+2023-03-03 18:41:30,269 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2298, simple_loss=0.3343, pruned_loss=0.06268, over 944034.00 frames. 
+2023-03-03 18:41:30,270 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-03 18:42:08,110 INFO [train.py:968] (0/2) Epoch 7, batch 21050, giga_loss[loss=0.2377, simple_loss=0.3231, pruned_loss=0.07614, over 28934.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3701, pruned_loss=0.1148, over 5721265.88 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.366, pruned_loss=0.1091, over 5732459.22 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3688, pruned_loss=0.1142, over 5711161.82 frames. ], batch size: 128, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:42:25,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.100e+02 1.089e+03 1.340e+03 1.793e+03 3.926e+03, threshold=2.681e+03, percent-clipped=8.0
+2023-03-03 18:42:33,263 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:42:36,246 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-03 18:42:45,887 INFO [train.py:968] (0/2) Epoch 7, batch 21100, giga_loss[loss=0.262, simple_loss=0.3359, pruned_loss=0.09402, over 28712.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3658, pruned_loss=0.1122, over 5716445.37 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3659, pruned_loss=0.1092, over 5734165.11 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3649, pruned_loss=0.1117, over 5706999.50 frames. ], batch size: 85, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:42:50,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:43:24,483 INFO [train.py:968] (0/2) Epoch 7, batch 21150, libri_loss[loss=0.3408, simple_loss=0.4095, pruned_loss=0.1361, over 28705.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3646, pruned_loss=0.1119, over 5720236.77 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3668, pruned_loss=0.1099, over 5738327.83 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.363, pruned_loss=0.1109, over 5708311.73 frames. ], batch size: 106, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:43:37,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:43:37,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2347, 1.3334, 1.1315, 0.9934], device='cuda:0'), covar=tensor([0.1378, 0.1328, 0.0930, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.1531, 0.1359, 0.1347, 0.1435], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 18:43:42,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.138e+02 1.081e+03 1.362e+03 1.816e+03 4.210e+03, threshold=2.724e+03, percent-clipped=8.0
+2023-03-03 18:43:58,279 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-03 18:43:59,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6618, 1.6510, 1.4332, 1.8149], device='cuda:0'), covar=tensor([0.2101, 0.2104, 0.2095, 0.2123], device='cuda:0'), in_proj_covar=tensor([0.1181, 0.0898, 0.1043, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:44:02,700 INFO [train.py:968] (0/2) Epoch 7, batch 21200, libri_loss[loss=0.2862, simple_loss=0.342, pruned_loss=0.1152, over 29377.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3639, pruned_loss=0.1119, over 5714330.65 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3672, pruned_loss=0.1104, over 5729465.92 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3622, pruned_loss=0.1107, over 5711808.42 frames. ], batch size: 67, lr: 4.62e-03, grad_scale: 8.0
+2023-03-03 18:44:28,998 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:44:41,617 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:44:41,805 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 18:44:42,666 INFO [train.py:968] (0/2) Epoch 7, batch 21250, giga_loss[loss=0.2813, simple_loss=0.3581, pruned_loss=0.1022, over 29048.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3649, pruned_loss=0.1129, over 5715225.59 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3676, pruned_loss=0.1108, over 5732487.69 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3631, pruned_loss=0.1117, over 5710223.34 frames. ], batch size: 164, lr: 4.62e-03, grad_scale: 8.0
+2023-03-03 18:44:43,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:45:03,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.385e+02 1.079e+03 1.403e+03 1.901e+03 6.225e+03, threshold=2.806e+03, percent-clipped=11.0
+2023-03-03 18:45:07,246 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:45:13,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:45:21,257 INFO [train.py:968] (0/2) Epoch 7, batch 21300, giga_loss[loss=0.2729, simple_loss=0.3501, pruned_loss=0.09782, over 29104.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3651, pruned_loss=0.1132, over 5714099.13 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3681, pruned_loss=0.1114, over 5737528.38 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3632, pruned_loss=0.1118, over 5704773.38 frames. ], batch size: 155, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:45:32,862 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:46:03,887 INFO [train.py:968] (0/2) Epoch 7, batch 21350, libri_loss[loss=0.3118, simple_loss=0.3767, pruned_loss=0.1234, over 29274.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3652, pruned_loss=0.1123, over 5717896.59 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3687, pruned_loss=0.1119, over 5737965.77 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.363, pruned_loss=0.1107, over 5709823.21 frames. ], batch size: 94, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:46:16,922 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 18:46:23,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.833e+02 1.071e+03 1.351e+03 2.067e+03 7.658e+03, threshold=2.702e+03, percent-clipped=17.0
+2023-03-03 18:46:45,021 INFO [train.py:968] (0/2) Epoch 7, batch 21400, giga_loss[loss=0.2649, simple_loss=0.34, pruned_loss=0.09493, over 28874.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3639, pruned_loss=0.1115, over 5715602.61 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3689, pruned_loss=0.1125, over 5741998.80 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3619, pruned_loss=0.1096, over 5704562.71 frames. ], batch size: 119, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:47:09,150 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:47:11,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:47:23,262 INFO [train.py:968] (0/2) Epoch 7, batch 21450, giga_loss[loss=0.2696, simple_loss=0.3417, pruned_loss=0.09878, over 29144.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3638, pruned_loss=0.1121, over 5711261.07 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3694, pruned_loss=0.1129, over 5742387.42 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3616, pruned_loss=0.1102, over 5701537.54 frames. ], batch size: 128, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:47:30,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:47:36,841 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:47:45,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.040e+02 9.966e+02 1.195e+03 1.526e+03 3.940e+03, threshold=2.389e+03, percent-clipped=4.0
+2023-03-03 18:48:05,231 INFO [train.py:968] (0/2) Epoch 7, batch 21500, giga_loss[loss=0.283, simple_loss=0.357, pruned_loss=0.1045, over 28618.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.362, pruned_loss=0.1116, over 5700779.98 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3696, pruned_loss=0.1132, over 5736234.09 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3598, pruned_loss=0.1098, over 5697786.32 frames. ], batch size: 85, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:48:13,603 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-03 18:48:36,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:48:43,458 INFO [train.py:968] (0/2) Epoch 7, batch 21550, giga_loss[loss=0.2552, simple_loss=0.3279, pruned_loss=0.09131, over 28655.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3586, pruned_loss=0.1099, over 5703030.73 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.37, pruned_loss=0.1137, over 5740414.83 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3562, pruned_loss=0.108, over 5696320.01 frames. ], batch size: 71, lr: 4.62e-03, grad_scale: 2.0
+2023-03-03 18:48:55,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8055, 1.1162, 3.4188, 2.8816], device='cuda:0'), covar=tensor([0.2176, 0.2764, 0.0713, 0.0928], device='cuda:0'), in_proj_covar=tensor([0.0590, 0.0548, 0.0771, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:49:05,443 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.982e+02 1.245e+03 1.751e+03 2.525e+03 1.787e+04, threshold=3.503e+03, percent-clipped=28.0
+2023-03-03 18:49:06,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4352, 1.6342, 1.3892, 1.5998], device='cuda:0'), covar=tensor([0.2442, 0.2360, 0.2439, 0.2164], device='cuda:0'), in_proj_covar=tensor([0.1180, 0.0902, 0.1049, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:49:22,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7938, 2.4634, 1.9833, 2.0746], device='cuda:0'), covar=tensor([0.0489, 0.0527, 0.0765, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0336, 0.0430, 0.0492, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 18:49:25,418 INFO [train.py:968] (0/2) Epoch 7, batch 21600, giga_loss[loss=0.2699, simple_loss=0.3438, pruned_loss=0.09806, over 28970.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3579, pruned_loss=0.1104, over 5697726.95 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3703, pruned_loss=0.114, over 5741019.22 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3557, pruned_loss=0.1086, over 5691351.17 frames. ], batch size: 164, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:49:25,653 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:49:27,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:49:29,489 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:49:50,449 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:50:03,796 INFO [train.py:968] (0/2) Epoch 7, batch 21650, giga_loss[loss=0.2984, simple_loss=0.3647, pruned_loss=0.116, over 28827.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3581, pruned_loss=0.1111, over 5678097.15 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3713, pruned_loss=0.115, over 5716341.42 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3552, pruned_loss=0.1087, over 5693464.26 frames. ], batch size: 199, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:50:05,524 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 18:50:10,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1282, 1.9200, 1.8708, 1.7035], device='cuda:0'), covar=tensor([0.1204, 0.2274, 0.1622, 0.1882], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0722, 0.0637, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 18:50:25,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.677e+02 1.019e+03 1.344e+03 1.831e+03 4.035e+03, threshold=2.688e+03, percent-clipped=1.0
+2023-03-03 18:50:31,328 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:50:33,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:50:34,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:50:45,394 INFO [train.py:968] (0/2) Epoch 7, batch 21700, giga_loss[loss=0.2677, simple_loss=0.3388, pruned_loss=0.09831, over 29006.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3554, pruned_loss=0.1098, over 5684726.31 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3716, pruned_loss=0.1152, over 5719931.12 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3526, pruned_loss=0.1075, over 5693061.56 frames. ], batch size: 136, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:50:57,009 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:51:06,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5515, 2.3902, 1.7649, 0.6239], device='cuda:0'), covar=tensor([0.2931, 0.1236, 0.1948, 0.2673], device='cuda:0'), in_proj_covar=tensor([0.1409, 0.1316, 0.1381, 0.1160], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 18:51:09,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:51:22,210 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:51:22,644 INFO [train.py:968] (0/2) Epoch 7, batch 21750, giga_loss[loss=0.2717, simple_loss=0.3457, pruned_loss=0.09886, over 28258.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3533, pruned_loss=0.1091, over 5693888.50 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3717, pruned_loss=0.1155, over 5723673.52 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3506, pruned_loss=0.1069, over 5696493.35 frames. ], batch size: 65, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:51:24,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:51:42,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.962e+02 1.011e+03 1.198e+03 1.711e+03 1.043e+04, threshold=2.397e+03, percent-clipped=12.0
+2023-03-03 18:51:48,738 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:51:59,638 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-03 18:52:01,907 INFO [train.py:968] (0/2) Epoch 7, batch 21800, giga_loss[loss=0.2507, simple_loss=0.3223, pruned_loss=0.08957, over 29089.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3511, pruned_loss=0.1078, over 5697799.04 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3726, pruned_loss=0.1163, over 5717979.73 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3478, pruned_loss=0.1053, over 5705126.05 frames. ], batch size: 128, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:52:10,727 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:52:19,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4020, 1.5375, 1.3056, 1.7330], device='cuda:0'), covar=tensor([0.2466, 0.2447, 0.2642, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.1185, 0.0903, 0.1050, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:52:25,964 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:52:29,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:52:41,779 INFO [train.py:968] (0/2) Epoch 7, batch 21850, giga_loss[loss=0.2656, simple_loss=0.3386, pruned_loss=0.0963, over 28982.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.349, pruned_loss=0.1067, over 5700389.56 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3728, pruned_loss=0.1169, over 5718094.56 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3455, pruned_loss=0.1039, over 5705402.73 frames. ], batch size: 145, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:52:52,048 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:53:02,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.322e+02 9.636e+02 1.225e+03 1.760e+03 3.247e+03, threshold=2.450e+03, percent-clipped=6.0
+2023-03-03 18:53:23,783 INFO [train.py:968] (0/2) Epoch 7, batch 21900, giga_loss[loss=0.2597, simple_loss=0.34, pruned_loss=0.08963, over 28927.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3506, pruned_loss=0.107, over 5705197.34 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3732, pruned_loss=0.1172, over 5720959.32 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3472, pruned_loss=0.1044, over 5706596.24 frames. ], batch size: 145, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:53:41,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9228, 0.9905, 3.5965, 2.9434], device='cuda:0'), covar=tensor([0.1712, 0.2604, 0.0381, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0586, 0.0545, 0.0767, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 18:53:42,574 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-03 18:54:06,833 INFO [train.py:968] (0/2) Epoch 7, batch 21950, giga_loss[loss=0.2805, simple_loss=0.3511, pruned_loss=0.1049, over 28794.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3539, pruned_loss=0.1084, over 5699539.75 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3735, pruned_loss=0.1174, over 5722389.99 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3509, pruned_loss=0.106, over 5699267.64 frames. ], batch size: 119, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:54:25,416 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:54:28,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7096, 1.7835, 1.6722, 1.6621], device='cuda:0'), covar=tensor([0.1177, 0.2141, 0.1587, 0.1629], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0725, 0.0636, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 18:54:30,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.748e+02 9.842e+02 1.183e+03 1.399e+03 3.967e+03, threshold=2.367e+03, percent-clipped=4.0
+2023-03-03 18:54:51,396 INFO [train.py:968] (0/2) Epoch 7, batch 22000, giga_loss[loss=0.2645, simple_loss=0.3435, pruned_loss=0.09276, over 28820.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3571, pruned_loss=0.1095, over 5688403.98 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3733, pruned_loss=0.1176, over 5717037.52 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3542, pruned_loss=0.1072, over 5692484.09 frames. ], batch size: 199, lr: 4.62e-03, grad_scale: 8.0
+2023-03-03 18:55:10,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:55:21,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-03 18:55:33,991 INFO [train.py:968] (0/2) Epoch 7, batch 22050, giga_loss[loss=0.2734, simple_loss=0.3599, pruned_loss=0.09347, over 28708.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.359, pruned_loss=0.11, over 5689371.95 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.374, pruned_loss=0.1183, over 5708611.41 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3557, pruned_loss=0.1073, over 5700421.18 frames. ], batch size: 243, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:55:38,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5912, 1.6526, 1.6367, 1.5319], device='cuda:0'), covar=tensor([0.1058, 0.1778, 0.1513, 0.1491], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0722, 0.0635, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 18:55:54,603 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:55:55,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.306e+02 1.141e+03 1.534e+03 2.312e+03 1.274e+04, threshold=3.068e+03, percent-clipped=22.0
+2023-03-03 18:56:13,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3060, 1.7887, 1.6369, 1.2717], device='cuda:0'), covar=tensor([0.1574, 0.1992, 0.1333, 0.1516], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0708, 0.0809, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 18:56:14,305 INFO [train.py:968] (0/2) Epoch 7, batch 22100, giga_loss[loss=0.2598, simple_loss=0.3381, pruned_loss=0.09076, over 28952.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3579, pruned_loss=0.1087, over 5697069.34 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3741, pruned_loss=0.1188, over 5711831.74 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3548, pruned_loss=0.1059, over 5702858.85 frames. ], batch size: 136, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:56:16,045 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:56:22,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:56:58,914 INFO [train.py:968] (0/2) Epoch 7, batch 22150, giga_loss[loss=0.2595, simple_loss=0.3374, pruned_loss=0.09074, over 28949.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3587, pruned_loss=0.1097, over 5688357.06 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3744, pruned_loss=0.1191, over 5712483.11 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3557, pruned_loss=0.107, over 5691891.82 frames. ], batch size: 164, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:57:23,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.526e+02 1.091e+03 1.359e+03 1.892e+03 5.438e+03, threshold=2.717e+03, percent-clipped=5.0
+2023-03-03 18:57:29,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:57:43,290 INFO [train.py:968] (0/2) Epoch 7, batch 22200, giga_loss[loss=0.3785, simple_loss=0.4174, pruned_loss=0.1698, over 26710.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3593, pruned_loss=0.1106, over 5694977.02 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3743, pruned_loss=0.1192, over 5714525.16 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3568, pruned_loss=0.1083, over 5695833.09 frames. ], batch size: 555, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:58:25,863 INFO [train.py:968] (0/2) Epoch 7, batch 22250, giga_loss[loss=0.2902, simple_loss=0.3612, pruned_loss=0.1096, over 28284.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3602, pruned_loss=0.1111, over 5694209.63 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3747, pruned_loss=0.1195, over 5714585.54 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3579, pruned_loss=0.1091, over 5694755.58 frames. ], batch size: 368, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:58:27,909 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:58:29,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:58:33,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4747, 1.7438, 1.7894, 1.4031], device='cuda:0'), covar=tensor([0.1514, 0.1901, 0.1235, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0709, 0.0810, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 18:58:46,757 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.231e+02 1.071e+03 1.412e+03 2.036e+03 6.113e+03, threshold=2.823e+03, percent-clipped=11.0
+2023-03-03 18:58:51,755 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:07,322 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:07,876 INFO [train.py:968] (0/2) Epoch 7, batch 22300, giga_loss[loss=0.2786, simple_loss=0.3589, pruned_loss=0.09911, over 29075.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3626, pruned_loss=0.1121, over 5697036.09 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.375, pruned_loss=0.1199, over 5708980.63 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3601, pruned_loss=0.1098, over 5701539.37 frames. ], batch size: 155, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:59:08,696 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295951.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 18:59:32,712 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:35,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:39,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.12 vs. limit=5.0
+2023-03-03 18:59:42,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:48,720 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-296000.pt
+2023-03-03 18:59:49,014 INFO [train.py:968] (0/2) Epoch 7, batch 22350, giga_loss[loss=0.2933, simple_loss=0.3586, pruned_loss=0.114, over 28510.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3647, pruned_loss=0.113, over 5694975.95 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3757, pruned_loss=0.1205, over 5701969.76 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3618, pruned_loss=0.1105, over 5705140.41 frames. ], batch size: 85, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:59:56,292 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:00:05,272 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296022.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:00:08,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.608e+02 1.249e+03 1.695e+03 2.374e+03 5.905e+03, threshold=3.391e+03, percent-clipped=17.0
+2023-03-03 19:00:22,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:00:25,464 INFO [train.py:968] (0/2) Epoch 7, batch 22400, giga_loss[loss=0.2895, simple_loss=0.3608, pruned_loss=0.109, over 28957.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3649, pruned_loss=0.1127, over 5707054.61 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3762, pruned_loss=0.1209, over 5708593.36 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3618, pruned_loss=0.1101, over 5709239.53 frames. ], batch size: 136, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:00:48,013 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:01:06,422 INFO [train.py:968] (0/2) Epoch 7, batch 22450, giga_loss[loss=0.2612, simple_loss=0.3328, pruned_loss=0.09481, over 28689.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3643, pruned_loss=0.1121, over 5711311.14 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3766, pruned_loss=0.1211, over 5704153.87 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3612, pruned_loss=0.1096, over 5716316.61 frames. ], batch size: 85, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:01:07,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:01:29,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.744e+02 1.176e+03 1.492e+03 2.149e+03 4.536e+03, threshold=2.985e+03, percent-clipped=5.0
+2023-03-03 19:01:30,561 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:01:38,150 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:01:40,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:01:49,809 INFO [train.py:968] (0/2) Epoch 7, batch 22500, giga_loss[loss=0.2855, simple_loss=0.3495, pruned_loss=0.1107, over 28548.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3645, pruned_loss=0.1122, over 5706158.84 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3766, pruned_loss=0.1212, over 5705342.56 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.362, pruned_loss=0.1101, over 5709185.00 frames. ], batch size: 60, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:02:04,193 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:02:22,737 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:02:25,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:02:30,734 INFO [train.py:968] (0/2) Epoch 7, batch 22550, giga_loss[loss=0.309, simple_loss=0.3714, pruned_loss=0.1233, over 27874.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.363, pruned_loss=0.1115, over 5702706.60 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3764, pruned_loss=0.1212, over 5700830.91 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3608, pruned_loss=0.1096, over 5709229.88 frames. ], batch size: 412, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:02:39,184 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:02:45,088 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:02:48,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.592e+02 1.128e+03 1.409e+03 1.755e+03 6.272e+03, threshold=2.818e+03, percent-clipped=6.0
+2023-03-03 19:03:04,272 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:07,056 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:09,289 INFO [train.py:968] (0/2) Epoch 7, batch 22600, libri_loss[loss=0.3805, simple_loss=0.4211, pruned_loss=0.1699, over 29201.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3622, pruned_loss=0.1117, over 5700661.30 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3776, pruned_loss=0.1222, over 5691744.61 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3589, pruned_loss=0.1088, over 5713374.52 frames. ], batch size: 97, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:03:20,555 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-03 19:03:21,860 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:24,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6848, 2.1774, 2.0513, 1.9994], device='cuda:0'), covar=tensor([0.0564, 0.0578, 0.0742, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0443, 0.0499, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:03:24,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:26,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:28,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:39,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9732, 4.8664, 2.0125, 2.0488], device='cuda:0'), covar=tensor([0.0750, 0.0192, 0.0787, 0.1089], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0485, 0.0316, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 19:03:48,928 INFO [train.py:968] (0/2) Epoch 7, batch 22650, giga_loss[loss=0.2626, simple_loss=0.334, pruned_loss=0.09561, over 28711.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3578, pruned_loss=0.1095, over 5704571.26 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3776, pruned_loss=0.1227, over 5695415.20 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3547, pruned_loss=0.1066, over 5711813.13 frames. ], batch size: 242, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:03:50,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:04:07,138 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:04:08,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296326.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:04:09,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.419e+02 1.062e+03 1.413e+03 2.138e+03 5.045e+03, threshold=2.825e+03, percent-clipped=13.0
+2023-03-03 19:04:27,910 INFO [train.py:968] (0/2) Epoch 7, batch 22700, giga_loss[loss=0.297, simple_loss=0.3624, pruned_loss=0.1158, over 28884.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3566, pruned_loss=0.1084, over 5694205.20 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3775, pruned_loss=0.1228, over 5680992.12 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3537, pruned_loss=0.1056, over 5714067.21 frames. ], batch size: 145, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:05:11,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296397.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:05:14,648 INFO [train.py:968] (0/2) Epoch 7, batch 22750, giga_loss[loss=0.3287, simple_loss=0.3909, pruned_loss=0.1332, over 26712.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3588, pruned_loss=0.1083, over 5693676.27 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3778, pruned_loss=0.123, over 5682338.95 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.356, pruned_loss=0.1058, over 5708127.31 frames. ], batch size: 555, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:05:37,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.064e+02 1.153e+03 1.481e+03 1.889e+03 5.875e+03, threshold=2.963e+03, percent-clipped=12.0
+2023-03-03 19:05:48,188 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:05:54,324 INFO [train.py:968] (0/2) Epoch 7, batch 22800, giga_loss[loss=0.2921, simple_loss=0.3668, pruned_loss=0.1086, over 28671.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3585, pruned_loss=0.1078, over 5706887.79 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3779, pruned_loss=0.1232, over 5687013.71 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3559, pruned_loss=0.1053, over 5714347.21 frames. ], batch size: 307, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:05:56,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:06:07,604 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:06:08,900 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296469.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:06:09,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:06:10,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296472.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:06:33,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:06:33,957 INFO [train.py:968] (0/2) Epoch 7, batch 22850, giga_loss[loss=0.2866, simple_loss=0.3471, pruned_loss=0.1131, over 28686.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3564, pruned_loss=0.1072, over 5717879.16 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3772, pruned_loss=0.1229, over 5691134.42 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3545, pruned_loss=0.1052, over 5720611.76 frames. ], batch size: 92, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:06:35,138 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296501.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:06:59,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.189e+02 1.046e+03 1.323e+03 1.941e+03 5.294e+03, threshold=2.645e+03, percent-clipped=10.0
+2023-03-03 19:07:08,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296540.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:07:09,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296543.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:07:14,962 INFO [train.py:968] (0/2) Epoch 7, batch 22900, giga_loss[loss=0.3249, simple_loss=0.382, pruned_loss=0.1339, over 28902.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3568, pruned_loss=0.1091, over 5720368.97 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3775, pruned_loss=0.1233, over 5697048.52 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3543, pruned_loss=0.1066, over 5718099.53 frames. ], batch size: 227, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:07:23,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6620, 4.4614, 4.2115, 2.0656], device='cuda:0'), covar=tensor([0.0449, 0.0567, 0.0712, 0.2006], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0862, 0.0781, 0.0620], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:07:32,957 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296572.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:07:35,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1058, 1.9538, 1.4754, 1.6389], device='cuda:0'), covar=tensor([0.0576, 0.0592, 0.0889, 0.0959], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0442, 0.0494, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:07:37,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4093, 1.6926, 1.6625, 1.3765], device='cuda:0'), covar=tensor([0.1036, 0.1467, 0.0915, 0.1011], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0704, 0.0811, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 19:07:45,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:07:53,195 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:07:56,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:07:57,385 INFO [train.py:968] (0/2) Epoch 7, batch 22950, giga_loss[loss=0.3238, simple_loss=0.3704, pruned_loss=0.1385, over 26784.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3561, pruned_loss=0.1106, over 5709654.29 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3778, pruned_loss=0.1237, over 5689610.26 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3535, pruned_loss=0.108, over 5714438.44 frames. ], batch size: 555, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:08:19,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:08:19,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.731e+02 1.079e+03 1.287e+03 1.812e+03 4.225e+03, threshold=2.573e+03, percent-clipped=6.0
+2023-03-03 19:08:28,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:08:30,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6340, 4.4168, 4.1665, 1.8894], device='cuda:0'), covar=tensor([0.0370, 0.0524, 0.0638, 0.2148], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0867, 0.0786, 0.0624], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:08:36,935 INFO [train.py:968] (0/2) Epoch 7, batch 23000, giga_loss[loss=0.2751, simple_loss=0.3444, pruned_loss=0.1029, over 28897.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3548, pruned_loss=0.1106, over 5716853.84 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3778, pruned_loss=0.1237, over 5693123.44 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3524, pruned_loss=0.1083, over 5717982.37 frames. ], batch size: 106, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:08:56,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3921, 1.5822, 1.5019, 1.3924], device='cuda:0'), covar=tensor([0.1166, 0.1495, 0.1613, 0.1574], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0725, 0.0649, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 19:09:14,917 INFO [train.py:968] (0/2) Epoch 7, batch 23050, libri_loss[loss=0.2948, simple_loss=0.3677, pruned_loss=0.1109, over 29538.00 frames. ], tot_loss[loss=0.286, simple_loss=0.353, pruned_loss=0.1095, over 5714947.73 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3778, pruned_loss=0.1237, over 5698103.92 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3505, pruned_loss=0.1073, over 5711784.60 frames. ], batch size: 81, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:09:21,767 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:09:37,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.269e+02 1.096e+03 1.337e+03 1.776e+03 9.073e+03, threshold=2.674e+03, percent-clipped=10.0
+2023-03-03 19:09:39,043 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:09:41,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:09:53,775 INFO [train.py:968] (0/2) Epoch 7, batch 23100, giga_loss[loss=0.2423, simple_loss=0.3188, pruned_loss=0.0829, over 28630.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3478, pruned_loss=0.1066, over 5715430.56 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3777, pruned_loss=0.1238, over 5692229.67 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3455, pruned_loss=0.1046, over 5717985.66 frames. ], batch size: 242, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:10:03,073 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:19,857 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:21,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:29,155 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4636, 4.3060, 4.0272, 1.8896], device='cuda:0'), covar=tensor([0.0454, 0.0570, 0.0661, 0.1976], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0864, 0.0781, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:10:32,368 INFO [train.py:968] (0/2) Epoch 7, batch 23150, giga_loss[loss=0.2438, simple_loss=0.3137, pruned_loss=0.08694, over 28881.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3443, pruned_loss=0.105, over 5708898.53 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3781, pruned_loss=0.1243, over 5687559.01 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3414, pruned_loss=0.1025, over 5714988.48 frames. ], batch size: 227, lr: 4.61e-03, grad_scale: 2.0
+2023-03-03 19:10:36,128 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 19:10:45,610 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:46,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:52,569 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:54,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.265e+02 1.145e+03 1.547e+03 2.051e+03 6.462e+03, threshold=3.095e+03, percent-clipped=13.0
+2023-03-03 19:11:10,292 INFO [train.py:968] (0/2) Epoch 7, batch 23200, giga_loss[loss=0.2857, simple_loss=0.3565, pruned_loss=0.1074, over 28941.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3447, pruned_loss=0.1056, over 5722001.78 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3792, pruned_loss=0.1254, over 5696679.39 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3399, pruned_loss=0.1018, over 5719699.02 frames. ], batch size: 213, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:11:40,458 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:11:43,492 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:11:51,541 INFO [train.py:968] (0/2) Epoch 7, batch 23250, giga_loss[loss=0.3195, simple_loss=0.3878, pruned_loss=0.1256, over 28727.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3479, pruned_loss=0.1067, over 5713378.89 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3793, pruned_loss=0.1256, over 5696706.77 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3435, pruned_loss=0.1033, over 5712127.37 frames. ], batch size: 284, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:11:56,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:12:13,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.292e+02 1.200e+03 1.416e+03 1.901e+03 4.345e+03, threshold=2.832e+03, percent-clipped=10.0
+2023-03-03 19:12:31,935 INFO [train.py:968] (0/2) Epoch 7, batch 23300, giga_loss[loss=0.2562, simple_loss=0.3405, pruned_loss=0.08589, over 29079.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3523, pruned_loss=0.109, over 5697493.71 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3797, pruned_loss=0.1258, over 5684106.58 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3476, pruned_loss=0.1056, over 5708060.49 frames. ], batch size: 155, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:12:39,919 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:12:41,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:13:06,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:13:07,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2254, 1.4267, 1.2274, 0.9397], device='cuda:0'), covar=tensor([0.1947, 0.1994, 0.2074, 0.1832], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.0894, 0.1038, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 19:13:11,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5688, 1.9959, 1.7689, 1.7999], device='cuda:0'), covar=tensor([0.0570, 0.0675, 0.0926, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0445, 0.0498, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:13:12,724 INFO [train.py:968] (0/2) Epoch 7, batch 23350, giga_loss[loss=0.2341, simple_loss=0.3167, pruned_loss=0.07578, over 28910.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3545, pruned_loss=0.1092, over 5700563.12 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3795, pruned_loss=0.1257, over 5685924.13 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3508, pruned_loss=0.1065, over 5707423.03 frames. ], batch size: 145, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:13:18,083 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:13:37,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.484e+02 1.083e+03 1.303e+03 1.648e+03 3.710e+03, threshold=2.607e+03, percent-clipped=6.0
+2023-03-03 19:13:53,675 INFO [train.py:968] (0/2) Epoch 7, batch 23400, giga_loss[loss=0.312, simple_loss=0.3816, pruned_loss=0.1212, over 28533.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3575, pruned_loss=0.1105, over 5697178.02 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3799, pruned_loss=0.1261, over 5677485.84 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3537, pruned_loss=0.1075, over 5711234.48 frames. ], batch size: 336, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:14:21,087 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:14:23,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:14:36,705 INFO [train.py:968] (0/2) Epoch 7, batch 23450, libri_loss[loss=0.3181, simple_loss=0.3696, pruned_loss=0.1333, over 29324.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3589, pruned_loss=0.1108, over 5708025.99 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3799, pruned_loss=0.1262, over 5680768.65 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3555, pruned_loss=0.1082, over 5716523.19 frames. ], batch size: 71, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:14:40,962 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.11 vs. limit=5.0
+2023-03-03 19:14:45,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3703, 1.7228, 1.7583, 1.3307], device='cuda:0'), covar=tensor([0.1325, 0.1845, 0.1086, 0.1289], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0714, 0.0818, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 19:14:46,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2862, 1.3467, 1.3034, 1.4847], device='cuda:0'), covar=tensor([0.0733, 0.0316, 0.0313, 0.0781], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0120, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0067, 0.0049, 0.0044, 0.0073], device='cuda:0')
+2023-03-03 19:15:02,652 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-03 19:15:05,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.807e+02 1.142e+03 1.519e+03 2.167e+03 5.684e+03, threshold=3.038e+03, percent-clipped=16.0
+2023-03-03 19:15:21,538 INFO [train.py:968] (0/2) Epoch 7, batch 23500, giga_loss[loss=0.3138, simple_loss=0.3771, pruned_loss=0.1252, over 28822.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3645, pruned_loss=0.1157, over 5688049.23 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3804, pruned_loss=0.1269, over 5665316.06 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3608, pruned_loss=0.1126, over 5710087.31 frames. ], batch size: 119, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:15:26,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5428, 1.8355, 1.8760, 1.4469], device='cuda:0'), covar=tensor([0.1427, 0.1817, 0.1148, 0.1303], device='cuda:0'), in_proj_covar=tensor([0.0787, 0.0711, 0.0814, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 19:16:12,079 INFO [train.py:968] (0/2) Epoch 7, batch 23550, giga_loss[loss=0.306, simple_loss=0.3776, pruned_loss=0.1172, over 28740.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3726, pruned_loss=0.1224, over 5678812.19 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3811, pruned_loss=0.1275, over 5668490.26 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3688, pruned_loss=0.1192, over 5693885.13 frames. ], batch size: 262, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:16:14,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:16:36,858 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:16:40,278 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:16:40,639 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.771e+02 1.774e+03 2.217e+03 3.235e+03 7.368e+03, threshold=4.435e+03, percent-clipped=27.0
+2023-03-03 19:16:59,553 INFO [train.py:968] (0/2) Epoch 7, batch 23600, giga_loss[loss=0.3184, simple_loss=0.3867, pruned_loss=0.1251, over 28836.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.38, pruned_loss=0.1282, over 5674817.40 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3815, pruned_loss=0.1279, over 5673522.79 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3765, pruned_loss=0.1252, over 5682554.97 frames. ], batch size: 199, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:17:08,989 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:17:10,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:17:17,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:17:17,994 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:17:21,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3441, 3.6939, 1.5014, 1.5488], device='cuda:0'), covar=tensor([0.0924, 0.0294, 0.0807, 0.1223], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0487, 0.0315, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 19:17:31,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:17:44,142 INFO [train.py:968] (0/2) Epoch 7, batch 23650, giga_loss[loss=0.3494, simple_loss=0.4115, pruned_loss=0.1436, over 29039.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3859, pruned_loss=0.1333, over 5677518.77 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3817, pruned_loss=0.128, over 5677482.62 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.383, pruned_loss=0.1309, over 5679865.51 frames. ], batch size: 155, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:18:15,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.667e+03 2.154e+03 2.873e+03 6.880e+03, threshold=4.308e+03, percent-clipped=8.0
+2023-03-03 19:18:16,492 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:18:32,386 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:18:34,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:18:36,569 INFO [train.py:968] (0/2) Epoch 7, batch 23700, giga_loss[loss=0.4245, simple_loss=0.4507, pruned_loss=0.1991, over 28671.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3934, pruned_loss=0.1397, over 5681289.37 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3818, pruned_loss=0.1281, over 5679997.53 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.391, pruned_loss=0.1378, over 5680959.03 frames. ], batch size: 262, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:19:01,368 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:05,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:23,770 INFO [train.py:968] (0/2) Epoch 7, batch 23750, giga_loss[loss=0.3553, simple_loss=0.4108, pruned_loss=0.1499, over 28791.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.3976, pruned_loss=0.1439, over 5677540.20 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3818, pruned_loss=0.1283, over 5687489.22 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3962, pruned_loss=0.1428, over 5670442.31 frames. ], batch size: 186, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:19:26,764 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:28,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:30,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:32,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:45,411 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:49,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:51,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.844e+02 1.622e+03 2.083e+03 3.146e+03 7.379e+03, threshold=4.167e+03, percent-clipped=5.0
+2023-03-03 19:19:55,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:59,365 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:20:09,704 INFO [train.py:968] (0/2) Epoch 7, batch 23800, giga_loss[loss=0.3687, simple_loss=0.4163, pruned_loss=0.1605, over 28608.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3989, pruned_loss=0.1461, over 5672001.89 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3816, pruned_loss=0.1283, over 5693944.14 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3983, pruned_loss=0.1456, over 5660287.07 frames. ], batch size: 336, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:20:15,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:20:15,932 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:21:00,352 INFO [train.py:968] (0/2) Epoch 7, batch 23850, libri_loss[loss=0.3513, simple_loss=0.4039, pruned_loss=0.1493, over 27661.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4021, pruned_loss=0.1501, over 5668750.62 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3819, pruned_loss=0.1286, over 5696591.28 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4018, pruned_loss=0.1498, over 5656632.13 frames. ], batch size: 116, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:21:24,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6602, 2.3252, 1.4697, 0.8420], device='cuda:0'), covar=tensor([0.3761, 0.2221, 0.2090, 0.3470], device='cuda:0'), in_proj_covar=tensor([0.1452, 0.1371, 0.1424, 0.1201], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 19:21:24,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:21:26,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:21:30,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.588e+03 2.113e+03 2.813e+03 6.148e+03, threshold=4.226e+03, percent-clipped=6.0
+2023-03-03 19:21:48,792 INFO [train.py:968] (0/2) Epoch 7, batch 23900, giga_loss[loss=0.3785, simple_loss=0.4156, pruned_loss=0.1707, over 28224.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4054, pruned_loss=0.1539, over 5654043.92 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3821, pruned_loss=0.129, over 5691828.06 frames. ], giga_tot_loss[loss=0.3566, simple_loss=0.4054, pruned_loss=0.1539, over 5647205.52 frames. ], batch size: 368, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:21:54,098 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:22:38,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:22:39,642 INFO [train.py:968] (0/2) Epoch 7, batch 23950, giga_loss[loss=0.3464, simple_loss=0.4085, pruned_loss=0.1421, over 28943.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.409, pruned_loss=0.157, over 5659480.48 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3816, pruned_loss=0.1291, over 5701760.42 frames. ], giga_tot_loss[loss=0.3634, simple_loss=0.4106, pruned_loss=0.1581, over 5643573.21 frames. ], batch size: 145, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:22:41,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:22:56,081 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:23:14,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.826e+02 1.914e+03 2.777e+03 4.407e+03 1.429e+04, threshold=5.554e+03, percent-clipped=29.0
+2023-03-03 19:23:14,809 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:23:24,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:23:32,882 INFO [train.py:968] (0/2) Epoch 7, batch 24000, giga_loss[loss=0.3764, simple_loss=0.4188, pruned_loss=0.167, over 28651.00 frames. ], tot_loss[loss=0.3617, simple_loss=0.4087, pruned_loss=0.1573, over 5660440.59 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3814, pruned_loss=0.1291, over 5707017.67 frames. ], giga_tot_loss[loss=0.3642, simple_loss=0.4108, pruned_loss=0.1588, over 5642244.74 frames. ], batch size: 262, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:23:32,886 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 19:23:41,159 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2251, simple_loss=0.3303, pruned_loss=0.06, over 944034.00 frames. 
+2023-03-03 19:23:41,160 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-03 19:24:27,490 INFO [train.py:968] (0/2) Epoch 7, batch 24050, giga_loss[loss=0.2998, simple_loss=0.3607, pruned_loss=0.1194, over 28801.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4072, pruned_loss=0.1569, over 5650722.35 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3816, pruned_loss=0.1294, over 5700129.63 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.4093, pruned_loss=0.1584, over 5640731.48 frames. ], batch size: 119, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:24:33,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:24:38,358 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-03 19:24:54,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.887e+03 2.256e+03 2.903e+03 6.739e+03, threshold=4.512e+03, percent-clipped=2.0
+2023-03-03 19:25:11,925 INFO [train.py:968] (0/2) Epoch 7, batch 24100, giga_loss[loss=0.3758, simple_loss=0.4204, pruned_loss=0.1656, over 28607.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.4065, pruned_loss=0.1561, over 5643075.38 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3817, pruned_loss=0.1297, over 5694585.84 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.409, pruned_loss=0.158, over 5638508.04 frames. ], batch size: 336, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:25:42,249 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:25:45,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:25:57,645 INFO [train.py:968] (0/2) Epoch 7, batch 24150, giga_loss[loss=0.3063, simple_loss=0.3777, pruned_loss=0.1175, over 28915.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4054, pruned_loss=0.1535, over 5650525.40 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3815, pruned_loss=0.1295, over 5700245.48 frames. ], giga_tot_loss[loss=0.3601, simple_loss=0.4083, pruned_loss=0.156, over 5640201.19 frames. ], batch size: 112, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:26:16,270 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:26:33,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.299e+02 1.810e+03 2.372e+03 3.325e+03 8.039e+03, threshold=4.743e+03, percent-clipped=10.0
+2023-03-03 19:26:49,991 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:26:50,420 INFO [train.py:968] (0/2) Epoch 7, batch 24200, libri_loss[loss=0.3817, simple_loss=0.4203, pruned_loss=0.1716, over 19641.00 frames. ], tot_loss[loss=0.3582, simple_loss=0.4071, pruned_loss=0.1547, over 5635630.63 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3818, pruned_loss=0.1297, over 5694252.03 frames. ], giga_tot_loss[loss=0.3617, simple_loss=0.4096, pruned_loss=0.1569, over 5632384.22 frames. ], batch size: 187, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:26:53,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:27:04,324 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-03 19:27:09,315 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 19:27:21,673 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:27:36,875 INFO [train.py:968] (0/2) Epoch 7, batch 24250, giga_loss[loss=0.3442, simple_loss=0.3975, pruned_loss=0.1454, over 28877.00 frames. ], tot_loss[loss=0.356, simple_loss=0.4052, pruned_loss=0.1534, over 5633642.64 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3807, pruned_loss=0.1291, over 5702117.84 frames. ], giga_tot_loss[loss=0.3613, simple_loss=0.4091, pruned_loss=0.1567, over 5621298.28 frames. ], batch size: 112, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:28:08,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.430e+02 1.477e+03 1.934e+03 2.569e+03 4.373e+03, threshold=3.869e+03, percent-clipped=0.0
+2023-03-03 19:28:24,524 INFO [train.py:968] (0/2) Epoch 7, batch 24300, giga_loss[loss=0.3341, simple_loss=0.3944, pruned_loss=0.1369, over 29017.00 frames. ], tot_loss[loss=0.3488, simple_loss=0.4008, pruned_loss=0.1484, over 5634987.92 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3806, pruned_loss=0.1292, over 5700168.10 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4048, pruned_loss=0.1517, over 5624607.64 frames. ], batch size: 213, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:28:30,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8018, 1.8028, 1.4050, 1.3781], device='cuda:0'), covar=tensor([0.0786, 0.0606, 0.1009, 0.1114], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0446, 0.0496, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:29:04,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:29:15,754 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-298000.pt
+2023-03-03 19:29:16,039 INFO [train.py:968] (0/2) Epoch 7, batch 24350, giga_loss[loss=0.3476, simple_loss=0.3978, pruned_loss=0.1487, over 27489.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3975, pruned_loss=0.1442, over 5635260.22 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3808, pruned_loss=0.1294, over 5692561.66 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.4006, pruned_loss=0.1468, over 5633088.83 frames. ], batch size: 472, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:29:23,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4860, 1.3629, 4.9300, 3.7261], device='cuda:0'), covar=tensor([0.1680, 0.2436, 0.0335, 0.0623], device='cuda:0'), in_proj_covar=tensor([0.0601, 0.0561, 0.0791, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 19:29:30,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2921, 1.9631, 1.4442, 1.4698], device='cuda:0'), covar=tensor([0.0718, 0.0300, 0.0311, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0117, 0.0120, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0068, 0.0049, 0.0044, 0.0074], device='cuda:0')
+2023-03-03 19:29:37,254 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0230, 1.3287, 1.0338, 0.2684], device='cuda:0'), covar=tensor([0.1974, 0.1718, 0.2718, 0.3537], device='cuda:0'), in_proj_covar=tensor([0.1459, 0.1375, 0.1431, 0.1207], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 19:29:45,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.417e+02 1.479e+03 1.989e+03 2.530e+03 7.018e+03, threshold=3.978e+03, percent-clipped=5.0
+2023-03-03 19:30:03,485 INFO [train.py:968] (0/2) Epoch 7, batch 24400, giga_loss[loss=0.3206, simple_loss=0.3807, pruned_loss=0.1303, over 28455.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3936, pruned_loss=0.1407, over 5648476.64 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3803, pruned_loss=0.1293, over 5686467.14 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3968, pruned_loss=0.143, over 5651348.97 frames. ], batch size: 78, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:30:54,863 INFO [train.py:968] (0/2) Epoch 7, batch 24450, giga_loss[loss=0.3276, simple_loss=0.3858, pruned_loss=0.1347, over 28643.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3917, pruned_loss=0.1392, over 5649549.88 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3806, pruned_loss=0.1296, over 5689855.32 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3942, pruned_loss=0.1409, over 5648483.46 frames. ], batch size: 262, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:31:26,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.121e+02 1.618e+03 2.135e+03 2.712e+03 5.768e+03, threshold=4.270e+03, percent-clipped=9.0
+2023-03-03 19:31:26,664 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=298133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:31:28,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=298136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:31:41,227 INFO [train.py:968] (0/2) Epoch 7, batch 24500, giga_loss[loss=0.3731, simple_loss=0.4145, pruned_loss=0.1659, over 27553.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3915, pruned_loss=0.1391, over 5665205.07 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3805, pruned_loss=0.1297, over 5694214.89 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3938, pruned_loss=0.1406, over 5659763.84 frames. ], batch size: 472, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:31:53,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1535, 4.9801, 4.6635, 2.0674], device='cuda:0'), covar=tensor([0.0372, 0.0513, 0.0600, 0.2147], device='cuda:0'), in_proj_covar=tensor([0.0961, 0.0902, 0.0808, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 19:31:54,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=298165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:32:30,117 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-03 19:32:33,776 INFO [train.py:968] (0/2) Epoch 7, batch 24550, giga_loss[loss=0.3294, simple_loss=0.3934, pruned_loss=0.1327, over 29026.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3914, pruned_loss=0.1388, over 5655417.28 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3805, pruned_loss=0.1298, over 5684764.50 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3934, pruned_loss=0.1401, over 5659017.74 frames. ], batch size: 155, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:33:07,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.453e+02 1.410e+03 1.932e+03 2.615e+03 9.359e+03, threshold=3.864e+03, percent-clipped=5.0
+2023-03-03 19:33:24,188 INFO [train.py:968] (0/2) Epoch 7, batch 24600, libri_loss[loss=0.2786, simple_loss=0.3426, pruned_loss=0.1073, over 29361.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3885, pruned_loss=0.1359, over 5662307.02 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3802, pruned_loss=0.1296, over 5688993.17 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3905, pruned_loss=0.1372, over 5660669.19 frames. ], batch size: 67, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:33:42,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8122, 1.7346, 1.2029, 1.4252], device='cuda:0'), covar=tensor([0.0666, 0.0571, 0.1016, 0.0984], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0446, 0.0496, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:34:03,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7579, 2.3253, 1.6066, 1.2188], device='cuda:0'), covar=tensor([0.1772, 0.1088, 0.1326, 0.1746], device='cuda:0'), in_proj_covar=tensor([0.1559, 0.1393, 0.1384, 0.1463], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 19:34:14,958 INFO [train.py:968] (0/2) Epoch 7, batch 24650, giga_loss[loss=0.3324, simple_loss=0.4001, pruned_loss=0.1323, over 28357.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3897, pruned_loss=0.1339, over 5669422.05 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3805, pruned_loss=0.1299, over 5681408.90 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3912, pruned_loss=0.1349, over 5674233.24 frames. ], batch size: 368, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:34:23,321 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 19:34:50,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.274e+02 1.590e+03 2.246e+03 3.984e+03 2.764e+04, threshold=4.492e+03, percent-clipped=25.0
+2023-03-03 19:35:08,771 INFO [train.py:968] (0/2) Epoch 7, batch 24700, giga_loss[loss=0.3028, simple_loss=0.3807, pruned_loss=0.1124, over 28944.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3912, pruned_loss=0.1348, over 5645031.06 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3809, pruned_loss=0.1303, over 5676297.91 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3923, pruned_loss=0.1353, over 5653002.85 frames. ], batch size: 136, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:35:09,055 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=298350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:35:42,254 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4418, 1.7976, 1.6960, 1.3512], device='cuda:0'), covar=tensor([0.1502, 0.1890, 0.1154, 0.1323], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0716, 0.0811, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 19:35:54,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.68 vs. limit=5.0
+2023-03-03 19:35:55,153 INFO [train.py:968] (0/2) Epoch 7, batch 24750, giga_loss[loss=0.3636, simple_loss=0.3978, pruned_loss=0.1647, over 23277.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3925, pruned_loss=0.1365, over 5629226.83 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.381, pruned_loss=0.1308, over 5655895.27 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3937, pruned_loss=0.1366, over 5653296.52 frames. ], batch size: 705, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:36:12,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2869, 1.4005, 1.4722, 1.3213], device='cuda:0'), covar=tensor([0.1110, 0.1332, 0.1579, 0.1301], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0727, 0.0645, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 19:36:28,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.953e+02 1.509e+03 1.932e+03 2.458e+03 5.729e+03, threshold=3.864e+03, percent-clipped=5.0
+2023-03-03 19:36:40,609 INFO [train.py:968] (0/2) Epoch 7, batch 24800, giga_loss[loss=0.2969, simple_loss=0.3667, pruned_loss=0.1135, over 28464.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.393, pruned_loss=0.1377, over 5639732.99 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3811, pruned_loss=0.1309, over 5659392.32 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.394, pruned_loss=0.1377, over 5655124.47 frames. ], batch size: 65, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:37:28,155 INFO [train.py:968] (0/2) Epoch 7, batch 24850, giga_loss[loss=0.358, simple_loss=0.4087, pruned_loss=0.1536, over 29063.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3908, pruned_loss=0.1375, over 5631448.51 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3815, pruned_loss=0.1313, over 5655060.12 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3915, pruned_loss=0.1373, over 5647397.81 frames. ], batch size: 128, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:37:44,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9724, 3.4443, 2.3227, 1.2091], device='cuda:0'), covar=tensor([0.3455, 0.1314, 0.1643, 0.2498], device='cuda:0'), in_proj_covar=tensor([0.1474, 0.1379, 0.1436, 0.1211], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 19:37:59,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.961e+02 1.635e+03 2.138e+03 3.542e+03 1.144e+04, threshold=4.277e+03, percent-clipped=16.0
+2023-03-03 19:38:13,701 INFO [train.py:968] (0/2) Epoch 7, batch 24900, giga_loss[loss=0.3471, simple_loss=0.3988, pruned_loss=0.1477, over 28903.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3887, pruned_loss=0.1363, over 5652397.60 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3811, pruned_loss=0.1311, over 5657546.17 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3896, pruned_loss=0.1364, over 5662707.00 frames. ], batch size: 186, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:38:48,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6011, 1.5978, 1.3252, 1.9718], device='cuda:0'), covar=tensor([0.2133, 0.2201, 0.2341, 0.2077], device='cuda:0'), in_proj_covar=tensor([0.1187, 0.0900, 0.1047, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 19:38:58,485 INFO [train.py:968] (0/2) Epoch 7, batch 24950, giga_loss[loss=0.3079, simple_loss=0.3775, pruned_loss=0.1191, over 28940.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3873, pruned_loss=0.1348, over 5661112.33 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3811, pruned_loss=0.1313, over 5662942.08 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3882, pruned_loss=0.1347, over 5664251.83 frames. ], batch size: 227, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:39:28,854 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.427e+02 1.526e+03 1.979e+03 2.345e+03 6.808e+03, threshold=3.959e+03, percent-clipped=3.0
+2023-03-03 19:39:43,057 INFO [train.py:968] (0/2) Epoch 7, batch 25000, giga_loss[loss=0.3492, simple_loss=0.4129, pruned_loss=0.1427, over 29038.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3872, pruned_loss=0.134, over 5669168.08 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3813, pruned_loss=0.1316, over 5669106.47 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.388, pruned_loss=0.1337, over 5666149.27 frames. ], batch size: 155, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:40:31,773 INFO [train.py:968] (0/2) Epoch 7, batch 25050, giga_loss[loss=0.3098, simple_loss=0.3819, pruned_loss=0.1188, over 29062.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3882, pruned_loss=0.1346, over 5670345.86 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3814, pruned_loss=0.1316, over 5672859.69 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3889, pruned_loss=0.1344, over 5664543.25 frames. ], batch size: 155, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:40:54,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=298725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:41:04,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.783e+02 1.611e+03 2.113e+03 3.140e+03 9.152e+03, threshold=4.226e+03, percent-clipped=14.0
+2023-03-03 19:41:17,638 INFO [train.py:968] (0/2) Epoch 7, batch 25100, giga_loss[loss=0.3021, simple_loss=0.3676, pruned_loss=0.1183, over 28930.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3872, pruned_loss=0.1341, over 5674722.66 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3818, pruned_loss=0.132, over 5677822.77 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3875, pruned_loss=0.1338, over 5665350.62 frames. ], batch size: 145, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:42:05,989 INFO [train.py:968] (0/2) Epoch 7, batch 25150, giga_loss[loss=0.2696, simple_loss=0.3447, pruned_loss=0.09724, over 28776.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3851, pruned_loss=0.1325, over 5687270.58 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3819, pruned_loss=0.1319, over 5679935.18 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3853, pruned_loss=0.1323, over 5678036.89 frames. ], batch size: 119, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:42:36,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.642e+03 1.967e+03 2.542e+03 7.172e+03, threshold=3.935e+03, percent-clipped=7.0
+2023-03-03 19:42:49,015 INFO [train.py:968] (0/2) Epoch 7, batch 25200, giga_loss[loss=0.3448, simple_loss=0.3759, pruned_loss=0.1568, over 23507.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3859, pruned_loss=0.1341, over 5687138.28 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3825, pruned_loss=0.1324, over 5688157.32 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3857, pruned_loss=0.1336, over 5672060.55 frames. ], batch size: 705, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:43:04,075 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=298868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:43:07,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=298871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:43:11,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=298875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:43:32,250 INFO [train.py:968] (0/2) Epoch 7, batch 25250, giga_loss[loss=0.3362, simple_loss=0.3823, pruned_loss=0.1451, over 27980.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.385, pruned_loss=0.134, over 5702903.84 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3824, pruned_loss=0.1325, over 5693708.87 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.385, pruned_loss=0.1336, over 5686008.51 frames. ], batch size: 412, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:43:32,591 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=298900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:43:55,016 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=298923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:44:06,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.875e+02 1.711e+03 2.278e+03 3.699e+03 8.069e+03, threshold=4.555e+03, percent-clipped=17.0
+2023-03-03 19:44:22,292 INFO [train.py:968] (0/2) Epoch 7, batch 25300, giga_loss[loss=0.411, simple_loss=0.4345, pruned_loss=0.1938, over 27636.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3835, pruned_loss=0.1335, over 5692446.54 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3825, pruned_loss=0.1325, over 5694881.67 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3835, pruned_loss=0.1332, over 5678069.86 frames. ], batch size: 472, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:44:48,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5180, 1.5310, 1.0795, 1.2150], device='cuda:0'), covar=tensor([0.0698, 0.0507, 0.1035, 0.0953], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0447, 0.0498, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:45:06,773 INFO [train.py:968] (0/2) Epoch 7, batch 25350, giga_loss[loss=0.3437, simple_loss=0.3935, pruned_loss=0.147, over 28613.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3817, pruned_loss=0.1324, over 5695962.44 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3826, pruned_loss=0.1325, over 5695857.98 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3817, pruned_loss=0.1322, over 5683641.81 frames. ], batch size: 78, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:45:30,308 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299022.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:45:43,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.645e+02 1.788e+03 2.212e+03 3.035e+03 1.144e+04, threshold=4.424e+03, percent-clipped=9.0
+2023-03-03 19:45:52,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5718, 4.3457, 4.0693, 1.7194], device='cuda:0'), covar=tensor([0.0517, 0.0733, 0.0838, 0.2098], device='cuda:0'), in_proj_covar=tensor([0.0949, 0.0899, 0.0801, 0.0627], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 19:45:56,321 INFO [train.py:968] (0/2) Epoch 7, batch 25400, giga_loss[loss=0.3151, simple_loss=0.3829, pruned_loss=0.1236, over 28632.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3821, pruned_loss=0.133, over 5682571.14 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3824, pruned_loss=0.1324, over 5689433.05 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3822, pruned_loss=0.1329, over 5678090.88 frames. ], batch size: 262, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:46:26,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:46:40,439 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-03 19:46:41,263 INFO [train.py:968] (0/2) Epoch 7, batch 25450, libri_loss[loss=0.3665, simple_loss=0.4057, pruned_loss=0.1636, over 19357.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3829, pruned_loss=0.1329, over 5679993.85 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3828, pruned_loss=0.1329, over 5686451.53 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3826, pruned_loss=0.1323, over 5679781.49 frames. ], batch size: 186, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:46:42,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3649, 1.8461, 1.2739, 0.7206], device='cuda:0'), covar=tensor([0.2703, 0.1529, 0.1843, 0.3317], device='cuda:0'), in_proj_covar=tensor([0.1464, 0.1376, 0.1419, 0.1197], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 19:47:11,527 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.930e+02 1.578e+03 2.063e+03 2.896e+03 6.514e+03, threshold=4.125e+03, percent-clipped=8.0
+2023-03-03 19:47:24,800 INFO [train.py:968] (0/2) Epoch 7, batch 25500, giga_loss[loss=0.3101, simple_loss=0.3781, pruned_loss=0.1211, over 28819.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3822, pruned_loss=0.1315, over 5683320.69 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3826, pruned_loss=0.1327, over 5690434.30 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3822, pruned_loss=0.1312, over 5679380.05 frames. ], batch size: 99, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:47:47,337 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-03 19:48:11,546 INFO [train.py:968] (0/2) Epoch 7, batch 25550, giga_loss[loss=0.471, simple_loss=0.4788, pruned_loss=0.2316, over 26475.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3829, pruned_loss=0.1315, over 5677480.37 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3828, pruned_loss=0.133, over 5686559.75 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3826, pruned_loss=0.131, over 5677145.11 frames. ], batch size: 555, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:48:18,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3222, 1.4360, 1.4256, 1.3302], device='cuda:0'), covar=tensor([0.1233, 0.1437, 0.1824, 0.1457], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0741, 0.0658, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-03 19:48:31,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8822, 1.7804, 1.2816, 1.4762], device='cuda:0'), covar=tensor([0.0612, 0.0511, 0.0881, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0450, 0.0497, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:48:43,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.578e+03 1.976e+03 2.856e+03 5.925e+03, threshold=3.952e+03, percent-clipped=7.0
+2023-03-03 19:48:49,375 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-03 19:48:55,722 INFO [train.py:968] (0/2) Epoch 7, batch 25600, giga_loss[loss=0.3481, simple_loss=0.4016, pruned_loss=0.1473, over 28558.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3858, pruned_loss=0.1344, over 5673862.04 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3828, pruned_loss=0.1332, over 5682163.06 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3856, pruned_loss=0.1338, over 5677954.19 frames. ], batch size: 307, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:48:55,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:49:38,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1484, 1.5576, 1.4505, 1.1502], device='cuda:0'), covar=tensor([0.1167, 0.1747, 0.0980, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0719, 0.0814, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 19:49:43,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:49:45,775 INFO [train.py:968] (0/2) Epoch 7, batch 25650, giga_loss[loss=0.3577, simple_loss=0.4008, pruned_loss=0.1572, over 28599.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3884, pruned_loss=0.1375, over 5678477.07 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.3827, pruned_loss=0.1331, over 5684511.86 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3884, pruned_loss=0.1371, over 5679719.89 frames. ], batch size: 307, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:50:09,935 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3256, 1.3872, 1.2240, 1.6153], device='cuda:0'), covar=tensor([0.0729, 0.0318, 0.0315, 0.0734], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0117, 0.0120, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0067, 0.0049, 0.0044, 0.0074], device='cuda:0')
+2023-03-03 19:50:19,245 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.712e+03 2.535e+03 4.057e+03 1.019e+04, threshold=5.070e+03, percent-clipped=25.0
+2023-03-03 19:50:20,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1449, 4.9872, 4.6966, 2.1546], device='cuda:0'), covar=tensor([0.0398, 0.0542, 0.0698, 0.1879], device='cuda:0'), in_proj_covar=tensor([0.0962, 0.0908, 0.0813, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 19:50:34,391 INFO [train.py:968] (0/2) Epoch 7, batch 25700, giga_loss[loss=0.2759, simple_loss=0.3467, pruned_loss=0.1026, over 28825.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3894, pruned_loss=0.1401, over 5669531.60 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3822, pruned_loss=0.1329, over 5680913.93 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3899, pruned_loss=0.1401, over 5672980.75 frames. ], batch size: 119, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:51:21,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:51:24,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:51:25,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299397.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:51:29,370 INFO [train.py:968] (0/2) Epoch 7, batch 25750, giga_loss[loss=0.3124, simple_loss=0.3777, pruned_loss=0.1235, over 28918.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3906, pruned_loss=0.1418, over 5662688.80 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3826, pruned_loss=0.1333, over 5670281.22 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3908, pruned_loss=0.1416, over 5674289.44 frames. ], batch size: 164, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:51:50,307 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:52:00,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.693e+03 2.080e+03 2.669e+03 1.021e+04, threshold=4.161e+03, percent-clipped=2.0
+2023-03-03 19:52:04,453 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:52:07,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:52:13,500 INFO [train.py:968] (0/2) Epoch 7, batch 25800, giga_loss[loss=0.2739, simple_loss=0.3414, pruned_loss=0.1032, over 28472.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3902, pruned_loss=0.1421, over 5653730.17 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.383, pruned_loss=0.1337, over 5662792.93 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.39, pruned_loss=0.1416, over 5669295.84 frames. ], batch size: 60, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:52:20,478 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:52:34,943 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:52:58,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8612, 1.8763, 1.3266, 1.5560], device='cuda:0'), covar=tensor([0.0718, 0.0649, 0.1030, 0.1027], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0453, 0.0503, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 19:52:59,700 INFO [train.py:968] (0/2) Epoch 7, batch 25850, giga_loss[loss=0.3491, simple_loss=0.4101, pruned_loss=0.1441, over 28500.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3889, pruned_loss=0.1407, over 5658948.99 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3832, pruned_loss=0.1338, over 5667388.26 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3887, pruned_loss=0.1403, over 5666887.03 frames. ], batch size: 336, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:53:16,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:53:31,728 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.590e+02 1.493e+03 1.868e+03 2.468e+03 5.474e+03, threshold=3.736e+03, percent-clipped=6.0
+2023-03-03 19:53:34,923 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299540.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:53:38,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299543.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:53:44,125 INFO [train.py:968] (0/2) Epoch 7, batch 25900, giga_loss[loss=0.2791, simple_loss=0.3569, pruned_loss=0.1007, over 28849.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3871, pruned_loss=0.1373, over 5669224.96 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3833, pruned_loss=0.1338, over 5671376.02 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3869, pruned_loss=0.1371, over 5671992.50 frames. ], batch size: 99, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:54:04,933 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299572.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:54:16,390 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:54:27,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4649, 2.0552, 1.5297, 0.7058], device='cuda:0'), covar=tensor([0.2723, 0.1843, 0.2225, 0.3273], device='cuda:0'), in_proj_covar=tensor([0.1440, 0.1367, 0.1405, 0.1189], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 19:54:29,808 INFO [train.py:968] (0/2) Epoch 7, batch 25950, giga_loss[loss=0.3289, simple_loss=0.387, pruned_loss=0.1354, over 28730.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3837, pruned_loss=0.1349, over 5652768.73 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3833, pruned_loss=0.1339, over 5664761.77 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3837, pruned_loss=0.1348, over 5660572.87 frames. ], batch size: 284, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 19:54:30,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:54:32,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:55:01,161 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:55:04,250 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:55:04,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.797e+02 1.559e+03 2.012e+03 3.427e+03 1.948e+04, threshold=4.024e+03, percent-clipped=22.0
+2023-03-03 19:55:09,952 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.08 vs. limit=2.0
+2023-03-03 19:55:14,720 INFO [train.py:968] (0/2) Epoch 7, batch 26000, giga_loss[loss=0.3501, simple_loss=0.3989, pruned_loss=0.1507, over 28269.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3815, pruned_loss=0.1336, over 5662414.14 frames. ], libri_tot_loss[loss=0.325, simple_loss=0.3829, pruned_loss=0.1335, over 5670633.36 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3817, pruned_loss=0.1338, over 5662896.06 frames. ], batch size: 368, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 19:56:05,126 INFO [train.py:968] (0/2) Epoch 7, batch 26050, giga_loss[loss=0.3081, simple_loss=0.3716, pruned_loss=0.1223, over 29041.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3817, pruned_loss=0.1348, over 5645739.87 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3826, pruned_loss=0.1333, over 5670680.49 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3822, pruned_loss=0.1352, over 5645860.61 frames. ], batch size: 155, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 19:56:20,453 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:56:38,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.605e+02 1.470e+03 1.945e+03 2.897e+03 8.225e+03, threshold=3.890e+03, percent-clipped=14.0
+2023-03-03 19:56:51,315 INFO [train.py:968] (0/2) Epoch 7, batch 26100, giga_loss[loss=0.3133, simple_loss=0.3816, pruned_loss=0.1225, over 28777.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3846, pruned_loss=0.1372, over 5650968.34 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3829, pruned_loss=0.1337, over 5672704.18 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3847, pruned_loss=0.1372, over 5648962.65 frames. ], batch size: 262, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 19:57:34,735 INFO [train.py:968] (0/2) Epoch 7, batch 26150, giga_loss[loss=0.3106, simple_loss=0.382, pruned_loss=0.1196, over 28764.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3899, pruned_loss=0.1385, over 5654154.01 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3837, pruned_loss=0.1343, over 5665129.02 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3894, pruned_loss=0.1381, over 5658082.72 frames. ], batch size: 284, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 19:58:12,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.381e+03 1.715e+03 2.416e+03 4.435e+03, threshold=3.430e+03, percent-clipped=5.0
+2023-03-03 19:58:21,878 INFO [train.py:968] (0/2) Epoch 7, batch 26200, libri_loss[loss=0.3191, simple_loss=0.3792, pruned_loss=0.1295, over 29678.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3932, pruned_loss=0.1387, over 5662085.05 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3837, pruned_loss=0.1343, over 5671673.64 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.393, pruned_loss=0.1384, over 5658938.78 frames. ], batch size: 91, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 19:58:33,030 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-03 19:58:36,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4886, 1.6076, 1.6104, 1.4677], device='cuda:0'), covar=tensor([0.1284, 0.1727, 0.1783, 0.1661], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0725, 0.0645, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 19:58:58,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:59:04,820 INFO [train.py:968] (0/2) Epoch 7, batch 26250, giga_loss[loss=0.3691, simple_loss=0.4208, pruned_loss=0.1587, over 28870.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3936, pruned_loss=0.1392, over 5660443.39 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3838, pruned_loss=0.1345, over 5675592.60 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3937, pruned_loss=0.139, over 5654077.69 frames. ], batch size: 285, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 19:59:43,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.396e+02 1.667e+03 2.132e+03 3.455e+03 7.227e+03, threshold=4.263e+03, percent-clipped=25.0
+2023-03-03 19:59:53,768 INFO [train.py:968] (0/2) Epoch 7, batch 26300, giga_loss[loss=0.4472, simple_loss=0.4621, pruned_loss=0.2162, over 27528.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3959, pruned_loss=0.1415, over 5660887.65 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3838, pruned_loss=0.1344, over 5679039.84 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3961, pruned_loss=0.1414, over 5652667.50 frames. ], batch size: 472, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 19:59:59,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2084, 0.9887, 4.6158, 3.4191], device='cuda:0'), covar=tensor([0.1721, 0.2642, 0.0357, 0.0760], device='cuda:0'), in_proj_covar=tensor([0.0605, 0.0561, 0.0797, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 20:00:00,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7677, 2.2635, 1.6641, 1.6647], device='cuda:0'), covar=tensor([0.0629, 0.0218, 0.0291, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0120, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0068, 0.0049, 0.0044, 0.0074], device='cuda:0')
+2023-03-03 20:00:00,700 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:00:40,038 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-300000.pt
+2023-03-03 20:00:40,344 INFO [train.py:968] (0/2) Epoch 7, batch 26350, giga_loss[loss=0.3131, simple_loss=0.3813, pruned_loss=0.1224, over 28692.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3948, pruned_loss=0.1415, over 5652273.84 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3833, pruned_loss=0.1342, over 5686262.16 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3958, pruned_loss=0.1419, over 5638661.16 frames. ], batch size: 60, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:00:50,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4059, 1.9490, 1.4164, 1.1707], device='cuda:0'), covar=tensor([0.1734, 0.1227, 0.1337, 0.1537], device='cuda:0'), in_proj_covar=tensor([0.1556, 0.1407, 0.1371, 0.1467], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 20:00:50,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:01:16,455 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:01:16,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.453e+02 1.326e+03 1.755e+03 2.403e+03 7.241e+03, threshold=3.510e+03, percent-clipped=4.0
+2023-03-03 20:01:20,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:01:33,181 INFO [train.py:968] (0/2) Epoch 7, batch 26400, giga_loss[loss=0.3884, simple_loss=0.4112, pruned_loss=0.1828, over 23665.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3928, pruned_loss=0.1405, over 5648993.66 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3833, pruned_loss=0.1342, over 5687374.85 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3936, pruned_loss=0.1408, over 5637091.49 frames. ], batch size: 705, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:01:48,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:02:08,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:02:14,118 INFO [train.py:968] (0/2) Epoch 7, batch 26450, giga_loss[loss=0.3365, simple_loss=0.3881, pruned_loss=0.1425, over 28702.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3898, pruned_loss=0.1384, over 5660414.80 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3834, pruned_loss=0.1341, over 5692916.14 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3905, pruned_loss=0.1389, over 5644968.22 frames. ], batch size: 242, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:02:18,718 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:02:20,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:02:46,064 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:02:48,990 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.543e+03 2.049e+03 2.806e+03 1.348e+04, threshold=4.097e+03, percent-clipped=18.0
+2023-03-03 20:02:51,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5582, 1.4888, 1.1467, 1.1866], device='cuda:0'), covar=tensor([0.0671, 0.0524, 0.1001, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0449, 0.0495, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 20:03:01,747 INFO [train.py:968] (0/2) Epoch 7, batch 26500, giga_loss[loss=0.4139, simple_loss=0.4474, pruned_loss=0.1902, over 27919.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3894, pruned_loss=0.1393, over 5661513.71 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.384, pruned_loss=0.1347, over 5693114.31 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3897, pruned_loss=0.1393, over 5647813.44 frames. ], batch size: 412, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:03:08,717 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:03:10,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:03:15,809 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=300162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:03:35,625 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:03:40,617 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=300191.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 20:03:50,082 INFO [train.py:968] (0/2) Epoch 7, batch 26550, giga_loss[loss=0.3484, simple_loss=0.396, pruned_loss=0.1504, over 28543.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3889, pruned_loss=0.1396, over 5639966.68 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3842, pruned_loss=0.1349, over 5678985.17 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3891, pruned_loss=0.1396, over 5639702.76 frames. ], batch size: 336, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:04:20,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:04:24,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:04:25,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.644e+02 1.445e+03 1.937e+03 2.655e+03 5.019e+03, threshold=3.875e+03, percent-clipped=6.0
+2023-03-03 20:04:35,253 INFO [train.py:968] (0/2) Epoch 7, batch 26600, giga_loss[loss=0.2661, simple_loss=0.3429, pruned_loss=0.09463, over 28402.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.389, pruned_loss=0.1396, over 5643803.49 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3844, pruned_loss=0.1349, over 5679151.89 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3891, pruned_loss=0.1397, over 5642199.48 frames. ], batch size: 60, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:04:50,578 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:05:19,880 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-03 20:05:21,162 INFO [train.py:968] (0/2) Epoch 7, batch 26650, giga_loss[loss=0.2847, simple_loss=0.354, pruned_loss=0.1077, over 28713.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3883, pruned_loss=0.1395, over 5659040.68 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3851, pruned_loss=0.1353, over 5678944.22 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3879, pruned_loss=0.1392, over 5657507.33 frames. ], batch size: 242, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:05:56,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.665e+02 1.722e+03 2.123e+03 2.970e+03 6.110e+03, threshold=4.247e+03, percent-clipped=9.0
+2023-03-03 20:06:06,776 INFO [train.py:968] (0/2) Epoch 7, batch 26700, giga_loss[loss=0.2841, simple_loss=0.3497, pruned_loss=0.1093, over 28531.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3864, pruned_loss=0.1382, over 5665319.06 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3843, pruned_loss=0.1348, over 5677501.14 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3869, pruned_loss=0.1387, over 5664078.93 frames. ], batch size: 78, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:06:14,029 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-03 20:06:21,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9893, 1.3119, 1.0502, 0.1579], device='cuda:0'), covar=tensor([0.1722, 0.1526, 0.2544, 0.3215], device='cuda:0'), in_proj_covar=tensor([0.1432, 0.1352, 0.1399, 0.1183], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 20:06:55,779 INFO [train.py:968] (0/2) Epoch 7, batch 26750, giga_loss[loss=0.2927, simple_loss=0.3719, pruned_loss=0.1068, over 28844.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3865, pruned_loss=0.1371, over 5662108.65 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3841, pruned_loss=0.1346, over 5676214.38 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.387, pruned_loss=0.1377, over 5661834.00 frames. ], batch size: 186, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:07:32,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.323e+02 1.630e+03 2.159e+03 3.113e+03 6.202e+03, threshold=4.317e+03, percent-clipped=10.0
+2023-03-03 20:07:43,492 INFO [train.py:968] (0/2) Epoch 7, batch 26800, giga_loss[loss=0.3342, simple_loss=0.3899, pruned_loss=0.1392, over 28528.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3909, pruned_loss=0.1403, over 5663026.66 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3849, pruned_loss=0.1352, over 5679204.31 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3907, pruned_loss=0.1402, over 5660020.99 frames. ], batch size: 78, lr: 4.58e-03, grad_scale: 8.0
+2023-03-03 20:07:45,727 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-03 20:08:01,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=300466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:08:33,075 INFO [train.py:968] (0/2) Epoch 7, batch 26850, giga_loss[loss=0.4525, simple_loss=0.4624, pruned_loss=0.2213, over 26694.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3916, pruned_loss=0.1417, over 5659464.96 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3855, pruned_loss=0.1356, over 5684069.00 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3912, pruned_loss=0.1415, over 5652111.91 frames. ], batch size: 555, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:09:07,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:09:08,817 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.448e+02 1.440e+03 1.871e+03 2.745e+03 8.762e+03, threshold=3.741e+03, percent-clipped=5.0
+2023-03-03 20:09:17,901 INFO [train.py:968] (0/2) Epoch 7, batch 26900, giga_loss[loss=0.3199, simple_loss=0.3827, pruned_loss=0.1286, over 28773.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3904, pruned_loss=0.1378, over 5673446.31 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3848, pruned_loss=0.1352, over 5685173.69 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3908, pruned_loss=0.1381, over 5666439.68 frames. ], batch size: 284, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:09:32,277 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300566.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 20:10:06,693 INFO [train.py:968] (0/2) Epoch 7, batch 26950, giga_loss[loss=0.3352, simple_loss=0.4015, pruned_loss=0.1345, over 28923.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3913, pruned_loss=0.1368, over 5676724.06 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3849, pruned_loss=0.1352, over 5690249.43 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3917, pruned_loss=0.1371, over 5666580.58 frames. ], batch size: 199, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:10:40,550 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.168e+02 1.243e+03 1.674e+03 2.372e+03 8.060e+03, threshold=3.348e+03, percent-clipped=5.0
+2023-03-03 20:10:49,787 INFO [train.py:968] (0/2) Epoch 7, batch 27000, giga_loss[loss=0.3533, simple_loss=0.4126, pruned_loss=0.1469, over 28707.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3932, pruned_loss=0.1365, over 5676704.45 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3848, pruned_loss=0.1351, over 5694985.93 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3937, pruned_loss=0.1368, over 5664361.77 frames. ], batch size: 242, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:10:49,792 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 20:10:57,442 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2878, 1.6168, 1.3321, 1.0513], device='cuda:0'), covar=tensor([0.1282, 0.0934, 0.0694, 0.0972], device='cuda:0'), in_proj_covar=tensor([0.1544, 0.1400, 0.1358, 0.1455], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 20:10:58,246 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2265, simple_loss=0.3303, pruned_loss=0.06131, over 944034.00 frames. 
+2023-03-03 20:10:58,247 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-03 20:11:27,352 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:11:29,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:11:34,606 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2358, 1.5638, 1.2458, 1.3402], device='cuda:0'), covar=tensor([0.2244, 0.2189, 0.2435, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.1195, 0.0906, 0.1054, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 20:11:45,757 INFO [train.py:968] (0/2) Epoch 7, batch 27050, giga_loss[loss=0.3662, simple_loss=0.4156, pruned_loss=0.1584, over 28209.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3969, pruned_loss=0.1402, over 5672047.45 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3852, pruned_loss=0.1354, over 5690527.40 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3971, pruned_loss=0.1403, over 5665843.02 frames. ], batch size: 368, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:11:56,181 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300709.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 20:11:58,203 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:11:58,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300712.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 20:12:26,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.550e+02 1.792e+03 2.272e+03 3.020e+03 6.972e+03, threshold=4.544e+03, percent-clipped=16.0
+2023-03-03 20:12:27,036 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300741.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 20:12:29,622 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 20:12:31,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 20:12:35,314 INFO [train.py:968] (0/2) Epoch 7, batch 27100, giga_loss[loss=0.465, simple_loss=0.4632, pruned_loss=0.2334, over 26518.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3985, pruned_loss=0.1425, over 5680063.53 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3857, pruned_loss=0.1359, over 5694818.93 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3986, pruned_loss=0.1423, over 5670848.81 frames. ], batch size: 555, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:13:11,064 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 20:13:24,777 INFO [train.py:968] (0/2) Epoch 7, batch 27150, libri_loss[loss=0.3273, simple_loss=0.3898, pruned_loss=0.1324, over 29379.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3978, pruned_loss=0.1426, over 5685654.07 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.386, pruned_loss=0.136, over 5700589.13 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.398, pruned_loss=0.1425, over 5672537.09 frames. ], batch size: 92, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:13:58,174 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-03 20:14:05,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.327e+02 1.548e+03 2.270e+03 3.016e+03 1.170e+04, threshold=4.540e+03, percent-clipped=5.0
+2023-03-03 20:14:07,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:14:13,606 INFO [train.py:968] (0/2) Epoch 7, batch 27200, giga_loss[loss=0.3081, simple_loss=0.3738, pruned_loss=0.1212, over 28803.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3968, pruned_loss=0.1422, over 5683899.86 frames. ], libri_tot_loss[loss=0.3291, simple_loss=0.3857, pruned_loss=0.1362, over 5706598.09 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3975, pruned_loss=0.1422, over 5667011.77 frames. ], batch size: 186, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:14:57,583 INFO [train.py:968] (0/2) Epoch 7, batch 27250, giga_loss[loss=0.3913, simple_loss=0.4117, pruned_loss=0.1854, over 23540.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3959, pruned_loss=0.1404, over 5678987.65 frames. ], libri_tot_loss[loss=0.3293, simple_loss=0.3858, pruned_loss=0.1364, over 5705274.62 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3966, pruned_loss=0.1403, over 5665711.93 frames. ], batch size: 705, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:15:34,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9382, 3.7724, 3.5502, 1.8889], device='cuda:0'), covar=tensor([0.0559, 0.0684, 0.0751, 0.1967], device='cuda:0'), in_proj_covar=tensor([0.0958, 0.0914, 0.0809, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 20:15:35,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.809e+02 1.599e+03 2.189e+03 2.855e+03 1.297e+04, threshold=4.378e+03, percent-clipped=8.0
+2023-03-03 20:15:44,657 INFO [train.py:968] (0/2) Epoch 7, batch 27300, libri_loss[loss=0.3228, simple_loss=0.3945, pruned_loss=0.1255, over 29549.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3972, pruned_loss=0.1404, over 5669309.72 frames. ], libri_tot_loss[loss=0.3292, simple_loss=0.3859, pruned_loss=0.1362, over 5709344.18 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.398, pruned_loss=0.1405, over 5654611.35 frames. ], batch size: 83, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:16:15,761 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:16:20,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:16:32,588 INFO [train.py:968] (0/2) Epoch 7, batch 27350, giga_loss[loss=0.3918, simple_loss=0.4268, pruned_loss=0.1784, over 27524.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3975, pruned_loss=0.1403, over 5676688.72 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3855, pruned_loss=0.1361, over 5709712.39 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3987, pruned_loss=0.1406, over 5663769.83 frames. ], batch size: 472, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:16:43,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=301010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:16:50,750 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=301016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:16:57,288 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-03 20:17:14,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.382e+02 1.557e+03 2.238e+03 3.001e+03 8.092e+03, threshold=4.476e+03, percent-clipped=9.0
+2023-03-03 20:17:21,311 INFO [train.py:968] (0/2) Epoch 7, batch 27400, giga_loss[loss=0.3375, simple_loss=0.4005, pruned_loss=0.1373, over 28908.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3965, pruned_loss=0.1405, over 5665382.12 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.385, pruned_loss=0.1357, over 5711005.89 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3982, pruned_loss=0.1412, over 5652916.74 frames. ], batch size: 174, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:18:07,954 INFO [train.py:968] (0/2) Epoch 7, batch 27450, giga_loss[loss=0.3641, simple_loss=0.4009, pruned_loss=0.1637, over 27572.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3958, pruned_loss=0.1409, over 5667582.48 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3854, pruned_loss=0.1361, over 5712738.72 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3972, pruned_loss=0.1413, over 5654424.54 frames. ], batch size: 472, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:18:49,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.319e+02 1.647e+03 2.082e+03 2.746e+03 8.662e+03, threshold=4.164e+03, percent-clipped=7.0
+2023-03-03 20:18:56,217 INFO [train.py:968] (0/2) Epoch 7, batch 27500, giga_loss[loss=0.3006, simple_loss=0.3691, pruned_loss=0.1161, over 29042.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3933, pruned_loss=0.1398, over 5680236.00 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3847, pruned_loss=0.1357, over 5714806.33 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3954, pruned_loss=0.1407, over 5666485.09 frames. ], batch size: 106, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:19:17,457 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 20:19:41,279 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=301195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:19:44,967 INFO [train.py:968] (0/2) Epoch 7, batch 27550, giga_loss[loss=0.3023, simple_loss=0.3642, pruned_loss=0.1202, over 28657.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3901, pruned_loss=0.1376, over 5684146.36 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3848, pruned_loss=0.1356, over 5722108.84 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.392, pruned_loss=0.1385, over 5665177.73 frames. ], batch size: 99, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:20:22,080 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.600e+02 1.618e+03 2.176e+03 3.996e+03 8.058e+03, threshold=4.353e+03, percent-clipped=23.0
+2023-03-03 20:20:29,846 INFO [train.py:968] (0/2) Epoch 7, batch 27600, giga_loss[loss=0.4214, simple_loss=0.4392, pruned_loss=0.2018, over 26656.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3886, pruned_loss=0.1373, over 5678040.50 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3842, pruned_loss=0.1351, over 5726716.41 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.391, pruned_loss=0.1385, over 5656244.07 frames. ], batch size: 555, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:21:13,428 INFO [train.py:968] (0/2) Epoch 7, batch 27650, giga_loss[loss=0.3164, simple_loss=0.3854, pruned_loss=0.1237, over 29055.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3885, pruned_loss=0.1382, over 5673520.20 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3839, pruned_loss=0.1351, over 5721598.01 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3908, pruned_loss=0.1393, over 5659985.50 frames. ], batch size: 164, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:21:40,220 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-03 20:21:43,038 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-03 20:21:51,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.927e+02 1.438e+03 1.974e+03 2.821e+03 5.777e+03, threshold=3.947e+03, percent-clipped=5.0
+2023-03-03 20:21:58,625 INFO [train.py:968] (0/2) Epoch 7, batch 27700, giga_loss[loss=0.2865, simple_loss=0.3575, pruned_loss=0.1077, over 28750.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3857, pruned_loss=0.1359, over 5673072.79 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3834, pruned_loss=0.1348, over 5725051.43 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3879, pruned_loss=0.1371, over 5658080.73 frames. ], batch size: 119, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:22:31,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=301385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:22:45,297 INFO [train.py:968] (0/2) Epoch 7, batch 27750, giga_loss[loss=0.2663, simple_loss=0.3364, pruned_loss=0.09807, over 28674.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3837, pruned_loss=0.1331, over 5661242.33 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3836, pruned_loss=0.135, over 5714254.07 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3852, pruned_loss=0.1338, over 5658439.52 frames. ], batch size: 99, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:23:05,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5080, 1.4977, 1.5403, 1.3539], device='cuda:0'), covar=tensor([0.1211, 0.1821, 0.1772, 0.1641], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0730, 0.0652, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 20:23:24,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.722e+02 1.384e+03 1.827e+03 2.615e+03 1.128e+04, threshold=3.654e+03, percent-clipped=9.0
+2023-03-03 20:23:33,603 INFO [train.py:968] (0/2) Epoch 7, batch 27800, giga_loss[loss=0.3354, simple_loss=0.3973, pruned_loss=0.1368, over 28923.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3826, pruned_loss=0.1319, over 5649585.38 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3842, pruned_loss=0.1354, over 5704077.99 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3834, pruned_loss=0.1321, over 5654512.12 frames. ], batch size: 213, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:23:58,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5219, 1.7131, 1.7978, 1.4225], device='cuda:0'), covar=tensor([0.1348, 0.1945, 0.1074, 0.1269], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0725, 0.0816, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 20:24:04,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3418, 1.7332, 1.7047, 1.2800], device='cuda:0'), covar=tensor([0.1476, 0.2047, 0.1166, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0725, 0.0816, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 20:24:07,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3365, 1.7111, 1.4948, 1.5303], device='cuda:0'), covar=tensor([0.0749, 0.0285, 0.0296, 0.0776], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0120, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0068, 0.0049, 0.0044, 0.0074], device='cuda:0')
+2023-03-03 20:24:27,132 INFO [train.py:968] (0/2) Epoch 7, batch 27850, giga_loss[loss=0.3352, simple_loss=0.3903, pruned_loss=0.14, over 28634.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3811, pruned_loss=0.131, over 5646867.96 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3843, pruned_loss=0.1356, over 5698104.80 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3815, pruned_loss=0.1309, over 5655495.13 frames. ], batch size: 336, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:24:54,819 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=301528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:24:58,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=301531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:24:59,970 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=301533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:25:07,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1247, 1.5129, 1.4473, 1.1418], device='cuda:0'), covar=tensor([0.1036, 0.1651, 0.0882, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0721, 0.0813, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 20:25:08,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.714e+02 1.567e+03 1.996e+03 3.255e+03 1.083e+04, threshold=3.993e+03, percent-clipped=19.0
+2023-03-03 20:25:13,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3431, 1.7423, 1.3169, 1.6009], device='cuda:0'), covar=tensor([0.0732, 0.0308, 0.0320, 0.0767], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0120, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0068, 0.0049, 0.0044, 0.0074], device='cuda:0')
+2023-03-03 20:25:19,755 INFO [train.py:968] (0/2) Epoch 7, batch 27900, giga_loss[loss=0.2699, simple_loss=0.3335, pruned_loss=0.1031, over 28529.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3787, pruned_loss=0.1308, over 5642431.09 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3844, pruned_loss=0.1355, over 5700487.58 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3788, pruned_loss=0.1307, over 5645404.65 frames. ], batch size: 85, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:25:29,643 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=301560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:25:37,708 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-03 20:25:40,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=301570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:26:10,183 INFO [train.py:968] (0/2) Epoch 7, batch 27950, giga_loss[loss=0.3866, simple_loss=0.4381, pruned_loss=0.1676, over 29041.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3792, pruned_loss=0.1316, over 5646800.51 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.385, pruned_loss=0.1361, over 5704350.88 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3786, pruned_loss=0.1308, over 5643946.92 frames. ], batch size: 155, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:26:22,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5613, 4.4025, 4.2050, 2.0074], device='cuda:0'), covar=tensor([0.0422, 0.0575, 0.0621, 0.2003], device='cuda:0'), in_proj_covar=tensor([0.0967, 0.0914, 0.0816, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 20:26:44,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.072e+02 1.551e+03 2.308e+03 3.329e+03 1.158e+04, threshold=4.616e+03, percent-clipped=19.0
+2023-03-03 20:26:54,638 INFO [train.py:968] (0/2) Epoch 7, batch 28000, giga_loss[loss=0.3149, simple_loss=0.3821, pruned_loss=0.1238, over 28674.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3801, pruned_loss=0.1309, over 5662044.66 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3845, pruned_loss=0.1356, over 5708192.99 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3799, pruned_loss=0.1305, over 5654738.63 frames. ], batch size: 284, lr: 4.57e-03, grad_scale: 8.0
+2023-03-03 20:27:31,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5120, 1.3718, 4.9479, 3.5200], device='cuda:0'), covar=tensor([0.1595, 0.2457, 0.0315, 0.0715], device='cuda:0'), in_proj_covar=tensor([0.0605, 0.0566, 0.0805, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 20:27:43,454 INFO [train.py:968] (0/2) Epoch 7, batch 28050, giga_loss[loss=0.3615, simple_loss=0.3956, pruned_loss=0.1637, over 23664.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3804, pruned_loss=0.1308, over 5656836.61 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3843, pruned_loss=0.1355, over 5712306.75 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3803, pruned_loss=0.1306, over 5646355.93 frames. ], batch size: 705, lr: 4.57e-03, grad_scale: 8.0
+2023-03-03 20:27:57,030 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=301713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:27:58,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=301716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:28:20,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.792e+02 1.407e+03 1.783e+03 2.316e+03 3.798e+03, threshold=3.566e+03, percent-clipped=0.0
+2023-03-03 20:28:25,346 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=301745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:28:30,429 INFO [train.py:968] (0/2) Epoch 7, batch 28100, giga_loss[loss=0.2894, simple_loss=0.3577, pruned_loss=0.1106, over 28669.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3799, pruned_loss=0.1306, over 5655414.84 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3841, pruned_loss=0.1353, over 5715839.12 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3799, pruned_loss=0.1305, over 5643029.20 frames. ], batch size: 242, lr: 4.57e-03, grad_scale: 8.0
+2023-03-03 20:29:14,752 INFO [train.py:968] (0/2) Epoch 7, batch 28150, giga_loss[loss=0.3405, simple_loss=0.3974, pruned_loss=0.1418, over 28571.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.381, pruned_loss=0.1319, over 5660810.38 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3843, pruned_loss=0.1353, over 5720730.57 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3807, pruned_loss=0.1317, over 5644935.50 frames. ], batch size: 307, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:29:36,244 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=301823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:29:54,299 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.595e+02 1.528e+03 1.981e+03 2.980e+03 7.085e+03, threshold=3.962e+03, percent-clipped=19.0
+2023-03-03 20:30:01,324 INFO [train.py:968] (0/2) Epoch 7, batch 28200, giga_loss[loss=0.3361, simple_loss=0.394, pruned_loss=0.1391, over 28953.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3827, pruned_loss=0.133, over 5652704.12 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3845, pruned_loss=0.1354, over 5715936.72 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3823, pruned_loss=0.1328, over 5642744.45 frames. ], batch size: 145, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:30:48,887 INFO [train.py:968] (0/2) Epoch 7, batch 28250, giga_loss[loss=0.3186, simple_loss=0.3778, pruned_loss=0.1297, over 28918.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3844, pruned_loss=0.1336, over 5653615.41 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3847, pruned_loss=0.1354, over 5710672.99 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3839, pruned_loss=0.1334, over 5648919.01 frames. ], batch size: 112, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:30:56,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=301908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:31:32,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.567e+03 2.341e+03 3.510e+03 1.151e+04, threshold=4.683e+03, percent-clipped=19.0
+2023-03-03 20:31:42,951 INFO [train.py:968] (0/2) Epoch 7, batch 28300, giga_loss[loss=0.3838, simple_loss=0.3992, pruned_loss=0.1843, over 23547.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3863, pruned_loss=0.1352, over 5648493.94 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.385, pruned_loss=0.1356, over 5711816.93 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3857, pruned_loss=0.1348, over 5643179.45 frames. ], batch size: 705, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:32:33,715 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-302000.pt
+2023-03-03 20:32:34,017 INFO [train.py:968] (0/2) Epoch 7, batch 28350, giga_loss[loss=0.3264, simple_loss=0.384, pruned_loss=0.1344, over 29003.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3875, pruned_loss=0.137, over 5638827.57 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.385, pruned_loss=0.1357, over 5703285.28 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.387, pruned_loss=0.1366, over 5641787.45 frames. ], batch size: 106, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:33:17,063 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.462e+03 1.936e+03 2.427e+03 5.316e+03, threshold=3.872e+03, percent-clipped=2.0
+2023-03-03 20:33:22,432 INFO [train.py:968] (0/2) Epoch 7, batch 28400, giga_loss[loss=0.3199, simple_loss=0.3893, pruned_loss=0.1253, over 28941.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3893, pruned_loss=0.137, over 5643142.66 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3849, pruned_loss=0.1356, over 5704902.16 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.389, pruned_loss=0.1367, over 5642555.31 frames. ], batch size: 164, lr: 4.57e-03, grad_scale: 8.0
+2023-03-03 20:33:23,322 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:33:27,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:33:57,308 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:34:10,533 INFO [train.py:968] (0/2) Epoch 7, batch 28450, giga_loss[loss=0.3814, simple_loss=0.4204, pruned_loss=0.1712, over 28057.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3886, pruned_loss=0.1363, over 5643206.53 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3846, pruned_loss=0.1355, over 5697840.32 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3888, pruned_loss=0.1362, over 5647840.40 frames. ], batch size: 412, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:34:50,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:34:59,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.327e+02 1.883e+03 2.422e+03 3.909e+03 2.626e+04, threshold=4.843e+03, percent-clipped=25.0
+2023-03-03 20:35:03,551 INFO [train.py:968] (0/2) Epoch 7, batch 28500, giga_loss[loss=0.3627, simple_loss=0.4017, pruned_loss=0.1619, over 28931.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3895, pruned_loss=0.1382, over 5630002.05 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3846, pruned_loss=0.1355, over 5698874.12 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3897, pruned_loss=0.1382, over 5632487.74 frames. ], batch size: 213, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:35:55,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:35:57,005 INFO [train.py:968] (0/2) Epoch 7, batch 28550, libri_loss[loss=0.3036, simple_loss=0.3628, pruned_loss=0.1222, over 29563.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3891, pruned_loss=0.1386, over 5624962.24 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3844, pruned_loss=0.1353, over 5692346.07 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3896, pruned_loss=0.1389, over 5629289.67 frames. ], batch size: 76, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:36:51,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.279e+02 1.501e+03 1.817e+03 2.391e+03 4.966e+03, threshold=3.633e+03, percent-clipped=1.0
+2023-03-03 20:36:56,404 INFO [train.py:968] (0/2) Epoch 7, batch 28600, giga_loss[loss=0.286, simple_loss=0.3591, pruned_loss=0.1064, over 28566.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3884, pruned_loss=0.1394, over 5617675.74 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3844, pruned_loss=0.1353, over 5693546.34 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3889, pruned_loss=0.1397, over 5619583.21 frames. ], batch size: 307, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:37:33,884 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:37:37,360 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4786, 1.6730, 1.3366, 1.6288], device='cuda:0'), covar=tensor([0.2360, 0.2217, 0.2391, 0.2099], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.0908, 0.1058, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 20:37:38,269 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:37:42,411 INFO [train.py:968] (0/2) Epoch 7, batch 28650, giga_loss[loss=0.3729, simple_loss=0.4134, pruned_loss=0.1662, over 28338.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3866, pruned_loss=0.1378, over 5642010.87 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3843, pruned_loss=0.1352, over 5699871.32 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3871, pruned_loss=0.1382, over 5636141.08 frames. ], batch size: 368, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:37:52,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6165, 1.6346, 1.5211, 1.4908], device='cuda:0'), covar=tensor([0.1124, 0.1780, 0.1778, 0.1585], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0734, 0.0654, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 20:38:22,631 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:38:25,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.622e+02 1.455e+03 1.911e+03 2.566e+03 7.332e+03, threshold=3.822e+03, percent-clipped=8.0
+2023-03-03 20:38:25,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:38:29,455 INFO [train.py:968] (0/2) Epoch 7, batch 28700, giga_loss[loss=0.2984, simple_loss=0.356, pruned_loss=0.1204, over 28686.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3861, pruned_loss=0.1378, over 5652346.81 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3841, pruned_loss=0.135, over 5703316.45 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3868, pruned_loss=0.1384, over 5642794.60 frames. ], batch size: 85, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:38:54,365 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:39:19,506 INFO [train.py:968] (0/2) Epoch 7, batch 28750, libri_loss[loss=0.2648, simple_loss=0.3322, pruned_loss=0.09867, over 29559.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3864, pruned_loss=0.138, over 5657687.06 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3843, pruned_loss=0.1351, over 5705542.53 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3868, pruned_loss=0.1384, over 5647220.06 frames. ], batch size: 74, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:39:23,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2411, 3.0617, 2.8777, 1.3417], device='cuda:0'), covar=tensor([0.0868, 0.0911, 0.0934, 0.2321], device='cuda:0'), in_proj_covar=tensor([0.0979, 0.0923, 0.0827, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-03 20:40:01,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.647e+02 1.635e+03 2.099e+03 2.580e+03 1.254e+04, threshold=4.198e+03, percent-clipped=11.0
+2023-03-03 20:40:09,858 INFO [train.py:968] (0/2) Epoch 7, batch 28800, giga_loss[loss=0.3572, simple_loss=0.4092, pruned_loss=0.1526, over 28659.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3862, pruned_loss=0.1375, over 5654343.62 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3845, pruned_loss=0.1353, over 5697217.13 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3864, pruned_loss=0.1377, over 5653005.44 frames. ], batch size: 307, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:40:20,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8509, 1.7258, 1.2735, 1.3813], device='cuda:0'), covar=tensor([0.0612, 0.0567, 0.0925, 0.0925], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0451, 0.0500, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 20:40:57,971 INFO [train.py:968] (0/2) Epoch 7, batch 28850, giga_loss[loss=0.3446, simple_loss=0.386, pruned_loss=0.1516, over 28794.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3888, pruned_loss=0.1395, over 5656467.12 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3846, pruned_loss=0.1353, over 5698582.79 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3888, pruned_loss=0.1397, over 5653596.01 frames. ], batch size: 119, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:41:12,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:41:40,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8301, 3.6441, 3.4631, 1.7304], device='cuda:0'), covar=tensor([0.0639, 0.0751, 0.0775, 0.2013], device='cuda:0'), in_proj_covar=tensor([0.0969, 0.0912, 0.0817, 0.0629], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-03 20:41:43,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.549e+02 1.616e+03 2.184e+03 2.883e+03 6.219e+03, threshold=4.369e+03, percent-clipped=6.0
+2023-03-03 20:41:48,344 INFO [train.py:968] (0/2) Epoch 7, batch 28900, giga_loss[loss=0.3335, simple_loss=0.3888, pruned_loss=0.1392, over 28581.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3905, pruned_loss=0.1412, over 5653873.72 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3848, pruned_loss=0.1354, over 5688830.88 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3905, pruned_loss=0.1412, over 5660858.42 frames. ], batch size: 336, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:42:29,062 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:42:34,767 INFO [train.py:968] (0/2) Epoch 7, batch 28950, libri_loss[loss=0.3631, simple_loss=0.4154, pruned_loss=0.1554, over 27750.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3901, pruned_loss=0.1411, over 5659719.77 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3851, pruned_loss=0.1358, over 5692552.05 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3898, pruned_loss=0.1409, over 5661373.40 frames. ], batch size: 116, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:42:56,867 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-03 20:43:15,847 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.122e+03 1.517e+03 2.019e+03 3.007e+03 6.470e+03, threshold=4.038e+03, percent-clipped=5.0
+2023-03-03 20:43:21,768 INFO [train.py:968] (0/2) Epoch 7, batch 29000, libri_loss[loss=0.3609, simple_loss=0.4127, pruned_loss=0.1546, over 29253.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3903, pruned_loss=0.1402, over 5671776.17 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.385, pruned_loss=0.1358, over 5696663.73 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3902, pruned_loss=0.1402, over 5668867.25 frames. ], batch size: 94, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:43:25,861 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:43:30,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:43:40,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:43:45,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:43:57,168 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:44:00,349 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:44:09,050 INFO [train.py:968] (0/2) Epoch 7, batch 29050, libri_loss[loss=0.3729, simple_loss=0.4199, pruned_loss=0.1629, over 29733.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3913, pruned_loss=0.1411, over 5672791.52 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3853, pruned_loss=0.1359, over 5702201.25 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3911, pruned_loss=0.1411, over 5664536.84 frames. ], batch size: 87, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:44:49,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.290e+02 1.647e+03 2.102e+03 2.784e+03 1.816e+04, threshold=4.203e+03, percent-clipped=11.0
+2023-03-03 20:44:52,654 INFO [train.py:968] (0/2) Epoch 7, batch 29100, giga_loss[loss=0.3154, simple_loss=0.3788, pruned_loss=0.126, over 28790.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3918, pruned_loss=0.1415, over 5663358.08 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3853, pruned_loss=0.136, over 5687685.75 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.392, pruned_loss=0.1415, over 5668381.30 frames. ], batch size: 99, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:44:57,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2938, 1.3545, 1.2121, 1.5284], device='cuda:0'), covar=tensor([0.0751, 0.0319, 0.0328, 0.0778], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0117, 0.0121, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0068, 0.0049, 0.0044, 0.0075], device='cuda:0')
+2023-03-03 20:45:14,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0585, 1.1655, 4.1496, 3.3522], device='cuda:0'), covar=tensor([0.1796, 0.2560, 0.0397, 0.0621], device='cuda:0'), in_proj_covar=tensor([0.0617, 0.0574, 0.0816, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 20:45:29,663 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-03 20:45:31,860 INFO [train.py:968] (0/2) Epoch 7, batch 29150, libri_loss[loss=0.3071, simple_loss=0.3515, pruned_loss=0.1314, over 29667.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3929, pruned_loss=0.1424, over 5657871.66 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.385, pruned_loss=0.1358, over 5683596.73 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3937, pruned_loss=0.143, over 5664406.82 frames. ], batch size: 69, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:45:41,810 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:45:43,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:45:46,554 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:45:48,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:46:03,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3230, 1.5228, 1.2073, 1.5072], device='cuda:0'), covar=tensor([0.2125, 0.2049, 0.2188, 0.1804], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.0907, 0.1055, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 20:46:09,860 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:46:12,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.670e+03 2.240e+03 2.856e+03 5.490e+03, threshold=4.479e+03, percent-clipped=6.0
+2023-03-03 20:46:14,382 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:46:16,033 INFO [train.py:968] (0/2) Epoch 7, batch 29200, giga_loss[loss=0.4006, simple_loss=0.4112, pruned_loss=0.195, over 23657.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3928, pruned_loss=0.1426, over 5659225.43 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3847, pruned_loss=0.1355, over 5688038.58 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3939, pruned_loss=0.1434, over 5659952.61 frames. ], batch size: 705, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:46:25,699 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 20:46:32,996 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:46:55,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7980, 1.6512, 1.2096, 1.4447], device='cuda:0'), covar=tensor([0.0623, 0.0633, 0.0910, 0.1078], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0448, 0.0498, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 20:47:01,787 INFO [train.py:968] (0/2) Epoch 7, batch 29250, giga_loss[loss=0.2875, simple_loss=0.366, pruned_loss=0.1044, over 28818.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3924, pruned_loss=0.1412, over 5661533.03 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3845, pruned_loss=0.1353, over 5695244.04 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3936, pruned_loss=0.1422, over 5654919.77 frames. ], batch size: 174, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:47:51,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.618e+02 1.358e+03 1.790e+03 2.298e+03 5.152e+03, threshold=3.581e+03, percent-clipped=3.0
+2023-03-03 20:47:56,167 INFO [train.py:968] (0/2) Epoch 7, batch 29300, giga_loss[loss=0.2911, simple_loss=0.3644, pruned_loss=0.1089, over 28519.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3924, pruned_loss=0.1406, over 5655295.16 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3845, pruned_loss=0.1352, over 5698637.25 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3935, pruned_loss=0.1416, over 5646673.68 frames. ], batch size: 65, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:48:15,774 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-03 20:48:17,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:48:44,289 INFO [train.py:968] (0/2) Epoch 7, batch 29350, giga_loss[loss=0.366, simple_loss=0.4067, pruned_loss=0.1626, over 29000.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3914, pruned_loss=0.1395, over 5657075.11 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3845, pruned_loss=0.1352, over 5701869.80 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3923, pruned_loss=0.1404, over 5646628.56 frames. ], batch size: 106, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:49:03,752 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5246, 2.1622, 2.2605, 2.0354], device='cuda:0'), covar=tensor([0.1081, 0.1984, 0.1495, 0.1692], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0728, 0.0648, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 20:49:09,004 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:49:27,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.406e+02 1.507e+03 2.055e+03 3.096e+03 7.240e+03, threshold=4.111e+03, percent-clipped=20.0
+2023-03-03 20:49:30,933 INFO [train.py:968] (0/2) Epoch 7, batch 29400, giga_loss[loss=0.2976, simple_loss=0.3625, pruned_loss=0.1164, over 28686.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3906, pruned_loss=0.1393, over 5664497.06 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3846, pruned_loss=0.1354, over 5706065.62 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3914, pruned_loss=0.1399, over 5651782.49 frames. ], batch size: 92, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:49:45,402 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:50:19,602 INFO [train.py:968] (0/2) Epoch 7, batch 29450, giga_loss[loss=0.362, simple_loss=0.4073, pruned_loss=0.1583, over 28940.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3897, pruned_loss=0.1387, over 5660809.18 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3847, pruned_loss=0.1355, over 5707743.71 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3902, pruned_loss=0.1391, over 5648793.70 frames. ], batch size: 213, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:50:29,554 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:50:32,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:51:04,466 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:51:05,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.889e+02 1.484e+03 1.825e+03 2.374e+03 6.909e+03, threshold=3.649e+03, percent-clipped=5.0
+2023-03-03 20:51:11,622 INFO [train.py:968] (0/2) Epoch 7, batch 29500, giga_loss[loss=0.343, simple_loss=0.3989, pruned_loss=0.1435, over 28818.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3905, pruned_loss=0.1394, over 5661979.05 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3845, pruned_loss=0.1354, over 5710859.71 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3912, pruned_loss=0.1399, over 5649096.59 frames. ], batch size: 186, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:51:26,569 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-03 20:51:39,785 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 20:51:58,410 INFO [train.py:968] (0/2) Epoch 7, batch 29550, libri_loss[loss=0.3153, simple_loss=0.3791, pruned_loss=0.1258, over 29527.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3896, pruned_loss=0.1393, over 5664219.82 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3843, pruned_loss=0.1352, over 5715388.58 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3906, pruned_loss=0.1399, over 5648141.18 frames. ], batch size: 89, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:52:05,648 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:52:08,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:52:33,601 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:52:39,280 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.655e+03 2.027e+03 2.464e+03 6.375e+03, threshold=4.054e+03, percent-clipped=9.0
+2023-03-03 20:52:39,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:52:43,633 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 20:52:45,283 INFO [train.py:968] (0/2) Epoch 7, batch 29600, giga_loss[loss=0.3299, simple_loss=0.3939, pruned_loss=0.133, over 28731.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3892, pruned_loss=0.1394, over 5675584.90 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.384, pruned_loss=0.135, over 5717619.44 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3903, pruned_loss=0.1402, over 5660235.59 frames. ], batch size: 92, lr: 4.56e-03, grad_scale: 8.0
+2023-03-03 20:53:08,247 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:53:33,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2542, 1.2833, 1.1393, 1.4890], device='cuda:0'), covar=tensor([0.0693, 0.0393, 0.0325, 0.0722], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0118, 0.0121, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0068, 0.0049, 0.0044, 0.0075], device='cuda:0')
+2023-03-03 20:53:34,032 INFO [train.py:968] (0/2) Epoch 7, batch 29650, giga_loss[loss=0.3932, simple_loss=0.4315, pruned_loss=0.1775, over 27632.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3916, pruned_loss=0.1415, over 5669565.68 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3842, pruned_loss=0.135, over 5718823.55 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3923, pruned_loss=0.1421, over 5655936.06 frames. ], batch size: 472, lr: 4.56e-03, grad_scale: 8.0
+2023-03-03 20:54:19,319 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.002e+02 1.407e+03 1.820e+03 2.308e+03 6.101e+03, threshold=3.640e+03, percent-clipped=4.0
+2023-03-03 20:54:24,251 INFO [train.py:968] (0/2) Epoch 7, batch 29700, giga_loss[loss=0.3323, simple_loss=0.3873, pruned_loss=0.1386, over 28227.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3928, pruned_loss=0.1427, over 5647198.38 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3844, pruned_loss=0.1351, over 5708844.63 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3932, pruned_loss=0.1431, over 5644817.12 frames. ], batch size: 77, lr: 4.56e-03, grad_scale: 8.0
+2023-03-03 20:54:50,876 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-03 20:54:53,322 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 20:55:00,840 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:55:02,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:55:10,365 INFO [train.py:968] (0/2) Epoch 7, batch 29750, giga_loss[loss=0.3251, simple_loss=0.3941, pruned_loss=0.1281, over 28507.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3927, pruned_loss=0.1427, over 5655266.75 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3844, pruned_loss=0.1353, over 5715966.55 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3934, pruned_loss=0.1432, over 5644802.04 frames. ], batch size: 71, lr: 4.56e-03, grad_scale: 8.0
+2023-03-03 20:55:10,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:55:29,500 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:55:32,830 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-03 20:55:39,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4115, 1.4867, 1.4965, 1.4294], device='cuda:0'), covar=tensor([0.0872, 0.1270, 0.1317, 0.1152], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0731, 0.0654, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 20:55:54,259 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:55:54,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.270e+02 1.579e+03 2.051e+03 2.997e+03 6.746e+03, threshold=4.101e+03, percent-clipped=13.0
+2023-03-03 20:55:59,325 INFO [train.py:968] (0/2) Epoch 7, batch 29800, giga_loss[loss=0.3277, simple_loss=0.391, pruned_loss=0.1322, over 28711.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3922, pruned_loss=0.1417, over 5657651.33 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3845, pruned_loss=0.1354, over 5717072.54 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3929, pruned_loss=0.1422, over 5646795.74 frames. ], batch size: 262, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:56:47,323 INFO [train.py:968] (0/2) Epoch 7, batch 29850, giga_loss[loss=0.2961, simple_loss=0.3708, pruned_loss=0.1107, over 28708.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3915, pruned_loss=0.1403, over 5656977.71 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3843, pruned_loss=0.1352, over 5716959.77 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3923, pruned_loss=0.141, over 5647394.84 frames. ], batch size: 284, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:57:31,048 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:57:33,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:57:33,421 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.636e+02 1.531e+03 2.018e+03 2.659e+03 8.011e+03, threshold=4.037e+03, percent-clipped=6.0
+2023-03-03 20:57:35,322 INFO [train.py:968] (0/2) Epoch 7, batch 29900, giga_loss[loss=0.4103, simple_loss=0.4395, pruned_loss=0.1906, over 28233.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3913, pruned_loss=0.1404, over 5653571.65 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3853, pruned_loss=0.1361, over 5710294.86 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3912, pruned_loss=0.1402, over 5649888.35 frames. ], batch size: 368, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 20:57:47,895 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7446, 2.7253, 1.7694, 0.9142], device='cuda:0'), covar=tensor([0.4240, 0.1953, 0.2366, 0.3788], device='cuda:0'), in_proj_covar=tensor([0.1481, 0.1400, 0.1451, 0.1221], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 20:57:58,197 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:58:00,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9326, 3.7609, 3.5404, 1.8042], device='cuda:0'), covar=tensor([0.0573, 0.0697, 0.0766, 0.2056], device='cuda:0'), in_proj_covar=tensor([0.0975, 0.0927, 0.0828, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-03 20:58:21,088 INFO [train.py:968] (0/2) Epoch 7, batch 29950, giga_loss[loss=0.3373, simple_loss=0.3713, pruned_loss=0.1517, over 23684.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3909, pruned_loss=0.1401, over 5662516.10 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3855, pruned_loss=0.1362, over 5707319.33 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3908, pruned_loss=0.14, over 5660124.75 frames. ], batch size: 705, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 20:58:33,848 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-03 20:58:47,517 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3831, 2.0877, 1.5928, 0.4847], device='cuda:0'), covar=tensor([0.2959, 0.1513, 0.2126, 0.3530], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1417, 0.1462, 0.1229], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 20:58:49,410 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-03 20:59:04,271 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.262e+02 1.549e+03 1.969e+03 2.863e+03 6.520e+03, threshold=3.937e+03, percent-clipped=8.0
+2023-03-03 20:59:07,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:59:08,156 INFO [train.py:968] (0/2) Epoch 7, batch 30000, giga_loss[loss=0.3694, simple_loss=0.4112, pruned_loss=0.1638, over 28594.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3862, pruned_loss=0.1367, over 5665961.76 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3846, pruned_loss=0.1356, over 5710965.56 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.387, pruned_loss=0.1372, over 5659742.63 frames. ], batch size: 336, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 20:59:08,162 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 20:59:16,778 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2258, simple_loss=0.3317, pruned_loss=0.05999, over 944034.00 frames. 
+2023-03-03 20:59:16,779 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-03 20:59:17,740 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:59:22,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-03 20:59:23,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3066, 3.1251, 2.9973, 1.4225], device='cuda:0'), covar=tensor([0.0832, 0.0982, 0.0931, 0.2092], device='cuda:0'), in_proj_covar=tensor([0.0976, 0.0927, 0.0828, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-03 21:00:04,885 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-03 21:00:06,435 INFO [train.py:968] (0/2) Epoch 7, batch 30050, giga_loss[loss=0.3156, simple_loss=0.3653, pruned_loss=0.133, over 28810.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3819, pruned_loss=0.1346, over 5658160.02 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3843, pruned_loss=0.1353, over 5715767.03 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3829, pruned_loss=0.1353, over 5647792.58 frames. ], batch size: 99, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:00:26,234 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 21:00:48,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.517e+02 1.818e+03 2.490e+03 4.051e+03 1.031e+04, threshold=4.981e+03, percent-clipped=27.0
+2023-03-03 21:00:50,966 INFO [train.py:968] (0/2) Epoch 7, batch 30100, libri_loss[loss=0.3182, simple_loss=0.3814, pruned_loss=0.1275, over 29647.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3816, pruned_loss=0.1352, over 5660424.07 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3848, pruned_loss=0.1357, over 5712376.86 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3817, pruned_loss=0.1354, over 5653392.65 frames. ], batch size: 88, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:01:31,242 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:01:33,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:01:34,343 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:01:38,064 INFO [train.py:968] (0/2) Epoch 7, batch 30150, giga_loss[loss=0.3797, simple_loss=0.4246, pruned_loss=0.1674, over 27918.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3803, pruned_loss=0.135, over 5656085.86 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3844, pruned_loss=0.1353, over 5713630.17 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3806, pruned_loss=0.1354, over 5648416.59 frames. ], batch size: 412, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:01:44,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4717, 1.8652, 1.7863, 1.3955], device='cuda:0'), covar=tensor([0.1531, 0.2007, 0.1207, 0.1345], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0717, 0.0814, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 21:01:59,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:02:03,826 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:02:11,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3875, 2.0664, 1.6156, 0.5108], device='cuda:0'), covar=tensor([0.2594, 0.1539, 0.2381, 0.3016], device='cuda:0'), in_proj_covar=tensor([0.1457, 0.1379, 0.1429, 0.1203], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 21:02:11,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4440, 1.7432, 1.3913, 1.6758], device='cuda:0'), covar=tensor([0.2115, 0.1992, 0.2164, 0.1767], device='cuda:0'), in_proj_covar=tensor([0.1199, 0.0910, 0.1055, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 21:02:25,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.343e+02 1.633e+03 2.244e+03 3.089e+03 1.313e+04, threshold=4.489e+03, percent-clipped=10.0
+2023-03-03 21:02:27,233 INFO [train.py:968] (0/2) Epoch 7, batch 30200, giga_loss[loss=0.3717, simple_loss=0.4252, pruned_loss=0.1591, over 27976.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3801, pruned_loss=0.134, over 5650276.31 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3842, pruned_loss=0.1352, over 5714946.12 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3805, pruned_loss=0.1345, over 5642259.19 frames. ], batch size: 412, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:03:06,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3637, 1.8405, 1.2966, 0.5922], device='cuda:0'), covar=tensor([0.2528, 0.1518, 0.2156, 0.3107], device='cuda:0'), in_proj_covar=tensor([0.1462, 0.1385, 0.1432, 0.1208], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 21:03:17,144 INFO [train.py:968] (0/2) Epoch 7, batch 30250, giga_loss[loss=0.2944, simple_loss=0.3635, pruned_loss=0.1126, over 28591.00 frames. ], tot_loss[loss=0.321, simple_loss=0.379, pruned_loss=0.1315, over 5639057.37 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.385, pruned_loss=0.136, over 5709011.84 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3785, pruned_loss=0.1311, over 5637015.79 frames. ], batch size: 307, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:04:06,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.943e+02 1.510e+03 1.930e+03 2.823e+03 6.496e+03, threshold=3.860e+03, percent-clipped=4.0
+2023-03-03 21:04:08,573 INFO [train.py:968] (0/2) Epoch 7, batch 30300, giga_loss[loss=0.2404, simple_loss=0.3267, pruned_loss=0.07704, over 29057.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.377, pruned_loss=0.1287, over 5642491.96 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3848, pruned_loss=0.1361, over 5705420.46 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3766, pruned_loss=0.1281, over 5641685.96 frames. ], batch size: 155, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:04:11,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 21:04:23,205 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:04:25,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:04:25,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:04:53,336 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:04:59,964 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-304000.pt
+2023-03-03 21:05:00,294 INFO [train.py:968] (0/2) Epoch 7, batch 30350, giga_loss[loss=0.3054, simple_loss=0.3728, pruned_loss=0.1191, over 28773.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3738, pruned_loss=0.1251, over 5624875.31 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3849, pruned_loss=0.1363, over 5685671.71 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3731, pruned_loss=0.1241, over 5640865.76 frames. ], batch size: 262, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:05:02,232 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-03 21:05:24,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:05:47,806 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.902e+02 1.382e+03 1.958e+03 2.705e+03 1.051e+04, threshold=3.916e+03, percent-clipped=8.0
+2023-03-03 21:05:49,235 INFO [train.py:968] (0/2) Epoch 7, batch 30400, giga_loss[loss=0.2686, simple_loss=0.3512, pruned_loss=0.09303, over 28751.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3696, pruned_loss=0.1211, over 5619893.24 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.385, pruned_loss=0.1365, over 5673738.23 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3688, pruned_loss=0.12, over 5643121.96 frames. ], batch size: 119, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:05:59,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3371, 1.5501, 1.5183, 1.5247], device='cuda:0'), covar=tensor([0.1217, 0.1439, 0.1487, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0717, 0.0639, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 21:06:02,891 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-03 21:06:40,579 INFO [train.py:968] (0/2) Epoch 7, batch 30450, giga_loss[loss=0.3626, simple_loss=0.4162, pruned_loss=0.1546, over 28542.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3674, pruned_loss=0.1168, over 5643346.76 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3843, pruned_loss=0.1363, over 5674384.68 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3671, pruned_loss=0.1158, over 5660319.99 frames. ], batch size: 336, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:06:48,456 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-03 21:07:32,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.992e+02 1.327e+03 1.705e+03 2.720e+03 5.684e+03, threshold=3.411e+03, percent-clipped=6.0
+2023-03-03 21:07:34,386 INFO [train.py:968] (0/2) Epoch 7, batch 30500, giga_loss[loss=0.3032, simple_loss=0.3657, pruned_loss=0.1204, over 27621.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3664, pruned_loss=0.1153, over 5640404.71 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3839, pruned_loss=0.1363, over 5666499.15 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3663, pruned_loss=0.1143, over 5660618.18 frames. ], batch size: 472, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:07:55,193 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:07:55,277 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-03 21:07:58,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:07:59,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:08:09,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6785, 1.9374, 1.1579, 1.6257], device='cuda:0'), covar=tensor([0.0755, 0.0562, 0.1127, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0439, 0.0492, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 21:08:25,224 INFO [train.py:968] (0/2) Epoch 7, batch 30550, giga_loss[loss=0.2731, simple_loss=0.3466, pruned_loss=0.09974, over 28819.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3643, pruned_loss=0.1135, over 5647334.72 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3833, pruned_loss=0.1361, over 5667315.26 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3643, pruned_loss=0.1125, over 5662212.76 frames. ], batch size: 199, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:08:27,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:08:29,808 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:09:15,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.588e+02 1.364e+03 1.737e+03 2.585e+03 9.569e+03, threshold=3.475e+03, percent-clipped=13.0
+2023-03-03 21:09:15,971 INFO [train.py:968] (0/2) Epoch 7, batch 30600, giga_loss[loss=0.2788, simple_loss=0.3503, pruned_loss=0.1036, over 29031.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3601, pruned_loss=0.1103, over 5652598.73 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.383, pruned_loss=0.136, over 5670577.52 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3601, pruned_loss=0.1092, over 5661220.64 frames. ], batch size: 128, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:09:57,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:10:05,847 INFO [train.py:968] (0/2) Epoch 7, batch 30650, giga_loss[loss=0.2587, simple_loss=0.3427, pruned_loss=0.0873, over 29033.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3579, pruned_loss=0.1092, over 5650183.34 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3823, pruned_loss=0.1355, over 5674733.68 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.358, pruned_loss=0.1081, over 5653242.80 frames. ], batch size: 136, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:10:19,297 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:10:21,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:10:35,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3443, 1.7201, 1.3853, 1.5793], device='cuda:0'), covar=tensor([0.0760, 0.0290, 0.0332, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0121, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0049, 0.0045, 0.0075], device='cuda:0')
+2023-03-03 21:10:45,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:10:49,094 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:10:50,994 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.117e+02 1.376e+03 1.900e+03 2.677e+03 7.807e+03, threshold=3.799e+03, percent-clipped=12.0
+2023-03-03 21:10:51,656 INFO [train.py:968] (0/2) Epoch 7, batch 30700, libri_loss[loss=0.3133, simple_loss=0.3668, pruned_loss=0.1299, over 29654.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3595, pruned_loss=0.1102, over 5653593.71 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3826, pruned_loss=0.136, over 5671618.73 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3588, pruned_loss=0.1082, over 5657298.26 frames. ], batch size: 73, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:11:10,226 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304367.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:11:14,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6305, 1.5627, 1.2831, 1.2801], device='cuda:0'), covar=tensor([0.0634, 0.0457, 0.0849, 0.0792], device='cuda:0'), in_proj_covar=tensor([0.0334, 0.0436, 0.0486, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 21:11:33,154 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 21:11:37,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2369, 4.0827, 2.3084, 2.2714], device='cuda:0'), covar=tensor([0.0591, 0.0316, 0.0600, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0494, 0.0322, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 21:11:41,296 INFO [train.py:968] (0/2) Epoch 7, batch 30750, giga_loss[loss=0.2815, simple_loss=0.3521, pruned_loss=0.1054, over 28955.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.358, pruned_loss=0.1091, over 5647325.14 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3824, pruned_loss=0.136, over 5665839.28 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3573, pruned_loss=0.1071, over 5655602.78 frames. ], batch size: 213, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:11:51,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7920, 1.7157, 1.4090, 2.0632], device='cuda:0'), covar=tensor([0.2200, 0.2224, 0.2329, 0.2075], device='cuda:0'), in_proj_covar=tensor([0.1196, 0.0903, 0.1061, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 21:12:27,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9239, 1.0367, 0.9741, 0.9171], device='cuda:0'), covar=tensor([0.0913, 0.0954, 0.0609, 0.0772], device='cuda:0'), in_proj_covar=tensor([0.1529, 0.1360, 0.1323, 0.1427], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 21:12:30,695 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.361e+02 1.264e+03 1.820e+03 2.930e+03 1.014e+04, threshold=3.640e+03, percent-clipped=12.0
+2023-03-03 21:12:31,357 INFO [train.py:968] (0/2) Epoch 7, batch 30800, giga_loss[loss=0.2456, simple_loss=0.326, pruned_loss=0.08258, over 28023.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3547, pruned_loss=0.1064, over 5654312.53 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3815, pruned_loss=0.1356, over 5670720.35 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3543, pruned_loss=0.1046, over 5656158.13 frames. ], batch size: 412, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:13:07,381 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:13:09,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:13:21,499 INFO [train.py:968] (0/2) Epoch 7, batch 30850, giga_loss[loss=0.2584, simple_loss=0.3306, pruned_loss=0.09307, over 28618.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.351, pruned_loss=0.1038, over 5669532.26 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3808, pruned_loss=0.1351, over 5675554.20 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3506, pruned_loss=0.1021, over 5666508.26 frames. ], batch size: 92, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:13:38,174 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:14:07,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.953e+02 1.234e+03 1.670e+03 2.196e+03 4.119e+03, threshold=3.340e+03, percent-clipped=3.0
+2023-03-03 21:14:08,302 INFO [train.py:968] (0/2) Epoch 7, batch 30900, giga_loss[loss=0.2699, simple_loss=0.3431, pruned_loss=0.09831, over 28570.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3495, pruned_loss=0.1037, over 5663139.04 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3806, pruned_loss=0.135, over 5672974.45 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3485, pruned_loss=0.1015, over 5663412.01 frames. ], batch size: 336, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:14:37,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:14:46,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2814, 1.7464, 1.3292, 0.6422], device='cuda:0'), covar=tensor([0.2576, 0.1445, 0.1677, 0.2972], device='cuda:0'), in_proj_covar=tensor([0.1442, 0.1363, 0.1422, 0.1193], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 21:14:59,400 INFO [train.py:968] (0/2) Epoch 7, batch 30950, giga_loss[loss=0.2576, simple_loss=0.3124, pruned_loss=0.1014, over 24145.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3486, pruned_loss=0.1036, over 5656563.41 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3805, pruned_loss=0.1349, over 5677457.54 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3473, pruned_loss=0.1013, over 5652455.26 frames. ], batch size: 705, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:15:30,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0459, 3.3151, 2.3978, 0.9823], device='cuda:0'), covar=tensor([0.3943, 0.1539, 0.1991, 0.3626], device='cuda:0'), in_proj_covar=tensor([0.1442, 0.1363, 0.1421, 0.1190], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 21:15:50,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.263e+02 1.355e+03 1.929e+03 2.765e+03 5.127e+03, threshold=3.858e+03, percent-clipped=15.0
+2023-03-03 21:15:52,069 INFO [train.py:968] (0/2) Epoch 7, batch 31000, giga_loss[loss=0.3086, simple_loss=0.3729, pruned_loss=0.1221, over 28855.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3488, pruned_loss=0.104, over 5643694.20 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3795, pruned_loss=0.1344, over 5673210.21 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3478, pruned_loss=0.1017, over 5644254.31 frames. ], batch size: 99, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:16:06,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:16:43,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3957, 2.0268, 1.7267, 1.3589], device='cuda:0'), covar=tensor([0.1723, 0.2099, 0.1453, 0.1618], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0703, 0.0812, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 21:16:44,989 INFO [train.py:968] (0/2) Epoch 7, batch 31050, libri_loss[loss=0.2537, simple_loss=0.3082, pruned_loss=0.09956, over 29485.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3524, pruned_loss=0.1058, over 5633984.37 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3793, pruned_loss=0.1345, over 5667995.75 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3507, pruned_loss=0.1029, over 5638240.35 frames. ], batch size: 70, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:17:12,118 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:17:15,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:17:37,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304742.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:17:46,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.842e+02 1.282e+03 1.734e+03 3.039e+03 8.521e+03, threshold=3.468e+03, percent-clipped=16.0
+2023-03-03 21:17:47,123 INFO [train.py:968] (0/2) Epoch 7, batch 31100, giga_loss[loss=0.2793, simple_loss=0.3553, pruned_loss=0.1016, over 28811.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3518, pruned_loss=0.1049, over 5630562.25 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.379, pruned_loss=0.1344, over 5671851.51 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3504, pruned_loss=0.1022, over 5629979.95 frames. ], batch size: 119, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:17:51,137 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:17:53,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4054, 1.6125, 1.5006, 1.4842], device='cuda:0'), covar=tensor([0.1358, 0.1865, 0.1670, 0.1560], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0703, 0.0629, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 21:18:50,163 INFO [train.py:968] (0/2) Epoch 7, batch 31150, giga_loss[loss=0.279, simple_loss=0.3526, pruned_loss=0.1027, over 28715.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3524, pruned_loss=0.1056, over 5636493.59 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3789, pruned_loss=0.1345, over 5673623.90 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3507, pruned_loss=0.1027, over 5633542.03 frames. ], batch size: 262, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:19:01,476 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:19:05,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:19:42,231 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:19:54,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.921e+02 1.204e+03 1.627e+03 2.332e+03 1.213e+04, threshold=3.254e+03, percent-clipped=11.0
+2023-03-03 21:19:55,365 INFO [train.py:968] (0/2) Epoch 7, batch 31200, giga_loss[loss=0.2951, simple_loss=0.3599, pruned_loss=0.1152, over 28425.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3494, pruned_loss=0.1026, over 5641441.90 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3788, pruned_loss=0.1344, over 5674801.78 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3481, pruned_loss=0.1003, over 5637986.94 frames. ], batch size: 369, lr: 4.55e-03, grad_scale: 8.0
+2023-03-03 21:20:43,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304885.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:20:46,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304888.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:20:59,388 INFO [train.py:968] (0/2) Epoch 7, batch 31250, giga_loss[loss=0.2365, simple_loss=0.3217, pruned_loss=0.07562, over 29029.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3471, pruned_loss=0.09971, over 5632440.60 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3785, pruned_loss=0.1343, over 5670505.08 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3453, pruned_loss=0.0969, over 5632136.67 frames. ], batch size: 213, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:21:14,103 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304911.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:21:20,232 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304917.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:22:04,482 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.784e+02 1.378e+03 1.838e+03 2.355e+03 7.365e+03, threshold=3.676e+03, percent-clipped=10.0
+2023-03-03 21:22:04,495 INFO [train.py:968] (0/2) Epoch 7, batch 31300, giga_loss[loss=0.2375, simple_loss=0.3133, pruned_loss=0.08086, over 28435.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3448, pruned_loss=0.09927, over 5646586.11 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3779, pruned_loss=0.1339, over 5672988.42 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3437, pruned_loss=0.09705, over 5643772.03 frames. ], batch size: 336, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:23:02,707 INFO [train.py:968] (0/2) Epoch 7, batch 31350, giga_loss[loss=0.2821, simple_loss=0.3563, pruned_loss=0.1039, over 28924.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3447, pruned_loss=0.1001, over 5649551.36 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3775, pruned_loss=0.1338, over 5667700.98 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3434, pruned_loss=0.09772, over 5652000.81 frames. ], batch size: 227, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:23:14,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:23:22,337 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305015.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:23:47,345 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:23:59,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.215e+02 1.345e+03 1.773e+03 2.353e+03 7.092e+03, threshold=3.547e+03, percent-clipped=7.0
+2023-03-03 21:23:59,449 INFO [train.py:968] (0/2) Epoch 7, batch 31400, giga_loss[loss=0.2957, simple_loss=0.3673, pruned_loss=0.112, over 28064.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3441, pruned_loss=0.09977, over 5664488.86 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3771, pruned_loss=0.1336, over 5672178.16 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3427, pruned_loss=0.09723, over 5662107.14 frames. ], batch size: 412, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:24:14,148 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:24:24,391 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:24:59,026 INFO [train.py:968] (0/2) Epoch 7, batch 31450, giga_loss[loss=0.2984, simple_loss=0.3719, pruned_loss=0.1125, over 28042.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3436, pruned_loss=0.09834, over 5666196.35 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3761, pruned_loss=0.1331, over 5676046.84 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3428, pruned_loss=0.09619, over 5660681.19 frames. ], batch size: 412, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:26:00,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.150e+02 1.333e+03 1.872e+03 2.616e+03 9.789e+03, threshold=3.744e+03, percent-clipped=11.0
+2023-03-03 21:26:00,929 INFO [train.py:968] (0/2) Epoch 7, batch 31500, giga_loss[loss=0.259, simple_loss=0.338, pruned_loss=0.09002, over 28847.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3454, pruned_loss=0.09963, over 5652335.72 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3753, pruned_loss=0.1329, over 5673148.91 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3444, pruned_loss=0.09704, over 5650013.84 frames. ], batch size: 227, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:26:45,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:27:02,286 INFO [train.py:968] (0/2) Epoch 7, batch 31550, giga_loss[loss=0.2952, simple_loss=0.3713, pruned_loss=0.1095, over 28651.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.343, pruned_loss=0.09796, over 5663727.35 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3754, pruned_loss=0.1329, over 5667938.12 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3414, pruned_loss=0.09505, over 5665746.99 frames. ], batch size: 262, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:28:11,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.401e+02 1.287e+03 1.600e+03 2.227e+03 7.519e+03, threshold=3.200e+03, percent-clipped=9.0
+2023-03-03 21:28:11,765 INFO [train.py:968] (0/2) Epoch 7, batch 31600, giga_loss[loss=0.2397, simple_loss=0.3256, pruned_loss=0.07694, over 28936.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3447, pruned_loss=0.09944, over 5661695.77 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.375, pruned_loss=0.1328, over 5670720.26 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3433, pruned_loss=0.09677, over 5660948.58 frames. ], batch size: 136, lr: 4.54e-03, grad_scale: 8.0
+2023-03-03 21:28:49,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305286.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:29:06,300 INFO [train.py:968] (0/2) Epoch 7, batch 31650, giga_loss[loss=0.2735, simple_loss=0.3646, pruned_loss=0.09117, over 28987.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3487, pruned_loss=0.1005, over 5675034.48 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3742, pruned_loss=0.1323, over 5679612.29 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3469, pruned_loss=0.09721, over 5666281.99 frames. ], batch size: 155, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:29:23,008 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-03 21:30:11,251 INFO [train.py:968] (0/2) Epoch 7, batch 31700, giga_loss[loss=0.2559, simple_loss=0.3488, pruned_loss=0.08147, over 28981.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3512, pruned_loss=0.09988, over 5664753.13 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3739, pruned_loss=0.1323, over 5675580.86 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3496, pruned_loss=0.09663, over 5660724.00 frames. ], batch size: 155, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:30:12,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.948e+02 1.477e+03 1.851e+03 2.671e+03 1.023e+04, threshold=3.702e+03, percent-clipped=20.0
+2023-03-03 21:30:52,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:30:57,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305390.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:31:03,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-03 21:31:09,096 INFO [train.py:968] (0/2) Epoch 7, batch 31750, giga_loss[loss=0.2911, simple_loss=0.3739, pruned_loss=0.1042, over 28874.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3514, pruned_loss=0.09873, over 5663641.21 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3733, pruned_loss=0.1318, over 5678450.90 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3499, pruned_loss=0.09553, over 5657629.18 frames. ], batch size: 174, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:31:23,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:31:23,242 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.35 vs. limit=5.0
+2023-03-03 21:31:38,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 21:31:42,253 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305429.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:31:47,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305432.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:31:53,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:32:00,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:32:05,691 INFO [train.py:968] (0/2) Epoch 7, batch 31800, giga_loss[loss=0.262, simple_loss=0.3551, pruned_loss=0.08444, over 28972.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3491, pruned_loss=0.09602, over 5677767.85 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3725, pruned_loss=0.1311, over 5684900.14 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3479, pruned_loss=0.09306, over 5666797.89 frames. ], batch size: 145, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:32:08,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.748e+02 1.281e+03 1.706e+03 2.477e+03 6.173e+03, threshold=3.411e+03, percent-clipped=10.0
+2023-03-03 21:32:22,109 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305461.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:32:31,666 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-03 21:33:11,190 INFO [train.py:968] (0/2) Epoch 7, batch 31850, giga_loss[loss=0.2544, simple_loss=0.3348, pruned_loss=0.08704, over 28944.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3504, pruned_loss=0.09764, over 5685163.16 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3724, pruned_loss=0.1311, over 5688223.06 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09491, over 5673493.36 frames. ], batch size: 136, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:33:44,649 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:33:47,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:33:52,140 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305533.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:33:57,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305536.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:34:17,061 INFO [train.py:968] (0/2) Epoch 7, batch 31900, giga_loss[loss=0.2922, simple_loss=0.3649, pruned_loss=0.1097, over 29021.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3497, pruned_loss=0.09879, over 5680705.05 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.372, pruned_loss=0.1308, over 5685849.78 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3485, pruned_loss=0.09585, over 5672923.20 frames. ], batch size: 155, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:34:19,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.760e+02 1.297e+03 1.753e+03 2.491e+03 6.157e+03, threshold=3.506e+03, percent-clipped=9.0
+2023-03-03 21:34:25,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1600, 2.3355, 1.2446, 1.2571], device='cuda:0'), covar=tensor([0.0863, 0.0394, 0.0824, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0491, 0.0325, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 21:34:26,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:34:33,927 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:34:33,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:34:38,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:34:44,899 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305565.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:35:02,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:05,906 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:13,733 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:13,902 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-03 21:35:20,698 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:23,563 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:29,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8882, 2.7806, 1.8879, 1.6965], device='cuda:0'), covar=tensor([0.1963, 0.0814, 0.0988, 0.1253], device='cuda:0'), in_proj_covar=tensor([0.1530, 0.1356, 0.1314, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 21:35:37,560 INFO [train.py:968] (0/2) Epoch 7, batch 31950, giga_loss[loss=0.2573, simple_loss=0.3401, pruned_loss=0.08722, over 28715.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3512, pruned_loss=0.1008, over 5673721.72 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3725, pruned_loss=0.1313, over 5679704.14 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3497, pruned_loss=0.09763, over 5672158.17 frames. ], batch size: 262, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:35:41,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4399, 1.8567, 1.7933, 1.4040], device='cuda:0'), covar=tensor([0.1524, 0.1880, 0.1182, 0.1348], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0699, 0.0806, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 21:35:53,094 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:54,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1601, 0.8722, 0.9185, 1.3317], device='cuda:0'), covar=tensor([0.0780, 0.0369, 0.0370, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0122, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0049, 0.0045, 0.0076], device='cuda:0')
+2023-03-03 21:36:03,512 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:36:32,331 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:36:48,952 INFO [train.py:968] (0/2) Epoch 7, batch 32000, giga_loss[loss=0.2374, simple_loss=0.3252, pruned_loss=0.07479, over 28864.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3461, pruned_loss=0.09775, over 5674546.46 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3721, pruned_loss=0.1311, over 5682902.55 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3447, pruned_loss=0.09487, over 5670350.68 frames. ], batch size: 164, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:36:51,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.591e+02 1.273e+03 1.652e+03 2.344e+03 1.156e+04, threshold=3.304e+03, percent-clipped=9.0
+2023-03-03 21:37:15,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2819, 3.1223, 2.9500, 1.3619], device='cuda:0'), covar=tensor([0.0825, 0.0909, 0.0932, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0879, 0.0784, 0.0615], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 21:37:50,567 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.94 vs. limit=5.0
+2023-03-03 21:37:54,022 INFO [train.py:968] (0/2) Epoch 7, batch 32050, giga_loss[loss=0.23, simple_loss=0.3096, pruned_loss=0.07518, over 28959.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3443, pruned_loss=0.09648, over 5675110.26 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3721, pruned_loss=0.1311, over 5685016.86 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3428, pruned_loss=0.09376, over 5669730.37 frames. ], batch size: 106, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:38:04,688 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:38:08,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:38:08,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8486, 2.6041, 1.7775, 0.8920], device='cuda:0'), covar=tensor([0.3481, 0.1959, 0.2207, 0.3526], device='cuda:0'), in_proj_covar=tensor([0.1428, 0.1351, 0.1421, 0.1193], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 21:38:15,479 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:38:34,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3702, 1.7589, 1.7419, 1.3586], device='cuda:0'), covar=tensor([0.1553, 0.1977, 0.1199, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0696, 0.0803, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 21:38:40,536 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:38:57,157 INFO [train.py:968] (0/2) Epoch 7, batch 32100, giga_loss[loss=0.286, simple_loss=0.3645, pruned_loss=0.1037, over 28836.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3429, pruned_loss=0.09639, over 5672320.44 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3718, pruned_loss=0.131, over 5676466.63 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3413, pruned_loss=0.09343, over 5674935.89 frames. ], batch size: 243, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:39:00,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.345e+02 1.478e+03 1.803e+03 2.623e+03 6.884e+03, threshold=3.606e+03, percent-clipped=12.0
+2023-03-03 21:39:16,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5572, 3.9102, 1.5859, 1.5407], device='cuda:0'), covar=tensor([0.0851, 0.0249, 0.0850, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0489, 0.0323, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 21:39:59,431 INFO [train.py:968] (0/2) Epoch 7, batch 32150, giga_loss[loss=0.3188, simple_loss=0.384, pruned_loss=0.1268, over 29106.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3471, pruned_loss=0.09844, over 5667718.41 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3717, pruned_loss=0.1309, over 5669929.83 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3455, pruned_loss=0.09557, over 5676020.72 frames. ], batch size: 200, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:40:57,757 INFO [train.py:968] (0/2) Epoch 7, batch 32200, giga_loss[loss=0.2412, simple_loss=0.3212, pruned_loss=0.08058, over 29015.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3466, pruned_loss=0.09939, over 5684099.76 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3707, pruned_loss=0.1303, over 5676182.55 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3454, pruned_loss=0.09661, over 5685464.07 frames. ], batch size: 285, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:41:01,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.229e+02 1.417e+03 1.778e+03 2.491e+03 4.957e+03, threshold=3.556e+03, percent-clipped=4.0
+2023-03-03 21:42:00,990 INFO [train.py:968] (0/2) Epoch 7, batch 32250, giga_loss[loss=0.2803, simple_loss=0.3531, pruned_loss=0.1037, over 27518.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3458, pruned_loss=0.09994, over 5683199.10 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3703, pruned_loss=0.1302, over 5681523.27 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3447, pruned_loss=0.09714, over 5679643.62 frames. ], batch size: 472, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:43:02,974 INFO [train.py:968] (0/2) Epoch 7, batch 32300, giga_loss[loss=0.2879, simple_loss=0.3589, pruned_loss=0.1084, over 28992.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3461, pruned_loss=0.1009, over 5671709.22 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3702, pruned_loss=0.1303, over 5671775.47 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3448, pruned_loss=0.09789, over 5677216.21 frames. ], batch size: 155, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:43:06,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.095e+02 1.291e+03 1.826e+03 2.720e+03 1.003e+04, threshold=3.651e+03, percent-clipped=5.0
+2023-03-03 21:43:49,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3851, 1.5790, 1.3571, 1.4650], device='cuda:0'), covar=tensor([0.0785, 0.0308, 0.0339, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0122, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0049, 0.0045, 0.0075], device='cuda:0')
+2023-03-03 21:44:12,002 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-306000.pt
+2023-03-03 21:44:12,323 INFO [train.py:968] (0/2) Epoch 7, batch 32350, giga_loss[loss=0.2665, simple_loss=0.3494, pruned_loss=0.09183, over 28676.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3472, pruned_loss=0.1008, over 5675123.15 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3698, pruned_loss=0.1302, over 5677456.93 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.346, pruned_loss=0.09782, over 5674310.71 frames. ], batch size: 85, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:44:29,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:44:37,520 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306016.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:45:25,030 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:45:31,653 INFO [train.py:968] (0/2) Epoch 7, batch 32400, giga_loss[loss=0.2471, simple_loss=0.3293, pruned_loss=0.08239, over 28659.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3474, pruned_loss=0.09956, over 5668448.63 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3698, pruned_loss=0.1302, over 5678804.64 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3464, pruned_loss=0.09712, over 5666513.87 frames. ], batch size: 307, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:45:36,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.222e+02 1.309e+03 1.706e+03 2.605e+03 9.589e+03, threshold=3.413e+03, percent-clipped=9.0
+2023-03-03 21:46:00,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3033, 1.6296, 1.2814, 1.3954], device='cuda:0'), covar=tensor([0.2113, 0.2003, 0.2180, 0.1731], device='cuda:0'), in_proj_covar=tensor([0.1189, 0.0888, 0.1056, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 21:46:32,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:46:33,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1106, 1.6551, 1.2112, 0.3537], device='cuda:0'), covar=tensor([0.2239, 0.1330, 0.2422, 0.3043], device='cuda:0'), in_proj_covar=tensor([0.1456, 0.1376, 0.1436, 0.1206], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 21:46:43,265 INFO [train.py:968] (0/2) Epoch 7, batch 32450, giga_loss[loss=0.2304, simple_loss=0.3085, pruned_loss=0.07613, over 28885.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3448, pruned_loss=0.09846, over 5674604.44 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3694, pruned_loss=0.1299, over 5682841.69 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3439, pruned_loss=0.09622, over 5669448.02 frames. ], batch size: 199, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:46:43,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3931, 4.2361, 4.0064, 1.7714], device='cuda:0'), covar=tensor([0.0442, 0.0609, 0.0601, 0.2175], device='cuda:0'), in_proj_covar=tensor([0.0919, 0.0869, 0.0770, 0.0610], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 21:47:31,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3858, 1.6535, 1.3378, 1.5834], device='cuda:0'), covar=tensor([0.0728, 0.0292, 0.0321, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0122, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0049, 0.0045, 0.0076], device='cuda:0')
+2023-03-03 21:47:42,818 INFO [train.py:968] (0/2) Epoch 7, batch 32500, giga_loss[loss=0.2035, simple_loss=0.287, pruned_loss=0.06003, over 28840.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.341, pruned_loss=0.09763, over 5680813.21 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3692, pruned_loss=0.1297, over 5688347.04 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3392, pruned_loss=0.09451, over 5671398.04 frames. ], batch size: 174, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:47:46,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.067e+02 1.350e+03 1.760e+03 2.204e+03 1.443e+04, threshold=3.520e+03, percent-clipped=11.0
+2023-03-03 21:47:46,744 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:47:50,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:48:28,336 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:48:46,986 INFO [train.py:968] (0/2) Epoch 7, batch 32550, giga_loss[loss=0.2863, simple_loss=0.35, pruned_loss=0.1114, over 28725.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3371, pruned_loss=0.09641, over 5666148.36 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3691, pruned_loss=0.1298, over 5681061.03 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.335, pruned_loss=0.09296, over 5665431.21 frames. ], batch size: 307, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:49:29,676 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:49:32,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:49:43,109 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:49:48,917 INFO [train.py:968] (0/2) Epoch 7, batch 32600, giga_loss[loss=0.2703, simple_loss=0.3448, pruned_loss=0.0979, over 28698.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3386, pruned_loss=0.09767, over 5659997.28 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3693, pruned_loss=0.1301, over 5673266.92 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3362, pruned_loss=0.09414, over 5666779.68 frames. ], batch size: 262, lr: 4.53e-03, grad_scale: 2.0
+2023-03-03 21:49:52,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.327e+02 1.506e+03 2.061e+03 2.733e+03 1.545e+04, threshold=4.123e+03, percent-clipped=11.0
+2023-03-03 21:50:03,705 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:50:07,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4074, 1.6838, 1.3261, 1.1736], device='cuda:0'), covar=tensor([0.1410, 0.1156, 0.0891, 0.1137], device='cuda:0'), in_proj_covar=tensor([0.1544, 0.1364, 0.1330, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 21:50:46,918 INFO [train.py:968] (0/2) Epoch 7, batch 32650, giga_loss[loss=0.2489, simple_loss=0.3306, pruned_loss=0.08361, over 28328.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3395, pruned_loss=0.09795, over 5667879.34 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3684, pruned_loss=0.1295, over 5675573.24 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3376, pruned_loss=0.0948, over 5671049.46 frames. ], batch size: 368, lr: 4.53e-03, grad_scale: 2.0
+2023-03-03 21:51:14,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4489, 2.1575, 1.6412, 0.6825], device='cuda:0'), covar=tensor([0.3179, 0.1558, 0.2307, 0.3413], device='cuda:0'), in_proj_covar=tensor([0.1455, 0.1388, 0.1438, 0.1201], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 21:51:49,870 INFO [train.py:968] (0/2) Epoch 7, batch 32700, giga_loss[loss=0.3099, simple_loss=0.3648, pruned_loss=0.1276, over 26754.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3369, pruned_loss=0.09523, over 5664905.89 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3683, pruned_loss=0.1294, over 5680107.65 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3349, pruned_loss=0.09221, over 5663365.18 frames. ], batch size: 555, lr: 4.53e-03, grad_scale: 2.0
+2023-03-03 21:51:55,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.096e+02 1.177e+03 1.593e+03 2.125e+03 7.178e+03, threshold=3.185e+03, percent-clipped=2.0
+2023-03-03 21:52:36,005 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-03 21:52:41,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306391.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:52:53,507 INFO [train.py:968] (0/2) Epoch 7, batch 32750, giga_loss[loss=0.2305, simple_loss=0.3097, pruned_loss=0.07563, over 28978.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3357, pruned_loss=0.09401, over 5664977.12 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3684, pruned_loss=0.1294, over 5682990.57 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3334, pruned_loss=0.09091, over 5660761.11 frames. ], batch size: 155, lr: 4.53e-03, grad_scale: 2.0
+2023-03-03 21:53:21,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:54:00,210 INFO [train.py:968] (0/2) Epoch 7, batch 32800, giga_loss[loss=0.2507, simple_loss=0.3347, pruned_loss=0.08338, over 28701.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3354, pruned_loss=0.09451, over 5675127.97 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3677, pruned_loss=0.129, over 5690634.40 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3331, pruned_loss=0.09123, over 5664303.16 frames. ], batch size: 243, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:54:05,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.726e+02 1.295e+03 1.676e+03 2.348e+03 6.961e+03, threshold=3.351e+03, percent-clipped=13.0
+2023-03-03 21:55:03,010 INFO [train.py:968] (0/2) Epoch 7, batch 32850, giga_loss[loss=0.2739, simple_loss=0.348, pruned_loss=0.09992, over 28040.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3372, pruned_loss=0.0951, over 5679098.21 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3679, pruned_loss=0.1294, over 5687715.18 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3342, pruned_loss=0.09103, over 5672806.42 frames. ], batch size: 412, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:55:46,167 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-03-03 21:55:50,031 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306534.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:55:54,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306537.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:56:10,807 INFO [train.py:968] (0/2) Epoch 7, batch 32900, giga_loss[loss=0.2812, simple_loss=0.3482, pruned_loss=0.1071, over 28950.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.338, pruned_loss=0.09585, over 5679586.62 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3679, pruned_loss=0.1294, over 5687638.88 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3352, pruned_loss=0.09221, over 5674703.30 frames. ], batch size: 199, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:56:15,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.087e+02 1.325e+03 1.713e+03 2.876e+03 9.123e+03, threshold=3.425e+03, percent-clipped=18.0
+2023-03-03 21:56:26,212 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:56:31,556 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306566.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:56:31,585 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:56:54,983 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:57:04,409 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:57:09,980 INFO [train.py:968] (0/2) Epoch 7, batch 32950, giga_loss[loss=0.3612, simple_loss=0.4024, pruned_loss=0.16, over 27682.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3394, pruned_loss=0.09754, over 5686769.80 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3675, pruned_loss=0.1292, over 5695542.66 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3365, pruned_loss=0.0937, over 5675486.76 frames. ], batch size: 472, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:57:13,028 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7180, 2.4158, 1.9657, 2.2797], device='cuda:0'), covar=tensor([0.0508, 0.0527, 0.0725, 0.0776], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0437, 0.0497, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 21:57:24,210 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2460, 3.1008, 2.9231, 1.4788], device='cuda:0'), covar=tensor([0.0844, 0.0852, 0.0876, 0.2142], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0879, 0.0781, 0.0617], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 21:57:36,781 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:58:10,148 INFO [train.py:968] (0/2) Epoch 7, batch 33000, giga_loss[loss=0.2754, simple_loss=0.3561, pruned_loss=0.09737, over 29007.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3385, pruned_loss=0.0965, over 5683793.05 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3672, pruned_loss=0.129, over 5700299.81 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3358, pruned_loss=0.09283, over 5670095.26 frames. ], batch size: 128, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:58:10,152 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 21:58:18,665 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2111, simple_loss=0.3099, pruned_loss=0.0561, over 944034.00 frames. 
+2023-03-03 21:58:18,665 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-03 21:58:23,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.246e+02 1.146e+03 1.495e+03 1.908e+03 4.640e+03, threshold=2.990e+03, percent-clipped=4.0
+2023-03-03 21:59:06,677 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 21:59:07,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2513, 1.3361, 1.1229, 1.0446], device='cuda:0'), covar=tensor([0.0615, 0.0362, 0.0819, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0434, 0.0493, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 21:59:11,105 INFO [train.py:968] (0/2) Epoch 7, batch 33050, giga_loss[loss=0.2666, simple_loss=0.3543, pruned_loss=0.08943, over 28957.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.34, pruned_loss=0.09638, over 5679883.14 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3664, pruned_loss=0.1286, over 5706931.89 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3374, pruned_loss=0.09256, over 5662012.11 frames. ], batch size: 155, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:59:13,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1309, 3.9651, 3.7005, 1.8420], device='cuda:0'), covar=tensor([0.0538, 0.0701, 0.0886, 0.2102], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0875, 0.0776, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:00:11,991 INFO [train.py:968] (0/2) Epoch 7, batch 33100, giga_loss[loss=0.2718, simple_loss=0.3458, pruned_loss=0.09887, over 28167.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3432, pruned_loss=0.09754, over 5679906.91 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3659, pruned_loss=0.1281, over 5711620.14 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3411, pruned_loss=0.0942, over 5660905.04 frames. ], batch size: 412, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:00:16,311 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.057e+02 1.442e+03 2.013e+03 2.717e+03 6.823e+03, threshold=4.027e+03, percent-clipped=19.0
+2023-03-03 22:00:25,821 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:00:28,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:00:43,019 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:01:05,352 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 22:01:06,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:01:14,940 INFO [train.py:968] (0/2) Epoch 7, batch 33150, giga_loss[loss=0.2809, simple_loss=0.3625, pruned_loss=0.09965, over 28670.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3438, pruned_loss=0.09753, over 5675732.60 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3657, pruned_loss=0.128, over 5705300.55 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3417, pruned_loss=0.0942, over 5664818.05 frames. ], batch size: 307, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:01:58,801 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-03 22:02:16,891 INFO [train.py:968] (0/2) Epoch 7, batch 33200, giga_loss[loss=0.2563, simple_loss=0.3287, pruned_loss=0.09193, over 28930.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3435, pruned_loss=0.09773, over 5665492.28 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3659, pruned_loss=0.1283, over 5696415.77 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3412, pruned_loss=0.09414, over 5663992.32 frames. ], batch size: 186, lr: 4.53e-03, grad_scale: 8.0
+2023-03-03 22:02:21,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.195e+02 1.295e+03 1.729e+03 2.286e+03 8.293e+03, threshold=3.457e+03, percent-clipped=3.0
+2023-03-03 22:02:43,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3214, 1.7231, 1.6878, 1.4097], device='cuda:0'), covar=tensor([0.1375, 0.1614, 0.1049, 0.1253], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0688, 0.0802, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 22:03:10,300 INFO [train.py:968] (0/2) Epoch 7, batch 33250, giga_loss[loss=0.2508, simple_loss=0.3263, pruned_loss=0.08763, over 28965.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3402, pruned_loss=0.09566, over 5676395.27 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3658, pruned_loss=0.1283, over 5703404.62 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3376, pruned_loss=0.09169, over 5668074.28 frames. ], batch size: 213, lr: 4.53e-03, grad_scale: 8.0
+2023-03-03 22:03:19,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3361, 3.1373, 1.4996, 1.3788], device='cuda:0'), covar=tensor([0.0936, 0.0296, 0.0908, 0.1372], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0488, 0.0322, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 22:04:02,796 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 22:04:12,378 INFO [train.py:968] (0/2) Epoch 7, batch 33300, giga_loss[loss=0.2474, simple_loss=0.328, pruned_loss=0.08342, over 28113.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3387, pruned_loss=0.0946, over 5668962.36 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3656, pruned_loss=0.1283, over 5696432.61 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3364, pruned_loss=0.09093, over 5667893.87 frames. ], batch size: 412, lr: 4.53e-03, grad_scale: 8.0
+2023-03-03 22:04:16,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.313e+02 1.212e+03 1.479e+03 1.909e+03 4.418e+03, threshold=2.957e+03, percent-clipped=5.0
+2023-03-03 22:04:30,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:05:14,561 INFO [train.py:968] (0/2) Epoch 7, batch 33350, giga_loss[loss=0.2156, simple_loss=0.2978, pruned_loss=0.0667, over 29048.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3373, pruned_loss=0.09446, over 5664687.74 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.366, pruned_loss=0.1286, over 5689795.99 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3347, pruned_loss=0.09079, over 5669593.08 frames. ], batch size: 128, lr: 4.53e-03, grad_scale: 8.0
+2023-03-03 22:05:32,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2830, 1.5864, 1.2124, 1.5586], device='cuda:0'), covar=tensor([0.2309, 0.2080, 0.2356, 0.1992], device='cuda:0'), in_proj_covar=tensor([0.1182, 0.0884, 0.1048, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 22:06:12,329 INFO [train.py:968] (0/2) Epoch 7, batch 33400, giga_loss[loss=0.2521, simple_loss=0.3396, pruned_loss=0.08229, over 28786.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3392, pruned_loss=0.0954, over 5671202.21 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.366, pruned_loss=0.1288, over 5694292.00 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3365, pruned_loss=0.09151, over 5670631.62 frames. ], batch size: 174, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:06:18,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.080e+02 1.349e+03 1.792e+03 2.532e+03 6.744e+03, threshold=3.585e+03, percent-clipped=17.0
+2023-03-03 22:06:58,585 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=307085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:07:16,345 INFO [train.py:968] (0/2) Epoch 7, batch 33450, giga_loss[loss=0.3035, simple_loss=0.3723, pruned_loss=0.1173, over 28927.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3417, pruned_loss=0.09701, over 5671121.93 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3655, pruned_loss=0.1284, over 5694388.81 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3395, pruned_loss=0.09356, over 5669951.62 frames. ], batch size: 199, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:07:22,175 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=307105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:07:26,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=307108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:07:37,415 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.21 vs. limit=5.0
+2023-03-03 22:07:48,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2513, 1.4073, 1.3075, 1.2941], device='cuda:0'), covar=tensor([0.1626, 0.1280, 0.0987, 0.1114], device='cuda:0'), in_proj_covar=tensor([0.1534, 0.1366, 0.1314, 0.1435], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 22:08:06,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=307137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:08:23,064 INFO [train.py:968] (0/2) Epoch 7, batch 33500, giga_loss[loss=0.2691, simple_loss=0.3451, pruned_loss=0.09651, over 29025.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3433, pruned_loss=0.09838, over 5667979.28 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3658, pruned_loss=0.1286, over 5697445.05 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3409, pruned_loss=0.09505, over 5663999.37 frames. ], batch size: 155, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:08:27,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=307152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:08:29,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.975e+02 1.364e+03 1.899e+03 2.781e+03 1.001e+04, threshold=3.798e+03, percent-clipped=17.0
+2023-03-03 22:08:40,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3919, 2.8655, 1.4553, 1.4133], device='cuda:0'), covar=tensor([0.0831, 0.0302, 0.0853, 0.1210], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0487, 0.0323, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 22:08:58,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4360, 1.6619, 1.3678, 1.8581], device='cuda:0'), covar=tensor([0.2211, 0.2090, 0.2220, 0.1893], device='cuda:0'), in_proj_covar=tensor([0.1184, 0.0886, 0.1050, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 22:09:22,976 INFO [train.py:968] (0/2) Epoch 7, batch 33550, giga_loss[loss=0.2802, simple_loss=0.3594, pruned_loss=0.1005, over 28895.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3472, pruned_loss=0.1012, over 5655203.46 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3649, pruned_loss=0.1281, over 5695613.29 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3452, pruned_loss=0.0978, over 5652345.51 frames. ], batch size: 213, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:10:13,583 INFO [train.py:968] (0/2) Epoch 7, batch 33600, giga_loss[loss=0.2489, simple_loss=0.3433, pruned_loss=0.07721, over 28836.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3505, pruned_loss=0.103, over 5662349.76 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.365, pruned_loss=0.1282, over 5694579.91 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.348, pruned_loss=0.09877, over 5659257.56 frames. ], batch size: 174, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:10:20,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.017e+02 1.335e+03 1.711e+03 2.396e+03 9.005e+03, threshold=3.422e+03, percent-clipped=10.0
+2023-03-03 22:11:18,560 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=307295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:11:22,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=307298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:11:23,579 INFO [train.py:968] (0/2) Epoch 7, batch 33650, giga_loss[loss=0.2873, simple_loss=0.3566, pruned_loss=0.1089, over 28940.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3518, pruned_loss=0.1034, over 5663726.20 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3649, pruned_loss=0.1281, over 5697751.68 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3498, pruned_loss=0.09981, over 5657928.54 frames. ], batch size: 228, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:12:01,036 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=307327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:12:32,202 INFO [train.py:968] (0/2) Epoch 7, batch 33700, giga_loss[loss=0.2491, simple_loss=0.325, pruned_loss=0.08664, over 28137.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3481, pruned_loss=0.1011, over 5673471.29 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3649, pruned_loss=0.1283, over 5703191.68 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3461, pruned_loss=0.09743, over 5663218.38 frames. ], batch size: 412, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:12:40,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.010e+02 1.349e+03 1.846e+03 2.632e+03 8.914e+03, threshold=3.691e+03, percent-clipped=8.0
+2023-03-03 22:13:34,250 INFO [train.py:968] (0/2) Epoch 7, batch 33750, giga_loss[loss=0.2658, simple_loss=0.3444, pruned_loss=0.09361, over 28919.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3468, pruned_loss=0.1001, over 5684307.54 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.365, pruned_loss=0.1284, over 5705473.60 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3448, pruned_loss=0.09654, over 5673656.48 frames. ], batch size: 186, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:14:41,568 INFO [train.py:968] (0/2) Epoch 7, batch 33800, giga_loss[loss=0.2493, simple_loss=0.3256, pruned_loss=0.08651, over 29080.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3458, pruned_loss=0.09968, over 5680893.11 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.365, pruned_loss=0.1284, over 5707212.90 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3438, pruned_loss=0.09628, over 5670424.13 frames. ], batch size: 136, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:14:51,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.002e+02 1.234e+03 1.771e+03 2.574e+03 6.935e+03, threshold=3.542e+03, percent-clipped=10.0
+2023-03-03 22:14:58,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=307460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:15:20,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7546, 5.5784, 5.3020, 2.3493], device='cuda:0'), covar=tensor([0.0364, 0.0599, 0.0736, 0.1900], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0879, 0.0778, 0.0615], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:15:39,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9382, 1.1976, 1.2661, 1.1289], device='cuda:0'), covar=tensor([0.1184, 0.1054, 0.1688, 0.1188], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0701, 0.0634, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 22:15:53,460 INFO [train.py:968] (0/2) Epoch 7, batch 33850, libri_loss[loss=0.3475, simple_loss=0.4057, pruned_loss=0.1446, over 29225.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3433, pruned_loss=0.0989, over 5677923.25 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3651, pruned_loss=0.1284, over 5705276.96 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3414, pruned_loss=0.09597, over 5671011.04 frames. ], batch size: 94, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:16:53,211 INFO [train.py:968] (0/2) Epoch 7, batch 33900, giga_loss[loss=0.2878, simple_loss=0.3583, pruned_loss=0.1087, over 28494.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3422, pruned_loss=0.0981, over 5686292.56 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3648, pruned_loss=0.1283, over 5709423.01 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3406, pruned_loss=0.09532, over 5676975.19 frames. ], batch size: 336, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:16:59,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.316e+02 1.367e+03 1.772e+03 2.602e+03 7.311e+03, threshold=3.543e+03, percent-clipped=12.0
+2023-03-03 22:17:53,186 INFO [train.py:968] (0/2) Epoch 7, batch 33950, giga_loss[loss=0.2163, simple_loss=0.3065, pruned_loss=0.06308, over 29002.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3415, pruned_loss=0.09691, over 5679200.69 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3649, pruned_loss=0.1283, over 5712782.50 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3396, pruned_loss=0.09389, over 5667972.08 frames. ], batch size: 199, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:17:57,381 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=307603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:18:01,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=307606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:18:33,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=307635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:18:49,194 INFO [train.py:968] (0/2) Epoch 7, batch 34000, giga_loss[loss=0.2416, simple_loss=0.3358, pruned_loss=0.07371, over 29067.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.09476, over 5683082.16 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3643, pruned_loss=0.1279, over 5712185.43 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3395, pruned_loss=0.09192, over 5673849.59 frames. ], batch size: 155, lr: 4.52e-03, grad_scale: 8.0
+2023-03-03 22:18:57,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.848e+02 1.364e+03 1.736e+03 2.656e+03 6.118e+03, threshold=3.472e+03, percent-clipped=6.0
+2023-03-03 22:19:47,190 INFO [train.py:968] (0/2) Epoch 7, batch 34050, giga_loss[loss=0.2951, simple_loss=0.3685, pruned_loss=0.1109, over 28876.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3426, pruned_loss=0.09413, over 5665274.61 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3643, pruned_loss=0.128, over 5692319.13 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3408, pruned_loss=0.09096, over 5676040.90 frames. ], batch size: 284, lr: 4.52e-03, grad_scale: 8.0
+2023-03-03 22:20:15,790 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 22:20:26,512 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=307735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:20:30,460 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-03 22:20:44,527 INFO [train.py:968] (0/2) Epoch 7, batch 34100, giga_loss[loss=0.256, simple_loss=0.3422, pruned_loss=0.08494, over 28666.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3441, pruned_loss=0.09521, over 5670539.72 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3639, pruned_loss=0.1277, over 5696620.29 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3426, pruned_loss=0.09217, over 5674743.93 frames. ], batch size: 262, lr: 4.52e-03, grad_scale: 8.0
+2023-03-03 22:20:51,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.608e+02 1.226e+03 1.518e+03 2.107e+03 6.601e+03, threshold=3.037e+03, percent-clipped=10.0
+2023-03-03 22:21:56,493 INFO [train.py:968] (0/2) Epoch 7, batch 34150, giga_loss[loss=0.2887, simple_loss=0.3704, pruned_loss=0.1035, over 28370.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3433, pruned_loss=0.0945, over 5670025.47 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3636, pruned_loss=0.1275, over 5698796.72 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3422, pruned_loss=0.092, over 5671403.52 frames. ], batch size: 368, lr: 4.52e-03, grad_scale: 8.0
+2023-03-03 22:22:57,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3273, 2.5537, 1.3001, 1.4570], device='cuda:0'), covar=tensor([0.0783, 0.0386, 0.0803, 0.1126], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0484, 0.0321, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 22:23:05,933 INFO [train.py:968] (0/2) Epoch 7, batch 34200, giga_loss[loss=0.2328, simple_loss=0.3249, pruned_loss=0.07035, over 28890.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3424, pruned_loss=0.09373, over 5668885.41 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3634, pruned_loss=0.1273, over 5702619.16 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3413, pruned_loss=0.09127, over 5665877.06 frames. ], batch size: 164, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:23:17,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.920e+02 1.419e+03 1.824e+03 2.764e+03 1.027e+04, threshold=3.648e+03, percent-clipped=19.0
+2023-03-03 22:23:28,979 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3837, 1.6131, 0.9808, 1.2214], device='cuda:0'), covar=tensor([0.0955, 0.0779, 0.1573, 0.1485], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0437, 0.0497, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:23:58,611 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-03 22:24:19,266 INFO [train.py:968] (0/2) Epoch 7, batch 34250, giga_loss[loss=0.2803, simple_loss=0.361, pruned_loss=0.09977, over 29009.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3425, pruned_loss=0.09277, over 5672201.40 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3631, pruned_loss=0.1271, over 5705668.34 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3416, pruned_loss=0.09059, over 5666753.48 frames. ], batch size: 285, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:24:56,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5194, 3.7113, 1.5543, 1.5197], device='cuda:0'), covar=tensor([0.0872, 0.0240, 0.0837, 0.1253], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0486, 0.0321, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 22:25:03,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5695, 3.7699, 1.6547, 1.5164], device='cuda:0'), covar=tensor([0.0849, 0.0222, 0.0822, 0.1252], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0485, 0.0321, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 22:25:27,326 INFO [train.py:968] (0/2) Epoch 7, batch 34300, giga_loss[loss=0.2476, simple_loss=0.3379, pruned_loss=0.07858, over 28697.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3444, pruned_loss=0.09455, over 5668919.89 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3628, pruned_loss=0.127, over 5702045.48 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3434, pruned_loss=0.09187, over 5666227.20 frames. ], batch size: 71, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:25:35,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.949e+02 1.452e+03 1.912e+03 2.619e+03 5.811e+03, threshold=3.824e+03, percent-clipped=9.0
+2023-03-03 22:25:38,791 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-03 22:26:02,650 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-03 22:26:21,660 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=307991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:26:27,288 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=307997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:26:28,988 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-308000.pt
+2023-03-03 22:26:29,294 INFO [train.py:968] (0/2) Epoch 7, batch 34350, libri_loss[loss=0.2985, simple_loss=0.3678, pruned_loss=0.1146, over 29526.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3487, pruned_loss=0.09698, over 5676185.53 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3628, pruned_loss=0.1269, over 5707521.59 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3473, pruned_loss=0.09385, over 5667813.69 frames. ], batch size: 81, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:26:46,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:27:32,683 INFO [train.py:968] (0/2) Epoch 7, batch 34400, libri_loss[loss=0.3219, simple_loss=0.3835, pruned_loss=0.1301, over 27972.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3493, pruned_loss=0.098, over 5679027.95 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3634, pruned_loss=0.1272, over 5706411.93 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3471, pruned_loss=0.09409, over 5672194.31 frames. ], batch size: 116, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:27:44,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.886e+02 1.587e+03 2.153e+03 3.356e+03 1.184e+04, threshold=4.306e+03, percent-clipped=18.0
+2023-03-03 22:28:29,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2873, 3.1235, 2.9459, 1.5844], device='cuda:0'), covar=tensor([0.0826, 0.0873, 0.0878, 0.2080], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0869, 0.0772, 0.0617], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:28:41,741 INFO [train.py:968] (0/2) Epoch 7, batch 34450, giga_loss[loss=0.2854, simple_loss=0.3518, pruned_loss=0.1095, over 28948.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3483, pruned_loss=0.09857, over 5672022.70 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3635, pruned_loss=0.1273, over 5700801.56 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.346, pruned_loss=0.09455, over 5670280.37 frames. ], batch size: 186, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:28:54,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:29:19,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0135, 1.2362, 3.6554, 3.0139], device='cuda:0'), covar=tensor([0.1644, 0.2442, 0.0377, 0.1099], device='cuda:0'), in_proj_covar=tensor([0.0595, 0.0557, 0.0783, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 22:29:54,520 INFO [train.py:968] (0/2) Epoch 7, batch 34500, giga_loss[loss=0.2429, simple_loss=0.3284, pruned_loss=0.07868, over 28835.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3456, pruned_loss=0.09658, over 5657175.88 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3638, pruned_loss=0.1274, over 5680899.24 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3433, pruned_loss=0.09283, over 5674100.97 frames. ], batch size: 99, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:30:06,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.868e+02 1.258e+03 1.566e+03 2.158e+03 4.959e+03, threshold=3.131e+03, percent-clipped=3.0
+2023-03-03 22:31:05,054 INFO [train.py:968] (0/2) Epoch 7, batch 34550, giga_loss[loss=0.2389, simple_loss=0.3091, pruned_loss=0.08436, over 24763.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3434, pruned_loss=0.09424, over 5661894.46 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3639, pruned_loss=0.1275, over 5680155.75 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3413, pruned_loss=0.09111, over 5675793.31 frames. ], batch size: 705, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:31:51,130 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-03 22:32:04,240 INFO [train.py:968] (0/2) Epoch 7, batch 34600, giga_loss[loss=0.2724, simple_loss=0.3492, pruned_loss=0.09783, over 28726.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3457, pruned_loss=0.09662, over 5664618.30 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3641, pruned_loss=0.1278, over 5687804.61 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3431, pruned_loss=0.09263, over 5668441.87 frames. ], batch size: 243, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:32:09,314 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:32:12,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:32:13,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.586e+02 1.219e+03 2.009e+03 2.696e+03 5.065e+03, threshold=4.019e+03, percent-clipped=15.0
+2023-03-03 22:32:47,515 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=308285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:33:04,924 INFO [train.py:968] (0/2) Epoch 7, batch 34650, giga_loss[loss=0.2783, simple_loss=0.3544, pruned_loss=0.1011, over 27594.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3478, pruned_loss=0.09732, over 5669569.76 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3637, pruned_loss=0.1274, over 5690277.13 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3458, pruned_loss=0.09389, over 5669868.86 frames. ], batch size: 472, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:34:06,599 INFO [train.py:968] (0/2) Epoch 7, batch 34700, giga_loss[loss=0.2349, simple_loss=0.3168, pruned_loss=0.07652, over 28687.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3466, pruned_loss=0.09688, over 5676096.80 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3633, pruned_loss=0.1272, over 5691270.86 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3453, pruned_loss=0.09416, over 5675324.79 frames. ], batch size: 307, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:34:17,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.452e+02 1.369e+03 1.870e+03 2.591e+03 6.398e+03, threshold=3.740e+03, percent-clipped=4.0
+2023-03-03 22:34:29,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:34:36,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:34:53,301 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:35:03,825 INFO [train.py:968] (0/2) Epoch 7, batch 34750, giga_loss[loss=0.2587, simple_loss=0.3385, pruned_loss=0.0894, over 28822.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3445, pruned_loss=0.09718, over 5661408.66 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3635, pruned_loss=0.1274, over 5684590.05 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.343, pruned_loss=0.09439, over 5666797.42 frames. ], batch size: 174, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:36:02,885 INFO [train.py:968] (0/2) Epoch 7, batch 34800, libri_loss[loss=0.2618, simple_loss=0.3282, pruned_loss=0.09764, over 29616.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3435, pruned_loss=0.09704, over 5665915.03 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3627, pruned_loss=0.1268, over 5689834.52 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3426, pruned_loss=0.0947, over 5664937.92 frames. ], batch size: 74, lr: 4.52e-03, grad_scale: 8.0
+2023-03-03 22:36:11,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.559e+02 1.561e+03 2.121e+03 2.771e+03 6.519e+03, threshold=4.241e+03, percent-clipped=11.0
+2023-03-03 22:36:54,828 INFO [train.py:968] (0/2) Epoch 7, batch 34850, libri_loss[loss=0.283, simple_loss=0.3382, pruned_loss=0.114, over 29551.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3514, pruned_loss=0.1026, over 5665455.59 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3631, pruned_loss=0.1273, over 5685742.11 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3498, pruned_loss=0.09942, over 5667027.49 frames. ], batch size: 76, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:37:03,180 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:04,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:07,712 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:09,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:21,064 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:23,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:31,113 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=308541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:36,694 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=308547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:38,392 INFO [train.py:968] (0/2) Epoch 7, batch 34900, giga_loss[loss=0.322, simple_loss=0.3943, pruned_loss=0.1248, over 29015.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3611, pruned_loss=0.1086, over 5660962.48 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3634, pruned_loss=0.1276, over 5680167.85 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3595, pruned_loss=0.1056, over 5666306.45 frames. ], batch size: 213, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:37:48,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.834e+02 1.267e+03 1.688e+03 2.511e+03 9.590e+03, threshold=3.375e+03, percent-clipped=5.0
+2023-03-03 22:37:51,567 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=308563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:38:17,223 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:38:21,308 INFO [train.py:968] (0/2) Epoch 7, batch 34950, giga_loss[loss=0.3477, simple_loss=0.4089, pruned_loss=0.1432, over 28823.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3662, pruned_loss=0.1126, over 5667620.86 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3637, pruned_loss=0.1278, over 5687125.19 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3647, pruned_loss=0.1094, over 5665192.33 frames. ], batch size: 199, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:38:30,911 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-03 22:39:06,556 INFO [train.py:968] (0/2) Epoch 7, batch 35000, giga_loss[loss=0.2749, simple_loss=0.3386, pruned_loss=0.1055, over 27778.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.362, pruned_loss=0.1114, over 5677596.39 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3637, pruned_loss=0.1277, over 5690532.15 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3608, pruned_loss=0.1087, over 5672262.97 frames. ], batch size: 472, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:39:14,688 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.971e+02 1.123e+03 1.479e+03 2.103e+03 1.111e+04, threshold=2.959e+03, percent-clipped=7.0
+2023-03-03 22:39:46,454 INFO [train.py:968] (0/2) Epoch 7, batch 35050, giga_loss[loss=0.2252, simple_loss=0.2991, pruned_loss=0.07564, over 28071.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3555, pruned_loss=0.1086, over 5676406.48 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3644, pruned_loss=0.1282, over 5679019.74 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3537, pruned_loss=0.1055, over 5681776.24 frames. ], batch size: 77, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:39:48,361 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:40:30,695 INFO [train.py:968] (0/2) Epoch 7, batch 35100, giga_loss[loss=0.2459, simple_loss=0.3089, pruned_loss=0.0914, over 28785.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3488, pruned_loss=0.1057, over 5677422.31 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3647, pruned_loss=0.1284, over 5682200.92 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3468, pruned_loss=0.1028, over 5678654.21 frames. ], batch size: 99, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:40:39,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.439e+02 1.118e+03 1.560e+03 2.427e+03 1.122e+04, threshold=3.120e+03, percent-clipped=16.0
+2023-03-03 22:41:12,932 INFO [train.py:968] (0/2) Epoch 7, batch 35150, giga_loss[loss=0.2332, simple_loss=0.3077, pruned_loss=0.07939, over 29003.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3432, pruned_loss=0.104, over 5684179.78 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3656, pruned_loss=0.1288, over 5686975.64 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3402, pruned_loss=0.1006, over 5680738.79 frames. ], batch size: 136, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:41:18,443 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4724, 5.2653, 4.9878, 2.4691], device='cuda:0'), covar=tensor([0.0289, 0.0415, 0.0497, 0.1808], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0879, 0.0777, 0.0617], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:41:28,185 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3271, 5.1164, 4.8543, 2.4226], device='cuda:0'), covar=tensor([0.0310, 0.0460, 0.0496, 0.1675], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0877, 0.0773, 0.0615], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:41:40,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4446, 2.1768, 1.7276, 0.6541], device='cuda:0'), covar=tensor([0.2936, 0.1777, 0.2521, 0.3453], device='cuda:0'), in_proj_covar=tensor([0.1430, 0.1379, 0.1410, 0.1179], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 22:41:55,264 INFO [train.py:968] (0/2) Epoch 7, batch 35200, giga_loss[loss=0.2181, simple_loss=0.2934, pruned_loss=0.07142, over 28241.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3344, pruned_loss=0.09945, over 5683972.50 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3657, pruned_loss=0.1288, over 5686645.49 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3316, pruned_loss=0.0963, over 5681374.57 frames. ], batch size: 368, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:42:03,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.158e+02 9.514e+02 1.240e+03 1.993e+03 5.890e+03, threshold=2.480e+03, percent-clipped=11.0
+2023-03-03 22:42:19,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2733, 2.1836, 1.8214, 1.9105], device='cuda:0'), covar=tensor([0.0415, 0.0369, 0.0643, 0.0674], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0436, 0.0496, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:42:37,285 INFO [train.py:968] (0/2) Epoch 7, batch 35250, giga_loss[loss=0.2188, simple_loss=0.2917, pruned_loss=0.07288, over 28548.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3301, pruned_loss=0.09735, over 5678165.76 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3663, pruned_loss=0.1291, over 5682965.99 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3266, pruned_loss=0.09395, over 5679463.49 frames. ], batch size: 336, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:43:18,779 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:43:21,968 INFO [train.py:968] (0/2) Epoch 7, batch 35300, libri_loss[loss=0.3268, simple_loss=0.3905, pruned_loss=0.1315, over 29379.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3275, pruned_loss=0.09582, over 5672607.17 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3669, pruned_loss=0.1294, over 5669812.86 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3231, pruned_loss=0.09203, over 5684763.27 frames. ], batch size: 92, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:43:30,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.394e+02 1.071e+03 1.441e+03 1.908e+03 4.492e+03, threshold=2.882e+03, percent-clipped=10.0
+2023-03-03 22:43:37,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:44:00,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4926, 4.3027, 4.0728, 1.9347], device='cuda:0'), covar=tensor([0.0449, 0.0607, 0.0634, 0.2108], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0880, 0.0773, 0.0616], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:44:01,379 INFO [train.py:968] (0/2) Epoch 7, batch 35350, giga_loss[loss=0.228, simple_loss=0.3016, pruned_loss=0.07723, over 28833.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3245, pruned_loss=0.09425, over 5685033.88 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3675, pruned_loss=0.1297, over 5671895.42 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3194, pruned_loss=0.08998, over 5692935.12 frames. ], batch size: 199, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:44:30,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6744, 1.5831, 1.8990, 1.4900], device='cuda:0'), covar=tensor([0.2021, 0.2959, 0.1498, 0.1639], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0715, 0.0823, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-03 22:44:47,709 INFO [train.py:968] (0/2) Epoch 7, batch 35400, giga_loss[loss=0.2223, simple_loss=0.2989, pruned_loss=0.07284, over 28707.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3211, pruned_loss=0.0924, over 5692645.90 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3679, pruned_loss=0.1299, over 5671073.71 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.316, pruned_loss=0.08826, over 5699720.22 frames. ], batch size: 242, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:44:53,970 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.850e+02 1.001e+03 1.203e+03 1.822e+03 5.745e+03, threshold=2.407e+03, percent-clipped=8.0
+2023-03-03 22:44:57,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=309062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:45:11,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=309077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:45:28,401 INFO [train.py:968] (0/2) Epoch 7, batch 35450, giga_loss[loss=0.2601, simple_loss=0.3224, pruned_loss=0.09892, over 27926.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3176, pruned_loss=0.0906, over 5691239.41 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3675, pruned_loss=0.1296, over 5667550.87 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3128, pruned_loss=0.08674, over 5699591.03 frames. ], batch size: 412, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:45:41,196 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=309113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:45:43,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=309116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:45:51,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=309125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:46:08,494 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:46:12,001 INFO [train.py:968] (0/2) Epoch 7, batch 35500, giga_loss[loss=0.2345, simple_loss=0.2996, pruned_loss=0.08474, over 28811.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3157, pruned_loss=0.08985, over 5687666.40 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3684, pruned_loss=0.1299, over 5669607.59 frames. ], giga_tot_loss[loss=0.2405, simple_loss=0.3099, pruned_loss=0.08554, over 5693167.10 frames. ], batch size: 186, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:46:19,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.454e+02 1.077e+03 1.297e+03 1.719e+03 4.806e+03, threshold=2.594e+03, percent-clipped=8.0
+2023-03-03 22:46:51,868 INFO [train.py:968] (0/2) Epoch 7, batch 35550, giga_loss[loss=0.2174, simple_loss=0.2913, pruned_loss=0.07176, over 28654.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3134, pruned_loss=0.08878, over 5686386.69 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.369, pruned_loss=0.1302, over 5673375.30 frames. ], giga_tot_loss[loss=0.2371, simple_loss=0.3066, pruned_loss=0.08377, over 5687811.94 frames. ], batch size: 60, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:47:08,652 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=309220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:47:11,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=309223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:47:17,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7671, 1.0821, 2.9134, 2.7226], device='cuda:0'), covar=tensor([0.1505, 0.2242, 0.0530, 0.0752], device='cuda:0'), in_proj_covar=tensor([0.0587, 0.0547, 0.0777, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 22:47:34,082 INFO [train.py:968] (0/2) Epoch 7, batch 35600, giga_loss[loss=0.2133, simple_loss=0.2835, pruned_loss=0.07159, over 29000.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3115, pruned_loss=0.08803, over 5685044.03 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3695, pruned_loss=0.1305, over 5666115.83 frames. ], giga_tot_loss[loss=0.2348, simple_loss=0.3041, pruned_loss=0.08274, over 5692706.36 frames. ], batch size: 128, lr: 4.51e-03, grad_scale: 8.0
+2023-03-03 22:47:36,232 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:47:43,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.650e+02 1.038e+03 1.483e+03 2.121e+03 8.997e+03, threshold=2.966e+03, percent-clipped=17.0
+2023-03-03 22:48:19,473 INFO [train.py:968] (0/2) Epoch 7, batch 35650, giga_loss[loss=0.2519, simple_loss=0.3201, pruned_loss=0.09186, over 29098.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3096, pruned_loss=0.08741, over 5684048.25 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3701, pruned_loss=0.1309, over 5661725.64 frames. ], giga_tot_loss[loss=0.2328, simple_loss=0.3019, pruned_loss=0.08186, over 5695452.37 frames. ], batch size: 128, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:48:31,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5203, 4.3110, 4.1023, 2.0714], device='cuda:0'), covar=tensor([0.0438, 0.0642, 0.0637, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0883, 0.0774, 0.0616], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:48:36,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=309320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:48:55,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5735, 1.6578, 1.4615, 1.2780], device='cuda:0'), covar=tensor([0.1253, 0.1220, 0.0934, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.1565, 0.1368, 0.1317, 0.1458], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 22:49:03,202 INFO [train.py:968] (0/2) Epoch 7, batch 35700, giga_loss[loss=0.2186, simple_loss=0.2901, pruned_loss=0.07356, over 28782.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3189, pruned_loss=0.09239, over 5659674.80 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3707, pruned_loss=0.131, over 5647748.14 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3101, pruned_loss=0.08633, over 5683771.97 frames. ], batch size: 99, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:49:11,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.578e+02 1.228e+03 1.558e+03 2.275e+03 8.703e+03, threshold=3.116e+03, percent-clipped=15.0
+2023-03-03 22:49:44,687 INFO [train.py:968] (0/2) Epoch 7, batch 35750, giga_loss[loss=0.313, simple_loss=0.3909, pruned_loss=0.1176, over 28954.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3306, pruned_loss=0.0989, over 5675608.26 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3703, pruned_loss=0.1308, over 5660005.88 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3208, pruned_loss=0.09188, over 5685519.13 frames. ], batch size: 136, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:50:14,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=309437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:50:26,146 INFO [train.py:968] (0/2) Epoch 7, batch 35800, giga_loss[loss=0.3301, simple_loss=0.4026, pruned_loss=0.1288, over 28775.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3437, pruned_loss=0.1063, over 5686884.21 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3701, pruned_loss=0.1307, over 5666423.81 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3352, pruned_loss=0.1, over 5689584.60 frames. ], batch size: 199, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:50:27,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0884, 1.3711, 3.7531, 3.0219], device='cuda:0'), covar=tensor([0.1660, 0.2410, 0.0386, 0.0968], device='cuda:0'), in_proj_covar=tensor([0.0590, 0.0548, 0.0778, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 22:50:36,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.412e+02 1.207e+03 1.513e+03 2.158e+03 5.450e+03, threshold=3.025e+03, percent-clipped=7.0
+2023-03-03 22:50:37,704 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=309463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:50:39,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=309466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:51:05,460 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:51:07,030 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 22:51:08,480 INFO [train.py:968] (0/2) Epoch 7, batch 35850, giga_loss[loss=0.3294, simple_loss=0.3972, pruned_loss=0.1308, over 27913.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.351, pruned_loss=0.1091, over 5681421.74 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3701, pruned_loss=0.1304, over 5663723.48 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3435, pruned_loss=0.1037, over 5686212.24 frames. ], batch size: 412, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:51:09,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=309500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:51:37,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9773, 5.6723, 5.4470, 2.7001], device='cuda:0'), covar=tensor([0.0410, 0.0610, 0.0747, 0.1569], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0881, 0.0775, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:51:50,337 INFO [train.py:968] (0/2) Epoch 7, batch 35900, libri_loss[loss=0.2865, simple_loss=0.349, pruned_loss=0.112, over 29592.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3541, pruned_loss=0.1091, over 5681508.82 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3699, pruned_loss=0.1302, over 5669392.96 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.348, pruned_loss=0.1045, over 5680589.45 frames. ], batch size: 74, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:51:59,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.686e+02 1.204e+03 1.575e+03 2.107e+03 5.354e+03, threshold=3.151e+03, percent-clipped=6.0
+2023-03-03 22:52:20,602 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=309580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:52:23,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=309583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:52:36,981 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-03 22:52:39,070 INFO [train.py:968] (0/2) Epoch 7, batch 35950, giga_loss[loss=0.2818, simple_loss=0.3598, pruned_loss=0.1019, over 28706.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3564, pruned_loss=0.109, over 5689446.12 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3704, pruned_loss=0.1304, over 5671375.04 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3509, pruned_loss=0.105, over 5687055.62 frames. ], batch size: 284, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:52:49,610 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:52:53,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4824, 1.4808, 1.2971, 1.3096], device='cuda:0'), covar=tensor([0.1222, 0.1201, 0.0983, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.1560, 0.1370, 0.1326, 0.1459], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-03 22:53:16,967 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=309643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:53:18,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=309646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:53:21,336 INFO [train.py:968] (0/2) Epoch 7, batch 36000, giga_loss[loss=0.2868, simple_loss=0.3493, pruned_loss=0.1121, over 28693.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3594, pruned_loss=0.1112, over 5697052.53 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3709, pruned_loss=0.1307, over 5677402.12 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3543, pruned_loss=0.1072, over 5690450.35 frames. ], batch size: 85, lr: 4.51e-03, grad_scale: 8.0
+2023-03-03 22:53:21,340 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 22:53:30,039 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2225, simple_loss=0.3279, pruned_loss=0.05855, over 944034.00 frames. 
+2023-03-03 22:53:30,040 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-03 22:53:39,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.803e+02 1.101e+03 1.411e+03 2.007e+03 4.591e+03, threshold=2.823e+03, percent-clipped=6.0
+2023-03-03 22:53:53,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:54:02,819 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-03 22:54:09,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6749, 1.5502, 1.2667, 1.1995], device='cuda:0'), covar=tensor([0.0670, 0.0548, 0.0961, 0.0866], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0436, 0.0493, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 22:54:13,663 INFO [train.py:968] (0/2) Epoch 7, batch 36050, giga_loss[loss=0.3013, simple_loss=0.3686, pruned_loss=0.117, over 28863.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3617, pruned_loss=0.1129, over 5687097.22 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3712, pruned_loss=0.1309, over 5676157.67 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3574, pruned_loss=0.1095, over 5683324.90 frames. ], batch size: 199, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:54:55,511 INFO [train.py:968] (0/2) Epoch 7, batch 36100, giga_loss[loss=0.2779, simple_loss=0.3539, pruned_loss=0.1009, over 28282.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3661, pruned_loss=0.1162, over 5689327.56 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3717, pruned_loss=0.1311, over 5681273.84 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.362, pruned_loss=0.113, over 5681840.74 frames. ], batch size: 77, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:55:05,114 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.486e+02 1.197e+03 1.610e+03 2.538e+03 1.101e+04, threshold=3.220e+03, percent-clipped=16.0
+2023-03-03 22:55:37,390 INFO [train.py:968] (0/2) Epoch 7, batch 36150, giga_loss[loss=0.2757, simple_loss=0.3573, pruned_loss=0.09707, over 28442.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3688, pruned_loss=0.1173, over 5699425.74 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3724, pruned_loss=0.1313, over 5687812.53 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3648, pruned_loss=0.1139, over 5687601.19 frames. ], batch size: 71, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:55:39,829 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=309804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:55:57,214 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-03 22:56:06,767 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-03 22:56:15,850 INFO [train.py:968] (0/2) Epoch 7, batch 36200, giga_loss[loss=0.267, simple_loss=0.3476, pruned_loss=0.09317, over 28555.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3702, pruned_loss=0.1173, over 5692510.09 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3724, pruned_loss=0.1313, over 5687551.11 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3669, pruned_loss=0.1144, over 5683521.15 frames. ], batch size: 85, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:56:28,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.152e+02 1.215e+03 1.546e+03 1.978e+03 7.026e+03, threshold=3.091e+03, percent-clipped=7.0
+2023-03-03 22:57:00,012 INFO [train.py:968] (0/2) Epoch 7, batch 36250, giga_loss[loss=0.3033, simple_loss=0.3693, pruned_loss=0.1186, over 28462.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3713, pruned_loss=0.1172, over 5681783.49 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3725, pruned_loss=0.1314, over 5679763.43 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3686, pruned_loss=0.1148, over 5681928.19 frames. ], batch size: 71, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:57:31,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7507, 1.0076, 3.3130, 2.7913], device='cuda:0'), covar=tensor([0.2579, 0.3239, 0.0806, 0.1053], device='cuda:0'), in_proj_covar=tensor([0.0593, 0.0546, 0.0771, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 22:57:39,573 INFO [train.py:968] (0/2) Epoch 7, batch 36300, giga_loss[loss=0.336, simple_loss=0.3948, pruned_loss=0.1386, over 28672.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3714, pruned_loss=0.1158, over 5694012.09 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3728, pruned_loss=0.1313, over 5685353.33 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.369, pruned_loss=0.1135, over 5689317.77 frames. ], batch size: 92, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:57:48,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.463e+02 1.085e+03 1.302e+03 1.722e+03 4.076e+03, threshold=2.605e+03, percent-clipped=4.0
+2023-03-03 22:57:53,336 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=309966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:58:20,556 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-310000.pt
+2023-03-03 22:58:20,871 INFO [train.py:968] (0/2) Epoch 7, batch 36350, giga_loss[loss=0.266, simple_loss=0.3512, pruned_loss=0.09041, over 29073.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3691, pruned_loss=0.113, over 5700154.37 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3732, pruned_loss=0.1316, over 5688606.37 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3668, pruned_loss=0.1108, over 5693837.31 frames. ], batch size: 136, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:59:01,121 INFO [train.py:968] (0/2) Epoch 7, batch 36400, giga_loss[loss=0.2701, simple_loss=0.3475, pruned_loss=0.09637, over 28533.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3672, pruned_loss=0.1113, over 5701344.62 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3739, pruned_loss=0.132, over 5683612.64 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3646, pruned_loss=0.1087, over 5702081.50 frames. ], batch size: 85, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:59:11,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.140e+02 1.179e+03 1.435e+03 1.959e+03 5.093e+03, threshold=2.870e+03, percent-clipped=11.0
+2023-03-03 22:59:12,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7939, 1.1158, 3.6754, 2.8121], device='cuda:0'), covar=tensor([0.1920, 0.2481, 0.0396, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0598, 0.0549, 0.0780, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 22:59:40,430 INFO [train.py:968] (0/2) Epoch 7, batch 36450, giga_loss[loss=0.318, simple_loss=0.381, pruned_loss=0.1275, over 28876.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3695, pruned_loss=0.1146, over 5703928.77 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3744, pruned_loss=0.1322, over 5692380.26 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3666, pruned_loss=0.1115, over 5697577.20 frames. ], batch size: 186, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 23:00:20,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4692, 3.4402, 1.6102, 1.3895], device='cuda:0'), covar=tensor([0.0858, 0.0262, 0.0764, 0.1293], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0483, 0.0316, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 23:00:26,251 INFO [train.py:968] (0/2) Epoch 7, batch 36500, giga_loss[loss=0.3062, simple_loss=0.3701, pruned_loss=0.1212, over 28588.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3741, pruned_loss=0.1206, over 5692983.88 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3751, pruned_loss=0.1325, over 5686215.87 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3713, pruned_loss=0.1176, over 5693733.05 frames. ], batch size: 307, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 23:00:37,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.020e+02 1.306e+03 1.794e+03 2.464e+03 8.881e+03, threshold=3.587e+03, percent-clipped=17.0
+2023-03-03 23:00:38,479 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 23:00:52,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=310179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:01:05,090 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 23:01:09,434 INFO [train.py:968] (0/2) Epoch 7, batch 36550, giga_loss[loss=0.3022, simple_loss=0.3468, pruned_loss=0.1288, over 24107.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3737, pruned_loss=0.122, over 5692940.36 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3747, pruned_loss=0.1322, over 5687385.05 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3718, pruned_loss=0.1198, over 5692374.32 frames. ], batch size: 705, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 23:01:17,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3155, 1.7254, 1.5937, 1.2197], device='cuda:0'), covar=tensor([0.1299, 0.1974, 0.1170, 0.1293], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0704, 0.0813, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 23:01:31,529 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 23:01:45,192 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=310239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:01:53,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=310248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:01:55,021 INFO [train.py:968] (0/2) Epoch 7, batch 36600, libri_loss[loss=0.2979, simple_loss=0.3661, pruned_loss=0.1149, over 29528.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3723, pruned_loss=0.1216, over 5689401.37 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3755, pruned_loss=0.1327, over 5673789.78 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.37, pruned_loss=0.1191, over 5701553.09 frames. ], batch size: 82, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 23:01:57,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4932, 1.6363, 1.4006, 1.5050], device='cuda:0'), covar=tensor([0.2024, 0.1972, 0.2058, 0.1962], device='cuda:0'), in_proj_covar=tensor([0.1177, 0.0886, 0.1036, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 23:02:06,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.064e+02 1.161e+03 1.363e+03 1.968e+03 4.372e+03, threshold=2.726e+03, percent-clipped=2.0
+2023-03-03 23:02:24,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2509, 1.5243, 1.2062, 1.1813], device='cuda:0'), covar=tensor([0.2093, 0.2039, 0.2195, 0.2010], device='cuda:0'), in_proj_covar=tensor([0.1184, 0.0892, 0.1044, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 23:02:31,437 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-03 23:02:34,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4556, 4.2542, 4.1051, 1.7593], device='cuda:0'), covar=tensor([0.0495, 0.0611, 0.0617, 0.2088], device='cuda:0'), in_proj_covar=tensor([0.0921, 0.0875, 0.0771, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 23:02:37,378 INFO [train.py:968] (0/2) Epoch 7, batch 36650, giga_loss[loss=0.2938, simple_loss=0.3621, pruned_loss=0.1127, over 28791.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3705, pruned_loss=0.121, over 5692353.03 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.375, pruned_loss=0.1323, over 5678535.42 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.369, pruned_loss=0.1191, over 5698130.35 frames. ], batch size: 284, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 23:03:00,272 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=310322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:03:02,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=310325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:03:14,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=310341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:03:24,191 INFO [train.py:968] (0/2) Epoch 7, batch 36700, giga_loss[loss=0.2755, simple_loss=0.3473, pruned_loss=0.1019, over 28928.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3681, pruned_loss=0.1187, over 5693734.34 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3748, pruned_loss=0.1321, over 5678929.76 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.367, pruned_loss=0.1173, over 5698078.28 frames. ], batch size: 136, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:03:27,184 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=310354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:03:34,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.219e+02 1.178e+03 1.387e+03 1.786e+03 5.290e+03, threshold=2.775e+03, percent-clipped=4.0
+2023-03-03 23:03:43,324 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=310374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:04:06,427 INFO [train.py:968] (0/2) Epoch 7, batch 36750, giga_loss[loss=0.3205, simple_loss=0.372, pruned_loss=0.1345, over 26513.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3671, pruned_loss=0.1174, over 5688830.77 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3754, pruned_loss=0.1323, over 5688253.90 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3653, pruned_loss=0.1155, over 5684347.22 frames. ], batch size: 555, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:04:18,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0642, 2.2932, 2.2621, 1.8824], device='cuda:0'), covar=tensor([0.1481, 0.1690, 0.1082, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0706, 0.0813, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 23:04:35,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3222, 3.1194, 1.4626, 1.4058], device='cuda:0'), covar=tensor([0.0927, 0.0244, 0.0835, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0484, 0.0315, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 23:04:49,962 INFO [train.py:968] (0/2) Epoch 7, batch 36800, libri_loss[loss=0.4327, simple_loss=0.4685, pruned_loss=0.1985, over 25683.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3631, pruned_loss=0.1143, over 5687642.44 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.376, pruned_loss=0.1325, over 5686410.11 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3611, pruned_loss=0.1123, over 5686053.56 frames. ], batch size: 136, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:04:56,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0953, 1.1368, 3.9063, 3.1598], device='cuda:0'), covar=tensor([0.2052, 0.2919, 0.0559, 0.1545], device='cuda:0'), in_proj_covar=tensor([0.0595, 0.0551, 0.0782, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 23:05:02,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.230e+02 1.033e+03 1.348e+03 1.843e+03 1.242e+04, threshold=2.697e+03, percent-clipped=10.0
+2023-03-03 23:05:22,517 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=310484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:05:28,116 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=310487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:05:37,875 INFO [train.py:968] (0/2) Epoch 7, batch 36850, giga_loss[loss=0.2751, simple_loss=0.3413, pruned_loss=0.1044, over 28206.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3558, pruned_loss=0.1103, over 5669494.49 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3758, pruned_loss=0.1323, over 5680474.28 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.354, pruned_loss=0.1084, over 5672627.66 frames. ], batch size: 368, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:05:52,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=310516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:06:06,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5646, 1.9000, 1.8901, 1.5081], device='cuda:0'), covar=tensor([0.1604, 0.1935, 0.1170, 0.1370], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0708, 0.0817, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 23:06:29,122 INFO [train.py:968] (0/2) Epoch 7, batch 36900, giga_loss[loss=0.3189, simple_loss=0.3593, pruned_loss=0.1393, over 26469.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.351, pruned_loss=0.1082, over 5657415.18 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3763, pruned_loss=0.1325, over 5684450.24 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3484, pruned_loss=0.1058, over 5655768.37 frames. ], batch size: 555, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:06:43,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.181e+02 8.833e+02 1.097e+03 1.419e+03 4.952e+03, threshold=2.195e+03, percent-clipped=5.0
+2023-03-03 23:07:17,128 INFO [train.py:968] (0/2) Epoch 7, batch 36950, libri_loss[loss=0.3214, simple_loss=0.3864, pruned_loss=0.1282, over 29723.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3482, pruned_loss=0.1065, over 5659512.65 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3766, pruned_loss=0.1325, over 5688108.76 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3452, pruned_loss=0.104, over 5654029.60 frames. ], batch size: 87, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:07:28,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=310614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:07:35,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=310623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:07:58,744 INFO [train.py:968] (0/2) Epoch 7, batch 37000, giga_loss[loss=0.2458, simple_loss=0.3269, pruned_loss=0.08239, over 29093.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3473, pruned_loss=0.1049, over 5669207.61 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3768, pruned_loss=0.1326, over 5685977.78 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3446, pruned_loss=0.1027, over 5666423.32 frames. ], batch size: 155, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:08:12,225 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.471e+02 1.025e+03 1.292e+03 1.703e+03 8.736e+03, threshold=2.585e+03, percent-clipped=17.0
+2023-03-03 23:08:25,959 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=310680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:08:41,676 INFO [train.py:968] (0/2) Epoch 7, batch 37050, giga_loss[loss=0.2725, simple_loss=0.3493, pruned_loss=0.09786, over 28635.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3474, pruned_loss=0.1049, over 5677647.83 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3778, pruned_loss=0.1332, over 5688372.52 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3438, pruned_loss=0.1021, over 5672978.63 frames. ], batch size: 307, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:09:23,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=310749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:09:23,952 INFO [train.py:968] (0/2) Epoch 7, batch 37100, giga_loss[loss=0.2734, simple_loss=0.3431, pruned_loss=0.1018, over 28966.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3457, pruned_loss=0.1037, over 5686958.24 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3787, pruned_loss=0.1335, over 5685890.54 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.341, pruned_loss=0.1003, over 5685347.76 frames. ], batch size: 213, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:09:28,536 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=310757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:09:31,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=310760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:09:35,045 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.889e+02 1.001e+03 1.144e+03 1.537e+03 5.046e+03, threshold=2.289e+03, percent-clipped=12.0
+2023-03-03 23:09:36,650 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=310766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:09:38,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=310769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:09:53,751 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=310789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:10:02,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=310798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:10:03,554 INFO [train.py:968] (0/2) Epoch 7, batch 37150, giga_loss[loss=0.2342, simple_loss=0.3122, pruned_loss=0.07814, over 28957.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3425, pruned_loss=0.1019, over 5706692.86 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3789, pruned_loss=0.1333, over 5691671.55 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3377, pruned_loss=0.09863, over 5700502.60 frames. ], batch size: 164, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:10:11,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6374, 1.7093, 1.4568, 2.0890], device='cuda:0'), covar=tensor([0.2175, 0.2106, 0.2244, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.1188, 0.0896, 0.1048, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 23:10:40,174 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-03 23:10:44,556 INFO [train.py:968] (0/2) Epoch 7, batch 37200, libri_loss[loss=0.3486, simple_loss=0.4228, pruned_loss=0.1373, over 29636.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3396, pruned_loss=0.1005, over 5708141.68 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.3791, pruned_loss=0.1332, over 5694805.07 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.335, pruned_loss=0.09749, over 5700415.69 frames. ], batch size: 91, lr: 4.50e-03, grad_scale: 8.0
+2023-03-03 23:10:55,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.127e+02 8.847e+02 1.130e+03 1.562e+03 5.668e+03, threshold=2.260e+03, percent-clipped=10.0
+2023-03-03 23:11:16,074 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=310892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:11:16,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5077, 3.6587, 1.6381, 1.6122], device='cuda:0'), covar=tensor([0.0811, 0.0319, 0.0771, 0.1090], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0485, 0.0315, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 23:11:17,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=310895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:11:20,899 INFO [train.py:968] (0/2) Epoch 7, batch 37250, giga_loss[loss=0.2351, simple_loss=0.3021, pruned_loss=0.08406, over 28412.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3373, pruned_loss=0.09957, over 5713107.91 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3791, pruned_loss=0.1329, over 5700525.71 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3324, pruned_loss=0.09651, over 5702092.63 frames. ], batch size: 71, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:11:21,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9753, 1.7986, 1.8705, 1.6395], device='cuda:0'), covar=tensor([0.0816, 0.1224, 0.1214, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0731, 0.0649, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 23:11:40,670 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=310924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:12:00,781 INFO [train.py:968] (0/2) Epoch 7, batch 37300, giga_loss[loss=0.2511, simple_loss=0.3212, pruned_loss=0.09049, over 28848.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3366, pruned_loss=0.09966, over 5724885.34 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3804, pruned_loss=0.1336, over 5705963.39 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3305, pruned_loss=0.0957, over 5711735.70 frames. ], batch size: 112, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:12:13,080 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.972e+02 1.004e+03 1.245e+03 2.095e+03 7.332e+03, threshold=2.491e+03, percent-clipped=21.0
+2023-03-03 23:12:29,052 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 23:12:41,065 INFO [train.py:968] (0/2) Epoch 7, batch 37350, giga_loss[loss=0.2531, simple_loss=0.3235, pruned_loss=0.09135, over 28698.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.334, pruned_loss=0.09794, over 5726635.26 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3808, pruned_loss=0.1337, over 5708571.84 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3281, pruned_loss=0.09423, over 5714055.36 frames. ], batch size: 242, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:13:13,504 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=311042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:13:18,783 INFO [train.py:968] (0/2) Epoch 7, batch 37400, giga_loss[loss=0.2742, simple_loss=0.3321, pruned_loss=0.1081, over 23977.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3338, pruned_loss=0.09808, over 5720251.96 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3814, pruned_loss=0.1336, over 5710353.86 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3261, pruned_loss=0.0933, over 5708983.20 frames. ], batch size: 705, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:13:22,200 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=311055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:13:30,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.239e+02 9.521e+02 1.253e+03 2.013e+03 1.179e+04, threshold=2.507e+03, percent-clipped=15.0
+2023-03-03 23:13:35,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9897, 1.0976, 3.8044, 3.1902], device='cuda:0'), covar=tensor([0.1815, 0.2666, 0.0414, 0.0718], device='cuda:0'), in_proj_covar=tensor([0.0596, 0.0552, 0.0781, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 23:13:58,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4385, 1.8991, 1.7820, 1.3359], device='cuda:0'), covar=tensor([0.1457, 0.2062, 0.1220, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0714, 0.0820, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 23:13:58,592 INFO [train.py:968] (0/2) Epoch 7, batch 37450, libri_loss[loss=0.3619, simple_loss=0.4241, pruned_loss=0.1498, over 29129.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.333, pruned_loss=0.09716, over 5716822.10 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.382, pruned_loss=0.1339, over 5704472.92 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3252, pruned_loss=0.09241, over 5712612.84 frames. ], batch size: 101, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:14:37,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1457, 1.2671, 3.7187, 3.0859], device='cuda:0'), covar=tensor([0.1399, 0.2083, 0.0397, 0.0830], device='cuda:0'), in_proj_covar=tensor([0.0590, 0.0546, 0.0772, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 23:14:37,703 INFO [train.py:968] (0/2) Epoch 7, batch 37500, libri_loss[loss=0.3798, simple_loss=0.4302, pruned_loss=0.1647, over 25788.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3331, pruned_loss=0.0975, over 5702116.13 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3828, pruned_loss=0.1343, over 5696742.62 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3246, pruned_loss=0.09204, over 5705863.91 frames. ], batch size: 136, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:14:51,802 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.124e+02 9.098e+02 1.165e+03 1.755e+03 7.987e+03, threshold=2.329e+03, percent-clipped=9.0
+2023-03-03 23:15:15,942 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=311196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:15:17,421 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=311198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:15:18,465 INFO [train.py:968] (0/2) Epoch 7, batch 37550, giga_loss[loss=0.2628, simple_loss=0.329, pruned_loss=0.09826, over 28737.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3348, pruned_loss=0.09835, over 5716935.22 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3832, pruned_loss=0.1343, over 5702482.87 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3264, pruned_loss=0.09306, over 5714771.79 frames. ], batch size: 92, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:15:19,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=311201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:15:45,640 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=311230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:16:04,003 INFO [train.py:968] (0/2) Epoch 7, batch 37600, giga_loss[loss=0.3136, simple_loss=0.3642, pruned_loss=0.1315, over 23638.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3403, pruned_loss=0.1022, over 5714779.60 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3831, pruned_loss=0.1342, over 5707357.24 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3326, pruned_loss=0.09738, over 5709136.75 frames. ], batch size: 705, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:16:16,646 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.357e+02 1.229e+03 1.599e+03 2.297e+03 9.088e+03, threshold=3.199e+03, percent-clipped=24.0
+2023-03-03 23:16:24,026 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-03 23:16:45,678 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-03 23:16:47,063 INFO [train.py:968] (0/2) Epoch 7, batch 37650, libri_loss[loss=0.3325, simple_loss=0.4008, pruned_loss=0.1321, over 29543.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3474, pruned_loss=0.1067, over 5720332.88 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3836, pruned_loss=0.1343, over 5714193.06 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3393, pruned_loss=0.1015, over 5709629.82 frames. ], batch size: 82, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:17:00,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8367, 1.6212, 1.3358, 1.4288], device='cuda:0'), covar=tensor([0.0535, 0.0551, 0.0793, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0441, 0.0498, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 23:17:35,049 INFO [train.py:968] (0/2) Epoch 7, batch 37700, giga_loss[loss=0.3454, simple_loss=0.3972, pruned_loss=0.1468, over 28654.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3558, pruned_loss=0.1126, over 5702818.68 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3839, pruned_loss=0.1346, over 5707988.35 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3485, pruned_loss=0.1077, over 5700083.40 frames. ], batch size: 92, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:17:49,897 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.118e+02 1.337e+03 1.725e+03 2.610e+03 9.909e+03, threshold=3.451e+03, percent-clipped=12.0
+2023-03-03 23:18:23,712 INFO [train.py:968] (0/2) Epoch 7, batch 37750, giga_loss[loss=0.2894, simple_loss=0.3656, pruned_loss=0.1066, over 28814.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3594, pruned_loss=0.1143, over 5682738.37 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3828, pruned_loss=0.134, over 5701881.98 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3538, pruned_loss=0.1104, over 5685869.94 frames. ], batch size: 99, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:18:38,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=311417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:18:41,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=311421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:19:09,567 INFO [train.py:968] (0/2) Epoch 7, batch 37800, giga_loss[loss=0.2741, simple_loss=0.3577, pruned_loss=0.09529, over 28553.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3648, pruned_loss=0.1165, over 5684852.91 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3832, pruned_loss=0.1342, over 5702675.66 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1127, over 5686339.89 frames. ], batch size: 78, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:19:22,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.263e+02 1.115e+03 1.435e+03 2.022e+03 9.727e+03, threshold=2.870e+03, percent-clipped=4.0
+2023-03-03 23:19:53,393 INFO [train.py:968] (0/2) Epoch 7, batch 37850, giga_loss[loss=0.3301, simple_loss=0.3963, pruned_loss=0.132, over 29083.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3709, pruned_loss=0.1203, over 5666768.28 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3836, pruned_loss=0.1346, over 5687813.53 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3659, pruned_loss=0.1166, over 5681454.75 frames. ], batch size: 128, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:19:56,365 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-03 23:20:37,237 INFO [train.py:968] (0/2) Epoch 7, batch 37900, giga_loss[loss=0.2965, simple_loss=0.3546, pruned_loss=0.1192, over 26766.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3678, pruned_loss=0.1177, over 5672492.39 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3836, pruned_loss=0.1346, over 5689074.50 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3638, pruned_loss=0.1147, over 5682840.92 frames. ], batch size: 555, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:20:43,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8218, 3.6057, 3.4135, 2.0048], device='cuda:0'), covar=tensor([0.0525, 0.0678, 0.0694, 0.1813], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0874, 0.0773, 0.0625], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 23:20:46,924 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=311560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:20:49,300 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 23:20:49,770 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=311563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:20:51,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.208e+02 1.071e+03 1.395e+03 1.923e+03 7.721e+03, threshold=2.789e+03, percent-clipped=13.0
+2023-03-03 23:20:54,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=311571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:21:11,480 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=311592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:21:17,748 INFO [train.py:968] (0/2) Epoch 7, batch 37950, giga_loss[loss=0.2653, simple_loss=0.3341, pruned_loss=0.09822, over 27606.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3638, pruned_loss=0.1144, over 5688219.31 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3838, pruned_loss=0.135, over 5696240.07 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3597, pruned_loss=0.1109, over 5689523.69 frames. ], batch size: 472, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:21:59,529 INFO [train.py:968] (0/2) Epoch 7, batch 38000, giga_loss[loss=0.292, simple_loss=0.3715, pruned_loss=0.1063, over 28711.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3623, pruned_loss=0.1128, over 5690280.63 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3841, pruned_loss=0.1352, over 5696076.96 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3583, pruned_loss=0.1093, over 5691175.72 frames. ], batch size: 284, lr: 4.50e-03, grad_scale: 8.0
+2023-03-03 23:22:14,927 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.629e+02 1.120e+03 1.351e+03 1.775e+03 5.752e+03, threshold=2.702e+03, percent-clipped=9.0
+2023-03-03 23:22:36,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2489, 1.5781, 1.2583, 1.5302], device='cuda:0'), covar=tensor([0.0777, 0.0316, 0.0313, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0119, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0049, 0.0044, 0.0075], device='cuda:0')
+2023-03-03 23:22:42,965 INFO [train.py:968] (0/2) Epoch 7, batch 38050, giga_loss[loss=0.2875, simple_loss=0.3632, pruned_loss=0.1059, over 28667.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3621, pruned_loss=0.1121, over 5688931.58 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3843, pruned_loss=0.1353, over 5687190.73 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3587, pruned_loss=0.1091, over 5696665.59 frames. ], batch size: 85, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:22:54,478 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=311714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:22:57,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=311717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:23:04,526 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=311725.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:23:22,387 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=311746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:23:25,399 INFO [train.py:968] (0/2) Epoch 7, batch 38100, giga_loss[loss=0.3115, simple_loss=0.3809, pruned_loss=0.121, over 28938.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3665, pruned_loss=0.1151, over 5677196.97 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3846, pruned_loss=0.1357, over 5672375.69 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3631, pruned_loss=0.1121, over 5695137.53 frames. ], batch size: 186, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:23:37,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.068e+02 1.279e+03 1.607e+03 2.095e+03 4.799e+03, threshold=3.215e+03, percent-clipped=10.0
+2023-03-03 23:23:49,580 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-03 23:24:03,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=311796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:24:05,967 INFO [train.py:968] (0/2) Epoch 7, batch 38150, giga_loss[loss=0.2974, simple_loss=0.3731, pruned_loss=0.1108, over 28874.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3692, pruned_loss=0.1171, over 5682924.18 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3853, pruned_loss=0.1361, over 5673468.00 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3653, pruned_loss=0.1136, over 5696501.10 frames. ], batch size: 186, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:24:52,159 INFO [train.py:968] (0/2) Epoch 7, batch 38200, giga_loss[loss=0.3045, simple_loss=0.3726, pruned_loss=0.1182, over 28796.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3703, pruned_loss=0.1184, over 5682450.05 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.385, pruned_loss=0.1358, over 5678020.39 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3672, pruned_loss=0.1156, over 5689266.65 frames. ], batch size: 119, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:25:05,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.618e+02 1.118e+03 1.402e+03 1.938e+03 4.385e+03, threshold=2.804e+03, percent-clipped=7.0
+2023-03-03 23:25:36,060 INFO [train.py:968] (0/2) Epoch 7, batch 38250, giga_loss[loss=0.2951, simple_loss=0.3682, pruned_loss=0.111, over 28880.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3707, pruned_loss=0.1188, over 5695640.68 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.385, pruned_loss=0.1357, over 5683701.67 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3678, pruned_loss=0.1162, over 5696243.59 frames. ], batch size: 112, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:26:06,099 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=311939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:26:08,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=311942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:26:15,103 INFO [train.py:968] (0/2) Epoch 7, batch 38300, libri_loss[loss=0.3709, simple_loss=0.4213, pruned_loss=0.1602, over 29075.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3716, pruned_loss=0.1201, over 5697337.99 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.386, pruned_loss=0.1366, over 5689747.48 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.368, pruned_loss=0.1166, over 5692497.17 frames. ], batch size: 101, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:26:22,501 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=311960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:26:26,929 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1612, 1.4823, 1.2127, 1.0548], device='cuda:0'), covar=tensor([0.2222, 0.2198, 0.2331, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.0903, 0.1051, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 23:26:28,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.226e+02 1.253e+03 1.491e+03 3.001e+03 8.360e+03, threshold=2.981e+03, percent-clipped=25.0
+2023-03-03 23:26:32,660 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=311971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:26:56,385 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-312000.pt
+2023-03-03 23:26:56,699 INFO [train.py:968] (0/2) Epoch 7, batch 38350, giga_loss[loss=0.2766, simple_loss=0.3624, pruned_loss=0.09542, over 29000.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3717, pruned_loss=0.1196, over 5699849.95 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3862, pruned_loss=0.1366, over 5692767.25 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3682, pruned_loss=0.1164, over 5693532.32 frames. ], batch size: 155, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:27:35,094 INFO [train.py:968] (0/2) Epoch 7, batch 38400, libri_loss[loss=0.3722, simple_loss=0.4218, pruned_loss=0.1613, over 25930.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3721, pruned_loss=0.1193, over 5700245.83 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3862, pruned_loss=0.137, over 5689095.60 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3685, pruned_loss=0.1154, over 5699524.77 frames. ], batch size: 136, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:27:48,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.816e+02 1.146e+03 1.676e+03 2.697e+03 7.547e+03, threshold=3.351e+03, percent-clipped=19.0
+2023-03-03 23:27:57,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=312078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:28:15,178 INFO [train.py:968] (0/2) Epoch 7, batch 38450, giga_loss[loss=0.3213, simple_loss=0.3901, pruned_loss=0.1263, over 28744.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3714, pruned_loss=0.1177, over 5707237.68 frames. ], libri_tot_loss[loss=0.3293, simple_loss=0.3857, pruned_loss=0.1365, over 5694966.34 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3686, pruned_loss=0.1146, over 5701900.91 frames. ], batch size: 284, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:28:15,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=312100.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:28:31,350 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=312119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:28:32,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9269, 4.8932, 1.9482, 1.9142], device='cuda:0'), covar=tensor([0.0709, 0.0308, 0.0717, 0.1046], device='cuda:0'), in_proj_covar=tensor([0.0337, 0.0479, 0.0309, 0.0372], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 23:28:58,075 INFO [train.py:968] (0/2) Epoch 7, batch 38500, giga_loss[loss=0.2676, simple_loss=0.3357, pruned_loss=0.09978, over 28886.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3693, pruned_loss=0.1163, over 5706302.16 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3858, pruned_loss=0.1366, over 5698782.06 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3666, pruned_loss=0.1133, over 5698885.79 frames. ], batch size: 186, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:29:12,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.114e+02 9.734e+02 1.176e+03 1.724e+03 5.923e+03, threshold=2.351e+03, percent-clipped=5.0
+2023-03-03 23:29:40,921 INFO [train.py:968] (0/2) Epoch 7, batch 38550, giga_loss[loss=0.2751, simple_loss=0.3423, pruned_loss=0.1039, over 27946.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3674, pruned_loss=0.1152, over 5703316.23 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3858, pruned_loss=0.1365, over 5697415.66 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.365, pruned_loss=0.1127, over 5698927.94 frames. ], batch size: 412, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:30:16,584 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=312243.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:30:18,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=312246.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 23:30:19,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=312247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:30:20,890 INFO [train.py:968] (0/2) Epoch 7, batch 38600, giga_loss[loss=0.2573, simple_loss=0.3315, pruned_loss=0.0915, over 28258.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3647, pruned_loss=0.1135, over 5707644.03 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3859, pruned_loss=0.1365, over 5700770.96 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3625, pruned_loss=0.1112, over 5701236.67 frames. ], batch size: 77, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:30:35,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.890e+02 9.820e+02 1.235e+03 1.890e+03 3.413e+03, threshold=2.470e+03, percent-clipped=14.0
+2023-03-03 23:30:40,348 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=312275.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:31:01,691 INFO [train.py:968] (0/2) Epoch 7, batch 38650, giga_loss[loss=0.2958, simple_loss=0.3631, pruned_loss=0.1143, over 28762.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3649, pruned_loss=0.1139, over 5714049.19 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3865, pruned_loss=0.1369, over 5705971.58 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3619, pruned_loss=0.111, over 5704445.64 frames. ], batch size: 92, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:31:05,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6415, 4.2736, 1.6499, 1.7064], device='cuda:0'), covar=tensor([0.0886, 0.0212, 0.0844, 0.1242], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0481, 0.0310, 0.0374], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:0')
+2023-03-03 23:31:31,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=312335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:31:42,379 INFO [train.py:968] (0/2) Epoch 7, batch 38700, giga_loss[loss=0.3085, simple_loss=0.3681, pruned_loss=0.1245, over 28538.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3646, pruned_loss=0.1138, over 5710139.01 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3862, pruned_loss=0.1367, over 5707747.20 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3623, pruned_loss=0.1115, over 5701024.91 frames. ], batch size: 85, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:31:56,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.055e+02 1.038e+03 1.436e+03 2.189e+03 1.163e+04, threshold=2.872e+03, percent-clipped=18.0
+2023-03-03 23:32:08,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5889, 2.3911, 1.5621, 0.7592], device='cuda:0'), covar=tensor([0.4729, 0.1997, 0.2509, 0.4089], device='cuda:0'), in_proj_covar=tensor([0.1452, 0.1361, 0.1427, 0.1188], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 23:32:21,461 INFO [train.py:968] (0/2) Epoch 7, batch 38750, giga_loss[loss=0.2802, simple_loss=0.3621, pruned_loss=0.09919, over 28554.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3647, pruned_loss=0.1131, over 5713447.21 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3864, pruned_loss=0.1367, over 5710143.10 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3622, pruned_loss=0.1106, over 5704065.21 frames. ], batch size: 65, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:32:59,967 INFO [train.py:968] (0/2) Epoch 7, batch 38800, giga_loss[loss=0.2541, simple_loss=0.3346, pruned_loss=0.08682, over 28779.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3639, pruned_loss=0.1117, over 5708245.39 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3868, pruned_loss=0.137, over 5701307.03 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3612, pruned_loss=0.1093, over 5709653.78 frames. ], batch size: 60, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:33:03,574 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=312453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:33:13,613 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.771e+02 9.077e+02 1.235e+03 1.720e+03 1.071e+04, threshold=2.470e+03, percent-clipped=6.0
+2023-03-03 23:33:23,418 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=312478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:33:26,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=312481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:33:33,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4471, 1.6370, 1.3746, 1.5696], device='cuda:0'), covar=tensor([0.0799, 0.0308, 0.0324, 0.0862], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0118, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0049, 0.0044, 0.0075], device='cuda:0')
+2023-03-03 23:33:35,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=312494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:33:39,356 INFO [train.py:968] (0/2) Epoch 7, batch 38850, libri_loss[loss=0.2889, simple_loss=0.3378, pruned_loss=0.12, over 27652.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3638, pruned_loss=0.1119, over 5702470.00 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3863, pruned_loss=0.1368, over 5692748.76 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3614, pruned_loss=0.1092, over 5711813.52 frames. ], batch size: 61, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:33:44,048 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2946, 1.6516, 1.5997, 1.2972], device='cuda:0'), covar=tensor([0.1174, 0.1636, 0.0959, 0.1125], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0718, 0.0818, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 23:33:47,725 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=312510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:34:22,365 INFO [train.py:968] (0/2) Epoch 7, batch 38900, giga_loss[loss=0.2438, simple_loss=0.3219, pruned_loss=0.08287, over 28342.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3631, pruned_loss=0.1123, over 5696089.04 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3861, pruned_loss=0.1366, over 5696465.21 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3609, pruned_loss=0.1097, over 5700086.28 frames. ], batch size: 71, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:34:35,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.197e+02 1.237e+03 1.681e+03 2.434e+03 1.374e+04, threshold=3.362e+03, percent-clipped=23.0
+2023-03-03 23:34:59,087 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=312596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:00,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=312599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:02,680 INFO [train.py:968] (0/2) Epoch 7, batch 38950, giga_loss[loss=0.2524, simple_loss=0.3203, pruned_loss=0.09223, over 28557.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3613, pruned_loss=0.1118, over 5700852.38 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3865, pruned_loss=0.1368, over 5699644.76 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3588, pruned_loss=0.1091, over 5701290.11 frames. ], batch size: 85, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:35:20,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=312622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:26,437 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=312628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:32,515 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=312637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:34,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=312640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:42,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3603, 1.5225, 1.2310, 1.0901], device='cuda:0'), covar=tensor([0.1609, 0.1332, 0.1203, 0.1423], device='cuda:0'), in_proj_covar=tensor([0.1546, 0.1388, 0.1362, 0.1474], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 23:35:43,222 INFO [train.py:968] (0/2) Epoch 7, batch 39000, giga_loss[loss=0.2899, simple_loss=0.3629, pruned_loss=0.1085, over 28646.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3569, pruned_loss=0.1089, over 5697318.49 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3868, pruned_loss=0.137, over 5692445.31 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3543, pruned_loss=0.1063, over 5703309.43 frames. ], batch size: 336, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:35:43,226 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-03 23:35:52,208 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2287, simple_loss=0.3346, pruned_loss=0.06143, over 944034.00 frames. 
+2023-03-03 23:35:52,208 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-03 23:36:05,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.476e+02 9.103e+02 1.175e+03 1.443e+03 4.302e+03, threshold=2.351e+03, percent-clipped=3.0
+2023-03-03 23:36:05,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=312669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:36:29,342 INFO [train.py:968] (0/2) Epoch 7, batch 39050, giga_loss[loss=0.2701, simple_loss=0.3412, pruned_loss=0.09949, over 28759.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3569, pruned_loss=0.1093, over 5697120.55 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.3868, pruned_loss=0.1368, over 5691314.44 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3532, pruned_loss=0.1059, over 5702819.24 frames. ], batch size: 119, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:37:11,661 INFO [train.py:968] (0/2) Epoch 7, batch 39100, giga_loss[loss=0.2535, simple_loss=0.3267, pruned_loss=0.09017, over 29006.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3573, pruned_loss=0.1102, over 5693880.96 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3872, pruned_loss=0.1372, over 5680214.78 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3533, pruned_loss=0.1065, over 5709097.50 frames. ], batch size: 213, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:37:25,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=312765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:37:26,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=312768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:37:27,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.757e+02 1.236e+03 1.581e+03 2.526e+03 7.274e+03, threshold=3.162e+03, percent-clipped=27.0
+2023-03-03 23:37:31,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0981, 1.1667, 3.9832, 3.0806], device='cuda:0'), covar=tensor([0.1602, 0.2393, 0.0334, 0.0698], device='cuda:0'), in_proj_covar=tensor([0.0605, 0.0557, 0.0792, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 23:37:52,428 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=312797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:37:54,416 INFO [train.py:968] (0/2) Epoch 7, batch 39150, giga_loss[loss=0.2453, simple_loss=0.317, pruned_loss=0.08679, over 28536.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3545, pruned_loss=0.1092, over 5696092.28 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3871, pruned_loss=0.1371, over 5682471.79 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3511, pruned_loss=0.106, over 5706189.20 frames. ], batch size: 60, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:38:20,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1768, 1.3681, 1.3768, 1.0700], device='cuda:0'), covar=tensor([0.1728, 0.2880, 0.1410, 0.1473], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0716, 0.0816, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-03 23:38:21,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2305, 3.0184, 2.8760, 1.2394], device='cuda:0'), covar=tensor([0.0913, 0.1070, 0.0983, 0.2534], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0879, 0.0778, 0.0621], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 23:38:33,414 INFO [train.py:968] (0/2) Epoch 7, batch 39200, giga_loss[loss=0.2492, simple_loss=0.3226, pruned_loss=0.08792, over 28574.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3535, pruned_loss=0.109, over 5688894.00 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3872, pruned_loss=0.1372, over 5671168.24 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3492, pruned_loss=0.1051, over 5707306.66 frames. ], batch size: 60, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:38:42,169 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-03 23:38:48,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.980e+02 9.982e+02 1.296e+03 1.768e+03 8.544e+03, threshold=2.593e+03, percent-clipped=7.0
+2023-03-03 23:39:14,618 INFO [train.py:968] (0/2) Epoch 7, batch 39250, giga_loss[loss=0.2904, simple_loss=0.3577, pruned_loss=0.1116, over 29081.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3497, pruned_loss=0.1071, over 5695312.68 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3869, pruned_loss=0.1371, over 5672200.39 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3464, pruned_loss=0.104, over 5708807.31 frames. ], batch size: 155, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:39:41,245 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-03 23:39:57,656 INFO [train.py:968] (0/2) Epoch 7, batch 39300, giga_loss[loss=0.2481, simple_loss=0.3279, pruned_loss=0.0842, over 29032.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.35, pruned_loss=0.1078, over 5691228.76 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3869, pruned_loss=0.137, over 5670894.96 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3461, pruned_loss=0.1045, over 5704539.37 frames. ], batch size: 136, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:40:12,127 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=312967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:40:13,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.410e+02 1.025e+03 1.209e+03 1.706e+03 4.794e+03, threshold=2.417e+03, percent-clipped=4.0
+2023-03-03 23:40:38,790 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-03 23:40:41,199 INFO [train.py:968] (0/2) Epoch 7, batch 39350, giga_loss[loss=0.3267, simple_loss=0.3922, pruned_loss=0.1306, over 28737.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3513, pruned_loss=0.1082, over 5698660.57 frames. ], libri_tot_loss[loss=0.3299, simple_loss=0.3865, pruned_loss=0.1367, over 5675550.03 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3479, pruned_loss=0.1053, over 5705396.75 frames. ], batch size: 262, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:41:07,700 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=313031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:41:27,526 INFO [train.py:968] (0/2) Epoch 7, batch 39400, giga_loss[loss=0.2905, simple_loss=0.3533, pruned_loss=0.1139, over 28595.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3538, pruned_loss=0.1087, over 5704132.39 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3864, pruned_loss=0.1366, over 5678076.15 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3509, pruned_loss=0.1062, over 5707466.46 frames. ], batch size: 78, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:41:45,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.241e+02 9.399e+02 1.206e+03 1.692e+03 4.280e+03, threshold=2.413e+03, percent-clipped=9.0
+2023-03-03 23:42:11,337 INFO [train.py:968] (0/2) Epoch 7, batch 39450, giga_loss[loss=0.2641, simple_loss=0.3479, pruned_loss=0.0902, over 28757.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3564, pruned_loss=0.1096, over 5699144.40 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3868, pruned_loss=0.1369, over 5679817.13 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3534, pruned_loss=0.1071, over 5700400.63 frames. ], batch size: 243, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:42:54,795 INFO [train.py:968] (0/2) Epoch 7, batch 39500, giga_loss[loss=0.2877, simple_loss=0.3595, pruned_loss=0.1079, over 28865.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3559, pruned_loss=0.1081, over 5698546.67 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3871, pruned_loss=0.1372, over 5684122.47 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3529, pruned_loss=0.1054, over 5696176.45 frames. ], batch size: 227, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:43:11,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.802e+02 9.639e+02 1.262e+03 1.732e+03 5.331e+03, threshold=2.524e+03, percent-clipped=9.0
+2023-03-03 23:43:21,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3791, 1.0599, 4.4860, 3.4395], device='cuda:0'), covar=tensor([0.1964, 0.3112, 0.0562, 0.0922], device='cuda:0'), in_proj_covar=tensor([0.0596, 0.0551, 0.0786, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 23:43:36,330 INFO [train.py:968] (0/2) Epoch 7, batch 39550, giga_loss[loss=0.2595, simple_loss=0.3416, pruned_loss=0.0887, over 28979.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3556, pruned_loss=0.1077, over 5683837.89 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3867, pruned_loss=0.137, over 5676799.70 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3521, pruned_loss=0.1045, over 5688594.17 frames. ], batch size: 164, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:44:13,953 INFO [train.py:968] (0/2) Epoch 7, batch 39600, giga_loss[loss=0.3076, simple_loss=0.3818, pruned_loss=0.1167, over 28837.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.356, pruned_loss=0.1077, over 5698997.43 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3862, pruned_loss=0.1365, over 5684010.66 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3526, pruned_loss=0.1045, over 5696997.61 frames. ], batch size: 145, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:44:31,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.420e+02 1.111e+03 1.546e+03 2.036e+03 5.507e+03, threshold=3.092e+03, percent-clipped=18.0
+2023-03-03 23:44:44,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5954, 1.7619, 1.6792, 1.2777], device='cuda:0'), covar=tensor([0.1785, 0.1233, 0.0990, 0.1506], device='cuda:0'), in_proj_covar=tensor([0.1571, 0.1388, 0.1373, 0.1489], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 23:44:54,419 INFO [train.py:968] (0/2) Epoch 7, batch 39650, giga_loss[loss=0.3152, simple_loss=0.3844, pruned_loss=0.123, over 28593.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3585, pruned_loss=0.1097, over 5699351.97 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3869, pruned_loss=0.1369, over 5689321.85 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3543, pruned_loss=0.1061, over 5693373.74 frames. ], batch size: 307, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:45:30,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=313342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:45:37,759 INFO [train.py:968] (0/2) Epoch 7, batch 39700, giga_loss[loss=0.3235, simple_loss=0.3865, pruned_loss=0.1303, over 27996.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3606, pruned_loss=0.1114, over 5686570.86 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.387, pruned_loss=0.137, over 5682102.58 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3566, pruned_loss=0.1078, over 5688187.02 frames. ], batch size: 412, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:45:52,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.550e+02 1.237e+03 1.559e+03 2.060e+03 5.076e+03, threshold=3.118e+03, percent-clipped=11.0
+2023-03-03 23:46:03,238 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 23:46:10,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=313391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:46:17,833 INFO [train.py:968] (0/2) Epoch 7, batch 39750, giga_loss[loss=0.3098, simple_loss=0.372, pruned_loss=0.1238, over 28620.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3623, pruned_loss=0.1123, over 5695940.79 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3866, pruned_loss=0.1367, over 5680705.64 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3591, pruned_loss=0.1093, over 5698596.51 frames. ], batch size: 92, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:46:21,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8220, 1.7041, 1.6946, 1.5741], device='cuda:0'), covar=tensor([0.1212, 0.2261, 0.1762, 0.1863], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0724, 0.0647, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 23:46:23,168 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=313406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:46:33,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0151, 1.2751, 1.2400, 1.2452], device='cuda:0'), covar=tensor([0.1218, 0.1065, 0.1846, 0.1136], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0725, 0.0649, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-03 23:46:54,304 INFO [train.py:968] (0/2) Epoch 7, batch 39800, giga_loss[loss=0.2872, simple_loss=0.3587, pruned_loss=0.1079, over 28955.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3661, pruned_loss=0.1146, over 5686455.74 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3871, pruned_loss=0.137, over 5665333.46 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.362, pruned_loss=0.1109, over 5703318.73 frames. ], batch size: 145, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:47:10,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.868e+02 1.152e+03 1.718e+03 2.235e+03 9.136e+03, threshold=3.436e+03, percent-clipped=10.0
+2023-03-03 23:47:22,261 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=313485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:47:25,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=313488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:47:35,130 INFO [train.py:968] (0/2) Epoch 7, batch 39850, giga_loss[loss=0.2917, simple_loss=0.3593, pruned_loss=0.1121, over 28969.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3664, pruned_loss=0.1141, over 5695781.60 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3873, pruned_loss=0.1371, over 5667519.23 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3629, pruned_loss=0.1109, over 5707408.16 frames. ], batch size: 213, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:47:52,851 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=313517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:48:20,231 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=313549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:48:20,695 INFO [train.py:968] (0/2) Epoch 7, batch 39900, giga_loss[loss=0.2792, simple_loss=0.3498, pruned_loss=0.1042, over 28753.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3666, pruned_loss=0.1142, over 5701550.10 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3873, pruned_loss=0.1369, over 5671124.45 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3635, pruned_loss=0.1115, over 5707827.74 frames. ], batch size: 119, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:48:22,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=313552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:48:35,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.264e+02 1.292e+03 1.547e+03 2.062e+03 4.731e+03, threshold=3.095e+03, percent-clipped=4.0
+2023-03-03 23:48:42,851 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=313581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:48:50,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4402, 3.5000, 1.4507, 1.5738], device='cuda:0'), covar=tensor([0.0818, 0.0297, 0.0838, 0.1169], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0490, 0.0314, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-03 23:48:57,710 INFO [train.py:968] (0/2) Epoch 7, batch 39950, giga_loss[loss=0.2507, simple_loss=0.3304, pruned_loss=0.0855, over 28948.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3658, pruned_loss=0.1136, over 5714674.87 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3872, pruned_loss=0.1366, over 5678123.42 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3629, pruned_loss=0.1112, over 5714313.43 frames. ], batch size: 136, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:49:07,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7245, 1.8782, 1.5456, 1.5164], device='cuda:0'), covar=tensor([0.1358, 0.1023, 0.0970, 0.1025], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1405, 0.1390, 0.1494], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-03 23:49:35,299 INFO [train.py:968] (0/2) Epoch 7, batch 40000, giga_loss[loss=0.2782, simple_loss=0.3552, pruned_loss=0.1006, over 28976.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3642, pruned_loss=0.1129, over 5714668.56 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.387, pruned_loss=0.1363, over 5682496.70 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3616, pruned_loss=0.1106, over 5711491.75 frames. ], batch size: 136, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:49:45,028 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=313662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:49:46,346 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=313664.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 23:49:51,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.435e+02 1.155e+03 1.423e+03 1.787e+03 5.511e+03, threshold=2.845e+03, percent-clipped=8.0
+2023-03-03 23:50:16,591 INFO [train.py:968] (0/2) Epoch 7, batch 40050, giga_loss[loss=0.3039, simple_loss=0.3638, pruned_loss=0.122, over 28403.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3609, pruned_loss=0.1113, over 5712187.22 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3872, pruned_loss=0.1365, over 5684839.09 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3583, pruned_loss=0.109, over 5708253.55 frames. ], batch size: 71, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:50:54,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=313743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:50:59,446 INFO [train.py:968] (0/2) Epoch 7, batch 40100, giga_loss[loss=0.2522, simple_loss=0.3424, pruned_loss=0.08093, over 28882.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.356, pruned_loss=0.1083, over 5714556.53 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3872, pruned_loss=0.1365, over 5684839.09 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3539, pruned_loss=0.1065, over 5711494.90 frames. ], batch size: 174, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:51:14,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=313766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:51:18,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.588e+02 9.358e+02 1.190e+03 1.491e+03 3.790e+03, threshold=2.381e+03, percent-clipped=3.0
+2023-03-03 23:51:18,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-03 23:51:40,874 INFO [train.py:968] (0/2) Epoch 7, batch 40150, giga_loss[loss=0.256, simple_loss=0.3399, pruned_loss=0.08606, over 28998.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3553, pruned_loss=0.1069, over 5714659.02 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3871, pruned_loss=0.1365, over 5686045.27 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3534, pruned_loss=0.1052, over 5711430.74 frames. ], batch size: 136, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:52:21,570 INFO [train.py:968] (0/2) Epoch 7, batch 40200, giga_loss[loss=0.2804, simple_loss=0.3635, pruned_loss=0.09864, over 28940.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3571, pruned_loss=0.1072, over 5695453.02 frames. ], libri_tot_loss[loss=0.3303, simple_loss=0.3872, pruned_loss=0.1368, over 5674494.66 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3547, pruned_loss=0.1048, over 5703470.08 frames. ], batch size: 213, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:52:38,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.227e+02 1.087e+03 1.407e+03 1.948e+03 5.360e+03, threshold=2.814e+03, percent-clipped=13.0
+2023-03-03 23:53:01,836 INFO [train.py:968] (0/2) Epoch 7, batch 40250, giga_loss[loss=0.3277, simple_loss=0.386, pruned_loss=0.1347, over 28713.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3579, pruned_loss=0.1079, over 5699571.38 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3875, pruned_loss=0.1369, over 5675766.22 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3553, pruned_loss=0.1055, over 5705005.94 frames. ], batch size: 262, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:53:09,385 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=313909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:53:11,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=313912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:53:13,498 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 23:53:37,697 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=313941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:53:41,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3979, 1.5203, 1.2506, 1.6212], device='cuda:0'), covar=tensor([0.2127, 0.2139, 0.2338, 0.2129], device='cuda:0'), in_proj_covar=tensor([0.1184, 0.0887, 0.1042, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-03 23:53:44,889 INFO [train.py:968] (0/2) Epoch 7, batch 40300, giga_loss[loss=0.2652, simple_loss=0.3408, pruned_loss=0.09477, over 29021.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3558, pruned_loss=0.1077, over 5706891.38 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3874, pruned_loss=0.1369, over 5676939.39 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3537, pruned_loss=0.1057, over 5710308.10 frames. ], batch size: 213, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:54:00,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.876e+02 1.138e+03 1.368e+03 1.765e+03 4.895e+03, threshold=2.736e+03, percent-clipped=10.0
+2023-03-03 23:54:25,883 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-314000.pt
+2023-03-03 23:54:26,215 INFO [train.py:968] (0/2) Epoch 7, batch 40350, libri_loss[loss=0.33, simple_loss=0.3944, pruned_loss=0.1328, over 29660.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3554, pruned_loss=0.1091, over 5717605.37 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3875, pruned_loss=0.1369, over 5682506.08 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3527, pruned_loss=0.1066, over 5716210.46 frames. ], batch size: 91, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:54:55,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:54:56,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314039.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 23:55:04,533 INFO [train.py:968] (0/2) Epoch 7, batch 40400, giga_loss[loss=0.2711, simple_loss=0.3322, pruned_loss=0.105, over 28387.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3549, pruned_loss=0.1106, over 5707110.39 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3877, pruned_loss=0.1372, over 5679212.74 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3518, pruned_loss=0.1077, over 5709503.57 frames. ], batch size: 78, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:55:23,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.846e+02 1.152e+03 1.374e+03 1.994e+03 6.096e+03, threshold=2.748e+03, percent-clipped=8.0
+2023-03-03 23:55:47,616 INFO [train.py:968] (0/2) Epoch 7, batch 40450, giga_loss[loss=0.3517, simple_loss=0.3937, pruned_loss=0.1549, over 26793.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3552, pruned_loss=0.112, over 5700948.27 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3874, pruned_loss=0.137, over 5681457.73 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3527, pruned_loss=0.1097, over 5701027.88 frames. ], batch size: 555, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:55:52,982 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-03 23:56:03,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:56:20,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2185, 1.7816, 1.4091, 0.3950], device='cuda:0'), covar=tensor([0.2470, 0.1562, 0.2701, 0.3439], device='cuda:0'), in_proj_covar=tensor([0.1451, 0.1358, 0.1424, 0.1192], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-03 23:56:28,073 INFO [train.py:968] (0/2) Epoch 7, batch 40500, giga_loss[loss=0.2329, simple_loss=0.3082, pruned_loss=0.07886, over 28885.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3519, pruned_loss=0.11, over 5709436.57 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3875, pruned_loss=0.1372, over 5687053.40 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.349, pruned_loss=0.1073, over 5704783.59 frames. ], batch size: 174, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:56:45,802 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.231e+02 1.111e+03 1.402e+03 2.051e+03 6.236e+03, threshold=2.804e+03, percent-clipped=10.0
+2023-03-03 23:56:52,894 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:56:54,937 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314182.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 23:56:55,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:56:56,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314185.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:56:57,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=314187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:57:08,968 INFO [train.py:968] (0/2) Epoch 7, batch 40550, libri_loss[loss=0.281, simple_loss=0.3481, pruned_loss=0.107, over 29567.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3485, pruned_loss=0.1082, over 5701359.94 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3875, pruned_loss=0.137, over 5681679.90 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3453, pruned_loss=0.1056, over 5703170.62 frames. ], batch size: 77, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:57:19,366 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:57:20,772 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314214.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:57:22,725 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=314217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:57:24,875 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8569, 4.6371, 4.3635, 1.8965], device='cuda:0'), covar=tensor([0.0347, 0.0475, 0.0564, 0.2160], device='cuda:0'), in_proj_covar=tensor([0.0947, 0.0880, 0.0789, 0.0623], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 23:57:50,335 INFO [train.py:968] (0/2) Epoch 7, batch 40600, libri_loss[loss=0.3301, simple_loss=0.3886, pruned_loss=0.1358, over 29512.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3443, pruned_loss=0.1058, over 5701486.04 frames. ], libri_tot_loss[loss=0.3309, simple_loss=0.3876, pruned_loss=0.1371, over 5674960.33 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3413, pruned_loss=0.1034, over 5709053.06 frames. ], batch size: 81, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:57:54,889 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=314257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:57:57,746 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:57:59,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:58:06,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.136e+02 1.231e+03 1.622e+03 2.244e+03 6.487e+03, threshold=3.245e+03, percent-clipped=12.0
+2023-03-03 23:58:22,194 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:58:29,250 INFO [train.py:968] (0/2) Epoch 7, batch 40650, giga_loss[loss=0.2687, simple_loss=0.3501, pruned_loss=0.09366, over 28907.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3457, pruned_loss=0.1063, over 5712616.76 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.3871, pruned_loss=0.1366, over 5680963.02 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3425, pruned_loss=0.1038, over 5714152.50 frames. ], batch size: 174, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:59:11,818 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 23:59:12,102 INFO [train.py:968] (0/2) Epoch 7, batch 40700, giga_loss[loss=0.2973, simple_loss=0.3584, pruned_loss=0.1181, over 28274.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3495, pruned_loss=0.1077, over 5707856.32 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3868, pruned_loss=0.1364, over 5683333.38 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3469, pruned_loss=0.1057, over 5707070.24 frames. ], batch size: 77, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:59:25,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9684, 1.9910, 1.4364, 1.7121], device='cuda:0'), covar=tensor([0.0565, 0.0467, 0.0816, 0.0796], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0446, 0.0499, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-03 23:59:29,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.603e+02 1.133e+03 1.387e+03 1.788e+03 3.936e+03, threshold=2.774e+03, percent-clipped=5.0
+2023-03-03 23:59:49,927 INFO [train.py:968] (0/2) Epoch 7, batch 40750, giga_loss[loss=0.3497, simple_loss=0.4065, pruned_loss=0.1465, over 27977.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3531, pruned_loss=0.1094, over 5708676.36 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.3873, pruned_loss=0.1368, over 5682044.65 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3491, pruned_loss=0.1063, over 5710451.55 frames. ], batch size: 412, lr: 4.48e-03, grad_scale: 4.0
+2023-03-04 00:00:00,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5431, 1.8839, 1.8542, 1.4277], device='cuda:0'), covar=tensor([0.1478, 0.1871, 0.1181, 0.1372], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0703, 0.0811, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 00:00:29,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1923, 1.3174, 3.1271, 2.9781], device='cuda:0'), covar=tensor([0.1315, 0.2166, 0.0403, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0600, 0.0556, 0.0794, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 00:00:30,318 INFO [train.py:968] (0/2) Epoch 7, batch 40800, giga_loss[loss=0.3317, simple_loss=0.3929, pruned_loss=0.1352, over 28268.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3567, pruned_loss=0.1109, over 5692947.23 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3876, pruned_loss=0.1372, over 5675869.16 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3525, pruned_loss=0.1076, over 5700384.28 frames. ], batch size: 77, lr: 4.48e-03, grad_scale: 8.0
+2023-03-04 00:00:49,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.931e+02 1.082e+03 1.331e+03 1.820e+03 1.145e+04, threshold=2.663e+03, percent-clipped=16.0
+2023-03-04 00:01:11,378 INFO [train.py:968] (0/2) Epoch 7, batch 40850, libri_loss[loss=0.3009, simple_loss=0.363, pruned_loss=0.1194, over 29552.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3592, pruned_loss=0.1117, over 5703246.89 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3877, pruned_loss=0.1372, over 5680451.93 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3553, pruned_loss=0.1086, over 5705449.33 frames. ], batch size: 75, lr: 4.47e-03, grad_scale: 8.0
+2023-03-04 00:01:55,871 INFO [train.py:968] (0/2) Epoch 7, batch 40900, giga_loss[loss=0.2754, simple_loss=0.3503, pruned_loss=0.1003, over 28483.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3634, pruned_loss=0.1149, over 5694879.46 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3885, pruned_loss=0.138, over 5671741.96 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3593, pruned_loss=0.1115, over 5704541.31 frames. ], batch size: 71, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:02:10,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:02:23,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.016e+02 1.315e+03 1.774e+03 2.152e+03 5.153e+03, threshold=3.549e+03, percent-clipped=18.0
+2023-03-04 00:02:41,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:02:44,321 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=314593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:02:50,453 INFO [train.py:968] (0/2) Epoch 7, batch 40950, giga_loss[loss=0.3375, simple_loss=0.401, pruned_loss=0.137, over 28665.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3713, pruned_loss=0.1223, over 5694805.97 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3887, pruned_loss=0.1381, over 5675277.96 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3676, pruned_loss=0.1192, over 5699545.55 frames. ], batch size: 262, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:03:20,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:03:37,973 INFO [train.py:968] (0/2) Epoch 7, batch 41000, giga_loss[loss=0.4097, simple_loss=0.4435, pruned_loss=0.1879, over 27658.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3788, pruned_loss=0.1275, over 5690024.98 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3891, pruned_loss=0.1384, over 5673110.78 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3753, pruned_loss=0.1247, over 5695997.30 frames. ], batch size: 472, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:03:50,975 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0405, 1.2746, 3.6700, 3.0957], device='cuda:0'), covar=tensor([0.1604, 0.2260, 0.0402, 0.0749], device='cuda:0'), in_proj_covar=tensor([0.0600, 0.0554, 0.0792, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 00:03:59,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.692e+02 1.598e+03 2.089e+03 2.833e+03 8.418e+03, threshold=4.177e+03, percent-clipped=15.0
+2023-03-04 00:04:19,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6270, 1.5655, 1.2264, 1.2212], device='cuda:0'), covar=tensor([0.0624, 0.0508, 0.0961, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0446, 0.0502, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 00:04:26,941 INFO [train.py:968] (0/2) Epoch 7, batch 41050, giga_loss[loss=0.3308, simple_loss=0.3931, pruned_loss=0.1342, over 29087.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3845, pruned_loss=0.1327, over 5686646.15 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3889, pruned_loss=0.1383, over 5675295.53 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3819, pruned_loss=0.1304, over 5689677.39 frames. ], batch size: 164, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:04:29,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6647, 1.2533, 4.8811, 3.4882], device='cuda:0'), covar=tensor([0.1480, 0.2467, 0.0322, 0.0694], device='cuda:0'), in_proj_covar=tensor([0.0598, 0.0553, 0.0788, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 00:04:31,343 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:04:34,411 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:04:56,496 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:04:58,561 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:04:59,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:05:10,428 INFO [train.py:968] (0/2) Epoch 7, batch 41100, giga_loss[loss=0.3679, simple_loss=0.4171, pruned_loss=0.1594, over 28911.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3905, pruned_loss=0.1375, over 5691746.67 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3892, pruned_loss=0.1384, over 5679646.87 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3881, pruned_loss=0.1356, over 5690625.63 frames. ], batch size: 227, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:05:26,795 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:05:34,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.955e+02 1.783e+03 2.200e+03 2.972e+03 6.852e+03, threshold=4.400e+03, percent-clipped=9.0
+2023-03-04 00:05:34,539 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:05:36,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:05:39,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7662, 2.1934, 1.4745, 1.4176], device='cuda:0'), covar=tensor([0.1521, 0.1172, 0.1432, 0.1523], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1410, 0.1400, 0.1516], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 00:05:59,212 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-04 00:05:59,513 INFO [train.py:968] (0/2) Epoch 7, batch 41150, giga_loss[loss=0.3358, simple_loss=0.3901, pruned_loss=0.1408, over 28834.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3973, pruned_loss=0.1436, over 5679966.27 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3892, pruned_loss=0.1384, over 5675879.47 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3954, pruned_loss=0.142, over 5682551.41 frames. ], batch size: 99, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:06:07,692 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:06:52,999 INFO [train.py:968] (0/2) Epoch 7, batch 41200, giga_loss[loss=0.3563, simple_loss=0.4095, pruned_loss=0.1516, over 28660.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3994, pruned_loss=0.1464, over 5664423.02 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3889, pruned_loss=0.1381, over 5679107.10 frames. ], giga_tot_loss[loss=0.3448, simple_loss=0.3983, pruned_loss=0.1456, over 5663343.22 frames. ], batch size: 307, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:07:21,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.828e+02 1.435e+03 2.037e+03 2.594e+03 7.195e+03, threshold=4.075e+03, percent-clipped=7.0
+2023-03-04 00:07:49,324 INFO [train.py:968] (0/2) Epoch 7, batch 41250, giga_loss[loss=0.4324, simple_loss=0.4571, pruned_loss=0.2039, over 27568.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.4014, pruned_loss=0.1494, over 5664756.37 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3887, pruned_loss=0.1378, over 5682430.53 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.4009, pruned_loss=0.1491, over 5660691.60 frames. ], batch size: 472, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:08:17,632 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=314929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:08:17,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3873, 2.2164, 1.6426, 0.6629], device='cuda:0'), covar=tensor([0.3605, 0.1706, 0.2666, 0.3822], device='cuda:0'), in_proj_covar=tensor([0.1477, 0.1387, 0.1442, 0.1211], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 00:08:39,198 INFO [train.py:968] (0/2) Epoch 7, batch 41300, giga_loss[loss=0.4546, simple_loss=0.4513, pruned_loss=0.229, over 23404.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4035, pruned_loss=0.1524, over 5642379.33 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.388, pruned_loss=0.1373, over 5686442.90 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4042, pruned_loss=0.1531, over 5634425.19 frames. ], batch size: 705, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:08:55,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9467, 2.0521, 1.9348, 1.9263], device='cuda:0'), covar=tensor([0.1111, 0.1470, 0.1455, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0731, 0.0654, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 00:08:56,014 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:09:02,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.350e+02 1.637e+03 2.008e+03 2.837e+03 8.324e+03, threshold=4.016e+03, percent-clipped=13.0
+2023-03-04 00:09:26,474 INFO [train.py:968] (0/2) Epoch 7, batch 41350, giga_loss[loss=0.4249, simple_loss=0.4371, pruned_loss=0.2064, over 23557.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4063, pruned_loss=0.1549, over 5654825.43 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3875, pruned_loss=0.1369, over 5698435.77 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.4083, pruned_loss=0.1568, over 5634993.61 frames. ], batch size: 705, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:09:40,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7867, 2.7298, 1.8184, 0.8628], device='cuda:0'), covar=tensor([0.4339, 0.2023, 0.2435, 0.4198], device='cuda:0'), in_proj_covar=tensor([0.1463, 0.1377, 0.1432, 0.1201], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 00:10:12,720 INFO [train.py:968] (0/2) Epoch 7, batch 41400, giga_loss[loss=0.3417, simple_loss=0.3903, pruned_loss=0.1465, over 28640.00 frames. ], tot_loss[loss=0.3592, simple_loss=0.4069, pruned_loss=0.1558, over 5633035.39 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.387, pruned_loss=0.1366, over 5686660.64 frames. ], giga_tot_loss[loss=0.3634, simple_loss=0.4099, pruned_loss=0.1584, over 5625584.08 frames. ], batch size: 242, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:10:15,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2553, 1.3800, 1.0717, 1.0999], device='cuda:0'), covar=tensor([0.0801, 0.0793, 0.0641, 0.0780], device='cuda:0'), in_proj_covar=tensor([0.1575, 0.1409, 0.1392, 0.1507], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 00:10:42,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.762e+03 2.216e+03 2.858e+03 7.946e+03, threshold=4.432e+03, percent-clipped=7.0
+2023-03-04 00:11:03,837 INFO [train.py:968] (0/2) Epoch 7, batch 41450, giga_loss[loss=0.3275, simple_loss=0.3779, pruned_loss=0.1385, over 28796.00 frames. ], tot_loss[loss=0.3589, simple_loss=0.406, pruned_loss=0.1559, over 5627270.85 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3869, pruned_loss=0.1366, over 5683560.57 frames. ], giga_tot_loss[loss=0.3634, simple_loss=0.4093, pruned_loss=0.1587, over 5622189.86 frames. ], batch size: 99, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:11:05,420 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 00:11:15,972 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=315111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:11:18,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=315114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:11:45,848 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=315143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:11:54,873 INFO [train.py:968] (0/2) Epoch 7, batch 41500, giga_loss[loss=0.2926, simple_loss=0.3629, pruned_loss=0.1111, over 28901.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.4043, pruned_loss=0.1543, over 5646355.17 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3866, pruned_loss=0.1363, over 5687584.11 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.4074, pruned_loss=0.1573, over 5637824.58 frames. ], batch size: 106, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:12:19,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.400e+02 1.604e+03 2.191e+03 2.746e+03 6.094e+03, threshold=4.381e+03, percent-clipped=2.0
+2023-03-04 00:12:43,686 INFO [train.py:968] (0/2) Epoch 7, batch 41550, giga_loss[loss=0.3376, simple_loss=0.396, pruned_loss=0.1396, over 28941.00 frames. ], tot_loss[loss=0.3535, simple_loss=0.4033, pruned_loss=0.1518, over 5655931.57 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.386, pruned_loss=0.1357, over 5693249.83 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4069, pruned_loss=0.1552, over 5642970.50 frames. ], batch size: 227, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:13:19,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9877, 4.7790, 4.5381, 1.8787], device='cuda:0'), covar=tensor([0.0380, 0.0545, 0.0619, 0.2166], device='cuda:0'), in_proj_covar=tensor([0.0976, 0.0917, 0.0814, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 00:13:38,784 INFO [train.py:968] (0/2) Epoch 7, batch 41600, giga_loss[loss=0.3482, simple_loss=0.4075, pruned_loss=0.1445, over 28642.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.403, pruned_loss=0.1506, over 5667359.76 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3856, pruned_loss=0.1354, over 5698972.18 frames. ], giga_tot_loss[loss=0.3574, simple_loss=0.4068, pruned_loss=0.154, over 5650810.08 frames. ], batch size: 92, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:13:45,231 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-04 00:14:06,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.675e+03 2.344e+03 3.099e+03 7.677e+03, threshold=4.688e+03, percent-clipped=8.0
+2023-03-04 00:14:17,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2312, 1.9624, 1.4036, 1.8990], device='cuda:0'), covar=tensor([0.0639, 0.0697, 0.1016, 0.0938], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0450, 0.0504, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 00:14:26,235 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-04 00:14:30,351 INFO [train.py:968] (0/2) Epoch 7, batch 41650, giga_loss[loss=0.4405, simple_loss=0.4531, pruned_loss=0.2139, over 24048.00 frames. ], tot_loss[loss=0.353, simple_loss=0.4037, pruned_loss=0.1511, over 5657806.82 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3848, pruned_loss=0.1349, over 5704091.85 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.4078, pruned_loss=0.1546, over 5639491.09 frames. ], batch size: 705, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:14:33,412 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-04 00:14:33,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0923, 1.3575, 3.2511, 3.1083], device='cuda:0'), covar=tensor([0.1403, 0.2054, 0.0425, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0607, 0.0555, 0.0798, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 00:14:34,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=315304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:14:35,511 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2734, 1.5790, 1.2627, 1.2341], device='cuda:0'), covar=tensor([0.1988, 0.1862, 0.1947, 0.1745], device='cuda:0'), in_proj_covar=tensor([0.1184, 0.0894, 0.1042, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 00:15:00,350 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:15:23,817 INFO [train.py:968] (0/2) Epoch 7, batch 41700, giga_loss[loss=0.3107, simple_loss=0.3947, pruned_loss=0.1134, over 28446.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4014, pruned_loss=0.1483, over 5657160.07 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3847, pruned_loss=0.1349, over 5706208.89 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4048, pruned_loss=0.1512, over 5640639.04 frames. ], batch size: 60, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:15:29,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:15:50,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.248e+02 1.536e+03 1.997e+03 2.652e+03 5.873e+03, threshold=3.994e+03, percent-clipped=4.0
+2023-03-04 00:16:11,809 INFO [train.py:968] (0/2) Epoch 7, batch 41750, giga_loss[loss=0.3281, simple_loss=0.3772, pruned_loss=0.1395, over 27468.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3982, pruned_loss=0.1446, over 5660796.95 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3847, pruned_loss=0.1348, over 5708492.80 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4015, pruned_loss=0.1473, over 5644044.17 frames. ], batch size: 472, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:16:31,410 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:16:58,431 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=315447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:17:00,044 INFO [train.py:968] (0/2) Epoch 7, batch 41800, giga_loss[loss=0.3199, simple_loss=0.3866, pruned_loss=0.1266, over 28636.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.395, pruned_loss=0.1417, over 5648570.89 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3843, pruned_loss=0.135, over 5685470.24 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3984, pruned_loss=0.1441, over 5653074.20 frames. ], batch size: 242, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:17:00,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=315450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:17:03,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4795, 1.8175, 1.2856, 1.5803], device='cuda:0'), covar=tensor([0.0691, 0.0269, 0.0311, 0.0759], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0119, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 00:17:23,352 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.045e+02 1.626e+03 2.137e+03 3.406e+03 1.030e+04, threshold=4.274e+03, percent-clipped=17.0
+2023-03-04 00:17:26,003 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=315479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:17:47,382 INFO [train.py:968] (0/2) Epoch 7, batch 41850, giga_loss[loss=0.3255, simple_loss=0.3855, pruned_loss=0.1328, over 28748.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3911, pruned_loss=0.1384, over 5657173.52 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3843, pruned_loss=0.1352, over 5692345.08 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3942, pruned_loss=0.1403, over 5653476.28 frames. ], batch size: 85, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:18:34,189 INFO [train.py:968] (0/2) Epoch 7, batch 41900, giga_loss[loss=0.357, simple_loss=0.4105, pruned_loss=0.1517, over 28878.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3887, pruned_loss=0.1374, over 5643850.71 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3835, pruned_loss=0.1345, over 5694701.85 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.392, pruned_loss=0.1395, over 5637993.07 frames. ], batch size: 112, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:18:50,290 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:19:02,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.711e+02 1.628e+03 2.209e+03 3.433e+03 5.995e+03, threshold=4.418e+03, percent-clipped=14.0
+2023-03-04 00:19:13,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5434, 1.7299, 1.8398, 1.4549], device='cuda:0'), covar=tensor([0.1421, 0.1838, 0.1113, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0711, 0.0810, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 00:19:22,584 INFO [train.py:968] (0/2) Epoch 7, batch 41950, giga_loss[loss=0.3223, simple_loss=0.3955, pruned_loss=0.1245, over 28853.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.39, pruned_loss=0.1375, over 5663114.10 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3837, pruned_loss=0.1347, over 5695474.21 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3925, pruned_loss=0.1392, over 5657363.32 frames. ], batch size: 119, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:19:49,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3566, 3.4269, 1.5692, 1.4351], device='cuda:0'), covar=tensor([0.0940, 0.0286, 0.0839, 0.1364], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0494, 0.0318, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 00:19:51,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4140, 1.8812, 1.5656, 1.4674], device='cuda:0'), covar=tensor([0.0746, 0.0267, 0.0285, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0119, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 00:20:12,753 INFO [train.py:968] (0/2) Epoch 7, batch 42000, giga_loss[loss=0.3446, simple_loss=0.3936, pruned_loss=0.1478, over 27585.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3896, pruned_loss=0.137, over 5663210.78 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3839, pruned_loss=0.1347, over 5693317.75 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3917, pruned_loss=0.1384, over 5659447.11 frames. ], batch size: 472, lr: 4.47e-03, grad_scale: 8.0
+2023-03-04 00:20:12,758 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 00:20:20,972 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2225, simple_loss=0.3271, pruned_loss=0.05891, over 944034.00 frames. 
+2023-03-04 00:20:20,973 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 00:20:47,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.355e+02 1.491e+03 1.906e+03 2.783e+03 6.042e+03, threshold=3.812e+03, percent-clipped=2.0
+2023-03-04 00:21:10,925 INFO [train.py:968] (0/2) Epoch 7, batch 42050, giga_loss[loss=0.2359, simple_loss=0.3289, pruned_loss=0.07142, over 28569.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.387, pruned_loss=0.1334, over 5676798.97 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3836, pruned_loss=0.1344, over 5698932.86 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.389, pruned_loss=0.1348, over 5668242.16 frames. ], batch size: 71, lr: 4.47e-03, grad_scale: 8.0
+2023-03-04 00:21:14,426 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=315702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:21:33,106 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:21:44,308 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=315728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:21:59,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9287, 2.0137, 1.4038, 1.6673], device='cuda:0'), covar=tensor([0.0623, 0.0477, 0.0845, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0445, 0.0498, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 00:22:10,249 INFO [train.py:968] (0/2) Epoch 7, batch 42100, giga_loss[loss=0.3257, simple_loss=0.3703, pruned_loss=0.1405, over 23556.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3886, pruned_loss=0.1324, over 5665782.46 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3837, pruned_loss=0.1346, over 5690994.87 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3902, pruned_loss=0.1333, over 5665845.35 frames. ], batch size: 705, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:22:24,014 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:22:32,600 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:22:33,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.626e+03 2.069e+03 2.844e+03 1.126e+04, threshold=4.138e+03, percent-clipped=14.0
+2023-03-04 00:22:53,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=315798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:22:55,205 INFO [train.py:968] (0/2) Epoch 7, batch 42150, giga_loss[loss=0.2679, simple_loss=0.3494, pruned_loss=0.09323, over 28495.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3899, pruned_loss=0.1334, over 5660928.62 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3838, pruned_loss=0.1348, over 5679216.42 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3911, pruned_loss=0.1339, over 5670098.37 frames. ], batch size: 60, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:23:38,400 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=315845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:23:41,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=315848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:23:43,096 INFO [train.py:968] (0/2) Epoch 7, batch 42200, giga_loss[loss=0.3269, simple_loss=0.3883, pruned_loss=0.1327, over 28654.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3919, pruned_loss=0.136, over 5659501.68 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.384, pruned_loss=0.1348, over 5675101.77 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.393, pruned_loss=0.1364, over 5669699.89 frames. ], batch size: 307, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:23:49,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5702, 1.7641, 1.4727, 1.8167], device='cuda:0'), covar=tensor([0.1763, 0.1581, 0.1499, 0.1593], device='cuda:0'), in_proj_covar=tensor([0.1191, 0.0897, 0.1049, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 00:24:01,157 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=315871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:24:04,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=315874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:24:06,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.884e+02 1.722e+03 2.103e+03 2.782e+03 8.108e+03, threshold=4.207e+03, percent-clipped=9.0
+2023-03-04 00:24:06,711 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=315877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:24:27,314 INFO [train.py:968] (0/2) Epoch 7, batch 42250, giga_loss[loss=0.3089, simple_loss=0.3698, pruned_loss=0.124, over 28887.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3907, pruned_loss=0.1359, over 5659266.41 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3841, pruned_loss=0.1349, over 5669288.41 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3915, pruned_loss=0.1362, over 5671830.52 frames. ], batch size: 186, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:24:31,755 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=315903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:25:06,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=315941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:25:06,800 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=315941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:25:09,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=315944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:25:13,283 INFO [train.py:968] (0/2) Epoch 7, batch 42300, giga_loss[loss=0.3047, simple_loss=0.3679, pruned_loss=0.1208, over 28959.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3889, pruned_loss=0.1358, over 5665039.81 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3843, pruned_loss=0.135, over 5675517.42 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3895, pruned_loss=0.1358, over 5669293.65 frames. ], batch size: 213, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:25:34,510 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=315973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:25:37,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.718e+03 2.120e+03 3.141e+03 1.234e+04, threshold=4.241e+03, percent-clipped=12.0
+2023-03-04 00:26:00,582 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-316000.pt
+2023-03-04 00:26:00,933 INFO [train.py:968] (0/2) Epoch 7, batch 42350, giga_loss[loss=0.3235, simple_loss=0.3841, pruned_loss=0.1315, over 28735.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3875, pruned_loss=0.1357, over 5656108.37 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3847, pruned_loss=0.1353, over 5672564.82 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3878, pruned_loss=0.1356, over 5661580.34 frames. ], batch size: 92, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:26:03,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2465, 1.4605, 1.1873, 0.9648], device='cuda:0'), covar=tensor([0.1336, 0.1246, 0.0917, 0.1269], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1417, 0.1392, 0.1508], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 00:26:48,665 INFO [train.py:968] (0/2) Epoch 7, batch 42400, giga_loss[loss=0.2993, simple_loss=0.3738, pruned_loss=0.1124, over 28608.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3866, pruned_loss=0.1337, over 5671106.74 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3846, pruned_loss=0.1352, over 5676636.30 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3869, pruned_loss=0.1336, over 5671681.34 frames. ], batch size: 242, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:27:13,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.794e+02 1.524e+03 2.158e+03 3.617e+03 8.943e+03, threshold=4.316e+03, percent-clipped=11.0
+2023-03-04 00:27:21,665 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:27:24,144 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:27:27,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=316092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:27:34,122 INFO [train.py:968] (0/2) Epoch 7, batch 42450, giga_loss[loss=0.3544, simple_loss=0.4015, pruned_loss=0.1536, over 28634.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3853, pruned_loss=0.1314, over 5678477.40 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3846, pruned_loss=0.1353, over 5680136.67 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3856, pruned_loss=0.1313, over 5675747.55 frames. ], batch size: 307, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:27:34,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5793, 2.3085, 1.6290, 0.6666], device='cuda:0'), covar=tensor([0.2419, 0.1472, 0.2387, 0.3053], device='cuda:0'), in_proj_covar=tensor([0.1474, 0.1398, 0.1438, 0.1205], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 00:27:51,800 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=316116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:27:55,566 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-04 00:28:16,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=316142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:28:24,334 INFO [train.py:968] (0/2) Epoch 7, batch 42500, giga_loss[loss=0.3113, simple_loss=0.3722, pruned_loss=0.1253, over 29017.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3854, pruned_loss=0.1315, over 5687514.34 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3845, pruned_loss=0.1353, over 5684429.53 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3857, pruned_loss=0.1313, over 5681639.56 frames. ], batch size: 213, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:28:24,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=316150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:28:51,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.503e+02 1.427e+03 1.759e+03 2.473e+03 7.122e+03, threshold=3.518e+03, percent-clipped=8.0
+2023-03-04 00:29:10,749 INFO [train.py:968] (0/2) Epoch 7, batch 42550, libri_loss[loss=0.3512, simple_loss=0.4157, pruned_loss=0.1434, over 29257.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3832, pruned_loss=0.1306, over 5686820.67 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3847, pruned_loss=0.1353, over 5688541.95 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3833, pruned_loss=0.1304, over 5678340.31 frames. ], batch size: 97, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:29:25,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2625, 1.2468, 3.9662, 3.1587], device='cuda:0'), covar=tensor([0.1620, 0.2448, 0.0420, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0607, 0.0556, 0.0800, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 00:29:42,604 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:29:45,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:29:56,497 INFO [train.py:968] (0/2) Epoch 7, batch 42600, giga_loss[loss=0.2812, simple_loss=0.3526, pruned_loss=0.1049, over 28928.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3827, pruned_loss=0.131, over 5680011.27 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3845, pruned_loss=0.1352, over 5683979.51 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3829, pruned_loss=0.1308, over 5676693.45 frames. ], batch size: 213, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:30:15,311 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=316267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:30:28,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.214e+02 1.365e+03 1.752e+03 2.311e+03 4.497e+03, threshold=3.505e+03, percent-clipped=6.0
+2023-03-04 00:30:33,269 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:30:35,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:30:41,154 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:30:44,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:30:47,693 INFO [train.py:968] (0/2) Epoch 7, batch 42650, giga_loss[loss=0.288, simple_loss=0.3575, pruned_loss=0.1092, over 28863.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3822, pruned_loss=0.1316, over 5676096.86 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3847, pruned_loss=0.1353, over 5686313.21 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3822, pruned_loss=0.1313, over 5671349.96 frames. ], batch size: 119, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:30:56,355 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=316306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:31:07,680 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=316317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:31:14,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=316325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:31:36,268 INFO [train.py:968] (0/2) Epoch 7, batch 42700, giga_loss[loss=0.2886, simple_loss=0.3599, pruned_loss=0.1087, over 29057.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3815, pruned_loss=0.1319, over 5670440.96 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3849, pruned_loss=0.1354, over 5686575.38 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3812, pruned_loss=0.1315, over 5666026.45 frames. ], batch size: 155, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:32:02,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.947e+02 1.577e+03 2.075e+03 3.336e+03 1.105e+04, threshold=4.150e+03, percent-clipped=20.0
+2023-03-04 00:32:22,870 INFO [train.py:968] (0/2) Epoch 7, batch 42750, libri_loss[loss=0.3365, simple_loss=0.3945, pruned_loss=0.1393, over 29548.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3812, pruned_loss=0.1319, over 5676730.20 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3854, pruned_loss=0.1357, over 5687974.54 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3804, pruned_loss=0.1312, over 5671856.32 frames. ], batch size: 89, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:32:29,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=316407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:32:29,554 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4215, 1.6926, 1.7554, 1.3400], device='cuda:0'), covar=tensor([0.1431, 0.1927, 0.1094, 0.1286], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0715, 0.0811, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 00:33:12,667 INFO [train.py:968] (0/2) Epoch 7, batch 42800, giga_loss[loss=0.3457, simple_loss=0.4018, pruned_loss=0.1448, over 28547.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.382, pruned_loss=0.1328, over 5683855.40 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3853, pruned_loss=0.1356, over 5691849.56 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3814, pruned_loss=0.1323, over 5676346.20 frames. ], batch size: 336, lr: 4.46e-03, grad_scale: 8.0
+2023-03-04 00:33:34,657 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-04 00:33:37,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4670, 4.2197, 3.9786, 1.8717], device='cuda:0'), covar=tensor([0.0577, 0.0915, 0.0962, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.0971, 0.0921, 0.0809, 0.0633], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 00:33:38,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.332e+02 1.624e+03 2.317e+03 3.110e+03 5.088e+03, threshold=4.634e+03, percent-clipped=7.0
+2023-03-04 00:33:57,845 INFO [train.py:968] (0/2) Epoch 7, batch 42850, giga_loss[loss=0.3178, simple_loss=0.3842, pruned_loss=0.1257, over 28864.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3823, pruned_loss=0.1322, over 5689727.78 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3847, pruned_loss=0.1351, over 5697888.70 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3823, pruned_loss=0.1321, over 5678053.25 frames. ], batch size: 199, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:34:18,279 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 00:34:31,525 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-04 00:34:42,926 INFO [train.py:968] (0/2) Epoch 7, batch 42900, giga_loss[loss=0.3346, simple_loss=0.3916, pruned_loss=0.1388, over 28282.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.383, pruned_loss=0.1319, over 5692725.23 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3849, pruned_loss=0.1352, over 5703167.97 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3828, pruned_loss=0.1317, over 5678284.67 frames. ], batch size: 368, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:34:51,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3124, 2.9233, 1.4347, 1.3782], device='cuda:0'), covar=tensor([0.0883, 0.0350, 0.0846, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0499, 0.0318, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 00:35:09,379 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.759e+02 1.663e+03 2.401e+03 3.228e+03 1.948e+04, threshold=4.802e+03, percent-clipped=8.0
+2023-03-04 00:35:20,487 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-04 00:35:22,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1194, 0.9461, 0.8526, 1.3512], device='cuda:0'), covar=tensor([0.0704, 0.0393, 0.0339, 0.0762], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0119, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 00:35:25,840 INFO [train.py:968] (0/2) Epoch 7, batch 42950, giga_loss[loss=0.2842, simple_loss=0.3629, pruned_loss=0.1027, over 28694.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3843, pruned_loss=0.1325, over 5691957.81 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3853, pruned_loss=0.1356, over 5709654.11 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3836, pruned_loss=0.1318, over 5674209.41 frames. ], batch size: 262, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:36:13,707 INFO [train.py:968] (0/2) Epoch 7, batch 43000, giga_loss[loss=0.3972, simple_loss=0.4324, pruned_loss=0.181, over 28802.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3861, pruned_loss=0.1348, over 5685863.45 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3854, pruned_loss=0.1358, over 5715793.60 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3855, pruned_loss=0.134, over 5665538.76 frames. ], batch size: 284, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:36:18,478 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5683, 1.6368, 1.1737, 1.3344], device='cuda:0'), covar=tensor([0.0622, 0.0428, 0.0936, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0441, 0.0498, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 00:36:46,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.350e+02 1.540e+03 2.087e+03 2.926e+03 8.528e+03, threshold=4.175e+03, percent-clipped=5.0
+2023-03-04 00:36:47,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=316681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:37:04,833 INFO [train.py:968] (0/2) Epoch 7, batch 43050, giga_loss[loss=0.3624, simple_loss=0.4126, pruned_loss=0.1561, over 28575.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3863, pruned_loss=0.1356, over 5677545.47 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.385, pruned_loss=0.1356, over 5715342.56 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3862, pruned_loss=0.1352, over 5661250.76 frames. ], batch size: 336, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:37:55,328 INFO [train.py:968] (0/2) Epoch 7, batch 43100, giga_loss[loss=0.3051, simple_loss=0.3663, pruned_loss=0.1219, over 28868.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3887, pruned_loss=0.1389, over 5673050.47 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.385, pruned_loss=0.1355, over 5717534.39 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3888, pruned_loss=0.1387, over 5657136.90 frames. ], batch size: 186, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:38:04,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5243, 1.8494, 1.4077, 1.3350], device='cuda:0'), covar=tensor([0.1212, 0.1002, 0.0760, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.1576, 0.1405, 0.1383, 0.1497], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 00:38:27,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.985e+03 2.502e+03 3.643e+03 1.216e+04, threshold=5.004e+03, percent-clipped=17.0
+2023-03-04 00:38:31,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=316782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:38:47,134 INFO [train.py:968] (0/2) Epoch 7, batch 43150, libri_loss[loss=0.3266, simple_loss=0.3872, pruned_loss=0.133, over 29541.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3901, pruned_loss=0.1414, over 5671474.81 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3848, pruned_loss=0.1354, over 5723633.72 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3904, pruned_loss=0.1415, over 5651569.95 frames. ], batch size: 81, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:39:11,082 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:39:14,145 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:39:14,968 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-04 00:39:36,606 INFO [train.py:968] (0/2) Epoch 7, batch 43200, giga_loss[loss=0.3312, simple_loss=0.3899, pruned_loss=0.1363, over 28224.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3908, pruned_loss=0.1421, over 5677739.94 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.385, pruned_loss=0.1355, over 5726537.70 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3909, pruned_loss=0.1421, over 5658762.58 frames. ], batch size: 368, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:39:41,454 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=316856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:40:00,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6943, 1.6670, 1.3773, 1.9119], device='cuda:0'), covar=tensor([0.2036, 0.2132, 0.2213, 0.2082], device='cuda:0'), in_proj_covar=tensor([0.1195, 0.0899, 0.1050, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 00:40:02,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.309e+02 1.563e+03 2.430e+03 3.241e+03 1.062e+04, threshold=4.861e+03, percent-clipped=7.0
+2023-03-04 00:40:21,136 INFO [train.py:968] (0/2) Epoch 7, batch 43250, giga_loss[loss=0.3873, simple_loss=0.4131, pruned_loss=0.1808, over 23941.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3886, pruned_loss=0.1403, over 5677202.92 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3847, pruned_loss=0.1352, over 5726461.64 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.389, pruned_loss=0.1408, over 5660812.03 frames. ], batch size: 705, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:40:30,504 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=316909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:40:40,928 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-04 00:40:44,985 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:40:47,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:41:07,425 INFO [train.py:968] (0/2) Epoch 7, batch 43300, giga_loss[loss=0.2952, simple_loss=0.3706, pruned_loss=0.1099, over 28895.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3867, pruned_loss=0.1377, over 5683963.62 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3844, pruned_loss=0.1348, over 5730244.80 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3875, pruned_loss=0.1384, over 5666678.51 frames. ], batch size: 145, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:41:12,034 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=316957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:41:32,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.524e+02 1.580e+03 2.041e+03 2.653e+03 5.993e+03, threshold=4.082e+03, percent-clipped=4.0
+2023-03-04 00:41:52,465 INFO [train.py:968] (0/2) Epoch 7, batch 43350, giga_loss[loss=0.3333, simple_loss=0.365, pruned_loss=0.1509, over 23909.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3843, pruned_loss=0.1342, over 5679793.26 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3847, pruned_loss=0.135, over 5720252.70 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3846, pruned_loss=0.1347, over 5673754.45 frames. ], batch size: 705, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:41:54,195 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4059, 1.5445, 1.4845, 1.5018], device='cuda:0'), covar=tensor([0.0897, 0.1128, 0.1327, 0.1017], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0730, 0.0651, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 00:42:19,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2957, 1.2594, 1.0467, 1.0411], device='cuda:0'), covar=tensor([0.0534, 0.0378, 0.0862, 0.0726], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0447, 0.0499, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 00:42:36,433 INFO [train.py:968] (0/2) Epoch 7, batch 43400, giga_loss[loss=0.3267, simple_loss=0.3833, pruned_loss=0.1351, over 28812.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.384, pruned_loss=0.1353, over 5659353.54 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3851, pruned_loss=0.1352, over 5713030.82 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3839, pruned_loss=0.1354, over 5659900.11 frames. ], batch size: 119, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:43:02,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.887e+02 1.603e+03 2.051e+03 2.730e+03 6.332e+03, threshold=4.101e+03, percent-clipped=10.0
+2023-03-04 00:43:24,722 INFO [train.py:968] (0/2) Epoch 7, batch 43450, giga_loss[loss=0.3092, simple_loss=0.3733, pruned_loss=0.1226, over 28699.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3821, pruned_loss=0.1347, over 5667288.75 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3844, pruned_loss=0.1348, over 5715958.63 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3825, pruned_loss=0.1352, over 5664540.32 frames. ], batch size: 242, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:43:26,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5497, 2.0487, 1.5141, 0.7445], device='cuda:0'), covar=tensor([0.1952, 0.1330, 0.2131, 0.2733], device='cuda:0'), in_proj_covar=tensor([0.1450, 0.1392, 0.1425, 0.1195], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 00:43:49,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2274, 1.4084, 3.2913, 3.0330], device='cuda:0'), covar=tensor([0.1288, 0.2086, 0.0434, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0610, 0.0561, 0.0805, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 00:44:12,468 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-04 00:44:13,395 INFO [train.py:968] (0/2) Epoch 7, batch 43500, giga_loss[loss=0.2781, simple_loss=0.3496, pruned_loss=0.1033, over 28847.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3833, pruned_loss=0.1361, over 5655466.17 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3849, pruned_loss=0.1352, over 5708516.70 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3832, pruned_loss=0.1361, over 5659638.10 frames. ], batch size: 119, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:44:29,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5539, 1.3539, 1.6188, 1.3225], device='cuda:0'), covar=tensor([0.1803, 0.2735, 0.1402, 0.1503], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0717, 0.0814, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 00:44:39,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.599e+03 2.101e+03 3.040e+03 8.933e+03, threshold=4.203e+03, percent-clipped=12.0
+2023-03-04 00:45:00,303 INFO [train.py:968] (0/2) Epoch 7, batch 43550, giga_loss[loss=0.3699, simple_loss=0.4304, pruned_loss=0.1547, over 28962.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3874, pruned_loss=0.138, over 5660384.31 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3849, pruned_loss=0.1351, over 5711813.87 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3873, pruned_loss=0.1382, over 5659902.54 frames. ], batch size: 199, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:45:22,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6752, 2.0529, 2.0089, 1.5873], device='cuda:0'), covar=tensor([0.1637, 0.1867, 0.1282, 0.1420], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0712, 0.0812, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 00:45:47,306 INFO [train.py:968] (0/2) Epoch 7, batch 43600, giga_loss[loss=0.2875, simple_loss=0.3781, pruned_loss=0.09839, over 28977.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.391, pruned_loss=0.1378, over 5648696.18 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3853, pruned_loss=0.1355, over 5698444.50 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3907, pruned_loss=0.1376, over 5659286.56 frames. ], batch size: 164, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:45:58,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2646, 1.7327, 1.6053, 1.1644], device='cuda:0'), covar=tensor([0.1582, 0.2057, 0.1298, 0.1473], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0714, 0.0814, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 00:46:15,211 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.860e+02 1.572e+03 2.102e+03 3.086e+03 8.754e+03, threshold=4.205e+03, percent-clipped=10.0
+2023-03-04 00:46:17,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=317284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:46:32,188 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=317295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:46:35,773 INFO [train.py:968] (0/2) Epoch 7, batch 43650, giga_loss[loss=0.357, simple_loss=0.4055, pruned_loss=0.1542, over 28716.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3919, pruned_loss=0.1375, over 5638262.42 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3848, pruned_loss=0.1353, over 5684829.77 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3924, pruned_loss=0.1375, over 5656182.42 frames. ], batch size: 242, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:47:04,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4408, 1.5747, 0.9523, 1.3817], device='cuda:0'), covar=tensor([0.0899, 0.0792, 0.1490, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0444, 0.0498, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 00:47:22,586 INFO [train.py:968] (0/2) Epoch 7, batch 43700, giga_loss[loss=0.2937, simple_loss=0.3553, pruned_loss=0.1161, over 28392.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3937, pruned_loss=0.1389, over 5654848.41 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3844, pruned_loss=0.1352, over 5687636.19 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3946, pruned_loss=0.1392, over 5665758.46 frames. ], batch size: 78, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:47:51,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.950e+02 1.748e+03 2.205e+03 3.333e+03 8.499e+03, threshold=4.409e+03, percent-clipped=16.0
+2023-03-04 00:48:08,180 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=317397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:48:10,135 INFO [train.py:968] (0/2) Epoch 7, batch 43750, giga_loss[loss=0.279, simple_loss=0.3439, pruned_loss=0.107, over 28383.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3949, pruned_loss=0.1403, over 5649132.96 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3842, pruned_loss=0.135, over 5679148.70 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.396, pruned_loss=0.1407, over 5665075.37 frames. ], batch size: 77, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:48:35,600 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=317427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:48:37,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=317430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:48:53,486 INFO [train.py:968] (0/2) Epoch 7, batch 43800, giga_loss[loss=0.3004, simple_loss=0.3687, pruned_loss=0.1161, over 29036.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3937, pruned_loss=0.1397, over 5659048.05 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3843, pruned_loss=0.135, over 5681234.89 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3947, pruned_loss=0.1401, over 5669625.18 frames. ], batch size: 136, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:49:02,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=317459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:49:22,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.866e+02 1.602e+03 1.983e+03 2.667e+03 4.696e+03, threshold=3.967e+03, percent-clipped=1.0
+2023-03-04 00:49:39,692 INFO [train.py:968] (0/2) Epoch 7, batch 43850, libri_loss[loss=0.307, simple_loss=0.3659, pruned_loss=0.1241, over 29572.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3933, pruned_loss=0.1405, over 5640579.85 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3846, pruned_loss=0.1353, over 5667828.22 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3942, pruned_loss=0.1408, over 5660966.31 frames. ], batch size: 75, lr: 4.45e-03, grad_scale: 2.0
+2023-03-04 00:50:26,805 INFO [train.py:968] (0/2) Epoch 7, batch 43900, giga_loss[loss=0.3708, simple_loss=0.4245, pruned_loss=0.1586, over 28460.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3918, pruned_loss=0.1405, over 5639303.81 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.385, pruned_loss=0.1356, over 5660311.10 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3922, pruned_loss=0.1406, over 5661956.65 frames. ], batch size: 78, lr: 4.45e-03, grad_scale: 2.0
+2023-03-04 00:50:57,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.675e+03 2.251e+03 4.345e+03 1.620e+04, threshold=4.501e+03, percent-clipped=30.0
+2023-03-04 00:51:05,226 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-04 00:51:12,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3917, 2.0771, 1.4679, 0.5743], device='cuda:0'), covar=tensor([0.3338, 0.1667, 0.2415, 0.3695], device='cuda:0'), in_proj_covar=tensor([0.1460, 0.1394, 0.1432, 0.1194], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 00:51:15,523 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-04 00:51:15,657 INFO [train.py:968] (0/2) Epoch 7, batch 43950, giga_loss[loss=0.4592, simple_loss=0.4505, pruned_loss=0.234, over 23437.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3915, pruned_loss=0.1411, over 5643418.04 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3855, pruned_loss=0.1359, over 5660202.91 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3915, pruned_loss=0.1409, over 5661252.01 frames. ], batch size: 705, lr: 4.45e-03, grad_scale: 2.0
+2023-03-04 00:52:02,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=317643.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 00:52:08,649 INFO [train.py:968] (0/2) Epoch 7, batch 44000, giga_loss[loss=0.4336, simple_loss=0.4552, pruned_loss=0.206, over 27909.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3923, pruned_loss=0.1426, over 5617954.41 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3859, pruned_loss=0.1364, over 5644138.41 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.392, pruned_loss=0.1421, over 5645801.78 frames. ], batch size: 412, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:52:30,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=317670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:52:40,701 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.054e+02 1.650e+03 2.066e+03 2.536e+03 5.257e+03, threshold=4.133e+03, percent-clipped=4.0
+2023-03-04 00:52:49,860 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=317691.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 00:52:59,744 INFO [train.py:968] (0/2) Epoch 7, batch 44050, giga_loss[loss=0.3123, simple_loss=0.3763, pruned_loss=0.1241, over 28960.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3895, pruned_loss=0.1409, over 5629343.61 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3858, pruned_loss=0.1361, over 5647224.16 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3895, pruned_loss=0.1409, over 5648388.56 frames. ], batch size: 145, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:53:41,764 INFO [train.py:968] (0/2) Epoch 7, batch 44100, giga_loss[loss=0.3474, simple_loss=0.3978, pruned_loss=0.1485, over 28564.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3881, pruned_loss=0.1398, over 5642102.83 frames. ], libri_tot_loss[loss=0.3291, simple_loss=0.386, pruned_loss=0.1361, over 5645975.89 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3881, pruned_loss=0.1399, over 5658757.67 frames. ], batch size: 307, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:53:47,558 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-04 00:54:01,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=317772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:54:09,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.421e+02 1.507e+03 1.934e+03 2.486e+03 4.754e+03, threshold=3.869e+03, percent-clipped=1.0
+2023-03-04 00:54:27,142 INFO [train.py:968] (0/2) Epoch 7, batch 44150, giga_loss[loss=0.3512, simple_loss=0.3823, pruned_loss=0.1601, over 23727.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3861, pruned_loss=0.138, over 5644082.24 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3856, pruned_loss=0.1359, over 5652224.01 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3863, pruned_loss=0.1384, over 5651781.73 frames. ], batch size: 705, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:54:37,411 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=317813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:54:40,340 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=317816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:54:51,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5149, 3.2097, 1.5955, 1.5227], device='cuda:0'), covar=tensor([0.0848, 0.0301, 0.0794, 0.1273], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0499, 0.0321, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 00:54:52,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5029, 1.5433, 1.2953, 1.7935], device='cuda:0'), covar=tensor([0.2243, 0.2286, 0.2424, 0.2306], device='cuda:0'), in_proj_covar=tensor([0.1194, 0.0905, 0.1053, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 00:55:10,566 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=317845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:55:14,534 INFO [train.py:968] (0/2) Epoch 7, batch 44200, giga_loss[loss=0.4551, simple_loss=0.4613, pruned_loss=0.2244, over 27615.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3875, pruned_loss=0.1379, over 5650899.06 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3854, pruned_loss=0.1357, over 5657782.33 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3879, pruned_loss=0.1384, over 5651961.60 frames. ], batch size: 472, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:55:22,454 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 00:55:45,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.330e+02 1.510e+03 1.952e+03 2.402e+03 1.316e+04, threshold=3.904e+03, percent-clipped=5.0
+2023-03-04 00:56:03,985 INFO [train.py:968] (0/2) Epoch 7, batch 44250, giga_loss[loss=0.2851, simple_loss=0.3568, pruned_loss=0.1067, over 28649.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3893, pruned_loss=0.1393, over 5650414.19 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3853, pruned_loss=0.1356, over 5660386.35 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3897, pruned_loss=0.1397, over 5648771.24 frames. ], batch size: 242, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:56:17,355 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=317915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:56:21,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=317918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:56:51,210 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=317947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:56:54,387 INFO [train.py:968] (0/2) Epoch 7, batch 44300, giga_loss[loss=0.3194, simple_loss=0.3879, pruned_loss=0.1255, over 28651.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3892, pruned_loss=0.1389, over 5661622.05 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3857, pruned_loss=0.136, over 5662544.74 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3892, pruned_loss=0.139, over 5658308.60 frames. ], batch size: 262, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:57:23,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.814e+02 1.469e+03 1.885e+03 2.769e+03 6.071e+03, threshold=3.770e+03, percent-clipped=7.0
+2023-03-04 00:57:38,745 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-318000.pt
+2023-03-04 00:57:39,064 INFO [train.py:968] (0/2) Epoch 7, batch 44350, giga_loss[loss=0.3072, simple_loss=0.3797, pruned_loss=0.1174, over 28722.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3897, pruned_loss=0.1362, over 5669103.45 frames. ], libri_tot_loss[loss=0.3291, simple_loss=0.3859, pruned_loss=0.1362, over 5666733.46 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3896, pruned_loss=0.136, over 5662585.19 frames. ], batch size: 262, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:57:53,917 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=318018.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 00:58:21,439 INFO [train.py:968] (0/2) Epoch 7, batch 44400, giga_loss[loss=0.3212, simple_loss=0.3913, pruned_loss=0.1256, over 28627.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3914, pruned_loss=0.1355, over 5671809.07 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3857, pruned_loss=0.1361, over 5673193.75 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3915, pruned_loss=0.1355, over 5661017.08 frames. ], batch size: 71, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:58:37,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=318066.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 00:58:56,025 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.341e+02 1.480e+03 1.973e+03 2.769e+03 1.091e+04, threshold=3.946e+03, percent-clipped=15.0
+2023-03-04 00:58:58,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3024, 2.1135, 1.7649, 1.6879], device='cuda:0'), covar=tensor([0.0741, 0.0246, 0.0272, 0.0758], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0118, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 00:59:08,344 INFO [train.py:968] (0/2) Epoch 7, batch 44450, giga_loss[loss=0.3234, simple_loss=0.3922, pruned_loss=0.1274, over 28869.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3936, pruned_loss=0.1374, over 5657860.21 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3853, pruned_loss=0.1359, over 5666072.18 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3943, pruned_loss=0.1376, over 5655710.45 frames. ], batch size: 112, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:59:21,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0119, 3.8355, 3.6277, 1.8645], device='cuda:0'), covar=tensor([0.0543, 0.0747, 0.0733, 0.2003], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.0926, 0.0812, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 00:59:57,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3981, 1.5791, 1.4675, 1.4568], device='cuda:0'), covar=tensor([0.0911, 0.0982, 0.1043, 0.1085], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0730, 0.0649, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 00:59:58,191 INFO [train.py:968] (0/2) Epoch 7, batch 44500, giga_loss[loss=0.4513, simple_loss=0.4535, pruned_loss=0.2245, over 23915.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3982, pruned_loss=0.1421, over 5657032.29 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3853, pruned_loss=0.1359, over 5666152.56 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3989, pruned_loss=0.1423, over 5655112.37 frames. ], batch size: 705, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:00:05,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3992, 3.1881, 3.0987, 1.8204], device='cuda:0'), covar=tensor([0.0663, 0.0833, 0.0763, 0.1800], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.0924, 0.0813, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 01:00:09,662 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=318161.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:00:13,567 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=318164.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:00:31,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.774e+03 2.324e+03 2.996e+03 9.184e+03, threshold=4.648e+03, percent-clipped=15.0
+2023-03-04 01:00:33,615 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=318185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:00:41,532 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=318193.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:00:47,050 INFO [train.py:968] (0/2) Epoch 7, batch 44550, giga_loss[loss=0.4584, simple_loss=0.4627, pruned_loss=0.2271, over 26600.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.4, pruned_loss=0.1445, over 5644156.43 frames. ], libri_tot_loss[loss=0.3291, simple_loss=0.3857, pruned_loss=0.1362, over 5647147.81 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.4004, pruned_loss=0.1445, over 5658587.81 frames. ], batch size: 555, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:00:54,893 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=318209.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:00:56,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=318212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:01:23,736 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=318241.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:01:32,316 INFO [train.py:968] (0/2) Epoch 7, batch 44600, giga_loss[loss=0.2953, simple_loss=0.3674, pruned_loss=0.1117, over 28657.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3985, pruned_loss=0.1437, over 5657007.19 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3862, pruned_loss=0.1366, over 5652938.40 frames. ], giga_tot_loss[loss=0.3429, simple_loss=0.3988, pruned_loss=0.1435, over 5663672.53 frames. ], batch size: 262, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:01:54,497 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=318277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:01:59,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.681e+03 2.284e+03 2.879e+03 5.010e+03, threshold=4.569e+03, percent-clipped=4.0
+2023-03-04 01:02:14,937 INFO [train.py:968] (0/2) Epoch 7, batch 44650, giga_loss[loss=0.3387, simple_loss=0.4116, pruned_loss=0.1329, over 28904.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3968, pruned_loss=0.1418, over 5649512.93 frames. ], libri_tot_loss[loss=0.3299, simple_loss=0.3863, pruned_loss=0.1368, over 5644769.54 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3971, pruned_loss=0.1417, over 5661932.43 frames. ], batch size: 99, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:02:36,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3481, 3.4312, 1.4960, 1.4158], device='cuda:0'), covar=tensor([0.0916, 0.0310, 0.0890, 0.1306], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0497, 0.0319, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 01:02:56,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2823, 1.1178, 0.9474, 1.4273], device='cuda:0'), covar=tensor([0.0781, 0.0352, 0.0363, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0118, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0070, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 01:03:01,003 INFO [train.py:968] (0/2) Epoch 7, batch 44700, libri_loss[loss=0.3264, simple_loss=0.3875, pruned_loss=0.1327, over 29489.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3966, pruned_loss=0.1393, over 5662137.98 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3865, pruned_loss=0.1368, over 5646910.01 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3969, pruned_loss=0.1392, over 5670180.48 frames. ], batch size: 85, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:03:29,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.019e+02 1.427e+03 1.938e+03 2.715e+03 6.027e+03, threshold=3.875e+03, percent-clipped=5.0
+2023-03-04 01:03:46,448 INFO [train.py:968] (0/2) Epoch 7, batch 44750, giga_loss[loss=0.3777, simple_loss=0.4264, pruned_loss=0.1645, over 28563.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3965, pruned_loss=0.1386, over 5677084.96 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3863, pruned_loss=0.1367, over 5652370.17 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3971, pruned_loss=0.1387, over 5679228.48 frames. ], batch size: 307, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:04:36,775 INFO [train.py:968] (0/2) Epoch 7, batch 44800, giga_loss[loss=0.3591, simple_loss=0.4187, pruned_loss=0.1498, over 28956.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.398, pruned_loss=0.1409, over 5654635.48 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.387, pruned_loss=0.1373, over 5647016.60 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.398, pruned_loss=0.1405, over 5661527.37 frames. ], batch size: 164, lr: 4.45e-03, grad_scale: 8.0
+2023-03-04 01:04:52,972 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.28 vs. limit=5.0
+2023-03-04 01:05:06,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.650e+03 2.085e+03 2.906e+03 5.876e+03, threshold=4.170e+03, percent-clipped=9.0
+2023-03-04 01:05:21,198 INFO [train.py:968] (0/2) Epoch 7, batch 44850, giga_loss[loss=0.2999, simple_loss=0.3694, pruned_loss=0.1153, over 28818.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3957, pruned_loss=0.1397, over 5637680.76 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.3864, pruned_loss=0.137, over 5633259.49 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3966, pruned_loss=0.1397, over 5655619.19 frames. ], batch size: 112, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:06:04,562 INFO [train.py:968] (0/2) Epoch 7, batch 44900, giga_loss[loss=0.4445, simple_loss=0.4482, pruned_loss=0.2205, over 26561.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3941, pruned_loss=0.1397, over 5648370.82 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.386, pruned_loss=0.1366, over 5634283.10 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3955, pruned_loss=0.1401, over 5661843.92 frames. ], batch size: 555, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:06:10,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6080, 3.3166, 1.5530, 1.5671], device='cuda:0'), covar=tensor([0.0842, 0.0379, 0.0823, 0.1284], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0497, 0.0320, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 01:06:15,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=318560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:06:26,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3907, 1.8417, 1.3374, 0.6222], device='cuda:0'), covar=tensor([0.2553, 0.1424, 0.1776, 0.3242], device='cuda:0'), in_proj_covar=tensor([0.1456, 0.1383, 0.1421, 0.1193], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 01:06:39,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.719e+02 1.655e+03 2.043e+03 2.772e+03 1.045e+04, threshold=4.086e+03, percent-clipped=6.0
+2023-03-04 01:06:51,509 INFO [train.py:968] (0/2) Epoch 7, batch 44950, giga_loss[loss=0.3101, simple_loss=0.3716, pruned_loss=0.1243, over 29042.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3924, pruned_loss=0.1395, over 5642669.80 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3864, pruned_loss=0.137, over 5628760.63 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3933, pruned_loss=0.1396, over 5659251.22 frames. ], batch size: 128, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:07:40,236 INFO [train.py:968] (0/2) Epoch 7, batch 45000, giga_loss[loss=0.3857, simple_loss=0.4013, pruned_loss=0.1851, over 23293.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.39, pruned_loss=0.1383, over 5620325.64 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3868, pruned_loss=0.1373, over 5602239.75 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3903, pruned_loss=0.1381, over 5657257.04 frames. ], batch size: 705, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:07:40,241 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 01:07:49,543 INFO [train.py:1012] (0/2) Epoch 7, validation: loss=0.2285, simple_loss=0.3357, pruned_loss=0.06063, over 944034.00 frames. 
+2023-03-04 01:07:49,544 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 01:07:51,035 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=318652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:08:07,681 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4943, 1.6281, 1.3523, 1.5067], device='cuda:0'), covar=tensor([0.2165, 0.2240, 0.2277, 0.2236], device='cuda:0'), in_proj_covar=tensor([0.1199, 0.0909, 0.1056, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 01:08:16,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.738e+02 1.459e+03 1.903e+03 2.624e+03 7.953e+03, threshold=3.807e+03, percent-clipped=5.0
+2023-03-04 01:08:25,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3100, 1.4062, 4.1206, 3.1889], device='cuda:0'), covar=tensor([0.1579, 0.2271, 0.0396, 0.0716], device='cuda:0'), in_proj_covar=tensor([0.0609, 0.0562, 0.0814, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 01:08:29,601 INFO [train.py:968] (0/2) Epoch 7, batch 45050, giga_loss[loss=0.3625, simple_loss=0.4096, pruned_loss=0.1577, over 28949.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3902, pruned_loss=0.1396, over 5569617.12 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.388, pruned_loss=0.1383, over 5536245.28 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3896, pruned_loss=0.1387, over 5656874.84 frames. ], batch size: 145, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:08:33,293 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=318703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:08:33,782 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=318704.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:08:35,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=318706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:09:02,778 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=318735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:09:08,985 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 01:09:11,486 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=318745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:09:15,595 INFO [train.py:968] (0/2) Epoch 7, batch 45100, giga_loss[loss=0.3051, simple_loss=0.3709, pruned_loss=0.1196, over 28868.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3907, pruned_loss=0.1408, over 5525579.59 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3886, pruned_loss=0.1388, over 5487225.89 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3897, pruned_loss=0.1395, over 5639853.14 frames. ], batch size: 186, lr: 4.44e-03, grad_scale: 2.0
+2023-03-04 01:09:26,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=318760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:09:26,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6390, 4.4383, 4.2087, 2.0402], device='cuda:0'), covar=tensor([0.0370, 0.0567, 0.0630, 0.2180], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.0926, 0.0818, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 01:09:43,841 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-04 01:09:46,285 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-7.pt
+2023-03-04 01:10:23,106 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.213e+02 1.503e+03 2.023e+03 3.013e+03 1.078e+04, threshold=4.046e+03, percent-clipped=22.0
+2023-03-04 01:10:32,076 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=318795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:10:34,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=318798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:10:59,339 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=318827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:11:00,346 INFO [train.py:968] (0/2) Epoch 8, batch 50, giga_loss[loss=0.3016, simple_loss=0.3806, pruned_loss=0.1113, over 29018.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.388, pruned_loss=0.1224, over 1263067.15 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3607, pruned_loss=0.1028, over 226499.00 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3935, pruned_loss=0.1263, over 1079198.70 frames. ], batch size: 136, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:11:50,662 INFO [train.py:968] (0/2) Epoch 8, batch 100, giga_loss[loss=0.2885, simple_loss=0.3588, pruned_loss=0.1091, over 28868.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3791, pruned_loss=0.1166, over 2242383.43 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3628, pruned_loss=0.1038, over 361724.17 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3817, pruned_loss=0.1186, over 2008358.21 frames. ], batch size: 99, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:11:54,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.139e+02 1.177e+03 1.418e+03 1.853e+03 3.948e+03, threshold=2.836e+03, percent-clipped=0.0
+2023-03-04 01:12:25,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8340, 3.6087, 3.4041, 1.9027], device='cuda:0'), covar=tensor([0.0609, 0.0854, 0.0828, 0.2039], device='cuda:0'), in_proj_covar=tensor([0.0960, 0.0914, 0.0807, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 01:12:31,825 INFO [train.py:968] (0/2) Epoch 8, batch 150, giga_loss[loss=0.2401, simple_loss=0.3097, pruned_loss=0.08527, over 28787.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3654, pruned_loss=0.11, over 3006936.41 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3653, pruned_loss=0.1066, over 573669.32 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.366, pruned_loss=0.1108, over 2705374.87 frames. ], batch size: 99, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:13:15,579 INFO [train.py:968] (0/2) Epoch 8, batch 200, libri_loss[loss=0.2728, simple_loss=0.3435, pruned_loss=0.101, over 29561.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3518, pruned_loss=0.1034, over 3613334.91 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3613, pruned_loss=0.1049, over 757451.56 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3514, pruned_loss=0.1037, over 3289052.09 frames. ], batch size: 76, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:13:19,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3125, 5.0587, 4.7888, 2.0103], device='cuda:0'), covar=tensor([0.0290, 0.0508, 0.0560, 0.2109], device='cuda:0'), in_proj_covar=tensor([0.0963, 0.0917, 0.0811, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 01:13:19,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.549e+02 9.861e+02 1.259e+03 1.572e+03 3.271e+03, threshold=2.517e+03, percent-clipped=3.0
+2023-03-04 01:13:41,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3745, 1.6761, 1.2534, 1.5612], device='cuda:0'), covar=tensor([0.0740, 0.0328, 0.0333, 0.0782], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0075], device='cuda:0')
+2023-03-04 01:13:54,310 INFO [train.py:968] (0/2) Epoch 8, batch 250, giga_loss[loss=0.2142, simple_loss=0.2886, pruned_loss=0.0699, over 29001.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3406, pruned_loss=0.09758, over 4080770.61 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3571, pruned_loss=0.1021, over 934209.31 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3397, pruned_loss=0.09776, over 3755801.97 frames. ], batch size: 136, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:14:15,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-04 01:14:38,967 INFO [train.py:968] (0/2) Epoch 8, batch 300, libri_loss[loss=0.2612, simple_loss=0.3469, pruned_loss=0.08774, over 29525.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3308, pruned_loss=0.09331, over 4443625.60 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3566, pruned_loss=0.1018, over 1008253.51 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3293, pruned_loss=0.09311, over 4165108.54 frames. ], batch size: 82, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:14:39,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319079.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:14:44,482 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.417e+02 9.108e+02 1.198e+03 1.648e+03 2.794e+03, threshold=2.396e+03, percent-clipped=2.0
+2023-03-04 01:15:16,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:15:20,971 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=319125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:15:24,947 INFO [train.py:968] (0/2) Epoch 8, batch 350, giga_loss[loss=0.2191, simple_loss=0.2934, pruned_loss=0.07239, over 28966.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3228, pruned_loss=0.08957, over 4726197.57 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3562, pruned_loss=0.102, over 1105676.53 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3206, pruned_loss=0.08892, over 4481267.41 frames. ], batch size: 213, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:15:30,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:16:03,269 INFO [train.py:968] (0/2) Epoch 8, batch 400, giga_loss[loss=0.2379, simple_loss=0.3037, pruned_loss=0.08603, over 28615.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3186, pruned_loss=0.08771, over 4946330.40 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3561, pruned_loss=0.102, over 1244865.83 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3156, pruned_loss=0.08669, over 4724199.94 frames. ], batch size: 78, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:16:07,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.353e+02 9.713e+02 1.210e+03 1.687e+03 4.292e+03, threshold=2.420e+03, percent-clipped=9.0
+2023-03-04 01:16:21,978 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1932, 3.1741, 2.0066, 1.6892], device='cuda:0'), covar=tensor([0.1736, 0.0749, 0.1126, 0.1427], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1405, 0.1401, 0.1487], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 01:16:39,023 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=319222.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:16:42,567 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=319225.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:16:45,249 INFO [train.py:968] (0/2) Epoch 8, batch 450, libri_loss[loss=0.2769, simple_loss=0.356, pruned_loss=0.09889, over 29527.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3155, pruned_loss=0.08592, over 5120606.70 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.356, pruned_loss=0.1012, over 1337161.74 frames. ], giga_tot_loss[loss=0.2411, simple_loss=0.3122, pruned_loss=0.08494, over 4928735.99 frames. ], batch size: 82, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:17:09,176 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=319254.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:17:15,370 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=319263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:17:17,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=319266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:17:28,208 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=319278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:17:28,549 INFO [train.py:968] (0/2) Epoch 8, batch 500, giga_loss[loss=0.2163, simple_loss=0.2965, pruned_loss=0.06802, over 28793.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3128, pruned_loss=0.0845, over 5255422.91 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3584, pruned_loss=0.1034, over 1445919.99 frames. ], giga_tot_loss[loss=0.2369, simple_loss=0.3085, pruned_loss=0.08268, over 5090031.95 frames. ], batch size: 145, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:17:31,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8607, 1.7304, 1.8421, 1.6408], device='cuda:0'), covar=tensor([0.1295, 0.1976, 0.1741, 0.1834], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0729, 0.0648, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 01:17:31,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=319281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:17:33,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.429e+02 9.342e+02 1.299e+03 1.752e+03 7.518e+03, threshold=2.598e+03, percent-clipped=10.0
+2023-03-04 01:17:44,131 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=319295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:17:55,499 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=319310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:18:12,614 INFO [train.py:968] (0/2) Epoch 8, batch 550, giga_loss[loss=0.2591, simple_loss=0.3356, pruned_loss=0.09126, over 28267.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3109, pruned_loss=0.08335, over 5355945.38 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3563, pruned_loss=0.1018, over 1578113.71 frames. ], giga_tot_loss[loss=0.2351, simple_loss=0.3066, pruned_loss=0.08176, over 5207279.19 frames. ], batch size: 368, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:18:20,929 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2043, 1.3603, 3.8603, 3.1754], device='cuda:0'), covar=tensor([0.1513, 0.2323, 0.0381, 0.0816], device='cuda:0'), in_proj_covar=tensor([0.0600, 0.0554, 0.0805, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 01:18:25,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4808, 4.2514, 3.9869, 2.0463], device='cuda:0'), covar=tensor([0.0383, 0.0586, 0.0617, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.0946, 0.0902, 0.0796, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 01:19:00,179 INFO [train.py:968] (0/2) Epoch 8, batch 600, giga_loss[loss=0.1954, simple_loss=0.2631, pruned_loss=0.06384, over 28541.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3086, pruned_loss=0.08278, over 5430019.45 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3559, pruned_loss=0.1013, over 1620374.70 frames. ], giga_tot_loss[loss=0.2339, simple_loss=0.3048, pruned_loss=0.08148, over 5309325.34 frames. ], batch size: 85, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:19:04,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.005e+02 1.052e+03 1.346e+03 2.108e+03 4.559e+03, threshold=2.693e+03, percent-clipped=17.0
+2023-03-04 01:19:49,811 INFO [train.py:968] (0/2) Epoch 8, batch 650, giga_loss[loss=0.2867, simple_loss=0.3443, pruned_loss=0.1146, over 28285.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3062, pruned_loss=0.08183, over 5479869.54 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3574, pruned_loss=0.1025, over 1662938.47 frames. ], giga_tot_loss[loss=0.2315, simple_loss=0.3024, pruned_loss=0.0803, over 5381250.19 frames. ], batch size: 369, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:20:23,454 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=319469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:20:33,682 INFO [train.py:968] (0/2) Epoch 8, batch 700, giga_loss[loss=0.1885, simple_loss=0.266, pruned_loss=0.0555, over 28059.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3039, pruned_loss=0.08051, over 5523168.68 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3575, pruned_loss=0.1031, over 1799746.70 frames. ], giga_tot_loss[loss=0.228, simple_loss=0.2991, pruned_loss=0.07841, over 5439956.85 frames. ], batch size: 77, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:20:38,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.755e+02 1.008e+03 1.284e+03 1.782e+03 5.530e+03, threshold=2.569e+03, percent-clipped=7.0
+2023-03-04 01:20:50,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:21:10,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1188, 2.1620, 1.9923, 2.0370], device='cuda:0'), covar=tensor([0.1411, 0.2132, 0.1801, 0.1857], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0727, 0.0647, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 01:21:15,880 INFO [train.py:968] (0/2) Epoch 8, batch 750, giga_loss[loss=0.2015, simple_loss=0.2782, pruned_loss=0.06242, over 28904.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3016, pruned_loss=0.07895, over 5575113.09 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3581, pruned_loss=0.1036, over 1901597.37 frames. ], giga_tot_loss[loss=0.2247, simple_loss=0.2963, pruned_loss=0.07657, over 5500644.07 frames. ], batch size: 145, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:21:36,278 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=319551.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:21:39,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4607, 1.5903, 1.3440, 1.7403], device='cuda:0'), covar=tensor([0.2508, 0.2315, 0.2552, 0.2089], device='cuda:0'), in_proj_covar=tensor([0.1206, 0.0906, 0.1059, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 01:21:40,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4208, 2.0766, 1.6500, 1.8401], device='cuda:0'), covar=tensor([0.0748, 0.0265, 0.0292, 0.0754], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 01:21:59,322 INFO [train.py:968] (0/2) Epoch 8, batch 800, giga_loss[loss=0.293, simple_loss=0.3505, pruned_loss=0.1177, over 28284.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3009, pruned_loss=0.07887, over 5605274.65 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3583, pruned_loss=0.1034, over 2037678.14 frames. ], giga_tot_loss[loss=0.2235, simple_loss=0.2947, pruned_loss=0.07617, over 5536991.45 frames. ], batch size: 368, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:22:03,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.256e+02 9.969e+02 1.349e+03 1.787e+03 3.567e+03, threshold=2.698e+03, percent-clipped=6.0
+2023-03-04 01:22:44,316 INFO [train.py:968] (0/2) Epoch 8, batch 850, giga_loss[loss=0.231, simple_loss=0.3062, pruned_loss=0.07797, over 28552.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3097, pruned_loss=0.08452, over 5609503.33 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.359, pruned_loss=0.104, over 2159723.53 frames. ], giga_tot_loss[loss=0.233, simple_loss=0.3029, pruned_loss=0.08153, over 5556108.00 frames. ], batch size: 60, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:22:51,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5848, 2.0227, 1.6758, 1.7753], device='cuda:0'), covar=tensor([0.0719, 0.0268, 0.0295, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0117, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0075], device='cuda:0')
+2023-03-04 01:22:59,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=319643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:23:01,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=319646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:23:14,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4851, 1.5783, 0.9948, 1.3136], device='cuda:0'), covar=tensor([0.0882, 0.0800, 0.1500, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0441, 0.0498, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 01:23:29,262 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=319675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:23:33,262 INFO [train.py:968] (0/2) Epoch 8, batch 900, giga_loss[loss=0.2656, simple_loss=0.3468, pruned_loss=0.09216, over 28863.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3229, pruned_loss=0.09165, over 5627003.29 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3587, pruned_loss=0.1037, over 2233357.63 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3169, pruned_loss=0.08912, over 5580179.14 frames. ], batch size: 145, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:23:39,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.533e+02 1.139e+03 1.429e+03 1.805e+03 4.548e+03, threshold=2.857e+03, percent-clipped=4.0
+2023-03-04 01:24:05,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7177, 1.8119, 1.6130, 1.6161], device='cuda:0'), covar=tensor([0.1438, 0.1937, 0.1753, 0.1750], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0724, 0.0643, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 01:24:14,257 INFO [train.py:968] (0/2) Epoch 8, batch 950, giga_loss[loss=0.3221, simple_loss=0.3811, pruned_loss=0.1315, over 28532.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3349, pruned_loss=0.09771, over 5652805.36 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3571, pruned_loss=0.1028, over 2342310.13 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3299, pruned_loss=0.0958, over 5608087.43 frames. ], batch size: 85, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:24:59,273 INFO [train.py:968] (0/2) Epoch 8, batch 1000, giga_loss[loss=0.2923, simple_loss=0.3682, pruned_loss=0.1082, over 28863.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3442, pruned_loss=0.1025, over 5662815.13 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3565, pruned_loss=0.1025, over 2425783.87 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3401, pruned_loss=0.101, over 5625629.61 frames. ], batch size: 186, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:25:05,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.448e+02 1.368e+03 1.836e+03 2.643e+03 5.798e+03, threshold=3.673e+03, percent-clipped=20.0
+2023-03-04 01:25:38,874 INFO [train.py:968] (0/2) Epoch 8, batch 1050, giga_loss[loss=0.2708, simple_loss=0.3477, pruned_loss=0.09691, over 28406.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3483, pruned_loss=0.1031, over 5675325.72 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3567, pruned_loss=0.1026, over 2443126.95 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.345, pruned_loss=0.102, over 5645248.19 frames. ], batch size: 71, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:25:54,580 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:26:17,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4330, 1.7093, 1.7056, 1.2844], device='cuda:0'), covar=tensor([0.1551, 0.2010, 0.1228, 0.1457], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0716, 0.0827, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 01:26:28,183 INFO [train.py:968] (0/2) Epoch 8, batch 1100, giga_loss[loss=0.2507, simple_loss=0.331, pruned_loss=0.08521, over 28647.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3505, pruned_loss=0.1031, over 5666932.96 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3572, pruned_loss=0.1028, over 2505553.78 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3476, pruned_loss=0.1022, over 5644621.70 frames. ], batch size: 92, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:26:29,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5233, 1.6483, 1.3886, 1.8132], device='cuda:0'), covar=tensor([0.2285, 0.2213, 0.2348, 0.2098], device='cuda:0'), in_proj_covar=tensor([0.1207, 0.0912, 0.1063, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 01:26:32,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.315e+02 1.056e+03 1.269e+03 1.537e+03 4.881e+03, threshold=2.537e+03, percent-clipped=2.0
+2023-03-04 01:27:08,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319926.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:27:10,475 INFO [train.py:968] (0/2) Epoch 8, batch 1150, giga_loss[loss=0.292, simple_loss=0.3622, pruned_loss=0.1109, over 28521.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3515, pruned_loss=0.1034, over 5683671.17 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.357, pruned_loss=0.1024, over 2555392.68 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3492, pruned_loss=0.1027, over 5663800.27 frames. ], batch size: 336, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:27:21,179 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 01:27:53,623 INFO [train.py:968] (0/2) Epoch 8, batch 1200, giga_loss[loss=0.3024, simple_loss=0.3698, pruned_loss=0.1175, over 28625.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3533, pruned_loss=0.1047, over 5665307.69 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3577, pruned_loss=0.1027, over 2627775.46 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.351, pruned_loss=0.1041, over 5654491.56 frames. ], batch size: 242, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:27:59,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.678e+02 1.009e+03 1.304e+03 1.778e+03 7.022e+03, threshold=2.608e+03, percent-clipped=12.0
+2023-03-04 01:28:00,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=319987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:28:04,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=319990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:28:11,981 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-320000.pt
+2023-03-04 01:28:29,370 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=320019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:28:36,797 INFO [train.py:968] (0/2) Epoch 8, batch 1250, giga_loss[loss=0.2904, simple_loss=0.3634, pruned_loss=0.1087, over 28939.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3558, pruned_loss=0.1062, over 5673143.25 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3584, pruned_loss=0.1031, over 2692984.47 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3536, pruned_loss=0.1057, over 5660248.51 frames. ], batch size: 145, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:28:53,660 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.24 vs. limit=5.0
+2023-03-04 01:29:12,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=320069.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:29:14,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=320072.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:29:20,662 INFO [train.py:968] (0/2) Epoch 8, batch 1300, giga_loss[loss=0.2951, simple_loss=0.3716, pruned_loss=0.1093, over 28616.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3594, pruned_loss=0.1081, over 5680706.12 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3589, pruned_loss=0.1034, over 2787713.40 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3575, pruned_loss=0.1077, over 5664572.44 frames. ], batch size: 336, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:29:26,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.211e+02 1.146e+03 1.483e+03 1.904e+03 4.273e+03, threshold=2.966e+03, percent-clipped=8.0
+2023-03-04 01:29:34,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4057, 1.6032, 1.2799, 1.5979], device='cuda:0'), covar=tensor([0.0723, 0.0270, 0.0313, 0.0767], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 01:29:38,881 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=320101.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:30:01,686 INFO [train.py:968] (0/2) Epoch 8, batch 1350, giga_loss[loss=0.2645, simple_loss=0.3485, pruned_loss=0.0902, over 28471.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3622, pruned_loss=0.1088, over 5699383.85 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3591, pruned_loss=0.1033, over 2865119.48 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3606, pruned_loss=0.1086, over 5681471.49 frames. ], batch size: 65, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:30:24,812 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=320156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:30:41,774 INFO [train.py:968] (0/2) Epoch 8, batch 1400, giga_loss[loss=0.2816, simple_loss=0.3599, pruned_loss=0.1017, over 28949.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3634, pruned_loss=0.109, over 5687136.48 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3597, pruned_loss=0.1036, over 2942529.60 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3619, pruned_loss=0.1089, over 5678256.36 frames. ], batch size: 174, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:30:47,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.669e+02 1.136e+03 1.448e+03 2.064e+03 8.365e+03, threshold=2.897e+03, percent-clipped=10.0
+2023-03-04 01:31:17,726 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-04 01:31:27,166 INFO [train.py:968] (0/2) Epoch 8, batch 1450, giga_loss[loss=0.2546, simple_loss=0.3454, pruned_loss=0.08193, over 28987.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3625, pruned_loss=0.1073, over 5692079.56 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3595, pruned_loss=0.1034, over 2972104.12 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3615, pruned_loss=0.1073, over 5683341.00 frames. ], batch size: 136, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:31:33,599 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=320237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:32:06,109 INFO [train.py:968] (0/2) Epoch 8, batch 1500, giga_loss[loss=0.2607, simple_loss=0.3438, pruned_loss=0.08875, over 27954.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3609, pruned_loss=0.1051, over 5699783.79 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3588, pruned_loss=0.1031, over 3014784.05 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3605, pruned_loss=0.1052, over 5691075.60 frames. ], batch size: 412, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:32:11,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.644e+02 9.482e+02 1.210e+03 1.450e+03 3.548e+03, threshold=2.420e+03, percent-clipped=3.0
+2023-03-04 01:32:42,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5577, 1.5333, 1.1853, 1.2061], device='cuda:0'), covar=tensor([0.0664, 0.0512, 0.0968, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0435, 0.0493, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 01:32:45,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6808, 2.0890, 2.0359, 1.5983], device='cuda:0'), covar=tensor([0.1556, 0.1712, 0.1132, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0711, 0.0825, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 01:32:46,073 INFO [train.py:968] (0/2) Epoch 8, batch 1550, giga_loss[loss=0.2891, simple_loss=0.3593, pruned_loss=0.1095, over 28868.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3594, pruned_loss=0.1038, over 5704130.41 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3594, pruned_loss=0.1036, over 3100137.57 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3588, pruned_loss=0.1037, over 5691883.50 frames. ], batch size: 186, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:33:28,024 INFO [train.py:968] (0/2) Epoch 8, batch 1600, giga_loss[loss=0.2763, simple_loss=0.3495, pruned_loss=0.1016, over 29140.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3598, pruned_loss=0.1051, over 5708368.85 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3587, pruned_loss=0.1031, over 3213241.87 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3596, pruned_loss=0.1054, over 5700059.99 frames. ], batch size: 155, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:33:34,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.266e+02 1.137e+03 1.412e+03 1.890e+03 6.116e+03, threshold=2.824e+03, percent-clipped=10.0
+2023-03-04 01:34:08,966 INFO [train.py:968] (0/2) Epoch 8, batch 1650, giga_loss[loss=0.3214, simple_loss=0.3782, pruned_loss=0.1323, over 28673.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3636, pruned_loss=0.1104, over 5705482.35 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.359, pruned_loss=0.1033, over 3306742.64 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3635, pruned_loss=0.1107, over 5703338.77 frames. ], batch size: 85, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:34:29,610 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 01:34:30,820 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=320453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:34:50,264 INFO [train.py:968] (0/2) Epoch 8, batch 1700, giga_loss[loss=0.358, simple_loss=0.402, pruned_loss=0.157, over 28700.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3649, pruned_loss=0.1136, over 5697979.61 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3586, pruned_loss=0.1032, over 3381702.23 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3652, pruned_loss=0.1141, over 5692539.42 frames. ], batch size: 284, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:34:57,846 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-04 01:34:58,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.511e+02 1.240e+03 1.608e+03 1.962e+03 6.322e+03, threshold=3.217e+03, percent-clipped=8.0
+2023-03-04 01:35:08,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6279, 2.5430, 1.6194, 0.8706], device='cuda:0'), covar=tensor([0.4796, 0.1986, 0.2671, 0.3935], device='cuda:0'), in_proj_covar=tensor([0.1460, 0.1378, 0.1428, 0.1199], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 01:35:35,349 INFO [train.py:968] (0/2) Epoch 8, batch 1750, giga_loss[loss=0.3087, simple_loss=0.377, pruned_loss=0.1202, over 28932.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3628, pruned_loss=0.113, over 5696743.93 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3578, pruned_loss=0.1028, over 3430687.62 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3635, pruned_loss=0.1138, over 5689595.15 frames. ], batch size: 106, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:35:38,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=320531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:35:59,070 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=320556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:36:18,370 INFO [train.py:968] (0/2) Epoch 8, batch 1800, giga_loss[loss=0.2975, simple_loss=0.3627, pruned_loss=0.1162, over 28869.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.361, pruned_loss=0.1122, over 5699973.03 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3579, pruned_loss=0.1029, over 3481303.40 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3616, pruned_loss=0.113, over 5700145.98 frames. ], batch size: 186, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:36:26,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.212e+02 1.192e+03 1.694e+03 2.416e+03 9.436e+03, threshold=3.387e+03, percent-clipped=12.0
+2023-03-04 01:36:45,421 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=320612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:37:00,384 INFO [train.py:968] (0/2) Epoch 8, batch 1850, giga_loss[loss=0.2746, simple_loss=0.3541, pruned_loss=0.09755, over 28754.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3607, pruned_loss=0.1115, over 5703870.46 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3578, pruned_loss=0.1028, over 3516176.19 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3613, pruned_loss=0.1124, over 5702442.81 frames. ], batch size: 262, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:37:10,308 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4299, 2.0282, 1.4942, 0.7034], device='cuda:0'), covar=tensor([0.2991, 0.1628, 0.2699, 0.3549], device='cuda:0'), in_proj_covar=tensor([0.1451, 0.1372, 0.1420, 0.1194], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 01:37:21,768 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=320654.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:37:39,442 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=320674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:37:43,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=320677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:37:44,547 INFO [train.py:968] (0/2) Epoch 8, batch 1900, giga_loss[loss=0.2737, simple_loss=0.3484, pruned_loss=0.09946, over 28970.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3598, pruned_loss=0.1105, over 5691633.12 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3579, pruned_loss=0.1029, over 3565236.43 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3602, pruned_loss=0.1112, over 5696031.30 frames. ], batch size: 136, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:37:53,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.529e+02 1.032e+03 1.239e+03 1.525e+03 4.940e+03, threshold=2.477e+03, percent-clipped=1.0
+2023-03-04 01:37:54,975 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-04 01:38:12,313 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=320706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:38:32,879 INFO [train.py:968] (0/2) Epoch 8, batch 1950, giga_loss[loss=0.2316, simple_loss=0.3137, pruned_loss=0.07478, over 29002.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3552, pruned_loss=0.1074, over 5688569.58 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3579, pruned_loss=0.1029, over 3565236.43 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3555, pruned_loss=0.1079, over 5691992.74 frames. ], batch size: 164, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:39:01,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=320755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:39:04,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=320758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:39:15,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 01:39:23,644 INFO [train.py:968] (0/2) Epoch 8, batch 2000, giga_loss[loss=0.2417, simple_loss=0.3208, pruned_loss=0.0813, over 28829.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3482, pruned_loss=0.1034, over 5677110.66 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3583, pruned_loss=0.1033, over 3600270.45 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3482, pruned_loss=0.1036, over 5676767.83 frames. ], batch size: 174, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:39:31,554 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=320787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:39:32,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.730e+02 8.923e+02 1.224e+03 1.502e+03 5.178e+03, threshold=2.447e+03, percent-clipped=6.0
+2023-03-04 01:39:35,931 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7821, 2.8363, 1.9713, 0.8825], device='cuda:0'), covar=tensor([0.5002, 0.1616, 0.2406, 0.4195], device='cuda:0'), in_proj_covar=tensor([0.1453, 0.1368, 0.1415, 0.1189], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 01:40:07,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9943, 1.2748, 3.9205, 3.0410], device='cuda:0'), covar=tensor([0.1666, 0.2355, 0.0358, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0596, 0.0555, 0.0791, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 01:40:07,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=320828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:40:08,426 INFO [train.py:968] (0/2) Epoch 8, batch 2050, giga_loss[loss=0.2626, simple_loss=0.3404, pruned_loss=0.09244, over 28941.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3432, pruned_loss=0.1009, over 5673124.95 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3589, pruned_loss=0.1038, over 3655291.70 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3424, pruned_loss=0.1008, over 5669495.82 frames. ], batch size: 164, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:40:55,056 INFO [train.py:968] (0/2) Epoch 8, batch 2100, libri_loss[loss=0.2917, simple_loss=0.3616, pruned_loss=0.1109, over 29586.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3424, pruned_loss=0.1, over 5687279.56 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3597, pruned_loss=0.1042, over 3732922.59 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3407, pruned_loss=0.09952, over 5677045.88 frames. ], batch size: 74, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:41:01,593 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-04 01:41:02,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.800e+02 1.017e+03 1.300e+03 1.837e+03 3.825e+03, threshold=2.600e+03, percent-clipped=6.0
+2023-03-04 01:41:03,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7667, 5.1041, 2.0064, 2.0198], device='cuda:0'), covar=tensor([0.0833, 0.0156, 0.0752, 0.1154], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0483, 0.0316, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 01:41:34,102 INFO [train.py:968] (0/2) Epoch 8, batch 2150, giga_loss[loss=0.2719, simple_loss=0.3453, pruned_loss=0.09923, over 28594.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.343, pruned_loss=0.09974, over 5696254.13 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3592, pruned_loss=0.1036, over 3815041.34 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3413, pruned_loss=0.0995, over 5683452.34 frames. ], batch size: 336, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:41:35,581 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=320931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:42:07,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=320971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:42:09,891 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=320974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:42:13,045 INFO [train.py:968] (0/2) Epoch 8, batch 2200, giga_loss[loss=0.2549, simple_loss=0.3375, pruned_loss=0.08617, over 28968.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3417, pruned_loss=0.09868, over 5702607.12 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3592, pruned_loss=0.1036, over 3856030.49 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3401, pruned_loss=0.09846, over 5689148.33 frames. ], batch size: 155, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:42:21,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.216e+02 9.608e+02 1.234e+03 1.596e+03 1.012e+04, threshold=2.469e+03, percent-clipped=4.0
+2023-03-04 01:42:37,306 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=321003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:42:56,994 INFO [train.py:968] (0/2) Epoch 8, batch 2250, giga_loss[loss=0.2985, simple_loss=0.3659, pruned_loss=0.1156, over 28645.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3391, pruned_loss=0.09759, over 5707326.85 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.359, pruned_loss=0.1033, over 3886400.43 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3376, pruned_loss=0.09749, over 5694300.58 frames. ], batch size: 336, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:42:57,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=321029.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:43:21,105 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:43:35,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=321074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:43:37,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=321077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:43:38,267 INFO [train.py:968] (0/2) Epoch 8, batch 2300, giga_loss[loss=0.2405, simple_loss=0.3191, pruned_loss=0.08094, over 28972.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3373, pruned_loss=0.09697, over 5704606.67 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3594, pruned_loss=0.1033, over 3932117.55 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3354, pruned_loss=0.09675, over 5694014.79 frames. ], batch size: 164, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:43:46,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.646e+02 9.958e+02 1.212e+03 1.786e+03 1.005e+04, threshold=2.424e+03, percent-clipped=13.0
+2023-03-04 01:43:59,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=321106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:44:16,165 INFO [train.py:968] (0/2) Epoch 8, batch 2350, giga_loss[loss=0.2739, simple_loss=0.3353, pruned_loss=0.1063, over 28915.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3365, pruned_loss=0.09635, over 5712462.48 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3599, pruned_loss=0.1033, over 4024873.97 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3335, pruned_loss=0.09581, over 5705679.08 frames. ], batch size: 112, lr: 4.17e-03, grad_scale: 2.0
+2023-03-04 01:44:37,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3948, 2.0986, 2.1160, 1.9328], device='cuda:0'), covar=tensor([0.1324, 0.2198, 0.1720, 0.1887], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0729, 0.0651, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 01:44:52,763 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=321172.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:44:54,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=321175.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:44:58,954 INFO [train.py:968] (0/2) Epoch 8, batch 2400, giga_loss[loss=0.2453, simple_loss=0.3118, pruned_loss=0.08943, over 28876.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3347, pruned_loss=0.09543, over 5720132.05 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3604, pruned_loss=0.1035, over 4061944.09 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3314, pruned_loss=0.09472, over 5711382.13 frames. ], batch size: 99, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:45:06,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.942e+02 9.114e+02 1.101e+03 1.645e+03 4.776e+03, threshold=2.201e+03, percent-clipped=9.0
+2023-03-04 01:45:16,675 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3941, 1.5687, 1.2122, 0.9848], device='cuda:0'), covar=tensor([0.1660, 0.1399, 0.1130, 0.1539], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1412, 0.1397, 0.1512], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 01:45:17,960 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=321204.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:45:37,026 INFO [train.py:968] (0/2) Epoch 8, batch 2450, giga_loss[loss=0.2606, simple_loss=0.3341, pruned_loss=0.09348, over 28025.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3336, pruned_loss=0.0948, over 5724177.32 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3616, pruned_loss=0.104, over 4136798.04 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3291, pruned_loss=0.09354, over 5715649.41 frames. ], batch size: 412, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:45:44,706 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0539, 1.2477, 1.4208, 1.0237], device='cuda:0'), covar=tensor([0.1132, 0.0975, 0.1514, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0731, 0.0652, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 01:46:08,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-04 01:46:13,021 INFO [train.py:968] (0/2) Epoch 8, batch 2500, giga_loss[loss=0.2204, simple_loss=0.302, pruned_loss=0.06942, over 28790.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3309, pruned_loss=0.09323, over 5732838.86 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3622, pruned_loss=0.1043, over 4188451.73 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3262, pruned_loss=0.09176, over 5722156.94 frames. ], batch size: 119, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:46:21,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.805e+02 9.010e+02 1.205e+03 1.675e+03 6.717e+03, threshold=2.409e+03, percent-clipped=10.0
+2023-03-04 01:46:35,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3979, 1.4801, 1.4393, 1.3759], device='cuda:0'), covar=tensor([0.1192, 0.1625, 0.1725, 0.1619], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0729, 0.0650, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 01:46:54,872 INFO [train.py:968] (0/2) Epoch 8, batch 2550, giga_loss[loss=0.263, simple_loss=0.3362, pruned_loss=0.09494, over 28301.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3274, pruned_loss=0.09168, over 5721895.09 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3623, pruned_loss=0.1044, over 4197072.47 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3234, pruned_loss=0.0904, over 5712792.49 frames. ], batch size: 368, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:47:28,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3641, 1.9221, 1.4661, 0.6043], device='cuda:0'), covar=tensor([0.2751, 0.1574, 0.2801, 0.3363], device='cuda:0'), in_proj_covar=tensor([0.1469, 0.1379, 0.1430, 0.1200], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 01:47:33,073 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.39 vs. limit=5.0
+2023-03-04 01:47:33,898 INFO [train.py:968] (0/2) Epoch 8, batch 2600, libri_loss[loss=0.2863, simple_loss=0.3709, pruned_loss=0.1009, over 29540.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3265, pruned_loss=0.09085, over 5731611.00 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3633, pruned_loss=0.1047, over 4247676.52 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3216, pruned_loss=0.08923, over 5719244.65 frames. ], batch size: 84, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:47:42,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.558e+02 9.470e+02 1.193e+03 1.730e+03 8.287e+03, threshold=2.385e+03, percent-clipped=13.0
+2023-03-04 01:47:59,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9620, 1.1302, 1.0185, 0.8068], device='cuda:0'), covar=tensor([0.1162, 0.1256, 0.0761, 0.1153], device='cuda:0'), in_proj_covar=tensor([0.1569, 0.1404, 0.1395, 0.1510], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 01:48:07,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2814, 1.6996, 1.6180, 1.2196], device='cuda:0'), covar=tensor([0.1559, 0.2141, 0.1242, 0.1404], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0718, 0.0832, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 01:48:13,043 INFO [train.py:968] (0/2) Epoch 8, batch 2650, giga_loss[loss=0.2498, simple_loss=0.3183, pruned_loss=0.09063, over 28737.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3247, pruned_loss=0.09017, over 5726928.06 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3635, pruned_loss=0.1047, over 4254065.14 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3206, pruned_loss=0.08882, over 5718115.31 frames. ], batch size: 99, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:48:17,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=321434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:48:32,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8892, 1.0696, 1.0041, 0.7613], device='cuda:0'), covar=tensor([0.1345, 0.1349, 0.0846, 0.1210], device='cuda:0'), in_proj_covar=tensor([0.1578, 0.1411, 0.1405, 0.1518], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 01:48:56,935 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3521, 4.1743, 3.9216, 2.0186], device='cuda:0'), covar=tensor([0.0411, 0.0526, 0.0589, 0.2156], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0888, 0.0785, 0.0623], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 01:48:57,311 INFO [train.py:968] (0/2) Epoch 8, batch 2700, giga_loss[loss=0.3044, simple_loss=0.3699, pruned_loss=0.1194, over 27556.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3268, pruned_loss=0.09163, over 5717528.03 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3639, pruned_loss=0.1048, over 4278690.67 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3227, pruned_loss=0.09034, over 5708204.10 frames. ], batch size: 472, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:49:08,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.292e+02 9.038e+02 1.109e+03 1.465e+03 3.783e+03, threshold=2.219e+03, percent-clipped=3.0
+2023-03-04 01:49:29,926 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:49:31,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:49:39,783 INFO [train.py:968] (0/2) Epoch 8, batch 2750, giga_loss[loss=0.2982, simple_loss=0.3675, pruned_loss=0.1144, over 28709.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3331, pruned_loss=0.09587, over 5712358.31 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.364, pruned_loss=0.1047, over 4316108.39 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3289, pruned_loss=0.09459, over 5709168.32 frames. ], batch size: 262, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:50:01,523 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:50:21,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=321577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:50:22,004 INFO [train.py:968] (0/2) Epoch 8, batch 2800, giga_loss[loss=0.312, simple_loss=0.3784, pruned_loss=0.1228, over 28972.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3399, pruned_loss=0.1004, over 5702519.41 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3633, pruned_loss=0.1042, over 4373944.49 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3361, pruned_loss=0.09954, over 5696844.04 frames. ], batch size: 164, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:50:24,079 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=321580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:50:32,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.991e+02 1.128e+03 1.522e+03 2.089e+03 4.685e+03, threshold=3.044e+03, percent-clipped=20.0
+2023-03-04 01:50:40,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6186, 1.8873, 1.9917, 1.5415], device='cuda:0'), covar=tensor([0.1510, 0.1935, 0.1129, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0719, 0.0830, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 01:50:49,842 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=321609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:51:06,412 INFO [train.py:968] (0/2) Epoch 8, batch 2850, giga_loss[loss=0.3033, simple_loss=0.3574, pruned_loss=0.1246, over 23675.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3483, pruned_loss=0.1061, over 5690992.82 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3625, pruned_loss=0.1037, over 4414227.95 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3453, pruned_loss=0.1057, over 5685431.87 frames. ], batch size: 705, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:51:49,977 INFO [train.py:968] (0/2) Epoch 8, batch 2900, giga_loss[loss=0.2897, simple_loss=0.372, pruned_loss=0.1037, over 28878.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3542, pruned_loss=0.1085, over 5689888.76 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3626, pruned_loss=0.1037, over 4458323.48 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3514, pruned_loss=0.1083, over 5679812.91 frames. ], batch size: 213, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:52:03,380 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.212e+02 1.136e+03 1.448e+03 2.118e+03 5.714e+03, threshold=2.896e+03, percent-clipped=10.0
+2023-03-04 01:52:27,265 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:52:36,599 INFO [train.py:968] (0/2) Epoch 8, batch 2950, giga_loss[loss=0.3154, simple_loss=0.3881, pruned_loss=0.1214, over 29025.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.359, pruned_loss=0.1107, over 5682930.50 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3621, pruned_loss=0.1035, over 4486907.20 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.357, pruned_loss=0.1108, over 5670878.65 frames. ], batch size: 136, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:53:23,931 INFO [train.py:968] (0/2) Epoch 8, batch 3000, giga_loss[loss=0.3275, simple_loss=0.3994, pruned_loss=0.1278, over 28707.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3655, pruned_loss=0.1146, over 5690428.18 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3624, pruned_loss=0.1037, over 4511261.15 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3637, pruned_loss=0.1147, over 5680294.20 frames. ], batch size: 242, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:53:23,937 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 01:53:31,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2342, 1.4954, 1.5395, 1.3316], device='cuda:0'), covar=tensor([0.1217, 0.1157, 0.1480, 0.1282], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0726, 0.0649, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 01:53:32,726 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2338, simple_loss=0.3367, pruned_loss=0.06548, over 944034.00 frames. 
+2023-03-04 01:53:32,726 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 01:53:43,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.125e+02 1.212e+03 1.570e+03 2.101e+03 7.244e+03, threshold=3.140e+03, percent-clipped=12.0
+2023-03-04 01:53:48,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2965, 1.5894, 1.2638, 1.4372], device='cuda:0'), covar=tensor([0.0766, 0.0317, 0.0316, 0.0829], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0070, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 01:54:04,221 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:54:13,264 INFO [train.py:968] (0/2) Epoch 8, batch 3050, giga_loss[loss=0.2466, simple_loss=0.3222, pruned_loss=0.08554, over 28718.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3635, pruned_loss=0.1134, over 5681592.97 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3625, pruned_loss=0.1039, over 4544196.53 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.362, pruned_loss=0.1136, over 5669719.93 frames. ], batch size: 284, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:54:44,159 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:54:54,760 INFO [train.py:968] (0/2) Epoch 8, batch 3100, giga_loss[loss=0.2706, simple_loss=0.354, pruned_loss=0.09357, over 28952.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3573, pruned_loss=0.1084, over 5688884.02 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3627, pruned_loss=0.1041, over 4571122.13 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3559, pruned_loss=0.1085, over 5675510.69 frames. ], batch size: 164, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:55:04,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.875e+02 1.013e+03 1.240e+03 1.729e+03 3.827e+03, threshold=2.479e+03, percent-clipped=4.0
+2023-03-04 01:55:05,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=321893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:55:05,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3635, 1.5528, 1.5942, 1.4572], device='cuda:0'), covar=tensor([0.1257, 0.1272, 0.1251, 0.1268], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0726, 0.0649, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 01:55:06,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=321894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:55:18,263 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-04 01:55:31,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2495, 1.6446, 1.3001, 1.4216], device='cuda:0'), covar=tensor([0.0777, 0.0300, 0.0314, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0070, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 01:55:37,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=321928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:55:37,759 INFO [train.py:968] (0/2) Epoch 8, batch 3150, libri_loss[loss=0.27, simple_loss=0.3354, pruned_loss=0.1023, over 29462.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.356, pruned_loss=0.1071, over 5682316.64 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3618, pruned_loss=0.1037, over 4613270.35 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3553, pruned_loss=0.1076, over 5672341.80 frames. ], batch size: 70, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:56:07,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.26 vs. limit=5.0
+2023-03-04 01:56:21,623 INFO [train.py:968] (0/2) Epoch 8, batch 3200, giga_loss[loss=0.269, simple_loss=0.3482, pruned_loss=0.09492, over 28905.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3569, pruned_loss=0.1075, over 5675888.83 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3616, pruned_loss=0.1037, over 4626248.05 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3564, pruned_loss=0.1079, over 5666095.50 frames. ], batch size: 164, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:56:29,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.507e+02 1.147e+03 1.494e+03 1.874e+03 4.226e+03, threshold=2.989e+03, percent-clipped=7.0
+2023-03-04 01:56:38,431 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-322000.pt
+2023-03-04 01:57:02,086 INFO [train.py:968] (0/2) Epoch 8, batch 3250, giga_loss[loss=0.3122, simple_loss=0.3875, pruned_loss=0.1185, over 28848.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3587, pruned_loss=0.1083, over 5683836.15 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3609, pruned_loss=0.1032, over 4664315.84 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3588, pruned_loss=0.1091, over 5669626.31 frames. ], batch size: 174, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:57:07,173 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:08,815 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:10,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:10,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:14,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3691, 4.2024, 3.9132, 1.8442], device='cuda:0'), covar=tensor([0.0465, 0.0594, 0.0636, 0.2030], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0876, 0.0767, 0.0611], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 01:57:33,891 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:34,660 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:36,589 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:39,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:42,514 INFO [train.py:968] (0/2) Epoch 8, batch 3300, giga_loss[loss=0.3229, simple_loss=0.3854, pruned_loss=0.1302, over 28341.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3615, pruned_loss=0.1098, over 5686475.55 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3614, pruned_loss=0.1034, over 4675290.47 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3612, pruned_loss=0.1105, over 5683444.27 frames. ], batch size: 65, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:57:53,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.691e+02 1.166e+03 1.514e+03 1.939e+03 3.866e+03, threshold=3.028e+03, percent-clipped=3.0
+2023-03-04 01:57:55,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2356, 1.8893, 1.5129, 0.5063], device='cuda:0'), covar=tensor([0.2735, 0.1465, 0.2223, 0.3186], device='cuda:0'), in_proj_covar=tensor([0.1447, 0.1359, 0.1425, 0.1188], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 01:57:56,418 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=322093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:58:03,977 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:58:04,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3258, 3.1584, 2.0342, 1.7966], device='cuda:0'), covar=tensor([0.1423, 0.0647, 0.0914, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.1583, 0.1415, 0.1404, 0.1511], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 01:58:08,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4844, 2.0619, 1.4274, 0.7406], device='cuda:0'), covar=tensor([0.2859, 0.1380, 0.2245, 0.3142], device='cuda:0'), in_proj_covar=tensor([0.1443, 0.1355, 0.1421, 0.1184], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 01:58:11,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6879, 1.8721, 1.9785, 1.5791], device='cuda:0'), covar=tensor([0.1662, 0.1923, 0.1195, 0.1337], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0714, 0.0827, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 01:58:25,348 INFO [train.py:968] (0/2) Epoch 8, batch 3350, giga_loss[loss=0.3044, simple_loss=0.3711, pruned_loss=0.1189, over 28465.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3636, pruned_loss=0.112, over 5683454.28 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3616, pruned_loss=0.1037, over 4695908.97 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3632, pruned_loss=0.1124, over 5680484.31 frames. ], batch size: 71, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:59:08,204 INFO [train.py:968] (0/2) Epoch 8, batch 3400, giga_loss[loss=0.3086, simple_loss=0.3621, pruned_loss=0.1275, over 28543.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3642, pruned_loss=0.113, over 5684062.92 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3623, pruned_loss=0.104, over 4719901.62 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3635, pruned_loss=0.1133, over 5677674.01 frames. ], batch size: 85, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:59:18,004 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=322192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:59:18,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.224e+02 1.315e+03 1.617e+03 2.082e+03 5.530e+03, threshold=3.234e+03, percent-clipped=5.0
+2023-03-04 01:59:47,758 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-04 01:59:48,729 INFO [train.py:968] (0/2) Epoch 8, batch 3450, giga_loss[loss=0.2943, simple_loss=0.3633, pruned_loss=0.1126, over 28786.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3649, pruned_loss=0.1135, over 5684600.01 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.363, pruned_loss=0.1042, over 4762616.64 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3638, pruned_loss=0.114, over 5675293.12 frames. ], batch size: 99, lr: 4.16e-03, grad_scale: 2.0
+2023-03-04 01:59:55,609 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:59:57,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:59:58,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=322241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:00:14,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9234, 1.1793, 3.4365, 2.8945], device='cuda:0'), covar=tensor([0.1669, 0.2406, 0.0443, 0.1119], device='cuda:0'), in_proj_covar=tensor([0.0599, 0.0553, 0.0788, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:00:22,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:00:31,542 INFO [train.py:968] (0/2) Epoch 8, batch 3500, giga_loss[loss=0.265, simple_loss=0.3457, pruned_loss=0.0922, over 28744.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3649, pruned_loss=0.1125, over 5693578.07 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3633, pruned_loss=0.1043, over 4789264.80 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3638, pruned_loss=0.113, over 5682627.65 frames. ], batch size: 99, lr: 4.16e-03, grad_scale: 2.0
+2023-03-04 02:00:40,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.707e+02 1.065e+03 1.344e+03 2.054e+03 7.016e+03, threshold=2.688e+03, percent-clipped=7.0
+2023-03-04 02:01:07,574 INFO [train.py:968] (0/2) Epoch 8, batch 3550, giga_loss[loss=0.2769, simple_loss=0.3511, pruned_loss=0.1014, over 28510.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.364, pruned_loss=0.1109, over 5702241.87 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3631, pruned_loss=0.1044, over 4840615.66 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3634, pruned_loss=0.1115, over 5686761.26 frames. ], batch size: 78, lr: 4.16e-03, grad_scale: 2.0
+2023-03-04 02:01:07,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=322329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:01:12,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:01:16,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:01:42,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:01:52,384 INFO [train.py:968] (0/2) Epoch 8, batch 3600, giga_loss[loss=0.2918, simple_loss=0.3404, pruned_loss=0.1216, over 23764.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3641, pruned_loss=0.1104, over 5696421.70 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3629, pruned_loss=0.1043, over 4850888.34 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3638, pruned_loss=0.1111, over 5682634.18 frames. ], batch size: 705, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:01:56,485 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:01:58,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:02:03,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.003e+02 9.709e+02 1.327e+03 1.820e+03 4.576e+03, threshold=2.654e+03, percent-clipped=4.0
+2023-03-04 02:02:13,481 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-04 02:02:21,597 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:02:31,127 INFO [train.py:968] (0/2) Epoch 8, batch 3650, giga_loss[loss=0.2361, simple_loss=0.3173, pruned_loss=0.07746, over 29010.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3616, pruned_loss=0.109, over 5708139.12 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3623, pruned_loss=0.104, over 4871652.30 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3618, pruned_loss=0.1098, over 5693573.52 frames. ], batch size: 128, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:02:52,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2789, 1.3729, 3.2665, 3.1327], device='cuda:0'), covar=tensor([0.1291, 0.2150, 0.0394, 0.0721], device='cuda:0'), in_proj_covar=tensor([0.0594, 0.0548, 0.0783, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:03:13,255 INFO [train.py:968] (0/2) Epoch 8, batch 3700, giga_loss[loss=0.281, simple_loss=0.3525, pruned_loss=0.1048, over 28834.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3585, pruned_loss=0.1077, over 5699220.49 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.362, pruned_loss=0.1039, over 4884098.91 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3589, pruned_loss=0.1085, over 5688014.28 frames. ], batch size: 174, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:03:26,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.998e+02 9.806e+02 1.272e+03 1.682e+03 2.568e+03, threshold=2.545e+03, percent-clipped=0.0
+2023-03-04 02:03:54,145 INFO [train.py:968] (0/2) Epoch 8, batch 3750, giga_loss[loss=0.2803, simple_loss=0.3549, pruned_loss=0.1029, over 28766.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3559, pruned_loss=0.106, over 5706046.05 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3618, pruned_loss=0.1038, over 4899242.56 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3562, pruned_loss=0.1067, over 5694750.76 frames. ], batch size: 99, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:04:37,728 INFO [train.py:968] (0/2) Epoch 8, batch 3800, giga_loss[loss=0.2891, simple_loss=0.3631, pruned_loss=0.1076, over 29047.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3566, pruned_loss=0.1069, over 5696460.84 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3617, pruned_loss=0.1038, over 4913737.60 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3569, pruned_loss=0.1075, over 5691828.31 frames. ], batch size: 155, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:04:47,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.859e+02 1.005e+03 1.323e+03 1.697e+03 4.604e+03, threshold=2.646e+03, percent-clipped=7.0
+2023-03-04 02:05:04,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5272, 1.7671, 1.7753, 1.6349], device='cuda:0'), covar=tensor([0.1163, 0.1187, 0.1364, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0719, 0.0644, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 02:05:17,415 INFO [train.py:968] (0/2) Epoch 8, batch 3850, giga_loss[loss=0.2899, simple_loss=0.365, pruned_loss=0.1074, over 29075.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3573, pruned_loss=0.1068, over 5697748.04 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3614, pruned_loss=0.1034, over 4939348.51 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3576, pruned_loss=0.1077, over 5695436.35 frames. ], batch size: 155, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:05:56,072 INFO [train.py:968] (0/2) Epoch 8, batch 3900, giga_loss[loss=0.2645, simple_loss=0.3468, pruned_loss=0.09114, over 28597.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3569, pruned_loss=0.1054, over 5697319.66 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3616, pruned_loss=0.1036, over 4949159.51 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3569, pruned_loss=0.1061, over 5699763.64 frames. ], batch size: 65, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:06:07,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.498e+02 1.012e+03 1.304e+03 1.772e+03 6.586e+03, threshold=2.609e+03, percent-clipped=6.0
+2023-03-04 02:06:15,688 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=322704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:06:36,227 INFO [train.py:968] (0/2) Epoch 8, batch 3950, libri_loss[loss=0.2425, simple_loss=0.3182, pruned_loss=0.08345, over 29373.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3567, pruned_loss=0.1049, over 5705866.50 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3612, pruned_loss=0.1032, over 4974829.77 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3569, pruned_loss=0.1057, over 5704707.25 frames. ], batch size: 67, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:06:38,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4487, 1.5643, 1.4869, 1.4973], device='cuda:0'), covar=tensor([0.1157, 0.1546, 0.1572, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0720, 0.0642, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 02:07:14,027 INFO [train.py:968] (0/2) Epoch 8, batch 4000, giga_loss[loss=0.2671, simple_loss=0.3429, pruned_loss=0.09564, over 28514.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3566, pruned_loss=0.1054, over 5705098.59 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3605, pruned_loss=0.1031, over 5006321.47 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3572, pruned_loss=0.1062, over 5698261.67 frames. ], batch size: 85, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:07:25,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.064e+02 1.020e+03 1.268e+03 1.928e+03 5.893e+03, threshold=2.536e+03, percent-clipped=8.0
+2023-03-04 02:07:55,617 INFO [train.py:968] (0/2) Epoch 8, batch 4050, libri_loss[loss=0.2642, simple_loss=0.343, pruned_loss=0.09271, over 28520.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3547, pruned_loss=0.1047, over 5710551.20 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3603, pruned_loss=0.103, over 5021173.81 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3552, pruned_loss=0.1055, over 5704279.32 frames. ], batch size: 63, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:07:56,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0692, 1.8551, 1.4498, 1.6446], device='cuda:0'), covar=tensor([0.0670, 0.0689, 0.0919, 0.1024], device='cuda:0'), in_proj_covar=tensor([0.0339, 0.0437, 0.0498, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 02:08:09,132 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:08:11,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:08:33,530 INFO [train.py:968] (0/2) Epoch 8, batch 4100, giga_loss[loss=0.2463, simple_loss=0.3291, pruned_loss=0.08176, over 27946.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3526, pruned_loss=0.1039, over 5714678.03 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3599, pruned_loss=0.1028, over 5046489.21 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3532, pruned_loss=0.1047, over 5705219.51 frames. ], batch size: 412, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:08:33,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:08:42,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.246e+02 1.011e+03 1.253e+03 1.788e+03 6.390e+03, threshold=2.506e+03, percent-clipped=7.0
+2023-03-04 02:09:09,388 INFO [train.py:968] (0/2) Epoch 8, batch 4150, giga_loss[loss=0.3171, simple_loss=0.3776, pruned_loss=0.1282, over 27665.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3517, pruned_loss=0.1037, over 5709556.85 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3607, pruned_loss=0.1033, over 5067688.97 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3513, pruned_loss=0.104, over 5706301.87 frames. ], batch size: 472, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:09:49,978 INFO [train.py:968] (0/2) Epoch 8, batch 4200, libri_loss[loss=0.3118, simple_loss=0.3883, pruned_loss=0.1176, over 29493.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3523, pruned_loss=0.1046, over 5712550.17 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3602, pruned_loss=0.1031, over 5091240.28 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.352, pruned_loss=0.105, over 5706040.88 frames. ], batch size: 85, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:10:00,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.609e+02 1.185e+03 1.515e+03 2.223e+03 4.659e+03, threshold=3.031e+03, percent-clipped=19.0
+2023-03-04 02:10:21,768 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 02:10:27,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9775, 1.8761, 1.4648, 1.5578], device='cuda:0'), covar=tensor([0.0635, 0.0616, 0.0912, 0.0960], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0436, 0.0497, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 02:10:27,550 INFO [train.py:968] (0/2) Epoch 8, batch 4250, giga_loss[loss=0.2482, simple_loss=0.3223, pruned_loss=0.08708, over 28573.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3515, pruned_loss=0.1047, over 5715417.49 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3601, pruned_loss=0.1031, over 5118279.39 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.351, pruned_loss=0.105, over 5705043.08 frames. ], batch size: 85, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:11:11,308 INFO [train.py:968] (0/2) Epoch 8, batch 4300, giga_loss[loss=0.2788, simple_loss=0.3466, pruned_loss=0.1055, over 28829.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3504, pruned_loss=0.1046, over 5713220.14 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3599, pruned_loss=0.1029, over 5137427.41 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.35, pruned_loss=0.1051, over 5700906.23 frames. ], batch size: 174, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:11:21,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.632e+02 1.119e+03 1.343e+03 1.776e+03 4.776e+03, threshold=2.685e+03, percent-clipped=5.0
+2023-03-04 02:11:50,788 INFO [train.py:968] (0/2) Epoch 8, batch 4350, giga_loss[loss=0.2487, simple_loss=0.3181, pruned_loss=0.08964, over 28384.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3474, pruned_loss=0.1033, over 5717902.26 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3601, pruned_loss=0.103, over 5155987.80 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3467, pruned_loss=0.1037, over 5704139.86 frames. ], batch size: 78, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:11:53,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4615, 4.2599, 4.0647, 1.9834], device='cuda:0'), covar=tensor([0.0481, 0.0674, 0.0726, 0.1936], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0886, 0.0780, 0.0624], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 02:12:27,976 INFO [train.py:968] (0/2) Epoch 8, batch 4400, giga_loss[loss=0.2644, simple_loss=0.3355, pruned_loss=0.09666, over 28828.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3449, pruned_loss=0.1019, over 5709919.53 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3602, pruned_loss=0.103, over 5171140.77 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3437, pruned_loss=0.1021, over 5703420.12 frames. ], batch size: 243, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:12:40,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.186e+02 1.011e+03 1.254e+03 1.840e+03 7.684e+03, threshold=2.508e+03, percent-clipped=9.0
+2023-03-04 02:12:52,123 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-04 02:13:07,566 INFO [train.py:968] (0/2) Epoch 8, batch 4450, giga_loss[loss=0.2908, simple_loss=0.3559, pruned_loss=0.1128, over 28875.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3438, pruned_loss=0.101, over 5703889.61 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3602, pruned_loss=0.1031, over 5178000.16 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3425, pruned_loss=0.1011, over 5704449.33 frames. ], batch size: 112, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:13:47,440 INFO [train.py:968] (0/2) Epoch 8, batch 4500, libri_loss[loss=0.294, simple_loss=0.3778, pruned_loss=0.1051, over 29214.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3471, pruned_loss=0.1027, over 5705688.41 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3602, pruned_loss=0.1031, over 5206448.28 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3455, pruned_loss=0.1027, over 5701278.80 frames. ], batch size: 97, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:14:03,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.073e+02 1.009e+03 1.318e+03 1.674e+03 3.389e+03, threshold=2.636e+03, percent-clipped=6.0
+2023-03-04 02:14:31,734 INFO [train.py:968] (0/2) Epoch 8, batch 4550, giga_loss[loss=0.2679, simple_loss=0.3488, pruned_loss=0.09346, over 28327.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.349, pruned_loss=0.1031, over 5719001.35 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3601, pruned_loss=0.1033, over 5227839.05 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3474, pruned_loss=0.103, over 5709584.69 frames. ], batch size: 368, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:14:38,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6368, 1.0857, 2.8472, 2.5836], device='cuda:0'), covar=tensor([0.1695, 0.2291, 0.0495, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0598, 0.0553, 0.0787, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:15:00,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9639, 3.0103, 2.0084, 0.9091], device='cuda:0'), covar=tensor([0.4180, 0.1660, 0.2476, 0.4271], device='cuda:0'), in_proj_covar=tensor([0.1458, 0.1357, 0.1425, 0.1196], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 02:15:06,308 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2137, 1.6978, 1.2778, 1.5173], device='cuda:0'), covar=tensor([0.2165, 0.2000, 0.2348, 0.1992], device='cuda:0'), in_proj_covar=tensor([0.1200, 0.0896, 0.1054, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:15:12,225 INFO [train.py:968] (0/2) Epoch 8, batch 4600, giga_loss[loss=0.2577, simple_loss=0.3324, pruned_loss=0.09149, over 28266.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3503, pruned_loss=0.1031, over 5719759.19 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3601, pruned_loss=0.1034, over 5239349.73 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3489, pruned_loss=0.1029, over 5710757.28 frames. ], batch size: 77, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:15:13,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=323380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:15:27,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.332e+02 1.000e+03 1.255e+03 1.755e+03 5.243e+03, threshold=2.510e+03, percent-clipped=5.0
+2023-03-04 02:15:33,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3034, 5.0903, 4.8245, 2.4832], device='cuda:0'), covar=tensor([0.0322, 0.0480, 0.0559, 0.1787], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0883, 0.0781, 0.0618], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 02:15:57,349 INFO [train.py:968] (0/2) Epoch 8, batch 4650, giga_loss[loss=0.2394, simple_loss=0.321, pruned_loss=0.07891, over 29073.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3519, pruned_loss=0.1035, over 5716925.86 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3601, pruned_loss=0.1033, over 5265834.47 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3504, pruned_loss=0.1034, over 5702522.60 frames. ], batch size: 128, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:16:13,470 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-04 02:16:42,203 INFO [train.py:968] (0/2) Epoch 8, batch 4700, giga_loss[loss=0.2728, simple_loss=0.36, pruned_loss=0.09286, over 28990.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3518, pruned_loss=0.1032, over 5708036.35 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3597, pruned_loss=0.1033, over 5279667.37 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3508, pruned_loss=0.1031, over 5694303.23 frames. ], batch size: 145, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:16:53,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.978e+02 1.021e+03 1.287e+03 1.595e+03 4.477e+03, threshold=2.573e+03, percent-clipped=12.0
+2023-03-04 02:17:21,644 INFO [train.py:968] (0/2) Epoch 8, batch 4750, giga_loss[loss=0.2646, simple_loss=0.3354, pruned_loss=0.09688, over 28685.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3522, pruned_loss=0.1033, over 5710147.92 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3597, pruned_loss=0.1033, over 5293459.64 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3511, pruned_loss=0.1032, over 5696112.83 frames. ], batch size: 99, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:18:02,552 INFO [train.py:968] (0/2) Epoch 8, batch 4800, giga_loss[loss=0.2965, simple_loss=0.3718, pruned_loss=0.1106, over 28907.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3547, pruned_loss=0.1052, over 5719085.62 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3607, pruned_loss=0.1038, over 5306691.05 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.353, pruned_loss=0.1047, over 5705408.51 frames. ], batch size: 199, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:18:15,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.186e+02 1.229e+03 1.666e+03 2.447e+03 6.023e+03, threshold=3.332e+03, percent-clipped=22.0
+2023-03-04 02:18:45,121 INFO [train.py:968] (0/2) Epoch 8, batch 4850, giga_loss[loss=0.2616, simple_loss=0.3384, pruned_loss=0.09243, over 28979.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3556, pruned_loss=0.1061, over 5714704.58 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3611, pruned_loss=0.104, over 5317928.03 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3538, pruned_loss=0.1056, over 5700971.47 frames. ], batch size: 145, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:19:30,542 INFO [train.py:968] (0/2) Epoch 8, batch 4900, giga_loss[loss=0.2841, simple_loss=0.3545, pruned_loss=0.1068, over 28812.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3572, pruned_loss=0.1069, over 5717527.43 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3616, pruned_loss=0.1043, over 5329684.83 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3552, pruned_loss=0.1063, over 5703178.61 frames. ], batch size: 213, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:19:39,139 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-04 02:19:41,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.009e+02 1.044e+03 1.341e+03 1.763e+03 3.574e+03, threshold=2.682e+03, percent-clipped=2.0
+2023-03-04 02:20:09,088 INFO [train.py:968] (0/2) Epoch 8, batch 4950, giga_loss[loss=0.2534, simple_loss=0.3433, pruned_loss=0.08175, over 28918.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3587, pruned_loss=0.1074, over 5721833.33 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3615, pruned_loss=0.1043, over 5340589.40 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3571, pruned_loss=0.107, over 5707697.22 frames. ], batch size: 174, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:20:31,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=323755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:20:36,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=323762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:20:39,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2379, 1.4633, 1.1848, 1.4038], device='cuda:0'), covar=tensor([0.0737, 0.0297, 0.0323, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 02:20:49,920 INFO [train.py:968] (0/2) Epoch 8, batch 5000, giga_loss[loss=0.3176, simple_loss=0.381, pruned_loss=0.1271, over 29027.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3598, pruned_loss=0.1078, over 5719573.72 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3618, pruned_loss=0.1044, over 5347427.06 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3583, pruned_loss=0.1074, over 5707351.16 frames. ], batch size: 155, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:21:01,927 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.498e+02 1.116e+03 1.421e+03 1.998e+03 6.474e+03, threshold=2.842e+03, percent-clipped=9.0
+2023-03-04 02:21:31,254 INFO [train.py:968] (0/2) Epoch 8, batch 5050, giga_loss[loss=0.339, simple_loss=0.4145, pruned_loss=0.1318, over 28754.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.361, pruned_loss=0.1086, over 5712315.28 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.362, pruned_loss=0.1045, over 5353004.90 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3596, pruned_loss=0.1083, over 5700822.32 frames. ], batch size: 262, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:21:47,722 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=323850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:21:55,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5490, 2.2343, 1.5413, 0.7122], device='cuda:0'), covar=tensor([0.2840, 0.1560, 0.2486, 0.3259], device='cuda:0'), in_proj_covar=tensor([0.1472, 0.1374, 0.1435, 0.1210], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 02:22:10,003 INFO [train.py:968] (0/2) Epoch 8, batch 5100, giga_loss[loss=0.2699, simple_loss=0.3406, pruned_loss=0.09957, over 28497.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3618, pruned_loss=0.1092, over 5716740.77 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3624, pruned_loss=0.1046, over 5373213.63 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3602, pruned_loss=0.1089, over 5702066.78 frames. ], batch size: 71, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:22:19,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1475, 1.3893, 1.1755, 1.0183], device='cuda:0'), covar=tensor([0.1841, 0.1841, 0.1919, 0.1665], device='cuda:0'), in_proj_covar=tensor([0.1195, 0.0900, 0.1048, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:22:20,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1086, 3.9036, 3.7134, 1.7285], device='cuda:0'), covar=tensor([0.0523, 0.0644, 0.0693, 0.2162], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0882, 0.0784, 0.0618], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 02:22:22,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.367e+02 1.186e+03 1.587e+03 2.256e+03 5.632e+03, threshold=3.175e+03, percent-clipped=12.0
+2023-03-04 02:22:22,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4048, 1.6122, 1.3860, 1.2213], device='cuda:0'), covar=tensor([0.1783, 0.1499, 0.1185, 0.1485], device='cuda:0'), in_proj_covar=tensor([0.1596, 0.1430, 0.1418, 0.1517], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 02:22:25,050 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=323898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:22:27,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=323901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:22:49,850 INFO [train.py:968] (0/2) Epoch 8, batch 5150, giga_loss[loss=0.3111, simple_loss=0.3702, pruned_loss=0.126, over 28747.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3609, pruned_loss=0.1088, over 5718660.79 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3624, pruned_loss=0.1046, over 5384512.11 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3597, pruned_loss=0.1088, over 5704378.40 frames. ], batch size: 99, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:22:50,715 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=323930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:23:31,253 INFO [train.py:968] (0/2) Epoch 8, batch 5200, libri_loss[loss=0.3093, simple_loss=0.3842, pruned_loss=0.1172, over 29665.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3585, pruned_loss=0.108, over 5713483.64 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3629, pruned_loss=0.1048, over 5397405.61 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.357, pruned_loss=0.1079, over 5700167.16 frames. ], batch size: 88, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:23:44,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.804e+02 1.178e+03 1.641e+03 2.486e+03 1.023e+04, threshold=3.282e+03, percent-clipped=17.0
+2023-03-04 02:23:48,412 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-324000.pt
+2023-03-04 02:23:51,311 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-04 02:24:11,590 INFO [train.py:968] (0/2) Epoch 8, batch 5250, giga_loss[loss=0.2826, simple_loss=0.3539, pruned_loss=0.1056, over 27952.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.355, pruned_loss=0.1062, over 5716383.01 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3628, pruned_loss=0.1047, over 5402539.64 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3539, pruned_loss=0.1062, over 5704134.38 frames. ], batch size: 412, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:24:27,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1458, 1.0249, 4.2223, 3.1943], device='cuda:0'), covar=tensor([0.1656, 0.2642, 0.0358, 0.1039], device='cuda:0'), in_proj_covar=tensor([0.0601, 0.0558, 0.0798, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:24:34,823 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=324057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:24:49,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1292, 1.3573, 4.1506, 3.1871], device='cuda:0'), covar=tensor([0.1943, 0.2387, 0.0748, 0.0973], device='cuda:0'), in_proj_covar=tensor([0.0600, 0.0559, 0.0800, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:24:50,987 INFO [train.py:968] (0/2) Epoch 8, batch 5300, giga_loss[loss=0.2782, simple_loss=0.3602, pruned_loss=0.0981, over 28913.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3554, pruned_loss=0.1055, over 5710690.04 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3629, pruned_loss=0.1049, over 5412949.09 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.354, pruned_loss=0.1053, over 5704256.84 frames. ], batch size: 227, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:25:05,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.609e+02 1.017e+03 1.245e+03 1.656e+03 3.489e+03, threshold=2.489e+03, percent-clipped=1.0
+2023-03-04 02:25:09,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4315, 3.5570, 1.4660, 1.6019], device='cuda:0'), covar=tensor([0.0849, 0.0242, 0.0892, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0497, 0.0321, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 02:25:25,784 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=324118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:25:35,167 INFO [train.py:968] (0/2) Epoch 8, batch 5350, giga_loss[loss=0.2303, simple_loss=0.3227, pruned_loss=0.069, over 29023.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3581, pruned_loss=0.1059, over 5713422.68 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3629, pruned_loss=0.1048, over 5420230.46 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.357, pruned_loss=0.1058, over 5705596.92 frames. ], batch size: 136, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:25:41,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:26:03,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=324163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:26:15,828 INFO [train.py:968] (0/2) Epoch 8, batch 5400, giga_loss[loss=0.2714, simple_loss=0.3458, pruned_loss=0.09851, over 29022.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3581, pruned_loss=0.1061, over 5709546.13 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3628, pruned_loss=0.1053, over 5420229.97 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3572, pruned_loss=0.1057, over 5713084.05 frames. ], batch size: 155, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:26:27,686 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.958e+02 1.196e+03 1.471e+03 1.875e+03 4.384e+03, threshold=2.942e+03, percent-clipped=11.0
+2023-03-04 02:26:52,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:26:56,218 INFO [train.py:968] (0/2) Epoch 8, batch 5450, giga_loss[loss=0.2721, simple_loss=0.3381, pruned_loss=0.1031, over 29081.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3561, pruned_loss=0.1066, over 5714322.00 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3629, pruned_loss=0.1053, over 5428406.24 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3552, pruned_loss=0.1063, over 5714896.11 frames. ], batch size: 128, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:27:38,269 INFO [train.py:968] (0/2) Epoch 8, batch 5500, giga_loss[loss=0.2629, simple_loss=0.3338, pruned_loss=0.09602, over 28839.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3547, pruned_loss=0.1071, over 5711796.12 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3628, pruned_loss=0.1053, over 5431698.21 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3538, pruned_loss=0.1069, over 5718661.60 frames. ], batch size: 186, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:27:38,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2088, 1.1740, 1.0365, 1.3282], device='cuda:0'), covar=tensor([0.0736, 0.0341, 0.0355, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0115, 0.0118, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 02:27:39,423 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:27:41,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:27:50,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.117e+02 1.174e+03 1.453e+03 1.942e+03 6.371e+03, threshold=2.906e+03, percent-clipped=9.0
+2023-03-04 02:28:04,173 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:28:17,336 INFO [train.py:968] (0/2) Epoch 8, batch 5550, giga_loss[loss=0.2846, simple_loss=0.3523, pruned_loss=0.1084, over 28881.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.353, pruned_loss=0.1073, over 5720863.33 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.363, pruned_loss=0.1054, over 5444097.20 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3519, pruned_loss=0.1072, over 5722943.10 frames. ], batch size: 174, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:28:50,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:28:53,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:29:00,275 INFO [train.py:968] (0/2) Epoch 8, batch 5600, giga_loss[loss=0.3404, simple_loss=0.395, pruned_loss=0.1429, over 28518.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3514, pruned_loss=0.1066, over 5722879.09 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3632, pruned_loss=0.1055, over 5450723.30 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3502, pruned_loss=0.1065, over 5722351.14 frames. ], batch size: 336, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:29:00,475 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=324379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:29:16,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.062e+02 1.097e+03 1.378e+03 1.790e+03 9.480e+03, threshold=2.756e+03, percent-clipped=6.0
+2023-03-04 02:29:20,413 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:29:43,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3253, 1.5448, 1.3157, 1.1314], device='cuda:0'), covar=tensor([0.1922, 0.1388, 0.1132, 0.1492], device='cuda:0'), in_proj_covar=tensor([0.1599, 0.1438, 0.1422, 0.1516], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 02:29:45,891 INFO [train.py:968] (0/2) Epoch 8, batch 5650, giga_loss[loss=0.2514, simple_loss=0.3201, pruned_loss=0.09133, over 28965.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3491, pruned_loss=0.1058, over 5715811.87 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.363, pruned_loss=0.1053, over 5452840.68 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3482, pruned_loss=0.1058, over 5714730.84 frames. ], batch size: 213, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:29:49,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:29:59,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8558, 2.0805, 1.8783, 1.9305], device='cuda:0'), covar=tensor([0.1167, 0.1395, 0.1464, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0722, 0.0647, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 02:30:03,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0436, 1.1951, 4.1065, 3.2910], device='cuda:0'), covar=tensor([0.1717, 0.2467, 0.0380, 0.0683], device='cuda:0'), in_proj_covar=tensor([0.0606, 0.0565, 0.0807, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:30:24,867 INFO [train.py:968] (0/2) Epoch 8, batch 5700, giga_loss[loss=0.2427, simple_loss=0.3118, pruned_loss=0.08684, over 28936.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3453, pruned_loss=0.1037, over 5716346.89 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.363, pruned_loss=0.1054, over 5465901.04 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3442, pruned_loss=0.1036, over 5710837.87 frames. ], batch size: 136, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:30:35,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:30:38,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.155e+02 1.049e+03 1.246e+03 1.525e+03 3.473e+03, threshold=2.493e+03, percent-clipped=2.0
+2023-03-04 02:31:01,215 INFO [train.py:968] (0/2) Epoch 8, batch 5750, giga_loss[loss=0.2695, simple_loss=0.332, pruned_loss=0.1035, over 28766.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.342, pruned_loss=0.1014, over 5717928.15 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3635, pruned_loss=0.1056, over 5480155.47 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3399, pruned_loss=0.101, over 5710682.77 frames. ], batch size: 99, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:31:09,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:31:40,867 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:31:42,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:31:42,957 INFO [train.py:968] (0/2) Epoch 8, batch 5800, giga_loss[loss=0.2668, simple_loss=0.3398, pruned_loss=0.09689, over 28697.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3416, pruned_loss=0.1015, over 5714675.14 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3635, pruned_loss=0.1056, over 5486755.12 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3397, pruned_loss=0.1012, over 5706379.50 frames. ], batch size: 262, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:31:55,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.253e+02 1.077e+03 1.440e+03 1.934e+03 4.576e+03, threshold=2.880e+03, percent-clipped=10.0
+2023-03-04 02:32:01,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8876, 1.0718, 0.8649, 0.1909], device='cuda:0'), covar=tensor([0.1932, 0.1627, 0.1945, 0.3317], device='cuda:0'), in_proj_covar=tensor([0.1479, 0.1377, 0.1441, 0.1215], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 02:32:05,046 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:32:21,941 INFO [train.py:968] (0/2) Epoch 8, batch 5850, libri_loss[loss=0.2576, simple_loss=0.3373, pruned_loss=0.089, over 29537.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3451, pruned_loss=0.1032, over 5711903.82 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3639, pruned_loss=0.1057, over 5492766.23 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3427, pruned_loss=0.1027, over 5704968.64 frames. ], batch size: 80, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:32:24,352 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-04 02:32:27,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:32:30,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:32:54,270 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:33:04,042 INFO [train.py:968] (0/2) Epoch 8, batch 5900, giga_loss[loss=0.2693, simple_loss=0.3454, pruned_loss=0.09658, over 28701.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3484, pruned_loss=0.1044, over 5711456.08 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.364, pruned_loss=0.1059, over 5498817.13 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3461, pruned_loss=0.1038, over 5703649.86 frames. ], batch size: 242, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:33:05,581 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:33:07,770 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:33:19,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.186e+02 1.203e+03 1.568e+03 1.967e+03 6.035e+03, threshold=3.135e+03, percent-clipped=8.0
+2023-03-04 02:33:29,505 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=324709.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 02:33:32,165 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:33:45,655 INFO [train.py:968] (0/2) Epoch 8, batch 5950, giga_loss[loss=0.3762, simple_loss=0.4172, pruned_loss=0.1676, over 26672.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3526, pruned_loss=0.106, over 5714624.29 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3644, pruned_loss=0.1061, over 5505396.24 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3502, pruned_loss=0.1054, over 5707229.17 frames. ], batch size: 555, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:34:05,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:34:27,756 INFO [train.py:968] (0/2) Epoch 8, batch 6000, giga_loss[loss=0.3312, simple_loss=0.3919, pruned_loss=0.1353, over 29009.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3552, pruned_loss=0.107, over 5703886.33 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3646, pruned_loss=0.1064, over 5502624.79 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3528, pruned_loss=0.1062, over 5706966.96 frames. ], batch size: 128, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:34:27,761 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 02:34:33,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1642, 1.4596, 1.1718, 0.9744], device='cuda:0'), covar=tensor([0.1342, 0.1106, 0.0863, 0.1132], device='cuda:0'), in_proj_covar=tensor([0.1593, 0.1431, 0.1418, 0.1506], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 02:34:36,370 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2286, simple_loss=0.3328, pruned_loss=0.06223, over 944034.00 frames. 
+2023-03-04 02:34:36,371 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 02:34:52,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.063e+02 1.089e+03 1.374e+03 1.841e+03 4.340e+03, threshold=2.749e+03, percent-clipped=8.0
+2023-03-04 02:35:01,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.30 vs. limit=5.0
+2023-03-04 02:35:19,171 INFO [train.py:968] (0/2) Epoch 8, batch 6050, giga_loss[loss=0.295, simple_loss=0.3636, pruned_loss=0.1132, over 28909.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.357, pruned_loss=0.1077, over 5706309.51 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3646, pruned_loss=0.1064, over 5512778.29 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3548, pruned_loss=0.1071, over 5704545.82 frames. ], batch size: 145, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:36:02,891 INFO [train.py:968] (0/2) Epoch 8, batch 6100, giga_loss[loss=0.2907, simple_loss=0.3568, pruned_loss=0.1123, over 28859.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3617, pruned_loss=0.1124, over 5701790.45 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3641, pruned_loss=0.1063, over 5523018.65 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3602, pruned_loss=0.112, over 5696156.26 frames. ], batch size: 112, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:36:21,994 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:36:22,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.009e+02 1.345e+03 1.795e+03 2.408e+03 7.136e+03, threshold=3.590e+03, percent-clipped=15.0
+2023-03-04 02:36:23,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:36:51,600 INFO [train.py:968] (0/2) Epoch 8, batch 6150, giga_loss[loss=0.3549, simple_loss=0.4121, pruned_loss=0.1488, over 28644.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3692, pruned_loss=0.1188, over 5688838.77 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3642, pruned_loss=0.1065, over 5517619.31 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3679, pruned_loss=0.1185, over 5691068.67 frames. ], batch size: 336, lr: 4.15e-03, grad_scale: 2.0
+2023-03-04 02:36:51,866 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:37:40,403 INFO [train.py:968] (0/2) Epoch 8, batch 6200, libri_loss[loss=0.2976, simple_loss=0.381, pruned_loss=0.1071, over 29081.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3775, pruned_loss=0.1252, over 5688136.88 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3644, pruned_loss=0.1065, over 5523030.50 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3765, pruned_loss=0.1252, over 5687450.34 frames. ], batch size: 101, lr: 4.15e-03, grad_scale: 2.0
+2023-03-04 02:37:57,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.853e+02 1.633e+03 2.083e+03 3.041e+03 1.916e+04, threshold=4.166e+03, percent-clipped=18.0
+2023-03-04 02:38:25,049 INFO [train.py:968] (0/2) Epoch 8, batch 6250, giga_loss[loss=0.3227, simple_loss=0.3872, pruned_loss=0.1291, over 28846.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3808, pruned_loss=0.1284, over 5699421.86 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.364, pruned_loss=0.1066, over 5538294.66 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3809, pruned_loss=0.1292, over 5691718.35 frames. ], batch size: 199, lr: 4.15e-03, grad_scale: 2.0
+2023-03-04 02:39:02,584 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 02:39:12,599 INFO [train.py:968] (0/2) Epoch 8, batch 6300, giga_loss[loss=0.3539, simple_loss=0.4189, pruned_loss=0.1444, over 28355.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3879, pruned_loss=0.1345, over 5690168.02 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3641, pruned_loss=0.1066, over 5540277.65 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3882, pruned_loss=0.1356, over 5685207.75 frames. ], batch size: 71, lr: 4.15e-03, grad_scale: 2.0
+2023-03-04 02:39:16,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=325084.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 02:39:17,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3485, 3.4106, 1.4220, 1.4993], device='cuda:0'), covar=tensor([0.0906, 0.0362, 0.0888, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0497, 0.0321, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 02:39:33,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.606e+03 2.022e+03 2.436e+03 5.060e+03, threshold=4.043e+03, percent-clipped=3.0
+2023-03-04 02:40:00,814 INFO [train.py:968] (0/2) Epoch 8, batch 6350, giga_loss[loss=0.3427, simple_loss=0.3982, pruned_loss=0.1436, over 28759.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.393, pruned_loss=0.1393, over 5686622.12 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3643, pruned_loss=0.1068, over 5546802.73 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3936, pruned_loss=0.1405, over 5679325.20 frames. ], batch size: 284, lr: 4.15e-03, grad_scale: 2.0
+2023-03-04 02:40:56,976 INFO [train.py:968] (0/2) Epoch 8, batch 6400, giga_loss[loss=0.4281, simple_loss=0.4371, pruned_loss=0.2096, over 23510.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3956, pruned_loss=0.1428, over 5665481.50 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3644, pruned_loss=0.1068, over 5548687.72 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3961, pruned_loss=0.1439, over 5658787.06 frames. ], batch size: 705, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:41:17,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.126e+03 1.646e+03 2.093e+03 3.192e+03 8.001e+03, threshold=4.186e+03, percent-clipped=14.0
+2023-03-04 02:41:50,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=325227.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 02:41:51,756 INFO [train.py:968] (0/2) Epoch 8, batch 6450, giga_loss[loss=0.356, simple_loss=0.4047, pruned_loss=0.1537, over 28651.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3993, pruned_loss=0.1468, over 5668765.71 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3642, pruned_loss=0.1066, over 5553665.04 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.4003, pruned_loss=0.1483, over 5660820.29 frames. ], batch size: 307, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:41:53,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=325230.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 02:42:19,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4990, 1.6065, 1.2599, 1.2633], device='cuda:0'), covar=tensor([0.1324, 0.1291, 0.1071, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.1608, 0.1455, 0.1428, 0.1528], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 02:42:22,789 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=325259.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 02:42:25,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.1250, 5.9385, 5.5067, 3.2578], device='cuda:0'), covar=tensor([0.0413, 0.0636, 0.0880, 0.1473], device='cuda:0'), in_proj_covar=tensor([0.0956, 0.0902, 0.0800, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 02:42:43,719 INFO [train.py:968] (0/2) Epoch 8, batch 6500, giga_loss[loss=0.4987, simple_loss=0.4778, pruned_loss=0.2598, over 23432.00 frames. ], tot_loss[loss=0.3508, simple_loss=0.4017, pruned_loss=0.15, over 5655398.18 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.364, pruned_loss=0.1065, over 5567377.99 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.404, pruned_loss=0.1529, over 5640316.73 frames. ], batch size: 705, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:43:05,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.305e+02 1.525e+03 2.075e+03 3.353e+03 1.398e+04, threshold=4.150e+03, percent-clipped=12.0
+2023-03-04 02:43:33,576 INFO [train.py:968] (0/2) Epoch 8, batch 6550, giga_loss[loss=0.563, simple_loss=0.5283, pruned_loss=0.2988, over 26607.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4044, pruned_loss=0.1522, over 5652723.38 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3644, pruned_loss=0.1066, over 5574119.27 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4068, pruned_loss=0.1552, over 5636212.03 frames. ], batch size: 555, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:43:45,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5836, 1.7520, 1.8705, 1.4294], device='cuda:0'), covar=tensor([0.1408, 0.1913, 0.1093, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0708, 0.0810, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 02:43:55,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 02:44:24,614 INFO [train.py:968] (0/2) Epoch 8, batch 6600, giga_loss[loss=0.3713, simple_loss=0.4171, pruned_loss=0.1628, over 28915.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.4036, pruned_loss=0.1525, over 5651920.11 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3643, pruned_loss=0.1066, over 5578614.19 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.406, pruned_loss=0.1555, over 5635956.62 frames. ], batch size: 174, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:44:43,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.098e+02 1.863e+03 2.245e+03 3.153e+03 7.287e+03, threshold=4.491e+03, percent-clipped=10.0
+2023-03-04 02:45:18,041 INFO [train.py:968] (0/2) Epoch 8, batch 6650, giga_loss[loss=0.3954, simple_loss=0.4155, pruned_loss=0.1877, over 23563.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4014, pruned_loss=0.1514, over 5637675.65 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3642, pruned_loss=0.1066, over 5583667.59 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.4038, pruned_loss=0.1544, over 5621923.85 frames. ], batch size: 705, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:46:10,407 INFO [train.py:968] (0/2) Epoch 8, batch 6700, giga_loss[loss=0.3291, simple_loss=0.3962, pruned_loss=0.131, over 29076.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4008, pruned_loss=0.1494, over 5650539.49 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3638, pruned_loss=0.1064, over 5586924.51 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4035, pruned_loss=0.1524, over 5635616.18 frames. ], batch size: 155, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:46:32,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.543e+03 2.063e+03 2.939e+03 8.003e+03, threshold=4.125e+03, percent-clipped=7.0
+2023-03-04 02:47:03,341 INFO [train.py:968] (0/2) Epoch 8, batch 6750, giga_loss[loss=0.4211, simple_loss=0.4534, pruned_loss=0.1944, over 27465.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4031, pruned_loss=0.1505, over 5634362.20 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3641, pruned_loss=0.1066, over 5575936.79 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4052, pruned_loss=0.153, over 5633805.19 frames. ], batch size: 472, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:47:58,662 INFO [train.py:968] (0/2) Epoch 8, batch 6800, giga_loss[loss=0.3944, simple_loss=0.4198, pruned_loss=0.1845, over 26354.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4032, pruned_loss=0.1505, over 5625530.04 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.364, pruned_loss=0.1066, over 5577700.11 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.4051, pruned_loss=0.1526, over 5623764.67 frames. ], batch size: 555, lr: 4.14e-03, grad_scale: 8.0
+2023-03-04 02:48:15,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.265e+02 1.587e+03 2.161e+03 3.147e+03 8.369e+03, threshold=4.322e+03, percent-clipped=13.0
+2023-03-04 02:48:21,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1278, 1.4734, 1.4335, 1.0768], device='cuda:0'), covar=tensor([0.1412, 0.2031, 0.1157, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0711, 0.0811, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 02:48:50,819 INFO [train.py:968] (0/2) Epoch 8, batch 6850, giga_loss[loss=0.3379, simple_loss=0.4008, pruned_loss=0.1375, over 28917.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.4007, pruned_loss=0.1473, over 5628607.64 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3643, pruned_loss=0.1067, over 5584967.25 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4029, pruned_loss=0.1499, over 5621978.54 frames. ], batch size: 213, lr: 4.14e-03, grad_scale: 8.0
+2023-03-04 02:49:41,314 INFO [train.py:968] (0/2) Epoch 8, batch 6900, libri_loss[loss=0.3073, simple_loss=0.3823, pruned_loss=0.1162, over 29525.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.399, pruned_loss=0.1447, over 5641468.78 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3643, pruned_loss=0.1067, over 5590599.97 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4013, pruned_loss=0.1475, over 5632010.80 frames. ], batch size: 84, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:49:53,392 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=325693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:49:58,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.098e+02 1.487e+03 2.092e+03 2.597e+03 6.919e+03, threshold=4.184e+03, percent-clipped=9.0
+2023-03-04 02:50:29,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4077, 1.3104, 4.9568, 3.5051], device='cuda:0'), covar=tensor([0.1590, 0.2459, 0.0326, 0.0721], device='cuda:0'), in_proj_covar=tensor([0.0601, 0.0556, 0.0801, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:50:31,326 INFO [train.py:968] (0/2) Epoch 8, batch 6950, giga_loss[loss=0.306, simple_loss=0.3739, pruned_loss=0.119, over 28558.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3938, pruned_loss=0.1403, over 5643802.62 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3639, pruned_loss=0.1066, over 5584861.69 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3964, pruned_loss=0.143, over 5641901.16 frames. ], batch size: 71, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:51:14,375 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-04 02:51:18,467 INFO [train.py:968] (0/2) Epoch 8, batch 7000, giga_loss[loss=0.3094, simple_loss=0.3697, pruned_loss=0.1245, over 28914.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3904, pruned_loss=0.1375, over 5646026.87 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3637, pruned_loss=0.1065, over 5591322.20 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3936, pruned_loss=0.1407, over 5640522.30 frames. ], batch size: 199, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:51:24,844 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=325787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:51:38,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.636e+02 1.719e+03 2.460e+03 3.512e+03 1.290e+04, threshold=4.921e+03, percent-clipped=15.0
+2023-03-04 02:52:04,742 INFO [train.py:968] (0/2) Epoch 8, batch 7050, giga_loss[loss=0.3354, simple_loss=0.3941, pruned_loss=0.1383, over 28127.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3887, pruned_loss=0.1362, over 5643038.04 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3636, pruned_loss=0.1065, over 5590708.09 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.392, pruned_loss=0.1396, over 5639983.65 frames. ], batch size: 77, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:52:54,880 INFO [train.py:968] (0/2) Epoch 8, batch 7100, giga_loss[loss=0.3506, simple_loss=0.4021, pruned_loss=0.1495, over 28776.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3885, pruned_loss=0.1362, over 5642618.18 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3637, pruned_loss=0.1068, over 5593831.72 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3916, pruned_loss=0.1394, over 5638233.59 frames. ], batch size: 262, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:53:05,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6416, 1.6432, 1.6761, 1.5304], device='cuda:0'), covar=tensor([0.1193, 0.1692, 0.1579, 0.1591], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0731, 0.0649, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 02:53:17,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.410e+02 1.299e+03 1.646e+03 2.393e+03 4.590e+03, threshold=3.292e+03, percent-clipped=0.0
+2023-03-04 02:53:29,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4926, 3.7262, 1.5677, 1.4901], device='cuda:0'), covar=tensor([0.0915, 0.0293, 0.0866, 0.1361], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0503, 0.0324, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 02:53:50,497 INFO [train.py:968] (0/2) Epoch 8, batch 7150, giga_loss[loss=0.2894, simple_loss=0.3631, pruned_loss=0.1079, over 29092.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3857, pruned_loss=0.1334, over 5648418.40 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3635, pruned_loss=0.1066, over 5601354.15 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3888, pruned_loss=0.1367, over 5639436.24 frames. ], batch size: 128, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:54:21,515 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=325960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:54:45,866 INFO [train.py:968] (0/2) Epoch 8, batch 7200, libri_loss[loss=0.2805, simple_loss=0.3504, pruned_loss=0.1052, over 29506.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3844, pruned_loss=0.1303, over 5657434.82 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3632, pruned_loss=0.1064, over 5605172.66 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3877, pruned_loss=0.1337, over 5647831.40 frames. ], batch size: 84, lr: 4.14e-03, grad_scale: 8.0
+2023-03-04 02:55:08,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.801e+02 1.445e+03 2.006e+03 2.730e+03 9.166e+03, threshold=4.013e+03, percent-clipped=14.0
+2023-03-04 02:55:08,513 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-326000.pt
+2023-03-04 02:55:20,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1536, 2.5698, 1.2031, 1.3716], device='cuda:0'), covar=tensor([0.0899, 0.0317, 0.0888, 0.1309], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0502, 0.0323, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 02:55:22,081 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7785, 2.0744, 2.0847, 1.6059], device='cuda:0'), covar=tensor([0.1510, 0.1933, 0.1193, 0.1398], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0717, 0.0814, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 02:55:34,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1958, 1.4901, 1.1120, 1.2789], device='cuda:0'), covar=tensor([0.2269, 0.2172, 0.2417, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.1212, 0.0907, 0.1061, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:55:35,291 INFO [train.py:968] (0/2) Epoch 8, batch 7250, giga_loss[loss=0.3553, simple_loss=0.4143, pruned_loss=0.1482, over 27946.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3863, pruned_loss=0.1298, over 5666365.00 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3629, pruned_loss=0.1062, over 5609436.63 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3896, pruned_loss=0.133, over 5655961.04 frames. ], batch size: 412, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:55:59,198 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=326052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:56:16,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=326068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:56:29,652 INFO [train.py:968] (0/2) Epoch 8, batch 7300, giga_loss[loss=0.3768, simple_loss=0.418, pruned_loss=0.1678, over 27629.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3891, pruned_loss=0.1323, over 5655902.17 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3631, pruned_loss=0.1063, over 5605143.41 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3919, pruned_loss=0.1353, over 5652385.66 frames. ], batch size: 474, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 02:56:50,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.450e+02 1.657e+03 2.260e+03 3.435e+03 6.346e+03, threshold=4.520e+03, percent-clipped=13.0
+2023-03-04 02:57:15,104 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-04 02:57:16,607 INFO [train.py:968] (0/2) Epoch 8, batch 7350, giga_loss[loss=0.2951, simple_loss=0.363, pruned_loss=0.1136, over 28887.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3888, pruned_loss=0.133, over 5654422.62 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3632, pruned_loss=0.1064, over 5607307.97 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3915, pruned_loss=0.1359, over 5651124.78 frames. ], batch size: 112, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 02:57:44,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=326162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:58:06,556 INFO [train.py:968] (0/2) Epoch 8, batch 7400, giga_loss[loss=0.281, simple_loss=0.3494, pruned_loss=0.1062, over 28938.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3863, pruned_loss=0.1317, over 5668736.81 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3632, pruned_loss=0.1063, over 5617934.26 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3895, pruned_loss=0.1352, over 5658575.88 frames. ], batch size: 199, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 02:58:16,655 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-04 02:58:21,827 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-04 02:58:25,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.647e+03 2.247e+03 2.998e+03 7.791e+03, threshold=4.493e+03, percent-clipped=8.0
+2023-03-04 02:58:35,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=326211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:58:37,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=326214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:58:42,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1381, 1.2057, 3.9313, 3.1932], device='cuda:0'), covar=tensor([0.1622, 0.2383, 0.0412, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0602, 0.0557, 0.0802, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 02:58:48,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2997, 2.7465, 1.4341, 1.3941], device='cuda:0'), covar=tensor([0.0823, 0.0334, 0.0742, 0.1126], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0500, 0.0322, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 02:58:53,101 INFO [train.py:968] (0/2) Epoch 8, batch 7450, giga_loss[loss=0.3278, simple_loss=0.3848, pruned_loss=0.1355, over 28930.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3853, pruned_loss=0.1325, over 5665370.39 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.363, pruned_loss=0.1062, over 5620160.48 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3883, pruned_loss=0.1356, over 5656122.24 frames. ], batch size: 227, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 02:59:05,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=326243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:59:18,745 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-04 02:59:40,846 INFO [train.py:968] (0/2) Epoch 8, batch 7500, giga_loss[loss=0.3723, simple_loss=0.4292, pruned_loss=0.1577, over 28950.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3843, pruned_loss=0.1314, over 5664096.69 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3633, pruned_loss=0.1061, over 5624482.64 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.387, pruned_loss=0.1347, over 5654279.06 frames. ], batch size: 227, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 02:59:48,330 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=326287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:59:53,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5621, 1.9779, 1.3182, 0.8880], device='cuda:0'), covar=tensor([0.3671, 0.2243, 0.1907, 0.3681], device='cuda:0'), in_proj_covar=tensor([0.1481, 0.1389, 0.1441, 0.1209], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 03:00:04,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.554e+02 1.482e+03 1.861e+03 2.347e+03 7.911e+03, threshold=3.723e+03, percent-clipped=4.0
+2023-03-04 03:00:09,015 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=326305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:00:13,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=326308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:00:30,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0567, 1.6291, 1.4009, 1.0319], device='cuda:0'), covar=tensor([0.1425, 0.2182, 0.1264, 0.1405], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0713, 0.0813, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 03:00:30,821 INFO [train.py:968] (0/2) Epoch 8, batch 7550, libri_loss[loss=0.308, simple_loss=0.3809, pruned_loss=0.1175, over 29540.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3842, pruned_loss=0.1302, over 5669263.97 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3636, pruned_loss=0.1063, over 5629999.50 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3865, pruned_loss=0.1331, over 5656847.04 frames. ], batch size: 83, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 03:00:35,794 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=326335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:00:38,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=326337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:00:50,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2240, 1.4490, 1.1900, 0.9803], device='cuda:0'), covar=tensor([0.1395, 0.1306, 0.0955, 0.1346], device='cuda:0'), in_proj_covar=tensor([0.1590, 0.1439, 0.1421, 0.1524], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 03:01:18,717 INFO [train.py:968] (0/2) Epoch 8, batch 7600, giga_loss[loss=0.3091, simple_loss=0.3822, pruned_loss=0.118, over 28921.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3854, pruned_loss=0.1307, over 5671105.11 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3641, pruned_loss=0.1067, over 5633810.66 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.387, pruned_loss=0.133, over 5658376.42 frames. ], batch size: 145, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:01:22,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5232, 1.6107, 1.7984, 1.4334], device='cuda:0'), covar=tensor([0.1316, 0.1846, 0.1068, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0715, 0.0816, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:0')
+2023-03-04 03:01:39,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.742e+02 1.396e+03 1.760e+03 2.461e+03 7.201e+03, threshold=3.519e+03, percent-clipped=7.0
+2023-03-04 03:02:00,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=326427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:02:01,423 INFO [train.py:968] (0/2) Epoch 8, batch 7650, giga_loss[loss=0.3466, simple_loss=0.3948, pruned_loss=0.1492, over 28791.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3833, pruned_loss=0.1288, over 5685570.07 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3643, pruned_loss=0.1068, over 5637735.66 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3851, pruned_loss=0.1312, over 5673127.06 frames. ], batch size: 99, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:02:23,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3891, 1.7278, 1.4970, 1.6272], device='cuda:0'), covar=tensor([0.0614, 0.0262, 0.0255, 0.0579], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0120, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0070, 0.0051, 0.0046, 0.0076], device='cuda:0')
+2023-03-04 03:02:39,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9229, 1.0834, 3.6157, 2.9299], device='cuda:0'), covar=tensor([0.1640, 0.2452, 0.0416, 0.0929], device='cuda:0'), in_proj_covar=tensor([0.0598, 0.0556, 0.0801, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 03:02:52,074 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=326478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:02:53,295 INFO [train.py:968] (0/2) Epoch 8, batch 7700, giga_loss[loss=0.2641, simple_loss=0.3306, pruned_loss=0.09875, over 28631.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3807, pruned_loss=0.1279, over 5678362.52 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3641, pruned_loss=0.1068, over 5640848.56 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3824, pruned_loss=0.1301, over 5666282.53 frames. ], batch size: 60, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:02:56,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=326481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:03:20,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.714e+02 1.536e+03 1.977e+03 3.066e+03 7.324e+03, threshold=3.954e+03, percent-clipped=19.0
+2023-03-04 03:03:27,204 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=326510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:03:42,496 INFO [train.py:968] (0/2) Epoch 8, batch 7750, giga_loss[loss=0.2903, simple_loss=0.3629, pruned_loss=0.1089, over 28724.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.379, pruned_loss=0.1273, over 5676402.90 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3638, pruned_loss=0.1063, over 5648712.38 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3813, pruned_loss=0.1302, over 5659898.80 frames. ], batch size: 262, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:03:59,439 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-04 03:04:24,472 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=326570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:04:26,891 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=326573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:04:32,712 INFO [train.py:968] (0/2) Epoch 8, batch 7800, giga_loss[loss=0.2957, simple_loss=0.3626, pruned_loss=0.1144, over 28759.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3793, pruned_loss=0.1288, over 5661270.36 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3637, pruned_loss=0.1063, over 5644652.07 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3813, pruned_loss=0.1315, over 5652030.74 frames. ], batch size: 243, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:04:57,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.627e+02 1.502e+03 2.026e+03 2.927e+03 5.586e+03, threshold=4.051e+03, percent-clipped=8.0
+2023-03-04 03:04:57,279 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=326602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:05:09,826 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6406, 4.4728, 4.2467, 1.8922], device='cuda:0'), covar=tensor([0.0532, 0.0689, 0.0749, 0.2054], device='cuda:0'), in_proj_covar=tensor([0.0952, 0.0904, 0.0799, 0.0628], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 03:05:14,541 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=326620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:05:23,815 INFO [train.py:968] (0/2) Epoch 8, batch 7850, libri_loss[loss=0.2727, simple_loss=0.3473, pruned_loss=0.09906, over 29599.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3787, pruned_loss=0.1294, over 5660937.16 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3637, pruned_loss=0.1063, over 5647352.89 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3805, pruned_loss=0.1317, over 5651243.67 frames. ], batch size: 74, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:05:55,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=326662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:06:08,071 INFO [train.py:968] (0/2) Epoch 8, batch 7900, libri_loss[loss=0.2477, simple_loss=0.326, pruned_loss=0.08473, over 29511.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.378, pruned_loss=0.129, over 5661461.31 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3635, pruned_loss=0.106, over 5654133.09 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3801, pruned_loss=0.1319, over 5647354.10 frames. ], batch size: 70, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:06:29,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.771e+02 1.502e+03 2.025e+03 2.746e+03 7.500e+03, threshold=4.049e+03, percent-clipped=10.0
+2023-03-04 03:06:53,217 INFO [train.py:968] (0/2) Epoch 8, batch 7950, giga_loss[loss=0.3246, simple_loss=0.3861, pruned_loss=0.1316, over 28305.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3788, pruned_loss=0.1291, over 5660305.12 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3637, pruned_loss=0.1064, over 5647105.42 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3807, pruned_loss=0.1318, over 5656264.25 frames. ], batch size: 368, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:07:41,747 INFO [train.py:968] (0/2) Epoch 8, batch 8000, giga_loss[loss=0.2952, simple_loss=0.3727, pruned_loss=0.1089, over 28721.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3807, pruned_loss=0.1299, over 5660813.52 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3641, pruned_loss=0.1067, over 5651126.86 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3822, pruned_loss=0.132, over 5654257.18 frames. ], batch size: 119, lr: 4.14e-03, grad_scale: 8.0
+2023-03-04 03:08:07,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.000e+02 1.825e+03 2.496e+03 3.421e+03 9.192e+03, threshold=4.992e+03, percent-clipped=17.0
+2023-03-04 03:08:09,036 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=326805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:08:11,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=326808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:08:31,071 INFO [train.py:968] (0/2) Epoch 8, batch 8050, giga_loss[loss=0.3226, simple_loss=0.3852, pruned_loss=0.13, over 28774.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3807, pruned_loss=0.1291, over 5673736.22 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3639, pruned_loss=0.1066, over 5653660.35 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3821, pruned_loss=0.131, over 5666520.66 frames. ], batch size: 242, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:08:36,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=326837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:09:16,863 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7980, 2.0570, 2.0208, 1.6021], device='cuda:0'), covar=tensor([0.1487, 0.1947, 0.1182, 0.1396], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0720, 0.0819, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 03:09:19,024 INFO [train.py:968] (0/2) Epoch 8, batch 8100, giga_loss[loss=0.3046, simple_loss=0.3697, pruned_loss=0.1197, over 28553.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3806, pruned_loss=0.1284, over 5680826.06 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3642, pruned_loss=0.1067, over 5658367.24 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3819, pruned_loss=0.1303, over 5671437.34 frames. ], batch size: 85, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:09:42,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.075e+02 1.444e+03 1.844e+03 2.484e+03 8.346e+03, threshold=3.688e+03, percent-clipped=1.0
+2023-03-04 03:10:04,149 INFO [train.py:968] (0/2) Epoch 8, batch 8150, giga_loss[loss=0.3904, simple_loss=0.4031, pruned_loss=0.1888, over 23447.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3823, pruned_loss=0.1301, over 5680709.19 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3647, pruned_loss=0.107, over 5662075.84 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3832, pruned_loss=0.1319, over 5670660.88 frames. ], batch size: 705, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:10:15,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3883, 1.8012, 1.2403, 0.7585], device='cuda:0'), covar=tensor([0.2963, 0.2198, 0.1751, 0.3041], device='cuda:0'), in_proj_covar=tensor([0.1468, 0.1386, 0.1439, 0.1203], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 03:10:53,688 INFO [train.py:968] (0/2) Epoch 8, batch 8200, giga_loss[loss=0.3309, simple_loss=0.3841, pruned_loss=0.1388, over 28980.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3844, pruned_loss=0.1324, over 5678107.66 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3649, pruned_loss=0.1071, over 5669236.15 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3855, pruned_loss=0.1347, over 5664069.52 frames. ], batch size: 106, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:11:13,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=326995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:11:21,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.509e+02 1.699e+03 2.365e+03 2.983e+03 7.613e+03, threshold=4.730e+03, percent-clipped=7.0
+2023-03-04 03:11:33,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5463, 1.7906, 1.5162, 1.2132], device='cuda:0'), covar=tensor([0.1688, 0.1415, 0.1092, 0.1599], device='cuda:0'), in_proj_covar=tensor([0.1590, 0.1457, 0.1418, 0.1534], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 03:11:46,558 INFO [train.py:968] (0/2) Epoch 8, batch 8250, giga_loss[loss=0.3569, simple_loss=0.4015, pruned_loss=0.1562, over 27948.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3856, pruned_loss=0.1355, over 5667358.78 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3644, pruned_loss=0.1068, over 5675062.00 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3875, pruned_loss=0.1382, over 5650813.85 frames. ], batch size: 412, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:12:37,721 INFO [train.py:968] (0/2) Epoch 8, batch 8300, giga_loss[loss=0.4368, simple_loss=0.4608, pruned_loss=0.2064, over 24133.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3866, pruned_loss=0.1369, over 5672136.67 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3647, pruned_loss=0.107, over 5680508.44 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3883, pruned_loss=0.1396, over 5654159.05 frames. ], batch size: 705, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:12:58,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.640e+03 2.201e+03 3.406e+03 1.299e+04, threshold=4.402e+03, percent-clipped=11.0
+2023-03-04 03:13:24,539 INFO [train.py:968] (0/2) Epoch 8, batch 8350, giga_loss[loss=0.3542, simple_loss=0.408, pruned_loss=0.1503, over 28605.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3861, pruned_loss=0.1366, over 5665991.81 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3642, pruned_loss=0.1068, over 5679524.57 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3886, pruned_loss=0.1399, over 5652593.37 frames. ], batch size: 336, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:13:33,548 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=327138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:13:35,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=327141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:14:02,757 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=327170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:14:08,456 INFO [train.py:968] (0/2) Epoch 8, batch 8400, giga_loss[loss=0.3383, simple_loss=0.3913, pruned_loss=0.1427, over 28972.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.384, pruned_loss=0.1345, over 5674538.13 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3642, pruned_loss=0.1067, over 5681920.56 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3868, pruned_loss=0.1382, over 5661231.61 frames. ], batch size: 199, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:14:27,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5881, 1.3281, 4.7977, 3.4956], device='cuda:0'), covar=tensor([0.2043, 0.2938, 0.0652, 0.0816], device='cuda:0'), in_proj_covar=tensor([0.0604, 0.0560, 0.0806, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 03:14:29,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.661e+03 2.071e+03 2.901e+03 7.816e+03, threshold=4.142e+03, percent-clipped=8.0
+2023-03-04 03:14:55,252 INFO [train.py:968] (0/2) Epoch 8, batch 8450, giga_loss[loss=0.2905, simple_loss=0.3729, pruned_loss=0.104, over 28862.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3828, pruned_loss=0.1316, over 5682287.19 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3641, pruned_loss=0.1066, over 5681778.26 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3853, pruned_loss=0.1349, over 5671953.47 frames. ], batch size: 174, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:14:59,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2346, 1.2459, 1.1509, 1.4621], device='cuda:0'), covar=tensor([0.0753, 0.0369, 0.0332, 0.0781], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0119, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0070, 0.0050, 0.0045, 0.0076], device='cuda:0')
+2023-03-04 03:15:07,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3587, 2.9225, 1.5056, 1.4607], device='cuda:0'), covar=tensor([0.0864, 0.0296, 0.0805, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0502, 0.0322, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:0')
+2023-03-04 03:15:41,924 INFO [train.py:968] (0/2) Epoch 8, batch 8500, giga_loss[loss=0.3078, simple_loss=0.3615, pruned_loss=0.127, over 27624.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3803, pruned_loss=0.1296, over 5678968.39 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3639, pruned_loss=0.1065, over 5685410.22 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3827, pruned_loss=0.1327, over 5667722.69 frames. ], batch size: 472, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:15:59,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.910e+02 1.678e+03 2.166e+03 3.141e+03 1.259e+04, threshold=4.331e+03, percent-clipped=14.0
+2023-03-04 03:16:04,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2555, 1.3695, 1.4584, 1.2775], device='cuda:0'), covar=tensor([0.1135, 0.1356, 0.1591, 0.1264], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0730, 0.0653, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 03:16:24,760 INFO [train.py:968] (0/2) Epoch 8, batch 8550, giga_loss[loss=0.3532, simple_loss=0.377, pruned_loss=0.1647, over 23620.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3794, pruned_loss=0.1297, over 5679347.48 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3638, pruned_loss=0.1064, over 5695316.79 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3821, pruned_loss=0.1333, over 5661005.43 frames. ], batch size: 705, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:17:12,907 INFO [train.py:968] (0/2) Epoch 8, batch 8600, giga_loss[loss=0.3222, simple_loss=0.3855, pruned_loss=0.1294, over 28904.00 frames. ], tot_loss[loss=0.321, simple_loss=0.38, pruned_loss=0.131, over 5684805.14 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3641, pruned_loss=0.1065, over 5694953.64 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3821, pruned_loss=0.134, over 5670466.28 frames. ], batch size: 174, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:17:14,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5150, 2.0375, 1.4027, 0.8515], device='cuda:0'), covar=tensor([0.2713, 0.1590, 0.1421, 0.2523], device='cuda:0'), in_proj_covar=tensor([0.1468, 0.1397, 0.1439, 0.1210], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 03:17:32,419 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=327398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:17:39,679 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.212e+02 1.634e+03 2.044e+03 3.240e+03 7.923e+03, threshold=4.087e+03, percent-clipped=14.0
+2023-03-04 03:18:05,310 INFO [train.py:968] (0/2) Epoch 8, batch 8650, giga_loss[loss=0.3546, simple_loss=0.4197, pruned_loss=0.1448, over 28797.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3812, pruned_loss=0.1319, over 5680017.01 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3639, pruned_loss=0.1064, over 5696806.32 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3834, pruned_loss=0.135, over 5666678.19 frames. ], batch size: 174, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:18:33,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6072, 1.8097, 1.9805, 1.4888], device='cuda:0'), covar=tensor([0.1569, 0.1976, 0.1168, 0.1404], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0717, 0.0818, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 03:18:44,533 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-04 03:18:51,222 INFO [train.py:968] (0/2) Epoch 8, batch 8700, libri_loss[loss=0.3509, simple_loss=0.4229, pruned_loss=0.1395, over 26234.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3836, pruned_loss=0.1324, over 5680728.04 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3639, pruned_loss=0.1064, over 5700414.05 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.386, pruned_loss=0.1356, over 5666529.21 frames. ], batch size: 136, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:19:18,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.396e+02 1.595e+03 2.275e+03 3.139e+03 1.166e+04, threshold=4.551e+03, percent-clipped=15.0
+2023-03-04 03:19:41,410 INFO [train.py:968] (0/2) Epoch 8, batch 8750, giga_loss[loss=0.314, simple_loss=0.3866, pruned_loss=0.1207, over 28756.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.387, pruned_loss=0.1319, over 5681367.46 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3638, pruned_loss=0.1064, over 5702476.63 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3891, pruned_loss=0.1347, over 5668066.05 frames. ], batch size: 243, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:19:54,187 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=327540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:20:23,736 INFO [train.py:968] (0/2) Epoch 8, batch 8800, giga_loss[loss=0.3674, simple_loss=0.4145, pruned_loss=0.1602, over 29056.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3857, pruned_loss=0.1301, over 5691924.82 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3628, pruned_loss=0.1059, over 5709359.10 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3896, pruned_loss=0.1342, over 5673227.88 frames. ], batch size: 136, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:20:48,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.600e+02 1.283e+03 1.755e+03 2.474e+03 4.171e+03, threshold=3.510e+03, percent-clipped=0.0
+2023-03-04 03:21:08,595 INFO [train.py:968] (0/2) Epoch 8, batch 8850, libri_loss[loss=0.2451, simple_loss=0.3168, pruned_loss=0.08669, over 29369.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3875, pruned_loss=0.1324, over 5694173.94 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3625, pruned_loss=0.1059, over 5712535.34 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3914, pruned_loss=0.1362, over 5675945.95 frames. ], batch size: 67, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:21:54,293 INFO [train.py:968] (0/2) Epoch 8, batch 8900, giga_loss[loss=0.294, simple_loss=0.3641, pruned_loss=0.1119, over 28802.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3889, pruned_loss=0.134, over 5686976.75 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3627, pruned_loss=0.1061, over 5706134.02 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3923, pruned_loss=0.1373, over 5677791.58 frames. ], batch size: 92, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:22:19,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.669e+02 1.730e+03 2.180e+03 2.938e+03 7.402e+03, threshold=4.360e+03, percent-clipped=11.0
+2023-03-04 03:22:41,384 INFO [train.py:968] (0/2) Epoch 8, batch 8950, giga_loss[loss=0.3971, simple_loss=0.4347, pruned_loss=0.1798, over 28714.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3902, pruned_loss=0.1364, over 5687091.64 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3627, pruned_loss=0.1061, over 5709319.43 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3934, pruned_loss=0.1395, over 5676549.97 frames. ], batch size: 262, lr: 4.13e-03, grad_scale: 2.0
+2023-03-04 03:22:48,101 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-04 03:23:25,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=327773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:23:30,975 INFO [train.py:968] (0/2) Epoch 8, batch 9000, giga_loss[loss=0.3102, simple_loss=0.3709, pruned_loss=0.1248, over 28995.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3893, pruned_loss=0.1366, over 5686175.00 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3628, pruned_loss=0.1062, over 5711514.71 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3926, pruned_loss=0.1399, over 5675377.44 frames. ], batch size: 128, lr: 4.13e-03, grad_scale: 2.0
+2023-03-04 03:23:30,979 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 03:23:39,747 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2244, simple_loss=0.3301, pruned_loss=0.05937, over 944034.00 frames. 
+2023-03-04 03:23:39,748 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 03:23:46,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3456, 1.5661, 1.4608, 1.1197], device='cuda:0'), covar=tensor([0.1877, 0.3141, 0.1612, 0.1620], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0721, 0.0823, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 03:23:55,018 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-04 03:23:59,994 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.835e+02 1.556e+03 1.951e+03 2.877e+03 8.968e+03, threshold=3.903e+03, percent-clipped=6.0
+2023-03-04 03:24:10,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5723, 1.9579, 1.9038, 1.4414], device='cuda:0'), covar=tensor([0.1561, 0.1957, 0.1185, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0720, 0.0822, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 03:24:24,822 INFO [train.py:968] (0/2) Epoch 8, batch 9050, giga_loss[loss=0.3197, simple_loss=0.3821, pruned_loss=0.1287, over 28955.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3875, pruned_loss=0.1362, over 5675092.17 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3628, pruned_loss=0.1062, over 5708294.22 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3907, pruned_loss=0.1398, over 5669427.11 frames. ], batch size: 136, lr: 4.13e-03, grad_scale: 2.0
+2023-03-04 03:24:38,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4315, 3.6502, 1.6141, 1.3598], device='cuda:0'), covar=tensor([0.0890, 0.0337, 0.0822, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0501, 0.0323, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 03:24:47,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1635, 1.3327, 1.1092, 1.0150], device='cuda:0'), covar=tensor([0.1373, 0.1332, 0.0915, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.1588, 0.1438, 0.1422, 0.1531], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 03:25:12,559 INFO [train.py:968] (0/2) Epoch 8, batch 9100, giga_loss[loss=0.3246, simple_loss=0.3853, pruned_loss=0.1319, over 28983.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3858, pruned_loss=0.1352, over 5677592.52 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.363, pruned_loss=0.1063, over 5711949.25 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3887, pruned_loss=0.1385, over 5669154.99 frames. ], batch size: 227, lr: 4.13e-03, grad_scale: 2.0
+2023-03-04 03:25:42,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.701e+02 1.607e+03 2.046e+03 2.792e+03 7.652e+03, threshold=4.091e+03, percent-clipped=9.0
+2023-03-04 03:25:43,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=327907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:25:52,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=327915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:25:52,817 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=327916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:25:55,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=327919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:26:05,140 INFO [train.py:968] (0/2) Epoch 8, batch 9150, giga_loss[loss=0.3431, simple_loss=0.3975, pruned_loss=0.1443, over 28799.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3866, pruned_loss=0.1362, over 5679469.81 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3628, pruned_loss=0.1062, over 5716065.08 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3894, pruned_loss=0.1394, over 5668514.26 frames. ], batch size: 284, lr: 4.13e-03, grad_scale: 2.0
+2023-03-04 03:26:23,686 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=327948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:26:30,783 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=327955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:26:53,763 INFO [train.py:968] (0/2) Epoch 8, batch 9200, libri_loss[loss=0.3311, simple_loss=0.3966, pruned_loss=0.1328, over 29750.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3852, pruned_loss=0.1354, over 5675754.16 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3632, pruned_loss=0.1064, over 5711912.82 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3876, pruned_loss=0.1385, over 5669371.17 frames. ], batch size: 87, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:27:15,855 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-328000.pt
+2023-03-04 03:27:21,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.612e+02 1.515e+03 1.884e+03 2.658e+03 1.614e+04, threshold=3.768e+03, percent-clipped=12.0
+2023-03-04 03:27:43,950 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-04 03:27:44,809 INFO [train.py:968] (0/2) Epoch 8, batch 9250, giga_loss[loss=0.3517, simple_loss=0.411, pruned_loss=0.1462, over 28246.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3833, pruned_loss=0.1342, over 5678863.67 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3628, pruned_loss=0.1062, over 5715192.50 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3859, pruned_loss=0.1373, over 5670238.55 frames. ], batch size: 368, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:28:14,513 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:28:17,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:28:30,159 INFO [train.py:968] (0/2) Epoch 8, batch 9300, giga_loss[loss=0.3479, simple_loss=0.3981, pruned_loss=0.1489, over 28013.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3836, pruned_loss=0.1341, over 5672805.80 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3632, pruned_loss=0.1064, over 5699822.42 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3857, pruned_loss=0.137, over 5678658.26 frames. ], batch size: 412, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:28:34,462 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-04 03:28:41,632 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:28:51,777 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 03:28:56,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1871, 1.6578, 1.3666, 1.5149], device='cuda:0'), covar=tensor([0.0750, 0.0320, 0.0316, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0119, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0070, 0.0051, 0.0046, 0.0077], device='cuda:0')
+2023-03-04 03:28:59,239 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.316e+02 1.546e+03 2.027e+03 3.113e+03 7.861e+03, threshold=4.054e+03, percent-clipped=15.0
+2023-03-04 03:29:13,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-04 03:29:16,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6892, 4.5453, 4.2627, 2.0684], device='cuda:0'), covar=tensor([0.0449, 0.0587, 0.0700, 0.1844], device='cuda:0'), in_proj_covar=tensor([0.0974, 0.0924, 0.0819, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 03:29:20,487 INFO [train.py:968] (0/2) Epoch 8, batch 9350, giga_loss[loss=0.3652, simple_loss=0.4181, pruned_loss=0.1561, over 28910.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3847, pruned_loss=0.1343, over 5670618.71 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3625, pruned_loss=0.106, over 5703085.12 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3878, pruned_loss=0.1379, over 5671127.79 frames. ], batch size: 186, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:29:33,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1386, 2.1823, 1.9656, 1.8752], device='cuda:0'), covar=tensor([0.1201, 0.1897, 0.1583, 0.1639], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0732, 0.0654, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 03:30:08,099 INFO [train.py:968] (0/2) Epoch 8, batch 9400, libri_loss[loss=0.2961, simple_loss=0.378, pruned_loss=0.1071, over 29227.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3863, pruned_loss=0.1356, over 5667692.50 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3625, pruned_loss=0.106, over 5706955.87 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3894, pruned_loss=0.1393, over 5663499.75 frames. ], batch size: 97, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:30:15,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4671, 2.0914, 1.6091, 0.7246], device='cuda:0'), covar=tensor([0.2923, 0.1585, 0.2021, 0.3317], device='cuda:0'), in_proj_covar=tensor([0.1466, 0.1389, 0.1433, 0.1203], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 03:30:25,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6641, 2.2860, 1.5439, 1.4330], device='cuda:0'), covar=tensor([0.1896, 0.1296, 0.1523, 0.1729], device='cuda:0'), in_proj_covar=tensor([0.1583, 0.1443, 0.1427, 0.1528], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 03:30:31,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.819e+02 1.532e+03 2.091e+03 2.770e+03 5.066e+03, threshold=4.182e+03, percent-clipped=7.0
+2023-03-04 03:30:53,188 INFO [train.py:968] (0/2) Epoch 8, batch 9450, giga_loss[loss=0.2962, simple_loss=0.3619, pruned_loss=0.1152, over 28370.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3848, pruned_loss=0.1344, over 5673003.16 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3621, pruned_loss=0.1057, over 5710109.33 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3882, pruned_loss=0.1384, over 5665735.74 frames. ], batch size: 71, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:31:28,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9563, 1.1613, 1.0829, 0.9306], device='cuda:0'), covar=tensor([0.1330, 0.1488, 0.0816, 0.1145], device='cuda:0'), in_proj_covar=tensor([0.1574, 0.1427, 0.1414, 0.1518], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 03:31:40,035 INFO [train.py:968] (0/2) Epoch 8, batch 9500, giga_loss[loss=0.3003, simple_loss=0.3802, pruned_loss=0.1102, over 29042.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3858, pruned_loss=0.1317, over 5684660.85 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.362, pruned_loss=0.1056, over 5712959.50 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.389, pruned_loss=0.1354, over 5675715.88 frames. ], batch size: 155, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:31:44,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:32:06,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.725e+02 1.322e+03 1.816e+03 2.235e+03 5.821e+03, threshold=3.631e+03, percent-clipped=2.0
+2023-03-04 03:32:12,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7452, 2.5923, 1.7207, 1.5674], device='cuda:0'), covar=tensor([0.2142, 0.1179, 0.1405, 0.1617], device='cuda:0'), in_proj_covar=tensor([0.1576, 0.1427, 0.1412, 0.1514], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 03:32:26,093 INFO [train.py:968] (0/2) Epoch 8, batch 9550, giga_loss[loss=0.3521, simple_loss=0.3996, pruned_loss=0.1523, over 28266.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3875, pruned_loss=0.131, over 5683666.66 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3621, pruned_loss=0.1057, over 5717095.39 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3904, pruned_loss=0.1343, over 5672129.22 frames. ], batch size: 368, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:32:27,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:32:41,461 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-04 03:33:15,020 INFO [train.py:968] (0/2) Epoch 8, batch 9600, giga_loss[loss=0.3509, simple_loss=0.4072, pruned_loss=0.1473, over 28985.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3908, pruned_loss=0.1336, over 5684568.34 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3617, pruned_loss=0.1054, over 5722078.07 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3941, pruned_loss=0.1371, over 5670469.84 frames. ], batch size: 213, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:33:20,002 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=328383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:33:26,561 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=328389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:33:31,536 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=328395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:33:41,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.444e+02 1.438e+03 1.927e+03 2.891e+03 6.950e+03, threshold=3.854e+03, percent-clipped=14.0
+2023-03-04 03:33:58,534 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=328423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:33:59,923 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:02,891 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:03,264 INFO [train.py:968] (0/2) Epoch 8, batch 9650, giga_loss[loss=0.3616, simple_loss=0.416, pruned_loss=0.1536, over 28745.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3922, pruned_loss=0.1357, over 5684331.14 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3615, pruned_loss=0.1053, over 5721629.68 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3951, pruned_loss=0.1386, over 5673539.42 frames. ], batch size: 199, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:34:18,214 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.31 vs. limit=5.0
+2023-03-04 03:34:26,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5707, 0.9480, 2.8590, 2.6443], device='cuda:0'), covar=tensor([0.1689, 0.2310, 0.0553, 0.0749], device='cuda:0'), in_proj_covar=tensor([0.0611, 0.0568, 0.0816, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 03:34:30,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:46,310 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:49,423 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:51,411 INFO [train.py:968] (0/2) Epoch 8, batch 9700, giga_loss[loss=0.4037, simple_loss=0.4204, pruned_loss=0.1935, over 23333.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3926, pruned_loss=0.1372, over 5667113.13 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3616, pruned_loss=0.1056, over 5713071.82 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3958, pruned_loss=0.1403, over 5664270.44 frames. ], batch size: 705, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:35:15,637 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-04 03:35:18,270 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:35:18,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.651e+02 1.620e+03 2.160e+03 2.809e+03 6.616e+03, threshold=4.320e+03, percent-clipped=10.0
+2023-03-04 03:35:38,892 INFO [train.py:968] (0/2) Epoch 8, batch 9750, giga_loss[loss=0.3333, simple_loss=0.3837, pruned_loss=0.1414, over 28866.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3909, pruned_loss=0.1361, over 5654328.06 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3615, pruned_loss=0.1054, over 5708511.00 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3945, pruned_loss=0.1398, over 5654662.22 frames. ], batch size: 112, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:36:02,887 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 03:36:12,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3507, 2.1777, 1.5211, 0.5571], device='cuda:0'), covar=tensor([0.3137, 0.1549, 0.2258, 0.2946], device='cuda:0'), in_proj_covar=tensor([0.1477, 0.1394, 0.1442, 0.1209], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 03:36:25,091 INFO [train.py:968] (0/2) Epoch 8, batch 9800, giga_loss[loss=0.2613, simple_loss=0.3515, pruned_loss=0.08548, over 28850.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3892, pruned_loss=0.1337, over 5660376.44 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3614, pruned_loss=0.1054, over 5710152.10 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3926, pruned_loss=0.1371, over 5658354.53 frames. ], batch size: 119, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:36:52,404 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.963e+02 1.295e+03 1.766e+03 2.760e+03 9.021e+03, threshold=3.531e+03, percent-clipped=7.0
+2023-03-04 03:36:54,345 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-04 03:36:54,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3387, 2.1493, 1.5087, 0.5463], device='cuda:0'), covar=tensor([0.3220, 0.1634, 0.2554, 0.3732], device='cuda:0'), in_proj_covar=tensor([0.1479, 0.1399, 0.1445, 0.1211], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 03:37:12,147 INFO [train.py:968] (0/2) Epoch 8, batch 9850, giga_loss[loss=0.2982, simple_loss=0.3821, pruned_loss=0.1071, over 29053.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3901, pruned_loss=0.1328, over 5663048.40 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.361, pruned_loss=0.1052, over 5712201.63 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3933, pruned_loss=0.1359, over 5659232.08 frames. ], batch size: 155, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:37:36,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8640, 1.0873, 3.2664, 2.9005], device='cuda:0'), covar=tensor([0.1650, 0.2404, 0.0507, 0.1044], device='cuda:0'), in_proj_covar=tensor([0.0610, 0.0567, 0.0814, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 03:37:43,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0123, 1.2756, 1.2349, 1.1514], device='cuda:0'), covar=tensor([0.1234, 0.1072, 0.1757, 0.1273], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0726, 0.0649, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 03:37:50,854 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-04 03:37:52,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5744, 1.7264, 1.5453, 1.5334], device='cuda:0'), covar=tensor([0.1350, 0.1869, 0.1729, 0.1689], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0725, 0.0649, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 03:38:00,398 INFO [train.py:968] (0/2) Epoch 8, batch 9900, libri_loss[loss=0.2786, simple_loss=0.3608, pruned_loss=0.09817, over 29394.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3905, pruned_loss=0.1326, over 5670565.99 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3612, pruned_loss=0.1051, over 5714740.09 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3933, pruned_loss=0.1355, over 5664439.70 frames. ], batch size: 92, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:38:29,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.435e+02 1.481e+03 1.986e+03 2.861e+03 8.059e+03, threshold=3.972e+03, percent-clipped=9.0
+2023-03-04 03:38:51,675 INFO [train.py:968] (0/2) Epoch 8, batch 9950, giga_loss[loss=0.3442, simple_loss=0.3933, pruned_loss=0.1475, over 28306.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3905, pruned_loss=0.1333, over 5663983.00 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3614, pruned_loss=0.1053, over 5718556.84 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3932, pruned_loss=0.136, over 5654636.46 frames. ], batch size: 368, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:39:18,234 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-04 03:39:20,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:39:26,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:39:32,585 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:39:40,616 INFO [train.py:968] (0/2) Epoch 8, batch 10000, giga_loss[loss=0.2816, simple_loss=0.3604, pruned_loss=0.1014, over 28938.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3895, pruned_loss=0.1331, over 5673587.82 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3615, pruned_loss=0.1052, over 5720868.66 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.392, pruned_loss=0.1358, over 5663328.48 frames. ], batch size: 145, lr: 4.12e-03, grad_scale: 8.0
+2023-03-04 03:40:00,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:40:10,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.533e+03 2.071e+03 2.978e+03 1.037e+04, threshold=4.142e+03, percent-clipped=15.0
+2023-03-04 03:40:31,579 INFO [train.py:968] (0/2) Epoch 8, batch 10050, giga_loss[loss=0.313, simple_loss=0.3792, pruned_loss=0.1235, over 28839.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3878, pruned_loss=0.1331, over 5666306.21 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3618, pruned_loss=0.1053, over 5723767.71 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3902, pruned_loss=0.1358, over 5654381.17 frames. ], batch size: 119, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:41:17,751 INFO [train.py:968] (0/2) Epoch 8, batch 10100, giga_loss[loss=0.2619, simple_loss=0.3447, pruned_loss=0.08959, over 28999.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3859, pruned_loss=0.1325, over 5676767.18 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3619, pruned_loss=0.1054, over 5728893.96 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3883, pruned_loss=0.1354, over 5661259.10 frames. ], batch size: 145, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:41:43,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:41:46,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:41:48,174 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:41:48,761 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.465e+02 1.446e+03 2.038e+03 3.083e+03 9.122e+03, threshold=4.076e+03, percent-clipped=14.0
+2023-03-04 03:41:50,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:41:54,211 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:41:58,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:01,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0031, 1.2365, 3.4232, 3.0948], device='cuda:0'), covar=tensor([0.1600, 0.2219, 0.0468, 0.0675], device='cuda:0'), in_proj_covar=tensor([0.0612, 0.0569, 0.0818, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 03:42:12,965 INFO [train.py:968] (0/2) Epoch 8, batch 10150, libri_loss[loss=0.278, simple_loss=0.3545, pruned_loss=0.1008, over 29536.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3836, pruned_loss=0.1318, over 5677629.43 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3618, pruned_loss=0.1053, over 5729786.28 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3862, pruned_loss=0.1348, over 5663230.81 frames. ], batch size: 80, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:42:17,146 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:21,024 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:22,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:27,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:28,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:49,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5377, 3.3579, 3.1879, 1.7672], device='cuda:0'), covar=tensor([0.0705, 0.0797, 0.0775, 0.2162], device='cuda:0'), in_proj_covar=tensor([0.0982, 0.0936, 0.0824, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 03:42:54,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:58,965 INFO [train.py:968] (0/2) Epoch 8, batch 10200, libri_loss[loss=0.3386, simple_loss=0.4031, pruned_loss=0.1371, over 29056.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3825, pruned_loss=0.1317, over 5677855.93 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3614, pruned_loss=0.1052, over 5735129.26 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3857, pruned_loss=0.1353, over 5659270.49 frames. ], batch size: 101, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:43:24,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.403e+02 1.503e+03 1.816e+03 2.505e+03 5.028e+03, threshold=3.632e+03, percent-clipped=4.0
+2023-03-04 03:43:43,823 INFO [train.py:968] (0/2) Epoch 8, batch 10250, giga_loss[loss=0.2881, simple_loss=0.3566, pruned_loss=0.1098, over 28681.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3808, pruned_loss=0.1305, over 5679344.61 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3613, pruned_loss=0.1049, over 5739228.74 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.384, pruned_loss=0.1341, over 5659599.07 frames. ], batch size: 92, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:44:30,011 INFO [train.py:968] (0/2) Epoch 8, batch 10300, giga_loss[loss=0.2921, simple_loss=0.3472, pruned_loss=0.1185, over 28977.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3776, pruned_loss=0.1269, over 5677693.76 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3607, pruned_loss=0.1046, over 5742688.72 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.381, pruned_loss=0.1305, over 5657737.41 frames. ], batch size: 100, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:44:40,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3564, 1.7033, 1.6843, 1.2824], device='cuda:0'), covar=tensor([0.1468, 0.2002, 0.1204, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0725, 0.0830, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 03:45:00,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.653e+02 1.300e+03 1.589e+03 2.131e+03 4.146e+03, threshold=3.178e+03, percent-clipped=2.0
+2023-03-04 03:45:15,799 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:45:16,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9349, 1.2898, 1.0754, 0.1202], device='cuda:0'), covar=tensor([0.2233, 0.1898, 0.2936, 0.3747], device='cuda:0'), in_proj_covar=tensor([0.1469, 0.1384, 0.1434, 0.1199], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 03:45:22,560 INFO [train.py:968] (0/2) Epoch 8, batch 10350, giga_loss[loss=0.3395, simple_loss=0.4067, pruned_loss=0.1361, over 28686.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3772, pruned_loss=0.1263, over 5670853.59 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3612, pruned_loss=0.105, over 5745280.09 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3797, pruned_loss=0.1291, over 5652157.76 frames. ], batch size: 307, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:45:33,476 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:46:08,472 INFO [train.py:968] (0/2) Epoch 8, batch 10400, giga_loss[loss=0.3302, simple_loss=0.363, pruned_loss=0.1487, over 23562.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3763, pruned_loss=0.1257, over 5677888.98 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3616, pruned_loss=0.1051, over 5749635.99 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3784, pruned_loss=0.1286, over 5656362.59 frames. ], batch size: 705, lr: 4.12e-03, grad_scale: 8.0
+2023-03-04 03:46:39,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.686e+02 1.481e+03 1.950e+03 2.879e+03 7.513e+03, threshold=3.900e+03, percent-clipped=19.0
+2023-03-04 03:46:53,126 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:47:00,944 INFO [train.py:968] (0/2) Epoch 8, batch 10450, giga_loss[loss=0.3159, simple_loss=0.3708, pruned_loss=0.1305, over 28867.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3733, pruned_loss=0.1251, over 5679401.40 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3615, pruned_loss=0.105, over 5751177.87 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3752, pruned_loss=0.1276, over 5660540.19 frames. ], batch size: 263, lr: 4.12e-03, grad_scale: 8.0
+2023-03-04 03:47:26,121 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:47:48,429 INFO [train.py:968] (0/2) Epoch 8, batch 10500, libri_loss[loss=0.2702, simple_loss=0.3489, pruned_loss=0.09573, over 29566.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3761, pruned_loss=0.1274, over 5682805.71 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3612, pruned_loss=0.1048, over 5751442.75 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3782, pruned_loss=0.1301, over 5665745.85 frames. ], batch size: 76, lr: 4.12e-03, grad_scale: 8.0
+2023-03-04 03:48:14,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3353, 5.1611, 4.9112, 2.1374], device='cuda:0'), covar=tensor([0.0316, 0.0485, 0.0517, 0.1832], device='cuda:0'), in_proj_covar=tensor([0.0964, 0.0924, 0.0811, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 03:48:15,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.582e+03 2.010e+03 2.611e+03 6.756e+03, threshold=4.019e+03, percent-clipped=7.0
+2023-03-04 03:48:36,337 INFO [train.py:968] (0/2) Epoch 8, batch 10550, giga_loss[loss=0.3279, simple_loss=0.3894, pruned_loss=0.1332, over 28644.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3789, pruned_loss=0.1283, over 5671799.50 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3614, pruned_loss=0.1049, over 5744176.52 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3805, pruned_loss=0.1306, over 5664787.53 frames. ], batch size: 307, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:49:27,736 INFO [train.py:968] (0/2) Epoch 8, batch 10600, giga_loss[loss=0.4309, simple_loss=0.4453, pruned_loss=0.2083, over 26670.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3795, pruned_loss=0.1292, over 5646351.06 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3618, pruned_loss=0.1051, over 5745849.63 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3806, pruned_loss=0.131, over 5638413.35 frames. ], batch size: 555, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:49:44,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2664, 1.6942, 1.6903, 1.2071], device='cuda:0'), covar=tensor([0.1584, 0.2250, 0.1227, 0.1480], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0731, 0.0834, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 03:49:57,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.545e+02 1.493e+03 1.834e+03 3.037e+03 9.768e+03, threshold=3.667e+03, percent-clipped=13.0
+2023-03-04 03:50:16,469 INFO [train.py:968] (0/2) Epoch 8, batch 10650, giga_loss[loss=0.3069, simple_loss=0.3709, pruned_loss=0.1214, over 28860.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3795, pruned_loss=0.1294, over 5640834.56 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.362, pruned_loss=0.1052, over 5749089.82 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3806, pruned_loss=0.1313, over 5629648.85 frames. ], batch size: 199, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:50:53,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:51:01,107 INFO [train.py:968] (0/2) Epoch 8, batch 10700, giga_loss[loss=0.3313, simple_loss=0.3694, pruned_loss=0.1466, over 23429.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3809, pruned_loss=0.1307, over 5651429.95 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3624, pruned_loss=0.1055, over 5752095.64 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3816, pruned_loss=0.1324, over 5638157.53 frames. ], batch size: 705, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:51:23,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=329498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:51:34,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.588e+02 1.723e+03 2.439e+03 3.295e+03 6.051e+03, threshold=4.877e+03, percent-clipped=18.0
+2023-03-04 03:51:44,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=329516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:51:55,328 INFO [train.py:968] (0/2) Epoch 8, batch 10750, giga_loss[loss=0.4105, simple_loss=0.4339, pruned_loss=0.1936, over 26615.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3832, pruned_loss=0.1322, over 5651739.15 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3625, pruned_loss=0.1055, over 5753309.54 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.384, pruned_loss=0.1338, over 5638694.50 frames. ], batch size: 555, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:52:43,776 INFO [train.py:968] (0/2) Epoch 8, batch 10800, giga_loss[loss=0.3224, simple_loss=0.388, pruned_loss=0.1284, over 28811.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3855, pruned_loss=0.1341, over 5655681.95 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3625, pruned_loss=0.1055, over 5754943.74 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3863, pruned_loss=0.1357, over 5643078.91 frames. ], batch size: 92, lr: 4.12e-03, grad_scale: 8.0
+2023-03-04 03:52:52,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5675, 1.7000, 1.6507, 1.5441], device='cuda:0'), covar=tensor([0.1295, 0.1675, 0.1770, 0.1582], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0739, 0.0657, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 03:52:59,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=329596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:53:11,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.679e+02 1.515e+03 2.063e+03 2.953e+03 5.297e+03, threshold=4.127e+03, percent-clipped=3.0
+2023-03-04 03:53:31,603 INFO [train.py:968] (0/2) Epoch 8, batch 10850, giga_loss[loss=0.3462, simple_loss=0.3988, pruned_loss=0.1468, over 28927.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3873, pruned_loss=0.1358, over 5657757.69 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3628, pruned_loss=0.1056, over 5757020.98 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3882, pruned_loss=0.1376, over 5643752.10 frames. ], batch size: 106, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:53:32,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=329629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:53:37,913 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9914, 1.3507, 1.0493, 0.2174], device='cuda:0'), covar=tensor([0.1976, 0.1555, 0.2667, 0.3384], device='cuda:0'), in_proj_covar=tensor([0.1482, 0.1394, 0.1446, 0.1214], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 03:53:45,740 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=329641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:53:47,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=329644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:53:59,083 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:04,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=329659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:06,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:06,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=329662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:16,650 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=329673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:20,928 INFO [train.py:968] (0/2) Epoch 8, batch 10900, giga_loss[loss=0.3067, simple_loss=0.3813, pruned_loss=0.1161, over 28869.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3879, pruned_loss=0.1364, over 5653547.13 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3626, pruned_loss=0.1055, over 5748596.80 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3896, pruned_loss=0.1388, over 5647461.06 frames. ], batch size: 174, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:54:32,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=329691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:52,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.735e+02 1.638e+03 2.188e+03 3.072e+03 1.108e+04, threshold=4.377e+03, percent-clipped=13.0
+2023-03-04 03:54:53,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5307, 1.5860, 1.5649, 1.4635], device='cuda:0'), covar=tensor([0.1113, 0.1841, 0.1542, 0.1568], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0733, 0.0650, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 03:55:09,749 INFO [train.py:968] (0/2) Epoch 8, batch 10950, giga_loss[loss=0.3488, simple_loss=0.3825, pruned_loss=0.1575, over 23831.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3867, pruned_loss=0.1336, over 5656274.51 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3621, pruned_loss=0.1053, over 5754168.11 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3894, pruned_loss=0.1368, over 5643129.38 frames. ], batch size: 705, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:55:14,589 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329733.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 03:55:20,257 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=329739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:24,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=329742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:53,142 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=329771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:54,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=329772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:55,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:55,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7456, 5.1634, 2.0196, 1.8427], device='cuda:0'), covar=tensor([0.0837, 0.0186, 0.0792, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0499, 0.0325, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 03:55:56,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=329775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:59,177 INFO [train.py:968] (0/2) Epoch 8, batch 11000, giga_loss[loss=0.3278, simple_loss=0.3872, pruned_loss=0.1342, over 28815.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3859, pruned_loss=0.1334, over 5655618.52 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3617, pruned_loss=0.1051, over 5759416.69 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3893, pruned_loss=0.1372, over 5637034.21 frames. ], batch size: 284, lr: 4.12e-03, grad_scale: 2.0
+2023-03-04 03:56:06,222 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-04 03:56:16,400 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-04 03:56:25,621 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=329804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:56:30,736 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.874e+02 1.651e+03 2.252e+03 3.015e+03 1.086e+04, threshold=4.505e+03, percent-clipped=9.0
+2023-03-04 03:56:52,371 INFO [train.py:968] (0/2) Epoch 8, batch 11050, libri_loss[loss=0.2732, simple_loss=0.3571, pruned_loss=0.09465, over 29777.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3848, pruned_loss=0.1332, over 5664184.98 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3618, pruned_loss=0.105, over 5757775.43 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3879, pruned_loss=0.1367, over 5649134.92 frames. ], batch size: 87, lr: 4.12e-03, grad_scale: 2.0
+2023-03-04 03:57:10,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=329845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:57:50,307 INFO [train.py:968] (0/2) Epoch 8, batch 11100, giga_loss[loss=0.2974, simple_loss=0.3636, pruned_loss=0.1156, over 28771.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3847, pruned_loss=0.1339, over 5649720.29 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3619, pruned_loss=0.1052, over 5750777.10 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3873, pruned_loss=0.1369, over 5643121.59 frames. ], batch size: 262, lr: 4.12e-03, grad_scale: 2.0
+2023-03-04 03:58:05,377 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:58:19,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.081e+02 1.479e+03 1.955e+03 2.937e+03 7.755e+03, threshold=3.910e+03, percent-clipped=2.0
+2023-03-04 03:58:36,590 INFO [train.py:968] (0/2) Epoch 8, batch 11150, giga_loss[loss=0.3457, simple_loss=0.3918, pruned_loss=0.1498, over 27984.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3827, pruned_loss=0.1324, over 5666772.98 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3616, pruned_loss=0.105, over 5753787.74 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3855, pruned_loss=0.1357, over 5656661.02 frames. ], batch size: 412, lr: 4.12e-03, grad_scale: 2.0
+2023-03-04 03:58:42,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:58:55,934 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:59:23,834 INFO [train.py:968] (0/2) Epoch 8, batch 11200, giga_loss[loss=0.3193, simple_loss=0.3828, pruned_loss=0.1279, over 28920.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3819, pruned_loss=0.1321, over 5672252.94 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3619, pruned_loss=0.105, over 5756668.09 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3843, pruned_loss=0.1352, over 5660213.59 frames. ], batch size: 227, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:59:34,568 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=329988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:59:36,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=329991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:59:46,584 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-330000.pt
+2023-03-04 03:59:55,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.790e+02 1.599e+03 2.086e+03 3.400e+03 1.101e+04, threshold=4.173e+03, percent-clipped=18.0
+2023-03-04 03:59:58,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3645, 1.1816, 4.3762, 3.4705], device='cuda:0'), covar=tensor([0.1635, 0.2501, 0.0359, 0.0797], device='cuda:0'), in_proj_covar=tensor([0.0617, 0.0567, 0.0815, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 04:00:04,169 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:00:10,875 INFO [train.py:968] (0/2) Epoch 8, batch 11250, giga_loss[loss=0.3297, simple_loss=0.3981, pruned_loss=0.1306, over 28950.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3813, pruned_loss=0.1319, over 5672265.78 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3614, pruned_loss=0.1048, over 5758876.15 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.384, pruned_loss=0.135, over 5658903.52 frames. ], batch size: 186, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 04:00:14,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:00:19,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:00:59,403 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 04:01:00,977 INFO [train.py:968] (0/2) Epoch 8, batch 11300, giga_loss[loss=0.3309, simple_loss=0.3879, pruned_loss=0.137, over 28904.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3824, pruned_loss=0.1328, over 5669532.73 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3615, pruned_loss=0.1049, over 5757770.04 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3847, pruned_loss=0.1355, over 5659298.48 frames. ], batch size: 227, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 04:01:20,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2465, 1.5153, 1.1883, 1.4248], device='cuda:0'), covar=tensor([0.0768, 0.0306, 0.0346, 0.0837], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0119, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0070, 0.0051, 0.0046, 0.0077], device='cuda:0')
+2023-03-04 04:01:31,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330108.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 04:01:32,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.966e+02 1.562e+03 1.997e+03 2.698e+03 6.747e+03, threshold=3.994e+03, percent-clipped=1.0
+2023-03-04 04:01:50,528 INFO [train.py:968] (0/2) Epoch 8, batch 11350, libri_loss[loss=0.3351, simple_loss=0.4014, pruned_loss=0.1344, over 29372.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3848, pruned_loss=0.1348, over 5675007.90 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3619, pruned_loss=0.1053, over 5761147.29 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3868, pruned_loss=0.1373, over 5661658.30 frames. ], batch size: 92, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:02:12,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:02:18,365 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-04 04:02:20,052 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-04 04:02:38,603 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:02:40,794 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:02:42,789 INFO [train.py:968] (0/2) Epoch 8, batch 11400, giga_loss[loss=0.3448, simple_loss=0.3912, pruned_loss=0.1492, over 28972.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.386, pruned_loss=0.1359, over 5673018.89 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3616, pruned_loss=0.1051, over 5762240.11 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.388, pruned_loss=0.1383, over 5660702.75 frames. ], batch size: 128, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:02:43,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 04:02:44,249 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:02:47,277 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:03:09,407 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:03:15,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.613e+03 2.143e+03 2.983e+03 1.383e+04, threshold=4.287e+03, percent-clipped=13.0
+2023-03-04 04:03:17,266 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:03:34,939 INFO [train.py:968] (0/2) Epoch 8, batch 11450, giga_loss[loss=0.3971, simple_loss=0.4198, pruned_loss=0.1872, over 26575.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3857, pruned_loss=0.1369, over 5658885.92 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3612, pruned_loss=0.1049, over 5763477.70 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.388, pruned_loss=0.1395, over 5646692.18 frames. ], batch size: 555, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:03:54,043 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330251.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 04:03:56,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330254.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 04:04:02,113 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=330261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:04:10,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:04:19,053 INFO [train.py:968] (0/2) Epoch 8, batch 11500, giga_loss[loss=0.3211, simple_loss=0.3774, pruned_loss=0.1324, over 28846.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3848, pruned_loss=0.1358, over 5664795.49 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3609, pruned_loss=0.1046, over 5766598.71 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3876, pruned_loss=0.139, over 5649631.86 frames. ], batch size: 112, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:04:24,116 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330283.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 04:04:33,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:04:36,474 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:04:52,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.149e+02 1.589e+03 2.150e+03 3.634e+03 1.156e+04, threshold=4.301e+03, percent-clipped=19.0
+2023-03-04 04:04:52,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:05:00,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:05:02,259 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:05:06,352 INFO [train.py:968] (0/2) Epoch 8, batch 11550, libri_loss[loss=0.247, simple_loss=0.3225, pruned_loss=0.08578, over 29668.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3844, pruned_loss=0.1346, over 5679699.43 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3611, pruned_loss=0.1048, over 5767949.74 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3873, pruned_loss=0.138, over 5663120.35 frames. ], batch size: 73, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:05:51,726 INFO [train.py:968] (0/2) Epoch 8, batch 11600, giga_loss[loss=0.3115, simple_loss=0.373, pruned_loss=0.1251, over 28722.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3844, pruned_loss=0.1339, over 5676442.31 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3615, pruned_loss=0.105, over 5769679.87 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3868, pruned_loss=0.137, over 5659991.32 frames. ], batch size: 92, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:05:59,988 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=330386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:06:00,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2123, 1.0626, 4.1207, 3.2456], device='cuda:0'), covar=tensor([0.1637, 0.2522, 0.0387, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0611, 0.0562, 0.0806, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 04:06:18,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.593e+03 2.147e+03 2.852e+03 6.235e+03, threshold=4.294e+03, percent-clipped=4.0
+2023-03-04 04:06:19,882 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:06:24,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:06:37,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1487, 1.2241, 4.0489, 3.2345], device='cuda:0'), covar=tensor([0.1686, 0.2307, 0.0415, 0.0672], device='cuda:0'), in_proj_covar=tensor([0.0611, 0.0562, 0.0806, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 04:06:40,257 INFO [train.py:968] (0/2) Epoch 8, batch 11650, giga_loss[loss=0.3101, simple_loss=0.3747, pruned_loss=0.1228, over 28865.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3849, pruned_loss=0.134, over 5688732.49 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3617, pruned_loss=0.1053, over 5773350.41 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3874, pruned_loss=0.1373, over 5668909.70 frames. ], batch size: 112, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:06:52,325 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:02,437 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:04,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:13,099 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:16,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:28,098 INFO [train.py:968] (0/2) Epoch 8, batch 11700, giga_loss[loss=0.4337, simple_loss=0.443, pruned_loss=0.2122, over 26562.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3867, pruned_loss=0.1356, over 5685307.25 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3617, pruned_loss=0.1051, over 5777476.01 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3898, pruned_loss=0.1396, over 5662304.03 frames. ], batch size: 555, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:07:35,531 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:44,747 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:59,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.608e+02 1.625e+03 2.101e+03 2.855e+03 6.891e+03, threshold=4.203e+03, percent-clipped=5.0
+2023-03-04 04:08:00,277 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 04:08:14,296 INFO [train.py:968] (0/2) Epoch 8, batch 11750, giga_loss[loss=0.3303, simple_loss=0.3871, pruned_loss=0.1368, over 28281.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3864, pruned_loss=0.1349, over 5695726.94 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3615, pruned_loss=0.1049, over 5779332.17 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3894, pruned_loss=0.1387, over 5674668.53 frames. ], batch size: 368, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:08:54,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2795, 2.1461, 1.6401, 1.7943], device='cuda:0'), covar=tensor([0.0695, 0.0667, 0.0943, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0448, 0.0496, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 04:08:59,251 INFO [train.py:968] (0/2) Epoch 8, batch 11800, giga_loss[loss=0.322, simple_loss=0.392, pruned_loss=0.1259, over 28864.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3868, pruned_loss=0.1343, over 5690198.03 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3615, pruned_loss=0.1048, over 5774140.82 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3899, pruned_loss=0.1384, over 5674814.51 frames. ], batch size: 199, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:09:27,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1160, 2.0004, 1.8746, 2.2811], device='cuda:0'), covar=tensor([0.1910, 0.2105, 0.2156, 0.1992], device='cuda:0'), in_proj_covar=tensor([0.1220, 0.0911, 0.1074, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 04:09:30,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1664, 1.2606, 4.2619, 3.3607], device='cuda:0'), covar=tensor([0.1668, 0.2419, 0.0355, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0612, 0.0563, 0.0810, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 04:09:33,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.442e+02 1.301e+03 1.816e+03 2.544e+03 6.969e+03, threshold=3.632e+03, percent-clipped=7.0
+2023-03-04 04:09:40,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2454, 1.4125, 1.3254, 1.4966], device='cuda:0'), covar=tensor([0.0762, 0.0330, 0.0313, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0119, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:0')
+2023-03-04 04:09:47,830 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.74 vs. limit=5.0
+2023-03-04 04:09:49,292 INFO [train.py:968] (0/2) Epoch 8, batch 11850, giga_loss[loss=0.2842, simple_loss=0.3479, pruned_loss=0.1102, over 28093.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3853, pruned_loss=0.1323, over 5673208.68 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3613, pruned_loss=0.1046, over 5766360.32 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3883, pruned_loss=0.1361, over 5667512.15 frames. ], batch size: 77, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:09:56,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:10:17,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2665, 1.4333, 1.5019, 1.3917], device='cuda:0'), covar=tensor([0.0974, 0.0882, 0.1166, 0.0924], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0745, 0.0664, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 04:10:36,069 INFO [train.py:968] (0/2) Epoch 8, batch 11900, giga_loss[loss=0.3078, simple_loss=0.372, pruned_loss=0.1217, over 28900.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3843, pruned_loss=0.1319, over 5674305.92 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3612, pruned_loss=0.1046, over 5767069.57 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3874, pruned_loss=0.1356, over 5666935.98 frames. ], batch size: 136, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:11:07,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.675e+02 1.489e+03 1.937e+03 2.506e+03 5.717e+03, threshold=3.873e+03, percent-clipped=11.0
+2023-03-04 04:11:22,426 INFO [train.py:968] (0/2) Epoch 8, batch 11950, giga_loss[loss=0.3112, simple_loss=0.375, pruned_loss=0.1237, over 28874.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3839, pruned_loss=0.1315, over 5685343.68 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3615, pruned_loss=0.1047, over 5767786.38 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3864, pruned_loss=0.1347, over 5677781.66 frames. ], batch size: 174, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:11:32,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2206, 1.4399, 1.1104, 0.9246], device='cuda:0'), covar=tensor([0.1183, 0.1146, 0.0880, 0.1153], device='cuda:0'), in_proj_covar=tensor([0.1602, 0.1449, 0.1421, 0.1532], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 04:11:57,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:12:12,497 INFO [train.py:968] (0/2) Epoch 8, batch 12000, giga_loss[loss=0.3415, simple_loss=0.4112, pruned_loss=0.1359, over 28794.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3849, pruned_loss=0.1332, over 5668069.23 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3614, pruned_loss=0.1047, over 5771264.81 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3875, pruned_loss=0.1364, over 5656885.70 frames. ], batch size: 284, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:12:12,502 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 04:12:21,416 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2256, simple_loss=0.3306, pruned_loss=0.06027, over 944034.00 frames. 
+2023-03-04 04:12:21,417 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 04:12:21,751 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:12:23,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:12:50,455 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:12:51,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.594e+02 1.554e+03 1.916e+03 3.068e+03 9.824e+03, threshold=3.832e+03, percent-clipped=14.0
+2023-03-04 04:13:08,721 INFO [train.py:968] (0/2) Epoch 8, batch 12050, giga_loss[loss=0.283, simple_loss=0.351, pruned_loss=0.1075, over 28530.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3867, pruned_loss=0.1342, over 5668288.08 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3616, pruned_loss=0.1047, over 5770199.78 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.389, pruned_loss=0.1373, over 5658695.90 frames. ], batch size: 85, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:13:13,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2087, 2.4158, 2.2749, 1.4475], device='cuda:0'), covar=tensor([0.0758, 0.0212, 0.0209, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0119, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:0')
+2023-03-04 04:13:53,487 INFO [train.py:968] (0/2) Epoch 8, batch 12100, giga_loss[loss=0.3288, simple_loss=0.3867, pruned_loss=0.1355, over 28447.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3862, pruned_loss=0.1346, over 5676914.88 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3619, pruned_loss=0.1048, over 5774469.89 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3886, pruned_loss=0.1379, over 5662266.68 frames. ], batch size: 71, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:14:20,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:14:22,474 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:14:28,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.056e+02 1.351e+03 1.810e+03 2.419e+03 6.441e+03, threshold=3.620e+03, percent-clipped=6.0
+2023-03-04 04:14:37,679 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=330924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:14:41,451 INFO [train.py:968] (0/2) Epoch 8, batch 12150, giga_loss[loss=0.291, simple_loss=0.3562, pruned_loss=0.1129, over 29007.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3864, pruned_loss=0.1356, over 5666479.90 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3618, pruned_loss=0.1048, over 5771730.45 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3887, pruned_loss=0.1387, over 5655884.18 frames. ], batch size: 106, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:14:49,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:15:30,221 INFO [train.py:968] (0/2) Epoch 8, batch 12200, giga_loss[loss=0.3008, simple_loss=0.3707, pruned_loss=0.1154, over 28966.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3877, pruned_loss=0.1366, over 5651334.59 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3623, pruned_loss=0.1051, over 5760561.94 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3895, pruned_loss=0.1392, over 5651875.29 frames. ], batch size: 136, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:15:30,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0593, 1.2463, 3.6344, 3.0237], device='cuda:0'), covar=tensor([0.1607, 0.2351, 0.0408, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0616, 0.0564, 0.0815, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 04:15:47,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5313, 4.4654, 1.7912, 1.6425], device='cuda:0'), covar=tensor([0.0883, 0.0233, 0.0784, 0.1236], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0504, 0.0325, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 04:16:00,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.469e+02 1.535e+03 2.006e+03 2.763e+03 8.917e+03, threshold=4.011e+03, percent-clipped=15.0
+2023-03-04 04:16:13,872 INFO [train.py:968] (0/2) Epoch 8, batch 12250, giga_loss[loss=0.3538, simple_loss=0.4102, pruned_loss=0.1487, over 28635.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3877, pruned_loss=0.1361, over 5659593.50 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3625, pruned_loss=0.1053, over 5762350.49 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3901, pruned_loss=0.1393, over 5653964.21 frames. ], batch size: 262, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:17:04,076 INFO [train.py:968] (0/2) Epoch 8, batch 12300, giga_loss[loss=0.3035, simple_loss=0.3703, pruned_loss=0.1184, over 28845.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3871, pruned_loss=0.1365, over 5642998.26 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3621, pruned_loss=0.105, over 5764677.11 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3899, pruned_loss=0.1399, over 5634688.15 frames. ], batch size: 199, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:17:37,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.567e+02 1.557e+03 2.032e+03 2.907e+03 8.242e+03, threshold=4.064e+03, percent-clipped=9.0
+2023-03-04 04:17:52,738 INFO [train.py:968] (0/2) Epoch 8, batch 12350, libri_loss[loss=0.2811, simple_loss=0.3668, pruned_loss=0.09769, over 29785.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3881, pruned_loss=0.1371, over 5640446.75 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3623, pruned_loss=0.1051, over 5754958.50 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3908, pruned_loss=0.1407, over 5638692.84 frames. ], batch size: 87, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:18:14,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4779, 1.7839, 1.2527, 1.2628], device='cuda:0'), covar=tensor([0.1539, 0.1135, 0.0963, 0.1177], device='cuda:0'), in_proj_covar=tensor([0.1602, 0.1452, 0.1431, 0.1535], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 04:18:34,586 INFO [train.py:968] (0/2) Epoch 8, batch 12400, giga_loss[loss=0.4087, simple_loss=0.4434, pruned_loss=0.1871, over 28265.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3879, pruned_loss=0.1361, over 5648137.98 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.362, pruned_loss=0.1048, over 5756030.06 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.391, pruned_loss=0.1401, over 5642930.94 frames. ], batch size: 368, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:18:39,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0929, 1.3444, 3.8625, 3.1192], device='cuda:0'), covar=tensor([0.1638, 0.2256, 0.0407, 0.0745], device='cuda:0'), in_proj_covar=tensor([0.0614, 0.0566, 0.0816, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 04:19:02,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8737, 2.5653, 1.6550, 1.5812], device='cuda:0'), covar=tensor([0.1599, 0.0926, 0.1284, 0.1464], device='cuda:0'), in_proj_covar=tensor([0.1593, 0.1441, 0.1422, 0.1523], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 04:19:03,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.515e+02 1.570e+03 2.094e+03 2.546e+03 6.188e+03, threshold=4.189e+03, percent-clipped=7.0
+2023-03-04 04:19:18,370 INFO [train.py:968] (0/2) Epoch 8, batch 12450, giga_loss[loss=0.3313, simple_loss=0.3893, pruned_loss=0.1366, over 28313.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3872, pruned_loss=0.1355, over 5651648.20 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3616, pruned_loss=0.1046, over 5756583.38 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3914, pruned_loss=0.1405, over 5641548.77 frames. ], batch size: 60, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:19:57,935 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4579, 4.2944, 4.0444, 1.9475], device='cuda:0'), covar=tensor([0.0462, 0.0656, 0.0659, 0.1935], device='cuda:0'), in_proj_covar=tensor([0.0976, 0.0929, 0.0825, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 04:20:01,554 INFO [train.py:968] (0/2) Epoch 8, batch 12500, giga_loss[loss=0.4324, simple_loss=0.4536, pruned_loss=0.2056, over 27970.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3852, pruned_loss=0.1341, over 5650952.78 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3617, pruned_loss=0.1047, over 5742242.47 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3893, pruned_loss=0.1391, over 5651248.78 frames. ], batch size: 412, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:20:19,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=331299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:20:38,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.027e+02 1.550e+03 2.080e+03 2.895e+03 6.224e+03, threshold=4.160e+03, percent-clipped=8.0
+2023-03-04 04:20:42,111 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-04 04:20:50,299 INFO [train.py:968] (0/2) Epoch 8, batch 12550, giga_loss[loss=0.3285, simple_loss=0.3813, pruned_loss=0.1378, over 27926.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3822, pruned_loss=0.1322, over 5655691.42 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3618, pruned_loss=0.1048, over 5736067.93 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.386, pruned_loss=0.1368, over 5658255.94 frames. ], batch size: 412, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:21:38,709 INFO [train.py:968] (0/2) Epoch 8, batch 12600, giga_loss[loss=0.3278, simple_loss=0.3827, pruned_loss=0.1365, over 28573.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3783, pruned_loss=0.1304, over 5652972.00 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3615, pruned_loss=0.1047, over 5738482.73 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3818, pruned_loss=0.1345, over 5651921.21 frames. ], batch size: 336, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:22:14,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.985e+02 1.587e+03 2.069e+03 3.154e+03 7.301e+03, threshold=4.137e+03, percent-clipped=11.0
+2023-03-04 04:22:17,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4870, 1.6042, 1.7299, 1.3547], device='cuda:0'), covar=tensor([0.1240, 0.1876, 0.1026, 0.1177], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0721, 0.0826, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 04:22:25,988 INFO [train.py:968] (0/2) Epoch 8, batch 12650, giga_loss[loss=0.3086, simple_loss=0.367, pruned_loss=0.1251, over 28917.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3785, pruned_loss=0.1319, over 5633699.09 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.362, pruned_loss=0.1051, over 5721768.10 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3814, pruned_loss=0.1356, over 5645321.89 frames. ], batch size: 227, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:22:40,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=331442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:22:42,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=331445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:23:12,678 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=331474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:23:18,548 INFO [train.py:968] (0/2) Epoch 8, batch 12700, giga_loss[loss=0.33, simple_loss=0.3694, pruned_loss=0.1453, over 23754.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.378, pruned_loss=0.1324, over 5630527.97 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3621, pruned_loss=0.1051, over 5722827.26 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3802, pruned_loss=0.1354, over 5638424.12 frames. ], batch size: 705, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:23:39,782 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-04 04:23:54,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.381e+02 1.536e+03 2.032e+03 3.034e+03 6.106e+03, threshold=4.064e+03, percent-clipped=10.0
+2023-03-04 04:24:07,905 INFO [train.py:968] (0/2) Epoch 8, batch 12750, giga_loss[loss=0.3242, simple_loss=0.379, pruned_loss=0.1346, over 28481.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3776, pruned_loss=0.1315, over 5641170.08 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3623, pruned_loss=0.1053, over 5725533.00 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3794, pruned_loss=0.1342, over 5643836.67 frames. ], batch size: 71, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:24:08,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0828, 1.5635, 1.4258, 1.0724], device='cuda:0'), covar=tensor([0.1319, 0.1955, 0.1110, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0718, 0.0824, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 04:24:24,791 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-04 04:24:51,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3725, 4.1977, 3.9475, 1.9400], device='cuda:0'), covar=tensor([0.0484, 0.0668, 0.0774, 0.2041], device='cuda:0'), in_proj_covar=tensor([0.0973, 0.0924, 0.0815, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 04:24:55,115 INFO [train.py:968] (0/2) Epoch 8, batch 12800, giga_loss[loss=0.3073, simple_loss=0.3874, pruned_loss=0.1136, over 28671.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.376, pruned_loss=0.1283, over 5648655.09 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3621, pruned_loss=0.1053, over 5731318.19 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3782, pruned_loss=0.1313, over 5642877.08 frames. ], batch size: 307, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:25:13,835 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-04 04:25:14,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2910, 1.6196, 1.3535, 1.4791], device='cuda:0'), covar=tensor([0.0782, 0.0301, 0.0323, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0119, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:0')
+2023-03-04 04:25:34,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.013e+02 1.454e+03 1.918e+03 2.644e+03 5.356e+03, threshold=3.837e+03, percent-clipped=5.0
+2023-03-04 04:25:38,585 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-04 04:25:48,228 INFO [train.py:968] (0/2) Epoch 8, batch 12850, giga_loss[loss=0.325, simple_loss=0.3958, pruned_loss=0.1271, over 28933.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3737, pruned_loss=0.1251, over 5650303.46 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.362, pruned_loss=0.1053, over 5736041.28 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3759, pruned_loss=0.1279, over 5639325.48 frames. ], batch size: 227, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:26:11,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3424, 3.2823, 1.3933, 1.4273], device='cuda:0'), covar=tensor([0.0904, 0.0279, 0.0918, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0502, 0.0326, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 04:26:33,826 INFO [train.py:968] (0/2) Epoch 8, batch 12900, giga_loss[loss=0.2637, simple_loss=0.3445, pruned_loss=0.09149, over 28876.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3695, pruned_loss=0.1209, over 5654153.57 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3607, pruned_loss=0.1048, over 5742267.88 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3729, pruned_loss=0.1245, over 5636359.16 frames. ], batch size: 199, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:26:51,508 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=331695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:27:12,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.967e+02 1.489e+03 2.110e+03 3.043e+03 7.151e+03, threshold=4.219e+03, percent-clipped=10.0
+2023-03-04 04:27:26,968 INFO [train.py:968] (0/2) Epoch 8, batch 12950, giga_loss[loss=0.3227, simple_loss=0.3666, pruned_loss=0.1394, over 26670.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3666, pruned_loss=0.1184, over 5647459.45 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3611, pruned_loss=0.1051, over 5742134.51 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1213, over 5631638.87 frames. ], batch size: 555, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:27:28,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5026, 1.7555, 1.3495, 1.3512], device='cuda:0'), covar=tensor([0.1419, 0.1040, 0.1018, 0.1128], device='cuda:0'), in_proj_covar=tensor([0.1581, 0.1434, 0.1406, 0.1509], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 04:27:33,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2050, 1.6906, 1.2450, 0.5765], device='cuda:0'), covar=tensor([0.2125, 0.1107, 0.1877, 0.2886], device='cuda:0'), in_proj_covar=tensor([0.1462, 0.1381, 0.1434, 0.1206], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 04:28:14,333 INFO [train.py:968] (0/2) Epoch 8, batch 13000, giga_loss[loss=0.2719, simple_loss=0.3555, pruned_loss=0.09418, over 28892.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3644, pruned_loss=0.1145, over 5657755.09 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3605, pruned_loss=0.1049, over 5745442.00 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3671, pruned_loss=0.1173, over 5639586.00 frames. ], batch size: 199, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:28:51,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.418e+02 1.217e+03 1.571e+03 2.335e+03 1.114e+04, threshold=3.142e+03, percent-clipped=6.0
+2023-03-04 04:29:05,849 INFO [train.py:968] (0/2) Epoch 8, batch 13050, giga_loss[loss=0.3406, simple_loss=0.3916, pruned_loss=0.1448, over 27630.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3638, pruned_loss=0.1127, over 5662963.77 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3601, pruned_loss=0.1048, over 5748528.72 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3664, pruned_loss=0.1152, over 5643844.94 frames. ], batch size: 472, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:29:41,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3119, 1.2505, 1.1916, 1.4557], device='cuda:0'), covar=tensor([0.0711, 0.0345, 0.0327, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0120, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:0')
+2023-03-04 04:29:51,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2223, 1.6588, 1.2822, 0.4416], device='cuda:0'), covar=tensor([0.1960, 0.1324, 0.2128, 0.2839], device='cuda:0'), in_proj_covar=tensor([0.1459, 0.1385, 0.1434, 0.1203], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 04:29:53,911 INFO [train.py:968] (0/2) Epoch 8, batch 13100, giga_loss[loss=0.2657, simple_loss=0.3413, pruned_loss=0.09505, over 28586.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.363, pruned_loss=0.1121, over 5655971.92 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3598, pruned_loss=0.1047, over 5744034.69 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3655, pruned_loss=0.1144, over 5641803.41 frames. ], batch size: 85, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:30:26,141 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0302, 3.0060, 2.0271, 0.9483], device='cuda:0'), covar=tensor([0.3693, 0.1714, 0.2230, 0.3769], device='cuda:0'), in_proj_covar=tensor([0.1453, 0.1379, 0.1427, 0.1200], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 04:30:31,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.922e+02 1.277e+03 1.597e+03 2.416e+03 8.269e+03, threshold=3.193e+03, percent-clipped=10.0
+2023-03-04 04:30:45,730 INFO [train.py:968] (0/2) Epoch 8, batch 13150, giga_loss[loss=0.2477, simple_loss=0.3335, pruned_loss=0.08093, over 29070.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3604, pruned_loss=0.11, over 5656124.76 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.36, pruned_loss=0.1051, over 5747208.13 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3622, pruned_loss=0.1117, over 5639749.19 frames. ], batch size: 136, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:31:35,414 INFO [train.py:968] (0/2) Epoch 8, batch 13200, libri_loss[loss=0.2592, simple_loss=0.3276, pruned_loss=0.09541, over 29530.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3576, pruned_loss=0.1086, over 5655117.17 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3592, pruned_loss=0.1047, over 5752611.69 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3598, pruned_loss=0.1104, over 5634144.10 frames. ], batch size: 79, lr: 4.10e-03, grad_scale: 8.0
+2023-03-04 04:31:41,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2880, 3.1085, 2.9551, 1.4869], device='cuda:0'), covar=tensor([0.0828, 0.0931, 0.0947, 0.2132], device='cuda:0'), in_proj_covar=tensor([0.0947, 0.0900, 0.0791, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 04:31:56,229 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-332000.pt
+2023-03-04 04:32:06,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.42 vs. limit=5.0
+2023-03-04 04:32:10,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.635e+02 1.459e+03 2.009e+03 2.787e+03 7.269e+03, threshold=4.019e+03, percent-clipped=17.0
+2023-03-04 04:32:22,573 INFO [train.py:968] (0/2) Epoch 8, batch 13250, giga_loss[loss=0.3237, simple_loss=0.3725, pruned_loss=0.1375, over 26723.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3571, pruned_loss=0.1081, over 5645105.46 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3588, pruned_loss=0.1045, over 5745486.67 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3593, pruned_loss=0.1099, over 5631810.62 frames. ], batch size: 555, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:32:40,463 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:33:03,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:33:12,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1119, 1.1670, 4.0325, 3.2699], device='cuda:0'), covar=tensor([0.1607, 0.2459, 0.0356, 0.0687], device='cuda:0'), in_proj_covar=tensor([0.0603, 0.0558, 0.0803, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 04:33:12,933 INFO [train.py:968] (0/2) Epoch 8, batch 13300, giga_loss[loss=0.2878, simple_loss=0.3615, pruned_loss=0.107, over 28943.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3571, pruned_loss=0.1079, over 5647612.05 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3588, pruned_loss=0.1047, over 5746991.80 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3587, pruned_loss=0.1092, over 5634454.16 frames. ], batch size: 145, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:33:28,256 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6590, 2.0440, 1.7446, 1.7911], device='cuda:0'), covar=tensor([0.1253, 0.1343, 0.1605, 0.1400], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0714, 0.0639, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 04:33:49,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.759e+02 1.208e+03 1.593e+03 2.087e+03 5.068e+03, threshold=3.185e+03, percent-clipped=1.0
+2023-03-04 04:34:01,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:34:04,025 INFO [train.py:968] (0/2) Epoch 8, batch 13350, giga_loss[loss=0.2586, simple_loss=0.3349, pruned_loss=0.09111, over 27826.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3534, pruned_loss=0.1049, over 5652789.28 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3582, pruned_loss=0.1046, over 5751125.82 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3552, pruned_loss=0.1061, over 5635845.71 frames. ], batch size: 412, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:34:54,669 INFO [train.py:968] (0/2) Epoch 8, batch 13400, giga_loss[loss=0.2706, simple_loss=0.3507, pruned_loss=0.09528, over 29041.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3503, pruned_loss=0.1026, over 5656446.05 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3584, pruned_loss=0.1048, over 5753628.25 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3513, pruned_loss=0.1033, over 5638960.57 frames. ], batch size: 155, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:35:06,561 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:35:27,603 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:35:30,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.328e+02 1.270e+03 1.771e+03 2.465e+03 6.040e+03, threshold=3.542e+03, percent-clipped=15.0
+2023-03-04 04:35:30,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:35:39,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8089, 3.6184, 3.4071, 1.7466], device='cuda:0'), covar=tensor([0.0602, 0.0800, 0.0816, 0.2074], device='cuda:0'), in_proj_covar=tensor([0.0945, 0.0898, 0.0788, 0.0621], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 04:35:44,442 INFO [train.py:968] (0/2) Epoch 8, batch 13450, giga_loss[loss=0.2656, simple_loss=0.3484, pruned_loss=0.09147, over 28840.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3461, pruned_loss=0.1003, over 5661372.35 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3575, pruned_loss=0.1044, over 5757604.98 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3474, pruned_loss=0.1011, over 5640146.84 frames. ], batch size: 186, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:36:04,686 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:36:04,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4920, 3.5162, 1.5552, 1.5468], device='cuda:0'), covar=tensor([0.0841, 0.0292, 0.0870, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0497, 0.0328, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 04:36:31,805 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 04:36:38,490 INFO [train.py:968] (0/2) Epoch 8, batch 13500, libri_loss[loss=0.2861, simple_loss=0.3535, pruned_loss=0.1093, over 29520.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.345, pruned_loss=0.1, over 5666287.66 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.357, pruned_loss=0.1043, over 5761135.98 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3462, pruned_loss=0.1007, over 5643936.01 frames. ], batch size: 81, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:36:49,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9716, 1.2626, 1.3509, 1.1528], device='cuda:0'), covar=tensor([0.1325, 0.0958, 0.1579, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0709, 0.0639, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 04:37:05,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7272, 1.9347, 1.6755, 1.6453], device='cuda:0'), covar=tensor([0.1154, 0.1611, 0.1426, 0.1451], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0706, 0.0637, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 04:37:07,928 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:37:13,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.453e+02 1.354e+03 1.822e+03 2.545e+03 6.448e+03, threshold=3.644e+03, percent-clipped=12.0
+2023-03-04 04:37:27,279 INFO [train.py:968] (0/2) Epoch 8, batch 13550, giga_loss[loss=0.2463, simple_loss=0.3278, pruned_loss=0.08244, over 28553.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3462, pruned_loss=0.1016, over 5657993.49 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.357, pruned_loss=0.1044, over 5763214.21 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3468, pruned_loss=0.1019, over 5635569.15 frames. ], batch size: 336, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:38:24,964 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:38:27,478 INFO [train.py:968] (0/2) Epoch 8, batch 13600, giga_loss[loss=0.231, simple_loss=0.3022, pruned_loss=0.07992, over 24211.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3479, pruned_loss=0.1017, over 5652517.86 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3564, pruned_loss=0.1043, over 5753645.99 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3487, pruned_loss=0.102, over 5639970.20 frames. ], batch size: 705, lr: 4.10e-03, grad_scale: 8.0
+2023-03-04 04:39:14,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.705e+02 1.419e+03 1.862e+03 2.289e+03 5.317e+03, threshold=3.724e+03, percent-clipped=7.0
+2023-03-04 04:39:23,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:39:31,647 INFO [train.py:968] (0/2) Epoch 8, batch 13650, giga_loss[loss=0.2816, simple_loss=0.3491, pruned_loss=0.1071, over 28658.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3497, pruned_loss=0.1015, over 5658821.69 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3562, pruned_loss=0.1041, over 5754713.77 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3504, pruned_loss=0.1019, over 5647171.93 frames. ], batch size: 242, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:40:29,297 INFO [train.py:968] (0/2) Epoch 8, batch 13700, libri_loss[loss=0.2667, simple_loss=0.3275, pruned_loss=0.1029, over 29502.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3506, pruned_loss=0.1023, over 5665650.82 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.356, pruned_loss=0.1042, over 5750017.16 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3512, pruned_loss=0.1025, over 5657355.40 frames. ], batch size: 70, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:40:50,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3423, 1.7965, 1.2557, 0.7346], device='cuda:0'), covar=tensor([0.3869, 0.2146, 0.2017, 0.3581], device='cuda:0'), in_proj_covar=tensor([0.1464, 0.1380, 0.1435, 0.1207], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 04:41:00,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:41:16,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.397e+02 1.343e+03 1.786e+03 2.414e+03 7.988e+03, threshold=3.571e+03, percent-clipped=5.0
+2023-03-04 04:41:30,241 INFO [train.py:968] (0/2) Epoch 8, batch 13750, giga_loss[loss=0.2684, simple_loss=0.341, pruned_loss=0.0979, over 28942.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3471, pruned_loss=0.09977, over 5665182.57 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3562, pruned_loss=0.1042, over 5751668.91 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3474, pruned_loss=0.09979, over 5655939.64 frames. ], batch size: 136, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:41:49,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:42:16,308 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:42:18,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:42:19,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:42:32,641 INFO [train.py:968] (0/2) Epoch 8, batch 13800, giga_loss[loss=0.2361, simple_loss=0.3243, pruned_loss=0.07399, over 28496.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3453, pruned_loss=0.09717, over 5666046.53 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3556, pruned_loss=0.1039, over 5754253.36 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3458, pruned_loss=0.09735, over 5654275.55 frames. ], batch size: 336, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:42:55,106 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:42:55,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2082, 3.2952, 1.3041, 1.4414], device='cuda:0'), covar=tensor([0.1102, 0.0387, 0.1019, 0.1439], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0491, 0.0327, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 04:43:22,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.771e+02 1.265e+03 1.499e+03 2.248e+03 6.007e+03, threshold=2.998e+03, percent-clipped=4.0
+2023-03-04 04:43:38,690 INFO [train.py:968] (0/2) Epoch 8, batch 13850, giga_loss[loss=0.2401, simple_loss=0.3222, pruned_loss=0.07902, over 28892.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.343, pruned_loss=0.09575, over 5664347.11 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3556, pruned_loss=0.1039, over 5755458.05 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3433, pruned_loss=0.09587, over 5653372.17 frames. ], batch size: 164, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:43:58,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:44:01,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:44:35,572 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:44:36,592 INFO [train.py:968] (0/2) Epoch 8, batch 13900, libri_loss[loss=0.2683, simple_loss=0.3469, pruned_loss=0.09485, over 29494.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3415, pruned_loss=0.09645, over 5660182.80 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.355, pruned_loss=0.1038, over 5746700.02 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3417, pruned_loss=0.09629, over 5655630.41 frames. ], batch size: 85, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:44:41,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:45:10,595 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:45:12,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:45:18,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.231e+02 1.442e+03 2.030e+03 2.881e+03 5.810e+03, threshold=4.061e+03, percent-clipped=23.0
+2023-03-04 04:45:28,642 INFO [train.py:968] (0/2) Epoch 8, batch 13950, giga_loss[loss=0.2494, simple_loss=0.3187, pruned_loss=0.09009, over 28655.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3408, pruned_loss=0.09643, over 5671749.66 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.355, pruned_loss=0.1039, over 5751341.63 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3405, pruned_loss=0.09595, over 5660905.55 frames. ], batch size: 92, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:45:44,547 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:45:56,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:46:03,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4127, 1.6655, 1.5639, 1.5097], device='cuda:0'), covar=tensor([0.1173, 0.1666, 0.1493, 0.1396], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0703, 0.0631, 0.0629], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 04:46:28,612 INFO [train.py:968] (0/2) Epoch 8, batch 14000, giga_loss[loss=0.226, simple_loss=0.2979, pruned_loss=0.07708, over 24419.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3412, pruned_loss=0.09646, over 5655897.11 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.355, pruned_loss=0.1039, over 5744690.28 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3407, pruned_loss=0.09588, over 5652482.74 frames. ], batch size: 705, lr: 4.10e-03, grad_scale: 8.0
+2023-03-04 04:46:57,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3618, 2.9563, 1.4914, 1.4736], device='cuda:0'), covar=tensor([0.0834, 0.0282, 0.0835, 0.1225], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0490, 0.0326, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 04:47:18,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.446e+02 1.375e+03 1.718e+03 2.485e+03 5.381e+03, threshold=3.436e+03, percent-clipped=5.0
+2023-03-04 04:47:32,704 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:47:33,005 INFO [train.py:968] (0/2) Epoch 8, batch 14050, giga_loss[loss=0.2636, simple_loss=0.3413, pruned_loss=0.09294, over 28153.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3436, pruned_loss=0.09703, over 5654574.53 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3549, pruned_loss=0.1039, over 5746389.98 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3432, pruned_loss=0.0965, over 5649618.40 frames. ], batch size: 412, lr: 4.10e-03, grad_scale: 8.0
+2023-03-04 04:47:38,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:48:18,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:48:43,236 INFO [train.py:968] (0/2) Epoch 8, batch 14100, giga_loss[loss=0.2527, simple_loss=0.3317, pruned_loss=0.08685, over 28725.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3411, pruned_loss=0.09513, over 5661633.25 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.355, pruned_loss=0.104, over 5747068.33 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3406, pruned_loss=0.09461, over 5656307.35 frames. ], batch size: 262, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:49:06,040 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:49:07,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:49:33,931 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332917.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 04:49:34,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.767e+02 1.278e+03 1.705e+03 2.231e+03 5.226e+03, threshold=3.409e+03, percent-clipped=6.0
+2023-03-04 04:49:34,824 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:49:46,325 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:49:50,076 INFO [train.py:968] (0/2) Epoch 8, batch 14150, giga_loss[loss=0.3028, simple_loss=0.3725, pruned_loss=0.1165, over 28913.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3418, pruned_loss=0.09586, over 5676125.94 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3547, pruned_loss=0.104, over 5750022.24 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3413, pruned_loss=0.09522, over 5667500.15 frames. ], batch size: 284, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:50:56,091 INFO [train.py:968] (0/2) Epoch 8, batch 14200, giga_loss[loss=0.2908, simple_loss=0.3733, pruned_loss=0.1041, over 28503.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3459, pruned_loss=0.0979, over 5680635.18 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3548, pruned_loss=0.1042, over 5754109.03 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.345, pruned_loss=0.09695, over 5668189.39 frames. ], batch size: 336, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:51:45,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.709e+02 1.442e+03 1.976e+03 2.589e+03 9.232e+03, threshold=3.952e+03, percent-clipped=12.0
+2023-03-04 04:51:59,648 INFO [train.py:968] (0/2) Epoch 8, batch 14250, giga_loss[loss=0.2875, simple_loss=0.3599, pruned_loss=0.1075, over 27528.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3495, pruned_loss=0.09744, over 5676673.81 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3545, pruned_loss=0.1039, over 5755834.92 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.349, pruned_loss=0.0968, over 5663945.57 frames. ], batch size: 472, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:52:40,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=333061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:52:43,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=333064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:53:00,829 INFO [train.py:968] (0/2) Epoch 8, batch 14300, giga_loss[loss=0.2788, simple_loss=0.3622, pruned_loss=0.09765, over 28502.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3499, pruned_loss=0.09637, over 5665860.30 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3542, pruned_loss=0.1038, over 5750001.05 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3496, pruned_loss=0.09577, over 5659752.20 frames. ], batch size: 85, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:53:17,294 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=333093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:53:31,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2825, 1.7274, 1.7030, 1.2605], device='cuda:0'), covar=tensor([0.1652, 0.2153, 0.1273, 0.1497], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0706, 0.0824, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 04:53:45,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.093e+02 1.278e+03 1.724e+03 2.357e+03 5.800e+03, threshold=3.448e+03, percent-clipped=7.0
+2023-03-04 04:53:57,775 INFO [train.py:968] (0/2) Epoch 8, batch 14350, giga_loss[loss=0.2327, simple_loss=0.3223, pruned_loss=0.07159, over 28884.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3487, pruned_loss=0.09453, over 5672643.51 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3537, pruned_loss=0.1035, over 5750383.73 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3488, pruned_loss=0.09418, over 5665555.66 frames. ], batch size: 164, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:55:02,943 INFO [train.py:968] (0/2) Epoch 8, batch 14400, giga_loss[loss=0.2359, simple_loss=0.3233, pruned_loss=0.07423, over 28887.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3482, pruned_loss=0.09487, over 5672866.32 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3535, pruned_loss=0.1034, over 5752920.51 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3485, pruned_loss=0.09456, over 5663794.39 frames. ], batch size: 174, lr: 4.10e-03, grad_scale: 8.0
+2023-03-04 04:55:06,761 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-04 04:55:52,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.691e+02 1.269e+03 1.696e+03 2.187e+03 3.768e+03, threshold=3.392e+03, percent-clipped=3.0
+2023-03-04 04:55:56,875 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=333222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:56:05,899 INFO [train.py:968] (0/2) Epoch 8, batch 14450, giga_loss[loss=0.3162, simple_loss=0.3659, pruned_loss=0.1333, over 26881.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3473, pruned_loss=0.09562, over 5680211.97 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3531, pruned_loss=0.1032, over 5753803.48 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3476, pruned_loss=0.09534, over 5670832.15 frames. ], batch size: 555, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:57:19,910 INFO [train.py:968] (0/2) Epoch 8, batch 14500, giga_loss[loss=0.241, simple_loss=0.3242, pruned_loss=0.07888, over 28749.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3493, pruned_loss=0.09778, over 5688736.96 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3525, pruned_loss=0.103, over 5755273.68 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.35, pruned_loss=0.09757, over 5678275.59 frames. ], batch size: 119, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:57:43,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=333292.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 04:57:45,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4255, 1.6996, 1.2836, 1.7947], device='cuda:0'), covar=tensor([0.2366, 0.2189, 0.2483, 0.1974], device='cuda:0'), in_proj_covar=tensor([0.1205, 0.0899, 0.1064, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 04:58:04,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3218, 1.5946, 1.6332, 1.4608], device='cuda:0'), covar=tensor([0.1117, 0.1266, 0.1233, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0708, 0.0639, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 04:58:31,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.057e+02 1.297e+03 1.884e+03 3.016e+03 1.009e+04, threshold=3.767e+03, percent-clipped=20.0
+2023-03-04 04:58:47,099 INFO [train.py:968] (0/2) Epoch 8, batch 14550, giga_loss[loss=0.2651, simple_loss=0.3426, pruned_loss=0.09376, over 28723.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3453, pruned_loss=0.09601, over 5681764.26 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3524, pruned_loss=0.1029, over 5756486.67 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3458, pruned_loss=0.09585, over 5671435.38 frames. ], batch size: 262, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 04:59:54,699 INFO [train.py:968] (0/2) Epoch 8, batch 14600, giga_loss[loss=0.2542, simple_loss=0.3343, pruned_loss=0.0871, over 28663.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3429, pruned_loss=0.09414, over 5681840.19 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3524, pruned_loss=0.103, over 5755259.31 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3433, pruned_loss=0.09387, over 5673940.75 frames. ], batch size: 307, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:00:19,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=333395.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:00:19,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5744, 1.7205, 1.6459, 1.5294], device='cuda:0'), covar=tensor([0.1066, 0.1617, 0.1398, 0.1500], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0709, 0.0639, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 05:00:35,180 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5094, 1.7324, 1.3084, 1.5640], device='cuda:0'), covar=tensor([0.0718, 0.0276, 0.0327, 0.0781], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0121, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0051, 0.0047, 0.0078], device='cuda:0')
+2023-03-04 05:00:47,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.216e+02 1.142e+03 1.443e+03 2.178e+03 4.767e+03, threshold=2.886e+03, percent-clipped=4.0
+2023-03-04 05:00:48,509 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=333420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:01:01,184 INFO [train.py:968] (0/2) Epoch 8, batch 14650, giga_loss[loss=0.2814, simple_loss=0.3638, pruned_loss=0.09944, over 28495.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3409, pruned_loss=0.09352, over 5681807.75 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.352, pruned_loss=0.1028, over 5758877.86 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3412, pruned_loss=0.09322, over 5670351.78 frames. ], batch size: 336, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:01:10,399 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=333435.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:01:15,737 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=333438.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:01:44,555 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=333467.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:02:00,777 INFO [train.py:968] (0/2) Epoch 8, batch 14700, giga_loss[loss=0.3175, simple_loss=0.3877, pruned_loss=0.1236, over 28408.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3461, pruned_loss=0.09696, over 5669861.19 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3519, pruned_loss=0.1027, over 5751448.50 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3462, pruned_loss=0.09657, over 5665258.92 frames. ], batch size: 369, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:02:07,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3085, 3.1458, 2.9962, 1.4162], device='cuda:0'), covar=tensor([0.0814, 0.0931, 0.0954, 0.2196], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0887, 0.0792, 0.0624], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 05:02:33,411 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 05:02:47,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.848e+02 1.639e+03 2.186e+03 3.185e+03 7.307e+03, threshold=4.372e+03, percent-clipped=32.0
+2023-03-04 05:03:00,591 INFO [train.py:968] (0/2) Epoch 8, batch 14750, giga_loss[loss=0.2771, simple_loss=0.343, pruned_loss=0.1056, over 28188.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3481, pruned_loss=0.09841, over 5682536.87 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3521, pruned_loss=0.1026, over 5754628.16 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3479, pruned_loss=0.09801, over 5673904.55 frames. ], batch size: 412, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:03:09,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5628, 1.9308, 1.4234, 1.3732], device='cuda:0'), covar=tensor([0.1514, 0.0987, 0.1050, 0.1147], device='cuda:0'), in_proj_covar=tensor([0.1568, 0.1395, 0.1357, 0.1479], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-04 05:04:00,602 INFO [train.py:968] (0/2) Epoch 8, batch 14800, giga_loss[loss=0.2569, simple_loss=0.3353, pruned_loss=0.08921, over 28868.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3447, pruned_loss=0.09745, over 5675658.98 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3513, pruned_loss=0.1023, over 5748952.88 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3452, pruned_loss=0.09736, over 5670881.35 frames. ], batch size: 227, lr: 4.09e-03, grad_scale: 8.0
+2023-03-04 05:04:26,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=333597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:04:48,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=333615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:04:51,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.934e+02 1.469e+03 1.952e+03 2.683e+03 6.653e+03, threshold=3.903e+03, percent-clipped=4.0
+2023-03-04 05:05:02,233 INFO [train.py:968] (0/2) Epoch 8, batch 14850, giga_loss[loss=0.2639, simple_loss=0.3427, pruned_loss=0.09258, over 28929.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3443, pruned_loss=0.09747, over 5684439.78 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3506, pruned_loss=0.1018, over 5752494.61 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3451, pruned_loss=0.09765, over 5675335.06 frames. ], batch size: 213, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:06:06,284 INFO [train.py:968] (0/2) Epoch 8, batch 14900, giga_loss[loss=0.2051, simple_loss=0.2771, pruned_loss=0.06656, over 24632.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3459, pruned_loss=0.09824, over 5683976.88 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3504, pruned_loss=0.1018, over 5756120.63 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3466, pruned_loss=0.09831, over 5671969.79 frames. ], batch size: 705, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:07:01,269 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.751e+02 1.370e+03 1.882e+03 2.917e+03 6.496e+03, threshold=3.763e+03, percent-clipped=10.0
+2023-03-04 05:07:13,680 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-04 05:07:16,143 INFO [train.py:968] (0/2) Epoch 8, batch 14950, giga_loss[loss=0.238, simple_loss=0.3254, pruned_loss=0.07531, over 28920.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3471, pruned_loss=0.098, over 5684236.05 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3501, pruned_loss=0.1018, over 5758926.74 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3479, pruned_loss=0.09808, over 5670917.49 frames. ], batch size: 227, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:07:32,599 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=333740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:07:37,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=333743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:08:16,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=333770.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:08:22,052 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=333772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:08:36,516 INFO [train.py:968] (0/2) Epoch 8, batch 15000, giga_loss[loss=0.29, simple_loss=0.3393, pruned_loss=0.1203, over 24359.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3457, pruned_loss=0.09706, over 5665690.01 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3499, pruned_loss=0.1018, over 5753207.85 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3464, pruned_loss=0.09701, over 5658198.30 frames. ], batch size: 705, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:08:36,521 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 05:08:45,239 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2108, simple_loss=0.3094, pruned_loss=0.05613, over 944034.00 frames. 
+2023-03-04 05:08:45,240 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 05:09:08,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=333795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:09:14,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=333799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:09:39,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.632e+02 1.444e+03 1.880e+03 2.657e+03 8.076e+03, threshold=3.760e+03, percent-clipped=11.0
+2023-03-04 05:09:50,333 INFO [train.py:968] (0/2) Epoch 8, batch 15050, giga_loss[loss=0.2505, simple_loss=0.3153, pruned_loss=0.09289, over 26918.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3422, pruned_loss=0.09636, over 5663273.76 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3497, pruned_loss=0.1017, over 5749899.46 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3428, pruned_loss=0.09627, over 5658064.80 frames. ], batch size: 555, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:09:51,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3683, 1.6447, 1.2191, 1.9118], device='cuda:0'), covar=tensor([0.2412, 0.2097, 0.2425, 0.2078], device='cuda:0'), in_proj_covar=tensor([0.1210, 0.0901, 0.1066, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:10:20,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8190, 4.6284, 4.3577, 1.9107], device='cuda:0'), covar=tensor([0.0518, 0.0694, 0.0868, 0.2133], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0883, 0.0782, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 05:10:37,555 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7967, 2.4953, 2.2302, 1.7306], device='cuda:0'), covar=tensor([0.1662, 0.1940, 0.1238, 0.1474], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0702, 0.0826, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 05:10:53,950 INFO [train.py:968] (0/2) Epoch 8, batch 15100, giga_loss[loss=0.2946, simple_loss=0.3606, pruned_loss=0.1142, over 28939.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.335, pruned_loss=0.09271, over 5666205.97 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3492, pruned_loss=0.1013, over 5751257.91 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3357, pruned_loss=0.09276, over 5658339.41 frames. ], batch size: 213, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:11:17,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4264, 2.9837, 1.5606, 1.5050], device='cuda:0'), covar=tensor([0.0775, 0.0290, 0.0791, 0.1140], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0488, 0.0326, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0027, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 05:11:34,075 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=333913.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:11:37,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=333916.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:11:41,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.850e+02 1.362e+03 1.686e+03 2.450e+03 4.291e+03, threshold=3.372e+03, percent-clipped=4.0
+2023-03-04 05:11:52,418 INFO [train.py:968] (0/2) Epoch 8, batch 15150, giga_loss[loss=0.2537, simple_loss=0.3387, pruned_loss=0.08439, over 28704.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3349, pruned_loss=0.09248, over 5666127.91 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3491, pruned_loss=0.1012, over 5746371.21 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3352, pruned_loss=0.09238, over 5661411.12 frames. ], batch size: 119, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:12:07,218 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=333938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:12:11,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=333941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:12:14,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=333945.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:12:38,821 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=333970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:12:47,573 INFO [train.py:968] (0/2) Epoch 8, batch 15200, giga_loss[loss=0.2506, simple_loss=0.3309, pruned_loss=0.08513, over 28500.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3367, pruned_loss=0.09418, over 5657966.21 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3486, pruned_loss=0.101, over 5741480.42 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3369, pruned_loss=0.09409, over 5655095.23 frames. ], batch size: 369, lr: 4.09e-03, grad_scale: 8.0
+2023-03-04 05:12:59,906 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=333990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:13:00,026 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-04 05:13:10,013 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-334000.pt
+2023-03-04 05:13:39,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.218e+02 1.324e+03 1.831e+03 2.890e+03 6.798e+03, threshold=3.662e+03, percent-clipped=16.0
+2023-03-04 05:13:49,185 INFO [train.py:968] (0/2) Epoch 8, batch 15250, giga_loss[loss=0.2416, simple_loss=0.3253, pruned_loss=0.07898, over 28709.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3355, pruned_loss=0.09308, over 5659550.16 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3485, pruned_loss=0.1009, over 5742646.64 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3355, pruned_loss=0.09294, over 5654690.55 frames. ], batch size: 262, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:14:32,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2972, 1.6202, 1.2398, 1.5585], device='cuda:0'), covar=tensor([0.2477, 0.2318, 0.2578, 0.2003], device='cuda:0'), in_proj_covar=tensor([0.1208, 0.0903, 0.1070, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:14:48,709 INFO [train.py:968] (0/2) Epoch 8, batch 15300, giga_loss[loss=0.2448, simple_loss=0.3318, pruned_loss=0.07887, over 28414.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3341, pruned_loss=0.09153, over 5651191.96 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3484, pruned_loss=0.1009, over 5735257.91 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3339, pruned_loss=0.09123, over 5651336.85 frames. ], batch size: 336, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:15:49,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.854e+02 1.337e+03 1.686e+03 2.469e+03 4.885e+03, threshold=3.371e+03, percent-clipped=4.0
+2023-03-04 05:15:59,786 INFO [train.py:968] (0/2) Epoch 8, batch 15350, libri_loss[loss=0.2444, simple_loss=0.3246, pruned_loss=0.0821, over 29640.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3324, pruned_loss=0.09101, over 5655782.23 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.348, pruned_loss=0.1007, over 5737430.30 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3324, pruned_loss=0.09087, over 5652550.57 frames. ], batch size: 88, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:16:06,450 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=334133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:16:06,567 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-04 05:16:11,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=334136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:16:50,002 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=334165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:17:01,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=334174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:17:05,648 INFO [train.py:968] (0/2) Epoch 8, batch 15400, giga_loss[loss=0.2429, simple_loss=0.3269, pruned_loss=0.07942, over 28709.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3326, pruned_loss=0.09071, over 5659913.16 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3477, pruned_loss=0.1005, over 5740873.03 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3327, pruned_loss=0.0906, over 5652692.36 frames. ], batch size: 262, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:17:40,463 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-04 05:17:44,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1245, 1.0545, 3.7279, 3.0287], device='cuda:0'), covar=tensor([0.1537, 0.2510, 0.0391, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0597, 0.0554, 0.0783, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:18:00,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.522e+02 1.285e+03 1.685e+03 2.412e+03 3.994e+03, threshold=3.370e+03, percent-clipped=3.0
+2023-03-04 05:18:13,366 INFO [train.py:968] (0/2) Epoch 8, batch 15450, giga_loss[loss=0.2596, simple_loss=0.3382, pruned_loss=0.09053, over 28419.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3326, pruned_loss=0.09062, over 5657227.81 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3476, pruned_loss=0.1004, over 5743170.16 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3325, pruned_loss=0.09044, over 5648286.02 frames. ], batch size: 369, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:18:32,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2277, 3.0587, 2.8563, 1.3970], device='cuda:0'), covar=tensor([0.0840, 0.0944, 0.0941, 0.2348], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0886, 0.0780, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 05:19:01,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7289, 1.0744, 2.8632, 2.7022], device='cuda:0'), covar=tensor([0.1580, 0.2242, 0.0543, 0.0903], device='cuda:0'), in_proj_covar=tensor([0.0596, 0.0552, 0.0784, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:19:13,333 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-04 05:19:17,821 INFO [train.py:968] (0/2) Epoch 8, batch 15500, giga_loss[loss=0.2258, simple_loss=0.3099, pruned_loss=0.07085, over 28857.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3335, pruned_loss=0.09213, over 5657470.39 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3469, pruned_loss=0.1001, over 5744172.21 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3337, pruned_loss=0.09212, over 5647399.12 frames. ], batch size: 136, lr: 4.09e-03, grad_scale: 2.0
+2023-03-04 05:19:59,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6335, 1.0656, 2.8671, 2.6646], device='cuda:0'), covar=tensor([0.1688, 0.2349, 0.0520, 0.0929], device='cuda:0'), in_proj_covar=tensor([0.0600, 0.0555, 0.0790, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:20:07,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=334317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:20:10,921 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=334320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:20:12,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.549e+02 1.256e+03 1.981e+03 3.190e+03 1.284e+04, threshold=3.962e+03, percent-clipped=19.0
+2023-03-04 05:20:18,926 INFO [train.py:968] (0/2) Epoch 8, batch 15550, libri_loss[loss=0.2963, simple_loss=0.3677, pruned_loss=0.1125, over 29492.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3325, pruned_loss=0.09069, over 5666618.66 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3467, pruned_loss=0.0999, over 5747357.50 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3326, pruned_loss=0.09068, over 5654071.50 frames. ], batch size: 85, lr: 4.09e-03, grad_scale: 2.0
+2023-03-04 05:20:30,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3153, 3.0945, 1.4441, 1.4396], device='cuda:0'), covar=tensor([0.0949, 0.0304, 0.0871, 0.1300], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0484, 0.0323, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0027, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 05:20:42,182 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=334349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:21:10,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2319, 2.5909, 1.3340, 1.3778], device='cuda:0'), covar=tensor([0.0918, 0.0360, 0.0843, 0.1285], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0485, 0.0323, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0027, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 05:21:19,530 INFO [train.py:968] (0/2) Epoch 8, batch 15600, giga_loss[loss=0.263, simple_loss=0.3505, pruned_loss=0.08777, over 28988.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3346, pruned_loss=0.08999, over 5676014.54 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3467, pruned_loss=0.09987, over 5749991.37 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3345, pruned_loss=0.08987, over 5662776.50 frames. ], batch size: 155, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:21:55,677 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=334409.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:22:08,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.288e+02 1.180e+03 1.545e+03 2.147e+03 4.338e+03, threshold=3.089e+03, percent-clipped=4.0
+2023-03-04 05:22:15,697 INFO [train.py:968] (0/2) Epoch 8, batch 15650, giga_loss[loss=0.3251, simple_loss=0.382, pruned_loss=0.1342, over 27699.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3375, pruned_loss=0.09204, over 5673113.98 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3462, pruned_loss=0.09963, over 5756459.42 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3375, pruned_loss=0.09184, over 5653788.20 frames. ], batch size: 472, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:23:16,177 INFO [train.py:968] (0/2) Epoch 8, batch 15700, giga_loss[loss=0.2548, simple_loss=0.3364, pruned_loss=0.08662, over 28496.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3395, pruned_loss=0.09295, over 5666248.22 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3462, pruned_loss=0.0997, over 5746820.60 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3395, pruned_loss=0.09266, over 5657779.52 frames. ], batch size: 336, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:23:44,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3676, 1.5449, 1.2586, 1.5317], device='cuda:0'), covar=tensor([0.2284, 0.2174, 0.2328, 0.1911], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.0894, 0.1061, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:24:04,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.219e+02 1.498e+03 1.916e+03 2.602e+03 9.100e+03, threshold=3.832e+03, percent-clipped=22.0
+2023-03-04 05:24:10,544 INFO [train.py:968] (0/2) Epoch 8, batch 15750, giga_loss[loss=0.3376, simple_loss=0.401, pruned_loss=0.1371, over 28854.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3396, pruned_loss=0.09296, over 5683822.49 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3458, pruned_loss=0.09946, over 5752079.44 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3396, pruned_loss=0.09264, over 5669161.34 frames. ], batch size: 284, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:25:10,449 INFO [train.py:968] (0/2) Epoch 8, batch 15800, giga_loss[loss=0.2172, simple_loss=0.2839, pruned_loss=0.07524, over 24807.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3376, pruned_loss=0.09156, over 5688477.13 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3457, pruned_loss=0.09945, over 5754195.57 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3375, pruned_loss=0.09118, over 5673770.28 frames. ], batch size: 705, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:25:29,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 05:25:47,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3044, 1.5153, 1.4054, 1.5626], device='cuda:0'), covar=tensor([0.0760, 0.0317, 0.0324, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0114, 0.0121, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0050, 0.0047, 0.0078], device='cuda:0')
+2023-03-04 05:26:07,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.388e+02 1.218e+03 1.459e+03 1.934e+03 3.534e+03, threshold=2.918e+03, percent-clipped=0.0
+2023-03-04 05:26:14,442 INFO [train.py:968] (0/2) Epoch 8, batch 15850, giga_loss[loss=0.2433, simple_loss=0.3193, pruned_loss=0.08369, over 27724.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3362, pruned_loss=0.09051, over 5682628.78 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3455, pruned_loss=0.09935, over 5747019.53 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3362, pruned_loss=0.09018, over 5676664.63 frames. ], batch size: 472, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:27:13,072 INFO [train.py:968] (0/2) Epoch 8, batch 15900, giga_loss[loss=0.2834, simple_loss=0.3572, pruned_loss=0.1048, over 28137.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.334, pruned_loss=0.08989, over 5680247.85 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3454, pruned_loss=0.09931, over 5749808.76 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3338, pruned_loss=0.08945, over 5671474.58 frames. ], batch size: 412, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:27:50,170 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-04 05:28:02,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.296e+02 1.351e+03 1.691e+03 2.600e+03 7.590e+03, threshold=3.383e+03, percent-clipped=19.0
+2023-03-04 05:28:10,118 INFO [train.py:968] (0/2) Epoch 8, batch 15950, giga_loss[loss=0.2679, simple_loss=0.3505, pruned_loss=0.09263, over 28685.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3358, pruned_loss=0.09104, over 5673999.51 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3448, pruned_loss=0.09899, over 5745404.47 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3358, pruned_loss=0.09064, over 5668907.15 frames. ], batch size: 307, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:29:14,320 INFO [train.py:968] (0/2) Epoch 8, batch 16000, giga_loss[loss=0.2911, simple_loss=0.3659, pruned_loss=0.1082, over 29022.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3378, pruned_loss=0.09205, over 5682787.93 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3444, pruned_loss=0.09878, over 5749037.72 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.338, pruned_loss=0.09176, over 5673808.96 frames. ], batch size: 285, lr: 4.09e-03, grad_scale: 8.0
+2023-03-04 05:29:21,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=334784.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:29:58,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 05:30:09,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.480e+02 1.304e+03 1.773e+03 2.641e+03 6.127e+03, threshold=3.547e+03, percent-clipped=9.0
+2023-03-04 05:30:19,681 INFO [train.py:968] (0/2) Epoch 8, batch 16050, giga_loss[loss=0.3085, simple_loss=0.3633, pruned_loss=0.1269, over 26870.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3394, pruned_loss=0.09384, over 5676466.94 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3445, pruned_loss=0.09893, over 5751387.46 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3393, pruned_loss=0.09332, over 5665421.90 frames. ], batch size: 555, lr: 4.09e-03, grad_scale: 8.0
+2023-03-04 05:31:13,974 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5498, 1.7148, 1.5268, 1.6934], device='cuda:0'), covar=tensor([0.2114, 0.2045, 0.2192, 0.1692], device='cuda:0'), in_proj_covar=tensor([0.1197, 0.0890, 0.1058, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:31:18,640 INFO [train.py:968] (0/2) Epoch 8, batch 16100, giga_loss[loss=0.2656, simple_loss=0.353, pruned_loss=0.0891, over 28898.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.343, pruned_loss=0.09559, over 5669666.61 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3447, pruned_loss=0.09902, over 5740130.32 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3428, pruned_loss=0.09508, over 5670458.34 frames. ], batch size: 164, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:31:36,039 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 05:32:09,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.165e+02 1.636e+03 2.218e+03 2.968e+03 6.533e+03, threshold=4.436e+03, percent-clipped=12.0
+2023-03-04 05:32:14,526 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=334927.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:32:15,517 INFO [train.py:968] (0/2) Epoch 8, batch 16150, giga_loss[loss=0.2793, simple_loss=0.365, pruned_loss=0.09673, over 28046.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3448, pruned_loss=0.09562, over 5676398.13 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3443, pruned_loss=0.09886, over 5742082.84 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3449, pruned_loss=0.09529, over 5674104.23 frames. ], batch size: 412, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:32:17,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=334930.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:32:30,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2933, 1.5248, 1.2032, 1.2205], device='cuda:0'), covar=tensor([0.1567, 0.1094, 0.0903, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.1581, 0.1392, 0.1346, 0.1498], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-04 05:32:49,428 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-04 05:32:51,683 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=334959.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:33:16,453 INFO [train.py:968] (0/2) Epoch 8, batch 16200, giga_loss[loss=0.3249, simple_loss=0.388, pruned_loss=0.1309, over 28976.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3459, pruned_loss=0.09651, over 5675544.51 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3444, pruned_loss=0.09914, over 5735640.15 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3459, pruned_loss=0.09588, over 5677386.65 frames. ], batch size: 285, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:33:50,452 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=335002.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:34:16,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.035e+02 1.474e+03 2.024e+03 2.697e+03 5.440e+03, threshold=4.047e+03, percent-clipped=6.0
+2023-03-04 05:34:24,313 INFO [train.py:968] (0/2) Epoch 8, batch 16250, giga_loss[loss=0.2826, simple_loss=0.3421, pruned_loss=0.1115, over 26891.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3437, pruned_loss=0.09576, over 5683181.74 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3444, pruned_loss=0.09912, over 5738910.32 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3438, pruned_loss=0.09522, over 5680837.92 frames. ], batch size: 555, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:34:25,421 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 05:34:33,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7387, 1.8155, 1.3007, 1.4923], device='cuda:0'), covar=tensor([0.0678, 0.0493, 0.0996, 0.0894], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0437, 0.0494, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 05:35:27,425 INFO [train.py:968] (0/2) Epoch 8, batch 16300, giga_loss[loss=0.2542, simple_loss=0.3384, pruned_loss=0.08499, over 28330.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3434, pruned_loss=0.09573, over 5683532.05 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3442, pruned_loss=0.09893, over 5742090.67 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3436, pruned_loss=0.09538, over 5677593.96 frames. ], batch size: 368, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:36:20,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.004e+02 1.333e+03 1.819e+03 2.582e+03 1.442e+04, threshold=3.637e+03, percent-clipped=7.0
+2023-03-04 05:36:27,902 INFO [train.py:968] (0/2) Epoch 8, batch 16350, giga_loss[loss=0.2599, simple_loss=0.3316, pruned_loss=0.09409, over 29030.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3415, pruned_loss=0.09493, over 5678083.36 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3442, pruned_loss=0.09883, over 5746887.28 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3416, pruned_loss=0.09462, over 5666855.71 frames. ], batch size: 136, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:37:28,780 INFO [train.py:968] (0/2) Epoch 8, batch 16400, giga_loss[loss=0.2002, simple_loss=0.2781, pruned_loss=0.06114, over 28419.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3381, pruned_loss=0.09401, over 5677000.61 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3437, pruned_loss=0.09854, over 5748474.03 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3386, pruned_loss=0.09394, over 5665094.23 frames. ], batch size: 71, lr: 4.08e-03, grad_scale: 8.0
+2023-03-04 05:38:19,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.1664, 5.9518, 5.6507, 2.8782], device='cuda:0'), covar=tensor([0.0347, 0.0592, 0.0788, 0.1551], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0882, 0.0785, 0.0620], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 05:38:21,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.640e+02 1.295e+03 1.666e+03 2.177e+03 5.573e+03, threshold=3.332e+03, percent-clipped=5.0
+2023-03-04 05:38:26,844 INFO [train.py:968] (0/2) Epoch 8, batch 16450, giga_loss[loss=0.2783, simple_loss=0.366, pruned_loss=0.09531, over 28887.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3379, pruned_loss=0.09351, over 5666319.25 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3437, pruned_loss=0.09869, over 5732677.74 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3381, pruned_loss=0.0932, over 5669545.80 frames. ], batch size: 284, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:38:41,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4548, 1.6786, 1.6410, 1.4927], device='cuda:0'), covar=tensor([0.1187, 0.1510, 0.1637, 0.1493], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0700, 0.0629, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 05:38:49,163 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4806, 1.8167, 1.8196, 1.3979], device='cuda:0'), covar=tensor([0.1639, 0.2180, 0.1284, 0.1429], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0697, 0.0818, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 05:39:23,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5434, 2.3376, 1.6770, 0.5927], device='cuda:0'), covar=tensor([0.3441, 0.1842, 0.2800, 0.3915], device='cuda:0'), in_proj_covar=tensor([0.1473, 0.1409, 0.1453, 0.1215], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 05:39:28,020 INFO [train.py:968] (0/2) Epoch 8, batch 16500, giga_loss[loss=0.2415, simple_loss=0.3066, pruned_loss=0.08818, over 24482.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3367, pruned_loss=0.09177, over 5658877.21 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3435, pruned_loss=0.09856, over 5727987.22 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3369, pruned_loss=0.09153, over 5663508.20 frames. ], batch size: 705, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:39:39,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1273, 1.1987, 3.6729, 3.1709], device='cuda:0'), covar=tensor([0.1600, 0.2447, 0.0383, 0.0963], device='cuda:0'), in_proj_covar=tensor([0.0599, 0.0556, 0.0787, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:39:55,425 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2620, 1.8640, 1.3846, 0.3496], device='cuda:0'), covar=tensor([0.2572, 0.1825, 0.2760, 0.3463], device='cuda:0'), in_proj_covar=tensor([0.1479, 0.1414, 0.1459, 0.1220], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 05:40:17,894 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.006e+02 1.443e+03 1.919e+03 2.750e+03 9.680e+03, threshold=3.838e+03, percent-clipped=18.0
+2023-03-04 05:40:23,731 INFO [train.py:968] (0/2) Epoch 8, batch 16550, giga_loss[loss=0.2679, simple_loss=0.3534, pruned_loss=0.09118, over 28178.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3377, pruned_loss=0.09126, over 5671681.84 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3435, pruned_loss=0.09868, over 5733414.12 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3377, pruned_loss=0.09071, over 5668243.86 frames. ], batch size: 412, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:40:30,699 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=335336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:40:40,252 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=335344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:41:17,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=335377.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:41:18,921 INFO [train.py:968] (0/2) Epoch 8, batch 16600, giga_loss[loss=0.2705, simple_loss=0.3542, pruned_loss=0.09335, over 28504.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3394, pruned_loss=0.09086, over 5659387.76 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3433, pruned_loss=0.09869, over 5725378.02 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3394, pruned_loss=0.0902, over 5662594.00 frames. ], batch size: 336, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:41:51,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=335407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:42:06,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.834e+02 1.194e+03 1.536e+03 2.183e+03 4.300e+03, threshold=3.072e+03, percent-clipped=2.0
+2023-03-04 05:42:12,453 INFO [train.py:968] (0/2) Epoch 8, batch 16650, giga_loss[loss=0.3058, simple_loss=0.3719, pruned_loss=0.1198, over 28297.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3406, pruned_loss=0.09079, over 5674032.59 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3434, pruned_loss=0.09851, over 5726052.55 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3404, pruned_loss=0.09014, over 5674086.57 frames. ], batch size: 368, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:42:53,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4161, 1.5511, 1.2921, 1.6231], device='cuda:0'), covar=tensor([0.2273, 0.2144, 0.2301, 0.1993], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.0895, 0.1063, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:43:16,132 INFO [train.py:968] (0/2) Epoch 8, batch 16700, giga_loss[loss=0.2526, simple_loss=0.3341, pruned_loss=0.0855, over 28711.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3417, pruned_loss=0.09176, over 5675065.90 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3434, pruned_loss=0.09842, over 5722064.78 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3415, pruned_loss=0.09123, over 5678192.58 frames. ], batch size: 262, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:44:10,785 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=335520.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:44:14,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=335523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:44:16,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.856e+02 1.396e+03 1.764e+03 2.517e+03 5.303e+03, threshold=3.528e+03, percent-clipped=14.0
+2023-03-04 05:44:23,029 INFO [train.py:968] (0/2) Epoch 8, batch 16750, libri_loss[loss=0.3514, simple_loss=0.3923, pruned_loss=0.1552, over 29677.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3423, pruned_loss=0.09204, over 5666205.99 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.344, pruned_loss=0.09888, over 5716485.14 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3415, pruned_loss=0.091, over 5672216.74 frames. ], batch size: 88, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:44:28,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0565, 1.2376, 3.5708, 2.9112], device='cuda:0'), covar=tensor([0.1516, 0.2336, 0.0388, 0.1071], device='cuda:0'), in_proj_covar=tensor([0.0598, 0.0556, 0.0790, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:44:56,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=335552.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:45:34,875 INFO [train.py:968] (0/2) Epoch 8, batch 16800, giga_loss[loss=0.2239, simple_loss=0.3225, pruned_loss=0.06259, over 28771.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3429, pruned_loss=0.09228, over 5665810.52 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3439, pruned_loss=0.09876, over 5719508.09 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3423, pruned_loss=0.09147, over 5667287.67 frames. ], batch size: 174, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:45:39,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=335582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:46:14,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7053, 4.5559, 4.2892, 1.8440], device='cuda:0'), covar=tensor([0.0422, 0.0516, 0.0646, 0.2044], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0871, 0.0768, 0.0609], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 05:46:23,887 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 05:46:41,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.473e+02 1.409e+03 1.823e+03 2.346e+03 9.394e+03, threshold=3.646e+03, percent-clipped=10.0
+2023-03-04 05:46:47,089 INFO [train.py:968] (0/2) Epoch 8, batch 16850, giga_loss[loss=0.3042, simple_loss=0.3772, pruned_loss=0.1156, over 28896.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3436, pruned_loss=0.09229, over 5675185.30 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.344, pruned_loss=0.09877, over 5722280.17 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3431, pruned_loss=0.0915, over 5672879.77 frames. ], batch size: 227, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:47:49,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4009, 1.7581, 1.4932, 1.6382], device='cuda:0'), covar=tensor([0.0721, 0.0263, 0.0308, 0.0781], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0121, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0050, 0.0047, 0.0078], device='cuda:0')
+2023-03-04 05:47:51,341 INFO [train.py:968] (0/2) Epoch 8, batch 16900, giga_loss[loss=0.2957, simple_loss=0.3778, pruned_loss=0.1068, over 28744.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3475, pruned_loss=0.09427, over 5682982.01 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3437, pruned_loss=0.09847, over 5728296.27 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3473, pruned_loss=0.09369, over 5674145.06 frames. ], batch size: 243, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:48:37,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=335711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:48:46,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=335719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:48:53,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.837e+02 1.243e+03 1.685e+03 2.163e+03 8.320e+03, threshold=3.369e+03, percent-clipped=4.0
+2023-03-04 05:49:01,066 INFO [train.py:968] (0/2) Epoch 8, batch 16950, giga_loss[loss=0.2762, simple_loss=0.3557, pruned_loss=0.09838, over 28766.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3463, pruned_loss=0.09308, over 5685091.33 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3434, pruned_loss=0.09831, over 5730773.33 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3465, pruned_loss=0.0927, over 5675496.08 frames. ], batch size: 262, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:50:08,494 INFO [train.py:968] (0/2) Epoch 8, batch 17000, giga_loss[loss=0.2856, simple_loss=0.3542, pruned_loss=0.1085, over 27769.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3443, pruned_loss=0.09277, over 5691136.21 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3437, pruned_loss=0.09833, over 5732667.02 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3442, pruned_loss=0.09228, over 5680787.95 frames. ], batch size: 474, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:50:14,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=335782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:51:12,817 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.056e+02 1.308e+03 1.739e+03 2.554e+03 9.624e+03, threshold=3.478e+03, percent-clipped=7.0
+2023-03-04 05:51:18,590 INFO [train.py:968] (0/2) Epoch 8, batch 17050, giga_loss[loss=0.2646, simple_loss=0.3433, pruned_loss=0.09299, over 28125.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3414, pruned_loss=0.09111, over 5696873.19 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3437, pruned_loss=0.09823, over 5736566.91 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3414, pruned_loss=0.09067, over 5684424.08 frames. ], batch size: 412, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:51:51,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8889, 3.7204, 3.4700, 1.6478], device='cuda:0'), covar=tensor([0.0552, 0.0621, 0.0674, 0.2270], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0862, 0.0771, 0.0612], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 05:51:58,563 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=335854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:52:01,463 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=335857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:52:10,124 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=335862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:52:14,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=335865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:52:29,851 INFO [train.py:968] (0/2) Epoch 8, batch 17100, giga_loss[loss=0.2353, simple_loss=0.3222, pruned_loss=0.07417, over 28906.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3389, pruned_loss=0.08857, over 5704659.96 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3432, pruned_loss=0.09789, over 5739195.20 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3393, pruned_loss=0.0884, over 5691999.09 frames. ], batch size: 174, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:52:40,544 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=335886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:52:50,255 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=335894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:53:26,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.830e+02 1.135e+03 1.450e+03 2.084e+03 4.484e+03, threshold=2.900e+03, percent-clipped=2.0
+2023-03-04 05:53:27,050 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=335925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:53:32,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=335928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:53:32,653 INFO [train.py:968] (0/2) Epoch 8, batch 17150, giga_loss[loss=0.2371, simple_loss=0.3153, pruned_loss=0.07945, over 28781.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3391, pruned_loss=0.08926, over 5696041.77 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3433, pruned_loss=0.098, over 5742957.91 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3392, pruned_loss=0.08881, over 5681488.16 frames. ], batch size: 99, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:54:01,374 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-04 05:54:06,369 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=335957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:54:06,409 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=335957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:54:10,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3163, 1.5706, 1.4474, 1.4379], device='cuda:0'), covar=tensor([0.1062, 0.1306, 0.1501, 0.1234], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0700, 0.0630, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 05:54:27,284 INFO [train.py:968] (0/2) Epoch 8, batch 17200, giga_loss[loss=0.2728, simple_loss=0.3521, pruned_loss=0.09671, over 28845.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3413, pruned_loss=0.09052, over 5701922.06 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3432, pruned_loss=0.09791, over 5748196.89 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3413, pruned_loss=0.08997, over 5684035.31 frames. ], batch size: 164, lr: 4.08e-03, grad_scale: 8.0
+2023-03-04 05:54:53,535 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-336000.pt
+2023-03-04 05:55:22,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.171e+02 1.424e+03 1.899e+03 2.642e+03 7.028e+03, threshold=3.797e+03, percent-clipped=19.0
+2023-03-04 05:55:24,573 INFO [train.py:968] (0/2) Epoch 8, batch 17250, libri_loss[loss=0.2656, simple_loss=0.3433, pruned_loss=0.09396, over 28603.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3432, pruned_loss=0.09227, over 5696323.18 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3432, pruned_loss=0.09791, over 5749994.82 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3432, pruned_loss=0.09173, over 5679487.55 frames. ], batch size: 106, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:56:24,781 INFO [train.py:968] (0/2) Epoch 8, batch 17300, giga_loss[loss=0.261, simple_loss=0.3407, pruned_loss=0.09062, over 28833.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3402, pruned_loss=0.09162, over 5693269.77 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3428, pruned_loss=0.0977, over 5752213.54 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3405, pruned_loss=0.09128, over 5677342.51 frames. ], batch size: 284, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:56:49,578 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=336100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:56:52,962 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=336103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:57:19,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.130e+02 1.500e+03 1.985e+03 2.671e+03 9.950e+03, threshold=3.969e+03, percent-clipped=11.0
+2023-03-04 05:57:21,327 INFO [train.py:968] (0/2) Epoch 8, batch 17350, giga_loss[loss=0.2472, simple_loss=0.3252, pruned_loss=0.08464, over 28922.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3405, pruned_loss=0.09303, over 5683130.06 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3421, pruned_loss=0.09732, over 5744554.30 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3414, pruned_loss=0.093, over 5675989.20 frames. ], batch size: 213, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:57:24,555 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=336132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:58:15,140 INFO [train.py:968] (0/2) Epoch 8, batch 17400, giga_loss[loss=0.2777, simple_loss=0.3595, pruned_loss=0.09792, over 28683.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3433, pruned_loss=0.09497, over 5692216.04 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.342, pruned_loss=0.09736, over 5748142.88 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3441, pruned_loss=0.0948, over 5681500.02 frames. ], batch size: 262, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:58:56,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9808, 2.4206, 2.3364, 2.3171], device='cuda:0'), covar=tensor([0.0868, 0.1903, 0.1527, 0.1468], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0702, 0.0629, 0.0629], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 05:59:02,979 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3815, 1.5279, 1.2868, 1.2548], device='cuda:0'), covar=tensor([0.2190, 0.2156, 0.2345, 0.1951], device='cuda:0'), in_proj_covar=tensor([0.1197, 0.0899, 0.1063, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 05:59:09,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.907e+02 1.291e+03 1.711e+03 2.296e+03 4.428e+03, threshold=3.423e+03, percent-clipped=3.0
+2023-03-04 05:59:10,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=336228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:59:10,929 INFO [train.py:968] (0/2) Epoch 8, batch 17450, giga_loss[loss=0.3896, simple_loss=0.4455, pruned_loss=0.1668, over 28942.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3543, pruned_loss=0.1021, over 5690249.73 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3419, pruned_loss=0.09734, over 5749030.16 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.355, pruned_loss=0.102, over 5680815.54 frames. ], batch size: 112, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:59:55,008 INFO [train.py:968] (0/2) Epoch 8, batch 17500, giga_loss[loss=0.2858, simple_loss=0.3584, pruned_loss=0.1067, over 29063.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3601, pruned_loss=0.1056, over 5696962.61 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.342, pruned_loss=0.09729, over 5750201.78 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.361, pruned_loss=0.1058, over 5686973.66 frames. ], batch size: 128, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 06:00:34,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.021e+02 1.188e+03 1.490e+03 2.067e+03 4.743e+03, threshold=2.981e+03, percent-clipped=3.0
+2023-03-04 06:00:36,644 INFO [train.py:968] (0/2) Epoch 8, batch 17550, giga_loss[loss=0.2445, simple_loss=0.3086, pruned_loss=0.09022, over 28990.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3569, pruned_loss=0.1048, over 5688933.36 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3419, pruned_loss=0.09721, over 5744861.10 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3584, pruned_loss=0.1053, over 5683986.25 frames. ], batch size: 106, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 06:01:22,896 INFO [train.py:968] (0/2) Epoch 8, batch 17600, giga_loss[loss=0.2482, simple_loss=0.3231, pruned_loss=0.08663, over 28732.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3499, pruned_loss=0.1017, over 5689556.30 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3421, pruned_loss=0.09719, over 5748337.16 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3512, pruned_loss=0.1023, over 5681289.19 frames. ], batch size: 242, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 06:02:07,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.971e+02 9.844e+02 1.210e+03 1.597e+03 3.178e+03, threshold=2.420e+03, percent-clipped=3.0
+2023-03-04 06:02:09,959 INFO [train.py:968] (0/2) Epoch 8, batch 17650, libri_loss[loss=0.2757, simple_loss=0.3541, pruned_loss=0.09863, over 29542.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.342, pruned_loss=0.09841, over 5685368.98 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3419, pruned_loss=0.09703, over 5750552.92 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3433, pruned_loss=0.09903, over 5675934.29 frames. ], batch size: 80, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 06:02:16,702 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5359, 3.1871, 1.9928, 1.7871], device='cuda:0'), covar=tensor([0.1668, 0.0975, 0.1250, 0.1620], device='cuda:0'), in_proj_covar=tensor([0.1597, 0.1406, 0.1367, 0.1512], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 06:02:48,399 INFO [train.py:968] (0/2) Epoch 8, batch 17700, giga_loss[loss=0.2149, simple_loss=0.2912, pruned_loss=0.06928, over 28852.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.335, pruned_loss=0.09496, over 5695050.33 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3418, pruned_loss=0.09681, over 5757574.04 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.336, pruned_loss=0.09564, over 5678345.01 frames. ], batch size: 174, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 06:03:27,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.072e+02 1.017e+03 1.334e+03 1.866e+03 3.486e+03, threshold=2.667e+03, percent-clipped=7.0
+2023-03-04 06:03:28,831 INFO [train.py:968] (0/2) Epoch 8, batch 17750, giga_loss[loss=0.2091, simple_loss=0.28, pruned_loss=0.06907, over 28703.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.329, pruned_loss=0.0922, over 5695655.27 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.342, pruned_loss=0.09678, over 5753337.19 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3292, pruned_loss=0.09266, over 5683583.93 frames. ], batch size: 92, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 06:04:08,245 INFO [train.py:968] (0/2) Epoch 8, batch 17800, giga_loss[loss=0.2621, simple_loss=0.3319, pruned_loss=0.0962, over 28602.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3236, pruned_loss=0.08955, over 5703659.77 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3417, pruned_loss=0.09646, over 5758697.30 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3236, pruned_loss=0.09005, over 5687344.08 frames. ], batch size: 307, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 06:04:29,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=336603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:04:48,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.642e+02 1.026e+03 1.394e+03 1.859e+03 4.944e+03, threshold=2.789e+03, percent-clipped=8.0
+2023-03-04 06:04:49,930 INFO [train.py:968] (0/2) Epoch 8, batch 17850, giga_loss[loss=0.2539, simple_loss=0.3218, pruned_loss=0.09303, over 28932.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3221, pruned_loss=0.08913, over 5701774.21 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3421, pruned_loss=0.0965, over 5757473.92 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3214, pruned_loss=0.08936, over 5688673.07 frames. ], batch size: 145, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:05:06,003 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 06:05:33,108 INFO [train.py:968] (0/2) Epoch 8, batch 17900, giga_loss[loss=0.2412, simple_loss=0.2902, pruned_loss=0.09612, over 23880.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3196, pruned_loss=0.08817, over 5703007.71 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3425, pruned_loss=0.0966, over 5762153.02 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3179, pruned_loss=0.08799, over 5686656.67 frames. ], batch size: 705, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:06:05,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2448, 2.8218, 1.3909, 1.3467], device='cuda:0'), covar=tensor([0.0926, 0.0392, 0.0877, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0484, 0.0322, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0027, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 06:06:12,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.151e+02 9.909e+02 1.372e+03 1.796e+03 6.003e+03, threshold=2.744e+03, percent-clipped=9.0
+2023-03-04 06:06:14,725 INFO [train.py:968] (0/2) Epoch 8, batch 17950, libri_loss[loss=0.2585, simple_loss=0.3447, pruned_loss=0.08613, over 29241.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3181, pruned_loss=0.08774, over 5710511.52 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3431, pruned_loss=0.09691, over 5765210.30 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3153, pruned_loss=0.08694, over 5692110.25 frames. ], batch size: 97, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:06:28,144 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=336746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:06:29,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=336749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:06:54,337 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=336778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:06:54,726 INFO [train.py:968] (0/2) Epoch 8, batch 18000, giga_loss[loss=0.2512, simple_loss=0.3175, pruned_loss=0.09242, over 27885.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3149, pruned_loss=0.0863, over 5706743.11 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3431, pruned_loss=0.09694, over 5768058.39 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3122, pruned_loss=0.08548, over 5688747.93 frames. ], batch size: 412, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:06:54,730 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 06:07:04,418 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2217, simple_loss=0.3257, pruned_loss=0.0588, over 944034.00 frames. 
+2023-03-04 06:07:04,419 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 06:07:21,816 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=336797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:07:47,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.500e+02 9.468e+02 1.236e+03 1.859e+03 1.174e+04, threshold=2.471e+03, percent-clipped=8.0
+2023-03-04 06:07:48,428 INFO [train.py:968] (0/2) Epoch 8, batch 18050, giga_loss[loss=0.211, simple_loss=0.2873, pruned_loss=0.06735, over 29028.00 frames. ], tot_loss[loss=0.242, simple_loss=0.313, pruned_loss=0.08551, over 5706881.58 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3439, pruned_loss=0.09743, over 5771054.39 frames. ], giga_tot_loss[loss=0.2389, simple_loss=0.3095, pruned_loss=0.08415, over 5688568.66 frames. ], batch size: 128, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:08:34,231 INFO [train.py:968] (0/2) Epoch 8, batch 18100, giga_loss[loss=0.2055, simple_loss=0.2716, pruned_loss=0.06971, over 29002.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3092, pruned_loss=0.08408, over 5701276.01 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3442, pruned_loss=0.09754, over 5772575.64 frames. ], giga_tot_loss[loss=0.2357, simple_loss=0.3059, pruned_loss=0.08277, over 5684774.06 frames. ], batch size: 136, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:08:43,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5959, 1.6999, 1.2814, 1.3141], device='cuda:0'), covar=tensor([0.1569, 0.1360, 0.1063, 0.1184], device='cuda:0'), in_proj_covar=tensor([0.1601, 0.1409, 0.1376, 0.1502], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 06:09:17,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.554e+02 9.919e+02 1.226e+03 1.751e+03 7.775e+03, threshold=2.453e+03, percent-clipped=10.0
+2023-03-04 06:09:17,851 INFO [train.py:968] (0/2) Epoch 8, batch 18150, giga_loss[loss=0.2272, simple_loss=0.3, pruned_loss=0.07727, over 28802.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3064, pruned_loss=0.08231, over 5699712.32 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3445, pruned_loss=0.09752, over 5773566.72 frames. ], giga_tot_loss[loss=0.2323, simple_loss=0.3026, pruned_loss=0.08097, over 5684016.96 frames. ], batch size: 262, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:10:03,565 INFO [train.py:968] (0/2) Epoch 8, batch 18200, giga_loss[loss=0.2481, simple_loss=0.3064, pruned_loss=0.09487, over 26671.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3027, pruned_loss=0.08077, over 5694351.37 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3443, pruned_loss=0.09737, over 5776331.25 frames. ], giga_tot_loss[loss=0.2291, simple_loss=0.2991, pruned_loss=0.07952, over 5677854.46 frames. ], batch size: 555, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:10:23,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2446, 1.2364, 1.1994, 1.4467], device='cuda:0'), covar=tensor([0.0701, 0.0456, 0.0337, 0.0725], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0121, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0051, 0.0047, 0.0078], device='cuda:0')
+2023-03-04 06:10:45,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.187e+02 1.001e+03 1.241e+03 1.682e+03 6.594e+03, threshold=2.483e+03, percent-clipped=13.0
+2023-03-04 06:10:48,859 INFO [train.py:968] (0/2) Epoch 8, batch 18250, giga_loss[loss=0.2892, simple_loss=0.3393, pruned_loss=0.1196, over 23664.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3084, pruned_loss=0.08445, over 5679958.79 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3444, pruned_loss=0.09743, over 5775829.20 frames. ], giga_tot_loss[loss=0.2349, simple_loss=0.3041, pruned_loss=0.08289, over 5664180.27 frames. ], batch size: 705, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:11:22,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7730, 5.5196, 5.2166, 2.4588], device='cuda:0'), covar=tensor([0.0321, 0.0532, 0.0649, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0878, 0.0774, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 06:11:35,282 INFO [train.py:968] (0/2) Epoch 8, batch 18300, giga_loss[loss=0.3061, simple_loss=0.3763, pruned_loss=0.118, over 28678.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3215, pruned_loss=0.09135, over 5679355.48 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3446, pruned_loss=0.09756, over 5768177.53 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3172, pruned_loss=0.08977, over 5671264.26 frames. ], batch size: 99, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:11:57,307 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:12:16,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.936e+02 1.301e+03 1.696e+03 2.201e+03 7.085e+03, threshold=3.393e+03, percent-clipped=20.0
+2023-03-04 06:12:17,327 INFO [train.py:968] (0/2) Epoch 8, batch 18350, giga_loss[loss=0.4074, simple_loss=0.4479, pruned_loss=0.1834, over 28379.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3354, pruned_loss=0.09894, over 5692400.55 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3448, pruned_loss=0.09747, over 5771238.38 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3315, pruned_loss=0.09773, over 5681271.68 frames. ], batch size: 71, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:12:53,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=337172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:12:57,990 INFO [train.py:968] (0/2) Epoch 8, batch 18400, giga_loss[loss=0.298, simple_loss=0.3698, pruned_loss=0.1131, over 28930.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3448, pruned_loss=0.1033, over 5689896.31 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3452, pruned_loss=0.09753, over 5765205.85 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3413, pruned_loss=0.1025, over 5683622.43 frames. ], batch size: 199, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:13:40,958 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.945e+02 1.320e+03 1.782e+03 2.475e+03 4.453e+03, threshold=3.563e+03, percent-clipped=8.0
+2023-03-04 06:13:40,971 INFO [train.py:968] (0/2) Epoch 8, batch 18450, giga_loss[loss=0.3009, simple_loss=0.3683, pruned_loss=0.1167, over 28420.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.351, pruned_loss=0.1053, over 5686188.75 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3456, pruned_loss=0.09768, over 5767768.98 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3478, pruned_loss=0.1046, over 5677392.90 frames. ], batch size: 71, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:14:04,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0952, 2.5023, 1.7064, 1.6581], device='cuda:0'), covar=tensor([0.1368, 0.1031, 0.1204, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.1605, 0.1427, 0.1392, 0.1521], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 06:14:20,349 INFO [train.py:968] (0/2) Epoch 8, batch 18500, giga_loss[loss=0.268, simple_loss=0.3491, pruned_loss=0.0935, over 28910.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3529, pruned_loss=0.1051, over 5686322.79 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3463, pruned_loss=0.09793, over 5770089.16 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3499, pruned_loss=0.1045, over 5674269.82 frames. ], batch size: 174, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:14:32,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:14:56,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=337315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:14:58,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=337318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:15:06,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.375e+02 1.133e+03 1.469e+03 2.074e+03 8.291e+03, threshold=2.938e+03, percent-clipped=4.0
+2023-03-04 06:15:06,872 INFO [train.py:968] (0/2) Epoch 8, batch 18550, libri_loss[loss=0.2546, simple_loss=0.3283, pruned_loss=0.09042, over 29540.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3553, pruned_loss=0.1065, over 5683293.14 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.347, pruned_loss=0.09839, over 5773467.78 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3525, pruned_loss=0.1059, over 5667440.64 frames. ], batch size: 77, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:15:22,347 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=337347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:15:26,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3169, 4.1140, 3.8872, 2.0638], device='cuda:0'), covar=tensor([0.0468, 0.0622, 0.0645, 0.1927], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0879, 0.0780, 0.0621], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 06:15:50,850 INFO [train.py:968] (0/2) Epoch 8, batch 18600, giga_loss[loss=0.2989, simple_loss=0.3681, pruned_loss=0.1148, over 27838.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3575, pruned_loss=0.1085, over 5681807.66 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3472, pruned_loss=0.09847, over 5772559.45 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3552, pruned_loss=0.108, over 5669785.10 frames. ], batch size: 412, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:16:33,352 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.906e+02 1.160e+03 1.633e+03 2.119e+03 4.932e+03, threshold=3.266e+03, percent-clipped=8.0
+2023-03-04 06:16:33,365 INFO [train.py:968] (0/2) Epoch 8, batch 18650, giga_loss[loss=0.2841, simple_loss=0.3613, pruned_loss=0.1035, over 29059.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3606, pruned_loss=0.1107, over 5686709.03 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3481, pruned_loss=0.09897, over 5774716.95 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3583, pruned_loss=0.1102, over 5672540.93 frames. ], batch size: 155, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:16:48,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6735, 1.5270, 1.2291, 1.3089], device='cuda:0'), covar=tensor([0.0501, 0.0383, 0.0728, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0438, 0.0497, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 06:17:03,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0221, 1.2406, 1.2301, 1.1373], device='cuda:0'), covar=tensor([0.1284, 0.1176, 0.1905, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0726, 0.0649, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 06:17:11,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=337476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:17:14,501 INFO [train.py:968] (0/2) Epoch 8, batch 18700, giga_loss[loss=0.2827, simple_loss=0.3593, pruned_loss=0.103, over 28372.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3632, pruned_loss=0.1118, over 5690409.18 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.348, pruned_loss=0.09888, over 5775862.98 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3618, pruned_loss=0.1118, over 5675538.87 frames. ], batch size: 77, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:17:41,280 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6773, 1.9328, 1.9800, 1.5543], device='cuda:0'), covar=tensor([0.1683, 0.1968, 0.1261, 0.1457], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0696, 0.0823, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 06:17:56,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.356e+02 1.098e+03 1.286e+03 1.804e+03 3.700e+03, threshold=2.571e+03, percent-clipped=2.0
+2023-03-04 06:17:56,755 INFO [train.py:968] (0/2) Epoch 8, batch 18750, giga_loss[loss=0.3307, simple_loss=0.3921, pruned_loss=0.1347, over 28889.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3648, pruned_loss=0.1115, over 5689947.76 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3483, pruned_loss=0.099, over 5778135.81 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3637, pruned_loss=0.1117, over 5674483.31 frames. ], batch size: 145, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:18:20,167 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9400, 1.2047, 1.0036, 0.1729], device='cuda:0'), covar=tensor([0.1991, 0.1616, 0.2217, 0.3281], device='cuda:0'), in_proj_covar=tensor([0.1458, 0.1405, 0.1446, 0.1199], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 06:18:37,156 INFO [train.py:968] (0/2) Epoch 8, batch 18800, giga_loss[loss=0.2926, simple_loss=0.3701, pruned_loss=0.1076, over 28254.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.367, pruned_loss=0.1122, over 5695802.95 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3493, pruned_loss=0.09963, over 5780143.57 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3658, pruned_loss=0.1121, over 5679289.74 frames. ], batch size: 77, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:19:07,031 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2667, 1.7909, 1.4892, 1.4997], device='cuda:0'), covar=tensor([0.0798, 0.0276, 0.0308, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0119, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0051, 0.0046, 0.0078], device='cuda:0')
+2023-03-04 06:19:10,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=337619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:19:12,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=337622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:19:16,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.158e+02 1.097e+03 1.411e+03 1.809e+03 4.459e+03, threshold=2.821e+03, percent-clipped=6.0
+2023-03-04 06:19:16,214 INFO [train.py:968] (0/2) Epoch 8, batch 18850, giga_loss[loss=0.2914, simple_loss=0.3721, pruned_loss=0.1053, over 29034.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3673, pruned_loss=0.1114, over 5699964.60 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3496, pruned_loss=0.09984, over 5782292.06 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3664, pruned_loss=0.1114, over 5682632.00 frames. ], batch size: 128, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:19:34,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=337651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:19:39,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8938, 1.8078, 1.1984, 1.4823], device='cuda:0'), covar=tensor([0.0676, 0.0619, 0.1016, 0.0972], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0439, 0.0496, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 06:19:46,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=337666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:19:59,522 INFO [train.py:968] (0/2) Epoch 8, batch 18900, giga_loss[loss=0.2962, simple_loss=0.3821, pruned_loss=0.1052, over 28514.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3662, pruned_loss=0.1089, over 5711420.33 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3499, pruned_loss=0.1, over 5782515.60 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3654, pruned_loss=0.1088, over 5697338.30 frames. ], batch size: 60, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:20:09,755 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:20:40,882 INFO [train.py:968] (0/2) Epoch 8, batch 18950, giga_loss[loss=0.2893, simple_loss=0.3585, pruned_loss=0.11, over 27956.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3646, pruned_loss=0.1076, over 5704671.72 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3504, pruned_loss=0.1003, over 5775441.16 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3637, pruned_loss=0.1074, over 5697975.24 frames. ], batch size: 412, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:20:41,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.444e+02 9.733e+02 1.208e+03 1.638e+03 3.804e+03, threshold=2.415e+03, percent-clipped=4.0
+2023-03-04 06:21:13,485 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9996, 1.1581, 1.2876, 1.0443], device='cuda:0'), covar=tensor([0.1225, 0.1133, 0.1741, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0729, 0.0650, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 06:21:19,154 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 06:21:21,412 INFO [train.py:968] (0/2) Epoch 8, batch 19000, giga_loss[loss=0.3709, simple_loss=0.4093, pruned_loss=0.1663, over 28859.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3657, pruned_loss=0.109, over 5711936.47 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3504, pruned_loss=0.1, over 5778293.64 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3653, pruned_loss=0.1093, over 5702589.71 frames. ], batch size: 199, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:21:49,727 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=337809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:21:52,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=337812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:21:57,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:22:06,949 INFO [train.py:968] (0/2) Epoch 8, batch 19050, libri_loss[loss=0.2837, simple_loss=0.3613, pruned_loss=0.1031, over 29541.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3684, pruned_loss=0.1136, over 5718566.99 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3506, pruned_loss=0.09999, over 5781940.15 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3686, pruned_loss=0.1142, over 5705644.89 frames. ], batch size: 89, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:22:08,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.134e+02 1.323e+03 1.758e+03 2.513e+03 1.188e+04, threshold=3.515e+03, percent-clipped=29.0
+2023-03-04 06:22:19,254 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=337841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:22:51,916 INFO [train.py:968] (0/2) Epoch 8, batch 19100, libri_loss[loss=0.2662, simple_loss=0.3416, pruned_loss=0.09537, over 29574.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.369, pruned_loss=0.1162, over 5714995.73 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3502, pruned_loss=0.09977, over 5784091.13 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3698, pruned_loss=0.1173, over 5701210.22 frames. ], batch size: 75, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:23:01,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8773, 1.0867, 3.8167, 3.0051], device='cuda:0'), covar=tensor([0.1771, 0.2545, 0.0394, 0.0796], device='cuda:0'), in_proj_covar=tensor([0.0595, 0.0553, 0.0793, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 06:23:18,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 06:23:32,197 INFO [train.py:968] (0/2) Epoch 8, batch 19150, giga_loss[loss=0.28, simple_loss=0.3465, pruned_loss=0.1068, over 28977.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.367, pruned_loss=0.1159, over 5711473.53 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3503, pruned_loss=0.09981, over 5786212.47 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3679, pruned_loss=0.117, over 5697715.29 frames. ], batch size: 136, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:23:32,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.149e+02 1.148e+03 1.390e+03 1.695e+03 4.320e+03, threshold=2.779e+03, percent-clipped=2.0
+2023-03-04 06:24:12,908 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-04 06:24:14,442 INFO [train.py:968] (0/2) Epoch 8, batch 19200, giga_loss[loss=0.2845, simple_loss=0.3585, pruned_loss=0.1053, over 28948.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3645, pruned_loss=0.1149, over 5708726.98 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.0996, over 5789083.85 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3657, pruned_loss=0.1162, over 5693553.52 frames. ], batch size: 174, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:24:16,273 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:24:24,940 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:24:26,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5450, 1.8237, 1.8586, 1.4484], device='cuda:0'), covar=tensor([0.1524, 0.1887, 0.1170, 0.1334], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0699, 0.0823, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 06:24:34,120 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-338000.pt
+2023-03-04 06:25:02,754 INFO [train.py:968] (0/2) Epoch 8, batch 19250, giga_loss[loss=0.291, simple_loss=0.3668, pruned_loss=0.1075, over 28311.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3631, pruned_loss=0.1134, over 5715716.88 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.0996, over 5789083.85 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3641, pruned_loss=0.1145, over 5703907.19 frames. ], batch size: 368, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:25:03,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.234e+02 1.210e+03 1.533e+03 1.972e+03 8.372e+03, threshold=3.065e+03, percent-clipped=10.0
+2023-03-04 06:25:17,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-04 06:25:32,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:25:43,842 INFO [train.py:968] (0/2) Epoch 8, batch 19300, giga_loss[loss=0.395, simple_loss=0.4263, pruned_loss=0.1819, over 23374.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3616, pruned_loss=0.1119, over 5712812.48 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3502, pruned_loss=0.09968, over 5790273.01 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3624, pruned_loss=0.1128, over 5701948.33 frames. ], batch size: 705, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:26:25,333 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-04 06:26:31,153 INFO [train.py:968] (0/2) Epoch 8, batch 19350, giga_loss[loss=0.243, simple_loss=0.3166, pruned_loss=0.08473, over 28584.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3574, pruned_loss=0.109, over 5700397.38 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3504, pruned_loss=0.09975, over 5790923.73 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.358, pruned_loss=0.1099, over 5689450.73 frames. ], batch size: 336, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:26:33,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.051e+02 1.058e+03 1.401e+03 1.933e+03 1.075e+04, threshold=2.802e+03, percent-clipped=6.0
+2023-03-04 06:27:19,575 INFO [train.py:968] (0/2) Epoch 8, batch 19400, giga_loss[loss=0.2449, simple_loss=0.3203, pruned_loss=0.08478, over 28554.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3509, pruned_loss=0.1054, over 5682657.72 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3505, pruned_loss=0.09976, over 5786332.88 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3513, pruned_loss=0.1062, over 5676498.27 frames. ], batch size: 336, lr: 4.07e-03, grad_scale: 2.0
+2023-03-04 06:27:29,839 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:27:43,649 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:27:46,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:28:02,410 INFO [train.py:968] (0/2) Epoch 8, batch 19450, giga_loss[loss=0.2366, simple_loss=0.306, pruned_loss=0.08354, over 28835.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3459, pruned_loss=0.1026, over 5694101.42 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3512, pruned_loss=0.1003, over 5789569.48 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3456, pruned_loss=0.1029, over 5683058.90 frames. ], batch size: 186, lr: 4.07e-03, grad_scale: 2.0
+2023-03-04 06:28:04,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.501e+02 8.724e+02 1.140e+03 1.485e+03 7.115e+03, threshold=2.279e+03, percent-clipped=6.0
+2023-03-04 06:28:15,957 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:28:18,989 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:28:50,448 INFO [train.py:968] (0/2) Epoch 8, batch 19500, giga_loss[loss=0.2547, simple_loss=0.3315, pruned_loss=0.089, over 29066.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.342, pruned_loss=0.1006, over 5656531.73 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3516, pruned_loss=0.1003, over 5781232.60 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3412, pruned_loss=0.1008, over 5652795.69 frames. ], batch size: 136, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:29:02,487 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-04 06:29:07,166 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:29:35,369 INFO [train.py:968] (0/2) Epoch 8, batch 19550, giga_loss[loss=0.2753, simple_loss=0.3522, pruned_loss=0.09918, over 28691.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3426, pruned_loss=0.1004, over 5657677.32 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3518, pruned_loss=0.1002, over 5779487.20 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3415, pruned_loss=0.1007, over 5653419.17 frames. ], batch size: 284, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:29:37,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.287e+02 9.694e+02 1.188e+03 1.573e+03 5.649e+03, threshold=2.375e+03, percent-clipped=9.0
+2023-03-04 06:29:43,077 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:29:44,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:29:50,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4628, 1.4464, 4.7653, 3.5110], device='cuda:0'), covar=tensor([0.1567, 0.2393, 0.0314, 0.0642], device='cuda:0'), in_proj_covar=tensor([0.0594, 0.0554, 0.0788, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 06:30:01,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:30:11,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:30:12,419 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:30:20,094 INFO [train.py:968] (0/2) Epoch 8, batch 19600, giga_loss[loss=0.2678, simple_loss=0.3396, pruned_loss=0.09802, over 29126.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3428, pruned_loss=0.1002, over 5669475.01 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3526, pruned_loss=0.1006, over 5782364.29 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3408, pruned_loss=0.1, over 5660553.81 frames. ], batch size: 128, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:30:37,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1777, 1.4506, 1.2091, 0.9689], device='cuda:0'), covar=tensor([0.2174, 0.2141, 0.2345, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.1211, 0.0908, 0.1067, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 06:30:47,231 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-04 06:31:01,791 INFO [train.py:968] (0/2) Epoch 8, batch 19650, giga_loss[loss=0.2482, simple_loss=0.3242, pruned_loss=0.08614, over 28755.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.342, pruned_loss=0.1003, over 5677634.80 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3533, pruned_loss=0.1009, over 5783859.76 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3396, pruned_loss=0.09984, over 5666854.79 frames. ], batch size: 262, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:31:04,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.834e+02 9.710e+02 1.271e+03 1.692e+03 1.010e+04, threshold=2.542e+03, percent-clipped=16.0
+2023-03-04 06:31:09,562 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:31:36,327 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:31:41,256 INFO [train.py:968] (0/2) Epoch 8, batch 19700, giga_loss[loss=0.2616, simple_loss=0.3367, pruned_loss=0.09322, over 28775.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3395, pruned_loss=0.09922, over 5687491.89 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3534, pruned_loss=0.1009, over 5785962.24 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3374, pruned_loss=0.09885, over 5675312.52 frames. ], batch size: 284, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:31:57,688 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:31:59,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:32:00,647 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-04 06:32:02,554 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 06:32:05,269 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:32:08,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:32:21,160 INFO [train.py:968] (0/2) Epoch 8, batch 19750, libri_loss[loss=0.3152, simple_loss=0.402, pruned_loss=0.1142, over 26323.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3371, pruned_loss=0.09764, over 5688067.55 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3538, pruned_loss=0.101, over 5776155.24 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3346, pruned_loss=0.0972, over 5684699.59 frames. ], batch size: 136, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:32:22,068 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:32:24,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.163e+02 9.253e+02 1.136e+03 1.576e+03 9.342e+03, threshold=2.273e+03, percent-clipped=8.0
+2023-03-04 06:32:30,650 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:33:03,691 INFO [train.py:968] (0/2) Epoch 8, batch 19800, giga_loss[loss=0.2707, simple_loss=0.3423, pruned_loss=0.09951, over 28300.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3353, pruned_loss=0.09677, over 5693136.76 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3542, pruned_loss=0.1012, over 5777914.50 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3325, pruned_loss=0.09617, over 5687180.58 frames. ], batch size: 368, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:33:37,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:33:43,941 INFO [train.py:968] (0/2) Epoch 8, batch 19850, giga_loss[loss=0.2523, simple_loss=0.321, pruned_loss=0.09176, over 28454.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3324, pruned_loss=0.09542, over 5706763.30 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3543, pruned_loss=0.101, over 5781164.34 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3296, pruned_loss=0.09498, over 5697345.64 frames. ], batch size: 71, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:33:46,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.845e+02 9.822e+02 1.243e+03 1.854e+03 4.079e+03, threshold=2.486e+03, percent-clipped=19.0
+2023-03-04 06:34:16,332 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:34:20,336 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 06:34:23,240 INFO [train.py:968] (0/2) Epoch 8, batch 19900, libri_loss[loss=0.3055, simple_loss=0.3914, pruned_loss=0.1098, over 29522.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3308, pruned_loss=0.09431, over 5714732.23 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3548, pruned_loss=0.101, over 5781446.46 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3273, pruned_loss=0.09374, over 5704451.23 frames. ], batch size: 83, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:34:37,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8488, 1.7086, 1.3403, 1.4978], device='cuda:0'), covar=tensor([0.0656, 0.0545, 0.0920, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0447, 0.0501, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 06:35:04,001 INFO [train.py:968] (0/2) Epoch 8, batch 19950, giga_loss[loss=0.2236, simple_loss=0.2981, pruned_loss=0.07451, over 28480.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3285, pruned_loss=0.09298, over 5720648.60 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3552, pruned_loss=0.1011, over 5784313.22 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3249, pruned_loss=0.09233, over 5708859.32 frames. ], batch size: 60, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:35:06,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.741e+02 9.470e+02 1.210e+03 1.663e+03 8.682e+03, threshold=2.420e+03, percent-clipped=9.0
+2023-03-04 06:35:29,863 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:35:32,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:35:35,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9164, 1.1457, 3.7252, 3.0292], device='cuda:0'), covar=tensor([0.1754, 0.2467, 0.0389, 0.0736], device='cuda:0'), in_proj_covar=tensor([0.0598, 0.0554, 0.0788, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 06:35:42,635 INFO [train.py:968] (0/2) Epoch 8, batch 20000, giga_loss[loss=0.2004, simple_loss=0.2791, pruned_loss=0.06084, over 28397.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3284, pruned_loss=0.09284, over 5719554.78 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3564, pruned_loss=0.1015, over 5784230.64 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3237, pruned_loss=0.09172, over 5708351.72 frames. ], batch size: 65, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:35:56,247 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:35:56,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1587, 1.7411, 1.3923, 0.3494], device='cuda:0'), covar=tensor([0.2464, 0.1416, 0.2510, 0.3473], device='cuda:0'), in_proj_covar=tensor([0.1458, 0.1382, 0.1442, 0.1197], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 06:35:56,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6469, 1.7367, 1.5588, 1.3440], device='cuda:0'), covar=tensor([0.1488, 0.1286, 0.1060, 0.1415], device='cuda:0'), in_proj_covar=tensor([0.1594, 0.1413, 0.1402, 0.1531], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 06:36:01,695 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:01,792 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9555, 2.4479, 1.7121, 1.3907], device='cuda:0'), covar=tensor([0.1985, 0.1143, 0.1482, 0.1830], device='cuda:0'), in_proj_covar=tensor([0.1596, 0.1416, 0.1404, 0.1534], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 06:36:09,870 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:10,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:11,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:20,686 INFO [train.py:968] (0/2) Epoch 8, batch 20050, libri_loss[loss=0.2818, simple_loss=0.373, pruned_loss=0.09527, over 29562.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3279, pruned_loss=0.0924, over 5724437.80 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3572, pruned_loss=0.1019, over 5786085.82 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3228, pruned_loss=0.09097, over 5712562.56 frames. ], batch size: 77, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:36:23,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.819e+02 9.902e+02 1.354e+03 2.113e+03 7.675e+03, threshold=2.708e+03, percent-clipped=13.0
+2023-03-04 06:36:32,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:34,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:58,459 INFO [train.py:968] (0/2) Epoch 8, batch 20100, giga_loss[loss=0.3018, simple_loss=0.3586, pruned_loss=0.1225, over 28558.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3308, pruned_loss=0.09377, over 5726838.33 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.358, pruned_loss=0.102, over 5790463.96 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3249, pruned_loss=0.09214, over 5710990.79 frames. ], batch size: 85, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:37:33,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5318, 3.2381, 2.0956, 1.7167], device='cuda:0'), covar=tensor([0.1328, 0.0650, 0.1058, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1433, 0.1418, 0.1547], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 06:37:40,455 INFO [train.py:968] (0/2) Epoch 8, batch 20150, giga_loss[loss=0.2669, simple_loss=0.344, pruned_loss=0.09489, over 28835.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3373, pruned_loss=0.09789, over 5728988.94 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3583, pruned_loss=0.102, over 5794398.90 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3315, pruned_loss=0.09636, over 5710774.18 frames. ], batch size: 119, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:37:44,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.222e+02 1.091e+03 1.346e+03 1.760e+03 6.148e+03, threshold=2.691e+03, percent-clipped=10.0
+2023-03-04 06:38:09,881 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:12,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:31,048 INFO [train.py:968] (0/2) Epoch 8, batch 20200, libri_loss[loss=0.3102, simple_loss=0.3874, pruned_loss=0.1165, over 27562.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3441, pruned_loss=0.1028, over 5708104.34 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3582, pruned_loss=0.1019, over 5791918.45 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3392, pruned_loss=0.1017, over 5694473.90 frames. ], batch size: 115, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:38:35,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4643, 1.5035, 1.5206, 1.4116], device='cuda:0'), covar=tensor([0.1052, 0.1669, 0.1518, 0.1434], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0724, 0.0651, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 06:38:39,245 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:40,523 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:40,631 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:43,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:39:09,299 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:39:15,435 INFO [train.py:968] (0/2) Epoch 8, batch 20250, giga_loss[loss=0.3744, simple_loss=0.4205, pruned_loss=0.1642, over 27959.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.353, pruned_loss=0.1091, over 5707587.59 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3585, pruned_loss=0.102, over 5793687.95 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3483, pruned_loss=0.1082, over 5691071.52 frames. ], batch size: 412, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:39:19,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.737e+02 1.298e+03 1.621e+03 2.119e+03 5.437e+03, threshold=3.242e+03, percent-clipped=13.0
+2023-03-04 06:39:44,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:39:59,798 INFO [train.py:968] (0/2) Epoch 8, batch 20300, giga_loss[loss=0.3008, simple_loss=0.3782, pruned_loss=0.1117, over 29064.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3578, pruned_loss=0.1111, over 5701887.27 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3589, pruned_loss=0.1023, over 5796154.67 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3537, pruned_loss=0.1104, over 5684469.34 frames. ], batch size: 128, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:40:08,299 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:40:19,359 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:40:43,626 INFO [train.py:968] (0/2) Epoch 8, batch 20350, libri_loss[loss=0.3045, simple_loss=0.3824, pruned_loss=0.1133, over 26001.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3616, pruned_loss=0.1121, over 5702525.71 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3593, pruned_loss=0.1025, over 5795776.94 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.358, pruned_loss=0.1116, over 5686065.70 frames. ], batch size: 136, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:40:46,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.701e+02 1.065e+03 1.414e+03 1.861e+03 6.537e+03, threshold=2.827e+03, percent-clipped=4.0
+2023-03-04 06:41:22,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:41:32,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:41:33,530 INFO [train.py:968] (0/2) Epoch 8, batch 20400, libri_loss[loss=0.282, simple_loss=0.3673, pruned_loss=0.09832, over 27541.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3679, pruned_loss=0.1161, over 5698167.05 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3591, pruned_loss=0.1024, over 5794713.29 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3653, pruned_loss=0.116, over 5685373.39 frames. ], batch size: 115, lr: 4.06e-03, grad_scale: 8.0
+2023-03-04 06:42:08,943 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-04 06:42:18,633 INFO [train.py:968] (0/2) Epoch 8, batch 20450, libri_loss[loss=0.2422, simple_loss=0.3168, pruned_loss=0.08387, over 28611.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3663, pruned_loss=0.1151, over 5692274.70 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3588, pruned_loss=0.1022, over 5794967.06 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3647, pruned_loss=0.1153, over 5681200.06 frames. ], batch size: 63, lr: 4.06e-03, grad_scale: 8.0
+2023-03-04 06:42:22,003 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.579e+02 1.280e+03 1.579e+03 2.108e+03 3.911e+03, threshold=3.157e+03, percent-clipped=13.0
+2023-03-04 06:42:44,342 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:43:03,298 INFO [train.py:968] (0/2) Epoch 8, batch 20500, giga_loss[loss=0.3106, simple_loss=0.3766, pruned_loss=0.1223, over 28758.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3603, pruned_loss=0.1103, over 5694034.95 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3589, pruned_loss=0.1025, over 5787769.48 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3589, pruned_loss=0.1104, over 5690305.23 frames. ], batch size: 66, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:43:24,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-04 06:43:38,641 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:43:40,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:43:45,748 INFO [train.py:968] (0/2) Epoch 8, batch 20550, giga_loss[loss=0.2787, simple_loss=0.3462, pruned_loss=0.1055, over 28532.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3594, pruned_loss=0.1094, over 5689055.39 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.359, pruned_loss=0.1025, over 5780290.95 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3583, pruned_loss=0.1095, over 5691204.07 frames. ], batch size: 85, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:43:50,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.497e+02 1.123e+03 1.622e+03 2.686e+03 1.828e+04, threshold=3.244e+03, percent-clipped=16.0
+2023-03-04 06:44:04,236 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:44:15,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:44:18,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1608, 1.5834, 1.5019, 1.1324], device='cuda:0'), covar=tensor([0.1405, 0.1940, 0.1122, 0.1280], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0698, 0.0821, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 06:44:26,334 INFO [train.py:968] (0/2) Epoch 8, batch 20600, giga_loss[loss=0.2898, simple_loss=0.3647, pruned_loss=0.1074, over 28779.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3593, pruned_loss=0.1087, over 5696266.27 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3593, pruned_loss=0.103, over 5783639.53 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3581, pruned_loss=0.1085, over 5692260.43 frames. ], batch size: 284, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:45:08,762 INFO [train.py:968] (0/2) Epoch 8, batch 20650, giga_loss[loss=0.3588, simple_loss=0.4201, pruned_loss=0.1487, over 28550.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3619, pruned_loss=0.1104, over 5687852.26 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.359, pruned_loss=0.103, over 5778586.00 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3611, pruned_loss=0.1105, over 5687285.80 frames. ], batch size: 336, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:45:12,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.585e+02 1.426e+03 1.767e+03 2.299e+03 6.199e+03, threshold=3.535e+03, percent-clipped=10.0
+2023-03-04 06:45:14,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:45:24,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3965, 2.0194, 1.4811, 0.6521], device='cuda:0'), covar=tensor([0.3599, 0.1803, 0.2481, 0.3736], device='cuda:0'), in_proj_covar=tensor([0.1453, 0.1385, 0.1438, 0.1194], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 06:45:34,138 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:45:34,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:45:47,344 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:45:50,466 INFO [train.py:968] (0/2) Epoch 8, batch 20700, giga_loss[loss=0.2973, simple_loss=0.3627, pruned_loss=0.1159, over 28634.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3647, pruned_loss=0.1124, over 5686203.76 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3597, pruned_loss=0.1033, over 5777601.74 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3637, pruned_loss=0.1124, over 5683412.80 frames. ], batch size: 242, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:46:13,710 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:46:16,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:46:34,518 INFO [train.py:968] (0/2) Epoch 8, batch 20750, giga_loss[loss=0.3431, simple_loss=0.392, pruned_loss=0.1471, over 28772.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3656, pruned_loss=0.1132, over 5699384.02 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3596, pruned_loss=0.1032, over 5778220.74 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3648, pruned_loss=0.1134, over 5696306.41 frames. ], batch size: 119, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:46:40,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.631e+02 1.189e+03 1.495e+03 1.923e+03 3.837e+03, threshold=2.990e+03, percent-clipped=3.0
+2023-03-04 06:46:44,740 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:46:46,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:46:46,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0260, 1.1628, 3.8063, 3.0860], device='cuda:0'), covar=tensor([0.1712, 0.2494, 0.0414, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0604, 0.0555, 0.0796, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 06:47:16,224 INFO [train.py:968] (0/2) Epoch 8, batch 20800, libri_loss[loss=0.2983, simple_loss=0.3735, pruned_loss=0.1115, over 29560.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3674, pruned_loss=0.1147, over 5711378.85 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3606, pruned_loss=0.1039, over 5779811.79 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3662, pruned_loss=0.1146, over 5704580.27 frames. ], batch size: 76, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:47:17,273 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:19,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:37,290 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:39,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:42,738 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:46,447 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:49,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:55,934 INFO [train.py:968] (0/2) Epoch 8, batch 20850, libri_loss[loss=0.3045, simple_loss=0.3814, pruned_loss=0.1138, over 27558.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3681, pruned_loss=0.1152, over 5709858.56 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3614, pruned_loss=0.1044, over 5783028.17 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3666, pruned_loss=0.1152, over 5698448.97 frames. ], batch size: 115, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:47:58,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:48:00,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.942e+02 1.106e+03 1.533e+03 2.389e+03 9.376e+03, threshold=3.066e+03, percent-clipped=15.0
+2023-03-04 06:48:01,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:48:11,069 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:48:34,733 INFO [train.py:968] (0/2) Epoch 8, batch 20900, giga_loss[loss=0.2748, simple_loss=0.3548, pruned_loss=0.0974, over 28429.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3673, pruned_loss=0.1141, over 5713810.27 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3615, pruned_loss=0.1046, over 5784796.91 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3661, pruned_loss=0.1141, over 5701772.52 frames. ], batch size: 65, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:48:39,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:48:41,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:48:43,281 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:49:04,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:49:13,913 INFO [train.py:968] (0/2) Epoch 8, batch 20950, giga_loss[loss=0.2784, simple_loss=0.3624, pruned_loss=0.09719, over 28929.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3676, pruned_loss=0.1132, over 5715504.41 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3616, pruned_loss=0.1045, over 5786761.89 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3667, pruned_loss=0.1135, over 5702958.63 frames. ], batch size: 174, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:49:20,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.433e+02 1.143e+03 1.474e+03 2.160e+03 6.188e+03, threshold=2.948e+03, percent-clipped=6.0
+2023-03-04 06:49:53,520 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:49:56,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:49:56,739 INFO [train.py:968] (0/2) Epoch 8, batch 21000, giga_loss[loss=0.2658, simple_loss=0.3389, pruned_loss=0.09636, over 28637.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3674, pruned_loss=0.1123, over 5710597.77 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3617, pruned_loss=0.1046, over 5777468.08 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3667, pruned_loss=0.1124, over 5708675.29 frames. ], batch size: 92, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:49:56,744 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 06:50:05,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3165, 1.6074, 1.2795, 1.3015], device='cuda:0'), covar=tensor([0.2457, 0.2214, 0.2425, 0.1895], device='cuda:0'), in_proj_covar=tensor([0.1213, 0.0913, 0.1068, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 06:50:06,714 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2291, simple_loss=0.3346, pruned_loss=0.06182, over 944034.00 frames. 
+2023-03-04 06:50:06,715 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 06:50:11,637 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:50:28,255 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:50:45,995 INFO [train.py:968] (0/2) Epoch 8, batch 21050, giga_loss[loss=0.3374, simple_loss=0.3814, pruned_loss=0.1467, over 26633.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3655, pruned_loss=0.1111, over 5714275.33 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3622, pruned_loss=0.105, over 5780661.44 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3645, pruned_loss=0.111, over 5708658.95 frames. ], batch size: 555, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:50:51,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.011e+02 9.593e+02 1.404e+03 1.886e+03 1.554e+04, threshold=2.808e+03, percent-clipped=9.0
+2023-03-04 06:50:51,903 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:51:25,018 INFO [train.py:968] (0/2) Epoch 8, batch 21100, giga_loss[loss=0.2687, simple_loss=0.3426, pruned_loss=0.09747, over 28839.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3628, pruned_loss=0.1101, over 5701395.53 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3625, pruned_loss=0.1055, over 5764120.08 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3618, pruned_loss=0.1098, over 5709482.57 frames. ], batch size: 99, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:52:04,561 INFO [train.py:968] (0/2) Epoch 8, batch 21150, giga_loss[loss=0.2761, simple_loss=0.3451, pruned_loss=0.1036, over 28924.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3622, pruned_loss=0.11, over 5704937.21 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3634, pruned_loss=0.1062, over 5767670.61 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3606, pruned_loss=0.1091, over 5706712.67 frames. ], batch size: 112, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:52:05,327 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:52:09,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.461e+02 1.026e+03 1.198e+03 1.595e+03 4.831e+03, threshold=2.396e+03, percent-clipped=5.0
+2023-03-04 06:52:18,623 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 06:52:30,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3155, 1.5079, 1.2441, 1.4652], device='cuda:0'), covar=tensor([0.0742, 0.0343, 0.0321, 0.0777], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:0')
+2023-03-04 06:52:47,584 INFO [train.py:968] (0/2) Epoch 8, batch 21200, giga_loss[loss=0.3081, simple_loss=0.3488, pruned_loss=0.1337, over 23613.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3621, pruned_loss=0.1105, over 5691243.14 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3637, pruned_loss=0.1066, over 5749158.21 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3605, pruned_loss=0.1094, over 5708815.79 frames. ], batch size: 705, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:52:47,833 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:52:50,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:53:04,180 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-340000.pt
+2023-03-04 06:53:13,392 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:53:29,288 INFO [train.py:968] (0/2) Epoch 8, batch 21250, giga_loss[loss=0.2775, simple_loss=0.3486, pruned_loss=0.1032, over 28713.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3631, pruned_loss=0.1114, over 5690957.79 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3642, pruned_loss=0.107, over 5751201.28 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3613, pruned_loss=0.1103, over 5702153.14 frames. ], batch size: 99, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:53:34,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.711e+02 1.010e+03 1.266e+03 1.798e+03 9.573e+03, threshold=2.533e+03, percent-clipped=18.0
+2023-03-04 06:53:57,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=340066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:54:07,039 INFO [train.py:968] (0/2) Epoch 8, batch 21300, giga_loss[loss=0.3175, simple_loss=0.3813, pruned_loss=0.1268, over 28433.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3622, pruned_loss=0.11, over 5704953.82 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3645, pruned_loss=0.1073, over 5754133.35 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3605, pruned_loss=0.1089, over 5710237.51 frames. ], batch size: 65, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:54:29,265 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=340107.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 06:54:47,452 INFO [train.py:968] (0/2) Epoch 8, batch 21350, giga_loss[loss=0.2881, simple_loss=0.3542, pruned_loss=0.1109, over 28881.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3606, pruned_loss=0.1081, over 5706336.00 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3649, pruned_loss=0.1076, over 5755379.58 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3589, pruned_loss=0.107, over 5708987.31 frames. ], batch size: 186, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:54:54,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.813e+02 9.451e+02 1.109e+03 1.325e+03 4.982e+03, threshold=2.217e+03, percent-clipped=7.0
+2023-03-04 06:54:59,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3262, 1.5068, 1.0736, 1.2089], device='cuda:0'), covar=tensor([0.1536, 0.1228, 0.1307, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1431, 0.1411, 0.1518], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 06:55:03,676 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=340145.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 06:55:06,543 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-04 06:55:07,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1785, 1.4378, 1.4736, 1.4085], device='cuda:0'), covar=tensor([0.1512, 0.1402, 0.1768, 0.1372], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0711, 0.0639, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 06:55:15,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=340161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:55:29,642 INFO [train.py:968] (0/2) Epoch 8, batch 21400, giga_loss[loss=0.2919, simple_loss=0.363, pruned_loss=0.1104, over 28980.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3602, pruned_loss=0.1086, over 5690295.83 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3651, pruned_loss=0.1078, over 5747024.01 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3585, pruned_loss=0.1076, over 5699048.68 frames. ], batch size: 145, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:55:55,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=340209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:55:57,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=340212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:56:00,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7884, 1.9241, 1.7881, 1.7520], device='cuda:0'), covar=tensor([0.1391, 0.1775, 0.1785, 0.1730], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0717, 0.0645, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 06:56:10,141 INFO [train.py:968] (0/2) Epoch 8, batch 21450, giga_loss[loss=0.2591, simple_loss=0.332, pruned_loss=0.0931, over 28329.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3589, pruned_loss=0.1087, over 5693638.86 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3658, pruned_loss=0.1085, over 5749003.56 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3569, pruned_loss=0.1071, over 5697947.34 frames. ], batch size: 71, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:56:15,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.728e+02 9.906e+02 1.272e+03 1.678e+03 9.998e+03, threshold=2.543e+03, percent-clipped=12.0
+2023-03-04 06:56:21,956 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:56:28,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2436, 3.0344, 2.8911, 1.4020], device='cuda:0'), covar=tensor([0.0826, 0.0946, 0.0865, 0.2417], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0878, 0.0774, 0.0624], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 06:56:53,706 INFO [train.py:968] (0/2) Epoch 8, batch 21500, giga_loss[loss=0.2842, simple_loss=0.3551, pruned_loss=0.1066, over 27910.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3555, pruned_loss=0.1072, over 5690174.21 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.366, pruned_loss=0.1087, over 5746296.54 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3538, pruned_loss=0.1059, over 5695794.61 frames. ], batch size: 412, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:57:07,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6648, 1.6432, 1.6529, 1.5812], device='cuda:0'), covar=tensor([0.1368, 0.1917, 0.1837, 0.1728], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0721, 0.0647, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 06:57:13,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=340304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:57:14,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=340305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:57:15,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=340307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:57:32,487 INFO [train.py:968] (0/2) Epoch 8, batch 21550, giga_loss[loss=0.3161, simple_loss=0.3817, pruned_loss=0.1252, over 28619.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3542, pruned_loss=0.1068, over 5681384.64 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3665, pruned_loss=0.1092, over 5742588.57 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3519, pruned_loss=0.1052, over 5687989.51 frames. ], batch size: 307, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:57:37,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.893e+02 9.973e+02 1.240e+03 1.788e+03 6.116e+03, threshold=2.479e+03, percent-clipped=9.0
+2023-03-04 06:57:37,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:57:39,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 06:58:13,032 INFO [train.py:968] (0/2) Epoch 8, batch 21600, giga_loss[loss=0.2749, simple_loss=0.3526, pruned_loss=0.09855, over 28436.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3538, pruned_loss=0.1068, over 5690574.17 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3665, pruned_loss=0.1092, over 5746632.19 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3517, pruned_loss=0.1054, over 5691116.47 frames. ], batch size: 65, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 06:58:49,744 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3445, 1.6146, 1.2605, 1.5372], device='cuda:0'), covar=tensor([0.2100, 0.1936, 0.2145, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.1211, 0.0910, 0.1063, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 06:58:51,664 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-04 06:58:52,818 INFO [train.py:968] (0/2) Epoch 8, batch 21650, giga_loss[loss=0.2817, simple_loss=0.345, pruned_loss=0.1092, over 28826.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3524, pruned_loss=0.1066, over 5693910.69 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3667, pruned_loss=0.1095, over 5744512.40 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3502, pruned_loss=0.1052, over 5694793.18 frames. ], batch size: 119, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 06:58:57,937 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 06:58:58,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.624e+02 1.168e+03 1.419e+03 1.812e+03 3.792e+03, threshold=2.838e+03, percent-clipped=9.0
+2023-03-04 06:59:07,165 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=340448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:59:09,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=340451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:59:30,571 INFO [train.py:968] (0/2) Epoch 8, batch 21700, giga_loss[loss=0.238, simple_loss=0.3163, pruned_loss=0.07984, over 28388.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.351, pruned_loss=0.1062, over 5684093.29 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3674, pruned_loss=0.1102, over 5731635.99 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.348, pruned_loss=0.1043, over 5694721.11 frames. ], batch size: 65, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:59:31,549 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:59:33,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=340482.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:00:02,813 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=340520.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:00:10,654 INFO [train.py:968] (0/2) Epoch 8, batch 21750, giga_loss[loss=0.3008, simple_loss=0.3593, pruned_loss=0.1211, over 29049.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3482, pruned_loss=0.105, over 5696247.25 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3672, pruned_loss=0.1104, over 5736010.22 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3456, pruned_loss=0.1032, over 5699898.23 frames. ], batch size: 128, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:00:16,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.463e+02 1.071e+03 1.469e+03 2.130e+03 4.872e+03, threshold=2.937e+03, percent-clipped=12.0
+2023-03-04 07:00:33,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8003, 1.7908, 1.8650, 1.8273], device='cuda:0'), covar=tensor([0.1243, 0.1794, 0.1617, 0.1510], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0723, 0.0649, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 07:00:48,754 INFO [train.py:968] (0/2) Epoch 8, batch 21800, giga_loss[loss=0.2505, simple_loss=0.3256, pruned_loss=0.0877, over 28879.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3466, pruned_loss=0.1042, over 5709544.78 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3679, pruned_loss=0.1109, over 5740288.96 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3435, pruned_loss=0.1022, over 5708162.25 frames. ], batch size: 145, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:01:19,176 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-04 07:01:27,851 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 07:01:29,514 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=340625.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:01:32,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=340628.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:01:32,587 INFO [train.py:968] (0/2) Epoch 8, batch 21850, giga_loss[loss=0.2715, simple_loss=0.3526, pruned_loss=0.09516, over 28935.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3473, pruned_loss=0.1047, over 5703080.30 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3684, pruned_loss=0.1115, over 5737332.15 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3441, pruned_loss=0.1025, over 5704101.59 frames. ], batch size: 174, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:01:38,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.408e+02 9.553e+02 1.208e+03 1.620e+03 1.164e+04, threshold=2.417e+03, percent-clipped=7.0
+2023-03-04 07:01:59,002 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340657.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:02:03,917 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=340663.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:02:07,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=340666.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:02:15,831 INFO [train.py:968] (0/2) Epoch 8, batch 21900, giga_loss[loss=0.2811, simple_loss=0.3606, pruned_loss=0.1008, over 29044.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3505, pruned_loss=0.1059, over 5704310.46 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3686, pruned_loss=0.1117, over 5739240.51 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3476, pruned_loss=0.104, over 5702930.41 frames. ], batch size: 164, lr: 4.05e-03, grad_scale: 2.0
+2023-03-04 07:02:30,275 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340695.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:03:01,437 INFO [train.py:968] (0/2) Epoch 8, batch 21950, giga_loss[loss=0.2602, simple_loss=0.344, pruned_loss=0.08814, over 28953.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3549, pruned_loss=0.1076, over 5690510.83 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3689, pruned_loss=0.1119, over 5737916.81 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3522, pruned_loss=0.1058, over 5690080.86 frames. ], batch size: 213, lr: 4.05e-03, grad_scale: 2.0
+2023-03-04 07:03:08,311 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.727e+02 1.048e+03 1.342e+03 1.855e+03 6.616e+03, threshold=2.685e+03, percent-clipped=12.0
+2023-03-04 07:03:33,903 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-04 07:03:41,996 INFO [train.py:968] (0/2) Epoch 8, batch 22000, giga_loss[loss=0.2808, simple_loss=0.3562, pruned_loss=0.1027, over 28889.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3559, pruned_loss=0.1072, over 5705535.02 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3692, pruned_loss=0.1123, over 5743825.64 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.353, pruned_loss=0.1053, over 5698549.94 frames. ], batch size: 119, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:03:54,621 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-04 07:04:04,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0094, 1.2551, 3.8229, 3.0491], device='cuda:0'), covar=tensor([0.1538, 0.2183, 0.0395, 0.0779], device='cuda:0'), in_proj_covar=tensor([0.0600, 0.0552, 0.0791, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:04:21,755 INFO [train.py:968] (0/2) Epoch 8, batch 22050, giga_loss[loss=0.2936, simple_loss=0.3583, pruned_loss=0.1145, over 26695.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3547, pruned_loss=0.106, over 5708108.40 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.369, pruned_loss=0.1125, over 5746365.03 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3522, pruned_loss=0.1041, over 5699542.04 frames. ], batch size: 555, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:04:30,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.449e+02 1.049e+03 1.332e+03 1.661e+03 3.361e+03, threshold=2.663e+03, percent-clipped=5.0
+2023-03-04 07:04:32,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5312, 1.8885, 1.8437, 1.4076], device='cuda:0'), covar=tensor([0.1689, 0.1879, 0.1333, 0.1429], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0701, 0.0819, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 07:05:03,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5664, 1.5767, 1.3763, 1.7981], device='cuda:0'), covar=tensor([0.2370, 0.2513, 0.2642, 0.2425], device='cuda:0'), in_proj_covar=tensor([0.1217, 0.0912, 0.1068, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:05:06,502 INFO [train.py:968] (0/2) Epoch 8, batch 22100, giga_loss[loss=0.31, simple_loss=0.3722, pruned_loss=0.1238, over 28955.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3544, pruned_loss=0.1058, over 5704881.08 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3695, pruned_loss=0.1129, over 5747493.98 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3519, pruned_loss=0.1039, over 5696680.99 frames. ], batch size: 136, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:05:36,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3557, 1.6203, 1.4150, 1.6026], device='cuda:0'), covar=tensor([0.0614, 0.0277, 0.0279, 0.0648], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0118, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0071, 0.0050, 0.0046, 0.0077], device='cuda:0')
+2023-03-04 07:05:39,128 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=340918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:05:47,264 INFO [train.py:968] (0/2) Epoch 8, batch 22150, giga_loss[loss=0.2879, simple_loss=0.3501, pruned_loss=0.1129, over 28735.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3544, pruned_loss=0.1063, over 5707752.51 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3689, pruned_loss=0.1128, over 5752130.87 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3525, pruned_loss=0.1046, over 5695735.10 frames. ], batch size: 92, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:05:53,993 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-04 07:05:54,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.825e+02 1.114e+03 1.410e+03 1.896e+03 6.631e+03, threshold=2.819e+03, percent-clipped=7.0
+2023-03-04 07:06:02,031 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-04 07:06:28,229 INFO [train.py:968] (0/2) Epoch 8, batch 22200, giga_loss[loss=0.3612, simple_loss=0.4211, pruned_loss=0.1507, over 28671.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3551, pruned_loss=0.1074, over 5697044.04 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3688, pruned_loss=0.113, over 5745574.32 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3534, pruned_loss=0.1058, over 5692045.61 frames. ], batch size: 262, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:06:32,506 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-04 07:06:41,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4842, 1.8092, 1.7960, 1.4139], device='cuda:0'), covar=tensor([0.1619, 0.1903, 0.1290, 0.1393], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0706, 0.0823, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 07:07:07,748 INFO [train.py:968] (0/2) Epoch 8, batch 22250, giga_loss[loss=0.2864, simple_loss=0.3587, pruned_loss=0.1071, over 28831.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3592, pruned_loss=0.1099, over 5710530.53 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3693, pruned_loss=0.1136, over 5750292.91 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3569, pruned_loss=0.1079, over 5700816.49 frames. ], batch size: 186, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:07:13,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.345e+02 1.131e+03 1.342e+03 2.078e+03 6.304e+03, threshold=2.685e+03, percent-clipped=12.0
+2023-03-04 07:07:30,827 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-04 07:07:44,166 INFO [train.py:968] (0/2) Epoch 8, batch 22300, giga_loss[loss=0.3122, simple_loss=0.3766, pruned_loss=0.124, over 28827.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3617, pruned_loss=0.1112, over 5718743.93 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3693, pruned_loss=0.1139, over 5754250.90 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3596, pruned_loss=0.1093, over 5706352.27 frames. ], batch size: 145, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:08:26,382 INFO [train.py:968] (0/2) Epoch 8, batch 22350, giga_loss[loss=0.2664, simple_loss=0.3512, pruned_loss=0.09082, over 28956.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3636, pruned_loss=0.1122, over 5719033.58 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3697, pruned_loss=0.1142, over 5756172.20 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3615, pruned_loss=0.1103, over 5707039.99 frames. ], batch size: 164, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:08:32,951 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-04 07:08:33,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.797e+02 1.408e+03 1.811e+03 2.545e+03 9.262e+03, threshold=3.622e+03, percent-clipped=19.0
+2023-03-04 07:08:44,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2534, 1.5888, 1.3556, 1.4722], device='cuda:0'), covar=tensor([0.0673, 0.0404, 0.0316, 0.0711], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0118, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:0')
+2023-03-04 07:09:05,548 INFO [train.py:968] (0/2) Epoch 8, batch 22400, giga_loss[loss=0.3327, simple_loss=0.3838, pruned_loss=0.1408, over 28418.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3631, pruned_loss=0.1113, over 5725597.89 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3696, pruned_loss=0.1142, over 5756942.33 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3615, pruned_loss=0.1099, over 5715368.10 frames. ], batch size: 71, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:09:32,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5136, 1.7005, 1.3933, 1.2578], device='cuda:0'), covar=tensor([0.1763, 0.1319, 0.1130, 0.1460], device='cuda:0'), in_proj_covar=tensor([0.1609, 0.1445, 0.1437, 0.1532], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 07:09:38,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5191, 2.1536, 1.5228, 1.6329], device='cuda:0'), covar=tensor([0.0686, 0.0224, 0.0298, 0.0733], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0118, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:0')
+2023-03-04 07:09:49,613 INFO [train.py:968] (0/2) Epoch 8, batch 22450, giga_loss[loss=0.2878, simple_loss=0.3576, pruned_loss=0.109, over 29120.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3631, pruned_loss=0.1112, over 5718964.57 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3698, pruned_loss=0.1144, over 5758099.62 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3615, pruned_loss=0.1098, over 5709302.01 frames. ], batch size: 128, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:09:57,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.327e+02 1.215e+03 1.522e+03 1.851e+03 5.376e+03, threshold=3.044e+03, percent-clipped=2.0
+2023-03-04 07:10:00,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2131, 3.0155, 2.8282, 1.4316], device='cuda:0'), covar=tensor([0.0840, 0.0877, 0.0838, 0.2262], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0883, 0.0783, 0.0623], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 07:10:30,923 INFO [train.py:968] (0/2) Epoch 8, batch 22500, giga_loss[loss=0.2405, simple_loss=0.3157, pruned_loss=0.08268, over 28810.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3623, pruned_loss=0.111, over 5724492.82 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3705, pruned_loss=0.115, over 5759388.46 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3602, pruned_loss=0.1093, over 5714605.16 frames. ], batch size: 119, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:10:43,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=341293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:11:13,324 INFO [train.py:968] (0/2) Epoch 8, batch 22550, giga_loss[loss=0.2507, simple_loss=0.3241, pruned_loss=0.08859, over 28944.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3595, pruned_loss=0.1096, over 5723020.01 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3705, pruned_loss=0.1151, over 5758918.11 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3577, pruned_loss=0.1081, over 5715073.09 frames. ], batch size: 106, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:11:22,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.292e+02 1.110e+03 1.433e+03 1.891e+03 4.365e+03, threshold=2.865e+03, percent-clipped=7.0
+2023-03-04 07:11:24,768 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-04 07:11:55,101 INFO [train.py:968] (0/2) Epoch 8, batch 22600, giga_loss[loss=0.2311, simple_loss=0.3067, pruned_loss=0.07772, over 28466.00 frames. ], tot_loss[loss=0.286, simple_loss=0.356, pruned_loss=0.108, over 5713407.80 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3705, pruned_loss=0.1152, over 5751654.68 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3544, pruned_loss=0.1066, over 5712805.26 frames. ], batch size: 71, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:12:10,086 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 07:12:14,311 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4279, 1.5825, 1.2596, 1.8849], device='cuda:0'), covar=tensor([0.2186, 0.2186, 0.2365, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.1207, 0.0905, 0.1060, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:12:33,980 INFO [train.py:968] (0/2) Epoch 8, batch 22650, giga_loss[loss=0.2962, simple_loss=0.3579, pruned_loss=0.1173, over 28823.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3545, pruned_loss=0.1068, over 5717785.50 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3703, pruned_loss=0.1153, over 5755640.84 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3531, pruned_loss=0.1054, over 5712923.86 frames. ], batch size: 99, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:12:40,699 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=341436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:12:43,295 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.367e+02 1.063e+03 1.296e+03 1.575e+03 3.472e+03, threshold=2.592e+03, percent-clipped=3.0
+2023-03-04 07:12:43,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=341439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:13:10,993 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=341468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:13:18,753 INFO [train.py:968] (0/2) Epoch 8, batch 22700, libri_loss[loss=0.2963, simple_loss=0.3577, pruned_loss=0.1174, over 29474.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3556, pruned_loss=0.1059, over 5706095.31 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3708, pruned_loss=0.1158, over 5745274.22 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3537, pruned_loss=0.1043, over 5709801.43 frames. ], batch size: 70, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:13:49,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0375, 1.0317, 3.7261, 3.1976], device='cuda:0'), covar=tensor([0.1573, 0.2511, 0.0376, 0.1005], device='cuda:0'), in_proj_covar=tensor([0.0603, 0.0555, 0.0800, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:13:58,323 INFO [train.py:968] (0/2) Epoch 8, batch 22750, giga_loss[loss=0.2677, simple_loss=0.3367, pruned_loss=0.09939, over 28685.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3573, pruned_loss=0.1066, over 5713082.06 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3713, pruned_loss=0.1164, over 5748893.21 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3548, pruned_loss=0.1043, over 5711635.92 frames. ], batch size: 92, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:14:03,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3613, 1.9629, 1.4261, 0.5975], device='cuda:0'), covar=tensor([0.3503, 0.1630, 0.2353, 0.3990], device='cuda:0'), in_proj_covar=tensor([0.1460, 0.1382, 0.1445, 0.1191], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 07:14:05,806 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.456e+02 1.051e+03 1.263e+03 1.550e+03 3.949e+03, threshold=2.525e+03, percent-clipped=5.0
+2023-03-04 07:14:38,277 INFO [train.py:968] (0/2) Epoch 8, batch 22800, giga_loss[loss=0.3308, simple_loss=0.3849, pruned_loss=0.1384, over 28905.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.356, pruned_loss=0.1065, over 5721391.10 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3713, pruned_loss=0.1165, over 5750500.35 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3539, pruned_loss=0.1045, over 5718585.62 frames. ], batch size: 174, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:15:21,419 INFO [train.py:968] (0/2) Epoch 8, batch 22850, giga_loss[loss=0.2533, simple_loss=0.326, pruned_loss=0.09028, over 29010.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3551, pruned_loss=0.1073, over 5721796.92 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3719, pruned_loss=0.117, over 5752264.48 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3528, pruned_loss=0.1052, over 5717786.46 frames. ], batch size: 128, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:15:27,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.693e+02 1.046e+03 1.403e+03 1.958e+03 3.764e+03, threshold=2.806e+03, percent-clipped=11.0
+2023-03-04 07:16:00,790 INFO [train.py:968] (0/2) Epoch 8, batch 22900, giga_loss[loss=0.2993, simple_loss=0.3635, pruned_loss=0.1175, over 28823.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3537, pruned_loss=0.1078, over 5719556.14 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.372, pruned_loss=0.1172, over 5752667.29 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3514, pruned_loss=0.1057, over 5715622.65 frames. ], batch size: 99, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:16:04,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7075, 1.6954, 1.2746, 1.4827], device='cuda:0'), covar=tensor([0.0672, 0.0610, 0.0904, 0.1051], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0444, 0.0490, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 07:16:05,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3280, 4.1728, 3.9107, 1.7689], device='cuda:0'), covar=tensor([0.0555, 0.0671, 0.0743, 0.2238], device='cuda:0'), in_proj_covar=tensor([0.0950, 0.0899, 0.0796, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 07:16:43,737 INFO [train.py:968] (0/2) Epoch 8, batch 22950, giga_loss[loss=0.2475, simple_loss=0.3124, pruned_loss=0.09133, over 28623.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3525, pruned_loss=0.1083, over 5723650.60 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3723, pruned_loss=0.1177, over 5753426.59 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3501, pruned_loss=0.1061, over 5719101.74 frames. ], batch size: 60, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:16:50,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.060e+02 1.123e+03 1.679e+03 2.281e+03 1.031e+04, threshold=3.358e+03, percent-clipped=16.0
+2023-03-04 07:17:20,849 INFO [train.py:968] (0/2) Epoch 8, batch 23000, giga_loss[loss=0.2466, simple_loss=0.3184, pruned_loss=0.08742, over 28951.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3512, pruned_loss=0.1078, over 5714836.41 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3723, pruned_loss=0.118, over 5747252.97 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3485, pruned_loss=0.1055, over 5714911.80 frames. ], batch size: 213, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:17:52,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3203, 1.4911, 1.5358, 1.3429], device='cuda:0'), covar=tensor([0.1064, 0.1181, 0.1507, 0.1188], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0720, 0.0648, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 07:17:58,772 INFO [train.py:968] (0/2) Epoch 8, batch 23050, giga_loss[loss=0.2464, simple_loss=0.3183, pruned_loss=0.08725, over 28917.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3482, pruned_loss=0.1066, over 5711432.32 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3726, pruned_loss=0.1183, over 5738845.86 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3451, pruned_loss=0.1041, over 5717602.62 frames. ], batch size: 174, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:18:06,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.880e+02 1.140e+03 1.473e+03 2.152e+03 5.908e+03, threshold=2.946e+03, percent-clipped=4.0
+2023-03-04 07:18:11,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2618, 1.5584, 1.1844, 1.6586], device='cuda:0'), covar=tensor([0.2272, 0.2101, 0.2482, 0.1981], device='cuda:0'), in_proj_covar=tensor([0.1208, 0.0899, 0.1059, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:18:32,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9948, 1.2029, 1.3083, 1.0980], device='cuda:0'), covar=tensor([0.1400, 0.1180, 0.1943, 0.1400], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0722, 0.0651, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 07:18:38,211 INFO [train.py:968] (0/2) Epoch 8, batch 23100, libri_loss[loss=0.3018, simple_loss=0.3695, pruned_loss=0.1171, over 29535.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3444, pruned_loss=0.1045, over 5704517.52 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3725, pruned_loss=0.1183, over 5732464.73 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3415, pruned_loss=0.1022, over 5714003.86 frames. ], batch size: 78, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:18:52,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3822, 3.3707, 1.5133, 1.3628], device='cuda:0'), covar=tensor([0.0912, 0.0321, 0.0910, 0.1342], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0500, 0.0324, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 07:19:15,986 INFO [train.py:968] (0/2) Epoch 8, batch 23150, giga_loss[loss=0.2477, simple_loss=0.3272, pruned_loss=0.08407, over 28998.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3403, pruned_loss=0.102, over 5713913.85 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3724, pruned_loss=0.1183, over 5734903.98 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3377, pruned_loss=0.1, over 5719121.80 frames. ], batch size: 155, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:19:25,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.041e+02 1.035e+03 1.398e+03 1.801e+03 4.729e+03, threshold=2.797e+03, percent-clipped=7.0
+2023-03-04 07:19:30,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0720, 2.0246, 1.9541, 1.7964], device='cuda:0'), covar=tensor([0.1220, 0.1898, 0.1625, 0.1861], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0722, 0.0650, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 07:19:57,973 INFO [train.py:968] (0/2) Epoch 8, batch 23200, giga_loss[loss=0.2832, simple_loss=0.356, pruned_loss=0.1052, over 28919.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3444, pruned_loss=0.1041, over 5707978.42 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3735, pruned_loss=0.1192, over 5735425.54 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3407, pruned_loss=0.1014, over 5711149.72 frames. ], batch size: 186, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:19:58,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5817, 1.6950, 1.3188, 1.3082], device='cuda:0'), covar=tensor([0.1613, 0.1332, 0.1137, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1463, 0.1437, 0.1545], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 07:20:14,769 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-342000.pt
+2023-03-04 07:20:37,178 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 07:20:40,443 INFO [train.py:968] (0/2) Epoch 8, batch 23250, giga_loss[loss=0.2986, simple_loss=0.3724, pruned_loss=0.1124, over 28557.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.347, pruned_loss=0.1049, over 5697890.42 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3735, pruned_loss=0.1192, over 5726059.41 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3441, pruned_loss=0.1028, over 5708077.19 frames. ], batch size: 336, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:20:49,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.337e+02 1.092e+03 1.341e+03 1.701e+03 4.689e+03, threshold=2.681e+03, percent-clipped=7.0
+2023-03-04 07:21:21,310 INFO [train.py:968] (0/2) Epoch 8, batch 23300, giga_loss[loss=0.3175, simple_loss=0.3753, pruned_loss=0.1298, over 26689.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3498, pruned_loss=0.1056, over 5699461.91 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3733, pruned_loss=0.1192, over 5727935.59 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3474, pruned_loss=0.1037, over 5705574.92 frames. ], batch size: 555, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:21:37,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6818, 2.3592, 1.5169, 0.8500], device='cuda:0'), covar=tensor([0.4380, 0.2348, 0.2476, 0.4071], device='cuda:0'), in_proj_covar=tensor([0.1463, 0.1382, 0.1442, 0.1196], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 07:22:00,727 INFO [train.py:968] (0/2) Epoch 8, batch 23350, giga_loss[loss=0.3634, simple_loss=0.4053, pruned_loss=0.1607, over 28728.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3544, pruned_loss=0.1082, over 5712387.61 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3743, pruned_loss=0.1202, over 5733545.72 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3508, pruned_loss=0.1053, over 5711257.77 frames. ], batch size: 92, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:22:10,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.260e+02 1.152e+03 1.478e+03 2.295e+03 1.004e+04, threshold=2.956e+03, percent-clipped=18.0
+2023-03-04 07:22:44,010 INFO [train.py:968] (0/2) Epoch 8, batch 23400, giga_loss[loss=0.3044, simple_loss=0.3692, pruned_loss=0.1198, over 28993.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3559, pruned_loss=0.1088, over 5719198.37 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3742, pruned_loss=0.1201, over 5735310.83 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.353, pruned_loss=0.1065, over 5716523.19 frames. ], batch size: 145, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:23:14,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6210, 1.9152, 1.4991, 1.7351], device='cuda:0'), covar=tensor([0.2240, 0.2083, 0.2364, 0.2259], device='cuda:0'), in_proj_covar=tensor([0.1216, 0.0908, 0.1063, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:23:22,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0956, 2.1566, 1.4696, 1.8826], device='cuda:0'), covar=tensor([0.0520, 0.0419, 0.0817, 0.0746], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0446, 0.0494, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 07:23:30,360 INFO [train.py:968] (0/2) Epoch 8, batch 23450, giga_loss[loss=0.3625, simple_loss=0.4214, pruned_loss=0.1518, over 28536.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3622, pruned_loss=0.1144, over 5710482.98 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3739, pruned_loss=0.1201, over 5737688.58 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3598, pruned_loss=0.1125, over 5705484.50 frames. ], batch size: 336, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:23:37,230 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=342234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:23:42,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.154e+02 1.184e+03 1.568e+03 2.303e+03 8.426e+03, threshold=3.137e+03, percent-clipped=13.0
+2023-03-04 07:24:23,443 INFO [train.py:968] (0/2) Epoch 8, batch 23500, giga_loss[loss=0.4197, simple_loss=0.4572, pruned_loss=0.1911, over 28535.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.1209, over 5698971.70 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3739, pruned_loss=0.1202, over 5738534.82 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3681, pruned_loss=0.1193, over 5693994.63 frames. ], batch size: 336, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:25:15,821 INFO [train.py:968] (0/2) Epoch 8, batch 23550, giga_loss[loss=0.3027, simple_loss=0.3699, pruned_loss=0.1178, over 28477.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3772, pruned_loss=0.1266, over 5689541.87 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3738, pruned_loss=0.1202, over 5741068.40 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3757, pruned_loss=0.1253, over 5682703.15 frames. ], batch size: 60, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:25:28,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.308e+02 1.609e+03 2.101e+03 2.973e+03 7.687e+03, threshold=4.203e+03, percent-clipped=21.0
+2023-03-04 07:26:02,524 INFO [train.py:968] (0/2) Epoch 8, batch 23600, giga_loss[loss=0.3544, simple_loss=0.4092, pruned_loss=0.1498, over 28605.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3822, pruned_loss=0.1307, over 5677476.54 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3738, pruned_loss=0.1202, over 5732012.48 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3811, pruned_loss=0.1297, over 5679831.09 frames. ], batch size: 307, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:26:04,356 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=342381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:26:18,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5018, 1.3870, 4.8626, 3.5421], device='cuda:0'), covar=tensor([0.1639, 0.2368, 0.0315, 0.0699], device='cuda:0'), in_proj_covar=tensor([0.0600, 0.0553, 0.0796, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:26:49,967 INFO [train.py:968] (0/2) Epoch 8, batch 23650, giga_loss[loss=0.3032, simple_loss=0.3733, pruned_loss=0.1166, over 28969.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3896, pruned_loss=0.1369, over 5684116.90 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3741, pruned_loss=0.1206, over 5736925.79 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3888, pruned_loss=0.1363, over 5679641.31 frames. ], batch size: 106, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:27:01,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.593e+02 1.593e+03 2.121e+03 2.832e+03 7.988e+03, threshold=4.243e+03, percent-clipped=6.0
+2023-03-04 07:27:40,037 INFO [train.py:968] (0/2) Epoch 8, batch 23700, giga_loss[loss=0.3562, simple_loss=0.4065, pruned_loss=0.1529, over 28299.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3931, pruned_loss=0.1401, over 5676602.07 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3749, pruned_loss=0.1212, over 5736727.25 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3921, pruned_loss=0.1394, over 5672066.62 frames. ], batch size: 368, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:28:29,558 INFO [train.py:968] (0/2) Epoch 8, batch 23750, libri_loss[loss=0.3129, simple_loss=0.3809, pruned_loss=0.1224, over 29557.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3947, pruned_loss=0.1424, over 5664021.19 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.375, pruned_loss=0.1214, over 5729542.43 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3941, pruned_loss=0.1419, over 5665434.29 frames. ], batch size: 76, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:28:42,717 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.862e+02 1.728e+03 2.151e+03 2.753e+03 7.888e+03, threshold=4.301e+03, percent-clipped=7.0
+2023-03-04 07:29:19,713 INFO [train.py:968] (0/2) Epoch 8, batch 23800, giga_loss[loss=0.2983, simple_loss=0.3713, pruned_loss=0.1127, over 29014.00 frames. ], tot_loss[loss=0.3484, simple_loss=0.4001, pruned_loss=0.1484, over 5653732.09 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.375, pruned_loss=0.1214, over 5734833.27 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4002, pruned_loss=0.1486, over 5648245.29 frames. ], batch size: 164, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:29:23,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8107, 1.7126, 1.3168, 1.4777], device='cuda:0'), covar=tensor([0.0665, 0.0645, 0.0893, 0.1061], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0450, 0.0497, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 07:29:47,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=342609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:30:07,921 INFO [train.py:968] (0/2) Epoch 8, batch 23850, giga_loss[loss=0.3722, simple_loss=0.4156, pruned_loss=0.1644, over 28784.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4022, pruned_loss=0.1511, over 5653423.50 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3754, pruned_loss=0.1219, over 5738699.55 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4031, pruned_loss=0.1521, over 5641305.69 frames. ], batch size: 284, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:30:13,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4583, 1.2291, 4.5753, 3.5099], device='cuda:0'), covar=tensor([0.1575, 0.2406, 0.0363, 0.0750], device='cuda:0'), in_proj_covar=tensor([0.0609, 0.0562, 0.0811, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:30:19,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.399e+02 1.712e+03 2.096e+03 2.792e+03 7.431e+03, threshold=4.191e+03, percent-clipped=6.0
+2023-03-04 07:31:01,244 INFO [train.py:968] (0/2) Epoch 8, batch 23900, giga_loss[loss=0.3692, simple_loss=0.4107, pruned_loss=0.1638, over 28526.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4061, pruned_loss=0.1532, over 5660497.73 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3758, pruned_loss=0.1225, over 5741682.16 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.4075, pruned_loss=0.1546, over 5645062.90 frames. ], batch size: 336, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:31:26,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=342704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:31:40,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2901, 2.9195, 1.4315, 1.3611], device='cuda:0'), covar=tensor([0.0851, 0.0292, 0.0753, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0499, 0.0323, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 07:31:53,545 INFO [train.py:968] (0/2) Epoch 8, batch 23950, giga_loss[loss=0.3722, simple_loss=0.4143, pruned_loss=0.165, over 27898.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4062, pruned_loss=0.1548, over 5651921.00 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3757, pruned_loss=0.1226, over 5743041.25 frames. ], giga_tot_loss[loss=0.36, simple_loss=0.4078, pruned_loss=0.1561, over 5637592.13 frames. ], batch size: 412, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:32:09,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.757e+03 2.216e+03 2.775e+03 5.699e+03, threshold=4.433e+03, percent-clipped=7.0
+2023-03-04 07:32:18,868 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=342752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:32:21,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=342755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:32:22,891 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=342756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:32:44,807 INFO [train.py:968] (0/2) Epoch 8, batch 24000, giga_loss[loss=0.331, simple_loss=0.3961, pruned_loss=0.133, over 29140.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4039, pruned_loss=0.1536, over 5652824.81 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3754, pruned_loss=0.1227, over 5744440.22 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.406, pruned_loss=0.1553, over 5638393.76 frames. ], batch size: 155, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:32:44,811 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 07:32:51,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3271, 1.6164, 1.2852, 1.4336], device='cuda:0'), covar=tensor([0.2707, 0.2593, 0.2899, 0.2102], device='cuda:0'), in_proj_covar=tensor([0.1212, 0.0908, 0.1063, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:32:52,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1972, 1.7177, 1.5077, 1.0987], device='cuda:0'), covar=tensor([0.1474, 0.2223, 0.1326, 0.1499], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0706, 0.0816, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 07:32:53,830 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2238, simple_loss=0.3302, pruned_loss=0.05867, over 944034.00 frames. 
+2023-03-04 07:32:53,831 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 07:32:54,392 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-04 07:32:58,295 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=342784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:33:06,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7660, 1.1215, 2.8458, 2.7140], device='cuda:0'), covar=tensor([0.1594, 0.2214, 0.0594, 0.1017], device='cuda:0'), in_proj_covar=tensor([0.0609, 0.0560, 0.0807, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:33:28,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-04 07:33:38,674 INFO [train.py:968] (0/2) Epoch 8, batch 24050, giga_loss[loss=0.3721, simple_loss=0.4256, pruned_loss=0.1593, over 28774.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4045, pruned_loss=0.1527, over 5653748.20 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3762, pruned_loss=0.1232, over 5742763.95 frames. ], giga_tot_loss[loss=0.3573, simple_loss=0.4061, pruned_loss=0.1543, over 5641462.29 frames. ], batch size: 284, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:33:52,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.401e+02 1.615e+03 2.019e+03 2.848e+03 1.148e+04, threshold=4.038e+03, percent-clipped=8.0
+2023-03-04 07:34:28,642 INFO [train.py:968] (0/2) Epoch 8, batch 24100, giga_loss[loss=0.3188, simple_loss=0.3852, pruned_loss=0.1262, over 28759.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.4049, pruned_loss=0.1528, over 5642257.48 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3762, pruned_loss=0.1235, over 5736814.96 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.407, pruned_loss=0.1546, over 5634861.55 frames. ], batch size: 284, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:34:48,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=342899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:34:50,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=342902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:35:15,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=342928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:35:15,736 INFO [train.py:968] (0/2) Epoch 8, batch 24150, giga_loss[loss=0.3127, simple_loss=0.3798, pruned_loss=0.1228, over 28946.00 frames. ], tot_loss[loss=0.3557, simple_loss=0.4054, pruned_loss=0.153, over 5625267.81 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3764, pruned_loss=0.1238, over 5728379.47 frames. ], giga_tot_loss[loss=0.3587, simple_loss=0.4077, pruned_loss=0.1549, over 5624654.57 frames. ], batch size: 213, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:35:17,714 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=342931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:35:20,770 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=342933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:35:32,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.208e+03 1.775e+03 2.412e+03 3.260e+03 7.649e+03, threshold=4.824e+03, percent-clipped=14.0
+2023-03-04 07:36:10,339 INFO [train.py:968] (0/2) Epoch 8, batch 24200, giga_loss[loss=0.3215, simple_loss=0.3846, pruned_loss=0.1292, over 28700.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.4025, pruned_loss=0.1503, over 5614171.51 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3764, pruned_loss=0.1239, over 5720608.58 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4047, pruned_loss=0.1521, over 5619704.32 frames. ], batch size: 284, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:36:33,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5908, 1.8381, 1.7862, 1.4176], device='cuda:0'), covar=tensor([0.1724, 0.2199, 0.1381, 0.1564], device='cuda:0'), in_proj_covar=tensor([0.0787, 0.0704, 0.0815, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 07:36:38,725 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-04 07:36:58,988 INFO [train.py:968] (0/2) Epoch 8, batch 24250, giga_loss[loss=0.4095, simple_loss=0.4311, pruned_loss=0.1939, over 26762.00 frames. ], tot_loss[loss=0.3447, simple_loss=0.3987, pruned_loss=0.1454, over 5629502.54 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3765, pruned_loss=0.124, over 5724184.66 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.4008, pruned_loss=0.1472, over 5628748.07 frames. ], batch size: 555, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:37:11,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.428e+02 1.606e+03 2.040e+03 2.988e+03 6.227e+03, threshold=4.080e+03, percent-clipped=3.0
+2023-03-04 07:37:45,814 INFO [train.py:968] (0/2) Epoch 8, batch 24300, giga_loss[loss=0.3151, simple_loss=0.3695, pruned_loss=0.1304, over 28741.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3955, pruned_loss=0.1419, over 5642142.86 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3767, pruned_loss=0.1243, over 5722528.70 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3977, pruned_loss=0.1437, over 5640319.91 frames. ], batch size: 92, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:37:46,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:37:51,765 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=343086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:38:09,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3293, 3.1359, 2.9566, 1.3327], device='cuda:0'), covar=tensor([0.0780, 0.0960, 0.0905, 0.2279], device='cuda:0'), in_proj_covar=tensor([0.0972, 0.0924, 0.0813, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 07:38:33,476 INFO [train.py:968] (0/2) Epoch 8, batch 24350, giga_loss[loss=0.3363, simple_loss=0.3944, pruned_loss=0.1392, over 29053.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.393, pruned_loss=0.1392, over 5662869.56 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3767, pruned_loss=0.1243, over 5726430.25 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.395, pruned_loss=0.1409, over 5656590.21 frames. ], batch size: 128, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:38:45,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.963e+02 1.435e+03 1.981e+03 2.889e+03 1.103e+04, threshold=3.962e+03, percent-clipped=11.0
+2023-03-04 07:39:05,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3146, 1.8423, 1.3866, 0.6255], device='cuda:0'), covar=tensor([0.2607, 0.1472, 0.2068, 0.3370], device='cuda:0'), in_proj_covar=tensor([0.1481, 0.1407, 0.1458, 0.1213], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 07:39:22,113 INFO [train.py:968] (0/2) Epoch 8, batch 24400, giga_loss[loss=0.2871, simple_loss=0.3653, pruned_loss=0.1045, over 28806.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3901, pruned_loss=0.1377, over 5656311.94 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3765, pruned_loss=0.1243, over 5727965.38 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.392, pruned_loss=0.1392, over 5649160.49 frames. ], batch size: 119, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:39:48,931 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=343206.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:40:03,464 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343222.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:40:06,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343225.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:40:11,825 INFO [train.py:968] (0/2) Epoch 8, batch 24450, giga_loss[loss=0.3506, simple_loss=0.3816, pruned_loss=0.1599, over 23725.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3899, pruned_loss=0.1372, over 5652846.66 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3768, pruned_loss=0.1246, over 5717918.38 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3913, pruned_loss=0.1383, over 5655841.63 frames. ], batch size: 705, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:40:25,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.797e+02 1.561e+03 2.154e+03 3.262e+03 1.415e+04, threshold=4.308e+03, percent-clipped=14.0
+2023-03-04 07:40:40,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=343254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:40:51,150 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-04 07:40:59,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4865, 1.5602, 1.2359, 1.2384], device='cuda:0'), covar=tensor([0.0744, 0.0554, 0.0971, 0.0944], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0451, 0.0495, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 07:41:04,763 INFO [train.py:968] (0/2) Epoch 8, batch 24500, giga_loss[loss=0.3113, simple_loss=0.3694, pruned_loss=0.1266, over 28824.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3892, pruned_loss=0.1363, over 5668828.10 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.377, pruned_loss=0.1249, over 5722465.13 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3905, pruned_loss=0.1372, over 5665942.63 frames. ], batch size: 284, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:41:17,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3646, 1.4953, 1.4633, 1.3193], device='cuda:0'), covar=tensor([0.1206, 0.1571, 0.1610, 0.1592], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0726, 0.0651, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 07:41:31,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:41:34,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:41:54,366 INFO [train.py:968] (0/2) Epoch 8, batch 24550, giga_loss[loss=0.339, simple_loss=0.4003, pruned_loss=0.1389, over 28677.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3865, pruned_loss=0.1332, over 5672281.68 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3767, pruned_loss=0.1248, over 5727762.56 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3881, pruned_loss=0.1342, over 5663475.79 frames. ], batch size: 92, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:41:59,620 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 07:42:11,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.726e+02 1.276e+03 1.611e+03 2.083e+03 4.815e+03, threshold=3.222e+03, percent-clipped=3.0
+2023-03-04 07:42:45,429 INFO [train.py:968] (0/2) Epoch 8, batch 24600, giga_loss[loss=0.3224, simple_loss=0.3956, pruned_loss=0.1246, over 28977.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3876, pruned_loss=0.1317, over 5666435.45 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3765, pruned_loss=0.125, over 5721709.47 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3892, pruned_loss=0.1325, over 5664303.76 frames. ], batch size: 227, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:43:35,139 INFO [train.py:968] (0/2) Epoch 8, batch 24650, giga_loss[loss=0.3897, simple_loss=0.4196, pruned_loss=0.1799, over 26743.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3893, pruned_loss=0.1331, over 5646732.90 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3768, pruned_loss=0.1255, over 5707044.72 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3907, pruned_loss=0.1335, over 5656757.81 frames. ], batch size: 555, lr: 4.03e-03, grad_scale: 1.0
+2023-03-04 07:43:38,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2427, 1.3507, 1.1585, 1.5097], device='cuda:0'), covar=tensor([0.0733, 0.0309, 0.0313, 0.0770], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0078], device='cuda:0')
+2023-03-04 07:43:42,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3866, 1.6251, 1.6577, 1.3057], device='cuda:0'), covar=tensor([0.1272, 0.1836, 0.1065, 0.1214], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0710, 0.0822, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 07:43:49,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.864e+02 1.686e+03 2.321e+03 3.006e+03 1.690e+04, threshold=4.643e+03, percent-clipped=20.0
+2023-03-04 07:43:50,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:43:52,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:43:55,706 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:43:58,261 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:44:05,941 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=343461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:44:05,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:44:23,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=343478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:44:23,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6495, 1.6861, 1.5898, 1.5406], device='cuda:0'), covar=tensor([0.1186, 0.1725, 0.1680, 0.1643], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0729, 0.0655, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 07:44:23,719 INFO [train.py:968] (0/2) Epoch 8, batch 24700, giga_loss[loss=0.3315, simple_loss=0.3994, pruned_loss=0.1318, over 28760.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3892, pruned_loss=0.1336, over 5647683.03 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3765, pruned_loss=0.1253, over 5708607.86 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3908, pruned_loss=0.1342, over 5653453.92 frames. ], batch size: 262, lr: 4.03e-03, grad_scale: 1.0
+2023-03-04 07:44:26,591 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=343483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:45:14,138 INFO [train.py:968] (0/2) Epoch 8, batch 24750, giga_loss[loss=0.3112, simple_loss=0.3728, pruned_loss=0.1248, over 28635.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.389, pruned_loss=0.1343, over 5639482.68 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3765, pruned_loss=0.1253, over 5710660.42 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3903, pruned_loss=0.1348, over 5641879.20 frames. ], batch size: 92, lr: 4.03e-03, grad_scale: 1.0
+2023-03-04 07:45:28,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.568e+03 2.107e+03 3.049e+03 7.981e+03, threshold=4.215e+03, percent-clipped=8.0
+2023-03-04 07:45:59,964 INFO [train.py:968] (0/2) Epoch 8, batch 24800, giga_loss[loss=0.317, simple_loss=0.3733, pruned_loss=0.1304, over 28923.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3873, pruned_loss=0.1344, over 5646385.02 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3766, pruned_loss=0.1255, over 5701276.50 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3885, pruned_loss=0.1349, over 5654801.30 frames. ], batch size: 136, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:46:01,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343581.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:46:20,639 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:46:24,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:46:44,463 INFO [train.py:968] (0/2) Epoch 8, batch 24850, giga_loss[loss=0.3803, simple_loss=0.4235, pruned_loss=0.1685, over 28828.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3871, pruned_loss=0.1345, over 5654186.48 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.377, pruned_loss=0.1257, over 5704829.28 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3879, pruned_loss=0.1348, over 5656612.26 frames. ], batch size: 199, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:46:50,454 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=343636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:46:58,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.469e+02 1.519e+03 1.916e+03 2.714e+03 5.776e+03, threshold=3.831e+03, percent-clipped=7.0
+2023-03-04 07:47:24,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5487, 2.0392, 1.5138, 1.2628], device='cuda:0'), covar=tensor([0.2006, 0.1403, 0.1549, 0.1742], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1473, 0.1435, 0.1542], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 07:47:28,551 INFO [train.py:968] (0/2) Epoch 8, batch 24900, giga_loss[loss=0.326, simple_loss=0.4032, pruned_loss=0.1244, over 29052.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3865, pruned_loss=0.1327, over 5672463.81 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3771, pruned_loss=0.1259, over 5707371.41 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3872, pruned_loss=0.1329, over 5671596.19 frames. ], batch size: 155, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:47:49,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3177, 1.4453, 1.0876, 1.1626], device='cuda:0'), covar=tensor([0.1268, 0.1165, 0.1006, 0.1111], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1475, 0.1439, 0.1543], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 07:48:13,087 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343724.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:48:15,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343727.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:48:17,365 INFO [train.py:968] (0/2) Epoch 8, batch 24950, giga_loss[loss=0.2777, simple_loss=0.3496, pruned_loss=0.1029, over 28470.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3861, pruned_loss=0.1324, over 5662237.44 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3777, pruned_loss=0.1264, over 5709570.75 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3862, pruned_loss=0.1323, over 5658687.76 frames. ], batch size: 65, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:48:32,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.723e+02 1.568e+03 2.130e+03 2.942e+03 7.510e+03, threshold=4.261e+03, percent-clipped=13.0
+2023-03-04 07:48:40,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=343756.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:49:01,521 INFO [train.py:968] (0/2) Epoch 8, batch 25000, giga_loss[loss=0.2928, simple_loss=0.3589, pruned_loss=0.1134, over 28962.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3852, pruned_loss=0.1313, over 5674897.82 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3776, pruned_loss=0.1265, over 5712529.60 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3857, pruned_loss=0.1312, over 5668036.28 frames. ], batch size: 106, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:49:52,328 INFO [train.py:968] (0/2) Epoch 8, batch 25050, giga_loss[loss=0.2851, simple_loss=0.3558, pruned_loss=0.1072, over 29189.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3842, pruned_loss=0.131, over 5679397.25 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3777, pruned_loss=0.1267, over 5709902.06 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3846, pruned_loss=0.1308, over 5675630.04 frames. ], batch size: 129, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:49:59,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:49:59,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3859, 2.0238, 1.4841, 0.6330], device='cuda:0'), covar=tensor([0.2631, 0.1507, 0.2401, 0.3034], device='cuda:0'), in_proj_covar=tensor([0.1467, 0.1394, 0.1432, 0.1195], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 07:50:10,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.561e+02 1.452e+03 1.926e+03 2.434e+03 8.031e+03, threshold=3.853e+03, percent-clipped=7.0
+2023-03-04 07:50:38,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7984, 2.3597, 1.8271, 2.4732], device='cuda:0'), covar=tensor([0.0482, 0.0513, 0.0793, 0.0738], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0449, 0.0495, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 07:50:41,020 INFO [train.py:968] (0/2) Epoch 8, batch 25100, giga_loss[loss=0.5018, simple_loss=0.4889, pruned_loss=0.2573, over 26642.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3833, pruned_loss=0.1314, over 5682760.05 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3781, pruned_loss=0.127, over 5715839.50 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3834, pruned_loss=0.1312, over 5672994.77 frames. ], batch size: 555, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:51:11,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3993, 1.4915, 1.3027, 1.1752], device='cuda:0'), covar=tensor([0.1447, 0.1363, 0.0947, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1475, 0.1440, 0.1542], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 07:51:27,340 INFO [train.py:968] (0/2) Epoch 8, batch 25150, giga_loss[loss=0.3579, simple_loss=0.413, pruned_loss=0.1513, over 28991.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3821, pruned_loss=0.1311, over 5685806.14 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3783, pruned_loss=0.1273, over 5708304.04 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3821, pruned_loss=0.1307, over 5683736.68 frames. ], batch size: 155, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:51:41,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.073e+02 1.636e+03 2.230e+03 3.220e+03 6.073e+03, threshold=4.460e+03, percent-clipped=10.0
+2023-03-04 07:51:55,485 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1166, 1.2180, 3.6864, 2.9899], device='cuda:0'), covar=tensor([0.1573, 0.2389, 0.0477, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.0613, 0.0566, 0.0819, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 07:52:10,863 INFO [train.py:968] (0/2) Epoch 8, batch 25200, giga_loss[loss=0.3155, simple_loss=0.376, pruned_loss=0.1275, over 28979.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3801, pruned_loss=0.13, over 5696661.35 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3779, pruned_loss=0.1271, over 5712976.19 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3806, pruned_loss=0.13, over 5690099.48 frames. ], batch size: 213, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:52:11,314 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:52:14,913 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:52:22,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5276, 1.9557, 1.8933, 1.4001], device='cuda:0'), covar=tensor([0.1418, 0.2106, 0.1193, 0.1439], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0713, 0.0825, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 07:52:22,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6451, 1.9807, 1.8594, 1.4624], device='cuda:0'), covar=tensor([0.1137, 0.1809, 0.1004, 0.1218], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0713, 0.0825, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 07:52:31,032 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-344000.pt
+2023-03-04 07:52:36,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6914, 2.3028, 2.0725, 1.5388], device='cuda:0'), covar=tensor([0.1584, 0.1945, 0.1290, 0.1511], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0712, 0.0825, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 07:52:38,723 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=344011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:52:53,900 INFO [train.py:968] (0/2) Epoch 8, batch 25250, giga_loss[loss=0.4233, simple_loss=0.4423, pruned_loss=0.2021, over 26756.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3782, pruned_loss=0.1294, over 5693351.72 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.378, pruned_loss=0.1272, over 5717259.95 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3785, pruned_loss=0.1293, over 5683289.06 frames. ], batch size: 555, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:52:57,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 07:53:10,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.779e+02 1.640e+03 2.448e+03 3.740e+03 9.748e+03, threshold=4.897e+03, percent-clipped=19.0
+2023-03-04 07:53:44,152 INFO [train.py:968] (0/2) Epoch 8, batch 25300, giga_loss[loss=0.3155, simple_loss=0.3796, pruned_loss=0.1257, over 28876.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3785, pruned_loss=0.1301, over 5690812.18 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.378, pruned_loss=0.1272, over 5719152.69 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3787, pruned_loss=0.1301, over 5680965.82 frames. ], batch size: 145, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:54:01,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2740, 2.9662, 1.9886, 1.6179], device='cuda:0'), covar=tensor([0.1707, 0.0871, 0.1293, 0.1628], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1475, 0.1442, 0.1548], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 07:54:06,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4301, 2.0518, 1.4881, 0.5827], device='cuda:0'), covar=tensor([0.3141, 0.1575, 0.2240, 0.3733], device='cuda:0'), in_proj_covar=tensor([0.1480, 0.1408, 0.1448, 0.1209], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 07:54:37,349 INFO [train.py:968] (0/2) Epoch 8, batch 25350, giga_loss[loss=0.2923, simple_loss=0.3695, pruned_loss=0.1076, over 29064.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3791, pruned_loss=0.1297, over 5689296.32 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3779, pruned_loss=0.1271, over 5720071.93 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3794, pruned_loss=0.1298, over 5680660.38 frames. ], batch size: 128, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:54:51,492 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.128e+02 1.455e+03 1.843e+03 2.315e+03 7.974e+03, threshold=3.686e+03, percent-clipped=1.0
+2023-03-04 07:55:20,973 INFO [train.py:968] (0/2) Epoch 8, batch 25400, giga_loss[loss=0.3285, simple_loss=0.3899, pruned_loss=0.1335, over 28755.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3802, pruned_loss=0.1299, over 5687754.77 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3781, pruned_loss=0.1274, over 5721090.20 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3802, pruned_loss=0.1298, over 5679777.29 frames. ], batch size: 284, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:55:21,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0826, 1.3449, 1.4152, 1.1879], device='cuda:0'), covar=tensor([0.1066, 0.0846, 0.1478, 0.1061], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0731, 0.0656, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 07:55:28,871 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-04 07:56:05,549 INFO [train.py:968] (0/2) Epoch 8, batch 25450, libri_loss[loss=0.2816, simple_loss=0.3463, pruned_loss=0.1085, over 29554.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3799, pruned_loss=0.1291, over 5683642.51 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3785, pruned_loss=0.1279, over 5717165.83 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3795, pruned_loss=0.1286, over 5680563.10 frames. ], batch size: 79, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:56:21,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.795e+02 1.485e+03 1.942e+03 2.867e+03 9.403e+03, threshold=3.884e+03, percent-clipped=11.0
+2023-03-04 07:56:28,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9816, 1.8180, 1.3898, 1.4833], device='cuda:0'), covar=tensor([0.0704, 0.0666, 0.0959, 0.1035], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0450, 0.0495, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 07:56:49,774 INFO [train.py:968] (0/2) Epoch 8, batch 25500, giga_loss[loss=0.2806, simple_loss=0.3575, pruned_loss=0.1018, over 28979.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3806, pruned_loss=0.1298, over 5687626.44 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3785, pruned_loss=0.1279, over 5720656.98 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3804, pruned_loss=0.1294, over 5681102.60 frames. ], batch size: 164, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:56:51,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=344281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:57:36,890 INFO [train.py:968] (0/2) Epoch 8, batch 25550, giga_loss[loss=0.3813, simple_loss=0.4193, pruned_loss=0.1717, over 28835.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3842, pruned_loss=0.1332, over 5687673.11 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3786, pruned_loss=0.128, over 5721896.64 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3841, pruned_loss=0.1329, over 5679990.15 frames. ], batch size: 285, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:57:53,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.708e+03 2.291e+03 2.950e+03 1.125e+04, threshold=4.582e+03, percent-clipped=12.0
+2023-03-04 07:58:29,237 INFO [train.py:968] (0/2) Epoch 8, batch 25600, giga_loss[loss=0.3712, simple_loss=0.4068, pruned_loss=0.1678, over 27608.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3853, pruned_loss=0.1355, over 5682115.69 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3784, pruned_loss=0.128, over 5723749.90 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3854, pruned_loss=0.1353, over 5674182.51 frames. ], batch size: 472, lr: 4.03e-03, grad_scale: 8.0
+2023-03-04 07:58:34,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4318, 1.6925, 1.3103, 1.5085], device='cuda:0'), covar=tensor([0.0730, 0.0295, 0.0318, 0.0776], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0116, 0.0119, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0051, 0.0046, 0.0078], device='cuda:0')
+2023-03-04 07:59:13,663 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-04 07:59:18,843 INFO [train.py:968] (0/2) Epoch 8, batch 25650, giga_loss[loss=0.3606, simple_loss=0.4028, pruned_loss=0.1592, over 28977.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3861, pruned_loss=0.1376, over 5680613.58 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3782, pruned_loss=0.1279, over 5725819.62 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3866, pruned_loss=0.1378, over 5671871.21 frames. ], batch size: 227, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:59:39,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.828e+03 2.280e+03 3.351e+03 1.251e+04, threshold=4.560e+03, percent-clipped=11.0
+2023-03-04 07:59:53,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3351, 1.0713, 4.7903, 3.6295], device='cuda:0'), covar=tensor([0.1667, 0.2700, 0.0328, 0.0682], device='cuda:0'), in_proj_covar=tensor([0.0611, 0.0565, 0.0815, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 08:00:04,922 INFO [train.py:968] (0/2) Epoch 8, batch 25700, giga_loss[loss=0.3309, simple_loss=0.3833, pruned_loss=0.1393, over 28380.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3864, pruned_loss=0.1376, over 5691651.18 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3785, pruned_loss=0.1281, over 5728741.49 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3867, pruned_loss=0.1378, over 5681100.56 frames. ], batch size: 71, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:00:43,488 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7715, 1.6428, 1.2557, 1.3653], device='cuda:0'), covar=tensor([0.0641, 0.0591, 0.0966, 0.0957], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0450, 0.0495, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 08:00:53,904 INFO [train.py:968] (0/2) Epoch 8, batch 25750, giga_loss[loss=0.2983, simple_loss=0.3654, pruned_loss=0.1156, over 28652.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3847, pruned_loss=0.1368, over 5666516.62 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3787, pruned_loss=0.1283, over 5720241.58 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3848, pruned_loss=0.1368, over 5666072.11 frames. ], batch size: 262, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:01:10,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.655e+03 2.049e+03 2.899e+03 7.842e+03, threshold=4.097e+03, percent-clipped=4.0
+2023-03-04 08:01:25,079 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=344565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:01:36,822 INFO [train.py:968] (0/2) Epoch 8, batch 25800, libri_loss[loss=0.3152, simple_loss=0.3696, pruned_loss=0.1304, over 29499.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.384, pruned_loss=0.1344, over 5678048.06 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.379, pruned_loss=0.1285, over 5722768.03 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3839, pruned_loss=0.1343, over 5674729.92 frames. ], batch size: 70, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:02:25,936 INFO [train.py:968] (0/2) Epoch 8, batch 25850, giga_loss[loss=0.3003, simple_loss=0.3698, pruned_loss=0.1154, over 28670.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3821, pruned_loss=0.1325, over 5666871.66 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3796, pruned_loss=0.1288, over 5722904.86 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3815, pruned_loss=0.1322, over 5663164.65 frames. ], batch size: 242, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:02:41,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.232e+02 1.662e+03 2.143e+03 3.170e+03 1.165e+04, threshold=4.286e+03, percent-clipped=14.0
+2023-03-04 08:02:49,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=344656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:02:54,455 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-04 08:03:12,697 INFO [train.py:968] (0/2) Epoch 8, batch 25900, giga_loss[loss=0.2933, simple_loss=0.3607, pruned_loss=0.113, over 28925.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3805, pruned_loss=0.1318, over 5669678.08 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3801, pruned_loss=0.1292, over 5724847.52 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3796, pruned_loss=0.1313, over 5663893.07 frames. ], batch size: 174, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:03:52,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 08:03:55,632 INFO [train.py:968] (0/2) Epoch 8, batch 25950, giga_loss[loss=0.248, simple_loss=0.3221, pruned_loss=0.08694, over 28226.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3803, pruned_loss=0.1324, over 5663514.98 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3803, pruned_loss=0.1294, over 5719382.81 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3794, pruned_loss=0.1319, over 5663039.74 frames. ], batch size: 77, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 08:04:09,003 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=344743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:04:12,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.504e+02 1.646e+03 2.051e+03 2.763e+03 7.431e+03, threshold=4.101e+03, percent-clipped=7.0
+2023-03-04 08:04:45,384 INFO [train.py:968] (0/2) Epoch 8, batch 26000, giga_loss[loss=0.3993, simple_loss=0.419, pruned_loss=0.1898, over 23525.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3819, pruned_loss=0.1346, over 5647847.91 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3799, pruned_loss=0.1294, over 5717411.19 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3816, pruned_loss=0.1343, over 5647719.30 frames. ], batch size: 705, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:05:02,348 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=344799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:05:04,277 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=344802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:05:20,937 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 08:05:27,905 INFO [train.py:968] (0/2) Epoch 8, batch 26050, giga_loss[loss=0.3606, simple_loss=0.4135, pruned_loss=0.1539, over 28975.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.384, pruned_loss=0.1351, over 5653821.27 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3802, pruned_loss=0.1298, over 5713609.77 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3836, pruned_loss=0.1347, over 5654171.21 frames. ], batch size: 213, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:05:29,732 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=344831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:05:42,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.992e+02 1.385e+03 1.895e+03 2.399e+03 9.634e+03, threshold=3.791e+03, percent-clipped=9.0
+2023-03-04 08:06:12,123 INFO [train.py:968] (0/2) Epoch 8, batch 26100, giga_loss[loss=0.3656, simple_loss=0.4277, pruned_loss=0.1517, over 28505.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3882, pruned_loss=0.1354, over 5663416.54 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3802, pruned_loss=0.1299, over 5717419.58 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.388, pruned_loss=0.1351, over 5658724.56 frames. ], batch size: 336, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:06:12,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3957, 1.6821, 1.4679, 1.6299], device='cuda:0'), covar=tensor([0.1700, 0.1627, 0.1632, 0.1472], device='cuda:0'), in_proj_covar=tensor([0.1218, 0.0913, 0.1066, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 08:06:45,556 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 08:06:57,845 INFO [train.py:968] (0/2) Epoch 8, batch 26150, giga_loss[loss=0.3283, simple_loss=0.3966, pruned_loss=0.13, over 28960.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3903, pruned_loss=0.136, over 5657562.74 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3808, pruned_loss=0.1306, over 5706437.00 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3899, pruned_loss=0.1354, over 5661302.29 frames. ], batch size: 145, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 08:07:09,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=344940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:07:16,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.759e+02 1.420e+03 1.824e+03 2.446e+03 6.737e+03, threshold=3.647e+03, percent-clipped=12.0
+2023-03-04 08:07:25,934 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=344960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:07:34,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-04 08:07:40,643 INFO [train.py:968] (0/2) Epoch 8, batch 26200, giga_loss[loss=0.3483, simple_loss=0.406, pruned_loss=0.1453, over 28837.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.392, pruned_loss=0.1378, over 5665551.44 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3812, pruned_loss=0.1312, over 5716376.28 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3919, pruned_loss=0.1371, over 5656319.36 frames. ], batch size: 199, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 08:08:26,822 INFO [train.py:968] (0/2) Epoch 8, batch 26250, giga_loss[loss=0.3492, simple_loss=0.4017, pruned_loss=0.1483, over 28862.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3924, pruned_loss=0.1386, over 5662248.82 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3812, pruned_loss=0.1312, over 5719155.54 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3926, pruned_loss=0.1382, over 5651132.42 frames. ], batch size: 119, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:08:44,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.394e+02 1.490e+03 1.968e+03 3.067e+03 1.061e+04, threshold=3.937e+03, percent-clipped=16.0
+2023-03-04 08:09:16,146 INFO [train.py:968] (0/2) Epoch 8, batch 26300, giga_loss[loss=0.3002, simple_loss=0.372, pruned_loss=0.1142, over 29066.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3912, pruned_loss=0.1384, over 5655222.93 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3811, pruned_loss=0.1311, over 5721132.26 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3916, pruned_loss=0.1383, over 5644154.82 frames. ], batch size: 136, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:09:20,031 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=345083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:09:22,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=345086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:09:51,747 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=345115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:09:56,084 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:09:56,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=345118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:09:59,313 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4730, 1.6838, 1.7712, 1.3660], device='cuda:0'), covar=tensor([0.1528, 0.2119, 0.1205, 0.1368], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0716, 0.0828, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 08:10:03,131 INFO [train.py:968] (0/2) Epoch 8, batch 26350, giga_loss[loss=0.2829, simple_loss=0.3459, pruned_loss=0.1099, over 28609.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3898, pruned_loss=0.138, over 5643223.16 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3813, pruned_loss=0.1313, over 5712807.19 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3902, pruned_loss=0.1379, over 5640400.89 frames. ], batch size: 78, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:10:19,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.982e+02 1.428e+03 2.101e+03 2.759e+03 5.066e+03, threshold=4.203e+03, percent-clipped=6.0
+2023-03-04 08:10:45,352 INFO [train.py:968] (0/2) Epoch 8, batch 26400, giga_loss[loss=0.3474, simple_loss=0.3947, pruned_loss=0.15, over 28837.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3871, pruned_loss=0.1367, over 5654874.98 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3806, pruned_loss=0.1309, over 5718310.66 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3882, pruned_loss=0.1372, over 5645553.91 frames. ], batch size: 112, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:11:33,247 INFO [train.py:968] (0/2) Epoch 8, batch 26450, giga_loss[loss=0.2799, simple_loss=0.3463, pruned_loss=0.1067, over 28577.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3854, pruned_loss=0.136, over 5656759.71 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3802, pruned_loss=0.1305, over 5721593.39 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3869, pruned_loss=0.1369, over 5643887.38 frames. ], batch size: 85, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:11:51,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.543e+02 1.746e+03 2.315e+03 3.231e+03 7.925e+03, threshold=4.630e+03, percent-clipped=15.0
+2023-03-04 08:12:02,970 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=345261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:12:04,967 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=345264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:12:17,818 INFO [train.py:968] (0/2) Epoch 8, batch 26500, giga_loss[loss=0.3564, simple_loss=0.4067, pruned_loss=0.153, over 28954.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3861, pruned_loss=0.137, over 5657115.70 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3802, pruned_loss=0.1306, over 5726913.78 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3874, pruned_loss=0.1378, over 5640030.49 frames. ], batch size: 213, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:12:27,376 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2818, 1.3108, 0.9988, 1.1389], device='cuda:0'), covar=tensor([0.1144, 0.1111, 0.0953, 0.1050], device='cuda:0'), in_proj_covar=tensor([0.1607, 0.1453, 0.1432, 0.1530], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 08:12:29,998 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=345293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:12:49,237 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:12:55,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8728, 3.6899, 3.5088, 1.8716], device='cuda:0'), covar=tensor([0.0636, 0.0800, 0.0805, 0.1901], device='cuda:0'), in_proj_covar=tensor([0.0984, 0.0933, 0.0822, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 08:13:00,727 INFO [train.py:968] (0/2) Epoch 8, batch 26550, giga_loss[loss=0.3135, simple_loss=0.3752, pruned_loss=0.1259, over 28950.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3855, pruned_loss=0.1367, over 5663994.59 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3796, pruned_loss=0.1302, over 5729176.28 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3872, pruned_loss=0.1378, over 5647168.04 frames. ], batch size: 145, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:13:07,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=345335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:13:16,681 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.10 vs. limit=5.0
+2023-03-04 08:13:20,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.198e+02 1.536e+03 1.896e+03 2.675e+03 7.634e+03, threshold=3.793e+03, percent-clipped=5.0
+2023-03-04 08:13:37,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3442, 1.5805, 1.3267, 1.2703], device='cuda:0'), covar=tensor([0.1965, 0.1950, 0.2002, 0.1753], device='cuda:0'), in_proj_covar=tensor([0.1223, 0.0921, 0.1075, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 08:13:45,056 INFO [train.py:968] (0/2) Epoch 8, batch 26600, giga_loss[loss=0.3866, simple_loss=0.4197, pruned_loss=0.1767, over 27673.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3844, pruned_loss=0.1364, over 5677172.28 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3804, pruned_loss=0.1308, over 5732669.59 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3851, pruned_loss=0.1368, over 5659203.99 frames. ], batch size: 472, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:14:32,230 INFO [train.py:968] (0/2) Epoch 8, batch 26650, giga_loss[loss=0.3654, simple_loss=0.4042, pruned_loss=0.1633, over 28803.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.384, pruned_loss=0.1359, over 5665346.19 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3804, pruned_loss=0.1309, over 5716470.92 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3846, pruned_loss=0.1363, over 5663810.05 frames. ], batch size: 99, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:14:49,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.550e+02 1.536e+03 1.927e+03 2.376e+03 1.118e+04, threshold=3.854e+03, percent-clipped=9.0
+2023-03-04 08:15:02,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3866, 1.7708, 1.5651, 1.5315], device='cuda:0'), covar=tensor([0.0771, 0.0295, 0.0297, 0.0815], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0119, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0078], device='cuda:0')
+2023-03-04 08:15:16,936 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=345478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:15:17,455 INFO [train.py:968] (0/2) Epoch 8, batch 26700, giga_loss[loss=0.3256, simple_loss=0.3883, pruned_loss=0.1315, over 28723.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3856, pruned_loss=0.1361, over 5663636.75 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3807, pruned_loss=0.131, over 5717349.70 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.386, pruned_loss=0.1364, over 5659993.84 frames. ], batch size: 119, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:15:20,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=345481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:15:30,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=345493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:15:46,513 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=345510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:16:05,304 INFO [train.py:968] (0/2) Epoch 8, batch 26750, giga_loss[loss=0.316, simple_loss=0.3774, pruned_loss=0.1273, over 28959.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3886, pruned_loss=0.1381, over 5665428.07 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3804, pruned_loss=0.1308, over 5718333.95 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3892, pruned_loss=0.1385, over 5661510.57 frames. ], batch size: 213, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:16:26,894 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.418e+02 1.594e+03 2.051e+03 2.894e+03 6.332e+03, threshold=4.102e+03, percent-clipped=9.0
+2023-03-04 08:16:36,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.22 vs. limit=5.0
+2023-03-04 08:16:55,433 INFO [train.py:968] (0/2) Epoch 8, batch 26800, giga_loss[loss=0.2999, simple_loss=0.3841, pruned_loss=0.1079, over 29004.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3881, pruned_loss=0.1385, over 5659157.94 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3805, pruned_loss=0.1309, over 5717370.28 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3885, pruned_loss=0.1389, over 5656661.79 frames. ], batch size: 106, lr: 4.02e-03, grad_scale: 8.0
+2023-03-04 08:17:34,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:17:41,295 INFO [train.py:968] (0/2) Epoch 8, batch 26850, giga_loss[loss=0.354, simple_loss=0.3924, pruned_loss=0.1578, over 26706.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3887, pruned_loss=0.1356, over 5674061.95 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3802, pruned_loss=0.1306, over 5720414.14 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3894, pruned_loss=0.1361, over 5668788.31 frames. ], batch size: 555, lr: 4.02e-03, grad_scale: 8.0
+2023-03-04 08:17:49,337 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=345636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:17:51,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=345639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:18:00,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.165e+02 1.374e+03 1.618e+03 2.160e+03 7.957e+03, threshold=3.235e+03, percent-clipped=2.0
+2023-03-04 08:18:15,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1531, 1.1220, 4.0699, 3.1920], device='cuda:0'), covar=tensor([0.1647, 0.2541, 0.0424, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0612, 0.0563, 0.0813, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 08:18:17,897 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=345668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:18:28,399 INFO [train.py:968] (0/2) Epoch 8, batch 26900, giga_loss[loss=0.3038, simple_loss=0.3911, pruned_loss=0.1082, over 28919.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3895, pruned_loss=0.1347, over 5663113.45 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3799, pruned_loss=0.1304, over 5715477.46 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3905, pruned_loss=0.1354, over 5662252.66 frames. ], batch size: 174, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:18:37,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=345689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:19:09,860 INFO [train.py:968] (0/2) Epoch 8, batch 26950, giga_loss[loss=0.3753, simple_loss=0.4202, pruned_loss=0.1652, over 28471.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3916, pruned_loss=0.1354, over 5673311.55 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3805, pruned_loss=0.1309, over 5720798.63 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3922, pruned_loss=0.1357, over 5666688.02 frames. ], batch size: 336, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:19:28,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.690e+02 1.496e+03 2.227e+03 3.410e+03 1.064e+04, threshold=4.454e+03, percent-clipped=25.0
+2023-03-04 08:19:56,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5074, 1.8571, 1.7951, 1.4135], device='cuda:0'), covar=tensor([0.1486, 0.1836, 0.1162, 0.1342], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0711, 0.0823, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 08:19:58,320 INFO [train.py:968] (0/2) Epoch 8, batch 27000, giga_loss[loss=0.3376, simple_loss=0.3946, pruned_loss=0.1404, over 28874.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3955, pruned_loss=0.1398, over 5671377.54 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3805, pruned_loss=0.1309, over 5722674.70 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3961, pruned_loss=0.1401, over 5664000.71 frames. ], batch size: 119, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:19:58,324 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 08:20:07,679 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2229, simple_loss=0.3279, pruned_loss=0.05895, over 944034.00 frames. 
+2023-03-04 08:20:07,680 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 08:20:49,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345824.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:20:52,537 INFO [train.py:968] (0/2) Epoch 8, batch 27050, giga_loss[loss=0.4075, simple_loss=0.4396, pruned_loss=0.1877, over 28603.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3965, pruned_loss=0.141, over 5670836.11 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3807, pruned_loss=0.131, over 5713824.08 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3971, pruned_loss=0.1414, over 5671968.43 frames. ], batch size: 336, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:20:56,679 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=345832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:21:00,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=345835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:21:09,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2009, 1.4658, 1.1650, 1.3004], device='cuda:0'), covar=tensor([0.2019, 0.1978, 0.2117, 0.1763], device='cuda:0'), in_proj_covar=tensor([0.1220, 0.0917, 0.1072, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 08:21:14,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.803e+03 2.503e+03 3.335e+03 7.728e+03, threshold=5.006e+03, percent-clipped=15.0
+2023-03-04 08:21:28,012 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=345864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:21:41,301 INFO [train.py:968] (0/2) Epoch 8, batch 27100, libri_loss[loss=0.4002, simple_loss=0.43, pruned_loss=0.1852, over 19851.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3959, pruned_loss=0.1413, over 5662941.73 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3808, pruned_loss=0.1311, over 5707424.59 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3967, pruned_loss=0.1417, over 5669093.36 frames. ], batch size: 187, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:22:28,043 INFO [train.py:968] (0/2) Epoch 8, batch 27150, giga_loss[loss=0.3419, simple_loss=0.3969, pruned_loss=0.1434, over 28720.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3935, pruned_loss=0.1391, over 5668817.63 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3802, pruned_loss=0.1309, over 5708017.78 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.395, pruned_loss=0.1399, over 5672190.30 frames. ], batch size: 78, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:22:35,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345936.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:22:46,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.329e+02 1.417e+03 1.868e+03 3.024e+03 8.303e+03, threshold=3.736e+03, percent-clipped=5.0
+2023-03-04 08:23:11,657 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:23:12,860 INFO [train.py:968] (0/2) Epoch 8, batch 27200, giga_loss[loss=0.2985, simple_loss=0.3739, pruned_loss=0.1116, over 28892.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.393, pruned_loss=0.1373, over 5667209.39 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3806, pruned_loss=0.1314, over 5713666.77 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3943, pruned_loss=0.1377, over 5663579.81 frames. ], batch size: 99, lr: 4.02e-03, grad_scale: 8.0
+2023-03-04 08:23:31,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=345997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:23:33,479 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-346000.pt
+2023-03-04 08:23:59,226 INFO [train.py:968] (0/2) Epoch 8, batch 27250, giga_loss[loss=0.3386, simple_loss=0.396, pruned_loss=0.1406, over 28900.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.393, pruned_loss=0.1366, over 5656056.39 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3804, pruned_loss=0.1314, over 5706068.02 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3944, pruned_loss=0.1371, over 5658775.90 frames. ], batch size: 106, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:24:19,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.578e+02 1.383e+03 1.899e+03 2.529e+03 5.368e+03, threshold=3.798e+03, percent-clipped=3.0
+2023-03-04 08:24:41,871 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-04 08:24:49,759 INFO [train.py:968] (0/2) Epoch 8, batch 27300, giga_loss[loss=0.3483, simple_loss=0.4036, pruned_loss=0.1465, over 27965.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3944, pruned_loss=0.1377, over 5655247.96 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3803, pruned_loss=0.1313, over 5709019.75 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3959, pruned_loss=0.1382, over 5654183.11 frames. ], batch size: 412, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:25:36,054 INFO [train.py:968] (0/2) Epoch 8, batch 27350, giga_loss[loss=0.2967, simple_loss=0.3612, pruned_loss=0.1161, over 28663.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3935, pruned_loss=0.1377, over 5648606.09 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3809, pruned_loss=0.1317, over 5701572.18 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3944, pruned_loss=0.1378, over 5653880.08 frames. ], batch size: 242, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:25:45,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=346140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:25:48,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=346143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:25:53,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.465e+02 1.500e+03 1.862e+03 2.438e+03 6.059e+03, threshold=3.724e+03, percent-clipped=6.0
+2023-03-04 08:26:17,606 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=346172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:26:24,069 INFO [train.py:968] (0/2) Epoch 8, batch 27400, giga_loss[loss=0.2986, simple_loss=0.3604, pruned_loss=0.1184, over 28701.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3925, pruned_loss=0.1381, over 5647357.54 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.381, pruned_loss=0.132, over 5695517.10 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3933, pruned_loss=0.1381, over 5656281.67 frames. ], batch size: 99, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:26:44,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=346199.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:27:10,874 INFO [train.py:968] (0/2) Epoch 8, batch 27450, libri_loss[loss=0.3186, simple_loss=0.3778, pruned_loss=0.1297, over 29543.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3916, pruned_loss=0.1382, over 5656444.17 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3813, pruned_loss=0.1322, over 5696427.44 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3924, pruned_loss=0.1382, over 5661530.98 frames. ], batch size: 77, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:27:34,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.673e+03 2.118e+03 2.946e+03 6.205e+03, threshold=4.236e+03, percent-clipped=11.0
+2023-03-04 08:27:58,897 INFO [train.py:968] (0/2) Epoch 8, batch 27500, giga_loss[loss=0.3607, simple_loss=0.4072, pruned_loss=0.1571, over 28535.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3889, pruned_loss=0.1368, over 5665157.58 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3813, pruned_loss=0.1323, over 5701701.91 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3897, pruned_loss=0.1369, over 5663554.78 frames. ], batch size: 307, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:28:29,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=346311.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:28:46,019 INFO [train.py:968] (0/2) Epoch 8, batch 27550, giga_loss[loss=0.3849, simple_loss=0.4255, pruned_loss=0.1722, over 28849.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3871, pruned_loss=0.1362, over 5664369.23 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3808, pruned_loss=0.132, over 5703098.91 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3882, pruned_loss=0.1366, over 5661290.31 frames. ], batch size: 112, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:28:58,600 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=346342.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:29:01,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=346345.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:29:05,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.335e+02 1.565e+03 2.209e+03 3.152e+03 8.949e+03, threshold=4.419e+03, percent-clipped=13.0
+2023-03-04 08:29:07,194 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=346352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:29:27,238 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=346374.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:29:31,326 INFO [train.py:968] (0/2) Epoch 8, batch 27600, giga_loss[loss=0.3913, simple_loss=0.4195, pruned_loss=0.1816, over 26557.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3872, pruned_loss=0.1368, over 5660070.63 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3816, pruned_loss=0.1325, over 5706263.79 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3875, pruned_loss=0.1367, over 5654139.72 frames. ], batch size: 555, lr: 4.02e-03, grad_scale: 8.0
+2023-03-04 08:30:15,789 INFO [train.py:968] (0/2) Epoch 8, batch 27650, giga_loss[loss=0.3082, simple_loss=0.3751, pruned_loss=0.1206, over 28810.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3846, pruned_loss=0.1335, over 5666988.24 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.3814, pruned_loss=0.1323, over 5708378.16 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.385, pruned_loss=0.1337, over 5660184.17 frames. ], batch size: 112, lr: 4.02e-03, grad_scale: 8.0
+2023-03-04 08:30:28,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9928, 1.1318, 3.2799, 2.8178], device='cuda:0'), covar=tensor([0.1560, 0.2486, 0.0413, 0.1672], device='cuda:0'), in_proj_covar=tensor([0.0615, 0.0568, 0.0811, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 08:30:36,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.483e+02 1.625e+03 1.925e+03 2.582e+03 5.913e+03, threshold=3.851e+03, percent-clipped=4.0
+2023-03-04 08:30:39,633 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=346454.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:30:41,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=346457.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:31:01,960 INFO [train.py:968] (0/2) Epoch 8, batch 27700, giga_loss[loss=0.2891, simple_loss=0.3634, pruned_loss=0.1074, over 28623.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3816, pruned_loss=0.1306, over 5655443.84 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.382, pruned_loss=0.1327, over 5699682.49 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3814, pruned_loss=0.1303, over 5657214.61 frames. ], batch size: 242, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:31:06,841 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=346486.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:31:15,135 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=346495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:31:15,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2749, 2.9406, 1.3259, 1.4281], device='cuda:0'), covar=tensor([0.0932, 0.0363, 0.0893, 0.1310], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0500, 0.0323, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 08:31:18,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=346498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:31:47,842 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=346527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:31:49,021 INFO [train.py:968] (0/2) Epoch 8, batch 27750, giga_loss[loss=0.2801, simple_loss=0.3504, pruned_loss=0.1049, over 28846.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3798, pruned_loss=0.1292, over 5643335.76 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3819, pruned_loss=0.1328, over 5685819.91 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3797, pruned_loss=0.1288, over 5656189.01 frames. ], batch size: 199, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:32:06,168 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=346547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:32:09,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.923e+02 1.472e+03 1.840e+03 2.635e+03 6.668e+03, threshold=3.680e+03, percent-clipped=4.0
+2023-03-04 08:32:34,861 INFO [train.py:968] (0/2) Epoch 8, batch 27800, giga_loss[loss=0.3436, simple_loss=0.3879, pruned_loss=0.1496, over 28857.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3778, pruned_loss=0.1288, over 5636722.80 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3817, pruned_loss=0.1326, over 5682386.33 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3778, pruned_loss=0.1286, over 5648566.94 frames. ], batch size: 186, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:33:22,832 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=346627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:33:23,837 INFO [train.py:968] (0/2) Epoch 8, batch 27850, giga_loss[loss=0.3067, simple_loss=0.3748, pruned_loss=0.1193, over 28881.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3752, pruned_loss=0.1276, over 5646714.14 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.3814, pruned_loss=0.1323, over 5688853.02 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3753, pruned_loss=0.1275, over 5648879.76 frames. ], batch size: 227, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:33:46,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.643e+02 1.697e+03 2.215e+03 3.560e+03 1.110e+04, threshold=4.429e+03, percent-clipped=22.0
+2023-03-04 08:34:09,127 INFO [train.py:968] (0/2) Epoch 8, batch 27900, giga_loss[loss=0.2891, simple_loss=0.3534, pruned_loss=0.1124, over 28237.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3774, pruned_loss=0.1284, over 5654116.72 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3814, pruned_loss=0.1322, over 5689618.30 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3774, pruned_loss=0.1283, over 5654579.49 frames. ], batch size: 77, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:34:56,718 INFO [train.py:968] (0/2) Epoch 8, batch 27950, giga_loss[loss=0.3318, simple_loss=0.3928, pruned_loss=0.1354, over 28653.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3788, pruned_loss=0.1289, over 5647104.40 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.381, pruned_loss=0.1319, over 5693921.68 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3791, pruned_loss=0.129, over 5642805.84 frames. ], batch size: 262, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:35:18,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2948, 3.1109, 2.9434, 1.4573], device='cuda:0'), covar=tensor([0.0781, 0.0880, 0.0839, 0.2262], device='cuda:0'), in_proj_covar=tensor([0.0987, 0.0932, 0.0827, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 08:35:19,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.366e+02 1.502e+03 1.885e+03 2.594e+03 1.118e+04, threshold=3.771e+03, percent-clipped=9.0
+2023-03-04 08:35:43,189 INFO [train.py:968] (0/2) Epoch 8, batch 28000, giga_loss[loss=0.2858, simple_loss=0.3517, pruned_loss=0.11, over 28436.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3787, pruned_loss=0.1289, over 5649175.10 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3813, pruned_loss=0.1322, over 5687656.32 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3787, pruned_loss=0.1287, over 5651403.23 frames. ], batch size: 71, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:35:50,783 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7676, 1.8544, 1.5366, 1.6471], device='cuda:0'), covar=tensor([0.1504, 0.2009, 0.1779, 0.1815], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0734, 0.0659, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 08:36:29,439 INFO [train.py:968] (0/2) Epoch 8, batch 28050, giga_loss[loss=0.2855, simple_loss=0.3555, pruned_loss=0.1078, over 28947.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3795, pruned_loss=0.1304, over 5637914.59 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3818, pruned_loss=0.1327, over 5686420.94 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3789, pruned_loss=0.1298, over 5640280.89 frames. ], batch size: 136, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:36:43,849 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=346846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:36:48,303 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 08:36:48,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.007e+02 1.566e+03 1.915e+03 2.803e+03 7.266e+03, threshold=3.830e+03, percent-clipped=13.0
+2023-03-04 08:37:14,257 INFO [train.py:968] (0/2) Epoch 8, batch 28100, giga_loss[loss=0.3639, simple_loss=0.4127, pruned_loss=0.1576, over 28987.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3808, pruned_loss=0.1318, over 5630281.94 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3823, pruned_loss=0.1331, over 5681408.77 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3798, pruned_loss=0.1309, over 5635545.67 frames. ], batch size: 136, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:37:19,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4381, 1.5488, 1.2668, 1.7645], device='cuda:0'), covar=tensor([0.2324, 0.2258, 0.2347, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.1229, 0.0924, 0.1078, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 08:37:20,105 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=346886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:37:22,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3710, 3.6899, 1.4772, 1.3924], device='cuda:0'), covar=tensor([0.0932, 0.0315, 0.0871, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0495, 0.0322, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 08:37:44,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6708, 1.5991, 1.2308, 1.3121], device='cuda:0'), covar=tensor([0.0667, 0.0594, 0.0962, 0.0949], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0446, 0.0493, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 08:37:44,373 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-04 08:37:53,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=346922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:37:54,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=346924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:37:58,315 INFO [train.py:968] (0/2) Epoch 8, batch 28150, giga_loss[loss=0.2944, simple_loss=0.361, pruned_loss=0.1139, over 28620.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3829, pruned_loss=0.1327, over 5635473.21 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3829, pruned_loss=0.1335, over 5677716.72 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3816, pruned_loss=0.1316, over 5641909.42 frames. ], batch size: 85, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:38:18,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.569e+03 1.987e+03 2.580e+03 5.139e+03, threshold=3.974e+03, percent-clipped=6.0
+2023-03-04 08:38:28,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8896, 1.0348, 1.0444, 0.8480], device='cuda:0'), covar=tensor([0.1157, 0.1277, 0.0738, 0.1043], device='cuda:0'), in_proj_covar=tensor([0.1630, 0.1475, 0.1453, 0.1550], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 08:38:42,893 INFO [train.py:968] (0/2) Epoch 8, batch 28200, giga_loss[loss=0.2756, simple_loss=0.3485, pruned_loss=0.1014, over 28312.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3857, pruned_loss=0.1351, over 5636523.95 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3829, pruned_loss=0.1335, over 5672411.09 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3847, pruned_loss=0.1341, over 5645237.83 frames. ], batch size: 60, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:38:56,925 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-04 08:39:06,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:39:33,310 INFO [train.py:968] (0/2) Epoch 8, batch 28250, libri_loss[loss=0.4044, simple_loss=0.4404, pruned_loss=0.1842, over 19114.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3866, pruned_loss=0.1361, over 5632925.36 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3825, pruned_loss=0.1333, over 5666828.30 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3861, pruned_loss=0.1356, over 5645361.49 frames. ], batch size: 188, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:39:53,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.956e+02 1.594e+03 2.106e+03 2.814e+03 5.783e+03, threshold=4.212e+03, percent-clipped=10.0
+2023-03-04 08:40:06,885 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:40:09,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:40:20,947 INFO [train.py:968] (0/2) Epoch 8, batch 28300, giga_loss[loss=0.29, simple_loss=0.3596, pruned_loss=0.1102, over 28771.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3866, pruned_loss=0.1362, over 5638060.78 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3821, pruned_loss=0.133, over 5671853.46 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3867, pruned_loss=0.1361, over 5642866.31 frames. ], batch size: 92, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:40:33,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2898, 1.5706, 1.2919, 1.0844], device='cuda:0'), covar=tensor([0.1641, 0.1375, 0.1076, 0.1412], device='cuda:0'), in_proj_covar=tensor([0.1634, 0.1475, 0.1453, 0.1553], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 08:40:42,668 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:41:13,661 INFO [train.py:968] (0/2) Epoch 8, batch 28350, giga_loss[loss=0.3812, simple_loss=0.4055, pruned_loss=0.1785, over 23586.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3866, pruned_loss=0.1347, over 5642092.14 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3823, pruned_loss=0.1331, over 5674380.43 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3866, pruned_loss=0.1346, over 5643000.77 frames. ], batch size: 705, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:41:28,028 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:41:30,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:41:36,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.264e+02 1.635e+03 2.052e+03 3.226e+03 7.988e+03, threshold=4.103e+03, percent-clipped=10.0
+2023-03-04 08:42:01,046 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:42:02,207 INFO [train.py:968] (0/2) Epoch 8, batch 28400, giga_loss[loss=0.3189, simple_loss=0.3852, pruned_loss=0.1263, over 28880.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3877, pruned_loss=0.1362, over 5641717.04 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3824, pruned_loss=0.1332, over 5679210.16 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3877, pruned_loss=0.136, over 5637691.05 frames. ], batch size: 174, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:42:44,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:42:53,930 INFO [train.py:968] (0/2) Epoch 8, batch 28450, giga_loss[loss=0.4039, simple_loss=0.4261, pruned_loss=0.1908, over 26568.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3872, pruned_loss=0.1366, over 5621297.13 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.3825, pruned_loss=0.1333, over 5673463.26 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3872, pruned_loss=0.1364, over 5622274.76 frames. ], batch size: 555, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:43:24,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.648e+02 1.521e+03 1.968e+03 2.598e+03 6.311e+03, threshold=3.937e+03, percent-clipped=13.0
+2023-03-04 08:43:31,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:43:51,935 INFO [train.py:968] (0/2) Epoch 8, batch 28500, giga_loss[loss=0.3486, simple_loss=0.3772, pruned_loss=0.16, over 23472.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3854, pruned_loss=0.1362, over 5623075.76 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3823, pruned_loss=0.1333, over 5675790.14 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3856, pruned_loss=0.1361, over 5621404.24 frames. ], batch size: 705, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:44:12,932 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:44:25,206 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-04 08:44:28,439 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=347317.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:44:39,073 INFO [train.py:968] (0/2) Epoch 8, batch 28550, giga_loss[loss=0.2766, simple_loss=0.3421, pruned_loss=0.1056, over 28746.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.385, pruned_loss=0.1363, over 5618590.25 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3825, pruned_loss=0.1334, over 5659592.87 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.385, pruned_loss=0.1361, over 5631405.38 frames. ], batch size: 99, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:45:02,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.902e+02 1.455e+03 1.923e+03 2.426e+03 4.751e+03, threshold=3.846e+03, percent-clipped=5.0
+2023-03-04 08:45:10,578 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:45:13,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:45:25,928 INFO [train.py:968] (0/2) Epoch 8, batch 28600, giga_loss[loss=0.3246, simple_loss=0.3857, pruned_loss=0.1318, over 29021.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3843, pruned_loss=0.136, over 5630486.87 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.3825, pruned_loss=0.1336, over 5657599.58 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3843, pruned_loss=0.1358, over 5641856.41 frames. ], batch size: 136, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:45:39,339 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:45:47,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:45:50,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:46:07,953 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3885, 1.7010, 1.5233, 1.4912], device='cuda:0'), covar=tensor([0.0734, 0.0296, 0.0283, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0118, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0078], device='cuda:0')
+2023-03-04 08:46:08,940 INFO [train.py:968] (0/2) Epoch 8, batch 28650, libri_loss[loss=0.35, simple_loss=0.3945, pruned_loss=0.1527, over 29524.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3839, pruned_loss=0.1362, over 5638465.95 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.3827, pruned_loss=0.1337, over 5663400.41 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3838, pruned_loss=0.1359, over 5641679.25 frames. ], batch size: 84, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:46:09,792 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2603, 2.0757, 1.7900, 1.8612], device='cuda:0'), covar=tensor([0.0457, 0.0417, 0.0657, 0.0759], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0446, 0.0495, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 08:46:13,896 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:46:19,733 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:46:21,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:46:30,427 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.757e+03 2.515e+03 3.748e+03 1.061e+04, threshold=5.030e+03, percent-clipped=23.0
+2023-03-04 08:46:48,678 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:46:51,754 INFO [train.py:968] (0/2) Epoch 8, batch 28700, libri_loss[loss=0.3391, simple_loss=0.3914, pruned_loss=0.1434, over 26178.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3835, pruned_loss=0.1351, over 5647301.86 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.3827, pruned_loss=0.1337, over 5659060.72 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3834, pruned_loss=0.135, over 5652668.05 frames. ], batch size: 136, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:47:01,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4724, 1.1850, 4.9304, 3.6242], device='cuda:0'), covar=tensor([0.1681, 0.2599, 0.0332, 0.0673], device='cuda:0'), in_proj_covar=tensor([0.0623, 0.0567, 0.0817, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 08:47:22,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3135, 1.6722, 1.4456, 1.4664], device='cuda:0'), covar=tensor([0.0723, 0.0284, 0.0312, 0.0815], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0116, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0078], device='cuda:0')
+2023-03-04 08:47:35,774 INFO [train.py:968] (0/2) Epoch 8, batch 28750, giga_loss[loss=0.2868, simple_loss=0.3548, pruned_loss=0.1094, over 28835.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.385, pruned_loss=0.1363, over 5654185.38 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3823, pruned_loss=0.1333, over 5662628.97 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3854, pruned_loss=0.1366, over 5655465.03 frames. ], batch size: 119, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:47:43,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=347536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:48:01,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.690e+02 1.724e+03 2.626e+03 3.763e+03 8.031e+03, threshold=5.253e+03, percent-clipped=10.0
+2023-03-04 08:48:20,269 INFO [train.py:968] (0/2) Epoch 8, batch 28800, giga_loss[loss=0.3515, simple_loss=0.401, pruned_loss=0.1509, over 28895.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3867, pruned_loss=0.1374, over 5664539.87 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.3824, pruned_loss=0.1333, over 5670654.99 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.387, pruned_loss=0.1378, over 5657818.72 frames. ], batch size: 199, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:49:03,186 INFO [train.py:968] (0/2) Epoch 8, batch 28850, giga_loss[loss=0.3684, simple_loss=0.4201, pruned_loss=0.1584, over 28572.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3887, pruned_loss=0.1394, over 5672954.31 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.383, pruned_loss=0.1336, over 5673160.40 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3886, pruned_loss=0.1395, over 5665610.94 frames. ], batch size: 336, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:49:26,805 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.487e+03 1.931e+03 3.194e+03 7.509e+03, threshold=3.862e+03, percent-clipped=3.0
+2023-03-04 08:49:33,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8031, 1.7766, 1.2545, 1.4782], device='cuda:0'), covar=tensor([0.0705, 0.0612, 0.0971, 0.1010], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0448, 0.0497, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 08:49:46,515 INFO [train.py:968] (0/2) Epoch 8, batch 28900, giga_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 28887.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3885, pruned_loss=0.1395, over 5673875.85 frames. ], libri_tot_loss[loss=0.325, simple_loss=0.3828, pruned_loss=0.1336, over 5677981.34 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3887, pruned_loss=0.1398, over 5663439.37 frames. ], batch size: 227, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:50:00,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347692.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:50:36,606 INFO [train.py:968] (0/2) Epoch 8, batch 28950, giga_loss[loss=0.3, simple_loss=0.369, pruned_loss=0.1156, over 28481.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3881, pruned_loss=0.1386, over 5674780.07 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.3828, pruned_loss=0.1337, over 5681263.52 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3884, pruned_loss=0.1389, over 5663556.58 frames. ], batch size: 71, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:50:58,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.429e+03 2.068e+03 2.932e+03 6.221e+03, threshold=4.135e+03, percent-clipped=12.0
+2023-03-04 08:51:13,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7427, 1.8319, 1.2559, 1.5836], device='cuda:0'), covar=tensor([0.0743, 0.0641, 0.1094, 0.1015], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0449, 0.0500, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 08:51:19,406 INFO [train.py:968] (0/2) Epoch 8, batch 29000, giga_loss[loss=0.2834, simple_loss=0.3635, pruned_loss=0.1017, over 28901.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.388, pruned_loss=0.1382, over 5681849.92 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3825, pruned_loss=0.1335, over 5687697.49 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3888, pruned_loss=0.1388, over 5666772.71 frames. ], batch size: 174, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:51:34,592 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 08:51:36,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4428, 3.6440, 1.5079, 1.4470], device='cuda:0'), covar=tensor([0.0936, 0.0320, 0.0876, 0.1337], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0501, 0.0326, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 08:52:06,202 INFO [train.py:968] (0/2) Epoch 8, batch 29050, giga_loss[loss=0.3723, simple_loss=0.4238, pruned_loss=0.1604, over 28941.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3901, pruned_loss=0.1402, over 5679824.17 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.3822, pruned_loss=0.1334, over 5688942.48 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.391, pruned_loss=0.1408, over 5666718.64 frames. ], batch size: 186, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:52:12,917 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347835.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:52:15,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347838.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:52:28,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.351e+02 1.762e+03 2.332e+03 3.364e+03 8.812e+03, threshold=4.664e+03, percent-clipped=17.0
+2023-03-04 08:52:33,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8877, 3.1389, 2.0575, 0.8927], device='cuda:0'), covar=tensor([0.4937, 0.1892, 0.2513, 0.4685], device='cuda:0'), in_proj_covar=tensor([0.1494, 0.1422, 0.1457, 0.1216], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 08:52:39,557 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347867.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:52:49,130 INFO [train.py:968] (0/2) Epoch 8, batch 29100, giga_loss[loss=0.3435, simple_loss=0.3914, pruned_loss=0.1478, over 28977.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3909, pruned_loss=0.1409, over 5662087.99 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3828, pruned_loss=0.1338, over 5675906.18 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3913, pruned_loss=0.1412, over 5662112.36 frames. ], batch size: 213, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:53:18,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:53:33,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9999, 1.8054, 1.4711, 1.5062], device='cuda:0'), covar=tensor([0.0672, 0.0684, 0.0931, 0.1024], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0450, 0.0501, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 08:53:33,793 INFO [train.py:968] (0/2) Epoch 8, batch 29150, giga_loss[loss=0.2935, simple_loss=0.3665, pruned_loss=0.1103, over 28747.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3898, pruned_loss=0.1396, over 5660576.50 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3831, pruned_loss=0.1338, over 5680184.40 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.39, pruned_loss=0.14, over 5656759.40 frames. ], batch size: 119, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:53:55,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6547, 1.7213, 1.3792, 2.1230], device='cuda:0'), covar=tensor([0.2172, 0.2247, 0.2384, 0.2055], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.0925, 0.1079, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 08:54:02,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.564e+02 1.656e+03 2.398e+03 3.428e+03 1.054e+04, threshold=4.796e+03, percent-clipped=12.0
+2023-03-04 08:54:27,305 INFO [train.py:968] (0/2) Epoch 8, batch 29200, giga_loss[loss=0.334, simple_loss=0.3902, pruned_loss=0.1389, over 27929.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3901, pruned_loss=0.139, over 5646654.87 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3832, pruned_loss=0.134, over 5682182.24 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3902, pruned_loss=0.1392, over 5641499.12 frames. ], batch size: 412, lr: 4.01e-03, grad_scale: 8.0
+2023-03-04 08:54:48,789 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-348000.pt
+2023-03-04 08:55:16,323 INFO [train.py:968] (0/2) Epoch 8, batch 29250, giga_loss[loss=0.3185, simple_loss=0.3755, pruned_loss=0.1307, over 29005.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3898, pruned_loss=0.1383, over 5648003.54 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3827, pruned_loss=0.1338, over 5683620.34 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3905, pruned_loss=0.1387, over 5641771.81 frames. ], batch size: 106, lr: 4.01e-03, grad_scale: 8.0
+2023-03-04 08:55:20,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8652, 2.2108, 2.2649, 1.7346], device='cuda:0'), covar=tensor([0.1661, 0.1875, 0.1193, 0.1398], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0721, 0.0833, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 08:55:39,247 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=348054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:55:40,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.188e+02 1.392e+03 1.768e+03 2.191e+03 4.145e+03, threshold=3.535e+03, percent-clipped=0.0
+2023-03-04 08:55:41,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=348057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:56:03,176 INFO [train.py:968] (0/2) Epoch 8, batch 29300, giga_loss[loss=0.3031, simple_loss=0.3693, pruned_loss=0.1184, over 28555.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3869, pruned_loss=0.1355, over 5656301.58 frames. ], libri_tot_loss[loss=0.325, simple_loss=0.3826, pruned_loss=0.1337, over 5684603.11 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3876, pruned_loss=0.1359, over 5650234.43 frames. ], batch size: 71, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:56:09,291 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=348086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:56:43,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-04 08:56:47,612 INFO [train.py:968] (0/2) Epoch 8, batch 29350, giga_loss[loss=0.3156, simple_loss=0.3768, pruned_loss=0.1272, over 28654.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3869, pruned_loss=0.1361, over 5660855.33 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3824, pruned_loss=0.1336, over 5689122.17 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3877, pruned_loss=0.1366, over 5651794.86 frames. ], batch size: 242, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:57:00,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4982, 1.8226, 1.8268, 1.3949], device='cuda:0'), covar=tensor([0.1544, 0.2079, 0.1185, 0.1401], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0718, 0.0830, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 08:57:12,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.756e+02 1.500e+03 1.943e+03 2.696e+03 7.692e+03, threshold=3.887e+03, percent-clipped=15.0
+2023-03-04 08:57:24,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9013, 1.0345, 3.6117, 3.0849], device='cuda:0'), covar=tensor([0.1749, 0.2593, 0.0439, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0621, 0.0568, 0.0817, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 08:57:34,103 INFO [train.py:968] (0/2) Epoch 8, batch 29400, giga_loss[loss=0.3575, simple_loss=0.4097, pruned_loss=0.1526, over 28695.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3883, pruned_loss=0.1371, over 5651455.15 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3826, pruned_loss=0.1336, over 5685241.56 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3888, pruned_loss=0.1375, over 5646924.98 frames. ], batch size: 85, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:58:19,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3969, 3.0521, 1.4661, 1.3868], device='cuda:0'), covar=tensor([0.0899, 0.0331, 0.0829, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0506, 0.0326, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 08:58:22,139 INFO [train.py:968] (0/2) Epoch 8, batch 29450, giga_loss[loss=0.3803, simple_loss=0.4143, pruned_loss=0.1731, over 27544.00 frames. ], tot_loss[loss=0.331, simple_loss=0.388, pruned_loss=0.137, over 5657202.42 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.3823, pruned_loss=0.1336, over 5682840.46 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3889, pruned_loss=0.1374, over 5654350.08 frames. ], batch size: 472, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:58:30,306 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 08:58:46,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.719e+02 1.580e+03 2.175e+03 3.215e+03 1.441e+04, threshold=4.349e+03, percent-clipped=15.0
+2023-03-04 08:59:06,063 INFO [train.py:968] (0/2) Epoch 8, batch 29500, giga_loss[loss=0.3968, simple_loss=0.4078, pruned_loss=0.1929, over 23600.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3862, pruned_loss=0.1366, over 5657166.79 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3815, pruned_loss=0.1327, over 5688193.03 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3881, pruned_loss=0.138, over 5648499.73 frames. ], batch size: 705, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:59:47,696 INFO [train.py:968] (0/2) Epoch 8, batch 29550, giga_loss[loss=0.328, simple_loss=0.3934, pruned_loss=0.1312, over 28937.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3863, pruned_loss=0.1365, over 5673679.30 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3815, pruned_loss=0.1327, over 5691953.76 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.388, pruned_loss=0.1379, over 5662230.93 frames. ], batch size: 174, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 09:00:08,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8843, 1.8846, 1.8916, 1.7436], device='cuda:0'), covar=tensor([0.1013, 0.1455, 0.1281, 0.1212], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0737, 0.0657, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 09:00:14,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.406e+02 1.493e+03 1.948e+03 2.297e+03 4.734e+03, threshold=3.895e+03, percent-clipped=3.0
+2023-03-04 09:00:31,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=348374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:00:33,321 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=348377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:00:35,889 INFO [train.py:968] (0/2) Epoch 8, batch 29600, giga_loss[loss=0.3969, simple_loss=0.4227, pruned_loss=0.1855, over 26530.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.39, pruned_loss=0.1401, over 5663681.25 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3812, pruned_loss=0.1325, over 5691477.71 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3916, pruned_loss=0.1414, over 5654749.91 frames. ], batch size: 555, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 09:00:41,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 09:01:18,206 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-04 09:01:25,821 INFO [train.py:968] (0/2) Epoch 8, batch 29650, libri_loss[loss=0.3513, simple_loss=0.4056, pruned_loss=0.1485, over 25852.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3904, pruned_loss=0.1406, over 5648670.86 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3814, pruned_loss=0.1326, over 5690696.32 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3917, pruned_loss=0.1416, over 5641848.96 frames. ], batch size: 136, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 09:01:45,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.358e+02 1.428e+03 1.872e+03 2.751e+03 6.436e+03, threshold=3.743e+03, percent-clipped=5.0
+2023-03-04 09:02:05,828 INFO [train.py:968] (0/2) Epoch 8, batch 29700, giga_loss[loss=0.3984, simple_loss=0.4326, pruned_loss=0.1821, over 28012.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3899, pruned_loss=0.1404, over 5652976.75 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3809, pruned_loss=0.1324, over 5692287.02 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3917, pruned_loss=0.1417, over 5644948.22 frames. ], batch size: 412, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:02:55,557 INFO [train.py:968] (0/2) Epoch 8, batch 29750, giga_loss[loss=0.3453, simple_loss=0.4073, pruned_loss=0.1417, over 28120.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3892, pruned_loss=0.1389, over 5656068.12 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3806, pruned_loss=0.1321, over 5693055.60 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.391, pruned_loss=0.1403, over 5648494.89 frames. ], batch size: 77, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:03:03,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0565, 1.0389, 3.7226, 2.9855], device='cuda:0'), covar=tensor([0.1621, 0.2476, 0.0471, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0619, 0.0568, 0.0815, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 09:03:19,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.961e+02 1.490e+03 1.887e+03 2.594e+03 1.212e+04, threshold=3.775e+03, percent-clipped=13.0
+2023-03-04 09:03:39,461 INFO [train.py:968] (0/2) Epoch 8, batch 29800, giga_loss[loss=0.3242, simple_loss=0.3885, pruned_loss=0.13, over 28663.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3893, pruned_loss=0.1384, over 5644423.55 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3812, pruned_loss=0.1325, over 5679597.05 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3906, pruned_loss=0.1394, over 5648890.80 frames. ], batch size: 262, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:04:26,074 INFO [train.py:968] (0/2) Epoch 8, batch 29850, giga_loss[loss=0.3355, simple_loss=0.394, pruned_loss=0.1384, over 28901.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3884, pruned_loss=0.1379, over 5656297.94 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3807, pruned_loss=0.1322, over 5688398.43 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3902, pruned_loss=0.1392, over 5650539.91 frames. ], batch size: 174, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:04:51,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.439e+03 2.012e+03 2.733e+03 5.742e+03, threshold=4.025e+03, percent-clipped=6.0
+2023-03-04 09:05:09,965 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2997, 4.1426, 3.9181, 1.9721], device='cuda:0'), covar=tensor([0.0541, 0.0677, 0.0710, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.1004, 0.0945, 0.0839, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 09:05:11,503 INFO [train.py:968] (0/2) Epoch 8, batch 29900, giga_loss[loss=0.3427, simple_loss=0.3982, pruned_loss=0.1436, over 28559.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.388, pruned_loss=0.1376, over 5665908.44 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3811, pruned_loss=0.1325, over 5692551.87 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3892, pruned_loss=0.1385, over 5657276.37 frames. ], batch size: 336, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:05:36,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3707, 1.8305, 1.4469, 1.6339], device='cuda:0'), covar=tensor([0.0738, 0.0273, 0.0306, 0.0777], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0118, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 09:05:55,207 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=348726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:05:56,930 INFO [train.py:968] (0/2) Epoch 8, batch 29950, giga_loss[loss=0.2673, simple_loss=0.3371, pruned_loss=0.0988, over 28584.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3855, pruned_loss=0.1364, over 5670408.37 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.381, pruned_loss=0.1324, over 5695743.62 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3865, pruned_loss=0.1372, over 5660582.37 frames. ], batch size: 71, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:06:21,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=348749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:06:24,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=348752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:06:30,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.876e+02 1.882e+03 2.464e+03 3.185e+03 8.072e+03, threshold=4.927e+03, percent-clipped=13.0
+2023-03-04 09:06:33,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=348761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:06:48,545 INFO [train.py:968] (0/2) Epoch 8, batch 30000, giga_loss[loss=0.3004, simple_loss=0.3582, pruned_loss=0.1213, over 28651.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3824, pruned_loss=0.1356, over 5651493.71 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3813, pruned_loss=0.1326, over 5688523.26 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.383, pruned_loss=0.1361, over 5649858.43 frames. ], batch size: 262, lr: 4.00e-03, grad_scale: 8.0
+2023-03-04 09:06:48,550 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 09:06:57,951 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2237, simple_loss=0.331, pruned_loss=0.05819, over 944034.00 frames. 
+2023-03-04 09:06:57,952 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 09:07:17,974 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=348805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:07:35,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7247, 2.2820, 1.7842, 1.4356], device='cuda:0'), covar=tensor([0.2192, 0.1345, 0.1399, 0.1703], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1492, 0.1461, 0.1563], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 09:07:41,482 INFO [train.py:968] (0/2) Epoch 8, batch 30050, giga_loss[loss=0.3098, simple_loss=0.3669, pruned_loss=0.1264, over 28400.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.38, pruned_loss=0.1341, over 5665322.25 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.381, pruned_loss=0.1323, over 5693039.28 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3807, pruned_loss=0.1348, over 5659236.74 frames. ], batch size: 65, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:07:46,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6049, 1.8457, 1.8905, 1.4515], device='cuda:0'), covar=tensor([0.1583, 0.2015, 0.1237, 0.1405], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0720, 0.0835, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 09:08:07,978 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6616, 1.9448, 1.6756, 1.6319], device='cuda:0'), covar=tensor([0.0689, 0.0259, 0.0266, 0.0735], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0116, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 09:08:09,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.792e+03 2.182e+03 2.837e+03 5.540e+03, threshold=4.365e+03, percent-clipped=3.0
+2023-03-04 09:08:32,506 INFO [train.py:968] (0/2) Epoch 8, batch 30100, giga_loss[loss=0.3425, simple_loss=0.3962, pruned_loss=0.1444, over 27879.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3789, pruned_loss=0.1338, over 5645501.13 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3813, pruned_loss=0.1325, over 5694029.02 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3792, pruned_loss=0.1342, over 5639405.94 frames. ], batch size: 412, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:08:47,401 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=348892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:08:49,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=348895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:08:49,423 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=348895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:08:51,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=348898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:09:12,951 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=348924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:09:16,990 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=348927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:09:18,225 INFO [train.py:968] (0/2) Epoch 8, batch 30150, giga_loss[loss=0.2723, simple_loss=0.3312, pruned_loss=0.1067, over 24055.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3786, pruned_loss=0.1318, over 5649099.02 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3813, pruned_loss=0.1325, over 5693597.90 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3788, pruned_loss=0.1322, over 5643458.68 frames. ], batch size: 705, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:09:48,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.771e+02 1.664e+03 2.286e+03 3.335e+03 6.527e+03, threshold=4.572e+03, percent-clipped=10.0
+2023-03-04 09:10:10,064 INFO [train.py:968] (0/2) Epoch 8, batch 30200, giga_loss[loss=0.2471, simple_loss=0.3328, pruned_loss=0.08073, over 28291.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3766, pruned_loss=0.1289, over 5645793.30 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3814, pruned_loss=0.1329, over 5699862.24 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3765, pruned_loss=0.1287, over 5634272.51 frames. ], batch size: 368, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:10:48,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3102, 4.1456, 3.9209, 1.8565], device='cuda:0'), covar=tensor([0.0531, 0.0733, 0.0837, 0.2178], device='cuda:0'), in_proj_covar=tensor([0.0992, 0.0932, 0.0824, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 09:10:59,319 INFO [train.py:968] (0/2) Epoch 8, batch 30250, giga_loss[loss=0.3215, simple_loss=0.3697, pruned_loss=0.1366, over 26725.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3738, pruned_loss=0.1255, over 5651001.61 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3812, pruned_loss=0.1328, over 5695890.14 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3738, pruned_loss=0.1253, over 5643808.19 frames. ], batch size: 555, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:11:26,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.285e+02 1.467e+03 1.862e+03 2.713e+03 5.076e+03, threshold=3.724e+03, percent-clipped=3.0
+2023-03-04 09:11:48,083 INFO [train.py:968] (0/2) Epoch 8, batch 30300, libri_loss[loss=0.2646, simple_loss=0.3377, pruned_loss=0.09577, over 29549.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3707, pruned_loss=0.1225, over 5659941.88 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3807, pruned_loss=0.1327, over 5705382.00 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3708, pruned_loss=0.122, over 5642929.03 frames. ], batch size: 83, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:12:06,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=349101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:12:31,985 INFO [train.py:968] (0/2) Epoch 8, batch 30350, giga_loss[loss=0.2612, simple_loss=0.3315, pruned_loss=0.09544, over 28910.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3662, pruned_loss=0.1184, over 5666526.68 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3799, pruned_loss=0.1325, over 5711633.73 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3666, pruned_loss=0.1178, over 5645432.56 frames. ], batch size: 213, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:12:39,480 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=349136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:12:39,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4565, 1.7823, 1.7896, 1.3441], device='cuda:0'), covar=tensor([0.1608, 0.2194, 0.1305, 0.1528], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0710, 0.0828, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 09:12:55,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=349153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:13:01,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.497e+02 1.335e+03 1.702e+03 2.524e+03 7.281e+03, threshold=3.404e+03, percent-clipped=7.0
+2023-03-04 09:13:24,253 INFO [train.py:968] (0/2) Epoch 8, batch 30400, giga_loss[loss=0.2619, simple_loss=0.3514, pruned_loss=0.08619, over 28684.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3639, pruned_loss=0.1137, over 5676999.12 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3798, pruned_loss=0.1326, over 5709861.26 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3641, pruned_loss=0.113, over 5661645.91 frames. ], batch size: 92, lr: 4.00e-03, grad_scale: 8.0
+2023-03-04 09:13:25,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=349180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:13:51,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3885, 1.7180, 1.6461, 1.2978], device='cuda:0'), covar=tensor([0.1352, 0.1908, 0.1191, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0700, 0.0823, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 09:14:18,091 INFO [train.py:968] (0/2) Epoch 8, batch 30450, giga_loss[loss=0.2594, simple_loss=0.3456, pruned_loss=0.0866, over 28953.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3637, pruned_loss=0.1131, over 5677199.39 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3797, pruned_loss=0.1327, over 5711983.99 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3638, pruned_loss=0.1122, over 5662927.94 frames. ], batch size: 164, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:14:36,379 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=349244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:14:38,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=349247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:14:50,604 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.409e+02 1.238e+03 1.896e+03 2.717e+03 1.433e+04, threshold=3.791e+03, percent-clipped=15.0
+2023-03-04 09:15:03,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6820, 1.5998, 1.3804, 1.4653], device='cuda:0'), covar=tensor([0.0551, 0.0350, 0.0694, 0.0773], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0441, 0.0497, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 09:15:06,289 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=349276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:08,834 INFO [train.py:968] (0/2) Epoch 8, batch 30500, giga_loss[loss=0.3071, simple_loss=0.3726, pruned_loss=0.1208, over 28243.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3618, pruned_loss=0.1116, over 5673942.59 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3791, pruned_loss=0.1324, over 5715045.84 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.362, pruned_loss=0.1107, over 5658954.68 frames. ], batch size: 368, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:15:09,115 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=349279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:12,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=349282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:39,595 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=349311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:52,316 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=349323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:55,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=349326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:57,739 INFO [train.py:968] (0/2) Epoch 8, batch 30550, giga_loss[loss=0.2411, simple_loss=0.3305, pruned_loss=0.07585, over 29027.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3587, pruned_loss=0.1089, over 5678068.37 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3788, pruned_loss=0.1322, over 5718378.11 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3587, pruned_loss=0.1079, over 5662675.82 frames. ], batch size: 155, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:16:22,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5968, 1.8061, 1.4367, 1.3445], device='cuda:0'), covar=tensor([0.1517, 0.1158, 0.1061, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1447, 0.1409, 0.1523], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 09:16:23,720 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=349355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:16:27,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.582e+02 1.316e+03 1.815e+03 2.669e+03 8.100e+03, threshold=3.631e+03, percent-clipped=10.0
+2023-03-04 09:16:30,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3045, 5.1097, 4.8141, 2.4923], device='cuda:0'), covar=tensor([0.0390, 0.0633, 0.0713, 0.1722], device='cuda:0'), in_proj_covar=tensor([0.0978, 0.0919, 0.0813, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 09:16:48,383 INFO [train.py:968] (0/2) Epoch 8, batch 30600, giga_loss[loss=0.2799, simple_loss=0.357, pruned_loss=0.1014, over 28985.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3567, pruned_loss=0.1079, over 5666672.41 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3785, pruned_loss=0.1321, over 5720224.03 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3566, pruned_loss=0.1068, over 5652211.35 frames. ], batch size: 213, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:17:35,064 INFO [train.py:968] (0/2) Epoch 8, batch 30650, libri_loss[loss=0.2993, simple_loss=0.3671, pruned_loss=0.1157, over 29522.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.357, pruned_loss=0.1076, over 5673828.85 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3784, pruned_loss=0.1319, over 5721301.62 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3566, pruned_loss=0.1063, over 5660139.10 frames. ], batch size: 84, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:17:35,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5607, 1.7343, 1.2237, 1.3596], device='cuda:0'), covar=tensor([0.0604, 0.0334, 0.0832, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0438, 0.0492, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 09:18:05,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.271e+02 1.263e+03 1.641e+03 2.215e+03 5.105e+03, threshold=3.282e+03, percent-clipped=2.0
+2023-03-04 09:18:24,846 INFO [train.py:968] (0/2) Epoch 8, batch 30700, giga_loss[loss=0.2413, simple_loss=0.3268, pruned_loss=0.0779, over 28629.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3554, pruned_loss=0.1069, over 5665888.29 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.378, pruned_loss=0.1321, over 5717078.08 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3547, pruned_loss=0.105, over 5657235.50 frames. ], batch size: 307, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:18:59,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3091, 1.5800, 1.2709, 1.2345], device='cuda:0'), covar=tensor([0.1727, 0.1241, 0.1187, 0.1328], device='cuda:0'), in_proj_covar=tensor([0.1581, 0.1438, 0.1399, 0.1517], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 09:19:15,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=349528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:19:15,838 INFO [train.py:968] (0/2) Epoch 8, batch 30750, giga_loss[loss=0.2694, simple_loss=0.3401, pruned_loss=0.09932, over 28638.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.353, pruned_loss=0.1052, over 5662425.79 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3783, pruned_loss=0.1325, over 5715729.29 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3516, pruned_loss=0.1026, over 5655303.68 frames. ], batch size: 307, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:19:47,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.184e+02 1.359e+03 1.684e+03 2.397e+03 6.793e+03, threshold=3.368e+03, percent-clipped=12.0
+2023-03-04 09:20:08,778 INFO [train.py:968] (0/2) Epoch 8, batch 30800, giga_loss[loss=0.2398, simple_loss=0.3196, pruned_loss=0.07994, over 29052.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3487, pruned_loss=0.1022, over 5674909.50 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3783, pruned_loss=0.1326, over 5717775.22 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3474, pruned_loss=0.09982, over 5666956.64 frames. ], batch size: 136, lr: 4.00e-03, grad_scale: 8.0
+2023-03-04 09:20:56,183 INFO [train.py:968] (0/2) Epoch 8, batch 30850, giga_loss[loss=0.258, simple_loss=0.3355, pruned_loss=0.09026, over 28534.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3463, pruned_loss=0.1014, over 5659670.84 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3776, pruned_loss=0.1324, over 5702303.89 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3452, pruned_loss=0.0989, over 5665780.30 frames. ], batch size: 336, lr: 4.00e-03, grad_scale: 8.0
+2023-03-04 09:20:59,241 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-04 09:21:02,216 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-04 09:21:26,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.210e+02 1.263e+03 1.703e+03 2.413e+03 4.636e+03, threshold=3.406e+03, percent-clipped=8.0
+2023-03-04 09:21:37,767 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=349671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:21:40,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=349674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:21:46,362 INFO [train.py:968] (0/2) Epoch 8, batch 30900, giga_loss[loss=0.25, simple_loss=0.34, pruned_loss=0.07995, over 28832.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3467, pruned_loss=0.1021, over 5629593.86 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3778, pruned_loss=0.1327, over 5676846.81 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.345, pruned_loss=0.09934, over 5655206.48 frames. ], batch size: 174, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:22:03,607 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=349693.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:22:13,213 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=349703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:22:16,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-04 09:22:24,959 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=349714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:22:32,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4051, 1.4945, 1.2344, 1.5484], device='cuda:0'), covar=tensor([0.0754, 0.0288, 0.0342, 0.0830], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0119, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 09:22:39,599 INFO [train.py:968] (0/2) Epoch 8, batch 30950, giga_loss[loss=0.2831, simple_loss=0.3484, pruned_loss=0.1089, over 26757.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3481, pruned_loss=0.1031, over 5621425.16 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3775, pruned_loss=0.1326, over 5672969.28 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3461, pruned_loss=0.1002, over 5643964.78 frames. ], batch size: 555, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:23:07,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9219, 1.1175, 1.0500, 0.8105], device='cuda:0'), covar=tensor([0.1063, 0.1246, 0.0669, 0.1019], device='cuda:0'), in_proj_covar=tensor([0.1559, 0.1420, 0.1367, 0.1491], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0')
+2023-03-04 09:23:16,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.511e+02 1.629e+03 2.046e+03 3.123e+03 7.771e+03, threshold=4.091e+03, percent-clipped=18.0
+2023-03-04 09:23:35,087 INFO [train.py:968] (0/2) Epoch 8, batch 31000, giga_loss[loss=0.2756, simple_loss=0.3557, pruned_loss=0.09772, over 28952.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3506, pruned_loss=0.1038, over 5627200.82 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3767, pruned_loss=0.1322, over 5676029.07 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3489, pruned_loss=0.101, over 5641296.71 frames. ], batch size: 164, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:23:40,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3128, 1.6914, 1.5962, 1.2242], device='cuda:0'), covar=tensor([0.1547, 0.2163, 0.1285, 0.1509], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0702, 0.0823, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 09:24:10,215 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0698, 4.8926, 4.6175, 2.0320], device='cuda:0'), covar=tensor([0.0357, 0.0521, 0.0609, 0.2025], device='cuda:0'), in_proj_covar=tensor([0.0965, 0.0905, 0.0798, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 09:24:35,741 INFO [train.py:968] (0/2) Epoch 8, batch 31050, giga_loss[loss=0.35, simple_loss=0.3996, pruned_loss=0.1502, over 27961.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.35, pruned_loss=0.1032, over 5623280.95 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3763, pruned_loss=0.132, over 5679062.62 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3486, pruned_loss=0.1007, over 5630780.59 frames. ], batch size: 412, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:25:13,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.159e+02 1.402e+03 1.811e+03 2.339e+03 7.842e+03, threshold=3.621e+03, percent-clipped=7.0
+2023-03-04 09:25:25,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=349870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:25:37,009 INFO [train.py:968] (0/2) Epoch 8, batch 31100, libri_loss[loss=0.3415, simple_loss=0.3835, pruned_loss=0.1498, over 29522.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3497, pruned_loss=0.1034, over 5621942.90 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3759, pruned_loss=0.1319, over 5668788.81 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3478, pruned_loss=0.1002, over 5635422.84 frames. ], batch size: 82, lr: 4.00e-03, grad_scale: 2.0
+2023-03-04 09:26:30,175 INFO [train.py:968] (0/2) Epoch 8, batch 31150, giga_loss[loss=0.2626, simple_loss=0.3412, pruned_loss=0.09196, over 27629.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3481, pruned_loss=0.102, over 5631166.45 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3758, pruned_loss=0.1319, over 5668675.34 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3456, pruned_loss=0.0982, over 5641051.38 frames. ], batch size: 472, lr: 4.00e-03, grad_scale: 2.0
+2023-03-04 09:27:07,880 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-04 09:27:15,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.774e+02 1.263e+03 1.685e+03 2.283e+03 1.112e+04, threshold=3.370e+03, percent-clipped=10.0
+2023-03-04 09:27:35,913 INFO [train.py:968] (0/2) Epoch 8, batch 31200, giga_loss[loss=0.2568, simple_loss=0.338, pruned_loss=0.08783, over 29029.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3448, pruned_loss=0.09828, over 5626891.95 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3753, pruned_loss=0.1316, over 5671163.04 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.343, pruned_loss=0.09526, over 5632136.67 frames. ], batch size: 213, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:27:45,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4089, 1.5064, 1.2537, 1.5519], device='cuda:0'), covar=tensor([0.2336, 0.2190, 0.2350, 0.2133], device='cuda:0'), in_proj_covar=tensor([0.1226, 0.0915, 0.1088, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 09:27:46,650 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=349989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:27:57,387 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-350000.pt
+2023-03-04 09:28:28,891 INFO [train.py:968] (0/2) Epoch 8, batch 31250, giga_loss[loss=0.2711, simple_loss=0.3393, pruned_loss=0.1014, over 28132.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3435, pruned_loss=0.09844, over 5631633.55 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3742, pruned_loss=0.1311, over 5666258.20 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3414, pruned_loss=0.09462, over 5637638.38 frames. ], batch size: 412, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:29:06,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.983e+02 1.418e+03 1.934e+03 2.837e+03 5.806e+03, threshold=3.867e+03, percent-clipped=16.0
+2023-03-04 09:29:16,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350068.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:29:19,522 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=350070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:29:28,578 INFO [train.py:968] (0/2) Epoch 8, batch 31300, giga_loss[loss=0.2744, simple_loss=0.3508, pruned_loss=0.099, over 28075.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3411, pruned_loss=0.09745, over 5651709.26 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3731, pruned_loss=0.1305, over 5671315.71 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3396, pruned_loss=0.09417, over 5651589.99 frames. ], batch size: 412, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:29:36,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9312, 3.7373, 3.5311, 1.8816], device='cuda:0'), covar=tensor([0.0629, 0.0857, 0.0905, 0.2003], device='cuda:0'), in_proj_covar=tensor([0.0965, 0.0903, 0.0800, 0.0627], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 09:29:40,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:30:26,277 INFO [train.py:968] (0/2) Epoch 8, batch 31350, giga_loss[loss=0.2802, simple_loss=0.3579, pruned_loss=0.1012, over 28509.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3408, pruned_loss=0.09768, over 5663132.99 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3731, pruned_loss=0.1306, over 5676531.39 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.339, pruned_loss=0.09427, over 5658056.60 frames. ], batch size: 336, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:31:02,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.994e+02 1.305e+03 1.672e+03 2.240e+03 4.199e+03, threshold=3.345e+03, percent-clipped=1.0
+2023-03-04 09:31:22,052 INFO [train.py:968] (0/2) Epoch 8, batch 31400, giga_loss[loss=0.264, simple_loss=0.3468, pruned_loss=0.09063, over 29011.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.341, pruned_loss=0.09673, over 5668075.88 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3726, pruned_loss=0.1303, over 5682093.88 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3391, pruned_loss=0.09352, over 5658700.93 frames. ], batch size: 285, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:32:03,616 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350211.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 09:32:07,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350214.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:32:21,473 INFO [train.py:968] (0/2) Epoch 8, batch 31450, giga_loss[loss=0.271, simple_loss=0.3495, pruned_loss=0.09627, over 29067.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3436, pruned_loss=0.09834, over 5659005.25 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.372, pruned_loss=0.13, over 5684065.48 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3415, pruned_loss=0.09462, over 5648497.76 frames. ], batch size: 128, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:32:27,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:32:29,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:32:39,781 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350243.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:32:41,561 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:32:59,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.099e+02 1.282e+03 1.734e+03 2.651e+03 7.041e+03, threshold=3.468e+03, percent-clipped=15.0
+2023-03-04 09:33:02,351 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:33:19,995 INFO [train.py:968] (0/2) Epoch 8, batch 31500, giga_loss[loss=0.232, simple_loss=0.3135, pruned_loss=0.07519, over 28675.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3406, pruned_loss=0.09623, over 5678028.22 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3707, pruned_loss=0.1292, over 5692097.14 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3391, pruned_loss=0.09289, over 5661938.86 frames. ], batch size: 307, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:33:31,540 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=350285.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:33:38,439 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-04 09:34:24,220 INFO [train.py:968] (0/2) Epoch 8, batch 31550, giga_loss[loss=0.3147, simple_loss=0.3835, pruned_loss=0.123, over 28544.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3428, pruned_loss=0.09858, over 5669089.33 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3699, pruned_loss=0.1289, over 5681443.41 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3413, pruned_loss=0.09506, over 5665507.39 frames. ], batch size: 85, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:34:31,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4229, 1.9125, 1.4205, 1.5529], device='cuda:0'), covar=tensor([0.0736, 0.0296, 0.0319, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0119, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 09:34:34,150 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 09:35:03,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.767e+02 1.403e+03 1.863e+03 2.804e+03 1.205e+04, threshold=3.726e+03, percent-clipped=12.0
+2023-03-04 09:35:10,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:35:22,625 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-04 09:35:28,254 INFO [train.py:968] (0/2) Epoch 8, batch 31600, giga_loss[loss=0.2733, simple_loss=0.3509, pruned_loss=0.09788, over 27571.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3454, pruned_loss=0.09912, over 5665311.48 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.37, pruned_loss=0.129, over 5680210.48 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3438, pruned_loss=0.09579, over 5663388.37 frames. ], batch size: 472, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:35:39,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:35:43,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:36:08,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4380, 2.0497, 1.3754, 0.7092], device='cuda:0'), covar=tensor([0.3552, 0.1706, 0.2712, 0.3660], device='cuda:0'), in_proj_covar=tensor([0.1475, 0.1404, 0.1439, 0.1210], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 09:36:20,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:36:24,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=350423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:36:33,479 INFO [train.py:968] (0/2) Epoch 8, batch 31650, giga_loss[loss=0.2589, simple_loss=0.353, pruned_loss=0.0824, over 28876.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3479, pruned_loss=0.09769, over 5659607.40 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.37, pruned_loss=0.1291, over 5681633.68 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3464, pruned_loss=0.09472, over 5656795.26 frames. ], batch size: 174, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:36:55,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:37:16,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.147e+02 1.350e+03 1.822e+03 2.572e+03 5.794e+03, threshold=3.644e+03, percent-clipped=8.0
+2023-03-04 09:37:38,187 INFO [train.py:968] (0/2) Epoch 8, batch 31700, giga_loss[loss=0.2895, simple_loss=0.3693, pruned_loss=0.1048, over 28466.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3494, pruned_loss=0.097, over 5658917.00 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3699, pruned_loss=0.129, over 5682760.51 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3482, pruned_loss=0.09458, over 5655566.01 frames. ], batch size: 336, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:38:07,274 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:38:10,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:38:33,376 INFO [train.py:968] (0/2) Epoch 8, batch 31750, giga_loss[loss=0.2395, simple_loss=0.326, pruned_loss=0.07651, over 29066.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3495, pruned_loss=0.09644, over 5672038.50 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3704, pruned_loss=0.1296, over 5687597.90 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3475, pruned_loss=0.09299, over 5664381.22 frames. ], batch size: 128, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:38:45,485 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:39:14,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.013e+02 1.287e+03 1.668e+03 2.258e+03 4.900e+03, threshold=3.336e+03, percent-clipped=11.0
+2023-03-04 09:39:35,427 INFO [train.py:968] (0/2) Epoch 8, batch 31800, libri_loss[loss=0.3001, simple_loss=0.3653, pruned_loss=0.1174, over 28612.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3503, pruned_loss=0.09781, over 5678518.16 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.37, pruned_loss=0.1294, over 5688306.31 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3485, pruned_loss=0.09448, over 5671711.97 frames. ], batch size: 106, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:39:37,877 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=350582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:39:47,707 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:39:52,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:40:28,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4645, 1.7220, 1.4112, 1.8057], device='cuda:0'), covar=tensor([0.2130, 0.1918, 0.2091, 0.2089], device='cuda:0'), in_proj_covar=tensor([0.1223, 0.0912, 0.1083, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 09:40:29,619 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:40:43,993 INFO [train.py:968] (0/2) Epoch 8, batch 31850, giga_loss[loss=0.2723, simple_loss=0.3485, pruned_loss=0.09805, over 28782.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3478, pruned_loss=0.09711, over 5679972.47 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3701, pruned_loss=0.1295, over 5691429.13 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.346, pruned_loss=0.0939, over 5671729.17 frames. ], batch size: 243, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:41:26,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350660.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:41:32,241 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.213e+02 1.349e+03 1.789e+03 2.534e+03 1.394e+04, threshold=3.578e+03, percent-clipped=16.0
+2023-03-04 09:41:50,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9565, 1.1356, 1.2104, 1.0908], device='cuda:0'), covar=tensor([0.0855, 0.0850, 0.1209, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0714, 0.0638, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 09:42:00,221 INFO [train.py:968] (0/2) Epoch 8, batch 31900, giga_loss[loss=0.2354, simple_loss=0.319, pruned_loss=0.07588, over 29159.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3498, pruned_loss=0.09974, over 5681686.69 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.37, pruned_loss=0.1294, over 5697456.68 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3479, pruned_loss=0.0963, over 5669219.48 frames. ], batch size: 199, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:43:10,465 INFO [train.py:968] (0/2) Epoch 8, batch 31950, giga_loss[loss=0.2338, simple_loss=0.318, pruned_loss=0.07478, over 28999.00 frames. ], tot_loss[loss=0.269, simple_loss=0.345, pruned_loss=0.09656, over 5686745.26 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3698, pruned_loss=0.1292, over 5699422.72 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3429, pruned_loss=0.09319, over 5674439.85 frames. ], batch size: 285, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:43:28,760 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 09:43:54,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.235e+02 1.170e+03 1.448e+03 2.219e+03 6.696e+03, threshold=2.897e+03, percent-clipped=7.0
+2023-03-04 09:44:13,353 INFO [train.py:968] (0/2) Epoch 8, batch 32000, giga_loss[loss=0.2984, simple_loss=0.3642, pruned_loss=0.1164, over 29096.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3429, pruned_loss=0.09527, over 5686658.17 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3695, pruned_loss=0.129, over 5703766.94 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.341, pruned_loss=0.09202, over 5672515.44 frames. ], batch size: 285, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:44:38,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:44:45,745 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350803.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:44:48,675 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=350805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:44:48,930 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-04 09:44:49,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350806.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:45:15,515 INFO [train.py:968] (0/2) Epoch 8, batch 32050, giga_loss[loss=0.2951, simple_loss=0.363, pruned_loss=0.1136, over 28471.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3415, pruned_loss=0.09534, over 5673391.39 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3692, pruned_loss=0.1289, over 5688529.86 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3392, pruned_loss=0.09166, over 5673860.78 frames. ], batch size: 336, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:45:24,119 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350835.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:46:00,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.157e+02 1.343e+03 1.601e+03 2.036e+03 5.789e+03, threshold=3.201e+03, percent-clipped=11.0
+2023-03-04 09:46:21,785 INFO [train.py:968] (0/2) Epoch 8, batch 32100, giga_loss[loss=0.2791, simple_loss=0.3644, pruned_loss=0.09692, over 28927.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3453, pruned_loss=0.09708, over 5677979.76 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3688, pruned_loss=0.1288, over 5691539.46 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3435, pruned_loss=0.09384, over 5675534.28 frames. ], batch size: 164, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:47:21,577 INFO [train.py:968] (0/2) Epoch 8, batch 32150, giga_loss[loss=0.2848, simple_loss=0.3512, pruned_loss=0.1092, over 28815.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3466, pruned_loss=0.09847, over 5690689.50 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3683, pruned_loss=0.1283, over 5694962.97 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3453, pruned_loss=0.09574, over 5685384.89 frames. ], batch size: 227, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:47:38,144 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:47:45,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:48:01,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:48:07,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.658e+02 1.378e+03 1.813e+03 2.546e+03 9.736e+03, threshold=3.627e+03, percent-clipped=11.0
+2023-03-04 09:48:17,706 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:48:23,898 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 09:48:26,996 INFO [train.py:968] (0/2) Epoch 8, batch 32200, giga_loss[loss=0.2503, simple_loss=0.3115, pruned_loss=0.09459, over 24887.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3448, pruned_loss=0.09848, over 5688896.60 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3676, pruned_loss=0.128, over 5697543.41 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3438, pruned_loss=0.09581, over 5681990.83 frames. ], batch size: 705, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:48:39,042 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=350992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:49:24,403 INFO [train.py:968] (0/2) Epoch 8, batch 32250, giga_loss[loss=0.263, simple_loss=0.3405, pruned_loss=0.09271, over 28625.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3454, pruned_loss=0.09987, over 5689489.87 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3668, pruned_loss=0.1276, over 5702137.23 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3445, pruned_loss=0.09715, over 5679403.81 frames. ], batch size: 307, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:49:25,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 09:49:59,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1081, 1.0999, 3.7824, 3.1106], device='cuda:0'), covar=tensor([0.1666, 0.2551, 0.0404, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0611, 0.0569, 0.0809, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 09:50:05,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.933e+02 1.445e+03 1.944e+03 3.035e+03 8.196e+03, threshold=3.887e+03, percent-clipped=16.0
+2023-03-04 09:50:25,714 INFO [train.py:968] (0/2) Epoch 8, batch 32300, giga_loss[loss=0.3059, simple_loss=0.3829, pruned_loss=0.1145, over 28707.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.345, pruned_loss=0.09927, over 5682814.05 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3657, pruned_loss=0.1269, over 5699807.17 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3445, pruned_loss=0.09656, over 5675619.78 frames. ], batch size: 262, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:50:33,939 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:50:54,564 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:50:59,488 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:51:33,924 INFO [train.py:968] (0/2) Epoch 8, batch 32350, libri_loss[loss=0.3193, simple_loss=0.3754, pruned_loss=0.1316, over 29539.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3468, pruned_loss=0.09906, over 5669380.72 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3656, pruned_loss=0.1268, over 5690403.76 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.346, pruned_loss=0.09635, over 5671373.33 frames. ], batch size: 81, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:51:40,659 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:52:00,574 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 09:52:19,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.348e+02 1.349e+03 1.938e+03 2.671e+03 6.224e+03, threshold=3.875e+03, percent-clipped=10.0
+2023-03-04 09:52:36,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8483, 1.1570, 5.5844, 3.8582], device='cuda:0'), covar=tensor([0.1524, 0.2531, 0.0339, 0.0746], device='cuda:0'), in_proj_covar=tensor([0.0607, 0.0566, 0.0805, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 09:52:42,490 INFO [train.py:968] (0/2) Epoch 8, batch 32400, giga_loss[loss=0.238, simple_loss=0.305, pruned_loss=0.08551, over 24373.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3458, pruned_loss=0.09889, over 5668042.75 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3647, pruned_loss=0.1262, over 5700342.73 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3451, pruned_loss=0.09588, over 5659400.42 frames. ], batch size: 705, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:52:43,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:52:45,172 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7100, 3.5267, 3.3122, 1.5321], device='cuda:0'), covar=tensor([0.0675, 0.0767, 0.0811, 0.2287], device='cuda:0'), in_proj_covar=tensor([0.0964, 0.0906, 0.0801, 0.0629], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 09:53:46,640 INFO [train.py:968] (0/2) Epoch 8, batch 32450, giga_loss[loss=0.2682, simple_loss=0.338, pruned_loss=0.09927, over 28642.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3416, pruned_loss=0.09723, over 5678887.19 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3641, pruned_loss=0.1258, over 5701450.08 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3411, pruned_loss=0.0945, over 5670693.10 frames. ], batch size: 307, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:53:58,487 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-04 09:54:02,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5401, 2.2148, 1.5732, 0.7250], device='cuda:0'), covar=tensor([0.2777, 0.1746, 0.2480, 0.3035], device='cuda:0'), in_proj_covar=tensor([0.1488, 0.1417, 0.1453, 0.1218], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 09:54:07,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0407, 1.3528, 1.0480, 0.3859], device='cuda:0'), covar=tensor([0.1988, 0.1771, 0.3197, 0.3218], device='cuda:0'), in_proj_covar=tensor([0.1489, 0.1419, 0.1454, 0.1219], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 09:54:21,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2319, 1.3895, 1.0875, 1.0386], device='cuda:0'), covar=tensor([0.0747, 0.0418, 0.1071, 0.0829], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0439, 0.0500, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 09:54:36,736 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.581e+02 1.327e+03 1.731e+03 2.525e+03 4.358e+03, threshold=3.461e+03, percent-clipped=5.0
+2023-03-04 09:54:38,647 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:54:57,161 INFO [train.py:968] (0/2) Epoch 8, batch 32500, giga_loss[loss=0.282, simple_loss=0.3452, pruned_loss=0.1094, over 28714.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3355, pruned_loss=0.09434, over 5680107.50 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3641, pruned_loss=0.1258, over 5702446.83 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.335, pruned_loss=0.09208, over 5672789.23 frames. ], batch size: 243, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:55:29,650 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4405, 1.2583, 4.7006, 3.3747], device='cuda:0'), covar=tensor([0.1537, 0.2449, 0.0335, 0.0754], device='cuda:0'), in_proj_covar=tensor([0.0603, 0.0565, 0.0799, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 09:55:49,242 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:55:51,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:55:53,378 INFO [train.py:968] (0/2) Epoch 8, batch 32550, giga_loss[loss=0.2706, simple_loss=0.3391, pruned_loss=0.101, over 28963.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.336, pruned_loss=0.0955, over 5664602.88 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3643, pruned_loss=0.1263, over 5692248.59 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3341, pruned_loss=0.09189, over 5666737.50 frames. ], batch size: 213, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 09:56:00,773 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-04 09:56:04,320 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:56:24,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:56:35,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.474e+02 1.592e+03 2.232e+03 3.215e+03 1.015e+04, threshold=4.463e+03, percent-clipped=18.0
+2023-03-04 09:56:37,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:56:49,520 INFO [train.py:968] (0/2) Epoch 8, batch 32600, giga_loss[loss=0.2633, simple_loss=0.341, pruned_loss=0.09286, over 28860.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3372, pruned_loss=0.09623, over 5673118.98 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3638, pruned_loss=0.126, over 5698800.76 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3354, pruned_loss=0.0928, over 5668286.80 frames. ], batch size: 174, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 09:57:16,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9309, 3.7513, 3.5221, 1.6995], device='cuda:0'), covar=tensor([0.0600, 0.0787, 0.0870, 0.2285], device='cuda:0'), in_proj_covar=tensor([0.0961, 0.0901, 0.0794, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 09:57:52,554 INFO [train.py:968] (0/2) Epoch 8, batch 32650, giga_loss[loss=0.2312, simple_loss=0.3125, pruned_loss=0.07492, over 28790.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3356, pruned_loss=0.09409, over 5671631.76 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3636, pruned_loss=0.1258, over 5699882.75 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3342, pruned_loss=0.09139, over 5666896.47 frames. ], batch size: 243, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 09:57:52,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6050, 2.2858, 1.6066, 0.7799], device='cuda:0'), covar=tensor([0.2781, 0.1621, 0.2609, 0.3158], device='cuda:0'), in_proj_covar=tensor([0.1487, 0.1410, 0.1451, 0.1215], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 09:58:00,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4581, 2.1405, 1.5752, 0.6164], device='cuda:0'), covar=tensor([0.3517, 0.1641, 0.2635, 0.3785], device='cuda:0'), in_proj_covar=tensor([0.1487, 0.1410, 0.1451, 0.1215], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 09:58:27,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:58:33,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.399e+02 1.312e+03 1.717e+03 2.384e+03 5.905e+03, threshold=3.434e+03, percent-clipped=3.0
+2023-03-04 09:58:47,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2163, 1.8403, 1.3436, 0.4268], device='cuda:0'), covar=tensor([0.2328, 0.1655, 0.2600, 0.2849], device='cuda:0'), in_proj_covar=tensor([0.1490, 0.1411, 0.1453, 0.1214], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 09:58:51,884 INFO [train.py:968] (0/2) Epoch 8, batch 32700, giga_loss[loss=0.261, simple_loss=0.3377, pruned_loss=0.0921, over 28900.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3352, pruned_loss=0.09344, over 5660475.14 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3635, pruned_loss=0.1258, over 5695111.15 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3331, pruned_loss=0.0901, over 5660755.64 frames. ], batch size: 227, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 09:58:53,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.40 vs. limit=5.0
+2023-03-04 09:59:31,919 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:59:35,489 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:59:37,728 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-04 09:59:55,002 INFO [train.py:968] (0/2) Epoch 8, batch 32750, giga_loss[loss=0.2639, simple_loss=0.338, pruned_loss=0.09492, over 28733.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3339, pruned_loss=0.09329, over 5656751.68 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3634, pruned_loss=0.1259, over 5687684.63 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3318, pruned_loss=0.08986, over 5662408.43 frames. ], batch size: 307, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:00:12,174 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:00:43,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.769e+02 1.368e+03 1.912e+03 2.844e+03 8.512e+03, threshold=3.823e+03, percent-clipped=12.0
+2023-03-04 10:01:04,840 INFO [train.py:968] (0/2) Epoch 8, batch 32800, giga_loss[loss=0.2646, simple_loss=0.3492, pruned_loss=0.08997, over 28813.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3337, pruned_loss=0.09247, over 5667333.43 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3634, pruned_loss=0.1259, over 5691902.90 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3315, pruned_loss=0.08912, over 5667539.73 frames. ], batch size: 243, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 10:01:36,002 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:01:38,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:01:55,374 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351617.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 10:02:08,170 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2314, 1.3225, 1.1938, 1.4417], device='cuda:0'), covar=tensor([0.0778, 0.0348, 0.0337, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0114, 0.0119, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:0')
+2023-03-04 10:02:11,375 INFO [train.py:968] (0/2) Epoch 8, batch 32850, giga_loss[loss=0.2799, simple_loss=0.3444, pruned_loss=0.1077, over 26982.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3354, pruned_loss=0.09336, over 5669753.01 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3636, pruned_loss=0.1261, over 5686861.23 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3328, pruned_loss=0.08982, over 5673180.64 frames. ], batch size: 555, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 10:02:14,344 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:02:23,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:02:42,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9008, 1.8101, 1.3757, 1.4264], device='cuda:0'), covar=tensor([0.0595, 0.0497, 0.0891, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0440, 0.0498, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 10:02:56,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.858e+02 1.202e+03 1.635e+03 2.218e+03 7.443e+03, threshold=3.270e+03, percent-clipped=3.0
+2023-03-04 10:03:11,983 INFO [train.py:968] (0/2) Epoch 8, batch 32900, giga_loss[loss=0.2139, simple_loss=0.3044, pruned_loss=0.06174, over 28917.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3359, pruned_loss=0.09418, over 5681310.26 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3634, pruned_loss=0.1261, over 5694121.77 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3329, pruned_loss=0.09014, over 5676690.19 frames. ], batch size: 164, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 10:03:12,624 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:03:49,457 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 10:03:53,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:04:10,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0332, 1.0693, 1.0149, 1.3034], device='cuda:0'), covar=tensor([0.0824, 0.0356, 0.0316, 0.0955], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0114, 0.0119, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:0')
+2023-03-04 10:04:11,321 INFO [train.py:968] (0/2) Epoch 8, batch 32950, libri_loss[loss=0.3549, simple_loss=0.3867, pruned_loss=0.1616, over 18587.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3345, pruned_loss=0.09393, over 5667647.43 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.363, pruned_loss=0.126, over 5687688.95 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3318, pruned_loss=0.08989, over 5670184.09 frames. ], batch size: 187, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:04:56,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.083e+02 1.464e+03 1.931e+03 2.817e+03 5.535e+03, threshold=3.861e+03, percent-clipped=14.0
+2023-03-04 10:05:10,321 INFO [train.py:968] (0/2) Epoch 8, batch 33000, giga_loss[loss=0.2724, simple_loss=0.3584, pruned_loss=0.09319, over 28723.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3359, pruned_loss=0.09341, over 5651042.72 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3634, pruned_loss=0.1263, over 5675592.60 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3327, pruned_loss=0.08915, over 5662618.96 frames. ], batch size: 243, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:05:10,326 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 10:05:18,695 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2087, simple_loss=0.3078, pruned_loss=0.05482, over 944034.00 frames. 
+2023-03-04 10:05:18,696 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 10:05:21,872 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:05:24,108 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:05:27,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:06:01,405 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:06:14,092 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:06:17,626 INFO [train.py:968] (0/2) Epoch 8, batch 33050, giga_loss[loss=0.2511, simple_loss=0.3383, pruned_loss=0.08199, over 28822.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3394, pruned_loss=0.09441, over 5650088.79 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3633, pruned_loss=0.1263, over 5677915.35 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3366, pruned_loss=0.09061, over 5656728.76 frames. ], batch size: 119, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:06:25,324 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 10:06:30,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3628, 3.1625, 3.0302, 1.3945], device='cuda:0'), covar=tensor([0.0846, 0.1050, 0.1041, 0.2242], device='cuda:0'), in_proj_covar=tensor([0.0947, 0.0890, 0.0790, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 10:06:50,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:06:53,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:07:03,059 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.083e+02 1.440e+03 1.856e+03 2.497e+03 7.262e+03, threshold=3.711e+03, percent-clipped=6.0
+2023-03-04 10:07:07,588 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-04 10:07:17,060 INFO [train.py:968] (0/2) Epoch 8, batch 33100, giga_loss[loss=0.3089, simple_loss=0.3803, pruned_loss=0.1187, over 28905.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3409, pruned_loss=0.09531, over 5665543.83 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3627, pruned_loss=0.126, over 5686138.49 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3385, pruned_loss=0.09153, over 5662636.64 frames. ], batch size: 284, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:07:24,984 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:07:27,528 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 10:07:46,909 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-04 10:08:23,106 INFO [train.py:968] (0/2) Epoch 8, batch 33150, giga_loss[loss=0.2671, simple_loss=0.3393, pruned_loss=0.09739, over 27585.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3415, pruned_loss=0.09558, over 5662985.66 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3623, pruned_loss=0.1256, over 5684991.83 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3395, pruned_loss=0.09238, over 5661294.87 frames. ], batch size: 472, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:08:44,845 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-04 10:09:07,347 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.131e+02 1.295e+03 1.740e+03 2.445e+03 8.429e+03, threshold=3.480e+03, percent-clipped=11.0
+2023-03-04 10:09:23,334 INFO [train.py:968] (0/2) Epoch 8, batch 33200, giga_loss[loss=0.2404, simple_loss=0.3305, pruned_loss=0.07518, over 28701.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3384, pruned_loss=0.09302, over 5673536.57 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3623, pruned_loss=0.1256, over 5687438.27 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3367, pruned_loss=0.09022, over 5669722.75 frames. ], batch size: 262, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 10:09:40,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351992.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 10:09:47,813 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-352000.pt
+2023-03-04 10:10:24,549 INFO [train.py:968] (0/2) Epoch 8, batch 33250, giga_loss[loss=0.2336, simple_loss=0.3216, pruned_loss=0.07281, over 28465.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3362, pruned_loss=0.09122, over 5679945.54 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3622, pruned_loss=0.1255, over 5691552.41 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3344, pruned_loss=0.08845, over 5673133.32 frames. ], batch size: 336, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:10:44,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3811, 3.1067, 1.4758, 1.4194], device='cuda:0'), covar=tensor([0.0872, 0.0274, 0.0869, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0491, 0.0327, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 10:10:49,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=352054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:11:03,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.615e+02 1.189e+03 1.704e+03 2.427e+03 5.707e+03, threshold=3.408e+03, percent-clipped=9.0
+2023-03-04 10:11:05,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9418, 1.0943, 0.9977, 0.7663], device='cuda:0'), covar=tensor([0.1248, 0.1416, 0.0853, 0.1157], device='cuda:0'), in_proj_covar=tensor([0.1590, 0.1430, 0.1379, 0.1497], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 10:11:05,259 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 10:11:18,597 INFO [train.py:968] (0/2) Epoch 8, batch 33300, giga_loss[loss=0.2479, simple_loss=0.3282, pruned_loss=0.0838, over 28651.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3356, pruned_loss=0.09218, over 5678638.82 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3626, pruned_loss=0.126, over 5688789.09 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3331, pruned_loss=0.0885, over 5675097.61 frames. ], batch size: 307, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:12:17,597 INFO [train.py:968] (0/2) Epoch 8, batch 33350, giga_loss[loss=0.3348, simple_loss=0.3942, pruned_loss=0.1377, over 28181.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3358, pruned_loss=0.09271, over 5669865.64 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3619, pruned_loss=0.1257, over 5683456.49 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3337, pruned_loss=0.08911, over 5672143.23 frames. ], batch size: 412, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:12:26,711 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352135.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 10:12:29,324 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352138.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 10:12:53,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=352160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:13:03,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.105e+02 1.289e+03 1.822e+03 2.500e+03 1.008e+04, threshold=3.644e+03, percent-clipped=14.0
+2023-03-04 10:13:04,610 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=352167.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 10:13:18,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=352178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:13:20,899 INFO [train.py:968] (0/2) Epoch 8, batch 33400, giga_loss[loss=0.2826, simple_loss=0.3489, pruned_loss=0.1082, over 28960.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3388, pruned_loss=0.09438, over 5673468.85 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3617, pruned_loss=0.1255, over 5690049.09 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3365, pruned_loss=0.09068, over 5669032.36 frames. ], batch size: 106, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:13:27,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6608, 4.4884, 4.2246, 1.8099], device='cuda:0'), covar=tensor([0.0423, 0.0595, 0.0635, 0.2325], device='cuda:0'), in_proj_covar=tensor([0.0950, 0.0889, 0.0795, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 10:13:44,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:13:46,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:13:47,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=352201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:14:23,304 INFO [train.py:968] (0/2) Epoch 8, batch 33450, giga_loss[loss=0.2913, simple_loss=0.3601, pruned_loss=0.1113, over 28927.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3392, pruned_loss=0.09441, over 5678115.53 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3612, pruned_loss=0.1251, over 5692327.67 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3376, pruned_loss=0.09153, over 5672507.58 frames. ], batch size: 227, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:14:26,320 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=352229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:14:46,445 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-04 10:14:54,577 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-04 10:15:06,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4983, 1.5403, 1.1906, 1.1808], device='cuda:0'), covar=tensor([0.0636, 0.0445, 0.0898, 0.0943], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0441, 0.0499, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 10:15:12,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.146e+02 1.326e+03 1.737e+03 2.212e+03 4.846e+03, threshold=3.474e+03, percent-clipped=6.0
+2023-03-04 10:15:29,581 INFO [train.py:968] (0/2) Epoch 8, batch 33500, giga_loss[loss=0.2266, simple_loss=0.2978, pruned_loss=0.07769, over 24303.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3423, pruned_loss=0.09729, over 5664609.11 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3609, pruned_loss=0.125, over 5697967.73 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3407, pruned_loss=0.09417, over 5654211.53 frames. ], batch size: 705, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:15:58,782 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:16:03,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:16:25,147 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=352326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:16:27,731 INFO [train.py:968] (0/2) Epoch 8, batch 33550, giga_loss[loss=0.2619, simple_loss=0.3451, pruned_loss=0.0894, over 28890.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3453, pruned_loss=0.09764, over 5667414.55 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3611, pruned_loss=0.1251, over 5698645.89 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3437, pruned_loss=0.09489, over 5658466.47 frames. ], batch size: 213, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:16:34,520 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=352335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:16:44,689 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:16:47,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:17:09,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.765e+02 1.422e+03 2.008e+03 2.947e+03 6.353e+03, threshold=4.016e+03, percent-clipped=13.0
+2023-03-04 10:17:24,720 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=352376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:17:27,929 INFO [train.py:968] (0/2) Epoch 8, batch 33600, giga_loss[loss=0.2376, simple_loss=0.3208, pruned_loss=0.07723, over 29056.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3476, pruned_loss=0.09937, over 5657933.28 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3615, pruned_loss=0.1254, over 5689739.52 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3454, pruned_loss=0.09577, over 5656609.62 frames. ], batch size: 136, lr: 3.98e-03, grad_scale: 8.0
+2023-03-04 10:18:37,664 INFO [train.py:968] (0/2) Epoch 8, batch 33650, giga_loss[loss=0.2116, simple_loss=0.2761, pruned_loss=0.07356, over 24812.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.346, pruned_loss=0.09842, over 5670723.69 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3612, pruned_loss=0.1251, over 5695856.47 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.344, pruned_loss=0.09507, over 5663552.58 frames. ], batch size: 705, lr: 3.98e-03, grad_scale: 8.0
+2023-03-04 10:19:30,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.930e+02 1.370e+03 1.697e+03 2.448e+03 5.417e+03, threshold=3.394e+03, percent-clipped=3.0
+2023-03-04 10:19:44,351 INFO [train.py:968] (0/2) Epoch 8, batch 33700, giga_loss[loss=0.2454, simple_loss=0.3318, pruned_loss=0.07954, over 28858.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3442, pruned_loss=0.09742, over 5670303.84 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3615, pruned_loss=0.1255, over 5688667.10 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3422, pruned_loss=0.09407, over 5671441.21 frames. ], batch size: 243, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:20:51,465 INFO [train.py:968] (0/2) Epoch 8, batch 33750, giga_loss[loss=0.2634, simple_loss=0.3312, pruned_loss=0.09781, over 26874.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3435, pruned_loss=0.09718, over 5665705.03 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3615, pruned_loss=0.1255, over 5688415.62 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3416, pruned_loss=0.09411, over 5666839.73 frames. ], batch size: 555, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:21:23,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=352553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:21:41,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.033e+02 1.221e+03 1.603e+03 2.190e+03 5.154e+03, threshold=3.205e+03, percent-clipped=9.0
+2023-03-04 10:21:57,506 INFO [train.py:968] (0/2) Epoch 8, batch 33800, giga_loss[loss=0.2609, simple_loss=0.331, pruned_loss=0.09535, over 27864.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3414, pruned_loss=0.09715, over 5667645.27 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3614, pruned_loss=0.1255, over 5690219.50 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3396, pruned_loss=0.09418, over 5666577.29 frames. ], batch size: 474, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:22:35,981 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-04 10:22:57,094 INFO [train.py:968] (0/2) Epoch 8, batch 33850, giga_loss[loss=0.2652, simple_loss=0.3429, pruned_loss=0.09376, over 28953.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3395, pruned_loss=0.09608, over 5678950.54 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3617, pruned_loss=0.1257, over 5694244.90 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3372, pruned_loss=0.09269, over 5674019.42 frames. ], batch size: 199, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:23:30,259 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-04 10:23:42,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.674e+02 1.415e+03 1.995e+03 3.092e+03 9.341e+03, threshold=3.990e+03, percent-clipped=20.0
+2023-03-04 10:23:55,128 INFO [train.py:968] (0/2) Epoch 8, batch 33900, giga_loss[loss=0.242, simple_loss=0.3295, pruned_loss=0.07731, over 28965.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3392, pruned_loss=0.09472, over 5684541.07 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.361, pruned_loss=0.1253, over 5700443.36 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3374, pruned_loss=0.09149, over 5674350.10 frames. ], batch size: 136, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:24:14,929 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:24:18,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:24:21,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=352701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:24:51,683 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=352728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:24:53,114 INFO [train.py:968] (0/2) Epoch 8, batch 33950, giga_loss[loss=0.2536, simple_loss=0.3447, pruned_loss=0.08123, over 28922.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3379, pruned_loss=0.0929, over 5667992.08 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3613, pruned_loss=0.1256, over 5684360.00 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3357, pruned_loss=0.0893, over 5672367.48 frames. ], batch size: 227, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:24:54,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 10:25:37,663 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.070e+02 1.479e+03 1.866e+03 2.775e+03 7.494e+03, threshold=3.732e+03, percent-clipped=12.0
+2023-03-04 10:25:49,101 INFO [train.py:968] (0/2) Epoch 8, batch 34000, libri_loss[loss=0.2981, simple_loss=0.3628, pruned_loss=0.1167, over 29297.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.34, pruned_loss=0.09215, over 5676298.73 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3614, pruned_loss=0.1255, over 5688799.23 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3376, pruned_loss=0.0886, over 5675456.43 frames. ], batch size: 94, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:26:19,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0355, 4.8308, 4.5700, 2.0426], device='cuda:0'), covar=tensor([0.0380, 0.0563, 0.0659, 0.2009], device='cuda:0'), in_proj_covar=tensor([0.0952, 0.0892, 0.0786, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 10:26:45,682 INFO [train.py:968] (0/2) Epoch 8, batch 34050, giga_loss[loss=0.2711, simple_loss=0.3515, pruned_loss=0.09538, over 28669.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3408, pruned_loss=0.09242, over 5682674.10 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3608, pruned_loss=0.1252, over 5694406.00 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3388, pruned_loss=0.08897, over 5676922.64 frames. ], batch size: 262, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:26:48,485 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6951, 1.8749, 1.7548, 1.6393], device='cuda:0'), covar=tensor([0.1189, 0.1769, 0.1641, 0.1661], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0713, 0.0637, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 10:27:01,401 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:27:04,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:27:34,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.918e+02 1.325e+03 1.761e+03 2.262e+03 6.571e+03, threshold=3.523e+03, percent-clipped=8.0
+2023-03-04 10:27:35,624 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=352868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:27:45,593 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=352876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:27:49,989 INFO [train.py:968] (0/2) Epoch 8, batch 34100, giga_loss[loss=0.285, simple_loss=0.348, pruned_loss=0.111, over 26724.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3406, pruned_loss=0.09287, over 5675436.91 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3596, pruned_loss=0.1244, over 5692921.11 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3393, pruned_loss=0.08934, over 5671355.92 frames. ], batch size: 555, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:28:58,848 INFO [train.py:968] (0/2) Epoch 8, batch 34150, giga_loss[loss=0.2695, simple_loss=0.3363, pruned_loss=0.1014, over 26805.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3412, pruned_loss=0.0936, over 5669347.41 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3598, pruned_loss=0.1244, over 5696172.40 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3395, pruned_loss=0.09002, over 5662911.29 frames. ], batch size: 555, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:29:48,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.661e+02 1.423e+03 2.063e+03 3.015e+03 6.159e+03, threshold=4.127e+03, percent-clipped=19.0
+2023-03-04 10:30:00,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2603, 1.6672, 1.5767, 1.1821], device='cuda:0'), covar=tensor([0.1360, 0.1914, 0.1107, 0.1348], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0691, 0.0819, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 10:30:03,477 INFO [train.py:968] (0/2) Epoch 8, batch 34200, giga_loss[loss=0.2398, simple_loss=0.3306, pruned_loss=0.07451, over 28954.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3422, pruned_loss=0.09415, over 5664491.91 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3597, pruned_loss=0.1243, over 5691738.53 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3406, pruned_loss=0.09074, over 5662148.67 frames. ], batch size: 106, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:30:04,023 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-04 10:30:23,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0993, 1.1989, 1.3266, 1.0496], device='cuda:0'), covar=tensor([0.0987, 0.1051, 0.1632, 0.1192], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0716, 0.0637, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 10:30:27,220 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=352998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:31:11,701 INFO [train.py:968] (0/2) Epoch 8, batch 34250, libri_loss[loss=0.3307, simple_loss=0.3789, pruned_loss=0.1413, over 29193.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3426, pruned_loss=0.09401, over 5664397.56 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3597, pruned_loss=0.1245, over 5690206.75 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3406, pruned_loss=0.08985, over 5662849.39 frames. ], batch size: 97, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:32:05,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.755e+02 1.283e+03 1.611e+03 2.111e+03 4.922e+03, threshold=3.223e+03, percent-clipped=4.0
+2023-03-04 10:32:18,450 INFO [train.py:968] (0/2) Epoch 8, batch 34300, giga_loss[loss=0.3339, simple_loss=0.3928, pruned_loss=0.1375, over 27707.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3456, pruned_loss=0.0955, over 5666976.17 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3597, pruned_loss=0.1246, over 5691858.62 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3437, pruned_loss=0.09169, over 5664032.10 frames. ], batch size: 472, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:33:02,014 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=353112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:33:25,482 INFO [train.py:968] (0/2) Epoch 8, batch 34350, giga_loss[loss=0.2804, simple_loss=0.3491, pruned_loss=0.1058, over 28991.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.347, pruned_loss=0.09582, over 5667289.54 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3598, pruned_loss=0.1246, over 5689060.73 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3453, pruned_loss=0.09229, over 5667047.28 frames. ], batch size: 186, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:33:46,997 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=353146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:34:23,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.405e+02 1.427e+03 2.109e+03 2.935e+03 8.342e+03, threshold=4.218e+03, percent-clipped=19.0
+2023-03-04 10:34:35,416 INFO [train.py:968] (0/2) Epoch 8, batch 34400, giga_loss[loss=0.2672, simple_loss=0.3448, pruned_loss=0.0948, over 28396.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3443, pruned_loss=0.09463, over 5662051.42 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3597, pruned_loss=0.1245, over 5683827.32 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3427, pruned_loss=0.09128, over 5665765.51 frames. ], batch size: 368, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:34:49,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2360, 3.2612, 1.4265, 1.3395], device='cuda:0'), covar=tensor([0.0936, 0.0366, 0.0875, 0.1334], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0488, 0.0326, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 10:35:40,637 INFO [train.py:968] (0/2) Epoch 8, batch 34450, giga_loss[loss=0.2605, simple_loss=0.3445, pruned_loss=0.08826, over 28444.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.342, pruned_loss=0.09353, over 5671299.53 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3599, pruned_loss=0.1246, over 5675720.52 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.34, pruned_loss=0.0898, over 5681494.74 frames. ], batch size: 368, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:36:02,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:36:42,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.268e+02 1.172e+03 1.620e+03 2.128e+03 6.070e+03, threshold=3.240e+03, percent-clipped=4.0
+2023-03-04 10:36:52,438 INFO [train.py:968] (0/2) Epoch 8, batch 34500, libri_loss[loss=0.2885, simple_loss=0.3543, pruned_loss=0.1113, over 29663.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3403, pruned_loss=0.09196, over 5674093.46 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.36, pruned_loss=0.1246, over 5681767.35 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3381, pruned_loss=0.08814, over 5676610.06 frames. ], batch size: 88, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:37:55,362 INFO [train.py:968] (0/2) Epoch 8, batch 34550, giga_loss[loss=0.2355, simple_loss=0.3192, pruned_loss=0.07589, over 27532.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3399, pruned_loss=0.09225, over 5667854.68 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3601, pruned_loss=0.1247, over 5686758.48 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3375, pruned_loss=0.08814, over 5665447.21 frames. ], batch size: 472, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:38:03,671 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=353336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:38:42,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.414e+02 1.278e+03 1.747e+03 2.714e+03 8.761e+03, threshold=3.493e+03, percent-clipped=16.0
+2023-03-04 10:38:48,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:38:55,262 INFO [train.py:968] (0/2) Epoch 8, batch 34600, giga_loss[loss=0.2531, simple_loss=0.3421, pruned_loss=0.08199, over 28941.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3437, pruned_loss=0.09459, over 5660591.03 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3598, pruned_loss=0.1245, over 5679462.85 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3417, pruned_loss=0.09102, over 5664382.50 frames. ], batch size: 199, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:39:00,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8543, 1.9395, 1.6801, 1.7605], device='cuda:0'), covar=tensor([0.1195, 0.1969, 0.1711, 0.1650], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0713, 0.0636, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 10:39:03,308 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:39:07,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:39:26,478 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=353404.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 10:39:45,051 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:39:56,364 INFO [train.py:968] (0/2) Epoch 8, batch 34650, libri_loss[loss=0.3571, simple_loss=0.3946, pruned_loss=0.1598, over 29273.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3463, pruned_loss=0.09587, over 5672843.42 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3601, pruned_loss=0.1248, over 5682287.99 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3442, pruned_loss=0.09234, over 5673136.15 frames. ], batch size: 94, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:40:49,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.214e+02 1.384e+03 1.797e+03 2.262e+03 7.335e+03, threshold=3.593e+03, percent-clipped=6.0
+2023-03-04 10:40:58,951 INFO [train.py:968] (0/2) Epoch 8, batch 34700, giga_loss[loss=0.2388, simple_loss=0.3161, pruned_loss=0.0808, over 28578.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.344, pruned_loss=0.09576, over 5664723.77 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3604, pruned_loss=0.125, over 5683406.55 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3421, pruned_loss=0.09269, over 5663751.93 frames. ], batch size: 336, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:40:59,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=353480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:41:07,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:41:40,503 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:41:43,835 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:41:45,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:41:53,280 INFO [train.py:968] (0/2) Epoch 8, batch 34750, giga_loss[loss=0.2549, simple_loss=0.3387, pruned_loss=0.08559, over 28909.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3438, pruned_loss=0.09634, over 5661039.84 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3605, pruned_loss=0.1251, over 5678552.72 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3418, pruned_loss=0.09323, over 5663570.04 frames. ], batch size: 186, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:42:14,070 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:42:40,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.309e+02 1.551e+03 2.205e+03 4.235e+03 1.003e+04, threshold=4.411e+03, percent-clipped=28.0
+2023-03-04 10:42:52,196 INFO [train.py:968] (0/2) Epoch 8, batch 34800, giga_loss[loss=0.3111, simple_loss=0.3807, pruned_loss=0.1208, over 28773.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3457, pruned_loss=0.09823, over 5663590.23 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3604, pruned_loss=0.1249, over 5683145.71 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.344, pruned_loss=0.09541, over 5661467.29 frames. ], batch size: 243, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:43:39,979 INFO [train.py:968] (0/2) Epoch 8, batch 34850, giga_loss[loss=0.376, simple_loss=0.4293, pruned_loss=0.1613, over 28945.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3568, pruned_loss=0.105, over 5675828.28 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3598, pruned_loss=0.1246, over 5685515.53 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3558, pruned_loss=0.1028, over 5672096.99 frames. ], batch size: 145, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:43:41,708 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:43:44,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:44:14,872 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:44:16,067 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:44:19,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:44:21,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.232e+02 1.361e+03 1.771e+03 2.543e+03 1.013e+04, threshold=3.542e+03, percent-clipped=8.0
+2023-03-04 10:44:28,829 INFO [train.py:968] (0/2) Epoch 8, batch 34900, giga_loss[loss=0.3184, simple_loss=0.3827, pruned_loss=0.1271, over 28502.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3646, pruned_loss=0.1104, over 5663943.19 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3601, pruned_loss=0.1247, over 5679431.70 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3636, pruned_loss=0.1082, over 5666618.26 frames. ], batch size: 60, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:44:44,676 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:44:56,083 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:45:08,875 INFO [train.py:968] (0/2) Epoch 8, batch 34950, giga_loss[loss=0.2458, simple_loss=0.3286, pruned_loss=0.0815, over 28712.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3624, pruned_loss=0.1102, over 5673219.36 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3591, pruned_loss=0.1237, over 5684281.78 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3628, pruned_loss=0.1087, over 5670237.59 frames. ], batch size: 262, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:45:42,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.473e+02 1.208e+03 1.596e+03 2.197e+03 6.908e+03, threshold=3.193e+03, percent-clipped=6.0
+2023-03-04 10:45:50,544 INFO [train.py:968] (0/2) Epoch 8, batch 35000, giga_loss[loss=0.24, simple_loss=0.3189, pruned_loss=0.08054, over 28942.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3568, pruned_loss=0.1081, over 5686651.89 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.359, pruned_loss=0.1234, over 5687848.62 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3571, pruned_loss=0.1069, over 5681133.64 frames. ], batch size: 227, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:45:50,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353779.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 10:46:35,853 INFO [train.py:968] (0/2) Epoch 8, batch 35050, giga_loss[loss=0.2343, simple_loss=0.3053, pruned_loss=0.08167, over 28613.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3506, pruned_loss=0.1059, over 5685800.54 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3591, pruned_loss=0.1234, over 5690012.06 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3506, pruned_loss=0.1046, over 5679446.38 frames. ], batch size: 71, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:46:56,562 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:46:57,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:46:59,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:47:09,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.025e+02 9.675e+02 1.187e+03 1.809e+03 4.699e+03, threshold=2.373e+03, percent-clipped=1.0
+2023-03-04 10:47:17,449 INFO [train.py:968] (0/2) Epoch 8, batch 35100, giga_loss[loss=0.2418, simple_loss=0.3027, pruned_loss=0.09045, over 28741.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3425, pruned_loss=0.1023, over 5688912.83 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3592, pruned_loss=0.1235, over 5692919.52 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3423, pruned_loss=0.1008, over 5681165.43 frames. ], batch size: 99, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:47:22,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:47:54,573 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353922.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 10:47:56,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353925.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 10:47:58,980 INFO [train.py:968] (0/2) Epoch 8, batch 35150, giga_loss[loss=0.2444, simple_loss=0.3129, pruned_loss=0.08792, over 28426.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3361, pruned_loss=0.09949, over 5691685.08 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3599, pruned_loss=0.1239, over 5696193.36 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3349, pruned_loss=0.09759, over 5682614.21 frames. ], batch size: 71, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:48:18,609 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353954.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 10:48:32,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.784e+02 1.063e+03 1.526e+03 2.253e+03 1.869e+04, threshold=3.052e+03, percent-clipped=22.0
+2023-03-04 10:48:40,360 INFO [train.py:968] (0/2) Epoch 8, batch 35200, giga_loss[loss=0.2857, simple_loss=0.3424, pruned_loss=0.1145, over 28790.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3321, pruned_loss=0.09817, over 5699806.47 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3602, pruned_loss=0.1239, over 5704548.32 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.33, pruned_loss=0.09571, over 5685038.33 frames. ], batch size: 285, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:48:58,450 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:48:59,552 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-354000.pt
+2023-03-04 10:49:00,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=354001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:49:03,060 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:49:24,816 INFO [train.py:968] (0/2) Epoch 8, batch 35250, giga_loss[loss=0.2171, simple_loss=0.293, pruned_loss=0.07061, over 29029.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.327, pruned_loss=0.09514, over 5694103.79 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3604, pruned_loss=0.124, over 5706569.10 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3248, pruned_loss=0.09282, over 5680533.12 frames. ], batch size: 164, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:49:25,634 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=354030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:49:38,462 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:49:50,247 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:50:00,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.136e+02 9.198e+02 1.191e+03 1.601e+03 4.650e+03, threshold=2.381e+03, percent-clipped=4.0
+2023-03-04 10:50:06,402 INFO [train.py:968] (0/2) Epoch 8, batch 35300, giga_loss[loss=0.227, simple_loss=0.298, pruned_loss=0.07793, over 28891.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3228, pruned_loss=0.09272, over 5701641.55 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3602, pruned_loss=0.1239, over 5709453.44 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3204, pruned_loss=0.09031, over 5688191.49 frames. ], batch size: 112, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:50:14,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9747, 1.3218, 1.0743, 0.1677], device='cuda:0'), covar=tensor([0.2371, 0.1972, 0.3194, 0.3816], device='cuda:0'), in_proj_covar=tensor([0.1453, 0.1387, 0.1416, 0.1182], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 10:50:16,362 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:50:37,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6131, 2.0466, 1.9347, 1.4964], device='cuda:0'), covar=tensor([0.1579, 0.2101, 0.1252, 0.1457], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0703, 0.0831, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 10:50:45,304 INFO [train.py:968] (0/2) Epoch 8, batch 35350, giga_loss[loss=0.2242, simple_loss=0.2993, pruned_loss=0.07452, over 28883.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3221, pruned_loss=0.09229, over 5714322.11 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3611, pruned_loss=0.1242, over 5714370.37 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3175, pruned_loss=0.08858, over 5698687.39 frames. ], batch size: 227, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:51:24,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.613e+02 1.003e+03 1.385e+03 1.959e+03 6.648e+03, threshold=2.770e+03, percent-clipped=15.0
+2023-03-04 10:51:29,446 INFO [train.py:968] (0/2) Epoch 8, batch 35400, giga_loss[loss=0.2546, simple_loss=0.3119, pruned_loss=0.0987, over 28877.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3187, pruned_loss=0.09075, over 5696202.54 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.361, pruned_loss=0.1242, over 5698414.45 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3146, pruned_loss=0.08727, over 5699454.33 frames. ], batch size: 112, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 10:51:41,469 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5963, 1.7153, 1.4116, 1.6830], device='cuda:0'), covar=tensor([0.2259, 0.2235, 0.2435, 0.2188], device='cuda:0'), in_proj_covar=tensor([0.1213, 0.0907, 0.1082, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 10:51:47,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4587, 1.8096, 1.4988, 1.5977], device='cuda:0'), covar=tensor([0.0734, 0.0283, 0.0302, 0.0774], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 10:52:14,640 INFO [train.py:968] (0/2) Epoch 8, batch 35450, giga_loss[loss=0.2228, simple_loss=0.2943, pruned_loss=0.07571, over 28742.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.314, pruned_loss=0.08793, over 5693402.66 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3613, pruned_loss=0.1244, over 5695788.38 frames. ], giga_tot_loss[loss=0.24, simple_loss=0.3102, pruned_loss=0.08486, over 5698459.43 frames. ], batch size: 92, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 10:52:47,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.534e+02 1.034e+03 1.346e+03 2.321e+03 7.501e+03, threshold=2.692e+03, percent-clipped=17.0
+2023-03-04 10:52:54,026 INFO [train.py:968] (0/2) Epoch 8, batch 35500, giga_loss[loss=0.2227, simple_loss=0.2935, pruned_loss=0.07595, over 28618.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3107, pruned_loss=0.08594, over 5687512.23 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3613, pruned_loss=0.1243, over 5689257.11 frames. ], giga_tot_loss[loss=0.2359, simple_loss=0.3065, pruned_loss=0.08264, over 5696660.66 frames. ], batch size: 336, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 10:52:56,719 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-04 10:53:34,215 INFO [train.py:968] (0/2) Epoch 8, batch 35550, libri_loss[loss=0.3749, simple_loss=0.4164, pruned_loss=0.1667, over 19807.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3096, pruned_loss=0.08573, over 5678204.27 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3622, pruned_loss=0.1248, over 5682361.86 frames. ], giga_tot_loss[loss=0.2331, simple_loss=0.3037, pruned_loss=0.0813, over 5692280.05 frames. ], batch size: 187, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 10:54:14,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.128e+02 9.736e+02 1.313e+03 1.847e+03 9.915e+03, threshold=2.627e+03, percent-clipped=12.0
+2023-03-04 10:54:19,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=354378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:54:19,933 INFO [train.py:968] (0/2) Epoch 8, batch 35600, giga_loss[loss=0.2255, simple_loss=0.2991, pruned_loss=0.07598, over 28902.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3057, pruned_loss=0.08381, over 5686538.47 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3613, pruned_loss=0.1241, over 5686908.64 frames. ], giga_tot_loss[loss=0.2302, simple_loss=0.3004, pruned_loss=0.07995, over 5693666.09 frames. ], batch size: 227, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:54:22,615 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0370, 1.3315, 3.6658, 2.8546], device='cuda:0'), covar=tensor([0.1684, 0.2456, 0.0435, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0612, 0.0566, 0.0803, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 10:54:56,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=354420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:54:59,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1501, 1.0495, 4.3669, 3.3249], device='cuda:0'), covar=tensor([0.1701, 0.2704, 0.0334, 0.0869], device='cuda:0'), in_proj_covar=tensor([0.0611, 0.0566, 0.0801, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 10:55:05,525 INFO [train.py:968] (0/2) Epoch 8, batch 35650, libri_loss[loss=0.3145, simple_loss=0.3794, pruned_loss=0.1248, over 29744.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3101, pruned_loss=0.08663, over 5677075.75 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3613, pruned_loss=0.1241, over 5681204.96 frames. ], giga_tot_loss[loss=0.2357, simple_loss=0.3052, pruned_loss=0.08312, over 5688243.09 frames. ], batch size: 87, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:55:07,678 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=354432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:55:40,531 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=354465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:55:46,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.743e+02 1.139e+03 1.422e+03 1.816e+03 4.766e+03, threshold=2.844e+03, percent-clipped=12.0
+2023-03-04 10:55:46,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8775, 1.9342, 1.7604, 1.8160], device='cuda:0'), covar=tensor([0.1130, 0.1507, 0.1563, 0.1325], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0725, 0.0645, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 10:55:52,103 INFO [train.py:968] (0/2) Epoch 8, batch 35700, giga_loss[loss=0.3247, simple_loss=0.3909, pruned_loss=0.1293, over 28307.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3235, pruned_loss=0.0941, over 5678464.17 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3616, pruned_loss=0.1242, over 5684362.61 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3188, pruned_loss=0.09071, over 5684246.46 frames. ], batch size: 368, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:56:06,273 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-04 10:56:31,546 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=354521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:56:35,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=354524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:56:39,013 INFO [train.py:968] (0/2) Epoch 8, batch 35750, giga_loss[loss=0.3016, simple_loss=0.3822, pruned_loss=0.1105, over 28875.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3383, pruned_loss=0.1022, over 5684945.32 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3622, pruned_loss=0.1247, over 5684572.43 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3337, pruned_loss=0.09886, over 5689255.88 frames. ], batch size: 145, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:57:01,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=354553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:09,733 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=354563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:12,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=354566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:18,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.120e+02 1.224e+03 1.611e+03 2.218e+03 6.497e+03, threshold=3.221e+03, percent-clipped=12.0
+2023-03-04 10:57:21,065 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=354575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:23,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=354578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:23,714 INFO [train.py:968] (0/2) Epoch 8, batch 35800, libri_loss[loss=0.3095, simple_loss=0.3699, pruned_loss=0.1246, over 29375.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3476, pruned_loss=0.1067, over 5683367.36 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3627, pruned_loss=0.125, over 5687191.70 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3432, pruned_loss=0.1034, over 5684296.18 frames. ], batch size: 92, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:57:38,269 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=354595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:48,943 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=354607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:49,686 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=354608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:51,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=354611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:58:07,348 INFO [train.py:968] (0/2) Epoch 8, batch 35850, giga_loss[loss=0.2835, simple_loss=0.3637, pruned_loss=0.1016, over 28865.00 frames. ], tot_loss[loss=0.282, simple_loss=0.351, pruned_loss=0.1065, over 5683659.51 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3627, pruned_loss=0.125, over 5690195.90 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3473, pruned_loss=0.1036, over 5681507.42 frames. ], batch size: 174, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:58:16,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=354640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:58:46,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.265e+02 1.073e+03 1.393e+03 1.900e+03 3.988e+03, threshold=2.787e+03, percent-clipped=4.0
+2023-03-04 10:58:52,024 INFO [train.py:968] (0/2) Epoch 8, batch 35900, giga_loss[loss=0.2959, simple_loss=0.3712, pruned_loss=0.1103, over 28513.00 frames. ], tot_loss[loss=0.283, simple_loss=0.353, pruned_loss=0.1066, over 5692830.58 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3625, pruned_loss=0.1248, over 5694679.94 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.35, pruned_loss=0.1041, over 5687133.34 frames. ], batch size: 336, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:58:58,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2481, 1.2633, 1.0300, 1.0855], device='cuda:0'), covar=tensor([0.0564, 0.0379, 0.0888, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0437, 0.0495, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 10:59:13,819 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-04 10:59:35,318 INFO [train.py:968] (0/2) Epoch 8, batch 35950, giga_loss[loss=0.3015, simple_loss=0.365, pruned_loss=0.119, over 28867.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3567, pruned_loss=0.1091, over 5698336.91 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3638, pruned_loss=0.1257, over 5697899.44 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3531, pruned_loss=0.106, over 5690865.10 frames. ], batch size: 119, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:00:10,853 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:00:14,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.398e+02 1.163e+03 1.530e+03 1.998e+03 5.441e+03, threshold=3.061e+03, percent-clipped=7.0
+2023-03-04 11:00:19,562 INFO [train.py:968] (0/2) Epoch 8, batch 36000, libri_loss[loss=0.3206, simple_loss=0.3842, pruned_loss=0.1285, over 29672.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3593, pruned_loss=0.1115, over 5690713.49 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3641, pruned_loss=0.1259, over 5699493.09 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.356, pruned_loss=0.1085, over 5682708.83 frames. ], batch size: 88, lr: 3.97e-03, grad_scale: 8.0
+2023-03-04 11:00:19,566 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 11:00:28,897 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2205, simple_loss=0.3268, pruned_loss=0.05714, over 944034.00 frames. 
+2023-03-04 11:00:28,897 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 11:01:07,941 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:01:12,496 INFO [train.py:968] (0/2) Epoch 8, batch 36050, giga_loss[loss=0.2986, simple_loss=0.3727, pruned_loss=0.1122, over 28995.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.362, pruned_loss=0.1131, over 5690532.21 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3646, pruned_loss=0.1263, over 5701689.30 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3589, pruned_loss=0.1102, over 5681968.59 frames. ], batch size: 128, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:01:23,129 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-04 11:01:47,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.703e+02 1.102e+03 1.588e+03 2.450e+03 7.927e+03, threshold=3.177e+03, percent-clipped=16.0
+2023-03-04 11:01:51,317 INFO [train.py:968] (0/2) Epoch 8, batch 36100, giga_loss[loss=0.3333, simple_loss=0.3943, pruned_loss=0.1361, over 28723.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3648, pruned_loss=0.1139, over 5702124.36 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3653, pruned_loss=0.1265, over 5709887.83 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3616, pruned_loss=0.1108, over 5687260.48 frames. ], batch size: 99, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:02:32,258 INFO [train.py:968] (0/2) Epoch 8, batch 36150, giga_loss[loss=0.2692, simple_loss=0.3546, pruned_loss=0.09197, over 28882.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3675, pruned_loss=0.1149, over 5686953.81 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3663, pruned_loss=0.1271, over 5701471.10 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3641, pruned_loss=0.1116, over 5682819.67 frames. ], batch size: 174, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:03:10,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.835e+02 1.132e+03 1.485e+03 1.933e+03 6.121e+03, threshold=2.970e+03, percent-clipped=5.0
+2023-03-04 11:03:12,500 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-04 11:03:13,372 INFO [train.py:968] (0/2) Epoch 8, batch 36200, giga_loss[loss=0.3512, simple_loss=0.4145, pruned_loss=0.1439, over 28591.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3684, pruned_loss=0.1147, over 5678507.92 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3661, pruned_loss=0.127, over 5695762.68 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3657, pruned_loss=0.1119, over 5680274.66 frames. ], batch size: 307, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:03:47,916 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 11:03:51,433 INFO [train.py:968] (0/2) Epoch 8, batch 36250, giga_loss[loss=0.2925, simple_loss=0.3652, pruned_loss=0.1099, over 27973.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3697, pruned_loss=0.1145, over 5691941.76 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3672, pruned_loss=0.1276, over 5701008.51 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3667, pruned_loss=0.1113, over 5688130.38 frames. ], batch size: 412, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:04:27,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.046e+02 1.108e+03 1.327e+03 1.815e+03 4.575e+03, threshold=2.653e+03, percent-clipped=7.0
+2023-03-04 11:04:31,517 INFO [train.py:968] (0/2) Epoch 8, batch 36300, giga_loss[loss=0.2821, simple_loss=0.3636, pruned_loss=0.1003, over 28837.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3684, pruned_loss=0.1127, over 5682679.14 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3674, pruned_loss=0.1277, over 5688561.92 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3659, pruned_loss=0.1098, over 5690271.61 frames. ], batch size: 119, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:04:33,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1880, 1.8739, 1.3704, 1.6102], device='cuda:0'), covar=tensor([0.0752, 0.0781, 0.1082, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0439, 0.0498, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 11:05:13,687 INFO [train.py:968] (0/2) Epoch 8, batch 36350, giga_loss[loss=0.2881, simple_loss=0.3609, pruned_loss=0.1077, over 28933.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3647, pruned_loss=0.109, over 5697602.02 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3676, pruned_loss=0.1277, over 5691468.95 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3626, pruned_loss=0.1064, over 5701020.30 frames. ], batch size: 145, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:05:24,307 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-04 11:05:24,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:05:51,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.603e+02 1.179e+03 1.422e+03 2.158e+03 9.944e+03, threshold=2.844e+03, percent-clipped=11.0
+2023-03-04 11:05:56,444 INFO [train.py:968] (0/2) Epoch 8, batch 36400, libri_loss[loss=0.3507, simple_loss=0.4011, pruned_loss=0.1502, over 29541.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.366, pruned_loss=0.1106, over 5712554.04 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3684, pruned_loss=0.1281, over 5699759.59 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3633, pruned_loss=0.1074, over 5708062.71 frames. ], batch size: 89, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:06:13,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:06:39,999 INFO [train.py:968] (0/2) Epoch 8, batch 36450, giga_loss[loss=0.3128, simple_loss=0.3726, pruned_loss=0.1265, over 28780.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3689, pruned_loss=0.1155, over 5708548.27 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3688, pruned_loss=0.1282, over 5704216.54 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3665, pruned_loss=0.1125, over 5700798.33 frames. ], batch size: 119, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:06:40,814 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:07:01,114 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 11:07:19,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.746e+02 1.335e+03 1.776e+03 2.433e+03 1.111e+04, threshold=3.552e+03, percent-clipped=16.0
+2023-03-04 11:07:23,102 INFO [train.py:968] (0/2) Epoch 8, batch 36500, giga_loss[loss=0.2941, simple_loss=0.3614, pruned_loss=0.1134, over 28829.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3713, pruned_loss=0.1193, over 5703343.87 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3691, pruned_loss=0.1284, over 5705326.95 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3691, pruned_loss=0.1167, over 5696168.84 frames. ], batch size: 145, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:07:28,428 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1886, 1.5518, 1.4244, 1.1122], device='cuda:0'), covar=tensor([0.1339, 0.1874, 0.1070, 0.1221], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0691, 0.0821, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 11:07:31,062 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=355286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:07:33,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=355289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:07:50,567 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 11:08:00,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=355318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:08:08,825 INFO [train.py:968] (0/2) Epoch 8, batch 36550, giga_loss[loss=0.3207, simple_loss=0.3731, pruned_loss=0.1341, over 27593.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3699, pruned_loss=0.1191, over 5709272.01 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.369, pruned_loss=0.1283, over 5708325.85 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3683, pruned_loss=0.117, over 5700878.68 frames. ], batch size: 472, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:08:19,971 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=355341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:08:23,808 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:08:23,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=355344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:08:43,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2188, 1.6361, 1.5864, 1.1434], device='cuda:0'), covar=tensor([0.1326, 0.2048, 0.1093, 0.1333], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0695, 0.0824, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 11:08:45,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=355373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:08:46,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.926e+02 1.138e+03 1.539e+03 2.175e+03 5.342e+03, threshold=3.078e+03, percent-clipped=4.0
+2023-03-04 11:08:49,571 INFO [train.py:968] (0/2) Epoch 8, batch 36600, libri_loss[loss=0.2792, simple_loss=0.3493, pruned_loss=0.1046, over 29571.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3677, pruned_loss=0.1185, over 5709586.45 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3697, pruned_loss=0.1287, over 5714319.62 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3658, pruned_loss=0.1162, over 5697110.84 frames. ], batch size: 78, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:09:32,592 INFO [train.py:968] (0/2) Epoch 8, batch 36650, giga_loss[loss=0.3007, simple_loss=0.3691, pruned_loss=0.1161, over 28481.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3677, pruned_loss=0.1179, over 5701163.13 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3704, pruned_loss=0.1291, over 5705312.31 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3656, pruned_loss=0.1156, over 5698347.34 frames. ], batch size: 336, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:09:40,802 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 11:09:44,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3862, 1.6823, 1.2764, 1.6160], device='cuda:0'), covar=tensor([0.2322, 0.2168, 0.2396, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.1211, 0.0913, 0.1078, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 11:10:12,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.471e+02 1.174e+03 1.632e+03 2.435e+03 9.259e+03, threshold=3.265e+03, percent-clipped=15.0
+2023-03-04 11:10:15,557 INFO [train.py:968] (0/2) Epoch 8, batch 36700, giga_loss[loss=0.2604, simple_loss=0.3439, pruned_loss=0.08844, over 28862.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3655, pruned_loss=0.1155, over 5699207.92 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3707, pruned_loss=0.129, over 5712147.49 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3634, pruned_loss=0.1134, over 5690608.18 frames. ], batch size: 174, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:10:27,719 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6776, 1.8255, 1.6745, 1.5973], device='cuda:0'), covar=tensor([0.1531, 0.1959, 0.2014, 0.1940], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0722, 0.0643, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 11:10:37,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3090, 1.7020, 1.2882, 1.4400], device='cuda:0'), covar=tensor([0.2163, 0.2035, 0.2244, 0.1878], device='cuda:0'), in_proj_covar=tensor([0.1208, 0.0909, 0.1077, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 11:10:58,950 INFO [train.py:968] (0/2) Epoch 8, batch 36750, giga_loss[loss=0.2424, simple_loss=0.323, pruned_loss=0.08095, over 28984.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3616, pruned_loss=0.1126, over 5696389.12 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3704, pruned_loss=0.1286, over 5718425.57 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3598, pruned_loss=0.1107, over 5683215.88 frames. ], batch size: 164, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:11:09,155 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:11:17,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6494, 1.8477, 1.9370, 1.4931], device='cuda:0'), covar=tensor([0.1520, 0.2061, 0.1154, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0702, 0.0828, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 11:11:20,592 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:11:37,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.866e+02 1.101e+03 1.481e+03 2.054e+03 8.648e+03, threshold=2.963e+03, percent-clipped=4.0
+2023-03-04 11:11:42,625 INFO [train.py:968] (0/2) Epoch 8, batch 36800, giga_loss[loss=0.2356, simple_loss=0.3079, pruned_loss=0.0817, over 28923.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3561, pruned_loss=0.1099, over 5666379.17 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3704, pruned_loss=0.1286, over 5700868.54 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3543, pruned_loss=0.1078, over 5669923.70 frames. ], batch size: 186, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:11:58,239 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:12:04,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:12:22,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5430, 4.2360, 1.7910, 1.5752], device='cuda:0'), covar=tensor([0.0915, 0.0194, 0.0827, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0487, 0.0323, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 11:12:33,346 INFO [train.py:968] (0/2) Epoch 8, batch 36850, giga_loss[loss=0.265, simple_loss=0.3345, pruned_loss=0.09772, over 28291.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3499, pruned_loss=0.1067, over 5659829.18 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3707, pruned_loss=0.1285, over 5705975.73 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3478, pruned_loss=0.1046, over 5657222.58 frames. ], batch size: 368, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:13:02,945 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 11:13:19,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.012e+02 8.780e+02 1.196e+03 1.986e+03 8.842e+03, threshold=2.393e+03, percent-clipped=8.0
+2023-03-04 11:13:23,207 INFO [train.py:968] (0/2) Epoch 8, batch 36900, giga_loss[loss=0.2451, simple_loss=0.3243, pruned_loss=0.08295, over 29032.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3454, pruned_loss=0.1045, over 5660536.07 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3704, pruned_loss=0.1282, over 5711016.17 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3436, pruned_loss=0.1026, over 5652950.93 frames. ], batch size: 128, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:13:57,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:14:05,115 INFO [train.py:968] (0/2) Epoch 8, batch 36950, giga_loss[loss=0.2693, simple_loss=0.3403, pruned_loss=0.09917, over 29002.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3463, pruned_loss=0.1042, over 5675478.76 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3705, pruned_loss=0.1282, over 5714015.24 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3441, pruned_loss=0.1022, over 5665807.89 frames. ], batch size: 213, lr: 3.96e-03, grad_scale: 2.0
+2023-03-04 11:14:21,632 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=355747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:14:25,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=355750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:14:45,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.601e+02 9.525e+02 1.125e+03 1.437e+03 5.070e+03, threshold=2.250e+03, percent-clipped=8.0
+2023-03-04 11:14:47,570 INFO [train.py:968] (0/2) Epoch 8, batch 37000, libri_loss[loss=0.2848, simple_loss=0.3452, pruned_loss=0.1121, over 29355.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.346, pruned_loss=0.1039, over 5680958.40 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.371, pruned_loss=0.1284, over 5720537.52 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.343, pruned_loss=0.1013, over 5665959.07 frames. ], batch size: 71, lr: 3.96e-03, grad_scale: 2.0
+2023-03-04 11:14:47,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=355779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:15:15,765 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-04 11:15:18,400 WARNING [optim.py:389] (0/2) Scaling gradients by 0.07581179589033127, model_norm_threshold=2250.47265625
+2023-03-04 11:15:18,486 INFO [optim.py:451] (0/2) Parameter Dominanting tot_sumsq module.encoder.encoder_embed.conv.0.weight with proportion 0.99, where dominant_sumsq=(grad_sumsq*orig_rms_sq)=8.767e+08, grad_sumsq = 2.087e+10, orig_rms_sq=4.200e-02
+2023-03-04 11:15:28,939 INFO [train.py:968] (0/2) Epoch 8, batch 37050, giga_loss[loss=0.2724, simple_loss=0.3524, pruned_loss=0.09616, over 28991.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3449, pruned_loss=0.1031, over 5689965.22 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3722, pruned_loss=0.1291, over 5715888.52 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3406, pruned_loss=0.09959, over 5681414.78 frames. ], batch size: 164, lr: 3.96e-03, grad_scale: 2.0
+2023-03-04 11:15:30,882 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-04 11:15:53,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5280, 1.8021, 1.8469, 1.4459], device='cuda:0'), covar=tensor([0.1681, 0.1985, 0.1249, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0701, 0.0829, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 11:15:54,578 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=355862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:15:56,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=355865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:16:05,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.533e+02 1.076e+03 1.607e+03 2.489e+03 2.968e+04, threshold=3.215e+03, percent-clipped=30.0
+2023-03-04 11:16:07,959 INFO [train.py:968] (0/2) Epoch 8, batch 37100, libri_loss[loss=0.3413, simple_loss=0.4046, pruned_loss=0.139, over 29369.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3432, pruned_loss=0.1026, over 5702904.78 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3727, pruned_loss=0.1294, over 5717470.39 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3383, pruned_loss=0.09861, over 5693809.28 frames. ], batch size: 92, lr: 3.96e-03, grad_scale: 2.0
+2023-03-04 11:16:13,922 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:16:18,894 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=355894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:16:29,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9400, 1.7532, 1.3902, 1.5127], device='cuda:0'), covar=tensor([0.0550, 0.0464, 0.0860, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0439, 0.0499, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 11:16:34,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:16:47,190 INFO [train.py:968] (0/2) Epoch 8, batch 37150, giga_loss[loss=0.2409, simple_loss=0.3108, pruned_loss=0.08549, over 28602.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3417, pruned_loss=0.1022, over 5694369.46 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3735, pruned_loss=0.1299, over 5706455.37 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3361, pruned_loss=0.0977, over 5695661.29 frames. ], batch size: 78, lr: 3.96e-03, grad_scale: 2.0
+2023-03-04 11:16:48,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:17:20,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3864, 1.3697, 4.6101, 3.4366], device='cuda:0'), covar=tensor([0.1667, 0.2523, 0.0317, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0605, 0.0557, 0.0793, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 11:17:20,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:17:23,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.356e+02 9.719e+02 1.169e+03 1.623e+03 6.331e+03, threshold=2.338e+03, percent-clipped=4.0
+2023-03-04 11:17:25,451 INFO [train.py:968] (0/2) Epoch 8, batch 37200, giga_loss[loss=0.2647, simple_loss=0.3256, pruned_loss=0.1019, over 28842.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3383, pruned_loss=0.1001, over 5698635.78 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3741, pruned_loss=0.13, over 5703623.65 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3328, pruned_loss=0.09609, over 5702008.93 frames. ], batch size: 99, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:17:42,826 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-356000.pt
+2023-03-04 11:18:07,153 INFO [train.py:968] (0/2) Epoch 8, batch 37250, giga_loss[loss=0.3081, simple_loss=0.3645, pruned_loss=0.1258, over 28700.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3373, pruned_loss=0.0994, over 5706544.51 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3747, pruned_loss=0.1298, over 5707310.10 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3309, pruned_loss=0.09516, over 5705817.26 frames. ], batch size: 262, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:18:28,843 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=356058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:18:30,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=356061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:18:41,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=356074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:18:42,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.460e+02 9.403e+02 1.197e+03 1.757e+03 6.565e+03, threshold=2.394e+03, percent-clipped=13.0
+2023-03-04 11:18:44,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=356077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:18:47,199 INFO [train.py:968] (0/2) Epoch 8, batch 37300, giga_loss[loss=0.2402, simple_loss=0.3149, pruned_loss=0.08271, over 28825.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3347, pruned_loss=0.09791, over 5714816.60 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3754, pruned_loss=0.13, over 5711291.94 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3283, pruned_loss=0.09375, over 5710879.68 frames. ], batch size: 186, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:18:55,448 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=356090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:19:10,964 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=356106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:19:18,065 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=356114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:19:19,924 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=356117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:19:27,562 INFO [train.py:968] (0/2) Epoch 8, batch 37350, giga_loss[loss=0.2371, simple_loss=0.3145, pruned_loss=0.07982, over 28738.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3316, pruned_loss=0.09569, over 5715318.44 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3761, pruned_loss=0.1303, over 5711940.93 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3253, pruned_loss=0.09168, over 5711636.97 frames. ], batch size: 262, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:19:41,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=356146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:20:03,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.185e+02 9.095e+02 1.199e+03 1.784e+03 4.401e+03, threshold=2.397e+03, percent-clipped=16.0
+2023-03-04 11:20:05,755 INFO [train.py:968] (0/2) Epoch 8, batch 37400, giga_loss[loss=0.2601, simple_loss=0.3294, pruned_loss=0.09537, over 29070.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3318, pruned_loss=0.09575, over 5710631.62 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3775, pruned_loss=0.1309, over 5705916.57 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3239, pruned_loss=0.09076, over 5713446.75 frames. ], batch size: 136, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:20:11,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5584, 2.2319, 1.6596, 0.6293], device='cuda:0'), covar=tensor([0.3385, 0.1519, 0.2923, 0.4095], device='cuda:0'), in_proj_covar=tensor([0.1489, 0.1402, 0.1457, 0.1210], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 11:20:47,459 INFO [train.py:968] (0/2) Epoch 8, batch 37450, giga_loss[loss=0.2279, simple_loss=0.3026, pruned_loss=0.07665, over 28774.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3298, pruned_loss=0.09452, over 5702251.66 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3779, pruned_loss=0.1311, over 5698110.56 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3229, pruned_loss=0.09018, over 5710714.29 frames. ], batch size: 99, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:21:16,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=356263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:21:27,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.193e+02 1.001e+03 1.227e+03 1.566e+03 3.560e+03, threshold=2.454e+03, percent-clipped=7.0
+2023-03-04 11:21:30,775 INFO [train.py:968] (0/2) Epoch 8, batch 37500, giga_loss[loss=0.2497, simple_loss=0.3206, pruned_loss=0.08945, over 28751.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3311, pruned_loss=0.09546, over 5701054.37 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3781, pruned_loss=0.131, over 5693707.09 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3244, pruned_loss=0.09128, over 5712208.78 frames. ], batch size: 119, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:21:47,679 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=356299.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 11:22:11,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2206, 1.5276, 1.2425, 1.0591], device='cuda:0'), covar=tensor([0.2235, 0.2208, 0.2422, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.1206, 0.0911, 0.1072, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 11:22:11,702 INFO [train.py:968] (0/2) Epoch 8, batch 37550, giga_loss[loss=0.2843, simple_loss=0.3402, pruned_loss=0.1142, over 28708.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3355, pruned_loss=0.09824, over 5708833.96 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3785, pruned_loss=0.131, over 5698056.20 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3286, pruned_loss=0.09396, over 5714333.04 frames. ], batch size: 92, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:22:57,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.419e+02 1.095e+03 1.402e+03 1.893e+03 5.075e+03, threshold=2.804e+03, percent-clipped=13.0
+2023-03-04 11:22:59,744 INFO [train.py:968] (0/2) Epoch 8, batch 37600, giga_loss[loss=0.3081, simple_loss=0.3754, pruned_loss=0.1205, over 28587.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3419, pruned_loss=0.1024, over 5705676.19 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3785, pruned_loss=0.131, over 5700171.31 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3361, pruned_loss=0.09878, over 5708192.41 frames. ], batch size: 307, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:23:29,624 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=356406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:23:31,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=356409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:23:49,599 INFO [train.py:968] (0/2) Epoch 8, batch 37650, giga_loss[loss=0.2735, simple_loss=0.3555, pruned_loss=0.09571, over 28884.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3523, pruned_loss=0.11, over 5696830.16 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3787, pruned_loss=0.1311, over 5698524.24 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3472, pruned_loss=0.1068, over 5700131.63 frames. ], batch size: 227, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:23:59,368 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=356438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:24:38,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.702e+02 1.250e+03 1.552e+03 2.168e+03 5.963e+03, threshold=3.105e+03, percent-clipped=14.0
+2023-03-04 11:24:39,512 INFO [train.py:968] (0/2) Epoch 8, batch 37700, giga_loss[loss=0.3149, simple_loss=0.3773, pruned_loss=0.1262, over 28827.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3578, pruned_loss=0.1128, over 5684611.84 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3787, pruned_loss=0.131, over 5700281.33 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3533, pruned_loss=0.11, over 5685483.36 frames. ], batch size: 99, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:24:58,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5733, 1.7115, 1.4248, 1.7111], device='cuda:0'), covar=tensor([0.2209, 0.2243, 0.2379, 0.2053], device='cuda:0'), in_proj_covar=tensor([0.1207, 0.0913, 0.1073, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 11:25:24,246 INFO [train.py:968] (0/2) Epoch 8, batch 37750, giga_loss[loss=0.3037, simple_loss=0.3814, pruned_loss=0.113, over 28775.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3622, pruned_loss=0.1142, over 5688675.10 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3788, pruned_loss=0.1309, over 5702470.77 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3585, pruned_loss=0.1118, over 5687149.15 frames. ], batch size: 119, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:26:07,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.526e+02 1.178e+03 1.534e+03 2.034e+03 8.290e+03, threshold=3.069e+03, percent-clipped=10.0
+2023-03-04 11:26:09,546 INFO [train.py:968] (0/2) Epoch 8, batch 37800, giga_loss[loss=0.367, simple_loss=0.4238, pruned_loss=0.1551, over 28919.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3691, pruned_loss=0.1186, over 5675591.44 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3788, pruned_loss=0.1309, over 5692396.76 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3658, pruned_loss=0.1164, over 5682126.40 frames. ], batch size: 199, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:26:32,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5764, 4.3740, 4.1242, 1.8848], device='cuda:0'), covar=tensor([0.0442, 0.0524, 0.0573, 0.2010], device='cuda:0'), in_proj_covar=tensor([0.0945, 0.0891, 0.0791, 0.0628], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 11:26:50,488 INFO [train.py:968] (0/2) Epoch 8, batch 37850, giga_loss[loss=0.355, simple_loss=0.3862, pruned_loss=0.1619, over 26766.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3667, pruned_loss=0.1167, over 5679654.18 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3794, pruned_loss=0.1314, over 5694429.22 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3633, pruned_loss=0.1144, over 5682840.92 frames. ], batch size: 555, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:27:27,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=356674.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 11:27:28,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7465, 2.5520, 1.5644, 0.8150], device='cuda:0'), covar=tensor([0.4967, 0.2142, 0.3035, 0.4454], device='cuda:0'), in_proj_covar=tensor([0.1491, 0.1410, 0.1459, 0.1209], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 11:27:28,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.621e+02 1.109e+03 1.444e+03 1.958e+03 4.932e+03, threshold=2.889e+03, percent-clipped=8.0
+2023-03-04 11:27:31,331 INFO [train.py:968] (0/2) Epoch 8, batch 37900, giga_loss[loss=0.2223, simple_loss=0.3054, pruned_loss=0.0696, over 28860.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3613, pruned_loss=0.1122, over 5690953.80 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3797, pruned_loss=0.1317, over 5697384.13 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3582, pruned_loss=0.1098, over 5690597.52 frames. ], batch size: 186, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:28:12,428 INFO [train.py:968] (0/2) Epoch 8, batch 37950, libri_loss[loss=0.3324, simple_loss=0.3937, pruned_loss=0.1356, over 29761.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3596, pruned_loss=0.1106, over 5689434.50 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3795, pruned_loss=0.1319, over 5693482.01 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3566, pruned_loss=0.1079, over 5691838.56 frames. ], batch size: 87, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:28:52,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.281e+02 1.275e+03 1.974e+03 3.076e+03 1.108e+04, threshold=3.948e+03, percent-clipped=29.0
+2023-03-04 11:28:53,923 INFO [train.py:968] (0/2) Epoch 8, batch 38000, giga_loss[loss=0.2595, simple_loss=0.3362, pruned_loss=0.09145, over 28470.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3598, pruned_loss=0.1105, over 5694838.17 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3795, pruned_loss=0.1318, over 5693924.48 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3571, pruned_loss=0.1079, over 5696480.99 frames. ], batch size: 71, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:29:27,706 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=356817.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 11:29:29,883 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=356820.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 11:29:36,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4872, 1.7257, 1.8324, 1.3976], device='cuda:0'), covar=tensor([0.1406, 0.1906, 0.1086, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0700, 0.0829, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 11:29:37,197 INFO [train.py:968] (0/2) Epoch 8, batch 38050, giga_loss[loss=0.3178, simple_loss=0.3846, pruned_loss=0.1255, over 28938.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3634, pruned_loss=0.1125, over 5696103.25 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3796, pruned_loss=0.1319, over 5696423.39 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3608, pruned_loss=0.11, over 5695137.53 frames. ], batch size: 186, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:29:54,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=356849.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 11:30:18,728 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.500e+02 1.246e+03 1.409e+03 1.779e+03 4.531e+03, threshold=2.819e+03, percent-clipped=1.0
+2023-03-04 11:30:21,013 INFO [train.py:968] (0/2) Epoch 8, batch 38100, libri_loss[loss=0.3719, simple_loss=0.4267, pruned_loss=0.1585, over 29531.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3656, pruned_loss=0.1141, over 5700853.63 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3796, pruned_loss=0.1319, over 5700625.04 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3631, pruned_loss=0.1117, over 5696363.02 frames. ], batch size: 81, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:30:35,446 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=356896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:31:05,492 INFO [train.py:968] (0/2) Epoch 8, batch 38150, giga_loss[loss=0.2904, simple_loss=0.3607, pruned_loss=0.11, over 28633.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3679, pruned_loss=0.1163, over 5697141.18 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3799, pruned_loss=0.1321, over 5704669.17 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3654, pruned_loss=0.1138, over 5690016.16 frames. ], batch size: 307, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:31:11,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3600, 1.7079, 1.6680, 1.2635], device='cuda:0'), covar=tensor([0.1448, 0.1947, 0.1146, 0.1299], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0700, 0.0827, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 11:31:13,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3018, 1.8324, 1.3575, 0.5757], device='cuda:0'), covar=tensor([0.2607, 0.1348, 0.2308, 0.3250], device='cuda:0'), in_proj_covar=tensor([0.1475, 0.1392, 0.1443, 0.1203], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 11:31:46,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.865e+02 1.162e+03 1.564e+03 1.987e+03 5.402e+03, threshold=3.128e+03, percent-clipped=8.0
+2023-03-04 11:31:48,152 INFO [train.py:968] (0/2) Epoch 8, batch 38200, giga_loss[loss=0.2893, simple_loss=0.3619, pruned_loss=0.1083, over 28857.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3699, pruned_loss=0.1183, over 5704326.81 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3804, pruned_loss=0.1322, over 5707466.43 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3674, pruned_loss=0.1161, over 5696133.19 frames. ], batch size: 119, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:32:11,886 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 11:32:28,679 INFO [train.py:968] (0/2) Epoch 8, batch 38250, giga_loss[loss=0.2829, simple_loss=0.3536, pruned_loss=0.1061, over 28952.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3698, pruned_loss=0.1182, over 5699720.96 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3804, pruned_loss=0.1321, over 5712351.55 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3674, pruned_loss=0.1162, over 5688944.93 frames. ], batch size: 213, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:33:06,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.784e+02 1.039e+03 1.398e+03 2.012e+03 7.714e+03, threshold=2.797e+03, percent-clipped=13.0
+2023-03-04 11:33:07,865 INFO [train.py:968] (0/2) Epoch 8, batch 38300, giga_loss[loss=0.2749, simple_loss=0.3534, pruned_loss=0.09816, over 28851.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3702, pruned_loss=0.1181, over 5706495.06 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.381, pruned_loss=0.1325, over 5716219.65 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3677, pruned_loss=0.1159, over 5694408.93 frames. ], batch size: 92, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:33:47,736 INFO [train.py:968] (0/2) Epoch 8, batch 38350, giga_loss[loss=0.2522, simple_loss=0.3326, pruned_loss=0.08586, over 28804.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.37, pruned_loss=0.1168, over 5708828.08 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.381, pruned_loss=0.1324, over 5720231.33 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3678, pruned_loss=0.1148, over 5695372.87 frames. ], batch size: 99, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:34:26,873 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.428e+02 1.022e+03 1.370e+03 2.227e+03 1.076e+04, threshold=2.741e+03, percent-clipped=17.0
+2023-03-04 11:34:26,886 INFO [train.py:968] (0/2) Epoch 8, batch 38400, giga_loss[loss=0.3433, simple_loss=0.4083, pruned_loss=0.1392, over 28601.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3707, pruned_loss=0.1162, over 5717182.69 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3817, pruned_loss=0.1329, over 5722614.31 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3682, pruned_loss=0.114, over 5704407.89 frames. ], batch size: 85, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:35:08,115 INFO [train.py:968] (0/2) Epoch 8, batch 38450, libri_loss[loss=0.339, simple_loss=0.3901, pruned_loss=0.1439, over 29671.00 frames. ], tot_loss[loss=0.299, simple_loss=0.368, pruned_loss=0.1151, over 5710923.80 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3815, pruned_loss=0.1328, over 5721021.04 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3656, pruned_loss=0.1127, over 5702290.69 frames. ], batch size: 88, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:35:43,598 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=357271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:35:48,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.347e+02 1.028e+03 1.224e+03 1.627e+03 3.446e+03, threshold=2.448e+03, percent-clipped=6.0
+2023-03-04 11:35:48,729 INFO [train.py:968] (0/2) Epoch 8, batch 38500, giga_loss[loss=0.2986, simple_loss=0.3706, pruned_loss=0.1133, over 29001.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.366, pruned_loss=0.1142, over 5704619.51 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3814, pruned_loss=0.1327, over 5716124.53 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3638, pruned_loss=0.1118, over 5701598.15 frames. ], batch size: 145, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:36:25,800 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 11:36:27,296 INFO [train.py:968] (0/2) Epoch 8, batch 38550, giga_loss[loss=0.2952, simple_loss=0.3655, pruned_loss=0.1125, over 28955.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3636, pruned_loss=0.1125, over 5702217.51 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3818, pruned_loss=0.1328, over 5713661.62 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3612, pruned_loss=0.1102, over 5701480.45 frames. ], batch size: 227, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:37:05,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.193e+02 1.036e+03 1.271e+03 1.797e+03 4.155e+03, threshold=2.542e+03, percent-clipped=9.0
+2023-03-04 11:37:05,882 INFO [train.py:968] (0/2) Epoch 8, batch 38600, giga_loss[loss=0.3192, simple_loss=0.3826, pruned_loss=0.1279, over 28766.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3627, pruned_loss=0.112, over 5703966.05 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3818, pruned_loss=0.1328, over 5714601.96 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3603, pruned_loss=0.1097, over 5702540.34 frames. ], batch size: 119, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:37:06,069 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=357379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:37:33,701 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=357414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:37:35,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=357417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:37:38,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4469, 1.8445, 1.4787, 1.6553], device='cuda:0'), covar=tensor([0.0647, 0.0247, 0.0267, 0.0648], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0117, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0051, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 11:37:44,357 INFO [train.py:968] (0/2) Epoch 8, batch 38650, giga_loss[loss=0.2903, simple_loss=0.3582, pruned_loss=0.1112, over 28830.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3621, pruned_loss=0.1112, over 5710340.79 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.3814, pruned_loss=0.1323, over 5721949.06 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3598, pruned_loss=0.109, over 5702143.81 frames. ], batch size: 119, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:37:58,017 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=357446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:38:02,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0580, 3.8810, 3.6419, 1.7813], device='cuda:0'), covar=tensor([0.0536, 0.0642, 0.0700, 0.2120], device='cuda:0'), in_proj_covar=tensor([0.0963, 0.0904, 0.0801, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 11:38:21,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.417e+02 1.059e+03 1.313e+03 1.641e+03 5.515e+03, threshold=2.627e+03, percent-clipped=7.0
+2023-03-04 11:38:21,976 INFO [train.py:968] (0/2) Epoch 8, batch 38700, giga_loss[loss=0.2542, simple_loss=0.3381, pruned_loss=0.08516, over 28770.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3631, pruned_loss=0.1113, over 5714041.26 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3813, pruned_loss=0.1322, over 5724372.11 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3611, pruned_loss=0.1094, over 5705191.77 frames. ], batch size: 99, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:38:22,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5416, 2.1170, 1.4449, 0.6933], device='cuda:0'), covar=tensor([0.3623, 0.1832, 0.2975, 0.4090], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1389, 0.1441, 0.1207], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 11:38:27,537 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=357486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:38:58,493 INFO [train.py:968] (0/2) Epoch 8, batch 38750, libri_loss[loss=0.3521, simple_loss=0.4, pruned_loss=0.1521, over 29549.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.363, pruned_loss=0.1109, over 5709296.52 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.381, pruned_loss=0.132, over 5716755.56 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3607, pruned_loss=0.1085, over 5709071.25 frames. ], batch size: 89, lr: 3.95e-03, grad_scale: 2.0
+2023-03-04 11:39:35,552 INFO [train.py:968] (0/2) Epoch 8, batch 38800, giga_loss[loss=0.2737, simple_loss=0.35, pruned_loss=0.09867, over 28874.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3621, pruned_loss=0.11, over 5715917.05 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3809, pruned_loss=0.1319, over 5720341.59 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3601, pruned_loss=0.1077, over 5712319.43 frames. ], batch size: 199, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:39:36,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.920e+02 1.204e+03 1.611e+03 2.232e+03 7.377e+03, threshold=3.221e+03, percent-clipped=13.0
+2023-03-04 11:40:08,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2038, 1.6253, 1.5645, 1.1495], device='cuda:0'), covar=tensor([0.1511, 0.2201, 0.1205, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0705, 0.0832, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 11:40:14,323 INFO [train.py:968] (0/2) Epoch 8, batch 38850, giga_loss[loss=0.3259, simple_loss=0.3808, pruned_loss=0.1355, over 27587.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3612, pruned_loss=0.1103, over 5699399.93 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3807, pruned_loss=0.1318, over 5715477.42 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3589, pruned_loss=0.1077, over 5699974.67 frames. ], batch size: 472, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:40:52,036 INFO [train.py:968] (0/2) Epoch 8, batch 38900, giga_loss[loss=0.2653, simple_loss=0.3349, pruned_loss=0.09785, over 28884.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3593, pruned_loss=0.1099, over 5696628.37 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3813, pruned_loss=0.1325, over 5712291.82 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3563, pruned_loss=0.1065, over 5700157.97 frames. ], batch size: 112, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:40:52,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.715e+02 1.029e+03 1.416e+03 2.500e+03 1.364e+04, threshold=2.831e+03, percent-clipped=16.0
+2023-03-04 11:41:09,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9526, 2.5406, 2.0364, 2.2514], device='cuda:0'), covar=tensor([0.0556, 0.0574, 0.0833, 0.0882], device='cuda:0'), in_proj_covar=tensor([0.0338, 0.0430, 0.0491, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 11:41:30,216 INFO [train.py:968] (0/2) Epoch 8, batch 38950, giga_loss[loss=0.2609, simple_loss=0.3382, pruned_loss=0.09178, over 28980.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.355, pruned_loss=0.1077, over 5703343.03 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.381, pruned_loss=0.1323, over 5715931.10 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3525, pruned_loss=0.1047, over 5702517.23 frames. ], batch size: 213, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:41:49,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=357754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:42:08,306 INFO [train.py:968] (0/2) Epoch 8, batch 39000, giga_loss[loss=0.2783, simple_loss=0.3471, pruned_loss=0.1048, over 28740.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3548, pruned_loss=0.1081, over 5707334.70 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.3812, pruned_loss=0.1324, over 5720284.59 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3519, pruned_loss=0.105, over 5702573.10 frames. ], batch size: 99, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:42:08,310 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 11:42:16,877 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2227, simple_loss=0.3279, pruned_loss=0.05873, over 944034.00 frames. 
+2023-03-04 11:42:16,878 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 11:42:18,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.018e+02 1.090e+03 1.315e+03 1.838e+03 4.349e+03, threshold=2.631e+03, percent-clipped=5.0
+2023-03-04 11:42:57,041 INFO [train.py:968] (0/2) Epoch 8, batch 39050, giga_loss[loss=0.2438, simple_loss=0.3186, pruned_loss=0.08449, over 29006.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3535, pruned_loss=0.1074, over 5717015.23 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.381, pruned_loss=0.1322, over 5724687.73 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3507, pruned_loss=0.1045, over 5709097.50 frames. ], batch size: 213, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:43:21,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=357861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:43:37,096 INFO [train.py:968] (0/2) Epoch 8, batch 39100, libri_loss[loss=0.2659, simple_loss=0.333, pruned_loss=0.09945, over 29498.00 frames. ], tot_loss[loss=0.282, simple_loss=0.351, pruned_loss=0.1065, over 5713482.27 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3805, pruned_loss=0.1319, over 5725928.27 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3487, pruned_loss=0.104, over 5705860.42 frames. ], batch size: 70, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:43:37,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.131e+02 1.051e+03 1.335e+03 1.681e+03 3.317e+03, threshold=2.671e+03, percent-clipped=7.0
+2023-03-04 11:43:51,090 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=357897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:43:53,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=357900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:44:15,316 INFO [train.py:968] (0/2) Epoch 8, batch 39150, libri_loss[loss=0.3226, simple_loss=0.3817, pruned_loss=0.1318, over 29554.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3496, pruned_loss=0.1062, over 5716997.56 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3805, pruned_loss=0.1318, over 5728862.43 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3467, pruned_loss=0.1033, over 5707660.13 frames. ], batch size: 77, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:44:15,637 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=357929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:44:51,768 INFO [train.py:968] (0/2) Epoch 8, batch 39200, giga_loss[loss=0.2491, simple_loss=0.3213, pruned_loss=0.08841, over 28730.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3467, pruned_loss=0.1047, over 5712557.37 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.381, pruned_loss=0.132, over 5725834.03 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3432, pruned_loss=0.1017, over 5707830.40 frames. ], batch size: 119, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:44:53,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.074e+02 1.072e+03 1.333e+03 2.055e+03 9.512e+03, threshold=2.666e+03, percent-clipped=11.0
+2023-03-04 11:45:11,632 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-358000.pt
+2023-03-04 11:45:14,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:45:16,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:45:22,021 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:45:35,685 INFO [train.py:968] (0/2) Epoch 8, batch 39250, giga_loss[loss=0.2495, simple_loss=0.3176, pruned_loss=0.09068, over 29032.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3467, pruned_loss=0.1051, over 5710849.23 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3816, pruned_loss=0.1325, over 5728590.91 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3428, pruned_loss=0.1018, over 5704539.37 frames. ], batch size: 136, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:45:41,180 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:45:53,357 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7397, 1.5544, 1.3242, 1.3785], device='cuda:0'), covar=tensor([0.0500, 0.0433, 0.0830, 0.0748], device='cuda:0'), in_proj_covar=tensor([0.0340, 0.0436, 0.0495, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 11:45:59,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9938, 1.2419, 1.3049, 1.1645], device='cuda:0'), covar=tensor([0.1286, 0.1091, 0.1752, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0722, 0.0651, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 11:46:09,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7892, 2.3401, 2.0860, 1.6441], device='cuda:0'), covar=tensor([0.1420, 0.1759, 0.1164, 0.1391], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0699, 0.0827, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 11:46:18,259 INFO [train.py:968] (0/2) Epoch 8, batch 39300, giga_loss[loss=0.2762, simple_loss=0.3573, pruned_loss=0.09751, over 28248.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3488, pruned_loss=0.1058, over 5710659.18 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3818, pruned_loss=0.1326, over 5728818.91 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.345, pruned_loss=0.1028, over 5705117.77 frames. ], batch size: 368, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:46:19,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.285e+02 9.273e+02 1.146e+03 1.654e+03 3.926e+03, threshold=2.291e+03, percent-clipped=5.0
+2023-03-04 11:46:31,911 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-04 11:47:02,554 INFO [train.py:968] (0/2) Epoch 8, batch 39350, giga_loss[loss=0.2982, simple_loss=0.3578, pruned_loss=0.1192, over 28595.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3504, pruned_loss=0.1057, over 5714166.10 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3813, pruned_loss=0.1322, over 5731176.41 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3475, pruned_loss=0.1032, over 5707466.46 frames. ], batch size: 78, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:47:17,442 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:47:42,306 INFO [train.py:968] (0/2) Epoch 8, batch 39400, giga_loss[loss=0.2547, simple_loss=0.3389, pruned_loss=0.08524, over 29083.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3541, pruned_loss=0.1077, over 5709740.57 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3819, pruned_loss=0.1327, over 5735018.48 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.35, pruned_loss=0.1043, over 5700235.92 frames. ], batch size: 136, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:47:43,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.331e+02 1.049e+03 1.379e+03 1.955e+03 6.436e+03, threshold=2.758e+03, percent-clipped=13.0
+2023-03-04 11:48:21,810 INFO [train.py:968] (0/2) Epoch 8, batch 39450, giga_loss[loss=0.2617, simple_loss=0.3303, pruned_loss=0.09655, over 28695.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3557, pruned_loss=0.108, over 5698957.18 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3816, pruned_loss=0.1327, over 5726920.38 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3518, pruned_loss=0.1044, over 5697777.82 frames. ], batch size: 99, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:49:03,824 INFO [train.py:968] (0/2) Epoch 8, batch 39500, giga_loss[loss=0.2727, simple_loss=0.3391, pruned_loss=0.1032, over 23908.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3557, pruned_loss=0.1076, over 5693400.74 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3817, pruned_loss=0.1327, over 5732225.93 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3518, pruned_loss=0.1041, over 5686904.14 frames. ], batch size: 705, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:49:05,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.358e+02 1.072e+03 1.373e+03 2.020e+03 9.466e+03, threshold=2.746e+03, percent-clipped=11.0
+2023-03-04 11:49:42,030 INFO [train.py:968] (0/2) Epoch 8, batch 39550, giga_loss[loss=0.2895, simple_loss=0.366, pruned_loss=0.1065, over 28720.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.355, pruned_loss=0.1066, over 5702513.65 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3815, pruned_loss=0.1325, over 5732919.14 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3517, pruned_loss=0.1036, over 5696395.81 frames. ], batch size: 262, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:49:45,092 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:49:51,400 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2552, 1.7755, 1.3205, 0.5684], device='cuda:0'), covar=tensor([0.2238, 0.1127, 0.1766, 0.2722], device='cuda:0'), in_proj_covar=tensor([0.1494, 0.1405, 0.1447, 0.1204], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 11:49:55,940 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-04 11:50:10,420 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:50:23,255 INFO [train.py:968] (0/2) Epoch 8, batch 39600, giga_loss[loss=0.2932, simple_loss=0.3581, pruned_loss=0.1142, over 28823.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3569, pruned_loss=0.1084, over 5701035.34 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3817, pruned_loss=0.1325, over 5735666.53 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3533, pruned_loss=0.1053, over 5692457.11 frames. ], batch size: 186, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:50:24,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.131e+02 1.104e+03 1.460e+03 2.110e+03 6.449e+03, threshold=2.919e+03, percent-clipped=13.0
+2023-03-04 11:50:29,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:50:37,393 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:51:03,879 INFO [train.py:968] (0/2) Epoch 8, batch 39650, libri_loss[loss=0.2592, simple_loss=0.3329, pruned_loss=0.09276, over 29551.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3589, pruned_loss=0.1101, over 5700242.09 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3818, pruned_loss=0.1325, over 5740346.08 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3552, pruned_loss=0.1068, over 5688041.30 frames. ], batch size: 77, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:51:15,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3621, 1.6908, 1.4480, 1.5042], device='cuda:0'), covar=tensor([0.0720, 0.0286, 0.0305, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 11:51:44,105 INFO [train.py:968] (0/2) Epoch 8, batch 39700, giga_loss[loss=0.2692, simple_loss=0.3434, pruned_loss=0.09746, over 28459.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3611, pruned_loss=0.1112, over 5710278.25 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.382, pruned_loss=0.1326, over 5742951.68 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3576, pruned_loss=0.1082, over 5697750.45 frames. ], batch size: 65, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:51:45,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.825e+02 1.131e+03 1.373e+03 1.711e+03 5.026e+03, threshold=2.746e+03, percent-clipped=8.0
+2023-03-04 11:51:53,635 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-04 11:52:01,486 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.20 vs. limit=5.0
+2023-03-04 11:52:17,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:52:23,261 INFO [train.py:968] (0/2) Epoch 8, batch 39750, giga_loss[loss=0.2724, simple_loss=0.3498, pruned_loss=0.09747, over 28987.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3633, pruned_loss=0.1122, over 5718025.90 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3823, pruned_loss=0.1327, over 5747058.88 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3599, pruned_loss=0.1092, over 5703789.13 frames. ], batch size: 213, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:52:24,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:52:27,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:52:49,691 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:53:01,844 INFO [train.py:968] (0/2) Epoch 8, batch 39800, giga_loss[loss=0.3691, simple_loss=0.4139, pruned_loss=0.1621, over 26764.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3633, pruned_loss=0.1116, over 5719726.87 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3822, pruned_loss=0.1327, over 5747675.20 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3602, pruned_loss=0.1088, over 5706925.66 frames. ], batch size: 555, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:53:03,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.680e+02 1.191e+03 1.427e+03 1.995e+03 5.381e+03, threshold=2.853e+03, percent-clipped=12.0
+2023-03-04 11:53:07,460 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:53:41,673 INFO [train.py:968] (0/2) Epoch 8, batch 39850, giga_loss[loss=0.2949, simple_loss=0.3715, pruned_loss=0.1092, over 28788.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3648, pruned_loss=0.1128, over 5706583.81 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.383, pruned_loss=0.1331, over 5730511.55 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3609, pruned_loss=0.1094, over 5711089.09 frames. ], batch size: 243, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:54:11,006 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:54:13,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:54:20,805 INFO [train.py:968] (0/2) Epoch 8, batch 39900, giga_loss[loss=0.3257, simple_loss=0.3931, pruned_loss=0.1292, over 28976.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3645, pruned_loss=0.1126, over 5710814.36 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3833, pruned_loss=0.1332, over 5731613.00 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.361, pruned_loss=0.1097, over 5713208.49 frames. ], batch size: 164, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:54:23,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.807e+02 1.299e+03 1.616e+03 2.191e+03 4.040e+03, threshold=3.232e+03, percent-clipped=9.0
+2023-03-04 11:54:25,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7453, 1.6011, 1.2613, 1.3303], device='cuda:0'), covar=tensor([0.0645, 0.0629, 0.0956, 0.0959], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0448, 0.0500, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 11:54:37,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9393, 1.0945, 3.5095, 3.0361], device='cuda:0'), covar=tensor([0.1727, 0.2558, 0.0412, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0611, 0.0560, 0.0807, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 11:54:37,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:54:43,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4138, 3.0474, 1.5920, 1.5380], device='cuda:0'), covar=tensor([0.0711, 0.0274, 0.0689, 0.1037], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0494, 0.0324, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 11:54:45,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:54:49,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3636, 1.7116, 1.2793, 1.7885], device='cuda:0'), covar=tensor([0.2117, 0.2026, 0.2357, 0.1853], device='cuda:0'), in_proj_covar=tensor([0.1207, 0.0912, 0.1073, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 11:55:00,373 INFO [train.py:968] (0/2) Epoch 8, batch 39950, giga_loss[loss=0.2708, simple_loss=0.3385, pruned_loss=0.1016, over 28726.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3617, pruned_loss=0.1109, over 5712158.51 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3837, pruned_loss=0.1335, over 5734766.60 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3583, pruned_loss=0.1079, over 5710683.53 frames. ], batch size: 92, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:55:05,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:55:34,568 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:55:38,475 INFO [train.py:968] (0/2) Epoch 8, batch 40000, libri_loss[loss=0.3328, simple_loss=0.3927, pruned_loss=0.1364, over 29196.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3614, pruned_loss=0.1114, over 5704516.12 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3842, pruned_loss=0.134, over 5729399.59 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.357, pruned_loss=0.1074, over 5706940.05 frames. ], batch size: 101, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:55:41,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.435e+02 1.151e+03 1.532e+03 2.203e+03 1.239e+04, threshold=3.065e+03, percent-clipped=9.0
+2023-03-04 11:56:22,096 INFO [train.py:968] (0/2) Epoch 8, batch 40050, giga_loss[loss=0.2355, simple_loss=0.3115, pruned_loss=0.07978, over 28552.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3562, pruned_loss=0.108, over 5714908.09 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3841, pruned_loss=0.1338, over 5731332.94 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3526, pruned_loss=0.1048, over 5714895.72 frames. ], batch size: 85, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:56:37,408 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:56:39,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:56:52,907 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-04 11:56:59,403 INFO [train.py:968] (0/2) Epoch 8, batch 40100, giga_loss[loss=0.288, simple_loss=0.3709, pruned_loss=0.1025, over 28929.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3558, pruned_loss=0.1073, over 5712740.45 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3837, pruned_loss=0.1336, over 5732934.75 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3525, pruned_loss=0.1043, over 5710893.75 frames. ], batch size: 145, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:57:00,768 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:02,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.940e+02 9.567e+02 1.275e+03 1.739e+03 6.944e+03, threshold=2.551e+03, percent-clipped=7.0
+2023-03-04 11:57:02,623 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:02,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:23,918 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:28,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:30,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:37,355 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-04 11:57:40,969 INFO [train.py:968] (0/2) Epoch 8, batch 40150, giga_loss[loss=0.2701, simple_loss=0.3384, pruned_loss=0.1009, over 24007.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3567, pruned_loss=0.1067, over 5696876.63 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3837, pruned_loss=0.1336, over 5725613.98 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3537, pruned_loss=0.1039, over 5700823.65 frames. ], batch size: 705, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:57:55,427 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:55,697 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-04 11:58:06,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1198, 1.1827, 3.8308, 3.0603], device='cuda:0'), covar=tensor([0.1597, 0.2525, 0.0377, 0.1526], device='cuda:0'), in_proj_covar=tensor([0.0606, 0.0553, 0.0802, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 11:58:07,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:58:20,621 INFO [train.py:968] (0/2) Epoch 8, batch 40200, giga_loss[loss=0.2467, simple_loss=0.3183, pruned_loss=0.08755, over 28581.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3571, pruned_loss=0.1068, over 5703811.38 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3836, pruned_loss=0.1334, over 5727179.19 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3547, pruned_loss=0.1045, over 5705283.92 frames. ], batch size: 71, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:58:23,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.502e+02 9.918e+02 1.221e+03 1.506e+03 2.842e+03, threshold=2.442e+03, percent-clipped=3.0
+2023-03-04 11:58:38,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-04 11:59:00,251 INFO [train.py:968] (0/2) Epoch 8, batch 40250, giga_loss[loss=0.2795, simple_loss=0.3486, pruned_loss=0.1052, over 28727.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3564, pruned_loss=0.1077, over 5706440.14 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3838, pruned_loss=0.1336, over 5727400.85 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3538, pruned_loss=0.1053, over 5707292.30 frames. ], batch size: 262, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:59:28,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6941, 1.6174, 1.2531, 1.3825], device='cuda:0'), covar=tensor([0.0642, 0.0571, 0.0976, 0.0934], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0447, 0.0499, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 11:59:36,678 INFO [train.py:968] (0/2) Epoch 8, batch 40300, libri_loss[loss=0.3165, simple_loss=0.3636, pruned_loss=0.1347, over 29645.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3558, pruned_loss=0.1091, over 5709867.47 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.384, pruned_loss=0.134, over 5721944.74 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3524, pruned_loss=0.1058, over 5715579.93 frames. ], batch size: 73, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:59:40,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.657e+02 1.107e+03 1.378e+03 1.968e+03 8.053e+03, threshold=2.756e+03, percent-clipped=18.0
+2023-03-04 11:59:57,582 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=359104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:00:00,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=359107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:00:18,250 INFO [train.py:968] (0/2) Epoch 8, batch 40350, giga_loss[loss=0.2634, simple_loss=0.3279, pruned_loss=0.09951, over 28330.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3534, pruned_loss=0.1088, over 5706677.56 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.384, pruned_loss=0.1339, over 5725345.40 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3501, pruned_loss=0.1058, over 5707991.83 frames. ], batch size: 77, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 12:00:22,809 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=359136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:00:22,824 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=359136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:00:56,324 INFO [train.py:968] (0/2) Epoch 8, batch 40400, giga_loss[loss=0.2664, simple_loss=0.336, pruned_loss=0.09837, over 29111.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3522, pruned_loss=0.1086, over 5704604.07 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3844, pruned_loss=0.134, over 5725932.58 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3486, pruned_loss=0.1056, over 5704691.37 frames. ], batch size: 113, lr: 3.94e-03, grad_scale: 8.0
+2023-03-04 12:00:58,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.756e+02 1.047e+03 1.355e+03 1.765e+03 5.577e+03, threshold=2.710e+03, percent-clipped=10.0
+2023-03-04 12:01:15,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2841, 1.6137, 1.2564, 1.2903], device='cuda:0'), covar=tensor([0.2233, 0.2200, 0.2494, 0.1961], device='cuda:0'), in_proj_covar=tensor([0.1211, 0.0910, 0.1078, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 12:01:34,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4503, 1.6403, 1.5331, 1.5549], device='cuda:0'), covar=tensor([0.1284, 0.1798, 0.1754, 0.1737], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0733, 0.0659, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 12:01:35,119 INFO [train.py:968] (0/2) Epoch 8, batch 40450, giga_loss[loss=0.2712, simple_loss=0.3414, pruned_loss=0.1005, over 28358.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3509, pruned_loss=0.1084, over 5695564.17 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3845, pruned_loss=0.1342, over 5719784.62 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3474, pruned_loss=0.1054, over 5700776.50 frames. ], batch size: 368, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:02:02,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1254, 1.5302, 1.1895, 0.9092], device='cuda:0'), covar=tensor([0.2061, 0.1992, 0.2210, 0.1693], device='cuda:0'), in_proj_covar=tensor([0.1218, 0.0917, 0.1081, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 12:02:15,133 INFO [train.py:968] (0/2) Epoch 8, batch 40500, giga_loss[loss=0.2679, simple_loss=0.3364, pruned_loss=0.09964, over 28776.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3474, pruned_loss=0.1065, over 5703053.13 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3845, pruned_loss=0.1342, over 5721354.95 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.344, pruned_loss=0.1036, over 5705503.97 frames. ], batch size: 262, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:02:18,143 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.564e+02 1.250e+03 1.472e+03 2.263e+03 9.982e+03, threshold=2.944e+03, percent-clipped=22.0
+2023-03-04 12:02:52,966 INFO [train.py:968] (0/2) Epoch 8, batch 40550, giga_loss[loss=0.268, simple_loss=0.3364, pruned_loss=0.09983, over 28939.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3429, pruned_loss=0.104, over 5705134.06 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3843, pruned_loss=0.1341, over 5724205.45 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3391, pruned_loss=0.1009, over 5704424.23 frames. ], batch size: 145, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:02:54,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3297, 1.7377, 1.6333, 1.2194], device='cuda:0'), covar=tensor([0.1462, 0.1982, 0.1199, 0.1365], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0700, 0.0825, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 12:03:29,174 INFO [train.py:968] (0/2) Epoch 8, batch 40600, giga_loss[loss=0.2944, simple_loss=0.3613, pruned_loss=0.1137, over 28600.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3416, pruned_loss=0.103, over 5708334.40 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3839, pruned_loss=0.1338, over 5718119.85 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3382, pruned_loss=0.1002, over 5712584.55 frames. ], batch size: 307, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:03:33,208 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.072e+02 1.048e+03 1.290e+03 1.645e+03 5.969e+03, threshold=2.579e+03, percent-clipped=3.0
+2023-03-04 12:04:10,729 INFO [train.py:968] (0/2) Epoch 8, batch 40650, giga_loss[loss=0.2765, simple_loss=0.345, pruned_loss=0.1039, over 29023.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.346, pruned_loss=0.1053, over 5707281.23 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3841, pruned_loss=0.1338, over 5722510.81 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3422, pruned_loss=0.1024, over 5706400.05 frames. ], batch size: 128, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:04:52,537 INFO [train.py:968] (0/2) Epoch 8, batch 40700, giga_loss[loss=0.2718, simple_loss=0.3442, pruned_loss=0.09974, over 28991.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3491, pruned_loss=0.1064, over 5703816.27 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.384, pruned_loss=0.1337, over 5716203.66 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3455, pruned_loss=0.1037, over 5708728.27 frames. ], batch size: 106, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:04:55,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.641e+02 1.109e+03 1.278e+03 1.660e+03 2.955e+03, threshold=2.556e+03, percent-clipped=3.0
+2023-03-04 12:05:16,215 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=359511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:05:17,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3291, 1.4742, 1.1816, 1.2430], device='cuda:0'), covar=tensor([0.1457, 0.1191, 0.1105, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.1611, 0.1490, 0.1462, 0.1559], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 12:05:29,351 INFO [train.py:968] (0/2) Epoch 8, batch 40750, giga_loss[loss=0.3009, simple_loss=0.3711, pruned_loss=0.1154, over 28688.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3531, pruned_loss=0.1084, over 5708122.87 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3841, pruned_loss=0.134, over 5719744.72 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3492, pruned_loss=0.1053, over 5708745.86 frames. ], batch size: 262, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:05:31,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1350, 1.5233, 1.2023, 0.5752], device='cuda:0'), covar=tensor([0.2077, 0.1363, 0.1589, 0.3422], device='cuda:0'), in_proj_covar=tensor([0.1500, 0.1413, 0.1452, 0.1217], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 12:06:10,893 INFO [train.py:968] (0/2) Epoch 8, batch 40800, giga_loss[loss=0.2568, simple_loss=0.3362, pruned_loss=0.08871, over 28884.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3577, pruned_loss=0.111, over 5706461.79 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3843, pruned_loss=0.1341, over 5725106.86 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3536, pruned_loss=0.1077, over 5701646.14 frames. ], batch size: 186, lr: 3.94e-03, grad_scale: 8.0
+2023-03-04 12:06:15,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.362e+02 1.149e+03 1.459e+03 2.071e+03 8.329e+03, threshold=2.917e+03, percent-clipped=15.0
+2023-03-04 12:06:52,636 INFO [train.py:968] (0/2) Epoch 8, batch 40850, giga_loss[loss=0.2897, simple_loss=0.3593, pruned_loss=0.1101, over 28748.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3594, pruned_loss=0.1115, over 5708056.39 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3845, pruned_loss=0.1342, over 5725458.77 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3556, pruned_loss=0.1084, over 5704066.63 frames. ], batch size: 119, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:06:58,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5548, 2.0705, 2.3272, 1.9867], device='cuda:0'), covar=tensor([0.1081, 0.2249, 0.1486, 0.1730], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0731, 0.0656, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 12:07:17,090 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=359654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:07:19,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=359657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:07:36,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4709, 1.7904, 1.7753, 1.3471], device='cuda:0'), covar=tensor([0.1625, 0.2019, 0.1306, 0.1470], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0700, 0.0826, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 12:07:42,607 INFO [train.py:968] (0/2) Epoch 8, batch 40900, giga_loss[loss=0.3199, simple_loss=0.3798, pruned_loss=0.1301, over 29086.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3643, pruned_loss=0.1161, over 5706878.93 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3848, pruned_loss=0.1345, over 5729039.30 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3606, pruned_loss=0.1131, over 5700007.56 frames. ], batch size: 164, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:07:47,921 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.610e+02 1.257e+03 1.817e+03 2.389e+03 6.544e+03, threshold=3.634e+03, percent-clipped=16.0
+2023-03-04 12:07:48,707 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=359686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:08:00,184 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=359700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:08:22,334 INFO [train.py:968] (0/2) Epoch 8, batch 40950, giga_loss[loss=0.3556, simple_loss=0.407, pruned_loss=0.1521, over 28076.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3698, pruned_loss=0.1209, over 5699430.29 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3839, pruned_loss=0.134, over 5722722.26 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3668, pruned_loss=0.1181, over 5699377.14 frames. ], batch size: 412, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:08:39,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9160, 1.9463, 1.7436, 1.7428], device='cuda:0'), covar=tensor([0.1246, 0.2064, 0.1725, 0.1728], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0728, 0.0655, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 12:09:06,439 INFO [train.py:968] (0/2) Epoch 8, batch 41000, giga_loss[loss=0.3078, simple_loss=0.374, pruned_loss=0.1208, over 28784.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3788, pruned_loss=0.1275, over 5698152.04 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3838, pruned_loss=0.1338, over 5726374.66 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3763, pruned_loss=0.1254, over 5694306.57 frames. ], batch size: 119, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:09:12,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.025e+02 1.810e+03 2.329e+03 3.224e+03 6.367e+03, threshold=4.659e+03, percent-clipped=15.0
+2023-03-04 12:09:23,124 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=359797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:09:49,035 INFO [train.py:968] (0/2) Epoch 8, batch 41050, giga_loss[loss=0.3472, simple_loss=0.4028, pruned_loss=0.1458, over 28966.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.382, pruned_loss=0.1304, over 5704792.15 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3831, pruned_loss=0.1332, over 5733142.06 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3805, pruned_loss=0.129, over 5694843.38 frames. ], batch size: 136, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:09:49,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2731, 1.5545, 1.2268, 1.4202], device='cuda:0'), covar=tensor([0.2432, 0.2276, 0.2563, 0.2054], device='cuda:0'), in_proj_covar=tensor([0.1215, 0.0906, 0.1077, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 12:10:34,500 INFO [train.py:968] (0/2) Epoch 8, batch 41100, giga_loss[loss=0.3212, simple_loss=0.3907, pruned_loss=0.1259, over 28938.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3879, pruned_loss=0.1354, over 5702789.77 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3827, pruned_loss=0.1327, over 5736401.68 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3872, pruned_loss=0.1347, over 5690935.51 frames. ], batch size: 186, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:10:39,007 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=359882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:10:40,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.540e+03 2.120e+03 2.949e+03 1.266e+04, threshold=4.241e+03, percent-clipped=7.0
+2023-03-04 12:11:26,634 INFO [train.py:968] (0/2) Epoch 8, batch 41150, giga_loss[loss=0.3744, simple_loss=0.4198, pruned_loss=0.1645, over 28628.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3934, pruned_loss=0.1407, over 5680648.35 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3828, pruned_loss=0.1329, over 5737664.22 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3928, pruned_loss=0.1401, over 5669397.06 frames. ], batch size: 336, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:12:19,638 INFO [train.py:968] (0/2) Epoch 8, batch 41200, libri_loss[loss=0.3465, simple_loss=0.4053, pruned_loss=0.1439, over 29538.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3967, pruned_loss=0.1444, over 5661965.83 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.383, pruned_loss=0.133, over 5729375.50 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3962, pruned_loss=0.144, over 5659702.80 frames. ], batch size: 81, lr: 3.94e-03, grad_scale: 8.0
+2023-03-04 12:12:26,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.541e+02 1.701e+03 2.185e+03 2.658e+03 5.859e+03, threshold=4.370e+03, percent-clipped=4.0
+2023-03-04 12:12:42,306 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-360000.pt
+2023-03-04 12:13:02,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3503, 1.5408, 1.2591, 1.2881], device='cuda:0'), covar=tensor([0.1306, 0.1267, 0.0973, 0.1014], device='cuda:0'), in_proj_covar=tensor([0.1609, 0.1476, 0.1448, 0.1549], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 12:13:12,944 INFO [train.py:968] (0/2) Epoch 8, batch 41250, giga_loss[loss=0.4117, simple_loss=0.4288, pruned_loss=0.1973, over 23652.00 frames. ], tot_loss[loss=0.3484, simple_loss=0.3998, pruned_loss=0.1485, over 5653671.14 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3825, pruned_loss=0.1326, over 5732087.46 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4001, pruned_loss=0.1487, over 5648423.36 frames. ], batch size: 705, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:13:57,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=360075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:14:02,631 INFO [train.py:968] (0/2) Epoch 8, batch 41300, giga_loss[loss=0.4675, simple_loss=0.4809, pruned_loss=0.227, over 27605.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.4043, pruned_loss=0.153, over 5642985.59 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3827, pruned_loss=0.1327, over 5735454.83 frames. ], giga_tot_loss[loss=0.3557, simple_loss=0.4046, pruned_loss=0.1534, over 5634343.75 frames. ], batch size: 472, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:14:08,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.807e+02 1.634e+03 2.210e+03 2.766e+03 8.042e+03, threshold=4.419e+03, percent-clipped=3.0
+2023-03-04 12:14:23,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0306, 5.0661, 2.2321, 2.0124], device='cuda:0'), covar=tensor([0.0833, 0.0182, 0.0702, 0.1207], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0499, 0.0325, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 12:14:51,541 INFO [train.py:968] (0/2) Epoch 8, batch 41350, giga_loss[loss=0.4151, simple_loss=0.4406, pruned_loss=0.1948, over 27590.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4062, pruned_loss=0.1548, over 5634554.31 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3822, pruned_loss=0.1324, over 5733908.22 frames. ], giga_tot_loss[loss=0.3596, simple_loss=0.4074, pruned_loss=0.1558, over 5627010.98 frames. ], batch size: 472, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:15:23,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3568, 1.3554, 1.4444, 1.2341], device='cuda:0'), covar=tensor([0.0762, 0.1012, 0.1213, 0.0986], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0731, 0.0657, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 12:15:36,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=360172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:15:42,890 INFO [train.py:968] (0/2) Epoch 8, batch 41400, giga_loss[loss=0.3866, simple_loss=0.4082, pruned_loss=0.1825, over 23716.00 frames. ], tot_loss[loss=0.3606, simple_loss=0.4071, pruned_loss=0.157, over 5628381.22 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3823, pruned_loss=0.1325, over 5737790.33 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.4086, pruned_loss=0.1583, over 5616726.92 frames. ], batch size: 705, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:15:47,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.948e+03 2.527e+03 3.905e+03 1.079e+04, threshold=5.053e+03, percent-clipped=21.0
+2023-03-04 12:16:05,684 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-04 12:16:18,601 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=360218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:16:21,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=360221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:16:29,855 INFO [train.py:968] (0/2) Epoch 8, batch 41450, giga_loss[loss=0.3376, simple_loss=0.3991, pruned_loss=0.138, over 28917.00 frames. ], tot_loss[loss=0.3575, simple_loss=0.405, pruned_loss=0.155, over 5647267.16 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3824, pruned_loss=0.1325, over 5739625.65 frames. ], giga_tot_loss[loss=0.3597, simple_loss=0.4064, pruned_loss=0.1564, over 5634718.67 frames. ], batch size: 174, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:16:45,339 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=360250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:16:50,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=360257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:17:10,668 INFO [train.py:968] (0/2) Epoch 8, batch 41500, giga_loss[loss=0.3205, simple_loss=0.3903, pruned_loss=0.1254, over 28790.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4033, pruned_loss=0.1517, over 5657364.39 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3827, pruned_loss=0.1325, over 5733822.11 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.4053, pruned_loss=0.1538, over 5647401.14 frames. ], batch size: 243, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:17:18,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.243e+02 1.543e+03 1.931e+03 2.732e+03 7.818e+03, threshold=3.863e+03, percent-clipped=2.0
+2023-03-04 12:17:36,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3234, 1.4504, 1.4439, 1.3337], device='cuda:0'), covar=tensor([0.1282, 0.1473, 0.1817, 0.1455], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0739, 0.0664, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 12:17:49,979 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=360315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:17:53,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=360318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:18:04,281 INFO [train.py:968] (0/2) Epoch 8, batch 41550, giga_loss[loss=0.4374, simple_loss=0.4536, pruned_loss=0.2106, over 27814.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.403, pruned_loss=0.1504, over 5658166.06 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3827, pruned_loss=0.1325, over 5734337.00 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4047, pruned_loss=0.1522, over 5649616.64 frames. ], batch size: 412, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:18:13,438 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 12:18:16,938 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=360343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:18:20,338 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=360347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:18:29,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4252, 2.0581, 1.5318, 0.6458], device='cuda:0'), covar=tensor([0.2980, 0.1696, 0.2524, 0.3678], device='cuda:0'), in_proj_covar=tensor([0.1525, 0.1444, 0.1472, 0.1238], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 12:18:53,105 INFO [train.py:968] (0/2) Epoch 8, batch 41600, giga_loss[loss=0.5935, simple_loss=0.5486, pruned_loss=0.3192, over 26352.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4037, pruned_loss=0.1506, over 5648382.59 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3824, pruned_loss=0.1324, over 5726314.90 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4059, pruned_loss=0.1527, over 5646346.61 frames. ], batch size: 555, lr: 3.94e-03, grad_scale: 8.0
+2023-03-04 12:18:59,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.486e+02 1.713e+03 2.250e+03 2.803e+03 9.332e+03, threshold=4.499e+03, percent-clipped=9.0
+2023-03-04 12:19:14,441 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=360400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:19:17,786 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=360403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:19:27,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=360411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:19:30,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2631, 1.3182, 1.1919, 1.3784], device='cuda:0'), covar=tensor([0.0784, 0.0337, 0.0336, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 12:19:41,553 INFO [train.py:968] (0/2) Epoch 8, batch 41650, libri_loss[loss=0.2957, simple_loss=0.3453, pruned_loss=0.123, over 29507.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.3999, pruned_loss=0.1472, over 5655619.42 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3822, pruned_loss=0.1322, over 5732051.72 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4026, pruned_loss=0.1495, over 5645998.66 frames. ], batch size: 70, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:19:44,474 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=360432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:20:24,768 INFO [train.py:968] (0/2) Epoch 8, batch 41700, giga_loss[loss=0.2742, simple_loss=0.3569, pruned_loss=0.09574, over 28924.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3979, pruned_loss=0.1444, over 5637201.66 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3822, pruned_loss=0.1322, over 5719049.40 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.4006, pruned_loss=0.1469, over 5638145.97 frames. ], batch size: 213, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:20:32,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.730e+02 1.596e+03 2.220e+03 3.123e+03 8.133e+03, threshold=4.441e+03, percent-clipped=4.0
+2023-03-04 12:20:39,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5049, 1.5840, 1.5037, 1.4750], device='cuda:0'), covar=tensor([0.1214, 0.1717, 0.1705, 0.1595], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0731, 0.0656, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 12:20:41,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8008, 2.6132, 1.7886, 1.3689], device='cuda:0'), covar=tensor([0.2233, 0.1076, 0.1368, 0.1785], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1477, 0.1446, 0.1553], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 12:20:44,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6397, 4.4295, 4.1791, 1.9763], device='cuda:0'), covar=tensor([0.0421, 0.0618, 0.0693, 0.2093], device='cuda:0'), in_proj_covar=tensor([0.0996, 0.0938, 0.0826, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 12:21:15,506 INFO [train.py:968] (0/2) Epoch 8, batch 41750, giga_loss[loss=0.2786, simple_loss=0.3532, pruned_loss=0.102, over 28501.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3948, pruned_loss=0.1407, over 5657640.43 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.382, pruned_loss=0.132, over 5721951.54 frames. ], giga_tot_loss[loss=0.3417, simple_loss=0.3974, pruned_loss=0.143, over 5654717.09 frames. ], batch size: 71, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:21:58,099 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1723, 1.7866, 1.3754, 0.4177], device='cuda:0'), covar=tensor([0.2587, 0.1592, 0.2653, 0.3365], device='cuda:0'), in_proj_covar=tensor([0.1511, 0.1430, 0.1460, 0.1228], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 12:22:02,178 INFO [train.py:968] (0/2) Epoch 8, batch 41800, libri_loss[loss=0.3199, simple_loss=0.3829, pruned_loss=0.1284, over 29540.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3908, pruned_loss=0.1377, over 5660805.54 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3818, pruned_loss=0.132, over 5727511.96 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3934, pruned_loss=0.1397, over 5650949.34 frames. ], batch size: 79, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:22:08,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.114e+02 1.500e+03 2.015e+03 2.939e+03 7.952e+03, threshold=4.031e+03, percent-clipped=8.0
+2023-03-04 12:22:50,480 INFO [train.py:968] (0/2) Epoch 8, batch 41850, giga_loss[loss=0.3507, simple_loss=0.4044, pruned_loss=0.1485, over 27645.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3874, pruned_loss=0.1357, over 5648617.03 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3815, pruned_loss=0.1317, over 5729787.28 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3899, pruned_loss=0.1377, over 5637778.08 frames. ], batch size: 472, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:23:19,871 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=360659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:23:28,608 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 12:23:38,175 INFO [train.py:968] (0/2) Epoch 8, batch 41900, giga_loss[loss=0.3329, simple_loss=0.4028, pruned_loss=0.1315, over 28888.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3886, pruned_loss=0.1363, over 5664827.20 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3813, pruned_loss=0.1315, over 5732379.90 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3908, pruned_loss=0.1381, over 5652896.61 frames. ], batch size: 145, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:23:43,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.447e+02 1.426e+03 1.804e+03 2.306e+03 4.126e+03, threshold=3.607e+03, percent-clipped=1.0
+2023-03-04 12:23:55,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5594, 1.8940, 1.7092, 1.8164], device='cuda:0'), covar=tensor([0.0725, 0.0268, 0.0279, 0.0778], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:0')
+2023-03-04 12:24:13,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=360718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:24:23,969 INFO [train.py:968] (0/2) Epoch 8, batch 41950, giga_loss[loss=0.2945, simple_loss=0.3632, pruned_loss=0.1129, over 28850.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3883, pruned_loss=0.1357, over 5645506.58 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3815, pruned_loss=0.1318, over 5706288.04 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3902, pruned_loss=0.137, over 5656962.85 frames. ], batch size: 112, lr: 3.94e-03, grad_scale: 2.0
+2023-03-04 12:24:37,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1198, 1.5456, 1.1264, 1.4814], device='cuda:0'), covar=tensor([0.2320, 0.2167, 0.2437, 0.1946], device='cuda:0'), in_proj_covar=tensor([0.1213, 0.0911, 0.1076, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 12:25:11,141 INFO [train.py:968] (0/2) Epoch 8, batch 42000, giga_loss[loss=0.2459, simple_loss=0.33, pruned_loss=0.0809, over 28933.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3854, pruned_loss=0.1323, over 5662023.68 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3816, pruned_loss=0.1317, over 5710351.44 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3869, pruned_loss=0.1334, over 5666119.84 frames. ], batch size: 112, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:25:11,145 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 12:25:16,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2257, 1.7730, 1.5967, 1.1388], device='cuda:0'), covar=tensor([0.1825, 0.2476, 0.1574, 0.1800], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0702, 0.0822, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 12:25:20,263 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2203, simple_loss=0.3245, pruned_loss=0.05808, over 944034.00 frames. 
+2023-03-04 12:25:20,264 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 12:25:25,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=360786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:25:26,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.369e+02 1.350e+03 1.765e+03 2.730e+03 1.626e+04, threshold=3.531e+03, percent-clipped=14.0
+2023-03-04 12:26:05,799 INFO [train.py:968] (0/2) Epoch 8, batch 42050, giga_loss[loss=0.3356, simple_loss=0.4098, pruned_loss=0.1307, over 28940.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3861, pruned_loss=0.1302, over 5655406.12 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3814, pruned_loss=0.1317, over 5693017.32 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3876, pruned_loss=0.1312, over 5673834.69 frames. ], batch size: 186, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:26:37,418 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=360861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:26:40,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=360864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:26:54,339 INFO [train.py:968] (0/2) Epoch 8, batch 42100, giga_loss[loss=0.3263, simple_loss=0.3854, pruned_loss=0.1337, over 28837.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3875, pruned_loss=0.1313, over 5654416.73 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3806, pruned_loss=0.1313, over 5695583.22 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3895, pruned_loss=0.1324, over 5666000.14 frames. ], batch size: 112, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:27:01,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.274e+02 1.640e+03 2.239e+03 3.134e+03 9.782e+03, threshold=4.479e+03, percent-clipped=20.0
+2023-03-04 12:27:05,928 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=360893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:27:41,479 INFO [train.py:968] (0/2) Epoch 8, batch 42150, giga_loss[loss=0.3374, simple_loss=0.3981, pruned_loss=0.1383, over 28921.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3887, pruned_loss=0.1332, over 5651681.88 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3809, pruned_loss=0.1316, over 5687860.49 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3901, pruned_loss=0.1338, over 5666552.85 frames. ], batch size: 227, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:27:41,813 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=360929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:27:44,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=360932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:28:12,268 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=360961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:28:27,798 INFO [train.py:968] (0/2) Epoch 8, batch 42200, libri_loss[loss=0.3442, simple_loss=0.3906, pruned_loss=0.1489, over 29584.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3858, pruned_loss=0.1314, over 5664748.42 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3805, pruned_loss=0.1314, over 5692006.74 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3874, pruned_loss=0.1321, over 5671772.65 frames. ], batch size: 76, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:28:35,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.711e+02 1.650e+03 1.998e+03 2.572e+03 8.436e+03, threshold=3.996e+03, percent-clipped=4.0
+2023-03-04 12:29:04,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2574, 1.1437, 1.0303, 1.4880], device='cuda:0'), covar=tensor([0.0733, 0.0346, 0.0334, 0.0768], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:0')
+2023-03-04 12:29:15,270 INFO [train.py:968] (0/2) Epoch 8, batch 42250, giga_loss[loss=0.372, simple_loss=0.4124, pruned_loss=0.1658, over 27964.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3846, pruned_loss=0.1321, over 5663164.32 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3802, pruned_loss=0.1311, over 5693395.13 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3862, pruned_loss=0.1329, over 5667213.76 frames. ], batch size: 412, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:29:19,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=361034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:29:58,731 INFO [train.py:968] (0/2) Epoch 8, batch 42300, giga_loss[loss=0.3423, simple_loss=0.3975, pruned_loss=0.1435, over 28926.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3833, pruned_loss=0.1321, over 5646014.67 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3803, pruned_loss=0.1311, over 5681179.97 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3847, pruned_loss=0.1328, over 5658745.57 frames. ], batch size: 174, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:30:07,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.872e+02 1.506e+03 1.943e+03 2.965e+03 8.575e+03, threshold=3.887e+03, percent-clipped=15.0
+2023-03-04 12:30:45,665 INFO [train.py:968] (0/2) Epoch 8, batch 42350, giga_loss[loss=0.3005, simple_loss=0.3754, pruned_loss=0.1128, over 28924.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3825, pruned_loss=0.1306, over 5661126.51 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3803, pruned_loss=0.1311, over 5684582.81 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3836, pruned_loss=0.1311, over 5667481.85 frames. ], batch size: 136, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:30:52,263 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=361137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:31:27,985 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=361177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:31:28,888 INFO [train.py:968] (0/2) Epoch 8, batch 42400, giga_loss[loss=0.2778, simple_loss=0.3568, pruned_loss=0.09936, over 28836.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3818, pruned_loss=0.1286, over 5675588.88 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3799, pruned_loss=0.1308, over 5688491.68 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3832, pruned_loss=0.1293, over 5676804.40 frames. ], batch size: 199, lr: 3.93e-03, grad_scale: 8.0
+2023-03-04 12:31:29,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=361180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:31:37,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.140e+02 1.548e+03 1.913e+03 2.753e+03 7.334e+03, threshold=3.827e+03, percent-clipped=10.0
+2023-03-04 12:31:58,801 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=361209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:32:16,931 INFO [train.py:968] (0/2) Epoch 8, batch 42450, giga_loss[loss=0.2727, simple_loss=0.3428, pruned_loss=0.1013, over 28580.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3823, pruned_loss=0.1286, over 5667320.56 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3803, pruned_loss=0.1312, over 5672631.03 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.383, pruned_loss=0.1288, over 5682371.61 frames. ], batch size: 85, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:33:00,800 INFO [train.py:968] (0/2) Epoch 8, batch 42500, libri_loss[loss=0.3719, simple_loss=0.4258, pruned_loss=0.159, over 29665.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3816, pruned_loss=0.1288, over 5671202.66 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3807, pruned_loss=0.1313, over 5678477.45 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3818, pruned_loss=0.1287, over 5677669.28 frames. ], batch size: 88, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:33:05,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1911, 1.1891, 1.0177, 0.9059], device='cuda:0'), covar=tensor([0.0744, 0.0512, 0.1016, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0447, 0.0495, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 12:33:09,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.652e+02 1.443e+03 1.946e+03 2.697e+03 7.683e+03, threshold=3.891e+03, percent-clipped=7.0
+2023-03-04 12:33:45,915 INFO [train.py:968] (0/2) Epoch 8, batch 42550, giga_loss[loss=0.3898, simple_loss=0.4281, pruned_loss=0.1758, over 27997.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3812, pruned_loss=0.1294, over 5673341.18 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.381, pruned_loss=0.1314, over 5683820.64 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3812, pruned_loss=0.1292, over 5673270.02 frames. ], batch size: 412, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:33:46,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3522, 1.8015, 1.7038, 1.2640], device='cuda:0'), covar=tensor([0.1358, 0.1997, 0.1102, 0.1312], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0709, 0.0830, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 12:34:30,845 INFO [train.py:968] (0/2) Epoch 8, batch 42600, giga_loss[loss=0.285, simple_loss=0.344, pruned_loss=0.113, over 28893.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3793, pruned_loss=0.1288, over 5668562.00 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3806, pruned_loss=0.1312, over 5677119.86 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3795, pruned_loss=0.1288, over 5674308.50 frames. ], batch size: 112, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:34:40,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.565e+03 2.115e+03 3.169e+03 7.935e+03, threshold=4.230e+03, percent-clipped=11.0
+2023-03-04 12:34:46,900 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=361394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:35:19,849 INFO [train.py:968] (0/2) Epoch 8, batch 42650, giga_loss[loss=0.3157, simple_loss=0.3674, pruned_loss=0.132, over 28570.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3808, pruned_loss=0.131, over 5661941.61 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3805, pruned_loss=0.1311, over 5678418.08 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.381, pruned_loss=0.131, over 5665164.88 frames. ], batch size: 85, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:35:24,046 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=361433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:35:26,593 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-04 12:36:05,398 INFO [train.py:968] (0/2) Epoch 8, batch 42700, giga_loss[loss=0.3239, simple_loss=0.3804, pruned_loss=0.1337, over 28943.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.379, pruned_loss=0.1298, over 5672511.11 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3809, pruned_loss=0.1312, over 5681605.90 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3789, pruned_loss=0.1297, over 5672010.76 frames. ], batch size: 106, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:36:13,245 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.950e+02 1.573e+03 2.137e+03 2.917e+03 7.672e+03, threshold=4.275e+03, percent-clipped=5.0
+2023-03-04 12:36:37,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=361512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:36:53,723 INFO [train.py:968] (0/2) Epoch 8, batch 42750, giga_loss[loss=0.3064, simple_loss=0.3702, pruned_loss=0.1213, over 28908.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3798, pruned_loss=0.1307, over 5680720.90 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3809, pruned_loss=0.1312, over 5685107.78 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3796, pruned_loss=0.1306, over 5677137.52 frames. ], batch size: 174, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:37:35,189 INFO [train.py:968] (0/2) Epoch 8, batch 42800, libri_loss[loss=0.3155, simple_loss=0.3737, pruned_loss=0.1286, over 29558.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3808, pruned_loss=0.1313, over 5690121.65 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3804, pruned_loss=0.1309, over 5694127.80 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.381, pruned_loss=0.1315, over 5678685.30 frames. ], batch size: 78, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:37:45,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.547e+03 2.079e+03 3.316e+03 1.981e+04, threshold=4.157e+03, percent-clipped=9.0
+2023-03-04 12:38:05,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4280, 3.4638, 1.4260, 1.4989], device='cuda:0'), covar=tensor([0.0915, 0.0319, 0.0877, 0.1291], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0504, 0.0327, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 12:38:17,175 INFO [train.py:968] (0/2) Epoch 8, batch 42850, giga_loss[loss=0.3303, simple_loss=0.3964, pruned_loss=0.1321, over 29062.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3801, pruned_loss=0.1299, over 5686890.42 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3793, pruned_loss=0.1302, over 5699476.94 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3813, pruned_loss=0.1306, over 5672584.57 frames. ], batch size: 128, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:38:28,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2182, 1.4394, 1.2035, 1.0489], device='cuda:0'), covar=tensor([0.1651, 0.1548, 0.1024, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.1622, 0.1496, 0.1461, 0.1564], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 12:38:28,721 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-04 12:38:39,950 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=361655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:38:40,816 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-04 12:38:42,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=361658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:38:44,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8880, 5.6828, 5.3498, 2.5372], device='cuda:0'), covar=tensor([0.0463, 0.0776, 0.0834, 0.1903], device='cuda:0'), in_proj_covar=tensor([0.0995, 0.0937, 0.0826, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 12:38:59,993 INFO [train.py:968] (0/2) Epoch 8, batch 42900, giga_loss[loss=0.3831, simple_loss=0.422, pruned_loss=0.1721, over 28629.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3812, pruned_loss=0.1301, over 5683241.51 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3795, pruned_loss=0.1303, over 5698147.98 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.382, pruned_loss=0.1305, over 5672677.97 frames. ], batch size: 92, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:39:08,116 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=361687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:39:10,266 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.351e+02 1.571e+03 1.973e+03 2.910e+03 2.430e+04, threshold=3.946e+03, percent-clipped=13.0
+2023-03-04 12:39:45,926 INFO [train.py:968] (0/2) Epoch 8, batch 42950, giga_loss[loss=0.3298, simple_loss=0.389, pruned_loss=0.1353, over 28836.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.382, pruned_loss=0.1305, over 5675392.64 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3797, pruned_loss=0.1306, over 5693878.52 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3826, pruned_loss=0.1307, over 5669748.77 frames. ], batch size: 199, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:40:02,328 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 12:40:28,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=361769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:40:34,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2975, 2.8782, 1.4289, 1.4581], device='cuda:0'), covar=tensor([0.0872, 0.0314, 0.0799, 0.1185], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0502, 0.0327, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 12:40:34,832 INFO [train.py:968] (0/2) Epoch 8, batch 43000, giga_loss[loss=0.3336, simple_loss=0.3885, pruned_loss=0.1393, over 28622.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3827, pruned_loss=0.1316, over 5673473.92 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3798, pruned_loss=0.1305, over 5698718.69 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3831, pruned_loss=0.1318, over 5663998.85 frames. ], batch size: 307, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:40:45,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.466e+03 1.927e+03 2.569e+03 5.447e+03, threshold=3.855e+03, percent-clipped=8.0
+2023-03-04 12:41:03,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=361808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:41:22,452 INFO [train.py:968] (0/2) Epoch 8, batch 43050, giga_loss[loss=0.3278, simple_loss=0.383, pruned_loss=0.1363, over 28924.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3862, pruned_loss=0.1356, over 5677223.30 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3796, pruned_loss=0.1305, over 5704254.21 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3868, pruned_loss=0.1359, over 5663875.56 frames. ], batch size: 136, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:42:14,926 INFO [train.py:968] (0/2) Epoch 8, batch 43100, giga_loss[loss=0.4954, simple_loss=0.4844, pruned_loss=0.2532, over 26671.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3881, pruned_loss=0.1388, over 5664280.67 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3801, pruned_loss=0.131, over 5705710.99 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3883, pruned_loss=0.1387, over 5651246.17 frames. ], batch size: 555, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:42:24,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.949e+02 1.885e+03 2.486e+03 3.209e+03 7.466e+03, threshold=4.971e+03, percent-clipped=15.0
+2023-03-04 12:42:44,034 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=361912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:42:48,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=361915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:43:03,513 INFO [train.py:968] (0/2) Epoch 8, batch 43150, giga_loss[loss=0.3539, simple_loss=0.4026, pruned_loss=0.1526, over 28824.00 frames. ], tot_loss[loss=0.336, simple_loss=0.39, pruned_loss=0.141, over 5656022.81 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3807, pruned_loss=0.1315, over 5696751.41 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3899, pruned_loss=0.1407, over 5652473.62 frames. ], batch size: 285, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:43:16,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=361944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:43:22,325 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=361951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:43:25,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=361954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:43:45,215 INFO [train.py:968] (0/2) Epoch 8, batch 43200, libri_loss[loss=0.3046, simple_loss=0.3746, pruned_loss=0.1172, over 29253.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3866, pruned_loss=0.1381, over 5673476.46 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3798, pruned_loss=0.1308, over 5703363.75 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3875, pruned_loss=0.1387, over 5663356.10 frames. ], batch size: 94, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:43:48,279 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=361983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:43:55,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.707e+02 1.527e+03 1.971e+03 2.750e+03 6.724e+03, threshold=3.942e+03, percent-clipped=6.0
+2023-03-04 12:43:59,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5779, 1.4716, 1.2358, 1.2841], device='cuda:0'), covar=tensor([0.0541, 0.0378, 0.0750, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0447, 0.0496, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 12:44:01,780 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=361998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:44:04,456 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-362000.pt
+2023-03-04 12:44:08,233 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 12:44:09,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5540, 1.5789, 1.5610, 1.4355], device='cuda:0'), covar=tensor([0.1246, 0.1955, 0.1769, 0.1745], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0735, 0.0654, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 12:44:15,931 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9921, 2.6089, 1.6809, 1.4652], device='cuda:0'), covar=tensor([0.1890, 0.1171, 0.1532, 0.1796], device='cuda:0'), in_proj_covar=tensor([0.1619, 0.1486, 0.1453, 0.1557], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 12:44:31,052 INFO [train.py:968] (0/2) Epoch 8, batch 43250, giga_loss[loss=0.3085, simple_loss=0.3702, pruned_loss=0.1234, over 28431.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3842, pruned_loss=0.1362, over 5677524.87 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3796, pruned_loss=0.1307, over 5707307.73 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3853, pruned_loss=0.137, over 5665450.90 frames. ], batch size: 78, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:45:16,142 INFO [train.py:968] (0/2) Epoch 8, batch 43300, giga_loss[loss=0.2928, simple_loss=0.3756, pruned_loss=0.105, over 28565.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3831, pruned_loss=0.133, over 5686857.81 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3802, pruned_loss=0.1311, over 5709507.74 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3834, pruned_loss=0.1332, over 5674959.50 frames. ], batch size: 71, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:45:26,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.603e+03 1.901e+03 2.945e+03 6.220e+03, threshold=3.801e+03, percent-clipped=14.0
+2023-03-04 12:46:02,728 INFO [train.py:968] (0/2) Epoch 8, batch 43350, giga_loss[loss=0.3382, simple_loss=0.3843, pruned_loss=0.146, over 28903.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3801, pruned_loss=0.1315, over 5666426.68 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3801, pruned_loss=0.1311, over 5701385.99 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3804, pruned_loss=0.1317, over 5665004.12 frames. ], batch size: 112, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:46:11,672 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3402, 1.8067, 1.4774, 1.5361], device='cuda:0'), covar=tensor([0.0651, 0.0248, 0.0268, 0.0642], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0117, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 12:46:12,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0135, 1.1959, 1.2561, 1.1109], device='cuda:0'), covar=tensor([0.1269, 0.1301, 0.1854, 0.1348], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0735, 0.0655, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 12:46:41,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2416, 0.8162, 0.8054, 1.3966], device='cuda:0'), covar=tensor([0.0738, 0.0331, 0.0338, 0.0769], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 12:46:43,101 INFO [train.py:968] (0/2) Epoch 8, batch 43400, giga_loss[loss=0.303, simple_loss=0.3645, pruned_loss=0.1207, over 28855.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3781, pruned_loss=0.1309, over 5665036.86 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3796, pruned_loss=0.1309, over 5697966.15 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3787, pruned_loss=0.1312, over 5665315.02 frames. ], batch size: 99, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:46:55,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.574e+02 1.581e+03 1.930e+03 2.659e+03 6.388e+03, threshold=3.861e+03, percent-clipped=7.0
+2023-03-04 12:47:07,283 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=362204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:47:28,523 INFO [train.py:968] (0/2) Epoch 8, batch 43450, libri_loss[loss=0.3609, simple_loss=0.4165, pruned_loss=0.1527, over 29536.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3772, pruned_loss=0.1307, over 5661446.21 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3799, pruned_loss=0.131, over 5699436.06 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3773, pruned_loss=0.1309, over 5659638.88 frames. ], batch size: 84, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:48:09,395 INFO [train.py:968] (0/2) Epoch 8, batch 43500, giga_loss[loss=0.322, simple_loss=0.3875, pruned_loss=0.1283, over 29017.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3796, pruned_loss=0.132, over 5669824.58 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3798, pruned_loss=0.1308, over 5705009.88 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3797, pruned_loss=0.1323, over 5662513.39 frames. ], batch size: 155, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:48:21,049 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.536e+03 1.886e+03 2.468e+03 6.162e+03, threshold=3.773e+03, percent-clipped=4.0
+2023-03-04 12:48:21,834 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=362292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:48:56,970 INFO [train.py:968] (0/2) Epoch 8, batch 43550, giga_loss[loss=0.316, simple_loss=0.3944, pruned_loss=0.1188, over 29027.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3841, pruned_loss=0.1335, over 5668219.06 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3798, pruned_loss=0.1309, over 5701680.77 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3843, pruned_loss=0.1337, over 5664509.35 frames. ], batch size: 136, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:49:41,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=362373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:49:44,550 INFO [train.py:968] (0/2) Epoch 8, batch 43600, giga_loss[loss=0.37, simple_loss=0.4364, pruned_loss=0.1518, over 28892.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3866, pruned_loss=0.133, over 5666666.84 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3792, pruned_loss=0.1307, over 5708423.31 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3875, pruned_loss=0.1333, over 5656182.01 frames. ], batch size: 112, lr: 3.93e-03, grad_scale: 8.0
+2023-03-04 12:49:57,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.584e+02 1.326e+03 1.756e+03 2.293e+03 8.539e+03, threshold=3.512e+03, percent-clipped=11.0
+2023-03-04 12:50:07,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2766, 1.2490, 1.0453, 0.9769], device='cuda:0'), covar=tensor([0.0665, 0.0473, 0.0973, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0444, 0.0491, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 12:50:28,306 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 12:50:30,624 INFO [train.py:968] (0/2) Epoch 8, batch 43650, giga_loss[loss=0.3109, simple_loss=0.3822, pruned_loss=0.1198, over 28526.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3883, pruned_loss=0.1338, over 5677744.73 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3792, pruned_loss=0.1307, over 5711637.95 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3891, pruned_loss=0.1342, over 5665810.35 frames. ], batch size: 336, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:50:51,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4292, 2.0114, 1.4623, 0.7591], device='cuda:0'), covar=tensor([0.2649, 0.1455, 0.2159, 0.3167], device='cuda:0'), in_proj_covar=tensor([0.1502, 0.1437, 0.1446, 0.1216], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 12:51:08,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4631, 1.6957, 1.7014, 1.3377], device='cuda:0'), covar=tensor([0.1572, 0.2083, 0.1213, 0.1408], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0708, 0.0831, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 12:51:16,697 INFO [train.py:968] (0/2) Epoch 8, batch 43700, giga_loss[loss=0.3873, simple_loss=0.4289, pruned_loss=0.1729, over 27913.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3906, pruned_loss=0.1357, over 5655609.53 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3798, pruned_loss=0.1312, over 5685290.69 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3909, pruned_loss=0.1356, over 5669219.99 frames. ], batch size: 412, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:51:30,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.586e+03 2.119e+03 2.592e+03 5.630e+03, threshold=4.237e+03, percent-clipped=7.0
+2023-03-04 12:51:47,826 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2856, 1.7358, 1.3223, 0.4752], device='cuda:0'), covar=tensor([0.2061, 0.1462, 0.2122, 0.3138], device='cuda:0'), in_proj_covar=tensor([0.1508, 0.1442, 0.1451, 0.1220], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 12:51:53,982 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=362516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:51:56,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=362519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:52:04,458 INFO [train.py:968] (0/2) Epoch 8, batch 43750, giga_loss[loss=0.3462, simple_loss=0.397, pruned_loss=0.1477, over 29049.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.391, pruned_loss=0.1367, over 5659218.91 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.38, pruned_loss=0.1314, over 5684052.72 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3912, pruned_loss=0.1365, over 5670928.89 frames. ], batch size: 136, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:52:20,522 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=362548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:52:49,336 INFO [train.py:968] (0/2) Epoch 8, batch 43800, giga_loss[loss=0.303, simple_loss=0.3679, pruned_loss=0.119, over 28770.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3901, pruned_loss=0.1371, over 5656808.31 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3802, pruned_loss=0.1314, over 5686148.31 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3904, pruned_loss=0.1371, over 5663375.79 frames. ], batch size: 119, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:52:49,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=362579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:52:55,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3303, 2.8352, 1.3851, 1.4375], device='cuda:0'), covar=tensor([0.0880, 0.0328, 0.0804, 0.1231], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0501, 0.0326, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 12:53:00,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.994e+02 1.623e+03 2.433e+03 3.465e+03 9.329e+03, threshold=4.865e+03, percent-clipped=19.0
+2023-03-04 12:53:35,201 INFO [train.py:968] (0/2) Epoch 8, batch 43850, giga_loss[loss=0.3064, simple_loss=0.3676, pruned_loss=0.1227, over 29004.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3868, pruned_loss=0.1355, over 5660792.04 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.38, pruned_loss=0.1315, over 5689343.18 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3873, pruned_loss=0.1355, over 5662861.80 frames. ], batch size: 145, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:53:35,744 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-04 12:54:09,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=362667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:54:13,053 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=362669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:54:22,139 INFO [train.py:968] (0/2) Epoch 8, batch 43900, giga_loss[loss=0.3378, simple_loss=0.3914, pruned_loss=0.1421, over 28590.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3857, pruned_loss=0.1355, over 5661946.58 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3798, pruned_loss=0.1312, over 5689128.53 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3864, pruned_loss=0.1358, over 5663358.03 frames. ], batch size: 78, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:54:33,212 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.438e+02 1.561e+03 2.034e+03 2.693e+03 7.161e+03, threshold=4.069e+03, percent-clipped=4.0
+2023-03-04 12:55:05,854 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=362722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:55:07,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=362725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:55:11,542 INFO [train.py:968] (0/2) Epoch 8, batch 43950, giga_loss[loss=0.381, simple_loss=0.4126, pruned_loss=0.1747, over 26540.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3867, pruned_loss=0.1371, over 5641646.02 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.38, pruned_loss=0.1314, over 5681514.65 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3871, pruned_loss=0.1373, over 5649250.28 frames. ], batch size: 555, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:55:31,587 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 12:55:36,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=362754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:55:57,412 INFO [train.py:968] (0/2) Epoch 8, batch 44000, giga_loss[loss=0.2894, simple_loss=0.3488, pruned_loss=0.115, over 28710.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3848, pruned_loss=0.1362, over 5640585.87 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3797, pruned_loss=0.1312, over 5677910.85 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3857, pruned_loss=0.1366, over 5648668.92 frames. ], batch size: 92, lr: 3.93e-03, grad_scale: 8.0
+2023-03-04 12:56:10,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.240e+02 1.892e+03 2.564e+03 3.417e+03 9.127e+03, threshold=5.129e+03, percent-clipped=11.0
+2023-03-04 12:56:26,920 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=362810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:56:28,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=362813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:56:44,095 INFO [train.py:968] (0/2) Epoch 8, batch 44050, giga_loss[loss=0.2952, simple_loss=0.3571, pruned_loss=0.1166, over 29014.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3819, pruned_loss=0.1341, over 5652232.90 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3797, pruned_loss=0.1312, over 5679195.30 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3826, pruned_loss=0.1345, over 5657363.18 frames. ], batch size: 128, lr: 3.93e-03, grad_scale: 8.0
+2023-03-04 12:56:51,553 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-04 12:56:56,894 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=362842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:57:27,710 INFO [train.py:968] (0/2) Epoch 8, batch 44100, giga_loss[loss=0.3363, simple_loss=0.3956, pruned_loss=0.1384, over 28619.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3813, pruned_loss=0.134, over 5654099.72 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3796, pruned_loss=0.1312, over 5683072.77 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.382, pruned_loss=0.1344, over 5654116.39 frames. ], batch size: 336, lr: 3.92e-03, grad_scale: 8.0
+2023-03-04 12:57:40,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.582e+02 1.422e+03 1.916e+03 2.650e+03 8.457e+03, threshold=3.832e+03, percent-clipped=2.0
+2023-03-04 12:58:18,922 INFO [train.py:968] (0/2) Epoch 8, batch 44150, giga_loss[loss=0.3797, simple_loss=0.4215, pruned_loss=0.169, over 28228.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3832, pruned_loss=0.1341, over 5653745.11 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3798, pruned_loss=0.1313, over 5684313.41 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3836, pruned_loss=0.1343, over 5652609.65 frames. ], batch size: 368, lr: 3.92e-03, grad_scale: 8.0
+2023-03-04 12:59:09,484 INFO [train.py:968] (0/2) Epoch 8, batch 44200, giga_loss[loss=0.2999, simple_loss=0.3646, pruned_loss=0.1176, over 28936.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3858, pruned_loss=0.1361, over 5650347.12 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3798, pruned_loss=0.1313, over 5684313.41 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3861, pruned_loss=0.1363, over 5649463.38 frames. ], batch size: 145, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 12:59:21,847 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.753e+02 1.546e+03 2.122e+03 2.740e+03 8.383e+03, threshold=4.243e+03, percent-clipped=9.0
+2023-03-04 12:59:53,530 INFO [train.py:968] (0/2) Epoch 8, batch 44250, giga_loss[loss=0.3629, simple_loss=0.4049, pruned_loss=0.1605, over 27542.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3844, pruned_loss=0.1352, over 5662332.77 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3797, pruned_loss=0.1311, over 5690111.39 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.385, pruned_loss=0.1357, over 5655050.77 frames. ], batch size: 472, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:00:05,083 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=363044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:00:06,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4034, 1.6335, 1.3508, 1.6087], device='cuda:0'), covar=tensor([0.2317, 0.2225, 0.2424, 0.1948], device='cuda:0'), in_proj_covar=tensor([0.1218, 0.0917, 0.1080, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 13:00:36,542 INFO [train.py:968] (0/2) Epoch 8, batch 44300, giga_loss[loss=0.441, simple_loss=0.473, pruned_loss=0.2045, over 28601.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3854, pruned_loss=0.1326, over 5669120.21 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3796, pruned_loss=0.1311, over 5685899.33 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3861, pruned_loss=0.1331, over 5666378.97 frames. ], batch size: 307, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:00:48,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.430e+02 1.446e+03 1.811e+03 2.230e+03 8.556e+03, threshold=3.623e+03, percent-clipped=4.0
+2023-03-04 13:01:19,079 INFO [train.py:968] (0/2) Epoch 8, batch 44350, giga_loss[loss=0.3664, simple_loss=0.4169, pruned_loss=0.1579, over 28948.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3869, pruned_loss=0.1318, over 5672346.81 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3796, pruned_loss=0.131, over 5692869.44 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3876, pruned_loss=0.1323, over 5663573.14 frames. ], batch size: 227, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:01:25,255 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3058, 1.8273, 1.4475, 1.5200], device='cuda:0'), covar=tensor([0.0765, 0.0287, 0.0308, 0.0802], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 13:01:25,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8062, 1.8393, 1.8376, 1.4932], device='cuda:0'), covar=tensor([0.1223, 0.1844, 0.1719, 0.2177], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0734, 0.0654, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 13:02:05,679 INFO [train.py:968] (0/2) Epoch 8, batch 44400, giga_loss[loss=0.4963, simple_loss=0.4875, pruned_loss=0.2525, over 23611.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3889, pruned_loss=0.1337, over 5658717.13 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3788, pruned_loss=0.1306, over 5690747.96 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3905, pruned_loss=0.1346, over 5652533.45 frames. ], batch size: 705, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:02:14,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=363187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:02:15,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=363190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:02:18,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.629e+02 1.536e+03 2.030e+03 2.995e+03 8.834e+03, threshold=4.061e+03, percent-clipped=18.0
+2023-03-04 13:02:21,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8941, 5.2184, 1.9932, 2.1296], device='cuda:0'), covar=tensor([0.0795, 0.0304, 0.0813, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0502, 0.0327, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 13:02:31,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8884, 1.3435, 5.2609, 3.8706], device='cuda:0'), covar=tensor([0.1493, 0.2555, 0.0355, 0.0596], device='cuda:0'), in_proj_covar=tensor([0.0623, 0.0572, 0.0823, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 13:02:44,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=363219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:02:53,135 INFO [train.py:968] (0/2) Epoch 8, batch 44450, giga_loss[loss=0.4315, simple_loss=0.4569, pruned_loss=0.2031, over 27585.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3922, pruned_loss=0.137, over 5666266.68 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3786, pruned_loss=0.1304, over 5696313.72 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3939, pruned_loss=0.138, over 5655595.44 frames. ], batch size: 472, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:03:00,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3075, 1.5374, 3.1799, 3.0124], device='cuda:0'), covar=tensor([0.1182, 0.1899, 0.0441, 0.0772], device='cuda:0'), in_proj_covar=tensor([0.0624, 0.0573, 0.0824, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 13:03:41,699 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3341, 1.8360, 1.3213, 0.5786], device='cuda:0'), covar=tensor([0.2250, 0.1294, 0.1926, 0.3056], device='cuda:0'), in_proj_covar=tensor([0.1490, 0.1435, 0.1442, 0.1208], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 13:03:42,004 INFO [train.py:968] (0/2) Epoch 8, batch 44500, giga_loss[loss=0.3721, simple_loss=0.4221, pruned_loss=0.161, over 28107.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3943, pruned_loss=0.1398, over 5656499.89 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3787, pruned_loss=0.1304, over 5688736.41 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.396, pruned_loss=0.1408, over 5654279.87 frames. ], batch size: 412, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:03:53,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.153e+03 1.606e+03 2.112e+03 3.224e+03 1.659e+04, threshold=4.224e+03, percent-clipped=13.0
+2023-03-04 13:03:57,309 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5845, 1.6934, 1.5293, 1.4739], device='cuda:0'), covar=tensor([0.1644, 0.1461, 0.1059, 0.1290], device='cuda:0'), in_proj_covar=tensor([0.1630, 0.1486, 0.1453, 0.1562], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 13:04:22,013 INFO [train.py:968] (0/2) Epoch 8, batch 44550, giga_loss[loss=0.3186, simple_loss=0.3859, pruned_loss=0.1257, over 28796.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3919, pruned_loss=0.1382, over 5666683.45 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3778, pruned_loss=0.1298, over 5686559.48 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3947, pruned_loss=0.14, over 5665549.22 frames. ], batch size: 119, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:04:43,332 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-04 13:04:47,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3239, 1.6729, 1.6360, 1.2380], device='cuda:0'), covar=tensor([0.1479, 0.2075, 0.1215, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0710, 0.0831, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 13:05:04,123 INFO [train.py:968] (0/2) Epoch 8, batch 44600, giga_loss[loss=0.2811, simple_loss=0.3616, pruned_loss=0.1003, over 28857.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3906, pruned_loss=0.1371, over 5666389.20 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3775, pruned_loss=0.1295, over 5692246.89 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3935, pruned_loss=0.139, over 5659982.60 frames. ], batch size: 145, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:05:17,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.705e+02 1.360e+03 1.840e+03 2.521e+03 4.134e+03, threshold=3.680e+03, percent-clipped=0.0
+2023-03-04 13:05:47,762 INFO [train.py:968] (0/2) Epoch 8, batch 44650, giga_loss[loss=0.2964, simple_loss=0.3796, pruned_loss=0.1066, over 28838.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3897, pruned_loss=0.1344, over 5681936.90 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3776, pruned_loss=0.1296, over 5696679.84 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3922, pruned_loss=0.136, over 5672491.25 frames. ], batch size: 174, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:06:04,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=363449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:06:14,971 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=363460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:06:31,469 INFO [train.py:968] (0/2) Epoch 8, batch 44700, giga_loss[loss=0.3419, simple_loss=0.4022, pruned_loss=0.1408, over 29165.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3902, pruned_loss=0.1335, over 5676621.27 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.378, pruned_loss=0.1299, over 5690558.30 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3921, pruned_loss=0.1345, over 5674648.88 frames. ], batch size: 113, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:06:44,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.874e+02 1.498e+03 2.000e+03 3.026e+03 7.044e+03, threshold=3.999e+03, percent-clipped=17.0
+2023-03-04 13:07:18,869 INFO [train.py:968] (0/2) Epoch 8, batch 44750, giga_loss[loss=0.3048, simple_loss=0.3784, pruned_loss=0.1156, over 28770.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3911, pruned_loss=0.1347, over 5665312.09 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3779, pruned_loss=0.1298, over 5692168.80 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3928, pruned_loss=0.1356, over 5662130.82 frames. ], batch size: 119, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:07:30,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=363540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:07:47,269 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=363556.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:08:08,182 INFO [train.py:968] (0/2) Epoch 8, batch 44800, giga_loss[loss=0.3012, simple_loss=0.3724, pruned_loss=0.115, over 28900.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3909, pruned_loss=0.1353, over 5661116.63 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3777, pruned_loss=0.1296, over 5693277.95 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3925, pruned_loss=0.1363, over 5657565.32 frames. ], batch size: 174, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:08:13,607 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 13:08:21,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.607e+02 1.632e+03 2.009e+03 2.831e+03 9.118e+03, threshold=4.017e+03, percent-clipped=11.0
+2023-03-04 13:08:52,958 INFO [train.py:968] (0/2) Epoch 8, batch 44850, giga_loss[loss=0.3546, simple_loss=0.4022, pruned_loss=0.1535, over 28904.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3883, pruned_loss=0.1344, over 5666437.12 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3774, pruned_loss=0.1293, over 5694727.43 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3901, pruned_loss=0.1355, over 5661464.79 frames. ], batch size: 174, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:09:20,844 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.45 vs. limit=5.0
+2023-03-04 13:09:40,387 INFO [train.py:968] (0/2) Epoch 8, batch 44900, giga_loss[loss=0.3424, simple_loss=0.3905, pruned_loss=0.1472, over 28901.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3869, pruned_loss=0.1345, over 5664788.40 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3772, pruned_loss=0.1292, over 5698673.51 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3887, pruned_loss=0.1357, over 5656804.05 frames. ], batch size: 199, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:09:55,271 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.576e+02 1.497e+03 2.399e+03 3.889e+03 1.621e+04, threshold=4.798e+03, percent-clipped=22.0
+2023-03-04 13:10:26,535 INFO [train.py:968] (0/2) Epoch 8, batch 44950, giga_loss[loss=0.3386, simple_loss=0.3923, pruned_loss=0.1425, over 28941.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3841, pruned_loss=0.1327, over 5671616.63 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3769, pruned_loss=0.1289, over 5703000.49 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.386, pruned_loss=0.134, over 5660685.87 frames. ], batch size: 199, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:11:00,534 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=363768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:11:08,978 INFO [train.py:968] (0/2) Epoch 8, batch 45000, giga_loss[loss=0.3069, simple_loss=0.3697, pruned_loss=0.1221, over 28636.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3849, pruned_loss=0.1347, over 5669498.12 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3777, pruned_loss=0.1293, over 5708812.45 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3861, pruned_loss=0.1355, over 5653961.94 frames. ], batch size: 242, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:11:08,984 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 13:11:17,317 INFO [train.py:1012] (0/2) Epoch 8, validation: loss=0.2227, simple_loss=0.3292, pruned_loss=0.05808, over 944034.00 frames. 
+2023-03-04 13:11:17,317 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 13:11:30,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.335e+02 1.524e+03 1.861e+03 2.648e+03 5.318e+03, threshold=3.722e+03, percent-clipped=2.0
+2023-03-04 13:11:53,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=363824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:11:59,684 INFO [train.py:968] (0/2) Epoch 8, batch 45050, giga_loss[loss=0.3514, simple_loss=0.4028, pruned_loss=0.15, over 28335.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3855, pruned_loss=0.1358, over 5662322.61 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3781, pruned_loss=0.1296, over 5714153.38 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3864, pruned_loss=0.1365, over 5643296.55 frames. ], batch size: 369, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:12:05,478 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=363835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:12:26,051 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-04 13:12:40,916 INFO [train.py:968] (0/2) Epoch 8, batch 45100, giga_loss[loss=0.3552, simple_loss=0.3994, pruned_loss=0.1556, over 26782.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.382, pruned_loss=0.1322, over 5664268.27 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3777, pruned_loss=0.1292, over 5719000.12 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3833, pruned_loss=0.1333, over 5642673.22 frames. ], batch size: 555, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:13:00,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.110e+02 1.325e+03 1.628e+03 2.212e+03 4.660e+03, threshold=3.257e+03, percent-clipped=2.0
+2023-03-04 13:13:16,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=363915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:13:27,610 INFO [train.py:968] (0/2) Epoch 8, batch 45150, giga_loss[loss=0.3162, simple_loss=0.3859, pruned_loss=0.1233, over 28236.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.377, pruned_loss=0.1268, over 5667799.97 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3772, pruned_loss=0.1288, over 5721960.23 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3785, pruned_loss=0.128, over 5647279.71 frames. ], batch size: 368, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:13:29,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=363931.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 13:14:03,427 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=363967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:06,407 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=363970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:14,894 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=363978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:15,977 INFO [train.py:968] (0/2) Epoch 8, batch 45200, giga_loss[loss=0.2919, simple_loss=0.3535, pruned_loss=0.1152, over 28772.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3768, pruned_loss=0.1267, over 5662470.40 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3773, pruned_loss=0.1288, over 5724390.06 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3779, pruned_loss=0.1276, over 5643363.11 frames. ], batch size: 99, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:14:18,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=363981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:30,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.168e+02 1.466e+03 1.941e+03 2.685e+03 8.116e+03, threshold=3.881e+03, percent-clipped=17.0
+2023-03-04 13:14:32,785 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=363999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:33,199 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-364000.pt
+2023-03-04 13:14:44,360 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=364010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:59,750 INFO [train.py:968] (0/2) Epoch 8, batch 45250, giga_loss[loss=0.3064, simple_loss=0.3661, pruned_loss=0.1234, over 28928.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3749, pruned_loss=0.1257, over 5670649.81 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3771, pruned_loss=0.1287, over 5718500.47 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3759, pruned_loss=0.1265, over 5659505.20 frames. ], batch size: 186, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:15:30,910 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=364058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:15:33,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=364061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:15:44,758 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=364074.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:15:46,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=364077.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 13:15:48,426 INFO [train.py:968] (0/2) Epoch 8, batch 45300, giga_loss[loss=0.2472, simple_loss=0.321, pruned_loss=0.08676, over 28768.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3734, pruned_loss=0.1259, over 5641206.41 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3775, pruned_loss=0.1291, over 5683032.91 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3737, pruned_loss=0.126, over 5664907.53 frames. ], batch size: 119, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:15:57,217 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=364090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:16:02,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.868e+02 1.552e+03 2.101e+03 2.963e+03 1.055e+04, threshold=4.201e+03, percent-clipped=9.0
+2023-03-04 13:16:11,181 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=364106.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 13:16:32,600 INFO [train.py:968] (0/2) Epoch 8, batch 45350, giga_loss[loss=0.299, simple_loss=0.3704, pruned_loss=0.1138, over 28672.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3762, pruned_loss=0.1274, over 5625509.41 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.378, pruned_loss=0.1295, over 5646499.95 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3759, pruned_loss=0.1271, over 5676563.58 frames. ], batch size: 262, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:16:40,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0175, 1.8817, 1.4388, 1.4508], device='cuda:0'), covar=tensor([0.0639, 0.0632, 0.0969, 0.0990], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0445, 0.0496, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 13:16:42,781 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=364143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:17:10,957 INFO [train.py:968] (0/2) Epoch 8, batch 45400, giga_loss[loss=0.3103, simple_loss=0.3784, pruned_loss=0.1211, over 28850.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3797, pruned_loss=0.1299, over 5551333.62 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3797, pruned_loss=0.1309, over 5561752.56 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3779, pruned_loss=0.1283, over 5670281.38 frames. ], batch size: 112, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:17:25,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.730e+02 1.444e+03 1.991e+03 2.931e+03 1.006e+04, threshold=3.981e+03, percent-clipped=13.0
+2023-03-04 13:17:28,608 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-04 13:17:30,933 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-8.pt
+2023-03-04 13:18:40,354 INFO [train.py:968] (0/2) Epoch 9, batch 50, giga_loss[loss=0.2635, simple_loss=0.3572, pruned_loss=0.08487, over 28851.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3892, pruned_loss=0.1226, over 1263385.95 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3754, pruned_loss=0.112, over 199064.33 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3916, pruned_loss=0.1244, over 1102653.71 frames. ], batch size: 174, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:18:48,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9255, 2.5754, 2.5979, 2.4621], device='cuda:0'), covar=tensor([0.1096, 0.1931, 0.1458, 0.1631], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0733, 0.0650, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 13:19:15,914 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=364286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:19:19,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=364289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:19:26,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.864e+02 1.168e+03 1.475e+03 1.995e+03 4.202e+03, threshold=2.951e+03, percent-clipped=1.0
+2023-03-04 13:19:28,052 INFO [train.py:968] (0/2) Epoch 9, batch 100, giga_loss[loss=0.2878, simple_loss=0.3675, pruned_loss=0.104, over 29047.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3794, pruned_loss=0.1173, over 2246933.21 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3665, pruned_loss=0.1065, over 312364.59 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3812, pruned_loss=0.1188, over 2045651.73 frames. ], batch size: 136, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:19:43,691 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=364318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:20:05,398 INFO [train.py:968] (0/2) Epoch 9, batch 150, giga_loss[loss=0.2539, simple_loss=0.318, pruned_loss=0.09496, over 28787.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3654, pruned_loss=0.111, over 3005570.08 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3618, pruned_loss=0.1068, over 572841.16 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3668, pruned_loss=0.1121, over 2705374.87 frames. ], batch size: 99, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:20:16,160 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3482, 1.5150, 1.4276, 1.1597], device='cuda:0'), covar=tensor([0.1948, 0.1511, 0.1037, 0.1526], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1476, 0.1442, 0.1558], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 13:20:43,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.023e+02 1.082e+03 1.483e+03 2.126e+03 1.214e+04, threshold=2.966e+03, percent-clipped=12.0
+2023-03-04 13:20:44,634 INFO [train.py:968] (0/2) Epoch 9, batch 200, libri_loss[loss=0.2683, simple_loss=0.3562, pruned_loss=0.09023, over 29653.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3517, pruned_loss=0.1039, over 3608918.78 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3592, pruned_loss=0.104, over 728405.55 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3518, pruned_loss=0.1045, over 3300474.83 frames. ], batch size: 88, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:20:47,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5946, 2.2189, 1.6411, 0.6569], device='cuda:0'), covar=tensor([0.3697, 0.2046, 0.2776, 0.4144], device='cuda:0'), in_proj_covar=tensor([0.1484, 0.1419, 0.1439, 0.1198], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 13:21:20,143 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 13:21:21,239 INFO [train.py:968] (0/2) Epoch 9, batch 250, giga_loss[loss=0.198, simple_loss=0.276, pruned_loss=0.05998, over 29001.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3398, pruned_loss=0.0971, over 4073138.33 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.358, pruned_loss=0.1034, over 925043.85 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3383, pruned_loss=0.09693, over 3755801.97 frames. ], batch size: 136, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:21:30,838 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-04 13:22:01,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=364494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:22:02,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.678e+02 9.624e+02 1.213e+03 1.538e+03 2.768e+03, threshold=2.427e+03, percent-clipped=1.0
+2023-03-04 13:22:04,675 INFO [train.py:968] (0/2) Epoch 9, batch 300, giga_loss[loss=0.2103, simple_loss=0.2839, pruned_loss=0.0683, over 29051.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3288, pruned_loss=0.09161, over 4437415.01 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3561, pruned_loss=0.102, over 999020.93 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3271, pruned_loss=0.09127, over 4165108.54 frames. ], batch size: 136, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:22:15,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7700, 1.8700, 1.7680, 1.6703], device='cuda:0'), covar=tensor([0.1479, 0.2035, 0.1922, 0.1857], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0728, 0.0649, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 13:22:50,644 INFO [train.py:968] (0/2) Epoch 9, batch 350, giga_loss[loss=0.2089, simple_loss=0.2859, pruned_loss=0.06593, over 28674.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3207, pruned_loss=0.08768, over 4721535.86 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3538, pruned_loss=0.1007, over 1120455.86 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3185, pruned_loss=0.08714, over 4474674.79 frames. ], batch size: 242, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:22:56,184 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=364556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:23:27,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.724e+02 8.969e+02 1.181e+03 1.749e+03 5.887e+03, threshold=2.362e+03, percent-clipped=11.0
+2023-03-04 13:23:28,318 INFO [train.py:968] (0/2) Epoch 9, batch 400, giga_loss[loss=0.2408, simple_loss=0.3103, pruned_loss=0.08558, over 28798.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3152, pruned_loss=0.08497, over 4942588.65 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3547, pruned_loss=0.1011, over 1190694.32 frames. ], giga_tot_loss[loss=0.2404, simple_loss=0.3126, pruned_loss=0.08415, over 4734399.70 frames. ], batch size: 119, lr: 3.71e-03, grad_scale: 8.0
+2023-03-04 13:23:30,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-04 13:24:09,308 INFO [train.py:968] (0/2) Epoch 9, batch 450, giga_loss[loss=0.2105, simple_loss=0.2857, pruned_loss=0.06762, over 28805.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3131, pruned_loss=0.08411, over 5118850.54 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3555, pruned_loss=0.1019, over 1307552.94 frames. ], giga_tot_loss[loss=0.2376, simple_loss=0.3097, pruned_loss=0.08277, over 4932897.31 frames. ], batch size: 186, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:24:09,764 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 13:24:49,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.549e+02 9.895e+02 1.351e+03 1.996e+03 5.741e+03, threshold=2.702e+03, percent-clipped=18.0
+2023-03-04 13:24:50,548 INFO [train.py:968] (0/2) Epoch 9, batch 500, giga_loss[loss=0.3246, simple_loss=0.3652, pruned_loss=0.142, over 26714.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3115, pruned_loss=0.08339, over 5256072.71 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3557, pruned_loss=0.102, over 1421779.25 frames. ], giga_tot_loss[loss=0.2356, simple_loss=0.3076, pruned_loss=0.08181, over 5091295.79 frames. ], batch size: 555, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:25:16,997 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 13:25:32,101 INFO [train.py:968] (0/2) Epoch 9, batch 550, giga_loss[loss=0.2566, simple_loss=0.3139, pruned_loss=0.09969, over 26730.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3095, pruned_loss=0.08247, over 5355122.34 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3547, pruned_loss=0.1012, over 1532517.36 frames. ], giga_tot_loss[loss=0.2337, simple_loss=0.3054, pruned_loss=0.08094, over 5210404.18 frames. ], batch size: 555, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:25:49,323 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-04 13:26:12,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.333e+02 1.048e+03 1.543e+03 2.141e+03 8.824e+03, threshold=3.086e+03, percent-clipped=16.0
+2023-03-04 13:26:12,797 INFO [train.py:968] (0/2) Epoch 9, batch 600, libri_loss[loss=0.2777, simple_loss=0.3489, pruned_loss=0.1033, over 29652.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.3084, pruned_loss=0.08221, over 5431139.69 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3551, pruned_loss=0.101, over 1636997.34 frames. ], giga_tot_loss[loss=0.2326, simple_loss=0.304, pruned_loss=0.08058, over 5307320.94 frames. ], batch size: 73, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:26:45,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3642, 1.2413, 1.1785, 1.5389], device='cuda:0'), covar=tensor([0.0779, 0.0342, 0.0341, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0117, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:0')
+2023-03-04 13:26:58,376 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 13:26:58,596 INFO [train.py:968] (0/2) Epoch 9, batch 650, giga_loss[loss=0.2391, simple_loss=0.306, pruned_loss=0.08605, over 27913.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3058, pruned_loss=0.08112, over 5481178.22 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3551, pruned_loss=0.1009, over 1679649.53 frames. ], giga_tot_loss[loss=0.2307, simple_loss=0.3019, pruned_loss=0.07972, over 5379864.51 frames. ], batch size: 412, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:27:17,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=364869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:27:39,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.751e+02 9.066e+02 1.125e+03 1.450e+03 3.055e+03, threshold=2.250e+03, percent-clipped=0.0
+2023-03-04 13:27:39,761 INFO [train.py:968] (0/2) Epoch 9, batch 700, giga_loss[loss=0.2057, simple_loss=0.2801, pruned_loss=0.06563, over 28926.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.303, pruned_loss=0.07987, over 5520700.57 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3558, pruned_loss=0.1011, over 1732686.04 frames. ], giga_tot_loss[loss=0.228, simple_loss=0.2991, pruned_loss=0.07844, over 5444506.27 frames. ], batch size: 145, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:28:04,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=364931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:28:15,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0378, 2.0215, 1.5071, 1.6401], device='cuda:0'), covar=tensor([0.0632, 0.0551, 0.0927, 0.0987], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0440, 0.0495, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 13:28:17,371 INFO [train.py:968] (0/2) Epoch 9, batch 750, libri_loss[loss=0.2777, simple_loss=0.3635, pruned_loss=0.09598, over 29378.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3016, pruned_loss=0.07885, over 5565484.64 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.356, pruned_loss=0.1013, over 1885960.88 frames. ], giga_tot_loss[loss=0.225, simple_loss=0.2963, pruned_loss=0.07679, over 5500644.07 frames. ], batch size: 92, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:28:52,388 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=364992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:28:55,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.262e+02 9.115e+02 1.194e+03 1.652e+03 4.200e+03, threshold=2.387e+03, percent-clipped=12.0
+2023-03-04 13:28:56,725 INFO [train.py:968] (0/2) Epoch 9, batch 800, giga_loss[loss=0.2303, simple_loss=0.2806, pruned_loss=0.08997, over 23812.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.2995, pruned_loss=0.0776, over 5597785.54 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3569, pruned_loss=0.1015, over 2041218.89 frames. ], giga_tot_loss[loss=0.2214, simple_loss=0.2928, pruned_loss=0.07498, over 5536389.40 frames. ], batch size: 705, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:29:08,876 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:29:10,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=365015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:29:39,182 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:29:41,885 INFO [train.py:968] (0/2) Epoch 9, batch 850, libri_loss[loss=0.2859, simple_loss=0.3572, pruned_loss=0.1073, over 29529.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3066, pruned_loss=0.08233, over 5610230.05 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3564, pruned_loss=0.1012, over 2117805.16 frames. ], giga_tot_loss[loss=0.2302, simple_loss=0.3006, pruned_loss=0.07996, over 5556266.46 frames. ], batch size: 81, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:30:02,390 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-04 13:30:04,805 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:30:06,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=365077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:30:24,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.658e+02 1.193e+03 1.484e+03 1.857e+03 6.976e+03, threshold=2.969e+03, percent-clipped=10.0
+2023-03-04 13:30:25,108 INFO [train.py:968] (0/2) Epoch 9, batch 900, libri_loss[loss=0.3469, simple_loss=0.4025, pruned_loss=0.1456, over 20166.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3202, pruned_loss=0.08942, over 5617210.94 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3575, pruned_loss=0.1021, over 2162658.13 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3146, pruned_loss=0.08712, over 5582152.49 frames. ], batch size: 187, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:30:31,298 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:30:39,192 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-04 13:31:05,200 INFO [train.py:968] (0/2) Epoch 9, batch 950, giga_loss[loss=0.3317, simple_loss=0.4021, pruned_loss=0.1307, over 28568.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.333, pruned_loss=0.0961, over 5643435.33 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3582, pruned_loss=0.1025, over 2254842.43 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3276, pruned_loss=0.09395, over 5609762.61 frames. ], batch size: 336, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:31:45,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.220e+02 1.361e+03 1.862e+03 2.467e+03 5.433e+03, threshold=3.725e+03, percent-clipped=17.0
+2023-03-04 13:31:46,052 INFO [train.py:968] (0/2) Epoch 9, batch 1000, giga_loss[loss=0.286, simple_loss=0.3697, pruned_loss=0.1011, over 28917.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3417, pruned_loss=0.09994, over 5646132.12 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3588, pruned_loss=0.1029, over 2310679.30 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3369, pruned_loss=0.09806, over 5629394.69 frames. ], batch size: 227, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:31:55,435 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2461, 1.4047, 3.4201, 3.2543], device='cuda:0'), covar=tensor([0.1352, 0.2403, 0.0388, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0610, 0.0563, 0.0807, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 13:32:23,895 INFO [train.py:968] (0/2) Epoch 9, batch 1050, giga_loss[loss=0.2494, simple_loss=0.3375, pruned_loss=0.0807, over 28939.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3456, pruned_loss=0.1002, over 5663958.16 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.358, pruned_loss=0.1025, over 2364334.29 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3419, pruned_loss=0.09887, over 5647422.01 frames. ], batch size: 164, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:32:30,109 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=365254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:32:39,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4619, 1.5164, 1.1964, 1.1270], device='cuda:0'), covar=tensor([0.0659, 0.0462, 0.0914, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0440, 0.0496, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 13:33:09,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.944e+02 1.081e+03 1.293e+03 1.659e+03 4.645e+03, threshold=2.586e+03, percent-clipped=1.0
+2023-03-04 13:33:10,450 INFO [train.py:968] (0/2) Epoch 9, batch 1100, giga_loss[loss=0.3618, simple_loss=0.4041, pruned_loss=0.1598, over 26607.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3481, pruned_loss=0.1006, over 5661232.70 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3586, pruned_loss=0.1026, over 2433958.29 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3447, pruned_loss=0.09942, over 5644083.83 frames. ], batch size: 555, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:33:19,060 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=365310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:33:31,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7293, 1.7202, 1.2263, 1.3893], device='cuda:0'), covar=tensor([0.0710, 0.0595, 0.1041, 0.0945], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0439, 0.0495, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 13:33:47,111 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-04 13:33:49,357 INFO [train.py:968] (0/2) Epoch 9, batch 1150, giga_loss[loss=0.2623, simple_loss=0.3318, pruned_loss=0.09638, over 28557.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3487, pruned_loss=0.1011, over 5682735.83 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3567, pruned_loss=0.1018, over 2551369.16 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3465, pruned_loss=0.1005, over 5663597.25 frames. ], batch size: 85, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:34:00,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=365361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:34:04,906 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=365367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:34:26,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2512, 1.4999, 1.1880, 1.4748], device='cuda:0'), covar=tensor([0.2286, 0.2109, 0.2257, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.1231, 0.0924, 0.1090, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 13:34:30,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.760e+02 1.196e+03 1.457e+03 2.144e+03 5.222e+03, threshold=2.914e+03, percent-clipped=12.0
+2023-03-04 13:34:31,585 INFO [train.py:968] (0/2) Epoch 9, batch 1200, giga_loss[loss=0.2906, simple_loss=0.3617, pruned_loss=0.1097, over 28625.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.351, pruned_loss=0.1029, over 5672967.01 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3568, pruned_loss=0.1014, over 2617061.17 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3491, pruned_loss=0.1025, over 5654491.56 frames. ], batch size: 242, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:34:47,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1582, 3.1401, 2.1435, 1.2364], device='cuda:0'), covar=tensor([0.4023, 0.1928, 0.2630, 0.3949], device='cuda:0'), in_proj_covar=tensor([0.1490, 0.1425, 0.1457, 0.1212], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 13:35:08,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2849, 3.2451, 1.4012, 1.3550], device='cuda:0'), covar=tensor([0.0987, 0.0226, 0.0882, 0.1393], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0488, 0.0323, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 13:35:12,909 INFO [train.py:968] (0/2) Epoch 9, batch 1250, giga_loss[loss=0.3212, simple_loss=0.3793, pruned_loss=0.1315, over 27514.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3535, pruned_loss=0.1045, over 5670480.32 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3567, pruned_loss=0.1011, over 2659107.68 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.352, pruned_loss=0.1044, over 5659461.27 frames. ], batch size: 472, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:35:53,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.611e+02 1.121e+03 1.395e+03 1.643e+03 3.824e+03, threshold=2.789e+03, percent-clipped=2.0
+2023-03-04 13:35:53,232 INFO [train.py:968] (0/2) Epoch 9, batch 1300, giga_loss[loss=0.2562, simple_loss=0.3351, pruned_loss=0.08861, over 28761.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3565, pruned_loss=0.1056, over 5662545.10 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3572, pruned_loss=0.1015, over 2780340.47 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.355, pruned_loss=0.1056, over 5663635.00 frames. ], batch size: 119, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:36:02,875 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:36:04,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=365513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:36:27,184 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:36:29,193 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1807, 1.4909, 1.1827, 0.9374], device='cuda:0'), covar=tensor([0.2479, 0.2251, 0.2602, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.0921, 0.1088, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 13:36:30,858 INFO [train.py:968] (0/2) Epoch 9, batch 1350, giga_loss[loss=0.2637, simple_loss=0.3491, pruned_loss=0.08913, over 28750.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3585, pruned_loss=0.1056, over 5685493.61 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3567, pruned_loss=0.101, over 2857482.82 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3575, pruned_loss=0.1058, over 5681382.44 frames. ], batch size: 284, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:36:58,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8760, 1.1296, 3.2916, 2.8486], device='cuda:0'), covar=tensor([0.1686, 0.2555, 0.0447, 0.0987], device='cuda:0'), in_proj_covar=tensor([0.0608, 0.0558, 0.0798, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 13:37:11,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.913e+02 1.105e+03 1.391e+03 1.881e+03 5.781e+03, threshold=2.783e+03, percent-clipped=9.0
+2023-03-04 13:37:11,683 INFO [train.py:968] (0/2) Epoch 9, batch 1400, giga_loss[loss=0.3309, simple_loss=0.3927, pruned_loss=0.1345, over 28565.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3605, pruned_loss=0.1066, over 5679515.98 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3569, pruned_loss=0.101, over 2995254.29 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3597, pruned_loss=0.107, over 5676027.44 frames. ], batch size: 336, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:37:12,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5050, 2.1633, 1.6702, 0.6656], device='cuda:0'), covar=tensor([0.3758, 0.2042, 0.2784, 0.4101], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1421, 0.1455, 0.1213], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 13:37:34,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=365629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:37:50,571 INFO [train.py:968] (0/2) Epoch 9, batch 1450, giga_loss[loss=0.2795, simple_loss=0.3587, pruned_loss=0.1001, over 28941.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.36, pruned_loss=0.1052, over 5686759.79 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3558, pruned_loss=0.1005, over 3066715.39 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.36, pruned_loss=0.1059, over 5679716.01 frames. ], batch size: 213, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:38:09,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1266, 1.9202, 1.4370, 1.5828], device='cuda:0'), covar=tensor([0.0642, 0.0586, 0.0947, 0.0966], device='cuda:0'), in_proj_covar=tensor([0.0341, 0.0435, 0.0492, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 13:38:11,002 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=365672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:38:20,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5499, 3.4841, 1.6228, 1.6697], device='cuda:0'), covar=tensor([0.0864, 0.0212, 0.0799, 0.1191], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0485, 0.0320, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 13:38:20,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=365685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:38:30,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.781e+02 1.061e+03 1.342e+03 1.739e+03 4.442e+03, threshold=2.683e+03, percent-clipped=5.0
+2023-03-04 13:38:30,603 INFO [train.py:968] (0/2) Epoch 9, batch 1500, giga_loss[loss=0.285, simple_loss=0.3606, pruned_loss=0.1047, over 28517.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3584, pruned_loss=0.1027, over 5696779.76 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3553, pruned_loss=0.1002, over 3108273.51 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3586, pruned_loss=0.1035, over 5689237.55 frames. ], batch size: 60, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:38:53,005 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8680, 1.1225, 3.6309, 3.0463], device='cuda:0'), covar=tensor([0.1777, 0.2561, 0.0418, 0.0775], device='cuda:0'), in_proj_covar=tensor([0.0609, 0.0559, 0.0805, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 13:39:00,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=365736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:39:09,512 INFO [train.py:968] (0/2) Epoch 9, batch 1550, giga_loss[loss=0.259, simple_loss=0.351, pruned_loss=0.08347, over 28378.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3567, pruned_loss=0.1013, over 5698503.51 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3561, pruned_loss=0.1007, over 3162086.48 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3565, pruned_loss=0.1017, over 5690379.75 frames. ], batch size: 77, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:39:17,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3705, 1.5117, 1.2297, 1.3990], device='cuda:0'), covar=tensor([0.1459, 0.1736, 0.1918, 0.1676], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0722, 0.0652, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 13:39:30,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:39:32,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=365775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:39:51,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.466e+02 1.114e+03 1.531e+03 2.296e+03 5.161e+03, threshold=3.061e+03, percent-clipped=16.0
+2023-03-04 13:39:51,877 INFO [train.py:968] (0/2) Epoch 9, batch 1600, libri_loss[loss=0.2608, simple_loss=0.3349, pruned_loss=0.09332, over 29575.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3571, pruned_loss=0.1024, over 5709522.35 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3563, pruned_loss=0.101, over 3229916.85 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3569, pruned_loss=0.1025, over 5698589.90 frames. ], batch size: 74, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:39:56,035 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:40:14,470 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:40:17,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=365831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:40:32,397 INFO [train.py:968] (0/2) Epoch 9, batch 1650, giga_loss[loss=0.2958, simple_loss=0.3612, pruned_loss=0.1152, over 28230.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3595, pruned_loss=0.1069, over 5715153.83 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.356, pruned_loss=0.1008, over 3282898.23 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3596, pruned_loss=0.1072, over 5703052.07 frames. ], batch size: 77, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:40:32,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6396, 1.8047, 1.9854, 1.5260], device='cuda:0'), covar=tensor([0.1625, 0.2002, 0.1218, 0.1442], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0704, 0.0837, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 13:40:42,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:40:58,658 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:41:01,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=365882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:41:10,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3710, 3.2793, 1.5376, 1.4221], device='cuda:0'), covar=tensor([0.0837, 0.0317, 0.0757, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0488, 0.0322, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 13:41:14,919 INFO [train.py:968] (0/2) Epoch 9, batch 1700, giga_loss[loss=0.2964, simple_loss=0.3683, pruned_loss=0.1122, over 28502.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3621, pruned_loss=0.1113, over 5701417.87 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3557, pruned_loss=0.1006, over 3331876.48 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3625, pruned_loss=0.1118, over 5690961.93 frames. ], batch size: 71, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:41:15,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.241e+02 1.131e+03 1.504e+03 2.227e+03 4.655e+03, threshold=3.007e+03, percent-clipped=8.0
+2023-03-04 13:41:24,522 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:41:54,102 INFO [train.py:968] (0/2) Epoch 9, batch 1750, giga_loss[loss=0.2709, simple_loss=0.3422, pruned_loss=0.09979, over 28888.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3605, pruned_loss=0.1109, over 5704143.80 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3557, pruned_loss=0.1004, over 3408106.29 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3609, pruned_loss=0.1118, over 5690035.17 frames. ], batch size: 145, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:42:33,802 INFO [train.py:968] (0/2) Epoch 9, batch 1800, giga_loss[loss=0.2744, simple_loss=0.3511, pruned_loss=0.09883, over 28937.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3596, pruned_loss=0.1107, over 5716750.92 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3562, pruned_loss=0.1006, over 3493946.28 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3598, pruned_loss=0.1116, over 5699775.85 frames. ], batch size: 136, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:42:34,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.856e+02 1.179e+03 1.431e+03 1.973e+03 4.506e+03, threshold=2.862e+03, percent-clipped=6.0
+2023-03-04 13:42:35,074 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-366000.pt
+2023-03-04 13:42:36,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7435, 1.6935, 1.7027, 1.6475], device='cuda:0'), covar=tensor([0.1187, 0.1671, 0.1588, 0.1479], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0716, 0.0644, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 13:42:56,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9451, 1.9232, 1.8051, 1.7178], device='cuda:0'), covar=tensor([0.1189, 0.1792, 0.1652, 0.1667], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0718, 0.0646, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 13:43:13,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:43:14,930 INFO [train.py:968] (0/2) Epoch 9, batch 1850, giga_loss[loss=0.282, simple_loss=0.3591, pruned_loss=0.1024, over 28754.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3589, pruned_loss=0.1098, over 5716710.98 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3566, pruned_loss=0.1007, over 3517115.00 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3589, pruned_loss=0.1105, over 5702442.81 frames. ], batch size: 262, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:43:55,453 INFO [train.py:968] (0/2) Epoch 9, batch 1900, giga_loss[loss=0.323, simple_loss=0.3746, pruned_loss=0.1357, over 23509.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3577, pruned_loss=0.1083, over 5699869.78 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3565, pruned_loss=0.1006, over 3574756.74 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3578, pruned_loss=0.1092, over 5695539.00 frames. ], batch size: 705, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:43:56,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.481e+02 1.015e+03 1.188e+03 1.674e+03 5.676e+03, threshold=2.375e+03, percent-clipped=1.0
+2023-03-04 13:44:28,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366135.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 13:44:36,852 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:44:39,376 INFO [train.py:968] (0/2) Epoch 9, batch 1950, giga_loss[loss=0.2518, simple_loss=0.3243, pruned_loss=0.08968, over 28951.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3539, pruned_loss=0.1056, over 5694706.69 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3566, pruned_loss=0.1005, over 3618947.68 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3538, pruned_loss=0.1064, over 5689554.48 frames. ], batch size: 136, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:44:52,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0358, 1.1318, 3.4723, 3.0393], device='cuda:0'), covar=tensor([0.1611, 0.2512, 0.0409, 0.1147], device='cuda:0'), in_proj_covar=tensor([0.0609, 0.0562, 0.0804, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 13:45:16,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:45:19,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:45:23,749 INFO [train.py:968] (0/2) Epoch 9, batch 2000, giga_loss[loss=0.2291, simple_loss=0.306, pruned_loss=0.07605, over 28411.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3474, pruned_loss=0.102, over 5685248.97 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3566, pruned_loss=0.1005, over 3660879.98 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3472, pruned_loss=0.1027, over 5680757.11 frames. ], batch size: 369, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:45:24,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.376e+02 1.022e+03 1.248e+03 1.592e+03 4.064e+03, threshold=2.496e+03, percent-clipped=13.0
+2023-03-04 13:45:46,436 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:46:10,281 INFO [train.py:968] (0/2) Epoch 9, batch 2050, giga_loss[loss=0.2545, simple_loss=0.3239, pruned_loss=0.09258, over 28987.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.341, pruned_loss=0.09885, over 5678280.12 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3571, pruned_loss=0.1009, over 3671825.58 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3405, pruned_loss=0.09919, over 5673978.21 frames. ], batch size: 227, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:46:51,649 INFO [train.py:968] (0/2) Epoch 9, batch 2100, giga_loss[loss=0.2487, simple_loss=0.3297, pruned_loss=0.08391, over 28936.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3396, pruned_loss=0.09778, over 5683044.71 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3577, pruned_loss=0.1013, over 3734365.64 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3384, pruned_loss=0.09775, over 5676677.27 frames. ], batch size: 213, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:46:52,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.367e+02 9.500e+02 1.250e+03 1.574e+03 4.341e+03, threshold=2.500e+03, percent-clipped=7.0
+2023-03-04 13:46:58,888 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:47:11,923 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:47:17,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5673, 1.4723, 1.7376, 1.3580], device='cuda:0'), covar=tensor([0.1535, 0.2221, 0.1201, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0705, 0.0837, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 13:47:21,980 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 13:47:31,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4775, 1.7037, 1.3756, 1.7796], device='cuda:0'), covar=tensor([0.2467, 0.2303, 0.2604, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1225, 0.0911, 0.1082, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 13:47:32,188 INFO [train.py:968] (0/2) Epoch 9, batch 2150, libri_loss[loss=0.2802, simple_loss=0.3687, pruned_loss=0.09584, over 29245.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.34, pruned_loss=0.09726, over 5694453.87 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3579, pruned_loss=0.1012, over 3788194.26 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3385, pruned_loss=0.09719, over 5684254.53 frames. ], batch size: 94, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:47:53,778 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:48:01,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4614, 1.6330, 1.2843, 1.7245], device='cuda:0'), covar=tensor([0.2233, 0.2217, 0.2506, 0.1991], device='cuda:0'), in_proj_covar=tensor([0.1225, 0.0910, 0.1082, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 13:48:09,700 INFO [train.py:968] (0/2) Epoch 9, batch 2200, giga_loss[loss=0.2933, simple_loss=0.3507, pruned_loss=0.1179, over 28841.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3398, pruned_loss=0.09712, over 5700800.20 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3581, pruned_loss=0.1012, over 3819305.22 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3383, pruned_loss=0.09702, over 5690393.62 frames. ], batch size: 99, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:48:10,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.894e+02 1.098e+03 1.261e+03 1.512e+03 3.943e+03, threshold=2.522e+03, percent-clipped=7.0
+2023-03-04 13:48:23,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6132, 2.2281, 1.6642, 1.2687], device='cuda:0'), covar=tensor([0.2498, 0.1359, 0.1521, 0.2200], device='cuda:0'), in_proj_covar=tensor([0.1606, 0.1471, 0.1450, 0.1554], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 13:48:48,121 INFO [train.py:968] (0/2) Epoch 9, batch 2250, libri_loss[loss=0.223, simple_loss=0.304, pruned_loss=0.07104, over 29417.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3384, pruned_loss=0.09633, over 5705177.25 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3588, pruned_loss=0.1013, over 3887107.19 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3361, pruned_loss=0.09602, over 5694126.22 frames. ], batch size: 67, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:48:50,657 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-04 13:49:02,372 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:49:26,045 INFO [train.py:968] (0/2) Epoch 9, batch 2300, giga_loss[loss=0.2221, simple_loss=0.3023, pruned_loss=0.07091, over 29016.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3356, pruned_loss=0.09497, over 5706481.54 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3585, pruned_loss=0.101, over 3936719.34 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3334, pruned_loss=0.09481, over 5693510.34 frames. ], batch size: 164, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:49:26,741 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.245e+02 9.369e+02 1.173e+03 1.563e+03 4.964e+03, threshold=2.346e+03, percent-clipped=6.0
+2023-03-04 13:49:34,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366510.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 13:49:43,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:49:46,555 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6778, 2.3719, 2.2161, 2.1926], device='cuda:0'), covar=tensor([0.1100, 0.1838, 0.1668, 0.1678], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0729, 0.0655, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0009], device='cuda:0')
+2023-03-04 13:49:55,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8234, 2.4774, 2.4978, 2.3706], device='cuda:0'), covar=tensor([0.1231, 0.2139, 0.1653, 0.1673], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0730, 0.0656, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 13:50:03,464 INFO [train.py:968] (0/2) Epoch 9, batch 2350, giga_loss[loss=0.3056, simple_loss=0.36, pruned_loss=0.1257, over 27558.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3333, pruned_loss=0.09423, over 5716964.20 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3584, pruned_loss=0.1009, over 3946711.74 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3315, pruned_loss=0.09415, over 5705789.28 frames. ], batch size: 472, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:50:16,042 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-04 13:50:19,949 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-04 13:50:42,963 INFO [train.py:968] (0/2) Epoch 9, batch 2400, giga_loss[loss=0.2169, simple_loss=0.3013, pruned_loss=0.06626, over 28863.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3308, pruned_loss=0.09315, over 5723329.31 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3588, pruned_loss=0.1009, over 3984154.11 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3288, pruned_loss=0.09299, over 5712492.18 frames. ], batch size: 174, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:50:43,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.530e+02 9.467e+02 1.165e+03 1.583e+03 5.336e+03, threshold=2.330e+03, percent-clipped=10.0
+2023-03-04 13:51:20,024 INFO [train.py:968] (0/2) Epoch 9, batch 2450, giga_loss[loss=0.2277, simple_loss=0.3012, pruned_loss=0.07713, over 28882.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3279, pruned_loss=0.09158, over 5731006.52 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3586, pruned_loss=0.1004, over 4022487.03 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3259, pruned_loss=0.09159, over 5718881.74 frames. ], batch size: 99, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:51:23,771 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366653.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 13:51:25,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366656.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:51:28,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7709, 1.2466, 2.7944, 2.7214], device='cuda:0'), covar=tensor([0.1701, 0.2291, 0.0577, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0613, 0.0560, 0.0801, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 13:51:30,166 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:51:32,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:51:43,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:51:46,794 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366685.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:51:49,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1737, 0.8442, 0.8398, 1.4560], device='cuda:0'), covar=tensor([0.0775, 0.0344, 0.0343, 0.0775], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:0')
+2023-03-04 13:51:51,064 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 13:51:53,989 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:51:55,914 INFO [train.py:968] (0/2) Epoch 9, batch 2500, giga_loss[loss=0.2565, simple_loss=0.3261, pruned_loss=0.09348, over 28844.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3257, pruned_loss=0.09011, over 5734286.39 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3579, pruned_loss=0.09993, over 4086803.54 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3235, pruned_loss=0.09012, over 5719617.54 frames. ], batch size: 199, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:51:56,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:51:56,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.539e+02 9.242e+02 1.246e+03 1.701e+03 4.471e+03, threshold=2.492e+03, percent-clipped=11.0
+2023-03-04 13:52:36,941 INFO [train.py:968] (0/2) Epoch 9, batch 2550, giga_loss[loss=0.2199, simple_loss=0.3032, pruned_loss=0.06833, over 28918.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3251, pruned_loss=0.08995, over 5731176.73 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3586, pruned_loss=0.1003, over 4140509.88 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3219, pruned_loss=0.08944, over 5714852.02 frames. ], batch size: 174, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:52:41,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:16,003 INFO [train.py:968] (0/2) Epoch 9, batch 2600, giga_loss[loss=0.2168, simple_loss=0.2948, pruned_loss=0.06941, over 28490.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3245, pruned_loss=0.08955, over 5730868.80 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3591, pruned_loss=0.1004, over 4175005.53 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.321, pruned_loss=0.0889, over 5715305.09 frames. ], batch size: 78, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:53:17,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.336e+02 9.299e+02 1.133e+03 1.457e+03 5.176e+03, threshold=2.266e+03, percent-clipped=7.0
+2023-03-04 13:53:26,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7184, 1.7132, 1.2175, 1.3856], device='cuda:0'), covar=tensor([0.0651, 0.0500, 0.0938, 0.0942], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0441, 0.0496, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 13:53:35,405 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:37,480 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:46,857 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:48,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:48,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:51,404 INFO [train.py:968] (0/2) Epoch 9, batch 2650, giga_loss[loss=0.2358, simple_loss=0.3088, pruned_loss=0.08143, over 28683.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3238, pruned_loss=0.08907, over 5732285.94 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3599, pruned_loss=0.1008, over 4221950.10 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3195, pruned_loss=0.08789, over 5719102.24 frames. ], batch size: 262, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:53:58,847 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:54:12,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:54:31,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:54:32,021 INFO [train.py:968] (0/2) Epoch 9, batch 2700, giga_loss[loss=0.2463, simple_loss=0.3183, pruned_loss=0.08714, over 29033.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3272, pruned_loss=0.09147, over 5721939.46 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3604, pruned_loss=0.1013, over 4269607.34 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3223, pruned_loss=0.08991, over 5708416.47 frames. ], batch size: 128, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:54:33,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:54:34,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.690e+02 9.651e+02 1.226e+03 1.671e+03 4.052e+03, threshold=2.452e+03, percent-clipped=11.0
+2023-03-04 13:54:34,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5477, 1.7774, 1.3651, 1.1797], device='cuda:0'), covar=tensor([0.1711, 0.1450, 0.1169, 0.1587], device='cuda:0'), in_proj_covar=tensor([0.1603, 0.1467, 0.1451, 0.1554], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 13:54:55,795 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:54:57,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8245, 1.6782, 1.2607, 1.3977], device='cuda:0'), covar=tensor([0.0653, 0.0624, 0.0952, 0.1003], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0442, 0.0496, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 13:55:11,682 INFO [train.py:968] (0/2) Epoch 9, batch 2750, giga_loss[loss=0.312, simple_loss=0.3816, pruned_loss=0.1212, over 28845.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3329, pruned_loss=0.09522, over 5723522.54 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3598, pruned_loss=0.1009, over 4302120.67 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3288, pruned_loss=0.09406, over 5709439.34 frames. ], batch size: 186, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 13:55:45,213 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:55:47,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:55:53,856 INFO [train.py:968] (0/2) Epoch 9, batch 2800, giga_loss[loss=0.329, simple_loss=0.3817, pruned_loss=0.1381, over 28909.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3403, pruned_loss=0.1004, over 5711757.22 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3593, pruned_loss=0.1006, over 4360934.36 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3366, pruned_loss=0.09956, over 5696774.52 frames. ], batch size: 186, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 13:55:55,282 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.759e+02 1.191e+03 1.544e+03 2.287e+03 5.714e+03, threshold=3.089e+03, percent-clipped=22.0
+2023-03-04 13:56:04,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2365, 1.2231, 1.0460, 0.9666], device='cuda:0'), covar=tensor([0.0620, 0.0430, 0.0906, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0343, 0.0442, 0.0496, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 13:56:09,927 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=367018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:56:35,276 INFO [train.py:968] (0/2) Epoch 9, batch 2850, giga_loss[loss=0.3076, simple_loss=0.3663, pruned_loss=0.1244, over 28548.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3469, pruned_loss=0.1046, over 5702728.38 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3588, pruned_loss=0.1006, over 4405610.80 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3438, pruned_loss=0.104, over 5686537.49 frames. ], batch size: 85, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 13:57:18,097 INFO [train.py:968] (0/2) Epoch 9, batch 2900, giga_loss[loss=0.335, simple_loss=0.3816, pruned_loss=0.1442, over 23819.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3523, pruned_loss=0.107, over 5690836.11 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3582, pruned_loss=0.1002, over 4446665.86 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.35, pruned_loss=0.1069, over 5675536.66 frames. ], batch size: 710, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 13:57:19,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.505e+02 1.114e+03 1.424e+03 2.128e+03 5.660e+03, threshold=2.847e+03, percent-clipped=7.0
+2023-03-04 13:57:44,636 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=367130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:57:58,446 INFO [train.py:968] (0/2) Epoch 9, batch 2950, giga_loss[loss=0.2758, simple_loss=0.3503, pruned_loss=0.1006, over 28770.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3571, pruned_loss=0.1087, over 5690372.04 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3579, pruned_loss=0.1001, over 4496747.33 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3553, pruned_loss=0.109, over 5671294.26 frames. ], batch size: 119, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 13:57:58,861 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-04 13:58:04,831 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=367157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:58:43,849 INFO [train.py:968] (0/2) Epoch 9, batch 3000, giga_loss[loss=0.3206, simple_loss=0.3945, pruned_loss=0.1233, over 28707.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.362, pruned_loss=0.1116, over 5700453.45 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3574, pruned_loss=0.09994, over 4531768.49 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3611, pruned_loss=0.1122, over 5680294.20 frames. ], batch size: 242, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 13:58:43,853 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 13:58:52,132 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2286, simple_loss=0.3318, pruned_loss=0.06271, over 944034.00 frames. 
+2023-03-04 13:58:52,132 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 13:58:53,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.841e+02 1.101e+03 1.346e+03 1.794e+03 5.662e+03, threshold=2.693e+03, percent-clipped=7.0
+2023-03-04 13:59:18,466 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1953, 2.2633, 1.2725, 1.3454], device='cuda:0'), covar=tensor([0.0747, 0.0350, 0.0685, 0.1095], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0487, 0.0320, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 13:59:31,083 INFO [train.py:968] (0/2) Epoch 9, batch 3050, giga_loss[loss=0.2422, simple_loss=0.3243, pruned_loss=0.07999, over 28866.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.359, pruned_loss=0.1092, over 5691251.56 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3566, pruned_loss=0.09938, over 4578132.87 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3589, pruned_loss=0.1105, over 5669102.18 frames. ], batch size: 112, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:00:05,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2807, 1.4868, 1.5656, 1.3562], device='cuda:0'), covar=tensor([0.1356, 0.1346, 0.1686, 0.1515], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0726, 0.0653, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 14:00:10,439 INFO [train.py:968] (0/2) Epoch 9, batch 3100, giga_loss[loss=0.2512, simple_loss=0.3311, pruned_loss=0.0857, over 28541.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.354, pruned_loss=0.1054, over 5695301.60 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3561, pruned_loss=0.09928, over 4619840.86 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3544, pruned_loss=0.1067, over 5674433.17 frames. ], batch size: 71, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:00:11,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4673, 1.7802, 1.4268, 1.4008], device='cuda:0'), covar=tensor([0.1906, 0.1401, 0.1120, 0.1519], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1471, 0.1455, 0.1565], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 14:00:11,808 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.061e+02 1.011e+03 1.251e+03 1.804e+03 4.497e+03, threshold=2.503e+03, percent-clipped=11.0
+2023-03-04 14:00:51,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4174, 1.5950, 1.2679, 1.6105], device='cuda:0'), covar=tensor([0.2247, 0.2111, 0.2344, 0.2004], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.0911, 0.1081, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 14:00:53,190 INFO [train.py:968] (0/2) Epoch 9, batch 3150, giga_loss[loss=0.3319, simple_loss=0.3794, pruned_loss=0.1422, over 23403.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3532, pruned_loss=0.1046, over 5685261.42 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3561, pruned_loss=0.09933, over 4626299.66 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3534, pruned_loss=0.1057, over 5667979.39 frames. ], batch size: 705, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:01:20,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=367383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:01:32,794 INFO [train.py:968] (0/2) Epoch 9, batch 3200, giga_loss[loss=0.2773, simple_loss=0.3549, pruned_loss=0.09981, over 28414.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3547, pruned_loss=0.1055, over 5675813.81 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3562, pruned_loss=0.09945, over 4654704.51 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3548, pruned_loss=0.1064, over 5669460.94 frames. ], batch size: 71, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:01:34,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.867e+02 1.113e+03 1.394e+03 2.065e+03 7.163e+03, threshold=2.788e+03, percent-clipped=17.0
+2023-03-04 14:01:35,198 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.08 vs. limit=2.0
+2023-03-04 14:01:56,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5028, 3.4071, 1.6695, 1.5997], device='cuda:0'), covar=tensor([0.0903, 0.0260, 0.0760, 0.1271], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0485, 0.0320, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 14:02:10,583 INFO [train.py:968] (0/2) Epoch 9, batch 3250, giga_loss[loss=0.3027, simple_loss=0.3776, pruned_loss=0.1139, over 28848.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3575, pruned_loss=0.1074, over 5667835.24 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3561, pruned_loss=0.09963, over 4662762.72 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3576, pruned_loss=0.108, over 5669626.31 frames. ], batch size: 174, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:02:38,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5324, 4.1739, 1.7127, 1.6355], device='cuda:0'), covar=tensor([0.0849, 0.0236, 0.0744, 0.1191], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0486, 0.0321, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 14:02:45,960 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2681, 3.1911, 1.4694, 1.3303], device='cuda:0'), covar=tensor([0.0901, 0.0272, 0.0784, 0.1282], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0486, 0.0321, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 14:02:50,646 INFO [train.py:968] (0/2) Epoch 9, batch 3300, giga_loss[loss=0.3355, simple_loss=0.4035, pruned_loss=0.1337, over 28341.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3598, pruned_loss=0.1087, over 5677134.01 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3566, pruned_loss=0.09991, over 4677857.31 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3595, pruned_loss=0.1091, over 5683444.27 frames. ], batch size: 65, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:02:52,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.304e+02 1.172e+03 1.466e+03 1.797e+03 6.655e+03, threshold=2.932e+03, percent-clipped=4.0
+2023-03-04 14:02:56,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=367505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:03:14,241 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 14:03:17,885 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=367532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:03:30,507 INFO [train.py:968] (0/2) Epoch 9, batch 3350, giga_loss[loss=0.2832, simple_loss=0.3481, pruned_loss=0.1091, over 28297.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3615, pruned_loss=0.1102, over 5680278.71 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3564, pruned_loss=0.09959, over 4713262.64 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3615, pruned_loss=0.1111, over 5680153.18 frames. ], batch size: 77, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:03:32,187 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=367550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:04:14,099 INFO [train.py:968] (0/2) Epoch 9, batch 3400, giga_loss[loss=0.2754, simple_loss=0.35, pruned_loss=0.1004, over 28815.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3622, pruned_loss=0.1112, over 5686125.39 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3562, pruned_loss=0.09934, over 4742297.97 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3626, pruned_loss=0.1123, over 5681363.37 frames. ], batch size: 284, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:04:16,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.349e+02 1.239e+03 1.608e+03 2.268e+03 5.010e+03, threshold=3.216e+03, percent-clipped=7.0
+2023-03-04 14:04:34,184 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=367622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:04:41,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3329, 3.1395, 2.8638, 1.8576], device='cuda:0'), covar=tensor([0.0715, 0.0871, 0.0827, 0.1977], device='cuda:0'), in_proj_covar=tensor([0.0963, 0.0902, 0.0806, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 14:04:54,092 INFO [train.py:968] (0/2) Epoch 9, batch 3450, giga_loss[loss=0.2768, simple_loss=0.3335, pruned_loss=0.1101, over 23798.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3627, pruned_loss=0.112, over 5665752.13 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3563, pruned_loss=0.09941, over 4751640.96 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.363, pruned_loss=0.113, over 5670714.65 frames. ], batch size: 705, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:04:54,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=367648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:04:56,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=367651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:05:14,239 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=367675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:05:16,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=367678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:05:17,602 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=367680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:05:31,013 INFO [train.py:968] (0/2) Epoch 9, batch 3500, giga_loss[loss=0.304, simple_loss=0.3843, pruned_loss=0.1118, over 28746.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3622, pruned_loss=0.1102, over 5684916.17 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3567, pruned_loss=0.09952, over 4795729.09 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3623, pruned_loss=0.1114, over 5681941.83 frames. ], batch size: 119, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:05:35,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.512e+02 1.138e+03 1.494e+03 1.934e+03 3.314e+03, threshold=2.988e+03, percent-clipped=1.0
+2023-03-04 14:05:38,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7618, 1.6919, 1.2627, 1.4134], device='cuda:0'), covar=tensor([0.0710, 0.0609, 0.1038, 0.0958], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0438, 0.0495, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 14:05:39,882 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=367707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:06:13,119 INFO [train.py:968] (0/2) Epoch 9, batch 3550, giga_loss[loss=0.2794, simple_loss=0.3687, pruned_loss=0.09501, over 28869.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3619, pruned_loss=0.1089, over 5683600.75 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3563, pruned_loss=0.09947, over 4807227.81 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3624, pruned_loss=0.1101, over 5685891.76 frames. ], batch size: 213, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:06:21,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=367758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:06:53,355 INFO [train.py:968] (0/2) Epoch 9, batch 3600, giga_loss[loss=0.2598, simple_loss=0.3422, pruned_loss=0.08869, over 28944.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3616, pruned_loss=0.1077, over 5690167.91 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3567, pruned_loss=0.09959, over 4842723.32 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3619, pruned_loss=0.1088, over 5687306.72 frames. ], batch size: 227, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:06:57,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.371e+02 1.050e+03 1.260e+03 1.558e+03 3.393e+03, threshold=2.519e+03, percent-clipped=2.0
+2023-03-04 14:07:31,441 INFO [train.py:968] (0/2) Epoch 9, batch 3650, giga_loss[loss=0.2801, simple_loss=0.3612, pruned_loss=0.0995, over 28842.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.36, pruned_loss=0.1069, over 5700902.93 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3568, pruned_loss=0.09991, over 4874329.44 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3602, pruned_loss=0.1077, over 5692583.12 frames. ], batch size: 199, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:07:48,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9866, 0.9544, 3.6563, 2.9876], device='cuda:0'), covar=tensor([0.1633, 0.2622, 0.0378, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0615, 0.0560, 0.0801, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 14:08:11,669 INFO [train.py:968] (0/2) Epoch 9, batch 3700, giga_loss[loss=0.2933, simple_loss=0.3575, pruned_loss=0.1146, over 27605.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3566, pruned_loss=0.1054, over 5698163.75 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3561, pruned_loss=0.09957, over 4909528.60 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3574, pruned_loss=0.1066, over 5685212.09 frames. ], batch size: 472, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:08:13,963 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=367901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:08:14,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.492e+02 1.034e+03 1.252e+03 1.769e+03 3.300e+03, threshold=2.503e+03, percent-clipped=8.0
+2023-03-04 14:08:15,473 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-04 14:08:16,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=367904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:08:33,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=367925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:08:39,530 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=367933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:08:50,168 INFO [train.py:968] (0/2) Epoch 9, batch 3750, giga_loss[loss=0.3212, simple_loss=0.3772, pruned_loss=0.1326, over 27883.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3538, pruned_loss=0.1037, over 5706867.78 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3561, pruned_loss=0.09952, over 4924138.09 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3545, pruned_loss=0.1048, over 5694177.84 frames. ], batch size: 412, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:08:53,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2575, 4.0064, 3.8256, 1.8305], device='cuda:0'), covar=tensor([0.0555, 0.0787, 0.0831, 0.2098], device='cuda:0'), in_proj_covar=tensor([0.0964, 0.0898, 0.0803, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 14:09:28,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=367997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:09:28,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4666, 1.8386, 1.4630, 1.3653], device='cuda:0'), covar=tensor([0.1644, 0.1240, 0.1303, 0.1471], device='cuda:0'), in_proj_covar=tensor([0.1620, 0.1481, 0.1456, 0.1552], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 14:09:29,354 INFO [train.py:968] (0/2) Epoch 9, batch 3800, giga_loss[loss=0.3122, simple_loss=0.3722, pruned_loss=0.1261, over 27664.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3544, pruned_loss=0.1044, over 5694672.86 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3568, pruned_loss=0.09987, over 4945947.16 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3543, pruned_loss=0.1051, over 5689562.35 frames. ], batch size: 472, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:09:31,890 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-368000.pt
+2023-03-04 14:09:32,933 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:09:35,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.722e+02 9.969e+02 1.147e+03 1.569e+03 3.340e+03, threshold=2.293e+03, percent-clipped=5.0
+2023-03-04 14:10:11,597 INFO [train.py:968] (0/2) Epoch 9, batch 3850, giga_loss[loss=0.2692, simple_loss=0.3503, pruned_loss=0.09406, over 28897.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3555, pruned_loss=0.1052, over 5698336.91 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3568, pruned_loss=0.09987, over 4945947.16 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3554, pruned_loss=0.1058, over 5694359.33 frames. ], batch size: 213, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:10:24,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:10:27,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:10:30,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:10:48,081 INFO [train.py:968] (0/2) Epoch 9, batch 3900, giga_loss[loss=0.2677, simple_loss=0.3395, pruned_loss=0.098, over 27994.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3551, pruned_loss=0.1043, over 5698340.30 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3567, pruned_loss=0.1, over 4972304.54 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3551, pruned_loss=0.1048, over 5698448.64 frames. ], batch size: 77, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:10:49,666 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:10:51,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.446e+02 1.015e+03 1.230e+03 1.807e+03 7.482e+03, threshold=2.459e+03, percent-clipped=14.0
+2023-03-04 14:11:20,395 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:11:22,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:11:24,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368146.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:11:27,119 INFO [train.py:968] (0/2) Epoch 9, batch 3950, libri_loss[loss=0.2664, simple_loss=0.3511, pruned_loss=0.09087, over 29520.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3547, pruned_loss=0.1035, over 5691280.91 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3566, pruned_loss=0.1, over 4981062.26 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3547, pruned_loss=0.104, over 5703001.27 frames. ], batch size: 81, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:11:44,080 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:12:02,258 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-04 14:12:04,082 INFO [train.py:968] (0/2) Epoch 9, batch 4000, giga_loss[loss=0.3136, simple_loss=0.3799, pruned_loss=0.1236, over 28823.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3551, pruned_loss=0.1044, over 5688372.04 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.357, pruned_loss=0.1003, over 4994931.12 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3548, pruned_loss=0.1046, over 5697790.12 frames. ], batch size: 199, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:12:07,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.415e+02 9.797e+02 1.296e+03 1.737e+03 7.741e+03, threshold=2.593e+03, percent-clipped=11.0
+2023-03-04 14:12:19,685 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:12:29,248 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.22 vs. limit=5.0
+2023-03-04 14:12:41,140 INFO [train.py:968] (0/2) Epoch 9, batch 4050, giga_loss[loss=0.274, simple_loss=0.3493, pruned_loss=0.09934, over 29041.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3539, pruned_loss=0.1041, over 5697058.49 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3571, pruned_loss=0.1004, over 5013485.23 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3535, pruned_loss=0.1043, over 5703467.36 frames. ], batch size: 128, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:12:46,069 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5049, 3.4612, 1.5973, 1.5478], device='cuda:0'), covar=tensor([0.0898, 0.0248, 0.0804, 0.1277], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0481, 0.0321, 0.0376], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 14:13:18,204 INFO [train.py:968] (0/2) Epoch 9, batch 4100, giga_loss[loss=0.2433, simple_loss=0.3159, pruned_loss=0.08537, over 28691.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3495, pruned_loss=0.1014, over 5706470.11 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.356, pruned_loss=0.09966, over 5043864.41 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3499, pruned_loss=0.1023, over 5705534.06 frames. ], batch size: 99, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:13:22,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.877e+02 1.015e+03 1.313e+03 1.675e+03 5.139e+03, threshold=2.626e+03, percent-clipped=14.0
+2023-03-04 14:13:34,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5983, 1.8513, 1.4388, 1.9787], device='cuda:0'), covar=tensor([0.2393, 0.2241, 0.2494, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.1227, 0.0907, 0.1084, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 14:13:56,517 INFO [train.py:968] (0/2) Epoch 9, batch 4150, giga_loss[loss=0.3326, simple_loss=0.3736, pruned_loss=0.1459, over 23260.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3473, pruned_loss=0.1003, over 5712788.05 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3553, pruned_loss=0.09925, over 5076562.15 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3479, pruned_loss=0.1013, over 5706369.83 frames. ], batch size: 705, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:14:16,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:14:33,167 INFO [train.py:968] (0/2) Epoch 9, batch 4200, giga_loss[loss=0.3384, simple_loss=0.399, pruned_loss=0.1389, over 27864.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3485, pruned_loss=0.1015, over 5713681.57 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3556, pruned_loss=0.09925, over 5092785.56 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3486, pruned_loss=0.1024, over 5705486.39 frames. ], batch size: 412, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:14:39,016 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.007e+02 1.119e+03 1.419e+03 1.928e+03 1.002e+04, threshold=2.839e+03, percent-clipped=11.0
+2023-03-04 14:15:14,235 INFO [train.py:968] (0/2) Epoch 9, batch 4250, giga_loss[loss=0.2712, simple_loss=0.3341, pruned_loss=0.1042, over 28573.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3477, pruned_loss=0.1016, over 5714410.74 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3558, pruned_loss=0.0993, over 5108136.02 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3475, pruned_loss=0.1023, over 5705043.08 frames. ], batch size: 85, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:15:15,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:15:16,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-04 14:15:37,489 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:15:52,850 INFO [train.py:968] (0/2) Epoch 9, batch 4300, giga_loss[loss=0.3252, simple_loss=0.3729, pruned_loss=0.1388, over 23641.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3474, pruned_loss=0.1024, over 5709324.00 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.356, pruned_loss=0.09952, over 5127411.77 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3468, pruned_loss=0.1029, over 5700580.13 frames. ], batch size: 705, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:15:57,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.376e+02 1.015e+03 1.244e+03 1.654e+03 3.183e+03, threshold=2.489e+03, percent-clipped=2.0
+2023-03-04 14:16:04,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2456, 3.0714, 1.3433, 1.3084], device='cuda:0'), covar=tensor([0.0913, 0.0381, 0.0882, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0490, 0.0323, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 14:16:07,533 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:16:09,840 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368521.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:16:10,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:16:11,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5984, 3.5881, 1.6487, 1.6148], device='cuda:0'), covar=tensor([0.0802, 0.0345, 0.0815, 0.1162], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0490, 0.0323, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 14:16:22,915 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:16:30,077 INFO [train.py:968] (0/2) Epoch 9, batch 4350, libri_loss[loss=0.2685, simple_loss=0.3398, pruned_loss=0.09858, over 29553.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3441, pruned_loss=0.1006, over 5716349.34 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3555, pruned_loss=0.09922, over 5150171.33 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3438, pruned_loss=0.1013, over 5704450.50 frames. ], batch size: 75, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:16:32,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:16:40,557 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4181, 3.5482, 1.4411, 1.4922], device='cuda:0'), covar=tensor([0.0878, 0.0340, 0.0838, 0.1223], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0490, 0.0323, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 14:16:50,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3947, 2.0816, 1.4277, 1.3151], device='cuda:0'), covar=tensor([0.2391, 0.1394, 0.1687, 0.1975], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1479, 0.1458, 0.1552], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 14:17:05,498 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:17:06,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:17:07,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:17:08,559 INFO [train.py:968] (0/2) Epoch 9, batch 4400, giga_loss[loss=0.2922, simple_loss=0.3589, pruned_loss=0.1127, over 28828.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3424, pruned_loss=0.09993, over 5715564.40 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3557, pruned_loss=0.09922, over 5166968.79 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3417, pruned_loss=0.1004, over 5703420.12 frames. ], batch size: 243, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:17:09,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6195, 2.4255, 1.5075, 0.8512], device='cuda:0'), covar=tensor([0.5135, 0.2134, 0.2610, 0.3769], device='cuda:0'), in_proj_covar=tensor([0.1494, 0.1405, 0.1448, 0.1210], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 14:17:12,457 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.169e+02 1.051e+03 1.290e+03 1.697e+03 5.010e+03, threshold=2.579e+03, percent-clipped=11.0
+2023-03-04 14:17:28,662 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:17:28,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4254, 2.1219, 1.5039, 0.6326], device='cuda:0'), covar=tensor([0.4000, 0.1877, 0.3123, 0.4185], device='cuda:0'), in_proj_covar=tensor([0.1489, 0.1400, 0.1442, 0.1207], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 14:17:42,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5273, 1.7712, 1.4521, 1.3339], device='cuda:0'), covar=tensor([0.2116, 0.1463, 0.1279, 0.1694], device='cuda:0'), in_proj_covar=tensor([0.1611, 0.1474, 0.1451, 0.1549], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 14:17:46,339 INFO [train.py:968] (0/2) Epoch 9, batch 4450, giga_loss[loss=0.2476, simple_loss=0.3261, pruned_loss=0.08457, over 28954.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3422, pruned_loss=0.09983, over 5716678.77 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3557, pruned_loss=0.09929, over 5174293.83 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3415, pruned_loss=0.1002, over 5705472.91 frames. ], batch size: 145, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:17:55,488 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:17:59,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368664.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:18:01,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368667.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:18:03,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:18:14,416 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:18:27,168 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368696.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:18:29,777 INFO [train.py:968] (0/2) Epoch 9, batch 4500, giga_loss[loss=0.2639, simple_loss=0.3492, pruned_loss=0.08936, over 28691.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3449, pruned_loss=0.1008, over 5714670.58 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.356, pruned_loss=0.09929, over 5188906.73 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3439, pruned_loss=0.1011, over 5702402.73 frames. ], batch size: 242, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:18:33,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.507e+02 9.454e+02 1.162e+03 1.674e+03 5.203e+03, threshold=2.323e+03, percent-clipped=7.0
+2023-03-04 14:19:00,099 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:19:02,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:19:09,241 INFO [train.py:968] (0/2) Epoch 9, batch 4550, giga_loss[loss=0.3397, simple_loss=0.405, pruned_loss=0.1372, over 28860.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.348, pruned_loss=0.1018, over 5725711.55 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3562, pruned_loss=0.09951, over 5203476.47 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3468, pruned_loss=0.1019, over 5712546.82 frames. ], batch size: 227, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:19:18,223 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:19:24,545 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:19:31,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1924, 1.4873, 1.2096, 0.9329], device='cuda:0'), covar=tensor([0.1589, 0.1379, 0.0918, 0.1429], device='cuda:0'), in_proj_covar=tensor([0.1611, 0.1477, 0.1459, 0.1550], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 14:19:49,620 INFO [train.py:968] (0/2) Epoch 9, batch 4600, giga_loss[loss=0.2834, simple_loss=0.3488, pruned_loss=0.109, over 28713.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3501, pruned_loss=0.1026, over 5717009.19 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3564, pruned_loss=0.09952, over 5216927.52 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3489, pruned_loss=0.1027, over 5705768.39 frames. ], batch size: 99, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:19:56,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.765e+02 9.838e+02 1.303e+03 1.658e+03 5.178e+03, threshold=2.606e+03, percent-clipped=16.0
+2023-03-04 14:20:34,820 INFO [train.py:968] (0/2) Epoch 9, batch 4650, giga_loss[loss=0.272, simple_loss=0.3523, pruned_loss=0.09585, over 28226.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3498, pruned_loss=0.1017, over 5705813.64 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3564, pruned_loss=0.09954, over 5226370.81 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3487, pruned_loss=0.1018, over 5695121.22 frames. ], batch size: 368, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:20:39,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:20:51,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2974, 1.9284, 1.4789, 0.5677], device='cuda:0'), covar=tensor([0.3172, 0.1801, 0.2650, 0.3990], device='cuda:0'), in_proj_covar=tensor([0.1487, 0.1399, 0.1442, 0.1208], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 14:21:14,271 INFO [train.py:968] (0/2) Epoch 9, batch 4700, giga_loss[loss=0.3749, simple_loss=0.404, pruned_loss=0.1729, over 28585.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3493, pruned_loss=0.1012, over 5703500.34 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3566, pruned_loss=0.09967, over 5232828.78 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3483, pruned_loss=0.1012, over 5693738.91 frames. ], batch size: 92, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:21:18,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.134e+02 1.016e+03 1.237e+03 1.580e+03 2.877e+03, threshold=2.474e+03, percent-clipped=2.0
+2023-03-04 14:21:23,718 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:21:25,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:21:27,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1928, 1.7558, 1.4046, 0.3868], device='cuda:0'), covar=tensor([0.2878, 0.1693, 0.2732, 0.3866], device='cuda:0'), in_proj_covar=tensor([0.1491, 0.1404, 0.1447, 0.1211], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 14:21:29,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:21:53,522 INFO [train.py:968] (0/2) Epoch 9, batch 4750, giga_loss[loss=0.3117, simple_loss=0.3731, pruned_loss=0.1251, over 28873.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3515, pruned_loss=0.1026, over 5711518.53 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3574, pruned_loss=0.1003, over 5249117.30 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3499, pruned_loss=0.1022, over 5699584.94 frames. ], batch size: 145, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:22:08,746 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-04 14:22:12,166 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:22:33,399 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:22:33,980 INFO [train.py:968] (0/2) Epoch 9, batch 4800, giga_loss[loss=0.3172, simple_loss=0.3802, pruned_loss=0.1271, over 27637.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3523, pruned_loss=0.1031, over 5715834.18 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3574, pruned_loss=0.1003, over 5264974.81 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3508, pruned_loss=0.1028, over 5702569.55 frames. ], batch size: 472, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:22:35,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:22:38,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.901e+02 1.246e+03 1.578e+03 2.381e+03 5.112e+03, threshold=3.157e+03, percent-clipped=23.0
+2023-03-04 14:22:54,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 14:22:57,798 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:03,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:11,293 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:12,854 INFO [train.py:968] (0/2) Epoch 9, batch 4850, giga_loss[loss=0.3331, simple_loss=0.3954, pruned_loss=0.1354, over 27960.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3555, pruned_loss=0.1058, over 5717417.49 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3577, pruned_loss=0.1005, over 5280865.52 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.354, pruned_loss=0.1054, over 5702593.72 frames. ], batch size: 412, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:23:18,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:19,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:21,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:44,335 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:52,818 INFO [train.py:968] (0/2) Epoch 9, batch 4900, giga_loss[loss=0.279, simple_loss=0.3459, pruned_loss=0.1061, over 28691.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3577, pruned_loss=0.1073, over 5715857.64 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3578, pruned_loss=0.1006, over 5281862.20 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3564, pruned_loss=0.107, over 5705196.17 frames. ], batch size: 92, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:23:58,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.440e+02 1.230e+03 1.583e+03 2.260e+03 5.620e+03, threshold=3.167e+03, percent-clipped=14.0
+2023-03-04 14:24:20,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:24:32,083 INFO [train.py:968] (0/2) Epoch 9, batch 4950, giga_loss[loss=0.2838, simple_loss=0.3592, pruned_loss=0.1042, over 28970.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3588, pruned_loss=0.1074, over 5711932.26 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3578, pruned_loss=0.1005, over 5287933.65 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3578, pruned_loss=0.1074, over 5708406.41 frames. ], batch size: 213, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:24:55,591 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:24:59,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:04,490 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:06,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:12,117 INFO [train.py:968] (0/2) Epoch 9, batch 5000, giga_loss[loss=0.2856, simple_loss=0.3507, pruned_loss=0.1102, over 28672.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3603, pruned_loss=0.1082, over 5713528.54 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.359, pruned_loss=0.101, over 5302123.54 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3586, pruned_loss=0.1078, over 5707203.97 frames. ], batch size: 92, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:25:14,592 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:16,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:16,757 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.685e+02 1.118e+03 1.364e+03 1.747e+03 5.611e+03, threshold=2.727e+03, percent-clipped=2.0
+2023-03-04 14:25:19,045 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:20,432 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:29,501 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:40,015 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:49,647 INFO [train.py:968] (0/2) Epoch 9, batch 5050, giga_loss[loss=0.3707, simple_loss=0.4166, pruned_loss=0.1624, over 27722.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3611, pruned_loss=0.1087, over 5713935.24 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.359, pruned_loss=0.101, over 5324419.48 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3597, pruned_loss=0.1088, over 5702800.70 frames. ], batch size: 472, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:26:01,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:11,821 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:15,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:19,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:24,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:28,833 INFO [train.py:968] (0/2) Epoch 9, batch 5100, libri_loss[loss=0.3138, simple_loss=0.3907, pruned_loss=0.1185, over 29174.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3602, pruned_loss=0.1077, over 5712109.25 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3588, pruned_loss=0.1008, over 5339289.16 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3592, pruned_loss=0.1082, over 5704901.47 frames. ], batch size: 97, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:26:34,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.670e+02 1.162e+03 1.438e+03 2.076e+03 4.730e+03, threshold=2.875e+03, percent-clipped=10.0
+2023-03-04 14:26:35,834 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:37,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:27:05,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:27:06,145 INFO [train.py:968] (0/2) Epoch 9, batch 5150, giga_loss[loss=0.2891, simple_loss=0.3578, pruned_loss=0.1102, over 28033.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3588, pruned_loss=0.1071, over 5712458.01 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3589, pruned_loss=0.1008, over 5346103.82 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3581, pruned_loss=0.1076, over 5705650.20 frames. ], batch size: 412, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:27:14,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1364, 1.5009, 1.4544, 1.0791], device='cuda:0'), covar=tensor([0.1311, 0.1890, 0.1137, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0704, 0.0834, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 14:27:45,642 INFO [train.py:968] (0/2) Epoch 9, batch 5200, giga_loss[loss=0.2831, simple_loss=0.3522, pruned_loss=0.107, over 28906.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3557, pruned_loss=0.1056, over 5710631.47 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3591, pruned_loss=0.1009, over 5358149.43 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3548, pruned_loss=0.106, over 5703776.84 frames. ], batch size: 227, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:27:52,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.223e+02 1.023e+03 1.328e+03 1.906e+03 4.699e+03, threshold=2.656e+03, percent-clipped=7.0
+2023-03-04 14:28:11,116 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:13,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:17,534 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:19,824 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:21,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7308, 1.7690, 1.2595, 1.4961], device='cuda:0'), covar=tensor([0.0656, 0.0543, 0.0975, 0.0957], device='cuda:0'), in_proj_covar=tensor([0.0342, 0.0442, 0.0493, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 14:28:26,278 INFO [train.py:968] (0/2) Epoch 9, batch 5250, giga_loss[loss=0.3042, simple_loss=0.3641, pruned_loss=0.1222, over 28777.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.353, pruned_loss=0.1044, over 5711712.67 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3593, pruned_loss=0.1011, over 5361996.00 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3522, pruned_loss=0.1046, over 5705916.82 frames. ], batch size: 99, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:28:36,186 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:42,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:59,148 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:29:02,133 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:29:06,119 INFO [train.py:968] (0/2) Epoch 9, batch 5300, giga_loss[loss=0.2563, simple_loss=0.3394, pruned_loss=0.08661, over 28846.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3552, pruned_loss=0.1046, over 5695606.46 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3593, pruned_loss=0.1013, over 5356186.24 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3544, pruned_loss=0.1047, over 5704636.80 frames. ], batch size: 112, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:29:12,579 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.407e+02 1.115e+03 1.345e+03 1.899e+03 4.843e+03, threshold=2.690e+03, percent-clipped=9.0
+2023-03-04 14:29:25,703 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:29:35,612 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5202, 1.6263, 1.8176, 1.3601], device='cuda:0'), covar=tensor([0.1715, 0.1969, 0.1342, 0.1523], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0698, 0.0830, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 14:29:46,842 INFO [train.py:968] (0/2) Epoch 9, batch 5350, giga_loss[loss=0.2871, simple_loss=0.3637, pruned_loss=0.1052, over 28667.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3564, pruned_loss=0.1039, over 5703454.15 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3593, pruned_loss=0.1012, over 5364333.71 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3556, pruned_loss=0.1041, over 5707802.44 frames. ], batch size: 284, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:29:53,739 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:30:07,533 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 14:30:13,559 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:30:22,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7023, 1.8188, 1.4931, 2.2136], device='cuda:0'), covar=tensor([0.2116, 0.2166, 0.2411, 0.2020], device='cuda:0'), in_proj_covar=tensor([0.1223, 0.0911, 0.1079, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 14:30:24,135 INFO [train.py:968] (0/2) Epoch 9, batch 5400, giga_loss[loss=0.2727, simple_loss=0.3468, pruned_loss=0.09928, over 28902.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3557, pruned_loss=0.1045, over 5707980.22 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3596, pruned_loss=0.1014, over 5371085.88 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3549, pruned_loss=0.1045, over 5711273.01 frames. ], batch size: 145, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:30:29,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.214e+02 1.080e+03 1.319e+03 1.783e+03 4.300e+03, threshold=2.638e+03, percent-clipped=7.0
+2023-03-04 14:30:54,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:31:02,283 INFO [train.py:968] (0/2) Epoch 9, batch 5450, giga_loss[loss=0.2599, simple_loss=0.3193, pruned_loss=0.1002, over 28596.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3544, pruned_loss=0.1052, over 5716177.24 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3596, pruned_loss=0.1014, over 5381018.39 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3537, pruned_loss=0.1053, over 5715925.99 frames. ], batch size: 85, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:31:27,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:31:33,469 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369689.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:31:40,009 INFO [train.py:968] (0/2) Epoch 9, batch 5500, giga_loss[loss=0.2913, simple_loss=0.3595, pruned_loss=0.1115, over 27987.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3532, pruned_loss=0.1055, over 5727319.92 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3599, pruned_loss=0.1017, over 5407567.31 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3521, pruned_loss=0.1056, over 5720983.22 frames. ], batch size: 412, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:31:46,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.209e+02 1.180e+03 1.483e+03 1.893e+03 4.008e+03, threshold=2.965e+03, percent-clipped=9.0
+2023-03-04 14:31:48,070 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:01,063 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:05,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:14,611 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:18,737 INFO [train.py:968] (0/2) Epoch 9, batch 5550, libri_loss[loss=0.3119, simple_loss=0.3786, pruned_loss=0.1226, over 29533.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3507, pruned_loss=0.1055, over 5732474.81 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3598, pruned_loss=0.1018, over 5420100.98 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3497, pruned_loss=0.1055, over 5723271.80 frames. ], batch size: 82, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:32:26,564 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:37,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369773.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:32:43,911 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:46,035 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:58,944 INFO [train.py:968] (0/2) Epoch 9, batch 5600, giga_loss[loss=0.3409, simple_loss=0.3884, pruned_loss=0.1467, over 23986.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3507, pruned_loss=0.1059, over 5723828.12 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3595, pruned_loss=0.1016, over 5430028.21 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3498, pruned_loss=0.1062, over 5714473.32 frames. ], batch size: 705, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:33:05,687 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.082e+02 1.115e+03 1.366e+03 1.706e+03 5.073e+03, threshold=2.732e+03, percent-clipped=4.0
+2023-03-04 14:33:10,525 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:33:21,441 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:33:23,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:33:32,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3590, 3.0396, 1.3180, 1.5700], device='cuda:0'), covar=tensor([0.0932, 0.0400, 0.0918, 0.1225], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0496, 0.0325, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 14:33:39,096 INFO [train.py:968] (0/2) Epoch 9, batch 5650, giga_loss[loss=0.2462, simple_loss=0.3242, pruned_loss=0.08412, over 28775.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3468, pruned_loss=0.1041, over 5720372.45 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3595, pruned_loss=0.1016, over 5432426.07 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3461, pruned_loss=0.1043, over 5712298.58 frames. ], batch size: 284, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:33:40,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2361, 0.7794, 0.7555, 1.4419], device='cuda:0'), covar=tensor([0.0724, 0.0332, 0.0361, 0.0797], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:0')
+2023-03-04 14:33:46,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:34:03,566 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:34:03,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3597, 2.2394, 2.0950, 2.0510], device='cuda:0'), covar=tensor([0.1149, 0.2083, 0.1580, 0.1782], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0727, 0.0651, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 14:34:16,182 INFO [train.py:968] (0/2) Epoch 9, batch 5700, giga_loss[loss=0.2265, simple_loss=0.3067, pruned_loss=0.07318, over 28874.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3416, pruned_loss=0.1011, over 5718356.35 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3597, pruned_loss=0.1018, over 5442034.17 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3406, pruned_loss=0.1011, over 5708375.65 frames. ], batch size: 186, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:34:20,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8826, 1.6567, 1.4531, 1.3561], device='cuda:0'), covar=tensor([0.0666, 0.0652, 0.0904, 0.1059], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0447, 0.0499, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 14:34:23,388 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.870e+02 1.027e+03 1.325e+03 1.818e+03 4.105e+03, threshold=2.650e+03, percent-clipped=4.0
+2023-03-04 14:34:44,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:34:56,319 INFO [train.py:968] (0/2) Epoch 9, batch 5750, giga_loss[loss=0.2504, simple_loss=0.3179, pruned_loss=0.09149, over 29051.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3392, pruned_loss=0.09992, over 5704506.44 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3607, pruned_loss=0.1027, over 5437117.38 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.337, pruned_loss=0.09912, over 5708614.20 frames. ], batch size: 128, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:35:33,104 INFO [train.py:968] (0/2) Epoch 9, batch 5800, giga_loss[loss=0.2478, simple_loss=0.3281, pruned_loss=0.08375, over 28816.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3397, pruned_loss=0.09996, over 5702315.29 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3607, pruned_loss=0.1028, over 5433992.16 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3378, pruned_loss=0.09923, over 5711115.63 frames. ], batch size: 119, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:35:34,531 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-370000.pt
+2023-03-04 14:35:41,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.144e+02 1.083e+03 1.420e+03 1.917e+03 1.191e+04, threshold=2.840e+03, percent-clipped=10.0
+2023-03-04 14:35:54,405 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=370023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:35:58,606 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-04 14:36:01,831 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=370031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:36:14,611 INFO [train.py:968] (0/2) Epoch 9, batch 5850, giga_loss[loss=0.2915, simple_loss=0.3676, pruned_loss=0.1077, over 29028.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3424, pruned_loss=0.1007, over 5705822.53 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3607, pruned_loss=0.1028, over 5443382.26 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3404, pruned_loss=0.1001, over 5709138.75 frames. ], batch size: 136, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:36:28,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370064.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:36:37,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:36:39,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:36:43,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:36:55,190 INFO [train.py:968] (0/2) Epoch 9, batch 5900, giga_loss[loss=0.2805, simple_loss=0.3575, pruned_loss=0.1017, over 28837.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3459, pruned_loss=0.1019, over 5709785.70 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3611, pruned_loss=0.1031, over 5456850.07 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3435, pruned_loss=0.101, over 5707026.02 frames. ], batch size: 186, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:37:01,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:37:01,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.954e+02 1.101e+03 1.387e+03 1.747e+03 3.610e+03, threshold=2.775e+03, percent-clipped=5.0
+2023-03-04 14:37:10,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:37:35,927 INFO [train.py:968] (0/2) Epoch 9, batch 5950, giga_loss[loss=0.2949, simple_loss=0.3663, pruned_loss=0.1118, over 28705.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3489, pruned_loss=0.1029, over 5714889.07 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3612, pruned_loss=0.1032, over 5463702.48 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3468, pruned_loss=0.1022, over 5709798.09 frames. ], batch size: 242, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:37:36,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370148.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:37:37,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-04 14:37:43,592 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=370156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:37:47,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4400, 4.2272, 3.9572, 1.9787], device='cuda:0'), covar=tensor([0.0490, 0.0635, 0.0690, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.0984, 0.0908, 0.0808, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 14:38:03,064 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 14:38:03,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=370183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:38:16,864 INFO [train.py:968] (0/2) Epoch 9, batch 6000, giga_loss[loss=0.3013, simple_loss=0.3563, pruned_loss=0.1231, over 23890.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3517, pruned_loss=0.1042, over 5709282.57 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3614, pruned_loss=0.1035, over 5472896.69 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3496, pruned_loss=0.1034, over 5702431.35 frames. ], batch size: 705, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:38:16,868 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 14:38:25,197 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2271, simple_loss=0.3316, pruned_loss=0.06128, over 944034.00 frames. 
+2023-03-04 14:38:25,197 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 14:38:32,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370207.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:38:33,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.569e+02 1.020e+03 1.383e+03 1.918e+03 6.026e+03, threshold=2.766e+03, percent-clipped=7.0
+2023-03-04 14:38:35,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370210.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:38:48,492 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:38:49,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3193, 3.0991, 2.9259, 1.3492], device='cuda:0'), covar=tensor([0.0802, 0.0994, 0.0911, 0.2317], device='cuda:0'), in_proj_covar=tensor([0.0989, 0.0913, 0.0811, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 14:38:50,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:00,986 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370239.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:39:05,965 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=370246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:07,086 INFO [train.py:968] (0/2) Epoch 9, batch 6050, giga_loss[loss=0.3519, simple_loss=0.404, pruned_loss=0.1499, over 29054.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3562, pruned_loss=0.1079, over 5709675.63 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3617, pruned_loss=0.1038, over 5485721.72 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.354, pruned_loss=0.107, over 5698518.80 frames. ], batch size: 155, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:39:13,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:16,107 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:20,292 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:22,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:47,105 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370291.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:39:48,801 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:49,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370294.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:39:53,290 INFO [train.py:968] (0/2) Epoch 9, batch 6100, giga_loss[loss=0.3578, simple_loss=0.3872, pruned_loss=0.1642, over 23584.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3639, pruned_loss=0.115, over 5695303.23 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3619, pruned_loss=0.1041, over 5488630.08 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.362, pruned_loss=0.1143, over 5686772.98 frames. ], batch size: 705, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:40:03,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.811e+02 1.391e+03 1.845e+03 2.573e+03 1.000e+04, threshold=3.690e+03, percent-clipped=22.0
+2023-03-04 14:40:15,261 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370323.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:40:39,657 INFO [train.py:968] (0/2) Epoch 9, batch 6150, giga_loss[loss=0.3376, simple_loss=0.3986, pruned_loss=0.1383, over 29025.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3705, pruned_loss=0.1194, over 5702756.36 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3618, pruned_loss=0.104, over 5494334.57 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3692, pruned_loss=0.119, over 5693759.17 frames. ], batch size: 155, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:41:25,254 INFO [train.py:968] (0/2) Epoch 9, batch 6200, giga_loss[loss=0.2788, simple_loss=0.3557, pruned_loss=0.1009, over 28922.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3769, pruned_loss=0.1248, over 5700044.63 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3616, pruned_loss=0.1039, over 5500233.44 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3762, pruned_loss=0.1249, over 5690554.34 frames. ], batch size: 174, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:41:25,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:41:25,636 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:41:29,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:41:32,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:41:35,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.096e+02 1.639e+03 2.107e+03 2.859e+03 8.722e+03, threshold=4.215e+03, percent-clipped=6.0
+2023-03-04 14:41:52,888 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:42:06,874 INFO [train.py:968] (0/2) Epoch 9, batch 6250, giga_loss[loss=0.3154, simple_loss=0.3816, pruned_loss=0.1246, over 28972.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3832, pruned_loss=0.1303, over 5700718.81 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3615, pruned_loss=0.1038, over 5506995.80 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3831, pruned_loss=0.131, over 5692500.14 frames. ], batch size: 213, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:42:25,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3734, 1.9789, 1.4071, 0.5805], device='cuda:0'), covar=tensor([0.2710, 0.1429, 0.2092, 0.3103], device='cuda:0'), in_proj_covar=tensor([0.1497, 0.1422, 0.1459, 0.1224], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 14:42:52,080 INFO [train.py:968] (0/2) Epoch 9, batch 6300, giga_loss[loss=0.3677, simple_loss=0.4227, pruned_loss=0.1563, over 28664.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3885, pruned_loss=0.1349, over 5687640.07 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3614, pruned_loss=0.1037, over 5511582.44 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3891, pruned_loss=0.1361, over 5680204.37 frames. ], batch size: 242, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:43:03,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.601e+03 2.119e+03 3.651e+03 1.753e+04, threshold=4.238e+03, percent-clipped=17.0
+2023-03-04 14:43:12,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9800, 1.2852, 1.3499, 1.0509], device='cuda:0'), covar=tensor([0.1181, 0.0908, 0.1586, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0729, 0.0654, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 14:43:25,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:43:27,002 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-04 14:43:34,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:43:37,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:43:41,192 INFO [train.py:968] (0/2) Epoch 9, batch 6350, giga_loss[loss=0.3739, simple_loss=0.396, pruned_loss=0.1759, over 23476.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3903, pruned_loss=0.1376, over 5670145.16 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3612, pruned_loss=0.1038, over 5516661.57 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3915, pruned_loss=0.1391, over 5663598.29 frames. ], batch size: 705, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:43:43,388 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:43:45,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:43:50,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:44:04,609 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:44:12,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:44:27,954 INFO [train.py:968] (0/2) Epoch 9, batch 6400, libri_loss[loss=0.2498, simple_loss=0.3218, pruned_loss=0.08885, over 28207.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3924, pruned_loss=0.1401, over 5672203.51 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3609, pruned_loss=0.1038, over 5526959.18 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3948, pruned_loss=0.1427, over 5662144.22 frames. ], batch size: 62, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:44:29,051 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.31 vs. limit=5.0
+2023-03-04 14:44:38,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.728e+02 1.479e+03 1.901e+03 3.064e+03 2.310e+04, threshold=3.803e+03, percent-clipped=11.0
+2023-03-04 14:44:42,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2154, 1.4306, 1.2388, 1.0239], device='cuda:0'), covar=tensor([0.1278, 0.1270, 0.0862, 0.1208], device='cuda:0'), in_proj_covar=tensor([0.1623, 0.1506, 0.1489, 0.1579], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 14:44:50,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:45:18,321 INFO [train.py:968] (0/2) Epoch 9, batch 6450, giga_loss[loss=0.2946, simple_loss=0.3637, pruned_loss=0.1128, over 28841.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3962, pruned_loss=0.1443, over 5657707.44 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3611, pruned_loss=0.1039, over 5521963.18 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.3981, pruned_loss=0.1465, over 5655547.11 frames. ], batch size: 112, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:45:45,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:45:49,035 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:46:08,342 INFO [train.py:968] (0/2) Epoch 9, batch 6500, giga_loss[loss=0.4042, simple_loss=0.4389, pruned_loss=0.1847, over 28890.00 frames. ], tot_loss[loss=0.3474, simple_loss=0.3997, pruned_loss=0.1475, over 5651153.99 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.361, pruned_loss=0.1039, over 5531145.38 frames. ], giga_tot_loss[loss=0.3515, simple_loss=0.4023, pruned_loss=0.1503, over 5643768.17 frames. ], batch size: 227, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:46:11,334 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:46:16,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:46:17,595 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:46:19,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.618e+03 2.139e+03 2.729e+03 9.477e+03, threshold=4.279e+03, percent-clipped=11.0
+2023-03-04 14:46:42,357 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:46:56,344 INFO [train.py:968] (0/2) Epoch 9, batch 6550, giga_loss[loss=0.3812, simple_loss=0.4188, pruned_loss=0.1718, over 27972.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.3999, pruned_loss=0.1482, over 5650770.60 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.361, pruned_loss=0.1038, over 5538483.59 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.4027, pruned_loss=0.1514, over 5640328.67 frames. ], batch size: 412, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:47:00,101 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-04 14:47:01,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6481, 1.6312, 1.5302, 1.5469], device='cuda:0'), covar=tensor([0.1086, 0.1720, 0.1755, 0.1496], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0727, 0.0656, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 14:47:11,848 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:47:14,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:47:44,087 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:47:46,201 INFO [train.py:968] (0/2) Epoch 9, batch 6600, giga_loss[loss=0.441, simple_loss=0.4596, pruned_loss=0.2112, over 28561.00 frames. ], tot_loss[loss=0.3491, simple_loss=0.3998, pruned_loss=0.1492, over 5645496.60 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3609, pruned_loss=0.1036, over 5542531.65 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4028, pruned_loss=0.1526, over 5635546.75 frames. ], batch size: 307, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:47:56,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.752e+02 1.616e+03 2.151e+03 2.892e+03 8.762e+03, threshold=4.302e+03, percent-clipped=7.0
+2023-03-04 14:48:33,786 INFO [train.py:968] (0/2) Epoch 9, batch 6650, giga_loss[loss=0.3324, simple_loss=0.3951, pruned_loss=0.1349, over 28926.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3977, pruned_loss=0.147, over 5636611.88 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3611, pruned_loss=0.1038, over 5541459.05 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4006, pruned_loss=0.1505, over 5632162.64 frames. ], batch size: 227, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:49:17,938 INFO [train.py:968] (0/2) Epoch 9, batch 6700, giga_loss[loss=0.3742, simple_loss=0.4263, pruned_loss=0.161, over 28980.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3978, pruned_loss=0.1456, over 5642782.37 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3615, pruned_loss=0.104, over 5545089.20 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.4007, pruned_loss=0.1492, over 5638113.59 frames. ], batch size: 227, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:49:28,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.130e+02 1.616e+03 2.069e+03 3.008e+03 1.489e+04, threshold=4.138e+03, percent-clipped=9.0
+2023-03-04 14:50:00,846 INFO [train.py:968] (0/2) Epoch 9, batch 6750, giga_loss[loss=0.3404, simple_loss=0.4035, pruned_loss=0.1386, over 28824.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.3989, pruned_loss=0.1459, over 5646304.80 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3611, pruned_loss=0.1038, over 5557524.42 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.403, pruned_loss=0.1506, over 5634533.66 frames. ], batch size: 199, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:50:49,122 INFO [train.py:968] (0/2) Epoch 9, batch 6800, giga_loss[loss=0.3093, simple_loss=0.3743, pruned_loss=0.1221, over 28988.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3969, pruned_loss=0.1447, over 5633800.02 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3609, pruned_loss=0.1035, over 5564418.75 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.4013, pruned_loss=0.1497, over 5619863.13 frames. ], batch size: 136, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 14:51:00,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.865e+02 1.470e+03 1.868e+03 2.742e+03 6.746e+03, threshold=3.736e+03, percent-clipped=8.0
+2023-03-04 14:51:38,755 INFO [train.py:968] (0/2) Epoch 9, batch 6850, libri_loss[loss=0.2842, simple_loss=0.36, pruned_loss=0.1042, over 29533.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3956, pruned_loss=0.1426, over 5639584.40 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3608, pruned_loss=0.1035, over 5566180.31 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.3994, pruned_loss=0.1469, over 5627592.76 frames. ], batch size: 80, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:51:40,694 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-04 14:52:09,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3561, 1.6180, 1.2946, 1.5898], device='cuda:0'), covar=tensor([0.2216, 0.2101, 0.2365, 0.1953], device='cuda:0'), in_proj_covar=tensor([0.1214, 0.0899, 0.1072, 0.0943], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 14:52:11,941 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-04 14:52:22,794 INFO [train.py:968] (0/2) Epoch 9, batch 6900, giga_loss[loss=0.3312, simple_loss=0.389, pruned_loss=0.1367, over 28775.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3925, pruned_loss=0.1389, over 5647323.81 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3609, pruned_loss=0.1035, over 5567534.73 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3959, pruned_loss=0.1428, over 5637309.07 frames. ], batch size: 243, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:52:34,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.857e+02 1.725e+03 2.527e+03 4.429e+03 1.287e+04, threshold=5.053e+03, percent-clipped=30.0
+2023-03-04 14:53:07,685 INFO [train.py:968] (0/2) Epoch 9, batch 6950, giga_loss[loss=0.3056, simple_loss=0.3704, pruned_loss=0.1204, over 28834.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3889, pruned_loss=0.1359, over 5654697.08 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3608, pruned_loss=0.1036, over 5572106.65 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.392, pruned_loss=0.1394, over 5643567.84 frames. ], batch size: 119, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:53:53,018 INFO [train.py:968] (0/2) Epoch 9, batch 7000, giga_loss[loss=0.2396, simple_loss=0.3209, pruned_loss=0.07912, over 28590.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3845, pruned_loss=0.1326, over 5656456.24 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3602, pruned_loss=0.1033, over 5583797.49 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3885, pruned_loss=0.1367, over 5639312.64 frames. ], batch size: 60, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:54:04,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.383e+02 1.650e+03 2.143e+03 2.741e+03 6.081e+03, threshold=4.286e+03, percent-clipped=5.0
+2023-03-04 14:54:26,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6692, 1.0562, 2.8628, 2.6828], device='cuda:0'), covar=tensor([0.1707, 0.2387, 0.0520, 0.0965], device='cuda:0'), in_proj_covar=tensor([0.0627, 0.0572, 0.0819, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 14:54:26,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2952, 1.2518, 1.1233, 0.9808], device='cuda:0'), covar=tensor([0.0660, 0.0473, 0.0961, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0443, 0.0493, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 14:54:34,970 INFO [train.py:968] (0/2) Epoch 9, batch 7050, giga_loss[loss=0.2937, simple_loss=0.3663, pruned_loss=0.1105, over 28880.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3834, pruned_loss=0.1314, over 5657933.81 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3597, pruned_loss=0.1029, over 5587239.15 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3875, pruned_loss=0.1357, over 5642792.30 frames. ], batch size: 174, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:55:25,782 INFO [train.py:968] (0/2) Epoch 9, batch 7100, giga_loss[loss=0.3446, simple_loss=0.4112, pruned_loss=0.139, over 28727.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3841, pruned_loss=0.132, over 5649765.28 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3599, pruned_loss=0.103, over 5583868.11 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3875, pruned_loss=0.1357, over 5641377.85 frames. ], batch size: 99, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:55:39,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.914e+02 1.453e+03 1.766e+03 2.419e+03 6.859e+03, threshold=3.532e+03, percent-clipped=1.0
+2023-03-04 14:55:44,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5275, 1.6550, 1.2572, 1.3348], device='cuda:0'), covar=tensor([0.0676, 0.0447, 0.0908, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0441, 0.0492, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 14:56:14,344 INFO [train.py:968] (0/2) Epoch 9, batch 7150, giga_loss[loss=0.3058, simple_loss=0.3711, pruned_loss=0.1202, over 28802.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3822, pruned_loss=0.1301, over 5657514.71 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3601, pruned_loss=0.1032, over 5591594.50 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3853, pruned_loss=0.1336, over 5645470.99 frames. ], batch size: 284, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:57:04,737 INFO [train.py:968] (0/2) Epoch 9, batch 7200, giga_loss[loss=0.342, simple_loss=0.4064, pruned_loss=0.1388, over 28858.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3824, pruned_loss=0.1274, over 5663462.17 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3604, pruned_loss=0.1034, over 5599508.08 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3854, pruned_loss=0.1308, over 5648382.80 frames. ], batch size: 119, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 14:57:16,929 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-04 14:57:20,469 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.769e+02 1.496e+03 1.889e+03 2.475e+03 4.824e+03, threshold=3.777e+03, percent-clipped=4.0
+2023-03-04 14:57:35,532 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 14:57:50,683 INFO [train.py:968] (0/2) Epoch 9, batch 7250, giga_loss[loss=0.3423, simple_loss=0.3996, pruned_loss=0.1425, over 27945.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3852, pruned_loss=0.1282, over 5676180.87 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.36, pruned_loss=0.1032, over 5606077.40 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3885, pruned_loss=0.1316, over 5659811.08 frames. ], batch size: 412, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 14:58:37,047 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 14:58:43,104 INFO [train.py:968] (0/2) Epoch 9, batch 7300, giga_loss[loss=0.2933, simple_loss=0.363, pruned_loss=0.1118, over 28844.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3861, pruned_loss=0.1296, over 5668266.23 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3601, pruned_loss=0.1033, over 5607505.02 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3887, pruned_loss=0.1324, over 5654442.36 frames. ], batch size: 119, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 14:58:55,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.718e+02 1.725e+03 2.383e+03 3.829e+03 8.314e+03, threshold=4.766e+03, percent-clipped=25.0
+2023-03-04 14:59:28,684 INFO [train.py:968] (0/2) Epoch 9, batch 7350, giga_loss[loss=0.3007, simple_loss=0.3635, pruned_loss=0.1189, over 28445.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3867, pruned_loss=0.1312, over 5670548.52 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3598, pruned_loss=0.1032, over 5610278.06 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3894, pruned_loss=0.1338, over 5658044.48 frames. ], batch size: 71, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 14:59:59,919 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-04 15:00:09,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 15:00:18,617 INFO [train.py:968] (0/2) Epoch 9, batch 7400, giga_loss[loss=0.3049, simple_loss=0.367, pruned_loss=0.1214, over 28700.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3841, pruned_loss=0.1306, over 5661336.12 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3597, pruned_loss=0.1032, over 5603979.03 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3866, pruned_loss=0.133, over 5658169.36 frames. ], batch size: 307, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:00:29,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.646e+03 2.211e+03 3.053e+03 5.592e+03, threshold=4.421e+03, percent-clipped=6.0
+2023-03-04 15:01:03,742 INFO [train.py:968] (0/2) Epoch 9, batch 7450, giga_loss[loss=0.2983, simple_loss=0.3701, pruned_loss=0.1133, over 28918.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3829, pruned_loss=0.1307, over 5667662.74 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3597, pruned_loss=0.1032, over 5603979.03 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3848, pruned_loss=0.1325, over 5665198.01 frames. ], batch size: 174, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:01:33,106 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=371679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:01:47,240 INFO [train.py:968] (0/2) Epoch 9, batch 7500, giga_loss[loss=0.3026, simple_loss=0.3775, pruned_loss=0.1139, over 28992.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3822, pruned_loss=0.1289, over 5666768.41 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.36, pruned_loss=0.1033, over 5611885.78 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3844, pruned_loss=0.1314, over 5659928.74 frames. ], batch size: 213, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:01:59,104 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.191e+02 1.332e+03 1.667e+03 2.388e+03 5.280e+03, threshold=3.334e+03, percent-clipped=1.0
+2023-03-04 15:02:30,463 INFO [train.py:968] (0/2) Epoch 9, batch 7550, giga_loss[loss=0.3566, simple_loss=0.4102, pruned_loss=0.1515, over 28828.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.381, pruned_loss=0.1271, over 5657000.83 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.36, pruned_loss=0.1033, over 5606614.91 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3834, pruned_loss=0.1298, over 5657481.61 frames. ], batch size: 284, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:02:35,799 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=371755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:03:03,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4381, 2.3815, 1.7837, 1.9888], device='cuda:0'), covar=tensor([0.0663, 0.0584, 0.0865, 0.0959], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0440, 0.0492, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:03:12,391 INFO [train.py:968] (0/2) Epoch 9, batch 7600, giga_loss[loss=0.294, simple_loss=0.3696, pruned_loss=0.1092, over 29002.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3811, pruned_loss=0.1269, over 5672302.57 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3597, pruned_loss=0.1031, over 5618131.28 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3841, pruned_loss=0.1301, over 5664225.11 frames. ], batch size: 155, lr: 3.67e-03, grad_scale: 8.0
+2023-03-04 15:03:22,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.916e+02 1.352e+03 1.706e+03 2.555e+03 5.697e+03, threshold=3.413e+03, percent-clipped=15.0
+2023-03-04 15:03:32,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6462, 1.7111, 1.4616, 1.7925], device='cuda:0'), covar=tensor([0.2212, 0.2227, 0.2462, 0.2272], device='cuda:0'), in_proj_covar=tensor([0.1236, 0.0917, 0.1092, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 15:03:53,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1208, 3.9414, 3.7279, 1.8517], device='cuda:0'), covar=tensor([0.0549, 0.0677, 0.0737, 0.1981], device='cuda:0'), in_proj_covar=tensor([0.1000, 0.0933, 0.0819, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 15:03:55,763 INFO [train.py:968] (0/2) Epoch 9, batch 7650, giga_loss[loss=0.3287, simple_loss=0.3898, pruned_loss=0.1339, over 28916.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3791, pruned_loss=0.1253, over 5687768.12 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3598, pruned_loss=0.1032, over 5620261.39 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3817, pruned_loss=0.1282, over 5680714.18 frames. ], batch size: 174, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:04:30,942 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 15:04:44,291 INFO [train.py:968] (0/2) Epoch 9, batch 7700, giga_loss[loss=0.3188, simple_loss=0.3637, pruned_loss=0.1369, over 23388.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3772, pruned_loss=0.1254, over 5665081.41 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3597, pruned_loss=0.1031, over 5626140.08 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3797, pruned_loss=0.1282, over 5655122.87 frames. ], batch size: 705, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:04:56,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.165e+02 1.521e+03 2.061e+03 2.494e+03 8.321e+03, threshold=4.122e+03, percent-clipped=8.0
+2023-03-04 15:05:30,209 INFO [train.py:968] (0/2) Epoch 9, batch 7750, giga_loss[loss=0.3172, simple_loss=0.3778, pruned_loss=0.1283, over 28950.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3769, pruned_loss=0.126, over 5673946.58 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.36, pruned_loss=0.1033, over 5628748.18 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3788, pruned_loss=0.1282, over 5664217.75 frames. ], batch size: 136, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:05:37,376 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 15:05:50,719 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-04 15:06:13,432 INFO [train.py:968] (0/2) Epoch 9, batch 7800, giga_loss[loss=0.3013, simple_loss=0.3664, pruned_loss=0.1181, over 28656.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.377, pruned_loss=0.1268, over 5669053.32 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3601, pruned_loss=0.1033, over 5637971.41 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.379, pruned_loss=0.1293, over 5654078.92 frames. ], batch size: 262, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:06:16,341 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-372000.pt
+2023-03-04 15:06:26,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.702e+02 1.614e+03 2.314e+03 3.368e+03 1.602e+04, threshold=4.627e+03, percent-clipped=16.0
+2023-03-04 15:06:57,094 INFO [train.py:968] (0/2) Epoch 9, batch 7850, giga_loss[loss=0.2741, simple_loss=0.3523, pruned_loss=0.09796, over 28958.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3757, pruned_loss=0.1264, over 5657965.47 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3605, pruned_loss=0.1036, over 5633213.73 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3773, pruned_loss=0.1288, over 5650898.09 frames. ], batch size: 145, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:07:00,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2547, 1.4690, 1.3187, 1.1232], device='cuda:0'), covar=tensor([0.1573, 0.1290, 0.0955, 0.1282], device='cuda:0'), in_proj_covar=tensor([0.1624, 0.1496, 0.1465, 0.1565], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 15:07:02,567 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=372054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:07:09,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7626, 3.5524, 3.3779, 1.6931], device='cuda:0'), covar=tensor([0.0693, 0.0821, 0.0789, 0.2232], device='cuda:0'), in_proj_covar=tensor([0.1006, 0.0943, 0.0828, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 15:07:29,009 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-04 15:07:31,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9996, 1.9209, 1.8023, 1.7162], device='cuda:0'), covar=tensor([0.1054, 0.1640, 0.1622, 0.1533], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0735, 0.0657, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 15:07:40,373 INFO [train.py:968] (0/2) Epoch 9, batch 7900, giga_loss[loss=0.3186, simple_loss=0.3797, pruned_loss=0.1287, over 28546.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3759, pruned_loss=0.1271, over 5656505.44 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3604, pruned_loss=0.1035, over 5636009.19 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3777, pruned_loss=0.1296, over 5648742.44 frames. ], batch size: 336, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:07:40,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4541, 1.6574, 1.2786, 1.8821], device='cuda:0'), covar=tensor([0.2366, 0.2380, 0.2549, 0.2374], device='cuda:0'), in_proj_covar=tensor([0.1236, 0.0920, 0.1089, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 15:07:51,721 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.609e+02 1.653e+03 2.356e+03 3.470e+03 8.684e+03, threshold=4.712e+03, percent-clipped=8.0
+2023-03-04 15:08:07,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=372130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:08:10,982 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-04 15:08:19,335 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 15:08:25,249 INFO [train.py:968] (0/2) Epoch 9, batch 7950, giga_loss[loss=0.3799, simple_loss=0.4311, pruned_loss=0.1643, over 28260.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3762, pruned_loss=0.127, over 5642521.51 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3604, pruned_loss=0.1037, over 5623198.71 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.378, pruned_loss=0.1294, over 5648917.86 frames. ], batch size: 368, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:08:40,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7156, 1.7733, 1.7516, 1.5286], device='cuda:0'), covar=tensor([0.1213, 0.1796, 0.1721, 0.1752], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0732, 0.0655, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 15:09:10,354 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=372197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:09:10,668 INFO [train.py:968] (0/2) Epoch 9, batch 8000, giga_loss[loss=0.3438, simple_loss=0.3989, pruned_loss=0.1443, over 28609.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3759, pruned_loss=0.1255, over 5654671.75 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3604, pruned_loss=0.1038, over 5625810.22 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3775, pruned_loss=0.1275, over 5657870.18 frames. ], batch size: 307, lr: 3.67e-03, grad_scale: 8.0
+2023-03-04 15:09:13,591 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=372200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:09:16,329 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-04 15:09:22,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.129e+02 1.433e+03 2.184e+03 3.317e+03 1.013e+04, threshold=4.368e+03, percent-clipped=7.0
+2023-03-04 15:09:38,164 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=372229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:09:53,622 INFO [train.py:968] (0/2) Epoch 9, batch 8050, giga_loss[loss=0.2749, simple_loss=0.3494, pruned_loss=0.1002, over 28855.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3753, pruned_loss=0.1237, over 5673404.54 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3601, pruned_loss=0.1036, over 5633893.28 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3773, pruned_loss=0.1262, over 5669755.16 frames. ], batch size: 186, lr: 3.67e-03, grad_scale: 8.0
+2023-03-04 15:10:17,273 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=372273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:10:20,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=372276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:10:38,666 INFO [train.py:968] (0/2) Epoch 9, batch 8100, libri_loss[loss=0.2567, simple_loss=0.3395, pruned_loss=0.08698, over 29546.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3757, pruned_loss=0.1241, over 5674639.69 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.36, pruned_loss=0.1036, over 5637526.71 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3776, pruned_loss=0.1265, over 5669102.87 frames. ], batch size: 77, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:10:40,848 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=372301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:10:45,618 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=372305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:10:51,231 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.868e+02 1.525e+03 1.971e+03 2.560e+03 8.728e+03, threshold=3.943e+03, percent-clipped=4.0
+2023-03-04 15:11:25,332 INFO [train.py:968] (0/2) Epoch 9, batch 8150, giga_loss[loss=0.3477, simple_loss=0.3976, pruned_loss=0.1488, over 28906.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3789, pruned_loss=0.1276, over 5671485.60 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3598, pruned_loss=0.1034, over 5644718.26 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3812, pruned_loss=0.1304, over 5661414.25 frames. ], batch size: 106, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:11:31,203 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 15:12:17,821 INFO [train.py:968] (0/2) Epoch 9, batch 8200, giga_loss[loss=0.397, simple_loss=0.4098, pruned_loss=0.1921, over 23793.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3814, pruned_loss=0.1313, over 5657836.04 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3601, pruned_loss=0.1034, over 5647010.14 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3832, pruned_loss=0.1337, over 5647992.41 frames. ], batch size: 705, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:12:31,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.000e+02 1.700e+03 2.383e+03 3.313e+03 6.609e+03, threshold=4.767e+03, percent-clipped=16.0
+2023-03-04 15:12:34,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8965, 3.6719, 3.4614, 1.5735], device='cuda:0'), covar=tensor([0.0768, 0.1067, 0.1093, 0.2326], device='cuda:0'), in_proj_covar=tensor([0.1006, 0.0943, 0.0830, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 15:12:45,463 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-04 15:13:01,046 INFO [train.py:968] (0/2) Epoch 9, batch 8250, giga_loss[loss=0.3279, simple_loss=0.3854, pruned_loss=0.1351, over 28333.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3829, pruned_loss=0.1333, over 5666261.95 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3603, pruned_loss=0.1034, over 5651603.64 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3846, pruned_loss=0.1358, over 5654322.87 frames. ], batch size: 368, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:13:11,515 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=372458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:13:48,882 INFO [train.py:968] (0/2) Epoch 9, batch 8300, giga_loss[loss=0.3925, simple_loss=0.4252, pruned_loss=0.1799, over 24189.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3849, pruned_loss=0.1355, over 5664774.57 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.36, pruned_loss=0.1034, over 5658427.41 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3871, pruned_loss=0.1384, over 5649396.17 frames. ], batch size: 705, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:14:02,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.734e+03 2.163e+03 2.917e+03 7.915e+03, threshold=4.326e+03, percent-clipped=7.0
+2023-03-04 15:14:33,416 INFO [train.py:968] (0/2) Epoch 9, batch 8350, giga_loss[loss=0.3142, simple_loss=0.3599, pruned_loss=0.1343, over 28475.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3837, pruned_loss=0.1348, over 5674186.48 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3599, pruned_loss=0.1034, over 5663353.75 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3862, pruned_loss=0.1379, over 5657336.52 frames. ], batch size: 85, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:14:36,632 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-04 15:15:17,450 INFO [train.py:968] (0/2) Epoch 9, batch 8400, giga_loss[loss=0.3013, simple_loss=0.383, pruned_loss=0.1098, over 28941.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3823, pruned_loss=0.1326, over 5674255.16 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3599, pruned_loss=0.1034, over 5655960.96 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3845, pruned_loss=0.1353, over 5668533.53 frames. ], batch size: 136, lr: 3.67e-03, grad_scale: 8.0
+2023-03-04 15:15:29,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.333e+02 1.551e+03 1.999e+03 2.781e+03 6.199e+03, threshold=3.997e+03, percent-clipped=6.0
+2023-03-04 15:16:00,699 INFO [train.py:968] (0/2) Epoch 9, batch 8450, giga_loss[loss=0.3053, simple_loss=0.3848, pruned_loss=0.1129, over 28647.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3798, pruned_loss=0.1288, over 5683829.42 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3596, pruned_loss=0.1033, over 5662074.34 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3825, pruned_loss=0.1319, over 5674242.53 frames. ], batch size: 71, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:16:06,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1346, 1.1297, 4.0214, 3.2045], device='cuda:0'), covar=tensor([0.1698, 0.2542, 0.0393, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0625, 0.0571, 0.0826, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:16:08,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3608, 1.2133, 4.7828, 3.4350], device='cuda:0'), covar=tensor([0.1706, 0.2620, 0.0337, 0.0741], device='cuda:0'), in_proj_covar=tensor([0.0625, 0.0570, 0.0825, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:16:23,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=372676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:16:42,878 INFO [train.py:968] (0/2) Epoch 9, batch 8500, giga_loss[loss=0.3279, simple_loss=0.3839, pruned_loss=0.1359, over 28933.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3784, pruned_loss=0.1283, over 5680367.25 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3597, pruned_loss=0.1034, over 5664070.09 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3806, pruned_loss=0.131, over 5671207.66 frames. ], batch size: 227, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:16:58,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.659e+03 2.420e+03 3.217e+03 9.608e+03, threshold=4.840e+03, percent-clipped=19.0
+2023-03-04 15:16:59,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2732, 1.7956, 1.4122, 0.4451], device='cuda:0'), covar=tensor([0.2620, 0.1928, 0.2630, 0.3502], device='cuda:0'), in_proj_covar=tensor([0.1492, 0.1420, 0.1451, 0.1224], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 15:17:25,222 INFO [train.py:968] (0/2) Epoch 9, batch 8550, giga_loss[loss=0.3124, simple_loss=0.367, pruned_loss=0.1289, over 28900.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3767, pruned_loss=0.1278, over 5678477.27 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3601, pruned_loss=0.1035, over 5667648.94 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3785, pruned_loss=0.1303, over 5668074.26 frames. ], batch size: 136, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:17:46,226 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-04 15:18:09,220 INFO [train.py:968] (0/2) Epoch 9, batch 8600, giga_loss[loss=0.2992, simple_loss=0.369, pruned_loss=0.1147, over 28941.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3759, pruned_loss=0.1276, over 5667767.11 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3603, pruned_loss=0.1036, over 5664465.40 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3776, pruned_loss=0.1301, over 5662802.05 frames. ], batch size: 155, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:18:25,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.722e+02 1.608e+03 1.991e+03 2.755e+03 9.394e+03, threshold=3.982e+03, percent-clipped=6.0
+2023-03-04 15:18:28,422 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3594, 5.1841, 4.8902, 2.4592], device='cuda:0'), covar=tensor([0.0383, 0.0554, 0.0602, 0.1787], device='cuda:0'), in_proj_covar=tensor([0.1001, 0.0935, 0.0826, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 15:18:30,226 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=372819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:18:33,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=372822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:18:39,129 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-04 15:18:42,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=372833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:18:57,257 INFO [train.py:968] (0/2) Epoch 9, batch 8650, giga_loss[loss=0.353, simple_loss=0.413, pruned_loss=0.1465, over 28901.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3789, pruned_loss=0.1293, over 5675284.24 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3598, pruned_loss=0.1033, over 5669598.07 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3809, pruned_loss=0.132, over 5666794.63 frames. ], batch size: 112, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:18:59,527 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=372851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:19:41,973 INFO [train.py:968] (0/2) Epoch 9, batch 8700, giga_loss[loss=0.3875, simple_loss=0.4422, pruned_loss=0.1664, over 28528.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3829, pruned_loss=0.1297, over 5674799.49 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3596, pruned_loss=0.1033, over 5673872.90 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3851, pruned_loss=0.1324, over 5664230.45 frames. ], batch size: 336, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:19:54,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.242e+02 1.396e+03 1.762e+03 2.250e+03 5.847e+03, threshold=3.524e+03, percent-clipped=4.0
+2023-03-04 15:20:28,425 INFO [train.py:968] (0/2) Epoch 9, batch 8750, libri_loss[loss=0.2754, simple_loss=0.3461, pruned_loss=0.1023, over 29549.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3857, pruned_loss=0.1302, over 5676634.06 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3595, pruned_loss=0.1033, over 5674908.41 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3881, pruned_loss=0.1329, over 5667401.58 frames. ], batch size: 80, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:20:53,747 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=372976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:20:55,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=372979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:21:11,563 INFO [train.py:968] (0/2) Epoch 9, batch 8800, giga_loss[loss=0.3348, simple_loss=0.3952, pruned_loss=0.1371, over 28956.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3876, pruned_loss=0.1319, over 5680475.41 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3594, pruned_loss=0.1033, over 5677070.19 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3897, pruned_loss=0.1342, over 5671249.01 frames. ], batch size: 227, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:21:18,488 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=373008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:21:23,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.495e+02 1.611e+03 1.824e+03 2.235e+03 6.138e+03, threshold=3.648e+03, percent-clipped=8.0
+2023-03-04 15:21:54,132 INFO [train.py:968] (0/2) Epoch 9, batch 8850, giga_loss[loss=0.3213, simple_loss=0.3878, pruned_loss=0.1274, over 28812.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3874, pruned_loss=0.1321, over 5691180.88 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3595, pruned_loss=0.1035, over 5683232.83 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3899, pruned_loss=0.1346, over 5678435.40 frames. ], batch size: 119, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:22:06,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8137, 1.7272, 1.6874, 1.6062], device='cuda:0'), covar=tensor([0.1311, 0.2110, 0.1859, 0.1762], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0731, 0.0656, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 15:22:26,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=373087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:22:33,715 INFO [train.py:968] (0/2) Epoch 9, batch 8900, libri_loss[loss=0.2567, simple_loss=0.3289, pruned_loss=0.0922, over 29677.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3877, pruned_loss=0.1334, over 5694297.04 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3595, pruned_loss=0.1035, over 5690282.11 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3907, pruned_loss=0.1364, over 5678020.91 frames. ], batch size: 69, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:22:49,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.365e+02 1.694e+03 2.197e+03 2.850e+03 5.724e+03, threshold=4.394e+03, percent-clipped=14.0
+2023-03-04 15:23:19,560 INFO [train.py:968] (0/2) Epoch 9, batch 8950, giga_loss[loss=0.303, simple_loss=0.3621, pruned_loss=0.1219, over 28876.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3869, pruned_loss=0.1334, over 5699515.68 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.36, pruned_loss=0.1037, over 5694976.78 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3894, pruned_loss=0.1363, over 5682313.67 frames. ], batch size: 199, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:23:50,220 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-04 15:23:57,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9165, 1.0869, 1.0052, 0.7772], device='cuda:0'), covar=tensor([0.1195, 0.1338, 0.0854, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.1626, 0.1504, 0.1460, 0.1561], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 15:24:02,675 INFO [train.py:968] (0/2) Epoch 9, batch 9000, giga_loss[loss=0.4027, simple_loss=0.4337, pruned_loss=0.1858, over 26595.00 frames. ], tot_loss[loss=0.324, simple_loss=0.384, pruned_loss=0.132, over 5695890.91 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3601, pruned_loss=0.1038, over 5700286.39 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3867, pruned_loss=0.1351, over 5677335.02 frames. ], batch size: 555, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:24:02,679 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 15:24:10,971 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.223, simple_loss=0.3297, pruned_loss=0.05818, over 944034.00 frames. 
+2023-03-04 15:24:10,972 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 15:24:24,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.545e+02 1.617e+03 2.055e+03 2.657e+03 5.552e+03, threshold=4.110e+03, percent-clipped=5.0
+2023-03-04 15:24:37,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3226, 1.5719, 1.2760, 1.4498], device='cuda:0'), covar=tensor([0.2511, 0.2389, 0.2620, 0.2106], device='cuda:0'), in_proj_covar=tensor([0.1245, 0.0930, 0.1096, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 15:24:54,064 INFO [train.py:968] (0/2) Epoch 9, batch 9050, giga_loss[loss=0.3184, simple_loss=0.3737, pruned_loss=0.1315, over 28702.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3829, pruned_loss=0.1319, over 5691828.77 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3601, pruned_loss=0.1037, over 5703668.00 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3856, pruned_loss=0.1351, over 5673849.52 frames. ], batch size: 71, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:25:37,579 INFO [train.py:968] (0/2) Epoch 9, batch 9100, giga_loss[loss=0.3416, simple_loss=0.3952, pruned_loss=0.144, over 27959.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3823, pruned_loss=0.1317, over 5687196.34 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3603, pruned_loss=0.104, over 5700273.37 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.385, pruned_loss=0.1348, over 5675869.72 frames. ], batch size: 412, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:25:53,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.670e+02 1.656e+03 2.208e+03 3.138e+03 7.099e+03, threshold=4.416e+03, percent-clipped=14.0
+2023-03-04 15:26:05,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=373329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:26:22,186 INFO [train.py:968] (0/2) Epoch 9, batch 9150, giga_loss[loss=0.2896, simple_loss=0.3508, pruned_loss=0.1142, over 28425.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3822, pruned_loss=0.1325, over 5681981.79 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.36, pruned_loss=0.1039, over 5703648.40 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3849, pruned_loss=0.1355, over 5669878.95 frames. ], batch size: 78, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:26:40,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1166, 1.6276, 1.4640, 1.0516], device='cuda:0'), covar=tensor([0.1458, 0.2349, 0.1333, 0.1495], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0708, 0.0834, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 15:26:51,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1832, 1.2093, 3.8525, 3.2436], device='cuda:0'), covar=tensor([0.1604, 0.2390, 0.0413, 0.0754], device='cuda:0'), in_proj_covar=tensor([0.0631, 0.0578, 0.0839, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:27:06,909 INFO [train.py:968] (0/2) Epoch 9, batch 9200, giga_loss[loss=0.2875, simple_loss=0.3611, pruned_loss=0.1069, over 29012.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3804, pruned_loss=0.1313, over 5687116.50 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3599, pruned_loss=0.1039, over 5709235.57 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3833, pruned_loss=0.1347, over 5671552.67 frames. ], batch size: 155, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:27:13,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7078, 1.7159, 1.2355, 1.3971], device='cuda:0'), covar=tensor([0.0617, 0.0516, 0.0888, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0451, 0.0499, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:27:23,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.717e+03 2.253e+03 3.450e+03 9.316e+03, threshold=4.506e+03, percent-clipped=16.0
+2023-03-04 15:27:42,388 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=373437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:27:53,018 INFO [train.py:968] (0/2) Epoch 9, batch 9250, giga_loss[loss=0.3161, simple_loss=0.3755, pruned_loss=0.1284, over 28262.00 frames. ], tot_loss[loss=0.32, simple_loss=0.379, pruned_loss=0.1305, over 5687651.71 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3596, pruned_loss=0.1037, over 5710438.02 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3818, pruned_loss=0.1337, over 5673968.26 frames. ], batch size: 368, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:28:04,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=373462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:28:34,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1261, 1.5346, 1.5337, 1.0713], device='cuda:0'), covar=tensor([0.1404, 0.2393, 0.1218, 0.1502], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0709, 0.0837, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 15:28:37,597 INFO [train.py:968] (0/2) Epoch 9, batch 9300, libri_loss[loss=0.2795, simple_loss=0.3585, pruned_loss=0.1002, over 29545.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3809, pruned_loss=0.1309, over 5683504.37 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3598, pruned_loss=0.1037, over 5717456.62 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3837, pruned_loss=0.1344, over 5665341.04 frames. ], batch size: 84, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:28:52,365 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.453e+02 1.579e+03 2.085e+03 2.628e+03 5.936e+03, threshold=4.170e+03, percent-clipped=7.0
+2023-03-04 15:29:17,989 INFO [train.py:968] (0/2) Epoch 9, batch 9350, giga_loss[loss=0.3751, simple_loss=0.4204, pruned_loss=0.1649, over 27778.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3834, pruned_loss=0.1325, over 5688824.27 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3595, pruned_loss=0.1036, over 5721961.25 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3864, pruned_loss=0.136, over 5669544.93 frames. ], batch size: 412, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:29:18,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4703, 3.2431, 1.5423, 1.5038], device='cuda:0'), covar=tensor([0.0819, 0.0312, 0.0775, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0500, 0.0327, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 15:29:20,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2011, 1.5797, 1.2456, 1.3491], device='cuda:0'), covar=tensor([0.2357, 0.2239, 0.2419, 0.1964], device='cuda:0'), in_proj_covar=tensor([0.1237, 0.0926, 0.1091, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 15:29:39,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4822, 3.4018, 1.4715, 1.5368], device='cuda:0'), covar=tensor([0.0862, 0.0300, 0.0805, 0.1244], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0500, 0.0328, 0.0380], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 15:30:03,357 INFO [train.py:968] (0/2) Epoch 9, batch 9400, libri_loss[loss=0.3403, simple_loss=0.4145, pruned_loss=0.1331, over 29655.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3827, pruned_loss=0.1321, over 5690487.78 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3598, pruned_loss=0.1037, over 5727572.42 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3855, pruned_loss=0.1357, over 5668692.89 frames. ], batch size: 91, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:30:10,097 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=373605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:30:12,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=373608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:30:19,978 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.324e+02 1.416e+03 1.739e+03 2.356e+03 4.757e+03, threshold=3.478e+03, percent-clipped=3.0
+2023-03-04 15:30:38,288 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=373637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:30:46,649 INFO [train.py:968] (0/2) Epoch 9, batch 9450, giga_loss[loss=0.3188, simple_loss=0.3844, pruned_loss=0.1266, over 28741.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3842, pruned_loss=0.1308, over 5692370.15 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3598, pruned_loss=0.1037, over 5730881.82 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.387, pruned_loss=0.1343, over 5671028.34 frames. ], batch size: 284, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:31:27,801 INFO [train.py:968] (0/2) Epoch 9, batch 9500, giga_loss[loss=0.432, simple_loss=0.4363, pruned_loss=0.2138, over 23600.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3854, pruned_loss=0.1296, over 5688337.22 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3597, pruned_loss=0.1036, over 5730348.16 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3884, pruned_loss=0.1332, over 5670337.52 frames. ], batch size: 705, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:31:32,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=373704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:31:32,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1254, 1.5374, 1.4728, 1.0578], device='cuda:0'), covar=tensor([0.1509, 0.2069, 0.1315, 0.1476], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0707, 0.0835, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 15:31:43,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.369e+02 1.383e+03 1.707e+03 2.436e+03 5.515e+03, threshold=3.413e+03, percent-clipped=12.0
+2023-03-04 15:31:53,406 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-04 15:32:04,545 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-04 15:32:11,523 INFO [train.py:968] (0/2) Epoch 9, batch 9550, giga_loss[loss=0.281, simple_loss=0.3634, pruned_loss=0.09925, over 28642.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3882, pruned_loss=0.1309, over 5685842.73 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3598, pruned_loss=0.1037, over 5732979.51 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.391, pruned_loss=0.1341, over 5668669.19 frames. ], batch size: 92, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:32:57,374 INFO [train.py:968] (0/2) Epoch 9, batch 9600, giga_loss[loss=0.3234, simple_loss=0.3897, pruned_loss=0.1285, over 28624.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3926, pruned_loss=0.1356, over 5679344.11 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.36, pruned_loss=0.1039, over 5731019.15 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3949, pruned_loss=0.1381, over 5666942.72 frames. ], batch size: 262, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:33:08,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=373812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:33:10,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.805e+02 1.547e+03 2.080e+03 2.675e+03 5.706e+03, threshold=4.161e+03, percent-clipped=6.0
+2023-03-04 15:33:38,422 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=373847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:33:38,743 INFO [train.py:968] (0/2) Epoch 9, batch 9650, libri_loss[loss=0.2992, simple_loss=0.3661, pruned_loss=0.1161, over 29548.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3937, pruned_loss=0.1375, over 5678652.81 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3605, pruned_loss=0.1044, over 5727630.81 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3961, pruned_loss=0.1402, over 5670222.40 frames. ], batch size: 80, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:33:40,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=373850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:33:46,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1247, 1.1660, 3.6505, 3.2023], device='cuda:0'), covar=tensor([0.1540, 0.2406, 0.0403, 0.1065], device='cuda:0'), in_proj_covar=tensor([0.0626, 0.0573, 0.0832, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:33:54,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4252, 1.5992, 1.3686, 1.4917], device='cuda:0'), covar=tensor([0.2071, 0.2023, 0.2141, 0.1800], device='cuda:0'), in_proj_covar=tensor([0.1238, 0.0928, 0.1093, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 15:34:05,808 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=373879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:34:09,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0377, 5.1591, 2.1111, 2.3550], device='cuda:0'), covar=tensor([0.0835, 0.0268, 0.0760, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0502, 0.0328, 0.0381], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 15:34:21,752 INFO [train.py:968] (0/2) Epoch 9, batch 9700, giga_loss[loss=0.3319, simple_loss=0.3846, pruned_loss=0.1396, over 28617.00 frames. ], tot_loss[loss=0.336, simple_loss=0.394, pruned_loss=0.139, over 5659946.55 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3609, pruned_loss=0.1047, over 5721144.63 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3964, pruned_loss=0.1418, over 5657975.38 frames. ], batch size: 78, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:34:39,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.692e+02 1.735e+03 2.441e+03 3.657e+03 1.577e+04, threshold=4.882e+03, percent-clipped=16.0
+2023-03-04 15:34:44,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=373922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:34:59,388 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=373938.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:35:06,134 INFO [train.py:968] (0/2) Epoch 9, batch 9750, libri_loss[loss=0.2538, simple_loss=0.3323, pruned_loss=0.08761, over 29557.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3916, pruned_loss=0.137, over 5655251.50 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3606, pruned_loss=0.1046, over 5722171.61 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3942, pruned_loss=0.1398, over 5651657.68 frames. ], batch size: 77, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:35:11,077 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=373955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:35:14,378 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=373958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:35:23,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1982, 2.6287, 1.2205, 1.2360], device='cuda:0'), covar=tensor([0.0945, 0.0347, 0.0900, 0.1395], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0501, 0.0326, 0.0379], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 15:35:40,008 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=373987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:35:49,925 INFO [train.py:968] (0/2) Epoch 9, batch 9800, giga_loss[loss=0.2817, simple_loss=0.3638, pruned_loss=0.09979, over 28826.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3898, pruned_loss=0.1338, over 5658202.07 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3606, pruned_loss=0.1046, over 5722265.98 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3921, pruned_loss=0.1363, over 5654620.44 frames. ], batch size: 92, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:35:51,281 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-374000.pt
+2023-03-04 15:35:56,582 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:36:05,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.190e+02 1.354e+03 2.025e+03 2.771e+03 5.960e+03, threshold=4.050e+03, percent-clipped=2.0
+2023-03-04 15:36:25,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-04 15:36:30,727 INFO [train.py:968] (0/2) Epoch 9, batch 9850, libri_loss[loss=0.2705, simple_loss=0.3461, pruned_loss=0.09742, over 29572.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.389, pruned_loss=0.1321, over 5666523.96 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.36, pruned_loss=0.1043, over 5723266.04 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3921, pruned_loss=0.135, over 5661277.17 frames. ], batch size: 79, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:37:00,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9046, 3.6960, 3.4497, 1.7616], device='cuda:0'), covar=tensor([0.0637, 0.0809, 0.0872, 0.2126], device='cuda:0'), in_proj_covar=tensor([0.0999, 0.0941, 0.0829, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 15:37:13,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8180, 1.7263, 1.2791, 1.4133], device='cuda:0'), covar=tensor([0.0654, 0.0613, 0.0982, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0448, 0.0498, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:37:16,840 INFO [train.py:968] (0/2) Epoch 9, batch 9900, giga_loss[loss=0.4068, simple_loss=0.4357, pruned_loss=0.1889, over 26701.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3884, pruned_loss=0.1318, over 5667999.03 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3597, pruned_loss=0.104, over 5728237.24 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3919, pruned_loss=0.135, over 5657990.53 frames. ], batch size: 555, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:37:24,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7052, 2.1012, 1.9219, 1.9015], device='cuda:0'), covar=tensor([0.0701, 0.0255, 0.0269, 0.0733], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0074, 0.0053, 0.0048, 0.0080], device='cuda:0')
+2023-03-04 15:37:27,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6631, 1.6146, 1.1409, 1.2725], device='cuda:0'), covar=tensor([0.0668, 0.0587, 0.0998, 0.0974], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0450, 0.0499, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:37:28,208 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:37:34,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:37:37,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.366e+02 1.447e+03 1.838e+03 2.435e+03 6.655e+03, threshold=3.676e+03, percent-clipped=5.0
+2023-03-04 15:38:04,528 INFO [train.py:968] (0/2) Epoch 9, batch 9950, giga_loss[loss=0.3529, simple_loss=0.4025, pruned_loss=0.1517, over 27654.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3886, pruned_loss=0.1322, over 5669673.13 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3602, pruned_loss=0.1043, over 5731684.48 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3916, pruned_loss=0.1351, over 5657178.81 frames. ], batch size: 472, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:38:37,514 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-04 15:38:48,178 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-04 15:38:49,135 INFO [train.py:968] (0/2) Epoch 9, batch 10000, libri_loss[loss=0.2434, simple_loss=0.3199, pruned_loss=0.08345, over 29359.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3872, pruned_loss=0.1323, over 5672655.76 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.36, pruned_loss=0.1041, over 5734665.54 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3903, pruned_loss=0.1354, over 5658891.35 frames. ], batch size: 67, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:39:04,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.743e+02 1.648e+03 2.144e+03 3.112e+03 1.037e+04, threshold=4.288e+03, percent-clipped=15.0
+2023-03-04 15:39:28,301 INFO [train.py:968] (0/2) Epoch 9, batch 10050, giga_loss[loss=0.3125, simple_loss=0.3658, pruned_loss=0.1296, over 28516.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3856, pruned_loss=0.1316, over 5673529.92 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3601, pruned_loss=0.104, over 5737644.05 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3894, pruned_loss=0.1357, over 5656529.39 frames. ], batch size: 78, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:39:58,978 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374282.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:40:11,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:40:12,523 INFO [train.py:968] (0/2) Epoch 9, batch 10100, giga_loss[loss=0.3262, simple_loss=0.3864, pruned_loss=0.133, over 28772.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3838, pruned_loss=0.1307, over 5680492.22 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.36, pruned_loss=0.1038, over 5741137.35 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3875, pruned_loss=0.1349, over 5662079.53 frames. ], batch size: 284, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:40:26,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374313.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:40:31,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.882e+02 1.522e+03 1.790e+03 2.388e+03 7.444e+03, threshold=3.579e+03, percent-clipped=3.0
+2023-03-04 15:40:54,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3558, 1.7718, 1.2159, 0.5829], device='cuda:0'), covar=tensor([0.2408, 0.1358, 0.1758, 0.3345], device='cuda:0'), in_proj_covar=tensor([0.1504, 0.1429, 0.1458, 0.1223], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 15:40:59,176 INFO [train.py:968] (0/2) Epoch 9, batch 10150, giga_loss[loss=0.3886, simple_loss=0.4069, pruned_loss=0.1852, over 23372.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3836, pruned_loss=0.1318, over 5672424.49 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3602, pruned_loss=0.1039, over 5738231.16 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3868, pruned_loss=0.1354, over 5659276.80 frames. ], batch size: 705, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:40:59,934 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:41:10,570 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-04 15:41:30,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:41:33,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5507, 1.8651, 1.8354, 1.3981], device='cuda:0'), covar=tensor([0.1591, 0.2069, 0.1248, 0.1437], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0712, 0.0841, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 15:41:43,557 INFO [train.py:968] (0/2) Epoch 9, batch 10200, libri_loss[loss=0.3088, simple_loss=0.382, pruned_loss=0.1178, over 29745.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.383, pruned_loss=0.1318, over 5679185.45 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3602, pruned_loss=0.104, over 5741981.73 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.386, pruned_loss=0.1351, over 5664187.36 frames. ], batch size: 87, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:42:01,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.697e+03 2.243e+03 3.090e+03 7.454e+03, threshold=4.485e+03, percent-clipped=15.0
+2023-03-04 15:42:23,954 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:42:25,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:42:29,622 INFO [train.py:968] (0/2) Epoch 9, batch 10250, giga_loss[loss=0.335, simple_loss=0.3961, pruned_loss=0.1369, over 28945.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3811, pruned_loss=0.1298, over 5670839.68 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3604, pruned_loss=0.1043, over 5744667.17 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3838, pruned_loss=0.1328, over 5654947.74 frames. ], batch size: 213, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:42:35,168 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-04 15:42:37,507 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374456.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:42:39,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374459.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:42:51,716 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:43:00,688 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:43:04,200 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374488.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:43:06,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:43:12,609 INFO [train.py:968] (0/2) Epoch 9, batch 10300, giga_loss[loss=0.297, simple_loss=0.3688, pruned_loss=0.1126, over 28217.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.377, pruned_loss=0.1256, over 5674599.88 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3604, pruned_loss=0.1041, over 5746607.88 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3797, pruned_loss=0.1288, over 5658132.09 frames. ], batch size: 368, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:43:28,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.949e+02 1.281e+03 1.553e+03 2.115e+03 5.128e+03, threshold=3.105e+03, percent-clipped=2.0
+2023-03-04 15:43:35,869 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:43:39,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:43:44,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5244, 2.0950, 1.6347, 0.7641], device='cuda:0'), covar=tensor([0.3318, 0.1889, 0.2302, 0.3719], device='cuda:0'), in_proj_covar=tensor([0.1515, 0.1436, 0.1471, 0.1232], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 15:43:57,563 INFO [train.py:968] (0/2) Epoch 9, batch 10350, giga_loss[loss=0.329, simple_loss=0.387, pruned_loss=0.1356, over 27943.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3754, pruned_loss=0.1242, over 5674825.17 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3597, pruned_loss=0.1037, over 5751620.54 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3786, pruned_loss=0.1278, over 5655280.30 frames. ], batch size: 412, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:44:02,005 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3402, 1.9586, 1.4540, 0.5184], device='cuda:0'), covar=tensor([0.2842, 0.1643, 0.2492, 0.3793], device='cuda:0'), in_proj_covar=tensor([0.1515, 0.1434, 0.1471, 0.1231], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 15:44:03,842 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374555.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:44:05,127 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:44:43,339 INFO [train.py:968] (0/2) Epoch 9, batch 10400, giga_loss[loss=0.3231, simple_loss=0.3603, pruned_loss=0.143, over 23473.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3741, pruned_loss=0.1245, over 5674170.89 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3599, pruned_loss=0.1037, over 5752665.05 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3768, pruned_loss=0.1278, over 5656321.70 frames. ], batch size: 705, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:45:04,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.332e+02 1.649e+03 2.229e+03 3.482e+03 1.052e+04, threshold=4.459e+03, percent-clipped=31.0
+2023-03-04 15:45:09,750 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:11,212 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 15:45:11,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:16,860 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:20,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:30,144 INFO [train.py:968] (0/2) Epoch 9, batch 10450, giga_loss[loss=0.3094, simple_loss=0.3716, pruned_loss=0.1236, over 28673.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3719, pruned_loss=0.1236, over 5666253.12 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3599, pruned_loss=0.1037, over 5742991.03 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3744, pruned_loss=0.1266, over 5658512.75 frames. ], batch size: 262, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:45:40,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374657.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:45:41,081 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:45,646 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:53,686 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:46:06,555 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:46:08,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374694.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:46:10,547 INFO [train.py:968] (0/2) Epoch 9, batch 10500, giga_loss[loss=0.2593, simple_loss=0.3379, pruned_loss=0.09036, over 28782.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3739, pruned_loss=0.1239, over 5674687.50 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3598, pruned_loss=0.1033, over 5748765.85 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3764, pruned_loss=0.1273, over 5660800.03 frames. ], batch size: 99, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:46:28,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.791e+02 1.371e+03 1.836e+03 2.784e+03 9.453e+03, threshold=3.673e+03, percent-clipped=5.0
+2023-03-04 15:46:29,397 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 15:46:32,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:46:53,866 INFO [train.py:968] (0/2) Epoch 9, batch 10550, giga_loss[loss=0.2939, simple_loss=0.3699, pruned_loss=0.109, over 28866.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3759, pruned_loss=0.1247, over 5680607.48 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3599, pruned_loss=0.1033, over 5753381.34 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3783, pruned_loss=0.1282, over 5662833.67 frames. ], batch size: 174, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:47:43,159 INFO [train.py:968] (0/2) Epoch 9, batch 10600, giga_loss[loss=0.3279, simple_loss=0.3817, pruned_loss=0.1371, over 27934.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3758, pruned_loss=0.125, over 5650874.11 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3598, pruned_loss=0.1031, over 5755536.21 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.378, pruned_loss=0.1283, over 5634018.17 frames. ], batch size: 412, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:47:45,358 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374800.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:47:47,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374803.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:48:04,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.034e+02 1.353e+03 1.684e+03 2.308e+03 5.954e+03, threshold=3.367e+03, percent-clipped=7.0
+2023-03-04 15:48:15,067 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374832.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:48:29,540 INFO [train.py:968] (0/2) Epoch 9, batch 10650, giga_loss[loss=0.3108, simple_loss=0.3751, pruned_loss=0.1232, over 29115.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3761, pruned_loss=0.1254, over 5652413.17 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3605, pruned_loss=0.1034, over 5760111.90 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3778, pruned_loss=0.1285, over 5631060.37 frames. ], batch size: 128, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:48:43,030 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:48:46,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:49:12,214 INFO [train.py:968] (0/2) Epoch 9, batch 10700, giga_loss[loss=0.3751, simple_loss=0.4027, pruned_loss=0.1737, over 23540.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3772, pruned_loss=0.1265, over 5654327.14 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3606, pruned_loss=0.1034, over 5760447.83 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.379, pruned_loss=0.1298, over 5633252.23 frames. ], batch size: 705, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:49:14,840 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:49:34,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.764e+02 1.529e+03 2.179e+03 2.998e+03 1.330e+04, threshold=4.358e+03, percent-clipped=21.0
+2023-03-04 15:49:43,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374930.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:49:58,434 INFO [train.py:968] (0/2) Epoch 9, batch 10750, giga_loss[loss=0.3291, simple_loss=0.3875, pruned_loss=0.1354, over 28696.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3789, pruned_loss=0.1272, over 5653934.29 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3607, pruned_loss=0.1035, over 5754444.08 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.381, pruned_loss=0.1308, over 5638203.16 frames. ], batch size: 307, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:50:13,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2435, 1.2188, 1.0460, 0.9467], device='cuda:0'), covar=tensor([0.0620, 0.0441, 0.0908, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0447, 0.0498, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:50:23,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8625, 2.8905, 1.9267, 0.7210], device='cuda:0'), covar=tensor([0.4560, 0.2001, 0.2547, 0.4755], device='cuda:0'), in_proj_covar=tensor([0.1507, 0.1420, 0.1461, 0.1225], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 15:50:43,655 INFO [train.py:968] (0/2) Epoch 9, batch 10800, giga_loss[loss=0.3433, simple_loss=0.3976, pruned_loss=0.1445, over 28241.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3799, pruned_loss=0.1279, over 5659948.29 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3607, pruned_loss=0.1034, over 5757961.67 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3822, pruned_loss=0.1316, over 5641219.65 frames. ], batch size: 368, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:51:04,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.349e+02 1.454e+03 1.776e+03 2.286e+03 8.204e+03, threshold=3.551e+03, percent-clipped=2.0
+2023-03-04 15:51:21,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4748, 1.4935, 1.5180, 1.4246], device='cuda:0'), covar=tensor([0.0967, 0.1410, 0.1424, 0.1307], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0739, 0.0654, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 15:51:29,880 INFO [train.py:968] (0/2) Epoch 9, batch 10850, giga_loss[loss=0.334, simple_loss=0.3886, pruned_loss=0.1397, over 27945.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3811, pruned_loss=0.1294, over 5655279.82 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3605, pruned_loss=0.1034, over 5752579.45 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3838, pruned_loss=0.1332, over 5641366.02 frames. ], batch size: 412, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:51:33,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:51:49,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:51:51,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375069.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:51:55,179 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375073.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:51:58,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375076.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:51:58,908 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4632, 1.6752, 1.3858, 1.8009], device='cuda:0'), covar=tensor([0.2071, 0.2066, 0.2225, 0.1880], device='cuda:0'), in_proj_covar=tensor([0.1244, 0.0929, 0.1095, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 15:52:17,895 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7021, 1.5849, 1.1866, 1.3366], device='cuda:0'), covar=tensor([0.0591, 0.0583, 0.0918, 0.0975], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0444, 0.0494, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:52:18,926 INFO [train.py:968] (0/2) Epoch 9, batch 10900, giga_loss[loss=0.3487, simple_loss=0.4024, pruned_loss=0.1475, over 28838.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3831, pruned_loss=0.1314, over 5657981.54 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3601, pruned_loss=0.1032, over 5754094.62 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3858, pruned_loss=0.1348, over 5644725.97 frames. ], batch size: 174, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:52:26,877 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375105.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:52:39,957 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.887e+02 1.884e+03 2.490e+03 3.435e+03 1.587e+04, threshold=4.980e+03, percent-clipped=21.0
+2023-03-04 15:53:06,796 INFO [train.py:968] (0/2) Epoch 9, batch 10950, giga_loss[loss=0.2909, simple_loss=0.3697, pruned_loss=0.1061, over 28906.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3829, pruned_loss=0.1295, over 5663564.26 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3597, pruned_loss=0.1029, over 5756591.84 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3861, pruned_loss=0.1333, over 5648360.94 frames. ], batch size: 174, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:53:09,621 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6762, 1.6941, 1.2280, 1.3366], device='cuda:0'), covar=tensor([0.0644, 0.0539, 0.0949, 0.0940], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0443, 0.0494, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:53:18,173 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 15:53:23,160 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4137, 1.5007, 1.2796, 1.5446], device='cuda:0'), covar=tensor([0.2211, 0.2200, 0.2333, 0.1988], device='cuda:0'), in_proj_covar=tensor([0.1241, 0.0926, 0.1092, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 15:53:53,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:53:57,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:53:57,708 INFO [train.py:968] (0/2) Epoch 9, batch 11000, giga_loss[loss=0.2925, simple_loss=0.3661, pruned_loss=0.1095, over 28938.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3817, pruned_loss=0.1296, over 5652038.58 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3589, pruned_loss=0.1026, over 5758485.43 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3853, pruned_loss=0.1334, over 5636401.22 frames. ], batch size: 164, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:54:04,429 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3448, 1.5098, 1.5195, 1.4721], device='cuda:0'), covar=tensor([0.1310, 0.1467, 0.1844, 0.1473], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0739, 0.0656, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 15:54:05,223 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:54:09,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:54:11,073 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:54:14,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:54:15,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375215.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:54:19,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.289e+02 1.665e+03 2.253e+03 3.042e+03 6.942e+03, threshold=4.506e+03, percent-clipped=6.0
+2023-03-04 15:54:24,949 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:54:42,210 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:54:43,959 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375244.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:54:47,286 INFO [train.py:968] (0/2) Epoch 9, batch 11050, giga_loss[loss=0.3097, simple_loss=0.3693, pruned_loss=0.1251, over 28919.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3807, pruned_loss=0.1299, over 5664842.53 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3592, pruned_loss=0.1028, over 5760859.56 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3836, pruned_loss=0.133, over 5649134.92 frames. ], batch size: 136, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:55:44,816 INFO [train.py:968] (0/2) Epoch 9, batch 11100, giga_loss[loss=0.2995, simple_loss=0.3609, pruned_loss=0.119, over 28636.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3815, pruned_loss=0.1315, over 5658530.93 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3592, pruned_loss=0.1027, over 5763105.70 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3842, pruned_loss=0.1346, over 5642563.41 frames. ], batch size: 307, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:55:51,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 15:56:01,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0527, 1.2655, 3.6328, 3.0645], device='cuda:0'), covar=tensor([0.1599, 0.2379, 0.0424, 0.0752], device='cuda:0'), in_proj_covar=tensor([0.0620, 0.0569, 0.0821, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 15:56:06,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.522e+02 1.564e+03 2.068e+03 2.963e+03 9.963e+03, threshold=4.135e+03, percent-clipped=7.0
+2023-03-04 15:56:31,170 INFO [train.py:968] (0/2) Epoch 9, batch 11150, giga_loss[loss=0.342, simple_loss=0.3889, pruned_loss=0.1475, over 27984.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3802, pruned_loss=0.1304, over 5670377.41 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3597, pruned_loss=0.1031, over 5762444.86 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3823, pruned_loss=0.133, over 5656661.02 frames. ], batch size: 412, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:56:42,424 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 15:56:49,721 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:57:13,655 INFO [train.py:968] (0/2) Epoch 9, batch 11200, giga_loss[loss=0.2737, simple_loss=0.346, pruned_loss=0.1007, over 28948.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3793, pruned_loss=0.1302, over 5675081.28 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3594, pruned_loss=0.1029, over 5765398.41 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3817, pruned_loss=0.1331, over 5659187.48 frames. ], batch size: 186, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 15:57:25,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9284, 3.6915, 3.4964, 1.7418], device='cuda:0'), covar=tensor([0.0810, 0.1123, 0.1170, 0.2193], device='cuda:0'), in_proj_covar=tensor([0.1011, 0.0956, 0.0840, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 15:57:35,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.093e+02 1.484e+03 1.817e+03 2.406e+03 4.820e+03, threshold=3.635e+03, percent-clipped=6.0
+2023-03-04 15:57:57,919 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:58:01,169 INFO [train.py:968] (0/2) Epoch 9, batch 11250, libri_loss[loss=0.295, simple_loss=0.3791, pruned_loss=0.1055, over 27485.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.38, pruned_loss=0.1307, over 5672473.53 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3599, pruned_loss=0.1031, over 5766246.86 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3819, pruned_loss=0.1334, over 5657640.95 frames. ], batch size: 115, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 15:58:29,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2424, 0.7809, 0.8138, 1.4249], device='cuda:0'), covar=tensor([0.0736, 0.0357, 0.0345, 0.0761], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0074, 0.0053, 0.0048, 0.0081], device='cuda:0')
+2023-03-04 15:58:48,803 INFO [train.py:968] (0/2) Epoch 9, batch 11300, libri_loss[loss=0.2708, simple_loss=0.3576, pruned_loss=0.09202, over 29544.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3818, pruned_loss=0.1322, over 5673980.39 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3601, pruned_loss=0.1032, over 5768897.20 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3835, pruned_loss=0.1347, over 5657969.99 frames. ], batch size: 84, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 15:59:11,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.579e+02 1.662e+03 2.046e+03 2.697e+03 7.221e+03, threshold=4.093e+03, percent-clipped=8.0
+2023-03-04 15:59:34,742 INFO [train.py:968] (0/2) Epoch 9, batch 11350, giga_loss[loss=0.3722, simple_loss=0.4147, pruned_loss=0.1648, over 28614.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3831, pruned_loss=0.1333, over 5675902.78 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3602, pruned_loss=0.1032, over 5770524.32 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.385, pruned_loss=0.1361, over 5659416.07 frames. ], batch size: 307, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 15:59:59,060 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:00:02,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:00:19,080 INFO [train.py:968] (0/2) Epoch 9, batch 11400, giga_loss[loss=0.3514, simple_loss=0.404, pruned_loss=0.1494, over 28621.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3818, pruned_loss=0.1317, over 5681566.48 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3595, pruned_loss=0.1028, over 5775023.12 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3849, pruned_loss=0.1355, over 5660487.08 frames. ], batch size: 242, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:00:42,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.879e+02 1.657e+03 2.337e+03 3.565e+03 9.147e+03, threshold=4.674e+03, percent-clipped=15.0
+2023-03-04 16:01:09,736 INFO [train.py:968] (0/2) Epoch 9, batch 11450, giga_loss[loss=0.3131, simple_loss=0.3657, pruned_loss=0.1302, over 28869.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3833, pruned_loss=0.1338, over 5674593.08 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3596, pruned_loss=0.1028, over 5775879.30 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3859, pruned_loss=0.1372, over 5656273.03 frames. ], batch size: 99, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:01:26,732 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-04 16:01:55,427 INFO [train.py:968] (0/2) Epoch 9, batch 11500, giga_loss[loss=0.3955, simple_loss=0.423, pruned_loss=0.184, over 26757.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3827, pruned_loss=0.1338, over 5661620.52 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3597, pruned_loss=0.1031, over 5769588.61 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3853, pruned_loss=0.1371, over 5650289.10 frames. ], batch size: 555, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:02:19,912 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.628e+03 2.309e+03 3.147e+03 6.217e+03, threshold=4.618e+03, percent-clipped=8.0
+2023-03-04 16:02:23,628 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:02:25,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:02:41,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:02:44,688 INFO [train.py:968] (0/2) Epoch 9, batch 11550, giga_loss[loss=0.327, simple_loss=0.3857, pruned_loss=0.1342, over 28874.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3831, pruned_loss=0.1336, over 5669654.68 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3595, pruned_loss=0.103, over 5767337.70 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3859, pruned_loss=0.1369, over 5661014.26 frames. ], batch size: 227, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:02:52,641 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:03:29,104 INFO [train.py:968] (0/2) Epoch 9, batch 11600, giga_loss[loss=0.3121, simple_loss=0.3798, pruned_loss=0.1222, over 28865.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3837, pruned_loss=0.1334, over 5667802.10 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3599, pruned_loss=0.1031, over 5767815.61 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3861, pruned_loss=0.1367, over 5657450.46 frames. ], batch size: 112, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:03:46,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3660, 1.7578, 1.7443, 1.2725], device='cuda:0'), covar=tensor([0.1539, 0.2154, 0.1202, 0.1442], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0719, 0.0842, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 16:03:49,531 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:03:51,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.470e+03 1.948e+03 2.532e+03 4.654e+03, threshold=3.895e+03, percent-clipped=2.0
+2023-03-04 16:03:54,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2911, 1.7113, 1.2756, 0.4527], device='cuda:0'), covar=tensor([0.1592, 0.1068, 0.1711, 0.2787], device='cuda:0'), in_proj_covar=tensor([0.1524, 0.1442, 0.1475, 0.1240], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 16:04:06,889 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:04:08,342 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:04:19,665 INFO [train.py:968] (0/2) Epoch 9, batch 11650, libri_loss[loss=0.2481, simple_loss=0.3374, pruned_loss=0.07942, over 29577.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3846, pruned_loss=0.1337, over 5679690.89 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3594, pruned_loss=0.1027, over 5770844.43 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3876, pruned_loss=0.1375, over 5666370.58 frames. ], batch size: 76, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:04:56,823 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:04:59,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:05:06,782 INFO [train.py:968] (0/2) Epoch 9, batch 11700, giga_loss[loss=0.309, simple_loss=0.3737, pruned_loss=0.1221, over 28904.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3866, pruned_loss=0.1357, over 5679716.53 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.359, pruned_loss=0.1024, over 5774581.01 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3902, pruned_loss=0.1399, over 5663431.50 frames. ], batch size: 145, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:05:23,220 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-04 16:05:25,418 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:05:27,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.626e+02 1.844e+03 2.318e+03 2.967e+03 7.912e+03, threshold=4.635e+03, percent-clipped=9.0
+2023-03-04 16:05:39,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-04 16:05:52,664 INFO [train.py:968] (0/2) Epoch 9, batch 11750, giga_loss[loss=0.3424, simple_loss=0.4003, pruned_loss=0.1423, over 28816.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3857, pruned_loss=0.1345, over 5681478.50 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.359, pruned_loss=0.1024, over 5767404.67 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3889, pruned_loss=0.1384, over 5673981.86 frames. ], batch size: 186, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:05:56,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:06:04,274 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:06:06,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:06:13,324 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-04 16:06:29,827 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:06:35,333 INFO [train.py:968] (0/2) Epoch 9, batch 11800, giga_loss[loss=0.3122, simple_loss=0.3808, pruned_loss=0.1218, over 28804.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3858, pruned_loss=0.1339, over 5689374.24 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3587, pruned_loss=0.1023, over 5773673.10 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3899, pruned_loss=0.1384, over 5674566.43 frames. ], batch size: 284, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:06:36,608 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-376000.pt
+2023-03-04 16:06:38,047 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-04 16:06:59,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.307e+02 1.516e+03 1.907e+03 2.504e+03 8.116e+03, threshold=3.813e+03, percent-clipped=4.0
+2023-03-04 16:07:21,078 INFO [train.py:968] (0/2) Epoch 9, batch 11850, giga_loss[loss=0.2974, simple_loss=0.373, pruned_loss=0.1109, over 28929.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3857, pruned_loss=0.1332, over 5677297.14 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3583, pruned_loss=0.1022, over 5765927.22 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3896, pruned_loss=0.1373, over 5671566.70 frames. ], batch size: 227, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:08:03,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=376095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:08:06,322 INFO [train.py:968] (0/2) Epoch 9, batch 11900, giga_loss[loss=0.3157, simple_loss=0.3786, pruned_loss=0.1265, over 28925.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3829, pruned_loss=0.1313, over 5674966.85 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3574, pruned_loss=0.1018, over 5768451.52 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3875, pruned_loss=0.1356, over 5665969.47 frames. ], batch size: 227, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:08:06,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=376098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:08:30,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.746e+02 1.406e+03 1.999e+03 2.696e+03 7.344e+03, threshold=3.998e+03, percent-clipped=10.0
+2023-03-04 16:08:35,635 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=376127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:08:52,541 INFO [train.py:968] (0/2) Epoch 9, batch 11950, giga_loss[loss=0.2953, simple_loss=0.367, pruned_loss=0.1117, over 29011.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3822, pruned_loss=0.1306, over 5688040.63 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3578, pruned_loss=0.102, over 5770293.75 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3861, pruned_loss=0.1345, over 5676910.91 frames. ], batch size: 164, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:09:38,983 INFO [train.py:968] (0/2) Epoch 9, batch 12000, giga_loss[loss=0.3243, simple_loss=0.3866, pruned_loss=0.131, over 28817.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3838, pruned_loss=0.1323, over 5664006.63 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3579, pruned_loss=0.102, over 5763139.25 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3876, pruned_loss=0.1363, over 5658876.85 frames. ], batch size: 284, lr: 3.65e-03, grad_scale: 8.0
+2023-03-04 16:09:38,988 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 16:09:47,545 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2231, simple_loss=0.328, pruned_loss=0.05912, over 944034.00 frames. 
+2023-03-04 16:09:47,545 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 16:09:58,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=376211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:10:01,591 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=376213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:10:11,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.909e+02 1.419e+03 2.028e+03 2.905e+03 7.469e+03, threshold=4.056e+03, percent-clipped=11.0
+2023-03-04 16:10:33,978 INFO [train.py:968] (0/2) Epoch 9, batch 12050, giga_loss[loss=0.356, simple_loss=0.4041, pruned_loss=0.154, over 28248.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3854, pruned_loss=0.1327, over 5672157.88 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.358, pruned_loss=0.102, over 5763690.50 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3886, pruned_loss=0.1363, over 5666275.76 frames. ], batch size: 368, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:11:24,681 INFO [train.py:968] (0/2) Epoch 9, batch 12100, giga_loss[loss=0.3131, simple_loss=0.3757, pruned_loss=0.1252, over 28911.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3862, pruned_loss=0.135, over 5668719.91 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3578, pruned_loss=0.1018, over 5764954.87 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3892, pruned_loss=0.1383, over 5662130.33 frames. ], batch size: 174, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:11:44,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.403e+02 1.606e+03 1.993e+03 2.667e+03 6.505e+03, threshold=3.986e+03, percent-clipped=4.0
+2023-03-04 16:11:56,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5355, 4.3313, 4.1334, 2.0807], device='cuda:0'), covar=tensor([0.0410, 0.0588, 0.0610, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.1017, 0.0954, 0.0844, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 16:12:05,508 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=376342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:10,182 INFO [train.py:968] (0/2) Epoch 9, batch 12150, giga_loss[loss=0.4177, simple_loss=0.433, pruned_loss=0.2012, over 26663.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.386, pruned_loss=0.1355, over 5662933.46 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.358, pruned_loss=0.1021, over 5765467.03 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.389, pruned_loss=0.1387, over 5654642.27 frames. ], batch size: 555, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:12:10,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3205, 3.1323, 2.9246, 1.3542], device='cuda:0'), covar=tensor([0.0869, 0.1089, 0.1067, 0.2274], device='cuda:0'), in_proj_covar=tensor([0.1015, 0.0953, 0.0843, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 16:12:15,449 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=376354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:16,774 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=376356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:17,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=376357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:18,925 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=376359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:46,406 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=376386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:48,667 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=376388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:56,230 INFO [train.py:968] (0/2) Epoch 9, batch 12200, giga_loss[loss=0.3644, simple_loss=0.4134, pruned_loss=0.1577, over 28581.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3854, pruned_loss=0.1345, over 5666461.32 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3581, pruned_loss=0.102, over 5762376.92 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3889, pruned_loss=0.1387, over 5658861.12 frames. ], batch size: 78, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:13:19,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.509e+03 2.027e+03 3.116e+03 9.638e+03, threshold=4.054e+03, percent-clipped=12.0
+2023-03-04 16:13:26,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3599, 2.0104, 1.5574, 0.5297], device='cuda:0'), covar=tensor([0.2629, 0.1639, 0.2462, 0.3356], device='cuda:0'), in_proj_covar=tensor([0.1525, 0.1447, 0.1477, 0.1238], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 16:13:38,986 INFO [train.py:968] (0/2) Epoch 9, batch 12250, giga_loss[loss=0.3106, simple_loss=0.3791, pruned_loss=0.1211, over 28793.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3858, pruned_loss=0.1344, over 5659013.51 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3579, pruned_loss=0.1018, over 5755484.86 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3897, pruned_loss=0.139, over 5655426.04 frames. ], batch size: 199, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:13:39,411 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5276, 1.6814, 1.3863, 1.6318], device='cuda:0'), covar=tensor([0.2206, 0.2250, 0.2419, 0.2089], device='cuda:0'), in_proj_covar=tensor([0.1242, 0.0928, 0.1096, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 16:14:29,876 INFO [train.py:968] (0/2) Epoch 9, batch 12300, giga_loss[loss=0.3335, simple_loss=0.3977, pruned_loss=0.1346, over 28880.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.386, pruned_loss=0.1354, over 5646233.27 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3579, pruned_loss=0.1017, over 5756473.43 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3892, pruned_loss=0.1392, over 5642102.26 frames. ], batch size: 119, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:14:57,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.364e+02 1.344e+03 1.968e+03 3.063e+03 1.603e+04, threshold=3.936e+03, percent-clipped=11.0
+2023-03-04 16:14:58,778 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-04 16:15:11,283 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-04 16:15:17,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9133, 1.0957, 3.5079, 3.0124], device='cuda:0'), covar=tensor([0.1715, 0.2510, 0.0420, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0629, 0.0572, 0.0833, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 16:15:19,815 INFO [train.py:968] (0/2) Epoch 9, batch 12350, giga_loss[loss=0.4486, simple_loss=0.4555, pruned_loss=0.2208, over 26521.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3862, pruned_loss=0.1357, over 5644694.09 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3574, pruned_loss=0.1014, over 5758551.07 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3895, pruned_loss=0.1394, over 5638073.43 frames. ], batch size: 555, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:16:05,538 INFO [train.py:968] (0/2) Epoch 9, batch 12400, giga_loss[loss=0.3146, simple_loss=0.3873, pruned_loss=0.121, over 28711.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3859, pruned_loss=0.135, over 5651785.00 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3572, pruned_loss=0.1013, over 5760391.79 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3891, pruned_loss=0.1385, over 5643427.29 frames. ], batch size: 119, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:16:23,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7436, 5.0111, 1.9876, 1.7833], device='cuda:0'), covar=tensor([0.0906, 0.0305, 0.0775, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0503, 0.0329, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 16:16:29,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.738e+02 1.626e+03 2.129e+03 2.701e+03 5.373e+03, threshold=4.257e+03, percent-clipped=10.0
+2023-03-04 16:16:49,195 INFO [train.py:968] (0/2) Epoch 9, batch 12450, giga_loss[loss=0.308, simple_loss=0.3686, pruned_loss=0.1237, over 28530.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3849, pruned_loss=0.1338, over 5641615.52 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3577, pruned_loss=0.1015, over 5746950.80 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3879, pruned_loss=0.1374, over 5643811.03 frames. ], batch size: 336, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:17:09,907 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 16:17:36,959 INFO [train.py:968] (0/2) Epoch 9, batch 12500, giga_loss[loss=0.351, simple_loss=0.3949, pruned_loss=0.1536, over 27590.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3822, pruned_loss=0.1319, over 5650946.69 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3576, pruned_loss=0.1015, over 5748604.47 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3851, pruned_loss=0.1354, over 5649386.02 frames. ], batch size: 472, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:17:59,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=376717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:18:07,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.104e+03 1.780e+03 2.292e+03 3.180e+03 7.182e+03, threshold=4.584e+03, percent-clipped=11.0
+2023-03-04 16:18:10,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3462, 2.8512, 1.4318, 1.4143], device='cuda:0'), covar=tensor([0.0787, 0.0321, 0.0748, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0504, 0.0330, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 16:18:11,292 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=376729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:18:27,867 INFO [train.py:968] (0/2) Epoch 9, batch 12550, giga_loss[loss=0.2924, simple_loss=0.3608, pruned_loss=0.112, over 28714.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3785, pruned_loss=0.1295, over 5664283.61 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3576, pruned_loss=0.1015, over 5749374.44 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3809, pruned_loss=0.1324, over 5662078.57 frames. ], batch size: 336, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:18:47,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6205, 1.9467, 1.4281, 2.0997], device='cuda:0'), covar=tensor([0.2163, 0.2146, 0.2417, 0.2160], device='cuda:0'), in_proj_covar=tensor([0.1240, 0.0924, 0.1097, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 16:19:16,174 INFO [train.py:968] (0/2) Epoch 9, batch 12600, giga_loss[loss=0.2957, simple_loss=0.365, pruned_loss=0.1132, over 28898.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3747, pruned_loss=0.1278, over 5648373.92 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3579, pruned_loss=0.1017, over 5748518.50 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3767, pruned_loss=0.1303, over 5645738.97 frames. ], batch size: 145, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:19:40,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.821e+02 1.504e+03 2.171e+03 2.876e+03 8.080e+03, threshold=4.341e+03, percent-clipped=7.0
+2023-03-04 16:20:02,925 INFO [train.py:968] (0/2) Epoch 9, batch 12650, giga_loss[loss=0.3092, simple_loss=0.3753, pruned_loss=0.1215, over 28596.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3761, pruned_loss=0.1301, over 5643992.74 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3581, pruned_loss=0.1019, over 5741321.87 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3777, pruned_loss=0.1324, over 5645647.40 frames. ], batch size: 307, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:20:16,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=376860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:20:18,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=376863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:20:47,243 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=376892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:20:54,691 INFO [train.py:968] (0/2) Epoch 9, batch 12700, giga_loss[loss=0.2816, simple_loss=0.3429, pruned_loss=0.1102, over 28927.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3746, pruned_loss=0.1293, over 5644673.26 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3578, pruned_loss=0.1017, over 5743007.46 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3764, pruned_loss=0.1317, over 5643476.84 frames. ], batch size: 106, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:21:19,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.741e+02 1.548e+03 1.924e+03 2.955e+03 8.647e+03, threshold=3.848e+03, percent-clipped=8.0
+2023-03-04 16:21:20,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3077, 1.6779, 1.6172, 1.2610], device='cuda:0'), covar=tensor([0.1312, 0.1907, 0.1149, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0713, 0.0842, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 16:21:39,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4288, 1.7704, 1.4141, 1.2335], device='cuda:0'), covar=tensor([0.1576, 0.1211, 0.1042, 0.1380], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1517, 0.1467, 0.1579], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 16:21:42,273 INFO [train.py:968] (0/2) Epoch 9, batch 12750, giga_loss[loss=0.3236, simple_loss=0.3953, pruned_loss=0.1259, over 28247.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3733, pruned_loss=0.127, over 5647696.82 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3577, pruned_loss=0.1016, over 5746235.56 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.375, pruned_loss=0.1294, over 5642029.69 frames. ], batch size: 368, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:22:33,542 INFO [train.py:968] (0/2) Epoch 9, batch 12800, giga_loss[loss=0.282, simple_loss=0.3497, pruned_loss=0.1072, over 28506.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3719, pruned_loss=0.1241, over 5649249.97 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3577, pruned_loss=0.1017, over 5747492.27 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3734, pruned_loss=0.1262, over 5642791.01 frames. ], batch size: 78, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:23:00,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.530e+02 1.450e+03 2.122e+03 2.797e+03 4.918e+03, threshold=4.244e+03, percent-clipped=10.0
+2023-03-04 16:23:23,953 INFO [train.py:968] (0/2) Epoch 9, batch 12850, giga_loss[loss=0.2531, simple_loss=0.3349, pruned_loss=0.08571, over 28843.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.369, pruned_loss=0.1207, over 5648499.02 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3573, pruned_loss=0.1015, over 5748752.79 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3707, pruned_loss=0.1228, over 5640885.60 frames. ], batch size: 145, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:23:57,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 16:24:01,039 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:24:17,599 INFO [train.py:968] (0/2) Epoch 9, batch 12900, giga_loss[loss=0.294, simple_loss=0.3463, pruned_loss=0.1208, over 26577.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3659, pruned_loss=0.1174, over 5649714.13 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.357, pruned_loss=0.1014, over 5749250.82 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3676, pruned_loss=0.1194, over 5641799.14 frames. ], batch size: 555, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:24:24,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=377104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:24:40,026 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:24:45,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.704e+02 1.313e+03 1.649e+03 2.335e+03 5.529e+03, threshold=3.299e+03, percent-clipped=3.0
+2023-03-04 16:24:48,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4747, 1.5507, 1.2110, 1.2045], device='cuda:0'), covar=tensor([0.0661, 0.0428, 0.0862, 0.0934], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0438, 0.0492, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 16:25:04,883 INFO [train.py:968] (0/2) Epoch 9, batch 12950, giga_loss[loss=0.273, simple_loss=0.3546, pruned_loss=0.09572, over 28023.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3615, pruned_loss=0.1129, over 5653446.79 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3565, pruned_loss=0.1012, over 5753843.29 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3636, pruned_loss=0.1151, over 5639962.02 frames. ], batch size: 412, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:25:13,349 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:25:55,601 INFO [train.py:968] (0/2) Epoch 9, batch 13000, giga_loss[loss=0.2907, simple_loss=0.3628, pruned_loss=0.1093, over 28911.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3604, pruned_loss=0.1096, over 5664421.37 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3565, pruned_loss=0.1012, over 5755009.29 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3622, pruned_loss=0.1114, over 5651411.63 frames. ], batch size: 186, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:26:05,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3634, 1.6364, 1.3588, 1.4159], device='cuda:0'), covar=tensor([0.2188, 0.1853, 0.1998, 0.1643], device='cuda:0'), in_proj_covar=tensor([0.1242, 0.0919, 0.1102, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 16:26:23,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.142e+02 1.188e+03 1.747e+03 2.432e+03 2.432e+04, threshold=3.494e+03, percent-clipped=13.0
+2023-03-04 16:26:44,701 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=377247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:26:45,127 INFO [train.py:968] (0/2) Epoch 9, batch 13050, libri_loss[loss=0.267, simple_loss=0.3328, pruned_loss=0.1005, over 29595.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3605, pruned_loss=0.1098, over 5661209.04 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3563, pruned_loss=0.1012, over 5757864.53 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3621, pruned_loss=0.1115, over 5646160.39 frames. ], batch size: 76, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:26:47,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=377250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:27:18,691 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=377279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:27:33,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3563, 1.6537, 1.6135, 1.4497], device='cuda:0'), covar=tensor([0.1316, 0.1422, 0.1544, 0.1434], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0722, 0.0649, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 16:27:37,496 INFO [train.py:968] (0/2) Epoch 9, batch 13100, giga_loss[loss=0.2764, simple_loss=0.3531, pruned_loss=0.09985, over 28534.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3581, pruned_loss=0.1079, over 5660004.83 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3563, pruned_loss=0.1013, over 5755965.90 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3595, pruned_loss=0.1092, over 5649049.82 frames. ], batch size: 307, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:28:02,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.018e+02 1.146e+03 1.567e+03 2.239e+03 7.546e+03, threshold=3.134e+03, percent-clipped=7.0
+2023-03-04 16:28:23,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1752, 1.2059, 3.4868, 3.0085], device='cuda:0'), covar=tensor([0.1555, 0.2542, 0.0440, 0.1364], device='cuda:0'), in_proj_covar=tensor([0.0622, 0.0567, 0.0822, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 16:28:26,709 INFO [train.py:968] (0/2) Epoch 9, batch 13150, giga_loss[loss=0.2984, simple_loss=0.3615, pruned_loss=0.1177, over 27502.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3543, pruned_loss=0.1057, over 5643062.34 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3564, pruned_loss=0.1014, over 5756936.03 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3553, pruned_loss=0.1068, over 5631904.92 frames. ], batch size: 472, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:29:17,194 INFO [train.py:968] (0/2) Epoch 9, batch 13200, giga_loss[loss=0.2734, simple_loss=0.3497, pruned_loss=0.09861, over 27906.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.355, pruned_loss=0.1062, over 5644325.49 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3566, pruned_loss=0.1016, over 5758077.58 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3556, pruned_loss=0.1069, over 5633118.71 frames. ], batch size: 412, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:29:43,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.550e+02 1.324e+03 1.728e+03 2.682e+03 6.418e+03, threshold=3.457e+03, percent-clipped=14.0
+2023-03-04 16:30:02,233 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-04 16:30:07,082 INFO [train.py:968] (0/2) Epoch 9, batch 13250, giga_loss[loss=0.2785, simple_loss=0.354, pruned_loss=0.1015, over 28526.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3547, pruned_loss=0.1058, over 5643214.76 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3567, pruned_loss=0.1018, over 5756887.58 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.355, pruned_loss=0.1063, over 5634083.11 frames. ], batch size: 336, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:30:17,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=377457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:30:28,208 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-04 16:30:53,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=377494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:30:55,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4725, 3.3384, 1.5500, 1.4529], device='cuda:0'), covar=tensor([0.0819, 0.0311, 0.0878, 0.1253], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0501, 0.0330, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 16:30:59,053 INFO [train.py:968] (0/2) Epoch 9, batch 13300, giga_loss[loss=0.2901, simple_loss=0.3588, pruned_loss=0.1107, over 27765.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3515, pruned_loss=0.1031, over 5639495.85 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3565, pruned_loss=0.1018, over 5750048.34 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3519, pruned_loss=0.1035, over 5635434.88 frames. ], batch size: 474, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:31:23,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.382e+02 1.181e+03 1.625e+03 2.199e+03 5.640e+03, threshold=3.251e+03, percent-clipped=5.0
+2023-03-04 16:31:29,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=377531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:31:46,068 INFO [train.py:968] (0/2) Epoch 9, batch 13350, libri_loss[loss=0.2749, simple_loss=0.3511, pruned_loss=0.09937, over 29333.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3482, pruned_loss=0.1005, over 5648278.79 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3558, pruned_loss=0.1016, over 5752888.61 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3488, pruned_loss=0.101, over 5638435.65 frames. ], batch size: 94, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:32:38,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3010, 1.6325, 1.5771, 1.5649], device='cuda:0'), covar=tensor([0.1318, 0.1117, 0.1233, 0.1124], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0722, 0.0647, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 16:32:41,318 INFO [train.py:968] (0/2) Epoch 9, batch 13400, giga_loss[loss=0.2851, simple_loss=0.353, pruned_loss=0.1086, over 28840.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3445, pruned_loss=0.09886, over 5646439.18 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3556, pruned_loss=0.1016, over 5750039.22 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3451, pruned_loss=0.09923, over 5640146.84 frames. ], batch size: 186, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:32:43,211 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=377600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:32:47,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:32:47,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=377603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:12,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.864e+02 1.332e+03 1.785e+03 2.507e+03 6.170e+03, threshold=3.571e+03, percent-clipped=13.0
+2023-03-04 16:33:18,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=377632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:24,630 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=377637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:26,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2694, 1.4188, 3.8347, 3.2057], device='cuda:0'), covar=tensor([0.1545, 0.2295, 0.0403, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0618, 0.0566, 0.0816, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 16:33:26,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=377640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:33,344 INFO [train.py:968] (0/2) Epoch 9, batch 13450, giga_loss[loss=0.2688, simple_loss=0.3299, pruned_loss=0.1038, over 26721.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3431, pruned_loss=0.0986, over 5643628.92 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.355, pruned_loss=0.1015, over 5743125.18 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3439, pruned_loss=0.09896, over 5642437.33 frames. ], batch size: 555, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:33:53,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:55,266 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=377669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:59,524 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=377674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:34:02,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=377677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:34:26,290 INFO [train.py:968] (0/2) Epoch 9, batch 13500, giga_loss[loss=0.2772, simple_loss=0.3521, pruned_loss=0.1011, over 28303.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3428, pruned_loss=0.09879, over 5639471.86 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3543, pruned_loss=0.1011, over 5746489.55 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.344, pruned_loss=0.09933, over 5633907.06 frames. ], batch size: 368, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:34:39,040 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=377706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:35:00,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.286e+02 1.355e+03 1.706e+03 2.257e+03 7.344e+03, threshold=3.412e+03, percent-clipped=8.0
+2023-03-04 16:35:23,322 INFO [train.py:968] (0/2) Epoch 9, batch 13550, libri_loss[loss=0.2081, simple_loss=0.2843, pruned_loss=0.06594, over 29641.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3444, pruned_loss=0.09838, over 5646967.94 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3535, pruned_loss=0.1009, over 5749772.87 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3458, pruned_loss=0.09902, over 5637183.05 frames. ], batch size: 73, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:35:33,661 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 16:35:46,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:36:20,381 INFO [train.py:968] (0/2) Epoch 9, batch 13600, giga_loss[loss=0.3109, simple_loss=0.3713, pruned_loss=0.1252, over 28658.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3466, pruned_loss=0.09861, over 5657468.78 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3534, pruned_loss=0.1007, over 5750284.61 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3476, pruned_loss=0.09919, over 5647171.93 frames. ], batch size: 242, lr: 3.64e-03, grad_scale: 8.0
+2023-03-04 16:36:54,686 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.685e+03 2.232e+03 3.171e+03 1.382e+04, threshold=4.463e+03, percent-clipped=20.0
+2023-03-04 16:36:56,528 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0496, 1.4195, 1.3082, 1.2342], device='cuda:0'), covar=tensor([0.1203, 0.1009, 0.1698, 0.1171], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0723, 0.0651, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 16:37:09,210 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-04 16:37:24,578 INFO [train.py:968] (0/2) Epoch 9, batch 13650, giga_loss[loss=0.269, simple_loss=0.3404, pruned_loss=0.09879, over 28964.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3467, pruned_loss=0.0984, over 5669538.64 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3531, pruned_loss=0.1006, over 5752396.54 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3477, pruned_loss=0.09898, over 5658369.50 frames. ], batch size: 145, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:37:43,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1895, 1.7574, 1.5076, 1.3533], device='cuda:0'), covar=tensor([0.0841, 0.0299, 0.0306, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0119, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0053, 0.0048, 0.0081], device='cuda:0')
+2023-03-04 16:38:26,395 INFO [train.py:968] (0/2) Epoch 9, batch 13700, giga_loss[loss=0.2451, simple_loss=0.3298, pruned_loss=0.08024, over 28920.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3434, pruned_loss=0.09629, over 5667906.54 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3527, pruned_loss=0.1004, over 5752994.56 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3445, pruned_loss=0.09688, over 5658218.99 frames. ], batch size: 213, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:39:02,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.103e+02 1.171e+03 1.454e+03 2.153e+03 6.553e+03, threshold=2.908e+03, percent-clipped=1.0
+2023-03-04 16:39:26,968 INFO [train.py:968] (0/2) Epoch 9, batch 13750, giga_loss[loss=0.2387, simple_loss=0.3221, pruned_loss=0.07767, over 28845.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3431, pruned_loss=0.09485, over 5666466.76 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3526, pruned_loss=0.1004, over 5752845.84 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3439, pruned_loss=0.09524, over 5656916.00 frames. ], batch size: 186, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:39:27,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2772, 1.6278, 1.5934, 1.2034], device='cuda:0'), covar=tensor([0.1721, 0.2331, 0.1386, 0.1574], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0696, 0.0832, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 16:40:02,892 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=377978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:40:23,997 INFO [train.py:968] (0/2) Epoch 9, batch 13800, giga_loss[loss=0.2578, simple_loss=0.3356, pruned_loss=0.09005, over 28892.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3414, pruned_loss=0.09428, over 5657900.30 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3523, pruned_loss=0.1003, over 5746287.63 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.342, pruned_loss=0.09447, over 5653372.17 frames. ], batch size: 164, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:40:24,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:40:25,835 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-378000.pt
+2023-03-04 16:40:59,653 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.554e+02 1.501e+03 2.056e+03 3.200e+03 1.135e+04, threshold=4.111e+03, percent-clipped=32.0
+2023-03-04 16:41:17,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=378041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:41:24,516 INFO [train.py:968] (0/2) Epoch 9, batch 13850, giga_loss[loss=0.2541, simple_loss=0.3345, pruned_loss=0.08688, over 29002.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.34, pruned_loss=0.09492, over 5656201.56 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3525, pruned_loss=0.1007, over 5738948.92 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3402, pruned_loss=0.09467, over 5657915.88 frames. ], batch size: 213, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:41:30,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0960, 3.9320, 3.7103, 1.7190], device='cuda:0'), covar=tensor([0.0490, 0.0653, 0.0682, 0.2345], device='cuda:0'), in_proj_covar=tensor([0.0973, 0.0907, 0.0802, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 16:42:01,033 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 16:42:20,691 INFO [train.py:968] (0/2) Epoch 9, batch 13900, giga_loss[loss=0.2625, simple_loss=0.3371, pruned_loss=0.09397, over 28097.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3387, pruned_loss=0.09445, over 5664932.83 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3526, pruned_loss=0.1007, over 5741634.04 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3384, pruned_loss=0.09403, over 5661794.94 frames. ], batch size: 412, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:42:46,459 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=378121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:42:50,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=378124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:42:53,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.696e+02 1.421e+03 1.907e+03 3.131e+03 1.330e+04, threshold=3.813e+03, percent-clipped=14.0
+2023-03-04 16:43:15,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=378145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:43:18,403 INFO [train.py:968] (0/2) Epoch 9, batch 13950, giga_loss[loss=0.2767, simple_loss=0.3586, pruned_loss=0.0974, over 28712.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3399, pruned_loss=0.09493, over 5648284.98 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3526, pruned_loss=0.1008, over 5734542.35 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3395, pruned_loss=0.09445, over 5650353.99 frames. ], batch size: 307, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:43:23,949 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=378153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:43:32,253 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-04 16:44:00,409 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=378184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:44:03,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=378187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:44:13,773 INFO [train.py:968] (0/2) Epoch 9, batch 14000, giga_loss[loss=0.2923, simple_loss=0.3747, pruned_loss=0.1049, over 28677.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3419, pruned_loss=0.09525, over 5657046.26 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.352, pruned_loss=0.1006, over 5740353.60 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3417, pruned_loss=0.09483, over 5650976.42 frames. ], batch size: 262, lr: 3.64e-03, grad_scale: 8.0
+2023-03-04 16:44:36,633 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=378216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:44:50,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.328e+02 1.319e+03 1.646e+03 2.286e+03 5.203e+03, threshold=3.291e+03, percent-clipped=5.0
+2023-03-04 16:45:15,678 INFO [train.py:968] (0/2) Epoch 9, batch 14050, giga_loss[loss=0.2319, simple_loss=0.3173, pruned_loss=0.07326, over 28725.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.338, pruned_loss=0.09269, over 5659250.39 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3516, pruned_loss=0.1007, over 5735913.83 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3378, pruned_loss=0.09212, over 5656307.35 frames. ], batch size: 262, lr: 3.64e-03, grad_scale: 8.0
+2023-03-04 16:46:03,522 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=378288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:46:07,549 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=378291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:46:13,472 INFO [train.py:968] (0/2) Epoch 9, batch 14100, giga_loss[loss=0.3003, simple_loss=0.3693, pruned_loss=0.1157, over 28777.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3379, pruned_loss=0.09278, over 5671037.73 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3509, pruned_loss=0.1003, over 5736453.03 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.338, pruned_loss=0.09236, over 5666348.50 frames. ], batch size: 243, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:46:46,830 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=378320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:46:54,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.567e+02 1.401e+03 1.988e+03 2.914e+03 6.412e+03, threshold=3.975e+03, percent-clipped=17.0
+2023-03-04 16:47:24,069 INFO [train.py:968] (0/2) Epoch 9, batch 14150, giga_loss[loss=0.2572, simple_loss=0.3499, pruned_loss=0.08225, over 28503.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3404, pruned_loss=0.09348, over 5667562.12 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3504, pruned_loss=0.1002, over 5730007.21 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3407, pruned_loss=0.09317, over 5668189.39 frames. ], batch size: 336, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:47:55,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=378373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:48:06,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6635, 3.9640, 1.6434, 1.6733], device='cuda:0'), covar=tensor([0.0821, 0.0174, 0.0852, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0493, 0.0328, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0028, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 16:48:24,792 INFO [train.py:968] (0/2) Epoch 9, batch 14200, giga_loss[loss=0.2658, simple_loss=0.3583, pruned_loss=0.08671, over 29031.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3437, pruned_loss=0.0927, over 5667365.12 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.35, pruned_loss=0.09998, over 5731705.05 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3442, pruned_loss=0.09251, over 5664741.28 frames. ], batch size: 155, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:48:45,425 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8081, 3.6347, 3.4467, 1.5776], device='cuda:0'), covar=tensor([0.0700, 0.0788, 0.0776, 0.2456], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.0911, 0.0802, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 16:48:54,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-04 16:48:55,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.570e+02 1.246e+03 1.530e+03 2.121e+03 4.869e+03, threshold=3.060e+03, percent-clipped=3.0
+2023-03-04 16:49:15,063 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3489, 1.5598, 1.4860, 1.4751], device='cuda:0'), covar=tensor([0.1267, 0.1670, 0.1803, 0.1522], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0719, 0.0646, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 16:49:23,813 INFO [train.py:968] (0/2) Epoch 9, batch 14250, giga_loss[loss=0.2721, simple_loss=0.3557, pruned_loss=0.09422, over 28134.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3446, pruned_loss=0.09229, over 5667689.91 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3496, pruned_loss=0.09976, over 5737178.55 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3453, pruned_loss=0.09213, over 5659317.53 frames. ], batch size: 412, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:49:50,484 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-04 16:50:21,750 INFO [train.py:968] (0/2) Epoch 9, batch 14300, libri_loss[loss=0.272, simple_loss=0.3511, pruned_loss=0.0965, over 29383.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3449, pruned_loss=0.09156, over 5665971.15 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3494, pruned_loss=0.09975, over 5729811.67 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3455, pruned_loss=0.09133, over 5665555.66 frames. ], batch size: 92, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:50:36,012 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 16:50:44,923 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=378516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:50:48,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=378519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:51:01,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2019, 1.6124, 1.5440, 1.1567], device='cuda:0'), covar=tensor([0.1270, 0.1888, 0.1047, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0697, 0.0836, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 16:51:01,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.525e+02 1.251e+03 1.679e+03 2.219e+03 7.242e+03, threshold=3.358e+03, percent-clipped=13.0
+2023-03-04 16:51:25,621 INFO [train.py:968] (0/2) Epoch 9, batch 14350, giga_loss[loss=0.2847, simple_loss=0.3587, pruned_loss=0.1053, over 27734.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3456, pruned_loss=0.09258, over 5665774.19 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3493, pruned_loss=0.09968, over 5731308.23 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.346, pruned_loss=0.09239, over 5663585.55 frames. ], batch size: 474, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:51:26,017 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=378548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:52:00,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9438, 2.9564, 1.9830, 1.2029], device='cuda:0'), covar=tensor([0.4442, 0.1882, 0.2619, 0.3909], device='cuda:0'), in_proj_covar=tensor([0.1509, 0.1426, 0.1467, 0.1232], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 16:52:28,667 INFO [train.py:968] (0/2) Epoch 9, batch 14400, libri_loss[loss=0.207, simple_loss=0.282, pruned_loss=0.06604, over 29501.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3447, pruned_loss=0.09352, over 5678345.54 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3489, pruned_loss=0.09959, over 5735796.37 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3454, pruned_loss=0.09331, over 5671124.99 frames. ], batch size: 70, lr: 3.64e-03, grad_scale: 8.0
+2023-03-04 16:52:46,985 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=378610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:52:53,962 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=378616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:53:09,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.707e+02 1.210e+03 1.636e+03 2.211e+03 7.668e+03, threshold=3.272e+03, percent-clipped=10.0
+2023-03-04 16:53:40,570 INFO [train.py:968] (0/2) Epoch 9, batch 14450, giga_loss[loss=0.2435, simple_loss=0.3247, pruned_loss=0.08117, over 28622.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.346, pruned_loss=0.09515, over 5688867.80 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3485, pruned_loss=0.09935, over 5740165.11 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3468, pruned_loss=0.09507, over 5678338.67 frames. ], batch size: 307, lr: 3.64e-03, grad_scale: 8.0
+2023-03-04 16:54:26,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7490, 3.5785, 3.3650, 1.5673], device='cuda:0'), covar=tensor([0.0656, 0.0752, 0.0774, 0.2307], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.0910, 0.0805, 0.0633], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 16:54:44,335 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=378688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:54:55,357 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-04 16:54:58,123 INFO [train.py:968] (0/2) Epoch 9, batch 14500, giga_loss[loss=0.2357, simple_loss=0.3212, pruned_loss=0.07511, over 28723.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3414, pruned_loss=0.0932, over 5673699.74 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3481, pruned_loss=0.0993, over 5731082.13 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3423, pruned_loss=0.09309, over 5671435.38 frames. ], batch size: 262, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:55:43,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.851e+02 1.337e+03 1.825e+03 2.536e+03 6.815e+03, threshold=3.651e+03, percent-clipped=12.0
+2023-03-04 16:56:04,017 INFO [train.py:968] (0/2) Epoch 9, batch 14550, libri_loss[loss=0.2702, simple_loss=0.3539, pruned_loss=0.09327, over 29121.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.34, pruned_loss=0.09218, over 5673143.04 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3481, pruned_loss=0.09926, over 5726784.84 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3405, pruned_loss=0.09196, over 5672943.60 frames. ], batch size: 97, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:56:08,825 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 16:56:33,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6108, 1.6425, 1.1758, 1.2107], device='cuda:0'), covar=tensor([0.0681, 0.0511, 0.0957, 0.1039], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0438, 0.0495, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 16:56:41,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1435, 1.3233, 3.4583, 3.0851], device='cuda:0'), covar=tensor([0.1496, 0.2318, 0.0407, 0.1455], device='cuda:0'), in_proj_covar=tensor([0.0613, 0.0568, 0.0811, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 16:56:53,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 16:57:05,081 INFO [train.py:968] (0/2) Epoch 9, batch 14600, giga_loss[loss=0.2747, simple_loss=0.3324, pruned_loss=0.1084, over 24263.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3376, pruned_loss=0.0916, over 5672189.44 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3478, pruned_loss=0.09914, over 5726933.11 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3381, pruned_loss=0.09132, over 5670605.55 frames. ], batch size: 705, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:57:41,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5353, 2.2106, 1.5870, 0.5856], device='cuda:0'), covar=tensor([0.2844, 0.1518, 0.2301, 0.3051], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1415, 0.1457, 0.1221], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 16:57:41,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.021e+02 1.272e+03 1.702e+03 2.585e+03 6.783e+03, threshold=3.403e+03, percent-clipped=11.0
+2023-03-04 16:58:00,792 INFO [train.py:968] (0/2) Epoch 9, batch 14650, libri_loss[loss=0.2237, simple_loss=0.2929, pruned_loss=0.07725, over 29356.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3406, pruned_loss=0.09315, over 5678049.82 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3469, pruned_loss=0.09867, over 5732101.72 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3415, pruned_loss=0.0931, over 5669888.10 frames. ], batch size: 67, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:58:07,115 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=378851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:58:13,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=378855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:59:05,760 INFO [train.py:968] (0/2) Epoch 9, batch 14700, giga_loss[loss=0.2579, simple_loss=0.3407, pruned_loss=0.08754, over 28670.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3425, pruned_loss=0.09455, over 5681786.43 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.347, pruned_loss=0.09863, over 5733010.27 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3431, pruned_loss=0.09451, over 5674086.98 frames. ], batch size: 307, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:59:40,417 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8114, 4.6508, 4.3715, 1.9523], device='cuda:0'), covar=tensor([0.0451, 0.0611, 0.0769, 0.1992], device='cuda:0'), in_proj_covar=tensor([0.0957, 0.0895, 0.0790, 0.0623], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 16:59:45,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.299e+02 1.403e+03 1.810e+03 2.654e+03 1.107e+04, threshold=3.620e+03, percent-clipped=15.0
+2023-03-04 17:00:05,562 INFO [train.py:968] (0/2) Epoch 9, batch 14750, giga_loss[loss=0.2313, simple_loss=0.3164, pruned_loss=0.07306, over 28948.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3403, pruned_loss=0.0944, over 5681920.40 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3463, pruned_loss=0.0983, over 5736694.22 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3412, pruned_loss=0.09459, over 5671474.94 frames. ], batch size: 100, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 17:00:53,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=378985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:00:56,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1220, 1.0927, 3.7495, 3.1336], device='cuda:0'), covar=tensor([0.1670, 0.2546, 0.0496, 0.1005], device='cuda:0'), in_proj_covar=tensor([0.0620, 0.0577, 0.0821, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 17:01:00,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=378991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:01:06,717 INFO [train.py:968] (0/2) Epoch 9, batch 14800, libri_loss[loss=0.2137, simple_loss=0.2893, pruned_loss=0.0691, over 28551.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3404, pruned_loss=0.09478, over 5687550.03 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3457, pruned_loss=0.09801, over 5738695.82 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3417, pruned_loss=0.09511, over 5675885.39 frames. ], batch size: 63, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 17:01:46,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.711e+02 1.341e+03 1.735e+03 2.352e+03 4.692e+03, threshold=3.471e+03, percent-clipped=7.0
+2023-03-04 17:02:10,676 INFO [train.py:968] (0/2) Epoch 9, batch 14850, giga_loss[loss=0.3245, simple_loss=0.3888, pruned_loss=0.1301, over 27596.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3424, pruned_loss=0.09595, over 5677356.24 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3458, pruned_loss=0.09829, over 5733416.46 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3432, pruned_loss=0.09591, over 5671205.94 frames. ], batch size: 472, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 17:02:28,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:03:15,846 INFO [train.py:968] (0/2) Epoch 9, batch 14900, giga_loss[loss=0.2976, simple_loss=0.3646, pruned_loss=0.1153, over 28112.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3447, pruned_loss=0.09645, over 5668176.19 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3453, pruned_loss=0.09807, over 5724362.53 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3456, pruned_loss=0.09656, over 5670349.23 frames. ], batch size: 412, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 17:04:04,420 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:04:06,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.252e+02 1.495e+03 2.303e+03 3.236e+03 8.232e+03, threshold=4.607e+03, percent-clipped=21.0
+2023-03-04 17:04:08,133 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:04:12,094 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:04:12,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3142, 1.5572, 1.1810, 1.4925], device='cuda:0'), covar=tensor([0.0726, 0.0328, 0.0341, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0120, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0053, 0.0049, 0.0082], device='cuda:0')
+2023-03-04 17:04:16,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:04:37,786 INFO [train.py:968] (0/2) Epoch 9, batch 14950, giga_loss[loss=0.2504, simple_loss=0.3243, pruned_loss=0.08826, over 28965.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3428, pruned_loss=0.09501, over 5659124.17 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.345, pruned_loss=0.09788, over 5725417.72 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3438, pruned_loss=0.09523, over 5659778.83 frames. ], batch size: 213, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 17:04:51,429 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:04:51,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2128, 3.0571, 2.1421, 1.7842], device='cuda:0'), covar=tensor([0.1761, 0.0789, 0.1054, 0.1382], device='cuda:0'), in_proj_covar=tensor([0.1623, 0.1458, 0.1407, 0.1534], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 17:05:02,476 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:05:41,805 INFO [train.py:968] (0/2) Epoch 9, batch 15000, giga_loss[loss=0.2723, simple_loss=0.3386, pruned_loss=0.103, over 28908.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3395, pruned_loss=0.09448, over 5664653.56 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3447, pruned_loss=0.09772, over 5730086.03 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3405, pruned_loss=0.09475, over 5659559.57 frames. ], batch size: 213, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:05:41,809 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 17:05:50,586 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2103, simple_loss=0.3087, pruned_loss=0.05592, over 944034.00 frames. 
+2023-03-04 17:05:50,587 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 17:06:00,444 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:06:05,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:06:31,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:06:34,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.952e+02 1.318e+03 1.696e+03 2.523e+03 8.877e+03, threshold=3.393e+03, percent-clipped=6.0
+2023-03-04 17:06:34,814 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:06:44,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:06:52,568 INFO [train.py:968] (0/2) Epoch 9, batch 15050, giga_loss[loss=0.2661, simple_loss=0.3376, pruned_loss=0.0973, over 29056.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3336, pruned_loss=0.09192, over 5665324.88 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3444, pruned_loss=0.09751, over 5731470.98 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3345, pruned_loss=0.09222, over 5658598.91 frames. ], batch size: 120, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:07:07,338 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:07:38,290 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:07:54,773 INFO [train.py:968] (0/2) Epoch 9, batch 15100, giga_loss[loss=0.2933, simple_loss=0.3592, pruned_loss=0.1137, over 28337.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3345, pruned_loss=0.09266, over 5662701.61 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3447, pruned_loss=0.09776, over 5732235.88 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3349, pruned_loss=0.09261, over 5656148.71 frames. ], batch size: 368, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:08:00,340 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.43 vs. limit=5.0
+2023-03-04 17:08:13,552 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:08:26,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.428e+02 1.454e+03 1.893e+03 2.845e+03 1.076e+04, threshold=3.786e+03, percent-clipped=18.0
+2023-03-04 17:08:46,983 INFO [train.py:968] (0/2) Epoch 9, batch 15150, giga_loss[loss=0.2942, simple_loss=0.3646, pruned_loss=0.1119, over 28615.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3366, pruned_loss=0.09404, over 5662827.77 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3449, pruned_loss=0.09792, over 5730712.51 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3365, pruned_loss=0.09377, over 5656725.98 frames. ], batch size: 336, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:09:15,256 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:09:18,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:09:19,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:09:24,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:09:44,220 INFO [train.py:968] (0/2) Epoch 9, batch 15200, giga_loss[loss=0.2941, simple_loss=0.3493, pruned_loss=0.1194, over 26911.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3337, pruned_loss=0.09154, over 5667814.27 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3445, pruned_loss=0.0977, over 5735877.77 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3335, pruned_loss=0.09133, over 5655730.53 frames. ], batch size: 555, lr: 3.63e-03, grad_scale: 8.0
+2023-03-04 17:09:44,703 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 17:09:50,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:09:56,264 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:10:21,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.794e+02 1.293e+03 1.667e+03 2.123e+03 3.795e+03, threshold=3.333e+03, percent-clipped=1.0
+2023-03-04 17:10:36,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1463, 3.9960, 3.7193, 1.9081], device='cuda:0'), covar=tensor([0.0558, 0.0716, 0.0830, 0.2025], device='cuda:0'), in_proj_covar=tensor([0.0962, 0.0900, 0.0794, 0.0623], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 17:10:38,671 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5797, 1.7437, 1.3801, 1.9882], device='cuda:0'), covar=tensor([0.2375, 0.2246, 0.2490, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.1229, 0.0909, 0.1093, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 17:10:41,253 INFO [train.py:968] (0/2) Epoch 9, batch 15250, giga_loss[loss=0.2423, simple_loss=0.3315, pruned_loss=0.07649, over 28167.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3327, pruned_loss=0.09024, over 5671411.88 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.344, pruned_loss=0.09731, over 5741705.42 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3327, pruned_loss=0.09019, over 5654090.18 frames. ], batch size: 412, lr: 3.63e-03, grad_scale: 8.0
+2023-03-04 17:11:10,446 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4807, 1.2285, 4.6093, 3.4382], device='cuda:0'), covar=tensor([0.1477, 0.2383, 0.0358, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0605, 0.0563, 0.0802, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 17:11:52,430 INFO [train.py:968] (0/2) Epoch 9, batch 15300, giga_loss[loss=0.3144, simple_loss=0.3619, pruned_loss=0.1335, over 26876.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3319, pruned_loss=0.09069, over 5662763.95 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3438, pruned_loss=0.09728, over 5741525.16 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3319, pruned_loss=0.09061, over 5648783.19 frames. ], batch size: 555, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:12:03,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5328, 2.3414, 1.8853, 2.0953], device='cuda:0'), covar=tensor([0.0635, 0.0554, 0.0802, 0.0907], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0435, 0.0495, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 17:12:31,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.370e+02 1.370e+03 1.786e+03 2.522e+03 9.506e+03, threshold=3.572e+03, percent-clipped=14.0
+2023-03-04 17:12:51,954 INFO [train.py:968] (0/2) Epoch 9, batch 15350, libri_loss[loss=0.2756, simple_loss=0.3527, pruned_loss=0.09927, over 29754.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.332, pruned_loss=0.09009, over 5659617.97 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3438, pruned_loss=0.09733, over 5738507.87 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3316, pruned_loss=0.08976, over 5648404.98 frames. ], batch size: 87, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:13:53,090 INFO [train.py:968] (0/2) Epoch 9, batch 15400, libri_loss[loss=0.3239, simple_loss=0.3894, pruned_loss=0.1292, over 29525.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3327, pruned_loss=0.09027, over 5664637.31 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3441, pruned_loss=0.09744, over 5743277.37 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3316, pruned_loss=0.0896, over 5648107.56 frames. ], batch size: 89, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:14:32,022 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 17:14:32,130 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.275e+02 1.270e+03 1.606e+03 2.438e+03 5.547e+03, threshold=3.212e+03, percent-clipped=6.0
+2023-03-04 17:14:35,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:14:54,684 INFO [train.py:968] (0/2) Epoch 9, batch 15450, giga_loss[loss=0.2203, simple_loss=0.3067, pruned_loss=0.06691, over 28857.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.334, pruned_loss=0.09209, over 5661451.02 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3439, pruned_loss=0.09727, over 5742850.92 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3333, pruned_loss=0.09163, over 5647399.12 frames. ], batch size: 136, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:15:05,121 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:15:19,583 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379670.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 17:15:48,278 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:15:53,261 INFO [train.py:968] (0/2) Epoch 9, batch 15500, giga_loss[loss=0.3185, simple_loss=0.3918, pruned_loss=0.1226, over 28727.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3324, pruned_loss=0.09035, over 5666187.22 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3436, pruned_loss=0.0971, over 5741952.72 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3318, pruned_loss=0.09003, over 5654071.50 frames. ], batch size: 263, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:15:56,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7376, 1.7193, 1.3266, 1.3609], device='cuda:0'), covar=tensor([0.0728, 0.0546, 0.0890, 0.0989], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0439, 0.0498, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 17:16:22,322 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:16:30,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.630e+02 1.299e+03 1.674e+03 2.123e+03 5.156e+03, threshold=3.349e+03, percent-clipped=6.0
+2023-03-04 17:16:52,096 INFO [train.py:968] (0/2) Epoch 9, batch 15550, giga_loss[loss=0.3081, simple_loss=0.3771, pruned_loss=0.1196, over 28895.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3352, pruned_loss=0.09029, over 5675918.47 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3433, pruned_loss=0.09688, over 5745774.16 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3348, pruned_loss=0.09005, over 5661357.78 frames. ], batch size: 227, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:17:25,582 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:17:28,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:17:49,000 INFO [train.py:968] (0/2) Epoch 9, batch 15600, giga_loss[loss=0.2798, simple_loss=0.3491, pruned_loss=0.1052, over 27699.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3384, pruned_loss=0.09206, over 5663288.50 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.343, pruned_loss=0.09677, over 5741770.42 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3382, pruned_loss=0.09182, over 5653788.20 frames. ], batch size: 472, lr: 3.63e-03, grad_scale: 8.0
+2023-03-04 17:17:51,190 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:17:55,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:18:01,681 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:18:27,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.482e+02 1.432e+03 2.048e+03 3.333e+03 8.326e+03, threshold=4.096e+03, percent-clipped=23.0
+2023-03-04 17:18:29,218 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:18:30,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:18:33,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:18:47,687 INFO [train.py:968] (0/2) Epoch 9, batch 15650, giga_loss[loss=0.2299, simple_loss=0.3162, pruned_loss=0.07176, over 28985.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3395, pruned_loss=0.09232, over 5669404.29 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3428, pruned_loss=0.09675, over 5745030.34 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3394, pruned_loss=0.09203, over 5657160.18 frames. ], batch size: 136, lr: 3.63e-03, grad_scale: 8.0
+2023-03-04 17:19:07,984 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:19:31,324 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:19:42,259 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:19:43,608 INFO [train.py:968] (0/2) Epoch 9, batch 15700, giga_loss[loss=0.3099, simple_loss=0.3791, pruned_loss=0.1204, over 28854.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3392, pruned_loss=0.09229, over 5684962.84 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3424, pruned_loss=0.09654, over 5749004.84 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3394, pruned_loss=0.09212, over 5669161.34 frames. ], batch size: 284, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:20:22,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.514e+02 1.236e+03 1.693e+03 2.368e+03 5.422e+03, threshold=3.386e+03, percent-clipped=3.0
+2023-03-04 17:20:39,174 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-04 17:20:40,377 INFO [train.py:968] (0/2) Epoch 9, batch 15750, giga_loss[loss=0.2822, simple_loss=0.3595, pruned_loss=0.1024, over 28682.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3358, pruned_loss=0.08995, over 5694211.73 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.342, pruned_loss=0.09648, over 5752229.64 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3362, pruned_loss=0.08972, over 5677051.29 frames. ], batch size: 307, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:21:44,285 INFO [train.py:968] (0/2) Epoch 9, batch 15800, giga_loss[loss=0.2818, simple_loss=0.3522, pruned_loss=0.1057, over 28922.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3345, pruned_loss=0.08899, over 5693195.48 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3417, pruned_loss=0.09631, over 5754551.57 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3348, pruned_loss=0.08879, over 5676160.15 frames. ], batch size: 227, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:21:46,345 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-380000.pt
+2023-03-04 17:22:00,141 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9881, 1.2386, 1.3015, 0.9743], device='cuda:0'), covar=tensor([0.1159, 0.1087, 0.1730, 0.1431], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0714, 0.0641, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 17:22:21,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.197e+02 1.356e+03 1.855e+03 2.728e+03 7.997e+03, threshold=3.710e+03, percent-clipped=15.0
+2023-03-04 17:22:37,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=380045.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 17:22:41,076 INFO [train.py:968] (0/2) Epoch 9, batch 15850, giga_loss[loss=0.267, simple_loss=0.3492, pruned_loss=0.0924, over 28926.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.333, pruned_loss=0.08923, over 5684067.43 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3416, pruned_loss=0.09639, over 5749253.58 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3332, pruned_loss=0.08881, over 5674486.78 frames. ], batch size: 284, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:23:38,875 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4587, 1.6885, 1.5478, 1.5175], device='cuda:0'), covar=tensor([0.1149, 0.1566, 0.1659, 0.1547], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0707, 0.0634, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 17:23:41,714 INFO [train.py:968] (0/2) Epoch 9, batch 15900, giga_loss[loss=0.2647, simple_loss=0.3511, pruned_loss=0.08917, over 28826.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3347, pruned_loss=0.0905, over 5678867.66 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3415, pruned_loss=0.09634, over 5750877.85 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3349, pruned_loss=0.09014, over 5669388.61 frames. ], batch size: 174, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:23:44,978 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=380099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:24:23,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.630e+02 1.228e+03 1.718e+03 2.161e+03 4.624e+03, threshold=3.435e+03, percent-clipped=3.0
+2023-03-04 17:24:46,406 INFO [train.py:968] (0/2) Epoch 9, batch 15950, giga_loss[loss=0.2369, simple_loss=0.301, pruned_loss=0.08637, over 24885.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3362, pruned_loss=0.09132, over 5682955.19 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3413, pruned_loss=0.09634, over 5754848.37 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3363, pruned_loss=0.09092, over 5670324.91 frames. ], batch size: 705, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:25:16,528 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5886, 2.1471, 1.4471, 0.8351], device='cuda:0'), covar=tensor([0.4534, 0.2660, 0.2702, 0.3981], device='cuda:0'), in_proj_covar=tensor([0.1489, 0.1428, 0.1456, 0.1223], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 17:25:32,295 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=380188.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 17:25:32,408 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-04 17:25:36,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=380191.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 17:25:44,049 INFO [train.py:968] (0/2) Epoch 9, batch 16000, giga_loss[loss=0.2752, simple_loss=0.3632, pruned_loss=0.09357, over 28899.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3379, pruned_loss=0.09282, over 5677399.37 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3415, pruned_loss=0.09656, over 5752886.01 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3376, pruned_loss=0.09213, over 5666886.34 frames. ], batch size: 136, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:26:08,033 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=380220.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 17:26:23,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.673e+02 1.317e+03 1.797e+03 2.337e+03 1.115e+04, threshold=3.595e+03, percent-clipped=11.0
+2023-03-04 17:26:33,951 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=380242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:26:39,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=380245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:26:40,984 INFO [train.py:968] (0/2) Epoch 9, batch 16050, giga_loss[loss=0.2672, simple_loss=0.3533, pruned_loss=0.09057, over 28982.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3423, pruned_loss=0.09525, over 5671519.88 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3419, pruned_loss=0.09674, over 5745650.94 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3418, pruned_loss=0.0945, over 5668685.78 frames. ], batch size: 155, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:26:55,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=380262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:27:06,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=380271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:27:10,461 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=380274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:27:37,048 INFO [train.py:968] (0/2) Epoch 9, batch 16100, giga_loss[loss=0.3055, simple_loss=0.3625, pruned_loss=0.1243, over 26849.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.344, pruned_loss=0.0951, over 5685113.44 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3418, pruned_loss=0.09676, over 5751019.49 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3437, pruned_loss=0.09442, over 5675923.26 frames. ], batch size: 555, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:28:21,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.146e+02 1.253e+03 1.728e+03 2.443e+03 9.083e+03, threshold=3.456e+03, percent-clipped=9.0
+2023-03-04 17:28:41,855 INFO [train.py:968] (0/2) Epoch 9, batch 16150, giga_loss[loss=0.2372, simple_loss=0.3193, pruned_loss=0.07754, over 29080.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3438, pruned_loss=0.09476, over 5686225.88 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3414, pruned_loss=0.09654, over 5752387.70 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3439, pruned_loss=0.0944, over 5676511.18 frames. ], batch size: 120, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:29:48,774 INFO [train.py:968] (0/2) Epoch 9, batch 16200, giga_loss[loss=0.2547, simple_loss=0.3312, pruned_loss=0.08912, over 28908.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3425, pruned_loss=0.09436, over 5685163.79 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3416, pruned_loss=0.09665, over 5746448.64 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3425, pruned_loss=0.09389, over 5681365.32 frames. ], batch size: 227, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:29:57,860 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=380405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:30:00,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=380408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:30:05,997 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=380414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:30:10,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=380417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:30:29,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.043e+02 1.420e+03 1.751e+03 2.686e+03 9.083e+03, threshold=3.502e+03, percent-clipped=12.0
+2023-03-04 17:30:38,173 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=380437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:30:47,342 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=380446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:30:50,329 INFO [train.py:968] (0/2) Epoch 9, batch 16250, giga_loss[loss=0.3256, simple_loss=0.375, pruned_loss=0.1381, over 26951.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3421, pruned_loss=0.09484, over 5682957.33 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3412, pruned_loss=0.09657, over 5748193.34 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3424, pruned_loss=0.09449, over 5677186.15 frames. ], batch size: 555, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:31:52,750 INFO [train.py:968] (0/2) Epoch 9, batch 16300, giga_loss[loss=0.2834, simple_loss=0.3546, pruned_loss=0.1062, over 28671.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3406, pruned_loss=0.09441, over 5670819.59 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3411, pruned_loss=0.09653, over 5750334.38 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.341, pruned_loss=0.09416, over 5663399.22 frames. ], batch size: 307, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:32:33,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7451, 1.7751, 1.2804, 1.4308], device='cuda:0'), covar=tensor([0.0706, 0.0562, 0.1022, 0.0953], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0435, 0.0494, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 17:32:34,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.196e+02 1.220e+03 1.548e+03 2.050e+03 4.297e+03, threshold=3.095e+03, percent-clipped=4.0
+2023-03-04 17:32:46,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9756, 2.7271, 1.9426, 1.6449], device='cuda:0'), covar=tensor([0.2555, 0.1238, 0.1455, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1462, 0.1412, 0.1539], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 17:32:51,228 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.85 vs. limit=5.0
+2023-03-04 17:32:53,251 INFO [train.py:968] (0/2) Epoch 9, batch 16350, giga_loss[loss=0.1986, simple_loss=0.2763, pruned_loss=0.06048, over 28419.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3375, pruned_loss=0.09343, over 5678905.63 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3405, pruned_loss=0.09622, over 5756241.10 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3383, pruned_loss=0.09342, over 5665094.23 frames. ], batch size: 71, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:33:18,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5184, 1.7483, 1.8313, 1.3728], device='cuda:0'), covar=tensor([0.1635, 0.2074, 0.1319, 0.1608], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0686, 0.0826, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 17:33:25,153 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=380576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:33:47,019 INFO [train.py:968] (0/2) Epoch 9, batch 16400, giga_loss[loss=0.302, simple_loss=0.3573, pruned_loss=0.1233, over 26661.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3378, pruned_loss=0.09338, over 5674397.51 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3404, pruned_loss=0.09609, over 5746607.88 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3385, pruned_loss=0.09343, over 5668430.97 frames. ], batch size: 555, lr: 3.63e-03, grad_scale: 8.0
+2023-03-04 17:34:29,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.887e+02 1.367e+03 1.904e+03 3.127e+03 8.881e+03, threshold=3.808e+03, percent-clipped=25.0
+2023-03-04 17:34:45,292 INFO [train.py:968] (0/2) Epoch 9, batch 16450, giga_loss[loss=0.2629, simple_loss=0.3522, pruned_loss=0.08683, over 28425.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.337, pruned_loss=0.09168, over 5676065.15 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.34, pruned_loss=0.09585, over 5749554.63 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3379, pruned_loss=0.09186, over 5667363.01 frames. ], batch size: 368, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:35:02,247 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=380660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:35:45,217 INFO [train.py:968] (0/2) Epoch 9, batch 16500, giga_loss[loss=0.2571, simple_loss=0.3448, pruned_loss=0.08467, over 28704.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3384, pruned_loss=0.09167, over 5673258.71 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3401, pruned_loss=0.09607, over 5743918.71 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3389, pruned_loss=0.09155, over 5669203.42 frames. ], batch size: 242, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:36:24,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.145e+02 1.274e+03 1.655e+03 2.477e+03 9.082e+03, threshold=3.311e+03, percent-clipped=11.0
+2023-03-04 17:36:38,508 INFO [train.py:968] (0/2) Epoch 9, batch 16550, giga_loss[loss=0.2523, simple_loss=0.3431, pruned_loss=0.08073, over 28919.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3408, pruned_loss=0.09183, over 5669411.78 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3399, pruned_loss=0.09592, over 5745600.99 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3415, pruned_loss=0.09175, over 5662498.58 frames. ], batch size: 136, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:37:29,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6655, 1.1357, 2.8806, 2.7199], device='cuda:0'), covar=tensor([0.1688, 0.2236, 0.0527, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0615, 0.0570, 0.0807, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 17:37:34,031 INFO [train.py:968] (0/2) Epoch 9, batch 16600, giga_loss[loss=0.23, simple_loss=0.3149, pruned_loss=0.07255, over 28654.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3414, pruned_loss=0.09146, over 5684846.19 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3401, pruned_loss=0.09606, over 5748161.57 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3418, pruned_loss=0.09122, over 5676207.78 frames. ], batch size: 92, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:38:01,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3292, 1.5832, 1.2593, 1.2943], device='cuda:0'), covar=tensor([0.2234, 0.2117, 0.2338, 0.1981], device='cuda:0'), in_proj_covar=tensor([0.1232, 0.0916, 0.1097, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 17:38:20,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.098e+02 1.167e+03 1.549e+03 2.156e+03 3.981e+03, threshold=3.098e+03, percent-clipped=2.0
+2023-03-04 17:38:39,631 INFO [train.py:968] (0/2) Epoch 9, batch 16650, giga_loss[loss=0.3151, simple_loss=0.3908, pruned_loss=0.1197, over 28472.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3412, pruned_loss=0.09152, over 5676095.67 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.34, pruned_loss=0.09602, over 5740227.81 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3416, pruned_loss=0.09134, over 5674956.80 frames. ], batch size: 336, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:38:44,623 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=380850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:39:11,864 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=380872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:39:46,149 INFO [train.py:968] (0/2) Epoch 9, batch 16700, giga_loss[loss=0.2916, simple_loss=0.3634, pruned_loss=0.1099, over 28058.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.341, pruned_loss=0.09114, over 5672926.72 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3399, pruned_loss=0.09591, over 5740667.11 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3414, pruned_loss=0.09101, over 5670506.71 frames. ], batch size: 412, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:40:17,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=380921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:40:32,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.342e+02 1.368e+03 1.769e+03 2.438e+03 6.342e+03, threshold=3.538e+03, percent-clipped=8.0
+2023-03-04 17:40:50,686 INFO [train.py:968] (0/2) Epoch 9, batch 16750, giga_loss[loss=0.2761, simple_loss=0.3443, pruned_loss=0.104, over 26944.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.341, pruned_loss=0.0908, over 5667027.33 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3398, pruned_loss=0.09581, over 5735651.91 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3414, pruned_loss=0.09065, over 5668478.69 frames. ], batch size: 555, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:40:56,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=380951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:41:01,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2963, 5.1332, 4.9109, 2.0595], device='cuda:0'), covar=tensor([0.0383, 0.0454, 0.0626, 0.2136], device='cuda:0'), in_proj_covar=tensor([0.0948, 0.0888, 0.0785, 0.0617], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 17:41:57,468 INFO [train.py:968] (0/2) Epoch 9, batch 16800, giga_loss[loss=0.2386, simple_loss=0.327, pruned_loss=0.07513, over 28942.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3413, pruned_loss=0.09082, over 5676345.23 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3404, pruned_loss=0.09611, over 5742725.06 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3412, pruned_loss=0.0902, over 5668304.83 frames. ], batch size: 136, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:42:22,354 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-04 17:42:44,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.795e+02 1.500e+03 2.005e+03 2.873e+03 7.972e+03, threshold=4.010e+03, percent-clipped=17.0
+2023-03-04 17:42:45,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:43:03,059 INFO [train.py:968] (0/2) Epoch 9, batch 16850, giga_loss[loss=0.2741, simple_loss=0.3589, pruned_loss=0.09472, over 28655.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3446, pruned_loss=0.09202, over 5681507.23 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3402, pruned_loss=0.09592, over 5743655.42 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3447, pruned_loss=0.09162, over 5673244.36 frames. ], batch size: 242, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:44:06,981 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:44:11,894 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:44:12,222 INFO [train.py:968] (0/2) Epoch 9, batch 16900, giga_loss[loss=0.2685, simple_loss=0.3443, pruned_loss=0.09635, over 28784.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3436, pruned_loss=0.09109, over 5680104.24 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3405, pruned_loss=0.09612, over 5744234.48 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3434, pruned_loss=0.09053, over 5672331.71 frames. ], batch size: 243, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:44:51,849 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:45:03,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.892e+02 1.276e+03 1.535e+03 2.063e+03 4.859e+03, threshold=3.070e+03, percent-clipped=3.0
+2023-03-04 17:45:18,612 INFO [train.py:968] (0/2) Epoch 9, batch 16950, giga_loss[loss=0.2312, simple_loss=0.3131, pruned_loss=0.07464, over 29058.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3411, pruned_loss=0.0904, over 5687258.32 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3402, pruned_loss=0.09598, over 5738242.51 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3413, pruned_loss=0.08997, over 5684504.68 frames. ], batch size: 128, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:46:05,023 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:46:09,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:46:14,717 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=381185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:46:30,840 INFO [train.py:968] (0/2) Epoch 9, batch 17000, giga_loss[loss=0.2711, simple_loss=0.36, pruned_loss=0.09109, over 28856.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3392, pruned_loss=0.08914, over 5689023.48 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.34, pruned_loss=0.09581, over 5739678.21 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3396, pruned_loss=0.08891, over 5684995.84 frames. ], batch size: 227, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:46:50,519 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:47:07,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:47:19,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.897e+02 1.239e+03 1.622e+03 2.570e+03 5.641e+03, threshold=3.244e+03, percent-clipped=14.0
+2023-03-04 17:47:34,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:47:34,850 INFO [train.py:968] (0/2) Epoch 9, batch 17050, giga_loss[loss=0.278, simple_loss=0.3427, pruned_loss=0.1067, over 26825.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3384, pruned_loss=0.08821, over 5689534.90 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3404, pruned_loss=0.09607, over 5734269.36 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3382, pruned_loss=0.08759, over 5690570.47 frames. ], batch size: 555, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:48:19,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5893, 1.7304, 1.4943, 1.7200], device='cuda:0'), covar=tensor([0.2262, 0.1901, 0.2022, 0.1850], device='cuda:0'), in_proj_covar=tensor([0.1226, 0.0914, 0.1095, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 17:48:31,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:48:33,487 INFO [train.py:968] (0/2) Epoch 9, batch 17100, giga_loss[loss=0.254, simple_loss=0.3248, pruned_loss=0.09164, over 24532.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3398, pruned_loss=0.08984, over 5675517.22 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3402, pruned_loss=0.09607, over 5728146.34 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3399, pruned_loss=0.08925, over 5680425.41 frames. ], batch size: 705, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:49:16,206 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.800e+02 1.432e+03 1.859e+03 2.550e+03 6.800e+03, threshold=3.719e+03, percent-clipped=11.0
+2023-03-04 17:49:32,147 INFO [train.py:968] (0/2) Epoch 9, batch 17150, giga_loss[loss=0.2744, simple_loss=0.3591, pruned_loss=0.09489, over 28808.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3433, pruned_loss=0.09215, over 5678939.31 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3401, pruned_loss=0.09607, over 5729406.50 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3435, pruned_loss=0.09162, over 5681207.66 frames. ], batch size: 243, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:49:55,061 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:49:59,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:50:17,039 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:50:21,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:50:26,034 INFO [train.py:968] (0/2) Epoch 9, batch 17200, giga_loss[loss=0.2809, simple_loss=0.3602, pruned_loss=0.1009, over 28462.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3437, pruned_loss=0.09296, over 5667357.19 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.34, pruned_loss=0.09597, over 5720401.08 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.344, pruned_loss=0.09256, over 5676466.59 frames. ], batch size: 336, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 17:50:29,449 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:50:56,202 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=381422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:50:56,216 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:51:05,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.297e+02 1.609e+03 2.040e+03 3.343e+03 9.836e+03, threshold=4.081e+03, percent-clipped=21.0
+2023-03-04 17:51:10,725 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:51:14,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:51:21,871 INFO [train.py:968] (0/2) Epoch 9, batch 17250, giga_loss[loss=0.3521, simple_loss=0.4016, pruned_loss=0.1513, over 28958.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3409, pruned_loss=0.09298, over 5671574.40 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3396, pruned_loss=0.09577, over 5721616.87 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3417, pruned_loss=0.09278, over 5676327.34 frames. ], batch size: 285, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:51:47,388 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-04 17:51:48,473 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:52:17,555 INFO [train.py:968] (0/2) Epoch 9, batch 17300, giga_loss[loss=0.2467, simple_loss=0.3206, pruned_loss=0.08645, over 28956.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3407, pruned_loss=0.0933, over 5676929.00 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3397, pruned_loss=0.09593, over 5723505.27 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3411, pruned_loss=0.09295, over 5678357.27 frames. ], batch size: 199, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:52:35,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=381512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:52:57,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.083e+02 1.401e+03 1.649e+03 2.363e+03 6.289e+03, threshold=3.299e+03, percent-clipped=1.0
+2023-03-04 17:53:11,356 INFO [train.py:968] (0/2) Epoch 9, batch 17350, giga_loss[loss=0.3722, simple_loss=0.41, pruned_loss=0.1672, over 26730.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3473, pruned_loss=0.09785, over 5674577.47 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3395, pruned_loss=0.09566, over 5726054.00 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.348, pruned_loss=0.09781, over 5672387.33 frames. ], batch size: 555, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:53:23,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:53:53,461 INFO [train.py:968] (0/2) Epoch 9, batch 17400, giga_loss[loss=0.296, simple_loss=0.379, pruned_loss=0.1064, over 28543.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3557, pruned_loss=0.1025, over 5676643.48 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3398, pruned_loss=0.0958, over 5718431.79 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3562, pruned_loss=0.1025, over 5681040.87 frames. ], batch size: 65, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:54:19,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3758, 1.5543, 1.3631, 1.5080], device='cuda:0'), covar=tensor([0.0773, 0.0310, 0.0322, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0119, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0053, 0.0049, 0.0082], device='cuda:0')
+2023-03-04 17:54:23,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.272e+02 1.182e+03 1.576e+03 2.204e+03 7.986e+03, threshold=3.152e+03, percent-clipped=5.0
+2023-03-04 17:54:27,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4477, 1.5783, 1.4537, 1.5320], device='cuda:0'), covar=tensor([0.1180, 0.1550, 0.1797, 0.1417], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0711, 0.0641, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 17:54:33,410 INFO [train.py:968] (0/2) Epoch 9, batch 17450, giga_loss[loss=0.3393, simple_loss=0.387, pruned_loss=0.1458, over 23697.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3588, pruned_loss=0.1042, over 5686168.23 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3395, pruned_loss=0.09548, over 5723940.12 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.36, pruned_loss=0.1047, over 5683734.06 frames. ], batch size: 705, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:55:15,550 INFO [train.py:968] (0/2) Epoch 9, batch 17500, giga_loss[loss=0.2708, simple_loss=0.3448, pruned_loss=0.09844, over 29027.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.355, pruned_loss=0.1032, over 5694970.23 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3393, pruned_loss=0.09535, over 5729876.19 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3568, pruned_loss=0.1039, over 5686328.75 frames. ], batch size: 155, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:55:20,408 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:55:21,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8011, 2.6646, 2.6760, 2.2985], device='cuda:0'), covar=tensor([0.1139, 0.1893, 0.1418, 0.1585], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0712, 0.0642, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 17:55:22,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:55:40,712 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-04 17:55:46,588 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:55:46,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.603e+02 1.075e+03 1.345e+03 1.794e+03 5.158e+03, threshold=2.689e+03, percent-clipped=7.0
+2023-03-04 17:55:56,716 INFO [train.py:968] (0/2) Epoch 9, batch 17550, giga_loss[loss=0.2721, simple_loss=0.3362, pruned_loss=0.104, over 28626.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3472, pruned_loss=0.09997, over 5686803.14 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3391, pruned_loss=0.09521, over 5729736.33 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.349, pruned_loss=0.1009, over 5679085.18 frames. ], batch size: 307, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:56:38,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:56:39,099 INFO [train.py:968] (0/2) Epoch 9, batch 17600, giga_loss[loss=0.2424, simple_loss=0.3118, pruned_loss=0.08647, over 29014.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3411, pruned_loss=0.09768, over 5684645.56 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3395, pruned_loss=0.0955, over 5726542.68 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3424, pruned_loss=0.09827, over 5679166.92 frames. ], batch size: 136, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 17:56:40,246 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 17:57:09,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.637e+02 1.104e+03 1.379e+03 1.938e+03 6.945e+03, threshold=2.759e+03, percent-clipped=10.0
+2023-03-04 17:57:21,071 INFO [train.py:968] (0/2) Epoch 9, batch 17650, giga_loss[loss=0.2316, simple_loss=0.3001, pruned_loss=0.08152, over 28652.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3346, pruned_loss=0.09496, over 5691442.95 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.34, pruned_loss=0.09572, over 5731733.52 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3351, pruned_loss=0.09526, over 5680820.64 frames. ], batch size: 92, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 17:57:52,462 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=381887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:57:52,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:58:00,480 INFO [train.py:968] (0/2) Epoch 9, batch 17700, giga_loss[loss=0.2609, simple_loss=0.3315, pruned_loss=0.09511, over 28022.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3287, pruned_loss=0.09227, over 5696837.75 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3401, pruned_loss=0.09561, over 5735926.71 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3287, pruned_loss=0.09253, over 5682855.70 frames. ], batch size: 77, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 17:58:05,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5442, 1.6434, 1.5843, 1.3381], device='cuda:0'), covar=tensor([0.2184, 0.1678, 0.1180, 0.1696], device='cuda:0'), in_proj_covar=tensor([0.1607, 0.1448, 0.1392, 0.1519], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 17:58:31,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.842e+02 1.006e+03 1.480e+03 2.350e+03 7.945e+03, threshold=2.960e+03, percent-clipped=14.0
+2023-03-04 17:58:34,028 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:58:35,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:58:39,001 INFO [train.py:968] (0/2) Epoch 9, batch 17750, giga_loss[loss=0.2875, simple_loss=0.3495, pruned_loss=0.1127, over 29084.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3246, pruned_loss=0.09058, over 5693242.50 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3403, pruned_loss=0.09567, over 5729640.26 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3242, pruned_loss=0.09066, over 5686433.50 frames. ], batch size: 155, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:58:57,162 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:59:16,800 INFO [train.py:968] (0/2) Epoch 9, batch 17800, libri_loss[loss=0.2336, simple_loss=0.3095, pruned_loss=0.07887, over 28481.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3226, pruned_loss=0.08958, over 5692811.44 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3404, pruned_loss=0.0957, over 5725232.90 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3216, pruned_loss=0.0894, over 5689917.07 frames. ], batch size: 63, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:59:19,150 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-382000.pt
+2023-03-04 17:59:34,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:59:43,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=382030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:59:47,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=382033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:59:51,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.042e+02 9.553e+02 1.253e+03 1.744e+03 4.749e+03, threshold=2.506e+03, percent-clipped=2.0
+2023-03-04 18:00:02,582 INFO [train.py:968] (0/2) Epoch 9, batch 17850, giga_loss[loss=0.2287, simple_loss=0.2973, pruned_loss=0.08002, over 28908.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3181, pruned_loss=0.08737, over 5691201.07 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3403, pruned_loss=0.09563, over 5728757.85 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.317, pruned_loss=0.08716, over 5685326.28 frames. ], batch size: 186, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:00:14,338 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=382062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:00:33,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5966, 1.8712, 1.8909, 1.4313], device='cuda:0'), covar=tensor([0.1475, 0.1919, 0.1201, 0.1359], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0699, 0.0844, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-04 18:00:44,089 INFO [train.py:968] (0/2) Epoch 9, batch 17900, giga_loss[loss=0.2319, simple_loss=0.2966, pruned_loss=0.08358, over 28763.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3149, pruned_loss=0.08604, over 5694498.82 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3405, pruned_loss=0.09579, over 5719549.06 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3138, pruned_loss=0.08573, over 5697291.99 frames. ], batch size: 92, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:01:20,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.922e+02 9.926e+02 1.252e+03 1.577e+03 4.238e+03, threshold=2.504e+03, percent-clipped=7.0
+2023-03-04 18:01:30,721 INFO [train.py:968] (0/2) Epoch 9, batch 17950, giga_loss[loss=0.2344, simple_loss=0.3051, pruned_loss=0.08184, over 28969.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3116, pruned_loss=0.08482, over 5680957.47 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3408, pruned_loss=0.0959, over 5720477.31 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3105, pruned_loss=0.08444, over 5682232.67 frames. ], batch size: 213, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:02:09,831 INFO [train.py:968] (0/2) Epoch 9, batch 18000, giga_loss[loss=0.2104, simple_loss=0.2826, pruned_loss=0.0691, over 29125.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3101, pruned_loss=0.08382, over 5676812.36 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3406, pruned_loss=0.09569, over 5711159.30 frames. ], giga_tot_loss[loss=0.2372, simple_loss=0.3079, pruned_loss=0.08319, over 5685202.51 frames. ], batch size: 128, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:02:09,836 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 18:02:17,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2458, 1.8001, 1.3332, 0.3955], device='cuda:0'), covar=tensor([0.2985, 0.2147, 0.3489, 0.3738], device='cuda:0'), in_proj_covar=tensor([0.1483, 0.1418, 0.1448, 0.1214], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 18:02:18,282 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2187, simple_loss=0.3224, pruned_loss=0.05749, over 944034.00 frames. 
+2023-03-04 18:02:18,282 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 18:02:32,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6494, 2.1626, 1.9551, 1.5150], device='cuda:0'), covar=tensor([0.1740, 0.2044, 0.1362, 0.1563], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0698, 0.0842, 0.0752], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-04 18:02:49,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.260e+02 9.982e+02 1.339e+03 1.946e+03 7.160e+03, threshold=2.678e+03, percent-clipped=10.0
+2023-03-04 18:02:57,767 INFO [train.py:968] (0/2) Epoch 9, batch 18050, giga_loss[loss=0.1975, simple_loss=0.2754, pruned_loss=0.05975, over 28527.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3063, pruned_loss=0.0817, over 5688211.87 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3407, pruned_loss=0.09567, over 5716636.16 frames. ], giga_tot_loss[loss=0.2326, simple_loss=0.3036, pruned_loss=0.08085, over 5689113.22 frames. ], batch size: 336, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:03:11,195 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-04 18:03:11,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:03:19,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7233, 1.6720, 1.2977, 1.2318], device='cuda:0'), covar=tensor([0.0624, 0.0490, 0.0913, 0.0988], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0436, 0.0496, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 18:03:40,674 INFO [train.py:968] (0/2) Epoch 9, batch 18100, giga_loss[loss=0.1983, simple_loss=0.2721, pruned_loss=0.06225, over 28973.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3061, pruned_loss=0.08159, over 5681087.66 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3415, pruned_loss=0.09594, over 5715417.28 frames. ], giga_tot_loss[loss=0.2309, simple_loss=0.3017, pruned_loss=0.08009, over 5680568.34 frames. ], batch size: 136, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:03:47,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7585, 2.6095, 1.7162, 0.9789], device='cuda:0'), covar=tensor([0.5275, 0.2319, 0.2909, 0.4231], device='cuda:0'), in_proj_covar=tensor([0.1489, 0.1425, 0.1454, 0.1217], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 18:04:12,736 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.760e+02 1.079e+03 1.498e+03 2.204e+03 5.517e+03, threshold=2.995e+03, percent-clipped=16.0
+2023-03-04 18:04:15,083 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:04:20,778 INFO [train.py:968] (0/2) Epoch 9, batch 18150, giga_loss[loss=0.2211, simple_loss=0.2895, pruned_loss=0.0763, over 28716.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3041, pruned_loss=0.08098, over 5680345.29 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3423, pruned_loss=0.09637, over 5716005.07 frames. ], giga_tot_loss[loss=0.2286, simple_loss=0.2991, pruned_loss=0.07905, over 5678560.03 frames. ], batch size: 262, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:04:42,975 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2235, 2.0943, 1.4487, 1.6522], device='cuda:0'), covar=tensor([0.0674, 0.0647, 0.0983, 0.1091], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0432, 0.0492, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 18:04:56,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6660, 1.5204, 1.2485, 1.2503], device='cuda:0'), covar=tensor([0.0556, 0.0408, 0.0785, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0344, 0.0432, 0.0492, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 18:05:09,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:05:13,967 INFO [train.py:968] (0/2) Epoch 9, batch 18200, giga_loss[loss=0.3136, simple_loss=0.3682, pruned_loss=0.1295, over 26737.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3113, pruned_loss=0.08577, over 5669504.61 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3429, pruned_loss=0.09658, over 5718022.73 frames. ], giga_tot_loss[loss=0.237, simple_loss=0.3063, pruned_loss=0.08388, over 5665544.23 frames. ], batch size: 555, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:05:19,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=382405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:05:21,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=382408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:05:48,104 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=382437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:05:48,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.547e+02 1.137e+03 1.388e+03 1.816e+03 4.127e+03, threshold=2.776e+03, percent-clipped=6.0
+2023-03-04 18:05:57,158 INFO [train.py:968] (0/2) Epoch 9, batch 18250, giga_loss[loss=0.2965, simple_loss=0.3736, pruned_loss=0.1097, over 28904.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3246, pruned_loss=0.0925, over 5666125.39 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.343, pruned_loss=0.09662, over 5707132.66 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3196, pruned_loss=0.09063, over 5670530.45 frames. ], batch size: 112, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:06:03,937 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382456.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:06:29,047 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:06:36,072 INFO [train.py:968] (0/2) Epoch 9, batch 18300, giga_loss[loss=0.2686, simple_loss=0.3469, pruned_loss=0.09515, over 28920.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3368, pruned_loss=0.09853, over 5685342.27 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3436, pruned_loss=0.09674, over 5712423.10 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3318, pruned_loss=0.0969, over 5682966.87 frames. ], batch size: 213, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:06:58,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6557, 2.2066, 1.7005, 1.8790], device='cuda:0'), covar=tensor([0.0598, 0.0616, 0.0867, 0.1002], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0436, 0.0498, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 18:07:04,562 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=382535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:07:06,241 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.254e+02 1.285e+03 1.712e+03 2.370e+03 6.570e+03, threshold=3.424e+03, percent-clipped=14.0
+2023-03-04 18:07:06,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=382538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:07:13,699 INFO [train.py:968] (0/2) Epoch 9, batch 18350, giga_loss[loss=0.3046, simple_loss=0.382, pruned_loss=0.1136, over 28290.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3447, pruned_loss=0.1021, over 5664789.36 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3441, pruned_loss=0.09689, over 5689516.50 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3401, pruned_loss=0.1007, over 5682032.95 frames. ], batch size: 369, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:07:30,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=382567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:07:40,549 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-04 18:07:57,249 INFO [train.py:968] (0/2) Epoch 9, batch 18400, giga_loss[loss=0.33, simple_loss=0.3868, pruned_loss=0.1366, over 28920.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3485, pruned_loss=0.1024, over 5661436.87 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3442, pruned_loss=0.09686, over 5685360.17 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3447, pruned_loss=0.1015, over 5678827.41 frames. ], batch size: 112, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:07:59,747 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:08:30,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.588e+02 1.092e+03 1.284e+03 1.683e+03 3.921e+03, threshold=2.567e+03, percent-clipped=6.0
+2023-03-04 18:08:39,255 INFO [train.py:968] (0/2) Epoch 9, batch 18450, giga_loss[loss=0.2937, simple_loss=0.3692, pruned_loss=0.1091, over 28920.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3499, pruned_loss=0.1022, over 5667410.42 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3441, pruned_loss=0.09688, over 5689737.78 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3471, pruned_loss=0.1015, over 5676785.36 frames. ], batch size: 213, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:09:22,069 INFO [train.py:968] (0/2) Epoch 9, batch 18500, giga_loss[loss=0.2258, simple_loss=0.3092, pruned_loss=0.07116, over 29003.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3517, pruned_loss=0.1035, over 5663626.02 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3447, pruned_loss=0.09724, over 5694342.54 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3491, pruned_loss=0.1028, over 5666371.11 frames. ], batch size: 145, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:09:35,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:09:53,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:09:56,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.334e+02 1.141e+03 1.526e+03 1.968e+03 6.796e+03, threshold=3.052e+03, percent-clipped=11.0
+2023-03-04 18:10:03,823 INFO [train.py:968] (0/2) Epoch 9, batch 18550, giga_loss[loss=0.2577, simple_loss=0.3392, pruned_loss=0.08808, over 28607.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3542, pruned_loss=0.1057, over 5672258.08 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3446, pruned_loss=0.09696, over 5696804.76 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3524, pruned_loss=0.1055, over 5671813.52 frames. ], batch size: 71, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:10:46,903 INFO [train.py:968] (0/2) Epoch 9, batch 18600, libri_loss[loss=0.2845, simple_loss=0.3613, pruned_loss=0.1039, over 29669.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3581, pruned_loss=0.1086, over 5678351.71 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3449, pruned_loss=0.09722, over 5700790.81 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3567, pruned_loss=0.1085, over 5673365.05 frames. ], batch size: 91, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:11:12,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382831.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:11:18,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.194e+02 1.186e+03 1.527e+03 2.364e+03 8.689e+03, threshold=3.054e+03, percent-clipped=10.0
+2023-03-04 18:11:27,179 INFO [train.py:968] (0/2) Epoch 9, batch 18650, giga_loss[loss=0.3241, simple_loss=0.385, pruned_loss=0.1317, over 28741.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3605, pruned_loss=0.1094, over 5683247.73 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3449, pruned_loss=0.09706, over 5703282.14 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3598, pruned_loss=0.1099, over 5676294.20 frames. ], batch size: 99, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:11:35,295 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=382858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:11:37,399 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=382861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:11:38,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:12:00,920 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=382890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:12:06,801 INFO [train.py:968] (0/2) Epoch 9, batch 18700, giga_loss[loss=0.3121, simple_loss=0.3837, pruned_loss=0.1203, over 29030.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3624, pruned_loss=0.1095, over 5679593.83 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.345, pruned_loss=0.09694, over 5707397.43 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3621, pruned_loss=0.1102, over 5670038.71 frames. ], batch size: 128, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:12:30,520 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-04 18:12:38,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.472e+02 1.124e+03 1.360e+03 1.735e+03 3.499e+03, threshold=2.721e+03, percent-clipped=3.0
+2023-03-04 18:12:44,699 INFO [train.py:968] (0/2) Epoch 9, batch 18750, giga_loss[loss=0.2939, simple_loss=0.3647, pruned_loss=0.1116, over 27897.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3643, pruned_loss=0.1097, over 5690186.56 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.346, pruned_loss=0.09747, over 5710086.88 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3636, pruned_loss=0.1101, over 5679804.21 frames. ], batch size: 412, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:13:06,178 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=382974.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:13:07,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:08,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=382977.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:13:22,277 INFO [train.py:968] (0/2) Epoch 9, batch 18800, giga_loss[loss=0.2659, simple_loss=0.3528, pruned_loss=0.08948, over 29123.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3639, pruned_loss=0.1083, over 5690280.56 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.346, pruned_loss=0.0974, over 5705857.28 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3638, pruned_loss=0.109, over 5684497.73 frames. ], batch size: 155, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:13:29,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:31,125 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383006.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:13:32,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:38,187 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 18:13:40,729 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:49,739 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:55,698 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:57,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.607e+02 1.070e+03 1.264e+03 1.764e+03 1.135e+04, threshold=2.528e+03, percent-clipped=15.0
+2023-03-04 18:14:04,309 INFO [train.py:968] (0/2) Epoch 9, batch 18850, giga_loss[loss=0.2621, simple_loss=0.3441, pruned_loss=0.09002, over 28677.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3621, pruned_loss=0.1059, over 5706738.93 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3456, pruned_loss=0.09709, over 5710746.73 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3629, pruned_loss=0.1071, over 5697310.86 frames. ], batch size: 92, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:14:16,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4525, 2.8347, 1.5700, 1.5657], device='cuda:0'), covar=tensor([0.0737, 0.0251, 0.0674, 0.1038], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0485, 0.0324, 0.0377], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 18:14:44,438 INFO [train.py:968] (0/2) Epoch 9, batch 18900, giga_loss[loss=0.2649, simple_loss=0.3468, pruned_loss=0.09148, over 28911.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3601, pruned_loss=0.1043, over 5706583.28 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3456, pruned_loss=0.09709, over 5710746.73 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3607, pruned_loss=0.1051, over 5699245.29 frames. ], batch size: 145, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:14:53,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:14:59,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1989, 1.4202, 1.3081, 0.9434], device='cuda:0'), covar=tensor([0.1791, 0.1614, 0.0986, 0.1567], device='cuda:0'), in_proj_covar=tensor([0.1626, 0.1485, 0.1435, 0.1554], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 18:15:03,147 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:15:05,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:15:18,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.148e+02 1.094e+03 1.282e+03 1.589e+03 3.499e+03, threshold=2.564e+03, percent-clipped=8.0
+2023-03-04 18:15:25,924 INFO [train.py:968] (0/2) Epoch 9, batch 18950, giga_loss[loss=0.3178, simple_loss=0.3828, pruned_loss=0.1264, over 28652.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3616, pruned_loss=0.1059, over 5705343.52 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3462, pruned_loss=0.09723, over 5706027.12 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.362, pruned_loss=0.1067, over 5703034.27 frames. ], batch size: 92, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:15:29,199 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:16:09,758 INFO [train.py:968] (0/2) Epoch 9, batch 19000, giga_loss[loss=0.3006, simple_loss=0.365, pruned_loss=0.1181, over 28576.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3637, pruned_loss=0.1098, over 5712539.12 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3461, pruned_loss=0.0972, over 5711269.59 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3644, pruned_loss=0.1107, over 5706051.04 frames. ], batch size: 307, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:16:43,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.202e+02 1.283e+03 1.645e+03 2.230e+03 6.382e+03, threshold=3.289e+03, percent-clipped=18.0
+2023-03-04 18:16:50,854 INFO [train.py:968] (0/2) Epoch 9, batch 19050, giga_loss[loss=0.3225, simple_loss=0.3759, pruned_loss=0.1346, over 28599.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3662, pruned_loss=0.1138, over 5694216.83 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3468, pruned_loss=0.09762, over 5696628.51 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3666, pruned_loss=0.1144, over 5701358.13 frames. ], batch size: 71, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:16:55,207 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1965, 1.2496, 1.1047, 0.8912], device='cuda:0'), covar=tensor([0.0753, 0.0525, 0.1027, 0.0907], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0439, 0.0497, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 18:16:55,893 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:16:57,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:17:21,379 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:17:31,177 INFO [train.py:968] (0/2) Epoch 9, batch 19100, giga_loss[loss=0.2646, simple_loss=0.3409, pruned_loss=0.09417, over 28741.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3649, pruned_loss=0.1137, over 5696412.11 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3474, pruned_loss=0.0978, over 5700205.40 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.365, pruned_loss=0.1144, over 5698776.01 frames. ], batch size: 119, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:17:33,400 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9157, 1.3034, 1.0845, 0.1165], device='cuda:0'), covar=tensor([0.2274, 0.1888, 0.3042, 0.3913], device='cuda:0'), in_proj_covar=tensor([0.1495, 0.1424, 0.1460, 0.1225], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 18:17:42,589 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:17:52,721 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-04 18:17:56,271 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:18:03,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.190e+02 1.131e+03 1.429e+03 2.189e+03 3.876e+03, threshold=2.857e+03, percent-clipped=5.0
+2023-03-04 18:18:11,162 INFO [train.py:968] (0/2) Epoch 9, batch 19150, giga_loss[loss=0.3013, simple_loss=0.3655, pruned_loss=0.1186, over 28755.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3631, pruned_loss=0.1131, over 5695530.77 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3479, pruned_loss=0.09793, over 5703146.24 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.363, pruned_loss=0.1138, over 5694823.62 frames. ], batch size: 99, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:18:52,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:18:54,912 INFO [train.py:968] (0/2) Epoch 9, batch 19200, libri_loss[loss=0.3236, simple_loss=0.3917, pruned_loss=0.1278, over 29132.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3625, pruned_loss=0.1119, over 5711466.35 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3488, pruned_loss=0.09821, over 5708440.20 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.362, pruned_loss=0.1126, over 5706024.89 frames. ], batch size: 101, lr: 3.61e-03, grad_scale: 8.0
+2023-03-04 18:19:01,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:19:26,840 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.361e+02 1.181e+03 1.795e+03 2.285e+03 4.861e+03, threshold=3.590e+03, percent-clipped=12.0
+2023-03-04 18:19:33,287 INFO [train.py:968] (0/2) Epoch 9, batch 19250, giga_loss[loss=0.2994, simple_loss=0.3641, pruned_loss=0.1173, over 28884.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3613, pruned_loss=0.1105, over 5705723.89 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3501, pruned_loss=0.09909, over 5701085.25 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3603, pruned_loss=0.1108, over 5707963.74 frames. ], batch size: 174, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:20:11,072 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:20:16,970 INFO [train.py:968] (0/2) Epoch 9, batch 19300, giga_loss[loss=0.2754, simple_loss=0.3411, pruned_loss=0.1049, over 27720.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3575, pruned_loss=0.1082, over 5692445.02 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3499, pruned_loss=0.09896, over 5705442.64 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3571, pruned_loss=0.1087, over 5689964.66 frames. ], batch size: 472, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:20:53,752 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:20:55,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.226e+02 1.045e+03 1.390e+03 1.931e+03 7.299e+03, threshold=2.781e+03, percent-clipped=5.0
+2023-03-04 18:20:55,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:21:02,261 INFO [train.py:968] (0/2) Epoch 9, batch 19350, giga_loss[loss=0.297, simple_loss=0.36, pruned_loss=0.117, over 28715.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3517, pruned_loss=0.1052, over 5675017.29 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3504, pruned_loss=0.09922, over 5695206.60 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.351, pruned_loss=0.1055, over 5681455.94 frames. ], batch size: 262, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:21:03,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:21:07,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:21:22,138 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:21:32,321 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:21:48,152 INFO [train.py:968] (0/2) Epoch 9, batch 19400, giga_loss[loss=0.2507, simple_loss=0.3199, pruned_loss=0.09076, over 29038.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3465, pruned_loss=0.1025, over 5677661.88 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3505, pruned_loss=0.09913, over 5696500.07 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3459, pruned_loss=0.103, over 5681330.97 frames. ], batch size: 128, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:22:24,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.430e+02 1.010e+03 1.357e+03 2.306e+03 8.793e+03, threshold=2.714e+03, percent-clipped=17.0
+2023-03-04 18:22:31,759 INFO [train.py:968] (0/2) Epoch 9, batch 19450, giga_loss[loss=0.3012, simple_loss=0.3698, pruned_loss=0.1164, over 28862.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3426, pruned_loss=0.1005, over 5648001.90 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3514, pruned_loss=0.09963, over 5690491.99 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3411, pruned_loss=0.1005, over 5655779.72 frames. ], batch size: 112, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:23:07,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:23:15,408 INFO [train.py:968] (0/2) Epoch 9, batch 19500, libri_loss[loss=0.2814, simple_loss=0.3629, pruned_loss=0.09996, over 29525.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3433, pruned_loss=0.1008, over 5651886.21 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3516, pruned_loss=0.09961, over 5693700.57 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3417, pruned_loss=0.1009, over 5654379.84 frames. ], batch size: 81, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:23:22,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:23:56,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.087e+02 9.489e+02 1.173e+03 1.469e+03 5.272e+03, threshold=2.347e+03, percent-clipped=5.0
+2023-03-04 18:24:01,289 INFO [train.py:968] (0/2) Epoch 9, batch 19550, giga_loss[loss=0.2776, simple_loss=0.356, pruned_loss=0.09965, over 28506.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3434, pruned_loss=0.101, over 5653551.57 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3518, pruned_loss=0.09961, over 5692633.99 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3419, pruned_loss=0.1011, over 5655930.30 frames. ], batch size: 336, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:24:38,073 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.31 vs. limit=5.0
+2023-03-04 18:24:38,890 INFO [train.py:968] (0/2) Epoch 9, batch 19600, giga_loss[loss=0.2717, simple_loss=0.3417, pruned_loss=0.1008, over 28921.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.343, pruned_loss=0.1006, over 5666580.46 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3529, pruned_loss=0.1001, over 5691815.32 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3403, pruned_loss=0.1002, over 5667651.90 frames. ], batch size: 227, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:24:58,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8324, 1.7938, 1.6135, 1.6299], device='cuda:0'), covar=tensor([0.1298, 0.2176, 0.1855, 0.1863], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0730, 0.0658, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 18:25:06,786 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:08,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:13,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4604, 1.7239, 1.3991, 1.7999], device='cuda:0'), covar=tensor([0.2376, 0.2299, 0.2489, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.1225, 0.0916, 0.1087, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 18:25:13,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.902e+02 1.028e+03 1.361e+03 1.887e+03 9.143e+03, threshold=2.721e+03, percent-clipped=15.0
+2023-03-04 18:25:18,484 INFO [train.py:968] (0/2) Epoch 9, batch 19650, giga_loss[loss=0.2846, simple_loss=0.3493, pruned_loss=0.1099, over 28848.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3406, pruned_loss=0.09893, over 5679465.84 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3532, pruned_loss=0.09995, over 5694962.24 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3378, pruned_loss=0.09874, over 5676864.61 frames. ], batch size: 174, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:25:19,610 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:22,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:29,708 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:31,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3832, 2.8491, 1.4750, 1.4007], device='cuda:0'), covar=tensor([0.0873, 0.0328, 0.0793, 0.1275], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0482, 0.0321, 0.0374], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:0')
+2023-03-04 18:25:31,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:36,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7098, 1.2745, 5.1836, 3.5478], device='cuda:0'), covar=tensor([0.1580, 0.2630, 0.0312, 0.0731], device='cuda:0'), in_proj_covar=tensor([0.0614, 0.0561, 0.0802, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 18:25:46,741 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:26:00,640 INFO [train.py:968] (0/2) Epoch 9, batch 19700, giga_loss[loss=0.2262, simple_loss=0.2972, pruned_loss=0.07756, over 28818.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3368, pruned_loss=0.09675, over 5686192.80 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3532, pruned_loss=0.09985, over 5697056.34 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3344, pruned_loss=0.09667, over 5682025.18 frames. ], batch size: 99, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:26:22,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:26:33,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.177e+02 1.002e+03 1.236e+03 1.650e+03 5.886e+03, threshold=2.472e+03, percent-clipped=7.0
+2023-03-04 18:26:35,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-04 18:26:39,545 INFO [train.py:968] (0/2) Epoch 9, batch 19750, giga_loss[loss=0.2804, simple_loss=0.3473, pruned_loss=0.1068, over 28559.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3351, pruned_loss=0.09576, over 5697674.53 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3534, pruned_loss=0.09992, over 5701431.32 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3326, pruned_loss=0.09556, over 5690360.13 frames. ], batch size: 307, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:26:48,307 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:26:54,058 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:27:00,655 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3400, 1.8167, 1.3578, 1.3916], device='cuda:0'), covar=tensor([0.0746, 0.0275, 0.0327, 0.0796], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0118, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0053, 0.0048, 0.0081], device='cuda:0')
+2023-03-04 18:27:17,429 INFO [train.py:968] (0/2) Epoch 9, batch 19800, giga_loss[loss=0.2546, simple_loss=0.3238, pruned_loss=0.0927, over 28912.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3324, pruned_loss=0.09463, over 5698994.35 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.354, pruned_loss=0.1003, over 5698156.17 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3295, pruned_loss=0.09406, over 5695570.03 frames. ], batch size: 227, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:27:20,843 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-384000.pt
+2023-03-04 18:27:26,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:27:28,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:27:50,390 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:27:51,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.734e+02 9.477e+02 1.251e+03 1.774e+03 5.242e+03, threshold=2.502e+03, percent-clipped=10.0
+2023-03-04 18:27:54,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7519, 2.0317, 1.5877, 2.3243], device='cuda:0'), covar=tensor([0.2032, 0.1976, 0.2229, 0.1878], device='cuda:0'), in_proj_covar=tensor([0.1234, 0.0921, 0.1094, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 18:27:56,109 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384046.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:27:57,209 INFO [train.py:968] (0/2) Epoch 9, batch 19850, giga_loss[loss=0.2533, simple_loss=0.327, pruned_loss=0.08986, over 29025.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3306, pruned_loss=0.09379, over 5704339.02 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3539, pruned_loss=0.09996, over 5697626.22 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3275, pruned_loss=0.09338, over 5702619.22 frames. ], batch size: 213, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:28:30,284 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:28:37,212 INFO [train.py:968] (0/2) Epoch 9, batch 19900, giga_loss[loss=0.2535, simple_loss=0.3233, pruned_loss=0.09179, over 28998.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3298, pruned_loss=0.09334, over 5711505.30 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3542, pruned_loss=0.09998, over 5701323.66 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3266, pruned_loss=0.09288, over 5707226.56 frames. ], batch size: 164, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:29:08,621 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.733e+02 1.066e+03 1.239e+03 1.704e+03 4.037e+03, threshold=2.477e+03, percent-clipped=7.0
+2023-03-04 18:29:13,267 INFO [train.py:968] (0/2) Epoch 9, batch 19950, giga_loss[loss=0.2526, simple_loss=0.3222, pruned_loss=0.09151, over 28881.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3298, pruned_loss=0.09324, over 5698810.52 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3554, pruned_loss=0.1006, over 5689200.83 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3251, pruned_loss=0.09205, over 5706454.71 frames. ], batch size: 174, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:29:40,887 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-04 18:29:51,528 INFO [train.py:968] (0/2) Epoch 9, batch 20000, giga_loss[loss=0.2206, simple_loss=0.303, pruned_loss=0.06912, over 28904.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.329, pruned_loss=0.09218, over 5707511.92 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3561, pruned_loss=0.1008, over 5696395.23 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3233, pruned_loss=0.09068, over 5707910.78 frames. ], batch size: 174, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:30:26,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.159e+02 1.005e+03 1.220e+03 1.584e+03 4.491e+03, threshold=2.441e+03, percent-clipped=10.0
+2023-03-04 18:30:31,120 INFO [train.py:968] (0/2) Epoch 9, batch 20050, giga_loss[loss=0.2649, simple_loss=0.3334, pruned_loss=0.09817, over 28662.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3287, pruned_loss=0.09204, over 5714757.52 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3566, pruned_loss=0.101, over 5699066.44 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3233, pruned_loss=0.09055, over 5713191.42 frames. ], batch size: 284, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:30:54,951 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:30:59,901 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-04 18:31:01,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:31:13,157 INFO [train.py:968] (0/2) Epoch 9, batch 20100, giga_loss[loss=0.2528, simple_loss=0.3353, pruned_loss=0.08515, over 28997.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3341, pruned_loss=0.09585, over 5712917.16 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3568, pruned_loss=0.1009, over 5703322.06 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3291, pruned_loss=0.09456, over 5708367.15 frames. ], batch size: 164, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:31:16,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:31:31,137 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-04 18:31:45,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:31:52,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:31:53,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.474e+02 1.142e+03 1.393e+03 1.957e+03 5.490e+03, threshold=2.787e+03, percent-clipped=6.0
+2023-03-04 18:31:58,034 INFO [train.py:968] (0/2) Epoch 9, batch 20150, libri_loss[loss=0.297, simple_loss=0.3759, pruned_loss=0.1091, over 29516.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3401, pruned_loss=0.09982, over 5696985.90 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3577, pruned_loss=0.1012, over 5698361.41 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3346, pruned_loss=0.09837, over 5697430.91 frames. ], batch size: 82, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:32:03,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9437, 1.9306, 1.3361, 1.5190], device='cuda:0'), covar=tensor([0.0707, 0.0615, 0.1003, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0443, 0.0497, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 18:32:14,338 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-04 18:32:46,762 INFO [train.py:968] (0/2) Epoch 9, batch 20200, giga_loss[loss=0.3126, simple_loss=0.3697, pruned_loss=0.1277, over 28818.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3478, pruned_loss=0.1051, over 5689509.99 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3575, pruned_loss=0.1011, over 5692688.03 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3431, pruned_loss=0.104, over 5695064.04 frames. ], batch size: 119, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:33:09,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384421.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:33:26,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.721e+02 1.189e+03 1.432e+03 1.900e+03 6.847e+03, threshold=2.864e+03, percent-clipped=12.0
+2023-03-04 18:33:28,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:33:31,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:33:32,041 INFO [train.py:968] (0/2) Epoch 9, batch 20250, giga_loss[loss=0.2655, simple_loss=0.3503, pruned_loss=0.09039, over 29099.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3537, pruned_loss=0.1085, over 5688478.77 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3574, pruned_loss=0.101, over 5695613.28 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.35, pruned_loss=0.1078, over 5690149.08 frames. ], batch size: 155, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:33:47,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:00,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:01,886 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:04,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:06,885 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:09,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:19,804 INFO [train.py:968] (0/2) Epoch 9, batch 20300, giga_loss[loss=0.3411, simple_loss=0.4082, pruned_loss=0.137, over 28992.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3593, pruned_loss=0.1111, over 5683483.62 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3575, pruned_loss=0.1011, over 5695804.21 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3563, pruned_loss=0.1105, over 5684596.83 frames. ], batch size: 213, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:34:30,276 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:33,696 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:40,352 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-04 18:34:58,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.392e+02 1.157e+03 1.485e+03 1.992e+03 5.297e+03, threshold=2.971e+03, percent-clipped=6.0
+2023-03-04 18:35:04,701 INFO [train.py:968] (0/2) Epoch 9, batch 20350, giga_loss[loss=0.3034, simple_loss=0.3723, pruned_loss=0.1172, over 28838.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3639, pruned_loss=0.1131, over 5684869.65 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3574, pruned_loss=0.101, over 5698626.43 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3616, pruned_loss=0.1129, over 5683109.67 frames. ], batch size: 199, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:35:19,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384564.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:35:21,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384567.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:35:43,005 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384596.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:35:44,472 INFO [train.py:968] (0/2) Epoch 9, batch 20400, giga_loss[loss=0.2832, simple_loss=0.3578, pruned_loss=0.1043, over 28778.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3657, pruned_loss=0.1141, over 5693829.01 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3566, pruned_loss=0.1006, over 5702574.63 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3649, pruned_loss=0.1147, over 5688289.82 frames. ], batch size: 119, lr: 3.61e-03, grad_scale: 8.0
+2023-03-04 18:35:50,943 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:35:52,862 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:36:20,637 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:36:24,761 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.865e+02 1.163e+03 1.446e+03 1.913e+03 3.780e+03, threshold=2.891e+03, percent-clipped=3.0
+2023-03-04 18:36:28,933 INFO [train.py:968] (0/2) Epoch 9, batch 20450, libri_loss[loss=0.2611, simple_loss=0.3338, pruned_loss=0.09424, over 28647.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3588, pruned_loss=0.1092, over 5685379.28 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3564, pruned_loss=0.1006, over 5695595.73 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3584, pruned_loss=0.11, over 5686298.05 frames. ], batch size: 63, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:36:32,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:36:39,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:36:55,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2692, 4.1145, 3.8556, 1.7375], device='cuda:0'), covar=tensor([0.0529, 0.0605, 0.0648, 0.2192], device='cuda:0'), in_proj_covar=tensor([0.0948, 0.0894, 0.0784, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 18:36:56,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3744, 2.9872, 1.3921, 1.5176], device='cuda:0'), covar=tensor([0.0878, 0.0282, 0.0811, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0485, 0.0322, 0.0378], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 18:37:12,532 INFO [train.py:968] (0/2) Epoch 9, batch 20500, giga_loss[loss=0.3244, simple_loss=0.3725, pruned_loss=0.1382, over 26667.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3562, pruned_loss=0.1068, over 5686616.74 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3565, pruned_loss=0.1006, over 5696681.76 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3559, pruned_loss=0.1074, over 5686419.49 frames. ], batch size: 555, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:37:15,012 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-04 18:37:25,840 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-04 18:37:35,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:37:48,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.172e+02 1.189e+03 1.556e+03 2.111e+03 7.933e+03, threshold=3.112e+03, percent-clipped=13.0
+2023-03-04 18:37:54,699 INFO [train.py:968] (0/2) Epoch 9, batch 20550, giga_loss[loss=0.2522, simple_loss=0.3386, pruned_loss=0.08288, over 28880.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3562, pruned_loss=0.1058, over 5685552.11 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.357, pruned_loss=0.1009, over 5695228.96 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3554, pruned_loss=0.1061, over 5686436.43 frames. ], batch size: 145, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:38:03,103 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:38:31,445 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:38:34,399 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:38:34,683 INFO [train.py:968] (0/2) Epoch 9, batch 20600, giga_loss[loss=0.3213, simple_loss=0.3859, pruned_loss=0.1284, over 27647.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3585, pruned_loss=0.1073, over 5687204.55 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3571, pruned_loss=0.101, over 5698365.76 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3578, pruned_loss=0.1076, over 5685088.30 frames. ], batch size: 472, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:38:39,411 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:38:42,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:38:59,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:39:07,805 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:39:12,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.284e+02 1.303e+03 1.677e+03 2.186e+03 5.770e+03, threshold=3.353e+03, percent-clipped=9.0
+2023-03-04 18:39:18,231 INFO [train.py:968] (0/2) Epoch 9, batch 20650, giga_loss[loss=0.2946, simple_loss=0.3658, pruned_loss=0.1117, over 28917.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.362, pruned_loss=0.1101, over 5687120.76 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3572, pruned_loss=0.101, over 5693262.12 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3614, pruned_loss=0.1105, over 5690252.03 frames. ], batch size: 112, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:40:01,628 INFO [train.py:968] (0/2) Epoch 9, batch 20700, giga_loss[loss=0.2959, simple_loss=0.3629, pruned_loss=0.1144, over 28223.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3625, pruned_loss=0.1106, over 5693978.84 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3571, pruned_loss=0.1008, over 5697071.10 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3623, pruned_loss=0.1113, over 5693020.14 frames. ], batch size: 368, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:40:47,067 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.810e+02 1.262e+03 1.619e+03 2.120e+03 7.034e+03, threshold=3.238e+03, percent-clipped=5.0
+2023-03-04 18:40:50,587 INFO [train.py:968] (0/2) Epoch 9, batch 20750, giga_loss[loss=0.2735, simple_loss=0.3509, pruned_loss=0.098, over 29019.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3633, pruned_loss=0.1113, over 5703772.55 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3573, pruned_loss=0.1009, over 5696344.75 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.363, pruned_loss=0.1118, over 5703638.51 frames. ], batch size: 164, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:41:30,417 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 18:41:33,354 INFO [train.py:968] (0/2) Epoch 9, batch 20800, giga_loss[loss=0.2902, simple_loss=0.3573, pruned_loss=0.1115, over 28824.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3641, pruned_loss=0.1124, over 5701272.12 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3574, pruned_loss=0.1008, over 5699239.33 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3637, pruned_loss=0.113, over 5698780.72 frames. ], batch size: 119, lr: 3.61e-03, grad_scale: 8.0
+2023-03-04 18:41:52,629 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:41:55,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6804, 1.5824, 4.8564, 3.6353], device='cuda:0'), covar=tensor([0.1423, 0.2323, 0.0324, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0621, 0.0563, 0.0807, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 18:42:08,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.725e+02 1.045e+03 1.234e+03 1.683e+03 4.321e+03, threshold=2.469e+03, percent-clipped=4.0
+2023-03-04 18:42:13,100 INFO [train.py:968] (0/2) Epoch 9, batch 20850, giga_loss[loss=0.2684, simple_loss=0.3485, pruned_loss=0.09413, over 28990.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3633, pruned_loss=0.1111, over 5703812.12 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3573, pruned_loss=0.1007, over 5698930.89 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3632, pruned_loss=0.1118, over 5702253.66 frames. ], batch size: 106, lr: 3.61e-03, grad_scale: 8.0
+2023-03-04 18:42:23,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-04 18:42:52,704 INFO [train.py:968] (0/2) Epoch 9, batch 20900, giga_loss[loss=0.2601, simple_loss=0.3385, pruned_loss=0.09085, over 28594.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3629, pruned_loss=0.1097, over 5705994.99 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.357, pruned_loss=0.1005, over 5700996.58 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3631, pruned_loss=0.1105, over 5703010.75 frames. ], batch size: 71, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:42:56,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=385102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:43:20,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8005, 1.1140, 1.1203, 0.9510], device='cuda:0'), covar=tensor([0.1604, 0.1306, 0.1958, 0.1501], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0728, 0.0657, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 18:43:22,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=385135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:43:28,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.007e+02 1.143e+03 1.320e+03 1.953e+03 5.139e+03, threshold=2.641e+03, percent-clipped=14.0
+2023-03-04 18:43:31,744 INFO [train.py:968] (0/2) Epoch 9, batch 20950, giga_loss[loss=0.2594, simple_loss=0.343, pruned_loss=0.0879, over 28246.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3638, pruned_loss=0.1093, over 5711365.65 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3575, pruned_loss=0.1007, over 5701907.70 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3637, pruned_loss=0.1099, over 5708377.91 frames. ], batch size: 77, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:43:57,039 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-04 18:44:10,103 INFO [train.py:968] (0/2) Epoch 9, batch 21000, giga_loss[loss=0.2472, simple_loss=0.3312, pruned_loss=0.08158, over 28594.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3617, pruned_loss=0.1082, over 5702067.05 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3578, pruned_loss=0.1011, over 5693560.97 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3614, pruned_loss=0.1085, over 5708709.66 frames. ], batch size: 78, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:44:10,107 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 18:44:18,485 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2261, simple_loss=0.331, pruned_loss=0.0606, over 944034.00 frames. 
+2023-03-04 18:44:18,486 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 18:44:30,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2408, 1.1711, 1.0777, 1.4216], device='cuda:0'), covar=tensor([0.0786, 0.0338, 0.0340, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0112, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0053, 0.0048, 0.0081], device='cuda:0')
+2023-03-04 18:44:52,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.702e+02 1.020e+03 1.335e+03 1.798e+03 7.012e+03, threshold=2.671e+03, percent-clipped=8.0
+2023-03-04 18:44:52,716 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=385245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:44:54,948 INFO [train.py:968] (0/2) Epoch 9, batch 21050, giga_loss[loss=0.253, simple_loss=0.3324, pruned_loss=0.08677, over 28994.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3591, pruned_loss=0.1068, over 5700568.27 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3578, pruned_loss=0.101, over 5690452.12 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.359, pruned_loss=0.1073, over 5709189.52 frames. ], batch size: 106, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:44:55,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=385248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:45:00,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1775, 0.7940, 0.9193, 1.2994], device='cuda:0'), covar=tensor([0.0792, 0.0359, 0.0346, 0.0840], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0053, 0.0048, 0.0081], device='cuda:0')
+2023-03-04 18:45:10,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:45:16,990 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=385277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:45:17,664 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=385278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:45:20,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=385281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:45:32,910 INFO [train.py:968] (0/2) Epoch 9, batch 21100, giga_loss[loss=0.2831, simple_loss=0.3603, pruned_loss=0.103, over 28715.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3582, pruned_loss=0.1063, over 5698638.68 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3583, pruned_loss=0.1013, over 5691800.81 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3577, pruned_loss=0.1066, over 5705075.28 frames. ], batch size: 242, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:45:41,739 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=385310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:46:10,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.727e+02 1.037e+03 1.307e+03 1.600e+03 2.976e+03, threshold=2.613e+03, percent-clipped=3.0
+2023-03-04 18:46:12,493 INFO [train.py:968] (0/2) Epoch 9, batch 21150, giga_loss[loss=0.2876, simple_loss=0.365, pruned_loss=0.1051, over 29066.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3576, pruned_loss=0.1065, over 5707490.38 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3584, pruned_loss=0.1015, over 5693400.65 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3571, pruned_loss=0.1067, over 5711854.16 frames. ], batch size: 155, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:46:52,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=385396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:46:53,682 INFO [train.py:968] (0/2) Epoch 9, batch 21200, libri_loss[loss=0.2932, simple_loss=0.3695, pruned_loss=0.1085, over 29764.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3592, pruned_loss=0.108, over 5702686.25 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3585, pruned_loss=0.1017, over 5694178.81 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3587, pruned_loss=0.1082, over 5705481.59 frames. ], batch size: 87, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:47:30,687 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.407e+02 1.001e+03 1.277e+03 1.880e+03 7.049e+03, threshold=2.555e+03, percent-clipped=10.0
+2023-03-04 18:47:33,029 INFO [train.py:968] (0/2) Epoch 9, batch 21250, giga_loss[loss=0.2605, simple_loss=0.3421, pruned_loss=0.08944, over 28858.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3579, pruned_loss=0.1064, over 5702637.68 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3586, pruned_loss=0.1019, over 5689228.37 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3574, pruned_loss=0.1065, over 5709364.15 frames. ], batch size: 119, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:47:59,177 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-04 18:48:12,227 INFO [train.py:968] (0/2) Epoch 9, batch 21300, giga_loss[loss=0.2517, simple_loss=0.3319, pruned_loss=0.0857, over 28609.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3584, pruned_loss=0.1059, over 5691951.21 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3591, pruned_loss=0.1022, over 5672709.55 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3575, pruned_loss=0.1059, over 5712537.44 frames. ], batch size: 60, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:48:47,125 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=385539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:48:49,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=385542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:48:51,427 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.786e+02 1.005e+03 1.261e+03 1.852e+03 6.010e+03, threshold=2.523e+03, percent-clipped=10.0
+2023-03-04 18:48:52,694 INFO [train.py:968] (0/2) Epoch 9, batch 21350, giga_loss[loss=0.3995, simple_loss=0.424, pruned_loss=0.1875, over 26584.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3575, pruned_loss=0.106, over 5687908.98 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3587, pruned_loss=0.1019, over 5676352.64 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3572, pruned_loss=0.1063, over 5701145.32 frames. ], batch size: 555, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:49:10,623 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=385571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:49:30,536 INFO [train.py:968] (0/2) Epoch 9, batch 21400, giga_loss[loss=0.2536, simple_loss=0.3321, pruned_loss=0.0875, over 28878.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3561, pruned_loss=0.1057, over 5686907.80 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.359, pruned_loss=0.1022, over 5679103.21 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3555, pruned_loss=0.1057, over 5695660.86 frames. ], batch size: 227, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:50:04,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=385642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:50:05,691 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:50:06,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.906e+02 1.106e+03 1.460e+03 2.150e+03 9.304e+03, threshold=2.920e+03, percent-clipped=16.0
+2023-03-04 18:50:09,365 INFO [train.py:968] (0/2) Epoch 9, batch 21450, giga_loss[loss=0.3062, simple_loss=0.3674, pruned_loss=0.1225, over 27634.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3537, pruned_loss=0.1047, over 5693909.64 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.359, pruned_loss=0.1022, over 5682212.54 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3532, pruned_loss=0.1048, over 5698214.61 frames. ], batch size: 472, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:50:10,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5108, 1.7446, 1.8198, 1.3752], device='cuda:0'), covar=tensor([0.1522, 0.2022, 0.1225, 0.1429], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0698, 0.0837, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 18:50:47,170 INFO [train.py:968] (0/2) Epoch 9, batch 21500, giga_loss[loss=0.2811, simple_loss=0.3521, pruned_loss=0.1051, over 28823.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3516, pruned_loss=0.1039, over 5687972.75 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3593, pruned_loss=0.1025, over 5678812.33 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3507, pruned_loss=0.1038, over 5694961.62 frames. ], batch size: 119, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:50:48,147 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-04 18:51:12,019 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385730.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:51:25,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.136e+02 1.170e+03 1.550e+03 1.936e+03 9.593e+03, threshold=3.101e+03, percent-clipped=9.0
+2023-03-04 18:51:26,765 INFO [train.py:968] (0/2) Epoch 9, batch 21550, giga_loss[loss=0.2862, simple_loss=0.352, pruned_loss=0.1102, over 29046.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3515, pruned_loss=0.1046, over 5683115.36 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3594, pruned_loss=0.1026, over 5679972.27 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3507, pruned_loss=0.1044, over 5687690.11 frames. ], batch size: 213, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:51:36,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8773, 2.7963, 1.9095, 0.8836], device='cuda:0'), covar=tensor([0.4246, 0.1828, 0.2448, 0.4151], device='cuda:0'), in_proj_covar=tensor([0.1498, 0.1412, 0.1455, 0.1214], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 18:51:55,043 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=385785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:51:57,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=385788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:52:07,154 INFO [train.py:968] (0/2) Epoch 9, batch 21600, giga_loss[loss=0.2686, simple_loss=0.3458, pruned_loss=0.09574, over 28674.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3495, pruned_loss=0.1037, over 5690014.68 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3591, pruned_loss=0.1025, over 5683562.90 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.349, pruned_loss=0.1037, over 5690594.19 frames. ], batch size: 336, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 18:52:22,354 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=385817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:52:25,508 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:52:39,369 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:52:45,475 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.283e+02 1.084e+03 1.327e+03 1.886e+03 7.035e+03, threshold=2.654e+03, percent-clipped=9.0
+2023-03-04 18:52:46,153 INFO [train.py:968] (0/2) Epoch 9, batch 21650, giga_loss[loss=0.2654, simple_loss=0.3339, pruned_loss=0.09849, over 28915.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3464, pruned_loss=0.1023, over 5703432.11 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3591, pruned_loss=0.1025, over 5689130.31 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3457, pruned_loss=0.1023, over 5699145.79 frames. ], batch size: 99, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:53:23,538 INFO [train.py:968] (0/2) Epoch 9, batch 21700, libri_loss[loss=0.2924, simple_loss=0.3666, pruned_loss=0.1091, over 29522.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3452, pruned_loss=0.102, over 5708759.18 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3594, pruned_loss=0.1028, over 5694613.97 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.344, pruned_loss=0.1018, over 5700733.98 frames. ], batch size: 84, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:54:01,957 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.910e+02 1.009e+03 1.285e+03 1.682e+03 5.097e+03, threshold=2.570e+03, percent-clipped=7.0
+2023-03-04 18:54:02,516 INFO [train.py:968] (0/2) Epoch 9, batch 21750, giga_loss[loss=0.2618, simple_loss=0.3353, pruned_loss=0.09411, over 28618.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3431, pruned_loss=0.1012, over 5705844.72 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3592, pruned_loss=0.1027, over 5688933.33 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3421, pruned_loss=0.101, over 5705082.08 frames. ], batch size: 307, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:54:41,771 INFO [train.py:968] (0/2) Epoch 9, batch 21800, giga_loss[loss=0.3083, simple_loss=0.3759, pruned_loss=0.1204, over 28252.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3428, pruned_loss=0.1011, over 5694239.33 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3592, pruned_loss=0.1029, over 5674387.92 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3417, pruned_loss=0.1007, over 5706581.78 frames. ], batch size: 368, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:54:43,477 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-386000.pt
+2023-03-04 18:54:58,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=386019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:55:21,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.292e+02 1.078e+03 1.255e+03 1.797e+03 6.601e+03, threshold=2.510e+03, percent-clipped=10.0
+2023-03-04 18:55:22,316 INFO [train.py:968] (0/2) Epoch 9, batch 21850, libri_loss[loss=0.2796, simple_loss=0.3414, pruned_loss=0.1089, over 29428.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3459, pruned_loss=0.1021, over 5697359.06 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3596, pruned_loss=0.1034, over 5679432.65 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3442, pruned_loss=0.1013, over 5703150.69 frames. ], batch size: 70, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:55:59,102 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-04 18:56:02,390 INFO [train.py:968] (0/2) Epoch 9, batch 21900, giga_loss[loss=0.2775, simple_loss=0.3576, pruned_loss=0.09867, over 28695.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3501, pruned_loss=0.104, over 5693665.13 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3602, pruned_loss=0.1039, over 5681232.84 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.348, pruned_loss=0.1029, over 5697002.57 frames. ], batch size: 284, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:56:09,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=386105.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:56:21,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3593, 1.2581, 4.3003, 3.4790], device='cuda:0'), covar=tensor([0.1543, 0.2560, 0.0319, 0.0788], device='cuda:0'), in_proj_covar=tensor([0.0617, 0.0561, 0.0811, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 18:56:43,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.154e+02 1.004e+03 1.193e+03 1.596e+03 9.745e+03, threshold=2.385e+03, percent-clipped=8.0
+2023-03-04 18:56:43,777 INFO [train.py:968] (0/2) Epoch 9, batch 21950, giga_loss[loss=0.3345, simple_loss=0.3959, pruned_loss=0.1366, over 28046.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3525, pruned_loss=0.1048, over 5695424.10 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3608, pruned_loss=0.1045, over 5684692.47 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3499, pruned_loss=0.1034, over 5695197.58 frames. ], batch size: 412, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:56:55,240 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=386162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:56:55,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3888, 3.1950, 3.1023, 1.5329], device='cuda:0'), covar=tensor([0.0806, 0.1026, 0.0997, 0.1926], device='cuda:0'), in_proj_covar=tensor([0.0967, 0.0905, 0.0802, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 18:56:57,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=386165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:56:59,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1829, 1.5800, 1.2294, 0.9645], device='cuda:0'), covar=tensor([0.2208, 0.2113, 0.2309, 0.1870], device='cuda:0'), in_proj_covar=tensor([0.1235, 0.0920, 0.1087, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 18:57:03,707 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 18:57:13,328 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 18:57:19,566 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=386194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:57:22,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=386197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:57:22,952 INFO [train.py:968] (0/2) Epoch 9, batch 22000, giga_loss[loss=0.291, simple_loss=0.3674, pruned_loss=0.1073, over 29080.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3531, pruned_loss=0.1047, over 5705596.64 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3613, pruned_loss=0.1053, over 5692701.92 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3504, pruned_loss=0.1028, over 5698728.41 frames. ], batch size: 155, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 18:57:35,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=386215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:58:04,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.823e+02 1.032e+03 1.318e+03 1.779e+03 5.407e+03, threshold=2.637e+03, percent-clipped=8.0
+2023-03-04 18:58:05,666 INFO [train.py:968] (0/2) Epoch 9, batch 22050, libri_loss[loss=0.3168, simple_loss=0.3893, pruned_loss=0.1222, over 29478.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3513, pruned_loss=0.103, over 5705372.65 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3611, pruned_loss=0.1054, over 5696065.86 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.349, pruned_loss=0.1014, over 5697175.96 frames. ], batch size: 85, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 18:58:05,998 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=386248.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:58:08,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=386251.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:58:33,082 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=386280.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:58:46,708 INFO [train.py:968] (0/2) Epoch 9, batch 22100, giga_loss[loss=0.277, simple_loss=0.3514, pruned_loss=0.1013, over 28944.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3518, pruned_loss=0.1037, over 5704533.34 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3611, pruned_loss=0.1056, over 5698140.79 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3499, pruned_loss=0.1023, over 5696430.14 frames. ], batch size: 106, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 18:59:18,815 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=386340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:59:21,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=386343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:59:24,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.095e+02 1.084e+03 1.362e+03 1.845e+03 5.712e+03, threshold=2.725e+03, percent-clipped=8.0
+2023-03-04 18:59:25,326 INFO [train.py:968] (0/2) Epoch 9, batch 22150, giga_loss[loss=0.2975, simple_loss=0.3524, pruned_loss=0.1213, over 23728.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3527, pruned_loss=0.1049, over 5693693.00 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3617, pruned_loss=0.1062, over 5691131.20 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3504, pruned_loss=0.1032, over 5693782.62 frames. ], batch size: 705, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 18:59:33,160 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=386358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:59:35,172 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=386361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:59:43,936 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=386372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:59:59,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=386390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:00:05,225 INFO [train.py:968] (0/2) Epoch 9, batch 22200, giga_loss[loss=0.2765, simple_loss=0.3497, pruned_loss=0.1016, over 28888.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3545, pruned_loss=0.1058, over 5698055.96 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.362, pruned_loss=0.1065, over 5693734.67 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3523, pruned_loss=0.1042, over 5696047.87 frames. ], batch size: 199, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:00:41,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.078e+02 1.137e+03 1.500e+03 2.004e+03 3.140e+03, threshold=3.000e+03, percent-clipped=3.0
+2023-03-04 19:00:41,766 INFO [train.py:968] (0/2) Epoch 9, batch 22250, giga_loss[loss=0.2998, simple_loss=0.3644, pruned_loss=0.1176, over 28694.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3585, pruned_loss=0.1079, over 5700573.64 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3625, pruned_loss=0.1069, over 5688849.23 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.356, pruned_loss=0.1062, over 5703406.37 frames. ], batch size: 92, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:01:07,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8810, 1.6251, 1.3477, 1.4029], device='cuda:0'), covar=tensor([0.0703, 0.0785, 0.0979, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0441, 0.0501, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 19:01:19,562 INFO [train.py:968] (0/2) Epoch 9, batch 22300, giga_loss[loss=0.2982, simple_loss=0.37, pruned_loss=0.1132, over 28942.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3619, pruned_loss=0.1104, over 5705272.97 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3629, pruned_loss=0.1073, over 5692512.94 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3595, pruned_loss=0.1087, over 5704546.17 frames. ], batch size: 112, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:01:23,493 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-04 19:01:57,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.271e+02 1.290e+03 1.712e+03 2.251e+03 5.087e+03, threshold=3.423e+03, percent-clipped=9.0
+2023-03-04 19:01:59,256 INFO [train.py:968] (0/2) Epoch 9, batch 22350, giga_loss[loss=0.2924, simple_loss=0.3637, pruned_loss=0.1106, over 28573.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3622, pruned_loss=0.1103, over 5710698.54 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3632, pruned_loss=0.1076, over 5691807.38 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.36, pruned_loss=0.1087, over 5710864.90 frames. ], batch size: 336, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:02:26,459 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-04 19:02:40,601 INFO [train.py:968] (0/2) Epoch 9, batch 22400, giga_loss[loss=0.2904, simple_loss=0.3597, pruned_loss=0.1106, over 28903.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.363, pruned_loss=0.1108, over 5704630.40 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3643, pruned_loss=0.1083, over 5686581.24 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3603, pruned_loss=0.109, over 5709368.81 frames. ], batch size: 112, lr: 3.60e-03, grad_scale: 8.0
+2023-03-04 19:03:10,570 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 19:03:19,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.850e+02 1.145e+03 1.639e+03 2.177e+03 9.040e+03, threshold=3.278e+03, percent-clipped=6.0
+2023-03-04 19:03:19,021 INFO [train.py:968] (0/2) Epoch 9, batch 22450, giga_loss[loss=0.2864, simple_loss=0.3581, pruned_loss=0.1073, over 28956.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3635, pruned_loss=0.1116, over 5712764.85 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3647, pruned_loss=0.1088, over 5692046.58 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.361, pruned_loss=0.1097, over 5712075.06 frames. ], batch size: 145, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:03:30,569 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=386661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:03:51,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=386685.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 19:04:00,730 INFO [train.py:968] (0/2) Epoch 9, batch 22500, giga_loss[loss=0.2894, simple_loss=0.3537, pruned_loss=0.1126, over 29129.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3612, pruned_loss=0.1103, over 5717123.25 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3652, pruned_loss=0.1093, over 5695731.57 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3587, pruned_loss=0.1084, over 5713556.79 frames. ], batch size: 128, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:04:41,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.628e+02 9.957e+02 1.352e+03 1.776e+03 7.960e+03, threshold=2.704e+03, percent-clipped=7.0
+2023-03-04 19:04:41,749 INFO [train.py:968] (0/2) Epoch 9, batch 22550, giga_loss[loss=0.2685, simple_loss=0.3344, pruned_loss=0.1013, over 28667.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3572, pruned_loss=0.1079, over 5715900.69 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3654, pruned_loss=0.1096, over 5691669.93 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3548, pruned_loss=0.1061, over 5717411.15 frames. ], batch size: 85, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:04:43,580 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-04 19:05:07,630 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=386782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:05:21,192 INFO [train.py:968] (0/2) Epoch 9, batch 22600, giga_loss[loss=0.2183, simple_loss=0.2932, pruned_loss=0.07172, over 28755.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3533, pruned_loss=0.1056, over 5714504.18 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3653, pruned_loss=0.1096, over 5693601.86 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3514, pruned_loss=0.1041, over 5714402.43 frames. ], batch size: 92, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:05:37,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3144, 3.0853, 2.9349, 1.4018], device='cuda:0'), covar=tensor([0.0782, 0.0950, 0.0858, 0.2276], device='cuda:0'), in_proj_covar=tensor([0.0968, 0.0907, 0.0806, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 19:05:47,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2829, 1.6071, 1.5968, 1.1762], device='cuda:0'), covar=tensor([0.1449, 0.2021, 0.1267, 0.1410], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0694, 0.0833, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 19:05:59,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.561e+02 1.072e+03 1.336e+03 1.761e+03 3.631e+03, threshold=2.672e+03, percent-clipped=6.0
+2023-03-04 19:05:59,239 INFO [train.py:968] (0/2) Epoch 9, batch 22650, giga_loss[loss=0.271, simple_loss=0.3476, pruned_loss=0.09716, over 28934.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3539, pruned_loss=0.1049, over 5708956.10 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3655, pruned_loss=0.1101, over 5688470.21 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3518, pruned_loss=0.1032, over 5713419.77 frames. ], batch size: 145, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:06:00,004 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=386849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:06:08,559 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 19:06:42,436 INFO [train.py:968] (0/2) Epoch 9, batch 22700, giga_loss[loss=0.2504, simple_loss=0.3403, pruned_loss=0.08022, over 28956.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3549, pruned_loss=0.1043, over 5708472.35 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3655, pruned_loss=0.1101, over 5690211.87 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3532, pruned_loss=0.103, over 5710639.55 frames. ], batch size: 164, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:07:20,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.399e+02 1.182e+03 1.465e+03 1.972e+03 8.295e+03, threshold=2.930e+03, percent-clipped=11.0
+2023-03-04 19:07:20,411 INFO [train.py:968] (0/2) Epoch 9, batch 22750, giga_loss[loss=0.2771, simple_loss=0.3507, pruned_loss=0.1018, over 28939.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3546, pruned_loss=0.1046, over 5711716.79 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3658, pruned_loss=0.1105, over 5685075.49 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3528, pruned_loss=0.103, over 5718808.17 frames. ], batch size: 227, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:07:35,982 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=386965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:07:44,937 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 19:08:03,246 INFO [train.py:968] (0/2) Epoch 9, batch 22800, giga_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1202, over 28001.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3539, pruned_loss=0.1054, over 5713309.40 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3663, pruned_loss=0.1111, over 5687492.73 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3518, pruned_loss=0.1035, over 5717325.67 frames. ], batch size: 412, lr: 3.60e-03, grad_scale: 8.0
+2023-03-04 19:08:33,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:08:42,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.966e+02 1.108e+03 1.357e+03 1.796e+03 4.597e+03, threshold=2.714e+03, percent-clipped=7.0
+2023-03-04 19:08:42,086 INFO [train.py:968] (0/2) Epoch 9, batch 22850, giga_loss[loss=0.2804, simple_loss=0.3376, pruned_loss=0.1116, over 28877.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3529, pruned_loss=0.1063, over 5714913.46 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3671, pruned_loss=0.1116, over 5690159.87 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3503, pruned_loss=0.1043, over 5716150.64 frames. ], batch size: 112, lr: 3.60e-03, grad_scale: 8.0
+2023-03-04 19:08:51,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387060.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 19:09:09,140 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 19:09:23,019 INFO [train.py:968] (0/2) Epoch 9, batch 22900, giga_loss[loss=0.3285, simple_loss=0.3872, pruned_loss=0.1349, over 27920.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3523, pruned_loss=0.1075, over 5717537.25 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3677, pruned_loss=0.1122, over 5694809.85 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3494, pruned_loss=0.1053, over 5714846.74 frames. ], batch size: 412, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:10:01,943 INFO [train.py:968] (0/2) Epoch 9, batch 22950, giga_loss[loss=0.3025, simple_loss=0.3704, pruned_loss=0.1173, over 28666.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3506, pruned_loss=0.1069, over 5718804.43 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3678, pruned_loss=0.1123, over 5695011.78 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3481, pruned_loss=0.1049, over 5716780.76 frames. ], batch size: 262, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:10:02,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.945e+02 1.088e+03 1.335e+03 1.815e+03 7.032e+03, threshold=2.669e+03, percent-clipped=6.0
+2023-03-04 19:10:08,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:10:25,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:10:27,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:10:32,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 19:10:36,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6079, 1.8830, 1.5861, 1.6815], device='cuda:0'), covar=tensor([0.1463, 0.1961, 0.1957, 0.1900], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0730, 0.0662, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 19:10:39,201 INFO [train.py:968] (0/2) Epoch 9, batch 23000, giga_loss[loss=0.2538, simple_loss=0.3274, pruned_loss=0.09009, over 28620.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3496, pruned_loss=0.1069, over 5718713.11 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3682, pruned_loss=0.1129, over 5700324.81 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3464, pruned_loss=0.1045, over 5713137.63 frames. ], batch size: 242, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 19:10:42,691 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387203.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 19:10:45,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387206.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 19:10:49,253 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:10:59,378 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:11:02,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7064, 3.1541, 1.5762, 1.6830], device='cuda:0'), covar=tensor([0.0763, 0.0310, 0.0874, 0.1124], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0500, 0.0329, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:0')
+2023-03-04 19:11:06,584 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387235.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 19:11:15,689 INFO [train.py:968] (0/2) Epoch 9, batch 23050, giga_loss[loss=0.2745, simple_loss=0.3459, pruned_loss=0.1015, over 28701.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3458, pruned_loss=0.1048, over 5727440.73 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3683, pruned_loss=0.1134, over 5706813.13 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3423, pruned_loss=0.1022, over 5718065.77 frames. ], batch size: 307, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 19:11:17,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.040e+02 1.148e+03 1.424e+03 2.185e+03 4.246e+03, threshold=2.849e+03, percent-clipped=10.0
+2023-03-04 19:11:55,451 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=387293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:11:58,532 INFO [train.py:968] (0/2) Epoch 9, batch 23100, giga_loss[loss=0.2317, simple_loss=0.3029, pruned_loss=0.08021, over 28291.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3405, pruned_loss=0.1017, over 5727417.77 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3685, pruned_loss=0.1136, over 5709793.19 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3371, pruned_loss=0.09929, over 5717369.92 frames. ], batch size: 65, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 19:11:59,936 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:01,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:24,761 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:29,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:36,463 INFO [train.py:968] (0/2) Epoch 9, batch 23150, giga_loss[loss=0.3294, simple_loss=0.3951, pruned_loss=0.1318, over 28180.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3412, pruned_loss=0.1018, over 5723437.06 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3691, pruned_loss=0.1142, over 5714041.86 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3374, pruned_loss=0.09891, over 5712025.77 frames. ], batch size: 367, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 19:12:36,643 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=387348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:37,763 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.592e+02 1.166e+03 1.712e+03 2.790e+03 1.739e+04, threshold=3.425e+03, percent-clipped=21.0
+2023-03-04 19:12:42,580 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=387357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:50,708 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:52,582 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:13:14,839 INFO [train.py:968] (0/2) Epoch 9, batch 23200, giga_loss[loss=0.2776, simple_loss=0.3516, pruned_loss=0.1018, over 28872.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3452, pruned_loss=0.1036, over 5708748.63 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3694, pruned_loss=0.1146, over 5700958.10 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3409, pruned_loss=0.1006, over 5711481.69 frames. ], batch size: 186, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:13:15,813 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:13:56,761 INFO [train.py:968] (0/2) Epoch 9, batch 23250, giga_loss[loss=0.2859, simple_loss=0.3598, pruned_loss=0.1061, over 28664.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3488, pruned_loss=0.1051, over 5701289.19 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3692, pruned_loss=0.1146, over 5694135.70 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3451, pruned_loss=0.1025, over 5710703.34 frames. ], batch size: 336, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:13:58,159 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.921e+02 1.149e+03 1.365e+03 1.947e+03 5.057e+03, threshold=2.730e+03, percent-clipped=3.0
+2023-03-04 19:14:17,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5341, 1.4813, 1.2240, 1.2777], device='cuda:0'), covar=tensor([0.0641, 0.0522, 0.0963, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0445, 0.0502, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 19:14:24,824 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:14:25,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0454, 1.8723, 1.5253, 1.6685], device='cuda:0'), covar=tensor([0.0650, 0.0692, 0.0907, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0444, 0.0501, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 19:14:27,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:14:36,058 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=387495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:14:38,505 INFO [train.py:968] (0/2) Epoch 9, batch 23300, libri_loss[loss=0.2968, simple_loss=0.3662, pruned_loss=0.1137, over 29509.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.353, pruned_loss=0.1071, over 5693332.40 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3693, pruned_loss=0.1147, over 5686191.48 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3498, pruned_loss=0.1049, over 5707725.93 frames. ], batch size: 82, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:14:44,163 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=387506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:14:50,932 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:15:18,168 INFO [train.py:968] (0/2) Epoch 9, batch 23350, giga_loss[loss=0.2542, simple_loss=0.331, pruned_loss=0.0887, over 28716.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3551, pruned_loss=0.1078, over 5707269.97 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3693, pruned_loss=0.1149, over 5690399.31 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3523, pruned_loss=0.1057, over 5715255.53 frames. ], batch size: 99, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:15:19,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.801e+02 1.177e+03 1.502e+03 1.873e+03 4.753e+03, threshold=3.003e+03, percent-clipped=9.0
+2023-03-04 19:15:35,918 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-04 19:15:58,053 INFO [train.py:968] (0/2) Epoch 9, batch 23400, giga_loss[loss=0.374, simple_loss=0.4099, pruned_loss=0.1691, over 29080.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3576, pruned_loss=0.1096, over 5711917.11 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3699, pruned_loss=0.1155, over 5691963.36 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3545, pruned_loss=0.1072, over 5717470.71 frames. ], batch size: 164, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:16:41,991 INFO [train.py:968] (0/2) Epoch 9, batch 23450, libri_loss[loss=0.2965, simple_loss=0.3659, pruned_loss=0.1136, over 27455.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3634, pruned_loss=0.1152, over 5702992.54 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3695, pruned_loss=0.1157, over 5698348.09 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3609, pruned_loss=0.113, over 5702305.83 frames. ], batch size: 115, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:16:44,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.041e+02 1.386e+03 1.781e+03 2.525e+03 9.224e+03, threshold=3.562e+03, percent-clipped=16.0
+2023-03-04 19:17:00,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:17:29,133 INFO [train.py:968] (0/2) Epoch 9, batch 23500, giga_loss[loss=0.3585, simple_loss=0.405, pruned_loss=0.156, over 28971.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3698, pruned_loss=0.1203, over 5683258.76 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3697, pruned_loss=0.116, over 5690214.96 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3675, pruned_loss=0.1183, over 5690699.89 frames. ], batch size: 106, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:17:51,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:18:02,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:18:15,845 INFO [train.py:968] (0/2) Epoch 9, batch 23550, giga_loss[loss=0.3235, simple_loss=0.3925, pruned_loss=0.1273, over 28675.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3779, pruned_loss=0.1266, over 5682733.15 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3698, pruned_loss=0.1164, over 5694708.42 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3761, pruned_loss=0.1249, over 5684471.36 frames. ], batch size: 307, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:18:17,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.648e+03 2.221e+03 2.920e+03 7.243e+03, threshold=4.441e+03, percent-clipped=15.0
+2023-03-04 19:19:05,242 INFO [train.py:968] (0/2) Epoch 9, batch 23600, giga_loss[loss=0.3499, simple_loss=0.4075, pruned_loss=0.1461, over 28889.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3829, pruned_loss=0.1309, over 5682811.28 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3701, pruned_loss=0.1167, over 5697730.70 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3814, pruned_loss=0.1294, over 5681193.65 frames. ], batch size: 164, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:19:18,139 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:19:20,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:19:30,158 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-04 19:19:33,322 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1643, 1.3200, 1.0624, 0.9822], device='cuda:0'), covar=tensor([0.0739, 0.0415, 0.1003, 0.0837], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0442, 0.0502, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 19:19:49,119 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:19:53,440 INFO [train.py:968] (0/2) Epoch 9, batch 23650, giga_loss[loss=0.3412, simple_loss=0.4055, pruned_loss=0.1384, over 28943.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3897, pruned_loss=0.1372, over 5679868.11 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3706, pruned_loss=0.1172, over 5701705.43 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3884, pruned_loss=0.1359, over 5674512.07 frames. ], batch size: 199, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:19:56,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.683e+03 2.044e+03 2.634e+03 6.860e+03, threshold=4.089e+03, percent-clipped=5.0
+2023-03-04 19:19:57,469 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-04 19:19:59,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.85 vs. limit=2.0
+2023-03-04 19:20:08,432 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:10,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:11,843 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:17,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:21,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:24,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:36,445 INFO [train.py:968] (0/2) Epoch 9, batch 23700, giga_loss[loss=0.3808, simple_loss=0.4216, pruned_loss=0.17, over 28001.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3915, pruned_loss=0.1391, over 5683091.95 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3704, pruned_loss=0.1173, over 5707438.39 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3912, pruned_loss=0.1385, over 5673226.44 frames. ], batch size: 412, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:20:36,638 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:46,000 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:21:26,540 INFO [train.py:968] (0/2) Epoch 9, batch 23750, giga_loss[loss=0.3296, simple_loss=0.3895, pruned_loss=0.1349, over 28823.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3939, pruned_loss=0.1423, over 5667912.53 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3707, pruned_loss=0.1175, over 5709156.67 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3937, pruned_loss=0.142, over 5658312.88 frames. ], batch size: 199, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:21:29,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.154e+03 1.746e+03 2.268e+03 3.201e+03 7.641e+03, threshold=4.536e+03, percent-clipped=10.0
+2023-03-04 19:22:17,264 INFO [train.py:968] (0/2) Epoch 9, batch 23800, giga_loss[loss=0.3488, simple_loss=0.396, pruned_loss=0.1508, over 28896.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3981, pruned_loss=0.147, over 5656105.62 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3706, pruned_loss=0.1176, over 5704939.89 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3985, pruned_loss=0.1472, over 5651253.08 frames. ], batch size: 227, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:22:19,678 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-388000.pt
+2023-03-04 19:22:22,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=388002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:22:35,258 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:22:39,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:22:48,455 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:22:52,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:23:11,245 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:23:14,928 INFO [train.py:968] (0/2) Epoch 9, batch 23850, giga_loss[loss=0.3678, simple_loss=0.422, pruned_loss=0.1568, over 28869.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4031, pruned_loss=0.1518, over 5647647.24 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3705, pruned_loss=0.1175, over 5705978.19 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.4037, pruned_loss=0.1522, over 5642615.84 frames. ], batch size: 227, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:23:21,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.919e+02 1.657e+03 2.067e+03 2.993e+03 6.008e+03, threshold=4.134e+03, percent-clipped=6.0
+2023-03-04 19:23:26,993 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:24:04,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=388091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:24:11,533 INFO [train.py:968] (0/2) Epoch 9, batch 23900, giga_loss[loss=0.3305, simple_loss=0.3789, pruned_loss=0.141, over 28940.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4033, pruned_loss=0.1516, over 5655461.89 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3707, pruned_loss=0.1178, over 5712006.79 frames. ], giga_tot_loss[loss=0.3549, simple_loss=0.4045, pruned_loss=0.1527, over 5644386.49 frames. ], batch size: 106, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:24:15,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=388101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:24:59,172 INFO [train.py:968] (0/2) Epoch 9, batch 23950, giga_loss[loss=0.3262, simple_loss=0.376, pruned_loss=0.1382, over 28498.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4022, pruned_loss=0.1515, over 5645229.94 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3705, pruned_loss=0.1177, over 5718101.24 frames. ], giga_tot_loss[loss=0.3559, simple_loss=0.4046, pruned_loss=0.1536, over 5628944.50 frames. ], batch size: 71, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:25:01,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.829e+03 2.524e+03 3.658e+03 1.119e+04, threshold=5.048e+03, percent-clipped=21.0
+2023-03-04 19:25:44,390 INFO [train.py:968] (0/2) Epoch 9, batch 24000, giga_loss[loss=0.3276, simple_loss=0.3865, pruned_loss=0.1343, over 28784.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.4008, pruned_loss=0.1505, over 5655578.85 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3705, pruned_loss=0.1179, over 5720186.39 frames. ], giga_tot_loss[loss=0.3543, simple_loss=0.4032, pruned_loss=0.1527, over 5639641.12 frames. ], batch size: 99, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:25:44,395 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 19:25:53,062 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.22, simple_loss=0.3265, pruned_loss=0.0568, over 944034.00 frames. 
+2023-03-04 19:25:53,063 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 19:25:53,485 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-04 19:26:41,022 INFO [train.py:968] (0/2) Epoch 9, batch 24050, giga_loss[loss=0.3678, simple_loss=0.4201, pruned_loss=0.1577, over 28866.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.401, pruned_loss=0.1492, over 5661175.12 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3708, pruned_loss=0.1181, over 5723204.62 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.403, pruned_loss=0.1512, over 5645229.34 frames. ], batch size: 112, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:26:44,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.785e+02 1.681e+03 2.123e+03 2.609e+03 5.420e+03, threshold=4.246e+03, percent-clipped=2.0
+2023-03-04 19:27:34,890 INFO [train.py:968] (0/2) Epoch 9, batch 24100, giga_loss[loss=0.3656, simple_loss=0.3922, pruned_loss=0.1696, over 23153.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4002, pruned_loss=0.1485, over 5637316.21 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3707, pruned_loss=0.1181, over 5716531.17 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.4027, pruned_loss=0.1509, over 5627882.97 frames. ], batch size: 705, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:28:16,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3469, 1.6472, 1.6199, 1.2438], device='cuda:0'), covar=tensor([0.1256, 0.1804, 0.1037, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0699, 0.0830, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 19:28:29,661 INFO [train.py:968] (0/2) Epoch 9, batch 24150, giga_loss[loss=0.3242, simple_loss=0.3858, pruned_loss=0.1313, over 28784.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.4017, pruned_loss=0.1494, over 5623023.75 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3708, pruned_loss=0.1182, over 5706965.07 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.4042, pruned_loss=0.1517, over 5622169.63 frames. ], batch size: 119, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:28:34,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.367e+02 1.771e+03 2.263e+03 3.023e+03 9.777e+03, threshold=4.525e+03, percent-clipped=10.0
+2023-03-04 19:28:40,220 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=388358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:28:59,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=388377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:29:24,876 INFO [train.py:968] (0/2) Epoch 9, batch 24200, giga_loss[loss=0.3186, simple_loss=0.3896, pruned_loss=0.1238, over 28338.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3977, pruned_loss=0.1449, over 5627440.41 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.371, pruned_loss=0.1184, over 5709084.84 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.3996, pruned_loss=0.1468, over 5624396.52 frames. ], batch size: 77, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:30:15,426 INFO [train.py:968] (0/2) Epoch 9, batch 24250, giga_loss[loss=0.2921, simple_loss=0.3639, pruned_loss=0.1102, over 28857.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3954, pruned_loss=0.1416, over 5629028.85 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3714, pruned_loss=0.1188, over 5701845.13 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.397, pruned_loss=0.1431, over 5632365.89 frames. ], batch size: 199, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:30:19,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.092e+02 1.675e+03 2.118e+03 3.030e+03 5.819e+03, threshold=4.236e+03, percent-clipped=5.0
+2023-03-04 19:30:35,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=388466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:30:45,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=388476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:31:03,521 INFO [train.py:968] (0/2) Epoch 9, batch 24300, giga_loss[loss=0.3023, simple_loss=0.3713, pruned_loss=0.1166, over 28836.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3913, pruned_loss=0.1377, over 5653351.14 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3716, pruned_loss=0.1191, over 5707220.36 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3928, pruned_loss=0.1391, over 5649660.29 frames. ], batch size: 199, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:31:25,716 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:31:30,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:31:55,146 INFO [train.py:968] (0/2) Epoch 9, batch 24350, giga_loss[loss=0.3218, simple_loss=0.3835, pruned_loss=0.13, over 28484.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3879, pruned_loss=0.1349, over 5656467.49 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3717, pruned_loss=0.1194, over 5712039.43 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3895, pruned_loss=0.1361, over 5648372.18 frames. ], batch size: 336, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:31:58,213 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5724, 1.5483, 1.5036, 1.4135], device='cuda:0'), covar=tensor([0.0968, 0.1492, 0.1476, 0.1410], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0725, 0.0657, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 19:32:00,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.650e+03 2.067e+03 2.824e+03 6.672e+03, threshold=4.135e+03, percent-clipped=8.0
+2023-03-04 19:32:00,241 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:32:12,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5072, 1.7578, 1.7970, 1.3762], device='cuda:0'), covar=tensor([0.1589, 0.2018, 0.1274, 0.1405], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0698, 0.0830, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 19:32:45,692 INFO [train.py:968] (0/2) Epoch 9, batch 24400, giga_loss[loss=0.3137, simple_loss=0.3792, pruned_loss=0.1242, over 28863.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3866, pruned_loss=0.1341, over 5665466.59 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3717, pruned_loss=0.1194, over 5712359.53 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3879, pruned_loss=0.1352, over 5658294.36 frames. ], batch size: 174, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:32:49,829 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=388603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:32:57,018 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:00,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:05,673 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:07,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:29,400 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:36,949 INFO [train.py:968] (0/2) Epoch 9, batch 24450, giga_loss[loss=0.2821, simple_loss=0.3636, pruned_loss=0.1003, over 28873.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3863, pruned_loss=0.1343, over 5667096.44 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3715, pruned_loss=0.1197, over 5718018.06 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3881, pruned_loss=0.1355, over 5654645.11 frames. ], batch size: 119, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:33:39,880 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:43,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.631e+02 1.571e+03 2.074e+03 2.869e+03 7.878e+03, threshold=4.149e+03, percent-clipped=10.0
+2023-03-04 19:34:28,437 INFO [train.py:968] (0/2) Epoch 9, batch 24500, giga_loss[loss=0.2931, simple_loss=0.3629, pruned_loss=0.1117, over 28137.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3854, pruned_loss=0.1329, over 5674093.56 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3719, pruned_loss=0.12, over 5720641.38 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3868, pruned_loss=0.1338, over 5661060.51 frames. ], batch size: 77, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:35:02,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=388733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:35:17,143 INFO [train.py:968] (0/2) Epoch 9, batch 24550, giga_loss[loss=0.2655, simple_loss=0.349, pruned_loss=0.09101, over 28553.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3841, pruned_loss=0.1297, over 5687083.60 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3715, pruned_loss=0.1199, over 5726698.33 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3859, pruned_loss=0.1308, over 5669762.77 frames. ], batch size: 71, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:35:21,927 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.727e+02 1.442e+03 1.944e+03 2.597e+03 6.749e+03, threshold=3.888e+03, percent-clipped=6.0
+2023-03-04 19:35:44,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3066, 1.7535, 1.5883, 1.1965], device='cuda:0'), covar=tensor([0.1352, 0.2087, 0.1193, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0697, 0.0830, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 19:36:08,444 INFO [train.py:968] (0/2) Epoch 9, batch 24600, giga_loss[loss=0.3483, simple_loss=0.4075, pruned_loss=0.1445, over 28888.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3857, pruned_loss=0.1296, over 5672693.97 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3715, pruned_loss=0.1201, over 5728751.86 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3875, pruned_loss=0.1306, over 5656172.38 frames. ], batch size: 186, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:36:57,017 INFO [train.py:968] (0/2) Epoch 9, batch 24650, giga_loss[loss=0.3641, simple_loss=0.4114, pruned_loss=0.1585, over 28738.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3872, pruned_loss=0.1313, over 5678103.92 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3715, pruned_loss=0.1204, over 5733058.38 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3889, pruned_loss=0.132, over 5660054.18 frames. ], batch size: 262, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:37:03,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.769e+03 2.292e+03 3.649e+03 1.042e+04, threshold=4.583e+03, percent-clipped=22.0
+2023-03-04 19:37:29,148 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:37:30,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:37:46,521 INFO [train.py:968] (0/2) Epoch 9, batch 24700, giga_loss[loss=0.3715, simple_loss=0.3996, pruned_loss=0.1716, over 23531.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3873, pruned_loss=0.1321, over 5662029.11 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3714, pruned_loss=0.1205, over 5728462.61 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3893, pruned_loss=0.133, over 5649851.58 frames. ], batch size: 705, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:37:54,389 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:38:09,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3964, 1.0594, 4.6890, 3.3523], device='cuda:0'), covar=tensor([0.1581, 0.2631, 0.0350, 0.0842], device='cuda:0'), in_proj_covar=tensor([0.0625, 0.0571, 0.0820, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 19:38:34,772 INFO [train.py:968] (0/2) Epoch 9, batch 24750, libri_loss[loss=0.3128, simple_loss=0.3834, pruned_loss=0.1211, over 29215.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3862, pruned_loss=0.1325, over 5660354.26 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3721, pruned_loss=0.1211, over 5731611.90 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3875, pruned_loss=0.1328, over 5646317.30 frames. ], batch size: 97, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:38:39,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.560e+02 1.632e+03 2.201e+03 3.016e+03 8.232e+03, threshold=4.402e+03, percent-clipped=7.0
+2023-03-04 19:39:03,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=388978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:39:20,138 INFO [train.py:968] (0/2) Epoch 9, batch 24800, giga_loss[loss=0.3695, simple_loss=0.4109, pruned_loss=0.164, over 28668.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3838, pruned_loss=0.132, over 5675532.43 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3717, pruned_loss=0.1211, over 5735109.21 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3854, pruned_loss=0.1325, over 5660002.72 frames. ], batch size: 262, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:40:03,881 INFO [train.py:968] (0/2) Epoch 9, batch 24850, giga_loss[loss=0.3163, simple_loss=0.3971, pruned_loss=0.1178, over 28794.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3828, pruned_loss=0.1315, over 5662437.18 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3716, pruned_loss=0.1212, over 5726793.16 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3845, pruned_loss=0.1322, over 5656301.51 frames. ], batch size: 119, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:40:09,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.701e+02 1.631e+03 2.182e+03 2.955e+03 6.889e+03, threshold=4.364e+03, percent-clipped=6.0
+2023-03-04 19:40:48,228 INFO [train.py:968] (0/2) Epoch 9, batch 24900, libri_loss[loss=0.3503, simple_loss=0.4083, pruned_loss=0.1461, over 29508.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3831, pruned_loss=0.1301, over 5684005.85 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3721, pruned_loss=0.1216, over 5733263.05 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3845, pruned_loss=0.1306, over 5671143.11 frames. ], batch size: 82, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:41:07,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-04 19:41:11,145 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=389121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:41:13,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=389124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:41:35,181 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 19:41:36,829 INFO [train.py:968] (0/2) Epoch 9, batch 24950, giga_loss[loss=0.3293, simple_loss=0.3901, pruned_loss=0.1342, over 28764.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3831, pruned_loss=0.13, over 5671746.23 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3726, pruned_loss=0.1221, over 5733069.41 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3839, pruned_loss=0.1301, over 5660641.68 frames. ], batch size: 99, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:41:37,627 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4471, 1.5926, 1.3606, 1.5086], device='cuda:0'), covar=tensor([0.2169, 0.2198, 0.2283, 0.2097], device='cuda:0'), in_proj_covar=tensor([0.1251, 0.0934, 0.1105, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 19:41:42,770 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=389153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:41:43,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.190e+02 1.330e+03 1.720e+03 2.355e+03 6.283e+03, threshold=3.440e+03, percent-clipped=2.0
+2023-03-04 19:42:21,983 INFO [train.py:968] (0/2) Epoch 9, batch 25000, giga_loss[loss=0.2692, simple_loss=0.3458, pruned_loss=0.09631, over 28942.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3828, pruned_loss=0.1295, over 5674798.60 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3731, pruned_loss=0.1225, over 5727174.69 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3834, pruned_loss=0.1294, over 5670189.81 frames. ], batch size: 145, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:42:55,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9535, 1.2584, 0.8312, 0.9611], device='cuda:0'), covar=tensor([0.0865, 0.0484, 0.1402, 0.0975], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0444, 0.0499, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 19:43:07,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5837, 3.7150, 1.5365, 1.6960], device='cuda:0'), covar=tensor([0.0861, 0.0294, 0.0839, 0.1218], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0510, 0.0334, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 19:43:12,804 INFO [train.py:968] (0/2) Epoch 9, batch 25050, giga_loss[loss=0.2995, simple_loss=0.3658, pruned_loss=0.1166, over 28266.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3804, pruned_loss=0.128, over 5683446.83 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3731, pruned_loss=0.1226, over 5728769.21 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.381, pruned_loss=0.128, over 5677651.14 frames. ], batch size: 368, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:43:18,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.811e+02 1.443e+03 1.886e+03 2.485e+03 6.571e+03, threshold=3.773e+03, percent-clipped=8.0
+2023-03-04 19:43:56,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4846, 1.6057, 1.6114, 1.4941], device='cuda:0'), covar=tensor([0.1465, 0.1821, 0.1942, 0.1687], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0732, 0.0666, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 19:43:59,566 INFO [train.py:968] (0/2) Epoch 9, batch 25100, giga_loss[loss=0.3816, simple_loss=0.4174, pruned_loss=0.1729, over 26573.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3797, pruned_loss=0.1282, over 5674518.42 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3734, pruned_loss=0.1229, over 5724634.80 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3801, pruned_loss=0.1281, over 5672385.12 frames. ], batch size: 555, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:44:06,421 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=389306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:44:14,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5260, 1.2640, 4.5967, 3.5421], device='cuda:0'), covar=tensor([0.1515, 0.2441, 0.0345, 0.0727], device='cuda:0'), in_proj_covar=tensor([0.0624, 0.0570, 0.0817, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 19:44:25,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9270, 1.1027, 3.4795, 2.9936], device='cuda:0'), covar=tensor([0.1688, 0.2520, 0.0442, 0.1064], device='cuda:0'), in_proj_covar=tensor([0.0626, 0.0572, 0.0820, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 19:44:28,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8581, 1.1152, 3.3215, 2.9174], device='cuda:0'), covar=tensor([0.1706, 0.2463, 0.0483, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0626, 0.0573, 0.0821, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 19:44:46,255 INFO [train.py:968] (0/2) Epoch 9, batch 25150, giga_loss[loss=0.3089, simple_loss=0.3661, pruned_loss=0.1258, over 28947.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3788, pruned_loss=0.128, over 5681801.82 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3735, pruned_loss=0.1229, over 5717732.49 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3792, pruned_loss=0.1279, over 5685230.88 frames. ], batch size: 106, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:44:50,873 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.529e+03 2.029e+03 2.943e+03 7.226e+03, threshold=4.057e+03, percent-clipped=13.0
+2023-03-04 19:45:28,484 INFO [train.py:968] (0/2) Epoch 9, batch 25200, libri_loss[loss=0.3065, simple_loss=0.3707, pruned_loss=0.1211, over 29521.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3779, pruned_loss=0.1278, over 5689491.48 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3742, pruned_loss=0.1236, over 5720933.61 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3777, pruned_loss=0.1273, over 5688345.98 frames. ], batch size: 80, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:46:16,306 INFO [train.py:968] (0/2) Epoch 9, batch 25250, giga_loss[loss=0.3384, simple_loss=0.393, pruned_loss=0.1419, over 28193.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3747, pruned_loss=0.1259, over 5687814.66 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3743, pruned_loss=0.1238, over 5724678.04 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3744, pruned_loss=0.1255, over 5682505.55 frames. ], batch size: 368, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:46:22,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.789e+02 1.757e+03 2.158e+03 2.680e+03 1.009e+04, threshold=4.315e+03, percent-clipped=10.0
+2023-03-04 19:46:27,787 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-04 19:47:02,943 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-04 19:47:03,176 INFO [train.py:968] (0/2) Epoch 9, batch 25300, giga_loss[loss=0.2816, simple_loss=0.3482, pruned_loss=0.1075, over 28390.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3755, pruned_loss=0.1271, over 5687449.27 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3746, pruned_loss=0.1241, over 5726860.18 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3751, pruned_loss=0.1265, over 5680053.55 frames. ], batch size: 65, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:47:51,607 INFO [train.py:968] (0/2) Epoch 9, batch 25350, giga_loss[loss=0.3684, simple_loss=0.4123, pruned_loss=0.1623, over 28867.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3773, pruned_loss=0.1279, over 5684143.14 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3748, pruned_loss=0.1243, over 5724512.91 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3768, pruned_loss=0.1273, over 5678961.02 frames. ], batch size: 99, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:47:58,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2774, 3.1034, 1.3297, 1.3473], device='cuda:0'), covar=tensor([0.0977, 0.0378, 0.0911, 0.1418], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0508, 0.0333, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 19:47:59,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.787e+03 2.328e+03 3.142e+03 1.165e+04, threshold=4.656e+03, percent-clipped=13.0
+2023-03-04 19:48:34,395 INFO [train.py:968] (0/2) Epoch 9, batch 25400, libri_loss[loss=0.3324, simple_loss=0.3821, pruned_loss=0.1414, over 29601.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3769, pruned_loss=0.1268, over 5694751.01 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3744, pruned_loss=0.1242, over 5729542.58 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.377, pruned_loss=0.1265, over 5684593.51 frames. ], batch size: 74, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:48:35,396 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=389599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:49:19,233 INFO [train.py:968] (0/2) Epoch 9, batch 25450, giga_loss[loss=0.2803, simple_loss=0.3574, pruned_loss=0.1016, over 28813.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3768, pruned_loss=0.1261, over 5683885.15 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3746, pruned_loss=0.1245, over 5716871.64 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3767, pruned_loss=0.1258, over 5685803.80 frames. ], batch size: 119, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:49:25,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.588e+02 1.523e+03 2.231e+03 3.361e+03 1.612e+04, threshold=4.462e+03, percent-clipped=9.0
+2023-03-04 19:49:50,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=389681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:50:03,362 INFO [train.py:968] (0/2) Epoch 9, batch 25500, giga_loss[loss=0.3692, simple_loss=0.408, pruned_loss=0.1652, over 27553.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3773, pruned_loss=0.1268, over 5685614.20 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3742, pruned_loss=0.1245, over 5722271.67 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3777, pruned_loss=0.1266, over 5680890.09 frames. ], batch size: 472, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:50:31,499 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=389729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:50:50,498 INFO [train.py:968] (0/2) Epoch 9, batch 25550, giga_loss[loss=0.3277, simple_loss=0.3891, pruned_loss=0.1331, over 28708.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3797, pruned_loss=0.1289, over 5688250.00 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3745, pruned_loss=0.1247, over 5722939.96 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3799, pruned_loss=0.1286, over 5682676.76 frames. ], batch size: 262, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:50:59,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.406e+02 1.755e+03 2.352e+03 3.112e+03 1.201e+04, threshold=4.704e+03, percent-clipped=9.0
+2023-03-04 19:51:32,466 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9256, 2.8579, 1.8153, 0.8085], device='cuda:0'), covar=tensor([0.4440, 0.2029, 0.2593, 0.4669], device='cuda:0'), in_proj_covar=tensor([0.1515, 0.1451, 0.1476, 0.1236], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 19:51:39,266 INFO [train.py:968] (0/2) Epoch 9, batch 25600, giga_loss[loss=0.3328, simple_loss=0.3898, pruned_loss=0.1379, over 28723.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3818, pruned_loss=0.1319, over 5684482.45 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3747, pruned_loss=0.1251, over 5725393.25 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3819, pruned_loss=0.1315, over 5676837.56 frames. ], batch size: 242, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:52:04,018 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=389824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:52:08,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=389827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:52:31,749 INFO [train.py:968] (0/2) Epoch 9, batch 25650, giga_loss[loss=0.2837, simple_loss=0.3449, pruned_loss=0.1113, over 28784.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3829, pruned_loss=0.1345, over 5678575.30 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3747, pruned_loss=0.1251, over 5726306.29 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.383, pruned_loss=0.1342, over 5671645.38 frames. ], batch size: 99, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 19:52:38,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3199, 2.8631, 1.3864, 1.4070], device='cuda:0'), covar=tensor([0.0912, 0.0401, 0.0826, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0506, 0.0331, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 19:52:41,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.871e+03 2.571e+03 3.270e+03 6.776e+03, threshold=5.142e+03, percent-clipped=9.0
+2023-03-04 19:52:41,617 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=389856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:53:21,973 INFO [train.py:968] (0/2) Epoch 9, batch 25700, libri_loss[loss=0.2852, simple_loss=0.3502, pruned_loss=0.1101, over 29565.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3844, pruned_loss=0.136, over 5689052.66 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3745, pruned_loss=0.125, over 5730965.39 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.385, pruned_loss=0.1362, over 5677938.39 frames. ], batch size: 77, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 19:53:50,927 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.87 vs. limit=5.0
+2023-03-04 19:54:03,812 INFO [train.py:968] (0/2) Epoch 9, batch 25750, giga_loss[loss=0.3236, simple_loss=0.3837, pruned_loss=0.1317, over 28762.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3823, pruned_loss=0.1346, over 5684798.68 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3753, pruned_loss=0.1258, over 5736458.37 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3824, pruned_loss=0.1344, over 5668816.10 frames. ], batch size: 284, lr: 3.58e-03, grad_scale: 2.0
+2023-03-04 19:54:11,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.723e+02 1.673e+03 2.026e+03 2.934e+03 7.083e+03, threshold=4.052e+03, percent-clipped=4.0
+2023-03-04 19:54:12,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=389957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:54:30,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=389974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:54:38,620 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=389983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:54:51,396 INFO [train.py:968] (0/2) Epoch 9, batch 25800, giga_loss[loss=0.292, simple_loss=0.3667, pruned_loss=0.1086, over 28942.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3823, pruned_loss=0.1337, over 5683839.02 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3757, pruned_loss=0.126, over 5735115.31 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3821, pruned_loss=0.1335, over 5671175.62 frames. ], batch size: 164, lr: 3.58e-03, grad_scale: 2.0
+2023-03-04 19:54:52,790 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-390000.pt
+2023-03-04 19:55:35,469 INFO [train.py:968] (0/2) Epoch 9, batch 25850, giga_loss[loss=0.3188, simple_loss=0.383, pruned_loss=0.1273, over 28756.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3823, pruned_loss=0.1328, over 5671760.19 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3759, pruned_loss=0.1263, over 5729181.57 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3821, pruned_loss=0.1325, over 5666496.89 frames. ], batch size: 99, lr: 3.58e-03, grad_scale: 2.0
+2023-03-04 19:55:44,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.772e+02 1.569e+03 2.200e+03 3.172e+03 9.506e+03, threshold=4.400e+03, percent-clipped=14.0
+2023-03-04 19:56:20,938 INFO [train.py:968] (0/2) Epoch 9, batch 25900, giga_loss[loss=0.3097, simple_loss=0.3732, pruned_loss=0.1232, over 28802.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3793, pruned_loss=0.1304, over 5662645.34 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.376, pruned_loss=0.1264, over 5721860.16 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3791, pruned_loss=0.1301, over 5663696.43 frames. ], batch size: 99, lr: 3.58e-03, grad_scale: 2.0
+2023-03-04 19:56:28,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=390104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:56:41,130 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:56:44,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:57:08,675 INFO [train.py:968] (0/2) Epoch 9, batch 25950, giga_loss[loss=0.3105, simple_loss=0.3693, pruned_loss=0.1259, over 28984.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3764, pruned_loss=0.1288, over 5663002.68 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3759, pruned_loss=0.1264, over 5715400.55 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3764, pruned_loss=0.1286, over 5667606.94 frames. ], batch size: 106, lr: 3.58e-03, grad_scale: 2.0
+2023-03-04 19:57:09,480 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:57:15,458 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=390156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:57:16,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.940e+02 1.577e+03 1.965e+03 3.007e+03 8.105e+03, threshold=3.930e+03, percent-clipped=10.0
+2023-03-04 19:58:03,320 INFO [train.py:968] (0/2) Epoch 9, batch 26000, giga_loss[loss=0.4047, simple_loss=0.4309, pruned_loss=0.1892, over 26683.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3775, pruned_loss=0.1304, over 5646964.04 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3759, pruned_loss=0.1264, over 5715400.55 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3775, pruned_loss=0.1303, over 5650547.60 frames. ], batch size: 555, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 19:58:45,133 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:58:45,509 INFO [train.py:968] (0/2) Epoch 9, batch 26050, libri_loss[loss=0.2972, simple_loss=0.364, pruned_loss=0.1152, over 29551.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3811, pruned_loss=0.1331, over 5638555.77 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3763, pruned_loss=0.1268, over 5700831.20 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.381, pruned_loss=0.1329, over 5651699.14 frames. ], batch size: 83, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 19:58:47,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:58:52,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.614e+02 1.479e+03 2.097e+03 2.844e+03 6.175e+03, threshold=4.195e+03, percent-clipped=11.0
+2023-03-04 19:59:13,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:59:23,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-04 19:59:33,526 INFO [train.py:968] (0/2) Epoch 9, batch 26100, giga_loss[loss=0.3322, simple_loss=0.4133, pruned_loss=0.1256, over 29043.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3844, pruned_loss=0.1327, over 5646113.58 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3762, pruned_loss=0.1269, over 5700251.17 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3844, pruned_loss=0.1326, over 5656523.81 frames. ], batch size: 155, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 19:59:43,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2246, 1.5332, 1.2559, 1.0891], device='cuda:0'), covar=tensor([0.2171, 0.2135, 0.2337, 0.1715], device='cuda:0'), in_proj_covar=tensor([0.1249, 0.0930, 0.1099, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 19:59:48,155 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=390315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:00:00,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7544, 1.8008, 1.2383, 1.4428], device='cuda:0'), covar=tensor([0.0719, 0.0576, 0.1005, 0.0980], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0438, 0.0497, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 20:00:05,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=390332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:00:20,418 INFO [train.py:968] (0/2) Epoch 9, batch 26150, giga_loss[loss=0.3401, simple_loss=0.3991, pruned_loss=0.1406, over 28791.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.388, pruned_loss=0.1335, over 5651805.99 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3764, pruned_loss=0.1272, over 5702374.50 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3879, pruned_loss=0.1332, over 5657578.79 frames. ], batch size: 284, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:00:29,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.209e+02 1.411e+03 1.988e+03 2.613e+03 9.405e+03, threshold=3.976e+03, percent-clipped=9.0
+2023-03-04 20:00:33,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=390358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:01:10,667 INFO [train.py:968] (0/2) Epoch 9, batch 26200, giga_loss[loss=0.3531, simple_loss=0.4132, pruned_loss=0.1465, over 28551.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3894, pruned_loss=0.1348, over 5640744.61 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3765, pruned_loss=0.1274, over 5694396.02 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3895, pruned_loss=0.1344, over 5652042.88 frames. ], batch size: 307, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:01:55,252 INFO [train.py:968] (0/2) Epoch 9, batch 26250, giga_loss[loss=0.3031, simple_loss=0.3745, pruned_loss=0.1159, over 28903.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3904, pruned_loss=0.1363, over 5645747.10 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.377, pruned_loss=0.1279, over 5695993.90 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3905, pruned_loss=0.1358, over 5652051.73 frames. ], batch size: 213, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:02:03,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9958, 1.7378, 1.3847, 1.4396], device='cuda:0'), covar=tensor([0.0656, 0.0655, 0.0951, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.0346, 0.0440, 0.0499, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 20:02:03,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.547e+02 1.686e+03 2.408e+03 3.420e+03 7.206e+03, threshold=4.816e+03, percent-clipped=13.0
+2023-03-04 20:02:22,416 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:02:26,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:02:44,340 INFO [train.py:968] (0/2) Epoch 9, batch 26300, libri_loss[loss=0.3353, simple_loss=0.3885, pruned_loss=0.1411, over 25853.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3905, pruned_loss=0.1374, over 5632168.99 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1279, over 5687658.82 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3909, pruned_loss=0.1373, over 5642910.90 frames. ], batch size: 136, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:02:47,495 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:02:49,843 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:02:52,622 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:03:06,898 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6808, 1.7039, 1.5842, 1.6028], device='cuda:0'), covar=tensor([0.1125, 0.1776, 0.1734, 0.1613], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0732, 0.0657, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 20:03:16,813 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=390531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:03:18,804 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:03:32,747 INFO [train.py:968] (0/2) Epoch 9, batch 26350, libri_loss[loss=0.2268, simple_loss=0.2973, pruned_loss=0.07815, over 28564.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3885, pruned_loss=0.1367, over 5627867.83 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3766, pruned_loss=0.1276, over 5681377.94 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3894, pruned_loss=0.137, over 5639991.85 frames. ], batch size: 63, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:03:39,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.626e+03 2.246e+03 3.366e+03 7.686e+03, threshold=4.492e+03, percent-clipped=8.0
+2023-03-04 20:04:16,982 INFO [train.py:968] (0/2) Epoch 9, batch 26400, libri_loss[loss=0.3826, simple_loss=0.4234, pruned_loss=0.1709, over 25768.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.386, pruned_loss=0.1353, over 5638240.90 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.377, pruned_loss=0.1279, over 5683012.52 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3867, pruned_loss=0.1355, over 5645553.91 frames. ], batch size: 136, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:04:47,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1608, 1.5018, 1.4945, 1.0786], device='cuda:0'), covar=tensor([0.1287, 0.2154, 0.1089, 0.1357], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0710, 0.0841, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 20:05:08,370 INFO [train.py:968] (0/2) Epoch 9, batch 26450, giga_loss[loss=0.2985, simple_loss=0.3575, pruned_loss=0.1197, over 28750.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3854, pruned_loss=0.136, over 5629268.70 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3772, pruned_loss=0.1281, over 5676096.21 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3858, pruned_loss=0.1361, over 5641362.42 frames. ], batch size: 99, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:05:18,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.751e+03 2.312e+03 2.967e+03 1.090e+04, threshold=4.624e+03, percent-clipped=9.0
+2023-03-04 20:05:35,885 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:05:38,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:05:49,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=390690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:05:57,374 INFO [train.py:968] (0/2) Epoch 9, batch 26500, giga_loss[loss=0.3831, simple_loss=0.423, pruned_loss=0.1716, over 27591.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3862, pruned_loss=0.1369, over 5634377.52 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3773, pruned_loss=0.1282, over 5679517.72 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3865, pruned_loss=0.137, over 5640275.72 frames. ], batch size: 472, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:06:03,828 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:06:40,350 INFO [train.py:968] (0/2) Epoch 9, batch 26550, libri_loss[loss=0.2684, simple_loss=0.3342, pruned_loss=0.1012, over 29580.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3842, pruned_loss=0.1355, over 5643944.56 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3773, pruned_loss=0.1282, over 5675455.58 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3847, pruned_loss=0.1358, over 5650342.94 frames. ], batch size: 74, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:06:49,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.300e+02 1.518e+03 2.049e+03 2.730e+03 6.062e+03, threshold=4.098e+03, percent-clipped=1.0
+2023-03-04 20:07:25,424 INFO [train.py:968] (0/2) Epoch 9, batch 26600, giga_loss[loss=0.3718, simple_loss=0.4134, pruned_loss=0.1651, over 27672.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3819, pruned_loss=0.1341, over 5654511.72 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3776, pruned_loss=0.1284, over 5672410.40 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3821, pruned_loss=0.1343, over 5661642.60 frames. ], batch size: 472, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:07:30,796 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=390804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:07:59,337 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:08:02,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:08:13,033 INFO [train.py:968] (0/2) Epoch 9, batch 26650, giga_loss[loss=0.3266, simple_loss=0.3821, pruned_loss=0.1355, over 28969.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.382, pruned_loss=0.1346, over 5660277.94 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3773, pruned_loss=0.1283, over 5676968.67 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3826, pruned_loss=0.135, over 5661586.15 frames. ], batch size: 112, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:08:22,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.892e+02 1.651e+03 2.226e+03 2.700e+03 8.527e+03, threshold=4.451e+03, percent-clipped=5.0
+2023-03-04 20:08:28,006 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:08:56,640 INFO [train.py:968] (0/2) Epoch 9, batch 26700, giga_loss[loss=0.2948, simple_loss=0.3717, pruned_loss=0.1089, over 28843.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.383, pruned_loss=0.1341, over 5663742.01 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3776, pruned_loss=0.1283, over 5682041.38 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3834, pruned_loss=0.1345, over 5659932.57 frames. ], batch size: 199, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:09:45,978 INFO [train.py:968] (0/2) Epoch 9, batch 26750, giga_loss[loss=0.3265, simple_loss=0.3875, pruned_loss=0.1327, over 28896.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3838, pruned_loss=0.134, over 5662919.33 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3774, pruned_loss=0.1281, over 5686788.74 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3843, pruned_loss=0.1346, over 5655784.13 frames. ], batch size: 112, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:09:55,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 20:09:57,365 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.325e+02 1.329e+03 1.818e+03 2.623e+03 6.348e+03, threshold=3.636e+03, percent-clipped=3.0
+2023-03-04 20:10:22,701 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=390983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:10:35,162 INFO [train.py:968] (0/2) Epoch 9, batch 26800, giga_loss[loss=0.3122, simple_loss=0.3926, pruned_loss=0.1159, over 29004.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3835, pruned_loss=0.1347, over 5667732.60 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3771, pruned_loss=0.128, over 5691747.48 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3844, pruned_loss=0.1354, over 5656661.79 frames. ], batch size: 106, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:11:15,066 INFO [train.py:968] (0/2) Epoch 9, batch 26850, libri_loss[loss=0.3093, simple_loss=0.3757, pruned_loss=0.1215, over 27981.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3845, pruned_loss=0.1324, over 5673430.98 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3773, pruned_loss=0.1282, over 5686293.95 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3854, pruned_loss=0.1331, over 5667822.31 frames. ], batch size: 116, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:11:24,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.043e+02 1.422e+03 1.859e+03 2.960e+03 7.573e+03, threshold=3.718e+03, percent-clipped=11.0
+2023-03-04 20:11:26,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:12:00,233 INFO [train.py:968] (0/2) Epoch 9, batch 26900, giga_loss[loss=0.3201, simple_loss=0.386, pruned_loss=0.1271, over 28491.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3847, pruned_loss=0.1311, over 5665706.35 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3768, pruned_loss=0.1281, over 5678593.73 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3862, pruned_loss=0.132, over 5667925.65 frames. ], batch size: 85, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:12:07,022 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:12:13,431 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:12:17,110 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:12:41,820 INFO [train.py:968] (0/2) Epoch 9, batch 26950, giga_loss[loss=0.3444, simple_loss=0.4051, pruned_loss=0.1419, over 28707.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3862, pruned_loss=0.1307, over 5652961.47 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.377, pruned_loss=0.1281, over 5666947.01 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3875, pruned_loss=0.1314, over 5664361.77 frames. ], batch size: 242, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:12:50,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.333e+02 1.398e+03 1.984e+03 2.735e+03 1.029e+04, threshold=3.969e+03, percent-clipped=15.0
+2023-03-04 20:13:09,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:13:28,319 INFO [train.py:968] (0/2) Epoch 9, batch 27000, giga_loss[loss=0.3125, simple_loss=0.3777, pruned_loss=0.1237, over 28745.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3889, pruned_loss=0.1337, over 5662956.08 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3765, pruned_loss=0.1278, over 5674071.17 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3906, pruned_loss=0.1346, over 5665484.26 frames. ], batch size: 99, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:13:28,323 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 20:13:37,447 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2172, simple_loss=0.3218, pruned_loss=0.05633, over 944034.00 frames. 
+2023-03-04 20:13:37,447 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 20:14:10,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2511, 0.8928, 0.8135, 1.3858], device='cuda:0'), covar=tensor([0.0705, 0.0354, 0.0342, 0.0774], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0081], device='cuda:0')
+2023-03-04 20:14:13,538 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:14:29,440 INFO [train.py:968] (0/2) Epoch 9, batch 27050, giga_loss[loss=0.2956, simple_loss=0.3669, pruned_loss=0.1122, over 29056.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3909, pruned_loss=0.136, over 5670328.43 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3765, pruned_loss=0.1279, over 5675365.81 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3924, pruned_loss=0.1367, over 5671161.28 frames. ], batch size: 155, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:14:43,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.783e+03 2.301e+03 3.416e+03 6.858e+03, threshold=4.602e+03, percent-clipped=15.0
+2023-03-04 20:14:56,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5241, 1.6781, 1.4346, 1.6885], device='cuda:0'), covar=tensor([0.0720, 0.0289, 0.0294, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0081], device='cuda:0')
+2023-03-04 20:15:22,092 INFO [train.py:968] (0/2) Epoch 9, batch 27100, giga_loss[loss=0.331, simple_loss=0.3947, pruned_loss=0.1336, over 29006.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3917, pruned_loss=0.1375, over 5668505.52 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3769, pruned_loss=0.1283, over 5677111.14 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3926, pruned_loss=0.1378, over 5667582.77 frames. ], batch size: 164, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:15:48,060 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:15:48,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 20:15:51,534 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:16:01,355 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:16:12,996 INFO [train.py:968] (0/2) Epoch 9, batch 27150, libri_loss[loss=0.2729, simple_loss=0.3363, pruned_loss=0.1047, over 29373.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3896, pruned_loss=0.1356, over 5674686.47 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3766, pruned_loss=0.1282, over 5680540.78 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3908, pruned_loss=0.136, over 5670584.77 frames. ], batch size: 71, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:16:18,592 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:16:22,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:16:23,245 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.094e+02 1.550e+03 2.207e+03 2.796e+03 5.527e+03, threshold=4.415e+03, percent-clipped=3.0
+2023-03-04 20:16:56,975 INFO [train.py:968] (0/2) Epoch 9, batch 27200, giga_loss[loss=0.3071, simple_loss=0.3932, pruned_loss=0.1106, over 28934.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3884, pruned_loss=0.1332, over 5669888.95 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3759, pruned_loss=0.1277, over 5684243.05 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3902, pruned_loss=0.1341, over 5663002.82 frames. ], batch size: 145, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:17:35,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:17:46,403 INFO [train.py:968] (0/2) Epoch 9, batch 27250, giga_loss[loss=0.3665, simple_loss=0.4213, pruned_loss=0.1558, over 28900.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3892, pruned_loss=0.1331, over 5665949.85 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3756, pruned_loss=0.1274, over 5685923.45 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.391, pruned_loss=0.1341, over 5658775.90 frames. ], batch size: 106, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:17:52,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3937, 1.7519, 1.3128, 1.4681], device='cuda:0'), covar=tensor([0.0724, 0.0297, 0.0321, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0082], device='cuda:0')
+2023-03-04 20:17:55,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.936e+02 1.502e+03 1.972e+03 2.535e+03 6.144e+03, threshold=3.943e+03, percent-clipped=4.0
+2023-03-04 20:18:17,864 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 20:18:18,184 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:18:23,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3555, 1.4112, 1.5099, 1.2997], device='cuda:0'), covar=tensor([0.1287, 0.1559, 0.1758, 0.1668], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0734, 0.0660, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 20:18:24,854 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391486.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 20:18:29,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:18:34,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:18:37,681 INFO [train.py:968] (0/2) Epoch 9, batch 27300, giga_loss[loss=0.3991, simple_loss=0.4374, pruned_loss=0.1804, over 27965.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3913, pruned_loss=0.1349, over 5663799.19 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3759, pruned_loss=0.1276, over 5689697.83 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3928, pruned_loss=0.1357, over 5654183.11 frames. ], batch size: 412, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:18:41,562 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:18:44,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:19:10,878 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:19:22,928 INFO [train.py:968] (0/2) Epoch 9, batch 27350, giga_loss[loss=0.3386, simple_loss=0.3987, pruned_loss=0.1392, over 28615.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3902, pruned_loss=0.1347, over 5665480.18 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3762, pruned_loss=0.1279, over 5693103.26 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3914, pruned_loss=0.1351, over 5654225.68 frames. ], batch size: 262, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:19:32,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.590e+03 1.971e+03 2.688e+03 1.387e+04, threshold=3.942e+03, percent-clipped=8.0
+2023-03-04 20:19:52,168 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:19:55,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:19:58,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6164, 1.5880, 1.2558, 1.1833], device='cuda:0'), covar=tensor([0.0698, 0.0551, 0.1011, 0.0958], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0447, 0.0503, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 20:20:11,371 INFO [train.py:968] (0/2) Epoch 9, batch 27400, giga_loss[loss=0.2954, simple_loss=0.3626, pruned_loss=0.1141, over 28574.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3895, pruned_loss=0.1352, over 5666894.43 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3767, pruned_loss=0.1283, over 5694231.89 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3903, pruned_loss=0.1354, over 5656574.26 frames. ], batch size: 307, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:20:21,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:23,508 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:34,070 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:35,970 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:42,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:46,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:47,039 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:50,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:53,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3056, 1.9530, 1.4276, 0.5683], device='cuda:0'), covar=tensor([0.2947, 0.1629, 0.2575, 0.3481], device='cuda:0'), in_proj_covar=tensor([0.1511, 0.1442, 0.1467, 0.1233], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 20:20:57,663 INFO [train.py:968] (0/2) Epoch 9, batch 27450, giga_loss[loss=0.3481, simple_loss=0.3957, pruned_loss=0.1502, over 28932.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3879, pruned_loss=0.1347, over 5667834.99 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3772, pruned_loss=0.1287, over 5690735.38 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3885, pruned_loss=0.1346, over 5662155.32 frames. ], batch size: 227, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:21:03,944 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:21:08,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.578e+03 2.158e+03 3.084e+03 1.430e+04, threshold=4.316e+03, percent-clipped=12.0
+2023-03-04 20:21:15,185 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:21:19,916 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:21:49,059 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-04 20:21:49,263 INFO [train.py:968] (0/2) Epoch 9, batch 27500, giga_loss[loss=0.3505, simple_loss=0.4024, pruned_loss=0.1493, over 28728.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3849, pruned_loss=0.1329, over 5666657.11 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3774, pruned_loss=0.1288, over 5693126.74 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3853, pruned_loss=0.1328, over 5659673.35 frames. ], batch size: 284, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:22:01,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:22:12,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3501, 3.0288, 1.4685, 1.4429], device='cuda:0'), covar=tensor([0.0887, 0.0306, 0.0819, 0.1274], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0505, 0.0330, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 20:22:36,043 INFO [train.py:968] (0/2) Epoch 9, batch 27550, giga_loss[loss=0.3299, simple_loss=0.3616, pruned_loss=0.1491, over 23273.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3842, pruned_loss=0.1334, over 5666542.68 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3774, pruned_loss=0.1288, over 5697895.15 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3846, pruned_loss=0.1334, over 5656256.86 frames. ], batch size: 705, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:22:38,135 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:22:40,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:22:46,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.814e+02 1.453e+03 1.841e+03 2.718e+03 1.199e+04, threshold=3.681e+03, percent-clipped=14.0
+2023-03-04 20:22:50,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0382, 1.0375, 3.8440, 3.0579], device='cuda:0'), covar=tensor([0.1741, 0.2602, 0.0419, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0638, 0.0579, 0.0839, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 20:22:50,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6987, 1.6202, 1.1920, 1.2051], device='cuda:0'), covar=tensor([0.0815, 0.0727, 0.1113, 0.1138], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0448, 0.0503, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 20:22:57,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4441, 1.8080, 1.3919, 1.3247], device='cuda:0'), covar=tensor([0.1782, 0.1400, 0.1625, 0.1709], device='cuda:0'), in_proj_covar=tensor([0.1641, 0.1526, 0.1501, 0.1602], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 20:23:06,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:23:20,082 INFO [train.py:968] (0/2) Epoch 9, batch 27600, giga_loss[loss=0.336, simple_loss=0.3966, pruned_loss=0.1377, over 28957.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3834, pruned_loss=0.1333, over 5665197.42 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3776, pruned_loss=0.1289, over 5699782.54 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3837, pruned_loss=0.1334, over 5654826.02 frames. ], batch size: 164, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:23:40,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2919, 1.6702, 1.2751, 1.4623], device='cuda:0'), covar=tensor([0.0766, 0.0363, 0.0341, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0082], device='cuda:0')
+2023-03-04 20:24:05,274 INFO [train.py:968] (0/2) Epoch 9, batch 27650, giga_loss[loss=0.328, simple_loss=0.3826, pruned_loss=0.1367, over 26562.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3805, pruned_loss=0.1299, over 5664273.17 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3774, pruned_loss=0.1288, over 5694907.14 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3809, pruned_loss=0.1301, over 5659073.90 frames. ], batch size: 555, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:24:11,513 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:24:14,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:24:17,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.479e+02 1.422e+03 1.866e+03 2.592e+03 8.876e+03, threshold=3.733e+03, percent-clipped=7.0
+2023-03-04 20:24:19,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391861.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 20:24:39,837 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:24:52,659 INFO [train.py:968] (0/2) Epoch 9, batch 27700, giga_loss[loss=0.2961, simple_loss=0.3696, pruned_loss=0.1113, over 29121.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3781, pruned_loss=0.1275, over 5660941.49 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3778, pruned_loss=0.1288, over 5695208.81 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3781, pruned_loss=0.1275, over 5656275.37 frames. ], batch size: 155, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:25:09,961 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:25:40,768 INFO [train.py:968] (0/2) Epoch 9, batch 27750, giga_loss[loss=0.3357, simple_loss=0.3915, pruned_loss=0.1399, over 28532.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3775, pruned_loss=0.127, over 5659719.27 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3783, pruned_loss=0.1292, over 5694805.17 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3771, pruned_loss=0.1267, over 5655621.12 frames. ], batch size: 78, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:25:54,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.298e+03 1.700e+03 2.557e+03 4.591e+03, threshold=3.399e+03, percent-clipped=7.0
+2023-03-04 20:26:35,465 INFO [train.py:968] (0/2) Epoch 9, batch 27800, giga_loss[loss=0.261, simple_loss=0.3254, pruned_loss=0.09825, over 28838.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3746, pruned_loss=0.1263, over 5654576.23 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.378, pruned_loss=0.129, over 5695824.14 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3745, pruned_loss=0.1262, over 5650237.24 frames. ], batch size: 99, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:26:37,049 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-392000.pt
+2023-03-04 20:26:41,851 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=392004.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 20:26:44,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=392007.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 20:27:11,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=392036.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 20:27:23,088 INFO [train.py:968] (0/2) Epoch 9, batch 27850, giga_loss[loss=0.3232, simple_loss=0.3794, pruned_loss=0.1335, over 28646.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3737, pruned_loss=0.1263, over 5649791.73 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3784, pruned_loss=0.1293, over 5688213.77 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3731, pruned_loss=0.1258, over 5650037.78 frames. ], batch size: 92, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:27:35,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.372e+02 1.608e+03 2.148e+03 3.076e+03 6.925e+03, threshold=4.295e+03, percent-clipped=15.0
+2023-03-04 20:28:03,001 INFO [train.py:968] (0/2) Epoch 9, batch 27900, giga_loss[loss=0.3072, simple_loss=0.3764, pruned_loss=0.119, over 28635.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3755, pruned_loss=0.1265, over 5659198.06 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.378, pruned_loss=0.1287, over 5692455.43 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3752, pruned_loss=0.1266, over 5654143.28 frames. ], batch size: 242, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:28:03,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=392098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:28:47,953 INFO [train.py:968] (0/2) Epoch 9, batch 27950, giga_loss[loss=0.2868, simple_loss=0.3583, pruned_loss=0.1076, over 29033.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3765, pruned_loss=0.127, over 5653465.84 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3773, pruned_loss=0.1282, over 5700902.18 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3769, pruned_loss=0.1274, over 5640575.82 frames. ], batch size: 136, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:28:58,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.730e+02 1.665e+03 2.084e+03 2.985e+03 1.004e+04, threshold=4.168e+03, percent-clipped=5.0
+2023-03-04 20:29:28,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2095, 0.8900, 0.8777, 1.3069], device='cuda:0'), covar=tensor([0.0761, 0.0348, 0.0331, 0.0788], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0082], device='cuda:0')
+2023-03-04 20:29:32,696 INFO [train.py:968] (0/2) Epoch 9, batch 28000, giga_loss[loss=0.3011, simple_loss=0.3665, pruned_loss=0.1179, over 29003.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3769, pruned_loss=0.127, over 5656663.30 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3778, pruned_loss=0.1286, over 5697581.71 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3767, pruned_loss=0.127, over 5648427.59 frames. ], batch size: 213, lr: 3.57e-03, grad_scale: 8.0
+2023-03-04 20:29:38,759 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-04 20:29:43,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3334, 1.5102, 1.1770, 1.1592], device='cuda:0'), covar=tensor([0.1238, 0.1195, 0.1208, 0.1220], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1523, 0.1490, 0.1597], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 20:29:56,777 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.89 vs. limit=5.0
+2023-03-04 20:30:19,373 INFO [train.py:968] (0/2) Epoch 9, batch 28050, giga_loss[loss=0.3408, simple_loss=0.3758, pruned_loss=0.1529, over 23488.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3772, pruned_loss=0.1278, over 5649050.37 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.378, pruned_loss=0.1287, over 5699384.09 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3769, pruned_loss=0.1277, over 5640196.36 frames. ], batch size: 705, lr: 3.57e-03, grad_scale: 8.0
+2023-03-04 20:30:25,145 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-04 20:30:31,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.043e+02 1.569e+03 1.947e+03 2.781e+03 4.249e+03, threshold=3.895e+03, percent-clipped=1.0
+2023-03-04 20:30:54,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=392290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:31:00,129 INFO [train.py:968] (0/2) Epoch 9, batch 28100, giga_loss[loss=0.3352, simple_loss=0.3883, pruned_loss=0.141, over 28006.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3789, pruned_loss=0.1296, over 5646021.08 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3789, pruned_loss=0.1295, over 5695505.78 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3778, pruned_loss=0.1288, over 5641113.59 frames. ], batch size: 412, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:31:42,321 INFO [train.py:968] (0/2) Epoch 9, batch 28150, giga_loss[loss=0.3453, simple_loss=0.4007, pruned_loss=0.145, over 28429.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3815, pruned_loss=0.1313, over 5654532.36 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3786, pruned_loss=0.1294, over 5697630.21 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3808, pruned_loss=0.1307, over 5646897.53 frames. ], batch size: 65, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:31:57,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.816e+02 1.529e+03 1.971e+03 2.837e+03 7.922e+03, threshold=3.941e+03, percent-clipped=18.0
+2023-03-04 20:32:11,005 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-04 20:32:31,548 INFO [train.py:968] (0/2) Epoch 9, batch 28200, giga_loss[loss=0.3212, simple_loss=0.3861, pruned_loss=0.1282, over 29043.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3835, pruned_loss=0.1328, over 5651605.19 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3786, pruned_loss=0.1293, over 5699831.74 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.383, pruned_loss=0.1325, over 5643164.27 frames. ], batch size: 136, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:32:56,635 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 20:33:02,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=392433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:33:04,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=392436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:33:15,433 INFO [train.py:968] (0/2) Epoch 9, batch 28250, giga_loss[loss=0.3736, simple_loss=0.4245, pruned_loss=0.1614, over 28586.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3858, pruned_loss=0.135, over 5631582.06 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3787, pruned_loss=0.1293, over 5678099.73 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3856, pruned_loss=0.1349, over 5641444.76 frames. ], batch size: 307, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:33:29,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.773e+03 2.333e+03 3.124e+03 1.411e+04, threshold=4.667e+03, percent-clipped=16.0
+2023-03-04 20:33:30,868 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=392465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:33:35,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 20:33:38,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=392473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:34:05,962 INFO [train.py:968] (0/2) Epoch 9, batch 28300, giga_loss[loss=0.3341, simple_loss=0.391, pruned_loss=0.1386, over 28950.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3859, pruned_loss=0.1355, over 5639419.92 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3788, pruned_loss=0.1295, over 5681468.70 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3858, pruned_loss=0.1354, over 5643709.04 frames. ], batch size: 128, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:34:51,717 INFO [train.py:968] (0/2) Epoch 9, batch 28350, giga_loss[loss=0.3203, simple_loss=0.3856, pruned_loss=0.1275, over 28828.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3857, pruned_loss=0.1338, over 5642870.22 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.379, pruned_loss=0.1296, over 5683838.62 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3856, pruned_loss=0.1338, over 5643141.16 frames. ], batch size: 227, lr: 3.57e-03, grad_scale: 1.0
+2023-03-04 20:35:08,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.688e+03 2.432e+03 3.483e+03 1.842e+04, threshold=4.864e+03, percent-clipped=10.0
+2023-03-04 20:35:10,926 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 20:35:40,413 INFO [train.py:968] (0/2) Epoch 9, batch 28400, giga_loss[loss=0.3193, simple_loss=0.3902, pruned_loss=0.1242, over 28664.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3859, pruned_loss=0.1345, over 5618829.43 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3788, pruned_loss=0.1295, over 5667525.10 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3862, pruned_loss=0.1346, over 5633200.31 frames. ], batch size: 243, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:35:56,300 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=392616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:36:00,968 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=392619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:36:27,509 INFO [train.py:968] (0/2) Epoch 9, batch 28450, giga_loss[loss=0.3695, simple_loss=0.412, pruned_loss=0.1635, over 27592.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3854, pruned_loss=0.1347, over 5621539.07 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3788, pruned_loss=0.1296, over 5670292.93 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3857, pruned_loss=0.1349, over 5629434.84 frames. ], batch size: 472, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:36:28,013 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=392648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:36:45,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.650e+02 1.544e+03 2.034e+03 2.648e+03 9.029e+03, threshold=4.068e+03, percent-clipped=1.0
+2023-03-04 20:36:54,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2946, 4.1319, 3.9344, 1.8980], device='cuda:0'), covar=tensor([0.0544, 0.0709, 0.0727, 0.1960], device='cuda:0'), in_proj_covar=tensor([0.1010, 0.0959, 0.0840, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 20:37:00,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3899, 1.5367, 1.2191, 1.2842], device='cuda:0'), covar=tensor([0.1340, 0.1199, 0.1208, 0.1230], device='cuda:0'), in_proj_covar=tensor([0.1623, 0.1518, 0.1481, 0.1590], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 20:37:28,655 INFO [train.py:968] (0/2) Epoch 9, batch 28500, giga_loss[loss=0.3643, simple_loss=0.4078, pruned_loss=0.1604, over 28571.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3845, pruned_loss=0.1348, over 5622144.64 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3789, pruned_loss=0.1296, over 5672860.07 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3847, pruned_loss=0.135, over 5625743.55 frames. ], batch size: 336, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:37:44,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3463, 1.6882, 1.4166, 1.5072], device='cuda:0'), covar=tensor([0.0749, 0.0280, 0.0296, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0082], device='cuda:0')
+2023-03-04 20:38:00,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2888, 4.1283, 3.9164, 1.8209], device='cuda:0'), covar=tensor([0.0588, 0.0761, 0.0811, 0.2065], device='cuda:0'), in_proj_covar=tensor([0.1015, 0.0964, 0.0844, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 20:38:16,311 INFO [train.py:968] (0/2) Epoch 9, batch 28550, giga_loss[loss=0.3433, simple_loss=0.4032, pruned_loss=0.1417, over 29085.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3839, pruned_loss=0.135, over 5624587.60 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3793, pruned_loss=0.1297, over 5674923.72 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3839, pruned_loss=0.1351, over 5623842.35 frames. ], batch size: 155, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:38:31,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.689e+02 1.616e+03 2.075e+03 3.239e+03 7.234e+03, threshold=4.149e+03, percent-clipped=15.0
+2023-03-04 20:38:40,262 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=392774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:38:52,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9554, 1.1330, 1.0703, 0.7616], device='cuda:0'), covar=tensor([0.1395, 0.1548, 0.0866, 0.1455], device='cuda:0'), in_proj_covar=tensor([0.1625, 0.1525, 0.1480, 0.1594], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 20:38:53,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4109, 1.6685, 1.4277, 1.5246], device='cuda:0'), covar=tensor([0.0706, 0.0330, 0.0290, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0082], device='cuda:0')
+2023-03-04 20:39:00,261 INFO [train.py:968] (0/2) Epoch 9, batch 28600, giga_loss[loss=0.2771, simple_loss=0.3491, pruned_loss=0.1026, over 28897.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3818, pruned_loss=0.1335, over 5648477.54 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.379, pruned_loss=0.1295, over 5680309.31 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3821, pruned_loss=0.1339, over 5641903.05 frames. ], batch size: 112, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:39:35,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1258, 3.2861, 1.2912, 1.2857], device='cuda:0'), covar=tensor([0.1189, 0.0420, 0.1013, 0.1621], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0507, 0.0332, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 20:39:46,255 INFO [train.py:968] (0/2) Epoch 9, batch 28650, giga_loss[loss=0.2859, simple_loss=0.3521, pruned_loss=0.1098, over 28954.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3827, pruned_loss=0.1346, over 5651256.56 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3793, pruned_loss=0.1297, over 5686981.73 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3829, pruned_loss=0.1349, over 5638855.39 frames. ], batch size: 106, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:40:02,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.464e+03 1.739e+03 2.257e+03 4.317e+03, threshold=3.477e+03, percent-clipped=1.0
+2023-03-04 20:40:34,340 INFO [train.py:968] (0/2) Epoch 9, batch 28700, giga_loss[loss=0.2859, simple_loss=0.3502, pruned_loss=0.1108, over 28224.00 frames. ], tot_loss[loss=0.326, simple_loss=0.383, pruned_loss=0.1345, over 5660892.79 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3795, pruned_loss=0.1298, over 5690928.68 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.383, pruned_loss=0.1348, over 5646897.02 frames. ], batch size: 77, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:40:51,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=392917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:41:21,029 INFO [train.py:968] (0/2) Epoch 9, batch 28750, giga_loss[loss=0.3377, simple_loss=0.3987, pruned_loss=0.1384, over 28507.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3843, pruned_loss=0.1357, over 5666024.12 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3797, pruned_loss=0.13, over 5694196.66 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3841, pruned_loss=0.1358, over 5651659.15 frames. ], batch size: 71, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:41:35,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.757e+02 1.750e+03 2.278e+03 3.481e+03 1.694e+04, threshold=4.555e+03, percent-clipped=25.0
+2023-03-04 20:41:46,271 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-04 20:42:10,619 INFO [train.py:968] (0/2) Epoch 9, batch 28800, giga_loss[loss=0.3476, simple_loss=0.4022, pruned_loss=0.1465, over 28740.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3861, pruned_loss=0.137, over 5667457.57 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3799, pruned_loss=0.13, over 5696103.83 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.386, pruned_loss=0.1372, over 5653896.24 frames. ], batch size: 242, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:42:57,374 INFO [train.py:968] (0/2) Epoch 9, batch 28850, libri_loss[loss=0.3336, simple_loss=0.3933, pruned_loss=0.137, over 28723.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3855, pruned_loss=0.1367, over 5676865.90 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3796, pruned_loss=0.1298, over 5699614.67 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3858, pruned_loss=0.1373, over 5662138.42 frames. ], batch size: 106, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:43:12,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.786e+03 2.564e+03 3.789e+03 8.513e+03, threshold=5.127e+03, percent-clipped=11.0
+2023-03-04 20:43:42,953 INFO [train.py:968] (0/2) Epoch 9, batch 28900, giga_loss[loss=0.3413, simple_loss=0.3897, pruned_loss=0.1464, over 28641.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3857, pruned_loss=0.1369, over 5673026.11 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3797, pruned_loss=0.1299, over 5696943.59 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3859, pruned_loss=0.1373, over 5663791.15 frames. ], batch size: 307, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:43:53,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6277, 4.3215, 1.7153, 1.7405], device='cuda:0'), covar=tensor([0.0876, 0.0311, 0.0846, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0507, 0.0332, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 20:44:31,012 INFO [train.py:968] (0/2) Epoch 9, batch 28950, giga_loss[loss=0.3343, simple_loss=0.3921, pruned_loss=0.1382, over 28571.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3863, pruned_loss=0.1369, over 5663968.36 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3801, pruned_loss=0.1301, over 5690628.71 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3862, pruned_loss=0.1371, over 5662443.70 frames. ], batch size: 307, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:44:32,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=393149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:44:45,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.202e+02 1.553e+03 2.017e+03 2.919e+03 9.345e+03, threshold=4.034e+03, percent-clipped=4.0
+2023-03-04 20:45:19,087 INFO [train.py:968] (0/2) Epoch 9, batch 29000, giga_loss[loss=0.3915, simple_loss=0.4214, pruned_loss=0.1808, over 26539.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.386, pruned_loss=0.1363, over 5657036.63 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3803, pruned_loss=0.1303, over 5679254.60 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3858, pruned_loss=0.1364, over 5664977.85 frames. ], batch size: 555, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:45:44,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2394, 1.1014, 4.1238, 3.2150], device='cuda:0'), covar=tensor([0.1708, 0.2689, 0.0434, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0638, 0.0578, 0.0834, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 20:46:02,688 INFO [train.py:968] (0/2) Epoch 9, batch 29050, giga_loss[loss=0.3762, simple_loss=0.4201, pruned_loss=0.1662, over 28750.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3876, pruned_loss=0.1378, over 5658361.03 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3808, pruned_loss=0.1306, over 5674989.77 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3872, pruned_loss=0.1377, over 5667808.23 frames. ], batch size: 262, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:46:17,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.705e+02 1.473e+03 1.806e+03 2.304e+03 4.213e+03, threshold=3.613e+03, percent-clipped=1.0
+2023-03-04 20:46:42,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=393292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:46:42,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=393292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:46:42,742 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=393292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:46:44,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=393295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:46:46,718 INFO [train.py:968] (0/2) Epoch 9, batch 29100, giga_loss[loss=0.3085, simple_loss=0.3787, pruned_loss=0.1192, over 28939.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3882, pruned_loss=0.1383, over 5656931.94 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3806, pruned_loss=0.1305, over 5676451.55 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3881, pruned_loss=0.1385, over 5662680.53 frames. ], batch size: 136, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:46:55,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3782, 1.0973, 4.0192, 3.1328], device='cuda:0'), covar=tensor([0.1534, 0.2617, 0.0442, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0634, 0.0576, 0.0831, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 20:47:05,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5710, 3.8890, 1.6259, 1.7361], device='cuda:0'), covar=tensor([0.0853, 0.0317, 0.0886, 0.1198], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0510, 0.0333, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 20:47:09,852 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=393324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:47:10,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2504, 1.8139, 1.3495, 0.3799], device='cuda:0'), covar=tensor([0.2467, 0.1786, 0.2693, 0.3416], device='cuda:0'), in_proj_covar=tensor([0.1532, 0.1458, 0.1473, 0.1242], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 20:47:14,647 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 20:47:19,247 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=393334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:47:25,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2436, 1.0848, 4.3924, 3.2588], device='cuda:0'), covar=tensor([0.1704, 0.2694, 0.0367, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.0635, 0.0576, 0.0831, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 20:47:30,690 INFO [train.py:968] (0/2) Epoch 9, batch 29150, giga_loss[loss=0.3221, simple_loss=0.3796, pruned_loss=0.1324, over 28868.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3871, pruned_loss=0.1376, over 5655368.80 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3801, pruned_loss=0.1303, over 5679714.23 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3877, pruned_loss=0.1382, over 5656293.87 frames. ], batch size: 199, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:47:43,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.726e+02 1.533e+03 1.987e+03 2.671e+03 7.838e+03, threshold=3.973e+03, percent-clipped=12.0
+2023-03-04 20:48:17,236 INFO [train.py:968] (0/2) Epoch 9, batch 29200, giga_loss[loss=0.2794, simple_loss=0.3601, pruned_loss=0.09939, over 28377.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3876, pruned_loss=0.1371, over 5641326.32 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3803, pruned_loss=0.1306, over 5674666.75 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.388, pruned_loss=0.1375, over 5645834.73 frames. ], batch size: 71, lr: 3.57e-03, grad_scale: 8.0
+2023-03-04 20:48:53,012 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=393435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:48:57,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=393438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:49:07,609 INFO [train.py:968] (0/2) Epoch 9, batch 29250, giga_loss[loss=0.2821, simple_loss=0.3567, pruned_loss=0.1037, over 28819.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3878, pruned_loss=0.1367, over 5641841.11 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3802, pruned_loss=0.1306, over 5680538.61 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3884, pruned_loss=0.1372, over 5639406.39 frames. ], batch size: 243, lr: 3.57e-03, grad_scale: 8.0
+2023-03-04 20:49:18,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.877e+02 1.415e+03 1.877e+03 2.703e+03 7.488e+03, threshold=3.754e+03, percent-clipped=6.0
+2023-03-04 20:49:21,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5342, 4.3615, 4.1289, 2.1257], device='cuda:0'), covar=tensor([0.0498, 0.0652, 0.0705, 0.1840], device='cuda:0'), in_proj_covar=tensor([0.1012, 0.0958, 0.0842, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-04 20:49:21,401 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=393467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:49:50,962 INFO [train.py:968] (0/2) Epoch 9, batch 29300, giga_loss[loss=0.3281, simple_loss=0.3844, pruned_loss=0.1358, over 28193.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3866, pruned_loss=0.1355, over 5650602.66 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3801, pruned_loss=0.1304, over 5681992.65 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3873, pruned_loss=0.136, over 5646992.23 frames. ], batch size: 368, lr: 3.57e-03, grad_scale: 8.0
+2023-03-04 20:49:59,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2581, 0.9659, 0.9532, 1.3603], device='cuda:0'), covar=tensor([0.0768, 0.0339, 0.0340, 0.0813], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0115, 0.0116, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0082], device='cuda:0')
+2023-03-04 20:50:12,965 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=393524.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 20:50:17,660 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 20:50:32,718 INFO [train.py:968] (0/2) Epoch 9, batch 29350, libri_loss[loss=0.2578, simple_loss=0.3197, pruned_loss=0.09793, over 29336.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3849, pruned_loss=0.1343, over 5665733.10 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3803, pruned_loss=0.1307, over 5690033.65 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3855, pruned_loss=0.1347, over 5654360.08 frames. ], batch size: 71, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:50:43,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.319e+02 1.616e+03 2.043e+03 2.959e+03 9.826e+03, threshold=4.086e+03, percent-clipped=9.0
+2023-03-04 20:51:15,866 INFO [train.py:968] (0/2) Epoch 9, batch 29400, giga_loss[loss=0.3737, simple_loss=0.4114, pruned_loss=0.168, over 27595.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3851, pruned_loss=0.1344, over 5658876.29 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.381, pruned_loss=0.1312, over 5695314.78 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3852, pruned_loss=0.1344, over 5644234.26 frames. ], batch size: 472, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:51:26,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2198, 1.5468, 1.2439, 1.0697], device='cuda:0'), covar=tensor([0.2265, 0.2144, 0.2371, 0.1890], device='cuda:0'), in_proj_covar=tensor([0.1252, 0.0932, 0.1109, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 20:52:08,156 INFO [train.py:968] (0/2) Epoch 9, batch 29450, giga_loss[loss=0.3093, simple_loss=0.3718, pruned_loss=0.1234, over 28957.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3862, pruned_loss=0.135, over 5665220.22 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.381, pruned_loss=0.1312, over 5695314.78 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3862, pruned_loss=0.135, over 5653824.15 frames. ], batch size: 128, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:52:14,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-04 20:52:22,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.389e+02 1.478e+03 1.890e+03 2.675e+03 1.004e+04, threshold=3.779e+03, percent-clipped=13.0
+2023-03-04 20:52:24,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=393667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:52:55,464 INFO [train.py:968] (0/2) Epoch 9, batch 29500, giga_loss[loss=0.3145, simple_loss=0.3695, pruned_loss=0.1298, over 29109.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3856, pruned_loss=0.1357, over 5661950.10 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3812, pruned_loss=0.1313, over 5698568.58 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3855, pruned_loss=0.1357, over 5649959.75 frames. ], batch size: 128, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:53:02,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6058, 4.1533, 1.6647, 1.6324], device='cuda:0'), covar=tensor([0.0852, 0.0309, 0.0843, 0.1262], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0507, 0.0331, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 20:53:05,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=393709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:53:14,060 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6798, 1.6627, 1.2552, 1.3440], device='cuda:0'), covar=tensor([0.0689, 0.0568, 0.0918, 0.1031], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0442, 0.0500, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 20:53:41,555 INFO [train.py:968] (0/2) Epoch 9, batch 29550, giga_loss[loss=0.3553, simple_loss=0.4162, pruned_loss=0.1472, over 28993.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3857, pruned_loss=0.1357, over 5675617.12 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.381, pruned_loss=0.1312, over 5700609.05 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3859, pruned_loss=0.1358, over 5664119.92 frames. ], batch size: 155, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:53:57,802 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.619e+03 1.963e+03 2.803e+03 1.139e+04, threshold=3.925e+03, percent-clipped=15.0
+2023-03-04 20:54:25,454 INFO [train.py:968] (0/2) Epoch 9, batch 29600, giga_loss[loss=0.2956, simple_loss=0.3624, pruned_loss=0.1144, over 28883.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.389, pruned_loss=0.1384, over 5673382.47 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3814, pruned_loss=0.1314, over 5706398.74 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.389, pruned_loss=0.1385, over 5658314.71 frames. ], batch size: 106, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:54:36,412 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=393810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:54:37,250 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 20:54:38,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=393813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:55:04,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=393842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:55:11,075 INFO [train.py:968] (0/2) Epoch 9, batch 29650, libri_loss[loss=0.3268, simple_loss=0.3937, pruned_loss=0.1299, over 28598.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3881, pruned_loss=0.1377, over 5658810.79 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3815, pruned_loss=0.1313, over 5709627.95 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3883, pruned_loss=0.1382, over 5641848.96 frames. ], batch size: 106, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:55:13,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=393852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:55:16,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=393855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:55:24,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.851e+02 1.603e+03 2.154e+03 2.859e+03 6.958e+03, threshold=4.309e+03, percent-clipped=8.0
+2023-03-04 20:55:40,840 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=393884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:55:53,900 INFO [train.py:968] (0/2) Epoch 9, batch 29700, giga_loss[loss=0.3353, simple_loss=0.3936, pruned_loss=0.1385, over 28675.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3887, pruned_loss=0.1381, over 5661033.14 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3816, pruned_loss=0.1314, over 5709434.48 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3888, pruned_loss=0.1384, over 5647322.12 frames. ], batch size: 242, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:55:54,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=393899.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 20:56:21,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2725, 1.5322, 1.1949, 1.6318], device='cuda:0'), covar=tensor([0.2366, 0.2218, 0.2597, 0.1888], device='cuda:0'), in_proj_covar=tensor([0.1259, 0.0938, 0.1112, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 20:56:40,558 INFO [train.py:968] (0/2) Epoch 9, batch 29750, giga_loss[loss=0.3651, simple_loss=0.4118, pruned_loss=0.1592, over 28604.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3885, pruned_loss=0.1374, over 5665394.53 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3815, pruned_loss=0.1313, over 5712575.77 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3887, pruned_loss=0.1378, over 5651237.41 frames. ], batch size: 307, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:56:57,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.915e+02 1.681e+03 2.037e+03 2.626e+03 6.238e+03, threshold=4.075e+03, percent-clipped=5.0
+2023-03-04 20:57:24,926 INFO [train.py:968] (0/2) Epoch 9, batch 29800, giga_loss[loss=0.3206, simple_loss=0.3818, pruned_loss=0.1297, over 28679.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3872, pruned_loss=0.1359, over 5668882.61 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3816, pruned_loss=0.1313, over 5718775.53 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3876, pruned_loss=0.1364, over 5650676.61 frames. ], batch size: 242, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:57:28,288 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-394000.pt
+2023-03-04 20:58:00,676 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=394042.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 20:58:03,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=394045.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 20:58:06,797 INFO [train.py:968] (0/2) Epoch 9, batch 29850, giga_loss[loss=0.4293, simple_loss=0.4418, pruned_loss=0.2084, over 26557.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3858, pruned_loss=0.135, over 5665344.98 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3813, pruned_loss=0.1311, over 5716539.87 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3866, pruned_loss=0.1358, over 5650151.40 frames. ], batch size: 555, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:58:07,166 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 20:58:22,815 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.708e+03 2.202e+03 3.239e+03 1.173e+04, threshold=4.404e+03, percent-clipped=15.0
+2023-03-04 20:58:28,875 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=394074.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 20:58:50,280 INFO [train.py:968] (0/2) Epoch 9, batch 29900, giga_loss[loss=0.3077, simple_loss=0.37, pruned_loss=0.1227, over 28773.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3841, pruned_loss=0.1337, over 5675138.06 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3808, pruned_loss=0.1308, over 5720512.28 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3852, pruned_loss=0.1346, over 5658804.21 frames. ], batch size: 119, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:59:34,005 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=394145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:59:35,634 INFO [train.py:968] (0/2) Epoch 9, batch 29950, giga_loss[loss=0.2841, simple_loss=0.3446, pruned_loss=0.1118, over 28584.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.381, pruned_loss=0.1319, over 5669002.81 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3807, pruned_loss=0.1306, over 5714410.51 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.382, pruned_loss=0.1328, over 5660582.37 frames. ], batch size: 71, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:59:51,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 1.709e+03 2.287e+03 3.335e+03 7.139e+03, threshold=4.573e+03, percent-clipped=11.0
+2023-03-04 20:59:57,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6309, 1.7157, 1.5906, 1.6018], device='cuda:0'), covar=tensor([0.1335, 0.1807, 0.1856, 0.1719], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0737, 0.0665, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 21:00:22,010 INFO [train.py:968] (0/2) Epoch 9, batch 30000, giga_loss[loss=0.3814, simple_loss=0.4061, pruned_loss=0.1784, over 26572.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3774, pruned_loss=0.1306, over 5659959.30 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.381, pruned_loss=0.1309, over 5717861.85 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.378, pruned_loss=0.1312, over 5649454.71 frames. ], batch size: 555, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:00:22,014 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 21:00:29,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6530, 1.6216, 1.2685, 1.3211], device='cuda:0'), covar=tensor([0.0695, 0.0451, 0.0964, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0438, 0.0495, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 21:00:30,275 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2199, simple_loss=0.3267, pruned_loss=0.05656, over 944034.00 frames. 
+2023-03-04 21:00:30,275 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 21:01:14,698 INFO [train.py:968] (0/2) Epoch 9, batch 30050, giga_loss[loss=0.2886, simple_loss=0.3551, pruned_loss=0.1111, over 28400.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3767, pruned_loss=0.1309, over 5659140.26 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3814, pruned_loss=0.1313, over 5707521.83 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3766, pruned_loss=0.131, over 5659236.74 frames. ], batch size: 65, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:01:33,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.787e+03 2.217e+03 3.243e+03 9.618e+03, threshold=4.434e+03, percent-clipped=9.0
+2023-03-04 21:02:02,386 INFO [train.py:968] (0/2) Epoch 9, batch 30100, giga_loss[loss=0.3722, simple_loss=0.4153, pruned_loss=0.1645, over 28751.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3758, pruned_loss=0.1311, over 5643576.80 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3815, pruned_loss=0.1313, over 5711269.22 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3756, pruned_loss=0.1311, over 5639345.29 frames. ], batch size: 284, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:02:19,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5696, 1.6389, 1.3438, 1.8422], device='cuda:0'), covar=tensor([0.2351, 0.2361, 0.2492, 0.2232], device='cuda:0'), in_proj_covar=tensor([0.1260, 0.0936, 0.1111, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 21:02:33,327 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 21:02:49,269 INFO [train.py:968] (0/2) Epoch 9, batch 30150, giga_loss[loss=0.2702, simple_loss=0.3535, pruned_loss=0.09349, over 28477.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3749, pruned_loss=0.1286, over 5646235.25 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3813, pruned_loss=0.1312, over 5705094.29 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3748, pruned_loss=0.1287, over 5647641.89 frames. ], batch size: 336, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:03:08,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.455e+02 1.545e+03 1.982e+03 2.561e+03 6.050e+03, threshold=3.963e+03, percent-clipped=3.0
+2023-03-04 21:03:42,472 INFO [train.py:968] (0/2) Epoch 9, batch 30200, giga_loss[loss=0.2853, simple_loss=0.3653, pruned_loss=0.1027, over 28965.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3713, pruned_loss=0.1246, over 5640613.18 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3801, pruned_loss=0.1307, over 5711393.32 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3721, pruned_loss=0.125, over 5634152.27 frames. ], batch size: 227, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:03:52,580 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4730, 3.7413, 1.6481, 1.5486], device='cuda:0'), covar=tensor([0.0916, 0.0295, 0.0882, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0506, 0.0331, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 21:04:27,192 INFO [train.py:968] (0/2) Epoch 9, batch 30250, giga_loss[loss=0.3095, simple_loss=0.3801, pruned_loss=0.1194, over 28327.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3693, pruned_loss=0.1217, over 5648106.86 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.38, pruned_loss=0.1309, over 5706595.42 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3697, pruned_loss=0.1215, over 5645284.76 frames. ], batch size: 368, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:04:34,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3935, 1.9993, 1.4325, 1.4822], device='cuda:0'), covar=tensor([0.0769, 0.0265, 0.0325, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0082], device='cuda:0')
+2023-03-04 21:04:43,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.318e+02 1.536e+03 1.934e+03 3.095e+03 9.518e+03, threshold=3.868e+03, percent-clipped=17.0
+2023-03-04 21:04:47,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-04 21:05:13,517 INFO [train.py:968] (0/2) Epoch 9, batch 30300, giga_loss[loss=0.2568, simple_loss=0.3392, pruned_loss=0.08719, over 28953.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3665, pruned_loss=0.1188, over 5654684.05 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3788, pruned_loss=0.1302, over 5715031.49 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3674, pruned_loss=0.1189, over 5642537.47 frames. ], batch size: 199, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:05:17,915 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-04 21:05:34,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=394520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:06:02,792 INFO [train.py:968] (0/2) Epoch 9, batch 30350, giga_loss[loss=0.2545, simple_loss=0.3277, pruned_loss=0.09061, over 27635.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1145, over 5657658.87 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3783, pruned_loss=0.1299, over 5716780.18 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3631, pruned_loss=0.1147, over 5646148.23 frames. ], batch size: 472, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:06:19,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.795e+02 1.216e+03 1.524e+03 2.376e+03 4.969e+03, threshold=3.048e+03, percent-clipped=3.0
+2023-03-04 21:06:32,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=394579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:06:52,311 INFO [train.py:968] (0/2) Epoch 9, batch 30400, giga_loss[loss=0.261, simple_loss=0.349, pruned_loss=0.08649, over 28680.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3609, pruned_loss=0.1108, over 5673505.56 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3781, pruned_loss=0.1298, over 5717977.55 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3617, pruned_loss=0.1108, over 5662826.22 frames. ], batch size: 262, lr: 3.56e-03, grad_scale: 8.0
+2023-03-04 21:07:33,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4588, 1.9111, 1.4735, 1.6330], device='cuda:0'), covar=tensor([0.0746, 0.0259, 0.0319, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0082], device='cuda:0')
+2023-03-04 21:07:43,680 INFO [train.py:968] (0/2) Epoch 9, batch 30450, giga_loss[loss=0.2524, simple_loss=0.33, pruned_loss=0.0874, over 28663.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3604, pruned_loss=0.1101, over 5657468.39 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3775, pruned_loss=0.1296, over 5700859.39 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3614, pruned_loss=0.1101, over 5663276.30 frames. ], batch size: 78, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:07:56,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3492, 1.6695, 1.6625, 1.2298], device='cuda:0'), covar=tensor([0.1550, 0.2166, 0.1276, 0.1511], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0698, 0.0835, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-04 21:07:59,886 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=394663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:08:02,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=394666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:08:04,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.387e+02 1.248e+03 1.694e+03 2.607e+03 5.414e+03, threshold=3.388e+03, percent-clipped=18.0
+2023-03-04 21:08:26,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9620, 1.2100, 1.3550, 1.0010], device='cuda:0'), covar=tensor([0.1364, 0.1157, 0.1656, 0.1435], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0722, 0.0651, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 21:08:32,255 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=394695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:08:34,123 INFO [train.py:968] (0/2) Epoch 9, batch 30500, giga_loss[loss=0.2895, simple_loss=0.3627, pruned_loss=0.1082, over 28958.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3587, pruned_loss=0.1086, over 5658027.37 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.377, pruned_loss=0.1293, over 5701980.63 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3597, pruned_loss=0.1086, over 5661204.55 frames. ], batch size: 136, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:08:37,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1872, 1.9610, 1.6815, 1.4971], device='cuda:0'), covar=tensor([0.0830, 0.0258, 0.0273, 0.1006], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0117, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0082], device='cuda:0')
+2023-03-04 21:09:23,817 INFO [train.py:968] (0/2) Epoch 9, batch 30550, giga_loss[loss=0.2937, simple_loss=0.3511, pruned_loss=0.1181, over 26722.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3557, pruned_loss=0.1067, over 5653706.47 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3766, pruned_loss=0.1294, over 5696110.56 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3564, pruned_loss=0.1062, over 5660110.61 frames. ], batch size: 555, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:09:38,190 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 21:09:41,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.798e+02 1.275e+03 1.777e+03 2.323e+03 9.514e+03, threshold=3.553e+03, percent-clipped=13.0
+2023-03-04 21:09:46,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8682, 3.6542, 3.4854, 1.6548], device='cuda:0'), covar=tensor([0.0739, 0.0946, 0.0918, 0.2047], device='cuda:0'), in_proj_covar=tensor([0.0992, 0.0937, 0.0813, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 21:10:09,101 INFO [train.py:968] (0/2) Epoch 9, batch 30600, giga_loss[loss=0.2597, simple_loss=0.3422, pruned_loss=0.08858, over 28758.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.355, pruned_loss=0.1069, over 5640241.66 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3764, pruned_loss=0.1296, over 5687592.43 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3551, pruned_loss=0.1055, over 5651483.77 frames. ], batch size: 284, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:10:54,715 INFO [train.py:968] (0/2) Epoch 9, batch 30650, giga_loss[loss=0.2688, simple_loss=0.3533, pruned_loss=0.09215, over 28900.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.355, pruned_loss=0.1064, over 5655455.99 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3757, pruned_loss=0.1293, over 5691245.94 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3552, pruned_loss=0.1051, over 5660168.95 frames. ], batch size: 136, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:11:15,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.303e+02 1.357e+03 1.778e+03 2.587e+03 7.846e+03, threshold=3.556e+03, percent-clipped=10.0
+2023-03-04 21:11:19,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2542, 1.3667, 1.0081, 1.1377], device='cuda:0'), covar=tensor([0.0983, 0.0968, 0.0867, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.1607, 0.1491, 0.1445, 0.1553], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 21:11:43,413 INFO [train.py:968] (0/2) Epoch 9, batch 30700, giga_loss[loss=0.2697, simple_loss=0.3526, pruned_loss=0.09341, over 29095.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3539, pruned_loss=0.1054, over 5641287.11 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3759, pruned_loss=0.1294, over 5680227.79 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3537, pruned_loss=0.104, over 5654483.72 frames. ], batch size: 128, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:12:09,423 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-04 21:12:31,088 INFO [train.py:968] (0/2) Epoch 9, batch 30750, giga_loss[loss=0.2488, simple_loss=0.3312, pruned_loss=0.08316, over 29078.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3503, pruned_loss=0.1026, over 5648032.76 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3753, pruned_loss=0.1292, over 5681473.30 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.35, pruned_loss=0.101, over 5656519.56 frames. ], batch size: 155, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:12:37,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=394954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:12:49,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.450e+02 1.400e+03 1.733e+03 2.452e+03 8.213e+03, threshold=3.466e+03, percent-clipped=9.0
+2023-03-04 21:13:01,791 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=394981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:13:16,802 INFO [train.py:968] (0/2) Epoch 9, batch 30800, giga_loss[loss=0.2415, simple_loss=0.3183, pruned_loss=0.08239, over 28863.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3465, pruned_loss=0.1003, over 5664518.85 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3748, pruned_loss=0.129, over 5686731.24 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3457, pruned_loss=0.09806, over 5665460.11 frames. ], batch size: 186, lr: 3.56e-03, grad_scale: 8.0
+2023-03-04 21:14:03,651 INFO [train.py:968] (0/2) Epoch 9, batch 30850, libri_loss[loss=0.2913, simple_loss=0.3557, pruned_loss=0.1134, over 29292.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3448, pruned_loss=0.1, over 5658921.81 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3742, pruned_loss=0.1288, over 5681560.99 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3441, pruned_loss=0.09789, over 5664327.21 frames. ], batch size: 94, lr: 3.56e-03, grad_scale: 8.0
+2023-03-04 21:14:22,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.121e+02 1.224e+03 1.750e+03 2.375e+03 3.812e+03, threshold=3.500e+03, percent-clipped=2.0
+2023-03-04 21:14:31,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 21:14:50,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=395097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:14:50,976 INFO [train.py:968] (0/2) Epoch 9, batch 30900, giga_loss[loss=0.2307, simple_loss=0.3097, pruned_loss=0.07587, over 28458.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3448, pruned_loss=0.101, over 5653810.29 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3732, pruned_loss=0.1283, over 5686134.22 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3443, pruned_loss=0.09898, over 5653292.49 frames. ], batch size: 71, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:14:52,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=395100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:15:19,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2569, 2.9954, 1.3997, 1.3719], device='cuda:0'), covar=tensor([0.0921, 0.0333, 0.0884, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0499, 0.0332, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 21:15:27,480 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=395129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:15:46,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=395147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:15:46,422 INFO [train.py:968] (0/2) Epoch 9, batch 30950, giga_loss[loss=0.2837, simple_loss=0.3503, pruned_loss=0.1085, over 26757.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3455, pruned_loss=0.1011, over 5637327.91 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.373, pruned_loss=0.1282, over 5678676.27 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3451, pruned_loss=0.09927, over 5643964.78 frames. ], batch size: 555, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:16:10,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.690e+02 1.347e+03 1.950e+03 2.596e+03 7.123e+03, threshold=3.900e+03, percent-clipped=9.0
+2023-03-04 21:16:40,283 INFO [train.py:968] (0/2) Epoch 9, batch 31000, giga_loss[loss=0.2648, simple_loss=0.3511, pruned_loss=0.08922, over 28952.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3478, pruned_loss=0.1014, over 5638187.91 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3724, pruned_loss=0.1279, over 5679727.36 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3473, pruned_loss=0.09948, over 5641296.71 frames. ], batch size: 164, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:17:17,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3662, 1.8208, 1.3055, 0.8090], device='cuda:0'), covar=tensor([0.4063, 0.2511, 0.2077, 0.3487], device='cuda:0'), in_proj_covar=tensor([0.1516, 0.1443, 0.1472, 0.1238], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 21:17:41,388 INFO [train.py:968] (0/2) Epoch 9, batch 31050, giga_loss[loss=0.2528, simple_loss=0.3135, pruned_loss=0.09604, over 24399.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3473, pruned_loss=0.1008, over 5628183.74 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3723, pruned_loss=0.1278, over 5682867.80 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3467, pruned_loss=0.099, over 5627025.69 frames. ], batch size: 705, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:17:56,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=395261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:18:06,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.410e+02 1.432e+03 1.825e+03 3.023e+03 6.941e+03, threshold=3.649e+03, percent-clipped=12.0
+2023-03-04 21:18:20,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2872, 1.1865, 0.9515, 1.3714], device='cuda:0'), covar=tensor([0.0704, 0.0323, 0.0352, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0117, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:0')
+2023-03-04 21:18:37,426 INFO [train.py:968] (0/2) Epoch 9, batch 31100, libri_loss[loss=0.2553, simple_loss=0.3177, pruned_loss=0.09642, over 29599.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3472, pruned_loss=0.1012, over 5639157.13 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3714, pruned_loss=0.1274, over 5683803.35 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3462, pruned_loss=0.09876, over 5634681.24 frames. ], batch size: 74, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:19:34,916 INFO [train.py:968] (0/2) Epoch 9, batch 31150, giga_loss[loss=0.2408, simple_loss=0.3266, pruned_loss=0.07748, over 28425.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3447, pruned_loss=0.09892, over 5643661.02 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.371, pruned_loss=0.1272, over 5685425.67 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3439, pruned_loss=0.0968, over 5637986.94 frames. ], batch size: 369, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:19:46,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=395356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:19:46,788 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-04 21:20:06,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.425e+02 1.432e+03 1.777e+03 2.793e+03 1.933e+04, threshold=3.554e+03, percent-clipped=12.0
+2023-03-04 21:20:37,369 INFO [train.py:968] (0/2) Epoch 9, batch 31200, giga_loss[loss=0.2872, simple_loss=0.3575, pruned_loss=0.1084, over 28867.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3426, pruned_loss=0.09616, over 5643636.49 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3709, pruned_loss=0.1273, over 5689814.59 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3415, pruned_loss=0.0938, over 5634152.64 frames. ], batch size: 284, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:21:35,282 INFO [train.py:968] (0/2) Epoch 9, batch 31250, libri_loss[loss=0.2877, simple_loss=0.3488, pruned_loss=0.1134, over 29595.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3397, pruned_loss=0.09533, over 5654402.38 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.37, pruned_loss=0.1269, over 5691685.27 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.339, pruned_loss=0.09307, over 5643772.03 frames. ], batch size: 75, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:21:48,221 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=395461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:21:57,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0619, 1.3738, 1.1115, 0.2820], device='cuda:0'), covar=tensor([0.1989, 0.2000, 0.3121, 0.3418], device='cuda:0'), in_proj_covar=tensor([0.1500, 0.1426, 0.1456, 0.1227], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 21:21:58,957 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.476e+02 1.296e+03 1.949e+03 2.739e+03 1.013e+04, threshold=3.898e+03, percent-clipped=14.0
+2023-03-04 21:22:24,860 INFO [train.py:968] (0/2) Epoch 9, batch 31300, giga_loss[loss=0.2572, simple_loss=0.3346, pruned_loss=0.08987, over 28852.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3402, pruned_loss=0.09666, over 5669528.82 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3687, pruned_loss=0.1262, over 5699268.50 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3391, pruned_loss=0.09371, over 5652142.95 frames. ], batch size: 174, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:22:26,137 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=395499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:22:29,531 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=395502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:22:53,933 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=395522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:23:02,839 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=395531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:23:22,294 INFO [train.py:968] (0/2) Epoch 9, batch 31350, giga_loss[loss=0.2778, simple_loss=0.3603, pruned_loss=0.09762, over 28815.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3393, pruned_loss=0.09591, over 5676184.01 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3688, pruned_loss=0.1264, over 5701771.07 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.338, pruned_loss=0.09301, over 5659939.30 frames. ], batch size: 174, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:23:45,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.672e+02 1.496e+03 1.971e+03 2.469e+03 1.441e+04, threshold=3.941e+03, percent-clipped=5.0
+2023-03-04 21:24:13,422 INFO [train.py:968] (0/2) Epoch 9, batch 31400, giga_loss[loss=0.2783, simple_loss=0.3588, pruned_loss=0.09896, over 28922.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3416, pruned_loss=0.09725, over 5675422.09 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3687, pruned_loss=0.1264, over 5704734.64 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3397, pruned_loss=0.09379, over 5658619.49 frames. ], batch size: 199, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:24:16,898 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5621, 3.3638, 3.1762, 1.6755], device='cuda:0'), covar=tensor([0.0738, 0.0889, 0.0915, 0.2507], device='cuda:0'), in_proj_covar=tensor([0.0978, 0.0919, 0.0807, 0.0630], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 21:24:59,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=395636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:25:14,563 INFO [train.py:968] (0/2) Epoch 9, batch 31450, giga_loss[loss=0.2296, simple_loss=0.3205, pruned_loss=0.06939, over 29053.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3433, pruned_loss=0.09772, over 5666854.55 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3686, pruned_loss=0.1266, over 5709181.08 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3412, pruned_loss=0.09416, over 5648549.07 frames. ], batch size: 155, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:25:35,198 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=395665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:25:38,191 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=395668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:25:39,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.295e+02 1.407e+03 1.822e+03 2.348e+03 9.847e+03, threshold=3.643e+03, percent-clipped=5.0
+2023-03-04 21:26:09,699 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=395697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:26:10,055 INFO [train.py:968] (0/2) Epoch 9, batch 31500, giga_loss[loss=0.253, simple_loss=0.3352, pruned_loss=0.0854, over 28906.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3405, pruned_loss=0.09605, over 5682817.47 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3681, pruned_loss=0.1264, over 5713951.95 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3381, pruned_loss=0.09199, over 5662225.88 frames. ], batch size: 145, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:27:15,877 INFO [train.py:968] (0/2) Epoch 9, batch 31550, giga_loss[loss=0.2507, simple_loss=0.3376, pruned_loss=0.08186, over 28963.00 frames. ], tot_loss[loss=0.268, simple_loss=0.342, pruned_loss=0.09696, over 5679997.68 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3681, pruned_loss=0.1264, over 5706688.14 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3398, pruned_loss=0.09336, over 5669547.12 frames. ], batch size: 213, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:27:40,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.698e+02 1.212e+03 1.764e+03 2.488e+03 5.420e+03, threshold=3.529e+03, percent-clipped=13.0
+2023-03-04 21:27:52,245 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=395779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:27:54,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=395782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:28:08,208 INFO [train.py:968] (0/2) Epoch 9, batch 31600, giga_loss[loss=0.2696, simple_loss=0.363, pruned_loss=0.08811, over 29016.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3429, pruned_loss=0.09669, over 5678725.40 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3671, pruned_loss=0.1258, over 5711514.16 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3408, pruned_loss=0.09293, over 5664387.97 frames. ], batch size: 136, lr: 3.56e-03, grad_scale: 8.0
+2023-03-04 21:28:19,990 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3854, 1.6318, 1.7002, 1.2800], device='cuda:0'), covar=tensor([0.1821, 0.2354, 0.1376, 0.1682], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0689, 0.0833, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 21:28:21,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6609, 3.8354, 1.6095, 1.8436], device='cuda:0'), covar=tensor([0.0802, 0.0307, 0.0829, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0498, 0.0334, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 21:28:26,195 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=395811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:28:57,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=395836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:29:13,464 INFO [train.py:968] (0/2) Epoch 9, batch 31650, giga_loss[loss=0.2634, simple_loss=0.354, pruned_loss=0.08637, over 28903.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3452, pruned_loss=0.09512, over 5670203.73 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3669, pruned_loss=0.1257, over 5712657.98 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3435, pruned_loss=0.09191, over 5657414.28 frames. ], batch size: 199, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:29:39,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.927e+02 1.311e+03 1.865e+03 2.709e+03 6.531e+03, threshold=3.731e+03, percent-clipped=11.0
+2023-03-04 21:30:03,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8840, 1.1134, 1.0381, 0.6966], device='cuda:0'), covar=tensor([0.1348, 0.1406, 0.0879, 0.1380], device='cuda:0'), in_proj_covar=tensor([0.1622, 0.1489, 0.1440, 0.1561], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 21:30:11,707 INFO [train.py:968] (0/2) Epoch 9, batch 31700, giga_loss[loss=0.2762, simple_loss=0.3666, pruned_loss=0.09289, over 28929.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3467, pruned_loss=0.09483, over 5657639.69 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3668, pruned_loss=0.1257, over 5703366.46 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.345, pruned_loss=0.09159, over 5655435.46 frames. ], batch size: 145, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:31:09,892 INFO [train.py:968] (0/2) Epoch 9, batch 31750, giga_loss[loss=0.2368, simple_loss=0.3345, pruned_loss=0.06958, over 28935.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3461, pruned_loss=0.09367, over 5669603.11 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3668, pruned_loss=0.1258, over 5707300.74 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3441, pruned_loss=0.09009, over 5663092.85 frames. ], batch size: 145, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:31:22,785 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-04 21:31:34,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.360e+02 1.281e+03 1.642e+03 2.082e+03 6.582e+03, threshold=3.284e+03, percent-clipped=8.0
+2023-03-04 21:31:36,086 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-04 21:31:43,219 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=395979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:31:46,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=395982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:32:04,771 INFO [train.py:968] (0/2) Epoch 9, batch 31800, giga_loss[loss=0.3061, simple_loss=0.3679, pruned_loss=0.1221, over 28523.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3476, pruned_loss=0.09564, over 5663043.63 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3664, pruned_loss=0.1257, over 5692388.05 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3456, pruned_loss=0.09165, over 5669783.01 frames. ], batch size: 370, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:32:06,569 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-396000.pt
+2023-03-04 21:32:16,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:32:51,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1064, 1.2380, 3.4634, 3.0136], device='cuda:0'), covar=tensor([0.1983, 0.2864, 0.0737, 0.1693], device='cuda:0'), in_proj_covar=tensor([0.0627, 0.0569, 0.0822, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 21:32:56,930 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-04 21:33:06,082 INFO [train.py:968] (0/2) Epoch 9, batch 31850, giga_loss[loss=0.2993, simple_loss=0.3547, pruned_loss=0.122, over 26785.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3453, pruned_loss=0.09562, over 5671196.00 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3657, pruned_loss=0.1254, over 5695754.78 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3438, pruned_loss=0.09192, over 5673067.79 frames. ], batch size: 555, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:33:39,851 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.524e+02 1.398e+03 1.869e+03 2.751e+03 7.712e+03, threshold=3.737e+03, percent-clipped=14.0
+2023-03-04 21:34:19,092 INFO [train.py:968] (0/2) Epoch 9, batch 31900, giga_loss[loss=0.2489, simple_loss=0.3296, pruned_loss=0.08411, over 28998.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3466, pruned_loss=0.09701, over 5675821.92 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3651, pruned_loss=0.1249, over 5700949.61 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3454, pruned_loss=0.09375, over 5672131.92 frames. ], batch size: 186, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:34:57,632 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 21:35:32,840 INFO [train.py:968] (0/2) Epoch 9, batch 31950, giga_loss[loss=0.2257, simple_loss=0.3102, pruned_loss=0.07065, over 28978.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3434, pruned_loss=0.0955, over 5677708.74 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3653, pruned_loss=0.125, over 5705012.46 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3419, pruned_loss=0.0922, over 5670747.92 frames. ], batch size: 199, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:35:39,078 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:35:46,347 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:35:59,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.372e+02 1.258e+03 1.771e+03 2.640e+03 8.260e+03, threshold=3.543e+03, percent-clipped=9.0
+2023-03-04 21:36:10,101 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:36:32,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4110, 2.5965, 1.5230, 1.5335], device='cuda:0'), covar=tensor([0.0751, 0.0296, 0.0727, 0.1102], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0500, 0.0332, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 21:36:35,716 INFO [train.py:968] (0/2) Epoch 9, batch 32000, giga_loss[loss=0.2627, simple_loss=0.3368, pruned_loss=0.09425, over 28826.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3422, pruned_loss=0.09498, over 5672013.33 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3657, pruned_loss=0.1255, over 5700013.82 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.34, pruned_loss=0.09115, over 5669815.73 frames. ], batch size: 112, lr: 3.56e-03, grad_scale: 8.0
+2023-03-04 21:37:33,750 INFO [train.py:968] (0/2) Epoch 9, batch 32050, giga_loss[loss=0.2708, simple_loss=0.3459, pruned_loss=0.09781, over 28963.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3396, pruned_loss=0.09414, over 5671783.73 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3658, pruned_loss=0.1255, over 5696792.09 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3371, pruned_loss=0.09007, over 5671957.33 frames. ], batch size: 213, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:37:34,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9758, 1.9300, 1.5146, 1.5142], device='cuda:0'), covar=tensor([0.0659, 0.0587, 0.0872, 0.1016], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0435, 0.0498, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 21:37:50,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:38:09,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.461e+02 1.260e+03 1.714e+03 2.358e+03 6.806e+03, threshold=3.428e+03, percent-clipped=11.0
+2023-03-04 21:38:18,339 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:38:24,487 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:38:38,995 INFO [train.py:968] (0/2) Epoch 9, batch 32100, giga_loss[loss=0.2822, simple_loss=0.3619, pruned_loss=0.1012, over 28936.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3424, pruned_loss=0.0954, over 5671389.21 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3658, pruned_loss=0.1256, over 5688327.94 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3398, pruned_loss=0.09134, over 5678479.94 frames. ], batch size: 213, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:39:37,807 INFO [train.py:968] (0/2) Epoch 9, batch 32150, giga_loss[loss=0.2558, simple_loss=0.3284, pruned_loss=0.09162, over 28466.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3443, pruned_loss=0.0967, over 5679596.27 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3657, pruned_loss=0.1256, over 5690095.27 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3419, pruned_loss=0.09295, over 5683486.78 frames. ], batch size: 369, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:40:05,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.957e+02 1.373e+03 1.668e+03 2.326e+03 5.995e+03, threshold=3.336e+03, percent-clipped=13.0
+2023-03-04 21:40:36,337 INFO [train.py:968] (0/2) Epoch 9, batch 32200, giga_loss[loss=0.2521, simple_loss=0.3343, pruned_loss=0.08497, over 28645.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3428, pruned_loss=0.09703, over 5685053.84 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3652, pruned_loss=0.1254, over 5695623.06 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3404, pruned_loss=0.09299, over 5682670.30 frames. ], batch size: 262, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:40:39,686 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:41:16,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396429.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 21:41:41,361 INFO [train.py:968] (0/2) Epoch 9, batch 32250, giga_loss[loss=0.2636, simple_loss=0.3432, pruned_loss=0.09202, over 28998.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3431, pruned_loss=0.09782, over 5680253.54 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.365, pruned_loss=0.1254, over 5696513.94 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3413, pruned_loss=0.09456, over 5677622.39 frames. ], batch size: 106, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:42:13,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.829e+02 1.410e+03 1.892e+03 2.525e+03 7.111e+03, threshold=3.783e+03, percent-clipped=13.0
+2023-03-04 21:42:50,712 INFO [train.py:968] (0/2) Epoch 9, batch 32300, giga_loss[loss=0.2668, simple_loss=0.347, pruned_loss=0.09333, over 28707.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3433, pruned_loss=0.09715, over 5677030.35 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3652, pruned_loss=0.1254, over 5695830.09 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3415, pruned_loss=0.09421, over 5675619.78 frames. ], batch size: 262, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:43:26,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5822, 2.2748, 1.8118, 0.6740], device='cuda:0'), covar=tensor([0.2729, 0.1819, 0.2403, 0.3200], device='cuda:0'), in_proj_covar=tensor([0.1500, 0.1428, 0.1462, 0.1227], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 21:43:27,506 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:43:37,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:43:52,877 INFO [train.py:968] (0/2) Epoch 9, batch 32350, giga_loss[loss=0.2425, simple_loss=0.305, pruned_loss=0.09, over 24336.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.346, pruned_loss=0.09875, over 5680067.32 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3639, pruned_loss=0.1249, over 5704257.24 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3446, pruned_loss=0.09545, over 5670123.26 frames. ], batch size: 705, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:43:57,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:44:35,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.471e+02 1.479e+03 2.063e+03 2.758e+03 1.071e+04, threshold=4.126e+03, percent-clipped=8.0
+2023-03-04 21:45:06,023 INFO [train.py:968] (0/2) Epoch 9, batch 32400, giga_loss[loss=0.3114, simple_loss=0.367, pruned_loss=0.1279, over 26803.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3456, pruned_loss=0.09856, over 5675786.07 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3631, pruned_loss=0.1242, over 5708085.75 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3448, pruned_loss=0.09575, over 5663344.14 frames. ], batch size: 555, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 21:45:42,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3277, 1.6788, 1.3264, 1.4262], device='cuda:0'), covar=tensor([0.2469, 0.2229, 0.2540, 0.1941], device='cuda:0'), in_proj_covar=tensor([0.1249, 0.0922, 0.1106, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 21:45:52,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:13,949 INFO [train.py:968] (0/2) Epoch 9, batch 32450, giga_loss[loss=0.2519, simple_loss=0.335, pruned_loss=0.08435, over 28642.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3404, pruned_loss=0.09587, over 5683732.76 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3627, pruned_loss=0.124, over 5711174.47 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3397, pruned_loss=0.0934, over 5670693.10 frames. ], batch size: 307, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 21:46:22,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:30,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:44,400 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:46,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.111e+02 1.290e+03 1.711e+03 2.402e+03 5.013e+03, threshold=3.423e+03, percent-clipped=1.0
+2023-03-04 21:46:46,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:51,217 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:55,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:47:13,564 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:47:15,813 INFO [train.py:968] (0/2) Epoch 9, batch 32500, giga_loss[loss=0.2262, simple_loss=0.3104, pruned_loss=0.07098, over 28900.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3362, pruned_loss=0.09459, over 5692765.94 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3627, pruned_loss=0.124, over 5716922.75 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3348, pruned_loss=0.09157, over 5676004.57 frames. ], batch size: 284, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 21:47:17,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:47:21,472 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:47:28,282 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:47:54,116 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:48:15,808 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 21:48:18,987 INFO [train.py:968] (0/2) Epoch 9, batch 32550, libri_loss[loss=0.2718, simple_loss=0.3207, pruned_loss=0.1115, over 29697.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3352, pruned_loss=0.09446, over 5683838.49 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3626, pruned_loss=0.124, over 5721042.76 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3336, pruned_loss=0.09137, over 5666105.02 frames. ], batch size: 73, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 21:48:45,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.078e+02 1.388e+03 1.748e+03 2.546e+03 6.048e+03, threshold=3.497e+03, percent-clipped=10.0
+2023-03-04 21:48:48,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:48:49,049 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:48:51,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:13,247 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:13,698 INFO [train.py:968] (0/2) Epoch 9, batch 32600, giga_loss[loss=0.2751, simple_loss=0.3494, pruned_loss=0.1004, over 28860.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3376, pruned_loss=0.09593, over 5679014.54 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3628, pruned_loss=0.1241, over 5716458.85 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3356, pruned_loss=0.09277, over 5668286.80 frames. ], batch size: 174, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 21:49:16,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:18,522 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:20,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396804.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 21:49:21,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:22,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2195, 3.0299, 1.2586, 1.4423], device='cuda:0'), covar=tensor([0.1172, 0.0461, 0.1085, 0.1497], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0500, 0.0333, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 21:49:26,069 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:51,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:57,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:50:11,683 INFO [train.py:968] (0/2) Epoch 9, batch 32650, libri_loss[loss=0.2498, simple_loss=0.305, pruned_loss=0.09734, over 29686.00 frames. ], tot_loss[loss=0.264, simple_loss=0.337, pruned_loss=0.09552, over 5684370.92 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3627, pruned_loss=0.1245, over 5722879.45 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3346, pruned_loss=0.09168, over 5668619.27 frames. ], batch size: 69, lr: 3.55e-03, grad_scale: 1.0
+2023-03-04 21:50:44,344 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.158e+02 1.337e+03 1.871e+03 2.456e+03 1.595e+04, threshold=3.742e+03, percent-clipped=14.0
+2023-03-04 21:51:11,965 INFO [train.py:968] (0/2) Epoch 9, batch 32700, giga_loss[loss=0.2614, simple_loss=0.3402, pruned_loss=0.09127, over 28661.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3349, pruned_loss=0.09345, over 5674988.48 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3623, pruned_loss=0.1243, over 5724229.13 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3328, pruned_loss=0.08986, over 5660162.85 frames. ], batch size: 262, lr: 3.55e-03, grad_scale: 1.0
+2023-03-04 21:51:38,546 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:51:43,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7141, 5.5231, 5.1871, 2.5844], device='cuda:0'), covar=tensor([0.0349, 0.0579, 0.0717, 0.1736], device='cuda:0'), in_proj_covar=tensor([0.0976, 0.0913, 0.0800, 0.0623], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 21:51:43,582 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:52:04,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6902, 2.1669, 1.4636, 1.8180], device='cuda:0'), covar=tensor([0.0633, 0.0227, 0.0292, 0.0750], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0118, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0055, 0.0049, 0.0084], device='cuda:0')
+2023-03-04 21:52:08,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8778, 3.6957, 3.4721, 1.6415], device='cuda:0'), covar=tensor([0.0663, 0.0805, 0.0847, 0.2155], device='cuda:0'), in_proj_covar=tensor([0.0975, 0.0912, 0.0799, 0.0621], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 21:52:14,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396947.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 21:52:14,991 INFO [train.py:968] (0/2) Epoch 9, batch 32750, giga_loss[loss=0.2773, simple_loss=0.3455, pruned_loss=0.1045, over 29082.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3326, pruned_loss=0.09223, over 5677008.67 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3616, pruned_loss=0.1239, over 5727803.18 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3308, pruned_loss=0.08892, over 5660714.06 frames. ], batch size: 120, lr: 3.55e-03, grad_scale: 1.0
+2023-03-04 21:52:18,049 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:52:18,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396950.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 21:52:53,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.612e+02 1.252e+03 1.633e+03 2.210e+03 5.279e+03, threshold=3.265e+03, percent-clipped=7.0
+2023-03-04 21:52:59,509 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396979.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 21:53:18,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6764, 4.4772, 4.2309, 2.1279], device='cuda:0'), covar=tensor([0.0516, 0.0793, 0.0802, 0.1995], device='cuda:0'), in_proj_covar=tensor([0.0981, 0.0915, 0.0803, 0.0628], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 21:53:20,886 INFO [train.py:968] (0/2) Epoch 9, batch 32800, giga_loss[loss=0.2349, simple_loss=0.3234, pruned_loss=0.07323, over 28901.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3323, pruned_loss=0.09168, over 5667555.91 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3619, pruned_loss=0.1242, over 5714498.47 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3298, pruned_loss=0.08797, over 5665882.98 frames. ], batch size: 227, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:54:30,653 INFO [train.py:968] (0/2) Epoch 9, batch 32850, giga_loss[loss=0.2362, simple_loss=0.3189, pruned_loss=0.07674, over 29005.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3332, pruned_loss=0.09172, over 5678946.47 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3616, pruned_loss=0.124, over 5718521.85 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.331, pruned_loss=0.08835, over 5673316.46 frames. ], batch size: 145, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:55:02,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.808e+02 1.172e+03 1.529e+03 2.454e+03 6.194e+03, threshold=3.058e+03, percent-clipped=8.0
+2023-03-04 21:55:33,237 INFO [train.py:968] (0/2) Epoch 9, batch 32900, giga_loss[loss=0.2696, simple_loss=0.3417, pruned_loss=0.09875, over 29078.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3347, pruned_loss=0.0931, over 5684006.16 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3613, pruned_loss=0.1239, over 5720215.82 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.333, pruned_loss=0.09026, over 5677847.96 frames. ], batch size: 285, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:56:25,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4913, 2.6088, 1.5554, 1.5895], device='cuda:0'), covar=tensor([0.0726, 0.0304, 0.0725, 0.1077], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0499, 0.0334, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 21:56:37,555 INFO [train.py:968] (0/2) Epoch 9, batch 32950, giga_loss[loss=0.2091, simple_loss=0.2753, pruned_loss=0.07144, over 24515.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3333, pruned_loss=0.09236, over 5674511.83 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3612, pruned_loss=0.1238, over 5722082.88 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3318, pruned_loss=0.08993, over 5667814.24 frames. ], batch size: 705, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:56:55,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3359, 1.7302, 1.4319, 1.5311], device='cuda:0'), covar=tensor([0.0703, 0.0393, 0.0319, 0.0717], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0118, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:0')
+2023-03-04 21:57:09,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.453e+02 1.245e+03 1.654e+03 2.417e+03 8.523e+03, threshold=3.307e+03, percent-clipped=15.0
+2023-03-04 21:57:33,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4246, 1.9838, 1.7829, 1.5271], device='cuda:0'), covar=tensor([0.0793, 0.0266, 0.0302, 0.0862], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0118, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:0')
+2023-03-04 21:57:37,714 INFO [train.py:968] (0/2) Epoch 9, batch 33000, giga_loss[loss=0.2341, simple_loss=0.3154, pruned_loss=0.07642, over 28371.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.334, pruned_loss=0.09117, over 5668955.53 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3613, pruned_loss=0.124, over 5724321.89 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3322, pruned_loss=0.08852, over 5660974.89 frames. ], batch size: 65, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:57:37,718 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 21:57:46,370 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2074, simple_loss=0.3068, pruned_loss=0.054, over 944034.00 frames. 
+2023-03-04 21:57:46,370 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 21:58:10,729 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=397219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:58:28,894 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-04 21:58:48,003 INFO [train.py:968] (0/2) Epoch 9, batch 33050, giga_loss[loss=0.257, simple_loss=0.3373, pruned_loss=0.08831, over 28150.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3382, pruned_loss=0.09274, over 5666430.11 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3613, pruned_loss=0.124, over 5725413.26 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3366, pruned_loss=0.09035, over 5658666.14 frames. ], batch size: 412, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:59:17,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5621, 1.7900, 1.5153, 1.9368], device='cuda:0'), covar=tensor([0.2128, 0.1890, 0.1939, 0.1856], device='cuda:0'), in_proj_covar=tensor([0.1244, 0.0922, 0.1107, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 21:59:22,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.199e+02 1.410e+03 1.895e+03 2.854e+03 8.763e+03, threshold=3.791e+03, percent-clipped=18.0
+2023-03-04 21:59:50,803 INFO [train.py:968] (0/2) Epoch 9, batch 33100, giga_loss[loss=0.2942, simple_loss=0.3673, pruned_loss=0.1105, over 28973.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3392, pruned_loss=0.09282, over 5673075.26 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3609, pruned_loss=0.1238, over 5727003.22 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3379, pruned_loss=0.09071, over 5664760.85 frames. ], batch size: 199, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:00:12,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1434, 3.0149, 1.2634, 1.3308], device='cuda:0'), covar=tensor([0.1197, 0.0433, 0.1054, 0.1581], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0499, 0.0334, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 22:00:50,879 INFO [train.py:968] (0/2) Epoch 9, batch 33150, giga_loss[loss=0.253, simple_loss=0.3281, pruned_loss=0.08893, over 28680.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3403, pruned_loss=0.09438, over 5641908.68 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3614, pruned_loss=0.1242, over 5699437.18 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3381, pruned_loss=0.09113, over 5659495.12 frames. ], batch size: 119, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:01:20,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.014e+02 1.476e+03 2.017e+03 2.650e+03 1.205e+04, threshold=4.035e+03, percent-clipped=9.0
+2023-03-04 22:01:46,785 INFO [train.py:968] (0/2) Epoch 9, batch 33200, giga_loss[loss=0.2495, simple_loss=0.3399, pruned_loss=0.07954, over 28926.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3374, pruned_loss=0.09217, over 5660048.26 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3614, pruned_loss=0.1242, over 5702875.17 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3351, pruned_loss=0.08896, over 5669954.26 frames. ], batch size: 284, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:02:47,404 INFO [train.py:968] (0/2) Epoch 9, batch 33250, giga_loss[loss=0.2717, simple_loss=0.3438, pruned_loss=0.09983, over 28660.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3357, pruned_loss=0.09101, over 5669118.22 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3611, pruned_loss=0.124, over 5706653.71 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3337, pruned_loss=0.08806, over 5673015.75 frames. ], batch size: 307, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:03:17,569 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.269e+02 1.373e+03 1.891e+03 2.619e+03 8.972e+03, threshold=3.783e+03, percent-clipped=9.0
+2023-03-04 22:03:38,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5372, 1.7265, 1.3983, 1.9284], device='cuda:0'), covar=tensor([0.2397, 0.2298, 0.2569, 0.2294], device='cuda:0'), in_proj_covar=tensor([0.1240, 0.0914, 0.1102, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 22:03:41,439 INFO [train.py:968] (0/2) Epoch 9, batch 33300, giga_loss[loss=0.2469, simple_loss=0.3309, pruned_loss=0.08149, over 28993.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3348, pruned_loss=0.09186, over 5679974.73 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3606, pruned_loss=0.1238, over 5713370.33 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3325, pruned_loss=0.08817, over 5675454.94 frames. ], batch size: 145, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:04:39,808 INFO [train.py:968] (0/2) Epoch 9, batch 33350, giga_loss[loss=0.2591, simple_loss=0.3349, pruned_loss=0.09166, over 27565.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3356, pruned_loss=0.09235, over 5669209.30 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3607, pruned_loss=0.1237, over 5705028.80 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.333, pruned_loss=0.08858, over 5671009.50 frames. ], batch size: 472, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:04:55,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9800, 1.3422, 1.1220, 0.1014], device='cuda:0'), covar=tensor([0.2139, 0.1923, 0.2821, 0.3855], device='cuda:0'), in_proj_covar=tensor([0.1497, 0.1427, 0.1452, 0.1223], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 22:05:01,421 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.34 vs. limit=5.0
+2023-03-04 22:05:15,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.078e+02 1.316e+03 1.567e+03 2.144e+03 4.406e+03, threshold=3.135e+03, percent-clipped=1.0
+2023-03-04 22:05:42,586 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=397594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:05:46,589 INFO [train.py:968] (0/2) Epoch 9, batch 33400, giga_loss[loss=0.2781, simple_loss=0.3487, pruned_loss=0.1037, over 28459.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3387, pruned_loss=0.0941, over 5660622.68 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3605, pruned_loss=0.1238, over 5694782.72 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3365, pruned_loss=0.09086, over 5670986.86 frames. ], batch size: 370, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:06:33,514 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 22:06:52,046 INFO [train.py:968] (0/2) Epoch 9, batch 33450, giga_loss[loss=0.2716, simple_loss=0.3518, pruned_loss=0.09571, over 28489.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3399, pruned_loss=0.09533, over 5652672.93 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3607, pruned_loss=0.124, over 5686330.80 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3376, pruned_loss=0.092, over 5667496.71 frames. ], batch size: 336, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:06:56,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3275, 1.7894, 1.3065, 0.6020], device='cuda:0'), covar=tensor([0.2337, 0.1377, 0.2242, 0.3296], device='cuda:0'), in_proj_covar=tensor([0.1501, 0.1435, 0.1460, 0.1231], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 22:07:33,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.063e+02 1.485e+03 2.119e+03 3.416e+03 9.410e+03, threshold=4.238e+03, percent-clipped=29.0
+2023-03-04 22:07:47,796 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 22:07:54,434 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=397692.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 22:07:58,595 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-04 22:08:00,253 INFO [train.py:968] (0/2) Epoch 9, batch 33500, giga_loss[loss=0.2903, simple_loss=0.3643, pruned_loss=0.1081, over 28112.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3437, pruned_loss=0.0979, over 5643972.61 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3609, pruned_loss=0.1241, over 5689194.09 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3413, pruned_loss=0.0947, over 5652470.19 frames. ], batch size: 412, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:08:44,972 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=397737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:08:48,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=397740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:08:58,640 INFO [train.py:968] (0/2) Epoch 9, batch 33550, giga_loss[loss=0.2945, simple_loss=0.3716, pruned_loss=0.1087, over 29042.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3458, pruned_loss=0.0978, over 5652002.32 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.361, pruned_loss=0.1242, over 5689010.73 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3437, pruned_loss=0.09488, over 5658616.95 frames. ], batch size: 199, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:09:23,923 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=397769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:09:34,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.207e+02 1.270e+03 1.605e+03 1.965e+03 4.227e+03, threshold=3.211e+03, percent-clipped=0.0
+2023-03-04 22:10:01,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5057, 1.6019, 1.2035, 1.2540], device='cuda:0'), covar=tensor([0.0756, 0.0464, 0.1018, 0.0862], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0434, 0.0498, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 22:10:02,059 INFO [train.py:968] (0/2) Epoch 9, batch 33600, libri_loss[loss=0.3345, simple_loss=0.3826, pruned_loss=0.1432, over 28725.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3466, pruned_loss=0.09846, over 5649353.20 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3611, pruned_loss=0.1243, over 5683613.35 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3444, pruned_loss=0.09513, over 5658662.23 frames. ], batch size: 106, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:10:31,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-04 22:11:10,548 INFO [train.py:968] (0/2) Epoch 9, batch 33650, giga_loss[loss=0.3224, simple_loss=0.3739, pruned_loss=0.1355, over 27575.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3447, pruned_loss=0.0976, over 5660079.39 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.361, pruned_loss=0.1244, over 5688453.52 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3427, pruned_loss=0.09426, over 5662584.29 frames. ], batch size: 472, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:11:46,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.576e+02 1.540e+03 2.162e+03 3.629e+03 2.144e+04, threshold=4.323e+03, percent-clipped=29.0
+2023-03-04 22:12:13,811 INFO [train.py:968] (0/2) Epoch 9, batch 33700, giga_loss[loss=0.2277, simple_loss=0.3175, pruned_loss=0.06891, over 27994.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.342, pruned_loss=0.09583, over 5674503.76 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3603, pruned_loss=0.1239, over 5692228.53 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3406, pruned_loss=0.09297, over 5672699.49 frames. ], batch size: 412, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:12:50,606 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=397925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:13:20,996 INFO [train.py:968] (0/2) Epoch 9, batch 33750, giga_loss[loss=0.2651, simple_loss=0.34, pruned_loss=0.09512, over 28531.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3414, pruned_loss=0.09576, over 5665820.12 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3603, pruned_loss=0.124, over 5685347.86 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.34, pruned_loss=0.09314, over 5669692.59 frames. ], batch size: 336, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:14:00,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.885e+02 1.413e+03 1.891e+03 2.597e+03 5.395e+03, threshold=3.783e+03, percent-clipped=4.0
+2023-03-04 22:14:12,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6538, 1.6804, 1.2431, 1.3331], device='cuda:0'), covar=tensor([0.0668, 0.0508, 0.0939, 0.1032], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0434, 0.0498, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 22:14:28,471 INFO [train.py:968] (0/2) Epoch 9, batch 33800, giga_loss[loss=0.2926, simple_loss=0.3617, pruned_loss=0.1118, over 28440.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.34, pruned_loss=0.09611, over 5670817.08 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3601, pruned_loss=0.1238, over 5689835.74 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3388, pruned_loss=0.09364, over 5669551.09 frames. ], batch size: 369, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:14:34,125 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-398000.pt
+2023-03-04 22:15:21,466 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3762, 1.7260, 1.2756, 1.5276], device='cuda:0'), covar=tensor([0.0747, 0.0302, 0.0344, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0119, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0054, 0.0049, 0.0084], device='cuda:0')
+2023-03-04 22:15:33,720 INFO [train.py:968] (0/2) Epoch 9, batch 33850, giga_loss[loss=0.232, simple_loss=0.3253, pruned_loss=0.06933, over 28759.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3388, pruned_loss=0.09496, over 5673223.40 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3599, pruned_loss=0.1237, over 5684469.95 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3377, pruned_loss=0.09264, over 5676082.82 frames. ], batch size: 243, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:15:56,091 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=398067.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 22:16:05,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.393e+02 1.257e+03 1.713e+03 2.348e+03 5.472e+03, threshold=3.426e+03, percent-clipped=7.0
+2023-03-04 22:16:32,687 INFO [train.py:968] (0/2) Epoch 9, batch 33900, giga_loss[loss=0.2519, simple_loss=0.3375, pruned_loss=0.08316, over 28952.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3381, pruned_loss=0.09371, over 5664269.52 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3599, pruned_loss=0.1237, over 5679851.29 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3366, pruned_loss=0.09104, over 5670863.63 frames. ], batch size: 155, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:17:17,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1687, 1.7742, 1.2527, 0.3343], device='cuda:0'), covar=tensor([0.2394, 0.1514, 0.2641, 0.2990], device='cuda:0'), in_proj_covar=tensor([0.1495, 0.1434, 0.1458, 0.1233], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 22:17:30,876 INFO [train.py:968] (0/2) Epoch 9, batch 33950, giga_loss[loss=0.2545, simple_loss=0.347, pruned_loss=0.08105, over 28778.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3385, pruned_loss=0.0922, over 5671376.36 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3595, pruned_loss=0.1234, over 5683364.02 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3373, pruned_loss=0.08988, over 5673148.33 frames. ], batch size: 243, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:18:04,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.507e+02 1.259e+03 1.711e+03 2.312e+03 4.589e+03, threshold=3.421e+03, percent-clipped=3.0
+2023-03-04 22:18:30,892 INFO [train.py:968] (0/2) Epoch 9, batch 34000, giga_loss[loss=0.2581, simple_loss=0.3398, pruned_loss=0.08818, over 27649.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3403, pruned_loss=0.09154, over 5663346.50 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3594, pruned_loss=0.1233, over 5673565.10 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3392, pruned_loss=0.0894, over 5673707.54 frames. ], batch size: 474, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:18:42,528 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=398210.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 22:18:47,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=398213.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 22:19:22,693 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=398242.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 22:19:30,744 INFO [train.py:968] (0/2) Epoch 9, batch 34050, giga_loss[loss=0.2714, simple_loss=0.356, pruned_loss=0.0934, over 28656.00 frames. ], tot_loss[loss=0.262, simple_loss=0.341, pruned_loss=0.09151, over 5668496.98 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3592, pruned_loss=0.1232, over 5675482.56 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.34, pruned_loss=0.08945, over 5675071.62 frames. ], batch size: 242, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:20:14,365 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.120e+02 1.214e+03 1.583e+03 2.193e+03 8.518e+03, threshold=3.167e+03, percent-clipped=5.0
+2023-03-04 22:20:17,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-04 22:20:36,385 INFO [train.py:968] (0/2) Epoch 9, batch 34100, giga_loss[loss=0.2667, simple_loss=0.3462, pruned_loss=0.09358, over 28586.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3415, pruned_loss=0.09284, over 5664320.38 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3594, pruned_loss=0.1233, over 5673656.66 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3398, pruned_loss=0.08994, over 5671065.80 frames. ], batch size: 78, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:20:39,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=398300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:21:15,250 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:21:40,423 INFO [train.py:968] (0/2) Epoch 9, batch 34150, giga_loss[loss=0.3052, simple_loss=0.3751, pruned_loss=0.1177, over 29032.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3412, pruned_loss=0.09271, over 5667503.49 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.359, pruned_loss=0.1229, over 5679556.93 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3398, pruned_loss=0.0899, over 5667603.62 frames. ], batch size: 285, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:22:21,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.199e+02 1.338e+03 2.027e+03 2.964e+03 1.696e+04, threshold=4.054e+03, percent-clipped=20.0
+2023-03-04 22:22:47,539 INFO [train.py:968] (0/2) Epoch 9, batch 34200, giga_loss[loss=0.2222, simple_loss=0.32, pruned_loss=0.06216, over 28870.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3423, pruned_loss=0.09277, over 5672129.84 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3593, pruned_loss=0.123, over 5684826.66 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3404, pruned_loss=0.08962, over 5667081.02 frames. ], batch size: 164, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:23:56,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=398443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:24:00,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=398446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:24:01,652 INFO [train.py:968] (0/2) Epoch 9, batch 34250, giga_loss[loss=0.2464, simple_loss=0.3301, pruned_loss=0.08136, over 28961.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3432, pruned_loss=0.09317, over 5664798.03 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3594, pruned_loss=0.1232, over 5679069.47 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3413, pruned_loss=0.09, over 5665603.54 frames. ], batch size: 186, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:24:37,103 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=398475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:24:40,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.675e+02 1.371e+03 2.014e+03 2.937e+03 9.053e+03, threshold=4.028e+03, percent-clipped=6.0
+2023-03-04 22:24:53,819 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:25:07,045 INFO [train.py:968] (0/2) Epoch 9, batch 34300, giga_loss[loss=0.2895, simple_loss=0.368, pruned_loss=0.1055, over 29077.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3468, pruned_loss=0.09534, over 5666855.47 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.359, pruned_loss=0.1231, over 5681109.31 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3452, pruned_loss=0.09202, over 5665299.83 frames. ], batch size: 200, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:25:42,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:26:06,132 INFO [train.py:968] (0/2) Epoch 9, batch 34350, giga_loss[loss=0.2582, simple_loss=0.3372, pruned_loss=0.08962, over 28185.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3454, pruned_loss=0.09422, over 5681826.15 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3589, pruned_loss=0.1228, over 5689212.19 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3437, pruned_loss=0.09076, over 5672900.55 frames. ], batch size: 412, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:26:50,650 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.631e+02 1.290e+03 1.688e+03 2.265e+03 1.091e+04, threshold=3.376e+03, percent-clipped=8.0
+2023-03-04 22:27:15,781 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-04 22:27:20,861 INFO [train.py:968] (0/2) Epoch 9, batch 34400, libri_loss[loss=0.3053, simple_loss=0.3712, pruned_loss=0.1197, over 28146.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3436, pruned_loss=0.09389, over 5678738.06 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3586, pruned_loss=0.1225, over 5692076.34 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3423, pruned_loss=0.09081, over 5669093.24 frames. ], batch size: 116, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:28:28,060 INFO [train.py:968] (0/2) Epoch 9, batch 34450, giga_loss[loss=0.2415, simple_loss=0.3329, pruned_loss=0.07507, over 27913.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3412, pruned_loss=0.09271, over 5676418.20 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3589, pruned_loss=0.1229, over 5685130.41 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3393, pruned_loss=0.08896, over 5675582.44 frames. ], batch size: 476, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:29:10,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.798e+02 1.138e+03 1.439e+03 1.898e+03 3.633e+03, threshold=2.878e+03, percent-clipped=1.0
+2023-03-04 22:29:18,922 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:29:34,117 INFO [train.py:968] (0/2) Epoch 9, batch 34500, giga_loss[loss=0.2402, simple_loss=0.3264, pruned_loss=0.077, over 28669.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3399, pruned_loss=0.09123, over 5674578.66 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3594, pruned_loss=0.1232, over 5680511.17 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3373, pruned_loss=0.08711, over 5678556.10 frames. ], batch size: 307, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:29:38,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=398702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:29:57,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8787, 3.7019, 3.4606, 1.8534], device='cuda:0'), covar=tensor([0.0660, 0.0750, 0.0794, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.0974, 0.0899, 0.0797, 0.0622], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 22:30:39,426 INFO [train.py:968] (0/2) Epoch 9, batch 34550, giga_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.0892, over 28897.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3398, pruned_loss=0.09134, over 5670886.31 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3592, pruned_loss=0.123, over 5684751.73 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3377, pruned_loss=0.08768, over 5670313.90 frames. ], batch size: 227, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:31:13,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.113e+02 1.314e+03 2.115e+03 2.885e+03 1.067e+04, threshold=4.231e+03, percent-clipped=24.0
+2023-03-04 22:31:36,046 INFO [train.py:968] (0/2) Epoch 9, batch 34600, giga_loss[loss=0.2955, simple_loss=0.3791, pruned_loss=0.1059, over 28731.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3429, pruned_loss=0.09337, over 5665310.71 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3589, pruned_loss=0.1228, over 5681579.70 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3409, pruned_loss=0.08966, over 5666695.69 frames. ], batch size: 262, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:32:25,349 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:32:33,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=398845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:32:36,977 INFO [train.py:968] (0/2) Epoch 9, batch 34650, giga_loss[loss=0.2768, simple_loss=0.3462, pruned_loss=0.1037, over 28650.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.344, pruned_loss=0.09415, over 5671875.52 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3591, pruned_loss=0.123, over 5677828.59 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.342, pruned_loss=0.09041, over 5676043.76 frames. ], batch size: 262, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:32:37,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=398848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:33:01,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=398865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:33:02,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8399, 4.6763, 4.3970, 2.0428], device='cuda:0'), covar=tensor([0.0387, 0.0549, 0.0610, 0.2094], device='cuda:0'), in_proj_covar=tensor([0.0967, 0.0898, 0.0794, 0.0619], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 22:33:15,254 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=398877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:33:15,610 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.688e+02 1.281e+03 1.574e+03 2.471e+03 5.875e+03, threshold=3.148e+03, percent-clipped=1.0
+2023-03-04 22:33:37,374 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-04 22:33:38,924 INFO [train.py:968] (0/2) Epoch 9, batch 34700, giga_loss[loss=0.224, simple_loss=0.3161, pruned_loss=0.066, over 28774.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3412, pruned_loss=0.09383, over 5661502.80 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3591, pruned_loss=0.1231, over 5678388.55 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3393, pruned_loss=0.09058, over 5664217.37 frames. ], batch size: 174, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:33:44,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=398903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:33:56,220 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:34:36,131 INFO [train.py:968] (0/2) Epoch 9, batch 34750, giga_loss[loss=0.2532, simple_loss=0.337, pruned_loss=0.08473, over 28735.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3398, pruned_loss=0.0936, over 5663329.90 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3586, pruned_loss=0.1226, over 5683087.35 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3386, pruned_loss=0.09087, over 5660832.60 frames. ], batch size: 262, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:35:14,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.120e+02 1.405e+03 1.982e+03 2.935e+03 1.080e+04, threshold=3.964e+03, percent-clipped=20.0
+2023-03-04 22:35:32,288 INFO [train.py:968] (0/2) Epoch 9, batch 34800, giga_loss[loss=0.3165, simple_loss=0.3933, pruned_loss=0.1199, over 29049.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3447, pruned_loss=0.09647, over 5671361.11 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3585, pruned_loss=0.1226, over 5686450.03 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3435, pruned_loss=0.09389, over 5666133.82 frames. ], batch size: 155, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:35:38,557 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6326, 1.8238, 1.8489, 1.4463], device='cuda:0'), covar=tensor([0.1700, 0.2090, 0.1285, 0.1473], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0686, 0.0833, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 22:35:43,501 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:35:46,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:35:49,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2791, 4.0320, 3.7844, 1.6939], device='cuda:0'), covar=tensor([0.0558, 0.0784, 0.0764, 0.2392], device='cuda:0'), in_proj_covar=tensor([0.0961, 0.0894, 0.0790, 0.0617], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 22:35:51,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4593, 1.5151, 1.0885, 1.2326], device='cuda:0'), covar=tensor([0.0656, 0.0450, 0.0977, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0433, 0.0498, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 22:36:09,341 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:36:17,375 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:36:18,326 INFO [train.py:968] (0/2) Epoch 9, batch 34850, giga_loss[loss=0.3069, simple_loss=0.3826, pruned_loss=0.1156, over 28785.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3544, pruned_loss=0.1028, over 5661608.43 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3586, pruned_loss=0.1227, over 5679491.54 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3531, pruned_loss=0.1002, over 5663797.51 frames. ], batch size: 119, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:36:20,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:36:22,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9592, 3.7324, 3.5326, 1.6829], device='cuda:0'), covar=tensor([0.0581, 0.0750, 0.0734, 0.2272], device='cuda:0'), in_proj_covar=tensor([0.0963, 0.0898, 0.0794, 0.0620], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 22:36:31,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=399061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:36:47,478 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:36:48,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.335e+02 1.268e+03 1.692e+03 2.290e+03 4.032e+03, threshold=3.383e+03, percent-clipped=1.0
+2023-03-04 22:37:06,051 INFO [train.py:968] (0/2) Epoch 9, batch 34900, giga_loss[loss=0.2658, simple_loss=0.3399, pruned_loss=0.09585, over 28771.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3601, pruned_loss=0.1067, over 5662561.93 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3587, pruned_loss=0.1227, over 5679594.63 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3591, pruned_loss=0.1042, over 5663846.36 frames. ], batch size: 119, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:37:47,719 INFO [train.py:968] (0/2) Epoch 9, batch 34950, giga_loss[loss=0.2385, simple_loss=0.3199, pruned_loss=0.07854, over 28606.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3581, pruned_loss=0.1068, over 5662774.41 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3593, pruned_loss=0.1232, over 5668458.74 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3567, pruned_loss=0.104, over 5674497.53 frames. ], batch size: 60, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:37:52,275 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=399154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:38:13,045 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.046e+02 1.099e+03 1.505e+03 2.092e+03 8.421e+03, threshold=3.009e+03, percent-clipped=8.0
+2023-03-04 22:38:30,926 INFO [train.py:968] (0/2) Epoch 9, batch 35000, giga_loss[loss=0.2314, simple_loss=0.3104, pruned_loss=0.07618, over 29117.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3516, pruned_loss=0.1044, over 5667923.85 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3598, pruned_loss=0.1235, over 5663145.16 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.35, pruned_loss=0.1016, over 5681201.82 frames. ], batch size: 155, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:38:36,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:38:39,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:38:45,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=399214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:39:05,263 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:39:08,637 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1956, 3.9989, 3.7780, 1.9819], device='cuda:0'), covar=tensor([0.0417, 0.0603, 0.0595, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.0968, 0.0903, 0.0800, 0.0624], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 22:39:15,559 INFO [train.py:968] (0/2) Epoch 9, batch 35050, giga_loss[loss=0.2366, simple_loss=0.3113, pruned_loss=0.08097, over 28575.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3439, pruned_loss=0.1007, over 5668081.29 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3601, pruned_loss=0.1236, over 5665064.41 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3422, pruned_loss=0.09825, over 5676887.18 frames. ], batch size: 336, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:39:16,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-04 22:39:23,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=399257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:39:39,576 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.875e+02 9.328e+02 1.354e+03 1.967e+03 4.879e+03, threshold=2.708e+03, percent-clipped=7.0
+2023-03-04 22:39:46,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=399288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:39:56,227 INFO [train.py:968] (0/2) Epoch 9, batch 35100, giga_loss[loss=0.2175, simple_loss=0.2905, pruned_loss=0.07224, over 28822.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3364, pruned_loss=0.09769, over 5677611.81 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3602, pruned_loss=0.1235, over 5671656.24 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3345, pruned_loss=0.09509, over 5679100.24 frames. ], batch size: 199, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:40:35,454 INFO [train.py:968] (0/2) Epoch 9, batch 35150, giga_loss[loss=0.2176, simple_loss=0.2935, pruned_loss=0.07088, over 28461.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3297, pruned_loss=0.09465, over 5671415.12 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3606, pruned_loss=0.1238, over 5657026.25 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3271, pruned_loss=0.09159, over 5686074.00 frames. ], batch size: 71, lr: 3.54e-03, grad_scale: 2.0
+2023-03-04 22:40:45,761 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:40:48,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:40:58,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6570, 2.2807, 1.4933, 0.8059], device='cuda:0'), covar=tensor([0.5236, 0.2444, 0.2708, 0.4679], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1428, 0.1455, 0.1224], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 22:40:58,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5667, 1.6765, 1.3974, 2.0530], device='cuda:0'), covar=tensor([0.2314, 0.2275, 0.2393, 0.2123], device='cuda:0'), in_proj_covar=tensor([0.1248, 0.0925, 0.1110, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 22:41:04,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.433e+02 1.049e+03 1.363e+03 1.829e+03 5.156e+03, threshold=2.726e+03, percent-clipped=10.0
+2023-03-04 22:41:05,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2870, 1.4218, 1.5454, 1.2863], device='cuda:0'), covar=tensor([0.1388, 0.1566, 0.1844, 0.1553], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0723, 0.0647, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 22:41:14,318 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:41:20,707 INFO [train.py:968] (0/2) Epoch 9, batch 35200, libri_loss[loss=0.3257, simple_loss=0.382, pruned_loss=0.1347, over 29387.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3239, pruned_loss=0.09209, over 5668170.54 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3608, pruned_loss=0.1239, over 5659392.63 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3213, pruned_loss=0.08923, over 5677510.66 frames. ], batch size: 92, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:41:28,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-04 22:41:31,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7640, 1.9203, 1.6045, 2.0086], device='cuda:0'), covar=tensor([0.2193, 0.2097, 0.2328, 0.1995], device='cuda:0'), in_proj_covar=tensor([0.1245, 0.0921, 0.1105, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 22:41:33,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1217, 4.9052, 4.6183, 2.2543], device='cuda:0'), covar=tensor([0.0394, 0.0610, 0.0687, 0.1904], device='cuda:0'), in_proj_covar=tensor([0.0968, 0.0903, 0.0799, 0.0626], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 22:41:49,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:41:51,581 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:42:03,238 INFO [train.py:968] (0/2) Epoch 9, batch 35250, giga_loss[loss=0.2315, simple_loss=0.2992, pruned_loss=0.08184, over 28606.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3213, pruned_loss=0.09049, over 5681305.09 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3612, pruned_loss=0.1239, over 5665336.97 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3179, pruned_loss=0.08751, over 5683751.65 frames. ], batch size: 71, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:42:15,548 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:42:23,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 22:42:28,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.258e+02 1.013e+03 1.366e+03 1.858e+03 8.297e+03, threshold=2.731e+03, percent-clipped=5.0
+2023-03-04 22:42:30,531 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=399481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:42:43,874 INFO [train.py:968] (0/2) Epoch 9, batch 35300, libri_loss[loss=0.3888, simple_loss=0.422, pruned_loss=0.1778, over 25530.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3185, pruned_loss=0.08902, over 5689127.49 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3623, pruned_loss=0.1246, over 5664755.15 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.3139, pruned_loss=0.08517, over 5692478.34 frames. ], batch size: 136, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:42:58,689 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3995, 1.7106, 1.4173, 1.5494], device='cuda:0'), covar=tensor([0.0712, 0.0330, 0.0312, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0117, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:0')
+2023-03-04 22:43:08,399 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=399529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:43:27,983 INFO [train.py:968] (0/2) Epoch 9, batch 35350, giga_loss[loss=0.2322, simple_loss=0.2988, pruned_loss=0.08277, over 28950.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3146, pruned_loss=0.08681, over 5699795.49 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.362, pruned_loss=0.1244, over 5668474.94 frames. ], giga_tot_loss[loss=0.2388, simple_loss=0.3105, pruned_loss=0.0835, over 5699510.77 frames. ], batch size: 213, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:43:53,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.563e+02 9.498e+02 1.196e+03 1.811e+03 7.159e+03, threshold=2.391e+03, percent-clipped=7.0
+2023-03-04 22:43:56,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4981, 3.3669, 1.5606, 1.5912], device='cuda:0'), covar=tensor([0.0888, 0.0282, 0.0855, 0.1259], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0496, 0.0331, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 22:44:08,647 INFO [train.py:968] (0/2) Epoch 9, batch 35400, libri_loss[loss=0.3225, simple_loss=0.3908, pruned_loss=0.1271, over 29546.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.311, pruned_loss=0.08483, over 5702648.58 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3626, pruned_loss=0.1246, over 5672024.49 frames. ], giga_tot_loss[loss=0.2347, simple_loss=0.3066, pruned_loss=0.08147, over 5699664.08 frames. ], batch size: 83, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:44:38,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=399632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:44:40,065 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-04 22:44:51,556 INFO [train.py:968] (0/2) Epoch 9, batch 35450, libri_loss[loss=0.2866, simple_loss=0.3562, pruned_loss=0.1085, over 29515.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3086, pruned_loss=0.08394, over 5696233.31 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3626, pruned_loss=0.1244, over 5672721.30 frames. ], giga_tot_loss[loss=0.2328, simple_loss=0.3042, pruned_loss=0.08073, over 5693727.18 frames. ], batch size: 81, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:45:10,742 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:45:13,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:45:17,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.103e+02 9.521e+02 1.213e+03 1.530e+03 4.529e+03, threshold=2.425e+03, percent-clipped=9.0
+2023-03-04 22:45:31,803 INFO [train.py:968] (0/2) Epoch 9, batch 35500, giga_loss[loss=0.2464, simple_loss=0.3155, pruned_loss=0.0886, over 28191.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3072, pruned_loss=0.08356, over 5694349.76 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3621, pruned_loss=0.1238, over 5678845.52 frames. ], giga_tot_loss[loss=0.231, simple_loss=0.302, pruned_loss=0.07999, over 5687563.88 frames. ], batch size: 368, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:45:36,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:45:50,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8510, 1.9026, 1.3107, 1.5070], device='cuda:0'), covar=tensor([0.0735, 0.0588, 0.1089, 0.1014], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0437, 0.0501, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 22:46:03,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8608, 2.5860, 1.7886, 0.7907], device='cuda:0'), covar=tensor([0.4675, 0.2365, 0.2601, 0.4690], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1423, 0.1448, 0.1218], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 22:46:18,181 INFO [train.py:968] (0/2) Epoch 9, batch 35550, giga_loss[loss=0.2075, simple_loss=0.2819, pruned_loss=0.06653, over 28665.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3036, pruned_loss=0.08195, over 5691077.79 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3622, pruned_loss=0.1239, over 5670636.39 frames. ], giga_tot_loss[loss=0.2282, simple_loss=0.299, pruned_loss=0.07874, over 5694119.44 frames. ], batch size: 242, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:46:33,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=399765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:46:43,048 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:46:45,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:46:47,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.387e+02 9.859e+02 1.295e+03 1.722e+03 4.994e+03, threshold=2.590e+03, percent-clipped=13.0
+2023-03-04 22:47:03,007 INFO [train.py:968] (0/2) Epoch 9, batch 35600, giga_loss[loss=0.3031, simple_loss=0.3598, pruned_loss=0.1232, over 23528.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3033, pruned_loss=0.08223, over 5685180.05 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3624, pruned_loss=0.1239, over 5674136.24 frames. ], giga_tot_loss[loss=0.2283, simple_loss=0.2985, pruned_loss=0.07906, over 5684603.68 frames. ], batch size: 705, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:47:10,306 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:47:47,924 INFO [train.py:968] (0/2) Epoch 9, batch 35650, giga_loss[loss=0.2766, simple_loss=0.3453, pruned_loss=0.104, over 28523.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3151, pruned_loss=0.08897, over 5676781.58 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3632, pruned_loss=0.1244, over 5667018.81 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3093, pruned_loss=0.08499, over 5684118.48 frames. ], batch size: 78, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:47:53,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=399856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:47:54,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5493, 1.6998, 1.5314, 1.3354], device='cuda:0'), covar=tensor([0.1783, 0.1463, 0.1187, 0.1529], device='cuda:0'), in_proj_covar=tensor([0.1651, 0.1487, 0.1444, 0.1578], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 22:47:54,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5273, 2.1540, 1.6291, 0.6104], device='cuda:0'), covar=tensor([0.3430, 0.1837, 0.2479, 0.3960], device='cuda:0'), in_proj_covar=tensor([0.1488, 0.1415, 0.1447, 0.1215], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 22:48:15,311 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.508e+02 1.275e+03 1.664e+03 2.130e+03 8.168e+03, threshold=3.329e+03, percent-clipped=12.0
+2023-03-04 22:48:25,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5992, 1.6404, 1.8602, 1.4426], device='cuda:0'), covar=tensor([0.1305, 0.1829, 0.1011, 0.1284], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0702, 0.0844, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-04 22:48:30,777 INFO [train.py:968] (0/2) Epoch 9, batch 35700, giga_loss[loss=0.3384, simple_loss=0.3836, pruned_loss=0.1466, over 23789.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3295, pruned_loss=0.09684, over 5673576.30 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3636, pruned_loss=0.1246, over 5662450.63 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.323, pruned_loss=0.09245, over 5683241.52 frames. ], batch size: 705, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:49:09,917 INFO [train.py:968] (0/2) Epoch 9, batch 35750, giga_loss[loss=0.3563, simple_loss=0.4172, pruned_loss=0.1477, over 28759.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3417, pruned_loss=0.1035, over 5680838.03 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3638, pruned_loss=0.1247, over 5661200.29 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3354, pruned_loss=0.09912, over 5689710.00 frames. ], batch size: 92, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:49:39,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.576e+02 1.196e+03 1.471e+03 2.211e+03 7.729e+03, threshold=2.942e+03, percent-clipped=7.0
+2023-03-04 22:49:45,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4948, 4.2901, 4.0578, 1.9839], device='cuda:0'), covar=tensor([0.0461, 0.0609, 0.0655, 0.1962], device='cuda:0'), in_proj_covar=tensor([0.0972, 0.0915, 0.0804, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 22:49:55,028 INFO [train.py:968] (0/2) Epoch 9, batch 35800, giga_loss[loss=0.2718, simple_loss=0.358, pruned_loss=0.09275, over 28774.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3471, pruned_loss=0.1046, over 5683256.70 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3638, pruned_loss=0.1247, over 5662427.29 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3422, pruned_loss=0.1011, over 5689261.30 frames. ], batch size: 243, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:49:57,183 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:49:57,658 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-400000.pt
+2023-03-04 22:49:59,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:50:07,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3976, 2.2449, 2.0351, 2.0821], device='cuda:0'), covar=tensor([0.1405, 0.1938, 0.1811, 0.1683], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0726, 0.0654, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 22:50:24,091 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=400031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:50:30,591 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.81 vs. limit=5.0
+2023-03-04 22:50:37,932 INFO [train.py:968] (0/2) Epoch 9, batch 35850, giga_loss[loss=0.2915, simple_loss=0.3694, pruned_loss=0.1069, over 27823.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3493, pruned_loss=0.1042, over 5677065.71 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3641, pruned_loss=0.1248, over 5657992.04 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3448, pruned_loss=0.1008, over 5686825.94 frames. ], batch size: 412, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:51:06,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.840e+02 1.098e+03 1.405e+03 2.004e+03 5.462e+03, threshold=2.809e+03, percent-clipped=9.0
+2023-03-04 22:51:25,603 INFO [train.py:968] (0/2) Epoch 9, batch 35900, giga_loss[loss=0.3342, simple_loss=0.3929, pruned_loss=0.1377, over 27671.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3515, pruned_loss=0.1048, over 5670808.26 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3643, pruned_loss=0.1249, over 5653359.52 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3473, pruned_loss=0.1016, over 5683044.10 frames. ], batch size: 472, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:51:58,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:52:05,603 INFO [train.py:968] (0/2) Epoch 9, batch 35950, giga_loss[loss=0.3038, simple_loss=0.3611, pruned_loss=0.1232, over 23845.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3544, pruned_loss=0.107, over 5676460.79 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3646, pruned_loss=0.1251, over 5655455.37 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3507, pruned_loss=0.104, over 5684508.69 frames. ], batch size: 705, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:52:35,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.714e+02 1.167e+03 1.521e+03 2.145e+03 7.875e+03, threshold=3.041e+03, percent-clipped=12.0
+2023-03-04 22:52:48,852 INFO [train.py:968] (0/2) Epoch 9, batch 36000, giga_loss[loss=0.2704, simple_loss=0.349, pruned_loss=0.0959, over 28626.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.359, pruned_loss=0.1106, over 5671821.25 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3649, pruned_loss=0.1252, over 5656934.43 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3557, pruned_loss=0.1079, over 5677425.62 frames. ], batch size: 92, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:52:48,857 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 22:52:57,485 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2197, simple_loss=0.3259, pruned_loss=0.05677, over 944034.00 frames. 
+2023-03-04 22:52:57,486 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 22:53:35,310 INFO [train.py:968] (0/2) Epoch 9, batch 36050, giga_loss[loss=0.3086, simple_loss=0.3764, pruned_loss=0.1205, over 29063.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3627, pruned_loss=0.1131, over 5667143.69 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3656, pruned_loss=0.1257, over 5643490.08 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3591, pruned_loss=0.11, over 5685561.41 frames. ], batch size: 128, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:53:59,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.178e+02 1.182e+03 1.736e+03 2.365e+03 7.653e+03, threshold=3.471e+03, percent-clipped=13.0
+2023-03-04 22:54:01,591 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:54:03,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:54:11,897 INFO [train.py:968] (0/2) Epoch 9, batch 36100, giga_loss[loss=0.3411, simple_loss=0.3978, pruned_loss=0.1422, over 28271.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3652, pruned_loss=0.1141, over 5672573.70 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.366, pruned_loss=0.1259, over 5644768.37 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3618, pruned_loss=0.1108, over 5687252.60 frames. ], batch size: 368, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:54:25,345 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=400315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:54:52,935 INFO [train.py:968] (0/2) Epoch 9, batch 36150, giga_loss[loss=0.2847, simple_loss=0.3662, pruned_loss=0.1016, over 28585.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3662, pruned_loss=0.1136, over 5677345.72 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3665, pruned_loss=0.1259, over 5649720.40 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3631, pruned_loss=0.1108, over 5685035.29 frames. ], batch size: 336, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:54:57,109 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-04 22:55:18,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=400377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:55:20,616 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=400380.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 22:55:22,482 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.015e+02 1.268e+03 1.517e+03 2.243e+03 5.849e+03, threshold=3.033e+03, percent-clipped=7.0
+2023-03-04 22:55:33,790 INFO [train.py:968] (0/2) Epoch 9, batch 36200, giga_loss[loss=0.2893, simple_loss=0.3646, pruned_loss=0.107, over 28714.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.367, pruned_loss=0.1129, over 5682383.73 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3666, pruned_loss=0.1257, over 5655275.85 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3644, pruned_loss=0.1105, over 5684155.18 frames. ], batch size: 92, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:55:49,088 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=400417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:56:04,952 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-04 22:56:14,375 INFO [train.py:968] (0/2) Epoch 9, batch 36250, giga_loss[loss=0.2659, simple_loss=0.3435, pruned_loss=0.09412, over 28932.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.366, pruned_loss=0.1109, over 5687289.00 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3669, pruned_loss=0.1257, over 5652260.91 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3638, pruned_loss=0.1087, over 5691889.98 frames. ], batch size: 112, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:56:18,487 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=400453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:56:20,275 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=400455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:56:41,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.175e+02 9.891e+02 1.218e+03 1.787e+03 6.944e+03, threshold=2.435e+03, percent-clipped=7.0
+2023-03-04 22:56:56,779 INFO [train.py:968] (0/2) Epoch 9, batch 36300, giga_loss[loss=0.2536, simple_loss=0.3374, pruned_loss=0.08496, over 28982.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.364, pruned_loss=0.1081, over 5693168.69 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3669, pruned_loss=0.1257, over 5652260.91 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3623, pruned_loss=0.1063, over 5696749.70 frames. ], batch size: 145, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:56:57,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3096, 3.0912, 2.9794, 1.2684], device='cuda:0'), covar=tensor([0.0795, 0.0975, 0.0854, 0.2385], device='cuda:0'), in_proj_covar=tensor([0.0965, 0.0909, 0.0798, 0.0632], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 22:57:37,366 INFO [train.py:968] (0/2) Epoch 9, batch 36350, giga_loss[loss=0.3171, simple_loss=0.385, pruned_loss=0.1246, over 28760.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3627, pruned_loss=0.1068, over 5703632.50 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3672, pruned_loss=0.1257, over 5656735.72 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3609, pruned_loss=0.105, over 5703540.49 frames. ], batch size: 262, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:57:51,031 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-04 22:58:04,878 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.588e+02 1.079e+03 1.385e+03 2.175e+03 1.319e+04, threshold=2.769e+03, percent-clipped=18.0
+2023-03-04 22:58:21,180 INFO [train.py:968] (0/2) Epoch 9, batch 36400, giga_loss[loss=0.3654, simple_loss=0.4014, pruned_loss=0.1647, over 28844.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3656, pruned_loss=0.1111, over 5695096.25 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3676, pruned_loss=0.1261, over 5649578.58 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3639, pruned_loss=0.1092, over 5701627.89 frames. ], batch size: 199, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:59:00,707 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-04 22:59:02,299 INFO [train.py:968] (0/2) Epoch 9, batch 36450, giga_loss[loss=0.3323, simple_loss=0.3848, pruned_loss=0.1399, over 28773.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3693, pruned_loss=0.1165, over 5688365.37 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3682, pruned_loss=0.1264, over 5649247.84 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3674, pruned_loss=0.1143, over 5695544.41 frames. ], batch size: 243, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:59:29,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5099, 4.1249, 1.6404, 1.5443], device='cuda:0'), covar=tensor([0.1069, 0.0297, 0.0907, 0.1498], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0497, 0.0330, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 22:59:30,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.635e+02 1.295e+03 1.681e+03 2.321e+03 5.991e+03, threshold=3.362e+03, percent-clipped=17.0
+2023-03-04 22:59:44,651 INFO [train.py:968] (0/2) Epoch 9, batch 36500, libri_loss[loss=0.2642, simple_loss=0.3329, pruned_loss=0.09775, over 29598.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3691, pruned_loss=0.1176, over 5691587.50 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3683, pruned_loss=0.1264, over 5651100.20 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3675, pruned_loss=0.1158, over 5695988.03 frames. ], batch size: 74, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:59:53,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7750, 2.0577, 1.6674, 1.5295], device='cuda:0'), covar=tensor([0.1771, 0.1471, 0.1446, 0.1758], device='cuda:0'), in_proj_covar=tensor([0.1640, 0.1496, 0.1447, 0.1574], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 23:00:21,506 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9624, 2.7761, 2.5883, 1.5748], device='cuda:0'), covar=tensor([0.0992, 0.1045, 0.0930, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.0967, 0.0910, 0.0806, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 23:00:30,097 INFO [train.py:968] (0/2) Epoch 9, batch 36550, giga_loss[loss=0.2946, simple_loss=0.3574, pruned_loss=0.1159, over 28790.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3661, pruned_loss=0.116, over 5693532.22 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3683, pruned_loss=0.1262, over 5650841.59 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3648, pruned_loss=0.1145, over 5698105.67 frames. ], batch size: 119, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 23:00:33,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:00:35,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400755.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:00:45,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7675, 4.5608, 4.4457, 1.8601], device='cuda:0'), covar=tensor([0.0476, 0.0678, 0.0703, 0.1863], device='cuda:0'), in_proj_covar=tensor([0.0969, 0.0914, 0.0807, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 23:00:58,299 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.341e+02 1.218e+03 1.502e+03 2.154e+03 8.461e+03, threshold=3.004e+03, percent-clipped=9.0
+2023-03-04 23:01:04,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:01:09,639 INFO [train.py:968] (0/2) Epoch 9, batch 36600, giga_loss[loss=0.332, simple_loss=0.3806, pruned_loss=0.1417, over 28616.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3657, pruned_loss=0.1161, over 5702675.53 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3684, pruned_loss=0.1261, over 5658301.67 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3645, pruned_loss=0.1148, over 5700525.00 frames. ], batch size: 92, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 23:01:33,463 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:01:35,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:01:50,910 INFO [train.py:968] (0/2) Epoch 9, batch 36650, giga_loss[loss=0.2621, simple_loss=0.3386, pruned_loss=0.09282, over 28599.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3641, pruned_loss=0.1142, over 5701822.86 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3685, pruned_loss=0.126, over 5660873.93 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.363, pruned_loss=0.113, over 5699111.41 frames. ], batch size: 85, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 23:02:24,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.299e+02 1.144e+03 1.439e+03 1.862e+03 7.684e+03, threshold=2.879e+03, percent-clipped=14.0
+2023-03-04 23:02:35,172 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:02:36,864 INFO [train.py:968] (0/2) Epoch 9, batch 36700, giga_loss[loss=0.2767, simple_loss=0.3488, pruned_loss=0.1022, over 28468.00 frames. ], tot_loss[loss=0.292, simple_loss=0.361, pruned_loss=0.1115, over 5692898.67 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3684, pruned_loss=0.126, over 5663146.01 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3601, pruned_loss=0.1104, over 5689175.09 frames. ], batch size: 60, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 23:02:37,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:02:38,044 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400898.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:02:39,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400901.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:03:06,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=400927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:09,541 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=400930.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:03:15,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:16,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:26,001 INFO [train.py:968] (0/2) Epoch 9, batch 36750, giga_loss[loss=0.2451, simple_loss=0.2981, pruned_loss=0.09609, over 23200.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.356, pruned_loss=0.1088, over 5676768.60 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3684, pruned_loss=0.126, over 5664404.28 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3552, pruned_loss=0.1079, over 5672978.61 frames. ], batch size: 705, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:03:45,142 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=400967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:48,484 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:49,705 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:50,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:51,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:57,049 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.157e+02 9.767e+02 1.242e+03 1.700e+03 7.286e+03, threshold=2.484e+03, percent-clipped=8.0
+2023-03-04 23:04:02,424 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 23:04:10,786 INFO [train.py:968] (0/2) Epoch 9, batch 36800, giga_loss[loss=0.2408, simple_loss=0.3121, pruned_loss=0.08477, over 28704.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3504, pruned_loss=0.1063, over 5670101.97 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3694, pruned_loss=0.1266, over 5662973.18 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3484, pruned_loss=0.1043, over 5668741.91 frames. ], batch size: 85, lr: 3.53e-03, grad_scale: 8.0
+2023-03-04 23:04:17,127 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:04:18,403 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:04:24,233 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:05:01,708 INFO [train.py:968] (0/2) Epoch 9, batch 36850, giga_loss[loss=0.2555, simple_loss=0.3277, pruned_loss=0.09163, over 28558.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3449, pruned_loss=0.1037, over 5660483.27 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3689, pruned_loss=0.1263, over 5668429.48 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3431, pruned_loss=0.1019, over 5654611.96 frames. ], batch size: 307, lr: 3.53e-03, grad_scale: 8.0
+2023-03-04 23:05:02,115 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6172, 1.6703, 1.4987, 1.3616], device='cuda:0'), covar=tensor([0.1881, 0.1571, 0.1204, 0.1664], device='cuda:0'), in_proj_covar=tensor([0.1630, 0.1487, 0.1436, 0.1574], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 23:05:17,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:05:23,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3107, 1.5780, 1.5464, 1.4250], device='cuda:0'), covar=tensor([0.1456, 0.1340, 0.1805, 0.1434], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0728, 0.0655, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 23:05:31,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.402e+02 8.806e+02 1.202e+03 1.753e+03 5.300e+03, threshold=2.405e+03, percent-clipped=10.0
+2023-03-04 23:05:45,105 INFO [train.py:968] (0/2) Epoch 9, batch 36900, giga_loss[loss=0.254, simple_loss=0.33, pruned_loss=0.08899, over 28717.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3445, pruned_loss=0.1031, over 5668261.61 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3694, pruned_loss=0.1267, over 5673796.71 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3419, pruned_loss=0.1006, over 5658557.62 frames. ], batch size: 262, lr: 3.53e-03, grad_scale: 8.0
+2023-03-04 23:05:49,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9028, 1.2746, 1.0681, 0.1095], device='cuda:0'), covar=tensor([0.2683, 0.1859, 0.3511, 0.4169], device='cuda:0'), in_proj_covar=tensor([0.1502, 0.1424, 0.1452, 0.1215], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 23:05:52,019 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.24 vs. limit=5.0
+2023-03-04 23:06:23,513 INFO [train.py:968] (0/2) Epoch 9, batch 36950, giga_loss[loss=0.2877, simple_loss=0.3479, pruned_loss=0.1137, over 28808.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3449, pruned_loss=0.1024, over 5675394.96 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.37, pruned_loss=0.1267, over 5675716.80 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3416, pruned_loss=0.09974, over 5665901.90 frames. ], batch size: 99, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:06:53,305 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.332e+02 1.030e+03 1.230e+03 1.561e+03 5.190e+03, threshold=2.459e+03, percent-clipped=8.0
+2023-03-04 23:07:05,837 INFO [train.py:968] (0/2) Epoch 9, batch 37000, giga_loss[loss=0.2807, simple_loss=0.3526, pruned_loss=0.1044, over 28027.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3428, pruned_loss=0.1009, over 5684296.84 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3705, pruned_loss=0.127, over 5674608.62 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3395, pruned_loss=0.09833, over 5677921.13 frames. ], batch size: 412, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:07:20,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7839, 1.6619, 1.2220, 1.3419], device='cuda:0'), covar=tensor([0.0689, 0.0628, 0.0942, 0.1166], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0434, 0.0495, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 23:07:37,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:07:46,167 INFO [train.py:968] (0/2) Epoch 9, batch 37050, giga_loss[loss=0.2427, simple_loss=0.311, pruned_loss=0.08714, over 28676.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.339, pruned_loss=0.09901, over 5694757.64 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.371, pruned_loss=0.1272, over 5671445.50 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3358, pruned_loss=0.0966, over 5692474.26 frames. ], batch size: 85, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:07:52,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5173, 1.5983, 1.4588, 1.2039], device='cuda:0'), covar=tensor([0.1973, 0.1574, 0.1149, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.1629, 0.1485, 0.1434, 0.1577], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 23:08:06,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401274.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:08:13,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.116e+02 9.645e+02 1.178e+03 1.529e+03 6.221e+03, threshold=2.355e+03, percent-clipped=5.0
+2023-03-04 23:08:24,914 INFO [train.py:968] (0/2) Epoch 9, batch 37100, giga_loss[loss=0.2368, simple_loss=0.3241, pruned_loss=0.07475, over 28894.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3369, pruned_loss=0.09793, over 5689334.24 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3716, pruned_loss=0.1274, over 5664006.11 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3334, pruned_loss=0.09551, over 5695104.50 frames. ], batch size: 174, lr: 3.53e-03, grad_scale: 1.0
+2023-03-04 23:08:48,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3375, 3.1077, 2.9111, 1.8346], device='cuda:0'), covar=tensor([0.0775, 0.0895, 0.0820, 0.2115], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.0907, 0.0811, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 23:08:54,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3153, 1.8310, 1.4034, 1.4372], device='cuda:0'), covar=tensor([0.0792, 0.0307, 0.0317, 0.0816], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0082], device='cuda:0')
+2023-03-04 23:09:00,025 INFO [train.py:968] (0/2) Epoch 9, batch 37150, giga_loss[loss=0.2299, simple_loss=0.303, pruned_loss=0.07836, over 28699.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3379, pruned_loss=0.0993, over 5697481.66 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3733, pruned_loss=0.1284, over 5667959.00 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.332, pruned_loss=0.09516, over 5699676.99 frames. ], batch size: 85, lr: 3.53e-03, grad_scale: 1.0
+2023-03-04 23:09:06,817 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:09:26,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2655, 3.0444, 1.3729, 1.3432], device='cuda:0'), covar=tensor([0.0976, 0.0351, 0.0849, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0495, 0.0328, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 23:09:27,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.316e+02 9.192e+02 1.433e+03 2.188e+03 1.163e+04, threshold=2.867e+03, percent-clipped=20.0
+2023-03-04 23:09:28,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=401386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:09:39,275 INFO [train.py:968] (0/2) Epoch 9, batch 37200, giga_loss[loss=0.2769, simple_loss=0.3496, pruned_loss=0.1021, over 27986.00 frames. ], tot_loss[loss=0.264, simple_loss=0.334, pruned_loss=0.09703, over 5705648.65 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3736, pruned_loss=0.1284, over 5671217.49 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3285, pruned_loss=0.09333, over 5704933.61 frames. ], batch size: 412, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:10:09,781 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=401439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:10:17,762 INFO [train.py:968] (0/2) Epoch 9, batch 37250, libri_loss[loss=0.3529, simple_loss=0.4223, pruned_loss=0.1417, over 29273.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3317, pruned_loss=0.09551, over 5707080.80 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3741, pruned_loss=0.1285, over 5663096.00 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3261, pruned_loss=0.09175, over 5714730.20 frames. ], batch size: 101, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:10:46,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.389e+02 9.599e+02 1.311e+03 2.036e+03 7.298e+03, threshold=2.622e+03, percent-clipped=9.0
+2023-03-04 23:10:55,281 INFO [train.py:968] (0/2) Epoch 9, batch 37300, giga_loss[loss=0.2291, simple_loss=0.3069, pruned_loss=0.07569, over 29048.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3328, pruned_loss=0.09646, over 5694542.52 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.376, pruned_loss=0.1296, over 5651494.94 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3245, pruned_loss=0.0909, over 5714744.87 frames. ], batch size: 164, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:11:19,853 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=401529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:11:22,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=401532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:11:33,463 INFO [train.py:968] (0/2) Epoch 9, batch 37350, giga_loss[loss=0.2357, simple_loss=0.3148, pruned_loss=0.07824, over 28767.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3296, pruned_loss=0.0942, over 5702106.53 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3766, pruned_loss=0.1297, over 5657326.54 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3214, pruned_loss=0.08883, over 5713823.63 frames. ], batch size: 119, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:11:44,437 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:11:59,403 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=401582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:12:02,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=401585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:12:03,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.893e+02 9.245e+02 1.241e+03 1.855e+03 1.759e+04, threshold=2.483e+03, percent-clipped=13.0
+2023-03-04 23:12:04,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7011, 5.5095, 5.1590, 2.9020], device='cuda:0'), covar=tensor([0.0387, 0.0496, 0.0622, 0.1642], device='cuda:0'), in_proj_covar=tensor([0.0982, 0.0911, 0.0816, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 23:12:11,454 INFO [train.py:968] (0/2) Epoch 9, batch 37400, giga_loss[loss=0.2198, simple_loss=0.2928, pruned_loss=0.07342, over 28422.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3293, pruned_loss=0.09397, over 5706743.62 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3771, pruned_loss=0.1299, over 5663360.24 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3212, pruned_loss=0.0887, over 5711723.39 frames. ], batch size: 65, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:12:11,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3211, 1.9323, 1.5577, 1.4719], device='cuda:0'), covar=tensor([0.0789, 0.0277, 0.0305, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0082], device='cuda:0')
+2023-03-04 23:12:22,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=401613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:12:23,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:12:52,714 INFO [train.py:968] (0/2) Epoch 9, batch 37450, giga_loss[loss=0.2366, simple_loss=0.3076, pruned_loss=0.08278, over 28308.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3292, pruned_loss=0.09408, over 5702068.78 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3776, pruned_loss=0.1301, over 5659951.67 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3212, pruned_loss=0.08886, over 5709535.63 frames. ], batch size: 65, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:12:53,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=401649.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:13:05,003 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401663.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:13:24,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.991e+02 1.028e+03 1.403e+03 2.038e+03 5.371e+03, threshold=2.806e+03, percent-clipped=12.0
+2023-03-04 23:13:29,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7160, 1.8405, 1.8276, 1.6443], device='cuda:0'), covar=tensor([0.1036, 0.1243, 0.1383, 0.1263], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0733, 0.0661, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 23:13:33,114 INFO [train.py:968] (0/2) Epoch 9, batch 37500, giga_loss[loss=0.2381, simple_loss=0.32, pruned_loss=0.07807, over 28967.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.333, pruned_loss=0.09642, over 5704631.93 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3781, pruned_loss=0.1301, over 5653888.72 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3243, pruned_loss=0.09086, over 5718833.83 frames. ], batch size: 164, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:14:03,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4939, 1.6621, 1.5733, 1.4557], device='cuda:0'), covar=tensor([0.1249, 0.1431, 0.1641, 0.1469], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0730, 0.0659, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 23:14:05,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=401733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:14:17,470 INFO [train.py:968] (0/2) Epoch 9, batch 37550, giga_loss[loss=0.2848, simple_loss=0.3519, pruned_loss=0.1089, over 28852.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3388, pruned_loss=0.1006, over 5699473.21 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3779, pruned_loss=0.13, over 5653587.90 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3313, pruned_loss=0.09575, over 5711471.41 frames. ], batch size: 106, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:14:24,795 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=401756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:14:26,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=401759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:14:52,579 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.079e+02 1.193e+03 1.678e+03 2.780e+03 7.171e+03, threshold=3.356e+03, percent-clipped=23.0
+2023-03-04 23:14:55,147 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:14:58,166 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=401792.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:15:00,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=401795.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:15:01,996 INFO [train.py:968] (0/2) Epoch 9, batch 37600, giga_loss[loss=0.3029, simple_loss=0.3771, pruned_loss=0.1144, over 28891.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3483, pruned_loss=0.1072, over 5696505.71 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3779, pruned_loss=0.1298, over 5661477.14 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.341, pruned_loss=0.1025, over 5700611.50 frames. ], batch size: 174, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:15:23,534 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401824.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:15:40,680 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5016, 1.5014, 1.2579, 1.5267], device='cuda:0'), covar=tensor([0.0741, 0.0305, 0.0323, 0.0783], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0115, 0.0209], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0082], device='cuda:0')
+2023-03-04 23:15:45,588 INFO [train.py:968] (0/2) Epoch 9, batch 37650, giga_loss[loss=0.2911, simple_loss=0.365, pruned_loss=0.1086, over 28970.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3552, pruned_loss=0.1113, over 5692393.91 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3784, pruned_loss=0.1302, over 5666696.15 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.348, pruned_loss=0.1067, over 5692074.85 frames. ], batch size: 136, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:16:10,078 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 23:16:13,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=401876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:16:15,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=401879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:16:20,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.756e+02 1.289e+03 1.636e+03 2.277e+03 6.477e+03, threshold=3.271e+03, percent-clipped=10.0
+2023-03-04 23:16:29,598 INFO [train.py:968] (0/2) Epoch 9, batch 37700, giga_loss[loss=0.2899, simple_loss=0.3713, pruned_loss=0.1042, over 28938.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3599, pruned_loss=0.1136, over 5670891.16 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3784, pruned_loss=0.1303, over 5652708.10 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3533, pruned_loss=0.109, over 5684081.41 frames. ], batch size: 145, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:16:36,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:17:14,829 INFO [train.py:968] (0/2) Epoch 9, batch 37750, giga_loss[loss=0.326, simple_loss=0.3925, pruned_loss=0.1298, over 28616.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3642, pruned_loss=0.115, over 5676073.86 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3783, pruned_loss=0.1301, over 5657585.18 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3588, pruned_loss=0.1112, over 5682677.02 frames. ], batch size: 336, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:17:45,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.228e+02 1.135e+03 1.486e+03 2.191e+03 6.842e+03, threshold=2.971e+03, percent-clipped=7.0
+2023-03-04 23:17:53,677 INFO [train.py:968] (0/2) Epoch 9, batch 37800, libri_loss[loss=0.3165, simple_loss=0.3678, pruned_loss=0.1326, over 29563.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3684, pruned_loss=0.1179, over 5676878.77 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3784, pruned_loss=0.1304, over 5659014.03 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3631, pruned_loss=0.1137, over 5682091.29 frames. ], batch size: 79, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:17:55,088 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-402000.pt
+2023-03-04 23:18:20,913 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4365, 1.7129, 1.7213, 1.3603], device='cuda:0'), covar=tensor([0.1689, 0.1975, 0.1275, 0.1588], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0695, 0.0840, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 23:18:27,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=402038.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:18:36,074 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-04 23:18:37,785 INFO [train.py:968] (0/2) Epoch 9, batch 37850, giga_loss[loss=0.2364, simple_loss=0.3168, pruned_loss=0.07803, over 28787.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3644, pruned_loss=0.1152, over 5677147.25 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.378, pruned_loss=0.1302, over 5660555.18 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3605, pruned_loss=0.112, over 5680172.22 frames. ], batch size: 242, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:18:55,727 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 23:19:05,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=402081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:19:09,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.357e+02 1.015e+03 1.295e+03 1.672e+03 3.625e+03, threshold=2.590e+03, percent-clipped=5.0
+2023-03-04 23:19:17,588 INFO [train.py:968] (0/2) Epoch 9, batch 37900, giga_loss[loss=0.2675, simple_loss=0.3417, pruned_loss=0.09666, over 28435.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3608, pruned_loss=0.1116, over 5694032.51 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3785, pruned_loss=0.1304, over 5665816.46 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3568, pruned_loss=0.1084, over 5692224.04 frames. ], batch size: 65, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:19:59,801 INFO [train.py:968] (0/2) Epoch 9, batch 37950, giga_loss[loss=0.297, simple_loss=0.3715, pruned_loss=0.1113, over 28219.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3602, pruned_loss=0.1105, over 5696442.19 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3788, pruned_loss=0.1306, over 5669340.99 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3562, pruned_loss=0.1072, over 5692587.91 frames. ], batch size: 368, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:20:29,055 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=402181.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:20:31,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=402184.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:20:33,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.923e+02 1.136e+03 1.551e+03 1.963e+03 8.054e+03, threshold=3.101e+03, percent-clipped=8.0
+2023-03-04 23:20:42,065 INFO [train.py:968] (0/2) Epoch 9, batch 38000, giga_loss[loss=0.2864, simple_loss=0.3632, pruned_loss=0.1048, over 28506.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3601, pruned_loss=0.11, over 5699129.52 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3787, pruned_loss=0.1305, over 5667952.39 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3568, pruned_loss=0.1071, over 5697611.13 frames. ], batch size: 336, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:20:56,029 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=402213.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:21:25,950 INFO [train.py:968] (0/2) Epoch 9, batch 38050, giga_loss[loss=0.3469, simple_loss=0.4012, pruned_loss=0.1463, over 27685.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3641, pruned_loss=0.1126, over 5693306.08 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3798, pruned_loss=0.1315, over 5664512.09 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.36, pruned_loss=0.1091, over 5696058.06 frames. ], batch size: 472, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:21:27,141 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-04 23:21:49,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 23:21:57,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.092e+02 1.444e+03 1.884e+03 2.436e+03 9.185e+03, threshold=3.769e+03, percent-clipped=10.0
+2023-03-04 23:22:04,452 INFO [train.py:968] (0/2) Epoch 9, batch 38100, giga_loss[loss=0.2948, simple_loss=0.3702, pruned_loss=0.1097, over 28712.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3665, pruned_loss=0.1146, over 5692089.19 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3804, pruned_loss=0.132, over 5663247.90 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3621, pruned_loss=0.1105, over 5697277.11 frames. ], batch size: 284, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:22:34,902 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 23:22:48,569 INFO [train.py:968] (0/2) Epoch 9, batch 38150, giga_loss[loss=0.2847, simple_loss=0.3529, pruned_loss=0.1082, over 28753.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3682, pruned_loss=0.1162, over 5683292.26 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3809, pruned_loss=0.1322, over 5660360.90 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3639, pruned_loss=0.1123, over 5690319.08 frames. ], batch size: 284, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:23:00,274 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=402365.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:23:18,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.236e+02 1.316e+03 1.779e+03 2.489e+03 1.284e+04, threshold=3.558e+03, percent-clipped=12.0
+2023-03-04 23:23:27,960 INFO [train.py:968] (0/2) Epoch 9, batch 38200, giga_loss[loss=0.2633, simple_loss=0.3387, pruned_loss=0.09397, over 28880.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3684, pruned_loss=0.1167, over 5694601.87 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3807, pruned_loss=0.1319, over 5665970.42 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3646, pruned_loss=0.1132, over 5696243.59 frames. ], batch size: 112, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:23:41,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2386, 1.4469, 1.1918, 1.0060], device='cuda:0'), covar=tensor([0.2086, 0.2102, 0.2301, 0.1946], device='cuda:0'), in_proj_covar=tensor([0.1251, 0.0931, 0.1110, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 23:23:56,123 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-04 23:24:09,220 INFO [train.py:968] (0/2) Epoch 9, batch 38250, giga_loss[loss=0.2676, simple_loss=0.3427, pruned_loss=0.09624, over 28551.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.368, pruned_loss=0.1168, over 5690682.73 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3804, pruned_loss=0.1316, over 5671043.59 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3649, pruned_loss=0.1139, over 5687970.39 frames. ], batch size: 78, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:24:15,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=402456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:24:16,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1198, 3.9436, 3.7208, 2.0226], device='cuda:0'), covar=tensor([0.0578, 0.0703, 0.0678, 0.1973], device='cuda:0'), in_proj_covar=tensor([0.0984, 0.0919, 0.0816, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 23:24:27,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4569, 1.6401, 1.6330, 1.5359], device='cuda:0'), covar=tensor([0.0949, 0.0956, 0.1142, 0.0965], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0731, 0.0657, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-04 23:24:43,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.861e+02 1.127e+03 1.488e+03 1.920e+03 5.811e+03, threshold=2.977e+03, percent-clipped=5.0
+2023-03-04 23:24:50,785 INFO [train.py:968] (0/2) Epoch 9, batch 38300, giga_loss[loss=0.2687, simple_loss=0.3449, pruned_loss=0.09629, over 28334.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3681, pruned_loss=0.1163, over 5689031.70 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3803, pruned_loss=0.1315, over 5664044.89 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3654, pruned_loss=0.1138, over 5693593.02 frames. ], batch size: 77, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:25:16,385 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-04 23:25:20,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2403, 1.4926, 1.2201, 1.0892], device='cuda:0'), covar=tensor([0.2390, 0.2202, 0.2512, 0.1876], device='cuda:0'), in_proj_covar=tensor([0.1253, 0.0930, 0.1109, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 23:25:30,295 INFO [train.py:968] (0/2) Epoch 9, batch 38350, giga_loss[loss=0.2624, simple_loss=0.3436, pruned_loss=0.09061, over 28348.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3672, pruned_loss=0.1145, over 5694831.06 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3798, pruned_loss=0.1311, over 5667387.77 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3653, pruned_loss=0.1125, over 5696099.39 frames. ], batch size: 77, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:25:41,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8759, 0.9778, 0.8996, 0.7756], device='cuda:0'), covar=tensor([0.1129, 0.1344, 0.0905, 0.1242], device='cuda:0'), in_proj_covar=tensor([0.1638, 0.1509, 0.1477, 0.1599], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 23:26:01,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.659e+02 1.191e+03 1.698e+03 2.390e+03 1.120e+04, threshold=3.396e+03, percent-clipped=16.0
+2023-03-04 23:26:09,645 INFO [train.py:968] (0/2) Epoch 9, batch 38400, giga_loss[loss=0.2695, simple_loss=0.3448, pruned_loss=0.09715, over 28758.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3677, pruned_loss=0.114, over 5698492.79 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.38, pruned_loss=0.1311, over 5663333.81 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3658, pruned_loss=0.112, over 5704263.18 frames. ], batch size: 92, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:26:10,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=402599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:26:13,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=402602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:26:36,569 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=402631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:26:49,939 INFO [train.py:968] (0/2) Epoch 9, batch 38450, giga_loss[loss=0.2705, simple_loss=0.3363, pruned_loss=0.1023, over 28775.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3652, pruned_loss=0.1125, over 5690740.10 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3799, pruned_loss=0.1309, over 5658205.34 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3634, pruned_loss=0.1106, over 5700921.95 frames. ], batch size: 99, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:26:51,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7615, 2.5139, 1.6854, 0.8757], device='cuda:0'), covar=tensor([0.4685, 0.2255, 0.2720, 0.4588], device='cuda:0'), in_proj_covar=tensor([0.1515, 0.1433, 0.1469, 0.1231], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-04 23:27:22,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.440e+02 9.804e+02 1.272e+03 1.724e+03 6.429e+03, threshold=2.545e+03, percent-clipped=4.0
+2023-03-04 23:27:30,439 INFO [train.py:968] (0/2) Epoch 9, batch 38500, giga_loss[loss=0.2469, simple_loss=0.3312, pruned_loss=0.08128, over 28970.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.362, pruned_loss=0.1106, over 5696935.36 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3793, pruned_loss=0.1305, over 5664238.71 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3608, pruned_loss=0.1091, over 5700344.55 frames. ], batch size: 136, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:27:35,429 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6629, 1.7337, 1.4304, 1.3306], device='cuda:0'), covar=tensor([0.1689, 0.1449, 0.1232, 0.1508], device='cuda:0'), in_proj_covar=tensor([0.1627, 0.1500, 0.1468, 0.1589], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-04 23:28:05,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=402740.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:28:11,381 INFO [train.py:968] (0/2) Epoch 9, batch 38550, giga_loss[loss=0.2536, simple_loss=0.3323, pruned_loss=0.08742, over 28589.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3596, pruned_loss=0.1091, over 5700358.10 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3791, pruned_loss=0.1304, over 5666741.38 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3585, pruned_loss=0.1079, over 5701030.60 frames. ], batch size: 85, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:28:44,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.694e+02 9.551e+02 1.169e+03 1.556e+03 6.402e+03, threshold=2.338e+03, percent-clipped=3.0
+2023-03-04 23:28:51,136 INFO [train.py:968] (0/2) Epoch 9, batch 38600, giga_loss[loss=0.2614, simple_loss=0.3382, pruned_loss=0.09228, over 28580.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.36, pruned_loss=0.1098, over 5694861.28 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3796, pruned_loss=0.1306, over 5658742.25 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3586, pruned_loss=0.1083, over 5702607.64 frames. ], batch size: 78, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:29:03,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-04 23:29:10,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2440, 3.0412, 2.8514, 1.4466], device='cuda:0'), covar=tensor([0.0858, 0.0996, 0.0929, 0.2381], device='cuda:0'), in_proj_covar=tensor([0.0988, 0.0918, 0.0815, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 23:29:27,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6026, 1.8409, 1.8618, 1.4762], device='cuda:0'), covar=tensor([0.1648, 0.2183, 0.1306, 0.1509], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0707, 0.0848, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-04 23:29:33,208 INFO [train.py:968] (0/2) Epoch 9, batch 38650, giga_loss[loss=0.2973, simple_loss=0.3612, pruned_loss=0.1167, over 28569.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3613, pruned_loss=0.1109, over 5694286.44 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3802, pruned_loss=0.1312, over 5658992.20 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3594, pruned_loss=0.1089, over 5700872.62 frames. ], batch size: 85, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:29:59,899 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=402883.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:30:01,843 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=402886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:30:03,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.102e+02 1.055e+03 1.319e+03 1.835e+03 1.884e+04, threshold=2.638e+03, percent-clipped=12.0
+2023-03-04 23:30:07,738 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-04 23:30:10,640 INFO [train.py:968] (0/2) Epoch 9, batch 38700, libri_loss[loss=0.2946, simple_loss=0.3528, pruned_loss=0.1181, over 29396.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.362, pruned_loss=0.111, over 5698262.57 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3802, pruned_loss=0.1314, over 5658604.25 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3598, pruned_loss=0.1086, over 5705553.82 frames. ], batch size: 67, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:30:24,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=402915.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:30:45,926 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=402945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:30:48,538 INFO [train.py:968] (0/2) Epoch 9, batch 38750, giga_loss[loss=0.2907, simple_loss=0.366, pruned_loss=0.1077, over 28803.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3605, pruned_loss=0.1091, over 5707256.65 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3804, pruned_loss=0.1313, over 5662228.89 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3583, pruned_loss=0.1068, over 5710480.59 frames. ], batch size: 99, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:31:18,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.190e+02 9.641e+02 1.145e+03 1.547e+03 1.132e+04, threshold=2.290e+03, percent-clipped=9.0
+2023-03-04 23:31:25,046 INFO [train.py:968] (0/2) Epoch 9, batch 38800, giga_loss[loss=0.296, simple_loss=0.3629, pruned_loss=0.1145, over 28746.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3607, pruned_loss=0.1092, over 5698637.81 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3806, pruned_loss=0.1314, over 5657110.13 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3582, pruned_loss=0.1068, over 5706316.85 frames. ], batch size: 119, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:32:05,770 INFO [train.py:968] (0/2) Epoch 9, batch 38850, giga_loss[loss=0.2583, simple_loss=0.3302, pruned_loss=0.09314, over 28570.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3595, pruned_loss=0.1092, over 5702207.69 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3802, pruned_loss=0.1312, over 5665992.62 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3571, pruned_loss=0.1066, over 5701820.06 frames. ], batch size: 85, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:32:37,802 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.434e+02 1.100e+03 1.409e+03 1.930e+03 8.140e+03, threshold=2.818e+03, percent-clipped=18.0
+2023-03-04 23:32:38,768 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:32:44,358 INFO [train.py:968] (0/2) Epoch 9, batch 38900, giga_loss[loss=0.2323, simple_loss=0.3082, pruned_loss=0.07824, over 28499.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3579, pruned_loss=0.1091, over 5701405.03 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3801, pruned_loss=0.1312, over 5667938.99 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3552, pruned_loss=0.1062, over 5700794.96 frames. ], batch size: 60, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:33:22,887 INFO [train.py:968] (0/2) Epoch 9, batch 38950, giga_loss[loss=0.2598, simple_loss=0.337, pruned_loss=0.09128, over 28626.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3539, pruned_loss=0.1066, over 5701518.77 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3798, pruned_loss=0.131, over 5666858.33 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3512, pruned_loss=0.1037, over 5702996.31 frames. ], batch size: 336, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:33:30,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2261, 1.1168, 4.0326, 3.2951], device='cuda:0'), covar=tensor([0.1558, 0.2653, 0.0328, 0.1126], device='cuda:0'), in_proj_covar=tensor([0.0625, 0.0565, 0.0816, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 23:33:39,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6909, 1.8288, 1.6645, 1.7189], device='cuda:0'), covar=tensor([0.1255, 0.1697, 0.1794, 0.1547], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0732, 0.0660, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-04 23:33:54,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.080e+02 1.065e+03 1.362e+03 1.894e+03 3.625e+03, threshold=2.725e+03, percent-clipped=8.0
+2023-03-04 23:33:56,286 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:34:02,013 INFO [train.py:968] (0/2) Epoch 9, batch 39000, giga_loss[loss=0.3019, simple_loss=0.3743, pruned_loss=0.1148, over 28315.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3529, pruned_loss=0.1064, over 5704604.28 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3794, pruned_loss=0.1306, over 5673342.35 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3505, pruned_loss=0.1037, over 5700974.20 frames. ], batch size: 368, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:34:02,017 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-04 23:34:10,205 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.219, simple_loss=0.325, pruned_loss=0.05654, over 944034.00 frames. 
+2023-03-04 23:34:10,206 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-04 23:34:37,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:34:49,169 INFO [train.py:968] (0/2) Epoch 9, batch 39050, giga_loss[loss=0.2579, simple_loss=0.3301, pruned_loss=0.09286, over 29052.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3515, pruned_loss=0.1059, over 5709155.92 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3792, pruned_loss=0.1304, over 5676253.86 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3491, pruned_loss=0.1034, over 5704434.39 frames. ], batch size: 128, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:35:21,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.954e+02 1.104e+03 1.390e+03 1.871e+03 4.254e+03, threshold=2.780e+03, percent-clipped=10.0
+2023-03-04 23:35:29,017 INFO [train.py:968] (0/2) Epoch 9, batch 39100, giga_loss[loss=0.2263, simple_loss=0.3014, pruned_loss=0.0756, over 28536.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3501, pruned_loss=0.1059, over 5689228.86 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3795, pruned_loss=0.1306, over 5653972.76 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3472, pruned_loss=0.1032, over 5706189.20 frames. ], batch size: 60, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:35:46,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:36:08,607 INFO [train.py:968] (0/2) Epoch 9, batch 39150, giga_loss[loss=0.2524, simple_loss=0.3256, pruned_loss=0.08963, over 28794.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3482, pruned_loss=0.1055, over 5696100.12 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3798, pruned_loss=0.1308, over 5657516.60 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3453, pruned_loss=0.1028, over 5706989.91 frames. ], batch size: 242, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:36:41,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.417e+02 1.044e+03 1.239e+03 1.960e+03 7.918e+03, threshold=2.478e+03, percent-clipped=11.0
+2023-03-04 23:36:49,398 INFO [train.py:968] (0/2) Epoch 9, batch 39200, giga_loss[loss=0.285, simple_loss=0.3515, pruned_loss=0.1093, over 27978.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3441, pruned_loss=0.103, over 5692485.85 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3798, pruned_loss=0.1309, over 5650627.78 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3413, pruned_loss=0.1004, over 5708197.14 frames. ], batch size: 412, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:37:27,715 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-04 23:37:31,516 INFO [train.py:968] (0/2) Epoch 9, batch 39250, giga_loss[loss=0.3164, simple_loss=0.3779, pruned_loss=0.1274, over 28053.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3433, pruned_loss=0.1025, over 5696417.21 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3799, pruned_loss=0.1309, over 5654648.58 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3406, pruned_loss=0.1, over 5705806.34 frames. ], batch size: 412, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:37:44,760 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=403463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:37:46,876 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:37:47,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=403466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:38:06,063 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4815, 3.4291, 1.5755, 1.4979], device='cuda:0'), covar=tensor([0.0840, 0.0233, 0.0882, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0497, 0.0328, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 23:38:07,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.988e+02 9.590e+02 1.163e+03 1.447e+03 3.404e+03, threshold=2.325e+03, percent-clipped=5.0
+2023-03-04 23:38:12,640 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:38:12,677 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=403495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:38:15,418 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:38:17,556 INFO [train.py:968] (0/2) Epoch 9, batch 39300, giga_loss[loss=0.2513, simple_loss=0.3358, pruned_loss=0.0834, over 28905.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3458, pruned_loss=0.1033, over 5695080.68 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3798, pruned_loss=0.1308, over 5648570.40 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3434, pruned_loss=0.1012, over 5707686.26 frames. ], batch size: 145, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:39:00,003 INFO [train.py:968] (0/2) Epoch 9, batch 39350, giga_loss[loss=0.2853, simple_loss=0.3479, pruned_loss=0.1113, over 28738.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3485, pruned_loss=0.1038, over 5700928.38 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.38, pruned_loss=0.1309, over 5653443.54 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3456, pruned_loss=0.1015, over 5707207.90 frames. ], batch size: 92, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:39:12,243 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-04 23:39:14,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3206, 1.5863, 1.2699, 1.3490], device='cuda:0'), covar=tensor([0.2473, 0.2407, 0.2794, 0.2129], device='cuda:0'), in_proj_covar=tensor([0.1254, 0.0929, 0.1111, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 23:39:15,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:39:35,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.018e+02 1.025e+03 1.371e+03 1.928e+03 6.325e+03, threshold=2.742e+03, percent-clipped=12.0
+2023-03-04 23:39:40,883 INFO [train.py:968] (0/2) Epoch 9, batch 39400, giga_loss[loss=0.2902, simple_loss=0.368, pruned_loss=0.1062, over 28623.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3513, pruned_loss=0.1047, over 5692936.61 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3804, pruned_loss=0.1311, over 5650690.10 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3477, pruned_loss=0.1018, over 5701714.43 frames. ], batch size: 307, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:39:49,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:39:50,739 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=403608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:39:55,264 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=403611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:40:02,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4161, 1.5598, 1.2529, 1.7675], device='cuda:0'), covar=tensor([0.2433, 0.2409, 0.2716, 0.2155], device='cuda:0'), in_proj_covar=tensor([0.1256, 0.0930, 0.1111, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 23:40:18,418 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=403640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:40:25,017 INFO [train.py:968] (0/2) Epoch 9, batch 39450, giga_loss[loss=0.2543, simple_loss=0.318, pruned_loss=0.09532, over 23765.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.351, pruned_loss=0.104, over 5685533.53 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3796, pruned_loss=0.1307, over 5654347.29 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3481, pruned_loss=0.1014, over 5690072.57 frames. ], batch size: 705, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:40:28,952 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403653.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:40:31,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7585, 1.1129, 2.8563, 2.8146], device='cuda:0'), covar=tensor([0.1599, 0.2356, 0.0542, 0.0748], device='cuda:0'), in_proj_covar=tensor([0.0626, 0.0571, 0.0820, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 23:41:00,978 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.457e+02 1.035e+03 1.524e+03 2.113e+03 6.645e+03, threshold=3.049e+03, percent-clipped=16.0
+2023-03-04 23:41:06,139 INFO [train.py:968] (0/2) Epoch 9, batch 39500, giga_loss[loss=0.289, simple_loss=0.3562, pruned_loss=0.1109, over 28870.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3507, pruned_loss=0.1032, over 5682252.25 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.38, pruned_loss=0.131, over 5647403.03 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3479, pruned_loss=0.1006, over 5692451.86 frames. ], batch size: 136, lr: 3.52e-03, grad_scale: 2.0
+2023-03-04 23:41:15,576 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=403709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:41:17,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=403712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:41:40,647 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=403741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:41:46,337 INFO [train.py:968] (0/2) Epoch 9, batch 39550, giga_loss[loss=0.2727, simple_loss=0.3521, pruned_loss=0.0967, over 28624.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3517, pruned_loss=0.1041, over 5683185.00 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3803, pruned_loss=0.131, over 5648639.58 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3485, pruned_loss=0.1013, over 5690994.30 frames. ], batch size: 336, lr: 3.52e-03, grad_scale: 2.0
+2023-03-04 23:41:47,796 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=403750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:41:49,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=403753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:42:13,099 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=403782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:42:20,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.264e+02 1.169e+03 1.459e+03 2.131e+03 4.424e+03, threshold=2.918e+03, percent-clipped=11.0
+2023-03-04 23:42:28,968 INFO [train.py:968] (0/2) Epoch 9, batch 39600, libri_loss[loss=0.2951, simple_loss=0.3622, pruned_loss=0.114, over 29537.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3523, pruned_loss=0.1049, over 5682622.49 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3804, pruned_loss=0.1312, over 5649185.94 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3494, pruned_loss=0.1024, over 5688401.54 frames. ], batch size: 78, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:42:51,156 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.90 vs. limit=5.0
+2023-03-04 23:43:08,899 INFO [train.py:968] (0/2) Epoch 9, batch 39650, giga_loss[loss=0.2698, simple_loss=0.3441, pruned_loss=0.0977, over 28710.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3551, pruned_loss=0.1067, over 5692268.63 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3801, pruned_loss=0.131, over 5653059.41 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3527, pruned_loss=0.1045, over 5693973.76 frames. ], batch size: 60, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:43:15,928 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-04 23:43:25,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:43:25,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:43:41,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.485e+02 1.136e+03 1.483e+03 2.098e+03 5.587e+03, threshold=2.965e+03, percent-clipped=11.0
+2023-03-04 23:43:47,117 INFO [train.py:968] (0/2) Epoch 9, batch 39700, giga_loss[loss=0.2822, simple_loss=0.3557, pruned_loss=0.1043, over 28852.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3588, pruned_loss=0.1092, over 5700488.74 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3808, pruned_loss=0.1313, over 5659629.16 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3555, pruned_loss=0.1064, over 5697872.50 frames. ], batch size: 112, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:44:25,656 INFO [train.py:968] (0/2) Epoch 9, batch 39750, giga_loss[loss=0.2535, simple_loss=0.3351, pruned_loss=0.0859, over 28975.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.36, pruned_loss=0.1089, over 5714387.23 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3804, pruned_loss=0.1309, over 5664746.34 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3572, pruned_loss=0.1065, over 5708588.56 frames. ], batch size: 136, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:44:43,991 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-04 23:44:59,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.163e+02 1.195e+03 1.619e+03 2.342e+03 5.247e+03, threshold=3.237e+03, percent-clipped=14.0
+2023-03-04 23:45:04,312 INFO [train.py:968] (0/2) Epoch 9, batch 39800, giga_loss[loss=0.2688, simple_loss=0.3512, pruned_loss=0.09321, over 28651.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3615, pruned_loss=0.1101, over 5707197.99 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.381, pruned_loss=0.1313, over 5661281.12 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3579, pruned_loss=0.1069, over 5707837.85 frames. ], batch size: 71, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:45:05,518 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-404000.pt
+2023-03-04 23:45:16,257 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=404013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:16,760 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=404014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:16,880 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-04 23:45:19,222 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=404016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:19,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=404017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:27,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=404028.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:45:30,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0921, 4.9804, 2.2451, 2.1669], device='cuda:0'), covar=tensor([0.0780, 0.0262, 0.0750, 0.1091], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0501, 0.0329, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 23:45:42,418 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=404045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:43,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=404046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:44,114 INFO [train.py:968] (0/2) Epoch 9, batch 39850, giga_loss[loss=0.2825, simple_loss=0.3593, pruned_loss=0.1028, over 28267.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3617, pruned_loss=0.1102, over 5691307.41 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3819, pruned_loss=0.1319, over 5644146.63 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3574, pruned_loss=0.1064, over 5707971.54 frames. ], batch size: 368, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:46:15,998 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.442e+02 1.194e+03 1.586e+03 2.345e+03 8.786e+03, threshold=3.172e+03, percent-clipped=12.0
+2023-03-04 23:46:21,442 INFO [train.py:968] (0/2) Epoch 9, batch 39900, giga_loss[loss=0.2703, simple_loss=0.3476, pruned_loss=0.09649, over 29080.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3607, pruned_loss=0.1096, over 5701387.67 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3815, pruned_loss=0.1317, over 5650863.24 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.357, pruned_loss=0.1061, over 5710478.77 frames. ], batch size: 155, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:46:56,886 INFO [train.py:968] (0/2) Epoch 9, batch 39950, giga_loss[loss=0.3476, simple_loss=0.4046, pruned_loss=0.1453, over 28738.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3592, pruned_loss=0.1089, over 5710917.93 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3816, pruned_loss=0.1317, over 5658140.59 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3554, pruned_loss=0.1054, over 5712874.84 frames. ], batch size: 284, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:47:12,600 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.51 vs. limit=5.0
+2023-03-04 23:47:16,856 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=404171.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:47:18,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=404174.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:47:32,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.376e+02 1.120e+03 1.415e+03 1.858e+03 8.427e+03, threshold=2.831e+03, percent-clipped=5.0
+2023-03-04 23:47:35,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3898, 1.8467, 1.4444, 1.6352], device='cuda:0'), covar=tensor([0.0691, 0.0263, 0.0301, 0.0760], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0116, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0075, 0.0054, 0.0049, 0.0083], device='cuda:0')
+2023-03-04 23:47:38,554 INFO [train.py:968] (0/2) Epoch 9, batch 40000, giga_loss[loss=0.2301, simple_loss=0.3148, pruned_loss=0.07269, over 28984.00 frames. ], tot_loss[loss=0.284, simple_loss=0.355, pruned_loss=0.1065, over 5712383.61 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3814, pruned_loss=0.1314, over 5664762.43 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3515, pruned_loss=0.1032, over 5709511.10 frames. ], batch size: 164, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:47:42,196 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=404203.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:47:47,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3199, 1.5500, 1.2290, 1.4614], device='cuda:0'), covar=tensor([0.2238, 0.2185, 0.2481, 0.2127], device='cuda:0'), in_proj_covar=tensor([0.1249, 0.0923, 0.1102, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-04 23:47:59,435 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2436, 1.3809, 1.4759, 1.3009], device='cuda:0'), covar=tensor([0.1394, 0.1593, 0.1956, 0.1659], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0733, 0.0665, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-04 23:48:18,645 INFO [train.py:968] (0/2) Epoch 9, batch 40050, giga_loss[loss=0.2988, simple_loss=0.3825, pruned_loss=0.1075, over 28882.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3512, pruned_loss=0.1042, over 5717020.99 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3806, pruned_loss=0.131, over 5669633.99 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3486, pruned_loss=0.1015, over 5711494.90 frames. ], batch size: 174, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:48:27,716 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3828, 3.3200, 1.6175, 1.4345], device='cuda:0'), covar=tensor([0.0807, 0.0290, 0.0793, 0.1210], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0501, 0.0329, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0021, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-04 23:48:31,650 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0370, 3.8157, 3.7190, 1.5415], device='cuda:0'), covar=tensor([0.0625, 0.0846, 0.0857, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.0984, 0.0918, 0.0817, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-04 23:48:52,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.018e+02 9.688e+02 1.280e+03 1.634e+03 4.508e+03, threshold=2.561e+03, percent-clipped=2.0
+2023-03-04 23:48:57,919 INFO [train.py:968] (0/2) Epoch 9, batch 40100, giga_loss[loss=0.2845, simple_loss=0.3612, pruned_loss=0.1039, over 28998.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3513, pruned_loss=0.1032, over 5714216.51 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3806, pruned_loss=0.131, over 5668212.79 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.349, pruned_loss=0.1009, over 5711430.74 frames. ], batch size: 136, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:49:01,300 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-04 23:49:25,788 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 23:49:39,160 INFO [train.py:968] (0/2) Epoch 9, batch 40150, giga_loss[loss=0.2756, simple_loss=0.3576, pruned_loss=0.09681, over 28579.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3538, pruned_loss=0.1039, over 5708843.83 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.381, pruned_loss=0.1311, over 5672967.36 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3508, pruned_loss=0.1012, over 5703531.73 frames. ], batch size: 336, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:49:59,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2237, 1.5617, 1.5639, 1.1294], device='cuda:0'), covar=tensor([0.1453, 0.2060, 0.1244, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0693, 0.0838, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 23:50:02,436 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 23:50:03,425 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=404379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:50:12,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.077e+02 1.177e+03 1.731e+03 2.337e+03 7.714e+03, threshold=3.463e+03, percent-clipped=17.0
+2023-03-04 23:50:16,547 INFO [train.py:968] (0/2) Epoch 9, batch 40200, giga_loss[loss=0.2802, simple_loss=0.3363, pruned_loss=0.112, over 28525.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3559, pruned_loss=0.1057, over 5713574.23 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3811, pruned_loss=0.1312, over 5679547.07 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3526, pruned_loss=0.1026, over 5704579.06 frames. ], batch size: 85, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:50:47,870 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=404438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:50:56,430 INFO [train.py:968] (0/2) Epoch 9, batch 40250, giga_loss[loss=0.283, simple_loss=0.3588, pruned_loss=0.1036, over 28989.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3539, pruned_loss=0.1056, over 5718752.21 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3811, pruned_loss=0.1311, over 5684812.97 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3506, pruned_loss=0.1025, over 5707757.09 frames. ], batch size: 174, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:51:09,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5158, 1.4806, 1.2725, 1.1226], device='cuda:0'), covar=tensor([0.0718, 0.0595, 0.1011, 0.0982], device='cuda:0'), in_proj_covar=tensor([0.0345, 0.0437, 0.0493, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-04 23:51:31,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.958e+02 1.044e+03 1.236e+03 1.749e+03 4.702e+03, threshold=2.471e+03, percent-clipped=3.0
+2023-03-04 23:51:34,771 INFO [train.py:968] (0/2) Epoch 9, batch 40300, giga_loss[loss=0.2625, simple_loss=0.3285, pruned_loss=0.09819, over 28953.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3515, pruned_loss=0.1054, over 5726653.96 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3806, pruned_loss=0.1308, over 5687474.53 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3487, pruned_loss=0.1027, over 5716298.10 frames. ], batch size: 213, lr: 3.52e-03, grad_scale: 2.0
+2023-03-04 23:52:17,120 INFO [train.py:968] (0/2) Epoch 9, batch 40350, giga_loss[loss=0.2524, simple_loss=0.3276, pruned_loss=0.0886, over 28924.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3496, pruned_loss=0.1057, over 5719260.40 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3804, pruned_loss=0.1306, over 5689561.05 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3473, pruned_loss=0.1035, over 5709455.17 frames. ], batch size: 227, lr: 3.52e-03, grad_scale: 2.0
+2023-03-04 23:52:51,044 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.860e+02 1.097e+03 1.531e+03 1.962e+03 4.759e+03, threshold=3.062e+03, percent-clipped=11.0
+2023-03-04 23:52:55,074 INFO [train.py:968] (0/2) Epoch 9, batch 40400, giga_loss[loss=0.2907, simple_loss=0.3411, pruned_loss=0.1201, over 28638.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3488, pruned_loss=0.1059, over 5716316.80 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3805, pruned_loss=0.1307, over 5690622.25 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3461, pruned_loss=0.1034, over 5707751.51 frames. ], batch size: 85, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:53:08,256 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=404613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:53:33,928 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 23:53:36,011 INFO [train.py:968] (0/2) Epoch 9, batch 40450, giga_loss[loss=0.2335, simple_loss=0.31, pruned_loss=0.07845, over 29003.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3475, pruned_loss=0.1053, over 5708744.80 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3807, pruned_loss=0.1307, over 5689794.90 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3444, pruned_loss=0.1028, over 5703104.49 frames. ], batch size: 164, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:54:13,233 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.810e+02 1.121e+03 1.591e+03 2.253e+03 1.428e+04, threshold=3.182e+03, percent-clipped=12.0
+2023-03-04 23:54:17,778 INFO [train.py:968] (0/2) Epoch 9, batch 40500, giga_loss[loss=0.2658, simple_loss=0.3196, pruned_loss=0.106, over 23877.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3442, pruned_loss=0.1039, over 5711332.01 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3809, pruned_loss=0.131, over 5695209.11 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3408, pruned_loss=0.1011, over 5702593.00 frames. ], batch size: 705, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:54:53,342 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-04 23:54:55,632 INFO [train.py:968] (0/2) Epoch 9, batch 40550, giga_loss[loss=0.2509, simple_loss=0.3211, pruned_loss=0.09035, over 28991.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3394, pruned_loss=0.1015, over 5698079.95 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.381, pruned_loss=0.1312, over 5678149.80 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3357, pruned_loss=0.09847, over 5707612.87 frames. ], batch size: 106, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:55:00,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=404754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:55:27,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0605, 2.2855, 2.3172, 1.8274], device='cuda:0'), covar=tensor([0.1568, 0.1901, 0.1216, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0695, 0.0836, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 23:55:31,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.589e+02 1.108e+03 1.582e+03 1.993e+03 5.992e+03, threshold=3.164e+03, percent-clipped=6.0
+2023-03-04 23:55:36,218 INFO [train.py:968] (0/2) Epoch 9, batch 40600, giga_loss[loss=0.3106, simple_loss=0.3722, pruned_loss=0.1246, over 28976.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3382, pruned_loss=0.1001, over 5707552.93 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3809, pruned_loss=0.1312, over 5680048.94 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3351, pruned_loss=0.09749, over 5713597.83 frames. ], batch size: 128, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:55:48,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=404813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:56:16,093 INFO [train.py:968] (0/2) Epoch 9, batch 40650, giga_loss[loss=0.2519, simple_loss=0.3344, pruned_loss=0.08473, over 28663.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.343, pruned_loss=0.1024, over 5707843.68 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3808, pruned_loss=0.131, over 5687445.77 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3392, pruned_loss=0.09948, over 5706977.80 frames. ], batch size: 242, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:56:39,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0845, 2.2432, 2.2945, 1.7981], device='cuda:0'), covar=tensor([0.1499, 0.1791, 0.1157, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0692, 0.0834, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-04 23:56:52,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.420e+02 1.160e+03 1.548e+03 2.346e+03 5.365e+03, threshold=3.097e+03, percent-clipped=10.0
+2023-03-04 23:56:56,365 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=404897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:56:56,725 INFO [train.py:968] (0/2) Epoch 9, batch 40700, giga_loss[loss=0.306, simple_loss=0.3708, pruned_loss=0.1206, over 28886.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3474, pruned_loss=0.105, over 5711547.79 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.381, pruned_loss=0.1313, over 5688381.23 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3433, pruned_loss=0.1016, over 5710795.73 frames. ], batch size: 227, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:56:58,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=404900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:57:22,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=404929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:57:33,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=404944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:57:37,518 INFO [train.py:968] (0/2) Epoch 9, batch 40750, giga_loss[loss=0.3087, simple_loss=0.3793, pruned_loss=0.1191, over 29076.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3508, pruned_loss=0.1062, over 5698829.30 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3807, pruned_loss=0.1312, over 5682030.87 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3472, pruned_loss=0.1032, over 5704610.00 frames. ], batch size: 128, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:57:45,017 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=404956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:57:46,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=404959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:57:53,434 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=404967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:58:09,229 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=404988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:58:09,277 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=404988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:58:12,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.394e+02 1.118e+03 1.400e+03 1.886e+03 4.159e+03, threshold=2.800e+03, percent-clipped=4.0
+2023-03-04 23:58:18,178 INFO [train.py:968] (0/2) Epoch 9, batch 40800, giga_loss[loss=0.286, simple_loss=0.3596, pruned_loss=0.1062, over 28742.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3544, pruned_loss=0.1078, over 5702381.61 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3807, pruned_loss=0.131, over 5686588.89 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3508, pruned_loss=0.1049, over 5703435.25 frames. ], batch size: 242, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:58:51,414 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-04 23:58:58,177 INFO [train.py:968] (0/2) Epoch 9, batch 40850, giga_loss[loss=0.3007, simple_loss=0.3639, pruned_loss=0.1188, over 28508.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3574, pruned_loss=0.1096, over 5709002.27 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3802, pruned_loss=0.1306, over 5692969.80 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3542, pruned_loss=0.1069, over 5704843.22 frames. ], batch size: 78, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:59:43,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.442e+02 1.287e+03 1.640e+03 2.269e+03 4.977e+03, threshold=3.280e+03, percent-clipped=11.0
+2023-03-04 23:59:46,797 INFO [train.py:968] (0/2) Epoch 9, batch 40900, giga_loss[loss=0.3617, simple_loss=0.4128, pruned_loss=0.1553, over 28897.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.363, pruned_loss=0.115, over 5700011.32 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3797, pruned_loss=0.1304, over 5687497.38 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3603, pruned_loss=0.1126, over 5701494.90 frames. ], batch size: 227, lr: 3.52e-03, grad_scale: 8.0
+2023-03-05 00:00:18,463 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=405131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:00:21,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=405134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:00:32,091 INFO [train.py:968] (0/2) Epoch 9, batch 40950, giga_loss[loss=0.3255, simple_loss=0.3918, pruned_loss=0.1296, over 28853.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3697, pruned_loss=0.1201, over 5691592.53 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3796, pruned_loss=0.1304, over 5683623.05 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3673, pruned_loss=0.1178, over 5696940.44 frames. ], batch size: 199, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:00:45,138 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=405163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:01:07,104 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-05 00:01:09,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.657e+03 2.285e+03 3.218e+03 9.756e+03, threshold=4.570e+03, percent-clipped=23.0
+2023-03-05 00:01:13,169 INFO [train.py:968] (0/2) Epoch 9, batch 41000, giga_loss[loss=0.3135, simple_loss=0.3784, pruned_loss=0.1243, over 28893.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3761, pruned_loss=0.1252, over 5688426.55 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.379, pruned_loss=0.1301, over 5680991.37 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3746, pruned_loss=0.1234, over 5695001.74 frames. ], batch size: 174, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:01:50,089 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-05 00:01:57,331 INFO [train.py:968] (0/2) Epoch 9, batch 41050, giga_loss[loss=0.3465, simple_loss=0.4029, pruned_loss=0.1451, over 28849.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3823, pruned_loss=0.1305, over 5677672.51 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3793, pruned_loss=0.1302, over 5669150.82 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3808, pruned_loss=0.1289, over 5692665.04 frames. ], batch size: 227, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:02:39,033 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.108e+02 1.529e+03 2.010e+03 2.735e+03 7.417e+03, threshold=4.020e+03, percent-clipped=4.0
+2023-03-05 00:02:43,356 INFO [train.py:968] (0/2) Epoch 9, batch 41100, giga_loss[loss=0.3251, simple_loss=0.3839, pruned_loss=0.1331, over 28762.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.39, pruned_loss=0.1374, over 5670124.34 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3791, pruned_loss=0.1301, over 5663529.41 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3891, pruned_loss=0.1363, over 5687102.23 frames. ], batch size: 99, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:02:57,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3632, 1.5199, 1.1919, 1.2107], device='cuda:0'), covar=tensor([0.1227, 0.1098, 0.0960, 0.1160], device='cuda:0'), in_proj_covar=tensor([0.1671, 0.1542, 0.1517, 0.1617], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 00:03:05,365 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=405319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:03:28,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=405342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:03:33,315 INFO [train.py:968] (0/2) Epoch 9, batch 41150, giga_loss[loss=0.288, simple_loss=0.3601, pruned_loss=0.1079, over 28456.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3932, pruned_loss=0.141, over 5649669.32 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3792, pruned_loss=0.1303, over 5659517.71 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3927, pruned_loss=0.1401, over 5667116.13 frames. ], batch size: 60, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:04:23,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.771e+02 1.656e+03 2.402e+03 2.985e+03 6.752e+03, threshold=4.803e+03, percent-clipped=10.0
+2023-03-05 00:04:27,759 INFO [train.py:968] (0/2) Epoch 9, batch 41200, giga_loss[loss=0.4037, simple_loss=0.4311, pruned_loss=0.1882, over 27874.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3957, pruned_loss=0.1441, over 5642730.63 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3793, pruned_loss=0.1304, over 5653374.43 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3954, pruned_loss=0.1435, over 5662263.35 frames. ], batch size: 412, lr: 3.52e-03, grad_scale: 8.0
+2023-03-05 00:05:19,463 INFO [train.py:968] (0/2) Epoch 9, batch 41250, giga_loss[loss=0.3967, simple_loss=0.4341, pruned_loss=0.1796, over 28633.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.3994, pruned_loss=0.1485, over 5634507.29 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3794, pruned_loss=0.1305, over 5657915.90 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.3994, pruned_loss=0.1482, over 5645800.11 frames. ], batch size: 336, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:05:35,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=405462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:05:39,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=405465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:05:58,112 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=405485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:06:00,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=405488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:06:07,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=405494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:06:07,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.732e+03 2.429e+03 3.779e+03 1.008e+04, threshold=4.858e+03, percent-clipped=13.0
+2023-03-05 00:06:09,798 INFO [train.py:968] (0/2) Epoch 9, batch 41300, giga_loss[loss=0.3611, simple_loss=0.4083, pruned_loss=0.157, over 28495.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4015, pruned_loss=0.151, over 5631565.50 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3788, pruned_loss=0.1302, over 5661899.60 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4027, pruned_loss=0.1517, over 5636072.24 frames. ], batch size: 65, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:06:31,110 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=405517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:06:31,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8831, 1.8031, 1.7313, 1.6635], device='cuda:0'), covar=tensor([0.1305, 0.2166, 0.1706, 0.1777], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0734, 0.0660, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 00:07:01,816 INFO [train.py:968] (0/2) Epoch 9, batch 41350, giga_loss[loss=0.3936, simple_loss=0.4119, pruned_loss=0.1876, over 23519.00 frames. ], tot_loss[loss=0.3532, simple_loss=0.4027, pruned_loss=0.1519, over 5623586.97 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3787, pruned_loss=0.1302, over 5664791.68 frames. ], giga_tot_loss[loss=0.3548, simple_loss=0.4041, pruned_loss=0.1528, over 5624117.18 frames. ], batch size: 705, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:07:37,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8117, 1.0877, 2.8571, 2.7414], device='cuda:0'), covar=tensor([0.1568, 0.2279, 0.0575, 0.1016], device='cuda:0'), in_proj_covar=tensor([0.0633, 0.0576, 0.0836, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 00:07:52,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.460e+02 1.751e+03 2.084e+03 2.664e+03 5.012e+03, threshold=4.168e+03, percent-clipped=1.0
+2023-03-05 00:07:55,670 INFO [train.py:968] (0/2) Epoch 9, batch 41400, giga_loss[loss=0.3156, simple_loss=0.3721, pruned_loss=0.1296, over 28848.00 frames. ], tot_loss[loss=0.3532, simple_loss=0.4021, pruned_loss=0.1522, over 5623594.29 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3787, pruned_loss=0.1301, over 5665754.72 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4033, pruned_loss=0.153, over 5622926.91 frames. ], batch size: 112, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:08:34,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3005, 5.1226, 4.8763, 2.6589], device='cuda:0'), covar=tensor([0.0404, 0.0552, 0.0639, 0.1628], device='cuda:0'), in_proj_covar=tensor([0.1004, 0.0938, 0.0831, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 00:08:41,880 INFO [train.py:968] (0/2) Epoch 9, batch 41450, giga_loss[loss=0.307, simple_loss=0.3701, pruned_loss=0.1219, over 28978.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.4006, pruned_loss=0.1512, over 5643541.22 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3787, pruned_loss=0.13, over 5671395.13 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4023, pruned_loss=0.1525, over 5637109.12 frames. ], batch size: 213, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:09:29,831 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.415e+02 1.639e+03 2.207e+03 2.998e+03 8.974e+03, threshold=4.415e+03, percent-clipped=5.0
+2023-03-05 00:09:32,893 INFO [train.py:968] (0/2) Epoch 9, batch 41500, giga_loss[loss=0.4349, simple_loss=0.4564, pruned_loss=0.2067, over 26572.00 frames. ], tot_loss[loss=0.3493, simple_loss=0.4002, pruned_loss=0.1492, over 5649626.10 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3786, pruned_loss=0.13, over 5673859.21 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.402, pruned_loss=0.1507, over 5641969.57 frames. ], batch size: 555, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:10:21,278 INFO [train.py:968] (0/2) Epoch 9, batch 41550, giga_loss[loss=0.33, simple_loss=0.3956, pruned_loss=0.1322, over 29028.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4005, pruned_loss=0.1483, over 5656450.41 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3785, pruned_loss=0.1299, over 5670502.20 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4026, pruned_loss=0.15, over 5652069.54 frames. ], batch size: 155, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:11:01,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4185, 1.8133, 1.4241, 1.5567], device='cuda:0'), covar=tensor([0.0650, 0.0365, 0.0308, 0.0688], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0082], device='cuda:0')
+2023-03-05 00:11:07,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.157e+03 1.563e+03 1.989e+03 2.863e+03 1.068e+04, threshold=3.978e+03, percent-clipped=4.0
+2023-03-05 00:11:10,335 INFO [train.py:968] (0/2) Epoch 9, batch 41600, giga_loss[loss=0.4299, simple_loss=0.447, pruned_loss=0.2064, over 24048.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4015, pruned_loss=0.1487, over 5646994.69 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3784, pruned_loss=0.13, over 5674810.79 frames. ], giga_tot_loss[loss=0.3525, simple_loss=0.4038, pruned_loss=0.1506, over 5639491.09 frames. ], batch size: 705, lr: 3.51e-03, grad_scale: 8.0
+2023-03-05 00:11:13,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2955, 1.4412, 1.5641, 1.4432], device='cuda:0'), covar=tensor([0.0906, 0.0763, 0.1080, 0.0797], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0742, 0.0669, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 00:12:01,040 INFO [train.py:968] (0/2) Epoch 9, batch 41650, giga_loss[loss=0.3199, simple_loss=0.3893, pruned_loss=0.1252, over 28866.00 frames. ], tot_loss[loss=0.3456, simple_loss=0.399, pruned_loss=0.1461, over 5641685.46 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3784, pruned_loss=0.1298, over 5669332.29 frames. ], giga_tot_loss[loss=0.3486, simple_loss=0.4012, pruned_loss=0.148, over 5639978.91 frames. ], batch size: 106, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:12:46,970 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.136e+02 1.461e+03 1.869e+03 2.538e+03 6.082e+03, threshold=3.737e+03, percent-clipped=6.0
+2023-03-05 00:12:48,324 INFO [train.py:968] (0/2) Epoch 9, batch 41700, giga_loss[loss=0.3264, simple_loss=0.3953, pruned_loss=0.1287, over 28868.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3973, pruned_loss=0.1432, over 5649203.51 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3783, pruned_loss=0.1297, over 5671824.35 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3994, pruned_loss=0.145, over 5645411.12 frames. ], batch size: 119, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:13:32,287 INFO [train.py:968] (0/2) Epoch 9, batch 41750, giga_loss[loss=0.2997, simple_loss=0.3736, pruned_loss=0.1129, over 28636.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3924, pruned_loss=0.1391, over 5662613.11 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3773, pruned_loss=0.1291, over 5678216.72 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3958, pruned_loss=0.1417, over 5653074.20 frames. ], batch size: 242, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:14:16,915 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.130e+02 1.450e+03 1.895e+03 2.770e+03 9.590e+03, threshold=3.789e+03, percent-clipped=12.0
+2023-03-05 00:14:19,571 INFO [train.py:968] (0/2) Epoch 9, batch 41800, giga_loss[loss=0.3348, simple_loss=0.377, pruned_loss=0.1463, over 23793.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3893, pruned_loss=0.1364, over 5658740.90 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3774, pruned_loss=0.1291, over 5680013.29 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3922, pruned_loss=0.1386, over 5649049.28 frames. ], batch size: 705, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:14:21,241 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-406000.pt
+2023-03-05 00:15:08,122 INFO [train.py:968] (0/2) Epoch 9, batch 41850, giga_loss[loss=0.3412, simple_loss=0.4006, pruned_loss=0.1409, over 28649.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3874, pruned_loss=0.1358, over 5652189.89 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3774, pruned_loss=0.1292, over 5684676.21 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3899, pruned_loss=0.1376, over 5640045.77 frames. ], batch size: 336, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:15:50,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.628e+02 1.530e+03 1.854e+03 2.431e+03 5.175e+03, threshold=3.707e+03, percent-clipped=5.0
+2023-03-05 00:15:51,919 INFO [train.py:968] (0/2) Epoch 9, batch 41900, giga_loss[loss=0.3757, simple_loss=0.4185, pruned_loss=0.1664, over 27909.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3863, pruned_loss=0.1344, over 5669151.29 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3773, pruned_loss=0.1291, over 5686260.58 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3886, pruned_loss=0.136, over 5657682.18 frames. ], batch size: 412, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:16:10,983 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=406117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:16:41,847 INFO [train.py:968] (0/2) Epoch 9, batch 41950, giga_loss[loss=0.2583, simple_loss=0.34, pruned_loss=0.0883, over 28575.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3854, pruned_loss=0.1332, over 5675417.51 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3777, pruned_loss=0.1294, over 5692077.94 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3872, pruned_loss=0.1344, over 5660801.03 frames. ], batch size: 60, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:17:30,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.386e+02 1.451e+03 2.058e+03 2.771e+03 6.692e+03, threshold=4.116e+03, percent-clipped=7.0
+2023-03-05 00:17:32,218 INFO [train.py:968] (0/2) Epoch 9, batch 42000, giga_loss[loss=0.2733, simple_loss=0.3542, pruned_loss=0.09618, over 29094.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.384, pruned_loss=0.1306, over 5677927.62 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3775, pruned_loss=0.1292, over 5688128.57 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3858, pruned_loss=0.1318, over 5669540.99 frames. ], batch size: 128, lr: 3.51e-03, grad_scale: 8.0
+2023-03-05 00:17:32,222 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 00:17:41,245 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2157, simple_loss=0.3201, pruned_loss=0.05563, over 944034.00 frames. 
+2023-03-05 00:17:41,246 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-05 00:17:46,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5012, 1.7133, 1.4737, 1.2291], device='cuda:0'), covar=tensor([0.2088, 0.1683, 0.1262, 0.1664], device='cuda:0'), in_proj_covar=tensor([0.1668, 0.1546, 0.1506, 0.1608], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 00:18:32,229 INFO [train.py:968] (0/2) Epoch 9, batch 42050, giga_loss[loss=0.3153, simple_loss=0.3896, pruned_loss=0.1205, over 28972.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3856, pruned_loss=0.1299, over 5682155.27 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3774, pruned_loss=0.1292, over 5693260.41 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3872, pruned_loss=0.1308, over 5670642.57 frames. ], batch size: 164, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:18:53,424 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-05 00:19:17,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.827e+02 1.604e+03 2.335e+03 3.289e+03 7.409e+03, threshold=4.669e+03, percent-clipped=12.0
+2023-03-05 00:19:18,638 INFO [train.py:968] (0/2) Epoch 9, batch 42100, giga_loss[loss=0.3112, simple_loss=0.375, pruned_loss=0.1237, over 28233.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.386, pruned_loss=0.1301, over 5683254.69 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3773, pruned_loss=0.1291, over 5697771.18 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3876, pruned_loss=0.131, over 5669980.88 frames. ], batch size: 77, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:19:23,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6335, 1.5549, 1.2903, 1.2749], device='cuda:0'), covar=tensor([0.0545, 0.0444, 0.0759, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0443, 0.0500, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 00:20:00,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8298, 1.8031, 1.3645, 1.4513], device='cuda:0'), covar=tensor([0.0686, 0.0591, 0.0952, 0.1019], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0441, 0.0499, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 00:20:05,189 INFO [train.py:968] (0/2) Epoch 9, batch 42150, giga_loss[loss=0.3759, simple_loss=0.4092, pruned_loss=0.1713, over 26592.00 frames. ], tot_loss[loss=0.324, simple_loss=0.386, pruned_loss=0.131, over 5683227.39 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3765, pruned_loss=0.1285, over 5702317.88 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3883, pruned_loss=0.1323, over 5667943.39 frames. ], batch size: 555, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:20:51,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.596e+03 1.943e+03 2.653e+03 5.859e+03, threshold=3.886e+03, percent-clipped=3.0
+2023-03-05 00:20:52,137 INFO [train.py:968] (0/2) Epoch 9, batch 42200, giga_loss[loss=0.3572, simple_loss=0.4066, pruned_loss=0.1539, over 28951.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3848, pruned_loss=0.1305, over 5685957.67 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3766, pruned_loss=0.1286, over 5704370.55 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3866, pruned_loss=0.1315, over 5671862.51 frames. ], batch size: 186, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:21:37,969 INFO [train.py:968] (0/2) Epoch 9, batch 42250, giga_loss[loss=0.3037, simple_loss=0.3673, pruned_loss=0.1201, over 28812.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3829, pruned_loss=0.1304, over 5687619.23 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3764, pruned_loss=0.1285, over 5710684.64 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3848, pruned_loss=0.1314, over 5669898.02 frames. ], batch size: 284, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:22:19,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=406492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:22:20,042 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=406493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:22:26,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.625e+02 1.690e+03 2.344e+03 3.117e+03 8.798e+03, threshold=4.688e+03, percent-clipped=16.0
+2023-03-05 00:22:26,751 INFO [train.py:968] (0/2) Epoch 9, batch 42300, giga_loss[loss=0.2808, simple_loss=0.3611, pruned_loss=0.1003, over 28667.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3816, pruned_loss=0.1298, over 5669304.11 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3766, pruned_loss=0.1286, over 5701728.22 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3829, pruned_loss=0.1305, over 5663224.03 frames. ], batch size: 66, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:22:58,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9727, 0.9733, 3.7950, 3.0265], device='cuda:0'), covar=tensor([0.1806, 0.2773, 0.0446, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0636, 0.0578, 0.0833, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 00:23:11,790 INFO [train.py:968] (0/2) Epoch 9, batch 42350, libri_loss[loss=0.3251, simple_loss=0.3867, pruned_loss=0.1318, over 27593.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3811, pruned_loss=0.1282, over 5680154.16 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3762, pruned_loss=0.1281, over 5704861.48 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3827, pruned_loss=0.1292, over 5670896.28 frames. ], batch size: 116, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:23:12,010 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=406548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:23:55,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.953e+02 1.324e+03 1.849e+03 2.783e+03 1.047e+04, threshold=3.698e+03, percent-clipped=7.0
+2023-03-05 00:23:56,139 INFO [train.py:968] (0/2) Epoch 9, batch 42400, giga_loss[loss=0.3483, simple_loss=0.4029, pruned_loss=0.1468, over 28817.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3815, pruned_loss=0.1278, over 5679801.60 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3764, pruned_loss=0.1284, over 5704689.37 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3827, pruned_loss=0.1283, over 5672238.79 frames. ], batch size: 186, lr: 3.51e-03, grad_scale: 8.0
+2023-03-05 00:24:12,235 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=406614.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:24:22,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7986, 5.6249, 5.2632, 2.4995], device='cuda:0'), covar=tensor([0.0392, 0.0568, 0.0701, 0.1885], device='cuda:0'), in_proj_covar=tensor([0.1008, 0.0943, 0.0830, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 00:24:29,932 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=406635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:24:32,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=406638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:24:40,910 INFO [train.py:968] (0/2) Epoch 9, batch 42450, giga_loss[loss=0.2936, simple_loss=0.3672, pruned_loss=0.11, over 28925.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3821, pruned_loss=0.1281, over 5683962.02 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3767, pruned_loss=0.1283, over 5699381.00 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.383, pruned_loss=0.1286, over 5682759.14 frames. ], batch size: 145, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:24:44,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4144, 1.6643, 1.2484, 1.7212], device='cuda:0'), covar=tensor([0.2337, 0.2369, 0.2647, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.1258, 0.0938, 0.1116, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 00:24:57,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=406667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:25:22,638 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6703, 1.8357, 1.9099, 1.4827], device='cuda:0'), covar=tensor([0.1645, 0.2148, 0.1268, 0.1463], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0701, 0.0838, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-05 00:25:23,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.506e+03 1.934e+03 2.803e+03 6.461e+03, threshold=3.868e+03, percent-clipped=8.0
+2023-03-05 00:25:23,710 INFO [train.py:968] (0/2) Epoch 9, batch 42500, giga_loss[loss=0.2854, simple_loss=0.3583, pruned_loss=0.1063, over 28722.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3804, pruned_loss=0.1278, over 5682159.46 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3767, pruned_loss=0.1283, over 5701761.99 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3812, pruned_loss=0.1283, over 5678965.42 frames. ], batch size: 262, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:25:54,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6852, 1.7382, 1.5249, 1.5847], device='cuda:0'), covar=tensor([0.1173, 0.1934, 0.1721, 0.1732], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0740, 0.0665, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 00:26:13,506 INFO [train.py:968] (0/2) Epoch 9, batch 42550, libri_loss[loss=0.2632, simple_loss=0.3267, pruned_loss=0.09986, over 29396.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3803, pruned_loss=0.1286, over 5685504.11 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3764, pruned_loss=0.128, over 5706896.62 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3813, pruned_loss=0.1293, over 5677297.99 frames. ], batch size: 67, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:26:35,378 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=406770.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:26:58,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.250e+02 1.533e+03 2.030e+03 2.844e+03 9.025e+03, threshold=4.059e+03, percent-clipped=11.0
+2023-03-05 00:26:58,881 INFO [train.py:968] (0/2) Epoch 9, batch 42600, giga_loss[loss=0.3204, simple_loss=0.381, pruned_loss=0.1298, over 28838.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3794, pruned_loss=0.1292, over 5682593.08 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3761, pruned_loss=0.1278, over 5711353.69 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3807, pruned_loss=0.1299, over 5671187.77 frames. ], batch size: 186, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:27:24,578 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2456, 3.0718, 2.9098, 1.3858], device='cuda:0'), covar=tensor([0.0907, 0.0963, 0.0873, 0.2364], device='cuda:0'), in_proj_covar=tensor([0.1025, 0.0957, 0.0842, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 00:27:44,453 INFO [train.py:968] (0/2) Epoch 9, batch 42650, giga_loss[loss=0.2743, simple_loss=0.3478, pruned_loss=0.1004, over 28934.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3783, pruned_loss=0.129, over 5674534.40 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3763, pruned_loss=0.1278, over 5709478.30 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3791, pruned_loss=0.1296, over 5666049.02 frames. ], batch size: 164, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:28:05,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8550, 4.5934, 1.8552, 1.8669], device='cuda:0'), covar=tensor([0.0801, 0.0292, 0.0801, 0.1148], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0505, 0.0331, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 00:28:05,579 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=406868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:28:10,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4965, 1.6886, 1.3734, 1.7159], device='cuda:0'), covar=tensor([0.2359, 0.2367, 0.2668, 0.2266], device='cuda:0'), in_proj_covar=tensor([0.1257, 0.0933, 0.1114, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 00:28:27,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0109, 1.2037, 3.3481, 2.8956], device='cuda:0'), covar=tensor([0.1592, 0.2359, 0.0493, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.0636, 0.0577, 0.0835, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 00:28:32,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.362e+02 1.591e+03 2.031e+03 2.942e+03 7.316e+03, threshold=4.062e+03, percent-clipped=10.0
+2023-03-05 00:28:32,346 INFO [train.py:968] (0/2) Epoch 9, batch 42700, giga_loss[loss=0.313, simple_loss=0.3777, pruned_loss=0.1242, over 28258.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3761, pruned_loss=0.1275, over 5676478.53 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.376, pruned_loss=0.1275, over 5708845.06 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3771, pruned_loss=0.1283, over 5669612.89 frames. ], batch size: 368, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:28:53,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=406923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:29:07,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4385, 3.7800, 1.6455, 1.4659], device='cuda:0'), covar=tensor([0.0914, 0.0309, 0.0807, 0.1306], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0506, 0.0331, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 00:29:14,173 INFO [train.py:968] (0/2) Epoch 9, batch 42750, giga_loss[loss=0.2858, simple_loss=0.3575, pruned_loss=0.107, over 28646.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3759, pruned_loss=0.1276, over 5676678.86 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3755, pruned_loss=0.1272, over 5698765.84 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3772, pruned_loss=0.1285, over 5677965.29 frames. ], batch size: 307, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:29:52,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=406989.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:29:58,760 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.507e+03 1.898e+03 3.023e+03 9.069e+03, threshold=3.796e+03, percent-clipped=13.0
+2023-03-05 00:29:58,772 INFO [train.py:968] (0/2) Epoch 9, batch 42800, giga_loss[loss=0.2718, simple_loss=0.3525, pruned_loss=0.09552, over 28831.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3765, pruned_loss=0.127, over 5683239.65 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3754, pruned_loss=0.1271, over 5702707.04 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3776, pruned_loss=0.1279, over 5680124.76 frames. ], batch size: 112, lr: 3.51e-03, grad_scale: 8.0
+2023-03-05 00:30:05,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0957, 2.8118, 1.8809, 1.5495], device='cuda:0'), covar=tensor([0.2097, 0.1124, 0.1489, 0.1902], device='cuda:0'), in_proj_covar=tensor([0.1652, 0.1547, 0.1496, 0.1594], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 00:30:07,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407005.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:30:12,608 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:30:15,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:30:34,889 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:30:42,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:30:46,793 INFO [train.py:968] (0/2) Epoch 9, batch 42850, giga_loss[loss=0.3022, simple_loss=0.3716, pruned_loss=0.1164, over 28920.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3766, pruned_loss=0.1263, over 5678188.24 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3752, pruned_loss=0.1269, over 5702077.64 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3776, pruned_loss=0.1271, over 5675977.41 frames. ], batch size: 285, lr: 3.51e-03, grad_scale: 8.0
+2023-03-05 00:31:04,063 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:31:05,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:31:06,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4234, 1.5264, 1.5101, 1.4169], device='cuda:0'), covar=tensor([0.1060, 0.1358, 0.1640, 0.1406], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0733, 0.0661, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 00:31:27,278 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 00:31:33,380 INFO [train.py:968] (0/2) Epoch 9, batch 42900, libri_loss[loss=0.3534, simple_loss=0.4036, pruned_loss=0.1516, over 19829.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3773, pruned_loss=0.1264, over 5658785.88 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3755, pruned_loss=0.1271, over 5684046.84 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3779, pruned_loss=0.1268, over 5673355.61 frames. ], batch size: 187, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:31:33,652 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:31:34,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.845e+02 1.655e+03 2.128e+03 3.020e+03 8.028e+03, threshold=4.256e+03, percent-clipped=11.0
+2023-03-05 00:32:06,346 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407132.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:32:08,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407135.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:32:13,741 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 00:32:18,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=407145.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:32:20,716 INFO [train.py:968] (0/2) Epoch 9, batch 42950, giga_loss[loss=0.3479, simple_loss=0.4003, pruned_loss=0.1477, over 28698.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3793, pruned_loss=0.1283, over 5655608.05 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3757, pruned_loss=0.1274, over 5680325.21 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3797, pruned_loss=0.1284, over 5669640.01 frames. ], batch size: 262, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:32:38,261 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407164.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:32:41,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1331, 2.8295, 1.2020, 1.1915], device='cuda:0'), covar=tensor([0.1121, 0.0438, 0.1012, 0.1590], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0505, 0.0332, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 00:32:44,400 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-05 00:33:08,770 INFO [train.py:968] (0/2) Epoch 9, batch 43000, giga_loss[loss=0.2952, simple_loss=0.3731, pruned_loss=0.1086, over 28912.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3812, pruned_loss=0.1303, over 5653950.28 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.376, pruned_loss=0.1276, over 5683724.65 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3813, pruned_loss=0.1303, over 5661549.57 frames. ], batch size: 199, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:33:09,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.798e+02 1.557e+03 1.821e+03 2.384e+03 5.858e+03, threshold=3.643e+03, percent-clipped=2.0
+2023-03-05 00:33:11,756 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-05 00:33:57,911 INFO [train.py:968] (0/2) Epoch 9, batch 43050, giga_loss[loss=0.357, simple_loss=0.4023, pruned_loss=0.1559, over 28251.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3851, pruned_loss=0.1351, over 5659791.67 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3757, pruned_loss=0.1274, over 5690371.58 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3858, pruned_loss=0.1354, over 5659165.21 frames. ], batch size: 368, lr: 3.51e-03, grad_scale: 2.0
+2023-03-05 00:34:38,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407288.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:34:43,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407291.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:34:49,743 INFO [train.py:968] (0/2) Epoch 9, batch 43100, giga_loss[loss=0.3304, simple_loss=0.3817, pruned_loss=0.1395, over 28960.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3862, pruned_loss=0.1376, over 5650775.03 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3752, pruned_loss=0.127, over 5689703.40 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3875, pruned_loss=0.1385, over 5649692.16 frames. ], batch size: 106, lr: 3.51e-03, grad_scale: 2.0
+2023-03-05 00:34:51,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.213e+03 1.847e+03 2.576e+03 4.008e+03 1.506e+04, threshold=5.152e+03, percent-clipped=29.0
+2023-03-05 00:35:12,937 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407320.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:35:41,278 INFO [train.py:968] (0/2) Epoch 9, batch 43150, giga_loss[loss=0.3915, simple_loss=0.4236, pruned_loss=0.1797, over 28008.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3863, pruned_loss=0.1379, over 5654582.00 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3751, pruned_loss=0.1269, over 5685765.17 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3878, pruned_loss=0.139, over 5656381.21 frames. ], batch size: 412, lr: 3.51e-03, grad_scale: 2.0
+2023-03-05 00:36:09,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=407380.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:36:23,082 INFO [train.py:968] (0/2) Epoch 9, batch 43200, giga_loss[loss=0.3313, simple_loss=0.3867, pruned_loss=0.1379, over 28933.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3841, pruned_loss=0.1358, over 5667283.80 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.375, pruned_loss=0.1268, over 5685490.94 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3855, pruned_loss=0.137, over 5668581.80 frames. ], batch size: 145, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:36:25,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.220e+02 1.713e+03 2.196e+03 3.103e+03 8.974e+03, threshold=4.392e+03, percent-clipped=5.0
+2023-03-05 00:36:34,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=407410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:37:08,870 INFO [train.py:968] (0/2) Epoch 9, batch 43250, giga_loss[loss=0.2912, simple_loss=0.3574, pruned_loss=0.1125, over 28923.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3813, pruned_loss=0.1329, over 5673095.18 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3745, pruned_loss=0.1264, over 5691151.51 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3831, pruned_loss=0.1344, over 5668191.20 frames. ], batch size: 227, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:37:41,621 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:37:41,950 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-05 00:37:47,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6813, 2.0832, 1.3531, 0.8851], device='cuda:0'), covar=tensor([0.3956, 0.3161, 0.1856, 0.4069], device='cuda:0'), in_proj_covar=tensor([0.1509, 0.1457, 0.1466, 0.1241], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 00:37:54,424 INFO [train.py:968] (0/2) Epoch 9, batch 43300, giga_loss[loss=0.3417, simple_loss=0.3877, pruned_loss=0.1479, over 28850.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3791, pruned_loss=0.1299, over 5681722.69 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3747, pruned_loss=0.1265, over 5693096.00 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3805, pruned_loss=0.1311, over 5675986.30 frames. ], batch size: 186, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:37:55,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.420e+02 1.582e+03 1.928e+03 2.598e+03 5.585e+03, threshold=3.856e+03, percent-clipped=3.0
+2023-03-05 00:38:18,949 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407523.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:38:20,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407526.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:38:39,814 INFO [train.py:968] (0/2) Epoch 9, batch 43350, libri_loss[loss=0.3275, simple_loss=0.3957, pruned_loss=0.1297, over 29647.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3776, pruned_loss=0.1292, over 5652509.42 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.375, pruned_loss=0.1266, over 5677085.74 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3785, pruned_loss=0.1301, over 5662539.20 frames. ], batch size: 91, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:38:44,740 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:38:46,093 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407555.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:38:46,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:38:51,418 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 00:39:10,638 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:39:23,698 INFO [train.py:968] (0/2) Epoch 9, batch 43400, giga_loss[loss=0.3257, simple_loss=0.3781, pruned_loss=0.1366, over 28955.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3764, pruned_loss=0.1291, over 5663267.06 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.375, pruned_loss=0.1266, over 5681424.32 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3771, pruned_loss=0.1299, over 5666961.11 frames. ], batch size: 106, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:39:26,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.086e+03 1.638e+03 2.117e+03 3.307e+03 7.529e+03, threshold=4.233e+03, percent-clipped=15.0
+2023-03-05 00:40:09,674 INFO [train.py:968] (0/2) Epoch 9, batch 43450, giga_loss[loss=0.3696, simple_loss=0.4102, pruned_loss=0.1645, over 27535.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3777, pruned_loss=0.131, over 5649878.73 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3753, pruned_loss=0.1267, over 5676930.41 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.378, pruned_loss=0.1316, over 5656627.16 frames. ], batch size: 472, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:40:51,455 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7183, 1.8561, 1.5473, 1.3674], device='cuda:0'), covar=tensor([0.1762, 0.1579, 0.1427, 0.1703], device='cuda:0'), in_proj_covar=tensor([0.1655, 0.1552, 0.1503, 0.1597], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 00:40:54,400 INFO [train.py:968] (0/2) Epoch 9, batch 43500, giga_loss[loss=0.3245, simple_loss=0.3929, pruned_loss=0.128, over 29062.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3808, pruned_loss=0.1327, over 5666367.91 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3755, pruned_loss=0.1271, over 5683603.08 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3809, pruned_loss=0.133, over 5665352.54 frames. ], batch size: 128, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:40:56,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.652e+02 1.588e+03 2.057e+03 2.628e+03 5.744e+03, threshold=4.114e+03, percent-clipped=5.0
+2023-03-05 00:41:42,896 INFO [train.py:968] (0/2) Epoch 9, batch 43550, giga_loss[loss=0.2817, simple_loss=0.3637, pruned_loss=0.09981, over 28593.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3838, pruned_loss=0.1323, over 5664982.21 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3752, pruned_loss=0.127, over 5685924.35 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3843, pruned_loss=0.1327, over 5661901.87 frames. ], batch size: 60, lr: 3.51e-03, grad_scale: 2.0
+2023-03-05 00:41:54,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3653, 3.4873, 1.4295, 1.4885], device='cuda:0'), covar=tensor([0.0933, 0.0345, 0.0940, 0.1347], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0505, 0.0332, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 00:42:26,784 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:42:34,431 INFO [train.py:968] (0/2) Epoch 9, batch 43600, giga_loss[loss=0.3, simple_loss=0.3657, pruned_loss=0.1172, over 28779.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.387, pruned_loss=0.1331, over 5659659.76 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3756, pruned_loss=0.1273, over 5686750.43 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3872, pruned_loss=0.1332, over 5655745.15 frames. ], batch size: 99, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:42:38,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.108e+02 1.478e+03 2.016e+03 2.670e+03 5.438e+03, threshold=4.033e+03, percent-clipped=9.0
+2023-03-05 00:42:52,441 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-05 00:43:19,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7270, 1.6699, 1.2679, 1.3445], device='cuda:0'), covar=tensor([0.0613, 0.0512, 0.0889, 0.1026], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0440, 0.0495, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 00:43:24,525 INFO [train.py:968] (0/2) Epoch 9, batch 43650, giga_loss[loss=0.3684, simple_loss=0.4186, pruned_loss=0.1591, over 28681.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3887, pruned_loss=0.1344, over 5670341.77 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3753, pruned_loss=0.1271, over 5687567.04 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3893, pruned_loss=0.1348, over 5666110.67 frames. ], batch size: 284, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:43:34,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=407859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:43:45,271 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:44:02,295 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:44:06,752 INFO [train.py:968] (0/2) Epoch 9, batch 43700, giga_loss[loss=0.3841, simple_loss=0.4233, pruned_loss=0.1724, over 28546.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3891, pruned_loss=0.1353, over 5667492.37 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3753, pruned_loss=0.1273, over 5685533.63 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.39, pruned_loss=0.1356, over 5666118.92 frames. ], batch size: 336, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:44:09,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.932e+02 1.669e+03 2.214e+03 3.008e+03 8.468e+03, threshold=4.427e+03, percent-clipped=16.0
+2023-03-05 00:44:16,478 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407908.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:44:23,926 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-05 00:44:48,841 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-05 00:44:49,719 INFO [train.py:968] (0/2) Epoch 9, batch 43750, giga_loss[loss=0.2939, simple_loss=0.3561, pruned_loss=0.1159, over 28606.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3874, pruned_loss=0.1346, over 5669649.39 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3751, pruned_loss=0.127, over 5686249.61 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3889, pruned_loss=0.1354, over 5667176.27 frames. ], batch size: 92, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:45:40,686 INFO [train.py:968] (0/2) Epoch 9, batch 43800, giga_loss[loss=0.3433, simple_loss=0.3823, pruned_loss=0.1522, over 23601.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3867, pruned_loss=0.1352, over 5658687.00 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3751, pruned_loss=0.127, over 5686249.61 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3879, pruned_loss=0.1358, over 5656762.14 frames. ], batch size: 705, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:45:40,865 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:45:41,993 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-408000.pt
+2023-03-05 00:45:42,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.248e+02 1.505e+03 1.799e+03 2.212e+03 1.117e+04, threshold=3.599e+03, percent-clipped=4.0
+2023-03-05 00:45:44,102 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:45:46,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:46:13,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:46:26,927 INFO [train.py:968] (0/2) Epoch 9, batch 43850, giga_loss[loss=0.3531, simple_loss=0.4053, pruned_loss=0.1504, over 28931.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3845, pruned_loss=0.1342, over 5668418.33 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3752, pruned_loss=0.127, over 5690424.23 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3854, pruned_loss=0.1348, over 5662785.04 frames. ], batch size: 145, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:47:14,826 INFO [train.py:968] (0/2) Epoch 9, batch 43900, giga_loss[loss=0.3419, simple_loss=0.3865, pruned_loss=0.1486, over 28507.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3827, pruned_loss=0.1335, over 5668884.83 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3752, pruned_loss=0.127, over 5693239.78 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3836, pruned_loss=0.1342, over 5661655.47 frames. ], batch size: 336, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:47:14,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8551, 3.6673, 3.4549, 2.0563], device='cuda:0'), covar=tensor([0.0500, 0.0675, 0.0682, 0.1706], device='cuda:0'), in_proj_covar=tensor([0.1014, 0.0961, 0.0835, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 00:47:17,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.369e+02 1.574e+03 1.908e+03 2.778e+03 9.831e+03, threshold=3.816e+03, percent-clipped=13.0
+2023-03-05 00:47:26,321 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-05 00:48:04,851 INFO [train.py:968] (0/2) Epoch 9, batch 43950, giga_loss[loss=0.3096, simple_loss=0.3747, pruned_loss=0.1223, over 28978.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3834, pruned_loss=0.1343, over 5646441.20 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1272, over 5684236.84 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.384, pruned_loss=0.1347, over 5647138.81 frames. ], batch size: 164, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:48:23,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:48:52,132 INFO [train.py:968] (0/2) Epoch 9, batch 44000, giga_loss[loss=0.3112, simple_loss=0.3736, pruned_loss=0.1244, over 28960.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3828, pruned_loss=0.1346, over 5650628.48 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3757, pruned_loss=0.1272, over 5686371.08 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3833, pruned_loss=0.1351, over 5648388.56 frames. ], batch size: 145, lr: 3.50e-03, grad_scale: 8.0
+2023-03-05 00:48:54,170 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.666e+03 2.223e+03 2.749e+03 7.588e+03, threshold=4.447e+03, percent-clipped=10.0
+2023-03-05 00:49:37,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:49:38,157 INFO [train.py:968] (0/2) Epoch 9, batch 44050, giga_loss[loss=0.3439, simple_loss=0.4062, pruned_loss=0.1408, over 28881.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3816, pruned_loss=0.1335, over 5666321.73 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3758, pruned_loss=0.1271, over 5691144.75 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.382, pruned_loss=0.1342, over 5659379.42 frames. ], batch size: 145, lr: 3.50e-03, grad_scale: 8.0
+2023-03-05 00:49:51,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:50:04,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408283.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:50:16,885 INFO [train.py:968] (0/2) Epoch 9, batch 44100, giga_loss[loss=0.3279, simple_loss=0.3858, pruned_loss=0.135, over 28630.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3806, pruned_loss=0.1325, over 5662408.04 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3756, pruned_loss=0.127, over 5690316.46 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3814, pruned_loss=0.1334, over 5656336.42 frames. ], batch size: 307, lr: 3.50e-03, grad_scale: 8.0
+2023-03-05 00:50:19,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.234e+02 1.480e+03 1.999e+03 2.594e+03 4.591e+03, threshold=3.998e+03, percent-clipped=2.0
+2023-03-05 00:50:25,667 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:50:27,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:50:33,019 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 00:50:41,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4150, 1.7147, 1.6512, 1.1946], device='cuda:0'), covar=tensor([0.1524, 0.2444, 0.1325, 0.1585], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0707, 0.0844, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 00:50:56,966 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:03,656 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-05 00:51:04,506 INFO [train.py:968] (0/2) Epoch 9, batch 44150, libri_loss[loss=0.3277, simple_loss=0.3903, pruned_loss=0.1326, over 29260.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3823, pruned_loss=0.1328, over 5662432.90 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3755, pruned_loss=0.1269, over 5695390.13 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3832, pruned_loss=0.1337, over 5651908.42 frames. ], batch size: 97, lr: 3.50e-03, grad_scale: 8.0
+2023-03-05 00:51:12,019 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=408356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:26,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:38,600 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:42,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:45,036 INFO [train.py:968] (0/2) Epoch 9, batch 44200, libri_loss[loss=0.3197, simple_loss=0.3762, pruned_loss=0.1316, over 29541.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3843, pruned_loss=0.1342, over 5668661.51 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3751, pruned_loss=0.1266, over 5701615.60 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3857, pruned_loss=0.1355, over 5653047.19 frames. ], batch size: 80, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:51:49,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.580e+03 2.042e+03 2.890e+03 6.473e+03, threshold=4.085e+03, percent-clipped=7.0
+2023-03-05 00:51:56,079 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:59,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:52:03,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2952, 2.7583, 1.4045, 1.3576], device='cuda:0'), covar=tensor([0.0870, 0.0335, 0.0804, 0.1269], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0504, 0.0331, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 00:52:08,791 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:52:11,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408426.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:52:14,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408429.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:52:16,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1452, 0.9864, 1.0824, 1.3953], device='cuda:0'), covar=tensor([0.0726, 0.0312, 0.0297, 0.0746], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0114, 0.0115, 0.0209], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:0')
+2023-03-05 00:52:21,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3230, 1.5801, 1.3067, 1.2720], device='cuda:0'), covar=tensor([0.2168, 0.1981, 0.2131, 0.1718], device='cuda:0'), in_proj_covar=tensor([0.1255, 0.0932, 0.1112, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 00:52:26,370 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:52:32,073 INFO [train.py:968] (0/2) Epoch 9, batch 44250, libri_loss[loss=0.3576, simple_loss=0.4056, pruned_loss=0.1548, over 29530.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.384, pruned_loss=0.1347, over 5677946.32 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.375, pruned_loss=0.1267, over 5707893.23 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3854, pruned_loss=0.1359, over 5658829.33 frames. ], batch size: 89, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:52:42,646 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408458.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:53:15,392 INFO [train.py:968] (0/2) Epoch 9, batch 44300, libri_loss[loss=0.2515, simple_loss=0.3162, pruned_loss=0.09339, over 29495.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3849, pruned_loss=0.1329, over 5660693.37 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3751, pruned_loss=0.1269, over 5687103.97 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3864, pruned_loss=0.134, over 5663154.17 frames. ], batch size: 70, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:53:19,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.877e+02 1.505e+03 1.908e+03 3.236e+03 6.892e+03, threshold=3.816e+03, percent-clipped=9.0
+2023-03-05 00:53:28,252 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=408511.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:53:31,578 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:53:33,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:53:55,985 INFO [train.py:968] (0/2) Epoch 9, batch 44350, giga_loss[loss=0.313, simple_loss=0.3974, pruned_loss=0.1142, over 28847.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3853, pruned_loss=0.1314, over 5655623.12 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3751, pruned_loss=0.1273, over 5684152.83 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3868, pruned_loss=0.1321, over 5659861.95 frames. ], batch size: 112, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:53:56,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:54:06,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=408559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:54:14,027 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=408567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:54:46,915 INFO [train.py:968] (0/2) Epoch 9, batch 44400, giga_loss[loss=0.2736, simple_loss=0.3482, pruned_loss=0.09953, over 28393.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3876, pruned_loss=0.1322, over 5651966.64 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3749, pruned_loss=0.1271, over 5685910.46 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3892, pruned_loss=0.133, over 5653544.74 frames. ], batch size: 71, lr: 3.50e-03, grad_scale: 8.0
+2023-03-05 00:54:50,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.294e+02 1.283e+03 1.618e+03 2.330e+03 1.009e+04, threshold=3.235e+03, percent-clipped=12.0
+2023-03-05 00:55:11,247 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 00:55:23,120 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-05 00:55:31,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4268, 1.6632, 1.3954, 1.3136], device='cuda:0'), covar=tensor([0.1542, 0.1392, 0.1413, 0.1417], device='cuda:0'), in_proj_covar=tensor([0.1660, 0.1562, 0.1506, 0.1606], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 00:55:35,110 INFO [train.py:968] (0/2) Epoch 9, batch 44450, giga_loss[loss=0.3786, simple_loss=0.4205, pruned_loss=0.1683, over 27921.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3917, pruned_loss=0.1364, over 5645511.90 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.375, pruned_loss=0.1271, over 5677092.99 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.393, pruned_loss=0.1371, over 5654498.20 frames. ], batch size: 412, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:55:46,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2834, 1.8552, 1.3818, 0.3614], device='cuda:0'), covar=tensor([0.2552, 0.1862, 0.2721, 0.3652], device='cuda:0'), in_proj_covar=tensor([0.1520, 0.1460, 0.1463, 0.1237], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 00:56:24,728 INFO [train.py:968] (0/2) Epoch 9, batch 44500, libri_loss[loss=0.4044, simple_loss=0.4331, pruned_loss=0.1878, over 18805.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3929, pruned_loss=0.1387, over 5630004.34 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3754, pruned_loss=0.1275, over 5661681.70 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3942, pruned_loss=0.1392, over 5650600.97 frames. ], batch size: 187, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:56:32,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.622e+02 1.562e+03 2.423e+03 3.928e+03 1.736e+04, threshold=4.847e+03, percent-clipped=33.0
+2023-03-05 00:56:55,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:57:13,166 INFO [train.py:968] (0/2) Epoch 9, batch 44550, giga_loss[loss=0.3564, simple_loss=0.3963, pruned_loss=0.1583, over 28838.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3919, pruned_loss=0.1383, over 5652671.34 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3761, pruned_loss=0.1281, over 5666287.27 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3927, pruned_loss=0.1384, over 5664568.45 frames. ], batch size: 99, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:57:52,030 INFO [train.py:968] (0/2) Epoch 9, batch 44600, libri_loss[loss=0.2522, simple_loss=0.3202, pruned_loss=0.09208, over 29633.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3902, pruned_loss=0.1371, over 5631914.22 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3759, pruned_loss=0.1282, over 5647783.08 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3916, pruned_loss=0.1375, over 5656489.91 frames. ], batch size: 69, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:57:56,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.697e+03 2.074e+03 3.231e+03 1.108e+04, threshold=4.148e+03, percent-clipped=8.0
+2023-03-05 00:58:29,802 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=408841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:58:34,970 INFO [train.py:968] (0/2) Epoch 9, batch 44650, libri_loss[loss=0.3795, simple_loss=0.4167, pruned_loss=0.1712, over 19222.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3879, pruned_loss=0.1334, over 5648333.61 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3758, pruned_loss=0.1282, over 5644493.61 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3894, pruned_loss=0.1339, over 5671786.59 frames. ], batch size: 187, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:59:00,441 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:59:03,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:59:03,826 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-05 00:59:13,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:59:23,700 INFO [train.py:968] (0/2) Epoch 9, batch 44700, giga_loss[loss=0.3164, simple_loss=0.3857, pruned_loss=0.1235, over 28754.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3886, pruned_loss=0.1323, over 5655739.54 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3758, pruned_loss=0.1282, over 5645565.14 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3899, pruned_loss=0.1327, over 5673242.36 frames. ], batch size: 284, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:59:28,535 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.441e+02 1.394e+03 1.758e+03 2.277e+03 5.548e+03, threshold=3.516e+03, percent-clipped=3.0
+2023-03-05 00:59:31,155 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:59:31,334 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 00:59:58,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:00:05,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:00:10,202 INFO [train.py:968] (0/2) Epoch 9, batch 44750, libri_loss[loss=0.3044, simple_loss=0.3753, pruned_loss=0.1167, over 27234.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3889, pruned_loss=0.1331, over 5650427.47 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3754, pruned_loss=0.1279, over 5651236.83 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3907, pruned_loss=0.1338, over 5659832.56 frames. ], batch size: 115, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:00:57,268 INFO [train.py:968] (0/2) Epoch 9, batch 44800, giga_loss[loss=0.3255, simple_loss=0.3883, pruned_loss=0.1313, over 28565.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3889, pruned_loss=0.1339, over 5651249.28 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3749, pruned_loss=0.1275, over 5656357.62 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.391, pruned_loss=0.1349, over 5654123.61 frames. ], batch size: 78, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 01:01:00,942 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.295e+02 1.611e+03 2.337e+03 3.442e+03 8.945e+03, threshold=4.673e+03, percent-clipped=23.0
+2023-03-05 01:01:11,089 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-05 01:01:21,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=409029.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 01:01:24,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=409032.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 01:01:37,452 INFO [train.py:968] (0/2) Epoch 9, batch 44850, giga_loss[loss=0.295, simple_loss=0.3544, pruned_loss=0.1178, over 28945.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.387, pruned_loss=0.1329, over 5665247.61 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3751, pruned_loss=0.1276, over 5663783.38 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3891, pruned_loss=0.1339, over 5660764.79 frames. ], batch size: 106, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 01:01:49,474 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=409061.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 01:01:49,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1708, 1.7805, 1.4586, 0.3267], device='cuda:0'), covar=tensor([0.2465, 0.1588, 0.2113, 0.3214], device='cuda:0'), in_proj_covar=tensor([0.1533, 0.1460, 0.1479, 0.1251], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 01:02:06,333 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=409077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:02:08,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=409080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:02:13,400 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=409085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:02:16,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=409088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:02:25,606 INFO [train.py:968] (0/2) Epoch 9, batch 44900, giga_loss[loss=0.3383, simple_loss=0.4016, pruned_loss=0.1375, over 28860.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.385, pruned_loss=0.1328, over 5660828.70 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.375, pruned_loss=0.1275, over 5661347.83 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3871, pruned_loss=0.1338, over 5658861.12 frames. ], batch size: 174, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 01:02:31,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.797e+02 1.482e+03 1.981e+03 2.686e+03 5.666e+03, threshold=3.962e+03, percent-clipped=2.0
+2023-03-05 01:02:38,922 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=409109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:02:44,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=409117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:03:16,359 INFO [train.py:968] (0/2) Epoch 9, batch 44950, giga_loss[loss=0.3129, simple_loss=0.3734, pruned_loss=0.1262, over 28759.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3837, pruned_loss=0.1325, over 5660534.34 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3748, pruned_loss=0.1273, over 5664860.90 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3856, pruned_loss=0.1336, over 5655959.61 frames. ], batch size: 284, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:04:04,859 INFO [train.py:968] (0/2) Epoch 9, batch 45000, giga_loss[loss=0.3192, simple_loss=0.3707, pruned_loss=0.1338, over 28903.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3825, pruned_loss=0.1326, over 5658808.85 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3748, pruned_loss=0.1273, over 5665997.40 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3842, pruned_loss=0.1335, over 5653922.86 frames. ], batch size: 213, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:04:04,864 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 01:04:13,172 INFO [train.py:1012] (0/2) Epoch 9, validation: loss=0.2212, simple_loss=0.3284, pruned_loss=0.05699, over 944034.00 frames. 
+2023-03-05 01:04:13,173 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-05 01:04:18,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-05 01:04:18,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.481e+02 1.699e+03 2.194e+03 3.609e+03 1.001e+04, threshold=4.388e+03, percent-clipped=19.0
+2023-03-05 01:04:27,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=409216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:04:58,173 INFO [train.py:968] (0/2) Epoch 9, batch 45050, giga_loss[loss=0.3114, simple_loss=0.3774, pruned_loss=0.1227, over 28847.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3833, pruned_loss=0.1346, over 5646175.88 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3745, pruned_loss=0.127, over 5668686.50 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3851, pruned_loss=0.1358, over 5639708.38 frames. ], batch size: 174, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:05:42,158 INFO [train.py:968] (0/2) Epoch 9, batch 45100, giga_loss[loss=0.2946, simple_loss=0.3698, pruned_loss=0.1098, over 28891.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3805, pruned_loss=0.1316, over 5652789.81 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3747, pruned_loss=0.1271, over 5673797.79 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.382, pruned_loss=0.1326, over 5642874.00 frames. ], batch size: 199, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:05:49,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.277e+02 1.569e+03 1.996e+03 2.908e+03 8.014e+03, threshold=3.993e+03, percent-clipped=11.0
+2023-03-05 01:06:19,642 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-05 01:06:28,571 INFO [train.py:968] (0/2) Epoch 9, batch 45150, giga_loss[loss=0.2964, simple_loss=0.3666, pruned_loss=0.1131, over 28644.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3773, pruned_loss=0.1274, over 5661442.30 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3753, pruned_loss=0.1275, over 5676848.21 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.378, pruned_loss=0.1279, over 5650165.25 frames. ], batch size: 78, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:06:39,915 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=409359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:06:42,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=409362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:06:47,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6058, 1.5878, 1.2747, 1.2699], device='cuda:0'), covar=tensor([0.0691, 0.0540, 0.1003, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0443, 0.0500, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 01:06:49,192 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=409369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:07:10,696 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=409391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:07:14,575 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-05 01:07:17,206 INFO [train.py:968] (0/2) Epoch 9, batch 45200, giga_loss[loss=0.2986, simple_loss=0.365, pruned_loss=0.1161, over 28662.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3766, pruned_loss=0.1268, over 5651104.15 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3753, pruned_loss=0.1275, over 5676875.97 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3771, pruned_loss=0.1272, over 5641824.84 frames. ], batch size: 92, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 01:07:25,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.303e+02 1.281e+03 1.576e+03 2.216e+03 8.584e+03, threshold=3.152e+03, percent-clipped=4.0
+2023-03-05 01:08:04,445 INFO [train.py:968] (0/2) Epoch 9, batch 45250, giga_loss[loss=0.3612, simple_loss=0.4158, pruned_loss=0.1533, over 28840.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3758, pruned_loss=0.1265, over 5651060.86 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3757, pruned_loss=0.1276, over 5659713.11 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.376, pruned_loss=0.1267, over 5657703.11 frames. ], batch size: 119, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 01:08:16,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1991, 1.7567, 1.3169, 0.3343], device='cuda:0'), covar=tensor([0.2547, 0.1625, 0.2805, 0.3622], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1451, 0.1468, 0.1240], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 01:08:55,608 INFO [train.py:968] (0/2) Epoch 9, batch 45300, giga_loss[loss=0.3685, simple_loss=0.4043, pruned_loss=0.1664, over 26734.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3743, pruned_loss=0.1267, over 5607887.18 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3768, pruned_loss=0.1287, over 5605932.98 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3733, pruned_loss=0.1259, over 5662584.80 frames. ], batch size: 555, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:09:01,170 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.327e+02 1.757e+03 2.642e+03 4.311e+03 1.062e+04, threshold=5.284e+03, percent-clipped=39.0
+2023-03-05 01:09:26,719 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 01:09:36,242 INFO [train.py:968] (0/2) Epoch 9, batch 45350, libri_loss[loss=0.3579, simple_loss=0.3966, pruned_loss=0.1596, over 19665.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3761, pruned_loss=0.1275, over 5570416.30 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3779, pruned_loss=0.1295, over 5540111.37 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3743, pruned_loss=0.126, over 5676531.92 frames. ], batch size: 186, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:10:04,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1696, 1.8069, 1.3336, 0.2990], device='cuda:0'), covar=tensor([0.2367, 0.1506, 0.2430, 0.2984], device='cuda:0'), in_proj_covar=tensor([0.1531, 0.1468, 0.1482, 0.1253], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 01:10:08,963 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5872, 3.3865, 3.2332, 1.9400], device='cuda:0'), covar=tensor([0.0624, 0.0778, 0.0757, 0.1811], device='cuda:0'), in_proj_covar=tensor([0.1029, 0.0964, 0.0848, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 01:10:20,881 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-05 01:10:23,497 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-9.pt
+2023-03-05 01:11:05,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.298e+02 1.628e+03 2.108e+03 2.932e+03 7.964e+03, threshold=4.216e+03, percent-clipped=7.0
+2023-03-05 01:11:09,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=409608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:11:41,255 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8076, 2.2634, 1.7996, 1.6201], device='cuda:0'), covar=tensor([0.2134, 0.1449, 0.1538, 0.1652], device='cuda:0'), in_proj_covar=tensor([0.1657, 0.1564, 0.1503, 0.1606], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 01:11:43,033 INFO [train.py:968] (0/2) Epoch 10, batch 50, giga_loss[loss=0.3699, simple_loss=0.4179, pruned_loss=0.1609, over 27623.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3896, pruned_loss=0.1216, over 1261866.38 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3688, pruned_loss=0.1031, over 88002.45 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3912, pruned_loss=0.1229, over 1191893.65 frames. ], batch size: 472, lr: 3.33e-03, grad_scale: 2.0
+2023-03-05 01:12:28,865 INFO [train.py:968] (0/2) Epoch 10, batch 100, giga_loss[loss=0.3052, simple_loss=0.38, pruned_loss=0.1151, over 29047.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3765, pruned_loss=0.1145, over 2237651.75 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3548, pruned_loss=0.09963, over 301740.02 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3796, pruned_loss=0.1167, over 2045651.73 frames. ], batch size: 136, lr: 3.33e-03, grad_scale: 2.0
+2023-03-05 01:12:36,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.077e+02 1.088e+03 1.383e+03 1.696e+03 3.321e+03, threshold=2.766e+03, percent-clipped=0.0
+2023-03-05 01:13:12,344 INFO [train.py:968] (0/2) Epoch 10, batch 150, giga_loss[loss=0.2292, simple_loss=0.3098, pruned_loss=0.07427, over 28892.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3618, pruned_loss=0.108, over 3006388.12 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3558, pruned_loss=0.1006, over 466760.50 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3633, pruned_loss=0.1093, over 2765758.52 frames. ], batch size: 145, lr: 3.33e-03, grad_scale: 2.0
+2023-03-05 01:13:13,592 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=409744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:13:52,956 INFO [train.py:968] (0/2) Epoch 10, batch 200, giga_loss[loss=0.2382, simple_loss=0.3143, pruned_loss=0.081, over 28945.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3486, pruned_loss=0.101, over 3614006.65 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3506, pruned_loss=0.09798, over 706762.50 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3495, pruned_loss=0.1021, over 3312917.46 frames. ], batch size: 227, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:14:01,818 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.012e+02 1.040e+03 1.321e+03 1.673e+03 4.065e+03, threshold=2.642e+03, percent-clipped=6.0
+2023-03-05 01:14:32,955 INFO [train.py:968] (0/2) Epoch 10, batch 250, giga_loss[loss=0.2254, simple_loss=0.3009, pruned_loss=0.07494, over 28799.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.337, pruned_loss=0.09493, over 4079909.43 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.352, pruned_loss=0.09827, over 833519.84 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3364, pruned_loss=0.09525, over 3795406.52 frames. ], batch size: 243, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:15:12,021 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=409887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:15:14,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=409890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:15:19,437 INFO [train.py:968] (0/2) Epoch 10, batch 300, giga_loss[loss=0.2195, simple_loss=0.2902, pruned_loss=0.07443, over 28451.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3261, pruned_loss=0.08985, over 4440570.48 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3515, pruned_loss=0.098, over 883614.39 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3249, pruned_loss=0.08978, over 4202816.57 frames. ], batch size: 71, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:15:30,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.355e+02 9.012e+02 1.194e+03 1.642e+03 2.957e+03, threshold=2.388e+03, percent-clipped=4.0
+2023-03-05 01:15:42,337 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=409919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:15:47,292 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5026, 1.6926, 1.4254, 1.7228], device='cuda:0'), covar=tensor([0.2505, 0.2505, 0.2649, 0.2349], device='cuda:0'), in_proj_covar=tensor([0.1265, 0.0940, 0.1121, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 01:16:06,453 INFO [train.py:968] (0/2) Epoch 10, batch 350, giga_loss[loss=0.1869, simple_loss=0.2645, pruned_loss=0.05463, over 28381.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3183, pruned_loss=0.08642, over 4718652.31 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3523, pruned_loss=0.09896, over 954923.82 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3163, pruned_loss=0.08585, over 4516377.53 frames. ], batch size: 78, lr: 3.32e-03, grad_scale: 1.0
+2023-03-05 01:16:38,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=409983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:16:45,211 INFO [train.py:968] (0/2) Epoch 10, batch 400, giga_loss[loss=0.2024, simple_loss=0.2795, pruned_loss=0.06267, over 28968.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.315, pruned_loss=0.08478, over 4937157.78 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3537, pruned_loss=0.09949, over 1115225.94 frames. ], giga_tot_loss[loss=0.2395, simple_loss=0.3117, pruned_loss=0.08363, over 4749876.49 frames. ], batch size: 106, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:16:49,541 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-410000.pt
+2023-03-05 01:16:56,604 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.911e+02 1.033e+03 1.408e+03 2.609e+03 1.377e+04, threshold=2.815e+03, percent-clipped=27.0
+2023-03-05 01:16:59,560 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=410010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:16:59,782 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-05 01:17:25,650 INFO [train.py:968] (0/2) Epoch 10, batch 450, giga_loss[loss=0.2413, simple_loss=0.3046, pruned_loss=0.08902, over 28480.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3122, pruned_loss=0.08324, over 5104843.54 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3543, pruned_loss=0.1002, over 1268076.74 frames. ], giga_tot_loss[loss=0.2356, simple_loss=0.308, pruned_loss=0.08162, over 4936712.82 frames. ], batch size: 78, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:18:09,590 INFO [train.py:968] (0/2) Epoch 10, batch 500, giga_loss[loss=0.2071, simple_loss=0.2897, pruned_loss=0.06224, over 28913.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3103, pruned_loss=0.0823, over 5243629.88 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3555, pruned_loss=0.1003, over 1337264.31 frames. ], giga_tot_loss[loss=0.2338, simple_loss=0.3061, pruned_loss=0.08073, over 5101104.40 frames. ], batch size: 213, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:18:22,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.617e+02 9.373e+02 1.244e+03 1.667e+03 7.706e+03, threshold=2.488e+03, percent-clipped=4.0
+2023-03-05 01:18:29,419 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-05 01:18:37,927 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=410126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:18:42,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=410129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:18:56,122 INFO [train.py:968] (0/2) Epoch 10, batch 550, giga_loss[loss=0.2035, simple_loss=0.2812, pruned_loss=0.0629, over 29060.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3078, pruned_loss=0.08092, over 5343908.95 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3563, pruned_loss=0.1006, over 1426323.08 frames. ], giga_tot_loss[loss=0.2309, simple_loss=0.3034, pruned_loss=0.07923, over 5221352.97 frames. ], batch size: 164, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:19:07,198 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=410158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:19:13,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=410163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:19:30,305 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=410185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:19:37,749 INFO [train.py:968] (0/2) Epoch 10, batch 600, giga_loss[loss=0.2114, simple_loss=0.276, pruned_loss=0.07338, over 28541.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.308, pruned_loss=0.0814, over 5423499.41 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3572, pruned_loss=0.1012, over 1598185.87 frames. ], giga_tot_loss[loss=0.2303, simple_loss=0.3023, pruned_loss=0.07912, over 5309325.34 frames. ], batch size: 85, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:19:50,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.099e+02 1.013e+03 1.211e+03 1.798e+03 5.961e+03, threshold=2.422e+03, percent-clipped=8.0
+2023-03-05 01:20:13,203 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-05 01:20:23,773 INFO [train.py:968] (0/2) Epoch 10, batch 650, giga_loss[loss=0.1896, simple_loss=0.2746, pruned_loss=0.05226, over 29070.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3066, pruned_loss=0.08094, over 5480107.16 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3576, pruned_loss=0.1016, over 1704359.86 frames. ], giga_tot_loss[loss=0.2289, simple_loss=0.3008, pruned_loss=0.07856, over 5379130.39 frames. ], batch size: 155, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:20:48,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9148, 1.0935, 3.5988, 2.9563], device='cuda:0'), covar=tensor([0.1755, 0.2572, 0.0437, 0.0949], device='cuda:0'), in_proj_covar=tensor([0.0628, 0.0569, 0.0823, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 01:21:10,119 INFO [train.py:968] (0/2) Epoch 10, batch 700, giga_loss[loss=0.2198, simple_loss=0.296, pruned_loss=0.07183, over 28238.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3021, pruned_loss=0.07885, over 5525586.15 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3569, pruned_loss=0.1012, over 1746076.28 frames. ], giga_tot_loss[loss=0.2255, simple_loss=0.2973, pruned_loss=0.07686, over 5442794.24 frames. ], batch size: 368, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:21:21,181 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.920e+02 8.926e+02 1.176e+03 1.622e+03 5.682e+03, threshold=2.352e+03, percent-clipped=6.0
+2023-03-05 01:21:55,030 INFO [train.py:968] (0/2) Epoch 10, batch 750, giga_loss[loss=0.1966, simple_loss=0.267, pruned_loss=0.0631, over 28471.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.2992, pruned_loss=0.07717, over 5572867.34 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3566, pruned_loss=0.1007, over 1827365.13 frames. ], giga_tot_loss[loss=0.2225, simple_loss=0.2944, pruned_loss=0.0753, over 5501945.41 frames. ], batch size: 65, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:22:00,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3514, 1.6731, 1.6097, 1.2123], device='cuda:0'), covar=tensor([0.1673, 0.2462, 0.1411, 0.1584], device='cuda:0'), in_proj_covar=tensor([0.0830, 0.0713, 0.0863, 0.0769], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 01:22:34,914 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=410385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:22:41,526 INFO [train.py:968] (0/2) Epoch 10, batch 800, libri_loss[loss=0.3047, simple_loss=0.3818, pruned_loss=0.1138, over 29503.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.297, pruned_loss=0.0762, over 5604761.48 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3574, pruned_loss=0.101, over 1907866.15 frames. ], giga_tot_loss[loss=0.2201, simple_loss=0.2918, pruned_loss=0.07419, over 5543239.87 frames. ], batch size: 85, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:22:52,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.610e+02 1.009e+03 1.272e+03 1.805e+03 5.594e+03, threshold=2.543e+03, percent-clipped=12.0
+2023-03-05 01:23:13,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6186, 1.9641, 1.9441, 1.4695], device='cuda:0'), covar=tensor([0.1672, 0.2155, 0.1332, 0.1510], device='cuda:0'), in_proj_covar=tensor([0.0828, 0.0711, 0.0861, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 01:23:29,628 INFO [train.py:968] (0/2) Epoch 10, batch 850, giga_loss[loss=0.2513, simple_loss=0.3288, pruned_loss=0.0869, over 28411.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3062, pruned_loss=0.08191, over 5608457.85 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3585, pruned_loss=0.1016, over 1977315.43 frames. ], giga_tot_loss[loss=0.2302, simple_loss=0.3008, pruned_loss=0.07978, over 5562929.07 frames. ], batch size: 71, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:23:43,801 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 01:23:45,356 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-05 01:24:06,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3533, 2.7615, 1.5032, 1.4167], device='cuda:0'), covar=tensor([0.0850, 0.0341, 0.0812, 0.1254], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0499, 0.0333, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 01:24:20,049 INFO [train.py:968] (0/2) Epoch 10, batch 900, giga_loss[loss=0.2695, simple_loss=0.3521, pruned_loss=0.09342, over 28390.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3209, pruned_loss=0.08965, over 5625164.39 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3587, pruned_loss=0.1017, over 2014842.02 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3164, pruned_loss=0.08786, over 5588378.90 frames. ], batch size: 71, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:24:34,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.518e+02 1.198e+03 1.462e+03 2.136e+03 6.859e+03, threshold=2.925e+03, percent-clipped=18.0
+2023-03-05 01:24:41,135 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 01:24:50,782 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=410528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:24:53,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=410531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:25:00,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=410538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:25:04,329 INFO [train.py:968] (0/2) Epoch 10, batch 950, giga_loss[loss=0.2915, simple_loss=0.3722, pruned_loss=0.1054, over 28836.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3326, pruned_loss=0.09552, over 5645773.38 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3583, pruned_loss=0.1016, over 2071002.33 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3288, pruned_loss=0.09406, over 5614892.55 frames. ], batch size: 99, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:25:17,948 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=410560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:25:18,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=410560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:25:44,530 INFO [train.py:968] (0/2) Epoch 10, batch 1000, giga_loss[loss=0.2884, simple_loss=0.3711, pruned_loss=0.1029, over 28274.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3395, pruned_loss=0.09818, over 5655814.84 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.357, pruned_loss=0.1012, over 2192915.42 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3362, pruned_loss=0.09699, over 5631348.51 frames. ], batch size: 368, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:25:51,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3331, 1.4945, 1.1737, 1.5786], device='cuda:0'), covar=tensor([0.0752, 0.0309, 0.0328, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:0')
+2023-03-05 01:25:55,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.719e+02 1.109e+03 1.496e+03 1.974e+03 6.191e+03, threshold=2.993e+03, percent-clipped=8.0
+2023-03-05 01:26:27,061 INFO [train.py:968] (0/2) Epoch 10, batch 1050, giga_loss[loss=0.3043, simple_loss=0.3834, pruned_loss=0.1126, over 27585.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.343, pruned_loss=0.09843, over 5664690.05 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3566, pruned_loss=0.1009, over 2248600.06 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3404, pruned_loss=0.09759, over 5642070.29 frames. ], batch size: 472, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:27:01,585 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=410681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:03,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=410684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:12,225 INFO [train.py:968] (0/2) Epoch 10, batch 1100, giga_loss[loss=0.2494, simple_loss=0.3283, pruned_loss=0.08522, over 28434.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3439, pruned_loss=0.09761, over 5670370.15 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3555, pruned_loss=0.1, over 2338075.95 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3419, pruned_loss=0.09717, over 5648063.62 frames. ], batch size: 65, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:27:19,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=410703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:23,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=410706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:23,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.826e+02 1.083e+03 1.333e+03 1.795e+03 5.197e+03, threshold=2.667e+03, percent-clipped=3.0
+2023-03-05 01:27:25,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6855, 2.2837, 1.8956, 1.7985], device='cuda:0'), covar=tensor([0.0714, 0.0245, 0.0259, 0.0766], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0055, 0.0049, 0.0083], device='cuda:0')
+2023-03-05 01:27:28,801 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=410713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:49,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=410735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:57,424 INFO [train.py:968] (0/2) Epoch 10, batch 1150, giga_loss[loss=0.2534, simple_loss=0.3233, pruned_loss=0.0917, over 28521.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3466, pruned_loss=0.0993, over 5684015.75 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3555, pruned_loss=0.09981, over 2387793.58 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3448, pruned_loss=0.09901, over 5666604.83 frames. ], batch size: 71, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:28:41,055 INFO [train.py:968] (0/2) Epoch 10, batch 1200, giga_loss[loss=0.2812, simple_loss=0.3589, pruned_loss=0.1018, over 28907.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3493, pruned_loss=0.1015, over 5668528.74 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3557, pruned_loss=0.09995, over 2545940.74 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3475, pruned_loss=0.1012, over 5656419.83 frames. ], batch size: 186, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:28:52,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.765e+02 1.139e+03 1.420e+03 1.839e+03 6.324e+03, threshold=2.839e+03, percent-clipped=12.0
+2023-03-05 01:29:24,813 INFO [train.py:968] (0/2) Epoch 10, batch 1250, giga_loss[loss=0.3079, simple_loss=0.3763, pruned_loss=0.1197, over 29036.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.352, pruned_loss=0.1031, over 5674039.13 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3564, pruned_loss=0.1004, over 2595154.87 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3502, pruned_loss=0.1027, over 5662255.71 frames. ], batch size: 128, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:30:10,362 INFO [train.py:968] (0/2) Epoch 10, batch 1300, giga_loss[loss=0.2684, simple_loss=0.3506, pruned_loss=0.09309, over 28713.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3553, pruned_loss=0.1044, over 5678558.89 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3561, pruned_loss=0.1002, over 2611735.09 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.354, pruned_loss=0.1042, over 5668272.54 frames. ], batch size: 242, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:30:14,063 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=410899.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 01:30:18,922 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.357e+02 1.139e+03 1.393e+03 1.858e+03 5.502e+03, threshold=2.785e+03, percent-clipped=7.0
+2023-03-05 01:30:50,761 INFO [train.py:968] (0/2) Epoch 10, batch 1350, giga_loss[loss=0.2861, simple_loss=0.3629, pruned_loss=0.1047, over 28772.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3574, pruned_loss=0.1043, over 5693207.03 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3561, pruned_loss=0.1002, over 2659343.63 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3563, pruned_loss=0.1042, over 5683541.66 frames. ], batch size: 119, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:30:55,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9799, 1.9185, 1.4387, 1.5576], device='cuda:0'), covar=tensor([0.0762, 0.0682, 0.1004, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0439, 0.0498, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 01:31:31,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3310, 1.7401, 1.3459, 1.3936], device='cuda:0'), covar=tensor([0.0798, 0.0278, 0.0312, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:0')
+2023-03-05 01:31:31,555 INFO [train.py:968] (0/2) Epoch 10, batch 1400, giga_loss[loss=0.2858, simple_loss=0.3577, pruned_loss=0.1069, over 28779.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3579, pruned_loss=0.1043, over 5697180.02 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3552, pruned_loss=0.1001, over 2787291.29 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3576, pruned_loss=0.1044, over 5681272.66 frames. ], batch size: 112, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:31:43,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.658e+02 1.176e+03 1.441e+03 1.769e+03 4.888e+03, threshold=2.881e+03, percent-clipped=7.0
+2023-03-05 01:32:01,643 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:32:16,217 INFO [train.py:968] (0/2) Epoch 10, batch 1450, giga_loss[loss=0.2632, simple_loss=0.3477, pruned_loss=0.08938, over 29061.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3578, pruned_loss=0.103, over 5702109.83 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3555, pruned_loss=0.1, over 2832697.59 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3574, pruned_loss=0.1032, over 5687611.34 frames. ], batch size: 136, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:32:16,400 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:32:18,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-05 01:32:35,017 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:32:56,074 INFO [train.py:968] (0/2) Epoch 10, batch 1500, giga_loss[loss=0.2729, simple_loss=0.3334, pruned_loss=0.1062, over 23526.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3547, pruned_loss=0.1002, over 5701520.29 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3539, pruned_loss=0.09919, over 2907233.36 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3552, pruned_loss=0.1008, over 5686751.43 frames. ], batch size: 705, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:33:05,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.383e+02 1.088e+03 1.247e+03 1.488e+03 4.406e+03, threshold=2.495e+03, percent-clipped=2.0
+2023-03-05 01:33:36,464 INFO [train.py:968] (0/2) Epoch 10, batch 1550, giga_loss[loss=0.262, simple_loss=0.3409, pruned_loss=0.09156, over 28948.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3531, pruned_loss=0.09889, over 5708810.21 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3532, pruned_loss=0.09856, over 2966627.82 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3538, pruned_loss=0.09967, over 5693576.93 frames. ], batch size: 145, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:33:49,058 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-05 01:34:15,410 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:34:21,287 INFO [train.py:968] (0/2) Epoch 10, batch 1600, giga_loss[loss=0.3131, simple_loss=0.3709, pruned_loss=0.1276, over 28886.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3557, pruned_loss=0.1027, over 5719478.87 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3524, pruned_loss=0.09787, over 3052446.28 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3567, pruned_loss=0.1037, over 5703102.27 frames. ], batch size: 199, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:34:32,050 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.632e+02 1.174e+03 1.573e+03 2.303e+03 6.782e+03, threshold=3.147e+03, percent-clipped=22.0
+2023-03-05 01:35:07,610 INFO [train.py:968] (0/2) Epoch 10, batch 1650, giga_loss[loss=0.2879, simple_loss=0.3508, pruned_loss=0.1125, over 28877.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3588, pruned_loss=0.1078, over 5711572.01 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3519, pruned_loss=0.09762, over 3109453.72 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.36, pruned_loss=0.1089, over 5695061.01 frames. ], batch size: 106, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:35:35,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411274.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 01:35:52,300 INFO [train.py:968] (0/2) Epoch 10, batch 1700, giga_loss[loss=0.3125, simple_loss=0.3716, pruned_loss=0.1267, over 28693.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3601, pruned_loss=0.11, over 5702329.93 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3523, pruned_loss=0.09787, over 3164152.67 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.361, pruned_loss=0.111, over 5686161.37 frames. ], batch size: 262, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:36:06,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.604e+02 1.229e+03 1.638e+03 2.616e+03 7.678e+03, threshold=3.275e+03, percent-clipped=15.0
+2023-03-05 01:36:39,620 INFO [train.py:968] (0/2) Epoch 10, batch 1750, giga_loss[loss=0.2539, simple_loss=0.3321, pruned_loss=0.0878, over 28988.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3583, pruned_loss=0.1094, over 5709152.04 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3528, pruned_loss=0.0982, over 3190869.49 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3588, pruned_loss=0.1101, over 5695226.03 frames. ], batch size: 136, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:36:54,868 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:37:21,557 INFO [train.py:968] (0/2) Epoch 10, batch 1800, giga_loss[loss=0.256, simple_loss=0.3386, pruned_loss=0.08671, over 28635.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3576, pruned_loss=0.1091, over 5705503.34 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3533, pruned_loss=0.09834, over 3238021.62 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3578, pruned_loss=0.1098, over 5697288.61 frames. ], batch size: 307, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:37:27,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:37:29,159 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:37:32,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.123e+02 1.095e+03 1.325e+03 1.831e+03 4.903e+03, threshold=2.650e+03, percent-clipped=4.0
+2023-03-05 01:37:39,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411417.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 01:37:40,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:37:41,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411420.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 01:38:01,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:38:01,873 INFO [train.py:968] (0/2) Epoch 10, batch 1850, libri_loss[loss=0.2283, simple_loss=0.3107, pruned_loss=0.07296, over 29572.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3568, pruned_loss=0.1073, over 5717102.61 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3529, pruned_loss=0.09794, over 3316816.55 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3573, pruned_loss=0.1084, over 5705546.10 frames. ], batch size: 76, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:38:06,261 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411449.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 01:38:46,388 INFO [train.py:968] (0/2) Epoch 10, batch 1900, giga_loss[loss=0.2624, simple_loss=0.3416, pruned_loss=0.09159, over 29069.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3543, pruned_loss=0.1053, over 5707864.29 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3527, pruned_loss=0.09752, over 3411982.02 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3549, pruned_loss=0.1067, over 5697981.62 frames. ], batch size: 128, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:39:04,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.726e+02 1.017e+03 1.469e+03 2.177e+03 1.129e+04, threshold=2.938e+03, percent-clipped=15.0
+2023-03-05 01:39:32,614 INFO [train.py:968] (0/2) Epoch 10, batch 1950, giga_loss[loss=0.2852, simple_loss=0.3509, pruned_loss=0.1098, over 28459.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3508, pruned_loss=0.1032, over 5700702.36 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3527, pruned_loss=0.09756, over 3473287.80 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3513, pruned_loss=0.1044, over 5688551.86 frames. ], batch size: 85, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:39:34,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:39:37,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:39:52,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:39:52,458 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:39:55,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:06,007 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:06,022 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:15,730 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:17,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:21,982 INFO [train.py:968] (0/2) Epoch 10, batch 2000, giga_loss[loss=0.2241, simple_loss=0.3035, pruned_loss=0.0724, over 28864.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3452, pruned_loss=0.1005, over 5674123.81 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3537, pruned_loss=0.09837, over 3501010.44 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.345, pruned_loss=0.1011, over 5678993.75 frames. ], batch size: 112, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:40:22,174 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:35,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.199e+02 9.518e+02 1.252e+03 1.949e+03 1.161e+04, threshold=2.504e+03, percent-clipped=14.0
+2023-03-05 01:40:43,294 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:41:01,527 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 01:41:07,380 INFO [train.py:968] (0/2) Epoch 10, batch 2050, giga_loss[loss=0.2266, simple_loss=0.3044, pruned_loss=0.07439, over 28606.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3394, pruned_loss=0.0975, over 5671478.35 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3538, pruned_loss=0.09808, over 3559595.74 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3389, pruned_loss=0.09815, over 5671025.13 frames. ], batch size: 85, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:41:49,756 INFO [train.py:968] (0/2) Epoch 10, batch 2100, giga_loss[loss=0.2731, simple_loss=0.3424, pruned_loss=0.1019, over 28443.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3405, pruned_loss=0.09769, over 5676274.96 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.355, pruned_loss=0.09883, over 3660322.36 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3387, pruned_loss=0.09773, over 5678996.18 frames. ], batch size: 85, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:41:52,082 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-05 01:41:58,197 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:42:00,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:42:01,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.852e+02 1.044e+03 1.252e+03 1.616e+03 5.770e+03, threshold=2.504e+03, percent-clipped=3.0
+2023-03-05 01:42:22,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-05 01:42:24,191 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:42:25,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:42:28,939 INFO [train.py:968] (0/2) Epoch 10, batch 2150, giga_loss[loss=0.2893, simple_loss=0.3613, pruned_loss=0.1086, over 28257.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3419, pruned_loss=0.09837, over 5688875.92 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.355, pruned_loss=0.09871, over 3716290.81 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3402, pruned_loss=0.09846, over 5685666.55 frames. ], batch size: 368, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:42:49,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5433, 2.3409, 1.5221, 0.7670], device='cuda:0'), covar=tensor([0.5456, 0.2546, 0.2923, 0.4706], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1449, 0.1471, 0.1243], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 01:42:52,805 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-05 01:42:58,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:43:12,804 INFO [train.py:968] (0/2) Epoch 10, batch 2200, giga_loss[loss=0.277, simple_loss=0.3488, pruned_loss=0.1026, over 28622.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3403, pruned_loss=0.09738, over 5692354.85 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3554, pruned_loss=0.09892, over 3737965.92 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3385, pruned_loss=0.09731, over 5688079.62 frames. ], batch size: 78, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:43:27,402 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.447e+02 9.937e+02 1.299e+03 1.809e+03 3.802e+03, threshold=2.597e+03, percent-clipped=7.0
+2023-03-05 01:43:52,429 INFO [train.py:968] (0/2) Epoch 10, batch 2250, giga_loss[loss=0.2159, simple_loss=0.2911, pruned_loss=0.07037, over 28635.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3369, pruned_loss=0.09554, over 5700328.63 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3556, pruned_loss=0.09882, over 3776836.34 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3351, pruned_loss=0.09551, over 5697000.78 frames. ], batch size: 71, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:44:11,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5907, 1.6796, 1.2898, 1.2772], device='cuda:0'), covar=tensor([0.0778, 0.0580, 0.0985, 0.0999], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0444, 0.0502, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 01:44:26,563 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:44:28,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:44:35,442 INFO [train.py:968] (0/2) Epoch 10, batch 2300, giga_loss[loss=0.2372, simple_loss=0.3051, pruned_loss=0.08464, over 28705.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3352, pruned_loss=0.09492, over 5704185.33 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3557, pruned_loss=0.09871, over 3829097.66 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3332, pruned_loss=0.09487, over 5697524.52 frames. ], batch size: 60, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:44:49,427 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.647e+02 9.224e+02 1.168e+03 1.632e+03 9.993e+03, threshold=2.336e+03, percent-clipped=5.0
+2023-03-05 01:44:52,485 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:45:00,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:45:03,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:45:08,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4315, 1.6241, 1.4698, 1.1775], device='cuda:0'), covar=tensor([0.2100, 0.1626, 0.1169, 0.1719], device='cuda:0'), in_proj_covar=tensor([0.1623, 0.1505, 0.1469, 0.1589], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 01:45:17,912 INFO [train.py:968] (0/2) Epoch 10, batch 2350, giga_loss[loss=0.249, simple_loss=0.3243, pruned_loss=0.08687, over 29016.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3327, pruned_loss=0.09379, over 5712199.60 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3564, pruned_loss=0.09897, over 3857813.48 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3304, pruned_loss=0.09352, over 5706250.51 frames. ], batch size: 155, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:45:24,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:45:26,941 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:45:28,353 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 01:45:59,453 INFO [train.py:968] (0/2) Epoch 10, batch 2400, giga_loss[loss=0.2251, simple_loss=0.2926, pruned_loss=0.07882, over 28921.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3286, pruned_loss=0.09163, over 5713443.16 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3568, pruned_loss=0.09905, over 3868766.44 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3264, pruned_loss=0.09134, over 5714686.52 frames. ], batch size: 106, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:46:04,364 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-412000.pt
+2023-03-05 01:46:13,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.630e+02 9.166e+02 1.099e+03 1.549e+03 6.023e+03, threshold=2.198e+03, percent-clipped=8.0
+2023-03-05 01:46:37,270 INFO [train.py:968] (0/2) Epoch 10, batch 2450, giga_loss[loss=0.2243, simple_loss=0.2973, pruned_loss=0.07562, over 29048.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3265, pruned_loss=0.09049, over 5723363.26 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3573, pruned_loss=0.09901, over 3899318.04 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3241, pruned_loss=0.09018, over 5721728.07 frames. ], batch size: 128, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:47:17,721 INFO [train.py:968] (0/2) Epoch 10, batch 2500, libri_loss[loss=0.3337, simple_loss=0.419, pruned_loss=0.1241, over 29289.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3258, pruned_loss=0.09012, over 5719997.65 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3585, pruned_loss=0.09939, over 3938846.27 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3225, pruned_loss=0.08947, over 5715671.45 frames. ], batch size: 94, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:47:18,632 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=412095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:47:21,405 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=412098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:47:29,915 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=412107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:47:32,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.807e+02 9.886e+02 1.285e+03 1.866e+03 7.759e+03, threshold=2.571e+03, percent-clipped=18.0
+2023-03-05 01:47:44,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=412127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:47:57,757 INFO [train.py:968] (0/2) Epoch 10, batch 2550, giga_loss[loss=0.2205, simple_loss=0.2985, pruned_loss=0.07125, over 28788.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3247, pruned_loss=0.08918, over 5722359.68 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3595, pruned_loss=0.09966, over 3987554.74 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3206, pruned_loss=0.08827, over 5714676.96 frames. ], batch size: 242, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:48:36,389 INFO [train.py:968] (0/2) Epoch 10, batch 2600, libri_loss[loss=0.2986, simple_loss=0.3866, pruned_loss=0.1053, over 29113.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.325, pruned_loss=0.08886, over 5718052.32 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.362, pruned_loss=0.1009, over 4048253.24 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3188, pruned_loss=0.08686, over 5719008.31 frames. ], batch size: 101, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:48:38,667 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=412197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:48:49,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.523e+02 9.540e+02 1.209e+03 1.619e+03 4.394e+03, threshold=2.418e+03, percent-clipped=6.0
+2023-03-05 01:49:07,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4389, 1.6470, 1.3110, 1.4637], device='cuda:0'), covar=tensor([0.2251, 0.2264, 0.2505, 0.2138], device='cuda:0'), in_proj_covar=tensor([0.1267, 0.0940, 0.1122, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 01:49:15,096 INFO [train.py:968] (0/2) Epoch 10, batch 2650, giga_loss[loss=0.2333, simple_loss=0.3106, pruned_loss=0.07795, over 28971.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3249, pruned_loss=0.08877, over 5719644.47 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3618, pruned_loss=0.1007, over 4118737.75 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3186, pruned_loss=0.08685, over 5716057.68 frames. ], batch size: 136, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:50:00,931 INFO [train.py:968] (0/2) Epoch 10, batch 2700, giga_loss[loss=0.2697, simple_loss=0.3446, pruned_loss=0.09741, over 28850.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3279, pruned_loss=0.09113, over 5709387.86 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3619, pruned_loss=0.1007, over 4127759.07 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3227, pruned_loss=0.08953, over 5711907.78 frames. ], batch size: 174, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:50:01,731 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=412295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:50:05,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-05 01:50:13,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.141e+02 9.962e+02 1.255e+03 1.774e+03 4.400e+03, threshold=2.510e+03, percent-clipped=9.0
+2023-03-05 01:50:16,319 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3007, 1.2974, 1.1369, 1.4893], device='cuda:0'), covar=tensor([0.0752, 0.0333, 0.0328, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:0')
+2023-03-05 01:50:43,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3649, 1.6483, 1.6341, 1.3046], device='cuda:0'), covar=tensor([0.1129, 0.1542, 0.0916, 0.1106], device='cuda:0'), in_proj_covar=tensor([0.0823, 0.0704, 0.0856, 0.0761], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 01:50:45,985 INFO [train.py:968] (0/2) Epoch 10, batch 2750, giga_loss[loss=0.296, simple_loss=0.3648, pruned_loss=0.1136, over 28698.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3336, pruned_loss=0.09512, over 5701490.48 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3612, pruned_loss=0.1003, over 4161684.61 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3293, pruned_loss=0.09394, over 5701159.62 frames. ], batch size: 284, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:51:31,613 INFO [train.py:968] (0/2) Epoch 10, batch 2800, giga_loss[loss=0.3477, simple_loss=0.3907, pruned_loss=0.1523, over 28687.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3419, pruned_loss=0.1008, over 5700022.21 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3613, pruned_loss=0.1004, over 4212334.56 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3377, pruned_loss=0.09977, over 5695054.38 frames. ], batch size: 242, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:51:47,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.395e+02 1.283e+03 1.789e+03 2.320e+03 4.211e+03, threshold=3.578e+03, percent-clipped=21.0
+2023-03-05 01:52:13,202 INFO [train.py:968] (0/2) Epoch 10, batch 2850, giga_loss[loss=0.2995, simple_loss=0.3795, pruned_loss=0.1097, over 28717.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3484, pruned_loss=0.1044, over 5693277.70 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3608, pruned_loss=0.1002, over 4269465.45 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3449, pruned_loss=0.1038, over 5683628.25 frames. ], batch size: 242, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:52:23,071 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=412454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:52:48,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=412482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:52:59,910 INFO [train.py:968] (0/2) Epoch 10, batch 2900, giga_loss[loss=0.2882, simple_loss=0.365, pruned_loss=0.1057, over 28483.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3539, pruned_loss=0.1072, over 5677907.58 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3606, pruned_loss=0.1003, over 4324530.24 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3508, pruned_loss=0.1068, over 5664804.57 frames. ], batch size: 65, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:53:12,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.358e+02 1.217e+03 1.513e+03 2.019e+03 6.693e+03, threshold=3.027e+03, percent-clipped=1.0
+2023-03-05 01:53:38,475 INFO [train.py:968] (0/2) Epoch 10, batch 2950, giga_loss[loss=0.2848, simple_loss=0.3622, pruned_loss=0.1037, over 29067.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3569, pruned_loss=0.1078, over 5678279.79 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3595, pruned_loss=0.09994, over 4388769.15 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3549, pruned_loss=0.108, over 5674972.23 frames. ], batch size: 128, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:54:09,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=412572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:54:28,021 INFO [train.py:968] (0/2) Epoch 10, batch 3000, giga_loss[loss=0.327, simple_loss=0.3866, pruned_loss=0.1337, over 27908.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3629, pruned_loss=0.1119, over 5677161.21 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3594, pruned_loss=0.09991, over 4418819.28 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3614, pruned_loss=0.1123, over 5670646.85 frames. ], batch size: 412, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:54:28,026 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 01:54:36,632 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2271, simple_loss=0.3302, pruned_loss=0.06204, over 944034.00 frames. 
+2023-03-05 01:54:36,633 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-05 01:54:50,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.882e+02 1.156e+03 1.533e+03 2.204e+03 7.684e+03, threshold=3.066e+03, percent-clipped=11.0
+2023-03-05 01:55:01,825 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=412625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:55:03,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=412628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:55:11,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9342, 1.9968, 1.4918, 1.5312], device='cuda:0'), covar=tensor([0.0726, 0.0496, 0.0898, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0440, 0.0501, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 01:55:17,811 INFO [train.py:968] (0/2) Epoch 10, batch 3050, giga_loss[loss=0.2678, simple_loss=0.341, pruned_loss=0.09732, over 28897.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3576, pruned_loss=0.1081, over 5682729.35 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3587, pruned_loss=0.09959, over 4466254.77 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3569, pruned_loss=0.109, over 5673977.82 frames. ], batch size: 106, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:55:28,668 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=412657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:55:41,083 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=412670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:55:44,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3497, 1.6187, 1.2803, 1.3392], device='cuda:0'), covar=tensor([0.2318, 0.2176, 0.2347, 0.2025], device='cuda:0'), in_proj_covar=tensor([0.1261, 0.0934, 0.1114, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 01:56:00,440 INFO [train.py:968] (0/2) Epoch 10, batch 3100, giga_loss[loss=0.3146, simple_loss=0.3713, pruned_loss=0.129, over 27992.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3528, pruned_loss=0.1044, over 5680946.16 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3585, pruned_loss=0.0997, over 4491868.50 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3524, pruned_loss=0.1051, over 5677473.11 frames. ], batch size: 412, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:56:17,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.305e+02 1.094e+03 1.471e+03 2.358e+03 9.089e+03, threshold=2.942e+03, percent-clipped=9.0
+2023-03-05 01:56:19,639 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=412715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:56:21,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=412718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:56:42,607 INFO [train.py:968] (0/2) Epoch 10, batch 3150, giga_loss[loss=0.2612, simple_loss=0.3376, pruned_loss=0.09245, over 29019.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3526, pruned_loss=0.1036, over 5676315.97 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3591, pruned_loss=0.1002, over 4524677.73 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3516, pruned_loss=0.104, over 5670298.41 frames. ], batch size: 106, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:56:45,412 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=412747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:57:19,405 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=412783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:57:26,844 INFO [train.py:968] (0/2) Epoch 10, batch 3200, giga_loss[loss=0.293, simple_loss=0.3603, pruned_loss=0.1128, over 28629.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3533, pruned_loss=0.1039, over 5668049.64 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3591, pruned_loss=0.1004, over 4545975.07 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3524, pruned_loss=0.1041, over 5669077.72 frames. ], batch size: 92, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 01:57:44,185 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.907e+02 1.188e+03 1.680e+03 2.160e+03 6.294e+03, threshold=3.360e+03, percent-clipped=12.0
+2023-03-05 01:57:45,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=412813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:57:47,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=412816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:57:58,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=412829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:58:11,233 INFO [train.py:968] (0/2) Epoch 10, batch 3250, giga_loss[loss=0.3018, simple_loss=0.3764, pruned_loss=0.1135, over 28808.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3561, pruned_loss=0.1058, over 5670383.38 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3589, pruned_loss=0.1003, over 4554964.03 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3555, pruned_loss=0.106, over 5673069.19 frames. ], batch size: 92, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 01:58:12,140 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=412845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:58:20,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2353, 3.3153, 1.3591, 1.4229], device='cuda:0'), covar=tensor([0.1199, 0.0399, 0.0989, 0.1592], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0494, 0.0328, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0028, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 01:58:55,565 INFO [train.py:968] (0/2) Epoch 10, batch 3300, giga_loss[loss=0.3022, simple_loss=0.3587, pruned_loss=0.1229, over 23785.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3594, pruned_loss=0.1083, over 5680208.24 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3589, pruned_loss=0.1002, over 4580781.72 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3589, pruned_loss=0.1087, over 5679291.32 frames. ], batch size: 705, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 01:58:55,980 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 01:59:11,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.804e+02 1.191e+03 1.560e+03 2.214e+03 5.262e+03, threshold=3.121e+03, percent-clipped=6.0
+2023-03-05 01:59:36,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1520, 2.5661, 1.1940, 1.2106], device='cuda:0'), covar=tensor([0.0960, 0.0321, 0.0851, 0.1404], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0493, 0.0329, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0028, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 01:59:37,506 INFO [train.py:968] (0/2) Epoch 10, batch 3350, giga_loss[loss=0.3099, simple_loss=0.3771, pruned_loss=0.1213, over 28946.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3605, pruned_loss=0.1091, over 5688558.10 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.359, pruned_loss=0.1001, over 4619872.88 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3601, pruned_loss=0.1098, over 5682038.16 frames. ], batch size: 145, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:00:00,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=412972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:00:03,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=412975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:00:20,066 INFO [train.py:968] (0/2) Epoch 10, batch 3400, libri_loss[loss=0.291, simple_loss=0.3697, pruned_loss=0.1061, over 29119.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.361, pruned_loss=0.11, over 5683588.59 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3584, pruned_loss=0.09964, over 4667469.46 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3612, pruned_loss=0.1112, over 5677506.00 frames. ], batch size: 101, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:00:28,528 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=413004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:00:34,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.558e+02 1.162e+03 1.449e+03 2.159e+03 5.296e+03, threshold=2.899e+03, percent-clipped=8.0
+2023-03-05 02:01:02,865 INFO [train.py:968] (0/2) Epoch 10, batch 3450, giga_loss[loss=0.2642, simple_loss=0.3461, pruned_loss=0.09118, over 29040.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3609, pruned_loss=0.1099, over 5681975.74 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3584, pruned_loss=0.09946, over 4706837.06 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3611, pruned_loss=0.1114, over 5672964.41 frames. ], batch size: 164, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:01:43,229 INFO [train.py:968] (0/2) Epoch 10, batch 3500, giga_loss[loss=0.3121, simple_loss=0.358, pruned_loss=0.1331, over 23632.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3607, pruned_loss=0.1089, over 5692513.86 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3576, pruned_loss=0.09905, over 4753181.64 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3615, pruned_loss=0.1107, over 5678323.57 frames. ], batch size: 705, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:01:58,728 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.521e+02 1.194e+03 1.369e+03 1.934e+03 4.840e+03, threshold=2.737e+03, percent-clipped=8.0
+2023-03-05 02:02:24,240 INFO [train.py:968] (0/2) Epoch 10, batch 3550, giga_loss[loss=0.2575, simple_loss=0.3152, pruned_loss=0.09991, over 23350.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.361, pruned_loss=0.1078, over 5697888.77 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3576, pruned_loss=0.09892, over 4781839.30 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3618, pruned_loss=0.1096, over 5681623.15 frames. ], batch size: 705, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:02:35,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2183, 0.7819, 0.7860, 1.3538], device='cuda:0'), covar=tensor([0.0765, 0.0373, 0.0358, 0.0785], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:0')
+2023-03-05 02:02:36,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=413158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:03:06,120 INFO [train.py:968] (0/2) Epoch 10, batch 3600, giga_loss[loss=0.2783, simple_loss=0.3546, pruned_loss=0.101, over 28925.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3608, pruned_loss=0.1071, over 5701959.00 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3568, pruned_loss=0.09845, over 4818857.97 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3621, pruned_loss=0.1092, over 5683893.82 frames. ], batch size: 186, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:03:19,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.372e+02 1.044e+03 1.340e+03 1.860e+03 4.044e+03, threshold=2.680e+03, percent-clipped=8.0
+2023-03-05 02:03:25,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1746, 1.3494, 1.2093, 0.9792], device='cuda:0'), covar=tensor([0.1590, 0.1707, 0.1069, 0.1484], device='cuda:0'), in_proj_covar=tensor([0.1624, 0.1520, 0.1477, 0.1591], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 02:03:43,455 INFO [train.py:968] (0/2) Epoch 10, batch 3650, giga_loss[loss=0.2963, simple_loss=0.3649, pruned_loss=0.1139, over 28680.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3589, pruned_loss=0.106, over 5709755.52 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3571, pruned_loss=0.09863, over 4841950.79 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3598, pruned_loss=0.1077, over 5694207.66 frames. ], batch size: 60, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:04:12,294 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 02:04:14,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5405, 1.7855, 1.8540, 1.4089], device='cuda:0'), covar=tensor([0.1594, 0.2200, 0.1229, 0.1461], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0700, 0.0847, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 02:04:21,152 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-05 02:04:26,920 INFO [train.py:968] (0/2) Epoch 10, batch 3700, giga_loss[loss=0.257, simple_loss=0.3347, pruned_loss=0.08963, over 28834.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3566, pruned_loss=0.1053, over 5695575.87 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3577, pruned_loss=0.09905, over 4857603.90 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.357, pruned_loss=0.1065, over 5688014.28 frames. ], batch size: 174, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:04:34,709 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=413301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:04:36,892 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=413304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:04:42,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.965e+02 1.066e+03 1.223e+03 1.542e+03 4.114e+03, threshold=2.446e+03, percent-clipped=5.0
+2023-03-05 02:04:58,390 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=413333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:05:06,675 INFO [train.py:968] (0/2) Epoch 10, batch 3750, giga_loss[loss=0.2711, simple_loss=0.3485, pruned_loss=0.09691, over 29044.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3542, pruned_loss=0.1039, over 5702974.49 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3578, pruned_loss=0.09898, over 4867890.04 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3544, pruned_loss=0.1049, over 5695321.00 frames. ], batch size: 164, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:05:49,136 INFO [train.py:968] (0/2) Epoch 10, batch 3800, giga_loss[loss=0.3117, simple_loss=0.377, pruned_loss=0.1233, over 28781.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3552, pruned_loss=0.1048, over 5703093.88 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3576, pruned_loss=0.09886, over 4903323.00 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3553, pruned_loss=0.1059, over 5691237.50 frames. ], batch size: 284, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:05:51,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3915, 1.4214, 1.1874, 1.5521], device='cuda:0'), covar=tensor([0.0752, 0.0332, 0.0319, 0.0779], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:0')
+2023-03-05 02:06:02,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.661e+02 9.983e+02 1.311e+03 1.765e+03 5.939e+03, threshold=2.621e+03, percent-clipped=13.0
+2023-03-05 02:06:27,769 INFO [train.py:968] (0/2) Epoch 10, batch 3850, giga_loss[loss=0.2731, simple_loss=0.353, pruned_loss=0.09656, over 28948.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3559, pruned_loss=0.1054, over 5699243.47 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3577, pruned_loss=0.09931, over 4924645.07 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3558, pruned_loss=0.1061, over 5694835.53 frames. ], batch size: 174, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:07:06,198 INFO [train.py:968] (0/2) Epoch 10, batch 3900, giga_loss[loss=0.2733, simple_loss=0.3614, pruned_loss=0.09258, over 28599.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3559, pruned_loss=0.1045, over 5703722.04 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3576, pruned_loss=0.09915, over 4944510.95 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3559, pruned_loss=0.1053, over 5699868.31 frames. ], batch size: 307, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:07:23,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.800e+02 9.492e+02 1.106e+03 1.463e+03 2.894e+03, threshold=2.212e+03, percent-clipped=4.0
+2023-03-05 02:07:50,195 INFO [train.py:968] (0/2) Epoch 10, batch 3950, libri_loss[loss=0.2933, simple_loss=0.3718, pruned_loss=0.1074, over 29393.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3548, pruned_loss=0.1034, over 5711824.86 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3574, pruned_loss=0.09911, over 4968201.73 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3549, pruned_loss=0.1042, over 5704309.80 frames. ], batch size: 92, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:08:11,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2730, 3.1997, 1.3682, 1.3739], device='cuda:0'), covar=tensor([0.0932, 0.0254, 0.0875, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0491, 0.0329, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0028, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 02:08:28,874 INFO [train.py:968] (0/2) Epoch 10, batch 4000, giga_loss[loss=0.2466, simple_loss=0.3284, pruned_loss=0.08242, over 28514.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3547, pruned_loss=0.1035, over 5709026.19 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3573, pruned_loss=0.099, over 4995987.00 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3548, pruned_loss=0.1044, over 5698261.67 frames. ], batch size: 85, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:08:42,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3530, 3.6086, 1.5108, 1.4256], device='cuda:0'), covar=tensor([0.0933, 0.0220, 0.0849, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0491, 0.0329, 0.0382], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0028, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 02:08:43,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.507e+02 1.012e+03 1.207e+03 1.580e+03 5.701e+03, threshold=2.414e+03, percent-clipped=11.0
+2023-03-05 02:09:05,158 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 02:09:09,558 INFO [train.py:968] (0/2) Epoch 10, batch 4050, giga_loss[loss=0.2408, simple_loss=0.3229, pruned_loss=0.07936, over 28988.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.353, pruned_loss=0.103, over 5715986.34 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3573, pruned_loss=0.09906, over 5009840.90 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.353, pruned_loss=0.1037, over 5704745.92 frames. ], batch size: 155, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:09:49,462 INFO [train.py:968] (0/2) Epoch 10, batch 4100, giga_loss[loss=0.3225, simple_loss=0.3729, pruned_loss=0.136, over 26796.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3504, pruned_loss=0.1021, over 5712792.13 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.357, pruned_loss=0.099, over 5019552.25 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3506, pruned_loss=0.1027, over 5704676.43 frames. ], batch size: 555, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:10:03,650 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.699e+02 1.012e+03 1.207e+03 1.655e+03 4.913e+03, threshold=2.414e+03, percent-clipped=12.0
+2023-03-05 02:10:28,267 INFO [train.py:968] (0/2) Epoch 10, batch 4150, giga_loss[loss=0.2672, simple_loss=0.3502, pruned_loss=0.09211, over 28615.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.348, pruned_loss=0.1005, over 5717158.86 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3572, pruned_loss=0.09905, over 5048219.73 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3477, pruned_loss=0.1011, over 5706606.75 frames. ], batch size: 336, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:10:29,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3070, 3.0855, 2.9055, 1.3979], device='cuda:0'), covar=tensor([0.0844, 0.0987, 0.0931, 0.2300], device='cuda:0'), in_proj_covar=tensor([0.0994, 0.0923, 0.0813, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 02:10:42,505 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6503, 1.7474, 1.4445, 1.7228], device='cuda:0'), covar=tensor([0.2458, 0.2411, 0.2690, 0.2257], device='cuda:0'), in_proj_covar=tensor([0.1265, 0.0941, 0.1118, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 02:11:08,662 INFO [train.py:968] (0/2) Epoch 10, batch 4200, giga_loss[loss=0.2746, simple_loss=0.3436, pruned_loss=0.1028, over 28197.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3477, pruned_loss=0.1005, over 5715047.85 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3572, pruned_loss=0.09902, over 5054605.71 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3474, pruned_loss=0.101, over 5706921.42 frames. ], batch size: 77, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:11:11,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4258, 4.2117, 3.9864, 1.8862], device='cuda:0'), covar=tensor([0.0497, 0.0663, 0.0684, 0.2037], device='cuda:0'), in_proj_covar=tensor([0.0992, 0.0922, 0.0810, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 02:11:23,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.669e+02 9.963e+02 1.330e+03 1.887e+03 7.322e+03, threshold=2.659e+03, percent-clipped=15.0
+2023-03-05 02:11:49,542 INFO [train.py:968] (0/2) Epoch 10, batch 4250, giga_loss[loss=0.2233, simple_loss=0.3091, pruned_loss=0.06874, over 29069.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.346, pruned_loss=0.1002, over 5717116.61 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3569, pruned_loss=0.09889, over 5067675.52 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3458, pruned_loss=0.1007, over 5707905.05 frames. ], batch size: 155, lr: 3.31e-03, grad_scale: 2.0
+2023-03-05 02:12:19,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4666, 4.2519, 4.0620, 1.9273], device='cuda:0'), covar=tensor([0.0461, 0.0609, 0.0639, 0.1985], device='cuda:0'), in_proj_covar=tensor([0.0994, 0.0926, 0.0813, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 02:12:33,338 INFO [train.py:968] (0/2) Epoch 10, batch 4300, giga_loss[loss=0.2597, simple_loss=0.3329, pruned_loss=0.09325, over 28933.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3446, pruned_loss=0.1002, over 5712926.80 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3572, pruned_loss=0.09913, over 5078381.11 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.344, pruned_loss=0.1004, over 5704762.03 frames. ], batch size: 128, lr: 3.31e-03, grad_scale: 2.0
+2023-03-05 02:12:49,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.704e+02 1.045e+03 1.410e+03 2.012e+03 7.190e+03, threshold=2.821e+03, percent-clipped=13.0
+2023-03-05 02:13:09,180 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1851, 1.5993, 1.5456, 1.0862], device='cuda:0'), covar=tensor([0.1439, 0.2236, 0.1241, 0.1451], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0698, 0.0845, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 02:13:11,832 INFO [train.py:968] (0/2) Epoch 10, batch 4350, giga_loss[loss=0.2466, simple_loss=0.3201, pruned_loss=0.0865, over 28816.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.344, pruned_loss=0.1004, over 5700459.79 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3577, pruned_loss=0.09953, over 5086203.90 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3428, pruned_loss=0.1003, over 5706619.72 frames. ], batch size: 199, lr: 3.31e-03, grad_scale: 2.0
+2023-03-05 02:13:25,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5037, 1.8507, 1.4965, 1.6403], device='cuda:0'), covar=tensor([0.0717, 0.0273, 0.0305, 0.0788], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:0')
+2023-03-05 02:13:48,977 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=413989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:13:51,815 INFO [train.py:968] (0/2) Epoch 10, batch 4400, giga_loss[loss=0.2756, simple_loss=0.3475, pruned_loss=0.1019, over 28943.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.343, pruned_loss=0.1004, over 5699894.08 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3578, pruned_loss=0.09943, over 5102342.35 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3417, pruned_loss=0.1003, over 5701219.59 frames. ], batch size: 227, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:13:56,475 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-414000.pt
+2023-03-05 02:14:03,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9982, 1.9282, 1.4796, 1.6168], device='cuda:0'), covar=tensor([0.0649, 0.0590, 0.0911, 0.0989], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0436, 0.0499, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 02:14:05,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5161, 1.6434, 1.5597, 1.4611], device='cuda:0'), covar=tensor([0.2039, 0.1678, 0.1269, 0.1480], device='cuda:0'), in_proj_covar=tensor([0.1619, 0.1519, 0.1496, 0.1595], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 02:14:08,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.041e+02 9.491e+02 1.320e+03 1.809e+03 7.967e+03, threshold=2.640e+03, percent-clipped=9.0
+2023-03-05 02:14:18,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8766, 1.9222, 1.3752, 1.5647], device='cuda:0'), covar=tensor([0.0707, 0.0640, 0.1025, 0.1078], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0438, 0.0501, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 02:14:32,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7168, 1.4572, 5.0671, 3.6159], device='cuda:0'), covar=tensor([0.1459, 0.2431, 0.0280, 0.0685], device='cuda:0'), in_proj_covar=tensor([0.0632, 0.0571, 0.0820, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 02:14:34,372 INFO [train.py:968] (0/2) Epoch 10, batch 4450, giga_loss[loss=0.3056, simple_loss=0.3808, pruned_loss=0.1153, over 29056.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.343, pruned_loss=0.09986, over 5703089.41 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3579, pruned_loss=0.0996, over 5110059.40 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3417, pruned_loss=0.09972, over 5702687.91 frames. ], batch size: 164, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:14:52,493 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-05 02:15:20,036 INFO [train.py:968] (0/2) Epoch 10, batch 4500, giga_loss[loss=0.3002, simple_loss=0.3721, pruned_loss=0.1141, over 27926.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3454, pruned_loss=0.101, over 5707555.07 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3578, pruned_loss=0.09957, over 5121603.40 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3443, pruned_loss=0.1009, over 5704778.97 frames. ], batch size: 412, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:15:22,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 02:15:23,852 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:15:38,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.829e+02 9.621e+02 1.196e+03 1.586e+03 6.622e+03, threshold=2.393e+03, percent-clipped=4.0
+2023-03-05 02:16:02,990 INFO [train.py:968] (0/2) Epoch 10, batch 4550, giga_loss[loss=0.2528, simple_loss=0.3247, pruned_loss=0.09047, over 28739.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3469, pruned_loss=0.1008, over 5714523.41 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3579, pruned_loss=0.09956, over 5129179.98 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3459, pruned_loss=0.1007, over 5710941.72 frames. ], batch size: 92, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:16:36,619 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:16:40,915 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:16:49,929 INFO [train.py:968] (0/2) Epoch 10, batch 4600, giga_loss[loss=0.3676, simple_loss=0.4153, pruned_loss=0.1599, over 27580.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3483, pruned_loss=0.1012, over 5706622.74 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3571, pruned_loss=0.09927, over 5152426.66 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3478, pruned_loss=0.1015, over 5697919.68 frames. ], batch size: 472, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:17:06,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0654, 1.1756, 3.6916, 3.1199], device='cuda:0'), covar=tensor([0.1543, 0.2513, 0.0341, 0.1059], device='cuda:0'), in_proj_covar=tensor([0.0629, 0.0569, 0.0819, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 02:17:09,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.746e+02 9.711e+02 1.174e+03 1.434e+03 4.410e+03, threshold=2.349e+03, percent-clipped=5.0
+2023-03-05 02:17:34,750 INFO [train.py:968] (0/2) Epoch 10, batch 4650, giga_loss[loss=0.251, simple_loss=0.3322, pruned_loss=0.08493, over 28989.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3471, pruned_loss=0.09968, over 5700465.00 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.357, pruned_loss=0.09915, over 5160028.15 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3467, pruned_loss=0.1, over 5694122.40 frames. ], batch size: 213, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:17:41,905 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 02:18:16,924 INFO [train.py:968] (0/2) Epoch 10, batch 4700, giga_loss[loss=0.2388, simple_loss=0.3258, pruned_loss=0.07591, over 29064.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3471, pruned_loss=0.09929, over 5702890.53 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3567, pruned_loss=0.09899, over 5174738.61 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3469, pruned_loss=0.09969, over 5696584.11 frames. ], batch size: 155, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:18:33,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.093e+02 1.062e+03 1.361e+03 1.693e+03 4.548e+03, threshold=2.722e+03, percent-clipped=5.0
+2023-03-05 02:18:56,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9123, 3.7158, 3.5116, 1.5499], device='cuda:0'), covar=tensor([0.0728, 0.0852, 0.0874, 0.2313], device='cuda:0'), in_proj_covar=tensor([0.0996, 0.0931, 0.0819, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 02:19:00,189 INFO [train.py:968] (0/2) Epoch 10, batch 4750, giga_loss[loss=0.271, simple_loss=0.3419, pruned_loss=0.09999, over 28801.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3484, pruned_loss=0.1004, over 5710110.15 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3566, pruned_loss=0.09891, over 5182061.97 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3482, pruned_loss=0.1008, over 5703329.38 frames. ], batch size: 199, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:19:15,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:19:40,006 INFO [train.py:968] (0/2) Epoch 10, batch 4800, giga_loss[loss=0.2944, simple_loss=0.3563, pruned_loss=0.1162, over 28565.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3488, pruned_loss=0.1011, over 5715629.02 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3559, pruned_loss=0.09857, over 5202885.70 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3489, pruned_loss=0.1017, over 5705869.25 frames. ], batch size: 85, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:20:00,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.584e+02 1.253e+03 1.474e+03 1.873e+03 4.562e+03, threshold=2.948e+03, percent-clipped=9.0
+2023-03-05 02:20:24,859 INFO [train.py:968] (0/2) Epoch 10, batch 4850, giga_loss[loss=0.2683, simple_loss=0.3425, pruned_loss=0.09706, over 28939.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3517, pruned_loss=0.1032, over 5715463.68 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3561, pruned_loss=0.09857, over 5209991.80 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3517, pruned_loss=0.1037, over 5706059.56 frames. ], batch size: 128, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:20:38,123 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:20:47,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:20:56,727 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:21:03,976 INFO [train.py:968] (0/2) Epoch 10, batch 4900, giga_loss[loss=0.3225, simple_loss=0.3966, pruned_loss=0.1242, over 28811.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3545, pruned_loss=0.1044, over 5710554.58 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3563, pruned_loss=0.09874, over 5236908.81 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3541, pruned_loss=0.105, over 5702777.13 frames. ], batch size: 174, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:21:10,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1642, 1.5527, 1.4196, 1.3111], device='cuda:0'), covar=tensor([0.0801, 0.0272, 0.0288, 0.0980], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:0')
+2023-03-05 02:21:15,285 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=414507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:21:17,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=414510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:21:20,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.357e+02 1.091e+03 1.362e+03 1.648e+03 3.342e+03, threshold=2.725e+03, percent-clipped=2.0
+2023-03-05 02:21:42,923 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=414539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:21:47,070 INFO [train.py:968] (0/2) Epoch 10, batch 4950, giga_loss[loss=0.3003, simple_loss=0.372, pruned_loss=0.1143, over 29011.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3557, pruned_loss=0.105, over 5714590.16 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3558, pruned_loss=0.09855, over 5252571.88 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3558, pruned_loss=0.1057, over 5704855.19 frames. ], batch size: 164, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:21:53,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:21:57,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:22:16,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-05 02:22:18,148 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414585.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 02:22:26,696 INFO [train.py:968] (0/2) Epoch 10, batch 5000, giga_loss[loss=0.3955, simple_loss=0.4348, pruned_loss=0.1781, over 23657.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3577, pruned_loss=0.1063, over 5711418.19 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3564, pruned_loss=0.09881, over 5261119.13 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3574, pruned_loss=0.1069, over 5704189.04 frames. ], batch size: 705, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:22:33,689 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2713, 1.4483, 1.2396, 1.0239], device='cuda:0'), covar=tensor([0.2251, 0.2224, 0.2515, 0.1973], device='cuda:0'), in_proj_covar=tensor([0.1255, 0.0932, 0.1110, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 02:22:43,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.758e+02 1.187e+03 1.491e+03 2.065e+03 7.214e+03, threshold=2.982e+03, percent-clipped=14.0
+2023-03-05 02:22:44,757 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=414616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:22:47,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=414619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:23:07,271 INFO [train.py:968] (0/2) Epoch 10, batch 5050, giga_loss[loss=0.3643, simple_loss=0.4179, pruned_loss=0.1554, over 27853.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3584, pruned_loss=0.1066, over 5709831.93 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3568, pruned_loss=0.09888, over 5278435.62 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3579, pruned_loss=0.1071, over 5700389.10 frames. ], batch size: 412, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:23:09,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8656, 2.1389, 2.1199, 1.6853], device='cuda:0'), covar=tensor([0.1628, 0.1881, 0.1247, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0695, 0.0842, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 02:23:09,971 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=414648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:23:24,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 02:23:47,348 INFO [train.py:968] (0/2) Epoch 10, batch 5100, giga_loss[loss=0.2482, simple_loss=0.3291, pruned_loss=0.0836, over 29022.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3572, pruned_loss=0.1059, over 5715244.50 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3566, pruned_loss=0.0988, over 5286119.67 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3569, pruned_loss=0.1065, over 5706761.70 frames. ], batch size: 155, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:23:48,961 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=414696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:23:52,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=414699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:23:54,065 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=414701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:23:55,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=414704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:24:05,902 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.729e+02 1.133e+03 1.477e+03 1.930e+03 5.370e+03, threshold=2.954e+03, percent-clipped=8.0
+2023-03-05 02:24:17,894 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=414728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:24:18,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1744, 1.8258, 1.4537, 1.3194], device='cuda:0'), covar=tensor([0.0783, 0.0273, 0.0286, 0.0988], device='cuda:0'), in_proj_covar=tensor([0.0178, 0.0113, 0.0115, 0.0209], device='cuda:0'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:0')
+2023-03-05 02:24:22,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=414733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:24:31,017 INFO [train.py:968] (0/2) Epoch 10, batch 5150, giga_loss[loss=0.2533, simple_loss=0.3308, pruned_loss=0.08793, over 29075.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3546, pruned_loss=0.1047, over 5704913.17 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3564, pruned_loss=0.0987, over 5286993.07 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3545, pruned_loss=0.1052, over 5699316.25 frames. ], batch size: 128, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:25:09,750 INFO [train.py:968] (0/2) Epoch 10, batch 5200, giga_loss[loss=0.2602, simple_loss=0.3384, pruned_loss=0.09097, over 28938.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3523, pruned_loss=0.1034, over 5710384.18 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3575, pruned_loss=0.09924, over 5300720.39 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3513, pruned_loss=0.1036, over 5704750.94 frames. ], batch size: 213, lr: 3.30e-03, grad_scale: 8.0
+2023-03-05 02:25:29,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.742e+02 1.074e+03 1.301e+03 1.800e+03 9.552e+03, threshold=2.603e+03, percent-clipped=7.0
+2023-03-05 02:25:41,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:25:48,367 INFO [train.py:968] (0/2) Epoch 10, batch 5250, libri_loss[loss=0.2997, simple_loss=0.3632, pruned_loss=0.1181, over 29550.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3519, pruned_loss=0.1027, over 5718489.47 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3579, pruned_loss=0.09962, over 5323297.59 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3506, pruned_loss=0.1027, over 5708051.09 frames. ], batch size: 78, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:26:00,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:26:27,595 INFO [train.py:968] (0/2) Epoch 10, batch 5300, libri_loss[loss=0.2584, simple_loss=0.3288, pruned_loss=0.09398, over 28515.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3529, pruned_loss=0.1022, over 5720252.61 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3577, pruned_loss=0.09976, over 5346298.00 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3518, pruned_loss=0.1022, over 5706695.24 frames. ], batch size: 63, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:26:48,194 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.868e+02 1.074e+03 1.300e+03 1.716e+03 3.649e+03, threshold=2.599e+03, percent-clipped=6.0
+2023-03-05 02:27:09,734 INFO [train.py:968] (0/2) Epoch 10, batch 5350, giga_loss[loss=0.2791, simple_loss=0.3565, pruned_loss=0.1008, over 28977.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3549, pruned_loss=0.1023, over 5723882.56 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.358, pruned_loss=0.1, over 5358572.64 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3536, pruned_loss=0.1021, over 5710429.77 frames. ], batch size: 213, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:27:22,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414960.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 02:27:36,035 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=414977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:27:38,138 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=414980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:27:48,643 INFO [train.py:968] (0/2) Epoch 10, batch 5400, giga_loss[loss=0.2858, simple_loss=0.3486, pruned_loss=0.1115, over 28858.00 frames. ], tot_loss[loss=0.279, simple_loss=0.353, pruned_loss=0.1025, over 5724431.27 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3581, pruned_loss=0.1002, over 5373377.08 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3518, pruned_loss=0.1023, over 5710211.60 frames. ], batch size: 112, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:27:58,093 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:27:59,765 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=415005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:27:59,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:28:02,350 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:28:09,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.179e+02 1.222e+03 1.621e+03 2.255e+03 5.370e+03, threshold=3.241e+03, percent-clipped=16.0
+2023-03-05 02:28:17,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9321, 1.1147, 1.0702, 0.8234], device='cuda:0'), covar=tensor([0.1600, 0.1726, 0.0907, 0.1439], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1540, 0.1499, 0.1603], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 02:28:23,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:28:28,351 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5255, 1.7488, 1.6270, 1.5102], device='cuda:0'), covar=tensor([0.1410, 0.1575, 0.1722, 0.1739], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0724, 0.0659, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 02:28:29,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=415041.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 02:28:32,498 INFO [train.py:968] (0/2) Epoch 10, batch 5450, giga_loss[loss=0.2861, simple_loss=0.3545, pruned_loss=0.1088, over 28952.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3522, pruned_loss=0.104, over 5720030.82 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3581, pruned_loss=0.1003, over 5371996.72 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3511, pruned_loss=0.1037, over 5715444.23 frames. ], batch size: 174, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:29:13,290 INFO [train.py:968] (0/2) Epoch 10, batch 5500, giga_loss[loss=0.2963, simple_loss=0.364, pruned_loss=0.1144, over 28879.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3495, pruned_loss=0.104, over 5725799.84 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3581, pruned_loss=0.1003, over 5374708.74 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3486, pruned_loss=0.1038, over 5721271.40 frames. ], batch size: 199, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:29:22,232 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415103.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 02:29:25,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415106.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 02:29:32,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.619e+02 1.165e+03 1.393e+03 1.888e+03 5.427e+03, threshold=2.785e+03, percent-clipped=6.0
+2023-03-05 02:29:49,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2282, 1.6699, 1.5486, 1.1283], device='cuda:0'), covar=tensor([0.1574, 0.2248, 0.1382, 0.1562], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0693, 0.0841, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 02:29:50,080 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415135.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 02:29:55,763 INFO [train.py:968] (0/2) Epoch 10, batch 5550, giga_loss[loss=0.295, simple_loss=0.3677, pruned_loss=0.1112, over 28379.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3479, pruned_loss=0.1038, over 5730793.99 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3585, pruned_loss=0.1006, over 5388125.30 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3465, pruned_loss=0.1035, over 5722665.96 frames. ], batch size: 368, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:30:32,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415184.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 02:30:37,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415187.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 02:30:41,868 INFO [train.py:968] (0/2) Epoch 10, batch 5600, giga_loss[loss=0.2967, simple_loss=0.3615, pruned_loss=0.116, over 28662.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3474, pruned_loss=0.1038, over 5720354.09 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3586, pruned_loss=0.1006, over 5395670.33 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3461, pruned_loss=0.1035, over 5711710.07 frames. ], batch size: 284, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:31:00,934 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=415216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:31:00,949 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415216.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 02:31:02,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.243e+02 1.037e+03 1.207e+03 1.722e+03 3.723e+03, threshold=2.413e+03, percent-clipped=3.0
+2023-03-05 02:31:23,671 INFO [train.py:968] (0/2) Epoch 10, batch 5650, giga_loss[loss=0.2632, simple_loss=0.3349, pruned_loss=0.09573, over 28602.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3435, pruned_loss=0.102, over 5715811.73 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3588, pruned_loss=0.1008, over 5399672.59 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3422, pruned_loss=0.1017, over 5708333.53 frames. ], batch size: 336, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:31:32,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2277, 2.5922, 1.4215, 1.3044], device='cuda:0'), covar=tensor([0.0911, 0.0372, 0.0839, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0502, 0.0332, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 02:31:39,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6265, 2.3398, 1.8049, 0.7960], device='cuda:0'), covar=tensor([0.3992, 0.1897, 0.2894, 0.4403], device='cuda:0'), in_proj_covar=tensor([0.1494, 0.1411, 0.1457, 0.1230], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 02:31:44,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3176, 4.1475, 3.8690, 1.7637], device='cuda:0'), covar=tensor([0.0485, 0.0587, 0.0650, 0.2083], device='cuda:0'), in_proj_covar=tensor([0.0997, 0.0929, 0.0820, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 02:32:04,538 INFO [train.py:968] (0/2) Epoch 10, batch 5700, giga_loss[loss=0.284, simple_loss=0.3549, pruned_loss=0.1066, over 28248.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3395, pruned_loss=0.09971, over 5710465.96 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3588, pruned_loss=0.1008, over 5401537.22 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3381, pruned_loss=0.09943, over 5709638.05 frames. ], batch size: 368, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:32:06,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9625, 3.7810, 3.5375, 1.6799], device='cuda:0'), covar=tensor([0.0597, 0.0731, 0.0798, 0.2038], device='cuda:0'), in_proj_covar=tensor([0.0994, 0.0927, 0.0817, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 02:32:24,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.854e+02 1.136e+03 1.403e+03 1.957e+03 6.758e+03, threshold=2.806e+03, percent-clipped=17.0
+2023-03-05 02:32:25,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6047, 2.4628, 2.4335, 2.0951], device='cuda:0'), covar=tensor([0.1117, 0.1676, 0.1297, 0.1437], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0734, 0.0667, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 02:32:46,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8476, 1.8526, 1.3604, 1.5763], device='cuda:0'), covar=tensor([0.0716, 0.0597, 0.1001, 0.0983], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0439, 0.0496, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 02:32:47,197 INFO [train.py:968] (0/2) Epoch 10, batch 5750, giga_loss[loss=0.2165, simple_loss=0.2985, pruned_loss=0.06726, over 28537.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3372, pruned_loss=0.09831, over 5713686.02 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3587, pruned_loss=0.1008, over 5409454.91 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3359, pruned_loss=0.09812, over 5711532.14 frames. ], batch size: 60, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:32:48,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9825, 2.7766, 2.6080, 1.5203], device='cuda:0'), covar=tensor([0.0973, 0.1056, 0.0904, 0.2164], device='cuda:0'), in_proj_covar=tensor([0.0998, 0.0927, 0.0818, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 02:32:50,388 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 02:33:07,104 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=415371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:33:14,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=415380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:33:25,640 INFO [train.py:968] (0/2) Epoch 10, batch 5800, giga_loss[loss=0.281, simple_loss=0.3582, pruned_loss=0.1019, over 28991.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3392, pruned_loss=0.09871, over 5713535.50 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3588, pruned_loss=0.1009, over 5422388.70 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3374, pruned_loss=0.09835, over 5708274.40 frames. ], batch size: 164, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:33:30,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 02:33:39,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1691, 3.3016, 1.3486, 1.3859], device='cuda:0'), covar=tensor([0.1126, 0.0388, 0.1052, 0.1521], device='cuda:0'), in_proj_covar=tensor([0.0347, 0.0500, 0.0331, 0.0383], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 02:33:44,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.869e+02 1.127e+03 1.583e+03 2.159e+03 5.647e+03, threshold=3.166e+03, percent-clipped=9.0
+2023-03-05 02:34:05,005 INFO [train.py:968] (0/2) Epoch 10, batch 5850, giga_loss[loss=0.3191, simple_loss=0.37, pruned_loss=0.1341, over 23882.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3438, pruned_loss=0.101, over 5703232.13 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3592, pruned_loss=0.1012, over 5428029.19 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3416, pruned_loss=0.1004, over 5703584.24 frames. ], batch size: 705, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:34:10,528 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-05 02:34:45,992 INFO [train.py:968] (0/2) Epoch 10, batch 5900, giga_loss[loss=0.2724, simple_loss=0.3526, pruned_loss=0.09608, over 28489.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3472, pruned_loss=0.1019, over 5713170.87 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3594, pruned_loss=0.1014, over 5436671.37 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3449, pruned_loss=0.1013, over 5710601.85 frames. ], batch size: 336, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:35:06,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.981e+02 1.082e+03 1.375e+03 1.919e+03 5.058e+03, threshold=2.749e+03, percent-clipped=3.0
+2023-03-05 02:35:13,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:35:14,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:35:32,258 INFO [train.py:968] (0/2) Epoch 10, batch 5950, giga_loss[loss=0.2829, simple_loss=0.3587, pruned_loss=0.1036, over 28562.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3507, pruned_loss=0.1035, over 5707968.61 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3591, pruned_loss=0.1013, over 5441346.97 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.349, pruned_loss=0.1031, over 5704084.00 frames. ], batch size: 336, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:35:41,890 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:36:13,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=415591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:36:17,410 INFO [train.py:968] (0/2) Epoch 10, batch 6000, libri_loss[loss=0.2985, simple_loss=0.3724, pruned_loss=0.1123, over 29555.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.354, pruned_loss=0.1057, over 5709896.61 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3593, pruned_loss=0.1013, over 5448350.55 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3525, pruned_loss=0.1054, over 5703912.39 frames. ], batch size: 89, lr: 3.30e-03, grad_scale: 8.0
+2023-03-05 02:36:17,415 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 02:36:25,634 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2254, simple_loss=0.3305, pruned_loss=0.06017, over 944034.00 frames. 
+2023-03-05 02:36:25,635 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-05 02:36:48,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.291e+02 1.163e+03 1.464e+03 1.913e+03 4.428e+03, threshold=2.927e+03, percent-clipped=6.0
+2023-03-05 02:37:10,701 INFO [train.py:968] (0/2) Epoch 10, batch 6050, giga_loss[loss=0.3116, simple_loss=0.376, pruned_loss=0.1236, over 28920.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3589, pruned_loss=0.1102, over 5697327.85 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3591, pruned_loss=0.1013, over 5454074.63 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3578, pruned_loss=0.1103, over 5694932.34 frames. ], batch size: 186, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:37:50,517 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=415689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:37:55,101 INFO [train.py:968] (0/2) Epoch 10, batch 6100, giga_loss[loss=0.2852, simple_loss=0.3614, pruned_loss=0.1045, over 28952.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3646, pruned_loss=0.1149, over 5689096.13 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3591, pruned_loss=0.1016, over 5459046.92 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3639, pruned_loss=0.1151, over 5690946.37 frames. ], batch size: 136, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:38:19,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.170e+02 1.603e+03 2.164e+03 2.971e+03 7.941e+03, threshold=4.328e+03, percent-clipped=27.0
+2023-03-05 02:38:35,231 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:38:37,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:38:42,548 INFO [train.py:968] (0/2) Epoch 10, batch 6150, giga_loss[loss=0.4756, simple_loss=0.4825, pruned_loss=0.2344, over 26579.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3714, pruned_loss=0.1195, over 5693065.61 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3593, pruned_loss=0.1017, over 5467201.52 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3707, pruned_loss=0.1198, over 5691509.94 frames. ], batch size: 555, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:38:44,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=415746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:39:04,034 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:39:32,916 INFO [train.py:968] (0/2) Epoch 10, batch 6200, giga_loss[loss=0.3248, simple_loss=0.3873, pruned_loss=0.1311, over 28872.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3768, pruned_loss=0.1245, over 5693389.53 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3592, pruned_loss=0.1016, over 5474374.76 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3767, pruned_loss=0.1252, over 5689698.13 frames. ], batch size: 136, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:39:56,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.946e+02 1.494e+03 1.911e+03 2.397e+03 9.734e+03, threshold=3.822e+03, percent-clipped=5.0
+2023-03-05 02:40:11,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6206, 2.3932, 1.6432, 0.8384], device='cuda:0'), covar=tensor([0.3899, 0.2075, 0.3142, 0.3967], device='cuda:0'), in_proj_covar=tensor([0.1513, 0.1429, 0.1477, 0.1234], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 02:40:12,525 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=415837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:40:18,205 INFO [train.py:968] (0/2) Epoch 10, batch 6250, giga_loss[loss=0.3049, simple_loss=0.3757, pruned_loss=0.117, over 28918.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3831, pruned_loss=0.13, over 5697665.65 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3588, pruned_loss=0.1015, over 5486654.72 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3839, pruned_loss=0.1314, over 5689322.97 frames. ], batch size: 106, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:41:01,719 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:41:03,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:41:06,515 INFO [train.py:968] (0/2) Epoch 10, batch 6300, libri_loss[loss=0.3452, simple_loss=0.4027, pruned_loss=0.1438, over 19532.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3874, pruned_loss=0.1337, over 5679313.60 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3595, pruned_loss=0.1018, over 5483852.73 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3882, pruned_loss=0.1354, over 5681411.69 frames. ], batch size: 189, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:41:21,004 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-05 02:41:32,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.710e+02 1.578e+03 2.173e+03 3.059e+03 9.205e+03, threshold=4.346e+03, percent-clipped=13.0
+2023-03-05 02:41:33,714 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:41:58,542 INFO [train.py:968] (0/2) Epoch 10, batch 6350, giga_loss[loss=0.4634, simple_loss=0.4655, pruned_loss=0.2307, over 26549.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3886, pruned_loss=0.1353, over 5667030.10 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3595, pruned_loss=0.1018, over 5490057.93 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3896, pruned_loss=0.1371, over 5665635.32 frames. ], batch size: 555, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:42:11,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3000, 1.3286, 1.1443, 1.0100], device='cuda:0'), covar=tensor([0.0660, 0.0431, 0.0936, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0440, 0.0496, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 02:42:51,618 INFO [train.py:968] (0/2) Epoch 10, batch 6400, giga_loss[loss=0.3609, simple_loss=0.421, pruned_loss=0.1504, over 28916.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3915, pruned_loss=0.1387, over 5665565.88 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.359, pruned_loss=0.1015, over 5496255.88 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3932, pruned_loss=0.1409, over 5661233.33 frames. ], batch size: 199, lr: 3.30e-03, grad_scale: 8.0
+2023-03-05 02:42:58,600 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-416000.pt
+2023-03-05 02:43:11,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3798, 1.7547, 1.6586, 1.2282], device='cuda:0'), covar=tensor([0.1637, 0.2304, 0.1388, 0.1623], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0691, 0.0834, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-05 02:43:17,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.331e+02 1.650e+03 2.133e+03 2.986e+03 8.153e+03, threshold=4.266e+03, percent-clipped=5.0
+2023-03-05 02:43:23,202 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-05 02:43:41,709 INFO [train.py:968] (0/2) Epoch 10, batch 6450, libri_loss[loss=0.2347, simple_loss=0.3139, pruned_loss=0.07775, over 28483.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3926, pruned_loss=0.1406, over 5663887.76 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3589, pruned_loss=0.1015, over 5508649.32 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3951, pruned_loss=0.1436, over 5654196.82 frames. ], batch size: 63, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:44:03,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:44:06,956 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-05 02:44:36,303 INFO [train.py:968] (0/2) Epoch 10, batch 6500, giga_loss[loss=0.3703, simple_loss=0.4089, pruned_loss=0.1659, over 28555.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3977, pruned_loss=0.1452, over 5644784.64 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3589, pruned_loss=0.1014, over 5510640.07 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.3998, pruned_loss=0.1478, over 5636206.31 frames. ], batch size: 336, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:45:01,689 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 02:45:03,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.643e+02 1.722e+03 2.297e+03 3.203e+03 6.304e+03, threshold=4.593e+03, percent-clipped=12.0
+2023-03-05 02:45:12,061 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=416129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:45:28,970 INFO [train.py:968] (0/2) Epoch 10, batch 6550, giga_loss[loss=0.4509, simple_loss=0.4542, pruned_loss=0.2238, over 26501.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.398, pruned_loss=0.1469, over 5646563.55 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3589, pruned_loss=0.1013, over 5514453.78 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4001, pruned_loss=0.1495, over 5637582.80 frames. ], batch size: 555, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:46:04,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4637, 1.6996, 1.6850, 1.2602], device='cuda:0'), covar=tensor([0.1532, 0.2180, 0.1226, 0.1446], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0693, 0.0836, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-05 02:46:18,009 INFO [train.py:968] (0/2) Epoch 10, batch 6600, giga_loss[loss=0.39, simple_loss=0.4205, pruned_loss=0.1797, over 27609.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3968, pruned_loss=0.1465, over 5642773.62 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3588, pruned_loss=0.1011, over 5518793.95 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.3993, pruned_loss=0.1496, over 5633910.19 frames. ], batch size: 472, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:46:33,546 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=416207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:46:36,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=416210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:46:37,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:46:46,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.777e+02 1.717e+03 2.669e+03 3.454e+03 9.093e+03, threshold=5.338e+03, percent-clipped=10.0
+2023-03-05 02:47:04,511 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=416239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:47:09,218 INFO [train.py:968] (0/2) Epoch 10, batch 6650, giga_loss[loss=0.3228, simple_loss=0.394, pruned_loss=0.1258, over 28981.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.398, pruned_loss=0.1465, over 5641019.10 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3593, pruned_loss=0.1014, over 5523361.77 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.4, pruned_loss=0.1493, over 5631448.92 frames. ], batch size: 136, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:47:19,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2297, 1.4675, 1.3733, 1.5016], device='cuda:0'), covar=tensor([0.0762, 0.0335, 0.0299, 0.0788], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0055, 0.0050, 0.0084], device='cuda:0')
+2023-03-05 02:47:28,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2613, 1.2528, 1.0985, 0.9625], device='cuda:0'), covar=tensor([0.0739, 0.0495, 0.1047, 0.0899], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0444, 0.0499, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 02:47:29,106 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7065, 1.9997, 2.0763, 1.5194], device='cuda:0'), covar=tensor([0.1590, 0.1946, 0.1187, 0.1443], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0694, 0.0836, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:0')
+2023-03-05 02:47:55,103 INFO [train.py:968] (0/2) Epoch 10, batch 6700, giga_loss[loss=0.4141, simple_loss=0.4456, pruned_loss=0.1912, over 27892.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.3967, pruned_loss=0.1443, over 5643237.55 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3598, pruned_loss=0.1017, over 5530335.21 frames. ], giga_tot_loss[loss=0.3473, simple_loss=0.3991, pruned_loss=0.1477, over 5633017.01 frames. ], batch size: 412, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:48:17,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.060e+02 1.600e+03 2.141e+03 2.796e+03 8.982e+03, threshold=4.281e+03, percent-clipped=2.0
+2023-03-05 02:48:43,386 INFO [train.py:968] (0/2) Epoch 10, batch 6750, giga_loss[loss=0.3689, simple_loss=0.3983, pruned_loss=0.1698, over 24015.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3975, pruned_loss=0.1446, over 5636682.93 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.36, pruned_loss=0.1018, over 5542097.91 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4003, pruned_loss=0.1483, over 5621197.10 frames. ], batch size: 705, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:48:45,378 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 02:48:50,482 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-05 02:48:55,529 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=416355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:48:57,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=416358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:49:26,847 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=416387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:49:28,621 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=416389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:49:33,859 INFO [train.py:968] (0/2) Epoch 10, batch 6800, giga_loss[loss=0.3411, simple_loss=0.3908, pruned_loss=0.1458, over 27625.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3939, pruned_loss=0.1409, over 5638505.42 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3603, pruned_loss=0.1021, over 5547091.11 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3965, pruned_loss=0.1444, over 5623530.51 frames. ], batch size: 472, lr: 3.30e-03, grad_scale: 8.0
+2023-03-05 02:49:43,322 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-05 02:50:01,045 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.728e+02 1.599e+03 1.974e+03 2.840e+03 1.062e+04, threshold=3.947e+03, percent-clipped=4.0
+2023-03-05 02:50:24,164 INFO [train.py:968] (0/2) Epoch 10, batch 6850, libri_loss[loss=0.3216, simple_loss=0.3842, pruned_loss=0.1296, over 18830.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3922, pruned_loss=0.1385, over 5635784.29 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3605, pruned_loss=0.1023, over 5544169.68 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3951, pruned_loss=0.1422, over 5630023.72 frames. ], batch size: 186, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:51:11,531 INFO [train.py:968] (0/2) Epoch 10, batch 6900, giga_loss[loss=0.2926, simple_loss=0.342, pruned_loss=0.1216, over 24005.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3889, pruned_loss=0.1356, over 5642257.67 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3601, pruned_loss=0.1023, over 5550038.17 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3924, pruned_loss=0.1395, over 5634953.09 frames. ], batch size: 705, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:51:20,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:51:28,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3390, 2.1945, 1.5697, 0.6066], device='cuda:0'), covar=tensor([0.3241, 0.1818, 0.2510, 0.3799], device='cuda:0'), in_proj_covar=tensor([0.1526, 0.1445, 0.1481, 0.1252], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 02:51:33,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2417, 1.7069, 1.2188, 0.6304], device='cuda:0'), covar=tensor([0.2987, 0.1524, 0.2061, 0.3573], device='cuda:0'), in_proj_covar=tensor([0.1526, 0.1445, 0.1481, 0.1251], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 02:51:33,978 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-05 02:51:37,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.649e+02 1.578e+03 2.104e+03 3.525e+03 7.613e+03, threshold=4.207e+03, percent-clipped=18.0
+2023-03-05 02:51:44,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=416529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:51:59,045 INFO [train.py:968] (0/2) Epoch 10, batch 6950, giga_loss[loss=0.3453, simple_loss=0.4016, pruned_loss=0.1445, over 28718.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3851, pruned_loss=0.1321, over 5644406.59 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3602, pruned_loss=0.1024, over 5546445.14 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3882, pruned_loss=0.1358, over 5642817.16 frames. ], batch size: 92, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:51:59,417 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9193, 2.2883, 1.7750, 1.5540], device='cuda:0'), covar=tensor([0.1784, 0.1509, 0.1654, 0.1781], device='cuda:0'), in_proj_covar=tensor([0.1637, 0.1541, 0.1495, 0.1601], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 02:52:01,952 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=416547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:52:03,684 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=416548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:52:12,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4119, 2.2560, 2.0395, 2.1782], device='cuda:0'), covar=tensor([0.1109, 0.1864, 0.1674, 0.1599], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0728, 0.0663, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 02:52:47,196 INFO [train.py:968] (0/2) Epoch 10, batch 7000, giga_loss[loss=0.3082, simple_loss=0.38, pruned_loss=0.1182, over 28837.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3838, pruned_loss=0.1314, over 5645210.46 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3604, pruned_loss=0.1025, over 5549800.17 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3866, pruned_loss=0.1349, over 5642864.48 frames. ], batch size: 199, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:53:09,688 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.185e+02 1.392e+03 1.681e+03 2.221e+03 4.062e+03, threshold=3.363e+03, percent-clipped=0.0
+2023-03-05 02:53:30,135 INFO [train.py:968] (0/2) Epoch 10, batch 7050, giga_loss[loss=0.3237, simple_loss=0.3594, pruned_loss=0.144, over 23529.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3812, pruned_loss=0.1296, over 5647523.76 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3601, pruned_loss=0.1025, over 5565509.48 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3848, pruned_loss=0.1335, over 5635253.03 frames. ], batch size: 705, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:53:32,671 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=416647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:53:34,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=416650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:54:05,470 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=416679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:54:20,805 INFO [train.py:968] (0/2) Epoch 10, batch 7100, giga_loss[loss=0.3049, simple_loss=0.3784, pruned_loss=0.1157, over 29073.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3818, pruned_loss=0.1297, over 5657684.58 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3597, pruned_loss=0.1023, over 5575958.01 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3855, pruned_loss=0.1339, over 5640772.74 frames. ], batch size: 155, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:54:28,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2269, 1.4355, 1.2973, 1.5283], device='cuda:0'), covar=tensor([0.0798, 0.0328, 0.0319, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0055, 0.0050, 0.0084], device='cuda:0')
+2023-03-05 02:54:51,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.435e+03 1.857e+03 2.706e+03 6.672e+03, threshold=3.713e+03, percent-clipped=12.0
+2023-03-05 02:55:03,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0594, 3.8836, 3.6599, 1.8250], device='cuda:0'), covar=tensor([0.0564, 0.0698, 0.0681, 0.2120], device='cuda:0'), in_proj_covar=tensor([0.1022, 0.0960, 0.0842, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 02:55:15,139 INFO [train.py:968] (0/2) Epoch 10, batch 7150, giga_loss[loss=0.3177, simple_loss=0.3961, pruned_loss=0.1196, over 28854.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3798, pruned_loss=0.1275, over 5659914.10 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3597, pruned_loss=0.1023, over 5580843.03 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3832, pruned_loss=0.1312, over 5643676.67 frames. ], batch size: 227, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:55:38,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:55:45,367 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5030, 3.7670, 1.5182, 1.7069], device='cuda:0'), covar=tensor([0.0902, 0.0313, 0.0904, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0508, 0.0336, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 02:56:08,274 INFO [train.py:968] (0/2) Epoch 10, batch 7200, libri_loss[loss=0.2678, simple_loss=0.3516, pruned_loss=0.09199, over 27950.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3802, pruned_loss=0.1254, over 5663962.50 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3595, pruned_loss=0.1023, over 5589363.08 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3838, pruned_loss=0.1292, over 5645801.90 frames. ], batch size: 116, lr: 3.30e-03, grad_scale: 8.0
+2023-03-05 02:56:32,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.428e+02 1.439e+03 1.878e+03 2.646e+03 6.712e+03, threshold=3.756e+03, percent-clipped=10.0
+2023-03-05 02:56:34,509 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-05 02:56:52,786 INFO [train.py:968] (0/2) Epoch 10, batch 7250, giga_loss[loss=0.3166, simple_loss=0.3901, pruned_loss=0.1215, over 29063.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3815, pruned_loss=0.1249, over 5683874.98 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3589, pruned_loss=0.1021, over 5598760.52 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3854, pruned_loss=0.1288, over 5662886.14 frames. ], batch size: 155, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:57:43,129 INFO [train.py:968] (0/2) Epoch 10, batch 7300, libri_loss[loss=0.2873, simple_loss=0.3628, pruned_loss=0.1059, over 29475.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3812, pruned_loss=0.1254, over 5673765.15 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3582, pruned_loss=0.1016, over 5604003.58 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.386, pruned_loss=0.1298, over 5654442.36 frames. ], batch size: 85, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:57:50,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:57:52,948 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=416907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:57:55,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=416910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:58:08,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:58:09,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.628e+03 2.092e+03 2.930e+03 1.019e+04, threshold=4.184e+03, percent-clipped=16.0
+2023-03-05 02:58:09,446 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:58:22,942 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=416939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:58:27,290 INFO [train.py:968] (0/2) Epoch 10, batch 7350, giga_loss[loss=0.3697, simple_loss=0.4178, pruned_loss=0.1608, over 28557.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3816, pruned_loss=0.1266, over 5676854.64 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3579, pruned_loss=0.1014, over 5610605.31 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3861, pruned_loss=0.1308, over 5657085.35 frames. ], batch size: 336, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:58:33,884 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 02:59:16,112 INFO [train.py:968] (0/2) Epoch 10, batch 7400, giga_loss[loss=0.2761, simple_loss=0.3508, pruned_loss=0.1007, over 29029.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3798, pruned_loss=0.1262, over 5668650.74 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3575, pruned_loss=0.1012, over 5604053.44 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3843, pruned_loss=0.1304, over 5660398.46 frames. ], batch size: 128, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:59:40,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.986e+02 1.596e+03 2.169e+03 2.895e+03 6.182e+03, threshold=4.338e+03, percent-clipped=9.0
+2023-03-05 02:59:51,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5089, 1.8601, 1.5660, 1.6973], device='cuda:0'), covar=tensor([0.0733, 0.0267, 0.0282, 0.0750], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0056, 0.0050, 0.0084], device='cuda:0')
+2023-03-05 02:59:58,963 INFO [train.py:968] (0/2) Epoch 10, batch 7450, giga_loss[loss=0.3558, simple_loss=0.4038, pruned_loss=0.1539, over 27589.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3798, pruned_loss=0.1276, over 5668216.43 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3575, pruned_loss=0.101, over 5609610.23 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3842, pruned_loss=0.1319, over 5658851.02 frames. ], batch size: 472, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 03:00:01,154 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:03,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:16,369 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:18,449 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:20,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:21,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:32,814 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:34,756 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:46,690 INFO [train.py:968] (0/2) Epoch 10, batch 7500, giga_loss[loss=0.2997, simple_loss=0.3791, pruned_loss=0.1102, over 28324.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3791, pruned_loss=0.1264, over 5668103.71 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3577, pruned_loss=0.1012, over 5615159.59 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3831, pruned_loss=0.1305, over 5657247.30 frames. ], batch size: 71, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 03:00:50,556 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:51,131 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:01:12,732 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.380e+02 1.524e+03 1.934e+03 2.897e+03 7.142e+03, threshold=3.867e+03, percent-clipped=9.0
+2023-03-05 03:01:18,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4768, 1.6369, 1.3410, 1.5373], device='cuda:0'), covar=tensor([0.2271, 0.2241, 0.2511, 0.2004], device='cuda:0'), in_proj_covar=tensor([0.1273, 0.0944, 0.1129, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 03:01:31,683 INFO [train.py:968] (0/2) Epoch 10, batch 7550, giga_loss[loss=0.2909, simple_loss=0.373, pruned_loss=0.1044, over 28879.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3786, pruned_loss=0.1252, over 5666560.17 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3576, pruned_loss=0.1012, over 5617738.33 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3824, pruned_loss=0.1292, over 5656708.96 frames. ], batch size: 174, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 03:02:13,896 INFO [train.py:968] (0/2) Epoch 10, batch 7600, giga_loss[loss=0.3172, simple_loss=0.3754, pruned_loss=0.1295, over 28548.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.379, pruned_loss=0.1252, over 5665441.05 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3575, pruned_loss=0.1013, over 5615869.43 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3829, pruned_loss=0.1291, over 5661000.30 frames. ], batch size: 85, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 03:02:38,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.709e+02 1.367e+03 1.737e+03 2.555e+03 1.427e+04, threshold=3.473e+03, percent-clipped=10.0
+2023-03-05 03:02:51,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:02:55,498 INFO [train.py:968] (0/2) Epoch 10, batch 7650, giga_loss[loss=0.3614, simple_loss=0.4074, pruned_loss=0.1577, over 28975.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3768, pruned_loss=0.1233, over 5683174.30 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3573, pruned_loss=0.1012, over 5623166.88 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3809, pruned_loss=0.1274, over 5675343.44 frames. ], batch size: 213, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 03:03:46,377 INFO [train.py:968] (0/2) Epoch 10, batch 7700, giga_loss[loss=0.3125, simple_loss=0.3801, pruned_loss=0.1224, over 29010.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3753, pruned_loss=0.1233, over 5677409.14 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3573, pruned_loss=0.1013, over 5627411.21 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3788, pruned_loss=0.1268, over 5668222.83 frames. ], batch size: 155, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:04:09,123 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-05 03:04:18,048 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.498e+02 1.534e+03 1.974e+03 2.461e+03 4.629e+03, threshold=3.948e+03, percent-clipped=3.0
+2023-03-05 03:04:36,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:04:36,439 INFO [train.py:968] (0/2) Epoch 10, batch 7750, giga_loss[loss=0.2974, simple_loss=0.3648, pruned_loss=0.115, over 28830.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3745, pruned_loss=0.1234, over 5669906.25 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3574, pruned_loss=0.1012, over 5626787.84 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3775, pruned_loss=0.1266, over 5663886.25 frames. ], batch size: 199, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:05:19,364 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4920, 1.6777, 1.5930, 1.3216], device='cuda:0'), covar=tensor([0.2079, 0.1612, 0.1332, 0.1703], device='cuda:0'), in_proj_covar=tensor([0.1665, 0.1560, 0.1519, 0.1628], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 03:05:27,481 INFO [train.py:968] (0/2) Epoch 10, batch 7800, giga_loss[loss=0.3016, simple_loss=0.3717, pruned_loss=0.1157, over 28894.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3751, pruned_loss=0.1253, over 5660171.90 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3576, pruned_loss=0.1013, over 5627014.66 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3775, pruned_loss=0.128, over 5655523.31 frames. ], batch size: 174, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:05:55,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.826e+02 1.630e+03 2.134e+03 3.167e+03 7.076e+03, threshold=4.269e+03, percent-clipped=18.0
+2023-03-05 03:06:01,571 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-05 03:06:17,271 INFO [train.py:968] (0/2) Epoch 10, batch 7850, giga_loss[loss=0.317, simple_loss=0.3757, pruned_loss=0.1291, over 28791.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3731, pruned_loss=0.1245, over 5656715.16 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3575, pruned_loss=0.1013, over 5633894.77 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3754, pruned_loss=0.1273, over 5647663.97 frames. ], batch size: 284, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:06:27,829 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=417456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:06:58,373 INFO [train.py:968] (0/2) Epoch 10, batch 7900, giga_loss[loss=0.328, simple_loss=0.3812, pruned_loss=0.1374, over 28751.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3729, pruned_loss=0.1245, over 5645429.27 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.358, pruned_loss=0.1017, over 5623207.07 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.375, pruned_loss=0.1273, over 5648535.96 frames. ], batch size: 99, lr: 3.29e-03, grad_scale: 2.0
+2023-03-05 03:07:28,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.171e+02 1.745e+03 2.790e+03 4.622e+03 2.495e+04, threshold=5.580e+03, percent-clipped=29.0
+2023-03-05 03:07:46,223 INFO [train.py:968] (0/2) Epoch 10, batch 7950, giga_loss[loss=0.3277, simple_loss=0.3892, pruned_loss=0.1331, over 28856.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3736, pruned_loss=0.1243, over 5643646.28 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3582, pruned_loss=0.1019, over 5610617.38 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3754, pruned_loss=0.127, over 5658027.25 frames. ], batch size: 284, lr: 3.29e-03, grad_scale: 2.0
+2023-03-05 03:08:06,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8690, 1.0324, 1.0541, 0.7512], device='cuda:0'), covar=tensor([0.1832, 0.1691, 0.0981, 0.1631], device='cuda:0'), in_proj_covar=tensor([0.1659, 0.1562, 0.1513, 0.1628], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 03:08:10,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-05 03:08:27,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4307, 1.7224, 1.7648, 1.2811], device='cuda:0'), covar=tensor([0.1748, 0.2356, 0.1426, 0.1661], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0705, 0.0847, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 03:08:29,581 INFO [train.py:968] (0/2) Epoch 10, batch 8000, giga_loss[loss=0.3572, simple_loss=0.4062, pruned_loss=0.1541, over 28874.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3753, pruned_loss=0.1251, over 5642337.61 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.358, pruned_loss=0.1017, over 5611465.83 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3775, pruned_loss=0.1281, over 5653805.21 frames. ], batch size: 186, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:08:31,544 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:08:34,795 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:08:38,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:08:52,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=417615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:09:00,854 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.789e+02 1.486e+03 2.041e+03 3.781e+03 9.095e+03, threshold=4.081e+03, percent-clipped=7.0
+2023-03-05 03:09:05,896 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:09:07,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-05 03:09:17,173 INFO [train.py:968] (0/2) Epoch 10, batch 8050, giga_loss[loss=0.2903, simple_loss=0.3599, pruned_loss=0.1103, over 28542.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3749, pruned_loss=0.1237, over 5660967.10 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3579, pruned_loss=0.1017, over 5617027.22 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3771, pruned_loss=0.1265, over 5665661.81 frames. ], batch size: 71, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:10:02,683 INFO [train.py:968] (0/2) Epoch 10, batch 8100, giga_loss[loss=0.3107, simple_loss=0.3732, pruned_loss=0.1241, over 28910.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3748, pruned_loss=0.1229, over 5673819.95 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3582, pruned_loss=0.1019, over 5621228.23 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3766, pruned_loss=0.1255, over 5674908.40 frames. ], batch size: 227, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:10:28,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=417718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:10:32,510 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:10:32,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.948e+02 1.451e+03 1.816e+03 2.361e+03 8.315e+03, threshold=3.633e+03, percent-clipped=5.0
+2023-03-05 03:10:51,004 INFO [train.py:968] (0/2) Epoch 10, batch 8150, giga_loss[loss=0.3447, simple_loss=0.3887, pruned_loss=0.1504, over 28782.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3764, pruned_loss=0.1246, over 5679262.57 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3586, pruned_loss=0.1021, over 5628414.93 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3779, pruned_loss=0.1269, over 5674696.10 frames. ], batch size: 99, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:11:07,309 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:11:09,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:11:28,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7267, 1.6200, 1.3671, 1.2644], device='cuda:0'), covar=tensor([0.0544, 0.0473, 0.0736, 0.0921], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0447, 0.0502, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 03:11:34,631 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:11:40,091 INFO [train.py:968] (0/2) Epoch 10, batch 8200, giga_loss[loss=0.3018, simple_loss=0.369, pruned_loss=0.1173, over 29003.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.378, pruned_loss=0.1267, over 5671257.00 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3584, pruned_loss=0.1019, over 5635019.56 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.38, pruned_loss=0.1295, over 5662737.24 frames. ], batch size: 136, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:11:45,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2184, 3.9953, 3.8369, 2.0252], device='cuda:0'), covar=tensor([0.0514, 0.0689, 0.0742, 0.2104], device='cuda:0'), in_proj_covar=tensor([0.1033, 0.0969, 0.0853, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 03:11:49,995 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3786, 2.0081, 1.6241, 1.5879], device='cuda:0'), covar=tensor([0.0706, 0.0255, 0.0271, 0.0768], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0055, 0.0049, 0.0084], device='cuda:0')
+2023-03-05 03:12:04,048 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 03:12:11,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.323e+02 1.639e+03 2.197e+03 2.769e+03 7.370e+03, threshold=4.395e+03, percent-clipped=10.0
+2023-03-05 03:12:22,948 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-05 03:12:29,627 INFO [train.py:968] (0/2) Epoch 10, batch 8250, giga_loss[loss=0.4397, simple_loss=0.4492, pruned_loss=0.2151, over 26566.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3782, pruned_loss=0.1288, over 5663173.83 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.358, pruned_loss=0.1017, over 5642234.58 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3807, pruned_loss=0.1319, over 5650347.28 frames. ], batch size: 555, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:12:48,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:12:50,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:13:16,061 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:13:16,504 INFO [train.py:968] (0/2) Epoch 10, batch 8300, giga_loss[loss=0.3318, simple_loss=0.3902, pruned_loss=0.1367, over 28566.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3796, pruned_loss=0.1307, over 5669031.05 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3578, pruned_loss=0.1016, over 5640749.35 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3825, pruned_loss=0.1342, over 5661228.71 frames. ], batch size: 85, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:13:34,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4230, 1.5751, 1.3292, 1.4823], device='cuda:0'), covar=tensor([0.0744, 0.0311, 0.0313, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0055, 0.0049, 0.0084], device='cuda:0')
+2023-03-05 03:13:45,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.790e+03 2.229e+03 3.104e+03 7.683e+03, threshold=4.459e+03, percent-clipped=7.0
+2023-03-05 03:13:57,562 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-05 03:14:04,455 INFO [train.py:968] (0/2) Epoch 10, batch 8350, giga_loss[loss=0.4663, simple_loss=0.4708, pruned_loss=0.2309, over 26615.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3804, pruned_loss=0.1318, over 5658579.01 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3574, pruned_loss=0.1014, over 5636597.47 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3836, pruned_loss=0.1355, over 5655820.74 frames. ], batch size: 555, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:14:29,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=417971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:14:47,763 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:14:49,926 INFO [train.py:968] (0/2) Epoch 10, batch 8400, giga_loss[loss=0.3033, simple_loss=0.376, pruned_loss=0.1153, over 28928.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.378, pruned_loss=0.1296, over 5665097.16 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3571, pruned_loss=0.1012, over 5639758.69 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3813, pruned_loss=0.1334, over 5660562.42 frames. ], batch size: 164, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:14:55,599 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-418000.pt
+2023-03-05 03:15:16,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.148e+02 1.380e+03 1.830e+03 2.450e+03 6.353e+03, threshold=3.660e+03, percent-clipped=3.0
+2023-03-05 03:15:20,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2036, 2.6062, 1.2881, 1.3352], device='cuda:0'), covar=tensor([0.0894, 0.0308, 0.0852, 0.1275], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0509, 0.0335, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 03:15:36,010 INFO [train.py:968] (0/2) Epoch 10, batch 8450, giga_loss[loss=0.2986, simple_loss=0.3773, pruned_loss=0.1099, over 28862.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3773, pruned_loss=0.1274, over 5677683.26 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3569, pruned_loss=0.1011, over 5642570.30 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3802, pruned_loss=0.1307, over 5671953.47 frames. ], batch size: 174, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:16:01,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7544, 1.7063, 1.3575, 1.3741], device='cuda:0'), covar=tensor([0.0693, 0.0634, 0.0857, 0.1086], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0445, 0.0498, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 03:16:05,608 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-05 03:16:21,585 INFO [train.py:968] (0/2) Epoch 10, batch 8500, giga_loss[loss=0.2933, simple_loss=0.3478, pruned_loss=0.1194, over 27624.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3756, pruned_loss=0.1257, over 5671450.33 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3573, pruned_loss=0.1014, over 5641936.76 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3778, pruned_loss=0.1284, over 5667722.69 frames. ], batch size: 472, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:16:25,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:16:29,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2562, 1.6163, 1.2422, 0.6439], device='cuda:0'), covar=tensor([0.2733, 0.2015, 0.1611, 0.3424], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1459, 0.1477, 0.1259], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 03:16:39,421 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:16:41,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:16:46,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.404e+02 1.496e+03 2.074e+03 2.872e+03 9.103e+03, threshold=4.149e+03, percent-clipped=13.0
+2023-03-05 03:17:08,503 INFO [train.py:968] (0/2) Epoch 10, batch 8550, giga_loss[loss=0.2864, simple_loss=0.3623, pruned_loss=0.1052, over 28991.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3731, pruned_loss=0.1243, over 5671109.77 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3572, pruned_loss=0.1013, over 5650037.22 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3755, pruned_loss=0.1272, over 5661826.47 frames. ], batch size: 164, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:17:09,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:17:32,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2312, 1.4488, 1.3351, 1.0661], device='cuda:0'), covar=tensor([0.1560, 0.1460, 0.0871, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.1655, 0.1553, 0.1501, 0.1618], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 03:17:35,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4060, 1.9058, 1.7073, 1.2687], device='cuda:0'), covar=tensor([0.1687, 0.2309, 0.1417, 0.1636], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0703, 0.0846, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 03:17:37,759 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:17:52,673 INFO [train.py:968] (0/2) Epoch 10, batch 8600, giga_loss[loss=0.3015, simple_loss=0.3652, pruned_loss=0.1189, over 28872.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.372, pruned_loss=0.1242, over 5681929.92 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3566, pruned_loss=0.101, over 5654480.61 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3747, pruned_loss=0.1273, over 5670985.95 frames. ], batch size: 199, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:18:23,858 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.238e+02 1.618e+03 1.985e+03 2.766e+03 8.678e+03, threshold=3.969e+03, percent-clipped=5.0
+2023-03-05 03:18:35,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 03:18:38,648 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:18:43,704 INFO [train.py:968] (0/2) Epoch 10, batch 8650, giga_loss[loss=0.3747, simple_loss=0.434, pruned_loss=0.1577, over 28797.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3732, pruned_loss=0.1251, over 5677892.28 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3557, pruned_loss=0.1005, over 5657846.74 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3767, pruned_loss=0.1287, over 5666678.19 frames. ], batch size: 174, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:18:44,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:19:09,291 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:19:30,415 INFO [train.py:968] (0/2) Epoch 10, batch 8700, giga_loss[loss=0.3442, simple_loss=0.4118, pruned_loss=0.1383, over 28776.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3757, pruned_loss=0.1255, over 5673637.66 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3556, pruned_loss=0.1003, over 5655796.22 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3793, pruned_loss=0.1294, over 5666529.21 frames. ], batch size: 99, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:19:58,912 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.101e+02 1.416e+03 1.838e+03 2.670e+03 7.551e+03, threshold=3.675e+03, percent-clipped=7.0
+2023-03-05 03:20:18,045 INFO [train.py:968] (0/2) Epoch 10, batch 8750, giga_loss[loss=0.3275, simple_loss=0.3969, pruned_loss=0.1291, over 28594.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3799, pruned_loss=0.1258, over 5676009.83 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3557, pruned_loss=0.1004, over 5659295.66 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3829, pruned_loss=0.1291, over 5667648.30 frames. ], batch size: 71, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:20:27,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9230, 3.6991, 3.4934, 1.7470], device='cuda:0'), covar=tensor([0.0772, 0.1040, 0.0983, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.1021, 0.0953, 0.0845, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 03:20:34,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3499, 1.5638, 1.2594, 1.4944], device='cuda:0'), covar=tensor([0.2248, 0.2195, 0.2356, 0.1944], device='cuda:0'), in_proj_covar=tensor([0.1269, 0.0936, 0.1122, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 03:20:38,395 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:20:44,951 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-05 03:21:03,375 INFO [train.py:968] (0/2) Epoch 10, batch 8800, giga_loss[loss=0.3249, simple_loss=0.3796, pruned_loss=0.1351, over 28949.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3811, pruned_loss=0.1262, over 5684414.69 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3556, pruned_loss=0.1006, over 5664929.65 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3846, pruned_loss=0.1297, over 5673125.05 frames. ], batch size: 106, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:21:12,811 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418402.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 03:21:32,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.426e+02 1.589e+03 2.092e+03 2.666e+03 8.345e+03, threshold=4.184e+03, percent-clipped=10.0
+2023-03-05 03:21:38,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:21:47,676 INFO [train.py:968] (0/2) Epoch 10, batch 8850, giga_loss[loss=0.286, simple_loss=0.355, pruned_loss=0.1085, over 28809.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3833, pruned_loss=0.1284, over 5687012.32 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3554, pruned_loss=0.1005, over 5666357.61 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3867, pruned_loss=0.1318, over 5676990.40 frames. ], batch size: 99, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:21:58,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4324, 1.7312, 1.2791, 1.8703], device='cuda:0'), covar=tensor([0.2305, 0.2296, 0.2559, 0.2337], device='cuda:0'), in_proj_covar=tensor([0.1266, 0.0936, 0.1122, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 03:22:04,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4703, 1.6937, 1.4093, 1.7331], device='cuda:0'), covar=tensor([0.1958, 0.1868, 0.1996, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.1267, 0.0936, 0.1122, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 03:22:34,776 INFO [train.py:968] (0/2) Epoch 10, batch 8900, giga_loss[loss=0.3693, simple_loss=0.3995, pruned_loss=0.1695, over 23901.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3838, pruned_loss=0.1295, over 5685171.74 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.355, pruned_loss=0.1001, over 5671289.16 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3873, pruned_loss=0.1329, over 5673459.88 frames. ], batch size: 705, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:22:36,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6967, 1.6661, 1.2549, 1.2467], device='cuda:0'), covar=tensor([0.0688, 0.0570, 0.0960, 0.1006], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0444, 0.0500, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 03:22:47,103 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:22:50,664 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:22:53,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:23:00,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5336, 1.7345, 1.8543, 1.4299], device='cuda:0'), covar=tensor([0.1582, 0.2144, 0.1234, 0.1458], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0699, 0.0845, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 03:23:05,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.493e+03 2.028e+03 2.683e+03 8.818e+03, threshold=4.057e+03, percent-clipped=5.0
+2023-03-05 03:23:22,234 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:23:24,756 INFO [train.py:968] (0/2) Epoch 10, batch 8950, giga_loss[loss=0.2937, simple_loss=0.3572, pruned_loss=0.1151, over 28970.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3836, pruned_loss=0.1304, over 5689913.21 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3551, pruned_loss=0.1002, over 5673641.17 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3866, pruned_loss=0.1333, over 5678742.12 frames. ], batch size: 213, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:23:34,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:24:04,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4778, 1.6114, 1.4646, 1.4486], device='cuda:0'), covar=tensor([0.1205, 0.1498, 0.1845, 0.1515], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0731, 0.0664, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 03:24:16,812 INFO [train.py:968] (0/2) Epoch 10, batch 9000, giga_loss[loss=0.2642, simple_loss=0.3411, pruned_loss=0.09362, over 28902.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.382, pruned_loss=0.1299, over 5689119.67 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3553, pruned_loss=0.1004, over 5677125.42 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3847, pruned_loss=0.1327, over 5677279.44 frames. ], batch size: 174, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:24:16,816 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 03:24:25,362 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2209, simple_loss=0.3273, pruned_loss=0.0572, over 944034.00 frames. 
+2023-03-05 03:24:25,363 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19633MB
+2023-03-05 03:24:55,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.785e+02 1.688e+03 2.114e+03 3.036e+03 6.053e+03, threshold=4.228e+03, percent-clipped=6.0
+2023-03-05 03:25:12,646 INFO [train.py:968] (0/2) Epoch 10, batch 9050, giga_loss[loss=0.3255, simple_loss=0.3858, pruned_loss=0.1326, over 28497.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3793, pruned_loss=0.1289, over 5685207.67 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3555, pruned_loss=0.1005, over 5682583.16 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3818, pruned_loss=0.1316, over 5671222.98 frames. ], batch size: 78, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:26:03,910 INFO [train.py:968] (0/2) Epoch 10, batch 9100, giga_loss[loss=0.3172, simple_loss=0.3802, pruned_loss=0.1271, over 29043.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3772, pruned_loss=0.1277, over 5682216.26 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3553, pruned_loss=0.1004, over 5683560.24 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3795, pruned_loss=0.1301, over 5670487.22 frames. ], batch size: 145, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:26:05,688 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:26:08,764 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:26:29,680 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 03:26:33,441 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:26:35,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.699e+03 2.145e+03 2.819e+03 7.493e+03, threshold=4.291e+03, percent-clipped=10.0
+2023-03-05 03:26:37,914 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:26:54,009 INFO [train.py:968] (0/2) Epoch 10, batch 9150, giga_loss[loss=0.311, simple_loss=0.3691, pruned_loss=0.1265, over 28809.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3788, pruned_loss=0.1293, over 5684234.96 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3561, pruned_loss=0.1012, over 5689335.92 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3804, pruned_loss=0.1311, over 5669710.32 frames. ], batch size: 119, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:26:54,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:27:23,820 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:27:25,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418777.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:27:30,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6439, 1.6534, 1.6953, 1.5258], device='cuda:0'), covar=tensor([0.1353, 0.1830, 0.1841, 0.1773], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0739, 0.0670, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 03:27:36,286 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:27:39,060 INFO [train.py:968] (0/2) Epoch 10, batch 9200, giga_loss[loss=0.3358, simple_loss=0.3753, pruned_loss=0.1482, over 23928.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3767, pruned_loss=0.1283, over 5685367.01 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.356, pruned_loss=0.1011, over 5695763.64 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3789, pruned_loss=0.1308, over 5667534.47 frames. ], batch size: 705, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:27:52,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:27:54,475 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:27:55,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5011, 2.0002, 1.6123, 1.2845], device='cuda:0'), covar=tensor([0.2040, 0.1411, 0.1397, 0.1759], device='cuda:0'), in_proj_covar=tensor([0.1662, 0.1552, 0.1518, 0.1629], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 03:28:05,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.759e+02 1.664e+03 2.235e+03 3.299e+03 6.862e+03, threshold=4.469e+03, percent-clipped=8.0
+2023-03-05 03:28:25,215 INFO [train.py:968] (0/2) Epoch 10, batch 9250, giga_loss[loss=0.2976, simple_loss=0.3621, pruned_loss=0.1166, over 28694.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3755, pruned_loss=0.1276, over 5674369.73 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3559, pruned_loss=0.1011, over 5681885.34 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3779, pruned_loss=0.1304, over 5672046.33 frames. ], batch size: 242, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:28:55,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:29:06,028 INFO [train.py:968] (0/2) Epoch 10, batch 9300, giga_loss[loss=0.3247, simple_loss=0.3928, pruned_loss=0.1283, over 28585.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.376, pruned_loss=0.1271, over 5687794.52 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3557, pruned_loss=0.101, over 5687182.81 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3785, pruned_loss=0.13, over 5680990.32 frames. ], batch size: 307, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:29:32,725 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418920.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:29:36,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418923.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:29:38,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.297e+02 1.358e+03 1.697e+03 2.090e+03 4.887e+03, threshold=3.393e+03, percent-clipped=1.0
+2023-03-05 03:29:49,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 03:29:53,610 INFO [train.py:968] (0/2) Epoch 10, batch 9350, giga_loss[loss=0.4055, simple_loss=0.4402, pruned_loss=0.1854, over 27578.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3784, pruned_loss=0.129, over 5680989.09 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3548, pruned_loss=0.1005, over 5692438.81 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3817, pruned_loss=0.1323, over 5670890.64 frames. ], batch size: 472, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:30:00,080 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:30:01,688 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418952.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:30:02,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:30:29,012 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:30:29,625 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:30:38,818 INFO [train.py:968] (0/2) Epoch 10, batch 9400, giga_loss[loss=0.3468, simple_loss=0.4007, pruned_loss=0.1465, over 28943.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3797, pruned_loss=0.1303, over 5678785.70 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3548, pruned_loss=0.1005, over 5695895.57 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3829, pruned_loss=0.1337, over 5667368.59 frames. ], batch size: 227, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:31:00,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6682, 2.5352, 2.0243, 2.2990], device='cuda:0'), covar=tensor([0.0631, 0.0604, 0.0848, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0440, 0.0495, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 03:31:05,757 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:31:08,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:31:08,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.585e+03 2.253e+03 2.927e+03 5.064e+03, threshold=4.505e+03, percent-clipped=14.0
+2023-03-05 03:31:23,953 INFO [train.py:968] (0/2) Epoch 10, batch 9450, giga_loss[loss=0.2709, simple_loss=0.3602, pruned_loss=0.09082, over 28157.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3804, pruned_loss=0.1299, over 5681264.29 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3545, pruned_loss=0.1004, over 5698939.69 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3835, pruned_loss=0.133, over 5669358.93 frames. ], batch size: 77, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:31:31,693 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:31:44,123 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-05 03:32:04,483 INFO [train.py:968] (0/2) Epoch 10, batch 9500, giga_loss[loss=0.2792, simple_loss=0.3639, pruned_loss=0.09726, over 28888.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.382, pruned_loss=0.1285, over 5685612.23 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3551, pruned_loss=0.1007, over 5704465.60 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3848, pruned_loss=0.1317, over 5670538.15 frames. ], batch size: 145, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:32:08,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:32:19,117 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:32:19,697 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419112.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:32:25,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:32:31,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.243e+02 1.356e+03 1.811e+03 2.398e+03 1.102e+04, threshold=3.622e+03, percent-clipped=8.0
+2023-03-05 03:32:45,292 INFO [train.py:968] (0/2) Epoch 10, batch 9550, giga_loss[loss=0.2993, simple_loss=0.3862, pruned_loss=0.1062, over 28880.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.383, pruned_loss=0.1273, over 5683396.50 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3547, pruned_loss=0.1007, over 5701702.93 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3866, pruned_loss=0.1308, over 5673815.73 frames. ], batch size: 174, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:32:49,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:33:06,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:33:12,388 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:33:20,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:33:30,384 INFO [train.py:968] (0/2) Epoch 10, batch 9600, giga_loss[loss=0.4048, simple_loss=0.4441, pruned_loss=0.1827, over 28558.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3864, pruned_loss=0.1299, over 5684742.47 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3548, pruned_loss=0.1007, over 5707712.20 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.39, pruned_loss=0.1334, over 5671330.85 frames. ], batch size: 336, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:34:02,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.218e+02 1.407e+03 1.995e+03 2.728e+03 4.715e+03, threshold=3.989e+03, percent-clipped=9.0
+2023-03-05 03:34:05,140 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:34:12,294 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:34:14,509 INFO [train.py:968] (0/2) Epoch 10, batch 9650, giga_loss[loss=0.3215, simple_loss=0.3805, pruned_loss=0.1312, over 28920.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3865, pruned_loss=0.1308, over 5688877.89 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3544, pruned_loss=0.1004, over 5713010.66 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3905, pruned_loss=0.1347, over 5672827.45 frames. ], batch size: 199, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:34:14,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:34:34,185 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:34:36,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:34:45,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:04,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:05,233 INFO [train.py:968] (0/2) Epoch 10, batch 9700, giga_loss[loss=0.4113, simple_loss=0.438, pruned_loss=0.1923, over 27524.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3881, pruned_loss=0.1336, over 5678524.15 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3543, pruned_loss=0.1003, over 5715068.66 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3917, pruned_loss=0.1371, over 5663726.46 frames. ], batch size: 472, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:35:05,425 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:06,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:18,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:20,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:30,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:32,938 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:34,492 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:34,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.881e+02 1.690e+03 2.363e+03 3.531e+03 1.367e+04, threshold=4.727e+03, percent-clipped=17.0
+2023-03-05 03:35:38,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:40,667 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:49,323 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:50,594 INFO [train.py:968] (0/2) Epoch 10, batch 9750, giga_loss[loss=0.356, simple_loss=0.3999, pruned_loss=0.156, over 28718.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3869, pruned_loss=0.1329, over 5674372.03 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3542, pruned_loss=0.1001, over 5721758.32 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3911, pruned_loss=0.1371, over 5655106.91 frames. ], batch size: 85, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:36:03,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:36:04,637 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:36:08,536 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:36:09,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2116, 3.0292, 2.8985, 1.5795], device='cuda:0'), covar=tensor([0.0979, 0.1046, 0.0970, 0.2212], device='cuda:0'), in_proj_covar=tensor([0.1027, 0.0968, 0.0853, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 03:36:14,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4793, 1.6887, 1.3351, 1.2392], device='cuda:0'), covar=tensor([0.1831, 0.1604, 0.1391, 0.1595], device='cuda:0'), in_proj_covar=tensor([0.1657, 0.1554, 0.1511, 0.1630], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 03:36:32,752 INFO [train.py:968] (0/2) Epoch 10, batch 9800, libri_loss[loss=0.2683, simple_loss=0.3485, pruned_loss=0.09405, over 25815.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3852, pruned_loss=0.1307, over 5673887.81 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3544, pruned_loss=0.1004, over 5721923.20 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3893, pruned_loss=0.1346, over 5657216.58 frames. ], batch size: 136, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:37:02,549 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.783e+02 1.447e+03 2.142e+03 3.573e+03 8.756e+03, threshold=4.283e+03, percent-clipped=12.0
+2023-03-05 03:37:16,062 INFO [train.py:968] (0/2) Epoch 10, batch 9850, giga_loss[loss=0.3541, simple_loss=0.4048, pruned_loss=0.1516, over 27622.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3844, pruned_loss=0.1283, over 5676410.61 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3542, pruned_loss=0.1003, over 5724659.92 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3885, pruned_loss=0.1322, over 5659686.08 frames. ], batch size: 472, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:37:29,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3939, 1.5667, 1.5476, 1.1273], device='cuda:0'), covar=tensor([0.1813, 0.3298, 0.1601, 0.1709], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0704, 0.0849, 0.0757], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 03:37:50,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:37:51,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419487.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:37:57,204 INFO [train.py:968] (0/2) Epoch 10, batch 9900, giga_loss[loss=0.3352, simple_loss=0.4013, pruned_loss=0.1346, over 28990.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.385, pruned_loss=0.128, over 5678292.95 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3543, pruned_loss=0.1004, over 5721715.63 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3894, pruned_loss=0.1321, over 5665972.33 frames. ], batch size: 128, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:37:59,128 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2316, 2.5524, 1.3022, 1.3508], device='cuda:0'), covar=tensor([0.0898, 0.0383, 0.0860, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0512, 0.0337, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 03:38:01,890 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:38:05,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:38:27,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.965e+02 1.581e+03 2.055e+03 2.928e+03 6.721e+03, threshold=4.110e+03, percent-clipped=3.0
+2023-03-05 03:38:31,484 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:38:44,068 INFO [train.py:968] (0/2) Epoch 10, batch 9950, giga_loss[loss=0.3048, simple_loss=0.374, pruned_loss=0.1178, over 28898.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3857, pruned_loss=0.1294, over 5671541.37 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3535, pruned_loss=0.1001, over 5725959.84 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.391, pruned_loss=0.1339, over 5656413.82 frames. ], batch size: 199, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:38:47,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:38:58,568 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-05 03:39:30,796 INFO [train.py:968] (0/2) Epoch 10, batch 10000, giga_loss[loss=0.3029, simple_loss=0.3652, pruned_loss=0.1203, over 29058.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3852, pruned_loss=0.129, over 5675778.72 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3541, pruned_loss=0.1003, over 5729336.07 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3896, pruned_loss=0.1331, over 5659524.87 frames. ], batch size: 136, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:39:40,705 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 03:39:41,171 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:39:44,499 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:40:03,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.082e+02 1.642e+03 2.275e+03 3.197e+03 5.917e+03, threshold=4.549e+03, percent-clipped=8.0
+2023-03-05 03:40:04,884 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:40:05,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419630.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 03:40:09,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:40:10,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419633.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:40:18,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4405, 1.8797, 1.7437, 1.2809], device='cuda:0'), covar=tensor([0.1603, 0.2149, 0.1328, 0.1646], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0701, 0.0848, 0.0757], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 03:40:20,772 INFO [train.py:968] (0/2) Epoch 10, batch 10050, giga_loss[loss=0.3342, simple_loss=0.398, pruned_loss=0.1351, over 28947.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3837, pruned_loss=0.129, over 5668947.46 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3542, pruned_loss=0.1003, over 5728363.60 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3875, pruned_loss=0.1326, over 5656615.73 frames. ], batch size: 136, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:40:33,877 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.44 vs. limit=5.0
+2023-03-05 03:40:34,317 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:40:35,854 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419662.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:41:04,604 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:06,258 INFO [train.py:968] (0/2) Epoch 10, batch 10100, giga_loss[loss=0.3524, simple_loss=0.3919, pruned_loss=0.1565, over 28701.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3821, pruned_loss=0.129, over 5670307.99 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3541, pruned_loss=0.1003, over 5729709.74 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3861, pruned_loss=0.1328, over 5657658.35 frames. ], batch size: 119, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:41:06,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:10,415 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:20,415 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:37,606 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:42,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.586e+03 2.212e+03 3.065e+03 5.423e+03, threshold=4.424e+03, percent-clipped=4.0
+2023-03-05 03:41:52,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:55,273 INFO [train.py:968] (0/2) Epoch 10, batch 10150, giga_loss[loss=0.4027, simple_loss=0.4341, pruned_loss=0.1856, over 28942.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3807, pruned_loss=0.1282, over 5679156.67 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3548, pruned_loss=0.1005, over 5732922.27 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3841, pruned_loss=0.1319, over 5664252.02 frames. ], batch size: 164, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:41:58,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:42:02,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:42:30,569 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:42:33,285 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:42:39,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4600, 1.6474, 1.3450, 1.7062], device='cuda:0'), covar=tensor([0.2115, 0.2035, 0.2117, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.1267, 0.0940, 0.1119, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 03:42:42,415 INFO [train.py:968] (0/2) Epoch 10, batch 10200, giga_loss[loss=0.3202, simple_loss=0.3795, pruned_loss=0.1304, over 28619.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.38, pruned_loss=0.1288, over 5674482.11 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3553, pruned_loss=0.1009, over 5732318.78 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3826, pruned_loss=0.1319, over 5662088.24 frames. ], batch size: 262, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:43:15,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.960e+02 1.514e+03 1.996e+03 2.744e+03 1.513e+04, threshold=3.991e+03, percent-clipped=5.0
+2023-03-05 03:43:25,281 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:43:29,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:43:29,548 INFO [train.py:968] (0/2) Epoch 10, batch 10250, giga_loss[loss=0.3075, simple_loss=0.3522, pruned_loss=0.1314, over 23648.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3786, pruned_loss=0.1282, over 5669700.16 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3553, pruned_loss=0.1009, over 5731585.12 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.381, pruned_loss=0.131, over 5659591.77 frames. ], batch size: 705, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:43:37,373 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:43:39,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:43:56,374 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:07,784 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:07,827 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:10,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:17,266 INFO [train.py:968] (0/2) Epoch 10, batch 10300, libri_loss[loss=0.2881, simple_loss=0.371, pruned_loss=0.1026, over 29379.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3768, pruned_loss=0.1256, over 5658877.52 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3557, pruned_loss=0.1009, over 5725975.59 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3788, pruned_loss=0.1283, over 5653988.66 frames. ], batch size: 92, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:44:17,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4683, 3.9411, 1.7337, 1.7046], device='cuda:0'), covar=tensor([0.0896, 0.0226, 0.0859, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0508, 0.0333, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 03:44:21,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:36,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:53,721 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.502e+02 1.328e+03 1.756e+03 2.375e+03 5.903e+03, threshold=3.513e+03, percent-clipped=3.0
+2023-03-05 03:45:08,711 INFO [train.py:968] (0/2) Epoch 10, batch 10350, giga_loss[loss=0.2938, simple_loss=0.3606, pruned_loss=0.1135, over 28907.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3741, pruned_loss=0.123, over 5661257.31 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3554, pruned_loss=0.1008, over 5725982.71 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3759, pruned_loss=0.1253, over 5657165.13 frames. ], batch size: 199, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:45:51,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:45:58,932 INFO [train.py:968] (0/2) Epoch 10, batch 10400, giga_loss[loss=0.2874, simple_loss=0.3357, pruned_loss=0.1195, over 23562.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3727, pruned_loss=0.1228, over 5660774.42 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3554, pruned_loss=0.1008, over 5726943.79 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3742, pruned_loss=0.1247, over 5656362.59 frames. ], batch size: 705, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 03:46:04,881 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-420000.pt
+2023-03-05 03:46:29,361 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:46:35,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.053e+02 1.379e+03 2.174e+03 3.560e+03 1.123e+04, threshold=4.348e+03, percent-clipped=25.0
+2023-03-05 03:46:47,404 INFO [train.py:968] (0/2) Epoch 10, batch 10450, giga_loss[loss=0.3107, simple_loss=0.375, pruned_loss=0.1232, over 28621.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3706, pruned_loss=0.1225, over 5663188.33 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3559, pruned_loss=0.1009, over 5725726.95 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3718, pruned_loss=0.1244, over 5659098.84 frames. ], batch size: 336, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:47:32,089 INFO [train.py:968] (0/2) Epoch 10, batch 10500, libri_loss[loss=0.2821, simple_loss=0.3605, pruned_loss=0.1018, over 29512.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.372, pruned_loss=0.1235, over 5674412.22 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.356, pruned_loss=0.1009, over 5730111.24 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3734, pruned_loss=0.1258, over 5664987.13 frames. ], batch size: 82, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:47:55,345 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-05 03:48:03,577 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:48:04,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.314e+02 1.547e+03 1.899e+03 2.509e+03 5.912e+03, threshold=3.798e+03, percent-clipped=5.0
+2023-03-05 03:48:05,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:48:18,695 INFO [train.py:968] (0/2) Epoch 10, batch 10550, giga_loss[loss=0.2724, simple_loss=0.3566, pruned_loss=0.09415, over 28992.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3744, pruned_loss=0.1246, over 5672339.86 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3561, pruned_loss=0.1009, over 5731865.72 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3756, pruned_loss=0.1266, over 5662844.40 frames. ], batch size: 164, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:48:32,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:48:35,383 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:48:36,423 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-05 03:48:41,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-05 03:49:02,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3973, 1.9562, 1.6152, 1.5619], device='cuda:0'), covar=tensor([0.0626, 0.0234, 0.0247, 0.0633], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0055, 0.0049, 0.0084], device='cuda:0')
+2023-03-05 03:49:04,162 INFO [train.py:968] (0/2) Epoch 10, batch 10600, giga_loss[loss=0.3294, simple_loss=0.3877, pruned_loss=0.1356, over 28960.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3743, pruned_loss=0.1238, over 5658310.63 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3565, pruned_loss=0.101, over 5733479.37 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3754, pruned_loss=0.126, over 5647293.67 frames. ], batch size: 213, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:49:37,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.332e+02 1.372e+03 1.847e+03 2.382e+03 4.820e+03, threshold=3.694e+03, percent-clipped=6.0
+2023-03-05 03:49:51,889 INFO [train.py:968] (0/2) Epoch 10, batch 10650, giga_loss[loss=0.3286, simple_loss=0.3671, pruned_loss=0.1451, over 23400.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3745, pruned_loss=0.1243, over 5641162.07 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3568, pruned_loss=0.101, over 5738157.58 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3757, pruned_loss=0.1269, over 5625462.25 frames. ], batch size: 705, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:50:19,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:50:36,636 INFO [train.py:968] (0/2) Epoch 10, batch 10700, giga_loss[loss=0.3234, simple_loss=0.3798, pruned_loss=0.1335, over 28514.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3748, pruned_loss=0.1249, over 5653665.33 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3564, pruned_loss=0.1007, over 5741826.34 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3766, pruned_loss=0.1279, over 5635414.56 frames. ], batch size: 336, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:50:44,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:50:46,287 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:50:47,704 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:51:04,032 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-05 03:51:10,747 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-05 03:51:12,263 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.647e+03 2.189e+03 3.218e+03 1.161e+04, threshold=4.378e+03, percent-clipped=13.0
+2023-03-05 03:51:17,329 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:51:27,551 INFO [train.py:968] (0/2) Epoch 10, batch 10750, giga_loss[loss=0.3072, simple_loss=0.3809, pruned_loss=0.1167, over 28968.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3771, pruned_loss=0.1268, over 5650995.40 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3567, pruned_loss=0.1009, over 5744659.73 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3786, pruned_loss=0.1293, over 5632917.46 frames. ], batch size: 164, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:51:48,800 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:04,744 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9693, 1.8673, 1.4245, 1.4508], device='cuda:0'), covar=tensor([0.0709, 0.0655, 0.0970, 0.1044], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0444, 0.0502, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 03:52:04,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5363, 2.2555, 1.6149, 0.6439], device='cuda:0'), covar=tensor([0.3893, 0.2026, 0.2959, 0.4270], device='cuda:0'), in_proj_covar=tensor([0.1525, 0.1435, 0.1459, 0.1246], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 03:52:09,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:15,892 INFO [train.py:968] (0/2) Epoch 10, batch 10800, giga_loss[loss=0.3419, simple_loss=0.397, pruned_loss=0.1434, over 27986.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.379, pruned_loss=0.1278, over 5655780.22 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.357, pruned_loss=0.101, over 5746910.09 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3803, pruned_loss=0.1301, over 5638379.92 frames. ], batch size: 412, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 03:52:18,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:34,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:36,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:42,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4758, 1.8039, 1.7501, 1.2888], device='cuda:0'), covar=tensor([0.1506, 0.1969, 0.1192, 0.1415], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0707, 0.0854, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 03:52:43,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.152e+02 1.454e+03 1.851e+03 2.537e+03 5.349e+03, threshold=3.703e+03, percent-clipped=7.0
+2023-03-05 03:52:53,623 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:55,516 INFO [train.py:968] (0/2) Epoch 10, batch 10850, libri_loss[loss=0.2885, simple_loss=0.3635, pruned_loss=0.1067, over 29508.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.378, pruned_loss=0.1271, over 5665825.32 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3564, pruned_loss=0.1009, over 5753650.81 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3806, pruned_loss=0.1303, over 5641600.31 frames. ], batch size: 81, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:53:02,630 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:53:45,541 INFO [train.py:968] (0/2) Epoch 10, batch 10900, giga_loss[loss=0.3122, simple_loss=0.3774, pruned_loss=0.1235, over 29028.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3793, pruned_loss=0.1287, over 5668341.19 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3567, pruned_loss=0.1011, over 5756076.43 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3814, pruned_loss=0.1314, over 5645967.70 frames. ], batch size: 155, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:54:18,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.754e+03 2.246e+03 2.887e+03 1.116e+04, threshold=4.491e+03, percent-clipped=11.0
+2023-03-05 03:54:28,719 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:54:30,299 INFO [train.py:968] (0/2) Epoch 10, batch 10950, libri_loss[loss=0.3181, simple_loss=0.3913, pruned_loss=0.1225, over 27892.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3814, pruned_loss=0.1288, over 5665509.49 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3574, pruned_loss=0.1016, over 5755334.95 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3834, pruned_loss=0.1317, over 5644388.18 frames. ], batch size: 116, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:54:30,585 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:54:58,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:55:01,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1999, 1.3792, 3.7530, 3.1749], device='cuda:0'), covar=tensor([0.1509, 0.2437, 0.0424, 0.1676], device='cuda:0'), in_proj_covar=tensor([0.0650, 0.0583, 0.0849, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 03:55:19,218 INFO [train.py:968] (0/2) Epoch 10, batch 11000, giga_loss[loss=0.3068, simple_loss=0.3718, pruned_loss=0.1209, over 28867.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3804, pruned_loss=0.1276, over 5664920.07 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.357, pruned_loss=0.1014, over 5758483.05 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3832, pruned_loss=0.1311, over 5642535.87 frames. ], batch size: 186, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:56:00,132 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.994e+02 1.568e+03 2.249e+03 3.318e+03 1.008e+04, threshold=4.498e+03, percent-clipped=11.0
+2023-03-05 03:56:11,686 INFO [train.py:968] (0/2) Epoch 10, batch 11050, giga_loss[loss=0.2771, simple_loss=0.3474, pruned_loss=0.1034, over 28977.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.378, pruned_loss=0.1266, over 5671349.58 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.357, pruned_loss=0.1013, over 5758923.67 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3805, pruned_loss=0.1297, over 5651796.24 frames. ], batch size: 213, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:56:34,227 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-05 03:56:47,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:56:57,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5069, 3.0400, 1.6406, 1.6548], device='cuda:0'), covar=tensor([0.0735, 0.0321, 0.0654, 0.1000], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0511, 0.0334, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 03:56:59,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9319, 1.1587, 3.2257, 2.8470], device='cuda:0'), covar=tensor([0.1678, 0.2525, 0.0511, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0647, 0.0581, 0.0846, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 03:57:03,260 INFO [train.py:968] (0/2) Epoch 10, batch 11100, libri_loss[loss=0.2447, simple_loss=0.3256, pruned_loss=0.08188, over 29564.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3775, pruned_loss=0.1267, over 5670805.57 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3568, pruned_loss=0.1012, over 5762766.18 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3806, pruned_loss=0.1304, over 5647680.05 frames. ], batch size: 74, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:57:41,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.413e+03 1.996e+03 2.791e+03 8.100e+03, threshold=3.992e+03, percent-clipped=8.0
+2023-03-05 03:57:52,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:57:54,016 INFO [train.py:968] (0/2) Epoch 10, batch 11150, giga_loss[loss=0.2869, simple_loss=0.361, pruned_loss=0.1064, over 29007.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3767, pruned_loss=0.1265, over 5675544.68 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3562, pruned_loss=0.1009, over 5766962.37 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3803, pruned_loss=0.1305, over 5650739.52 frames. ], batch size: 155, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:58:11,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:58:43,008 INFO [train.py:968] (0/2) Epoch 10, batch 11200, libri_loss[loss=0.2595, simple_loss=0.3477, pruned_loss=0.08564, over 29176.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3762, pruned_loss=0.127, over 5679686.39 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3559, pruned_loss=0.1006, over 5768290.88 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.38, pruned_loss=0.1311, over 5656182.47 frames. ], batch size: 101, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 03:59:02,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:59:11,701 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:59:14,011 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:59:17,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.593e+02 1.533e+03 2.020e+03 3.029e+03 7.177e+03, threshold=4.041e+03, percent-clipped=12.0
+2023-03-05 03:59:32,482 INFO [train.py:968] (0/2) Epoch 10, batch 11250, giga_loss[loss=0.2917, simple_loss=0.3625, pruned_loss=0.1105, over 28900.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.376, pruned_loss=0.1274, over 5676708.65 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3555, pruned_loss=0.1004, over 5771207.02 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3796, pruned_loss=0.1314, over 5654040.38 frames. ], batch size: 199, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:59:43,965 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:05,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0498, 1.9251, 1.4311, 1.6214], device='cuda:0'), covar=tensor([0.0732, 0.0720, 0.0968, 0.1087], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0442, 0.0499, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 04:00:11,909 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:15,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:23,281 INFO [train.py:968] (0/2) Epoch 10, batch 11300, giga_loss[loss=0.3023, simple_loss=0.3723, pruned_loss=0.1161, over 29031.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3778, pruned_loss=0.1291, over 5666429.27 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3559, pruned_loss=0.1005, over 5768698.38 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3807, pruned_loss=0.1325, over 5649592.25 frames. ], batch size: 128, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:00:34,390 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:36,317 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:40,175 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:40,426 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-05 04:00:43,368 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:55,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.628e+03 2.103e+03 2.979e+03 7.670e+03, threshold=4.207e+03, percent-clipped=6.0
+2023-03-05 04:01:03,548 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:01:10,528 INFO [train.py:968] (0/2) Epoch 10, batch 11350, giga_loss[loss=0.318, simple_loss=0.3783, pruned_loss=0.1289, over 28769.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.379, pruned_loss=0.1302, over 5675110.51 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3553, pruned_loss=0.1002, over 5770365.23 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3822, pruned_loss=0.1337, over 5658595.11 frames. ], batch size: 243, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:01:25,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:01:27,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:01:53,675 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:01:56,955 INFO [train.py:968] (0/2) Epoch 10, batch 11400, giga_loss[loss=0.3133, simple_loss=0.3717, pruned_loss=0.1274, over 28659.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3807, pruned_loss=0.131, over 5675935.04 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3557, pruned_loss=0.1004, over 5772284.34 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3834, pruned_loss=0.1342, over 5659548.35 frames. ], batch size: 92, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:01:57,721 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:02:38,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.583e+03 2.295e+03 3.375e+03 8.864e+03, threshold=4.591e+03, percent-clipped=15.0
+2023-03-05 04:02:48,479 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 04:02:48,709 INFO [train.py:968] (0/2) Epoch 10, batch 11450, giga_loss[loss=0.3566, simple_loss=0.4087, pruned_loss=0.1522, over 27976.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3834, pruned_loss=0.1342, over 5662607.92 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3558, pruned_loss=0.1005, over 5773455.04 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3857, pruned_loss=0.137, over 5647725.20 frames. ], batch size: 412, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:03:41,453 INFO [train.py:968] (0/2) Epoch 10, batch 11500, giga_loss[loss=0.3228, simple_loss=0.3882, pruned_loss=0.1287, over 28882.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3847, pruned_loss=0.136, over 5654473.30 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3559, pruned_loss=0.1006, over 5764525.77 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3865, pruned_loss=0.1381, over 5650467.25 frames. ], batch size: 145, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:03:52,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:04:18,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.634e+03 2.071e+03 2.694e+03 5.456e+03, threshold=4.143e+03, percent-clipped=2.0
+2023-03-05 04:04:32,954 INFO [train.py:968] (0/2) Epoch 10, batch 11550, giga_loss[loss=0.3044, simple_loss=0.3699, pruned_loss=0.1195, over 28981.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3824, pruned_loss=0.1333, over 5667252.75 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3561, pruned_loss=0.1006, over 5767868.64 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3843, pruned_loss=0.1358, over 5658647.31 frames. ], batch size: 106, lr: 3.28e-03, grad_scale: 2.0
+2023-03-05 04:05:18,271 INFO [train.py:968] (0/2) Epoch 10, batch 11600, giga_loss[loss=0.343, simple_loss=0.3963, pruned_loss=0.1448, over 28571.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3828, pruned_loss=0.1331, over 5669055.56 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3562, pruned_loss=0.1008, over 5770509.80 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3848, pruned_loss=0.1357, over 5657457.29 frames. ], batch size: 307, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:05:25,214 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-05 04:05:56,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.680e+03 2.052e+03 2.840e+03 1.007e+04, threshold=4.104e+03, percent-clipped=8.0
+2023-03-05 04:06:07,053 INFO [train.py:968] (0/2) Epoch 10, batch 11650, giga_loss[loss=0.3441, simple_loss=0.4017, pruned_loss=0.1433, over 29106.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.383, pruned_loss=0.1329, over 5674691.17 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.356, pruned_loss=0.1008, over 5772101.09 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3853, pruned_loss=0.1356, over 5662105.17 frames. ], batch size: 155, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:06:23,145 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-05 04:06:54,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:06:58,695 INFO [train.py:968] (0/2) Epoch 10, batch 11700, giga_loss[loss=0.2912, simple_loss=0.3639, pruned_loss=0.1093, over 29031.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3839, pruned_loss=0.1331, over 5672014.35 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3564, pruned_loss=0.101, over 5763325.65 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3862, pruned_loss=0.136, over 5666747.23 frames. ], batch size: 106, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:07:35,233 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.492e+03 2.262e+03 3.160e+03 5.482e+03, threshold=4.524e+03, percent-clipped=9.0
+2023-03-05 04:07:46,723 INFO [train.py:968] (0/2) Epoch 10, batch 11750, giga_loss[loss=0.3066, simple_loss=0.3703, pruned_loss=0.1214, over 28695.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.384, pruned_loss=0.1334, over 5677337.92 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3564, pruned_loss=0.1009, over 5765219.79 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3862, pruned_loss=0.1362, over 5670264.16 frames. ], batch size: 262, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:08:11,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:08:34,167 INFO [train.py:968] (0/2) Epoch 10, batch 11800, giga_loss[loss=0.3314, simple_loss=0.3928, pruned_loss=0.135, over 28733.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3843, pruned_loss=0.1327, over 5675373.54 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3565, pruned_loss=0.101, over 5756486.40 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3862, pruned_loss=0.1352, over 5675951.28 frames. ], batch size: 99, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:08:44,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6989, 1.6624, 1.2436, 1.3990], device='cuda:0'), covar=tensor([0.0712, 0.0680, 0.0943, 0.1140], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0450, 0.0506, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 04:09:15,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=421432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:09:15,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.856e+02 1.527e+03 2.080e+03 3.078e+03 8.881e+03, threshold=4.160e+03, percent-clipped=11.0
+2023-03-05 04:09:17,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=421435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:09:24,495 INFO [train.py:968] (0/2) Epoch 10, batch 11850, giga_loss[loss=0.2658, simple_loss=0.3453, pruned_loss=0.09313, over 28228.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3846, pruned_loss=0.1319, over 5674239.04 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3568, pruned_loss=0.1013, over 5758828.67 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3863, pruned_loss=0.134, over 5671380.50 frames. ], batch size: 77, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:09:48,827 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=421464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:10:01,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:10:14,196 INFO [train.py:968] (0/2) Epoch 10, batch 11900, giga_loss[loss=0.2753, simple_loss=0.3481, pruned_loss=0.1012, over 28605.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3827, pruned_loss=0.1304, over 5668109.60 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3568, pruned_loss=0.1012, over 5759044.72 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3842, pruned_loss=0.1325, over 5664804.35 frames. ], batch size: 307, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:10:34,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=421513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:10:36,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=421516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:10:42,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2542, 1.2758, 1.2112, 1.5064], device='cuda:0'), covar=tensor([0.0744, 0.0338, 0.0312, 0.0781], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0056, 0.0050, 0.0085], device='cuda:0')
+2023-03-05 04:10:48,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.418e+02 1.393e+03 1.892e+03 2.764e+03 8.259e+03, threshold=3.784e+03, percent-clipped=6.0
+2023-03-05 04:11:00,807 INFO [train.py:968] (0/2) Epoch 10, batch 11950, giga_loss[loss=0.3143, simple_loss=0.3752, pruned_loss=0.1267, over 29053.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3818, pruned_loss=0.1298, over 5674583.28 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3568, pruned_loss=0.1012, over 5752421.14 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3837, pruned_loss=0.1321, over 5675174.89 frames. ], batch size: 106, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:11:01,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=421545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:11:37,696 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:11:49,404 INFO [train.py:968] (0/2) Epoch 10, batch 12000, libri_loss[loss=0.2647, simple_loss=0.3472, pruned_loss=0.09111, over 29555.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3827, pruned_loss=0.1307, over 5666337.67 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3572, pruned_loss=0.1014, over 5757616.28 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3847, pruned_loss=0.1334, over 5659547.12 frames. ], batch size: 78, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 04:11:49,408 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 04:11:57,828 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2216, simple_loss=0.3273, pruned_loss=0.05798, over 944034.00 frames. 
+2023-03-05 04:11:57,829 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 04:12:12,940 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:12:23,325 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:12:27,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=421623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:12:30,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=421626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:12:37,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.877e+02 1.412e+03 1.797e+03 2.510e+03 1.117e+04, threshold=3.594e+03, percent-clipped=7.0
+2023-03-05 04:12:48,666 INFO [train.py:968] (0/2) Epoch 10, batch 12050, giga_loss[loss=0.3726, simple_loss=0.4105, pruned_loss=0.1673, over 27520.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3846, pruned_loss=0.132, over 5673408.81 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3572, pruned_loss=0.1015, over 5759167.70 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3863, pruned_loss=0.1342, over 5665845.05 frames. ], batch size: 472, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 04:12:49,645 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421645.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 04:12:58,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1509, 1.0100, 0.9327, 1.3371], device='cuda:0'), covar=tensor([0.0760, 0.0314, 0.0326, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0077, 0.0056, 0.0050, 0.0085], device='cuda:0')
+2023-03-05 04:12:59,134 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=421655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:13:37,443 INFO [train.py:968] (0/2) Epoch 10, batch 12100, giga_loss[loss=0.3933, simple_loss=0.4195, pruned_loss=0.1835, over 26651.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3854, pruned_loss=0.1343, over 5667335.62 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3578, pruned_loss=0.1019, over 5759191.23 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.387, pruned_loss=0.1365, over 5659055.71 frames. ], batch size: 555, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 04:13:46,990 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:13:59,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:14:18,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.611e+02 1.486e+03 1.993e+03 2.623e+03 7.028e+03, threshold=3.986e+03, percent-clipped=14.0
+2023-03-05 04:14:28,020 INFO [train.py:968] (0/2) Epoch 10, batch 12150, giga_loss[loss=0.3634, simple_loss=0.4063, pruned_loss=0.1602, over 28859.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3847, pruned_loss=0.1342, over 5659576.22 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3579, pruned_loss=0.1019, over 5751494.85 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3864, pruned_loss=0.1364, over 5657846.54 frames. ], batch size: 227, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:15:08,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2007, 1.2103, 1.0268, 0.9075], device='cuda:0'), covar=tensor([0.0546, 0.0349, 0.0733, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0446, 0.0503, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 04:15:17,174 INFO [train.py:968] (0/2) Epoch 10, batch 12200, giga_loss[loss=0.338, simple_loss=0.3968, pruned_loss=0.1396, over 28870.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3862, pruned_loss=0.1355, over 5664581.23 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3576, pruned_loss=0.1017, over 5753597.50 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.388, pruned_loss=0.1379, over 5660221.76 frames. ], batch size: 136, lr: 3.28e-03, grad_scale: 2.0
+2023-03-05 04:15:31,403 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 04:15:58,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.308e+02 1.546e+03 2.033e+03 2.794e+03 1.069e+04, threshold=4.067e+03, percent-clipped=12.0
+2023-03-05 04:16:06,260 INFO [train.py:968] (0/2) Epoch 10, batch 12250, giga_loss[loss=0.3749, simple_loss=0.4184, pruned_loss=0.1657, over 28733.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3864, pruned_loss=0.1356, over 5660640.56 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3576, pruned_loss=0.1018, over 5756462.21 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3885, pruned_loss=0.1381, over 5652800.20 frames. ], batch size: 284, lr: 3.28e-03, grad_scale: 2.0
+2023-03-05 04:16:20,521 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 04:16:53,523 INFO [train.py:968] (0/2) Epoch 10, batch 12300, giga_loss[loss=0.4311, simple_loss=0.4364, pruned_loss=0.2129, over 23777.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3846, pruned_loss=0.1343, over 5655659.74 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3575, pruned_loss=0.1017, over 5761212.25 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3874, pruned_loss=0.1376, over 5641429.40 frames. ], batch size: 705, lr: 3.28e-03, grad_scale: 2.0
+2023-03-05 04:17:34,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.309e+02 1.527e+03 2.097e+03 3.053e+03 1.721e+04, threshold=4.194e+03, percent-clipped=13.0
+2023-03-05 04:17:40,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2327, 1.4949, 1.2318, 1.3266], device='cuda:0'), covar=tensor([0.1820, 0.1761, 0.1944, 0.1489], device='cuda:0'), in_proj_covar=tensor([0.1270, 0.0943, 0.1121, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 04:17:42,784 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:17:43,219 INFO [train.py:968] (0/2) Epoch 10, batch 12350, giga_loss[loss=0.4078, simple_loss=0.4305, pruned_loss=0.1925, over 26661.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3854, pruned_loss=0.1348, over 5646355.67 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3578, pruned_loss=0.102, over 5755168.55 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.388, pruned_loss=0.138, over 5638550.76 frames. ], batch size: 555, lr: 3.28e-03, grad_scale: 2.0
+2023-03-05 04:17:58,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:18:23,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:18:27,718 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-05 04:18:30,484 INFO [train.py:968] (0/2) Epoch 10, batch 12400, giga_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 28644.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3857, pruned_loss=0.1346, over 5652876.13 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3577, pruned_loss=0.102, over 5758281.08 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3884, pruned_loss=0.1378, over 5641659.34 frames. ], batch size: 262, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:18:30,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:18:34,566 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-422000.pt
+2023-03-05 04:18:37,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3194, 1.6618, 1.2974, 1.2447], device='cuda:0'), covar=tensor([0.2060, 0.1925, 0.2120, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.1269, 0.0943, 0.1120, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 04:18:51,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422020.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 04:19:05,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.490e+02 1.376e+03 1.769e+03 2.339e+03 1.217e+04, threshold=3.539e+03, percent-clipped=7.0
+2023-03-05 04:19:14,030 INFO [train.py:968] (0/2) Epoch 10, batch 12450, giga_loss[loss=0.3032, simple_loss=0.3684, pruned_loss=0.119, over 29014.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3836, pruned_loss=0.1326, over 5650965.64 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3576, pruned_loss=0.1018, over 5752321.70 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3872, pruned_loss=0.1368, over 5641686.42 frames. ], batch size: 128, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:19:45,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:19:53,191 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 04:19:56,056 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:19:56,161 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4480, 1.8624, 1.3409, 2.0536], device='cuda:0'), covar=tensor([0.2578, 0.2499, 0.2742, 0.2327], device='cuda:0'), in_proj_covar=tensor([0.1273, 0.0945, 0.1124, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 04:19:57,786 INFO [train.py:968] (0/2) Epoch 10, batch 12500, giga_loss[loss=0.3131, simple_loss=0.3743, pruned_loss=0.126, over 28916.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3813, pruned_loss=0.1309, over 5667648.45 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.357, pruned_loss=0.1014, over 5758442.80 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3857, pruned_loss=0.1358, over 5651039.66 frames. ], batch size: 112, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:20:05,260 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:07,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:32,496 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:35,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:37,166 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:37,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.756e+02 1.661e+03 2.114e+03 2.814e+03 9.190e+03, threshold=4.229e+03, percent-clipped=12.0
+2023-03-05 04:20:39,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:41,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:43,993 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3664, 1.6639, 1.6779, 1.2201], device='cuda:0'), covar=tensor([0.1551, 0.2186, 0.1243, 0.1449], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0704, 0.0849, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 04:20:44,962 INFO [train.py:968] (0/2) Epoch 10, batch 12550, libri_loss[loss=0.2639, simple_loss=0.3418, pruned_loss=0.09297, over 29568.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3781, pruned_loss=0.1287, over 5677537.14 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3569, pruned_loss=0.1013, over 5762743.26 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3824, pruned_loss=0.1334, over 5657798.04 frames. ], batch size: 76, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:20:58,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:21:01,567 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422163.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 04:21:04,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=422165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:21:04,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422166.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 04:21:07,926 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:21:29,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1688, 2.5368, 1.2252, 1.3567], device='cuda:0'), covar=tensor([0.0934, 0.0355, 0.0833, 0.1258], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0511, 0.0336, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 04:21:32,674 INFO [train.py:968] (0/2) Epoch 10, batch 12600, giga_loss[loss=0.3028, simple_loss=0.3446, pruned_loss=0.1306, over 23687.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.375, pruned_loss=0.1271, over 5661290.32 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3567, pruned_loss=0.1012, over 5755892.19 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3788, pruned_loss=0.1315, over 5650450.78 frames. ], batch size: 705, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:21:33,631 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422195.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 04:22:01,647 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:04,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:04,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5118, 1.7531, 1.8172, 1.3494], device='cuda:0'), covar=tensor([0.1569, 0.2059, 0.1220, 0.1412], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0704, 0.0849, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 04:22:14,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:14,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.117e+02 1.713e+03 2.287e+03 3.540e+03 9.174e+03, threshold=4.575e+03, percent-clipped=11.0
+2023-03-05 04:22:16,138 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:22,072 INFO [train.py:968] (0/2) Epoch 10, batch 12650, giga_loss[loss=0.2958, simple_loss=0.3467, pruned_loss=0.1225, over 28101.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3746, pruned_loss=0.128, over 5658618.36 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3571, pruned_loss=0.1013, over 5759256.82 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3777, pruned_loss=0.132, over 5644628.03 frames. ], batch size: 77, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:22:28,414 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 04:22:29,794 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:43,236 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:52,669 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=422275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:23:10,515 INFO [train.py:968] (0/2) Epoch 10, batch 12700, giga_loss[loss=0.3517, simple_loss=0.4052, pruned_loss=0.1491, over 28833.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3745, pruned_loss=0.1287, over 5656511.89 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3576, pruned_loss=0.1018, over 5761839.73 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3769, pruned_loss=0.132, over 5641206.55 frames. ], batch size: 199, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:23:33,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=422317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:23:34,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:23:51,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.464e+02 1.798e+03 2.629e+03 3.797e+03 1.056e+04, threshold=5.258e+03, percent-clipped=16.0
+2023-03-05 04:23:57,808 INFO [train.py:968] (0/2) Epoch 10, batch 12750, giga_loss[loss=0.274, simple_loss=0.3477, pruned_loss=0.1002, over 28910.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3736, pruned_loss=0.1273, over 5659306.89 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3578, pruned_loss=0.1017, over 5762556.41 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3757, pruned_loss=0.1308, over 5643122.22 frames. ], batch size: 174, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:24:10,871 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=422355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:24:48,990 INFO [train.py:968] (0/2) Epoch 10, batch 12800, giga_loss[loss=0.384, simple_loss=0.4356, pruned_loss=0.1662, over 28849.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3724, pruned_loss=0.1251, over 5657897.48 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3573, pruned_loss=0.1015, over 5765273.61 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.375, pruned_loss=0.1285, over 5640741.01 frames. ], batch size: 186, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:24:49,411 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.89 vs. limit=5.0
+2023-03-05 04:25:05,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=422410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:25:33,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.966e+02 1.363e+03 1.837e+03 2.520e+03 5.967e+03, threshold=3.675e+03, percent-clipped=3.0
+2023-03-05 04:25:42,732 INFO [train.py:968] (0/2) Epoch 10, batch 12850, giga_loss[loss=0.2741, simple_loss=0.3502, pruned_loss=0.09903, over 28874.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.1209, over 5654274.10 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3571, pruned_loss=0.1014, over 5766803.40 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3715, pruned_loss=0.1238, over 5638585.41 frames. ], batch size: 174, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:25:50,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0681, 1.2743, 3.3197, 3.0095], device='cuda:0'), covar=tensor([0.1612, 0.2484, 0.0480, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0653, 0.0585, 0.0853, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 04:25:59,927 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:26:04,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:26:34,404 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:26:34,807 INFO [train.py:968] (0/2) Epoch 10, batch 12900, giga_loss[loss=0.2478, simple_loss=0.3298, pruned_loss=0.08293, over 28985.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3664, pruned_loss=0.1175, over 5655481.95 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3573, pruned_loss=0.1017, over 5768701.94 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3682, pruned_loss=0.1199, over 5639205.45 frames. ], batch size: 128, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:26:55,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-05 04:27:17,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.279e+02 1.278e+03 1.605e+03 2.231e+03 9.378e+03, threshold=3.209e+03, percent-clipped=4.0
+2023-03-05 04:27:21,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4122, 1.6259, 1.5282, 1.4792], device='cuda:0'), covar=tensor([0.1262, 0.1606, 0.1657, 0.1520], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0722, 0.0658, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 04:27:22,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:27:24,997 INFO [train.py:968] (0/2) Epoch 10, batch 12950, giga_loss[loss=0.2613, simple_loss=0.3453, pruned_loss=0.08861, over 29057.00 frames. ], tot_loss[loss=0.295, simple_loss=0.362, pruned_loss=0.114, over 5651473.68 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3563, pruned_loss=0.1012, over 5770534.58 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3647, pruned_loss=0.1169, over 5632537.68 frames. ], batch size: 155, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:27:42,336 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-05 04:28:13,263 INFO [train.py:968] (0/2) Epoch 10, batch 13000, giga_loss[loss=0.2555, simple_loss=0.345, pruned_loss=0.083, over 28886.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.359, pruned_loss=0.1097, over 5658170.08 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3557, pruned_loss=0.1011, over 5770559.65 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3619, pruned_loss=0.1125, over 5638888.44 frames. ], batch size: 199, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:28:39,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7491, 1.8542, 1.2889, 1.5371], device='cuda:0'), covar=tensor([0.0709, 0.0513, 0.0972, 0.0956], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0439, 0.0500, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 04:28:53,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.780e+02 1.340e+03 1.710e+03 2.309e+03 4.232e+03, threshold=3.420e+03, percent-clipped=9.0
+2023-03-05 04:29:02,687 INFO [train.py:968] (0/2) Epoch 10, batch 13050, giga_loss[loss=0.2777, simple_loss=0.3301, pruned_loss=0.1126, over 24263.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3589, pruned_loss=0.1084, over 5665485.25 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3551, pruned_loss=0.1009, over 5774241.90 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3619, pruned_loss=0.1111, over 5643945.86 frames. ], batch size: 705, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:29:10,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:29:42,362 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:29:45,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:29:51,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:29:52,693 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-05 04:29:52,860 INFO [train.py:968] (0/2) Epoch 10, batch 13100, giga_loss[loss=0.2609, simple_loss=0.3424, pruned_loss=0.08973, over 28629.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3586, pruned_loss=0.1082, over 5665121.84 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3547, pruned_loss=0.1008, over 5777998.06 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3614, pruned_loss=0.1106, over 5641424.53 frames. ], batch size: 307, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:30:07,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5430, 1.9124, 1.8844, 1.3991], device='cuda:0'), covar=tensor([0.1719, 0.2161, 0.1370, 0.1641], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0692, 0.0839, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 04:30:11,205 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:30:23,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:30:26,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.650e+02 1.276e+03 1.570e+03 2.024e+03 5.438e+03, threshold=3.139e+03, percent-clipped=3.0
+2023-03-05 04:30:37,220 INFO [train.py:968] (0/2) Epoch 10, batch 13150, giga_loss[loss=0.2669, simple_loss=0.3434, pruned_loss=0.09524, over 28841.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3552, pruned_loss=0.1058, over 5660418.24 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3536, pruned_loss=0.1006, over 5765963.19 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3586, pruned_loss=0.1083, over 5646543.91 frames. ], batch size: 106, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:31:16,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:31:25,834 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:31:26,098 INFO [train.py:968] (0/2) Epoch 10, batch 13200, giga_loss[loss=0.27, simple_loss=0.3426, pruned_loss=0.09873, over 28249.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3524, pruned_loss=0.1045, over 5638842.36 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.353, pruned_loss=0.1003, over 5757501.95 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3556, pruned_loss=0.1069, over 5632492.40 frames. ], batch size: 368, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:31:29,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:31:59,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:32:08,799 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:32:09,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.658e+02 1.330e+03 1.642e+03 2.311e+03 7.505e+03, threshold=3.285e+03, percent-clipped=13.0
+2023-03-05 04:32:11,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:32:17,724 INFO [train.py:968] (0/2) Epoch 10, batch 13250, giga_loss[loss=0.311, simple_loss=0.3645, pruned_loss=0.1287, over 26723.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3529, pruned_loss=0.1051, over 5640039.90 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3526, pruned_loss=0.1002, over 5759360.27 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3559, pruned_loss=0.1072, over 5631810.62 frames. ], batch size: 555, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:32:29,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5853, 4.0643, 1.6196, 1.6697], device='cuda:0'), covar=tensor([0.0845, 0.0239, 0.0858, 0.1193], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0507, 0.0334, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 04:32:42,587 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:32:48,250 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:32:51,587 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:33:09,109 INFO [train.py:968] (0/2) Epoch 10, batch 13300, giga_loss[loss=0.2513, simple_loss=0.3351, pruned_loss=0.08374, over 28943.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3519, pruned_loss=0.104, over 5644698.29 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3527, pruned_loss=0.1003, over 5761572.90 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3542, pruned_loss=0.1056, over 5634454.16 frames. ], batch size: 145, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:33:21,307 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:33:43,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:33:48,453 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:33:53,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.902e+02 1.270e+03 1.719e+03 2.371e+03 5.806e+03, threshold=3.438e+03, percent-clipped=13.0
+2023-03-05 04:34:02,473 INFO [train.py:968] (0/2) Epoch 10, batch 13350, giga_loss[loss=0.3117, simple_loss=0.3612, pruned_loss=0.1311, over 26569.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3489, pruned_loss=0.1014, over 5641146.58 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3521, pruned_loss=0.1, over 5757374.49 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3512, pruned_loss=0.1029, over 5634235.48 frames. ], batch size: 555, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:34:19,207 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:34:46,067 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-05 04:34:52,640 INFO [train.py:968] (0/2) Epoch 10, batch 13400, libri_loss[loss=0.239, simple_loss=0.3112, pruned_loss=0.08342, over 29578.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3451, pruned_loss=0.09855, over 5648349.33 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3515, pruned_loss=0.09982, over 5758815.32 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3474, pruned_loss=0.09998, over 5638960.57 frames. ], batch size: 76, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:35:37,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.573e+02 1.188e+03 1.643e+03 2.428e+03 6.004e+03, threshold=3.286e+03, percent-clipped=8.0
+2023-03-05 04:35:42,039 INFO [train.py:968] (0/2) Epoch 10, batch 13450, giga_loss[loss=0.2608, simple_loss=0.3321, pruned_loss=0.09472, over 27936.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3409, pruned_loss=0.0963, over 5655166.82 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3505, pruned_loss=0.0995, over 5762802.49 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3433, pruned_loss=0.09767, over 5639504.36 frames. ], batch size: 412, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:36:40,064 INFO [train.py:968] (0/2) Epoch 10, batch 13500, giga_loss[loss=0.3059, simple_loss=0.3526, pruned_loss=0.1296, over 26721.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3401, pruned_loss=0.09632, over 5656190.36 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3503, pruned_loss=0.09937, over 5762862.99 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3421, pruned_loss=0.09748, over 5642437.33 frames. ], batch size: 555, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:36:42,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4583, 1.8053, 1.4733, 1.3454], device='cuda:0'), covar=tensor([0.1692, 0.1274, 0.1097, 0.1414], device='cuda:0'), in_proj_covar=tensor([0.1616, 0.1503, 0.1467, 0.1569], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 04:37:24,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.908e+02 1.349e+03 1.789e+03 2.574e+03 6.264e+03, threshold=3.578e+03, percent-clipped=13.0
+2023-03-05 04:37:33,895 INFO [train.py:968] (0/2) Epoch 10, batch 13550, giga_loss[loss=0.2765, simple_loss=0.3404, pruned_loss=0.1062, over 26684.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3397, pruned_loss=0.0963, over 5651922.82 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3498, pruned_loss=0.09929, over 5767170.77 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3414, pruned_loss=0.09724, over 5634075.30 frames. ], batch size: 555, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:38:26,566 INFO [train.py:968] (0/2) Epoch 10, batch 13600, libri_loss[loss=0.2589, simple_loss=0.3322, pruned_loss=0.09282, over 29562.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3432, pruned_loss=0.09729, over 5666036.69 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3496, pruned_loss=0.09919, over 5772036.08 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3446, pruned_loss=0.09806, over 5643979.10 frames. ], batch size: 77, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:39:02,677 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=423223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:39:18,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.164e+02 1.412e+03 1.688e+03 2.364e+03 4.528e+03, threshold=3.376e+03, percent-clipped=5.0
+2023-03-05 04:39:25,372 INFO [train.py:968] (0/2) Epoch 10, batch 13650, giga_loss[loss=0.2208, simple_loss=0.3059, pruned_loss=0.06789, over 28429.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3448, pruned_loss=0.09716, over 5666674.94 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3494, pruned_loss=0.09912, over 5773725.95 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3459, pruned_loss=0.09779, over 5644121.72 frames. ], batch size: 65, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:39:38,015 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=423255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:40:23,804 INFO [train.py:968] (0/2) Epoch 10, batch 13700, giga_loss[loss=0.3012, simple_loss=0.3657, pruned_loss=0.1184, over 28366.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3447, pruned_loss=0.09715, over 5677179.28 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3489, pruned_loss=0.09901, over 5775626.91 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.346, pruned_loss=0.09772, over 5656206.32 frames. ], batch size: 368, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:41:16,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.501e+02 1.443e+03 1.718e+03 2.187e+03 4.057e+03, threshold=3.435e+03, percent-clipped=4.0
+2023-03-05 04:41:23,524 INFO [train.py:968] (0/2) Epoch 10, batch 13750, giga_loss[loss=0.2869, simple_loss=0.3583, pruned_loss=0.1078, over 28663.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3423, pruned_loss=0.0956, over 5676791.45 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.349, pruned_loss=0.09901, over 5776559.08 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3433, pruned_loss=0.09601, over 5657502.47 frames. ], batch size: 307, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:42:25,800 INFO [train.py:968] (0/2) Epoch 10, batch 13800, libri_loss[loss=0.2824, simple_loss=0.3606, pruned_loss=0.1021, over 28698.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.342, pruned_loss=0.09415, over 5672538.42 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3488, pruned_loss=0.09894, over 5773108.61 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3426, pruned_loss=0.09441, over 5657359.84 frames. ], batch size: 106, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:43:15,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.821e+02 1.237e+03 1.501e+03 2.022e+03 4.353e+03, threshold=3.001e+03, percent-clipped=5.0
+2023-03-05 04:43:22,586 INFO [train.py:968] (0/2) Epoch 10, batch 13850, giga_loss[loss=0.2759, simple_loss=0.3473, pruned_loss=0.1022, over 28640.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3408, pruned_loss=0.09361, over 5661371.50 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3488, pruned_loss=0.09918, over 5764220.13 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.341, pruned_loss=0.09338, over 5653964.49 frames. ], batch size: 307, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:43:47,750 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-05 04:43:51,461 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=423468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:43:59,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0420, 5.8231, 5.5139, 3.2156], device='cuda:0'), covar=tensor([0.0455, 0.0633, 0.0731, 0.1454], device='cuda:0'), in_proj_covar=tensor([0.1001, 0.0936, 0.0819, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 04:44:20,193 INFO [train.py:968] (0/2) Epoch 10, batch 13900, giga_loss[loss=0.2673, simple_loss=0.3505, pruned_loss=0.09208, over 28814.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3374, pruned_loss=0.09262, over 5667835.58 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3474, pruned_loss=0.09846, over 5761579.96 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3384, pruned_loss=0.09283, over 5659081.05 frames. ], batch size: 174, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:44:20,996 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 04:45:10,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.549e+02 1.410e+03 1.722e+03 2.071e+03 8.478e+03, threshold=3.444e+03, percent-clipped=10.0
+2023-03-05 04:45:18,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0269, 3.8483, 3.6306, 1.8992], device='cuda:0'), covar=tensor([0.0528, 0.0680, 0.0763, 0.2364], device='cuda:0'), in_proj_covar=tensor([0.1003, 0.0939, 0.0822, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 04:45:18,669 INFO [train.py:968] (0/2) Epoch 10, batch 13950, giga_loss[loss=0.3126, simple_loss=0.363, pruned_loss=0.1311, over 27675.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.337, pruned_loss=0.09295, over 5660184.07 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3476, pruned_loss=0.09872, over 5754681.66 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3374, pruned_loss=0.09273, over 5656447.64 frames. ], batch size: 472, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:46:15,222 INFO [train.py:968] (0/2) Epoch 10, batch 14000, libri_loss[loss=0.2556, simple_loss=0.3409, pruned_loss=0.08516, over 29470.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3361, pruned_loss=0.09219, over 5659783.60 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3471, pruned_loss=0.09838, over 5754232.75 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3366, pruned_loss=0.0922, over 5655619.77 frames. ], batch size: 85, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:46:20,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=423598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:47:01,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=423630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:47:11,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.397e+02 1.378e+03 1.940e+03 3.068e+03 9.866e+03, threshold=3.880e+03, percent-clipped=19.0
+2023-03-05 04:47:19,356 INFO [train.py:968] (0/2) Epoch 10, batch 14050, giga_loss[loss=0.2746, simple_loss=0.3555, pruned_loss=0.09684, over 28633.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3384, pruned_loss=0.09296, over 5652770.03 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3472, pruned_loss=0.09835, over 5756782.46 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3387, pruned_loss=0.09291, over 5645769.57 frames. ], batch size: 307, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:48:24,367 INFO [train.py:968] (0/2) Epoch 10, batch 14100, giga_loss[loss=0.2463, simple_loss=0.326, pruned_loss=0.0833, over 28639.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3372, pruned_loss=0.09172, over 5662466.78 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3467, pruned_loss=0.09811, over 5758516.97 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3376, pruned_loss=0.09176, over 5653147.03 frames. ], batch size: 307, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:48:36,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6772, 1.8581, 1.4266, 2.0331], device='cuda:0'), covar=tensor([0.2362, 0.2333, 0.2601, 0.2244], device='cuda:0'), in_proj_covar=tensor([0.1266, 0.0937, 0.1127, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 04:49:16,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.988e+02 1.350e+03 1.898e+03 2.779e+03 7.429e+03, threshold=3.796e+03, percent-clipped=16.0
+2023-03-05 04:49:22,333 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=423741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:49:22,866 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=423742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:49:26,848 INFO [train.py:968] (0/2) Epoch 10, batch 14150, giga_loss[loss=0.2884, simple_loss=0.3513, pruned_loss=0.1128, over 27613.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.336, pruned_loss=0.09144, over 5668489.55 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3461, pruned_loss=0.09795, over 5755851.58 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3364, pruned_loss=0.0913, over 5658494.93 frames. ], batch size: 472, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:49:27,144 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=423744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:49:49,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1272, 1.1115, 3.7711, 3.1215], device='cuda:0'), covar=tensor([0.2035, 0.3072, 0.0715, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0637, 0.0576, 0.0833, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 04:50:02,008 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=423773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:50:02,080 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=423773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:50:04,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=423776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:50:09,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3367, 3.4437, 1.5479, 1.5104], device='cuda:0'), covar=tensor([0.0917, 0.0273, 0.0851, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0500, 0.0333, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 04:50:31,190 INFO [train.py:968] (0/2) Epoch 10, batch 14200, giga_loss[loss=0.324, simple_loss=0.3869, pruned_loss=0.1305, over 26919.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3388, pruned_loss=0.09286, over 5683496.43 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3458, pruned_loss=0.09788, over 5759217.25 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3392, pruned_loss=0.0927, over 5671189.69 frames. ], batch size: 555, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:50:31,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9698, 2.7417, 1.7969, 0.8236], device='cuda:0'), covar=tensor([0.4603, 0.2360, 0.3075, 0.4719], device='cuda:0'), in_proj_covar=tensor([0.1534, 0.1452, 0.1476, 0.1264], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 04:50:48,845 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=423805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:51:30,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.849e+02 1.358e+03 1.777e+03 2.624e+03 1.093e+04, threshold=3.554e+03, percent-clipped=11.0
+2023-03-05 04:51:36,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=423843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:51:36,644 INFO [train.py:968] (0/2) Epoch 10, batch 14250, giga_loss[loss=0.3213, simple_loss=0.3899, pruned_loss=0.1263, over 27704.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3423, pruned_loss=0.09272, over 5677474.96 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3453, pruned_loss=0.09757, over 5761870.22 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3431, pruned_loss=0.0928, over 5663555.35 frames. ], batch size: 474, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:51:43,858 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-05 04:52:32,591 INFO [train.py:968] (0/2) Epoch 10, batch 14300, libri_loss[loss=0.2957, simple_loss=0.3633, pruned_loss=0.1141, over 19304.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3438, pruned_loss=0.09204, over 5675439.66 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3449, pruned_loss=0.09761, over 5756368.84 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3447, pruned_loss=0.09185, over 5666589.49 frames. ], batch size: 186, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:53:22,206 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-03-05 04:53:30,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.426e+02 1.388e+03 1.718e+03 2.472e+03 8.953e+03, threshold=3.436e+03, percent-clipped=8.0
+2023-03-05 04:53:34,984 INFO [train.py:968] (0/2) Epoch 10, batch 14350, giga_loss[loss=0.2747, simple_loss=0.3543, pruned_loss=0.09756, over 28472.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3437, pruned_loss=0.0907, over 5671951.41 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3446, pruned_loss=0.09753, over 5757758.01 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3446, pruned_loss=0.09056, over 5662893.21 frames. ], batch size: 336, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:54:26,490 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=423986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:54:30,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=423989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:54:36,999 INFO [train.py:968] (0/2) Epoch 10, batch 14400, giga_loss[loss=0.2758, simple_loss=0.354, pruned_loss=0.09874, over 28562.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3429, pruned_loss=0.09037, over 5674743.06 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3441, pruned_loss=0.09725, over 5761337.28 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09031, over 5661739.86 frames. ], batch size: 307, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:54:45,800 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-424000.pt
+2023-03-05 04:55:04,692 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=424018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:55:30,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.083e+02 1.181e+03 1.446e+03 2.041e+03 4.953e+03, threshold=2.891e+03, percent-clipped=3.0
+2023-03-05 04:55:33,819 INFO [train.py:968] (0/2) Epoch 10, batch 14450, giga_loss[loss=0.2565, simple_loss=0.3331, pruned_loss=0.09, over 29021.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3418, pruned_loss=0.09074, over 5679601.07 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.344, pruned_loss=0.09721, over 5761111.58 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3429, pruned_loss=0.09052, over 5666491.57 frames. ], batch size: 136, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:56:41,666 INFO [train.py:968] (0/2) Epoch 10, batch 14500, giga_loss[loss=0.3521, simple_loss=0.3995, pruned_loss=0.1523, over 28800.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3424, pruned_loss=0.09188, over 5694345.52 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.344, pruned_loss=0.09733, over 5765237.02 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.343, pruned_loss=0.09143, over 5678390.90 frames. ], batch size: 243, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:56:52,744 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0286, 2.6492, 2.0828, 1.4567], device='cuda:0'), covar=tensor([0.3413, 0.2071, 0.2006, 0.3499], device='cuda:0'), in_proj_covar=tensor([0.1522, 0.1445, 0.1472, 0.1253], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 04:57:13,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=424117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:57:43,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.579e+02 1.288e+03 1.630e+03 2.284e+03 5.179e+03, threshold=3.259e+03, percent-clipped=6.0
+2023-03-05 04:57:53,545 INFO [train.py:968] (0/2) Epoch 10, batch 14550, giga_loss[loss=0.2496, simple_loss=0.3294, pruned_loss=0.08486, over 28254.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.343, pruned_loss=0.09339, over 5678412.70 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.344, pruned_loss=0.09735, over 5749351.92 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3436, pruned_loss=0.09283, over 5676413.88 frames. ], batch size: 412, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:58:24,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3380, 3.0738, 1.5197, 1.4581], device='cuda:0'), covar=tensor([0.0885, 0.0278, 0.0840, 0.1212], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0500, 0.0333, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 04:58:27,262 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-05 04:59:03,312 INFO [train.py:968] (0/2) Epoch 10, batch 14600, libri_loss[loss=0.2425, simple_loss=0.3216, pruned_loss=0.08175, over 29550.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3389, pruned_loss=0.0912, over 5683838.71 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3433, pruned_loss=0.09703, over 5755840.75 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3398, pruned_loss=0.09082, over 5673196.61 frames. ], batch size: 83, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:59:05,931 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-05 05:00:01,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.494e+02 1.178e+03 1.533e+03 2.319e+03 7.885e+03, threshold=3.065e+03, percent-clipped=11.0
+2023-03-05 05:00:10,842 INFO [train.py:968] (0/2) Epoch 10, batch 14650, giga_loss[loss=0.2239, simple_loss=0.2901, pruned_loss=0.07885, over 24134.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3375, pruned_loss=0.09038, over 5683235.22 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3434, pruned_loss=0.09712, over 5756543.40 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3381, pruned_loss=0.08989, over 5672703.99 frames. ], batch size: 705, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:00:33,997 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=424260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:00:36,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=424263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:01:10,353 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=424292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:01:11,469 INFO [train.py:968] (0/2) Epoch 10, batch 14700, libri_loss[loss=0.2453, simple_loss=0.3265, pruned_loss=0.08208, over 28556.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3366, pruned_loss=0.09057, over 5682712.80 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3428, pruned_loss=0.09677, over 5759914.67 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3374, pruned_loss=0.09023, over 5668224.91 frames. ], batch size: 106, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:01:53,933 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-05 05:02:07,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.443e+02 1.285e+03 2.200e+03 2.871e+03 7.315e+03, threshold=4.400e+03, percent-clipped=19.0
+2023-03-05 05:02:15,422 INFO [train.py:968] (0/2) Epoch 10, batch 14750, giga_loss[loss=0.2851, simple_loss=0.361, pruned_loss=0.1046, over 28798.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3418, pruned_loss=0.09345, over 5682399.90 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3431, pruned_loss=0.09706, over 5760326.74 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3421, pruned_loss=0.09286, over 5669518.38 frames. ], batch size: 119, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:02:32,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4434, 3.2944, 1.5315, 1.5612], device='cuda:0'), covar=tensor([0.0860, 0.0364, 0.0832, 0.1174], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0498, 0.0333, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 05:03:16,116 INFO [train.py:968] (0/2) Epoch 10, batch 14800, giga_loss[loss=0.2872, simple_loss=0.3541, pruned_loss=0.1101, over 28655.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3406, pruned_loss=0.09352, over 5687963.86 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3431, pruned_loss=0.09708, over 5762668.85 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3408, pruned_loss=0.09295, over 5673640.09 frames. ], batch size: 307, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 05:04:11,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.502e+02 1.311e+03 1.658e+03 1.966e+03 5.634e+03, threshold=3.316e+03, percent-clipped=3.0
+2023-03-05 05:04:17,411 INFO [train.py:968] (0/2) Epoch 10, batch 14850, giga_loss[loss=0.2756, simple_loss=0.3502, pruned_loss=0.1005, over 28918.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3406, pruned_loss=0.0943, over 5690401.13 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.343, pruned_loss=0.09698, over 5762250.60 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3408, pruned_loss=0.09385, over 5676813.72 frames. ], batch size: 213, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 05:05:14,495 INFO [train.py:968] (0/2) Epoch 10, batch 14900, libri_loss[loss=0.282, simple_loss=0.3639, pruned_loss=0.1001, over 29515.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3407, pruned_loss=0.0947, over 5692642.97 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3427, pruned_loss=0.09681, over 5766278.20 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.341, pruned_loss=0.09439, over 5674987.87 frames. ], batch size: 81, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:05:37,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1342, 1.7332, 1.2422, 0.3436], device='cuda:0'), covar=tensor([0.2547, 0.1630, 0.2456, 0.3375], device='cuda:0'), in_proj_covar=tensor([0.1522, 0.1450, 0.1470, 0.1249], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 05:06:13,078 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.569e+02 1.444e+03 1.949e+03 2.866e+03 5.472e+03, threshold=3.898e+03, percent-clipped=16.0
+2023-03-05 05:06:16,622 INFO [train.py:968] (0/2) Epoch 10, batch 14950, giga_loss[loss=0.2577, simple_loss=0.3416, pruned_loss=0.08696, over 28687.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3426, pruned_loss=0.09514, over 5690509.27 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3427, pruned_loss=0.09689, over 5766346.83 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3428, pruned_loss=0.09478, over 5674161.57 frames. ], batch size: 242, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:07:32,500 INFO [train.py:968] (0/2) Epoch 10, batch 15000, giga_loss[loss=0.2529, simple_loss=0.3383, pruned_loss=0.08372, over 28958.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3424, pruned_loss=0.09404, over 5675989.74 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3427, pruned_loss=0.09683, over 5757814.61 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3427, pruned_loss=0.09379, over 5668718.67 frames. ], batch size: 136, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:07:32,504 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 05:07:41,003 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2076, simple_loss=0.3063, pruned_loss=0.05445, over 944034.00 frames. 
+2023-03-05 05:07:41,004 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 05:08:49,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.438e+02 1.286e+03 2.049e+03 3.124e+03 7.710e+03, threshold=4.098e+03, percent-clipped=17.0
+2023-03-05 05:08:52,383 INFO [train.py:968] (0/2) Epoch 10, batch 15050, giga_loss[loss=0.2457, simple_loss=0.3173, pruned_loss=0.08704, over 28143.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3412, pruned_loss=0.09379, over 5665079.31 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3421, pruned_loss=0.09657, over 5751533.34 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3421, pruned_loss=0.09377, over 5662699.61 frames. ], batch size: 412, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:10:00,005 INFO [train.py:968] (0/2) Epoch 10, batch 15100, giga_loss[loss=0.2581, simple_loss=0.3312, pruned_loss=0.09247, over 29043.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3383, pruned_loss=0.09401, over 5654938.51 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3419, pruned_loss=0.09674, over 5743261.09 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3391, pruned_loss=0.0938, over 5657599.90 frames. ], batch size: 155, lr: 3.27e-03, grad_scale: 2.0
+2023-03-05 05:11:01,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.289e+02 1.348e+03 1.850e+03 2.664e+03 6.452e+03, threshold=3.701e+03, percent-clipped=10.0
+2023-03-05 05:11:03,856 INFO [train.py:968] (0/2) Epoch 10, batch 15150, giga_loss[loss=0.2748, simple_loss=0.3538, pruned_loss=0.09792, over 28129.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3328, pruned_loss=0.09116, over 5662410.04 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3418, pruned_loss=0.09668, over 5746209.00 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3334, pruned_loss=0.09093, over 5660053.97 frames. ], batch size: 412, lr: 3.27e-03, grad_scale: 2.0
+2023-03-05 05:11:08,874 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.85 vs. limit=5.0
+2023-03-05 05:11:27,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9406, 3.7376, 3.5687, 1.7066], device='cuda:0'), covar=tensor([0.0581, 0.0743, 0.0734, 0.1997], device='cuda:0'), in_proj_covar=tensor([0.0988, 0.0920, 0.0811, 0.0633], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 05:12:02,895 INFO [train.py:968] (0/2) Epoch 10, batch 15200, giga_loss[loss=0.3231, simple_loss=0.371, pruned_loss=0.1376, over 26874.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3338, pruned_loss=0.09206, over 5663603.31 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3418, pruned_loss=0.09671, over 5751172.91 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.334, pruned_loss=0.09168, over 5654741.96 frames. ], batch size: 555, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:12:11,557 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2699, 1.4881, 1.4930, 1.3431], device='cuda:0'), covar=tensor([0.1254, 0.1271, 0.1706, 0.1309], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0713, 0.0649, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 05:12:57,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.675e+03 2.210e+03 3.641e+03 2.904e+04, threshold=4.421e+03, percent-clipped=24.0
+2023-03-05 05:12:57,484 INFO [train.py:968] (0/2) Epoch 10, batch 15250, giga_loss[loss=0.249, simple_loss=0.3273, pruned_loss=0.08535, over 28898.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3363, pruned_loss=0.09378, over 5668430.22 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3417, pruned_loss=0.09658, over 5752838.75 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3364, pruned_loss=0.09355, over 5658954.17 frames. ], batch size: 227, lr: 3.27e-03, grad_scale: 1.0
+2023-03-05 05:13:40,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2038, 1.5443, 1.4854, 1.4357], device='cuda:0'), covar=tensor([0.1293, 0.1209, 0.1652, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0710, 0.0646, 0.0633], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 05:14:04,791 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=424892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:14:06,034 INFO [train.py:968] (0/2) Epoch 10, batch 15300, giga_loss[loss=0.2574, simple_loss=0.343, pruned_loss=0.08586, over 28667.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3333, pruned_loss=0.09164, over 5655638.66 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3415, pruned_loss=0.09653, over 5752549.55 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3335, pruned_loss=0.09147, over 5648101.47 frames. ], batch size: 262, lr: 3.27e-03, grad_scale: 1.0
+2023-03-05 05:14:44,302 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 05:15:02,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.904e+02 1.157e+03 1.513e+03 1.811e+03 3.038e+03, threshold=3.026e+03, percent-clipped=0.0
+2023-03-05 05:15:02,517 INFO [train.py:968] (0/2) Epoch 10, batch 15350, giga_loss[loss=0.2522, simple_loss=0.3266, pruned_loss=0.08888, over 28745.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3319, pruned_loss=0.08974, over 5669589.40 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3419, pruned_loss=0.09685, over 5755095.46 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3315, pruned_loss=0.08918, over 5658879.71 frames. ], batch size: 243, lr: 3.27e-03, grad_scale: 1.0
+2023-03-05 05:15:45,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-05 05:16:14,890 INFO [train.py:968] (0/2) Epoch 10, batch 15400, libri_loss[loss=0.25, simple_loss=0.3307, pruned_loss=0.08468, over 29500.00 frames. ], tot_loss[loss=0.254, simple_loss=0.33, pruned_loss=0.08894, over 5662445.91 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3419, pruned_loss=0.09685, over 5758547.79 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3295, pruned_loss=0.08835, over 5648937.54 frames. ], batch size: 85, lr: 3.26e-03, grad_scale: 1.0
+2023-03-05 05:16:37,387 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 05:17:21,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.481e+02 1.119e+03 1.471e+03 2.133e+03 6.455e+03, threshold=2.942e+03, percent-clipped=12.0
+2023-03-05 05:17:21,721 INFO [train.py:968] (0/2) Epoch 10, batch 15450, libri_loss[loss=0.2606, simple_loss=0.3294, pruned_loss=0.09591, over 29533.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3306, pruned_loss=0.08903, over 5658434.12 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3412, pruned_loss=0.09647, over 5761204.46 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3305, pruned_loss=0.08867, over 5642514.74 frames. ], batch size: 76, lr: 3.26e-03, grad_scale: 1.0
+2023-03-05 05:18:26,213 INFO [train.py:968] (0/2) Epoch 10, batch 15500, giga_loss[loss=0.2318, simple_loss=0.3188, pruned_loss=0.07239, over 28734.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3302, pruned_loss=0.08877, over 5663413.33 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3413, pruned_loss=0.09649, over 5762753.28 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3299, pruned_loss=0.08835, over 5648281.76 frames. ], batch size: 243, lr: 3.26e-03, grad_scale: 1.0
+2023-03-05 05:19:05,160 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=425120.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 05:19:35,493 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.914e+02 1.268e+03 1.576e+03 2.285e+03 6.992e+03, threshold=3.151e+03, percent-clipped=14.0
+2023-03-05 05:19:35,506 INFO [train.py:968] (0/2) Epoch 10, batch 15550, giga_loss[loss=0.2569, simple_loss=0.3403, pruned_loss=0.08674, over 28977.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3311, pruned_loss=0.09004, over 5658666.43 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3412, pruned_loss=0.09646, over 5763467.51 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3308, pruned_loss=0.08972, over 5645706.86 frames. ], batch size: 227, lr: 3.26e-03, grad_scale: 1.0
+2023-03-05 05:20:34,924 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-05 05:20:35,072 INFO [train.py:968] (0/2) Epoch 10, batch 15600, giga_loss[loss=0.2538, simple_loss=0.3437, pruned_loss=0.08194, over 29142.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3317, pruned_loss=0.0889, over 5661624.62 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3414, pruned_loss=0.0966, over 5753149.69 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3311, pruned_loss=0.08833, over 5658179.64 frames. ], batch size: 113, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:21:02,041 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=425215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:21:39,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.985e+02 1.222e+03 1.560e+03 1.947e+03 5.119e+03, threshold=3.121e+03, percent-clipped=4.0
+2023-03-05 05:21:39,157 INFO [train.py:968] (0/2) Epoch 10, batch 15650, giga_loss[loss=0.293, simple_loss=0.3651, pruned_loss=0.1105, over 27556.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3342, pruned_loss=0.08924, over 5662706.17 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3412, pruned_loss=0.09649, over 5754554.07 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3338, pruned_loss=0.08881, over 5657864.13 frames. ], batch size: 472, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:21:57,039 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 05:22:10,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=425267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:22:41,678 INFO [train.py:968] (0/2) Epoch 10, batch 15700, giga_loss[loss=0.2495, simple_loss=0.3305, pruned_loss=0.0842, over 28086.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3359, pruned_loss=0.09017, over 5653808.59 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.341, pruned_loss=0.09653, over 5748184.55 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3356, pruned_loss=0.08967, over 5654378.87 frames. ], batch size: 412, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:23:17,395 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-05 05:23:41,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.842e+02 1.555e+03 2.059e+03 3.090e+03 8.698e+03, threshold=4.117e+03, percent-clipped=24.0
+2023-03-05 05:23:41,684 INFO [train.py:968] (0/2) Epoch 10, batch 15750, libri_loss[loss=0.2569, simple_loss=0.3359, pruned_loss=0.08897, over 29521.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3368, pruned_loss=0.09054, over 5666023.90 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3409, pruned_loss=0.09635, over 5752338.20 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3367, pruned_loss=0.09017, over 5660736.95 frames. ], batch size: 84, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:24:42,662 INFO [train.py:968] (0/2) Epoch 10, batch 15800, giga_loss[loss=0.2439, simple_loss=0.3006, pruned_loss=0.09366, over 24644.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3352, pruned_loss=0.08918, over 5678268.09 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3405, pruned_loss=0.09613, over 5754192.29 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3353, pruned_loss=0.08901, over 5671385.14 frames. ], batch size: 705, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:25:01,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=425410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:25:06,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=425413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:25:42,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4272, 1.7452, 1.3564, 1.4238], device='cuda:0'), covar=tensor([0.2458, 0.2299, 0.2704, 0.2094], device='cuda:0'), in_proj_covar=tensor([0.1249, 0.0923, 0.1116, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 05:25:43,329 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=425442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:25:45,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.900e+02 1.290e+03 1.798e+03 2.456e+03 5.550e+03, threshold=3.596e+03, percent-clipped=5.0
+2023-03-05 05:25:45,517 INFO [train.py:968] (0/2) Epoch 10, batch 15850, giga_loss[loss=0.2541, simple_loss=0.3325, pruned_loss=0.08786, over 27625.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3322, pruned_loss=0.08694, over 5685072.49 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3403, pruned_loss=0.09602, over 5754020.34 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3323, pruned_loss=0.08673, over 5678363.32 frames. ], batch size: 472, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:25:52,470 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-05 05:26:49,283 INFO [train.py:968] (0/2) Epoch 10, batch 15900, giga_loss[loss=0.2543, simple_loss=0.3301, pruned_loss=0.08927, over 28952.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.33, pruned_loss=0.08637, over 5676260.27 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3403, pruned_loss=0.09594, over 5755341.08 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.33, pruned_loss=0.0862, over 5669344.33 frames. ], batch size: 186, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:26:50,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=425495.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 05:27:48,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.918e+02 1.252e+03 1.622e+03 2.266e+03 4.782e+03, threshold=3.243e+03, percent-clipped=5.0
+2023-03-05 05:27:48,052 INFO [train.py:968] (0/2) Epoch 10, batch 15950, giga_loss[loss=0.2529, simple_loss=0.3411, pruned_loss=0.08234, over 28109.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3303, pruned_loss=0.08701, over 5680989.03 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.34, pruned_loss=0.09582, over 5759719.79 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3303, pruned_loss=0.08673, over 5669602.73 frames. ], batch size: 412, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:28:10,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3062, 1.6312, 1.4064, 1.2524], device='cuda:0'), covar=tensor([0.1897, 0.1304, 0.1063, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.1640, 0.1513, 0.1440, 0.1580], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 05:28:46,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=425590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:28:52,333 INFO [train.py:968] (0/2) Epoch 10, batch 16000, giga_loss[loss=0.2918, simple_loss=0.3548, pruned_loss=0.1144, over 27637.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3342, pruned_loss=0.08943, over 5671381.45 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3399, pruned_loss=0.09585, over 5753119.32 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3342, pruned_loss=0.08907, over 5667563.91 frames. ], batch size: 472, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:29:04,870 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 05:29:09,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1656, 4.9537, 4.6668, 2.2012], device='cuda:0'), covar=tensor([0.0382, 0.0570, 0.0637, 0.1981], device='cuda:0'), in_proj_covar=tensor([0.0987, 0.0920, 0.0814, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 05:29:43,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3918, 1.7211, 1.4504, 1.4824], device='cuda:0'), covar=tensor([0.1467, 0.1799, 0.1972, 0.1755], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0705, 0.0639, 0.0629], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 05:29:51,092 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=425638.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 05:29:56,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=425641.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 05:29:58,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.319e+02 1.311e+03 1.764e+03 2.786e+03 6.735e+03, threshold=3.529e+03, percent-clipped=17.0
+2023-03-05 05:29:58,832 INFO [train.py:968] (0/2) Epoch 10, batch 16050, giga_loss[loss=0.2502, simple_loss=0.3291, pruned_loss=0.08566, over 28910.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3352, pruned_loss=0.09026, over 5673025.26 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3402, pruned_loss=0.09599, over 5755835.96 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3348, pruned_loss=0.08971, over 5665841.68 frames. ], batch size: 164, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:30:34,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=425670.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 05:30:47,350 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=425681.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 05:31:01,953 INFO [train.py:968] (0/2) Epoch 10, batch 16100, giga_loss[loss=0.2727, simple_loss=0.3473, pruned_loss=0.09904, over 28939.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.338, pruned_loss=0.09189, over 5678780.42 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3401, pruned_loss=0.09596, over 5757928.73 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3377, pruned_loss=0.0914, over 5669999.36 frames. ], batch size: 199, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:31:27,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4630, 1.7104, 1.7273, 1.3158], device='cuda:0'), covar=tensor([0.1615, 0.2273, 0.1353, 0.1575], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0678, 0.0839, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 05:31:43,251 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=425733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:31:46,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=425736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:31:54,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.263e+02 1.339e+03 1.828e+03 2.243e+03 4.454e+03, threshold=3.656e+03, percent-clipped=2.0
+2023-03-05 05:31:54,477 INFO [train.py:968] (0/2) Epoch 10, batch 16150, giga_loss[loss=0.2804, simple_loss=0.3641, pruned_loss=0.09831, over 28570.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3407, pruned_loss=0.09309, over 5678623.88 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3395, pruned_loss=0.09575, over 5753211.86 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3411, pruned_loss=0.09278, over 5673206.36 frames. ], batch size: 307, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:32:19,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=425765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:32:20,744 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4935, 1.7347, 1.4094, 1.6323], device='cuda:0'), covar=tensor([0.2401, 0.2268, 0.2586, 0.2094], device='cuda:0'), in_proj_covar=tensor([0.1249, 0.0921, 0.1115, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 05:32:39,963 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3502, 3.2929, 1.4844, 1.5030], device='cuda:0'), covar=tensor([0.0900, 0.0300, 0.0924, 0.1210], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0498, 0.0335, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 05:32:53,689 INFO [train.py:968] (0/2) Epoch 10, batch 16200, giga_loss[loss=0.2534, simple_loss=0.3319, pruned_loss=0.08749, over 27670.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3416, pruned_loss=0.09339, over 5675982.12 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3387, pruned_loss=0.09543, over 5747123.50 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3427, pruned_loss=0.09335, over 5674138.05 frames. ], batch size: 474, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:33:40,528 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5931, 1.7433, 1.5683, 1.5651], device='cuda:0'), covar=tensor([0.1206, 0.2166, 0.1808, 0.1753], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0711, 0.0642, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 05:34:05,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.369e+02 1.316e+03 1.669e+03 2.149e+03 4.226e+03, threshold=3.337e+03, percent-clipped=2.0
+2023-03-05 05:34:05,508 INFO [train.py:968] (0/2) Epoch 10, batch 16250, libri_loss[loss=0.2411, simple_loss=0.3121, pruned_loss=0.08503, over 29640.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3401, pruned_loss=0.09229, over 5680573.29 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3384, pruned_loss=0.09528, over 5747382.36 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3412, pruned_loss=0.09236, over 5677373.16 frames. ], batch size: 69, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:34:07,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4278, 1.6710, 1.2909, 1.9130], device='cuda:0'), covar=tensor([0.2463, 0.2378, 0.2582, 0.2361], device='cuda:0'), in_proj_covar=tensor([0.1255, 0.0925, 0.1119, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 05:34:14,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2262, 0.8276, 0.8184, 1.4292], device='cuda:0'), covar=tensor([0.0739, 0.0345, 0.0357, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0113, 0.0117, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0055, 0.0050, 0.0086], device='cuda:0')
+2023-03-05 05:34:46,811 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 05:35:09,397 INFO [train.py:968] (0/2) Epoch 10, batch 16300, giga_loss[loss=0.2365, simple_loss=0.3212, pruned_loss=0.07593, over 28990.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3393, pruned_loss=0.09274, over 5691755.76 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3382, pruned_loss=0.09515, over 5750501.23 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3404, pruned_loss=0.09285, over 5685442.25 frames. ], batch size: 136, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:36:14,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.579e+02 1.367e+03 1.751e+03 2.505e+03 5.954e+03, threshold=3.503e+03, percent-clipped=11.0
+2023-03-05 05:36:14,573 INFO [train.py:968] (0/2) Epoch 10, batch 16350, giga_loss[loss=0.3044, simple_loss=0.3709, pruned_loss=0.119, over 28911.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3391, pruned_loss=0.09288, over 5676651.25 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3384, pruned_loss=0.09518, over 5753114.91 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3399, pruned_loss=0.09286, over 5667344.10 frames. ], batch size: 145, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:37:15,206 INFO [train.py:968] (0/2) Epoch 10, batch 16400, giga_loss[loss=0.2489, simple_loss=0.3303, pruned_loss=0.08374, over 28840.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3376, pruned_loss=0.09272, over 5674776.40 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3379, pruned_loss=0.09486, over 5747644.91 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3387, pruned_loss=0.09293, over 5669887.49 frames. ], batch size: 284, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 05:37:21,531 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-426000.pt
+2023-03-05 05:37:49,880 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:38:15,972 INFO [train.py:968] (0/2) Epoch 10, batch 16450, giga_loss[loss=0.3044, simple_loss=0.3721, pruned_loss=0.1183, over 28978.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.337, pruned_loss=0.09312, over 5663598.23 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3382, pruned_loss=0.09493, over 5740990.33 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3376, pruned_loss=0.09318, over 5663645.13 frames. ], batch size: 284, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:38:16,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.463e+02 1.489e+03 1.931e+03 2.827e+03 7.993e+03, threshold=3.862e+03, percent-clipped=10.0
+2023-03-05 05:38:32,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426056.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 05:38:37,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6215, 1.8163, 1.7096, 1.5037], device='cuda:0'), covar=tensor([0.1340, 0.1920, 0.1712, 0.1797], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0706, 0.0639, 0.0631], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 05:39:20,185 INFO [train.py:968] (0/2) Epoch 10, batch 16500, giga_loss[loss=0.2319, simple_loss=0.2984, pruned_loss=0.08268, over 24484.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3371, pruned_loss=0.09221, over 5661803.86 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.338, pruned_loss=0.09485, over 5735611.43 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3377, pruned_loss=0.09228, over 5665266.33 frames. ], batch size: 705, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:40:17,462 INFO [train.py:968] (0/2) Epoch 10, batch 16550, giga_loss[loss=0.2749, simple_loss=0.3673, pruned_loss=0.09131, over 28698.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3365, pruned_loss=0.09114, over 5666555.58 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.338, pruned_loss=0.09497, over 5735811.77 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3369, pruned_loss=0.09102, over 5667560.86 frames. ], batch size: 307, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:40:19,497 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.121e+02 1.342e+03 1.751e+03 2.822e+03 8.934e+03, threshold=3.502e+03, percent-clipped=12.0
+2023-03-05 05:40:41,229 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:40:54,946 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2139, 1.4458, 1.3823, 1.2667], device='cuda:0'), covar=tensor([0.2242, 0.1397, 0.1146, 0.1490], device='cuda:0'), in_proj_covar=tensor([0.1641, 0.1502, 0.1435, 0.1577], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 05:41:14,839 INFO [train.py:968] (0/2) Epoch 10, batch 16600, giga_loss[loss=0.2345, simple_loss=0.3291, pruned_loss=0.06996, over 28980.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3381, pruned_loss=0.08997, over 5672609.10 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3381, pruned_loss=0.09495, over 5738215.04 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3384, pruned_loss=0.08979, over 5669857.01 frames. ], batch size: 136, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:41:21,804 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=426199.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 05:41:24,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=426202.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 05:41:57,084 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=426231.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 05:42:10,365 INFO [train.py:968] (0/2) Epoch 10, batch 16650, giga_loss[loss=0.2349, simple_loss=0.3172, pruned_loss=0.07625, over 28662.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3394, pruned_loss=0.0899, over 5677386.85 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3376, pruned_loss=0.09459, over 5742315.45 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.34, pruned_loss=0.08987, over 5669217.37 frames. ], batch size: 85, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:42:11,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.625e+02 1.207e+03 1.644e+03 2.576e+03 5.313e+03, threshold=3.287e+03, percent-clipped=8.0
+2023-03-05 05:42:33,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:42:47,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6003, 2.3368, 1.6139, 0.6057], device='cuda:0'), covar=tensor([0.3283, 0.1753, 0.2638, 0.3417], device='cuda:0'), in_proj_covar=tensor([0.1520, 0.1446, 0.1467, 0.1251], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 05:43:10,180 INFO [train.py:968] (0/2) Epoch 10, batch 16700, giga_loss[loss=0.2627, simple_loss=0.3412, pruned_loss=0.09207, over 28949.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3393, pruned_loss=0.08964, over 5675071.97 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3379, pruned_loss=0.09477, over 5733473.37 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3396, pruned_loss=0.08934, over 5675936.16 frames. ], batch size: 227, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:44:15,248 INFO [train.py:968] (0/2) Epoch 10, batch 16750, giga_loss[loss=0.2889, simple_loss=0.3604, pruned_loss=0.1087, over 28184.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3393, pruned_loss=0.08999, over 5664642.04 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3376, pruned_loss=0.09473, over 5729813.34 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3398, pruned_loss=0.08962, over 5667820.07 frames. ], batch size: 412, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:44:16,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.769e+02 1.311e+03 1.741e+03 2.144e+03 4.266e+03, threshold=3.482e+03, percent-clipped=6.0
+2023-03-05 05:44:46,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6322, 2.2834, 1.4903, 0.9321], device='cuda:0'), covar=tensor([0.5291, 0.2693, 0.2773, 0.4066], device='cuda:0'), in_proj_covar=tensor([0.1517, 0.1443, 0.1461, 0.1251], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 05:45:05,317 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:45:26,468 INFO [train.py:968] (0/2) Epoch 10, batch 16800, giga_loss[loss=0.2515, simple_loss=0.3362, pruned_loss=0.0834, over 28563.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3399, pruned_loss=0.09017, over 5667437.40 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3376, pruned_loss=0.09475, over 5731211.71 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3403, pruned_loss=0.08976, over 5667434.01 frames. ], batch size: 78, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 05:45:33,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:46:23,888 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 05:46:42,077 INFO [train.py:968] (0/2) Epoch 10, batch 16850, giga_loss[loss=0.2845, simple_loss=0.3455, pruned_loss=0.1117, over 26793.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3401, pruned_loss=0.08922, over 5669206.40 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3379, pruned_loss=0.09495, over 5728779.65 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3401, pruned_loss=0.08867, over 5671261.34 frames. ], batch size: 555, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 05:46:42,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.457e+02 1.400e+03 1.844e+03 2.568e+03 4.881e+03, threshold=3.687e+03, percent-clipped=10.0
+2023-03-05 05:47:05,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4544, 1.7342, 1.4063, 1.6218], device='cuda:0'), covar=tensor([0.2399, 0.2309, 0.2583, 0.2128], device='cuda:0'), in_proj_covar=tensor([0.1247, 0.0924, 0.1118, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 05:47:49,750 INFO [train.py:968] (0/2) Epoch 10, batch 16900, giga_loss[loss=0.3093, simple_loss=0.3705, pruned_loss=0.1241, over 26835.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3427, pruned_loss=0.09098, over 5668719.09 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3382, pruned_loss=0.09508, over 5730459.41 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3426, pruned_loss=0.09025, over 5667103.32 frames. ], batch size: 555, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:48:48,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:48:52,868 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:48:52,958 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=426541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:48:54,986 INFO [train.py:968] (0/2) Epoch 10, batch 16950, giga_loss[loss=0.2136, simple_loss=0.3064, pruned_loss=0.06041, over 28795.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3443, pruned_loss=0.09132, over 5676668.81 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3381, pruned_loss=0.09499, over 5727312.42 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3446, pruned_loss=0.09068, over 5676888.66 frames. ], batch size: 119, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:48:55,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=426544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:48:56,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.600e+02 1.305e+03 1.726e+03 2.523e+03 9.349e+03, threshold=3.453e+03, percent-clipped=9.0
+2023-03-05 05:49:32,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=426573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:49:33,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1236, 1.4772, 1.3915, 1.0544], device='cuda:0'), covar=tensor([0.1333, 0.1896, 0.1099, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0677, 0.0839, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 05:49:51,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0161, 4.8187, 4.5721, 2.1595], device='cuda:0'), covar=tensor([0.0441, 0.0571, 0.0642, 0.1961], device='cuda:0'), in_proj_covar=tensor([0.0992, 0.0922, 0.0812, 0.0633], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 05:49:56,590 INFO [train.py:968] (0/2) Epoch 10, batch 17000, libri_loss[loss=0.2363, simple_loss=0.3173, pruned_loss=0.07768, over 29506.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3434, pruned_loss=0.09139, over 5670273.36 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3382, pruned_loss=0.09492, over 5718240.93 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3436, pruned_loss=0.09082, over 5676010.71 frames. ], batch size: 82, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:50:56,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:51:02,077 INFO [train.py:968] (0/2) Epoch 10, batch 17050, giga_loss[loss=0.2657, simple_loss=0.3435, pruned_loss=0.0939, over 28411.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3386, pruned_loss=0.08908, over 5686609.38 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3369, pruned_loss=0.09412, over 5723562.48 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.34, pruned_loss=0.08921, over 5685134.65 frames. ], batch size: 368, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:51:08,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.476e+02 1.301e+03 1.719e+03 2.330e+03 5.595e+03, threshold=3.439e+03, percent-clipped=5.0
+2023-03-05 05:51:55,234 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=426680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:51:58,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=426683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:52:12,333 INFO [train.py:968] (0/2) Epoch 10, batch 17100, giga_loss[loss=0.2331, simple_loss=0.3223, pruned_loss=0.07194, over 28915.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3368, pruned_loss=0.08755, over 5673702.62 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3371, pruned_loss=0.09414, over 5706592.87 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3379, pruned_loss=0.08747, over 5686766.35 frames. ], batch size: 145, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:52:40,783 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=426712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:53:21,321 INFO [train.py:968] (0/2) Epoch 10, batch 17150, giga_loss[loss=0.2434, simple_loss=0.3281, pruned_loss=0.0793, over 28986.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3358, pruned_loss=0.08676, over 5681795.41 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3371, pruned_loss=0.0941, over 5710060.24 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3367, pruned_loss=0.0866, over 5688563.24 frames. ], batch size: 155, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:53:26,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.893e+02 1.130e+03 1.584e+03 2.329e+03 6.684e+03, threshold=3.167e+03, percent-clipped=7.0
+2023-03-05 05:53:35,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:54:10,922 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=426782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:54:14,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=426785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:54:15,165 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-05 05:54:18,574 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:54:24,945 INFO [train.py:968] (0/2) Epoch 10, batch 17200, giga_loss[loss=0.2862, simple_loss=0.3655, pruned_loss=0.1035, over 28109.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.338, pruned_loss=0.08839, over 5677609.49 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3372, pruned_loss=0.09416, over 5710811.08 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3385, pruned_loss=0.08814, over 5681921.92 frames. ], batch size: 412, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:54:49,124 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=426814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:55:24,098 INFO [train.py:968] (0/2) Epoch 10, batch 17250, giga_loss[loss=0.2228, simple_loss=0.2962, pruned_loss=0.07472, over 24489.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3396, pruned_loss=0.08921, over 5668007.58 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3371, pruned_loss=0.09405, over 5705711.05 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3401, pruned_loss=0.08902, over 5675378.90 frames. ], batch size: 705, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:55:26,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.626e+02 1.318e+03 1.749e+03 2.438e+03 2.388e+04, threshold=3.498e+03, percent-clipped=13.0
+2023-03-05 05:56:18,483 INFO [train.py:968] (0/2) Epoch 10, batch 17300, giga_loss[loss=0.2547, simple_loss=0.3328, pruned_loss=0.0883, over 28667.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3386, pruned_loss=0.08938, over 5669811.93 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3369, pruned_loss=0.09403, over 5706652.81 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3392, pruned_loss=0.08913, over 5674162.43 frames. ], batch size: 262, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:56:22,791 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=426897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:56:24,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=426900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:56:44,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:56:58,042 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=426929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:57:17,919 INFO [train.py:968] (0/2) Epoch 10, batch 17350, giga_loss[loss=0.2648, simple_loss=0.3474, pruned_loss=0.09112, over 28907.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3375, pruned_loss=0.08986, over 5671831.79 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3368, pruned_loss=0.09393, over 5710496.86 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3381, pruned_loss=0.08966, over 5671285.29 frames. ], batch size: 227, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:57:20,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.372e+02 1.430e+03 1.934e+03 2.692e+03 5.011e+03, threshold=3.869e+03, percent-clipped=12.0
+2023-03-05 05:58:14,411 INFO [train.py:968] (0/2) Epoch 10, batch 17400, giga_loss[loss=0.3211, simple_loss=0.3824, pruned_loss=0.1299, over 28886.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3374, pruned_loss=0.09032, over 5679215.91 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3368, pruned_loss=0.09405, over 5711402.87 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3379, pruned_loss=0.08995, over 5677342.66 frames. ], batch size: 106, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:58:22,029 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=427001.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 05:59:11,115 INFO [train.py:968] (0/2) Epoch 10, batch 17450, giga_loss[loss=0.3347, simple_loss=0.4014, pruned_loss=0.134, over 28522.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3466, pruned_loss=0.0969, over 5669914.21 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3365, pruned_loss=0.09398, over 5705985.81 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3473, pruned_loss=0.09662, over 5672466.85 frames. ], batch size: 336, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:59:13,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.050e+02 1.261e+03 1.853e+03 2.518e+03 8.873e+03, threshold=3.706e+03, percent-clipped=10.0
+2023-03-05 05:59:24,246 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=427059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:59:27,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=427062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:59:39,604 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 05:59:53,570 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=427091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:59:55,342 INFO [train.py:968] (0/2) Epoch 10, batch 17500, giga_loss[loss=0.3328, simple_loss=0.4156, pruned_loss=0.1251, over 28713.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3554, pruned_loss=0.1018, over 5677676.29 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3365, pruned_loss=0.09392, over 5704900.64 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3562, pruned_loss=0.1018, over 5679861.98 frames. ], batch size: 242, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:00:31,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4570, 1.6752, 1.6068, 1.5811], device='cuda:0'), covar=tensor([0.1181, 0.1431, 0.1352, 0.1228], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0712, 0.0646, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 06:00:38,711 INFO [train.py:968] (0/2) Epoch 10, batch 17550, giga_loss[loss=0.237, simple_loss=0.3161, pruned_loss=0.079, over 29073.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3568, pruned_loss=0.1031, over 5674551.64 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3369, pruned_loss=0.09407, over 5697332.46 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3574, pruned_loss=0.1031, over 5682878.60 frames. ], batch size: 136, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:00:40,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.819e+02 1.272e+03 1.553e+03 2.020e+03 5.941e+03, threshold=3.106e+03, percent-clipped=6.0
+2023-03-05 06:00:56,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=427164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:01:23,594 INFO [train.py:968] (0/2) Epoch 10, batch 17600, libri_loss[loss=0.2531, simple_loss=0.3369, pruned_loss=0.08462, over 26251.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3506, pruned_loss=0.1009, over 5675686.99 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3365, pruned_loss=0.09378, over 5699797.97 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.352, pruned_loss=0.1013, over 5679928.36 frames. ], batch size: 136, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 06:01:44,478 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0923, 1.4589, 1.3133, 0.9837], device='cuda:0'), covar=tensor([0.1698, 0.1366, 0.0894, 0.1477], device='cuda:0'), in_proj_covar=tensor([0.1635, 0.1489, 0.1438, 0.1566], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 06:02:05,562 INFO [train.py:968] (0/2) Epoch 10, batch 17650, giga_loss[loss=0.2246, simple_loss=0.2986, pruned_loss=0.07528, over 28710.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3435, pruned_loss=0.09767, over 5683805.39 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.337, pruned_loss=0.09396, over 5706726.27 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3445, pruned_loss=0.09805, over 5680464.34 frames. ], batch size: 92, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 06:02:10,159 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.901e+02 1.123e+03 1.455e+03 2.364e+03 6.674e+03, threshold=2.909e+03, percent-clipped=15.0
+2023-03-05 06:02:48,635 INFO [train.py:968] (0/2) Epoch 10, batch 17700, giga_loss[loss=0.2318, simple_loss=0.2953, pruned_loss=0.08412, over 28870.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3369, pruned_loss=0.09482, over 5690540.36 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3371, pruned_loss=0.09387, over 5711942.64 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3377, pruned_loss=0.09527, over 5682379.92 frames. ], batch size: 99, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 06:03:03,362 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=427307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:03:05,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=427310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:03:26,878 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-05 06:03:31,187 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=427339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:03:35,935 INFO [train.py:968] (0/2) Epoch 10, batch 17750, giga_loss[loss=0.2153, simple_loss=0.287, pruned_loss=0.07185, over 28858.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3294, pruned_loss=0.09158, over 5690476.47 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3375, pruned_loss=0.09404, over 5716112.69 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3296, pruned_loss=0.09175, over 5679692.84 frames. ], batch size: 199, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 06:03:37,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.093e+02 9.320e+02 1.176e+03 1.558e+03 3.490e+03, threshold=2.352e+03, percent-clipped=3.0
+2023-03-05 06:03:40,213 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2264, 1.2818, 3.7533, 3.0731], device='cuda:0'), covar=tensor([0.1485, 0.2464, 0.0370, 0.1630], device='cuda:0'), in_proj_covar=tensor([0.0638, 0.0575, 0.0831, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 06:04:05,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=427376.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 06:04:08,858 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=427381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:04:18,153 INFO [train.py:968] (0/2) Epoch 10, batch 17800, giga_loss[loss=0.2203, simple_loss=0.2999, pruned_loss=0.07033, over 29086.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3247, pruned_loss=0.08943, over 5697672.89 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3381, pruned_loss=0.09425, over 5721148.35 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3238, pruned_loss=0.08923, over 5683418.69 frames. ], batch size: 155, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:04:33,279 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 06:04:57,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6653, 2.4529, 2.1074, 2.0669], device='cuda:0'), covar=tensor([0.0625, 0.0587, 0.0832, 0.0972], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0431, 0.0498, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 06:05:00,678 INFO [train.py:968] (0/2) Epoch 10, batch 17850, giga_loss[loss=0.2514, simple_loss=0.3144, pruned_loss=0.09419, over 28854.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3215, pruned_loss=0.08835, over 5700814.37 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3382, pruned_loss=0.09434, over 5724685.54 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3205, pruned_loss=0.08802, over 5685951.40 frames. ], batch size: 186, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:05:03,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.004e+02 1.064e+03 1.469e+03 2.001e+03 9.002e+03, threshold=2.939e+03, percent-clipped=18.0
+2023-03-05 06:05:03,709 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=427448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:05:39,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5335, 2.5159, 2.4870, 2.2231], device='cuda:0'), covar=tensor([0.1250, 0.1788, 0.1386, 0.1407], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0724, 0.0657, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 06:05:43,579 INFO [train.py:968] (0/2) Epoch 10, batch 17900, libri_loss[loss=0.3206, simple_loss=0.3854, pruned_loss=0.1279, over 29540.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3189, pruned_loss=0.08715, over 5704040.93 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3385, pruned_loss=0.09441, over 5729331.85 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3173, pruned_loss=0.08664, over 5687273.94 frames. ], batch size: 83, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:05:49,865 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 06:06:06,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=427519.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 06:06:08,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=427522.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 06:06:25,792 INFO [train.py:968] (0/2) Epoch 10, batch 17950, giga_loss[loss=0.2262, simple_loss=0.2966, pruned_loss=0.07787, over 28598.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3158, pruned_loss=0.08597, over 5705683.38 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3387, pruned_loss=0.09455, over 5731884.55 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.314, pruned_loss=0.08531, over 5689657.43 frames. ], batch size: 92, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:06:29,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.404e+02 9.753e+02 1.258e+03 1.658e+03 2.751e+03, threshold=2.516e+03, percent-clipped=0.0
+2023-03-05 06:06:32,278 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=427551.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 06:06:44,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=427564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:07:11,167 INFO [train.py:968] (0/2) Epoch 10, batch 18000, giga_loss[loss=0.2399, simple_loss=0.3187, pruned_loss=0.08053, over 28850.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.313, pruned_loss=0.08469, over 5706729.76 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3388, pruned_loss=0.09454, over 5733794.85 frames. ], giga_tot_loss[loss=0.2398, simple_loss=0.3113, pruned_loss=0.08411, over 5692404.35 frames. ], batch size: 174, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:07:11,171 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 06:07:19,741 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2203, simple_loss=0.3251, pruned_loss=0.05777, over 944034.00 frames. 
+2023-03-05 06:07:19,742 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 06:08:03,698 INFO [train.py:968] (0/2) Epoch 10, batch 18050, giga_loss[loss=0.2468, simple_loss=0.3167, pruned_loss=0.08849, over 28557.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3108, pruned_loss=0.0837, over 5701340.29 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3394, pruned_loss=0.09477, over 5737135.44 frames. ], giga_tot_loss[loss=0.237, simple_loss=0.3084, pruned_loss=0.0828, over 5686237.79 frames. ], batch size: 336, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:08:06,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.583e+02 9.908e+02 1.286e+03 1.710e+03 4.832e+03, threshold=2.572e+03, percent-clipped=7.0
+2023-03-05 06:08:21,055 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2644, 2.0542, 1.7098, 1.4370], device='cuda:0'), covar=tensor([0.0812, 0.0289, 0.0281, 0.0990], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0078, 0.0055, 0.0050, 0.0085], device='cuda:0')
+2023-03-05 06:08:22,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7600, 2.0585, 2.0901, 1.5967], device='cuda:0'), covar=tensor([0.1658, 0.2121, 0.1283, 0.1485], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0688, 0.0852, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 06:08:47,925 INFO [train.py:968] (0/2) Epoch 10, batch 18100, giga_loss[loss=0.2226, simple_loss=0.2902, pruned_loss=0.07753, over 28927.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3085, pruned_loss=0.08286, over 5704192.03 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3399, pruned_loss=0.09502, over 5741529.64 frames. ], giga_tot_loss[loss=0.2344, simple_loss=0.3055, pruned_loss=0.08166, over 5687686.84 frames. ], batch size: 106, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:09:08,505 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 06:09:15,987 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-05 06:09:27,707 INFO [train.py:968] (0/2) Epoch 10, batch 18150, giga_loss[loss=0.2532, simple_loss=0.3289, pruned_loss=0.08879, over 28619.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3068, pruned_loss=0.0818, over 5708669.22 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3404, pruned_loss=0.09517, over 5748442.49 frames. ], giga_tot_loss[loss=0.2313, simple_loss=0.3024, pruned_loss=0.08009, over 5687328.21 frames. ], batch size: 242, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:09:33,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.257e+02 1.002e+03 1.297e+03 1.782e+03 4.209e+03, threshold=2.593e+03, percent-clipped=6.0
+2023-03-05 06:09:40,935 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=427756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:09:46,966 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-05 06:09:52,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5500, 1.7320, 1.4227, 1.5409], device='cuda:0'), covar=tensor([0.2372, 0.2357, 0.2617, 0.2283], device='cuda:0'), in_proj_covar=tensor([0.1258, 0.0936, 0.1123, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 06:10:17,702 INFO [train.py:968] (0/2) Epoch 10, batch 18200, libri_loss[loss=0.286, simple_loss=0.3653, pruned_loss=0.1033, over 29246.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3033, pruned_loss=0.08022, over 5694760.06 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3409, pruned_loss=0.09534, over 5750784.03 frames. ], giga_tot_loss[loss=0.2277, simple_loss=0.2986, pruned_loss=0.07842, over 5674680.77 frames. ], batch size: 94, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:10:17,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4344, 1.5566, 1.5421, 1.3968], device='cuda:0'), covar=tensor([0.1278, 0.1798, 0.1737, 0.1671], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0716, 0.0651, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 06:10:21,916 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 06:10:42,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=427823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:11:01,386 INFO [train.py:968] (0/2) Epoch 10, batch 18250, giga_loss[loss=0.236, simple_loss=0.3073, pruned_loss=0.08237, over 28778.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3038, pruned_loss=0.08102, over 5692539.93 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3408, pruned_loss=0.09529, over 5753071.95 frames. ], giga_tot_loss[loss=0.2293, simple_loss=0.2997, pruned_loss=0.07943, over 5673712.61 frames. ], batch size: 112, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:11:07,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.565e+02 1.023e+03 1.464e+03 2.247e+03 8.701e+03, threshold=2.928e+03, percent-clipped=21.0
+2023-03-05 06:11:34,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7004, 1.7673, 1.6296, 1.5695], device='cuda:0'), covar=tensor([0.1232, 0.1698, 0.1807, 0.1638], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0720, 0.0654, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 06:11:54,654 INFO [train.py:968] (0/2) Epoch 10, batch 18300, giga_loss[loss=0.2968, simple_loss=0.3664, pruned_loss=0.1137, over 28875.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3156, pruned_loss=0.08711, over 5687115.87 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3409, pruned_loss=0.09528, over 5752020.29 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3118, pruned_loss=0.08572, over 5672294.16 frames. ], batch size: 145, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:12:00,750 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=427899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:12:03,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=427902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:12:28,990 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=427931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:12:34,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=427939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:12:39,200 INFO [train.py:968] (0/2) Epoch 10, batch 18350, giga_loss[loss=0.3478, simple_loss=0.405, pruned_loss=0.1453, over 27587.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3297, pruned_loss=0.09489, over 5693699.65 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3412, pruned_loss=0.09544, over 5755107.53 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3261, pruned_loss=0.09358, over 5678081.76 frames. ], batch size: 472, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:12:43,326 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.180e+02 1.218e+03 1.595e+03 2.149e+03 6.796e+03, threshold=3.190e+03, percent-clipped=8.0
+2023-03-05 06:12:55,890 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=427966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:12:59,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=427969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:13:19,423 INFO [train.py:968] (0/2) Epoch 10, batch 18400, giga_loss[loss=0.3098, simple_loss=0.3759, pruned_loss=0.1219, over 29023.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3396, pruned_loss=0.09951, over 5688811.90 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3412, pruned_loss=0.09538, over 5748141.57 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3366, pruned_loss=0.09857, over 5680817.00 frames. ], batch size: 106, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:13:23,520 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=427998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:13:24,611 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-428000.pt
+2023-03-05 06:14:04,667 INFO [train.py:968] (0/2) Epoch 10, batch 18450, giga_loss[loss=0.3112, simple_loss=0.3881, pruned_loss=0.1171, over 28607.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.345, pruned_loss=0.101, over 5683647.81 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3416, pruned_loss=0.09555, over 5749536.45 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3423, pruned_loss=0.1002, over 5675062.80 frames. ], batch size: 307, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:14:07,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.344e+02 1.194e+03 1.377e+03 1.909e+03 4.736e+03, threshold=2.753e+03, percent-clipped=2.0
+2023-03-05 06:14:34,889 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=428082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:14:37,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=428085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:14:45,806 INFO [train.py:968] (0/2) Epoch 10, batch 18500, libri_loss[loss=0.2537, simple_loss=0.3297, pruned_loss=0.08888, over 29650.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3461, pruned_loss=0.09994, over 5684161.49 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3418, pruned_loss=0.09556, over 5751108.71 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3439, pruned_loss=0.09941, over 5673911.39 frames. ], batch size: 73, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:14:47,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-05 06:15:04,878 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=428114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:15:13,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=428124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:15:21,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5643, 1.6355, 1.7589, 1.3537], device='cuda:0'), covar=tensor([0.1554, 0.2242, 0.1253, 0.1537], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0686, 0.0850, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 06:15:23,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-05 06:15:30,493 INFO [train.py:968] (0/2) Epoch 10, batch 18550, giga_loss[loss=0.3064, simple_loss=0.3686, pruned_loss=0.1221, over 26724.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3492, pruned_loss=0.1015, over 5670728.45 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3423, pruned_loss=0.0959, over 5746000.33 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3471, pruned_loss=0.1009, over 5665933.51 frames. ], batch size: 555, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:15:36,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6333, 1.7613, 1.9116, 1.4709], device='cuda:0'), covar=tensor([0.1573, 0.2083, 0.1223, 0.1431], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0685, 0.0849, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 06:15:36,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.227e+02 1.083e+03 1.237e+03 1.604e+03 3.476e+03, threshold=2.474e+03, percent-clipped=3.0
+2023-03-05 06:16:16,621 INFO [train.py:968] (0/2) Epoch 10, batch 18600, giga_loss[loss=0.2864, simple_loss=0.3585, pruned_loss=0.1071, over 28954.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3521, pruned_loss=0.1043, over 5674790.15 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3425, pruned_loss=0.09602, over 5746805.33 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3503, pruned_loss=0.1038, over 5669998.49 frames. ], batch size: 136, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:17:03,691 INFO [train.py:968] (0/2) Epoch 10, batch 18650, giga_loss[loss=0.2901, simple_loss=0.3658, pruned_loss=0.1072, over 28537.00 frames. ], tot_loss[loss=0.285, simple_loss=0.356, pruned_loss=0.107, over 5670286.19 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3429, pruned_loss=0.09615, over 5740217.96 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3544, pruned_loss=0.1066, over 5671566.97 frames. ], batch size: 307, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:17:08,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.998e+02 1.123e+03 1.374e+03 1.866e+03 7.938e+03, threshold=2.748e+03, percent-clipped=9.0
+2023-03-05 06:17:34,528 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2953, 2.0712, 1.7060, 1.6817], device='cuda:0'), covar=tensor([0.0743, 0.0652, 0.0891, 0.1003], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0430, 0.0498, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 06:17:42,587 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 06:17:43,628 INFO [train.py:968] (0/2) Epoch 10, batch 18700, giga_loss[loss=0.3292, simple_loss=0.4033, pruned_loss=0.1276, over 29031.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3597, pruned_loss=0.1086, over 5677879.51 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3439, pruned_loss=0.0965, over 5742115.84 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3578, pruned_loss=0.1082, over 5675544.59 frames. ], batch size: 164, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:18:15,464 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5067, 1.6736, 1.7542, 1.3473], device='cuda:0'), covar=tensor([0.1572, 0.2068, 0.1277, 0.1513], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0687, 0.0851, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 06:18:25,182 INFO [train.py:968] (0/2) Epoch 10, batch 18750, giga_loss[loss=0.2737, simple_loss=0.3584, pruned_loss=0.09455, over 28713.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3619, pruned_loss=0.1088, over 5672420.93 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3441, pruned_loss=0.0966, over 5735735.00 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3607, pruned_loss=0.1087, over 5675314.87 frames. ], batch size: 262, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:18:32,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.474e+02 1.100e+03 1.469e+03 1.875e+03 1.044e+04, threshold=2.937e+03, percent-clipped=14.0
+2023-03-05 06:18:42,592 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=428363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:19:07,287 INFO [train.py:968] (0/2) Epoch 10, batch 18800, giga_loss[loss=0.2985, simple_loss=0.3739, pruned_loss=0.1116, over 28254.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3634, pruned_loss=0.109, over 5679297.96 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3445, pruned_loss=0.09672, over 5737347.59 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3624, pruned_loss=0.109, over 5679289.74 frames. ], batch size: 77, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:19:48,331 INFO [train.py:968] (0/2) Epoch 10, batch 18850, giga_loss[loss=0.2952, simple_loss=0.3794, pruned_loss=0.1055, over 28916.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3627, pruned_loss=0.1075, over 5681067.89 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3444, pruned_loss=0.09654, over 5733624.49 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3623, pruned_loss=0.1079, over 5683465.03 frames. ], batch size: 227, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:19:54,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.666e+02 1.083e+03 1.434e+03 1.720e+03 3.240e+03, threshold=2.869e+03, percent-clipped=1.0
+2023-03-05 06:20:16,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5639, 1.7701, 1.4155, 1.8169], device='cuda:0'), covar=tensor([0.2333, 0.2324, 0.2583, 0.2187], device='cuda:0'), in_proj_covar=tensor([0.1260, 0.0942, 0.1123, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 06:20:24,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0607, 2.5638, 1.1492, 1.1465], device='cuda:0'), covar=tensor([0.1062, 0.0288, 0.0915, 0.1500], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0492, 0.0329, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 06:20:28,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7205, 1.7048, 1.6150, 1.4259], device='cuda:0'), covar=tensor([0.2193, 0.1899, 0.1538, 0.1721], device='cuda:0'), in_proj_covar=tensor([0.1649, 0.1521, 0.1500, 0.1603], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 06:20:29,250 INFO [train.py:968] (0/2) Epoch 10, batch 18900, giga_loss[loss=0.2731, simple_loss=0.3536, pruned_loss=0.0963, over 29059.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3607, pruned_loss=0.1047, over 5700962.00 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3443, pruned_loss=0.09643, over 5739038.60 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.361, pruned_loss=0.1053, over 5696819.55 frames. ], batch size: 128, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:20:32,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=428499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:20:53,375 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-05 06:21:08,907 INFO [train.py:968] (0/2) Epoch 10, batch 18950, libri_loss[loss=0.3427, simple_loss=0.4074, pruned_loss=0.1389, over 19708.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3591, pruned_loss=0.103, over 5697979.28 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3447, pruned_loss=0.09656, over 5734862.43 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3593, pruned_loss=0.1036, over 5697845.86 frames. ], batch size: 187, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:21:13,482 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.889e+02 1.097e+03 1.429e+03 1.939e+03 5.531e+03, threshold=2.858e+03, percent-clipped=5.0
+2023-03-05 06:21:22,876 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=428561.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 06:21:47,267 INFO [train.py:968] (0/2) Epoch 10, batch 19000, libri_loss[loss=0.2946, simple_loss=0.3859, pruned_loss=0.1016, over 29531.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3593, pruned_loss=0.1033, over 5702779.50 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3454, pruned_loss=0.09675, over 5736482.03 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3594, pruned_loss=0.1039, over 5699930.66 frames. ], batch size: 89, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:22:27,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=428642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:22:29,202 INFO [train.py:968] (0/2) Epoch 10, batch 19050, libri_loss[loss=0.3101, simple_loss=0.369, pruned_loss=0.1256, over 29542.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3619, pruned_loss=0.1073, over 5709758.63 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3459, pruned_loss=0.09698, over 5738733.80 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.362, pruned_loss=0.1079, over 5704219.15 frames. ], batch size: 77, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:22:30,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=428645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:22:33,321 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.058e+02 1.230e+03 1.662e+03 2.294e+03 7.030e+03, threshold=3.324e+03, percent-clipped=10.0
+2023-03-05 06:22:40,298 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-05 06:22:56,369 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=428674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:23:13,412 INFO [train.py:968] (0/2) Epoch 10, batch 19100, giga_loss[loss=0.3206, simple_loss=0.38, pruned_loss=0.1307, over 28874.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3649, pruned_loss=0.1122, over 5706051.99 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3462, pruned_loss=0.09723, over 5732051.76 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.365, pruned_loss=0.1126, over 5706327.24 frames. ], batch size: 119, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:23:48,409 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=428738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:23:49,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4894, 1.8591, 1.4457, 1.7756], device='cuda:0'), covar=tensor([0.2447, 0.2286, 0.2585, 0.2043], device='cuda:0'), in_proj_covar=tensor([0.1254, 0.0936, 0.1118, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 06:23:52,314 INFO [train.py:968] (0/2) Epoch 10, batch 19150, giga_loss[loss=0.2718, simple_loss=0.3253, pruned_loss=0.1091, over 23860.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3637, pruned_loss=0.1124, over 5705481.93 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3465, pruned_loss=0.09721, over 5738148.85 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3642, pruned_loss=0.1133, over 5699273.16 frames. ], batch size: 705, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:23:59,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.346e+02 1.283e+03 1.802e+03 2.302e+03 5.688e+03, threshold=3.604e+03, percent-clipped=8.0
+2023-03-05 06:24:34,304 INFO [train.py:968] (0/2) Epoch 10, batch 19200, giga_loss[loss=0.269, simple_loss=0.3376, pruned_loss=0.1002, over 29097.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3613, pruned_loss=0.1114, over 5703809.78 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3469, pruned_loss=0.09746, over 5739680.92 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3617, pruned_loss=0.1123, over 5696623.05 frames. ], batch size: 128, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:24:59,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-05 06:25:18,195 INFO [train.py:968] (0/2) Epoch 10, batch 19250, giga_loss[loss=0.2821, simple_loss=0.3566, pruned_loss=0.1038, over 28497.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3606, pruned_loss=0.1109, over 5712200.98 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3469, pruned_loss=0.09729, over 5743880.49 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3614, pruned_loss=0.1121, over 5701795.95 frames. ], batch size: 71, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:25:18,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5410, 1.7578, 1.8492, 1.3925], device='cuda:0'), covar=tensor([0.1623, 0.2069, 0.1280, 0.1507], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0686, 0.0848, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 06:25:26,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.965e+02 1.133e+03 1.519e+03 1.920e+03 5.988e+03, threshold=3.039e+03, percent-clipped=5.0
+2023-03-05 06:25:44,740 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=428874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:25:49,459 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=428881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:25:51,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=428884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:26:00,359 INFO [train.py:968] (0/2) Epoch 10, batch 19300, giga_loss[loss=0.264, simple_loss=0.3431, pruned_loss=0.09245, over 28889.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3589, pruned_loss=0.1086, over 5717678.81 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.347, pruned_loss=0.09723, over 5745415.97 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3596, pruned_loss=0.1098, over 5707448.82 frames. ], batch size: 199, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:26:16,583 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-05 06:26:17,778 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=428913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:26:40,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=428936.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 06:26:48,866 INFO [train.py:968] (0/2) Epoch 10, batch 19350, giga_loss[loss=0.2332, simple_loss=0.3212, pruned_loss=0.07266, over 28902.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3557, pruned_loss=0.1065, over 5701186.53 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3467, pruned_loss=0.09687, over 5749480.26 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3569, pruned_loss=0.1081, over 5688052.17 frames. ], batch size: 174, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:26:54,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.172e+02 1.020e+03 1.278e+03 1.690e+03 3.655e+03, threshold=2.556e+03, percent-clipped=1.0
+2023-03-05 06:27:14,098 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.11 vs. limit=2.0
+2023-03-05 06:27:19,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=428979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:27:31,488 INFO [train.py:968] (0/2) Epoch 10, batch 19400, giga_loss[loss=0.261, simple_loss=0.325, pruned_loss=0.09845, over 26663.00 frames. ], tot_loss[loss=0.279, simple_loss=0.351, pruned_loss=0.1035, over 5696835.09 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3474, pruned_loss=0.09723, over 5743195.89 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3514, pruned_loss=0.1047, over 5690361.88 frames. ], batch size: 555, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:28:21,026 INFO [train.py:968] (0/2) Epoch 10, batch 19450, giga_loss[loss=0.2331, simple_loss=0.3088, pruned_loss=0.07875, over 28779.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3446, pruned_loss=0.1004, over 5684568.38 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3474, pruned_loss=0.09723, over 5743834.04 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3449, pruned_loss=0.1013, over 5678441.71 frames. ], batch size: 99, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:28:27,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.247e+02 9.230e+02 1.286e+03 1.695e+03 5.940e+03, threshold=2.572e+03, percent-clipped=10.0
+2023-03-05 06:28:55,713 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429079.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 06:28:58,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429082.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 06:29:09,031 INFO [train.py:968] (0/2) Epoch 10, batch 19500, libri_loss[loss=0.31, simple_loss=0.388, pruned_loss=0.116, over 28615.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3398, pruned_loss=0.09818, over 5669240.72 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3478, pruned_loss=0.0974, over 5745410.08 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3395, pruned_loss=0.09884, over 5660932.41 frames. ], batch size: 106, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:29:26,059 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429111.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 06:29:55,100 INFO [train.py:968] (0/2) Epoch 10, batch 19550, giga_loss[loss=0.2817, simple_loss=0.3508, pruned_loss=0.1063, over 28983.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3392, pruned_loss=0.09785, over 5662547.70 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3476, pruned_loss=0.0974, over 5746651.87 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3391, pruned_loss=0.09837, over 5654087.83 frames. ], batch size: 155, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:30:01,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.416e+02 1.056e+03 1.383e+03 1.741e+03 5.363e+03, threshold=2.765e+03, percent-clipped=8.0
+2023-03-05 06:30:16,351 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-05 06:30:41,890 INFO [train.py:968] (0/2) Epoch 10, batch 19600, libri_loss[loss=0.3406, simple_loss=0.4048, pruned_loss=0.1382, over 19572.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3398, pruned_loss=0.0981, over 5656929.84 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3481, pruned_loss=0.09759, over 5739593.24 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3391, pruned_loss=0.09834, over 5656147.32 frames. ], batch size: 188, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:31:23,056 INFO [train.py:968] (0/2) Epoch 10, batch 19650, giga_loss[loss=0.322, simple_loss=0.3733, pruned_loss=0.1354, over 27660.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3397, pruned_loss=0.09798, over 5667293.77 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3489, pruned_loss=0.09784, over 5739255.97 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3382, pruned_loss=0.09796, over 5665166.38 frames. ], batch size: 472, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:31:25,664 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 06:31:26,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:31:29,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.249e+02 1.013e+03 1.262e+03 1.709e+03 4.253e+03, threshold=2.525e+03, percent-clipped=5.0
+2023-03-05 06:31:46,941 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=429272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:32:00,894 INFO [train.py:968] (0/2) Epoch 10, batch 19700, giga_loss[loss=0.2459, simple_loss=0.3149, pruned_loss=0.08842, over 28851.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3381, pruned_loss=0.09722, over 5678894.12 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3493, pruned_loss=0.09793, over 5740025.84 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3363, pruned_loss=0.09711, over 5675144.37 frames. ], batch size: 99, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:32:06,996 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=429300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:32:34,542 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=429334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:32:41,048 INFO [train.py:968] (0/2) Epoch 10, batch 19750, giga_loss[loss=0.262, simple_loss=0.3362, pruned_loss=0.09389, over 29013.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3368, pruned_loss=0.09653, over 5678856.05 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3502, pruned_loss=0.09825, over 5734002.61 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3343, pruned_loss=0.0961, over 5680466.98 frames. ], batch size: 145, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:32:49,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.234e+02 9.899e+02 1.342e+03 2.195e+03 9.204e+03, threshold=2.684e+03, percent-clipped=21.0
+2023-03-05 06:32:49,933 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:33:21,190 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:33:22,152 INFO [train.py:968] (0/2) Epoch 10, batch 19800, giga_loss[loss=0.2308, simple_loss=0.3074, pruned_loss=0.07711, over 29017.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.334, pruned_loss=0.09524, over 5689401.41 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3504, pruned_loss=0.09834, over 5735567.03 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3317, pruned_loss=0.09481, over 5688941.43 frames. ], batch size: 164, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:33:23,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:33:49,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:34:02,742 INFO [train.py:968] (0/2) Epoch 10, batch 19850, giga_loss[loss=0.2716, simple_loss=0.3444, pruned_loss=0.0994, over 28766.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3312, pruned_loss=0.09335, over 5699864.07 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.351, pruned_loss=0.09831, over 5740213.18 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3283, pruned_loss=0.09292, over 5693888.13 frames. ], batch size: 284, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:34:07,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3248, 1.4540, 1.2393, 1.4586], device='cuda:0'), covar=tensor([0.0767, 0.0333, 0.0329, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:0')
+2023-03-05 06:34:11,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.400e+02 9.762e+02 1.223e+03 1.654e+03 6.072e+03, threshold=2.447e+03, percent-clipped=8.0
+2023-03-05 06:34:43,523 INFO [train.py:968] (0/2) Epoch 10, batch 19900, giga_loss[loss=0.2326, simple_loss=0.3109, pruned_loss=0.07714, over 28781.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3288, pruned_loss=0.09208, over 5711131.34 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3511, pruned_loss=0.09811, over 5744033.27 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3257, pruned_loss=0.09175, over 5702104.74 frames. ], batch size: 284, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:34:45,895 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:34:47,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:35:09,860 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:35:16,743 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=429538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:35:20,695 INFO [train.py:968] (0/2) Epoch 10, batch 19950, giga_loss[loss=0.2568, simple_loss=0.3317, pruned_loss=0.0909, over 28835.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3289, pruned_loss=0.09206, over 5725295.91 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3525, pruned_loss=0.09872, over 5751114.45 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3242, pruned_loss=0.09096, over 5710239.33 frames. ], batch size: 199, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:35:30,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.661e+02 1.134e+03 1.349e+03 1.818e+03 3.556e+03, threshold=2.698e+03, percent-clipped=11.0
+2023-03-05 06:36:01,304 INFO [train.py:968] (0/2) Epoch 10, batch 20000, libri_loss[loss=0.2673, simple_loss=0.3542, pruned_loss=0.09018, over 29569.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3274, pruned_loss=0.09128, over 5718851.21 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3528, pruned_loss=0.09864, over 5750816.70 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3225, pruned_loss=0.09023, over 5705872.42 frames. ], batch size: 77, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:36:39,433 INFO [train.py:968] (0/2) Epoch 10, batch 20050, giga_loss[loss=0.2136, simple_loss=0.2998, pruned_loss=0.06369, over 28904.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3269, pruned_loss=0.09087, over 5722752.16 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.353, pruned_loss=0.0987, over 5754822.57 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3219, pruned_loss=0.08975, over 5707910.78 frames. ], batch size: 174, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:36:42,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:36:47,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.861e+02 1.030e+03 1.367e+03 1.993e+03 8.026e+03, threshold=2.734e+03, percent-clipped=12.0
+2023-03-05 06:37:03,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:37:05,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7830, 4.4223, 1.8873, 1.9728], device='cuda:0'), covar=tensor([0.0843, 0.0276, 0.0783, 0.1142], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0493, 0.0329, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 06:37:18,046 INFO [train.py:968] (0/2) Epoch 10, batch 20100, giga_loss[loss=0.2441, simple_loss=0.3112, pruned_loss=0.08848, over 28713.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3252, pruned_loss=0.08991, over 5725895.80 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3531, pruned_loss=0.09875, over 5755618.45 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3211, pruned_loss=0.08896, over 5713504.89 frames. ], batch size: 119, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:37:34,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:37:35,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9177, 1.0999, 3.3620, 3.0035], device='cuda:0'), covar=tensor([0.1785, 0.2600, 0.0488, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.0637, 0.0568, 0.0825, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 06:38:04,544 INFO [train.py:968] (0/2) Epoch 10, batch 20150, giga_loss[loss=0.2675, simple_loss=0.3437, pruned_loss=0.09567, over 28883.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3305, pruned_loss=0.09346, over 5720934.66 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3536, pruned_loss=0.09899, over 5757558.21 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3265, pruned_loss=0.09239, over 5708935.89 frames. ], batch size: 119, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:38:13,212 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.219e+02 1.025e+03 1.343e+03 1.770e+03 5.590e+03, threshold=2.685e+03, percent-clipped=6.0
+2023-03-05 06:38:47,551 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:38:49,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:38:49,808 INFO [train.py:968] (0/2) Epoch 10, batch 20200, giga_loss[loss=0.2874, simple_loss=0.3529, pruned_loss=0.111, over 28751.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3374, pruned_loss=0.09785, over 5703426.55 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.354, pruned_loss=0.09906, over 5750817.43 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.333, pruned_loss=0.09676, over 5698127.55 frames. ], batch size: 99, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:39:13,794 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:39:16,875 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3547, 4.1670, 3.9376, 1.6976], device='cuda:0'), covar=tensor([0.0509, 0.0668, 0.0622, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.0999, 0.0937, 0.0819, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 06:39:16,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:39:17,906 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:39:38,349 INFO [train.py:968] (0/2) Epoch 10, batch 20250, giga_loss[loss=0.3008, simple_loss=0.3662, pruned_loss=0.1177, over 28893.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3453, pruned_loss=0.103, over 5704459.49 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3545, pruned_loss=0.09933, over 5754378.06 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3408, pruned_loss=0.1019, over 5695357.14 frames. ], batch size: 199, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:39:45,583 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:39:47,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:39:47,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.946e+02 1.381e+03 1.762e+03 2.398e+03 9.876e+03, threshold=3.524e+03, percent-clipped=20.0
+2023-03-05 06:39:50,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:40:01,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7597, 1.7314, 1.3328, 1.3785], device='cuda:0'), covar=tensor([0.0690, 0.0563, 0.0894, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0435, 0.0501, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 06:40:14,823 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:40:16,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7024, 1.1483, 5.1219, 3.3363], device='cuda:0'), covar=tensor([0.1510, 0.2722, 0.0338, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.0640, 0.0573, 0.0830, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 06:40:21,971 INFO [train.py:968] (0/2) Epoch 10, batch 20300, giga_loss[loss=0.2755, simple_loss=0.3538, pruned_loss=0.09857, over 28824.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.35, pruned_loss=0.1052, over 5702046.42 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3546, pruned_loss=0.09932, over 5758255.98 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.346, pruned_loss=0.1045, over 5688997.56 frames. ], batch size: 186, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:40:37,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:40:57,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5327, 1.7546, 1.8052, 1.3554], device='cuda:0'), covar=tensor([0.1587, 0.2132, 0.1318, 0.1517], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0686, 0.0848, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 06:41:07,299 INFO [train.py:968] (0/2) Epoch 10, batch 20350, giga_loss[loss=0.2852, simple_loss=0.3598, pruned_loss=0.1053, over 28646.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3548, pruned_loss=0.1069, over 5703549.10 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3544, pruned_loss=0.09913, over 5762012.35 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3517, pruned_loss=0.1068, over 5687930.96 frames. ], batch size: 242, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:41:15,205 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.844e+02 1.096e+03 1.307e+03 1.697e+03 3.991e+03, threshold=2.615e+03, percent-clipped=3.0
+2023-03-05 06:41:52,092 INFO [train.py:968] (0/2) Epoch 10, batch 20400, giga_loss[loss=0.3532, simple_loss=0.4077, pruned_loss=0.1493, over 27655.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3586, pruned_loss=0.1083, over 5700858.02 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3543, pruned_loss=0.09899, over 5764869.79 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3563, pruned_loss=0.1086, over 5684118.08 frames. ], batch size: 472, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:41:57,603 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-430000.pt
+2023-03-05 06:42:35,752 INFO [train.py:968] (0/2) Epoch 10, batch 20450, giga_loss[loss=0.2682, simple_loss=0.347, pruned_loss=0.09464, over 28931.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3619, pruned_loss=0.1101, over 5704070.55 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3543, pruned_loss=0.09889, over 5767699.19 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3603, pruned_loss=0.1105, over 5687159.28 frames. ], batch size: 136, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:42:44,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5077, 2.2355, 1.6290, 0.7214], device='cuda:0'), covar=tensor([0.4193, 0.2133, 0.3305, 0.4403], device='cuda:0'), in_proj_covar=tensor([0.1541, 0.1461, 0.1481, 0.1259], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 06:42:45,187 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.755e+02 1.076e+03 1.402e+03 1.730e+03 6.707e+03, threshold=2.804e+03, percent-clipped=5.0
+2023-03-05 06:42:46,765 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=430056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:42:48,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=430059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:43:04,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=430073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:43:16,313 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=430088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:43:21,604 INFO [train.py:968] (0/2) Epoch 10, batch 20500, giga_loss[loss=0.2452, simple_loss=0.3247, pruned_loss=0.08282, over 28724.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3567, pruned_loss=0.1067, over 5700239.56 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3548, pruned_loss=0.09928, over 5768812.11 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.355, pruned_loss=0.1068, over 5685298.57 frames. ], batch size: 262, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:43:21,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9817, 2.1814, 2.2005, 1.7543], device='cuda:0'), covar=tensor([0.1681, 0.2030, 0.1303, 0.1566], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0686, 0.0849, 0.0758], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 06:44:03,964 INFO [train.py:968] (0/2) Epoch 10, batch 20550, giga_loss[loss=0.2628, simple_loss=0.3401, pruned_loss=0.09275, over 28631.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3547, pruned_loss=0.1047, over 5703739.82 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3552, pruned_loss=0.0995, over 5767736.13 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3529, pruned_loss=0.1047, over 5691299.89 frames. ], batch size: 85, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:44:14,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.142e+02 1.057e+03 1.379e+03 2.236e+03 4.908e+03, threshold=2.759e+03, percent-clipped=9.0
+2023-03-05 06:44:47,495 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=430193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:44:48,935 INFO [train.py:968] (0/2) Epoch 10, batch 20600, giga_loss[loss=0.2286, simple_loss=0.3181, pruned_loss=0.06959, over 28920.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3539, pruned_loss=0.1038, over 5697445.44 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3552, pruned_loss=0.09964, over 5769932.85 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3526, pruned_loss=0.1037, over 5684665.50 frames. ], batch size: 174, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:45:30,643 INFO [train.py:968] (0/2) Epoch 10, batch 20650, giga_loss[loss=0.2962, simple_loss=0.3705, pruned_loss=0.111, over 28972.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3551, pruned_loss=0.1044, over 5695696.11 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3552, pruned_loss=0.09983, over 5764409.62 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.354, pruned_loss=0.1043, over 5689431.02 frames. ], batch size: 164, lr: 3.24e-03, grad_scale: 2.0
+2023-03-05 06:45:33,246 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-05 06:45:40,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.888e+02 1.200e+03 1.580e+03 2.252e+03 7.312e+03, threshold=3.161e+03, percent-clipped=18.0
+2023-03-05 06:46:13,529 INFO [train.py:968] (0/2) Epoch 10, batch 20700, giga_loss[loss=0.2953, simple_loss=0.3667, pruned_loss=0.1119, over 28786.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3585, pruned_loss=0.1072, over 5688440.93 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3553, pruned_loss=0.1, over 5757918.29 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3575, pruned_loss=0.107, over 5688101.98 frames. ], batch size: 186, lr: 3.24e-03, grad_scale: 2.0
+2023-03-05 06:46:38,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1593, 1.5902, 1.4957, 1.0926], device='cuda:0'), covar=tensor([0.1434, 0.2070, 0.1153, 0.1320], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0692, 0.0851, 0.0761], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 06:46:47,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3925, 1.5103, 1.3953, 1.2565], device='cuda:0'), covar=tensor([0.1452, 0.1319, 0.1121, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.1620, 0.1503, 0.1495, 0.1592], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 06:46:55,140 INFO [train.py:968] (0/2) Epoch 10, batch 20750, giga_loss[loss=0.2737, simple_loss=0.3461, pruned_loss=0.1007, over 28769.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3612, pruned_loss=0.1096, over 5694726.59 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3561, pruned_loss=0.1007, over 5759720.94 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3598, pruned_loss=0.1092, over 5689985.10 frames. ], batch size: 99, lr: 3.24e-03, grad_scale: 2.0
+2023-03-05 06:47:05,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.752e+02 1.301e+03 1.706e+03 2.410e+03 5.916e+03, threshold=3.413e+03, percent-clipped=12.0
+2023-03-05 06:47:43,329 INFO [train.py:968] (0/2) Epoch 10, batch 20800, giga_loss[loss=0.2916, simple_loss=0.3664, pruned_loss=0.1083, over 28056.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3611, pruned_loss=0.1094, over 5697414.58 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3562, pruned_loss=0.1007, over 5751328.85 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3599, pruned_loss=0.1091, over 5700920.57 frames. ], batch size: 412, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:48:26,629 INFO [train.py:968] (0/2) Epoch 10, batch 20850, giga_loss[loss=0.2793, simple_loss=0.3538, pruned_loss=0.1024, over 28882.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3621, pruned_loss=0.1104, over 5699047.81 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3565, pruned_loss=0.1009, over 5752698.34 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.361, pruned_loss=0.1102, over 5699284.02 frames. ], batch size: 112, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:48:31,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=430448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:48:32,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8209, 2.1532, 1.7270, 1.4324], device='cuda:0'), covar=tensor([0.1912, 0.1521, 0.1854, 0.1846], device='cuda:0'), in_proj_covar=tensor([0.1628, 0.1508, 0.1501, 0.1603], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 06:48:37,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.892e+02 1.160e+03 1.416e+03 2.052e+03 3.709e+03, threshold=2.831e+03, percent-clipped=3.0
+2023-03-05 06:48:39,785 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.36 vs. limit=5.0
+2023-03-05 06:48:44,935 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5641, 1.6593, 1.4099, 1.3112], device='cuda:0'), covar=tensor([0.1663, 0.1468, 0.1416, 0.1565], device='cuda:0'), in_proj_covar=tensor([0.1628, 0.1510, 0.1502, 0.1605], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 06:49:07,867 INFO [train.py:968] (0/2) Epoch 10, batch 20900, giga_loss[loss=0.2582, simple_loss=0.3418, pruned_loss=0.08726, over 29054.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3628, pruned_loss=0.1106, over 5708612.43 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3571, pruned_loss=0.1014, over 5755793.97 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3615, pruned_loss=0.1102, over 5705014.61 frames. ], batch size: 164, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:49:16,069 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2605, 1.4014, 1.2841, 1.5478], device='cuda:0'), covar=tensor([0.0758, 0.0314, 0.0311, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0078, 0.0055, 0.0050, 0.0085], device='cuda:0')
+2023-03-05 06:49:47,864 INFO [train.py:968] (0/2) Epoch 10, batch 20950, giga_loss[loss=0.279, simple_loss=0.3537, pruned_loss=0.1022, over 28614.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3626, pruned_loss=0.1096, over 5686563.51 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3578, pruned_loss=0.1018, over 5733331.97 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3611, pruned_loss=0.1092, over 5701289.63 frames. ], batch size: 85, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:49:58,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.316e+02 1.149e+03 1.592e+03 2.385e+03 9.477e+03, threshold=3.185e+03, percent-clipped=12.0
+2023-03-05 06:50:06,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=430568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:50:07,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5501, 1.6656, 1.1937, 1.3133], device='cuda:0'), covar=tensor([0.0664, 0.0458, 0.0957, 0.0893], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0430, 0.0497, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 06:50:24,841 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=430591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:50:26,692 INFO [train.py:968] (0/2) Epoch 10, batch 21000, giga_loss[loss=0.2867, simple_loss=0.3609, pruned_loss=0.1062, over 28531.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3637, pruned_loss=0.1093, over 5699326.87 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3578, pruned_loss=0.1019, over 5736821.41 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3626, pruned_loss=0.109, over 5707159.23 frames. ], batch size: 71, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:50:26,696 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 06:50:35,599 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2214, simple_loss=0.327, pruned_loss=0.05794, over 944034.00 frames. 
+2023-03-05 06:50:35,599 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 06:50:35,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=430594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:50:52,888 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=430616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:51:00,597 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=430623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:51:16,750 INFO [train.py:968] (0/2) Epoch 10, batch 21050, giga_loss[loss=0.2672, simple_loss=0.3447, pruned_loss=0.09484, over 28823.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3633, pruned_loss=0.109, over 5708957.61 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3578, pruned_loss=0.1022, over 5739602.65 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3625, pruned_loss=0.1087, over 5712420.29 frames. ], batch size: 199, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:51:24,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.841e+02 1.044e+03 1.434e+03 1.939e+03 6.488e+03, threshold=2.868e+03, percent-clipped=6.0
+2023-03-05 06:51:49,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5871, 2.2240, 2.5931, 2.0796], device='cuda:0'), covar=tensor([0.1022, 0.1746, 0.1173, 0.1374], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0722, 0.0658, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 06:51:55,283 INFO [train.py:968] (0/2) Epoch 10, batch 21100, giga_loss[loss=0.288, simple_loss=0.3579, pruned_loss=0.109, over 28232.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3608, pruned_loss=0.1083, over 5705348.86 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3581, pruned_loss=0.1025, over 5741778.46 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3599, pruned_loss=0.1078, over 5705585.11 frames. ], batch size: 368, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:52:09,726 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=430711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:52:12,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=430714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:52:35,147 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=430743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:52:36,348 INFO [train.py:968] (0/2) Epoch 10, batch 21150, giga_loss[loss=0.2564, simple_loss=0.339, pruned_loss=0.08692, over 29120.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3589, pruned_loss=0.1072, over 5707587.90 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3581, pruned_loss=0.1025, over 5744297.86 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3582, pruned_loss=0.1069, over 5704937.34 frames. ], batch size: 128, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:52:46,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.523e+02 1.088e+03 1.346e+03 1.981e+03 4.519e+03, threshold=2.691e+03, percent-clipped=5.0
+2023-03-05 06:52:47,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=430757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:53:16,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5624, 2.0358, 1.8805, 1.4394], device='cuda:0'), covar=tensor([0.1690, 0.2275, 0.1360, 0.1662], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0693, 0.0848, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 06:53:17,217 INFO [train.py:968] (0/2) Epoch 10, batch 21200, giga_loss[loss=0.2792, simple_loss=0.3481, pruned_loss=0.1052, over 28576.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3582, pruned_loss=0.1073, over 5707698.50 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3585, pruned_loss=0.1028, over 5738747.18 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3573, pruned_loss=0.1068, over 5710539.17 frames. ], batch size: 336, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 06:53:58,155 INFO [train.py:968] (0/2) Epoch 10, batch 21250, giga_loss[loss=0.2817, simple_loss=0.3549, pruned_loss=0.1043, over 28431.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3589, pruned_loss=0.1079, over 5706888.09 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3584, pruned_loss=0.1029, over 5741557.92 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3582, pruned_loss=0.1076, over 5706275.45 frames. ], batch size: 71, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 06:54:10,991 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.006e+02 9.760e+02 1.291e+03 1.796e+03 3.565e+03, threshold=2.582e+03, percent-clipped=6.0
+2023-03-05 06:54:39,595 INFO [train.py:968] (0/2) Epoch 10, batch 21300, giga_loss[loss=0.2702, simple_loss=0.3529, pruned_loss=0.09372, over 28966.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3601, pruned_loss=0.1084, over 5700895.48 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3591, pruned_loss=0.1034, over 5733586.36 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.359, pruned_loss=0.1077, over 5706995.65 frames. ], batch size: 136, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 06:55:20,844 INFO [train.py:968] (0/2) Epoch 10, batch 21350, giga_loss[loss=0.2435, simple_loss=0.3287, pruned_loss=0.07914, over 28620.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3584, pruned_loss=0.1066, over 5710797.45 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.359, pruned_loss=0.1036, over 5736905.23 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3576, pruned_loss=0.106, over 5712174.33 frames. ], batch size: 60, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:55:32,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.736e+02 9.922e+02 1.230e+03 1.598e+03 5.853e+03, threshold=2.460e+03, percent-clipped=6.0
+2023-03-05 06:56:01,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=430991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:56:03,184 INFO [train.py:968] (0/2) Epoch 10, batch 21400, giga_loss[loss=0.2443, simple_loss=0.3263, pruned_loss=0.08118, over 28967.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3562, pruned_loss=0.1049, over 5703924.15 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3593, pruned_loss=0.1039, over 5740314.00 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3552, pruned_loss=0.1042, over 5701606.62 frames. ], batch size: 164, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:56:07,318 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=430998.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 06:56:44,934 INFO [train.py:968] (0/2) Epoch 10, batch 21450, giga_loss[loss=0.2445, simple_loss=0.3268, pruned_loss=0.08111, over 28856.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3555, pruned_loss=0.1053, over 5698793.61 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3595, pruned_loss=0.1041, over 5741836.42 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3546, pruned_loss=0.1046, over 5695259.15 frames. ], batch size: 145, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:56:54,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.336e+02 1.029e+03 1.261e+03 1.658e+03 8.033e+03, threshold=2.522e+03, percent-clipped=7.0
+2023-03-05 06:57:23,742 INFO [train.py:968] (0/2) Epoch 10, batch 21500, giga_loss[loss=0.3007, simple_loss=0.3619, pruned_loss=0.1197, over 27597.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3535, pruned_loss=0.1046, over 5703677.75 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3596, pruned_loss=0.1042, over 5743515.85 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3526, pruned_loss=0.104, over 5698679.39 frames. ], batch size: 472, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:57:53,421 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 06:57:56,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=431132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:57:57,389 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=431134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:58:00,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=431137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:58:04,656 INFO [train.py:968] (0/2) Epoch 10, batch 21550, giga_loss[loss=0.2521, simple_loss=0.3312, pruned_loss=0.08645, over 28498.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3508, pruned_loss=0.1036, over 5697308.41 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3601, pruned_loss=0.1045, over 5740857.17 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3495, pruned_loss=0.1029, over 5694984.81 frames. ], batch size: 78, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:58:15,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.555e+02 1.119e+03 1.374e+03 1.902e+03 3.231e+03, threshold=2.747e+03, percent-clipped=9.0
+2023-03-05 06:58:24,063 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=431166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:58:46,305 INFO [train.py:968] (0/2) Epoch 10, batch 21600, giga_loss[loss=0.2954, simple_loss=0.3621, pruned_loss=0.1143, over 29046.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3516, pruned_loss=0.105, over 5691129.23 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3605, pruned_loss=0.1048, over 5742103.44 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3501, pruned_loss=0.1041, over 5687690.11 frames. ], batch size: 213, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 06:58:47,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4399, 2.0726, 1.5511, 0.6200], device='cuda:0'), covar=tensor([0.3799, 0.1902, 0.2669, 0.4089], device='cuda:0'), in_proj_covar=tensor([0.1512, 0.1426, 0.1462, 0.1244], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 06:59:26,563 INFO [train.py:968] (0/2) Epoch 10, batch 21650, giga_loss[loss=0.264, simple_loss=0.336, pruned_loss=0.096, over 28966.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3501, pruned_loss=0.1043, over 5697086.44 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3606, pruned_loss=0.105, over 5745548.82 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3486, pruned_loss=0.1034, over 5690372.05 frames. ], batch size: 164, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 06:59:37,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.842e+02 1.176e+03 1.512e+03 1.941e+03 4.099e+03, threshold=3.025e+03, percent-clipped=10.0
+2023-03-05 06:59:53,602 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=431275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:59:56,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=431278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:00:07,623 INFO [train.py:968] (0/2) Epoch 10, batch 21700, giga_loss[loss=0.2579, simple_loss=0.3259, pruned_loss=0.09495, over 28915.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3471, pruned_loss=0.103, over 5708181.34 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3608, pruned_loss=0.1052, over 5748696.57 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3455, pruned_loss=0.102, over 5699145.79 frames. ], batch size: 99, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:00:19,266 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=431307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:00:50,342 INFO [train.py:968] (0/2) Epoch 10, batch 21750, giga_loss[loss=0.2371, simple_loss=0.3064, pruned_loss=0.08393, over 28908.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3441, pruned_loss=0.1016, over 5708249.87 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3609, pruned_loss=0.1053, over 5749555.09 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3427, pruned_loss=0.1008, over 5700233.42 frames. ], batch size: 106, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:01:00,413 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.146e+02 1.098e+03 1.342e+03 1.809e+03 5.048e+03, threshold=2.683e+03, percent-clipped=5.0
+2023-03-05 07:01:15,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=431373.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 07:01:30,075 INFO [train.py:968] (0/2) Epoch 10, batch 21800, giga_loss[loss=0.2714, simple_loss=0.3466, pruned_loss=0.09816, over 29013.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3411, pruned_loss=0.09975, over 5714564.06 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3611, pruned_loss=0.1055, over 5749513.70 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3396, pruned_loss=0.09881, over 5707738.39 frames. ], batch size: 164, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:01:34,524 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=431400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:01:46,080 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-05 07:01:59,452 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-05 07:02:09,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3260, 3.0858, 1.3043, 1.6198], device='cuda:0'), covar=tensor([0.0884, 0.0323, 0.0923, 0.1207], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0497, 0.0330, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:0')
+2023-03-05 07:02:09,683 INFO [train.py:968] (0/2) Epoch 10, batch 21850, giga_loss[loss=0.2575, simple_loss=0.3385, pruned_loss=0.08819, over 28993.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3424, pruned_loss=0.1005, over 5706385.89 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3616, pruned_loss=0.1061, over 5741464.23 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.34, pruned_loss=0.09905, over 5707382.42 frames. ], batch size: 164, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:02:20,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.778e+02 1.074e+03 1.370e+03 2.008e+03 4.452e+03, threshold=2.741e+03, percent-clipped=10.0
+2023-03-05 07:02:34,611 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=431476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:02:38,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6587, 4.4596, 4.2334, 1.9493], device='cuda:0'), covar=tensor([0.0452, 0.0623, 0.0637, 0.2104], device='cuda:0'), in_proj_covar=tensor([0.1000, 0.0939, 0.0820, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 07:02:50,828 INFO [train.py:968] (0/2) Epoch 10, batch 21900, giga_loss[loss=0.3184, simple_loss=0.3889, pruned_loss=0.1239, over 28699.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3451, pruned_loss=0.1016, over 5694380.99 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3618, pruned_loss=0.1063, over 5732301.54 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3427, pruned_loss=0.1001, over 5702398.53 frames. ], batch size: 262, lr: 3.24e-03, grad_scale: 2.0
+2023-03-05 07:03:01,063 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4109, 4.2051, 3.9972, 1.8688], device='cuda:0'), covar=tensor([0.0412, 0.0578, 0.0571, 0.2091], device='cuda:0'), in_proj_covar=tensor([0.0998, 0.0937, 0.0819, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 07:03:10,196 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=431516.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 07:03:12,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=431519.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 07:03:34,899 INFO [train.py:968] (0/2) Epoch 10, batch 21950, giga_loss[loss=0.3029, simple_loss=0.3698, pruned_loss=0.118, over 28572.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3491, pruned_loss=0.1034, over 5692320.11 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3618, pruned_loss=0.1066, over 5737265.48 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3467, pruned_loss=0.1019, over 5693337.44 frames. ], batch size: 336, lr: 3.24e-03, grad_scale: 2.0
+2023-03-05 07:03:39,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=431548.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 07:03:48,688 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.637e+02 9.839e+02 1.392e+03 1.934e+03 8.092e+03, threshold=2.784e+03, percent-clipped=13.0
+2023-03-05 07:04:03,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4583, 1.7553, 1.4175, 1.6399], device='cuda:0'), covar=tensor([0.2346, 0.2236, 0.2544, 0.2015], device='cuda:0'), in_proj_covar=tensor([0.1260, 0.0939, 0.1116, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 07:04:16,594 INFO [train.py:968] (0/2) Epoch 10, batch 22000, giga_loss[loss=0.2403, simple_loss=0.3146, pruned_loss=0.08294, over 28422.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3511, pruned_loss=0.1038, over 5690229.04 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3618, pruned_loss=0.1067, over 5731155.86 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.349, pruned_loss=0.1024, over 5695780.14 frames. ], batch size: 65, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:04:35,505 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 07:05:00,225 INFO [train.py:968] (0/2) Epoch 10, batch 22050, giga_loss[loss=0.2298, simple_loss=0.3086, pruned_loss=0.0755, over 28604.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3517, pruned_loss=0.1032, over 5698253.81 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3623, pruned_loss=0.1072, over 5734275.18 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3493, pruned_loss=0.1015, over 5698951.78 frames. ], batch size: 85, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:05:13,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.308e+02 1.022e+03 1.210e+03 1.682e+03 7.277e+03, threshold=2.420e+03, percent-clipped=7.0
+2023-03-05 07:05:44,593 INFO [train.py:968] (0/2) Epoch 10, batch 22100, giga_loss[loss=0.2579, simple_loss=0.3296, pruned_loss=0.09304, over 28810.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3512, pruned_loss=0.1029, over 5689954.90 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3626, pruned_loss=0.1077, over 5728855.52 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3488, pruned_loss=0.1011, over 5695373.45 frames. ], batch size: 112, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:06:23,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4013, 3.3137, 1.5671, 1.4145], device='cuda:0'), covar=tensor([0.0870, 0.0336, 0.0816, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0499, 0.0331, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 07:06:26,260 INFO [train.py:968] (0/2) Epoch 10, batch 22150, giga_loss[loss=0.3454, simple_loss=0.3898, pruned_loss=0.1504, over 28696.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3523, pruned_loss=0.104, over 5698711.15 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3631, pruned_loss=0.1082, over 5733448.80 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3496, pruned_loss=0.1019, over 5697874.59 frames. ], batch size: 99, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:06:35,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6153, 3.6065, 1.5878, 1.6120], device='cuda:0'), covar=tensor([0.0794, 0.0357, 0.0837, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0498, 0.0331, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 07:06:40,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.078e+02 1.186e+03 1.508e+03 2.065e+03 1.532e+04, threshold=3.016e+03, percent-clipped=19.0
+2023-03-05 07:06:53,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=431775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:07:01,058 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3655, 2.9352, 1.4952, 1.4410], device='cuda:0'), covar=tensor([0.0812, 0.0284, 0.0809, 0.1178], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0500, 0.0331, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 07:07:09,746 INFO [train.py:968] (0/2) Epoch 10, batch 22200, giga_loss[loss=0.2635, simple_loss=0.3434, pruned_loss=0.09182, over 28996.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3523, pruned_loss=0.1043, over 5690993.29 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3633, pruned_loss=0.1085, over 5726593.02 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3498, pruned_loss=0.1023, over 5696188.47 frames. ], batch size: 164, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:07:54,091 INFO [train.py:968] (0/2) Epoch 10, batch 22250, giga_loss[loss=0.2706, simple_loss=0.3494, pruned_loss=0.09586, over 28525.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3548, pruned_loss=0.1061, over 5695202.35 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3636, pruned_loss=0.1089, over 5728354.97 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3522, pruned_loss=0.104, over 5696925.21 frames. ], batch size: 65, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:07:54,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8716, 2.5472, 1.7974, 1.3512], device='cuda:0'), covar=tensor([0.2248, 0.1266, 0.1723, 0.2253], device='cuda:0'), in_proj_covar=tensor([0.1650, 0.1532, 0.1519, 0.1619], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 07:07:59,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=431851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:08:06,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.960e+02 1.187e+03 1.570e+03 2.329e+03 6.838e+03, threshold=3.140e+03, percent-clipped=12.0
+2023-03-05 07:08:11,396 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-05 07:08:15,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6427, 4.4448, 1.7516, 1.8469], device='cuda:0'), covar=tensor([0.0853, 0.0267, 0.0828, 0.1129], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0499, 0.0331, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 07:08:29,585 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=431885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:08:35,766 INFO [train.py:968] (0/2) Epoch 10, batch 22300, libri_loss[loss=0.3152, simple_loss=0.3923, pruned_loss=0.119, over 29537.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3583, pruned_loss=0.1077, over 5707008.23 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3645, pruned_loss=0.1096, over 5731624.43 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3552, pruned_loss=0.1054, over 5704695.51 frames. ], batch size: 80, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:08:56,830 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=431918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:09:00,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=431921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:09:05,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5033, 4.2943, 4.0471, 1.7338], device='cuda:0'), covar=tensor([0.0435, 0.0584, 0.0654, 0.2280], device='cuda:0'), in_proj_covar=tensor([0.0998, 0.0933, 0.0824, 0.0637], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 07:09:18,125 INFO [train.py:968] (0/2) Epoch 10, batch 22350, giga_loss[loss=0.2909, simple_loss=0.3681, pruned_loss=0.1069, over 28591.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3593, pruned_loss=0.1079, over 5709951.45 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3647, pruned_loss=0.1097, over 5732393.81 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3567, pruned_loss=0.1059, over 5707335.23 frames. ], batch size: 242, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:09:20,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7207, 1.2447, 5.1408, 3.6682], device='cuda:0'), covar=tensor([0.1495, 0.2545, 0.0323, 0.0676], device='cuda:0'), in_proj_covar=tensor([0.0643, 0.0577, 0.0834, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 07:09:22,965 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=431950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:09:29,686 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.693e+02 1.108e+03 1.351e+03 1.963e+03 3.849e+03, threshold=2.702e+03, percent-clipped=3.0
+2023-03-05 07:09:54,958 INFO [train.py:968] (0/2) Epoch 10, batch 22400, giga_loss[loss=0.2788, simple_loss=0.351, pruned_loss=0.1033, over 28934.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.36, pruned_loss=0.1082, over 5713187.22 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3657, pruned_loss=0.1106, over 5730160.47 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3568, pruned_loss=0.1058, over 5711644.37 frames. ], batch size: 106, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:09:55,212 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=431994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:09:57,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=431997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:09:58,898 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-432000.pt
+2023-03-05 07:10:20,070 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=432026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:10:36,321 INFO [train.py:968] (0/2) Epoch 10, batch 22450, giga_loss[loss=0.2768, simple_loss=0.3563, pruned_loss=0.09868, over 28969.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3606, pruned_loss=0.1088, over 5712462.37 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3665, pruned_loss=0.1113, over 5731997.64 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3571, pruned_loss=0.1062, over 5708918.51 frames. ], batch size: 213, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:10:49,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.544e+02 1.260e+03 1.569e+03 1.958e+03 4.244e+03, threshold=3.137e+03, percent-clipped=10.0
+2023-03-05 07:11:15,223 INFO [train.py:968] (0/2) Epoch 10, batch 22500, libri_loss[loss=0.356, simple_loss=0.4174, pruned_loss=0.1473, over 29482.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3608, pruned_loss=0.1092, over 5720586.05 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3677, pruned_loss=0.1124, over 5737877.73 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3566, pruned_loss=0.1059, over 5711677.45 frames. ], batch size: 85, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:11:55,249 INFO [train.py:968] (0/2) Epoch 10, batch 22550, giga_loss[loss=0.2304, simple_loss=0.3186, pruned_loss=0.07112, over 28884.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3589, pruned_loss=0.108, over 5721058.52 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3683, pruned_loss=0.113, over 5739164.60 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3548, pruned_loss=0.1047, over 5712330.67 frames. ], batch size: 174, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:12:08,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.589e+02 1.244e+03 1.588e+03 2.267e+03 7.625e+03, threshold=3.176e+03, percent-clipped=9.0
+2023-03-05 07:12:37,829 INFO [train.py:968] (0/2) Epoch 10, batch 22600, libri_loss[loss=0.2808, simple_loss=0.3441, pruned_loss=0.1088, over 28189.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3556, pruned_loss=0.1065, over 5718623.06 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3681, pruned_loss=0.1132, over 5733838.53 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3521, pruned_loss=0.1034, over 5716768.70 frames. ], batch size: 62, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:13:16,103 INFO [train.py:968] (0/2) Epoch 10, batch 22650, giga_loss[loss=0.219, simple_loss=0.3014, pruned_loss=0.06826, over 28400.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3524, pruned_loss=0.1049, over 5703609.26 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3679, pruned_loss=0.1133, over 5721553.11 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3494, pruned_loss=0.1022, over 5712565.44 frames. ], batch size: 65, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:13:28,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.001e+02 1.052e+03 1.390e+03 1.952e+03 6.956e+03, threshold=2.780e+03, percent-clipped=8.0
+2023-03-05 07:13:29,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=432260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:13:36,841 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5987, 1.7959, 1.5805, 1.5648], device='cuda:0'), covar=tensor([0.1307, 0.1777, 0.1965, 0.1825], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0733, 0.0663, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 07:13:40,403 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 07:13:55,617 INFO [train.py:968] (0/2) Epoch 10, batch 22700, giga_loss[loss=0.2866, simple_loss=0.3716, pruned_loss=0.1008, over 28638.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3523, pruned_loss=0.1038, over 5707551.59 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3682, pruned_loss=0.1137, over 5723759.32 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3492, pruned_loss=0.1011, over 5712385.35 frames. ], batch size: 307, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:14:14,089 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=432311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:14:20,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1190, 1.1210, 4.1964, 3.1412], device='cuda:0'), covar=tensor([0.1570, 0.2399, 0.0381, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0642, 0.0575, 0.0836, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 07:14:37,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=432339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:14:41,242 INFO [train.py:968] (0/2) Epoch 10, batch 22750, giga_loss[loss=0.2895, simple_loss=0.3727, pruned_loss=0.1032, over 29060.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3533, pruned_loss=0.1031, over 5707770.22 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3683, pruned_loss=0.1138, over 5725601.07 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3507, pruned_loss=0.1007, over 5709691.03 frames. ], batch size: 164, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:14:55,263 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.719e+02 1.038e+03 1.295e+03 1.843e+03 4.442e+03, threshold=2.591e+03, percent-clipped=6.0
+2023-03-05 07:15:23,391 INFO [train.py:968] (0/2) Epoch 10, batch 22800, giga_loss[loss=0.2519, simple_loss=0.3237, pruned_loss=0.09007, over 28927.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3537, pruned_loss=0.1033, over 5718471.58 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3687, pruned_loss=0.1143, over 5729027.32 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.351, pruned_loss=0.1008, over 5716720.82 frames. ], batch size: 106, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:15:29,348 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=432403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:15:31,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=432406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:15:40,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1192, 3.9058, 3.6848, 1.6337], device='cuda:0'), covar=tensor([0.0563, 0.0727, 0.0721, 0.2465], device='cuda:0'), in_proj_covar=tensor([0.1004, 0.0938, 0.0827, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 07:15:57,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=432435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:16:05,080 INFO [train.py:968] (0/2) Epoch 10, batch 22850, giga_loss[loss=0.2546, simple_loss=0.3225, pruned_loss=0.09333, over 28058.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3526, pruned_loss=0.1036, over 5721924.30 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3691, pruned_loss=0.1146, over 5733296.63 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3498, pruned_loss=0.1011, over 5716540.80 frames. ], batch size: 77, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:16:11,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4262, 1.6675, 1.7034, 1.2667], device='cuda:0'), covar=tensor([0.1603, 0.2092, 0.1317, 0.1490], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0689, 0.0844, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 07:16:16,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.795e+02 1.117e+03 1.438e+03 1.808e+03 4.774e+03, threshold=2.877e+03, percent-clipped=6.0
+2023-03-05 07:16:45,612 INFO [train.py:968] (0/2) Epoch 10, batch 22900, giga_loss[loss=0.2778, simple_loss=0.3458, pruned_loss=0.1049, over 28809.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.351, pruned_loss=0.1043, over 5724628.37 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3692, pruned_loss=0.1148, over 5734217.99 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3483, pruned_loss=0.102, over 5719386.01 frames. ], batch size: 199, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:16:57,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5601, 1.7127, 1.8585, 1.4038], device='cuda:0'), covar=tensor([0.1567, 0.1915, 0.1275, 0.1402], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0689, 0.0843, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 07:17:17,135 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 07:17:26,251 INFO [train.py:968] (0/2) Epoch 10, batch 22950, giga_loss[loss=0.277, simple_loss=0.3467, pruned_loss=0.1036, over 28706.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3506, pruned_loss=0.1055, over 5722547.22 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3695, pruned_loss=0.115, over 5735777.10 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3478, pruned_loss=0.1032, over 5716660.39 frames. ], batch size: 92, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:17:37,331 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-05 07:17:39,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.264e+02 1.205e+03 1.521e+03 2.105e+03 4.870e+03, threshold=3.043e+03, percent-clipped=11.0
+2023-03-05 07:18:07,865 INFO [train.py:968] (0/2) Epoch 10, batch 23000, giga_loss[loss=0.239, simple_loss=0.3085, pruned_loss=0.08476, over 28763.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3492, pruned_loss=0.1052, over 5727063.30 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3699, pruned_loss=0.1154, over 5739347.26 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3463, pruned_loss=0.1029, over 5719076.99 frames. ], batch size: 99, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:18:21,050 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=432609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:18:33,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4792, 1.6598, 1.3802, 1.3408], device='cuda:0'), covar=tensor([0.2006, 0.1630, 0.1308, 0.1604], device='cuda:0'), in_proj_covar=tensor([0.1673, 0.1551, 0.1540, 0.1651], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 07:18:47,073 INFO [train.py:968] (0/2) Epoch 10, batch 23050, giga_loss[loss=0.2672, simple_loss=0.3423, pruned_loss=0.09604, over 28879.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3464, pruned_loss=0.1041, over 5721057.95 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3696, pruned_loss=0.1153, over 5739636.88 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3439, pruned_loss=0.1021, over 5713993.88 frames. ], batch size: 199, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:19:00,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.575e+02 1.171e+03 1.575e+03 2.348e+03 5.721e+03, threshold=3.150e+03, percent-clipped=15.0
+2023-03-05 07:19:20,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=432686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:19:26,149 INFO [train.py:968] (0/2) Epoch 10, batch 23100, giga_loss[loss=0.333, simple_loss=0.386, pruned_loss=0.14, over 27855.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3429, pruned_loss=0.1022, over 5722916.32 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3701, pruned_loss=0.1158, over 5738641.49 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3399, pruned_loss=0.09996, over 5717954.54 frames. ], batch size: 412, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:19:32,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1890, 1.1462, 1.1451, 0.8545], device='cuda:0'), covar=tensor([0.0762, 0.0579, 0.1024, 0.1021], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0438, 0.0496, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 07:19:42,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=432714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:20:03,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8393, 1.8364, 1.4312, 1.5080], device='cuda:0'), covar=tensor([0.0750, 0.0663, 0.0989, 0.1088], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0441, 0.0500, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 07:20:04,577 INFO [train.py:968] (0/2) Epoch 10, batch 23150, giga_loss[loss=0.2438, simple_loss=0.317, pruned_loss=0.08535, over 29014.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3389, pruned_loss=0.09996, over 5716011.01 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3701, pruned_loss=0.1159, over 5732836.03 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3356, pruned_loss=0.09762, over 5716487.46 frames. ], batch size: 155, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:20:09,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3130, 1.6016, 1.6004, 1.1924], device='cuda:0'), covar=tensor([0.1452, 0.2008, 0.1205, 0.1395], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0691, 0.0844, 0.0758], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 07:20:17,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.881e+02 1.161e+03 1.496e+03 1.894e+03 6.822e+03, threshold=2.992e+03, percent-clipped=3.0
+2023-03-05 07:20:34,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2435, 1.8856, 1.4708, 0.4603], device='cuda:0'), covar=tensor([0.3358, 0.1683, 0.2556, 0.4079], device='cuda:0'), in_proj_covar=tensor([0.1534, 0.1435, 0.1471, 0.1257], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 07:20:43,976 INFO [train.py:968] (0/2) Epoch 10, batch 23200, giga_loss[loss=0.2817, simple_loss=0.3579, pruned_loss=0.1027, over 29059.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.338, pruned_loss=0.09891, over 5720695.48 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3702, pruned_loss=0.1161, over 5735275.01 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3344, pruned_loss=0.09638, over 5718567.58 frames. ], batch size: 113, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:20:50,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3986, 1.6701, 1.7177, 1.2814], device='cuda:0'), covar=tensor([0.1493, 0.1971, 0.1196, 0.1411], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0691, 0.0844, 0.0757], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 07:21:13,747 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=432829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:15,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=432832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:18,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8461, 1.8440, 1.5409, 1.6056], device='cuda:0'), covar=tensor([0.0782, 0.0722, 0.0949, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0440, 0.0497, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 07:21:25,895 INFO [train.py:968] (0/2) Epoch 10, batch 23250, giga_loss[loss=0.2476, simple_loss=0.3245, pruned_loss=0.08542, over 28425.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3418, pruned_loss=0.1009, over 5707900.54 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3703, pruned_loss=0.1163, over 5729835.52 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3382, pruned_loss=0.09836, over 5710997.88 frames. ], batch size: 60, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:21:35,736 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=432856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:37,369 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=432857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:39,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=432860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:40,240 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=432861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:40,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.988e+02 1.290e+03 1.677e+03 2.317e+03 6.631e+03, threshold=3.354e+03, percent-clipped=13.0
+2023-03-05 07:21:51,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=432874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:22:04,468 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=432889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:22:08,160 INFO [train.py:968] (0/2) Epoch 10, batch 23300, giga_loss[loss=0.2655, simple_loss=0.3374, pruned_loss=0.09682, over 28619.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3452, pruned_loss=0.1025, over 5712164.19 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3701, pruned_loss=0.1164, over 5735210.98 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3417, pruned_loss=0.09993, over 5709422.92 frames. ], batch size: 85, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:22:47,614 INFO [train.py:968] (0/2) Epoch 10, batch 23350, giga_loss[loss=0.3036, simple_loss=0.3748, pruned_loss=0.1162, over 28190.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3483, pruned_loss=0.1035, over 5715711.37 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3697, pruned_loss=0.1163, over 5740914.55 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.345, pruned_loss=0.101, over 5707660.01 frames. ], batch size: 368, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:23:01,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.924e+02 1.169e+03 1.414e+03 1.875e+03 5.763e+03, threshold=2.827e+03, percent-clipped=5.0
+2023-03-05 07:23:19,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=432984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:23:25,936 INFO [train.py:968] (0/2) Epoch 10, batch 23400, giga_loss[loss=0.2665, simple_loss=0.345, pruned_loss=0.09396, over 28932.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3507, pruned_loss=0.1042, over 5722248.23 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3695, pruned_loss=0.1163, over 5742581.55 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3479, pruned_loss=0.102, over 5713800.43 frames. ], batch size: 145, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:24:01,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7838, 1.1761, 2.8301, 2.8069], device='cuda:0'), covar=tensor([0.1562, 0.2308, 0.0538, 0.0971], device='cuda:0'), in_proj_covar=tensor([0.0646, 0.0576, 0.0841, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 07:24:10,652 INFO [train.py:968] (0/2) Epoch 10, batch 23450, giga_loss[loss=0.3211, simple_loss=0.3747, pruned_loss=0.1338, over 28901.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3529, pruned_loss=0.1055, over 5722364.27 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3695, pruned_loss=0.1164, over 5741339.26 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3504, pruned_loss=0.1035, over 5716861.20 frames. ], batch size: 99, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:24:25,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.283e+02 1.239e+03 1.513e+03 2.191e+03 1.200e+04, threshold=3.027e+03, percent-clipped=16.0
+2023-03-05 07:24:28,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4564, 3.6587, 1.6170, 1.5612], device='cuda:0'), covar=tensor([0.0892, 0.0350, 0.0839, 0.1247], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0505, 0.0332, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 07:24:57,914 INFO [train.py:968] (0/2) Epoch 10, batch 23500, giga_loss[loss=0.3181, simple_loss=0.3812, pruned_loss=0.1275, over 28950.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3599, pruned_loss=0.1119, over 5700360.97 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.37, pruned_loss=0.1169, over 5735196.15 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3571, pruned_loss=0.1096, over 5701551.60 frames. ], batch size: 186, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:25:32,518 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=433127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:25:37,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=433130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:25:43,622 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.22 vs. limit=5.0
+2023-03-05 07:25:51,360 INFO [train.py:968] (0/2) Epoch 10, batch 23550, giga_loss[loss=0.3346, simple_loss=0.3976, pruned_loss=0.1358, over 28852.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3659, pruned_loss=0.1165, over 5690181.70 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.37, pruned_loss=0.117, over 5736086.17 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3636, pruned_loss=0.1147, over 5690179.79 frames. ], batch size: 112, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:26:07,923 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=433159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:26:10,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.551e+02 1.472e+03 1.918e+03 2.714e+03 7.612e+03, threshold=3.837e+03, percent-clipped=16.0
+2023-03-05 07:26:31,678 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-05 07:26:41,578 INFO [train.py:968] (0/2) Epoch 10, batch 23600, giga_loss[loss=0.3392, simple_loss=0.4012, pruned_loss=0.1386, over 28253.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3739, pruned_loss=0.1228, over 5688405.98 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3698, pruned_loss=0.117, over 5739108.78 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3723, pruned_loss=0.1214, over 5684719.55 frames. ], batch size: 368, lr: 3.23e-03, grad_scale: 8.0
+2023-03-05 07:26:46,835 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-05 07:27:20,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=433231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:27:28,787 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-05 07:27:33,477 INFO [train.py:968] (0/2) Epoch 10, batch 23650, giga_loss[loss=0.3976, simple_loss=0.4351, pruned_loss=0.18, over 28210.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3802, pruned_loss=0.1283, over 5686282.56 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3701, pruned_loss=0.1172, over 5740627.88 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3787, pruned_loss=0.1271, over 5681764.23 frames. ], batch size: 368, lr: 3.23e-03, grad_scale: 8.0
+2023-03-05 07:27:40,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=433249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:27:53,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.531e+02 1.596e+03 1.939e+03 2.594e+03 4.748e+03, threshold=3.877e+03, percent-clipped=6.0
+2023-03-05 07:28:22,614 INFO [train.py:968] (0/2) Epoch 10, batch 23700, giga_loss[loss=0.3229, simple_loss=0.392, pruned_loss=0.1269, over 28939.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3869, pruned_loss=0.1341, over 5681392.05 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3703, pruned_loss=0.1173, over 5741216.78 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.386, pruned_loss=0.1334, over 5675871.76 frames. ], batch size: 213, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:29:09,007 INFO [train.py:968] (0/2) Epoch 10, batch 23750, giga_loss[loss=0.3735, simple_loss=0.4141, pruned_loss=0.1664, over 27569.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3898, pruned_loss=0.1371, over 5679834.51 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3706, pruned_loss=0.1177, over 5741851.51 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3892, pruned_loss=0.1367, over 5673368.03 frames. ], batch size: 472, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:29:12,165 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3830, 1.6800, 1.3484, 1.4221], device='cuda:0'), covar=tensor([0.2033, 0.1915, 0.2081, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.1260, 0.0938, 0.1121, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 07:29:29,663 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.078e+03 1.673e+03 2.184e+03 3.479e+03 9.165e+03, threshold=4.368e+03, percent-clipped=19.0
+2023-03-05 07:29:39,945 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=433374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:29:41,885 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=433377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:29:57,293 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=433392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:29:59,255 INFO [train.py:968] (0/2) Epoch 10, batch 23800, giga_loss[loss=0.3424, simple_loss=0.3941, pruned_loss=0.1454, over 28839.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3931, pruned_loss=0.1412, over 5668801.84 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3712, pruned_loss=0.1182, over 5744969.89 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3925, pruned_loss=0.1407, over 5659467.08 frames. ], batch size: 174, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:30:01,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=433395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:30:09,113 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=433406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:30:25,576 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=433424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:30:45,673 INFO [train.py:968] (0/2) Epoch 10, batch 23850, giga_loss[loss=0.3917, simple_loss=0.4308, pruned_loss=0.1763, over 27854.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3967, pruned_loss=0.1454, over 5657525.95 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3714, pruned_loss=0.1186, over 5740817.77 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.397, pruned_loss=0.1456, over 5650218.22 frames. ], batch size: 412, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:30:49,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9911, 1.1267, 1.3287, 0.9368], device='cuda:0'), covar=tensor([0.1338, 0.1311, 0.1825, 0.1480], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0733, 0.0669, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 07:31:05,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.667e+02 1.721e+03 2.226e+03 2.949e+03 7.221e+03, threshold=4.452e+03, percent-clipped=5.0
+2023-03-05 07:31:38,644 INFO [train.py:968] (0/2) Epoch 10, batch 23900, giga_loss[loss=0.4998, simple_loss=0.4906, pruned_loss=0.2545, over 26575.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4004, pruned_loss=0.1493, over 5642695.36 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3714, pruned_loss=0.1188, over 5735371.89 frames. ], giga_tot_loss[loss=0.3505, simple_loss=0.4011, pruned_loss=0.1499, over 5639205.88 frames. ], batch size: 555, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:32:35,288 INFO [train.py:968] (0/2) Epoch 10, batch 23950, giga_loss[loss=0.3267, simple_loss=0.3842, pruned_loss=0.1346, over 28654.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.402, pruned_loss=0.1501, over 5650628.09 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3712, pruned_loss=0.1189, over 5739629.66 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4035, pruned_loss=0.1513, over 5642255.95 frames. ], batch size: 262, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:32:56,211 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.816e+02 1.760e+03 2.960e+03 4.259e+03 1.231e+04, threshold=5.920e+03, percent-clipped=19.0
+2023-03-05 07:33:00,010 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=433567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:33:27,150 INFO [train.py:968] (0/2) Epoch 10, batch 24000, giga_loss[loss=0.4048, simple_loss=0.4406, pruned_loss=0.1845, over 27973.00 frames. ], tot_loss[loss=0.3513, simple_loss=0.4015, pruned_loss=0.1506, over 5640869.35 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3712, pruned_loss=0.1189, over 5744603.93 frames. ], giga_tot_loss[loss=0.3542, simple_loss=0.4036, pruned_loss=0.1524, over 5627135.00 frames. ], batch size: 412, lr: 3.23e-03, grad_scale: 8.0
+2023-03-05 07:33:27,155 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 07:33:35,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2664, 1.8531, 1.6264, 1.1665], device='cuda:0'), covar=tensor([0.1642, 0.2431, 0.1458, 0.1672], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0693, 0.0842, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 07:33:35,817 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.217, simple_loss=0.3234, pruned_loss=0.05525, over 944034.00 frames. 
+2023-03-05 07:33:35,817 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 07:34:23,987 INFO [train.py:968] (0/2) Epoch 10, batch 24050, giga_loss[loss=0.3642, simple_loss=0.4117, pruned_loss=0.1583, over 27893.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.4009, pruned_loss=0.1507, over 5652346.35 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3713, pruned_loss=0.119, over 5746697.75 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4029, pruned_loss=0.1526, over 5637868.58 frames. ], batch size: 412, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:34:44,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.660e+03 2.096e+03 2.912e+03 9.066e+03, threshold=4.192e+03, percent-clipped=3.0
+2023-03-05 07:35:10,329 INFO [train.py:968] (0/2) Epoch 10, batch 24100, giga_loss[loss=0.3435, simple_loss=0.4024, pruned_loss=0.1423, over 28786.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4001, pruned_loss=0.1489, over 5655488.98 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3711, pruned_loss=0.119, over 5745517.19 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4023, pruned_loss=0.1508, over 5643926.82 frames. ], batch size: 284, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:36:04,141 INFO [train.py:968] (0/2) Epoch 10, batch 24150, giga_loss[loss=0.3742, simple_loss=0.421, pruned_loss=0.1637, over 28787.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.3995, pruned_loss=0.1479, over 5643542.17 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3713, pruned_loss=0.1193, over 5738837.05 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4017, pruned_loss=0.1497, over 5637874.82 frames. ], batch size: 99, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:36:24,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.506e+03 1.852e+03 2.421e+03 5.147e+03, threshold=3.704e+03, percent-clipped=4.0
+2023-03-05 07:36:50,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6134, 1.6789, 1.6413, 1.4877], device='cuda:0'), covar=tensor([0.1314, 0.1914, 0.1786, 0.1752], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0739, 0.0674, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 07:36:53,667 INFO [train.py:968] (0/2) Epoch 10, batch 24200, libri_loss[loss=0.3777, simple_loss=0.4226, pruned_loss=0.1664, over 29373.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.4011, pruned_loss=0.1488, over 5635060.72 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3715, pruned_loss=0.1197, over 5742596.34 frames. ], giga_tot_loss[loss=0.3525, simple_loss=0.4034, pruned_loss=0.1508, over 5624342.43 frames. ], batch size: 92, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:37:00,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-05 07:37:08,840 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=433809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:37:45,831 INFO [train.py:968] (0/2) Epoch 10, batch 24250, giga_loss[loss=0.3752, simple_loss=0.415, pruned_loss=0.1677, over 27583.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3968, pruned_loss=0.1446, over 5623728.91 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3721, pruned_loss=0.1201, over 5733550.98 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.3988, pruned_loss=0.1463, over 5619923.94 frames. ], batch size: 472, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:37:47,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2647, 3.1075, 1.3899, 1.4249], device='cuda:0'), covar=tensor([0.0941, 0.0328, 0.0868, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0508, 0.0334, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 07:38:03,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.922e+02 1.777e+03 2.160e+03 3.515e+03 9.549e+03, threshold=4.321e+03, percent-clipped=22.0
+2023-03-05 07:38:21,525 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=433881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:38:32,860 INFO [train.py:968] (0/2) Epoch 10, batch 24300, libri_loss[loss=0.2817, simple_loss=0.3414, pruned_loss=0.111, over 29576.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.393, pruned_loss=0.1401, over 5637191.51 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3718, pruned_loss=0.1202, over 5738345.75 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3954, pruned_loss=0.1419, over 5627425.91 frames. ], batch size: 75, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:39:20,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=433942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:39:22,013 INFO [train.py:968] (0/2) Epoch 10, batch 24350, giga_loss[loss=0.31, simple_loss=0.3739, pruned_loss=0.123, over 28021.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3893, pruned_loss=0.1362, over 5657864.93 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3718, pruned_loss=0.1203, over 5741004.76 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3914, pruned_loss=0.1378, over 5646492.60 frames. ], batch size: 412, lr: 3.23e-03, grad_scale: 1.0
+2023-03-05 07:39:40,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.420e+02 1.600e+03 2.548e+03 4.309e+03 2.273e+04, threshold=5.096e+03, percent-clipped=24.0
+2023-03-05 07:40:08,731 INFO [train.py:968] (0/2) Epoch 10, batch 24400, giga_loss[loss=0.2941, simple_loss=0.3455, pruned_loss=0.1214, over 23679.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3867, pruned_loss=0.1337, over 5670131.76 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.1199, over 5744316.96 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3896, pruned_loss=0.1358, over 5655439.07 frames. ], batch size: 705, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:40:16,466 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-434000.pt
+2023-03-05 07:40:59,318 INFO [train.py:968] (0/2) Epoch 10, batch 24450, giga_loss[loss=0.2928, simple_loss=0.3612, pruned_loss=0.1122, over 28942.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3844, pruned_loss=0.1325, over 5667266.36 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3707, pruned_loss=0.1197, over 5747019.08 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3874, pruned_loss=0.1347, over 5651520.41 frames. ], batch size: 227, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:41:22,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.622e+02 1.521e+03 1.895e+03 2.458e+03 5.498e+03, threshold=3.789e+03, percent-clipped=1.0
+2023-03-05 07:41:41,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:41:47,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:41:52,469 INFO [train.py:968] (0/2) Epoch 10, batch 24500, giga_loss[loss=0.2867, simple_loss=0.3653, pruned_loss=0.1041, over 28857.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3851, pruned_loss=0.133, over 5657792.54 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3705, pruned_loss=0.1196, over 5739540.69 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3878, pruned_loss=0.1349, over 5651880.04 frames. ], batch size: 174, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:42:13,919 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4087, 1.6281, 1.7075, 1.3076], device='cuda:0'), covar=tensor([0.1515, 0.2183, 0.1242, 0.1458], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0702, 0.0846, 0.0758], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 07:42:17,994 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:42:38,460 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:42:41,854 INFO [train.py:968] (0/2) Epoch 10, batch 24550, giga_loss[loss=0.359, simple_loss=0.4026, pruned_loss=0.1577, over 28922.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3842, pruned_loss=0.1319, over 5673737.64 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3703, pruned_loss=0.1197, over 5742911.51 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.387, pruned_loss=0.1338, over 5663888.78 frames. ], batch size: 213, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:42:46,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1879, 1.4242, 3.4723, 3.1979], device='cuda:0'), covar=tensor([0.1750, 0.2565, 0.0716, 0.0948], device='cuda:0'), in_proj_covar=tensor([0.0652, 0.0581, 0.0851, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 07:43:06,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.470e+03 2.096e+03 3.057e+03 1.325e+04, threshold=4.191e+03, percent-clipped=19.0
+2023-03-05 07:43:14,600 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 07:43:20,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9015, 1.1681, 5.2439, 3.7649], device='cuda:0'), covar=tensor([0.1474, 0.2679, 0.0400, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0653, 0.0583, 0.0854, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 07:43:25,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:43:34,442 INFO [train.py:968] (0/2) Epoch 10, batch 24600, libri_loss[loss=0.2587, simple_loss=0.3276, pruned_loss=0.09496, over 29571.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3841, pruned_loss=0.13, over 5680260.40 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3703, pruned_loss=0.1197, over 5747144.53 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3867, pruned_loss=0.1318, over 5666838.86 frames. ], batch size: 77, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:43:43,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1570, 1.5254, 1.3717, 1.2692], device='cuda:0'), covar=tensor([0.0828, 0.0317, 0.0288, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:0')
+2023-03-05 07:44:26,009 INFO [train.py:968] (0/2) Epoch 10, batch 24650, giga_loss[loss=0.3438, simple_loss=0.4094, pruned_loss=0.1391, over 28830.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3852, pruned_loss=0.1291, over 5666651.85 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3706, pruned_loss=0.12, over 5738940.36 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3872, pruned_loss=0.1303, over 5662751.98 frames. ], batch size: 119, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:44:30,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3106, 1.5005, 1.5087, 1.3401], device='cuda:0'), covar=tensor([0.1323, 0.1394, 0.1814, 0.1479], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0730, 0.0664, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 07:44:39,261 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:44:50,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.697e+02 1.566e+03 1.953e+03 2.661e+03 5.499e+03, threshold=3.906e+03, percent-clipped=1.0
+2023-03-05 07:45:14,198 INFO [train.py:968] (0/2) Epoch 10, batch 24700, libri_loss[loss=0.3167, simple_loss=0.3782, pruned_loss=0.1277, over 29469.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.386, pruned_loss=0.1298, over 5664715.01 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3708, pruned_loss=0.1202, over 5737446.08 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3879, pruned_loss=0.1309, over 5660572.05 frames. ], batch size: 85, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:45:48,963 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:45:49,397 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:45:51,317 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:45:57,511 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:46:04,506 INFO [train.py:968] (0/2) Epoch 10, batch 24750, giga_loss[loss=0.3561, simple_loss=0.4077, pruned_loss=0.1522, over 28308.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3865, pruned_loss=0.131, over 5661455.13 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3707, pruned_loss=0.1202, over 5739155.10 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3884, pruned_loss=0.1321, over 5655514.74 frames. ], batch size: 368, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:46:19,788 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:46:20,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:46:24,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.708e+03 2.372e+03 3.409e+03 9.190e+03, threshold=4.743e+03, percent-clipped=18.0
+2023-03-05 07:46:38,434 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3220, 1.7990, 1.4066, 1.2834], device='cuda:0'), covar=tensor([0.2414, 0.2275, 0.2452, 0.2140], device='cuda:0'), in_proj_covar=tensor([0.1270, 0.0947, 0.1128, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 07:46:46,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8741, 3.6845, 3.5148, 1.6369], device='cuda:0'), covar=tensor([0.0648, 0.0787, 0.0783, 0.2078], device='cuda:0'), in_proj_covar=tensor([0.1030, 0.0967, 0.0848, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 07:46:53,716 INFO [train.py:968] (0/2) Epoch 10, batch 24800, giga_loss[loss=0.3117, simple_loss=0.3716, pruned_loss=0.1259, over 29003.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3835, pruned_loss=0.1299, over 5657800.77 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3705, pruned_loss=0.1202, over 5745215.07 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3856, pruned_loss=0.1311, over 5644483.32 frames. ], batch size: 136, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:46:59,463 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:47:02,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:47:27,265 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:47:37,393 INFO [train.py:968] (0/2) Epoch 10, batch 24850, giga_loss[loss=0.3186, simple_loss=0.386, pruned_loss=0.1256, over 28919.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3819, pruned_loss=0.1296, over 5670640.79 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3706, pruned_loss=0.1203, over 5746684.76 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3838, pruned_loss=0.1308, over 5656760.91 frames. ], batch size: 199, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:47:41,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6940, 1.6909, 1.2022, 1.3073], device='cuda:0'), covar=tensor([0.0676, 0.0596, 0.0981, 0.1064], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0441, 0.0497, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 07:47:51,232 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:47:56,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.402e+02 1.678e+03 1.994e+03 2.709e+03 6.402e+03, threshold=3.988e+03, percent-clipped=2.0
+2023-03-05 07:48:21,399 INFO [train.py:968] (0/2) Epoch 10, batch 24900, giga_loss[loss=0.3176, simple_loss=0.3761, pruned_loss=0.1296, over 28005.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3815, pruned_loss=0.1299, over 5671618.76 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3707, pruned_loss=0.1204, over 5749006.90 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3831, pruned_loss=0.1309, over 5657538.57 frames. ], batch size: 412, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:48:41,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:49:04,082 INFO [train.py:968] (0/2) Epoch 10, batch 24950, giga_loss[loss=0.2922, simple_loss=0.3715, pruned_loss=0.1064, over 28954.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3804, pruned_loss=0.1278, over 5685328.61 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3708, pruned_loss=0.1206, over 5749301.00 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3821, pruned_loss=0.1288, over 5670898.69 frames. ], batch size: 145, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:49:27,728 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.046e+02 1.306e+03 1.807e+03 2.870e+03 1.131e+04, threshold=3.613e+03, percent-clipped=13.0
+2023-03-05 07:49:54,708 INFO [train.py:968] (0/2) Epoch 10, batch 25000, giga_loss[loss=0.3055, simple_loss=0.3835, pruned_loss=0.1137, over 29022.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3805, pruned_loss=0.1277, over 5667026.24 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3709, pruned_loss=0.1206, over 5744330.58 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3818, pruned_loss=0.1284, over 5659416.32 frames. ], batch size: 136, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:50:21,462 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 07:50:41,715 INFO [train.py:968] (0/2) Epoch 10, batch 25050, giga_loss[loss=0.2909, simple_loss=0.3645, pruned_loss=0.1087, over 28740.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3805, pruned_loss=0.1271, over 5679492.92 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3708, pruned_loss=0.1207, over 5747515.83 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3819, pruned_loss=0.1278, over 5668779.45 frames. ], batch size: 92, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:50:57,866 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:51:00,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:51:04,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.481e+02 1.371e+03 1.928e+03 2.585e+03 8.263e+03, threshold=3.855e+03, percent-clipped=8.0
+2023-03-05 07:51:24,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:51:30,392 INFO [train.py:968] (0/2) Epoch 10, batch 25100, giga_loss[loss=0.3162, simple_loss=0.3696, pruned_loss=0.1315, over 28649.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3808, pruned_loss=0.128, over 5688339.48 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3707, pruned_loss=0.1207, over 5750069.75 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3822, pruned_loss=0.1286, over 5675900.89 frames. ], batch size: 99, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:51:39,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:51:40,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2958, 1.2991, 1.2058, 1.4926], device='cuda:0'), covar=tensor([0.0701, 0.0355, 0.0308, 0.0725], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:0')
+2023-03-05 07:51:49,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:52:09,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:52:18,327 INFO [train.py:968] (0/2) Epoch 10, batch 25150, giga_loss[loss=0.2754, simple_loss=0.3482, pruned_loss=0.1013, over 29037.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3808, pruned_loss=0.1291, over 5662824.08 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3714, pruned_loss=0.1213, over 5725258.53 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3815, pruned_loss=0.1293, over 5673666.80 frames. ], batch size: 155, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:52:38,054 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.519e+02 1.568e+03 2.118e+03 2.735e+03 1.203e+04, threshold=4.235e+03, percent-clipped=12.0
+2023-03-05 07:53:01,505 INFO [train.py:968] (0/2) Epoch 10, batch 25200, giga_loss[loss=0.2944, simple_loss=0.3625, pruned_loss=0.1131, over 28902.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3796, pruned_loss=0.1285, over 5667738.89 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3718, pruned_loss=0.1216, over 5717549.54 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3801, pruned_loss=0.1286, over 5682252.44 frames. ], batch size: 186, lr: 3.23e-03, grad_scale: 8.0
+2023-03-05 07:53:14,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:53:42,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:53:50,155 INFO [train.py:968] (0/2) Epoch 10, batch 25250, giga_loss[loss=0.2763, simple_loss=0.3512, pruned_loss=0.1007, over 28979.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3781, pruned_loss=0.1278, over 5680558.26 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3721, pruned_loss=0.122, over 5719199.71 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3782, pruned_loss=0.1276, over 5690099.48 frames. ], batch size: 213, lr: 3.23e-03, grad_scale: 8.0
+2023-03-05 07:53:53,781 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:53:57,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:05,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:08,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:14,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.979e+02 1.553e+03 2.221e+03 3.066e+03 8.674e+03, threshold=4.441e+03, percent-clipped=5.0
+2023-03-05 07:54:23,750 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:24,372 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:24,417 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:27,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:35,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:40,605 INFO [train.py:968] (0/2) Epoch 10, batch 25300, giga_loss[loss=0.2787, simple_loss=0.3509, pruned_loss=0.1033, over 29018.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3756, pruned_loss=0.1269, over 5676557.20 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3721, pruned_loss=0.1219, over 5720177.71 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3758, pruned_loss=0.1268, over 5682936.80 frames. ], batch size: 128, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:54:55,186 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:55:26,395 INFO [train.py:968] (0/2) Epoch 10, batch 25350, giga_loss[loss=0.3437, simple_loss=0.3911, pruned_loss=0.1481, over 28839.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3747, pruned_loss=0.1266, over 5683738.36 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3722, pruned_loss=0.1221, over 5727517.13 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3748, pruned_loss=0.1265, over 5680492.28 frames. ], batch size: 112, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:55:48,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.400e+02 1.734e+03 2.255e+03 3.279e+03 7.558e+03, threshold=4.510e+03, percent-clipped=11.0
+2023-03-05 07:55:55,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:55:59,932 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:56:10,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5006, 1.5892, 1.5819, 1.4062], device='cuda:0'), covar=tensor([0.1816, 0.1563, 0.1042, 0.1439], device='cuda:0'), in_proj_covar=tensor([0.1651, 0.1546, 0.1537, 0.1636], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 07:56:15,120 INFO [train.py:968] (0/2) Epoch 10, batch 25400, giga_loss[loss=0.2998, simple_loss=0.3661, pruned_loss=0.1167, over 28867.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3767, pruned_loss=0.1275, over 5686598.50 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3731, pruned_loss=0.1227, over 5732086.95 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.1271, over 5678961.02 frames. ], batch size: 99, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:56:19,375 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-05 07:56:27,759 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=435009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:56:43,109 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5244, 1.1440, 4.5618, 3.5399], device='cuda:0'), covar=tensor([0.1568, 0.2722, 0.0375, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0650, 0.0581, 0.0853, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 07:56:58,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9301, 1.0872, 0.8724, 0.2341], device='cuda:0'), covar=tensor([0.1995, 0.1741, 0.1856, 0.3534], device='cuda:0'), in_proj_covar=tensor([0.1538, 0.1464, 0.1478, 0.1265], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 07:56:59,081 INFO [train.py:968] (0/2) Epoch 10, batch 25450, giga_loss[loss=0.3755, simple_loss=0.395, pruned_loss=0.178, over 23679.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.377, pruned_loss=0.1268, over 5687033.74 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.373, pruned_loss=0.1226, over 5731328.33 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3768, pruned_loss=0.1266, over 5680794.00 frames. ], batch size: 705, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:57:21,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.256e+02 1.463e+03 1.783e+03 2.247e+03 7.347e+03, threshold=3.566e+03, percent-clipped=6.0
+2023-03-05 07:57:43,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2444, 4.0781, 3.8547, 1.9231], device='cuda:0'), covar=tensor([0.0553, 0.0679, 0.0744, 0.1874], device='cuda:0'), in_proj_covar=tensor([0.1032, 0.0974, 0.0848, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 07:57:45,191 INFO [train.py:968] (0/2) Epoch 10, batch 25500, giga_loss[loss=0.3549, simple_loss=0.3981, pruned_loss=0.1558, over 28831.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3776, pruned_loss=0.1264, over 5693643.22 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3729, pruned_loss=0.1226, over 5734280.86 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3776, pruned_loss=0.1264, over 5685257.38 frames. ], batch size: 284, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:58:32,041 INFO [train.py:968] (0/2) Epoch 10, batch 25550, giga_loss[loss=0.2979, simple_loss=0.3638, pruned_loss=0.116, over 28649.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.377, pruned_loss=0.1262, over 5683844.37 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3722, pruned_loss=0.1222, over 5727718.16 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3777, pruned_loss=0.1266, over 5681346.08 frames. ], batch size: 242, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:58:36,028 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-05 07:58:52,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9123, 2.2468, 2.2291, 1.8020], device='cuda:0'), covar=tensor([0.1678, 0.1909, 0.1291, 0.1428], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0698, 0.0844, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 07:58:55,125 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.878e+02 1.677e+03 2.122e+03 3.313e+03 9.544e+03, threshold=4.243e+03, percent-clipped=22.0
+2023-03-05 07:59:05,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=435180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:59:17,865 INFO [train.py:968] (0/2) Epoch 10, batch 25600, giga_loss[loss=0.2903, simple_loss=0.3541, pruned_loss=0.1132, over 28769.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3809, pruned_loss=0.1297, over 5687100.92 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3722, pruned_loss=0.1223, over 5731309.47 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3816, pruned_loss=0.1301, over 5680359.20 frames. ], batch size: 99, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:00:09,432 INFO [train.py:968] (0/2) Epoch 10, batch 25650, giga_loss[loss=0.3053, simple_loss=0.3707, pruned_loss=0.12, over 28814.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3823, pruned_loss=0.1326, over 5681957.36 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3717, pruned_loss=0.1221, over 5732966.25 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3835, pruned_loss=0.1332, over 5674625.60 frames. ], batch size: 119, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:00:15,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=435252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:00:34,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.178e+03 1.907e+03 2.468e+03 3.218e+03 6.695e+03, threshold=4.936e+03, percent-clipped=11.0
+2023-03-05 08:01:00,619 INFO [train.py:968] (0/2) Epoch 10, batch 25700, giga_loss[loss=0.3065, simple_loss=0.3686, pruned_loss=0.1221, over 28865.00 frames. ], tot_loss[loss=0.326, simple_loss=0.383, pruned_loss=0.1345, over 5672490.69 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3722, pruned_loss=0.1225, over 5725602.85 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3836, pruned_loss=0.1347, over 5672881.96 frames. ], batch size: 199, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:01:33,342 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=435323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:01:35,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=435326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:01:49,036 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=435342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:01:50,171 INFO [train.py:968] (0/2) Epoch 10, batch 25750, giga_loss[loss=0.2916, simple_loss=0.3517, pruned_loss=0.1157, over 28755.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3836, pruned_loss=0.1349, over 5680278.96 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3722, pruned_loss=0.1224, over 5727865.26 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3842, pruned_loss=0.1354, over 5677957.65 frames. ], batch size: 66, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:01:58,940 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=435355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:02:05,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3844, 1.6452, 1.4614, 1.5145], device='cuda:0'), covar=tensor([0.0758, 0.0301, 0.0304, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:0')
+2023-03-05 08:02:10,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.696e+03 2.193e+03 2.791e+03 8.876e+03, threshold=4.386e+03, percent-clipped=3.0
+2023-03-05 08:02:35,759 INFO [train.py:968] (0/2) Epoch 10, batch 25800, giga_loss[loss=0.2762, simple_loss=0.3402, pruned_loss=0.1061, over 28556.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3827, pruned_loss=0.1349, over 5671460.05 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3725, pruned_loss=0.1228, over 5730308.79 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3831, pruned_loss=0.1352, over 5666297.75 frames. ], batch size: 85, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:02:36,909 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=435395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:02:39,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=435398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:02:44,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3998, 1.5507, 1.4402, 1.2000], device='cuda:0'), covar=tensor([0.1646, 0.1530, 0.1027, 0.1487], device='cuda:0'), in_proj_covar=tensor([0.1650, 0.1560, 0.1540, 0.1636], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 08:02:51,773 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-05 08:03:06,317 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=435427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:03:23,085 INFO [train.py:968] (0/2) Epoch 10, batch 25850, giga_loss[loss=0.3726, simple_loss=0.4104, pruned_loss=0.1674, over 23808.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.382, pruned_loss=0.1326, over 5676267.26 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3727, pruned_loss=0.1229, over 5731937.90 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3823, pruned_loss=0.1329, over 5670164.27 frames. ], batch size: 705, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:03:45,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.691e+02 1.563e+03 2.145e+03 2.780e+03 9.977e+03, threshold=4.289e+03, percent-clipped=8.0
+2023-03-05 08:04:08,264 INFO [train.py:968] (0/2) Epoch 10, batch 25900, giga_loss[loss=0.351, simple_loss=0.4061, pruned_loss=0.148, over 28652.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3798, pruned_loss=0.1304, over 5665654.94 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3735, pruned_loss=0.1236, over 5727736.85 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3795, pruned_loss=0.1302, over 5662808.69 frames. ], batch size: 85, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:04:38,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1868, 0.9091, 0.9310, 1.3073], device='cuda:0'), covar=tensor([0.0664, 0.0461, 0.0326, 0.0669], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:0')
+2023-03-05 08:04:55,370 INFO [train.py:968] (0/2) Epoch 10, batch 25950, giga_loss[loss=0.3877, simple_loss=0.4201, pruned_loss=0.1777, over 28597.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.378, pruned_loss=0.1296, over 5668555.82 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3734, pruned_loss=0.1236, over 5729391.99 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3779, pruned_loss=0.1295, over 5664170.60 frames. ], batch size: 307, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:05:16,080 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=435567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:05:16,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.571e+02 1.472e+03 1.982e+03 2.612e+03 6.095e+03, threshold=3.965e+03, percent-clipped=6.0
+2023-03-05 08:05:38,861 INFO [train.py:968] (0/2) Epoch 10, batch 26000, libri_loss[loss=0.2881, simple_loss=0.3407, pruned_loss=0.1178, over 29650.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3765, pruned_loss=0.1292, over 5665218.68 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3737, pruned_loss=0.1239, over 5724809.64 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3763, pruned_loss=0.129, over 5663039.74 frames. ], batch size: 69, lr: 3.22e-03, grad_scale: 8.0
+2023-03-05 08:06:28,610 INFO [train.py:968] (0/2) Epoch 10, batch 26050, libri_loss[loss=0.2944, simple_loss=0.3485, pruned_loss=0.1201, over 29386.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3777, pruned_loss=0.1305, over 5650681.75 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3743, pruned_loss=0.1243, over 5718784.11 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3771, pruned_loss=0.1302, over 5652456.59 frames. ], batch size: 67, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:06:50,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.542e+02 1.659e+03 2.308e+03 2.952e+03 7.387e+03, threshold=4.616e+03, percent-clipped=13.0
+2023-03-05 08:07:12,230 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 08:07:15,130 INFO [train.py:968] (0/2) Epoch 10, batch 26100, giga_loss[loss=0.2874, simple_loss=0.365, pruned_loss=0.1049, over 28985.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3804, pruned_loss=0.1313, over 5656620.63 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3741, pruned_loss=0.1241, over 5720614.47 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3801, pruned_loss=0.1313, over 5655961.55 frames. ], batch size: 136, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:07:33,156 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 08:07:35,598 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=435717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:08:01,492 INFO [train.py:968] (0/2) Epoch 10, batch 26150, giga_loss[loss=0.3272, simple_loss=0.3929, pruned_loss=0.1307, over 28812.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3849, pruned_loss=0.1317, over 5661736.36 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3743, pruned_loss=0.1242, over 5721980.84 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3847, pruned_loss=0.1318, over 5658481.44 frames. ], batch size: 186, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:08:19,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5202, 4.3356, 4.0810, 1.8614], device='cuda:0'), covar=tensor([0.0525, 0.0745, 0.0784, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.1033, 0.0975, 0.0854, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 08:08:20,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8352, 1.7823, 1.3472, 1.3800], device='cuda:0'), covar=tensor([0.0774, 0.0631, 0.0916, 0.1100], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0440, 0.0497, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 08:08:28,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.422e+02 1.352e+03 1.694e+03 2.705e+03 1.366e+04, threshold=3.387e+03, percent-clipped=5.0
+2023-03-05 08:08:37,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8014, 4.6404, 4.3506, 1.8537], device='cuda:0'), covar=tensor([0.0486, 0.0655, 0.0782, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.1033, 0.0976, 0.0855, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 08:08:52,443 INFO [train.py:968] (0/2) Epoch 10, batch 26200, giga_loss[loss=0.2943, simple_loss=0.3627, pruned_loss=0.1129, over 28913.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3853, pruned_loss=0.1307, over 5651969.38 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3743, pruned_loss=0.1243, over 5713536.59 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3852, pruned_loss=0.1308, over 5656325.67 frames. ], batch size: 106, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:09:40,044 INFO [train.py:968] (0/2) Epoch 10, batch 26250, giga_loss[loss=0.2797, simple_loss=0.3552, pruned_loss=0.1021, over 28449.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3873, pruned_loss=0.1332, over 5650515.09 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.374, pruned_loss=0.1242, over 5717491.44 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3878, pruned_loss=0.1335, over 5649073.37 frames. ], batch size: 71, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:09:52,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2622, 1.8002, 1.3959, 0.4292], device='cuda:0'), covar=tensor([0.2962, 0.1908, 0.3018, 0.4145], device='cuda:0'), in_proj_covar=tensor([0.1541, 0.1475, 0.1484, 0.1270], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 08:09:55,105 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=435860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:09:58,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=435863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:09:58,961 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.99 vs. limit=2.0
+2023-03-05 08:10:03,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.206e+02 1.493e+03 1.982e+03 2.647e+03 7.855e+03, threshold=3.965e+03, percent-clipped=15.0
+2023-03-05 08:10:23,012 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=435892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:10:25,207 INFO [train.py:968] (0/2) Epoch 10, batch 26300, giga_loss[loss=0.3879, simple_loss=0.427, pruned_loss=0.1744, over 29052.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3882, pruned_loss=0.1345, over 5654579.38 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3739, pruned_loss=0.1243, over 5709309.06 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3888, pruned_loss=0.1348, over 5658981.89 frames. ], batch size: 128, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:10:27,259 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 08:11:12,840 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=435942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:11:14,642 INFO [train.py:968] (0/2) Epoch 10, batch 26350, giga_loss[loss=0.2943, simple_loss=0.3608, pruned_loss=0.1139, over 29040.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3876, pruned_loss=0.1349, over 5651374.38 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3742, pruned_loss=0.1244, over 5711407.96 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3881, pruned_loss=0.1352, over 5651637.71 frames. ], batch size: 136, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:11:44,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.888e+02 1.597e+03 1.944e+03 2.812e+03 5.591e+03, threshold=3.887e+03, percent-clipped=4.0
+2023-03-05 08:12:04,731 INFO [train.py:968] (0/2) Epoch 10, batch 26400, giga_loss[loss=0.3099, simple_loss=0.3738, pruned_loss=0.123, over 28256.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3861, pruned_loss=0.1344, over 5647482.30 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3745, pruned_loss=0.1246, over 5712808.95 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3863, pruned_loss=0.1346, over 5645684.73 frames. ], batch size: 368, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:12:09,629 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-436000.pt
+2023-03-05 08:12:26,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=436017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:12:52,367 INFO [train.py:968] (0/2) Epoch 10, batch 26450, giga_loss[loss=0.352, simple_loss=0.398, pruned_loss=0.1529, over 28718.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3843, pruned_loss=0.1339, over 5654032.81 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3747, pruned_loss=0.1246, over 5714043.01 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3845, pruned_loss=0.1341, over 5650885.02 frames. ], batch size: 262, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:13:21,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.612e+03 2.076e+03 3.206e+03 8.032e+03, threshold=4.153e+03, percent-clipped=16.0
+2023-03-05 08:13:35,382 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=436085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:13:38,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=436088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:13:45,156 INFO [train.py:968] (0/2) Epoch 10, batch 26500, giga_loss[loss=0.3989, simple_loss=0.4436, pruned_loss=0.1771, over 28256.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3839, pruned_loss=0.1345, over 5643801.16 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3752, pruned_loss=0.1251, over 5714021.91 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3838, pruned_loss=0.1344, over 5640261.26 frames. ], batch size: 368, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:14:08,826 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=436117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:14:30,356 INFO [train.py:968] (0/2) Epoch 10, batch 26550, giga_loss[loss=0.2905, simple_loss=0.3663, pruned_loss=0.1074, over 28804.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3839, pruned_loss=0.1351, over 5643639.99 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3752, pruned_loss=0.1253, over 5710407.77 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3841, pruned_loss=0.1351, over 5642662.42 frames. ], batch size: 174, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:14:51,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.715e+03 2.193e+03 2.751e+03 5.914e+03, threshold=4.386e+03, percent-clipped=7.0
+2023-03-05 08:15:11,620 INFO [train.py:968] (0/2) Epoch 10, batch 26600, giga_loss[loss=0.329, simple_loss=0.3798, pruned_loss=0.1391, over 28854.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3812, pruned_loss=0.1332, over 5650615.62 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3754, pruned_loss=0.1256, over 5706929.52 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3814, pruned_loss=0.1333, over 5651212.27 frames. ], batch size: 99, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:15:40,746 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0928, 2.5666, 2.0568, 1.5578], device='cuda:0'), covar=tensor([0.2044, 0.1492, 0.1452, 0.1982], device='cuda:0'), in_proj_covar=tensor([0.1664, 0.1568, 0.1547, 0.1652], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 08:15:58,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2041, 1.5995, 1.5696, 1.1246], device='cuda:0'), covar=tensor([0.1494, 0.2155, 0.1199, 0.1390], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0705, 0.0851, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 08:16:01,092 INFO [train.py:968] (0/2) Epoch 10, batch 26650, giga_loss[loss=0.2998, simple_loss=0.3572, pruned_loss=0.1212, over 28838.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3804, pruned_loss=0.1327, over 5665081.18 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3756, pruned_loss=0.1257, over 5707875.87 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3803, pruned_loss=0.1326, over 5664389.72 frames. ], batch size: 99, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:16:03,559 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=436246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:16:23,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.580e+03 2.205e+03 3.056e+03 1.020e+04, threshold=4.411e+03, percent-clipped=11.0
+2023-03-05 08:16:40,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-05 08:16:44,847 INFO [train.py:968] (0/2) Epoch 10, batch 26700, giga_loss[loss=0.4029, simple_loss=0.4339, pruned_loss=0.186, over 27921.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3807, pruned_loss=0.1324, over 5662213.48 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3757, pruned_loss=0.1257, over 5706138.03 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3807, pruned_loss=0.1326, over 5662389.39 frames. ], batch size: 412, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:17:11,006 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-05 08:17:13,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5550, 1.7213, 1.8188, 1.3825], device='cuda:0'), covar=tensor([0.1465, 0.2031, 0.1176, 0.1425], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0704, 0.0851, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 08:17:35,148 INFO [train.py:968] (0/2) Epoch 10, batch 26750, giga_loss[loss=0.2832, simple_loss=0.3609, pruned_loss=0.1028, over 28991.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3826, pruned_loss=0.1328, over 5660511.53 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3756, pruned_loss=0.1257, over 5706740.61 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3828, pruned_loss=0.1331, over 5659715.10 frames. ], batch size: 155, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:17:38,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8070, 4.5534, 4.3199, 2.1471], device='cuda:0'), covar=tensor([0.0568, 0.0843, 0.0911, 0.1828], device='cuda:0'), in_proj_covar=tensor([0.1027, 0.0975, 0.0853, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 08:18:01,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.555e+03 2.115e+03 3.270e+03 8.684e+03, threshold=4.229e+03, percent-clipped=12.0
+2023-03-05 08:18:01,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2771, 4.1050, 3.8943, 1.8540], device='cuda:0'), covar=tensor([0.0572, 0.0698, 0.0749, 0.2014], device='cuda:0'), in_proj_covar=tensor([0.1028, 0.0975, 0.0854, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 08:18:24,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=436392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:18:25,309 INFO [train.py:968] (0/2) Epoch 10, batch 26800, libri_loss[loss=0.3993, simple_loss=0.4424, pruned_loss=0.1781, over 29681.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3816, pruned_loss=0.1323, over 5654032.82 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3762, pruned_loss=0.1263, over 5702604.53 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3813, pruned_loss=0.1321, over 5655131.89 frames. ], batch size: 88, lr: 3.22e-03, grad_scale: 8.0
+2023-03-05 08:18:26,554 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-05 08:18:30,709 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 08:18:40,623 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 08:18:46,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6873, 1.7925, 1.9755, 1.4996], device='cuda:0'), covar=tensor([0.1528, 0.1998, 0.1163, 0.1407], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0704, 0.0850, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 08:19:09,486 INFO [train.py:968] (0/2) Epoch 10, batch 26850, giga_loss[loss=0.2971, simple_loss=0.3836, pruned_loss=0.1053, over 28980.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3832, pruned_loss=0.1321, over 5663871.04 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3764, pruned_loss=0.1264, over 5704448.25 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.383, pruned_loss=0.1321, over 5661516.01 frames. ], batch size: 106, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:19:33,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.661e+02 1.531e+03 1.995e+03 2.657e+03 1.093e+04, threshold=3.989e+03, percent-clipped=4.0
+2023-03-05 08:19:56,215 INFO [train.py:968] (0/2) Epoch 10, batch 26900, giga_loss[loss=0.2656, simple_loss=0.3528, pruned_loss=0.08923, over 28509.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3837, pruned_loss=0.1299, over 5662876.46 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3764, pruned_loss=0.1264, over 5696868.60 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3836, pruned_loss=0.1299, over 5668298.49 frames. ], batch size: 78, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:20:32,486 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=436535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:20:35,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=436538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:20:39,976 INFO [train.py:968] (0/2) Epoch 10, batch 26950, giga_loss[loss=0.3518, simple_loss=0.4039, pruned_loss=0.1498, over 27915.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3843, pruned_loss=0.1285, over 5664646.41 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3758, pruned_loss=0.1262, over 5693872.17 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3849, pruned_loss=0.1287, over 5670239.63 frames. ], batch size: 412, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:20:59,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=436567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:21:01,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.484e+02 1.361e+03 1.761e+03 2.466e+03 1.023e+04, threshold=3.521e+03, percent-clipped=8.0
+2023-03-05 08:21:22,433 INFO [train.py:968] (0/2) Epoch 10, batch 27000, giga_loss[loss=0.3175, simple_loss=0.3799, pruned_loss=0.1275, over 29062.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3871, pruned_loss=0.1307, over 5672260.35 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3766, pruned_loss=0.127, over 5697980.81 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3872, pruned_loss=0.1303, over 5672266.83 frames. ], batch size: 164, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:21:22,439 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 08:21:30,946 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2162, simple_loss=0.3214, pruned_loss=0.05557, over 944034.00 frames. 
+2023-03-05 08:21:30,947 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 08:21:56,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=436621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:22:19,772 INFO [train.py:968] (0/2) Epoch 10, batch 27050, giga_loss[loss=0.3296, simple_loss=0.3902, pruned_loss=0.1345, over 28737.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3904, pruned_loss=0.1349, over 5671110.10 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3765, pruned_loss=0.1269, over 5697829.80 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3908, pruned_loss=0.1348, over 5670759.89 frames. ], batch size: 262, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:22:50,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.565e+03 2.213e+03 3.129e+03 1.948e+04, threshold=4.426e+03, percent-clipped=18.0
+2023-03-05 08:23:13,767 INFO [train.py:968] (0/2) Epoch 10, batch 27100, giga_loss[loss=0.3506, simple_loss=0.394, pruned_loss=0.1536, over 28955.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.39, pruned_loss=0.1354, over 5675139.91 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3761, pruned_loss=0.1267, over 5700039.69 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3908, pruned_loss=0.1356, over 5672599.44 frames. ], batch size: 213, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:23:26,719 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3826, 1.7301, 1.3446, 1.4430], device='cuda:0'), covar=tensor([0.2362, 0.2191, 0.2475, 0.1962], device='cuda:0'), in_proj_covar=tensor([0.1268, 0.0948, 0.1125, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 08:23:59,449 INFO [train.py:968] (0/2) Epoch 10, batch 27150, giga_loss[loss=0.3055, simple_loss=0.377, pruned_loss=0.117, over 28966.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.389, pruned_loss=0.1349, over 5665230.33 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.376, pruned_loss=0.1266, over 5698150.27 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3903, pruned_loss=0.1355, over 5664607.34 frames. ], batch size: 106, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:23:59,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 08:24:07,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7998, 1.8850, 1.2973, 1.5296], device='cuda:0'), covar=tensor([0.0786, 0.0626, 0.1036, 0.1038], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0443, 0.0502, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 08:24:21,823 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=436764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:24:24,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=436767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:24:27,482 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.578e+02 1.484e+03 1.914e+03 2.545e+03 8.158e+03, threshold=3.829e+03, percent-clipped=8.0
+2023-03-05 08:24:47,790 INFO [train.py:968] (0/2) Epoch 10, batch 27200, giga_loss[loss=0.2912, simple_loss=0.3779, pruned_loss=0.1023, over 28511.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3864, pruned_loss=0.1322, over 5664404.45 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3753, pruned_loss=0.1262, over 5690667.76 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3884, pruned_loss=0.1331, over 5670853.24 frames. ], batch size: 60, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:24:49,501 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=436796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:25:15,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1999, 1.3872, 3.6551, 3.1065], device='cuda:0'), covar=tensor([0.1614, 0.2438, 0.0483, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0649, 0.0585, 0.0851, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 08:25:36,421 INFO [train.py:968] (0/2) Epoch 10, batch 27250, giga_loss[loss=0.3409, simple_loss=0.4108, pruned_loss=0.1355, over 28898.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3865, pruned_loss=0.1307, over 5645668.80 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3758, pruned_loss=0.1266, over 5682781.44 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3877, pruned_loss=0.1311, over 5658080.00 frames. ], batch size: 174, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:26:05,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.573e+02 1.432e+03 1.826e+03 2.362e+03 1.314e+04, threshold=3.653e+03, percent-clipped=6.0
+2023-03-05 08:26:26,478 INFO [train.py:968] (0/2) Epoch 10, batch 27300, giga_loss[loss=0.3115, simple_loss=0.3824, pruned_loss=0.1203, over 28704.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3883, pruned_loss=0.1314, over 5648607.50 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.376, pruned_loss=0.1268, over 5675082.88 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3891, pruned_loss=0.1317, over 5665533.82 frames. ], batch size: 262, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:26:30,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2257, 1.2815, 1.3123, 1.4212], device='cuda:0'), covar=tensor([0.0754, 0.0331, 0.0302, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0078, 0.0056, 0.0051, 0.0085], device='cuda:0')
+2023-03-05 08:27:19,189 INFO [train.py:968] (0/2) Epoch 10, batch 27350, giga_loss[loss=0.3223, simple_loss=0.3848, pruned_loss=0.1299, over 28732.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3889, pruned_loss=0.1331, over 5639174.18 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3756, pruned_loss=0.1264, over 5678551.99 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3902, pruned_loss=0.1336, over 5648938.63 frames. ], batch size: 262, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:27:43,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.591e+02 1.553e+03 2.128e+03 3.508e+03 1.523e+04, threshold=4.256e+03, percent-clipped=23.0
+2023-03-05 08:28:05,611 INFO [train.py:968] (0/2) Epoch 10, batch 27400, giga_loss[loss=0.3243, simple_loss=0.3848, pruned_loss=0.1319, over 28728.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3874, pruned_loss=0.1322, over 5646433.63 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3758, pruned_loss=0.1265, over 5677046.76 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3884, pruned_loss=0.1327, over 5655128.18 frames. ], batch size: 262, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:28:46,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1465, 1.4376, 1.2571, 0.9753], device='cuda:0'), covar=tensor([0.1787, 0.1643, 0.1096, 0.1567], device='cuda:0'), in_proj_covar=tensor([0.1663, 0.1572, 0.1557, 0.1656], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 08:28:57,088 INFO [train.py:968] (0/2) Epoch 10, batch 27450, giga_loss[loss=0.3438, simple_loss=0.3958, pruned_loss=0.1459, over 28256.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3845, pruned_loss=0.131, over 5662737.24 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3751, pruned_loss=0.1261, over 5680207.09 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.386, pruned_loss=0.1317, over 5666408.67 frames. ], batch size: 368, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:29:26,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.565e+03 1.994e+03 2.945e+03 7.942e+03, threshold=3.988e+03, percent-clipped=6.0
+2023-03-05 08:29:35,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3344, 1.6362, 1.2323, 1.6760], device='cuda:0'), covar=tensor([0.0770, 0.0291, 0.0322, 0.0788], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:0')
+2023-03-05 08:29:48,727 INFO [train.py:968] (0/2) Epoch 10, batch 27500, giga_loss[loss=0.2808, simple_loss=0.3524, pruned_loss=0.1046, over 28753.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3815, pruned_loss=0.1296, over 5662536.87 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3753, pruned_loss=0.1262, over 5683176.99 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3827, pruned_loss=0.1302, over 5662552.87 frames. ], batch size: 119, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:30:35,237 INFO [train.py:968] (0/2) Epoch 10, batch 27550, giga_loss[loss=0.3947, simple_loss=0.43, pruned_loss=0.1797, over 27910.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3812, pruned_loss=0.13, over 5659359.98 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3759, pruned_loss=0.1266, over 5679684.71 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3818, pruned_loss=0.1302, over 5661242.83 frames. ], batch size: 412, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:31:00,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.533e+03 1.984e+03 2.500e+03 8.636e+03, threshold=3.967e+03, percent-clipped=8.0
+2023-03-05 08:31:18,427 INFO [train.py:968] (0/2) Epoch 10, batch 27600, libri_loss[loss=0.255, simple_loss=0.3241, pruned_loss=0.09293, over 29656.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3819, pruned_loss=0.1315, over 5647544.01 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3764, pruned_loss=0.1269, over 5664598.61 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3821, pruned_loss=0.1315, over 5661529.48 frames. ], batch size: 73, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:31:58,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:32:00,264 INFO [train.py:968] (0/2) Epoch 10, batch 27650, giga_loss[loss=0.3043, simple_loss=0.3694, pruned_loss=0.1196, over 28953.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3792, pruned_loss=0.1295, over 5654748.37 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3762, pruned_loss=0.1269, over 5670430.02 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3797, pruned_loss=0.1297, over 5660049.66 frames. ], batch size: 227, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:32:19,872 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:32:28,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.037e+02 1.345e+03 1.865e+03 2.529e+03 9.492e+03, threshold=3.730e+03, percent-clipped=6.0
+2023-03-05 08:32:47,442 INFO [train.py:968] (0/2) Epoch 10, batch 27700, giga_loss[loss=0.2619, simple_loss=0.3421, pruned_loss=0.09084, over 28969.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3766, pruned_loss=0.1262, over 5660129.10 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3764, pruned_loss=0.1271, over 5675095.88 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3768, pruned_loss=0.1262, over 5660106.17 frames. ], batch size: 227, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:33:00,666 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 08:33:34,941 INFO [train.py:968] (0/2) Epoch 10, batch 27750, giga_loss[loss=0.3405, simple_loss=0.4001, pruned_loss=0.1405, over 28610.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3755, pruned_loss=0.1252, over 5659172.45 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3764, pruned_loss=0.127, over 5677884.56 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3757, pruned_loss=0.1252, over 5656339.70 frames. ], batch size: 307, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:34:06,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.993e+02 1.261e+03 1.849e+03 2.967e+03 1.345e+04, threshold=3.698e+03, percent-clipped=20.0
+2023-03-05 08:34:25,111 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-05 08:34:27,662 INFO [train.py:968] (0/2) Epoch 10, batch 27800, giga_loss[loss=0.2567, simple_loss=0.3304, pruned_loss=0.09151, over 28977.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3729, pruned_loss=0.1238, over 5663241.72 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.376, pruned_loss=0.1268, over 5681291.19 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3734, pruned_loss=0.1239, over 5657682.09 frames. ], batch size: 164, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:34:40,324 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-05 08:35:17,585 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:35:26,098 INFO [train.py:968] (0/2) Epoch 10, batch 27850, giga_loss[loss=0.2915, simple_loss=0.3669, pruned_loss=0.108, over 28935.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3696, pruned_loss=0.1231, over 5653574.98 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3756, pruned_loss=0.1267, over 5682308.88 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3702, pruned_loss=0.1233, over 5648073.75 frames. ], batch size: 145, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:35:56,760 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.254e+02 1.854e+03 2.208e+03 3.190e+03 9.341e+03, threshold=4.416e+03, percent-clipped=16.0
+2023-03-05 08:36:14,682 INFO [train.py:968] (0/2) Epoch 10, batch 27900, giga_loss[loss=0.2996, simple_loss=0.3704, pruned_loss=0.1145, over 28924.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3704, pruned_loss=0.1235, over 5659003.22 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3755, pruned_loss=0.1266, over 5688571.26 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3709, pruned_loss=0.1237, over 5648223.41 frames. ], batch size: 164, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:36:45,744 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5306, 2.0782, 1.7292, 1.8276], device='cuda:0'), covar=tensor([0.0665, 0.0708, 0.0958, 0.1030], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0440, 0.0500, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 08:37:03,576 INFO [train.py:968] (0/2) Epoch 10, batch 27950, libri_loss[loss=0.2865, simple_loss=0.3614, pruned_loss=0.1057, over 29552.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3722, pruned_loss=0.1236, over 5667099.95 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3758, pruned_loss=0.1267, over 5693508.55 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3722, pruned_loss=0.1235, over 5653347.27 frames. ], batch size: 77, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:37:10,616 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6915, 1.8942, 1.9203, 1.5239], device='cuda:0'), covar=tensor([0.1699, 0.2127, 0.1275, 0.1509], device='cuda:0'), in_proj_covar=tensor([0.0821, 0.0708, 0.0858, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 08:37:31,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.667e+02 1.480e+03 1.882e+03 2.487e+03 4.320e+03, threshold=3.764e+03, percent-clipped=0.0
+2023-03-05 08:37:54,537 INFO [train.py:968] (0/2) Epoch 10, batch 28000, giga_loss[loss=0.3226, simple_loss=0.3856, pruned_loss=0.1298, over 28900.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.372, pruned_loss=0.1231, over 5657858.35 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.376, pruned_loss=0.1268, over 5694700.00 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3719, pruned_loss=0.123, over 5645963.31 frames. ], batch size: 186, lr: 3.22e-03, grad_scale: 8.0
+2023-03-05 08:38:15,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=437616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:38:37,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=437639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:38:40,972 INFO [train.py:968] (0/2) Epoch 10, batch 28050, giga_loss[loss=0.3655, simple_loss=0.3976, pruned_loss=0.1667, over 23466.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3728, pruned_loss=0.1237, over 5655953.21 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3763, pruned_loss=0.1268, over 5697501.76 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3723, pruned_loss=0.1235, over 5642325.25 frames. ], batch size: 705, lr: 3.22e-03, grad_scale: 8.0
+2023-03-05 08:38:50,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4570, 2.2376, 1.5238, 0.6607], device='cuda:0'), covar=tensor([0.3294, 0.1907, 0.2978, 0.3633], device='cuda:0'), in_proj_covar=tensor([0.1538, 0.1453, 0.1468, 0.1267], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 08:39:06,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.376e+02 1.565e+03 1.866e+03 2.435e+03 8.394e+03, threshold=3.731e+03, percent-clipped=6.0
+2023-03-05 08:39:11,872 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:39:22,848 INFO [train.py:968] (0/2) Epoch 10, batch 28100, giga_loss[loss=0.3302, simple_loss=0.3897, pruned_loss=0.1353, over 29036.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3737, pruned_loss=0.1251, over 5648878.30 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3764, pruned_loss=0.127, over 5689377.38 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.373, pruned_loss=0.1246, over 5643429.96 frames. ], batch size: 155, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:39:32,337 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:11,399 INFO [train.py:968] (0/2) Epoch 10, batch 28150, giga_loss[loss=0.4043, simple_loss=0.4256, pruned_loss=0.1915, over 26439.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.376, pruned_loss=0.1267, over 5641906.75 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3766, pruned_loss=0.1271, over 5681884.52 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3753, pruned_loss=0.1262, over 5644689.76 frames. ], batch size: 555, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:40:25,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=437759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:29,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=437762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:37,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.513e+02 1.596e+03 1.925e+03 2.678e+03 5.809e+03, threshold=3.849e+03, percent-clipped=6.0
+2023-03-05 08:40:42,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6271, 1.6545, 1.2347, 1.3255], device='cuda:0'), covar=tensor([0.0727, 0.0642, 0.0934, 0.1179], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0441, 0.0499, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 08:40:45,170 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=437782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:49,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=437785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:53,987 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=437791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:56,955 INFO [train.py:968] (0/2) Epoch 10, batch 28200, giga_loss[loss=0.3089, simple_loss=0.3662, pruned_loss=0.1259, over 28605.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3779, pruned_loss=0.1279, over 5645259.08 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3765, pruned_loss=0.1271, over 5675758.71 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3774, pruned_loss=0.1276, over 5651788.24 frames. ], batch size: 92, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:41:14,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=437810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:41:15,339 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:41:18,462 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=437814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:41:49,545 INFO [train.py:968] (0/2) Epoch 10, batch 28250, giga_loss[loss=0.3355, simple_loss=0.391, pruned_loss=0.1399, over 28534.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3804, pruned_loss=0.1304, over 5630131.08 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3763, pruned_loss=0.127, over 5669600.34 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3802, pruned_loss=0.1303, over 5639789.59 frames. ], batch size: 336, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:42:15,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.184e+03 1.481e+03 2.059e+03 2.781e+03 5.372e+03, threshold=4.118e+03, percent-clipped=5.0
+2023-03-05 08:42:24,534 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 08:42:40,463 INFO [train.py:968] (0/2) Epoch 10, batch 28300, giga_loss[loss=0.3301, simple_loss=0.3754, pruned_loss=0.1425, over 23660.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3803, pruned_loss=0.1308, over 5636565.46 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3762, pruned_loss=0.1269, over 5674190.66 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3804, pruned_loss=0.1308, over 5639672.03 frames. ], batch size: 705, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:43:30,437 INFO [train.py:968] (0/2) Epoch 10, batch 28350, giga_loss[loss=0.2925, simple_loss=0.368, pruned_loss=0.1085, over 28745.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3798, pruned_loss=0.1287, over 5648780.60 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3755, pruned_loss=0.1267, over 5679208.14 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3806, pruned_loss=0.1291, over 5645721.78 frames. ], batch size: 243, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:43:39,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=437953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:43:41,168 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=437956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:43:57,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3024, 2.8637, 1.3786, 1.4667], device='cuda:0'), covar=tensor([0.0873, 0.0354, 0.0845, 0.1212], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0512, 0.0336, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 08:44:00,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.703e+03 2.329e+03 3.168e+03 8.593e+03, threshold=4.657e+03, percent-clipped=14.0
+2023-03-05 08:44:10,026 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=437985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:44:19,704 INFO [train.py:968] (0/2) Epoch 10, batch 28400, giga_loss[loss=0.3297, simple_loss=0.3668, pruned_loss=0.1463, over 23735.00 frames. ], tot_loss[loss=0.321, simple_loss=0.381, pruned_loss=0.1305, over 5633807.73 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3754, pruned_loss=0.1269, over 5673889.13 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3819, pruned_loss=0.1306, over 5635734.84 frames. ], batch size: 705, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:44:27,005 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-438000.pt
+2023-03-05 08:44:52,330 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:45:04,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4080, 4.2336, 4.0050, 1.9182], device='cuda:0'), covar=tensor([0.0531, 0.0705, 0.0740, 0.2078], device='cuda:0'), in_proj_covar=tensor([0.1050, 0.0993, 0.0871, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 08:45:07,508 INFO [train.py:968] (0/2) Epoch 10, batch 28450, giga_loss[loss=0.3744, simple_loss=0.4026, pruned_loss=0.1731, over 26642.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3805, pruned_loss=0.1308, over 5635948.94 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.375, pruned_loss=0.1267, over 5679237.28 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3817, pruned_loss=0.1312, over 5631656.40 frames. ], batch size: 555, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:45:18,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:45:40,753 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-05 08:45:40,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.559e+03 2.043e+03 2.775e+03 5.825e+03, threshold=4.087e+03, percent-clipped=5.0
+2023-03-05 08:45:45,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:46:03,506 INFO [train.py:968] (0/2) Epoch 10, batch 28500, giga_loss[loss=0.3758, simple_loss=0.4086, pruned_loss=0.1715, over 26653.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3802, pruned_loss=0.1309, over 5637285.66 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3755, pruned_loss=0.1271, over 5681389.40 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3809, pruned_loss=0.1311, over 5630647.43 frames. ], batch size: 555, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:46:27,976 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:46:58,365 INFO [train.py:968] (0/2) Epoch 10, batch 28550, giga_loss[loss=0.3253, simple_loss=0.3814, pruned_loss=0.1346, over 28566.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3789, pruned_loss=0.1308, over 5629695.05 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3755, pruned_loss=0.1271, over 5684785.89 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3795, pruned_loss=0.131, over 5619583.21 frames. ], batch size: 307, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:47:02,844 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438148.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 08:47:05,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:47:27,172 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3790, 1.7124, 1.7211, 1.2549], device='cuda:0'), covar=tensor([0.1434, 0.2123, 0.1149, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0820, 0.0707, 0.0856, 0.0766], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 08:47:30,016 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.549e+02 1.428e+03 1.919e+03 2.678e+03 7.372e+03, threshold=3.837e+03, percent-clipped=5.0
+2023-03-05 08:47:38,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:47:46,503 INFO [train.py:968] (0/2) Epoch 10, batch 28600, giga_loss[loss=0.2932, simple_loss=0.3551, pruned_loss=0.1156, over 28914.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3776, pruned_loss=0.1297, over 5649094.74 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3755, pruned_loss=0.1269, over 5687767.95 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3782, pruned_loss=0.1301, over 5637701.65 frames. ], batch size: 186, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:47:48,967 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:47:51,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:47:53,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4133, 1.5372, 1.4181, 1.3992], device='cuda:0'), covar=tensor([0.1210, 0.1563, 0.1913, 0.1502], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0739, 0.0669, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 08:48:06,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2277, 1.2469, 1.1346, 0.9057], device='cuda:0'), covar=tensor([0.0764, 0.0514, 0.0964, 0.0944], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0442, 0.0497, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 08:48:12,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:48:15,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:48:18,876 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:48:22,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4812, 1.6195, 1.4458, 1.2753], device='cuda:0'), covar=tensor([0.2013, 0.1836, 0.1343, 0.1753], device='cuda:0'), in_proj_covar=tensor([0.1675, 0.1598, 0.1560, 0.1652], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 08:48:33,135 INFO [train.py:968] (0/2) Epoch 10, batch 28650, libri_loss[loss=0.2703, simple_loss=0.3442, pruned_loss=0.09816, over 29564.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3777, pruned_loss=0.1301, over 5660510.43 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3754, pruned_loss=0.1269, over 5693874.41 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3783, pruned_loss=0.1306, over 5644916.28 frames. ], batch size: 76, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:48:38,593 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-05 08:48:42,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:49:02,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.628e+03 1.920e+03 2.779e+03 1.508e+04, threshold=3.840e+03, percent-clipped=13.0
+2023-03-05 08:49:20,895 INFO [train.py:968] (0/2) Epoch 10, batch 28700, libri_loss[loss=0.2857, simple_loss=0.3532, pruned_loss=0.1092, over 29574.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3785, pruned_loss=0.1303, over 5656207.15 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3753, pruned_loss=0.1267, over 5687874.90 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3791, pruned_loss=0.1309, over 5647220.06 frames. ], batch size: 74, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:49:55,615 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:49:55,737 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-05 08:49:58,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:50:11,173 INFO [train.py:968] (0/2) Epoch 10, batch 28750, giga_loss[loss=0.3167, simple_loss=0.3791, pruned_loss=0.1271, over 28856.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3795, pruned_loss=0.1311, over 5661669.85 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3754, pruned_loss=0.1267, over 5688966.52 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3799, pruned_loss=0.1316, over 5653596.42 frames. ], batch size: 227, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:50:25,629 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:50:40,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.158e+03 1.854e+03 2.294e+03 3.937e+03 9.752e+03, threshold=4.587e+03, percent-clipped=26.0
+2023-03-05 08:50:41,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-05 08:50:57,333 INFO [train.py:968] (0/2) Epoch 10, batch 28800, giga_loss[loss=0.2906, simple_loss=0.3631, pruned_loss=0.109, over 28421.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3824, pruned_loss=0.1334, over 5664169.44 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3759, pruned_loss=0.1271, over 5693873.15 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3824, pruned_loss=0.1336, over 5652547.08 frames. ], batch size: 65, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 08:51:05,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:51:43,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6601, 3.3879, 1.5540, 1.6768], device='cuda:0'), covar=tensor([0.0771, 0.0331, 0.0838, 0.1132], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0508, 0.0333, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 08:51:43,453 INFO [train.py:968] (0/2) Epoch 10, batch 28850, libri_loss[loss=0.3332, simple_loss=0.3921, pruned_loss=0.1371, over 29519.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3836, pruned_loss=0.1351, over 5671983.44 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3761, pruned_loss=0.1273, over 5697280.98 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3836, pruned_loss=0.1352, over 5659069.97 frames. ], batch size: 82, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:51:55,668 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:52:15,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.930e+02 1.769e+03 2.553e+03 3.874e+03 8.782e+03, threshold=5.105e+03, percent-clipped=16.0
+2023-03-05 08:52:24,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:52:29,736 INFO [train.py:968] (0/2) Epoch 10, batch 28900, giga_loss[loss=0.2927, simple_loss=0.3552, pruned_loss=0.1151, over 28756.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3843, pruned_loss=0.1358, over 5678602.32 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3767, pruned_loss=0.1278, over 5700886.87 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.384, pruned_loss=0.1356, over 5664511.80 frames. ], batch size: 99, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:52:35,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1368, 1.1519, 3.7444, 2.9894], device='cuda:0'), covar=tensor([0.1614, 0.2630, 0.0476, 0.1021], device='cuda:0'), in_proj_covar=tensor([0.0651, 0.0583, 0.0853, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 08:52:55,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 08:52:57,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:53:14,154 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:53:14,556 INFO [train.py:968] (0/2) Epoch 10, batch 28950, giga_loss[loss=0.2888, simple_loss=0.3634, pruned_loss=0.1071, over 28853.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3843, pruned_loss=0.1347, over 5677613.48 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3771, pruned_loss=0.128, over 5692928.00 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3838, pruned_loss=0.1346, over 5672628.89 frames. ], batch size: 199, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:53:15,521 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:53:19,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:53:48,373 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:53:48,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.891e+02 1.492e+03 1.841e+03 2.756e+03 1.012e+04, threshold=3.683e+03, percent-clipped=2.0
+2023-03-05 08:54:06,783 INFO [train.py:968] (0/2) Epoch 10, batch 29000, libri_loss[loss=0.3017, simple_loss=0.3541, pruned_loss=0.1247, over 28582.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3852, pruned_loss=0.1358, over 5670959.61 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3768, pruned_loss=0.1279, over 5697622.65 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3853, pruned_loss=0.136, over 5661922.42 frames. ], batch size: 63, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:54:23,954 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-05 08:54:41,021 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:54:43,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:54:54,708 INFO [train.py:968] (0/2) Epoch 10, batch 29050, giga_loss[loss=0.3034, simple_loss=0.3676, pruned_loss=0.1196, over 28799.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3851, pruned_loss=0.1355, over 5674695.10 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.377, pruned_loss=0.128, over 5694929.54 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.385, pruned_loss=0.1355, over 5669856.30 frames. ], batch size: 119, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:55:08,464 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 08:55:13,837 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:55:16,690 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 08:55:18,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:55:18,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438669.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 08:55:19,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:55:23,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.086e+03 1.718e+03 2.133e+03 3.302e+03 6.606e+03, threshold=4.267e+03, percent-clipped=18.0
+2023-03-05 08:55:39,259 INFO [train.py:968] (0/2) Epoch 10, batch 29100, giga_loss[loss=0.4101, simple_loss=0.4221, pruned_loss=0.1991, over 23726.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3868, pruned_loss=0.1372, over 5673818.44 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3772, pruned_loss=0.1282, over 5703030.24 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3869, pruned_loss=0.1374, over 5661533.01 frames. ], batch size: 705, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:55:42,099 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438698.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 08:55:43,312 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:56:22,685 INFO [train.py:968] (0/2) Epoch 10, batch 29150, libri_loss[loss=0.321, simple_loss=0.3907, pruned_loss=0.1256, over 29634.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3863, pruned_loss=0.1366, over 5669645.62 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3775, pruned_loss=0.1284, over 5698488.09 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3862, pruned_loss=0.1367, over 5662639.76 frames. ], batch size: 91, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:56:39,816 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-05 08:56:54,421 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.675e+03 2.209e+03 3.311e+03 1.115e+04, threshold=4.419e+03, percent-clipped=10.0
+2023-03-05 08:57:11,834 INFO [train.py:968] (0/2) Epoch 10, batch 29200, giga_loss[loss=0.318, simple_loss=0.3824, pruned_loss=0.1268, over 29009.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3855, pruned_loss=0.1351, over 5645759.84 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3777, pruned_loss=0.1285, over 5691543.87 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3854, pruned_loss=0.1353, over 5644819.32 frames. ], batch size: 213, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 08:57:25,432 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 08:57:40,111 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 08:57:49,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:58:01,500 INFO [train.py:968] (0/2) Epoch 10, batch 29250, giga_loss[loss=0.3423, simple_loss=0.4039, pruned_loss=0.1403, over 29044.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3858, pruned_loss=0.1348, over 5642434.00 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3776, pruned_loss=0.1284, over 5689338.76 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3862, pruned_loss=0.1352, over 5642219.67 frames. ], batch size: 155, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 08:58:31,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.507e+02 1.407e+03 1.955e+03 2.562e+03 4.601e+03, threshold=3.911e+03, percent-clipped=2.0
+2023-03-05 08:58:45,861 INFO [train.py:968] (0/2) Epoch 10, batch 29300, giga_loss[loss=0.2791, simple_loss=0.351, pruned_loss=0.1036, over 29080.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3845, pruned_loss=0.1331, over 5651596.00 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3778, pruned_loss=0.1286, over 5691373.12 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3846, pruned_loss=0.1334, over 5649284.29 frames. ], batch size: 155, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 08:59:03,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-05 08:59:10,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:59:31,441 INFO [train.py:968] (0/2) Epoch 10, batch 29350, giga_loss[loss=0.3411, simple_loss=0.3949, pruned_loss=0.1437, over 28577.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3826, pruned_loss=0.1319, over 5662430.98 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3775, pruned_loss=0.1285, over 5697001.44 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3831, pruned_loss=0.1323, over 5654697.49 frames. ], batch size: 85, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 08:59:47,075 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438959.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 08:59:56,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1074, 1.2345, 3.4815, 2.9098], device='cuda:0'), covar=tensor([0.1617, 0.2410, 0.0442, 0.1231], device='cuda:0'), in_proj_covar=tensor([0.0654, 0.0580, 0.0857, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 08:59:59,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:00:01,190 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.847e+02 1.642e+03 2.260e+03 2.932e+03 9.636e+03, threshold=4.521e+03, percent-clipped=8.0
+2023-03-05 09:00:02,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:00:19,634 INFO [train.py:968] (0/2) Epoch 10, batch 29400, giga_loss[loss=0.3434, simple_loss=0.4019, pruned_loss=0.1424, over 28252.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3832, pruned_loss=0.1325, over 5659973.29 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3772, pruned_loss=0.1282, over 5701824.78 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.384, pruned_loss=0.133, over 5648786.83 frames. ], batch size: 368, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:00:31,031 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:00:32,410 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-05 09:01:10,718 INFO [train.py:968] (0/2) Epoch 10, batch 29450, giga_loss[loss=0.2893, simple_loss=0.3476, pruned_loss=0.1154, over 28591.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3846, pruned_loss=0.1337, over 5650604.06 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3771, pruned_loss=0.1282, over 5691952.00 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3853, pruned_loss=0.1342, over 5649146.11 frames. ], batch size: 85, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:01:11,200 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-05 09:01:28,180 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=439061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:01:28,789 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4474, 2.8234, 1.6085, 1.5581], device='cuda:0'), covar=tensor([0.0695, 0.0325, 0.0651, 0.1003], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0513, 0.0335, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 09:01:31,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=439064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:01:40,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.527e+03 2.182e+03 2.959e+03 1.086e+04, threshold=4.364e+03, percent-clipped=7.0
+2023-03-05 09:01:41,322 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-05 09:01:56,831 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:01:57,273 INFO [train.py:968] (0/2) Epoch 10, batch 29500, giga_loss[loss=0.3097, simple_loss=0.3713, pruned_loss=0.1241, over 28594.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3841, pruned_loss=0.134, over 5655124.25 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3775, pruned_loss=0.1282, over 5695602.43 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3845, pruned_loss=0.1346, over 5649578.96 frames. ], batch size: 92, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:02:01,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4261, 1.5696, 1.4180, 1.2489], device='cuda:0'), covar=tensor([0.1475, 0.1435, 0.1014, 0.1437], device='cuda:0'), in_proj_covar=tensor([0.1668, 0.1592, 0.1547, 0.1651], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 09:02:44,039 INFO [train.py:968] (0/2) Epoch 10, batch 29550, giga_loss[loss=0.303, simple_loss=0.3678, pruned_loss=0.1191, over 28886.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.384, pruned_loss=0.1343, over 5668041.10 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3773, pruned_loss=0.1282, over 5697121.45 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3846, pruned_loss=0.1349, over 5661527.68 frames. ], batch size: 227, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:03:13,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.910e+02 1.533e+03 1.838e+03 2.203e+03 5.851e+03, threshold=3.676e+03, percent-clipped=5.0
+2023-03-05 09:03:29,745 INFO [train.py:968] (0/2) Epoch 10, batch 29600, giga_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1224, over 28742.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3856, pruned_loss=0.1359, over 5668110.29 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3772, pruned_loss=0.128, over 5704334.53 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3864, pruned_loss=0.1368, over 5655487.00 frames. ], batch size: 99, lr: 3.21e-03, grad_scale: 8.0
+2023-03-05 09:03:53,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8013, 0.8850, 0.7808, 0.7670], device='cuda:0'), covar=tensor([0.0970, 0.1272, 0.0798, 0.1118], device='cuda:0'), in_proj_covar=tensor([0.1666, 0.1588, 0.1545, 0.1649], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 09:04:18,656 INFO [train.py:968] (0/2) Epoch 10, batch 29650, giga_loss[loss=0.2833, simple_loss=0.3577, pruned_loss=0.1045, over 28792.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3861, pruned_loss=0.1363, over 5660651.08 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3775, pruned_loss=0.1282, over 5705638.17 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3867, pruned_loss=0.1369, over 5649045.73 frames. ], batch size: 242, lr: 3.21e-03, grad_scale: 8.0
+2023-03-05 09:04:51,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.525e+02 1.377e+03 1.666e+03 2.409e+03 7.631e+03, threshold=3.332e+03, percent-clipped=6.0
+2023-03-05 09:05:08,838 INFO [train.py:968] (0/2) Epoch 10, batch 29700, giga_loss[loss=0.299, simple_loss=0.367, pruned_loss=0.1154, over 28909.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3859, pruned_loss=0.1363, over 5657358.34 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3775, pruned_loss=0.1283, over 5707714.50 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3864, pruned_loss=0.1368, over 5646148.69 frames. ], batch size: 186, lr: 3.21e-03, grad_scale: 8.0
+2023-03-05 09:05:27,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3823, 1.5122, 1.3910, 1.5691], device='cuda:0'), covar=tensor([0.0613, 0.0275, 0.0271, 0.0614], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 09:05:49,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=439334.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 09:05:56,216 INFO [train.py:968] (0/2) Epoch 10, batch 29750, giga_loss[loss=0.2995, simple_loss=0.3694, pruned_loss=0.1148, over 28445.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3858, pruned_loss=0.1356, over 5658130.56 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3777, pruned_loss=0.1283, over 5707158.43 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3862, pruned_loss=0.1362, over 5648353.38 frames. ], batch size: 60, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:06:26,802 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.409e+02 1.538e+03 2.138e+03 3.077e+03 6.114e+03, threshold=4.277e+03, percent-clipped=20.0
+2023-03-05 09:06:41,264 INFO [train.py:968] (0/2) Epoch 10, batch 29800, giga_loss[loss=0.3477, simple_loss=0.4093, pruned_loss=0.143, over 28708.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3843, pruned_loss=0.1338, over 5661331.19 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3771, pruned_loss=0.1278, over 5712298.70 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3854, pruned_loss=0.1349, over 5647394.84 frames. ], batch size: 284, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:06:46,117 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=439400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:07:29,469 INFO [train.py:968] (0/2) Epoch 10, batch 29850, giga_loss[loss=0.337, simple_loss=0.3898, pruned_loss=0.1421, over 28233.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3827, pruned_loss=0.1326, over 5658458.73 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3768, pruned_loss=0.1276, over 5707353.27 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.384, pruned_loss=0.1338, over 5649888.35 frames. ], batch size: 368, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:08:01,098 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=439477.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 09:08:01,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.440e+02 1.692e+03 2.097e+03 2.751e+03 9.408e+03, threshold=4.194e+03, percent-clipped=6.0
+2023-03-05 09:08:04,962 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=439480.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 09:08:16,245 INFO [train.py:968] (0/2) Epoch 10, batch 29900, giga_loss[loss=0.3201, simple_loss=0.3753, pruned_loss=0.1325, over 28868.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3827, pruned_loss=0.1326, over 5659907.55 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3774, pruned_loss=0.128, over 5701977.08 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3832, pruned_loss=0.1333, over 5657002.38 frames. ], batch size: 106, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:08:29,886 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439509.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 09:09:01,043 INFO [train.py:968] (0/2) Epoch 10, batch 29950, giga_loss[loss=0.3098, simple_loss=0.3676, pruned_loss=0.126, over 28973.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3794, pruned_loss=0.1305, over 5665818.84 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3771, pruned_loss=0.1278, over 5706193.96 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3802, pruned_loss=0.1314, over 5659089.62 frames. ], batch size: 136, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:09:37,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.581e+02 1.749e+03 2.303e+03 3.249e+03 1.223e+04, threshold=4.607e+03, percent-clipped=16.0
+2023-03-05 09:09:50,993 INFO [train.py:968] (0/2) Epoch 10, batch 30000, giga_loss[loss=0.3131, simple_loss=0.3777, pruned_loss=0.1242, over 28919.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3762, pruned_loss=0.1293, over 5649097.10 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3776, pruned_loss=0.1282, over 5698020.99 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3765, pruned_loss=0.1297, over 5650485.33 frames. ], batch size: 145, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:09:50,997 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 09:09:59,370 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2196, simple_loss=0.3266, pruned_loss=0.05631, over 944034.00 frames. 
+2023-03-05 09:09:59,370 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 09:10:47,253 INFO [train.py:968] (0/2) Epoch 10, batch 30050, giga_loss[loss=0.2835, simple_loss=0.3445, pruned_loss=0.1112, over 28591.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3749, pruned_loss=0.1292, over 5660865.61 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3773, pruned_loss=0.1279, over 5701078.32 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3753, pruned_loss=0.1297, over 5658699.90 frames. ], batch size: 92, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:10:52,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0277, 3.2079, 2.2923, 1.1057], device='cuda:0'), covar=tensor([0.5081, 0.1943, 0.2547, 0.4824], device='cuda:0'), in_proj_covar=tensor([0.1547, 0.1474, 0.1487, 0.1274], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 09:11:20,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.011e+02 1.587e+03 2.163e+03 2.961e+03 1.021e+04, threshold=4.326e+03, percent-clipped=8.0
+2023-03-05 09:11:28,947 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=439686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:11:35,877 INFO [train.py:968] (0/2) Epoch 10, batch 30100, giga_loss[loss=0.314, simple_loss=0.3698, pruned_loss=0.1291, over 28636.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3736, pruned_loss=0.1291, over 5640683.85 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3777, pruned_loss=0.1282, over 5700307.66 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3736, pruned_loss=0.1292, over 5638788.23 frames. ], batch size: 71, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:11:49,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 09:12:19,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0172, 1.3588, 1.0447, 0.1881], device='cuda:0'), covar=tensor([0.2132, 0.1982, 0.2930, 0.3730], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1475, 0.1488, 0.1276], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 09:12:25,523 INFO [train.py:968] (0/2) Epoch 10, batch 30150, giga_loss[loss=0.2705, simple_loss=0.3535, pruned_loss=0.09375, over 28477.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3727, pruned_loss=0.1262, over 5651549.73 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3774, pruned_loss=0.128, over 5702314.18 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3728, pruned_loss=0.1265, over 5647641.89 frames. ], batch size: 336, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:12:58,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=439775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:13:02,311 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-05 09:13:02,460 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.005e+02 1.547e+03 1.962e+03 2.851e+03 7.075e+03, threshold=3.923e+03, percent-clipped=5.0
+2023-03-05 09:13:02,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4784, 1.9727, 1.5988, 1.3929], device='cuda:0'), covar=tensor([0.1914, 0.1237, 0.1432, 0.1572], device='cuda:0'), in_proj_covar=tensor([0.1663, 0.1587, 0.1542, 0.1652], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 09:13:21,500 INFO [train.py:968] (0/2) Epoch 10, batch 30200, giga_loss[loss=0.2788, simple_loss=0.3554, pruned_loss=0.1011, over 28338.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3696, pruned_loss=0.1223, over 5636534.62 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3776, pruned_loss=0.1283, over 5700181.11 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3694, pruned_loss=0.1222, over 5634439.15 frames. ], batch size: 368, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:14:01,557 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0786, 1.5119, 1.3747, 0.9887], device='cuda:0'), covar=tensor([0.1312, 0.2093, 0.1130, 0.1443], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0704, 0.0854, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 09:14:10,896 INFO [train.py:968] (0/2) Epoch 10, batch 30250, giga_loss[loss=0.3133, simple_loss=0.3695, pruned_loss=0.1286, over 26725.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3669, pruned_loss=0.1189, over 5645262.53 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3768, pruned_loss=0.1279, over 5697782.28 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3672, pruned_loss=0.119, over 5643808.19 frames. ], batch size: 555, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:14:40,445 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=439873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:14:45,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.082e+02 1.328e+03 1.703e+03 2.551e+03 5.899e+03, threshold=3.405e+03, percent-clipped=4.0
+2023-03-05 09:14:58,617 INFO [train.py:968] (0/2) Epoch 10, batch 30300, giga_loss[loss=0.2733, simple_loss=0.3488, pruned_loss=0.09888, over 28034.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3638, pruned_loss=0.1163, over 5642801.87 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3752, pruned_loss=0.1273, over 5694537.93 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.365, pruned_loss=0.1165, over 5642748.38 frames. ], batch size: 412, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:15:16,779 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=439918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:15:18,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=439921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:15:34,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2973, 1.4845, 1.3047, 1.5373], device='cuda:0'), covar=tensor([0.0743, 0.0354, 0.0347, 0.0784], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0117, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 09:15:41,898 INFO [train.py:968] (0/2) Epoch 10, batch 30350, giga_loss[loss=0.2827, simple_loss=0.3563, pruned_loss=0.1045, over 28954.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1144, over 5652105.13 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3744, pruned_loss=0.1272, over 5700402.36 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3626, pruned_loss=0.1141, over 5644746.29 frames. ], batch size: 227, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:15:47,723 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:16:13,299 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 09:16:13,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.021e+02 1.538e+03 2.172e+03 3.259e+03 1.166e+04, threshold=4.343e+03, percent-clipped=18.0
+2023-03-05 09:16:21,375 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=439988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:16:28,433 INFO [train.py:968] (0/2) Epoch 10, batch 30400, giga_loss[loss=0.2541, simple_loss=0.3408, pruned_loss=0.08364, over 28709.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3609, pruned_loss=0.1118, over 5663481.54 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.374, pruned_loss=0.1272, over 5696884.08 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3618, pruned_loss=0.1111, over 5660078.38 frames. ], batch size: 242, lr: 3.21e-03, grad_scale: 8.0
+2023-03-05 09:16:34,872 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-440000.pt
+2023-03-05 09:17:24,161 INFO [train.py:968] (0/2) Epoch 10, batch 30450, giga_loss[loss=0.2881, simple_loss=0.3604, pruned_loss=0.1079, over 28296.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3598, pruned_loss=0.1098, over 5664283.43 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3737, pruned_loss=0.127, over 5697886.66 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3606, pruned_loss=0.1093, over 5660611.09 frames. ], batch size: 368, lr: 3.21e-03, grad_scale: 8.0
+2023-03-05 09:17:32,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=440052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:17:42,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:17:48,802 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=440068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:17:59,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.176e+02 1.264e+03 1.564e+03 2.249e+03 4.095e+03, threshold=3.128e+03, percent-clipped=0.0
+2023-03-05 09:18:07,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1211, 1.2087, 3.1946, 2.8960], device='cuda:0'), covar=tensor([0.1467, 0.2472, 0.0446, 0.1719], device='cuda:0'), in_proj_covar=tensor([0.0649, 0.0579, 0.0848, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 09:18:13,212 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=440092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:18:14,387 INFO [train.py:968] (0/2) Epoch 10, batch 30500, giga_loss[loss=0.2773, simple_loss=0.3487, pruned_loss=0.1029, over 28819.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3578, pruned_loss=0.1081, over 5658600.74 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.373, pruned_loss=0.1267, over 5689748.52 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3588, pruned_loss=0.1076, over 5662212.76 frames. ], batch size: 199, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:19:02,187 INFO [train.py:968] (0/2) Epoch 10, batch 30550, giga_loss[loss=0.2861, simple_loss=0.3502, pruned_loss=0.111, over 27936.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3546, pruned_loss=0.1061, over 5654148.26 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3725, pruned_loss=0.1267, over 5685813.15 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3553, pruned_loss=0.1049, over 5660017.32 frames. ], batch size: 412, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:19:10,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3672, 1.8316, 1.5301, 1.4838], device='cuda:0'), covar=tensor([0.1144, 0.1158, 0.1265, 0.1435], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0722, 0.0655, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 09:19:17,706 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3219, 1.6977, 1.6044, 1.1696], device='cuda:0'), covar=tensor([0.1622, 0.2378, 0.1384, 0.1569], device='cuda:0'), in_proj_covar=tensor([0.0817, 0.0695, 0.0848, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 09:19:32,878 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-05 09:19:39,930 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.776e+02 1.206e+03 1.628e+03 2.223e+03 2.514e+04, threshold=3.255e+03, percent-clipped=14.0
+2023-03-05 09:19:55,729 INFO [train.py:968] (0/2) Epoch 10, batch 30600, giga_loss[loss=0.2642, simple_loss=0.3452, pruned_loss=0.09165, over 28571.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3523, pruned_loss=0.1048, over 5646645.93 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3721, pruned_loss=0.1265, over 5685540.33 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3529, pruned_loss=0.1037, over 5651375.09 frames. ], batch size: 336, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:19:55,993 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2304, 4.0592, 3.8339, 1.7617], device='cuda:0'), covar=tensor([0.0596, 0.0766, 0.0817, 0.2084], device='cuda:0'), in_proj_covar=tensor([0.1019, 0.0964, 0.0837, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 09:20:04,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:20:07,801 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:20:35,976 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:20:38,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8100, 2.6095, 1.6891, 1.0625], device='cuda:0'), covar=tensor([0.5101, 0.2751, 0.3040, 0.4275], device='cuda:0'), in_proj_covar=tensor([0.1550, 0.1469, 0.1481, 0.1270], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 09:20:44,267 INFO [train.py:968] (0/2) Epoch 10, batch 30650, giga_loss[loss=0.2686, simple_loss=0.3456, pruned_loss=0.09586, over 28936.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3514, pruned_loss=0.1034, over 5656327.65 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3718, pruned_loss=0.1264, over 5687743.49 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.352, pruned_loss=0.1025, over 5657947.77 frames. ], batch size: 213, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:20:48,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:21:20,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.454e+02 1.195e+03 1.548e+03 2.085e+03 8.702e+03, threshold=3.095e+03, percent-clipped=2.0
+2023-03-05 09:21:33,143 INFO [train.py:968] (0/2) Epoch 10, batch 30700, giga_loss[loss=0.2509, simple_loss=0.3278, pruned_loss=0.08704, over 28955.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3508, pruned_loss=0.1028, over 5657894.15 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3714, pruned_loss=0.1263, over 5691083.39 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3512, pruned_loss=0.1016, over 5655602.78 frames. ], batch size: 213, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:22:25,418 INFO [train.py:968] (0/2) Epoch 10, batch 30750, giga_loss[loss=0.2738, simple_loss=0.3345, pruned_loss=0.1066, over 27714.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3483, pruned_loss=0.1009, over 5652322.96 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.371, pruned_loss=0.1264, over 5683693.82 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3486, pruned_loss=0.09962, over 5655591.34 frames. ], batch size: 472, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:22:44,972 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:23:01,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8313, 2.6006, 2.1431, 1.7334], device='cuda:0'), covar=tensor([0.1806, 0.0938, 0.0973, 0.1364], device='cuda:0'), in_proj_covar=tensor([0.1626, 0.1522, 0.1487, 0.1596], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 09:23:02,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.084e+02 1.328e+03 1.754e+03 2.593e+03 7.891e+03, threshold=3.508e+03, percent-clipped=16.0
+2023-03-05 09:23:14,188 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:23:16,370 INFO [train.py:968] (0/2) Epoch 10, batch 30800, giga_loss[loss=0.2514, simple_loss=0.317, pruned_loss=0.09295, over 26746.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3449, pruned_loss=0.09887, over 5662697.39 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3705, pruned_loss=0.1261, over 5683418.94 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.345, pruned_loss=0.09746, over 5664921.72 frames. ], batch size: 555, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:23:16,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:23:21,872 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-05 09:23:28,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4907, 1.9146, 1.6453, 1.5221], device='cuda:0'), covar=tensor([0.1735, 0.1251, 0.1259, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.1624, 0.1522, 0.1486, 0.1595], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 09:23:38,361 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 09:23:46,812 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:23:49,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:24:05,194 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:24:05,715 INFO [train.py:968] (0/2) Epoch 10, batch 30850, giga_loss[loss=0.2974, simple_loss=0.3662, pruned_loss=0.1142, over 28908.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3432, pruned_loss=0.09828, over 5667801.79 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3701, pruned_loss=0.1259, over 5688731.15 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3431, pruned_loss=0.09668, over 5664462.53 frames. ], batch size: 227, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:24:31,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:24:44,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5028, 1.9702, 1.7155, 1.4974], device='cuda:0'), covar=tensor([0.1787, 0.1260, 0.1305, 0.1424], device='cuda:0'), in_proj_covar=tensor([0.1634, 0.1528, 0.1495, 0.1600], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 09:24:44,475 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.968e+02 1.385e+03 1.915e+03 2.697e+03 1.358e+04, threshold=3.830e+03, percent-clipped=15.0
+2023-03-05 09:24:56,213 INFO [train.py:968] (0/2) Epoch 10, batch 30900, giga_loss[loss=0.2331, simple_loss=0.3155, pruned_loss=0.07534, over 28458.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3431, pruned_loss=0.09895, over 5650311.89 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3696, pruned_loss=0.1257, over 5681875.00 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3429, pruned_loss=0.09732, over 5653292.49 frames. ], batch size: 71, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:25:11,124 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:25:14,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:25:38,287 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0907, 1.9378, 1.3977, 1.5820], device='cuda:0'), covar=tensor([0.0573, 0.0480, 0.0876, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0442, 0.0503, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 09:25:45,798 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:25:50,764 INFO [train.py:968] (0/2) Epoch 10, batch 30950, giga_loss[loss=0.3054, simple_loss=0.3796, pruned_loss=0.1156, over 28823.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3439, pruned_loss=0.09954, over 5646972.02 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.369, pruned_loss=0.1254, over 5686299.65 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3437, pruned_loss=0.09799, over 5644824.07 frames. ], batch size: 186, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:26:25,240 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:26:29,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:26:37,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.680e+02 1.390e+03 1.681e+03 2.342e+03 7.242e+03, threshold=3.362e+03, percent-clipped=2.0
+2023-03-05 09:26:41,597 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:26:43,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:26:47,968 INFO [train.py:968] (0/2) Epoch 10, batch 31000, giga_loss[loss=0.2736, simple_loss=0.3527, pruned_loss=0.09724, over 28001.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.346, pruned_loss=0.0995, over 5643194.09 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3691, pruned_loss=0.1254, over 5686357.86 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3452, pruned_loss=0.0976, over 5640547.45 frames. ], batch size: 412, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:26:59,844 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:27:09,230 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:27:09,435 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 09:27:11,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:27:17,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:27:46,291 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:27:47,961 INFO [train.py:968] (0/2) Epoch 10, batch 31050, giga_loss[loss=0.3033, simple_loss=0.3625, pruned_loss=0.1221, over 27679.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3461, pruned_loss=0.0999, over 5637416.89 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3686, pruned_loss=0.1252, over 5690830.81 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3452, pruned_loss=0.09772, over 5629846.00 frames. ], batch size: 472, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:28:38,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.184e+02 1.321e+03 1.602e+03 2.265e+03 6.593e+03, threshold=3.204e+03, percent-clipped=9.0
+2023-03-05 09:28:46,944 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 09:28:54,361 INFO [train.py:968] (0/2) Epoch 10, batch 31100, giga_loss[loss=0.2307, simple_loss=0.2951, pruned_loss=0.08313, over 24415.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3453, pruned_loss=0.09949, over 5642330.40 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3681, pruned_loss=0.125, over 5695380.00 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3446, pruned_loss=0.0975, over 5631660.72 frames. ], batch size: 705, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:29:45,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2149, 1.4420, 1.2537, 1.1669], device='cuda:0'), covar=tensor([0.1538, 0.1375, 0.1081, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.1621, 0.1525, 0.1481, 0.1598], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 09:29:59,144 INFO [train.py:968] (0/2) Epoch 10, batch 31150, giga_loss[loss=0.2528, simple_loss=0.3399, pruned_loss=0.08283, over 28060.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3432, pruned_loss=0.09686, over 5640974.13 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3682, pruned_loss=0.1251, over 5686816.10 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3425, pruned_loss=0.09512, over 5639636.42 frames. ], batch size: 412, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:30:49,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.632e+02 1.221e+03 1.618e+03 2.500e+03 9.362e+03, threshold=3.235e+03, percent-clipped=10.0
+2023-03-05 09:31:02,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5724, 1.8482, 1.6521, 1.5955], device='cuda:0'), covar=tensor([0.1428, 0.2098, 0.1791, 0.1890], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0713, 0.0650, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 09:31:03,188 INFO [train.py:968] (0/2) Epoch 10, batch 31200, giga_loss[loss=0.2321, simple_loss=0.3195, pruned_loss=0.0724, over 28368.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3406, pruned_loss=0.09377, over 5632701.99 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3683, pruned_loss=0.1253, over 5680439.80 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3397, pruned_loss=0.09191, over 5636321.73 frames. ], batch size: 368, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:32:04,112 INFO [train.py:968] (0/2) Epoch 10, batch 31250, giga_loss[loss=0.2589, simple_loss=0.325, pruned_loss=0.09643, over 28461.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3385, pruned_loss=0.09378, over 5643290.65 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3677, pruned_loss=0.1251, over 5682312.98 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3375, pruned_loss=0.09166, over 5643350.30 frames. ], batch size: 85, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:32:19,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0008, 1.9463, 1.4295, 1.6048], device='cuda:0'), covar=tensor([0.0631, 0.0548, 0.0879, 0.0977], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0438, 0.0498, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 09:32:46,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.097e+02 1.363e+03 1.790e+03 2.464e+03 4.151e+03, threshold=3.581e+03, percent-clipped=9.0
+2023-03-05 09:32:59,475 INFO [train.py:968] (0/2) Epoch 10, batch 31300, giga_loss[loss=0.2696, simple_loss=0.3425, pruned_loss=0.09836, over 28954.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3384, pruned_loss=0.09417, over 5650454.97 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3674, pruned_loss=0.1249, over 5678466.88 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.337, pruned_loss=0.09165, over 5653410.21 frames. ], batch size: 155, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:33:54,281 INFO [train.py:968] (0/2) Epoch 10, batch 31350, giga_loss[loss=0.2662, simple_loss=0.3356, pruned_loss=0.09837, over 26821.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3392, pruned_loss=0.09479, over 5664950.14 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3675, pruned_loss=0.1252, over 5685084.08 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3371, pruned_loss=0.09166, over 5660617.60 frames. ], batch size: 555, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:34:00,208 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 09:34:15,364 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1480, 2.6312, 1.1918, 1.3333], device='cuda:0'), covar=tensor([0.0931, 0.0304, 0.0940, 0.1361], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0507, 0.0338, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 09:34:24,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3130, 1.5757, 1.2870, 1.2744], device='cuda:0'), covar=tensor([0.2068, 0.1970, 0.2080, 0.1882], device='cuda:0'), in_proj_covar=tensor([0.1277, 0.0944, 0.1133, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 09:34:40,686 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.087e+02 1.338e+03 1.861e+03 2.823e+03 6.263e+03, threshold=3.723e+03, percent-clipped=11.0
+2023-03-05 09:34:55,748 INFO [train.py:968] (0/2) Epoch 10, batch 31400, giga_loss[loss=0.2515, simple_loss=0.3176, pruned_loss=0.09269, over 24342.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3407, pruned_loss=0.09507, over 5662478.30 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3676, pruned_loss=0.1252, over 5687022.10 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3385, pruned_loss=0.09201, over 5656719.78 frames. ], batch size: 705, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:35:33,032 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.48 vs. limit=5.0
+2023-03-05 09:35:58,607 INFO [train.py:968] (0/2) Epoch 10, batch 31450, libri_loss[loss=0.274, simple_loss=0.3456, pruned_loss=0.1012, over 29533.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3409, pruned_loss=0.09502, over 5654008.65 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3666, pruned_loss=0.1247, over 5684093.88 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3392, pruned_loss=0.09211, over 5650857.78 frames. ], batch size: 82, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:36:47,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.779e+02 1.197e+03 1.768e+03 2.313e+03 4.574e+03, threshold=3.536e+03, percent-clipped=4.0
+2023-03-05 09:37:04,691 INFO [train.py:968] (0/2) Epoch 10, batch 31500, giga_loss[loss=0.245, simple_loss=0.3259, pruned_loss=0.082, over 28994.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.337, pruned_loss=0.09202, over 5673731.17 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3664, pruned_loss=0.1245, over 5687586.54 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3355, pruned_loss=0.08939, over 5667824.91 frames. ], batch size: 199, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:38:13,891 INFO [train.py:968] (0/2) Epoch 10, batch 31550, giga_loss[loss=0.2724, simple_loss=0.3512, pruned_loss=0.0968, over 28984.00 frames. ], tot_loss[loss=0.263, simple_loss=0.339, pruned_loss=0.09348, over 5671207.60 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3664, pruned_loss=0.1246, over 5690805.01 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3374, pruned_loss=0.09097, over 5663455.72 frames. ], batch size: 128, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:38:14,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5557, 1.6810, 1.2330, 1.2755], device='cuda:0'), covar=tensor([0.0787, 0.0492, 0.0948, 0.1006], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0437, 0.0495, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 09:38:35,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1228, 1.1784, 3.7368, 2.9260], device='cuda:0'), covar=tensor([0.1650, 0.2505, 0.0385, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0642, 0.0577, 0.0841, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 09:39:01,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.213e+02 1.375e+03 1.776e+03 2.403e+03 5.256e+03, threshold=3.552e+03, percent-clipped=4.0
+2023-03-05 09:39:16,633 INFO [train.py:968] (0/2) Epoch 10, batch 31600, giga_loss[loss=0.2608, simple_loss=0.3378, pruned_loss=0.09196, over 26812.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3429, pruned_loss=0.09375, over 5674130.63 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3661, pruned_loss=0.1246, over 5693998.90 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3415, pruned_loss=0.09128, over 5664757.48 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 09:40:24,937 INFO [train.py:968] (0/2) Epoch 10, batch 31650, giga_loss[loss=0.2542, simple_loss=0.3405, pruned_loss=0.08398, over 28052.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3429, pruned_loss=0.09191, over 5664603.35 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3655, pruned_loss=0.1242, over 5697415.35 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3421, pruned_loss=0.08985, over 5653665.90 frames. ], batch size: 412, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 09:40:27,121 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 09:40:47,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5610, 1.8091, 1.4608, 1.7515], device='cuda:0'), covar=tensor([0.2375, 0.2247, 0.2472, 0.2079], device='cuda:0'), in_proj_covar=tensor([0.1278, 0.0944, 0.1135, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 09:41:07,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.666e+02 1.264e+03 1.909e+03 2.610e+03 5.380e+03, threshold=3.818e+03, percent-clipped=5.0
+2023-03-05 09:41:17,974 INFO [train.py:968] (0/2) Epoch 10, batch 31700, giga_loss[loss=0.2515, simple_loss=0.3398, pruned_loss=0.08163, over 28063.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3441, pruned_loss=0.0916, over 5663626.51 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3653, pruned_loss=0.1244, over 5691555.94 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3429, pruned_loss=0.08875, over 5659633.10 frames. ], batch size: 412, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:42:20,821 INFO [train.py:968] (0/2) Epoch 10, batch 31750, giga_loss[loss=0.2639, simple_loss=0.3386, pruned_loss=0.09458, over 28812.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3435, pruned_loss=0.09113, over 5662268.08 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3649, pruned_loss=0.1242, over 5684094.42 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3425, pruned_loss=0.08846, over 5665414.24 frames. ], batch size: 99, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:43:04,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.901e+02 1.288e+03 1.743e+03 2.634e+03 9.477e+03, threshold=3.486e+03, percent-clipped=5.0
+2023-03-05 09:43:18,974 INFO [train.py:968] (0/2) Epoch 10, batch 31800, giga_loss[loss=0.2597, simple_loss=0.337, pruned_loss=0.09117, over 28749.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.343, pruned_loss=0.09218, over 5667451.62 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3643, pruned_loss=0.1239, over 5678311.74 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08917, over 5675084.41 frames. ], batch size: 243, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:43:59,492 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=441425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:44:26,899 INFO [train.py:968] (0/2) Epoch 10, batch 31850, giga_loss[loss=0.2698, simple_loss=0.3492, pruned_loss=0.09521, over 28160.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3426, pruned_loss=0.0935, over 5671747.07 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3631, pruned_loss=0.1231, over 5685471.87 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.342, pruned_loss=0.09047, over 5671507.93 frames. ], batch size: 412, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:45:27,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.874e+02 1.291e+03 1.753e+03 2.363e+03 8.386e+03, threshold=3.506e+03, percent-clipped=8.0
+2023-03-05 09:45:32,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6635, 1.6674, 1.2071, 1.3038], device='cuda:0'), covar=tensor([0.0737, 0.0607, 0.0982, 0.1079], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0438, 0.0495, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 09:45:43,666 INFO [train.py:968] (0/2) Epoch 10, batch 31900, giga_loss[loss=0.2349, simple_loss=0.3191, pruned_loss=0.07538, over 28725.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3426, pruned_loss=0.09417, over 5678559.48 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3628, pruned_loss=0.123, over 5690763.17 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3419, pruned_loss=0.09121, over 5673314.10 frames. ], batch size: 307, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:46:00,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=441507.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 09:46:29,765 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 09:46:51,763 INFO [train.py:968] (0/2) Epoch 10, batch 31950, libri_loss[loss=0.2606, simple_loss=0.3258, pruned_loss=0.09769, over 29550.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.338, pruned_loss=0.09146, over 5682375.25 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3621, pruned_loss=0.1226, over 5697229.92 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3374, pruned_loss=0.08858, over 5671916.99 frames. ], batch size: 77, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:47:38,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.976e+02 1.143e+03 1.483e+03 2.082e+03 6.667e+03, threshold=2.967e+03, percent-clipped=8.0
+2023-03-05 09:47:53,684 INFO [train.py:968] (0/2) Epoch 10, batch 32000, giga_loss[loss=0.2199, simple_loss=0.3022, pruned_loss=0.06882, over 28959.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.336, pruned_loss=0.09067, over 5684409.44 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3612, pruned_loss=0.1222, over 5704032.21 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3357, pruned_loss=0.08779, over 5669730.37 frames. ], batch size: 106, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 09:48:27,109 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4763, 1.7262, 1.3789, 1.7616], device='cuda:0'), covar=tensor([0.2495, 0.2275, 0.2586, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.1273, 0.0938, 0.1131, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 09:49:01,294 INFO [train.py:968] (0/2) Epoch 10, batch 32050, giga_loss[loss=0.204, simple_loss=0.2858, pruned_loss=0.06111, over 28584.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3352, pruned_loss=0.09085, over 5689430.57 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3615, pruned_loss=0.1224, over 5706309.32 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3341, pruned_loss=0.08776, over 5675524.63 frames. ], batch size: 78, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:49:48,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5721, 4.3857, 4.1682, 2.3132], device='cuda:0'), covar=tensor([0.0511, 0.0670, 0.0751, 0.1620], device='cuda:0'), in_proj_covar=tensor([0.1011, 0.0946, 0.0830, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 09:49:53,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8311, 1.8802, 1.3244, 1.5617], device='cuda:0'), covar=tensor([0.0744, 0.0573, 0.0967, 0.0921], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0438, 0.0496, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 09:49:53,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.545e+02 1.355e+03 1.722e+03 2.273e+03 4.785e+03, threshold=3.444e+03, percent-clipped=8.0
+2023-03-05 09:49:55,983 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=441684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:50:05,870 INFO [train.py:968] (0/2) Epoch 10, batch 32100, giga_loss[loss=0.2775, simple_loss=0.36, pruned_loss=0.09749, over 28939.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3395, pruned_loss=0.09229, over 5688551.04 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3612, pruned_loss=0.1222, over 5706875.14 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3388, pruned_loss=0.08973, over 5676943.15 frames. ], batch size: 145, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:51:10,424 INFO [train.py:968] (0/2) Epoch 10, batch 32150, giga_loss[loss=0.2869, simple_loss=0.3471, pruned_loss=0.1133, over 26879.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3383, pruned_loss=0.0927, over 5691921.93 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3608, pruned_loss=0.122, over 5709106.85 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3376, pruned_loss=0.09038, over 5680280.13 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:51:58,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.321e+02 1.253e+03 1.867e+03 2.743e+03 5.887e+03, threshold=3.734e+03, percent-clipped=11.0
+2023-03-05 09:52:09,221 INFO [train.py:968] (0/2) Epoch 10, batch 32200, giga_loss[loss=0.3597, simple_loss=0.4082, pruned_loss=0.1556, over 28147.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3383, pruned_loss=0.09365, over 5693700.69 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3603, pruned_loss=0.1217, over 5713877.01 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3378, pruned_loss=0.09137, over 5679760.78 frames. ], batch size: 412, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:52:17,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=441800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:52:38,341 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=441818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:52:39,626 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=441819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:53:09,007 INFO [train.py:968] (0/2) Epoch 10, batch 32250, giga_loss[loss=0.2438, simple_loss=0.3233, pruned_loss=0.08214, over 28604.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3393, pruned_loss=0.09474, over 5696303.52 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3606, pruned_loss=0.122, over 5719413.27 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3379, pruned_loss=0.09185, over 5679498.24 frames. ], batch size: 85, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:54:07,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=441882.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 09:54:09,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.243e+02 1.430e+03 1.810e+03 2.550e+03 9.566e+03, threshold=3.620e+03, percent-clipped=9.0
+2023-03-05 09:54:20,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3823, 1.6501, 1.6813, 1.2725], device='cuda:0'), covar=tensor([0.1661, 0.2301, 0.1325, 0.1553], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0690, 0.0848, 0.0761], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 09:54:25,425 INFO [train.py:968] (0/2) Epoch 10, batch 32300, giga_loss[loss=0.2982, simple_loss=0.3729, pruned_loss=0.1117, over 28397.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3412, pruned_loss=0.09459, over 5692781.31 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3608, pruned_loss=0.1221, over 5722113.68 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3397, pruned_loss=0.09178, over 5676675.45 frames. ], batch size: 368, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:55:01,002 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9071, 1.0816, 1.0187, 0.8582], device='cuda:0'), covar=tensor([0.1244, 0.1426, 0.0779, 0.1099], device='cuda:0'), in_proj_covar=tensor([0.1645, 0.1539, 0.1488, 0.1606], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 09:55:41,256 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=441943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:55:41,535 INFO [train.py:968] (0/2) Epoch 10, batch 32350, giga_loss[loss=0.2853, simple_loss=0.3481, pruned_loss=0.1113, over 26918.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3425, pruned_loss=0.09485, over 5682205.57 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.361, pruned_loss=0.1222, over 5723305.78 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3407, pruned_loss=0.09202, over 5667628.94 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:55:45,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=441946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:55:58,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6812, 1.0682, 2.8578, 2.6989], device='cuda:0'), covar=tensor([0.1615, 0.2390, 0.0529, 0.0982], device='cuda:0'), in_proj_covar=tensor([0.0638, 0.0574, 0.0835, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 09:56:01,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3838, 1.8268, 1.2719, 1.6871], device='cuda:0'), covar=tensor([0.2481, 0.2262, 0.2775, 0.2222], device='cuda:0'), in_proj_covar=tensor([0.1273, 0.0936, 0.1133, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 09:56:26,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=441975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:56:38,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.117e+02 1.429e+03 1.799e+03 2.543e+03 7.083e+03, threshold=3.599e+03, percent-clipped=10.0
+2023-03-05 09:56:52,940 INFO [train.py:968] (0/2) Epoch 10, batch 32400, giga_loss[loss=0.2569, simple_loss=0.3253, pruned_loss=0.09425, over 28823.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3396, pruned_loss=0.09376, over 5685154.93 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3602, pruned_loss=0.1217, over 5725793.88 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3386, pruned_loss=0.09146, over 5670622.05 frames. ], batch size: 119, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 09:57:01,387 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-442000.pt
+2023-03-05 09:57:30,728 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 09:57:36,682 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442025.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 09:57:39,923 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442028.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 09:57:58,889 INFO [train.py:968] (0/2) Epoch 10, batch 32450, giga_loss[loss=0.2519, simple_loss=0.3149, pruned_loss=0.09444, over 26799.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3339, pruned_loss=0.09178, over 5687374.35 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3599, pruned_loss=0.1216, over 5729182.88 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3329, pruned_loss=0.08957, over 5671998.97 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 09:58:08,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1336, 1.4461, 1.4125, 1.3793], device='cuda:0'), covar=tensor([0.1353, 0.1436, 0.1852, 0.1365], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0713, 0.0650, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 09:58:16,051 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442057.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 09:58:17,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:58:31,758 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-05 09:58:54,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.985e+02 1.473e+03 1.845e+03 2.696e+03 5.950e+03, threshold=3.690e+03, percent-clipped=7.0
+2023-03-05 09:59:07,114 INFO [train.py:968] (0/2) Epoch 10, batch 32500, giga_loss[loss=0.2829, simple_loss=0.3551, pruned_loss=0.1054, over 28748.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3317, pruned_loss=0.09116, over 5672948.23 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3596, pruned_loss=0.1215, over 5721691.88 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3308, pruned_loss=0.08913, over 5666365.94 frames. ], batch size: 262, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:59:29,022 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2353, 1.4731, 1.2635, 1.2874], device='cuda:0'), covar=tensor([0.1427, 0.1195, 0.1054, 0.1102], device='cuda:0'), in_proj_covar=tensor([0.1641, 0.1537, 0.1484, 0.1602], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 09:59:30,903 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:00:07,693 INFO [train.py:968] (0/2) Epoch 10, batch 32550, giga_loss[loss=0.2686, simple_loss=0.3464, pruned_loss=0.09542, over 28917.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3344, pruned_loss=0.0929, over 5677007.58 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3595, pruned_loss=0.1214, over 5722657.42 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3337, pruned_loss=0.09128, over 5670838.22 frames. ], batch size: 186, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:00:52,300 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.117e+02 1.682e+03 2.389e+03 3.673e+03 7.848e+03, threshold=4.778e+03, percent-clipped=23.0
+2023-03-05 10:01:02,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:01:02,775 INFO [train.py:968] (0/2) Epoch 10, batch 32600, giga_loss[loss=0.2445, simple_loss=0.3265, pruned_loss=0.0813, over 28330.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3351, pruned_loss=0.09344, over 5685059.70 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3591, pruned_loss=0.1212, over 5726548.05 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.334, pruned_loss=0.09129, over 5675030.90 frames. ], batch size: 368, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:01:03,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:01:11,883 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:01:16,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:01:49,912 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:02:00,201 INFO [train.py:968] (0/2) Epoch 10, batch 32650, giga_loss[loss=0.2418, simple_loss=0.3261, pruned_loss=0.07869, over 28861.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3329, pruned_loss=0.09146, over 5674917.94 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3592, pruned_loss=0.1213, over 5730148.15 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3312, pruned_loss=0.08876, over 5662370.43 frames. ], batch size: 284, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:02:51,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.861e+02 1.187e+03 1.546e+03 2.114e+03 8.260e+03, threshold=3.092e+03, percent-clipped=1.0
+2023-03-05 10:03:03,491 INFO [train.py:968] (0/2) Epoch 10, batch 32700, libri_loss[loss=0.2716, simple_loss=0.3287, pruned_loss=0.1073, over 29579.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3319, pruned_loss=0.09103, over 5679140.80 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3587, pruned_loss=0.121, over 5735204.65 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3301, pruned_loss=0.08825, over 5662716.62 frames. ], batch size: 75, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:03:11,111 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3034, 1.4657, 1.2481, 1.5759], device='cuda:0'), covar=tensor([0.0760, 0.0305, 0.0333, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0117, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0056, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 10:03:31,919 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:03:36,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5770, 1.5823, 1.2382, 1.2433], device='cuda:0'), covar=tensor([0.0768, 0.0502, 0.0962, 0.0927], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0433, 0.0493, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 10:04:02,175 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:03,515 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:05,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:06,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:10,422 INFO [train.py:968] (0/2) Epoch 10, batch 32750, giga_loss[loss=0.2474, simple_loss=0.3283, pruned_loss=0.08323, over 28539.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3301, pruned_loss=0.08977, over 5672032.83 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3587, pruned_loss=0.1211, over 5728007.42 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3284, pruned_loss=0.08709, over 5664192.16 frames. ], batch size: 370, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:04:43,262 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:44,825 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:05:06,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.432e+02 1.286e+03 1.703e+03 2.227e+03 4.369e+03, threshold=3.406e+03, percent-clipped=9.0
+2023-03-05 10:05:08,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4035, 1.5336, 1.0998, 1.2197], device='cuda:0'), covar=tensor([0.0716, 0.0485, 0.0981, 0.0960], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0434, 0.0494, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 10:05:16,936 INFO [train.py:968] (0/2) Epoch 10, batch 32800, giga_loss[loss=0.2883, simple_loss=0.3564, pruned_loss=0.1101, over 28630.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.332, pruned_loss=0.09011, over 5687864.91 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3588, pruned_loss=0.1211, over 5733220.53 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3298, pruned_loss=0.08702, over 5675428.99 frames. ], batch size: 307, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 10:06:20,167 INFO [train.py:968] (0/2) Epoch 10, batch 32850, giga_loss[loss=0.2902, simple_loss=0.3561, pruned_loss=0.1121, over 28624.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3325, pruned_loss=0.09059, over 5680461.65 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3585, pruned_loss=0.1209, over 5728742.22 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3306, pruned_loss=0.08795, over 5674260.40 frames. ], batch size: 307, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 10:07:11,897 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.716e+02 1.262e+03 1.488e+03 2.041e+03 5.964e+03, threshold=2.977e+03, percent-clipped=6.0
+2023-03-05 10:07:13,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:07:21,217 INFO [train.py:968] (0/2) Epoch 10, batch 32900, libri_loss[loss=0.3306, simple_loss=0.3833, pruned_loss=0.139, over 25973.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3322, pruned_loss=0.09084, over 5684029.59 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3588, pruned_loss=0.1213, over 5729156.28 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3298, pruned_loss=0.08772, over 5678037.49 frames. ], batch size: 136, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:08:05,645 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442531.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:08:21,033 INFO [train.py:968] (0/2) Epoch 10, batch 32950, giga_loss[loss=0.219, simple_loss=0.2909, pruned_loss=0.07349, over 24053.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3316, pruned_loss=0.08979, over 5668923.83 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3584, pruned_loss=0.1211, over 5730215.37 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3295, pruned_loss=0.08676, over 5662095.18 frames. ], batch size: 705, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:08:35,491 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 10:08:44,671 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:09:00,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3535, 1.7652, 1.4542, 1.4996], device='cuda:0'), covar=tensor([0.0722, 0.0326, 0.0317, 0.0792], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0117, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0056, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 10:09:09,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.576e+02 1.328e+03 1.818e+03 2.999e+03 1.435e+04, threshold=3.636e+03, percent-clipped=26.0
+2023-03-05 10:09:16,663 INFO [train.py:968] (0/2) Epoch 10, batch 33000, giga_loss[loss=0.2759, simple_loss=0.3554, pruned_loss=0.09818, over 28455.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3356, pruned_loss=0.09129, over 5660521.81 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3583, pruned_loss=0.1211, over 5721488.17 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3333, pruned_loss=0.08803, over 5661853.09 frames. ], batch size: 369, lr: 3.20e-03, grad_scale: 1.0
+2023-03-05 10:09:16,667 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 10:09:25,195 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2044, simple_loss=0.3044, pruned_loss=0.05216, over 944034.00 frames. 
+2023-03-05 10:09:25,196 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 10:10:07,130 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:10:11,320 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:10:13,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:10:24,939 INFO [train.py:968] (0/2) Epoch 10, batch 33050, libri_loss[loss=0.297, simple_loss=0.3547, pruned_loss=0.1196, over 29540.00 frames. ], tot_loss[loss=0.26, simple_loss=0.337, pruned_loss=0.09148, over 5662602.53 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3579, pruned_loss=0.1209, over 5724050.85 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3351, pruned_loss=0.08856, over 5660047.87 frames. ], batch size: 83, lr: 3.20e-03, grad_scale: 1.0
+2023-03-05 10:10:48,504 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:10:50,411 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:11:21,271 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.062e+02 1.450e+03 1.970e+03 3.015e+03 7.735e+03, threshold=3.940e+03, percent-clipped=16.0
+2023-03-05 10:11:22,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:11:24,572 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:11:27,732 INFO [train.py:968] (0/2) Epoch 10, batch 33100, giga_loss[loss=0.2692, simple_loss=0.3429, pruned_loss=0.09776, over 28197.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3385, pruned_loss=0.09269, over 5659223.34 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3579, pruned_loss=0.1211, over 5717442.40 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3366, pruned_loss=0.08949, over 5661940.25 frames. ], batch size: 412, lr: 3.20e-03, grad_scale: 1.0
+2023-03-05 10:12:26,075 INFO [train.py:968] (0/2) Epoch 10, batch 33150, libri_loss[loss=0.2735, simple_loss=0.3424, pruned_loss=0.1023, over 27777.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3388, pruned_loss=0.09326, over 5660465.48 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3578, pruned_loss=0.1212, over 5710750.20 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3368, pruned_loss=0.08974, over 5666652.71 frames. ], batch size: 115, lr: 3.20e-03, grad_scale: 1.0
+2023-03-05 10:13:14,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.714e+02 1.285e+03 1.596e+03 2.180e+03 5.356e+03, threshold=3.191e+03, percent-clipped=4.0
+2023-03-05 10:13:22,195 INFO [train.py:968] (0/2) Epoch 10, batch 33200, giga_loss[loss=0.277, simple_loss=0.3485, pruned_loss=0.1027, over 26972.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3351, pruned_loss=0.09083, over 5662144.01 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3576, pruned_loss=0.121, over 5706946.32 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3331, pruned_loss=0.08727, over 5669236.74 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:14:10,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:14:13,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:14:23,057 INFO [train.py:968] (0/2) Epoch 10, batch 33250, giga_loss[loss=0.228, simple_loss=0.301, pruned_loss=0.07748, over 28757.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3345, pruned_loss=0.09078, over 5661889.48 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3576, pruned_loss=0.1211, over 5701027.47 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3324, pruned_loss=0.08733, over 5672953.65 frames. ], batch size: 99, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:14:51,055 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:15:19,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.886e+02 1.239e+03 1.871e+03 3.079e+03 7.933e+03, threshold=3.742e+03, percent-clipped=22.0
+2023-03-05 10:15:26,498 INFO [train.py:968] (0/2) Epoch 10, batch 33300, giga_loss[loss=0.2648, simple_loss=0.346, pruned_loss=0.09179, over 28861.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3339, pruned_loss=0.09073, over 5659047.36 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.358, pruned_loss=0.1214, over 5699604.09 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3317, pruned_loss=0.08746, over 5668690.86 frames. ], batch size: 227, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:15:44,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442906.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:15:44,883 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-05 10:16:31,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:16:34,136 INFO [train.py:968] (0/2) Epoch 10, batch 33350, giga_loss[loss=0.3175, simple_loss=0.3855, pruned_loss=0.1248, over 28880.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3363, pruned_loss=0.09166, over 5663624.91 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.358, pruned_loss=0.1214, over 5701697.80 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3344, pruned_loss=0.08876, over 5668975.96 frames. ], batch size: 227, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:16:53,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1039, 1.0028, 1.0169, 1.3290], device='cuda:0'), covar=tensor([0.0822, 0.0347, 0.0312, 0.1019], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0118, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0056, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 10:17:29,556 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.267e+02 1.182e+03 1.478e+03 1.943e+03 3.965e+03, threshold=2.956e+03, percent-clipped=1.0
+2023-03-05 10:17:37,698 INFO [train.py:968] (0/2) Epoch 10, batch 33400, giga_loss[loss=0.2725, simple_loss=0.3434, pruned_loss=0.1008, over 28061.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3374, pruned_loss=0.09264, over 5663013.43 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3583, pruned_loss=0.1217, over 5695665.16 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3352, pruned_loss=0.08963, over 5671717.80 frames. ], batch size: 412, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:17:51,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=443005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:18:30,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=443036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:18:39,690 INFO [train.py:968] (0/2) Epoch 10, batch 33450, giga_loss[loss=0.3316, simple_loss=0.3824, pruned_loss=0.1404, over 26877.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3401, pruned_loss=0.0952, over 5645570.31 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3583, pruned_loss=0.1218, over 5687884.93 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3378, pruned_loss=0.09172, over 5657668.94 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:18:48,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443049.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:18:51,922 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443052.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:19:09,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=443066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:19:24,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443081.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:19:26,744 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:19:28,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:19:30,185 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.043e+02 1.355e+03 1.853e+03 2.525e+03 1.532e+04, threshold=3.705e+03, percent-clipped=18.0
+2023-03-05 10:19:37,237 INFO [train.py:968] (0/2) Epoch 10, batch 33500, giga_loss[loss=0.2429, simple_loss=0.3335, pruned_loss=0.0762, over 28835.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3433, pruned_loss=0.09619, over 5649377.53 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3584, pruned_loss=0.1219, over 5691167.96 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3411, pruned_loss=0.09295, over 5655422.44 frames. ], batch size: 145, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:19:43,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9337, 1.0953, 0.9834, 0.8734], device='cuda:0'), covar=tensor([0.1145, 0.1347, 0.0865, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.1637, 0.1518, 0.1484, 0.1589], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 10:20:02,679 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:20:39,042 INFO [train.py:968] (0/2) Epoch 10, batch 33550, giga_loss[loss=0.2842, simple_loss=0.3596, pruned_loss=0.1044, over 28901.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3451, pruned_loss=0.09703, over 5644106.33 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.358, pruned_loss=0.1217, over 5683659.46 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3433, pruned_loss=0.09395, over 5654585.36 frames. ], batch size: 227, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:20:45,731 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:20:51,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:21:30,409 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:21:31,523 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:21:33,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:21:41,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.404e+02 1.271e+03 1.790e+03 2.484e+03 1.179e+04, threshold=3.579e+03, percent-clipped=11.0
+2023-03-05 10:21:49,820 INFO [train.py:968] (0/2) Epoch 10, batch 33600, giga_loss[loss=0.2486, simple_loss=0.3232, pruned_loss=0.08706, over 27566.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3433, pruned_loss=0.09571, over 5661971.90 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3579, pruned_loss=0.1217, over 5687020.02 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3418, pruned_loss=0.09301, over 5666907.50 frames. ], batch size: 472, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:22:12,989 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:22:15,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:22:16,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:22:52,474 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:22:54,565 INFO [train.py:968] (0/2) Epoch 10, batch 33650, giga_loss[loss=0.2535, simple_loss=0.3289, pruned_loss=0.08904, over 28906.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3415, pruned_loss=0.09507, over 5673677.24 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3577, pruned_loss=0.1216, over 5688036.59 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3401, pruned_loss=0.0924, over 5676111.42 frames. ], batch size: 106, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:23:26,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2333, 1.8188, 1.3457, 0.3948], device='cuda:0'), covar=tensor([0.3030, 0.2005, 0.3180, 0.3977], device='cuda:0'), in_proj_covar=tensor([0.1547, 0.1483, 0.1481, 0.1269], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 10:23:30,773 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=443272.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:23:50,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.224e+02 1.540e+03 2.008e+03 2.957e+03 6.593e+03, threshold=4.016e+03, percent-clipped=14.0
+2023-03-05 10:23:58,838 INFO [train.py:968] (0/2) Epoch 10, batch 33700, giga_loss[loss=0.2379, simple_loss=0.3158, pruned_loss=0.07995, over 27908.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3413, pruned_loss=0.09533, over 5666725.27 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3578, pruned_loss=0.1216, over 5684038.30 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3398, pruned_loss=0.09257, over 5672219.62 frames. ], batch size: 412, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:24:39,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5247, 1.8102, 1.4218, 1.6207], device='cuda:0'), covar=tensor([0.2329, 0.2196, 0.2562, 0.2070], device='cuda:0'), in_proj_covar=tensor([0.1260, 0.0930, 0.1122, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 10:24:42,514 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-05 10:24:58,749 INFO [train.py:968] (0/2) Epoch 10, batch 33750, libri_loss[loss=0.2488, simple_loss=0.3152, pruned_loss=0.09118, over 29569.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3408, pruned_loss=0.09636, over 5671998.41 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3573, pruned_loss=0.1212, over 5690124.07 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3393, pruned_loss=0.09344, over 5669988.40 frames. ], batch size: 75, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:25:20,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7795, 1.4731, 5.3799, 3.4490], device='cuda:0'), covar=tensor([0.1501, 0.2368, 0.0364, 0.0974], device='cuda:0'), in_proj_covar=tensor([0.0640, 0.0572, 0.0831, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 10:25:33,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9679, 2.0096, 1.9078, 1.7636], device='cuda:0'), covar=tensor([0.1313, 0.2115, 0.1766, 0.1915], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0707, 0.0649, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 10:25:51,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.057e+02 1.369e+03 1.988e+03 2.582e+03 1.254e+04, threshold=3.977e+03, percent-clipped=11.0
+2023-03-05 10:25:58,359 INFO [train.py:968] (0/2) Epoch 10, batch 33800, giga_loss[loss=0.214, simple_loss=0.2955, pruned_loss=0.06628, over 29039.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.338, pruned_loss=0.09488, over 5685008.19 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3574, pruned_loss=0.1212, over 5694573.82 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3362, pruned_loss=0.09178, over 5678992.37 frames. ], batch size: 120, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:26:57,729 INFO [train.py:968] (0/2) Epoch 10, batch 33850, libri_loss[loss=0.2633, simple_loss=0.3167, pruned_loss=0.105, over 29643.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3384, pruned_loss=0.09392, over 5674187.65 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3571, pruned_loss=0.1211, over 5687163.76 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.337, pruned_loss=0.09125, over 5675651.02 frames. ], batch size: 69, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:27:49,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.616e+02 1.321e+03 1.773e+03 2.432e+03 5.436e+03, threshold=3.546e+03, percent-clipped=5.0
+2023-03-05 10:27:57,512 INFO [train.py:968] (0/2) Epoch 10, batch 33900, giga_loss[loss=0.2107, simple_loss=0.3014, pruned_loss=0.05998, over 28901.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3366, pruned_loss=0.09224, over 5673967.31 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3574, pruned_loss=0.1213, over 5694304.42 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3346, pruned_loss=0.08903, over 5668426.83 frames. ], batch size: 186, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:27:57,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=443494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:28:46,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4334, 1.8518, 1.6043, 1.2748], device='cuda:0'), covar=tensor([0.2555, 0.1378, 0.1198, 0.1867], device='cuda:0'), in_proj_covar=tensor([0.1643, 0.1519, 0.1477, 0.1594], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 10:28:49,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1685, 1.7294, 1.2476, 0.3149], device='cuda:0'), covar=tensor([0.2576, 0.1574, 0.2640, 0.3682], device='cuda:0'), in_proj_covar=tensor([0.1536, 0.1472, 0.1474, 0.1260], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 10:28:50,044 INFO [train.py:968] (0/2) Epoch 10, batch 33950, giga_loss[loss=0.2854, simple_loss=0.3635, pruned_loss=0.1037, over 28409.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3388, pruned_loss=0.09148, over 5679111.70 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.357, pruned_loss=0.121, over 5696519.80 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.337, pruned_loss=0.08837, over 5672033.78 frames. ], batch size: 368, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:29:01,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9917, 1.9809, 1.4235, 1.5524], device='cuda:0'), covar=tensor([0.0714, 0.0545, 0.0914, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0432, 0.0497, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 10:29:32,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.38 vs. limit=5.0
+2023-03-05 10:29:43,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.932e+02 1.268e+03 1.767e+03 2.281e+03 5.011e+03, threshold=3.535e+03, percent-clipped=5.0
+2023-03-05 10:29:48,008 INFO [train.py:968] (0/2) Epoch 10, batch 34000, giga_loss[loss=0.2495, simple_loss=0.3351, pruned_loss=0.08191, over 28757.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3393, pruned_loss=0.09048, over 5684807.47 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3568, pruned_loss=0.1208, over 5698680.30 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3379, pruned_loss=0.08795, over 5677193.33 frames. ], batch size: 262, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:29:59,193 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 10:30:44,978 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3104, 1.5682, 1.5341, 1.4355], device='cuda:0'), covar=tensor([0.1174, 0.1437, 0.1603, 0.1332], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0707, 0.0649, 0.0635], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 10:30:51,159 INFO [train.py:968] (0/2) Epoch 10, batch 34050, giga_loss[loss=0.2334, simple_loss=0.3235, pruned_loss=0.07162, over 28849.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3397, pruned_loss=0.09123, over 5676292.18 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3568, pruned_loss=0.1209, over 5695171.34 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3381, pruned_loss=0.08826, over 5672238.50 frames. ], batch size: 112, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:30:57,506 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=443647.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:31:08,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-05 10:31:56,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.648e+02 1.444e+03 1.961e+03 3.126e+03 1.021e+04, threshold=3.922e+03, percent-clipped=20.0
+2023-03-05 10:32:04,539 INFO [train.py:968] (0/2) Epoch 10, batch 34100, giga_loss[loss=0.2697, simple_loss=0.3377, pruned_loss=0.1008, over 26805.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3393, pruned_loss=0.09082, over 5667465.66 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3564, pruned_loss=0.1207, over 5696505.02 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3382, pruned_loss=0.0884, over 5662911.29 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:32:45,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5240, 4.2850, 1.6389, 1.7044], device='cuda:0'), covar=tensor([0.0894, 0.0234, 0.0876, 0.1252], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0500, 0.0338, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 10:33:11,214 INFO [train.py:968] (0/2) Epoch 10, batch 34150, giga_loss[loss=0.2408, simple_loss=0.3302, pruned_loss=0.07573, over 28954.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3406, pruned_loss=0.09144, over 5662448.99 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3566, pruned_loss=0.1206, over 5691389.36 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3392, pruned_loss=0.0889, over 5662148.67 frames. ], batch size: 106, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:34:19,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.556e+02 1.255e+03 1.707e+03 2.309e+03 6.773e+03, threshold=3.414e+03, percent-clipped=4.0
+2023-03-05 10:34:20,061 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443790.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:34:22,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443793.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:34:22,470 INFO [train.py:968] (0/2) Epoch 10, batch 34200, giga_loss[loss=0.2809, simple_loss=0.3576, pruned_loss=0.1021, over 28716.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3405, pruned_loss=0.09084, over 5665234.79 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3564, pruned_loss=0.1205, over 5692906.18 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3392, pruned_loss=0.08823, over 5662761.31 frames. ], batch size: 307, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:34:59,725 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443822.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:35:26,942 INFO [train.py:968] (0/2) Epoch 10, batch 34250, giga_loss[loss=0.2926, simple_loss=0.3727, pruned_loss=0.1063, over 28916.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3431, pruned_loss=0.0919, over 5668688.86 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3561, pruned_loss=0.1203, over 5694845.60 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3422, pruned_loss=0.08976, over 5664754.59 frames. ], batch size: 186, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:35:57,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=443869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:36:05,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5296, 2.9308, 1.5397, 1.5931], device='cuda:0'), covar=tensor([0.0745, 0.0363, 0.0775, 0.1110], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0497, 0.0336, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 10:36:20,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.138e+02 1.326e+03 1.819e+03 2.776e+03 5.229e+03, threshold=3.638e+03, percent-clipped=15.0
+2023-03-05 10:36:25,365 INFO [train.py:968] (0/2) Epoch 10, batch 34300, giga_loss[loss=0.2444, simple_loss=0.3265, pruned_loss=0.08114, over 28991.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3446, pruned_loss=0.09302, over 5671295.13 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.356, pruned_loss=0.1203, over 5695274.39 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3434, pruned_loss=0.09015, over 5667047.28 frames. ], batch size: 186, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:36:34,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3439, 3.3910, 1.4693, 1.4673], device='cuda:0'), covar=tensor([0.0904, 0.0295, 0.0883, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0498, 0.0336, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 10:37:33,866 INFO [train.py:968] (0/2) Epoch 10, batch 34350, giga_loss[loss=0.2575, simple_loss=0.336, pruned_loss=0.0895, over 28992.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3427, pruned_loss=0.0926, over 5671091.56 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3564, pruned_loss=0.1204, over 5695922.47 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3412, pruned_loss=0.08966, over 5666428.68 frames. ], batch size: 199, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:37:53,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3641, 2.0961, 1.4576, 0.4897], device='cuda:0'), covar=tensor([0.3287, 0.1868, 0.3280, 0.4435], device='cuda:0'), in_proj_covar=tensor([0.1534, 0.1476, 0.1482, 0.1262], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 10:38:19,006 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=443986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:38:20,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2779, 3.1202, 1.3795, 1.3997], device='cuda:0'), covar=tensor([0.0914, 0.0336, 0.0900, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0499, 0.0337, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 10:38:22,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.592e+02 1.445e+03 2.037e+03 2.956e+03 9.500e+03, threshold=4.074e+03, percent-clipped=12.0
+2023-03-05 10:38:28,135 INFO [train.py:968] (0/2) Epoch 10, batch 34400, libri_loss[loss=0.2998, simple_loss=0.3571, pruned_loss=0.1213, over 29569.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3408, pruned_loss=0.0925, over 5693714.92 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.356, pruned_loss=0.1202, over 5704769.71 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3391, pruned_loss=0.08871, over 5680885.36 frames. ], batch size: 89, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:38:33,574 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-444000.pt
+2023-03-05 10:38:52,685 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=444012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:38:55,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=444015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:39:35,241 INFO [train.py:968] (0/2) Epoch 10, batch 34450, libri_loss[loss=0.2483, simple_loss=0.3069, pruned_loss=0.09484, over 29359.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.338, pruned_loss=0.09025, over 5690730.78 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3552, pruned_loss=0.1197, over 5710453.09 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.337, pruned_loss=0.08691, over 5675271.74 frames. ], batch size: 67, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:39:35,592 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=444044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:39:46,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 10:39:58,727 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-05 10:40:26,517 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:40:28,566 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444089.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:40:28,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.955e+02 1.173e+03 1.552e+03 2.253e+03 4.862e+03, threshold=3.105e+03, percent-clipped=0.0
+2023-03-05 10:40:35,024 INFO [train.py:968] (0/2) Epoch 10, batch 34500, giga_loss[loss=0.2485, simple_loss=0.3171, pruned_loss=0.08997, over 24085.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3365, pruned_loss=0.08941, over 5683466.41 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3546, pruned_loss=0.1194, over 5714355.72 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3357, pruned_loss=0.08617, over 5667219.01 frames. ], batch size: 705, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:41:36,012 INFO [train.py:968] (0/2) Epoch 10, batch 34550, libri_loss[loss=0.3043, simple_loss=0.3708, pruned_loss=0.1189, over 28628.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.34, pruned_loss=0.09169, over 5682466.00 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3548, pruned_loss=0.1195, over 5716603.05 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3389, pruned_loss=0.08852, over 5666880.26 frames. ], batch size: 106, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:42:30,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.640e+02 1.434e+03 1.933e+03 2.920e+03 1.218e+04, threshold=3.867e+03, percent-clipped=22.0
+2023-03-05 10:42:32,136 INFO [train.py:968] (0/2) Epoch 10, batch 34600, libri_loss[loss=0.2937, simple_loss=0.3565, pruned_loss=0.1155, over 27691.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3418, pruned_loss=0.09236, over 5685201.44 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3551, pruned_loss=0.1197, over 5716783.68 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3402, pruned_loss=0.08882, over 5671527.59 frames. ], batch size: 116, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:43:31,602 INFO [train.py:968] (0/2) Epoch 10, batch 34650, giga_loss[loss=0.234, simple_loss=0.3135, pruned_loss=0.07729, over 28628.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3385, pruned_loss=0.09116, over 5675904.14 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.355, pruned_loss=0.1196, over 5718656.00 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3373, pruned_loss=0.08822, over 5663196.96 frames. ], batch size: 262, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:43:59,897 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-05 10:44:18,007 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.114e+02 1.452e+03 1.891e+03 2.692e+03 1.221e+04, threshold=3.782e+03, percent-clipped=9.0
+2023-03-05 10:44:20,040 INFO [train.py:968] (0/2) Epoch 10, batch 34700, libri_loss[loss=0.2866, simple_loss=0.3573, pruned_loss=0.1079, over 29546.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3388, pruned_loss=0.09272, over 5672714.46 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3553, pruned_loss=0.1198, over 5713658.07 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3367, pruned_loss=0.08885, over 5664742.56 frames. ], batch size: 89, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:45:13,245 INFO [train.py:968] (0/2) Epoch 10, batch 34750, giga_loss[loss=0.2733, simple_loss=0.3465, pruned_loss=0.09998, over 28587.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3387, pruned_loss=0.0931, over 5659343.35 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3551, pruned_loss=0.1196, over 5697938.64 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3369, pruned_loss=0.08963, over 5667354.22 frames. ], batch size: 92, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:45:28,839 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=444361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:45:55,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.970e+02 1.447e+03 1.840e+03 2.546e+03 7.694e+03, threshold=3.679e+03, percent-clipped=7.0
+2023-03-05 10:45:58,773 INFO [train.py:968] (0/2) Epoch 10, batch 34800, giga_loss[loss=0.2978, simple_loss=0.3749, pruned_loss=0.1104, over 28063.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3465, pruned_loss=0.09813, over 5656038.25 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3548, pruned_loss=0.1194, over 5693846.21 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3448, pruned_loss=0.09459, over 5665306.26 frames. ], batch size: 412, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:46:40,904 INFO [train.py:968] (0/2) Epoch 10, batch 34850, giga_loss[loss=0.3163, simple_loss=0.3882, pruned_loss=0.1222, over 28267.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3549, pruned_loss=0.1032, over 5651093.87 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3547, pruned_loss=0.1193, over 5686034.64 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3535, pruned_loss=0.1, over 5664970.55 frames. ], batch size: 368, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:46:56,264 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=444462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:46:57,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=444464.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:47:21,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.083e+02 1.181e+03 1.554e+03 2.380e+03 1.249e+04, threshold=3.108e+03, percent-clipped=8.0
+2023-03-05 10:47:24,466 INFO [train.py:968] (0/2) Epoch 10, batch 34900, giga_loss[loss=0.2523, simple_loss=0.3292, pruned_loss=0.08774, over 28576.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3585, pruned_loss=0.106, over 5657962.41 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3548, pruned_loss=0.1194, over 5689489.13 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3574, pruned_loss=0.1032, over 5665281.46 frames. ], batch size: 60, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:47:33,283 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=444504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:47:35,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=444507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:47:56,216 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:48:00,437 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=444536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:48:06,512 INFO [train.py:968] (0/2) Epoch 10, batch 34950, giga_loss[loss=0.2381, simple_loss=0.3168, pruned_loss=0.0797, over 28826.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3531, pruned_loss=0.1038, over 5673231.99 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3551, pruned_loss=0.1195, over 5692098.54 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3519, pruned_loss=0.1012, over 5676084.75 frames. ], batch size: 199, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:48:17,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4322, 2.9496, 1.4725, 1.4666], device='cuda:0'), covar=tensor([0.0863, 0.0307, 0.0854, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0500, 0.0336, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 10:48:43,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.694e+02 1.041e+03 1.292e+03 1.810e+03 6.262e+03, threshold=2.584e+03, percent-clipped=6.0
+2023-03-05 10:48:46,403 INFO [train.py:968] (0/2) Epoch 10, batch 35000, giga_loss[loss=0.3476, simple_loss=0.3804, pruned_loss=0.1574, over 26663.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3472, pruned_loss=0.1017, over 5668397.13 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3555, pruned_loss=0.1196, over 5679929.75 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3457, pruned_loss=0.09893, over 5681929.83 frames. ], batch size: 555, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:48:56,149 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=444605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:48:59,526 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=444607.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:49:00,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=444608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:49:02,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=444610.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:49:02,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8166, 2.5748, 1.7450, 1.0505], device='cuda:0'), covar=tensor([0.5396, 0.2797, 0.2879, 0.4662], device='cuda:0'), in_proj_covar=tensor([0.1546, 0.1494, 0.1484, 0.1270], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 10:49:21,029 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=444637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:49:22,427 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=444639.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:49:26,907 INFO [train.py:968] (0/2) Epoch 10, batch 35050, giga_loss[loss=0.2807, simple_loss=0.3434, pruned_loss=0.1089, over 27850.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3413, pruned_loss=0.09956, over 5663518.43 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3559, pruned_loss=0.1199, over 5674850.82 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3395, pruned_loss=0.0967, over 5679202.55 frames. ], batch size: 412, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:49:41,479 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:50:04,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.931e+02 1.013e+03 1.377e+03 1.922e+03 6.255e+03, threshold=2.754e+03, percent-clipped=13.0
+2023-03-05 10:50:06,849 INFO [train.py:968] (0/2) Epoch 10, batch 35100, giga_loss[loss=0.2081, simple_loss=0.2849, pruned_loss=0.0657, over 28733.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3353, pruned_loss=0.0972, over 5676012.28 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3562, pruned_loss=0.1198, over 5681526.09 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3329, pruned_loss=0.09418, over 5682414.29 frames. ], batch size: 60, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:50:47,146 INFO [train.py:968] (0/2) Epoch 10, batch 35150, giga_loss[loss=0.2126, simple_loss=0.2905, pruned_loss=0.06736, over 29033.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3283, pruned_loss=0.09395, over 5674758.60 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3558, pruned_loss=0.1194, over 5677489.59 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3263, pruned_loss=0.09136, over 5682480.42 frames. ], batch size: 128, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:50:52,829 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:50:59,223 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444761.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:51:23,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.370e+02 1.035e+03 1.365e+03 1.904e+03 6.409e+03, threshold=2.731e+03, percent-clipped=9.0
+2023-03-05 10:51:25,241 INFO [train.py:968] (0/2) Epoch 10, batch 35200, giga_loss[loss=0.213, simple_loss=0.2884, pruned_loss=0.06881, over 28548.00 frames. ], tot_loss[loss=0.254, simple_loss=0.324, pruned_loss=0.09202, over 5676484.22 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3558, pruned_loss=0.1192, over 5680969.14 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3214, pruned_loss=0.08923, over 5679463.49 frames. ], batch size: 336, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:51:31,293 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 10:51:39,403 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444812.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:52:06,607 INFO [train.py:968] (0/2) Epoch 10, batch 35250, giga_loss[loss=0.2217, simple_loss=0.2926, pruned_loss=0.07541, over 28808.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3214, pruned_loss=0.09062, over 5676333.59 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3557, pruned_loss=0.1192, over 5673727.42 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3186, pruned_loss=0.08773, over 5684763.27 frames. ], batch size: 199, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:52:44,402 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.868e+02 1.107e+03 1.465e+03 1.990e+03 5.674e+03, threshold=2.931e+03, percent-clipped=15.0
+2023-03-05 10:52:46,914 INFO [train.py:968] (0/2) Epoch 10, batch 35300, giga_loss[loss=0.2129, simple_loss=0.2897, pruned_loss=0.06805, over 29009.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3182, pruned_loss=0.0886, over 5687595.13 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3563, pruned_loss=0.1194, over 5674541.53 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3149, pruned_loss=0.08565, over 5693646.93 frames. ], batch size: 128, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:52:56,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=444906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:53:25,958 INFO [train.py:968] (0/2) Epoch 10, batch 35350, giga_loss[loss=0.2295, simple_loss=0.2984, pruned_loss=0.08028, over 27791.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3155, pruned_loss=0.08721, over 5687933.43 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3564, pruned_loss=0.1193, over 5669097.39 frames. ], giga_tot_loss[loss=0.2394, simple_loss=0.3113, pruned_loss=0.08375, over 5699056.05 frames. ], batch size: 474, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:53:50,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-05 10:54:02,597 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.724e+02 1.044e+03 1.378e+03 2.007e+03 8.432e+03, threshold=2.757e+03, percent-clipped=10.0
+2023-03-05 10:54:04,142 INFO [train.py:968] (0/2) Epoch 10, batch 35400, giga_loss[loss=0.2032, simple_loss=0.279, pruned_loss=0.06373, over 28571.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3109, pruned_loss=0.08459, over 5690889.41 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3563, pruned_loss=0.1193, over 5669324.29 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.3073, pruned_loss=0.08162, over 5699664.08 frames. ], batch size: 60, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:54:28,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=445024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:54:38,094 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:54:42,491 INFO [train.py:968] (0/2) Epoch 10, batch 35450, giga_loss[loss=0.1947, simple_loss=0.2801, pruned_loss=0.05459, over 29050.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3095, pruned_loss=0.08426, over 5690931.75 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3562, pruned_loss=0.1192, over 5677148.11 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.3044, pruned_loss=0.08027, over 5691581.98 frames. ], batch size: 164, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:54:45,360 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-05 10:54:47,150 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:54:49,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:55:13,032 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:55:23,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.964e+02 9.835e+02 1.237e+03 1.704e+03 6.166e+03, threshold=2.473e+03, percent-clipped=6.0
+2023-03-05 10:55:24,449 INFO [train.py:968] (0/2) Epoch 10, batch 35500, giga_loss[loss=0.2259, simple_loss=0.299, pruned_loss=0.07635, over 28724.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3048, pruned_loss=0.08157, over 5688473.72 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3562, pruned_loss=0.1191, over 5678441.37 frames. ], giga_tot_loss[loss=0.2286, simple_loss=0.3006, pruned_loss=0.0783, over 5687850.06 frames. ], batch size: 284, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:55:34,827 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2377, 1.6084, 1.5638, 1.1930], device='cuda:0'), covar=tensor([0.1456, 0.2080, 0.1168, 0.1337], device='cuda:0'), in_proj_covar=tensor([0.0831, 0.0698, 0.0861, 0.0773], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 10:55:37,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5890, 1.7387, 1.3644, 1.4434], device='cuda:0'), covar=tensor([0.2077, 0.1662, 0.1569, 0.1703], device='cuda:0'), in_proj_covar=tensor([0.1678, 0.1548, 0.1518, 0.1644], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 10:55:53,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:55:53,310 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6368, 2.4396, 1.5283, 0.9298], device='cuda:0'), covar=tensor([0.6018, 0.2663, 0.3420, 0.4773], device='cuda:0'), in_proj_covar=tensor([0.1544, 0.1483, 0.1473, 0.1262], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 10:56:01,409 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445136.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:56:07,661 INFO [train.py:968] (0/2) Epoch 10, batch 35550, giga_loss[loss=0.2333, simple_loss=0.2982, pruned_loss=0.08421, over 28979.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3021, pruned_loss=0.08058, over 5698040.77 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3563, pruned_loss=0.1192, over 5682600.92 frames. ], giga_tot_loss[loss=0.226, simple_loss=0.2977, pruned_loss=0.07712, over 5693924.06 frames. ], batch size: 106, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:56:39,308 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:56:41,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:56:44,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445187.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:56:49,241 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.043e+02 9.734e+02 1.153e+03 1.554e+03 1.683e+04, threshold=2.306e+03, percent-clipped=12.0
+2023-03-05 10:56:49,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9350, 2.2812, 1.8356, 2.1635], device='cuda:0'), covar=tensor([0.2049, 0.2077, 0.2244, 0.2223], device='cuda:0'), in_proj_covar=tensor([0.1259, 0.0939, 0.1120, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 10:56:51,966 INFO [train.py:968] (0/2) Epoch 10, batch 35600, giga_loss[loss=0.2253, simple_loss=0.3027, pruned_loss=0.07394, over 28883.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3021, pruned_loss=0.08128, over 5694802.86 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3568, pruned_loss=0.1195, over 5685144.55 frames. ], giga_tot_loss[loss=0.2264, simple_loss=0.2974, pruned_loss=0.07776, over 5689523.29 frames. ], batch size: 199, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:57:06,443 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=445212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:57:06,458 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:57:28,693 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2848, 1.7737, 1.3421, 0.4301], device='cuda:0'), covar=tensor([0.2555, 0.1720, 0.2488, 0.3588], device='cuda:0'), in_proj_covar=tensor([0.1541, 0.1484, 0.1473, 0.1261], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 10:57:34,191 INFO [train.py:968] (0/2) Epoch 10, batch 35650, giga_loss[loss=0.3289, simple_loss=0.3778, pruned_loss=0.14, over 26653.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3132, pruned_loss=0.08745, over 5693930.00 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3573, pruned_loss=0.1198, over 5692079.84 frames. ], giga_tot_loss[loss=0.2369, simple_loss=0.3074, pruned_loss=0.08318, over 5683687.99 frames. ], batch size: 555, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:57:46,956 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 10:57:56,770 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:57:59,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:58:05,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445279.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:58:09,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445282.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:58:16,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.615e+02 1.227e+03 1.572e+03 1.988e+03 5.476e+03, threshold=3.143e+03, percent-clipped=17.0
+2023-03-05 10:58:18,239 INFO [train.py:968] (0/2) Epoch 10, batch 35700, libri_loss[loss=0.3745, simple_loss=0.4153, pruned_loss=0.1669, over 19140.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3274, pruned_loss=0.09523, over 5685504.17 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3577, pruned_loss=0.1202, over 5686769.61 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3215, pruned_loss=0.09087, over 5683241.52 frames. ], batch size: 186, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:58:23,561 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:58:32,265 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445311.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:58:45,886 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445330.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:58:48,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445333.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:58:55,517 INFO [train.py:968] (0/2) Epoch 10, batch 35750, libri_loss[loss=0.2583, simple_loss=0.3248, pruned_loss=0.09587, over 29376.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3385, pruned_loss=0.1009, over 5696341.67 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3577, pruned_loss=0.12, over 5693082.07 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3328, pruned_loss=0.09657, over 5688789.94 frames. ], batch size: 67, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:59:04,223 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4118, 2.1997, 1.5684, 0.5415], device='cuda:0'), covar=tensor([0.3445, 0.1972, 0.2858, 0.3866], device='cuda:0'), in_proj_covar=tensor([0.1546, 0.1492, 0.1478, 0.1266], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 10:59:10,021 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445362.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:59:26,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2071, 1.8056, 1.3492, 0.3685], device='cuda:0'), covar=tensor([0.3169, 0.1969, 0.3227, 0.4052], device='cuda:0'), in_proj_covar=tensor([0.1553, 0.1497, 0.1483, 0.1272], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 10:59:35,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.930e+02 1.113e+03 1.532e+03 2.198e+03 7.485e+03, threshold=3.064e+03, percent-clipped=12.0
+2023-03-05 10:59:37,822 INFO [train.py:968] (0/2) Epoch 10, batch 35800, giga_loss[loss=0.2866, simple_loss=0.3608, pruned_loss=0.1062, over 28232.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3449, pruned_loss=0.1031, over 5689756.57 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3578, pruned_loss=0.1201, over 5687868.47 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.34, pruned_loss=0.09932, over 5688164.87 frames. ], batch size: 77, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:59:41,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:00:20,539 INFO [train.py:968] (0/2) Epoch 10, batch 35850, giga_loss[loss=0.2628, simple_loss=0.3377, pruned_loss=0.09398, over 28997.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3473, pruned_loss=0.1027, over 5687292.43 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3583, pruned_loss=0.1204, over 5687266.15 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3429, pruned_loss=0.09928, over 5686477.67 frames. ], batch size: 106, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:00:21,640 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 11:00:47,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5711, 1.8017, 1.8712, 1.4150], device='cuda:0'), covar=tensor([0.1846, 0.2230, 0.1413, 0.1660], device='cuda:0'), in_proj_covar=tensor([0.0824, 0.0694, 0.0857, 0.0771], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 11:01:07,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.428e+02 1.145e+03 1.377e+03 1.757e+03 3.758e+03, threshold=2.755e+03, percent-clipped=2.0
+2023-03-05 11:01:09,561 INFO [train.py:968] (0/2) Epoch 10, batch 35900, giga_loss[loss=0.3015, simple_loss=0.3703, pruned_loss=0.1164, over 27671.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3506, pruned_loss=0.1039, over 5685303.75 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3587, pruned_loss=0.1207, over 5689242.20 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3466, pruned_loss=0.1007, over 5683044.10 frames. ], batch size: 472, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:01:47,314 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:01:48,271 INFO [train.py:968] (0/2) Epoch 10, batch 35950, libri_loss[loss=0.3108, simple_loss=0.3796, pruned_loss=0.121, over 29241.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3533, pruned_loss=0.1059, over 5677981.72 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3595, pruned_loss=0.1211, over 5676585.15 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3491, pruned_loss=0.1025, over 5687600.55 frames. ], batch size: 94, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:01:49,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:02:10,838 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-05 11:02:13,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:02:25,344 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:02:25,482 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-05 11:02:30,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.946e+02 1.158e+03 1.410e+03 2.250e+03 1.084e+04, threshold=2.821e+03, percent-clipped=20.0
+2023-03-05 11:02:30,668 INFO [train.py:968] (0/2) Epoch 10, batch 36000, giga_loss[loss=0.3275, simple_loss=0.3716, pruned_loss=0.1417, over 23396.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.356, pruned_loss=0.1083, over 5667689.05 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3596, pruned_loss=0.1213, over 5675368.02 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3525, pruned_loss=0.1052, over 5676126.57 frames. ], batch size: 705, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 11:02:30,671 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 11:02:39,708 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2162, simple_loss=0.3232, pruned_loss=0.0546, over 944034.00 frames. 
+2023-03-05 11:02:39,708 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 11:03:17,257 INFO [train.py:968] (0/2) Epoch 10, batch 36050, giga_loss[loss=0.275, simple_loss=0.3548, pruned_loss=0.09765, over 28628.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3591, pruned_loss=0.1099, over 5682221.53 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3601, pruned_loss=0.1214, over 5680999.09 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3557, pruned_loss=0.1068, over 5684083.43 frames. ], batch size: 242, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:03:56,003 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.533e+02 1.135e+03 1.533e+03 2.235e+03 4.549e+03, threshold=3.067e+03, percent-clipped=16.0
+2023-03-05 11:03:56,015 INFO [train.py:968] (0/2) Epoch 10, batch 36100, giga_loss[loss=0.3032, simple_loss=0.3802, pruned_loss=0.1131, over 28194.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3623, pruned_loss=0.1112, over 5672472.68 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3603, pruned_loss=0.1213, over 5666505.28 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3595, pruned_loss=0.1086, over 5687010.33 frames. ], batch size: 368, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:04:24,231 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=445728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:04:24,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3673, 1.6299, 1.6240, 1.2125], device='cuda:0'), covar=tensor([0.1514, 0.2290, 0.1251, 0.1513], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0692, 0.0853, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 11:04:25,614 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:04:27,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:04:31,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 11:04:35,986 INFO [train.py:968] (0/2) Epoch 10, batch 36150, libri_loss[loss=0.382, simple_loss=0.4176, pruned_loss=0.1733, over 25513.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3642, pruned_loss=0.1119, over 5676110.90 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3612, pruned_loss=0.1218, over 5671910.39 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3611, pruned_loss=0.1088, over 5683606.98 frames. ], batch size: 136, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:04:50,913 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:05:15,670 INFO [train.py:968] (0/2) Epoch 10, batch 36200, libri_loss[loss=0.2866, simple_loss=0.3429, pruned_loss=0.1152, over 29339.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3659, pruned_loss=0.1122, over 5673988.92 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3614, pruned_loss=0.1219, over 5668087.47 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3634, pruned_loss=0.1095, over 5682779.80 frames. ], batch size: 71, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:05:16,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.952e+02 1.219e+03 1.570e+03 2.204e+03 5.655e+03, threshold=3.140e+03, percent-clipped=12.0
+2023-03-05 11:05:52,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4053, 2.0460, 1.4927, 0.7158], device='cuda:0'), covar=tensor([0.4109, 0.2246, 0.3241, 0.4417], device='cuda:0'), in_proj_covar=tensor([0.1532, 0.1474, 0.1471, 0.1257], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 11:05:55,077 INFO [train.py:968] (0/2) Epoch 10, batch 36250, giga_loss[loss=0.2844, simple_loss=0.3617, pruned_loss=0.1035, over 28807.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3651, pruned_loss=0.1102, over 5686159.82 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3617, pruned_loss=0.122, over 5670415.40 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3629, pruned_loss=0.1078, over 5691201.06 frames. ], batch size: 99, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:06:34,214 INFO [train.py:968] (0/2) Epoch 10, batch 36300, giga_loss[loss=0.2682, simple_loss=0.3516, pruned_loss=0.09238, over 28625.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3639, pruned_loss=0.1084, over 5683611.37 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3624, pruned_loss=0.1223, over 5664752.85 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3615, pruned_loss=0.1058, over 5693993.12 frames. ], batch size: 85, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:06:34,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.758e+02 1.000e+03 1.362e+03 1.938e+03 5.157e+03, threshold=2.723e+03, percent-clipped=6.0
+2023-03-05 11:06:45,868 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=445908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:06:53,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-05 11:07:14,730 INFO [train.py:968] (0/2) Epoch 10, batch 36350, giga_loss[loss=0.2847, simple_loss=0.3648, pruned_loss=0.1023, over 28841.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3617, pruned_loss=0.1063, over 5698180.25 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3627, pruned_loss=0.1224, over 5669422.27 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3596, pruned_loss=0.1038, over 5702743.02 frames. ], batch size: 112, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:07:23,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7923, 2.3646, 2.0631, 1.6258], device='cuda:0'), covar=tensor([0.1518, 0.2030, 0.1277, 0.1579], device='cuda:0'), in_proj_covar=tensor([0.0825, 0.0693, 0.0857, 0.0770], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 11:07:37,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3223, 1.6270, 1.5610, 1.1709], device='cuda:0'), covar=tensor([0.1540, 0.2159, 0.1243, 0.1551], device='cuda:0'), in_proj_covar=tensor([0.0825, 0.0694, 0.0857, 0.0770], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 11:07:44,156 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=445980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:07:50,727 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 11:07:56,594 INFO [train.py:968] (0/2) Epoch 10, batch 36400, giga_loss[loss=0.3139, simple_loss=0.3826, pruned_loss=0.1226, over 29003.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3627, pruned_loss=0.1083, over 5696814.84 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3624, pruned_loss=0.1222, over 5669524.40 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3612, pruned_loss=0.1064, over 5700736.70 frames. ], batch size: 164, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:07:57,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.064e+02 1.030e+03 1.320e+03 1.840e+03 5.348e+03, threshold=2.640e+03, percent-clipped=6.0
+2023-03-05 11:08:02,933 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-446000.pt
+2023-03-05 11:08:26,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7997, 2.3918, 2.0980, 1.6336], device='cuda:0'), covar=tensor([0.1558, 0.1854, 0.1257, 0.1509], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0688, 0.0851, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 11:08:40,512 INFO [train.py:968] (0/2) Epoch 10, batch 36450, giga_loss[loss=0.2871, simple_loss=0.3611, pruned_loss=0.1065, over 28960.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3656, pruned_loss=0.1133, over 5699250.09 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3628, pruned_loss=0.1224, over 5677696.18 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3642, pruned_loss=0.1112, over 5695763.96 frames. ], batch size: 164, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:08:49,131 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 11:08:54,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5906, 4.3334, 4.1677, 1.8864], device='cuda:0'), covar=tensor([0.0506, 0.0685, 0.0676, 0.2105], device='cuda:0'), in_proj_covar=tensor([0.0999, 0.0941, 0.0821, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 11:09:21,302 INFO [train.py:968] (0/2) Epoch 10, batch 36500, giga_loss[loss=0.2951, simple_loss=0.3611, pruned_loss=0.1145, over 29036.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3663, pruned_loss=0.1152, over 5698135.66 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3632, pruned_loss=0.1226, over 5680782.37 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3649, pruned_loss=0.1133, over 5693194.79 frames. ], batch size: 164, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:09:22,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.432e+02 1.392e+03 1.815e+03 2.397e+03 7.821e+03, threshold=3.631e+03, percent-clipped=17.0
+2023-03-05 11:09:26,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4409, 1.5858, 1.3088, 1.5567], device='cuda:0'), covar=tensor([0.0735, 0.0308, 0.0314, 0.0769], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 11:09:29,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:09:37,231 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=446108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:10:06,909 INFO [train.py:968] (0/2) Epoch 10, batch 36550, giga_loss[loss=0.2607, simple_loss=0.3355, pruned_loss=0.09298, over 29058.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3637, pruned_loss=0.1141, over 5703093.22 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3632, pruned_loss=0.1226, over 5681531.88 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3626, pruned_loss=0.1126, over 5698763.28 frames. ], batch size: 136, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:10:47,840 INFO [train.py:968] (0/2) Epoch 10, batch 36600, giga_loss[loss=0.3096, simple_loss=0.3833, pruned_loss=0.118, over 28731.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3627, pruned_loss=0.1135, over 5706564.87 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3638, pruned_loss=0.1225, over 5688497.83 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3613, pruned_loss=0.1119, over 5697366.07 frames. ], batch size: 119, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:10:48,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.677e+02 1.099e+03 1.384e+03 1.873e+03 3.752e+03, threshold=2.768e+03, percent-clipped=1.0
+2023-03-05 11:11:29,802 INFO [train.py:968] (0/2) Epoch 10, batch 36650, giga_loss[loss=0.2742, simple_loss=0.3506, pruned_loss=0.09894, over 28569.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.361, pruned_loss=0.1116, over 5709826.62 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3639, pruned_loss=0.1225, over 5693696.01 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3597, pruned_loss=0.1102, over 5698447.04 frames. ], batch size: 85, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:11:32,528 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=446246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:11:34,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=446249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:11:59,855 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=446278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:12:04,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:12:13,007 INFO [train.py:968] (0/2) Epoch 10, batch 36700, giga_loss[loss=0.2848, simple_loss=0.3588, pruned_loss=0.1054, over 28713.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3595, pruned_loss=0.1102, over 5681769.82 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3645, pruned_loss=0.1228, over 5680415.11 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3578, pruned_loss=0.1084, over 5685768.21 frames. ], batch size: 119, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:12:15,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.509e+02 1.208e+03 1.607e+03 2.358e+03 7.897e+03, threshold=3.214e+03, percent-clipped=14.0
+2023-03-05 11:12:28,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0852, 1.0800, 4.1341, 3.1421], device='cuda:0'), covar=tensor([0.1616, 0.2578, 0.0402, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0641, 0.0576, 0.0841, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 11:12:57,257 INFO [train.py:968] (0/2) Epoch 10, batch 36750, giga_loss[loss=0.2997, simple_loss=0.361, pruned_loss=0.1192, over 28976.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.357, pruned_loss=0.1088, over 5677459.83 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3657, pruned_loss=0.1236, over 5675193.24 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3545, pruned_loss=0.1064, over 5685435.80 frames. ], batch size: 213, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:13:06,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:13:41,213 INFO [train.py:968] (0/2) Epoch 10, batch 36800, giga_loss[loss=0.2399, simple_loss=0.3143, pruned_loss=0.08282, over 28206.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3511, pruned_loss=0.1057, over 5654564.82 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3667, pruned_loss=0.1241, over 5660986.72 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3478, pruned_loss=0.1028, over 5672627.66 frames. ], batch size: 368, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:13:42,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.771e+02 1.032e+03 1.354e+03 1.963e+03 1.212e+04, threshold=2.708e+03, percent-clipped=12.0
+2023-03-05 11:13:44,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6809, 1.7648, 1.5307, 1.3976], device='cuda:0'), covar=tensor([0.1677, 0.1551, 0.1404, 0.1670], device='cuda:0'), in_proj_covar=tensor([0.1669, 0.1558, 0.1530, 0.1644], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 11:14:12,073 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=446426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:14:16,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=446429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:14:28,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5595, 2.4231, 1.7633, 2.0826], device='cuda:0'), covar=tensor([0.0639, 0.0556, 0.0904, 0.0916], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0432, 0.0493, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 11:14:31,743 INFO [train.py:968] (0/2) Epoch 10, batch 36850, giga_loss[loss=0.2387, simple_loss=0.3187, pruned_loss=0.07937, over 29045.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3459, pruned_loss=0.1036, over 5647916.77 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3669, pruned_loss=0.1243, over 5667258.84 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3426, pruned_loss=0.1006, over 5656534.53 frames. ], batch size: 164, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:14:44,009 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=446458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:07,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:09,698 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=446486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:15,428 INFO [train.py:968] (0/2) Epoch 10, batch 36900, libri_loss[loss=0.2739, simple_loss=0.3337, pruned_loss=0.107, over 29343.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3429, pruned_loss=0.1019, over 5651803.87 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3669, pruned_loss=0.1243, over 5671823.33 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3398, pruned_loss=0.09918, over 5654323.45 frames. ], batch size: 71, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:15:17,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.743e+02 9.477e+02 1.225e+03 1.568e+03 8.568e+03, threshold=2.449e+03, percent-clipped=4.0
+2023-03-05 11:15:19,117 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=446498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:21,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=446501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:21,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=446501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:45,929 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=446530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:56,620 INFO [train.py:968] (0/2) Epoch 10, batch 36950, giga_loss[loss=0.2655, simple_loss=0.3397, pruned_loss=0.09562, over 28962.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3429, pruned_loss=0.101, over 5650208.89 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3673, pruned_loss=0.1244, over 5656650.31 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3396, pruned_loss=0.09826, over 5665093.08 frames. ], batch size: 227, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:16:34,632 INFO [train.py:968] (0/2) Epoch 10, batch 37000, giga_loss[loss=0.2803, simple_loss=0.3545, pruned_loss=0.1031, over 28523.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3438, pruned_loss=0.1017, over 5660465.84 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3678, pruned_loss=0.1245, over 5658739.88 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3398, pruned_loss=0.09842, over 5670308.93 frames. ], batch size: 307, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:16:35,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.263e+02 1.006e+03 1.472e+03 2.135e+03 4.655e+03, threshold=2.943e+03, percent-clipped=18.0
+2023-03-05 11:17:01,894 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=446626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:17:04,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=446629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:17:15,352 INFO [train.py:968] (0/2) Epoch 10, batch 37050, giga_loss[loss=0.2614, simple_loss=0.3337, pruned_loss=0.09452, over 28721.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3419, pruned_loss=0.1005, over 5679671.13 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3684, pruned_loss=0.1248, over 5663303.02 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3374, pruned_loss=0.09702, over 5683785.67 frames. ], batch size: 284, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:17:26,638 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=446658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:17:51,549 INFO [train.py:968] (0/2) Epoch 10, batch 37100, libri_loss[loss=0.3507, simple_loss=0.4148, pruned_loss=0.1433, over 29268.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3404, pruned_loss=0.09973, over 5687092.45 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3691, pruned_loss=0.125, over 5656361.03 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3352, pruned_loss=0.09586, over 5698109.28 frames. ], batch size: 94, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:17:52,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.473e+02 1.021e+03 1.453e+03 2.628e+03 1.263e+04, threshold=2.905e+03, percent-clipped=20.0
+2023-03-05 11:18:21,598 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=446732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:18:29,815 INFO [train.py:968] (0/2) Epoch 10, batch 37150, giga_loss[loss=0.2282, simple_loss=0.3056, pruned_loss=0.0754, over 28950.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3363, pruned_loss=0.09763, over 5693449.31 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3692, pruned_loss=0.125, over 5660420.76 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3317, pruned_loss=0.0942, over 5698897.12 frames. ], batch size: 227, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:18:30,827 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=446745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:19:08,311 INFO [train.py:968] (0/2) Epoch 10, batch 37200, giga_loss[loss=0.2399, simple_loss=0.3134, pruned_loss=0.08323, over 28794.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3329, pruned_loss=0.0959, over 5698869.38 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3692, pruned_loss=0.125, over 5662809.26 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3289, pruned_loss=0.09296, over 5701380.50 frames. ], batch size: 199, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:19:09,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.727e+02 8.493e+02 1.069e+03 1.401e+03 4.074e+03, threshold=2.138e+03, percent-clipped=3.0
+2023-03-05 11:19:47,327 INFO [train.py:968] (0/2) Epoch 10, batch 37250, libri_loss[loss=0.413, simple_loss=0.4563, pruned_loss=0.1849, over 29268.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.333, pruned_loss=0.09614, over 5707391.15 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3712, pruned_loss=0.126, over 5662568.37 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3268, pruned_loss=0.09192, over 5710811.15 frames. ], batch size: 94, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:19:58,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:20:12,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:20:25,619 INFO [train.py:968] (0/2) Epoch 10, batch 37300, libri_loss[loss=0.3039, simple_loss=0.3746, pruned_loss=0.1166, over 29562.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3308, pruned_loss=0.09464, over 5712751.05 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3717, pruned_loss=0.126, over 5666512.39 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3243, pruned_loss=0.09042, over 5712862.17 frames. ], batch size: 79, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:20:27,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.889e+02 9.654e+02 1.234e+03 1.694e+03 5.926e+03, threshold=2.467e+03, percent-clipped=15.0
+2023-03-05 11:21:01,196 INFO [train.py:968] (0/2) Epoch 10, batch 37350, giga_loss[loss=0.2259, simple_loss=0.2971, pruned_loss=0.07731, over 28760.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3303, pruned_loss=0.0942, over 5713036.92 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3719, pruned_loss=0.1255, over 5671446.97 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3223, pruned_loss=0.08924, over 5711557.21 frames. ], batch size: 99, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:21:40,338 INFO [train.py:968] (0/2) Epoch 10, batch 37400, giga_loss[loss=0.2713, simple_loss=0.3415, pruned_loss=0.1006, over 28851.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3287, pruned_loss=0.09294, over 5716668.55 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3723, pruned_loss=0.1254, over 5677091.05 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.321, pruned_loss=0.08832, over 5711275.76 frames. ], batch size: 199, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:21:43,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.186e+02 9.458e+02 1.202e+03 1.973e+03 1.488e+04, threshold=2.405e+03, percent-clipped=17.0
+2023-03-05 11:21:48,737 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:21:50,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:00,951 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:02,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:11,935 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:12,536 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:18,915 INFO [train.py:968] (0/2) Epoch 10, batch 37450, giga_loss[loss=0.2591, simple_loss=0.319, pruned_loss=0.09954, over 24175.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3279, pruned_loss=0.09264, over 5714822.93 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3725, pruned_loss=0.1254, over 5682446.37 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3204, pruned_loss=0.08813, over 5706377.19 frames. ], batch size: 705, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:22:19,106 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:23,828 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:51,695 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:58,713 INFO [train.py:968] (0/2) Epoch 10, batch 37500, giga_loss[loss=0.2821, simple_loss=0.3518, pruned_loss=0.1062, over 28969.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3285, pruned_loss=0.09257, over 5712712.92 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3727, pruned_loss=0.1253, over 5674497.97 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3214, pruned_loss=0.08841, over 5713157.43 frames. ], batch size: 164, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:23:00,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.650e+02 9.166e+02 1.177e+03 1.564e+03 5.302e+03, threshold=2.353e+03, percent-clipped=6.0
+2023-03-05 11:23:09,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:23:18,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:23:38,302 INFO [train.py:968] (0/2) Epoch 10, batch 37550, giga_loss[loss=0.2698, simple_loss=0.3383, pruned_loss=0.1006, over 28922.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3348, pruned_loss=0.09689, over 5716548.71 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3738, pruned_loss=0.1258, over 5679099.90 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3268, pruned_loss=0.09211, over 5714026.96 frames. ], batch size: 213, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:24:21,172 INFO [train.py:968] (0/2) Epoch 10, batch 37600, giga_loss[loss=0.307, simple_loss=0.3726, pruned_loss=0.1207, over 28900.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3421, pruned_loss=0.1017, over 5702202.54 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3743, pruned_loss=0.1259, over 5671267.86 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3343, pruned_loss=0.09697, over 5708146.14 frames. ], batch size: 145, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:24:25,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.851e+02 1.186e+03 1.561e+03 2.452e+03 6.199e+03, threshold=3.122e+03, percent-clipped=26.0
+2023-03-05 11:24:35,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:24:55,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1727, 2.9053, 1.3116, 1.2917], device='cuda:0'), covar=tensor([0.1010, 0.0357, 0.0871, 0.1420], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0498, 0.0334, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 11:25:09,939 INFO [train.py:968] (0/2) Epoch 10, batch 37650, giga_loss[loss=0.2723, simple_loss=0.3485, pruned_loss=0.09807, over 28839.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3499, pruned_loss=0.1071, over 5696920.89 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3742, pruned_loss=0.1259, over 5673655.25 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3435, pruned_loss=0.1032, over 5699746.36 frames. ], batch size: 186, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:25:16,449 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:25:19,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:25:27,727 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:25:31,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:25:47,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:26:00,547 INFO [train.py:968] (0/2) Epoch 10, batch 37700, giga_loss[loss=0.2867, simple_loss=0.366, pruned_loss=0.1037, over 28720.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3554, pruned_loss=0.11, over 5683178.68 frames. ], libri_tot_loss[loss=0.3133, simple_loss=0.3746, pruned_loss=0.126, over 5674519.83 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3496, pruned_loss=0.1065, over 5685081.76 frames. ], batch size: 284, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:26:01,395 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:26:01,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5148, 1.7977, 1.7627, 1.3187], device='cuda:0'), covar=tensor([0.1582, 0.2079, 0.1266, 0.1496], device='cuda:0'), in_proj_covar=tensor([0.0823, 0.0694, 0.0858, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 11:26:04,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.506e+02 1.192e+03 1.520e+03 2.553e+03 8.055e+03, threshold=3.040e+03, percent-clipped=15.0
+2023-03-05 11:26:09,627 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:26:44,610 INFO [train.py:968] (0/2) Epoch 10, batch 37750, giga_loss[loss=0.3136, simple_loss=0.3832, pruned_loss=0.122, over 28313.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3597, pruned_loss=0.1116, over 5687231.30 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3746, pruned_loss=0.1261, over 5678195.06 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3546, pruned_loss=0.1082, over 5685693.62 frames. ], batch size: 368, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:26:45,521 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 11:27:31,169 INFO [train.py:968] (0/2) Epoch 10, batch 37800, giga_loss[loss=0.3309, simple_loss=0.3908, pruned_loss=0.1355, over 28544.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3663, pruned_loss=0.1159, over 5677084.29 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3749, pruned_loss=0.1264, over 5672725.77 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3617, pruned_loss=0.1127, over 5680780.14 frames. ], batch size: 60, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:27:34,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.546e+02 1.096e+03 1.508e+03 2.305e+03 1.218e+04, threshold=3.017e+03, percent-clipped=15.0
+2023-03-05 11:27:45,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:27:52,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:28:13,739 INFO [train.py:968] (0/2) Epoch 10, batch 37850, giga_loss[loss=0.2761, simple_loss=0.3424, pruned_loss=0.1049, over 27675.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.365, pruned_loss=0.1147, over 5683700.61 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3746, pruned_loss=0.1262, over 5677280.74 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3614, pruned_loss=0.1122, over 5682792.95 frames. ], batch size: 472, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:28:27,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:28:34,958 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-05 11:28:54,903 INFO [train.py:968] (0/2) Epoch 10, batch 37900, giga_loss[loss=0.2338, simple_loss=0.3226, pruned_loss=0.07251, over 28985.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3596, pruned_loss=0.1104, over 5684718.34 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3754, pruned_loss=0.127, over 5670251.08 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3558, pruned_loss=0.1073, over 5689981.44 frames. ], batch size: 164, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:28:58,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.382e+02 1.037e+03 1.290e+03 1.996e+03 4.648e+03, threshold=2.581e+03, percent-clipped=8.0
+2023-03-05 11:29:40,304 INFO [train.py:968] (0/2) Epoch 10, batch 37950, giga_loss[loss=0.2904, simple_loss=0.3715, pruned_loss=0.1046, over 28711.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3572, pruned_loss=0.108, over 5689188.97 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3755, pruned_loss=0.1272, over 5672803.85 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3538, pruned_loss=0.1052, over 5691175.72 frames. ], batch size: 284, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:29:47,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:29:51,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:29:56,844 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:29:58,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:30:14,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:30:16,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:30:22,838 INFO [train.py:968] (0/2) Epoch 10, batch 38000, giga_loss[loss=0.2784, simple_loss=0.3496, pruned_loss=0.1036, over 28667.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3569, pruned_loss=0.1074, over 5697010.53 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3757, pruned_loss=0.1273, over 5675147.89 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3539, pruned_loss=0.105, over 5696665.59 frames. ], batch size: 85, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:30:23,421 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:30:26,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.417e+02 1.146e+03 1.490e+03 2.014e+03 5.827e+03, threshold=2.980e+03, percent-clipped=14.0
+2023-03-05 11:30:31,574 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:30:33,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:31:00,383 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:31:06,759 INFO [train.py:968] (0/2) Epoch 10, batch 38050, giga_loss[loss=0.3177, simple_loss=0.3864, pruned_loss=0.1245, over 28929.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3608, pruned_loss=0.1098, over 5695069.78 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3761, pruned_loss=0.1275, over 5673851.54 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3579, pruned_loss=0.1075, over 5696343.32 frames. ], batch size: 213, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:31:34,818 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-05 11:31:35,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:31:49,182 INFO [train.py:968] (0/2) Epoch 10, batch 38100, giga_loss[loss=0.3155, simple_loss=0.3811, pruned_loss=0.1249, over 28712.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3622, pruned_loss=0.1108, over 5697734.67 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3765, pruned_loss=0.1278, over 5676215.18 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3593, pruned_loss=0.1084, over 5697277.11 frames. ], batch size: 284, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:31:54,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.453e+02 1.375e+03 1.727e+03 2.505e+03 5.489e+03, threshold=3.454e+03, percent-clipped=18.0
+2023-03-05 11:32:17,548 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:32:19,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:32:35,164 INFO [train.py:968] (0/2) Epoch 10, batch 38150, libri_loss[loss=0.3401, simple_loss=0.3995, pruned_loss=0.1403, over 29530.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3641, pruned_loss=0.1127, over 5695688.46 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3771, pruned_loss=0.1282, over 5681773.75 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.361, pruned_loss=0.1102, over 5690736.22 frames. ], batch size: 83, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:32:36,083 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:32:45,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:33:17,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3581, 1.1985, 4.8897, 3.3140], device='cuda:0'), covar=tensor([0.1701, 0.2619, 0.0333, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0643, 0.0573, 0.0836, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 11:33:18,310 INFO [train.py:968] (0/2) Epoch 10, batch 38200, giga_loss[loss=0.2725, simple_loss=0.3475, pruned_loss=0.0988, over 28730.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.364, pruned_loss=0.113, over 5704212.05 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3767, pruned_loss=0.128, over 5686476.91 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3616, pruned_loss=0.1107, over 5696382.52 frames. ], batch size: 119, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:33:19,372 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3020, 1.3969, 1.5381, 1.3896], device='cuda:0'), covar=tensor([0.1137, 0.1253, 0.1257, 0.1095], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0734, 0.0667, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 11:33:22,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.527e+02 1.141e+03 1.557e+03 2.371e+03 5.791e+03, threshold=3.113e+03, percent-clipped=6.0
+2023-03-05 11:33:37,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 11:33:40,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:33:42,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:34:01,289 INFO [train.py:968] (0/2) Epoch 10, batch 38250, giga_loss[loss=0.2917, simple_loss=0.3626, pruned_loss=0.1104, over 28904.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.364, pruned_loss=0.1132, over 5691452.25 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3772, pruned_loss=0.1283, over 5680984.94 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3614, pruned_loss=0.1108, over 5689902.97 frames. ], batch size: 186, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:34:09,644 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:34:43,337 INFO [train.py:968] (0/2) Epoch 10, batch 38300, giga_loss[loss=0.2678, simple_loss=0.351, pruned_loss=0.09223, over 28382.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3646, pruned_loss=0.113, over 5698958.24 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3774, pruned_loss=0.1285, over 5684102.12 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3622, pruned_loss=0.1108, over 5694945.69 frames. ], batch size: 65, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:34:48,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.087e+02 1.058e+03 1.303e+03 1.736e+03 1.325e+04, threshold=2.606e+03, percent-clipped=9.0
+2023-03-05 11:34:51,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4613, 4.2836, 4.0678, 1.9078], device='cuda:0'), covar=tensor([0.0553, 0.0728, 0.0715, 0.2174], device='cuda:0'), in_proj_covar=tensor([0.0999, 0.0941, 0.0821, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 11:34:57,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1237, 1.0348, 3.7894, 3.2151], device='cuda:0'), covar=tensor([0.1706, 0.2773, 0.0424, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0641, 0.0570, 0.0836, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 11:35:24,774 INFO [train.py:968] (0/2) Epoch 10, batch 38350, giga_loss[loss=0.3138, simple_loss=0.3874, pruned_loss=0.1201, over 28664.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3653, pruned_loss=0.1122, over 5699516.98 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3778, pruned_loss=0.1286, over 5685184.81 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3627, pruned_loss=0.1098, over 5695560.01 frames. ], batch size: 92, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:36:03,999 INFO [train.py:968] (0/2) Epoch 10, batch 38400, giga_loss[loss=0.2668, simple_loss=0.3383, pruned_loss=0.09761, over 28678.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3656, pruned_loss=0.1117, over 5713361.10 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3778, pruned_loss=0.1285, over 5691797.99 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.363, pruned_loss=0.1094, over 5704891.81 frames. ], batch size: 92, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:36:07,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.235e+02 1.094e+03 1.436e+03 2.143e+03 7.236e+03, threshold=2.871e+03, percent-clipped=18.0
+2023-03-05 11:36:09,969 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-448000.pt
+2023-03-05 11:36:27,538 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=448019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:36:41,414 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-05 11:36:47,406 INFO [train.py:968] (0/2) Epoch 10, batch 38450, giga_loss[loss=0.2922, simple_loss=0.3621, pruned_loss=0.1111, over 28713.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.363, pruned_loss=0.1102, over 5710220.65 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3776, pruned_loss=0.1284, over 5692837.00 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3611, pruned_loss=0.1084, over 5702774.02 frames. ], batch size: 284, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:37:27,468 INFO [train.py:968] (0/2) Epoch 10, batch 38500, giga_loss[loss=0.2546, simple_loss=0.3345, pruned_loss=0.08737, over 28689.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3606, pruned_loss=0.1091, over 5712381.75 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3777, pruned_loss=0.1283, over 5697006.75 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3587, pruned_loss=0.1074, over 5703067.43 frames. ], batch size: 85, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:37:32,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.015e+02 1.031e+03 1.205e+03 1.557e+03 5.045e+03, threshold=2.409e+03, percent-clipped=4.0
+2023-03-05 11:37:49,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=448120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:38:07,092 INFO [train.py:968] (0/2) Epoch 10, batch 38550, giga_loss[loss=0.2709, simple_loss=0.3488, pruned_loss=0.09654, over 28361.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3587, pruned_loss=0.1082, over 5712923.64 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3774, pruned_loss=0.1281, over 5700992.27 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3569, pruned_loss=0.1064, over 5702240.33 frames. ], batch size: 71, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:38:45,390 INFO [train.py:968] (0/2) Epoch 10, batch 38600, giga_loss[loss=0.2909, simple_loss=0.3728, pruned_loss=0.1045, over 28939.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3597, pruned_loss=0.1092, over 5704656.86 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3773, pruned_loss=0.128, over 5694025.02 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.358, pruned_loss=0.1074, over 5703365.11 frames. ], batch size: 164, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:38:47,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3552, 1.7587, 1.3252, 1.5091], device='cuda:0'), covar=tensor([0.0752, 0.0306, 0.0317, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0055, 0.0050, 0.0086], device='cuda:0')
+2023-03-05 11:38:50,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.766e+02 1.032e+03 1.227e+03 1.835e+03 6.051e+03, threshold=2.453e+03, percent-clipped=16.0
+2023-03-05 11:39:21,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2517, 1.6639, 1.2434, 0.4765], device='cuda:0'), covar=tensor([0.2361, 0.1258, 0.1755, 0.3225], device='cuda:0'), in_proj_covar=tensor([0.1540, 0.1455, 0.1463, 0.1247], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 11:39:24,748 INFO [train.py:968] (0/2) Epoch 10, batch 38650, giga_loss[loss=0.2552, simple_loss=0.3346, pruned_loss=0.08791, over 28459.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3584, pruned_loss=0.1081, over 5705691.20 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3766, pruned_loss=0.1276, over 5697870.94 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3573, pruned_loss=0.1068, over 5701367.59 frames. ], batch size: 71, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:39:40,233 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=448263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:39:40,382 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-05 11:39:42,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=448266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:39:56,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7242, 1.6610, 4.7387, 3.3824], device='cuda:0'), covar=tensor([0.1582, 0.2405, 0.0317, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0641, 0.0571, 0.0834, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 11:40:03,574 INFO [train.py:968] (0/2) Epoch 10, batch 38700, giga_loss[loss=0.2715, simple_loss=0.3527, pruned_loss=0.09514, over 28973.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3598, pruned_loss=0.1087, over 5713670.15 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.377, pruned_loss=0.1279, over 5701965.51 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3584, pruned_loss=0.107, over 5706872.84 frames. ], batch size: 164, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:40:04,481 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=448295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:40:09,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.351e+02 9.986e+02 1.292e+03 1.751e+03 8.495e+03, threshold=2.583e+03, percent-clipped=11.0
+2023-03-05 11:40:15,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9409, 2.0328, 1.8380, 1.7263], device='cuda:0'), covar=tensor([0.1523, 0.2045, 0.1818, 0.1998], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0724, 0.0662, 0.0644], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 11:40:42,597 INFO [train.py:968] (0/2) Epoch 10, batch 38750, giga_loss[loss=0.2629, simple_loss=0.3397, pruned_loss=0.09303, over 28918.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3576, pruned_loss=0.1065, over 5717376.14 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1279, over 5703006.68 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3565, pruned_loss=0.1051, over 5711234.19 frames. ], batch size: 112, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:41:20,394 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-05 11:41:24,307 INFO [train.py:968] (0/2) Epoch 10, batch 38800, giga_loss[loss=0.3527, simple_loss=0.3889, pruned_loss=0.1582, over 26704.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3571, pruned_loss=0.1066, over 5701428.75 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3769, pruned_loss=0.1279, over 5697864.76 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3557, pruned_loss=0.105, over 5701399.21 frames. ], batch size: 555, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:41:24,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=448394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:41:28,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.798e+02 9.254e+02 1.144e+03 1.637e+03 5.410e+03, threshold=2.288e+03, percent-clipped=5.0
+2023-03-05 11:42:01,405 INFO [train.py:968] (0/2) Epoch 10, batch 38850, giga_loss[loss=0.3109, simple_loss=0.377, pruned_loss=0.1224, over 28521.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3557, pruned_loss=0.1065, over 5695175.20 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3776, pruned_loss=0.1286, over 5689413.89 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3535, pruned_loss=0.1041, over 5703779.60 frames. ], batch size: 336, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:42:41,733 INFO [train.py:968] (0/2) Epoch 10, batch 38900, libri_loss[loss=0.3077, simple_loss=0.3738, pruned_loss=0.1207, over 27690.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3534, pruned_loss=0.1055, over 5700918.05 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3778, pruned_loss=0.1287, over 5693515.59 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3508, pruned_loss=0.1028, over 5704449.41 frames. ], batch size: 115, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:42:46,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.834e+02 1.120e+03 1.365e+03 1.905e+03 1.132e+04, threshold=2.730e+03, percent-clipped=19.0
+2023-03-05 11:43:13,071 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=448537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:43:13,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6456, 2.3610, 1.6419, 0.7140], device='cuda:0'), covar=tensor([0.4064, 0.1784, 0.3273, 0.4585], device='cuda:0'), in_proj_covar=tensor([0.1524, 0.1441, 0.1454, 0.1234], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 11:43:15,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=448540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:43:18,289 INFO [train.py:968] (0/2) Epoch 10, batch 38950, giga_loss[loss=0.3235, simple_loss=0.3893, pruned_loss=0.1288, over 28666.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3504, pruned_loss=0.1038, over 5706749.61 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3777, pruned_loss=0.1288, over 5696740.34 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.348, pruned_loss=0.1013, over 5706891.12 frames. ], batch size: 336, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:43:41,601 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=448569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:43:41,808 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-05 11:44:03,044 INFO [train.py:968] (0/2) Epoch 10, batch 39000, giga_loss[loss=0.2563, simple_loss=0.3269, pruned_loss=0.09287, over 28583.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3502, pruned_loss=0.1044, over 5699307.27 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3779, pruned_loss=0.1288, over 5691149.98 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3475, pruned_loss=0.1018, over 5704267.52 frames. ], batch size: 78, lr: 3.18e-03, grad_scale: 2.0
+2023-03-05 11:44:03,048 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 11:44:10,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2389, 1.6852, 1.5301, 1.0898], device='cuda:0'), covar=tensor([0.1578, 0.2352, 0.1399, 0.1593], device='cuda:0'), in_proj_covar=tensor([0.0821, 0.0699, 0.0857, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 11:44:11,824 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2204, simple_loss=0.3259, pruned_loss=0.05747, over 944034.00 frames. 
+2023-03-05 11:44:11,824 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 11:44:17,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.313e+02 1.106e+03 1.382e+03 1.716e+03 3.999e+03, threshold=2.765e+03, percent-clipped=7.0
+2023-03-05 11:44:46,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3636, 4.1681, 3.9301, 1.9715], device='cuda:0'), covar=tensor([0.0512, 0.0667, 0.0712, 0.2024], device='cuda:0'), in_proj_covar=tensor([0.1006, 0.0944, 0.0822, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 11:44:48,369 INFO [train.py:968] (0/2) Epoch 10, batch 39050, giga_loss[loss=0.2578, simple_loss=0.3333, pruned_loss=0.09114, over 28867.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3481, pruned_loss=0.1036, over 5701772.05 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3775, pruned_loss=0.1285, over 5687099.67 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3458, pruned_loss=0.1012, over 5708672.11 frames. ], batch size: 174, lr: 3.18e-03, grad_scale: 2.0
+2023-03-05 11:44:57,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-05 11:45:29,890 INFO [train.py:968] (0/2) Epoch 10, batch 39100, giga_loss[loss=0.2423, simple_loss=0.3198, pruned_loss=0.08241, over 28905.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3456, pruned_loss=0.1026, over 5701884.30 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3772, pruned_loss=0.1283, over 5690308.40 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3435, pruned_loss=0.1005, over 5704710.15 frames. ], batch size: 112, lr: 3.18e-03, grad_scale: 2.0
+2023-03-05 11:45:35,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.971e+02 1.010e+03 1.195e+03 1.674e+03 5.322e+03, threshold=2.391e+03, percent-clipped=9.0
+2023-03-05 11:46:07,658 INFO [train.py:968] (0/2) Epoch 10, batch 39150, giga_loss[loss=0.2709, simple_loss=0.329, pruned_loss=0.1063, over 28793.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.343, pruned_loss=0.1017, over 5698575.15 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3775, pruned_loss=0.1286, over 5684958.04 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3405, pruned_loss=0.09938, over 5705130.36 frames. ], batch size: 99, lr: 3.18e-03, grad_scale: 2.0
+2023-03-05 11:46:50,599 INFO [train.py:968] (0/2) Epoch 10, batch 39200, giga_loss[loss=0.2195, simple_loss=0.294, pruned_loss=0.07248, over 28636.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3406, pruned_loss=0.1004, over 5702730.06 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3773, pruned_loss=0.1284, over 5687395.90 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3385, pruned_loss=0.09851, over 5705853.05 frames. ], batch size: 71, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:46:56,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.346e+02 1.028e+03 1.218e+03 1.747e+03 8.819e+03, threshold=2.435e+03, percent-clipped=10.0
+2023-03-05 11:47:05,565 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-05 11:47:29,673 INFO [train.py:968] (0/2) Epoch 10, batch 39250, giga_loss[loss=0.2791, simple_loss=0.3545, pruned_loss=0.1018, over 28818.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3437, pruned_loss=0.1024, over 5711463.27 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.378, pruned_loss=0.1289, over 5693480.30 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3405, pruned_loss=0.09976, over 5708981.01 frames. ], batch size: 199, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:48:11,675 INFO [train.py:968] (0/2) Epoch 10, batch 39300, giga_loss[loss=0.2491, simple_loss=0.3346, pruned_loss=0.0818, over 28897.00 frames. ], tot_loss[loss=0.276, simple_loss=0.346, pruned_loss=0.103, over 5706072.39 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3784, pruned_loss=0.1293, over 5691090.23 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.342, pruned_loss=0.09971, over 5706115.53 frames. ], batch size: 174, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:48:16,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.317e+02 1.076e+03 1.329e+03 2.064e+03 1.836e+04, threshold=2.657e+03, percent-clipped=19.0
+2023-03-05 11:48:19,340 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=448904.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 11:48:58,188 INFO [train.py:968] (0/2) Epoch 10, batch 39350, giga_loss[loss=0.316, simple_loss=0.3809, pruned_loss=0.1256, over 27605.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.349, pruned_loss=0.1039, over 5706987.76 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3786, pruned_loss=0.1293, over 5693329.43 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3453, pruned_loss=0.1009, over 5705178.61 frames. ], batch size: 472, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:49:12,617 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=448962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:49:24,335 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=448977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:49:40,088 INFO [train.py:968] (0/2) Epoch 10, batch 39400, giga_loss[loss=0.3437, simple_loss=0.4163, pruned_loss=0.1355, over 28554.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3524, pruned_loss=0.1054, over 5705635.68 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3793, pruned_loss=0.1299, over 5699417.64 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.348, pruned_loss=0.1018, over 5699148.63 frames. ], batch size: 336, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:49:48,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.265e+02 9.474e+02 1.247e+03 1.581e+03 6.099e+03, threshold=2.495e+03, percent-clipped=3.0
+2023-03-05 11:49:48,790 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=449003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:50:00,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7344, 1.0910, 2.8610, 2.7832], device='cuda:0'), covar=tensor([0.1640, 0.2410, 0.0532, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0638, 0.0573, 0.0834, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 11:50:22,264 INFO [train.py:968] (0/2) Epoch 10, batch 39450, giga_loss[loss=0.2733, simple_loss=0.3545, pruned_loss=0.09602, over 28709.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3512, pruned_loss=0.1041, over 5698374.02 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3791, pruned_loss=0.1299, over 5702738.78 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3472, pruned_loss=0.1006, over 5690125.34 frames. ], batch size: 242, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:50:23,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3733, 2.0579, 1.7924, 1.8219], device='cuda:0'), covar=tensor([0.0712, 0.0802, 0.0892, 0.1136], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0437, 0.0497, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 11:50:37,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4754, 1.7761, 1.4040, 1.7173], device='cuda:0'), covar=tensor([0.2368, 0.2261, 0.2542, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.1276, 0.0946, 0.1130, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 11:51:02,540 INFO [train.py:968] (0/2) Epoch 10, batch 39500, giga_loss[loss=0.2696, simple_loss=0.3365, pruned_loss=0.1013, over 28849.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3522, pruned_loss=0.1043, over 5708593.30 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3797, pruned_loss=0.1303, over 5707036.76 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.348, pruned_loss=0.1007, over 5698019.90 frames. ], batch size: 112, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:51:08,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.338e+02 1.008e+03 1.342e+03 1.853e+03 9.165e+03, threshold=2.683e+03, percent-clipped=9.0
+2023-03-05 11:51:44,545 INFO [train.py:968] (0/2) Epoch 10, batch 39550, giga_loss[loss=0.3332, simple_loss=0.3883, pruned_loss=0.1391, over 28647.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3531, pruned_loss=0.1055, over 5702017.16 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3793, pruned_loss=0.1299, over 5711537.52 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3493, pruned_loss=0.1022, over 5689326.29 frames. ], batch size: 262, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:52:27,390 INFO [train.py:968] (0/2) Epoch 10, batch 39600, giga_loss[loss=0.32, simple_loss=0.378, pruned_loss=0.131, over 28825.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3541, pruned_loss=0.1062, over 5700687.54 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3795, pruned_loss=0.1299, over 5715240.11 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3505, pruned_loss=0.1032, over 5687318.20 frames. ], batch size: 119, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:52:29,721 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=449197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:52:32,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.826e+02 1.195e+03 1.409e+03 2.152e+03 5.996e+03, threshold=2.818e+03, percent-clipped=11.0
+2023-03-05 11:52:49,320 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-05 11:53:09,496 INFO [train.py:968] (0/2) Epoch 10, batch 39650, giga_loss[loss=0.2779, simple_loss=0.3551, pruned_loss=0.1003, over 29073.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3573, pruned_loss=0.108, over 5705679.82 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3796, pruned_loss=0.1299, over 5715955.12 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3538, pruned_loss=0.1051, over 5694448.55 frames. ], batch size: 128, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:53:37,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449279.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 11:53:48,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1423, 1.0548, 4.0431, 3.2324], device='cuda:0'), covar=tensor([0.1682, 0.2767, 0.0368, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.0641, 0.0574, 0.0838, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 11:53:49,316 INFO [train.py:968] (0/2) Epoch 10, batch 39700, giga_loss[loss=0.2829, simple_loss=0.3509, pruned_loss=0.1074, over 28594.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.361, pruned_loss=0.1102, over 5703252.62 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3802, pruned_loss=0.1303, over 5707384.60 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3574, pruned_loss=0.1073, over 5701759.42 frames. ], batch size: 85, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:53:55,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6470, 2.0913, 1.7757, 1.4230], device='cuda:0'), covar=tensor([0.2505, 0.1531, 0.1753, 0.2099], device='cuda:0'), in_proj_covar=tensor([0.1660, 0.1567, 0.1536, 0.1643], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 11:53:56,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.845e+02 1.274e+03 1.690e+03 2.677e+03 7.626e+03, threshold=3.381e+03, percent-clipped=23.0
+2023-03-05 11:54:13,713 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=449324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:54:24,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:54:30,039 INFO [train.py:968] (0/2) Epoch 10, batch 39750, giga_loss[loss=0.2731, simple_loss=0.3533, pruned_loss=0.09641, over 28932.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3611, pruned_loss=0.1095, over 5710960.51 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3801, pruned_loss=0.1303, over 5710241.04 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3581, pruned_loss=0.107, over 5707289.08 frames. ], batch size: 227, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:54:33,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6526, 1.7008, 1.9576, 1.4537], device='cuda:0'), covar=tensor([0.1653, 0.2134, 0.1270, 0.1487], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0696, 0.0851, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 11:54:37,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:54:59,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:55:12,941 INFO [train.py:968] (0/2) Epoch 10, batch 39800, libri_loss[loss=0.2992, simple_loss=0.3688, pruned_loss=0.1148, over 29541.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3612, pruned_loss=0.1096, over 5709235.34 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3805, pruned_loss=0.1305, over 5704608.13 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3581, pruned_loss=0.107, over 5710821.33 frames. ], batch size: 79, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:55:18,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.597e+02 1.246e+03 1.469e+03 2.072e+03 4.408e+03, threshold=2.937e+03, percent-clipped=5.0
+2023-03-05 11:55:24,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2907, 1.2175, 1.1338, 0.9435], device='cuda:0'), covar=tensor([0.0661, 0.0499, 0.0951, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0348, 0.0435, 0.0496, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 11:55:35,496 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449422.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 11:55:37,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449425.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 11:55:51,804 INFO [train.py:968] (0/2) Epoch 10, batch 39850, giga_loss[loss=0.2706, simple_loss=0.34, pruned_loss=0.1006, over 28645.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3611, pruned_loss=0.1097, over 5705034.60 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3808, pruned_loss=0.1307, over 5697615.09 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3582, pruned_loss=0.1073, over 5711781.11 frames. ], batch size: 85, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 11:55:55,851 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-05 11:56:00,151 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449454.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 11:56:21,256 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:22,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:29,739 INFO [train.py:968] (0/2) Epoch 10, batch 39900, giga_loss[loss=0.2774, simple_loss=0.35, pruned_loss=0.1024, over 28801.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3597, pruned_loss=0.1089, over 5711071.94 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3809, pruned_loss=0.1307, over 5704623.67 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3565, pruned_loss=0.1062, over 5710133.30 frames. ], batch size: 186, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 11:56:30,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:32,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:36,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.720e+02 1.114e+03 1.372e+03 1.726e+03 4.210e+03, threshold=2.745e+03, percent-clipped=4.0
+2023-03-05 11:56:42,720 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:49,425 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:51,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:53,777 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:57:08,008 INFO [train.py:968] (0/2) Epoch 10, batch 39950, giga_loss[loss=0.293, simple_loss=0.3588, pruned_loss=0.1136, over 28691.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3577, pruned_loss=0.1081, over 5715797.23 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3807, pruned_loss=0.1306, over 5710476.50 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3547, pruned_loss=0.1053, over 5710247.65 frames. ], batch size: 242, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 11:57:16,327 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:57:32,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:57:51,658 INFO [train.py:968] (0/2) Epoch 10, batch 40000, giga_loss[loss=0.3031, simple_loss=0.37, pruned_loss=0.1181, over 28265.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3526, pruned_loss=0.1048, over 5719815.02 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3807, pruned_loss=0.1306, over 5711299.12 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3501, pruned_loss=0.1025, over 5714771.81 frames. ], batch size: 368, lr: 3.17e-03, grad_scale: 8.0
+2023-03-05 11:58:00,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.566e+02 1.044e+03 1.293e+03 1.646e+03 4.797e+03, threshold=2.585e+03, percent-clipped=10.0
+2023-03-05 11:58:09,716 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5565, 1.8084, 1.4197, 1.2685], device='cuda:0'), covar=tensor([0.1925, 0.1443, 0.1475, 0.1803], device='cuda:0'), in_proj_covar=tensor([0.1653, 0.1567, 0.1540, 0.1637], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 11:58:31,397 INFO [train.py:968] (0/2) Epoch 10, batch 40050, giga_loss[loss=0.2666, simple_loss=0.3533, pruned_loss=0.09, over 28628.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3516, pruned_loss=0.1042, over 5701013.99 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3809, pruned_loss=0.1308, over 5697647.37 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3487, pruned_loss=0.1016, over 5709276.83 frames. ], batch size: 307, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 11:59:11,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2996, 1.9899, 1.4769, 0.4844], device='cuda:0'), covar=tensor([0.3107, 0.1731, 0.2798, 0.4054], device='cuda:0'), in_proj_covar=tensor([0.1544, 0.1450, 0.1468, 0.1248], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 11:59:12,467 INFO [train.py:968] (0/2) Epoch 10, batch 40100, giga_loss[loss=0.3552, simple_loss=0.4185, pruned_loss=0.1459, over 28591.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3539, pruned_loss=0.1042, over 5703000.18 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3811, pruned_loss=0.1308, over 5700289.96 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3511, pruned_loss=0.1017, over 5707208.18 frames. ], batch size: 307, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 11:59:19,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:59:23,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.299e+02 1.068e+03 1.298e+03 1.735e+03 3.644e+03, threshold=2.595e+03, percent-clipped=9.0
+2023-03-05 11:59:30,669 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:59:33,115 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:59:53,944 INFO [train.py:968] (0/2) Epoch 10, batch 40150, giga_loss[loss=0.2957, simple_loss=0.3678, pruned_loss=0.1118, over 27620.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3551, pruned_loss=0.1047, over 5703771.24 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3812, pruned_loss=0.1309, over 5704498.77 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3519, pruned_loss=0.1018, over 5703400.33 frames. ], batch size: 472, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 11:59:56,480 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:00:31,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9783, 1.2779, 1.0221, 0.2066], device='cuda:0'), covar=tensor([0.2356, 0.1917, 0.3243, 0.4425], device='cuda:0'), in_proj_covar=tensor([0.1535, 0.1447, 0.1467, 0.1245], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 12:00:32,647 INFO [train.py:968] (0/2) Epoch 10, batch 40200, giga_loss[loss=0.2378, simple_loss=0.3059, pruned_loss=0.08488, over 28557.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3528, pruned_loss=0.1041, over 5714435.16 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3805, pruned_loss=0.1305, over 5708445.80 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3504, pruned_loss=0.1017, over 5710679.88 frames. ], batch size: 85, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:00:43,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.302e+02 1.113e+03 1.410e+03 1.902e+03 5.243e+03, threshold=2.820e+03, percent-clipped=10.0
+2023-03-05 12:01:10,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8866, 1.0714, 1.0277, 0.8078], device='cuda:0'), covar=tensor([0.1550, 0.1757, 0.0954, 0.1505], device='cuda:0'), in_proj_covar=tensor([0.1659, 0.1564, 0.1542, 0.1640], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 12:01:12,247 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:01:13,296 INFO [train.py:968] (0/2) Epoch 10, batch 40250, giga_loss[loss=0.2817, simple_loss=0.3479, pruned_loss=0.1077, over 28942.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3514, pruned_loss=0.1048, over 5717911.45 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3804, pruned_loss=0.1304, over 5708903.83 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3492, pruned_loss=0.1026, over 5714684.52 frames. ], batch size: 186, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:01:14,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:01:40,927 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:01:58,458 INFO [train.py:968] (0/2) Epoch 10, batch 40300, libri_loss[loss=0.2594, simple_loss=0.3284, pruned_loss=0.09523, over 29667.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3501, pruned_loss=0.1059, over 5701756.19 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3806, pruned_loss=0.1306, over 5698863.45 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3478, pruned_loss=0.1035, over 5707414.45 frames. ], batch size: 69, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:02:02,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5829, 1.7570, 1.8590, 1.3924], device='cuda:0'), covar=tensor([0.1583, 0.2102, 0.1303, 0.1421], device='cuda:0'), in_proj_covar=tensor([0.0817, 0.0695, 0.0851, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 12:02:06,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.777e+02 1.153e+03 1.452e+03 2.382e+03 6.302e+03, threshold=2.904e+03, percent-clipped=18.0
+2023-03-05 12:02:20,790 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=449923.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:02:37,660 INFO [train.py:968] (0/2) Epoch 10, batch 40350, giga_loss[loss=0.2504, simple_loss=0.3252, pruned_loss=0.08776, over 28779.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3501, pruned_loss=0.1071, over 5694502.76 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3806, pruned_loss=0.1306, over 5695152.70 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3473, pruned_loss=0.1046, over 5703158.85 frames. ], batch size: 199, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:03:01,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5610, 1.5088, 1.2279, 1.0869], device='cuda:0'), covar=tensor([0.0792, 0.0622, 0.1071, 0.1079], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0439, 0.0498, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 12:03:21,331 INFO [train.py:968] (0/2) Epoch 10, batch 40400, giga_loss[loss=0.2583, simple_loss=0.3257, pruned_loss=0.09549, over 29084.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3477, pruned_loss=0.1057, over 5691185.71 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3803, pruned_loss=0.1304, over 5692321.46 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3455, pruned_loss=0.1037, over 5700574.16 frames. ], batch size: 128, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:03:25,175 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-450000.pt
+2023-03-05 12:03:28,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.924e+02 1.074e+03 1.224e+03 1.585e+03 4.690e+03, threshold=2.447e+03, percent-clipped=6.0
+2023-03-05 12:03:59,402 INFO [train.py:968] (0/2) Epoch 10, batch 40450, libri_loss[loss=0.3327, simple_loss=0.3971, pruned_loss=0.1342, over 29555.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3445, pruned_loss=0.1039, over 5702468.83 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3805, pruned_loss=0.1305, over 5695843.72 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3419, pruned_loss=0.1017, over 5706847.63 frames. ], batch size: 81, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:04:07,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7731, 1.8134, 1.3607, 1.4225], device='cuda:0'), covar=tensor([0.0726, 0.0622, 0.0967, 0.1077], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0438, 0.0497, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 12:04:10,149 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:04:14,619 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:04:40,827 INFO [train.py:968] (0/2) Epoch 10, batch 40500, giga_loss[loss=0.3254, simple_loss=0.3766, pruned_loss=0.1371, over 26602.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3407, pruned_loss=0.102, over 5694939.83 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3807, pruned_loss=0.1305, over 5689949.20 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3372, pruned_loss=0.09925, over 5704248.49 frames. ], batch size: 555, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:04:49,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.272e+02 1.111e+03 1.390e+03 1.845e+03 3.935e+03, threshold=2.779e+03, percent-clipped=9.0
+2023-03-05 12:05:14,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 12:05:18,278 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-05 12:05:20,552 INFO [train.py:968] (0/2) Epoch 10, batch 40550, giga_loss[loss=0.2301, simple_loss=0.2953, pruned_loss=0.08243, over 28469.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3385, pruned_loss=0.1002, over 5704446.75 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3806, pruned_loss=0.1306, over 5690328.82 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3349, pruned_loss=0.09739, over 5711478.06 frames. ], batch size: 78, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:05:23,360 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 12:06:04,288 INFO [train.py:968] (0/2) Epoch 10, batch 40600, giga_loss[loss=0.3156, simple_loss=0.3802, pruned_loss=0.1255, over 27928.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.341, pruned_loss=0.1009, over 5701364.71 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3808, pruned_loss=0.1307, over 5691795.17 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3377, pruned_loss=0.09834, over 5705742.29 frames. ], batch size: 412, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:06:13,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.402e+02 1.109e+03 1.349e+03 1.692e+03 3.437e+03, threshold=2.698e+03, percent-clipped=2.0
+2023-03-05 12:06:49,334 INFO [train.py:968] (0/2) Epoch 10, batch 40650, giga_loss[loss=0.2595, simple_loss=0.3359, pruned_loss=0.09155, over 28991.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3448, pruned_loss=0.1027, over 5706831.19 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3806, pruned_loss=0.1305, over 5693631.72 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.342, pruned_loss=0.1006, over 5708728.27 frames. ], batch size: 106, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:07:19,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2134, 1.5965, 1.5335, 1.1074], device='cuda:0'), covar=tensor([0.1488, 0.2083, 0.1233, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0696, 0.0850, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 12:07:27,866 INFO [train.py:968] (0/2) Epoch 10, batch 40700, giga_loss[loss=0.2726, simple_loss=0.3377, pruned_loss=0.1037, over 23701.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3481, pruned_loss=0.104, over 5705735.26 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3806, pruned_loss=0.1304, over 5696984.75 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3454, pruned_loss=0.102, over 5704385.89 frames. ], batch size: 705, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:07:31,138 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=450298.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:07:33,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5373, 1.6475, 1.3740, 1.6609], device='cuda:0'), covar=tensor([0.2528, 0.2493, 0.2730, 0.2327], device='cuda:0'), in_proj_covar=tensor([0.1263, 0.0934, 0.1119, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 12:07:36,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.868e+02 1.154e+03 1.489e+03 2.318e+03 9.185e+03, threshold=2.978e+03, percent-clipped=11.0
+2023-03-05 12:07:46,746 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-05 12:08:11,576 INFO [train.py:968] (0/2) Epoch 10, batch 40750, giga_loss[loss=0.2839, simple_loss=0.3586, pruned_loss=0.1046, over 28920.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.352, pruned_loss=0.1058, over 5688703.07 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3809, pruned_loss=0.1307, over 5680287.84 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3493, pruned_loss=0.1037, over 5702661.43 frames. ], batch size: 227, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:08:30,927 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7082, 4.5166, 4.2300, 1.9709], device='cuda:0'), covar=tensor([0.0391, 0.0552, 0.0576, 0.2116], device='cuda:0'), in_proj_covar=tensor([0.1020, 0.0953, 0.0839, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 12:08:54,703 INFO [train.py:968] (0/2) Epoch 10, batch 40800, giga_loss[loss=0.2952, simple_loss=0.3667, pruned_loss=0.1118, over 29039.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3544, pruned_loss=0.1071, over 5699994.09 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3811, pruned_loss=0.1309, over 5687697.99 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3511, pruned_loss=0.1045, over 5705004.38 frames. ], batch size: 128, lr: 3.17e-03, grad_scale: 8.0
+2023-03-05 12:08:54,860 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:09:05,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.052e+02 1.209e+03 1.532e+03 2.182e+03 5.317e+03, threshold=3.063e+03, percent-clipped=16.0
+2023-03-05 12:09:33,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=450430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:09:40,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=450437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:09:44,925 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=450441.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:09:46,998 INFO [train.py:968] (0/2) Epoch 10, batch 40850, giga_loss[loss=0.3577, simple_loss=0.4026, pruned_loss=0.1564, over 29005.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3607, pruned_loss=0.113, over 5700103.16 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3813, pruned_loss=0.131, over 5691200.27 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3575, pruned_loss=0.1103, over 5701103.42 frames. ], batch size: 136, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:09:47,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=450444.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:10:16,756 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:10:16,815 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=450473.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:10:34,565 INFO [train.py:968] (0/2) Epoch 10, batch 40900, giga_loss[loss=0.307, simple_loss=0.3758, pruned_loss=0.1192, over 28853.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3689, pruned_loss=0.1196, over 5696891.69 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3816, pruned_loss=0.1312, over 5692301.94 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3658, pruned_loss=0.1171, over 5696940.44 frames. ], batch size: 199, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:10:47,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.640e+02 1.571e+03 2.074e+03 2.898e+03 6.117e+03, threshold=4.148e+03, percent-clipped=21.0
+2023-03-05 12:11:24,914 INFO [train.py:968] (0/2) Epoch 10, batch 40950, giga_loss[loss=0.4079, simple_loss=0.4325, pruned_loss=0.1917, over 26586.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.375, pruned_loss=0.124, over 5693463.08 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3812, pruned_loss=0.131, over 5694372.91 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3728, pruned_loss=0.1221, over 5691623.00 frames. ], batch size: 555, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:11:53,115 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=450573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:11:55,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=450576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:11:58,099 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=450580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:12:02,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=450583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:12:11,436 INFO [train.py:968] (0/2) Epoch 10, batch 41000, libri_loss[loss=0.3732, simple_loss=0.4197, pruned_loss=0.1633, over 29538.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3808, pruned_loss=0.129, over 5696474.86 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3816, pruned_loss=0.1313, over 5696592.38 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3787, pruned_loss=0.1273, over 5692915.34 frames. ], batch size: 82, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:12:23,253 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=450605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:12:23,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.204e+03 1.653e+03 1.998e+03 2.456e+03 4.467e+03, threshold=3.996e+03, percent-clipped=2.0
+2023-03-05 12:12:30,181 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=450612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:12:45,851 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:12:58,790 INFO [train.py:968] (0/2) Epoch 10, batch 41050, giga_loss[loss=0.3606, simple_loss=0.415, pruned_loss=0.1531, over 28533.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3881, pruned_loss=0.1352, over 5694297.37 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3815, pruned_loss=0.1312, over 5698702.01 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3866, pruned_loss=0.1339, over 5689520.18 frames. ], batch size: 336, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:13:52,612 INFO [train.py:968] (0/2) Epoch 10, batch 41100, giga_loss[loss=0.3369, simple_loss=0.3944, pruned_loss=0.1396, over 28845.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.392, pruned_loss=0.1398, over 5668479.59 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3814, pruned_loss=0.1312, over 5701949.29 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.391, pruned_loss=0.1389, over 5661537.37 frames. ], batch size: 99, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:14:04,059 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.108e+02 1.833e+03 2.211e+03 3.083e+03 6.584e+03, threshold=4.421e+03, percent-clipped=16.0
+2023-03-05 12:14:43,806 INFO [train.py:968] (0/2) Epoch 10, batch 41150, giga_loss[loss=0.4261, simple_loss=0.4473, pruned_loss=0.2025, over 28232.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3945, pruned_loss=0.1428, over 5659681.80 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3819, pruned_loss=0.1315, over 5693240.37 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3935, pruned_loss=0.142, over 5661753.05 frames. ], batch size: 368, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:15:10,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=450769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:15:34,498 INFO [train.py:968] (0/2) Epoch 10, batch 41200, giga_loss[loss=0.3493, simple_loss=0.3973, pruned_loss=0.1506, over 28527.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.398, pruned_loss=0.1471, over 5644511.24 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.382, pruned_loss=0.1315, over 5697681.45 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.3976, pruned_loss=0.1469, over 5640961.51 frames. ], batch size: 60, lr: 3.17e-03, grad_scale: 8.0
+2023-03-05 12:15:47,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.153e+03 1.883e+03 2.709e+03 4.251e+03 1.261e+04, threshold=5.419e+03, percent-clipped=23.0
+2023-03-05 12:16:24,864 INFO [train.py:968] (0/2) Epoch 10, batch 41250, giga_loss[loss=0.4149, simple_loss=0.4378, pruned_loss=0.196, over 27952.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4016, pruned_loss=0.1503, over 5646101.77 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.382, pruned_loss=0.1315, over 5701493.18 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4016, pruned_loss=0.1505, over 5638798.78 frames. ], batch size: 412, lr: 3.17e-03, grad_scale: 8.0
+2023-03-05 12:16:28,898 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3930, 1.8038, 1.2870, 0.9172], device='cuda:0'), covar=tensor([0.3345, 0.2329, 0.1880, 0.3209], device='cuda:0'), in_proj_covar=tensor([0.1552, 0.1475, 0.1491, 0.1274], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 12:16:31,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=450848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:16:55,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8815, 1.8797, 1.8214, 1.7234], device='cuda:0'), covar=tensor([0.1303, 0.1909, 0.1790, 0.1712], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0731, 0.0669, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 12:17:18,295 INFO [train.py:968] (0/2) Epoch 10, batch 41300, giga_loss[loss=0.2989, simple_loss=0.3699, pruned_loss=0.1139, over 28962.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.4024, pruned_loss=0.1514, over 5625423.10 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3818, pruned_loss=0.1314, over 5695511.85 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.403, pruned_loss=0.152, over 5622692.66 frames. ], batch size: 164, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:17:19,798 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450895.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:17:35,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.804e+03 2.343e+03 3.075e+03 7.183e+03, threshold=4.686e+03, percent-clipped=7.0
+2023-03-05 12:17:40,724 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=450912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:17:43,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=450915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:18:10,522 INFO [train.py:968] (0/2) Epoch 10, batch 41350, libri_loss[loss=0.3517, simple_loss=0.4052, pruned_loss=0.1491, over 29525.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.4013, pruned_loss=0.1512, over 5625607.65 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.382, pruned_loss=0.1315, over 5692777.33 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4019, pruned_loss=0.152, over 5624359.43 frames. ], batch size: 83, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:18:11,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=450944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:18:58,338 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=450991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:19:01,162 INFO [train.py:968] (0/2) Epoch 10, batch 41400, giga_loss[loss=0.3663, simple_loss=0.4134, pruned_loss=0.1596, over 28269.00 frames. ], tot_loss[loss=0.3491, simple_loss=0.3995, pruned_loss=0.1494, over 5646332.78 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.382, pruned_loss=0.1314, over 5695486.67 frames. ], giga_tot_loss[loss=0.3505, simple_loss=0.4003, pruned_loss=0.1503, over 5642101.74 frames. ], batch size: 368, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:19:01,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=450994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:19:11,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=451004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:19:14,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.191e+02 1.772e+03 2.309e+03 3.209e+03 5.923e+03, threshold=4.619e+03, percent-clipped=6.0
+2023-03-05 12:19:29,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=451023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:19:50,914 INFO [train.py:968] (0/2) Epoch 10, batch 41450, giga_loss[loss=0.3517, simple_loss=0.407, pruned_loss=0.1482, over 28948.00 frames. ], tot_loss[loss=0.347, simple_loss=0.399, pruned_loss=0.1475, over 5629555.13 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.382, pruned_loss=0.1314, over 5678757.45 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4001, pruned_loss=0.1487, over 5638980.20 frames. ], batch size: 106, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:20:22,303 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-05 12:20:34,945 INFO [train.py:968] (0/2) Epoch 10, batch 41500, libri_loss[loss=0.2492, simple_loss=0.3218, pruned_loss=0.08832, over 29350.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3964, pruned_loss=0.1444, over 5656232.00 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3804, pruned_loss=0.1304, over 5687755.27 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.3995, pruned_loss=0.147, over 5654203.04 frames. ], batch size: 71, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:20:46,443 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.661e+03 2.086e+03 2.808e+03 6.867e+03, threshold=4.173e+03, percent-clipped=3.0
+2023-03-05 12:21:23,296 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=451142.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:21:24,921 INFO [train.py:968] (0/2) Epoch 10, batch 41550, giga_loss[loss=0.315, simple_loss=0.378, pruned_loss=0.1259, over 28861.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3987, pruned_loss=0.1462, over 5640802.62 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3807, pruned_loss=0.1307, over 5689972.49 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4013, pruned_loss=0.1484, over 5636165.46 frames. ], batch size: 186, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:21:29,350 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=451147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:21:32,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=451150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:21:53,946 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-05 12:21:59,609 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=451179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:22:12,484 INFO [train.py:968] (0/2) Epoch 10, batch 41600, giga_loss[loss=0.3363, simple_loss=0.4039, pruned_loss=0.1344, over 28616.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3967, pruned_loss=0.1437, over 5650544.36 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3805, pruned_loss=0.1305, over 5695828.92 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3996, pruned_loss=0.1461, over 5640395.01 frames. ], batch size: 262, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:22:25,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.732e+02 1.590e+03 2.217e+03 3.381e+03 7.909e+03, threshold=4.435e+03, percent-clipped=14.0
+2023-03-05 12:22:59,616 INFO [train.py:968] (0/2) Epoch 10, batch 41650, giga_loss[loss=0.3456, simple_loss=0.4091, pruned_loss=0.141, over 28798.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3942, pruned_loss=0.1405, over 5653780.25 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3804, pruned_loss=0.1304, over 5695378.24 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3971, pruned_loss=0.1429, over 5644800.17 frames. ], batch size: 284, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:23:06,799 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3538, 2.0593, 1.5505, 0.4851], device='cuda:0'), covar=tensor([0.3362, 0.1882, 0.2695, 0.4418], device='cuda:0'), in_proj_covar=tensor([0.1546, 0.1471, 0.1483, 0.1267], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 12:23:17,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=451262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:23:24,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=451270.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:23:47,815 INFO [train.py:968] (0/2) Epoch 10, batch 41700, giga_loss[loss=0.2934, simple_loss=0.3592, pruned_loss=0.1138, over 28454.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3903, pruned_loss=0.1367, over 5658192.51 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3802, pruned_loss=0.1304, over 5690787.53 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.393, pruned_loss=0.1388, over 5653554.99 frames. ], batch size: 60, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:24:01,457 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.651e+03 2.011e+03 2.802e+03 1.404e+04, threshold=4.022e+03, percent-clipped=11.0
+2023-03-05 12:24:36,348 INFO [train.py:968] (0/2) Epoch 10, batch 41750, giga_loss[loss=0.3607, simple_loss=0.4058, pruned_loss=0.1578, over 27969.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3877, pruned_loss=0.1347, over 5656920.63 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3803, pruned_loss=0.1305, over 5693825.12 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.39, pruned_loss=0.1365, over 5650038.35 frames. ], batch size: 412, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:25:25,590 INFO [train.py:968] (0/2) Epoch 10, batch 41800, giga_loss[loss=0.4064, simple_loss=0.4428, pruned_loss=0.185, over 28764.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3852, pruned_loss=0.1334, over 5651294.81 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3802, pruned_loss=0.1304, over 5696935.15 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3872, pruned_loss=0.1348, over 5642467.70 frames. ], batch size: 99, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:25:29,908 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6905, 1.4735, 1.7794, 1.3117], device='cuda:0'), covar=tensor([0.1488, 0.2391, 0.1202, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0817, 0.0701, 0.0850, 0.0761], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 12:25:41,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.500e+02 1.436e+03 1.838e+03 2.639e+03 6.828e+03, threshold=3.677e+03, percent-clipped=8.0
+2023-03-05 12:25:48,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=451413.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:25:50,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=451416.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:26:15,463 INFO [train.py:968] (0/2) Epoch 10, batch 41850, giga_loss[loss=0.3131, simple_loss=0.3806, pruned_loss=0.1229, over 28682.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3859, pruned_loss=0.1334, over 5660513.89 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3798, pruned_loss=0.1302, over 5688846.56 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3878, pruned_loss=0.1348, over 5659979.29 frames. ], batch size: 242, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:26:17,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=451445.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:26:20,459 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 12:26:51,990 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3896, 2.1164, 1.6039, 1.4918], device='cuda:0'), covar=tensor([0.0785, 0.0264, 0.0306, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 12:27:02,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3536, 1.5864, 1.2887, 1.9069], device='cuda:0'), covar=tensor([0.2277, 0.2185, 0.2333, 0.1808], device='cuda:0'), in_proj_covar=tensor([0.1272, 0.0949, 0.1128, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 12:27:02,488 INFO [train.py:968] (0/2) Epoch 10, batch 41900, giga_loss[loss=0.3213, simple_loss=0.3794, pruned_loss=0.1317, over 28288.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3843, pruned_loss=0.132, over 5659764.70 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3801, pruned_loss=0.1305, over 5684572.83 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3858, pruned_loss=0.1329, over 5661561.74 frames. ], batch size: 368, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:27:22,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.742e+02 1.289e+03 1.742e+03 2.322e+03 1.080e+04, threshold=3.484e+03, percent-clipped=8.0
+2023-03-05 12:27:32,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=451517.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:27:37,973 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3162, 3.5470, 1.4664, 1.4663], device='cuda:0'), covar=tensor([0.1181, 0.0388, 0.1006, 0.1547], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0507, 0.0336, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 12:27:59,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3019, 1.8798, 1.4472, 1.4700], device='cuda:0'), covar=tensor([0.0773, 0.0286, 0.0313, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 12:28:02,756 INFO [train.py:968] (0/2) Epoch 10, batch 41950, giga_loss[loss=0.3818, simple_loss=0.4286, pruned_loss=0.1675, over 26709.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3827, pruned_loss=0.1287, over 5664871.18 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3802, pruned_loss=0.1305, over 5686787.50 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3838, pruned_loss=0.1294, over 5664080.66 frames. ], batch size: 555, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:28:53,575 INFO [train.py:968] (0/2) Epoch 10, batch 42000, libri_loss[loss=0.2881, simple_loss=0.3443, pruned_loss=0.116, over 29385.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.383, pruned_loss=0.1271, over 5661434.57 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3798, pruned_loss=0.1305, over 5678780.81 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3843, pruned_loss=0.1275, over 5667859.66 frames. ], batch size: 67, lr: 3.17e-03, grad_scale: 8.0
+2023-03-05 12:28:53,581 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 12:29:01,941 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2138, simple_loss=0.3185, pruned_loss=0.05458, over 944034.00 frames. 
+2023-03-05 12:29:01,941 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 12:29:15,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.719e+03 2.186e+03 3.065e+03 6.999e+03, threshold=4.372e+03, percent-clipped=17.0
+2023-03-05 12:29:44,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=451637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:29:50,416 INFO [train.py:968] (0/2) Epoch 10, batch 42050, giga_loss[loss=0.3203, simple_loss=0.3792, pruned_loss=0.1307, over 28768.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3835, pruned_loss=0.128, over 5662909.06 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3795, pruned_loss=0.1302, over 5681987.35 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3849, pruned_loss=0.1286, over 5664627.26 frames. ], batch size: 92, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:30:06,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=451660.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:30:08,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=451663.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:30:19,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 12:30:35,790 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=451692.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:30:36,869 INFO [train.py:968] (0/2) Epoch 10, batch 42100, giga_loss[loss=0.3545, simple_loss=0.4059, pruned_loss=0.1515, over 28278.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3857, pruned_loss=0.1306, over 5667467.88 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3793, pruned_loss=0.1302, over 5684903.33 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.387, pruned_loss=0.1311, over 5666022.02 frames. ], batch size: 368, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:30:49,130 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.134e+03 1.693e+03 2.166e+03 3.200e+03 9.447e+03, threshold=4.332e+03, percent-clipped=13.0
+2023-03-05 12:31:01,840 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=451723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:31:19,984 INFO [train.py:968] (0/2) Epoch 10, batch 42150, giga_loss[loss=0.3071, simple_loss=0.3664, pruned_loss=0.1239, over 28668.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3838, pruned_loss=0.13, over 5673797.52 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3791, pruned_loss=0.1301, over 5683038.40 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3852, pruned_loss=0.1304, over 5673631.32 frames. ], batch size: 242, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:31:39,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6711, 1.8991, 1.9463, 1.4845], device='cuda:0'), covar=tensor([0.1467, 0.1998, 0.1149, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0704, 0.0854, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 12:31:57,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=451780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:32:01,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=451783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:32:10,453 INFO [train.py:968] (0/2) Epoch 10, batch 42200, giga_loss[loss=0.2988, simple_loss=0.3549, pruned_loss=0.1214, over 28655.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3819, pruned_loss=0.13, over 5670057.64 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3794, pruned_loss=0.1302, over 5686333.48 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3829, pruned_loss=0.1303, over 5666795.09 frames. ], batch size: 85, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:32:19,215 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0790, 1.1803, 3.5154, 2.9849], device='cuda:0'), covar=tensor([0.1702, 0.2547, 0.0491, 0.1120], device='cuda:0'), in_proj_covar=tensor([0.0655, 0.0583, 0.0856, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 12:32:24,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.640e+03 2.205e+03 2.943e+03 5.516e+03, threshold=4.410e+03, percent-clipped=7.0
+2023-03-05 12:32:28,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=451812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:32:58,137 INFO [train.py:968] (0/2) Epoch 10, batch 42250, giga_loss[loss=0.3054, simple_loss=0.3689, pruned_loss=0.121, over 28566.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3807, pruned_loss=0.1291, over 5669604.36 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3795, pruned_loss=0.1302, over 5688964.01 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3814, pruned_loss=0.1293, over 5664132.48 frames. ], batch size: 336, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:33:33,944 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2752, 1.2325, 1.0699, 0.9574], device='cuda:0'), covar=tensor([0.0790, 0.0562, 0.1079, 0.0950], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0442, 0.0499, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 12:33:35,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4031, 1.8327, 1.6731, 1.5262], device='cuda:0'), covar=tensor([0.0741, 0.0290, 0.0288, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 12:33:44,745 INFO [train.py:968] (0/2) Epoch 10, batch 42300, giga_loss[loss=0.2999, simple_loss=0.3768, pruned_loss=0.1115, over 28763.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3805, pruned_loss=0.1275, over 5682613.50 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3794, pruned_loss=0.1301, over 5693812.35 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3812, pruned_loss=0.1277, over 5673566.05 frames. ], batch size: 119, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:33:57,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.853e+02 1.365e+03 1.932e+03 2.933e+03 6.280e+03, threshold=3.864e+03, percent-clipped=4.0
+2023-03-05 12:34:06,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1864, 1.0108, 4.1906, 3.2967], device='cuda:0'), covar=tensor([0.1738, 0.2740, 0.0446, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0653, 0.0582, 0.0855, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 12:34:30,419 INFO [train.py:968] (0/2) Epoch 10, batch 42350, giga_loss[loss=0.357, simple_loss=0.4121, pruned_loss=0.151, over 28698.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3808, pruned_loss=0.1272, over 5685308.05 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3796, pruned_loss=0.1302, over 5699149.23 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3812, pruned_loss=0.1272, over 5673043.95 frames. ], batch size: 262, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:34:45,367 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=451956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:35:19,672 INFO [train.py:968] (0/2) Epoch 10, batch 42400, libri_loss[loss=0.3237, simple_loss=0.3842, pruned_loss=0.1316, over 29532.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3803, pruned_loss=0.1267, over 5697160.08 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3802, pruned_loss=0.1307, over 5702583.80 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3801, pruned_loss=0.1261, over 5684092.38 frames. ], batch size: 82, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:35:27,167 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-452000.pt
+2023-03-05 12:35:35,579 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.753e+02 1.633e+03 2.280e+03 3.367e+03 1.310e+04, threshold=4.560e+03, percent-clipped=16.0
+2023-03-05 12:36:06,242 INFO [train.py:968] (0/2) Epoch 10, batch 42450, giga_loss[loss=0.3096, simple_loss=0.3722, pruned_loss=0.1235, over 28603.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3778, pruned_loss=0.1254, over 5683712.85 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3804, pruned_loss=0.1307, over 5699033.94 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3773, pruned_loss=0.1249, over 5676093.87 frames. ], batch size: 307, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:36:36,472 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:36:48,816 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-05 12:36:53,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4663, 1.6276, 1.7471, 1.3914], device='cuda:0'), covar=tensor([0.1199, 0.1720, 0.1001, 0.1228], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0706, 0.0857, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 12:36:53,505 INFO [train.py:968] (0/2) Epoch 10, batch 42500, giga_loss[loss=0.3379, simple_loss=0.3783, pruned_loss=0.1488, over 23494.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3774, pruned_loss=0.126, over 5671089.48 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3802, pruned_loss=0.1305, over 5693166.91 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3772, pruned_loss=0.1256, over 5669107.59 frames. ], batch size: 705, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:36:56,582 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:36:58,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:37:11,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.482e+02 1.651e+03 2.120e+03 2.838e+03 5.616e+03, threshold=4.241e+03, percent-clipped=2.0
+2023-03-05 12:37:33,378 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4965, 4.3282, 4.1098, 1.8451], device='cuda:0'), covar=tensor([0.0500, 0.0641, 0.0679, 0.2061], device='cuda:0'), in_proj_covar=tensor([0.1038, 0.0976, 0.0852, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 12:37:44,534 INFO [train.py:968] (0/2) Epoch 10, batch 42550, giga_loss[loss=0.2767, simple_loss=0.3402, pruned_loss=0.1066, over 28520.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3768, pruned_loss=0.127, over 5668712.39 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3801, pruned_loss=0.1306, over 5696353.27 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3766, pruned_loss=0.1267, over 5664047.16 frames. ], batch size: 78, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:38:01,382 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-05 12:38:29,291 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 12:38:32,055 INFO [train.py:968] (0/2) Epoch 10, batch 42600, libri_loss[loss=0.3138, simple_loss=0.3705, pruned_loss=0.1285, over 29542.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.376, pruned_loss=0.1274, over 5668096.14 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3793, pruned_loss=0.1301, over 5691478.86 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3765, pruned_loss=0.1274, over 5668564.89 frames. ], batch size: 77, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:38:46,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.632e+03 2.045e+03 2.825e+03 8.733e+03, threshold=4.090e+03, percent-clipped=8.0
+2023-03-05 12:39:18,597 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:39:20,989 INFO [train.py:968] (0/2) Epoch 10, batch 42650, giga_loss[loss=0.3642, simple_loss=0.4093, pruned_loss=0.1595, over 28332.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3755, pruned_loss=0.1273, over 5677019.06 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3793, pruned_loss=0.1301, over 5693578.81 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3759, pruned_loss=0.1273, over 5675493.75 frames. ], batch size: 369, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:39:21,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:39:50,648 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:39:58,091 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9985, 1.1586, 1.2025, 1.0009], device='cuda:0'), covar=tensor([0.1382, 0.1532, 0.0840, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.1672, 0.1583, 0.1549, 0.1664], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 12:40:11,172 INFO [train.py:968] (0/2) Epoch 10, batch 42700, giga_loss[loss=0.2699, simple_loss=0.3531, pruned_loss=0.09333, over 28917.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3766, pruned_loss=0.1282, over 5673286.20 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3798, pruned_loss=0.1305, over 5687382.90 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3764, pruned_loss=0.1278, over 5677538.10 frames. ], batch size: 145, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:40:20,969 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452305.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:40:25,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.509e+02 1.622e+03 2.176e+03 2.709e+03 9.285e+03, threshold=4.352e+03, percent-clipped=9.0
+2023-03-05 12:40:46,611 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:40:57,069 INFO [train.py:968] (0/2) Epoch 10, batch 42750, libri_loss[loss=0.3171, simple_loss=0.3753, pruned_loss=0.1295, over 29564.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3761, pruned_loss=0.1269, over 5680425.05 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3795, pruned_loss=0.1304, over 5691593.39 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.376, pruned_loss=0.1265, over 5679601.52 frames. ], batch size: 77, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:41:04,054 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-05 12:41:13,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6660, 1.0536, 2.8183, 2.6778], device='cuda:0'), covar=tensor([0.1822, 0.2493, 0.0613, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0652, 0.0581, 0.0854, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 12:41:40,694 INFO [train.py:968] (0/2) Epoch 10, batch 42800, giga_loss[loss=0.3055, simple_loss=0.3717, pruned_loss=0.1197, over 28572.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3762, pruned_loss=0.1262, over 5684854.26 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.379, pruned_loss=0.1302, over 5695706.56 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3765, pruned_loss=0.126, over 5679997.62 frames. ], batch size: 307, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:41:57,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.616e+02 1.431e+03 1.937e+03 2.983e+03 8.680e+03, threshold=3.873e+03, percent-clipped=9.0
+2023-03-05 12:42:04,577 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452419.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:42:25,438 INFO [train.py:968] (0/2) Epoch 10, batch 42850, giga_loss[loss=0.3169, simple_loss=0.3837, pruned_loss=0.1251, over 29012.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3756, pruned_loss=0.125, over 5683908.37 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3786, pruned_loss=0.1298, over 5703191.55 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3762, pruned_loss=0.125, over 5672765.37 frames. ], batch size: 155, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:42:32,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:42:57,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:42:58,517 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:43:00,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:43:19,853 INFO [train.py:968] (0/2) Epoch 10, batch 42900, libri_loss[loss=0.2602, simple_loss=0.332, pruned_loss=0.09416, over 29349.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3768, pruned_loss=0.1266, over 5673660.55 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3783, pruned_loss=0.1295, over 5704161.45 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3775, pruned_loss=0.1268, over 5663378.81 frames. ], batch size: 71, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:43:31,361 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:43:37,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.784e+02 1.535e+03 2.088e+03 2.780e+03 9.375e+03, threshold=4.177e+03, percent-clipped=9.0
+2023-03-05 12:44:07,778 INFO [train.py:968] (0/2) Epoch 10, batch 42950, giga_loss[loss=0.4602, simple_loss=0.4755, pruned_loss=0.2225, over 26589.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.379, pruned_loss=0.1292, over 5672974.99 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3783, pruned_loss=0.1297, over 5707773.52 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3796, pruned_loss=0.1293, over 5660772.57 frames. ], batch size: 555, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:44:10,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0009, 3.7964, 3.6041, 1.8276], device='cuda:0'), covar=tensor([0.0610, 0.0758, 0.0741, 0.2052], device='cuda:0'), in_proj_covar=tensor([0.1048, 0.0985, 0.0865, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 12:44:56,111 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:44:56,514 INFO [train.py:968] (0/2) Epoch 10, batch 43000, libri_loss[loss=0.3658, simple_loss=0.42, pruned_loss=0.1558, over 29529.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3806, pruned_loss=0.1319, over 5661995.74 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3779, pruned_loss=0.1294, over 5705796.89 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3816, pruned_loss=0.1323, over 5652278.23 frames. ], batch size: 82, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:44:58,165 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:45:14,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.763e+03 2.316e+03 2.956e+03 5.312e+03, threshold=4.632e+03, percent-clipped=5.0
+2023-03-05 12:45:18,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:45:21,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:45:28,705 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:45:44,896 INFO [train.py:968] (0/2) Epoch 10, batch 43050, giga_loss[loss=0.2931, simple_loss=0.3614, pruned_loss=0.1124, over 29111.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3809, pruned_loss=0.1333, over 5659029.59 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3774, pruned_loss=0.1293, over 5701367.62 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3823, pruned_loss=0.1338, over 5653284.48 frames. ], batch size: 128, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:45:47,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:46:04,640 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:46:18,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452680.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:46:32,739 INFO [train.py:968] (0/2) Epoch 10, batch 43100, giga_loss[loss=0.3136, simple_loss=0.3861, pruned_loss=0.1205, over 28731.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3818, pruned_loss=0.134, over 5667459.91 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3767, pruned_loss=0.1287, over 5704771.79 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3837, pruned_loss=0.1352, over 5658405.77 frames. ], batch size: 243, lr: 3.16e-03, grad_scale: 2.0
+2023-03-05 12:46:50,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.604e+03 2.046e+03 2.709e+03 5.655e+03, threshold=4.093e+03, percent-clipped=4.0
+2023-03-05 12:47:15,546 INFO [train.py:968] (0/2) Epoch 10, batch 43150, giga_loss[loss=0.3326, simple_loss=0.389, pruned_loss=0.1381, over 28039.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3813, pruned_loss=0.1332, over 5669878.15 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3772, pruned_loss=0.1289, over 5701155.52 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3825, pruned_loss=0.1341, over 5664465.86 frames. ], batch size: 412, lr: 3.16e-03, grad_scale: 2.0
+2023-03-05 12:47:21,818 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:47:50,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8029, 4.6090, 4.3518, 1.9574], device='cuda:0'), covar=tensor([0.0604, 0.0866, 0.0947, 0.2126], device='cuda:0'), in_proj_covar=tensor([0.1052, 0.0987, 0.0865, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 12:48:03,679 INFO [train.py:968] (0/2) Epoch 10, batch 43200, giga_loss[loss=0.2972, simple_loss=0.3644, pruned_loss=0.115, over 28924.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3796, pruned_loss=0.1311, over 5673715.38 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3769, pruned_loss=0.1286, over 5704980.18 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3808, pruned_loss=0.1321, over 5665404.95 frames. ], batch size: 119, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:48:04,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452794.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:48:12,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:48:16,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4917, 1.6340, 1.3191, 1.4804], device='cuda:0'), covar=tensor([0.0753, 0.0304, 0.0326, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 12:48:17,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.096e+02 1.717e+03 2.112e+03 3.051e+03 1.852e+04, threshold=4.223e+03, percent-clipped=7.0
+2023-03-05 12:48:28,031 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452823.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:48:30,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452826.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:48:46,014 INFO [train.py:968] (0/2) Epoch 10, batch 43250, libri_loss[loss=0.2656, simple_loss=0.3394, pruned_loss=0.09592, over 29533.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.378, pruned_loss=0.1284, over 5689399.13 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3774, pruned_loss=0.1289, over 5711334.31 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3786, pruned_loss=0.129, over 5676080.37 frames. ], batch size: 80, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:48:56,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452855.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:49:33,400 INFO [train.py:968] (0/2) Epoch 10, batch 43300, giga_loss[loss=0.3192, simple_loss=0.3559, pruned_loss=0.1413, over 23892.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3759, pruned_loss=0.1275, over 5671914.56 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3778, pruned_loss=0.1292, over 5713801.16 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.376, pruned_loss=0.1277, over 5658680.98 frames. ], batch size: 705, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:49:48,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.431e+02 1.433e+03 1.870e+03 2.570e+03 6.243e+03, threshold=3.739e+03, percent-clipped=10.0
+2023-03-05 12:50:11,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452937.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:50:14,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452940.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:50:19,759 INFO [train.py:968] (0/2) Epoch 10, batch 43350, giga_loss[loss=0.3074, simple_loss=0.3453, pruned_loss=0.1348, over 23801.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.375, pruned_loss=0.1277, over 5668410.99 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3783, pruned_loss=0.1295, over 5707934.08 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3746, pruned_loss=0.1275, over 5663117.04 frames. ], batch size: 705, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:50:40,229 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452969.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:50:49,144 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1291, 1.3734, 1.4217, 1.3160], device='cuda:0'), covar=tensor([0.1138, 0.0870, 0.1207, 0.1003], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0728, 0.0668, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 12:51:02,098 INFO [train.py:968] (0/2) Epoch 10, batch 43400, giga_loss[loss=0.3195, simple_loss=0.3835, pruned_loss=0.1278, over 28910.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3749, pruned_loss=0.1282, over 5652101.64 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3777, pruned_loss=0.1291, over 5696127.18 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3749, pruned_loss=0.1284, over 5656792.02 frames. ], batch size: 145, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:51:12,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-05 12:51:18,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.621e+03 2.182e+03 3.081e+03 1.113e+04, threshold=4.364e+03, percent-clipped=15.0
+2023-03-05 12:51:46,437 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:51:49,045 INFO [train.py:968] (0/2) Epoch 10, batch 43450, giga_loss[loss=0.3239, simple_loss=0.3909, pruned_loss=0.1285, over 29062.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3782, pruned_loss=0.1301, over 5665721.78 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.378, pruned_loss=0.1291, over 5700048.95 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3779, pruned_loss=0.1301, over 5665352.54 frames. ], batch size: 128, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:52:30,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1381, 1.1060, 3.9687, 3.1994], device='cuda:0'), covar=tensor([0.1737, 0.2741, 0.0420, 0.0820], device='cuda:0'), in_proj_covar=tensor([0.0653, 0.0583, 0.0851, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 12:52:35,509 INFO [train.py:968] (0/2) Epoch 10, batch 43500, giga_loss[loss=0.3301, simple_loss=0.3769, pruned_loss=0.1416, over 23956.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3819, pruned_loss=0.1304, over 5663485.74 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3777, pruned_loss=0.1291, over 5700262.54 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.382, pruned_loss=0.1306, over 5662365.79 frames. ], batch size: 705, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:52:54,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.554e+02 1.547e+03 2.062e+03 2.746e+03 8.711e+03, threshold=4.123e+03, percent-clipped=6.0
+2023-03-05 12:53:00,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 12:53:08,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:53:30,986 INFO [train.py:968] (0/2) Epoch 10, batch 43550, giga_loss[loss=0.3082, simple_loss=0.3779, pruned_loss=0.1192, over 28869.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3832, pruned_loss=0.1295, over 5660006.06 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3774, pruned_loss=0.1288, over 5704647.75 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3836, pruned_loss=0.1299, over 5654586.29 frames. ], batch size: 186, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:53:46,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 12:54:07,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:54:08,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:54:10,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:54:16,189 INFO [train.py:968] (0/2) Epoch 10, batch 43600, giga_loss[loss=0.3082, simple_loss=0.3817, pruned_loss=0.1174, over 28990.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3855, pruned_loss=0.1314, over 5674121.42 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3767, pruned_loss=0.1283, over 5708676.55 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3867, pruned_loss=0.1321, over 5664837.05 frames. ], batch size: 164, lr: 3.16e-03, grad_scale: 8.0
+2023-03-05 12:54:32,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.524e+02 1.668e+03 2.334e+03 3.477e+03 7.805e+03, threshold=4.668e+03, percent-clipped=15.0
+2023-03-05 12:54:36,280 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:55:01,971 INFO [train.py:968] (0/2) Epoch 10, batch 43650, giga_loss[loss=0.371, simple_loss=0.421, pruned_loss=0.1605, over 28752.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3876, pruned_loss=0.1335, over 5656891.34 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.377, pruned_loss=0.1285, over 5689397.58 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3887, pruned_loss=0.1341, over 5664322.78 frames. ], batch size: 284, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:55:05,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2937, 1.4295, 1.5039, 1.3238], device='cuda:0'), covar=tensor([0.1214, 0.1376, 0.1680, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0728, 0.0666, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 12:55:26,468 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:55:28,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:55:41,160 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:55:46,636 INFO [train.py:968] (0/2) Epoch 10, batch 43700, giga_loss[loss=0.3096, simple_loss=0.3747, pruned_loss=0.1223, over 28440.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3867, pruned_loss=0.1335, over 5668475.90 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3765, pruned_loss=0.1281, over 5692093.78 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3883, pruned_loss=0.1346, over 5671039.21 frames. ], batch size: 85, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:55:52,013 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:56:01,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:56:02,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.772e+02 1.524e+03 1.920e+03 2.359e+03 6.358e+03, threshold=3.840e+03, percent-clipped=6.0
+2023-03-05 12:56:10,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.2324, 6.0325, 5.7043, 3.1372], device='cuda:0'), covar=tensor([0.0383, 0.0580, 0.0617, 0.1618], device='cuda:0'), in_proj_covar=tensor([0.1044, 0.0983, 0.0861, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 12:56:13,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4600, 1.5667, 1.5842, 1.5219], device='cuda:0'), covar=tensor([0.1146, 0.1474, 0.1365, 0.1365], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0728, 0.0665, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 12:56:13,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:56:19,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:56:32,854 INFO [train.py:968] (0/2) Epoch 10, batch 43750, giga_loss[loss=0.3312, simple_loss=0.39, pruned_loss=0.1362, over 28591.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3862, pruned_loss=0.1341, over 5667373.65 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3765, pruned_loss=0.1279, over 5698237.96 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3877, pruned_loss=0.1352, over 5663340.01 frames. ], batch size: 307, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:56:34,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4179, 1.6024, 1.7070, 1.2925], device='cuda:0'), covar=tensor([0.1274, 0.2117, 0.1091, 0.1375], device='cuda:0'), in_proj_covar=tensor([0.0820, 0.0704, 0.0855, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 12:56:42,277 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:56:46,931 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:57:00,524 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:57:06,520 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 12:57:12,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4172, 1.6019, 1.3376, 1.5437], device='cuda:0'), covar=tensor([0.0684, 0.0307, 0.0296, 0.0751], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 12:57:18,817 INFO [train.py:968] (0/2) Epoch 10, batch 43800, giga_loss[loss=0.3241, simple_loss=0.3817, pruned_loss=0.1332, over 28904.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.384, pruned_loss=0.1333, over 5661097.01 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3764, pruned_loss=0.128, over 5690316.07 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3856, pruned_loss=0.1343, over 5663451.49 frames. ], batch size: 213, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:57:36,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.194e+02 1.813e+03 2.465e+03 3.642e+03 6.965e+03, threshold=4.929e+03, percent-clipped=21.0
+2023-03-05 12:58:06,554 INFO [train.py:968] (0/2) Epoch 10, batch 43850, giga_loss[loss=0.2969, simple_loss=0.3708, pruned_loss=0.1115, over 29025.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.383, pruned_loss=0.1331, over 5664832.20 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3764, pruned_loss=0.1278, over 5693634.10 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3844, pruned_loss=0.1342, over 5663083.45 frames. ], batch size: 164, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:58:59,679 INFO [train.py:968] (0/2) Epoch 10, batch 43900, giga_loss[loss=0.3703, simple_loss=0.407, pruned_loss=0.1668, over 26540.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3833, pruned_loss=0.1343, over 5653074.41 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3758, pruned_loss=0.1274, over 5696058.69 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.385, pruned_loss=0.1356, over 5649250.28 frames. ], batch size: 555, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:59:17,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.174e+02 1.808e+03 2.226e+03 2.927e+03 6.536e+03, threshold=4.452e+03, percent-clipped=3.0
+2023-03-05 12:59:50,034 INFO [train.py:968] (0/2) Epoch 10, batch 43950, giga_loss[loss=0.4181, simple_loss=0.4274, pruned_loss=0.2044, over 23887.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3819, pruned_loss=0.1334, over 5644506.94 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3762, pruned_loss=0.1276, over 5691727.42 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3831, pruned_loss=0.1344, over 5644312.58 frames. ], batch size: 705, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:00:32,649 INFO [train.py:968] (0/2) Epoch 10, batch 44000, giga_loss[loss=0.3458, simple_loss=0.4005, pruned_loss=0.1455, over 28672.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.381, pruned_loss=0.133, over 5658414.20 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3761, pruned_loss=0.1274, over 5694189.58 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3822, pruned_loss=0.1341, over 5654998.89 frames. ], batch size: 307, lr: 3.16e-03, grad_scale: 8.0
+2023-03-05 13:00:52,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.736e+03 2.221e+03 3.106e+03 6.160e+03, threshold=4.442e+03, percent-clipped=10.0
+2023-03-05 13:01:04,711 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3064, 1.2455, 1.1356, 1.4644], device='cuda:0'), covar=tensor([0.0737, 0.0341, 0.0328, 0.0788], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 13:01:20,778 INFO [train.py:968] (0/2) Epoch 10, batch 44050, giga_loss[loss=0.3037, simple_loss=0.3637, pruned_loss=0.1218, over 28434.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3796, pruned_loss=0.1321, over 5657722.62 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.376, pruned_loss=0.1274, over 5695309.64 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3806, pruned_loss=0.1329, over 5653766.23 frames. ], batch size: 71, lr: 3.16e-03, grad_scale: 8.0
+2023-03-05 13:01:37,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:01:39,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6336, 1.7136, 1.2499, 1.3836], device='cuda:0'), covar=tensor([0.0876, 0.0647, 0.1128, 0.1070], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0448, 0.0499, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 13:01:56,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2692, 1.3971, 1.3603, 1.1551], device='cuda:0'), covar=tensor([0.1880, 0.1691, 0.1203, 0.1607], device='cuda:0'), in_proj_covar=tensor([0.1677, 0.1590, 0.1561, 0.1663], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 13:01:58,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:02:07,682 INFO [train.py:968] (0/2) Epoch 10, batch 44100, giga_loss[loss=0.3115, simple_loss=0.3849, pruned_loss=0.119, over 28981.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3802, pruned_loss=0.1314, over 5660623.58 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3754, pruned_loss=0.1269, over 5701836.84 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3817, pruned_loss=0.1328, over 5650273.45 frames. ], batch size: 164, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:02:28,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.675e+02 1.557e+03 1.947e+03 2.629e+03 4.464e+03, threshold=3.894e+03, percent-clipped=1.0
+2023-03-05 13:02:37,305 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-05 13:02:42,374 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:02:47,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6095, 1.6343, 1.7355, 1.4524], device='cuda:0'), covar=tensor([0.1471, 0.1955, 0.1859, 0.1898], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0737, 0.0675, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 13:02:56,811 INFO [train.py:968] (0/2) Epoch 10, batch 44150, giga_loss[loss=0.285, simple_loss=0.3577, pruned_loss=0.1061, over 28583.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3842, pruned_loss=0.1339, over 5664106.66 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3763, pruned_loss=0.1276, over 5704539.59 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3847, pruned_loss=0.1344, over 5653047.19 frames. ], batch size: 78, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:03:03,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:03:07,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4561, 4.3111, 4.0824, 1.9953], device='cuda:0'), covar=tensor([0.0485, 0.0625, 0.0662, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.1047, 0.0983, 0.0862, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 13:03:46,059 INFO [train.py:968] (0/2) Epoch 10, batch 44200, giga_loss[loss=0.3101, simple_loss=0.3744, pruned_loss=0.1229, over 28914.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3828, pruned_loss=0.1335, over 5662842.48 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3759, pruned_loss=0.1273, over 5696586.04 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3838, pruned_loss=0.1343, over 5659775.39 frames. ], batch size: 199, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:03:58,103 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:00,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:03,825 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.461e+02 1.452e+03 1.944e+03 2.965e+03 6.010e+03, threshold=3.888e+03, percent-clipped=10.0
+2023-03-05 13:04:18,851 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:21,709 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:21,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:26,012 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:27,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4261, 1.8085, 1.5969, 1.3477], device='cuda:0'), covar=tensor([0.2564, 0.1725, 0.1393, 0.2044], device='cuda:0'), in_proj_covar=tensor([0.1672, 0.1585, 0.1558, 0.1658], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 13:04:32,265 INFO [train.py:968] (0/2) Epoch 10, batch 44250, giga_loss[loss=0.2913, simple_loss=0.3773, pruned_loss=0.1026, over 29053.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3841, pruned_loss=0.1316, over 5668091.25 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3756, pruned_loss=0.1271, over 5696932.52 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3853, pruned_loss=0.1325, over 5664397.57 frames. ], batch size: 136, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:04:34,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:45,727 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:55,302 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:57,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:05:11,611 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:05:13,020 INFO [train.py:968] (0/2) Epoch 10, batch 44300, giga_loss[loss=0.3604, simple_loss=0.4176, pruned_loss=0.1515, over 28930.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3853, pruned_loss=0.1304, over 5668572.48 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1272, over 5700230.69 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3865, pruned_loss=0.1311, over 5661818.78 frames. ], batch size: 199, lr: 3.16e-03, grad_scale: 2.0
+2023-03-05 13:05:15,786 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:05:21,867 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:05:29,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.649e+02 1.397e+03 1.926e+03 2.690e+03 9.664e+03, threshold=3.853e+03, percent-clipped=5.0
+2023-03-05 13:05:38,436 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:05:53,429 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6894, 1.9816, 1.7162, 1.8071], device='cuda:0'), covar=tensor([0.1370, 0.1504, 0.1798, 0.1603], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0731, 0.0668, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 13:05:59,605 INFO [train.py:968] (0/2) Epoch 10, batch 44350, giga_loss[loss=0.423, simple_loss=0.4583, pruned_loss=0.1939, over 27910.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3882, pruned_loss=0.1323, over 5655036.55 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3759, pruned_loss=0.1273, over 5694232.21 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3894, pruned_loss=0.133, over 5653418.84 frames. ], batch size: 412, lr: 3.16e-03, grad_scale: 2.0
+2023-03-05 13:06:15,415 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-05 13:06:48,863 INFO [train.py:968] (0/2) Epoch 10, batch 44400, giga_loss[loss=0.2942, simple_loss=0.3638, pruned_loss=0.1123, over 28749.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.392, pruned_loss=0.1359, over 5658385.43 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3759, pruned_loss=0.1274, over 5696171.93 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3931, pruned_loss=0.1365, over 5654974.71 frames. ], batch size: 99, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:06:54,892 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-454000.pt
+2023-03-05 13:07:03,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5041, 1.9658, 1.5421, 1.8396], device='cuda:0'), covar=tensor([0.0704, 0.0251, 0.0299, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 13:07:08,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.600e+02 1.612e+03 2.134e+03 3.073e+03 6.295e+03, threshold=4.268e+03, percent-clipped=11.0
+2023-03-05 13:07:40,967 INFO [train.py:968] (0/2) Epoch 10, batch 44450, giga_loss[loss=0.3145, simple_loss=0.3822, pruned_loss=0.1234, over 28943.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3935, pruned_loss=0.1386, over 5659326.22 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3758, pruned_loss=0.1275, over 5699074.52 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3947, pruned_loss=0.1391, over 5653610.15 frames. ], batch size: 174, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:08:06,312 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:08:26,748 INFO [train.py:968] (0/2) Epoch 10, batch 44500, giga_loss[loss=0.3009, simple_loss=0.3665, pruned_loss=0.1177, over 28503.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3916, pruned_loss=0.1375, over 5673393.40 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3759, pruned_loss=0.1276, over 5701479.00 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3928, pruned_loss=0.1381, over 5665856.58 frames. ], batch size: 71, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:08:44,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.104e+02 1.518e+03 1.853e+03 2.620e+03 6.749e+03, threshold=3.706e+03, percent-clipped=7.0
+2023-03-05 13:09:07,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7396, 2.0422, 1.6062, 1.9098], device='cuda:0'), covar=tensor([0.0705, 0.0257, 0.0284, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 13:09:13,802 INFO [train.py:968] (0/2) Epoch 10, batch 44550, giga_loss[loss=0.3365, simple_loss=0.3899, pruned_loss=0.1416, over 27956.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.39, pruned_loss=0.1364, over 5665974.21 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3755, pruned_loss=0.1273, over 5703621.39 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3914, pruned_loss=0.1372, over 5658011.56 frames. ], batch size: 412, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:09:16,658 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:09:30,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4681, 1.7483, 1.7086, 1.5102], device='cuda:0'), covar=tensor([0.1536, 0.1757, 0.1817, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0732, 0.0666, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 13:09:45,816 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 13:09:57,335 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-05 13:09:58,392 INFO [train.py:968] (0/2) Epoch 10, batch 44600, giga_loss[loss=0.2962, simple_loss=0.3762, pruned_loss=0.1081, over 28818.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3885, pruned_loss=0.1328, over 5674479.47 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3757, pruned_loss=0.1275, over 5698642.70 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3897, pruned_loss=0.1333, over 5671300.04 frames. ], batch size: 119, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:10:12,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:10:18,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.157e+02 1.589e+03 2.371e+03 3.073e+03 7.254e+03, threshold=4.741e+03, percent-clipped=15.0
+2023-03-05 13:10:24,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:10:43,002 INFO [train.py:968] (0/2) Epoch 10, batch 44650, libri_loss[loss=0.2958, simple_loss=0.3521, pruned_loss=0.1198, over 29577.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3878, pruned_loss=0.1313, over 5683760.55 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3752, pruned_loss=0.1272, over 5702520.58 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3897, pruned_loss=0.1322, over 5676720.11 frames. ], batch size: 74, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:10:43,175 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:10:50,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:11:34,753 INFO [train.py:968] (0/2) Epoch 10, batch 44700, giga_loss[loss=0.4691, simple_loss=0.4625, pruned_loss=0.2378, over 23334.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3891, pruned_loss=0.1331, over 5665731.82 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3752, pruned_loss=0.1271, over 5704947.97 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3909, pruned_loss=0.134, over 5657537.03 frames. ], batch size: 705, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:11:56,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.644e+02 1.660e+03 2.283e+03 3.945e+03 1.378e+04, threshold=4.566e+03, percent-clipped=17.0
+2023-03-05 13:12:21,184 INFO [train.py:968] (0/2) Epoch 10, batch 44750, giga_loss[loss=0.2826, simple_loss=0.3606, pruned_loss=0.1023, over 28900.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3875, pruned_loss=0.1325, over 5658016.50 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3749, pruned_loss=0.127, over 5697052.46 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3895, pruned_loss=0.1335, over 5657565.32 frames. ], batch size: 174, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:12:28,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:12:30,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:12:39,052 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:12:41,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:12:46,490 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:12:56,846 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:13:03,357 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2742, 2.7382, 1.4750, 1.3637], device='cuda:0'), covar=tensor([0.0857, 0.0388, 0.0754, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0510, 0.0337, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 13:13:08,243 INFO [train.py:968] (0/2) Epoch 10, batch 44800, giga_loss[loss=0.2798, simple_loss=0.3528, pruned_loss=0.1034, over 28904.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3854, pruned_loss=0.1315, over 5664521.87 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3749, pruned_loss=0.1268, over 5698992.49 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3873, pruned_loss=0.1325, over 5661464.79 frames. ], batch size: 174, lr: 3.16e-03, grad_scale: 8.0
+2023-03-05 13:13:10,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:13:11,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6699, 1.6189, 1.2048, 1.2982], device='cuda:0'), covar=tensor([0.0733, 0.0580, 0.1013, 0.0992], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0445, 0.0498, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 13:13:30,266 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.628e+02 1.496e+03 2.101e+03 3.162e+03 7.830e+03, threshold=4.202e+03, percent-clipped=6.0
+2023-03-05 13:13:34,814 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:13:56,195 INFO [train.py:968] (0/2) Epoch 10, batch 44850, giga_loss[loss=0.3022, simple_loss=0.3636, pruned_loss=0.1204, over 28638.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.384, pruned_loss=0.1318, over 5664101.84 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3747, pruned_loss=0.1268, over 5704184.93 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3859, pruned_loss=0.1327, over 5656183.97 frames. ], batch size: 307, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:14:00,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:14:33,382 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:14:43,611 INFO [train.py:968] (0/2) Epoch 10, batch 44900, giga_loss[loss=0.3104, simple_loss=0.3692, pruned_loss=0.1258, over 28744.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3815, pruned_loss=0.1305, over 5662017.36 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3744, pruned_loss=0.1266, over 5697383.38 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3834, pruned_loss=0.1315, over 5660190.05 frames. ], batch size: 262, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:15:03,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.656e+02 1.549e+03 2.111e+03 2.844e+03 7.364e+03, threshold=4.221e+03, percent-clipped=8.0
+2023-03-05 13:15:08,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:15:27,232 INFO [train.py:968] (0/2) Epoch 10, batch 44950, giga_loss[loss=0.3034, simple_loss=0.3688, pruned_loss=0.1189, over 28615.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3811, pruned_loss=0.1314, over 5661537.77 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3738, pruned_loss=0.1262, over 5703254.94 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3835, pruned_loss=0.1327, over 5653593.91 frames. ], batch size: 262, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:15:41,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2767, 1.6057, 1.3189, 1.4320], device='cuda:0'), covar=tensor([0.0743, 0.0321, 0.0306, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0057, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 13:16:13,682 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:14,831 INFO [train.py:968] (0/2) Epoch 10, batch 45000, giga_loss[loss=0.3336, simple_loss=0.3896, pruned_loss=0.1388, over 28361.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3808, pruned_loss=0.1317, over 5651755.76 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3743, pruned_loss=0.1265, over 5702644.63 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3823, pruned_loss=0.1326, over 5645124.94 frames. ], batch size: 71, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:16:14,836 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 13:16:18,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2800, 1.8798, 1.4252, 0.3956], device='cuda:0'), covar=tensor([0.3063, 0.2274, 0.3448, 0.4200], device='cuda:0'), in_proj_covar=tensor([0.1538, 0.1465, 0.1475, 0.1266], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 13:16:23,769 INFO [train.py:1012] (0/2) Epoch 10, validation: loss=0.2189, simple_loss=0.3261, pruned_loss=0.05584, over 944034.00 frames. 
+2023-03-05 13:16:23,770 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 13:16:24,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:39,127 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:42,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.579e+03 1.907e+03 2.463e+03 5.397e+03, threshold=3.814e+03, percent-clipped=5.0
+2023-03-05 13:16:45,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:48,849 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:50,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:17:06,582 INFO [train.py:968] (0/2) Epoch 10, batch 45050, giga_loss[loss=0.31, simple_loss=0.3783, pruned_loss=0.1209, over 28854.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3793, pruned_loss=0.1296, over 5640066.23 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3747, pruned_loss=0.1269, over 5688561.11 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3803, pruned_loss=0.1301, over 5645789.30 frames. ], batch size: 186, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:17:28,242 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:17:31,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:17:50,755 INFO [train.py:968] (0/2) Epoch 10, batch 45100, giga_loss[loss=0.2586, simple_loss=0.3373, pruned_loss=0.08994, over 28886.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3752, pruned_loss=0.1254, over 5647856.71 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3749, pruned_loss=0.1271, over 5690575.36 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3759, pruned_loss=0.1256, over 5649176.36 frames. ], batch size: 227, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:17:53,029 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:17:59,210 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3142, 1.3618, 1.1875, 1.3063], device='cuda:0'), covar=tensor([0.1437, 0.1260, 0.1261, 0.1232], device='cuda:0'), in_proj_covar=tensor([0.1668, 0.1584, 0.1549, 0.1651], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 13:18:06,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5178, 1.8131, 1.8413, 1.3741], device='cuda:0'), covar=tensor([0.1629, 0.2166, 0.1276, 0.1496], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0703, 0.0858, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 13:18:10,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.961e+02 1.334e+03 1.733e+03 2.314e+03 6.226e+03, threshold=3.465e+03, percent-clipped=7.0
+2023-03-05 13:18:40,275 INFO [train.py:968] (0/2) Epoch 10, batch 45150, giga_loss[loss=0.2841, simple_loss=0.3597, pruned_loss=0.1042, over 28750.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.374, pruned_loss=0.1245, over 5645850.71 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3748, pruned_loss=0.1269, over 5692664.36 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3747, pruned_loss=0.1247, over 5644255.67 frames. ], batch size: 284, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:18:42,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:18:50,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3269, 3.0860, 1.3653, 1.4391], device='cuda:0'), covar=tensor([0.0905, 0.0319, 0.0855, 0.1304], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0510, 0.0337, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 13:18:55,888 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:18:56,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5526, 1.6326, 1.6128, 1.4170], device='cuda:0'), covar=tensor([0.1554, 0.2228, 0.2008, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0738, 0.0672, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 13:18:57,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:02,168 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:05,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:24,965 INFO [train.py:968] (0/2) Epoch 10, batch 45200, giga_loss[loss=0.3603, simple_loss=0.4042, pruned_loss=0.1582, over 27907.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3724, pruned_loss=0.1237, over 5666591.18 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3748, pruned_loss=0.1269, over 5698048.20 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3728, pruned_loss=0.1238, over 5659538.03 frames. ], batch size: 412, lr: 3.16e-03, grad_scale: 8.0
+2023-03-05 13:19:25,180 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:25,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:28,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4232, 3.4060, 1.5838, 1.4955], device='cuda:0'), covar=tensor([0.0881, 0.0254, 0.0810, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0510, 0.0337, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 13:19:33,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:51,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.817e+02 1.636e+03 2.313e+03 3.505e+03 8.597e+03, threshold=4.627e+03, percent-clipped=27.0
+2023-03-05 13:19:55,301 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-05 13:20:17,588 INFO [train.py:968] (0/2) Epoch 10, batch 45250, giga_loss[loss=0.3813, simple_loss=0.4216, pruned_loss=0.1705, over 28751.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3707, pruned_loss=0.1235, over 5673539.26 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3749, pruned_loss=0.1269, over 5699081.96 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.371, pruned_loss=0.1235, over 5667047.70 frames. ], batch size: 284, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:20:30,171 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:20:59,094 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:01,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:03,120 INFO [train.py:968] (0/2) Epoch 10, batch 45300, giga_loss[loss=0.2757, simple_loss=0.352, pruned_loss=0.09969, over 28833.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3712, pruned_loss=0.1233, over 5674532.20 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3748, pruned_loss=0.1269, over 5689335.93 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3714, pruned_loss=0.1232, over 5678213.06 frames. ], batch size: 119, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:21:23,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.534e+03 1.881e+03 2.627e+03 7.857e+03, threshold=3.763e+03, percent-clipped=7.0
+2023-03-05 13:21:27,019 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:43,802 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:44,599 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:46,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:48,193 INFO [train.py:968] (0/2) Epoch 10, batch 45350, giga_loss[loss=0.3055, simple_loss=0.3748, pruned_loss=0.118, over 28699.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3741, pruned_loss=0.125, over 5667893.72 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.375, pruned_loss=0.127, over 5691078.69 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3741, pruned_loss=0.1249, over 5669003.15 frames. ], batch size: 262, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:22:14,286 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:22:30,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:22:35,532 INFO [train.py:968] (0/2) Epoch 10, batch 45400, giga_loss[loss=0.3519, simple_loss=0.405, pruned_loss=0.1494, over 28674.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3751, pruned_loss=0.126, over 5668010.59 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3748, pruned_loss=0.1269, over 5694482.98 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3752, pruned_loss=0.126, over 5665362.39 frames. ], batch size: 284, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:22:42,567 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:22:44,964 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:22:55,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.484e+03 1.970e+03 2.511e+03 6.636e+03, threshold=3.940e+03, percent-clipped=7.0
+2023-03-05 13:23:08,894 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:23:19,144 INFO [train.py:968] (0/2) Epoch 10, batch 45450, giga_loss[loss=0.3564, simple_loss=0.414, pruned_loss=0.1493, over 28307.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3765, pruned_loss=0.1274, over 5667247.56 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3753, pruned_loss=0.1272, over 5697418.83 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.1271, over 5662015.72 frames. ], batch size: 368, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:23:53,706 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5275, 1.7386, 1.8125, 1.3749], device='cuda:0'), covar=tensor([0.1454, 0.2104, 0.1164, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0825, 0.0706, 0.0862, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 13:24:03,409 INFO [train.py:968] (0/2) Epoch 10, batch 45500, giga_loss[loss=0.2827, simple_loss=0.3516, pruned_loss=0.1069, over 28584.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3773, pruned_loss=0.1282, over 5649453.67 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3756, pruned_loss=0.1274, over 5685723.05 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3768, pruned_loss=0.1278, over 5654862.16 frames. ], batch size: 60, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:24:22,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:24:24,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.621e+03 2.179e+03 3.206e+03 1.523e+04, threshold=4.357e+03, percent-clipped=17.0
+2023-03-05 13:24:27,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3698, 1.7961, 1.4311, 1.5525], device='cuda:0'), covar=tensor([0.0627, 0.0250, 0.0275, 0.0641], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 13:24:39,245 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:24:41,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:24:45,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8135, 1.8907, 2.1225, 1.6228], device='cuda:0'), covar=tensor([0.1652, 0.2194, 0.1256, 0.1549], device='cuda:0'), in_proj_covar=tensor([0.0826, 0.0708, 0.0862, 0.0769], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 13:24:50,410 INFO [train.py:968] (0/2) Epoch 10, batch 45550, giga_loss[loss=0.3036, simple_loss=0.3744, pruned_loss=0.1163, over 28957.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3796, pruned_loss=0.1295, over 5630247.95 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3763, pruned_loss=0.1278, over 5678227.51 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3786, pruned_loss=0.1289, over 5640679.72 frames. ], batch size: 112, lr: 3.15e-03, grad_scale: 2.0
+2023-03-05 13:25:07,921 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:25:35,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455193.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 13:25:36,028 INFO [train.py:968] (0/2) Epoch 10, batch 45600, giga_loss[loss=0.3144, simple_loss=0.3724, pruned_loss=0.1282, over 28798.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.381, pruned_loss=0.1309, over 5641549.68 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3763, pruned_loss=0.1278, over 5683869.34 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3803, pruned_loss=0.1304, over 5643434.09 frames. ], batch size: 92, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:25:40,270 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:26:00,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.597e+02 1.596e+03 2.208e+03 3.063e+03 7.987e+03, threshold=4.417e+03, percent-clipped=9.0
+2023-03-05 13:26:13,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0357, 4.8354, 4.5746, 2.3472], device='cuda:0'), covar=tensor([0.0458, 0.0612, 0.0668, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.1056, 0.0988, 0.0869, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 13:26:27,850 INFO [train.py:968] (0/2) Epoch 10, batch 45650, giga_loss[loss=0.2879, simple_loss=0.3565, pruned_loss=0.1097, over 28770.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3816, pruned_loss=0.1313, over 5648818.05 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3765, pruned_loss=0.1279, over 5687222.41 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.381, pruned_loss=0.1309, over 5646645.15 frames. ], batch size: 119, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:26:55,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7223, 2.1740, 2.4394, 2.0496], device='cuda:0'), covar=tensor([0.1153, 0.2139, 0.1502, 0.1763], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0733, 0.0669, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 13:27:13,972 INFO [train.py:968] (0/2) Epoch 10, batch 45700, giga_loss[loss=0.2958, simple_loss=0.3705, pruned_loss=0.1106, over 28846.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3822, pruned_loss=0.1313, over 5656580.85 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.376, pruned_loss=0.1275, over 5695636.16 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3823, pruned_loss=0.1315, over 5646085.38 frames. ], batch size: 186, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:27:31,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:27:37,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.590e+03 2.192e+03 2.762e+03 9.100e+03, threshold=4.384e+03, percent-clipped=8.0
+2023-03-05 13:28:09,933 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:28:10,344 INFO [train.py:968] (0/2) Epoch 10, batch 45750, giga_loss[loss=0.263, simple_loss=0.3561, pruned_loss=0.085, over 28591.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3815, pruned_loss=0.1297, over 5661197.24 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3759, pruned_loss=0.1275, over 5699131.56 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3818, pruned_loss=0.1299, over 5649392.71 frames. ], batch size: 336, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:28:31,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2511, 1.4337, 1.2544, 1.4555], device='cuda:0'), covar=tensor([0.0775, 0.0330, 0.0326, 0.0808], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 13:28:33,134 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-05 13:28:57,723 INFO [train.py:968] (0/2) Epoch 10, batch 45800, giga_loss[loss=0.2693, simple_loss=0.3433, pruned_loss=0.0977, over 29031.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3815, pruned_loss=0.1303, over 5640873.96 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3758, pruned_loss=0.1276, over 5674542.03 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3819, pruned_loss=0.1304, over 5653138.66 frames. ], batch size: 155, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:29:21,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.042e+02 1.529e+03 1.928e+03 2.698e+03 1.003e+04, threshold=3.856e+03, percent-clipped=8.0
+2023-03-05 13:29:28,140 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-05 13:29:45,974 INFO [train.py:968] (0/2) Epoch 10, batch 45850, giga_loss[loss=0.3349, simple_loss=0.3911, pruned_loss=0.1394, over 28564.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3821, pruned_loss=0.1315, over 5619667.51 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3764, pruned_loss=0.1282, over 5639560.49 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3819, pruned_loss=0.131, over 5660611.76 frames. ], batch size: 307, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:29:59,618 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:30:01,695 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:30:35,793 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:30:39,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:30:41,478 INFO [train.py:968] (0/2) Epoch 10, batch 45900, giga_loss[loss=0.2984, simple_loss=0.3633, pruned_loss=0.1168, over 28810.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3808, pruned_loss=0.131, over 5579904.57 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.377, pruned_loss=0.1286, over 5587267.45 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3802, pruned_loss=0.1303, over 5656916.66 frames. ], batch size: 186, lr: 3.15e-03, grad_scale: 2.0
+2023-03-05 13:31:07,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.672e+03 2.360e+03 4.017e+03 1.842e+04, threshold=4.720e+03, percent-clipped=27.0
+2023-03-05 13:31:30,723 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-05 13:31:32,269 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-10.pt
+2023-03-05 13:32:31,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455568.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 13:32:35,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:32:40,253 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-05 13:32:50,179 INFO [train.py:968] (0/2) Epoch 11, batch 50, giga_loss[loss=0.2919, simple_loss=0.3739, pruned_loss=0.105, over 28695.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3814, pruned_loss=0.1157, over 1266427.11 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3418, pruned_loss=0.09055, over 174544.49 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3872, pruned_loss=0.1193, over 1125835.44 frames. ], batch size: 242, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:32:59,272 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-05 13:33:05,974 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:33:16,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.295e+02 1.221e+03 1.515e+03 1.929e+03 5.658e+03, threshold=3.030e+03, percent-clipped=1.0
+2023-03-05 13:33:30,231 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:33:34,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:33:37,271 INFO [train.py:968] (0/2) Epoch 11, batch 100, giga_loss[loss=0.2421, simple_loss=0.3215, pruned_loss=0.08135, over 28422.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3719, pruned_loss=0.1109, over 2250110.05 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3458, pruned_loss=0.09331, over 344302.86 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3757, pruned_loss=0.1134, over 2026738.42 frames. ], batch size: 71, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:33:52,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2407, 1.4741, 1.4090, 1.3785], device='cuda:0'), covar=tensor([0.1220, 0.1248, 0.1693, 0.1271], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0731, 0.0660, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 13:34:01,566 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:34:11,360 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-05 13:34:21,685 INFO [train.py:968] (0/2) Epoch 11, batch 150, libri_loss[loss=0.2628, simple_loss=0.353, pruned_loss=0.08626, over 29513.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.359, pruned_loss=0.1051, over 3016320.28 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3561, pruned_loss=0.1005, over 454402.15 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3595, pruned_loss=0.1057, over 2780978.73 frames. ], batch size: 84, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:34:39,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8051, 1.7137, 1.3062, 1.5131], device='cuda:0'), covar=tensor([0.0756, 0.0725, 0.1002, 0.1227], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0442, 0.0494, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 13:34:42,409 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455711.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:34:45,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455714.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 13:34:46,699 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:34:47,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:34:49,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.897e+02 1.022e+03 1.320e+03 1.747e+03 4.788e+03, threshold=2.639e+03, percent-clipped=4.0
+2023-03-05 13:34:49,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:35:03,977 INFO [train.py:968] (0/2) Epoch 11, batch 200, libri_loss[loss=0.2703, simple_loss=0.3511, pruned_loss=0.09474, over 29527.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3465, pruned_loss=0.09897, over 3620431.41 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3572, pruned_loss=0.1007, over 615979.41 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3455, pruned_loss=0.0989, over 3362064.81 frames. ], batch size: 83, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:35:06,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455743.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:35:10,099 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5717, 1.8725, 1.4892, 2.1990], device='cuda:0'), covar=tensor([0.2438, 0.2392, 0.2629, 0.2106], device='cuda:0'), in_proj_covar=tensor([0.1286, 0.0955, 0.1139, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 13:35:10,685 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:35:26,531 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:35:45,554 INFO [train.py:968] (0/2) Epoch 11, batch 250, libri_loss[loss=0.2773, simple_loss=0.3607, pruned_loss=0.09693, over 29275.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3347, pruned_loss=0.0928, over 4086861.67 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3553, pruned_loss=0.09925, over 772434.57 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.333, pruned_loss=0.09248, over 3825198.43 frames. ], batch size: 94, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:35:58,144 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 13:36:07,671 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-05 13:36:10,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.408e+02 9.513e+02 1.287e+03 1.585e+03 3.537e+03, threshold=2.574e+03, percent-clipped=4.0
+2023-03-05 13:36:29,743 INFO [train.py:968] (0/2) Epoch 11, batch 300, giga_loss[loss=0.2321, simple_loss=0.3079, pruned_loss=0.07816, over 29140.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3251, pruned_loss=0.08883, over 4446552.56 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3558, pruned_loss=0.09954, over 823888.92 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.323, pruned_loss=0.08824, over 4226170.45 frames. ], batch size: 128, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:36:48,914 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:36:51,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:37:15,825 INFO [train.py:968] (0/2) Epoch 11, batch 350, giga_loss[loss=0.2139, simple_loss=0.2912, pruned_loss=0.06826, over 29003.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3173, pruned_loss=0.08519, over 4726260.87 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3528, pruned_loss=0.09734, over 947944.08 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.315, pruned_loss=0.08465, over 4522798.64 frames. ], batch size: 136, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:37:16,794 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.45 vs. limit=5.0
+2023-03-05 13:37:18,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:37:40,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1878, 1.6477, 1.2350, 0.4995], device='cuda:0'), covar=tensor([0.2627, 0.1500, 0.1975, 0.3667], device='cuda:0'), in_proj_covar=tensor([0.1523, 0.1461, 0.1464, 0.1251], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 13:37:41,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.140e+02 9.736e+02 1.264e+03 1.670e+03 3.410e+03, threshold=2.527e+03, percent-clipped=2.0
+2023-03-05 13:37:47,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7422, 2.1114, 1.8122, 1.9547], device='cuda:0'), covar=tensor([0.0721, 0.0261, 0.0283, 0.0745], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 13:37:59,489 INFO [train.py:968] (0/2) Epoch 11, batch 400, giga_loss[loss=0.2088, simple_loss=0.2861, pruned_loss=0.06578, over 28543.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3116, pruned_loss=0.08244, over 4942579.75 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3516, pruned_loss=0.09667, over 996459.44 frames. ], giga_tot_loss[loss=0.2366, simple_loss=0.3094, pruned_loss=0.08192, over 4773986.20 frames. ], batch size: 71, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:38:02,345 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:38:18,614 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0052, 1.3388, 1.0857, 0.1567], device='cuda:0'), covar=tensor([0.2737, 0.2497, 0.3793, 0.4684], device='cuda:0'), in_proj_covar=tensor([0.1521, 0.1458, 0.1464, 0.1253], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 13:38:20,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8952, 3.6798, 3.4899, 1.6421], device='cuda:0'), covar=tensor([0.0701, 0.0934, 0.0904, 0.2285], device='cuda:0'), in_proj_covar=tensor([0.1040, 0.0977, 0.0858, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 13:38:34,347 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:38:40,568 INFO [train.py:968] (0/2) Epoch 11, batch 450, giga_loss[loss=0.2415, simple_loss=0.3116, pruned_loss=0.08573, over 28704.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3099, pruned_loss=0.08186, over 5115182.35 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3546, pruned_loss=0.09849, over 1114327.57 frames. ], giga_tot_loss[loss=0.234, simple_loss=0.3066, pruned_loss=0.08072, over 4964504.58 frames. ], batch size: 262, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:38:51,565 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-456000.pt
+2023-03-05 13:38:54,518 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=456004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:39:07,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.582e+02 8.974e+02 1.142e+03 1.471e+03 4.544e+03, threshold=2.285e+03, percent-clipped=3.0
+2023-03-05 13:39:26,875 INFO [train.py:968] (0/2) Epoch 11, batch 500, giga_loss[loss=0.2313, simple_loss=0.3003, pruned_loss=0.08108, over 29094.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3063, pruned_loss=0.0799, over 5246161.06 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3525, pruned_loss=0.09739, over 1182857.99 frames. ], giga_tot_loss[loss=0.2307, simple_loss=0.3035, pruned_loss=0.07893, over 5120322.79 frames. ], batch size: 136, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:40:09,695 INFO [train.py:968] (0/2) Epoch 11, batch 550, giga_loss[loss=0.2364, simple_loss=0.3068, pruned_loss=0.08306, over 28711.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3053, pruned_loss=0.07967, over 5351583.09 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3549, pruned_loss=0.09918, over 1297333.87 frames. ], giga_tot_loss[loss=0.2288, simple_loss=0.3014, pruned_loss=0.07808, over 5239698.66 frames. ], batch size: 262, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:40:37,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.697e+02 9.627e+02 1.196e+03 1.682e+03 5.981e+03, threshold=2.392e+03, percent-clipped=16.0
+2023-03-05 13:40:45,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=456126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:40:47,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=456129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:40:59,313 INFO [train.py:968] (0/2) Epoch 11, batch 600, giga_loss[loss=0.2272, simple_loss=0.2849, pruned_loss=0.08476, over 23886.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3032, pruned_loss=0.07894, over 5420550.16 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3551, pruned_loss=0.09927, over 1343412.87 frames. ], giga_tot_loss[loss=0.2272, simple_loss=0.2995, pruned_loss=0.07749, over 5327944.80 frames. ], batch size: 705, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:41:03,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=456143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:41:14,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=456158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:41:35,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0224, 1.2587, 1.0342, 0.3446], device='cuda:0'), covar=tensor([0.2136, 0.1928, 0.2809, 0.3894], device='cuda:0'), in_proj_covar=tensor([0.1515, 0.1450, 0.1460, 0.1251], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 13:41:39,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7606, 1.9012, 1.3306, 1.5181], device='cuda:0'), covar=tensor([0.0855, 0.0660, 0.1154, 0.1083], device='cuda:0'), in_proj_covar=tensor([0.0349, 0.0438, 0.0491, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 13:41:44,580 INFO [train.py:968] (0/2) Epoch 11, batch 650, giga_loss[loss=0.228, simple_loss=0.2808, pruned_loss=0.08761, over 23913.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3029, pruned_loss=0.07872, over 5479916.30 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3567, pruned_loss=0.1003, over 1500505.80 frames. ], giga_tot_loss[loss=0.2255, simple_loss=0.2978, pruned_loss=0.07658, over 5391333.37 frames. ], batch size: 705, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:41:47,722 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 13:42:11,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.034e+02 9.496e+02 1.171e+03 1.617e+03 8.187e+03, threshold=2.342e+03, percent-clipped=12.0
+2023-03-05 13:42:32,221 INFO [train.py:968] (0/2) Epoch 11, batch 700, giga_loss[loss=0.2083, simple_loss=0.2863, pruned_loss=0.06519, over 28976.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3001, pruned_loss=0.07739, over 5532698.41 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3572, pruned_loss=0.1002, over 1586849.53 frames. ], giga_tot_loss[loss=0.2228, simple_loss=0.295, pruned_loss=0.07532, over 5456297.61 frames. ], batch size: 213, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:42:55,225 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 13:43:04,993 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2023, 1.6148, 1.5527, 1.1346], device='cuda:0'), covar=tensor([0.1630, 0.2291, 0.1344, 0.1489], device='cuda:0'), in_proj_covar=tensor([0.0841, 0.0714, 0.0877, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 13:43:16,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=456286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:43:20,247 INFO [train.py:968] (0/2) Epoch 11, batch 750, giga_loss[loss=0.1965, simple_loss=0.2699, pruned_loss=0.06151, over 28741.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2965, pruned_loss=0.07548, over 5576900.65 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3571, pruned_loss=0.1001, over 1650900.90 frames. ], giga_tot_loss[loss=0.2194, simple_loss=0.2917, pruned_loss=0.07357, over 5512394.63 frames. ], batch size: 284, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:43:20,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=456289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:43:45,203 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=456318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:43:45,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=456318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:43:45,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.948e+02 9.126e+02 1.135e+03 2.061e+03 7.630e+03, threshold=2.269e+03, percent-clipped=19.0
+2023-03-05 13:43:51,710 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 13:44:04,149 INFO [train.py:968] (0/2) Epoch 11, batch 800, libri_loss[loss=0.3049, simple_loss=0.3835, pruned_loss=0.1132, over 25998.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.2962, pruned_loss=0.07608, over 5598999.11 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3574, pruned_loss=0.1005, over 1732239.76 frames. ], giga_tot_loss[loss=0.2195, simple_loss=0.2911, pruned_loss=0.07398, over 5545090.33 frames. ], batch size: 136, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:44:43,423 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=456379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:44:54,257 INFO [train.py:968] (0/2) Epoch 11, batch 850, giga_loss[loss=0.2814, simple_loss=0.3617, pruned_loss=0.1006, over 28730.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3076, pruned_loss=0.08208, over 5612380.25 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3569, pruned_loss=0.09973, over 1834023.38 frames. ], giga_tot_loss[loss=0.2314, simple_loss=0.3025, pruned_loss=0.08018, over 5563917.37 frames. ], batch size: 242, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:45:19,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3027, 1.6141, 1.3222, 1.5266], device='cuda:0'), covar=tensor([0.0711, 0.0402, 0.0330, 0.0727], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 13:45:24,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.735e+02 1.274e+03 1.568e+03 2.486e+03 6.544e+03, threshold=3.136e+03, percent-clipped=29.0
+2023-03-05 13:45:40,375 INFO [train.py:968] (0/2) Epoch 11, batch 900, giga_loss[loss=0.3232, simple_loss=0.3827, pruned_loss=0.1319, over 28668.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.321, pruned_loss=0.089, over 5633907.01 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.355, pruned_loss=0.09871, over 1915312.35 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3168, pruned_loss=0.08761, over 5589930.16 frames. ], batch size: 99, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:45:59,636 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=456461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:46:02,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0020, 1.2020, 3.4372, 2.9169], device='cuda:0'), covar=tensor([0.1645, 0.2669, 0.0445, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0650, 0.0581, 0.0849, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 13:46:02,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=456464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:46:22,316 INFO [train.py:968] (0/2) Epoch 11, batch 950, libri_loss[loss=0.2428, simple_loss=0.3216, pruned_loss=0.08196, over 29575.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3315, pruned_loss=0.09391, over 5659527.19 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3535, pruned_loss=0.09785, over 2033201.89 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3281, pruned_loss=0.09296, over 5617363.21 frames. ], batch size: 75, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:46:27,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=456493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:46:29,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3188, 1.1529, 3.9116, 3.2246], device='cuda:0'), covar=tensor([0.1618, 0.2835, 0.0432, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0651, 0.0581, 0.0850, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 13:46:46,103 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=456519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:46:46,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.037e+02 1.264e+03 1.584e+03 2.317e+03 7.526e+03, threshold=3.167e+03, percent-clipped=14.0
+2023-03-05 13:46:48,419 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=456522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:46:50,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=456525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:47:02,173 INFO [train.py:968] (0/2) Epoch 11, batch 1000, giga_loss[loss=0.2933, simple_loss=0.3667, pruned_loss=0.11, over 28941.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3389, pruned_loss=0.097, over 5664360.81 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.353, pruned_loss=0.09763, over 2172687.27 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3357, pruned_loss=0.09628, over 5633376.66 frames. ], batch size: 199, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:47:04,515 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 13:47:10,225 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=456549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:47:14,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=456554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:47:26,091 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.85 vs. limit=5.0
+2023-03-05 13:47:41,223 INFO [train.py:968] (0/2) Epoch 11, batch 1050, giga_loss[loss=0.3148, simple_loss=0.389, pruned_loss=0.1203, over 28767.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3432, pruned_loss=0.09776, over 5670403.22 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3541, pruned_loss=0.09837, over 2262370.33 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.34, pruned_loss=0.09688, over 5642626.94 frames. ], batch size: 284, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:48:12,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.646e+02 1.096e+03 1.396e+03 1.978e+03 5.581e+03, threshold=2.791e+03, percent-clipped=5.0
+2023-03-05 13:48:23,156 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 13:48:27,960 INFO [train.py:968] (0/2) Epoch 11, batch 1100, giga_loss[loss=0.2848, simple_loss=0.3507, pruned_loss=0.1094, over 28944.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3442, pruned_loss=0.09762, over 5674923.34 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3525, pruned_loss=0.09758, over 2333763.83 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3421, pruned_loss=0.09721, over 5649279.41 frames. ], batch size: 106, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:48:41,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3288, 1.5085, 1.4674, 1.3521], device='cuda:0'), covar=tensor([0.1400, 0.1531, 0.1948, 0.1567], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0730, 0.0664, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 13:48:45,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5683, 2.1085, 1.5595, 0.7810], device='cuda:0'), covar=tensor([0.3647, 0.2056, 0.2749, 0.4007], device='cuda:0'), in_proj_covar=tensor([0.1522, 0.1453, 0.1475, 0.1258], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 13:49:13,840 INFO [train.py:968] (0/2) Epoch 11, batch 1150, giga_loss[loss=0.2863, simple_loss=0.3583, pruned_loss=0.1072, over 28673.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3463, pruned_loss=0.09901, over 5685066.14 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3519, pruned_loss=0.09727, over 2369242.83 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3448, pruned_loss=0.09881, over 5663073.48 frames. ], batch size: 242, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:49:32,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4537, 1.5789, 1.2644, 1.6195], device='cuda:0'), covar=tensor([0.2321, 0.2242, 0.2366, 0.2104], device='cuda:0'), in_proj_covar=tensor([0.1283, 0.0955, 0.1137, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 13:49:41,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.032e+02 1.041e+03 1.232e+03 1.542e+03 4.928e+03, threshold=2.465e+03, percent-clipped=5.0
+2023-03-05 13:49:58,613 INFO [train.py:968] (0/2) Epoch 11, batch 1200, giga_loss[loss=0.3026, simple_loss=0.3726, pruned_loss=0.1163, over 28712.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3494, pruned_loss=0.1013, over 5678256.36 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3518, pruned_loss=0.09707, over 2452312.53 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3481, pruned_loss=0.1013, over 5659707.82 frames. ], batch size: 284, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:50:36,986 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 13:50:40,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3529, 1.5862, 1.6715, 1.2082], device='cuda:0'), covar=tensor([0.1660, 0.2341, 0.1288, 0.1535], device='cuda:0'), in_proj_covar=tensor([0.0833, 0.0705, 0.0870, 0.0775], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 13:50:41,099 INFO [train.py:968] (0/2) Epoch 11, batch 1250, giga_loss[loss=0.2758, simple_loss=0.3562, pruned_loss=0.09772, over 28964.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3517, pruned_loss=0.1027, over 5673230.81 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3519, pruned_loss=0.0975, over 2544077.68 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3507, pruned_loss=0.1027, over 5661699.91 frames. ], batch size: 136, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:51:03,412 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=456815.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 13:51:07,206 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.881e+02 1.088e+03 1.451e+03 1.780e+03 3.608e+03, threshold=2.902e+03, percent-clipped=10.0
+2023-03-05 13:51:21,608 INFO [train.py:968] (0/2) Epoch 11, batch 1300, giga_loss[loss=0.2603, simple_loss=0.3392, pruned_loss=0.09066, over 28882.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3541, pruned_loss=0.1034, over 5681911.32 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3515, pruned_loss=0.09751, over 2657245.22 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3535, pruned_loss=0.1036, over 5667899.03 frames. ], batch size: 60, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:51:29,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4655, 1.6604, 1.3837, 1.2792], device='cuda:0'), covar=tensor([0.2367, 0.2349, 0.2545, 0.2233], device='cuda:0'), in_proj_covar=tensor([0.1287, 0.0956, 0.1142, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 13:52:04,318 INFO [train.py:968] (0/2) Epoch 11, batch 1350, giga_loss[loss=0.2654, simple_loss=0.3549, pruned_loss=0.088, over 28469.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3561, pruned_loss=0.1035, over 5694318.12 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3517, pruned_loss=0.09776, over 2702621.79 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3556, pruned_loss=0.1036, over 5683185.59 frames. ], batch size: 60, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:52:10,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=456894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:52:31,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.521e+02 1.149e+03 1.408e+03 1.974e+03 5.393e+03, threshold=2.816e+03, percent-clipped=8.0
+2023-03-05 13:52:32,087 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=456921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:52:34,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=456924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:52:45,012 INFO [train.py:968] (0/2) Epoch 11, batch 1400, giga_loss[loss=0.3043, simple_loss=0.3789, pruned_loss=0.1149, over 28566.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.357, pruned_loss=0.1033, over 5693644.82 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3514, pruned_loss=0.09764, over 2780531.17 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3568, pruned_loss=0.1036, over 5681354.33 frames. ], batch size: 71, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:53:29,648 INFO [train.py:968] (0/2) Epoch 11, batch 1450, giga_loss[loss=0.287, simple_loss=0.3605, pruned_loss=0.1067, over 28761.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3563, pruned_loss=0.1016, over 5701567.96 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3513, pruned_loss=0.09741, over 2842651.66 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3563, pruned_loss=0.102, over 5687934.28 frames. ], batch size: 92, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:53:49,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3376, 3.3627, 1.6647, 1.5183], device='cuda:0'), covar=tensor([0.1016, 0.0226, 0.0831, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0500, 0.0334, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 13:53:52,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.780e+02 1.010e+03 1.216e+03 1.635e+03 5.311e+03, threshold=2.433e+03, percent-clipped=6.0
+2023-03-05 13:54:02,064 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-05 13:54:06,817 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=457037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:07,864 INFO [train.py:968] (0/2) Epoch 11, batch 1500, giga_loss[loss=0.2332, simple_loss=0.3221, pruned_loss=0.07218, over 28452.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.354, pruned_loss=0.09906, over 5700034.44 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3508, pruned_loss=0.09688, over 2938046.04 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3544, pruned_loss=0.09972, over 5691683.85 frames. ], batch size: 65, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:54:09,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=457040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:21,122 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=457057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:28,795 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=457067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:30,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=457069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:31,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=457070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:38,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6495, 1.5844, 1.3097, 1.2656], device='cuda:0'), covar=tensor([0.0775, 0.0592, 0.0953, 0.1013], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0438, 0.0495, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 13:54:47,724 INFO [train.py:968] (0/2) Epoch 11, batch 1550, giga_loss[loss=0.2277, simple_loss=0.32, pruned_loss=0.06772, over 28985.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.352, pruned_loss=0.09767, over 5706354.89 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3499, pruned_loss=0.09638, over 3053220.23 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3528, pruned_loss=0.09849, over 5693487.41 frames. ], batch size: 145, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:54:56,887 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=457099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:55:15,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.283e+02 9.798e+02 1.396e+03 1.890e+03 5.209e+03, threshold=2.792e+03, percent-clipped=7.0
+2023-03-05 13:55:31,311 INFO [train.py:968] (0/2) Epoch 11, batch 1600, giga_loss[loss=0.3073, simple_loss=0.3688, pruned_loss=0.1229, over 28908.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3541, pruned_loss=0.1005, over 5719125.57 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3501, pruned_loss=0.09649, over 3151680.89 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3548, pruned_loss=0.1012, over 5702375.73 frames. ], batch size: 186, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:56:21,412 INFO [train.py:968] (0/2) Epoch 11, batch 1650, giga_loss[loss=0.2711, simple_loss=0.3458, pruned_loss=0.09821, over 28881.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3578, pruned_loss=0.1063, over 5708121.36 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.35, pruned_loss=0.09643, over 3165472.48 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3583, pruned_loss=0.1069, over 5694172.88 frames. ], batch size: 174, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:56:22,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=457190.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:56:37,706 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 13:56:44,343 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5347, 1.6401, 1.5193, 1.4996], device='cuda:0'), covar=tensor([0.1422, 0.1709, 0.1832, 0.1613], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0723, 0.0662, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 13:56:45,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4620, 1.7232, 1.2319, 1.6372], device='cuda:0'), covar=tensor([0.0724, 0.0275, 0.0326, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 13:56:46,243 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.752e+02 1.230e+03 1.627e+03 2.067e+03 5.891e+03, threshold=3.255e+03, percent-clipped=13.0
+2023-03-05 13:56:54,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2208, 1.8151, 5.3535, 4.1506], device='cuda:0'), covar=tensor([0.1695, 0.2602, 0.0520, 0.0643], device='cuda:0'), in_proj_covar=tensor([0.0641, 0.0570, 0.0834, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 13:57:00,395 INFO [train.py:968] (0/2) Epoch 11, batch 1700, giga_loss[loss=0.2802, simple_loss=0.3507, pruned_loss=0.1048, over 28836.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3583, pruned_loss=0.1084, over 5705843.83 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3482, pruned_loss=0.09573, over 3284276.55 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.36, pruned_loss=0.1097, over 5688689.76 frames. ], batch size: 199, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:57:44,216 INFO [train.py:968] (0/2) Epoch 11, batch 1750, giga_loss[loss=0.3132, simple_loss=0.3681, pruned_loss=0.1292, over 28738.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3573, pruned_loss=0.1084, over 5702962.75 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3482, pruned_loss=0.09578, over 3363321.18 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.359, pruned_loss=0.1099, over 5692420.23 frames. ], batch size: 119, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 13:57:50,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=457296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:58:09,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.976e+02 1.116e+03 1.468e+03 1.895e+03 5.009e+03, threshold=2.936e+03, percent-clipped=7.0
+2023-03-05 13:58:20,472 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=457333.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:58:23,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=457336.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 13:58:25,003 INFO [train.py:968] (0/2) Epoch 11, batch 1800, libri_loss[loss=0.2429, simple_loss=0.3249, pruned_loss=0.08046, over 29591.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3563, pruned_loss=0.108, over 5704483.74 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3486, pruned_loss=0.09575, over 3421855.58 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3576, pruned_loss=0.1095, over 5695714.26 frames. ], batch size: 75, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 13:58:45,833 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=457365.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:58:55,376 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3256, 1.9009, 1.4029, 0.4754], device='cuda:0'), covar=tensor([0.3367, 0.1871, 0.2978, 0.4142], device='cuda:0'), in_proj_covar=tensor([0.1514, 0.1440, 0.1462, 0.1247], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 13:59:04,941 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-05 13:59:05,084 INFO [train.py:968] (0/2) Epoch 11, batch 1850, giga_loss[loss=0.3185, simple_loss=0.3851, pruned_loss=0.126, over 28594.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3562, pruned_loss=0.107, over 5714644.07 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3492, pruned_loss=0.09595, over 3506189.13 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3572, pruned_loss=0.1084, over 5702533.17 frames. ], batch size: 307, lr: 3.00e-03, grad_scale: 2.0
+2023-03-05 13:59:34,132 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.237e+02 1.159e+03 1.574e+03 2.636e+03 2.767e+04, threshold=3.147e+03, percent-clipped=21.0
+2023-03-05 13:59:45,970 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=457432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:59:52,244 INFO [train.py:968] (0/2) Epoch 11, batch 1900, giga_loss[loss=0.2532, simple_loss=0.332, pruned_loss=0.08721, over 28406.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3539, pruned_loss=0.105, over 5706951.39 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3488, pruned_loss=0.09573, over 3538796.79 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.355, pruned_loss=0.1064, over 5697746.42 frames. ], batch size: 65, lr: 3.00e-03, grad_scale: 2.0
+2023-03-05 13:59:52,496 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=457439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:59:54,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=457442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:00:23,582 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=457471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:00:31,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5408, 2.1892, 1.7864, 0.8148], device='cuda:0'), covar=tensor([0.4500, 0.2362, 0.2668, 0.4325], device='cuda:0'), in_proj_covar=tensor([0.1513, 0.1437, 0.1462, 0.1248], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 14:00:37,402 INFO [train.py:968] (0/2) Epoch 11, batch 1950, giga_loss[loss=0.2569, simple_loss=0.3306, pruned_loss=0.09162, over 28863.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3504, pruned_loss=0.1026, over 5701035.83 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3492, pruned_loss=0.09593, over 3616154.81 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3512, pruned_loss=0.1038, over 5690887.42 frames. ], batch size: 227, lr: 3.00e-03, grad_scale: 2.0
+2023-03-05 14:00:56,428 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4339, 1.9071, 1.5662, 1.6506], device='cuda:0'), covar=tensor([0.0670, 0.0260, 0.0268, 0.0639], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 14:01:10,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.593e+02 1.015e+03 1.293e+03 1.658e+03 1.662e+04, threshold=2.586e+03, percent-clipped=7.0
+2023-03-05 14:01:26,516 INFO [train.py:968] (0/2) Epoch 11, batch 2000, giga_loss[loss=0.2327, simple_loss=0.3121, pruned_loss=0.07665, over 28969.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3438, pruned_loss=0.0988, over 5684831.35 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3497, pruned_loss=0.09622, over 3651977.33 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3441, pruned_loss=0.09974, over 5680784.05 frames. ], batch size: 155, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:01:45,319 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-05 14:01:59,952 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=457575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:02:00,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6755, 3.5034, 1.6759, 1.6779], device='cuda:0'), covar=tensor([0.0837, 0.0243, 0.0818, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0496, 0.0333, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 14:02:03,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=457578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:02:13,219 INFO [train.py:968] (0/2) Epoch 11, batch 2050, giga_loss[loss=0.2442, simple_loss=0.3147, pruned_loss=0.08687, over 28987.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3385, pruned_loss=0.09635, over 5678270.31 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3499, pruned_loss=0.09631, over 3693299.26 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3385, pruned_loss=0.09706, over 5673978.21 frames. ], batch size: 227, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:02:30,413 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=457607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:02:32,400 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-05 14:02:34,011 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=457612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:02:46,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.803e+02 8.784e+02 1.034e+03 1.443e+03 3.650e+03, threshold=2.068e+03, percent-clipped=4.0
+2023-03-05 14:03:01,855 INFO [train.py:968] (0/2) Epoch 11, batch 2100, giga_loss[loss=0.2984, simple_loss=0.3557, pruned_loss=0.1205, over 27610.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3372, pruned_loss=0.09537, over 5686124.33 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3501, pruned_loss=0.0964, over 3769078.69 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3366, pruned_loss=0.09585, over 5676121.88 frames. ], batch size: 472, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:03:09,182 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=457648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:03:41,067 INFO [train.py:968] (0/2) Epoch 11, batch 2150, giga_loss[loss=0.2848, simple_loss=0.3639, pruned_loss=0.1029, over 28594.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3373, pruned_loss=0.09459, over 5696516.37 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3504, pruned_loss=0.09645, over 3831570.46 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3363, pruned_loss=0.09491, over 5683452.34 frames. ], batch size: 336, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:04:05,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.913e+02 9.902e+02 1.223e+03 1.597e+03 2.660e+03, threshold=2.447e+03, percent-clipped=7.0
+2023-03-05 14:04:17,925 INFO [train.py:968] (0/2) Epoch 11, batch 2200, giga_loss[loss=0.2486, simple_loss=0.3237, pruned_loss=0.0867, over 28859.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3378, pruned_loss=0.09432, over 5704667.61 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3506, pruned_loss=0.09611, over 3957163.48 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3361, pruned_loss=0.09469, over 5686705.46 frames. ], batch size: 199, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:04:56,060 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-05 14:04:59,383 INFO [train.py:968] (0/2) Epoch 11, batch 2250, giga_loss[loss=0.2766, simple_loss=0.3307, pruned_loss=0.1112, over 28643.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3375, pruned_loss=0.09462, over 5708644.36 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3512, pruned_loss=0.09629, over 3995667.92 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3355, pruned_loss=0.09477, over 5690970.07 frames. ], batch size: 85, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:04:59,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3653, 4.2105, 3.9736, 1.9908], device='cuda:0'), covar=tensor([0.0497, 0.0592, 0.0600, 0.2043], device='cuda:0'), in_proj_covar=tensor([0.1001, 0.0935, 0.0826, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 14:05:03,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0619, 1.2022, 1.1147, 0.9096], device='cuda:0'), covar=tensor([0.1421, 0.1568, 0.0878, 0.1129], device='cuda:0'), in_proj_covar=tensor([0.1650, 0.1563, 0.1527, 0.1653], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 14:05:06,041 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=457797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:05:24,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.225e+02 1.017e+03 1.237e+03 1.730e+03 4.662e+03, threshold=2.474e+03, percent-clipped=10.0
+2023-03-05 14:05:39,285 INFO [train.py:968] (0/2) Epoch 11, batch 2300, giga_loss[loss=0.2918, simple_loss=0.356, pruned_loss=0.1138, over 28562.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3352, pruned_loss=0.09367, over 5717722.98 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3509, pruned_loss=0.09599, over 4069376.30 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3332, pruned_loss=0.09392, over 5697874.62 frames. ], batch size: 336, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:05:59,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1745, 2.0155, 1.5955, 1.6243], device='cuda:0'), covar=tensor([0.0784, 0.0685, 0.0958, 0.1148], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0439, 0.0495, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 14:06:22,690 INFO [train.py:968] (0/2) Epoch 11, batch 2350, giga_loss[loss=0.2266, simple_loss=0.3075, pruned_loss=0.07288, over 29042.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3322, pruned_loss=0.09223, over 5722294.98 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3506, pruned_loss=0.09564, over 4104402.34 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3305, pruned_loss=0.09259, over 5704562.79 frames. ], batch size: 136, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:06:51,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.585e+02 8.930e+02 1.202e+03 1.759e+03 5.512e+03, threshold=2.404e+03, percent-clipped=12.0
+2023-03-05 14:07:07,236 INFO [train.py:968] (0/2) Epoch 11, batch 2400, giga_loss[loss=0.2246, simple_loss=0.3024, pruned_loss=0.07344, over 28915.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3298, pruned_loss=0.09138, over 5726426.05 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3509, pruned_loss=0.09574, over 4122625.03 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3281, pruned_loss=0.09157, over 5710625.73 frames. ], batch size: 174, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:07:17,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5217, 1.6716, 1.4597, 1.2165], device='cuda:0'), covar=tensor([0.2435, 0.1708, 0.1500, 0.1988], device='cuda:0'), in_proj_covar=tensor([0.1663, 0.1569, 0.1540, 0.1668], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 14:07:30,959 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=457969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:07:46,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=457987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:07:47,455 INFO [train.py:968] (0/2) Epoch 11, batch 2450, giga_loss[loss=0.2513, simple_loss=0.3168, pruned_loss=0.0929, over 28689.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3278, pruned_loss=0.09042, over 5733143.33 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3513, pruned_loss=0.09567, over 4183291.26 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3255, pruned_loss=0.09047, over 5715866.77 frames. ], batch size: 92, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:07:55,142 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-458000.pt
+2023-03-05 14:08:12,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.742e+02 9.626e+02 1.163e+03 1.431e+03 3.265e+03, threshold=2.326e+03, percent-clipped=5.0
+2023-03-05 14:08:13,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:08:20,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:08:24,392 INFO [train.py:968] (0/2) Epoch 11, batch 2500, giga_loss[loss=0.248, simple_loss=0.3198, pruned_loss=0.08808, over 28591.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3255, pruned_loss=0.08891, over 5738504.25 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3516, pruned_loss=0.09556, over 4232719.71 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3227, pruned_loss=0.08885, over 5721688.27 frames. ], batch size: 307, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:09:00,765 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-05 14:09:06,625 INFO [train.py:968] (0/2) Epoch 11, batch 2550, giga_loss[loss=0.2224, simple_loss=0.2963, pruned_loss=0.07429, over 28420.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3243, pruned_loss=0.08884, over 5723177.30 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3523, pruned_loss=0.09589, over 4260418.03 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3211, pruned_loss=0.08844, over 5711979.67 frames. ], batch size: 71, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:09:32,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.096e+02 9.694e+02 1.290e+03 1.815e+03 6.482e+03, threshold=2.580e+03, percent-clipped=14.0
+2023-03-05 14:09:39,482 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:09:41,402 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:09:45,344 INFO [train.py:968] (0/2) Epoch 11, batch 2600, giga_loss[loss=0.2531, simple_loss=0.3192, pruned_loss=0.09347, over 28791.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3226, pruned_loss=0.08775, over 5727010.56 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3532, pruned_loss=0.09627, over 4289287.12 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.319, pruned_loss=0.08703, over 5718522.61 frames. ], batch size: 99, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:10:03,023 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:06,399 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:10,184 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:12,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:14,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:26,735 INFO [train.py:968] (0/2) Epoch 11, batch 2650, giga_loss[loss=0.2156, simple_loss=0.2983, pruned_loss=0.06649, over 28931.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3218, pruned_loss=0.08707, over 5727346.97 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3542, pruned_loss=0.09648, over 4329048.82 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3175, pruned_loss=0.08614, over 5716715.84 frames. ], batch size: 174, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:10:34,550 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:56,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.878e+02 8.753e+02 1.144e+03 1.449e+03 4.301e+03, threshold=2.288e+03, percent-clipped=6.0
+2023-03-05 14:11:08,928 INFO [train.py:968] (0/2) Epoch 11, batch 2700, giga_loss[loss=0.244, simple_loss=0.3216, pruned_loss=0.08324, over 28342.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.323, pruned_loss=0.08776, over 5725257.84 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3544, pruned_loss=0.09652, over 4382189.03 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3183, pruned_loss=0.08668, over 5712232.26 frames. ], batch size: 65, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:11:29,995 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-05 14:11:46,214 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:11:55,062 INFO [train.py:968] (0/2) Epoch 11, batch 2750, giga_loss[loss=0.2583, simple_loss=0.3376, pruned_loss=0.08951, over 28889.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3275, pruned_loss=0.09102, over 5721394.52 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3541, pruned_loss=0.09656, over 4404910.70 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3236, pruned_loss=0.09005, over 5708562.17 frames. ], batch size: 174, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:12:19,336 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:12:21,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:12:26,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.493e+02 1.191e+03 1.727e+03 2.544e+03 5.223e+03, threshold=3.453e+03, percent-clipped=26.0
+2023-03-05 14:12:40,937 INFO [train.py:968] (0/2) Epoch 11, batch 2800, giga_loss[loss=0.344, simple_loss=0.3966, pruned_loss=0.1457, over 27587.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3348, pruned_loss=0.09576, over 5701538.84 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3543, pruned_loss=0.09653, over 4431449.57 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3311, pruned_loss=0.09494, over 5696214.55 frames. ], batch size: 472, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:12:45,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:12:48,387 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:13:27,247 INFO [train.py:968] (0/2) Epoch 11, batch 2850, giga_loss[loss=0.2909, simple_loss=0.3628, pruned_loss=0.1095, over 28730.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3433, pruned_loss=0.1017, over 5694004.96 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3542, pruned_loss=0.09649, over 4456939.20 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3402, pruned_loss=0.1011, over 5689453.17 frames. ], batch size: 284, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:13:31,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3241, 4.1135, 3.8610, 1.9312], device='cuda:0'), covar=tensor([0.0511, 0.0657, 0.0666, 0.2045], device='cuda:0'), in_proj_covar=tensor([0.1019, 0.0948, 0.0836, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 14:13:31,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4672, 1.7381, 1.4280, 1.5440], device='cuda:0'), covar=tensor([0.2319, 0.2272, 0.2475, 0.2116], device='cuda:0'), in_proj_covar=tensor([0.1280, 0.0952, 0.1136, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 14:13:43,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:13:58,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.796e+02 1.213e+03 1.508e+03 2.006e+03 4.711e+03, threshold=3.016e+03, percent-clipped=3.0
+2023-03-05 14:14:12,069 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458438.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:14:12,517 INFO [train.py:968] (0/2) Epoch 11, batch 2900, giga_loss[loss=0.3395, simple_loss=0.4026, pruned_loss=0.1383, over 28724.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3492, pruned_loss=0.1042, over 5694916.16 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3546, pruned_loss=0.09686, over 4506654.44 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3462, pruned_loss=0.1037, over 5685091.84 frames. ], batch size: 284, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:14:57,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:14:58,794 INFO [train.py:968] (0/2) Epoch 11, batch 2950, libri_loss[loss=0.273, simple_loss=0.3668, pruned_loss=0.08955, over 29148.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3542, pruned_loss=0.1062, over 5681522.24 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3549, pruned_loss=0.09667, over 4559561.99 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3514, pruned_loss=0.1062, over 5667808.33 frames. ], batch size: 101, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:14:59,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:15:24,109 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:15:32,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.165e+02 1.038e+03 1.297e+03 1.753e+03 7.628e+03, threshold=2.594e+03, percent-clipped=12.0
+2023-03-05 14:15:40,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:15:45,347 INFO [train.py:968] (0/2) Epoch 11, batch 3000, giga_loss[loss=0.3167, simple_loss=0.3823, pruned_loss=0.1255, over 28909.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3587, pruned_loss=0.1079, over 5700227.78 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3541, pruned_loss=0.09612, over 4592804.78 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.357, pruned_loss=0.1086, over 5684470.07 frames. ], batch size: 186, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:15:45,352 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 14:15:51,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2496, 1.7738, 1.6062, 1.1380], device='cuda:0'), covar=tensor([0.1923, 0.2559, 0.1635, 0.1829], device='cuda:0'), in_proj_covar=tensor([0.0829, 0.0703, 0.0866, 0.0771], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 14:15:53,825 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2243, simple_loss=0.3286, pruned_loss=0.05998, over 944034.00 frames. 
+2023-03-05 14:15:53,825 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 14:15:54,113 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:16:02,705 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:16:05,862 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:16:30,330 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:16:34,266 INFO [train.py:968] (0/2) Epoch 11, batch 3050, giga_loss[loss=0.2283, simple_loss=0.3164, pruned_loss=0.07011, over 28794.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.359, pruned_loss=0.1087, over 5689427.33 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3534, pruned_loss=0.09609, over 4631827.28 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3584, pruned_loss=0.1097, over 5671059.52 frames. ], batch size: 174, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:16:41,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 14:17:06,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.818e+02 1.274e+03 1.691e+03 2.726e+03 9.588e+03, threshold=3.381e+03, percent-clipped=29.0
+2023-03-05 14:17:07,031 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4636, 4.5816, 1.7188, 1.6499], device='cuda:0'), covar=tensor([0.0963, 0.0218, 0.0836, 0.1342], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0498, 0.0332, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 14:17:17,914 INFO [train.py:968] (0/2) Epoch 11, batch 3100, libri_loss[loss=0.2629, simple_loss=0.3396, pruned_loss=0.09312, over 29533.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3534, pruned_loss=0.1045, over 5688177.96 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3535, pruned_loss=0.09633, over 4660368.12 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3529, pruned_loss=0.1053, over 5676685.63 frames. ], batch size: 79, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:17:19,620 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 14:17:31,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:17:37,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4078, 1.7235, 1.4648, 1.6487], device='cuda:0'), covar=tensor([0.0674, 0.0283, 0.0279, 0.0643], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0112, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 14:17:49,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:17:51,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:18:01,591 INFO [train.py:968] (0/2) Epoch 11, batch 3150, giga_loss[loss=0.3091, simple_loss=0.3805, pruned_loss=0.1189, over 28983.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.351, pruned_loss=0.1024, over 5682891.45 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3535, pruned_loss=0.09641, over 4685110.71 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3505, pruned_loss=0.1031, over 5669662.67 frames. ], batch size: 145, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:18:04,654 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5453, 4.3800, 4.0976, 2.0830], device='cuda:0'), covar=tensor([0.0506, 0.0652, 0.0671, 0.2029], device='cuda:0'), in_proj_covar=tensor([0.1023, 0.0957, 0.0835, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 14:18:15,730 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:18:29,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.412e+02 1.073e+03 1.424e+03 1.981e+03 5.708e+03, threshold=2.848e+03, percent-clipped=2.0
+2023-03-05 14:18:43,752 INFO [train.py:968] (0/2) Epoch 11, batch 3200, libri_loss[loss=0.3043, simple_loss=0.381, pruned_loss=0.1138, over 29225.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3513, pruned_loss=0.1025, over 5676498.40 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3532, pruned_loss=0.09641, over 4711022.61 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.351, pruned_loss=0.1032, over 5668571.50 frames. ], batch size: 94, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:18:45,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5805, 1.7070, 1.4103, 1.8268], device='cuda:0'), covar=tensor([0.2440, 0.2449, 0.2707, 0.2227], device='cuda:0'), in_proj_covar=tensor([0.1281, 0.0951, 0.1137, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 14:18:57,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6332, 1.8061, 1.9537, 1.4349], device='cuda:0'), covar=tensor([0.1720, 0.2173, 0.1290, 0.1558], device='cuda:0'), in_proj_covar=tensor([0.0832, 0.0703, 0.0870, 0.0775], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 14:19:09,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4363, 4.2659, 4.0120, 1.7906], device='cuda:0'), covar=tensor([0.0547, 0.0698, 0.0676, 0.2320], device='cuda:0'), in_proj_covar=tensor([0.1024, 0.0961, 0.0839, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 14:19:19,780 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:19:24,374 INFO [train.py:968] (0/2) Epoch 11, batch 3250, giga_loss[loss=0.3011, simple_loss=0.3626, pruned_loss=0.1198, over 28589.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3533, pruned_loss=0.1034, over 5679695.29 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3532, pruned_loss=0.09639, over 4731178.33 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3531, pruned_loss=0.1041, over 5672785.61 frames. ], batch size: 92, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:19:25,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0893, 1.2133, 3.8967, 3.0154], device='cuda:0'), covar=tensor([0.1699, 0.2572, 0.0400, 0.0928], device='cuda:0'), in_proj_covar=tensor([0.0650, 0.0575, 0.0841, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 14:19:27,602 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 14:19:33,639 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:19:35,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:19:44,770 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458813.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:19:55,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.621e+02 1.160e+03 1.450e+03 1.934e+03 5.995e+03, threshold=2.899e+03, percent-clipped=6.0
+2023-03-05 14:19:59,070 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:20:06,489 INFO [train.py:968] (0/2) Epoch 11, batch 3300, libri_loss[loss=0.3121, simple_loss=0.3934, pruned_loss=0.1154, over 26159.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3561, pruned_loss=0.1049, over 5691162.03 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3535, pruned_loss=0.09648, over 4785525.75 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3558, pruned_loss=0.1057, over 5678626.84 frames. ], batch size: 136, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:20:42,424 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-05 14:20:45,182 INFO [train.py:968] (0/2) Epoch 11, batch 3350, giga_loss[loss=0.2796, simple_loss=0.3512, pruned_loss=0.104, over 28921.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3577, pruned_loss=0.1062, over 5684169.25 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3541, pruned_loss=0.09697, over 4827314.09 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3571, pruned_loss=0.1069, over 5676757.38 frames. ], batch size: 112, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:21:06,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:21:16,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.101e+02 1.255e+03 1.631e+03 2.310e+03 5.963e+03, threshold=3.262e+03, percent-clipped=11.0
+2023-03-05 14:21:25,155 INFO [train.py:968] (0/2) Epoch 11, batch 3400, giga_loss[loss=0.2979, simple_loss=0.3724, pruned_loss=0.1117, over 29009.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3585, pruned_loss=0.1071, over 5697513.13 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3538, pruned_loss=0.09688, over 4853728.74 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3583, pruned_loss=0.108, over 5686893.88 frames. ], batch size: 136, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:21:42,057 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458956.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:21:44,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458959.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 14:22:11,048 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458988.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 14:22:11,451 INFO [train.py:968] (0/2) Epoch 11, batch 3450, giga_loss[loss=0.312, simple_loss=0.3746, pruned_loss=0.1247, over 28776.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3595, pruned_loss=0.1086, over 5691918.33 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3536, pruned_loss=0.09666, over 4879583.06 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3596, pruned_loss=0.1097, over 5678938.47 frames. ], batch size: 119, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:22:42,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.702e+02 1.184e+03 1.491e+03 2.026e+03 6.153e+03, threshold=2.982e+03, percent-clipped=4.0
+2023-03-05 14:22:46,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-05 14:22:52,970 INFO [train.py:968] (0/2) Epoch 11, batch 3500, libri_loss[loss=0.2647, simple_loss=0.3565, pruned_loss=0.08644, over 29669.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3592, pruned_loss=0.108, over 5690748.70 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3534, pruned_loss=0.09642, over 4904959.95 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3595, pruned_loss=0.1094, over 5675526.15 frames. ], batch size: 91, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:23:07,456 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=459057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:23:09,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=459060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:23:31,441 INFO [train.py:968] (0/2) Epoch 11, batch 3550, giga_loss[loss=0.2916, simple_loss=0.3702, pruned_loss=0.1065, over 29012.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3587, pruned_loss=0.1067, over 5700382.21 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3532, pruned_loss=0.09635, over 4932946.08 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3593, pruned_loss=0.1082, over 5684099.56 frames. ], batch size: 227, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:23:31,708 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=459089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:24:04,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.697e+02 1.061e+03 1.266e+03 1.805e+03 3.383e+03, threshold=2.532e+03, percent-clipped=5.0
+2023-03-05 14:24:09,352 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=459131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:24:14,902 INFO [train.py:968] (0/2) Epoch 11, batch 3600, giga_loss[loss=0.2783, simple_loss=0.3604, pruned_loss=0.09808, over 28950.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3597, pruned_loss=0.1064, over 5701866.80 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3532, pruned_loss=0.0963, over 4952019.80 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3603, pruned_loss=0.1078, over 5685225.77 frames. ], batch size: 213, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:24:16,669 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-05 14:24:20,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5241, 2.2099, 1.6751, 0.7274], device='cuda:0'), covar=tensor([0.4231, 0.2075, 0.2914, 0.4250], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1436, 0.1460, 0.1248], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 14:24:28,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=459157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:24:42,462 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 14:24:52,702 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4506, 1.6022, 1.3130, 1.8203], device='cuda:0'), covar=tensor([0.2304, 0.2208, 0.2350, 0.2075], device='cuda:0'), in_proj_covar=tensor([0.1278, 0.0951, 0.1133, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 14:24:54,524 INFO [train.py:968] (0/2) Epoch 11, batch 3650, giga_loss[loss=0.2529, simple_loss=0.3341, pruned_loss=0.08588, over 28966.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3591, pruned_loss=0.1057, over 5708822.23 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3535, pruned_loss=0.09652, over 4984793.93 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3595, pruned_loss=0.1069, over 5689268.99 frames. ], batch size: 213, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:25:23,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.276e+02 1.043e+03 1.386e+03 2.045e+03 7.456e+03, threshold=2.772e+03, percent-clipped=15.0
+2023-03-05 14:25:28,706 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=459233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:25:34,860 INFO [train.py:968] (0/2) Epoch 11, batch 3700, giga_loss[loss=0.2797, simple_loss=0.3492, pruned_loss=0.1051, over 28580.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3563, pruned_loss=0.1046, over 5693948.74 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3532, pruned_loss=0.09639, over 5002160.25 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.357, pruned_loss=0.1059, over 5681481.44 frames. ], batch size: 307, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:26:06,455 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.23 vs. limit=5.0
+2023-03-05 14:26:08,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4032, 1.6665, 0.9827, 1.2478], device='cuda:0'), covar=tensor([0.1012, 0.0772, 0.1629, 0.1268], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0434, 0.0495, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 14:26:15,249 INFO [train.py:968] (0/2) Epoch 11, batch 3750, giga_loss[loss=0.27, simple_loss=0.3504, pruned_loss=0.09483, over 28908.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3541, pruned_loss=0.103, over 5705663.43 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3534, pruned_loss=0.09646, over 5023751.95 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3546, pruned_loss=0.1042, over 5694697.21 frames. ], batch size: 145, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:26:22,821 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=459300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:26:23,007 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-05 14:26:23,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4476, 1.5615, 1.5312, 1.2808], device='cuda:0'), covar=tensor([0.1913, 0.1612, 0.1343, 0.1747], device='cuda:0'), in_proj_covar=tensor([0.1684, 0.1597, 0.1558, 0.1685], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 14:26:25,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=459303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:26:42,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.567e+02 1.007e+03 1.302e+03 2.041e+03 4.114e+03, threshold=2.605e+03, percent-clipped=12.0
+2023-03-05 14:26:48,283 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=459332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:26:53,306 INFO [train.py:968] (0/2) Epoch 11, batch 3800, giga_loss[loss=0.2742, simple_loss=0.3481, pruned_loss=0.1001, over 28756.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.353, pruned_loss=0.1028, over 5708041.36 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3539, pruned_loss=0.09677, over 5035781.53 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.353, pruned_loss=0.1036, over 5697805.24 frames. ], batch size: 284, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:27:31,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9569, 3.2924, 2.2381, 1.0103], device='cuda:0'), covar=tensor([0.5285, 0.2080, 0.2826, 0.4869], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1431, 0.1463, 0.1248], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 14:27:35,319 INFO [train.py:968] (0/2) Epoch 11, batch 3850, giga_loss[loss=0.2818, simple_loss=0.3578, pruned_loss=0.1029, over 28787.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3532, pruned_loss=0.1031, over 5710845.15 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3525, pruned_loss=0.09596, over 5072727.37 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3543, pruned_loss=0.1048, over 5695533.99 frames. ], batch size: 119, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:28:03,858 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.720e+02 1.063e+03 1.445e+03 2.083e+03 7.585e+03, threshold=2.890e+03, percent-clipped=17.0
+2023-03-05 14:28:12,195 INFO [train.py:968] (0/2) Epoch 11, batch 3900, libri_loss[loss=0.2979, simple_loss=0.37, pruned_loss=0.1129, over 19897.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3541, pruned_loss=0.1036, over 5710366.94 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3525, pruned_loss=0.09625, over 5103215.19 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3551, pruned_loss=0.105, over 5699292.28 frames. ], batch size: 186, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:28:55,565 INFO [train.py:968] (0/2) Epoch 11, batch 3950, giga_loss[loss=0.2401, simple_loss=0.3297, pruned_loss=0.07521, over 28990.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3541, pruned_loss=0.1031, over 5713086.29 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3522, pruned_loss=0.09605, over 5115176.73 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3552, pruned_loss=0.1045, over 5701824.32 frames. ], batch size: 136, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:29:11,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=459506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:29:27,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.597e+02 1.005e+03 1.188e+03 1.502e+03 4.574e+03, threshold=2.376e+03, percent-clipped=2.0
+2023-03-05 14:29:36,017 INFO [train.py:968] (0/2) Epoch 11, batch 4000, giga_loss[loss=0.251, simple_loss=0.3136, pruned_loss=0.09425, over 23451.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3537, pruned_loss=0.1023, over 5711954.14 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3532, pruned_loss=0.09679, over 5138120.75 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3537, pruned_loss=0.103, over 5700954.18 frames. ], batch size: 705, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:30:17,900 INFO [train.py:968] (0/2) Epoch 11, batch 4050, giga_loss[loss=0.2658, simple_loss=0.3477, pruned_loss=0.09196, over 28814.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3544, pruned_loss=0.1035, over 5698945.46 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3533, pruned_loss=0.09688, over 5146277.48 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3544, pruned_loss=0.1041, over 5697617.11 frames. ], batch size: 145, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:30:19,187 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-05 14:30:19,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5002, 1.7510, 1.4689, 1.5123], device='cuda:0'), covar=tensor([0.2006, 0.1817, 0.2049, 0.1678], device='cuda:0'), in_proj_covar=tensor([0.1271, 0.0945, 0.1126, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 14:30:33,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=459608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:30:48,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.692e+02 1.016e+03 1.242e+03 1.593e+03 7.726e+03, threshold=2.484e+03, percent-clipped=10.0
+2023-03-05 14:30:58,235 INFO [train.py:968] (0/2) Epoch 11, batch 4100, giga_loss[loss=0.2685, simple_loss=0.3452, pruned_loss=0.09587, over 28676.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.352, pruned_loss=0.1024, over 5704315.67 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3534, pruned_loss=0.09709, over 5152297.60 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3518, pruned_loss=0.1028, over 5706935.24 frames. ], batch size: 242, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:31:06,750 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=459649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:31:08,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=459652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:31:20,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=459666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:31:32,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=459681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:31:38,053 INFO [train.py:968] (0/2) Epoch 11, batch 4150, giga_loss[loss=0.2785, simple_loss=0.3517, pruned_loss=0.1026, over 27860.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3481, pruned_loss=0.1001, over 5708789.27 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3533, pruned_loss=0.09701, over 5163853.88 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.348, pruned_loss=0.1006, over 5707789.80 frames. ], batch size: 412, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:32:05,335 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-05 14:32:07,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.323e+02 1.116e+03 1.319e+03 1.802e+03 3.811e+03, threshold=2.638e+03, percent-clipped=8.0
+2023-03-05 14:32:17,276 INFO [train.py:968] (0/2) Epoch 11, batch 4200, giga_loss[loss=0.2256, simple_loss=0.3086, pruned_loss=0.0713, over 28444.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.346, pruned_loss=0.09909, over 5710294.11 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3528, pruned_loss=0.09671, over 5181539.23 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3462, pruned_loss=0.09979, over 5707951.13 frames. ], batch size: 60, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:32:28,583 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=459751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:32:30,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=459754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:32:54,224 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=459783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:32:59,686 INFO [train.py:968] (0/2) Epoch 11, batch 4250, giga_loss[loss=0.2904, simple_loss=0.3549, pruned_loss=0.113, over 27691.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3462, pruned_loss=0.09959, over 5712812.58 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3527, pruned_loss=0.09663, over 5188590.46 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3463, pruned_loss=0.1002, over 5709414.50 frames. ], batch size: 472, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:33:09,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2541, 1.3997, 1.2125, 1.2059], device='cuda:0'), covar=tensor([0.1665, 0.1464, 0.1183, 0.1419], device='cuda:0'), in_proj_covar=tensor([0.1683, 0.1603, 0.1571, 0.1691], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 14:33:31,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.538e+02 1.152e+03 1.413e+03 1.806e+03 3.844e+03, threshold=2.826e+03, percent-clipped=7.0
+2023-03-05 14:33:42,126 INFO [train.py:968] (0/2) Epoch 11, batch 4300, giga_loss[loss=0.2252, simple_loss=0.3025, pruned_loss=0.0739, over 28746.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3456, pruned_loss=0.1001, over 5713605.27 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3529, pruned_loss=0.09684, over 5208954.36 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3454, pruned_loss=0.1005, over 5706582.51 frames. ], batch size: 112, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:33:52,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4139, 1.6477, 1.5977, 1.5435], device='cuda:0'), covar=tensor([0.1407, 0.1515, 0.1756, 0.1515], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0722, 0.0663, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 14:34:24,510 INFO [train.py:968] (0/2) Epoch 11, batch 4350, giga_loss[loss=0.2471, simple_loss=0.3268, pruned_loss=0.08368, over 28276.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.343, pruned_loss=0.09915, over 5714791.95 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3525, pruned_loss=0.09663, over 5219392.03 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3431, pruned_loss=0.0997, over 5706681.96 frames. ], batch size: 60, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:34:49,516 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-05 14:34:50,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7897, 5.5620, 5.2560, 2.8437], device='cuda:0'), covar=tensor([0.0382, 0.0580, 0.0652, 0.1649], device='cuda:0'), in_proj_covar=tensor([0.1021, 0.0956, 0.0839, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 14:34:55,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.498e+02 1.005e+03 1.257e+03 1.650e+03 6.240e+03, threshold=2.513e+03, percent-clipped=3.0
+2023-03-05 14:35:02,858 INFO [train.py:968] (0/2) Epoch 11, batch 4400, giga_loss[loss=0.2163, simple_loss=0.2964, pruned_loss=0.06811, over 29029.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.341, pruned_loss=0.09829, over 5717479.08 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3529, pruned_loss=0.0968, over 5232841.54 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3405, pruned_loss=0.09861, over 5707770.06 frames. ], batch size: 155, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:35:44,413 INFO [train.py:968] (0/2) Epoch 11, batch 4450, giga_loss[loss=0.2693, simple_loss=0.3458, pruned_loss=0.09638, over 28711.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3395, pruned_loss=0.09749, over 5702900.57 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3529, pruned_loss=0.09673, over 5231791.11 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3389, pruned_loss=0.0978, over 5703440.37 frames. ], batch size: 262, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:35:53,288 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-460000.pt
+2023-03-05 14:36:17,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.636e+02 9.999e+02 1.189e+03 1.449e+03 3.683e+03, threshold=2.378e+03, percent-clipped=5.0
+2023-03-05 14:36:26,601 INFO [train.py:968] (0/2) Epoch 11, batch 4500, libri_loss[loss=0.3019, simple_loss=0.383, pruned_loss=0.1104, over 28590.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3422, pruned_loss=0.09898, over 5700239.43 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.353, pruned_loss=0.09692, over 5250186.77 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3413, pruned_loss=0.0991, over 5698654.84 frames. ], batch size: 106, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:36:29,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=460041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:36:56,178 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-05 14:37:09,982 INFO [train.py:968] (0/2) Epoch 11, batch 4550, giga_loss[loss=0.2827, simple_loss=0.3671, pruned_loss=0.0992, over 29014.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3446, pruned_loss=0.09978, over 5713544.92 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3529, pruned_loss=0.09685, over 5266611.48 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3437, pruned_loss=0.09999, over 5707780.00 frames. ], batch size: 213, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:37:43,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.864e+02 9.417e+02 1.252e+03 1.864e+03 1.076e+04, threshold=2.505e+03, percent-clipped=13.0
+2023-03-05 14:37:52,431 INFO [train.py:968] (0/2) Epoch 11, batch 4600, giga_loss[loss=0.2768, simple_loss=0.3598, pruned_loss=0.09691, over 28989.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3462, pruned_loss=0.09986, over 5718725.34 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3526, pruned_loss=0.09676, over 5277575.72 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3456, pruned_loss=0.1002, over 5712170.23 frames. ], batch size: 227, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:38:30,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3314, 1.6859, 1.4450, 1.5494], device='cuda:0'), covar=tensor([0.0623, 0.0262, 0.0279, 0.0666], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 14:38:32,636 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=460184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:38:35,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=460187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:38:37,275 INFO [train.py:968] (0/2) Epoch 11, batch 4650, giga_loss[loss=0.2423, simple_loss=0.3219, pruned_loss=0.08135, over 28887.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3474, pruned_loss=0.1002, over 5709919.80 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3526, pruned_loss=0.09693, over 5299097.06 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3467, pruned_loss=0.1004, over 5698920.50 frames. ], batch size: 213, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:38:44,418 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-05 14:39:04,111 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=460216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:39:14,660 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.036e+02 9.536e+02 1.185e+03 1.491e+03 3.699e+03, threshold=2.370e+03, percent-clipped=5.0
+2023-03-05 14:39:24,020 INFO [train.py:968] (0/2) Epoch 11, batch 4700, giga_loss[loss=0.284, simple_loss=0.3687, pruned_loss=0.0996, over 28653.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3455, pruned_loss=0.0982, over 5702245.97 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3524, pruned_loss=0.09679, over 5299746.58 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3451, pruned_loss=0.09848, over 5694832.04 frames. ], batch size: 336, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:40:03,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5296, 1.7488, 1.7983, 1.3707], device='cuda:0'), covar=tensor([0.1757, 0.2147, 0.1393, 0.1530], device='cuda:0'), in_proj_covar=tensor([0.0817, 0.0693, 0.0856, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 14:40:07,334 INFO [train.py:968] (0/2) Epoch 11, batch 4750, giga_loss[loss=0.2475, simple_loss=0.3172, pruned_loss=0.08891, over 28486.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3468, pruned_loss=0.099, over 5701633.67 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3523, pruned_loss=0.09671, over 5304734.00 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3464, pruned_loss=0.09931, over 5694991.36 frames. ], batch size: 71, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:40:41,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.341e+02 1.120e+03 1.417e+03 1.944e+03 4.682e+03, threshold=2.835e+03, percent-clipped=13.0
+2023-03-05 14:40:48,592 INFO [train.py:968] (0/2) Epoch 11, batch 4800, giga_loss[loss=0.2673, simple_loss=0.3431, pruned_loss=0.09572, over 28891.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3475, pruned_loss=0.09933, over 5709905.37 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3525, pruned_loss=0.09684, over 5310489.51 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3471, pruned_loss=0.0995, over 5705026.64 frames. ], batch size: 227, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:41:24,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1452, 1.3361, 1.3415, 1.0231], device='cuda:0'), covar=tensor([0.2090, 0.1805, 0.1081, 0.1649], device='cuda:0'), in_proj_covar=tensor([0.1680, 0.1600, 0.1564, 0.1678], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 14:41:32,723 INFO [train.py:968] (0/2) Epoch 11, batch 4850, giga_loss[loss=0.2642, simple_loss=0.3448, pruned_loss=0.0918, over 28730.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3486, pruned_loss=0.1007, over 5706446.78 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3521, pruned_loss=0.0966, over 5320759.21 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3485, pruned_loss=0.1011, over 5700494.95 frames. ], batch size: 262, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:42:04,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.724e+02 1.175e+03 1.482e+03 1.833e+03 4.122e+03, threshold=2.963e+03, percent-clipped=5.0
+2023-03-05 14:42:14,583 INFO [train.py:968] (0/2) Epoch 11, batch 4900, giga_loss[loss=0.3084, simple_loss=0.3699, pruned_loss=0.1235, over 28703.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3514, pruned_loss=0.1026, over 5717165.82 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3521, pruned_loss=0.09667, over 5335748.14 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3513, pruned_loss=0.1029, over 5707684.44 frames. ], batch size: 92, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:42:57,195 INFO [train.py:968] (0/2) Epoch 11, batch 4950, giga_loss[loss=0.2769, simple_loss=0.3572, pruned_loss=0.09824, over 28931.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3539, pruned_loss=0.1039, over 5718408.18 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3522, pruned_loss=0.09667, over 5347476.91 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3537, pruned_loss=0.1043, over 5707248.04 frames. ], batch size: 227, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:43:04,503 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:43:28,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.409e+02 1.314e+03 1.644e+03 2.290e+03 5.493e+03, threshold=3.287e+03, percent-clipped=10.0
+2023-03-05 14:43:36,382 INFO [train.py:968] (0/2) Epoch 11, batch 5000, giga_loss[loss=0.2907, simple_loss=0.3656, pruned_loss=0.1079, over 28806.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3554, pruned_loss=0.1043, over 5715608.18 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3523, pruned_loss=0.09663, over 5360922.14 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3552, pruned_loss=0.1049, over 5706017.73 frames. ], batch size: 243, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:44:14,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9077, 1.8457, 1.5044, 1.4730], device='cuda:0'), covar=tensor([0.0699, 0.0525, 0.0887, 0.0916], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0436, 0.0495, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 14:44:19,583 INFO [train.py:968] (0/2) Epoch 11, batch 5050, giga_loss[loss=0.3151, simple_loss=0.3886, pruned_loss=0.1208, over 28927.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3561, pruned_loss=0.1046, over 5713998.33 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3526, pruned_loss=0.0967, over 5369093.83 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3559, pruned_loss=0.1052, over 5703918.65 frames. ], batch size: 227, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:44:25,829 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-05 14:44:34,807 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:44:49,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.327e+02 1.087e+03 1.438e+03 2.239e+03 4.960e+03, threshold=2.876e+03, percent-clipped=11.0
+2023-03-05 14:44:58,720 INFO [train.py:968] (0/2) Epoch 11, batch 5100, giga_loss[loss=0.3289, simple_loss=0.3713, pruned_loss=0.1433, over 23588.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3558, pruned_loss=0.1044, over 5703104.34 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3524, pruned_loss=0.09662, over 5375952.99 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3558, pruned_loss=0.1051, over 5699889.99 frames. ], batch size: 705, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:45:12,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3252, 1.5530, 1.4925, 1.4088], device='cuda:0'), covar=tensor([0.1291, 0.1582, 0.1794, 0.1595], device='cuda:0'), in_proj_covar=tensor([0.0437, 0.0735, 0.0677, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 14:45:40,217 INFO [train.py:968] (0/2) Epoch 11, batch 5150, giga_loss[loss=0.2398, simple_loss=0.3276, pruned_loss=0.07597, over 29065.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3546, pruned_loss=0.1038, over 5708730.16 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3523, pruned_loss=0.09661, over 5384032.65 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3548, pruned_loss=0.1045, over 5703372.41 frames. ], batch size: 164, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:45:51,944 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-05 14:46:12,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.353e+02 1.024e+03 1.273e+03 1.610e+03 5.974e+03, threshold=2.547e+03, percent-clipped=3.0
+2023-03-05 14:46:20,938 INFO [train.py:968] (0/2) Epoch 11, batch 5200, giga_loss[loss=0.2121, simple_loss=0.3033, pruned_loss=0.06042, over 28975.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3527, pruned_loss=0.1034, over 5709709.21 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.353, pruned_loss=0.09703, over 5403868.29 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3523, pruned_loss=0.1038, over 5698762.44 frames. ], batch size: 174, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:46:31,738 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460754.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:46:57,168 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:47:01,159 INFO [train.py:968] (0/2) Epoch 11, batch 5250, libri_loss[loss=0.2506, simple_loss=0.3343, pruned_loss=0.08344, over 29552.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3502, pruned_loss=0.1024, over 5709755.28 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3535, pruned_loss=0.09757, over 5411547.33 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3493, pruned_loss=0.1025, over 5702463.24 frames. ], batch size: 79, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:47:21,854 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:47:31,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.104e+02 1.061e+03 1.296e+03 1.788e+03 4.984e+03, threshold=2.593e+03, percent-clipped=8.0
+2023-03-05 14:47:38,975 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:47:40,395 INFO [train.py:968] (0/2) Epoch 11, batch 5300, giga_loss[loss=0.2835, simple_loss=0.3671, pruned_loss=0.09993, over 28674.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3495, pruned_loss=0.1011, over 5714092.41 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3535, pruned_loss=0.09755, over 5424220.97 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3487, pruned_loss=0.1012, over 5704963.14 frames. ], batch size: 262, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:48:02,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5236, 1.8286, 1.6430, 1.4808], device='cuda:0'), covar=tensor([0.2360, 0.1850, 0.1958, 0.1869], device='cuda:0'), in_proj_covar=tensor([0.1682, 0.1603, 0.1569, 0.1675], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 14:48:11,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=460872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:48:26,289 INFO [train.py:968] (0/2) Epoch 11, batch 5350, giga_loss[loss=0.3061, simple_loss=0.3603, pruned_loss=0.1259, over 23868.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3511, pruned_loss=0.1004, over 5713505.58 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3538, pruned_loss=0.09775, over 5429855.18 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3502, pruned_loss=0.1004, over 5705000.61 frames. ], batch size: 705, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:48:58,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.344e+02 1.070e+03 1.403e+03 1.948e+03 6.477e+03, threshold=2.806e+03, percent-clipped=8.0
+2023-03-05 14:49:09,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-05 14:49:09,273 INFO [train.py:968] (0/2) Epoch 11, batch 5400, giga_loss[loss=0.3195, simple_loss=0.383, pruned_loss=0.1279, over 28833.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3522, pruned_loss=0.101, over 5721575.95 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3538, pruned_loss=0.09784, over 5436704.89 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3514, pruned_loss=0.101, over 5712625.17 frames. ], batch size: 285, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:49:22,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2446, 1.6549, 1.5817, 1.1237], device='cuda:0'), covar=tensor([0.1373, 0.2086, 0.1177, 0.1345], device='cuda:0'), in_proj_covar=tensor([0.0821, 0.0696, 0.0859, 0.0769], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 14:49:40,651 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-05 14:49:47,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=460985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:49:50,331 INFO [train.py:968] (0/2) Epoch 11, batch 5450, giga_loss[loss=0.2929, simple_loss=0.359, pruned_loss=0.1134, over 28968.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3511, pruned_loss=0.1022, over 5723834.88 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3542, pruned_loss=0.0981, over 5445156.41 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3502, pruned_loss=0.102, over 5714387.04 frames. ], batch size: 213, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:49:52,782 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3504, 1.6137, 1.3217, 1.4962], device='cuda:0'), covar=tensor([0.0687, 0.0326, 0.0318, 0.0754], device='cuda:0'), in_proj_covar=tensor([0.0178, 0.0112, 0.0115, 0.0209], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 14:50:14,058 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:50:15,818 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:50:24,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.343e+02 1.189e+03 1.559e+03 2.002e+03 7.552e+03, threshold=3.119e+03, percent-clipped=9.0
+2023-03-05 14:50:33,014 INFO [train.py:968] (0/2) Epoch 11, batch 5500, giga_loss[loss=0.2691, simple_loss=0.3378, pruned_loss=0.1002, over 28838.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3505, pruned_loss=0.1033, over 5729731.79 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3548, pruned_loss=0.09846, over 5453938.60 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3492, pruned_loss=0.1029, over 5719265.18 frames. ], batch size: 186, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:50:39,841 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:51:13,382 INFO [train.py:968] (0/2) Epoch 11, batch 5550, giga_loss[loss=0.2408, simple_loss=0.3049, pruned_loss=0.0884, over 28447.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.348, pruned_loss=0.1032, over 5729425.12 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3553, pruned_loss=0.09891, over 5460689.79 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3464, pruned_loss=0.1026, over 5721263.89 frames. ], batch size: 71, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:51:19,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0165, 1.3388, 1.0758, 0.2113], device='cuda:0'), covar=tensor([0.2743, 0.2144, 0.3614, 0.4763], device='cuda:0'), in_proj_covar=tensor([0.1545, 0.1451, 0.1474, 0.1263], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 14:51:44,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:51:45,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.654e+02 1.100e+03 1.411e+03 1.888e+03 4.859e+03, threshold=2.821e+03, percent-clipped=6.0
+2023-03-05 14:51:46,149 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461129.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 14:51:47,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:51:48,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3843, 4.2157, 3.9635, 1.9388], device='cuda:0'), covar=tensor([0.0501, 0.0650, 0.0698, 0.2207], device='cuda:0'), in_proj_covar=tensor([0.1030, 0.0959, 0.0843, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 14:51:53,447 INFO [train.py:968] (0/2) Epoch 11, batch 5600, libri_loss[loss=0.2924, simple_loss=0.3705, pruned_loss=0.1072, over 29550.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3485, pruned_loss=0.1041, over 5733236.68 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3562, pruned_loss=0.0996, over 5471527.73 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3462, pruned_loss=0.1032, over 5722756.38 frames. ], batch size: 83, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:52:11,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:52:11,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:52:21,683 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=461171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:52:35,713 INFO [train.py:968] (0/2) Epoch 11, batch 5650, giga_loss[loss=0.2538, simple_loss=0.325, pruned_loss=0.09136, over 28948.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3465, pruned_loss=0.1034, over 5725130.98 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3565, pruned_loss=0.09985, over 5478128.89 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3443, pruned_loss=0.1025, over 5714337.52 frames. ], batch size: 213, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:52:39,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:52:40,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 14:52:54,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:53:09,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.450e+02 1.118e+03 1.369e+03 2.001e+03 6.081e+03, threshold=2.739e+03, percent-clipped=9.0
+2023-03-05 14:53:16,841 INFO [train.py:968] (0/2) Epoch 11, batch 5700, giga_loss[loss=0.2552, simple_loss=0.3202, pruned_loss=0.09512, over 28936.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.342, pruned_loss=0.1008, over 5723449.86 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3562, pruned_loss=0.09971, over 5488950.87 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3402, pruned_loss=0.1003, over 5710837.87 frames. ], batch size: 136, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:53:36,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4098, 2.0808, 1.7307, 1.7505], device='cuda:0'), covar=tensor([0.0717, 0.0244, 0.0286, 0.0756], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0209], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 14:53:45,021 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461272.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:53:48,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461275.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:53:56,195 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4221, 1.6160, 1.4421, 1.2564], device='cuda:0'), covar=tensor([0.2474, 0.1833, 0.1544, 0.2057], device='cuda:0'), in_proj_covar=tensor([0.1687, 0.1601, 0.1571, 0.1670], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 14:53:59,092 INFO [train.py:968] (0/2) Epoch 11, batch 5750, giga_loss[loss=0.2603, simple_loss=0.3342, pruned_loss=0.09316, over 28221.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3385, pruned_loss=0.09907, over 5719909.66 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3559, pruned_loss=0.09954, over 5497563.53 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.337, pruned_loss=0.0988, over 5706438.46 frames. ], batch size: 368, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:54:09,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1047, 1.2297, 3.7320, 3.0251], device='cuda:0'), covar=tensor([0.1622, 0.2417, 0.0478, 0.0956], device='cuda:0'), in_proj_covar=tensor([0.0648, 0.0579, 0.0844, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 14:54:10,369 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:10,972 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 14:54:11,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-05 14:54:12,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:32,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.829e+02 1.078e+03 1.478e+03 2.345e+03 8.466e+03, threshold=2.955e+03, percent-clipped=20.0
+2023-03-05 14:54:36,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:36,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:38,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:38,677 INFO [train.py:968] (0/2) Epoch 11, batch 5800, giga_loss[loss=0.2278, simple_loss=0.3108, pruned_loss=0.07242, over 28534.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3377, pruned_loss=0.09874, over 5721431.44 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3559, pruned_loss=0.09964, over 5507690.34 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3361, pruned_loss=0.0984, over 5706540.30 frames. ], batch size: 71, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:54:50,992 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:52,925 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:55:01,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:55:19,000 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:55:20,194 INFO [train.py:968] (0/2) Epoch 11, batch 5850, giga_loss[loss=0.2924, simple_loss=0.3566, pruned_loss=0.114, over 28935.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3401, pruned_loss=0.09965, over 5720938.42 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.356, pruned_loss=0.09976, over 5515910.58 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3385, pruned_loss=0.09927, over 5705590.89 frames. ], batch size: 106, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 14:55:24,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5350, 1.5935, 1.3247, 1.2632], device='cuda:0'), covar=tensor([0.0695, 0.0472, 0.0916, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0350, 0.0441, 0.0494, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 14:55:24,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3038, 1.5023, 1.6382, 1.3373], device='cuda:0'), covar=tensor([0.1290, 0.1328, 0.1605, 0.1400], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0732, 0.0672, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 14:55:54,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.965e+02 1.183e+03 1.533e+03 2.072e+03 3.917e+03, threshold=3.066e+03, percent-clipped=8.0
+2023-03-05 14:56:00,959 INFO [train.py:968] (0/2) Epoch 11, batch 5900, giga_loss[loss=0.3072, simple_loss=0.3756, pruned_loss=0.1194, over 28654.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3444, pruned_loss=0.1012, over 5720613.16 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3563, pruned_loss=0.09991, over 5525927.36 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3425, pruned_loss=0.1007, over 5704098.79 frames. ], batch size: 307, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 14:56:21,831 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-05 14:56:21,957 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-05 14:56:40,472 INFO [train.py:968] (0/2) Epoch 11, batch 5950, giga_loss[loss=0.2846, simple_loss=0.3588, pruned_loss=0.1052, over 28893.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3491, pruned_loss=0.103, over 5727073.80 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3571, pruned_loss=0.1005, over 5538745.71 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3466, pruned_loss=0.1023, over 5709102.69 frames. ], batch size: 199, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 14:57:19,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.987e+02 1.143e+03 1.416e+03 1.917e+03 4.043e+03, threshold=2.832e+03, percent-clipped=3.0
+2023-03-05 14:57:26,239 INFO [train.py:968] (0/2) Epoch 11, batch 6000, giga_loss[loss=0.3665, simple_loss=0.4094, pruned_loss=0.1618, over 26616.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3523, pruned_loss=0.1043, over 5722548.71 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3573, pruned_loss=0.1006, over 5545651.37 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.35, pruned_loss=0.1036, over 5705385.60 frames. ], batch size: 555, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:57:26,243 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 14:57:36,261 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2223, simple_loss=0.3277, pruned_loss=0.05847, over 944034.00 frames. 
+2023-03-05 14:57:36,261 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 14:57:41,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:58:19,649 INFO [train.py:968] (0/2) Epoch 11, batch 6050, giga_loss[loss=0.2953, simple_loss=0.3612, pruned_loss=0.1147, over 28574.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3531, pruned_loss=0.1045, over 5713429.17 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3571, pruned_loss=0.1005, over 5544287.87 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3513, pruned_loss=0.1041, over 5704248.87 frames. ], batch size: 71, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:58:48,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=461619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:58:56,652 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=461627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:59:00,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.515e+02 1.191e+03 1.453e+03 2.121e+03 5.864e+03, threshold=2.906e+03, percent-clipped=7.0
+2023-03-05 14:59:07,944 INFO [train.py:968] (0/2) Epoch 11, batch 6100, giga_loss[loss=0.3431, simple_loss=0.4017, pruned_loss=0.1422, over 28329.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3599, pruned_loss=0.1106, over 5698339.37 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3572, pruned_loss=0.1005, over 5538875.89 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3584, pruned_loss=0.1104, over 5696753.05 frames. ], batch size: 368, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:59:36,527 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=461669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:59:55,334 INFO [train.py:968] (0/2) Epoch 11, batch 6150, giga_loss[loss=0.3109, simple_loss=0.3755, pruned_loss=0.1232, over 28776.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3672, pruned_loss=0.1167, over 5695234.49 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3573, pruned_loss=0.1008, over 5545393.04 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3662, pruned_loss=0.1166, over 5692444.16 frames. ], batch size: 66, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 14:59:55,601 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:59:57,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:00:25,952 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:00:35,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.521e+02 1.758e+03 2.251e+03 2.981e+03 1.554e+04, threshold=4.502e+03, percent-clipped=28.0
+2023-03-05 15:00:40,988 INFO [train.py:968] (0/2) Epoch 11, batch 6200, giga_loss[loss=0.2883, simple_loss=0.3592, pruned_loss=0.1087, over 28863.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3731, pruned_loss=0.1207, over 5695226.58 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3569, pruned_loss=0.1005, over 5556581.17 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3731, pruned_loss=0.1215, over 5687450.34 frames. ], batch size: 119, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:01:31,186 INFO [train.py:968] (0/2) Epoch 11, batch 6250, giga_loss[loss=0.3701, simple_loss=0.4202, pruned_loss=0.16, over 28693.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3794, pruned_loss=0.1265, over 5690188.30 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3573, pruned_loss=0.1008, over 5550321.38 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3792, pruned_loss=0.127, over 5690406.99 frames. ], batch size: 262, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:02:10,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.666e+03 2.109e+03 2.819e+03 9.215e+03, threshold=4.218e+03, percent-clipped=2.0
+2023-03-05 15:02:14,831 INFO [train.py:968] (0/2) Epoch 11, batch 6300, giga_loss[loss=0.3352, simple_loss=0.4097, pruned_loss=0.1303, over 28941.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3844, pruned_loss=0.1305, over 5690078.43 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3572, pruned_loss=0.1007, over 5560681.74 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.385, pruned_loss=0.1319, over 5684630.00 frames. ], batch size: 164, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:02:51,163 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5075, 1.5466, 1.1373, 1.2012], device='cuda:0'), covar=tensor([0.0710, 0.0529, 0.0982, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0440, 0.0494, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 15:03:08,762 INFO [train.py:968] (0/2) Epoch 11, batch 6350, giga_loss[loss=0.4507, simple_loss=0.466, pruned_loss=0.2177, over 27567.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3881, pruned_loss=0.134, over 5681852.27 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.357, pruned_loss=0.1005, over 5564298.22 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.389, pruned_loss=0.1355, over 5675419.46 frames. ], batch size: 472, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:03:39,654 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0113, 1.1918, 1.2260, 1.1067], device='cuda:0'), covar=tensor([0.1211, 0.1077, 0.1672, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0733, 0.0671, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 15:03:53,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=461932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:03:53,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.581e+03 2.152e+03 3.760e+03 1.164e+04, threshold=4.304e+03, percent-clipped=20.0
+2023-03-05 15:03:59,054 INFO [train.py:968] (0/2) Epoch 11, batch 6400, giga_loss[loss=0.3387, simple_loss=0.3933, pruned_loss=0.142, over 28707.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3896, pruned_loss=0.1363, over 5667458.90 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.357, pruned_loss=0.1004, over 5570256.41 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3911, pruned_loss=0.1384, over 5659461.79 frames. ], batch size: 242, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:04:44,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4321, 1.6674, 1.3805, 1.2807], device='cuda:0'), covar=tensor([0.2255, 0.2220, 0.2449, 0.1976], device='cuda:0'), in_proj_covar=tensor([0.1274, 0.0953, 0.1132, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 15:04:49,354 INFO [train.py:968] (0/2) Epoch 11, batch 6450, giga_loss[loss=0.3523, simple_loss=0.4005, pruned_loss=0.1521, over 28651.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3931, pruned_loss=0.1404, over 5666161.10 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3571, pruned_loss=0.1005, over 5572014.82 frames. ], giga_tot_loss[loss=0.3406, simple_loss=0.3952, pruned_loss=0.143, over 5660820.29 frames. ], batch size: 307, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:04:56,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:05:03,113 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-462000.pt
+2023-03-05 15:05:06,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=462002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:05:33,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 1.683e+03 2.307e+03 3.188e+03 1.525e+04, threshold=4.614e+03, percent-clipped=8.0
+2023-03-05 15:05:41,341 INFO [train.py:968] (0/2) Epoch 11, batch 6500, giga_loss[loss=0.4621, simple_loss=0.4583, pruned_loss=0.2329, over 23440.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.396, pruned_loss=0.1437, over 5656674.23 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3574, pruned_loss=0.1008, over 5583856.03 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.399, pruned_loss=0.1474, over 5645110.28 frames. ], batch size: 705, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:05:45,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=462044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:06:31,753 INFO [train.py:968] (0/2) Epoch 11, batch 6550, giga_loss[loss=0.2797, simple_loss=0.3501, pruned_loss=0.1046, over 28626.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3977, pruned_loss=0.145, over 5649086.98 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3575, pruned_loss=0.1009, over 5587687.94 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.4009, pruned_loss=0.1489, over 5637973.87 frames. ], batch size: 92, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:06:52,256 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0462, 1.1871, 3.8187, 3.2233], device='cuda:0'), covar=tensor([0.1653, 0.2399, 0.0412, 0.0746], device='cuda:0'), in_proj_covar=tensor([0.0651, 0.0583, 0.0849, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 15:07:15,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.970e+02 1.620e+03 2.074e+03 2.899e+03 7.331e+03, threshold=4.148e+03, percent-clipped=4.0
+2023-03-05 15:07:20,078 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=462137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:07:21,034 INFO [train.py:968] (0/2) Epoch 11, batch 6600, giga_loss[loss=0.3431, simple_loss=0.3949, pruned_loss=0.1457, over 28695.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3972, pruned_loss=0.1456, over 5650089.61 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3574, pruned_loss=0.1009, over 5592965.16 frames. ], giga_tot_loss[loss=0.3503, simple_loss=0.4009, pruned_loss=0.1499, over 5638089.89 frames. ], batch size: 262, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:07:22,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=462140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:07:26,284 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=462145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:07:28,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=462148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:07:33,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3023, 1.7081, 1.5399, 1.4209], device='cuda:0'), covar=tensor([0.1608, 0.1504, 0.1882, 0.1667], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0731, 0.0668, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 15:07:50,610 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=462169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:07:57,291 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=462177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:08:08,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=462187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:08:10,352 INFO [train.py:968] (0/2) Epoch 11, batch 6650, giga_loss[loss=0.4356, simple_loss=0.4555, pruned_loss=0.2079, over 27609.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3958, pruned_loss=0.1452, over 5645030.24 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3572, pruned_loss=0.1007, over 5596103.82 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.3998, pruned_loss=0.1497, over 5633910.19 frames. ], batch size: 472, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:08:11,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=462190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:08:40,258 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=462219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:08:55,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.294e+02 1.730e+03 2.079e+03 2.626e+03 5.576e+03, threshold=4.157e+03, percent-clipped=10.0
+2023-03-05 15:09:00,887 INFO [train.py:968] (0/2) Epoch 11, batch 6700, giga_loss[loss=0.2892, simple_loss=0.3649, pruned_loss=0.1068, over 28964.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3965, pruned_loss=0.1451, over 5635452.34 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3571, pruned_loss=0.1006, over 5592429.33 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4004, pruned_loss=0.1496, over 5630621.03 frames. ], batch size: 213, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:09:49,004 INFO [train.py:968] (0/2) Epoch 11, batch 6750, giga_loss[loss=0.4356, simple_loss=0.4506, pruned_loss=0.2102, over 27892.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3955, pruned_loss=0.1434, over 5644073.66 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3568, pruned_loss=0.1004, over 5601797.65 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.4001, pruned_loss=0.1484, over 5633017.01 frames. ], batch size: 412, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:10:07,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=462307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:10:33,663 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.599e+03 2.051e+03 2.863e+03 1.234e+04, threshold=4.103e+03, percent-clipped=8.0
+2023-03-05 15:10:39,979 INFO [train.py:968] (0/2) Epoch 11, batch 6800, giga_loss[loss=0.2742, simple_loss=0.3543, pruned_loss=0.09701, over 28956.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3966, pruned_loss=0.144, over 5632682.64 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3569, pruned_loss=0.1006, over 5603105.91 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4007, pruned_loss=0.1483, over 5623098.66 frames. ], batch size: 164, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:11:32,908 INFO [train.py:968] (0/2) Epoch 11, batch 6850, libri_loss[loss=0.3369, simple_loss=0.4055, pruned_loss=0.1342, over 27955.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.394, pruned_loss=0.1411, over 5631973.37 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3573, pruned_loss=0.1008, over 5608641.81 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.3977, pruned_loss=0.1454, over 5620076.71 frames. ], batch size: 116, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:12:13,044 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.018e+02 1.517e+03 1.924e+03 2.576e+03 8.721e+03, threshold=3.849e+03, percent-clipped=10.0
+2023-03-05 15:12:16,206 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=462436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:12:18,462 INFO [train.py:968] (0/2) Epoch 11, batch 6900, giga_loss[loss=0.3051, simple_loss=0.3816, pruned_loss=0.1143, over 28893.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3917, pruned_loss=0.1381, over 5639644.16 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3572, pruned_loss=0.1008, over 5608357.31 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3961, pruned_loss=0.1429, over 5630766.61 frames. ], batch size: 145, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:12:28,636 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=462450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:12:32,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=462453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:12:48,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7303, 4.6442, 1.9596, 1.6904], device='cuda:0'), covar=tensor([0.0885, 0.0361, 0.0803, 0.1322], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0505, 0.0335, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 15:12:59,933 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=462482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:13:09,620 INFO [train.py:968] (0/2) Epoch 11, batch 6950, giga_loss[loss=0.4133, simple_loss=0.4331, pruned_loss=0.1968, over 26586.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.388, pruned_loss=0.1346, over 5637861.34 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3569, pruned_loss=0.1007, over 5604955.02 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3925, pruned_loss=0.1393, over 5634203.34 frames. ], batch size: 555, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:13:52,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.791e+02 1.798e+03 2.220e+03 3.021e+03 9.676e+03, threshold=4.440e+03, percent-clipped=14.0
+2023-03-05 15:13:58,121 INFO [train.py:968] (0/2) Epoch 11, batch 7000, giga_loss[loss=0.3048, simple_loss=0.3749, pruned_loss=0.1173, over 28616.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3839, pruned_loss=0.131, over 5651138.31 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3568, pruned_loss=0.1007, over 5609870.41 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3879, pruned_loss=0.1351, over 5644409.51 frames. ], batch size: 336, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:14:08,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9616, 2.7041, 2.0046, 1.5285], device='cuda:0'), covar=tensor([0.2285, 0.1282, 0.1665, 0.2105], device='cuda:0'), in_proj_covar=tensor([0.1677, 0.1605, 0.1564, 0.1672], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 15:14:45,991 INFO [train.py:968] (0/2) Epoch 11, batch 7050, giga_loss[loss=0.3046, simple_loss=0.3741, pruned_loss=0.1176, over 28853.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3825, pruned_loss=0.1306, over 5650019.75 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3565, pruned_loss=0.1006, over 5617040.86 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3865, pruned_loss=0.1346, over 5639313.34 frames. ], batch size: 284, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:15:24,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.506e+03 1.997e+03 2.697e+03 6.592e+03, threshold=3.995e+03, percent-clipped=3.0
+2023-03-05 15:15:27,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=462633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:15:32,099 INFO [train.py:968] (0/2) Epoch 11, batch 7100, giga_loss[loss=0.3224, simple_loss=0.3904, pruned_loss=0.1272, over 29101.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3819, pruned_loss=0.1301, over 5654939.24 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.356, pruned_loss=0.1003, over 5627337.35 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3863, pruned_loss=0.1346, over 5638399.11 frames. ], batch size: 128, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:16:22,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-05 15:16:31,178 INFO [train.py:968] (0/2) Epoch 11, batch 7150, giga_loss[loss=0.3232, simple_loss=0.3826, pruned_loss=0.1319, over 28933.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3814, pruned_loss=0.1297, over 5645802.78 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3562, pruned_loss=0.1005, over 5619219.27 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3851, pruned_loss=0.1334, over 5639518.32 frames. ], batch size: 164, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:16:36,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2849, 1.5351, 1.2950, 1.2570], device='cuda:0'), covar=tensor([0.1824, 0.1399, 0.1476, 0.1509], device='cuda:0'), in_proj_covar=tensor([0.1681, 0.1602, 0.1562, 0.1673], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 15:16:38,808 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=462695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:16:43,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6929, 1.6821, 1.2735, 1.3559], device='cuda:0'), covar=tensor([0.0781, 0.0670, 0.1078, 0.1050], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0443, 0.0494, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 15:16:55,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-05 15:17:19,347 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.612e+02 1.495e+03 1.964e+03 2.446e+03 6.739e+03, threshold=3.928e+03, percent-clipped=5.0
+2023-03-05 15:17:24,825 INFO [train.py:968] (0/2) Epoch 11, batch 7200, giga_loss[loss=0.3825, simple_loss=0.4461, pruned_loss=0.1594, over 28649.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3813, pruned_loss=0.128, over 5653147.04 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3564, pruned_loss=0.1006, over 5621630.25 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3843, pruned_loss=0.1312, over 5646352.25 frames. ], batch size: 262, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 15:18:21,761 INFO [train.py:968] (0/2) Epoch 11, batch 7250, giga_loss[loss=0.2962, simple_loss=0.377, pruned_loss=0.1078, over 28892.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3812, pruned_loss=0.1257, over 5653638.52 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3556, pruned_loss=0.1003, over 5618927.12 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3849, pruned_loss=0.1292, over 5651221.92 frames. ], batch size: 213, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:18:43,248 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=462811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:19:05,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.696e+03 2.517e+03 3.240e+03 6.999e+03, threshold=5.035e+03, percent-clipped=13.0
+2023-03-05 15:19:10,700 INFO [train.py:968] (0/2) Epoch 11, batch 7300, giga_loss[loss=0.4107, simple_loss=0.4481, pruned_loss=0.1866, over 27863.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3828, pruned_loss=0.1262, over 5658271.88 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3557, pruned_loss=0.1003, over 5614904.66 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3862, pruned_loss=0.1294, over 5661051.73 frames. ], batch size: 412, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:20:01,505 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-05 15:20:02,667 INFO [train.py:968] (0/2) Epoch 11, batch 7350, giga_loss[loss=0.3979, simple_loss=0.425, pruned_loss=0.1854, over 27520.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3824, pruned_loss=0.1268, over 5654612.48 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3554, pruned_loss=0.1001, over 5619891.24 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3859, pruned_loss=0.1301, over 5652751.67 frames. ], batch size: 472, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:20:45,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.271e+02 1.451e+03 1.820e+03 2.361e+03 6.040e+03, threshold=3.640e+03, percent-clipped=3.0
+2023-03-05 15:20:49,671 INFO [train.py:968] (0/2) Epoch 11, batch 7400, giga_loss[loss=0.3447, simple_loss=0.3901, pruned_loss=0.1497, over 28922.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3816, pruned_loss=0.1268, over 5670066.21 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3558, pruned_loss=0.1003, over 5629210.52 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3849, pruned_loss=0.1301, over 5661263.35 frames. ], batch size: 106, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:21:08,062 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=462954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:21:11,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=462957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:21:17,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6614, 5.4431, 5.1767, 2.6352], device='cuda:0'), covar=tensor([0.0457, 0.0648, 0.0670, 0.1765], device='cuda:0'), in_proj_covar=tensor([0.1052, 0.0990, 0.0866, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 15:21:30,333 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3628, 1.9276, 1.5849, 1.6147], device='cuda:0'), covar=tensor([0.0614, 0.0232, 0.0247, 0.0616], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 15:21:35,805 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=462986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:21:39,152 INFO [train.py:968] (0/2) Epoch 11, batch 7450, giga_loss[loss=0.2882, simple_loss=0.3458, pruned_loss=0.1152, over 28746.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3807, pruned_loss=0.1273, over 5672468.77 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3561, pruned_loss=0.1003, over 5637093.68 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3838, pruned_loss=0.1308, over 5659426.78 frames. ], batch size: 99, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:21:58,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:22:22,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.936e+02 1.676e+03 2.057e+03 2.880e+03 7.308e+03, threshold=4.114e+03, percent-clipped=13.0
+2023-03-05 15:22:26,937 INFO [train.py:968] (0/2) Epoch 11, batch 7500, giga_loss[loss=0.2891, simple_loss=0.3648, pruned_loss=0.1067, over 28800.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3794, pruned_loss=0.1272, over 5676405.32 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3559, pruned_loss=0.1002, over 5639792.76 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3822, pruned_loss=0.1302, over 5664099.78 frames. ], batch size: 119, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:23:01,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5411, 3.3629, 3.1771, 1.6592], device='cuda:0'), covar=tensor([0.0738, 0.0823, 0.0798, 0.2478], device='cuda:0'), in_proj_covar=tensor([0.1045, 0.0980, 0.0859, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 15:23:02,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:23:19,569 INFO [train.py:968] (0/2) Epoch 11, batch 7550, giga_loss[loss=0.3133, simple_loss=0.3813, pruned_loss=0.1227, over 28640.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3789, pruned_loss=0.1259, over 5668410.36 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3556, pruned_loss=0.09999, over 5646232.74 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.382, pruned_loss=0.1291, over 5653500.18 frames. ], batch size: 242, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:23:23,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:24:00,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.041e+02 1.262e+03 1.681e+03 2.134e+03 5.872e+03, threshold=3.363e+03, percent-clipped=2.0
+2023-03-05 15:24:04,578 INFO [train.py:968] (0/2) Epoch 11, batch 7600, giga_loss[loss=0.3092, simple_loss=0.3767, pruned_loss=0.1208, over 28932.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3788, pruned_loss=0.1247, over 5679857.27 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3558, pruned_loss=0.1001, over 5653582.94 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3817, pruned_loss=0.1278, over 5661897.40 frames. ], batch size: 213, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 15:24:19,525 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:24:19,583 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:24:21,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=463154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:24:38,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9258, 3.7213, 3.5301, 1.5234], device='cuda:0'), covar=tensor([0.0665, 0.0817, 0.0837, 0.2284], device='cuda:0'), in_proj_covar=tensor([0.1041, 0.0974, 0.0853, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 15:24:47,649 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=463183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:24:52,883 INFO [train.py:968] (0/2) Epoch 11, batch 7650, giga_loss[loss=0.2933, simple_loss=0.3657, pruned_loss=0.1105, over 28852.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3791, pruned_loss=0.1248, over 5683647.55 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3557, pruned_loss=0.09999, over 5653890.32 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3818, pruned_loss=0.1278, over 5669505.66 frames. ], batch size: 199, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 15:25:14,862 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:25:18,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=463216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:25:37,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.322e+02 1.498e+03 2.077e+03 2.661e+03 5.525e+03, threshold=4.154e+03, percent-clipped=11.0
+2023-03-05 15:25:43,243 INFO [train.py:968] (0/2) Epoch 11, batch 7700, giga_loss[loss=0.3143, simple_loss=0.3754, pruned_loss=0.1266, over 28649.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3771, pruned_loss=0.1239, over 5683842.36 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3556, pruned_loss=0.09984, over 5655874.87 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3797, pruned_loss=0.1267, over 5671484.88 frames. ], batch size: 242, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 15:25:47,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8479, 3.6544, 3.4608, 2.0372], device='cuda:0'), covar=tensor([0.0561, 0.0703, 0.0677, 0.1785], device='cuda:0'), in_proj_covar=tensor([0.1042, 0.0974, 0.0854, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 15:25:49,915 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=463245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:26:34,936 INFO [train.py:968] (0/2) Epoch 11, batch 7750, giga_loss[loss=0.3008, simple_loss=0.3662, pruned_loss=0.1177, over 29009.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3749, pruned_loss=0.1235, over 5667088.30 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3556, pruned_loss=0.09985, over 5656785.49 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.377, pruned_loss=0.1257, over 5656780.36 frames. ], batch size: 164, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:27:07,349 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-05 15:27:18,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.900e+02 1.728e+03 2.431e+03 3.124e+03 7.029e+03, threshold=4.862e+03, percent-clipped=13.0
+2023-03-05 15:27:21,548 INFO [train.py:968] (0/2) Epoch 11, batch 7800, giga_loss[loss=0.2691, simple_loss=0.3463, pruned_loss=0.09594, over 28876.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3745, pruned_loss=0.124, over 5659664.70 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3561, pruned_loss=0.1, over 5655922.44 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3764, pruned_loss=0.1264, over 5653051.92 frames. ], batch size: 119, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:27:34,494 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-05 15:27:35,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6803, 1.0690, 2.8133, 2.5389], device='cuda:0'), covar=tensor([0.2063, 0.2565, 0.0994, 0.1733], device='cuda:0'), in_proj_covar=tensor([0.0654, 0.0588, 0.0855, 0.0757], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 15:27:39,935 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2603, 4.0593, 3.7986, 1.7109], device='cuda:0'), covar=tensor([0.0702, 0.0842, 0.0990, 0.2309], device='cuda:0'), in_proj_covar=tensor([0.1052, 0.0985, 0.0863, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 15:27:47,473 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:28:12,329 INFO [train.py:968] (0/2) Epoch 11, batch 7850, giga_loss[loss=0.2753, simple_loss=0.3493, pruned_loss=0.1007, over 28814.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3743, pruned_loss=0.1245, over 5659703.88 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3563, pruned_loss=0.1001, over 5659988.47 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.376, pruned_loss=0.1269, over 5650723.33 frames. ], batch size: 243, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:28:52,214 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.38 vs. limit=5.0
+2023-03-05 15:28:53,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.278e+02 1.686e+03 2.160e+03 3.366e+03 9.226e+03, threshold=4.320e+03, percent-clipped=5.0
+2023-03-05 15:28:55,798 INFO [train.py:968] (0/2) Epoch 11, batch 7900, giga_loss[loss=0.2928, simple_loss=0.3501, pruned_loss=0.1177, over 28612.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3723, pruned_loss=0.1233, over 5663445.77 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.356, pruned_loss=0.09981, over 5667250.41 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3747, pruned_loss=0.1266, over 5649277.19 frames. ], batch size: 85, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:29:09,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:29:24,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:29:41,917 INFO [train.py:968] (0/2) Epoch 11, batch 7950, giga_loss[loss=0.3205, simple_loss=0.3904, pruned_loss=0.1254, over 28715.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.372, pruned_loss=0.1231, over 5663010.72 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3563, pruned_loss=0.1001, over 5670809.99 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.374, pruned_loss=0.1261, over 5648442.32 frames. ], batch size: 262, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:29:49,282 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-05 15:29:54,692 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:30:19,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:30:28,721 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.104e+02 1.521e+03 1.993e+03 2.615e+03 5.738e+03, threshold=3.985e+03, percent-clipped=6.0
+2023-03-05 15:30:32,118 INFO [train.py:968] (0/2) Epoch 11, batch 8000, giga_loss[loss=0.2956, simple_loss=0.3675, pruned_loss=0.1118, over 28900.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3737, pruned_loss=0.124, over 5664100.32 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3565, pruned_loss=0.1003, over 5672390.15 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3753, pruned_loss=0.1264, over 5651255.20 frames. ], batch size: 145, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:31:20,384 INFO [train.py:968] (0/2) Epoch 11, batch 8050, giga_loss[loss=0.3131, simple_loss=0.3827, pruned_loss=0.1218, over 28676.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3736, pruned_loss=0.1228, over 5672004.59 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3564, pruned_loss=0.1002, over 5675843.14 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3752, pruned_loss=0.1252, over 5658685.88 frames. ], batch size: 262, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:31:35,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5042, 3.4816, 1.6020, 1.6075], device='cuda:0'), covar=tensor([0.0895, 0.0361, 0.0852, 0.1268], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0508, 0.0337, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 15:31:41,236 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:31:43,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=463615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:32:06,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.410e+02 1.435e+03 1.840e+03 2.321e+03 8.689e+03, threshold=3.679e+03, percent-clipped=3.0
+2023-03-05 15:32:08,270 INFO [train.py:968] (0/2) Epoch 11, batch 8100, libri_loss[loss=0.2774, simple_loss=0.3634, pruned_loss=0.09565, over 29711.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3723, pruned_loss=0.1209, over 5686510.41 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3559, pruned_loss=0.09991, over 5683117.80 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3746, pruned_loss=0.1237, over 5669263.49 frames. ], batch size: 91, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:32:11,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:32:12,042 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=463644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:32:33,996 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:32:37,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=463672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:32:53,763 INFO [train.py:968] (0/2) Epoch 11, batch 8150, giga_loss[loss=0.3516, simple_loss=0.4009, pruned_loss=0.1512, over 28426.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3732, pruned_loss=0.122, over 5685724.83 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3558, pruned_loss=0.09983, over 5683834.14 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3754, pruned_loss=0.1246, over 5671468.58 frames. ], batch size: 368, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:33:05,806 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=463701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:33:14,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1499, 0.8803, 0.8910, 1.3646], device='cuda:0'), covar=tensor([0.0691, 0.0434, 0.0330, 0.0715], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 15:33:42,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.086e+03 1.538e+03 1.864e+03 2.751e+03 1.770e+04, threshold=3.728e+03, percent-clipped=16.0
+2023-03-05 15:33:43,392 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 15:33:43,830 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:33:44,191 INFO [train.py:968] (0/2) Epoch 11, batch 8200, giga_loss[loss=0.3402, simple_loss=0.402, pruned_loss=0.1392, over 28653.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.376, pruned_loss=0.1249, over 5678813.02 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3553, pruned_loss=0.09956, over 5688074.94 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3786, pruned_loss=0.1277, over 5663569.51 frames. ], batch size: 262, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:34:39,299 INFO [train.py:968] (0/2) Epoch 11, batch 8250, giga_loss[loss=0.2714, simple_loss=0.3369, pruned_loss=0.1029, over 28532.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3786, pruned_loss=0.1292, over 5661687.49 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3552, pruned_loss=0.09943, over 5689603.13 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.381, pruned_loss=0.1319, over 5648096.76 frames. ], batch size: 71, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:34:56,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0939, 4.8925, 4.6264, 2.0388], device='cuda:0'), covar=tensor([0.0490, 0.0683, 0.0749, 0.2174], device='cuda:0'), in_proj_covar=tensor([0.1050, 0.0987, 0.0866, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 15:35:04,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3402, 1.7244, 1.4508, 1.6885], device='cuda:0'), covar=tensor([0.0754, 0.0296, 0.0300, 0.0784], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0079, 0.0057, 0.0051, 0.0086], device='cuda:0')
+2023-03-05 15:35:21,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:35:28,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.161e+02 1.660e+03 1.963e+03 2.713e+03 6.548e+03, threshold=3.925e+03, percent-clipped=7.0
+2023-03-05 15:35:29,592 INFO [train.py:968] (0/2) Epoch 11, batch 8300, giga_loss[loss=0.3104, simple_loss=0.3639, pruned_loss=0.1285, over 28600.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3782, pruned_loss=0.1294, over 5672053.56 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3551, pruned_loss=0.09928, over 5688297.97 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3808, pruned_loss=0.1324, over 5661910.26 frames. ], batch size: 92, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:35:48,473 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:36:09,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:36:13,372 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:36:16,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=463884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:36:20,568 INFO [train.py:968] (0/2) Epoch 11, batch 8350, giga_loss[loss=0.3009, simple_loss=0.3633, pruned_loss=0.1192, over 28260.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3803, pruned_loss=0.1314, over 5666721.86 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3551, pruned_loss=0.09928, over 5690218.86 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3825, pruned_loss=0.1341, over 5656724.69 frames. ], batch size: 368, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:36:43,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=463913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:37:03,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.189e+02 1.560e+03 2.077e+03 3.015e+03 6.597e+03, threshold=4.154e+03, percent-clipped=13.0
+2023-03-05 15:37:06,138 INFO [train.py:968] (0/2) Epoch 11, batch 8400, giga_loss[loss=0.3541, simple_loss=0.4058, pruned_loss=0.1512, over 28593.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3786, pruned_loss=0.1296, over 5672323.37 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3553, pruned_loss=0.09934, over 5697052.99 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.381, pruned_loss=0.1327, over 5657510.14 frames. ], batch size: 307, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:37:16,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:37:36,395 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:37:39,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=463975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:37:50,962 INFO [train.py:968] (0/2) Epoch 11, batch 8450, giga_loss[loss=0.302, simple_loss=0.3599, pruned_loss=0.122, over 27651.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3763, pruned_loss=0.1264, over 5684690.31 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3548, pruned_loss=0.09908, over 5700401.95 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.379, pruned_loss=0.1295, over 5669571.23 frames. ], batch size: 472, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:38:02,435 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-464000.pt
+2023-03-05 15:38:06,368 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:38:19,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:38:20,982 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:38:23,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:38:36,475 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.250e+02 1.452e+03 1.798e+03 2.459e+03 7.075e+03, threshold=3.597e+03, percent-clipped=4.0
+2023-03-05 15:38:37,991 INFO [train.py:968] (0/2) Epoch 11, batch 8500, giga_loss[loss=0.2818, simple_loss=0.3557, pruned_loss=0.1039, over 28824.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3745, pruned_loss=0.1245, over 5683582.57 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3552, pruned_loss=0.09938, over 5705663.41 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3767, pruned_loss=0.1273, over 5666588.43 frames. ], batch size: 213, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:38:49,742 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:39:23,139 INFO [train.py:968] (0/2) Epoch 11, batch 8550, giga_loss[loss=0.3428, simple_loss=0.397, pruned_loss=0.1443, over 28884.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3723, pruned_loss=0.1231, over 5688184.74 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3554, pruned_loss=0.09949, over 5710540.28 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3743, pruned_loss=0.1258, over 5669731.43 frames. ], batch size: 199, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:39:36,203 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:40:00,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7519, 1.8275, 1.6550, 1.6817], device='cuda:0'), covar=tensor([0.1478, 0.2085, 0.2136, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0733, 0.0669, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 15:40:07,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.305e+02 1.516e+03 2.004e+03 3.057e+03 1.256e+04, threshold=4.008e+03, percent-clipped=14.0
+2023-03-05 15:40:08,257 INFO [train.py:968] (0/2) Epoch 11, batch 8600, giga_loss[loss=0.3196, simple_loss=0.382, pruned_loss=0.1286, over 28895.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3707, pruned_loss=0.1224, over 5685854.09 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3558, pruned_loss=0.09965, over 5713921.76 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3722, pruned_loss=0.1248, over 5667620.04 frames. ], batch size: 145, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:40:19,869 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-05 15:40:32,021 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:40:34,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:40:56,870 INFO [train.py:968] (0/2) Epoch 11, batch 8650, giga_loss[loss=0.3423, simple_loss=0.4053, pruned_loss=0.1397, over 28784.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.123, over 5686013.36 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3556, pruned_loss=0.0995, over 5720489.69 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3733, pruned_loss=0.126, over 5663899.97 frames. ], batch size: 284, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:40:57,088 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:41:03,116 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:41:05,549 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 15:41:08,876 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:41:40,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:41:46,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.064e+02 1.496e+03 1.964e+03 2.961e+03 1.977e+04, threshold=3.929e+03, percent-clipped=14.0
+2023-03-05 15:41:46,976 INFO [train.py:968] (0/2) Epoch 11, batch 8700, giga_loss[loss=0.3464, simple_loss=0.4039, pruned_loss=0.1445, over 28881.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3765, pruned_loss=0.1264, over 5669022.64 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3561, pruned_loss=0.1, over 5704518.31 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3779, pruned_loss=0.1287, over 5665287.08 frames. ], batch size: 186, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:41:47,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.32 vs. limit=5.0
+2023-03-05 15:42:19,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1660, 1.5451, 1.2264, 0.5896], device='cuda:0'), covar=tensor([0.2234, 0.1604, 0.1923, 0.3239], device='cuda:0'), in_proj_covar=tensor([0.1538, 0.1461, 0.1478, 0.1259], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 15:42:34,381 INFO [train.py:968] (0/2) Epoch 11, batch 8750, giga_loss[loss=0.3075, simple_loss=0.3832, pruned_loss=0.1159, over 28809.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3794, pruned_loss=0.1255, over 5669915.22 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3561, pruned_loss=0.1001, over 5701875.41 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3814, pruned_loss=0.1281, over 5667542.85 frames. ], batch size: 284, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:43:07,078 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-05 15:43:07,489 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:43:15,853 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.366e+02 1.420e+03 1.832e+03 2.395e+03 5.088e+03, threshold=3.664e+03, percent-clipped=3.0
+2023-03-05 15:43:16,929 INFO [train.py:968] (0/2) Epoch 11, batch 8800, giga_loss[loss=0.3329, simple_loss=0.3949, pruned_loss=0.1355, over 28748.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3795, pruned_loss=0.1249, over 5679995.19 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3548, pruned_loss=0.09956, over 5710187.63 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3835, pruned_loss=0.1288, over 5668736.94 frames. ], batch size: 262, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:43:51,028 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:43:52,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:44:01,467 INFO [train.py:968] (0/2) Epoch 11, batch 8850, giga_loss[loss=0.3344, simple_loss=0.3939, pruned_loss=0.1375, over 28956.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3809, pruned_loss=0.1258, over 5685518.89 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3546, pruned_loss=0.09941, over 5714790.16 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.385, pruned_loss=0.1298, over 5671249.01 frames. ], batch size: 227, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:44:15,779 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:44:46,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.313e+02 1.603e+03 1.952e+03 2.606e+03 9.830e+03, threshold=3.903e+03, percent-clipped=7.0
+2023-03-05 15:44:47,241 INFO [train.py:968] (0/2) Epoch 11, batch 8900, giga_loss[loss=0.3053, simple_loss=0.3688, pruned_loss=0.1209, over 28789.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3811, pruned_loss=0.1264, over 5691573.83 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3548, pruned_loss=0.09949, over 5715484.86 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3846, pruned_loss=0.1298, over 5679415.91 frames. ], batch size: 112, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:44:54,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9606, 1.2543, 3.4060, 2.9377], device='cuda:0'), covar=tensor([0.1750, 0.2505, 0.0510, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.0661, 0.0593, 0.0864, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 15:45:17,056 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:45:19,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:45:25,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:45:35,938 INFO [train.py:968] (0/2) Epoch 11, batch 8950, giga_loss[loss=0.362, simple_loss=0.4127, pruned_loss=0.1556, over 28996.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3813, pruned_loss=0.1281, over 5686975.01 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3545, pruned_loss=0.09934, over 5718136.11 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3847, pruned_loss=0.1313, over 5674562.74 frames. ], batch size: 227, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:45:46,549 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:46:01,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5787, 2.7069, 1.8822, 2.2707], device='cuda:0'), covar=tensor([0.0664, 0.0472, 0.0904, 0.0916], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0445, 0.0499, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 15:46:20,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.711e+02 1.506e+03 2.056e+03 2.779e+03 6.029e+03, threshold=4.112e+03, percent-clipped=8.0
+2023-03-05 15:46:20,107 INFO [train.py:968] (0/2) Epoch 11, batch 9000, giga_loss[loss=0.3014, simple_loss=0.3658, pruned_loss=0.1185, over 28794.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3801, pruned_loss=0.1274, over 5697637.52 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3547, pruned_loss=0.09943, over 5726704.23 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3839, pruned_loss=0.1313, over 5678375.44 frames. ], batch size: 119, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:46:20,110 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 15:46:28,969 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2151, simple_loss=0.3204, pruned_loss=0.05491, over 944034.00 frames. 
+2023-03-05 15:46:28,970 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 15:46:43,822 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-05 15:46:51,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:46:53,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3369, 1.8901, 1.4685, 0.6019], device='cuda:0'), covar=tensor([0.3006, 0.1738, 0.2640, 0.3850], device='cuda:0'), in_proj_covar=tensor([0.1535, 0.1464, 0.1478, 0.1259], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 15:47:00,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:47:16,829 INFO [train.py:968] (0/2) Epoch 11, batch 9050, giga_loss[loss=0.2901, simple_loss=0.3575, pruned_loss=0.1113, over 28556.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3791, pruned_loss=0.128, over 5686567.23 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3548, pruned_loss=0.09951, over 5723857.71 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3821, pruned_loss=0.1312, over 5673930.65 frames. ], batch size: 85, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:47:26,244 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464597.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 15:47:50,461 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:47:55,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:48:08,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.721e+03 2.083e+03 2.643e+03 4.802e+03, threshold=4.167e+03, percent-clipped=3.0
+2023-03-05 15:48:08,123 INFO [train.py:968] (0/2) Epoch 11, batch 9100, giga_loss[loss=0.3383, simple_loss=0.3953, pruned_loss=0.1407, over 27964.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3777, pruned_loss=0.128, over 5679954.68 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3546, pruned_loss=0.09948, over 5726518.35 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3805, pruned_loss=0.1309, over 5666906.11 frames. ], batch size: 412, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:48:23,207 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:48:56,114 INFO [train.py:968] (0/2) Epoch 11, batch 9150, giga_loss[loss=0.3343, simple_loss=0.3854, pruned_loss=0.1416, over 28285.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3773, pruned_loss=0.1274, over 5687570.57 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3547, pruned_loss=0.09945, over 5728791.79 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3803, pruned_loss=0.1307, over 5673560.03 frames. ], batch size: 77, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:49:13,341 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:49:16,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:49:23,640 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:49:26,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:49:29,744 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9372, 1.0553, 1.0531, 0.8487], device='cuda:0'), covar=tensor([0.1701, 0.1819, 0.1003, 0.1538], device='cuda:0'), in_proj_covar=tensor([0.1692, 0.1611, 0.1572, 0.1688], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 15:49:42,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.619e+03 2.094e+03 3.352e+03 6.196e+03, threshold=4.188e+03, percent-clipped=12.0
+2023-03-05 15:49:42,950 INFO [train.py:968] (0/2) Epoch 11, batch 9200, libri_loss[loss=0.2642, simple_loss=0.3493, pruned_loss=0.08953, over 28680.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.377, pruned_loss=0.1277, over 5682555.22 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3544, pruned_loss=0.09928, over 5732819.65 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3806, pruned_loss=0.1316, over 5665956.58 frames. ], batch size: 106, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:49:43,176 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:49:52,351 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:50:28,569 INFO [train.py:968] (0/2) Epoch 11, batch 9250, libri_loss[loss=0.2554, simple_loss=0.3413, pruned_loss=0.08474, over 29515.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.375, pruned_loss=0.1264, over 5689197.24 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3541, pruned_loss=0.09909, over 5736059.82 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3784, pruned_loss=0.1301, over 5672275.23 frames. ], batch size: 82, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:50:50,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5756, 1.7127, 1.4233, 1.3435], device='cuda:0'), covar=tensor([0.1984, 0.1722, 0.1509, 0.1761], device='cuda:0'), in_proj_covar=tensor([0.1702, 0.1623, 0.1585, 0.1699], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 15:50:54,988 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-05 15:51:15,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.553e+03 1.950e+03 2.386e+03 5.707e+03, threshold=3.901e+03, percent-clipped=4.0
+2023-03-05 15:51:15,471 INFO [train.py:968] (0/2) Epoch 11, batch 9300, giga_loss[loss=0.3448, simple_loss=0.407, pruned_loss=0.1413, over 28880.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.375, pruned_loss=0.1264, over 5689938.69 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3538, pruned_loss=0.09882, over 5736562.11 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3789, pruned_loss=0.1308, over 5673839.99 frames. ], batch size: 199, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:51:16,250 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:51:46,138 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9957, 2.2896, 1.9399, 1.9055], device='cuda:0'), covar=tensor([0.1838, 0.1601, 0.1642, 0.1444], device='cuda:0'), in_proj_covar=tensor([0.1693, 0.1610, 0.1571, 0.1686], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 15:52:03,997 INFO [train.py:968] (0/2) Epoch 11, batch 9350, giga_loss[loss=0.3899, simple_loss=0.4239, pruned_loss=0.1779, over 27594.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3769, pruned_loss=0.1273, over 5683711.64 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3538, pruned_loss=0.09882, over 5741310.68 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3806, pruned_loss=0.1315, over 5665406.43 frames. ], batch size: 472, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:52:12,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1825, 1.5366, 1.4124, 1.3221], device='cuda:0'), covar=tensor([0.0824, 0.0294, 0.0277, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 15:52:37,440 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:52:48,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.627e+02 1.489e+03 2.069e+03 2.967e+03 7.200e+03, threshold=4.139e+03, percent-clipped=10.0
+2023-03-05 15:52:48,801 INFO [train.py:968] (0/2) Epoch 11, batch 9400, giga_loss[loss=0.2986, simple_loss=0.3701, pruned_loss=0.1135, over 28856.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3788, pruned_loss=0.1284, over 5690104.40 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3539, pruned_loss=0.09885, over 5743749.27 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.382, pruned_loss=0.1321, over 5672746.77 frames. ], batch size: 199, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:53:18,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464972.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 15:53:34,306 INFO [train.py:968] (0/2) Epoch 11, batch 9450, libri_loss[loss=0.3116, simple_loss=0.3736, pruned_loss=0.1248, over 19924.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3781, pruned_loss=0.1282, over 5673542.30 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3538, pruned_loss=0.09875, over 5735750.95 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3815, pruned_loss=0.1322, over 5665493.62 frames. ], batch size: 186, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:53:49,428 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2828, 1.6868, 1.3352, 1.3341], device='cuda:0'), covar=tensor([0.1575, 0.1583, 0.1805, 0.1823], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0737, 0.0671, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 15:54:16,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.824e+02 1.432e+03 1.753e+03 2.269e+03 8.970e+03, threshold=3.507e+03, percent-clipped=4.0
+2023-03-05 15:54:16,286 INFO [train.py:968] (0/2) Epoch 11, batch 9500, giga_loss[loss=0.3045, simple_loss=0.3767, pruned_loss=0.1162, over 28741.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.379, pruned_loss=0.1264, over 5678241.33 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3538, pruned_loss=0.09878, over 5732784.71 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3829, pruned_loss=0.1309, over 5671028.34 frames. ], batch size: 284, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:54:58,654 INFO [train.py:968] (0/2) Epoch 11, batch 9550, giga_loss[loss=0.2906, simple_loss=0.3721, pruned_loss=0.1045, over 28562.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3798, pruned_loss=0.125, over 5676755.36 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.354, pruned_loss=0.09904, over 5726523.67 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3836, pruned_loss=0.1293, over 5675113.09 frames. ], batch size: 78, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:55:00,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9981, 1.3442, 1.0760, 0.1334], device='cuda:0'), covar=tensor([0.2367, 0.2085, 0.3042, 0.4479], device='cuda:0'), in_proj_covar=tensor([0.1524, 0.1451, 0.1469, 0.1251], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 15:55:22,669 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=465115.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 15:55:24,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=465118.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 15:55:45,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.407e+02 1.498e+03 2.191e+03 2.940e+03 7.980e+03, threshold=4.382e+03, percent-clipped=12.0
+2023-03-05 15:55:45,826 INFO [train.py:968] (0/2) Epoch 11, batch 9600, giga_loss[loss=0.4632, simple_loss=0.4625, pruned_loss=0.2319, over 23496.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3831, pruned_loss=0.1265, over 5672372.38 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3541, pruned_loss=0.09904, over 5728940.41 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3867, pruned_loss=0.1306, over 5667627.06 frames. ], batch size: 705, lr: 2.98e-03, grad_scale: 8.0
+2023-03-05 15:55:52,636 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=465147.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 15:55:56,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9402, 5.1195, 1.9857, 2.1464], device='cuda:0'), covar=tensor([0.0816, 0.0210, 0.0814, 0.1182], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0510, 0.0336, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 15:56:31,623 INFO [train.py:968] (0/2) Epoch 11, batch 9650, giga_loss[loss=0.2852, simple_loss=0.3525, pruned_loss=0.1089, over 28604.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.386, pruned_loss=0.1296, over 5679888.48 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.354, pruned_loss=0.099, over 5732476.57 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3894, pruned_loss=0.1333, over 5672157.63 frames. ], batch size: 92, lr: 2.98e-03, grad_scale: 8.0
+2023-03-05 15:56:39,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3553, 1.5005, 1.3508, 1.5208], device='cuda:0'), covar=tensor([0.0775, 0.0314, 0.0312, 0.0829], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 15:56:53,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=465215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:57:15,647 INFO [train.py:968] (0/2) Epoch 11, batch 9700, giga_loss[loss=0.3206, simple_loss=0.3867, pruned_loss=0.1272, over 29026.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3875, pruned_loss=0.1316, over 5678514.41 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3549, pruned_loss=0.09959, over 5732464.73 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3902, pruned_loss=0.1348, over 5671392.72 frames. ], batch size: 155, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:57:16,309 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.688e+02 1.516e+03 2.214e+03 2.975e+03 8.408e+03, threshold=4.427e+03, percent-clipped=9.0
+2023-03-05 15:58:05,296 INFO [train.py:968] (0/2) Epoch 11, batch 9750, giga_loss[loss=0.3246, simple_loss=0.382, pruned_loss=0.1336, over 28617.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3864, pruned_loss=0.1321, over 5666606.41 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3548, pruned_loss=0.09961, over 5734827.55 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3891, pruned_loss=0.1351, over 5657975.38 frames. ], batch size: 78, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:58:15,469 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=465300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:58:48,027 INFO [train.py:968] (0/2) Epoch 11, batch 9800, giga_loss[loss=0.2951, simple_loss=0.3818, pruned_loss=0.1043, over 28476.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3842, pruned_loss=0.1301, over 5659996.34 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3548, pruned_loss=0.09958, over 5736128.22 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3873, pruned_loss=0.1336, over 5649340.02 frames. ], batch size: 60, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:58:48,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.402e+02 1.658e+03 2.275e+03 3.269e+03 6.759e+03, threshold=4.549e+03, percent-clipped=9.0
+2023-03-05 15:59:04,807 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=465358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:59:06,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=465361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:59:30,181 INFO [train.py:968] (0/2) Epoch 11, batch 9850, giga_loss[loss=0.3006, simple_loss=0.3771, pruned_loss=0.112, over 28675.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3832, pruned_loss=0.1278, over 5671140.30 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3549, pruned_loss=0.09976, over 5740567.34 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3862, pruned_loss=0.131, over 5656954.03 frames. ], batch size: 92, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:59:31,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=465390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:00:07,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4432, 1.5951, 1.2638, 1.7484], device='cuda:0'), covar=tensor([0.2531, 0.2413, 0.2757, 0.2250], device='cuda:0'), in_proj_covar=tensor([0.1289, 0.0956, 0.1142, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 16:00:14,580 INFO [train.py:968] (0/2) Epoch 11, batch 9900, giga_loss[loss=0.3021, simple_loss=0.3732, pruned_loss=0.1155, over 28891.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3833, pruned_loss=0.1266, over 5678456.56 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3547, pruned_loss=0.09965, over 5742719.89 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3863, pruned_loss=0.1297, over 5664342.26 frames. ], batch size: 199, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:00:15,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.286e+02 1.471e+03 1.969e+03 2.888e+03 6.533e+03, threshold=3.937e+03, percent-clipped=7.0
+2023-03-05 16:00:18,773 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=465443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:00:19,254 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=465444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:00:20,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=465446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:00:31,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7773, 2.6620, 1.6368, 0.9059], device='cuda:0'), covar=tensor([0.5607, 0.2670, 0.3362, 0.5108], device='cuda:0'), in_proj_covar=tensor([0.1534, 0.1458, 0.1478, 0.1254], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 16:00:51,276 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=465475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:01:02,982 INFO [train.py:968] (0/2) Epoch 11, batch 9950, giga_loss[loss=0.2987, simple_loss=0.369, pruned_loss=0.1142, over 28864.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3834, pruned_loss=0.1269, over 5672414.14 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3546, pruned_loss=0.09956, over 5744452.10 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3862, pruned_loss=0.1297, over 5658978.36 frames. ], batch size: 199, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:01:03,237 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=465489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:01:54,411 INFO [train.py:968] (0/2) Epoch 11, batch 10000, giga_loss[loss=0.3946, simple_loss=0.4282, pruned_loss=0.1805, over 27654.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3833, pruned_loss=0.1273, over 5669786.92 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3545, pruned_loss=0.09946, over 5745495.70 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3861, pruned_loss=0.13, over 5657178.81 frames. ], batch size: 472, lr: 2.98e-03, grad_scale: 8.0
+2023-03-05 16:01:55,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.405e+02 1.485e+03 1.826e+03 2.282e+03 5.028e+03, threshold=3.653e+03, percent-clipped=5.0
+2023-03-05 16:02:42,211 INFO [train.py:968] (0/2) Epoch 11, batch 10050, giga_loss[loss=0.3317, simple_loss=0.3686, pruned_loss=0.1474, over 23424.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3826, pruned_loss=0.1278, over 5672558.15 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3545, pruned_loss=0.09938, over 5748082.97 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3854, pruned_loss=0.1307, over 5658608.39 frames. ], batch size: 705, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:03:28,086 INFO [train.py:968] (0/2) Epoch 11, batch 10100, giga_loss[loss=0.3114, simple_loss=0.3725, pruned_loss=0.1252, over 28971.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3814, pruned_loss=0.1281, over 5673943.40 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3545, pruned_loss=0.09935, over 5750459.85 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3849, pruned_loss=0.1316, over 5657217.74 frames. ], batch size: 106, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:03:31,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.946e+02 1.659e+03 2.292e+03 3.315e+03 1.085e+04, threshold=4.584e+03, percent-clipped=21.0
+2023-03-05 16:03:43,148 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=465652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:04:21,573 INFO [train.py:968] (0/2) Epoch 11, batch 10150, giga_loss[loss=0.3099, simple_loss=0.375, pruned_loss=0.1224, over 28688.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3788, pruned_loss=0.1266, over 5678427.28 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3544, pruned_loss=0.09927, over 5751973.55 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3818, pruned_loss=0.1297, over 5663188.74 frames. ], batch size: 242, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:05:05,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3753, 1.3842, 3.9455, 3.3352], device='cuda:0'), covar=tensor([0.1976, 0.2791, 0.0850, 0.1033], device='cuda:0'), in_proj_covar=tensor([0.0656, 0.0590, 0.0857, 0.0757], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 16:05:11,461 INFO [train.py:968] (0/2) Epoch 11, batch 10200, giga_loss[loss=0.3348, simple_loss=0.3876, pruned_loss=0.141, over 28609.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3789, pruned_loss=0.128, over 5665787.10 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3547, pruned_loss=0.09942, over 5741796.45 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3815, pruned_loss=0.131, over 5661366.72 frames. ], batch size: 307, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:05:15,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.585e+02 1.524e+03 1.841e+03 2.825e+03 6.901e+03, threshold=3.681e+03, percent-clipped=7.0
+2023-03-05 16:05:48,226 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 16:05:52,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2414, 1.4056, 1.3582, 1.1385], device='cuda:0'), covar=tensor([0.1568, 0.1565, 0.0931, 0.1391], device='cuda:0'), in_proj_covar=tensor([0.1708, 0.1616, 0.1580, 0.1699], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 16:05:59,588 INFO [train.py:968] (0/2) Epoch 11, batch 10250, giga_loss[loss=0.348, simple_loss=0.3756, pruned_loss=0.1602, over 23765.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3781, pruned_loss=0.1278, over 5669044.51 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3544, pruned_loss=0.09927, over 5745259.24 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.381, pruned_loss=0.1308, over 5661029.99 frames. ], batch size: 705, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:06:27,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=465819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:06:42,950 INFO [train.py:968] (0/2) Epoch 11, batch 10300, giga_loss[loss=0.2741, simple_loss=0.3582, pruned_loss=0.09502, over 28798.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3747, pruned_loss=0.1242, over 5658930.39 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3542, pruned_loss=0.09905, over 5739179.49 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3778, pruned_loss=0.1275, over 5656176.24 frames. ], batch size: 174, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:06:46,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.956e+02 1.500e+03 1.859e+03 2.487e+03 6.610e+03, threshold=3.719e+03, percent-clipped=5.0
+2023-03-05 16:07:05,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=465864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:07:15,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4922, 1.7259, 1.7955, 1.3397], device='cuda:0'), covar=tensor([0.1592, 0.2102, 0.1299, 0.1496], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0704, 0.0864, 0.0772], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 16:07:15,974 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-05 16:07:30,547 INFO [train.py:968] (0/2) Epoch 11, batch 10350, giga_loss[loss=0.3687, simple_loss=0.4161, pruned_loss=0.1606, over 27869.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3724, pruned_loss=0.1219, over 5660089.04 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.354, pruned_loss=0.0989, over 5740628.59 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3756, pruned_loss=0.1254, over 5654212.66 frames. ], batch size: 412, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:07:35,511 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=465893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:08:20,588 INFO [train.py:968] (0/2) Epoch 11, batch 10400, giga_loss[loss=0.2815, simple_loss=0.3501, pruned_loss=0.1065, over 28741.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3721, pruned_loss=0.1215, over 5664633.52 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3542, pruned_loss=0.09896, over 5743054.47 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3747, pruned_loss=0.1245, over 5656441.08 frames. ], batch size: 243, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:08:23,340 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.833e+02 1.427e+03 1.953e+03 2.840e+03 8.370e+03, threshold=3.906e+03, percent-clipped=9.0
+2023-03-05 16:08:29,414 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 16:08:43,277 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=465962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:08:45,818 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=465965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:09:04,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.98 vs. limit=2.0
+2023-03-05 16:09:10,901 INFO [train.py:968] (0/2) Epoch 11, batch 10450, giga_loss[loss=0.2514, simple_loss=0.332, pruned_loss=0.08539, over 29055.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3691, pruned_loss=0.1204, over 5660555.27 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3544, pruned_loss=0.09907, over 5742572.13 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3713, pruned_loss=0.1231, over 5653121.04 frames. ], batch size: 136, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:09:16,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=465994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:09:21,323 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-466000.pt
+2023-03-05 16:09:29,431 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=466007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:09:33,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=466010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:09:50,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=466027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:10:01,695 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=466038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:10:02,108 INFO [train.py:968] (0/2) Epoch 11, batch 10500, giga_loss[loss=0.3295, simple_loss=0.3912, pruned_loss=0.1339, over 28962.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3687, pruned_loss=0.121, over 5667303.45 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3546, pruned_loss=0.09914, over 5744034.46 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3703, pruned_loss=0.1233, over 5659431.74 frames. ], batch size: 136, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:10:02,320 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=466039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:10:05,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.586e+02 1.756e+03 2.448e+03 3.307e+03 1.133e+04, threshold=4.896e+03, percent-clipped=18.0
+2023-03-05 16:10:43,120 INFO [train.py:968] (0/2) Epoch 11, batch 10550, giga_loss[loss=0.2812, simple_loss=0.3603, pruned_loss=0.101, over 28902.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3705, pruned_loss=0.1212, over 5676924.76 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3541, pruned_loss=0.09874, over 5748656.97 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3727, pruned_loss=0.124, over 5664599.09 frames. ], batch size: 145, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:10:44,205 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=466090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:10:51,195 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-05 16:11:24,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3902, 1.6850, 1.6210, 1.2931], device='cuda:0'), covar=tensor([0.2043, 0.1439, 0.1182, 0.1544], device='cuda:0'), in_proj_covar=tensor([0.1710, 0.1616, 0.1567, 0.1691], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 16:11:28,030 INFO [train.py:968] (0/2) Epoch 11, batch 10600, libri_loss[loss=0.2707, simple_loss=0.3479, pruned_loss=0.09672, over 29579.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3727, pruned_loss=0.1223, over 5672274.34 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3551, pruned_loss=0.09933, over 5753881.07 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3742, pruned_loss=0.1248, over 5655106.44 frames. ], batch size: 76, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:11:31,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.920e+02 1.457e+03 1.844e+03 2.440e+03 6.042e+03, threshold=3.688e+03, percent-clipped=1.0
+2023-03-05 16:11:59,486 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=466170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:12:01,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3754, 1.4913, 1.6135, 1.4358], device='cuda:0'), covar=tensor([0.1327, 0.1511, 0.1416, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0732, 0.0667, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 16:12:01,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=466173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:12:13,877 INFO [train.py:968] (0/2) Epoch 11, batch 10650, giga_loss[loss=0.284, simple_loss=0.3565, pruned_loss=0.1057, over 28624.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.373, pruned_loss=0.1227, over 5647120.42 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3555, pruned_loss=0.09948, over 5748032.74 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3742, pruned_loss=0.1251, over 5636933.41 frames. ], batch size: 78, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:12:26,654 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=466202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:12:47,610 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-05 16:12:58,328 INFO [train.py:968] (0/2) Epoch 11, batch 10700, giga_loss[loss=0.3317, simple_loss=0.3941, pruned_loss=0.1346, over 28655.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3747, pruned_loss=0.1244, over 5635008.80 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3563, pruned_loss=0.09989, over 5737743.19 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3755, pruned_loss=0.1267, over 5632336.65 frames. ], batch size: 262, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:13:00,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.877e+02 1.399e+03 1.795e+03 2.309e+03 8.000e+03, threshold=3.589e+03, percent-clipped=5.0
+2023-03-05 16:13:24,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=466268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:13:39,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1496, 1.4273, 1.3454, 1.2761], device='cuda:0'), covar=tensor([0.1565, 0.1517, 0.2035, 0.1574], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0734, 0.0671, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 16:13:47,436 INFO [train.py:968] (0/2) Epoch 11, batch 10750, giga_loss[loss=0.3009, simple_loss=0.3665, pruned_loss=0.1177, over 29004.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.376, pruned_loss=0.1259, over 5634838.55 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3564, pruned_loss=0.09996, over 5741088.60 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.377, pruned_loss=0.1281, over 5627977.84 frames. ], batch size: 106, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:14:06,318 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5989, 1.6458, 1.3674, 1.8003], device='cuda:0'), covar=tensor([0.2233, 0.2326, 0.2469, 0.2259], device='cuda:0'), in_proj_covar=tensor([0.1293, 0.0963, 0.1145, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 16:14:36,641 INFO [train.py:968] (0/2) Epoch 11, batch 10800, giga_loss[loss=0.3826, simple_loss=0.4202, pruned_loss=0.1725, over 27487.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3774, pruned_loss=0.1264, over 5647793.07 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3564, pruned_loss=0.0999, over 5743664.86 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3785, pruned_loss=0.1285, over 5638764.62 frames. ], batch size: 472, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:14:39,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.593e+02 1.818e+03 2.491e+03 3.213e+03 8.932e+03, threshold=4.983e+03, percent-clipped=17.0
+2023-03-05 16:15:22,810 INFO [train.py:968] (0/2) Epoch 11, batch 10850, giga_loss[loss=0.2936, simple_loss=0.3692, pruned_loss=0.109, over 28928.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3788, pruned_loss=0.1278, over 5648724.42 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3559, pruned_loss=0.0997, over 5746872.71 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3805, pruned_loss=0.1303, over 5636708.16 frames. ], batch size: 174, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:15:43,780 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=466411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:15:45,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=466413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:15:47,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=466414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:16:09,713 INFO [train.py:968] (0/2) Epoch 11, batch 10900, giga_loss[loss=0.342, simple_loss=0.399, pruned_loss=0.1424, over 28698.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3814, pruned_loss=0.13, over 5656780.63 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.356, pruned_loss=0.09974, over 5747763.52 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.383, pruned_loss=0.1325, over 5644886.90 frames. ], batch size: 242, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:16:14,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.459e+02 1.640e+03 2.192e+03 2.923e+03 8.086e+03, threshold=4.384e+03, percent-clipped=7.0
+2023-03-05 16:16:14,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=466443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:16:36,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=466465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:16:59,013 INFO [train.py:968] (0/2) Epoch 11, batch 10950, giga_loss[loss=0.2911, simple_loss=0.3702, pruned_loss=0.106, over 28938.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3819, pruned_loss=0.13, over 5659334.73 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3557, pruned_loss=0.09959, over 5750308.97 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3839, pruned_loss=0.1325, over 5646384.23 frames. ], batch size: 213, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:17:01,698 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=466491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:17:09,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9897, 3.8133, 3.5921, 1.9328], device='cuda:0'), covar=tensor([0.0638, 0.0758, 0.0838, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1062, 0.0991, 0.0873, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 16:17:37,114 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-05 16:17:52,748 INFO [train.py:968] (0/2) Epoch 11, batch 11000, giga_loss[loss=0.2721, simple_loss=0.346, pruned_loss=0.09908, over 28876.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3825, pruned_loss=0.1297, over 5651674.74 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.356, pruned_loss=0.09986, over 5742860.80 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3841, pruned_loss=0.1319, over 5646185.55 frames. ], batch size: 112, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:17:58,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.896e+02 1.543e+03 2.133e+03 2.639e+03 8.652e+03, threshold=4.267e+03, percent-clipped=5.0
+2023-03-05 16:18:11,579 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=466556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:18:12,960 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7285, 2.6459, 1.7173, 0.8111], device='cuda:0'), covar=tensor([0.5297, 0.2600, 0.2990, 0.5074], device='cuda:0'), in_proj_covar=tensor([0.1541, 0.1461, 0.1476, 0.1256], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 16:18:16,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=466559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:18:39,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8855, 3.6987, 3.5163, 1.4676], device='cuda:0'), covar=tensor([0.0653, 0.0775, 0.0798, 0.2210], device='cuda:0'), in_proj_covar=tensor([0.1061, 0.0991, 0.0871, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 16:18:42,436 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=466588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:18:43,541 INFO [train.py:968] (0/2) Epoch 11, batch 11050, giga_loss[loss=0.2935, simple_loss=0.3525, pruned_loss=0.1172, over 28759.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3804, pruned_loss=0.1289, over 5656696.15 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3557, pruned_loss=0.09983, over 5744719.88 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3825, pruned_loss=0.1312, over 5648938.03 frames. ], batch size: 99, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:19:02,940 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=466608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:19:05,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6366, 3.7530, 1.6874, 1.6677], device='cuda:0'), covar=tensor([0.0804, 0.0331, 0.0773, 0.1159], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0513, 0.0338, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 16:19:05,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=466611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:19:09,252 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 16:19:34,617 INFO [train.py:968] (0/2) Epoch 11, batch 11100, giga_loss[loss=0.3527, simple_loss=0.4109, pruned_loss=0.1473, over 28561.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3788, pruned_loss=0.1283, over 5659091.02 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3554, pruned_loss=0.09972, over 5748996.78 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3812, pruned_loss=0.1309, over 5647187.40 frames. ], batch size: 307, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:19:36,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=466640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:19:40,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.554e+02 1.661e+03 2.180e+03 3.234e+03 6.740e+03, threshold=4.360e+03, percent-clipped=7.0
+2023-03-05 16:19:53,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4678, 1.5864, 1.4615, 1.3170], device='cuda:0'), covar=tensor([0.1767, 0.1790, 0.1212, 0.1527], device='cuda:0'), in_proj_covar=tensor([0.1710, 0.1622, 0.1573, 0.1691], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 16:20:13,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=466672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:20:29,765 INFO [train.py:968] (0/2) Epoch 11, batch 11150, giga_loss[loss=0.3013, simple_loss=0.3705, pruned_loss=0.1161, over 28668.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3779, pruned_loss=0.1278, over 5654507.53 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3553, pruned_loss=0.09969, over 5741488.52 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3802, pruned_loss=0.1303, over 5651153.83 frames. ], batch size: 242, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:21:17,279 INFO [train.py:968] (0/2) Epoch 11, batch 11200, libri_loss[loss=0.2936, simple_loss=0.3691, pruned_loss=0.109, over 29291.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3775, pruned_loss=0.128, over 5663619.92 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3557, pruned_loss=0.09984, over 5743540.71 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3795, pruned_loss=0.1305, over 5657407.30 frames. ], batch size: 94, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:21:19,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.651e+03 2.005e+03 2.829e+03 6.072e+03, threshold=4.010e+03, percent-clipped=10.0
+2023-03-05 16:21:27,334 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=2.04 vs. limit=2.0
+2023-03-05 16:22:03,458 INFO [train.py:968] (0/2) Epoch 11, batch 11250, giga_loss[loss=0.4115, simple_loss=0.4342, pruned_loss=0.1944, over 27697.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3778, pruned_loss=0.1284, over 5658112.15 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3563, pruned_loss=0.1001, over 5741241.53 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3793, pruned_loss=0.1307, over 5653467.18 frames. ], batch size: 474, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:22:44,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3048, 1.6153, 1.3077, 1.5468], device='cuda:0'), covar=tensor([0.0721, 0.0319, 0.0303, 0.0756], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 16:22:53,064 INFO [train.py:968] (0/2) Epoch 11, batch 11300, giga_loss[loss=0.3163, simple_loss=0.373, pruned_loss=0.1298, over 28767.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3785, pruned_loss=0.129, over 5647410.33 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3567, pruned_loss=0.1002, over 5733647.32 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3796, pruned_loss=0.1311, over 5650111.29 frames. ], batch size: 99, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:22:58,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.499e+03 1.847e+03 2.568e+03 5.257e+03, threshold=3.694e+03, percent-clipped=4.0
+2023-03-05 16:23:16,794 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=466866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:23:38,965 INFO [train.py:968] (0/2) Epoch 11, batch 11350, giga_loss[loss=0.3183, simple_loss=0.3775, pruned_loss=0.1295, over 28769.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3792, pruned_loss=0.1291, over 5661429.34 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.357, pruned_loss=0.1004, over 5737374.51 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3802, pruned_loss=0.1313, over 5658595.11 frames. ], batch size: 243, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:23:59,552 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 16:24:04,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5823, 1.8068, 1.7985, 1.3819], device='cuda:0'), covar=tensor([0.1561, 0.2043, 0.1274, 0.1513], device='cuda:0'), in_proj_covar=tensor([0.0821, 0.0701, 0.0864, 0.0770], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 16:24:25,046 INFO [train.py:968] (0/2) Epoch 11, batch 11400, giga_loss[loss=0.3717, simple_loss=0.419, pruned_loss=0.1622, over 28557.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3808, pruned_loss=0.1302, over 5665003.87 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3571, pruned_loss=0.1004, over 5739751.22 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3819, pruned_loss=0.1323, over 5659807.61 frames. ], batch size: 336, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:24:29,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.830e+02 1.668e+03 2.084e+03 2.582e+03 9.003e+03, threshold=4.169e+03, percent-clipped=8.0
+2023-03-05 16:25:15,829 INFO [train.py:968] (0/2) Epoch 11, batch 11450, giga_loss[loss=0.3149, simple_loss=0.3689, pruned_loss=0.1305, over 28889.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3826, pruned_loss=0.1327, over 5654633.51 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.357, pruned_loss=0.1004, over 5741471.27 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3837, pruned_loss=0.1346, over 5648375.57 frames. ], batch size: 99, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:25:34,585 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=467009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:25:37,051 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=467012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:26:01,398 INFO [train.py:968] (0/2) Epoch 11, batch 11500, giga_loss[loss=0.3333, simple_loss=0.3844, pruned_loss=0.1411, over 27962.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.382, pruned_loss=0.1327, over 5653859.66 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3569, pruned_loss=0.1004, over 5736702.87 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3838, pruned_loss=0.1353, over 5650680.59 frames. ], batch size: 412, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:26:03,297 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=467041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:26:06,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.608e+03 2.243e+03 2.736e+03 4.878e+03, threshold=4.485e+03, percent-clipped=8.0
+2023-03-05 16:26:08,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=467047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:26:48,439 INFO [train.py:968] (0/2) Epoch 11, batch 11550, giga_loss[loss=0.3255, simple_loss=0.3842, pruned_loss=0.1334, over 28735.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3813, pruned_loss=0.132, over 5664508.64 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3568, pruned_loss=0.1003, over 5739564.30 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3834, pruned_loss=0.1348, over 5657588.95 frames. ], batch size: 284, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:27:11,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-05 16:27:28,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=467134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:27:34,626 INFO [train.py:968] (0/2) Epoch 11, batch 11600, giga_loss[loss=0.311, simple_loss=0.3709, pruned_loss=0.1256, over 28679.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.381, pruned_loss=0.131, over 5662253.36 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.357, pruned_loss=0.1005, over 5739390.19 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.383, pruned_loss=0.1338, over 5655062.75 frames. ], batch size: 92, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:27:39,159 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.164e+02 1.551e+03 2.088e+03 2.941e+03 1.085e+04, threshold=4.177e+03, percent-clipped=14.0
+2023-03-05 16:28:22,269 INFO [train.py:968] (0/2) Epoch 11, batch 11650, giga_loss[loss=0.3304, simple_loss=0.39, pruned_loss=0.1354, over 28820.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3819, pruned_loss=0.1314, over 5668045.23 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3573, pruned_loss=0.1005, over 5741874.27 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3836, pruned_loss=0.134, over 5659250.62 frames. ], batch size: 174, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:28:23,299 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=467190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:28:25,278 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=467193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:28:55,805 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=467222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:29:15,763 INFO [train.py:968] (0/2) Epoch 11, batch 11700, giga_loss[loss=0.3221, simple_loss=0.3821, pruned_loss=0.1311, over 29031.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3832, pruned_loss=0.1325, over 5674895.38 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3572, pruned_loss=0.1003, over 5742636.08 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3848, pruned_loss=0.1349, over 5666747.23 frames. ], batch size: 106, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:29:20,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4677, 1.7341, 1.8066, 1.3545], device='cuda:0'), covar=tensor([0.1567, 0.2170, 0.1248, 0.1512], device='cuda:0'), in_proj_covar=tensor([0.0820, 0.0701, 0.0861, 0.0769], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 16:29:22,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.593e+03 2.055e+03 2.885e+03 7.701e+03, threshold=4.110e+03, percent-clipped=8.0
+2023-03-05 16:30:02,801 INFO [train.py:968] (0/2) Epoch 11, batch 11750, giga_loss[loss=0.3414, simple_loss=0.3973, pruned_loss=0.1428, over 28923.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3824, pruned_loss=0.1322, over 5678949.00 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3567, pruned_loss=0.09993, over 5744331.89 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3847, pruned_loss=0.1353, over 5669498.25 frames. ], batch size: 213, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:30:44,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4728, 1.7611, 1.3660, 1.8586], device='cuda:0'), covar=tensor([0.2278, 0.2180, 0.2373, 0.2093], device='cuda:0'), in_proj_covar=tensor([0.1290, 0.0962, 0.1139, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 16:30:48,068 INFO [train.py:968] (0/2) Epoch 11, batch 11800, giga_loss[loss=0.3393, simple_loss=0.3976, pruned_loss=0.1405, over 28490.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3828, pruned_loss=0.1317, over 5686823.35 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3568, pruned_loss=0.1002, over 5748368.69 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3851, pruned_loss=0.1347, over 5674122.66 frames. ], batch size: 78, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:30:52,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.514e+02 1.524e+03 2.275e+03 3.145e+03 9.412e+03, threshold=4.550e+03, percent-clipped=14.0
+2023-03-05 16:31:18,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4264, 1.9918, 1.4117, 0.6772], device='cuda:0'), covar=tensor([0.3221, 0.1933, 0.2729, 0.4049], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1479, 0.1486, 0.1264], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 16:31:32,644 INFO [train.py:968] (0/2) Epoch 11, batch 11850, giga_loss[loss=0.2922, simple_loss=0.3617, pruned_loss=0.1113, over 28565.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3808, pruned_loss=0.1292, over 5684565.15 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.356, pruned_loss=0.0996, over 5752583.32 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3843, pruned_loss=0.1333, over 5668140.17 frames. ], batch size: 71, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:31:54,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8207, 1.2091, 1.1351, 1.0099], device='cuda:0'), covar=tensor([0.1439, 0.1130, 0.1800, 0.1369], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0735, 0.0673, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 16:32:21,511 INFO [train.py:968] (0/2) Epoch 11, batch 11900, giga_loss[loss=0.288, simple_loss=0.3508, pruned_loss=0.1126, over 28635.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3796, pruned_loss=0.1282, over 5676236.31 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3561, pruned_loss=0.0997, over 5754021.37 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3825, pruned_loss=0.1316, over 5661366.24 frames. ], batch size: 92, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:32:27,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.392e+03 1.954e+03 2.556e+03 6.643e+03, threshold=3.908e+03, percent-clipped=6.0
+2023-03-05 16:32:37,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9249, 2.5323, 1.9647, 1.5913], device='cuda:0'), covar=tensor([0.2204, 0.1417, 0.1700, 0.1972], device='cuda:0'), in_proj_covar=tensor([0.1700, 0.1619, 0.1570, 0.1684], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 16:32:46,059 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-05 16:33:05,947 INFO [train.py:968] (0/2) Epoch 11, batch 11950, giga_loss[loss=0.3788, simple_loss=0.4075, pruned_loss=0.1751, over 26695.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3791, pruned_loss=0.1277, over 5687913.79 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3571, pruned_loss=0.1004, over 5753549.35 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3813, pruned_loss=0.1307, over 5674326.52 frames. ], batch size: 555, lr: 2.97e-03, grad_scale: 2.0
+2023-03-05 16:33:23,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=467509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:33:50,995 INFO [train.py:968] (0/2) Epoch 11, batch 12000, giga_loss[loss=0.2856, simple_loss=0.3538, pruned_loss=0.1087, over 28247.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.379, pruned_loss=0.1278, over 5681527.67 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3568, pruned_loss=0.1002, over 5758786.95 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3819, pruned_loss=0.1313, over 5663376.28 frames. ], batch size: 77, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:33:50,999 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 16:33:59,455 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.218, simple_loss=0.3234, pruned_loss=0.05624, over 944034.00 frames. 
+2023-03-05 16:33:59,456 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 16:34:06,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.406e+03 1.989e+03 3.057e+03 9.686e+03, threshold=3.979e+03, percent-clipped=16.0
+2023-03-05 16:34:08,784 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 16:34:13,101 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-05 16:34:44,945 INFO [train.py:968] (0/2) Epoch 11, batch 12050, libri_loss[loss=0.2607, simple_loss=0.3407, pruned_loss=0.09032, over 29534.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3807, pruned_loss=0.1288, over 5681399.90 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3562, pruned_loss=0.09983, over 5762007.39 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3844, pruned_loss=0.133, over 5661193.39 frames. ], batch size: 80, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:35:09,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=467612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:35:32,065 INFO [train.py:968] (0/2) Epoch 11, batch 12100, libri_loss[loss=0.2515, simple_loss=0.3223, pruned_loss=0.09038, over 29362.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3811, pruned_loss=0.1297, over 5677264.94 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3562, pruned_loss=0.09986, over 5755726.98 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3846, pruned_loss=0.1337, over 5665035.64 frames. ], batch size: 67, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:35:37,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.758e+02 1.353e+03 1.994e+03 3.017e+03 9.094e+03, threshold=3.989e+03, percent-clipped=9.0
+2023-03-05 16:35:48,179 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=467652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:35:49,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=467655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:36:19,358 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=467684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:36:22,225 INFO [train.py:968] (0/2) Epoch 11, batch 12150, giga_loss[loss=0.3227, simple_loss=0.3869, pruned_loss=0.1292, over 28948.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3813, pruned_loss=0.1312, over 5665849.57 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3557, pruned_loss=0.09963, over 5757920.83 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3849, pruned_loss=0.135, over 5652896.23 frames. ], batch size: 174, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:37:08,493 INFO [train.py:968] (0/2) Epoch 11, batch 12200, giga_loss[loss=0.3863, simple_loss=0.4292, pruned_loss=0.1717, over 28289.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3822, pruned_loss=0.1318, over 5671826.88 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3557, pruned_loss=0.09971, over 5757723.85 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3857, pruned_loss=0.1355, over 5659524.24 frames. ], batch size: 368, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:37:16,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.899e+02 1.521e+03 1.936e+03 2.627e+03 6.934e+03, threshold=3.873e+03, percent-clipped=4.0
+2023-03-05 16:37:19,507 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-05 16:37:43,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-05 16:37:45,254 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2156, 0.8318, 0.8553, 1.4453], device='cuda:0'), covar=tensor([0.0714, 0.0372, 0.0336, 0.0766], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 16:37:55,253 INFO [train.py:968] (0/2) Epoch 11, batch 12250, libri_loss[loss=0.2903, simple_loss=0.3745, pruned_loss=0.103, over 29558.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3838, pruned_loss=0.1329, over 5655890.80 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3558, pruned_loss=0.0997, over 5750036.98 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3874, pruned_loss=0.137, over 5650159.82 frames. ], batch size: 83, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:38:26,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-05 16:38:39,400 INFO [train.py:968] (0/2) Epoch 11, batch 12300, giga_loss[loss=0.3236, simple_loss=0.3872, pruned_loss=0.13, over 28748.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3828, pruned_loss=0.1321, over 5658736.07 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.356, pruned_loss=0.0998, over 5753496.64 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.386, pruned_loss=0.1358, over 5649501.40 frames. ], batch size: 284, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:38:46,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.308e+02 1.632e+03 2.118e+03 2.837e+03 7.667e+03, threshold=4.235e+03, percent-clipped=10.0
+2023-03-05 16:38:56,974 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=467856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:38:58,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2434, 1.4357, 3.1669, 2.9278], device='cuda:0'), covar=tensor([0.1308, 0.2191, 0.0468, 0.0914], device='cuda:0'), in_proj_covar=tensor([0.0663, 0.0595, 0.0863, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 16:39:16,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2497, 4.0704, 3.8422, 1.7873], device='cuda:0'), covar=tensor([0.0568, 0.0670, 0.0729, 0.2200], device='cuda:0'), in_proj_covar=tensor([0.1076, 0.1004, 0.0884, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 16:39:31,590 INFO [train.py:968] (0/2) Epoch 11, batch 12350, giga_loss[loss=0.2901, simple_loss=0.3664, pruned_loss=0.1069, over 29082.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3828, pruned_loss=0.1322, over 5646259.54 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3558, pruned_loss=0.09977, over 5755081.71 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3856, pruned_loss=0.1355, over 5636989.85 frames. ], batch size: 128, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:39:44,939 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 16:40:03,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3065, 1.4583, 1.2348, 1.2271], device='cuda:0'), covar=tensor([0.1587, 0.1425, 0.1268, 0.1423], device='cuda:0'), in_proj_covar=tensor([0.1695, 0.1613, 0.1570, 0.1676], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 16:40:17,467 INFO [train.py:968] (0/2) Epoch 11, batch 12400, giga_loss[loss=0.3097, simple_loss=0.3767, pruned_loss=0.1213, over 28835.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3828, pruned_loss=0.1317, over 5650219.56 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3561, pruned_loss=0.0999, over 5758703.99 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3854, pruned_loss=0.1349, over 5637457.57 frames. ], batch size: 186, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:40:24,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.535e+02 1.538e+03 1.891e+03 2.534e+03 4.995e+03, threshold=3.782e+03, percent-clipped=4.0
+2023-03-05 16:41:00,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=467987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:41:02,044 INFO [train.py:968] (0/2) Epoch 11, batch 12450, giga_loss[loss=0.3359, simple_loss=0.389, pruned_loss=0.1414, over 28252.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3819, pruned_loss=0.1308, over 5651143.97 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3562, pruned_loss=0.1001, over 5752096.50 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3844, pruned_loss=0.1337, over 5644720.72 frames. ], batch size: 368, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:41:15,189 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-468000.pt
+2023-03-05 16:41:48,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5134, 3.4144, 1.6274, 1.4968], device='cuda:0'), covar=tensor([0.0868, 0.0347, 0.0811, 0.1276], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0514, 0.0338, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 16:41:54,310 INFO [train.py:968] (0/2) Epoch 11, batch 12500, giga_loss[loss=0.3353, simple_loss=0.3943, pruned_loss=0.1382, over 28916.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3818, pruned_loss=0.1314, over 5657154.81 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3562, pruned_loss=0.1001, over 5752841.02 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3838, pruned_loss=0.1339, over 5651039.66 frames. ], batch size: 112, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:42:02,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.597e+03 1.983e+03 2.799e+03 8.005e+03, threshold=3.966e+03, percent-clipped=11.0
+2023-03-05 16:42:29,109 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5938, 1.8017, 1.3868, 2.0020], device='cuda:0'), covar=tensor([0.2381, 0.2349, 0.2638, 0.2099], device='cuda:0'), in_proj_covar=tensor([0.1288, 0.0961, 0.1137, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 16:42:43,459 INFO [train.py:968] (0/2) Epoch 11, batch 12550, giga_loss[loss=0.3371, simple_loss=0.3838, pruned_loss=0.1452, over 28932.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3781, pruned_loss=0.129, over 5666572.54 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3563, pruned_loss=0.1001, over 5754408.67 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3798, pruned_loss=0.1312, over 5659519.25 frames. ], batch size: 213, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:43:22,736 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=468130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:43:24,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=468133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:43:30,996 INFO [train.py:968] (0/2) Epoch 11, batch 12600, giga_loss[loss=0.2916, simple_loss=0.3519, pruned_loss=0.1157, over 28910.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3748, pruned_loss=0.1277, over 5661311.89 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3562, pruned_loss=0.1001, over 5756377.10 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3765, pruned_loss=0.1299, over 5652571.60 frames. ], batch size: 106, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:43:40,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.638e+02 1.747e+03 2.420e+03 3.376e+03 8.476e+03, threshold=4.841e+03, percent-clipped=17.0
+2023-03-05 16:43:54,330 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=468162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:44:04,431 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 16:44:17,152 INFO [train.py:968] (0/2) Epoch 11, batch 12650, giga_loss[loss=0.31, simple_loss=0.3686, pruned_loss=0.1257, over 28853.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3738, pruned_loss=0.1278, over 5652633.35 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3571, pruned_loss=0.1006, over 5752405.08 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3748, pruned_loss=0.1297, over 5646760.95 frames. ], batch size: 112, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:44:34,005 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=468203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:44:41,568 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=468213.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 16:44:59,551 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=468231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:45:06,897 INFO [train.py:968] (0/2) Epoch 11, batch 12700, giga_loss[loss=0.3647, simple_loss=0.4021, pruned_loss=0.1636, over 26573.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3716, pruned_loss=0.1267, over 5641117.36 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.357, pruned_loss=0.1005, over 5750698.60 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3727, pruned_loss=0.1287, over 5636805.00 frames. ], batch size: 555, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:45:13,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.385e+02 1.524e+03 2.039e+03 2.846e+03 5.098e+03, threshold=4.078e+03, percent-clipped=4.0
+2023-03-05 16:45:51,711 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8609, 1.9862, 1.4459, 1.5890], device='cuda:0'), covar=tensor([0.0816, 0.0616, 0.0919, 0.1137], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0443, 0.0498, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 16:45:53,638 INFO [train.py:968] (0/2) Epoch 11, batch 12750, giga_loss[loss=0.3112, simple_loss=0.3766, pruned_loss=0.1229, over 28627.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3703, pruned_loss=0.1251, over 5643353.20 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3571, pruned_loss=0.1007, over 5743969.95 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3712, pruned_loss=0.1268, over 5644244.49 frames. ], batch size: 307, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:46:41,356 INFO [train.py:968] (0/2) Epoch 11, batch 12800, giga_loss[loss=0.2817, simple_loss=0.3608, pruned_loss=0.1013, over 28671.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3684, pruned_loss=0.1216, over 5634176.03 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3571, pruned_loss=0.1009, over 5731740.89 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3696, pruned_loss=0.1236, over 5642877.08 frames. ], batch size: 307, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:46:47,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.291e+02 1.452e+03 2.111e+03 2.800e+03 4.632e+03, threshold=4.222e+03, percent-clipped=2.0
+2023-03-05 16:47:17,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=468374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:47:20,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=468377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:47:32,526 INFO [train.py:968] (0/2) Epoch 11, batch 12850, libri_loss[loss=0.3086, simple_loss=0.3759, pruned_loss=0.1207, over 29541.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.367, pruned_loss=0.1191, over 5635051.96 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3567, pruned_loss=0.1008, over 5732505.96 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3687, pruned_loss=0.1214, over 5637900.92 frames. ], batch size: 83, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:47:46,496 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=468406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:48:17,633 INFO [train.py:968] (0/2) Epoch 11, batch 12900, giga_loss[loss=0.3412, simple_loss=0.3859, pruned_loss=0.1482, over 26747.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3635, pruned_loss=0.1155, over 5639453.59 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3556, pruned_loss=0.1005, over 5735503.30 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3663, pruned_loss=0.1184, over 5634597.15 frames. ], batch size: 555, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:48:24,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.389e+02 1.337e+03 1.669e+03 2.720e+03 6.239e+03, threshold=3.339e+03, percent-clipped=6.0
+2023-03-05 16:48:47,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=468468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:49:10,677 INFO [train.py:968] (0/2) Epoch 11, batch 12950, giga_loss[loss=0.2717, simple_loss=0.3574, pruned_loss=0.09298, over 28866.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3591, pruned_loss=0.1115, over 5639842.27 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3555, pruned_loss=0.1005, over 5737381.12 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3615, pruned_loss=0.114, over 5633134.55 frames. ], batch size: 285, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:49:36,895 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2332, 1.5016, 1.4463, 1.1425], device='cuda:0'), covar=tensor([0.1625, 0.1547, 0.0902, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.1664, 0.1583, 0.1541, 0.1641], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 16:49:40,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8945, 2.0836, 1.8417, 1.6605], device='cuda:0'), covar=tensor([0.1514, 0.1965, 0.1731, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0719, 0.0659, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 16:49:56,620 INFO [train.py:968] (0/2) Epoch 11, batch 13000, giga_loss[loss=0.2718, simple_loss=0.3552, pruned_loss=0.09417, over 28886.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3584, pruned_loss=0.1089, over 5650891.07 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3552, pruned_loss=0.1006, over 5741985.35 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3606, pruned_loss=0.111, over 5638888.44 frames. ], batch size: 199, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:50:04,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.124e+02 1.328e+03 1.647e+03 2.352e+03 5.353e+03, threshold=3.294e+03, percent-clipped=6.0
+2023-03-05 16:50:16,603 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 16:50:34,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=468578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:50:45,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=468588.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 16:50:46,029 INFO [train.py:968] (0/2) Epoch 11, batch 13050, giga_loss[loss=0.2996, simple_loss=0.3464, pruned_loss=0.1264, over 24263.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3584, pruned_loss=0.1077, over 5659468.52 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3547, pruned_loss=0.1005, over 5746199.46 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3608, pruned_loss=0.1097, over 5643945.86 frames. ], batch size: 705, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:51:39,170 INFO [train.py:968] (0/2) Epoch 11, batch 13100, giga_loss[loss=0.2751, simple_loss=0.3495, pruned_loss=0.1003, over 28905.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.358, pruned_loss=0.1075, over 5652425.26 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3542, pruned_loss=0.1003, over 5746794.93 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3604, pruned_loss=0.1094, over 5638592.01 frames. ], batch size: 213, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:51:47,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.890e+02 1.255e+03 1.690e+03 2.340e+03 4.988e+03, threshold=3.381e+03, percent-clipped=8.0
+2023-03-05 16:51:53,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1812, 1.8377, 1.4516, 0.3686], device='cuda:0'), covar=tensor([0.2922, 0.1862, 0.2785, 0.3950], device='cuda:0'), in_proj_covar=tensor([0.1551, 0.1470, 0.1478, 0.1261], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 16:52:27,359 INFO [train.py:968] (0/2) Epoch 11, batch 13150, giga_loss[loss=0.25, simple_loss=0.3271, pruned_loss=0.08646, over 27930.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3557, pruned_loss=0.1058, over 5655158.63 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3545, pruned_loss=0.1005, over 5749611.86 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3574, pruned_loss=0.1072, over 5640140.52 frames. ], batch size: 412, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:52:58,381 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=468721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:53:01,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=468724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:53:09,938 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=468731.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 16:53:11,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=468734.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 16:53:14,967 INFO [train.py:968] (0/2) Epoch 11, batch 13200, giga_loss[loss=0.2592, simple_loss=0.3352, pruned_loss=0.09166, over 28284.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3529, pruned_loss=0.1042, over 5652258.70 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3539, pruned_loss=0.1003, over 5750351.94 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3548, pruned_loss=0.1057, over 5635381.86 frames. ], batch size: 368, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:53:24,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.171e+02 1.424e+03 1.987e+03 2.638e+03 6.379e+03, threshold=3.974e+03, percent-clipped=15.0
+2023-03-05 16:53:30,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=468753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:53:41,772 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=468763.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 16:54:07,426 INFO [train.py:968] (0/2) Epoch 11, batch 13250, libri_loss[loss=0.3154, simple_loss=0.392, pruned_loss=0.1194, over 25729.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3532, pruned_loss=0.1043, over 5643218.92 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3539, pruned_loss=0.1004, over 5748899.99 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3546, pruned_loss=0.1054, over 5629565.21 frames. ], batch size: 136, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:54:42,708 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-05 16:54:56,052 INFO [train.py:968] (0/2) Epoch 11, batch 13300, giga_loss[loss=0.2761, simple_loss=0.3497, pruned_loss=0.1013, over 28587.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3521, pruned_loss=0.1031, over 5652757.18 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3538, pruned_loss=0.1002, over 5751788.15 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3533, pruned_loss=0.1043, over 5637517.87 frames. ], batch size: 336, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:55:00,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=468843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:55:03,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.370e+02 1.390e+03 1.985e+03 2.713e+03 5.811e+03, threshold=3.970e+03, percent-clipped=4.0
+2023-03-05 16:55:04,046 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-05 16:55:15,120 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-05 16:55:48,830 INFO [train.py:968] (0/2) Epoch 11, batch 13350, giga_loss[loss=0.2354, simple_loss=0.3217, pruned_loss=0.07453, over 28673.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3488, pruned_loss=0.1005, over 5652266.02 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3535, pruned_loss=0.1001, over 5753162.44 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.35, pruned_loss=0.1015, over 5638149.84 frames. ], batch size: 242, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:56:08,811 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 16:56:40,026 INFO [train.py:968] (0/2) Epoch 11, batch 13400, giga_loss[loss=0.2583, simple_loss=0.3322, pruned_loss=0.09223, over 28648.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3445, pruned_loss=0.09765, over 5650748.39 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3531, pruned_loss=0.0999, over 5752582.02 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3458, pruned_loss=0.09861, over 5638364.11 frames. ], batch size: 262, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:56:52,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.346e+02 1.190e+03 1.608e+03 2.096e+03 6.194e+03, threshold=3.216e+03, percent-clipped=4.0
+2023-03-05 16:57:32,222 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=468986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:57:34,451 INFO [train.py:968] (0/2) Epoch 11, batch 13450, giga_loss[loss=0.252, simple_loss=0.3373, pruned_loss=0.08334, over 28979.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3422, pruned_loss=0.09683, over 5660572.37 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3528, pruned_loss=0.09975, over 5754867.73 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3433, pruned_loss=0.09768, over 5646759.93 frames. ], batch size: 164, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:57:34,740 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=468989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:58:01,142 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=469018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:58:23,713 INFO [train.py:968] (0/2) Epoch 11, batch 13500, giga_loss[loss=0.2737, simple_loss=0.3212, pruned_loss=0.1132, over 23884.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3417, pruned_loss=0.09738, over 5647954.78 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3526, pruned_loss=0.09981, over 5749632.07 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3424, pruned_loss=0.09795, over 5638810.08 frames. ], batch size: 705, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:58:34,352 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:58:34,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.518e+02 1.355e+03 1.934e+03 2.916e+03 9.283e+03, threshold=3.867e+03, percent-clipped=16.0
+2023-03-05 16:58:38,549 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-05 16:59:23,606 INFO [train.py:968] (0/2) Epoch 11, batch 13550, giga_loss[loss=0.2444, simple_loss=0.3368, pruned_loss=0.07606, over 28896.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3425, pruned_loss=0.09758, over 5640689.41 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3525, pruned_loss=0.09983, over 5740937.58 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.343, pruned_loss=0.09797, over 5638973.29 frames. ], batch size: 174, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:59:25,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7689, 1.9462, 1.2970, 1.5533], device='cuda:0'), covar=tensor([0.0771, 0.0499, 0.0961, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0438, 0.0496, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 17:00:20,377 INFO [train.py:968] (0/2) Epoch 11, batch 13600, giga_loss[loss=0.2541, simple_loss=0.3392, pruned_loss=0.08446, over 28636.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3452, pruned_loss=0.09805, over 5640606.38 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3523, pruned_loss=0.09981, over 5743418.17 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3456, pruned_loss=0.09836, over 5635583.00 frames. ], batch size: 242, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 17:00:29,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.942e+02 1.363e+03 1.672e+03 2.602e+03 4.250e+03, threshold=3.345e+03, percent-clipped=2.0
+2023-03-05 17:01:19,330 INFO [train.py:968] (0/2) Epoch 11, batch 13650, giga_loss[loss=0.2674, simple_loss=0.3425, pruned_loss=0.09614, over 28892.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3459, pruned_loss=0.09771, over 5649351.22 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3523, pruned_loss=0.09979, over 5739914.47 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3462, pruned_loss=0.09795, over 5647485.97 frames. ], batch size: 199, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 17:01:42,371 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 17:02:23,285 INFO [train.py:968] (0/2) Epoch 11, batch 13700, giga_loss[loss=0.2795, simple_loss=0.3557, pruned_loss=0.1017, over 28709.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3457, pruned_loss=0.09783, over 5648639.39 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3521, pruned_loss=0.09977, over 5733044.46 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3461, pruned_loss=0.09802, over 5652016.22 frames. ], batch size: 262, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:02:34,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.282e+02 1.414e+03 2.018e+03 2.566e+03 6.816e+03, threshold=4.036e+03, percent-clipped=13.0
+2023-03-05 17:03:20,820 INFO [train.py:968] (0/2) Epoch 11, batch 13750, giga_loss[loss=0.2908, simple_loss=0.3578, pruned_loss=0.1119, over 26782.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3435, pruned_loss=0.09602, over 5660255.69 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3516, pruned_loss=0.09952, over 5736691.27 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3441, pruned_loss=0.09635, over 5657864.07 frames. ], batch size: 555, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:04:18,969 INFO [train.py:968] (0/2) Epoch 11, batch 13800, giga_loss[loss=0.2488, simple_loss=0.3296, pruned_loss=0.08397, over 28992.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3411, pruned_loss=0.09295, over 5666325.03 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3508, pruned_loss=0.09914, over 5740004.94 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3421, pruned_loss=0.09345, over 5659878.96 frames. ], batch size: 285, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:04:30,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.171e+02 1.192e+03 1.492e+03 2.106e+03 4.838e+03, threshold=2.985e+03, percent-clipped=4.0
+2023-03-05 17:05:19,850 INFO [train.py:968] (0/2) Epoch 11, batch 13850, giga_loss[loss=0.3044, simple_loss=0.3551, pruned_loss=0.1268, over 28993.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3386, pruned_loss=0.09267, over 5662024.71 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3501, pruned_loss=0.0988, over 5741536.19 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3397, pruned_loss=0.09317, over 5653164.29 frames. ], batch size: 186, lr: 2.97e-03, grad_scale: 2.0
+2023-03-05 17:05:26,022 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:05:33,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4100, 2.0403, 1.4517, 0.6766], device='cuda:0'), covar=tensor([0.3517, 0.1796, 0.2788, 0.4078], device='cuda:0'), in_proj_covar=tensor([0.1540, 0.1457, 0.1470, 0.1261], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 17:05:59,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=469423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:06:03,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9638, 2.0493, 1.4101, 1.6319], device='cuda:0'), covar=tensor([0.0792, 0.0568, 0.0958, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0436, 0.0493, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 17:06:07,996 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469431.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:06:18,902 INFO [train.py:968] (0/2) Epoch 11, batch 13900, giga_loss[loss=0.2684, simple_loss=0.3448, pruned_loss=0.09606, over 29018.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3376, pruned_loss=0.09279, over 5669609.80 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3498, pruned_loss=0.09873, over 5742185.91 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3384, pruned_loss=0.09311, over 5660513.79 frames. ], batch size: 128, lr: 2.97e-03, grad_scale: 2.0
+2023-03-05 17:06:31,902 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.450e+02 1.305e+03 1.761e+03 2.661e+03 9.463e+03, threshold=3.522e+03, percent-clipped=17.0
+2023-03-05 17:07:16,853 INFO [train.py:968] (0/2) Epoch 11, batch 13950, giga_loss[loss=0.2245, simple_loss=0.2933, pruned_loss=0.07783, over 24422.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.337, pruned_loss=0.09308, over 5654524.91 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3496, pruned_loss=0.09869, over 5734581.59 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3377, pruned_loss=0.09326, over 5652326.18 frames. ], batch size: 705, lr: 2.97e-03, grad_scale: 2.0
+2023-03-05 17:07:26,364 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9651, 1.2341, 0.9540, 0.2490], device='cuda:0'), covar=tensor([0.2182, 0.1857, 0.2943, 0.3850], device='cuda:0'), in_proj_covar=tensor([0.1532, 0.1452, 0.1467, 0.1257], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 17:08:05,832 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:08:12,098 INFO [train.py:968] (0/2) Epoch 11, batch 14000, giga_loss[loss=0.2598, simple_loss=0.3344, pruned_loss=0.09256, over 27461.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3387, pruned_loss=0.09353, over 5656657.42 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3493, pruned_loss=0.09865, over 5737546.78 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3392, pruned_loss=0.09356, over 5649864.55 frames. ], batch size: 472, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:08:29,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.227e+02 1.440e+03 1.904e+03 2.546e+03 8.615e+03, threshold=3.808e+03, percent-clipped=9.0
+2023-03-05 17:08:46,616 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=469566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:08:50,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=469569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:09:15,275 INFO [train.py:968] (0/2) Epoch 11, batch 14050, giga_loss[loss=0.233, simple_loss=0.302, pruned_loss=0.08203, over 24473.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.34, pruned_loss=0.09337, over 5656623.51 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3493, pruned_loss=0.09865, over 5739347.98 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3402, pruned_loss=0.09333, over 5648600.11 frames. ], batch size: 705, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:09:19,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8499, 1.9115, 1.3400, 1.4692], device='cuda:0'), covar=tensor([0.0709, 0.0485, 0.0900, 0.0972], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0436, 0.0494, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 17:09:27,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=469598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:10:08,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:10:21,428 INFO [train.py:968] (0/2) Epoch 11, batch 14100, giga_loss[loss=0.2852, simple_loss=0.3555, pruned_loss=0.1074, over 27744.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3362, pruned_loss=0.09088, over 5671707.22 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3489, pruned_loss=0.09849, over 5742834.16 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3366, pruned_loss=0.09085, over 5660762.75 frames. ], batch size: 474, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:10:41,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.933e+02 1.236e+03 1.645e+03 2.117e+03 4.295e+03, threshold=3.290e+03, percent-clipped=2.0
+2023-03-05 17:10:42,780 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-05 17:11:27,644 INFO [train.py:968] (0/2) Epoch 11, batch 14150, giga_loss[loss=0.2789, simple_loss=0.3606, pruned_loss=0.09856, over 28965.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3366, pruned_loss=0.09117, over 5682416.76 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3483, pruned_loss=0.09819, over 5745055.23 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3373, pruned_loss=0.09128, over 5670520.24 frames. ], batch size: 155, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:11:46,472 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:12:06,276 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-05 17:12:20,682 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-05 17:12:27,888 INFO [train.py:968] (0/2) Epoch 11, batch 14200, giga_loss[loss=0.2697, simple_loss=0.3652, pruned_loss=0.08714, over 28708.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3413, pruned_loss=0.09247, over 5681673.57 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3484, pruned_loss=0.09838, over 5749196.78 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3414, pruned_loss=0.09217, over 5666083.67 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:12:41,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.683e+02 1.419e+03 1.773e+03 2.623e+03 6.612e+03, threshold=3.545e+03, percent-clipped=16.0
+2023-03-05 17:12:54,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4297, 1.8128, 1.5367, 1.3769], device='cuda:0'), covar=tensor([0.1853, 0.1493, 0.1166, 0.1327], device='cuda:0'), in_proj_covar=tensor([0.1654, 0.1554, 0.1509, 0.1606], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 17:13:03,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=469768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:13:28,203 INFO [train.py:968] (0/2) Epoch 11, batch 14250, giga_loss[loss=0.2619, simple_loss=0.3494, pruned_loss=0.08723, over 28635.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3445, pruned_loss=0.0925, over 5681114.92 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.348, pruned_loss=0.09831, over 5750722.59 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3448, pruned_loss=0.09218, over 5665406.61 frames. ], batch size: 242, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:13:45,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3188, 1.6795, 1.6209, 1.2063], device='cuda:0'), covar=tensor([0.1778, 0.2404, 0.1427, 0.1721], device='cuda:0'), in_proj_covar=tensor([0.0825, 0.0689, 0.0863, 0.0770], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 17:13:49,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=469806.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:14:30,735 INFO [train.py:968] (0/2) Epoch 11, batch 14300, giga_loss[loss=0.2631, simple_loss=0.3554, pruned_loss=0.0854, over 28758.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3459, pruned_loss=0.09207, over 5678979.25 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3479, pruned_loss=0.09824, over 5752307.35 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3462, pruned_loss=0.09182, over 5664603.52 frames. ], batch size: 119, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:14:42,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.997e+02 1.408e+03 1.855e+03 2.470e+03 6.436e+03, threshold=3.711e+03, percent-clipped=7.0
+2023-03-05 17:14:47,441 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:14:55,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1642, 1.4695, 1.3711, 1.3494], device='cuda:0'), covar=tensor([0.1167, 0.1190, 0.1701, 0.1194], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0713, 0.0656, 0.0640], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 17:15:28,666 INFO [train.py:968] (0/2) Epoch 11, batch 14350, giga_loss[loss=0.2664, simple_loss=0.3491, pruned_loss=0.09185, over 28994.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3444, pruned_loss=0.0911, over 5673944.78 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3476, pruned_loss=0.09824, over 5746767.10 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3448, pruned_loss=0.09075, over 5665492.73 frames. ], batch size: 199, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:15:35,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4509, 1.7174, 1.5521, 1.4751], device='cuda:0'), covar=tensor([0.1373, 0.1826, 0.1801, 0.1693], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0716, 0.0658, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 17:15:53,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=469907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:15:59,495 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=469911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:16:02,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=469914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:16:34,324 INFO [train.py:968] (0/2) Epoch 11, batch 14400, giga_loss[loss=0.247, simple_loss=0.3272, pruned_loss=0.08337, over 28985.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3454, pruned_loss=0.09263, over 5670716.35 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3476, pruned_loss=0.09824, over 5745690.26 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3457, pruned_loss=0.09233, over 5664744.60 frames. ], batch size: 145, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 17:16:39,201 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=469943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:16:45,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=469949.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:16:50,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.864e+02 1.380e+03 2.065e+03 3.015e+03 6.559e+03, threshold=4.130e+03, percent-clipped=10.0
+2023-03-05 17:16:51,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=469952.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:17:06,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8645, 1.9249, 1.3034, 1.5044], device='cuda:0'), covar=tensor([0.0741, 0.0565, 0.0969, 0.0983], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0434, 0.0493, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 17:17:26,537 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=469981.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:17:35,776 INFO [train.py:968] (0/2) Epoch 11, batch 14450, giga_loss[loss=0.2538, simple_loss=0.3311, pruned_loss=0.08824, over 28938.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3455, pruned_loss=0.0937, over 5686798.86 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3473, pruned_loss=0.09813, over 5749351.40 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.346, pruned_loss=0.09343, over 5676852.59 frames. ], batch size: 106, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:17:52,872 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-470000.pt
+2023-03-05 17:17:55,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:18:56,058 INFO [train.py:968] (0/2) Epoch 11, batch 14500, giga_loss[loss=0.2381, simple_loss=0.3194, pruned_loss=0.07845, over 28254.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3446, pruned_loss=0.09409, over 5687046.42 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.347, pruned_loss=0.09801, over 5751623.86 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3452, pruned_loss=0.09394, over 5676413.88 frames. ], batch size: 412, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:19:17,390 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:19:19,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.030e+02 1.368e+03 1.948e+03 2.798e+03 7.382e+03, threshold=3.895e+03, percent-clipped=12.0
+2023-03-05 17:19:21,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:19:52,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:19:59,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:20:08,778 INFO [train.py:968] (0/2) Epoch 11, batch 14550, giga_loss[loss=0.2425, simple_loss=0.3259, pruned_loss=0.07952, over 29042.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3404, pruned_loss=0.09181, over 5688937.07 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.347, pruned_loss=0.09835, over 5756224.45 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3408, pruned_loss=0.09122, over 5674112.44 frames. ], batch size: 285, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:21:14,694 INFO [train.py:968] (0/2) Epoch 11, batch 14600, giga_loss[loss=0.2293, simple_loss=0.294, pruned_loss=0.08226, over 24134.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3376, pruned_loss=0.08996, over 5690028.83 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3465, pruned_loss=0.09805, over 5760218.38 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3381, pruned_loss=0.08954, over 5672703.99 frames. ], batch size: 705, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:21:19,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2599, 1.6635, 1.4594, 1.1880], device='cuda:0'), covar=tensor([0.2351, 0.1518, 0.1146, 0.1549], device='cuda:0'), in_proj_covar=tensor([0.1658, 0.1551, 0.1501, 0.1607], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 17:21:20,020 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:21:22,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:21:26,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.407e+02 1.236e+03 1.579e+03 2.047e+03 3.620e+03, threshold=3.158e+03, percent-clipped=0.0
+2023-03-05 17:21:57,282 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:22:14,803 INFO [train.py:968] (0/2) Epoch 11, batch 14650, giga_loss[loss=0.2819, simple_loss=0.3485, pruned_loss=0.1076, over 28950.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3371, pruned_loss=0.09066, over 5686874.40 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3463, pruned_loss=0.09801, over 5762545.24 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3374, pruned_loss=0.09007, over 5668224.91 frames. ], batch size: 106, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:22:36,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3684, 5.1906, 4.9349, 2.2920], device='cuda:0'), covar=tensor([0.0363, 0.0555, 0.0729, 0.1806], device='cuda:0'), in_proj_covar=tensor([0.1020, 0.0956, 0.0835, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 17:22:53,297 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:22:56,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:23:03,737 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:23:17,128 INFO [train.py:968] (0/2) Epoch 11, batch 14700, giga_loss[loss=0.2843, simple_loss=0.3642, pruned_loss=0.1022, over 28798.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3414, pruned_loss=0.09266, over 5685903.65 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.346, pruned_loss=0.09788, over 5763224.62 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3418, pruned_loss=0.09223, over 5669518.38 frames. ], batch size: 119, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:23:21,583 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:23:27,331 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:23:32,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.097e+02 1.411e+03 1.716e+03 2.277e+03 5.599e+03, threshold=3.432e+03, percent-clipped=9.0
+2023-03-05 17:23:33,364 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:23:38,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7172, 1.9319, 1.8456, 1.6785], device='cuda:0'), covar=tensor([0.1371, 0.1895, 0.1696, 0.1774], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0707, 0.0651, 0.0638], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 17:23:52,911 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470268.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:24:17,161 INFO [train.py:968] (0/2) Epoch 11, batch 14750, giga_loss[loss=0.2668, simple_loss=0.3435, pruned_loss=0.095, over 28826.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3396, pruned_loss=0.0925, over 5690787.27 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3453, pruned_loss=0.09756, over 5765718.30 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3405, pruned_loss=0.09235, over 5674097.89 frames. ], batch size: 263, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:24:37,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0798, 1.5532, 1.4059, 1.0662], device='cuda:0'), covar=tensor([0.1373, 0.2041, 0.1130, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0685, 0.0852, 0.0761], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 17:24:52,789 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:25:25,559 INFO [train.py:968] (0/2) Epoch 11, batch 14800, giga_loss[loss=0.2742, simple_loss=0.3509, pruned_loss=0.0988, over 28965.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3405, pruned_loss=0.0939, over 5693379.46 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3451, pruned_loss=0.09738, over 5767093.65 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3413, pruned_loss=0.09392, over 5678227.59 frames. ], batch size: 284, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:25:42,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.050e+02 1.435e+03 1.989e+03 3.088e+03 1.122e+04, threshold=3.978e+03, percent-clipped=22.0
+2023-03-05 17:25:43,994 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-05 17:26:06,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:26:11,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:26:26,978 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6142, 2.3529, 1.8419, 0.7270], device='cuda:0'), covar=tensor([0.3950, 0.2056, 0.2669, 0.4193], device='cuda:0'), in_proj_covar=tensor([0.1524, 0.1445, 0.1464, 0.1247], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 17:26:29,149 INFO [train.py:968] (0/2) Epoch 11, batch 14850, libri_loss[loss=0.3212, simple_loss=0.3805, pruned_loss=0.131, over 29485.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3413, pruned_loss=0.0947, over 5689881.85 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3453, pruned_loss=0.09755, over 5767743.19 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3417, pruned_loss=0.09453, over 5676988.91 frames. ], batch size: 85, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:26:35,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0111, 3.8543, 3.6085, 1.7177], device='cuda:0'), covar=tensor([0.0592, 0.0666, 0.0722, 0.2239], device='cuda:0'), in_proj_covar=tensor([0.1021, 0.0956, 0.0834, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 17:26:48,886 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:27:02,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7019, 2.5471, 1.6524, 0.9242], device='cuda:0'), covar=tensor([0.4983, 0.2600, 0.2833, 0.4426], device='cuda:0'), in_proj_covar=tensor([0.1533, 0.1452, 0.1470, 0.1253], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 17:27:33,214 INFO [train.py:968] (0/2) Epoch 11, batch 14900, giga_loss[loss=0.323, simple_loss=0.3923, pruned_loss=0.1268, over 28975.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3434, pruned_loss=0.09503, over 5690156.69 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3446, pruned_loss=0.09727, over 5770000.50 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3442, pruned_loss=0.09509, over 5676633.84 frames. ], batch size: 120, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:27:53,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.824e+02 1.452e+03 1.955e+03 2.514e+03 6.526e+03, threshold=3.911e+03, percent-clipped=7.0
+2023-03-05 17:28:14,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9311, 3.7819, 3.5467, 1.7896], device='cuda:0'), covar=tensor([0.0506, 0.0654, 0.0697, 0.2311], device='cuda:0'), in_proj_covar=tensor([0.1019, 0.0949, 0.0829, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 17:28:45,259 INFO [train.py:968] (0/2) Epoch 11, batch 14950, giga_loss[loss=0.2477, simple_loss=0.3299, pruned_loss=0.08274, over 28623.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3424, pruned_loss=0.09389, over 5687288.40 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3441, pruned_loss=0.09715, over 5772061.77 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3435, pruned_loss=0.09395, over 5671406.16 frames. ], batch size: 242, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:28:59,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:29:29,282 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 17:29:58,972 INFO [train.py:968] (0/2) Epoch 11, batch 15000, giga_loss[loss=0.2612, simple_loss=0.3438, pruned_loss=0.08929, over 28363.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3398, pruned_loss=0.09309, over 5668022.02 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.344, pruned_loss=0.09705, over 5763217.56 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3408, pruned_loss=0.09313, over 5660904.75 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:29:58,977 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 17:30:08,327 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2039, simple_loss=0.3039, pruned_loss=0.052, over 944034.00 frames. 
+2023-03-05 17:30:08,328 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 17:30:24,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.876e+02 1.326e+03 1.743e+03 2.662e+03 6.929e+03, threshold=3.485e+03, percent-clipped=8.0
+2023-03-05 17:30:31,500 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-05 17:31:12,219 INFO [train.py:968] (0/2) Epoch 11, batch 15050, giga_loss[loss=0.2246, simple_loss=0.3067, pruned_loss=0.07126, over 28685.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3351, pruned_loss=0.09179, over 5649747.26 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3437, pruned_loss=0.09693, over 5745170.09 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.336, pruned_loss=0.09186, over 5658563.94 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:31:48,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:31:55,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:32:15,155 INFO [train.py:968] (0/2) Epoch 11, batch 15100, giga_loss[loss=0.2288, simple_loss=0.3095, pruned_loss=0.07409, over 29061.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3318, pruned_loss=0.09013, over 5654083.91 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3438, pruned_loss=0.09699, over 5745141.60 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3322, pruned_loss=0.09002, over 5659188.08 frames. ], batch size: 214, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:32:19,463 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470643.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:32:31,340 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.681e+02 1.461e+03 1.849e+03 2.611e+03 9.028e+03, threshold=3.698e+03, percent-clipped=12.0
+2023-03-05 17:32:57,526 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470673.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:33:00,873 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 17:33:14,062 INFO [train.py:968] (0/2) Epoch 11, batch 15150, giga_loss[loss=0.2797, simple_loss=0.3544, pruned_loss=0.1025, over 28879.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3318, pruned_loss=0.09002, over 5646884.94 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3437, pruned_loss=0.09695, over 5739493.72 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3321, pruned_loss=0.08984, over 5654347.18 frames. ], batch size: 284, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:33:16,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:34:09,174 INFO [train.py:968] (0/2) Epoch 11, batch 15200, giga_loss[loss=0.2204, simple_loss=0.3083, pruned_loss=0.06629, over 28675.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3314, pruned_loss=0.0894, over 5659236.75 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3432, pruned_loss=0.09667, over 5742094.23 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3317, pruned_loss=0.08931, over 5660693.15 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:34:29,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.676e+02 1.500e+03 1.839e+03 2.637e+03 9.048e+03, threshold=3.677e+03, percent-clipped=7.0
+2023-03-05 17:34:38,223 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:34:42,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:34:42,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5306, 2.1303, 1.5721, 0.7976], device='cuda:0'), covar=tensor([0.3301, 0.1915, 0.3131, 0.3907], device='cuda:0'), in_proj_covar=tensor([0.1525, 0.1449, 0.1467, 0.1247], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 17:34:46,037 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:34:50,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:35:12,349 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470786.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:35:15,501 INFO [train.py:968] (0/2) Epoch 11, batch 15250, giga_loss[loss=0.2394, simple_loss=0.307, pruned_loss=0.0859, over 24445.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3298, pruned_loss=0.0881, over 5648293.93 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.343, pruned_loss=0.09668, over 5744801.58 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3299, pruned_loss=0.08789, over 5645883.53 frames. ], batch size: 705, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:35:16,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470789.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:35:20,184 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:35:26,489 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:35:30,835 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:35:49,476 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470818.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:35:59,795 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-05 17:36:10,864 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:36:12,706 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3295, 1.4769, 1.3989, 1.3807], device='cuda:0'), covar=tensor([0.1739, 0.1471, 0.1210, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.1656, 0.1544, 0.1492, 0.1601], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 17:36:13,876 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:36:17,224 INFO [train.py:968] (0/2) Epoch 11, batch 15300, giga_loss[loss=0.2497, simple_loss=0.3227, pruned_loss=0.08834, over 29012.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3306, pruned_loss=0.08844, over 5663869.39 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3435, pruned_loss=0.09698, over 5746523.94 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3299, pruned_loss=0.08778, over 5658444.11 frames. ], batch size: 136, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:36:20,538 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:36:36,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.545e+02 1.191e+03 1.515e+03 1.973e+03 4.974e+03, threshold=3.031e+03, percent-clipped=2.0
+2023-03-05 17:36:55,835 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:37:06,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:37:21,515 INFO [train.py:968] (0/2) Epoch 11, batch 15350, giga_loss[loss=0.2368, simple_loss=0.3263, pruned_loss=0.07368, over 28420.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3299, pruned_loss=0.08841, over 5651159.09 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3432, pruned_loss=0.09689, over 5738412.51 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3292, pruned_loss=0.08768, over 5652001.76 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:38:26,169 INFO [train.py:968] (0/2) Epoch 11, batch 15400, giga_loss[loss=0.2553, simple_loss=0.3361, pruned_loss=0.08724, over 28723.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3311, pruned_loss=0.08861, over 5653530.25 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3432, pruned_loss=0.09687, over 5742229.33 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3303, pruned_loss=0.08783, over 5648581.63 frames. ], batch size: 262, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:38:44,245 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.971e+02 1.210e+03 1.595e+03 2.037e+03 6.211e+03, threshold=3.190e+03, percent-clipped=9.0
+2023-03-05 17:38:46,590 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:39:27,911 INFO [train.py:968] (0/2) Epoch 11, batch 15450, giga_loss[loss=0.2702, simple_loss=0.3408, pruned_loss=0.09984, over 27718.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3312, pruned_loss=0.08912, over 5649865.39 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3426, pruned_loss=0.09658, over 5734178.97 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3307, pruned_loss=0.08851, over 5650246.97 frames. ], batch size: 474, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:39:32,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3355, 1.5416, 0.9853, 1.1211], device='cuda:0'), covar=tensor([0.0910, 0.0628, 0.1471, 0.1305], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0433, 0.0491, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 17:39:42,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6121, 1.7285, 1.1876, 1.2862], device='cuda:0'), covar=tensor([0.0702, 0.0454, 0.0950, 0.0940], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0434, 0.0492, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 17:39:47,485 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1308, 3.9674, 3.7265, 1.7178], device='cuda:0'), covar=tensor([0.0554, 0.0694, 0.0775, 0.2287], device='cuda:0'), in_proj_covar=tensor([0.1015, 0.0944, 0.0821, 0.0641], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 17:40:05,521 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:40:12,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:40:34,086 INFO [train.py:968] (0/2) Epoch 11, batch 15500, giga_loss[loss=0.2532, simple_loss=0.3323, pruned_loss=0.0871, over 28753.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3308, pruned_loss=0.08968, over 5651792.41 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3425, pruned_loss=0.09662, over 5737533.88 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3303, pruned_loss=0.08903, over 5647607.25 frames. ], batch size: 262, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:40:48,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:40:48,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471048.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:40:57,291 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.803e+02 1.311e+03 1.653e+03 2.507e+03 6.136e+03, threshold=3.307e+03, percent-clipped=14.0
+2023-03-05 17:41:38,252 INFO [train.py:968] (0/2) Epoch 11, batch 15550, giga_loss[loss=0.2241, simple_loss=0.3214, pruned_loss=0.06337, over 28773.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3319, pruned_loss=0.08848, over 5663050.33 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3425, pruned_loss=0.09662, over 5737533.88 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3315, pruned_loss=0.08798, over 5659792.96 frames. ], batch size: 243, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:42:43,732 INFO [train.py:968] (0/2) Epoch 11, batch 15600, giga_loss[loss=0.2771, simple_loss=0.3541, pruned_loss=0.1, over 28422.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3348, pruned_loss=0.0893, over 5662496.85 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3423, pruned_loss=0.09652, over 5738515.21 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3346, pruned_loss=0.08897, over 5658591.80 frames. ], batch size: 369, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 17:43:03,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.448e+02 1.455e+03 1.768e+03 2.533e+03 7.878e+03, threshold=3.537e+03, percent-clipped=11.0
+2023-03-05 17:43:30,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:43:44,408 INFO [train.py:968] (0/2) Epoch 11, batch 15650, giga_loss[loss=0.263, simple_loss=0.3388, pruned_loss=0.09361, over 28079.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3366, pruned_loss=0.09043, over 5661665.86 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3421, pruned_loss=0.09641, over 5742150.21 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3365, pruned_loss=0.09013, over 5653758.59 frames. ], batch size: 412, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 17:43:47,030 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471191.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:43:49,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471194.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:44:15,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:44:24,045 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471223.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:44:42,001 INFO [train.py:968] (0/2) Epoch 11, batch 15700, giga_loss[loss=0.2933, simple_loss=0.3633, pruned_loss=0.1117, over 28581.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3366, pruned_loss=0.08996, over 5675348.45 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3421, pruned_loss=0.09623, over 5745067.05 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3364, pruned_loss=0.08977, over 5665276.73 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 17:44:58,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.447e+02 1.274e+03 1.709e+03 2.419e+03 4.469e+03, threshold=3.418e+03, percent-clipped=4.0
+2023-03-05 17:45:15,591 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=471269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:45:37,755 INFO [train.py:968] (0/2) Epoch 11, batch 15750, giga_loss[loss=0.2177, simple_loss=0.305, pruned_loss=0.06521, over 28466.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3347, pruned_loss=0.08882, over 5678212.10 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3416, pruned_loss=0.09596, over 5740607.48 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3348, pruned_loss=0.08864, over 5671140.09 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 17:46:14,350 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:46:18,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:46:23,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2793, 1.3078, 1.3628, 1.4606], device='cuda:0'), covar=tensor([0.0783, 0.0350, 0.0315, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0116, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:0')
+2023-03-05 17:46:32,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:46:43,170 INFO [train.py:968] (0/2) Epoch 11, batch 15800, giga_loss[loss=0.2381, simple_loss=0.3215, pruned_loss=0.07728, over 28578.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3314, pruned_loss=0.08626, over 5680988.38 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3413, pruned_loss=0.09582, over 5739575.87 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3316, pruned_loss=0.08613, over 5675731.34 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:47:00,351 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:47:03,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.803e+02 1.259e+03 1.681e+03 2.431e+03 8.120e+03, threshold=3.363e+03, percent-clipped=16.0
+2023-03-05 17:47:09,593 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:47:12,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:47:42,790 INFO [train.py:968] (0/2) Epoch 11, batch 15850, giga_loss[loss=0.2763, simple_loss=0.3554, pruned_loss=0.09858, over 28973.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.33, pruned_loss=0.08661, over 5680855.70 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3403, pruned_loss=0.09528, over 5743504.23 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3309, pruned_loss=0.08681, over 5671502.56 frames. ], batch size: 284, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:47:45,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:48:37,750 INFO [train.py:968] (0/2) Epoch 11, batch 15900, giga_loss[loss=0.2541, simple_loss=0.3381, pruned_loss=0.08502, over 28950.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3305, pruned_loss=0.08744, over 5681675.74 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3399, pruned_loss=0.09519, over 5745280.20 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3312, pruned_loss=0.08736, over 5670204.71 frames. ], batch size: 199, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:48:55,293 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=471454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:48:57,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.271e+02 1.281e+03 1.654e+03 2.336e+03 6.297e+03, threshold=3.308e+03, percent-clipped=5.0
+2023-03-05 17:49:22,915 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:49:24,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:49:29,310 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1935, 2.3221, 2.1980, 2.1492], device='cuda:0'), covar=tensor([0.1370, 0.1962, 0.1608, 0.1562], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0710, 0.0654, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 17:49:36,261 INFO [train.py:968] (0/2) Epoch 11, batch 15950, giga_loss[loss=0.2952, simple_loss=0.3734, pruned_loss=0.1085, over 28658.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3342, pruned_loss=0.08921, over 5681364.71 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3399, pruned_loss=0.09506, over 5747903.75 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3346, pruned_loss=0.08911, over 5668268.25 frames. ], batch size: 262, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:49:57,771 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:50:43,402 INFO [train.py:968] (0/2) Epoch 11, batch 16000, giga_loss[loss=0.2925, simple_loss=0.359, pruned_loss=0.1129, over 28936.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3346, pruned_loss=0.08977, over 5677246.56 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3396, pruned_loss=0.0948, over 5748007.10 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.335, pruned_loss=0.08979, over 5665257.97 frames. ], batch size: 199, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:50:56,473 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=471549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:51:03,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.179e+02 1.303e+03 1.806e+03 2.467e+03 8.615e+03, threshold=3.611e+03, percent-clipped=11.0
+2023-03-05 17:51:41,975 INFO [train.py:968] (0/2) Epoch 11, batch 16050, giga_loss[loss=0.2566, simple_loss=0.34, pruned_loss=0.08664, over 28917.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.337, pruned_loss=0.09094, over 5680969.38 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3395, pruned_loss=0.09465, over 5749271.35 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3374, pruned_loss=0.09101, over 5669088.84 frames. ], batch size: 186, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:52:22,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6779, 1.6915, 1.2665, 1.3266], device='cuda:0'), covar=tensor([0.0791, 0.0624, 0.0994, 0.1083], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0435, 0.0495, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 17:52:28,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6378, 1.5777, 1.1951, 1.2198], device='cuda:0'), covar=tensor([0.0680, 0.0520, 0.0916, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0435, 0.0495, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 17:52:38,760 INFO [train.py:968] (0/2) Epoch 11, batch 16100, giga_loss[loss=0.2829, simple_loss=0.364, pruned_loss=0.1009, over 28570.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.34, pruned_loss=0.09214, over 5686443.32 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3387, pruned_loss=0.09423, over 5751569.26 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.341, pruned_loss=0.09249, over 5673206.36 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:52:44,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:52:56,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.605e+02 1.338e+03 1.836e+03 2.591e+03 5.610e+03, threshold=3.672e+03, percent-clipped=8.0
+2023-03-05 17:53:38,444 INFO [train.py:968] (0/2) Epoch 11, batch 16150, giga_loss[loss=0.2799, simple_loss=0.3522, pruned_loss=0.1038, over 28176.00 frames. ], tot_loss[loss=0.263, simple_loss=0.341, pruned_loss=0.09254, over 5689705.14 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3382, pruned_loss=0.09397, over 5755308.38 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3423, pruned_loss=0.09302, over 5674461.89 frames. ], batch size: 412, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:54:46,197 INFO [train.py:968] (0/2) Epoch 11, batch 16200, giga_loss[loss=0.2549, simple_loss=0.3312, pruned_loss=0.08928, over 28641.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3395, pruned_loss=0.09158, over 5693966.19 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3377, pruned_loss=0.09368, over 5757745.23 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.341, pruned_loss=0.09219, over 5678161.74 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:55:08,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.311e+02 1.464e+03 1.871e+03 2.605e+03 7.099e+03, threshold=3.741e+03, percent-clipped=14.0
+2023-03-05 17:55:47,204 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3027, 1.6468, 1.2460, 1.4017], device='cuda:0'), covar=tensor([0.2303, 0.2048, 0.2464, 0.1897], device='cuda:0'), in_proj_covar=tensor([0.1283, 0.0946, 0.1140, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 17:55:47,727 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:55:48,964 INFO [train.py:968] (0/2) Epoch 11, batch 16250, giga_loss[loss=0.2766, simple_loss=0.3528, pruned_loss=0.1002, over 28480.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3389, pruned_loss=0.09207, over 5699242.95 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3376, pruned_loss=0.09359, over 5758145.55 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3402, pruned_loss=0.09259, over 5685461.00 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:55:51,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:56:25,505 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:56:37,523 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:56:49,796 INFO [train.py:968] (0/2) Epoch 11, batch 16300, libri_loss[loss=0.2997, simple_loss=0.3684, pruned_loss=0.1155, over 26248.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3374, pruned_loss=0.09134, over 5673277.01 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3371, pruned_loss=0.09339, over 5749402.58 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.339, pruned_loss=0.09189, over 5666766.94 frames. ], batch size: 136, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:56:58,018 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 17:57:02,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1471, 1.4398, 1.3730, 1.2633], device='cuda:0'), covar=tensor([0.1281, 0.1310, 0.1772, 0.1398], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0712, 0.0656, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 17:57:03,428 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=471851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:57:10,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.374e+02 1.215e+03 1.648e+03 2.244e+03 9.347e+03, threshold=3.297e+03, percent-clipped=9.0
+2023-03-05 17:57:24,279 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=471867.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:57:50,672 INFO [train.py:968] (0/2) Epoch 11, batch 16350, giga_loss[loss=0.2522, simple_loss=0.3263, pruned_loss=0.08902, over 28840.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3363, pruned_loss=0.09149, over 5679513.45 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3367, pruned_loss=0.09318, over 5752421.97 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.338, pruned_loss=0.09208, over 5669887.49 frames. ], batch size: 284, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:58:36,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:58:56,461 INFO [train.py:968] (0/2) Epoch 11, batch 16400, giga_loss[loss=0.2578, simple_loss=0.333, pruned_loss=0.09128, over 28155.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3362, pruned_loss=0.09252, over 5663228.22 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3369, pruned_loss=0.09334, over 5741080.85 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3373, pruned_loss=0.09282, over 5664834.59 frames. ], batch size: 412, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:59:16,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.910e+02 1.432e+03 2.131e+03 2.802e+03 9.805e+03, threshold=4.263e+03, percent-clipped=17.0
+2023-03-05 17:59:37,185 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:59:40,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:59:40,561 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-05 17:59:53,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3887, 1.4151, 1.1717, 1.5279], device='cuda:0'), covar=tensor([0.0795, 0.0312, 0.0347, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0116, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:0')
+2023-03-05 17:59:55,087 INFO [train.py:968] (0/2) Epoch 11, batch 16450, giga_loss[loss=0.2411, simple_loss=0.3226, pruned_loss=0.07985, over 27661.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3368, pruned_loss=0.09168, over 5670617.46 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3369, pruned_loss=0.09326, over 5745151.95 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3377, pruned_loss=0.09197, over 5665971.17 frames. ], batch size: 472, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:00:07,205 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-472000.pt
+2023-03-05 18:00:11,008 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:00:48,523 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-05 18:00:52,168 INFO [train.py:968] (0/2) Epoch 11, batch 16500, libri_loss[loss=0.2638, simple_loss=0.3517, pruned_loss=0.08795, over 29655.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3349, pruned_loss=0.08968, over 5674151.95 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3363, pruned_loss=0.09295, over 5747625.16 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3361, pruned_loss=0.09012, over 5666026.06 frames. ], batch size: 91, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:01:12,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.218e+02 1.428e+03 2.005e+03 3.086e+03 6.878e+03, threshold=4.009e+03, percent-clipped=7.0
+2023-03-05 18:01:23,718 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:01:27,437 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:01:50,480 INFO [train.py:968] (0/2) Epoch 11, batch 16550, giga_loss[loss=0.2793, simple_loss=0.3637, pruned_loss=0.09741, over 28471.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3368, pruned_loss=0.08882, over 5671030.96 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3362, pruned_loss=0.09289, over 5741107.53 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3378, pruned_loss=0.08912, over 5669157.57 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:02:02,769 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:02:05,615 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:02:45,295 INFO [train.py:968] (0/2) Epoch 11, batch 16600, giga_loss[loss=0.2716, simple_loss=0.3528, pruned_loss=0.0952, over 28404.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3375, pruned_loss=0.08826, over 5669324.81 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3366, pruned_loss=0.09314, over 5735243.21 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.338, pruned_loss=0.08816, over 5671014.47 frames. ], batch size: 369, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:02:52,608 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 18:03:05,111 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:03:05,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.831e+02 1.283e+03 1.663e+03 2.445e+03 5.698e+03, threshold=3.325e+03, percent-clipped=4.0
+2023-03-05 18:03:48,021 INFO [train.py:968] (0/2) Epoch 11, batch 16650, giga_loss[loss=0.2407, simple_loss=0.3282, pruned_loss=0.07664, over 28907.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3378, pruned_loss=0.08823, over 5677854.75 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3364, pruned_loss=0.09307, over 5737807.35 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3383, pruned_loss=0.08813, over 5676147.43 frames. ], batch size: 164, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:03:53,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4519, 1.6486, 1.3278, 1.6480], device='cuda:0'), covar=tensor([0.2208, 0.1999, 0.2249, 0.1905], device='cuda:0'), in_proj_covar=tensor([0.1282, 0.0943, 0.1140, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 18:03:56,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:04:39,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:04:53,070 INFO [train.py:968] (0/2) Epoch 11, batch 16700, giga_loss[loss=0.2409, simple_loss=0.3265, pruned_loss=0.07761, over 28729.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3382, pruned_loss=0.08899, over 5666886.10 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3365, pruned_loss=0.0931, over 5732791.29 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3386, pruned_loss=0.08877, over 5668810.91 frames. ], batch size: 243, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:04:56,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472242.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 18:05:14,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.748e+02 1.504e+03 1.928e+03 2.678e+03 5.722e+03, threshold=3.856e+03, percent-clipped=10.0
+2023-03-05 18:06:00,313 INFO [train.py:968] (0/2) Epoch 11, batch 16750, giga_loss[loss=0.245, simple_loss=0.3308, pruned_loss=0.07956, over 29012.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3388, pruned_loss=0.08928, over 5668118.51 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3364, pruned_loss=0.09307, over 5732743.33 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3393, pruned_loss=0.089, over 5667980.95 frames. ], batch size: 199, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:07:05,359 INFO [train.py:968] (0/2) Epoch 11, batch 16800, giga_loss[loss=0.2557, simple_loss=0.3367, pruned_loss=0.0874, over 29006.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3387, pruned_loss=0.08837, over 5680262.05 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3363, pruned_loss=0.093, over 5738130.73 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3393, pruned_loss=0.08804, over 5672802.52 frames. ], batch size: 285, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:07:30,763 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.645e+02 1.392e+03 1.769e+03 2.314e+03 6.309e+03, threshold=3.538e+03, percent-clipped=4.0
+2023-03-05 18:07:47,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5194, 1.5091, 1.2293, 1.6521], device='cuda:0'), covar=tensor([0.0740, 0.0296, 0.0326, 0.0830], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:0')
+2023-03-05 18:07:48,676 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:07:52,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:08:10,140 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472385.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 18:08:17,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472388.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 18:08:17,523 INFO [train.py:968] (0/2) Epoch 11, batch 16850, giga_loss[loss=0.3076, simple_loss=0.3794, pruned_loss=0.1179, over 27594.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3409, pruned_loss=0.08982, over 5676344.30 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3358, pruned_loss=0.0928, over 5741489.63 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3418, pruned_loss=0.08967, over 5666361.80 frames. ], batch size: 472, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:08:32,361 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:08:33,708 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:08:54,753 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472417.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 18:09:21,051 INFO [train.py:968] (0/2) Epoch 11, batch 16900, giga_loss[loss=0.2379, simple_loss=0.3301, pruned_loss=0.0729, over 28909.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3437, pruned_loss=0.09084, over 5684172.00 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3361, pruned_loss=0.09292, over 5737095.53 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3444, pruned_loss=0.09058, over 5678017.79 frames. ], batch size: 174, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:09:26,117 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:09:47,741 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.261e+02 1.311e+03 1.659e+03 2.091e+03 5.018e+03, threshold=3.317e+03, percent-clipped=3.0
+2023-03-05 18:09:59,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2957, 1.5639, 1.5782, 1.1622], device='cuda:0'), covar=tensor([0.1571, 0.2204, 0.1323, 0.1482], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0678, 0.0853, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:0')
+2023-03-05 18:10:10,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:10:27,164 INFO [train.py:968] (0/2) Epoch 11, batch 16950, giga_loss[loss=0.2606, simple_loss=0.3406, pruned_loss=0.09026, over 28123.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3424, pruned_loss=0.09088, over 5687530.94 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3362, pruned_loss=0.09317, over 5740951.90 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.343, pruned_loss=0.09038, over 5677804.26 frames. ], batch size: 412, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:11:31,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:11:40,123 INFO [train.py:968] (0/2) Epoch 11, batch 17000, giga_loss[loss=0.2492, simple_loss=0.3316, pruned_loss=0.08337, over 28866.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3385, pruned_loss=0.08906, over 5693809.20 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3358, pruned_loss=0.09293, over 5743386.88 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3394, pruned_loss=0.08884, over 5683224.23 frames. ], batch size: 174, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:12:01,152 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:12:04,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.007e+02 1.243e+03 1.739e+03 2.766e+03 1.092e+04, threshold=3.477e+03, percent-clipped=13.0
+2023-03-05 18:12:05,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1737, 1.0912, 3.9714, 3.2914], device='cuda:0'), covar=tensor([0.1715, 0.2856, 0.0351, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0650, 0.0586, 0.0837, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 18:12:14,313 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-05 18:12:19,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:12:50,055 INFO [train.py:968] (0/2) Epoch 11, batch 17050, giga_loss[loss=0.2604, simple_loss=0.3468, pruned_loss=0.08701, over 28905.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3368, pruned_loss=0.08714, over 5702313.85 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3359, pruned_loss=0.09294, over 5747476.56 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3375, pruned_loss=0.08681, over 5688994.56 frames. ], batch size: 284, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:13:27,268 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:13:30,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:13:51,991 INFO [train.py:968] (0/2) Epoch 11, batch 17100, giga_loss[loss=0.2431, simple_loss=0.3263, pruned_loss=0.07999, over 29022.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3367, pruned_loss=0.08718, over 5694397.26 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3358, pruned_loss=0.09292, over 5740125.38 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3374, pruned_loss=0.08686, over 5689869.26 frames. ], batch size: 155, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:14:05,927 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:14:16,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.066e+02 1.128e+03 1.516e+03 2.447e+03 5.079e+03, threshold=3.032e+03, percent-clipped=5.0
+2023-03-05 18:14:37,454 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:14:40,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:14:52,610 INFO [train.py:968] (0/2) Epoch 11, batch 17150, giga_loss[loss=0.2521, simple_loss=0.3339, pruned_loss=0.08516, over 28570.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3386, pruned_loss=0.08822, over 5689876.98 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.336, pruned_loss=0.09313, over 5743590.06 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3389, pruned_loss=0.08769, over 5682342.90 frames. ], batch size: 60, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:14:59,367 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:15:14,412 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:15:22,243 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:15:25,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:15:51,308 INFO [train.py:968] (0/2) Epoch 11, batch 17200, giga_loss[loss=0.2605, simple_loss=0.3397, pruned_loss=0.0906, over 28744.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3421, pruned_loss=0.09083, over 5685107.10 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3363, pruned_loss=0.09327, over 5745469.10 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3422, pruned_loss=0.09022, over 5676755.59 frames. ], batch size: 243, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 18:15:58,216 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:16:13,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.615e+02 1.371e+03 1.720e+03 2.498e+03 4.743e+03, threshold=3.439e+03, percent-clipped=9.0
+2023-03-05 18:16:30,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4345, 1.6282, 1.6092, 1.5095], device='cuda:0'), covar=tensor([0.1370, 0.1788, 0.1825, 0.1654], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0711, 0.0653, 0.0636], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 18:16:38,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:16:52,317 INFO [train.py:968] (0/2) Epoch 11, batch 17250, giga_loss[loss=0.2544, simple_loss=0.333, pruned_loss=0.08786, over 28922.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3404, pruned_loss=0.09099, over 5681921.45 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3363, pruned_loss=0.09327, over 5745469.10 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3404, pruned_loss=0.09051, over 5675421.37 frames. ], batch size: 213, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 18:17:25,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:17:46,586 INFO [train.py:968] (0/2) Epoch 11, batch 17300, giga_loss[loss=0.2592, simple_loss=0.3383, pruned_loss=0.09002, over 28678.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3398, pruned_loss=0.0915, over 5686004.21 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3362, pruned_loss=0.09318, over 5748850.41 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3401, pruned_loss=0.09116, over 5675539.80 frames. ], batch size: 262, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:17:57,311 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6075, 2.3112, 1.6275, 0.6880], device='cuda:0'), covar=tensor([0.3432, 0.2087, 0.3518, 0.4129], device='cuda:0'), in_proj_covar=tensor([0.1528, 0.1465, 0.1482, 0.1258], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 18:18:09,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.479e+02 1.453e+03 1.923e+03 2.957e+03 5.396e+03, threshold=3.847e+03, percent-clipped=16.0
+2023-03-05 18:18:42,603 INFO [train.py:968] (0/2) Epoch 11, batch 17350, giga_loss[loss=0.2617, simple_loss=0.3369, pruned_loss=0.09327, over 27498.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3409, pruned_loss=0.09252, over 5694099.89 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3358, pruned_loss=0.09298, over 5752713.77 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3416, pruned_loss=0.0924, over 5680759.60 frames. ], batch size: 472, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:19:19,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:19:21,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:19:27,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:19:34,757 INFO [train.py:968] (0/2) Epoch 11, batch 17400, giga_loss[loss=0.3229, simple_loss=0.3916, pruned_loss=0.1271, over 28972.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3488, pruned_loss=0.09761, over 5693576.79 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3355, pruned_loss=0.0928, over 5757194.45 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3498, pruned_loss=0.09771, over 5677247.63 frames. ], batch size: 164, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:19:47,005 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:19:53,404 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.391e+02 1.328e+03 1.686e+03 2.150e+03 5.295e+03, threshold=3.371e+03, percent-clipped=1.0
+2023-03-05 18:19:54,214 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:19:56,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:20:14,255 INFO [train.py:968] (0/2) Epoch 11, batch 17450, giga_loss[loss=0.3297, simple_loss=0.3981, pruned_loss=0.1306, over 28805.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3565, pruned_loss=0.1024, over 5692074.30 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3358, pruned_loss=0.09292, over 5751818.97 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3578, pruned_loss=0.1027, over 5681671.21 frames. ], batch size: 99, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:20:18,100 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:20:59,256 INFO [train.py:968] (0/2) Epoch 11, batch 17500, libri_loss[loss=0.2977, simple_loss=0.3749, pruned_loss=0.1102, over 28604.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.357, pruned_loss=0.1033, over 5695622.31 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3359, pruned_loss=0.09296, over 5752389.14 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3581, pruned_loss=0.1036, over 5686509.47 frames. ], batch size: 106, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:21:19,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.345e+02 1.133e+03 1.441e+03 1.960e+03 5.091e+03, threshold=2.882e+03, percent-clipped=3.0
+2023-03-05 18:21:29,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=473070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:21:31,393 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=473072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:21:33,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=473075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:21:44,971 INFO [train.py:968] (0/2) Epoch 11, batch 17550, libri_loss[loss=0.2438, simple_loss=0.3251, pruned_loss=0.08127, over 29565.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3498, pruned_loss=0.1002, over 5694139.77 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3361, pruned_loss=0.09296, over 5755258.58 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.351, pruned_loss=0.1007, over 5682598.78 frames. ], batch size: 80, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:21:58,299 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=473104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:22:29,159 INFO [train.py:968] (0/2) Epoch 11, batch 17600, giga_loss[loss=0.221, simple_loss=0.2947, pruned_loss=0.07365, over 28762.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3428, pruned_loss=0.09733, over 5688268.00 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3359, pruned_loss=0.09281, over 5755642.93 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3441, pruned_loss=0.09794, over 5677260.35 frames. ], batch size: 119, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:22:46,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.277e+02 1.065e+03 1.408e+03 1.818e+03 5.480e+03, threshold=2.817e+03, percent-clipped=8.0
+2023-03-05 18:23:12,147 INFO [train.py:968] (0/2) Epoch 11, batch 17650, giga_loss[loss=0.2099, simple_loss=0.286, pruned_loss=0.06691, over 29026.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3354, pruned_loss=0.09399, over 5688689.04 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3364, pruned_loss=0.09294, over 5760254.45 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3361, pruned_loss=0.09443, over 5674079.87 frames. ], batch size: 155, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:23:33,217 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=473213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:23:37,094 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=473216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:23:53,737 INFO [train.py:968] (0/2) Epoch 11, batch 17700, giga_loss[loss=0.2179, simple_loss=0.2889, pruned_loss=0.07342, over 28710.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3282, pruned_loss=0.09068, over 5695498.89 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3364, pruned_loss=0.0927, over 5761911.56 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3287, pruned_loss=0.09123, over 5680909.56 frames. ], batch size: 262, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:23:59,524 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=473245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:24:13,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.278e+02 1.036e+03 1.402e+03 1.977e+03 1.013e+04, threshold=2.805e+03, percent-clipped=9.0
+2023-03-05 18:24:37,163 INFO [train.py:968] (0/2) Epoch 11, batch 17750, giga_loss[loss=0.2325, simple_loss=0.3003, pruned_loss=0.08234, over 28817.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3222, pruned_loss=0.08788, over 5698978.85 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3364, pruned_loss=0.09266, over 5761793.61 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3223, pruned_loss=0.08831, over 5687177.97 frames. ], batch size: 119, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:24:54,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=473309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:25:19,186 INFO [train.py:968] (0/2) Epoch 11, batch 17800, giga_loss[loss=0.2254, simple_loss=0.3025, pruned_loss=0.07409, over 27968.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3195, pruned_loss=0.08667, over 5690990.69 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3365, pruned_loss=0.09263, over 5752083.68 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3192, pruned_loss=0.08691, over 5689108.52 frames. ], batch size: 412, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:25:35,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.381e+02 1.023e+03 1.261e+03 1.635e+03 3.263e+03, threshold=2.521e+03, percent-clipped=2.0
+2023-03-05 18:25:35,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=473360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:26:04,230 INFO [train.py:968] (0/2) Epoch 11, batch 17850, giga_loss[loss=0.2133, simple_loss=0.2955, pruned_loss=0.06553, over 28912.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3157, pruned_loss=0.08519, over 5691996.62 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3366, pruned_loss=0.0927, over 5754847.83 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3151, pruned_loss=0.08522, over 5687135.39 frames. ], batch size: 174, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:26:39,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0467, 3.2713, 2.1443, 0.9413], device='cuda:0'), covar=tensor([0.4984, 0.1816, 0.2850, 0.5044], device='cuda:0'), in_proj_covar=tensor([0.1509, 0.1453, 0.1463, 0.1240], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 18:26:49,039 INFO [train.py:968] (0/2) Epoch 11, batch 17900, giga_loss[loss=0.2079, simple_loss=0.281, pruned_loss=0.0674, over 28591.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3124, pruned_loss=0.08344, over 5699537.37 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3369, pruned_loss=0.09276, over 5755391.38 frames. ], giga_tot_loss[loss=0.2389, simple_loss=0.3112, pruned_loss=0.08328, over 5694215.45 frames. ], batch size: 78, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:27:05,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.192e+02 1.067e+03 1.411e+03 1.958e+03 5.666e+03, threshold=2.823e+03, percent-clipped=17.0
+2023-03-05 18:27:22,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5040, 2.2140, 1.6316, 0.6395], device='cuda:0'), covar=tensor([0.4309, 0.2232, 0.3117, 0.4924], device='cuda:0'), in_proj_covar=tensor([0.1519, 0.1461, 0.1468, 0.1249], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 18:27:27,747 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=473486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:27:29,489 INFO [train.py:968] (0/2) Epoch 11, batch 17950, giga_loss[loss=0.2416, simple_loss=0.3052, pruned_loss=0.089, over 28484.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3105, pruned_loss=0.08269, over 5696943.76 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3376, pruned_loss=0.09309, over 5758398.39 frames. ], giga_tot_loss[loss=0.2362, simple_loss=0.3084, pruned_loss=0.08204, over 5688655.20 frames. ], batch size: 60, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:28:09,981 INFO [train.py:968] (0/2) Epoch 11, batch 18000, giga_loss[loss=0.2055, simple_loss=0.2837, pruned_loss=0.06367, over 29028.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3095, pruned_loss=0.08221, over 5703123.49 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3378, pruned_loss=0.09299, over 5764343.29 frames. ], giga_tot_loss[loss=0.2346, simple_loss=0.3065, pruned_loss=0.08136, over 5688568.66 frames. ], batch size: 128, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:28:09,986 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 18:28:15,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9150, 3.6796, 3.5621, 1.6714], device='cuda:0'), covar=tensor([0.0701, 0.0848, 0.0807, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.1012, 0.0942, 0.0825, 0.0642], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 18:28:17,488 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6460, 1.6810, 1.3277, 1.3310], device='cuda:0'), covar=tensor([0.0797, 0.0470, 0.1006, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0432, 0.0498, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 18:28:18,982 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2151, simple_loss=0.3191, pruned_loss=0.05558, over 944034.00 frames. 
+2023-03-05 18:28:18,983 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 18:28:38,854 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.640e+02 1.007e+03 1.458e+03 2.252e+03 9.762e+03, threshold=2.917e+03, percent-clipped=16.0
+2023-03-05 18:29:03,925 INFO [train.py:968] (0/2) Epoch 11, batch 18050, giga_loss[loss=0.3177, simple_loss=0.355, pruned_loss=0.1402, over 26535.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3071, pruned_loss=0.08129, over 5696387.28 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3382, pruned_loss=0.09292, over 5764300.47 frames. ], giga_tot_loss[loss=0.2321, simple_loss=0.3035, pruned_loss=0.08034, over 5682886.84 frames. ], batch size: 555, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:29:46,276 INFO [train.py:968] (0/2) Epoch 11, batch 18100, giga_loss[loss=0.2025, simple_loss=0.2796, pruned_loss=0.0627, over 28972.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3043, pruned_loss=0.07963, over 5700666.55 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3383, pruned_loss=0.09293, over 5769334.24 frames. ], giga_tot_loss[loss=0.2283, simple_loss=0.3, pruned_loss=0.07836, over 5682664.38 frames. ], batch size: 164, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:29:57,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2687, 1.6348, 1.6078, 1.1860], device='cuda:0'), covar=tensor([0.1552, 0.2221, 0.1278, 0.1426], device='cuda:0'), in_proj_covar=tensor([0.0833, 0.0692, 0.0871, 0.0778], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 18:30:09,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.650e+02 9.567e+02 1.181e+03 1.601e+03 7.112e+03, threshold=2.361e+03, percent-clipped=7.0
+2023-03-05 18:30:24,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2315, 1.3903, 1.2841, 1.1454], device='cuda:0'), covar=tensor([0.1913, 0.1787, 0.0998, 0.1530], device='cuda:0'), in_proj_covar=tensor([0.1696, 0.1562, 0.1512, 0.1638], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 18:30:27,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=473684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:30:30,105 INFO [train.py:968] (0/2) Epoch 11, batch 18150, libri_loss[loss=0.2843, simple_loss=0.364, pruned_loss=0.1023, over 25977.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3022, pruned_loss=0.07863, over 5693391.18 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3388, pruned_loss=0.09323, over 5769959.05 frames. ], giga_tot_loss[loss=0.2249, simple_loss=0.2964, pruned_loss=0.07665, over 5674816.14 frames. ], batch size: 136, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:31:09,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=473735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:31:13,680 INFO [train.py:968] (0/2) Epoch 11, batch 18200, giga_loss[loss=0.2599, simple_loss=0.3325, pruned_loss=0.09362, over 28533.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3029, pruned_loss=0.07953, over 5689945.68 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.339, pruned_loss=0.09323, over 5771421.29 frames. ], giga_tot_loss[loss=0.2267, simple_loss=0.2978, pruned_loss=0.07779, over 5673149.14 frames. ], batch size: 71, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:31:35,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.831e+02 1.013e+03 1.371e+03 1.981e+03 4.809e+03, threshold=2.742e+03, percent-clipped=17.0
+2023-03-05 18:31:44,422 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-05 18:31:48,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4768, 1.6290, 1.2912, 1.8679], device='cuda:0'), covar=tensor([0.2601, 0.2619, 0.2922, 0.2059], device='cuda:0'), in_proj_covar=tensor([0.1291, 0.0957, 0.1149, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 18:32:00,925 INFO [train.py:968] (0/2) Epoch 11, batch 18250, giga_loss[loss=0.3244, simple_loss=0.3863, pruned_loss=0.1313, over 27551.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3153, pruned_loss=0.08642, over 5674236.97 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3395, pruned_loss=0.09345, over 5756465.26 frames. ], giga_tot_loss[loss=0.2392, simple_loss=0.3095, pruned_loss=0.08441, over 5671589.22 frames. ], batch size: 472, lr: 2.95e-03, grad_scale: 1.0
+2023-03-05 18:32:36,403 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=473827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:32:39,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=473830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:32:43,767 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=473836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:32:45,652 INFO [train.py:968] (0/2) Epoch 11, batch 18300, giga_loss[loss=0.2839, simple_loss=0.3592, pruned_loss=0.1043, over 28710.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.329, pruned_loss=0.09389, over 5679246.92 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3399, pruned_loss=0.09359, over 5754713.06 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3236, pruned_loss=0.09207, over 5677033.90 frames. ], batch size: 99, lr: 2.95e-03, grad_scale: 1.0
+2023-03-05 18:33:01,099 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=473859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:33:02,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=473861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:33:04,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.270e+03 1.632e+03 3.148e+03 1.452e+04, threshold=3.265e+03, percent-clipped=26.0
+2023-03-05 18:33:16,705 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=473878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:33:20,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=473881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:33:21,046 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 18:33:26,280 INFO [train.py:968] (0/2) Epoch 11, batch 18350, libri_loss[loss=0.2678, simple_loss=0.3533, pruned_loss=0.09117, over 29513.00 frames. ], tot_loss[loss=0.267, simple_loss=0.338, pruned_loss=0.09804, over 5684821.62 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3398, pruned_loss=0.09341, over 5759130.31 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3335, pruned_loss=0.09683, over 5677082.61 frames. ], batch size: 84, lr: 2.95e-03, grad_scale: 1.0
+2023-03-05 18:33:43,083 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=473910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:34:07,530 INFO [train.py:968] (0/2) Epoch 11, batch 18400, giga_loss[loss=0.2537, simple_loss=0.338, pruned_loss=0.0847, over 29019.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.344, pruned_loss=0.1002, over 5688371.25 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3398, pruned_loss=0.09342, over 5762182.81 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3405, pruned_loss=0.09934, over 5678032.32 frames. ], batch size: 128, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:34:28,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.225e+02 1.109e+03 1.323e+03 1.862e+03 5.231e+03, threshold=2.647e+03, percent-clipped=4.0
+2023-03-05 18:34:50,926 INFO [train.py:968] (0/2) Epoch 11, batch 18450, giga_loss[loss=0.3087, simple_loss=0.3838, pruned_loss=0.1168, over 28486.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3455, pruned_loss=0.09934, over 5693915.94 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3399, pruned_loss=0.09345, over 5763707.71 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3428, pruned_loss=0.09875, over 5683450.72 frames. ], batch size: 336, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:35:02,756 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-474000.pt
+2023-03-05 18:35:06,796 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:35:09,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:35:35,360 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=474036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:35:38,345 INFO [train.py:968] (0/2) Epoch 11, batch 18500, giga_loss[loss=0.309, simple_loss=0.3683, pruned_loss=0.1249, over 26724.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3471, pruned_loss=0.09976, over 5668387.82 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3403, pruned_loss=0.09376, over 5755942.46 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3446, pruned_loss=0.09906, over 5665933.51 frames. ], batch size: 555, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:35:58,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.468e+02 1.058e+03 1.333e+03 1.541e+03 1.655e+04, threshold=2.666e+03, percent-clipped=10.0
+2023-03-05 18:36:24,018 INFO [train.py:968] (0/2) Epoch 11, batch 18550, giga_loss[loss=0.3022, simple_loss=0.3654, pruned_loss=0.1195, over 27687.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3498, pruned_loss=0.1022, over 5673966.83 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3406, pruned_loss=0.09389, over 5758638.37 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3478, pruned_loss=0.1016, over 5668157.06 frames. ], batch size: 472, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:37:08,129 INFO [train.py:968] (0/2) Epoch 11, batch 18600, giga_loss[loss=0.3138, simple_loss=0.3765, pruned_loss=0.1256, over 28636.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.354, pruned_loss=0.1053, over 5679899.80 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3411, pruned_loss=0.0941, over 5761532.29 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3521, pruned_loss=0.1049, over 5671017.69 frames. ], batch size: 85, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:37:26,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.782e+02 1.162e+03 1.541e+03 1.930e+03 4.293e+03, threshold=3.082e+03, percent-clipped=9.0
+2023-03-05 18:37:49,211 INFO [train.py:968] (0/2) Epoch 11, batch 18650, giga_loss[loss=0.3227, simple_loss=0.3896, pruned_loss=0.1279, over 28940.00 frames. ], tot_loss[loss=0.285, simple_loss=0.357, pruned_loss=0.1065, over 5682944.84 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3413, pruned_loss=0.094, over 5761919.95 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3557, pruned_loss=0.1066, over 5673605.03 frames. ], batch size: 136, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:37:49,418 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:38:01,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5942, 4.3872, 4.1965, 1.9565], device='cuda:0'), covar=tensor([0.0474, 0.0646, 0.0661, 0.2077], device='cuda:0'), in_proj_covar=tensor([0.1004, 0.0937, 0.0822, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 18:38:08,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:38:29,245 INFO [train.py:968] (0/2) Epoch 11, batch 18700, libri_loss[loss=0.216, simple_loss=0.2927, pruned_loss=0.06966, over 29332.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.359, pruned_loss=0.1069, over 5693257.41 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.341, pruned_loss=0.09376, over 5766908.32 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3588, pruned_loss=0.1076, over 5678848.14 frames. ], batch size: 67, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:38:30,152 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:38:40,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0059, 1.2664, 1.2977, 1.1049], device='cuda:0'), covar=tensor([0.1398, 0.1086, 0.1910, 0.1366], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0723, 0.0665, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 18:38:41,224 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5986, 1.8017, 1.4333, 1.7522], device='cuda:0'), covar=tensor([0.2343, 0.2326, 0.2550, 0.2200], device='cuda:0'), in_proj_covar=tensor([0.1289, 0.0955, 0.1145, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 18:38:50,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.744e+02 1.139e+03 1.347e+03 1.617e+03 3.572e+03, threshold=2.695e+03, percent-clipped=4.0
+2023-03-05 18:39:11,675 INFO [train.py:968] (0/2) Epoch 11, batch 18750, giga_loss[loss=0.2855, simple_loss=0.3653, pruned_loss=0.1028, over 28951.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3614, pruned_loss=0.1077, over 5684857.34 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3413, pruned_loss=0.09405, over 5760326.57 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3612, pruned_loss=0.1082, over 5678406.01 frames. ], batch size: 145, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:39:11,866 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:39:15,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9951, 1.2507, 1.2774, 1.0945], device='cuda:0'), covar=tensor([0.1438, 0.1157, 0.2009, 0.1440], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0729, 0.0669, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 18:39:17,071 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:39:46,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4868, 1.6488, 1.5255, 1.2584], device='cuda:0'), covar=tensor([0.2155, 0.2025, 0.1456, 0.1900], device='cuda:0'), in_proj_covar=tensor([0.1683, 0.1556, 0.1524, 0.1631], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 18:39:50,798 INFO [train.py:968] (0/2) Epoch 11, batch 18800, giga_loss[loss=0.2509, simple_loss=0.3453, pruned_loss=0.07828, over 28686.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3627, pruned_loss=0.1078, over 5687822.09 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.342, pruned_loss=0.09436, over 5761329.85 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3625, pruned_loss=0.1082, over 5680146.26 frames. ], batch size: 242, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:40:03,704 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:40:05,818 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:40:09,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.530e+02 1.109e+03 1.473e+03 1.809e+03 3.997e+03, threshold=2.945e+03, percent-clipped=7.0
+2023-03-05 18:40:28,742 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=474386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:40:31,144 INFO [train.py:968] (0/2) Epoch 11, batch 18850, giga_loss[loss=0.3053, simple_loss=0.376, pruned_loss=0.1174, over 28650.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3615, pruned_loss=0.1055, over 5694079.88 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.342, pruned_loss=0.09428, over 5752544.08 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3615, pruned_loss=0.1061, over 5695185.28 frames. ], batch size: 92, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:41:11,007 INFO [train.py:968] (0/2) Epoch 11, batch 18900, giga_loss[loss=0.3281, simple_loss=0.3798, pruned_loss=0.1382, over 26654.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3591, pruned_loss=0.103, over 5681778.80 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3422, pruned_loss=0.09445, over 5735532.55 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3591, pruned_loss=0.1034, over 5697845.86 frames. ], batch size: 555, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:41:28,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.472e+02 1.199e+03 1.559e+03 1.964e+03 5.851e+03, threshold=3.118e+03, percent-clipped=7.0
+2023-03-05 18:41:48,359 INFO [train.py:968] (0/2) Epoch 11, batch 18950, giga_loss[loss=0.2809, simple_loss=0.3598, pruned_loss=0.1009, over 29009.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3598, pruned_loss=0.104, over 5692492.21 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3425, pruned_loss=0.09459, over 5739539.13 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.36, pruned_loss=0.1045, over 5700440.00 frames. ], batch size: 227, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:41:48,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:41:52,372 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3655, 3.9035, 1.6060, 1.5585], device='cuda:0'), covar=tensor([0.0978, 0.0239, 0.0829, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0496, 0.0333, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 18:41:57,253 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-05 18:42:31,296 INFO [train.py:968] (0/2) Epoch 11, batch 19000, giga_loss[loss=0.3671, simple_loss=0.4137, pruned_loss=0.1602, over 28847.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3608, pruned_loss=0.1067, over 5705248.20 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.343, pruned_loss=0.09495, over 5745468.31 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3613, pruned_loss=0.1071, over 5704545.05 frames. ], batch size: 106, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:42:54,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.035e+02 1.402e+03 1.747e+03 2.248e+03 7.319e+03, threshold=3.494e+03, percent-clipped=8.0
+2023-03-05 18:42:56,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:43:16,244 INFO [train.py:968] (0/2) Epoch 11, batch 19050, giga_loss[loss=0.3939, simple_loss=0.4225, pruned_loss=0.1826, over 26554.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3628, pruned_loss=0.1106, over 5708340.74 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3428, pruned_loss=0.09483, over 5747864.88 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3636, pruned_loss=0.1114, over 5704896.22 frames. ], batch size: 555, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:43:37,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:43:55,927 INFO [train.py:968] (0/2) Epoch 11, batch 19100, libri_loss[loss=0.2455, simple_loss=0.3158, pruned_loss=0.08757, over 29577.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3614, pruned_loss=0.1105, over 5709704.93 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.343, pruned_loss=0.0948, over 5754259.83 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3626, pruned_loss=0.1117, over 5699683.80 frames. ], batch size: 74, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:44:14,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.849e+02 1.214e+03 1.483e+03 2.378e+03 6.842e+03, threshold=2.965e+03, percent-clipped=9.0
+2023-03-05 18:44:15,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:44:22,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:44:29,515 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:44:37,355 INFO [train.py:968] (0/2) Epoch 11, batch 19150, giga_loss[loss=0.3018, simple_loss=0.3609, pruned_loss=0.1214, over 28766.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3592, pruned_loss=0.1095, over 5707065.31 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3433, pruned_loss=0.09484, over 5755307.10 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3604, pruned_loss=0.1109, over 5696905.93 frames. ], batch size: 99, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:44:53,873 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:44:55,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:45:07,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4771, 1.7191, 1.3586, 1.6345], device='cuda:0'), covar=tensor([0.2288, 0.2201, 0.2367, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.1287, 0.0952, 0.1140, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 18:45:12,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7698, 1.9376, 1.9583, 1.6460], device='cuda:0'), covar=tensor([0.1677, 0.2038, 0.1293, 0.1468], device='cuda:0'), in_proj_covar=tensor([0.0826, 0.0692, 0.0864, 0.0771], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 18:45:19,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4723, 1.5693, 1.5742, 1.4582], device='cuda:0'), covar=tensor([0.1340, 0.1712, 0.1868, 0.1609], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0728, 0.0667, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 18:45:21,847 INFO [train.py:968] (0/2) Epoch 11, batch 19200, giga_loss[loss=0.26, simple_loss=0.3383, pruned_loss=0.09089, over 28497.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3575, pruned_loss=0.1081, over 5713238.65 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3433, pruned_loss=0.09457, over 5757424.36 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.359, pruned_loss=0.1098, over 5701795.95 frames. ], batch size: 71, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:45:22,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=474739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:45:39,352 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:45:42,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:45:43,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.846e+02 1.155e+03 1.426e+03 1.753e+03 4.492e+03, threshold=2.852e+03, percent-clipped=1.0
+2023-03-05 18:45:43,697 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-05 18:45:50,519 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-05 18:46:02,867 INFO [train.py:968] (0/2) Epoch 11, batch 19250, giga_loss[loss=0.2477, simple_loss=0.3328, pruned_loss=0.08133, over 28889.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3563, pruned_loss=0.1064, over 5719994.79 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3433, pruned_loss=0.09446, over 5760358.40 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3576, pruned_loss=0.1081, over 5707448.82 frames. ], batch size: 199, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:46:03,739 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=474790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:09,079 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-05 18:46:18,126 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:21,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:25,520 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:29,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:47,932 INFO [train.py:968] (0/2) Epoch 11, batch 19300, giga_loss[loss=0.2398, simple_loss=0.3257, pruned_loss=0.07693, over 28902.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3537, pruned_loss=0.1044, over 5700870.03 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3439, pruned_loss=0.09463, over 5762221.50 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3545, pruned_loss=0.1059, over 5688052.17 frames. ], batch size: 174, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:46:48,169 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=474839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:53,273 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=474847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:47:08,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.426e+02 1.068e+03 1.350e+03 1.813e+03 6.733e+03, threshold=2.699e+03, percent-clipped=5.0
+2023-03-05 18:47:08,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:47:29,457 INFO [train.py:968] (0/2) Epoch 11, batch 19350, giga_loss[loss=0.2308, simple_loss=0.3122, pruned_loss=0.07464, over 28977.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3483, pruned_loss=0.101, over 5708152.86 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3439, pruned_loss=0.09454, over 5766318.03 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3492, pruned_loss=0.1026, over 5691852.94 frames. ], batch size: 227, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:48:08,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5375, 3.4400, 1.6332, 1.5534], device='cuda:0'), covar=tensor([0.0891, 0.0246, 0.0844, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0498, 0.0333, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 18:48:12,771 INFO [train.py:968] (0/2) Epoch 11, batch 19400, libri_loss[loss=0.2276, simple_loss=0.3049, pruned_loss=0.07516, over 29654.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3437, pruned_loss=0.09922, over 5696143.43 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3438, pruned_loss=0.09451, over 5771007.46 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3446, pruned_loss=0.1007, over 5676058.04 frames. ], batch size: 69, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:48:33,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.135e+02 1.017e+03 1.248e+03 2.083e+03 1.039e+04, threshold=2.496e+03, percent-clipped=15.0
+2023-03-05 18:48:57,254 INFO [train.py:968] (0/2) Epoch 11, batch 19450, giga_loss[loss=0.2564, simple_loss=0.3278, pruned_loss=0.09243, over 28385.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3387, pruned_loss=0.09672, over 5676637.99 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3443, pruned_loss=0.0947, over 5766878.21 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3389, pruned_loss=0.0979, over 5661985.67 frames. ], batch size: 369, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:49:13,605 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=475007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:49:15,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=475010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:49:39,519 INFO [train.py:968] (0/2) Epoch 11, batch 19500, giga_loss[loss=0.2882, simple_loss=0.3602, pruned_loss=0.1081, over 28772.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3386, pruned_loss=0.09683, over 5658278.13 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3447, pruned_loss=0.09473, over 5763980.55 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3381, pruned_loss=0.09785, over 5645156.23 frames. ], batch size: 284, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:49:39,727 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=475039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:49:52,184 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=475053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:50:01,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.142e+02 1.010e+03 1.273e+03 1.801e+03 5.890e+03, threshold=2.546e+03, percent-clipped=11.0
+2023-03-05 18:50:24,211 INFO [train.py:968] (0/2) Epoch 11, batch 19550, giga_loss[loss=0.224, simple_loss=0.3055, pruned_loss=0.07129, over 28628.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3392, pruned_loss=0.09728, over 5668007.09 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3445, pruned_loss=0.09463, over 5765231.11 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3389, pruned_loss=0.0982, over 5655667.86 frames. ], batch size: 60, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:51:00,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6191, 1.8000, 1.5179, 1.3564], device='cuda:0'), covar=tensor([0.2098, 0.1672, 0.1440, 0.1935], device='cuda:0'), in_proj_covar=tensor([0.1685, 0.1555, 0.1520, 0.1648], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 18:51:07,318 INFO [train.py:968] (0/2) Epoch 11, batch 19600, giga_loss[loss=0.2435, simple_loss=0.3211, pruned_loss=0.08292, over 28970.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3388, pruned_loss=0.09655, over 5672683.99 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3451, pruned_loss=0.09468, over 5759716.47 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3378, pruned_loss=0.09735, over 5664028.18 frames. ], batch size: 213, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:51:28,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.546e+02 1.147e+03 1.527e+03 2.226e+03 1.017e+04, threshold=3.053e+03, percent-clipped=16.0
+2023-03-05 18:51:49,847 INFO [train.py:968] (0/2) Epoch 11, batch 19650, giga_loss[loss=0.2762, simple_loss=0.3478, pruned_loss=0.1023, over 28028.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3378, pruned_loss=0.09624, over 5677298.76 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3457, pruned_loss=0.09485, over 5759158.88 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3364, pruned_loss=0.09675, over 5669274.16 frames. ], batch size: 412, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:51:55,027 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=475196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:51:57,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=475199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:52:02,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=475205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:52:20,747 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=475228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:52:29,128 INFO [train.py:968] (0/2) Epoch 11, batch 19700, giga_loss[loss=0.2987, simple_loss=0.3621, pruned_loss=0.1176, over 27962.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3361, pruned_loss=0.09527, over 5689833.55 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3462, pruned_loss=0.09499, over 5764004.64 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3342, pruned_loss=0.09559, over 5676304.52 frames. ], batch size: 412, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:52:47,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.500e+02 1.044e+03 1.378e+03 2.068e+03 5.031e+03, threshold=2.755e+03, percent-clipped=8.0
+2023-03-05 18:53:06,589 INFO [train.py:968] (0/2) Epoch 11, batch 19750, giga_loss[loss=0.2214, simple_loss=0.3055, pruned_loss=0.06864, over 28923.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3339, pruned_loss=0.09364, over 5694955.73 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3466, pruned_loss=0.09499, over 5759184.07 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3316, pruned_loss=0.09389, over 5685298.46 frames. ], batch size: 174, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:53:20,172 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3958, 1.4670, 1.5521, 1.3703], device='cuda:0'), covar=tensor([0.1410, 0.1852, 0.1899, 0.1795], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0728, 0.0668, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 18:53:48,990 INFO [train.py:968] (0/2) Epoch 11, batch 19800, giga_loss[loss=0.2857, simple_loss=0.3489, pruned_loss=0.1113, over 26680.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3323, pruned_loss=0.09302, over 5696117.59 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3475, pruned_loss=0.09544, over 5760954.76 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3294, pruned_loss=0.09274, over 5685618.41 frames. ], batch size: 555, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:54:07,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.065e+02 9.520e+02 1.146e+03 1.658e+03 5.546e+03, threshold=2.292e+03, percent-clipped=10.0
+2023-03-05 18:54:28,667 INFO [train.py:968] (0/2) Epoch 11, batch 19850, giga_loss[loss=0.2282, simple_loss=0.3055, pruned_loss=0.0755, over 28935.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3279, pruned_loss=0.09106, over 5708986.59 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3477, pruned_loss=0.09551, over 5762414.45 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3253, pruned_loss=0.09073, over 5698893.95 frames. ], batch size: 227, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:55:03,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4778, 1.6847, 1.7712, 1.3187], device='cuda:0'), covar=tensor([0.1689, 0.2121, 0.1307, 0.1482], device='cuda:0'), in_proj_covar=tensor([0.0832, 0.0693, 0.0870, 0.0776], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 18:55:09,239 INFO [train.py:968] (0/2) Epoch 11, batch 19900, giga_loss[loss=0.2405, simple_loss=0.3138, pruned_loss=0.08362, over 28708.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3267, pruned_loss=0.09006, over 5719785.64 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3487, pruned_loss=0.09586, over 5766594.51 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.323, pruned_loss=0.08932, over 5706577.05 frames. ], batch size: 92, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:55:21,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3263, 1.5408, 1.3294, 1.4493], device='cuda:0'), covar=tensor([0.0780, 0.0311, 0.0319, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0112, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 18:55:28,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.646e+02 9.748e+02 1.353e+03 2.142e+03 5.832e+03, threshold=2.705e+03, percent-clipped=20.0
+2023-03-05 18:55:46,893 INFO [train.py:968] (0/2) Epoch 11, batch 19950, giga_loss[loss=0.2117, simple_loss=0.2926, pruned_loss=0.06538, over 28938.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3247, pruned_loss=0.08914, over 5720870.43 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3494, pruned_loss=0.09614, over 5765813.14 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3207, pruned_loss=0.08818, over 5709836.03 frames. ], batch size: 186, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:56:24,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2714, 1.4359, 1.2548, 1.4798], device='cuda:0'), covar=tensor([0.0787, 0.0328, 0.0331, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:0')
+2023-03-05 18:56:28,041 INFO [train.py:968] (0/2) Epoch 11, batch 20000, giga_loss[loss=0.2403, simple_loss=0.3101, pruned_loss=0.08521, over 28719.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3228, pruned_loss=0.08848, over 5720704.93 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3494, pruned_loss=0.09607, over 5767103.21 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3195, pruned_loss=0.08773, over 5710559.56 frames. ], batch size: 66, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:56:49,530 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.831e+02 9.431e+02 1.112e+03 1.453e+03 3.057e+03, threshold=2.223e+03, percent-clipped=3.0
+2023-03-05 18:57:01,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=475580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:57:08,417 INFO [train.py:968] (0/2) Epoch 11, batch 20050, giga_loss[loss=0.2828, simple_loss=0.355, pruned_loss=0.1053, over 28582.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3236, pruned_loss=0.08863, over 5723900.05 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3501, pruned_loss=0.0964, over 5768653.02 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3199, pruned_loss=0.08764, over 5713923.51 frames. ], batch size: 307, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:57:51,071 INFO [train.py:968] (0/2) Epoch 11, batch 20100, giga_loss[loss=0.2795, simple_loss=0.3512, pruned_loss=0.1039, over 28767.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3275, pruned_loss=0.09155, over 5720281.45 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3505, pruned_loss=0.09666, over 5770072.93 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.324, pruned_loss=0.09047, over 5710603.41 frames. ], batch size: 262, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:58:08,376 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2803, 1.4869, 1.1141, 1.1499], device='cuda:0'), covar=tensor([0.0994, 0.0600, 0.1358, 0.1108], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0441, 0.0504, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 18:58:12,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.239e+02 1.130e+03 1.426e+03 1.951e+03 5.206e+03, threshold=2.852e+03, percent-clipped=14.0
+2023-03-05 18:58:37,206 INFO [train.py:968] (0/2) Epoch 11, batch 20150, giga_loss[loss=0.3269, simple_loss=0.3926, pruned_loss=0.1306, over 28765.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3341, pruned_loss=0.09582, over 5710465.91 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3511, pruned_loss=0.09687, over 5773722.87 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3303, pruned_loss=0.09467, over 5698153.07 frames. ], batch size: 284, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:59:10,444 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=475723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:59:13,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=475726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:59:25,780 INFO [train.py:968] (0/2) Epoch 11, batch 20200, giga_loss[loss=0.2783, simple_loss=0.3443, pruned_loss=0.1062, over 28629.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3414, pruned_loss=0.1002, over 5700998.32 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3514, pruned_loss=0.09679, over 5766832.02 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3377, pruned_loss=0.09936, over 5695681.25 frames. ], batch size: 85, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:59:39,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=475755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:59:48,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.321e+02 1.291e+03 1.575e+03 2.331e+03 5.356e+03, threshold=3.151e+03, percent-clipped=15.0
+2023-03-05 19:00:09,260 INFO [train.py:968] (0/2) Epoch 11, batch 20250, giga_loss[loss=0.2974, simple_loss=0.3648, pruned_loss=0.115, over 28796.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3475, pruned_loss=0.1041, over 5695502.97 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3516, pruned_loss=0.09704, over 5769449.78 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3442, pruned_loss=0.1033, over 5687787.02 frames. ], batch size: 119, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 19:00:25,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=475806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:00:58,994 INFO [train.py:968] (0/2) Epoch 11, batch 20300, giga_loss[loss=0.3373, simple_loss=0.3829, pruned_loss=0.1459, over 23651.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3532, pruned_loss=0.1063, over 5686888.78 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3516, pruned_loss=0.09703, over 5766184.54 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3506, pruned_loss=0.1057, over 5683142.30 frames. ], batch size: 705, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 19:01:13,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9251, 1.1347, 3.5707, 2.8707], device='cuda:0'), covar=tensor([0.1839, 0.2736, 0.0454, 0.1004], device='cuda:0'), in_proj_covar=tensor([0.0648, 0.0578, 0.0832, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 19:01:22,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.873e+02 1.070e+03 1.381e+03 1.718e+03 5.867e+03, threshold=2.763e+03, percent-clipped=2.0
+2023-03-05 19:01:46,329 INFO [train.py:968] (0/2) Epoch 11, batch 20350, giga_loss[loss=0.2796, simple_loss=0.3598, pruned_loss=0.0997, over 28778.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3581, pruned_loss=0.1085, over 5686362.46 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3519, pruned_loss=0.09713, over 5768450.60 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3558, pruned_loss=0.1081, over 5680131.05 frames. ], batch size: 262, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 19:01:47,282 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=475890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:02:28,329 INFO [train.py:968] (0/2) Epoch 11, batch 20400, libri_loss[loss=0.2877, simple_loss=0.3658, pruned_loss=0.1048, over 26129.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3609, pruned_loss=0.1097, over 5694923.74 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3518, pruned_loss=0.09703, over 5769132.64 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3593, pruned_loss=0.1099, over 5687673.49 frames. ], batch size: 136, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 19:02:33,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6170, 3.4328, 3.1917, 1.6127], device='cuda:0'), covar=tensor([0.0728, 0.0789, 0.0766, 0.2419], device='cuda:0'), in_proj_covar=tensor([0.1023, 0.0953, 0.0834, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 19:02:36,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4185, 1.6887, 1.4652, 1.2701], device='cuda:0'), covar=tensor([0.1724, 0.1499, 0.1030, 0.1543], device='cuda:0'), in_proj_covar=tensor([0.1686, 0.1561, 0.1534, 0.1660], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 19:02:48,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.664e+02 1.084e+03 1.478e+03 2.023e+03 7.815e+03, threshold=2.955e+03, percent-clipped=13.0
+2023-03-05 19:03:08,860 INFO [train.py:968] (0/2) Epoch 11, batch 20450, giga_loss[loss=0.23, simple_loss=0.3104, pruned_loss=0.07479, over 28802.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3565, pruned_loss=0.1068, over 5687504.89 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3518, pruned_loss=0.09704, over 5763735.68 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3555, pruned_loss=0.1073, over 5683670.50 frames. ], batch size: 186, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 19:03:20,053 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-476000.pt
+2023-03-05 19:03:47,615 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8225, 3.6303, 3.4456, 1.7674], device='cuda:0'), covar=tensor([0.0681, 0.0799, 0.0729, 0.2220], device='cuda:0'), in_proj_covar=tensor([0.1022, 0.0953, 0.0831, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 19:03:49,980 INFO [train.py:968] (0/2) Epoch 11, batch 20500, giga_loss[loss=0.2583, simple_loss=0.3383, pruned_loss=0.08915, over 28540.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3536, pruned_loss=0.104, over 5702906.19 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3517, pruned_loss=0.09708, over 5768136.71 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3529, pruned_loss=0.1046, over 5693975.14 frames. ], batch size: 307, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 19:03:50,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5390, 1.7964, 1.8269, 1.3660], device='cuda:0'), covar=tensor([0.1825, 0.2237, 0.1461, 0.1710], device='cuda:0'), in_proj_covar=tensor([0.0828, 0.0690, 0.0868, 0.0772], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 19:04:14,049 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.160e+02 1.022e+03 1.295e+03 1.729e+03 3.904e+03, threshold=2.590e+03, percent-clipped=2.0
+2023-03-05 19:04:31,601 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-05 19:04:34,458 INFO [train.py:968] (0/2) Epoch 11, batch 20550, giga_loss[loss=0.2629, simple_loss=0.3451, pruned_loss=0.09032, over 28947.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3522, pruned_loss=0.1027, over 5691111.51 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3517, pruned_loss=0.09702, over 5768554.34 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3518, pruned_loss=0.1033, over 5682813.56 frames. ], batch size: 128, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 19:05:08,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1299, 1.1644, 4.1153, 3.3278], device='cuda:0'), covar=tensor([0.1746, 0.2716, 0.0405, 0.0747], device='cuda:0'), in_proj_covar=tensor([0.0653, 0.0581, 0.0842, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 19:05:14,816 INFO [train.py:968] (0/2) Epoch 11, batch 20600, giga_loss[loss=0.3103, simple_loss=0.3744, pruned_loss=0.1231, over 28819.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3537, pruned_loss=0.1031, over 5689191.31 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3519, pruned_loss=0.09722, over 5760756.41 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3532, pruned_loss=0.1035, over 5688599.46 frames. ], batch size: 284, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:05:39,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.590e+02 1.098e+03 1.399e+03 1.965e+03 4.178e+03, threshold=2.798e+03, percent-clipped=8.0
+2023-03-05 19:05:52,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=476181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:05:59,414 INFO [train.py:968] (0/2) Epoch 11, batch 20650, giga_loss[loss=0.2777, simple_loss=0.3585, pruned_loss=0.09845, over 28870.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3569, pruned_loss=0.1056, over 5682444.05 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3522, pruned_loss=0.09744, over 5754325.58 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3563, pruned_loss=0.1059, over 5685767.09 frames. ], batch size: 174, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:06:31,324 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 19:06:35,040 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=476227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:06:43,284 INFO [train.py:968] (0/2) Epoch 11, batch 20700, giga_loss[loss=0.2895, simple_loss=0.358, pruned_loss=0.1105, over 28693.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3579, pruned_loss=0.1061, over 5692267.70 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3522, pruned_loss=0.0973, over 5758964.20 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3576, pruned_loss=0.1068, over 5688458.49 frames. ], batch size: 92, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:07:09,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=476265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:07:10,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.275e+02 1.184e+03 1.516e+03 1.914e+03 3.334e+03, threshold=3.033e+03, percent-clipped=8.0
+2023-03-05 19:07:13,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2949, 1.8357, 1.5756, 1.4470], device='cuda:0'), covar=tensor([0.0757, 0.0308, 0.0285, 0.0793], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:0')
+2023-03-05 19:07:13,713 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=476270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:07:31,926 INFO [train.py:968] (0/2) Epoch 11, batch 20750, giga_loss[loss=0.2944, simple_loss=0.3702, pruned_loss=0.1093, over 28551.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3585, pruned_loss=0.1066, over 5700684.24 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3524, pruned_loss=0.09733, over 5755841.38 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3581, pruned_loss=0.1071, over 5700460.63 frames. ], batch size: 307, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:08:05,456 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=476324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:08:08,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=476327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:08:16,748 INFO [train.py:968] (0/2) Epoch 11, batch 20800, giga_loss[loss=0.2974, simple_loss=0.3629, pruned_loss=0.116, over 28920.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3599, pruned_loss=0.1085, over 5701097.84 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3523, pruned_loss=0.09729, over 5757496.98 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3598, pruned_loss=0.109, over 5698896.50 frames. ], batch size: 213, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:08:31,732 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=476356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:08:39,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.901e+02 1.238e+03 1.541e+03 2.217e+03 4.263e+03, threshold=3.081e+03, percent-clipped=9.0
+2023-03-05 19:08:40,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1756, 3.9971, 3.7443, 1.7740], device='cuda:0'), covar=tensor([0.0602, 0.0757, 0.0758, 0.2209], device='cuda:0'), in_proj_covar=tensor([0.1024, 0.0961, 0.0840, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 19:08:52,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3481, 1.3310, 1.2252, 1.4727], device='cuda:0'), covar=tensor([0.0784, 0.0328, 0.0319, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0057, 0.0051, 0.0088], device='cuda:0')
+2023-03-05 19:08:55,014 INFO [train.py:968] (0/2) Epoch 11, batch 20850, giga_loss[loss=0.296, simple_loss=0.3705, pruned_loss=0.1108, over 28892.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3603, pruned_loss=0.1084, over 5711373.13 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3522, pruned_loss=0.09716, over 5762276.69 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3605, pruned_loss=0.1094, over 5703516.64 frames. ], batch size: 199, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:09:10,227 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=476408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:09:12,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=476411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:09:36,322 INFO [train.py:968] (0/2) Epoch 11, batch 20900, giga_loss[loss=0.2678, simple_loss=0.3465, pruned_loss=0.09452, over 28614.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3597, pruned_loss=0.1071, over 5710559.27 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3521, pruned_loss=0.09712, over 5764779.05 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.36, pruned_loss=0.108, over 5701289.63 frames. ], batch size: 85, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:09:37,406 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=476440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:09:58,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.229e+02 1.085e+03 1.423e+03 2.129e+03 7.087e+03, threshold=2.846e+03, percent-clipped=7.0
+2023-03-05 19:10:16,042 INFO [train.py:968] (0/2) Epoch 11, batch 20950, libri_loss[loss=0.3269, simple_loss=0.392, pruned_loss=0.1309, over 20571.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3613, pruned_loss=0.1073, over 5710209.76 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3524, pruned_loss=0.09728, over 5759545.88 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3615, pruned_loss=0.1081, over 5707164.05 frames. ], batch size: 186, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:10:25,606 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=476501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:10:54,530 INFO [train.py:968] (0/2) Epoch 11, batch 21000, libri_loss[loss=0.2865, simple_loss=0.3682, pruned_loss=0.1024, over 29110.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3615, pruned_loss=0.1072, over 5719794.95 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3525, pruned_loss=0.09734, over 5764385.51 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.362, pruned_loss=0.1082, over 5711161.82 frames. ], batch size: 101, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:10:54,535 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 19:10:58,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9607, 1.1877, 3.5417, 3.0912], device='cuda:0'), covar=tensor([0.1884, 0.2806, 0.0448, 0.0820], device='cuda:0'), in_proj_covar=tensor([0.0647, 0.0576, 0.0833, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 19:11:03,128 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2182, simple_loss=0.3234, pruned_loss=0.05645, over 944034.00 frames. 
+2023-03-05 19:11:03,128 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 19:11:24,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.209e+02 1.017e+03 1.212e+03 1.619e+03 4.956e+03, threshold=2.425e+03, percent-clipped=5.0
+2023-03-05 19:11:39,947 INFO [train.py:968] (0/2) Epoch 11, batch 21050, giga_loss[loss=0.2925, simple_loss=0.364, pruned_loss=0.1105, over 28809.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3589, pruned_loss=0.106, over 5706728.85 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3528, pruned_loss=0.09765, over 5755104.92 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3592, pruned_loss=0.1067, over 5706620.87 frames. ], batch size: 243, lr: 2.94e-03, grad_scale: 2.0
+2023-03-05 19:11:49,814 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=476602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:12:19,539 INFO [train.py:968] (0/2) Epoch 11, batch 21100, giga_loss[loss=0.2921, simple_loss=0.364, pruned_loss=0.1101, over 28693.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3555, pruned_loss=0.1043, over 5710506.17 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3525, pruned_loss=0.09761, over 5756744.78 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.356, pruned_loss=0.105, over 5707979.63 frames. ], batch size: 99, lr: 2.94e-03, grad_scale: 2.0
+2023-03-05 19:12:26,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=476645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:12:32,592 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-05 19:12:44,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.094e+02 9.408e+02 1.202e+03 1.713e+03 4.550e+03, threshold=2.403e+03, percent-clipped=10.0
+2023-03-05 19:13:00,405 INFO [train.py:968] (0/2) Epoch 11, batch 21150, giga_loss[loss=0.3224, simple_loss=0.3949, pruned_loss=0.1249, over 28933.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3551, pruned_loss=0.1041, over 5716507.32 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3528, pruned_loss=0.09774, over 5758565.10 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3553, pruned_loss=0.1047, over 5712166.41 frames. ], batch size: 145, lr: 2.94e-03, grad_scale: 2.0
+2023-03-05 19:13:42,911 INFO [train.py:968] (0/2) Epoch 11, batch 21200, giga_loss[loss=0.2737, simple_loss=0.3525, pruned_loss=0.09749, over 28830.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3551, pruned_loss=0.1044, over 5717092.35 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3526, pruned_loss=0.09753, over 5760879.39 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3555, pruned_loss=0.1051, over 5710584.69 frames. ], batch size: 199, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:13:49,212 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=476745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:13:51,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=476748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:14:10,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.456e+02 1.015e+03 1.251e+03 1.595e+03 3.593e+03, threshold=2.502e+03, percent-clipped=6.0
+2023-03-05 19:14:12,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9474, 1.0585, 1.0813, 0.9175], device='cuda:0'), covar=tensor([0.1651, 0.1856, 0.0957, 0.1328], device='cuda:0'), in_proj_covar=tensor([0.1689, 0.1580, 0.1556, 0.1663], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 19:14:16,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=476777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:14:25,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=476788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:14:26,169 INFO [train.py:968] (0/2) Epoch 11, batch 21250, giga_loss[loss=0.2568, simple_loss=0.3424, pruned_loss=0.08556, over 28952.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3551, pruned_loss=0.1043, over 5713946.51 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3525, pruned_loss=0.09748, over 5762881.61 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3554, pruned_loss=0.105, over 5706202.45 frames. ], batch size: 213, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:14:27,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=476791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:14:48,772 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=476820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:15:04,825 INFO [train.py:968] (0/2) Epoch 11, batch 21300, giga_loss[loss=0.2855, simple_loss=0.3609, pruned_loss=0.105, over 28565.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3546, pruned_loss=0.1034, over 5723542.88 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3526, pruned_loss=0.09781, over 5769389.58 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3549, pruned_loss=0.104, over 5709839.09 frames. ], batch size: 336, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:15:28,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.055e+02 1.031e+03 1.256e+03 1.921e+03 8.433e+03, threshold=2.511e+03, percent-clipped=16.0
+2023-03-05 19:15:34,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=476876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:15:44,293 INFO [train.py:968] (0/2) Epoch 11, batch 21350, giga_loss[loss=0.2451, simple_loss=0.324, pruned_loss=0.08314, over 29038.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3548, pruned_loss=0.1034, over 5720355.47 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3533, pruned_loss=0.09845, over 5772399.08 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3544, pruned_loss=0.1035, over 5705077.89 frames. ], batch size: 136, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:16:22,598 INFO [train.py:968] (0/2) Epoch 11, batch 21400, giga_loss[loss=0.2568, simple_loss=0.3322, pruned_loss=0.09066, over 28414.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3546, pruned_loss=0.1041, over 5715324.09 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3535, pruned_loss=0.09868, over 5773346.57 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3543, pruned_loss=0.104, over 5701775.47 frames. ], batch size: 71, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:16:36,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6911, 1.9186, 1.9439, 1.5017], device='cuda:0'), covar=tensor([0.1672, 0.2133, 0.1306, 0.1516], device='cuda:0'), in_proj_covar=tensor([0.0827, 0.0694, 0.0867, 0.0770], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 19:16:47,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.486e+02 9.615e+02 1.248e+03 1.721e+03 5.492e+03, threshold=2.497e+03, percent-clipped=9.0
+2023-03-05 19:16:47,911 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=476970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:17:01,249 INFO [train.py:968] (0/2) Epoch 11, batch 21450, giga_loss[loss=0.262, simple_loss=0.3362, pruned_loss=0.09396, over 29008.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3532, pruned_loss=0.1038, over 5713950.17 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.354, pruned_loss=0.09939, over 5776235.82 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3526, pruned_loss=0.1033, over 5697656.60 frames. ], batch size: 164, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:17:24,951 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=477019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:17:27,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=477022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:17:30,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5492, 1.8513, 1.5556, 1.3766], device='cuda:0'), covar=tensor([0.1855, 0.1435, 0.1147, 0.1422], device='cuda:0'), in_proj_covar=tensor([0.1679, 0.1574, 0.1546, 0.1652], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 19:17:30,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4298, 1.8494, 1.7089, 1.2781], device='cuda:0'), covar=tensor([0.1636, 0.2374, 0.1440, 0.1651], device='cuda:0'), in_proj_covar=tensor([0.0827, 0.0696, 0.0869, 0.0772], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 19:17:40,083 INFO [train.py:968] (0/2) Epoch 11, batch 21500, giga_loss[loss=0.2746, simple_loss=0.3493, pruned_loss=0.09993, over 28611.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3504, pruned_loss=0.1026, over 5714588.56 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3539, pruned_loss=0.0997, over 5780334.64 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3499, pruned_loss=0.102, over 5696083.16 frames. ], batch size: 336, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:17:50,520 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=477051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:18:01,743 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 19:18:03,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.199e+02 1.073e+03 1.334e+03 1.841e+03 4.654e+03, threshold=2.669e+03, percent-clipped=12.0
+2023-03-05 19:18:04,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-05 19:18:21,410 INFO [train.py:968] (0/2) Epoch 11, batch 21550, giga_loss[loss=0.2881, simple_loss=0.3546, pruned_loss=0.1108, over 28970.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3482, pruned_loss=0.1017, over 5705383.09 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3538, pruned_loss=0.09973, over 5779473.96 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3478, pruned_loss=0.1012, over 5691351.17 frames. ], batch size: 164, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:18:42,475 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-05 19:18:57,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2120, 0.7879, 0.8489, 1.3677], device='cuda:0'), covar=tensor([0.0764, 0.0358, 0.0362, 0.0829], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:0')
+2023-03-05 19:19:00,162 INFO [train.py:968] (0/2) Epoch 11, batch 21600, giga_loss[loss=0.2628, simple_loss=0.3371, pruned_loss=0.09425, over 28922.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3485, pruned_loss=0.1025, over 5709230.70 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3543, pruned_loss=0.1002, over 5782364.91 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3477, pruned_loss=0.1018, over 5693918.94 frames. ], batch size: 227, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:19:10,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1696, 3.9810, 3.7697, 1.9024], device='cuda:0'), covar=tensor([0.0504, 0.0648, 0.0689, 0.2080], device='cuda:0'), in_proj_covar=tensor([0.1022, 0.0962, 0.0843, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 19:19:24,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.773e+02 1.127e+03 1.481e+03 2.456e+03 5.917e+03, threshold=2.963e+03, percent-clipped=20.0
+2023-03-05 19:19:28,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5298, 1.5311, 1.6121, 1.5517], device='cuda:0'), covar=tensor([0.1851, 0.1624, 0.1226, 0.1459], device='cuda:0'), in_proj_covar=tensor([0.1684, 0.1576, 0.1549, 0.1658], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 19:19:36,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=477182.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 19:19:40,777 INFO [train.py:968] (0/2) Epoch 11, batch 21650, giga_loss[loss=0.264, simple_loss=0.3329, pruned_loss=0.09752, over 28888.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3458, pruned_loss=0.1013, over 5708268.56 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3545, pruned_loss=0.1005, over 5783921.26 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3448, pruned_loss=0.1005, over 5693484.25 frames. ], batch size: 199, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:20:20,839 INFO [train.py:968] (0/2) Epoch 11, batch 21700, giga_loss[loss=0.2607, simple_loss=0.3409, pruned_loss=0.09028, over 28914.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3437, pruned_loss=0.1006, over 5711341.19 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3553, pruned_loss=0.1011, over 5784306.57 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3419, pruned_loss=0.09939, over 5696924.89 frames. ], batch size: 145, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:20:45,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.065e+02 1.070e+03 1.356e+03 1.797e+03 4.832e+03, threshold=2.713e+03, percent-clipped=6.0
+2023-03-05 19:21:00,952 INFO [train.py:968] (0/2) Epoch 11, batch 21750, giga_loss[loss=0.2686, simple_loss=0.3359, pruned_loss=0.1006, over 28603.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3417, pruned_loss=0.09948, over 5712855.70 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3557, pruned_loss=0.1014, over 5778385.74 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3397, pruned_loss=0.09828, over 5705203.42 frames. ], batch size: 262, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:21:41,042 INFO [train.py:968] (0/2) Epoch 11, batch 21800, giga_loss[loss=0.2812, simple_loss=0.3567, pruned_loss=0.1028, over 28730.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3409, pruned_loss=0.09935, over 5707673.19 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3558, pruned_loss=0.1016, over 5772048.42 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3389, pruned_loss=0.09811, over 5705882.32 frames. ], batch size: 262, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:21:43,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3550, 1.9554, 1.3811, 0.4834], device='cuda:0'), covar=tensor([0.4002, 0.2003, 0.2587, 0.4430], device='cuda:0'), in_proj_covar=tensor([0.1541, 0.1458, 0.1482, 0.1262], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 19:21:44,768 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-05 19:21:45,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=477345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:22:00,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2803, 3.4536, 1.4785, 1.4037], device='cuda:0'), covar=tensor([0.0908, 0.0263, 0.0924, 0.1314], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0498, 0.0333, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 19:22:05,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.886e+02 1.036e+03 1.523e+03 2.344e+03 6.469e+03, threshold=3.047e+03, percent-clipped=13.0
+2023-03-05 19:22:21,646 INFO [train.py:968] (0/2) Epoch 11, batch 21850, giga_loss[loss=0.2914, simple_loss=0.3711, pruned_loss=0.1058, over 28867.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3422, pruned_loss=0.09961, over 5708777.20 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3562, pruned_loss=0.1022, over 5771491.05 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3399, pruned_loss=0.09808, over 5706200.24 frames. ], batch size: 186, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:23:03,065 INFO [train.py:968] (0/2) Epoch 11, batch 21900, giga_loss[loss=0.2516, simple_loss=0.3191, pruned_loss=0.09204, over 28185.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3459, pruned_loss=0.1016, over 5703715.38 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3564, pruned_loss=0.1027, over 5774393.73 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3435, pruned_loss=0.09989, over 5697542.58 frames. ], batch size: 77, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:23:29,595 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.316e+02 1.074e+03 1.341e+03 1.878e+03 9.785e+03, threshold=2.682e+03, percent-clipped=6.0
+2023-03-05 19:23:44,430 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=477488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:23:44,846 INFO [train.py:968] (0/2) Epoch 11, batch 21950, giga_loss[loss=0.3344, simple_loss=0.3989, pruned_loss=0.1349, over 27621.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3504, pruned_loss=0.1034, over 5690660.74 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3573, pruned_loss=0.1034, over 5766363.41 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3475, pruned_loss=0.1013, over 5691875.09 frames. ], batch size: 472, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:23:47,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=477491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:24:05,771 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=477513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:24:12,836 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=477520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:24:26,716 INFO [train.py:968] (0/2) Epoch 11, batch 22000, giga_loss[loss=0.2932, simple_loss=0.3736, pruned_loss=0.1064, over 28708.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3509, pruned_loss=0.1028, over 5702410.73 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3573, pruned_loss=0.1037, over 5768465.00 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3485, pruned_loss=0.1009, over 5700421.18 frames. ], batch size: 243, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:24:37,452 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4115, 4.2222, 3.9973, 1.8047], device='cuda:0'), covar=tensor([0.0470, 0.0643, 0.0671, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.1021, 0.0959, 0.0844, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 19:24:41,962 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=477557.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 19:24:52,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.107e+02 1.065e+03 1.293e+03 1.779e+03 1.380e+04, threshold=2.586e+03, percent-clipped=11.0
+2023-03-05 19:25:05,816 INFO [train.py:968] (0/2) Epoch 11, batch 22050, giga_loss[loss=0.2304, simple_loss=0.3228, pruned_loss=0.06902, over 28966.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3497, pruned_loss=0.1016, over 5708178.43 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3572, pruned_loss=0.1039, over 5770384.42 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3475, pruned_loss=0.09976, over 5702175.09 frames. ], batch size: 164, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:25:12,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7366, 5.0619, 1.9791, 2.1389], device='cuda:0'), covar=tensor([0.0823, 0.0281, 0.0785, 0.1073], device='cuda:0'), in_proj_covar=tensor([0.0351, 0.0500, 0.0333, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 19:25:46,895 INFO [train.py:968] (0/2) Epoch 11, batch 22100, giga_loss[loss=0.3439, simple_loss=0.3972, pruned_loss=0.1452, over 26802.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3496, pruned_loss=0.1018, over 5699708.16 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3575, pruned_loss=0.1042, over 5765956.66 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3474, pruned_loss=0.09999, over 5695501.65 frames. ], batch size: 555, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:26:11,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.845e+02 1.114e+03 1.371e+03 1.867e+03 4.685e+03, threshold=2.741e+03, percent-clipped=7.0
+2023-03-05 19:26:25,714 INFO [train.py:968] (0/2) Epoch 11, batch 22150, libri_loss[loss=0.2654, simple_loss=0.3405, pruned_loss=0.09517, over 29559.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3505, pruned_loss=0.1026, over 5695169.21 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3581, pruned_loss=0.1048, over 5758824.65 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3479, pruned_loss=0.1006, over 5695519.43 frames. ], batch size: 75, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:26:35,001 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=477700.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 19:26:38,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=477703.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 19:27:02,554 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=477732.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 19:27:10,155 INFO [train.py:968] (0/2) Epoch 11, batch 22200, giga_loss[loss=0.2909, simple_loss=0.3685, pruned_loss=0.1067, over 29011.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3518, pruned_loss=0.1042, over 5687570.62 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3581, pruned_loss=0.1048, over 5751355.12 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3497, pruned_loss=0.1025, over 5693014.05 frames. ], batch size: 136, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:27:36,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.315e+02 1.217e+03 1.472e+03 1.889e+03 3.729e+03, threshold=2.944e+03, percent-clipped=6.0
+2023-03-05 19:27:51,219 INFO [train.py:968] (0/2) Epoch 11, batch 22250, giga_loss[loss=0.3146, simple_loss=0.3864, pruned_loss=0.1215, over 28954.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3545, pruned_loss=0.1054, over 5697150.74 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3582, pruned_loss=0.1049, over 5751876.58 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3526, pruned_loss=0.1039, over 5700003.74 frames. ], batch size: 227, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:28:32,149 INFO [train.py:968] (0/2) Epoch 11, batch 22300, giga_loss[loss=0.2948, simple_loss=0.368, pruned_loss=0.1108, over 28813.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3568, pruned_loss=0.1065, over 5701153.07 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3583, pruned_loss=0.1051, over 5753841.31 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3552, pruned_loss=0.1052, over 5700682.97 frames. ], batch size: 99, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:28:43,868 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=477852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:28:57,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.418e+02 1.186e+03 1.421e+03 1.842e+03 6.402e+03, threshold=2.841e+03, percent-clipped=6.0
+2023-03-05 19:29:10,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=477888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:29:11,327 INFO [train.py:968] (0/2) Epoch 11, batch 22350, giga_loss[loss=0.2756, simple_loss=0.3484, pruned_loss=0.1014, over 28685.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3593, pruned_loss=0.1081, over 5712655.84 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3593, pruned_loss=0.1059, over 5757496.97 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.357, pruned_loss=0.1064, over 5707747.15 frames. ], batch size: 71, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:29:38,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=477923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:29:49,259 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=477937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:29:51,348 INFO [train.py:968] (0/2) Epoch 11, batch 22400, giga_loss[loss=0.3052, simple_loss=0.3758, pruned_loss=0.1173, over 28886.00 frames. ], tot_loss[loss=0.287, simple_loss=0.359, pruned_loss=0.1075, over 5712687.67 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3599, pruned_loss=0.1065, over 5749674.30 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3567, pruned_loss=0.1056, over 5715948.06 frames. ], batch size: 199, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:30:01,912 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-03-05 19:30:20,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.783e+02 1.246e+03 1.420e+03 1.924e+03 5.464e+03, threshold=2.840e+03, percent-clipped=11.0
+2023-03-05 19:30:35,750 INFO [train.py:968] (0/2) Epoch 11, batch 22450, giga_loss[loss=0.3017, simple_loss=0.3767, pruned_loss=0.1134, over 29014.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3593, pruned_loss=0.1077, over 5709271.98 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3598, pruned_loss=0.1065, over 5751375.11 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3575, pruned_loss=0.1062, over 5709833.22 frames. ], batch size: 164, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:30:45,205 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-478000.pt
+2023-03-05 19:30:56,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4708, 1.5945, 1.6896, 1.2942], device='cuda:0'), covar=tensor([0.1563, 0.2116, 0.1313, 0.1492], device='cuda:0'), in_proj_covar=tensor([0.0824, 0.0690, 0.0862, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 19:31:10,002 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=478031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:31:13,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=478034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:31:16,446 INFO [train.py:968] (0/2) Epoch 11, batch 22500, giga_loss[loss=0.279, simple_loss=0.3403, pruned_loss=0.1089, over 24011.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3573, pruned_loss=0.1066, over 5705630.77 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3601, pruned_loss=0.1067, over 5744778.99 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3556, pruned_loss=0.1052, over 5710504.53 frames. ], batch size: 705, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:31:21,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2139, 1.2826, 4.2870, 3.4078], device='cuda:0'), covar=tensor([0.1694, 0.2637, 0.0358, 0.0918], device='cuda:0'), in_proj_covar=tensor([0.0653, 0.0581, 0.0848, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 19:31:35,312 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=478063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:31:41,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.788e+02 1.221e+03 1.738e+03 2.499e+03 6.217e+03, threshold=3.475e+03, percent-clipped=14.0
+2023-03-05 19:32:00,280 INFO [train.py:968] (0/2) Epoch 11, batch 22550, giga_loss[loss=0.2614, simple_loss=0.3373, pruned_loss=0.09278, over 28785.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3565, pruned_loss=0.1067, over 5708267.71 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.361, pruned_loss=0.1074, over 5742669.59 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3544, pruned_loss=0.105, over 5713230.74 frames. ], batch size: 262, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:32:11,952 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=478103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:32:40,496 INFO [train.py:968] (0/2) Epoch 11, batch 22600, libri_loss[loss=0.3346, simple_loss=0.3959, pruned_loss=0.1367, over 29541.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3537, pruned_loss=0.1052, over 5712537.65 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.362, pruned_loss=0.1082, over 5746942.77 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.351, pruned_loss=0.1032, over 5712026.74 frames. ], batch size: 84, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:33:04,159 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.779e+02 1.028e+03 1.345e+03 1.826e+03 8.644e+03, threshold=2.690e+03, percent-clipped=6.0
+2023-03-05 19:33:17,313 INFO [train.py:968] (0/2) Epoch 11, batch 22650, giga_loss[loss=0.284, simple_loss=0.3643, pruned_loss=0.1019, over 27912.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3523, pruned_loss=0.1041, over 5708970.84 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3625, pruned_loss=0.1087, over 5741893.52 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3493, pruned_loss=0.1018, over 5712664.18 frames. ], batch size: 412, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:33:51,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=478227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:34:01,617 INFO [train.py:968] (0/2) Epoch 11, batch 22700, giga_loss[loss=0.2641, simple_loss=0.3438, pruned_loss=0.09218, over 28959.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3533, pruned_loss=0.1035, over 5709511.03 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3633, pruned_loss=0.1095, over 5744345.02 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3501, pruned_loss=0.1009, over 5709801.43 frames. ], batch size: 106, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:34:10,203 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-05 19:34:29,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.931e+02 1.073e+03 1.377e+03 1.842e+03 5.230e+03, threshold=2.753e+03, percent-clipped=10.0
+2023-03-05 19:34:42,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4420, 1.6426, 1.3122, 1.6068], device='cuda:0'), covar=tensor([0.2368, 0.2347, 0.2727, 0.2238], device='cuda:0'), in_proj_covar=tensor([0.1300, 0.0960, 0.1146, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 19:34:44,740 INFO [train.py:968] (0/2) Epoch 11, batch 22750, giga_loss[loss=0.252, simple_loss=0.3271, pruned_loss=0.08841, over 29069.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3546, pruned_loss=0.1037, over 5714914.98 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3633, pruned_loss=0.1096, over 5745929.45 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3519, pruned_loss=0.1015, over 5713453.51 frames. ], batch size: 136, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:34:51,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=478298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:35:01,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=478312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:35:22,989 INFO [train.py:968] (0/2) Epoch 11, batch 22800, giga_loss[loss=0.2571, simple_loss=0.3318, pruned_loss=0.09121, over 29017.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3531, pruned_loss=0.1035, over 5724302.91 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3634, pruned_loss=0.11, over 5749271.16 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3508, pruned_loss=0.1013, over 5719454.53 frames. ], batch size: 164, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:35:49,116 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=478370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:35:50,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.684e+02 1.131e+03 1.416e+03 2.089e+03 9.626e+03, threshold=2.832e+03, percent-clipped=14.0
+2023-03-05 19:35:52,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=478373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:36:04,394 INFO [train.py:968] (0/2) Epoch 11, batch 22850, giga_loss[loss=0.2815, simple_loss=0.344, pruned_loss=0.1095, over 29010.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3524, pruned_loss=0.1048, over 5713157.36 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3639, pruned_loss=0.1105, over 5739986.29 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3497, pruned_loss=0.1024, over 5717786.46 frames. ], batch size: 128, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:36:08,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6259, 1.9991, 1.7327, 1.5419], device='cuda:0'), covar=tensor([0.2617, 0.1884, 0.1964, 0.2162], device='cuda:0'), in_proj_covar=tensor([0.1683, 0.1570, 0.1547, 0.1657], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 19:36:12,874 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=478402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:36:42,845 INFO [train.py:968] (0/2) Epoch 11, batch 22900, giga_loss[loss=0.2747, simple_loss=0.3374, pruned_loss=0.106, over 28940.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3513, pruned_loss=0.1057, over 5712138.08 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3645, pruned_loss=0.111, over 5739538.53 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3486, pruned_loss=0.1033, over 5715984.54 frames. ], batch size: 186, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:36:46,239 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=478441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:36:47,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1615, 3.9654, 3.7220, 1.8371], device='cuda:0'), covar=tensor([0.0631, 0.0789, 0.0781, 0.2224], device='cuda:0'), in_proj_covar=tensor([0.1027, 0.0961, 0.0845, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 19:36:48,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=478444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:36:55,360 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=478455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:36:57,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=478458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:37:04,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3152, 1.9581, 1.4445, 0.5287], device='cuda:0'), covar=tensor([0.3357, 0.1755, 0.3088, 0.4596], device='cuda:0'), in_proj_covar=tensor([0.1537, 0.1455, 0.1488, 0.1259], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 19:37:10,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.937e+02 1.057e+03 1.298e+03 1.670e+03 6.223e+03, threshold=2.596e+03, percent-clipped=4.0
+2023-03-05 19:37:10,446 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=478473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:37:15,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=478478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:37:21,166 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=478487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:37:23,435 INFO [train.py:968] (0/2) Epoch 11, batch 22950, giga_loss[loss=0.3393, simple_loss=0.3849, pruned_loss=0.1469, over 26792.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3495, pruned_loss=0.1054, over 5708676.98 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3649, pruned_loss=0.1113, over 5733685.36 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3467, pruned_loss=0.1032, over 5717298.23 frames. ], batch size: 555, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:37:24,241 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=478490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:37:41,599 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 19:38:01,112 INFO [train.py:968] (0/2) Epoch 11, batch 23000, libri_loss[loss=0.2843, simple_loss=0.3592, pruned_loss=0.1047, over 29481.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3484, pruned_loss=0.1051, over 5709646.58 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3655, pruned_loss=0.1121, over 5734784.51 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3449, pruned_loss=0.1023, over 5714622.24 frames. ], batch size: 85, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:38:26,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.106e+02 1.165e+03 1.595e+03 2.522e+03 7.847e+03, threshold=3.190e+03, percent-clipped=21.0
+2023-03-05 19:38:39,355 INFO [train.py:968] (0/2) Epoch 11, batch 23050, giga_loss[loss=0.3463, simple_loss=0.3877, pruned_loss=0.1524, over 26592.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3445, pruned_loss=0.103, over 5708100.88 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3657, pruned_loss=0.1122, over 5729127.28 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3412, pruned_loss=0.1004, over 5716452.61 frames. ], batch size: 555, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:39:05,011 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=478621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:39:06,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5079, 2.8584, 1.5617, 1.7129], device='cuda:0'), covar=tensor([0.0702, 0.0278, 0.0709, 0.0989], device='cuda:0'), in_proj_covar=tensor([0.0352, 0.0503, 0.0334, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:0')
+2023-03-05 19:39:08,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=478624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:39:20,468 INFO [train.py:968] (0/2) Epoch 11, batch 23100, giga_loss[loss=0.2299, simple_loss=0.2979, pruned_loss=0.08091, over 28449.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3398, pruned_loss=0.1008, over 5708035.92 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3657, pruned_loss=0.1126, over 5729752.87 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3368, pruned_loss=0.0983, over 5713879.40 frames. ], batch size: 85, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:39:31,452 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=478653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:39:44,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.552e+02 1.078e+03 1.305e+03 1.711e+03 4.173e+03, threshold=2.610e+03, percent-clipped=3.0
+2023-03-05 19:39:58,974 INFO [train.py:968] (0/2) Epoch 11, batch 23150, giga_loss[loss=0.3145, simple_loss=0.3778, pruned_loss=0.1256, over 28736.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3373, pruned_loss=0.09924, over 5701741.32 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3661, pruned_loss=0.1129, over 5714128.72 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3341, pruned_loss=0.09677, over 5719836.52 frames. ], batch size: 262, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:39:59,990 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3988, 1.5975, 1.5905, 1.4731], device='cuda:0'), covar=tensor([0.1537, 0.1779, 0.2061, 0.1783], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0725, 0.0669, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 19:40:22,837 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-05 19:40:32,160 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4932, 1.7760, 1.7950, 1.3220], device='cuda:0'), covar=tensor([0.1672, 0.2242, 0.1340, 0.1526], device='cuda:0'), in_proj_covar=tensor([0.0825, 0.0696, 0.0864, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 19:40:40,325 INFO [train.py:968] (0/2) Epoch 11, batch 23200, giga_loss[loss=0.2647, simple_loss=0.345, pruned_loss=0.09217, over 28727.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3395, pruned_loss=0.09979, over 5695338.10 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.366, pruned_loss=0.1131, over 5710765.26 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3367, pruned_loss=0.09748, over 5712127.37 frames. ], batch size: 284, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:40:58,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9434, 3.0635, 2.0873, 0.8518], device='cuda:0'), covar=tensor([0.5594, 0.1983, 0.2894, 0.5343], device='cuda:0'), in_proj_covar=tensor([0.1537, 0.1456, 0.1486, 0.1257], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 19:41:08,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.090e+02 1.101e+03 1.394e+03 1.756e+03 3.434e+03, threshold=2.788e+03, percent-clipped=2.0
+2023-03-05 19:41:16,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8994, 2.1393, 2.1855, 1.7028], device='cuda:0'), covar=tensor([0.1669, 0.1977, 0.1290, 0.1464], device='cuda:0'), in_proj_covar=tensor([0.0823, 0.0693, 0.0862, 0.0766], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 19:41:22,480 INFO [train.py:968] (0/2) Epoch 11, batch 23250, libri_loss[loss=0.3209, simple_loss=0.3933, pruned_loss=0.1243, over 29773.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3434, pruned_loss=0.1015, over 5690225.12 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3663, pruned_loss=0.1134, over 5706017.36 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3403, pruned_loss=0.09918, over 5707223.80 frames. ], batch size: 87, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:42:02,688 INFO [train.py:968] (0/2) Epoch 11, batch 23300, giga_loss[loss=0.265, simple_loss=0.3312, pruned_loss=0.09944, over 23815.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3475, pruned_loss=0.1031, over 5698441.61 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3672, pruned_loss=0.114, over 5711166.41 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3437, pruned_loss=0.1003, over 5706954.04 frames. ], batch size: 705, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:42:26,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=478865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:42:31,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.350e+02 1.150e+03 1.354e+03 1.851e+03 4.336e+03, threshold=2.707e+03, percent-clipped=6.0
+2023-03-05 19:42:43,799 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3171, 1.5044, 1.4825, 1.3589], device='cuda:0'), covar=tensor([0.1218, 0.1464, 0.1709, 0.1486], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0724, 0.0668, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 19:42:45,660 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=478887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:42:46,692 INFO [train.py:968] (0/2) Epoch 11, batch 23350, libri_loss[loss=0.2842, simple_loss=0.3561, pruned_loss=0.1062, over 29499.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3492, pruned_loss=0.1032, over 5707334.12 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3669, pruned_loss=0.1139, over 5714012.33 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3461, pruned_loss=0.1009, over 5711234.48 frames. ], batch size: 81, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:43:06,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1432, 0.8941, 0.9299, 1.3948], device='cuda:0'), covar=tensor([0.0757, 0.0355, 0.0344, 0.0807], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:0')
+2023-03-05 19:43:27,138 INFO [train.py:968] (0/2) Epoch 11, batch 23400, giga_loss[loss=0.2549, simple_loss=0.3301, pruned_loss=0.08991, over 28919.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3521, pruned_loss=0.1049, over 5714927.44 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3671, pruned_loss=0.1142, over 5716285.29 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.349, pruned_loss=0.1025, over 5716066.95 frames. ], batch size: 66, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:43:30,093 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=478942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:43:33,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0562, 2.0991, 1.5328, 1.8129], device='cuda:0'), covar=tensor([0.0555, 0.0411, 0.0817, 0.0789], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0442, 0.0501, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 19:43:55,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=478970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:43:59,311 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.205e+02 1.259e+03 1.620e+03 2.290e+03 4.493e+03, threshold=3.239e+03, percent-clipped=11.0
+2023-03-05 19:44:13,889 INFO [train.py:968] (0/2) Epoch 11, batch 23450, giga_loss[loss=0.3382, simple_loss=0.3918, pruned_loss=0.1423, over 29084.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3579, pruned_loss=0.11, over 5708918.01 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3674, pruned_loss=0.1146, over 5716718.14 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3547, pruned_loss=0.1075, over 5709090.90 frames. ], batch size: 155, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:44:28,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1756, 1.6827, 1.3043, 0.3897], device='cuda:0'), covar=tensor([0.2331, 0.1660, 0.1992, 0.3580], device='cuda:0'), in_proj_covar=tensor([0.1545, 0.1474, 0.1495, 0.1264], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 19:44:32,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5429, 2.3498, 1.7600, 0.7406], device='cuda:0'), covar=tensor([0.3020, 0.1965, 0.2659, 0.3531], device='cuda:0'), in_proj_covar=tensor([0.1543, 0.1471, 0.1493, 0.1263], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 19:44:33,380 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=479008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:44:36,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=479011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:45:00,994 INFO [train.py:968] (0/2) Epoch 11, batch 23500, giga_loss[loss=0.3457, simple_loss=0.406, pruned_loss=0.1427, over 28863.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3643, pruned_loss=0.1156, over 5709611.80 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3667, pruned_loss=0.1144, over 5723602.81 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3622, pruned_loss=0.1138, over 5702869.81 frames. ], batch size: 284, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:45:03,547 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=479040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:45:38,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.662e+03 2.367e+03 3.117e+03 9.585e+03, threshold=4.734e+03, percent-clipped=22.0
+2023-03-05 19:45:51,007 INFO [train.py:968] (0/2) Epoch 11, batch 23550, giga_loss[loss=0.325, simple_loss=0.3931, pruned_loss=0.1285, over 28607.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3711, pruned_loss=0.1208, over 5695154.34 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3666, pruned_loss=0.1146, over 5725552.57 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3696, pruned_loss=0.1192, over 5687419.24 frames. ], batch size: 307, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:45:57,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:46:27,393 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 19:46:41,379 INFO [train.py:968] (0/2) Epoch 11, batch 23600, giga_loss[loss=0.3188, simple_loss=0.3841, pruned_loss=0.1267, over 29000.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3782, pruned_loss=0.1269, over 5680057.47 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3671, pruned_loss=0.115, over 5718774.95 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3768, pruned_loss=0.1255, over 5679222.62 frames. ], batch size: 213, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:46:54,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3854, 1.6536, 1.2996, 1.3726], device='cuda:0'), covar=tensor([0.2091, 0.2081, 0.2332, 0.1849], device='cuda:0'), in_proj_covar=tensor([0.1292, 0.0957, 0.1146, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 19:47:16,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.679e+03 2.347e+03 3.114e+03 8.832e+03, threshold=4.695e+03, percent-clipped=7.0
+2023-03-05 19:47:31,841 INFO [train.py:968] (0/2) Epoch 11, batch 23650, giga_loss[loss=0.4241, simple_loss=0.4611, pruned_loss=0.1936, over 28251.00 frames. ], tot_loss[loss=0.325, simple_loss=0.385, pruned_loss=0.1326, over 5683189.70 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3672, pruned_loss=0.1151, over 5720505.53 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3838, pruned_loss=0.1315, over 5680691.49 frames. ], batch size: 368, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:47:51,993 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-05 19:48:24,101 INFO [train.py:968] (0/2) Epoch 11, batch 23700, giga_loss[loss=0.3418, simple_loss=0.3945, pruned_loss=0.1446, over 27990.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3889, pruned_loss=0.136, over 5672163.34 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3676, pruned_loss=0.1154, over 5719705.64 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.388, pruned_loss=0.1352, over 5670303.76 frames. ], batch size: 412, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:48:46,374 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=479262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:48:59,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.060e+02 1.537e+03 2.013e+03 2.734e+03 8.076e+03, threshold=4.026e+03, percent-clipped=4.0
+2023-03-05 19:49:13,184 INFO [train.py:968] (0/2) Epoch 11, batch 23750, giga_loss[loss=0.3219, simple_loss=0.3834, pruned_loss=0.1302, over 28843.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3899, pruned_loss=0.1377, over 5664000.25 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.368, pruned_loss=0.1159, over 5718450.33 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3891, pruned_loss=0.1369, over 5663123.23 frames. ], batch size: 186, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:49:40,770 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=479317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:50:05,561 INFO [train.py:968] (0/2) Epoch 11, batch 23800, giga_loss[loss=0.4971, simple_loss=0.498, pruned_loss=0.2481, over 26512.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3953, pruned_loss=0.144, over 5645931.82 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3678, pruned_loss=0.1159, over 5711796.33 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3951, pruned_loss=0.1437, over 5649676.27 frames. ], batch size: 555, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:50:13,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=479345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:50:41,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.635e+03 2.076e+03 3.030e+03 7.215e+03, threshold=4.152e+03, percent-clipped=6.0
+2023-03-05 19:50:56,484 INFO [train.py:968] (0/2) Epoch 11, batch 23850, libri_loss[loss=0.2852, simple_loss=0.356, pruned_loss=0.1071, over 29524.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.3994, pruned_loss=0.1485, over 5638925.88 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3682, pruned_loss=0.1165, over 5706898.02 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4, pruned_loss=0.1488, over 5644267.46 frames. ], batch size: 80, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:51:11,303 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=479405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:51:16,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=479408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:51:49,272 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=479437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:51:53,334 INFO [train.py:968] (0/2) Epoch 11, batch 23900, giga_loss[loss=0.3044, simple_loss=0.3743, pruned_loss=0.1173, over 28934.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.4019, pruned_loss=0.1493, over 5647230.19 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3684, pruned_loss=0.1167, over 5711031.32 frames. ], giga_tot_loss[loss=0.3515, simple_loss=0.4028, pruned_loss=0.1501, over 5646399.70 frames. ], batch size: 106, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:52:09,695 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2147, 1.1602, 3.6743, 3.1533], device='cuda:0'), covar=tensor([0.1605, 0.2452, 0.0450, 0.0748], device='cuda:0'), in_proj_covar=tensor([0.0664, 0.0589, 0.0862, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 19:52:14,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=479460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:52:18,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=479463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:52:25,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=479469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:52:30,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.164e+03 1.782e+03 2.325e+03 2.951e+03 1.090e+04, threshold=4.650e+03, percent-clipped=9.0
+2023-03-05 19:52:44,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=479488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:52:44,466 INFO [train.py:968] (0/2) Epoch 11, batch 23950, giga_loss[loss=0.3521, simple_loss=0.4047, pruned_loss=0.1498, over 29093.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4021, pruned_loss=0.1506, over 5650346.93 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3689, pruned_loss=0.1173, over 5717845.26 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4036, pruned_loss=0.1518, over 5641599.33 frames. ], batch size: 128, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:52:47,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=479491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:52:48,275 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=479492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:53:16,267 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=479520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:53:23,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1986, 5.0318, 4.7993, 2.3559], device='cuda:0'), covar=tensor([0.0414, 0.0570, 0.0621, 0.1811], device='cuda:0'), in_proj_covar=tensor([0.1044, 0.0984, 0.0865, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 19:53:31,653 INFO [train.py:968] (0/2) Epoch 11, batch 24000, giga_loss[loss=0.4893, simple_loss=0.4879, pruned_loss=0.2454, over 26719.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.3993, pruned_loss=0.1492, over 5656156.55 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.368, pruned_loss=0.117, over 5724252.57 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4026, pruned_loss=0.1517, over 5640282.55 frames. ], batch size: 555, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 19:53:31,657 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 19:53:38,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2687, 1.2537, 1.1323, 1.4594], device='cuda:0'), covar=tensor([0.0799, 0.0331, 0.0335, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:0')
+2023-03-05 19:53:38,927 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0021, 1.1668, 3.4208, 3.1276], device='cuda:0'), covar=tensor([0.1998, 0.2948, 0.0480, 0.0767], device='cuda:0'), in_proj_covar=tensor([0.0666, 0.0591, 0.0864, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 19:53:40,831 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2145, simple_loss=0.3204, pruned_loss=0.05423, over 944034.00 frames. 
+2023-03-05 19:53:40,832 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 19:53:47,155 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479547.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 19:54:15,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.778e+02 1.792e+03 2.344e+03 3.463e+03 8.527e+03, threshold=4.688e+03, percent-clipped=12.0
+2023-03-05 19:54:27,344 INFO [train.py:968] (0/2) Epoch 11, batch 24050, giga_loss[loss=0.3314, simple_loss=0.4017, pruned_loss=0.1306, over 28954.00 frames. ], tot_loss[loss=0.349, simple_loss=0.3997, pruned_loss=0.1492, over 5639911.91 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3683, pruned_loss=0.1175, over 5708758.03 frames. ], giga_tot_loss[loss=0.3529, simple_loss=0.4028, pruned_loss=0.1515, over 5638841.52 frames. ], batch size: 164, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:54:48,709 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=479612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:54:50,814 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=479615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:55:16,795 INFO [train.py:968] (0/2) Epoch 11, batch 24100, giga_loss[loss=0.3528, simple_loss=0.4061, pruned_loss=0.1497, over 28298.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.3993, pruned_loss=0.1478, over 5639910.51 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3686, pruned_loss=0.1178, over 5712755.65 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4023, pruned_loss=0.1501, over 5634216.71 frames. ], batch size: 368, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:55:21,348 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=479644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:55:29,391 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-05 19:55:50,989 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479675.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 19:55:51,344 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.838e+03 2.517e+03 3.397e+03 8.748e+03, threshold=5.034e+03, percent-clipped=13.0
+2023-03-05 19:56:06,631 INFO [train.py:968] (0/2) Epoch 11, batch 24150, giga_loss[loss=0.3516, simple_loss=0.3824, pruned_loss=0.1604, over 23570.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4007, pruned_loss=0.1486, over 5633795.46 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3694, pruned_loss=0.1185, over 5717378.40 frames. ], giga_tot_loss[loss=0.352, simple_loss=0.4031, pruned_loss=0.1505, over 5623637.40 frames. ], batch size: 710, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:56:51,564 INFO [train.py:968] (0/2) Epoch 11, batch 24200, giga_loss[loss=0.2878, simple_loss=0.3519, pruned_loss=0.1118, over 28652.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3988, pruned_loss=0.1469, over 5643553.74 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3696, pruned_loss=0.1189, over 5726372.74 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.402, pruned_loss=0.1495, over 5623500.33 frames. ], batch size: 92, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:57:12,062 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:57:27,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.547e+03 2.063e+03 2.856e+03 7.319e+03, threshold=4.125e+03, percent-clipped=3.0
+2023-03-05 19:57:39,181 INFO [train.py:968] (0/2) Epoch 11, batch 24250, libri_loss[loss=0.3131, simple_loss=0.3791, pruned_loss=0.1235, over 29372.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3941, pruned_loss=0.1417, over 5642666.87 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3701, pruned_loss=0.1196, over 5723383.03 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.3975, pruned_loss=0.1443, over 5624607.64 frames. ], batch size: 92, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:57:45,011 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-05 19:58:08,409 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:58:27,098 INFO [train.py:968] (0/2) Epoch 11, batch 24300, giga_loss[loss=0.3197, simple_loss=0.3792, pruned_loss=0.1301, over 27637.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3921, pruned_loss=0.1387, over 5651001.68 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3705, pruned_loss=0.1201, over 5726070.19 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3948, pruned_loss=0.1407, over 5632983.59 frames. ], batch size: 472, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:58:34,094 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:58:55,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6021, 1.7047, 1.6241, 1.4862], device='cuda:0'), covar=tensor([0.1373, 0.1801, 0.1921, 0.1834], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0726, 0.0665, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 19:58:58,837 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:59:00,621 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.473e+02 1.754e+03 2.158e+03 3.103e+03 1.039e+04, threshold=4.315e+03, percent-clipped=9.0
+2023-03-05 19:59:13,125 INFO [train.py:968] (0/2) Epoch 11, batch 24350, giga_loss[loss=0.304, simple_loss=0.3767, pruned_loss=0.1156, over 28900.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3888, pruned_loss=0.1356, over 5670559.98 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3703, pruned_loss=0.12, over 5730517.17 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3916, pruned_loss=0.1377, over 5650768.56 frames. ], batch size: 112, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:59:45,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=479922.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:00:03,225 INFO [train.py:968] (0/2) Epoch 11, batch 24400, giga_loss[loss=0.3466, simple_loss=0.3732, pruned_loss=0.1601, over 23628.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3864, pruned_loss=0.134, over 5666674.65 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3698, pruned_loss=0.1198, over 5734190.88 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3893, pruned_loss=0.1361, over 5646751.33 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 20:00:23,970 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 20:00:39,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.425e+02 1.621e+03 2.233e+03 3.420e+03 8.622e+03, threshold=4.466e+03, percent-clipped=13.0
+2023-03-05 20:00:51,083 INFO [train.py:968] (0/2) Epoch 11, batch 24450, giga_loss[loss=0.3078, simple_loss=0.3771, pruned_loss=0.1193, over 28227.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3849, pruned_loss=0.1327, over 5678392.64 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3698, pruned_loss=0.1197, over 5736179.93 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3873, pruned_loss=0.1345, over 5660513.41 frames. ], batch size: 77, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:00:58,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:01:02,533 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-480000.pt
+2023-03-05 20:01:46,554 INFO [train.py:968] (0/2) Epoch 11, batch 24500, libri_loss[loss=0.3047, simple_loss=0.3662, pruned_loss=0.1216, over 29589.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3862, pruned_loss=0.1337, over 5667205.87 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3702, pruned_loss=0.1199, over 5728722.94 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3883, pruned_loss=0.1353, over 5657981.06 frames. ], batch size: 75, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:01:58,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480050.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:02:12,380 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480065.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:02:15,968 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480068.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:02:23,372 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:02:24,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.469e+03 1.854e+03 2.423e+03 9.405e+03, threshold=3.707e+03, percent-clipped=4.0
+2023-03-05 20:02:36,028 INFO [train.py:968] (0/2) Epoch 11, batch 24550, giga_loss[loss=0.2936, simple_loss=0.3702, pruned_loss=0.1085, over 28859.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3833, pruned_loss=0.1308, over 5672716.33 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3703, pruned_loss=0.1201, over 5733281.87 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3854, pruned_loss=0.1323, over 5659242.78 frames. ], batch size: 112, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:02:43,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480097.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:03:18,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:03:25,789 INFO [train.py:968] (0/2) Epoch 11, batch 24600, giga_loss[loss=0.2919, simple_loss=0.3753, pruned_loss=0.1043, over 28971.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3827, pruned_loss=0.1278, over 5690012.12 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3701, pruned_loss=0.1203, over 5738296.60 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3848, pruned_loss=0.1291, over 5673270.93 frames. ], batch size: 164, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:04:05,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.395e+02 1.571e+03 1.991e+03 2.807e+03 1.376e+04, threshold=3.982e+03, percent-clipped=7.0
+2023-03-05 20:04:16,995 INFO [train.py:968] (0/2) Epoch 11, batch 24650, giga_loss[loss=0.3147, simple_loss=0.3817, pruned_loss=0.1238, over 28000.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3834, pruned_loss=0.1273, over 5672865.48 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3699, pruned_loss=0.1202, over 5742424.71 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3855, pruned_loss=0.1286, over 5654320.75 frames. ], batch size: 412, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:04:22,499 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480193.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:04:23,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:04:24,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480196.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:04:48,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:04:50,281 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-05 20:04:50,706 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480225.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:05:03,878 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:05:05,033 INFO [train.py:968] (0/2) Epoch 11, batch 24700, libri_loss[loss=0.2913, simple_loss=0.3597, pruned_loss=0.1114, over 29482.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3846, pruned_loss=0.1289, over 5677531.00 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3698, pruned_loss=0.1203, over 5745993.91 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.387, pruned_loss=0.1301, over 5657010.54 frames. ], batch size: 85, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:05:09,696 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-05 20:05:14,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:05:37,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:05:39,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.898e+02 1.792e+03 2.339e+03 3.217e+03 7.311e+03, threshold=4.679e+03, percent-clipped=10.0
+2023-03-05 20:05:42,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:05:50,580 INFO [train.py:968] (0/2) Epoch 11, batch 24750, giga_loss[loss=0.2687, simple_loss=0.3501, pruned_loss=0.09364, over 29014.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3858, pruned_loss=0.1304, over 5664220.48 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3702, pruned_loss=0.1207, over 5737983.02 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3878, pruned_loss=0.1314, over 5653031.38 frames. ], batch size: 128, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:06:08,888 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:06:38,523 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:06:39,501 INFO [train.py:968] (0/2) Epoch 11, batch 24800, giga_loss[loss=0.2987, simple_loss=0.364, pruned_loss=0.1166, over 28381.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3839, pruned_loss=0.1304, over 5661859.08 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3704, pruned_loss=0.121, over 5740134.43 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3855, pruned_loss=0.131, over 5649245.50 frames. ], batch size: 71, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 20:06:41,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:06:59,170 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-05 20:07:04,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:06,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:08,949 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:09,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:15,550 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.661e+03 2.189e+03 3.140e+03 6.651e+03, threshold=4.377e+03, percent-clipped=8.0
+2023-03-05 20:07:25,348 INFO [train.py:968] (0/2) Epoch 11, batch 24850, giga_loss[loss=0.3413, simple_loss=0.3889, pruned_loss=0.1469, over 28603.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3827, pruned_loss=0.1302, over 5669390.24 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5740961.76 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3839, pruned_loss=0.1307, over 5658457.99 frames. ], batch size: 336, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 20:07:27,011 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:30,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:31,143 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-05 20:07:32,650 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:56,140 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:08:10,098 INFO [train.py:968] (0/2) Epoch 11, batch 24900, libri_loss[loss=0.3103, simple_loss=0.3797, pruned_loss=0.1205, over 27758.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3816, pruned_loss=0.1295, over 5656630.00 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3707, pruned_loss=0.1214, over 5724850.01 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3828, pruned_loss=0.1298, over 5660855.83 frames. ], batch size: 116, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:08:11,929 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5754, 3.2790, 2.4353, 2.0571], device='cuda:0'), covar=tensor([0.1420, 0.0845, 0.1168, 0.1273], device='cuda:0'), in_proj_covar=tensor([0.1705, 0.1603, 0.1575, 0.1682], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 20:08:19,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:08:44,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.313e+02 1.533e+03 2.339e+03 3.652e+03 1.356e+04, threshold=4.679e+03, percent-clipped=16.0
+2023-03-05 20:08:53,169 INFO [train.py:968] (0/2) Epoch 11, batch 24950, giga_loss[loss=0.2653, simple_loss=0.3483, pruned_loss=0.09115, over 28883.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3807, pruned_loss=0.1277, over 5671018.28 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.371, pruned_loss=0.1217, over 5728765.96 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3817, pruned_loss=0.1279, over 5669676.66 frames. ], batch size: 227, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:09:19,544 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:09:21,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:09:41,570 INFO [train.py:968] (0/2) Epoch 11, batch 25000, giga_loss[loss=0.3342, simple_loss=0.3976, pruned_loss=0.1354, over 28700.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3814, pruned_loss=0.1283, over 5653701.50 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.372, pruned_loss=0.1225, over 5719731.99 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3816, pruned_loss=0.128, over 5658385.70 frames. ], batch size: 242, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:09:47,219 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:10:18,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.984e+02 1.504e+03 1.741e+03 2.339e+03 7.943e+03, threshold=3.482e+03, percent-clipped=2.0
+2023-03-05 20:10:23,114 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:10:30,266 INFO [train.py:968] (0/2) Epoch 11, batch 25050, giga_loss[loss=0.3776, simple_loss=0.4053, pruned_loss=0.175, over 26505.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3803, pruned_loss=0.127, over 5664045.11 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3721, pruned_loss=0.1224, over 5722241.80 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3805, pruned_loss=0.1269, over 5664592.58 frames. ], batch size: 555, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:10:34,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:10:37,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:10:51,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:11:06,638 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:11:22,081 INFO [train.py:968] (0/2) Epoch 11, batch 25100, giga_loss[loss=0.2903, simple_loss=0.3533, pruned_loss=0.1137, over 28506.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3775, pruned_loss=0.1254, over 5680550.08 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.372, pruned_loss=0.1223, over 5723770.30 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3777, pruned_loss=0.1254, over 5679216.85 frames. ], batch size: 85, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:11:48,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1326, 5.1202, 2.2584, 2.1911], device='cuda:0'), covar=tensor([0.0803, 0.0189, 0.0710, 0.1105], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0513, 0.0338, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 20:11:48,584 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-05 20:11:58,360 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7625, 1.0653, 2.8362, 2.6885], device='cuda:0'), covar=tensor([0.1611, 0.2356, 0.0573, 0.1327], device='cuda:0'), in_proj_covar=tensor([0.0662, 0.0588, 0.0859, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 20:12:00,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.751e+02 1.593e+03 2.094e+03 2.928e+03 1.008e+04, threshold=4.188e+03, percent-clipped=20.0
+2023-03-05 20:12:09,587 INFO [train.py:968] (0/2) Epoch 11, batch 25150, giga_loss[loss=0.3326, simple_loss=0.3926, pruned_loss=0.1363, over 28838.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3765, pruned_loss=0.1255, over 5665242.91 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3724, pruned_loss=0.1227, over 5713363.50 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3765, pruned_loss=0.1252, over 5672893.16 frames. ], batch size: 199, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:12:56,186 INFO [train.py:968] (0/2) Epoch 11, batch 25200, giga_loss[loss=0.3262, simple_loss=0.3862, pruned_loss=0.1331, over 28868.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3748, pruned_loss=0.125, over 5678672.51 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.372, pruned_loss=0.1226, over 5715986.88 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3752, pruned_loss=0.1249, over 5681850.70 frames. ], batch size: 199, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:12:59,291 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-05 20:13:11,464 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:13:14,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:13:21,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4159, 1.7078, 1.3660, 1.1555], device='cuda:0'), covar=tensor([0.1960, 0.1567, 0.1239, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.1692, 0.1593, 0.1569, 0.1680], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 20:13:33,121 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:13:34,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.373e+02 1.603e+03 1.968e+03 2.918e+03 9.973e+03, threshold=3.937e+03, percent-clipped=11.0
+2023-03-05 20:13:43,522 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:13:44,585 INFO [train.py:968] (0/2) Epoch 11, batch 25250, libri_loss[loss=0.3292, simple_loss=0.3981, pruned_loss=0.1302, over 29751.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.374, pruned_loss=0.1247, over 5680874.54 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3723, pruned_loss=0.1228, over 5720089.91 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3741, pruned_loss=0.1245, over 5678438.51 frames. ], batch size: 87, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:14:06,566 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:14:15,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2119, 1.4872, 1.2816, 0.9835], device='cuda:0'), covar=tensor([0.1911, 0.1740, 0.1172, 0.1747], device='cuda:0'), in_proj_covar=tensor([0.1696, 0.1592, 0.1568, 0.1679], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 20:14:30,185 INFO [train.py:968] (0/2) Epoch 11, batch 25300, giga_loss[loss=0.3119, simple_loss=0.3755, pruned_loss=0.1241, over 28837.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3733, pruned_loss=0.1248, over 5686313.99 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3725, pruned_loss=0.1229, over 5722192.17 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3732, pruned_loss=0.1245, over 5682157.95 frames. ], batch size: 186, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:15:06,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.748e+03 2.212e+03 3.270e+03 1.201e+04, threshold=4.424e+03, percent-clipped=16.0
+2023-03-05 20:15:15,938 INFO [train.py:968] (0/2) Epoch 11, batch 25350, giga_loss[loss=0.2816, simple_loss=0.3569, pruned_loss=0.1032, over 28936.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3738, pruned_loss=0.1256, over 5682517.26 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3724, pruned_loss=0.1229, over 5723070.28 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3739, pruned_loss=0.1255, over 5677322.74 frames. ], batch size: 145, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:15:40,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3491, 2.0579, 1.5214, 0.5506], device='cuda:0'), covar=tensor([0.4026, 0.2061, 0.2909, 0.4604], device='cuda:0'), in_proj_covar=tensor([0.1556, 0.1484, 0.1493, 0.1272], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 20:16:01,243 INFO [train.py:968] (0/2) Epoch 11, batch 25400, giga_loss[loss=0.2969, simple_loss=0.3765, pruned_loss=0.1086, over 28951.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3754, pruned_loss=0.1259, over 5682423.61 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3724, pruned_loss=0.123, over 5715463.78 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3755, pruned_loss=0.1258, over 5683937.91 frames. ], batch size: 145, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:16:16,883 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:16:18,680 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:16:35,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.951e+02 1.474e+03 1.968e+03 3.165e+03 8.207e+03, threshold=3.937e+03, percent-clipped=8.0
+2023-03-05 20:16:44,126 INFO [train.py:968] (0/2) Epoch 11, batch 25450, giga_loss[loss=0.3102, simple_loss=0.3759, pruned_loss=0.1222, over 28952.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3748, pruned_loss=0.1248, over 5680724.38 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3721, pruned_loss=0.1227, over 5718486.36 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3752, pruned_loss=0.1251, over 5678349.01 frames. ], batch size: 227, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:17:02,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4594, 1.2037, 4.7003, 3.4338], device='cuda:0'), covar=tensor([0.1658, 0.2766, 0.0380, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0665, 0.0591, 0.0864, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 20:17:34,066 INFO [train.py:968] (0/2) Epoch 11, batch 25500, libri_loss[loss=0.2753, simple_loss=0.3384, pruned_loss=0.1061, over 28120.00 frames. ], tot_loss[loss=0.312, simple_loss=0.375, pruned_loss=0.1245, over 5683299.47 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3719, pruned_loss=0.1227, over 5720854.96 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3755, pruned_loss=0.1248, over 5678407.48 frames. ], batch size: 62, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:17:55,320 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481063.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:18:09,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.727e+02 1.633e+03 2.373e+03 3.522e+03 8.337e+03, threshold=4.746e+03, percent-clipped=20.0
+2023-03-05 20:18:17,696 INFO [train.py:968] (0/2) Epoch 11, batch 25550, giga_loss[loss=0.3956, simple_loss=0.4406, pruned_loss=0.1753, over 28563.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3771, pruned_loss=0.1265, over 5687953.92 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3715, pruned_loss=0.1224, over 5725908.22 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.378, pruned_loss=0.1271, over 5678458.55 frames. ], batch size: 336, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:18:27,979 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:18:33,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:18:39,271 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:19:00,968 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:19:06,239 INFO [train.py:968] (0/2) Epoch 11, batch 25600, giga_loss[loss=0.2886, simple_loss=0.3608, pruned_loss=0.1083, over 29108.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3787, pruned_loss=0.1284, over 5692028.32 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3711, pruned_loss=0.1222, over 5729618.20 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.38, pruned_loss=0.1292, over 5680337.36 frames. ], batch size: 155, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:19:08,947 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-05 20:19:18,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:19:35,190 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 20:19:47,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.425e+02 1.630e+03 2.025e+03 2.705e+03 8.579e+03, threshold=4.050e+03, percent-clipped=4.0
+2023-03-05 20:19:57,538 INFO [train.py:968] (0/2) Epoch 11, batch 25650, giga_loss[loss=0.34, simple_loss=0.3723, pruned_loss=0.1538, over 23658.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3804, pruned_loss=0.1314, over 5677069.85 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3716, pruned_loss=0.1226, over 5723381.19 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3812, pruned_loss=0.1318, over 5672150.09 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:20:00,009 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:20:48,027 INFO [train.py:968] (0/2) Epoch 11, batch 25700, giga_loss[loss=0.3257, simple_loss=0.3832, pruned_loss=0.1341, over 28753.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3807, pruned_loss=0.1325, over 5686530.46 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3715, pruned_loss=0.1227, over 5729463.98 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3816, pruned_loss=0.1331, over 5675803.39 frames. ], batch size: 284, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:21:19,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.815e+02 1.607e+03 2.247e+03 3.379e+03 8.351e+03, threshold=4.494e+03, percent-clipped=19.0
+2023-03-05 20:21:26,118 INFO [train.py:968] (0/2) Epoch 11, batch 25750, giga_loss[loss=0.3398, simple_loss=0.3957, pruned_loss=0.1419, over 28951.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3798, pruned_loss=0.1317, over 5686556.62 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3714, pruned_loss=0.1228, over 5725630.55 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3813, pruned_loss=0.1327, over 5678808.61 frames. ], batch size: 213, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:21:33,095 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:21:35,567 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:21:46,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 20:22:04,602 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:22:10,564 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:22:11,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:22:12,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:22:15,325 INFO [train.py:968] (0/2) Epoch 11, batch 25800, giga_loss[loss=0.3729, simple_loss=0.4188, pruned_loss=0.1635, over 27649.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3785, pruned_loss=0.1311, over 5663478.21 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3715, pruned_loss=0.1229, over 5714865.99 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3797, pruned_loss=0.1318, over 5667024.51 frames. ], batch size: 472, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:22:24,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2338, 1.3925, 1.1571, 1.1783], device='cuda:0'), covar=tensor([0.1548, 0.1381, 0.1091, 0.1273], device='cuda:0'), in_proj_covar=tensor([0.1706, 0.1608, 0.1580, 0.1693], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 20:22:38,741 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:22:49,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.942e+02 1.648e+03 1.956e+03 2.698e+03 6.467e+03, threshold=3.912e+03, percent-clipped=1.0
+2023-03-05 20:22:58,694 INFO [train.py:968] (0/2) Epoch 11, batch 25850, giga_loss[loss=0.3001, simple_loss=0.372, pruned_loss=0.114, over 29112.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3776, pruned_loss=0.1285, over 5672418.13 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3708, pruned_loss=0.1225, over 5717782.30 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3792, pruned_loss=0.1295, over 5671904.47 frames. ], batch size: 128, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:23:11,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-05 20:23:47,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481438.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:23:48,276 INFO [train.py:968] (0/2) Epoch 11, batch 25900, giga_loss[loss=0.3053, simple_loss=0.3807, pruned_loss=0.1149, over 29043.00 frames. ], tot_loss[loss=0.315, simple_loss=0.376, pruned_loss=0.127, over 5652143.81 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1227, over 5709435.23 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3771, pruned_loss=0.1278, over 5658963.68 frames. ], batch size: 136, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:24:25,598 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:24:26,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.057e+02 1.528e+03 1.956e+03 2.860e+03 9.950e+03, threshold=3.912e+03, percent-clipped=13.0
+2023-03-05 20:24:27,720 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 20:24:28,200 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:24:32,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:24:36,768 INFO [train.py:968] (0/2) Epoch 11, batch 25950, giga_loss[loss=0.2846, simple_loss=0.3535, pruned_loss=0.1078, over 28753.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.374, pruned_loss=0.1262, over 5650023.00 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3717, pruned_loss=0.1233, over 5703639.80 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3744, pruned_loss=0.1263, over 5659361.48 frames. ], batch size: 242, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:24:45,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5308, 1.6045, 1.7680, 1.3866], device='cuda:0'), covar=tensor([0.1467, 0.2036, 0.1161, 0.1378], device='cuda:0'), in_proj_covar=tensor([0.0829, 0.0702, 0.0869, 0.0773], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 20:24:55,113 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:25:18,398 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-05 20:25:22,363 INFO [train.py:968] (0/2) Epoch 11, batch 26000, giga_loss[loss=0.2709, simple_loss=0.3458, pruned_loss=0.09806, over 29088.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3736, pruned_loss=0.1267, over 5655456.70 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3717, pruned_loss=0.1232, over 5707923.27 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3739, pruned_loss=0.1269, over 5658102.62 frames. ], batch size: 128, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 20:25:53,818 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:25:58,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3749, 1.6761, 1.2997, 1.2906], device='cuda:0'), covar=tensor([0.2364, 0.2222, 0.2581, 0.2010], device='cuda:0'), in_proj_covar=tensor([0.1302, 0.0966, 0.1151, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 20:26:00,642 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:26:04,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.171e+03 1.518e+03 2.066e+03 2.800e+03 6.029e+03, threshold=4.132e+03, percent-clipped=8.0
+2023-03-05 20:26:04,517 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481581.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:26:06,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481584.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:26:09,579 INFO [train.py:968] (0/2) Epoch 11, batch 26050, libri_loss[loss=0.2443, simple_loss=0.3105, pruned_loss=0.089, over 29396.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3752, pruned_loss=0.1282, over 5649423.09 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3717, pruned_loss=0.1233, over 5704587.13 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3756, pruned_loss=0.1285, over 5651857.67 frames. ], batch size: 67, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:26:12,819 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:26:34,391 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481613.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:26:46,611 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:26:48,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:26:56,227 INFO [train.py:968] (0/2) Epoch 11, batch 26100, libri_loss[loss=0.3317, simple_loss=0.3966, pruned_loss=0.1334, over 27675.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3777, pruned_loss=0.1288, over 5655067.74 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.372, pruned_loss=0.1235, over 5704068.52 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3778, pruned_loss=0.129, over 5656494.07 frames. ], batch size: 115, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:27:15,707 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:27:36,344 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.888e+02 1.499e+03 2.030e+03 2.838e+03 8.058e+03, threshold=4.060e+03, percent-clipped=13.0
+2023-03-05 20:27:45,165 INFO [train.py:968] (0/2) Epoch 11, batch 26150, giga_loss[loss=0.3452, simple_loss=0.4131, pruned_loss=0.1386, over 29050.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3825, pruned_loss=0.1294, over 5659054.98 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3719, pruned_loss=0.1234, over 5704772.51 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3827, pruned_loss=0.1296, over 5659186.33 frames. ], batch size: 136, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:27:56,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2063, 1.2643, 1.0728, 0.9829], device='cuda:0'), covar=tensor([0.0787, 0.0487, 0.0999, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0445, 0.0499, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 20:28:35,889 INFO [train.py:968] (0/2) Epoch 11, batch 26200, giga_loss[loss=0.3626, simple_loss=0.4165, pruned_loss=0.1544, over 28967.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3829, pruned_loss=0.1291, over 5650320.41 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3719, pruned_loss=0.1236, over 5697304.48 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3832, pruned_loss=0.1292, over 5656842.25 frames. ], batch size: 213, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:28:42,572 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-05 20:29:15,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.614e+03 1.916e+03 2.421e+03 6.030e+03, threshold=3.831e+03, percent-clipped=1.0
+2023-03-05 20:29:23,123 INFO [train.py:968] (0/2) Epoch 11, batch 26250, giga_loss[loss=0.3158, simple_loss=0.385, pruned_loss=0.1233, over 28985.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3849, pruned_loss=0.1311, over 5652270.36 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.372, pruned_loss=0.1236, over 5702305.04 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3853, pruned_loss=0.1313, over 5651948.89 frames. ], batch size: 164, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:29:41,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4742, 1.6655, 1.7388, 1.3054], device='cuda:0'), covar=tensor([0.1612, 0.2302, 0.1316, 0.1526], device='cuda:0'), in_proj_covar=tensor([0.0824, 0.0698, 0.0865, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 20:29:46,548 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481818.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:30:06,965 INFO [train.py:968] (0/2) Epoch 11, batch 26300, giga_loss[loss=0.4121, simple_loss=0.4301, pruned_loss=0.1971, over 23510.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.385, pruned_loss=0.1319, over 5651339.04 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.372, pruned_loss=0.1236, over 5707146.33 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3857, pruned_loss=0.1323, over 5645514.50 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:30:15,434 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2009, 2.5313, 1.1710, 1.4255], device='cuda:0'), covar=tensor([0.0935, 0.0376, 0.0859, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0514, 0.0339, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 20:30:43,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.845e+02 1.767e+03 2.132e+03 2.986e+03 8.635e+03, threshold=4.265e+03, percent-clipped=10.0
+2023-03-05 20:30:49,382 INFO [train.py:968] (0/2) Epoch 11, batch 26350, giga_loss[loss=0.3165, simple_loss=0.3844, pruned_loss=0.1243, over 28888.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.384, pruned_loss=0.1321, over 5653328.10 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3719, pruned_loss=0.1237, over 5700401.66 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3852, pruned_loss=0.1326, over 5652267.52 frames. ], batch size: 199, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:30:52,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5835, 2.5903, 1.7830, 2.0499], device='cuda:0'), covar=tensor([0.0749, 0.0666, 0.1004, 0.1061], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0445, 0.0501, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 20:31:38,059 INFO [train.py:968] (0/2) Epoch 11, batch 26400, giga_loss[loss=0.3277, simple_loss=0.3875, pruned_loss=0.134, over 28590.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3819, pruned_loss=0.1312, over 5650991.60 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3721, pruned_loss=0.1239, over 5704356.88 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3829, pruned_loss=0.1316, over 5645657.02 frames. ], batch size: 336, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 20:31:42,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:31:50,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:31:54,408 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:32:04,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:32:17,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.560e+02 1.672e+03 2.283e+03 3.292e+03 7.554e+03, threshold=4.565e+03, percent-clipped=14.0
+2023-03-05 20:32:24,256 INFO [train.py:968] (0/2) Epoch 11, batch 26450, giga_loss[loss=0.2685, simple_loss=0.3455, pruned_loss=0.09578, over 28834.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3796, pruned_loss=0.1306, over 5659810.36 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3718, pruned_loss=0.1239, over 5710927.40 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.381, pruned_loss=0.1312, over 5648160.25 frames. ], batch size: 174, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:32:35,583 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-482000.pt
+2023-03-05 20:33:14,102 INFO [train.py:968] (0/2) Epoch 11, batch 26500, giga_loss[loss=0.3272, simple_loss=0.3924, pruned_loss=0.131, over 28917.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.379, pruned_loss=0.1307, over 5650755.30 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.372, pruned_loss=0.1241, over 5706710.45 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3801, pruned_loss=0.1312, over 5644064.78 frames. ], batch size: 106, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:33:42,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6095, 1.6278, 1.8560, 1.4431], device='cuda:0'), covar=tensor([0.1372, 0.1941, 0.1094, 0.1340], device='cuda:0'), in_proj_covar=tensor([0.0825, 0.0699, 0.0866, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 20:33:54,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5666, 2.2269, 2.0353, 2.1137], device='cuda:0'), covar=tensor([0.1173, 0.2007, 0.1834, 0.1721], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0728, 0.0670, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 20:33:54,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.678e+03 2.010e+03 3.114e+03 8.431e+03, threshold=4.021e+03, percent-clipped=8.0
+2023-03-05 20:33:57,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7233, 1.0369, 2.8776, 2.6908], device='cuda:0'), covar=tensor([0.1699, 0.2398, 0.0570, 0.0949], device='cuda:0'), in_proj_covar=tensor([0.0669, 0.0594, 0.0863, 0.0766], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 20:33:58,963 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:00,948 INFO [train.py:968] (0/2) Epoch 11, batch 26550, giga_loss[loss=0.2802, simple_loss=0.3415, pruned_loss=0.1094, over 28697.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3799, pruned_loss=0.1316, over 5648793.49 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3721, pruned_loss=0.1241, over 5706214.61 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3808, pruned_loss=0.1321, over 5643348.73 frames. ], batch size: 92, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:34:01,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=482089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:09,386 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:11,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=482099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:21,361 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:23,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=482113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:28,953 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=482118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:38,242 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=482128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:47,666 INFO [train.py:968] (0/2) Epoch 11, batch 26600, giga_loss[loss=0.2982, simple_loss=0.3625, pruned_loss=0.1169, over 28802.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3778, pruned_loss=0.1305, over 5659618.60 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.372, pruned_loss=0.1241, over 5709300.77 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3787, pruned_loss=0.131, over 5651758.21 frames. ], batch size: 285, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:34:49,856 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=482142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:35:29,452 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.620e+03 2.079e+03 2.453e+03 6.878e+03, threshold=4.159e+03, percent-clipped=5.0
+2023-03-05 20:35:29,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-05 20:35:34,922 INFO [train.py:968] (0/2) Epoch 11, batch 26650, giga_loss[loss=0.2948, simple_loss=0.367, pruned_loss=0.1113, over 28903.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3771, pruned_loss=0.1301, over 5655466.73 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3725, pruned_loss=0.1244, over 5691897.15 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3774, pruned_loss=0.1302, over 5663422.19 frames. ], batch size: 174, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:35:39,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=482193.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:36:20,098 INFO [train.py:968] (0/2) Epoch 11, batch 26700, giga_loss[loss=0.3255, simple_loss=0.387, pruned_loss=0.132, over 28696.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3772, pruned_loss=0.1296, over 5659351.74 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3723, pruned_loss=0.1242, over 5693887.62 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3778, pruned_loss=0.1301, over 5662773.45 frames. ], batch size: 262, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:36:37,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7154, 1.7637, 1.7373, 1.7545], device='cuda:0'), covar=tensor([0.1533, 0.2112, 0.1931, 0.1713], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0734, 0.0672, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 20:36:42,571 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-05 20:37:04,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.618e+02 1.527e+03 2.069e+03 2.940e+03 1.270e+04, threshold=4.138e+03, percent-clipped=9.0
+2023-03-05 20:37:09,594 INFO [train.py:968] (0/2) Epoch 11, batch 26750, giga_loss[loss=0.2873, simple_loss=0.3632, pruned_loss=0.1057, over 28991.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3799, pruned_loss=0.1307, over 5657684.59 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3726, pruned_loss=0.1244, over 5694287.81 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3802, pruned_loss=0.131, over 5659715.10 frames. ], batch size: 155, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:37:56,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=482332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:37:59,040 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482336.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:38:00,908 INFO [train.py:968] (0/2) Epoch 11, batch 26800, giga_loss[loss=0.2943, simple_loss=0.3663, pruned_loss=0.1111, over 28891.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3787, pruned_loss=0.1299, over 5658265.86 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3726, pruned_loss=0.1243, over 5697564.09 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3791, pruned_loss=0.1303, over 5656319.70 frames. ], batch size: 199, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:38:02,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=482339.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:38:21,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7786, 1.7131, 1.2338, 1.3844], device='cuda:0'), covar=tensor([0.0705, 0.0616, 0.0987, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0445, 0.0501, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 20:38:29,330 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=482368.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:38:42,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.735e+03 2.272e+03 3.746e+03 1.125e+04, threshold=4.545e+03, percent-clipped=17.0
+2023-03-05 20:38:46,923 INFO [train.py:968] (0/2) Epoch 11, batch 26850, giga_loss[loss=0.2921, simple_loss=0.3831, pruned_loss=0.1006, over 28971.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3813, pruned_loss=0.1303, over 5670253.02 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3726, pruned_loss=0.1243, over 5702019.21 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3817, pruned_loss=0.1307, over 5663861.41 frames. ], batch size: 164, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:39:33,123 INFO [train.py:968] (0/2) Epoch 11, batch 26900, giga_loss[loss=0.4095, simple_loss=0.4275, pruned_loss=0.1957, over 23619.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3829, pruned_loss=0.1293, over 5667339.63 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.373, pruned_loss=0.1246, over 5698173.03 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3829, pruned_loss=0.1294, over 5665292.99 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:40:04,841 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:40:08,164 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=482478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:40:09,510 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=482480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:40:11,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.903e+02 1.448e+03 1.818e+03 2.306e+03 1.046e+04, threshold=3.635e+03, percent-clipped=5.0
+2023-03-05 20:40:16,633 INFO [train.py:968] (0/2) Epoch 11, batch 26950, giga_loss[loss=0.4211, simple_loss=0.434, pruned_loss=0.2041, over 23684.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3838, pruned_loss=0.128, over 5672744.90 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3729, pruned_loss=0.1245, over 5701998.85 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3843, pruned_loss=0.1283, over 5666729.27 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:40:32,303 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=482507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:41:00,992 INFO [train.py:968] (0/2) Epoch 11, batch 27000, giga_loss[loss=0.4434, simple_loss=0.4556, pruned_loss=0.2156, over 26436.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.387, pruned_loss=0.1313, over 5669388.31 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3729, pruned_loss=0.1248, over 5696764.59 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3876, pruned_loss=0.1314, over 5669315.82 frames. ], batch size: 555, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:41:00,997 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 20:41:09,806 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2156, simple_loss=0.321, pruned_loss=0.05513, over 944034.00 frames. 
+2023-03-05 20:41:09,807 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 20:41:12,144 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5216, 1.6311, 1.5755, 1.5022], device='cuda:0'), covar=tensor([0.1391, 0.1672, 0.1949, 0.1662], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0730, 0.0667, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 20:41:52,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.814e+02 1.653e+03 2.161e+03 3.039e+03 8.778e+03, threshold=4.322e+03, percent-clipped=12.0
+2023-03-05 20:41:56,152 INFO [train.py:968] (0/2) Epoch 11, batch 27050, giga_loss[loss=0.3975, simple_loss=0.4185, pruned_loss=0.1882, over 23584.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.389, pruned_loss=0.1338, over 5664232.21 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3729, pruned_loss=0.1247, over 5691786.74 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.39, pruned_loss=0.1342, over 5666805.30 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:42:29,292 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2864, 1.5201, 1.2511, 1.4956], device='cuda:0'), covar=tensor([0.0738, 0.0312, 0.0310, 0.0789], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0058, 0.0052, 0.0088], device='cuda:0')
+2023-03-05 20:42:46,787 INFO [train.py:968] (0/2) Epoch 11, batch 27100, giga_loss[loss=0.3397, simple_loss=0.3936, pruned_loss=0.1429, over 28574.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3896, pruned_loss=0.135, over 5675129.71 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3727, pruned_loss=0.1246, over 5696087.98 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3908, pruned_loss=0.1356, over 5672810.44 frames. ], batch size: 336, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 20:43:30,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.558e+03 1.959e+03 2.757e+03 6.486e+03, threshold=3.918e+03, percent-clipped=8.0
+2023-03-05 20:43:36,842 INFO [train.py:968] (0/2) Epoch 11, batch 27150, giga_loss[loss=0.2917, simple_loss=0.3635, pruned_loss=0.11, over 28874.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3884, pruned_loss=0.1344, over 5667604.76 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3724, pruned_loss=0.1244, over 5698026.47 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.39, pruned_loss=0.1352, over 5663818.05 frames. ], batch size: 112, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 20:44:22,625 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 20:44:24,262 INFO [train.py:968] (0/2) Epoch 11, batch 27200, giga_loss[loss=0.3506, simple_loss=0.4187, pruned_loss=0.1413, over 28970.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3867, pruned_loss=0.1316, over 5674574.77 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3722, pruned_loss=0.1243, over 5699966.87 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3882, pruned_loss=0.1324, over 5669636.51 frames. ], batch size: 227, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:45:07,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.521e+02 1.534e+03 2.229e+03 3.517e+03 1.575e+04, threshold=4.459e+03, percent-clipped=22.0
+2023-03-05 20:45:11,578 INFO [train.py:968] (0/2) Epoch 11, batch 27250, giga_loss[loss=0.3529, simple_loss=0.405, pruned_loss=0.1504, over 27606.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3883, pruned_loss=0.1323, over 5651693.93 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3727, pruned_loss=0.1248, over 5693177.85 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3893, pruned_loss=0.1325, over 5653416.73 frames. ], batch size: 472, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:45:54,750 INFO [train.py:968] (0/2) Epoch 11, batch 27300, giga_loss[loss=0.3571, simple_loss=0.3877, pruned_loss=0.1632, over 23525.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3872, pruned_loss=0.131, over 5659921.77 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3725, pruned_loss=0.1247, over 5691998.83 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3889, pruned_loss=0.1317, over 5660731.16 frames. ], batch size: 705, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:46:11,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=482855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:46:42,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.061e+02 1.623e+03 2.117e+03 3.188e+03 9.306e+03, threshold=4.234e+03, percent-clipped=8.0
+2023-03-05 20:46:46,278 INFO [train.py:968] (0/2) Epoch 11, batch 27350, giga_loss[loss=0.2804, simple_loss=0.353, pruned_loss=0.1039, over 28618.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3886, pruned_loss=0.1329, over 5652638.79 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3722, pruned_loss=0.1245, over 5695053.57 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3905, pruned_loss=0.1337, over 5649958.00 frames. ], batch size: 92, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:47:05,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3875, 3.3340, 1.5116, 1.4802], device='cuda:0'), covar=tensor([0.0931, 0.0331, 0.0875, 0.1305], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0517, 0.0343, 0.0396], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-05 20:47:31,052 INFO [train.py:968] (0/2) Epoch 11, batch 27400, giga_loss[loss=0.3727, simple_loss=0.3943, pruned_loss=0.1755, over 23476.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3875, pruned_loss=0.1325, over 5644111.62 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3721, pruned_loss=0.1245, over 5680496.92 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3894, pruned_loss=0.1334, over 5654673.55 frames. ], batch size: 705, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:48:04,752 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-05 20:48:16,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.796e+02 1.571e+03 1.978e+03 3.000e+03 5.880e+03, threshold=3.956e+03, percent-clipped=4.0
+2023-03-05 20:48:22,193 INFO [train.py:968] (0/2) Epoch 11, batch 27450, giga_loss[loss=0.3124, simple_loss=0.373, pruned_loss=0.1259, over 28256.00 frames. ], tot_loss[loss=0.324, simple_loss=0.385, pruned_loss=0.1314, over 5661381.77 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3719, pruned_loss=0.1243, over 5683703.28 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3869, pruned_loss=0.1323, over 5666408.67 frames. ], batch size: 368, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:48:23,374 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-05 20:48:29,628 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:48:32,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=483001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:49:02,196 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=483030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:49:09,588 INFO [train.py:968] (0/2) Epoch 11, batch 27500, giga_loss[loss=0.2893, simple_loss=0.3623, pruned_loss=0.1081, over 28946.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3834, pruned_loss=0.1309, over 5665201.81 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3725, pruned_loss=0.1247, over 5685931.47 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3848, pruned_loss=0.1315, over 5666350.18 frames. ], batch size: 227, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:49:53,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.808e+02 1.679e+03 2.269e+03 3.169e+03 1.088e+04, threshold=4.538e+03, percent-clipped=14.0
+2023-03-05 20:49:56,672 INFO [train.py:968] (0/2) Epoch 11, batch 27550, giga_loss[loss=0.338, simple_loss=0.3881, pruned_loss=0.1439, over 28994.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3813, pruned_loss=0.1304, over 5665339.77 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3718, pruned_loss=0.1244, over 5692309.46 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3834, pruned_loss=0.1314, over 5659406.59 frames. ], batch size: 227, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:49:59,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 20:50:30,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3239, 1.7979, 1.2793, 0.7627], device='cuda:0'), covar=tensor([0.3323, 0.1759, 0.2027, 0.3754], device='cuda:0'), in_proj_covar=tensor([0.1574, 0.1496, 0.1504, 0.1282], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 20:50:42,430 INFO [train.py:968] (0/2) Epoch 11, batch 27600, giga_loss[loss=0.3495, simple_loss=0.3914, pruned_loss=0.1538, over 27526.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3816, pruned_loss=0.1314, over 5667865.35 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3724, pruned_loss=0.1248, over 5694180.34 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3829, pruned_loss=0.1319, over 5661272.55 frames. ], batch size: 472, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 20:51:23,181 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 20:51:23,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.080e+02 1.461e+03 2.068e+03 3.062e+03 8.067e+03, threshold=4.136e+03, percent-clipped=11.0
+2023-03-05 20:51:27,657 INFO [train.py:968] (0/2) Epoch 11, batch 27650, giga_loss[loss=0.2795, simple_loss=0.352, pruned_loss=0.1035, over 28737.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.38, pruned_loss=0.1301, over 5669504.46 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3723, pruned_loss=0.1248, over 5696815.15 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3813, pruned_loss=0.1307, over 5661281.50 frames. ], batch size: 262, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:52:09,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5844, 1.8561, 1.8364, 1.3985], device='cuda:0'), covar=tensor([0.1927, 0.2307, 0.1515, 0.1713], device='cuda:0'), in_proj_covar=tensor([0.0831, 0.0704, 0.0874, 0.0774], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 20:52:15,944 INFO [train.py:968] (0/2) Epoch 11, batch 27700, giga_loss[loss=0.257, simple_loss=0.3382, pruned_loss=0.08795, over 28834.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3751, pruned_loss=0.125, over 5670726.96 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3718, pruned_loss=0.1244, over 5698863.19 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3766, pruned_loss=0.1258, over 5662153.01 frames. ], batch size: 199, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:52:26,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4074, 1.6535, 1.2999, 1.4551], device='cuda:0'), covar=tensor([0.2581, 0.2480, 0.2852, 0.2197], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.0962, 0.1150, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 20:53:03,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.990e+02 1.285e+03 1.617e+03 2.557e+03 8.452e+03, threshold=3.233e+03, percent-clipped=5.0
+2023-03-05 20:53:06,204 INFO [train.py:968] (0/2) Epoch 11, batch 27750, libri_loss[loss=0.2868, simple_loss=0.3506, pruned_loss=0.1115, over 29322.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3745, pruned_loss=0.1246, over 5653542.99 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3716, pruned_loss=0.1242, over 5692384.95 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.376, pruned_loss=0.1254, over 5652120.15 frames. ], batch size: 71, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:53:57,137 INFO [train.py:968] (0/2) Epoch 11, batch 27800, giga_loss[loss=0.3062, simple_loss=0.3678, pruned_loss=0.1223, over 28595.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3727, pruned_loss=0.1239, over 5649976.80 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3714, pruned_loss=0.1242, over 5686800.00 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.374, pruned_loss=0.1246, over 5653133.16 frames. ], batch size: 307, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:54:18,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1928, 1.5192, 1.2935, 1.0201], device='cuda:0'), covar=tensor([0.2069, 0.1800, 0.1243, 0.1746], device='cuda:0'), in_proj_covar=tensor([0.1720, 0.1623, 0.1592, 0.1693], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 20:54:33,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=483370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:54:46,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.671e+02 1.702e+03 2.122e+03 2.845e+03 8.389e+03, threshold=4.245e+03, percent-clipped=20.0
+2023-03-05 20:54:51,819 INFO [train.py:968] (0/2) Epoch 11, batch 27850, giga_loss[loss=0.269, simple_loss=0.3474, pruned_loss=0.09531, over 28876.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3691, pruned_loss=0.1223, over 5651690.29 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3716, pruned_loss=0.1244, over 5690012.04 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.37, pruned_loss=0.1227, over 5650812.72 frames. ], batch size: 174, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:55:39,904 INFO [train.py:968] (0/2) Epoch 11, batch 27900, giga_loss[loss=0.2914, simple_loss=0.3629, pruned_loss=0.11, over 28844.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.371, pruned_loss=0.1233, over 5655941.44 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3714, pruned_loss=0.1242, over 5694585.18 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3718, pruned_loss=0.1237, over 5650092.57 frames. ], batch size: 119, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:56:19,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4153, 2.9032, 1.6082, 1.5622], device='cuda:0'), covar=tensor([0.0743, 0.0270, 0.0674, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0514, 0.0340, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-05 20:56:20,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.354e+02 1.589e+03 2.089e+03 2.932e+03 9.371e+03, threshold=4.179e+03, percent-clipped=13.0
+2023-03-05 20:56:25,296 INFO [train.py:968] (0/2) Epoch 11, batch 27950, giga_loss[loss=0.3687, simple_loss=0.394, pruned_loss=0.1717, over 23237.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3727, pruned_loss=0.1239, over 5632930.81 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3717, pruned_loss=0.1243, over 5679634.20 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3731, pruned_loss=0.1241, over 5639742.85 frames. ], batch size: 705, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:56:29,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6511, 1.6148, 1.2158, 1.2484], device='cuda:0'), covar=tensor([0.0685, 0.0457, 0.0926, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0441, 0.0501, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-05 20:56:56,782 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3149, 1.9411, 1.5039, 0.5682], device='cuda:0'), covar=tensor([0.3285, 0.1903, 0.2743, 0.3963], device='cuda:0'), in_proj_covar=tensor([0.1561, 0.1480, 0.1490, 0.1270], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 20:57:12,755 INFO [train.py:968] (0/2) Epoch 11, batch 28000, libri_loss[loss=0.3358, simple_loss=0.4027, pruned_loss=0.1345, over 25932.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3732, pruned_loss=0.1237, over 5643398.81 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3723, pruned_loss=0.1246, over 5680178.18 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.373, pruned_loss=0.1236, over 5647607.47 frames. ], batch size: 136, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 20:57:57,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.780e+02 1.401e+03 1.763e+03 2.335e+03 6.383e+03, threshold=3.526e+03, percent-clipped=4.0
+2023-03-05 20:57:58,928 INFO [train.py:968] (0/2) Epoch 11, batch 28050, giga_loss[loss=0.2983, simple_loss=0.369, pruned_loss=0.1138, over 28764.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3733, pruned_loss=0.124, over 5646834.98 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3727, pruned_loss=0.1246, over 5685661.04 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3727, pruned_loss=0.1238, over 5644117.57 frames. ], batch size: 284, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:58:05,746 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5537, 1.4346, 1.3006, 1.1295], device='cuda:0'), covar=tensor([0.0568, 0.0334, 0.0712, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0439, 0.0499, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 20:58:22,532 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 20:58:46,460 INFO [train.py:968] (0/2) Epoch 11, batch 28100, giga_loss[loss=0.3595, simple_loss=0.4058, pruned_loss=0.1566, over 28006.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3739, pruned_loss=0.1249, over 5644350.15 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3727, pruned_loss=0.1246, over 5686855.74 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3735, pruned_loss=0.1248, over 5641113.59 frames. ], batch size: 412, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:58:46,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3621, 4.2058, 3.9523, 1.8538], device='cuda:0'), covar=tensor([0.0548, 0.0625, 0.0704, 0.2081], device='cuda:0'), in_proj_covar=tensor([0.1059, 0.0996, 0.0872, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 20:59:33,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.049e+02 1.511e+03 2.096e+03 3.196e+03 1.086e+04, threshold=4.192e+03, percent-clipped=18.0
+2023-03-05 20:59:37,144 INFO [train.py:968] (0/2) Epoch 11, batch 28150, giga_loss[loss=0.31, simple_loss=0.3781, pruned_loss=0.121, over 29054.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3762, pruned_loss=0.1262, over 5648267.09 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3726, pruned_loss=0.1245, over 5687983.46 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.376, pruned_loss=0.1262, over 5644516.78 frames. ], batch size: 128, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:59:39,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3143, 3.0389, 1.4431, 1.4395], device='cuda:0'), covar=tensor([0.0943, 0.0352, 0.0854, 0.1325], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0516, 0.0341, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-05 21:00:25,839 INFO [train.py:968] (0/2) Epoch 11, batch 28200, giga_loss[loss=0.3161, simple_loss=0.3774, pruned_loss=0.1274, over 28696.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3788, pruned_loss=0.1285, over 5650331.89 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3728, pruned_loss=0.1247, over 5690196.90 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3785, pruned_loss=0.1283, over 5645151.59 frames. ], batch size: 242, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:00:34,543 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=483745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:00:34,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=483745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:01:14,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.757e+03 2.210e+03 3.350e+03 9.179e+03, threshold=4.419e+03, percent-clipped=12.0
+2023-03-05 21:01:16,464 INFO [train.py:968] (0/2) Epoch 11, batch 28250, libri_loss[loss=0.3161, simple_loss=0.3884, pruned_loss=0.1219, over 29283.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3798, pruned_loss=0.1297, over 5652682.74 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3727, pruned_loss=0.1246, over 5693267.08 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3797, pruned_loss=0.1297, over 5645361.49 frames. ], batch size: 97, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:01:18,974 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6342, 1.8586, 1.5722, 1.5188], device='cuda:0'), covar=tensor([0.2021, 0.1675, 0.1434, 0.1496], device='cuda:0'), in_proj_covar=tensor([0.1712, 0.1617, 0.1592, 0.1689], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 21:02:09,840 INFO [train.py:968] (0/2) Epoch 11, batch 28300, libri_loss[loss=0.273, simple_loss=0.3308, pruned_loss=0.1076, over 29632.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3794, pruned_loss=0.1291, over 5653497.37 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3723, pruned_loss=0.1245, over 5696501.64 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3798, pruned_loss=0.1294, over 5644075.76 frames. ], batch size: 69, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:02:52,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4312, 1.9121, 1.5080, 0.8402], device='cuda:0'), covar=tensor([0.2754, 0.1936, 0.2105, 0.3099], device='cuda:0'), in_proj_covar=tensor([0.1576, 0.1492, 0.1496, 0.1279], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 21:02:55,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.071e+02 1.646e+03 2.222e+03 2.990e+03 1.037e+04, threshold=4.445e+03, percent-clipped=9.0
+2023-03-05 21:03:00,233 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=483888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:03:00,683 INFO [train.py:968] (0/2) Epoch 11, batch 28350, giga_loss[loss=0.3628, simple_loss=0.39, pruned_loss=0.1678, over 23586.00 frames. ], tot_loss[loss=0.316, simple_loss=0.378, pruned_loss=0.127, over 5646864.52 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3721, pruned_loss=0.1246, over 5691670.35 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3787, pruned_loss=0.1272, over 5643000.77 frames. ], batch size: 705, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:03:02,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=483891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:03:21,563 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4152, 1.7102, 1.3881, 1.6074], device='cuda:0'), covar=tensor([0.2194, 0.1968, 0.2149, 0.1875], device='cuda:0'), in_proj_covar=tensor([0.1296, 0.0959, 0.1147, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 21:03:31,039 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=483920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:03:45,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0274, 1.3275, 1.0789, 0.1975], device='cuda:0'), covar=tensor([0.2356, 0.2087, 0.3018, 0.4104], device='cuda:0'), in_proj_covar=tensor([0.1576, 0.1496, 0.1498, 0.1279], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 21:03:46,112 INFO [train.py:968] (0/2) Epoch 11, batch 28400, giga_loss[loss=0.3892, simple_loss=0.4264, pruned_loss=0.176, over 28050.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.379, pruned_loss=0.1286, over 5635165.76 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3721, pruned_loss=0.1246, over 5684576.68 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3798, pruned_loss=0.1288, over 5637204.46 frames. ], batch size: 412, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:04:42,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.071e+02 1.642e+03 2.162e+03 2.993e+03 9.712e+03, threshold=4.323e+03, percent-clipped=7.0
+2023-03-05 21:04:43,582 INFO [train.py:968] (0/2) Epoch 11, batch 28450, giga_loss[loss=0.2824, simple_loss=0.3483, pruned_loss=0.1083, over 28700.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3793, pruned_loss=0.1297, over 5623486.57 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3722, pruned_loss=0.1247, over 5685680.80 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3798, pruned_loss=0.1298, over 5623825.89 frames. ], batch size: 85, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:04:55,163 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-484000.pt
+2023-03-05 21:05:40,361 INFO [train.py:968] (0/2) Epoch 11, batch 28500, giga_loss[loss=0.3224, simple_loss=0.3751, pruned_loss=0.1348, over 28247.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3776, pruned_loss=0.1292, over 5630724.86 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3716, pruned_loss=0.1242, over 5691032.19 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3788, pruned_loss=0.1299, over 5625061.84 frames. ], batch size: 368, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:06:27,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.884e+02 1.505e+03 1.831e+03 2.531e+03 1.092e+04, threshold=3.663e+03, percent-clipped=7.0
+2023-03-05 21:06:28,729 INFO [train.py:968] (0/2) Epoch 11, batch 28550, giga_loss[loss=0.3165, simple_loss=0.3776, pruned_loss=0.1277, over 28898.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3771, pruned_loss=0.1294, over 5631058.46 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3717, pruned_loss=0.1244, over 5687973.84 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3781, pruned_loss=0.1299, over 5627277.36 frames. ], batch size: 186, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:06:29,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5195, 1.6621, 1.6184, 1.5197], device='cuda:0'), covar=tensor([0.1401, 0.1666, 0.1842, 0.1644], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0734, 0.0674, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 21:06:55,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=484120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:07:13,352 INFO [train.py:968] (0/2) Epoch 11, batch 28600, libri_loss[loss=0.3153, simple_loss=0.3815, pruned_loss=0.1246, over 29502.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3759, pruned_loss=0.1285, over 5645244.41 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3718, pruned_loss=0.1244, over 5691086.75 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3768, pruned_loss=0.129, over 5637977.08 frames. ], batch size: 84, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:07:32,787 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 21:07:58,174 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.567e+03 2.141e+03 3.080e+03 1.163e+04, threshold=4.281e+03, percent-clipped=17.0
+2023-03-05 21:07:59,015 INFO [train.py:968] (0/2) Epoch 11, batch 28650, giga_loss[loss=0.2977, simple_loss=0.3555, pruned_loss=0.1199, over 28532.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3747, pruned_loss=0.1276, over 5636437.79 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3718, pruned_loss=0.1243, over 5680171.74 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3756, pruned_loss=0.1282, over 5638294.84 frames. ], batch size: 60, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:08:45,210 INFO [train.py:968] (0/2) Epoch 11, batch 28700, giga_loss[loss=0.2714, simple_loss=0.3394, pruned_loss=0.1017, over 28832.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3763, pruned_loss=0.1287, over 5655972.46 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3717, pruned_loss=0.1243, over 5687712.18 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3771, pruned_loss=0.1293, over 5649681.28 frames. ], batch size: 119, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:09:08,224 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=484263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:09:10,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=484266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:09:30,808 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.864e+02 1.740e+03 2.293e+03 3.059e+03 1.214e+04, threshold=4.586e+03, percent-clipped=17.0
+2023-03-05 21:09:31,358 INFO [train.py:968] (0/2) Epoch 11, batch 28750, giga_loss[loss=0.4072, simple_loss=0.4417, pruned_loss=0.1863, over 28398.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3793, pruned_loss=0.1313, over 5650889.46 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3721, pruned_loss=0.1246, over 5677533.20 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3797, pruned_loss=0.1316, over 5653750.21 frames. ], batch size: 369, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:09:38,410 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=484295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:10:19,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3750, 1.6310, 1.3123, 1.5143], device='cuda:0'), covar=tensor([0.2352, 0.2283, 0.2565, 0.2095], device='cuda:0'), in_proj_covar=tensor([0.1298, 0.0961, 0.1150, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 21:10:21,581 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-05 21:10:21,918 INFO [train.py:968] (0/2) Epoch 11, batch 28800, giga_loss[loss=0.3085, simple_loss=0.3788, pruned_loss=0.1191, over 28925.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.38, pruned_loss=0.1315, over 5657583.21 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3721, pruned_loss=0.1245, over 5679966.94 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3804, pruned_loss=0.132, over 5657209.77 frames. ], batch size: 213, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:11:07,492 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.564e+03 2.269e+03 3.315e+03 7.597e+03, threshold=4.539e+03, percent-clipped=17.0
+2023-03-05 21:11:09,244 INFO [train.py:968] (0/2) Epoch 11, batch 28850, giga_loss[loss=0.3076, simple_loss=0.3796, pruned_loss=0.1178, over 29102.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3806, pruned_loss=0.1323, over 5669061.78 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3719, pruned_loss=0.1244, over 5683256.70 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3812, pruned_loss=0.1329, over 5665365.77 frames. ], batch size: 155, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:11:53,705 INFO [train.py:968] (0/2) Epoch 11, batch 28900, giga_loss[loss=0.4008, simple_loss=0.4268, pruned_loss=0.1874, over 26691.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3791, pruned_loss=0.131, over 5668938.54 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3719, pruned_loss=0.1242, over 5686667.75 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3798, pruned_loss=0.1318, over 5662866.70 frames. ], batch size: 555, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:12:42,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.842e+02 1.496e+03 2.108e+03 3.331e+03 1.129e+04, threshold=4.216e+03, percent-clipped=8.0
+2023-03-05 21:12:42,888 INFO [train.py:968] (0/2) Epoch 11, batch 28950, giga_loss[loss=0.4019, simple_loss=0.4331, pruned_loss=0.1853, over 28656.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3808, pruned_loss=0.1321, over 5661596.90 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.372, pruned_loss=0.1242, over 5680304.29 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3815, pruned_loss=0.1329, over 5662787.48 frames. ], batch size: 307, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:13:27,507 INFO [train.py:968] (0/2) Epoch 11, batch 29000, giga_loss[loss=0.2674, simple_loss=0.3498, pruned_loss=0.09248, over 28901.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3823, pruned_loss=0.1329, over 5662940.58 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3721, pruned_loss=0.1243, over 5676071.63 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.383, pruned_loss=0.1336, over 5666772.71 frames. ], batch size: 174, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:14:15,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.767e+02 1.397e+03 1.739e+03 2.552e+03 4.009e+03, threshold=3.477e+03, percent-clipped=0.0
+2023-03-05 21:14:16,518 INFO [train.py:968] (0/2) Epoch 11, batch 29050, giga_loss[loss=0.3444, simple_loss=0.4022, pruned_loss=0.1433, over 28314.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3839, pruned_loss=0.1344, over 5662032.86 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3723, pruned_loss=0.1245, over 5673403.27 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3843, pruned_loss=0.1349, over 5667069.59 frames. ], batch size: 368, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:14:20,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3146, 3.1354, 2.9830, 1.3692], device='cuda:0'), covar=tensor([0.0852, 0.1006, 0.0887, 0.2231], device='cuda:0'), in_proj_covar=tensor([0.1060, 0.0999, 0.0878, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 21:15:03,620 INFO [train.py:968] (0/2) Epoch 11, batch 29100, giga_loss[loss=0.2756, simple_loss=0.3472, pruned_loss=0.102, over 28351.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3849, pruned_loss=0.1358, over 5660848.93 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3722, pruned_loss=0.1246, over 5676511.15 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3854, pruned_loss=0.1362, over 5661813.18 frames. ], batch size: 71, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:15:41,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8788, 3.6825, 3.4847, 1.6800], device='cuda:0'), covar=tensor([0.0676, 0.0833, 0.0821, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.1065, 0.1005, 0.0881, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 21:15:44,369 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.456e+02 1.525e+03 1.876e+03 2.931e+03 1.034e+04, threshold=3.751e+03, percent-clipped=13.0
+2023-03-05 21:15:45,491 INFO [train.py:968] (0/2) Epoch 11, batch 29150, giga_loss[loss=0.3625, simple_loss=0.4064, pruned_loss=0.1593, over 27510.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3844, pruned_loss=0.1349, over 5659639.02 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3727, pruned_loss=0.1248, over 5680289.68 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3848, pruned_loss=0.1354, over 5656623.58 frames. ], batch size: 472, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:15:55,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8524, 1.1140, 1.0725, 0.7407], device='cuda:0'), covar=tensor([0.1661, 0.1670, 0.0965, 0.1642], device='cuda:0'), in_proj_covar=tensor([0.1701, 0.1608, 0.1575, 0.1681], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 21:16:25,927 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5414, 1.4146, 1.6498, 1.2468], device='cuda:0'), covar=tensor([0.1593, 0.2612, 0.1338, 0.1457], device='cuda:0'), in_proj_covar=tensor([0.0831, 0.0704, 0.0872, 0.0772], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 21:16:38,168 INFO [train.py:968] (0/2) Epoch 11, batch 29200, giga_loss[loss=0.3413, simple_loss=0.4101, pruned_loss=0.1363, over 29009.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3845, pruned_loss=0.1343, over 5647451.03 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3725, pruned_loss=0.1248, over 5683616.50 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3851, pruned_loss=0.1348, over 5641731.98 frames. ], batch size: 145, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:16:45,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3619, 1.4989, 1.5119, 1.3745], device='cuda:0'), covar=tensor([0.1375, 0.1592, 0.1783, 0.1550], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0739, 0.0678, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 21:16:50,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.38 vs. limit=5.0
+2023-03-05 21:17:27,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.440e+02 1.575e+03 2.031e+03 2.794e+03 6.104e+03, threshold=4.061e+03, percent-clipped=11.0
+2023-03-05 21:17:28,183 INFO [train.py:968] (0/2) Epoch 11, batch 29250, giga_loss[loss=0.3233, simple_loss=0.3823, pruned_loss=0.1321, over 28619.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3837, pruned_loss=0.1332, over 5651587.46 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3722, pruned_loss=0.1248, over 5688901.02 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3849, pruned_loss=0.1339, over 5640971.67 frames. ], batch size: 85, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:17:31,993 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-05 21:18:11,692 INFO [train.py:968] (0/2) Epoch 11, batch 29300, giga_loss[loss=0.2845, simple_loss=0.353, pruned_loss=0.108, over 28725.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3821, pruned_loss=0.1314, over 5660984.61 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3725, pruned_loss=0.1249, over 5692902.91 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.383, pruned_loss=0.132, over 5648189.76 frames. ], batch size: 92, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:18:12,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3815, 1.6355, 1.4620, 1.2156], device='cuda:0'), covar=tensor([0.2228, 0.1708, 0.1333, 0.1824], device='cuda:0'), in_proj_covar=tensor([0.1707, 0.1610, 0.1579, 0.1684], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 21:18:48,004 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=484881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:18:54,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.688e+02 1.472e+03 1.839e+03 2.542e+03 6.327e+03, threshold=3.678e+03, percent-clipped=8.0
+2023-03-05 21:18:55,475 INFO [train.py:968] (0/2) Epoch 11, batch 29350, giga_loss[loss=0.3523, simple_loss=0.3821, pruned_loss=0.1613, over 23620.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3808, pruned_loss=0.1308, over 5652590.19 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3724, pruned_loss=0.1249, over 5685290.40 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3818, pruned_loss=0.1314, over 5648547.37 frames. ], batch size: 705, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:18:56,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=484891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:19:00,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6122, 1.7689, 1.4450, 1.9276], device='cuda:0'), covar=tensor([0.2415, 0.2324, 0.2592, 0.2160], device='cuda:0'), in_proj_covar=tensor([0.1297, 0.0960, 0.1149, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 21:19:21,111 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 21:19:43,637 INFO [train.py:968] (0/2) Epoch 11, batch 29400, giga_loss[loss=0.3063, simple_loss=0.3808, pruned_loss=0.1158, over 29036.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3814, pruned_loss=0.1312, over 5647785.52 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3723, pruned_loss=0.1247, over 5684734.32 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3824, pruned_loss=0.1319, over 5644628.02 frames. ], batch size: 155, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:20:32,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.956e+02 1.456e+03 1.736e+03 2.364e+03 4.931e+03, threshold=3.473e+03, percent-clipped=6.0
+2023-03-05 21:20:32,517 INFO [train.py:968] (0/2) Epoch 11, batch 29450, giga_loss[loss=0.3017, simple_loss=0.3703, pruned_loss=0.1166, over 28620.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3815, pruned_loss=0.1308, over 5660450.72 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3725, pruned_loss=0.1248, over 5687524.52 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3823, pruned_loss=0.1315, over 5654816.57 frames. ], batch size: 242, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:20:50,143 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:21:09,260 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:21:23,533 INFO [train.py:968] (0/2) Epoch 11, batch 29500, giga_loss[loss=0.3641, simple_loss=0.4182, pruned_loss=0.155, over 28926.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3811, pruned_loss=0.1318, over 5657565.40 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3724, pruned_loss=0.1248, over 5690242.15 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3819, pruned_loss=0.1324, over 5650353.93 frames. ], batch size: 227, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:21:28,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6208, 1.7433, 1.7446, 1.4608], device='cuda:0'), covar=tensor([0.1873, 0.1677, 0.1200, 0.1674], device='cuda:0'), in_proj_covar=tensor([0.1720, 0.1621, 0.1588, 0.1691], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 21:22:03,914 INFO [train.py:968] (0/2) Epoch 11, batch 29550, giga_loss[loss=0.3548, simple_loss=0.4034, pruned_loss=0.1531, over 28551.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3807, pruned_loss=0.1311, over 5675612.80 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3725, pruned_loss=0.1245, over 5696184.32 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3816, pruned_loss=0.132, over 5663396.45 frames. ], batch size: 336, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:22:04,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.771e+03 2.335e+03 3.301e+03 1.130e+04, threshold=4.671e+03, percent-clipped=22.0
+2023-03-05 21:22:35,693 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-05 21:22:37,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:22:49,853 INFO [train.py:968] (0/2) Epoch 11, batch 29600, libri_loss[loss=0.3175, simple_loss=0.3869, pruned_loss=0.124, over 29541.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3839, pruned_loss=0.1337, over 5671178.51 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.373, pruned_loss=0.1248, over 5699729.52 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3844, pruned_loss=0.1345, over 5657281.72 frames. ], batch size: 81, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:23:27,960 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-05 21:23:38,911 INFO [train.py:968] (0/2) Epoch 11, batch 29650, giga_loss[loss=0.2657, simple_loss=0.3364, pruned_loss=0.09755, over 28631.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.384, pruned_loss=0.1342, over 5659024.70 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3731, pruned_loss=0.1249, over 5704474.29 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3847, pruned_loss=0.1349, over 5643115.23 frames. ], batch size: 85, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:23:39,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.745e+03 2.263e+03 3.307e+03 6.979e+03, threshold=4.526e+03, percent-clipped=8.0
+2023-03-05 21:24:19,373 INFO [train.py:968] (0/2) Epoch 11, batch 29700, libri_loss[loss=0.3671, simple_loss=0.4098, pruned_loss=0.1622, over 19234.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3849, pruned_loss=0.1349, over 5637857.52 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3739, pruned_loss=0.1255, over 5681359.40 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3852, pruned_loss=0.1354, over 5644583.95 frames. ], batch size: 188, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:24:37,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:24:48,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:25:09,448 INFO [train.py:968] (0/2) Epoch 11, batch 29750, giga_loss[loss=0.3429, simple_loss=0.4028, pruned_loss=0.1415, over 28120.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3853, pruned_loss=0.1346, over 5636363.58 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3737, pruned_loss=0.1254, over 5673358.71 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3858, pruned_loss=0.1351, over 5648494.89 frames. ], batch size: 77, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:25:10,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.514e+03 2.058e+03 2.670e+03 4.853e+03, threshold=4.116e+03, percent-clipped=1.0
+2023-03-05 21:25:11,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:25:55,712 INFO [train.py:968] (0/2) Epoch 11, batch 29800, giga_loss[loss=0.3017, simple_loss=0.3691, pruned_loss=0.1172, over 28740.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3837, pruned_loss=0.1328, over 5644734.25 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3732, pruned_loss=0.1248, over 5678842.33 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3848, pruned_loss=0.134, over 5648746.11 frames. ], batch size: 284, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:26:05,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6713, 1.7987, 1.8819, 1.4633], device='cuda:0'), covar=tensor([0.1624, 0.2074, 0.1301, 0.1503], device='cuda:0'), in_proj_covar=tensor([0.0829, 0.0703, 0.0870, 0.0772], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 21:26:34,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:26:37,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:26:45,466 INFO [train.py:968] (0/2) Epoch 11, batch 29850, giga_loss[loss=0.3182, simple_loss=0.3783, pruned_loss=0.1291, over 28865.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3822, pruned_loss=0.132, over 5650344.58 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3731, pruned_loss=0.1249, over 5680234.57 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3833, pruned_loss=0.1329, over 5651970.93 frames. ], batch size: 186, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:26:46,205 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.398e+02 1.571e+03 2.043e+03 2.549e+03 4.645e+03, threshold=4.087e+03, percent-clipped=4.0
+2023-03-05 21:26:55,860 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:26:56,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:26:58,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:27:04,477 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:27:06,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:27:22,135 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:27:30,711 INFO [train.py:968] (0/2) Epoch 11, batch 29900, giga_loss[loss=0.4131, simple_loss=0.432, pruned_loss=0.1971, over 26653.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3815, pruned_loss=0.1314, over 5659496.76 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3734, pruned_loss=0.1251, over 5683498.82 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3823, pruned_loss=0.1321, over 5656951.36 frames. ], batch size: 555, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:27:33,315 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:27:42,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4962, 1.6335, 1.3456, 1.6540], device='cuda:0'), covar=tensor([0.2206, 0.2238, 0.2381, 0.2110], device='cuda:0'), in_proj_covar=tensor([0.1294, 0.0960, 0.1149, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 21:27:59,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1979, 1.4673, 1.2819, 1.0162], device='cuda:0'), covar=tensor([0.2224, 0.2000, 0.1219, 0.1708], device='cuda:0'), in_proj_covar=tensor([0.1719, 0.1620, 0.1585, 0.1684], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 21:28:17,111 INFO [train.py:968] (0/2) Epoch 11, batch 29950, giga_loss[loss=0.2648, simple_loss=0.3389, pruned_loss=0.09538, over 28910.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3774, pruned_loss=0.1289, over 5668247.88 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3733, pruned_loss=0.125, over 5686774.42 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3782, pruned_loss=0.1297, over 5663132.27 frames. ], batch size: 136, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:28:18,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.821e+03 2.368e+03 3.512e+03 8.881e+03, threshold=4.736e+03, percent-clipped=21.0
+2023-03-05 21:28:22,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:28:50,746 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:28:53,138 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:04,826 INFO [train.py:968] (0/2) Epoch 11, batch 30000, giga_loss[loss=0.2672, simple_loss=0.33, pruned_loss=0.1022, over 28611.00 frames. ], tot_loss[loss=0.315, simple_loss=0.374, pruned_loss=0.128, over 5653364.24 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3733, pruned_loss=0.1249, over 5689174.94 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3746, pruned_loss=0.1287, over 5646962.31 frames. ], batch size: 92, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:29:04,830 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 21:29:13,228 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2178, simple_loss=0.3253, pruned_loss=0.05515, over 944034.00 frames. 
+2023-03-05 21:29:13,229 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 21:29:16,108 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:18,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:25,660 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:42,102 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:49,367 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:54,326 INFO [train.py:968] (0/2) Epoch 11, batch 30050, giga_loss[loss=0.3057, simple_loss=0.3562, pruned_loss=0.1275, over 28832.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3723, pruned_loss=0.1273, over 5664928.79 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.373, pruned_loss=0.1246, over 5691822.93 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3732, pruned_loss=0.1282, over 5656021.79 frames. ], batch size: 99, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:29:55,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.699e+03 2.074e+03 2.865e+03 1.214e+04, threshold=4.147e+03, percent-clipped=4.0
+2023-03-05 21:30:27,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3798, 3.2229, 3.0351, 1.8635], device='cuda:0'), covar=tensor([0.0720, 0.0858, 0.0801, 0.1863], device='cuda:0'), in_proj_covar=tensor([0.1076, 0.1008, 0.0883, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 21:30:41,685 INFO [train.py:968] (0/2) Epoch 11, batch 30100, giga_loss[loss=0.335, simple_loss=0.3994, pruned_loss=0.1353, over 28577.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3714, pruned_loss=0.1272, over 5649043.72 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3727, pruned_loss=0.1245, over 5692730.08 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3723, pruned_loss=0.1281, over 5639725.57 frames. ], batch size: 336, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:30:42,648 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:30:45,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:30:56,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3671, 3.2015, 2.9835, 1.8820], device='cuda:0'), covar=tensor([0.0757, 0.0922, 0.0916, 0.1952], device='cuda:0'), in_proj_covar=tensor([0.1070, 0.1003, 0.0879, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 21:31:07,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:31:13,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:31:26,757 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 21:31:30,144 INFO [train.py:968] (0/2) Epoch 11, batch 30150, giga_loss[loss=0.326, simple_loss=0.3696, pruned_loss=0.1411, over 24055.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3714, pruned_loss=0.1257, over 5655301.47 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3722, pruned_loss=0.1243, over 5696822.83 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3725, pruned_loss=0.1267, over 5643458.68 frames. ], batch size: 705, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:31:33,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.212e+02 1.714e+03 2.353e+03 3.199e+03 7.278e+03, threshold=4.706e+03, percent-clipped=15.0
+2023-03-05 21:31:43,492 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-05 21:32:14,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4618, 1.6723, 1.7635, 1.2864], device='cuda:0'), covar=tensor([0.1713, 0.2503, 0.1457, 0.1709], device='cuda:0'), in_proj_covar=tensor([0.0829, 0.0699, 0.0869, 0.0771], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 21:32:20,343 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:32:27,187 INFO [train.py:968] (0/2) Epoch 11, batch 30200, giga_loss[loss=0.2567, simple_loss=0.3426, pruned_loss=0.08544, over 28884.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3688, pruned_loss=0.1219, over 5646397.17 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3721, pruned_loss=0.1243, over 5699810.89 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3698, pruned_loss=0.1227, over 5633679.15 frames. ], batch size: 174, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:32:40,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:33:16,189 INFO [train.py:968] (0/2) Epoch 11, batch 30250, giga_loss[loss=0.2771, simple_loss=0.361, pruned_loss=0.09659, over 28826.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.366, pruned_loss=0.1183, over 5657264.28 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3718, pruned_loss=0.1243, over 5700657.27 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3669, pruned_loss=0.1187, over 5645049.43 frames. ], batch size: 186, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:33:18,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.623e+02 1.338e+03 1.742e+03 2.337e+03 5.952e+03, threshold=3.484e+03, percent-clipped=4.0
+2023-03-05 21:33:18,516 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:33:36,885 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:33:40,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:34:06,377 INFO [train.py:968] (0/2) Epoch 11, batch 30300, giga_loss[loss=0.2811, simple_loss=0.355, pruned_loss=0.1037, over 28900.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.364, pruned_loss=0.1158, over 5658695.29 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3721, pruned_loss=0.1246, over 5703868.57 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3644, pruned_loss=0.1158, over 5645744.60 frames. ], batch size: 227, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:34:08,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:34:54,785 INFO [train.py:968] (0/2) Epoch 11, batch 30350, giga_loss[loss=0.2593, simple_loss=0.3404, pruned_loss=0.08905, over 28743.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3608, pruned_loss=0.1123, over 5654504.49 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.372, pruned_loss=0.1247, over 5697997.03 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.361, pruned_loss=0.112, over 5648287.81 frames. ], batch size: 99, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:34:56,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.491e+02 1.269e+03 1.856e+03 2.641e+03 9.825e+03, threshold=3.712e+03, percent-clipped=12.0
+2023-03-05 21:35:01,345 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:35:02,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4467, 2.1179, 1.5870, 0.6903], device='cuda:0'), covar=tensor([0.3796, 0.2009, 0.2983, 0.4213], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1476, 0.1488, 0.1269], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 21:35:04,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:35:22,995 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-05 21:35:34,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:35:49,452 INFO [train.py:968] (0/2) Epoch 11, batch 30400, giga_loss[loss=0.3169, simple_loss=0.3836, pruned_loss=0.1251, over 28948.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3594, pruned_loss=0.1086, over 5659025.53 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3718, pruned_loss=0.1246, over 5689737.54 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3597, pruned_loss=0.1084, over 5661288.60 frames. ], batch size: 213, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:36:09,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:36:37,196 INFO [train.py:968] (0/2) Epoch 11, batch 30450, giga_loss[loss=0.2239, simple_loss=0.3107, pruned_loss=0.06849, over 28507.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3584, pruned_loss=0.1084, over 5664256.29 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3711, pruned_loss=0.1245, over 5695184.06 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3588, pruned_loss=0.1077, over 5660030.13 frames. ], batch size: 71, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:36:39,867 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.082e+02 1.258e+03 2.005e+03 2.675e+03 7.066e+03, threshold=4.010e+03, percent-clipped=7.0
+2023-03-05 21:36:46,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2914, 0.9314, 1.0445, 1.4436], device='cuda:0'), covar=tensor([0.0776, 0.0335, 0.0348, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0112, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:0')
+2023-03-05 21:36:47,473 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-486000.pt
+2023-03-05 21:36:49,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1358, 1.5520, 1.4642, 1.0514], device='cuda:0'), covar=tensor([0.1662, 0.2457, 0.1405, 0.1589], device='cuda:0'), in_proj_covar=tensor([0.0825, 0.0692, 0.0862, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 21:36:55,739 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=486007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:37:05,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7435, 4.5585, 4.3004, 2.0798], device='cuda:0'), covar=tensor([0.0526, 0.0754, 0.0867, 0.2009], device='cuda:0'), in_proj_covar=tensor([0.1051, 0.0990, 0.0862, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 21:37:23,054 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=486035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:37:26,173 INFO [train.py:968] (0/2) Epoch 11, batch 30500, giga_loss[loss=0.2597, simple_loss=0.3411, pruned_loss=0.08919, over 28754.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3569, pruned_loss=0.1075, over 5659896.22 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.371, pruned_loss=0.1248, over 5687577.65 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.357, pruned_loss=0.1062, over 5663347.40 frames. ], batch size: 262, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:38:15,095 INFO [train.py:968] (0/2) Epoch 11, batch 30550, libri_loss[loss=0.3349, simple_loss=0.3807, pruned_loss=0.1445, over 19999.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.355, pruned_loss=0.1061, over 5653155.19 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3709, pruned_loss=0.1248, over 5680209.49 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.355, pruned_loss=0.1049, over 5663044.93 frames. ], batch size: 187, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:38:17,990 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.236e+02 1.498e+03 2.005e+03 2.936e+03 8.060e+03, threshold=4.011e+03, percent-clipped=13.0
+2023-03-05 21:38:26,289 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:38:30,611 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=486103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:38:35,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=486109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:38:53,799 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6800, 1.7341, 1.2313, 1.3851], device='cuda:0'), covar=tensor([0.0700, 0.0467, 0.0977, 0.0966], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0438, 0.0497, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 21:38:58,668 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=486132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:38:58,919 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-05 21:39:04,057 INFO [train.py:968] (0/2) Epoch 11, batch 30600, giga_loss[loss=0.3558, simple_loss=0.3956, pruned_loss=0.1579, over 26828.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3534, pruned_loss=0.1053, over 5643214.21 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.37, pruned_loss=0.1244, over 5677475.64 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3537, pruned_loss=0.1043, over 5653570.62 frames. ], batch size: 555, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:39:27,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=486166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:39:50,896 INFO [train.py:968] (0/2) Epoch 11, batch 30650, giga_loss[loss=0.2529, simple_loss=0.334, pruned_loss=0.08593, over 28832.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3538, pruned_loss=0.1052, over 5654160.98 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3696, pruned_loss=0.1242, over 5679424.43 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.354, pruned_loss=0.1041, over 5659835.87 frames. ], batch size: 92, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:39:53,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.785e+02 1.529e+03 1.980e+03 2.658e+03 5.585e+03, threshold=3.959e+03, percent-clipped=6.0
+2023-03-05 21:40:31,686 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=486231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:40:39,641 INFO [train.py:968] (0/2) Epoch 11, batch 30700, giga_loss[loss=0.2631, simple_loss=0.3422, pruned_loss=0.09199, over 28689.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3516, pruned_loss=0.1035, over 5645480.47 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3696, pruned_loss=0.1245, over 5673558.09 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3516, pruned_loss=0.1021, over 5654982.39 frames. ], batch size: 242, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:40:54,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:40:56,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=486255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:41:25,373 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=486284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:41:29,575 INFO [train.py:968] (0/2) Epoch 11, batch 30750, giga_loss[loss=0.2722, simple_loss=0.3524, pruned_loss=0.09593, over 28548.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3486, pruned_loss=0.1012, over 5645538.04 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3693, pruned_loss=0.1245, over 5668266.69 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3484, pruned_loss=0.09959, over 5658005.87 frames. ], batch size: 336, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:41:31,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.087e+02 1.261e+03 1.840e+03 2.634e+03 7.636e+03, threshold=3.680e+03, percent-clipped=6.0
+2023-03-05 21:41:48,893 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:41:51,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=486312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:42:18,807 INFO [train.py:968] (0/2) Epoch 11, batch 30800, libri_loss[loss=0.3016, simple_loss=0.3654, pruned_loss=0.1189, over 29535.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3454, pruned_loss=0.0997, over 5660853.79 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3694, pruned_loss=0.1247, over 5675112.37 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3446, pruned_loss=0.09757, over 5664151.20 frames. ], batch size: 81, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 21:42:20,356 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=486341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:42:57,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=486382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:43:04,719 INFO [train.py:968] (0/2) Epoch 11, batch 30850, giga_loss[loss=0.258, simple_loss=0.3338, pruned_loss=0.0911, over 28706.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3429, pruned_loss=0.0984, over 5655052.87 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3688, pruned_loss=0.1244, over 5669654.49 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3419, pruned_loss=0.09603, over 5662056.97 frames. ], batch size: 262, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 21:43:07,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.642e+02 1.293e+03 1.716e+03 2.436e+03 6.491e+03, threshold=3.432e+03, percent-clipped=6.0
+2023-03-05 21:43:26,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=486410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:43:55,685 INFO [train.py:968] (0/2) Epoch 11, batch 30900, giga_loss[loss=0.2253, simple_loss=0.3032, pruned_loss=0.07366, over 28865.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3429, pruned_loss=0.09876, over 5652736.70 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3685, pruned_loss=0.1243, over 5674355.41 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.342, pruned_loss=0.09655, over 5653683.85 frames. ], batch size: 106, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:44:26,938 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-05 21:44:51,959 INFO [train.py:968] (0/2) Epoch 11, batch 30950, libri_loss[loss=0.3191, simple_loss=0.3759, pruned_loss=0.1312, over 29498.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3453, pruned_loss=0.1001, over 5643059.12 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3682, pruned_loss=0.1243, over 5670276.61 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3442, pruned_loss=0.09785, over 5646175.66 frames. ], batch size: 82, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:44:57,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.691e+02 1.390e+03 1.841e+03 2.671e+03 5.919e+03, threshold=3.682e+03, percent-clipped=17.0
+2023-03-05 21:45:18,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2274, 1.4858, 1.3318, 1.4586], device='cuda:0'), covar=tensor([0.0726, 0.0391, 0.0333, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0057, 0.0052, 0.0089], device='cuda:0')
+2023-03-05 21:45:21,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7746, 1.1267, 2.8449, 2.6771], device='cuda:0'), covar=tensor([0.1572, 0.2380, 0.0521, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.0660, 0.0586, 0.0850, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 21:45:34,140 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:45:36,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=486528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:45:48,939 INFO [train.py:968] (0/2) Epoch 11, batch 31000, giga_loss[loss=0.2429, simple_loss=0.3115, pruned_loss=0.08711, over 24434.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3473, pruned_loss=0.1005, over 5634343.87 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.368, pruned_loss=0.1242, over 5671652.39 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3461, pruned_loss=0.09826, over 5635039.77 frames. ], batch size: 705, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:46:07,199 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:46:09,793 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=486556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:46:11,030 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=486557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:46:46,190 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=486585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:46:50,865 INFO [train.py:968] (0/2) Epoch 11, batch 31050, libri_loss[loss=0.3301, simple_loss=0.3765, pruned_loss=0.1418, over 29529.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3472, pruned_loss=0.1006, over 5617185.36 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3677, pruned_loss=0.1241, over 5657119.72 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3461, pruned_loss=0.09834, over 5628484.78 frames. ], batch size: 83, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:46:57,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.881e+02 1.321e+03 1.645e+03 2.407e+03 6.417e+03, threshold=3.289e+03, percent-clipped=6.0
+2023-03-05 21:47:13,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=486606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:47:55,317 INFO [train.py:968] (0/2) Epoch 11, batch 31100, giga_loss[loss=0.2463, simple_loss=0.3304, pruned_loss=0.08111, over 28927.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3458, pruned_loss=0.09956, over 5630446.04 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3675, pruned_loss=0.124, over 5660694.40 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3448, pruned_loss=0.0975, over 5635386.55 frames. ], batch size: 213, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:48:53,236 INFO [train.py:968] (0/2) Epoch 11, batch 31150, giga_loss[loss=0.2417, simple_loss=0.3284, pruned_loss=0.07745, over 27597.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3441, pruned_loss=0.09781, over 5636262.95 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3672, pruned_loss=0.1241, over 5668186.17 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3428, pruned_loss=0.09509, over 5632263.06 frames. ], batch size: 472, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:48:55,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.657e+02 1.328e+03 1.798e+03 2.700e+03 9.247e+03, threshold=3.596e+03, percent-clipped=12.0
+2023-03-05 21:49:53,321 INFO [train.py:968] (0/2) Epoch 11, batch 31200, giga_loss[loss=0.2136, simple_loss=0.2851, pruned_loss=0.07108, over 24446.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3414, pruned_loss=0.09507, over 5642793.81 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3668, pruned_loss=0.124, over 5674616.03 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3402, pruned_loss=0.09246, over 5633746.87 frames. ], batch size: 705, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 21:50:07,396 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:50:10,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=486752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:50:43,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2843, 1.3150, 1.2293, 1.4868], device='cuda:0'), covar=tensor([0.0668, 0.0428, 0.0335, 0.0720], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0112, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0057, 0.0052, 0.0089], device='cuda:0')
+2023-03-05 21:50:46,007 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=486781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:50:52,488 INFO [train.py:968] (0/2) Epoch 11, batch 31250, giga_loss[loss=0.2253, simple_loss=0.3023, pruned_loss=0.07416, over 27542.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3384, pruned_loss=0.0943, over 5647692.38 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3661, pruned_loss=0.1237, over 5669109.70 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3374, pruned_loss=0.0918, over 5645421.21 frames. ], batch size: 472, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:51:01,978 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.783e+02 1.188e+03 1.472e+03 2.010e+03 3.757e+03, threshold=2.945e+03, percent-clipped=3.0
+2023-03-05 21:51:21,382 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=486810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:51:32,705 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9586, 1.1788, 1.1307, 0.9002], device='cuda:0'), covar=tensor([0.1465, 0.1624, 0.0870, 0.1249], device='cuda:0'), in_proj_covar=tensor([0.1672, 0.1574, 0.1528, 0.1625], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 21:51:53,164 INFO [train.py:968] (0/2) Epoch 11, batch 31300, giga_loss[loss=0.279, simple_loss=0.3513, pruned_loss=0.1034, over 28941.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3386, pruned_loss=0.09469, over 5654860.65 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3657, pruned_loss=0.1236, over 5664283.91 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3376, pruned_loss=0.09222, over 5657209.98 frames. ], batch size: 199, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:52:49,098 INFO [train.py:968] (0/2) Epoch 11, batch 31350, giga_loss[loss=0.2452, simple_loss=0.3329, pruned_loss=0.07875, over 28158.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3393, pruned_loss=0.09499, over 5666405.72 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3655, pruned_loss=0.1236, over 5670616.50 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3381, pruned_loss=0.09231, over 5662420.13 frames. ], batch size: 412, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:52:55,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.765e+02 1.440e+03 1.786e+03 2.410e+03 6.965e+03, threshold=3.573e+03, percent-clipped=10.0
+2023-03-05 21:53:47,751 INFO [train.py:968] (0/2) Epoch 11, batch 31400, giga_loss[loss=0.2901, simple_loss=0.3691, pruned_loss=0.1055, over 28865.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3396, pruned_loss=0.09437, over 5657488.01 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.365, pruned_loss=0.1233, over 5670838.72 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3385, pruned_loss=0.09187, over 5653712.79 frames. ], batch size: 227, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:54:49,783 INFO [train.py:968] (0/2) Epoch 11, batch 31450, giga_loss[loss=0.2767, simple_loss=0.3564, pruned_loss=0.0985, over 28715.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.34, pruned_loss=0.0943, over 5663360.41 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3643, pruned_loss=0.123, over 5676530.06 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3389, pruned_loss=0.0916, over 5654617.46 frames. ], batch size: 262, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:54:50,637 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2124, 1.5975, 1.4721, 1.1584], device='cuda:0'), covar=tensor([0.1371, 0.2030, 0.1115, 0.1413], device='cuda:0'), in_proj_covar=tensor([0.0828, 0.0687, 0.0866, 0.0771], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 21:54:56,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.438e+02 1.329e+03 1.644e+03 2.368e+03 5.447e+03, threshold=3.289e+03, percent-clipped=6.0
+2023-03-05 21:54:58,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7502, 2.2323, 1.8246, 1.3296], device='cuda:0'), covar=tensor([0.2391, 0.1630, 0.1962, 0.2731], device='cuda:0'), in_proj_covar=tensor([0.1546, 0.1468, 0.1478, 0.1262], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 21:55:36,685 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=487029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:55:48,486 INFO [train.py:968] (0/2) Epoch 11, batch 31500, giga_loss[loss=0.2966, simple_loss=0.3562, pruned_loss=0.1185, over 26751.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3385, pruned_loss=0.09398, over 5674586.82 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3636, pruned_loss=0.1228, over 5676829.84 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3372, pruned_loss=0.09075, over 5666236.79 frames. ], batch size: 555, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:56:02,743 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-05 21:56:07,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8635, 1.9676, 1.2858, 1.5311], device='cuda:0'), covar=tensor([0.0711, 0.0491, 0.1012, 0.0910], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0438, 0.0501, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 21:56:48,470 INFO [train.py:968] (0/2) Epoch 11, batch 31550, libri_loss[loss=0.3229, simple_loss=0.3776, pruned_loss=0.1341, over 29517.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3405, pruned_loss=0.09607, over 5668155.72 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3632, pruned_loss=0.1226, over 5678065.94 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3388, pruned_loss=0.09251, over 5660314.15 frames. ], batch size: 89, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:56:53,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.405e+02 1.495e+03 1.916e+03 2.796e+03 4.332e+03, threshold=3.833e+03, percent-clipped=13.0
+2023-03-05 21:57:10,700 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=487106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:57:47,502 INFO [train.py:968] (0/2) Epoch 11, batch 31600, libri_loss[loss=0.3196, simple_loss=0.3599, pruned_loss=0.1397, over 29612.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3443, pruned_loss=0.09615, over 5674847.38 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3629, pruned_loss=0.1225, over 5679983.62 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3428, pruned_loss=0.09279, over 5666543.72 frames. ], batch size: 75, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 21:58:13,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8061, 1.1987, 2.8464, 2.6616], device='cuda:0'), covar=tensor([0.1635, 0.2464, 0.0514, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0662, 0.0593, 0.0856, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 21:58:31,185 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3204, 1.4698, 1.3166, 1.3310], device='cuda:0'), covar=tensor([0.1849, 0.1648, 0.1464, 0.1536], device='cuda:0'), in_proj_covar=tensor([0.1670, 0.1571, 0.1527, 0.1623], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 21:58:44,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=487185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:58:50,321 INFO [train.py:968] (0/2) Epoch 11, batch 31650, giga_loss[loss=0.2737, simple_loss=0.3427, pruned_loss=0.1023, over 26908.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3458, pruned_loss=0.09495, over 5667792.07 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3626, pruned_loss=0.1224, over 5681616.63 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3444, pruned_loss=0.09172, over 5659446.07 frames. ], batch size: 555, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 21:58:55,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.622e+02 1.417e+03 1.853e+03 2.863e+03 8.837e+03, threshold=3.707e+03, percent-clipped=10.0
+2023-03-05 21:59:12,521 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-05 21:59:26,980 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=487222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:59:36,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2373, 1.7492, 1.2758, 0.4507], device='cuda:0'), covar=tensor([0.3006, 0.1817, 0.3130, 0.3752], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1491, 0.1493, 0.1275], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 21:59:46,229 INFO [train.py:968] (0/2) Epoch 11, batch 31700, giga_loss[loss=0.249, simple_loss=0.3416, pruned_loss=0.07817, over 28923.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3466, pruned_loss=0.09427, over 5659626.97 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3628, pruned_loss=0.1227, over 5675988.02 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3449, pruned_loss=0.09072, over 5658171.89 frames. ], batch size: 112, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:00:44,488 INFO [train.py:968] (0/2) Epoch 11, batch 31750, giga_loss[loss=0.2902, simple_loss=0.3645, pruned_loss=0.1079, over 28791.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3449, pruned_loss=0.09261, over 5670676.34 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3626, pruned_loss=0.1226, over 5680785.47 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3433, pruned_loss=0.089, over 5664999.12 frames. ], batch size: 263, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:00:49,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.612e+02 1.355e+03 1.655e+03 2.319e+03 6.113e+03, threshold=3.310e+03, percent-clipped=4.0
+2023-03-05 22:01:31,544 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:01:34,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:01:41,264 INFO [train.py:968] (0/2) Epoch 11, batch 31800, giga_loss[loss=0.2322, simple_loss=0.3147, pruned_loss=0.07484, over 29021.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3451, pruned_loss=0.09359, over 5673255.09 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3623, pruned_loss=0.1225, over 5674281.53 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3435, pruned_loss=0.08997, over 5674146.89 frames. ], batch size: 93, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:02:08,530 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=487360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:02:16,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4063, 2.1513, 1.4867, 0.5337], device='cuda:0'), covar=tensor([0.3274, 0.1988, 0.3355, 0.3880], device='cuda:0'), in_proj_covar=tensor([0.1558, 0.1482, 0.1489, 0.1271], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 22:02:52,493 INFO [train.py:968] (0/2) Epoch 11, batch 31850, giga_loss[loss=0.2303, simple_loss=0.315, pruned_loss=0.07277, over 28774.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3437, pruned_loss=0.0939, over 5675185.96 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3626, pruned_loss=0.1227, over 5678592.05 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.342, pruned_loss=0.09034, over 5672276.70 frames. ], batch size: 99, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:02:59,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.753e+02 1.252e+03 1.618e+03 2.007e+03 4.648e+03, threshold=3.236e+03, percent-clipped=4.0
+2023-03-05 22:03:17,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=487404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:03:31,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2181, 1.5302, 1.3372, 1.0575], device='cuda:0'), covar=tensor([0.1936, 0.1472, 0.0929, 0.1475], device='cuda:0'), in_proj_covar=tensor([0.1662, 0.1564, 0.1516, 0.1616], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 22:04:11,608 INFO [train.py:968] (0/2) Epoch 11, batch 31900, giga_loss[loss=0.2178, simple_loss=0.3056, pruned_loss=0.06496, over 29056.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3438, pruned_loss=0.09453, over 5678909.88 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3621, pruned_loss=0.1225, over 5680948.34 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3425, pruned_loss=0.09151, over 5674363.69 frames. ], batch size: 214, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:04:31,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=487451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:04:57,109 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=487470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:05:11,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=487481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:05:21,659 INFO [train.py:968] (0/2) Epoch 11, batch 31950, giga_loss[loss=0.235, simple_loss=0.3203, pruned_loss=0.07482, over 28399.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3385, pruned_loss=0.0915, over 5679914.62 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3613, pruned_loss=0.122, over 5686024.86 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3377, pruned_loss=0.08884, over 5671564.81 frames. ], batch size: 368, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:05:28,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.437e+02 1.240e+03 1.697e+03 2.506e+03 8.302e+03, threshold=3.394e+03, percent-clipped=13.0
+2023-03-05 22:05:36,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4521, 1.7968, 1.3813, 1.7347], device='cuda:0'), covar=tensor([0.2315, 0.2175, 0.2487, 0.2029], device='cuda:0'), in_proj_covar=tensor([0.1293, 0.0956, 0.1154, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 22:06:24,302 INFO [train.py:968] (0/2) Epoch 11, batch 32000, giga_loss[loss=0.249, simple_loss=0.3307, pruned_loss=0.08368, over 28370.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3369, pruned_loss=0.09044, over 5677119.19 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3615, pruned_loss=0.1221, over 5686903.44 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3357, pruned_loss=0.08764, over 5669862.51 frames. ], batch size: 368, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:06:36,320 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:06:39,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:07:12,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=487579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:07:23,950 INFO [train.py:968] (0/2) Epoch 11, batch 32050, giga_loss[loss=0.2856, simple_loss=0.3644, pruned_loss=0.1035, over 28836.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3353, pruned_loss=0.09021, over 5678053.02 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3607, pruned_loss=0.1216, over 5684160.19 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3341, pruned_loss=0.08722, over 5674935.89 frames. ], batch size: 243, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:07:35,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.477e+02 1.393e+03 1.805e+03 2.459e+03 4.764e+03, threshold=3.610e+03, percent-clipped=11.0
+2023-03-05 22:07:36,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=487597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:08:12,989 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:08:16,986 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:08:27,213 INFO [train.py:968] (0/2) Epoch 11, batch 32100, giga_loss[loss=0.2747, simple_loss=0.3542, pruned_loss=0.09758, over 29106.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3404, pruned_loss=0.09306, over 5679495.08 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3603, pruned_loss=0.1215, over 5685091.50 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3395, pruned_loss=0.09028, over 5676020.72 frames. ], batch size: 200, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:08:46,880 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=487656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:09:24,552 INFO [train.py:968] (0/2) Epoch 11, batch 32150, giga_loss[loss=0.2675, simple_loss=0.339, pruned_loss=0.09796, over 29015.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3422, pruned_loss=0.09568, over 5669341.59 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3609, pruned_loss=0.1221, over 5663238.22 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3404, pruned_loss=0.09215, over 5685464.07 frames. ], batch size: 285, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:09:34,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.847e+02 1.626e+03 1.976e+03 3.137e+03 5.206e+03, threshold=3.953e+03, percent-clipped=16.0
+2023-03-05 22:10:26,449 INFO [train.py:968] (0/2) Epoch 11, batch 32200, giga_loss[loss=0.3061, simple_loss=0.3719, pruned_loss=0.1202, over 28926.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3416, pruned_loss=0.09615, over 5671034.24 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3609, pruned_loss=0.1221, over 5667184.41 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3396, pruned_loss=0.09263, over 5680528.26 frames. ], batch size: 284, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:10:27,910 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:10:30,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:11:07,829 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=487772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:11:17,913 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 22:11:26,427 INFO [train.py:968] (0/2) Epoch 11, batch 32250, giga_loss[loss=0.2371, simple_loss=0.3234, pruned_loss=0.07543, over 28992.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3414, pruned_loss=0.09666, over 5674083.94 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3603, pruned_loss=0.1217, over 5671881.73 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3401, pruned_loss=0.09372, over 5677216.21 frames. ], batch size: 155, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:11:32,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.497e+02 1.432e+03 1.961e+03 2.750e+03 5.254e+03, threshold=3.921e+03, percent-clipped=8.0
+2023-03-05 22:12:12,147 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=487826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:12:29,952 INFO [train.py:968] (0/2) Epoch 11, batch 32300, giga_loss[loss=0.282, simple_loss=0.3666, pruned_loss=0.09873, over 29024.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3421, pruned_loss=0.09621, over 5674595.65 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3604, pruned_loss=0.122, over 5674085.02 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3404, pruned_loss=0.09281, over 5675412.80 frames. ], batch size: 199, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:12:37,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=487845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:13:11,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-05 22:13:39,863 INFO [train.py:968] (0/2) Epoch 11, batch 32350, giga_loss[loss=0.27, simple_loss=0.3438, pruned_loss=0.0981, over 28705.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3435, pruned_loss=0.09636, over 5666152.01 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3607, pruned_loss=0.1222, over 5672084.56 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3413, pruned_loss=0.09262, over 5668796.12 frames. ], batch size: 262, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:13:51,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.474e+02 1.475e+03 2.164e+03 3.077e+03 6.661e+03, threshold=4.328e+03, percent-clipped=13.0
+2023-03-05 22:13:59,913 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4531, 1.6999, 1.3761, 1.6892], device='cuda:0'), covar=tensor([0.2513, 0.2411, 0.2673, 0.2039], device='cuda:0'), in_proj_covar=tensor([0.1293, 0.0955, 0.1150, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 22:14:56,602 INFO [train.py:968] (0/2) Epoch 11, batch 32400, giga_loss[loss=0.2374, simple_loss=0.3197, pruned_loss=0.07757, over 28077.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3423, pruned_loss=0.09573, over 5656288.55 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3607, pruned_loss=0.1222, over 5663839.46 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3404, pruned_loss=0.09245, over 5664901.26 frames. ], batch size: 412, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:15:24,457 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=487962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:15:33,372 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:15:37,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:15:56,822 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:15:57,239 INFO [train.py:968] (0/2) Epoch 11, batch 32450, giga_loss[loss=0.2136, simple_loss=0.2972, pruned_loss=0.06502, over 29023.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3387, pruned_loss=0.0952, over 5668795.59 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3601, pruned_loss=0.1219, over 5670120.31 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3368, pruned_loss=0.0917, over 5669768.23 frames. ], batch size: 155, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:15:59,164 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:16:02,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4921, 1.8349, 1.7185, 1.5739], device='cuda:0'), covar=tensor([0.1423, 0.1794, 0.1828, 0.1734], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0719, 0.0657, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-05 22:16:05,763 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.116e+02 1.314e+03 1.700e+03 2.399e+03 5.513e+03, threshold=3.400e+03, percent-clipped=4.0
+2023-03-05 22:16:08,492 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-488000.pt
+2023-03-05 22:16:09,877 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=488001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:16:38,915 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=488020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:17:02,390 INFO [train.py:968] (0/2) Epoch 11, batch 32500, giga_loss[loss=0.2122, simple_loss=0.2791, pruned_loss=0.07262, over 24046.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3332, pruned_loss=0.0925, over 5670448.59 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3599, pruned_loss=0.1217, over 5672750.65 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3315, pruned_loss=0.08945, over 5668879.84 frames. ], batch size: 705, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:18:01,275 INFO [train.py:968] (0/2) Epoch 11, batch 32550, giga_loss[loss=0.2629, simple_loss=0.3387, pruned_loss=0.09356, over 28980.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.333, pruned_loss=0.09232, over 5672054.16 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3593, pruned_loss=0.1214, over 5678797.86 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3316, pruned_loss=0.08953, over 5665293.52 frames. ], batch size: 128, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:18:13,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.364e+02 1.452e+03 1.915e+03 2.718e+03 7.536e+03, threshold=3.829e+03, percent-clipped=13.0
+2023-03-05 22:18:26,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 22:18:26,699 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4138, 1.7301, 1.3500, 1.6841], device='cuda:0'), covar=tensor([0.2259, 0.2156, 0.2469, 0.1840], device='cuda:0'), in_proj_covar=tensor([0.1290, 0.0954, 0.1146, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 22:19:01,412 INFO [train.py:968] (0/2) Epoch 11, batch 32600, giga_loss[loss=0.2861, simple_loss=0.3525, pruned_loss=0.1099, over 27609.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3347, pruned_loss=0.09321, over 5679574.77 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3596, pruned_loss=0.1217, over 5682187.74 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3329, pruned_loss=0.09032, over 5671076.85 frames. ], batch size: 472, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:20:00,252 INFO [train.py:968] (0/2) Epoch 11, batch 32650, giga_loss[loss=0.2466, simple_loss=0.3359, pruned_loss=0.07861, over 28760.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3326, pruned_loss=0.09103, over 5670908.80 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3594, pruned_loss=0.1216, over 5679111.56 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3309, pruned_loss=0.08822, over 5666411.92 frames. ], batch size: 243, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:20:11,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.704e+02 1.292e+03 1.680e+03 2.643e+03 5.285e+03, threshold=3.359e+03, percent-clipped=5.0
+2023-03-05 22:21:05,383 INFO [train.py:968] (0/2) Epoch 11, batch 32700, giga_loss[loss=0.2094, simple_loss=0.2976, pruned_loss=0.06064, over 28452.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3321, pruned_loss=0.09087, over 5663659.83 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3594, pruned_loss=0.1217, over 5678890.19 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3302, pruned_loss=0.08788, over 5659889.69 frames. ], batch size: 71, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:22:06,000 INFO [train.py:968] (0/2) Epoch 11, batch 32750, giga_loss[loss=0.2423, simple_loss=0.3172, pruned_loss=0.08371, over 28627.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3306, pruned_loss=0.09074, over 5668638.51 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3583, pruned_loss=0.1213, over 5683277.21 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3287, pruned_loss=0.08739, over 5661440.57 frames. ], batch size: 78, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:22:19,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.273e+02 1.237e+03 1.702e+03 2.834e+03 8.744e+03, threshold=3.403e+03, percent-clipped=13.0
+2023-03-05 22:23:10,354 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=488337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:23:13,076 INFO [train.py:968] (0/2) Epoch 11, batch 32800, giga_loss[loss=0.2311, simple_loss=0.3193, pruned_loss=0.07147, over 29032.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.331, pruned_loss=0.08962, over 5681917.28 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3586, pruned_loss=0.1214, over 5686547.95 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3289, pruned_loss=0.08637, over 5673132.08 frames. ], batch size: 136, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:24:20,542 INFO [train.py:968] (0/2) Epoch 11, batch 32850, giga_loss[loss=0.2797, simple_loss=0.3476, pruned_loss=0.1059, over 28394.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3318, pruned_loss=0.09012, over 5676436.39 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3584, pruned_loss=0.1213, over 5680634.62 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.33, pruned_loss=0.08731, over 5674123.92 frames. ], batch size: 369, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:24:29,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.048e+02 1.169e+03 1.458e+03 1.961e+03 5.496e+03, threshold=2.917e+03, percent-clipped=7.0
+2023-03-05 22:25:08,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3986, 1.7185, 1.4132, 1.4106], device='cuda:0'), covar=tensor([0.2079, 0.1871, 0.1918, 0.1643], device='cuda:0'), in_proj_covar=tensor([0.1297, 0.0958, 0.1153, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 22:25:16,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3237, 1.4502, 1.2597, 1.3130], device='cuda:0'), covar=tensor([0.1717, 0.1439, 0.1446, 0.1427], device='cuda:0'), in_proj_covar=tensor([0.1687, 0.1564, 0.1527, 0.1638], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 22:25:23,613 INFO [train.py:968] (0/2) Epoch 11, batch 32900, giga_loss[loss=0.2397, simple_loss=0.3253, pruned_loss=0.07703, over 28587.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3334, pruned_loss=0.09173, over 5679124.67 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3588, pruned_loss=0.1216, over 5681982.62 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3312, pruned_loss=0.08886, over 5676004.92 frames. ], batch size: 307, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:26:10,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5157, 1.5495, 1.2730, 1.1523], device='cuda:0'), covar=tensor([0.0689, 0.0435, 0.0855, 0.1023], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0434, 0.0498, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 22:26:11,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=488480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:26:14,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4982, 3.5105, 1.5867, 1.5173], device='cuda:0'), covar=tensor([0.0887, 0.0271, 0.0859, 0.1259], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0506, 0.0341, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 22:26:15,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=488483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:26:19,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=488486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:26:22,889 INFO [train.py:968] (0/2) Epoch 11, batch 32950, giga_loss[loss=0.228, simple_loss=0.3149, pruned_loss=0.07061, over 28057.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.332, pruned_loss=0.09036, over 5676267.01 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3581, pruned_loss=0.1213, over 5687106.57 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3302, pruned_loss=0.08749, over 5669115.76 frames. ], batch size: 412, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:26:33,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.315e+02 1.340e+03 1.657e+03 2.370e+03 5.292e+03, threshold=3.315e+03, percent-clipped=14.0
+2023-03-05 22:26:49,055 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=488512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:27:00,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3081, 3.1947, 1.5066, 1.3939], device='cuda:0'), covar=tensor([0.0932, 0.0300, 0.0865, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0507, 0.0342, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 22:27:18,048 INFO [train.py:968] (0/2) Epoch 11, batch 33000, giga_loss[loss=0.2528, simple_loss=0.3365, pruned_loss=0.08452, over 28902.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3338, pruned_loss=0.09027, over 5665951.12 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3576, pruned_loss=0.121, over 5684677.69 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3321, pruned_loss=0.08739, over 5661812.88 frames. ], batch size: 227, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:27:18,053 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 22:27:25,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3856, 1.7196, 1.3606, 1.3766], device='cuda:0'), covar=tensor([0.2510, 0.2399, 0.2481, 0.2066], device='cuda:0'), in_proj_covar=tensor([0.1294, 0.0959, 0.1150, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 22:27:26,905 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.202, simple_loss=0.3022, pruned_loss=0.05088, over 944034.00 frames. 
+2023-03-05 22:27:26,906 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 22:28:23,439 INFO [train.py:968] (0/2) Epoch 11, batch 33050, giga_loss[loss=0.2788, simple_loss=0.3556, pruned_loss=0.101, over 28903.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3376, pruned_loss=0.0923, over 5671353.64 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3577, pruned_loss=0.1212, over 5690750.43 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3356, pruned_loss=0.08892, over 5661784.11 frames. ], batch size: 213, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:28:34,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.348e+02 1.460e+03 2.157e+03 3.220e+03 9.187e+03, threshold=4.314e+03, percent-clipped=20.0
+2023-03-05 22:28:44,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4336, 1.6608, 1.3687, 1.4738], device='cuda:0'), covar=tensor([0.0780, 0.0304, 0.0327, 0.0850], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0058, 0.0053, 0.0090], device='cuda:0')
+2023-03-05 22:29:07,159 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=488624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:29:22,140 INFO [train.py:968] (0/2) Epoch 11, batch 33100, giga_loss[loss=0.2555, simple_loss=0.3379, pruned_loss=0.08659, over 28671.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3388, pruned_loss=0.09272, over 5671486.75 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.358, pruned_loss=0.1213, over 5687188.22 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3364, pruned_loss=0.08913, over 5666179.70 frames. ], batch size: 307, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:30:11,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3467, 3.4025, 1.4315, 1.4952], device='cuda:0'), covar=tensor([0.0972, 0.0422, 0.0924, 0.1374], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0508, 0.0343, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-05 22:30:12,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6401, 3.8275, 1.6545, 1.6859], device='cuda:0'), covar=tensor([0.0887, 0.0394, 0.0905, 0.1306], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0509, 0.0343, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-05 22:30:27,983 INFO [train.py:968] (0/2) Epoch 11, batch 33150, giga_loss[loss=0.255, simple_loss=0.3326, pruned_loss=0.08872, over 28904.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3385, pruned_loss=0.09283, over 5674795.93 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3579, pruned_loss=0.1214, over 5691787.56 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3363, pruned_loss=0.08938, over 5665997.15 frames. ], batch size: 186, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:30:30,370 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=488690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:30:37,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.113e+02 1.488e+03 1.840e+03 2.614e+03 6.614e+03, threshold=3.680e+03, percent-clipped=5.0
+2023-03-05 22:30:51,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2616, 1.4716, 1.2781, 1.4311], device='cuda:0'), covar=tensor([0.0772, 0.0337, 0.0343, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:0')
+2023-03-05 22:31:02,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6955, 1.9354, 1.5382, 2.2836], device='cuda:0'), covar=tensor([0.2429, 0.2457, 0.2691, 0.2169], device='cuda:0'), in_proj_covar=tensor([0.1297, 0.0959, 0.1155, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 22:31:13,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5193, 1.8128, 1.7882, 1.3417], device='cuda:0'), covar=tensor([0.1851, 0.2416, 0.1517, 0.1699], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0682, 0.0863, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 22:31:19,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7374, 2.2306, 1.7496, 1.6186], device='cuda:0'), covar=tensor([0.2318, 0.1594, 0.1689, 0.1656], device='cuda:0'), in_proj_covar=tensor([0.1694, 0.1573, 0.1532, 0.1645], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 22:31:23,242 INFO [train.py:968] (0/2) Epoch 11, batch 33200, giga_loss[loss=0.2407, simple_loss=0.3276, pruned_loss=0.07695, over 28441.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3349, pruned_loss=0.09, over 5672666.69 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.358, pruned_loss=0.1214, over 5686565.41 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3325, pruned_loss=0.08654, over 5670608.65 frames. ], batch size: 370, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:31:44,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6776, 3.5076, 3.2933, 1.6077], device='cuda:0'), covar=tensor([0.0775, 0.0885, 0.0876, 0.2498], device='cuda:0'), in_proj_covar=tensor([0.1029, 0.0966, 0.0839, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 22:32:17,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=488786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:32:18,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5692, 2.0407, 1.6769, 1.6145], device='cuda:0'), covar=tensor([0.0750, 0.0250, 0.0298, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0090], device='cuda:0')
+2023-03-05 22:32:20,723 INFO [train.py:968] (0/2) Epoch 11, batch 33250, giga_loss[loss=0.2489, simple_loss=0.3276, pruned_loss=0.08515, over 28165.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3336, pruned_loss=0.08939, over 5674978.43 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3575, pruned_loss=0.121, over 5691117.83 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3313, pruned_loss=0.08598, over 5668936.16 frames. ], batch size: 412, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:32:32,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.811e+02 1.270e+03 1.784e+03 2.551e+03 8.288e+03, threshold=3.567e+03, percent-clipped=11.0
+2023-03-05 22:33:20,418 INFO [train.py:968] (0/2) Epoch 11, batch 33300, giga_loss[loss=0.2498, simple_loss=0.3306, pruned_loss=0.08445, over 28752.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3325, pruned_loss=0.08948, over 5660364.64 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3574, pruned_loss=0.121, over 5675166.17 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3303, pruned_loss=0.08624, over 5670884.27 frames. ], batch size: 307, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:33:40,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=488861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:34:17,529 INFO [train.py:968] (0/2) Epoch 11, batch 33350, giga_loss[loss=0.249, simple_loss=0.3338, pruned_loss=0.08208, over 28910.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3336, pruned_loss=0.0897, over 5661577.90 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3576, pruned_loss=0.1209, over 5672557.11 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.331, pruned_loss=0.0862, over 5671188.47 frames. ], batch size: 227, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:34:29,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.473e+02 1.242e+03 1.480e+03 1.934e+03 6.575e+03, threshold=2.959e+03, percent-clipped=3.0
+2023-03-05 22:34:57,578 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=488919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:35:20,385 INFO [train.py:968] (0/2) Epoch 11, batch 33400, giga_loss[loss=0.2997, simple_loss=0.3695, pruned_loss=0.115, over 28641.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3353, pruned_loss=0.09073, over 5667499.91 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3572, pruned_loss=0.1205, over 5677100.71 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3332, pruned_loss=0.08771, over 5670867.97 frames. ], batch size: 307, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:36:21,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3820, 1.8245, 1.3203, 0.8197], device='cuda:0'), covar=tensor([0.4363, 0.2603, 0.2438, 0.4034], device='cuda:0'), in_proj_covar=tensor([0.1568, 0.1500, 0.1502, 0.1283], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 22:36:25,964 INFO [train.py:968] (0/2) Epoch 11, batch 33450, giga_loss[loss=0.2936, simple_loss=0.3662, pruned_loss=0.1105, over 28126.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3355, pruned_loss=0.09152, over 5667055.35 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3564, pruned_loss=0.1201, over 5684006.57 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3339, pruned_loss=0.0887, over 5663290.83 frames. ], batch size: 412, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:36:37,483 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-05 22:36:41,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.489e+02 1.454e+03 1.980e+03 2.680e+03 1.082e+04, threshold=3.961e+03, percent-clipped=21.0
+2023-03-05 22:36:41,695 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=488999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:36:49,820 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:36:53,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:37:30,204 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:37:32,039 INFO [train.py:968] (0/2) Epoch 11, batch 33500, libri_loss[loss=0.2753, simple_loss=0.337, pruned_loss=0.1068, over 29554.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3394, pruned_loss=0.0939, over 5661204.79 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3558, pruned_loss=0.1197, over 5688589.69 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3384, pruned_loss=0.09149, over 5653394.41 frames. ], batch size: 76, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:37:58,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=489065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:38:26,111 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3376, 1.7095, 1.3666, 1.4771], device='cuda:0'), covar=tensor([0.0762, 0.0298, 0.0331, 0.0869], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0116, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:0')
+2023-03-05 22:38:28,528 INFO [train.py:968] (0/2) Epoch 11, batch 33550, giga_loss[loss=0.2817, simple_loss=0.3604, pruned_loss=0.1015, over 28134.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3417, pruned_loss=0.09422, over 5667214.80 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3561, pruned_loss=0.1199, over 5691495.15 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3404, pruned_loss=0.09167, over 5658032.21 frames. ], batch size: 412, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:38:38,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.182e+02 1.470e+03 1.997e+03 2.754e+03 7.283e+03, threshold=3.994e+03, percent-clipped=8.0
+2023-03-05 22:39:17,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8010, 2.0870, 1.6746, 2.3226], device='cuda:0'), covar=tensor([0.2328, 0.2223, 0.2533, 0.2069], device='cuda:0'), in_proj_covar=tensor([0.1296, 0.0959, 0.1153, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 22:39:28,923 INFO [train.py:968] (0/2) Epoch 11, batch 33600, libri_loss[loss=0.3572, simple_loss=0.3914, pruned_loss=0.1615, over 18835.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3425, pruned_loss=0.09522, over 5658071.82 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3559, pruned_loss=0.1199, over 5679774.15 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.341, pruned_loss=0.09215, over 5660445.26 frames. ], batch size: 187, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:39:33,762 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:39:41,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:39:55,514 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=489156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:40:01,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=489161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:40:19,688 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:40:38,416 INFO [train.py:968] (0/2) Epoch 11, batch 33650, libri_loss[loss=0.2294, simple_loss=0.298, pruned_loss=0.08041, over 29383.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3393, pruned_loss=0.09347, over 5666700.10 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3556, pruned_loss=0.1198, over 5683121.87 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3381, pruned_loss=0.09052, over 5665030.32 frames. ], batch size: 71, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:40:50,379 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.982e+02 1.278e+03 1.609e+03 2.398e+03 9.404e+03, threshold=3.218e+03, percent-clipped=7.0
+2023-03-05 22:41:03,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:41:07,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:41:25,086 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 22:41:39,427 INFO [train.py:968] (0/2) Epoch 11, batch 33700, giga_loss[loss=0.2501, simple_loss=0.3325, pruned_loss=0.08389, over 28758.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3389, pruned_loss=0.09318, over 5679984.11 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3556, pruned_loss=0.1197, over 5687313.66 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3375, pruned_loss=0.09037, over 5674572.13 frames. ], batch size: 263, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:41:41,696 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:42:45,488 INFO [train.py:968] (0/2) Epoch 11, batch 33750, giga_loss[loss=0.2793, simple_loss=0.3488, pruned_loss=0.1049, over 29014.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3377, pruned_loss=0.09312, over 5661823.09 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3553, pruned_loss=0.1196, over 5671787.70 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3366, pruned_loss=0.09047, over 5671304.47 frames. ], batch size: 199, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:42:47,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-05 22:42:51,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=489294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:42:59,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.396e+02 1.362e+03 1.990e+03 2.893e+03 6.430e+03, threshold=3.980e+03, percent-clipped=19.0
+2023-03-05 22:43:04,724 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:43:09,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:43:39,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3623, 1.4711, 1.2362, 1.4770], device='cuda:0'), covar=tensor([0.0707, 0.0349, 0.0349, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0058, 0.0053, 0.0090], device='cuda:0')
+2023-03-05 22:43:45,754 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:43:47,631 INFO [train.py:968] (0/2) Epoch 11, batch 33800, giga_loss[loss=0.2885, simple_loss=0.3632, pruned_loss=0.1069, over 28479.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3364, pruned_loss=0.09336, over 5648781.35 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3559, pruned_loss=0.1202, over 5656357.02 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3344, pruned_loss=0.08998, over 5671134.99 frames. ], batch size: 336, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:44:46,092 INFO [train.py:968] (0/2) Epoch 11, batch 33850, giga_loss[loss=0.2457, simple_loss=0.3295, pruned_loss=0.08093, over 28494.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3345, pruned_loss=0.09162, over 5661104.26 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3555, pruned_loss=0.1199, over 5658644.42 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3328, pruned_loss=0.08853, over 5676975.19 frames. ], batch size: 336, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:44:57,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.003e+02 1.252e+03 1.909e+03 2.492e+03 5.078e+03, threshold=3.818e+03, percent-clipped=7.0
+2023-03-05 22:45:16,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5565, 1.6607, 1.2796, 1.9001], device='cuda:0'), covar=tensor([0.2577, 0.2546, 0.2852, 0.2445], device='cuda:0'), in_proj_covar=tensor([0.1293, 0.0959, 0.1150, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 22:45:45,352 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:45:47,115 INFO [train.py:968] (0/2) Epoch 11, batch 33900, giga_loss[loss=0.1892, simple_loss=0.2681, pruned_loss=0.05517, over 28564.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3328, pruned_loss=0.08963, over 5660636.33 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3552, pruned_loss=0.1198, over 5663749.97 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3314, pruned_loss=0.08678, over 5668886.89 frames. ], batch size: 60, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:45:49,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:46:21,874 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:46:43,473 INFO [train.py:968] (0/2) Epoch 11, batch 33950, giga_loss[loss=0.2527, simple_loss=0.344, pruned_loss=0.0807, over 28873.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3338, pruned_loss=0.08826, over 5661058.31 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3551, pruned_loss=0.1197, over 5658142.14 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3323, pruned_loss=0.08551, over 5672004.65 frames. ], batch size: 227, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:46:56,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.684e+02 1.264e+03 1.636e+03 2.891e+03 1.053e+04, threshold=3.272e+03, percent-clipped=16.0
+2023-03-05 22:47:35,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=489531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:47:37,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=489533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:47:44,357 INFO [train.py:968] (0/2) Epoch 11, batch 34000, giga_loss[loss=0.279, simple_loss=0.3563, pruned_loss=0.1008, over 28961.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3361, pruned_loss=0.08799, over 5667967.93 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3551, pruned_loss=0.1197, over 5659413.43 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3349, pruned_loss=0.08575, over 5675398.17 frames. ], batch size: 285, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:48:30,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-05 22:48:44,768 INFO [train.py:968] (0/2) Epoch 11, batch 34050, giga_loss[loss=0.2605, simple_loss=0.344, pruned_loss=0.08853, over 28627.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3371, pruned_loss=0.08858, over 5670800.51 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3546, pruned_loss=0.1195, over 5662820.25 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3364, pruned_loss=0.08656, over 5673569.64 frames. ], batch size: 307, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:49:05,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.654e+02 1.257e+03 1.681e+03 1.960e+03 4.949e+03, threshold=3.361e+03, percent-clipped=3.0
+2023-03-05 22:49:29,214 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 22:49:51,388 INFO [train.py:968] (0/2) Epoch 11, batch 34100, giga_loss[loss=0.2834, simple_loss=0.3526, pruned_loss=0.1071, over 26805.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3375, pruned_loss=0.0894, over 5662336.54 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3546, pruned_loss=0.1196, over 5656246.68 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3364, pruned_loss=0.0867, over 5671054.64 frames. ], batch size: 555, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:50:01,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4498, 1.7342, 1.3152, 1.7273], device='cuda:0'), covar=tensor([0.2445, 0.2326, 0.2641, 0.2066], device='cuda:0'), in_proj_covar=tensor([0.1291, 0.0957, 0.1150, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 22:50:40,520 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:50:46,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:51:03,826 INFO [train.py:968] (0/2) Epoch 11, batch 34150, giga_loss[loss=0.2541, simple_loss=0.3392, pruned_loss=0.08453, over 28886.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3371, pruned_loss=0.08899, over 5660312.83 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3544, pruned_loss=0.1194, over 5658714.00 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3361, pruned_loss=0.08672, over 5664847.12 frames. ], batch size: 227, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:51:20,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.455e+02 1.425e+03 1.737e+03 2.307e+03 1.216e+04, threshold=3.475e+03, percent-clipped=11.0
+2023-03-05 22:51:25,905 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:52:14,067 INFO [train.py:968] (0/2) Epoch 11, batch 34200, giga_loss[loss=0.2489, simple_loss=0.3106, pruned_loss=0.09361, over 24651.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3381, pruned_loss=0.08899, over 5654889.07 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3547, pruned_loss=0.1196, over 5652722.44 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3367, pruned_loss=0.08637, over 5663366.21 frames. ], batch size: 705, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:53:22,148 INFO [train.py:968] (0/2) Epoch 11, batch 34250, giga_loss[loss=0.3192, simple_loss=0.3752, pruned_loss=0.1316, over 26875.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3394, pruned_loss=0.08971, over 5662337.97 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3547, pruned_loss=0.1197, over 5656546.49 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3381, pruned_loss=0.08715, over 5665743.76 frames. ], batch size: 555, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:53:36,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.823e+02 1.569e+03 1.933e+03 2.789e+03 7.926e+03, threshold=3.866e+03, percent-clipped=12.0
+2023-03-05 22:54:09,193 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3780, 1.6916, 1.4582, 1.5921], device='cuda:0'), covar=tensor([0.0746, 0.0298, 0.0317, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0058, 0.0053, 0.0090], device='cuda:0')
+2023-03-05 22:54:21,985 INFO [train.py:968] (0/2) Epoch 11, batch 34300, giga_loss[loss=0.2748, simple_loss=0.3596, pruned_loss=0.09505, over 28460.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3421, pruned_loss=0.09097, over 5666352.53 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3543, pruned_loss=0.1194, over 5654528.92 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.341, pruned_loss=0.08829, over 5670418.29 frames. ], batch size: 336, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:54:58,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7944, 5.2218, 1.9298, 2.2058], device='cuda:0'), covar=tensor([0.0836, 0.0355, 0.0817, 0.1114], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0507, 0.0344, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-05 22:55:08,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3094, 1.6054, 1.4342, 1.4647], device='cuda:0'), covar=tensor([0.1237, 0.1600, 0.1896, 0.1607], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0705, 0.0646, 0.0634], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 22:55:12,502 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-05 22:55:25,212 INFO [train.py:968] (0/2) Epoch 11, batch 34350, giga_loss[loss=0.2448, simple_loss=0.3243, pruned_loss=0.08262, over 28993.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.341, pruned_loss=0.09105, over 5662592.05 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3537, pruned_loss=0.1189, over 5651719.40 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3402, pruned_loss=0.08833, over 5669626.00 frames. ], batch size: 199, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:55:41,805 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.304e+02 1.392e+03 1.720e+03 2.259e+03 1.058e+04, threshold=3.439e+03, percent-clipped=3.0
+2023-03-05 22:55:48,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=489908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:56:21,102 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=489933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:56:26,055 INFO [train.py:968] (0/2) Epoch 11, batch 34400, giga_loss[loss=0.2382, simple_loss=0.3236, pruned_loss=0.07643, over 28973.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3408, pruned_loss=0.09218, over 5676889.08 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3535, pruned_loss=0.1184, over 5664054.59 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3397, pruned_loss=0.08891, over 5672191.89 frames. ], batch size: 228, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:56:42,722 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-05 22:57:33,462 INFO [train.py:968] (0/2) Epoch 11, batch 34450, giga_loss[loss=0.2251, simple_loss=0.3144, pruned_loss=0.06791, over 28671.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3379, pruned_loss=0.09022, over 5684094.88 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3532, pruned_loss=0.1183, over 5671369.67 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3368, pruned_loss=0.08694, over 5674401.47 frames. ], batch size: 307, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:57:52,434 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-490000.pt
+2023-03-05 22:57:54,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.366e+02 1.219e+03 1.520e+03 2.011e+03 4.253e+03, threshold=3.039e+03, percent-clipped=3.0
+2023-03-05 22:58:07,452 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 22:58:41,131 INFO [train.py:968] (0/2) Epoch 11, batch 34500, giga_loss[loss=0.2477, simple_loss=0.3239, pruned_loss=0.08577, over 24763.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.336, pruned_loss=0.08827, over 5685776.01 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3534, pruned_loss=0.1184, over 5673959.02 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3348, pruned_loss=0.08529, over 5675793.31 frames. ], batch size: 705, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:58:58,539 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=490051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:59:00,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=490054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:59:38,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=490083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:59:43,069 INFO [train.py:968] (0/2) Epoch 11, batch 34550, libri_loss[loss=0.2866, simple_loss=0.3513, pruned_loss=0.111, over 27930.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3385, pruned_loss=0.09006, over 5678087.34 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3536, pruned_loss=0.1182, over 5675151.42 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3369, pruned_loss=0.08702, over 5669285.84 frames. ], batch size: 116, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:59:58,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.511e+02 1.122e+03 1.620e+03 2.616e+03 8.014e+03, threshold=3.239e+03, percent-clipped=22.0
+2023-03-05 23:00:07,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4115, 1.6459, 1.4713, 1.4880], device='cuda:0'), covar=tensor([0.1254, 0.1697, 0.1800, 0.1553], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0710, 0.0651, 0.0639], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:0')
+2023-03-05 23:00:12,718 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 23:00:37,510 INFO [train.py:968] (0/2) Epoch 11, batch 34600, giga_loss[loss=0.2481, simple_loss=0.3332, pruned_loss=0.0815, over 29106.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3413, pruned_loss=0.09186, over 5684547.85 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3529, pruned_loss=0.1179, over 5683232.21 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3401, pruned_loss=0.08873, over 5669998.99 frames. ], batch size: 136, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:01:24,082 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=490176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:01:35,957 INFO [train.py:968] (0/2) Epoch 11, batch 34650, libri_loss[loss=0.2644, simple_loss=0.3219, pruned_loss=0.1035, over 29579.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3397, pruned_loss=0.09091, over 5691711.50 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3526, pruned_loss=0.1175, over 5688543.58 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3388, pruned_loss=0.08808, over 5675082.96 frames. ], batch size: 76, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:01:49,817 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.134e+02 1.380e+03 1.803e+03 2.494e+03 5.545e+03, threshold=3.607e+03, percent-clipped=9.0
+2023-03-05 23:02:29,262 INFO [train.py:968] (0/2) Epoch 11, batch 34700, giga_loss[loss=0.2591, simple_loss=0.3371, pruned_loss=0.0905, over 28947.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3383, pruned_loss=0.09181, over 5677849.77 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3531, pruned_loss=0.1179, over 5687828.25 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3367, pruned_loss=0.08849, over 5665021.17 frames. ], batch size: 213, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:02:47,598 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-05 23:03:21,393 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=490288.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:03:21,837 INFO [train.py:968] (0/2) Epoch 11, batch 34750, giga_loss[loss=0.2929, simple_loss=0.3623, pruned_loss=0.1118, over 28735.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3378, pruned_loss=0.09238, over 5668514.22 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3528, pruned_loss=0.1177, over 5684102.65 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3363, pruned_loss=0.08912, over 5660832.60 frames. ], batch size: 262, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:03:39,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.402e+02 1.508e+03 1.991e+03 2.793e+03 1.009e+04, threshold=3.982e+03, percent-clipped=13.0
+2023-03-05 23:03:46,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=490308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:03:49,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7145, 1.6996, 1.2496, 1.2634], device='cuda:0'), covar=tensor([0.0769, 0.0577, 0.0945, 0.1109], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0432, 0.0497, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 23:04:11,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3851, 1.4820, 1.3560, 1.3977], device='cuda:0'), covar=tensor([0.1608, 0.1528, 0.1303, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.1689, 0.1549, 0.1516, 0.1637], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 23:04:17,285 INFO [train.py:968] (0/2) Epoch 11, batch 34800, giga_loss[loss=0.314, simple_loss=0.3873, pruned_loss=0.1203, over 28495.00 frames. ], tot_loss[loss=0.267, simple_loss=0.343, pruned_loss=0.09552, over 5672804.93 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3527, pruned_loss=0.1176, over 5685340.71 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3418, pruned_loss=0.09254, over 5665411.87 frames. ], batch size: 336, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:05:03,933 INFO [train.py:968] (0/2) Epoch 11, batch 34850, giga_loss[loss=0.2818, simple_loss=0.3684, pruned_loss=0.09762, over 28820.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3514, pruned_loss=0.1007, over 5662560.37 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3523, pruned_loss=0.1175, over 5677756.16 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3505, pruned_loss=0.09804, over 5663329.16 frames. ], batch size: 227, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:05:04,338 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-05 23:05:12,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4446, 2.1490, 1.5873, 0.5934], device='cuda:0'), covar=tensor([0.3824, 0.2093, 0.2957, 0.4326], device='cuda:0'), in_proj_covar=tensor([0.1570, 0.1500, 0.1498, 0.1283], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 23:05:14,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.409e+02 1.273e+03 1.721e+03 2.292e+03 5.557e+03, threshold=3.442e+03, percent-clipped=2.0
+2023-03-05 23:05:32,045 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 23:05:44,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4945, 4.2939, 4.0786, 1.8371], device='cuda:0'), covar=tensor([0.0507, 0.0676, 0.0681, 0.2145], device='cuda:0'), in_proj_covar=tensor([0.1035, 0.0960, 0.0843, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 23:05:47,425 INFO [train.py:968] (0/2) Epoch 11, batch 34900, giga_loss[loss=0.3469, simple_loss=0.4086, pruned_loss=0.1426, over 28022.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3584, pruned_loss=0.1053, over 5658693.60 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3529, pruned_loss=0.1178, over 5673524.31 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3573, pruned_loss=0.1026, over 5663392.32 frames. ], batch size: 412, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:05:58,061 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=490451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:06:00,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=490454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:06:25,700 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=490483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:06:26,451 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=490484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:06:29,757 INFO [train.py:968] (0/2) Epoch 11, batch 34950, libri_loss[loss=0.24, simple_loss=0.3092, pruned_loss=0.08534, over 29591.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3554, pruned_loss=0.1046, over 5673980.88 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3525, pruned_loss=0.1175, over 5676668.73 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3549, pruned_loss=0.1024, over 5674756.04 frames. ], batch size: 74, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:06:40,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.438e+02 1.074e+03 1.346e+03 1.832e+03 5.752e+03, threshold=2.692e+03, percent-clipped=6.0
+2023-03-05 23:07:06,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5149, 1.7005, 1.7567, 1.3534], device='cuda:0'), covar=tensor([0.1426, 0.1861, 0.1114, 0.1312], device='cuda:0'), in_proj_covar=tensor([0.0828, 0.0685, 0.0864, 0.0769], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 23:07:11,833 INFO [train.py:968] (0/2) Epoch 11, batch 35000, giga_loss[loss=0.2474, simple_loss=0.3103, pruned_loss=0.09219, over 28510.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3491, pruned_loss=0.1022, over 5684579.56 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.353, pruned_loss=0.1177, over 5681004.60 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3483, pruned_loss=0.1, over 5681305.81 frames. ], batch size: 85, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:07:22,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=490551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:07:56,778 INFO [train.py:968] (0/2) Epoch 11, batch 35050, giga_loss[loss=0.2455, simple_loss=0.3192, pruned_loss=0.08593, over 28575.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3416, pruned_loss=0.09887, over 5673181.02 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.353, pruned_loss=0.1177, over 5674485.67 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3409, pruned_loss=0.09699, over 5676887.18 frames. ], batch size: 336, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:08:02,904 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=490598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:08:07,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.075e+02 9.615e+02 1.324e+03 1.798e+03 5.896e+03, threshold=2.648e+03, percent-clipped=8.0
+2023-03-05 23:08:11,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8421, 1.9080, 1.6236, 2.1442], device='cuda:0'), covar=tensor([0.2287, 0.2339, 0.2588, 0.2214], device='cuda:0'), in_proj_covar=tensor([0.1291, 0.0955, 0.1148, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 23:08:36,786 INFO [train.py:968] (0/2) Epoch 11, batch 35100, giga_loss[loss=0.2362, simple_loss=0.3063, pruned_loss=0.08305, over 27992.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3351, pruned_loss=0.09631, over 5680874.43 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.353, pruned_loss=0.1175, over 5680509.02 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.334, pruned_loss=0.09433, over 5678671.60 frames. ], batch size: 412, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:08:45,639 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=490652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:08:52,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=490663.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:09:03,696 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-05 23:09:15,057 INFO [train.py:968] (0/2) Epoch 11, batch 35150, giga_loss[loss=0.2157, simple_loss=0.2864, pruned_loss=0.07247, over 28890.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3285, pruned_loss=0.09343, over 5684209.47 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3533, pruned_loss=0.1176, over 5676682.61 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3267, pruned_loss=0.09102, over 5685596.20 frames. ], batch size: 106, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:09:20,151 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=490694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:09:22,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=490697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:09:27,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.794e+02 1.010e+03 1.364e+03 2.199e+03 4.836e+03, threshold=2.729e+03, percent-clipped=18.0
+2023-03-05 23:09:44,120 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=490726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:09:55,609 INFO [train.py:968] (0/2) Epoch 11, batch 35200, giga_loss[loss=0.2597, simple_loss=0.3276, pruned_loss=0.09585, over 28824.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3238, pruned_loss=0.0911, over 5682802.70 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3533, pruned_loss=0.1173, over 5673374.76 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3215, pruned_loss=0.08863, over 5687108.88 frames. ], batch size: 112, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:10:12,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4675, 1.5733, 1.3112, 1.5964], device='cuda:0'), covar=tensor([0.0755, 0.0313, 0.0332, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0117, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:0')
+2023-03-05 23:10:20,372 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5382, 4.3519, 1.6816, 1.7543], device='cuda:0'), covar=tensor([0.0932, 0.0248, 0.0884, 0.1224], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0504, 0.0340, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 23:10:37,247 INFO [train.py:968] (0/2) Epoch 11, batch 35250, giga_loss[loss=0.2332, simple_loss=0.306, pruned_loss=0.08026, over 28198.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3203, pruned_loss=0.08924, over 5681127.81 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3532, pruned_loss=0.1172, over 5676910.78 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3178, pruned_loss=0.0868, over 5681458.69 frames. ], batch size: 368, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:10:51,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.139e+02 9.718e+02 1.245e+03 1.774e+03 6.417e+03, threshold=2.490e+03, percent-clipped=5.0
+2023-03-05 23:10:52,688 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=490806.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:10:55,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=490809.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:10:56,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0633, 2.3955, 2.4053, 1.9050], device='cuda:0'), covar=tensor([0.1255, 0.1709, 0.1003, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0832, 0.0691, 0.0870, 0.0773], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 23:11:18,619 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=490838.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:11:19,060 INFO [train.py:968] (0/2) Epoch 11, batch 35300, giga_loss[loss=0.2214, simple_loss=0.2989, pruned_loss=0.0719, over 28891.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3168, pruned_loss=0.08762, over 5682793.05 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3539, pruned_loss=0.1177, over 5669572.28 frames. ], giga_tot_loss[loss=0.2413, simple_loss=0.3135, pruned_loss=0.08452, over 5689924.96 frames. ], batch size: 174, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:11:26,356 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-05 23:11:30,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4142, 3.0582, 1.4607, 1.5717], device='cuda:0'), covar=tensor([0.0937, 0.0350, 0.0842, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0504, 0.0339, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 23:11:33,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=490859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:12:01,241 INFO [train.py:968] (0/2) Epoch 11, batch 35350, giga_loss[loss=0.2313, simple_loss=0.2987, pruned_loss=0.08196, over 28779.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3138, pruned_loss=0.08604, over 5697699.64 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3538, pruned_loss=0.1177, over 5674379.27 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3105, pruned_loss=0.08311, over 5699552.31 frames. ], batch size: 284, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:12:12,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.981e+02 1.029e+03 1.399e+03 1.837e+03 7.504e+03, threshold=2.797e+03, percent-clipped=9.0
+2023-03-05 23:12:42,006 INFO [train.py:968] (0/2) Epoch 11, batch 35400, giga_loss[loss=0.2275, simple_loss=0.2951, pruned_loss=0.0799, over 29050.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3113, pruned_loss=0.08487, over 5696906.08 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3543, pruned_loss=0.1178, over 5676843.50 frames. ], giga_tot_loss[loss=0.2351, simple_loss=0.3072, pruned_loss=0.08156, over 5697114.93 frames. ], batch size: 106, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:13:07,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=490973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:13:21,506 INFO [train.py:968] (0/2) Epoch 11, batch 35450, giga_loss[loss=0.1902, simple_loss=0.2638, pruned_loss=0.05831, over 28410.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3087, pruned_loss=0.08341, over 5697966.51 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3545, pruned_loss=0.1178, over 5680387.21 frames. ], giga_tot_loss[loss=0.2312, simple_loss=0.3035, pruned_loss=0.07943, over 5695553.01 frames. ], batch size: 60, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:13:32,845 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=491002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:13:35,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.664e+02 1.065e+03 1.387e+03 1.985e+03 7.694e+03, threshold=2.773e+03, percent-clipped=11.0
+2023-03-05 23:13:36,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=491005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:13:53,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=491027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:13:59,034 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=491034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:14:03,065 INFO [train.py:968] (0/2) Epoch 11, batch 35500, giga_loss[loss=0.2019, simple_loss=0.2764, pruned_loss=0.06372, over 28992.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.306, pruned_loss=0.08208, over 5696494.73 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3546, pruned_loss=0.1179, over 5680306.61 frames. ], giga_tot_loss[loss=0.2289, simple_loss=0.3011, pruned_loss=0.07832, over 5694944.36 frames. ], batch size: 136, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:14:44,806 INFO [train.py:968] (0/2) Epoch 11, batch 35550, giga_loss[loss=0.2013, simple_loss=0.2713, pruned_loss=0.06562, over 28748.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3033, pruned_loss=0.08085, over 5693345.07 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3549, pruned_loss=0.1179, over 5683170.91 frames. ], giga_tot_loss[loss=0.2262, simple_loss=0.2981, pruned_loss=0.07714, over 5689742.63 frames. ], batch size: 112, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:14:53,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8664, 5.1380, 1.8642, 2.1849], device='cuda:0'), covar=tensor([0.0798, 0.0290, 0.0820, 0.1063], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0502, 0.0338, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 23:14:57,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.573e+02 1.028e+03 1.272e+03 1.742e+03 7.406e+03, threshold=2.545e+03, percent-clipped=8.0
+2023-03-05 23:15:07,146 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=491116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:15:09,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=491119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:15:27,664 INFO [train.py:968] (0/2) Epoch 11, batch 35600, giga_loss[loss=0.2502, simple_loss=0.3158, pruned_loss=0.09235, over 28757.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3022, pruned_loss=0.08091, over 5696781.49 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3553, pruned_loss=0.1181, over 5686560.07 frames. ], giga_tot_loss[loss=0.2243, simple_loss=0.2958, pruned_loss=0.07641, over 5691510.51 frames. ], batch size: 284, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:15:34,335 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=491148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:15:54,675 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=491170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:15:55,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3142, 1.1907, 1.1708, 1.4353], device='cuda:0'), covar=tensor([0.0747, 0.0379, 0.0339, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0116, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:0')
+2023-03-05 23:15:57,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=491173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:16:01,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4673, 1.6422, 1.3878, 1.5363], device='cuda:0'), covar=tensor([0.0758, 0.0315, 0.0326, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0116, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:0')
+2023-03-05 23:16:11,440 INFO [train.py:968] (0/2) Epoch 11, batch 35650, giga_loss[loss=0.3132, simple_loss=0.3812, pruned_loss=0.1226, over 28933.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3098, pruned_loss=0.08504, over 5693144.31 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3556, pruned_loss=0.1183, over 5690143.48 frames. ], giga_tot_loss[loss=0.2323, simple_loss=0.3035, pruned_loss=0.08059, over 5685873.10 frames. ], batch size: 213, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:16:24,414 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=491202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:16:25,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.847e+02 1.006e+03 1.398e+03 2.209e+03 9.892e+03, threshold=2.796e+03, percent-clipped=19.0
+2023-03-05 23:16:58,763 INFO [train.py:968] (0/2) Epoch 11, batch 35700, giga_loss[loss=0.289, simple_loss=0.3622, pruned_loss=0.108, over 28874.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3225, pruned_loss=0.09173, over 5694634.99 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3557, pruned_loss=0.1183, over 5692490.25 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3167, pruned_loss=0.08774, over 5686676.43 frames. ], batch size: 119, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:17:41,998 INFO [train.py:968] (0/2) Epoch 11, batch 35750, giga_loss[loss=0.3304, simple_loss=0.3964, pruned_loss=0.1322, over 28783.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3349, pruned_loss=0.09794, over 5700067.22 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3555, pruned_loss=0.118, over 5695960.40 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3301, pruned_loss=0.09465, over 5690507.30 frames. ], batch size: 284, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:17:52,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.443e+02 1.218e+03 1.485e+03 1.962e+03 6.486e+03, threshold=2.971e+03, percent-clipped=9.0
+2023-03-05 23:18:21,950 INFO [train.py:968] (0/2) Epoch 11, batch 35800, giga_loss[loss=0.2658, simple_loss=0.3554, pruned_loss=0.08808, over 28933.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3428, pruned_loss=0.1016, over 5693258.86 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3561, pruned_loss=0.1183, over 5696266.08 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3377, pruned_loss=0.09798, over 5685207.21 frames. ], batch size: 164, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:18:27,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4196, 3.4846, 1.5781, 1.5517], device='cuda:0'), covar=tensor([0.0933, 0.0298, 0.0848, 0.1300], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0503, 0.0337, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 23:19:02,125 INFO [train.py:968] (0/2) Epoch 11, batch 35850, libri_loss[loss=0.3651, simple_loss=0.4179, pruned_loss=0.1562, over 29281.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3468, pruned_loss=0.1025, over 5692633.51 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3568, pruned_loss=0.1186, over 5702218.97 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3417, pruned_loss=0.09857, over 5680403.46 frames. ], batch size: 94, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:19:10,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5089, 3.5117, 1.7106, 1.6040], device='cuda:0'), covar=tensor([0.0920, 0.0270, 0.0840, 0.1302], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0505, 0.0338, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 23:19:14,921 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.764e+02 1.200e+03 1.538e+03 2.011e+03 1.478e+04, threshold=3.077e+03, percent-clipped=9.0
+2023-03-05 23:19:47,323 INFO [train.py:968] (0/2) Epoch 11, batch 35900, giga_loss[loss=0.2735, simple_loss=0.3484, pruned_loss=0.09931, over 28884.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3492, pruned_loss=0.1024, over 5691250.06 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3574, pruned_loss=0.1189, over 5695542.29 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3444, pruned_loss=0.09871, over 5686770.71 frames. ], batch size: 145, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:20:09,926 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=491461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:20:31,115 INFO [train.py:968] (0/2) Epoch 11, batch 35950, giga_loss[loss=0.3135, simple_loss=0.3758, pruned_loss=0.1256, over 28934.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3514, pruned_loss=0.1034, over 5696193.74 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3579, pruned_loss=0.1192, over 5698086.63 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.347, pruned_loss=0.1, over 5690208.39 frames. ], batch size: 213, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:20:47,650 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.110e+02 1.116e+03 1.513e+03 2.481e+03 1.305e+04, threshold=3.026e+03, percent-clipped=8.0
+2023-03-05 23:21:13,933 INFO [train.py:968] (0/2) Epoch 11, batch 36000, libri_loss[loss=0.2918, simple_loss=0.3527, pruned_loss=0.1154, over 29561.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3543, pruned_loss=0.1061, over 5691216.60 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3584, pruned_loss=0.1193, over 5703483.87 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.35, pruned_loss=0.1027, over 5680552.20 frames. ], batch size: 76, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:21:13,937 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-05 23:21:22,141 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2555, 1.5657, 1.4265, 1.2048], device='cuda:0'), covar=tensor([0.1791, 0.1618, 0.0907, 0.1200], device='cuda:0'), in_proj_covar=tensor([0.1718, 0.1589, 0.1546, 0.1657], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 23:21:22,667 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2113, simple_loss=0.3168, pruned_loss=0.0529, over 944034.00 frames. 
+2023-03-05 23:21:22,668 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-05 23:21:24,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7980, 1.9091, 2.0562, 1.6453], device='cuda:0'), covar=tensor([0.1560, 0.2004, 0.1178, 0.1400], device='cuda:0'), in_proj_covar=tensor([0.0833, 0.0690, 0.0870, 0.0773], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 23:21:36,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2703, 1.5939, 1.2241, 1.4649], device='cuda:0'), covar=tensor([0.2231, 0.2019, 0.2223, 0.1858], device='cuda:0'), in_proj_covar=tensor([0.1291, 0.0963, 0.1148, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 23:21:46,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=491568.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:22:04,258 INFO [train.py:968] (0/2) Epoch 11, batch 36050, giga_loss[loss=0.3014, simple_loss=0.366, pruned_loss=0.1184, over 28261.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3587, pruned_loss=0.1092, over 5683470.01 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3593, pruned_loss=0.1195, over 5698475.06 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3545, pruned_loss=0.1059, over 5679929.96 frames. ], batch size: 368, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:22:11,323 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 23:22:17,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.681e+02 1.364e+03 1.634e+03 2.179e+03 5.379e+03, threshold=3.268e+03, percent-clipped=9.0
+2023-03-05 23:22:41,616 INFO [train.py:968] (0/2) Epoch 11, batch 36100, giga_loss[loss=0.3093, simple_loss=0.3727, pruned_loss=0.1229, over 28652.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3624, pruned_loss=0.1113, over 5693768.00 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3599, pruned_loss=0.1197, over 5705314.36 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3584, pruned_loss=0.1082, over 5684295.05 frames. ], batch size: 85, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:23:24,537 INFO [train.py:968] (0/2) Epoch 11, batch 36150, giga_loss[loss=0.2756, simple_loss=0.3561, pruned_loss=0.09761, over 28704.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3641, pruned_loss=0.1115, over 5690454.67 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3599, pruned_loss=0.1196, over 5706350.79 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.361, pruned_loss=0.109, over 5682186.63 frames. ], batch size: 262, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:23:28,652 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=491695.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:23:39,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.389e+02 1.102e+03 1.423e+03 1.899e+03 3.826e+03, threshold=2.845e+03, percent-clipped=1.0
+2023-03-05 23:24:07,563 INFO [train.py:968] (0/2) Epoch 11, batch 36200, giga_loss[loss=0.2797, simple_loss=0.3632, pruned_loss=0.0981, over 28925.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3656, pruned_loss=0.112, over 5687823.11 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.36, pruned_loss=0.1197, over 5708618.39 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3632, pruned_loss=0.1097, over 5678645.25 frames. ], batch size: 174, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:24:43,147 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=491785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:24:46,844 INFO [train.py:968] (0/2) Epoch 11, batch 36250, giga_loss[loss=0.2608, simple_loss=0.3524, pruned_loss=0.08456, over 28985.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3665, pruned_loss=0.1114, over 5687960.99 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3608, pruned_loss=0.1203, over 5699331.21 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.364, pruned_loss=0.1089, over 5688547.12 frames. ], batch size: 155, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:25:00,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.926e+02 1.123e+03 1.541e+03 2.051e+03 9.579e+03, threshold=3.081e+03, percent-clipped=14.0
+2023-03-05 23:25:23,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=491836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:25:25,455 INFO [train.py:968] (0/2) Epoch 11, batch 36300, giga_loss[loss=0.2517, simple_loss=0.3409, pruned_loss=0.08126, over 28845.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3642, pruned_loss=0.1085, over 5699217.25 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3609, pruned_loss=0.1201, over 5705315.58 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3622, pruned_loss=0.1065, over 5694077.12 frames. ], batch size: 186, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:26:00,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1713, 3.1120, 1.4016, 1.3768], device='cuda:0'), covar=tensor([0.1082, 0.0304, 0.0900, 0.1455], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0500, 0.0335, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0029, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 23:26:05,913 INFO [train.py:968] (0/2) Epoch 11, batch 36350, giga_loss[loss=0.2953, simple_loss=0.3633, pruned_loss=0.1137, over 28957.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3617, pruned_loss=0.1062, over 5707776.08 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3612, pruned_loss=0.1202, over 5707572.99 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.36, pruned_loss=0.1041, over 5701533.02 frames. ], batch size: 106, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:26:08,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1109, 2.9091, 1.2316, 1.2677], device='cuda:0'), covar=tensor([0.1257, 0.0349, 0.1053, 0.1683], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0500, 0.0335, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0029, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 23:26:21,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.402e+02 1.047e+03 1.248e+03 1.703e+03 8.232e+03, threshold=2.497e+03, percent-clipped=7.0
+2023-03-05 23:26:45,963 INFO [train.py:968] (0/2) Epoch 11, batch 36400, giga_loss[loss=0.3368, simple_loss=0.3899, pruned_loss=0.1419, over 29096.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3617, pruned_loss=0.1065, over 5702537.43 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3616, pruned_loss=0.1205, over 5696164.89 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.36, pruned_loss=0.1045, over 5707749.46 frames. ], batch size: 128, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:26:51,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=491943.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:27:24,629 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=491979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:27:28,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=491982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:27:31,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3746, 4.1505, 3.9569, 1.8071], device='cuda:0'), covar=tensor([0.0606, 0.0791, 0.0809, 0.2233], device='cuda:0'), in_proj_covar=tensor([0.1020, 0.0956, 0.0838, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 23:27:32,562 INFO [train.py:968] (0/2) Epoch 11, batch 36450, libri_loss[loss=0.3171, simple_loss=0.38, pruned_loss=0.1271, over 18997.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3636, pruned_loss=0.1106, over 5689588.94 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3616, pruned_loss=0.1205, over 5689532.20 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3623, pruned_loss=0.1088, over 5700798.33 frames. ], batch size: 189, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:27:41,633 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-492000.pt
+2023-03-05 23:27:47,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.678e+02 1.305e+03 1.588e+03 2.265e+03 5.168e+03, threshold=3.175e+03, percent-clipped=19.0
+2023-03-05 23:27:50,529 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:28:13,908 INFO [train.py:968] (0/2) Epoch 11, batch 36500, giga_loss[loss=0.3251, simple_loss=0.3864, pruned_loss=0.1319, over 29000.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3649, pruned_loss=0.1133, over 5686977.87 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3617, pruned_loss=0.1203, over 5691177.67 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3638, pruned_loss=0.1117, over 5694798.03 frames. ], batch size: 227, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:28:36,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:28:40,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=492070.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:28:45,265 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7442, 1.7811, 1.2854, 1.3605], device='cuda:0'), covar=tensor([0.0688, 0.0511, 0.0938, 0.0958], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0436, 0.0499, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 23:28:53,401 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492086.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:28:55,049 INFO [train.py:968] (0/2) Epoch 11, batch 36550, giga_loss[loss=0.2715, simple_loss=0.3393, pruned_loss=0.1018, over 28214.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.364, pruned_loss=0.1132, over 5689895.07 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3624, pruned_loss=0.1207, over 5687715.19 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3625, pruned_loss=0.1114, over 5699964.48 frames. ], batch size: 77, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:28:55,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492089.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:28:57,526 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 23:29:09,369 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.169e+02 1.213e+03 1.541e+03 2.117e+03 6.296e+03, threshold=3.082e+03, percent-clipped=9.0
+2023-03-05 23:29:19,518 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492118.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:29:30,223 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7769, 1.7520, 1.3421, 1.2731], device='cuda:0'), covar=tensor([0.0815, 0.0627, 0.1065, 0.1116], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0439, 0.0502, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 23:29:34,430 INFO [train.py:968] (0/2) Epoch 11, batch 36600, giga_loss[loss=0.2501, simple_loss=0.3203, pruned_loss=0.08998, over 28501.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.361, pruned_loss=0.1119, over 5690518.43 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.363, pruned_loss=0.121, over 5688513.20 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3593, pruned_loss=0.1099, over 5697615.47 frames. ], batch size: 85, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:29:48,364 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:29:50,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=492160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:30:16,037 INFO [train.py:968] (0/2) Epoch 11, batch 36650, giga_loss[loss=0.2608, simple_loss=0.3377, pruned_loss=0.09193, over 28432.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3605, pruned_loss=0.1114, over 5698418.44 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3631, pruned_loss=0.1211, over 5696027.78 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3589, pruned_loss=0.1094, over 5697676.55 frames. ], batch size: 65, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:30:31,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.180e+02 1.214e+03 1.482e+03 2.297e+03 8.574e+03, threshold=2.963e+03, percent-clipped=12.0
+2023-03-05 23:30:37,361 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492213.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:30:40,261 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492216.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:30:49,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3143, 3.4153, 1.4885, 1.4573], device='cuda:0'), covar=tensor([0.0991, 0.0241, 0.0889, 0.1452], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0508, 0.0339, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-05 23:30:57,801 INFO [train.py:968] (0/2) Epoch 11, batch 36700, giga_loss[loss=0.259, simple_loss=0.341, pruned_loss=0.08847, over 28678.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3597, pruned_loss=0.1102, over 5691819.72 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3635, pruned_loss=0.1212, over 5690742.15 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.358, pruned_loss=0.1083, over 5695240.61 frames. ], batch size: 284, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:31:03,872 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492245.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:31:08,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2070, 1.3366, 3.2593, 3.1375], device='cuda:0'), covar=tensor([0.1389, 0.2370, 0.0393, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0655, 0.0581, 0.0849, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 23:31:08,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2605, 1.4597, 1.2424, 1.4625], device='cuda:0'), covar=tensor([0.0798, 0.0324, 0.0331, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-05 23:31:26,608 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492268.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:31:46,598 INFO [train.py:968] (0/2) Epoch 11, batch 36750, giga_loss[loss=0.2444, simple_loss=0.3269, pruned_loss=0.08091, over 29011.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3565, pruned_loss=0.1078, over 5682702.15 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3634, pruned_loss=0.121, over 5694146.10 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3552, pruned_loss=0.1063, over 5682239.22 frames. ], batch size: 145, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:31:47,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2204, 4.0672, 3.8289, 1.7534], device='cuda:0'), covar=tensor([0.0545, 0.0645, 0.0652, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.1032, 0.0968, 0.0848, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 23:31:47,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4769, 1.7071, 1.7639, 1.3576], device='cuda:0'), covar=tensor([0.1682, 0.2296, 0.1292, 0.1516], device='cuda:0'), in_proj_covar=tensor([0.0830, 0.0693, 0.0869, 0.0771], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-05 23:31:58,496 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:32:00,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:32:01,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.559e+02 1.052e+03 1.243e+03 1.740e+03 4.744e+03, threshold=2.487e+03, percent-clipped=4.0
+2023-03-05 23:32:21,449 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:32:28,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:32:31,229 INFO [train.py:968] (0/2) Epoch 11, batch 36800, giga_loss[loss=0.2441, simple_loss=0.3147, pruned_loss=0.08677, over 28746.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3511, pruned_loss=0.1051, over 5658014.40 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3639, pruned_loss=0.1215, over 5675704.06 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3494, pruned_loss=0.1031, over 5673329.98 frames. ], batch size: 284, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:33:18,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3897, 3.2095, 3.1003, 1.2627], device='cuda:0'), covar=tensor([0.0852, 0.0976, 0.0914, 0.2145], device='cuda:0'), in_proj_covar=tensor([0.1027, 0.0965, 0.0843, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 23:33:22,098 INFO [train.py:968] (0/2) Epoch 11, batch 36850, libri_loss[loss=0.2871, simple_loss=0.3609, pruned_loss=0.1066, over 29513.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3459, pruned_loss=0.1027, over 5654975.62 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3645, pruned_loss=0.1216, over 5683394.21 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3434, pruned_loss=0.1002, over 5659773.02 frames. ], batch size: 82, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:33:43,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.608e+02 9.300e+02 1.225e+03 1.569e+03 6.746e+03, threshold=2.450e+03, percent-clipped=3.0
+2023-03-05 23:34:15,848 INFO [train.py:968] (0/2) Epoch 11, batch 36900, giga_loss[loss=0.2785, simple_loss=0.3473, pruned_loss=0.1049, over 28814.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3407, pruned_loss=0.1002, over 5648063.67 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3646, pruned_loss=0.1218, over 5684607.24 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3385, pruned_loss=0.09806, over 5650626.60 frames. ], batch size: 99, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:34:19,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=492442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:35:01,484 INFO [train.py:968] (0/2) Epoch 11, batch 36950, giga_loss[loss=0.3031, simple_loss=0.375, pruned_loss=0.1156, over 29061.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3419, pruned_loss=0.1001, over 5666727.44 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3653, pruned_loss=0.1222, over 5688961.49 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3391, pruned_loss=0.09771, over 5664335.04 frames. ], batch size: 164, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:35:16,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.975e+02 1.081e+03 1.581e+03 2.193e+03 9.834e+03, threshold=3.162e+03, percent-clipped=20.0
+2023-03-05 23:35:19,674 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-05 23:35:39,499 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-05 23:35:40,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=492532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:35:44,420 INFO [train.py:968] (0/2) Epoch 11, batch 37000, giga_loss[loss=0.2537, simple_loss=0.3218, pruned_loss=0.09274, over 28911.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3399, pruned_loss=0.09858, over 5671377.28 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.365, pruned_loss=0.1219, over 5692056.98 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3377, pruned_loss=0.09655, over 5666540.27 frames. ], batch size: 99, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:35:50,369 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-05 23:36:26,001 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:36:27,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:36:28,081 INFO [train.py:968] (0/2) Epoch 11, batch 37050, giga_loss[loss=0.2735, simple_loss=0.3397, pruned_loss=0.1037, over 29015.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3386, pruned_loss=0.09775, over 5687049.12 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3662, pruned_loss=0.1226, over 5692576.39 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3352, pruned_loss=0.09508, over 5682634.80 frames. ], batch size: 155, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:36:33,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9159, 1.0697, 1.0311, 0.8301], device='cuda:0'), covar=tensor([0.1223, 0.1325, 0.0783, 0.1153], device='cuda:0'), in_proj_covar=tensor([0.1707, 0.1582, 0.1548, 0.1663], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-05 23:36:34,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7099, 1.7114, 1.2119, 1.2884], device='cuda:0'), covar=tensor([0.0790, 0.0562, 0.1070, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0440, 0.0503, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 23:36:45,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.801e+02 9.511e+02 1.351e+03 1.935e+03 1.501e+04, threshold=2.701e+03, percent-clipped=7.0
+2023-03-05 23:36:51,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:37:08,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1030, 1.9769, 2.0322, 1.7193], device='cuda:0'), covar=tensor([0.1386, 0.2320, 0.1789, 0.2109], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0738, 0.0672, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 23:37:09,894 INFO [train.py:968] (0/2) Epoch 11, batch 37100, giga_loss[loss=0.236, simple_loss=0.3154, pruned_loss=0.07833, over 28970.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3356, pruned_loss=0.09597, over 5691931.53 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3668, pruned_loss=0.1228, over 5682556.65 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.332, pruned_loss=0.09329, over 5696832.25 frames. ], batch size: 213, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:37:13,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=492643.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:37:39,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5483, 4.3636, 4.1312, 2.0179], device='cuda:0'), covar=tensor([0.0480, 0.0628, 0.0611, 0.2041], device='cuda:0'), in_proj_covar=tensor([0.1023, 0.0958, 0.0839, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-05 23:37:41,278 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:37:43,965 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:37:52,338 INFO [train.py:968] (0/2) Epoch 11, batch 37150, giga_loss[loss=0.2322, simple_loss=0.3076, pruned_loss=0.07838, over 28950.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3317, pruned_loss=0.09422, over 5695082.83 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3668, pruned_loss=0.1228, over 5682556.65 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3289, pruned_loss=0.09214, over 5698897.12 frames. ], batch size: 227, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:37:53,041 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:37:53,234 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-05 23:38:04,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=492704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:38:07,380 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:38:08,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.769e+02 9.631e+02 1.226e+03 1.578e+03 4.548e+03, threshold=2.452e+03, percent-clipped=4.0
+2023-03-05 23:38:31,779 INFO [train.py:968] (0/2) Epoch 11, batch 37200, giga_loss[loss=0.2342, simple_loss=0.3019, pruned_loss=0.0833, over 28724.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3295, pruned_loss=0.09331, over 5699617.41 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3672, pruned_loss=0.1229, over 5685040.31 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3262, pruned_loss=0.09103, over 5700710.51 frames. ], batch size: 99, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:39:11,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492786.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:39:12,968 INFO [train.py:968] (0/2) Epoch 11, batch 37250, giga_loss[loss=0.228, simple_loss=0.2944, pruned_loss=0.08083, over 28667.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3274, pruned_loss=0.09223, over 5709221.86 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3674, pruned_loss=0.1229, over 5683585.90 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3243, pruned_loss=0.09015, over 5711400.93 frames. ], batch size: 92, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:39:13,185 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492789.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:39:28,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.460e+02 9.472e+02 1.195e+03 1.766e+03 8.416e+03, threshold=2.390e+03, percent-clipped=13.0
+2023-03-05 23:39:32,212 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492814.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:39:36,936 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492818.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:39:37,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5384, 1.8041, 1.4482, 1.6943], device='cuda:0'), covar=tensor([0.2370, 0.2408, 0.2599, 0.2332], device='cuda:0'), in_proj_covar=tensor([0.1294, 0.0960, 0.1147, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 23:39:51,347 INFO [train.py:968] (0/2) Epoch 11, batch 37300, giga_loss[loss=0.224, simple_loss=0.304, pruned_loss=0.07198, over 29032.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3272, pruned_loss=0.09217, over 5709178.01 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3689, pruned_loss=0.1236, over 5680645.45 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3221, pruned_loss=0.08892, over 5713677.80 frames. ], batch size: 155, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:39:58,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:39:59,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3723, 3.6450, 1.5054, 1.6017], device='cuda:0'), covar=tensor([0.0875, 0.0301, 0.0852, 0.1227], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0506, 0.0339, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 23:40:01,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:40:06,978 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9227, 1.1075, 3.6471, 2.9188], device='cuda:0'), covar=tensor([0.1898, 0.2739, 0.0450, 0.0921], device='cuda:0'), in_proj_covar=tensor([0.0657, 0.0585, 0.0852, 0.0752], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 23:40:23,141 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:40:30,295 INFO [train.py:968] (0/2) Epoch 11, batch 37350, giga_loss[loss=0.2148, simple_loss=0.2908, pruned_loss=0.06946, over 29031.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3272, pruned_loss=0.09234, over 5706864.67 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3702, pruned_loss=0.124, over 5679347.05 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3204, pruned_loss=0.08822, over 5712946.60 frames. ], batch size: 128, lr: 2.89e-03, grad_scale: 1.0
+2023-03-05 23:40:49,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.752e+02 9.743e+02 1.253e+03 1.772e+03 6.092e+03, threshold=2.506e+03, percent-clipped=11.0
+2023-03-05 23:41:14,211 INFO [train.py:968] (0/2) Epoch 11, batch 37400, giga_loss[loss=0.2219, simple_loss=0.3029, pruned_loss=0.07047, over 29077.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3264, pruned_loss=0.09162, over 5708133.24 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3714, pruned_loss=0.1245, over 5681142.62 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3189, pruned_loss=0.0872, over 5711705.32 frames. ], batch size: 128, lr: 2.89e-03, grad_scale: 1.0
+2023-03-05 23:41:54,605 INFO [train.py:968] (0/2) Epoch 11, batch 37450, giga_loss[loss=0.234, simple_loss=0.3079, pruned_loss=0.08005, over 28826.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3247, pruned_loss=0.09071, over 5706575.16 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.371, pruned_loss=0.1242, over 5684306.44 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3186, pruned_loss=0.08706, over 5706869.26 frames. ], batch size: 112, lr: 2.89e-03, grad_scale: 1.0
+2023-03-05 23:42:15,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.482e+02 9.810e+02 1.181e+03 1.679e+03 6.162e+03, threshold=2.361e+03, percent-clipped=7.0
+2023-03-05 23:42:32,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5948, 4.5167, 1.8658, 1.6674], device='cuda:0'), covar=tensor([0.0948, 0.0246, 0.0833, 0.1326], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0506, 0.0339, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 23:42:36,414 INFO [train.py:968] (0/2) Epoch 11, batch 37500, giga_loss[loss=0.224, simple_loss=0.3111, pruned_loss=0.0684, over 28863.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3268, pruned_loss=0.09164, over 5718098.57 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3716, pruned_loss=0.1243, over 5688139.53 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3202, pruned_loss=0.08781, over 5715329.98 frames. ], batch size: 174, lr: 2.89e-03, grad_scale: 1.0
+2023-03-05 23:42:45,688 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=493047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:42:47,271 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-05 23:42:48,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3470, 1.4914, 1.2708, 1.5771], device='cuda:0'), covar=tensor([0.0789, 0.0335, 0.0343, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:0')
+2023-03-05 23:43:01,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=493065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:43:22,464 INFO [train.py:968] (0/2) Epoch 11, batch 37550, giga_loss[loss=0.3271, simple_loss=0.3947, pruned_loss=0.1298, over 28808.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3326, pruned_loss=0.09556, over 5715080.07 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.372, pruned_loss=0.1246, over 5691377.34 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3265, pruned_loss=0.09186, over 5710297.56 frames. ], batch size: 199, lr: 2.89e-03, grad_scale: 1.0
+2023-03-05 23:43:28,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6698, 2.0360, 1.5936, 1.7823], device='cuda:0'), covar=tensor([0.2085, 0.2066, 0.2257, 0.1838], device='cuda:0'), in_proj_covar=tensor([0.1296, 0.0961, 0.1149, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 23:43:40,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.913e+02 1.230e+03 1.645e+03 2.373e+03 7.153e+03, threshold=3.290e+03, percent-clipped=26.0
+2023-03-05 23:44:09,258 INFO [train.py:968] (0/2) Epoch 11, batch 37600, giga_loss[loss=0.3311, simple_loss=0.3946, pruned_loss=0.1338, over 29087.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.341, pruned_loss=0.1013, over 5708032.00 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3718, pruned_loss=0.1243, over 5692661.20 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3356, pruned_loss=0.09797, over 5703536.60 frames. ], batch size: 155, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:44:56,924 INFO [train.py:968] (0/2) Epoch 11, batch 37650, libri_loss[loss=0.3154, simple_loss=0.3871, pruned_loss=0.1218, over 29495.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3486, pruned_loss=0.1062, over 5701358.17 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3719, pruned_loss=0.1243, over 5697060.61 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3436, pruned_loss=0.1032, over 5693705.77 frames. ], batch size: 85, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:44:57,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=493189.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:45:18,882 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=493208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:45:22,621 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.498e+02 1.180e+03 1.482e+03 1.827e+03 4.775e+03, threshold=2.964e+03, percent-clipped=7.0
+2023-03-05 23:45:22,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=493211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:45:32,758 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-05 23:45:45,920 INFO [train.py:968] (0/2) Epoch 11, batch 37700, giga_loss[loss=0.3741, simple_loss=0.4161, pruned_loss=0.1661, over 26539.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3531, pruned_loss=0.1079, over 5691567.31 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3718, pruned_loss=0.1242, over 5700253.83 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.349, pruned_loss=0.1054, over 5682536.20 frames. ], batch size: 555, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:45:46,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=493240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:46:28,980 INFO [train.py:968] (0/2) Epoch 11, batch 37750, giga_loss[loss=0.3173, simple_loss=0.3966, pruned_loss=0.119, over 28850.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3588, pruned_loss=0.1106, over 5695959.78 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3716, pruned_loss=0.1239, over 5708114.47 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3551, pruned_loss=0.1082, over 5681577.71 frames. ], batch size: 174, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:46:31,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7869, 2.2356, 1.7936, 1.2030], device='cuda:0'), covar=tensor([0.2864, 0.1956, 0.2152, 0.3070], device='cuda:0'), in_proj_covar=tensor([0.1551, 0.1473, 0.1493, 0.1266], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-05 23:46:43,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=493306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:46:48,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.193e+02 1.100e+03 1.389e+03 1.980e+03 7.015e+03, threshold=2.778e+03, percent-clipped=9.0
+2023-03-05 23:47:07,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=493332.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:47:09,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=493335.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:47:12,976 INFO [train.py:968] (0/2) Epoch 11, batch 37800, giga_loss[loss=0.3003, simple_loss=0.3708, pruned_loss=0.1149, over 28819.00 frames. ], tot_loss[loss=0.297, simple_loss=0.365, pruned_loss=0.1145, over 5680406.94 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.372, pruned_loss=0.1243, over 5692179.39 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3614, pruned_loss=0.1121, over 5682978.97 frames. ], batch size: 174, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:47:29,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6021, 3.6692, 1.7909, 1.6179], device='cuda:0'), covar=tensor([0.0897, 0.0242, 0.0855, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0508, 0.0337, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-05 23:47:33,062 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=493364.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:47:39,580 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3607, 1.4920, 1.3077, 1.5332], device='cuda:0'), covar=tensor([0.0806, 0.0313, 0.0318, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-05 23:47:49,457 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-05 23:47:52,188 INFO [train.py:968] (0/2) Epoch 11, batch 37850, giga_loss[loss=0.2266, simple_loss=0.312, pruned_loss=0.07062, over 28898.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3619, pruned_loss=0.1122, over 5677685.01 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3722, pruned_loss=0.1245, over 5686270.89 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3585, pruned_loss=0.1096, over 5684315.15 frames. ], batch size: 145, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:48:09,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.800e+02 1.171e+03 1.640e+03 2.236e+03 1.033e+04, threshold=3.279e+03, percent-clipped=14.0
+2023-03-05 23:48:19,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=493422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:48:33,754 INFO [train.py:968] (0/2) Epoch 11, batch 37900, giga_loss[loss=0.2686, simple_loss=0.3448, pruned_loss=0.0962, over 28165.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3576, pruned_loss=0.1084, over 5688324.90 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3722, pruned_loss=0.1246, over 5688924.16 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3547, pruned_loss=0.1061, over 5691174.18 frames. ], batch size: 368, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:49:01,645 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=493467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:49:21,183 INFO [train.py:968] (0/2) Epoch 11, batch 37950, giga_loss[loss=0.2431, simple_loss=0.3218, pruned_loss=0.08218, over 29047.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3552, pruned_loss=0.106, over 5691557.99 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3721, pruned_loss=0.1245, over 5689743.48 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.353, pruned_loss=0.1041, over 5693171.97 frames. ], batch size: 136, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:49:38,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.453e+02 1.118e+03 1.524e+03 2.057e+03 8.474e+03, threshold=3.047e+03, percent-clipped=9.0
+2023-03-05 23:49:41,048 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-05 23:49:56,944 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-05 23:50:01,970 INFO [train.py:968] (0/2) Epoch 11, batch 38000, libri_loss[loss=0.3353, simple_loss=0.3942, pruned_loss=0.1382, over 29551.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3551, pruned_loss=0.1055, over 5701956.83 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3724, pruned_loss=0.1247, over 5696262.88 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3526, pruned_loss=0.1033, over 5697593.09 frames. ], batch size: 77, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:50:23,261 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=493565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:50:26,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=493568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:50:42,446 INFO [train.py:968] (0/2) Epoch 11, batch 38050, giga_loss[loss=0.2779, simple_loss=0.3564, pruned_loss=0.09971, over 28908.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3596, pruned_loss=0.1085, over 5677607.46 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3731, pruned_loss=0.1253, over 5670706.77 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3564, pruned_loss=0.1057, over 5696769.60 frames. ], batch size: 106, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:50:49,137 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=493597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:50:59,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.129e+02 1.188e+03 1.535e+03 1.878e+03 6.293e+03, threshold=3.070e+03, percent-clipped=6.0
+2023-03-05 23:51:24,168 INFO [train.py:968] (0/2) Epoch 11, batch 38100, giga_loss[loss=0.2763, simple_loss=0.352, pruned_loss=0.1003, over 28965.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3606, pruned_loss=0.1092, over 5682527.92 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3728, pruned_loss=0.125, over 5671709.10 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.358, pruned_loss=0.1068, over 5697050.36 frames. ], batch size: 136, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:52:05,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=493681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:52:12,351 INFO [train.py:968] (0/2) Epoch 11, batch 38150, giga_loss[loss=0.2647, simple_loss=0.3392, pruned_loss=0.09509, over 28595.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3626, pruned_loss=0.1111, over 5681399.46 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3729, pruned_loss=0.1249, over 5673832.13 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3604, pruned_loss=0.1091, over 5691064.91 frames. ], batch size: 60, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:52:31,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.990e+02 1.121e+03 1.411e+03 1.839e+03 5.173e+03, threshold=2.822e+03, percent-clipped=2.0
+2023-03-05 23:52:56,634 INFO [train.py:968] (0/2) Epoch 11, batch 38200, giga_loss[loss=0.2811, simple_loss=0.3491, pruned_loss=0.1065, over 28542.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3628, pruned_loss=0.1117, over 5688497.09 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3729, pruned_loss=0.1249, over 5678183.92 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3608, pruned_loss=0.1099, over 5692421.83 frames. ], batch size: 85, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:53:39,587 INFO [train.py:968] (0/2) Epoch 11, batch 38250, giga_loss[loss=0.3199, simple_loss=0.3819, pruned_loss=0.1289, over 28875.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3634, pruned_loss=0.1124, over 5688508.99 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3733, pruned_loss=0.1251, over 5680961.32 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3612, pruned_loss=0.1105, over 5689164.10 frames. ], batch size: 186, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:53:44,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2668, 1.4358, 1.4906, 1.3338], device='cuda:0'), covar=tensor([0.1697, 0.1822, 0.2192, 0.1782], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0737, 0.0673, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-05 23:53:55,408 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=493807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:53:57,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.884e+02 1.111e+03 1.398e+03 1.928e+03 6.026e+03, threshold=2.796e+03, percent-clipped=6.0
+2023-03-05 23:54:08,830 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=493824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:54:08,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8029, 1.8315, 1.5788, 2.0006], device='cuda:0'), covar=tensor([0.2282, 0.2367, 0.2505, 0.2235], device='cuda:0'), in_proj_covar=tensor([0.1299, 0.0968, 0.1151, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 23:54:11,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=493827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:54:21,679 INFO [train.py:968] (0/2) Epoch 11, batch 38300, giga_loss[loss=0.3062, simple_loss=0.3659, pruned_loss=0.1233, over 26593.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.364, pruned_loss=0.1122, over 5696849.18 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3738, pruned_loss=0.1256, over 5685597.87 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3617, pruned_loss=0.1101, over 5693552.77 frames. ], batch size: 555, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:54:23,189 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 23:54:24,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=493842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:54:34,964 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=493856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:55:03,314 INFO [train.py:968] (0/2) Epoch 11, batch 38350, giga_loss[loss=0.2643, simple_loss=0.3457, pruned_loss=0.09148, over 28861.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3639, pruned_loss=0.1107, over 5702398.10 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3743, pruned_loss=0.1258, over 5689085.38 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3614, pruned_loss=0.1086, over 5696968.31 frames. ], batch size: 186, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:55:20,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.045e+02 1.003e+03 1.271e+03 1.586e+03 3.409e+03, threshold=2.541e+03, percent-clipped=6.0
+2023-03-05 23:55:24,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2087, 1.1956, 3.9099, 3.1211], device='cuda:0'), covar=tensor([0.1602, 0.2594, 0.0403, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0655, 0.0577, 0.0848, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-05 23:55:41,910 INFO [train.py:968] (0/2) Epoch 11, batch 38400, giga_loss[loss=0.2851, simple_loss=0.3636, pruned_loss=0.1033, over 29014.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.364, pruned_loss=0.1101, over 5704562.12 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.374, pruned_loss=0.1256, over 5684297.85 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.362, pruned_loss=0.1082, over 5705356.51 frames. ], batch size: 155, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:56:21,417 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=493985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:56:23,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=493988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:56:23,688 INFO [train.py:968] (0/2) Epoch 11, batch 38450, giga_loss[loss=0.2635, simple_loss=0.3442, pruned_loss=0.09139, over 28885.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3606, pruned_loss=0.1083, over 5705996.43 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3739, pruned_loss=0.1256, over 5687255.63 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.359, pruned_loss=0.1066, over 5704074.53 frames. ], batch size: 199, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:56:33,824 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-494000.pt
+2023-03-05 23:56:46,190 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.717e+02 1.074e+03 1.286e+03 1.797e+03 5.691e+03, threshold=2.573e+03, percent-clipped=10.0
+2023-03-05 23:56:50,872 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=494017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:57:08,699 INFO [train.py:968] (0/2) Epoch 11, batch 38500, giga_loss[loss=0.3041, simple_loss=0.3611, pruned_loss=0.1236, over 27592.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3584, pruned_loss=0.1076, over 5702408.08 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3736, pruned_loss=0.1254, over 5691728.25 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3571, pruned_loss=0.106, over 5697087.38 frames. ], batch size: 472, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:57:13,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4234, 1.6330, 1.3143, 1.5418], device='cuda:0'), covar=tensor([0.2381, 0.2270, 0.2460, 0.2190], device='cuda:0'), in_proj_covar=tensor([0.1288, 0.0958, 0.1139, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-05 23:57:46,681 INFO [train.py:968] (0/2) Epoch 11, batch 38550, giga_loss[loss=0.2783, simple_loss=0.358, pruned_loss=0.09924, over 28951.00 frames. ], tot_loss[loss=0.284, simple_loss=0.356, pruned_loss=0.106, over 5710786.92 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3736, pruned_loss=0.1254, over 5696532.84 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3546, pruned_loss=0.1043, over 5702577.71 frames. ], batch size: 164, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:58:06,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.610e+02 1.146e+03 1.392e+03 2.139e+03 9.735e+03, threshold=2.785e+03, percent-clipped=18.0
+2023-03-05 23:58:21,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=494129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:58:29,360 INFO [train.py:968] (0/2) Epoch 11, batch 38600, giga_loss[loss=0.2863, simple_loss=0.3561, pruned_loss=0.1082, over 28620.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3566, pruned_loss=0.1069, over 5702676.80 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3736, pruned_loss=0.1254, over 5692429.02 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3553, pruned_loss=0.1053, over 5699844.38 frames. ], batch size: 92, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:59:03,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=494182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:59:08,915 INFO [train.py:968] (0/2) Epoch 11, batch 38650, giga_loss[loss=0.2806, simple_loss=0.362, pruned_loss=0.09955, over 28985.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3568, pruned_loss=0.1068, over 5708003.44 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3742, pruned_loss=0.1259, over 5695691.58 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.355, pruned_loss=0.1048, over 5703145.20 frames. ], batch size: 145, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:59:28,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.926e+02 1.007e+03 1.236e+03 1.534e+03 6.270e+03, threshold=2.471e+03, percent-clipped=5.0
+2023-03-05 23:59:49,591 INFO [train.py:968] (0/2) Epoch 11, batch 38700, giga_loss[loss=0.2743, simple_loss=0.3532, pruned_loss=0.09777, over 28684.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3566, pruned_loss=0.1059, over 5711067.34 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3746, pruned_loss=0.1262, over 5699861.09 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3545, pruned_loss=0.1038, over 5703866.73 frames. ], batch size: 99, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:59:54,141 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=494244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:00:10,586 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 00:00:28,752 INFO [train.py:968] (0/2) Epoch 11, batch 38750, giga_loss[loss=0.3678, simple_loss=0.4144, pruned_loss=0.1606, over 27509.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3548, pruned_loss=0.1042, over 5712473.90 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3741, pruned_loss=0.126, over 5695809.35 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3532, pruned_loss=0.1022, over 5710528.48 frames. ], batch size: 472, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:00:46,460 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.763e+02 9.713e+02 1.222e+03 1.561e+03 5.616e+03, threshold=2.443e+03, percent-clipped=5.0
+2023-03-06 00:01:00,649 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=494325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:01:02,986 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=494328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:01:10,635 INFO [train.py:968] (0/2) Epoch 11, batch 38800, giga_loss[loss=0.264, simple_loss=0.3384, pruned_loss=0.09486, over 28748.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3553, pruned_loss=0.105, over 5705942.17 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3742, pruned_loss=0.1259, over 5697815.18 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3537, pruned_loss=0.1032, over 5702716.11 frames. ], batch size: 92, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:01:25,410 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=494357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:01:37,292 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5301, 1.7843, 1.4203, 1.5940], device='cuda:0'), covar=tensor([0.2387, 0.2309, 0.2550, 0.2285], device='cuda:0'), in_proj_covar=tensor([0.1285, 0.0955, 0.1135, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 00:01:49,502 INFO [train.py:968] (0/2) Epoch 11, batch 38850, giga_loss[loss=0.262, simple_loss=0.3389, pruned_loss=0.09258, over 28432.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3523, pruned_loss=0.1033, over 5711361.69 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3742, pruned_loss=0.1258, over 5702095.42 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3507, pruned_loss=0.1016, over 5705238.26 frames. ], batch size: 71, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:02:01,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1065, 1.0382, 3.7996, 3.2201], device='cuda:0'), covar=tensor([0.2175, 0.3215, 0.0731, 0.1030], device='cuda:0'), in_proj_covar=tensor([0.0654, 0.0576, 0.0846, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 00:02:09,409 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6231, 1.7741, 1.6415, 1.5102], device='cuda:0'), covar=tensor([0.1887, 0.1459, 0.1272, 0.1382], device='cuda:0'), in_proj_covar=tensor([0.1723, 0.1614, 0.1588, 0.1683], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 00:02:10,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.784e+02 1.085e+03 1.246e+03 1.634e+03 7.654e+03, threshold=2.493e+03, percent-clipped=9.0
+2023-03-06 00:02:30,684 INFO [train.py:968] (0/2) Epoch 11, batch 38900, giga_loss[loss=0.2693, simple_loss=0.3412, pruned_loss=0.09869, over 28995.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3494, pruned_loss=0.1022, over 5705608.96 frames. ], libri_tot_loss[loss=0.3133, simple_loss=0.3744, pruned_loss=0.1261, over 5704774.88 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3473, pruned_loss=0.1001, over 5698434.78 frames. ], batch size: 164, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:03:12,045 INFO [train.py:968] (0/2) Epoch 11, batch 38950, giga_loss[loss=0.2863, simple_loss=0.3626, pruned_loss=0.105, over 28683.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3479, pruned_loss=0.1016, over 5707500.54 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3746, pruned_loss=0.1262, over 5702151.01 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3461, pruned_loss=0.09974, over 5704239.93 frames. ], batch size: 242, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:03:14,621 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 00:03:27,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=494504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:03:34,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.335e+02 1.141e+03 1.443e+03 2.083e+03 5.731e+03, threshold=2.885e+03, percent-clipped=15.0
+2023-03-06 00:03:54,262 INFO [train.py:968] (0/2) Epoch 11, batch 39000, giga_loss[loss=0.278, simple_loss=0.3519, pruned_loss=0.102, over 28128.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3481, pruned_loss=0.1025, over 5710065.24 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3747, pruned_loss=0.1265, over 5704986.51 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3454, pruned_loss=0.09986, over 5705241.61 frames. ], batch size: 77, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:03:54,266 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 00:04:02,711 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2177, simple_loss=0.3231, pruned_loss=0.05612, over 944034.00 frames. 
+2023-03-06 00:04:02,712 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 00:04:39,627 INFO [train.py:968] (0/2) Epoch 11, batch 39050, libri_loss[loss=0.2723, simple_loss=0.3423, pruned_loss=0.1012, over 29573.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3465, pruned_loss=0.1021, over 5709683.68 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3749, pruned_loss=0.1265, over 5699762.72 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3437, pruned_loss=0.09938, over 5710458.53 frames. ], batch size: 76, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:04:46,417 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 00:04:58,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.747e+02 1.132e+03 1.522e+03 2.216e+03 8.067e+03, threshold=3.044e+03, percent-clipped=14.0
+2023-03-06 00:05:05,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=494619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:05:20,552 INFO [train.py:968] (0/2) Epoch 11, batch 39100, giga_loss[loss=0.2429, simple_loss=0.3155, pruned_loss=0.08515, over 28834.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3449, pruned_loss=0.102, over 5709130.11 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3748, pruned_loss=0.1264, over 5704748.22 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.342, pruned_loss=0.09941, over 5705348.47 frames. ], batch size: 119, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:05:26,909 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=494647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:05:28,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=494650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:05:31,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5729, 1.7922, 1.5783, 1.3479], device='cuda:0'), covar=tensor([0.2181, 0.1526, 0.1383, 0.1854], device='cuda:0'), in_proj_covar=tensor([0.1708, 0.1602, 0.1578, 0.1674], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 00:05:32,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4530, 1.5479, 1.5827, 1.4746], device='cuda:0'), covar=tensor([0.1443, 0.1673, 0.1885, 0.1660], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0731, 0.0669, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 00:05:51,682 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=494679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:06:00,206 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.01 vs. limit=5.0
+2023-03-06 00:06:00,365 INFO [train.py:968] (0/2) Epoch 11, batch 39150, giga_loss[loss=0.2833, simple_loss=0.3516, pruned_loss=0.1075, over 27909.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3409, pruned_loss=0.09987, over 5699830.43 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3747, pruned_loss=0.1265, over 5694147.73 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3381, pruned_loss=0.0973, over 5705968.69 frames. ], batch size: 412, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:06:21,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.404e+02 1.050e+03 1.282e+03 1.731e+03 6.170e+03, threshold=2.563e+03, percent-clipped=5.0
+2023-03-06 00:06:26,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=494719.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:06:45,292 INFO [train.py:968] (0/2) Epoch 11, batch 39200, giga_loss[loss=0.349, simple_loss=0.4037, pruned_loss=0.1471, over 26653.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3397, pruned_loss=0.09964, over 5693791.42 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3751, pruned_loss=0.1267, over 5688057.51 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3366, pruned_loss=0.09694, over 5704638.89 frames. ], batch size: 555, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:07:02,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=494762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:07:04,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=494765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:07:26,137 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-06 00:07:27,125 INFO [train.py:968] (0/2) Epoch 11, batch 39250, giga_loss[loss=0.2528, simple_loss=0.3235, pruned_loss=0.09103, over 28691.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3406, pruned_loss=0.09981, over 5704622.84 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3752, pruned_loss=0.1267, over 5693097.77 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3372, pruned_loss=0.09713, over 5709151.12 frames. ], batch size: 78, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:07:31,751 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=494794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:07:34,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1798, 1.1481, 3.7406, 3.2078], device='cuda:0'), covar=tensor([0.1523, 0.2677, 0.0363, 0.1717], device='cuda:0'), in_proj_covar=tensor([0.0651, 0.0575, 0.0845, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 00:07:51,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.180e+02 9.889e+02 1.269e+03 1.754e+03 6.862e+03, threshold=2.538e+03, percent-clipped=6.0
+2023-03-06 00:07:58,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=494821.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:07:59,763 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=494823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:08:12,074 INFO [train.py:968] (0/2) Epoch 11, batch 39300, giga_loss[loss=0.3757, simple_loss=0.4229, pruned_loss=0.1643, over 27740.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3436, pruned_loss=0.1008, over 5706520.05 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3754, pruned_loss=0.1269, over 5696660.25 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.34, pruned_loss=0.09779, over 5707069.24 frames. ], batch size: 472, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:08:29,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4709, 1.5504, 1.2509, 1.1709], device='cuda:0'), covar=tensor([0.0788, 0.0545, 0.0986, 0.1057], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0437, 0.0500, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 00:08:33,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4184, 1.5904, 1.5763, 1.3431], device='cuda:0'), covar=tensor([0.1385, 0.1846, 0.1929, 0.1944], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0729, 0.0668, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 00:08:38,871 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-06 00:08:42,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2052, 3.9904, 3.7962, 2.0252], device='cuda:0'), covar=tensor([0.0551, 0.0717, 0.0685, 0.1974], device='cuda:0'), in_proj_covar=tensor([0.1040, 0.0977, 0.0847, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 00:08:53,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4444, 3.3878, 1.5020, 1.5763], device='cuda:0'), covar=tensor([0.0897, 0.0342, 0.0882, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0353, 0.0505, 0.0335, 0.0384], device='cuda:0'), out_proj_covar=tensor([0.0022, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-06 00:08:57,528 INFO [train.py:968] (0/2) Epoch 11, batch 39350, giga_loss[loss=0.2992, simple_loss=0.3709, pruned_loss=0.1137, over 28951.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3465, pruned_loss=0.1018, over 5705841.67 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3753, pruned_loss=0.1267, over 5700867.99 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3432, pruned_loss=0.09907, over 5702795.83 frames. ], batch size: 213, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:09:19,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.072e+02 1.064e+03 1.231e+03 1.767e+03 6.062e+03, threshold=2.461e+03, percent-clipped=9.0
+2023-03-06 00:09:31,065 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-06 00:09:40,874 INFO [train.py:968] (0/2) Epoch 11, batch 39400, giga_loss[loss=0.2792, simple_loss=0.3623, pruned_loss=0.09802, over 28809.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3483, pruned_loss=0.1021, over 5689262.45 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3752, pruned_loss=0.1268, over 5686682.35 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.345, pruned_loss=0.09916, over 5700561.58 frames. ], batch size: 284, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:10:24,485 INFO [train.py:968] (0/2) Epoch 11, batch 39450, giga_loss[loss=0.3275, simple_loss=0.3933, pruned_loss=0.1309, over 28602.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3486, pruned_loss=0.1018, over 5675609.98 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3755, pruned_loss=0.127, over 5679674.14 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3455, pruned_loss=0.09908, over 5690292.35 frames. ], batch size: 336, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:10:42,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.070e+02 1.072e+03 1.354e+03 1.782e+03 5.270e+03, threshold=2.709e+03, percent-clipped=11.0
+2023-03-06 00:10:47,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9892, 1.0932, 3.5643, 3.0053], device='cuda:0'), covar=tensor([0.1718, 0.2698, 0.0449, 0.0941], device='cuda:0'), in_proj_covar=tensor([0.0657, 0.0579, 0.0852, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 00:11:02,149 INFO [train.py:968] (0/2) Epoch 11, batch 39500, giga_loss[loss=0.2675, simple_loss=0.3412, pruned_loss=0.09694, over 28803.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3493, pruned_loss=0.1021, over 5690734.14 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.375, pruned_loss=0.1267, over 5686258.55 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3462, pruned_loss=0.09914, over 5696799.31 frames. ], batch size: 199, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:11:09,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3451, 3.5494, 1.4734, 1.4300], device='cuda:0'), covar=tensor([0.0951, 0.0310, 0.0926, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0509, 0.0338, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-06 00:11:36,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=495081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:11:43,207 INFO [train.py:968] (0/2) Epoch 11, batch 39550, libri_loss[loss=0.2815, simple_loss=0.3385, pruned_loss=0.1123, over 29650.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3513, pruned_loss=0.104, over 5692388.97 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3756, pruned_loss=0.1273, over 5695082.59 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3472, pruned_loss=0.1001, over 5689326.29 frames. ], batch size: 73, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:11:46,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=495094.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:12:01,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.427e+02 1.260e+03 1.740e+03 2.823e+03 8.209e+03, threshold=3.480e+03, percent-clipped=26.0
+2023-03-06 00:12:25,358 INFO [train.py:968] (0/2) Epoch 11, batch 39600, giga_loss[loss=0.3714, simple_loss=0.4179, pruned_loss=0.1624, over 26647.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3527, pruned_loss=0.1052, over 5687135.07 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3757, pruned_loss=0.1271, over 5692882.03 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3485, pruned_loss=0.1015, over 5686286.63 frames. ], batch size: 555, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:13:05,733 INFO [train.py:968] (0/2) Epoch 11, batch 39650, giga_loss[loss=0.3378, simple_loss=0.3953, pruned_loss=0.1402, over 27684.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3555, pruned_loss=0.1067, over 5688643.87 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3761, pruned_loss=0.1273, over 5686379.16 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3515, pruned_loss=0.1033, over 5693708.32 frames. ], batch size: 472, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:13:12,245 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=495196.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:13:13,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=495198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:13:27,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.507e+02 1.208e+03 1.498e+03 1.908e+03 4.888e+03, threshold=2.996e+03, percent-clipped=2.0
+2023-03-06 00:13:44,761 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=495237.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:13:45,700 INFO [train.py:968] (0/2) Epoch 11, batch 39700, giga_loss[loss=0.3084, simple_loss=0.3876, pruned_loss=0.1146, over 28924.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3586, pruned_loss=0.1084, over 5702290.06 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3762, pruned_loss=0.1273, over 5691903.99 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3546, pruned_loss=0.1049, over 5701435.20 frames. ], batch size: 174, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:13:46,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=495240.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:13:53,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9174, 1.1318, 1.0730, 0.7837], device='cuda:0'), covar=tensor([0.1678, 0.1784, 0.1006, 0.1645], device='cuda:0'), in_proj_covar=tensor([0.1728, 0.1625, 0.1592, 0.1690], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 00:14:10,021 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=495269.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:14:26,923 INFO [train.py:968] (0/2) Epoch 11, batch 39750, giga_loss[loss=0.2996, simple_loss=0.363, pruned_loss=0.1181, over 28760.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3599, pruned_loss=0.1083, over 5715249.20 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3764, pruned_loss=0.1273, over 5695904.63 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3563, pruned_loss=0.1052, over 5711237.07 frames. ], batch size: 99, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:14:50,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.634e+02 1.222e+03 1.739e+03 2.182e+03 6.146e+03, threshold=3.479e+03, percent-clipped=10.0
+2023-03-06 00:15:11,185 INFO [train.py:968] (0/2) Epoch 11, batch 39800, giga_loss[loss=0.2481, simple_loss=0.3217, pruned_loss=0.08721, over 28607.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3605, pruned_loss=0.1088, over 5704777.32 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3766, pruned_loss=0.1275, over 5686531.11 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3574, pruned_loss=0.1062, over 5710499.37 frames. ], batch size: 85, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:15:11,513 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=495339.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:15:12,687 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=495341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:15:13,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=495342.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:15:15,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=495344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:15:35,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2959, 3.0908, 2.9792, 1.3415], device='cuda:0'), covar=tensor([0.0796, 0.0984, 0.0863, 0.2178], device='cuda:0'), in_proj_covar=tensor([0.1033, 0.0971, 0.0843, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 00:15:38,601 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=495371.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:15:39,934 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=495373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:15:52,771 INFO [train.py:968] (0/2) Epoch 11, batch 39850, giga_loss[loss=0.3515, simple_loss=0.4134, pruned_loss=0.1448, over 29014.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3601, pruned_loss=0.1085, over 5707110.63 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.377, pruned_loss=0.1277, over 5688049.82 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.357, pruned_loss=0.1059, over 5710538.59 frames. ], batch size: 155, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:16:01,506 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-06 00:16:04,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1394, 1.1908, 3.3738, 2.9746], device='cuda:0'), covar=tensor([0.1559, 0.2567, 0.0476, 0.1372], device='cuda:0'), in_proj_covar=tensor([0.0658, 0.0580, 0.0855, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 00:16:13,143 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.407e+02 1.126e+03 1.422e+03 1.888e+03 3.915e+03, threshold=2.843e+03, percent-clipped=3.0
+2023-03-06 00:16:17,699 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-06 00:16:30,733 INFO [train.py:968] (0/2) Epoch 11, batch 39900, giga_loss[loss=0.2639, simple_loss=0.3383, pruned_loss=0.09476, over 29002.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3589, pruned_loss=0.108, over 5711839.23 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.377, pruned_loss=0.1277, over 5693393.75 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.356, pruned_loss=0.1054, over 5710042.08 frames. ], batch size: 155, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:16:44,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=495456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:16:46,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7191, 1.6814, 1.2042, 1.4613], device='cuda:0'), covar=tensor([0.0675, 0.0593, 0.0999, 0.1052], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0437, 0.0498, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 00:16:52,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5886, 1.7216, 1.5558, 1.5153], device='cuda:0'), covar=tensor([0.1480, 0.2003, 0.2102, 0.1954], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0733, 0.0671, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 00:16:59,228 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-06 00:17:10,376 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2789, 0.8825, 0.9352, 1.4034], device='cuda:0'), covar=tensor([0.0737, 0.0348, 0.0346, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0114, 0.0209], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:0')
+2023-03-06 00:17:11,415 INFO [train.py:968] (0/2) Epoch 11, batch 39950, giga_loss[loss=0.2641, simple_loss=0.3363, pruned_loss=0.0959, over 29021.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3579, pruned_loss=0.1078, over 5712159.97 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3777, pruned_loss=0.1281, over 5694646.97 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3544, pruned_loss=0.1048, over 5710107.19 frames. ], batch size: 128, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:17:30,622 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-06 00:17:31,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.290e+02 1.152e+03 1.406e+03 2.160e+03 9.480e+03, threshold=2.812e+03, percent-clipped=13.0
+2023-03-06 00:17:32,441 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-06 00:17:52,820 INFO [train.py:968] (0/2) Epoch 11, batch 40000, giga_loss[loss=0.2372, simple_loss=0.3086, pruned_loss=0.08291, over 28439.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.353, pruned_loss=0.1048, over 5716803.61 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3779, pruned_loss=0.1283, over 5696163.63 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3496, pruned_loss=0.1019, over 5714068.03 frames. ], batch size: 60, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:18:09,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6221, 1.0079, 2.8386, 2.7642], device='cuda:0'), covar=tensor([0.1785, 0.2561, 0.0558, 0.0862], device='cuda:0'), in_proj_covar=tensor([0.0658, 0.0580, 0.0855, 0.0758], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 00:18:31,689 INFO [train.py:968] (0/2) Epoch 11, batch 40050, giga_loss[loss=0.236, simple_loss=0.3207, pruned_loss=0.07559, over 28853.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3517, pruned_loss=0.1042, over 5710551.53 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3781, pruned_loss=0.1284, over 5694856.88 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3479, pruned_loss=0.1009, over 5709483.81 frames. ], batch size: 112, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:18:39,050 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=495599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:18:41,839 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=495602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:18:51,181 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.084e+02 1.147e+03 1.345e+03 1.846e+03 5.137e+03, threshold=2.690e+03, percent-clipped=5.0
+2023-03-06 00:19:00,076 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=495626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:19:03,124 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=495631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:19:08,772 INFO [train.py:968] (0/2) Epoch 11, batch 40100, libri_loss[loss=0.347, simple_loss=0.4035, pruned_loss=0.1453, over 29642.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3532, pruned_loss=0.1037, over 5714255.76 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3775, pruned_loss=0.128, over 5699459.62 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3497, pruned_loss=0.1005, over 5709673.33 frames. ], batch size: 88, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:19:52,427 INFO [train.py:968] (0/2) Epoch 11, batch 40150, giga_loss[loss=0.2852, simple_loss=0.3622, pruned_loss=0.1041, over 28721.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3543, pruned_loss=0.1037, over 5707548.31 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3774, pruned_loss=0.1278, over 5701624.28 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3515, pruned_loss=0.1011, over 5702077.08 frames. ], batch size: 242, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:20:12,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.375e+02 1.202e+03 1.468e+03 1.920e+03 4.661e+03, threshold=2.936e+03, percent-clipped=7.0
+2023-03-06 00:20:30,752 INFO [train.py:968] (0/2) Epoch 11, batch 40200, giga_loss[loss=0.2661, simple_loss=0.3269, pruned_loss=0.1027, over 28528.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3546, pruned_loss=0.1049, over 5707201.00 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3781, pruned_loss=0.1282, over 5696188.43 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.351, pruned_loss=0.1017, over 5708529.20 frames. ], batch size: 78, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:21:10,163 INFO [train.py:968] (0/2) Epoch 11, batch 40250, giga_loss[loss=0.2769, simple_loss=0.3466, pruned_loss=0.1036, over 28707.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3525, pruned_loss=0.105, over 5716936.54 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3778, pruned_loss=0.1281, over 5700141.92 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3494, pruned_loss=0.102, over 5714597.31 frames. ], batch size: 284, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:21:24,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=495806.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:21:32,244 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 00:21:32,469 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.509e+02 1.241e+03 1.665e+03 2.267e+03 1.230e+04, threshold=3.330e+03, percent-clipped=13.0
+2023-03-06 00:21:43,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6371, 4.4387, 4.2410, 1.9138], device='cuda:0'), covar=tensor([0.0543, 0.0759, 0.0705, 0.2070], device='cuda:0'), in_proj_covar=tensor([0.1037, 0.0971, 0.0847, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 00:21:48,843 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=495835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:21:48,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1282, 1.4196, 1.2146, 0.9853], device='cuda:0'), covar=tensor([0.2398, 0.1833, 0.1340, 0.1898], device='cuda:0'), in_proj_covar=tensor([0.1732, 0.1627, 0.1590, 0.1698], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 00:21:51,405 INFO [train.py:968] (0/2) Epoch 11, batch 40300, giga_loss[loss=0.218, simple_loss=0.2913, pruned_loss=0.07232, over 28169.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.351, pruned_loss=0.1061, over 5705945.24 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3779, pruned_loss=0.1283, over 5690196.05 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3478, pruned_loss=0.103, over 5713597.53 frames. ], batch size: 77, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:22:35,023 INFO [train.py:968] (0/2) Epoch 11, batch 40350, libri_loss[loss=0.3401, simple_loss=0.4047, pruned_loss=0.1378, over 29097.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3489, pruned_loss=0.1058, over 5695509.50 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.378, pruned_loss=0.1283, over 5688475.97 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3459, pruned_loss=0.1031, over 5703206.53 frames. ], batch size: 101, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:22:35,476 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 00:22:57,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.278e+02 1.100e+03 1.314e+03 1.721e+03 5.775e+03, threshold=2.628e+03, percent-clipped=5.0
+2023-03-06 00:23:14,604 INFO [train.py:968] (0/2) Epoch 11, batch 40400, giga_loss[loss=0.2485, simple_loss=0.3243, pruned_loss=0.0864, over 29079.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3479, pruned_loss=0.1052, over 5703682.03 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3777, pruned_loss=0.1281, over 5693656.39 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3452, pruned_loss=0.1028, over 5705245.76 frames. ], batch size: 128, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:23:45,038 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=495976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:23:55,481 INFO [train.py:968] (0/2) Epoch 11, batch 40450, giga_loss[loss=0.2792, simple_loss=0.3485, pruned_loss=0.105, over 28915.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3448, pruned_loss=0.1035, over 5706230.23 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3778, pruned_loss=0.1282, over 5695084.39 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.342, pruned_loss=0.1011, over 5706278.01 frames. ], batch size: 227, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:24:04,486 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-496000.pt
+2023-03-06 00:24:07,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=496001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:24:19,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.278e+02 1.194e+03 1.572e+03 2.350e+03 7.227e+03, threshold=3.144e+03, percent-clipped=22.0
+2023-03-06 00:24:34,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3526, 2.0565, 1.6431, 0.5896], device='cuda:0'), covar=tensor([0.3604, 0.1949, 0.3225, 0.4325], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1468, 0.1488, 0.1272], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 00:24:35,010 INFO [train.py:968] (0/2) Epoch 11, batch 40500, giga_loss[loss=0.2069, simple_loss=0.2818, pruned_loss=0.06598, over 28661.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3419, pruned_loss=0.1023, over 5712471.40 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3779, pruned_loss=0.1281, over 5704320.02 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.338, pruned_loss=0.09926, over 5704532.04 frames. ], batch size: 60, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:24:52,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-06 00:25:15,881 INFO [train.py:968] (0/2) Epoch 11, batch 40550, giga_loss[loss=0.2668, simple_loss=0.3349, pruned_loss=0.0993, over 28926.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3376, pruned_loss=0.09941, over 5717807.01 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3772, pruned_loss=0.1277, over 5707459.78 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3344, pruned_loss=0.09688, over 5708797.49 frames. ], batch size: 213, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:25:37,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.388e+02 1.064e+03 1.441e+03 1.799e+03 4.172e+03, threshold=2.882e+03, percent-clipped=5.0
+2023-03-06 00:25:55,776 INFO [train.py:968] (0/2) Epoch 11, batch 40600, giga_loss[loss=0.2773, simple_loss=0.3512, pruned_loss=0.1017, over 29062.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3385, pruned_loss=0.0994, over 5715182.43 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3771, pruned_loss=0.1277, over 5704192.19 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3348, pruned_loss=0.09648, over 5711681.82 frames. ], batch size: 155, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:25:59,499 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=496144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:26:01,488 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=496147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:26:26,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=496176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:26:30,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=496181.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:26:36,381 INFO [train.py:968] (0/2) Epoch 11, batch 40650, giga_loss[loss=0.2614, simple_loss=0.3457, pruned_loss=0.0885, over 28667.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3426, pruned_loss=0.1011, over 5710266.12 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3769, pruned_loss=0.1276, over 5703422.69 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3391, pruned_loss=0.09834, over 5708200.68 frames. ], batch size: 262, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:26:54,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4475, 2.3645, 2.1561, 2.2187], device='cuda:0'), covar=tensor([0.1328, 0.2047, 0.1760, 0.1804], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0733, 0.0675, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 00:26:56,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=496210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:27:00,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.896e+02 1.176e+03 1.454e+03 2.034e+03 6.607e+03, threshold=2.907e+03, percent-clipped=11.0
+2023-03-06 00:27:17,936 INFO [train.py:968] (0/2) Epoch 11, batch 40700, giga_loss[loss=0.279, simple_loss=0.3579, pruned_loss=0.09999, over 28922.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3456, pruned_loss=0.1025, over 5713166.83 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3766, pruned_loss=0.1274, over 5708668.75 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3423, pruned_loss=0.0998, over 5706816.13 frames. ], batch size: 145, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:27:18,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1847, 1.1805, 3.9053, 3.1194], device='cuda:0'), covar=tensor([0.1661, 0.2696, 0.0446, 0.1002], device='cuda:0'), in_proj_covar=tensor([0.0661, 0.0582, 0.0858, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 00:27:26,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2982, 4.0960, 3.8881, 1.8502], device='cuda:0'), covar=tensor([0.0536, 0.0679, 0.0647, 0.2055], device='cuda:0'), in_proj_covar=tensor([0.1043, 0.0974, 0.0849, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 00:27:41,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6082, 1.6488, 1.6462, 1.4925], device='cuda:0'), covar=tensor([0.1371, 0.1891, 0.1918, 0.1799], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0735, 0.0677, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 00:27:57,862 INFO [train.py:968] (0/2) Epoch 11, batch 40750, giga_loss[loss=0.2834, simple_loss=0.3497, pruned_loss=0.1085, over 28441.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3495, pruned_loss=0.1044, over 5709263.24 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3765, pruned_loss=0.1274, over 5713363.82 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.346, pruned_loss=0.1014, over 5699979.51 frames. ], batch size: 78, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:28:21,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.057e+02 1.135e+03 1.397e+03 2.061e+03 3.979e+03, threshold=2.794e+03, percent-clipped=8.0
+2023-03-06 00:28:29,483 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=496324.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:28:31,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=496327.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:28:36,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 00:28:41,787 INFO [train.py:968] (0/2) Epoch 11, batch 40800, libri_loss[loss=0.34, simple_loss=0.3964, pruned_loss=0.1418, over 29256.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3518, pruned_loss=0.105, over 5718131.37 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.1272, over 5717123.32 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3489, pruned_loss=0.1025, over 5707325.93 frames. ], batch size: 94, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 00:28:50,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=496351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:28:52,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=496353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:28:57,231 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=496356.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:28:57,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=496356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:29:24,001 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=496385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:29:27,610 INFO [train.py:968] (0/2) Epoch 11, batch 40850, giga_loss[loss=0.3011, simple_loss=0.3707, pruned_loss=0.1158, over 29070.00 frames. ], tot_loss[loss=0.288, simple_loss=0.357, pruned_loss=0.1095, over 5710566.27 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.376, pruned_loss=0.1271, over 5719516.40 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3544, pruned_loss=0.1072, over 5699939.74 frames. ], batch size: 155, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:29:53,389 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.32 vs. limit=5.0
+2023-03-06 00:29:58,869 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.972e+02 1.436e+03 1.959e+03 3.222e+03 1.053e+04, threshold=3.918e+03, percent-clipped=28.0
+2023-03-06 00:30:14,326 INFO [train.py:968] (0/2) Epoch 11, batch 40900, libri_loss[loss=0.3035, simple_loss=0.354, pruned_loss=0.1265, over 29457.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.363, pruned_loss=0.1149, over 5712802.31 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3758, pruned_loss=0.1269, over 5722937.75 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3603, pruned_loss=0.1124, over 5700038.82 frames. ], batch size: 70, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:30:58,906 INFO [train.py:968] (0/2) Epoch 11, batch 40950, giga_loss[loss=0.3255, simple_loss=0.3928, pruned_loss=0.1291, over 28688.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3685, pruned_loss=0.1188, over 5711182.23 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3747, pruned_loss=0.1262, over 5727327.81 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.367, pruned_loss=0.1173, over 5696823.42 frames. ], batch size: 242, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:31:06,448 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=496494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:31:09,418 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=496497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:31:13,705 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-06 00:31:26,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.716e+03 2.133e+03 3.099e+03 7.016e+03, threshold=4.266e+03, percent-clipped=10.0
+2023-03-06 00:31:34,062 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=496526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:31:44,890 INFO [train.py:968] (0/2) Epoch 11, batch 41000, giga_loss[loss=0.3619, simple_loss=0.4135, pruned_loss=0.1551, over 27884.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3739, pruned_loss=0.1229, over 5706231.95 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3745, pruned_loss=0.126, over 5732720.19 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3728, pruned_loss=0.1217, over 5689270.09 frames. ], batch size: 412, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:31:58,949 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-06 00:32:31,114 INFO [train.py:968] (0/2) Epoch 11, batch 41050, giga_loss[loss=0.5275, simple_loss=0.5159, pruned_loss=0.2696, over 26580.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3801, pruned_loss=0.1282, over 5696940.10 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.375, pruned_loss=0.1263, over 5725191.22 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3789, pruned_loss=0.1269, over 5690790.86 frames. ], batch size: 555, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:32:55,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8744, 1.9290, 1.3989, 1.5247], device='cuda:0'), covar=tensor([0.0744, 0.0556, 0.0932, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0441, 0.0499, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 00:32:57,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.913e+02 1.599e+03 2.017e+03 2.796e+03 5.194e+03, threshold=4.035e+03, percent-clipped=3.0
+2023-03-06 00:33:14,310 INFO [train.py:968] (0/2) Epoch 11, batch 41100, giga_loss[loss=0.4307, simple_loss=0.4563, pruned_loss=0.2025, over 27446.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3848, pruned_loss=0.1319, over 5685547.53 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3751, pruned_loss=0.1262, over 5713900.05 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3841, pruned_loss=0.1312, over 5688191.69 frames. ], batch size: 472, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:33:55,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2708, 1.7010, 1.3957, 1.4490], device='cuda:0'), covar=tensor([0.0748, 0.0334, 0.0312, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0081, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 00:34:06,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2351, 1.4292, 1.3065, 1.0730], device='cuda:0'), covar=tensor([0.1816, 0.1825, 0.1213, 0.1641], device='cuda:0'), in_proj_covar=tensor([0.1728, 0.1632, 0.1601, 0.1695], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 00:34:06,856 INFO [train.py:968] (0/2) Epoch 11, batch 41150, giga_loss[loss=0.329, simple_loss=0.3941, pruned_loss=0.1319, over 28973.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3877, pruned_loss=0.1355, over 5664181.16 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3744, pruned_loss=0.1257, over 5717539.20 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.388, pruned_loss=0.1356, over 5662202.69 frames. ], batch size: 136, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:34:10,991 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 00:34:14,526 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=496696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:34:41,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.479e+02 1.709e+03 2.169e+03 2.892e+03 1.212e+04, threshold=4.338e+03, percent-clipped=11.0
+2023-03-06 00:35:06,128 INFO [train.py:968] (0/2) Epoch 11, batch 41200, giga_loss[loss=0.3583, simple_loss=0.4061, pruned_loss=0.1553, over 28825.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.39, pruned_loss=0.1384, over 5665272.32 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3744, pruned_loss=0.1257, over 5717539.20 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3902, pruned_loss=0.1384, over 5663732.44 frames. ], batch size: 199, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:36:00,746 INFO [train.py:968] (0/2) Epoch 11, batch 41250, giga_loss[loss=0.3602, simple_loss=0.4059, pruned_loss=0.1573, over 28868.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3946, pruned_loss=0.1439, over 5629325.40 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3743, pruned_loss=0.1257, over 5709591.07 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.395, pruned_loss=0.1441, over 5635121.06 frames. ], batch size: 186, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:36:31,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.184e+02 1.781e+03 2.374e+03 3.107e+03 7.115e+03, threshold=4.749e+03, percent-clipped=12.0
+2023-03-06 00:36:52,280 INFO [train.py:968] (0/2) Epoch 11, batch 41300, giga_loss[loss=0.3716, simple_loss=0.4221, pruned_loss=0.1606, over 28932.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3986, pruned_loss=0.1474, over 5630196.51 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3745, pruned_loss=0.1258, over 5713532.93 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.3993, pruned_loss=0.1479, over 5629401.56 frames. ], batch size: 213, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:37:24,615 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-06 00:37:43,403 INFO [train.py:968] (0/2) Epoch 11, batch 41350, giga_loss[loss=0.3658, simple_loss=0.4052, pruned_loss=0.1632, over 28518.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3981, pruned_loss=0.1477, over 5629207.00 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3738, pruned_loss=0.1254, over 5718590.06 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.4002, pruned_loss=0.1492, over 5621483.01 frames. ], batch size: 336, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:37:47,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7181, 1.8857, 1.9874, 1.4977], device='cuda:0'), covar=tensor([0.1713, 0.2114, 0.1289, 0.1536], device='cuda:0'), in_proj_covar=tensor([0.0824, 0.0694, 0.0863, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 00:38:06,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=496911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:38:12,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.854e+03 2.444e+03 3.285e+03 1.162e+04, threshold=4.888e+03, percent-clipped=10.0
+2023-03-06 00:38:31,208 INFO [train.py:968] (0/2) Epoch 11, batch 41400, libri_loss[loss=0.3411, simple_loss=0.4053, pruned_loss=0.1384, over 29526.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3971, pruned_loss=0.1477, over 5636810.53 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3732, pruned_loss=0.1249, over 5723125.63 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.3997, pruned_loss=0.1498, over 5624743.06 frames. ], batch size: 89, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:39:04,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4197, 3.4837, 1.5219, 1.4578], device='cuda:0'), covar=tensor([0.0923, 0.0295, 0.0821, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0511, 0.0338, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-06 00:39:17,665 INFO [train.py:968] (0/2) Epoch 11, batch 41450, giga_loss[loss=0.4038, simple_loss=0.4211, pruned_loss=0.1933, over 23399.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3956, pruned_loss=0.1461, over 5645440.70 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3732, pruned_loss=0.1249, over 5717530.18 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3985, pruned_loss=0.1486, over 5637989.22 frames. ], batch size: 705, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:39:49,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.811e+02 1.670e+03 2.056e+03 2.809e+03 1.207e+04, threshold=4.112e+03, percent-clipped=5.0
+2023-03-06 00:39:49,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-06 00:39:58,299 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 00:40:10,026 INFO [train.py:968] (0/2) Epoch 11, batch 41500, giga_loss[loss=0.2996, simple_loss=0.3727, pruned_loss=0.1132, over 29053.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.395, pruned_loss=0.1443, over 5643340.08 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3729, pruned_loss=0.1248, over 5710278.67 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.3978, pruned_loss=0.1465, over 5642081.64 frames. ], batch size: 155, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:40:43,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:40:43,403 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-06 00:41:00,266 INFO [train.py:968] (0/2) Epoch 11, batch 41550, libri_loss[loss=0.3122, simple_loss=0.3769, pruned_loss=0.1238, over 26010.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3959, pruned_loss=0.1436, over 5657090.95 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3733, pruned_loss=0.125, over 5709286.10 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3981, pruned_loss=0.1455, over 5656250.03 frames. ], batch size: 136, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:41:34,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.932e+02 1.660e+03 2.053e+03 2.800e+03 7.829e+03, threshold=4.107e+03, percent-clipped=9.0
+2023-03-06 00:41:59,051 INFO [train.py:968] (0/2) Epoch 11, batch 41600, giga_loss[loss=0.3415, simple_loss=0.3942, pruned_loss=0.1444, over 28742.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3965, pruned_loss=0.1444, over 5645220.65 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3732, pruned_loss=0.1249, over 5712891.34 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.3989, pruned_loss=0.1464, over 5640197.42 frames. ], batch size: 262, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 00:42:12,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4596, 2.0964, 1.4347, 0.6344], device='cuda:0'), covar=tensor([0.4370, 0.2162, 0.3262, 0.4868], device='cuda:0'), in_proj_covar=tensor([0.1573, 0.1504, 0.1510, 0.1297], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 00:42:46,678 INFO [train.py:968] (0/2) Epoch 11, batch 41650, giga_loss[loss=0.343, simple_loss=0.3937, pruned_loss=0.1461, over 27701.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3926, pruned_loss=0.14, over 5638937.14 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.373, pruned_loss=0.1247, over 5703416.25 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3952, pruned_loss=0.1421, over 5640964.09 frames. ], batch size: 474, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 00:42:53,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3490, 3.1000, 1.4593, 1.4511], device='cuda:0'), covar=tensor([0.0899, 0.0334, 0.0842, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0511, 0.0339, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-06 00:43:11,251 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=497214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:43:13,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=497217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:43:15,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.084e+02 1.489e+03 1.907e+03 2.577e+03 7.794e+03, threshold=3.814e+03, percent-clipped=7.0
+2023-03-06 00:43:33,384 INFO [train.py:968] (0/2) Epoch 11, batch 41700, giga_loss[loss=0.3179, simple_loss=0.3791, pruned_loss=0.1283, over 28726.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.389, pruned_loss=0.1363, over 5652733.53 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3725, pruned_loss=0.1247, over 5710500.52 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3922, pruned_loss=0.1386, over 5645884.92 frames. ], batch size: 92, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 00:43:34,257 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:43:40,031 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=497246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:44:22,641 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497286.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:44:22,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:44:24,519 INFO [train.py:968] (0/2) Epoch 11, batch 41750, giga_loss[loss=0.361, simple_loss=0.4099, pruned_loss=0.156, over 28611.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3848, pruned_loss=0.1323, over 5661586.91 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3724, pruned_loss=0.1246, over 5712097.50 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3875, pruned_loss=0.1343, over 5654072.82 frames. ], batch size: 336, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 00:44:54,194 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.052e+02 1.850e+03 2.233e+03 3.097e+03 5.694e+03, threshold=4.466e+03, percent-clipped=13.0
+2023-03-06 00:45:14,064 INFO [train.py:968] (0/2) Epoch 11, batch 41800, giga_loss[loss=0.323, simple_loss=0.379, pruned_loss=0.1336, over 28976.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3816, pruned_loss=0.13, over 5652823.89 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3726, pruned_loss=0.1248, over 5705520.37 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3838, pruned_loss=0.1315, over 5651967.76 frames. ], batch size: 213, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:45:30,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:45:46,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 00:46:04,758 INFO [train.py:968] (0/2) Epoch 11, batch 41850, giga_loss[loss=0.3992, simple_loss=0.4341, pruned_loss=0.1822, over 26708.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3824, pruned_loss=0.131, over 5653587.09 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3727, pruned_loss=0.1248, over 5712815.41 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3844, pruned_loss=0.1325, over 5643890.24 frames. ], batch size: 555, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:46:34,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.184e+02 1.609e+03 2.168e+03 2.997e+03 1.562e+04, threshold=4.336e+03, percent-clipped=8.0
+2023-03-06 00:46:43,291 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=497429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:46:45,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=497432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:46:52,721 INFO [train.py:968] (0/2) Epoch 11, batch 41900, giga_loss[loss=0.2827, simple_loss=0.3624, pruned_loss=0.1016, over 28932.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3831, pruned_loss=0.1313, over 5658247.41 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3726, pruned_loss=0.1248, over 5706533.37 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.385, pruned_loss=0.1325, over 5656212.98 frames. ], batch size: 174, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:46:55,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2905, 3.0763, 2.9142, 1.4496], device='cuda:0'), covar=tensor([0.0878, 0.1049, 0.0960, 0.2284], device='cuda:0'), in_proj_covar=tensor([0.1063, 0.1002, 0.0871, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 00:47:16,577 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=497461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:47:33,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3417, 4.1304, 3.8923, 1.8094], device='cuda:0'), covar=tensor([0.0536, 0.0738, 0.0755, 0.2135], device='cuda:0'), in_proj_covar=tensor([0.1063, 0.1001, 0.0870, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 00:47:49,615 INFO [train.py:968] (0/2) Epoch 11, batch 41950, giga_loss[loss=0.2903, simple_loss=0.3556, pruned_loss=0.1125, over 28874.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3796, pruned_loss=0.1278, over 5663307.51 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3728, pruned_loss=0.1249, over 5704333.61 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3809, pruned_loss=0.1287, over 5662976.18 frames. ], batch size: 227, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:48:22,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.132e+02 1.367e+03 1.791e+03 2.831e+03 1.075e+04, threshold=3.581e+03, percent-clipped=8.0
+2023-03-06 00:48:39,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6377, 1.7551, 1.7869, 1.5058], device='cuda:0'), covar=tensor([0.1351, 0.1555, 0.1714, 0.1588], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0732, 0.0675, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 00:48:39,705 INFO [train.py:968] (0/2) Epoch 11, batch 42000, libri_loss[loss=0.2859, simple_loss=0.3496, pruned_loss=0.1111, over 29524.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3799, pruned_loss=0.1257, over 5662852.84 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.373, pruned_loss=0.1251, over 5697486.16 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.381, pruned_loss=0.1263, over 5667856.08 frames. ], batch size: 81, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:48:39,709 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 00:48:47,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3601, 1.7072, 1.3401, 1.3287], device='cuda:0'), covar=tensor([0.2734, 0.2488, 0.2758, 0.2169], device='cuda:0'), in_proj_covar=tensor([0.1296, 0.0964, 0.1151, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 00:48:48,332 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2129, simple_loss=0.317, pruned_loss=0.0544, over 944034.00 frames. 
+2023-03-06 00:48:48,332 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 00:49:38,860 INFO [train.py:968] (0/2) Epoch 11, batch 42050, giga_loss[loss=0.3204, simple_loss=0.3943, pruned_loss=0.1232, over 28954.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3823, pruned_loss=0.1261, over 5657190.07 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3736, pruned_loss=0.1256, over 5688613.20 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3827, pruned_loss=0.1261, over 5668833.71 frames. ], batch size: 136, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:49:51,446 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:50:03,423 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:50:07,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6446, 1.7049, 1.6712, 1.4832], device='cuda:0'), covar=tensor([0.1426, 0.1817, 0.1938, 0.1828], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0732, 0.0673, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 00:50:08,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.493e+02 1.581e+03 2.057e+03 2.634e+03 7.999e+03, threshold=4.115e+03, percent-clipped=8.0
+2023-03-06 00:50:15,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3381, 1.6819, 1.2624, 1.5001], device='cuda:0'), covar=tensor([0.2306, 0.2204, 0.2457, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.0962, 0.1148, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 00:50:28,859 INFO [train.py:968] (0/2) Epoch 11, batch 42100, giga_loss[loss=0.2959, simple_loss=0.3631, pruned_loss=0.1143, over 28911.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3827, pruned_loss=0.1277, over 5658956.67 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3731, pruned_loss=0.1254, over 5692436.40 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3836, pruned_loss=0.1278, over 5664265.59 frames. ], batch size: 106, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:50:50,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497661.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:51:11,257 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:51:14,195 INFO [train.py:968] (0/2) Epoch 11, batch 42150, giga_loss[loss=0.3651, simple_loss=0.4056, pruned_loss=0.1623, over 28933.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3827, pruned_loss=0.128, over 5667364.95 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.373, pruned_loss=0.1253, over 5696710.31 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3837, pruned_loss=0.1283, over 5667279.55 frames. ], batch size: 199, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:51:44,831 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.759e+03 2.211e+03 2.775e+03 5.977e+03, threshold=4.422e+03, percent-clipped=5.0
+2023-03-06 00:51:51,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:52:00,642 INFO [train.py:968] (0/2) Epoch 11, batch 42200, giga_loss[loss=0.2761, simple_loss=0.3432, pruned_loss=0.1045, over 28575.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3797, pruned_loss=0.1268, over 5672497.80 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3728, pruned_loss=0.125, over 5700181.37 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3809, pruned_loss=0.1274, over 5668623.52 frames. ], batch size: 85, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:52:19,600 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=497758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:52:21,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=497761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:52:48,060 INFO [train.py:968] (0/2) Epoch 11, batch 42250, giga_loss[loss=0.2761, simple_loss=0.3495, pruned_loss=0.1014, over 28967.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3776, pruned_loss=0.1264, over 5661957.15 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3727, pruned_loss=0.1247, over 5699903.00 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3788, pruned_loss=0.1271, over 5658320.81 frames. ], batch size: 106, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:52:48,901 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=497790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:52:53,317 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-06 00:53:03,584 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=497804.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:53:05,686 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=497807.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:53:22,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.373e+02 1.675e+03 2.322e+03 3.608e+03 1.342e+04, threshold=4.643e+03, percent-clipped=10.0
+2023-03-06 00:53:25,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1064, 1.1015, 1.1351, 1.3615], device='cuda:0'), covar=tensor([0.0839, 0.0326, 0.0286, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 00:53:37,466 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=497836.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:53:40,359 INFO [train.py:968] (0/2) Epoch 11, batch 42300, giga_loss[loss=0.3641, simple_loss=0.4122, pruned_loss=0.158, over 28944.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3782, pruned_loss=0.1264, over 5652292.12 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3729, pruned_loss=0.125, over 5681548.58 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3789, pruned_loss=0.1267, over 5664974.36 frames. ], batch size: 106, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:53:52,453 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:54:06,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=497871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:54:08,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=497874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:54:22,160 INFO [train.py:968] (0/2) Epoch 11, batch 42350, giga_loss[loss=0.3171, simple_loss=0.3848, pruned_loss=0.1246, over 28706.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3781, pruned_loss=0.125, over 5676572.45 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3726, pruned_loss=0.1248, over 5689202.25 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3791, pruned_loss=0.1255, over 5679356.41 frames. ], batch size: 242, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:54:36,013 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=497903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:54:47,510 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 00:54:51,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.710e+02 1.342e+03 1.681e+03 2.212e+03 5.656e+03, threshold=3.361e+03, percent-clipped=2.0
+2023-03-06 00:55:09,122 INFO [train.py:968] (0/2) Epoch 11, batch 42400, giga_loss[loss=0.3119, simple_loss=0.3749, pruned_loss=0.1244, over 27932.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3785, pruned_loss=0.1248, over 5670957.31 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3731, pruned_loss=0.1249, over 5684717.69 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3791, pruned_loss=0.1251, over 5676449.80 frames. ], batch size: 412, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:55:40,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:55:54,441 INFO [train.py:968] (0/2) Epoch 11, batch 42450, giga_loss[loss=0.2751, simple_loss=0.3488, pruned_loss=0.1007, over 28564.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3765, pruned_loss=0.1237, over 5681727.18 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3731, pruned_loss=0.1248, over 5693135.37 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3772, pruned_loss=0.124, over 5678534.11 frames. ], batch size: 307, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:56:02,400 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-498000.pt
+2023-03-06 00:56:21,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.549e+02 1.617e+03 2.146e+03 2.985e+03 1.086e+04, threshold=4.291e+03, percent-clipped=20.0
+2023-03-06 00:56:24,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3041, 1.6813, 1.2141, 0.7638], device='cuda:0'), covar=tensor([0.3552, 0.2281, 0.2237, 0.3649], device='cuda:0'), in_proj_covar=tensor([0.1575, 0.1498, 0.1502, 0.1292], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 00:56:39,067 INFO [train.py:968] (0/2) Epoch 11, batch 42500, giga_loss[loss=0.3474, simple_loss=0.4065, pruned_loss=0.1442, over 28650.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3756, pruned_loss=0.1236, over 5667157.55 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3737, pruned_loss=0.1252, over 5677758.37 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3758, pruned_loss=0.1235, over 5677069.89 frames. ], batch size: 307, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:56:59,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=498059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:57:27,605 INFO [train.py:968] (0/2) Epoch 11, batch 42550, giga_loss[loss=0.2959, simple_loss=0.3559, pruned_loss=0.118, over 28847.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3747, pruned_loss=0.1242, over 5663130.70 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3733, pruned_loss=0.1249, over 5681516.76 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3752, pruned_loss=0.1243, over 5667407.06 frames. ], batch size: 99, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:57:57,724 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=498119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:58:02,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=498122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:58:02,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.857e+02 1.692e+03 2.253e+03 2.864e+03 6.081e+03, threshold=4.505e+03, percent-clipped=6.0
+2023-03-06 00:58:17,880 INFO [train.py:968] (0/2) Epoch 11, batch 42600, giga_loss[loss=0.3191, simple_loss=0.3792, pruned_loss=0.1295, over 28623.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3737, pruned_loss=0.1243, over 5658224.77 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3735, pruned_loss=0.125, over 5679933.27 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3739, pruned_loss=0.1243, over 5662775.93 frames. ], batch size: 336, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:58:29,317 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=498151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:59:07,379 INFO [train.py:968] (0/2) Epoch 11, batch 42650, giga_loss[loss=0.2843, simple_loss=0.3522, pruned_loss=0.1082, over 28905.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1234, over 5671759.41 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3737, pruned_loss=0.1252, over 5682563.83 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3716, pruned_loss=0.1232, over 5672469.23 frames. ], batch size: 145, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:59:12,431 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3606, 3.6409, 1.5609, 1.4717], device='cuda:0'), covar=tensor([0.0896, 0.0341, 0.0823, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0515, 0.0341, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 00:59:20,424 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-06 00:59:20,902 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=498202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:59:22,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=498205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:59:38,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.713e+03 2.130e+03 3.143e+03 8.141e+03, threshold=4.260e+03, percent-clipped=9.0
+2023-03-06 00:59:44,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=498228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:59:50,682 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=498234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:59:55,592 INFO [train.py:968] (0/2) Epoch 11, batch 42700, giga_loss[loss=0.4106, simple_loss=0.4313, pruned_loss=0.1949, over 26655.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3721, pruned_loss=0.1239, over 5675367.46 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3738, pruned_loss=0.1253, over 5684438.89 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3719, pruned_loss=0.1237, over 5674070.88 frames. ], batch size: 555, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:00:44,024 INFO [train.py:968] (0/2) Epoch 11, batch 42750, giga_loss[loss=0.3066, simple_loss=0.3695, pruned_loss=0.1219, over 29022.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3719, pruned_loss=0.1234, over 5680195.22 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3736, pruned_loss=0.1251, over 5686536.09 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3719, pruned_loss=0.1234, over 5677066.20 frames. ], batch size: 106, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:01:16,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.374e+02 1.570e+03 1.906e+03 2.573e+03 6.239e+03, threshold=3.812e+03, percent-clipped=8.0
+2023-03-06 01:01:32,005 INFO [train.py:968] (0/2) Epoch 11, batch 42800, giga_loss[loss=0.3259, simple_loss=0.3832, pruned_loss=0.1343, over 28192.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.373, pruned_loss=0.1233, over 5675987.08 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3735, pruned_loss=0.125, over 5688471.66 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3731, pruned_loss=0.1234, over 5671881.98 frames. ], batch size: 368, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:02:00,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=498371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:02:03,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=498374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:02:03,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1096, 1.2215, 1.0874, 1.0794], device='cuda:0'), covar=tensor([0.1528, 0.1496, 0.1123, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.1732, 0.1649, 0.1613, 0.1705], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 01:02:18,638 INFO [train.py:968] (0/2) Epoch 11, batch 42850, libri_loss[loss=0.3033, simple_loss=0.3695, pruned_loss=0.1186, over 29525.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3741, pruned_loss=0.1236, over 5675178.36 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3739, pruned_loss=0.1253, over 5691782.68 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3738, pruned_loss=0.1233, over 5668733.17 frames. ], batch size: 83, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:02:30,668 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=498403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:02:38,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3590, 1.2598, 4.6567, 3.4602], device='cuda:0'), covar=tensor([0.1702, 0.2634, 0.0360, 0.0850], device='cuda:0'), in_proj_covar=tensor([0.0664, 0.0583, 0.0860, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 01:02:50,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.561e+03 1.948e+03 2.991e+03 7.703e+03, threshold=3.897e+03, percent-clipped=14.0
+2023-03-06 01:03:04,897 INFO [train.py:968] (0/2) Epoch 11, batch 42900, giga_loss[loss=0.3214, simple_loss=0.3794, pruned_loss=0.1317, over 28598.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.375, pruned_loss=0.1242, over 5680723.21 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3736, pruned_loss=0.125, over 5696317.52 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3751, pruned_loss=0.1242, over 5671085.99 frames. ], batch size: 307, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:03:53,812 INFO [train.py:968] (0/2) Epoch 11, batch 42950, giga_loss[loss=0.3408, simple_loss=0.3964, pruned_loss=0.1426, over 28612.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3764, pruned_loss=0.1262, over 5664257.36 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3728, pruned_loss=0.1244, over 5691736.14 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3773, pruned_loss=0.1267, over 5659908.88 frames. ], batch size: 307, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:04:27,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.769e+03 2.277e+03 3.009e+03 7.235e+03, threshold=4.553e+03, percent-clipped=13.0
+2023-03-06 01:04:42,437 INFO [train.py:968] (0/2) Epoch 11, batch 43000, giga_loss[loss=0.3139, simple_loss=0.3689, pruned_loss=0.1295, over 28722.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3795, pruned_loss=0.1295, over 5663624.29 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3732, pruned_loss=0.1247, over 5685204.04 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.38, pruned_loss=0.1298, over 5665082.67 frames. ], batch size: 99, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:05:09,531 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=498562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:05:10,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4924, 1.7005, 1.3965, 1.5828], device='cuda:0'), covar=tensor([0.2175, 0.2175, 0.2366, 0.1969], device='cuda:0'), in_proj_covar=tensor([0.1297, 0.0965, 0.1147, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 01:05:37,134 INFO [train.py:968] (0/2) Epoch 11, batch 43050, giga_loss[loss=0.2825, simple_loss=0.355, pruned_loss=0.105, over 28820.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3791, pruned_loss=0.1308, over 5657286.81 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.1241, over 5691375.07 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3806, pruned_loss=0.1319, over 5651809.94 frames. ], batch size: 119, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:06:00,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7446, 1.7612, 1.2599, 1.3573], device='cuda:0'), covar=tensor([0.0726, 0.0571, 0.1024, 0.1063], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0445, 0.0504, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 01:06:13,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.226e+03 1.766e+03 2.498e+03 3.535e+03 9.783e+03, threshold=4.997e+03, percent-clipped=11.0
+2023-03-06 01:06:28,045 INFO [train.py:968] (0/2) Epoch 11, batch 43100, giga_loss[loss=0.3217, simple_loss=0.3738, pruned_loss=0.1347, over 28910.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3809, pruned_loss=0.1327, over 5659183.27 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3728, pruned_loss=0.1244, over 5692498.84 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3818, pruned_loss=0.1334, over 5653121.25 frames. ], batch size: 112, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:07:10,704 INFO [train.py:968] (0/2) Epoch 11, batch 43150, giga_loss[loss=0.264, simple_loss=0.3426, pruned_loss=0.09272, over 29068.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3794, pruned_loss=0.1312, over 5675853.01 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3732, pruned_loss=0.1246, over 5699133.68 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.38, pruned_loss=0.1318, over 5664107.32 frames. ], batch size: 155, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:07:41,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.495e+02 1.617e+03 1.945e+03 2.628e+03 1.081e+04, threshold=3.889e+03, percent-clipped=6.0
+2023-03-06 01:07:52,984 INFO [train.py:968] (0/2) Epoch 11, batch 43200, libri_loss[loss=0.2899, simple_loss=0.3621, pruned_loss=0.1088, over 29260.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3774, pruned_loss=0.1298, over 5682162.53 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3728, pruned_loss=0.1244, over 5707249.97 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3784, pruned_loss=0.1308, over 5664497.14 frames. ], batch size: 97, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:08:36,767 INFO [train.py:968] (0/2) Epoch 11, batch 43250, giga_loss[loss=0.3024, simple_loss=0.3598, pruned_loss=0.1225, over 27514.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3761, pruned_loss=0.1269, over 5681059.38 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3722, pruned_loss=0.124, over 5698806.65 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3775, pruned_loss=0.128, over 5673716.35 frames. ], batch size: 472, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:09:10,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2470, 1.2264, 1.0978, 0.9107], device='cuda:0'), covar=tensor([0.0622, 0.0367, 0.0745, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0443, 0.0503, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 01:09:11,921 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.530e+03 2.077e+03 2.854e+03 9.132e+03, threshold=4.154e+03, percent-clipped=10.0
+2023-03-06 01:09:26,341 INFO [train.py:968] (0/2) Epoch 11, batch 43300, giga_loss[loss=0.3029, simple_loss=0.3465, pruned_loss=0.1296, over 23539.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3742, pruned_loss=0.1257, over 5672713.86 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3725, pruned_loss=0.1243, over 5699533.62 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3751, pruned_loss=0.1265, over 5666315.96 frames. ], batch size: 705, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:10:08,535 INFO [train.py:968] (0/2) Epoch 11, batch 43350, giga_loss[loss=0.3447, simple_loss=0.3919, pruned_loss=0.1487, over 27486.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3733, pruned_loss=0.1256, over 5666021.92 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3724, pruned_loss=0.124, over 5695714.84 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3742, pruned_loss=0.1265, over 5663250.22 frames. ], batch size: 472, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:10:43,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.697e+03 2.097e+03 3.280e+03 7.293e+03, threshold=4.194e+03, percent-clipped=16.0
+2023-03-06 01:10:58,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=498937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:10:59,267 INFO [train.py:968] (0/2) Epoch 11, batch 43400, giga_loss[loss=0.3107, simple_loss=0.362, pruned_loss=0.1297, over 28719.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3718, pruned_loss=0.1254, over 5661938.85 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3724, pruned_loss=0.124, over 5695732.27 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3725, pruned_loss=0.1261, over 5659638.88 frames. ], batch size: 92, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:11:43,502 INFO [train.py:968] (0/2) Epoch 11, batch 43450, giga_loss[loss=0.329, simple_loss=0.3915, pruned_loss=0.1332, over 28865.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3755, pruned_loss=0.1281, over 5667583.78 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3724, pruned_loss=0.1239, over 5697303.71 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3761, pruned_loss=0.1288, over 5663705.05 frames. ], batch size: 112, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:11:47,060 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=498993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:12:08,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3887, 2.0106, 1.4889, 1.6210], device='cuda:0'), covar=tensor([0.0757, 0.0267, 0.0321, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0114, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 01:12:16,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.572e+02 1.578e+03 2.083e+03 3.025e+03 1.153e+04, threshold=4.166e+03, percent-clipped=13.0
+2023-03-06 01:12:30,391 INFO [train.py:968] (0/2) Epoch 11, batch 43500, giga_loss[loss=0.32, simple_loss=0.3955, pruned_loss=0.1222, over 28642.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3795, pruned_loss=0.1291, over 5662703.13 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3723, pruned_loss=0.1239, over 5690758.10 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3802, pruned_loss=0.1297, over 5664828.80 frames. ], batch size: 307, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:13:10,030 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=499080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:13:12,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=499083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:13:17,546 INFO [train.py:968] (0/2) Epoch 11, batch 43550, giga_loss[loss=0.3205, simple_loss=0.3932, pruned_loss=0.1239, over 28892.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3806, pruned_loss=0.1275, over 5661957.20 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3711, pruned_loss=0.123, over 5698082.07 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3826, pruned_loss=0.129, over 5656182.01 frames. ], batch size: 112, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:13:40,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=499112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:13:50,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.330e+02 1.507e+03 1.877e+03 2.578e+03 7.810e+03, threshold=3.754e+03, percent-clipped=8.0
+2023-03-06 01:14:02,417 INFO [train.py:968] (0/2) Epoch 11, batch 43600, giga_loss[loss=0.3061, simple_loss=0.3752, pruned_loss=0.1185, over 28980.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3816, pruned_loss=0.1286, over 5664330.48 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3706, pruned_loss=0.123, over 5692634.63 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3841, pruned_loss=0.1302, over 5662280.07 frames. ], batch size: 136, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 01:14:48,839 INFO [train.py:968] (0/2) Epoch 11, batch 43650, giga_loss[loss=0.4262, simple_loss=0.4494, pruned_loss=0.2016, over 26666.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.384, pruned_loss=0.1304, over 5675168.48 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3703, pruned_loss=0.1227, over 5697290.55 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3865, pruned_loss=0.132, over 5669071.18 frames. ], batch size: 555, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:15:09,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8634, 4.6811, 4.4041, 2.1883], device='cuda:0'), covar=tensor([0.0531, 0.0733, 0.0735, 0.1970], device='cuda:0'), in_proj_covar=tensor([0.1061, 0.1000, 0.0870, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 01:15:26,404 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.918e+02 1.648e+03 2.412e+03 3.483e+03 8.243e+03, threshold=4.824e+03, percent-clipped=18.0
+2023-03-06 01:15:36,912 INFO [train.py:968] (0/2) Epoch 11, batch 43700, giga_loss[loss=0.3203, simple_loss=0.3808, pruned_loss=0.1299, over 28993.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3848, pruned_loss=0.1318, over 5658318.50 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3704, pruned_loss=0.1228, over 5682240.04 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3871, pruned_loss=0.1331, over 5666259.11 frames. ], batch size: 227, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:16:24,365 INFO [train.py:968] (0/2) Epoch 11, batch 43750, giga_loss[loss=0.3834, simple_loss=0.4062, pruned_loss=0.1803, over 23570.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3833, pruned_loss=0.1316, over 5653906.19 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3702, pruned_loss=0.1228, over 5675456.05 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3855, pruned_loss=0.1328, over 5665806.67 frames. ], batch size: 705, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:16:50,338 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-06 01:16:51,061 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-06 01:17:02,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.627e+03 2.006e+03 2.823e+03 1.504e+04, threshold=4.012e+03, percent-clipped=11.0
+2023-03-06 01:17:14,426 INFO [train.py:968] (0/2) Epoch 11, batch 43800, giga_loss[loss=0.301, simple_loss=0.3672, pruned_loss=0.1174, over 28681.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3827, pruned_loss=0.1319, over 5643092.85 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3706, pruned_loss=0.1232, over 5664373.73 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3841, pruned_loss=0.1326, over 5662354.75 frames. ], batch size: 262, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:17:42,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=499368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:18:03,149 INFO [train.py:968] (0/2) Epoch 11, batch 43850, giga_loss[loss=0.3088, simple_loss=0.3737, pruned_loss=0.1219, over 28721.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3814, pruned_loss=0.1319, over 5649283.36 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3705, pruned_loss=0.1231, over 5668346.14 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3828, pruned_loss=0.1327, over 5660722.71 frames. ], batch size: 242, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:18:37,984 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-06 01:18:41,143 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.616e+03 2.014e+03 2.830e+03 4.279e+03, threshold=4.028e+03, percent-clipped=1.0
+2023-03-06 01:18:46,214 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=499430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:18:54,428 INFO [train.py:968] (0/2) Epoch 11, batch 43900, giga_loss[loss=0.3575, simple_loss=0.4136, pruned_loss=0.1507, over 28221.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3815, pruned_loss=0.1325, over 5643627.65 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3708, pruned_loss=0.1231, over 5672990.54 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3826, pruned_loss=0.1333, over 5648178.50 frames. ], batch size: 368, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:19:41,886 INFO [train.py:968] (0/2) Epoch 11, batch 43950, giga_loss[loss=0.3024, simple_loss=0.3695, pruned_loss=0.1177, over 28929.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3812, pruned_loss=0.1329, over 5649227.88 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3714, pruned_loss=0.1238, over 5680707.17 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.382, pruned_loss=0.1333, over 5644736.19 frames. ], batch size: 145, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:20:05,665 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=499511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:20:07,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=499514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:20:15,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.114e+02 1.847e+03 2.572e+03 3.662e+03 2.185e+04, threshold=5.144e+03, percent-clipped=20.0
+2023-03-06 01:20:26,843 INFO [train.py:968] (0/2) Epoch 11, batch 44000, giga_loss[loss=0.2715, simple_loss=0.3432, pruned_loss=0.09986, over 28678.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3786, pruned_loss=0.1311, over 5667828.09 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.371, pruned_loss=0.1236, over 5686587.48 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3799, pruned_loss=0.1318, over 5658190.26 frames. ], batch size: 119, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:20:29,867 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=499543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:21:11,706 INFO [train.py:968] (0/2) Epoch 11, batch 44050, giga_loss[loss=0.301, simple_loss=0.359, pruned_loss=0.1215, over 28714.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3774, pruned_loss=0.13, over 5659839.00 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3713, pruned_loss=0.1237, over 5680382.34 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3782, pruned_loss=0.1306, over 5657901.40 frames. ], batch size: 92, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:21:46,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.681e+02 1.566e+03 2.018e+03 2.522e+03 5.119e+03, threshold=4.036e+03, percent-clipped=0.0
+2023-03-06 01:21:56,857 INFO [train.py:968] (0/2) Epoch 11, batch 44100, giga_loss[loss=0.3174, simple_loss=0.3902, pruned_loss=0.1223, over 28899.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3777, pruned_loss=0.1297, over 5661774.09 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3712, pruned_loss=0.1236, over 5686769.73 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3786, pruned_loss=0.1304, over 5653674.43 frames. ], batch size: 199, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:22:51,093 INFO [train.py:968] (0/2) Epoch 11, batch 44150, giga_loss[loss=0.3675, simple_loss=0.4291, pruned_loss=0.153, over 28678.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3804, pruned_loss=0.131, over 5649427.50 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3707, pruned_loss=0.1233, over 5681079.57 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3817, pruned_loss=0.132, over 5648258.04 frames. ], batch size: 262, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:23:15,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0929, 1.1255, 3.6974, 3.0318], device='cuda:0'), covar=tensor([0.1736, 0.2666, 0.0433, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0675, 0.0589, 0.0867, 0.0769], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 01:23:25,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.632e+03 2.143e+03 3.083e+03 8.211e+03, threshold=4.286e+03, percent-clipped=9.0
+2023-03-06 01:23:40,976 INFO [train.py:968] (0/2) Epoch 11, batch 44200, giga_loss[loss=0.3118, simple_loss=0.3712, pruned_loss=0.1262, over 28950.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3798, pruned_loss=0.1308, over 5659600.39 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3708, pruned_loss=0.1234, over 5684416.79 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3808, pruned_loss=0.1316, over 5655389.87 frames. ], batch size: 213, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:23:45,941 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=499745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:24:24,594 INFO [train.py:968] (0/2) Epoch 11, batch 44250, giga_loss[loss=0.2598, simple_loss=0.354, pruned_loss=0.08278, over 28407.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3811, pruned_loss=0.13, over 5667205.64 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.371, pruned_loss=0.1236, over 5687654.43 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.382, pruned_loss=0.1307, over 5660375.97 frames. ], batch size: 60, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:24:39,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=499805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:24:56,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.453e+02 1.530e+03 1.973e+03 2.946e+03 9.292e+03, threshold=3.946e+03, percent-clipped=7.0
+2023-03-06 01:25:08,160 INFO [train.py:968] (0/2) Epoch 11, batch 44300, giga_loss[loss=0.2864, simple_loss=0.3751, pruned_loss=0.09882, over 28931.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3812, pruned_loss=0.1276, over 5672354.60 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3706, pruned_loss=0.1233, over 5692164.71 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3825, pruned_loss=0.1285, over 5662768.15 frames. ], batch size: 136, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:25:53,292 INFO [train.py:968] (0/2) Epoch 11, batch 44350, giga_loss[loss=0.3179, simple_loss=0.3747, pruned_loss=0.1305, over 28681.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3833, pruned_loss=0.1281, over 5674262.78 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3707, pruned_loss=0.1235, over 5695777.55 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3847, pruned_loss=0.1289, over 5662638.85 frames. ], batch size: 92, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:26:09,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0208, 2.5159, 1.0502, 1.2915], device='cuda:0'), covar=tensor([0.1213, 0.0507, 0.1063, 0.1556], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0514, 0.0340, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 01:26:10,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5453, 1.7925, 1.8372, 1.3849], device='cuda:0'), covar=tensor([0.1582, 0.2053, 0.1271, 0.1500], device='cuda:0'), in_proj_covar=tensor([0.0832, 0.0701, 0.0870, 0.0773], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 01:26:11,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4850, 1.8219, 1.5812, 1.3307], device='cuda:0'), covar=tensor([0.1645, 0.1278, 0.0990, 0.1282], device='cuda:0'), in_proj_covar=tensor([0.1719, 0.1622, 0.1597, 0.1684], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 01:26:31,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.710e+02 1.348e+03 1.728e+03 2.479e+03 1.012e+04, threshold=3.455e+03, percent-clipped=13.0
+2023-03-06 01:26:43,195 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3420, 4.1602, 3.9625, 1.9073], device='cuda:0'), covar=tensor([0.0524, 0.0685, 0.0728, 0.2112], device='cuda:0'), in_proj_covar=tensor([0.1068, 0.1003, 0.0873, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 01:26:47,490 INFO [train.py:968] (0/2) Epoch 11, batch 44400, giga_loss[loss=0.3803, simple_loss=0.4195, pruned_loss=0.1706, over 29003.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3867, pruned_loss=0.1314, over 5667959.84 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3707, pruned_loss=0.1234, over 5699772.00 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.388, pruned_loss=0.1322, over 5654880.17 frames. ], batch size: 213, lr: 2.87e-03, grad_scale: 8.0
+2023-03-06 01:26:55,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=499948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:26:58,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=499951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:27:24,539 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=499980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:27:26,155 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=499981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:27:33,667 INFO [train.py:968] (0/2) Epoch 11, batch 44450, giga_loss[loss=0.3685, simple_loss=0.4239, pruned_loss=0.1566, over 28594.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.39, pruned_loss=0.1356, over 5655818.12 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.371, pruned_loss=0.1238, over 5696126.47 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3915, pruned_loss=0.1362, over 5646595.56 frames. ], batch size: 307, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:27:43,306 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-500000.pt
+2023-03-06 01:28:10,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.879e+02 1.768e+03 2.592e+03 3.949e+03 8.105e+03, threshold=5.185e+03, percent-clipped=30.0
+2023-03-06 01:28:19,528 INFO [train.py:968] (0/2) Epoch 11, batch 44500, giga_loss[loss=0.2787, simple_loss=0.3486, pruned_loss=0.1045, over 28861.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3892, pruned_loss=0.1354, over 5676194.61 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3711, pruned_loss=0.1239, over 5699838.70 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3907, pruned_loss=0.136, over 5665023.93 frames. ], batch size: 145, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:28:37,091 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0583, 1.3592, 1.2442, 1.1588], device='cuda:0'), covar=tensor([0.1448, 0.1402, 0.1804, 0.1479], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0731, 0.0671, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 01:29:07,205 INFO [train.py:968] (0/2) Epoch 11, batch 44550, giga_loss[loss=0.339, simple_loss=0.3927, pruned_loss=0.1426, over 27521.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3872, pruned_loss=0.1339, over 5673808.98 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3712, pruned_loss=0.124, over 5701482.21 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3885, pruned_loss=0.1344, over 5663146.29 frames. ], batch size: 472, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:29:36,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=500120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:29:42,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.944e+02 1.562e+03 2.049e+03 2.653e+03 5.067e+03, threshold=4.098e+03, percent-clipped=0.0
+2023-03-06 01:29:51,006 INFO [train.py:968] (0/2) Epoch 11, batch 44600, giga_loss[loss=0.2594, simple_loss=0.3498, pruned_loss=0.0845, over 28486.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3845, pruned_loss=0.1299, over 5680832.69 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3707, pruned_loss=0.1236, over 5703350.66 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3861, pruned_loss=0.1308, over 5670288.77 frames. ], batch size: 71, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:30:23,471 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-06 01:30:25,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:30:42,038 INFO [train.py:968] (0/2) Epoch 11, batch 44650, giga_loss[loss=0.3091, simple_loss=0.3841, pruned_loss=0.117, over 28828.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3856, pruned_loss=0.1293, over 5681700.78 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.371, pruned_loss=0.1238, over 5704823.26 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3868, pruned_loss=0.1299, over 5671736.37 frames. ], batch size: 119, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:31:19,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.199e+02 1.484e+03 1.880e+03 2.487e+03 1.891e+04, threshold=3.760e+03, percent-clipped=8.0
+2023-03-06 01:31:29,980 INFO [train.py:968] (0/2) Epoch 11, batch 44700, libri_loss[loss=0.2705, simple_loss=0.3312, pruned_loss=0.1049, over 28075.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3861, pruned_loss=0.1303, over 5670104.39 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3707, pruned_loss=0.1236, over 5706227.44 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3878, pruned_loss=0.1311, over 5660136.91 frames. ], batch size: 62, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:31:46,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4447, 3.7106, 1.5908, 1.6396], device='cuda:0'), covar=tensor([0.0962, 0.0340, 0.0882, 0.1264], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0519, 0.0342, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 01:31:52,266 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:31:52,319 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=500263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:31:55,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=500266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:32:17,663 INFO [train.py:968] (0/2) Epoch 11, batch 44750, giga_loss[loss=0.3574, simple_loss=0.412, pruned_loss=0.1514, over 28588.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.386, pruned_loss=0.1308, over 5656356.65 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3708, pruned_loss=0.1238, over 5701695.88 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3877, pruned_loss=0.1316, over 5651934.17 frames. ], batch size: 336, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:32:22,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=500295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:32:50,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.454e+03 1.828e+03 2.544e+03 1.241e+04, threshold=3.656e+03, percent-clipped=8.0
+2023-03-06 01:33:01,498 INFO [train.py:968] (0/2) Epoch 11, batch 44800, giga_loss[loss=0.2845, simple_loss=0.3587, pruned_loss=0.1052, over 28205.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3843, pruned_loss=0.1298, over 5669776.03 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3713, pruned_loss=0.124, over 5707975.70 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3858, pruned_loss=0.1304, over 5658550.31 frames. ], batch size: 77, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:33:05,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0161, 1.4050, 1.1346, 0.1782], device='cuda:0'), covar=tensor([0.2730, 0.2274, 0.3348, 0.4175], device='cuda:0'), in_proj_covar=tensor([0.1583, 0.1510, 0.1507, 0.1294], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 01:33:15,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=500356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:33:40,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2955, 1.4388, 1.5605, 1.3787], device='cuda:0'), covar=tensor([0.1015, 0.0896, 0.1253, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0738, 0.0678, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 01:33:47,370 INFO [train.py:968] (0/2) Epoch 11, batch 44850, giga_loss[loss=0.2984, simple_loss=0.3629, pruned_loss=0.117, over 29048.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3826, pruned_loss=0.1298, over 5669896.26 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3715, pruned_loss=0.1241, over 5710968.91 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.384, pruned_loss=0.1304, over 5657059.74 frames. ], batch size: 136, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:34:00,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-06 01:34:23,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.261e+02 1.613e+03 2.587e+03 3.897e+03 2.898e+04, threshold=5.173e+03, percent-clipped=26.0
+2023-03-06 01:34:32,429 INFO [train.py:968] (0/2) Epoch 11, batch 44900, giga_loss[loss=0.2849, simple_loss=0.3548, pruned_loss=0.1075, over 29115.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3812, pruned_loss=0.1297, over 5658221.57 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3713, pruned_loss=0.124, over 5697889.25 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3828, pruned_loss=0.1305, over 5657043.38 frames. ], batch size: 155, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:35:17,383 INFO [train.py:968] (0/2) Epoch 11, batch 44950, giga_loss[loss=0.2949, simple_loss=0.3551, pruned_loss=0.1174, over 29006.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3805, pruned_loss=0.1297, over 5646756.57 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.372, pruned_loss=0.1244, over 5686127.53 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3816, pruned_loss=0.1301, over 5654892.56 frames. ], batch size: 106, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:35:25,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=500499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:35:28,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=500502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:35:50,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.185e+03 1.542e+03 1.945e+03 3.002e+03 6.478e+03, threshold=3.889e+03, percent-clipped=3.0
+2023-03-06 01:35:52,560 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=500531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:36:02,773 INFO [train.py:968] (0/2) Epoch 11, batch 45000, giga_loss[loss=0.3299, simple_loss=0.3833, pruned_loss=0.1382, over 28700.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3812, pruned_loss=0.1316, over 5650670.61 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3721, pruned_loss=0.1244, over 5691435.86 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3822, pruned_loss=0.1322, over 5651725.47 frames. ], batch size: 99, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:36:02,778 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 01:36:11,324 INFO [train.py:1012] (0/2) Epoch 11, validation: loss=0.2157, simple_loss=0.3231, pruned_loss=0.0542, over 944034.00 frames. 
+2023-03-06 01:36:11,325 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 01:36:18,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=500547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:36:53,255 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:36:54,300 INFO [train.py:968] (0/2) Epoch 11, batch 45050, giga_loss[loss=0.2691, simple_loss=0.34, pruned_loss=0.0991, over 28868.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3797, pruned_loss=0.1306, over 5643002.06 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.372, pruned_loss=0.1243, over 5691157.89 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3809, pruned_loss=0.1314, over 5642167.54 frames. ], batch size: 112, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:37:32,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.191e+02 1.494e+03 1.758e+03 2.550e+03 5.480e+03, threshold=3.517e+03, percent-clipped=5.0
+2023-03-06 01:37:40,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=500638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:37:41,203 INFO [train.py:968] (0/2) Epoch 11, batch 45100, giga_loss[loss=0.2819, simple_loss=0.3564, pruned_loss=0.1037, over 28243.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3759, pruned_loss=0.1258, over 5656181.92 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3721, pruned_loss=0.1243, over 5695167.92 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3769, pruned_loss=0.1264, over 5651178.50 frames. ], batch size: 368, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:38:13,454 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500674.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 01:38:26,651 INFO [train.py:968] (0/2) Epoch 11, batch 45150, giga_loss[loss=0.2916, simple_loss=0.3673, pruned_loss=0.1079, over 28816.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3718, pruned_loss=0.1222, over 5649413.33 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3722, pruned_loss=0.1244, over 5693909.79 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1226, over 5646138.41 frames. ], batch size: 119, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:38:28,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=500690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:38:30,688 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=500693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:39:04,552 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=500722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:39:11,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.696e+02 1.324e+03 1.670e+03 2.197e+03 5.271e+03, threshold=3.340e+03, percent-clipped=8.0
+2023-03-06 01:39:19,296 INFO [train.py:968] (0/2) Epoch 11, batch 45200, libri_loss[loss=0.2519, simple_loss=0.319, pruned_loss=0.09242, over 29646.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3718, pruned_loss=0.1225, over 5655489.05 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3719, pruned_loss=0.1242, over 5693357.06 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3727, pruned_loss=0.123, over 5652354.39 frames. ], batch size: 69, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:39:52,113 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:40:01,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=500781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:40:04,852 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=500784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:40:11,214 INFO [train.py:968] (0/2) Epoch 11, batch 45250, giga_loss[loss=0.2609, simple_loss=0.3335, pruned_loss=0.0942, over 28805.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3702, pruned_loss=0.1225, over 5663968.06 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3721, pruned_loss=0.1242, over 5688473.28 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3708, pruned_loss=0.1228, over 5664799.08 frames. ], batch size: 284, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:40:14,450 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-06 01:40:20,092 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-06 01:40:34,103 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=500813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:40:46,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.713e+03 2.130e+03 3.352e+03 1.143e+04, threshold=4.260e+03, percent-clipped=25.0
+2023-03-06 01:40:55,494 INFO [train.py:968] (0/2) Epoch 11, batch 45300, giga_loss[loss=0.265, simple_loss=0.3395, pruned_loss=0.0953, over 28595.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3697, pruned_loss=0.1223, over 5671961.26 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.372, pruned_loss=0.1243, over 5690489.57 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3702, pruned_loss=0.1226, over 5670219.73 frames. ], batch size: 78, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:41:38,013 INFO [train.py:968] (0/2) Epoch 11, batch 45350, giga_loss[loss=0.3103, simple_loss=0.3806, pruned_loss=0.12, over 29065.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3719, pruned_loss=0.1231, over 5679740.68 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3721, pruned_loss=0.1246, over 5685813.77 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3721, pruned_loss=0.123, over 5682166.56 frames. ], batch size: 155, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:41:53,474 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5882, 5.4015, 5.1350, 2.3643], device='cuda:0'), covar=tensor([0.0344, 0.0480, 0.0530, 0.1771], device='cuda:0'), in_proj_covar=tensor([0.1074, 0.1008, 0.0879, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 01:42:16,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.516e+03 2.145e+03 2.961e+03 1.085e+04, threshold=4.291e+03, percent-clipped=9.0
+2023-03-06 01:42:25,419 INFO [train.py:968] (0/2) Epoch 11, batch 45400, giga_loss[loss=0.2874, simple_loss=0.3551, pruned_loss=0.1098, over 28687.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3739, pruned_loss=0.1244, over 5662687.20 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3727, pruned_loss=0.1248, over 5683715.17 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3736, pruned_loss=0.124, over 5665810.80 frames. ], batch size: 99, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:42:41,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3260, 1.4678, 1.4812, 1.3251], device='cuda:0'), covar=tensor([0.1549, 0.1753, 0.2173, 0.1775], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0737, 0.0672, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 01:42:42,036 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-06 01:42:46,399 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=500962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:42:54,859 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:43:01,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4989, 1.7705, 1.4961, 1.5212], device='cuda:0'), covar=tensor([0.0762, 0.0294, 0.0300, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 01:43:04,701 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:43:10,074 INFO [train.py:968] (0/2) Epoch 11, batch 45450, giga_loss[loss=0.2934, simple_loss=0.3579, pruned_loss=0.1144, over 28708.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3739, pruned_loss=0.1245, over 5672033.69 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3727, pruned_loss=0.125, over 5691277.58 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3736, pruned_loss=0.124, over 5666838.02 frames. ], batch size: 242, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:43:47,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.694e+02 1.258e+03 1.776e+03 2.448e+03 6.853e+03, threshold=3.551e+03, percent-clipped=8.0
+2023-03-06 01:43:56,858 INFO [train.py:968] (0/2) Epoch 11, batch 45500, giga_loss[loss=0.2988, simple_loss=0.3622, pruned_loss=0.1177, over 28462.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3756, pruned_loss=0.1263, over 5649257.36 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3733, pruned_loss=0.1254, over 5674306.30 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3748, pruned_loss=0.1255, over 5660741.36 frames. ], batch size: 71, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:44:08,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=501049.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 01:44:42,513 INFO [train.py:968] (0/2) Epoch 11, batch 45550, giga_loss[loss=0.3316, simple_loss=0.3875, pruned_loss=0.1379, over 27824.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3779, pruned_loss=0.1284, over 5604649.38 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3742, pruned_loss=0.1261, over 5638333.00 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3766, pruned_loss=0.1272, over 5646943.85 frames. ], batch size: 412, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:44:59,935 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=501105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:45:01,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=501108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:45:18,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.103e+02 1.703e+03 2.070e+03 2.940e+03 9.829e+03, threshold=4.140e+03, percent-clipped=20.0
+2023-03-06 01:45:21,776 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-06 01:45:25,459 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=501137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:45:26,624 INFO [train.py:968] (0/2) Epoch 11, batch 45600, giga_loss[loss=0.3196, simple_loss=0.3827, pruned_loss=0.1282, over 28416.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3812, pruned_loss=0.1308, over 5539215.74 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3757, pruned_loss=0.1274, over 5560930.66 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3789, pruned_loss=0.1287, over 5641793.91 frames. ], batch size: 71, lr: 2.87e-03, grad_scale: 8.0
+2023-03-06 01:45:38,489 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=501150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:45:38,875 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-06 01:46:12,828 INFO [train.py:968] (0/2) Epoch 11, batch 45650, giga_loss[loss=0.3443, simple_loss=0.398, pruned_loss=0.1453, over 28352.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3814, pruned_loss=0.1306, over 5547726.13 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.376, pruned_loss=0.1276, over 5536734.84 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3793, pruned_loss=0.1288, over 5651246.62 frames. ], batch size: 369, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:46:18,013 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-06 01:46:20,670 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-11.pt
+2023-03-06 01:46:49,856 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=501192.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 01:46:55,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=501195.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 01:47:22,661 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=501224.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 01:47:27,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.536e+02 1.318e+03 1.738e+03 2.239e+03 5.406e+03, threshold=3.477e+03, percent-clipped=5.0
+2023-03-06 01:47:40,803 INFO [train.py:968] (0/2) Epoch 12, batch 50, giga_loss[loss=0.3271, simple_loss=0.3852, pruned_loss=0.1345, over 26795.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3798, pruned_loss=0.1145, over 1263211.06 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3476, pruned_loss=0.09413, over 145124.41 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3836, pruned_loss=0.1169, over 1147001.26 frames. ], batch size: 555, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:48:29,098 INFO [train.py:968] (0/2) Epoch 12, batch 100, libri_loss[loss=0.2957, simple_loss=0.3807, pruned_loss=0.1053, over 27863.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3712, pruned_loss=0.1098, over 2248493.76 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3537, pruned_loss=0.0988, over 368848.74 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3739, pruned_loss=0.1115, over 2008358.21 frames. ], batch size: 116, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:48:30,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=501293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:48:32,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=501296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:48:56,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=501325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:49:00,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.638e+02 1.100e+03 1.358e+03 1.818e+03 3.678e+03, threshold=2.716e+03, percent-clipped=1.0
+2023-03-06 01:49:10,939 INFO [train.py:968] (0/2) Epoch 12, batch 150, giga_loss[loss=0.2523, simple_loss=0.326, pruned_loss=0.08935, over 28839.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3583, pruned_loss=0.104, over 3005017.18 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3489, pruned_loss=0.09556, over 599506.66 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3606, pruned_loss=0.1058, over 2690038.06 frames. ], batch size: 186, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:49:14,248 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=501345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:49:25,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=501358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:49:45,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3399, 1.4813, 1.1969, 1.5289], device='cuda:0'), covar=tensor([0.0755, 0.0361, 0.0347, 0.0819], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 01:49:53,382 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-06 01:49:55,400 INFO [train.py:968] (0/2) Epoch 12, batch 200, giga_loss[loss=0.2518, simple_loss=0.3137, pruned_loss=0.09498, over 28666.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3451, pruned_loss=0.09844, over 3604771.48 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3519, pruned_loss=0.09785, over 648500.62 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3451, pruned_loss=0.09885, over 3337467.72 frames. ], batch size: 85, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:50:26,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.879e+02 9.737e+02 1.311e+03 1.815e+03 4.619e+03, threshold=2.622e+03, percent-clipped=5.0
+2023-03-06 01:50:36,758 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 01:50:36,911 INFO [train.py:968] (0/2) Epoch 12, batch 250, giga_loss[loss=0.2176, simple_loss=0.2933, pruned_loss=0.07095, over 28923.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3328, pruned_loss=0.09162, over 4066743.74 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3496, pruned_loss=0.09575, over 794498.58 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3319, pruned_loss=0.09177, over 3805352.48 frames. ], batch size: 213, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:51:14,998 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=501488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:51:16,490 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-06 01:51:17,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=501491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:51:17,738 INFO [train.py:968] (0/2) Epoch 12, batch 300, giga_loss[loss=0.2382, simple_loss=0.3013, pruned_loss=0.08757, over 28591.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3249, pruned_loss=0.08799, over 4433646.20 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3494, pruned_loss=0.09502, over 969141.53 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3229, pruned_loss=0.08778, over 4172874.00 frames. ], batch size: 336, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:51:27,109 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=501501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:51:29,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=501504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:51:35,168 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4257, 1.6487, 1.6130, 1.5264], device='cuda:0'), covar=tensor([0.1449, 0.1673, 0.1782, 0.1639], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0734, 0.0668, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 01:51:43,549 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=501520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:51:53,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.945e+02 9.198e+02 1.214e+03 1.585e+03 4.091e+03, threshold=2.427e+03, percent-clipped=3.0
+2023-03-06 01:51:56,168 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=501533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:52:05,418 INFO [train.py:968] (0/2) Epoch 12, batch 350, giga_loss[loss=0.2811, simple_loss=0.332, pruned_loss=0.1151, over 26622.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3182, pruned_loss=0.08538, over 4707242.43 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.351, pruned_loss=0.0964, over 1057560.18 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3155, pruned_loss=0.08461, over 4485483.08 frames. ], batch size: 555, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:52:06,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2372, 1.3806, 1.4316, 1.0509], device='cuda:0'), covar=tensor([0.1882, 0.3229, 0.1504, 0.1562], device='cuda:0'), in_proj_covar=tensor([0.0845, 0.0707, 0.0881, 0.0782], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 01:52:23,928 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=501566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:52:37,557 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-06 01:52:43,996 INFO [train.py:968] (0/2) Epoch 12, batch 400, giga_loss[loss=0.209, simple_loss=0.2816, pruned_loss=0.06821, over 28615.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3148, pruned_loss=0.08415, over 4924637.38 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3525, pruned_loss=0.09763, over 1212014.44 frames. ], giga_tot_loss[loss=0.2383, simple_loss=0.3109, pruned_loss=0.08278, over 4724199.94 frames. ], batch size: 78, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 01:53:15,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.527e+02 1.042e+03 1.270e+03 1.742e+03 3.660e+03, threshold=2.541e+03, percent-clipped=8.0
+2023-03-06 01:53:24,742 INFO [train.py:968] (0/2) Epoch 12, batch 450, libri_loss[loss=0.2558, simple_loss=0.3336, pruned_loss=0.08895, over 29586.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.311, pruned_loss=0.08212, over 5102720.53 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3522, pruned_loss=0.09812, over 1304014.20 frames. ], giga_tot_loss[loss=0.2342, simple_loss=0.3072, pruned_loss=0.0806, over 4928735.99 frames. ], batch size: 75, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 01:53:56,138 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8076, 1.8134, 1.3186, 1.4708], device='cuda:0'), covar=tensor([0.0750, 0.0615, 0.1064, 0.1073], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0441, 0.0502, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 01:54:06,795 INFO [train.py:968] (0/2) Epoch 12, batch 500, giga_loss[loss=0.2273, simple_loss=0.3009, pruned_loss=0.07689, over 29003.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3088, pruned_loss=0.08082, over 5246021.02 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3532, pruned_loss=0.09849, over 1440398.99 frames. ], giga_tot_loss[loss=0.231, simple_loss=0.3041, pruned_loss=0.07895, over 5086672.31 frames. ], batch size: 213, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 01:54:31,956 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=501720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:54:38,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5820, 1.8273, 1.5107, 1.7106], device='cuda:0'), covar=tensor([0.2311, 0.2238, 0.2365, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.1310, 0.0969, 0.1158, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 01:54:41,801 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.285e+02 9.903e+02 1.226e+03 1.549e+03 4.823e+03, threshold=2.453e+03, percent-clipped=4.0
+2023-03-06 01:54:50,688 INFO [train.py:968] (0/2) Epoch 12, batch 550, giga_loss[loss=0.2266, simple_loss=0.3042, pruned_loss=0.07446, over 28267.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3064, pruned_loss=0.07961, over 5329202.01 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3524, pruned_loss=0.09805, over 1530562.43 frames. ], giga_tot_loss[loss=0.2287, simple_loss=0.3018, pruned_loss=0.07782, over 5207279.19 frames. ], batch size: 368, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:55:37,150 INFO [train.py:968] (0/2) Epoch 12, batch 600, giga_loss[loss=0.2171, simple_loss=0.2737, pruned_loss=0.08025, over 23826.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3036, pruned_loss=0.07817, over 5410836.60 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3521, pruned_loss=0.09773, over 1595909.46 frames. ], giga_tot_loss[loss=0.2263, simple_loss=0.2994, pruned_loss=0.07657, over 5307320.94 frames. ], batch size: 705, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:56:02,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2839, 1.7938, 1.3208, 0.5251], device='cuda:0'), covar=tensor([0.3919, 0.2267, 0.3185, 0.4827], device='cuda:0'), in_proj_covar=tensor([0.1577, 0.1501, 0.1500, 0.1287], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 01:56:10,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.097e+02 1.031e+03 1.324e+03 2.071e+03 5.206e+03, threshold=2.649e+03, percent-clipped=15.0
+2023-03-06 01:56:20,482 INFO [train.py:968] (0/2) Epoch 12, batch 650, giga_loss[loss=0.2043, simple_loss=0.2777, pruned_loss=0.06539, over 28561.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3023, pruned_loss=0.07729, over 5464143.07 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.352, pruned_loss=0.09756, over 1775220.60 frames. ], giga_tot_loss[loss=0.2237, simple_loss=0.2969, pruned_loss=0.07522, over 5371253.03 frames. ], batch size: 78, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:56:46,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=501867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:56:52,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4380, 1.6286, 1.4489, 1.2668], device='cuda:0'), covar=tensor([0.2289, 0.1794, 0.1271, 0.1739], device='cuda:0'), in_proj_covar=tensor([0.1726, 0.1618, 0.1593, 0.1681], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 01:57:09,644 INFO [train.py:968] (0/2) Epoch 12, batch 700, giga_loss[loss=0.2357, simple_loss=0.2878, pruned_loss=0.09177, over 24034.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.2989, pruned_loss=0.076, over 5511714.52 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3516, pruned_loss=0.09719, over 1795901.50 frames. ], giga_tot_loss[loss=0.2217, simple_loss=0.2946, pruned_loss=0.07439, over 5437172.17 frames. ], batch size: 705, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:57:43,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.267e+02 1.050e+03 1.456e+03 1.855e+03 5.572e+03, threshold=2.912e+03, percent-clipped=9.0
+2023-03-06 01:57:52,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=501941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:57:52,944 INFO [train.py:968] (0/2) Epoch 12, batch 750, giga_loss[loss=0.1946, simple_loss=0.2725, pruned_loss=0.05835, over 28904.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.2965, pruned_loss=0.07486, over 5563098.56 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3511, pruned_loss=0.09663, over 1855659.00 frames. ], giga_tot_loss[loss=0.2197, simple_loss=0.2924, pruned_loss=0.07344, over 5500644.07 frames. ], batch size: 145, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:57:58,363 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8191, 1.8127, 1.5717, 2.1359], device='cuda:0'), covar=tensor([0.2331, 0.2573, 0.2675, 0.2329], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.0971, 0.1161, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 01:58:19,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9410, 1.0903, 3.3326, 2.8470], device='cuda:0'), covar=tensor([0.1701, 0.2670, 0.0491, 0.0975], device='cuda:0'), in_proj_covar=tensor([0.0671, 0.0590, 0.0869, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 01:58:37,791 INFO [train.py:968] (0/2) Epoch 12, batch 800, giga_loss[loss=0.2019, simple_loss=0.2696, pruned_loss=0.06707, over 28612.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2937, pruned_loss=0.07346, over 5592501.77 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3511, pruned_loss=0.09642, over 1952023.25 frames. ], giga_tot_loss[loss=0.2164, simple_loss=0.2891, pruned_loss=0.07183, over 5538902.59 frames. ], batch size: 85, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 01:58:42,949 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-502000.pt
+2023-03-06 01:58:43,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6010, 5.3779, 5.1190, 2.6842], device='cuda:0'), covar=tensor([0.0466, 0.0698, 0.0762, 0.1694], device='cuda:0'), in_proj_covar=tensor([0.1056, 0.0985, 0.0860, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 01:58:58,963 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7787, 1.7281, 1.3712, 1.4667], device='cuda:0'), covar=tensor([0.0804, 0.0679, 0.0995, 0.1159], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0438, 0.0499, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 01:59:09,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.498e+02 1.054e+03 1.315e+03 1.774e+03 4.984e+03, threshold=2.630e+03, percent-clipped=10.0
+2023-03-06 01:59:24,336 INFO [train.py:968] (0/2) Epoch 12, batch 850, giga_loss[loss=0.3572, simple_loss=0.3994, pruned_loss=0.1575, over 26673.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3021, pruned_loss=0.07835, over 5606818.54 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3513, pruned_loss=0.09654, over 2105116.07 frames. ], giga_tot_loss[loss=0.2244, simple_loss=0.2963, pruned_loss=0.07626, over 5555000.46 frames. ], batch size: 555, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 01:59:30,869 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=502051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:00:04,296 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=502084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:00:06,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=502087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:00:12,223 INFO [train.py:968] (0/2) Epoch 12, batch 900, giga_loss[loss=0.2992, simple_loss=0.3704, pruned_loss=0.114, over 28692.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3158, pruned_loss=0.08572, over 5626177.56 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.35, pruned_loss=0.09586, over 2179555.69 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3111, pruned_loss=0.08408, over 5580970.25 frames. ], batch size: 307, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 02:00:14,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=502095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:00:32,993 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=502116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:00:46,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.457e+02 1.259e+03 1.631e+03 2.293e+03 3.913e+03, threshold=3.263e+03, percent-clipped=17.0
+2023-03-06 02:00:54,234 INFO [train.py:968] (0/2) Epoch 12, batch 950, giga_loss[loss=0.3081, simple_loss=0.3925, pruned_loss=0.1119, over 29019.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3284, pruned_loss=0.09196, over 5652142.29 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3502, pruned_loss=0.09583, over 2289565.01 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3238, pruned_loss=0.0905, over 5609066.00 frames. ], batch size: 164, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:01:24,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=502176.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 02:01:36,050 INFO [train.py:968] (0/2) Epoch 12, batch 1000, giga_loss[loss=0.293, simple_loss=0.3691, pruned_loss=0.1084, over 28913.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3373, pruned_loss=0.09634, over 5666205.57 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3495, pruned_loss=0.09537, over 2379008.00 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3335, pruned_loss=0.09534, over 5626414.46 frames. ], batch size: 199, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:02:04,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3670, 1.8626, 1.5117, 0.5429], device='cuda:0'), covar=tensor([0.2745, 0.2126, 0.3043, 0.3952], device='cuda:0'), in_proj_covar=tensor([0.1574, 0.1497, 0.1501, 0.1287], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 02:02:08,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.354e+02 1.236e+03 1.590e+03 2.249e+03 6.064e+03, threshold=3.180e+03, percent-clipped=5.0
+2023-03-06 02:02:13,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=502238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:02:15,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=502241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:02:16,228 INFO [train.py:968] (0/2) Epoch 12, batch 1050, giga_loss[loss=0.3101, simple_loss=0.3835, pruned_loss=0.1184, over 28855.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3413, pruned_loss=0.09687, over 5676408.17 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3498, pruned_loss=0.09518, over 2477916.38 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3379, pruned_loss=0.0962, over 5642863.67 frames. ], batch size: 199, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:02:16,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=502242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:02:41,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=502270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:02:56,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0778, 1.4964, 1.3946, 1.0243], device='cuda:0'), covar=tensor([0.1606, 0.2259, 0.1343, 0.1583], device='cuda:0'), in_proj_covar=tensor([0.0843, 0.0700, 0.0880, 0.0781], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 02:03:01,322 INFO [train.py:968] (0/2) Epoch 12, batch 1100, giga_loss[loss=0.2965, simple_loss=0.3647, pruned_loss=0.1141, over 28685.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3432, pruned_loss=0.09663, over 5673544.39 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3496, pruned_loss=0.09499, over 2577528.51 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3404, pruned_loss=0.09621, over 5642259.54 frames. ], batch size: 242, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:03:34,497 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.065e+02 1.039e+03 1.287e+03 1.841e+03 3.540e+03, threshold=2.574e+03, percent-clipped=3.0
+2023-03-06 02:03:44,725 INFO [train.py:968] (0/2) Epoch 12, batch 1150, giga_loss[loss=0.2506, simple_loss=0.3288, pruned_loss=0.08615, over 28912.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3445, pruned_loss=0.09725, over 5688762.92 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3494, pruned_loss=0.09489, over 2610897.35 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3424, pruned_loss=0.09698, over 5662032.07 frames. ], batch size: 112, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:04:23,072 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=502385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:04:24,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=502388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:04:27,079 INFO [train.py:968] (0/2) Epoch 12, batch 1200, giga_loss[loss=0.2917, simple_loss=0.3607, pruned_loss=0.1114, over 28878.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3483, pruned_loss=0.1001, over 5684379.43 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3513, pruned_loss=0.09596, over 2722897.58 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3456, pruned_loss=0.09952, over 5657302.60 frames. ], batch size: 145, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:04:49,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=502417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:04:59,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=502426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:05:02,072 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-06 02:05:03,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.666e+02 1.131e+03 1.399e+03 2.078e+03 9.467e+03, threshold=2.798e+03, percent-clipped=13.0
+2023-03-06 02:05:10,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6425, 1.6331, 1.2135, 1.2409], device='cuda:0'), covar=tensor([0.0806, 0.0622, 0.1004, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0438, 0.0502, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 02:05:11,952 INFO [train.py:968] (0/2) Epoch 12, batch 1250, giga_loss[loss=0.338, simple_loss=0.3815, pruned_loss=0.1472, over 23287.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3517, pruned_loss=0.1026, over 5679385.46 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3516, pruned_loss=0.09604, over 2782539.41 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3494, pruned_loss=0.1022, over 5656873.52 frames. ], batch size: 705, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:05:57,802 INFO [train.py:968] (0/2) Epoch 12, batch 1300, giga_loss[loss=0.2799, simple_loss=0.3596, pruned_loss=0.1, over 28752.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3549, pruned_loss=0.1041, over 5682907.10 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3515, pruned_loss=0.09587, over 2813644.04 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3532, pruned_loss=0.104, over 5663189.95 frames. ], batch size: 262, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:06:30,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.145e+02 1.118e+03 1.335e+03 1.686e+03 2.878e+03, threshold=2.670e+03, percent-clipped=3.0
+2023-03-06 02:06:35,905 INFO [train.py:968] (0/2) Epoch 12, batch 1350, libri_loss[loss=0.3542, simple_loss=0.4041, pruned_loss=0.1521, over 29528.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3565, pruned_loss=0.1039, over 5693326.37 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3516, pruned_loss=0.09638, over 2896151.52 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3551, pruned_loss=0.1037, over 5680432.80 frames. ], batch size: 84, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:06:43,344 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=502551.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 02:07:00,584 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=502569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:07:03,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=502572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:07:21,827 INFO [train.py:968] (0/2) Epoch 12, batch 1400, giga_loss[loss=0.273, simple_loss=0.354, pruned_loss=0.09606, over 28745.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3579, pruned_loss=0.1042, over 5689695.43 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.352, pruned_loss=0.09666, over 2911406.77 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3567, pruned_loss=0.104, over 5678610.08 frames. ], batch size: 78, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:07:27,676 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=502601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:07:32,047 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-06 02:07:53,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.505e+02 1.169e+03 1.388e+03 1.894e+03 3.810e+03, threshold=2.777e+03, percent-clipped=4.0
+2023-03-06 02:07:59,960 INFO [train.py:968] (0/2) Epoch 12, batch 1450, libri_loss[loss=0.3273, simple_loss=0.4, pruned_loss=0.1273, over 26192.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3583, pruned_loss=0.1037, over 5686613.91 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3526, pruned_loss=0.0971, over 3030296.03 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3572, pruned_loss=0.1036, over 5680735.66 frames. ], batch size: 136, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:08:29,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4210, 1.6370, 1.3196, 1.6136], device='cuda:0'), covar=tensor([0.0717, 0.0302, 0.0309, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 02:08:40,476 INFO [train.py:968] (0/2) Epoch 12, batch 1500, libri_loss[loss=0.2647, simple_loss=0.3541, pruned_loss=0.08764, over 28616.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3571, pruned_loss=0.1016, over 5696710.16 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3527, pruned_loss=0.09705, over 3086112.73 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3563, pruned_loss=0.1016, over 5689312.73 frames. ], batch size: 106, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:08:42,882 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=502694.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 02:08:44,891 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=502697.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 02:09:08,066 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=502726.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 02:09:12,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.890e+02 1.055e+03 1.420e+03 1.977e+03 6.048e+03, threshold=2.840e+03, percent-clipped=9.0
+2023-03-06 02:09:20,106 INFO [train.py:968] (0/2) Epoch 12, batch 1550, giga_loss[loss=0.25, simple_loss=0.3354, pruned_loss=0.08233, over 28758.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3549, pruned_loss=0.09963, over 5682954.29 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3533, pruned_loss=0.09772, over 3162436.78 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3541, pruned_loss=0.09938, over 5690454.02 frames. ], batch size: 242, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:09:27,472 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=502751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:10:04,246 INFO [train.py:968] (0/2) Epoch 12, batch 1600, giga_loss[loss=0.277, simple_loss=0.3484, pruned_loss=0.1028, over 28796.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3553, pruned_loss=0.1007, over 5686330.40 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3537, pruned_loss=0.09795, over 3205088.76 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3544, pruned_loss=0.1005, over 5698589.90 frames. ], batch size: 112, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 02:10:07,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2618, 2.6636, 1.9910, 1.6707], device='cuda:0'), covar=tensor([0.1875, 0.1391, 0.1708, 0.1899], device='cuda:0'), in_proj_covar=tensor([0.1712, 0.1608, 0.1593, 0.1680], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 02:10:39,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.985e+02 1.225e+03 1.470e+03 1.912e+03 8.980e+03, threshold=2.940e+03, percent-clipped=7.0
+2023-03-06 02:10:42,068 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=502837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:10:47,586 INFO [train.py:968] (0/2) Epoch 12, batch 1650, giga_loss[loss=0.309, simple_loss=0.3746, pruned_loss=0.1217, over 28594.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3585, pruned_loss=0.1058, over 5698699.53 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3532, pruned_loss=0.09769, over 3284737.65 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3582, pruned_loss=0.1058, over 5703181.67 frames. ], batch size: 60, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 02:11:31,042 INFO [train.py:968] (0/2) Epoch 12, batch 1700, giga_loss[loss=0.2919, simple_loss=0.3593, pruned_loss=0.1123, over 28767.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3594, pruned_loss=0.1085, over 5678756.58 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3527, pruned_loss=0.09758, over 3321550.11 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3595, pruned_loss=0.1087, over 5692843.73 frames. ], batch size: 92, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 02:12:05,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.875e+02 1.360e+03 1.665e+03 2.200e+03 9.527e+03, threshold=3.329e+03, percent-clipped=11.0
+2023-03-06 02:12:11,864 INFO [train.py:968] (0/2) Epoch 12, batch 1750, giga_loss[loss=0.2902, simple_loss=0.3623, pruned_loss=0.1091, over 29013.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3577, pruned_loss=0.1084, over 5683501.11 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.352, pruned_loss=0.09711, over 3397822.41 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3584, pruned_loss=0.1091, over 5689108.69 frames. ], batch size: 164, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:12:58,004 INFO [train.py:968] (0/2) Epoch 12, batch 1800, giga_loss[loss=0.3, simple_loss=0.3627, pruned_loss=0.1186, over 28610.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3559, pruned_loss=0.1075, over 5697909.22 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3521, pruned_loss=0.09714, over 3410484.30 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3563, pruned_loss=0.1082, over 5701180.65 frames. ], batch size: 78, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:13:34,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.293e+02 1.178e+03 1.465e+03 1.947e+03 4.085e+03, threshold=2.930e+03, percent-clipped=5.0
+2023-03-06 02:13:40,771 INFO [train.py:968] (0/2) Epoch 12, batch 1850, giga_loss[loss=0.299, simple_loss=0.3758, pruned_loss=0.1111, over 28330.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3555, pruned_loss=0.1067, over 5703231.43 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.352, pruned_loss=0.09697, over 3459799.90 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.356, pruned_loss=0.1075, over 5702451.43 frames. ], batch size: 368, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:14:25,271 INFO [train.py:968] (0/2) Epoch 12, batch 1900, giga_loss[loss=0.249, simple_loss=0.3304, pruned_loss=0.08383, over 28948.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3538, pruned_loss=0.1048, over 5702731.53 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3512, pruned_loss=0.09645, over 3532461.63 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3547, pruned_loss=0.106, over 5696499.15 frames. ], batch size: 164, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:14:58,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=503126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:15:03,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.075e+02 1.003e+03 1.273e+03 1.945e+03 3.601e+03, threshold=2.546e+03, percent-clipped=5.0
+2023-03-06 02:15:10,535 INFO [train.py:968] (0/2) Epoch 12, batch 1950, giga_loss[loss=0.2534, simple_loss=0.3254, pruned_loss=0.09066, over 28951.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3503, pruned_loss=0.1023, over 5700080.03 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3509, pruned_loss=0.09624, over 3602460.05 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3512, pruned_loss=0.1036, over 5689554.48 frames. ], batch size: 136, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:15:24,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6922, 1.8544, 1.9437, 1.5239], device='cuda:0'), covar=tensor([0.1743, 0.2297, 0.1371, 0.1588], device='cuda:0'), in_proj_covar=tensor([0.0840, 0.0698, 0.0879, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 02:15:58,032 INFO [train.py:968] (0/2) Epoch 12, batch 2000, giga_loss[loss=0.229, simple_loss=0.3068, pruned_loss=0.07559, over 28926.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3445, pruned_loss=0.09894, over 5692469.87 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3509, pruned_loss=0.0963, over 3658112.62 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3451, pruned_loss=0.09998, over 5680749.86 frames. ], batch size: 213, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:16:03,461 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=503197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:16:17,643 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=503212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:16:37,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.948e+02 9.843e+02 1.344e+03 1.998e+03 4.300e+03, threshold=2.688e+03, percent-clipped=14.0
+2023-03-06 02:16:44,195 INFO [train.py:968] (0/2) Epoch 12, batch 2050, giga_loss[loss=0.2362, simple_loss=0.3113, pruned_loss=0.0806, over 28930.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3383, pruned_loss=0.09599, over 5683580.21 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3517, pruned_loss=0.09685, over 3680260.74 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3382, pruned_loss=0.09648, over 5672684.42 frames. ], batch size: 213, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:17:12,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=503269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:17:15,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=503272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:17:20,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2226, 1.3121, 4.7484, 3.4330], device='cuda:0'), covar=tensor([0.1807, 0.2703, 0.0344, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0661, 0.0585, 0.0857, 0.0761], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 02:17:34,344 INFO [train.py:968] (0/2) Epoch 12, batch 2100, libri_loss[loss=0.3238, simple_loss=0.3926, pruned_loss=0.1275, over 29520.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3358, pruned_loss=0.09432, over 5687873.13 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3517, pruned_loss=0.09694, over 3712532.90 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3355, pruned_loss=0.0946, over 5677045.88 frames. ], batch size: 84, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:17:40,611 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=503301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:17:45,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=503308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:18:06,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.329e+02 9.557e+02 1.123e+03 1.549e+03 3.320e+03, threshold=2.246e+03, percent-clipped=2.0
+2023-03-06 02:18:13,769 INFO [train.py:968] (0/2) Epoch 12, batch 2150, giga_loss[loss=0.2586, simple_loss=0.3322, pruned_loss=0.09247, over 28964.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3369, pruned_loss=0.09455, over 5695340.87 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3507, pruned_loss=0.09618, over 3774115.16 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3368, pruned_loss=0.09513, over 5684254.53 frames. ], batch size: 213, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:18:24,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=503355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:18:26,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=503358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:18:48,644 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=503387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:18:52,663 INFO [train.py:968] (0/2) Epoch 12, batch 2200, giga_loss[loss=0.2414, simple_loss=0.3115, pruned_loss=0.0857, over 28860.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3364, pruned_loss=0.09413, over 5699794.70 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3509, pruned_loss=0.09618, over 3813229.75 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.336, pruned_loss=0.09458, over 5690002.64 frames. ], batch size: 112, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:19:06,742 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=503407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:19:27,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.837e+02 9.864e+02 1.305e+03 1.888e+03 6.740e+03, threshold=2.610e+03, percent-clipped=18.0
+2023-03-06 02:19:34,238 INFO [train.py:968] (0/2) Epoch 12, batch 2250, libri_loss[loss=0.2839, simple_loss=0.3495, pruned_loss=0.1091, over 29494.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3348, pruned_loss=0.09328, over 5689880.52 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3517, pruned_loss=0.0966, over 3866089.18 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3334, pruned_loss=0.09328, over 5693678.61 frames. ], batch size: 70, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:20:04,694 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 02:20:07,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0827, 5.8447, 5.5732, 3.0326], device='cuda:0'), covar=tensor([0.0348, 0.0521, 0.0522, 0.1539], device='cuda:0'), in_proj_covar=tensor([0.1035, 0.0962, 0.0845, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 02:20:16,800 INFO [train.py:968] (0/2) Epoch 12, batch 2300, giga_loss[loss=0.2187, simple_loss=0.302, pruned_loss=0.06771, over 28972.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3327, pruned_loss=0.09245, over 5689746.45 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3523, pruned_loss=0.09672, over 3892182.72 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.331, pruned_loss=0.09232, over 5694014.79 frames. ], batch size: 164, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:20:17,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3495, 3.6285, 1.5575, 1.5429], device='cuda:0'), covar=tensor([0.0997, 0.0277, 0.0886, 0.1293], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0510, 0.0342, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 02:20:19,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0316, 1.3345, 1.1209, 0.2907], device='cuda:0'), covar=tensor([0.2259, 0.2147, 0.3065, 0.3877], device='cuda:0'), in_proj_covar=tensor([0.1571, 0.1486, 0.1498, 0.1286], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 02:20:34,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3107, 3.1224, 2.9644, 1.4788], device='cuda:0'), covar=tensor([0.0833, 0.0906, 0.0799, 0.2169], device='cuda:0'), in_proj_covar=tensor([0.1033, 0.0961, 0.0844, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 02:20:51,614 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.867e+02 1.033e+03 1.415e+03 2.034e+03 6.752e+03, threshold=2.830e+03, percent-clipped=13.0
+2023-03-06 02:20:57,011 INFO [train.py:968] (0/2) Epoch 12, batch 2350, giga_loss[loss=0.2298, simple_loss=0.3018, pruned_loss=0.07894, over 28991.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3305, pruned_loss=0.09143, over 5706112.13 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3529, pruned_loss=0.09695, over 3930889.18 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3284, pruned_loss=0.09109, over 5706790.05 frames. ], batch size: 213, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:21:22,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=503572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:21:40,422 INFO [train.py:968] (0/2) Epoch 12, batch 2400, giga_loss[loss=0.2484, simple_loss=0.3116, pruned_loss=0.09267, over 28601.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3289, pruned_loss=0.09101, over 5715146.65 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3539, pruned_loss=0.09753, over 3970143.06 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3263, pruned_loss=0.0903, over 5712283.55 frames. ], batch size: 85, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:22:12,990 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.610e+02 9.774e+02 1.246e+03 1.567e+03 4.324e+03, threshold=2.493e+03, percent-clipped=6.0
+2023-03-06 02:22:18,538 INFO [train.py:968] (0/2) Epoch 12, batch 2450, giga_loss[loss=0.2297, simple_loss=0.3111, pruned_loss=0.07414, over 28917.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3265, pruned_loss=0.09002, over 5714950.43 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.355, pruned_loss=0.09824, over 3989440.69 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3234, pruned_loss=0.08892, over 5718592.70 frames. ], batch size: 145, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:22:49,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=503683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:22:56,075 INFO [train.py:968] (0/2) Epoch 12, batch 2500, giga_loss[loss=0.2203, simple_loss=0.2951, pruned_loss=0.07277, over 28420.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3239, pruned_loss=0.08828, over 5719839.70 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3554, pruned_loss=0.09807, over 4027344.26 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3206, pruned_loss=0.08733, over 5719586.62 frames. ], batch size: 71, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:23:15,882 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=503715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:23:18,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=503718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:23:22,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5793, 1.7689, 1.8368, 1.3953], device='cuda:0'), covar=tensor([0.1861, 0.2237, 0.1436, 0.1552], device='cuda:0'), in_proj_covar=tensor([0.0843, 0.0698, 0.0881, 0.0779], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 02:23:32,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.235e+02 9.569e+02 1.168e+03 1.469e+03 5.746e+03, threshold=2.336e+03, percent-clipped=2.0
+2023-03-06 02:23:39,410 INFO [train.py:968] (0/2) Epoch 12, batch 2550, giga_loss[loss=0.3097, simple_loss=0.3624, pruned_loss=0.1285, over 26647.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3218, pruned_loss=0.08768, over 5713245.45 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3553, pruned_loss=0.09804, over 4036678.54 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3192, pruned_loss=0.08692, over 5712340.24 frames. ], batch size: 555, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:23:43,572 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=503747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:24:10,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=503782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:24:18,158 INFO [train.py:968] (0/2) Epoch 12, batch 2600, giga_loss[loss=0.216, simple_loss=0.2894, pruned_loss=0.07128, over 28436.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3204, pruned_loss=0.08648, over 5720347.08 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3554, pruned_loss=0.09776, over 4081246.84 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3174, pruned_loss=0.0858, over 5717044.48 frames. ], batch size: 65, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:24:44,867 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=503826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:24:47,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=503829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:24:51,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.240e+02 9.606e+02 1.191e+03 1.719e+03 6.821e+03, threshold=2.382e+03, percent-clipped=9.0
+2023-03-06 02:24:52,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5055, 1.7336, 1.8338, 1.3449], device='cuda:0'), covar=tensor([0.1651, 0.2129, 0.1313, 0.1471], device='cuda:0'), in_proj_covar=tensor([0.0845, 0.0700, 0.0882, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 02:24:57,426 INFO [train.py:968] (0/2) Epoch 12, batch 2650, giga_loss[loss=0.2605, simple_loss=0.3385, pruned_loss=0.09131, over 28773.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3204, pruned_loss=0.08655, over 5711778.34 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3567, pruned_loss=0.0984, over 4115609.58 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3164, pruned_loss=0.08534, over 5715508.96 frames. ], batch size: 284, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:25:10,559 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=503858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:25:28,811 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=503879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:25:37,353 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=503886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:25:42,360 INFO [train.py:968] (0/2) Epoch 12, batch 2700, giga_loss[loss=0.2534, simple_loss=0.3273, pruned_loss=0.08972, over 28820.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3238, pruned_loss=0.08861, over 5712273.60 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3572, pruned_loss=0.09855, over 4168615.69 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3193, pruned_loss=0.08722, over 5710095.11 frames. ], batch size: 199, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:25:42,879 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 02:26:11,321 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=503925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:26:14,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=503928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:26:19,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.824e+02 1.147e+03 1.401e+03 1.697e+03 3.898e+03, threshold=2.801e+03, percent-clipped=7.0
+2023-03-06 02:26:27,288 INFO [train.py:968] (0/2) Epoch 12, batch 2750, giga_loss[loss=0.2991, simple_loss=0.3668, pruned_loss=0.1157, over 28568.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3296, pruned_loss=0.09224, over 5711454.66 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3566, pruned_loss=0.09807, over 4203063.81 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3258, pruned_loss=0.09129, over 5706445.39 frames. ], batch size: 336, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:26:41,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=503957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:27:14,799 INFO [train.py:968] (0/2) Epoch 12, batch 2800, libri_loss[loss=0.2527, simple_loss=0.3242, pruned_loss=0.09059, over 29474.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3383, pruned_loss=0.09836, over 5703002.80 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3567, pruned_loss=0.09818, over 4228779.88 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3349, pruned_loss=0.09751, over 5696083.88 frames. ], batch size: 70, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:27:22,012 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-504000.pt
+2023-03-06 02:27:29,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0006, 1.3288, 1.0396, 0.2361], device='cuda:0'), covar=tensor([0.2449, 0.2056, 0.3280, 0.4116], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1486, 0.1499, 0.1277], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 02:27:53,187 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.494e+02 1.293e+03 1.722e+03 2.568e+03 5.203e+03, threshold=3.444e+03, percent-clipped=21.0
+2023-03-06 02:27:57,981 INFO [train.py:968] (0/2) Epoch 12, batch 2850, giga_loss[loss=0.3073, simple_loss=0.3722, pruned_loss=0.1212, over 28538.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3443, pruned_loss=0.1016, over 5694195.42 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3564, pruned_loss=0.09802, over 4306049.88 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3411, pruned_loss=0.1012, over 5683300.42 frames. ], batch size: 307, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:28:49,775 INFO [train.py:968] (0/2) Epoch 12, batch 2900, giga_loss[loss=0.2879, simple_loss=0.3668, pruned_loss=0.1045, over 28784.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3509, pruned_loss=0.105, over 5675075.83 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3564, pruned_loss=0.09802, over 4321666.58 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3482, pruned_loss=0.1047, over 5664966.63 frames. ], batch size: 186, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:29:08,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=504115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:29:25,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.632e+02 1.097e+03 1.292e+03 1.853e+03 3.767e+03, threshold=2.583e+03, percent-clipped=2.0
+2023-03-06 02:29:29,096 INFO [train.py:968] (0/2) Epoch 12, batch 2950, giga_loss[loss=0.3216, simple_loss=0.3885, pruned_loss=0.1274, over 28727.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3555, pruned_loss=0.1066, over 5689577.64 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3551, pruned_loss=0.09751, over 4375029.18 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3542, pruned_loss=0.107, over 5675425.99 frames. ], batch size: 262, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:30:09,025 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 02:30:09,429 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=504182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:30:17,775 INFO [train.py:968] (0/2) Epoch 12, batch 3000, giga_loss[loss=0.2824, simple_loss=0.3587, pruned_loss=0.1031, over 28797.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3609, pruned_loss=0.1105, over 5684225.18 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3548, pruned_loss=0.09746, over 4397951.18 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3602, pruned_loss=0.111, over 5670016.49 frames. ], batch size: 186, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:30:17,780 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 02:30:24,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1343, 1.5257, 1.5408, 1.3290], device='cuda:0'), covar=tensor([0.1510, 0.1343, 0.1784, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0729, 0.0673, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 02:30:26,066 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2252, simple_loss=0.3291, pruned_loss=0.06071, over 944034.00 frames. 
+2023-03-06 02:30:26,067 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 02:31:05,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.374e+02 1.220e+03 1.556e+03 2.001e+03 6.144e+03, threshold=3.111e+03, percent-clipped=16.0
+2023-03-06 02:31:09,484 INFO [train.py:968] (0/2) Epoch 12, batch 3050, giga_loss[loss=0.2428, simple_loss=0.3245, pruned_loss=0.08054, over 28623.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3557, pruned_loss=0.1064, over 5690281.29 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3545, pruned_loss=0.09742, over 4420753.51 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3554, pruned_loss=0.107, over 5675905.64 frames. ], batch size: 336, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:31:18,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=504254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:31:26,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=504261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:31:53,526 INFO [train.py:968] (0/2) Epoch 12, batch 3100, giga_loss[loss=0.2463, simple_loss=0.328, pruned_loss=0.08231, over 28947.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3529, pruned_loss=0.104, over 5689709.14 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3548, pruned_loss=0.09759, over 4435698.83 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3524, pruned_loss=0.1044, over 5676397.82 frames. ], batch size: 174, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:32:29,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6864, 4.7796, 1.8617, 1.9355], device='cuda:0'), covar=tensor([0.0935, 0.0187, 0.0822, 0.1226], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0508, 0.0342, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 02:32:32,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.862e+02 1.104e+03 1.511e+03 2.022e+03 4.319e+03, threshold=3.021e+03, percent-clipped=6.0
+2023-03-06 02:32:36,610 INFO [train.py:968] (0/2) Epoch 12, batch 3150, giga_loss[loss=0.3004, simple_loss=0.3718, pruned_loss=0.1145, over 28741.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3521, pruned_loss=0.103, over 5677231.79 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3541, pruned_loss=0.09726, over 4485944.01 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3522, pruned_loss=0.1038, over 5671355.20 frames. ], batch size: 262, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:32:52,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4740, 1.6761, 1.3207, 1.4743], device='cuda:0'), covar=tensor([0.2421, 0.2420, 0.2689, 0.2216], device='cuda:0'), in_proj_covar=tensor([0.1300, 0.0970, 0.1150, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 02:33:15,410 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=504387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:18,560 INFO [train.py:968] (0/2) Epoch 12, batch 3200, giga_loss[loss=0.2802, simple_loss=0.3582, pruned_loss=0.1011, over 28864.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3528, pruned_loss=0.103, over 5679695.07 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3538, pruned_loss=0.097, over 4525194.01 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.353, pruned_loss=0.104, over 5671436.09 frames. ], batch size: 112, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:33:23,193 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=504397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:27,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=504400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:31,379 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=504404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:34,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=504407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:52,139 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=504429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:57,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=504436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:58,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.654e+02 1.319e+03 1.652e+03 2.418e+03 8.835e+03, threshold=3.305e+03, percent-clipped=18.0
+2023-03-06 02:34:01,581 INFO [train.py:968] (0/2) Epoch 12, batch 3250, giga_loss[loss=0.2733, simple_loss=0.3506, pruned_loss=0.09795, over 28877.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3558, pruned_loss=0.1049, over 5684749.14 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3541, pruned_loss=0.09713, over 4552448.58 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3557, pruned_loss=0.1058, over 5674143.90 frames. ], batch size: 227, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:34:46,163 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=504490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:34:47,155 INFO [train.py:968] (0/2) Epoch 12, batch 3300, giga_loss[loss=0.2899, simple_loss=0.3599, pruned_loss=0.1099, over 28670.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3579, pruned_loss=0.1065, over 5690647.92 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3542, pruned_loss=0.09722, over 4571833.30 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3578, pruned_loss=0.1072, over 5680263.24 frames. ], batch size: 262, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:35:25,998 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.637e+02 1.181e+03 1.450e+03 1.788e+03 3.113e+03, threshold=2.900e+03, percent-clipped=0.0
+2023-03-06 02:35:29,653 INFO [train.py:968] (0/2) Epoch 12, batch 3350, giga_loss[loss=0.2931, simple_loss=0.3658, pruned_loss=0.1102, over 29020.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3582, pruned_loss=0.107, over 5693866.70 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3544, pruned_loss=0.09726, over 4578492.13 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3581, pruned_loss=0.1076, over 5684860.66 frames. ], batch size: 128, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:35:45,919 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=504557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:35:46,705 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 02:36:16,246 INFO [train.py:968] (0/2) Epoch 12, batch 3400, giga_loss[loss=0.3083, simple_loss=0.3731, pruned_loss=0.1218, over 28044.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3594, pruned_loss=0.1087, over 5679810.61 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3544, pruned_loss=0.09725, over 4588657.01 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3593, pruned_loss=0.1093, over 5678040.45 frames. ], batch size: 412, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:36:50,264 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=504633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:36:52,965 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=504636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:36:55,233 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.238e+02 1.171e+03 1.548e+03 2.062e+03 5.520e+03, threshold=3.097e+03, percent-clipped=6.0
+2023-03-06 02:36:56,729 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=504640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:36:57,856 INFO [train.py:968] (0/2) Epoch 12, batch 3450, giga_loss[loss=0.2742, simple_loss=0.3545, pruned_loss=0.09691, over 29081.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3594, pruned_loss=0.1083, over 5669437.16 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3545, pruned_loss=0.09724, over 4608536.72 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3593, pruned_loss=0.1091, over 5674295.56 frames. ], batch size: 155, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:37:16,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=504665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:37:24,456 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-06 02:37:38,624 INFO [train.py:968] (0/2) Epoch 12, batch 3500, giga_loss[loss=0.2845, simple_loss=0.3624, pruned_loss=0.1033, over 28344.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3593, pruned_loss=0.1071, over 5683261.39 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3545, pruned_loss=0.09715, over 4628028.10 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3594, pruned_loss=0.1079, over 5684023.07 frames. ], batch size: 368, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:37:46,086 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=504700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:37:48,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=504703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:38:13,429 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=504732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:38:19,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.310e+02 1.094e+03 1.323e+03 1.701e+03 5.426e+03, threshold=2.647e+03, percent-clipped=7.0
+2023-03-06 02:38:22,375 INFO [train.py:968] (0/2) Epoch 12, batch 3550, giga_loss[loss=0.283, simple_loss=0.3624, pruned_loss=0.1018, over 28950.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.358, pruned_loss=0.1052, over 5687590.35 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3537, pruned_loss=0.09668, over 4657424.32 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3588, pruned_loss=0.1064, over 5685225.77 frames. ], batch size: 213, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:38:42,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=504762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:39:04,386 INFO [train.py:968] (0/2) Epoch 12, batch 3600, giga_loss[loss=0.2736, simple_loss=0.3426, pruned_loss=0.1023, over 28989.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3569, pruned_loss=0.1042, over 5696199.81 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3537, pruned_loss=0.09669, over 4682242.78 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3576, pruned_loss=0.1053, over 5690318.65 frames. ], batch size: 155, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:39:39,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2677, 1.4746, 1.4484, 1.3246], device='cuda:0'), covar=tensor([0.1433, 0.1476, 0.1970, 0.1498], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0730, 0.0676, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 02:39:43,653 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.172e+02 9.933e+02 1.314e+03 1.751e+03 4.886e+03, threshold=2.629e+03, percent-clipped=8.0
+2023-03-06 02:39:45,731 INFO [train.py:968] (0/2) Epoch 12, batch 3650, giga_loss[loss=0.2477, simple_loss=0.327, pruned_loss=0.08425, over 28474.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3552, pruned_loss=0.1035, over 5691247.09 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3544, pruned_loss=0.09709, over 4711682.77 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3552, pruned_loss=0.1043, over 5682361.42 frames. ], batch size: 60, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:39:46,972 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-06 02:40:01,243 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=504862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:40:23,581 INFO [train.py:968] (0/2) Epoch 12, batch 3700, giga_loss[loss=0.277, simple_loss=0.3544, pruned_loss=0.09987, over 28791.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3525, pruned_loss=0.1017, over 5709850.94 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3551, pruned_loss=0.09762, over 4758633.14 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3519, pruned_loss=0.102, over 5695216.23 frames. ], batch size: 284, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:40:33,503 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=504905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:40:35,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=504908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:40:57,857 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=504937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:40:59,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.018e+02 9.734e+02 1.225e+03 1.695e+03 5.014e+03, threshold=2.450e+03, percent-clipped=9.0
+2023-03-06 02:41:01,875 INFO [train.py:968] (0/2) Epoch 12, batch 3750, giga_loss[loss=0.2648, simple_loss=0.3406, pruned_loss=0.09447, over 28553.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3514, pruned_loss=0.1012, over 5712065.28 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.355, pruned_loss=0.09743, over 4780312.48 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3509, pruned_loss=0.1017, over 5697869.22 frames. ], batch size: 78, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:41:29,905 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-06 02:41:35,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4320, 3.3195, 1.5727, 1.5544], device='cuda:0'), covar=tensor([0.0934, 0.0274, 0.0812, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0503, 0.0339, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-06 02:41:47,551 INFO [train.py:968] (0/2) Epoch 12, batch 3800, giga_loss[loss=0.3096, simple_loss=0.3745, pruned_loss=0.1223, over 27989.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3534, pruned_loss=0.1032, over 5708743.44 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3546, pruned_loss=0.09733, over 4802549.40 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3533, pruned_loss=0.1037, over 5693961.82 frames. ], batch size: 412, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:41:54,476 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-06 02:42:04,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=505015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:42:07,637 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5343, 1.7241, 1.8513, 1.3609], device='cuda:0'), covar=tensor([0.1746, 0.2444, 0.1406, 0.1634], device='cuda:0'), in_proj_covar=tensor([0.0834, 0.0696, 0.0878, 0.0778], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 02:42:21,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.935e+02 1.003e+03 1.210e+03 1.466e+03 2.967e+03, threshold=2.419e+03, percent-clipped=4.0
+2023-03-06 02:42:25,249 INFO [train.py:968] (0/2) Epoch 12, batch 3850, giga_loss[loss=0.2426, simple_loss=0.3302, pruned_loss=0.0775, over 28796.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3526, pruned_loss=0.1021, over 5702961.78 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3545, pruned_loss=0.09725, over 4815905.18 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3526, pruned_loss=0.1027, over 5695220.69 frames. ], batch size: 119, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:42:38,450 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=505058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:43:04,935 INFO [train.py:968] (0/2) Epoch 12, batch 3900, giga_loss[loss=0.2781, simple_loss=0.3591, pruned_loss=0.09851, over 28658.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3511, pruned_loss=0.1004, over 5711555.24 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.354, pruned_loss=0.09705, over 4841203.74 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3514, pruned_loss=0.1011, over 5702538.23 frames. ], batch size: 66, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:43:20,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9977, 1.1673, 1.3206, 0.9949], device='cuda:0'), covar=tensor([0.1544, 0.1275, 0.1959, 0.1611], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0723, 0.0671, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-06 02:43:41,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4796, 2.0296, 1.7313, 1.6478], device='cuda:0'), covar=tensor([0.0760, 0.0265, 0.0281, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0209], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 02:43:45,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.600e+02 9.586e+02 1.282e+03 1.785e+03 6.403e+03, threshold=2.563e+03, percent-clipped=11.0
+2023-03-06 02:43:48,286 INFO [train.py:968] (0/2) Epoch 12, batch 3950, giga_loss[loss=0.2783, simple_loss=0.3558, pruned_loss=0.1005, over 28949.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3508, pruned_loss=0.1003, over 5706318.28 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3538, pruned_loss=0.09686, over 4853307.01 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3512, pruned_loss=0.101, over 5699961.89 frames. ], batch size: 145, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:43:52,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6886, 4.4978, 4.2739, 1.9035], device='cuda:0'), covar=tensor([0.0446, 0.0643, 0.0675, 0.2149], device='cuda:0'), in_proj_covar=tensor([0.1037, 0.0967, 0.0844, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 02:44:00,472 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=505158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:44:03,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=505161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:44:10,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4296, 1.7060, 1.6017, 1.5313], device='cuda:0'), covar=tensor([0.1810, 0.1896, 0.2098, 0.1951], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0726, 0.0671, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 02:44:27,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=505190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:44:28,362 INFO [train.py:968] (0/2) Epoch 12, batch 4000, giga_loss[loss=0.2571, simple_loss=0.3259, pruned_loss=0.09411, over 28538.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3511, pruned_loss=0.1011, over 5709705.74 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3544, pruned_loss=0.09722, over 4878454.44 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3508, pruned_loss=0.1015, over 5700836.39 frames. ], batch size: 71, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:45:03,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2065, 1.6572, 1.2244, 0.4927], device='cuda:0'), covar=tensor([0.3050, 0.1906, 0.2384, 0.4362], device='cuda:0'), in_proj_covar=tensor([0.1537, 0.1452, 0.1472, 0.1266], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 02:45:04,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=505237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:45:06,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.741e+02 9.803e+02 1.221e+03 1.749e+03 6.697e+03, threshold=2.442e+03, percent-clipped=10.0
+2023-03-06 02:45:09,269 INFO [train.py:968] (0/2) Epoch 12, batch 4050, giga_loss[loss=0.2997, simple_loss=0.3692, pruned_loss=0.1151, over 27559.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3482, pruned_loss=0.09962, over 5708589.72 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3546, pruned_loss=0.0973, over 4889799.01 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3478, pruned_loss=0.09989, over 5702706.10 frames. ], batch size: 472, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:45:47,766 INFO [train.py:968] (0/2) Epoch 12, batch 4100, giga_loss[loss=0.2519, simple_loss=0.3253, pruned_loss=0.08922, over 28923.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3446, pruned_loss=0.09747, over 5717512.86 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3537, pruned_loss=0.09668, over 4916811.24 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3447, pruned_loss=0.09822, over 5710237.94 frames. ], batch size: 112, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:46:27,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.792e+02 1.050e+03 1.283e+03 1.829e+03 4.791e+03, threshold=2.565e+03, percent-clipped=6.0
+2023-03-06 02:46:29,268 INFO [train.py:968] (0/2) Epoch 12, batch 4150, giga_loss[loss=0.2484, simple_loss=0.3301, pruned_loss=0.08333, over 29019.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3442, pruned_loss=0.09779, over 5706032.06 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3534, pruned_loss=0.09665, over 4931721.44 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3443, pruned_loss=0.09842, over 5704371.98 frames. ], batch size: 164, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:46:47,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7587, 1.7963, 1.3418, 1.4143], device='cuda:0'), covar=tensor([0.0683, 0.0536, 0.0979, 0.0970], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0432, 0.0496, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 02:46:59,666 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=505380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:47:02,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=505383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:47:10,576 INFO [train.py:968] (0/2) Epoch 12, batch 4200, giga_loss[loss=0.2433, simple_loss=0.3207, pruned_loss=0.08294, over 28850.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3443, pruned_loss=0.09847, over 5707990.21 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3536, pruned_loss=0.09668, over 4949891.17 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3441, pruned_loss=0.09897, over 5704344.84 frames. ], batch size: 112, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:47:26,084 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=505412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:47:45,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=505433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:47:50,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.252e+02 1.059e+03 1.313e+03 1.648e+03 5.608e+03, threshold=2.627e+03, percent-clipped=6.0
+2023-03-06 02:47:51,663 INFO [train.py:968] (0/2) Epoch 12, batch 4250, giga_loss[loss=0.2455, simple_loss=0.3225, pruned_loss=0.08423, over 28646.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.343, pruned_loss=0.0981, over 5707130.66 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3534, pruned_loss=0.09647, over 4967720.75 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3427, pruned_loss=0.09872, over 5707703.74 frames. ], batch size: 262, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:48:01,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6337, 2.3446, 1.7079, 0.8638], device='cuda:0'), covar=tensor([0.3317, 0.1971, 0.2629, 0.3340], device='cuda:0'), in_proj_covar=tensor([0.1537, 0.1456, 0.1476, 0.1266], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 02:48:06,113 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=505461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:48:09,979 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3118, 1.6193, 1.4046, 1.2174], device='cuda:0'), covar=tensor([0.2369, 0.1885, 0.1475, 0.1925], device='cuda:0'), in_proj_covar=tensor([0.1715, 0.1628, 0.1601, 0.1681], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 02:48:29,156 INFO [train.py:968] (0/2) Epoch 12, batch 4300, giga_loss[loss=0.2855, simple_loss=0.3562, pruned_loss=0.1075, over 28577.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3415, pruned_loss=0.09745, over 5714329.94 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3536, pruned_loss=0.09661, over 5011720.33 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3407, pruned_loss=0.09793, over 5707624.70 frames. ], batch size: 336, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:49:06,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.677e+02 1.046e+03 1.260e+03 1.582e+03 5.700e+03, threshold=2.520e+03, percent-clipped=11.0
+2023-03-06 02:49:08,486 INFO [train.py:968] (0/2) Epoch 12, batch 4350, giga_loss[loss=0.2412, simple_loss=0.323, pruned_loss=0.07973, over 28972.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3383, pruned_loss=0.09592, over 5715850.84 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3537, pruned_loss=0.0968, over 5024238.11 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3373, pruned_loss=0.09615, over 5708717.10 frames. ], batch size: 136, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:49:13,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1903, 1.4146, 1.3868, 1.1941], device='cuda:0'), covar=tensor([0.2239, 0.1867, 0.1083, 0.1543], device='cuda:0'), in_proj_covar=tensor([0.1712, 0.1628, 0.1600, 0.1681], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 02:49:36,228 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=505576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:49:39,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=505579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:49:48,124 INFO [train.py:968] (0/2) Epoch 12, batch 4400, libri_loss[loss=0.2312, simple_loss=0.3116, pruned_loss=0.07538, over 29574.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3372, pruned_loss=0.09511, over 5716165.77 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3535, pruned_loss=0.09655, over 5054680.95 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.336, pruned_loss=0.09547, over 5703920.44 frames. ], batch size: 75, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:49:55,236 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=505602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:49:59,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=505608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:50:15,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4520, 2.4009, 2.4403, 2.0182], device='cuda:0'), covar=tensor([0.1369, 0.1875, 0.1490, 0.1809], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0730, 0.0672, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 02:50:28,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.187e+02 1.014e+03 1.365e+03 1.853e+03 3.098e+03, threshold=2.730e+03, percent-clipped=7.0
+2023-03-06 02:50:29,753 INFO [train.py:968] (0/2) Epoch 12, batch 4450, libri_loss[loss=0.3165, simple_loss=0.3939, pruned_loss=0.1195, over 28672.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3404, pruned_loss=0.09697, over 5713287.99 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3535, pruned_loss=0.09661, over 5079237.62 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.339, pruned_loss=0.0972, over 5698917.66 frames. ], batch size: 106, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:50:36,389 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-06 02:51:12,458 INFO [train.py:968] (0/2) Epoch 12, batch 4500, giga_loss[loss=0.2971, simple_loss=0.3735, pruned_loss=0.1104, over 28596.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3427, pruned_loss=0.09785, over 5723553.22 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3537, pruned_loss=0.09689, over 5103042.09 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3412, pruned_loss=0.09781, over 5707662.08 frames. ], batch size: 336, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:51:27,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4125, 4.1004, 1.6175, 1.6699], device='cuda:0'), covar=tensor([0.0913, 0.0201, 0.0896, 0.1230], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0506, 0.0338, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-06 02:51:31,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9142, 2.5899, 1.9952, 1.5844], device='cuda:0'), covar=tensor([0.2608, 0.1542, 0.1854, 0.2156], device='cuda:0'), in_proj_covar=tensor([0.1714, 0.1626, 0.1601, 0.1679], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 02:51:53,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.138e+02 1.049e+03 1.345e+03 1.748e+03 3.182e+03, threshold=2.690e+03, percent-clipped=5.0
+2023-03-06 02:51:56,848 INFO [train.py:968] (0/2) Epoch 12, batch 4550, libri_loss[loss=0.2731, simple_loss=0.3524, pruned_loss=0.09687, over 29538.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3438, pruned_loss=0.09762, over 5721858.59 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3534, pruned_loss=0.09684, over 5114170.48 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3426, pruned_loss=0.09764, over 5712620.38 frames. ], batch size: 89, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:52:41,505 INFO [train.py:968] (0/2) Epoch 12, batch 4600, giga_loss[loss=0.2609, simple_loss=0.3451, pruned_loss=0.08842, over 28624.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3449, pruned_loss=0.09803, over 5710696.70 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3531, pruned_loss=0.09677, over 5129600.45 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.344, pruned_loss=0.09811, over 5700429.28 frames. ], batch size: 262, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:53:21,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=505836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:53:23,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.872e+02 9.912e+02 1.185e+03 1.593e+03 3.194e+03, threshold=2.370e+03, percent-clipped=4.0
+2023-03-06 02:53:25,482 INFO [train.py:968] (0/2) Epoch 12, batch 4650, giga_loss[loss=0.2494, simple_loss=0.3298, pruned_loss=0.08456, over 28888.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3436, pruned_loss=0.09648, over 5704819.05 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.353, pruned_loss=0.0967, over 5139519.20 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3428, pruned_loss=0.0966, over 5695598.95 frames. ], batch size: 186, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:54:04,702 INFO [train.py:968] (0/2) Epoch 12, batch 4700, giga_loss[loss=0.2784, simple_loss=0.3512, pruned_loss=0.1028, over 28897.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3446, pruned_loss=0.09686, over 5714547.15 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3527, pruned_loss=0.09669, over 5180469.04 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3438, pruned_loss=0.09697, over 5697276.71 frames. ], batch size: 199, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:54:43,922 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.383e+02 1.373e+03 1.803e+03 2.671e+03 6.642e+03, threshold=3.606e+03, percent-clipped=31.0
+2023-03-06 02:54:45,471 INFO [train.py:968] (0/2) Epoch 12, batch 4750, giga_loss[loss=0.2896, simple_loss=0.364, pruned_loss=0.1076, over 28005.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3459, pruned_loss=0.09819, over 5719820.46 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3522, pruned_loss=0.0965, over 5194716.98 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3456, pruned_loss=0.09846, over 5702817.73 frames. ], batch size: 412, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:54:52,664 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=505951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:55:06,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9661, 1.3092, 1.0419, 0.2626], device='cuda:0'), covar=tensor([0.2568, 0.2140, 0.3522, 0.4572], device='cuda:0'), in_proj_covar=tensor([0.1538, 0.1451, 0.1475, 0.1262], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 02:55:07,411 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 02:55:08,436 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-06 02:55:15,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=505977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:55:16,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=505979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:55:18,580 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=505982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:55:26,601 INFO [train.py:968] (0/2) Epoch 12, batch 4800, giga_loss[loss=0.2973, simple_loss=0.3669, pruned_loss=0.1138, over 28879.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3483, pruned_loss=0.09994, over 5714362.14 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3533, pruned_loss=0.09734, over 5195569.67 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.347, pruned_loss=0.09946, over 5707232.45 frames. ], batch size: 186, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:55:34,602 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-506000.pt
+2023-03-06 02:55:44,089 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-06 02:55:45,996 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:55:52,138 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-06 02:55:55,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5938, 2.2288, 1.7850, 1.8564], device='cuda:0'), covar=tensor([0.0722, 0.0242, 0.0287, 0.0785], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 02:56:07,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.927e+02 1.197e+03 1.476e+03 1.908e+03 4.317e+03, threshold=2.952e+03, percent-clipped=2.0
+2023-03-06 02:56:09,796 INFO [train.py:968] (0/2) Epoch 12, batch 4850, giga_loss[loss=0.2859, simple_loss=0.3604, pruned_loss=0.1057, over 28857.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.351, pruned_loss=0.1018, over 5713498.72 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3536, pruned_loss=0.0975, over 5209518.83 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3496, pruned_loss=0.1013, over 5706386.26 frames. ], batch size: 186, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:56:25,394 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:56:50,511 INFO [train.py:968] (0/2) Epoch 12, batch 4900, giga_loss[loss=0.2784, simple_loss=0.3579, pruned_loss=0.09951, over 28656.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3529, pruned_loss=0.1025, over 5712010.03 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3536, pruned_loss=0.09734, over 5224489.13 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3518, pruned_loss=0.1023, over 5704380.31 frames. ], batch size: 242, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:57:15,220 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=506120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:57:17,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=506123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:57:32,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.639e+02 1.161e+03 1.321e+03 1.781e+03 5.224e+03, threshold=2.643e+03, percent-clipped=4.0
+2023-03-06 02:57:32,655 INFO [train.py:968] (0/2) Epoch 12, batch 4950, giga_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.08993, over 28639.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3551, pruned_loss=0.1036, over 5709997.28 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3539, pruned_loss=0.09742, over 5233956.16 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.354, pruned_loss=0.1035, over 5704215.75 frames. ], batch size: 307, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:57:41,091 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:57:41,318 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-06 02:58:15,634 INFO [train.py:968] (0/2) Epoch 12, batch 5000, libri_loss[loss=0.3035, simple_loss=0.3651, pruned_loss=0.1209, over 29646.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3565, pruned_loss=0.1047, over 5708937.09 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3541, pruned_loss=0.09761, over 5240427.55 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3554, pruned_loss=0.1046, over 5702789.78 frames. ], batch size: 73, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:58:54,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.296e+02 1.049e+03 1.300e+03 1.747e+03 3.716e+03, threshold=2.600e+03, percent-clipped=8.0
+2023-03-06 02:58:56,483 INFO [train.py:968] (0/2) Epoch 12, batch 5050, giga_loss[loss=0.3089, simple_loss=0.3635, pruned_loss=0.1272, over 23588.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3556, pruned_loss=0.104, over 5706597.77 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3543, pruned_loss=0.09769, over 5249662.76 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3547, pruned_loss=0.1039, over 5699889.99 frames. ], batch size: 705, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:59:07,773 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:59:10,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3866, 1.5433, 1.5010, 1.3714], device='cuda:0'), covar=tensor([0.1238, 0.1741, 0.1758, 0.1691], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0726, 0.0672, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 02:59:37,448 INFO [train.py:968] (0/2) Epoch 12, batch 5100, giga_loss[loss=0.2438, simple_loss=0.3288, pruned_loss=0.07939, over 29065.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3554, pruned_loss=0.1038, over 5710979.16 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.355, pruned_loss=0.0981, over 5259260.22 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3542, pruned_loss=0.1036, over 5703372.41 frames. ], batch size: 164, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 03:00:07,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=506326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:00:17,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.668e+02 1.134e+03 1.506e+03 2.089e+03 7.924e+03, threshold=3.013e+03, percent-clipped=15.0
+2023-03-06 03:00:18,227 INFO [train.py:968] (0/2) Epoch 12, batch 5150, giga_loss[loss=0.2253, simple_loss=0.3008, pruned_loss=0.07492, over 28411.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3519, pruned_loss=0.1022, over 5709094.13 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3553, pruned_loss=0.09829, over 5271935.66 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3506, pruned_loss=0.1019, over 5699649.41 frames. ], batch size: 60, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 03:00:30,691 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:00:59,063 INFO [train.py:968] (0/2) Epoch 12, batch 5200, giga_loss[loss=0.2534, simple_loss=0.3307, pruned_loss=0.08808, over 28814.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3484, pruned_loss=0.1004, over 5715275.18 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3553, pruned_loss=0.09834, over 5287390.21 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3472, pruned_loss=0.1001, over 5703794.22 frames. ], batch size: 186, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 03:01:09,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6434, 1.8583, 1.5820, 1.8478], device='cuda:0'), covar=tensor([0.2346, 0.2310, 0.2564, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.1302, 0.0965, 0.1148, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 03:01:34,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=506436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:01:38,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.530e+02 1.018e+03 1.261e+03 1.539e+03 4.002e+03, threshold=2.523e+03, percent-clipped=3.0
+2023-03-06 03:01:40,138 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-06 03:01:40,307 INFO [train.py:968] (0/2) Epoch 12, batch 5250, giga_loss[loss=0.2764, simple_loss=0.3609, pruned_loss=0.0959, over 28318.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3491, pruned_loss=0.0997, over 5714643.00 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3555, pruned_loss=0.09846, over 5295429.17 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3479, pruned_loss=0.09944, over 5704227.40 frames. ], batch size: 368, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 03:02:04,284 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=506469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:02:06,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=506472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:02:23,795 INFO [train.py:968] (0/2) Epoch 12, batch 5300, giga_loss[loss=0.2512, simple_loss=0.3371, pruned_loss=0.08265, over 28899.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3513, pruned_loss=0.09975, over 5717051.27 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3554, pruned_loss=0.09834, over 5304785.28 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3504, pruned_loss=0.09966, over 5706221.67 frames. ], batch size: 164, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 03:02:31,275 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:03:05,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.776e+02 1.109e+03 1.409e+03 2.019e+03 4.084e+03, threshold=2.818e+03, percent-clipped=9.0
+2023-03-06 03:03:06,086 INFO [train.py:968] (0/2) Epoch 12, batch 5350, giga_loss[loss=0.2809, simple_loss=0.345, pruned_loss=0.1084, over 28605.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3515, pruned_loss=0.1001, over 5721253.93 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3549, pruned_loss=0.0981, over 5318645.86 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3511, pruned_loss=0.1003, over 5709296.13 frames. ], batch size: 92, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 03:03:37,984 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=506579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:03:39,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=506582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:03:48,537 INFO [train.py:968] (0/2) Epoch 12, batch 5400, giga_loss[loss=0.2789, simple_loss=0.3526, pruned_loss=0.1026, over 28695.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.349, pruned_loss=0.1003, over 5726348.14 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3551, pruned_loss=0.09818, over 5324620.16 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3484, pruned_loss=0.1004, over 5715364.50 frames. ], batch size: 262, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 03:03:57,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:04:03,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3451, 1.6661, 1.3412, 1.2880], device='cuda:0'), covar=tensor([0.1897, 0.1840, 0.2001, 0.1866], device='cuda:0'), in_proj_covar=tensor([0.1300, 0.0962, 0.1147, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 03:04:05,476 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:04:22,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=506631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:04:30,987 INFO [train.py:968] (0/2) Epoch 12, batch 5450, giga_loss[loss=0.283, simple_loss=0.3527, pruned_loss=0.1066, over 28850.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3483, pruned_loss=0.1018, over 5714748.15 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3552, pruned_loss=0.09854, over 5321875.34 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3476, pruned_loss=0.1016, over 5720308.88 frames. ], batch size: 227, lr: 2.74e-03, grad_scale: 2.0
+2023-03-06 03:04:31,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.091e+02 1.110e+03 1.537e+03 2.341e+03 5.162e+03, threshold=3.073e+03, percent-clipped=14.0
+2023-03-06 03:04:49,206 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:05:12,475 INFO [train.py:968] (0/2) Epoch 12, batch 5500, giga_loss[loss=0.2393, simple_loss=0.3129, pruned_loss=0.08286, over 29029.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3452, pruned_loss=0.1009, over 5719268.15 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3555, pruned_loss=0.09871, over 5330105.06 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3443, pruned_loss=0.1006, over 5723024.78 frames. ], batch size: 128, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:05:46,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=506732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:05:55,850 INFO [train.py:968] (0/2) Epoch 12, batch 5550, giga_loss[loss=0.3016, simple_loss=0.3532, pruned_loss=0.125, over 28751.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3442, pruned_loss=0.1007, over 5719823.77 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3553, pruned_loss=0.0987, over 5338816.74 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3435, pruned_loss=0.1005, over 5720463.99 frames. ], batch size: 99, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:05:58,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.845e+02 1.150e+03 1.430e+03 2.032e+03 4.854e+03, threshold=2.860e+03, percent-clipped=10.0
+2023-03-06 03:06:24,816 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=506774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:06:26,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=506777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:06:39,291 INFO [train.py:968] (0/2) Epoch 12, batch 5600, giga_loss[loss=0.2414, simple_loss=0.3155, pruned_loss=0.08367, over 28800.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3415, pruned_loss=0.09896, over 5714205.21 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3555, pruned_loss=0.09868, over 5352549.47 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3404, pruned_loss=0.09888, over 5710676.28 frames. ], batch size: 145, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:06:51,191 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:07:03,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-06 03:07:21,098 INFO [train.py:968] (0/2) Epoch 12, batch 5650, giga_loss[loss=0.2369, simple_loss=0.3162, pruned_loss=0.07876, over 28770.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3375, pruned_loss=0.09725, over 5716016.80 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3551, pruned_loss=0.09855, over 5359589.86 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3368, pruned_loss=0.09728, over 5711722.68 frames. ], batch size: 284, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:07:21,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.002e+02 1.156e+03 1.369e+03 1.844e+03 4.744e+03, threshold=2.737e+03, percent-clipped=7.0
+2023-03-06 03:07:29,970 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:07:47,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=506875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:07:50,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=506878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:08:00,694 INFO [train.py:968] (0/2) Epoch 12, batch 5700, giga_loss[loss=0.2389, simple_loss=0.3159, pruned_loss=0.08092, over 28931.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3347, pruned_loss=0.09586, over 5714324.54 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.355, pruned_loss=0.09854, over 5373055.25 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3337, pruned_loss=0.09584, over 5706554.27 frames. ], batch size: 227, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:08:14,755 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:08:41,712 INFO [train.py:968] (0/2) Epoch 12, batch 5750, giga_loss[loss=0.2765, simple_loss=0.3335, pruned_loss=0.1098, over 29018.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3353, pruned_loss=0.09635, over 5717233.76 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3549, pruned_loss=0.09851, over 5378156.09 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3345, pruned_loss=0.09633, over 5709705.84 frames. ], batch size: 128, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:08:42,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.143e+02 1.144e+03 1.500e+03 2.295e+03 5.454e+03, threshold=3.001e+03, percent-clipped=11.0
+2023-03-06 03:09:05,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-06 03:09:10,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=506978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:09:19,769 INFO [train.py:968] (0/2) Epoch 12, batch 5800, giga_loss[loss=0.2601, simple_loss=0.3391, pruned_loss=0.09052, over 28960.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3398, pruned_loss=0.09864, over 5719198.82 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.355, pruned_loss=0.09855, over 5394034.56 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3386, pruned_loss=0.09859, over 5707727.33 frames. ], batch size: 145, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:09:52,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6898, 1.8274, 1.5449, 1.5676], device='cuda:0'), covar=tensor([0.1360, 0.2165, 0.1915, 0.2026], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0728, 0.0671, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 03:09:57,592 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:09:59,167 INFO [train.py:968] (0/2) Epoch 12, batch 5850, giga_loss[loss=0.2565, simple_loss=0.3413, pruned_loss=0.08585, over 28785.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3433, pruned_loss=0.09995, over 5718843.43 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.355, pruned_loss=0.09852, over 5406243.06 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.342, pruned_loss=0.09997, over 5705787.06 frames. ], batch size: 243, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:10:00,650 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.467e+02 1.207e+03 1.609e+03 2.144e+03 6.231e+03, threshold=3.219e+03, percent-clipped=14.0
+2023-03-06 03:10:12,915 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:10:18,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3026, 5.1172, 4.8127, 2.4610], device='cuda:0'), covar=tensor([0.0379, 0.0526, 0.0583, 0.1806], device='cuda:0'), in_proj_covar=tensor([0.1053, 0.0981, 0.0858, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 03:10:42,827 INFO [train.py:968] (0/2) Epoch 12, batch 5900, giga_loss[loss=0.2815, simple_loss=0.3617, pruned_loss=0.1007, over 28627.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3461, pruned_loss=0.1004, over 5720917.11 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3553, pruned_loss=0.09883, over 5413397.61 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3447, pruned_loss=0.1002, over 5709741.80 frames. ], batch size: 307, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:11:05,068 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-06 03:11:06,710 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=507121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:11:09,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=507124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:11:23,961 INFO [train.py:968] (0/2) Epoch 12, batch 5950, giga_loss[loss=0.2483, simple_loss=0.3243, pruned_loss=0.0862, over 28588.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3486, pruned_loss=0.1014, over 5723157.90 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3553, pruned_loss=0.09903, over 5435163.26 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3471, pruned_loss=0.1011, over 5706600.61 frames. ], batch size: 85, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:11:24,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.290e+02 1.118e+03 1.416e+03 2.131e+03 5.260e+03, threshold=2.833e+03, percent-clipped=6.0
+2023-03-06 03:11:35,197 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=507153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:12:00,845 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=507182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:12:03,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=507185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:12:11,491 INFO [train.py:968] (0/2) Epoch 12, batch 6000, giga_loss[loss=0.3369, simple_loss=0.387, pruned_loss=0.1434, over 28609.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3523, pruned_loss=0.1044, over 5711551.35 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3551, pruned_loss=0.09901, over 5439724.50 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3512, pruned_loss=0.1043, over 5697004.09 frames. ], batch size: 85, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:12:11,495 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 03:12:21,240 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2226, simple_loss=0.3276, pruned_loss=0.05882, over 944034.00 frames. 
+2023-03-06 03:12:21,241 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 03:12:41,462 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=507214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:12:53,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:13:08,578 INFO [train.py:968] (0/2) Epoch 12, batch 6050, giga_loss[loss=0.2814, simple_loss=0.3499, pruned_loss=0.1065, over 28992.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3581, pruned_loss=0.1093, over 5710998.59 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3551, pruned_loss=0.09894, over 5446649.76 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3573, pruned_loss=0.1094, over 5696983.95 frames. ], batch size: 106, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:13:10,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.960e+02 1.273e+03 1.673e+03 2.296e+03 4.224e+03, threshold=3.347e+03, percent-clipped=10.0
+2023-03-06 03:13:27,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7644, 1.8756, 1.5942, 2.1780], device='cuda:0'), covar=tensor([0.2197, 0.2379, 0.2450, 0.2110], device='cuda:0'), in_proj_covar=tensor([0.1301, 0.0968, 0.1152, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 03:13:40,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:13:59,860 INFO [train.py:968] (0/2) Epoch 12, batch 6100, giga_loss[loss=0.334, simple_loss=0.4028, pruned_loss=0.1326, over 28989.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3647, pruned_loss=0.1142, over 5707706.00 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3552, pruned_loss=0.09902, over 5454046.58 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.364, pruned_loss=0.1144, over 5694515.70 frames. ], batch size: 164, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:14:47,787 INFO [train.py:968] (0/2) Epoch 12, batch 6150, giga_loss[loss=0.3259, simple_loss=0.3914, pruned_loss=0.1303, over 28807.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3717, pruned_loss=0.1193, over 5700523.80 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3557, pruned_loss=0.09918, over 5460753.00 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3709, pruned_loss=0.1196, over 5687453.45 frames. ], batch size: 199, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:14:49,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.565e+03 1.981e+03 2.444e+03 7.348e+03, threshold=3.961e+03, percent-clipped=7.0
+2023-03-06 03:15:14,959 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=507370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:15:18,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=507373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:15:37,569 INFO [train.py:968] (0/2) Epoch 12, batch 6200, giga_loss[loss=0.4574, simple_loss=0.4629, pruned_loss=0.226, over 26560.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.379, pruned_loss=0.126, over 5705071.32 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3561, pruned_loss=0.09961, over 5471590.01 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3784, pruned_loss=0.1264, over 5690415.36 frames. ], batch size: 555, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:15:45,496 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=507402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:16:04,154 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7953, 2.6330, 1.6976, 0.9279], device='cuda:0'), covar=tensor([0.5418, 0.2753, 0.2926, 0.4944], device='cuda:0'), in_proj_covar=tensor([0.1576, 0.1495, 0.1502, 0.1298], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 03:16:13,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:16:22,493 INFO [train.py:968] (0/2) Epoch 12, batch 6250, giga_loss[loss=0.3329, simple_loss=0.3879, pruned_loss=0.1389, over 28578.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3844, pruned_loss=0.1301, over 5695588.80 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3567, pruned_loss=0.09977, over 5478607.14 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3841, pruned_loss=0.1311, over 5683431.29 frames. ], batch size: 307, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:16:25,721 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.669e+03 2.103e+03 3.031e+03 6.491e+03, threshold=4.206e+03, percent-clipped=15.0
+2023-03-06 03:17:17,469 INFO [train.py:968] (0/2) Epoch 12, batch 6300, giga_loss[loss=0.3805, simple_loss=0.4259, pruned_loss=0.1675, over 28708.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3884, pruned_loss=0.1344, over 5681848.42 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3565, pruned_loss=0.09971, over 5480554.10 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3886, pruned_loss=0.1354, over 5671485.39 frames. ], batch size: 242, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:17:53,797 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:18:07,942 INFO [train.py:968] (0/2) Epoch 12, batch 6350, libri_loss[loss=0.2231, simple_loss=0.3028, pruned_loss=0.07171, over 29665.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3906, pruned_loss=0.1375, over 5670228.41 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3562, pruned_loss=0.09939, over 5487087.52 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3919, pruned_loss=0.1396, over 5661047.90 frames. ], batch size: 69, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:18:08,874 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:18:09,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.614e+03 2.119e+03 2.719e+03 5.439e+03, threshold=4.238e+03, percent-clipped=10.0
+2023-03-06 03:18:44,633 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=507576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:18:47,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=507579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:19:02,180 INFO [train.py:968] (0/2) Epoch 12, batch 6400, giga_loss[loss=0.3104, simple_loss=0.3724, pruned_loss=0.1242, over 28975.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3943, pruned_loss=0.1419, over 5655632.17 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3563, pruned_loss=0.09947, over 5482271.58 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3956, pruned_loss=0.1438, over 5654866.67 frames. ], batch size: 128, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:19:20,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=507608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:19:57,636 INFO [train.py:968] (0/2) Epoch 12, batch 6450, giga_loss[loss=0.3668, simple_loss=0.4161, pruned_loss=0.1588, over 28569.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.3975, pruned_loss=0.1454, over 5646088.26 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3561, pruned_loss=0.09947, over 5487914.27 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.3993, pruned_loss=0.1476, over 5642487.07 frames. ], batch size: 336, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:19:59,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.609e+03 2.108e+03 3.059e+03 6.645e+03, threshold=4.216e+03, percent-clipped=7.0
+2023-03-06 03:20:04,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:20:18,654 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-06 03:20:34,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507678.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 03:20:48,752 INFO [train.py:968] (0/2) Epoch 12, batch 6500, giga_loss[loss=0.3712, simple_loss=0.4114, pruned_loss=0.1655, over 27972.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.397, pruned_loss=0.1453, over 5645776.44 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3563, pruned_loss=0.09959, over 5492344.90 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.3985, pruned_loss=0.1474, over 5640328.67 frames. ], batch size: 412, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:21:40,300 INFO [train.py:968] (0/2) Epoch 12, batch 6550, giga_loss[loss=0.3748, simple_loss=0.4213, pruned_loss=0.1642, over 28561.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.3971, pruned_loss=0.1466, over 5644054.42 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3564, pruned_loss=0.09964, over 5499828.45 frames. ], giga_tot_loss[loss=0.3486, simple_loss=0.3991, pruned_loss=0.149, over 5635546.75 frames. ], batch size: 307, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:21:42,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.723e+02 1.859e+03 2.619e+03 3.326e+03 1.017e+04, threshold=5.238e+03, percent-clipped=16.0
+2023-03-06 03:21:42,804 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 03:21:55,981 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.94 vs. limit=2.0
+2023-03-06 03:22:27,714 INFO [train.py:968] (0/2) Epoch 12, batch 6600, giga_loss[loss=0.3602, simple_loss=0.4131, pruned_loss=0.1536, over 28500.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.394, pruned_loss=0.1438, over 5641567.26 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3558, pruned_loss=0.09937, over 5509566.81 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.3973, pruned_loss=0.1475, over 5630578.50 frames. ], batch size: 78, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:22:28,003 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=507792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:22:31,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=507795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:23:00,218 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=507824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:23:02,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7185, 1.8883, 1.4952, 2.2290], device='cuda:0'), covar=tensor([0.2254, 0.2342, 0.2606, 0.2137], device='cuda:0'), in_proj_covar=tensor([0.1297, 0.0964, 0.1153, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 03:23:07,713 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7445, 4.4254, 1.8192, 1.8338], device='cuda:0'), covar=tensor([0.0882, 0.0245, 0.0822, 0.1234], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0514, 0.0343, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 03:23:17,618 INFO [train.py:968] (0/2) Epoch 12, batch 6650, giga_loss[loss=0.3351, simple_loss=0.3976, pruned_loss=0.1363, over 28980.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3935, pruned_loss=0.1419, over 5650194.04 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3555, pruned_loss=0.09917, over 5515317.40 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3969, pruned_loss=0.1456, over 5638113.59 frames. ], batch size: 227, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:23:19,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.880e+02 1.708e+03 2.415e+03 3.503e+03 1.178e+04, threshold=4.830e+03, percent-clipped=11.0
+2023-03-06 03:24:08,905 INFO [train.py:968] (0/2) Epoch 12, batch 6700, giga_loss[loss=0.3251, simple_loss=0.391, pruned_loss=0.1296, over 28881.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3952, pruned_loss=0.143, over 5643120.35 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3554, pruned_loss=0.09909, over 5521341.60 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3986, pruned_loss=0.1466, over 5630107.86 frames. ], batch size: 145, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:24:19,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:24:33,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:24:50,873 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:24:57,277 INFO [train.py:968] (0/2) Epoch 12, batch 6750, giga_loss[loss=0.2698, simple_loss=0.3518, pruned_loss=0.09395, over 28980.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3925, pruned_loss=0.1408, over 5639066.76 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3552, pruned_loss=0.09905, over 5531577.37 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3965, pruned_loss=0.145, over 5622491.44 frames. ], batch size: 164, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:24:58,521 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 03:24:59,379 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.290e+02 1.527e+03 1.991e+03 2.464e+03 6.054e+03, threshold=3.981e+03, percent-clipped=3.0
+2023-03-06 03:25:04,425 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5661, 3.6569, 1.7416, 1.6906], device='cuda:0'), covar=tensor([0.0865, 0.0228, 0.0871, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0512, 0.0344, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 03:25:27,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9511, 1.1966, 1.2334, 1.0757], device='cuda:0'), covar=tensor([0.1318, 0.1128, 0.1762, 0.1274], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0733, 0.0675, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 03:25:48,972 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:25:50,587 INFO [train.py:968] (0/2) Epoch 12, batch 6800, libri_loss[loss=0.3436, simple_loss=0.4096, pruned_loss=0.1388, over 29677.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3912, pruned_loss=0.1382, over 5648277.92 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3553, pruned_loss=0.09915, over 5541320.77 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3951, pruned_loss=0.1423, over 5628417.80 frames. ], batch size: 91, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:25:56,919 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-508000.pt
+2023-03-06 03:26:11,361 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=508014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:26:13,665 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=508017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:26:38,076 INFO [train.py:968] (0/2) Epoch 12, batch 6850, giga_loss[loss=0.3474, simple_loss=0.3946, pruned_loss=0.1501, over 26699.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3898, pruned_loss=0.1364, over 5655332.21 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3552, pruned_loss=0.09922, over 5547109.05 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3934, pruned_loss=0.1402, over 5635821.52 frames. ], batch size: 555, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:26:42,064 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:26:42,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.624e+03 2.406e+03 3.358e+03 9.127e+03, threshold=4.813e+03, percent-clipped=13.0
+2023-03-06 03:26:44,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:26:50,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508053.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 03:26:58,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:27:00,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:27:03,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8173, 1.0312, 1.0701, 0.7664], device='cuda:0'), covar=tensor([0.1776, 0.1834, 0.1047, 0.1625], device='cuda:0'), in_proj_covar=tensor([0.1732, 0.1646, 0.1608, 0.1701], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 03:27:12,449 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-06 03:27:13,942 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:27:28,070 INFO [train.py:968] (0/2) Epoch 12, batch 6900, giga_loss[loss=0.3191, simple_loss=0.3856, pruned_loss=0.1262, over 28928.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3859, pruned_loss=0.133, over 5662142.08 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3552, pruned_loss=0.0992, over 5551573.18 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3893, pruned_loss=0.1365, over 5644278.00 frames. ], batch size: 145, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:27:29,046 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:28:08,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8544, 1.8668, 1.7988, 1.6218], device='cuda:0'), covar=tensor([0.1412, 0.1990, 0.1971, 0.1937], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0730, 0.0672, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 03:28:14,913 INFO [train.py:968] (0/2) Epoch 12, batch 6950, libri_loss[loss=0.2459, simple_loss=0.3279, pruned_loss=0.08199, over 29521.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3824, pruned_loss=0.1303, over 5654932.05 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.355, pruned_loss=0.09897, over 5556119.60 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3864, pruned_loss=0.1345, over 5639312.64 frames. ], batch size: 80, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:28:18,208 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.287e+02 1.441e+03 1.815e+03 2.594e+03 7.490e+03, threshold=3.630e+03, percent-clipped=3.0
+2023-03-06 03:28:37,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3730, 3.5711, 1.5796, 1.4942], device='cuda:0'), covar=tensor([0.0979, 0.0316, 0.0841, 0.1314], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0513, 0.0344, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 03:28:59,195 INFO [train.py:968] (0/2) Epoch 12, batch 7000, giga_loss[loss=0.2816, simple_loss=0.3556, pruned_loss=0.1039, over 28885.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3813, pruned_loss=0.1296, over 5661297.65 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3547, pruned_loss=0.09903, over 5566647.88 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3855, pruned_loss=0.1338, over 5642122.91 frames. ], batch size: 145, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:29:04,721 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508196.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 03:29:06,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508199.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 03:29:42,524 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508228.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 03:29:56,525 INFO [train.py:968] (0/2) Epoch 12, batch 7050, libri_loss[loss=0.2856, simple_loss=0.3648, pruned_loss=0.1032, over 27955.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3809, pruned_loss=0.1291, over 5658576.39 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3547, pruned_loss=0.09901, over 5571935.55 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3848, pruned_loss=0.133, over 5640195.10 frames. ], batch size: 116, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:30:01,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.525e+03 1.808e+03 2.512e+03 8.469e+03, threshold=3.617e+03, percent-clipped=10.0
+2023-03-06 03:30:45,673 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-06 03:30:47,683 INFO [train.py:968] (0/2) Epoch 12, batch 7100, giga_loss[loss=0.2845, simple_loss=0.3554, pruned_loss=0.1068, over 28802.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3787, pruned_loss=0.1267, over 5660760.61 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3548, pruned_loss=0.09897, over 5574348.08 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3822, pruned_loss=0.1305, over 5645470.99 frames. ], batch size: 284, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:30:59,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9009, 5.6898, 5.3556, 2.8877], device='cuda:0'), covar=tensor([0.0397, 0.0603, 0.0616, 0.1597], device='cuda:0'), in_proj_covar=tensor([0.1070, 0.1004, 0.0875, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 03:31:04,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:31:42,222 INFO [train.py:968] (0/2) Epoch 12, batch 7150, giga_loss[loss=0.3047, simple_loss=0.3738, pruned_loss=0.1178, over 28808.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3776, pruned_loss=0.1235, over 5659163.54 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3548, pruned_loss=0.09913, over 5575838.88 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3812, pruned_loss=0.1271, over 5647763.61 frames. ], batch size: 99, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:31:47,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.854e+02 1.473e+03 1.827e+03 2.513e+03 6.900e+03, threshold=3.655e+03, percent-clipped=8.0
+2023-03-06 03:31:53,970 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=508351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:32:04,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:32:27,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:32:29,328 INFO [train.py:968] (0/2) Epoch 12, batch 7200, giga_loss[loss=0.3215, simple_loss=0.3879, pruned_loss=0.1275, over 28010.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.38, pruned_loss=0.1235, over 5671401.83 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3547, pruned_loss=0.09913, over 5581662.99 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3836, pruned_loss=0.1271, over 5659679.31 frames. ], batch size: 412, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:32:29,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:33:24,636 INFO [train.py:968] (0/2) Epoch 12, batch 7250, giga_loss[loss=0.3505, simple_loss=0.4052, pruned_loss=0.1479, over 28265.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3818, pruned_loss=0.1256, over 5664343.95 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3547, pruned_loss=0.09905, over 5587396.05 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3851, pruned_loss=0.129, over 5651479.88 frames. ], batch size: 368, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:33:29,050 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.041e+02 1.529e+03 2.014e+03 2.661e+03 5.403e+03, threshold=4.029e+03, percent-clipped=11.0
+2023-03-06 03:33:33,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:33:37,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:33:48,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9749, 5.7576, 5.4494, 2.9375], device='cuda:0'), covar=tensor([0.0589, 0.0825, 0.0974, 0.1790], device='cuda:0'), in_proj_covar=tensor([0.1080, 0.1010, 0.0880, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 03:34:04,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:11,701 INFO [train.py:968] (0/2) Epoch 12, batch 7300, giga_loss[loss=0.3142, simple_loss=0.3808, pruned_loss=0.1238, over 28878.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3813, pruned_loss=0.1259, over 5671570.96 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3546, pruned_loss=0.09898, over 5592020.96 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3844, pruned_loss=0.1291, over 5658124.97 frames. ], batch size: 186, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:34:25,848 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:28,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:54,537 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:56,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:56,768 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:59,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:35:00,665 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:35:02,418 INFO [train.py:968] (0/2) Epoch 12, batch 7350, libri_loss[loss=0.343, simple_loss=0.4138, pruned_loss=0.1361, over 18467.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3803, pruned_loss=0.1265, over 5669338.80 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3548, pruned_loss=0.09905, over 5588821.57 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3832, pruned_loss=0.1295, over 5664235.98 frames. ], batch size: 188, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:35:06,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.540e+02 1.605e+03 2.266e+03 3.583e+03 1.154e+04, threshold=4.532e+03, percent-clipped=16.0
+2023-03-06 03:35:20,989 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:35:24,617 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:35:46,364 INFO [train.py:968] (0/2) Epoch 12, batch 7400, giga_loss[loss=0.2904, simple_loss=0.3551, pruned_loss=0.1128, over 28841.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3789, pruned_loss=0.1266, over 5666593.86 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3546, pruned_loss=0.09889, over 5594326.77 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3821, pruned_loss=0.13, over 5659559.81 frames. ], batch size: 186, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:35:50,974 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-06 03:36:38,442 INFO [train.py:968] (0/2) Epoch 12, batch 7450, giga_loss[loss=0.2543, simple_loss=0.3392, pruned_loss=0.08471, over 28489.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3776, pruned_loss=0.1247, over 5666938.67 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3547, pruned_loss=0.09898, over 5600265.62 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3806, pruned_loss=0.128, over 5657450.06 frames. ], batch size: 71, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:36:40,400 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.53 vs. limit=5.0
+2023-03-06 03:36:41,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.807e+02 1.366e+03 1.795e+03 2.435e+03 1.005e+04, threshold=3.590e+03, percent-clipped=2.0
+2023-03-06 03:36:50,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3368, 1.6999, 1.2904, 1.5661], device='cuda:0'), covar=tensor([0.0769, 0.0307, 0.0327, 0.0812], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 03:37:26,868 INFO [train.py:968] (0/2) Epoch 12, batch 7500, giga_loss[loss=0.3081, simple_loss=0.3826, pruned_loss=0.1168, over 28997.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3774, pruned_loss=0.1239, over 5664325.52 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3547, pruned_loss=0.09889, over 5604509.16 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3802, pruned_loss=0.1269, over 5653847.63 frames. ], batch size: 128, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:37:51,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=508717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:37:59,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:38:13,076 INFO [train.py:968] (0/2) Epoch 12, batch 7550, giga_loss[loss=0.295, simple_loss=0.3629, pruned_loss=0.1136, over 28863.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3779, pruned_loss=0.1241, over 5665167.58 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3545, pruned_loss=0.09883, over 5599780.30 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3806, pruned_loss=0.127, over 5663110.36 frames. ], batch size: 199, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:38:16,757 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.068e+03 1.576e+03 1.934e+03 2.350e+03 5.210e+03, threshold=3.868e+03, percent-clipped=7.0
+2023-03-06 03:38:56,925 INFO [train.py:968] (0/2) Epoch 12, batch 7600, giga_loss[loss=0.2595, simple_loss=0.3372, pruned_loss=0.09091, over 29038.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3739, pruned_loss=0.1205, over 5686034.99 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.354, pruned_loss=0.09854, over 5609641.92 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3773, pruned_loss=0.124, over 5678009.36 frames. ], batch size: 136, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:39:38,137 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-06 03:39:43,494 INFO [train.py:968] (0/2) Epoch 12, batch 7650, giga_loss[loss=0.3016, simple_loss=0.3564, pruned_loss=0.1234, over 28645.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3714, pruned_loss=0.1193, over 5678750.33 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.354, pruned_loss=0.09849, over 5614279.52 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3746, pruned_loss=0.1226, over 5669914.77 frames. ], batch size: 85, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:39:52,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.301e+02 1.574e+03 2.235e+03 2.761e+03 8.303e+03, threshold=4.470e+03, percent-clipped=11.0
+2023-03-06 03:40:01,974 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=508858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:40:14,327 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:40:16,424 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:40:34,069 INFO [train.py:968] (0/2) Epoch 12, batch 7700, giga_loss[loss=0.313, simple_loss=0.3689, pruned_loss=0.1286, over 28850.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3714, pruned_loss=0.1206, over 5671418.32 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3545, pruned_loss=0.09869, over 5618972.23 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3738, pruned_loss=0.1235, over 5660961.85 frames. ], batch size: 227, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:40:43,989 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:41:25,358 INFO [train.py:968] (0/2) Epoch 12, batch 7750, giga_loss[loss=0.3588, simple_loss=0.4052, pruned_loss=0.1563, over 27471.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3718, pruned_loss=0.1223, over 5664497.80 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3541, pruned_loss=0.09835, over 5623301.87 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3743, pruned_loss=0.1252, over 5652859.75 frames. ], batch size: 472, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:41:30,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.681e+02 1.604e+03 2.031e+03 3.023e+03 6.155e+03, threshold=4.063e+03, percent-clipped=5.0
+2023-03-06 03:41:55,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9727, 2.4711, 1.0673, 1.1988], device='cuda:0'), covar=tensor([0.1242, 0.0508, 0.1087, 0.1614], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0518, 0.0345, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 03:42:14,526 INFO [train.py:968] (0/2) Epoch 12, batch 7800, giga_loss[loss=0.3299, simple_loss=0.3921, pruned_loss=0.1339, over 28924.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3713, pruned_loss=0.1225, over 5661854.82 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3546, pruned_loss=0.0986, over 5629368.59 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3733, pruned_loss=0.1252, over 5647996.07 frames. ], batch size: 227, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:42:27,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1109, 1.3139, 3.7489, 3.0525], device='cuda:0'), covar=tensor([0.1698, 0.2497, 0.0475, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0670, 0.0595, 0.0873, 0.0772], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 03:42:31,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1089, 1.2144, 3.6664, 3.1124], device='cuda:0'), covar=tensor([0.1980, 0.2764, 0.0792, 0.1084], device='cuda:0'), in_proj_covar=tensor([0.0670, 0.0595, 0.0874, 0.0773], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 03:42:54,264 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=509035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:43:03,044 INFO [train.py:968] (0/2) Epoch 12, batch 7850, giga_loss[loss=0.3438, simple_loss=0.3969, pruned_loss=0.1453, over 28841.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.371, pruned_loss=0.1227, over 5658229.22 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3549, pruned_loss=0.0989, over 5628532.08 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3724, pruned_loss=0.1248, over 5648438.63 frames. ], batch size: 174, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:43:08,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.611e+02 1.757e+03 2.228e+03 3.127e+03 9.372e+03, threshold=4.456e+03, percent-clipped=16.0
+2023-03-06 03:43:52,393 INFO [train.py:968] (0/2) Epoch 12, batch 7900, giga_loss[loss=0.319, simple_loss=0.3795, pruned_loss=0.1292, over 28678.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3728, pruned_loss=0.1238, over 5658902.41 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3551, pruned_loss=0.09895, over 5631198.19 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.374, pruned_loss=0.1257, over 5649351.27 frames. ], batch size: 262, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:43:52,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=509092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:44:36,368 INFO [train.py:968] (0/2) Epoch 12, batch 7950, libri_loss[loss=0.2416, simple_loss=0.3219, pruned_loss=0.08066, over 29468.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.372, pruned_loss=0.1215, over 5675040.71 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3551, pruned_loss=0.09894, over 5642490.10 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3736, pruned_loss=0.124, over 5658586.08 frames. ], batch size: 70, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:44:36,791 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 03:44:42,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.444e+02 1.673e+03 2.297e+03 3.191e+03 1.020e+04, threshold=4.594e+03, percent-clipped=14.0
+2023-03-06 03:45:19,120 INFO [train.py:968] (0/2) Epoch 12, batch 8000, giga_loss[loss=0.3031, simple_loss=0.3821, pruned_loss=0.112, over 28900.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3716, pruned_loss=0.1201, over 5691457.07 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3553, pruned_loss=0.09904, over 5651730.73 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3733, pruned_loss=0.1228, over 5670628.19 frames. ], batch size: 186, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:45:35,387 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-06 03:45:59,568 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=509233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:46:01,343 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=509235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:46:04,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=509238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:46:06,559 INFO [train.py:968] (0/2) Epoch 12, batch 8050, giga_loss[loss=0.3025, simple_loss=0.3701, pruned_loss=0.1174, over 28888.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.373, pruned_loss=0.1214, over 5691202.70 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3555, pruned_loss=0.09928, over 5655317.48 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3744, pruned_loss=0.1236, over 5672098.24 frames. ], batch size: 186, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:46:14,233 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.390e+02 1.508e+03 1.894e+03 2.527e+03 1.364e+04, threshold=3.788e+03, percent-clipped=6.0
+2023-03-06 03:46:32,446 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=509267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:46:57,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7894, 3.6252, 3.4616, 1.8520], device='cuda:0'), covar=tensor([0.0671, 0.0787, 0.0809, 0.2377], device='cuda:0'), in_proj_covar=tensor([0.1085, 0.1012, 0.0883, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 03:46:59,300 INFO [train.py:968] (0/2) Epoch 12, batch 8100, giga_loss[loss=0.3111, simple_loss=0.3657, pruned_loss=0.1282, over 28465.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3756, pruned_loss=0.1244, over 5679398.94 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3557, pruned_loss=0.09942, over 5659545.93 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3768, pruned_loss=0.1264, over 5660974.19 frames. ], batch size: 85, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:47:12,671 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3350, 1.5824, 1.3547, 1.1902], device='cuda:0'), covar=tensor([0.2173, 0.1708, 0.1303, 0.1705], device='cuda:0'), in_proj_covar=tensor([0.1742, 0.1646, 0.1617, 0.1713], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 03:47:47,132 INFO [train.py:968] (0/2) Epoch 12, batch 8150, giga_loss[loss=0.3628, simple_loss=0.3952, pruned_loss=0.1652, over 26646.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3761, pruned_loss=0.1261, over 5677930.86 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3551, pruned_loss=0.09919, over 5667276.59 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3784, pruned_loss=0.1289, over 5656836.06 frames. ], batch size: 555, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:47:53,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.342e+02 1.681e+03 1.971e+03 2.783e+03 6.672e+03, threshold=3.941e+03, percent-clipped=10.0
+2023-03-06 03:48:19,832 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=509376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:48:22,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=509379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:48:32,600 INFO [train.py:968] (0/2) Epoch 12, batch 8200, giga_loss[loss=0.3465, simple_loss=0.3921, pruned_loss=0.1504, over 28897.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.127, over 5678282.27 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3546, pruned_loss=0.09886, over 5675841.60 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3796, pruned_loss=0.131, over 5653285.03 frames. ], batch size: 227, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:48:48,927 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=509408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:48:52,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=509410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:48:55,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3906, 3.3165, 1.5627, 1.4892], device='cuda:0'), covar=tensor([0.0901, 0.0348, 0.0824, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0519, 0.0344, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 03:49:24,311 INFO [train.py:968] (0/2) Epoch 12, batch 8250, giga_loss[loss=0.4591, simple_loss=0.4742, pruned_loss=0.222, over 24287.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3786, pruned_loss=0.13, over 5676205.29 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3545, pruned_loss=0.09901, over 5681661.39 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3819, pruned_loss=0.1337, over 5651320.19 frames. ], batch size: 705, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:49:31,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.308e+02 1.671e+03 2.269e+03 3.299e+03 1.498e+04, threshold=4.539e+03, percent-clipped=17.0
+2023-03-06 03:49:38,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0624, 1.3404, 1.2426, 1.0950], device='cuda:0'), covar=tensor([0.1681, 0.1458, 0.1017, 0.1370], device='cuda:0'), in_proj_covar=tensor([0.1747, 0.1650, 0.1617, 0.1711], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 03:49:56,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3912, 1.9463, 1.4187, 0.6525], device='cuda:0'), covar=tensor([0.3332, 0.1686, 0.2571, 0.4129], device='cuda:0'), in_proj_covar=tensor([0.1561, 0.1492, 0.1488, 0.1296], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 03:50:11,467 INFO [train.py:968] (0/2) Epoch 12, batch 8300, giga_loss[loss=0.3113, simple_loss=0.3709, pruned_loss=0.1259, over 28730.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3789, pruned_loss=0.1301, over 5674626.65 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3546, pruned_loss=0.09894, over 5685393.98 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3819, pruned_loss=0.1338, over 5651252.64 frames. ], batch size: 85, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:50:11,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4595, 2.1786, 1.5363, 0.6083], device='cuda:0'), covar=tensor([0.3988, 0.1993, 0.3209, 0.4651], device='cuda:0'), in_proj_covar=tensor([0.1560, 0.1490, 0.1485, 0.1293], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 03:50:54,616 INFO [train.py:968] (0/2) Epoch 12, batch 8350, giga_loss[loss=0.284, simple_loss=0.3597, pruned_loss=0.1041, over 29041.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3769, pruned_loss=0.1282, over 5687468.80 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3544, pruned_loss=0.09882, over 5691005.38 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.38, pruned_loss=0.1319, over 5663642.28 frames. ], batch size: 155, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:51:01,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.508e+03 1.983e+03 3.139e+03 8.394e+03, threshold=3.966e+03, percent-clipped=10.0
+2023-03-06 03:51:03,504 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=509553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:51:06,094 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=509556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:51:30,949 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=509585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:51:36,707 INFO [train.py:968] (0/2) Epoch 12, batch 8400, giga_loss[loss=0.2857, simple_loss=0.3674, pruned_loss=0.102, over 28965.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3762, pruned_loss=0.1257, over 5697935.90 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3546, pruned_loss=0.09898, over 5695740.58 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3792, pruned_loss=0.1293, over 5674690.77 frames. ], batch size: 155, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:51:37,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5926, 2.0901, 1.7848, 1.5475], device='cuda:0'), covar=tensor([0.2447, 0.1544, 0.1357, 0.1721], device='cuda:0'), in_proj_covar=tensor([0.1742, 0.1641, 0.1610, 0.1703], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 03:51:38,515 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 03:52:19,851 INFO [train.py:968] (0/2) Epoch 12, batch 8450, giga_loss[loss=0.2947, simple_loss=0.3611, pruned_loss=0.1142, over 28939.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3748, pruned_loss=0.1244, over 5689903.74 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3549, pruned_loss=0.0991, over 5697249.26 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3774, pruned_loss=0.1278, over 5669888.12 frames. ], batch size: 136, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:52:27,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.079e+02 1.636e+03 2.081e+03 2.499e+03 6.617e+03, threshold=4.162e+03, percent-clipped=8.0
+2023-03-06 03:53:04,015 INFO [train.py:968] (0/2) Epoch 12, batch 8500, giga_loss[loss=0.2776, simple_loss=0.3533, pruned_loss=0.101, over 28991.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3726, pruned_loss=0.1229, over 5683968.11 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3557, pruned_loss=0.09953, over 5703474.86 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3749, pruned_loss=0.1265, over 5661826.47 frames. ], batch size: 164, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:53:09,197 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 03:53:09,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5255, 1.6952, 1.7459, 1.3225], device='cuda:0'), covar=tensor([0.1562, 0.2340, 0.1266, 0.1513], device='cuda:0'), in_proj_covar=tensor([0.0832, 0.0698, 0.0872, 0.0775], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 03:53:20,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4996, 1.7250, 1.4045, 1.7657], device='cuda:0'), covar=tensor([0.2158, 0.2050, 0.2130, 0.2018], device='cuda:0'), in_proj_covar=tensor([0.1308, 0.0973, 0.1155, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 03:53:47,729 INFO [train.py:968] (0/2) Epoch 12, batch 8550, giga_loss[loss=0.2892, simple_loss=0.359, pruned_loss=0.1097, over 28904.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3699, pruned_loss=0.1214, over 5693605.87 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3552, pruned_loss=0.09923, over 5708791.93 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3726, pruned_loss=0.1252, over 5670466.28 frames. ], batch size: 174, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:53:57,300 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.064e+02 1.581e+03 2.022e+03 2.884e+03 7.873e+03, threshold=4.043e+03, percent-clipped=13.0
+2023-03-06 03:54:41,365 INFO [train.py:968] (0/2) Epoch 12, batch 8600, giga_loss[loss=0.316, simple_loss=0.3793, pruned_loss=0.1263, over 28769.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3724, pruned_loss=0.1236, over 5689325.55 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3549, pruned_loss=0.09907, over 5712853.40 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3751, pruned_loss=0.1271, over 5667113.80 frames. ], batch size: 119, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:55:27,909 INFO [train.py:968] (0/2) Epoch 12, batch 8650, giga_loss[loss=0.3023, simple_loss=0.3869, pruned_loss=0.1088, over 29007.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3757, pruned_loss=0.1248, over 5679480.35 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3549, pruned_loss=0.09908, over 5705652.05 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3786, pruned_loss=0.1285, over 5667203.57 frames. ], batch size: 136, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:55:34,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.782e+02 1.583e+03 1.964e+03 2.887e+03 6.545e+03, threshold=3.927e+03, percent-clipped=7.0
+2023-03-06 03:56:14,860 INFO [train.py:968] (0/2) Epoch 12, batch 8700, giga_loss[loss=0.2927, simple_loss=0.3715, pruned_loss=0.1069, over 28244.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3781, pruned_loss=0.1238, over 5680028.74 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3548, pruned_loss=0.0989, over 5707862.10 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3811, pruned_loss=0.1275, over 5667391.25 frames. ], batch size: 368, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:56:47,490 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-06 03:57:03,617 INFO [train.py:968] (0/2) Epoch 12, batch 8750, giga_loss[loss=0.452, simple_loss=0.4649, pruned_loss=0.2195, over 26652.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3807, pruned_loss=0.125, over 5685350.66 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3549, pruned_loss=0.09884, over 5710197.35 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3833, pruned_loss=0.1283, over 5672839.24 frames. ], batch size: 555, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:57:11,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.283e+02 1.458e+03 2.013e+03 2.750e+03 1.432e+04, threshold=4.027e+03, percent-clipped=12.0
+2023-03-06 03:57:48,968 INFO [train.py:968] (0/2) Epoch 12, batch 8800, giga_loss[loss=0.3043, simple_loss=0.3709, pruned_loss=0.1189, over 28840.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3826, pruned_loss=0.1272, over 5690476.74 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3544, pruned_loss=0.09865, over 5712837.33 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3856, pruned_loss=0.1305, over 5677683.05 frames. ], batch size: 199, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:57:52,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6550, 1.6983, 1.9342, 1.4394], device='cuda:0'), covar=tensor([0.1650, 0.2215, 0.1269, 0.1536], device='cuda:0'), in_proj_covar=tensor([0.0833, 0.0699, 0.0873, 0.0775], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 03:57:55,342 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-510000.pt
+2023-03-06 03:58:35,040 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=510039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:58:36,869 INFO [train.py:968] (0/2) Epoch 12, batch 8850, libri_loss[loss=0.2526, simple_loss=0.3328, pruned_loss=0.08619, over 29535.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3822, pruned_loss=0.1276, over 5690044.67 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3545, pruned_loss=0.0987, over 5718543.74 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3855, pruned_loss=0.1312, over 5673728.58 frames. ], batch size: 77, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:58:43,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4031, 1.7247, 1.3391, 1.6033], device='cuda:0'), covar=tensor([0.2647, 0.2581, 0.2892, 0.2389], device='cuda:0'), in_proj_covar=tensor([0.1308, 0.0972, 0.1160, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 03:58:44,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.012e+02 1.519e+03 1.998e+03 2.929e+03 1.008e+04, threshold=3.996e+03, percent-clipped=14.0
+2023-03-06 03:59:22,762 INFO [train.py:968] (0/2) Epoch 12, batch 8900, giga_loss[loss=0.3626, simple_loss=0.413, pruned_loss=0.1561, over 28944.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3822, pruned_loss=0.1283, over 5694113.94 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3545, pruned_loss=0.09852, over 5721829.92 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3855, pruned_loss=0.132, over 5677708.47 frames. ], batch size: 174, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:00:04,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3461, 1.1895, 1.1327, 1.4814], device='cuda:0'), covar=tensor([0.0768, 0.0355, 0.0332, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0059, 0.0053, 0.0089], device='cuda:0')
+2023-03-06 04:00:14,638 INFO [train.py:968] (0/2) Epoch 12, batch 8950, giga_loss[loss=0.2344, simple_loss=0.3086, pruned_loss=0.08008, over 28539.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3804, pruned_loss=0.128, over 5692469.75 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3543, pruned_loss=0.09842, over 5724658.45 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3835, pruned_loss=0.1314, over 5676761.25 frames. ], batch size: 71, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:00:21,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.091e+02 1.565e+03 1.927e+03 2.472e+03 5.505e+03, threshold=3.854e+03, percent-clipped=3.0
+2023-03-06 04:00:49,627 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-06 04:00:57,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3638, 2.0399, 1.6021, 0.5329], device='cuda:0'), covar=tensor([0.3758, 0.1977, 0.3163, 0.4560], device='cuda:0'), in_proj_covar=tensor([0.1560, 0.1494, 0.1481, 0.1293], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 04:00:59,273 INFO [train.py:968] (0/2) Epoch 12, batch 9000, giga_loss[loss=0.2804, simple_loss=0.3444, pruned_loss=0.1082, over 28601.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3787, pruned_loss=0.1275, over 5676397.70 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3549, pruned_loss=0.09883, over 5714585.97 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3816, pruned_loss=0.131, over 5670540.83 frames. ], batch size: 71, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:00:59,277 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 04:01:07,815 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2162, simple_loss=0.3223, pruned_loss=0.05508, over 944034.00 frames. 
+2023-03-06 04:01:07,816 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 04:01:08,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9905, 1.3430, 1.1038, 0.2145], device='cuda:0'), covar=tensor([0.2755, 0.2141, 0.3396, 0.4511], device='cuda:0'), in_proj_covar=tensor([0.1560, 0.1495, 0.1481, 0.1294], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 04:01:10,262 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-06 04:01:12,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=510199.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 04:01:54,083 INFO [train.py:968] (0/2) Epoch 12, batch 9050, giga_loss[loss=0.3361, simple_loss=0.3917, pruned_loss=0.1403, over 27988.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3769, pruned_loss=0.1265, over 5670066.22 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.355, pruned_loss=0.09895, over 5709467.04 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3797, pruned_loss=0.1298, over 5668795.44 frames. ], batch size: 412, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:02:04,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.161e+02 1.632e+03 2.192e+03 2.953e+03 9.909e+03, threshold=4.385e+03, percent-clipped=15.0
+2023-03-06 04:02:22,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7979, 2.0456, 1.9577, 1.6301], device='cuda:0'), covar=tensor([0.1328, 0.1536, 0.1591, 0.1735], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0736, 0.0675, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 04:02:33,106 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-06 04:02:42,901 INFO [train.py:968] (0/2) Epoch 12, batch 9100, giga_loss[loss=0.3416, simple_loss=0.394, pruned_loss=0.1446, over 28799.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3766, pruned_loss=0.1262, over 5671597.33 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3547, pruned_loss=0.09864, over 5710807.34 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3797, pruned_loss=0.1299, over 5668514.26 frames. ], batch size: 284, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:03:32,530 INFO [train.py:968] (0/2) Epoch 12, batch 9150, giga_loss[loss=0.3456, simple_loss=0.395, pruned_loss=0.1481, over 28649.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3757, pruned_loss=0.1264, over 5674930.39 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3545, pruned_loss=0.09854, over 5711901.30 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3783, pruned_loss=0.1295, over 5671442.57 frames. ], batch size: 307, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:03:42,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.825e+02 1.581e+03 2.192e+03 3.283e+03 8.283e+03, threshold=4.385e+03, percent-clipped=9.0
+2023-03-06 04:04:08,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5959, 2.0727, 1.6906, 1.5255], device='cuda:0'), covar=tensor([0.2090, 0.1694, 0.1918, 0.1825], device='cuda:0'), in_proj_covar=tensor([0.1738, 0.1646, 0.1615, 0.1706], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 04:04:23,228 INFO [train.py:968] (0/2) Epoch 12, batch 9200, giga_loss[loss=0.2917, simple_loss=0.3609, pruned_loss=0.1113, over 28823.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.375, pruned_loss=0.1264, over 5677021.22 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3547, pruned_loss=0.09857, over 5714105.70 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3773, pruned_loss=0.1293, over 5671710.89 frames. ], batch size: 285, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:04:40,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=510414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:05:05,917 INFO [train.py:968] (0/2) Epoch 12, batch 9250, libri_loss[loss=0.2255, simple_loss=0.3014, pruned_loss=0.07477, over 29647.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3755, pruned_loss=0.1262, over 5688437.47 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3551, pruned_loss=0.09883, over 5717038.73 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3776, pruned_loss=0.1289, over 5680809.37 frames. ], batch size: 69, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:05:17,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.482e+02 1.614e+03 2.059e+03 2.855e+03 8.996e+03, threshold=4.118e+03, percent-clipped=8.0
+2023-03-06 04:05:18,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4268, 4.2750, 4.0113, 1.8888], device='cuda:0'), covar=tensor([0.0508, 0.0645, 0.0688, 0.2104], device='cuda:0'), in_proj_covar=tensor([0.1086, 0.1015, 0.0887, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 04:05:58,194 INFO [train.py:968] (0/2) Epoch 12, batch 9300, giga_loss[loss=0.3924, simple_loss=0.4317, pruned_loss=0.1766, over 27578.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3776, pruned_loss=0.1273, over 5680289.21 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.355, pruned_loss=0.09886, over 5720045.07 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3796, pruned_loss=0.13, over 5670890.64 frames. ], batch size: 472, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:06:42,987 INFO [train.py:968] (0/2) Epoch 12, batch 9350, giga_loss[loss=0.3354, simple_loss=0.3848, pruned_loss=0.143, over 28982.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3788, pruned_loss=0.1284, over 5677874.01 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3552, pruned_loss=0.09905, over 5722986.30 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3812, pruned_loss=0.1315, over 5666102.46 frames. ], batch size: 136, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:06:51,867 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.572e+03 2.103e+03 3.274e+03 9.620e+03, threshold=4.207e+03, percent-clipped=10.0
+2023-03-06 04:06:56,492 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=510557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:06:58,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=510560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:07:14,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=510574.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 04:07:29,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=510589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:07:31,085 INFO [train.py:968] (0/2) Epoch 12, batch 9400, libri_loss[loss=0.2271, simple_loss=0.3063, pruned_loss=0.07395, over 29663.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3796, pruned_loss=0.1288, over 5680986.76 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3549, pruned_loss=0.09889, over 5725840.86 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3821, pruned_loss=0.1318, over 5668559.29 frames. ], batch size: 73, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:08:16,927 INFO [train.py:968] (0/2) Epoch 12, batch 9450, giga_loss[loss=0.3374, simple_loss=0.4014, pruned_loss=0.1367, over 28700.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3794, pruned_loss=0.1258, over 5682579.35 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3552, pruned_loss=0.09906, over 5727793.03 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3819, pruned_loss=0.1288, over 5670000.15 frames. ], batch size: 262, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:08:26,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.524e+02 1.328e+03 1.628e+03 2.206e+03 1.067e+04, threshold=3.256e+03, percent-clipped=1.0
+2023-03-06 04:08:52,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9950, 3.8045, 3.6222, 1.9709], device='cuda:0'), covar=tensor([0.0640, 0.0809, 0.0826, 0.2107], device='cuda:0'), in_proj_covar=tensor([0.1081, 0.1011, 0.0884, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 04:08:58,365 INFO [train.py:968] (0/2) Epoch 12, batch 9500, giga_loss[loss=0.2951, simple_loss=0.3721, pruned_loss=0.1091, over 28966.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3802, pruned_loss=0.124, over 5688906.58 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.355, pruned_loss=0.099, over 5731156.32 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3828, pruned_loss=0.127, over 5675017.53 frames. ], batch size: 106, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:09:25,826 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=510717.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:09:28,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=510720.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:09:49,650 INFO [train.py:968] (0/2) Epoch 12, batch 9550, giga_loss[loss=0.3089, simple_loss=0.3759, pruned_loss=0.1209, over 29043.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3847, pruned_loss=0.1279, over 5688491.48 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3548, pruned_loss=0.09891, over 5734718.84 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3874, pruned_loss=0.1308, over 5673456.16 frames. ], batch size: 155, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:09:55,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-06 04:09:57,128 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=510749.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:09:59,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.675e+03 2.063e+03 3.267e+03 8.444e+03, threshold=4.126e+03, percent-clipped=26.0
+2023-03-06 04:10:04,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9335, 3.7560, 3.5737, 2.0058], device='cuda:0'), covar=tensor([0.0562, 0.0707, 0.0712, 0.2109], device='cuda:0'), in_proj_covar=tensor([0.1079, 0.1010, 0.0882, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 04:10:07,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8014, 1.8857, 1.7218, 1.7613], device='cuda:0'), covar=tensor([0.1423, 0.1833, 0.1937, 0.1719], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0729, 0.0668, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 04:10:12,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3816, 0.9999, 4.2581, 3.2395], device='cuda:0'), covar=tensor([0.1669, 0.2896, 0.0385, 0.0922], device='cuda:0'), in_proj_covar=tensor([0.0670, 0.0592, 0.0873, 0.0771], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 04:10:15,240 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=510767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:10:17,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4988, 1.7648, 1.6691, 1.4940], device='cuda:0'), covar=tensor([0.1829, 0.1513, 0.1029, 0.1280], device='cuda:0'), in_proj_covar=tensor([0.1736, 0.1641, 0.1606, 0.1697], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 04:10:33,407 INFO [train.py:968] (0/2) Epoch 12, batch 9600, libri_loss[loss=0.325, simple_loss=0.3974, pruned_loss=0.1262, over 29488.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3852, pruned_loss=0.1296, over 5693262.42 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3542, pruned_loss=0.09878, over 5741754.57 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3892, pruned_loss=0.1334, over 5672618.31 frames. ], batch size: 85, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:10:51,243 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-06 04:11:19,984 INFO [train.py:968] (0/2) Epoch 12, batch 9650, giga_loss[loss=0.3724, simple_loss=0.3986, pruned_loss=0.1731, over 23333.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3858, pruned_loss=0.1313, over 5684312.64 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.354, pruned_loss=0.09857, over 5744547.47 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3899, pruned_loss=0.1352, over 5664270.44 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:11:30,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.675e+03 2.256e+03 2.910e+03 5.829e+03, threshold=4.511e+03, percent-clipped=9.0
+2023-03-06 04:12:10,363 INFO [train.py:968] (0/2) Epoch 12, batch 9700, giga_loss[loss=0.3549, simple_loss=0.3836, pruned_loss=0.163, over 23636.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3856, pruned_loss=0.1321, over 5668703.54 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3536, pruned_loss=0.09845, over 5745153.17 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3894, pruned_loss=0.1355, over 5651884.44 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:12:21,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3727, 1.6104, 1.4319, 1.3238], device='cuda:0'), covar=tensor([0.2291, 0.1787, 0.1700, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.1747, 0.1650, 0.1616, 0.1709], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 04:12:31,590 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-06 04:12:44,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3384, 1.7180, 1.3464, 1.5954], device='cuda:0'), covar=tensor([0.2607, 0.2550, 0.2846, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.1311, 0.0972, 0.1160, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 04:12:54,304 INFO [train.py:968] (0/2) Epoch 12, batch 9750, giga_loss[loss=0.3391, simple_loss=0.373, pruned_loss=0.1526, over 23910.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3842, pruned_loss=0.1297, over 5664843.54 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3542, pruned_loss=0.09887, over 5738897.96 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3874, pruned_loss=0.1328, over 5655331.51 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:13:02,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.417e+02 1.505e+03 2.031e+03 2.845e+03 9.611e+03, threshold=4.062e+03, percent-clipped=5.0
+2023-03-06 04:13:37,783 INFO [train.py:968] (0/2) Epoch 12, batch 9800, giga_loss[loss=0.2734, simple_loss=0.3623, pruned_loss=0.09223, over 29053.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3836, pruned_loss=0.1279, over 5663112.65 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.354, pruned_loss=0.09889, over 5732353.60 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3869, pruned_loss=0.1309, over 5659232.08 frames. ], batch size: 155, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:13:49,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0712, 3.8822, 3.6920, 1.8350], device='cuda:0'), covar=tensor([0.0649, 0.0784, 0.0793, 0.2154], device='cuda:0'), in_proj_covar=tensor([0.1082, 0.1011, 0.0883, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 04:13:59,357 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-06 04:14:25,548 INFO [train.py:968] (0/2) Epoch 12, batch 9850, giga_loss[loss=0.3185, simple_loss=0.382, pruned_loss=0.1274, over 28770.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3845, pruned_loss=0.1277, over 5667779.70 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3544, pruned_loss=0.09899, over 5731816.35 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3874, pruned_loss=0.1305, over 5664264.02 frames. ], batch size: 262, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:14:37,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.987e+02 1.437e+03 1.733e+03 2.388e+03 4.730e+03, threshold=3.466e+03, percent-clipped=3.0
+2023-03-06 04:14:54,699 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 04:15:14,168 INFO [train.py:968] (0/2) Epoch 12, batch 9900, giga_loss[loss=0.3212, simple_loss=0.3771, pruned_loss=0.1326, over 27967.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3848, pruned_loss=0.1284, over 5661054.42 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3545, pruned_loss=0.09901, over 5734976.75 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3878, pruned_loss=0.1315, over 5653611.64 frames. ], batch size: 412, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:15:21,312 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=511100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:16:01,135 INFO [train.py:968] (0/2) Epoch 12, batch 9950, giga_loss[loss=0.3083, simple_loss=0.3706, pruned_loss=0.123, over 28776.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3836, pruned_loss=0.128, over 5672077.11 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3544, pruned_loss=0.09891, over 5739561.69 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3868, pruned_loss=0.1312, over 5660592.40 frames. ], batch size: 284, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:16:01,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=511142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:16:12,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.967e+02 1.512e+03 1.920e+03 2.693e+03 7.322e+03, threshold=3.840e+03, percent-clipped=19.0
+2023-03-06 04:16:22,257 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-06 04:16:42,278 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 04:16:54,986 INFO [train.py:968] (0/2) Epoch 12, batch 10000, giga_loss[loss=0.4137, simple_loss=0.4329, pruned_loss=0.1972, over 26655.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3821, pruned_loss=0.1283, over 5664936.71 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3544, pruned_loss=0.0989, over 5741373.88 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.385, pruned_loss=0.1312, over 5653313.79 frames. ], batch size: 555, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:16:56,157 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 04:17:42,014 INFO [train.py:968] (0/2) Epoch 12, batch 10050, giga_loss[loss=0.3105, simple_loss=0.3746, pruned_loss=0.1231, over 28999.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3801, pruned_loss=0.1275, over 5669862.43 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3546, pruned_loss=0.09891, over 5739136.78 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3827, pruned_loss=0.1304, over 5661259.10 frames. ], batch size: 145, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:17:53,973 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.596e+02 1.760e+03 2.450e+03 3.904e+03 1.075e+04, threshold=4.900e+03, percent-clipped=25.0
+2023-03-06 04:18:30,534 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=511285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:18:34,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=511288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:18:38,604 INFO [train.py:968] (0/2) Epoch 12, batch 10100, giga_loss[loss=0.2713, simple_loss=0.3438, pruned_loss=0.09941, over 29001.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3777, pruned_loss=0.1265, over 5668275.02 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3545, pruned_loss=0.09888, over 5741015.69 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3802, pruned_loss=0.1292, over 5659056.92 frames. ], batch size: 112, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:18:53,102 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 04:19:01,349 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=511317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:19:26,596 INFO [train.py:968] (0/2) Epoch 12, batch 10150, giga_loss[loss=0.4147, simple_loss=0.4338, pruned_loss=0.1978, over 26623.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3782, pruned_loss=0.1279, over 5669461.92 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3547, pruned_loss=0.09913, over 5743889.87 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3805, pruned_loss=0.1304, over 5657935.31 frames. ], batch size: 555, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:19:36,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.637e+03 2.119e+03 3.050e+03 1.052e+04, threshold=4.238e+03, percent-clipped=11.0
+2023-03-06 04:20:11,645 INFO [train.py:968] (0/2) Epoch 12, batch 10200, giga_loss[loss=0.2676, simple_loss=0.3435, pruned_loss=0.09584, over 28194.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3761, pruned_loss=0.1258, over 5669419.34 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3549, pruned_loss=0.09914, over 5745338.58 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3782, pruned_loss=0.1283, over 5657634.51 frames. ], batch size: 368, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:20:56,305 INFO [train.py:968] (0/2) Epoch 12, batch 10250, giga_loss[loss=0.2977, simple_loss=0.3657, pruned_loss=0.1149, over 28865.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3726, pruned_loss=0.1216, over 5672484.10 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3552, pruned_loss=0.09929, over 5746323.08 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3746, pruned_loss=0.1242, over 5659731.85 frames. ], batch size: 186, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:21:03,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4179, 1.6342, 1.3398, 1.5413], device='cuda:0'), covar=tensor([0.2329, 0.2274, 0.2538, 0.2170], device='cuda:0'), in_proj_covar=tensor([0.1315, 0.0976, 0.1163, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 04:21:07,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.708e+02 1.238e+03 1.554e+03 2.107e+03 6.963e+03, threshold=3.108e+03, percent-clipped=2.0
+2023-03-06 04:21:30,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=511475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:21:42,437 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=511487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:21:45,214 INFO [train.py:968] (0/2) Epoch 12, batch 10300, giga_loss[loss=0.2973, simple_loss=0.3685, pruned_loss=0.113, over 29066.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3711, pruned_loss=0.1201, over 5665207.47 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3557, pruned_loss=0.09957, over 5746517.20 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3727, pruned_loss=0.1224, over 5652962.97 frames. ], batch size: 155, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:22:29,291 INFO [train.py:968] (0/2) Epoch 12, batch 10350, giga_loss[loss=0.3956, simple_loss=0.4183, pruned_loss=0.1865, over 26657.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3703, pruned_loss=0.1197, over 5678632.02 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3553, pruned_loss=0.09911, over 5751537.37 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3725, pruned_loss=0.1229, over 5661106.12 frames. ], batch size: 555, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:22:36,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3751, 1.6565, 1.3428, 1.3353], device='cuda:0'), covar=tensor([0.2173, 0.2111, 0.2332, 0.1915], device='cuda:0'), in_proj_covar=tensor([0.1317, 0.0977, 0.1165, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 04:22:39,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.104e+02 1.359e+03 1.781e+03 2.255e+03 6.402e+03, threshold=3.561e+03, percent-clipped=10.0
+2023-03-06 04:22:54,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0026, 2.8347, 2.6930, 1.6484], device='cuda:0'), covar=tensor([0.0957, 0.1012, 0.0924, 0.1813], device='cuda:0'), in_proj_covar=tensor([0.1072, 0.1006, 0.0878, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 04:23:14,589 INFO [train.py:968] (0/2) Epoch 12, batch 10400, giga_loss[loss=0.2627, simple_loss=0.3432, pruned_loss=0.09115, over 28949.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3671, pruned_loss=0.1185, over 5669263.36 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3553, pruned_loss=0.09903, over 5745324.49 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3695, pruned_loss=0.1221, over 5657099.10 frames. ], batch size: 128, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:23:32,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2617, 1.3689, 1.1538, 1.1928], device='cuda:0'), covar=tensor([0.1575, 0.1352, 0.1324, 0.1374], device='cuda:0'), in_proj_covar=tensor([0.1730, 0.1645, 0.1604, 0.1703], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 04:23:42,247 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=511618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:23:44,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=511621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:24:04,240 INFO [train.py:968] (0/2) Epoch 12, batch 10450, giga_loss[loss=0.3744, simple_loss=0.4194, pruned_loss=0.1647, over 28699.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3686, pruned_loss=0.1202, over 5675097.73 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3555, pruned_loss=0.09917, over 5745952.20 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3704, pruned_loss=0.123, over 5664408.18 frames. ], batch size: 262, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:24:07,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2093, 4.0225, 3.7925, 1.8977], device='cuda:0'), covar=tensor([0.0489, 0.0636, 0.0680, 0.2320], device='cuda:0'), in_proj_covar=tensor([0.1074, 0.1006, 0.0876, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 04:24:11,621 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=511650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:24:14,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.788e+02 1.655e+03 2.176e+03 3.209e+03 5.101e+03, threshold=4.353e+03, percent-clipped=16.0
+2023-03-06 04:24:19,010 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=511659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:24:48,797 INFO [train.py:968] (0/2) Epoch 12, batch 10500, giga_loss[loss=0.3007, simple_loss=0.3727, pruned_loss=0.1144, over 28801.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3703, pruned_loss=0.1199, over 5682179.68 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3554, pruned_loss=0.09899, over 5750029.88 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5668198.36 frames. ], batch size: 199, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:25:17,960 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5367, 1.6621, 1.2494, 1.2817], device='cuda:0'), covar=tensor([0.0805, 0.0466, 0.0976, 0.0903], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0438, 0.0498, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 04:25:33,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6195, 1.6633, 1.5680, 1.4970], device='cuda:0'), covar=tensor([0.1831, 0.1972, 0.1684, 0.1677], device='cuda:0'), in_proj_covar=tensor([0.1733, 0.1654, 0.1611, 0.1709], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 04:25:35,028 INFO [train.py:968] (0/2) Epoch 12, batch 10550, giga_loss[loss=0.3213, simple_loss=0.3651, pruned_loss=0.1388, over 23616.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3713, pruned_loss=0.1205, over 5666994.73 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3555, pruned_loss=0.09895, over 5754294.99 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3731, pruned_loss=0.1236, over 5650125.70 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:25:38,335 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.14 vs. limit=5.0
+2023-03-06 04:25:47,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.061e+02 1.332e+03 1.687e+03 2.081e+03 5.735e+03, threshold=3.374e+03, percent-clipped=3.0
+2023-03-06 04:26:15,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3830, 2.0487, 1.4645, 0.5467], device='cuda:0'), covar=tensor([0.3672, 0.1751, 0.2829, 0.4268], device='cuda:0'), in_proj_covar=tensor([0.1559, 0.1478, 0.1479, 0.1280], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 04:26:24,293 INFO [train.py:968] (0/2) Epoch 12, batch 10600, giga_loss[loss=0.3065, simple_loss=0.3709, pruned_loss=0.121, over 28849.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3713, pruned_loss=0.1212, over 5652210.79 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3553, pruned_loss=0.09881, over 5758782.58 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3735, pruned_loss=0.1245, over 5631485.50 frames. ], batch size: 186, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:27:09,152 INFO [train.py:968] (0/2) Epoch 12, batch 10650, giga_loss[loss=0.3866, simple_loss=0.4206, pruned_loss=0.1763, over 28705.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3721, pruned_loss=0.1224, over 5657121.22 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3554, pruned_loss=0.09875, over 5762513.44 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3742, pruned_loss=0.1256, over 5634744.27 frames. ], batch size: 92, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:27:15,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5832, 2.4609, 2.5720, 2.1138], device='cuda:0'), covar=tensor([0.1332, 0.1894, 0.1425, 0.1687], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0735, 0.0676, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 04:27:21,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.633e+03 2.113e+03 2.917e+03 1.249e+04, threshold=4.227e+03, percent-clipped=15.0
+2023-03-06 04:27:24,231 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=511859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:27:27,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=511862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:27:57,923 INFO [train.py:968] (0/2) Epoch 12, batch 10700, giga_loss[loss=0.314, simple_loss=0.3759, pruned_loss=0.1261, over 28869.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3732, pruned_loss=0.1232, over 5636235.75 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3556, pruned_loss=0.09886, over 5748303.15 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3753, pruned_loss=0.1265, over 5627958.27 frames. ], batch size: 199, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:28:45,159 INFO [train.py:968] (0/2) Epoch 12, batch 10750, giga_loss[loss=0.2933, simple_loss=0.3758, pruned_loss=0.1054, over 28938.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3742, pruned_loss=0.1232, over 5649908.57 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3558, pruned_loss=0.09889, over 5751734.30 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3761, pruned_loss=0.1264, over 5637480.86 frames. ], batch size: 145, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:28:56,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.124e+02 1.453e+03 1.774e+03 2.636e+03 6.282e+03, threshold=3.548e+03, percent-clipped=3.0
+2023-03-06 04:29:14,058 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6225, 1.7208, 1.9308, 1.4450], device='cuda:0'), covar=tensor([0.1682, 0.2225, 0.1303, 0.1591], device='cuda:0'), in_proj_covar=tensor([0.0836, 0.0705, 0.0880, 0.0783], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 04:29:27,920 INFO [train.py:968] (0/2) Epoch 12, batch 10800, giga_loss[loss=0.315, simple_loss=0.3733, pruned_loss=0.1284, over 29066.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3745, pruned_loss=0.1236, over 5653694.20 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3557, pruned_loss=0.09884, over 5751260.64 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3767, pruned_loss=0.1269, over 5641240.17 frames. ], batch size: 128, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:29:36,871 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-512000.pt
+2023-03-06 04:29:42,658 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=512005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:29:44,922 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=512008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:30:07,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=512034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:30:10,871 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=512037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:30:14,618 INFO [train.py:968] (0/2) Epoch 12, batch 10850, libri_loss[loss=0.2364, simple_loss=0.3155, pruned_loss=0.07866, over 29643.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.375, pruned_loss=0.1242, over 5665818.77 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3551, pruned_loss=0.09846, over 5757064.26 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3781, pruned_loss=0.1283, over 5646831.30 frames. ], batch size: 69, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:30:20,323 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=512048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:30:29,902 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.287e+02 1.726e+03 2.131e+03 3.071e+03 1.272e+04, threshold=4.263e+03, percent-clipped=15.0
+2023-03-06 04:31:03,032 INFO [train.py:968] (0/2) Epoch 12, batch 10900, giga_loss[loss=0.3014, simple_loss=0.376, pruned_loss=0.1134, over 28634.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3769, pruned_loss=0.1249, over 5655478.81 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3555, pruned_loss=0.09874, over 5748271.16 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3794, pruned_loss=0.1283, over 5646232.17 frames. ], batch size: 307, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:31:19,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9240, 1.1522, 1.0927, 0.8932], device='cuda:0'), covar=tensor([0.1570, 0.1680, 0.1006, 0.1322], device='cuda:0'), in_proj_covar=tensor([0.1750, 0.1666, 0.1617, 0.1724], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 04:31:36,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4143, 1.9909, 1.5410, 1.6222], device='cuda:0'), covar=tensor([0.0625, 0.0241, 0.0269, 0.0633], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0059, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 04:31:52,689 INFO [train.py:968] (0/2) Epoch 12, batch 10950, giga_loss[loss=0.274, simple_loss=0.3533, pruned_loss=0.09733, over 28892.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3768, pruned_loss=0.1241, over 5650235.47 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3551, pruned_loss=0.09849, over 5743137.70 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3797, pruned_loss=0.1277, over 5645537.24 frames. ], batch size: 145, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:32:03,465 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-06 04:32:05,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.943e+02 1.725e+03 2.231e+03 2.984e+03 7.954e+03, threshold=4.463e+03, percent-clipped=7.0
+2023-03-06 04:32:27,384 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=512177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:32:30,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=512180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:32:42,765 INFO [train.py:968] (0/2) Epoch 12, batch 11000, libri_loss[loss=0.3204, simple_loss=0.3817, pruned_loss=0.1296, over 19748.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3759, pruned_loss=0.1243, over 5648644.52 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3547, pruned_loss=0.09849, over 5738935.12 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3792, pruned_loss=0.128, over 5646432.25 frames. ], batch size: 186, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:32:55,166 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-06 04:32:58,332 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=512209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:32:59,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 04:33:18,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=512227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:33:25,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=512234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:33:26,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2710, 1.5241, 1.2708, 0.9308], device='cuda:0'), covar=tensor([0.2226, 0.2192, 0.2468, 0.1903], device='cuda:0'), in_proj_covar=tensor([0.1314, 0.0973, 0.1162, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 04:33:35,736 INFO [train.py:968] (0/2) Epoch 12, batch 11050, giga_loss[loss=0.3908, simple_loss=0.4135, pruned_loss=0.184, over 23601.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.375, pruned_loss=0.1244, over 5654067.14 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3545, pruned_loss=0.09832, over 5738293.00 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3783, pruned_loss=0.128, over 5651240.47 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:33:51,708 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.860e+03 2.364e+03 3.143e+03 1.092e+04, threshold=4.728e+03, percent-clipped=13.0
+2023-03-06 04:34:31,046 INFO [train.py:968] (0/2) Epoch 12, batch 11100, giga_loss[loss=0.2696, simple_loss=0.3497, pruned_loss=0.09478, over 28844.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3754, pruned_loss=0.1254, over 5652401.44 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3547, pruned_loss=0.09845, over 5737648.06 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3779, pruned_loss=0.1283, over 5649982.44 frames. ], batch size: 174, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:35:16,535 INFO [train.py:968] (0/2) Epoch 12, batch 11150, giga_loss[loss=0.366, simple_loss=0.4011, pruned_loss=0.1654, over 28569.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3748, pruned_loss=0.1257, over 5663526.15 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3546, pruned_loss=0.09834, over 5742434.98 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3775, pruned_loss=0.1289, over 5655511.02 frames. ], batch size: 336, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:35:26,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.469e+03 1.872e+03 2.512e+03 6.672e+03, threshold=3.744e+03, percent-clipped=4.0
+2023-03-06 04:35:30,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3969, 1.7633, 1.4006, 1.4662], device='cuda:0'), covar=tensor([0.2449, 0.2312, 0.2518, 0.1903], device='cuda:0'), in_proj_covar=tensor([0.1312, 0.0971, 0.1158, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 04:35:48,783 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=512377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:35:51,168 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=512380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:36:02,530 INFO [train.py:968] (0/2) Epoch 12, batch 11200, giga_loss[loss=0.298, simple_loss=0.3654, pruned_loss=0.1153, over 28900.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3759, pruned_loss=0.1271, over 5654932.68 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3546, pruned_loss=0.09834, over 5734533.80 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3783, pruned_loss=0.1299, over 5654040.38 frames. ], batch size: 199, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:36:18,600 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=512409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:36:35,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=512423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:36:55,528 INFO [train.py:968] (0/2) Epoch 12, batch 11250, giga_loss[loss=0.284, simple_loss=0.352, pruned_loss=0.108, over 28767.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3759, pruned_loss=0.1273, over 5652996.90 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3546, pruned_loss=0.09829, over 5736462.68 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3779, pruned_loss=0.1298, over 5650111.29 frames. ], batch size: 99, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:37:09,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.646e+03 2.056e+03 2.837e+03 8.278e+03, threshold=4.112e+03, percent-clipped=10.0
+2023-03-06 04:37:40,624 INFO [train.py:968] (0/2) Epoch 12, batch 11300, giga_loss[loss=0.3791, simple_loss=0.419, pruned_loss=0.1696, over 27529.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3767, pruned_loss=0.1278, over 5659100.61 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3545, pruned_loss=0.09826, over 5732147.65 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.379, pruned_loss=0.1306, over 5658116.70 frames. ], batch size: 472, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:38:27,469 INFO [train.py:968] (0/2) Epoch 12, batch 11350, giga_loss[loss=0.3448, simple_loss=0.3887, pruned_loss=0.1504, over 28659.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3785, pruned_loss=0.1285, over 5653393.52 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3548, pruned_loss=0.09837, over 5724100.39 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3804, pruned_loss=0.1311, over 5659548.35 frames. ], batch size: 92, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:38:41,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.984e+02 1.580e+03 2.339e+03 3.366e+03 9.060e+03, threshold=4.678e+03, percent-clipped=13.0
+2023-03-06 04:38:48,569 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=512566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:38:51,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=512569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:39:14,184 INFO [train.py:968] (0/2) Epoch 12, batch 11400, giga_loss[loss=0.3705, simple_loss=0.408, pruned_loss=0.1665, over 27883.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3786, pruned_loss=0.1294, over 5654354.66 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3543, pruned_loss=0.09816, over 5730918.75 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3815, pruned_loss=0.1329, over 5650526.43 frames. ], batch size: 412, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:39:19,924 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=512598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:39:22,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=512602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:40:01,548 INFO [train.py:968] (0/2) Epoch 12, batch 11450, libri_loss[loss=0.255, simple_loss=0.3283, pruned_loss=0.09084, over 29635.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3772, pruned_loss=0.1286, over 5659014.71 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3544, pruned_loss=0.09837, over 5736471.76 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3804, pruned_loss=0.1323, over 5648165.21 frames. ], batch size: 73, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:40:14,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.779e+02 1.527e+03 2.013e+03 2.706e+03 6.021e+03, threshold=4.026e+03, percent-clipped=2.0
+2023-03-06 04:40:45,529 INFO [train.py:968] (0/2) Epoch 12, batch 11500, giga_loss[loss=0.265, simple_loss=0.3406, pruned_loss=0.09475, over 28623.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3763, pruned_loss=0.1275, over 5663409.52 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3544, pruned_loss=0.0983, over 5737075.02 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3794, pruned_loss=0.1313, over 5652489.98 frames. ], batch size: 262, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:41:28,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3284, 2.0589, 1.6760, 0.5035], device='cuda:0'), covar=tensor([0.3766, 0.2076, 0.2629, 0.4528], device='cuda:0'), in_proj_covar=tensor([0.1587, 0.1509, 0.1508, 0.1301], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 04:41:33,521 INFO [train.py:968] (0/2) Epoch 12, batch 11550, giga_loss[loss=0.3293, simple_loss=0.3661, pruned_loss=0.1463, over 23375.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3768, pruned_loss=0.1272, over 5669990.46 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3548, pruned_loss=0.09863, over 5741840.00 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3798, pruned_loss=0.1311, over 5654463.94 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:41:35,738 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=512745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:41:37,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=512748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:41:45,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.841e+02 1.653e+03 2.033e+03 2.834e+03 7.184e+03, threshold=4.066e+03, percent-clipped=8.0
+2023-03-06 04:42:04,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=512777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:42:16,873 INFO [train.py:968] (0/2) Epoch 12, batch 11600, libri_loss[loss=0.2564, simple_loss=0.3471, pruned_loss=0.08286, over 29162.00 frames. ], tot_loss[loss=0.314, simple_loss=0.376, pruned_loss=0.126, over 5667874.51 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3542, pruned_loss=0.09831, over 5738147.36 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3799, pruned_loss=0.1307, over 5655511.03 frames. ], batch size: 97, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:42:28,925 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 04:43:04,064 INFO [train.py:968] (0/2) Epoch 12, batch 11650, giga_loss[loss=0.3718, simple_loss=0.4213, pruned_loss=0.1612, over 28288.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3768, pruned_loss=0.1263, over 5675876.67 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.354, pruned_loss=0.09825, over 5731958.86 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3811, pruned_loss=0.1312, over 5669304.33 frames. ], batch size: 368, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:43:07,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-06 04:43:18,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.148e+02 1.514e+03 1.969e+03 2.691e+03 7.791e+03, threshold=3.938e+03, percent-clipped=14.0
+2023-03-06 04:43:29,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1347, 1.1264, 3.9121, 3.1802], device='cuda:0'), covar=tensor([0.1697, 0.2670, 0.0441, 0.0829], device='cuda:0'), in_proj_covar=tensor([0.0671, 0.0593, 0.0873, 0.0775], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 04:43:48,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3709, 3.7942, 1.4878, 1.5785], device='cuda:0'), covar=tensor([0.1051, 0.0392, 0.0946, 0.1428], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0516, 0.0342, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 04:43:52,540 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=512890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:43:53,551 INFO [train.py:968] (0/2) Epoch 12, batch 11700, giga_loss[loss=0.3828, simple_loss=0.4202, pruned_loss=0.1727, over 28558.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3811, pruned_loss=0.1305, over 5673391.25 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3541, pruned_loss=0.09838, over 5735402.86 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3849, pruned_loss=0.1348, over 5663936.62 frames. ], batch size: 60, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:44:06,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4962, 1.7907, 1.6522, 1.6527], device='cuda:0'), covar=tensor([0.1557, 0.1626, 0.1962, 0.1704], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0736, 0.0677, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 04:44:40,274 INFO [train.py:968] (0/2) Epoch 12, batch 11750, giga_loss[loss=0.4102, simple_loss=0.4294, pruned_loss=0.1955, over 26747.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3803, pruned_loss=0.1296, over 5685856.01 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3543, pruned_loss=0.09853, over 5739851.11 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3839, pruned_loss=0.1338, over 5672992.38 frames. ], batch size: 555, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:44:54,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.860e+02 1.708e+03 2.300e+03 3.362e+03 9.546e+03, threshold=4.601e+03, percent-clipped=15.0
+2023-03-06 04:45:18,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2759, 1.2618, 1.1107, 1.4445], device='cuda:0'), covar=tensor([0.0748, 0.0341, 0.0337, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0113, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0059, 0.0052, 0.0089], device='cuda:0')
+2023-03-06 04:45:26,886 INFO [train.py:968] (0/2) Epoch 12, batch 11800, libri_loss[loss=0.2889, simple_loss=0.3719, pruned_loss=0.1029, over 29295.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3813, pruned_loss=0.1293, over 5680446.92 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3543, pruned_loss=0.09852, over 5733514.76 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3847, pruned_loss=0.1334, over 5675016.57 frames. ], batch size: 94, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:46:13,413 INFO [train.py:968] (0/2) Epoch 12, batch 11850, giga_loss[loss=0.2744, simple_loss=0.3543, pruned_loss=0.09728, over 29094.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3796, pruned_loss=0.1276, over 5673431.84 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.354, pruned_loss=0.09835, over 5737229.42 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3835, pruned_loss=0.1319, over 5663877.01 frames. ], batch size: 155, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:46:18,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2402, 1.1267, 3.8762, 3.2558], device='cuda:0'), covar=tensor([0.1609, 0.2661, 0.0426, 0.1523], device='cuda:0'), in_proj_covar=tensor([0.0668, 0.0591, 0.0869, 0.0773], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 04:46:29,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.022e+02 1.387e+03 1.836e+03 2.594e+03 1.038e+04, threshold=3.672e+03, percent-clipped=7.0
+2023-03-06 04:46:59,296 INFO [train.py:968] (0/2) Epoch 12, batch 11900, giga_loss[loss=0.2772, simple_loss=0.3551, pruned_loss=0.09968, over 28984.00 frames. ], tot_loss[loss=0.315, simple_loss=0.378, pruned_loss=0.126, over 5673567.74 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3544, pruned_loss=0.09869, over 5732091.59 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3813, pruned_loss=0.1299, over 5668441.34 frames. ], batch size: 174, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:47:45,577 INFO [train.py:968] (0/2) Epoch 12, batch 11950, libri_loss[loss=0.2389, simple_loss=0.3122, pruned_loss=0.08284, over 29622.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.376, pruned_loss=0.1248, over 5686909.34 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3539, pruned_loss=0.09843, over 5735663.86 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3795, pruned_loss=0.1286, over 5678538.59 frames. ], batch size: 69, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:47:59,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.373e+02 1.512e+03 2.009e+03 3.175e+03 8.039e+03, threshold=4.017e+03, percent-clipped=16.0
+2023-03-06 04:48:30,199 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4763, 1.6985, 1.4148, 1.6063], device='cuda:0'), covar=tensor([0.1962, 0.1880, 0.1965, 0.1713], device='cuda:0'), in_proj_covar=tensor([0.1315, 0.0978, 0.1160, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 04:48:34,066 INFO [train.py:968] (0/2) Epoch 12, batch 12000, libri_loss[loss=0.2998, simple_loss=0.3828, pruned_loss=0.1083, over 25941.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3776, pruned_loss=0.1265, over 5665787.78 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.354, pruned_loss=0.09851, over 5735900.07 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3811, pruned_loss=0.1304, over 5656885.70 frames. ], batch size: 136, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:48:34,072 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 04:48:41,311 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1860, 1.8065, 1.4111, 0.3923], device='cuda:0'), covar=tensor([0.2782, 0.1818, 0.2863, 0.3782], device='cuda:0'), in_proj_covar=tensor([0.1585, 0.1512, 0.1506, 0.1301], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 04:48:41,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1997, 1.7055, 1.5154, 1.0680], device='cuda:0'), covar=tensor([0.1844, 0.2603, 0.1497, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0835, 0.0702, 0.0879, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 04:48:42,400 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2158, simple_loss=0.3219, pruned_loss=0.05489, over 944034.00 frames. 
+2023-03-06 04:48:42,401 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 04:49:14,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4712, 1.7028, 1.5120, 1.3328], device='cuda:0'), covar=tensor([0.2020, 0.1677, 0.1437, 0.1804], device='cuda:0'), in_proj_covar=tensor([0.1736, 0.1660, 0.1603, 0.1717], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 04:49:28,434 INFO [train.py:968] (0/2) Epoch 12, batch 12050, giga_loss[loss=0.383, simple_loss=0.4268, pruned_loss=0.1696, over 28272.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3786, pruned_loss=0.1268, over 5653463.44 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3539, pruned_loss=0.09848, over 5721038.87 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3821, pruned_loss=0.1305, over 5658458.19 frames. ], batch size: 368, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:49:34,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4774, 1.8108, 1.7160, 1.2985], device='cuda:0'), covar=tensor([0.1797, 0.2492, 0.1537, 0.1737], device='cuda:0'), in_proj_covar=tensor([0.0836, 0.0702, 0.0879, 0.0781], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 04:49:42,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.515e+03 1.892e+03 2.450e+03 7.589e+03, threshold=3.783e+03, percent-clipped=3.0
+2023-03-06 04:49:48,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=513265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:50:03,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=513279.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:50:16,577 INFO [train.py:968] (0/2) Epoch 12, batch 12100, giga_loss[loss=0.3429, simple_loss=0.3953, pruned_loss=0.1452, over 27958.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3791, pruned_loss=0.1283, over 5656917.59 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3541, pruned_loss=0.09862, over 5723523.37 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.382, pruned_loss=0.1316, over 5657487.26 frames. ], batch size: 412, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:51:03,609 INFO [train.py:968] (0/2) Epoch 12, batch 12150, libri_loss[loss=0.289, simple_loss=0.3717, pruned_loss=0.1031, over 29739.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.379, pruned_loss=0.1286, over 5660995.55 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.354, pruned_loss=0.09851, over 5728246.33 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3822, pruned_loss=0.1323, over 5655152.94 frames. ], batch size: 87, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:51:20,143 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.482e+02 1.509e+03 1.874e+03 2.680e+03 5.499e+03, threshold=3.749e+03, percent-clipped=7.0
+2023-03-06 04:51:41,271 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=513379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:51:50,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4286, 1.5484, 1.6282, 1.5340], device='cuda:0'), covar=tensor([0.1123, 0.1223, 0.1250, 0.1173], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0737, 0.0677, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 04:51:54,228 INFO [train.py:968] (0/2) Epoch 12, batch 12200, giga_loss[loss=0.2888, simple_loss=0.3612, pruned_loss=0.1082, over 28966.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3804, pruned_loss=0.1301, over 5659291.68 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3538, pruned_loss=0.09843, over 5730737.21 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3834, pruned_loss=0.1334, over 5651875.29 frames. ], batch size: 136, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:52:10,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=513408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:52:12,434 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=513411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:52:38,885 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=513440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:52:41,695 INFO [train.py:968] (0/2) Epoch 12, batch 12250, giga_loss[loss=0.312, simple_loss=0.3755, pruned_loss=0.1243, over 28900.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3814, pruned_loss=0.1305, over 5661020.34 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3542, pruned_loss=0.09876, over 5727550.97 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3842, pruned_loss=0.1338, over 5655297.00 frames. ], batch size: 186, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:52:56,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.547e+02 1.496e+03 2.035e+03 2.526e+03 5.185e+03, threshold=4.069e+03, percent-clipped=4.0
+2023-03-06 04:53:30,677 INFO [train.py:968] (0/2) Epoch 12, batch 12300, libri_loss[loss=0.3094, simple_loss=0.3854, pruned_loss=0.1167, over 27751.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3811, pruned_loss=0.1309, over 5638277.23 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3541, pruned_loss=0.09871, over 5729492.80 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3839, pruned_loss=0.1341, over 5630677.70 frames. ], batch size: 116, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:54:09,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2832, 1.7579, 1.3352, 1.4787], device='cuda:0'), covar=tensor([0.0677, 0.0383, 0.0319, 0.0711], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0059, 0.0053, 0.0089], device='cuda:0')
+2023-03-06 04:54:14,786 INFO [train.py:968] (0/2) Epoch 12, batch 12350, giga_loss[loss=0.2876, simple_loss=0.3508, pruned_loss=0.1121, over 28680.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.382, pruned_loss=0.1311, over 5649423.50 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3547, pruned_loss=0.099, over 5730547.57 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3847, pruned_loss=0.1347, over 5639179.37 frames. ], batch size: 92, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:54:33,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.856e+02 1.521e+03 1.952e+03 2.671e+03 7.137e+03, threshold=3.904e+03, percent-clipped=6.0
+2023-03-06 04:54:35,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3245, 5.1535, 4.8825, 2.4214], device='cuda:0'), covar=tensor([0.0393, 0.0532, 0.0611, 0.1738], device='cuda:0'), in_proj_covar=tensor([0.1093, 0.1017, 0.0894, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 04:54:50,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2381, 1.2326, 3.7249, 2.9583], device='cuda:0'), covar=tensor([0.1621, 0.2591, 0.0455, 0.0869], device='cuda:0'), in_proj_covar=tensor([0.0676, 0.0594, 0.0876, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 04:55:02,192 INFO [train.py:968] (0/2) Epoch 12, batch 12400, giga_loss[loss=0.3221, simple_loss=0.3858, pruned_loss=0.1292, over 28702.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3819, pruned_loss=0.1304, over 5646752.41 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3551, pruned_loss=0.09925, over 5722804.07 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.384, pruned_loss=0.1333, over 5645088.62 frames. ], batch size: 242, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:55:13,039 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-06 04:55:43,782 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 04:55:48,803 INFO [train.py:968] (0/2) Epoch 12, batch 12450, giga_loss[loss=0.4093, simple_loss=0.4326, pruned_loss=0.193, over 26613.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3805, pruned_loss=0.1296, over 5640696.57 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3551, pruned_loss=0.09921, over 5716076.16 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3831, pruned_loss=0.1331, over 5642712.37 frames. ], batch size: 555, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:56:01,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=513654.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 04:56:06,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.434e+02 1.496e+03 2.191e+03 3.181e+03 1.167e+04, threshold=4.382e+03, percent-clipped=16.0
+2023-03-06 04:56:31,810 INFO [train.py:968] (0/2) Epoch 12, batch 12500, giga_loss[loss=0.3485, simple_loss=0.3956, pruned_loss=0.1507, over 27590.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3779, pruned_loss=0.1275, over 5652826.24 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3553, pruned_loss=0.09924, over 5719581.49 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3805, pruned_loss=0.1311, over 5649386.02 frames. ], batch size: 472, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 04:56:58,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5298, 4.3834, 4.1681, 1.9461], device='cuda:0'), covar=tensor([0.0516, 0.0660, 0.0666, 0.1837], device='cuda:0'), in_proj_covar=tensor([0.1090, 0.1018, 0.0893, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 04:57:13,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3407, 1.2668, 1.2609, 1.4458], device='cuda:0'), covar=tensor([0.0768, 0.0343, 0.0324, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 04:57:21,660 INFO [train.py:968] (0/2) Epoch 12, batch 12550, giga_loss[loss=0.2776, simple_loss=0.3383, pruned_loss=0.1084, over 28533.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3747, pruned_loss=0.1258, over 5667699.84 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3553, pruned_loss=0.09915, over 5721665.23 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3771, pruned_loss=0.1291, over 5662383.86 frames. ], batch size: 85, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 04:57:32,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=513754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:57:40,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.707e+03 2.277e+03 3.333e+03 8.493e+03, threshold=4.554e+03, percent-clipped=11.0
+2023-03-06 04:57:47,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2040, 1.2371, 3.2888, 3.1612], device='cuda:0'), covar=tensor([0.1374, 0.2437, 0.0441, 0.0926], device='cuda:0'), in_proj_covar=tensor([0.0675, 0.0594, 0.0877, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 04:58:09,385 INFO [train.py:968] (0/2) Epoch 12, batch 12600, giga_loss[loss=0.309, simple_loss=0.34, pruned_loss=0.139, over 23325.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3708, pruned_loss=0.1239, over 5657098.14 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3552, pruned_loss=0.09913, over 5724156.18 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3734, pruned_loss=0.1275, over 5648607.61 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 04:58:12,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=513797.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 04:58:14,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=513800.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:58:33,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3469, 3.0957, 1.3848, 1.4632], device='cuda:0'), covar=tensor([0.0954, 0.0436, 0.0862, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0520, 0.0345, 0.0396], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 04:58:40,756 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=513829.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:58:51,619 INFO [train.py:968] (0/2) Epoch 12, batch 12650, giga_loss[loss=0.3162, simple_loss=0.3771, pruned_loss=0.1276, over 28907.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3699, pruned_loss=0.1244, over 5663401.37 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3542, pruned_loss=0.09856, over 5730248.11 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3734, pruned_loss=0.1287, over 5648937.53 frames. ], batch size: 174, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 04:59:10,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4015, 1.7158, 1.6439, 1.4368], device='cuda:0'), covar=tensor([0.1612, 0.1607, 0.1919, 0.1848], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0732, 0.0678, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 04:59:11,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.202e+02 1.618e+03 2.224e+03 3.580e+03 1.304e+04, threshold=4.449e+03, percent-clipped=16.0
+2023-03-06 04:59:16,512 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.55 vs. limit=5.0
+2023-03-06 04:59:20,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4158, 1.5801, 1.3220, 1.5037], device='cuda:0'), covar=tensor([0.2564, 0.2485, 0.2665, 0.2120], device='cuda:0'), in_proj_covar=tensor([0.1319, 0.0977, 0.1164, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 04:59:39,118 INFO [train.py:968] (0/2) Epoch 12, batch 12700, libri_loss[loss=0.2908, simple_loss=0.3696, pruned_loss=0.106, over 29526.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3699, pruned_loss=0.1244, over 5652188.86 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3551, pruned_loss=0.09912, over 5723815.23 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3726, pruned_loss=0.1285, over 5642884.54 frames. ], batch size: 89, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 04:59:42,582 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=513897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:59:44,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=513900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:00:15,606 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=513929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:00:28,212 INFO [train.py:968] (0/2) Epoch 12, batch 12750, giga_loss[loss=0.2985, simple_loss=0.3737, pruned_loss=0.1117, over 28689.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3695, pruned_loss=0.124, over 5654175.87 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.355, pruned_loss=0.0991, over 5726524.31 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3718, pruned_loss=0.1275, over 5643573.54 frames. ], batch size: 262, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 05:00:47,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.763e+02 1.409e+03 1.809e+03 2.624e+03 6.829e+03, threshold=3.617e+03, percent-clipped=5.0
+2023-03-06 05:00:56,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1476, 4.9719, 4.7099, 2.3941], device='cuda:0'), covar=tensor([0.0391, 0.0557, 0.0666, 0.1862], device='cuda:0'), in_proj_covar=tensor([0.1079, 0.1007, 0.0880, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 05:01:18,991 INFO [train.py:968] (0/2) Epoch 12, batch 12800, giga_loss[loss=0.2952, simple_loss=0.3683, pruned_loss=0.1111, over 28361.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3675, pruned_loss=0.1204, over 5652105.79 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3549, pruned_loss=0.09908, over 5725462.16 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3696, pruned_loss=0.1235, over 5643373.60 frames. ], batch size: 368, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 05:01:27,366 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-514000.pt
+2023-03-06 05:02:12,218 INFO [train.py:968] (0/2) Epoch 12, batch 12850, giga_loss[loss=0.2933, simple_loss=0.3644, pruned_loss=0.1111, over 28311.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3653, pruned_loss=0.117, over 5649546.46 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3546, pruned_loss=0.0989, over 5727783.06 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3675, pruned_loss=0.1199, over 5639554.60 frames. ], batch size: 368, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 05:02:32,185 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2326, 3.0293, 1.3011, 1.4639], device='cuda:0'), covar=tensor([0.0981, 0.0354, 0.0991, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0521, 0.0346, 0.0396], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 05:02:32,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.941e+02 1.449e+03 1.796e+03 2.427e+03 8.186e+03, threshold=3.592e+03, percent-clipped=8.0
+2023-03-06 05:03:05,055 INFO [train.py:968] (0/2) Epoch 12, batch 12900, giga_loss[loss=0.2686, simple_loss=0.3425, pruned_loss=0.09731, over 28763.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3626, pruned_loss=0.114, over 5651509.86 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3548, pruned_loss=0.09924, over 5729645.02 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3641, pruned_loss=0.1161, over 5641172.40 frames. ], batch size: 284, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 05:03:28,792 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-06 05:03:44,208 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:03:56,175 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:03:58,804 INFO [train.py:968] (0/2) Epoch 12, batch 12950, giga_loss[loss=0.2521, simple_loss=0.3359, pruned_loss=0.0841, over 28470.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3589, pruned_loss=0.1109, over 5646451.99 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3545, pruned_loss=0.09919, over 5732455.47 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3605, pruned_loss=0.1128, over 5634697.12 frames. ], batch size: 60, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 05:04:16,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.396e+02 1.426e+03 1.845e+03 3.037e+03 7.559e+03, threshold=3.690e+03, percent-clipped=17.0
+2023-03-06 05:04:23,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3089, 1.6269, 1.3019, 1.4790], device='cuda:0'), covar=tensor([0.0738, 0.0350, 0.0344, 0.0789], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 05:04:46,027 INFO [train.py:968] (0/2) Epoch 12, batch 13000, giga_loss[loss=0.2862, simple_loss=0.3653, pruned_loss=0.1036, over 28972.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3576, pruned_loss=0.1076, over 5661569.04 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3541, pruned_loss=0.09924, over 5735989.38 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3594, pruned_loss=0.1094, over 5647007.59 frames. ], batch size: 186, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:04:51,066 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-06 05:05:06,816 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:05:37,386 INFO [train.py:968] (0/2) Epoch 12, batch 13050, giga_loss[loss=0.2476, simple_loss=0.337, pruned_loss=0.07909, over 28889.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3582, pruned_loss=0.1078, over 5652758.81 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3539, pruned_loss=0.09928, over 5731697.61 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3599, pruned_loss=0.1094, over 5642229.74 frames. ], batch size: 145, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:05:55,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.854e+02 1.202e+03 1.550e+03 2.026e+03 5.210e+03, threshold=3.100e+03, percent-clipped=4.0
+2023-03-06 05:06:26,623 INFO [train.py:968] (0/2) Epoch 12, batch 13100, libri_loss[loss=0.2784, simple_loss=0.3537, pruned_loss=0.1015, over 29552.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.357, pruned_loss=0.1068, over 5656968.17 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3536, pruned_loss=0.09926, over 5736038.89 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3588, pruned_loss=0.1083, over 5642575.33 frames. ], batch size: 79, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:06:52,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3698, 1.7921, 1.2489, 1.5823], device='cuda:0'), covar=tensor([0.2583, 0.2248, 0.2927, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.1316, 0.0972, 0.1166, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 05:07:15,025 INFO [train.py:968] (0/2) Epoch 12, batch 13150, giga_loss[loss=0.2422, simple_loss=0.3239, pruned_loss=0.08024, over 28351.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3532, pruned_loss=0.1042, over 5653114.42 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3531, pruned_loss=0.09906, over 5737521.45 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3551, pruned_loss=0.1057, over 5638905.43 frames. ], batch size: 71, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:07:20,698 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:07:34,511 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.363e+02 1.371e+03 1.781e+03 2.796e+03 1.253e+04, threshold=3.562e+03, percent-clipped=22.0
+2023-03-06 05:07:36,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0824, 1.5021, 1.3748, 1.0363], device='cuda:0'), covar=tensor([0.1347, 0.2036, 0.1116, 0.1401], device='cuda:0'), in_proj_covar=tensor([0.0832, 0.0690, 0.0870, 0.0776], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 05:07:58,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3792, 5.1988, 4.9176, 2.3370], device='cuda:0'), covar=tensor([0.0438, 0.0604, 0.0781, 0.1892], device='cuda:0'), in_proj_covar=tensor([0.1057, 0.0990, 0.0858, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 05:08:05,208 INFO [train.py:968] (0/2) Epoch 12, batch 13200, giga_loss[loss=0.3118, simple_loss=0.3868, pruned_loss=0.1183, over 28721.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.352, pruned_loss=0.1037, over 5643971.22 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3527, pruned_loss=0.09904, over 5731075.97 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3538, pruned_loss=0.105, over 5637344.20 frames. ], batch size: 284, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:08:07,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4878, 5.3007, 5.0339, 2.5014], device='cuda:0'), covar=tensor([0.0425, 0.0579, 0.0720, 0.1722], device='cuda:0'), in_proj_covar=tensor([0.1056, 0.0988, 0.0857, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 05:08:55,448 INFO [train.py:968] (0/2) Epoch 12, batch 13250, libri_loss[loss=0.2507, simple_loss=0.3356, pruned_loss=0.08285, over 29205.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3513, pruned_loss=0.1029, over 5646240.09 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3524, pruned_loss=0.09896, over 5734336.59 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.353, pruned_loss=0.1041, over 5636068.50 frames. ], batch size: 97, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:09:15,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.313e+02 1.341e+03 1.768e+03 2.415e+03 5.390e+03, threshold=3.537e+03, percent-clipped=7.0
+2023-03-06 05:09:43,381 INFO [train.py:968] (0/2) Epoch 12, batch 13300, giga_loss[loss=0.2503, simple_loss=0.3273, pruned_loss=0.08669, over 27583.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.349, pruned_loss=0.1006, over 5654406.34 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3522, pruned_loss=0.09893, over 5734096.70 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3504, pruned_loss=0.1016, over 5644939.98 frames. ], batch size: 472, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:09:50,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3941, 1.8514, 1.2546, 0.8557], device='cuda:0'), covar=tensor([0.4280, 0.2523, 0.2596, 0.3996], device='cuda:0'), in_proj_covar=tensor([0.1561, 0.1490, 0.1495, 0.1290], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 05:09:54,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:10:08,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:10:16,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:10:30,442 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 05:10:33,656 INFO [train.py:968] (0/2) Epoch 12, batch 13350, giga_loss[loss=0.2362, simple_loss=0.3202, pruned_loss=0.07612, over 28988.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3451, pruned_loss=0.09756, over 5652630.81 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3514, pruned_loss=0.09845, over 5738287.24 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3469, pruned_loss=0.09884, over 5638499.45 frames. ], batch size: 128, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:10:53,613 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.526e+02 1.241e+03 1.426e+03 1.846e+03 3.951e+03, threshold=2.853e+03, percent-clipped=1.0
+2023-03-06 05:10:57,969 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:11:12,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:11:20,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:11:24,873 INFO [train.py:968] (0/2) Epoch 12, batch 13400, giga_loss[loss=0.2556, simple_loss=0.3295, pruned_loss=0.09083, over 28849.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3411, pruned_loss=0.09503, over 5653377.92 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3511, pruned_loss=0.09827, over 5741863.22 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3426, pruned_loss=0.09615, over 5637032.31 frames. ], batch size: 186, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:12:21,188 INFO [train.py:968] (0/2) Epoch 12, batch 13450, giga_loss[loss=0.2624, simple_loss=0.3384, pruned_loss=0.09321, over 28703.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3398, pruned_loss=0.09466, over 5665456.31 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3509, pruned_loss=0.09833, over 5744844.37 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3409, pruned_loss=0.09542, over 5648485.54 frames. ], batch size: 262, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:12:25,027 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=514645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:12:28,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=514648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:12:37,529 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=514657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:12:37,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4865, 1.7618, 1.4330, 1.6251], device='cuda:0'), covar=tensor([0.2706, 0.2319, 0.2660, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.0966, 0.1165, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 05:12:40,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=514660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:12:41,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.344e+02 1.321e+03 1.635e+03 2.066e+03 5.674e+03, threshold=3.271e+03, percent-clipped=11.0
+2023-03-06 05:12:56,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=514677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:13:10,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=514689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:13:12,549 INFO [train.py:968] (0/2) Epoch 12, batch 13500, libri_loss[loss=0.2539, simple_loss=0.323, pruned_loss=0.09243, over 29561.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3396, pruned_loss=0.0955, over 5654251.79 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3507, pruned_loss=0.09832, over 5747834.16 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3404, pruned_loss=0.09605, over 5635839.94 frames. ], batch size: 76, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:13:44,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:13:45,605 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-06 05:13:58,993 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=514731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:14:01,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=514734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:14:12,089 INFO [train.py:968] (0/2) Epoch 12, batch 13550, giga_loss[loss=0.3089, simple_loss=0.3699, pruned_loss=0.124, over 27647.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3409, pruned_loss=0.09598, over 5652032.07 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3501, pruned_loss=0.09793, over 5747860.91 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3419, pruned_loss=0.09671, over 5636264.13 frames. ], batch size: 472, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:14:34,806 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.986e+02 1.455e+03 1.939e+03 3.186e+03 1.021e+04, threshold=3.878e+03, percent-clipped=23.0
+2023-03-06 05:14:35,255 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=514763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:15:09,810 INFO [train.py:968] (0/2) Epoch 12, batch 13600, giga_loss[loss=0.2654, simple_loss=0.3496, pruned_loss=0.09063, over 28339.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3434, pruned_loss=0.09593, over 5655105.02 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3498, pruned_loss=0.09779, over 5750064.62 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3444, pruned_loss=0.09661, over 5639393.66 frames. ], batch size: 368, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:16:08,520 INFO [train.py:968] (0/2) Epoch 12, batch 13650, giga_loss[loss=0.299, simple_loss=0.3709, pruned_loss=0.1135, over 28866.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3437, pruned_loss=0.09561, over 5660525.30 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3499, pruned_loss=0.09792, over 5744213.79 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3443, pruned_loss=0.09598, over 5650242.84 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:16:30,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.139e+02 1.405e+03 1.921e+03 3.116e+03 9.572e+03, threshold=3.842e+03, percent-clipped=14.0
+2023-03-06 05:16:32,816 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=514864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:16:37,424 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=514867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:16:49,063 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 05:17:07,775 INFO [train.py:968] (0/2) Epoch 12, batch 13700, giga_loss[loss=0.2339, simple_loss=0.3155, pruned_loss=0.07609, over 29113.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3415, pruned_loss=0.09443, over 5666675.70 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3491, pruned_loss=0.09773, over 5747034.08 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3425, pruned_loss=0.09484, over 5654368.35 frames. ], batch size: 200, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:17:13,001 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=514896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:17:13,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:17:52,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-06 05:18:09,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:18:11,146 INFO [train.py:968] (0/2) Epoch 12, batch 13750, giga_loss[loss=0.2505, simple_loss=0.3354, pruned_loss=0.08282, over 28464.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3404, pruned_loss=0.09322, over 5656511.20 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3491, pruned_loss=0.09774, over 5737143.36 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3411, pruned_loss=0.09349, over 5655093.11 frames. ], batch size: 369, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:18:28,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:18:39,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.230e+02 1.289e+03 1.586e+03 2.222e+03 5.811e+03, threshold=3.172e+03, percent-clipped=5.0
+2023-03-06 05:19:12,826 INFO [train.py:968] (0/2) Epoch 12, batch 13800, giga_loss[loss=0.2669, simple_loss=0.34, pruned_loss=0.09693, over 28950.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3396, pruned_loss=0.09195, over 5657685.74 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3491, pruned_loss=0.09796, over 5730319.36 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.34, pruned_loss=0.09187, over 5661611.37 frames. ], batch size: 199, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:19:14,128 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.22 vs. limit=5.0
+2023-03-06 05:20:14,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=515040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:20:15,915 INFO [train.py:968] (0/2) Epoch 12, batch 13850, giga_loss[loss=0.2517, simple_loss=0.3276, pruned_loss=0.08786, over 28944.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3362, pruned_loss=0.09126, over 5656007.88 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3489, pruned_loss=0.09795, over 5730888.37 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3366, pruned_loss=0.09114, over 5657486.81 frames. ], batch size: 155, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:20:17,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=515043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:20:41,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.898e+02 1.214e+03 1.534e+03 2.233e+03 8.472e+03, threshold=3.068e+03, percent-clipped=9.0
+2023-03-06 05:20:49,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5005, 1.4662, 1.1493, 1.5566], device='cuda:0'), covar=tensor([0.0726, 0.0298, 0.0339, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0112, 0.0115, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0058, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 05:20:52,223 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=515072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:05,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=515082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:07,965 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=515085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:07,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4953, 1.5554, 1.3118, 1.6021], device='cuda:0'), covar=tensor([0.2520, 0.2462, 0.2752, 0.2296], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.0964, 0.1163, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 05:21:15,322 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-06 05:21:17,653 INFO [train.py:968] (0/2) Epoch 12, batch 13900, giga_loss[loss=0.2934, simple_loss=0.3566, pruned_loss=0.1151, over 27675.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3349, pruned_loss=0.09099, over 5658395.54 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3484, pruned_loss=0.0977, over 5733355.67 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3355, pruned_loss=0.09104, over 5656447.64 frames. ], batch size: 472, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:21:23,022 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=515098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:25,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=515101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:39,591 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=515114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:59,656 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=515130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:22:13,257 INFO [train.py:968] (0/2) Epoch 12, batch 13950, libri_loss[loss=0.2393, simple_loss=0.3181, pruned_loss=0.08026, over 29562.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3346, pruned_loss=0.09096, over 5660517.89 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3478, pruned_loss=0.0975, over 5736612.80 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3352, pruned_loss=0.091, over 5654216.47 frames. ], batch size: 75, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:22:18,313 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:22:38,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.859e+02 1.236e+03 1.615e+03 2.217e+03 8.568e+03, threshold=3.229e+03, percent-clipped=10.0
+2023-03-06 05:22:55,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2954, 1.5795, 1.5804, 1.2210], device='cuda:0'), covar=tensor([0.1492, 0.2184, 0.1255, 0.1538], device='cuda:0'), in_proj_covar=tensor([0.0833, 0.0688, 0.0874, 0.0778], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 05:23:18,018 INFO [train.py:968] (0/2) Epoch 12, batch 14000, giga_loss[loss=0.2377, simple_loss=0.3258, pruned_loss=0.07475, over 28483.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3367, pruned_loss=0.09135, over 5652469.22 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3477, pruned_loss=0.09742, over 5738428.83 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3371, pruned_loss=0.09138, over 5645363.39 frames. ], batch size: 336, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:23:50,611 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-06 05:24:01,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1456, 3.2462, 1.3115, 1.4217], device='cuda:0'), covar=tensor([0.1179, 0.0347, 0.1012, 0.1477], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0508, 0.0344, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 05:24:21,040 INFO [train.py:968] (0/2) Epoch 12, batch 14050, libri_loss[loss=0.2419, simple_loss=0.3169, pruned_loss=0.08344, over 29565.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3359, pruned_loss=0.09077, over 5653733.39 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.347, pruned_loss=0.09722, over 5731410.98 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3367, pruned_loss=0.09084, over 5652098.38 frames. ], batch size: 77, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:24:49,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.121e+02 1.302e+03 1.631e+03 2.365e+03 7.532e+03, threshold=3.262e+03, percent-clipped=11.0
+2023-03-06 05:25:27,506 INFO [train.py:968] (0/2) Epoch 12, batch 14100, libri_loss[loss=0.2352, simple_loss=0.3104, pruned_loss=0.08001, over 29668.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3336, pruned_loss=0.08926, over 5658438.00 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3466, pruned_loss=0.09702, over 5726051.73 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3343, pruned_loss=0.08934, over 5660410.92 frames. ], batch size: 73, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:26:20,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8878, 1.1247, 1.2523, 0.9386], device='cuda:0'), covar=tensor([0.1405, 0.1319, 0.1792, 0.1533], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0711, 0.0658, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-06 05:26:24,719 INFO [train.py:968] (0/2) Epoch 12, batch 14150, giga_loss[loss=0.2402, simple_loss=0.3339, pruned_loss=0.07325, over 28886.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.335, pruned_loss=0.08978, over 5678979.84 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3457, pruned_loss=0.09659, over 5732761.48 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3359, pruned_loss=0.08993, over 5672108.41 frames. ], batch size: 145, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:26:56,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.634e+02 1.384e+03 2.124e+03 3.002e+03 9.679e+03, threshold=4.248e+03, percent-clipped=18.0
+2023-03-06 05:27:17,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2773, 5.0626, 4.8289, 2.1564], device='cuda:0'), covar=tensor([0.0359, 0.0532, 0.0550, 0.2059], device='cuda:0'), in_proj_covar=tensor([0.1051, 0.0978, 0.0851, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 05:27:32,827 INFO [train.py:968] (0/2) Epoch 12, batch 14200, giga_loss[loss=0.2383, simple_loss=0.3385, pruned_loss=0.06903, over 28386.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3391, pruned_loss=0.08982, over 5672740.84 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3455, pruned_loss=0.09654, over 5735620.16 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3399, pruned_loss=0.0899, over 5664172.21 frames. ], batch size: 336, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:28:29,569 INFO [train.py:968] (0/2) Epoch 12, batch 14250, giga_loss[loss=0.2862, simple_loss=0.3717, pruned_loss=0.1003, over 28712.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3411, pruned_loss=0.08923, over 5680538.45 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3451, pruned_loss=0.09638, over 5741402.16 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3419, pruned_loss=0.08919, over 5666589.49 frames. ], batch size: 262, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:28:57,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.765e+02 1.426e+03 1.842e+03 2.649e+03 8.507e+03, threshold=3.683e+03, percent-clipped=5.0
+2023-03-06 05:29:28,255 INFO [train.py:968] (0/2) Epoch 12, batch 14300, giga_loss[loss=0.2468, simple_loss=0.3373, pruned_loss=0.0781, over 28758.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3419, pruned_loss=0.08875, over 5679385.76 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3452, pruned_loss=0.09646, over 5744630.22 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3424, pruned_loss=0.08845, over 5663647.73 frames. ], batch size: 243, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:29:33,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2550, 1.9699, 1.4958, 1.3418], device='cuda:0'), covar=tensor([0.0810, 0.0324, 0.0305, 0.1004], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:0')
+2023-03-06 05:29:59,893 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-06 05:30:03,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=515521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:30:19,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5802, 1.4915, 1.7457, 1.3009], device='cuda:0'), covar=tensor([0.1965, 0.3072, 0.1521, 0.1687], device='cuda:0'), in_proj_covar=tensor([0.0831, 0.0686, 0.0873, 0.0776], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 05:30:31,815 INFO [train.py:968] (0/2) Epoch 12, batch 14350, giga_loss[loss=0.2547, simple_loss=0.3229, pruned_loss=0.09327, over 24351.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3425, pruned_loss=0.08961, over 5669296.52 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.345, pruned_loss=0.09642, over 5737680.70 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08929, over 5661485.29 frames. ], batch size: 705, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:30:49,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:30:57,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8844, 2.9207, 1.7878, 1.0364], device='cuda:0'), covar=tensor([0.4922, 0.2076, 0.2895, 0.3945], device='cuda:0'), in_proj_covar=tensor([0.1573, 0.1498, 0.1510, 0.1296], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 05:30:57,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.600e+02 1.278e+03 1.610e+03 2.162e+03 5.345e+03, threshold=3.219e+03, percent-clipped=7.0
+2023-03-06 05:31:32,018 INFO [train.py:968] (0/2) Epoch 12, batch 14400, giga_loss[loss=0.2119, simple_loss=0.2971, pruned_loss=0.06332, over 29069.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3419, pruned_loss=0.09026, over 5678729.20 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3447, pruned_loss=0.09631, over 5741827.65 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3425, pruned_loss=0.08997, over 5667228.11 frames. ], batch size: 128, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:32:22,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7013, 1.7419, 1.2387, 1.4100], device='cuda:0'), covar=tensor([0.0747, 0.0516, 0.0971, 0.0919], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0436, 0.0499, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 05:32:37,840 INFO [train.py:968] (0/2) Epoch 12, batch 14450, giga_loss[loss=0.2856, simple_loss=0.3582, pruned_loss=0.1066, over 28958.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3423, pruned_loss=0.09143, over 5692075.56 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3439, pruned_loss=0.0959, over 5745062.35 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3434, pruned_loss=0.09143, over 5678956.94 frames. ], batch size: 213, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:32:50,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0667, 1.3175, 3.5064, 2.9400], device='cuda:0'), covar=tensor([0.1891, 0.2617, 0.0871, 0.1124], device='cuda:0'), in_proj_covar=tensor([0.0662, 0.0590, 0.0858, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 05:33:12,571 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=515664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:33:12,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.551e+02 1.302e+03 1.707e+03 2.294e+03 5.758e+03, threshold=3.415e+03, percent-clipped=8.0
+2023-03-06 05:33:15,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=515667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:33:57,290 INFO [train.py:968] (0/2) Epoch 12, batch 14500, giga_loss[loss=0.2528, simple_loss=0.3364, pruned_loss=0.08463, over 28941.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3417, pruned_loss=0.09216, over 5687350.65 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3439, pruned_loss=0.09596, over 5748274.04 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3426, pruned_loss=0.092, over 5672415.15 frames. ], batch size: 285, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:34:07,640 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=515696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:35:07,357 INFO [train.py:968] (0/2) Epoch 12, batch 14550, giga_loss[loss=0.2677, simple_loss=0.3543, pruned_loss=0.09052, over 28850.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3388, pruned_loss=0.09045, over 5690314.24 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3441, pruned_loss=0.0962, over 5749163.30 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3392, pruned_loss=0.08992, over 5675385.13 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:35:33,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.338e+02 1.230e+03 1.622e+03 2.267e+03 5.621e+03, threshold=3.244e+03, percent-clipped=10.0
+2023-03-06 05:35:51,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2155, 4.0379, 3.8331, 1.7049], device='cuda:0'), covar=tensor([0.0534, 0.0718, 0.0763, 0.2154], device='cuda:0'), in_proj_covar=tensor([0.1053, 0.0979, 0.0857, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 05:36:11,000 INFO [train.py:968] (0/2) Epoch 12, batch 14600, giga_loss[loss=0.232, simple_loss=0.3151, pruned_loss=0.07445, over 28158.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3369, pruned_loss=0.08979, over 5677305.00 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3439, pruned_loss=0.09629, over 5739750.68 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3372, pruned_loss=0.08916, over 5672946.75 frames. ], batch size: 412, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:36:20,846 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:36:35,270 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:37:15,092 INFO [train.py:968] (0/2) Epoch 12, batch 14650, giga_loss[loss=0.2608, simple_loss=0.3381, pruned_loss=0.0917, over 28839.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3363, pruned_loss=0.08983, over 5673813.27 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3435, pruned_loss=0.09611, over 5742122.08 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3368, pruned_loss=0.0894, over 5667580.57 frames. ], batch size: 119, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:37:23,541 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:37:49,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.674e+02 1.230e+03 1.674e+03 2.505e+03 6.196e+03, threshold=3.348e+03, percent-clipped=12.0
+2023-03-06 05:38:24,382 INFO [train.py:968] (0/2) Epoch 12, batch 14700, giga_loss[loss=0.2234, simple_loss=0.318, pruned_loss=0.0644, over 29029.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3405, pruned_loss=0.09179, over 5676238.76 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3435, pruned_loss=0.09611, over 5742122.08 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3408, pruned_loss=0.09146, over 5671387.77 frames. ], batch size: 145, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:38:33,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1150, 3.2993, 2.2225, 1.2671], device='cuda:0'), covar=tensor([0.5022, 0.2340, 0.2779, 0.4529], device='cuda:0'), in_proj_covar=tensor([0.1571, 0.1497, 0.1509, 0.1295], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 05:38:39,906 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:39:10,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4749, 1.7444, 1.6539, 1.4716], device='cuda:0'), covar=tensor([0.2296, 0.1615, 0.1247, 0.1557], device='cuda:0'), in_proj_covar=tensor([0.1685, 0.1580, 0.1531, 0.1653], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 05:39:12,222 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=515931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:39:26,295 INFO [train.py:968] (0/2) Epoch 12, batch 14750, giga_loss[loss=0.263, simple_loss=0.3354, pruned_loss=0.09528, over 28972.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3385, pruned_loss=0.09197, over 5677331.70 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.343, pruned_loss=0.09579, over 5742776.55 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3393, pruned_loss=0.09191, over 5671489.17 frames. ], batch size: 186, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:39:54,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5477, 2.2367, 1.5928, 0.7383], device='cuda:0'), covar=tensor([0.4282, 0.2129, 0.3234, 0.4624], device='cuda:0'), in_proj_covar=tensor([0.1569, 0.1497, 0.1510, 0.1294], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 05:39:54,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-06 05:39:58,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.297e+02 1.371e+03 1.906e+03 2.373e+03 8.875e+03, threshold=3.811e+03, percent-clipped=10.0
+2023-03-06 05:40:22,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4985, 4.3337, 4.0865, 1.7928], device='cuda:0'), covar=tensor([0.0531, 0.0695, 0.0802, 0.2187], device='cuda:0'), in_proj_covar=tensor([0.1040, 0.0966, 0.0847, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 05:40:30,402 INFO [train.py:968] (0/2) Epoch 12, batch 14800, giga_loss[loss=0.2706, simple_loss=0.3431, pruned_loss=0.099, over 28879.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3388, pruned_loss=0.09283, over 5679302.65 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3431, pruned_loss=0.0959, over 5742816.48 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3392, pruned_loss=0.09262, over 5673550.63 frames. ], batch size: 112, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:40:39,084 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-516000.pt
+2023-03-06 05:40:42,081 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=516001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:41:28,143 INFO [train.py:968] (0/2) Epoch 12, batch 14850, giga_loss[loss=0.3131, simple_loss=0.3852, pruned_loss=0.1205, over 28409.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3403, pruned_loss=0.09373, over 5687757.26 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3427, pruned_loss=0.09567, over 5746450.55 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3408, pruned_loss=0.09369, over 5677869.39 frames. ], batch size: 336, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:42:01,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.199e+02 1.502e+03 2.103e+03 3.191e+03 9.543e+03, threshold=4.205e+03, percent-clipped=16.0
+2023-03-06 05:42:11,467 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:42:15,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:42:36,216 INFO [train.py:968] (0/2) Epoch 12, batch 14900, libri_loss[loss=0.2725, simple_loss=0.3526, pruned_loss=0.09617, over 29657.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3432, pruned_loss=0.09432, over 5686636.65 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3429, pruned_loss=0.09575, over 5745909.18 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3435, pruned_loss=0.09418, over 5678210.46 frames. ], batch size: 88, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:42:56,811 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:43:15,960 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-06 05:43:53,031 INFO [train.py:968] (0/2) Epoch 12, batch 14950, giga_loss[loss=0.2422, simple_loss=0.3192, pruned_loss=0.0826, over 28857.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3426, pruned_loss=0.09371, over 5674479.02 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3426, pruned_loss=0.09563, over 5748235.21 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.343, pruned_loss=0.09366, over 5664647.34 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:44:29,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.250e+02 1.345e+03 1.932e+03 2.721e+03 5.988e+03, threshold=3.864e+03, percent-clipped=2.0
+2023-03-06 05:44:30,152 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=516166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:44:36,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4182, 1.6987, 1.7432, 1.3308], device='cuda:0'), covar=tensor([0.1557, 0.2237, 0.1292, 0.1587], device='cuda:0'), in_proj_covar=tensor([0.0828, 0.0686, 0.0872, 0.0777], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 05:44:39,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:45:05,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:45:11,465 INFO [train.py:968] (0/2) Epoch 12, batch 15000, giga_loss[loss=0.2648, simple_loss=0.3329, pruned_loss=0.09832, over 28834.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3402, pruned_loss=0.09369, over 5663230.99 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3425, pruned_loss=0.0956, over 5741781.40 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3406, pruned_loss=0.09365, over 5659502.59 frames. ], batch size: 174, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:45:11,470 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 05:45:20,525 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.205, simple_loss=0.3042, pruned_loss=0.05287, over 944034.00 frames. 
+2023-03-06 05:45:20,526 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 05:45:52,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2261, 4.0481, 3.8257, 1.9803], device='cuda:0'), covar=tensor([0.0518, 0.0666, 0.0708, 0.2146], device='cuda:0'), in_proj_covar=tensor([0.1035, 0.0962, 0.0843, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 05:46:00,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:46:26,685 INFO [train.py:968] (0/2) Epoch 12, batch 15050, giga_loss[loss=0.2079, simple_loss=0.2956, pruned_loss=0.06006, over 29065.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3336, pruned_loss=0.09064, over 5665033.49 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3422, pruned_loss=0.09539, over 5745587.00 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3341, pruned_loss=0.09071, over 5657185.50 frames. ], batch size: 285, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:46:52,922 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.646e+02 1.308e+03 1.640e+03 2.214e+03 5.629e+03, threshold=3.280e+03, percent-clipped=7.0
+2023-03-06 05:47:07,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:47:18,077 INFO [train.py:968] (0/2) Epoch 12, batch 15100, giga_loss[loss=0.2639, simple_loss=0.3428, pruned_loss=0.09252, over 28324.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3322, pruned_loss=0.09012, over 5676815.27 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.342, pruned_loss=0.09549, over 5751877.12 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3323, pruned_loss=0.08985, over 5660981.52 frames. ], batch size: 368, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:47:36,130 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=516304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:47:49,679 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:47:53,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:04,659 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:07,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:15,424 INFO [train.py:968] (0/2) Epoch 12, batch 15150, giga_loss[loss=0.2658, simple_loss=0.342, pruned_loss=0.0948, over 28763.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3326, pruned_loss=0.09062, over 5674043.19 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3411, pruned_loss=0.0949, over 5756190.98 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3333, pruned_loss=0.09082, over 5654992.93 frames. ], batch size: 243, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:48:22,849 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:25,816 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-06 05:48:39,414 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:44,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.427e+02 1.517e+03 2.038e+03 2.909e+03 9.320e+03, threshold=4.076e+03, percent-clipped=17.0
+2023-03-06 05:48:44,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:47,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:52,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:49:09,854 INFO [train.py:968] (0/2) Epoch 12, batch 15200, giga_loss[loss=0.2429, simple_loss=0.3165, pruned_loss=0.08461, over 27559.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3322, pruned_loss=0.09014, over 5681764.53 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3404, pruned_loss=0.09462, over 5760532.79 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3331, pruned_loss=0.09046, over 5660169.11 frames. ], batch size: 472, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:49:22,441 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:49:50,727 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:49:55,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:50:14,359 INFO [train.py:968] (0/2) Epoch 12, batch 15250, giga_loss[loss=0.2545, simple_loss=0.3349, pruned_loss=0.087, over 28947.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3304, pruned_loss=0.08837, over 5658485.33 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3405, pruned_loss=0.09464, over 5752577.79 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.331, pruned_loss=0.08855, over 5647623.37 frames. ], batch size: 186, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:50:29,887 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:50:45,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.450e+02 1.355e+03 1.721e+03 2.814e+03 7.595e+03, threshold=3.443e+03, percent-clipped=11.0
+2023-03-06 05:51:20,917 INFO [train.py:968] (0/2) Epoch 12, batch 15300, giga_loss[loss=0.2452, simple_loss=0.3208, pruned_loss=0.08482, over 29034.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3288, pruned_loss=0.08731, over 5671611.44 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.34, pruned_loss=0.09448, over 5755026.93 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3295, pruned_loss=0.08749, over 5659613.90 frames. ], batch size: 199, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:51:23,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=516494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:52:02,504 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:52:03,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-06 05:52:05,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:52:14,306 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=516530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:52:30,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:52:30,723 INFO [train.py:968] (0/2) Epoch 12, batch 15350, giga_loss[loss=0.219, simple_loss=0.2883, pruned_loss=0.07479, over 24624.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.328, pruned_loss=0.0871, over 5663884.45 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.34, pruned_loss=0.09469, over 5759214.63 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3283, pruned_loss=0.08688, over 5648555.64 frames. ], batch size: 705, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:52:42,320 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:53:05,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.234e+02 1.201e+03 1.591e+03 2.231e+03 6.154e+03, threshold=3.182e+03, percent-clipped=8.0
+2023-03-06 05:53:08,201 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3958, 1.7144, 1.3269, 1.6853], device='cuda:0'), covar=tensor([0.2423, 0.2291, 0.2671, 0.2023], device='cuda:0'), in_proj_covar=tensor([0.1308, 0.0963, 0.1158, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 05:53:33,883 INFO [train.py:968] (0/2) Epoch 12, batch 15400, libri_loss[loss=0.2697, simple_loss=0.3446, pruned_loss=0.09745, over 29659.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3286, pruned_loss=0.08703, over 5664564.92 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3401, pruned_loss=0.09474, over 5762846.94 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3285, pruned_loss=0.08666, over 5647267.10 frames. ], batch size: 88, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:54:38,874 INFO [train.py:968] (0/2) Epoch 12, batch 15450, giga_loss[loss=0.2365, simple_loss=0.317, pruned_loss=0.07798, over 28982.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3288, pruned_loss=0.08769, over 5671332.94 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3394, pruned_loss=0.09441, over 5765621.45 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3292, pruned_loss=0.08757, over 5653592.05 frames. ], batch size: 145, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:55:10,123 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-06 05:55:10,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.052e+02 1.256e+03 1.542e+03 2.103e+03 6.874e+03, threshold=3.084e+03, percent-clipped=11.0
+2023-03-06 05:55:25,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:55:36,180 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:55:38,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:55:45,388 INFO [train.py:968] (0/2) Epoch 12, batch 15500, giga_loss[loss=0.2424, simple_loss=0.3242, pruned_loss=0.08034, over 28955.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3274, pruned_loss=0.08685, over 5665281.19 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3391, pruned_loss=0.09422, over 5767123.21 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3278, pruned_loss=0.08685, over 5649318.69 frames. ], batch size: 213, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:56:11,410 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:56:30,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8542, 1.1500, 1.0446, 0.7875], device='cuda:0'), covar=tensor([0.1546, 0.1585, 0.0997, 0.1483], device='cuda:0'), in_proj_covar=tensor([0.1706, 0.1594, 0.1542, 0.1654], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 05:56:38,716 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-06 05:56:39,531 INFO [train.py:968] (0/2) Epoch 12, batch 15550, giga_loss[loss=0.264, simple_loss=0.3504, pruned_loss=0.08884, over 28909.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3293, pruned_loss=0.08641, over 5677930.87 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3385, pruned_loss=0.09393, over 5768558.49 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3298, pruned_loss=0.08644, over 5660681.69 frames. ], batch size: 199, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:56:59,251 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 05:57:10,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.213e+02 1.232e+03 1.496e+03 1.940e+03 5.431e+03, threshold=2.991e+03, percent-clipped=3.0
+2023-03-06 05:57:43,739 INFO [train.py:968] (0/2) Epoch 12, batch 15600, giga_loss[loss=0.254, simple_loss=0.3426, pruned_loss=0.08266, over 28854.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3317, pruned_loss=0.08693, over 5668305.49 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3384, pruned_loss=0.09386, over 5770051.58 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3322, pruned_loss=0.08695, over 5652788.34 frames. ], batch size: 145, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:57:50,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3897, 1.7222, 1.4585, 1.2792], device='cuda:0'), covar=tensor([0.1942, 0.1446, 0.1110, 0.1479], device='cuda:0'), in_proj_covar=tensor([0.1707, 0.1594, 0.1540, 0.1652], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 05:58:23,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:58:26,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:58:43,981 INFO [train.py:968] (0/2) Epoch 12, batch 15650, giga_loss[loss=0.2629, simple_loss=0.3506, pruned_loss=0.08758, over 28953.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3341, pruned_loss=0.08791, over 5669416.88 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3383, pruned_loss=0.09384, over 5770168.74 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3345, pruned_loss=0.08785, over 5655591.71 frames. ], batch size: 136, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:58:59,684 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:59:17,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.447e+02 1.400e+03 1.746e+03 2.525e+03 7.600e+03, threshold=3.493e+03, percent-clipped=13.0
+2023-03-06 05:59:18,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:59:44,850 INFO [train.py:968] (0/2) Epoch 12, batch 15700, giga_loss[loss=0.2866, simple_loss=0.3531, pruned_loss=0.11, over 28987.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3351, pruned_loss=0.08827, over 5681939.44 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3386, pruned_loss=0.09401, over 5771975.52 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3351, pruned_loss=0.088, over 5668165.30 frames. ], batch size: 213, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:00:02,091 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:00:44,905 INFO [train.py:968] (0/2) Epoch 12, batch 15750, giga_loss[loss=0.2186, simple_loss=0.3043, pruned_loss=0.06646, over 28932.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3325, pruned_loss=0.08661, over 5690408.24 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3386, pruned_loss=0.09411, over 5773699.39 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3324, pruned_loss=0.08622, over 5676753.05 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:01:17,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.714e+02 1.220e+03 1.552e+03 2.037e+03 4.269e+03, threshold=3.103e+03, percent-clipped=3.0
+2023-03-06 06:01:50,952 INFO [train.py:968] (0/2) Epoch 12, batch 15800, giga_loss[loss=0.2736, simple_loss=0.3501, pruned_loss=0.09851, over 28457.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.331, pruned_loss=0.08571, over 5681886.95 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3384, pruned_loss=0.09399, over 5766417.67 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.331, pruned_loss=0.0854, over 5676236.35 frames. ], batch size: 336, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:02:15,549 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=517012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:02:17,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=517015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:02:46,591 INFO [train.py:968] (0/2) Epoch 12, batch 15850, giga_loss[loss=0.2469, simple_loss=0.3281, pruned_loss=0.08281, over 28948.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3296, pruned_loss=0.08593, over 5673114.88 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3387, pruned_loss=0.09427, over 5757196.06 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3292, pruned_loss=0.08521, over 5675062.35 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:02:50,099 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=517044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:02:55,999 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=517048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:03:00,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=517051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:03:19,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.165e+02 1.300e+03 1.832e+03 2.461e+03 6.534e+03, threshold=3.665e+03, percent-clipped=19.0
+2023-03-06 06:03:19,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=517068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:03:33,770 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=517080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:03:43,744 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-03-06 06:03:46,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5699, 1.8312, 1.8925, 1.3878], device='cuda:0'), covar=tensor([0.1780, 0.2315, 0.1446, 0.1727], device='cuda:0'), in_proj_covar=tensor([0.0827, 0.0681, 0.0868, 0.0775], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 06:03:46,992 INFO [train.py:968] (0/2) Epoch 12, batch 15900, giga_loss[loss=0.2626, simple_loss=0.3466, pruned_loss=0.0893, over 29081.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3304, pruned_loss=0.08654, over 5667166.71 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3381, pruned_loss=0.09398, over 5755272.78 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3303, pruned_loss=0.08602, over 5667939.67 frames. ], batch size: 285, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:04:14,372 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7938, 1.9640, 1.2697, 1.5478], device='cuda:0'), covar=tensor([0.0784, 0.0534, 0.1011, 0.1032], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0435, 0.0500, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:04:53,675 INFO [train.py:968] (0/2) Epoch 12, batch 15950, giga_loss[loss=0.256, simple_loss=0.3157, pruned_loss=0.09818, over 24885.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3327, pruned_loss=0.08788, over 5671876.33 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3383, pruned_loss=0.09405, over 5756789.19 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3325, pruned_loss=0.08737, over 5670324.91 frames. ], batch size: 705, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:05:26,995 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.221e+02 1.444e+03 1.783e+03 2.557e+03 1.592e+04, threshold=3.565e+03, percent-clipped=11.0
+2023-03-06 06:05:54,970 INFO [train.py:968] (0/2) Epoch 12, batch 16000, libri_loss[loss=0.2386, simple_loss=0.3206, pruned_loss=0.07829, over 29508.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3336, pruned_loss=0.08882, over 5674915.47 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3382, pruned_loss=0.09402, over 5761324.02 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3333, pruned_loss=0.08826, over 5666886.34 frames. ], batch size: 81, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 06:06:55,616 INFO [train.py:968] (0/2) Epoch 12, batch 16050, giga_loss[loss=0.2852, simple_loss=0.3629, pruned_loss=0.1037, over 28965.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3372, pruned_loss=0.0908, over 5674963.63 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3381, pruned_loss=0.094, over 5759239.84 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3372, pruned_loss=0.09035, over 5669909.89 frames. ], batch size: 199, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:07:28,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.741e+02 1.392e+03 2.034e+03 3.172e+03 7.766e+03, threshold=4.068e+03, percent-clipped=23.0
+2023-03-06 06:07:55,363 INFO [train.py:968] (0/2) Epoch 12, batch 16100, giga_loss[loss=0.2974, simple_loss=0.3753, pruned_loss=0.1097, over 28529.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3408, pruned_loss=0.09205, over 5680470.61 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.338, pruned_loss=0.09408, over 5760970.09 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3408, pruned_loss=0.09161, over 5674262.71 frames. ], batch size: 370, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:08:57,062 INFO [train.py:968] (0/2) Epoch 12, batch 16150, giga_loss[loss=0.2614, simple_loss=0.3393, pruned_loss=0.09171, over 28992.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3405, pruned_loss=0.09198, over 5690405.33 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3373, pruned_loss=0.09367, over 5765896.48 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3413, pruned_loss=0.09191, over 5677991.72 frames. ], batch size: 199, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:09:35,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.729e+02 1.238e+03 1.548e+03 2.242e+03 4.095e+03, threshold=3.096e+03, percent-clipped=1.0
+2023-03-06 06:09:57,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7624, 1.6948, 1.2556, 1.2974], device='cuda:0'), covar=tensor([0.0711, 0.0556, 0.0956, 0.1093], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0436, 0.0502, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:10:04,071 INFO [train.py:968] (0/2) Epoch 12, batch 16200, giga_loss[loss=0.2694, simple_loss=0.3358, pruned_loss=0.1015, over 26891.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.338, pruned_loss=0.09067, over 5691769.62 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3372, pruned_loss=0.09356, over 5765104.62 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3387, pruned_loss=0.09066, over 5680837.92 frames. ], batch size: 555, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:11:01,030 INFO [train.py:968] (0/2) Epoch 12, batch 16250, giga_loss[loss=0.2571, simple_loss=0.3368, pruned_loss=0.0887, over 28943.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3375, pruned_loss=0.09094, over 5687669.68 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3366, pruned_loss=0.09326, over 5762016.02 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3386, pruned_loss=0.09108, over 5678628.29 frames. ], batch size: 186, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:11:04,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=517443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:11:30,299 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-06 06:11:30,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5127, 1.8685, 1.6134, 1.4219], device='cuda:0'), covar=tensor([0.2272, 0.1618, 0.1476, 0.1788], device='cuda:0'), in_proj_covar=tensor([0.1726, 0.1615, 0.1557, 0.1668], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 06:11:34,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.940e+02 1.456e+03 1.825e+03 2.427e+03 1.089e+04, threshold=3.651e+03, percent-clipped=16.0
+2023-03-06 06:12:02,608 INFO [train.py:968] (0/2) Epoch 12, batch 16300, giga_loss[loss=0.265, simple_loss=0.3405, pruned_loss=0.09472, over 28537.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3356, pruned_loss=0.08991, over 5674928.08 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3366, pruned_loss=0.0932, over 5762243.70 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3366, pruned_loss=0.09002, over 5664916.50 frames. ], batch size: 336, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:12:13,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7342, 1.8702, 1.2588, 1.4870], device='cuda:0'), covar=tensor([0.0739, 0.0588, 0.1035, 0.1068], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0434, 0.0499, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:12:13,909 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-06 06:12:28,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2768, 1.6579, 1.2920, 0.9788], device='cuda:0'), covar=tensor([0.2509, 0.2378, 0.2808, 0.2094], device='cuda:0'), in_proj_covar=tensor([0.1308, 0.0964, 0.1158, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 06:13:06,293 INFO [train.py:968] (0/2) Epoch 12, batch 16350, giga_loss[loss=0.2155, simple_loss=0.2857, pruned_loss=0.07262, over 28721.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.334, pruned_loss=0.09025, over 5675945.97 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3366, pruned_loss=0.09323, over 5764601.98 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3347, pruned_loss=0.09027, over 5665000.23 frames. ], batch size: 78, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:13:37,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.019e+02 1.226e+03 1.573e+03 2.153e+03 5.339e+03, threshold=3.147e+03, percent-clipped=9.0
+2023-03-06 06:13:57,768 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=517586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:14:00,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=517589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:14:03,367 INFO [train.py:968] (0/2) Epoch 12, batch 16400, giga_loss[loss=0.2352, simple_loss=0.3246, pruned_loss=0.07297, over 28915.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3345, pruned_loss=0.09055, over 5680717.59 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3366, pruned_loss=0.09321, over 5765427.86 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3351, pruned_loss=0.09051, over 5669003.81 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 06:14:21,289 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-06 06:14:40,695 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=517618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:14:59,842 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.01 vs. limit=5.0
+2023-03-06 06:15:06,207 INFO [train.py:968] (0/2) Epoch 12, batch 16450, libri_loss[loss=0.2726, simple_loss=0.3417, pruned_loss=0.1017, over 29566.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3329, pruned_loss=0.08845, over 5675703.32 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3364, pruned_loss=0.09322, over 5766559.95 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3335, pruned_loss=0.08834, over 5664231.66 frames. ], batch size: 75, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 06:15:34,375 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.481e+02 1.340e+03 1.630e+03 2.108e+03 4.055e+03, threshold=3.259e+03, percent-clipped=4.0
+2023-03-06 06:15:59,485 INFO [train.py:968] (0/2) Epoch 12, batch 16500, giga_loss[loss=0.245, simple_loss=0.3366, pruned_loss=0.07668, over 28178.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3336, pruned_loss=0.08784, over 5685955.36 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3362, pruned_loss=0.09313, over 5771832.91 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3341, pruned_loss=0.08762, over 5668243.86 frames. ], batch size: 412, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 06:16:10,269 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=517701.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:16:36,110 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 06:16:53,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9196, 1.2553, 2.8655, 2.7419], device='cuda:0'), covar=tensor([0.1572, 0.2353, 0.0553, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.0663, 0.0585, 0.0847, 0.0752], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:16:54,235 INFO [train.py:968] (0/2) Epoch 12, batch 16550, giga_loss[loss=0.3403, simple_loss=0.4148, pruned_loss=0.1329, over 28504.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3354, pruned_loss=0.08733, over 5672880.53 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.336, pruned_loss=0.09298, over 5764490.86 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.336, pruned_loss=0.08715, over 5662594.00 frames. ], batch size: 336, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:17:22,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0784, 1.2298, 3.7768, 3.1344], device='cuda:0'), covar=tensor([0.1757, 0.2696, 0.0410, 0.0968], device='cuda:0'), in_proj_covar=tensor([0.0663, 0.0584, 0.0847, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:17:27,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.421e+02 1.352e+03 1.589e+03 2.410e+03 9.484e+03, threshold=3.179e+03, percent-clipped=10.0
+2023-03-06 06:17:35,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5913, 1.9870, 1.5704, 1.9067], device='cuda:0'), covar=tensor([0.0735, 0.0250, 0.0300, 0.0754], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:0')
+2023-03-06 06:17:51,461 INFO [train.py:968] (0/2) Epoch 12, batch 16600, giga_loss[loss=0.2414, simple_loss=0.3319, pruned_loss=0.07551, over 28545.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3362, pruned_loss=0.08684, over 5685248.95 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.336, pruned_loss=0.09291, over 5765020.37 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3366, pruned_loss=0.08671, over 5675933.45 frames. ], batch size: 78, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:18:57,308 INFO [train.py:968] (0/2) Epoch 12, batch 16650, giga_loss[loss=0.2631, simple_loss=0.3458, pruned_loss=0.09021, over 28397.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3362, pruned_loss=0.0871, over 5686834.62 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3358, pruned_loss=0.09276, over 5767302.80 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3368, pruned_loss=0.08705, over 5676350.85 frames. ], batch size: 368, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:19:04,187 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=517845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:19:05,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=517846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:19:37,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.485e+02 1.394e+03 1.938e+03 2.593e+03 6.845e+03, threshold=3.876e+03, percent-clipped=12.0
+2023-03-06 06:20:09,042 INFO [train.py:968] (0/2) Epoch 12, batch 16700, giga_loss[loss=0.2491, simple_loss=0.3365, pruned_loss=0.0808, over 28058.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3368, pruned_loss=0.08724, over 5680414.16 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3359, pruned_loss=0.09281, over 5768287.47 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3372, pruned_loss=0.0871, over 5670506.71 frames. ], batch size: 412, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:21:10,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4423, 1.7452, 1.7188, 1.2861], device='cuda:0'), covar=tensor([0.1821, 0.2420, 0.1468, 0.1703], device='cuda:0'), in_proj_covar=tensor([0.0824, 0.0676, 0.0865, 0.0772], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 06:21:16,672 INFO [train.py:968] (0/2) Epoch 12, batch 16750, giga_loss[loss=0.2795, simple_loss=0.3705, pruned_loss=0.09422, over 28711.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3369, pruned_loss=0.08669, over 5680415.39 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3362, pruned_loss=0.09286, over 5770002.28 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.337, pruned_loss=0.08643, over 5669332.46 frames. ], batch size: 262, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:21:51,595 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.062e+02 1.247e+03 1.649e+03 2.469e+03 4.722e+03, threshold=3.298e+03, percent-clipped=8.0
+2023-03-06 06:22:18,783 INFO [train.py:968] (0/2) Epoch 12, batch 16800, giga_loss[loss=0.2535, simple_loss=0.3335, pruned_loss=0.08675, over 27437.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3361, pruned_loss=0.08607, over 5678706.68 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.336, pruned_loss=0.09254, over 5766884.88 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3363, pruned_loss=0.08583, over 5667701.33 frames. ], batch size: 472, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:22:28,985 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-518000.pt
+2023-03-06 06:23:25,546 INFO [train.py:968] (0/2) Epoch 12, batch 16850, giga_loss[loss=0.2826, simple_loss=0.3629, pruned_loss=0.1011, over 28942.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3399, pruned_loss=0.08804, over 5686492.55 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3356, pruned_loss=0.09244, over 5770462.00 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3405, pruned_loss=0.08781, over 5672948.87 frames. ], batch size: 213, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:24:03,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.885e+02 1.429e+03 1.889e+03 2.507e+03 5.397e+03, threshold=3.779e+03, percent-clipped=9.0
+2023-03-06 06:24:11,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518076.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:24:13,348 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-06 06:24:31,397 INFO [train.py:968] (0/2) Epoch 12, batch 16900, giga_loss[loss=0.2194, simple_loss=0.3065, pruned_loss=0.06615, over 29098.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3394, pruned_loss=0.0874, over 5689458.29 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3356, pruned_loss=0.09238, over 5771794.66 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3399, pruned_loss=0.08718, over 5676064.53 frames. ], batch size: 128, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:24:56,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7481, 1.0547, 2.8754, 2.6566], device='cuda:0'), covar=tensor([0.1634, 0.2382, 0.0562, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0664, 0.0587, 0.0853, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:25:25,052 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-06 06:25:40,742 INFO [train.py:968] (0/2) Epoch 12, batch 16950, giga_loss[loss=0.2548, simple_loss=0.3353, pruned_loss=0.08712, over 28951.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3379, pruned_loss=0.08774, over 5697086.32 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3356, pruned_loss=0.09242, over 5773355.94 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3385, pruned_loss=0.08747, over 5683866.01 frames. ], batch size: 227, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:26:24,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.446e+02 1.272e+03 1.771e+03 2.334e+03 4.466e+03, threshold=3.542e+03, percent-clipped=1.0
+2023-03-06 06:26:51,655 INFO [train.py:968] (0/2) Epoch 12, batch 17000, giga_loss[loss=0.2828, simple_loss=0.3546, pruned_loss=0.1055, over 26954.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3366, pruned_loss=0.08692, over 5696871.38 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3357, pruned_loss=0.09243, over 5775053.55 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3369, pruned_loss=0.08657, over 5683051.74 frames. ], batch size: 555, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:26:59,264 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-06 06:27:10,071 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 06:27:32,705 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518219.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:27:33,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:27:34,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:27:36,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518222.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:28:00,923 INFO [train.py:968] (0/2) Epoch 12, batch 17050, giga_loss[loss=0.2821, simple_loss=0.3472, pruned_loss=0.1085, over 26825.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3342, pruned_loss=0.08467, over 5703112.87 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3354, pruned_loss=0.0923, over 5776160.53 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3346, pruned_loss=0.08445, over 5690570.47 frames. ], batch size: 555, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:28:01,216 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518242.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:28:09,460 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518251.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:28:32,130 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.474e+02 1.228e+03 1.516e+03 2.087e+03 9.629e+03, threshold=3.031e+03, percent-clipped=2.0
+2023-03-06 06:28:55,176 INFO [train.py:968] (0/2) Epoch 12, batch 17100, giga_loss[loss=0.2345, simple_loss=0.3176, pruned_loss=0.07571, over 29093.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3343, pruned_loss=0.08544, over 5698767.64 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3349, pruned_loss=0.09201, over 5778692.03 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3352, pruned_loss=0.08526, over 5683779.40 frames. ], batch size: 200, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:29:00,343 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2300, 1.2804, 3.4871, 3.2068], device='cuda:0'), covar=tensor([0.1402, 0.2534, 0.0401, 0.0924], device='cuda:0'), in_proj_covar=tensor([0.0663, 0.0586, 0.0852, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:29:17,013 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 06:29:24,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0727, 2.9360, 1.1532, 1.4230], device='cuda:0'), covar=tensor([0.1206, 0.0432, 0.1062, 0.1470], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0504, 0.0343, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 06:29:31,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3773, 1.7771, 1.6019, 1.4782], device='cuda:0'), covar=tensor([0.0731, 0.0267, 0.0290, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0058, 0.0053, 0.0091], device='cuda:0')
+2023-03-06 06:29:42,036 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 06:29:48,452 INFO [train.py:968] (0/2) Epoch 12, batch 17150, giga_loss[loss=0.2541, simple_loss=0.3346, pruned_loss=0.08677, over 27682.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3365, pruned_loss=0.08676, over 5695231.89 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3346, pruned_loss=0.09186, over 5775229.16 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3375, pruned_loss=0.08655, over 5682941.43 frames. ], batch size: 472, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:30:13,182 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:16,938 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:18,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:18,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:23,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.545e+02 1.415e+03 1.977e+03 3.147e+03 8.693e+03, threshold=3.953e+03, percent-clipped=27.0
+2023-03-06 06:30:45,597 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-06 06:30:46,507 INFO [train.py:968] (0/2) Epoch 12, batch 17200, giga_loss[loss=0.23, simple_loss=0.3163, pruned_loss=0.07183, over 28856.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3377, pruned_loss=0.08803, over 5679332.88 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3348, pruned_loss=0.09197, over 5766574.68 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3383, pruned_loss=0.08773, over 5676343.20 frames. ], batch size: 174, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:30:50,442 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:51,544 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:55,784 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:56,685 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:31:00,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5410, 1.6892, 1.7846, 1.4188], device='cuda:0'), covar=tensor([0.1374, 0.2023, 0.1154, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0675, 0.0863, 0.0771], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 06:31:44,333 INFO [train.py:968] (0/2) Epoch 12, batch 17250, giga_loss[loss=0.313, simple_loss=0.3699, pruned_loss=0.1281, over 27626.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3354, pruned_loss=0.08791, over 5678711.60 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3346, pruned_loss=0.09186, over 5767235.81 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3361, pruned_loss=0.08772, over 5674566.84 frames. ], batch size: 472, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:31:56,753 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:32:19,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.231e+02 1.406e+03 1.934e+03 3.173e+03 1.460e+04, threshold=3.869e+03, percent-clipped=15.0
+2023-03-06 06:32:39,473 INFO [train.py:968] (0/2) Epoch 12, batch 17300, giga_loss[loss=0.2427, simple_loss=0.3164, pruned_loss=0.08446, over 28984.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3343, pruned_loss=0.08783, over 5686447.18 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3344, pruned_loss=0.09175, over 5771317.79 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3351, pruned_loss=0.08767, over 5676541.68 frames. ], batch size: 213, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:32:41,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1369, 2.3201, 1.7053, 1.8731], device='cuda:0'), covar=tensor([0.0782, 0.0573, 0.0878, 0.1042], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0433, 0.0497, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:33:18,522 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:33:32,389 INFO [train.py:968] (0/2) Epoch 12, batch 17350, giga_loss[loss=0.3172, simple_loss=0.373, pruned_loss=0.1307, over 26746.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.338, pruned_loss=0.09033, over 5685715.68 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3341, pruned_loss=0.09156, over 5765900.56 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.339, pruned_loss=0.0903, over 5680277.81 frames. ], batch size: 555, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:33:59,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3760, 3.3474, 1.5547, 1.4447], device='cuda:0'), covar=tensor([0.0948, 0.0290, 0.0891, 0.1303], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0503, 0.0343, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 06:34:04,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.079e+02 1.305e+03 1.661e+03 2.253e+03 4.313e+03, threshold=3.321e+03, percent-clipped=2.0
+2023-03-06 06:34:13,256 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:34:20,090 INFO [train.py:968] (0/2) Epoch 12, batch 17400, giga_loss[loss=0.299, simple_loss=0.3844, pruned_loss=0.1068, over 28803.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3465, pruned_loss=0.09578, over 5690552.95 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3334, pruned_loss=0.09125, over 5769340.48 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3482, pruned_loss=0.09609, over 5680481.63 frames. ], batch size: 199, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:34:37,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8249, 2.2016, 1.7162, 1.5389], device='cuda:0'), covar=tensor([0.2268, 0.1658, 0.2235, 0.2052], device='cuda:0'), in_proj_covar=tensor([0.1712, 0.1605, 0.1543, 0.1671], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 06:34:43,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518617.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:35:03,283 INFO [train.py:968] (0/2) Epoch 12, batch 17450, giga_loss[loss=0.2774, simple_loss=0.3437, pruned_loss=0.1055, over 28871.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3532, pruned_loss=0.09948, over 5693163.01 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3334, pruned_loss=0.0912, over 5766698.83 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.355, pruned_loss=0.09992, over 5686249.62 frames. ], batch size: 99, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:35:06,908 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:35:30,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.098e+02 1.176e+03 1.559e+03 2.302e+03 5.668e+03, threshold=3.118e+03, percent-clipped=9.0
+2023-03-06 06:35:46,495 INFO [train.py:968] (0/2) Epoch 12, batch 17500, giga_loss[loss=0.2255, simple_loss=0.3061, pruned_loss=0.07243, over 28536.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3514, pruned_loss=0.09951, over 5685255.86 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3338, pruned_loss=0.09144, over 5759017.91 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3528, pruned_loss=0.09982, over 5684026.77 frames. ], batch size: 65, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:36:08,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3060, 2.5003, 1.3158, 1.5000], device='cuda:0'), covar=tensor([0.0942, 0.0334, 0.0850, 0.1244], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0502, 0.0341, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 06:36:30,125 INFO [train.py:968] (0/2) Epoch 12, batch 17550, giga_loss[loss=0.2612, simple_loss=0.3396, pruned_loss=0.09139, over 28732.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3442, pruned_loss=0.0963, over 5688721.41 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3334, pruned_loss=0.0911, over 5763779.97 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3462, pruned_loss=0.09709, over 5681289.19 frames. ], batch size: 242, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:36:45,560 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518760.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:36:48,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518763.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:36:57,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.298e+02 1.015e+03 1.276e+03 1.746e+03 4.368e+03, threshold=2.551e+03, percent-clipped=3.0
+2023-03-06 06:36:59,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:37:00,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:37:17,551 INFO [train.py:968] (0/2) Epoch 12, batch 17600, giga_loss[loss=0.2168, simple_loss=0.2998, pruned_loss=0.06691, over 28897.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3382, pruned_loss=0.09417, over 5685283.87 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.334, pruned_loss=0.09142, over 5766506.44 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3394, pruned_loss=0.0946, over 5675697.78 frames. ], batch size: 174, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:37:17,812 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518792.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:37:46,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:37:57,811 INFO [train.py:968] (0/2) Epoch 12, batch 17650, giga_loss[loss=0.2321, simple_loss=0.305, pruned_loss=0.07962, over 29011.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3316, pruned_loss=0.09119, over 5692330.84 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3347, pruned_loss=0.09166, over 5769317.95 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3318, pruned_loss=0.09132, over 5679728.15 frames. ], batch size: 155, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:38:13,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3325, 1.1323, 4.2158, 3.2900], device='cuda:0'), covar=tensor([0.1555, 0.2608, 0.0363, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0658, 0.0580, 0.0848, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:38:25,016 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.783e+02 1.012e+03 1.241e+03 1.900e+03 3.275e+03, threshold=2.481e+03, percent-clipped=8.0
+2023-03-06 06:38:40,340 INFO [train.py:968] (0/2) Epoch 12, batch 17700, giga_loss[loss=0.2151, simple_loss=0.2915, pruned_loss=0.06937, over 28646.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3252, pruned_loss=0.08833, over 5697724.06 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3347, pruned_loss=0.09149, over 5770278.36 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3252, pruned_loss=0.08854, over 5684656.01 frames. ], batch size: 85, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:38:49,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:38:50,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4193, 3.3507, 1.5233, 1.5412], device='cuda:0'), covar=tensor([0.0934, 0.0261, 0.0822, 0.1267], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0503, 0.0342, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 06:39:01,535 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:02,200 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:04,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:05,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:22,136 INFO [train.py:968] (0/2) Epoch 12, batch 17750, giga_loss[loss=0.2195, simple_loss=0.2978, pruned_loss=0.07058, over 29156.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3195, pruned_loss=0.08574, over 5698599.22 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3349, pruned_loss=0.09147, over 5772497.46 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3192, pruned_loss=0.08586, over 5685416.37 frames. ], batch size: 128, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:39:27,615 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:29,107 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:37,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:41,886 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-06 06:39:44,413 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:47,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:49,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.335e+02 1.010e+03 1.342e+03 1.954e+03 8.483e+03, threshold=2.684e+03, percent-clipped=10.0
+2023-03-06 06:40:02,293 INFO [train.py:968] (0/2) Epoch 12, batch 17800, giga_loss[loss=0.2054, simple_loss=0.2824, pruned_loss=0.06417, over 28682.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3173, pruned_loss=0.08494, over 5693795.03 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3352, pruned_loss=0.09147, over 5763129.46 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3165, pruned_loss=0.08491, over 5690370.92 frames. ], batch size: 262, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:40:08,707 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:40:28,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:40:48,179 INFO [train.py:968] (0/2) Epoch 12, batch 17850, giga_loss[loss=0.2316, simple_loss=0.3106, pruned_loss=0.07626, over 28998.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3134, pruned_loss=0.08315, over 5689513.69 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3354, pruned_loss=0.09151, over 5764056.76 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3122, pruned_loss=0.08296, over 5684842.49 frames. ], batch size: 155, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:40:50,707 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:40:52,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:40:54,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5752, 1.7005, 1.6138, 1.4162], device='cuda:0'), covar=tensor([0.2287, 0.1943, 0.1426, 0.1860], device='cuda:0'), in_proj_covar=tensor([0.1733, 0.1621, 0.1565, 0.1683], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 06:41:14,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.923e+02 9.363e+02 1.310e+03 1.984e+03 5.464e+03, threshold=2.620e+03, percent-clipped=8.0
+2023-03-06 06:41:17,806 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:41:29,319 INFO [train.py:968] (0/2) Epoch 12, batch 17900, giga_loss[loss=0.2274, simple_loss=0.3001, pruned_loss=0.07732, over 28486.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3099, pruned_loss=0.08137, over 5701344.22 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3356, pruned_loss=0.09162, over 5764538.65 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.3086, pruned_loss=0.08102, over 5696676.44 frames. ], batch size: 336, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:41:40,236 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:41:43,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:42:07,333 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:42:09,473 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:42:15,717 INFO [train.py:968] (0/2) Epoch 12, batch 17950, giga_loss[loss=0.2284, simple_loss=0.3104, pruned_loss=0.07316, over 28763.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3076, pruned_loss=0.08022, over 5692677.96 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3362, pruned_loss=0.09192, over 5767559.24 frames. ], giga_tot_loss[loss=0.232, simple_loss=0.3052, pruned_loss=0.07939, over 5684304.45 frames. ], batch size: 262, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:42:19,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4013, 2.4347, 2.0192, 2.0242], device='cuda:0'), covar=tensor([0.0762, 0.0670, 0.0888, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0431, 0.0497, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:42:33,057 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:42:35,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:42:40,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.040e+02 8.811e+02 1.115e+03 1.501e+03 3.863e+03, threshold=2.231e+03, percent-clipped=5.0
+2023-03-06 06:42:57,540 INFO [train.py:968] (0/2) Epoch 12, batch 18000, giga_loss[loss=0.2087, simple_loss=0.2786, pruned_loss=0.06939, over 28344.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3046, pruned_loss=0.07885, over 5691243.41 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3363, pruned_loss=0.09192, over 5765751.08 frames. ], giga_tot_loss[loss=0.2293, simple_loss=0.3024, pruned_loss=0.07809, over 5685623.95 frames. ], batch size: 78, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:42:57,545 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 06:43:06,968 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2157, simple_loss=0.3215, pruned_loss=0.0549, over 944034.00 frames. 
+2023-03-06 06:43:06,969 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 06:43:07,156 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:43:10,675 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:43:24,829 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:43:49,607 INFO [train.py:968] (0/2) Epoch 12, batch 18050, giga_loss[loss=0.2048, simple_loss=0.2784, pruned_loss=0.06566, over 28933.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3012, pruned_loss=0.07749, over 5691533.51 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3363, pruned_loss=0.09186, over 5768453.18 frames. ], giga_tot_loss[loss=0.2261, simple_loss=0.2988, pruned_loss=0.07669, over 5683385.86 frames. ], batch size: 213, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:44:14,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.527e+02 9.551e+02 1.179e+03 1.596e+03 6.920e+03, threshold=2.358e+03, percent-clipped=13.0
+2023-03-06 06:44:16,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6078, 1.6901, 1.8473, 1.4601], device='cuda:0'), covar=tensor([0.1377, 0.2013, 0.1163, 0.1388], device='cuda:0'), in_proj_covar=tensor([0.0843, 0.0690, 0.0882, 0.0786], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 06:44:17,506 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5225, 1.6101, 1.2469, 1.2204], device='cuda:0'), covar=tensor([0.0809, 0.0585, 0.1093, 0.1016], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0431, 0.0495, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 06:44:17,659 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-06 06:44:35,160 INFO [train.py:968] (0/2) Epoch 12, batch 18100, giga_loss[loss=0.1857, simple_loss=0.2565, pruned_loss=0.05745, over 28466.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.2992, pruned_loss=0.0768, over 5684553.94 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3369, pruned_loss=0.0921, over 5769108.44 frames. ], giga_tot_loss[loss=0.2236, simple_loss=0.296, pruned_loss=0.07563, over 5675996.64 frames. ], batch size: 85, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:45:15,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1184, 5.2035, 2.3050, 2.4365], device='cuda:0'), covar=tensor([0.0808, 0.0193, 0.0779, 0.1053], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0504, 0.0343, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 06:45:20,188 INFO [train.py:968] (0/2) Epoch 12, batch 18150, giga_loss[loss=0.1981, simple_loss=0.2666, pruned_loss=0.06474, over 28069.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2972, pruned_loss=0.07621, over 5685756.30 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.337, pruned_loss=0.09207, over 5770319.62 frames. ], giga_tot_loss[loss=0.2223, simple_loss=0.2942, pruned_loss=0.07516, over 5677140.02 frames. ], batch size: 77, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:45:51,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.107e+02 1.025e+03 1.220e+03 1.517e+03 5.766e+03, threshold=2.440e+03, percent-clipped=4.0
+2023-03-06 06:45:52,330 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 06:46:08,230 INFO [train.py:968] (0/2) Epoch 12, batch 18200, giga_loss[loss=0.2479, simple_loss=0.3347, pruned_loss=0.08054, over 28856.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3076, pruned_loss=0.08199, over 5678392.92 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3374, pruned_loss=0.09212, over 5771215.14 frames. ], giga_tot_loss[loss=0.2329, simple_loss=0.3041, pruned_loss=0.0808, over 5669043.64 frames. ], batch size: 174, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:46:49,382 INFO [train.py:968] (0/2) Epoch 12, batch 18250, giga_loss[loss=0.2792, simple_loss=0.3507, pruned_loss=0.1039, over 28485.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3208, pruned_loss=0.08889, over 5688023.07 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3371, pruned_loss=0.09181, over 5775542.51 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3174, pruned_loss=0.08795, over 5672975.36 frames. ], batch size: 60, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:47:14,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.822e+02 1.318e+03 1.896e+03 2.594e+03 7.077e+03, threshold=3.792e+03, percent-clipped=28.0
+2023-03-06 06:47:27,737 INFO [train.py:968] (0/2) Epoch 12, batch 18300, giga_loss[loss=0.3034, simple_loss=0.3845, pruned_loss=0.1112, over 28638.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3331, pruned_loss=0.09483, over 5698143.21 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3376, pruned_loss=0.09186, over 5775505.54 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3296, pruned_loss=0.09405, over 5683819.83 frames. ], batch size: 262, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:47:30,144 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-06 06:47:39,824 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:47:45,749 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:47:47,729 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:48:04,867 INFO [train.py:968] (0/2) Epoch 12, batch 18350, giga_loss[loss=0.3013, simple_loss=0.384, pruned_loss=0.1093, over 28654.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3395, pruned_loss=0.09732, over 5695646.41 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3382, pruned_loss=0.09218, over 5770647.53 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.336, pruned_loss=0.09658, over 5683978.13 frames. ], batch size: 307, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:48:26,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:48:32,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.070e+02 1.358e+03 1.643e+03 2.461e+03 7.125e+03, threshold=3.287e+03, percent-clipped=3.0
+2023-03-06 06:48:40,768 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:48:43,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:48:45,908 INFO [train.py:968] (0/2) Epoch 12, batch 18400, giga_loss[loss=0.3202, simple_loss=0.3923, pruned_loss=0.124, over 27587.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3434, pruned_loss=0.09807, over 5682209.27 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3383, pruned_loss=0.09232, over 5762029.23 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3405, pruned_loss=0.09745, over 5679920.31 frames. ], batch size: 472, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:49:29,533 INFO [train.py:968] (0/2) Epoch 12, batch 18450, giga_loss[loss=0.2992, simple_loss=0.3661, pruned_loss=0.1162, over 28916.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3452, pruned_loss=0.09809, over 5681189.11 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3386, pruned_loss=0.0925, over 5763898.47 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3427, pruned_loss=0.0975, over 5676714.18 frames. ], batch size: 164, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:49:39,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:49:41,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:49:58,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.675e+02 1.131e+03 1.392e+03 1.833e+03 9.427e+03, threshold=2.785e+03, percent-clipped=3.0
+2023-03-06 06:50:02,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3764, 1.6545, 1.3467, 1.3595], device='cuda:0'), covar=tensor([0.2495, 0.2386, 0.2656, 0.2056], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.0967, 0.1157, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 06:50:05,306 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:12,857 INFO [train.py:968] (0/2) Epoch 12, batch 18500, giga_loss[loss=0.2765, simple_loss=0.3532, pruned_loss=0.09985, over 28219.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3471, pruned_loss=0.09944, over 5675974.48 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3388, pruned_loss=0.09246, over 5767550.29 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3451, pruned_loss=0.09917, over 5667054.22 frames. ], batch size: 368, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:50:14,649 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-06 06:50:29,301 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:32,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:48,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:50,129 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:56,916 INFO [train.py:968] (0/2) Epoch 12, batch 18550, giga_loss[loss=0.2759, simple_loss=0.3614, pruned_loss=0.09516, over 28402.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3506, pruned_loss=0.1023, over 5677156.18 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3388, pruned_loss=0.09246, over 5767550.29 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.349, pruned_loss=0.1021, over 5670213.43 frames. ], batch size: 71, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:50:58,123 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:58,926 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-06 06:51:15,113 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:51:25,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.563e+02 1.055e+03 1.286e+03 1.771e+03 4.487e+03, threshold=2.571e+03, percent-clipped=3.0
+2023-03-06 06:51:39,834 INFO [train.py:968] (0/2) Epoch 12, batch 18600, giga_loss[loss=0.2691, simple_loss=0.3536, pruned_loss=0.09233, over 28564.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.354, pruned_loss=0.1046, over 5676638.69 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3392, pruned_loss=0.09269, over 5769328.06 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3527, pruned_loss=0.1045, over 5667878.32 frames. ], batch size: 336, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:52:01,129 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:52:21,115 INFO [train.py:968] (0/2) Epoch 12, batch 18650, giga_loss[loss=0.3062, simple_loss=0.3861, pruned_loss=0.1132, over 28763.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3573, pruned_loss=0.1056, over 5683014.71 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3395, pruned_loss=0.09289, over 5771187.64 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3562, pruned_loss=0.1055, over 5673462.39 frames. ], batch size: 262, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:52:45,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.431e+02 1.193e+03 1.414e+03 1.754e+03 4.161e+03, threshold=2.828e+03, percent-clipped=4.0
+2023-03-06 06:52:59,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:52:59,810 INFO [train.py:968] (0/2) Epoch 12, batch 18700, giga_loss[loss=0.321, simple_loss=0.3785, pruned_loss=0.1317, over 26528.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3588, pruned_loss=0.1057, over 5687531.29 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.34, pruned_loss=0.09303, over 5775544.31 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3582, pruned_loss=0.1059, over 5672506.33 frames. ], batch size: 555, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:53:01,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:53:36,181 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-06 06:53:37,039 INFO [train.py:968] (0/2) Epoch 12, batch 18750, giga_loss[loss=0.3209, simple_loss=0.3831, pruned_loss=0.1293, over 28357.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3597, pruned_loss=0.1055, over 5699987.78 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.34, pruned_loss=0.09317, over 5779351.63 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3597, pruned_loss=0.1059, over 5682325.51 frames. ], batch size: 77, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:53:54,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:54:04,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.574e+02 1.172e+03 1.623e+03 2.197e+03 6.113e+03, threshold=3.246e+03, percent-clipped=9.0
+2023-03-06 06:54:14,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1619, 0.8398, 0.9246, 1.3523], device='cuda:0'), covar=tensor([0.0810, 0.0379, 0.0359, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 06:54:20,136 INFO [train.py:968] (0/2) Epoch 12, batch 18800, giga_loss[loss=0.2548, simple_loss=0.3461, pruned_loss=0.08175, over 29002.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3601, pruned_loss=0.1045, over 5701280.26 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.34, pruned_loss=0.09312, over 5777347.65 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3602, pruned_loss=0.1049, over 5688485.31 frames. ], batch size: 136, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:54:24,329 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519998.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:54:25,581 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-520000.pt
+2023-03-06 06:54:55,577 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:54:57,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=520037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:54:57,537 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:54:59,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=520040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:55:00,637 INFO [train.py:968] (0/2) Epoch 12, batch 18850, giga_loss[loss=0.2529, simple_loss=0.3416, pruned_loss=0.08207, over 28739.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3579, pruned_loss=0.1017, over 5708786.36 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3406, pruned_loss=0.09347, over 5779330.39 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3578, pruned_loss=0.1019, over 5695993.95 frames. ], batch size: 92, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:55:20,096 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:55:22,564 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:55:27,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.403e+02 1.003e+03 1.228e+03 1.503e+03 3.643e+03, threshold=2.456e+03, percent-clipped=1.0
+2023-03-06 06:55:34,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9050, 1.0621, 1.0526, 0.8216], device='cuda:0'), covar=tensor([0.1835, 0.2268, 0.1111, 0.1625], device='cuda:0'), in_proj_covar=tensor([0.1745, 0.1638, 0.1579, 0.1699], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 06:55:41,102 INFO [train.py:968] (0/2) Epoch 12, batch 18900, giga_loss[loss=0.3096, simple_loss=0.3781, pruned_loss=0.1206, over 28884.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3574, pruned_loss=0.1014, over 5710095.28 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.341, pruned_loss=0.09355, over 5780694.91 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3572, pruned_loss=0.1016, over 5697428.30 frames. ], batch size: 136, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:55:47,411 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5116, 1.8273, 1.6300, 1.4305], device='cuda:0'), covar=tensor([0.2194, 0.1776, 0.1813, 0.1869], device='cuda:0'), in_proj_covar=tensor([0.1741, 0.1636, 0.1575, 0.1696], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 06:55:49,896 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:55:51,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=520105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:56:08,390 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 06:56:17,980 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:56:23,659 INFO [train.py:968] (0/2) Epoch 12, batch 18950, giga_loss[loss=0.3248, simple_loss=0.3759, pruned_loss=0.1369, over 28244.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3594, pruned_loss=0.1044, over 5716430.99 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3413, pruned_loss=0.09359, over 5783448.29 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3593, pruned_loss=0.1047, over 5702774.16 frames. ], batch size: 368, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:56:30,512 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=520149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:56:31,849 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2556, 0.8689, 1.0303, 1.4383], device='cuda:0'), covar=tensor([0.0747, 0.0363, 0.0315, 0.0782], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 06:56:56,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.075e+02 1.306e+03 1.550e+03 2.093e+03 5.022e+03, threshold=3.100e+03, percent-clipped=18.0
+2023-03-06 06:56:59,499 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-06 06:57:10,484 INFO [train.py:968] (0/2) Epoch 12, batch 19000, giga_loss[loss=0.3631, simple_loss=0.4115, pruned_loss=0.1573, over 28262.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3619, pruned_loss=0.1088, over 5711617.30 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3416, pruned_loss=0.09365, over 5777503.36 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3623, pruned_loss=0.1095, over 5702844.80 frames. ], batch size: 368, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:57:11,914 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=520194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:57:48,720 INFO [train.py:968] (0/2) Epoch 12, batch 19050, giga_loss[loss=0.2668, simple_loss=0.3374, pruned_loss=0.09808, over 28414.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3607, pruned_loss=0.1092, over 5710537.60 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3417, pruned_loss=0.09371, over 5774836.63 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3611, pruned_loss=0.1099, over 5705010.72 frames. ], batch size: 65, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:57:50,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9366, 3.0217, 2.1504, 0.9247], device='cuda:0'), covar=tensor([0.5431, 0.2150, 0.2717, 0.5006], device='cuda:0'), in_proj_covar=tensor([0.1558, 0.1476, 0.1484, 0.1274], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 06:58:16,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.704e+02 1.253e+03 1.658e+03 2.227e+03 1.365e+04, threshold=3.316e+03, percent-clipped=6.0
+2023-03-06 06:58:21,215 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-06 06:58:26,664 INFO [train.py:968] (0/2) Epoch 12, batch 19100, libri_loss[loss=0.3066, simple_loss=0.3832, pruned_loss=0.115, over 29297.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3591, pruned_loss=0.1088, over 5704940.82 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3422, pruned_loss=0.09383, over 5769526.72 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3597, pruned_loss=0.1099, over 5701684.70 frames. ], batch size: 94, lr: 2.70e-03, grad_scale: 1.0
+2023-03-06 06:58:32,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4267, 1.6055, 1.5460, 1.5559], device='cuda:0'), covar=tensor([0.1297, 0.1431, 0.1481, 0.1410], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0728, 0.0673, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 06:59:06,044 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:59:08,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=520340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:59:11,617 INFO [train.py:968] (0/2) Epoch 12, batch 19150, giga_loss[loss=0.2713, simple_loss=0.3489, pruned_loss=0.09689, over 28869.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3565, pruned_loss=0.1078, over 5702773.99 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3418, pruned_loss=0.09357, over 5770106.49 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3575, pruned_loss=0.109, over 5699035.57 frames. ], batch size: 145, lr: 2.70e-03, grad_scale: 1.0
+2023-03-06 06:59:34,638 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-06 06:59:35,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:59:38,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=520373.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:59:42,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.442e+02 1.193e+03 1.485e+03 1.972e+03 8.383e+03, threshold=2.970e+03, percent-clipped=4.0
+2023-03-06 06:59:54,361 INFO [train.py:968] (0/2) Epoch 12, batch 19200, giga_loss[loss=0.2841, simple_loss=0.3567, pruned_loss=0.1058, over 28877.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.356, pruned_loss=0.1064, over 5712354.69 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3422, pruned_loss=0.09362, over 5772525.17 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3567, pruned_loss=0.1077, over 5705920.03 frames. ], batch size: 199, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:00:37,133 INFO [train.py:968] (0/2) Epoch 12, batch 19250, giga_loss[loss=0.2656, simple_loss=0.3392, pruned_loss=0.09598, over 28719.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.355, pruned_loss=0.1054, over 5683215.97 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3425, pruned_loss=0.09368, over 5751892.40 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3556, pruned_loss=0.1066, over 5696840.06 frames. ], batch size: 262, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:01:06,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.727e+02 1.116e+03 1.464e+03 2.540e+03 6.259e+03, threshold=2.927e+03, percent-clipped=14.0
+2023-03-06 07:01:19,715 INFO [train.py:968] (0/2) Epoch 12, batch 19300, giga_loss[loss=0.3287, simple_loss=0.3777, pruned_loss=0.1398, over 28561.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3506, pruned_loss=0.1027, over 5682275.89 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3425, pruned_loss=0.09348, over 5756362.24 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3514, pruned_loss=0.1041, over 5687097.94 frames. ], batch size: 78, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:01:30,970 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-06 07:01:41,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520516.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:01:45,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=520519.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 07:01:49,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=520524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:02:07,074 INFO [train.py:968] (0/2) Epoch 12, batch 19350, giga_loss[loss=0.2354, simple_loss=0.3076, pruned_loss=0.08163, over 28676.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3441, pruned_loss=0.09924, over 5678477.99 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3429, pruned_loss=0.0937, over 5758632.57 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3445, pruned_loss=0.1003, over 5679348.27 frames. ], batch size: 85, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:02:12,927 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520548.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:02:39,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.195e+02 8.652e+02 1.029e+03 1.442e+03 4.920e+03, threshold=2.059e+03, percent-clipped=4.0
+2023-03-06 07:02:54,451 INFO [train.py:968] (0/2) Epoch 12, batch 19400, giga_loss[loss=0.2192, simple_loss=0.2771, pruned_loss=0.0806, over 23393.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3393, pruned_loss=0.09704, over 5669900.98 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3436, pruned_loss=0.09412, over 5760603.89 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3389, pruned_loss=0.09756, over 5666838.12 frames. ], batch size: 705, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:03:25,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5924, 1.5685, 1.2786, 1.2566], device='cuda:0'), covar=tensor([0.0725, 0.0529, 0.0971, 0.0995], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0436, 0.0499, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:03:39,270 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=520638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:03:42,936 INFO [train.py:968] (0/2) Epoch 12, batch 19450, giga_loss[loss=0.2753, simple_loss=0.3494, pruned_loss=0.1006, over 28243.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3381, pruned_loss=0.09668, over 5652733.16 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3441, pruned_loss=0.09436, over 5762557.15 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3372, pruned_loss=0.09692, over 5647241.54 frames. ], batch size: 368, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:04:03,741 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:04:05,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3668, 2.4024, 1.7043, 2.0683], device='cuda:0'), covar=tensor([0.0742, 0.0604, 0.0934, 0.0996], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0435, 0.0498, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:04:06,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=520670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:04:07,005 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=520671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:04:10,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.986e+02 9.524e+02 1.209e+03 1.618e+03 4.744e+03, threshold=2.418e+03, percent-clipped=11.0
+2023-03-06 07:04:24,876 INFO [train.py:968] (0/2) Epoch 12, batch 19500, giga_loss[loss=0.2655, simple_loss=0.3322, pruned_loss=0.09937, over 28700.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3389, pruned_loss=0.09707, over 5662154.27 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3445, pruned_loss=0.0945, over 5761771.32 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3378, pruned_loss=0.09717, over 5656691.15 frames. ], batch size: 92, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:04:32,464 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:04:32,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-06 07:04:40,020 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-06 07:04:45,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9839, 1.1118, 3.6685, 2.8503], device='cuda:0'), covar=tensor([0.1823, 0.2773, 0.0431, 0.1010], device='cuda:0'), in_proj_covar=tensor([0.0661, 0.0584, 0.0852, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:05:07,018 INFO [train.py:968] (0/2) Epoch 12, batch 19550, giga_loss[loss=0.321, simple_loss=0.3741, pruned_loss=0.134, over 26664.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3388, pruned_loss=0.09674, over 5667664.92 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3446, pruned_loss=0.09447, over 5757734.36 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3375, pruned_loss=0.09697, over 5662372.04 frames. ], batch size: 555, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:05:32,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.373e+02 1.048e+03 1.479e+03 1.992e+03 1.062e+04, threshold=2.959e+03, percent-clipped=18.0
+2023-03-06 07:05:45,344 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 07:05:46,914 INFO [train.py:968] (0/2) Epoch 12, batch 19600, giga_loss[loss=0.2762, simple_loss=0.3476, pruned_loss=0.1024, over 28028.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3374, pruned_loss=0.09573, over 5666946.73 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3454, pruned_loss=0.09476, over 5748316.94 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3356, pruned_loss=0.09568, over 5669274.16 frames. ], batch size: 412, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:05:50,988 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-06 07:06:26,075 INFO [train.py:968] (0/2) Epoch 12, batch 19650, giga_loss[loss=0.2463, simple_loss=0.3157, pruned_loss=0.0885, over 28746.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3363, pruned_loss=0.09521, over 5663631.43 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3463, pruned_loss=0.09512, over 5733001.60 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3338, pruned_loss=0.09486, over 5677137.10 frames. ], batch size: 92, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:06:55,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.552e+02 9.347e+02 1.147e+03 1.565e+03 3.405e+03, threshold=2.294e+03, percent-clipped=2.0
+2023-03-06 07:06:57,520 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 07:07:04,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2696, 2.5349, 1.3849, 1.2845], device='cuda:0'), covar=tensor([0.0946, 0.0381, 0.0838, 0.1369], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0506, 0.0342, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 07:07:06,616 INFO [train.py:968] (0/2) Epoch 12, batch 19700, giga_loss[loss=0.2423, simple_loss=0.3103, pruned_loss=0.08714, over 28526.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3329, pruned_loss=0.09329, over 5680347.36 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3467, pruned_loss=0.09525, over 5736033.94 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3305, pruned_loss=0.09289, over 5687545.12 frames. ], batch size: 85, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:07:47,895 INFO [train.py:968] (0/2) Epoch 12, batch 19750, giga_loss[loss=0.2525, simple_loss=0.3255, pruned_loss=0.08972, over 29036.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3302, pruned_loss=0.09188, over 5686961.72 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3469, pruned_loss=0.0953, over 5739385.31 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3275, pruned_loss=0.09144, over 5688057.96 frames. ], batch size: 136, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:08:15,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.940e+02 1.049e+03 1.387e+03 1.789e+03 4.049e+03, threshold=2.774e+03, percent-clipped=8.0
+2023-03-06 07:08:17,202 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=520979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:08:26,538 INFO [train.py:968] (0/2) Epoch 12, batch 19800, giga_loss[loss=0.2724, simple_loss=0.337, pruned_loss=0.1039, over 28895.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3273, pruned_loss=0.09051, over 5703532.86 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3472, pruned_loss=0.09526, over 5743832.03 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3245, pruned_loss=0.09011, over 5699412.42 frames. ], batch size: 174, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:08:42,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=521013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:09:06,086 INFO [train.py:968] (0/2) Epoch 12, batch 19850, giga_loss[loss=0.2247, simple_loss=0.2929, pruned_loss=0.07822, over 28693.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3254, pruned_loss=0.08964, over 5703730.30 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3479, pruned_loss=0.09551, over 5735097.04 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3221, pruned_loss=0.08896, over 5707482.39 frames. ], batch size: 92, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:09:08,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=521046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:09:35,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.930e+02 9.733e+02 1.409e+03 2.097e+03 1.061e+04, threshold=2.818e+03, percent-clipped=14.0
+2023-03-06 07:09:49,226 INFO [train.py:968] (0/2) Epoch 12, batch 19900, giga_loss[loss=0.3411, simple_loss=0.3838, pruned_loss=0.1491, over 27645.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3232, pruned_loss=0.08859, over 5701475.02 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3483, pruned_loss=0.09558, over 5734505.90 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3199, pruned_loss=0.08792, over 5704409.03 frames. ], batch size: 472, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:10:27,036 INFO [train.py:968] (0/2) Epoch 12, batch 19950, giga_loss[loss=0.24, simple_loss=0.319, pruned_loss=0.08055, over 28370.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.322, pruned_loss=0.08793, over 5708899.96 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3486, pruned_loss=0.09564, over 5734534.63 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3185, pruned_loss=0.08719, over 5710527.20 frames. ], batch size: 368, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:10:38,978 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=521156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:10:39,095 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-06 07:10:40,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=521159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:10:46,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=521167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:10:51,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4244, 1.7810, 1.3916, 1.5507], device='cuda:0'), covar=tensor([0.2413, 0.2373, 0.2691, 0.2191], device='cuda:0'), in_proj_covar=tensor([0.1320, 0.0975, 0.1162, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 07:10:54,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.493e+02 8.664e+02 1.052e+03 1.210e+03 4.368e+03, threshold=2.105e+03, percent-clipped=3.0
+2023-03-06 07:11:02,094 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=521188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:11:02,780 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=521189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:11:04,353 INFO [train.py:968] (0/2) Epoch 12, batch 20000, giga_loss[loss=0.2548, simple_loss=0.3342, pruned_loss=0.08767, over 28582.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3218, pruned_loss=0.0872, over 5716170.62 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3492, pruned_loss=0.09583, over 5737683.30 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3178, pruned_loss=0.0862, over 5713923.51 frames. ], batch size: 307, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:11:04,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=521192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:11:28,314 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=521221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:11:42,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4241, 4.3835, 1.6953, 1.5459], device='cuda:0'), covar=tensor([0.1195, 0.0343, 0.0945, 0.1555], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0504, 0.0341, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 07:11:46,899 INFO [train.py:968] (0/2) Epoch 12, batch 20050, giga_loss[loss=0.4439, simple_loss=0.4566, pruned_loss=0.2156, over 26663.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.326, pruned_loss=0.09025, over 5714854.82 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3495, pruned_loss=0.09586, over 5740631.23 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3222, pruned_loss=0.08933, over 5710388.35 frames. ], batch size: 555, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:12:18,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.559e+02 1.123e+03 1.416e+03 1.992e+03 4.655e+03, threshold=2.832e+03, percent-clipped=23.0
+2023-03-06 07:12:34,251 INFO [train.py:968] (0/2) Epoch 12, batch 20100, giga_loss[loss=0.3075, simple_loss=0.3719, pruned_loss=0.1216, over 28765.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3326, pruned_loss=0.09509, over 5696184.33 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3495, pruned_loss=0.09586, over 5733778.29 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3292, pruned_loss=0.09429, over 5698153.07 frames. ], batch size: 284, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:13:22,196 INFO [train.py:968] (0/2) Epoch 12, batch 20150, giga_loss[loss=0.3197, simple_loss=0.3736, pruned_loss=0.1329, over 28629.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3396, pruned_loss=0.09943, over 5699314.99 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3495, pruned_loss=0.09582, over 5738597.31 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3365, pruned_loss=0.09882, over 5695681.25 frames. ], batch size: 85, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:13:31,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=521354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:13:43,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5527, 1.8137, 1.4652, 1.5133], device='cuda:0'), covar=tensor([0.2301, 0.2258, 0.2533, 0.2216], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.0969, 0.1159, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 07:13:52,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.545e+02 1.269e+03 1.664e+03 2.211e+03 6.248e+03, threshold=3.327e+03, percent-clipped=11.0
+2023-03-06 07:13:53,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9742, 1.2947, 1.0300, 0.2876], device='cuda:0'), covar=tensor([0.2404, 0.2004, 0.2987, 0.3908], device='cuda:0'), in_proj_covar=tensor([0.1551, 0.1470, 0.1486, 0.1262], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 07:14:04,354 INFO [train.py:968] (0/2) Epoch 12, batch 20200, giga_loss[loss=0.2558, simple_loss=0.3386, pruned_loss=0.08651, over 29015.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3453, pruned_loss=0.1023, over 5695501.98 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3499, pruned_loss=0.09594, over 5734816.29 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.342, pruned_loss=0.1019, over 5693725.13 frames. ], batch size: 164, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:14:50,718 INFO [train.py:968] (0/2) Epoch 12, batch 20250, giga_loss[loss=0.3002, simple_loss=0.386, pruned_loss=0.1072, over 28954.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3513, pruned_loss=0.1049, over 5685125.45 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3503, pruned_loss=0.09608, over 5731417.91 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3483, pruned_loss=0.1046, over 5686196.59 frames. ], batch size: 145, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:14:54,092 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-06 07:15:20,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.559e+02 1.072e+03 1.309e+03 1.758e+03 3.349e+03, threshold=2.619e+03, percent-clipped=1.0
+2023-03-06 07:15:28,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5466, 1.1306, 4.7332, 3.5149], device='cuda:0'), covar=tensor([0.1684, 0.2889, 0.0339, 0.0815], device='cuda:0'), in_proj_covar=tensor([0.0659, 0.0581, 0.0850, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:15:35,470 INFO [train.py:968] (0/2) Epoch 12, batch 20300, giga_loss[loss=0.3246, simple_loss=0.3976, pruned_loss=0.1258, over 28358.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3561, pruned_loss=0.1066, over 5680536.22 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3507, pruned_loss=0.09617, over 5726555.83 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3534, pruned_loss=0.1065, over 5684837.35 frames. ], batch size: 368, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:15:40,018 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=521497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:15:43,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=521500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:16:10,926 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=521529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:16:21,747 INFO [train.py:968] (0/2) Epoch 12, batch 20350, giga_loss[loss=0.3005, simple_loss=0.3656, pruned_loss=0.1177, over 28815.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3604, pruned_loss=0.109, over 5692808.39 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3505, pruned_loss=0.09605, over 5730410.97 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3586, pruned_loss=0.1093, over 5692011.52 frames. ], batch size: 186, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:16:21,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=521542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:16:45,154 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-06 07:16:49,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.600e+02 1.214e+03 1.529e+03 2.086e+03 4.625e+03, threshold=3.057e+03, percent-clipped=12.0
+2023-03-06 07:16:59,696 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=521587.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:17:03,964 INFO [train.py:968] (0/2) Epoch 12, batch 20400, giga_loss[loss=0.2547, simple_loss=0.3266, pruned_loss=0.09138, over 28283.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3572, pruned_loss=0.1069, over 5689329.46 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3506, pruned_loss=0.09613, over 5737179.31 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.356, pruned_loss=0.1075, over 5681581.91 frames. ], batch size: 368, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:17:31,635 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 07:17:45,640 INFO [train.py:968] (0/2) Epoch 12, batch 20450, giga_loss[loss=0.2985, simple_loss=0.375, pruned_loss=0.111, over 28948.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3539, pruned_loss=0.104, over 5693511.02 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3508, pruned_loss=0.09646, over 5731163.14 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3529, pruned_loss=0.1044, over 5691780.13 frames. ], batch size: 106, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:18:15,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.293e+02 1.212e+03 1.499e+03 2.033e+03 4.029e+03, threshold=2.998e+03, percent-clipped=6.0
+2023-03-06 07:18:21,120 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=521685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:18:24,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=521688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:18:29,894 INFO [train.py:968] (0/2) Epoch 12, batch 20500, giga_loss[loss=0.2469, simple_loss=0.3266, pruned_loss=0.08365, over 28537.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.353, pruned_loss=0.1029, over 5691756.75 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3512, pruned_loss=0.09665, over 5733560.88 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3518, pruned_loss=0.1032, over 5687800.57 frames. ], batch size: 71, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:18:51,052 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=521717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:19:11,722 INFO [train.py:968] (0/2) Epoch 12, batch 20550, giga_loss[loss=0.2844, simple_loss=0.3588, pruned_loss=0.105, over 28688.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3537, pruned_loss=0.1029, over 5692511.07 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3513, pruned_loss=0.09659, over 5736487.10 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3528, pruned_loss=0.1033, over 5685904.20 frames. ], batch size: 66, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:19:41,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3474, 1.5468, 1.2719, 1.5361], device='cuda:0'), covar=tensor([0.0740, 0.0338, 0.0311, 0.0772], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 07:19:43,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.828e+02 1.411e+03 1.898e+03 2.529e+03 9.081e+03, threshold=3.797e+03, percent-clipped=14.0
+2023-03-06 07:19:54,751 INFO [train.py:968] (0/2) Epoch 12, batch 20600, giga_loss[loss=0.4058, simple_loss=0.4366, pruned_loss=0.1875, over 27466.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.356, pruned_loss=0.1044, over 5690411.47 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3514, pruned_loss=0.0967, over 5735047.21 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3552, pruned_loss=0.1048, over 5685834.61 frames. ], batch size: 472, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:20:36,821 INFO [train.py:968] (0/2) Epoch 12, batch 20650, giga_loss[loss=0.2805, simple_loss=0.3548, pruned_loss=0.1031, over 28858.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3589, pruned_loss=0.1067, over 5695057.21 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.352, pruned_loss=0.09687, over 5740273.57 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3579, pruned_loss=0.107, over 5685384.26 frames. ], batch size: 119, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:20:43,839 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=521852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:21:08,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.231e+02 1.232e+03 1.720e+03 2.371e+03 4.698e+03, threshold=3.440e+03, percent-clipped=4.0
+2023-03-06 07:21:21,102 INFO [train.py:968] (0/2) Epoch 12, batch 20700, libri_loss[loss=0.2225, simple_loss=0.3123, pruned_loss=0.06636, over 29535.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3588, pruned_loss=0.1064, over 5705720.75 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3518, pruned_loss=0.0965, over 5739961.73 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3583, pruned_loss=0.1073, over 5697092.69 frames. ], batch size: 79, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:22:04,287 INFO [train.py:968] (0/2) Epoch 12, batch 20750, giga_loss[loss=0.3122, simple_loss=0.3764, pruned_loss=0.124, over 28620.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3595, pruned_loss=0.1071, over 5714110.31 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3519, pruned_loss=0.0965, over 5745898.71 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3593, pruned_loss=0.1082, over 5700703.92 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:22:23,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=521962.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:22:36,867 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.255e+02 1.192e+03 1.519e+03 2.065e+03 8.855e+03, threshold=3.037e+03, percent-clipped=7.0
+2023-03-06 07:22:47,607 INFO [train.py:968] (0/2) Epoch 12, batch 20800, giga_loss[loss=0.3557, simple_loss=0.4127, pruned_loss=0.1493, over 29068.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.361, pruned_loss=0.109, over 5712539.60 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3516, pruned_loss=0.09622, over 5749131.17 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3613, pruned_loss=0.1103, over 5698430.89 frames. ], batch size: 128, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:22:54,224 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-522000.pt
+2023-03-06 07:23:08,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0842, 3.1792, 2.1240, 1.1646], device='cuda:0'), covar=tensor([0.5577, 0.2211, 0.2959, 0.4852], device='cuda:0'), in_proj_covar=tensor([0.1542, 0.1459, 0.1481, 0.1257], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 07:23:25,923 INFO [train.py:968] (0/2) Epoch 12, batch 20850, giga_loss[loss=0.2517, simple_loss=0.3371, pruned_loss=0.08315, over 28388.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3597, pruned_loss=0.1073, over 5707992.33 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3515, pruned_loss=0.09613, over 5739601.69 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3602, pruned_loss=0.1087, over 5704717.59 frames. ], batch size: 65, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:23:55,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.270e+02 1.046e+03 1.218e+03 1.688e+03 3.857e+03, threshold=2.437e+03, percent-clipped=2.0
+2023-03-06 07:23:59,300 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=522083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:24:05,745 INFO [train.py:968] (0/2) Epoch 12, batch 20900, giga_loss[loss=0.295, simple_loss=0.3736, pruned_loss=0.1081, over 28723.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3602, pruned_loss=0.1068, over 5709424.68 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3513, pruned_loss=0.09605, over 5742729.95 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3609, pruned_loss=0.1082, over 5703396.88 frames. ], batch size: 262, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:24:16,309 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=522105.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:24:19,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=522108.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:24:43,064 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=522137.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 07:24:46,081 INFO [train.py:968] (0/2) Epoch 12, batch 20950, giga_loss[loss=0.2569, simple_loss=0.3388, pruned_loss=0.0875, over 28866.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.361, pruned_loss=0.1067, over 5717359.35 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3513, pruned_loss=0.09593, over 5746625.80 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3618, pruned_loss=0.1083, over 5708153.78 frames. ], batch size: 174, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:25:14,457 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.166e+02 1.067e+03 1.354e+03 1.728e+03 5.421e+03, threshold=2.709e+03, percent-clipped=12.0
+2023-03-06 07:25:22,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4756, 3.7329, 1.6789, 1.4674], device='cuda:0'), covar=tensor([0.0937, 0.0228, 0.0821, 0.1359], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0501, 0.0339, 0.0388], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 07:25:26,185 INFO [train.py:968] (0/2) Epoch 12, batch 21000, giga_loss[loss=0.284, simple_loss=0.3523, pruned_loss=0.1078, over 28665.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3584, pruned_loss=0.1053, over 5717401.19 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3512, pruned_loss=0.09584, over 5747188.31 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3593, pruned_loss=0.1068, over 5709276.31 frames. ], batch size: 92, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:25:26,190 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 07:25:34,736 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.217, simple_loss=0.323, pruned_loss=0.05553, over 944034.00 frames. 
+2023-03-06 07:25:34,737 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 07:25:51,115 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.89 vs. limit=5.0
+2023-03-06 07:26:00,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=522227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:26:13,520 INFO [train.py:968] (0/2) Epoch 12, batch 21050, giga_loss[loss=0.2897, simple_loss=0.3638, pruned_loss=0.1078, over 28502.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3557, pruned_loss=0.104, over 5715660.34 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3507, pruned_loss=0.09546, over 5751044.27 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.357, pruned_loss=0.1057, over 5704823.44 frames. ], batch size: 85, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:26:42,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.568e+02 1.078e+03 1.365e+03 1.965e+03 8.945e+03, threshold=2.729e+03, percent-clipped=16.0
+2023-03-06 07:26:46,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8837, 1.0679, 3.3699, 2.8586], device='cuda:0'), covar=tensor([0.1815, 0.2741, 0.0480, 0.0901], device='cuda:0'), in_proj_covar=tensor([0.0660, 0.0581, 0.0849, 0.0757], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:26:51,020 INFO [train.py:968] (0/2) Epoch 12, batch 21100, giga_loss[loss=0.277, simple_loss=0.3481, pruned_loss=0.103, over 29290.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3539, pruned_loss=0.103, over 5716146.96 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3508, pruned_loss=0.09553, over 5749555.25 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.355, pruned_loss=0.1045, over 5707989.80 frames. ], batch size: 113, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:27:13,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1810, 1.7719, 1.4329, 0.4075], device='cuda:0'), covar=tensor([0.2773, 0.1727, 0.2737, 0.3235], device='cuda:0'), in_proj_covar=tensor([0.1548, 0.1456, 0.1480, 0.1258], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 07:27:31,728 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-06 07:27:35,309 INFO [train.py:968] (0/2) Epoch 12, batch 21150, giga_loss[loss=0.2886, simple_loss=0.3594, pruned_loss=0.1089, over 28957.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3547, pruned_loss=0.1042, over 5717516.19 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3509, pruned_loss=0.09559, over 5751152.61 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3555, pruned_loss=0.1054, over 5709394.51 frames. ], batch size: 106, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:27:58,601 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=522370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:28:01,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=522373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:28:08,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.552e+02 1.075e+03 1.370e+03 2.038e+03 7.374e+03, threshold=2.741e+03, percent-clipped=9.0
+2023-03-06 07:28:15,814 INFO [train.py:968] (0/2) Epoch 12, batch 21200, giga_loss[loss=0.2773, simple_loss=0.3484, pruned_loss=0.1031, over 28691.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3543, pruned_loss=0.1042, over 5704259.28 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3507, pruned_loss=0.09558, over 5743543.03 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3553, pruned_loss=0.1053, over 5703127.62 frames. ], batch size: 85, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:28:23,614 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=522402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:28:45,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2420, 1.9911, 1.5212, 1.7584], device='cuda:0'), covar=tensor([0.0793, 0.0784, 0.1066, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0435, 0.0499, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:28:53,956 INFO [train.py:968] (0/2) Epoch 12, batch 21250, giga_loss[loss=0.2977, simple_loss=0.3675, pruned_loss=0.1139, over 28899.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3533, pruned_loss=0.1028, over 5696532.47 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3505, pruned_loss=0.09574, over 5723932.20 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3544, pruned_loss=0.104, over 5710356.30 frames. ], batch size: 186, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:29:08,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=522458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:29:25,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.028e+02 9.874e+02 1.295e+03 1.882e+03 1.006e+04, threshold=2.589e+03, percent-clipped=13.0
+2023-03-06 07:29:34,546 INFO [train.py:968] (0/2) Epoch 12, batch 21300, giga_loss[loss=0.297, simple_loss=0.3797, pruned_loss=0.1071, over 28554.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3526, pruned_loss=0.1015, over 5699749.34 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3505, pruned_loss=0.09582, over 5726169.52 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3535, pruned_loss=0.1025, over 5708269.86 frames. ], batch size: 78, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:29:34,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3032, 1.6556, 1.5067, 1.2316], device='cuda:0'), covar=tensor([0.1857, 0.1390, 0.1018, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.1706, 0.1604, 0.1561, 0.1679], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 07:29:59,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2237, 3.0200, 1.3359, 1.3814], device='cuda:0'), covar=tensor([0.0985, 0.0275, 0.0898, 0.1347], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0499, 0.0338, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0029, 0.0021, 0.0025], device='cuda:0')
+2023-03-06 07:30:14,042 INFO [train.py:968] (0/2) Epoch 12, batch 21350, giga_loss[loss=0.2935, simple_loss=0.3399, pruned_loss=0.1235, over 23616.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3535, pruned_loss=0.1031, over 5697144.36 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3514, pruned_loss=0.09667, over 5731932.69 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3536, pruned_loss=0.1034, over 5697774.95 frames. ], batch size: 705, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:30:41,332 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-06 07:30:45,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.694e+02 1.128e+03 1.409e+03 1.887e+03 5.220e+03, threshold=2.818e+03, percent-clipped=11.0
+2023-03-06 07:30:47,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4840, 1.7740, 1.7312, 1.3472], device='cuda:0'), covar=tensor([0.1710, 0.2406, 0.1386, 0.1571], device='cuda:0'), in_proj_covar=tensor([0.0834, 0.0693, 0.0876, 0.0781], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 07:30:54,483 INFO [train.py:968] (0/2) Epoch 12, batch 21400, giga_loss[loss=0.2776, simple_loss=0.3603, pruned_loss=0.09746, over 28510.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3517, pruned_loss=0.1023, over 5700093.27 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3509, pruned_loss=0.09645, over 5735769.35 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3521, pruned_loss=0.1029, over 5696554.54 frames. ], batch size: 60, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:31:01,544 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=522601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:31:03,388 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=522604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:31:26,452 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=522633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:31:33,178 INFO [train.py:968] (0/2) Epoch 12, batch 21450, giga_loss[loss=0.2614, simple_loss=0.334, pruned_loss=0.09438, over 28762.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3501, pruned_loss=0.1016, over 5689727.16 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3513, pruned_loss=0.09667, over 5720408.37 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3501, pruned_loss=0.1019, over 5699495.11 frames. ], batch size: 99, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:32:05,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.721e+02 1.053e+03 1.296e+03 1.737e+03 5.516e+03, threshold=2.593e+03, percent-clipped=4.0
+2023-03-06 07:32:12,961 INFO [train.py:968] (0/2) Epoch 12, batch 21500, giga_loss[loss=0.222, simple_loss=0.3047, pruned_loss=0.06972, over 28539.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3472, pruned_loss=0.1003, over 5690633.47 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3517, pruned_loss=0.09699, over 5721924.53 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3469, pruned_loss=0.1004, over 5696369.13 frames. ], batch size: 71, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:32:51,532 INFO [train.py:968] (0/2) Epoch 12, batch 21550, giga_loss[loss=0.2577, simple_loss=0.3354, pruned_loss=0.09006, over 28709.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3477, pruned_loss=0.1011, over 5686606.64 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3518, pruned_loss=0.0971, over 5723239.33 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3472, pruned_loss=0.1011, over 5689203.28 frames. ], batch size: 119, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:33:25,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.333e+02 1.076e+03 1.343e+03 1.810e+03 5.221e+03, threshold=2.686e+03, percent-clipped=11.0
+2023-03-06 07:33:34,014 INFO [train.py:968] (0/2) Epoch 12, batch 21600, giga_loss[loss=0.2641, simple_loss=0.3355, pruned_loss=0.09637, over 28809.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.346, pruned_loss=0.1001, over 5694836.41 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3516, pruned_loss=0.09695, over 5728698.37 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3457, pruned_loss=0.1004, over 5690950.22 frames. ], batch size: 284, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:34:14,621 INFO [train.py:968] (0/2) Epoch 12, batch 21650, giga_loss[loss=0.253, simple_loss=0.3302, pruned_loss=0.08792, over 28501.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3445, pruned_loss=0.09985, over 5706663.98 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3522, pruned_loss=0.09751, over 5732883.78 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3435, pruned_loss=0.09966, over 5699151.06 frames. ], batch size: 60, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:34:43,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2767, 1.6175, 1.2785, 1.5027], device='cuda:0'), covar=tensor([0.0715, 0.0311, 0.0330, 0.0775], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0090], device='cuda:0')
+2023-03-06 07:34:44,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.095e+02 1.074e+03 1.312e+03 1.643e+03 4.899e+03, threshold=2.624e+03, percent-clipped=9.0
+2023-03-06 07:34:53,150 INFO [train.py:968] (0/2) Epoch 12, batch 21700, giga_loss[loss=0.2893, simple_loss=0.3541, pruned_loss=0.1123, over 28286.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3427, pruned_loss=0.0995, over 5708326.37 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3525, pruned_loss=0.09771, over 5735405.95 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3416, pruned_loss=0.09922, over 5699824.54 frames. ], batch size: 368, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:35:30,270 INFO [train.py:968] (0/2) Epoch 12, batch 21750, giga_loss[loss=0.219, simple_loss=0.3014, pruned_loss=0.06833, over 28955.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3412, pruned_loss=0.09878, over 5718876.91 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3531, pruned_loss=0.09824, over 5741925.87 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3393, pruned_loss=0.09811, over 5704989.03 frames. ], batch size: 164, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:36:01,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.552e+02 1.084e+03 1.561e+03 2.291e+03 6.965e+03, threshold=3.121e+03, percent-clipped=19.0
+2023-03-06 07:36:10,039 INFO [train.py:968] (0/2) Epoch 12, batch 21800, giga_loss[loss=0.2503, simple_loss=0.3295, pruned_loss=0.08551, over 28785.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.341, pruned_loss=0.0988, over 5720056.23 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3537, pruned_loss=0.09877, over 5744101.11 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3388, pruned_loss=0.0978, over 5706833.87 frames. ], batch size: 119, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:36:46,170 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4562, 1.7682, 1.4011, 1.5868], device='cuda:0'), covar=tensor([0.2358, 0.2290, 0.2673, 0.2137], device='cuda:0'), in_proj_covar=tensor([0.1312, 0.0969, 0.1152, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 07:36:50,744 INFO [train.py:968] (0/2) Epoch 12, batch 21850, libri_loss[loss=0.3368, simple_loss=0.4048, pruned_loss=0.1344, over 29086.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3444, pruned_loss=0.1002, over 5708989.50 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3546, pruned_loss=0.09953, over 5737433.90 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3415, pruned_loss=0.09869, over 5703494.77 frames. ], batch size: 101, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:36:57,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2987, 1.8269, 0.9491, 1.3278], device='cuda:0'), covar=tensor([0.0997, 0.0635, 0.1588, 0.1188], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0436, 0.0498, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:37:25,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.940e+02 9.999e+02 1.342e+03 1.900e+03 8.587e+03, threshold=2.684e+03, percent-clipped=8.0
+2023-03-06 07:37:28,512 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-06 07:37:32,588 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:37:32,968 INFO [train.py:968] (0/2) Epoch 12, batch 21900, giga_loss[loss=0.26, simple_loss=0.3484, pruned_loss=0.08576, over 28695.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3479, pruned_loss=0.1018, over 5700044.30 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3555, pruned_loss=0.1003, over 5735258.20 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3446, pruned_loss=0.09993, over 5697002.57 frames. ], batch size: 284, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:38:06,057 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 07:38:15,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3348, 1.8958, 1.3700, 0.5576], device='cuda:0'), covar=tensor([0.4290, 0.2105, 0.3269, 0.4836], device='cuda:0'), in_proj_covar=tensor([0.1554, 0.1461, 0.1484, 0.1265], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 07:38:15,715 INFO [train.py:968] (0/2) Epoch 12, batch 21950, giga_loss[loss=0.2653, simple_loss=0.3455, pruned_loss=0.0926, over 28554.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3501, pruned_loss=0.1023, over 5697904.54 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3556, pruned_loss=0.1006, over 5734071.53 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3472, pruned_loss=0.1006, over 5695639.78 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:38:34,352 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3696, 1.9111, 1.4160, 0.5712], device='cuda:0'), covar=tensor([0.3590, 0.1879, 0.3175, 0.4438], device='cuda:0'), in_proj_covar=tensor([0.1551, 0.1455, 0.1479, 0.1261], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 07:38:49,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.541e+02 9.809e+02 1.146e+03 1.504e+03 4.161e+03, threshold=2.291e+03, percent-clipped=5.0
+2023-03-06 07:38:57,225 INFO [train.py:968] (0/2) Epoch 12, batch 22000, giga_loss[loss=0.2602, simple_loss=0.34, pruned_loss=0.09017, over 28740.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.35, pruned_loss=0.1014, over 5702300.32 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3555, pruned_loss=0.1006, over 5736075.09 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3476, pruned_loss=0.09997, over 5698102.43 frames. ], batch size: 284, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:39:08,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:39:18,005 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-06 07:39:39,210 INFO [train.py:968] (0/2) Epoch 12, batch 22050, giga_loss[loss=0.2609, simple_loss=0.3371, pruned_loss=0.09239, over 28833.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3491, pruned_loss=0.1008, over 5701321.70 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3559, pruned_loss=0.1012, over 5737625.45 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3467, pruned_loss=0.09917, over 5695969.79 frames. ], batch size: 199, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:40:13,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.246e+02 1.123e+03 1.587e+03 2.184e+03 9.255e+03, threshold=3.173e+03, percent-clipped=20.0
+2023-03-06 07:40:17,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2512, 2.1039, 1.5865, 1.7797], device='cuda:0'), covar=tensor([0.0763, 0.0725, 0.0999, 0.1102], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0437, 0.0497, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:40:21,078 INFO [train.py:968] (0/2) Epoch 12, batch 22100, giga_loss[loss=0.2793, simple_loss=0.3483, pruned_loss=0.1051, over 28696.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3494, pruned_loss=0.1011, over 5702307.97 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.356, pruned_loss=0.1013, over 5737430.21 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3472, pruned_loss=0.09969, over 5697874.59 frames. ], batch size: 99, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:40:48,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0293, 1.0960, 3.7555, 3.1140], device='cuda:0'), covar=tensor([0.1776, 0.2705, 0.0455, 0.0737], device='cuda:0'), in_proj_covar=tensor([0.0667, 0.0590, 0.0859, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:41:03,645 INFO [train.py:968] (0/2) Epoch 12, batch 22150, giga_loss[loss=0.2659, simple_loss=0.3412, pruned_loss=0.09534, over 28994.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3488, pruned_loss=0.1011, over 5700727.55 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.356, pruned_loss=0.1013, over 5737430.21 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3472, pruned_loss=0.09999, over 5697276.99 frames. ], batch size: 136, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:41:37,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.829e+02 1.223e+03 1.524e+03 1.780e+03 4.676e+03, threshold=3.048e+03, percent-clipped=3.0
+2023-03-06 07:41:42,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2719, 2.0194, 1.8142, 1.6666], device='cuda:0'), covar=tensor([0.0730, 0.0726, 0.0857, 0.1110], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0437, 0.0497, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:41:44,967 INFO [train.py:968] (0/2) Epoch 12, batch 22200, giga_loss[loss=0.3322, simple_loss=0.3998, pruned_loss=0.1323, over 28712.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3524, pruned_loss=0.1037, over 5705793.33 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3561, pruned_loss=0.1015, over 5740523.39 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3509, pruned_loss=0.1027, over 5699483.33 frames. ], batch size: 284, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:41:49,002 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7254, 1.7254, 1.3718, 1.3473], device='cuda:0'), covar=tensor([0.0749, 0.0570, 0.0913, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0437, 0.0498, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:42:06,669 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:42:14,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-06 07:42:18,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2502, 3.4612, 1.5166, 1.3217], device='cuda:0'), covar=tensor([0.0962, 0.0278, 0.0884, 0.1351], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0506, 0.0341, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 07:42:27,797 INFO [train.py:968] (0/2) Epoch 12, batch 22250, giga_loss[loss=0.3654, simple_loss=0.4122, pruned_loss=0.1593, over 26760.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3552, pruned_loss=0.1051, over 5710429.79 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3563, pruned_loss=0.1017, over 5741346.78 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3538, pruned_loss=0.1041, over 5704580.50 frames. ], batch size: 555, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:42:49,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=523466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:42:51,447 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-06 07:42:57,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-06 07:43:00,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.565e+02 1.180e+03 1.501e+03 1.910e+03 4.710e+03, threshold=3.002e+03, percent-clipped=8.0
+2023-03-06 07:43:02,719 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 07:43:05,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6993, 1.0186, 2.8478, 2.6929], device='cuda:0'), covar=tensor([0.1713, 0.2590, 0.0576, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0668, 0.0592, 0.0865, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:43:08,419 INFO [train.py:968] (0/2) Epoch 12, batch 22300, giga_loss[loss=0.251, simple_loss=0.3283, pruned_loss=0.0869, over 28854.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3572, pruned_loss=0.1064, over 5715008.60 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3572, pruned_loss=0.1024, over 5744286.77 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3554, pruned_loss=0.1051, over 5707358.79 frames. ], batch size: 106, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:43:47,064 INFO [train.py:968] (0/2) Epoch 12, batch 22350, giga_loss[loss=0.3512, simple_loss=0.4099, pruned_loss=0.1463, over 28775.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3572, pruned_loss=0.106, over 5724742.64 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3579, pruned_loss=0.1029, over 5747098.76 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3551, pruned_loss=0.1045, over 5715219.15 frames. ], batch size: 262, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:44:04,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:44:20,964 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=523581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:44:22,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.436e+02 1.250e+03 1.863e+03 2.481e+03 1.046e+04, threshold=3.726e+03, percent-clipped=16.0
+2023-03-06 07:44:29,890 INFO [train.py:968] (0/2) Epoch 12, batch 22400, giga_loss[loss=0.2968, simple_loss=0.3753, pruned_loss=0.1092, over 28735.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3576, pruned_loss=0.1062, over 5719178.50 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3584, pruned_loss=0.1033, over 5750707.58 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3554, pruned_loss=0.1047, over 5708008.34 frames. ], batch size: 284, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:44:45,482 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=523609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:44:48,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=523612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:45:11,466 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=523641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:45:12,411 INFO [train.py:968] (0/2) Epoch 12, batch 22450, giga_loss[loss=0.2585, simple_loss=0.338, pruned_loss=0.08947, over 29044.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3552, pruned_loss=0.1047, over 5722163.10 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3585, pruned_loss=0.1035, over 5749183.86 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3534, pruned_loss=0.1035, over 5714366.99 frames. ], batch size: 128, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:45:31,007 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:45:48,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.758e+02 1.173e+03 1.461e+03 1.914e+03 3.931e+03, threshold=2.923e+03, percent-clipped=1.0
+2023-03-06 07:45:57,006 INFO [train.py:968] (0/2) Epoch 12, batch 22500, giga_loss[loss=0.3505, simple_loss=0.3953, pruned_loss=0.1528, over 26703.00 frames. ], tot_loss[loss=0.279, simple_loss=0.352, pruned_loss=0.103, over 5720637.58 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3585, pruned_loss=0.1034, over 5749993.94 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3506, pruned_loss=0.1021, over 5713656.07 frames. ], batch size: 555, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:46:22,771 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=523724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:46:24,688 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=523727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:46:37,139 INFO [train.py:968] (0/2) Epoch 12, batch 22550, giga_loss[loss=0.2839, simple_loss=0.3549, pruned_loss=0.1065, over 28568.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3489, pruned_loss=0.1015, over 5719131.26 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3588, pruned_loss=0.1038, over 5751633.37 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3474, pruned_loss=0.1004, over 5712034.61 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:46:48,566 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=523756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:47:00,087 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:47:07,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.386e+02 1.042e+03 1.304e+03 1.857e+03 4.229e+03, threshold=2.607e+03, percent-clipped=4.0
+2023-03-06 07:47:11,648 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:47:15,658 INFO [train.py:968] (0/2) Epoch 12, batch 22600, giga_loss[loss=0.2742, simple_loss=0.3504, pruned_loss=0.09902, over 28774.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.348, pruned_loss=0.1005, over 5723592.32 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3596, pruned_loss=0.1046, over 5755603.89 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3457, pruned_loss=0.09883, over 5713530.63 frames. ], batch size: 112, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:47:15,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=523792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:47:59,202 INFO [train.py:968] (0/2) Epoch 12, batch 22650, giga_loss[loss=0.3137, simple_loss=0.3856, pruned_loss=0.1208, over 27625.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3492, pruned_loss=0.09979, over 5719508.31 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3595, pruned_loss=0.1046, over 5757130.94 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3474, pruned_loss=0.09845, over 5709837.32 frames. ], batch size: 472, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:48:32,802 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.685e+02 1.026e+03 1.271e+03 1.743e+03 4.949e+03, threshold=2.542e+03, percent-clipped=9.0
+2023-03-06 07:48:39,448 INFO [train.py:968] (0/2) Epoch 12, batch 22700, giga_loss[loss=0.2616, simple_loss=0.3426, pruned_loss=0.09028, over 29036.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3496, pruned_loss=0.09934, over 5724182.96 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3601, pruned_loss=0.105, over 5759047.99 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3475, pruned_loss=0.09777, over 5713922.25 frames. ], batch size: 128, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:49:13,707 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=523935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:49:15,023 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-06 07:49:15,587 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=523938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:49:16,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=523939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:49:18,104 INFO [train.py:968] (0/2) Epoch 12, batch 22750, giga_loss[loss=0.2448, simple_loss=0.3268, pruned_loss=0.08147, over 28998.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3494, pruned_loss=0.1, over 5729757.56 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3603, pruned_loss=0.1052, over 5760094.30 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3473, pruned_loss=0.09849, over 5720116.67 frames. ], batch size: 155, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:49:40,388 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=523967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:49:54,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.116e+02 1.077e+03 1.362e+03 1.849e+03 1.111e+04, threshold=2.724e+03, percent-clipped=14.0
+2023-03-06 07:50:00,801 INFO [train.py:968] (0/2) Epoch 12, batch 22800, giga_loss[loss=0.2982, simple_loss=0.376, pruned_loss=0.1102, over 28561.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3487, pruned_loss=0.1013, over 5727725.99 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3603, pruned_loss=0.1054, over 5762280.02 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3469, pruned_loss=0.09984, over 5717800.88 frames. ], batch size: 307, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:50:06,654 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-524000.pt
+2023-03-06 07:50:12,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2724, 1.9919, 1.6174, 1.6934], device='cuda:0'), covar=tensor([0.0704, 0.0756, 0.0946, 0.1165], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0436, 0.0494, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 07:50:36,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=524037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:50:40,035 INFO [train.py:968] (0/2) Epoch 12, batch 22850, giga_loss[loss=0.267, simple_loss=0.3321, pruned_loss=0.1009, over 28940.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3483, pruned_loss=0.1026, over 5708028.38 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3612, pruned_loss=0.1062, over 5743809.22 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3457, pruned_loss=0.1006, over 5715984.54 frames. ], batch size: 186, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:51:14,103 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:51:15,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.650e+02 1.196e+03 1.505e+03 2.164e+03 7.500e+03, threshold=3.011e+03, percent-clipped=12.0
+2023-03-06 07:51:16,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:51:17,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=524087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:51:20,276 INFO [train.py:968] (0/2) Epoch 12, batch 22900, giga_loss[loss=0.3332, simple_loss=0.3899, pruned_loss=0.1383, over 28625.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3485, pruned_loss=0.1039, over 5714156.05 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3623, pruned_loss=0.1072, over 5744422.92 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.345, pruned_loss=0.1013, over 5719074.11 frames. ], batch size: 307, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:51:36,317 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:51:54,194 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=524137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:51:57,056 INFO [train.py:968] (0/2) Epoch 12, batch 22950, libri_loss[loss=0.346, simple_loss=0.4054, pruned_loss=0.1433, over 29643.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3479, pruned_loss=0.1041, over 5716427.46 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3628, pruned_loss=0.1078, over 5749604.57 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.344, pruned_loss=0.1012, over 5714533.47 frames. ], batch size: 91, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:52:00,531 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=524147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:52:04,559 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-06 07:52:12,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=524162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:52:23,823 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=524175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:52:27,725 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:52:31,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:52:32,562 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.516e+02 1.089e+03 1.414e+03 1.983e+03 4.232e+03, threshold=2.828e+03, percent-clipped=6.0
+2023-03-06 07:52:37,300 INFO [train.py:968] (0/2) Epoch 12, batch 23000, giga_loss[loss=0.3053, simple_loss=0.36, pruned_loss=0.1252, over 26592.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3439, pruned_loss=0.1021, over 5718577.54 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3633, pruned_loss=0.1083, over 5749945.93 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3401, pruned_loss=0.0993, over 5716452.61 frames. ], batch size: 555, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:52:51,772 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:53:15,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4963, 2.1918, 1.7027, 0.7925], device='cuda:0'), covar=tensor([0.3967, 0.1981, 0.3249, 0.4560], device='cuda:0'), in_proj_covar=tensor([0.1568, 0.1468, 0.1485, 0.1272], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 07:53:16,420 INFO [train.py:968] (0/2) Epoch 12, batch 23050, giga_loss[loss=0.2427, simple_loss=0.3196, pruned_loss=0.08293, over 28561.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3394, pruned_loss=0.09989, over 5720058.15 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3628, pruned_loss=0.1082, over 5754048.68 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3363, pruned_loss=0.09759, over 5713964.98 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:53:23,931 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5075, 1.8306, 1.7882, 1.3114], device='cuda:0'), covar=tensor([0.1376, 0.2156, 0.1237, 0.1478], device='cuda:0'), in_proj_covar=tensor([0.0831, 0.0688, 0.0870, 0.0777], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 07:53:49,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.478e+02 1.158e+03 1.407e+03 1.913e+03 3.849e+03, threshold=2.814e+03, percent-clipped=4.0
+2023-03-06 07:53:53,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:53:54,487 INFO [train.py:968] (0/2) Epoch 12, batch 23100, giga_loss[loss=0.3105, simple_loss=0.3707, pruned_loss=0.1251, over 28555.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3356, pruned_loss=0.09779, over 5724070.40 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3624, pruned_loss=0.1081, over 5753441.45 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.333, pruned_loss=0.09586, over 5719081.19 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:53:55,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:54:05,123 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:54:07,571 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:54:18,632 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:54:30,974 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:54:35,124 INFO [train.py:968] (0/2) Epoch 12, batch 23150, libri_loss[loss=0.3181, simple_loss=0.3824, pruned_loss=0.1269, over 29749.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3404, pruned_loss=0.1006, over 5716685.38 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3627, pruned_loss=0.1088, over 5754882.45 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3368, pruned_loss=0.09795, over 5709758.28 frames. ], batch size: 87, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:55:08,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.557e+02 1.147e+03 1.573e+03 2.062e+03 8.199e+03, threshold=3.145e+03, percent-clipped=13.0
+2023-03-06 07:55:15,482 INFO [train.py:968] (0/2) Epoch 12, batch 23200, giga_loss[loss=0.2495, simple_loss=0.3281, pruned_loss=0.08545, over 28813.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3431, pruned_loss=0.1015, over 5715201.37 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3628, pruned_loss=0.1089, over 5757268.77 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3399, pruned_loss=0.09922, over 5707149.49 frames. ], batch size: 119, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:55:34,129 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5477, 1.6804, 1.4791, 1.3575], device='cuda:0'), covar=tensor([0.2301, 0.1864, 0.1639, 0.1854], device='cuda:0'), in_proj_covar=tensor([0.1732, 0.1637, 0.1598, 0.1706], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 07:55:57,973 INFO [train.py:968] (0/2) Epoch 12, batch 23250, giga_loss[loss=0.2566, simple_loss=0.3394, pruned_loss=0.08688, over 29094.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3453, pruned_loss=0.1013, over 5719086.12 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3628, pruned_loss=0.1089, over 5758593.18 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3426, pruned_loss=0.09946, over 5711079.71 frames. ], batch size: 155, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:56:15,293 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=524462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:56:33,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.858e+02 1.142e+03 1.420e+03 1.841e+03 4.867e+03, threshold=2.840e+03, percent-clipped=2.0
+2023-03-06 07:56:38,958 INFO [train.py:968] (0/2) Epoch 12, batch 23300, giga_loss[loss=0.3097, simple_loss=0.3768, pruned_loss=0.1213, over 27605.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3481, pruned_loss=0.1024, over 5720860.55 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3628, pruned_loss=0.1089, over 5761129.11 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3458, pruned_loss=0.1007, over 5711309.25 frames. ], batch size: 472, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:56:50,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7365, 4.5443, 4.3622, 1.8480], device='cuda:0'), covar=tensor([0.0526, 0.0724, 0.0812, 0.1923], device='cuda:0'), in_proj_covar=tensor([0.1056, 0.0982, 0.0857, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 07:56:56,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=524512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:57:22,511 INFO [train.py:968] (0/2) Epoch 12, batch 23350, giga_loss[loss=0.2592, simple_loss=0.3333, pruned_loss=0.09255, over 28589.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3505, pruned_loss=0.1036, over 5709751.94 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.363, pruned_loss=0.1092, over 5745755.21 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.348, pruned_loss=0.1018, over 5715230.94 frames. ], batch size: 71, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:57:27,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=524550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:57:57,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.915e+02 1.356e+03 1.737e+03 2.680e+03 7.977e+03, threshold=3.474e+03, percent-clipped=22.0
+2023-03-06 07:58:02,743 INFO [train.py:968] (0/2) Epoch 12, batch 23400, libri_loss[loss=0.3187, simple_loss=0.3807, pruned_loss=0.1284, over 25884.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3547, pruned_loss=0.1074, over 5709673.28 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3631, pruned_loss=0.1096, over 5750105.77 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.352, pruned_loss=0.1054, over 5708543.47 frames. ], batch size: 136, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:58:14,698 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:58:17,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:58:41,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1063, 3.1296, 2.0858, 1.0041], device='cuda:0'), covar=tensor([0.5403, 0.2171, 0.3039, 0.5296], device='cuda:0'), in_proj_covar=tensor([0.1565, 0.1469, 0.1493, 0.1269], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 07:58:47,737 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:58:51,564 INFO [train.py:968] (0/2) Epoch 12, batch 23450, libri_loss[loss=0.2571, simple_loss=0.3158, pruned_loss=0.09918, over 29504.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3602, pruned_loss=0.1121, over 5705834.70 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3627, pruned_loss=0.1095, over 5753290.22 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3583, pruned_loss=0.1106, over 5701280.67 frames. ], batch size: 70, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:59:05,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:59:08,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:59:37,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.782e+02 1.607e+03 2.142e+03 2.672e+03 1.168e+04, threshold=4.284e+03, percent-clipped=13.0
+2023-03-06 07:59:39,402 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:59:42,945 INFO [train.py:968] (0/2) Epoch 12, batch 23500, giga_loss[loss=0.361, simple_loss=0.4134, pruned_loss=0.1543, over 28527.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.368, pruned_loss=0.1179, over 5684337.91 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3626, pruned_loss=0.1095, over 5745064.33 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3666, pruned_loss=0.1167, over 5687248.48 frames. ], batch size: 78, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:59:43,986 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:59:44,023 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3730, 1.5871, 1.2970, 1.2879], device='cuda:0'), covar=tensor([0.2127, 0.2087, 0.2325, 0.1874], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.0969, 0.1152, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 07:59:48,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:00:11,122 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=524718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:00:17,283 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:00:33,480 INFO [train.py:968] (0/2) Epoch 12, batch 23550, giga_loss[loss=0.3375, simple_loss=0.3923, pruned_loss=0.1414, over 28960.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3744, pruned_loss=0.1234, over 5676657.63 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3628, pruned_loss=0.1098, over 5744492.40 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3732, pruned_loss=0.1223, over 5678615.70 frames. ], batch size: 106, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:01:17,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.934e+02 1.619e+03 1.946e+03 2.469e+03 6.392e+03, threshold=3.893e+03, percent-clipped=5.0
+2023-03-06 08:01:20,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=524788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:01:22,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1726, 1.3932, 1.2248, 1.0701], device='cuda:0'), covar=tensor([0.1841, 0.1769, 0.1199, 0.1567], device='cuda:0'), in_proj_covar=tensor([0.1729, 0.1643, 0.1593, 0.1708], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 08:01:23,066 INFO [train.py:968] (0/2) Epoch 12, batch 23600, giga_loss[loss=0.3299, simple_loss=0.3928, pruned_loss=0.1335, over 28905.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3808, pruned_loss=0.1287, over 5670717.17 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.363, pruned_loss=0.1101, over 5734576.30 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3798, pruned_loss=0.1278, over 5679741.94 frames. ], batch size: 199, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 08:02:13,801 INFO [train.py:968] (0/2) Epoch 12, batch 23650, giga_loss[loss=0.355, simple_loss=0.4103, pruned_loss=0.1498, over 28791.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3863, pruned_loss=0.1335, over 5657424.18 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3631, pruned_loss=0.1102, over 5726908.36 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3858, pruned_loss=0.133, over 5670442.31 frames. ], batch size: 186, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:02:51,517 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9619, 1.2780, 0.9540, 0.1607], device='cuda:0'), covar=tensor([0.2170, 0.1684, 0.2418, 0.4116], device='cuda:0'), in_proj_covar=tensor([0.1584, 0.1490, 0.1508, 0.1285], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 08:02:56,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.336e+02 1.693e+03 2.177e+03 2.908e+03 1.177e+04, threshold=4.355e+03, percent-clipped=7.0
+2023-03-06 08:03:00,945 INFO [train.py:968] (0/2) Epoch 12, batch 23700, giga_loss[loss=0.3578, simple_loss=0.407, pruned_loss=0.1543, over 28608.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3888, pruned_loss=0.1367, over 5653606.58 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3635, pruned_loss=0.1107, over 5728982.78 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3886, pruned_loss=0.1365, over 5660287.07 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:03:50,961 INFO [train.py:968] (0/2) Epoch 12, batch 23750, giga_loss[loss=0.3796, simple_loss=0.4249, pruned_loss=0.1672, over 28230.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3913, pruned_loss=0.1399, over 5653569.18 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3632, pruned_loss=0.1107, over 5730132.55 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.392, pruned_loss=0.1404, over 5656144.86 frames. ], batch size: 368, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 08:03:58,256 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8129, 2.8490, 2.0029, 0.9029], device='cuda:0'), covar=tensor([0.5624, 0.2441, 0.2703, 0.5170], device='cuda:0'), in_proj_covar=tensor([0.1582, 0.1488, 0.1504, 0.1281], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 08:04:07,273 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 08:04:32,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.287e+02 1.665e+03 2.095e+03 2.597e+03 7.611e+03, threshold=4.190e+03, percent-clipped=9.0
+2023-03-06 08:04:35,606 INFO [train.py:968] (0/2) Epoch 12, batch 23800, giga_loss[loss=0.3654, simple_loss=0.4124, pruned_loss=0.1593, over 28960.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.394, pruned_loss=0.1432, over 5643892.71 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3636, pruned_loss=0.1116, over 5725505.77 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3954, pruned_loss=0.1441, over 5646516.00 frames. ], batch size: 227, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 08:05:33,252 INFO [train.py:968] (0/2) Epoch 12, batch 23850, libri_loss[loss=0.2614, simple_loss=0.3302, pruned_loss=0.09629, over 29552.00 frames. ], tot_loss[loss=0.345, simple_loss=0.398, pruned_loss=0.146, over 5647521.03 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3638, pruned_loss=0.1117, over 5729925.88 frames. ], giga_tot_loss[loss=0.3473, simple_loss=0.3998, pruned_loss=0.1474, over 5643573.21 frames. ], batch size: 76, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 08:06:23,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.131e+03 1.926e+03 2.679e+03 4.660e+03 1.936e+04, threshold=5.358e+03, percent-clipped=28.0
+2023-03-06 08:06:27,133 INFO [train.py:968] (0/2) Epoch 12, batch 23900, giga_loss[loss=0.3339, simple_loss=0.3894, pruned_loss=0.1392, over 28651.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.3985, pruned_loss=0.1469, over 5640619.33 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3639, pruned_loss=0.112, over 5722909.95 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4006, pruned_loss=0.1485, over 5642244.74 frames. ], batch size: 262, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 08:06:29,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:06:31,634 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:06:31,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4253, 2.0637, 1.5376, 0.4708], device='cuda:0'), covar=tensor([0.3586, 0.1974, 0.3166, 0.4707], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1493, 0.1504, 0.1283], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 08:07:07,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4498, 1.7002, 1.2725, 1.5928], device='cuda:0'), covar=tensor([0.0694, 0.0282, 0.0316, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 08:07:18,752 INFO [train.py:968] (0/2) Epoch 12, batch 23950, giga_loss[loss=0.3611, simple_loss=0.4146, pruned_loss=0.1538, over 28554.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3962, pruned_loss=0.1457, over 5644552.95 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3638, pruned_loss=0.112, over 5726340.36 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.3987, pruned_loss=0.1476, over 5641081.83 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 08:07:20,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5838, 1.8189, 1.5060, 1.8589], device='cuda:0'), covar=tensor([0.2040, 0.1907, 0.1999, 0.1764], device='cuda:0'), in_proj_covar=tensor([0.1311, 0.0971, 0.1155, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 08:07:39,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:08:02,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.707e+03 2.118e+03 2.783e+03 6.525e+03, threshold=4.236e+03, percent-clipped=2.0
+2023-03-06 08:08:07,015 INFO [train.py:968] (0/2) Epoch 12, batch 24000, giga_loss[loss=0.3442, simple_loss=0.3969, pruned_loss=0.1458, over 28847.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.395, pruned_loss=0.1448, over 5643515.81 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3636, pruned_loss=0.1122, over 5727435.78 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.3978, pruned_loss=0.1468, over 5638077.91 frames. ], batch size: 99, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:08:07,019 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 08:08:14,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9784, 1.1135, 3.3913, 2.9802], device='cuda:0'), covar=tensor([0.1942, 0.3114, 0.0542, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0672, 0.0596, 0.0872, 0.0775], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 08:08:15,697 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2144, simple_loss=0.3211, pruned_loss=0.05382, over 944034.00 frames. 
+2023-03-06 08:08:15,698 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 08:08:24,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:08:57,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:08:59,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:09:02,688 INFO [train.py:968] (0/2) Epoch 12, batch 24050, giga_loss[loss=0.4253, simple_loss=0.4534, pruned_loss=0.1986, over 27944.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3956, pruned_loss=0.1442, over 5644686.49 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3639, pruned_loss=0.1126, over 5732238.75 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.3983, pruned_loss=0.1463, over 5634089.15 frames. ], batch size: 412, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:09:29,371 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:09:50,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.660e+02 1.711e+03 2.272e+03 3.909e+03 7.917e+03, threshold=4.544e+03, percent-clipped=20.0
+2023-03-06 08:09:53,635 INFO [train.py:968] (0/2) Epoch 12, batch 24100, giga_loss[loss=0.3367, simple_loss=0.3961, pruned_loss=0.1387, over 28504.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3977, pruned_loss=0.1457, over 5633779.92 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3641, pruned_loss=0.1128, over 5726435.02 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4004, pruned_loss=0.1479, over 5628208.44 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:10:08,045 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:10:10,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:10:22,130 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525319.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 08:10:40,002 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:10:42,363 INFO [train.py:968] (0/2) Epoch 12, batch 24150, giga_loss[loss=0.2901, simple_loss=0.3662, pruned_loss=0.107, over 28999.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3967, pruned_loss=0.1446, over 5609456.66 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3648, pruned_loss=0.1134, over 5701431.39 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.3992, pruned_loss=0.1466, over 5625187.28 frames. ], batch size: 128, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:10:59,403 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:11:06,902 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:11:29,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.663e+03 2.111e+03 2.866e+03 6.940e+03, threshold=4.221e+03, percent-clipped=2.0
+2023-03-06 08:11:32,635 INFO [train.py:968] (0/2) Epoch 12, batch 24200, giga_loss[loss=0.3203, simple_loss=0.3961, pruned_loss=0.1222, over 28742.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3931, pruned_loss=0.14, over 5620463.68 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3646, pruned_loss=0.1134, over 5703779.44 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3956, pruned_loss=0.142, over 5629826.95 frames. ], batch size: 119, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:11:41,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-06 08:12:26,606 INFO [train.py:968] (0/2) Epoch 12, batch 24250, giga_loss[loss=0.3544, simple_loss=0.3975, pruned_loss=0.1557, over 26639.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3896, pruned_loss=0.1364, over 5632249.62 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3645, pruned_loss=0.1134, over 5704825.54 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3918, pruned_loss=0.1381, over 5638246.79 frames. ], batch size: 555, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:12:51,611 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:13:05,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.964e+02 1.600e+03 2.162e+03 3.008e+03 1.152e+04, threshold=4.324e+03, percent-clipped=11.0
+2023-03-06 08:13:10,611 INFO [train.py:968] (0/2) Epoch 12, batch 24300, giga_loss[loss=0.3116, simple_loss=0.3728, pruned_loss=0.1252, over 28833.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3864, pruned_loss=0.1337, over 5653774.42 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3642, pruned_loss=0.1137, over 5708420.45 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3894, pruned_loss=0.1357, over 5653330.85 frames. ], batch size: 119, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:13:27,928 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-06 08:14:00,103 INFO [train.py:968] (0/2) Epoch 12, batch 24350, libri_loss[loss=0.3299, simple_loss=0.3922, pruned_loss=0.1337, over 29684.00 frames. ], tot_loss[loss=0.324, simple_loss=0.384, pruned_loss=0.132, over 5646779.45 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.365, pruned_loss=0.1144, over 5702326.19 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.386, pruned_loss=0.1333, over 5650644.93 frames. ], batch size: 88, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:14:22,399 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:14:36,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:14:46,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.273e+02 1.686e+03 2.244e+03 3.087e+03 8.982e+03, threshold=4.487e+03, percent-clipped=9.0
+2023-03-06 08:14:48,816 INFO [train.py:968] (0/2) Epoch 12, batch 24400, giga_loss[loss=0.3301, simple_loss=0.3908, pruned_loss=0.1347, over 28625.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3824, pruned_loss=0.1306, over 5661914.98 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3649, pruned_loss=0.1145, over 5705365.43 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3843, pruned_loss=0.1317, over 5661775.27 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 08:15:13,825 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:15:17,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:15:41,107 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 08:15:44,909 INFO [train.py:968] (0/2) Epoch 12, batch 24450, giga_loss[loss=0.2631, simple_loss=0.3433, pruned_loss=0.09147, over 29010.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3823, pruned_loss=0.1302, over 5662487.49 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.365, pruned_loss=0.1146, over 5707569.17 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.384, pruned_loss=0.1313, over 5659825.68 frames. ], batch size: 136, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:15:47,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:16:16,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8368, 3.6512, 3.4521, 1.7154], device='cuda:0'), covar=tensor([0.0678, 0.0788, 0.0765, 0.2052], device='cuda:0'), in_proj_covar=tensor([0.1071, 0.1004, 0.0871, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 08:16:21,597 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:16:30,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.222e+02 1.499e+03 1.841e+03 2.288e+03 5.442e+03, threshold=3.682e+03, percent-clipped=6.0
+2023-03-06 08:16:33,741 INFO [train.py:968] (0/2) Epoch 12, batch 24500, giga_loss[loss=0.3793, simple_loss=0.4088, pruned_loss=0.1749, over 26683.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3798, pruned_loss=0.128, over 5669529.91 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3655, pruned_loss=0.115, over 5713254.22 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3813, pruned_loss=0.129, over 5660285.32 frames. ], batch size: 555, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:16:37,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525694.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 08:16:44,812 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:17:06,883 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:17:08,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:17:18,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:17:23,248 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:17:25,282 INFO [train.py:968] (0/2) Epoch 12, batch 24550, giga_loss[loss=0.3455, simple_loss=0.4148, pruned_loss=0.1381, over 28725.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3804, pruned_loss=0.1256, over 5686998.58 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3653, pruned_loss=0.1151, over 5717296.93 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3822, pruned_loss=0.1265, over 5674911.28 frames. ], batch size: 243, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:17:40,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:18:14,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.607e+02 1.570e+03 2.093e+03 2.762e+03 8.201e+03, threshold=4.186e+03, percent-clipped=12.0
+2023-03-06 08:18:18,060 INFO [train.py:968] (0/2) Epoch 12, batch 24600, giga_loss[loss=0.2863, simple_loss=0.3603, pruned_loss=0.1061, over 28942.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3821, pruned_loss=0.1264, over 5659988.62 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3659, pruned_loss=0.1156, over 5712363.82 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3834, pruned_loss=0.127, over 5653679.84 frames. ], batch size: 119, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:18:33,525 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-06 08:18:39,322 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 08:19:02,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525837.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 08:19:06,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525840.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 08:19:06,937 INFO [train.py:968] (0/2) Epoch 12, batch 24650, giga_loss[loss=0.2966, simple_loss=0.3669, pruned_loss=0.1131, over 28869.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3839, pruned_loss=0.1285, over 5652190.48 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3666, pruned_loss=0.1162, over 5703174.20 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3846, pruned_loss=0.1286, over 5653892.42 frames. ], batch size: 199, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:19:31,538 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525869.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 08:19:36,457 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:19:39,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:19:42,705 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:19:45,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:19:48,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.948e+03 2.845e+03 4.213e+03 1.497e+04, threshold=5.690e+03, percent-clipped=27.0
+2023-03-06 08:19:51,713 INFO [train.py:968] (0/2) Epoch 12, batch 24700, giga_loss[loss=0.321, simple_loss=0.3643, pruned_loss=0.1388, over 23302.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.385, pruned_loss=0.1302, over 5649482.15 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3671, pruned_loss=0.1168, over 5701799.15 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3855, pruned_loss=0.1301, over 5651121.63 frames. ], batch size: 705, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:20:09,347 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:20:14,269 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:20:19,587 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 08:20:38,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:20:40,455 INFO [train.py:968] (0/2) Epoch 12, batch 24750, giga_loss[loss=0.308, simple_loss=0.3655, pruned_loss=0.1253, over 27916.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3823, pruned_loss=0.1295, over 5652103.54 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3675, pruned_loss=0.1171, over 5706745.28 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3828, pruned_loss=0.1296, over 5647543.81 frames. ], batch size: 412, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:21:17,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.510e+02 1.728e+03 2.120e+03 2.945e+03 6.535e+03, threshold=4.240e+03, percent-clipped=2.0
+2023-03-06 08:21:18,373 INFO [train.py:968] (0/2) Epoch 12, batch 24800, giga_loss[loss=0.3025, simple_loss=0.3663, pruned_loss=0.1194, over 28775.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3805, pruned_loss=0.1287, over 5658583.45 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3672, pruned_loss=0.117, over 5700118.01 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3819, pruned_loss=0.1294, over 5658940.70 frames. ], batch size: 284, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:21:25,353 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-526000.pt
+2023-03-06 08:22:04,262 INFO [train.py:968] (0/2) Epoch 12, batch 24850, libri_loss[loss=0.2784, simple_loss=0.342, pruned_loss=0.1074, over 29578.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3793, pruned_loss=0.1277, over 5648813.69 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3669, pruned_loss=0.117, over 5685301.68 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3809, pruned_loss=0.1285, over 5661165.13 frames. ], batch size: 77, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:22:12,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:22:36,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:22:37,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9119, 1.8656, 1.4291, 1.5216], device='cuda:0'), covar=tensor([0.0803, 0.0676, 0.1002, 0.1091], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0442, 0.0502, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 08:22:40,876 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:22:43,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:22:46,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.023e+02 1.309e+03 1.639e+03 2.114e+03 5.046e+03, threshold=3.277e+03, percent-clipped=2.0
+2023-03-06 08:22:48,121 INFO [train.py:968] (0/2) Epoch 12, batch 24900, giga_loss[loss=0.3514, simple_loss=0.3966, pruned_loss=0.1531, over 27603.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3782, pruned_loss=0.126, over 5661928.47 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3667, pruned_loss=0.1171, over 5687313.35 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.38, pruned_loss=0.1267, over 5669342.82 frames. ], batch size: 472, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:23:12,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:23:13,203 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 08:23:35,072 INFO [train.py:968] (0/2) Epoch 12, batch 24950, giga_loss[loss=0.3098, simple_loss=0.3784, pruned_loss=0.1206, over 28725.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3787, pruned_loss=0.1259, over 5661508.21 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3674, pruned_loss=0.1176, over 5692379.55 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3797, pruned_loss=0.1262, over 5662011.96 frames. ], batch size: 262, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:23:37,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3115, 1.2855, 1.2265, 1.4231], device='cuda:0'), covar=tensor([0.0771, 0.0354, 0.0333, 0.0870], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0091], device='cuda:0')
+2023-03-06 08:24:17,563 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=526184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:23,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.342e+02 1.541e+03 1.990e+03 2.722e+03 8.244e+03, threshold=3.981e+03, percent-clipped=11.0
+2023-03-06 08:24:25,385 INFO [train.py:968] (0/2) Epoch 12, batch 25000, giga_loss[loss=0.2864, simple_loss=0.3585, pruned_loss=0.1071, over 28611.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3779, pruned_loss=0.1254, over 5657284.40 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3678, pruned_loss=0.1178, over 5684328.75 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3786, pruned_loss=0.1256, over 5664510.89 frames. ], batch size: 78, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:24:26,232 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=526193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:28,487 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:31,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:51,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:54,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:58,762 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:25:01,339 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.82 vs. limit=5.0
+2023-03-06 08:25:10,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=526240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:25:12,204 INFO [train.py:968] (0/2) Epoch 12, batch 25050, giga_loss[loss=0.344, simple_loss=0.3941, pruned_loss=0.147, over 28506.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.377, pruned_loss=0.125, over 5669604.67 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3678, pruned_loss=0.118, over 5680204.57 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3778, pruned_loss=0.1252, over 5679216.85 frames. ], batch size: 85, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:25:23,453 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:25:58,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.996e+02 1.405e+03 1.754e+03 2.504e+03 5.704e+03, threshold=3.508e+03, percent-clipped=2.0
+2023-03-06 08:26:01,355 INFO [train.py:968] (0/2) Epoch 12, batch 25100, libri_loss[loss=0.2916, simple_loss=0.3697, pruned_loss=0.1067, over 29756.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3758, pruned_loss=0.1248, over 5670101.73 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3679, pruned_loss=0.1179, over 5685035.07 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3765, pruned_loss=0.1252, over 5672893.16 frames. ], batch size: 87, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:26:36,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=526329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:26:48,048 INFO [train.py:968] (0/2) Epoch 12, batch 25150, giga_loss[loss=0.323, simple_loss=0.3837, pruned_loss=0.1312, over 28553.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3741, pruned_loss=0.124, over 5677621.75 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3681, pruned_loss=0.118, over 5682481.89 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3745, pruned_loss=0.1243, over 5682320.56 frames. ], batch size: 336, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:27:05,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2749, 1.5484, 1.2731, 1.5489], device='cuda:0'), covar=tensor([0.0755, 0.0320, 0.0319, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 08:27:37,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.657e+03 2.250e+03 3.028e+03 6.657e+03, threshold=4.501e+03, percent-clipped=18.0
+2023-03-06 08:27:38,951 INFO [train.py:968] (0/2) Epoch 12, batch 25200, giga_loss[loss=0.2764, simple_loss=0.3419, pruned_loss=0.1054, over 28755.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.373, pruned_loss=0.1236, over 5679360.83 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.368, pruned_loss=0.1179, over 5684878.89 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3735, pruned_loss=0.124, over 5681012.07 frames. ], batch size: 85, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:28:25,201 INFO [train.py:968] (0/2) Epoch 12, batch 25250, giga_loss[loss=0.3673, simple_loss=0.3931, pruned_loss=0.1707, over 23580.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3726, pruned_loss=0.1242, over 5662950.85 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.368, pruned_loss=0.118, over 5668813.45 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.373, pruned_loss=0.1245, over 5678185.16 frames. ], batch size: 705, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:29:12,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.963e+02 1.634e+03 2.301e+03 3.536e+03 1.077e+04, threshold=4.602e+03, percent-clipped=14.0
+2023-03-06 08:29:13,237 INFO [train.py:968] (0/2) Epoch 12, batch 25300, giga_loss[loss=0.2914, simple_loss=0.3608, pruned_loss=0.111, over 28882.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3731, pruned_loss=0.1244, over 5670003.75 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3679, pruned_loss=0.1179, over 5671552.85 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3736, pruned_loss=0.1248, over 5679351.34 frames. ], batch size: 186, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:30:00,599 INFO [train.py:968] (0/2) Epoch 12, batch 25350, giga_loss[loss=0.3461, simple_loss=0.4062, pruned_loss=0.143, over 28599.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3743, pruned_loss=0.1245, over 5666265.53 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3683, pruned_loss=0.1182, over 5663237.05 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3744, pruned_loss=0.1246, over 5681117.67 frames. ], batch size: 307, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:30:16,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:30:24,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:30:29,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4402, 3.5466, 1.5718, 1.5247], device='cuda:0'), covar=tensor([0.0943, 0.0359, 0.0886, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0517, 0.0345, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 08:30:46,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.807e+02 1.488e+03 1.940e+03 2.823e+03 7.059e+03, threshold=3.881e+03, percent-clipped=8.0
+2023-03-06 08:30:47,415 INFO [train.py:968] (0/2) Epoch 12, batch 25400, giga_loss[loss=0.2927, simple_loss=0.363, pruned_loss=0.1113, over 28983.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3736, pruned_loss=0.1231, over 5672391.02 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3683, pruned_loss=0.1182, over 5665642.61 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3737, pruned_loss=0.1233, over 5681995.41 frames. ], batch size: 136, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:31:11,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:31:29,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 08:31:34,669 INFO [train.py:968] (0/2) Epoch 12, batch 25450, giga_loss[loss=0.2952, simple_loss=0.365, pruned_loss=0.1127, over 28899.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3736, pruned_loss=0.1226, over 5676615.66 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3686, pruned_loss=0.1184, over 5668469.80 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3735, pruned_loss=0.1226, over 5681731.47 frames. ], batch size: 227, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:31:44,803 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=526655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:16,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.654e+02 1.481e+03 1.963e+03 2.524e+03 1.047e+04, threshold=3.925e+03, percent-clipped=9.0
+2023-03-06 08:32:17,612 INFO [train.py:968] (0/2) Epoch 12, batch 25500, libri_loss[loss=0.2793, simple_loss=0.3396, pruned_loss=0.1094, over 28248.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3746, pruned_loss=0.124, over 5675935.24 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3688, pruned_loss=0.1188, over 5668691.34 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3746, pruned_loss=0.1239, over 5680638.30 frames. ], batch size: 62, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:32:27,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:30,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:31,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:37,959 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:40,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:58,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:33:04,828 INFO [train.py:968] (0/2) Epoch 12, batch 25550, giga_loss[loss=0.2635, simple_loss=0.3362, pruned_loss=0.09541, over 28625.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3782, pruned_loss=0.128, over 5679251.37 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3696, pruned_loss=0.1196, over 5673647.72 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3777, pruned_loss=0.1274, over 5678611.41 frames. ], batch size: 60, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:33:05,879 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:33:23,912 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:33:27,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:33:52,967 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:33:55,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.997e+03 2.568e+03 4.189e+03 1.197e+04, threshold=5.135e+03, percent-clipped=25.0
+2023-03-06 08:33:55,883 INFO [train.py:968] (0/2) Epoch 12, batch 25600, giga_loss[loss=0.4759, simple_loss=0.4767, pruned_loss=0.2375, over 26474.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3791, pruned_loss=0.1304, over 5666776.38 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3698, pruned_loss=0.1198, over 5665665.80 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3786, pruned_loss=0.1299, over 5674158.68 frames. ], batch size: 555, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:34:11,859 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-06 08:34:48,308 INFO [train.py:968] (0/2) Epoch 12, batch 25650, giga_loss[loss=0.2702, simple_loss=0.3373, pruned_loss=0.1015, over 28523.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3793, pruned_loss=0.1314, over 5675260.87 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3697, pruned_loss=0.1198, over 5671414.83 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3792, pruned_loss=0.1311, over 5676272.78 frames. ], batch size: 78, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:34:53,753 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:34:56,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:35:15,783 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3067, 1.5737, 1.3146, 1.4831], device='cuda:0'), covar=tensor([0.0721, 0.0387, 0.0332, 0.0757], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 08:35:22,544 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:35:34,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.861e+02 1.488e+03 2.058e+03 2.852e+03 5.973e+03, threshold=4.117e+03, percent-clipped=5.0
+2023-03-06 08:35:34,366 INFO [train.py:968] (0/2) Epoch 12, batch 25700, giga_loss[loss=0.3418, simple_loss=0.3814, pruned_loss=0.1511, over 24304.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3772, pruned_loss=0.1299, over 5668273.22 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3695, pruned_loss=0.1198, over 5672449.76 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3773, pruned_loss=0.1299, over 5668175.96 frames. ], batch size: 705, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:36:05,351 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 08:36:17,507 INFO [train.py:968] (0/2) Epoch 12, batch 25750, libri_loss[loss=0.2806, simple_loss=0.3546, pruned_loss=0.1033, over 27835.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3767, pruned_loss=0.1292, over 5653371.85 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3701, pruned_loss=0.1204, over 5658103.70 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3766, pruned_loss=0.129, over 5666887.03 frames. ], batch size: 116, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:36:31,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3339, 1.6316, 1.4385, 1.3650], device='cuda:0'), covar=tensor([0.1798, 0.2000, 0.2081, 0.2258], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0731, 0.0671, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 08:36:56,672 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9115, 1.8325, 1.2939, 1.6390], device='cuda:0'), covar=tensor([0.0668, 0.0512, 0.0973, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0442, 0.0502, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 08:37:00,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.131e+03 1.723e+03 2.570e+03 3.761e+03 1.131e+04, threshold=5.141e+03, percent-clipped=20.0
+2023-03-06 08:37:00,472 INFO [train.py:968] (0/2) Epoch 12, batch 25800, giga_loss[loss=0.3368, simple_loss=0.3771, pruned_loss=0.1482, over 26545.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3758, pruned_loss=0.1268, over 5665882.95 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3699, pruned_loss=0.1204, over 5663714.26 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.376, pruned_loss=0.1268, over 5671501.01 frames. ], batch size: 555, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:37:28,852 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=527021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:37:38,014 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:37:47,774 INFO [train.py:968] (0/2) Epoch 12, batch 25850, giga_loss[loss=0.2899, simple_loss=0.3562, pruned_loss=0.1118, over 28776.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3733, pruned_loss=0.125, over 5656535.15 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 5664608.85 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3735, pruned_loss=0.1251, over 5660143.59 frames. ], batch size: 92, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:38:34,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.728e+02 1.571e+03 1.931e+03 2.429e+03 6.010e+03, threshold=3.862e+03, percent-clipped=2.0
+2023-03-06 08:38:34,387 INFO [train.py:968] (0/2) Epoch 12, batch 25900, giga_loss[loss=0.2759, simple_loss=0.35, pruned_loss=0.1009, over 29009.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3714, pruned_loss=0.1241, over 5662114.44 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 5665952.31 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3716, pruned_loss=0.1243, over 5663590.58 frames. ], batch size: 164, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:39:26,651 INFO [train.py:968] (0/2) Epoch 12, batch 25950, giga_loss[loss=0.3304, simple_loss=0.3841, pruned_loss=0.1384, over 27894.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3715, pruned_loss=0.1252, over 5649169.79 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.37, pruned_loss=0.1205, over 5669264.97 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3717, pruned_loss=0.1253, over 5647542.05 frames. ], batch size: 412, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:39:55,048 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=527173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:39:58,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=527176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:40:12,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-06 08:40:15,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.559e+02 1.635e+03 2.208e+03 3.260e+03 6.780e+03, threshold=4.416e+03, percent-clipped=16.0
+2023-03-06 08:40:15,683 INFO [train.py:968] (0/2) Epoch 12, batch 26000, giga_loss[loss=0.3029, simple_loss=0.3721, pruned_loss=0.1169, over 28957.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3742, pruned_loss=0.1271, over 5655689.12 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3697, pruned_loss=0.1203, over 5673616.58 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3747, pruned_loss=0.1274, over 5650297.89 frames. ], batch size: 213, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:40:16,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-06 08:40:23,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=527202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:40:26,287 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=527205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:40:41,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3475, 1.5669, 1.2640, 1.2517], device='cuda:0'), covar=tensor([0.2590, 0.2559, 0.2914, 0.2221], device='cuda:0'), in_proj_covar=tensor([0.1321, 0.0979, 0.1164, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 08:41:00,575 INFO [train.py:968] (0/2) Epoch 12, batch 26050, giga_loss[loss=0.3038, simple_loss=0.3811, pruned_loss=0.1133, over 28772.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3786, pruned_loss=0.1277, over 5665109.29 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3698, pruned_loss=0.1205, over 5674806.57 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3789, pruned_loss=0.1279, over 5659464.62 frames. ], batch size: 119, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:41:33,603 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=527275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:41:47,606 INFO [train.py:968] (0/2) Epoch 12, batch 26100, giga_loss[loss=0.3352, simple_loss=0.3954, pruned_loss=0.1375, over 28727.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3821, pruned_loss=0.128, over 5668579.94 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3702, pruned_loss=0.1207, over 5677346.08 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3822, pruned_loss=0.1281, over 5661694.63 frames. ], batch size: 262, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:41:50,114 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.925e+02 1.454e+03 1.858e+03 2.516e+03 7.953e+03, threshold=3.717e+03, percent-clipped=5.0
+2023-03-06 08:42:27,015 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-06 08:42:36,693 INFO [train.py:968] (0/2) Epoch 12, batch 26150, libri_loss[loss=0.3787, simple_loss=0.418, pruned_loss=0.1697, over 29050.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3845, pruned_loss=0.1304, over 5668159.17 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3707, pruned_loss=0.1213, over 5683521.10 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3845, pruned_loss=0.1301, over 5656464.49 frames. ], batch size: 101, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:43:10,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2620, 5.1067, 4.7808, 2.1148], device='cuda:0'), covar=tensor([0.0414, 0.0542, 0.0588, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.1081, 0.1013, 0.0883, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 08:43:20,643 INFO [train.py:968] (0/2) Epoch 12, batch 26200, giga_loss[loss=0.3648, simple_loss=0.4095, pruned_loss=0.1601, over 28261.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3862, pruned_loss=0.1324, over 5660439.15 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3713, pruned_loss=0.1219, over 5685077.20 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.386, pruned_loss=0.1319, over 5649527.48 frames. ], batch size: 368, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:43:21,995 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.713e+02 1.628e+03 2.203e+03 3.368e+03 1.682e+04, threshold=4.405e+03, percent-clipped=20.0
+2023-03-06 08:43:23,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:43:38,960 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8214, 1.0622, 2.8243, 2.6323], device='cuda:0'), covar=tensor([0.1603, 0.2434, 0.0607, 0.1303], device='cuda:0'), in_proj_covar=tensor([0.0676, 0.0598, 0.0878, 0.0782], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 08:44:03,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3287, 1.6536, 1.2783, 1.5102], device='cuda:0'), covar=tensor([0.0684, 0.0356, 0.0319, 0.0733], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0209], device='cuda:0'), out_proj_covar=tensor([0.0082, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 08:44:09,703 INFO [train.py:968] (0/2) Epoch 12, batch 26250, giga_loss[loss=0.2551, simple_loss=0.3335, pruned_loss=0.08836, over 28678.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3861, pruned_loss=0.1332, over 5653579.96 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3711, pruned_loss=0.1219, over 5687331.61 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3862, pruned_loss=0.133, over 5642775.78 frames. ], batch size: 92, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:44:31,849 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=527467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:44:58,194 INFO [train.py:968] (0/2) Epoch 12, batch 26300, giga_loss[loss=0.2818, simple_loss=0.3579, pruned_loss=0.1029, over 29059.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3847, pruned_loss=0.1329, over 5645699.82 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3712, pruned_loss=0.122, over 5682531.86 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.385, pruned_loss=0.1328, over 5639631.00 frames. ], batch size: 164, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:44:59,443 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.090e+02 1.630e+03 2.205e+03 3.037e+03 9.497e+03, threshold=4.410e+03, percent-clipped=7.0
+2023-03-06 08:45:21,741 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-06 08:45:36,589 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-06 08:45:39,091 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5384, 1.5335, 1.1725, 1.1558], device='cuda:0'), covar=tensor([0.0672, 0.0466, 0.0910, 0.1196], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0443, 0.0502, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 08:45:40,967 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=527539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:45:42,689 INFO [train.py:968] (0/2) Epoch 12, batch 26350, giga_loss[loss=0.29, simple_loss=0.3523, pruned_loss=0.1139, over 29044.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3809, pruned_loss=0.1308, over 5636014.91 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3713, pruned_loss=0.1222, over 5664711.16 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3813, pruned_loss=0.1307, over 5646370.14 frames. ], batch size: 128, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:45:43,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=527542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:46:09,459 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=527571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:46:16,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:46:18,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2803, 2.8748, 1.4760, 1.3738], device='cuda:0'), covar=tensor([0.0934, 0.0318, 0.0811, 0.1285], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0518, 0.0345, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 08:46:30,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=527587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:46:35,449 INFO [train.py:968] (0/2) Epoch 12, batch 26400, giga_loss[loss=0.2727, simple_loss=0.3482, pruned_loss=0.09862, over 28749.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3794, pruned_loss=0.1307, over 5638522.01 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3711, pruned_loss=0.1223, over 5670724.31 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3801, pruned_loss=0.1308, over 5640816.50 frames. ], batch size: 60, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:46:36,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.482e+02 1.598e+03 1.938e+03 2.869e+03 1.079e+04, threshold=3.877e+03, percent-clipped=5.0
+2023-03-06 08:47:22,537 INFO [train.py:968] (0/2) Epoch 12, batch 26450, libri_loss[loss=0.317, simple_loss=0.384, pruned_loss=0.125, over 29540.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3796, pruned_loss=0.1308, over 5643956.20 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3714, pruned_loss=0.1223, over 5676584.17 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.38, pruned_loss=0.1311, over 5639503.98 frames. ], batch size: 83, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:47:28,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:48:02,706 INFO [train.py:968] (0/2) Epoch 12, batch 26500, giga_loss[loss=0.3096, simple_loss=0.369, pruned_loss=0.1251, over 28669.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3776, pruned_loss=0.1294, over 5649113.11 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3711, pruned_loss=0.1222, over 5673919.33 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3784, pruned_loss=0.13, over 5647601.20 frames. ], batch size: 85, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:48:04,048 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.438e+02 1.624e+03 2.121e+03 3.162e+03 1.022e+04, threshold=4.242e+03, percent-clipped=16.0
+2023-03-06 08:48:27,122 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=527720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:48:29,839 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=527723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:48:47,330 INFO [train.py:968] (0/2) Epoch 12, batch 26550, giga_loss[loss=0.3458, simple_loss=0.392, pruned_loss=0.1498, over 27673.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3763, pruned_loss=0.1288, over 5665756.66 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3715, pruned_loss=0.1225, over 5678936.06 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3768, pruned_loss=0.1292, over 5659203.99 frames. ], batch size: 472, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:48:58,071 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=527752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:49:38,360 INFO [train.py:968] (0/2) Epoch 12, batch 26600, giga_loss[loss=0.4166, simple_loss=0.4439, pruned_loss=0.1946, over 28024.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3764, pruned_loss=0.1289, over 5672934.41 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3717, pruned_loss=0.1225, over 5683280.55 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3767, pruned_loss=0.1293, over 5663515.00 frames. ], batch size: 412, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:49:40,999 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=527793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:49:41,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.566e+02 1.779e+03 2.265e+03 2.945e+03 5.847e+03, threshold=4.529e+03, percent-clipped=10.0
+2023-03-06 08:49:43,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=527796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:50:09,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=527825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:50:25,723 INFO [train.py:968] (0/2) Epoch 12, batch 26650, giga_loss[loss=0.3862, simple_loss=0.4149, pruned_loss=0.1787, over 26644.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3767, pruned_loss=0.1282, over 5669089.62 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3712, pruned_loss=0.1222, over 5686438.95 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3775, pruned_loss=0.1289, over 5658546.78 frames. ], batch size: 555, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:50:26,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:50:36,298 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 08:51:04,728 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-06 08:51:15,729 INFO [train.py:968] (0/2) Epoch 12, batch 26700, giga_loss[loss=0.2995, simple_loss=0.3557, pruned_loss=0.1217, over 28377.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3801, pruned_loss=0.1306, over 5668237.38 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3714, pruned_loss=0.1225, over 5689779.66 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3806, pruned_loss=0.131, over 5656853.22 frames. ], batch size: 71, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:51:19,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.986e+02 1.496e+03 1.721e+03 2.141e+03 4.517e+03, threshold=3.443e+03, percent-clipped=0.0
+2023-03-06 08:51:53,132 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 08:51:54,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8072, 1.1232, 2.8232, 2.6733], device='cuda:0'), covar=tensor([0.1588, 0.2347, 0.0599, 0.1468], device='cuda:0'), in_proj_covar=tensor([0.0679, 0.0600, 0.0878, 0.0785], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 08:52:01,407 INFO [train.py:968] (0/2) Epoch 12, batch 26750, giga_loss[loss=0.3001, simple_loss=0.3656, pruned_loss=0.1173, over 28688.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3797, pruned_loss=0.1312, over 5667130.42 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3707, pruned_loss=0.1221, over 5692461.00 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.381, pruned_loss=0.1321, over 5654691.42 frames. ], batch size: 284, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:52:14,350 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-06 08:52:18,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:52:37,476 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=527985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:52:39,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=527988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:52:43,465 INFO [train.py:968] (0/2) Epoch 12, batch 26800, libri_loss[loss=0.3258, simple_loss=0.3892, pruned_loss=0.1311, over 28926.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3804, pruned_loss=0.1285, over 5674387.79 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3711, pruned_loss=0.1225, over 5687936.18 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3814, pruned_loss=0.1291, over 5667822.31 frames. ], batch size: 107, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:52:44,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.582e+02 1.451e+03 1.800e+03 2.537e+03 9.313e+03, threshold=3.600e+03, percent-clipped=11.0
+2023-03-06 08:52:50,044 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-528000.pt
+2023-03-06 08:53:04,497 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=528017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:53:23,168 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4445, 1.5492, 1.3157, 1.6418], device='cuda:0'), covar=tensor([0.2566, 0.2549, 0.2845, 0.2030], device='cuda:0'), in_proj_covar=tensor([0.1318, 0.0978, 0.1160, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 08:53:26,083 INFO [train.py:968] (0/2) Epoch 12, batch 26850, giga_loss[loss=0.3199, simple_loss=0.3887, pruned_loss=0.1256, over 28726.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3824, pruned_loss=0.1289, over 5665091.31 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.1231, over 5676442.00 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.383, pruned_loss=0.129, over 5668883.52 frames. ], batch size: 242, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:53:29,468 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=528044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:53:31,253 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-06 08:53:57,400 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7104, 1.7494, 1.6273, 1.5073], device='cuda:0'), covar=tensor([0.1293, 0.1944, 0.1712, 0.1862], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0735, 0.0673, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 08:54:07,979 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-06 08:54:10,130 INFO [train.py:968] (0/2) Epoch 12, batch 26900, giga_loss[loss=0.3049, simple_loss=0.3718, pruned_loss=0.119, over 28859.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3831, pruned_loss=0.1277, over 5659185.71 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3714, pruned_loss=0.123, over 5670933.37 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.384, pruned_loss=0.128, over 5667084.20 frames. ], batch size: 186, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:54:12,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.527e+02 1.453e+03 1.879e+03 2.280e+03 6.723e+03, threshold=3.758e+03, percent-clipped=4.0
+2023-03-06 08:54:20,720 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=528105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:54:24,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=528108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:54:53,726 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=528137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:54:58,644 INFO [train.py:968] (0/2) Epoch 12, batch 26950, giga_loss[loss=0.3626, simple_loss=0.4205, pruned_loss=0.1523, over 28626.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3866, pruned_loss=0.1316, over 5660636.21 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.1229, over 5675883.35 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3877, pruned_loss=0.132, over 5662495.93 frames. ], batch size: 92, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:55:47,932 INFO [train.py:968] (0/2) Epoch 12, batch 27000, giga_loss[loss=0.2839, simple_loss=0.3597, pruned_loss=0.104, over 28923.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3871, pruned_loss=0.1327, over 5672388.42 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.371, pruned_loss=0.1228, over 5677715.04 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3883, pruned_loss=0.1332, over 5672176.31 frames. ], batch size: 174, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:55:47,937 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 08:55:56,465 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2134, simple_loss=0.3197, pruned_loss=0.05357, over 944034.00 frames. 
+2023-03-06 08:55:56,466 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 08:55:58,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.677e+03 2.139e+03 2.937e+03 1.019e+04, threshold=4.277e+03, percent-clipped=15.0
+2023-03-06 08:56:36,320 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=528236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:56:42,933 INFO [train.py:968] (0/2) Epoch 12, batch 27050, giga_loss[loss=0.3387, simple_loss=0.3998, pruned_loss=0.1389, over 29014.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3884, pruned_loss=0.1346, over 5672493.72 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3709, pruned_loss=0.1228, over 5681651.13 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3898, pruned_loss=0.1353, over 5668673.86 frames. ], batch size: 128, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:57:29,186 INFO [train.py:968] (0/2) Epoch 12, batch 27100, libri_loss[loss=0.2884, simple_loss=0.3622, pruned_loss=0.1072, over 29526.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3851, pruned_loss=0.1317, over 5676439.41 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3701, pruned_loss=0.1221, over 5685598.34 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3878, pruned_loss=0.1334, over 5669105.55 frames. ], batch size: 81, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:57:31,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.777e+02 1.554e+03 1.871e+03 2.770e+03 5.860e+03, threshold=3.742e+03, percent-clipped=2.0
+2023-03-06 08:58:17,030 INFO [train.py:968] (0/2) Epoch 12, batch 27150, giga_loss[loss=0.2778, simple_loss=0.366, pruned_loss=0.09483, over 28928.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3843, pruned_loss=0.1296, over 5669123.06 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3697, pruned_loss=0.1218, over 5687889.85 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3869, pruned_loss=0.1313, over 5661188.69 frames. ], batch size: 106, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:58:31,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4465, 2.4755, 2.3414, 2.3087], device='cuda:0'), covar=tensor([0.1475, 0.2146, 0.1806, 0.2037], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0730, 0.0671, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 08:59:04,584 INFO [train.py:968] (0/2) Epoch 12, batch 27200, giga_loss[loss=0.3847, simple_loss=0.4316, pruned_loss=0.1688, over 27937.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3852, pruned_loss=0.1292, over 5661330.85 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3699, pruned_loss=0.122, over 5685443.81 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3873, pruned_loss=0.1305, over 5657212.01 frames. ], batch size: 412, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:59:07,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.634e+02 1.538e+03 1.828e+03 2.603e+03 6.431e+03, threshold=3.657e+03, percent-clipped=7.0
+2023-03-06 08:59:12,157 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=528400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:59:33,426 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=528419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:59:55,610 INFO [train.py:968] (0/2) Epoch 12, batch 27250, giga_loss[loss=0.4096, simple_loss=0.4471, pruned_loss=0.1861, over 28552.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3876, pruned_loss=0.1311, over 5660082.92 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5688579.01 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.389, pruned_loss=0.132, over 5653756.34 frames. ], batch size: 336, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:00:03,825 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-06 09:00:44,101 INFO [train.py:968] (0/2) Epoch 12, batch 27300, giga_loss[loss=0.2589, simple_loss=0.3399, pruned_loss=0.08895, over 29054.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3869, pruned_loss=0.131, over 5661192.84 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3705, pruned_loss=0.1223, over 5689711.11 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.388, pruned_loss=0.1317, over 5655008.55 frames. ], batch size: 128, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:00:47,957 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.919e+02 1.509e+03 1.954e+03 2.897e+03 1.301e+04, threshold=3.907e+03, percent-clipped=13.0
+2023-03-06 09:01:16,588 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=528524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:01:34,234 INFO [train.py:968] (0/2) Epoch 12, batch 27350, giga_loss[loss=0.3587, simple_loss=0.4049, pruned_loss=0.1563, over 28574.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3857, pruned_loss=0.1315, over 5658454.88 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3707, pruned_loss=0.1225, over 5686007.24 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3869, pruned_loss=0.1322, over 5656574.26 frames. ], batch size: 307, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:01:52,943 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=528562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:01:55,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=528565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:02:20,142 INFO [train.py:968] (0/2) Epoch 12, batch 27400, giga_loss[loss=0.3275, simple_loss=0.3878, pruned_loss=0.1336, over 28631.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3847, pruned_loss=0.1317, over 5661162.03 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3709, pruned_loss=0.1226, over 5682208.52 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3858, pruned_loss=0.1323, over 5663147.39 frames. ], batch size: 242, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:02:22,785 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=528594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:02:24,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.757e+03 2.278e+03 3.134e+03 6.699e+03, threshold=4.555e+03, percent-clipped=12.0
+2023-03-06 09:02:25,438 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 09:02:42,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=528611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:03:04,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9614, 1.1008, 3.4143, 3.0137], device='cuda:0'), covar=tensor([0.1685, 0.2562, 0.0503, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0675, 0.0602, 0.0875, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 09:03:13,938 INFO [train.py:968] (0/2) Epoch 12, batch 27450, giga_loss[loss=0.3044, simple_loss=0.3729, pruned_loss=0.118, over 28842.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3814, pruned_loss=0.1297, over 5660818.89 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3707, pruned_loss=0.1225, over 5684625.64 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3825, pruned_loss=0.1303, over 5659843.04 frames. ], batch size: 174, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:03:46,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3755, 1.6081, 1.6246, 1.2701], device='cuda:0'), covar=tensor([0.1157, 0.1688, 0.0942, 0.1142], device='cuda:0'), in_proj_covar=tensor([0.0832, 0.0699, 0.0873, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 09:03:49,474 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-06 09:04:00,026 INFO [train.py:968] (0/2) Epoch 12, batch 27500, giga_loss[loss=0.3574, simple_loss=0.4035, pruned_loss=0.1556, over 28805.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3802, pruned_loss=0.1298, over 5662600.17 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1226, over 5687527.67 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3809, pruned_loss=0.1303, over 5658850.55 frames. ], batch size: 285, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:04:04,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.435e+02 1.609e+03 2.147e+03 2.908e+03 6.610e+03, threshold=4.294e+03, percent-clipped=6.0
+2023-03-06 09:04:46,621 INFO [train.py:968] (0/2) Epoch 12, batch 27550, giga_loss[loss=0.2751, simple_loss=0.3504, pruned_loss=0.09993, over 28915.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3792, pruned_loss=0.1294, over 5660734.10 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3716, pruned_loss=0.1232, over 5688690.42 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3794, pruned_loss=0.1294, over 5656361.45 frames. ], batch size: 136, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:04:57,814 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=528754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:04:59,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=528757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:05:18,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=528775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:05:26,352 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4656, 1.8553, 1.7375, 1.2599], device='cuda:0'), covar=tensor([0.1585, 0.2444, 0.1404, 0.1685], device='cuda:0'), in_proj_covar=tensor([0.0834, 0.0700, 0.0874, 0.0782], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 09:05:27,823 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=528786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:05:33,892 INFO [train.py:968] (0/2) Epoch 12, batch 27600, giga_loss[loss=0.3152, simple_loss=0.3798, pruned_loss=0.1253, over 27966.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3774, pruned_loss=0.1268, over 5664344.47 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3719, pruned_loss=0.1234, over 5692152.66 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3773, pruned_loss=0.1267, over 5657278.01 frames. ], batch size: 412, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 09:05:36,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.765e+02 1.448e+03 1.966e+03 3.110e+03 9.834e+03, threshold=3.932e+03, percent-clipped=12.0
+2023-03-06 09:06:18,355 INFO [train.py:968] (0/2) Epoch 12, batch 27650, giga_loss[loss=0.2736, simple_loss=0.3517, pruned_loss=0.09777, over 28983.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3746, pruned_loss=0.1241, over 5671944.87 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3722, pruned_loss=0.1234, over 5699912.25 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3744, pruned_loss=0.124, over 5658376.13 frames. ], batch size: 136, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:06:28,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4372, 1.9176, 1.5662, 1.6753], device='cuda:0'), covar=tensor([0.0643, 0.0242, 0.0263, 0.0632], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 09:07:10,521 INFO [train.py:968] (0/2) Epoch 12, batch 27700, giga_loss[loss=0.3105, simple_loss=0.3535, pruned_loss=0.1337, over 23507.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.374, pruned_loss=0.1239, over 5659846.08 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1234, over 5695969.23 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3739, pruned_loss=0.1239, over 5651674.12 frames. ], batch size: 705, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:07:15,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.281e+02 1.454e+03 1.864e+03 2.760e+03 9.398e+03, threshold=3.727e+03, percent-clipped=6.0
+2023-03-06 09:07:17,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=528899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:07:36,410 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=528918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:07:38,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=528921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:07:45,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=528928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:07:59,842 INFO [train.py:968] (0/2) Epoch 12, batch 27750, giga_loss[loss=0.2819, simple_loss=0.3492, pruned_loss=0.1073, over 28810.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3718, pruned_loss=0.1234, over 5662670.28 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3725, pruned_loss=0.1234, over 5702020.56 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3715, pruned_loss=0.1234, over 5649647.48 frames. ], batch size: 284, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:08:09,769 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-06 09:08:13,195 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=528950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:08:49,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3034, 1.6730, 1.5417, 1.5042], device='cuda:0'), covar=tensor([0.0697, 0.0356, 0.0285, 0.0749], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 09:08:53,978 INFO [train.py:968] (0/2) Epoch 12, batch 27800, giga_loss[loss=0.3156, simple_loss=0.3804, pruned_loss=0.1254, over 28940.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3696, pruned_loss=0.1226, over 5661689.01 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3728, pruned_loss=0.1236, over 5706406.70 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.369, pruned_loss=0.1224, over 5646338.73 frames. ], batch size: 213, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 09:08:58,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.131e+03 1.718e+03 2.228e+03 3.363e+03 1.480e+04, threshold=4.455e+03, percent-clipped=19.0
+2023-03-06 09:09:30,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6421, 2.1049, 1.6781, 1.4351], device='cuda:0'), covar=tensor([0.2180, 0.1675, 0.1965, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.1721, 0.1630, 0.1579, 0.1693], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 09:09:40,060 INFO [train.py:968] (0/2) Epoch 12, batch 27850, libri_loss[loss=0.3319, simple_loss=0.3964, pruned_loss=0.1337, over 29296.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3709, pruned_loss=0.1223, over 5669848.99 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3728, pruned_loss=0.1235, over 5709804.77 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3703, pruned_loss=0.1222, over 5653769.35 frames. ], batch size: 94, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 09:09:40,316 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=529042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:09:42,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=529045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:10:10,583 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=529074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:10:18,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4713, 3.4985, 1.5096, 1.5928], device='cuda:0'), covar=tensor([0.0912, 0.0240, 0.0859, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0515, 0.0344, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 09:10:26,574 INFO [train.py:968] (0/2) Epoch 12, batch 27900, giga_loss[loss=0.3122, simple_loss=0.3782, pruned_loss=0.1231, over 29026.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3723, pruned_loss=0.1229, over 5656572.59 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3732, pruned_loss=0.1237, over 5705934.93 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3715, pruned_loss=0.1226, over 5645790.55 frames. ], batch size: 155, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 09:10:31,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.277e+02 1.458e+03 1.932e+03 2.681e+03 6.546e+03, threshold=3.865e+03, percent-clipped=6.0
+2023-03-06 09:11:11,623 INFO [train.py:968] (0/2) Epoch 12, batch 27950, giga_loss[loss=0.3616, simple_loss=0.3958, pruned_loss=0.1637, over 23794.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3734, pruned_loss=0.1237, over 5656815.74 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3742, pruned_loss=0.1244, over 5711099.87 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3718, pruned_loss=0.1228, over 5641918.52 frames. ], batch size: 705, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 09:11:55,043 INFO [train.py:968] (0/2) Epoch 12, batch 28000, giga_loss[loss=0.3056, simple_loss=0.3655, pruned_loss=0.1228, over 28554.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3735, pruned_loss=0.1244, over 5662210.40 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3742, pruned_loss=0.1245, over 5720656.76 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3721, pruned_loss=0.1236, over 5638164.04 frames. ], batch size: 78, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:11:59,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.093e+02 1.419e+03 1.786e+03 2.349e+03 5.595e+03, threshold=3.572e+03, percent-clipped=5.0
+2023-03-06 09:12:07,417 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=529205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:12:36,990 INFO [train.py:968] (0/2) Epoch 12, batch 28050, giga_loss[loss=0.349, simple_loss=0.4076, pruned_loss=0.1452, over 28744.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3749, pruned_loss=0.1255, over 5662472.51 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3748, pruned_loss=0.125, over 5722395.81 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3731, pruned_loss=0.1244, over 5640331.78 frames. ], batch size: 284, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:13:21,749 INFO [train.py:968] (0/2) Epoch 12, batch 28100, giga_loss[loss=0.3224, simple_loss=0.3795, pruned_loss=0.1327, over 27987.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.378, pruned_loss=0.1279, over 5662811.07 frames. ], libri_tot_loss[loss=0.3133, simple_loss=0.3754, pruned_loss=0.1256, over 5720878.54 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3762, pruned_loss=0.1264, over 5644520.41 frames. ], batch size: 412, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:13:28,309 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.171e+03 1.604e+03 2.083e+03 2.695e+03 1.242e+04, threshold=4.166e+03, percent-clipped=11.0
+2023-03-06 09:13:31,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=529303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:13:35,023 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2982, 3.1235, 2.9741, 1.4629], device='cuda:0'), covar=tensor([0.0897, 0.0965, 0.0921, 0.2149], device='cuda:0'), in_proj_covar=tensor([0.1085, 0.1014, 0.0884, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 09:13:56,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6596, 1.6796, 1.2980, 1.2590], device='cuda:0'), covar=tensor([0.0731, 0.0528, 0.0915, 0.1015], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0445, 0.0502, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 09:14:08,008 INFO [train.py:968] (0/2) Epoch 12, batch 28150, giga_loss[loss=0.2994, simple_loss=0.3641, pruned_loss=0.1173, over 28664.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3783, pruned_loss=0.1283, over 5664996.82 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3747, pruned_loss=0.1252, over 5726065.55 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3775, pruned_loss=0.1275, over 5643998.98 frames. ], batch size: 85, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:14:58,062 INFO [train.py:968] (0/2) Epoch 12, batch 28200, libri_loss[loss=0.3407, simple_loss=0.4029, pruned_loss=0.1393, over 29183.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3796, pruned_loss=0.1293, over 5657500.93 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3751, pruned_loss=0.1255, over 5719024.75 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3787, pruned_loss=0.1285, over 5644189.20 frames. ], batch size: 101, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:15:01,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.827e+02 1.722e+03 2.106e+03 2.709e+03 9.076e+03, threshold=4.213e+03, percent-clipped=5.0
+2023-03-06 09:15:39,256 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=529436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:15:44,838 INFO [train.py:968] (0/2) Epoch 12, batch 28250, giga_loss[loss=0.3573, simple_loss=0.4044, pruned_loss=0.1551, over 26607.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3808, pruned_loss=0.1308, over 5650274.98 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3746, pruned_loss=0.1254, over 5710833.45 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3806, pruned_loss=0.1305, over 5644200.48 frames. ], batch size: 555, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:15:50,774 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=529446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:15:53,144 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=529449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:15:55,676 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=529452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:16:24,436 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=529478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:16:32,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9762, 2.1218, 2.2862, 1.7923], device='cuda:0'), covar=tensor([0.1641, 0.2096, 0.1234, 0.1492], device='cuda:0'), in_proj_covar=tensor([0.0835, 0.0699, 0.0876, 0.0782], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 09:16:38,626 INFO [train.py:968] (0/2) Epoch 12, batch 28300, giga_loss[loss=0.2928, simple_loss=0.369, pruned_loss=0.1083, over 28553.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3795, pruned_loss=0.1282, over 5656166.98 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3745, pruned_loss=0.1254, over 5714820.01 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3796, pruned_loss=0.128, over 5646729.04 frames. ], batch size: 336, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:16:42,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1286, 1.1569, 3.2943, 2.8649], device='cuda:0'), covar=tensor([0.1516, 0.2563, 0.0506, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0674, 0.0597, 0.0871, 0.0778], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 09:16:45,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.726e+03 2.347e+03 3.357e+03 6.579e+03, threshold=4.694e+03, percent-clipped=14.0
+2023-03-06 09:17:00,424 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1333, 1.2098, 3.6698, 3.1072], device='cuda:0'), covar=tensor([0.1995, 0.2791, 0.0784, 0.1244], device='cuda:0'), in_proj_covar=tensor([0.0672, 0.0596, 0.0869, 0.0776], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 09:17:28,419 INFO [train.py:968] (0/2) Epoch 12, batch 28350, giga_loss[loss=0.3241, simple_loss=0.3831, pruned_loss=0.1326, over 28050.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3804, pruned_loss=0.1297, over 5638480.34 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3742, pruned_loss=0.1253, over 5706916.98 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3808, pruned_loss=0.1297, over 5637204.46 frames. ], batch size: 412, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:18:08,144 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=529580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:18:19,281 INFO [train.py:968] (0/2) Epoch 12, batch 28400, libri_loss[loss=0.396, simple_loss=0.4314, pruned_loss=0.1803, over 29394.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3812, pruned_loss=0.131, over 5624224.57 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3747, pruned_loss=0.1257, over 5701469.28 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3813, pruned_loss=0.1307, over 5626471.95 frames. ], batch size: 92, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 09:18:25,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.603e+03 1.973e+03 2.752e+03 9.719e+03, threshold=3.947e+03, percent-clipped=5.0
+2023-03-06 09:19:17,394 INFO [train.py:968] (0/2) Epoch 12, batch 28450, giga_loss[loss=0.383, simple_loss=0.421, pruned_loss=0.1725, over 27709.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3797, pruned_loss=0.1307, over 5626476.01 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3747, pruned_loss=0.1256, over 5706631.91 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3798, pruned_loss=0.1307, over 5621645.71 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:20:09,858 INFO [train.py:968] (0/2) Epoch 12, batch 28500, giga_loss[loss=0.3151, simple_loss=0.3798, pruned_loss=0.1252, over 28685.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3784, pruned_loss=0.13, over 5636551.96 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3742, pruned_loss=0.1251, over 5711911.24 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3791, pruned_loss=0.1306, over 5626087.67 frames. ], batch size: 60, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:20:14,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.580e+03 1.988e+03 2.745e+03 6.532e+03, threshold=3.976e+03, percent-clipped=7.0
+2023-03-06 09:20:37,813 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=529723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:20:41,481 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=529726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:20:46,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4348, 1.3109, 1.2811, 1.5987], device='cuda:0'), covar=tensor([0.0718, 0.0326, 0.0302, 0.0763], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:0')
+2023-03-06 09:20:54,483 INFO [train.py:968] (0/2) Epoch 12, batch 28550, giga_loss[loss=0.4008, simple_loss=0.4259, pruned_loss=0.1879, over 27578.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.378, pruned_loss=0.1298, over 5652535.82 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.374, pruned_loss=0.1248, over 5716961.17 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3789, pruned_loss=0.1306, over 5638061.82 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:21:06,649 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=529755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:21:20,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2387, 1.2459, 4.0555, 3.3252], device='cuda:0'), covar=tensor([0.1697, 0.2645, 0.0399, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0675, 0.0595, 0.0870, 0.0776], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 09:21:21,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7600, 1.8905, 1.6976, 1.7651], device='cuda:0'), covar=tensor([0.1426, 0.1916, 0.1994, 0.1770], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0732, 0.0670, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 09:21:37,750 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=529788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:21:42,605 INFO [train.py:968] (0/2) Epoch 12, batch 28600, giga_loss[loss=0.306, simple_loss=0.3621, pruned_loss=0.125, over 28576.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3772, pruned_loss=0.1297, over 5640086.64 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3742, pruned_loss=0.1251, over 5706487.39 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3778, pruned_loss=0.1303, over 5637134.27 frames. ], batch size: 85, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:21:49,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.585e+03 2.073e+03 2.492e+03 7.530e+03, threshold=4.146e+03, percent-clipped=3.0
+2023-03-06 09:22:00,121 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=529811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:22:16,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=529827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:22:28,642 INFO [train.py:968] (0/2) Epoch 12, batch 28650, giga_loss[loss=0.2568, simple_loss=0.3288, pruned_loss=0.09237, over 28832.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3778, pruned_loss=0.1298, over 5655195.74 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3743, pruned_loss=0.1253, over 5708232.25 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3783, pruned_loss=0.1302, over 5649681.28 frames. ], batch size: 119, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:23:05,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5049, 1.7210, 1.7523, 1.3292], device='cuda:0'), covar=tensor([0.1557, 0.2214, 0.1271, 0.1551], device='cuda:0'), in_proj_covar=tensor([0.0833, 0.0698, 0.0873, 0.0782], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 09:23:15,884 INFO [train.py:968] (0/2) Epoch 12, batch 28700, giga_loss[loss=0.3035, simple_loss=0.3762, pruned_loss=0.1154, over 29051.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3788, pruned_loss=0.1306, over 5650701.25 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3741, pruned_loss=0.1252, over 5699654.41 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3794, pruned_loss=0.131, over 5653024.43 frames. ], batch size: 155, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:23:22,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.643e+03 2.379e+03 2.991e+03 8.528e+03, threshold=4.759e+03, percent-clipped=10.0
+2023-03-06 09:24:05,382 INFO [train.py:968] (0/2) Epoch 12, batch 28750, giga_loss[loss=0.3138, simple_loss=0.3788, pruned_loss=0.1244, over 29030.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3798, pruned_loss=0.1315, over 5659979.96 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3742, pruned_loss=0.1254, over 5704967.29 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3804, pruned_loss=0.1319, over 5655963.57 frames. ], batch size: 128, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:24:16,545 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=529954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:24:18,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=529957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:24:31,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=529970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:24:33,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=529973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:24:45,606 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=529986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:24:50,512 INFO [train.py:968] (0/2) Epoch 12, batch 28800, giga_loss[loss=0.3209, simple_loss=0.3775, pruned_loss=0.1322, over 28244.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.38, pruned_loss=0.1321, over 5661321.65 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3743, pruned_loss=0.1256, over 5696753.60 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3805, pruned_loss=0.1322, over 5664586.70 frames. ], batch size: 368, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:24:58,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.524e+02 1.530e+03 2.175e+03 2.908e+03 7.333e+03, threshold=4.351e+03, percent-clipped=6.0
+2023-03-06 09:24:58,792 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-530000.pt
+2023-03-06 09:25:01,017 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:25:13,445 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:25:33,614 INFO [train.py:968] (0/2) Epoch 12, batch 28850, giga_loss[loss=0.3138, simple_loss=0.3825, pruned_loss=0.1226, over 29026.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3785, pruned_loss=0.1303, over 5667860.49 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3743, pruned_loss=0.1255, over 5702157.48 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3789, pruned_loss=0.1307, over 5664498.19 frames. ], batch size: 155, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:25:47,654 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:25:55,028 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-06 09:26:22,818 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:26:23,230 INFO [train.py:968] (0/2) Epoch 12, batch 28900, giga_loss[loss=0.3138, simple_loss=0.3766, pruned_loss=0.1255, over 28571.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.38, pruned_loss=0.1313, over 5661440.90 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.374, pruned_loss=0.1253, over 5697045.54 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3806, pruned_loss=0.1319, over 5662443.70 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:26:33,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.813e+02 1.531e+03 2.023e+03 3.296e+03 7.946e+03, threshold=4.046e+03, percent-clipped=10.0
+2023-03-06 09:27:10,127 INFO [train.py:968] (0/2) Epoch 12, batch 28950, giga_loss[loss=0.3001, simple_loss=0.3716, pruned_loss=0.1143, over 28901.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3808, pruned_loss=0.1315, over 5652716.07 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3745, pruned_loss=0.1256, over 5681281.78 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3809, pruned_loss=0.1317, over 5666772.71 frames. ], batch size: 174, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:27:28,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:27:54,631 INFO [train.py:968] (0/2) Epoch 12, batch 29000, giga_loss[loss=0.3043, simple_loss=0.3563, pruned_loss=0.1262, over 28723.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3809, pruned_loss=0.1317, over 5660887.23 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.374, pruned_loss=0.1251, over 5685878.48 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3817, pruned_loss=0.1325, over 5667395.20 frames. ], batch size: 92, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:28:04,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.518e+03 1.826e+03 2.655e+03 9.715e+03, threshold=3.652e+03, percent-clipped=11.0
+2023-03-06 09:28:38,437 INFO [train.py:968] (0/2) Epoch 12, batch 29050, giga_loss[loss=0.284, simple_loss=0.3547, pruned_loss=0.1066, over 28836.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3818, pruned_loss=0.1326, over 5650800.09 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3735, pruned_loss=0.1248, over 5679367.54 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3831, pruned_loss=0.1338, over 5660910.12 frames. ], batch size: 174, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:29:23,649 INFO [train.py:968] (0/2) Epoch 12, batch 29100, giga_loss[loss=0.3137, simple_loss=0.3778, pruned_loss=0.1248, over 28868.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3812, pruned_loss=0.1322, over 5651457.23 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3733, pruned_loss=0.1247, over 5682600.63 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3825, pruned_loss=0.1333, over 5656293.87 frames. ], batch size: 199, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:29:31,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.875e+02 1.546e+03 1.796e+03 2.481e+03 5.658e+03, threshold=3.593e+03, percent-clipped=9.0
+2023-03-06 09:29:37,805 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=530306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:29:39,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=530309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:30:08,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:30:11,709 INFO [train.py:968] (0/2) Epoch 12, batch 29150, giga_loss[loss=0.3074, simple_loss=0.3732, pruned_loss=0.1208, over 28534.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3812, pruned_loss=0.1314, over 5644835.79 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3729, pruned_loss=0.1243, over 5685450.84 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3829, pruned_loss=0.1329, over 5645686.16 frames. ], batch size: 336, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:30:29,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2364, 0.8405, 0.9271, 1.4321], device='cuda:0'), covar=tensor([0.0712, 0.0395, 0.0337, 0.0770], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 09:30:46,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5169, 2.8417, 1.5924, 1.7074], device='cuda:0'), covar=tensor([0.0722, 0.0309, 0.0704, 0.0959], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0518, 0.0346, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 09:31:02,515 INFO [train.py:968] (0/2) Epoch 12, batch 29200, giga_loss[loss=0.3694, simple_loss=0.3931, pruned_loss=0.1729, over 23494.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3815, pruned_loss=0.1312, over 5634441.72 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3728, pruned_loss=0.1243, over 5680671.47 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3832, pruned_loss=0.1326, over 5638709.85 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:31:02,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:31:09,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.912e+02 1.382e+03 1.859e+03 2.875e+03 7.183e+03, threshold=3.718e+03, percent-clipped=14.0
+2023-03-06 09:31:28,703 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:31:39,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:31:48,812 INFO [train.py:968] (0/2) Epoch 12, batch 29250, giga_loss[loss=0.3251, simple_loss=0.3809, pruned_loss=0.1347, over 28946.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3805, pruned_loss=0.1302, over 5644381.33 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3728, pruned_loss=0.1243, over 5680671.47 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3818, pruned_loss=0.1312, over 5647703.27 frames. ], batch size: 112, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:32:08,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:32:34,263 INFO [train.py:968] (0/2) Epoch 12, batch 29300, giga_loss[loss=0.4051, simple_loss=0.4162, pruned_loss=0.197, over 23620.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3796, pruned_loss=0.1299, over 5643512.09 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3728, pruned_loss=0.1244, over 5677001.79 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3808, pruned_loss=0.1308, over 5648547.37 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:32:42,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.482e+03 2.047e+03 2.901e+03 7.843e+03, threshold=4.094e+03, percent-clipped=11.0
+2023-03-06 09:33:03,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4296, 2.3610, 1.9461, 2.2143], device='cuda:0'), covar=tensor([0.0645, 0.0526, 0.0762, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0440, 0.0496, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 09:33:13,828 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=530535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:33:15,924 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=530538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:33:18,380 INFO [train.py:968] (0/2) Epoch 12, batch 29350, giga_loss[loss=0.3699, simple_loss=0.412, pruned_loss=0.1639, over 27532.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3797, pruned_loss=0.1294, over 5643352.43 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3734, pruned_loss=0.1246, over 5676581.23 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3802, pruned_loss=0.13, over 5646888.18 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:33:46,333 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:33:52,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:33:55,378 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=530575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:33:58,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=530578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:34:00,873 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 09:34:09,569 INFO [train.py:968] (0/2) Epoch 12, batch 29400, giga_loss[loss=0.3232, simple_loss=0.3811, pruned_loss=0.1327, over 28620.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3809, pruned_loss=0.1303, over 5644372.79 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3736, pruned_loss=0.1248, over 5668241.85 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3812, pruned_loss=0.1307, over 5654816.57 frames. ], batch size: 242, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:34:19,991 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.508e+02 1.468e+03 1.976e+03 2.716e+03 1.093e+04, threshold=3.952e+03, percent-clipped=9.0
+2023-03-06 09:34:25,282 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:34:26,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=530609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:34:28,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=530612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:34:51,195 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3593, 1.6441, 1.2838, 1.5001], device='cuda:0'), covar=tensor([0.2508, 0.2480, 0.2742, 0.2163], device='cuda:0'), in_proj_covar=tensor([0.1334, 0.0984, 0.1171, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 09:34:56,882 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:34:57,241 INFO [train.py:968] (0/2) Epoch 12, batch 29450, giga_loss[loss=0.3205, simple_loss=0.3716, pruned_loss=0.1348, over 29109.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3804, pruned_loss=0.1309, over 5648835.73 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3732, pruned_loss=0.1244, over 5675598.24 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3812, pruned_loss=0.1318, over 5649959.75 frames. ], batch size: 128, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:35:09,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4537, 1.7634, 1.3693, 1.7250], device='cuda:0'), covar=tensor([0.2405, 0.2370, 0.2675, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.1332, 0.0983, 0.1170, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 09:35:43,197 INFO [train.py:968] (0/2) Epoch 12, batch 29500, giga_loss[loss=0.2759, simple_loss=0.3397, pruned_loss=0.106, over 28785.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3798, pruned_loss=0.1304, over 5667708.32 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.373, pruned_loss=0.1241, over 5680462.81 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3808, pruned_loss=0.1314, over 5663864.46 frames. ], batch size: 99, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:35:51,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.099e+03 1.561e+03 1.897e+03 2.332e+03 6.341e+03, threshold=3.793e+03, percent-clipped=5.0
+2023-03-06 09:36:29,638 INFO [train.py:968] (0/2) Epoch 12, batch 29550, libri_loss[loss=0.3356, simple_loss=0.379, pruned_loss=0.1461, over 29571.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3825, pruned_loss=0.1328, over 5663120.94 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3732, pruned_loss=0.1244, over 5682223.37 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3833, pruned_loss=0.1335, over 5657720.31 frames. ], batch size: 74, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:37:19,559 INFO [train.py:968] (0/2) Epoch 12, batch 29600, giga_loss[loss=0.2744, simple_loss=0.3513, pruned_loss=0.09872, over 28376.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3815, pruned_loss=0.1321, over 5647118.81 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.373, pruned_loss=0.1242, over 5682538.49 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3824, pruned_loss=0.133, over 5642128.09 frames. ], batch size: 71, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:37:24,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:37:27,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.211e+02 1.600e+03 2.056e+03 3.056e+03 7.900e+03, threshold=4.113e+03, percent-clipped=18.0
+2023-03-06 09:37:44,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4677, 1.6792, 1.6834, 1.2571], device='cuda:0'), covar=tensor([0.1555, 0.2493, 0.1396, 0.1648], device='cuda:0'), in_proj_covar=tensor([0.0833, 0.0697, 0.0873, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 09:37:49,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1430, 1.0715, 3.9590, 3.0276], device='cuda:0'), covar=tensor([0.1719, 0.2764, 0.0462, 0.1017], device='cuda:0'), in_proj_covar=tensor([0.0676, 0.0595, 0.0867, 0.0776], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 09:38:03,436 INFO [train.py:968] (0/2) Epoch 12, batch 29650, giga_loss[loss=0.3191, simple_loss=0.3803, pruned_loss=0.129, over 28910.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3822, pruned_loss=0.1324, over 5641047.74 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3731, pruned_loss=0.1242, over 5669186.26 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3832, pruned_loss=0.1334, over 5647995.50 frames. ], batch size: 186, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:38:12,142 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-06 09:38:15,285 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:38:51,248 INFO [train.py:968] (0/2) Epoch 12, batch 29700, giga_loss[loss=0.3108, simple_loss=0.3725, pruned_loss=0.1245, over 28604.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3821, pruned_loss=0.1317, over 5639294.73 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3731, pruned_loss=0.1242, over 5660249.19 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.383, pruned_loss=0.1327, over 5651237.41 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:39:02,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.922e+03 2.612e+03 3.509e+03 1.124e+04, threshold=5.225e+03, percent-clipped=14.0
+2023-03-06 09:39:31,399 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-06 09:39:38,506 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=530940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:39:39,371 INFO [train.py:968] (0/2) Epoch 12, batch 29750, libri_loss[loss=0.4177, simple_loss=0.4532, pruned_loss=0.1911, over 29213.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3822, pruned_loss=0.1317, over 5634999.56 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3731, pruned_loss=0.1243, over 5654227.68 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.383, pruned_loss=0.1324, over 5649651.30 frames. ], batch size: 97, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:39:40,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=530943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:39:44,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:39:54,043 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 09:40:08,956 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:40:25,421 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:40:28,029 INFO [train.py:968] (0/2) Epoch 12, batch 29800, giga_loss[loss=0.4199, simple_loss=0.4452, pruned_loss=0.1973, over 27493.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3817, pruned_loss=0.1312, over 5644208.34 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3732, pruned_loss=0.1243, over 5656227.39 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3824, pruned_loss=0.1319, over 5653725.89 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:40:37,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.094e+03 2.047e+03 2.583e+03 3.355e+03 9.995e+03, threshold=5.165e+03, percent-clipped=7.0
+2023-03-06 09:41:15,848 INFO [train.py:968] (0/2) Epoch 12, batch 29850, giga_loss[loss=0.2871, simple_loss=0.3574, pruned_loss=0.1084, over 28634.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3796, pruned_loss=0.1297, over 5655086.85 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3735, pruned_loss=0.1244, over 5660094.00 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.38, pruned_loss=0.1302, over 5659053.55 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:41:24,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0774, 1.4265, 1.3401, 1.0504], device='cuda:0'), covar=tensor([0.2135, 0.1720, 0.1100, 0.1674], device='cuda:0'), in_proj_covar=tensor([0.1744, 0.1653, 0.1597, 0.1709], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 09:41:30,626 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-06 09:41:57,822 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=531090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:41:58,790 INFO [train.py:968] (0/2) Epoch 12, batch 29900, giga_loss[loss=0.3314, simple_loss=0.3809, pruned_loss=0.141, over 28859.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3756, pruned_loss=0.1274, over 5664896.76 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3736, pruned_loss=0.1245, over 5664307.47 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.376, pruned_loss=0.1279, over 5664279.33 frames. ], batch size: 243, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:41:59,818 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=531093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:42:09,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.299e+02 1.632e+03 2.174e+03 2.734e+03 6.351e+03, threshold=4.349e+03, percent-clipped=1.0
+2023-03-06 09:42:29,251 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=531122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:42:31,724 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=531126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:42:46,338 INFO [train.py:968] (0/2) Epoch 12, batch 29950, giga_loss[loss=0.278, simple_loss=0.3438, pruned_loss=0.1061, over 28907.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3724, pruned_loss=0.1264, over 5653979.96 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3742, pruned_loss=0.1249, over 5669970.50 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3722, pruned_loss=0.1265, over 5648137.16 frames. ], batch size: 199, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:43:08,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1196, 1.2289, 3.4646, 3.0028], device='cuda:0'), covar=tensor([0.1557, 0.2529, 0.0444, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0678, 0.0599, 0.0870, 0.0778], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 09:43:31,121 INFO [train.py:968] (0/2) Epoch 12, batch 30000, giga_loss[loss=0.2976, simple_loss=0.362, pruned_loss=0.1166, over 28877.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3725, pruned_loss=0.1273, over 5663166.25 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3743, pruned_loss=0.1248, over 5671667.51 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3722, pruned_loss=0.1275, over 5656902.15 frames. ], batch size: 136, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 09:43:31,126 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 09:43:34,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8520, 3.6215, 3.4495, 1.6994], device='cuda:0'), covar=tensor([0.0727, 0.0985, 0.0936, 0.2687], device='cuda:0'), in_proj_covar=tensor([0.1090, 0.1021, 0.0889, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 09:43:39,671 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.214, simple_loss=0.3216, pruned_loss=0.05325, over 944034.00 frames. 
+2023-03-06 09:43:39,672 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 09:43:44,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4058, 1.8298, 1.3217, 0.8411], device='cuda:0'), covar=tensor([0.3730, 0.2422, 0.2099, 0.4231], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1504, 0.1503, 0.1284], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 09:43:48,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.784e+03 2.329e+03 3.239e+03 7.150e+03, threshold=4.657e+03, percent-clipped=8.0
+2023-03-06 09:44:10,079 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 09:44:15,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=531228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:44:29,483 INFO [train.py:968] (0/2) Epoch 12, batch 30050, libri_loss[loss=0.2459, simple_loss=0.3141, pruned_loss=0.08882, over 28482.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3713, pruned_loss=0.1268, over 5643985.75 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3741, pruned_loss=0.1248, over 5669026.59 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3711, pruned_loss=0.127, over 5640122.91 frames. ], batch size: 63, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:44:42,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6938, 1.7640, 1.2243, 1.3693], device='cuda:0'), covar=tensor([0.0781, 0.0565, 0.1017, 0.1071], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0440, 0.0498, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 09:45:15,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4567, 2.9053, 1.5976, 1.7026], device='cuda:0'), covar=tensor([0.0765, 0.0326, 0.0766, 0.1065], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0519, 0.0346, 0.0396], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 09:45:16,633 INFO [train.py:968] (0/2) Epoch 12, batch 30100, giga_loss[loss=0.3124, simple_loss=0.3823, pruned_loss=0.1212, over 28864.00 frames. ], tot_loss[loss=0.309, simple_loss=0.37, pruned_loss=0.124, over 5652044.47 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3739, pruned_loss=0.1245, over 5673785.59 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.37, pruned_loss=0.1245, over 5644105.39 frames. ], batch size: 174, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:45:28,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.996e+02 1.521e+03 2.133e+03 3.082e+03 1.333e+04, threshold=4.267e+03, percent-clipped=7.0
+2023-03-06 09:45:48,481 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=531321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:46:14,536 INFO [train.py:968] (0/2) Epoch 12, batch 30150, giga_loss[loss=0.2795, simple_loss=0.3584, pruned_loss=0.1003, over 28828.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3666, pruned_loss=0.1194, over 5633607.37 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3739, pruned_loss=0.1246, over 5665060.67 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3666, pruned_loss=0.1198, over 5634459.28 frames. ], batch size: 199, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:46:36,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=531364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:46:42,038 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=531371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:46:45,488 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=531374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:47:00,086 INFO [train.py:968] (0/2) Epoch 12, batch 30200, giga_loss[loss=0.2619, simple_loss=0.34, pruned_loss=0.09192, over 28826.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3653, pruned_loss=0.1172, over 5639382.90 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3742, pruned_loss=0.125, over 5658951.53 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3647, pruned_loss=0.1169, over 5645049.43 frames. ], batch size: 186, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:47:11,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.268e+02 1.458e+03 1.797e+03 2.537e+03 6.801e+03, threshold=3.594e+03, percent-clipped=4.0
+2023-03-06 09:47:12,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=531403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:47:52,809 INFO [train.py:968] (0/2) Epoch 12, batch 30250, giga_loss[loss=0.2573, simple_loss=0.3391, pruned_loss=0.08773, over 28309.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3621, pruned_loss=0.1139, over 5643144.78 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.1249, over 5661517.53 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3618, pruned_loss=0.1135, over 5645069.95 frames. ], batch size: 368, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:48:09,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2474, 1.3984, 1.2127, 1.2242], device='cuda:0'), covar=tensor([0.1568, 0.1238, 0.1026, 0.1146], device='cuda:0'), in_proj_covar=tensor([0.1708, 0.1623, 0.1566, 0.1668], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 09:48:28,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4531, 1.2635, 3.9177, 3.2981], device='cuda:0'), covar=tensor([0.1498, 0.2682, 0.0457, 0.0726], device='cuda:0'), in_proj_covar=tensor([0.0681, 0.0602, 0.0874, 0.0782], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 09:48:41,703 INFO [train.py:968] (0/2) Epoch 12, batch 30300, giga_loss[loss=0.2824, simple_loss=0.3641, pruned_loss=0.1003, over 28850.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3591, pruned_loss=0.1105, over 5649619.44 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3742, pruned_loss=0.1253, over 5663797.67 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3584, pruned_loss=0.1097, over 5648896.37 frames. ], batch size: 186, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:48:51,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=531501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:48:51,836 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-06 09:48:53,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.315e+02 1.420e+03 1.887e+03 2.923e+03 7.173e+03, threshold=3.774e+03, percent-clipped=10.0
+2023-03-06 09:48:57,598 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=531507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:48:58,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7089, 2.2178, 1.5816, 0.8328], device='cuda:0'), covar=tensor([0.4196, 0.2323, 0.3132, 0.4590], device='cuda:0'), in_proj_covar=tensor([0.1568, 0.1488, 0.1491, 0.1279], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 09:48:59,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=531510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:49:29,088 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=531539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:49:32,770 INFO [train.py:968] (0/2) Epoch 12, batch 30350, giga_loss[loss=0.2895, simple_loss=0.3687, pruned_loss=0.1051, over 28961.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3581, pruned_loss=0.1075, over 5656944.33 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3739, pruned_loss=0.1253, over 5659762.55 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3575, pruned_loss=0.1066, over 5660414.89 frames. ], batch size: 227, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:49:48,048 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4417, 1.7160, 1.5171, 1.4078], device='cuda:0'), covar=tensor([0.1765, 0.1309, 0.1141, 0.1329], device='cuda:0'), in_proj_covar=tensor([0.1705, 0.1621, 0.1565, 0.1666], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 09:50:18,966 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 09:50:25,769 INFO [train.py:968] (0/2) Epoch 12, batch 30400, giga_loss[loss=0.2796, simple_loss=0.3587, pruned_loss=0.1002, over 28967.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3573, pruned_loss=0.107, over 5660071.65 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3733, pruned_loss=0.125, over 5662184.20 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3571, pruned_loss=0.1061, over 5660696.98 frames. ], batch size: 164, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 09:50:27,204 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7303, 2.2391, 1.8601, 1.8354], device='cuda:0'), covar=tensor([0.1191, 0.1115, 0.1551, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0731, 0.0671, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 09:50:34,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.425e+02 1.341e+03 1.737e+03 2.445e+03 4.511e+03, threshold=3.473e+03, percent-clipped=2.0
+2023-03-06 09:51:14,761 INFO [train.py:968] (0/2) Epoch 12, batch 30450, giga_loss[loss=0.3109, simple_loss=0.383, pruned_loss=0.1194, over 29035.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.356, pruned_loss=0.1058, over 5666179.16 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3729, pruned_loss=0.125, over 5665848.74 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3559, pruned_loss=0.1049, over 5663617.70 frames. ], batch size: 164, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:51:16,876 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=531644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:51:19,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=531647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:51:19,911 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 09:51:47,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=531676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:52:01,095 INFO [train.py:968] (0/2) Epoch 12, batch 30500, giga_loss[loss=0.2658, simple_loss=0.3438, pruned_loss=0.09387, over 28938.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3522, pruned_loss=0.1032, over 5662363.86 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3722, pruned_loss=0.1246, over 5662036.28 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3522, pruned_loss=0.1023, over 5663044.93 frames. ], batch size: 186, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:52:06,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=531696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:52:09,312 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=531699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:52:14,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.413e+02 1.289e+03 1.805e+03 2.967e+03 9.558e+03, threshold=3.611e+03, percent-clipped=21.0
+2023-03-06 09:52:30,552 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=531716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:52:53,748 INFO [train.py:968] (0/2) Epoch 12, batch 30550, giga_loss[loss=0.2776, simple_loss=0.3545, pruned_loss=0.1004, over 28599.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3517, pruned_loss=0.103, over 5658313.29 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3719, pruned_loss=0.1245, over 5665942.33 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3517, pruned_loss=0.1021, over 5655280.38 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:53:43,222 INFO [train.py:968] (0/2) Epoch 12, batch 30600, giga_loss[loss=0.2583, simple_loss=0.3396, pruned_loss=0.08854, over 28766.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3515, pruned_loss=0.103, over 5661620.78 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3712, pruned_loss=0.1242, over 5668301.39 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3518, pruned_loss=0.1021, over 5656899.78 frames. ], batch size: 284, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:53:55,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.172e+02 1.197e+03 1.535e+03 2.090e+03 5.306e+03, threshold=3.071e+03, percent-clipped=4.0
+2023-03-06 09:53:58,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3558, 1.6903, 1.3106, 1.3854], device='cuda:0'), covar=tensor([0.2670, 0.2470, 0.2848, 0.2129], device='cuda:0'), in_proj_covar=tensor([0.1328, 0.0976, 0.1171, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 09:54:31,923 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=531839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:54:33,885 INFO [train.py:968] (0/2) Epoch 12, batch 30650, giga_loss[loss=0.2485, simple_loss=0.3346, pruned_loss=0.08124, over 28993.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3481, pruned_loss=0.09998, over 5661888.78 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.371, pruned_loss=0.1241, over 5669514.89 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3484, pruned_loss=0.09931, over 5656988.76 frames. ], batch size: 155, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:54:34,163 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=531842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:55:04,212 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=531871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:55:20,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2820, 1.7806, 1.3069, 0.4207], device='cuda:0'), covar=tensor([0.2674, 0.1881, 0.3004, 0.4119], device='cuda:0'), in_proj_covar=tensor([0.1555, 0.1471, 0.1480, 0.1268], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 09:55:25,202 INFO [train.py:968] (0/2) Epoch 12, batch 30700, giga_loss[loss=0.228, simple_loss=0.318, pruned_loss=0.06896, over 28620.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3447, pruned_loss=0.09749, over 5667742.35 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3705, pruned_loss=0.1237, over 5672394.67 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.345, pruned_loss=0.09678, over 5661182.15 frames. ], batch size: 262, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:55:34,870 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=531900.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 09:55:38,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.627e+02 1.339e+03 1.934e+03 2.833e+03 8.803e+03, threshold=3.868e+03, percent-clipped=19.0
+2023-03-06 09:56:05,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4329, 1.7401, 1.3541, 1.4957], device='cuda:0'), covar=tensor([0.2552, 0.2298, 0.2657, 0.2154], device='cuda:0'), in_proj_covar=tensor([0.1329, 0.0974, 0.1172, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 09:56:16,484 INFO [train.py:968] (0/2) Epoch 12, batch 30750, giga_loss[loss=0.2668, simple_loss=0.3462, pruned_loss=0.09369, over 29037.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3414, pruned_loss=0.09626, over 5672208.51 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3697, pruned_loss=0.1233, over 5677320.06 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3419, pruned_loss=0.0956, over 5662621.40 frames. ], batch size: 155, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:57:05,347 INFO [train.py:968] (0/2) Epoch 12, batch 30800, giga_loss[loss=0.2756, simple_loss=0.3345, pruned_loss=0.1083, over 23996.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3399, pruned_loss=0.09578, over 5663180.67 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3692, pruned_loss=0.1231, over 5672476.81 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3401, pruned_loss=0.09501, over 5659466.66 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 09:57:11,814 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-532000.pt
+2023-03-06 09:57:17,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.918e+02 1.360e+03 1.811e+03 3.036e+03 1.083e+04, threshold=3.622e+03, percent-clipped=10.0
+2023-03-06 09:57:17,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=532004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:57:57,350 INFO [train.py:968] (0/2) Epoch 12, batch 30850, giga_loss[loss=0.2793, simple_loss=0.3384, pruned_loss=0.1101, over 26677.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3395, pruned_loss=0.09583, over 5647121.10 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3687, pruned_loss=0.1229, over 5672944.64 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3395, pruned_loss=0.09479, over 5643802.29 frames. ], batch size: 555, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 09:58:34,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:58:44,014 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1970, 0.8570, 0.8560, 1.3805], device='cuda:0'), covar=tensor([0.0727, 0.0418, 0.0366, 0.0794], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 09:58:55,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:58:56,162 INFO [train.py:968] (0/2) Epoch 12, batch 30900, giga_loss[loss=0.277, simple_loss=0.3509, pruned_loss=0.1016, over 27663.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3422, pruned_loss=0.09654, over 5648404.38 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3686, pruned_loss=0.1228, over 5676214.89 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.342, pruned_loss=0.09534, over 5642367.60 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:59:09,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.544e+02 1.430e+03 1.787e+03 2.317e+03 5.997e+03, threshold=3.573e+03, percent-clipped=7.0
+2023-03-06 09:59:36,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6050, 1.9690, 1.8896, 1.4381], device='cuda:0'), covar=tensor([0.1251, 0.1829, 0.1062, 0.1424], device='cuda:0'), in_proj_covar=tensor([0.0835, 0.0688, 0.0873, 0.0782], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 09:59:52,043 INFO [train.py:968] (0/2) Epoch 12, batch 30950, giga_loss[loss=0.2761, simple_loss=0.3517, pruned_loss=0.1002, over 28620.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3421, pruned_loss=0.09569, over 5632866.17 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3675, pruned_loss=0.1223, over 5669831.17 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3423, pruned_loss=0.09461, over 5633578.11 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:00:04,058 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=532152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:00:52,732 INFO [train.py:968] (0/2) Epoch 12, batch 31000, giga_loss[loss=0.2628, simple_loss=0.3414, pruned_loss=0.09208, over 29141.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3428, pruned_loss=0.09664, over 5636636.01 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.367, pruned_loss=0.122, over 5677129.12 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3426, pruned_loss=0.0953, over 5629247.90 frames. ], batch size: 128, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:01:11,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.813e+02 1.387e+03 1.904e+03 2.796e+03 8.495e+03, threshold=3.808e+03, percent-clipped=18.0
+2023-03-06 10:01:25,277 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:01:29,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:01:30,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1717, 1.5305, 1.3775, 1.2608], device='cuda:0'), covar=tensor([0.1789, 0.1804, 0.1960, 0.1883], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0725, 0.0668, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 10:01:41,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5194, 4.4670, 1.7913, 1.6410], device='cuda:0'), covar=tensor([0.0923, 0.0272, 0.0871, 0.1277], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0514, 0.0347, 0.0396], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 10:01:47,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:01:49,408 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:01:53,955 INFO [train.py:968] (0/2) Epoch 12, batch 31050, libri_loss[loss=0.2887, simple_loss=0.3501, pruned_loss=0.1137, over 29542.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3403, pruned_loss=0.09521, over 5650255.10 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3664, pruned_loss=0.1218, over 5683238.87 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3401, pruned_loss=0.09358, over 5637607.48 frames. ], batch size: 82, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:02:01,724 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:02:20,739 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:02:33,764 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532275.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:02:56,832 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-06 10:02:57,073 INFO [train.py:968] (0/2) Epoch 12, batch 31100, libri_loss[loss=0.2759, simple_loss=0.3465, pruned_loss=0.1027, over 29780.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.339, pruned_loss=0.09307, over 5644479.77 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3659, pruned_loss=0.1216, over 5687892.21 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3385, pruned_loss=0.09132, over 5629369.03 frames. ], batch size: 87, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:03:15,027 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.988e+02 1.069e+03 1.418e+03 2.351e+03 4.966e+03, threshold=2.836e+03, percent-clipped=4.0
+2023-03-06 10:03:41,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5122, 1.5696, 1.3634, 1.6038], device='cuda:0'), covar=tensor([0.0677, 0.0281, 0.0302, 0.0771], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 10:03:58,474 INFO [train.py:968] (0/2) Epoch 12, batch 31150, giga_loss[loss=0.2534, simple_loss=0.3255, pruned_loss=0.09066, over 27646.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3371, pruned_loss=0.09156, over 5640637.30 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3658, pruned_loss=0.1217, over 5679180.56 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3364, pruned_loss=0.08954, over 5635618.13 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 10:04:01,475 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-06 10:04:40,573 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:05:00,946 INFO [train.py:968] (0/2) Epoch 12, batch 31200, giga_loss[loss=0.2805, simple_loss=0.359, pruned_loss=0.1009, over 27705.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3352, pruned_loss=0.09138, over 5648692.27 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3658, pruned_loss=0.1219, over 5672793.25 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3341, pruned_loss=0.08901, over 5648912.19 frames. ], batch size: 474, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:05:18,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.018e+02 1.291e+03 1.646e+03 2.385e+03 1.554e+04, threshold=3.293e+03, percent-clipped=19.0
+2023-03-06 10:05:32,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532418.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:05:34,388 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532421.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:05:51,019 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=532436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:05:52,957 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-06 10:05:57,029 INFO [train.py:968] (0/2) Epoch 12, batch 31250, libri_loss[loss=0.2646, simple_loss=0.3183, pruned_loss=0.1054, over 29656.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3358, pruned_loss=0.0924, over 5652255.87 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3653, pruned_loss=0.1218, over 5664562.38 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3347, pruned_loss=0.08993, over 5659375.94 frames. ], batch size: 69, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:06:07,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532450.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:06:25,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1779, 1.5660, 1.1802, 0.3734], device='cuda:0'), covar=tensor([0.2742, 0.1627, 0.2512, 0.4264], device='cuda:0'), in_proj_covar=tensor([0.1579, 0.1488, 0.1498, 0.1288], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 10:06:54,582 INFO [train.py:968] (0/2) Epoch 12, batch 31300, giga_loss[loss=0.2587, simple_loss=0.3468, pruned_loss=0.08525, over 28940.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.338, pruned_loss=0.09346, over 5655192.29 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3653, pruned_loss=0.1219, over 5668067.57 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3367, pruned_loss=0.09099, over 5657488.30 frames. ], batch size: 164, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:07:12,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.030e+02 1.391e+03 1.778e+03 2.579e+03 5.595e+03, threshold=3.555e+03, percent-clipped=15.0
+2023-03-06 10:07:32,941 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:07:37,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:07:40,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:07:47,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-06 10:07:53,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2386, 1.2307, 1.1201, 1.4660], device='cuda:0'), covar=tensor([0.0772, 0.0358, 0.0361, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0116, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 10:07:58,750 INFO [train.py:968] (0/2) Epoch 12, batch 31350, giga_loss[loss=0.2921, simple_loss=0.3685, pruned_loss=0.1078, over 29045.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3399, pruned_loss=0.09347, over 5650527.76 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3653, pruned_loss=0.122, over 5669403.24 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3388, pruned_loss=0.09138, over 5651131.76 frames. ], batch size: 106, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:08:18,708 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:08:34,756 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-06 10:09:04,872 INFO [train.py:968] (0/2) Epoch 12, batch 31400, giga_loss[loss=0.2331, simple_loss=0.3075, pruned_loss=0.07931, over 27771.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3353, pruned_loss=0.09022, over 5663258.39 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.365, pruned_loss=0.1219, over 5671656.19 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3344, pruned_loss=0.08842, over 5661628.02 frames. ], batch size: 474, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:09:23,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.885e+02 1.343e+03 1.711e+03 2.318e+03 6.365e+03, threshold=3.421e+03, percent-clipped=10.0
+2023-03-06 10:10:08,118 INFO [train.py:968] (0/2) Epoch 12, batch 31450, giga_loss[loss=0.2686, simple_loss=0.3462, pruned_loss=0.09551, over 28663.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3376, pruned_loss=0.09247, over 5663880.99 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.365, pruned_loss=0.122, over 5666777.21 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3359, pruned_loss=0.08976, over 5666387.50 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:10:44,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5860, 1.8003, 1.5873, 1.6184], device='cuda:0'), covar=tensor([0.1504, 0.2169, 0.1884, 0.1862], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0719, 0.0662, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 10:10:44,430 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:10:50,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:11:11,510 INFO [train.py:968] (0/2) Epoch 12, batch 31500, giga_loss[loss=0.2866, simple_loss=0.3648, pruned_loss=0.1042, over 29016.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3399, pruned_loss=0.09273, over 5666776.10 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3647, pruned_loss=0.1218, over 5671636.19 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3383, pruned_loss=0.09026, over 5664387.97 frames. ], batch size: 136, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:11:23,542 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=532702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:11:23,558 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:11:29,404 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.836e+02 1.282e+03 1.911e+03 2.794e+03 8.130e+03, threshold=3.822e+03, percent-clipped=12.0
+2023-03-06 10:11:32,941 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-06 10:12:14,798 INFO [train.py:968] (0/2) Epoch 12, batch 31550, libri_loss[loss=0.3763, simple_loss=0.4144, pruned_loss=0.1691, over 25756.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3428, pruned_loss=0.09245, over 5651918.94 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3648, pruned_loss=0.1222, over 5664525.94 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3408, pruned_loss=0.08934, over 5656779.16 frames. ], batch size: 136, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:13:18,211 INFO [train.py:968] (0/2) Epoch 12, batch 31600, giga_loss[loss=0.2362, simple_loss=0.3308, pruned_loss=0.07079, over 28929.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3427, pruned_loss=0.09065, over 5653758.48 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3646, pruned_loss=0.1221, over 5667066.00 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3411, pruned_loss=0.08802, over 5655435.46 frames. ], batch size: 145, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 10:13:33,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.277e+02 1.539e+03 2.048e+03 2.878e+03 9.147e+03, threshold=4.096e+03, percent-clipped=6.0
+2023-03-06 10:13:38,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:14:09,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1168, 3.9263, 3.6994, 1.8560], device='cuda:0'), covar=tensor([0.0589, 0.0752, 0.0765, 0.2300], device='cuda:0'), in_proj_covar=tensor([0.1038, 0.0970, 0.0844, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 10:14:17,372 INFO [train.py:968] (0/2) Epoch 12, batch 31650, giga_loss[loss=0.2917, simple_loss=0.3712, pruned_loss=0.1061, over 28856.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3427, pruned_loss=0.08968, over 5667813.56 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3641, pruned_loss=0.1217, over 5672980.92 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3414, pruned_loss=0.08718, over 5663633.39 frames. ], batch size: 174, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:14:35,007 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=532856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:15:17,406 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 10:15:19,614 INFO [train.py:968] (0/2) Epoch 12, batch 31700, giga_loss[loss=0.2727, simple_loss=0.3434, pruned_loss=0.101, over 27774.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3441, pruned_loss=0.09163, over 5674414.12 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3641, pruned_loss=0.1218, over 5672859.11 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08922, over 5671127.41 frames. ], batch size: 474, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:15:22,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2529, 1.4739, 1.1291, 1.3156], device='cuda:0'), covar=tensor([0.2913, 0.2714, 0.3293, 0.2268], device='cuda:0'), in_proj_covar=tensor([0.1328, 0.0971, 0.1169, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 10:15:40,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.158e+02 1.266e+03 1.604e+03 2.190e+03 5.168e+03, threshold=3.208e+03, percent-clipped=5.0
+2023-03-06 10:16:11,038 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=532931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:16:21,956 INFO [train.py:968] (0/2) Epoch 12, batch 31750, giga_loss[loss=0.2786, simple_loss=0.3635, pruned_loss=0.09687, over 28845.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3408, pruned_loss=0.0912, over 5670247.99 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3626, pruned_loss=0.1209, over 5670726.55 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3403, pruned_loss=0.08881, over 5669270.51 frames. ], batch size: 174, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:16:25,676 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3709, 1.6695, 1.4840, 1.6688], device='cuda:0'), covar=tensor([0.0754, 0.0291, 0.0310, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0116, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 10:16:41,673 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:16:42,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9330, 2.7835, 1.8515, 0.8806], device='cuda:0'), covar=tensor([0.5988, 0.2874, 0.3378, 0.5645], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1491, 0.1502, 0.1290], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 10:16:47,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:17:00,474 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-06 10:17:37,996 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:17:45,481 INFO [train.py:968] (0/2) Epoch 12, batch 31800, giga_loss[loss=0.2423, simple_loss=0.3249, pruned_loss=0.07982, over 28973.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3429, pruned_loss=0.09332, over 5671599.79 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3627, pruned_loss=0.121, over 5671894.92 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3423, pruned_loss=0.09125, over 5669846.39 frames. ], batch size: 136, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:18:03,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6334, 1.6806, 1.2412, 1.3575], device='cuda:0'), covar=tensor([0.0779, 0.0563, 0.0957, 0.1075], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0434, 0.0498, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 10:18:06,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.916e+02 1.401e+03 1.948e+03 2.740e+03 7.855e+03, threshold=3.896e+03, percent-clipped=21.0
+2023-03-06 10:18:45,408 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-06 10:18:56,511 INFO [train.py:968] (0/2) Epoch 12, batch 31850, libri_loss[loss=0.2667, simple_loss=0.3367, pruned_loss=0.09836, over 29478.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3391, pruned_loss=0.09113, over 5672576.04 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3625, pruned_loss=0.1208, over 5668093.56 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3382, pruned_loss=0.08886, over 5674439.85 frames. ], batch size: 85, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:19:41,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=533077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:20:01,891 INFO [train.py:968] (0/2) Epoch 12, batch 31900, giga_loss[loss=0.2266, simple_loss=0.2931, pruned_loss=0.08, over 23415.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3362, pruned_loss=0.08934, over 5661772.02 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3623, pruned_loss=0.1207, over 5662223.17 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3354, pruned_loss=0.08725, over 5668255.83 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:20:20,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.491e+02 1.296e+03 1.621e+03 2.441e+03 1.117e+04, threshold=3.243e+03, percent-clipped=14.0
+2023-03-06 10:21:03,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5537, 2.0563, 1.7064, 1.5044], device='cuda:0'), covar=tensor([0.2456, 0.1606, 0.1651, 0.1815], device='cuda:0'), in_proj_covar=tensor([0.1688, 0.1605, 0.1532, 0.1644], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 10:21:05,791 INFO [train.py:968] (0/2) Epoch 12, batch 31950, giga_loss[loss=0.2733, simple_loss=0.3487, pruned_loss=0.09896, over 28715.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3345, pruned_loss=0.08913, over 5672759.18 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.362, pruned_loss=0.1207, over 5664510.76 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3335, pruned_loss=0.08679, over 5676125.68 frames. ], batch size: 263, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:21:58,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6688, 3.4858, 3.2882, 1.9124], device='cuda:0'), covar=tensor([0.0637, 0.0842, 0.0911, 0.1958], device='cuda:0'), in_proj_covar=tensor([0.1042, 0.0973, 0.0852, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 10:22:01,965 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1215, 1.4559, 1.2999, 1.0509], device='cuda:0'), covar=tensor([0.2009, 0.1591, 0.1065, 0.1494], device='cuda:0'), in_proj_covar=tensor([0.1692, 0.1607, 0.1537, 0.1648], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 10:22:12,815 INFO [train.py:968] (0/2) Epoch 12, batch 32000, giga_loss[loss=0.239, simple_loss=0.3068, pruned_loss=0.08562, over 24366.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3377, pruned_loss=0.09036, over 5674367.83 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3617, pruned_loss=0.1206, over 5667094.08 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.337, pruned_loss=0.08834, over 5674581.38 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 10:22:29,095 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.482e+02 1.355e+03 1.674e+03 2.311e+03 4.556e+03, threshold=3.349e+03, percent-clipped=7.0
+2023-03-06 10:22:38,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6415, 1.9085, 1.9840, 1.4837], device='cuda:0'), covar=tensor([0.1607, 0.2248, 0.1291, 0.1554], device='cuda:0'), in_proj_covar=tensor([0.0835, 0.0687, 0.0874, 0.0783], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 10:22:45,999 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=533220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:22:49,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=533223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:22:56,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=533231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:23:08,475 INFO [train.py:968] (0/2) Epoch 12, batch 32050, giga_loss[loss=0.2759, simple_loss=0.3477, pruned_loss=0.102, over 27600.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3379, pruned_loss=0.09154, over 5681559.88 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3609, pruned_loss=0.1201, over 5665046.19 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3374, pruned_loss=0.08943, over 5683600.14 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:23:27,469 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=533252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:23:48,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4579, 1.5024, 1.1438, 1.1103], device='cuda:0'), covar=tensor([0.0767, 0.0518, 0.1009, 0.1065], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0438, 0.0503, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 10:24:10,409 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4698, 1.6931, 1.6655, 1.5115], device='cuda:0'), covar=tensor([0.1355, 0.1650, 0.1648, 0.1584], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0714, 0.0658, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-06 10:24:15,289 INFO [train.py:968] (0/2) Epoch 12, batch 32100, giga_loss[loss=0.2734, simple_loss=0.3282, pruned_loss=0.1093, over 24325.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3366, pruned_loss=0.09167, over 5678220.05 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.361, pruned_loss=0.1203, over 5665514.90 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3356, pruned_loss=0.08926, over 5679747.01 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:24:31,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=533306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:24:34,194 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.101e+02 1.380e+03 1.685e+03 2.351e+03 4.932e+03, threshold=3.370e+03, percent-clipped=9.0
+2023-03-06 10:25:11,649 INFO [train.py:968] (0/2) Epoch 12, batch 32150, giga_loss[loss=0.2583, simple_loss=0.3412, pruned_loss=0.08773, over 28927.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3372, pruned_loss=0.09261, over 5676142.22 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3608, pruned_loss=0.1202, over 5661297.03 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.336, pruned_loss=0.09003, over 5681917.69 frames. ], batch size: 145, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:25:59,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6332, 1.9336, 1.6697, 1.6429], device='cuda:0'), covar=tensor([0.1443, 0.1996, 0.1850, 0.1822], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0711, 0.0657, 0.0646], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-06 10:25:59,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=533374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:26:03,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=533377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:26:22,766 INFO [train.py:968] (0/2) Epoch 12, batch 32200, giga_loss[loss=0.2617, simple_loss=0.3537, pruned_loss=0.08486, over 29024.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.339, pruned_loss=0.09303, over 5674487.07 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3604, pruned_loss=0.12, over 5665268.93 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3381, pruned_loss=0.09079, over 5675412.80 frames. ], batch size: 199, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:26:45,349 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=533406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:26:47,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.980e+02 1.562e+03 1.998e+03 2.582e+03 6.622e+03, threshold=3.996e+03, percent-clipped=14.0
+2023-03-06 10:27:43,134 INFO [train.py:968] (0/2) Epoch 12, batch 32250, giga_loss[loss=0.2352, simple_loss=0.3198, pruned_loss=0.07534, over 28458.00 frames. ], tot_loss[loss=0.264, simple_loss=0.341, pruned_loss=0.09348, over 5663255.74 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3604, pruned_loss=0.12, over 5664092.42 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3401, pruned_loss=0.09148, over 5665293.13 frames. ], batch size: 60, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:27:56,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=533449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:28:01,168 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=533452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:28:42,344 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=533481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:28:56,973 INFO [train.py:968] (0/2) Epoch 12, batch 32300, giga_loss[loss=0.2547, simple_loss=0.3251, pruned_loss=0.09216, over 29120.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3391, pruned_loss=0.0926, over 5667412.91 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.36, pruned_loss=0.1198, over 5665667.77 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3384, pruned_loss=0.09073, over 5667834.76 frames. ], batch size: 113, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:29:15,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.574e+02 1.386e+03 1.662e+03 2.732e+03 1.178e+04, threshold=3.324e+03, percent-clipped=10.0
+2023-03-06 10:29:59,670 INFO [train.py:968] (0/2) Epoch 12, batch 32350, giga_loss[loss=0.2546, simple_loss=0.3145, pruned_loss=0.09734, over 26966.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3348, pruned_loss=0.09148, over 5671927.49 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3598, pruned_loss=0.1197, over 5668728.26 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3338, pruned_loss=0.08933, over 5670007.05 frames. ], batch size: 555, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:30:06,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5428, 1.7360, 1.6589, 1.5698], device='cuda:0'), covar=tensor([0.1443, 0.1961, 0.1857, 0.1708], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0720, 0.0663, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 10:31:06,176 INFO [train.py:968] (0/2) Epoch 12, batch 32400, giga_loss[loss=0.2511, simple_loss=0.3314, pruned_loss=0.08542, over 28901.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3303, pruned_loss=0.08961, over 5672272.60 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3598, pruned_loss=0.1197, over 5673623.50 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3289, pruned_loss=0.08722, over 5666558.12 frames. ], batch size: 164, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 10:31:20,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=533604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:31:25,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.092e+02 1.362e+03 1.675e+03 2.149e+03 5.728e+03, threshold=3.350e+03, percent-clipped=9.0
+2023-03-06 10:31:46,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5099, 1.6445, 1.5245, 1.4016], device='cuda:0'), covar=tensor([0.1764, 0.1748, 0.1369, 0.1578], device='cuda:0'), in_proj_covar=tensor([0.1705, 0.1617, 0.1542, 0.1652], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 10:32:05,169 INFO [train.py:968] (0/2) Epoch 12, batch 32450, giga_loss[loss=0.2558, simple_loss=0.3312, pruned_loss=0.09022, over 28782.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3322, pruned_loss=0.0913, over 5666350.13 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3595, pruned_loss=0.1198, over 5668598.28 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3307, pruned_loss=0.0887, over 5667227.78 frames. ], batch size: 119, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:33:02,278 INFO [train.py:968] (0/2) Epoch 12, batch 32500, giga_loss[loss=0.2258, simple_loss=0.3095, pruned_loss=0.07106, over 28414.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3339, pruned_loss=0.09213, over 5665868.01 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3594, pruned_loss=0.1198, over 5663360.03 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3322, pruned_loss=0.08942, over 5671393.54 frames. ], batch size: 368, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:33:23,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.576e+02 1.531e+03 2.237e+03 3.321e+03 9.273e+03, threshold=4.473e+03, percent-clipped=23.0
+2023-03-06 10:34:02,061 INFO [train.py:968] (0/2) Epoch 12, batch 32550, giga_loss[loss=0.2876, simple_loss=0.3433, pruned_loss=0.1159, over 26856.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3313, pruned_loss=0.09033, over 5664371.25 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3593, pruned_loss=0.12, over 5667768.47 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3292, pruned_loss=0.08704, over 5664935.86 frames. ], batch size: 555, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:34:34,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3049, 1.8331, 1.3333, 0.5377], device='cuda:0'), covar=tensor([0.3011, 0.1621, 0.2741, 0.3510], device='cuda:0'), in_proj_covar=tensor([0.1581, 0.1501, 0.1502, 0.1295], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 10:35:05,001 INFO [train.py:968] (0/2) Epoch 12, batch 32600, giga_loss[loss=0.2249, simple_loss=0.3062, pruned_loss=0.07184, over 28753.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3304, pruned_loss=0.08951, over 5665434.47 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3591, pruned_loss=0.1198, over 5673928.37 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3283, pruned_loss=0.08635, over 5660343.24 frames. ], batch size: 119, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:35:25,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.152e+02 1.418e+03 1.721e+03 2.591e+03 9.081e+03, threshold=3.443e+03, percent-clipped=7.0
+2023-03-06 10:35:27,351 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=533812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:36:10,042 INFO [train.py:968] (0/2) Epoch 12, batch 32650, giga_loss[loss=0.2322, simple_loss=0.3143, pruned_loss=0.07502, over 29026.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3293, pruned_loss=0.08976, over 5674519.37 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3585, pruned_loss=0.1197, over 5682529.43 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.327, pruned_loss=0.08621, over 5662651.69 frames. ], batch size: 285, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:36:25,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7474, 2.1997, 1.9823, 1.5956], device='cuda:0'), covar=tensor([0.2415, 0.1568, 0.1663, 0.2035], device='cuda:0'), in_proj_covar=tensor([0.1703, 0.1614, 0.1541, 0.1653], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 10:36:28,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1613, 1.3380, 1.1026, 1.0064], device='cuda:0'), covar=tensor([0.0878, 0.0480, 0.1027, 0.0962], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0433, 0.0499, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 10:36:58,028 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8987, 2.1100, 1.4719, 1.5407], device='cuda:0'), covar=tensor([0.0795, 0.0513, 0.0914, 0.1008], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0434, 0.0500, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 10:37:09,995 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4390, 3.6761, 1.6367, 1.5196], device='cuda:0'), covar=tensor([0.0914, 0.0353, 0.0901, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0507, 0.0345, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 10:37:12,251 INFO [train.py:968] (0/2) Epoch 12, batch 32700, giga_loss[loss=0.2623, simple_loss=0.3392, pruned_loss=0.09272, over 28928.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3304, pruned_loss=0.08951, over 5685333.68 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3581, pruned_loss=0.1196, over 5686214.45 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3281, pruned_loss=0.08591, over 5672462.40 frames. ], batch size: 186, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:37:35,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.383e+02 1.288e+03 1.712e+03 2.286e+03 6.994e+03, threshold=3.424e+03, percent-clipped=7.0
+2023-03-06 10:37:50,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-06 10:38:14,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4443, 1.6438, 1.5832, 1.5028], device='cuda:0'), covar=tensor([0.1269, 0.1731, 0.1768, 0.1588], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0715, 0.0658, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:0')
+2023-03-06 10:38:16,304 INFO [train.py:968] (0/2) Epoch 12, batch 32750, giga_loss[loss=0.2407, simple_loss=0.3231, pruned_loss=0.07917, over 28799.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3316, pruned_loss=0.0906, over 5689131.87 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3578, pruned_loss=0.1195, over 5692225.81 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3293, pruned_loss=0.08698, over 5673461.14 frames. ], batch size: 243, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:38:46,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5176, 1.8995, 1.5551, 1.7311], device='cuda:0'), covar=tensor([0.0731, 0.0271, 0.0309, 0.0773], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0117, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0055, 0.0092], device='cuda:0')
+2023-03-06 10:38:59,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=533979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:39:11,574 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=533989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:39:14,153 INFO [train.py:968] (0/2) Epoch 12, batch 32800, giga_loss[loss=0.2309, simple_loss=0.3139, pruned_loss=0.07397, over 28355.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3327, pruned_loss=0.09197, over 5695415.61 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3576, pruned_loss=0.1195, over 5699895.47 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3302, pruned_loss=0.08814, over 5675990.18 frames. ], batch size: 368, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:39:27,147 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-534000.pt
+2023-03-06 10:39:36,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.688e+02 1.225e+03 1.640e+03 2.297e+03 1.137e+04, threshold=3.279e+03, percent-clipped=10.0
+2023-03-06 10:40:19,366 INFO [train.py:968] (0/2) Epoch 12, batch 32850, giga_loss[loss=0.2329, simple_loss=0.3208, pruned_loss=0.07249, over 28695.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3313, pruned_loss=0.09038, over 5687761.26 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3576, pruned_loss=0.1195, over 5701218.41 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3287, pruned_loss=0.0867, over 5670591.48 frames. ], batch size: 262, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:40:23,303 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=534045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:40:51,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6710, 2.2098, 1.8487, 1.5768], device='cuda:0'), covar=tensor([0.2223, 0.1488, 0.1540, 0.1698], device='cuda:0'), in_proj_covar=tensor([0.1713, 0.1624, 0.1545, 0.1655], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 10:41:18,064 INFO [train.py:968] (0/2) Epoch 12, batch 32900, giga_loss[loss=0.2684, simple_loss=0.3569, pruned_loss=0.08991, over 28957.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3327, pruned_loss=0.08947, over 5678442.50 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3573, pruned_loss=0.1193, over 5703958.02 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3305, pruned_loss=0.08635, over 5662012.11 frames. ], batch size: 155, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:41:39,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.196e+02 1.405e+03 1.793e+03 2.637e+03 1.204e+04, threshold=3.586e+03, percent-clipped=11.0
+2023-03-06 10:41:58,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:42:00,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:42:19,260 INFO [train.py:968] (0/2) Epoch 12, batch 32950, giga_loss[loss=0.2442, simple_loss=0.332, pruned_loss=0.07821, over 28903.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3358, pruned_loss=0.09052, over 5676544.22 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3573, pruned_loss=0.1192, over 5705793.23 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.334, pruned_loss=0.08783, over 5661784.11 frames. ], batch size: 213, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:42:33,894 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:43:13,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:43:19,620 INFO [train.py:968] (0/2) Epoch 12, batch 33000, giga_loss[loss=0.2562, simple_loss=0.3493, pruned_loss=0.08155, over 28950.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3371, pruned_loss=0.09147, over 5674810.00 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3567, pruned_loss=0.119, over 5701575.56 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3355, pruned_loss=0.08863, over 5665837.89 frames. ], batch size: 213, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:43:19,625 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 10:43:28,201 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2005, simple_loss=0.3015, pruned_loss=0.04977, over 944034.00 frames. 
+2023-03-06 10:43:28,201 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 10:43:28,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=534192.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 10:43:46,319 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.666e+02 1.284e+03 1.617e+03 2.160e+03 5.254e+03, threshold=3.235e+03, percent-clipped=7.0
+2023-03-06 10:44:14,489 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 10:44:14,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3415, 1.2813, 3.9744, 3.0944], device='cuda:0'), covar=tensor([0.1484, 0.2516, 0.0444, 0.0957], device='cuda:0'), in_proj_covar=tensor([0.0667, 0.0597, 0.0855, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 10:44:24,287 INFO [train.py:968] (0/2) Epoch 12, batch 33050, giga_loss[loss=0.2612, simple_loss=0.3499, pruned_loss=0.08622, over 28703.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3375, pruned_loss=0.09218, over 5667306.97 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3563, pruned_loss=0.1188, over 5697593.97 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3358, pruned_loss=0.08907, over 5663379.21 frames. ], batch size: 307, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:45:01,730 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=534275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:45:20,306 INFO [train.py:968] (0/2) Epoch 12, batch 33100, giga_loss[loss=0.2094, simple_loss=0.2973, pruned_loss=0.06075, over 28965.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3334, pruned_loss=0.08947, over 5674940.98 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3556, pruned_loss=0.1184, over 5700348.29 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.332, pruned_loss=0.08642, over 5668074.28 frames. ], batch size: 213, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:45:43,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.322e+02 1.221e+03 1.499e+03 2.050e+03 4.768e+03, threshold=2.997e+03, percent-clipped=6.0
+2023-03-06 10:46:09,952 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:46:12,829 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:46:23,839 INFO [train.py:968] (0/2) Epoch 12, batch 33150, giga_loss[loss=0.2498, simple_loss=0.3286, pruned_loss=0.08549, over 28113.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3328, pruned_loss=0.08898, over 5659075.22 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3558, pruned_loss=0.1186, over 5683558.54 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3312, pruned_loss=0.08595, over 5667893.87 frames. ], batch size: 412, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:46:39,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3701, 3.0824, 2.3803, 1.9164], device='cuda:0'), covar=tensor([0.2074, 0.1039, 0.1387, 0.1719], device='cuda:0'), in_proj_covar=tensor([0.1710, 0.1617, 0.1544, 0.1649], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 10:46:46,073 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:46:47,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:47:04,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3482, 1.5461, 1.3862, 1.5141], device='cuda:0'), covar=tensor([0.0769, 0.0310, 0.0325, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0117, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 10:47:21,816 INFO [train.py:968] (0/2) Epoch 12, batch 33200, giga_loss[loss=0.2606, simple_loss=0.3485, pruned_loss=0.08634, over 28101.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3311, pruned_loss=0.08846, over 5664855.97 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3554, pruned_loss=0.1184, over 5685910.78 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3295, pruned_loss=0.08543, over 5669724.43 frames. ], batch size: 412, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 10:47:44,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.412e+02 1.326e+03 1.697e+03 2.254e+03 4.615e+03, threshold=3.393e+03, percent-clipped=10.0
+2023-03-06 10:47:54,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:48:21,957 INFO [train.py:968] (0/2) Epoch 12, batch 33250, libri_loss[loss=0.2906, simple_loss=0.353, pruned_loss=0.1141, over 29207.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3325, pruned_loss=0.08915, over 5667591.25 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.355, pruned_loss=0.1181, over 5688039.09 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3311, pruned_loss=0.08627, over 5668830.51 frames. ], batch size: 97, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 10:49:24,886 INFO [train.py:968] (0/2) Epoch 12, batch 33300, giga_loss[loss=0.2703, simple_loss=0.3499, pruned_loss=0.09533, over 29087.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3352, pruned_loss=0.09055, over 5672553.47 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3556, pruned_loss=0.1185, over 5691531.69 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3331, pruned_loss=0.0873, over 5669929.93 frames. ], batch size: 285, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:49:45,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:49:50,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:49:50,829 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.180e+02 1.434e+03 2.095e+03 2.742e+03 9.638e+03, threshold=4.189e+03, percent-clipped=17.0
+2023-03-06 10:49:55,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4178, 1.8006, 1.4594, 1.4378], device='cuda:0'), covar=tensor([0.2373, 0.2276, 0.2560, 0.1851], device='cuda:0'), in_proj_covar=tensor([0.1318, 0.0970, 0.1164, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 10:50:18,786 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-06 10:50:23,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:50:28,481 INFO [train.py:968] (0/2) Epoch 12, batch 33350, giga_loss[loss=0.2388, simple_loss=0.3275, pruned_loss=0.07507, over 28803.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3358, pruned_loss=0.09178, over 5673702.02 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3547, pruned_loss=0.118, over 5697965.76 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3342, pruned_loss=0.08872, over 5665048.97 frames. ], batch size: 119, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:50:57,660 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:51:00,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:51:03,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534567.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:51:33,478 INFO [train.py:968] (0/2) Epoch 12, batch 33400, giga_loss[loss=0.2869, simple_loss=0.3629, pruned_loss=0.1055, over 27563.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3389, pruned_loss=0.09372, over 5661224.37 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3542, pruned_loss=0.1177, over 5701141.76 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3379, pruned_loss=0.09107, over 5650966.89 frames. ], batch size: 474, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:51:33,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9219, 1.9905, 1.4474, 1.5770], device='cuda:0'), covar=tensor([0.0690, 0.0487, 0.0895, 0.0944], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0433, 0.0501, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 10:51:37,530 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:51:48,973 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3630, 3.7472, 1.5847, 1.5755], device='cuda:0'), covar=tensor([0.0964, 0.0341, 0.0909, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0506, 0.0345, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 10:51:50,573 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=534607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:51:54,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.121e+02 1.398e+03 1.729e+03 2.196e+03 6.055e+03, threshold=3.458e+03, percent-clipped=3.0
+2023-03-06 10:52:15,680 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 10:52:29,613 INFO [train.py:968] (0/2) Epoch 12, batch 33450, giga_loss[loss=0.2437, simple_loss=0.3407, pruned_loss=0.07339, over 28943.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3414, pruned_loss=0.09405, over 5670417.16 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.354, pruned_loss=0.1176, over 5703813.95 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3406, pruned_loss=0.09166, over 5659102.38 frames. ], batch size: 155, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:52:39,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:53:33,396 INFO [train.py:968] (0/2) Epoch 12, batch 33500, libri_loss[loss=0.3132, simple_loss=0.3663, pruned_loss=0.13, over 29537.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.343, pruned_loss=0.09528, over 5674563.40 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3542, pruned_loss=0.1178, over 5709576.19 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3418, pruned_loss=0.09253, over 5659285.57 frames. ], batch size: 79, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:54:00,435 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534710.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 10:54:03,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.244e+02 1.559e+03 2.008e+03 2.771e+03 1.619e+04, threshold=4.017e+03, percent-clipped=14.0
+2023-03-06 10:54:04,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534713.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 10:54:44,291 INFO [train.py:968] (0/2) Epoch 12, batch 33550, libri_loss[loss=0.3174, simple_loss=0.385, pruned_loss=0.1249, over 27478.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3412, pruned_loss=0.09476, over 5680418.32 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.354, pruned_loss=0.1176, over 5714076.82 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.34, pruned_loss=0.09194, over 5663218.38 frames. ], batch size: 115, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:54:44,639 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534742.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:55:46,371 INFO [train.py:968] (0/2) Epoch 12, batch 33600, giga_loss[loss=0.2391, simple_loss=0.3245, pruned_loss=0.07691, over 28896.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3389, pruned_loss=0.09304, over 5686538.42 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3539, pruned_loss=0.1176, over 5713941.86 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3378, pruned_loss=0.09055, over 5672865.54 frames. ], batch size: 155, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:55:49,401 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:55:53,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:56:17,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.792e+02 1.227e+03 1.621e+03 2.168e+03 7.047e+03, threshold=3.241e+03, percent-clipped=7.0
+2023-03-06 10:56:25,380 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-06 10:56:28,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9203, 3.7437, 3.4805, 1.9604], device='cuda:0'), covar=tensor([0.0623, 0.0761, 0.0758, 0.2003], device='cuda:0'), in_proj_covar=tensor([0.1037, 0.0968, 0.0841, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 10:56:36,123 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:57:00,220 INFO [train.py:968] (0/2) Epoch 12, batch 33650, giga_loss[loss=0.2592, simple_loss=0.3331, pruned_loss=0.0927, over 29014.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3374, pruned_loss=0.09227, over 5682999.22 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3537, pruned_loss=0.1174, over 5714868.15 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3366, pruned_loss=0.09032, over 5671304.47 frames. ], batch size: 199, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:57:04,442 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=534845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:57:09,941 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 10:58:06,507 INFO [train.py:968] (0/2) Epoch 12, batch 33700, giga_loss[loss=0.2808, simple_loss=0.3502, pruned_loss=0.1057, over 29022.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3349, pruned_loss=0.09148, over 5683891.03 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3536, pruned_loss=0.1174, over 5716150.01 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3341, pruned_loss=0.08962, over 5672962.07 frames. ], batch size: 199, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:58:28,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.626e+02 1.405e+03 1.886e+03 2.218e+03 5.999e+03, threshold=3.772e+03, percent-clipped=8.0
+2023-03-06 10:59:08,215 INFO [train.py:968] (0/2) Epoch 12, batch 33750, giga_loss[loss=0.2781, simple_loss=0.3569, pruned_loss=0.09963, over 28697.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3342, pruned_loss=0.09082, over 5690365.61 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3535, pruned_loss=0.1174, over 5719834.32 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3333, pruned_loss=0.08888, over 5678146.12 frames. ], batch size: 307, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:59:58,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:00:09,762 INFO [train.py:968] (0/2) Epoch 12, batch 33800, giga_loss[loss=0.2437, simple_loss=0.3285, pruned_loss=0.07948, over 28511.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3332, pruned_loss=0.08973, over 5675508.02 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3537, pruned_loss=0.1176, over 5714771.63 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3319, pruned_loss=0.08737, over 5669041.75 frames. ], batch size: 369, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:00:12,758 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-06 11:00:22,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2549, 1.4228, 1.2651, 1.4972], device='cuda:0'), covar=tensor([0.0782, 0.0360, 0.0348, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0117, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 11:00:34,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.368e+02 1.470e+03 1.851e+03 2.999e+03 1.133e+04, threshold=3.701e+03, percent-clipped=11.0
+2023-03-06 11:00:59,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1040, 1.1828, 4.1646, 3.0914], device='cuda:0'), covar=tensor([0.1821, 0.2731, 0.0336, 0.0900], device='cuda:0'), in_proj_covar=tensor([0.0662, 0.0591, 0.0847, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:01:06,301 INFO [train.py:968] (0/2) Epoch 12, batch 33850, giga_loss[loss=0.2741, simple_loss=0.3505, pruned_loss=0.09888, over 28732.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3349, pruned_loss=0.08901, over 5682198.18 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3535, pruned_loss=0.1175, over 5719223.68 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3336, pruned_loss=0.08658, over 5672205.46 frames. ], batch size: 99, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:01:22,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3727, 3.5584, 1.6143, 1.4751], device='cuda:0'), covar=tensor([0.0876, 0.0287, 0.0862, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0506, 0.0347, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 11:01:30,597 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-06 11:01:36,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-06 11:02:02,048 INFO [train.py:968] (0/2) Epoch 12, batch 33900, giga_loss[loss=0.2488, simple_loss=0.3353, pruned_loss=0.08117, over 29073.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3368, pruned_loss=0.08907, over 5672740.65 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3539, pruned_loss=0.1178, over 5706260.40 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3349, pruned_loss=0.08599, over 5674579.62 frames. ], batch size: 128, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:02:25,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.629e+02 1.369e+03 1.782e+03 2.473e+03 5.940e+03, threshold=3.564e+03, percent-clipped=5.0
+2023-03-06 11:02:40,030 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=535125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:02:42,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=535128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:02:50,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-06 11:02:59,417 INFO [train.py:968] (0/2) Epoch 12, batch 33950, giga_loss[loss=0.2386, simple_loss=0.3204, pruned_loss=0.0784, over 27768.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3374, pruned_loss=0.08917, over 5676551.05 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3536, pruned_loss=0.1175, over 5709204.74 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3356, pruned_loss=0.08602, over 5674260.87 frames. ], batch size: 474, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:03:22,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=535157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:04:02,602 INFO [train.py:968] (0/2) Epoch 12, batch 34000, giga_loss[loss=0.278, simple_loss=0.3598, pruned_loss=0.0981, over 28669.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3373, pruned_loss=0.08951, over 5671458.63 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3535, pruned_loss=0.1173, over 5706172.53 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3353, pruned_loss=0.08612, over 5671035.15 frames. ], batch size: 262, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:04:35,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.000e+02 1.373e+03 1.844e+03 2.336e+03 6.012e+03, threshold=3.688e+03, percent-clipped=7.0
+2023-03-06 11:04:46,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=535220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:05:11,372 INFO [train.py:968] (0/2) Epoch 12, batch 34050, giga_loss[loss=0.2609, simple_loss=0.3421, pruned_loss=0.08984, over 29032.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3367, pruned_loss=0.08882, over 5670382.12 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3536, pruned_loss=0.1173, over 5708153.72 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3347, pruned_loss=0.08564, over 5667603.62 frames. ], batch size: 285, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:05:21,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5768, 1.6606, 1.2448, 1.2208], device='cuda:0'), covar=tensor([0.0873, 0.0575, 0.1041, 0.1107], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0432, 0.0499, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:05:51,504 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=535270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:06:18,772 INFO [train.py:968] (0/2) Epoch 12, batch 34100, giga_loss[loss=0.2445, simple_loss=0.3368, pruned_loss=0.0761, over 28870.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3377, pruned_loss=0.08917, over 5661827.05 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3536, pruned_loss=0.1175, over 5698683.30 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3357, pruned_loss=0.08565, over 5667081.02 frames. ], batch size: 164, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:06:53,750 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.968e+02 1.268e+03 1.667e+03 2.860e+03 9.327e+03, threshold=3.334e+03, percent-clipped=16.0
+2023-03-06 11:07:24,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3433, 1.5627, 1.3735, 1.4902], device='cuda:0'), covar=tensor([0.0755, 0.0327, 0.0322, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0116, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0059, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 11:07:31,118 INFO [train.py:968] (0/2) Epoch 12, batch 34150, giga_loss[loss=0.2659, simple_loss=0.3466, pruned_loss=0.09262, over 28969.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3387, pruned_loss=0.08987, over 5650495.98 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3537, pruned_loss=0.1178, over 5684477.69 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3366, pruned_loss=0.08611, over 5666302.89 frames. ], batch size: 199, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:07:55,992 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=535363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:07:58,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=535366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:08:29,395 INFO [train.py:968] (0/2) Epoch 12, batch 34200, giga_loss[loss=0.2462, simple_loss=0.3377, pruned_loss=0.07742, over 28972.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3424, pruned_loss=0.09189, over 5656917.09 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3538, pruned_loss=0.1178, over 5683685.80 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.34, pruned_loss=0.08774, over 5668922.53 frames. ], batch size: 186, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:08:35,658 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=535395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:08:56,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.768e+02 1.476e+03 2.303e+03 3.507e+03 1.074e+04, threshold=4.605e+03, percent-clipped=27.0
+2023-03-06 11:09:17,580 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=535429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:09:34,346 INFO [train.py:968] (0/2) Epoch 12, batch 34250, giga_loss[loss=0.259, simple_loss=0.3349, pruned_loss=0.09148, over 28941.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3424, pruned_loss=0.09175, over 5662317.65 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3541, pruned_loss=0.118, over 5682418.39 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3399, pruned_loss=0.08752, over 5672343.68 frames. ], batch size: 186, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:10:51,526 INFO [train.py:968] (0/2) Epoch 12, batch 34300, giga_loss[loss=0.2676, simple_loss=0.3407, pruned_loss=0.09721, over 27800.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3413, pruned_loss=0.09194, over 5664146.02 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3539, pruned_loss=0.1179, over 5685894.79 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3393, pruned_loss=0.08831, over 5668805.26 frames. ], batch size: 476, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:11:18,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.499e+02 1.388e+03 1.679e+03 2.215e+03 5.858e+03, threshold=3.358e+03, percent-clipped=3.0
+2023-03-06 11:11:23,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7143, 2.4285, 2.0787, 1.6129], device='cuda:0'), covar=tensor([0.3031, 0.1543, 0.1451, 0.1987], device='cuda:0'), in_proj_covar=tensor([0.1723, 0.1596, 0.1534, 0.1647], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 11:12:00,020 INFO [train.py:968] (0/2) Epoch 12, batch 34350, giga_loss[loss=0.2708, simple_loss=0.3502, pruned_loss=0.09566, over 27913.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3385, pruned_loss=0.09033, over 5674536.39 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3539, pruned_loss=0.1179, over 5688727.47 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3365, pruned_loss=0.08669, over 5675582.44 frames. ], batch size: 476, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:13:08,134 INFO [train.py:968] (0/2) Epoch 12, batch 34400, giga_loss[loss=0.2494, simple_loss=0.3329, pruned_loss=0.08295, over 28633.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3368, pruned_loss=0.08835, over 5673520.02 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3541, pruned_loss=0.118, over 5682135.43 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3349, pruned_loss=0.08504, over 5679398.96 frames. ], batch size: 307, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:13:41,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.144e+02 1.300e+03 1.881e+03 2.806e+03 9.626e+03, threshold=3.761e+03, percent-clipped=18.0
+2023-03-06 11:13:47,764 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=535618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:14:16,704 INFO [train.py:968] (0/2) Epoch 12, batch 34450, giga_loss[loss=0.262, simple_loss=0.3442, pruned_loss=0.08997, over 28128.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3377, pruned_loss=0.08912, over 5664902.81 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3538, pruned_loss=0.1179, over 5684016.17 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3361, pruned_loss=0.08621, over 5667695.62 frames. ], batch size: 412, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:14:20,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=535645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:15:01,943 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=535682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:15:15,570 INFO [train.py:968] (0/2) Epoch 12, batch 34500, giga_loss[loss=0.2285, simple_loss=0.3198, pruned_loss=0.06858, over 28991.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3408, pruned_loss=0.09097, over 5670903.84 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3537, pruned_loss=0.1178, over 5687866.10 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3394, pruned_loss=0.08814, over 5669239.19 frames. ], batch size: 164, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:15:20,869 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 11:15:40,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.049e+02 1.396e+03 2.147e+03 3.011e+03 6.453e+03, threshold=4.294e+03, percent-clipped=13.0
+2023-03-06 11:16:12,750 INFO [train.py:968] (0/2) Epoch 12, batch 34550, giga_loss[loss=0.2483, simple_loss=0.3302, pruned_loss=0.08323, over 28930.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3404, pruned_loss=0.09097, over 5680990.79 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3539, pruned_loss=0.1177, over 5693434.35 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3387, pruned_loss=0.08798, over 5674212.96 frames. ], batch size: 227, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:16:16,138 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-06 11:16:34,343 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4775, 1.6507, 1.6848, 1.3137], device='cuda:0'), covar=tensor([0.1618, 0.2288, 0.1342, 0.1612], device='cuda:0'), in_proj_covar=tensor([0.0834, 0.0681, 0.0873, 0.0781], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 11:16:53,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9750, 1.3533, 1.0572, 0.1804], device='cuda:0'), covar=tensor([0.2684, 0.2157, 0.3455, 0.4195], device='cuda:0'), in_proj_covar=tensor([0.1571, 0.1494, 0.1498, 0.1286], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 11:17:07,855 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=535788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:17:09,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=535791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:17:09,997 INFO [train.py:968] (0/2) Epoch 12, batch 34600, giga_loss[loss=0.2562, simple_loss=0.3344, pruned_loss=0.08896, over 28329.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3373, pruned_loss=0.09077, over 5670489.43 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3535, pruned_loss=0.1175, over 5693854.86 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3359, pruned_loss=0.08778, over 5664225.28 frames. ], batch size: 368, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:17:15,630 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-06 11:17:23,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=535804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:17:36,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.986e+02 1.502e+03 2.052e+03 3.039e+03 9.560e+03, threshold=4.103e+03, percent-clipped=11.0
+2023-03-06 11:17:44,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=535820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:18:06,071 INFO [train.py:968] (0/2) Epoch 12, batch 34650, giga_loss[loss=0.2672, simple_loss=0.3432, pruned_loss=0.0956, over 28735.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3367, pruned_loss=0.09114, over 5659714.75 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3536, pruned_loss=0.1175, over 5686848.58 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3352, pruned_loss=0.08817, over 5660832.60 frames. ], batch size: 262, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:18:57,861 INFO [train.py:968] (0/2) Epoch 12, batch 34700, giga_loss[loss=0.3351, simple_loss=0.3961, pruned_loss=0.137, over 28669.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3416, pruned_loss=0.09457, over 5666786.86 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3527, pruned_loss=0.1171, over 5688757.31 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3408, pruned_loss=0.09187, over 5665280.49 frames. ], batch size: 92, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:19:13,948 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=535910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:19:17,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.648e+02 1.403e+03 1.853e+03 2.581e+03 8.505e+03, threshold=3.707e+03, percent-clipped=6.0
+2023-03-06 11:19:44,050 INFO [train.py:968] (0/2) Epoch 12, batch 34750, libri_loss[loss=0.2326, simple_loss=0.2984, pruned_loss=0.08344, over 29390.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3499, pruned_loss=0.09962, over 5666545.61 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.352, pruned_loss=0.1167, over 5685397.58 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3496, pruned_loss=0.09721, over 5667372.50 frames. ], batch size: 67, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:19:50,652 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=535947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:19:52,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=535950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:20:20,765 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=535979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:20:33,207 INFO [train.py:968] (0/2) Epoch 12, batch 34800, libri_loss[loss=0.286, simple_loss=0.3435, pruned_loss=0.1142, over 19635.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3561, pruned_loss=0.1034, over 5654561.73 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3521, pruned_loss=0.1167, over 5676694.16 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3558, pruned_loss=0.1013, over 5663688.77 frames. ], batch size: 186, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:20:34,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=535993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:20:39,801 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-536000.pt
+2023-03-06 11:20:44,309 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6531, 4.4238, 4.2263, 1.9163], device='cuda:0'), covar=tensor([0.0560, 0.0817, 0.0788, 0.1996], device='cuda:0'), in_proj_covar=tensor([0.1041, 0.0965, 0.0839, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 11:20:50,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.321e+02 1.266e+03 1.701e+03 2.344e+03 5.166e+03, threshold=3.402e+03, percent-clipped=5.0
+2023-03-06 11:21:12,161 INFO [train.py:968] (0/2) Epoch 12, batch 34850, giga_loss[loss=0.3028, simple_loss=0.3551, pruned_loss=0.1252, over 26791.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3546, pruned_loss=0.1038, over 5669296.36 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3518, pruned_loss=0.1164, over 5683186.13 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3549, pruned_loss=0.1017, over 5670166.40 frames. ], batch size: 555, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:21:13,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3043, 1.2837, 4.3551, 3.4093], device='cuda:0'), covar=tensor([0.2116, 0.3206, 0.0624, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0667, 0.0597, 0.0864, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:21:20,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0061, 2.0444, 1.4634, 1.6790], device='cuda:0'), covar=tensor([0.0791, 0.0715, 0.0995, 0.1139], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0433, 0.0498, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:21:20,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6058, 2.0569, 1.6627, 1.4221], device='cuda:0'), covar=tensor([0.2685, 0.1968, 0.1909, 0.2302], device='cuda:0'), in_proj_covar=tensor([0.1714, 0.1594, 0.1538, 0.1654], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 11:21:24,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=536057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:21:50,002 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9674, 1.1764, 1.2658, 1.0772], device='cuda:0'), covar=tensor([0.1445, 0.1289, 0.1981, 0.1405], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0722, 0.0667, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 11:21:52,133 INFO [train.py:968] (0/2) Epoch 12, batch 34900, giga_loss[loss=0.2974, simple_loss=0.3489, pruned_loss=0.123, over 26642.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3479, pruned_loss=0.1008, over 5676971.18 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.352, pruned_loss=0.1165, over 5679479.86 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3479, pruned_loss=0.09869, over 5680892.54 frames. ], batch size: 555, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:21:58,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2129, 2.3296, 2.4100, 1.9844], device='cuda:0'), covar=tensor([0.1594, 0.1993, 0.1232, 0.1451], device='cuda:0'), in_proj_covar=tensor([0.0838, 0.0687, 0.0876, 0.0784], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 11:22:12,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.299e+02 9.657e+02 1.330e+03 1.714e+03 4.050e+03, threshold=2.660e+03, percent-clipped=4.0
+2023-03-06 11:22:32,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=536136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:22:34,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=536139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:22:35,761 INFO [train.py:968] (0/2) Epoch 12, batch 34950, giga_loss[loss=0.2459, simple_loss=0.3178, pruned_loss=0.08701, over 28724.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3413, pruned_loss=0.09816, over 5679498.73 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3517, pruned_loss=0.1163, over 5681019.68 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3413, pruned_loss=0.09634, over 5681114.14 frames. ], batch size: 85, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:22:59,013 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=536168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:23:19,916 INFO [train.py:968] (0/2) Epoch 12, batch 35000, giga_loss[loss=0.2414, simple_loss=0.3121, pruned_loss=0.08531, over 28650.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3331, pruned_loss=0.0944, over 5681306.38 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3517, pruned_loss=0.1162, over 5682871.12 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3331, pruned_loss=0.09284, over 5680951.83 frames. ], batch size: 307, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:23:28,000 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=536200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:23:29,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=536203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:23:38,942 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.850e+02 9.735e+02 1.292e+03 1.753e+03 6.559e+03, threshold=2.584e+03, percent-clipped=7.0
+2023-03-06 11:23:53,063 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=536232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:23:59,909 INFO [train.py:968] (0/2) Epoch 12, batch 35050, giga_loss[loss=0.2096, simple_loss=0.2832, pruned_loss=0.06802, over 28675.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3259, pruned_loss=0.0911, over 5680469.06 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3515, pruned_loss=0.116, over 5676493.51 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3256, pruned_loss=0.08966, over 5684818.51 frames. ], batch size: 92, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:24:40,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=536285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:24:47,350 INFO [train.py:968] (0/2) Epoch 12, batch 35100, giga_loss[loss=0.2457, simple_loss=0.3173, pruned_loss=0.08704, over 28824.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3219, pruned_loss=0.08979, over 5686050.98 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3524, pruned_loss=0.1166, over 5679113.14 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3205, pruned_loss=0.08774, over 5687108.88 frames. ], batch size: 112, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:25:06,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.248e+02 9.782e+02 1.363e+03 1.970e+03 6.192e+03, threshold=2.727e+03, percent-clipped=13.0
+2023-03-06 11:25:30,331 INFO [train.py:968] (0/2) Epoch 12, batch 35150, giga_loss[loss=0.2202, simple_loss=0.2986, pruned_loss=0.07085, over 28198.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3202, pruned_loss=0.08933, over 5685475.79 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3528, pruned_loss=0.1167, over 5684618.81 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3179, pruned_loss=0.08695, over 5681458.69 frames. ], batch size: 368, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:25:31,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4047, 3.3217, 1.4923, 1.5020], device='cuda:0'), covar=tensor([0.0937, 0.0336, 0.0901, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0508, 0.0345, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 11:26:11,167 INFO [train.py:968] (0/2) Epoch 12, batch 35200, giga_loss[loss=0.2267, simple_loss=0.2899, pruned_loss=0.08171, over 28371.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3167, pruned_loss=0.0874, over 5697488.94 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.353, pruned_loss=0.1167, over 5690111.99 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3137, pruned_loss=0.08469, over 5689481.37 frames. ], batch size: 65, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:26:28,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.684e+02 1.025e+03 1.393e+03 2.049e+03 7.509e+03, threshold=2.786e+03, percent-clipped=14.0
+2023-03-06 11:26:38,984 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=536428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:26:43,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=536431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:26:46,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6512, 1.7741, 1.2989, 1.4159], device='cuda:0'), covar=tensor([0.0762, 0.0604, 0.0935, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0434, 0.0498, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:26:54,182 INFO [train.py:968] (0/2) Epoch 12, batch 35250, giga_loss[loss=0.214, simple_loss=0.2779, pruned_loss=0.07509, over 28446.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3142, pruned_loss=0.08595, over 5706285.61 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3536, pruned_loss=0.1171, over 5691023.97 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3107, pruned_loss=0.08304, over 5699269.66 frames. ], batch size: 85, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:27:12,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=536460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:27:39,343 INFO [train.py:968] (0/2) Epoch 12, batch 35300, giga_loss[loss=0.1931, simple_loss=0.272, pruned_loss=0.05709, over 29030.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3107, pruned_loss=0.08419, over 5706566.43 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3541, pruned_loss=0.1172, over 5695063.05 frames. ], giga_tot_loss[loss=0.2345, simple_loss=0.3067, pruned_loss=0.08111, over 5697659.36 frames. ], batch size: 136, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:27:44,381 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-06 11:27:59,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.298e+02 8.814e+02 1.106e+03 1.507e+03 4.576e+03, threshold=2.212e+03, percent-clipped=2.0
+2023-03-06 11:28:19,289 INFO [train.py:968] (0/2) Epoch 12, batch 35350, giga_loss[loss=0.2131, simple_loss=0.2837, pruned_loss=0.07128, over 28722.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3084, pruned_loss=0.08318, over 5682192.10 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3545, pruned_loss=0.1176, over 5673548.54 frames. ], giga_tot_loss[loss=0.2309, simple_loss=0.3033, pruned_loss=0.07923, over 5695797.25 frames. ], batch size: 86, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:28:49,802 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=536579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:29:02,796 INFO [train.py:968] (0/2) Epoch 12, batch 35400, giga_loss[loss=0.2113, simple_loss=0.2672, pruned_loss=0.07769, over 23711.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3046, pruned_loss=0.08108, over 5670093.02 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3546, pruned_loss=0.1177, over 5663590.35 frames. ], giga_tot_loss[loss=0.2273, simple_loss=0.2998, pruned_loss=0.0774, over 5690180.64 frames. ], batch size: 705, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:29:24,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.543e+02 1.049e+03 1.473e+03 1.946e+03 8.284e+03, threshold=2.946e+03, percent-clipped=17.0
+2023-03-06 11:29:39,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1847, 3.9854, 3.7860, 2.4535], device='cuda:0'), covar=tensor([0.0633, 0.0826, 0.0828, 0.1627], device='cuda:0'), in_proj_covar=tensor([0.1048, 0.0982, 0.0849, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 11:29:47,222 INFO [train.py:968] (0/2) Epoch 12, batch 35450, giga_loss[loss=0.2025, simple_loss=0.2764, pruned_loss=0.06427, over 28410.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3007, pruned_loss=0.07906, over 5667412.70 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3547, pruned_loss=0.1177, over 5656667.22 frames. ], giga_tot_loss[loss=0.2238, simple_loss=0.2962, pruned_loss=0.07571, over 5690257.42 frames. ], batch size: 65, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:29:57,945 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=536653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:30:05,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3269, 1.5721, 1.3228, 1.5283], device='cuda:0'), covar=tensor([0.0736, 0.0336, 0.0324, 0.0807], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0116, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 11:30:30,843 INFO [train.py:968] (0/2) Epoch 12, batch 35500, giga_loss[loss=0.2389, simple_loss=0.3028, pruned_loss=0.08747, over 28744.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.2987, pruned_loss=0.0783, over 5672550.79 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3552, pruned_loss=0.1179, over 5652444.82 frames. ], giga_tot_loss[loss=0.2216, simple_loss=0.2937, pruned_loss=0.07473, over 5694473.70 frames. ], batch size: 92, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:30:52,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.762e+02 1.028e+03 1.289e+03 1.761e+03 4.957e+03, threshold=2.579e+03, percent-clipped=7.0
+2023-03-06 11:31:00,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4803, 1.6211, 1.2376, 1.2035], device='cuda:0'), covar=tensor([0.0840, 0.0519, 0.1051, 0.1005], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0437, 0.0502, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:31:18,054 INFO [train.py:968] (0/2) Epoch 12, batch 35550, giga_loss[loss=0.3061, simple_loss=0.3706, pruned_loss=0.1208, over 29026.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3083, pruned_loss=0.08371, over 5670879.34 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3555, pruned_loss=0.1179, over 5658598.06 frames. ], giga_tot_loss[loss=0.2314, simple_loss=0.3028, pruned_loss=0.07997, over 5683407.01 frames. ], batch size: 106, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:32:06,823 INFO [train.py:968] (0/2) Epoch 12, batch 35600, giga_loss[loss=0.3127, simple_loss=0.379, pruned_loss=0.1232, over 27869.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3217, pruned_loss=0.09062, over 5676939.18 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3559, pruned_loss=0.118, over 5658202.06 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3163, pruned_loss=0.087, over 5687705.88 frames. ], batch size: 412, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:32:31,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.745e+02 1.236e+03 1.578e+03 2.083e+03 7.169e+03, threshold=3.155e+03, percent-clipped=15.0
+2023-03-06 11:32:47,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6686, 3.0327, 2.6241, 2.1814], device='cuda:0'), covar=tensor([0.1614, 0.1163, 0.1203, 0.1627], device='cuda:0'), in_proj_covar=tensor([0.1724, 0.1604, 0.1553, 0.1672], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 11:32:52,564 INFO [train.py:968] (0/2) Epoch 12, batch 35650, giga_loss[loss=0.2858, simple_loss=0.3662, pruned_loss=0.1027, over 29052.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3348, pruned_loss=0.09742, over 5681888.57 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3556, pruned_loss=0.1179, over 5659291.05 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3307, pruned_loss=0.09458, over 5689397.99 frames. ], batch size: 136, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:33:33,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4140, 3.1640, 2.3850, 1.8638], device='cuda:0'), covar=tensor([0.1843, 0.0972, 0.1305, 0.1668], device='cuda:0'), in_proj_covar=tensor([0.1726, 0.1605, 0.1555, 0.1671], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 11:33:36,421 INFO [train.py:968] (0/2) Epoch 12, batch 35700, giga_loss[loss=0.2652, simple_loss=0.346, pruned_loss=0.0922, over 28685.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3421, pruned_loss=0.1006, over 5670689.38 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.356, pruned_loss=0.118, over 5647244.07 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.338, pruned_loss=0.09772, over 5687884.16 frames. ], batch size: 60, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:33:47,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5958, 1.7385, 1.5825, 1.4007], device='cuda:0'), covar=tensor([0.2233, 0.1816, 0.1430, 0.1932], device='cuda:0'), in_proj_covar=tensor([0.1726, 0.1603, 0.1554, 0.1669], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 11:33:59,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.752e+02 1.159e+03 1.567e+03 2.058e+03 4.082e+03, threshold=3.133e+03, percent-clipped=5.0
+2023-03-06 11:34:12,144 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=536932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:34:21,960 INFO [train.py:968] (0/2) Epoch 12, batch 35750, giga_loss[loss=0.2871, simple_loss=0.3645, pruned_loss=0.1049, over 28702.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3449, pruned_loss=0.1005, over 5664471.81 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.356, pruned_loss=0.118, over 5639414.47 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3414, pruned_loss=0.098, over 5685731.88 frames. ], batch size: 284, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:34:34,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=536954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:34:37,257 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=536958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:35:08,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0503, 1.3684, 1.3249, 1.2275], device='cuda:0'), covar=tensor([0.1065, 0.0894, 0.1631, 0.1059], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0729, 0.0671, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 11:35:08,526 INFO [train.py:968] (0/2) Epoch 12, batch 35800, libri_loss[loss=0.4411, simple_loss=0.4623, pruned_loss=0.2099, over 29114.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3482, pruned_loss=0.1019, over 5675352.24 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3568, pruned_loss=0.1186, over 5646755.36 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3444, pruned_loss=0.09882, over 5686759.48 frames. ], batch size: 101, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:35:12,863 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2254, 0.8571, 0.9014, 1.3714], device='cuda:0'), covar=tensor([0.0772, 0.0395, 0.0349, 0.0820], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 11:35:32,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4901, 1.6434, 1.3823, 1.6999], device='cuda:0'), covar=tensor([0.2461, 0.2394, 0.2560, 0.2033], device='cuda:0'), in_proj_covar=tensor([0.1319, 0.0976, 0.1165, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 11:35:32,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.278e+02 1.236e+03 1.723e+03 2.448e+03 6.712e+03, threshold=3.446e+03, percent-clipped=14.0
+2023-03-06 11:35:40,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=537028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:35:53,146 INFO [train.py:968] (0/2) Epoch 12, batch 35850, giga_loss[loss=0.2867, simple_loss=0.3615, pruned_loss=0.106, over 28677.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3507, pruned_loss=0.1034, over 5686037.68 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3568, pruned_loss=0.1184, over 5650123.21 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3476, pruned_loss=0.1009, over 5692485.98 frames. ], batch size: 307, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:36:37,998 INFO [train.py:968] (0/2) Epoch 12, batch 35900, libri_loss[loss=0.2766, simple_loss=0.336, pruned_loss=0.1085, over 29650.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3537, pruned_loss=0.106, over 5678510.07 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3562, pruned_loss=0.1181, over 5653668.79 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3516, pruned_loss=0.104, over 5680794.23 frames. ], batch size: 69, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:36:42,365 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=537097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:36:45,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=537100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:36:57,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.380e+02 1.159e+03 1.426e+03 1.929e+03 4.582e+03, threshold=2.853e+03, percent-clipped=6.0
+2023-03-06 11:37:07,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=537129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:37:17,512 INFO [train.py:968] (0/2) Epoch 12, batch 35950, libri_loss[loss=0.3521, simple_loss=0.4117, pruned_loss=0.1463, over 29517.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3569, pruned_loss=0.108, over 5688858.62 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3568, pruned_loss=0.1183, over 5662721.40 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3546, pruned_loss=0.1058, over 5683516.39 frames. ], batch size: 84, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:37:43,554 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=537171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:37:45,378 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=537174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:37:52,392 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=537185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:37:56,784 INFO [train.py:968] (0/2) Epoch 12, batch 36000, giga_loss[loss=0.2947, simple_loss=0.3742, pruned_loss=0.1076, over 28220.00 frames. ], tot_loss[loss=0.289, simple_loss=0.36, pruned_loss=0.109, over 5693798.54 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3578, pruned_loss=0.1188, over 5666264.36 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3573, pruned_loss=0.1065, over 5686961.50 frames. ], batch size: 77, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:37:56,788 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 11:38:05,159 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2107, simple_loss=0.317, pruned_loss=0.05223, over 944034.00 frames. 
+2023-03-06 11:38:05,160 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 11:38:14,704 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=537203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:38:22,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3666, 1.6142, 1.3235, 1.7042], device='cuda:0'), covar=tensor([0.0794, 0.0302, 0.0331, 0.0837], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 11:38:25,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.255e+02 1.093e+03 1.516e+03 1.963e+03 6.436e+03, threshold=3.032e+03, percent-clipped=10.0
+2023-03-06 11:38:47,396 INFO [train.py:968] (0/2) Epoch 12, batch 36050, giga_loss[loss=0.2787, simple_loss=0.3489, pruned_loss=0.1043, over 28726.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3614, pruned_loss=0.1094, over 5690805.54 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3583, pruned_loss=0.1191, over 5668930.71 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3588, pruned_loss=0.1068, over 5683606.98 frames. ], batch size: 92, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:39:15,206 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3027, 3.2974, 1.3344, 1.5510], device='cuda:0'), covar=tensor([0.1044, 0.0246, 0.0949, 0.1352], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0503, 0.0342, 0.0391], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 11:39:26,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0320, 3.8052, 3.5978, 1.5429], device='cuda:0'), covar=tensor([0.0587, 0.0762, 0.0771, 0.2456], device='cuda:0'), in_proj_covar=tensor([0.1027, 0.0961, 0.0832, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 11:39:29,485 INFO [train.py:968] (0/2) Epoch 12, batch 36100, libri_loss[loss=0.3402, simple_loss=0.4023, pruned_loss=0.1391, over 29558.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3619, pruned_loss=0.1083, over 5693022.81 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3583, pruned_loss=0.119, over 5673530.21 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3599, pruned_loss=0.1062, over 5683481.90 frames. ], batch size: 89, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:39:33,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8435, 1.1773, 2.8497, 2.7409], device='cuda:0'), covar=tensor([0.1582, 0.2475, 0.0519, 0.1231], device='cuda:0'), in_proj_covar=tensor([0.0664, 0.0593, 0.0855, 0.0761], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:39:42,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=537307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:39:49,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.055e+02 1.108e+03 1.305e+03 1.676e+03 5.881e+03, threshold=2.611e+03, percent-clipped=5.0
+2023-03-06 11:40:05,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=537333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:40:12,122 INFO [train.py:968] (0/2) Epoch 12, batch 36150, giga_loss[loss=0.2593, simple_loss=0.3404, pruned_loss=0.08908, over 28932.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3614, pruned_loss=0.1068, over 5701967.33 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3582, pruned_loss=0.1189, over 5676816.07 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3599, pruned_loss=0.105, over 5691889.98 frames. ], batch size: 112, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:40:54,518 INFO [train.py:968] (0/2) Epoch 12, batch 36200, giga_loss[loss=0.265, simple_loss=0.3494, pruned_loss=0.09027, over 29085.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3599, pruned_loss=0.1046, over 5704590.36 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3587, pruned_loss=0.119, over 5677281.22 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3584, pruned_loss=0.103, over 5696606.48 frames. ], batch size: 155, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:41:15,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.568e+02 1.142e+03 1.432e+03 2.049e+03 5.281e+03, threshold=2.864e+03, percent-clipped=13.0
+2023-03-06 11:41:36,540 INFO [train.py:968] (0/2) Epoch 12, batch 36250, giga_loss[loss=0.2476, simple_loss=0.3372, pruned_loss=0.07899, over 28568.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.358, pruned_loss=0.1031, over 5713590.11 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3587, pruned_loss=0.119, over 5681356.45 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3568, pruned_loss=0.1014, over 5704197.94 frames. ], batch size: 65, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:41:43,662 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=537450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:41:45,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=537453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:42:02,987 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=537476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:42:06,559 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=537479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:42:09,358 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=537482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:42:16,795 INFO [train.py:968] (0/2) Epoch 12, batch 36300, giga_loss[loss=0.2772, simple_loss=0.3603, pruned_loss=0.09701, over 29003.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3594, pruned_loss=0.1057, over 5713762.92 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3588, pruned_loss=0.1191, over 5688221.41 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3583, pruned_loss=0.1038, over 5700736.70 frames. ], batch size: 164, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:42:30,462 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=537508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:42:30,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1904, 1.3609, 1.0685, 0.9662], device='cuda:0'), covar=tensor([0.0867, 0.0462, 0.1038, 0.0925], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0432, 0.0498, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:42:38,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.040e+02 1.151e+03 1.496e+03 2.003e+03 5.999e+03, threshold=2.992e+03, percent-clipped=6.0
+2023-03-06 11:42:50,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4049, 1.8518, 1.5696, 1.5849], device='cuda:0'), covar=tensor([0.0642, 0.0256, 0.0256, 0.0652], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:0')
+2023-03-06 11:42:58,294 INFO [train.py:968] (0/2) Epoch 12, batch 36350, giga_loss[loss=0.4043, simple_loss=0.4356, pruned_loss=0.1865, over 28960.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3631, pruned_loss=0.1108, over 5709691.01 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3593, pruned_loss=0.1192, over 5692908.49 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3618, pruned_loss=0.1088, over 5695763.96 frames. ], batch size: 164, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:43:13,695 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=537560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:43:24,255 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=537571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:43:41,947 INFO [train.py:968] (0/2) Epoch 12, batch 36400, giga_loss[loss=0.2616, simple_loss=0.3393, pruned_loss=0.09195, over 28687.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3633, pruned_loss=0.1125, over 5705681.27 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3595, pruned_loss=0.1194, over 5693584.95 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3621, pruned_loss=0.1108, over 5694303.11 frames. ], batch size: 242, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:44:08,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.185e+02 1.420e+03 1.846e+03 2.509e+03 5.702e+03, threshold=3.692e+03, percent-clipped=13.0
+2023-03-06 11:44:30,037 INFO [train.py:968] (0/2) Epoch 12, batch 36450, giga_loss[loss=0.3068, simple_loss=0.3604, pruned_loss=0.1266, over 26595.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.361, pruned_loss=0.1117, over 5709247.42 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3595, pruned_loss=0.1193, over 5697823.17 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3602, pruned_loss=0.1103, over 5696864.46 frames. ], batch size: 555, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:45:10,656 INFO [train.py:968] (0/2) Epoch 12, batch 36500, giga_loss[loss=0.2965, simple_loss=0.3603, pruned_loss=0.1163, over 28877.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3598, pruned_loss=0.1116, over 5710624.30 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3602, pruned_loss=0.1198, over 5701017.08 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3585, pruned_loss=0.1098, over 5698365.35 frames. ], batch size: 112, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:45:19,347 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=537703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:45:21,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=537706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:45:27,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9841, 0.9847, 3.6161, 2.9003], device='cuda:0'), covar=tensor([0.1756, 0.2822, 0.0433, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0665, 0.0596, 0.0862, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:45:30,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.491e+02 1.223e+03 1.554e+03 2.164e+03 5.354e+03, threshold=3.109e+03, percent-clipped=4.0
+2023-03-06 11:45:42,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=537735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:45:48,628 INFO [train.py:968] (0/2) Epoch 12, batch 36550, giga_loss[loss=0.2479, simple_loss=0.3368, pruned_loss=0.07951, over 28842.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3593, pruned_loss=0.1108, over 5710085.14 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.361, pruned_loss=0.1202, over 5703506.44 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3574, pruned_loss=0.1083, over 5698697.61 frames. ], batch size: 145, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:46:05,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6013, 2.2369, 1.7471, 0.7037], device='cuda:0'), covar=tensor([0.4107, 0.2105, 0.2945, 0.4902], device='cuda:0'), in_proj_covar=tensor([0.1572, 0.1498, 0.1494, 0.1285], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 11:46:36,577 INFO [train.py:968] (0/2) Epoch 12, batch 36600, giga_loss[loss=0.2665, simple_loss=0.3411, pruned_loss=0.09597, over 28612.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3576, pruned_loss=0.109, over 5695007.58 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3612, pruned_loss=0.1203, over 5705411.32 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3559, pruned_loss=0.1069, over 5684537.48 frames. ], batch size: 85, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:47:01,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.437e+02 1.051e+03 1.377e+03 2.117e+03 8.156e+03, threshold=2.753e+03, percent-clipped=9.0
+2023-03-06 11:47:24,242 INFO [train.py:968] (0/2) Epoch 12, batch 36650, giga_loss[loss=0.2472, simple_loss=0.3211, pruned_loss=0.08666, over 28519.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3521, pruned_loss=0.105, over 5695640.79 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3611, pruned_loss=0.1201, over 5708100.96 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3508, pruned_loss=0.1033, over 5684884.22 frames. ], batch size: 71, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:47:48,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5021, 4.3174, 4.0560, 2.2758], device='cuda:0'), covar=tensor([0.0448, 0.0583, 0.0604, 0.2027], device='cuda:0'), in_proj_covar=tensor([0.1039, 0.0972, 0.0845, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 11:48:09,166 INFO [train.py:968] (0/2) Epoch 12, batch 36700, giga_loss[loss=0.2351, simple_loss=0.3047, pruned_loss=0.08272, over 28615.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3462, pruned_loss=0.1019, over 5685710.33 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3612, pruned_loss=0.12, over 5713081.42 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3447, pruned_loss=0.1001, over 5672183.88 frames. ], batch size: 92, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:48:39,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.504e+02 9.069e+02 1.185e+03 1.787e+03 6.939e+03, threshold=2.370e+03, percent-clipped=12.0
+2023-03-06 11:49:03,085 INFO [train.py:968] (0/2) Epoch 12, batch 36750, giga_loss[loss=0.3021, simple_loss=0.351, pruned_loss=0.1266, over 26469.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3422, pruned_loss=0.1004, over 5652825.99 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3615, pruned_loss=0.12, over 5697967.06 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3402, pruned_loss=0.0983, over 5655768.37 frames. ], batch size: 555, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:49:06,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=537946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:49:52,680 INFO [train.py:968] (0/2) Epoch 12, batch 36800, giga_loss[loss=0.2887, simple_loss=0.3636, pruned_loss=0.1069, over 29005.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3382, pruned_loss=0.09792, over 5649439.09 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3612, pruned_loss=0.1199, over 5698812.23 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3367, pruned_loss=0.09632, over 5650890.93 frames. ], batch size: 136, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:50:00,239 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-538000.pt
+2023-03-06 11:50:15,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.450e+02 9.668e+02 1.261e+03 1.725e+03 8.855e+03, threshold=2.521e+03, percent-clipped=9.0
+2023-03-06 11:50:33,916 INFO [train.py:968] (0/2) Epoch 12, batch 36850, giga_loss[loss=0.2489, simple_loss=0.327, pruned_loss=0.08535, over 28962.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3391, pruned_loss=0.0979, over 5635964.08 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3618, pruned_loss=0.1203, over 5666161.18 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3366, pruned_loss=0.09552, over 5665093.08 frames. ], batch size: 227, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:51:11,561 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=538089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:51:13,157 INFO [train.py:968] (0/2) Epoch 12, batch 36900, giga_loss[loss=0.2956, simple_loss=0.3448, pruned_loss=0.1232, over 28715.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3393, pruned_loss=0.09775, over 5648869.19 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3622, pruned_loss=0.1205, over 5666426.35 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3365, pruned_loss=0.09525, over 5671488.83 frames. ], batch size: 92, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:51:13,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=538092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:51:39,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.918e+02 1.067e+03 1.342e+03 1.778e+03 7.046e+03, threshold=2.683e+03, percent-clipped=11.0
+2023-03-06 11:51:39,917 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=538121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:51:56,959 INFO [train.py:968] (0/2) Epoch 12, batch 36950, giga_loss[loss=0.2562, simple_loss=0.33, pruned_loss=0.09119, over 28966.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3374, pruned_loss=0.09696, over 5662724.25 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3625, pruned_loss=0.1208, over 5661297.38 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3345, pruned_loss=0.09429, over 5685347.76 frames. ], batch size: 213, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:52:37,288 INFO [train.py:968] (0/2) Epoch 12, batch 37000, libri_loss[loss=0.374, simple_loss=0.422, pruned_loss=0.163, over 20118.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3355, pruned_loss=0.09616, over 5667615.06 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3636, pruned_loss=0.1213, over 5646995.16 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3316, pruned_loss=0.09292, over 5700045.83 frames. ], batch size: 187, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:52:52,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6188, 1.7014, 1.7538, 1.5598], device='cuda:0'), covar=tensor([0.1641, 0.2027, 0.2120, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0742, 0.0681, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 11:52:58,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1931, 1.4045, 3.4389, 3.2309], device='cuda:0'), covar=tensor([0.1498, 0.2494, 0.0413, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.0668, 0.0596, 0.0861, 0.0766], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:52:59,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.854e+02 9.734e+02 1.423e+03 2.008e+03 7.114e+03, threshold=2.845e+03, percent-clipped=15.0
+2023-03-06 11:53:15,229 INFO [train.py:968] (0/2) Epoch 12, batch 37050, giga_loss[loss=0.2264, simple_loss=0.3016, pruned_loss=0.07556, over 28642.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3338, pruned_loss=0.09538, over 5680906.00 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3641, pruned_loss=0.1214, over 5656009.05 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3291, pruned_loss=0.09184, over 5699665.83 frames. ], batch size: 92, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:53:46,752 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-06 11:53:52,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4773, 1.5401, 1.1941, 1.1416], device='cuda:0'), covar=tensor([0.0890, 0.0544, 0.1083, 0.1037], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0435, 0.0503, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 11:53:52,749 INFO [train.py:968] (0/2) Epoch 12, batch 37100, giga_loss[loss=0.2811, simple_loss=0.3435, pruned_loss=0.1094, over 28885.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3327, pruned_loss=0.09493, over 5691386.10 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3649, pruned_loss=0.1216, over 5662770.44 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3274, pruned_loss=0.09121, over 5701091.96 frames. ], batch size: 199, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:54:16,675 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=538320.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 11:54:17,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.356e+02 9.776e+02 1.204e+03 1.716e+03 1.286e+04, threshold=2.409e+03, percent-clipped=10.0
+2023-03-06 11:54:33,476 INFO [train.py:968] (0/2) Epoch 12, batch 37150, giga_loss[loss=0.2322, simple_loss=0.3141, pruned_loss=0.07517, over 28903.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3302, pruned_loss=0.09354, over 5707407.54 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3651, pruned_loss=0.1214, over 5667743.82 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3252, pruned_loss=0.09023, over 5711290.38 frames. ], batch size: 174, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:55:15,001 INFO [train.py:968] (0/2) Epoch 12, batch 37200, giga_loss[loss=0.2451, simple_loss=0.3227, pruned_loss=0.08376, over 29015.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3272, pruned_loss=0.09194, over 5705424.79 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3654, pruned_loss=0.1215, over 5660729.51 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3226, pruned_loss=0.08886, over 5714438.55 frames. ], batch size: 164, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:55:38,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.532e+02 9.418e+02 1.170e+03 1.616e+03 4.411e+03, threshold=2.339e+03, percent-clipped=6.0
+2023-03-06 11:55:55,833 INFO [train.py:968] (0/2) Epoch 12, batch 37250, giga_loss[loss=0.259, simple_loss=0.3259, pruned_loss=0.096, over 28605.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3258, pruned_loss=0.09127, over 5700870.24 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3659, pruned_loss=0.1215, over 5660391.57 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3206, pruned_loss=0.08804, over 5709487.14 frames. ], batch size: 85, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:56:05,978 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=538456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:56:35,758 INFO [train.py:968] (0/2) Epoch 12, batch 37300, giga_loss[loss=0.2498, simple_loss=0.318, pruned_loss=0.09077, over 28178.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3238, pruned_loss=0.08967, over 5708526.00 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3661, pruned_loss=0.1215, over 5662653.41 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3191, pruned_loss=0.08683, over 5713676.67 frames. ], batch size: 77, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:57:00,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.484e+02 9.582e+02 1.239e+03 1.815e+03 7.620e+03, threshold=2.477e+03, percent-clipped=11.0
+2023-03-06 11:57:07,356 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=538530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:57:18,046 INFO [train.py:968] (0/2) Epoch 12, batch 37350, libri_loss[loss=0.3246, simple_loss=0.4016, pruned_loss=0.1238, over 29657.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3247, pruned_loss=0.09017, over 5708227.47 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3669, pruned_loss=0.1216, over 5670830.32 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3188, pruned_loss=0.08666, over 5706672.75 frames. ], batch size: 91, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:57:19,594 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=538544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:57:36,879 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-06 11:57:55,000 INFO [train.py:968] (0/2) Epoch 12, batch 37400, giga_loss[loss=0.2736, simple_loss=0.3496, pruned_loss=0.09874, over 28749.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3267, pruned_loss=0.09105, over 5722178.78 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3676, pruned_loss=0.1218, over 5679302.29 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3194, pruned_loss=0.08672, over 5714946.93 frames. ], batch size: 284, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:58:06,981 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 11:58:23,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.198e+02 1.123e+03 1.408e+03 2.061e+03 7.926e+03, threshold=2.817e+03, percent-clipped=18.0
+2023-03-06 11:58:40,467 INFO [train.py:968] (0/2) Epoch 12, batch 37450, giga_loss[loss=0.338, simple_loss=0.3802, pruned_loss=0.1479, over 23638.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3323, pruned_loss=0.09495, over 5719658.32 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.368, pruned_loss=0.1219, over 5685773.00 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3251, pruned_loss=0.09056, over 5709136.75 frames. ], batch size: 705, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:58:46,207 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7081, 1.7047, 1.7216, 1.5972], device='cuda:0'), covar=tensor([0.1488, 0.2153, 0.2036, 0.1834], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0742, 0.0681, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 11:59:23,534 INFO [train.py:968] (0/2) Epoch 12, batch 37500, libri_loss[loss=0.3166, simple_loss=0.3823, pruned_loss=0.1255, over 29766.00 frames. ], tot_loss[loss=0.27, simple_loss=0.34, pruned_loss=0.09998, over 5719858.29 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3689, pruned_loss=0.1226, over 5689186.61 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3313, pruned_loss=0.09433, over 5709013.95 frames. ], batch size: 87, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:59:25,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=538695.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 11:59:30,195 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 11:59:55,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.866e+02 1.189e+03 1.511e+03 2.075e+03 6.460e+03, threshold=3.022e+03, percent-clipped=8.0
+2023-03-06 12:00:12,473 INFO [train.py:968] (0/2) Epoch 12, batch 37550, giga_loss[loss=0.2895, simple_loss=0.3666, pruned_loss=0.1062, over 28884.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3485, pruned_loss=0.1059, over 5711628.07 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3695, pruned_loss=0.1229, over 5692740.97 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3407, pruned_loss=0.1007, over 5700131.63 frames. ], batch size: 227, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 12:01:02,505 INFO [train.py:968] (0/2) Epoch 12, batch 37600, giga_loss[loss=0.3542, simple_loss=0.4171, pruned_loss=0.1457, over 27964.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3528, pruned_loss=0.1079, over 5685717.69 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3694, pruned_loss=0.1229, over 5683750.13 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.346, pruned_loss=0.1034, over 5684785.69 frames. ], batch size: 412, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:01:19,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1654, 3.9608, 3.7352, 1.7520], device='cuda:0'), covar=tensor([0.0605, 0.0756, 0.0742, 0.2326], device='cuda:0'), in_proj_covar=tensor([0.1053, 0.0979, 0.0854, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 12:01:27,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.349e+02 1.217e+03 1.583e+03 2.118e+03 6.549e+03, threshold=3.165e+03, percent-clipped=15.0
+2023-03-06 12:01:36,745 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=538831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:01:43,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=538838.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 12:01:45,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=538841.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 12:01:45,981 INFO [train.py:968] (0/2) Epoch 12, batch 37650, giga_loss[loss=0.3311, simple_loss=0.3953, pruned_loss=0.1334, over 28313.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3568, pruned_loss=0.109, over 5688088.61 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.369, pruned_loss=0.1224, over 5685848.60 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3515, pruned_loss=0.1054, over 5685693.62 frames. ], batch size: 368, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:02:09,959 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=538870.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 12:02:21,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5136, 3.6407, 1.6834, 1.7140], device='cuda:0'), covar=tensor([0.0808, 0.0340, 0.0793, 0.1139], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0506, 0.0340, 0.0389], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 12:02:26,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0221, 1.1597, 3.3532, 2.8210], device='cuda:0'), covar=tensor([0.1747, 0.2811, 0.0494, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.0665, 0.0593, 0.0857, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:02:28,765 INFO [train.py:968] (0/2) Epoch 12, batch 37700, giga_loss[loss=0.349, simple_loss=0.4148, pruned_loss=0.1417, over 28598.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.363, pruned_loss=0.1128, over 5689741.53 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.369, pruned_loss=0.1225, over 5693448.88 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3584, pruned_loss=0.1095, over 5680708.69 frames. ], batch size: 336, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:02:39,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=538905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:02:50,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=538919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:02:52,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.548e+02 1.262e+03 1.857e+03 3.026e+03 1.233e+04, threshold=3.715e+03, percent-clipped=23.0
+2023-03-06 12:03:08,223 INFO [train.py:968] (0/2) Epoch 12, batch 37750, giga_loss[loss=0.3355, simple_loss=0.3765, pruned_loss=0.1473, over 23716.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3626, pruned_loss=0.1126, over 5688326.20 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3683, pruned_loss=0.1222, over 5692892.26 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3592, pruned_loss=0.1097, over 5681784.33 frames. ], batch size: 705, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:03:34,959 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=538974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:03:37,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=538977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:03:49,780 INFO [train.py:968] (0/2) Epoch 12, batch 37800, giga_loss[loss=0.2785, simple_loss=0.3523, pruned_loss=0.1024, over 28670.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3573, pruned_loss=0.1082, over 5696284.29 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3682, pruned_loss=0.1221, over 5696203.59 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3545, pruned_loss=0.1058, over 5688259.50 frames. ], batch size: 307, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:04:01,896 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7000, 1.0789, 2.8502, 2.6411], device='cuda:0'), covar=tensor([0.1746, 0.2601, 0.0536, 0.0903], device='cuda:0'), in_proj_covar=tensor([0.0666, 0.0592, 0.0857, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:04:02,415 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=539006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:04:18,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.038e+02 1.017e+03 1.263e+03 1.565e+03 3.566e+03, threshold=2.526e+03, percent-clipped=1.0
+2023-03-06 12:04:36,544 INFO [train.py:968] (0/2) Epoch 12, batch 37850, giga_loss[loss=0.2495, simple_loss=0.3322, pruned_loss=0.08346, over 28878.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3537, pruned_loss=0.1048, over 5697332.34 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3682, pruned_loss=0.1221, over 5696203.59 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3516, pruned_loss=0.1029, over 5691086.54 frames. ], batch size: 106, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:04:42,639 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=539048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:04:44,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=539051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:04:45,128 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2698, 1.8376, 1.4247, 0.3980], device='cuda:0'), covar=tensor([0.3477, 0.2151, 0.3460, 0.4787], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1485, 0.1494, 0.1278], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 12:04:54,034 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=539062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:04:56,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=539065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:05:08,627 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=539080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:05:18,405 INFO [train.py:968] (0/2) Epoch 12, batch 37900, giga_loss[loss=0.2442, simple_loss=0.3236, pruned_loss=0.08238, over 28517.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.354, pruned_loss=0.1048, over 5706984.14 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3687, pruned_loss=0.1224, over 5703495.75 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3512, pruned_loss=0.1023, over 5695221.00 frames. ], batch size: 60, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:05:19,952 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=539094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:05:31,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=539109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:05:40,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.806e+02 1.145e+03 1.572e+03 2.161e+03 6.725e+03, threshold=3.144e+03, percent-clipped=18.0
+2023-03-06 12:05:58,601 INFO [train.py:968] (0/2) Epoch 12, batch 37950, giga_loss[loss=0.3159, simple_loss=0.3879, pruned_loss=0.122, over 28914.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3569, pruned_loss=0.1067, over 5700635.70 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3694, pruned_loss=0.1231, over 5701578.31 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3535, pruned_loss=0.1033, over 5693051.23 frames. ], batch size: 199, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:06:41,210 INFO [train.py:968] (0/2) Epoch 12, batch 38000, giga_loss[loss=0.3365, simple_loss=0.3929, pruned_loss=0.14, over 28567.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3598, pruned_loss=0.1089, over 5708487.29 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3692, pruned_loss=0.1231, over 5705466.41 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.357, pruned_loss=0.1059, over 5699223.08 frames. ], batch size: 336, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:06:43,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5160, 2.1085, 1.6585, 0.7218], device='cuda:0'), covar=tensor([0.4634, 0.2146, 0.2983, 0.4752], device='cuda:0'), in_proj_covar=tensor([0.1567, 0.1491, 0.1498, 0.1283], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 12:07:04,233 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-06 12:07:12,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.009e+02 1.288e+03 1.653e+03 2.314e+03 4.802e+03, threshold=3.306e+03, percent-clipped=7.0
+2023-03-06 12:07:19,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=539232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:07:29,011 INFO [train.py:968] (0/2) Epoch 12, batch 38050, giga_loss[loss=0.3389, simple_loss=0.387, pruned_loss=0.1454, over 26617.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3623, pruned_loss=0.1112, over 5704732.26 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3687, pruned_loss=0.1228, over 5708414.88 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3604, pruned_loss=0.1091, over 5694789.99 frames. ], batch size: 555, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:07:38,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2960, 1.5489, 1.1601, 1.2853], device='cuda:0'), covar=tensor([0.2382, 0.2353, 0.2677, 0.2137], device='cuda:0'), in_proj_covar=tensor([0.1326, 0.0978, 0.1167, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 12:07:43,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2204, 3.8723, 1.5989, 1.3868], device='cuda:0'), covar=tensor([0.1056, 0.0308, 0.0863, 0.1426], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0505, 0.0343, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 12:08:11,470 INFO [train.py:968] (0/2) Epoch 12, batch 38100, giga_loss[loss=0.2713, simple_loss=0.3488, pruned_loss=0.09696, over 28912.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3618, pruned_loss=0.1111, over 5696503.83 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3689, pruned_loss=0.1228, over 5703072.07 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3598, pruned_loss=0.1089, over 5693066.95 frames. ], batch size: 145, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:08:39,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.488e+02 1.112e+03 1.458e+03 1.944e+03 5.419e+03, threshold=2.916e+03, percent-clipped=5.0
+2023-03-06 12:08:41,116 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-06 12:08:54,444 INFO [train.py:968] (0/2) Epoch 12, batch 38150, giga_loss[loss=0.276, simple_loss=0.3547, pruned_loss=0.09863, over 28804.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3622, pruned_loss=0.1114, over 5700415.20 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3692, pruned_loss=0.1228, over 5704208.28 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3604, pruned_loss=0.1096, over 5696693.73 frames. ], batch size: 119, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:09:39,188 INFO [train.py:968] (0/2) Epoch 12, batch 38200, giga_loss[loss=0.3115, simple_loss=0.3851, pruned_loss=0.1189, over 28973.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3622, pruned_loss=0.1114, over 5697447.23 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3689, pruned_loss=0.1226, over 5708148.65 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3608, pruned_loss=0.1098, over 5690939.55 frames. ], batch size: 164, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:10:02,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.391e+02 1.121e+03 1.385e+03 1.881e+03 4.471e+03, threshold=2.770e+03, percent-clipped=9.0
+2023-03-06 12:10:14,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9904, 2.1817, 1.4859, 1.7299], device='cuda:0'), covar=tensor([0.0812, 0.0659, 0.1010, 0.1062], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0432, 0.0499, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:10:16,307 INFO [train.py:968] (0/2) Epoch 12, batch 38250, giga_loss[loss=0.287, simple_loss=0.3667, pruned_loss=0.1036, over 28941.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3624, pruned_loss=0.1101, over 5705601.13 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3691, pruned_loss=0.1225, over 5708861.29 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.361, pruned_loss=0.1087, over 5699508.57 frames. ], batch size: 213, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:10:32,168 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-06 12:10:52,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=539484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:10:58,775 INFO [train.py:968] (0/2) Epoch 12, batch 38300, giga_loss[loss=0.266, simple_loss=0.3488, pruned_loss=0.0916, over 28923.00 frames. ], tot_loss[loss=0.29, simple_loss=0.362, pruned_loss=0.109, over 5710493.60 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3692, pruned_loss=0.1225, over 5713754.16 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3606, pruned_loss=0.1075, over 5701052.72 frames. ], batch size: 145, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:11:23,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.156e+02 1.016e+03 1.364e+03 1.862e+03 7.122e+03, threshold=2.727e+03, percent-clipped=6.0
+2023-03-06 12:11:39,440 INFO [train.py:968] (0/2) Epoch 12, batch 38350, giga_loss[loss=0.289, simple_loss=0.3602, pruned_loss=0.1089, over 28882.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3608, pruned_loss=0.1084, over 5705043.97 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3695, pruned_loss=0.1226, over 5713710.55 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3592, pruned_loss=0.1068, over 5697566.68 frames. ], batch size: 186, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:12:21,312 INFO [train.py:968] (0/2) Epoch 12, batch 38400, giga_loss[loss=0.2657, simple_loss=0.3429, pruned_loss=0.0942, over 28623.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3589, pruned_loss=0.1076, over 5708067.10 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3696, pruned_loss=0.1226, over 5718301.54 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3574, pruned_loss=0.1059, over 5697919.38 frames. ], batch size: 307, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:12:31,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=539607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:12:45,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.706e+02 9.721e+02 1.255e+03 1.636e+03 1.339e+04, threshold=2.510e+03, percent-clipped=3.0
+2023-03-06 12:12:48,408 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=539627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:12:51,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=539630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:12:58,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4626, 1.5808, 1.6002, 1.6728], device='cuda:0'), covar=tensor([0.0783, 0.0314, 0.0285, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0112, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:0')
+2023-03-06 12:13:01,055 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5334, 1.6121, 1.1545, 1.1876], device='cuda:0'), covar=tensor([0.0832, 0.0593, 0.1085, 0.1176], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0433, 0.0502, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:13:02,176 INFO [train.py:968] (0/2) Epoch 12, batch 38450, giga_loss[loss=0.2438, simple_loss=0.329, pruned_loss=0.07934, over 29066.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3574, pruned_loss=0.1067, over 5701958.19 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3696, pruned_loss=0.1227, over 5711323.65 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3558, pruned_loss=0.1051, over 5700515.05 frames. ], batch size: 136, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:13:05,330 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 12:13:15,628 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=539659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:13:17,673 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=539662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:13:19,827 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 12:13:40,895 INFO [train.py:968] (0/2) Epoch 12, batch 38500, giga_loss[loss=0.3006, simple_loss=0.3702, pruned_loss=0.1155, over 28803.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3562, pruned_loss=0.1062, over 5708875.08 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3694, pruned_loss=0.1224, over 5715123.89 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3548, pruned_loss=0.1046, over 5703995.77 frames. ], batch size: 199, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:14:05,452 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.100e+02 1.016e+03 1.425e+03 1.918e+03 7.057e+03, threshold=2.851e+03, percent-clipped=17.0
+2023-03-06 12:14:20,063 INFO [train.py:968] (0/2) Epoch 12, batch 38550, giga_loss[loss=0.274, simple_loss=0.3463, pruned_loss=0.1008, over 28873.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3571, pruned_loss=0.1068, over 5702646.35 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3696, pruned_loss=0.1225, over 5709270.83 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3554, pruned_loss=0.105, over 5704281.52 frames. ], batch size: 112, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:14:27,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=539750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:14:27,429 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 12:14:30,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=539753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:14:52,755 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=539782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:15:00,111 INFO [train.py:968] (0/2) Epoch 12, batch 38600, giga_loss[loss=0.2819, simple_loss=0.3523, pruned_loss=0.1058, over 28917.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3569, pruned_loss=0.1065, over 5705305.23 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3695, pruned_loss=0.1225, over 5711773.54 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3554, pruned_loss=0.1048, over 5704080.42 frames. ], batch size: 112, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:15:10,239 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-06 12:15:11,258 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=539807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:15:24,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.741e+02 9.715e+02 1.184e+03 1.481e+03 4.068e+03, threshold=2.368e+03, percent-clipped=2.0
+2023-03-06 12:15:34,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3939, 3.3002, 1.5095, 1.5277], device='cuda:0'), covar=tensor([0.0970, 0.0239, 0.0952, 0.1369], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0498, 0.0340, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 12:15:38,991 INFO [train.py:968] (0/2) Epoch 12, batch 38650, libri_loss[loss=0.3466, simple_loss=0.4132, pruned_loss=0.14, over 29209.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3564, pruned_loss=0.1054, over 5712513.69 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3699, pruned_loss=0.1226, over 5717390.72 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3544, pruned_loss=0.1034, over 5706103.63 frames. ], batch size: 97, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:16:18,851 INFO [train.py:968] (0/2) Epoch 12, batch 38700, giga_loss[loss=0.2554, simple_loss=0.339, pruned_loss=0.08595, over 28621.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3553, pruned_loss=0.1042, over 5716437.33 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3701, pruned_loss=0.1227, over 5719288.30 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3534, pruned_loss=0.1023, over 5709556.91 frames. ], batch size: 71, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:16:27,292 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5684, 2.5810, 1.9424, 2.0905], device='cuda:0'), covar=tensor([0.0733, 0.0628, 0.0919, 0.1003], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0435, 0.0504, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 12:16:45,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.454e+02 1.016e+03 1.304e+03 1.740e+03 1.243e+04, threshold=2.608e+03, percent-clipped=11.0
+2023-03-06 12:16:47,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4397, 3.4582, 1.5884, 1.5838], device='cuda:0'), covar=tensor([0.0943, 0.0241, 0.0862, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0497, 0.0338, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-06 12:16:58,465 INFO [train.py:968] (0/2) Epoch 12, batch 38750, libri_loss[loss=0.3142, simple_loss=0.3704, pruned_loss=0.129, over 29581.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3541, pruned_loss=0.1039, over 5713839.95 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3703, pruned_loss=0.1229, over 5722772.43 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3521, pruned_loss=0.1018, over 5704921.45 frames. ], batch size: 76, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:17:25,689 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4237, 1.6246, 1.6935, 1.2817], device='cuda:0'), covar=tensor([0.1578, 0.2366, 0.1316, 0.1599], device='cuda:0'), in_proj_covar=tensor([0.0846, 0.0696, 0.0885, 0.0788], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 12:17:37,868 INFO [train.py:968] (0/2) Epoch 12, batch 38800, giga_loss[loss=0.2559, simple_loss=0.3257, pruned_loss=0.093, over 28827.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3523, pruned_loss=0.1037, over 5700282.81 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3708, pruned_loss=0.1232, over 5715903.15 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.35, pruned_loss=0.1014, over 5699772.83 frames. ], batch size: 99, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:17:44,832 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-540000.pt
+2023-03-06 12:18:02,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.685e+02 1.006e+03 1.204e+03 1.693e+03 6.415e+03, threshold=2.407e+03, percent-clipped=9.0
+2023-03-06 12:18:12,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=540037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:18:15,535 INFO [train.py:968] (0/2) Epoch 12, batch 38850, giga_loss[loss=0.3092, simple_loss=0.3706, pruned_loss=0.1239, over 28686.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3491, pruned_loss=0.1021, over 5707916.51 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3709, pruned_loss=0.1231, over 5722323.06 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3463, pruned_loss=0.09963, over 5701007.38 frames. ], batch size: 242, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:18:43,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:18:49,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8027, 4.6023, 4.4426, 1.9951], device='cuda:0'), covar=tensor([0.0567, 0.0732, 0.0871, 0.1865], device='cuda:0'), in_proj_covar=tensor([0.1052, 0.0981, 0.0851, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 12:18:53,293 INFO [train.py:968] (0/2) Epoch 12, batch 38900, giga_loss[loss=0.2485, simple_loss=0.3206, pruned_loss=0.08823, over 28839.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3488, pruned_loss=0.1023, over 5714590.49 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3713, pruned_loss=0.1234, over 5726040.51 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3452, pruned_loss=0.0991, over 5705297.64 frames. ], batch size: 99, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:18:59,728 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:19:24,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.568e+02 1.170e+03 1.564e+03 2.304e+03 6.098e+03, threshold=3.128e+03, percent-clipped=22.0
+2023-03-06 12:19:36,784 INFO [train.py:968] (0/2) Epoch 12, batch 38950, giga_loss[loss=0.2384, simple_loss=0.3172, pruned_loss=0.07978, over 28407.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3473, pruned_loss=0.1019, over 5717624.67 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3707, pruned_loss=0.123, over 5728840.47 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3447, pruned_loss=0.09948, over 5707484.47 frames. ], batch size: 65, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:19:41,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4675, 3.2812, 1.4791, 1.6621], device='cuda:0'), covar=tensor([0.0877, 0.0345, 0.0849, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0499, 0.0338, 0.0385], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:0')
+2023-03-06 12:20:01,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5681, 1.6506, 1.8067, 1.3604], device='cuda:0'), covar=tensor([0.1717, 0.2196, 0.1351, 0.1538], device='cuda:0'), in_proj_covar=tensor([0.0842, 0.0692, 0.0882, 0.0786], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 12:20:06,008 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=540180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:20:07,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=540182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:20:09,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=540183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:20:15,669 INFO [train.py:968] (0/2) Epoch 12, batch 39000, giga_loss[loss=0.273, simple_loss=0.3463, pruned_loss=0.0998, over 27939.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3454, pruned_loss=0.1012, over 5719322.58 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3711, pruned_loss=0.1233, over 5729591.20 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3426, pruned_loss=0.09872, over 5710598.50 frames. ], batch size: 412, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:20:15,674 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 12:20:24,144 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2153, simple_loss=0.3223, pruned_loss=0.05414, over 944034.00 frames. 
+2023-03-06 12:20:24,144 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 12:20:39,514 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=540212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:20:40,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8762, 1.8742, 1.4301, 1.5402], device='cuda:0'), covar=tensor([0.0739, 0.0650, 0.1013, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0434, 0.0504, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:20:49,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.377e+02 1.052e+03 1.223e+03 1.610e+03 5.417e+03, threshold=2.447e+03, percent-clipped=5.0
+2023-03-06 12:20:57,996 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:21:01,661 INFO [train.py:968] (0/2) Epoch 12, batch 39050, giga_loss[loss=0.2765, simple_loss=0.3443, pruned_loss=0.1044, over 28890.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3431, pruned_loss=0.1003, over 5703757.92 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.371, pruned_loss=0.1232, over 5717883.39 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3405, pruned_loss=0.09807, over 5707660.13 frames. ], batch size: 213, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:21:28,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5307, 2.1541, 1.6377, 0.8729], device='cuda:0'), covar=tensor([0.4683, 0.2512, 0.3357, 0.4696], device='cuda:0'), in_proj_covar=tensor([0.1573, 0.1483, 0.1496, 0.1282], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 12:21:43,916 INFO [train.py:968] (0/2) Epoch 12, batch 39100, giga_loss[loss=0.2265, simple_loss=0.3109, pruned_loss=0.07102, over 29081.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3396, pruned_loss=0.09857, over 5707651.65 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3711, pruned_loss=0.1232, over 5719707.10 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3372, pruned_loss=0.09656, over 5708807.31 frames. ], batch size: 155, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:22:12,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.885e+02 1.016e+03 1.331e+03 1.705e+03 5.542e+03, threshold=2.662e+03, percent-clipped=7.0
+2023-03-06 12:22:13,713 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=540325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:22:16,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=540328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:22:28,382 INFO [train.py:968] (0/2) Epoch 12, batch 39150, giga_loss[loss=0.3003, simple_loss=0.3699, pruned_loss=0.1153, over 27611.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3388, pruned_loss=0.09839, over 5704008.86 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3712, pruned_loss=0.1232, over 5719770.60 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3366, pruned_loss=0.09655, over 5704772.82 frames. ], batch size: 472, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:22:42,238 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=540357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:22:43,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1438, 1.3381, 3.6635, 3.0147], device='cuda:0'), covar=tensor([0.1597, 0.2480, 0.0392, 0.0931], device='cuda:0'), in_proj_covar=tensor([0.0659, 0.0589, 0.0854, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:23:10,770 INFO [train.py:968] (0/2) Epoch 12, batch 39200, giga_loss[loss=0.2717, simple_loss=0.3441, pruned_loss=0.09967, over 28791.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3406, pruned_loss=0.09904, over 5709510.95 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3708, pruned_loss=0.123, over 5723499.66 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3386, pruned_loss=0.09735, over 5706329.33 frames. ], batch size: 119, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:23:17,235 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-06 12:23:38,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.252e+02 1.075e+03 1.289e+03 1.804e+03 5.320e+03, threshold=2.577e+03, percent-clipped=9.0
+2023-03-06 12:23:56,729 INFO [train.py:968] (0/2) Epoch 12, batch 39250, libri_loss[loss=0.3204, simple_loss=0.3816, pruned_loss=0.1296, over 29520.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3439, pruned_loss=0.1003, over 5715355.40 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3709, pruned_loss=0.123, over 5728865.20 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3413, pruned_loss=0.09814, over 5707466.46 frames. ], batch size: 81, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:24:05,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=540452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:24:25,965 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=540474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:24:38,934 INFO [train.py:968] (0/2) Epoch 12, batch 39300, libri_loss[loss=0.3042, simple_loss=0.3765, pruned_loss=0.116, over 29662.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3465, pruned_loss=0.1013, over 5701411.90 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3708, pruned_loss=0.1229, over 5721257.43 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3438, pruned_loss=0.09904, over 5701013.73 frames. ], batch size: 91, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:24:46,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2169, 1.8429, 1.4247, 0.4493], device='cuda:0'), covar=tensor([0.3898, 0.1992, 0.3111, 0.4748], device='cuda:0'), in_proj_covar=tensor([0.1573, 0.1484, 0.1501, 0.1288], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 12:25:08,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.705e+02 9.875e+02 1.289e+03 1.666e+03 4.967e+03, threshold=2.578e+03, percent-clipped=7.0
+2023-03-06 12:25:13,109 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3499, 2.9173, 1.5331, 1.4968], device='cuda:0'), covar=tensor([0.0803, 0.0313, 0.0783, 0.1149], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0504, 0.0341, 0.0386], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 12:25:20,114 INFO [train.py:968] (0/2) Epoch 12, batch 39350, giga_loss[loss=0.2396, simple_loss=0.3258, pruned_loss=0.07669, over 28689.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3471, pruned_loss=0.1008, over 5699577.87 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3707, pruned_loss=0.1229, over 5721609.35 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3442, pruned_loss=0.09825, over 5698410.86 frames. ], batch size: 336, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:26:02,463 INFO [train.py:968] (0/2) Epoch 12, batch 39400, giga_loss[loss=0.2536, simple_loss=0.3328, pruned_loss=0.0872, over 28902.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3467, pruned_loss=0.1003, over 5681699.77 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3707, pruned_loss=0.1229, over 5712655.37 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3438, pruned_loss=0.09766, over 5687269.61 frames. ], batch size: 199, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:26:05,645 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=540595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:07,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=540598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:16,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=540611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:21,822 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=540617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:24,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=540620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:27,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.524e+02 1.175e+03 1.543e+03 2.117e+03 4.824e+03, threshold=3.086e+03, percent-clipped=12.0
+2023-03-06 12:26:30,155 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=540627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:40,413 INFO [train.py:968] (0/2) Epoch 12, batch 39450, giga_loss[loss=0.2377, simple_loss=0.319, pruned_loss=0.0782, over 28673.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3476, pruned_loss=0.1005, over 5688614.50 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3708, pruned_loss=0.1231, over 5709105.24 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3443, pruned_loss=0.09754, over 5695897.08 frames. ], batch size: 78, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:26:40,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3316, 2.0214, 1.5568, 0.5070], device='cuda:0'), covar=tensor([0.3438, 0.1783, 0.2989, 0.4278], device='cuda:0'), in_proj_covar=tensor([0.1563, 0.1481, 0.1496, 0.1280], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 12:26:49,048 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=540649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:55,802 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-06 12:27:20,963 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0589, 1.8903, 1.5426, 1.5284], device='cuda:0'), covar=tensor([0.0643, 0.0527, 0.0852, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0435, 0.0500, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:27:21,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6486, 1.2703, 5.2685, 3.5095], device='cuda:0'), covar=tensor([0.1603, 0.2650, 0.0352, 0.0762], device='cuda:0'), in_proj_covar=tensor([0.0660, 0.0590, 0.0857, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:27:22,838 INFO [train.py:968] (0/2) Epoch 12, batch 39500, giga_loss[loss=0.2666, simple_loss=0.3326, pruned_loss=0.1003, over 28524.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3494, pruned_loss=0.1026, over 5684009.79 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3706, pruned_loss=0.1229, over 5706332.55 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3462, pruned_loss=0.09962, over 5691213.14 frames. ], batch size: 78, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:27:44,985 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:27:53,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.243e+02 1.188e+03 1.585e+03 2.237e+03 6.342e+03, threshold=3.169e+03, percent-clipped=10.0
+2023-03-06 12:28:02,478 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:28:05,217 INFO [train.py:968] (0/2) Epoch 12, batch 39550, giga_loss[loss=0.2833, simple_loss=0.3607, pruned_loss=0.103, over 28888.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3506, pruned_loss=0.1035, over 5688515.32 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3709, pruned_loss=0.123, over 5712115.81 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3472, pruned_loss=0.1003, over 5688435.46 frames. ], batch size: 186, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:28:14,807 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=540754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:28:16,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=540757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:28:25,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4258, 1.6669, 1.3399, 1.4139], device='cuda:0'), covar=tensor([0.2306, 0.2276, 0.2622, 0.2197], device='cuda:0'), in_proj_covar=tensor([0.1319, 0.0969, 0.1157, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 12:28:43,759 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=540786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:28:49,268 INFO [train.py:968] (0/2) Epoch 12, batch 39600, giga_loss[loss=0.2919, simple_loss=0.3584, pruned_loss=0.1127, over 28955.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3533, pruned_loss=0.1047, over 5694060.30 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.371, pruned_loss=0.1231, over 5707139.55 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3503, pruned_loss=0.1021, over 5697780.35 frames. ], batch size: 106, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:28:56,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1863, 1.3375, 1.4629, 1.1551], device='cuda:0'), covar=tensor([0.1446, 0.1626, 0.1833, 0.1754], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0729, 0.0672, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 12:29:16,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.101e+02 1.187e+03 1.445e+03 2.083e+03 6.792e+03, threshold=2.890e+03, percent-clipped=4.0
+2023-03-06 12:29:16,425 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:29:28,519 INFO [train.py:968] (0/2) Epoch 12, batch 39650, giga_loss[loss=0.3044, simple_loss=0.3775, pruned_loss=0.1156, over 28753.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3561, pruned_loss=0.106, over 5706795.52 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3711, pruned_loss=0.123, over 5715078.94 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3531, pruned_loss=0.1033, over 5702059.67 frames. ], batch size: 262, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:30:09,646 INFO [train.py:968] (0/2) Epoch 12, batch 39700, giga_loss[loss=0.2976, simple_loss=0.3731, pruned_loss=0.1111, over 28964.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3575, pruned_loss=0.1065, over 5711752.74 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3712, pruned_loss=0.1232, over 5715561.80 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3546, pruned_loss=0.1038, over 5707553.74 frames. ], batch size: 164, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:30:13,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6792, 2.3159, 1.6788, 0.7208], device='cuda:0'), covar=tensor([0.4638, 0.2165, 0.3464, 0.5119], device='cuda:0'), in_proj_covar=tensor([0.1576, 0.1492, 0.1507, 0.1290], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 12:30:30,824 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-06 12:30:40,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.326e+02 1.169e+03 1.477e+03 1.971e+03 7.670e+03, threshold=2.954e+03, percent-clipped=13.0
+2023-03-06 12:30:49,579 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540937.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 12:30:52,994 INFO [train.py:968] (0/2) Epoch 12, batch 39750, giga_loss[loss=0.2688, simple_loss=0.3486, pruned_loss=0.09445, over 28958.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3569, pruned_loss=0.1058, over 5717081.30 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3708, pruned_loss=0.1229, over 5717409.57 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3548, pruned_loss=0.1038, over 5712155.68 frames. ], batch size: 136, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:31:02,798 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7729, 2.1569, 1.5934, 2.2780], device='cuda:0'), covar=tensor([0.2150, 0.2129, 0.2540, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.1321, 0.0972, 0.1160, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 12:31:24,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5699, 1.6730, 1.3774, 1.6563], device='cuda:0'), covar=tensor([0.2255, 0.2353, 0.2626, 0.2308], device='cuda:0'), in_proj_covar=tensor([0.1320, 0.0972, 0.1158, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 12:31:31,681 INFO [train.py:968] (0/2) Epoch 12, batch 39800, giga_loss[loss=0.2618, simple_loss=0.3469, pruned_loss=0.08835, over 28976.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3576, pruned_loss=0.1068, over 5719336.43 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3709, pruned_loss=0.1229, over 5720000.71 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3558, pruned_loss=0.1049, over 5713208.49 frames. ], batch size: 164, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:31:59,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.762e+02 1.088e+03 1.366e+03 1.845e+03 4.049e+03, threshold=2.733e+03, percent-clipped=5.0
+2023-03-06 12:32:11,491 INFO [train.py:968] (0/2) Epoch 12, batch 39850, libri_loss[loss=0.332, simple_loss=0.3928, pruned_loss=0.1356, over 29373.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3574, pruned_loss=0.1072, over 5711332.70 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3719, pruned_loss=0.1237, over 5715374.05 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3545, pruned_loss=0.1045, over 5710110.01 frames. ], batch size: 92, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:32:22,179 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1572, 2.1651, 1.6223, 1.6993], device='cuda:0'), covar=tensor([0.0727, 0.0674, 0.0918, 0.1110], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0433, 0.0498, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:32:50,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=541091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:32:51,111 INFO [train.py:968] (0/2) Epoch 12, batch 39900, giga_loss[loss=0.2617, simple_loss=0.3286, pruned_loss=0.09743, over 28493.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3545, pruned_loss=0.1057, over 5710254.11 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.372, pruned_loss=0.1238, over 5717518.29 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3519, pruned_loss=0.1032, over 5707318.04 frames. ], batch size: 71, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:33:02,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4589, 1.9823, 1.4727, 0.7147], device='cuda:0'), covar=tensor([0.4315, 0.2182, 0.3064, 0.4730], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1479, 0.1494, 0.1280], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 12:33:07,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1682, 1.5238, 1.2109, 1.0913], device='cuda:0'), covar=tensor([0.2299, 0.2266, 0.2531, 0.2067], device='cuda:0'), in_proj_covar=tensor([0.1324, 0.0976, 0.1161, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 12:33:09,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=541114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:33:20,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.282e+02 1.085e+03 1.522e+03 1.944e+03 4.524e+03, threshold=3.044e+03, percent-clipped=14.0
+2023-03-06 12:33:34,110 INFO [train.py:968] (0/2) Epoch 12, batch 39950, giga_loss[loss=0.293, simple_loss=0.3491, pruned_loss=0.1184, over 23928.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3511, pruned_loss=0.104, over 5716008.35 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3727, pruned_loss=0.1244, over 5721367.80 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3479, pruned_loss=0.101, over 5710249.24 frames. ], batch size: 705, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:33:37,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8192, 1.6879, 1.4106, 1.3153], device='cuda:0'), covar=tensor([0.0752, 0.0660, 0.0943, 0.1155], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0434, 0.0499, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:33:51,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1688, 1.2970, 3.2121, 2.8407], device='cuda:0'), covar=tensor([0.1436, 0.2452, 0.0461, 0.1981], device='cuda:0'), in_proj_covar=tensor([0.0665, 0.0594, 0.0861, 0.0774], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:34:11,281 INFO [train.py:968] (0/2) Epoch 12, batch 40000, giga_loss[loss=0.2446, simple_loss=0.3358, pruned_loss=0.07673, over 28929.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3511, pruned_loss=0.1037, over 5700760.14 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3734, pruned_loss=0.125, over 5704068.73 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3472, pruned_loss=0.1002, over 5710893.75 frames. ], batch size: 145, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:34:18,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=541201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:34:38,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.691e+02 1.061e+03 1.297e+03 1.883e+03 7.530e+03, threshold=2.594e+03, percent-clipped=10.0
+2023-03-06 12:34:45,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=541234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:34:49,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=541237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:34:54,657 INFO [train.py:968] (0/2) Epoch 12, batch 40050, giga_loss[loss=0.2809, simple_loss=0.3444, pruned_loss=0.1088, over 24007.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3524, pruned_loss=0.1032, over 5695274.37 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3734, pruned_loss=0.125, over 5706907.67 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.349, pruned_loss=0.1001, over 5700823.65 frames. ], batch size: 705, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:35:06,551 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=541257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:35:09,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=541260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:35:16,259 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=541266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:35:34,485 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=541289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:35:37,490 INFO [train.py:968] (0/2) Epoch 12, batch 40100, giga_loss[loss=0.2393, simple_loss=0.3118, pruned_loss=0.08334, over 28581.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3523, pruned_loss=0.1029, over 5702999.58 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3731, pruned_loss=0.1248, over 5709032.04 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3498, pruned_loss=0.1004, over 5705283.92 frames. ], batch size: 71, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:35:51,840 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=541312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:35:51,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=541312.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 12:36:05,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.910e+02 1.124e+03 1.555e+03 2.048e+03 5.100e+03, threshold=3.110e+03, percent-clipped=12.0
+2023-03-06 12:36:16,872 INFO [train.py:968] (0/2) Epoch 12, batch 40150, giga_loss[loss=0.2602, simple_loss=0.3375, pruned_loss=0.09149, over 28727.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3508, pruned_loss=0.1031, over 5708844.71 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3733, pruned_loss=0.1249, over 5712949.46 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3482, pruned_loss=0.1007, over 5707292.30 frames. ], batch size: 262, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:36:20,723 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=541344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:36:22,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=541347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:36:43,455 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=541376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:36:57,284 INFO [train.py:968] (0/2) Epoch 12, batch 40200, giga_loss[loss=0.2866, simple_loss=0.3583, pruned_loss=0.1074, over 28728.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3493, pruned_loss=0.1035, over 5709094.33 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3736, pruned_loss=0.1251, over 5703946.12 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3464, pruned_loss=0.1008, over 5715811.84 frames. ], batch size: 242, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:37:29,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.856e+02 1.174e+03 1.475e+03 2.080e+03 5.545e+03, threshold=2.950e+03, percent-clipped=5.0
+2023-03-06 12:37:39,667 INFO [train.py:968] (0/2) Epoch 12, batch 40250, libri_loss[loss=0.3059, simple_loss=0.3557, pruned_loss=0.1281, over 29629.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3483, pruned_loss=0.1045, over 5709011.87 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3736, pruned_loss=0.1251, over 5710909.57 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3448, pruned_loss=0.1014, over 5707906.37 frames. ], batch size: 69, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 12:37:49,422 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=541455.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 12:37:51,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=541458.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 12:37:52,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3551, 1.3782, 1.1348, 1.5382], device='cuda:0'), covar=tensor([0.0714, 0.0323, 0.0329, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 12:38:14,952 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-06 12:38:16,888 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=541487.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 12:38:21,391 INFO [train.py:968] (0/2) Epoch 12, batch 40300, giga_loss[loss=0.2817, simple_loss=0.3576, pruned_loss=0.1029, over 28672.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3468, pruned_loss=0.1042, over 5708038.55 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3733, pruned_loss=0.1249, over 5715388.31 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3437, pruned_loss=0.1015, over 5703467.50 frames. ], batch size: 336, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 12:38:33,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4623, 2.7992, 1.5549, 1.5514], device='cuda:0'), covar=tensor([0.0731, 0.0298, 0.0755, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0507, 0.0342, 0.0387], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 12:38:51,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.770e+02 1.152e+03 1.577e+03 2.152e+03 1.004e+04, threshold=3.154e+03, percent-clipped=9.0
+2023-03-06 12:39:00,436 INFO [train.py:968] (0/2) Epoch 12, batch 40350, giga_loss[loss=0.295, simple_loss=0.3616, pruned_loss=0.1142, over 27588.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3454, pruned_loss=0.1036, over 5698623.55 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3733, pruned_loss=0.125, over 5707958.67 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3422, pruned_loss=0.1008, over 5701683.53 frames. ], batch size: 472, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 12:39:38,671 INFO [train.py:968] (0/2) Epoch 12, batch 40400, giga_loss[loss=0.2065, simple_loss=0.2892, pruned_loss=0.06189, over 28991.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.342, pruned_loss=0.1017, over 5698516.99 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3734, pruned_loss=0.1249, over 5700467.36 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3384, pruned_loss=0.09871, over 5707604.22 frames. ], batch size: 164, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:39:58,355 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=541613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:40:11,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.040e+02 1.231e+03 1.677e+03 2.599e+03 1.156e+04, threshold=3.354e+03, percent-clipped=14.0
+2023-03-06 12:40:21,647 INFO [train.py:968] (0/2) Epoch 12, batch 40450, giga_loss[loss=0.2779, simple_loss=0.3366, pruned_loss=0.1096, over 28859.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3374, pruned_loss=0.09938, over 5702597.39 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3733, pruned_loss=0.1249, over 5704497.75 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.334, pruned_loss=0.09663, over 5706107.91 frames. ], batch size: 186, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:40:49,377 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-06 12:40:55,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=541687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:40:58,363 INFO [train.py:968] (0/2) Epoch 12, batch 40500, giga_loss[loss=0.2515, simple_loss=0.3284, pruned_loss=0.08731, over 28979.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3367, pruned_loss=0.09855, over 5710634.05 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3727, pruned_loss=0.1245, over 5708703.82 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3328, pruned_loss=0.09546, over 5709856.37 frames. ], batch size: 106, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:41:08,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2794, 4.1241, 3.8976, 1.8272], device='cuda:0'), covar=tensor([0.0579, 0.0716, 0.0735, 0.2072], device='cuda:0'), in_proj_covar=tensor([0.1068, 0.0987, 0.0862, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 12:41:29,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.234e+02 1.109e+03 1.424e+03 2.089e+03 4.932e+03, threshold=2.847e+03, percent-clipped=6.0
+2023-03-06 12:41:40,010 INFO [train.py:968] (0/2) Epoch 12, batch 40550, giga_loss[loss=0.2859, simple_loss=0.3568, pruned_loss=0.1075, over 28514.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3379, pruned_loss=0.09869, over 5704607.73 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3726, pruned_loss=0.1244, over 5702375.27 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3341, pruned_loss=0.0957, over 5709834.66 frames. ], batch size: 85, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:42:08,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7347, 2.1499, 2.0447, 1.7610], device='cuda:0'), covar=tensor([0.0684, 0.0244, 0.0262, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 12:42:08,777 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-06 12:42:21,734 INFO [train.py:968] (0/2) Epoch 12, batch 40600, giga_loss[loss=0.3488, simple_loss=0.3985, pruned_loss=0.1496, over 26768.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3433, pruned_loss=0.1011, over 5694829.11 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3728, pruned_loss=0.1245, over 5692123.29 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3396, pruned_loss=0.09828, over 5707489.21 frames. ], batch size: 555, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:42:24,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9677, 5.7829, 5.4941, 2.7182], device='cuda:0'), covar=tensor([0.0358, 0.0527, 0.0615, 0.1757], device='cuda:0'), in_proj_covar=tensor([0.1071, 0.0986, 0.0862, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 12:42:48,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2859, 1.7328, 1.3795, 1.4354], device='cuda:0'), covar=tensor([0.0697, 0.0396, 0.0333, 0.0759], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 12:42:51,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.792e+02 1.164e+03 1.481e+03 1.849e+03 6.717e+03, threshold=2.962e+03, percent-clipped=5.0
+2023-03-06 12:42:52,187 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=541830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:42:56,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=541833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:43:01,996 INFO [train.py:968] (0/2) Epoch 12, batch 40650, giga_loss[loss=0.2932, simple_loss=0.3663, pruned_loss=0.1101, over 27638.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3463, pruned_loss=0.1027, over 5703715.63 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3722, pruned_loss=0.1242, over 5700943.99 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3426, pruned_loss=0.0997, over 5706148.13 frames. ], batch size: 472, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:43:17,758 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=541862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:43:44,900 INFO [train.py:968] (0/2) Epoch 12, batch 40700, giga_loss[loss=0.2609, simple_loss=0.3368, pruned_loss=0.09246, over 28647.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3507, pruned_loss=0.105, over 5684694.29 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3726, pruned_loss=0.1245, over 5688050.91 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3471, pruned_loss=0.1021, over 5698257.31 frames. ], batch size: 92, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:44:13,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9899, 1.1528, 3.3255, 2.9151], device='cuda:0'), covar=tensor([0.1625, 0.2644, 0.0468, 0.0984], device='cuda:0'), in_proj_covar=tensor([0.0663, 0.0592, 0.0860, 0.0772], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:44:14,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.822e+02 1.170e+03 1.444e+03 1.821e+03 9.267e+03, threshold=2.889e+03, percent-clipped=9.0
+2023-03-06 12:44:23,281 INFO [train.py:968] (0/2) Epoch 12, batch 40750, giga_loss[loss=0.2614, simple_loss=0.3375, pruned_loss=0.09259, over 28692.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3532, pruned_loss=0.1058, over 5687753.52 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3728, pruned_loss=0.1245, over 5683095.82 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3492, pruned_loss=0.1027, over 5702922.86 frames. ], batch size: 92, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:44:54,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3600, 1.6270, 1.3723, 1.5085], device='cuda:0'), covar=tensor([0.0700, 0.0338, 0.0315, 0.0777], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 12:45:08,418 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=541988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:45:13,290 INFO [train.py:968] (0/2) Epoch 12, batch 40800, giga_loss[loss=0.2939, simple_loss=0.362, pruned_loss=0.1129, over 28757.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3571, pruned_loss=0.1092, over 5686032.95 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3727, pruned_loss=0.1244, over 5681780.20 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.354, pruned_loss=0.1066, over 5698886.62 frames. ], batch size: 119, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:45:15,631 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-06 12:45:23,338 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-542000.pt
+2023-03-06 12:45:26,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4910, 1.7148, 1.3649, 1.7368], device='cuda:0'), covar=tensor([0.2282, 0.2219, 0.2518, 0.1919], device='cuda:0'), in_proj_covar=tensor([0.1327, 0.0977, 0.1164, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 12:45:29,556 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-06 12:45:53,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.515e+02 1.296e+03 1.851e+03 2.561e+03 6.544e+03, threshold=3.701e+03, percent-clipped=18.0
+2023-03-06 12:46:04,695 INFO [train.py:968] (0/2) Epoch 12, batch 40850, giga_loss[loss=0.3601, simple_loss=0.4142, pruned_loss=0.153, over 28578.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3627, pruned_loss=0.1142, over 5690091.21 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3725, pruned_loss=0.1243, over 5684827.77 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3601, pruned_loss=0.112, over 5697734.13 frames. ], batch size: 336, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:46:50,336 INFO [train.py:968] (0/2) Epoch 12, batch 40900, giga_loss[loss=0.3739, simple_loss=0.4171, pruned_loss=0.1653, over 28715.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3714, pruned_loss=0.1209, over 5682877.86 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3729, pruned_loss=0.1248, over 5682021.66 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3687, pruned_loss=0.1185, over 5691465.81 frames. ], batch size: 284, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:47:26,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.172e+03 1.648e+03 1.960e+03 2.567e+03 5.430e+03, threshold=3.920e+03, percent-clipped=7.0
+2023-03-06 12:47:27,510 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=542131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:47:29,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=542134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:47:35,414 INFO [train.py:968] (0/2) Epoch 12, batch 40950, giga_loss[loss=0.3474, simple_loss=0.402, pruned_loss=0.1464, over 28987.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3752, pruned_loss=0.1237, over 5686228.49 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3729, pruned_loss=0.1246, over 5684076.74 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3731, pruned_loss=0.1219, over 5691214.98 frames. ], batch size: 66, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:47:55,664 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=542163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:48:15,128 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-06 12:48:21,771 INFO [train.py:968] (0/2) Epoch 12, batch 41000, giga_loss[loss=0.445, simple_loss=0.4685, pruned_loss=0.2107, over 27617.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3811, pruned_loss=0.1286, over 5688301.65 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3731, pruned_loss=0.1248, over 5685653.30 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3793, pruned_loss=0.127, over 5691154.18 frames. ], batch size: 472, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:48:47,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9273, 1.9391, 1.7800, 1.6327], device='cuda:0'), covar=tensor([0.1357, 0.2043, 0.1944, 0.1989], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0724, 0.0671, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 12:48:55,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.878e+03 2.282e+03 2.954e+03 7.280e+03, threshold=4.564e+03, percent-clipped=13.0
+2023-03-06 12:49:07,082 INFO [train.py:968] (0/2) Epoch 12, batch 41050, giga_loss[loss=0.3064, simple_loss=0.3675, pruned_loss=0.1226, over 28346.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3857, pruned_loss=0.1327, over 5682580.58 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3733, pruned_loss=0.1248, over 5693313.58 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3844, pruned_loss=0.1316, over 5678286.54 frames. ], batch size: 65, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:49:22,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3023, 1.2693, 1.2926, 1.5133], device='cuda:0'), covar=tensor([0.0710, 0.0364, 0.0301, 0.0749], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 12:49:51,827 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8443, 2.4178, 2.6402, 2.2097], device='cuda:0'), covar=tensor([0.1125, 0.1668, 0.1294, 0.1540], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0727, 0.0673, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 12:49:59,797 INFO [train.py:968] (0/2) Epoch 12, batch 41100, giga_loss[loss=0.3202, simple_loss=0.3823, pruned_loss=0.1291, over 28672.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3879, pruned_loss=0.1357, over 5671856.61 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3734, pruned_loss=0.125, over 5697822.50 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3872, pruned_loss=0.1349, over 5663808.24 frames. ], batch size: 262, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:50:12,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3924, 1.7064, 1.3125, 1.5984], device='cuda:0'), covar=tensor([0.2289, 0.2142, 0.2445, 0.2040], device='cuda:0'), in_proj_covar=tensor([0.1319, 0.0972, 0.1159, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 12:50:42,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.746e+03 2.221e+03 2.785e+03 7.750e+03, threshold=4.442e+03, percent-clipped=2.0
+2023-03-06 12:50:52,750 INFO [train.py:968] (0/2) Epoch 12, batch 41150, giga_loss[loss=0.4988, simple_loss=0.4824, pruned_loss=0.2576, over 26398.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3907, pruned_loss=0.1395, over 5674065.41 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3729, pruned_loss=0.1245, over 5703489.62 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3911, pruned_loss=0.1397, over 5661430.75 frames. ], batch size: 555, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:51:27,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0963, 1.3149, 3.2920, 2.9042], device='cuda:0'), covar=tensor([0.1544, 0.2356, 0.0493, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.0666, 0.0593, 0.0864, 0.0776], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:51:45,311 INFO [train.py:968] (0/2) Epoch 12, batch 41200, giga_loss[loss=0.304, simple_loss=0.3673, pruned_loss=0.1204, over 28950.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.395, pruned_loss=0.1442, over 5653697.80 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3733, pruned_loss=0.1248, over 5706448.10 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3956, pruned_loss=0.1447, over 5639217.27 frames. ], batch size: 227, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:52:24,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.741e+03 2.332e+03 3.116e+03 6.054e+03, threshold=4.665e+03, percent-clipped=8.0
+2023-03-06 12:52:24,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0980, 3.9222, 3.7428, 1.7871], device='cuda:0'), covar=tensor([0.0602, 0.0761, 0.0748, 0.2191], device='cuda:0'), in_proj_covar=tensor([0.1076, 0.0995, 0.0870, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 12:52:39,579 INFO [train.py:968] (0/2) Epoch 12, batch 41250, giga_loss[loss=0.3927, simple_loss=0.4407, pruned_loss=0.1724, over 28917.00 frames. ], tot_loss[loss=0.3481, simple_loss=0.3995, pruned_loss=0.1484, over 5643349.03 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3732, pruned_loss=0.1249, over 5709018.98 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.4005, pruned_loss=0.1491, over 5628612.62 frames. ], batch size: 227, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:53:13,797 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-06 12:53:35,874 INFO [train.py:968] (0/2) Epoch 12, batch 41300, giga_loss[loss=0.3926, simple_loss=0.429, pruned_loss=0.1781, over 27962.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.401, pruned_loss=0.1503, over 5629961.19 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3733, pruned_loss=0.1251, over 5705201.33 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4021, pruned_loss=0.1511, over 5620982.84 frames. ], batch size: 412, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:53:45,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1609, 2.1726, 1.5760, 1.9507], device='cuda:0'), covar=tensor([0.0775, 0.0662, 0.0972, 0.0980], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0440, 0.0501, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 12:53:55,193 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-06 12:54:16,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.716e+03 2.281e+03 3.842e+03 1.040e+04, threshold=4.562e+03, percent-clipped=13.0
+2023-03-06 12:54:27,208 INFO [train.py:968] (0/2) Epoch 12, batch 41350, giga_loss[loss=0.3276, simple_loss=0.3942, pruned_loss=0.1306, over 28812.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.3993, pruned_loss=0.1496, over 5636330.81 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3733, pruned_loss=0.125, over 5707036.71 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4003, pruned_loss=0.1504, over 5626991.96 frames. ], batch size: 119, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:55:16,802 INFO [train.py:968] (0/2) Epoch 12, batch 41400, giga_loss[loss=0.3926, simple_loss=0.4318, pruned_loss=0.1767, over 28711.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.3984, pruned_loss=0.148, over 5646603.81 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.374, pruned_loss=0.1255, over 5701975.55 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.3992, pruned_loss=0.1488, over 5641918.90 frames. ], batch size: 307, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:55:53,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4573, 2.1724, 1.6855, 0.6102], device='cuda:0'), covar=tensor([0.4183, 0.2336, 0.3187, 0.4881], device='cuda:0'), in_proj_covar=tensor([0.1609, 0.1527, 0.1521, 0.1326], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-06 12:55:59,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.789e+03 2.329e+03 3.271e+03 7.879e+03, threshold=4.658e+03, percent-clipped=7.0
+2023-03-06 12:56:09,179 INFO [train.py:968] (0/2) Epoch 12, batch 41450, giga_loss[loss=0.416, simple_loss=0.4449, pruned_loss=0.1936, over 27431.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3974, pruned_loss=0.1456, over 5642283.65 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3743, pruned_loss=0.1257, over 5694646.72 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.398, pruned_loss=0.1463, over 5644823.28 frames. ], batch size: 472, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:57:00,852 INFO [train.py:968] (0/2) Epoch 12, batch 41500, giga_loss[loss=0.3515, simple_loss=0.4092, pruned_loss=0.1468, over 28726.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3971, pruned_loss=0.1444, over 5660930.51 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3739, pruned_loss=0.1253, over 5700810.13 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.3988, pruned_loss=0.146, over 5655953.05 frames. ], batch size: 119, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:57:41,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.655e+02 1.695e+03 2.170e+03 3.093e+03 1.058e+04, threshold=4.340e+03, percent-clipped=8.0
+2023-03-06 12:57:50,421 INFO [train.py:968] (0/2) Epoch 12, batch 41550, giga_loss[loss=0.3148, simple_loss=0.3824, pruned_loss=0.1236, over 28879.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3958, pruned_loss=0.1434, over 5645036.65 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3735, pruned_loss=0.1251, over 5698738.48 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.3984, pruned_loss=0.1456, over 5640875.43 frames. ], batch size: 186, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:58:05,732 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=542757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:58:40,912 INFO [train.py:968] (0/2) Epoch 12, batch 41600, libri_loss[loss=0.2978, simple_loss=0.3655, pruned_loss=0.1151, over 29299.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3933, pruned_loss=0.1401, over 5648406.48 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.373, pruned_loss=0.1247, over 5702716.72 frames. ], giga_tot_loss[loss=0.3406, simple_loss=0.3962, pruned_loss=0.1425, over 5640641.57 frames. ], batch size: 94, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:59:21,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.523e+02 1.512e+03 1.988e+03 2.713e+03 7.118e+03, threshold=3.976e+03, percent-clipped=2.0
+2023-03-06 12:59:29,087 INFO [train.py:968] (0/2) Epoch 12, batch 41650, giga_loss[loss=0.2798, simple_loss=0.3584, pruned_loss=0.1006, over 28574.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.389, pruned_loss=0.1356, over 5650531.00 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3728, pruned_loss=0.1247, over 5697295.10 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3919, pruned_loss=0.1377, over 5648573.10 frames. ], batch size: 78, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:59:53,918 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8272, 1.0818, 2.7963, 2.5869], device='cuda:0'), covar=tensor([0.1597, 0.2468, 0.0572, 0.1614], device='cuda:0'), in_proj_covar=tensor([0.0668, 0.0595, 0.0866, 0.0773], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 12:59:54,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6313, 2.4855, 1.5582, 0.9045], device='cuda:0'), covar=tensor([0.5311, 0.2443, 0.2706, 0.4291], device='cuda:0'), in_proj_covar=tensor([0.1594, 0.1516, 0.1509, 0.1314], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 12:59:57,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5103, 1.8378, 1.7448, 1.3078], device='cuda:0'), covar=tensor([0.1693, 0.2447, 0.1488, 0.1729], device='cuda:0'), in_proj_covar=tensor([0.0832, 0.0688, 0.0873, 0.0777], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 13:00:20,783 INFO [train.py:968] (0/2) Epoch 12, batch 41700, giga_loss[loss=0.2906, simple_loss=0.3598, pruned_loss=0.1107, over 28870.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3851, pruned_loss=0.1322, over 5656499.60 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3725, pruned_loss=0.1244, over 5700842.40 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.388, pruned_loss=0.1344, over 5650776.79 frames. ], batch size: 199, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:00:57,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.761e+03 2.326e+03 3.329e+03 6.085e+03, threshold=4.652e+03, percent-clipped=15.0
+2023-03-06 13:01:04,598 INFO [train.py:968] (0/2) Epoch 12, batch 41750, giga_loss[loss=0.333, simple_loss=0.3885, pruned_loss=0.1388, over 27505.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.382, pruned_loss=0.13, over 5656696.51 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3723, pruned_loss=0.1244, over 5700166.70 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3849, pruned_loss=0.132, over 5650951.76 frames. ], batch size: 472, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:01:58,742 INFO [train.py:968] (0/2) Epoch 12, batch 41800, libri_loss[loss=0.2771, simple_loss=0.349, pruned_loss=0.1026, over 29576.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3818, pruned_loss=0.1305, over 5653104.42 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3722, pruned_loss=0.1243, over 5703075.56 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3844, pruned_loss=0.1323, over 5645066.97 frames. ], batch size: 78, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:02:35,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.541e+02 1.434e+03 1.858e+03 2.796e+03 7.158e+03, threshold=3.716e+03, percent-clipped=5.0
+2023-03-06 13:02:46,980 INFO [train.py:968] (0/2) Epoch 12, batch 41850, giga_loss[loss=0.3003, simple_loss=0.3767, pruned_loss=0.1119, over 28862.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3821, pruned_loss=0.1306, over 5661673.59 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3719, pruned_loss=0.1242, over 5702385.76 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3846, pruned_loss=0.1322, over 5655500.06 frames. ], batch size: 199, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:03:13,090 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=543069.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:03:36,884 INFO [train.py:968] (0/2) Epoch 12, batch 41900, giga_loss[loss=0.2836, simple_loss=0.3557, pruned_loss=0.1058, over 28981.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3791, pruned_loss=0.1276, over 5669578.87 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3715, pruned_loss=0.1238, over 5703291.28 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3818, pruned_loss=0.1294, over 5662598.81 frames. ], batch size: 164, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:03:37,355 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=543092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:04:19,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.996e+02 1.427e+03 1.799e+03 2.480e+03 8.589e+03, threshold=3.599e+03, percent-clipped=9.0
+2023-03-06 13:04:21,506 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=543132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:04:23,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1606, 2.0596, 1.6140, 1.7793], device='cuda:0'), covar=tensor([0.0604, 0.0494, 0.0819, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0441, 0.0504, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 13:04:29,481 INFO [train.py:968] (0/2) Epoch 12, batch 41950, giga_loss[loss=0.2847, simple_loss=0.3673, pruned_loss=0.1011, over 28914.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3788, pruned_loss=0.1252, over 5678207.68 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3716, pruned_loss=0.1238, over 5710033.79 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3811, pruned_loss=0.1267, over 5665520.17 frames. ], batch size: 136, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:05:20,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7368, 5.1423, 1.9145, 2.0540], device='cuda:0'), covar=tensor([0.1048, 0.0429, 0.0932, 0.1386], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0517, 0.0347, 0.0396], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 13:05:21,354 INFO [train.py:968] (0/2) Epoch 12, batch 42000, giga_loss[loss=0.3378, simple_loss=0.3907, pruned_loss=0.1425, over 27581.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3804, pruned_loss=0.1249, over 5679908.68 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3711, pruned_loss=0.1235, over 5712422.78 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3828, pruned_loss=0.1265, over 5667090.90 frames. ], batch size: 472, lr: 2.64e-03, grad_scale: 8.0
+2023-03-06 13:05:21,359 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 13:05:29,511 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2102, simple_loss=0.3152, pruned_loss=0.05262, over 944034.00 frames. 
+2023-03-06 13:05:29,512 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 13:06:07,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.414e+02 1.638e+03 2.186e+03 2.823e+03 6.131e+03, threshold=4.373e+03, percent-clipped=12.0
+2023-03-06 13:06:17,143 INFO [train.py:968] (0/2) Epoch 12, batch 42050, giga_loss[loss=0.45, simple_loss=0.4569, pruned_loss=0.2216, over 26620.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3823, pruned_loss=0.1278, over 5677017.19 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3707, pruned_loss=0.1234, over 5715514.42 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3847, pruned_loss=0.1291, over 5663767.95 frames. ], batch size: 555, lr: 2.64e-03, grad_scale: 8.0
+2023-03-06 13:06:47,312 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=543275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:06:50,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=543278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:07:02,069 INFO [train.py:968] (0/2) Epoch 12, batch 42100, giga_loss[loss=0.2863, simple_loss=0.3564, pruned_loss=0.1081, over 28847.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3814, pruned_loss=0.1274, over 5683353.66 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3704, pruned_loss=0.123, over 5718048.16 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.384, pruned_loss=0.129, over 5669299.86 frames. ], batch size: 199, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:07:14,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=543307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:07:37,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.996e+02 1.923e+03 2.812e+03 3.721e+03 8.256e+03, threshold=5.625e+03, percent-clipped=13.0
+2023-03-06 13:07:46,224 INFO [train.py:968] (0/2) Epoch 12, batch 42150, giga_loss[loss=0.3618, simple_loss=0.395, pruned_loss=0.1643, over 23494.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.38, pruned_loss=0.1274, over 5689141.98 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3704, pruned_loss=0.123, over 5724555.58 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3824, pruned_loss=0.1287, over 5670964.11 frames. ], batch size: 705, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:08:00,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9384, 2.1652, 2.2208, 1.7252], device='cuda:0'), covar=tensor([0.1761, 0.2057, 0.1321, 0.1548], device='cuda:0'), in_proj_covar=tensor([0.0837, 0.0697, 0.0879, 0.0784], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 13:08:35,189 INFO [train.py:968] (0/2) Epoch 12, batch 42200, libri_loss[loss=0.307, simple_loss=0.3723, pruned_loss=0.1209, over 29745.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3779, pruned_loss=0.127, over 5675742.29 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3704, pruned_loss=0.1229, over 5725625.42 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.38, pruned_loss=0.1283, over 5659171.91 frames. ], batch size: 87, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:08:50,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=543409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:09:14,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.740e+02 1.581e+03 1.926e+03 2.605e+03 1.034e+04, threshold=3.853e+03, percent-clipped=4.0
+2023-03-06 13:09:23,665 INFO [train.py:968] (0/2) Epoch 12, batch 42250, giga_loss[loss=0.3452, simple_loss=0.4055, pruned_loss=0.1425, over 27943.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3768, pruned_loss=0.1257, over 5682067.87 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3704, pruned_loss=0.1227, over 5728204.88 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3788, pruned_loss=0.127, over 5665101.77 frames. ], batch size: 412, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:09:25,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=543444.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:09:48,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=543467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:10:11,066 INFO [train.py:968] (0/2) Epoch 12, batch 42300, giga_loss[loss=0.2963, simple_loss=0.3689, pruned_loss=0.1119, over 28848.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3759, pruned_loss=0.1233, over 5685365.45 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3706, pruned_loss=0.1229, over 5720677.27 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3773, pruned_loss=0.1242, over 5677196.01 frames. ], batch size: 186, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:10:50,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.665e+02 1.540e+03 2.175e+03 3.097e+03 1.210e+04, threshold=4.351e+03, percent-clipped=11.0
+2023-03-06 13:11:00,206 INFO [train.py:968] (0/2) Epoch 12, batch 42350, giga_loss[loss=0.3396, simple_loss=0.3977, pruned_loss=0.1407, over 27977.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3768, pruned_loss=0.1237, over 5683191.06 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3707, pruned_loss=0.1229, over 5722243.67 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.378, pruned_loss=0.1244, over 5674908.02 frames. ], batch size: 412, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:11:07,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6573, 2.2274, 1.4854, 0.8811], device='cuda:0'), covar=tensor([0.4980, 0.2901, 0.2759, 0.4736], device='cuda:0'), in_proj_covar=tensor([0.1596, 0.1519, 0.1512, 0.1311], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 13:11:13,818 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-06 13:11:43,257 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=543587.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:11:46,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=543590.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:11:46,423 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-06 13:11:47,305 INFO [train.py:968] (0/2) Epoch 12, batch 42400, giga_loss[loss=0.2831, simple_loss=0.3571, pruned_loss=0.1045, over 28861.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.377, pruned_loss=0.1241, over 5687350.62 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3714, pruned_loss=0.1234, over 5725555.42 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3774, pruned_loss=0.1243, over 5677371.15 frames. ], batch size: 112, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:12:03,757 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=543610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:12:06,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=543613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:12:08,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7889, 2.1214, 1.9966, 1.9389], device='cuda:0'), covar=tensor([0.1201, 0.1146, 0.1306, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0734, 0.0680, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 13:12:09,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3563, 1.6934, 1.3599, 1.4767], device='cuda:0'), covar=tensor([0.0744, 0.0307, 0.0312, 0.0812], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0083, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 13:12:10,575 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=543619.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:12:25,369 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.637e+03 2.231e+03 2.872e+03 7.979e+03, threshold=4.461e+03, percent-clipped=7.0
+2023-03-06 13:12:34,704 INFO [train.py:968] (0/2) Epoch 12, batch 42450, libri_loss[loss=0.3209, simple_loss=0.3815, pruned_loss=0.1301, over 29560.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3758, pruned_loss=0.124, over 5677576.43 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3714, pruned_loss=0.1234, over 5718436.23 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3762, pruned_loss=0.1242, over 5674853.52 frames. ], batch size: 77, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:12:34,984 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=543642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:13:25,178 INFO [train.py:968] (0/2) Epoch 12, batch 42500, giga_loss[loss=0.323, simple_loss=0.3645, pruned_loss=0.1408, over 23722.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3746, pruned_loss=0.124, over 5671707.68 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3714, pruned_loss=0.1234, over 5719880.36 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.375, pruned_loss=0.1241, over 5668018.60 frames. ], batch size: 705, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:14:07,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.597e+03 2.022e+03 3.562e+03 9.199e+03, threshold=4.044e+03, percent-clipped=11.0
+2023-03-06 13:14:17,502 INFO [train.py:968] (0/2) Epoch 12, batch 42550, giga_loss[loss=0.3693, simple_loss=0.4041, pruned_loss=0.1673, over 26743.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3745, pruned_loss=0.125, over 5666391.72 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3711, pruned_loss=0.1231, over 5721423.35 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3751, pruned_loss=0.1253, over 5661671.34 frames. ], batch size: 555, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:14:34,515 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=543760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:14:59,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=543784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:15:05,969 INFO [train.py:968] (0/2) Epoch 12, batch 42600, giga_loss[loss=0.3622, simple_loss=0.4012, pruned_loss=0.1616, over 28070.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3736, pruned_loss=0.1249, over 5679448.52 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3706, pruned_loss=0.1227, over 5725776.06 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3746, pruned_loss=0.1256, over 5670519.17 frames. ], batch size: 412, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:15:08,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4195, 3.8665, 1.6317, 1.4870], device='cuda:0'), covar=tensor([0.0950, 0.0330, 0.0870, 0.1336], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0516, 0.0346, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 13:15:47,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.623e+03 2.107e+03 2.889e+03 7.915e+03, threshold=4.214e+03, percent-clipped=5.0
+2023-03-06 13:15:51,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2865, 1.5086, 1.4401, 1.5500], device='cuda:0'), covar=tensor([0.0759, 0.0309, 0.0302, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 13:15:54,238 INFO [train.py:968] (0/2) Epoch 12, batch 42650, giga_loss[loss=0.2785, simple_loss=0.3431, pruned_loss=0.107, over 28844.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3724, pruned_loss=0.1244, over 5687328.33 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3703, pruned_loss=0.1226, over 5730499.20 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3735, pruned_loss=0.1251, over 5674628.72 frames. ], batch size: 99, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:16:07,254 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=543856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:16:13,264 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-06 13:16:45,600 INFO [train.py:968] (0/2) Epoch 12, batch 42700, giga_loss[loss=0.2974, simple_loss=0.3642, pruned_loss=0.1153, over 29022.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3738, pruned_loss=0.1253, over 5687723.40 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3701, pruned_loss=0.1225, over 5731202.70 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3748, pruned_loss=0.126, over 5677066.20 frames. ], batch size: 106, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:16:59,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2621, 1.2435, 1.1322, 0.8885], device='cuda:0'), covar=tensor([0.0769, 0.0494, 0.0994, 0.1005], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0443, 0.0504, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 13:17:18,856 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=543927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:17:21,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=543930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:17:26,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.459e+02 1.640e+03 2.086e+03 2.881e+03 7.021e+03, threshold=4.172e+03, percent-clipped=11.0
+2023-03-06 13:17:31,397 INFO [train.py:968] (0/2) Epoch 12, batch 42750, libri_loss[loss=0.3187, simple_loss=0.3846, pruned_loss=0.1264, over 29742.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3742, pruned_loss=0.1246, over 5679448.45 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3705, pruned_loss=0.1226, over 5723775.48 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3748, pruned_loss=0.1251, over 5676077.38 frames. ], batch size: 87, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:17:37,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2931, 1.6858, 1.4669, 1.5177], device='cuda:0'), covar=tensor([0.0794, 0.0310, 0.0314, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 13:17:46,988 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=543959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:18:16,258 INFO [train.py:968] (0/2) Epoch 12, batch 42800, giga_loss[loss=0.3259, simple_loss=0.3919, pruned_loss=0.13, over 28852.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3733, pruned_loss=0.1231, over 5675595.75 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3703, pruned_loss=0.1224, over 5724363.51 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.374, pruned_loss=0.1237, over 5671641.83 frames. ], batch size: 199, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:18:24,448 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-544000.pt
+2023-03-06 13:18:57,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.518e+03 2.041e+03 2.830e+03 6.210e+03, threshold=4.081e+03, percent-clipped=4.0
+2023-03-06 13:18:59,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5398, 1.6907, 1.3086, 1.7102], device='cuda:0'), covar=tensor([0.2503, 0.2508, 0.2898, 0.2246], device='cuda:0'), in_proj_covar=tensor([0.1325, 0.0979, 0.1162, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 13:19:05,387 INFO [train.py:968] (0/2) Epoch 12, batch 42850, giga_loss[loss=0.3301, simple_loss=0.3674, pruned_loss=0.1464, over 23579.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3743, pruned_loss=0.1234, over 5674196.73 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3702, pruned_loss=0.1224, over 5725781.06 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.375, pruned_loss=0.1239, over 5668795.04 frames. ], batch size: 705, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:19:50,350 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-06 13:19:54,762 INFO [train.py:968] (0/2) Epoch 12, batch 42900, giga_loss[loss=0.3177, simple_loss=0.3849, pruned_loss=0.1253, over 28954.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3761, pruned_loss=0.1257, over 5671907.30 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3704, pruned_loss=0.1224, over 5728536.07 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3767, pruned_loss=0.1261, over 5663587.89 frames. ], batch size: 199, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:20:38,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.196e+03 1.562e+03 2.075e+03 2.886e+03 6.970e+03, threshold=4.150e+03, percent-clipped=13.0
+2023-03-06 13:20:39,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=544135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:20:47,321 INFO [train.py:968] (0/2) Epoch 12, batch 42950, giga_loss[loss=0.3836, simple_loss=0.4178, pruned_loss=0.1746, over 27575.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3794, pruned_loss=0.1292, over 5664017.40 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3706, pruned_loss=0.1226, over 5720317.30 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3797, pruned_loss=0.1294, over 5663267.90 frames. ], batch size: 472, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:21:21,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3838, 1.6432, 1.3135, 1.6297], device='cuda:0'), covar=tensor([0.0764, 0.0299, 0.0319, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 13:21:42,254 INFO [train.py:968] (0/2) Epoch 12, batch 43000, giga_loss[loss=0.4324, simple_loss=0.4469, pruned_loss=0.2089, over 26671.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3808, pruned_loss=0.1321, over 5656230.41 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3709, pruned_loss=0.1229, over 5724227.61 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.381, pruned_loss=0.1321, over 5651246.17 frames. ], batch size: 555, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:22:14,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5914, 1.8397, 1.7835, 1.6005], device='cuda:0'), covar=tensor([0.1154, 0.1233, 0.1497, 0.1291], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0732, 0.0677, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 13:22:21,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=544231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:22:25,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.652e+02 1.847e+03 2.382e+03 3.186e+03 1.295e+04, threshold=4.765e+03, percent-clipped=14.0
+2023-03-06 13:22:32,529 INFO [train.py:968] (0/2) Epoch 12, batch 43050, giga_loss[loss=0.3472, simple_loss=0.3995, pruned_loss=0.1474, over 28824.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3825, pruned_loss=0.1342, over 5657042.81 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.371, pruned_loss=0.1229, over 5723486.41 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3828, pruned_loss=0.1345, over 5652473.62 frames. ], batch size: 285, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:23:06,185 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=544278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:23:08,132 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=544281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:23:18,501 INFO [train.py:968] (0/2) Epoch 12, batch 43100, giga_loss[loss=0.2898, simple_loss=0.3558, pruned_loss=0.1118, over 28667.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3808, pruned_loss=0.1329, over 5671088.74 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.371, pruned_loss=0.1229, over 5725401.01 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3812, pruned_loss=0.1333, over 5664753.74 frames. ], batch size: 242, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:23:34,382 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=544310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:23:57,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.014e+02 1.612e+03 2.128e+03 3.305e+03 7.841e+03, threshold=4.257e+03, percent-clipped=10.0
+2023-03-06 13:24:04,186 INFO [train.py:968] (0/2) Epoch 12, batch 43150, giga_loss[loss=0.2589, simple_loss=0.3447, pruned_loss=0.08653, over 28915.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3777, pruned_loss=0.1302, over 5677357.74 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3712, pruned_loss=0.1231, over 5730853.06 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3781, pruned_loss=0.1305, over 5666038.64 frames. ], batch size: 164, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:24:31,030 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=544374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:24:33,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=544377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:24:45,241 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=544390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:24:46,317 INFO [train.py:968] (0/2) Epoch 12, batch 43200, giga_loss[loss=0.2799, simple_loss=0.3546, pruned_loss=0.1026, over 28924.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3764, pruned_loss=0.1271, over 5687224.99 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3717, pruned_loss=0.1236, over 5733215.06 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3765, pruned_loss=0.1271, over 5674271.76 frames. ], batch size: 174, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:24:49,700 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.33 vs. limit=2.0
+2023-03-06 13:24:59,831 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=544406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:25:06,253 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 13:25:25,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.035e+02 1.659e+03 2.053e+03 3.049e+03 1.350e+04, threshold=4.106e+03, percent-clipped=14.0
+2023-03-06 13:25:31,750 INFO [train.py:968] (0/2) Epoch 12, batch 43250, giga_loss[loss=0.3046, simple_loss=0.3636, pruned_loss=0.1228, over 28712.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.374, pruned_loss=0.1254, over 5658937.57 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3721, pruned_loss=0.1239, over 5710011.76 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3738, pruned_loss=0.1252, over 5667920.45 frames. ], batch size: 92, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:25:44,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5554, 2.1444, 2.2954, 1.8870], device='cuda:0'), covar=tensor([0.1301, 0.2309, 0.1731, 0.2045], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0732, 0.0678, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 13:26:16,717 INFO [train.py:968] (0/2) Epoch 12, batch 43300, giga_loss[loss=0.3036, simple_loss=0.3674, pruned_loss=0.1199, over 28794.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3731, pruned_loss=0.1253, over 5651907.98 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3727, pruned_loss=0.1244, over 5703977.32 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3724, pruned_loss=0.1248, over 5663203.24 frames. ], batch size: 119, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:26:31,613 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=544509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:26:58,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.754e+02 1.677e+03 2.074e+03 2.901e+03 7.115e+03, threshold=4.148e+03, percent-clipped=10.0
+2023-03-06 13:27:04,076 INFO [train.py:968] (0/2) Epoch 12, batch 43350, giga_loss[loss=0.3588, simple_loss=0.3876, pruned_loss=0.165, over 23474.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3714, pruned_loss=0.1251, over 5648669.54 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3729, pruned_loss=0.1247, over 5697599.02 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3708, pruned_loss=0.1244, over 5662220.89 frames. ], batch size: 705, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:27:49,649 INFO [train.py:968] (0/2) Epoch 12, batch 43400, giga_loss[loss=0.3052, simple_loss=0.3709, pruned_loss=0.1197, over 28258.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3726, pruned_loss=0.1261, over 5642414.34 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3727, pruned_loss=0.1246, over 5691001.71 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3722, pruned_loss=0.1256, over 5658258.14 frames. ], batch size: 368, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:28:30,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.649e+03 2.039e+03 3.061e+03 6.886e+03, threshold=4.079e+03, percent-clipped=6.0
+2023-03-06 13:28:35,503 INFO [train.py:968] (0/2) Epoch 12, batch 43450, libri_loss[loss=0.2655, simple_loss=0.3357, pruned_loss=0.09759, over 29533.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3762, pruned_loss=0.1274, over 5658714.90 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3722, pruned_loss=0.1243, over 5696499.58 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3764, pruned_loss=0.1274, over 5664976.38 frames. ], batch size: 80, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:29:24,775 INFO [train.py:968] (0/2) Epoch 12, batch 43500, giga_loss[loss=0.3671, simple_loss=0.3988, pruned_loss=0.1677, over 23682.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3788, pruned_loss=0.1261, over 5652703.22 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3723, pruned_loss=0.1244, over 5698562.85 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3789, pruned_loss=0.1261, over 5655333.17 frames. ], batch size: 705, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:29:37,159 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=544704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:29:46,517 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4762, 1.5284, 1.2352, 1.1515], device='cuda:0'), covar=tensor([0.0761, 0.0505, 0.0929, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0444, 0.0503, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 13:30:08,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.380e+02 1.495e+03 1.977e+03 2.841e+03 5.830e+03, threshold=3.955e+03, percent-clipped=9.0
+2023-03-06 13:30:13,470 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3373, 3.2189, 1.5327, 1.5102], device='cuda:0'), covar=tensor([0.0926, 0.0368, 0.0857, 0.1247], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0517, 0.0348, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 13:30:14,456 INFO [train.py:968] (0/2) Epoch 12, batch 43550, giga_loss[loss=0.3358, simple_loss=0.4017, pruned_loss=0.1349, over 28329.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3806, pruned_loss=0.1274, over 5657135.90 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3723, pruned_loss=0.1246, over 5695197.24 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.381, pruned_loss=0.1273, over 5660384.11 frames. ], batch size: 369, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:30:36,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=544765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:30:54,427 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=544783.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:31:00,808 INFO [train.py:968] (0/2) Epoch 12, batch 43600, giga_loss[loss=0.2969, simple_loss=0.3679, pruned_loss=0.1129, over 28488.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.383, pruned_loss=0.129, over 5671547.98 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3723, pruned_loss=0.1245, over 5699594.97 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3835, pruned_loss=0.129, over 5669708.25 frames. ], batch size: 85, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:31:30,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=544823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:31:38,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3904, 1.4947, 1.5295, 1.4597], device='cuda:0'), covar=tensor([0.1133, 0.1257, 0.1506, 0.1196], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0733, 0.0679, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 13:31:41,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.663e+03 2.131e+03 2.920e+03 6.599e+03, threshold=4.262e+03, percent-clipped=10.0
+2023-03-06 13:31:45,788 INFO [train.py:968] (0/2) Epoch 12, batch 43650, libri_loss[loss=0.2883, simple_loss=0.3565, pruned_loss=0.11, over 29523.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3845, pruned_loss=0.1312, over 5673413.78 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.372, pruned_loss=0.1244, over 5708190.34 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3858, pruned_loss=0.1316, over 5662334.75 frames. ], batch size: 81, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:32:20,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=544884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:32:27,174 INFO [train.py:968] (0/2) Epoch 12, batch 43700, giga_loss[loss=0.3153, simple_loss=0.3773, pruned_loss=0.1267, over 28880.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3839, pruned_loss=0.1313, over 5682615.33 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3722, pruned_loss=0.1246, over 5707399.46 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.385, pruned_loss=0.1316, over 5673716.19 frames. ], batch size: 199, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:32:42,640 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=544908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:32:45,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=544911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:33:08,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.938e+02 1.615e+03 2.019e+03 2.883e+03 7.459e+03, threshold=4.037e+03, percent-clipped=7.0
+2023-03-06 13:33:11,864 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=544940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:33:13,049 INFO [train.py:968] (0/2) Epoch 12, batch 43750, giga_loss[loss=0.3757, simple_loss=0.4193, pruned_loss=0.166, over 27567.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3821, pruned_loss=0.131, over 5673938.95 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3717, pruned_loss=0.1241, over 5712971.13 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.384, pruned_loss=0.1321, over 5660223.34 frames. ], batch size: 472, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:33:50,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2379, 1.1355, 3.9695, 3.2511], device='cuda:0'), covar=tensor([0.1726, 0.2725, 0.0432, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0678, 0.0604, 0.0878, 0.0787], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 13:34:01,624 INFO [train.py:968] (0/2) Epoch 12, batch 43800, giga_loss[loss=0.3698, simple_loss=0.4119, pruned_loss=0.1639, over 27513.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3802, pruned_loss=0.1306, over 5663759.85 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3713, pruned_loss=0.1239, over 5704292.86 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3821, pruned_loss=0.1316, over 5660168.15 frames. ], batch size: 472, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:34:37,074 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=545027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:34:41,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=545030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:34:49,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.724e+03 2.337e+03 3.482e+03 7.496e+03, threshold=4.674e+03, percent-clipped=16.0
+2023-03-06 13:34:54,030 INFO [train.py:968] (0/2) Epoch 12, batch 43850, giga_loss[loss=0.2999, simple_loss=0.3654, pruned_loss=0.1171, over 27953.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3796, pruned_loss=0.1307, over 5654584.03 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3713, pruned_loss=0.1238, over 5706922.16 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3812, pruned_loss=0.1317, over 5648691.16 frames. ], batch size: 412, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:35:11,319 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-06 13:35:11,802 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=545059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:35:32,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=545079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:35:46,446 INFO [train.py:968] (0/2) Epoch 12, batch 43900, giga_loss[loss=0.311, simple_loss=0.3747, pruned_loss=0.1236, over 29065.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3791, pruned_loss=0.1309, over 5643032.22 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3709, pruned_loss=0.1235, over 5700643.50 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.381, pruned_loss=0.1321, over 5643683.54 frames. ], batch size: 155, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:36:28,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.554e+02 1.585e+03 1.992e+03 2.821e+03 6.624e+03, threshold=3.983e+03, percent-clipped=4.0
+2023-03-06 13:36:31,103 INFO [train.py:968] (0/2) Epoch 12, batch 43950, giga_loss[loss=0.2787, simple_loss=0.3523, pruned_loss=0.1025, over 28808.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3773, pruned_loss=0.1295, over 5659703.68 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3712, pruned_loss=0.1235, over 5705988.72 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3786, pruned_loss=0.1307, over 5653807.97 frames. ], batch size: 174, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:36:35,754 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=545146.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:36:47,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=545158.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:37:17,022 INFO [train.py:968] (0/2) Epoch 12, batch 44000, giga_loss[loss=0.3014, simple_loss=0.3692, pruned_loss=0.1168, over 28769.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3764, pruned_loss=0.1288, over 5667157.85 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3715, pruned_loss=0.1238, over 5708766.20 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3773, pruned_loss=0.1296, over 5659099.01 frames. ], batch size: 284, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:37:24,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=545198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:37:44,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2050, 1.4797, 1.1789, 0.9020], device='cuda:0'), covar=tensor([0.2477, 0.2377, 0.2662, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1331, 0.0983, 0.1172, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 13:37:47,351 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=545222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:37:50,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=545225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:38:02,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.686e+02 1.482e+03 2.006e+03 2.911e+03 7.508e+03, threshold=4.012e+03, percent-clipped=12.0
+2023-03-06 13:38:07,701 INFO [train.py:968] (0/2) Epoch 12, batch 44050, giga_loss[loss=0.401, simple_loss=0.4327, pruned_loss=0.1846, over 27829.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3766, pruned_loss=0.1287, over 5663348.76 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3717, pruned_loss=0.1238, over 5709474.37 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3772, pruned_loss=0.1293, over 5656245.68 frames. ], batch size: 412, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:38:19,074 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=545254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:38:59,480 INFO [train.py:968] (0/2) Epoch 12, batch 44100, giga_loss[loss=0.2911, simple_loss=0.3627, pruned_loss=0.1097, over 28586.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3784, pruned_loss=0.1294, over 5658369.19 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3713, pruned_loss=0.1236, over 5712207.04 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3794, pruned_loss=0.1303, over 5648424.90 frames. ], batch size: 60, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:39:00,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4454, 3.5179, 1.6254, 1.4702], device='cuda:0'), covar=tensor([0.0917, 0.0340, 0.0842, 0.1340], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0518, 0.0347, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 13:39:05,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6365, 1.8138, 1.6044, 1.4559], device='cuda:0'), covar=tensor([0.2033, 0.1797, 0.1621, 0.1858], device='cuda:0'), in_proj_covar=tensor([0.1764, 0.1666, 0.1627, 0.1714], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 13:39:06,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=545301.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:39:10,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=545304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:39:38,920 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=545333.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:39:41,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.771e+02 1.516e+03 2.038e+03 2.526e+03 5.177e+03, threshold=4.076e+03, percent-clipped=5.0
+2023-03-06 13:39:45,361 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=545341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:39:45,760 INFO [train.py:968] (0/2) Epoch 12, batch 44150, giga_loss[loss=0.296, simple_loss=0.3648, pruned_loss=0.1136, over 28805.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3789, pruned_loss=0.1299, over 5659898.76 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3711, pruned_loss=0.1235, over 5713503.50 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3801, pruned_loss=0.1308, over 5649852.79 frames. ], batch size: 145, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:39:49,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=545344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:40:12,099 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=545373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:40:28,643 INFO [train.py:968] (0/2) Epoch 12, batch 44200, giga_loss[loss=0.2983, simple_loss=0.3878, pruned_loss=0.1044, over 28463.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3791, pruned_loss=0.1295, over 5672780.86 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3712, pruned_loss=0.1236, over 5715826.73 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3804, pruned_loss=0.1305, over 5659643.81 frames. ], batch size: 85, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:41:09,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.891e+02 1.500e+03 2.087e+03 3.241e+03 8.724e+03, threshold=4.175e+03, percent-clipped=14.0
+2023-03-06 13:41:12,290 INFO [train.py:968] (0/2) Epoch 12, batch 44250, giga_loss[loss=0.2917, simple_loss=0.3721, pruned_loss=0.1057, over 28907.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3808, pruned_loss=0.128, over 5671333.08 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3712, pruned_loss=0.1237, over 5712612.78 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.382, pruned_loss=0.1289, over 5663179.26 frames. ], batch size: 186, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:41:18,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2585, 2.3407, 1.8466, 2.0000], device='cuda:0'), covar=tensor([0.0802, 0.0690, 0.0910, 0.1083], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0443, 0.0502, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 13:41:56,699 INFO [train.py:968] (0/2) Epoch 12, batch 44300, giga_loss[loss=0.26, simple_loss=0.3472, pruned_loss=0.08639, over 28540.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3825, pruned_loss=0.1271, over 5671747.49 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3715, pruned_loss=0.1239, over 5715344.51 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3835, pruned_loss=0.1277, over 5661648.91 frames. ], batch size: 85, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:42:25,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=545521.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:42:40,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.222e+02 1.574e+03 2.071e+03 2.761e+03 6.528e+03, threshold=4.142e+03, percent-clipped=5.0
+2023-03-06 13:42:44,782 INFO [train.py:968] (0/2) Epoch 12, batch 44350, giga_loss[loss=0.3685, simple_loss=0.3973, pruned_loss=0.1698, over 23426.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3835, pruned_loss=0.1283, over 5666537.60 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3707, pruned_loss=0.1233, over 5721327.64 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3854, pruned_loss=0.1294, over 5651524.00 frames. ], batch size: 705, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:43:36,191 INFO [train.py:968] (0/2) Epoch 12, batch 44400, giga_loss[loss=0.3167, simple_loss=0.3763, pruned_loss=0.1285, over 28980.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3867, pruned_loss=0.1318, over 5664344.23 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3707, pruned_loss=0.1233, over 5722274.00 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3883, pruned_loss=0.1327, over 5651500.55 frames. ], batch size: 136, lr: 2.64e-03, grad_scale: 8.0
+2023-03-06 13:44:14,564 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=545631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:44:20,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.696e+03 2.144e+03 3.058e+03 5.982e+03, threshold=4.289e+03, percent-clipped=6.0
+2023-03-06 13:44:23,988 INFO [train.py:968] (0/2) Epoch 12, batch 44450, giga_loss[loss=0.3099, simple_loss=0.3765, pruned_loss=0.1216, over 28985.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3879, pruned_loss=0.1334, over 5671578.53 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.371, pruned_loss=0.1236, over 5722671.31 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3894, pruned_loss=0.1342, over 5659279.87 frames. ], batch size: 174, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:44:42,197 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=545660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:44:47,181 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=545664.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:44:50,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=545667.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:44:51,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5854, 1.7957, 1.4970, 1.7759], device='cuda:0'), covar=tensor([0.2339, 0.2382, 0.2583, 0.2205], device='cuda:0'), in_proj_covar=tensor([0.1331, 0.0979, 0.1172, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 13:45:12,254 INFO [train.py:968] (0/2) Epoch 12, batch 44500, giga_loss[loss=0.2959, simple_loss=0.3703, pruned_loss=0.1108, over 28530.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3851, pruned_loss=0.1317, over 5670256.98 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3707, pruned_loss=0.1234, over 5723703.96 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3866, pruned_loss=0.1326, over 5659576.78 frames. ], batch size: 78, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:45:17,036 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=545696.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:45:57,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.702e+03 2.317e+03 3.408e+03 1.155e+04, threshold=4.634e+03, percent-clipped=15.0
+2023-03-06 13:45:59,845 INFO [train.py:968] (0/2) Epoch 12, batch 44550, giga_loss[loss=0.3372, simple_loss=0.3973, pruned_loss=0.1386, over 28016.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3837, pruned_loss=0.1299, over 5675156.46 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3707, pruned_loss=0.1234, over 5725525.41 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3852, pruned_loss=0.1307, over 5664299.28 frames. ], batch size: 412, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:46:04,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-06 13:46:09,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3348, 1.2645, 3.9767, 3.2989], device='cuda:0'), covar=tensor([0.1671, 0.2668, 0.0441, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0678, 0.0604, 0.0877, 0.0788], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 13:46:47,308 INFO [train.py:968] (0/2) Epoch 12, batch 44600, giga_loss[loss=0.2918, simple_loss=0.3738, pruned_loss=0.1049, over 28802.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3846, pruned_loss=0.1283, over 5683455.00 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3709, pruned_loss=0.1236, over 5726570.78 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3855, pruned_loss=0.1288, over 5673740.95 frames. ], batch size: 66, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:47:01,695 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 13:47:12,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5235, 1.7316, 1.3466, 1.6387], device='cuda:0'), covar=tensor([0.0679, 0.0275, 0.0312, 0.0756], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 13:47:25,736 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=545832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:47:35,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.771e+02 1.500e+03 1.888e+03 2.871e+03 6.385e+03, threshold=3.776e+03, percent-clipped=7.0
+2023-03-06 13:47:37,276 INFO [train.py:968] (0/2) Epoch 12, batch 44650, giga_loss[loss=0.3623, simple_loss=0.3956, pruned_loss=0.1645, over 23566.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.385, pruned_loss=0.1288, over 5672996.91 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3707, pruned_loss=0.1234, over 5728929.71 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3861, pruned_loss=0.1294, over 5662501.09 frames. ], batch size: 705, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:48:29,583 INFO [train.py:968] (0/2) Epoch 12, batch 44700, giga_loss[loss=0.2996, simple_loss=0.3625, pruned_loss=0.1184, over 28829.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3857, pruned_loss=0.1302, over 5665671.17 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3708, pruned_loss=0.1235, over 5731478.32 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3868, pruned_loss=0.1307, over 5654406.72 frames. ], batch size: 112, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:48:51,751 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-06 13:49:10,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.424e+02 1.572e+03 2.058e+03 3.010e+03 7.107e+03, threshold=4.116e+03, percent-clipped=12.0
+2023-03-06 13:49:13,988 INFO [train.py:968] (0/2) Epoch 12, batch 44750, giga_loss[loss=0.3406, simple_loss=0.394, pruned_loss=0.1436, over 28982.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3847, pruned_loss=0.13, over 5663993.55 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3714, pruned_loss=0.1239, over 5726055.35 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3853, pruned_loss=0.1301, over 5658638.50 frames. ], batch size: 145, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:49:36,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7462, 1.7597, 1.6666, 1.5448], device='cuda:0'), covar=tensor([0.1475, 0.2056, 0.2006, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0737, 0.0681, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 13:49:40,250 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=545972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:50:02,032 INFO [train.py:968] (0/2) Epoch 12, batch 44800, giga_loss[loss=0.2893, simple_loss=0.3633, pruned_loss=0.1076, over 28677.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3831, pruned_loss=0.1302, over 5630113.38 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3717, pruned_loss=0.1241, over 5689881.80 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3836, pruned_loss=0.1302, over 5656293.21 frames. ], batch size: 262, lr: 2.63e-03, grad_scale: 8.0
+2023-03-06 13:50:07,649 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-546000.pt
+2023-03-06 13:50:14,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:50:39,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:50:43,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.733e+02 1.666e+03 2.111e+03 3.249e+03 6.720e+03, threshold=4.222e+03, percent-clipped=7.0
+2023-03-06 13:50:46,061 INFO [train.py:968] (0/2) Epoch 12, batch 44850, giga_loss[loss=0.3009, simple_loss=0.3754, pruned_loss=0.1132, over 28975.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.382, pruned_loss=0.1304, over 5644797.20 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3719, pruned_loss=0.1242, over 5697326.99 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3825, pruned_loss=0.1306, over 5657733.16 frames. ], batch size: 164, lr: 2.63e-03, grad_scale: 8.0
+2023-03-06 13:50:56,495 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:51:05,787 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:51:35,589 INFO [train.py:968] (0/2) Epoch 12, batch 44900, giga_loss[loss=0.3477, simple_loss=0.4003, pruned_loss=0.1476, over 27800.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3803, pruned_loss=0.1298, over 5648784.11 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3721, pruned_loss=0.1244, over 5701086.45 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3807, pruned_loss=0.1299, over 5654976.65 frames. ], batch size: 412, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:52:03,684 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:52:18,599 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:52:23,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.567e+03 2.096e+03 3.092e+03 5.337e+03, threshold=4.192e+03, percent-clipped=9.0
+2023-03-06 13:52:25,805 INFO [train.py:968] (0/2) Epoch 12, batch 44950, giga_loss[loss=0.2792, simple_loss=0.3558, pruned_loss=0.1013, over 28584.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.38, pruned_loss=0.1307, over 5641487.76 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3724, pruned_loss=0.1245, over 5703821.79 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3802, pruned_loss=0.1308, over 5643244.67 frames. ], batch size: 60, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:52:33,262 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:52:35,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:52:54,965 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1553, 1.5313, 1.2093, 0.9801], device='cuda:0'), covar=tensor([0.2265, 0.2149, 0.2391, 0.2007], device='cuda:0'), in_proj_covar=tensor([0.1332, 0.0981, 0.1174, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 13:52:57,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5442, 1.7429, 1.4385, 2.0241], device='cuda:0'), covar=tensor([0.2491, 0.2581, 0.2803, 0.2342], device='cuda:0'), in_proj_covar=tensor([0.1332, 0.0981, 0.1175, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 13:53:02,337 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:53:05,125 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:53:05,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:53:13,657 INFO [train.py:968] (0/2) Epoch 12, batch 45000, giga_loss[loss=0.2792, simple_loss=0.3549, pruned_loss=0.1018, over 28555.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.378, pruned_loss=0.1292, over 5635363.76 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3724, pruned_loss=0.1245, over 5701064.68 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3781, pruned_loss=0.1293, over 5639074.41 frames. ], batch size: 336, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:53:13,661 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 13:53:22,189 INFO [train.py:1012] (0/2) Epoch 12, validation: loss=0.2161, simple_loss=0.3243, pruned_loss=0.05396, over 944034.00 frames. 
+2023-03-06 13:53:22,190 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 13:53:38,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:53:41,177 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:53:52,998 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:54:05,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.954e+02 1.370e+03 1.827e+03 2.828e+03 7.741e+03, threshold=3.654e+03, percent-clipped=5.0
+2023-03-06 13:54:06,953 INFO [train.py:968] (0/2) Epoch 12, batch 45050, giga_loss[loss=0.2791, simple_loss=0.3426, pruned_loss=0.1078, over 28530.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3739, pruned_loss=0.1244, over 5648885.07 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3723, pruned_loss=0.1245, over 5704781.27 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3742, pruned_loss=0.1245, over 5647278.19 frames. ], batch size: 85, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:54:53,764 INFO [train.py:968] (0/2) Epoch 12, batch 45100, giga_loss[loss=0.3836, simple_loss=0.4155, pruned_loss=0.1759, over 26777.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3714, pruned_loss=0.1221, over 5648703.47 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3719, pruned_loss=0.124, over 5712610.10 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3721, pruned_loss=0.1226, over 5638171.87 frames. ], batch size: 555, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:55:17,690 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-06 13:55:38,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.620e+02 1.572e+03 2.113e+03 2.781e+03 6.524e+03, threshold=4.226e+03, percent-clipped=9.0
+2023-03-06 13:55:40,524 INFO [train.py:968] (0/2) Epoch 12, batch 45150, giga_loss[loss=0.3292, simple_loss=0.3824, pruned_loss=0.138, over 28696.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3711, pruned_loss=0.1221, over 5667318.14 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3717, pruned_loss=0.1239, over 5715485.40 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3719, pruned_loss=0.1225, over 5655043.24 frames. ], batch size: 242, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:55:46,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:55:49,228 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:55:52,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:56:11,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4915, 1.7490, 1.6150, 1.4631], device='cuda:0'), covar=tensor([0.2095, 0.1895, 0.1919, 0.1869], device='cuda:0'), in_proj_covar=tensor([0.1747, 0.1644, 0.1611, 0.1701], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 13:56:23,039 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:56:31,056 INFO [train.py:968] (0/2) Epoch 12, batch 45200, giga_loss[loss=0.3343, simple_loss=0.3887, pruned_loss=0.14, over 28286.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3679, pruned_loss=0.1208, over 5680649.88 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3712, pruned_loss=0.1236, over 5718819.83 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3688, pruned_loss=0.1213, over 5666880.47 frames. ], batch size: 368, lr: 2.63e-03, grad_scale: 8.0
+2023-03-06 13:57:03,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:57:11,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:57:14,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.822e+03 2.441e+03 3.646e+03 1.059e+04, threshold=4.881e+03, percent-clipped=17.0
+2023-03-06 13:57:16,475 INFO [train.py:968] (0/2) Epoch 12, batch 45250, giga_loss[loss=0.3488, simple_loss=0.4082, pruned_loss=0.1447, over 28942.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3686, pruned_loss=0.1212, over 5688920.18 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3706, pruned_loss=0.1232, over 5721784.28 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3698, pruned_loss=0.1219, over 5674782.21 frames. ], batch size: 213, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:58:02,173 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:03,112 INFO [train.py:968] (0/2) Epoch 12, batch 45300, giga_loss[loss=0.3704, simple_loss=0.4213, pruned_loss=0.1598, over 28926.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3711, pruned_loss=0.1226, over 5683278.31 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3706, pruned_loss=0.1232, over 5716369.89 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.372, pruned_loss=0.1231, over 5675953.22 frames. ], batch size: 227, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:58:04,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:06,244 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:09,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:19,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:29,757 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-06 13:58:31,518 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:51,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.607e+02 1.515e+03 1.967e+03 2.769e+03 5.769e+03, threshold=3.935e+03, percent-clipped=3.0
+2023-03-06 13:58:52,181 INFO [train.py:968] (0/2) Epoch 12, batch 45350, giga_loss[loss=0.2796, simple_loss=0.3537, pruned_loss=0.1027, over 28958.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3722, pruned_loss=0.1227, over 5677678.20 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3707, pruned_loss=0.1231, over 5720288.28 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3729, pruned_loss=0.1233, over 5667475.62 frames. ], batch size: 145, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:59:17,943 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:20,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:26,821 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:28,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:37,744 INFO [train.py:968] (0/2) Epoch 12, batch 45400, giga_loss[loss=0.3061, simple_loss=0.3711, pruned_loss=0.1206, over 28857.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3738, pruned_loss=0.124, over 5682001.37 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3714, pruned_loss=0.1234, over 5724569.88 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3739, pruned_loss=0.1241, over 5668840.14 frames. ], batch size: 227, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:59:44,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:47,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:54,255 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:22,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.747e+02 1.357e+03 1.788e+03 2.423e+03 5.139e+03, threshold=3.576e+03, percent-clipped=9.0
+2023-03-06 14:00:26,147 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:26,503 INFO [train.py:968] (0/2) Epoch 12, batch 45450, giga_loss[loss=0.3076, simple_loss=0.3546, pruned_loss=0.1303, over 23575.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3743, pruned_loss=0.125, over 5657078.28 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3717, pruned_loss=0.1237, over 5715687.21 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3741, pruned_loss=0.1249, over 5653597.22 frames. ], batch size: 705, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:00:28,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:32,772 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:35,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:37,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:53,856 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:01:05,126 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:01:13,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5512, 1.6226, 1.5089, 1.4327], device='cuda:0'), covar=tensor([0.1367, 0.1890, 0.2026, 0.1862], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0746, 0.0690, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 14:01:14,390 INFO [train.py:968] (0/2) Epoch 12, batch 45500, giga_loss[loss=0.2938, simple_loss=0.3667, pruned_loss=0.1105, over 28359.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3773, pruned_loss=0.1278, over 5647812.93 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3718, pruned_loss=0.1237, over 5718469.62 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.377, pruned_loss=0.1277, over 5641823.66 frames. ], batch size: 60, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:02:00,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.421e+02 1.647e+03 2.223e+03 2.840e+03 9.788e+03, threshold=4.447e+03, percent-clipped=14.0
+2023-03-06 14:02:00,532 INFO [train.py:968] (0/2) Epoch 12, batch 45550, giga_loss[loss=0.3129, simple_loss=0.3754, pruned_loss=0.1252, over 29038.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3784, pruned_loss=0.1282, over 5659140.53 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3714, pruned_loss=0.1236, over 5721493.05 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3787, pruned_loss=0.1283, over 5650515.43 frames. ], batch size: 128, lr: 2.63e-03, grad_scale: 2.0
+2023-03-06 14:02:02,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:02:04,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:02:32,976 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:02:49,432 INFO [train.py:968] (0/2) Epoch 12, batch 45600, giga_loss[loss=0.3254, simple_loss=0.3815, pruned_loss=0.1347, over 28679.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3789, pruned_loss=0.1287, over 5656411.00 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3714, pruned_loss=0.1235, over 5723915.39 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3793, pruned_loss=0.1289, over 5646719.44 frames. ], batch size: 242, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:03:41,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.551e+02 1.500e+03 1.899e+03 2.483e+03 6.198e+03, threshold=3.798e+03, percent-clipped=4.0
+2023-03-06 14:03:41,578 INFO [train.py:968] (0/2) Epoch 12, batch 45650, giga_loss[loss=0.2688, simple_loss=0.3434, pruned_loss=0.09711, over 28966.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3781, pruned_loss=0.1282, over 5663383.92 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3708, pruned_loss=0.123, over 5725313.20 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3791, pruned_loss=0.129, over 5652980.95 frames. ], batch size: 128, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:04:13,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:04:40,435 INFO [train.py:968] (0/2) Epoch 12, batch 45700, giga_loss[loss=0.3019, simple_loss=0.3575, pruned_loss=0.1231, over 28750.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3785, pruned_loss=0.1272, over 5654893.56 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3709, pruned_loss=0.1232, over 5726171.64 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3793, pruned_loss=0.1277, over 5645664.92 frames. ], batch size: 92, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:05:27,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.926e+02 1.828e+03 2.519e+03 4.117e+03 1.014e+04, threshold=5.038e+03, percent-clipped=30.0
+2023-03-06 14:05:27,732 INFO [train.py:968] (0/2) Epoch 12, batch 45750, giga_loss[loss=0.2833, simple_loss=0.3501, pruned_loss=0.1083, over 28839.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3777, pruned_loss=0.1264, over 5658985.98 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3704, pruned_loss=0.1231, over 5721611.44 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3789, pruned_loss=0.127, over 5654268.51 frames. ], batch size: 112, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:06:14,626 INFO [train.py:968] (0/2) Epoch 12, batch 45800, giga_loss[loss=0.283, simple_loss=0.3503, pruned_loss=0.1078, over 28926.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3774, pruned_loss=0.1272, over 5658774.84 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3706, pruned_loss=0.1232, over 5715383.44 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3783, pruned_loss=0.1276, over 5660006.90 frames. ], batch size: 136, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:06:40,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7186, 1.6938, 1.3315, 1.4390], device='cuda:0'), covar=tensor([0.0593, 0.0448, 0.0810, 0.0945], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0447, 0.0506, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 14:06:41,687 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=547013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:06:43,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=547016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:06:56,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=547025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:07:15,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.650e+02 1.441e+03 1.806e+03 2.614e+03 8.466e+03, threshold=3.613e+03, percent-clipped=2.0
+2023-03-06 14:07:15,823 INFO [train.py:968] (0/2) Epoch 12, batch 45850, giga_loss[loss=0.2688, simple_loss=0.3383, pruned_loss=0.09963, over 28379.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3763, pruned_loss=0.1268, over 5658398.92 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3709, pruned_loss=0.1235, over 5716368.52 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3768, pruned_loss=0.1269, over 5658113.22 frames. ], batch size: 71, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:07:19,928 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=547045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:08:07,710 INFO [train.py:968] (0/2) Epoch 12, batch 45900, giga_loss[loss=0.3111, simple_loss=0.3695, pruned_loss=0.1264, over 29155.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.375, pruned_loss=0.1263, over 5666441.84 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3704, pruned_loss=0.1231, over 5718787.87 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.376, pruned_loss=0.1268, over 5662363.55 frames. ], batch size: 113, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:08:56,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.345e+02 1.522e+03 1.860e+03 2.314e+03 5.801e+03, threshold=3.721e+03, percent-clipped=5.0
+2023-03-06 14:08:56,710 INFO [train.py:968] (0/2) Epoch 12, batch 45950, giga_loss[loss=0.3144, simple_loss=0.3813, pruned_loss=0.1237, over 28932.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.375, pruned_loss=0.1263, over 5660008.10 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3707, pruned_loss=0.1232, over 5716778.96 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3756, pruned_loss=0.1267, over 5657521.81 frames. ], batch size: 128, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:09:13,049 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=547161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:09:15,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1966, 2.6489, 2.2001, 1.8303], device='cuda:0'), covar=tensor([0.2157, 0.1538, 0.1620, 0.1906], device='cuda:0'), in_proj_covar=tensor([0.1768, 0.1659, 0.1621, 0.1715], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 14:09:19,214 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=547168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:09:22,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=547171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:09:41,067 INFO [train.py:968] (0/2) Epoch 12, batch 46000, giga_loss[loss=0.286, simple_loss=0.3539, pruned_loss=0.1091, over 28828.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1273, over 5668220.47 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3706, pruned_loss=0.1231, over 5721360.09 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3762, pruned_loss=0.1277, over 5660869.82 frames. ], batch size: 99, lr: 2.63e-03, grad_scale: 8.0
+2023-03-06 14:09:47,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=547200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:10:23,265 INFO [train.py:968] (0/2) Epoch 12, batch 46050, giga_loss[loss=0.2826, simple_loss=0.3588, pruned_loss=0.1032, over 28708.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3775, pruned_loss=0.1291, over 5656859.28 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3716, pruned_loss=0.1241, over 5708452.89 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3773, pruned_loss=0.1286, over 5661510.55 frames. ], batch size: 92, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:10:25,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.576e+02 1.716e+03 2.313e+03 3.149e+03 9.521e+03, threshold=4.627e+03, percent-clipped=15.0
+2023-03-06 14:11:19,811 INFO [train.py:968] (0/2) Epoch 12, batch 46100, giga_loss[loss=0.3444, simple_loss=0.397, pruned_loss=0.1459, over 28606.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3785, pruned_loss=0.1302, over 5657091.70 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3714, pruned_loss=0.1241, over 5709571.63 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3785, pruned_loss=0.1299, over 5659626.74 frames. ], batch size: 336, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:11:37,665 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-06 14:11:41,159 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-06 14:11:44,792 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 14:12:04,572 INFO [train.py:968] (0/2) Epoch 12, batch 46150, giga_loss[loss=0.3091, simple_loss=0.371, pruned_loss=0.1236, over 28867.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3787, pruned_loss=0.1302, over 5651929.31 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3716, pruned_loss=0.1242, over 5700143.05 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3787, pruned_loss=0.13, over 5661439.38 frames. ], batch size: 99, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:12:05,379 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.711e+03 2.235e+03 3.049e+03 9.213e+03, threshold=4.470e+03, percent-clipped=10.0
+2023-03-06 14:12:49,539 INFO [train.py:968] (0/2) Epoch 12, batch 46200, libri_loss[loss=0.4205, simple_loss=0.4368, pruned_loss=0.202, over 19309.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3798, pruned_loss=0.1319, over 5606837.92 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3723, pruned_loss=0.1248, over 5652118.35 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3794, pruned_loss=0.1314, over 5656806.30 frames. ], batch size: 186, lr: 2.63e-03, grad_scale: 2.0
+2023-03-06 14:13:33,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3348, 1.5417, 1.5776, 1.2084], device='cuda:0'), covar=tensor([0.1354, 0.2026, 0.1119, 0.1322], device='cuda:0'), in_proj_covar=tensor([0.0842, 0.0702, 0.0883, 0.0786], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 14:13:34,462 INFO [train.py:968] (0/2) Epoch 12, batch 46250, giga_loss[loss=0.3271, simple_loss=0.382, pruned_loss=0.1361, over 28839.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3808, pruned_loss=0.133, over 5570064.98 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3728, pruned_loss=0.1253, over 5601959.46 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3801, pruned_loss=0.1322, over 5655920.51 frames. ], batch size: 199, lr: 2.63e-03, grad_scale: 2.0
+2023-03-06 14:13:36,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.770e+03 2.327e+03 3.657e+03 9.301e+03, threshold=4.654e+03, percent-clipped=11.0
+2023-03-06 14:13:48,930 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9983, 1.1217, 1.1277, 1.0139], device='cuda:0'), covar=tensor([0.1583, 0.1984, 0.1038, 0.1314], device='cuda:0'), in_proj_covar=tensor([0.1771, 0.1671, 0.1630, 0.1723], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 14:13:55,283 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-06 14:13:57,928 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-12.pt
+2023-03-06 14:14:58,064 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-06 14:15:15,478 INFO [train.py:968] (0/2) Epoch 13, batch 50, giga_loss[loss=0.3151, simple_loss=0.3892, pruned_loss=0.1205, over 28855.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3793, pruned_loss=0.1141, over 1263389.71 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3504, pruned_loss=0.09745, over 116882.34 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3819, pruned_loss=0.1156, over 1170121.25 frames. ], batch size: 199, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:15:28,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5093, 1.6697, 1.5074, 1.4473], device='cuda:0'), covar=tensor([0.1404, 0.2094, 0.1926, 0.2007], device='cuda:0'), in_proj_covar=tensor([0.0439, 0.0743, 0.0685, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 14:15:34,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=547536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:15:44,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.268e+02 1.312e+03 1.601e+03 2.255e+03 6.771e+03, threshold=3.203e+03, percent-clipped=6.0
+2023-03-06 14:16:04,146 INFO [train.py:968] (0/2) Epoch 13, batch 100, giga_loss[loss=0.2988, simple_loss=0.3655, pruned_loss=0.116, over 28710.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3696, pruned_loss=0.1096, over 2240694.96 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3433, pruned_loss=0.09364, over 250039.13 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3726, pruned_loss=0.1113, over 2082669.07 frames. ], batch size: 92, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:16:50,974 INFO [train.py:968] (0/2) Epoch 13, batch 150, giga_loss[loss=0.2339, simple_loss=0.3113, pruned_loss=0.07831, over 28625.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3559, pruned_loss=0.1039, over 2995085.69 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3467, pruned_loss=0.09504, over 296387.99 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3569, pruned_loss=0.1047, over 2850485.06 frames. ], batch size: 307, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:16:58,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=547624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:17:14,208 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.712e+02 1.080e+03 1.539e+03 2.188e+03 6.145e+03, threshold=3.078e+03, percent-clipped=8.0
+2023-03-06 14:17:30,622 INFO [train.py:968] (0/2) Epoch 13, batch 200, giga_loss[loss=0.2274, simple_loss=0.3044, pruned_loss=0.0752, over 28876.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3415, pruned_loss=0.0964, over 3604310.49 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3464, pruned_loss=0.09478, over 407338.52 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3417, pruned_loss=0.09683, over 3443630.13 frames. ], batch size: 112, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:17:43,108 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=547679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:17:45,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=547682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:17:53,271 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 14:18:11,654 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=547711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:18:15,219 INFO [train.py:968] (0/2) Epoch 13, batch 250, giga_loss[loss=0.2488, simple_loss=0.3252, pruned_loss=0.08617, over 28830.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3299, pruned_loss=0.09029, over 4067224.78 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.345, pruned_loss=0.09339, over 511408.61 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3295, pruned_loss=0.09048, over 3907320.81 frames. ], batch size: 174, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:18:40,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.912e+02 8.962e+02 1.097e+03 1.502e+03 6.448e+03, threshold=2.194e+03, percent-clipped=5.0
+2023-03-06 14:18:58,921 INFO [train.py:968] (0/2) Epoch 13, batch 300, giga_loss[loss=0.272, simple_loss=0.3295, pruned_loss=0.1073, over 27589.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.321, pruned_loss=0.08619, over 4425038.77 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3503, pruned_loss=0.09768, over 609186.09 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3194, pruned_loss=0.08557, over 4277498.65 frames. ], batch size: 472, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:19:42,613 INFO [train.py:968] (0/2) Epoch 13, batch 350, giga_loss[loss=0.2381, simple_loss=0.3075, pruned_loss=0.08438, over 28820.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3163, pruned_loss=0.08443, over 4708345.92 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3525, pruned_loss=0.09855, over 810597.58 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3131, pruned_loss=0.08318, over 4546605.49 frames. ], batch size: 186, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:19:58,274 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=547835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:20:05,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.945e+02 1.066e+03 1.344e+03 1.910e+03 6.467e+03, threshold=2.687e+03, percent-clipped=19.0
+2023-03-06 14:20:14,046 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=547854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:20:24,071 INFO [train.py:968] (0/2) Epoch 13, batch 400, giga_loss[loss=0.2047, simple_loss=0.2762, pruned_loss=0.06663, over 28429.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.3113, pruned_loss=0.08221, over 4929187.65 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3511, pruned_loss=0.09774, over 885423.26 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.3083, pruned_loss=0.08111, over 4788203.75 frames. ], batch size: 71, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:21:01,044 INFO [train.py:968] (0/2) Epoch 13, batch 450, giga_loss[loss=0.2084, simple_loss=0.2863, pruned_loss=0.06525, over 28945.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.311, pruned_loss=0.0819, over 5108982.58 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3527, pruned_loss=0.09851, over 1127815.63 frames. ], giga_tot_loss[loss=0.2332, simple_loss=0.3063, pruned_loss=0.08004, over 4956810.65 frames. ], batch size: 164, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:21:13,646 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0380, 2.0973, 1.5259, 1.6820], device='cuda:0'), covar=tensor([0.0855, 0.0724, 0.1060, 0.1210], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0442, 0.0503, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 14:21:22,801 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 14:21:23,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3820, 3.0793, 1.4945, 1.4897], device='cuda:0'), covar=tensor([0.0961, 0.0312, 0.0883, 0.1352], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0514, 0.0349, 0.0396], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 14:21:28,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.943e+02 1.050e+03 1.390e+03 2.055e+03 5.500e+03, threshold=2.780e+03, percent-clipped=10.0
+2023-03-06 14:21:47,848 INFO [train.py:968] (0/2) Epoch 13, batch 500, giga_loss[loss=0.2294, simple_loss=0.2999, pruned_loss=0.07942, over 28509.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3076, pruned_loss=0.08037, over 5243030.57 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3524, pruned_loss=0.09853, over 1175706.47 frames. ], giga_tot_loss[loss=0.2305, simple_loss=0.3036, pruned_loss=0.07873, over 5116812.86 frames. ], batch size: 336, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:22:17,395 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=547999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:22:17,822 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-548000.pt
+2023-03-06 14:22:31,533 INFO [train.py:968] (0/2) Epoch 13, batch 550, giga_loss[loss=0.2009, simple_loss=0.2796, pruned_loss=0.06113, over 28940.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3058, pruned_loss=0.07946, over 5350975.05 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3537, pruned_loss=0.09957, over 1292174.71 frames. ], giga_tot_loss[loss=0.228, simple_loss=0.3011, pruned_loss=0.07744, over 5237173.53 frames. ], batch size: 213, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:22:41,034 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 14:22:44,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8561, 3.6461, 3.4499, 1.6820], device='cuda:0'), covar=tensor([0.0670, 0.0873, 0.0823, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.1063, 0.0987, 0.0862, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 14:22:58,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.797e+02 9.888e+02 1.326e+03 1.768e+03 6.511e+03, threshold=2.653e+03, percent-clipped=6.0
+2023-03-06 14:23:10,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 14:23:18,032 INFO [train.py:968] (0/2) Epoch 13, batch 600, giga_loss[loss=0.2181, simple_loss=0.2883, pruned_loss=0.07393, over 28519.00 frames. ], tot_loss[loss=0.23, simple_loss=0.303, pruned_loss=0.07847, over 5424790.37 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.353, pruned_loss=0.09893, over 1338452.44 frames. ], giga_tot_loss[loss=0.2264, simple_loss=0.2991, pruned_loss=0.07684, over 5330712.36 frames. ], batch size: 336, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:24:04,560 INFO [train.py:968] (0/2) Epoch 13, batch 650, giga_loss[loss=0.2095, simple_loss=0.2829, pruned_loss=0.06809, over 28835.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3014, pruned_loss=0.07768, over 5469371.36 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.353, pruned_loss=0.09931, over 1441643.01 frames. ], giga_tot_loss[loss=0.2243, simple_loss=0.297, pruned_loss=0.07577, over 5393211.71 frames. ], batch size: 99, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:24:05,737 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-06 14:24:07,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7096, 1.8641, 1.5774, 1.8726], device='cuda:0'), covar=tensor([0.2418, 0.2489, 0.2713, 0.2539], device='cuda:0'), in_proj_covar=tensor([0.1342, 0.0989, 0.1182, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 14:24:25,718 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=548142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:24:27,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.403e+02 9.529e+02 1.244e+03 1.872e+03 6.441e+03, threshold=2.487e+03, percent-clipped=8.0
+2023-03-06 14:24:28,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=548145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:24:46,194 INFO [train.py:968] (0/2) Epoch 13, batch 700, giga_loss[loss=0.1826, simple_loss=0.2597, pruned_loss=0.05273, over 28306.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.2997, pruned_loss=0.07666, over 5526222.00 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3528, pruned_loss=0.09867, over 1615834.63 frames. ], giga_tot_loss[loss=0.2216, simple_loss=0.2942, pruned_loss=0.07445, over 5451194.42 frames. ], batch size: 65, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:24:52,889 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=548174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:25:19,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2005, 1.7546, 1.4775, 0.3590], device='cuda:0'), covar=tensor([0.3741, 0.2399, 0.3511, 0.5110], device='cuda:0'), in_proj_covar=tensor([0.1591, 0.1515, 0.1500, 0.1308], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 14:25:25,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:25:29,930 INFO [train.py:968] (0/2) Epoch 13, batch 750, libri_loss[loss=0.2859, simple_loss=0.3687, pruned_loss=0.1015, over 29763.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.2968, pruned_loss=0.07507, over 5570470.42 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3527, pruned_loss=0.0982, over 1697963.93 frames. ], giga_tot_loss[loss=0.2187, simple_loss=0.2914, pruned_loss=0.07299, over 5507189.54 frames. ], batch size: 87, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:25:41,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:25:54,045 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.245e+02 1.013e+03 1.246e+03 1.843e+03 4.617e+03, threshold=2.493e+03, percent-clipped=8.0
+2023-03-06 14:26:06,275 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-06 14:26:13,084 INFO [train.py:968] (0/2) Epoch 13, batch 800, giga_loss[loss=0.1849, simple_loss=0.261, pruned_loss=0.05437, over 28896.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2951, pruned_loss=0.0747, over 5594081.71 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3541, pruned_loss=0.09903, over 1777778.17 frames. ], giga_tot_loss[loss=0.217, simple_loss=0.2893, pruned_loss=0.07234, over 5541500.81 frames. ], batch size: 213, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:26:43,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3169, 1.5960, 1.3240, 0.9947], device='cuda:0'), covar=tensor([0.2328, 0.2319, 0.2604, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.1335, 0.0983, 0.1178, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 14:26:51,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5740, 2.1711, 1.7141, 0.7467], device='cuda:0'), covar=tensor([0.4275, 0.2396, 0.3057, 0.4936], device='cuda:0'), in_proj_covar=tensor([0.1591, 0.1512, 0.1500, 0.1309], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 14:26:55,300 INFO [train.py:968] (0/2) Epoch 13, batch 850, giga_loss[loss=0.2761, simple_loss=0.3553, pruned_loss=0.09849, over 28827.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3044, pruned_loss=0.07962, over 5608896.51 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3527, pruned_loss=0.09798, over 1949099.13 frames. ], giga_tot_loss[loss=0.2262, simple_loss=0.298, pruned_loss=0.07719, over 5563941.43 frames. ], batch size: 119, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:27:08,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5805, 1.7791, 1.9070, 1.4005], device='cuda:0'), covar=tensor([0.1626, 0.2303, 0.1302, 0.1541], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0703, 0.0895, 0.0797], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-06 14:27:19,641 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-06 14:27:24,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.464e+02 1.111e+03 1.393e+03 1.749e+03 4.169e+03, threshold=2.786e+03, percent-clipped=7.0
+2023-03-06 14:27:31,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=548353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:27:31,281 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=548353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:27:35,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=548356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:27:42,080 INFO [train.py:968] (0/2) Epoch 13, batch 900, giga_loss[loss=0.2705, simple_loss=0.3556, pruned_loss=0.09264, over 28893.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3177, pruned_loss=0.08634, over 5617576.75 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3518, pruned_loss=0.09729, over 2052590.44 frames. ], giga_tot_loss[loss=0.2402, simple_loss=0.3119, pruned_loss=0.08427, over 5586402.39 frames. ], batch size: 174, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:27:48,452 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=548372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:27:51,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=548375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:28:00,597 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=548385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:28:14,199 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=548404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:28:24,156 INFO [train.py:968] (0/2) Epoch 13, batch 950, giga_loss[loss=0.2807, simple_loss=0.3557, pruned_loss=0.1028, over 28789.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3288, pruned_loss=0.09157, over 5636966.45 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3506, pruned_loss=0.09671, over 2193400.03 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3235, pruned_loss=0.08985, over 5611058.20 frames. ], batch size: 92, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:28:48,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.818e+02 1.220e+03 1.742e+03 2.368e+03 7.281e+03, threshold=3.484e+03, percent-clipped=13.0
+2023-03-06 14:29:06,851 INFO [train.py:968] (0/2) Epoch 13, batch 1000, libri_loss[loss=0.2888, simple_loss=0.3618, pruned_loss=0.1079, over 29222.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3372, pruned_loss=0.0953, over 5656276.65 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3502, pruned_loss=0.09657, over 2285084.15 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3328, pruned_loss=0.09392, over 5630018.72 frames. ], batch size: 97, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:29:31,509 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 14:29:37,993 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=548504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:29:46,507 INFO [train.py:968] (0/2) Epoch 13, batch 1050, giga_loss[loss=0.2674, simple_loss=0.3458, pruned_loss=0.09449, over 28876.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3408, pruned_loss=0.09585, over 5670730.81 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3501, pruned_loss=0.09642, over 2371759.87 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3371, pruned_loss=0.09481, over 5646716.59 frames. ], batch size: 112, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:30:16,973 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.648e+02 1.005e+03 1.324e+03 1.927e+03 6.577e+03, threshold=2.648e+03, percent-clipped=5.0
+2023-03-06 14:30:32,003 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=548561.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:30:34,348 INFO [train.py:968] (0/2) Epoch 13, batch 1100, giga_loss[loss=0.2793, simple_loss=0.3543, pruned_loss=0.1021, over 28881.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3427, pruned_loss=0.09612, over 5664130.55 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3498, pruned_loss=0.0961, over 2388006.07 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.34, pruned_loss=0.09544, over 5645558.90 frames. ], batch size: 186, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:31:16,796 INFO [train.py:968] (0/2) Epoch 13, batch 1150, giga_loss[loss=0.2928, simple_loss=0.3641, pruned_loss=0.1107, over 28686.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3448, pruned_loss=0.09769, over 5674685.86 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3493, pruned_loss=0.09601, over 2430436.06 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3428, pruned_loss=0.09718, over 5666183.62 frames. ], batch size: 262, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:31:18,881 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=548615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:31:45,233 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.073e+02 1.194e+03 1.519e+03 1.924e+03 4.704e+03, threshold=3.038e+03, percent-clipped=8.0
+2023-03-06 14:31:47,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4467, 1.4705, 1.4168, 1.4022], device='cuda:0'), covar=tensor([0.1729, 0.1894, 0.1248, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.1747, 0.1656, 0.1621, 0.1712], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 14:31:53,243 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 14:32:02,586 INFO [train.py:968] (0/2) Epoch 13, batch 1200, libri_loss[loss=0.2474, simple_loss=0.3329, pruned_loss=0.08092, over 29555.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3477, pruned_loss=0.1003, over 5651570.31 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.35, pruned_loss=0.09653, over 2479312.99 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3458, pruned_loss=0.09972, over 5656921.09 frames. ], batch size: 84, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:32:07,207 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 14:32:35,974 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6326, 2.8281, 2.5048, 2.5484], device='cuda:0'), covar=tensor([0.1366, 0.1514, 0.1559, 0.1547], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0735, 0.0678, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 14:32:43,839 INFO [train.py:968] (0/2) Epoch 13, batch 1250, giga_loss[loss=0.2783, simple_loss=0.3506, pruned_loss=0.103, over 29036.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3502, pruned_loss=0.1018, over 5660200.22 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3497, pruned_loss=0.09608, over 2560348.35 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3488, pruned_loss=0.1016, over 5662255.71 frames. ], batch size: 128, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:32:54,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:33:10,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.118e+02 1.211e+03 1.436e+03 1.746e+03 3.737e+03, threshold=2.872e+03, percent-clipped=5.0
+2023-03-06 14:33:26,187 INFO [train.py:968] (0/2) Epoch 13, batch 1300, libri_loss[loss=0.2324, simple_loss=0.3094, pruned_loss=0.07775, over 29664.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3531, pruned_loss=0.1026, over 5672565.94 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3501, pruned_loss=0.09632, over 2691768.16 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3518, pruned_loss=0.1027, over 5665801.71 frames. ], batch size: 73, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:34:05,660 INFO [train.py:968] (0/2) Epoch 13, batch 1350, giga_loss[loss=0.2673, simple_loss=0.3508, pruned_loss=0.09186, over 28887.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3547, pruned_loss=0.1024, over 5683389.92 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.35, pruned_loss=0.0962, over 2729943.42 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3539, pruned_loss=0.1026, over 5684116.17 frames. ], batch size: 227, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:34:12,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 14:34:33,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.709e+02 1.154e+03 1.367e+03 1.901e+03 4.165e+03, threshold=2.733e+03, percent-clipped=3.0
+2023-03-06 14:34:48,435 INFO [train.py:968] (0/2) Epoch 13, batch 1400, giga_loss[loss=0.2789, simple_loss=0.3657, pruned_loss=0.0961, over 28726.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3556, pruned_loss=0.1025, over 5681866.57 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3492, pruned_loss=0.09586, over 2805916.51 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3554, pruned_loss=0.103, over 5679605.86 frames. ], batch size: 284, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:34:52,662 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=548871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:34:54,506 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=548874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:34:58,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:35:18,843 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=548903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:35:30,004 INFO [train.py:968] (0/2) Epoch 13, batch 1450, giga_loss[loss=0.2665, simple_loss=0.3482, pruned_loss=0.09245, over 28797.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3542, pruned_loss=0.1005, over 5691944.41 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3491, pruned_loss=0.0959, over 2897818.26 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3543, pruned_loss=0.101, over 5684118.59 frames. ], batch size: 199, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:35:47,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548936.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 14:35:53,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.051e+02 1.056e+03 1.252e+03 1.584e+03 3.676e+03, threshold=2.505e+03, percent-clipped=7.0
+2023-03-06 14:36:09,637 INFO [train.py:968] (0/2) Epoch 13, batch 1500, giga_loss[loss=0.3218, simple_loss=0.3816, pruned_loss=0.131, over 27954.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3527, pruned_loss=0.09844, over 5701343.45 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3492, pruned_loss=0.09598, over 2971441.66 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3528, pruned_loss=0.09887, over 5691075.60 frames. ], batch size: 412, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:36:28,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:36:37,216 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-06 14:36:46,405 INFO [train.py:968] (0/2) Epoch 13, batch 1550, giga_loss[loss=0.2706, simple_loss=0.3514, pruned_loss=0.09488, over 28777.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3516, pruned_loss=0.09737, over 5706015.40 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3495, pruned_loss=0.09621, over 3098692.59 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3517, pruned_loss=0.09768, over 5691471.77 frames. ], batch size: 119, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:36:52,912 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=549022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:36:55,014 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=549025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:37:13,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.451e+02 1.048e+03 1.411e+03 2.014e+03 5.914e+03, threshold=2.821e+03, percent-clipped=12.0
+2023-03-06 14:37:21,709 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=549054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:37:29,193 INFO [train.py:968] (0/2) Epoch 13, batch 1600, giga_loss[loss=0.2772, simple_loss=0.3518, pruned_loss=0.1013, over 29140.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3528, pruned_loss=0.09926, over 5716452.24 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3502, pruned_loss=0.09652, over 3181305.92 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3526, pruned_loss=0.09941, over 5700059.99 frames. ], batch size: 155, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:37:30,144 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9356, 1.2620, 1.1027, 0.1375], device='cuda:0'), covar=tensor([0.2738, 0.2244, 0.3218, 0.4607], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1490, 0.1493, 0.1295], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 14:37:40,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=549079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:37:43,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=549082.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:38:05,859 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=549111.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 14:38:08,119 INFO [train.py:968] (0/2) Epoch 13, batch 1650, giga_loss[loss=0.3024, simple_loss=0.3652, pruned_loss=0.1198, over 28230.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3568, pruned_loss=0.1044, over 5719329.48 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3509, pruned_loss=0.09683, over 3271779.83 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3565, pruned_loss=0.1045, over 5703052.07 frames. ], batch size: 77, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:38:23,537 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=549133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:38:27,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=549136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:38:36,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.797e+02 1.274e+03 1.657e+03 2.181e+03 6.961e+03, threshold=3.314e+03, percent-clipped=12.0
+2023-03-06 14:38:44,557 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6142, 1.8625, 1.9196, 1.4697], device='cuda:0'), covar=tensor([0.1669, 0.2175, 0.1294, 0.1536], device='cuda:0'), in_proj_covar=tensor([0.0847, 0.0696, 0.0888, 0.0792], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 14:38:53,124 INFO [train.py:968] (0/2) Epoch 13, batch 1700, giga_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.114, over 28670.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.359, pruned_loss=0.108, over 5705650.45 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3511, pruned_loss=0.097, over 3297785.49 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3586, pruned_loss=0.1081, over 5691591.23 frames. ], batch size: 262, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:38:53,378 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=549165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:39:25,906 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=549203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:39:38,849 INFO [train.py:968] (0/2) Epoch 13, batch 1750, libri_loss[loss=0.2821, simple_loss=0.3629, pruned_loss=0.1006, over 29287.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3579, pruned_loss=0.1083, over 5706497.38 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3514, pruned_loss=0.09711, over 3348902.08 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3576, pruned_loss=0.1085, over 5692067.28 frames. ], batch size: 97, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:40:06,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.011e+02 1.191e+03 1.532e+03 1.993e+03 4.750e+03, threshold=3.063e+03, percent-clipped=4.0
+2023-03-06 14:40:21,416 INFO [train.py:968] (0/2) Epoch 13, batch 1800, giga_loss[loss=0.2826, simple_loss=0.3388, pruned_loss=0.1132, over 23778.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3558, pruned_loss=0.1075, over 5708168.34 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3511, pruned_loss=0.09688, over 3407247.17 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3558, pruned_loss=0.108, over 5697014.03 frames. ], batch size: 705, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:40:23,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2967, 1.2011, 4.6449, 3.3259], device='cuda:0'), covar=tensor([0.1705, 0.2725, 0.0330, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0666, 0.0593, 0.0861, 0.0778], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 14:41:04,380 INFO [train.py:968] (0/2) Epoch 13, batch 1850, libri_loss[loss=0.2262, simple_loss=0.3067, pruned_loss=0.07286, over 29351.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3545, pruned_loss=0.1058, over 5716687.45 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3505, pruned_loss=0.09653, over 3481302.47 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.355, pruned_loss=0.1067, over 5702451.43 frames. ], batch size: 71, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:41:04,752 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7447, 1.8244, 1.3447, 1.4930], device='cuda:0'), covar=tensor([0.0880, 0.0723, 0.1023, 0.1253], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0442, 0.0507, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 14:41:18,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4040, 1.7557, 1.4765, 1.6606], device='cuda:0'), covar=tensor([0.0756, 0.0292, 0.0300, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 14:41:29,660 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.188e+02 1.152e+03 1.562e+03 2.180e+03 8.128e+03, threshold=3.123e+03, percent-clipped=14.0
+2023-03-06 14:41:31,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=549349.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:41:48,560 INFO [train.py:968] (0/2) Epoch 13, batch 1900, giga_loss[loss=0.2683, simple_loss=0.3444, pruned_loss=0.09614, over 28970.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3524, pruned_loss=0.1037, over 5701995.01 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3503, pruned_loss=0.09639, over 3552534.88 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.353, pruned_loss=0.1048, over 5696031.30 frames. ], batch size: 136, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:42:35,460 INFO [train.py:968] (0/2) Epoch 13, batch 1950, giga_loss[loss=0.2375, simple_loss=0.3195, pruned_loss=0.07778, over 28874.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.349, pruned_loss=0.1012, over 5697817.83 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.351, pruned_loss=0.09667, over 3598620.85 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3491, pruned_loss=0.1021, over 5689980.71 frames. ], batch size: 186, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:43:07,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.694e+02 1.056e+03 1.345e+03 1.797e+03 4.289e+03, threshold=2.691e+03, percent-clipped=2.0
+2023-03-06 14:43:22,683 INFO [train.py:968] (0/2) Epoch 13, batch 2000, giga_loss[loss=0.2317, simple_loss=0.308, pruned_loss=0.07774, over 28411.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3422, pruned_loss=0.09726, over 5691859.34 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3506, pruned_loss=0.09631, over 3655708.95 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3424, pruned_loss=0.09817, over 5680757.11 frames. ], batch size: 369, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:44:06,047 INFO [train.py:968] (0/2) Epoch 13, batch 2050, giga_loss[loss=0.3198, simple_loss=0.3644, pruned_loss=0.1376, over 28247.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3387, pruned_loss=0.09579, over 5690672.66 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3511, pruned_loss=0.09656, over 3751440.65 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3381, pruned_loss=0.09639, over 5676273.59 frames. ], batch size: 368, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:44:37,365 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.363e+02 9.653e+02 1.200e+03 1.595e+03 4.948e+03, threshold=2.399e+03, percent-clipped=12.0
+2023-03-06 14:44:56,608 INFO [train.py:968] (0/2) Epoch 13, batch 2100, giga_loss[loss=0.2757, simple_loss=0.3526, pruned_loss=0.09943, over 28873.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3355, pruned_loss=0.09404, over 5691506.51 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.351, pruned_loss=0.0965, over 3772950.31 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.335, pruned_loss=0.09452, over 5678205.11 frames. ], batch size: 136, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:44:57,713 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-06 14:45:05,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=549578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:45:29,356 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-06 14:45:34,729 INFO [train.py:968] (0/2) Epoch 13, batch 2150, giga_loss[loss=0.2571, simple_loss=0.3274, pruned_loss=0.09337, over 28696.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.336, pruned_loss=0.09389, over 5698723.89 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3496, pruned_loss=0.09555, over 3863721.44 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3358, pruned_loss=0.09477, over 5682410.44 frames. ], batch size: 78, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:45:59,958 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.843e+02 1.008e+03 1.271e+03 1.603e+03 3.454e+03, threshold=2.543e+03, percent-clipped=2.0
+2023-03-06 14:46:14,744 INFO [train.py:968] (0/2) Epoch 13, batch 2200, giga_loss[loss=0.2875, simple_loss=0.3661, pruned_loss=0.1044, over 27951.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3367, pruned_loss=0.09431, over 5693785.57 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3503, pruned_loss=0.09595, over 3902878.54 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3359, pruned_loss=0.09472, over 5686734.83 frames. ], batch size: 412, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:46:57,664 INFO [train.py:968] (0/2) Epoch 13, batch 2250, libri_loss[loss=0.2987, simple_loss=0.372, pruned_loss=0.1127, over 20505.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3358, pruned_loss=0.09384, over 5685308.01 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3512, pruned_loss=0.09635, over 3952572.66 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.334, pruned_loss=0.09385, over 5690780.97 frames. ], batch size: 187, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:47:02,083 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=549721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:47:04,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=549724.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 14:47:04,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=549724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:47:24,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.981e+02 9.990e+02 1.195e+03 1.493e+03 3.390e+03, threshold=2.390e+03, percent-clipped=4.0
+2023-03-06 14:47:29,039 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=549753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:47:38,215 INFO [train.py:968] (0/2) Epoch 13, batch 2300, giga_loss[loss=0.2505, simple_loss=0.3249, pruned_loss=0.0881, over 28562.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3331, pruned_loss=0.09235, over 5699777.76 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3499, pruned_loss=0.09535, over 4019281.87 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.332, pruned_loss=0.09292, over 5697874.62 frames. ], batch size: 336, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:47:54,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=549785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:48:18,385 INFO [train.py:968] (0/2) Epoch 13, batch 2350, giga_loss[loss=0.2271, simple_loss=0.3042, pruned_loss=0.07497, over 28914.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3312, pruned_loss=0.09154, over 5702091.60 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3504, pruned_loss=0.09564, over 4066113.09 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3295, pruned_loss=0.09172, over 5703927.64 frames. ], batch size: 186, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:48:47,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.404e+02 1.014e+03 1.413e+03 2.051e+03 9.608e+03, threshold=2.826e+03, percent-clipped=20.0
+2023-03-06 14:49:01,036 INFO [train.py:968] (0/2) Epoch 13, batch 2400, giga_loss[loss=0.2437, simple_loss=0.3196, pruned_loss=0.08395, over 28877.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3285, pruned_loss=0.09055, over 5698856.09 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3509, pruned_loss=0.09604, over 4072300.58 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3267, pruned_loss=0.09039, over 5709926.68 frames. ], batch size: 227, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:49:02,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=549867.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 14:49:07,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=549870.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:49:28,159 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=549899.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:49:43,284 INFO [train.py:968] (0/2) Epoch 13, batch 2450, libri_loss[loss=0.2389, simple_loss=0.3213, pruned_loss=0.07821, over 29501.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3268, pruned_loss=0.08956, over 5712487.56 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3511, pruned_loss=0.09603, over 4134725.36 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3244, pruned_loss=0.08922, over 5715671.92 frames. ], batch size: 70, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:49:53,010 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=549929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:50:06,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.162e+02 9.367e+02 1.155e+03 1.478e+03 5.332e+03, threshold=2.311e+03, percent-clipped=4.0
+2023-03-06 14:50:20,003 INFO [train.py:968] (0/2) Epoch 13, batch 2500, giga_loss[loss=0.219, simple_loss=0.3001, pruned_loss=0.06895, over 28790.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3236, pruned_loss=0.08798, over 5721370.79 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3511, pruned_loss=0.09593, over 4152446.66 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3215, pruned_loss=0.0877, over 5722156.94 frames. ], batch size: 119, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:50:47,034 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-06 14:50:50,146 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-550000.pt
+2023-03-06 14:51:03,518 INFO [train.py:968] (0/2) Epoch 13, batch 2550, giga_loss[loss=0.2534, simple_loss=0.3277, pruned_loss=0.08951, over 28301.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3212, pruned_loss=0.0871, over 5712822.96 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3513, pruned_loss=0.09591, over 4161057.43 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3193, pruned_loss=0.08687, over 5712792.49 frames. ], batch size: 368, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:51:30,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.437e+02 9.032e+02 1.053e+03 1.555e+03 3.613e+03, threshold=2.107e+03, percent-clipped=8.0
+2023-03-06 14:51:43,314 INFO [train.py:968] (0/2) Epoch 13, batch 2600, giga_loss[loss=0.2454, simple_loss=0.3215, pruned_loss=0.0847, over 28884.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.32, pruned_loss=0.08592, over 5722584.70 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3514, pruned_loss=0.09549, over 4218732.40 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3175, pruned_loss=0.08569, over 5718880.05 frames. ], batch size: 174, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:51:57,075 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:52:16,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9340, 1.1400, 1.0559, 0.8675], device='cuda:0'), covar=tensor([0.1952, 0.2011, 0.1118, 0.1649], device='cuda:0'), in_proj_covar=tensor([0.1739, 0.1642, 0.1611, 0.1712], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 14:52:22,440 INFO [train.py:968] (0/2) Epoch 13, batch 2650, giga_loss[loss=0.225, simple_loss=0.3007, pruned_loss=0.07468, over 28733.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3194, pruned_loss=0.08552, over 5717227.51 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3515, pruned_loss=0.0954, over 4259908.73 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3165, pruned_loss=0.08515, over 5716865.26 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:52:52,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.428e+02 9.793e+02 1.221e+03 1.702e+03 5.002e+03, threshold=2.442e+03, percent-clipped=13.0
+2023-03-06 14:53:01,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:53:06,394 INFO [train.py:968] (0/2) Epoch 13, batch 2700, giga_loss[loss=0.2033, simple_loss=0.285, pruned_loss=0.06086, over 28356.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3217, pruned_loss=0.08697, over 5711227.58 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3519, pruned_loss=0.09546, over 4292572.23 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3185, pruned_loss=0.08646, over 5707485.40 frames. ], batch size: 65, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:53:08,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6099, 4.2731, 1.7558, 1.6333], device='cuda:0'), covar=tensor([0.0971, 0.0269, 0.0872, 0.1337], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0503, 0.0344, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 14:53:16,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7325, 1.7658, 1.3595, 1.3255], device='cuda:0'), covar=tensor([0.0777, 0.0590, 0.0952, 0.1081], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0439, 0.0503, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 14:53:28,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7577, 1.9956, 1.7867, 1.7499], device='cuda:0'), covar=tensor([0.1648, 0.2022, 0.1830, 0.1882], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0735, 0.0680, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 14:53:52,288 INFO [train.py:968] (0/2) Epoch 13, batch 2750, giga_loss[loss=0.2695, simple_loss=0.34, pruned_loss=0.09946, over 28987.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3272, pruned_loss=0.09078, over 5713020.22 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.352, pruned_loss=0.09546, over 4308668.86 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3243, pruned_loss=0.09032, over 5708225.82 frames. ], batch size: 136, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:54:17,480 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:54:24,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.790e+02 1.069e+03 1.324e+03 1.815e+03 4.118e+03, threshold=2.648e+03, percent-clipped=8.0
+2023-03-06 14:54:38,872 INFO [train.py:968] (0/2) Epoch 13, batch 2800, giga_loss[loss=0.3411, simple_loss=0.401, pruned_loss=0.1406, over 28777.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3348, pruned_loss=0.09584, over 5704580.01 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3522, pruned_loss=0.09542, over 4354479.70 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3318, pruned_loss=0.09544, over 5696353.81 frames. ], batch size: 284, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:54:42,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2746, 4.1023, 3.8519, 1.8778], device='cuda:0'), covar=tensor([0.0542, 0.0689, 0.0682, 0.2093], device='cuda:0'), in_proj_covar=tensor([0.1049, 0.0972, 0.0851, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 14:55:07,968 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550293.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:55:16,521 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:55:17,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:55:18,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:55:26,531 INFO [train.py:968] (0/2) Epoch 13, batch 2850, giga_loss[loss=0.2883, simple_loss=0.3608, pruned_loss=0.1079, over 28874.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3428, pruned_loss=0.101, over 5694584.40 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3523, pruned_loss=0.09541, over 4376504.37 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3402, pruned_loss=0.1008, over 5686333.26 frames. ], batch size: 199, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:55:45,778 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:55:57,106 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.905e+02 1.363e+03 1.727e+03 2.313e+03 5.120e+03, threshold=3.454e+03, percent-clipped=14.0
+2023-03-06 14:55:57,398 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:56:12,907 INFO [train.py:968] (0/2) Epoch 13, batch 2900, giga_loss[loss=0.3139, simple_loss=0.3813, pruned_loss=0.1232, over 29002.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3478, pruned_loss=0.1028, over 5690196.36 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.352, pruned_loss=0.09533, over 4412900.57 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3458, pruned_loss=0.1028, over 5680118.25 frames. ], batch size: 106, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:56:38,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6128, 1.7122, 1.2993, 1.4332], device='cuda:0'), covar=tensor([0.0712, 0.0419, 0.0963, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0438, 0.0502, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 14:56:58,329 INFO [train.py:968] (0/2) Epoch 13, batch 2950, giga_loss[loss=0.2828, simple_loss=0.3595, pruned_loss=0.1031, over 29063.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3526, pruned_loss=0.1048, over 5685209.80 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3518, pruned_loss=0.09527, over 4449767.29 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3511, pruned_loss=0.1051, over 5672000.78 frames. ], batch size: 128, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:57:03,064 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:27,239 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:28,231 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.894e+02 1.135e+03 1.480e+03 2.075e+03 4.713e+03, threshold=2.960e+03, percent-clipped=2.0
+2023-03-06 14:57:29,323 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:33,022 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:36,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:40,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:43,688 INFO [train.py:968] (0/2) Epoch 13, batch 3000, giga_loss[loss=0.3123, simple_loss=0.38, pruned_loss=0.1223, over 28717.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.357, pruned_loss=0.1068, over 5696437.03 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3509, pruned_loss=0.09468, over 4499123.17 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3565, pruned_loss=0.1078, over 5679987.14 frames. ], batch size: 284, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:57:43,692 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 14:57:48,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5862, 1.6774, 1.2940, 1.3615], device='cuda:0'), covar=tensor([0.0818, 0.0470, 0.1009, 0.0924], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0441, 0.0505, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 14:57:52,534 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2205, simple_loss=0.3251, pruned_loss=0.05791, over 944034.00 frames. 
+2023-03-06 14:57:52,534 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 14:58:03,808 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:58:35,506 INFO [train.py:968] (0/2) Epoch 13, batch 3050, giga_loss[loss=0.2426, simple_loss=0.3224, pruned_loss=0.08136, over 28592.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3567, pruned_loss=0.1069, over 5687078.29 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3514, pruned_loss=0.09517, over 4533863.30 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3561, pruned_loss=0.1077, over 5669348.67 frames. ], batch size: 307, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:59:04,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.704e+02 1.167e+03 1.683e+03 2.403e+03 7.257e+03, threshold=3.367e+03, percent-clipped=12.0
+2023-03-06 14:59:18,043 INFO [train.py:968] (0/2) Epoch 13, batch 3100, giga_loss[loss=0.2848, simple_loss=0.3536, pruned_loss=0.108, over 27660.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3518, pruned_loss=0.1031, over 5691614.85 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3519, pruned_loss=0.09562, over 4566151.96 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.351, pruned_loss=0.1035, over 5674410.41 frames. ], batch size: 472, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:59:50,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:59:54,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:00:03,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:00:04,058 INFO [train.py:968] (0/2) Epoch 13, batch 3150, giga_loss[loss=0.2713, simple_loss=0.3458, pruned_loss=0.09839, over 28216.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3507, pruned_loss=0.1022, over 5683773.00 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3519, pruned_loss=0.09585, over 4585726.43 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3501, pruned_loss=0.1025, over 5667855.49 frames. ], batch size: 368, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:00:19,891 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:00:23,495 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-06 15:00:36,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.353e+02 1.046e+03 1.304e+03 1.756e+03 5.436e+03, threshold=2.608e+03, percent-clipped=2.0
+2023-03-06 15:00:49,828 INFO [train.py:968] (0/2) Epoch 13, batch 3200, giga_loss[loss=0.304, simple_loss=0.3727, pruned_loss=0.1176, over 28902.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.351, pruned_loss=0.102, over 5680311.00 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3519, pruned_loss=0.09585, over 4585726.43 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3505, pruned_loss=0.1022, over 5667922.21 frames. ], batch size: 199, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:00:52,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550668.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:01:08,653 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.65 vs. limit=5.0
+2023-03-06 15:01:30,021 INFO [train.py:968] (0/2) Epoch 13, batch 3250, giga_loss[loss=0.248, simple_loss=0.3323, pruned_loss=0.0818, over 28924.00 frames. ], tot_loss[loss=0.278, simple_loss=0.352, pruned_loss=0.102, over 5686570.35 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3515, pruned_loss=0.09561, over 4618242.25 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3518, pruned_loss=0.1025, over 5671997.15 frames. ], batch size: 106, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:01:31,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3687, 1.5748, 1.3110, 1.4477], device='cuda:0'), covar=tensor([0.0776, 0.0321, 0.0327, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 15:01:39,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:02:02,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.449e+02 1.214e+03 1.629e+03 2.255e+03 5.347e+03, threshold=3.257e+03, percent-clipped=15.0
+2023-03-06 15:02:08,854 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:02:11,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:02:14,969 INFO [train.py:968] (0/2) Epoch 13, batch 3300, libri_loss[loss=0.2665, simple_loss=0.3556, pruned_loss=0.0887, over 29244.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3543, pruned_loss=0.1034, over 5696877.80 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3519, pruned_loss=0.09575, over 4646890.16 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.354, pruned_loss=0.1038, over 5683398.66 frames. ], batch size: 97, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:02:36,513 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:02:41,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:02:53,038 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550811.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:02:55,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550814.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:02:55,507 INFO [train.py:968] (0/2) Epoch 13, batch 3350, giga_loss[loss=0.3011, simple_loss=0.3709, pruned_loss=0.1157, over 28465.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3559, pruned_loss=0.1048, over 5697990.70 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3523, pruned_loss=0.09608, over 4699928.98 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3555, pruned_loss=0.1053, over 5680484.31 frames. ], batch size: 71, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:03:08,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:03:13,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:03:20,847 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550843.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:03:28,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.304e+02 1.232e+03 1.570e+03 2.222e+03 3.824e+03, threshold=3.139e+03, percent-clipped=4.0
+2023-03-06 15:03:42,893 INFO [train.py:968] (0/2) Epoch 13, batch 3400, giga_loss[loss=0.2644, simple_loss=0.3361, pruned_loss=0.09632, over 28841.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3565, pruned_loss=0.1061, over 5692161.87 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.352, pruned_loss=0.09593, over 4711971.34 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3564, pruned_loss=0.1067, over 5676468.60 frames. ], batch size: 119, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:03:44,499 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:03:46,332 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:04:13,417 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:04:16,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3082, 4.1130, 3.9588, 1.8345], device='cuda:0'), covar=tensor([0.0619, 0.0801, 0.0900, 0.1951], device='cuda:0'), in_proj_covar=tensor([0.1051, 0.0980, 0.0857, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 15:04:23,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6151, 4.6490, 1.9290, 1.6517], device='cuda:0'), covar=tensor([0.0969, 0.0194, 0.0779, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0508, 0.0344, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 15:04:26,672 INFO [train.py:968] (0/2) Epoch 13, batch 3450, giga_loss[loss=0.2886, simple_loss=0.3537, pruned_loss=0.1117, over 28877.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3564, pruned_loss=0.1062, over 5688079.43 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3523, pruned_loss=0.09602, over 4723711.58 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3562, pruned_loss=0.1068, over 5674064.81 frames. ], batch size: 112, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:04:47,704 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:04:48,832 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550941.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:04:49,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:04:53,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7459, 1.8382, 1.7067, 1.6168], device='cuda:0'), covar=tensor([0.1656, 0.2392, 0.1994, 0.2125], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0734, 0.0678, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 15:04:56,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.879e+02 1.089e+03 1.352e+03 1.855e+03 4.338e+03, threshold=2.704e+03, percent-clipped=4.0
+2023-03-06 15:05:07,811 INFO [train.py:968] (0/2) Epoch 13, batch 3500, giga_loss[loss=0.2387, simple_loss=0.3317, pruned_loss=0.07287, over 29086.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3562, pruned_loss=0.1052, over 5695044.63 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3522, pruned_loss=0.09587, over 4752050.80 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3562, pruned_loss=0.1059, over 5680018.41 frames. ], batch size: 155, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:05:13,480 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:14,340 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:16,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:20,716 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:22,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:39,155 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=551001.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:05:41,297 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=551004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:48,126 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=551011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:51,489 INFO [train.py:968] (0/2) Epoch 13, batch 3550, giga_loss[loss=0.2706, simple_loss=0.3484, pruned_loss=0.0964, over 28954.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.356, pruned_loss=0.1042, over 5700931.31 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3518, pruned_loss=0.09567, over 4780760.01 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3565, pruned_loss=0.1051, over 5684173.53 frames. ], batch size: 174, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:05:59,804 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-06 15:06:08,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6100, 1.9990, 1.9332, 1.4473], device='cuda:0'), covar=tensor([0.1844, 0.2466, 0.1511, 0.1801], device='cuda:0'), in_proj_covar=tensor([0.0849, 0.0702, 0.0890, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 15:06:24,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.392e+02 1.060e+03 1.367e+03 1.931e+03 1.083e+04, threshold=2.733e+03, percent-clipped=12.0
+2023-03-06 15:06:37,357 INFO [train.py:968] (0/2) Epoch 13, batch 3600, giga_loss[loss=0.286, simple_loss=0.3515, pruned_loss=0.1103, over 27559.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3562, pruned_loss=0.1035, over 5699422.99 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3518, pruned_loss=0.09556, over 4795347.60 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3566, pruned_loss=0.1045, over 5685457.96 frames. ], batch size: 472, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:07:16,055 INFO [train.py:968] (0/2) Epoch 13, batch 3650, giga_loss[loss=0.2573, simple_loss=0.3322, pruned_loss=0.09122, over 28692.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3546, pruned_loss=0.1027, over 5708490.88 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3513, pruned_loss=0.0953, over 4839193.96 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3554, pruned_loss=0.1039, over 5690051.35 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:07:48,492 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.319e+02 1.067e+03 1.315e+03 1.664e+03 4.026e+03, threshold=2.630e+03, percent-clipped=9.0
+2023-03-06 15:08:02,351 INFO [train.py:968] (0/2) Epoch 13, batch 3700, giga_loss[loss=0.2322, simple_loss=0.3154, pruned_loss=0.07455, over 29070.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3517, pruned_loss=0.1012, over 5708277.89 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3513, pruned_loss=0.09519, over 4859857.44 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3523, pruned_loss=0.1023, over 5690599.34 frames. ], batch size: 155, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:08:39,440 INFO [train.py:968] (0/2) Epoch 13, batch 3750, giga_loss[loss=0.3191, simple_loss=0.3649, pruned_loss=0.1367, over 26594.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3496, pruned_loss=0.1002, over 5711890.41 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3517, pruned_loss=0.09556, over 4893842.88 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3497, pruned_loss=0.101, over 5693471.63 frames. ], batch size: 555, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:08:56,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4566, 1.6110, 1.3095, 1.6847], device='cuda:0'), covar=tensor([0.0767, 0.0304, 0.0318, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:0')
+2023-03-06 15:09:13,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.179e+02 1.037e+03 1.307e+03 1.772e+03 4.190e+03, threshold=2.614e+03, percent-clipped=4.0
+2023-03-06 15:09:23,639 INFO [train.py:968] (0/2) Epoch 13, batch 3800, giga_loss[loss=0.2791, simple_loss=0.3513, pruned_loss=0.1034, over 28966.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3513, pruned_loss=0.1018, over 5710362.54 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3514, pruned_loss=0.09542, over 4909174.75 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3515, pruned_loss=0.1026, over 5693175.29 frames. ], batch size: 106, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:10:06,159 INFO [train.py:968] (0/2) Epoch 13, batch 3850, libri_loss[loss=0.2617, simple_loss=0.3526, pruned_loss=0.08536, over 29520.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3519, pruned_loss=0.1018, over 5713668.84 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.352, pruned_loss=0.09572, over 4927646.89 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3516, pruned_loss=0.1023, over 5697717.11 frames. ], batch size: 84, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:10:08,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=551316.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:10:34,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.252e+02 1.087e+03 1.403e+03 2.098e+03 6.351e+03, threshold=2.805e+03, percent-clipped=13.0
+2023-03-06 15:10:45,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4763, 2.1179, 1.5438, 0.6679], device='cuda:0'), covar=tensor([0.4465, 0.2006, 0.3009, 0.4839], device='cuda:0'), in_proj_covar=tensor([0.1566, 0.1476, 0.1480, 0.1284], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 15:10:46,104 INFO [train.py:968] (0/2) Epoch 13, batch 3900, giga_loss[loss=0.2616, simple_loss=0.3471, pruned_loss=0.08803, over 28828.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3518, pruned_loss=0.1013, over 5716000.30 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3516, pruned_loss=0.0959, over 4967472.09 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3519, pruned_loss=0.1018, over 5698354.08 frames. ], batch size: 174, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:10:54,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=551376.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:11:27,915 INFO [train.py:968] (0/2) Epoch 13, batch 3950, giga_loss[loss=0.3256, simple_loss=0.377, pruned_loss=0.1371, over 26674.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3507, pruned_loss=0.1003, over 5720352.07 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3514, pruned_loss=0.09575, over 4988989.71 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3509, pruned_loss=0.1009, over 5703690.27 frames. ], batch size: 555, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:11:32,021 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-06 15:11:42,338 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=551433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:11:57,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.944e+02 9.128e+02 1.190e+03 1.636e+03 4.629e+03, threshold=2.380e+03, percent-clipped=7.0
+2023-03-06 15:12:02,941 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-06 15:12:03,932 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=551459.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:12:05,200 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 15:12:07,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=551462.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:12:09,319 INFO [train.py:968] (0/2) Epoch 13, batch 4000, giga_loss[loss=0.2662, simple_loss=0.3441, pruned_loss=0.09414, over 28912.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3507, pruned_loss=0.1004, over 5706257.28 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3521, pruned_loss=0.09621, over 4997471.07 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3503, pruned_loss=0.1007, over 5697790.64 frames. ], batch size: 227, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:12:17,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5352, 3.5791, 1.5891, 1.6227], device='cuda:0'), covar=tensor([0.0918, 0.0237, 0.0850, 0.1283], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0502, 0.0342, 0.0390], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 15:12:28,795 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=551491.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:12:49,082 INFO [train.py:968] (0/2) Epoch 13, batch 4050, giga_loss[loss=0.247, simple_loss=0.3226, pruned_loss=0.08568, over 28730.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3492, pruned_loss=0.1002, over 5713945.17 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3521, pruned_loss=0.09627, over 5006517.94 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3489, pruned_loss=0.1004, over 5705713.93 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:12:52,140 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=551519.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:12:54,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=551522.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:13:11,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8656, 3.6875, 3.4584, 1.5839], device='cuda:0'), covar=tensor([0.0618, 0.0755, 0.0753, 0.2248], device='cuda:0'), in_proj_covar=tensor([0.1059, 0.0980, 0.0858, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 15:13:17,335 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=551551.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:13:17,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.964e+02 1.080e+03 1.352e+03 1.948e+03 5.393e+03, threshold=2.703e+03, percent-clipped=18.0
+2023-03-06 15:13:25,538 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-06 15:13:28,219 INFO [train.py:968] (0/2) Epoch 13, batch 4100, giga_loss[loss=0.2331, simple_loss=0.3133, pruned_loss=0.07647, over 28929.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3468, pruned_loss=0.09918, over 5718161.35 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3524, pruned_loss=0.09646, over 5029241.55 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3462, pruned_loss=0.09923, over 5706839.36 frames. ], batch size: 145, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:14:08,983 INFO [train.py:968] (0/2) Epoch 13, batch 4150, giga_loss[loss=0.27, simple_loss=0.3369, pruned_loss=0.1015, over 28801.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3447, pruned_loss=0.09794, over 5719817.97 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3523, pruned_loss=0.09633, over 5049189.98 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3441, pruned_loss=0.09812, over 5707965.72 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:14:22,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7201, 1.0645, 2.8545, 2.7927], device='cuda:0'), covar=tensor([0.1640, 0.2494, 0.0514, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0668, 0.0593, 0.0857, 0.0779], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 15:14:40,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.844e+02 1.113e+03 1.370e+03 1.698e+03 3.666e+03, threshold=2.740e+03, percent-clipped=7.0
+2023-03-06 15:14:49,797 INFO [train.py:968] (0/2) Epoch 13, batch 4200, giga_loss[loss=0.2808, simple_loss=0.3493, pruned_loss=0.1061, over 28643.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3444, pruned_loss=0.09805, over 5717675.68 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3525, pruned_loss=0.09638, over 5058404.29 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3436, pruned_loss=0.09817, over 5708989.95 frames. ], batch size: 307, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:15:11,621 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8029, 1.7649, 1.2802, 1.4613], device='cuda:0'), covar=tensor([0.0719, 0.0646, 0.1092, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0440, 0.0504, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 15:15:29,908 INFO [train.py:968] (0/2) Epoch 13, batch 4250, giga_loss[loss=0.2793, simple_loss=0.3499, pruned_loss=0.1043, over 28661.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3433, pruned_loss=0.09754, over 5720989.57 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3525, pruned_loss=0.0963, over 5086933.99 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3424, pruned_loss=0.09775, over 5708832.17 frames. ], batch size: 307, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:16:03,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.068e+02 1.078e+03 1.462e+03 2.330e+03 6.609e+03, threshold=2.923e+03, percent-clipped=17.0
+2023-03-06 15:16:13,373 INFO [train.py:968] (0/2) Epoch 13, batch 4300, giga_loss[loss=0.2224, simple_loss=0.2986, pruned_loss=0.0731, over 28754.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3429, pruned_loss=0.0983, over 5710592.50 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3528, pruned_loss=0.09635, over 5096870.28 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3417, pruned_loss=0.09848, over 5704992.22 frames. ], batch size: 99, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:16:27,419 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 15:16:47,080 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=551808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:16:52,372 INFO [train.py:968] (0/2) Epoch 13, batch 4350, giga_loss[loss=0.2479, simple_loss=0.3165, pruned_loss=0.08968, over 28727.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3413, pruned_loss=0.09814, over 5705642.73 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3533, pruned_loss=0.0969, over 5106682.09 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3397, pruned_loss=0.09786, over 5706132.59 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:16:54,759 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.54 vs. limit=5.0
+2023-03-06 15:17:24,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.717e+02 1.121e+03 1.456e+03 2.044e+03 7.473e+03, threshold=2.912e+03, percent-clipped=9.0
+2023-03-06 15:17:33,722 INFO [train.py:968] (0/2) Epoch 13, batch 4400, giga_loss[loss=0.2494, simple_loss=0.3158, pruned_loss=0.09152, over 28932.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3408, pruned_loss=0.09824, over 5698302.36 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3538, pruned_loss=0.0971, over 5117128.86 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3388, pruned_loss=0.09788, over 5702042.40 frames. ], batch size: 106, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:17:39,235 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=551872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:18:15,342 INFO [train.py:968] (0/2) Epoch 13, batch 4450, giga_loss[loss=0.2516, simple_loss=0.3372, pruned_loss=0.08301, over 29056.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.342, pruned_loss=0.09856, over 5700540.76 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3543, pruned_loss=0.09741, over 5129794.11 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3398, pruned_loss=0.09802, over 5702687.91 frames. ], batch size: 164, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:18:29,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5973, 4.4372, 4.1714, 2.1368], device='cuda:0'), covar=tensor([0.0463, 0.0597, 0.0660, 0.1964], device='cuda:0'), in_proj_covar=tensor([0.1059, 0.0981, 0.0857, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 15:18:34,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.7763, 1.2172, 1.1595, 1.0858], device='cuda:0'), covar=tensor([0.1737, 0.1159, 0.1915, 0.1431], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0735, 0.0677, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 15:18:49,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=551951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:18:51,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.855e+02 1.045e+03 1.283e+03 1.711e+03 3.623e+03, threshold=2.565e+03, percent-clipped=3.0
+2023-03-06 15:18:52,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=551954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:18:57,082 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-06 15:19:00,966 INFO [train.py:968] (0/2) Epoch 13, batch 4500, giga_loss[loss=0.2743, simple_loss=0.3568, pruned_loss=0.09596, over 28661.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3441, pruned_loss=0.0996, over 5708346.80 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3547, pruned_loss=0.09782, over 5148928.37 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3417, pruned_loss=0.09887, over 5705519.47 frames. ], batch size: 242, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:19:16,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=551983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:19:29,625 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-552000.pt
+2023-03-06 15:19:41,450 INFO [train.py:968] (0/2) Epoch 13, batch 4550, giga_loss[loss=0.2624, simple_loss=0.3397, pruned_loss=0.09249, over 28742.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3469, pruned_loss=0.1003, over 5713836.98 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.355, pruned_loss=0.09811, over 5168845.20 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3445, pruned_loss=0.09955, over 5708824.73 frames. ], batch size: 119, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:20:16,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.898e+02 1.101e+03 1.357e+03 1.859e+03 4.907e+03, threshold=2.714e+03, percent-clipped=12.0
+2023-03-06 15:20:26,587 INFO [train.py:968] (0/2) Epoch 13, batch 4600, giga_loss[loss=0.2595, simple_loss=0.3386, pruned_loss=0.09016, over 28846.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3485, pruned_loss=0.1007, over 5709178.51 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3553, pruned_loss=0.09834, over 5182608.76 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3462, pruned_loss=0.09992, over 5702537.00 frames. ], batch size: 186, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:20:46,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3941, 4.2128, 3.9844, 2.0646], device='cuda:0'), covar=tensor([0.0503, 0.0648, 0.0668, 0.1934], device='cuda:0'), in_proj_covar=tensor([0.1053, 0.0978, 0.0855, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 15:20:53,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7766, 2.5995, 1.6871, 0.8187], device='cuda:0'), covar=tensor([0.6377, 0.3038, 0.3283, 0.5859], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1473, 0.1480, 0.1283], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 15:21:09,414 INFO [train.py:968] (0/2) Epoch 13, batch 4650, giga_loss[loss=0.245, simple_loss=0.3228, pruned_loss=0.08364, over 28764.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.347, pruned_loss=0.09906, over 5707382.36 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3547, pruned_loss=0.098, over 5209520.49 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3454, pruned_loss=0.09879, over 5696031.23 frames. ], batch size: 119, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:21:44,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.300e+02 1.175e+03 1.545e+03 2.261e+03 8.282e+03, threshold=3.090e+03, percent-clipped=12.0
+2023-03-06 15:21:46,341 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=552157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:21:53,553 INFO [train.py:968] (0/2) Epoch 13, batch 4700, giga_loss[loss=0.2741, simple_loss=0.3521, pruned_loss=0.09803, over 28693.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3468, pruned_loss=0.09874, over 5705538.36 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3546, pruned_loss=0.09791, over 5216169.87 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3456, pruned_loss=0.09861, over 5695071.19 frames. ], batch size: 262, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:22:33,183 INFO [train.py:968] (0/2) Epoch 13, batch 4750, giga_loss[loss=0.2506, simple_loss=0.3344, pruned_loss=0.08341, over 28537.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3467, pruned_loss=0.09831, over 5715548.57 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3546, pruned_loss=0.09784, over 5239521.61 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3454, pruned_loss=0.09828, over 5701281.41 frames. ], batch size: 307, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:22:55,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5651, 1.8526, 1.4935, 1.6688], device='cuda:0'), covar=tensor([0.2342, 0.2295, 0.2616, 0.2258], device='cuda:0'), in_proj_covar=tensor([0.1340, 0.0981, 0.1178, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 15:23:00,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=552247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:23:05,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.305e+02 1.187e+03 1.540e+03 1.895e+03 5.317e+03, threshold=3.080e+03, percent-clipped=2.0
+2023-03-06 15:23:14,915 INFO [train.py:968] (0/2) Epoch 13, batch 4800, giga_loss[loss=0.2695, simple_loss=0.3452, pruned_loss=0.09694, over 29031.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3466, pruned_loss=0.09831, over 5718194.03 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3545, pruned_loss=0.09765, over 5254913.51 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3455, pruned_loss=0.09846, over 5703603.11 frames. ], batch size: 128, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:23:56,781 INFO [train.py:968] (0/2) Epoch 13, batch 4850, giga_loss[loss=0.2831, simple_loss=0.3558, pruned_loss=0.1052, over 29058.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3496, pruned_loss=0.1004, over 5718723.77 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3544, pruned_loss=0.09752, over 5271096.24 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3487, pruned_loss=0.1007, over 5704737.99 frames. ], batch size: 128, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:24:12,769 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 15:24:30,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.163e+02 1.224e+03 1.611e+03 2.094e+03 5.410e+03, threshold=3.221e+03, percent-clipped=8.0
+2023-03-06 15:24:40,243 INFO [train.py:968] (0/2) Epoch 13, batch 4900, giga_loss[loss=0.2568, simple_loss=0.3375, pruned_loss=0.08808, over 29029.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3525, pruned_loss=0.1024, over 5719021.92 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3547, pruned_loss=0.09763, over 5280100.71 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3514, pruned_loss=0.1026, over 5705751.94 frames. ], batch size: 128, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:25:00,542 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=552390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:25:04,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=552393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:25:21,525 INFO [train.py:968] (0/2) Epoch 13, batch 4950, giga_loss[loss=0.2596, simple_loss=0.334, pruned_loss=0.0926, over 28884.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3548, pruned_loss=0.1035, over 5716696.43 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3549, pruned_loss=0.09776, over 5290773.91 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3538, pruned_loss=0.1037, over 5704332.11 frames. ], batch size: 106, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:25:28,359 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=552422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:25:33,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-06 15:25:50,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3463, 1.6556, 1.6692, 1.1847], device='cuda:0'), covar=tensor([0.1646, 0.2292, 0.1352, 0.1567], device='cuda:0'), in_proj_covar=tensor([0.0843, 0.0694, 0.0882, 0.0788], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 15:25:54,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.754e+02 1.208e+03 1.572e+03 2.220e+03 5.390e+03, threshold=3.144e+03, percent-clipped=9.0
+2023-03-06 15:26:02,494 INFO [train.py:968] (0/2) Epoch 13, batch 5000, giga_loss[loss=0.2539, simple_loss=0.3307, pruned_loss=0.0885, over 28703.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3555, pruned_loss=0.1038, over 5720439.27 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3549, pruned_loss=0.09774, over 5297073.56 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3547, pruned_loss=0.104, over 5709000.91 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:26:02,962 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=552465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:26:09,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1830, 1.6129, 1.1843, 0.4251], device='cuda:0'), covar=tensor([0.2893, 0.1778, 0.2579, 0.4218], device='cuda:0'), in_proj_covar=tensor([0.1572, 0.1479, 0.1486, 0.1291], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 15:26:46,234 INFO [train.py:968] (0/2) Epoch 13, batch 5050, giga_loss[loss=0.2717, simple_loss=0.3532, pruned_loss=0.09507, over 28984.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3567, pruned_loss=0.1049, over 5712190.13 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3548, pruned_loss=0.09766, over 5307656.75 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3561, pruned_loss=0.1052, over 5701041.31 frames. ], batch size: 164, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:26:48,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6377, 1.6578, 1.2194, 1.3165], device='cuda:0'), covar=tensor([0.0860, 0.0714, 0.1176, 0.1239], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0440, 0.0501, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 15:26:56,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2809, 1.1992, 1.2128, 1.4282], device='cuda:0'), covar=tensor([0.0731, 0.0340, 0.0328, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:0')
+2023-03-06 15:27:00,740 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=552532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:27:19,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.170e+02 1.202e+03 1.524e+03 2.008e+03 5.299e+03, threshold=3.049e+03, percent-clipped=6.0
+2023-03-06 15:27:27,223 INFO [train.py:968] (0/2) Epoch 13, batch 5100, giga_loss[loss=0.2457, simple_loss=0.32, pruned_loss=0.08574, over 29021.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3568, pruned_loss=0.1048, over 5718131.22 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3548, pruned_loss=0.0976, over 5319277.98 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3564, pruned_loss=0.1053, over 5706127.69 frames. ], batch size: 128, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:28:03,645 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=552604.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:28:12,260 INFO [train.py:968] (0/2) Epoch 13, batch 5150, giga_loss[loss=0.297, simple_loss=0.3509, pruned_loss=0.1215, over 23896.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3533, pruned_loss=0.103, over 5709852.12 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3549, pruned_loss=0.09761, over 5328060.09 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.353, pruned_loss=0.1034, over 5697635.79 frames. ], batch size: 705, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:28:42,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.562e+02 1.013e+03 1.252e+03 1.639e+03 5.503e+03, threshold=2.504e+03, percent-clipped=3.0
+2023-03-06 15:28:51,008 INFO [train.py:968] (0/2) Epoch 13, batch 5200, giga_loss[loss=0.2304, simple_loss=0.3097, pruned_loss=0.07549, over 28308.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3495, pruned_loss=0.101, over 5718539.12 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3553, pruned_loss=0.09794, over 5343690.44 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3487, pruned_loss=0.1012, over 5705014.54 frames. ], batch size: 60, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:28:59,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9131, 1.0778, 0.8651, 0.3031], device='cuda:0'), covar=tensor([0.2524, 0.1864, 0.2461, 0.4351], device='cuda:0'), in_proj_covar=tensor([0.1569, 0.1474, 0.1482, 0.1284], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 15:29:00,857 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=552675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:29:02,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=552678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:29:23,216 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 15:29:26,508 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=552707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:29:32,225 INFO [train.py:968] (0/2) Epoch 13, batch 5250, giga_loss[loss=0.2482, simple_loss=0.3213, pruned_loss=0.08758, over 28419.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3487, pruned_loss=0.1002, over 5712951.88 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3554, pruned_loss=0.09809, over 5342613.91 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3479, pruned_loss=0.1003, over 5707929.84 frames. ], batch size: 71, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:30:06,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.907e+02 1.014e+03 1.237e+03 1.765e+03 5.721e+03, threshold=2.474e+03, percent-clipped=9.0
+2023-03-06 15:30:14,950 INFO [train.py:968] (0/2) Epoch 13, batch 5300, giga_loss[loss=0.2755, simple_loss=0.3646, pruned_loss=0.09318, over 28724.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3491, pruned_loss=0.09909, over 5715599.69 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3548, pruned_loss=0.09789, over 5355752.99 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3489, pruned_loss=0.09935, over 5707743.21 frames. ], batch size: 262, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:30:58,397 INFO [train.py:968] (0/2) Epoch 13, batch 5350, giga_loss[loss=0.3524, simple_loss=0.4026, pruned_loss=0.1511, over 26681.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3495, pruned_loss=0.09863, over 5710783.48 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3547, pruned_loss=0.0979, over 5354605.50 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3493, pruned_loss=0.09883, over 5710683.57 frames. ], batch size: 555, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:31:19,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=552840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:31:33,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.114e+02 1.138e+03 1.444e+03 2.104e+03 4.497e+03, threshold=2.888e+03, percent-clipped=15.0
+2023-03-06 15:31:40,087 INFO [train.py:968] (0/2) Epoch 13, batch 5400, giga_loss[loss=0.2716, simple_loss=0.3419, pruned_loss=0.1006, over 29005.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3487, pruned_loss=0.09958, over 5715410.17 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3551, pruned_loss=0.0981, over 5364681.57 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3482, pruned_loss=0.09961, over 5712618.26 frames. ], batch size: 136, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:32:18,223 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1771, 2.0913, 2.0531, 1.8436], device='cuda:0'), covar=tensor([0.1527, 0.2349, 0.1921, 0.2227], device='cuda:0'), in_proj_covar=tensor([0.0442, 0.0742, 0.0682, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 15:32:21,284 INFO [train.py:968] (0/2) Epoch 13, batch 5450, libri_loss[loss=0.3089, simple_loss=0.3858, pruned_loss=0.116, over 27916.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3494, pruned_loss=0.1017, over 5719106.09 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3558, pruned_loss=0.09856, over 5377758.64 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3482, pruned_loss=0.1014, over 5716050.24 frames. ], batch size: 116, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:32:53,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.972e+02 1.194e+03 1.413e+03 2.005e+03 4.459e+03, threshold=2.826e+03, percent-clipped=3.0
+2023-03-06 15:33:00,610 INFO [train.py:968] (0/2) Epoch 13, batch 5500, giga_loss[loss=0.2869, simple_loss=0.3474, pruned_loss=0.1132, over 28879.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.347, pruned_loss=0.1016, over 5728374.94 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3561, pruned_loss=0.09881, over 5393211.12 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3456, pruned_loss=0.1012, over 5721271.40 frames. ], batch size: 199, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:33:13,374 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=552979.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:33:15,914 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=552983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:33:19,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=552986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:33:29,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=552998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:33:42,388 INFO [train.py:968] (0/2) Epoch 13, batch 5550, giga_loss[loss=0.2823, simple_loss=0.3592, pruned_loss=0.1027, over 28379.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3456, pruned_loss=0.1018, over 5731621.81 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3563, pruned_loss=0.09903, over 5405093.02 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3441, pruned_loss=0.1014, over 5722665.96 frames. ], batch size: 368, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:33:42,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=553015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:34:11,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9992, 3.8273, 3.6216, 1.9136], device='cuda:0'), covar=tensor([0.0566, 0.0678, 0.0671, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.1065, 0.0986, 0.0859, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 15:34:19,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.793e+02 1.253e+03 1.666e+03 2.221e+03 7.867e+03, threshold=3.331e+03, percent-clipped=14.0
+2023-03-06 15:34:22,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6513, 1.2271, 4.8832, 3.5526], device='cuda:0'), covar=tensor([0.1523, 0.2744, 0.0365, 0.0769], device='cuda:0'), in_proj_covar=tensor([0.0669, 0.0596, 0.0862, 0.0783], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 15:34:28,027 INFO [train.py:968] (0/2) Epoch 13, batch 5600, giga_loss[loss=0.2472, simple_loss=0.323, pruned_loss=0.08571, over 28662.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3454, pruned_loss=0.102, over 5720145.56 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3565, pruned_loss=0.09913, over 5411626.67 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3438, pruned_loss=0.1016, over 5711710.07 frames. ], batch size: 284, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:34:41,816 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.53 vs. limit=5.0
+2023-03-06 15:35:08,266 INFO [train.py:968] (0/2) Epoch 13, batch 5650, giga_loss[loss=0.2114, simple_loss=0.2818, pruned_loss=0.07049, over 28395.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3409, pruned_loss=0.09955, over 5700028.96 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3564, pruned_loss=0.09917, over 5407395.85 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3393, pruned_loss=0.09925, over 5707076.74 frames. ], batch size: 65, lr: 2.52e-03, grad_scale: 2.0
+2023-03-06 15:35:15,609 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=553122.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:35:17,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=553125.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:35:42,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=553154.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:35:44,700 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-06 15:35:45,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.643e+02 1.114e+03 1.380e+03 1.856e+03 3.689e+03, threshold=2.761e+03, percent-clipped=1.0
+2023-03-06 15:35:50,569 INFO [train.py:968] (0/2) Epoch 13, batch 5700, giga_loss[loss=0.2491, simple_loss=0.3221, pruned_loss=0.08805, over 28651.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3361, pruned_loss=0.09666, over 5707200.42 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.356, pruned_loss=0.09889, over 5417258.16 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3349, pruned_loss=0.09664, over 5709190.76 frames. ], batch size: 262, lr: 2.52e-03, grad_scale: 2.0
+2023-03-06 15:36:15,779 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 15:36:29,478 INFO [train.py:968] (0/2) Epoch 13, batch 5750, giga_loss[loss=0.2632, simple_loss=0.3358, pruned_loss=0.09532, over 28948.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3344, pruned_loss=0.09543, over 5712675.15 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3556, pruned_loss=0.09861, over 5429753.67 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3331, pruned_loss=0.0956, over 5710681.95 frames. ], batch size: 136, lr: 2.52e-03, grad_scale: 2.0
+2023-03-06 15:37:04,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.817e+02 1.163e+03 1.598e+03 2.042e+03 6.149e+03, threshold=3.195e+03, percent-clipped=10.0
+2023-03-06 15:37:08,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3100, 1.4075, 1.2267, 1.5564], device='cuda:0'), covar=tensor([0.0730, 0.0328, 0.0344, 0.0829], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 15:37:09,135 INFO [train.py:968] (0/2) Epoch 13, batch 5800, giga_loss[loss=0.2484, simple_loss=0.3302, pruned_loss=0.08333, over 28843.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3355, pruned_loss=0.09583, over 5711047.13 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3555, pruned_loss=0.09847, over 5434683.85 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3345, pruned_loss=0.09605, over 5707532.17 frames. ], batch size: 119, lr: 2.52e-03, grad_scale: 2.0
+2023-03-06 15:37:25,579 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=553286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:37:43,731 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=553309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:37:47,534 INFO [train.py:968] (0/2) Epoch 13, batch 5850, libri_loss[loss=0.2927, simple_loss=0.3704, pruned_loss=0.1075, over 29543.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3401, pruned_loss=0.09783, over 5708641.26 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3559, pruned_loss=0.0988, over 5444348.59 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3382, pruned_loss=0.09762, over 5707456.57 frames. ], batch size: 83, lr: 2.52e-03, grad_scale: 2.0
+2023-03-06 15:38:01,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4751, 1.6947, 1.7303, 1.2893], device='cuda:0'), covar=tensor([0.1705, 0.2355, 0.1439, 0.1648], device='cuda:0'), in_proj_covar=tensor([0.0838, 0.0690, 0.0879, 0.0785], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 15:38:24,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.725e+02 1.108e+03 1.341e+03 1.871e+03 5.106e+03, threshold=2.682e+03, percent-clipped=4.0
+2023-03-06 15:38:30,106 INFO [train.py:968] (0/2) Epoch 13, batch 5900, giga_loss[loss=0.2623, simple_loss=0.3406, pruned_loss=0.09205, over 28992.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3441, pruned_loss=0.09947, over 5708326.32 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3564, pruned_loss=0.09931, over 5446465.55 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3418, pruned_loss=0.09889, over 5710601.77 frames. ], batch size: 136, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:38:36,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=553373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:38:47,756 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-06 15:39:11,703 INFO [train.py:968] (0/2) Epoch 13, batch 5950, giga_loss[loss=0.3546, simple_loss=0.4126, pruned_loss=0.1483, over 27985.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3475, pruned_loss=0.1007, over 5709442.96 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3563, pruned_loss=0.09929, over 5462002.58 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3453, pruned_loss=0.1003, over 5707871.90 frames. ], batch size: 412, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:39:19,647 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-06 15:39:30,357 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 15:39:53,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.852e+02 1.261e+03 1.522e+03 2.123e+03 4.264e+03, threshold=3.045e+03, percent-clipped=12.0
+2023-03-06 15:39:59,851 INFO [train.py:968] (0/2) Epoch 13, batch 6000, giga_loss[loss=0.2854, simple_loss=0.36, pruned_loss=0.1054, over 28917.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.35, pruned_loss=0.1022, over 5704255.48 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3565, pruned_loss=0.09946, over 5462951.64 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3481, pruned_loss=0.1017, over 5704028.01 frames. ], batch size: 213, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:39:59,856 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 15:40:08,375 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2192, simple_loss=0.3255, pruned_loss=0.05648, over 944034.00 frames. 
+2023-03-06 15:40:08,376 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 15:40:52,490 INFO [train.py:968] (0/2) Epoch 13, batch 6050, giga_loss[loss=0.3683, simple_loss=0.4217, pruned_loss=0.1574, over 28617.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3551, pruned_loss=0.1065, over 5693044.50 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3565, pruned_loss=0.09959, over 5467393.83 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3534, pruned_loss=0.1061, over 5695747.69 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:40:54,598 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=553516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:40:56,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=553519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:41:21,749 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=553548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:41:30,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.217e+02 1.318e+03 1.712e+03 2.272e+03 5.514e+03, threshold=3.424e+03, percent-clipped=12.0
+2023-03-06 15:41:36,778 INFO [train.py:968] (0/2) Epoch 13, batch 6100, giga_loss[loss=0.3362, simple_loss=0.3997, pruned_loss=0.1363, over 28737.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3609, pruned_loss=0.1117, over 5693858.76 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3561, pruned_loss=0.09956, over 5486750.63 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3599, pruned_loss=0.1118, over 5687527.98 frames. ], batch size: 262, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:42:25,868 INFO [train.py:968] (0/2) Epoch 13, batch 6150, giga_loss[loss=0.2819, simple_loss=0.3593, pruned_loss=0.1022, over 29017.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3672, pruned_loss=0.116, over 5695336.20 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3557, pruned_loss=0.09939, over 5486606.15 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.367, pruned_loss=0.1165, over 5694307.37 frames. ], batch size: 128, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:43:08,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.790e+02 1.522e+03 2.055e+03 2.491e+03 7.580e+03, threshold=4.111e+03, percent-clipped=10.0
+2023-03-06 15:43:15,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=553661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:43:19,562 INFO [train.py:968] (0/2) Epoch 13, batch 6200, giga_loss[loss=0.3556, simple_loss=0.4102, pruned_loss=0.1505, over 28589.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3727, pruned_loss=0.1204, over 5692999.56 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3556, pruned_loss=0.0994, over 5488641.12 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3726, pruned_loss=0.1209, over 5691290.60 frames. ], batch size: 336, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:43:37,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=553684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:44:06,903 INFO [train.py:968] (0/2) Epoch 13, batch 6250, giga_loss[loss=0.3164, simple_loss=0.3886, pruned_loss=0.1221, over 28900.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3794, pruned_loss=0.1263, over 5693107.13 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3553, pruned_loss=0.09923, over 5497174.88 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.38, pruned_loss=0.1273, over 5687864.92 frames. ], batch size: 174, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:44:48,571 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-06 15:44:48,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.904e+02 1.630e+03 2.056e+03 2.795e+03 7.677e+03, threshold=4.112e+03, percent-clipped=7.0
+2023-03-06 15:44:54,606 INFO [train.py:968] (0/2) Epoch 13, batch 6300, giga_loss[loss=0.3284, simple_loss=0.3895, pruned_loss=0.1336, over 29102.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3839, pruned_loss=0.13, over 5687800.12 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3554, pruned_loss=0.09917, over 5506314.40 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.385, pruned_loss=0.1317, over 5680285.29 frames. ], batch size: 128, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:45:37,836 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=553804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:45:40,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=553807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:45:48,809 INFO [train.py:968] (0/2) Epoch 13, batch 6350, giga_loss[loss=0.4124, simple_loss=0.4436, pruned_loss=0.1906, over 28031.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3848, pruned_loss=0.1317, over 5673411.33 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3551, pruned_loss=0.09892, over 5512454.61 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3865, pruned_loss=0.1337, over 5664332.08 frames. ], batch size: 412, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:45:51,177 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=553817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:46:02,589 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=553827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:46:04,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9385, 1.1628, 1.3141, 1.0054], device='cuda:0'), covar=tensor([0.1507, 0.1325, 0.1940, 0.1506], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0735, 0.0681, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 15:46:05,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=553830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:46:10,806 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=553836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:46:32,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.774e+02 1.523e+03 2.102e+03 2.667e+03 1.068e+04, threshold=4.204e+03, percent-clipped=8.0
+2023-03-06 15:46:35,496 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=553859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:46:38,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3746, 1.5942, 1.4340, 1.2186], device='cuda:0'), covar=tensor([0.2049, 0.1815, 0.1350, 0.1725], device='cuda:0'), in_proj_covar=tensor([0.1750, 0.1662, 0.1633, 0.1721], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 15:46:43,303 INFO [train.py:968] (0/2) Epoch 13, batch 6400, giga_loss[loss=0.416, simple_loss=0.4557, pruned_loss=0.1882, over 28644.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3888, pruned_loss=0.1355, over 5677875.15 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3555, pruned_loss=0.09915, over 5522069.92 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3904, pruned_loss=0.1377, over 5665708.22 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 8.0
+2023-03-06 15:47:11,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4182, 1.8153, 1.2997, 0.9424], device='cuda:0'), covar=tensor([0.3139, 0.2315, 0.1678, 0.3434], device='cuda:0'), in_proj_covar=tensor([0.1597, 0.1506, 0.1504, 0.1311], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 15:47:35,249 INFO [train.py:968] (0/2) Epoch 13, batch 6450, giga_loss[loss=0.3362, simple_loss=0.3868, pruned_loss=0.1428, over 28698.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3916, pruned_loss=0.1393, over 5656557.11 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3556, pruned_loss=0.09921, over 5518306.21 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3934, pruned_loss=0.1416, over 5652809.92 frames. ], batch size: 262, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:47:51,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4567, 1.6289, 1.5731, 1.5015], device='cuda:0'), covar=tensor([0.1218, 0.1297, 0.1615, 0.1294], device='cuda:0'), in_proj_covar=tensor([0.0439, 0.0736, 0.0680, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 15:48:20,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1950, 1.6064, 1.3719, 1.3516], device='cuda:0'), covar=tensor([0.0797, 0.0319, 0.0293, 0.0990], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 15:48:23,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.962e+02 1.998e+03 2.656e+03 4.601e+03 1.232e+04, threshold=5.312e+03, percent-clipped=28.0
+2023-03-06 15:48:30,508 INFO [train.py:968] (0/2) Epoch 13, batch 6500, giga_loss[loss=0.2805, simple_loss=0.3553, pruned_loss=0.1028, over 28951.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3961, pruned_loss=0.1434, over 5646228.75 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3558, pruned_loss=0.09928, over 5524373.60 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3979, pruned_loss=0.1458, over 5639416.17 frames. ], batch size: 106, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:49:05,631 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-554000.pt
+2023-03-06 15:49:16,393 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 15:49:19,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6883, 2.0991, 1.6984, 1.5197], device='cuda:0'), covar=tensor([0.2529, 0.1801, 0.2008, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.1758, 0.1673, 0.1647, 0.1731], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 15:49:19,830 INFO [train.py:968] (0/2) Epoch 13, batch 6550, giga_loss[loss=0.3416, simple_loss=0.3999, pruned_loss=0.1416, over 28873.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3944, pruned_loss=0.1427, over 5646323.88 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3558, pruned_loss=0.09923, over 5528075.80 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3964, pruned_loss=0.1453, over 5639078.23 frames. ], batch size: 199, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:49:53,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5403, 1.6115, 1.8036, 1.3693], device='cuda:0'), covar=tensor([0.1400, 0.2041, 0.1165, 0.1421], device='cuda:0'), in_proj_covar=tensor([0.0830, 0.0690, 0.0872, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 15:50:06,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.122e+02 1.849e+03 2.255e+03 3.238e+03 9.374e+03, threshold=4.511e+03, percent-clipped=6.0
+2023-03-06 15:50:11,393 INFO [train.py:968] (0/2) Epoch 13, batch 6600, giga_loss[loss=0.3979, simple_loss=0.4268, pruned_loss=0.1845, over 27522.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3918, pruned_loss=0.1413, over 5643950.76 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3557, pruned_loss=0.09926, over 5530133.08 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3946, pruned_loss=0.1445, over 5638647.57 frames. ], batch size: 472, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:50:47,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0188, 1.1277, 3.3809, 2.9231], device='cuda:0'), covar=tensor([0.1666, 0.2609, 0.0516, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0678, 0.0605, 0.0874, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 15:51:03,624 INFO [train.py:968] (0/2) Epoch 13, batch 6650, giga_loss[loss=0.3778, simple_loss=0.4079, pruned_loss=0.1738, over 23434.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3909, pruned_loss=0.1401, over 5639338.83 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3553, pruned_loss=0.09901, over 5542626.12 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3945, pruned_loss=0.1443, over 5627271.54 frames. ], batch size: 705, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:51:45,742 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.282e+02 1.537e+03 2.154e+03 2.834e+03 7.872e+03, threshold=4.307e+03, percent-clipped=2.0
+2023-03-06 15:51:46,473 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 15:51:51,264 INFO [train.py:968] (0/2) Epoch 13, batch 6700, giga_loss[loss=0.3061, simple_loss=0.3745, pruned_loss=0.1189, over 28916.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3912, pruned_loss=0.1391, over 5656260.18 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3558, pruned_loss=0.09929, over 5551844.39 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3945, pruned_loss=0.1431, over 5640725.95 frames. ], batch size: 136, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:52:19,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=554192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:52:43,075 INFO [train.py:968] (0/2) Epoch 13, batch 6750, giga_loss[loss=0.3915, simple_loss=0.417, pruned_loss=0.183, over 26498.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3933, pruned_loss=0.1408, over 5643161.92 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3554, pruned_loss=0.09909, over 5555905.92 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.397, pruned_loss=0.1449, over 5628792.26 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:52:45,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4331, 1.6786, 1.5178, 1.6311], device='cuda:0'), covar=tensor([0.0621, 0.0274, 0.0272, 0.0638], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 15:53:25,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.968e+02 1.542e+03 1.991e+03 2.567e+03 7.034e+03, threshold=3.982e+03, percent-clipped=9.0
+2023-03-06 15:53:28,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:53:30,038 INFO [train.py:968] (0/2) Epoch 13, batch 6800, giga_loss[loss=0.2989, simple_loss=0.3734, pruned_loss=0.1122, over 28980.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3905, pruned_loss=0.1385, over 5635213.72 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3554, pruned_loss=0.09919, over 5560720.64 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.395, pruned_loss=0.1434, over 5622491.44 frames. ], batch size: 164, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:54:25,517 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:54:26,840 INFO [train.py:968] (0/2) Epoch 13, batch 6850, libri_loss[loss=0.2669, simple_loss=0.3401, pruned_loss=0.09689, over 29541.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3882, pruned_loss=0.1352, over 5642143.05 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3553, pruned_loss=0.09918, over 5562458.04 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3919, pruned_loss=0.1392, over 5630854.42 frames. ], batch size: 75, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:54:44,630 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=554335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:54:46,611 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=554338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:54:49,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5656, 1.7652, 1.4037, 1.7915], device='cuda:0'), covar=tensor([0.2361, 0.2356, 0.2599, 0.2347], device='cuda:0'), in_proj_covar=tensor([0.1325, 0.0977, 0.1166, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 15:54:57,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4092, 2.8775, 1.4388, 1.5194], device='cuda:0'), covar=tensor([0.0852, 0.0359, 0.0832, 0.1181], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0515, 0.0347, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 15:55:06,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.743e+02 1.557e+03 2.362e+03 3.420e+03 8.184e+03, threshold=4.725e+03, percent-clipped=14.0
+2023-03-06 15:55:10,392 INFO [train.py:968] (0/2) Epoch 13, batch 6900, giga_loss[loss=0.3173, simple_loss=0.3889, pruned_loss=0.1229, over 28775.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3857, pruned_loss=0.1324, over 5652653.76 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3557, pruned_loss=0.09966, over 5572603.28 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3896, pruned_loss=0.1365, over 5637309.07 frames. ], batch size: 243, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:55:14,304 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=554367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:55:30,882 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-06 15:55:55,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8709, 1.9990, 1.8629, 1.7315], device='cuda:0'), covar=tensor([0.1909, 0.1626, 0.1331, 0.1428], device='cuda:0'), in_proj_covar=tensor([0.1750, 0.1674, 0.1644, 0.1729], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 15:56:01,904 INFO [train.py:968] (0/2) Epoch 13, batch 6950, giga_loss[loss=0.276, simple_loss=0.3518, pruned_loss=0.1001, over 28834.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3822, pruned_loss=0.1295, over 5658871.82 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3553, pruned_loss=0.09949, over 5577385.15 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3859, pruned_loss=0.1332, over 5643567.84 frames. ], batch size: 119, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:56:48,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.737e+03 2.326e+03 3.901e+03 1.655e+04, threshold=4.652e+03, percent-clipped=15.0
+2023-03-06 15:56:52,491 INFO [train.py:968] (0/2) Epoch 13, batch 7000, giga_loss[loss=0.3367, simple_loss=0.392, pruned_loss=0.1407, over 28877.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3798, pruned_loss=0.1281, over 5655138.87 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.355, pruned_loss=0.0993, over 5582185.12 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3835, pruned_loss=0.1316, over 5639770.50 frames. ], batch size: 227, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:57:00,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:57:00,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1701, 1.5598, 1.1869, 0.4355], device='cuda:0'), covar=tensor([0.2381, 0.1510, 0.2022, 0.3849], device='cuda:0'), in_proj_covar=tensor([0.1597, 0.1507, 0.1500, 0.1309], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 15:57:41,697 INFO [train.py:968] (0/2) Epoch 13, batch 7050, giga_loss[loss=0.2884, simple_loss=0.3665, pruned_loss=0.1052, over 28860.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3786, pruned_loss=0.1275, over 5653585.59 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3549, pruned_loss=0.0994, over 5586740.50 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3819, pruned_loss=0.1306, over 5638349.54 frames. ], batch size: 145, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:58:16,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9718, 1.2681, 2.7379, 2.6873], device='cuda:0'), covar=tensor([0.1371, 0.2181, 0.0530, 0.0995], device='cuda:0'), in_proj_covar=tensor([0.0678, 0.0603, 0.0873, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 15:58:36,097 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.084e+02 1.481e+03 1.944e+03 2.310e+03 6.201e+03, threshold=3.889e+03, percent-clipped=6.0
+2023-03-06 15:58:39,332 INFO [train.py:968] (0/2) Epoch 13, batch 7100, giga_loss[loss=0.3192, simple_loss=0.387, pruned_loss=0.1257, over 29073.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3791, pruned_loss=0.1277, over 5654734.41 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3549, pruned_loss=0.09934, over 5589656.08 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.382, pruned_loss=0.1307, over 5640772.74 frames. ], batch size: 155, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:58:47,735 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5138, 2.1722, 1.5051, 0.7793], device='cuda:0'), covar=tensor([0.4171, 0.2217, 0.3778, 0.4776], device='cuda:0'), in_proj_covar=tensor([0.1589, 0.1501, 0.1496, 0.1303], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 15:59:34,349 INFO [train.py:968] (0/2) Epoch 13, batch 7150, giga_loss[loss=0.3034, simple_loss=0.3758, pruned_loss=0.1155, over 28967.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3763, pruned_loss=0.1248, over 5658374.98 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3547, pruned_loss=0.09927, over 5593912.50 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3791, pruned_loss=0.1276, over 5644425.29 frames. ], batch size: 213, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:59:37,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:00:03,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=554638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:00:29,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.804e+02 1.434e+03 1.955e+03 2.750e+03 7.598e+03, threshold=3.910e+03, percent-clipped=9.0
+2023-03-06 16:00:36,390 INFO [train.py:968] (0/2) Epoch 13, batch 7200, giga_loss[loss=0.2922, simple_loss=0.3743, pruned_loss=0.105, over 28644.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3782, pruned_loss=0.1236, over 5661294.81 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3547, pruned_loss=0.09929, over 5595576.94 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3805, pruned_loss=0.1259, over 5649224.15 frames. ], batch size: 242, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:00:59,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=554688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:01:24,743 INFO [train.py:968] (0/2) Epoch 13, batch 7250, giga_loss[loss=0.3439, simple_loss=0.4041, pruned_loss=0.1419, over 28631.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3796, pruned_loss=0.1234, over 5671881.43 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3551, pruned_loss=0.09949, over 5598169.12 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3817, pruned_loss=0.1256, over 5661259.05 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:01:39,748 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:02:09,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.649e+02 1.709e+03 2.239e+03 2.995e+03 6.401e+03, threshold=4.477e+03, percent-clipped=14.0
+2023-03-06 16:02:13,487 INFO [train.py:968] (0/2) Epoch 13, batch 7300, giga_loss[loss=0.3885, simple_loss=0.4175, pruned_loss=0.1798, over 23409.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3804, pruned_loss=0.1246, over 5654338.20 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3552, pruned_loss=0.09949, over 5593738.47 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3826, pruned_loss=0.127, over 5651214.54 frames. ], batch size: 705, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:02:29,737 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=554781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:02:32,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=554784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:02:58,239 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=554813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:02:59,987 INFO [train.py:968] (0/2) Epoch 13, batch 7350, giga_loss[loss=0.2702, simple_loss=0.3429, pruned_loss=0.09873, over 28777.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3796, pruned_loss=0.1247, over 5658822.94 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3552, pruned_loss=0.09946, over 5590350.35 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3818, pruned_loss=0.127, over 5659928.12 frames. ], batch size: 92, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:03:22,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=554831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:03:26,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=554834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:03:30,952 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:03:35,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=554846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:03:50,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.756e+02 1.738e+03 2.058e+03 3.330e+03 1.345e+04, threshold=4.117e+03, percent-clipped=16.0
+2023-03-06 16:03:51,594 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=554863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:03:52,866 INFO [train.py:968] (0/2) Epoch 13, batch 7400, giga_loss[loss=0.2988, simple_loss=0.3626, pruned_loss=0.1175, over 28754.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3771, pruned_loss=0.1242, over 5654910.63 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3547, pruned_loss=0.09919, over 5588504.19 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3798, pruned_loss=0.1269, over 5658353.43 frames. ], batch size: 119, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:04:39,541 INFO [train.py:968] (0/2) Epoch 13, batch 7450, giga_loss[loss=0.2585, simple_loss=0.3432, pruned_loss=0.08694, over 28360.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3769, pruned_loss=0.125, over 5655379.13 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.355, pruned_loss=0.0995, over 5581493.59 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3789, pruned_loss=0.127, over 5665223.99 frames. ], batch size: 65, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:04:53,862 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:05:28,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.516e+02 1.458e+03 1.699e+03 2.321e+03 4.208e+03, threshold=3.398e+03, percent-clipped=1.0
+2023-03-06 16:05:34,389 INFO [train.py:968] (0/2) Epoch 13, batch 7500, giga_loss[loss=0.3574, simple_loss=0.4192, pruned_loss=0.1478, over 28937.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3772, pruned_loss=0.1242, over 5649583.72 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3552, pruned_loss=0.09956, over 5587876.07 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.379, pruned_loss=0.1263, over 5653125.81 frames. ], batch size: 145, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:05:58,682 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=554989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:06:00,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=554992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:06:01,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=554993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:06:22,419 INFO [train.py:968] (0/2) Epoch 13, batch 7550, giga_loss[loss=0.393, simple_loss=0.4262, pruned_loss=0.1798, over 26613.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3769, pruned_loss=0.123, over 5661230.78 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3552, pruned_loss=0.09953, over 5593664.38 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3788, pruned_loss=0.125, over 5660200.91 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:06:30,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=555021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:07:05,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.847e+02 1.463e+03 1.851e+03 2.671e+03 6.981e+03, threshold=3.702e+03, percent-clipped=14.0
+2023-03-06 16:07:09,821 INFO [train.py:968] (0/2) Epoch 13, batch 7600, giga_loss[loss=0.293, simple_loss=0.365, pruned_loss=0.1105, over 28852.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3756, pruned_loss=0.1218, over 5674753.31 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3553, pruned_loss=0.09962, over 5600031.09 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3774, pruned_loss=0.1238, over 5669505.66 frames. ], batch size: 199, lr: 2.51e-03, grad_scale: 8.0
+2023-03-06 16:07:39,406 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-06 16:07:46,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=555103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:07:56,819 INFO [train.py:968] (0/2) Epoch 13, batch 7650, giga_loss[loss=0.3222, simple_loss=0.385, pruned_loss=0.1297, over 28890.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3745, pruned_loss=0.1216, over 5681490.70 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3551, pruned_loss=0.09948, over 5608932.74 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3768, pruned_loss=0.124, over 5671686.86 frames. ], batch size: 227, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:08:16,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=555136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:08:19,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=555139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:08:46,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.945e+02 1.606e+03 2.035e+03 2.854e+03 6.653e+03, threshold=4.070e+03, percent-clipped=11.0
+2023-03-06 16:08:49,813 INFO [train.py:968] (0/2) Epoch 13, batch 7700, giga_loss[loss=0.3917, simple_loss=0.4198, pruned_loss=0.1818, over 26573.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3732, pruned_loss=0.1219, over 5669004.35 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3548, pruned_loss=0.09925, over 5613093.42 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3755, pruned_loss=0.1243, over 5658705.26 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:08:54,460 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=555168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:08:55,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5553, 1.7366, 1.4064, 1.8766], device='cuda:0'), covar=tensor([0.2366, 0.2462, 0.2654, 0.2286], device='cuda:0'), in_proj_covar=tensor([0.1337, 0.0989, 0.1176, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 16:09:41,727 INFO [train.py:968] (0/2) Epoch 13, batch 7750, giga_loss[loss=0.2993, simple_loss=0.3655, pruned_loss=0.1166, over 28726.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3714, pruned_loss=0.1215, over 5666289.75 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3547, pruned_loss=0.09905, over 5618725.92 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3736, pruned_loss=0.124, over 5653773.56 frames. ], batch size: 242, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:09:43,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=555216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:10:14,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=555246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:10:18,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=555249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:10:29,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.772e+02 1.640e+03 2.073e+03 3.338e+03 8.548e+03, threshold=4.147e+03, percent-clipped=18.0
+2023-03-06 16:10:33,775 INFO [train.py:968] (0/2) Epoch 13, batch 7800, giga_loss[loss=0.393, simple_loss=0.4149, pruned_loss=0.1855, over 23893.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3697, pruned_loss=0.1209, over 5663410.05 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3545, pruned_loss=0.09892, over 5623063.50 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3719, pruned_loss=0.1234, over 5650160.13 frames. ], batch size: 705, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:10:44,878 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=555278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:10:50,414 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-06 16:11:09,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=555302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:11:20,962 INFO [train.py:968] (0/2) Epoch 13, batch 7850, giga_loss[loss=0.3542, simple_loss=0.3851, pruned_loss=0.1617, over 23710.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3692, pruned_loss=0.1212, over 5662064.39 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3544, pruned_loss=0.0989, over 5630358.55 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3714, pruned_loss=0.1237, over 5645665.96 frames. ], batch size: 705, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:11:23,304 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 16:12:04,971 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=555359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:12:07,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=555362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:12:07,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.968e+02 1.682e+03 2.518e+03 3.769e+03 1.146e+04, threshold=5.036e+03, percent-clipped=20.0
+2023-03-06 16:12:09,324 INFO [train.py:968] (0/2) Epoch 13, batch 7900, libri_loss[loss=0.2968, simple_loss=0.375, pruned_loss=0.1093, over 29232.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3697, pruned_loss=0.1216, over 5666728.57 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3546, pruned_loss=0.09905, over 5633727.73 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3714, pruned_loss=0.1237, over 5651116.82 frames. ], batch size: 97, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:12:36,481 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=555391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:12:50,754 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-06 16:12:57,152 INFO [train.py:968] (0/2) Epoch 13, batch 7950, giga_loss[loss=0.2841, simple_loss=0.3679, pruned_loss=0.1002, over 28871.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3703, pruned_loss=0.1213, over 5664015.73 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3547, pruned_loss=0.09925, over 5633916.88 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3722, pruned_loss=0.1236, over 5652443.77 frames. ], batch size: 174, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:13:07,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5525, 2.2639, 1.6889, 0.6259], device='cuda:0'), covar=tensor([0.3582, 0.2166, 0.3102, 0.4610], device='cuda:0'), in_proj_covar=tensor([0.1598, 0.1517, 0.1507, 0.1308], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 16:13:27,035 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=555445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:13:31,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=555448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:13:33,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1960, 1.8160, 1.4160, 0.3847], device='cuda:0'), covar=tensor([0.3350, 0.2264, 0.3232, 0.4551], device='cuda:0'), in_proj_covar=tensor([0.1603, 0.1523, 0.1510, 0.1310], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 16:13:45,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.781e+02 1.463e+03 1.964e+03 2.789e+03 5.315e+03, threshold=3.928e+03, percent-clipped=3.0
+2023-03-06 16:13:46,629 INFO [train.py:968] (0/2) Epoch 13, batch 8000, giga_loss[loss=0.3305, simple_loss=0.3841, pruned_loss=0.1384, over 28339.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3714, pruned_loss=0.121, over 5670991.71 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3548, pruned_loss=0.09925, over 5636358.89 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3729, pruned_loss=0.123, over 5660024.02 frames. ], batch size: 368, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:13:47,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5614, 4.4442, 1.6594, 1.8344], device='cuda:0'), covar=tensor([0.0928, 0.0393, 0.0852, 0.1194], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0523, 0.0350, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 16:13:58,484 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=555477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:14:34,339 INFO [train.py:968] (0/2) Epoch 13, batch 8050, giga_loss[loss=0.3902, simple_loss=0.4402, pruned_loss=0.1701, over 28843.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3714, pruned_loss=0.1201, over 5683248.27 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3545, pruned_loss=0.09898, over 5640501.73 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3731, pruned_loss=0.1222, over 5671450.69 frames. ], batch size: 199, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:15:00,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 16:15:19,742 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.164e+02 1.462e+03 1.926e+03 2.674e+03 5.116e+03, threshold=3.852e+03, percent-clipped=12.0
+2023-03-06 16:15:21,332 INFO [train.py:968] (0/2) Epoch 13, batch 8100, libri_loss[loss=0.3331, simple_loss=0.4067, pruned_loss=0.1297, over 29564.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3717, pruned_loss=0.1205, over 5689572.66 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3547, pruned_loss=0.09905, over 5650900.00 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3735, pruned_loss=0.1229, over 5671884.76 frames. ], batch size: 83, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:16:12,733 INFO [train.py:968] (0/2) Epoch 13, batch 8150, giga_loss[loss=0.4601, simple_loss=0.4545, pruned_loss=0.2329, over 23621.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.376, pruned_loss=0.1251, over 5674958.52 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3544, pruned_loss=0.09897, over 5654360.91 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3781, pruned_loss=0.1276, over 5658099.70 frames. ], batch size: 705, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:17:03,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.960e+02 1.742e+03 2.346e+03 3.487e+03 1.050e+04, threshold=4.692e+03, percent-clipped=21.0
+2023-03-06 16:17:05,870 INFO [train.py:968] (0/2) Epoch 13, batch 8200, giga_loss[loss=0.2509, simple_loss=0.3226, pruned_loss=0.08958, over 28532.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3759, pruned_loss=0.1268, over 5666987.91 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.354, pruned_loss=0.09862, over 5660427.67 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3786, pruned_loss=0.1298, over 5648096.76 frames. ], batch size: 71, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:17:37,909 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-06 16:17:45,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2199, 1.2891, 1.2028, 1.4662], device='cuda:0'), covar=tensor([0.0730, 0.0383, 0.0331, 0.0763], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 16:17:49,470 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0340, 3.1176, 2.1587, 0.9866], device='cuda:0'), covar=tensor([0.5049, 0.1965, 0.2384, 0.5109], device='cuda:0'), in_proj_covar=tensor([0.1597, 0.1514, 0.1503, 0.1306], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 16:17:54,056 INFO [train.py:968] (0/2) Epoch 13, batch 8250, giga_loss[loss=0.2964, simple_loss=0.3615, pruned_loss=0.1157, over 28706.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3772, pruned_loss=0.128, over 5682578.26 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3542, pruned_loss=0.09868, over 5668272.63 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.38, pruned_loss=0.1315, over 5660472.53 frames. ], batch size: 242, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:18:34,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3477, 3.0813, 1.4850, 1.4193], device='cuda:0'), covar=tensor([0.0920, 0.0342, 0.0829, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0521, 0.0348, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 16:18:41,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.648e+03 2.391e+03 3.343e+03 1.287e+04, threshold=4.782e+03, percent-clipped=11.0
+2023-03-06 16:18:42,667 INFO [train.py:968] (0/2) Epoch 13, batch 8300, giga_loss[loss=0.2668, simple_loss=0.3452, pruned_loss=0.09419, over 28842.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3777, pruned_loss=0.129, over 5675696.76 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3538, pruned_loss=0.0985, over 5672960.68 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3809, pruned_loss=0.1327, over 5653838.81 frames. ], batch size: 174, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:19:30,542 INFO [train.py:968] (0/2) Epoch 13, batch 8350, giga_loss[loss=0.3654, simple_loss=0.4022, pruned_loss=0.1643, over 29079.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3764, pruned_loss=0.1277, over 5680110.82 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.354, pruned_loss=0.09851, over 5678396.88 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3792, pruned_loss=0.1314, over 5657616.95 frames. ], batch size: 128, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:20:02,554 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1403, 1.1536, 3.8725, 3.1750], device='cuda:0'), covar=tensor([0.1802, 0.2838, 0.0476, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0605, 0.0878, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 16:20:12,818 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.237e+02 1.459e+03 1.958e+03 2.782e+03 5.377e+03, threshold=3.915e+03, percent-clipped=2.0
+2023-03-06 16:20:14,138 INFO [train.py:968] (0/2) Epoch 13, batch 8400, giga_loss[loss=0.3099, simple_loss=0.3787, pruned_loss=0.1206, over 28811.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3753, pruned_loss=0.1253, over 5689696.50 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3539, pruned_loss=0.09841, over 5681977.22 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3782, pruned_loss=0.1289, over 5668629.80 frames. ], batch size: 284, lr: 2.51e-03, grad_scale: 8.0
+2023-03-06 16:20:48,315 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-06 16:20:52,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 16:21:00,697 INFO [train.py:968] (0/2) Epoch 13, batch 8450, giga_loss[loss=0.2816, simple_loss=0.3463, pruned_loss=0.1084, over 28646.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3731, pruned_loss=0.1228, over 5688924.27 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3539, pruned_loss=0.09838, over 5683614.74 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.376, pruned_loss=0.1264, over 5670553.54 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:21:23,911 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 16:21:24,321 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=555942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:21:42,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.461e+02 1.581e+03 2.049e+03 2.797e+03 7.547e+03, threshold=4.098e+03, percent-clipped=11.0
+2023-03-06 16:21:43,471 INFO [train.py:968] (0/2) Epoch 13, batch 8500, giga_loss[loss=0.2887, simple_loss=0.3517, pruned_loss=0.1128, over 28727.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.371, pruned_loss=0.1212, over 5693930.79 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3539, pruned_loss=0.09831, over 5689212.62 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3737, pruned_loss=0.1247, over 5674345.12 frames. ], batch size: 92, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:22:12,862 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-06 16:22:19,309 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-556000.pt
+2023-03-06 16:22:25,669 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-06 16:22:33,849 INFO [train.py:968] (0/2) Epoch 13, batch 8550, giga_loss[loss=0.2871, simple_loss=0.3504, pruned_loss=0.1119, over 28653.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3686, pruned_loss=0.1205, over 5689669.03 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3535, pruned_loss=0.09817, over 5692669.84 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3714, pruned_loss=0.1238, over 5670998.32 frames. ], batch size: 85, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:22:36,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9298, 1.1243, 1.0829, 0.8151], device='cuda:0'), covar=tensor([0.1626, 0.1925, 0.1078, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.1754, 0.1672, 0.1642, 0.1733], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 16:23:21,174 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.308e+02 1.515e+03 1.973e+03 2.690e+03 1.063e+04, threshold=3.945e+03, percent-clipped=11.0
+2023-03-06 16:23:22,420 INFO [train.py:968] (0/2) Epoch 13, batch 8600, giga_loss[loss=0.3391, simple_loss=0.4006, pruned_loss=0.1388, over 28564.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3689, pruned_loss=0.121, over 5674087.65 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3537, pruned_loss=0.09832, over 5687184.07 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3715, pruned_loss=0.1242, over 5663747.16 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:24:07,849 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-06 16:24:17,042 INFO [train.py:968] (0/2) Epoch 13, batch 8650, giga_loss[loss=0.4432, simple_loss=0.4671, pruned_loss=0.2097, over 28741.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3726, pruned_loss=0.1233, over 5673589.77 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3531, pruned_loss=0.09796, over 5687911.50 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3755, pruned_loss=0.1266, over 5664456.49 frames. ], batch size: 284, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:24:20,214 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=556118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:25:01,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.334e+02 1.593e+03 2.497e+03 3.705e+03 1.320e+04, threshold=4.994e+03, percent-clipped=23.0
+2023-03-06 16:25:02,603 INFO [train.py:968] (0/2) Epoch 13, batch 8700, giga_loss[loss=0.3044, simple_loss=0.3843, pruned_loss=0.1123, over 28522.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3762, pruned_loss=0.1228, over 5680412.81 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3533, pruned_loss=0.09803, over 5695690.75 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.379, pruned_loss=0.1262, over 5665782.32 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:25:51,026 INFO [train.py:968] (0/2) Epoch 13, batch 8750, giga_loss[loss=0.3509, simple_loss=0.4094, pruned_loss=0.1462, over 28276.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.378, pruned_loss=0.1229, over 5683908.23 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3528, pruned_loss=0.09779, over 5699704.75 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3812, pruned_loss=0.1264, over 5668330.60 frames. ], batch size: 368, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:26:36,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.455e+02 1.489e+03 2.248e+03 3.358e+03 9.282e+03, threshold=4.496e+03, percent-clipped=13.0
+2023-03-06 16:26:38,068 INFO [train.py:968] (0/2) Epoch 13, batch 8800, giga_loss[loss=0.3293, simple_loss=0.3952, pruned_loss=0.1318, over 28723.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3802, pruned_loss=0.1248, over 5687776.97 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3529, pruned_loss=0.0978, over 5703560.59 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3831, pruned_loss=0.1281, over 5671615.76 frames. ], batch size: 243, lr: 2.51e-03, grad_scale: 8.0
+2023-03-06 16:27:24,126 INFO [train.py:968] (0/2) Epoch 13, batch 8850, giga_loss[loss=0.3207, simple_loss=0.383, pruned_loss=0.1292, over 28776.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3795, pruned_loss=0.1247, over 5696351.39 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3525, pruned_loss=0.09749, over 5708066.79 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3828, pruned_loss=0.1282, over 5679022.02 frames. ], batch size: 243, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:27:25,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=556317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:28:11,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.189e+02 1.517e+03 2.221e+03 3.368e+03 9.927e+03, threshold=4.442e+03, percent-clipped=9.0
+2023-03-06 16:28:11,112 INFO [train.py:968] (0/2) Epoch 13, batch 8900, giga_loss[loss=0.3342, simple_loss=0.3903, pruned_loss=0.1391, over 28697.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3805, pruned_loss=0.127, over 5695488.50 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3526, pruned_loss=0.09763, over 5711475.82 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3836, pruned_loss=0.1302, over 5678634.60 frames. ], batch size: 242, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:29:06,681 INFO [train.py:968] (0/2) Epoch 13, batch 8950, giga_loss[loss=0.3752, simple_loss=0.4138, pruned_loss=0.1683, over 26662.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3792, pruned_loss=0.1267, over 5689699.91 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3523, pruned_loss=0.09745, over 5713890.94 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3822, pruned_loss=0.1298, over 5673974.07 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:29:31,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5478, 1.8775, 1.8403, 1.3492], device='cuda:0'), covar=tensor([0.1683, 0.2347, 0.1416, 0.1664], device='cuda:0'), in_proj_covar=tensor([0.0844, 0.0699, 0.0885, 0.0789], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 16:29:43,392 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=556453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:29:51,113 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=556460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:29:53,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=556463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:29:55,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.584e+03 2.065e+03 2.834e+03 6.813e+03, threshold=4.130e+03, percent-clipped=11.0
+2023-03-06 16:29:55,881 INFO [train.py:968] (0/2) Epoch 13, batch 9000, giga_loss[loss=0.3911, simple_loss=0.4231, pruned_loss=0.1796, over 27966.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3779, pruned_loss=0.1271, over 5689127.72 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3526, pruned_loss=0.0976, over 5716083.05 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3804, pruned_loss=0.1298, over 5674665.63 frames. ], batch size: 412, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:29:55,885 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 16:30:04,609 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2177, simple_loss=0.325, pruned_loss=0.05519, over 944034.00 frames. 
+2023-03-06 16:30:04,609 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 16:30:10,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4239, 3.5969, 1.6288, 1.4637], device='cuda:0'), covar=tensor([0.0923, 0.0344, 0.0807, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0523, 0.0348, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 16:30:29,094 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=556492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:30:30,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=556493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:30:54,228 INFO [train.py:968] (0/2) Epoch 13, batch 9050, giga_loss[loss=0.2797, simple_loss=0.3434, pruned_loss=0.108, over 28963.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.376, pruned_loss=0.1263, over 5673178.69 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3528, pruned_loss=0.0978, over 5707792.93 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3781, pruned_loss=0.1287, over 5668100.20 frames. ], batch size: 106, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:31:07,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2778, 1.4137, 3.6846, 3.0717], device='cuda:0'), covar=tensor([0.1528, 0.2362, 0.0448, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0682, 0.0602, 0.0875, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 16:31:35,348 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 16:31:45,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.433e+02 1.691e+03 2.074e+03 2.959e+03 6.461e+03, threshold=4.148e+03, percent-clipped=12.0
+2023-03-06 16:31:45,876 INFO [train.py:968] (0/2) Epoch 13, batch 9100, giga_loss[loss=0.2928, simple_loss=0.3587, pruned_loss=0.1134, over 28739.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3772, pruned_loss=0.1274, over 5664535.19 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3533, pruned_loss=0.09812, over 5699577.16 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3788, pruned_loss=0.1294, over 5667923.60 frames. ], batch size: 99, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:32:35,676 INFO [train.py:968] (0/2) Epoch 13, batch 9150, giga_loss[loss=0.2637, simple_loss=0.3298, pruned_loss=0.09881, over 28269.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3755, pruned_loss=0.1266, over 5669624.78 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3528, pruned_loss=0.09785, over 5703513.15 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3778, pruned_loss=0.1293, over 5667913.07 frames. ], batch size: 77, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:32:56,328 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=556636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:33:01,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=556639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:33:26,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.860e+02 1.628e+03 2.065e+03 2.941e+03 9.698e+03, threshold=4.131e+03, percent-clipped=10.0
+2023-03-06 16:33:26,958 INFO [train.py:968] (0/2) Epoch 13, batch 9200, giga_loss[loss=0.331, simple_loss=0.3847, pruned_loss=0.1387, over 28293.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3748, pruned_loss=0.1267, over 5672501.06 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3527, pruned_loss=0.09774, over 5704377.59 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3769, pruned_loss=0.129, over 5670344.27 frames. ], batch size: 77, lr: 2.51e-03, grad_scale: 8.0
+2023-03-06 16:33:30,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=556668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:34:10,568 INFO [train.py:968] (0/2) Epoch 13, batch 9250, giga_loss[loss=0.2969, simple_loss=0.3672, pruned_loss=0.1133, over 28596.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3747, pruned_loss=0.1261, over 5687947.29 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.353, pruned_loss=0.0981, over 5713001.14 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.377, pruned_loss=0.1288, over 5677124.95 frames. ], batch size: 242, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:34:15,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4833, 1.6105, 1.5458, 1.5268], device='cuda:0'), covar=tensor([0.1414, 0.1750, 0.1847, 0.1657], device='cuda:0'), in_proj_covar=tensor([0.0442, 0.0734, 0.0679, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 16:34:26,716 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=556731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:35:00,405 INFO [train.py:968] (0/2) Epoch 13, batch 9300, giga_loss[loss=0.2798, simple_loss=0.353, pruned_loss=0.1033, over 29069.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3752, pruned_loss=0.1258, over 5674791.67 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3523, pruned_loss=0.0978, over 5707635.44 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3782, pruned_loss=0.129, over 5670512.38 frames. ], batch size: 113, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:35:01,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.332e+02 1.493e+03 1.991e+03 2.514e+03 5.707e+03, threshold=3.982e+03, percent-clipped=4.0
+2023-03-06 16:35:29,890 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=556796.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 16:35:48,487 INFO [train.py:968] (0/2) Epoch 13, batch 9350, giga_loss[loss=0.3557, simple_loss=0.4055, pruned_loss=0.1529, over 27423.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3768, pruned_loss=0.1269, over 5669773.44 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3522, pruned_loss=0.09769, over 5709774.23 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3801, pruned_loss=0.1305, over 5663140.36 frames. ], batch size: 472, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:35:57,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5635, 5.4042, 5.1084, 2.7338], device='cuda:0'), covar=tensor([0.0403, 0.0549, 0.0615, 0.1561], device='cuda:0'), in_proj_covar=tensor([0.1092, 0.1018, 0.0885, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 16:35:59,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2231, 0.8065, 0.8459, 1.3798], device='cuda:0'), covar=tensor([0.0759, 0.0351, 0.0344, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 16:35:59,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=556828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:36:34,866 INFO [train.py:968] (0/2) Epoch 13, batch 9400, giga_loss[loss=0.305, simple_loss=0.377, pruned_loss=0.1165, over 28961.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3765, pruned_loss=0.1268, over 5670512.07 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3522, pruned_loss=0.09765, over 5708342.72 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3797, pruned_loss=0.1304, over 5665757.33 frames. ], batch size: 136, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:36:35,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.644e+03 2.195e+03 3.100e+03 9.291e+03, threshold=4.390e+03, percent-clipped=12.0
+2023-03-06 16:37:21,495 INFO [train.py:968] (0/2) Epoch 13, batch 9450, giga_loss[loss=0.2873, simple_loss=0.3659, pruned_loss=0.1044, over 28553.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.378, pruned_loss=0.1251, over 5669590.95 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3527, pruned_loss=0.098, over 5699375.55 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3806, pruned_loss=0.1282, over 5672854.88 frames. ], batch size: 85, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:37:38,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3232, 3.2420, 1.4742, 1.4488], device='cuda:0'), covar=tensor([0.0923, 0.0400, 0.0882, 0.1309], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0521, 0.0348, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 16:38:06,026 INFO [train.py:968] (0/2) Epoch 13, batch 9500, giga_loss[loss=0.3167, simple_loss=0.3946, pruned_loss=0.1194, over 28957.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3791, pruned_loss=0.1238, over 5664205.00 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3526, pruned_loss=0.09794, over 5693343.40 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3819, pruned_loss=0.127, over 5670993.88 frames. ], batch size: 106, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:38:07,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.740e+02 1.312e+03 1.927e+03 2.613e+03 1.356e+04, threshold=3.853e+03, percent-clipped=15.0
+2023-03-06 16:38:11,202 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=556971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:38:13,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=556974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:38:42,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=557003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:38:53,573 INFO [train.py:968] (0/2) Epoch 13, batch 9550, giga_loss[loss=0.4469, simple_loss=0.467, pruned_loss=0.2134, over 27513.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3827, pruned_loss=0.1265, over 5667937.11 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3523, pruned_loss=0.09774, over 5697780.12 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3858, pruned_loss=0.1298, over 5668716.92 frames. ], batch size: 472, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:39:32,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-06 16:39:41,928 INFO [train.py:968] (0/2) Epoch 13, batch 9600, giga_loss[loss=0.3081, simple_loss=0.3704, pruned_loss=0.1229, over 28602.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3837, pruned_loss=0.1278, over 5673586.22 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3521, pruned_loss=0.09762, over 5701147.12 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3868, pruned_loss=0.1311, over 5670686.63 frames. ], batch size: 85, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:39:44,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.033e+02 1.574e+03 1.986e+03 2.594e+03 9.594e+03, threshold=3.972e+03, percent-clipped=7.0
+2023-03-06 16:39:47,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3078, 1.9061, 1.4266, 0.4362], device='cuda:0'), covar=tensor([0.3085, 0.2184, 0.3199, 0.4180], device='cuda:0'), in_proj_covar=tensor([0.1587, 0.1510, 0.1494, 0.1307], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 16:40:22,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=557106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:40:31,080 INFO [train.py:968] (0/2) Epoch 13, batch 9650, giga_loss[loss=0.278, simple_loss=0.3535, pruned_loss=0.1012, over 28877.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3849, pruned_loss=0.1297, over 5672129.87 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3523, pruned_loss=0.09759, over 5703029.64 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.388, pruned_loss=0.133, over 5667442.76 frames. ], batch size: 174, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:41:16,003 INFO [train.py:968] (0/2) Epoch 13, batch 9700, giga_loss[loss=0.3315, simple_loss=0.3927, pruned_loss=0.1351, over 28785.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3829, pruned_loss=0.1288, over 5663408.81 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.352, pruned_loss=0.09734, over 5705701.89 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3868, pruned_loss=0.1328, over 5655870.60 frames. ], batch size: 186, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:41:18,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.060e+02 1.670e+03 2.260e+03 3.580e+03 1.018e+04, threshold=4.519e+03, percent-clipped=15.0
+2023-03-06 16:41:20,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=557171.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 16:41:49,743 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7466, 2.0277, 1.5472, 1.7344], device='cuda:0'), covar=tensor([0.1729, 0.1883, 0.1987, 0.1933], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0733, 0.0676, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 16:42:00,526 INFO [train.py:968] (0/2) Epoch 13, batch 9750, giga_loss[loss=0.3178, simple_loss=0.3804, pruned_loss=0.1276, over 28803.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3809, pruned_loss=0.1268, over 5666782.27 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3512, pruned_loss=0.09699, over 5710519.39 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3853, pruned_loss=0.1309, over 5655778.32 frames. ], batch size: 284, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:42:15,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4124, 1.5524, 1.5842, 1.4966], device='cuda:0'), covar=tensor([0.1138, 0.1120, 0.1413, 0.1179], device='cuda:0'), in_proj_covar=tensor([0.0439, 0.0732, 0.0675, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 16:42:31,341 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=557249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:42:33,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=557252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:42:44,364 INFO [train.py:968] (0/2) Epoch 13, batch 9800, giga_loss[loss=0.272, simple_loss=0.3616, pruned_loss=0.09121, over 28957.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3814, pruned_loss=0.1254, over 5666220.45 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3516, pruned_loss=0.09736, over 5709490.91 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3854, pruned_loss=0.1292, over 5657529.09 frames. ], batch size: 164, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:42:46,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.729e+02 1.298e+03 1.734e+03 2.428e+03 8.224e+03, threshold=3.467e+03, percent-clipped=3.0
+2023-03-06 16:42:55,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2126, 1.5163, 1.1935, 1.0623], device='cuda:0'), covar=tensor([0.2565, 0.2580, 0.2951, 0.2145], device='cuda:0'), in_proj_covar=tensor([0.1343, 0.0996, 0.1186, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 16:42:59,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=557281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:43:30,062 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=557314.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 16:43:30,409 INFO [train.py:968] (0/2) Epoch 13, batch 9850, giga_loss[loss=0.3313, simple_loss=0.4019, pruned_loss=0.1303, over 28657.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3809, pruned_loss=0.1242, over 5675899.93 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3511, pruned_loss=0.09713, over 5713960.96 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3852, pruned_loss=0.1281, over 5663905.16 frames. ], batch size: 336, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:43:31,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=557317.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 16:43:56,666 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=557338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:44:03,891 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=557346.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 16:44:22,133 INFO [train.py:968] (0/2) Epoch 13, batch 9900, giga_loss[loss=0.3126, simple_loss=0.377, pruned_loss=0.1241, over 28926.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.382, pruned_loss=0.1256, over 5668096.21 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.351, pruned_loss=0.09696, over 5717165.94 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3863, pruned_loss=0.1295, over 5654949.40 frames. ], batch size: 227, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:44:26,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.885e+02 1.548e+03 1.980e+03 2.999e+03 8.037e+03, threshold=3.960e+03, percent-clipped=16.0
+2023-03-06 16:44:37,524 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-06 16:45:14,706 INFO [train.py:968] (0/2) Epoch 13, batch 9950, libri_loss[loss=0.2744, simple_loss=0.3544, pruned_loss=0.09718, over 29460.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3821, pruned_loss=0.1263, over 5671831.60 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3512, pruned_loss=0.097, over 5719634.00 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3858, pruned_loss=0.1298, over 5658512.24 frames. ], batch size: 85, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:46:07,217 INFO [train.py:968] (0/2) Epoch 13, batch 10000, giga_loss[loss=0.3712, simple_loss=0.4157, pruned_loss=0.1633, over 28600.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3815, pruned_loss=0.1275, over 5659389.12 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3515, pruned_loss=0.0972, over 5711042.56 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3843, pruned_loss=0.1302, over 5655948.47 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:46:09,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.070e+02 1.526e+03 2.153e+03 2.915e+03 9.046e+03, threshold=4.307e+03, percent-clipped=17.0
+2023-03-06 16:46:52,718 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-06 16:46:54,037 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 16:46:56,945 INFO [train.py:968] (0/2) Epoch 13, batch 10050, giga_loss[loss=0.2952, simple_loss=0.3611, pruned_loss=0.1147, over 28528.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.379, pruned_loss=0.1262, over 5661791.27 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3516, pruned_loss=0.09711, over 5711413.92 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3817, pruned_loss=0.129, over 5657652.00 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:47:37,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3181, 1.6255, 1.3068, 1.2794], device='cuda:0'), covar=tensor([0.2470, 0.2402, 0.2672, 0.2097], device='cuda:0'), in_proj_covar=tensor([0.1338, 0.0993, 0.1181, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 16:47:47,099 INFO [train.py:968] (0/2) Epoch 13, batch 10100, giga_loss[loss=0.3079, simple_loss=0.3668, pruned_loss=0.1245, over 28514.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3754, pruned_loss=0.1239, over 5672734.26 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3512, pruned_loss=0.09682, over 5714668.18 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3784, pruned_loss=0.127, over 5665550.34 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:47:51,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.230e+02 1.718e+03 2.313e+03 3.363e+03 1.145e+04, threshold=4.626e+03, percent-clipped=12.0
+2023-03-06 16:48:03,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5139, 3.3509, 3.1837, 2.0795], device='cuda:0'), covar=tensor([0.0682, 0.0820, 0.0773, 0.1622], device='cuda:0'), in_proj_covar=tensor([0.1097, 0.1022, 0.0889, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 16:48:34,249 INFO [train.py:968] (0/2) Epoch 13, batch 10150, giga_loss[loss=0.3244, simple_loss=0.3912, pruned_loss=0.1289, over 28586.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3759, pruned_loss=0.1255, over 5673014.51 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3517, pruned_loss=0.09706, over 5718778.09 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3784, pruned_loss=0.1285, over 5662665.84 frames. ], batch size: 242, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:49:23,129 INFO [train.py:968] (0/2) Epoch 13, batch 10200, giga_loss[loss=0.2986, simple_loss=0.3621, pruned_loss=0.1176, over 28314.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3748, pruned_loss=0.1249, over 5662614.34 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.352, pruned_loss=0.09716, over 5711766.17 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.377, pruned_loss=0.1276, over 5659607.81 frames. ], batch size: 368, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:49:25,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.742e+02 1.655e+03 2.289e+03 3.285e+03 1.314e+04, threshold=4.578e+03, percent-clipped=6.0
+2023-03-06 16:50:11,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=557713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:50:13,060 INFO [train.py:968] (0/2) Epoch 13, batch 10250, giga_loss[loss=0.2655, simple_loss=0.3443, pruned_loss=0.09334, over 28838.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3715, pruned_loss=0.1212, over 5660617.43 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3519, pruned_loss=0.09701, over 5714772.00 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3738, pruned_loss=0.124, over 5654556.72 frames. ], batch size: 186, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:50:33,210 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-06 16:50:59,107 INFO [train.py:968] (0/2) Epoch 13, batch 10300, giga_loss[loss=0.2742, simple_loss=0.3489, pruned_loss=0.09981, over 28540.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3683, pruned_loss=0.1181, over 5668975.87 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3517, pruned_loss=0.09694, over 5723925.61 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3713, pruned_loss=0.1216, over 5653237.51 frames. ], batch size: 71, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 16:51:02,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.547e+02 1.252e+03 1.638e+03 2.177e+03 5.715e+03, threshold=3.277e+03, percent-clipped=5.0
+2023-03-06 16:51:42,800 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-06 16:51:49,028 INFO [train.py:968] (0/2) Epoch 13, batch 10350, giga_loss[loss=0.2702, simple_loss=0.3378, pruned_loss=0.1013, over 28803.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.368, pruned_loss=0.1181, over 5672295.66 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3519, pruned_loss=0.09712, over 5723846.44 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3704, pruned_loss=0.1212, over 5658574.62 frames. ], batch size: 119, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 16:52:33,196 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=557856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:52:36,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=557859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:52:38,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5240, 1.7137, 1.3927, 1.6994], device='cuda:0'), covar=tensor([0.2293, 0.2318, 0.2500, 0.2133], device='cuda:0'), in_proj_covar=tensor([0.1347, 0.0999, 0.1190, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 16:52:43,159 INFO [train.py:968] (0/2) Epoch 13, batch 10400, giga_loss[loss=0.2552, simple_loss=0.3362, pruned_loss=0.08709, over 28812.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3655, pruned_loss=0.1178, over 5668554.73 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3519, pruned_loss=0.09707, over 5726027.22 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3677, pruned_loss=0.1204, over 5655416.81 frames. ], batch size: 186, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:52:46,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.692e+02 1.639e+03 2.205e+03 3.078e+03 5.673e+03, threshold=4.410e+03, percent-clipped=20.0
+2023-03-06 16:53:08,244 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=557888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:53:31,248 INFO [train.py:968] (0/2) Epoch 13, batch 10450, giga_loss[loss=0.2632, simple_loss=0.3351, pruned_loss=0.0956, over 28634.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3655, pruned_loss=0.1181, over 5676237.71 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.352, pruned_loss=0.09722, over 5727383.12 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3673, pruned_loss=0.1204, over 5663491.18 frames. ], batch size: 78, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:54:17,418 INFO [train.py:968] (0/2) Epoch 13, batch 10500, giga_loss[loss=0.2932, simple_loss=0.3675, pruned_loss=0.1094, over 28964.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3684, pruned_loss=0.119, over 5680913.37 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3519, pruned_loss=0.09714, over 5730927.88 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3702, pruned_loss=0.1214, over 5666541.39 frames. ], batch size: 213, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:54:21,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.640e+02 1.497e+03 1.885e+03 2.884e+03 6.547e+03, threshold=3.769e+03, percent-clipped=8.0
+2023-03-06 16:54:48,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1599, 1.1970, 3.8656, 3.1801], device='cuda:0'), covar=tensor([0.1963, 0.2790, 0.0680, 0.0916], device='cuda:0'), in_proj_covar=tensor([0.0681, 0.0601, 0.0872, 0.0792], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 16:54:53,820 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-558000.pt
+2023-03-06 16:55:07,550 INFO [train.py:968] (0/2) Epoch 13, batch 10550, giga_loss[loss=0.3708, simple_loss=0.3975, pruned_loss=0.1721, over 23616.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3698, pruned_loss=0.12, over 5664786.95 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3521, pruned_loss=0.09721, over 5733570.54 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3713, pruned_loss=0.1221, over 5650125.70 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:55:29,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9150, 1.0571, 0.8873, 0.3023], device='cuda:0'), covar=tensor([0.2101, 0.1754, 0.2005, 0.3715], device='cuda:0'), in_proj_covar=tensor([0.1585, 0.1509, 0.1495, 0.1303], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 16:55:57,781 INFO [train.py:968] (0/2) Epoch 13, batch 10600, giga_loss[loss=0.3244, simple_loss=0.3847, pruned_loss=0.132, over 27889.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3695, pruned_loss=0.1198, over 5648368.40 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3524, pruned_loss=0.09731, over 5737436.52 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3709, pruned_loss=0.1219, over 5631217.08 frames. ], batch size: 412, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:56:03,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.441e+02 1.371e+03 1.804e+03 2.503e+03 7.050e+03, threshold=3.609e+03, percent-clipped=9.0
+2023-03-06 16:56:10,778 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=558078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:56:14,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4238, 1.8247, 1.7093, 1.3756], device='cuda:0'), covar=tensor([0.2198, 0.1570, 0.1095, 0.1511], device='cuda:0'), in_proj_covar=tensor([0.1763, 0.1675, 0.1652, 0.1739], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 16:56:43,811 INFO [train.py:968] (0/2) Epoch 13, batch 10650, giga_loss[loss=0.3024, simple_loss=0.375, pruned_loss=0.1149, over 28705.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3702, pruned_loss=0.1207, over 5649821.94 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3524, pruned_loss=0.09725, over 5736066.29 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3718, pruned_loss=0.1231, over 5634744.27 frames. ], batch size: 92, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:57:35,852 INFO [train.py:968] (0/2) Epoch 13, batch 10700, giga_loss[loss=0.311, simple_loss=0.3779, pruned_loss=0.122, over 28948.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3723, pruned_loss=0.1222, over 5645743.13 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3525, pruned_loss=0.09733, over 5738390.40 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3737, pruned_loss=0.1244, over 5630639.58 frames. ], batch size: 227, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:57:41,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.128e+02 1.498e+03 2.043e+03 2.723e+03 7.618e+03, threshold=4.087e+03, percent-clipped=12.0
+2023-03-06 16:58:26,004 INFO [train.py:968] (0/2) Epoch 13, batch 10750, giga_loss[loss=0.3379, simple_loss=0.3885, pruned_loss=0.1436, over 28964.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3749, pruned_loss=0.1237, over 5651884.82 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3526, pruned_loss=0.09739, over 5740161.49 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3763, pruned_loss=0.1258, over 5636457.75 frames. ], batch size: 112, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:58:26,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2350, 1.3585, 1.2015, 1.0697], device='cuda:0'), covar=tensor([0.0802, 0.0437, 0.0955, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0444, 0.0507, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 16:59:03,332 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 16:59:13,898 INFO [train.py:968] (0/2) Epoch 13, batch 10800, giga_loss[loss=0.3019, simple_loss=0.3703, pruned_loss=0.1167, over 28537.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3755, pruned_loss=0.1243, over 5658195.76 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3527, pruned_loss=0.09747, over 5742561.46 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3769, pruned_loss=0.1264, over 5642528.29 frames. ], batch size: 336, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 16:59:18,646 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.599e+03 1.927e+03 2.591e+03 4.914e+03, threshold=3.853e+03, percent-clipped=2.0
+2023-03-06 17:00:06,199 INFO [train.py:968] (0/2) Epoch 13, batch 10850, giga_loss[loss=0.4138, simple_loss=0.4282, pruned_loss=0.1997, over 23655.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3766, pruned_loss=0.1259, over 5658023.52 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3528, pruned_loss=0.09753, over 5745534.23 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3782, pruned_loss=0.128, over 5641392.86 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:00:14,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4572, 1.5327, 1.0930, 1.2136], device='cuda:0'), covar=tensor([0.0654, 0.0428, 0.0932, 0.0978], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0441, 0.0503, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 17:00:57,681 INFO [train.py:968] (0/2) Epoch 13, batch 10900, giga_loss[loss=0.2748, simple_loss=0.3511, pruned_loss=0.09925, over 28671.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3789, pruned_loss=0.126, over 5655993.17 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3534, pruned_loss=0.09798, over 5738071.63 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3799, pruned_loss=0.1276, over 5647776.81 frames. ], batch size: 92, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:01:03,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.662e+03 2.401e+03 3.671e+03 8.435e+03, threshold=4.803e+03, percent-clipped=22.0
+2023-03-06 17:01:10,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3705, 1.5131, 1.2817, 1.2701], device='cuda:0'), covar=tensor([0.1677, 0.1452, 0.1426, 0.1312], device='cuda:0'), in_proj_covar=tensor([0.1767, 0.1679, 0.1658, 0.1744], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 17:01:50,169 INFO [train.py:968] (0/2) Epoch 13, batch 10950, giga_loss[loss=0.3768, simple_loss=0.4227, pruned_loss=0.1655, over 27527.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3778, pruned_loss=0.1256, over 5648047.96 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3535, pruned_loss=0.09827, over 5741253.24 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.379, pruned_loss=0.1272, over 5637145.21 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:02:29,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=558453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:02:44,235 INFO [train.py:968] (0/2) Epoch 13, batch 11000, giga_loss[loss=0.3819, simple_loss=0.4137, pruned_loss=0.1751, over 26463.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3752, pruned_loss=0.1244, over 5663745.46 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.353, pruned_loss=0.09799, over 5744512.19 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.377, pruned_loss=0.1265, over 5650261.97 frames. ], batch size: 555, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:02:48,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.730e+02 1.661e+03 2.078e+03 2.770e+03 8.120e+03, threshold=4.156e+03, percent-clipped=2.0
+2023-03-06 17:03:35,258 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=558507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:03:44,822 INFO [train.py:968] (0/2) Epoch 13, batch 11050, libri_loss[loss=0.2863, simple_loss=0.3676, pruned_loss=0.1025, over 29667.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3742, pruned_loss=0.1241, over 5658753.53 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3531, pruned_loss=0.09797, over 5745407.78 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3758, pruned_loss=0.1261, over 5646119.66 frames. ], batch size: 88, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:04:19,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4049, 1.7378, 1.3694, 1.4947], device='cuda:0'), covar=tensor([0.2254, 0.2118, 0.2322, 0.1944], device='cuda:0'), in_proj_covar=tensor([0.1334, 0.0988, 0.1176, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 17:04:20,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2065, 0.8508, 0.9656, 1.3997], device='cuda:0'), covar=tensor([0.0771, 0.0356, 0.0337, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 17:04:30,968 INFO [train.py:968] (0/2) Epoch 13, batch 11100, giga_loss[loss=0.3848, simple_loss=0.4196, pruned_loss=0.175, over 28963.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3737, pruned_loss=0.124, over 5670184.13 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3527, pruned_loss=0.09758, over 5748530.75 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3759, pruned_loss=0.1266, over 5655509.28 frames. ], batch size: 145, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:04:37,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.696e+03 2.249e+03 3.200e+03 1.177e+04, threshold=4.498e+03, percent-clipped=10.0
+2023-03-06 17:04:59,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=558596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:05:01,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=558599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:05:16,802 INFO [train.py:968] (0/2) Epoch 13, batch 11150, giga_loss[loss=0.3041, simple_loss=0.3598, pruned_loss=0.1242, over 28489.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3727, pruned_loss=0.124, over 5666248.89 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3528, pruned_loss=0.09763, over 5742588.66 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3749, pruned_loss=0.1266, over 5658047.40 frames. ], batch size: 78, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:05:29,440 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=558628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:05:52,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5304, 1.6207, 1.6277, 1.5255], device='cuda:0'), covar=tensor([0.1497, 0.1862, 0.2062, 0.1719], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0738, 0.0680, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 17:06:03,165 INFO [train.py:968] (0/2) Epoch 13, batch 11200, giga_loss[loss=0.3338, simple_loss=0.3854, pruned_loss=0.1411, over 28613.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3729, pruned_loss=0.1244, over 5652802.94 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3529, pruned_loss=0.09763, over 5732463.70 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.375, pruned_loss=0.127, over 5653182.10 frames. ], batch size: 336, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:06:08,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.180e+02 1.518e+03 1.914e+03 2.842e+03 5.787e+03, threshold=3.829e+03, percent-clipped=5.0
+2023-03-06 17:06:54,561 INFO [train.py:968] (0/2) Epoch 13, batch 11250, giga_loss[loss=0.3468, simple_loss=0.4013, pruned_loss=0.1461, over 28430.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.374, pruned_loss=0.1253, over 5654635.03 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.353, pruned_loss=0.0978, over 5732323.64 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3758, pruned_loss=0.1277, over 5653837.98 frames. ], batch size: 85, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:07:39,157 INFO [train.py:968] (0/2) Epoch 13, batch 11300, giga_loss[loss=0.3421, simple_loss=0.3965, pruned_loss=0.1439, over 28903.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3747, pruned_loss=0.1257, over 5651249.99 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3532, pruned_loss=0.09775, over 5721917.88 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3769, pruned_loss=0.1287, over 5656742.91 frames. ], batch size: 186, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:07:45,025 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=558770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:07:45,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.230e+02 1.558e+03 2.072e+03 2.938e+03 9.906e+03, threshold=4.143e+03, percent-clipped=11.0
+2023-03-06 17:07:49,452 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4943, 1.5751, 1.2231, 1.2020], device='cuda:0'), covar=tensor([0.0816, 0.0569, 0.1047, 0.1006], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0445, 0.0504, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 17:08:17,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3739, 1.7505, 1.3619, 1.5784], device='cuda:0'), covar=tensor([0.0745, 0.0312, 0.0318, 0.0792], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0115, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0061, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 17:08:26,699 INFO [train.py:968] (0/2) Epoch 13, batch 11350, giga_loss[loss=0.3592, simple_loss=0.3936, pruned_loss=0.1624, over 23272.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3756, pruned_loss=0.1261, over 5658999.31 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3528, pruned_loss=0.09759, over 5725967.31 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.378, pruned_loss=0.1291, over 5658839.54 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:08:29,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3044, 1.1929, 4.1948, 3.2594], device='cuda:0'), covar=tensor([0.1738, 0.2707, 0.0395, 0.0878], device='cuda:0'), in_proj_covar=tensor([0.0686, 0.0606, 0.0879, 0.0797], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 17:09:15,354 INFO [train.py:968] (0/2) Epoch 13, batch 11400, giga_loss[loss=0.318, simple_loss=0.375, pruned_loss=0.1306, over 29051.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.376, pruned_loss=0.1271, over 5656179.02 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3526, pruned_loss=0.09746, over 5729761.67 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3786, pruned_loss=0.1303, over 5651290.68 frames. ], batch size: 128, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:09:22,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.566e+03 2.049e+03 2.954e+03 7.935e+03, threshold=4.097e+03, percent-clipped=7.0
+2023-03-06 17:09:33,574 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=558881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:09:34,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=558882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:09:34,851 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=558883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:10:05,122 INFO [train.py:968] (0/2) Epoch 13, batch 11450, giga_loss[loss=0.3776, simple_loss=0.3978, pruned_loss=0.1788, over 23731.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3762, pruned_loss=0.1276, over 5655838.74 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3527, pruned_loss=0.09749, over 5733184.69 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3785, pruned_loss=0.1307, over 5647643.36 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:10:48,017 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-06 17:10:53,806 INFO [train.py:968] (0/2) Epoch 13, batch 11500, giga_loss[loss=0.282, simple_loss=0.3617, pruned_loss=0.1012, over 29070.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3757, pruned_loss=0.1268, over 5659903.84 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.353, pruned_loss=0.09767, over 5723709.67 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3779, pruned_loss=0.1297, over 5660720.14 frames. ], batch size: 155, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:10:59,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.585e+02 1.494e+03 1.952e+03 2.823e+03 8.208e+03, threshold=3.904e+03, percent-clipped=8.0
+2023-03-06 17:11:22,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5906, 4.4260, 4.1472, 2.2100], device='cuda:0'), covar=tensor([0.0516, 0.0681, 0.0752, 0.1823], device='cuda:0'), in_proj_covar=tensor([0.1104, 0.1029, 0.0894, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 17:11:44,548 INFO [train.py:968] (0/2) Epoch 13, batch 11550, giga_loss[loss=0.3354, simple_loss=0.3949, pruned_loss=0.1379, over 29045.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.377, pruned_loss=0.1276, over 5649379.82 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3533, pruned_loss=0.09788, over 5715836.81 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3786, pruned_loss=0.1299, over 5656038.87 frames. ], batch size: 136, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:11:55,235 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:11:57,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:12:25,049 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:12:34,009 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=559064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:12:34,430 INFO [train.py:968] (0/2) Epoch 13, batch 11600, giga_loss[loss=0.3193, simple_loss=0.3787, pruned_loss=0.13, over 28960.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3772, pruned_loss=0.1269, over 5665462.40 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3538, pruned_loss=0.09821, over 5720209.34 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3785, pruned_loss=0.1291, over 5665533.10 frames. ], batch size: 106, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:12:43,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.186e+02 1.460e+03 1.902e+03 2.878e+03 1.196e+04, threshold=3.803e+03, percent-clipped=8.0
+2023-03-06 17:13:29,145 INFO [train.py:968] (0/2) Epoch 13, batch 11650, giga_loss[loss=0.3247, simple_loss=0.387, pruned_loss=0.1312, over 28887.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3808, pruned_loss=0.1307, over 5663718.57 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3536, pruned_loss=0.09812, over 5721093.30 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3821, pruned_loss=0.1326, over 5662841.71 frames. ], batch size: 174, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:13:59,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:14:09,030 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 17:14:16,890 INFO [train.py:968] (0/2) Epoch 13, batch 11700, giga_loss[loss=0.3124, simple_loss=0.3847, pruned_loss=0.12, over 28778.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3798, pruned_loss=0.1293, over 5675703.51 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3536, pruned_loss=0.09807, over 5721780.42 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3813, pruned_loss=0.1315, over 5673533.53 frames. ], batch size: 92, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:14:22,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.573e+03 2.051e+03 2.843e+03 7.916e+03, threshold=4.102e+03, percent-clipped=8.0
+2023-03-06 17:14:44,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2767, 1.5686, 1.3775, 1.4693], device='cuda:0'), covar=tensor([0.0788, 0.0326, 0.0323, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 17:14:47,987 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-06 17:15:03,045 INFO [train.py:968] (0/2) Epoch 13, batch 11750, giga_loss[loss=0.2849, simple_loss=0.3625, pruned_loss=0.1037, over 28151.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3796, pruned_loss=0.1283, over 5672905.08 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3536, pruned_loss=0.09807, over 5716839.19 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3815, pruned_loss=0.1308, over 5674745.75 frames. ], batch size: 77, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:15:23,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9528, 3.7564, 3.5531, 1.6513], device='cuda:0'), covar=tensor([0.0746, 0.0949, 0.1015, 0.2156], device='cuda:0'), in_proj_covar=tensor([0.1110, 0.1033, 0.0896, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-06 17:15:42,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:15:44,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:15:51,595 INFO [train.py:968] (0/2) Epoch 13, batch 11800, giga_loss[loss=0.2845, simple_loss=0.3584, pruned_loss=0.1053, over 28850.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3796, pruned_loss=0.1274, over 5670250.38 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3535, pruned_loss=0.098, over 5715794.12 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3816, pruned_loss=0.1299, over 5672289.87 frames. ], batch size: 186, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:15:59,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.499e+02 1.636e+03 2.086e+03 3.053e+03 9.376e+03, threshold=4.172e+03, percent-clipped=8.0
+2023-03-06 17:16:16,399 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:16:18,261 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:16:22,593 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=559296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:16:42,760 INFO [train.py:968] (0/2) Epoch 13, batch 11850, giga_loss[loss=0.2935, simple_loss=0.3634, pruned_loss=0.1119, over 28937.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3788, pruned_loss=0.1275, over 5667333.46 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3533, pruned_loss=0.09791, over 5716807.15 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3806, pruned_loss=0.1296, over 5667796.32 frames. ], batch size: 227, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:16:49,750 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:17:04,468 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 17:17:18,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6237, 1.5594, 1.2749, 1.2606], device='cuda:0'), covar=tensor([0.0656, 0.0538, 0.0851, 0.1172], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0448, 0.0507, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 17:17:27,206 INFO [train.py:968] (0/2) Epoch 13, batch 11900, giga_loss[loss=0.2708, simple_loss=0.3475, pruned_loss=0.09709, over 28899.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3777, pruned_loss=0.1264, over 5684686.71 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3537, pruned_loss=0.09806, over 5724559.21 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3797, pruned_loss=0.129, over 5676583.86 frames. ], batch size: 227, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:17:32,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.232e+02 1.520e+03 1.845e+03 2.328e+03 5.392e+03, threshold=3.690e+03, percent-clipped=3.0
+2023-03-06 17:18:02,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:04,295 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:04,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:07,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:14,127 INFO [train.py:968] (0/2) Epoch 13, batch 11950, giga_loss[loss=0.3555, simple_loss=0.4047, pruned_loss=0.1531, over 27881.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3783, pruned_loss=0.1272, over 5670400.67 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3537, pruned_loss=0.09803, over 5730057.50 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3806, pruned_loss=0.1303, over 5657167.24 frames. ], batch size: 412, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:18:22,758 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5530, 1.7209, 1.8096, 1.3457], device='cuda:0'), covar=tensor([0.1700, 0.2277, 0.1382, 0.1619], device='cuda:0'), in_proj_covar=tensor([0.0844, 0.0702, 0.0887, 0.0789], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 17:18:31,185 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:34,367 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:40,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:50,596 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4641, 3.1380, 1.5844, 1.6105], device='cuda:0'), covar=tensor([0.0926, 0.0336, 0.0848, 0.1286], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0526, 0.0352, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-06 17:19:06,053 INFO [train.py:968] (0/2) Epoch 13, batch 12000, giga_loss[loss=0.3569, simple_loss=0.3842, pruned_loss=0.1648, over 23300.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3791, pruned_loss=0.1273, over 5670969.81 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3536, pruned_loss=0.09796, over 5731871.48 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3813, pruned_loss=0.13, over 5658478.59 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:19:06,058 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 17:19:14,775 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.218, simple_loss=0.3246, pruned_loss=0.05571, over 944034.00 frames. 
+2023-03-06 17:19:14,775 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 17:19:20,311 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.510e+03 2.025e+03 2.753e+03 9.981e+03, threshold=4.050e+03, percent-clipped=12.0
+2023-03-06 17:19:53,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3530, 1.4928, 1.4388, 1.4852], device='cuda:0'), covar=tensor([0.0722, 0.0325, 0.0290, 0.0784], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 17:20:01,612 INFO [train.py:968] (0/2) Epoch 13, batch 12050, giga_loss[loss=0.3692, simple_loss=0.3898, pruned_loss=0.1743, over 23863.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3793, pruned_loss=0.1286, over 5671688.08 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3535, pruned_loss=0.09784, over 5735059.20 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3816, pruned_loss=0.1314, over 5657818.35 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:20:49,262 INFO [train.py:968] (0/2) Epoch 13, batch 12100, giga_loss[loss=0.3143, simple_loss=0.3785, pruned_loss=0.125, over 28530.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3799, pruned_loss=0.1298, over 5667144.47 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3529, pruned_loss=0.0976, over 5734515.38 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.383, pruned_loss=0.1332, over 5654840.14 frames. ], batch size: 71, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:20:54,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.760e+02 1.554e+03 1.945e+03 2.552e+03 4.989e+03, threshold=3.889e+03, percent-clipped=7.0
+2023-03-06 17:21:04,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:21:09,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:21:36,635 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:21:37,082 INFO [train.py:968] (0/2) Epoch 13, batch 12150, giga_loss[loss=0.3324, simple_loss=0.3841, pruned_loss=0.1404, over 28939.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3803, pruned_loss=0.1295, over 5667370.97 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3535, pruned_loss=0.09779, over 5728671.78 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3831, pruned_loss=0.1329, over 5660185.93 frames. ], batch size: 106, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:21:43,856 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=559622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:21:46,355 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=559624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:21:53,506 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6178, 1.6848, 1.0984, 1.3066], device='cuda:0'), covar=tensor([0.0855, 0.0649, 0.1156, 0.1174], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0451, 0.0511, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 17:22:24,594 INFO [train.py:968] (0/2) Epoch 13, batch 12200, giga_loss[loss=0.3171, simple_loss=0.38, pruned_loss=0.1271, over 29061.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3803, pruned_loss=0.1294, over 5664753.94 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3533, pruned_loss=0.09771, over 5733097.73 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3833, pruned_loss=0.133, over 5653597.20 frames. ], batch size: 155, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:22:30,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5170, 1.0487, 4.4039, 3.2973], device='cuda:0'), covar=tensor([0.1551, 0.2809, 0.0398, 0.1044], device='cuda:0'), in_proj_covar=tensor([0.0689, 0.0608, 0.0885, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 17:22:31,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:22:32,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.772e+02 1.786e+03 2.288e+03 3.579e+03 9.281e+03, threshold=4.576e+03, percent-clipped=21.0
+2023-03-06 17:22:55,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2681, 0.7711, 0.8682, 1.4491], device='cuda:0'), covar=tensor([0.0758, 0.0375, 0.0346, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0061, 0.0054, 0.0092], device='cuda:0')
+2023-03-06 17:23:11,593 INFO [train.py:968] (0/2) Epoch 13, batch 12250, giga_loss[loss=0.3957, simple_loss=0.4111, pruned_loss=0.1901, over 23588.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3791, pruned_loss=0.1288, over 5653212.58 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3529, pruned_loss=0.0974, over 5737811.47 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3826, pruned_loss=0.1329, over 5638128.97 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:24:04,850 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=559763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:24:04,953 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6447, 2.3347, 1.6138, 0.8440], device='cuda:0'), covar=tensor([0.4219, 0.2234, 0.3285, 0.4538], device='cuda:0'), in_proj_covar=tensor([0.1596, 0.1521, 0.1510, 0.1305], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 17:24:05,835 INFO [train.py:968] (0/2) Epoch 13, batch 12300, giga_loss[loss=0.3244, simple_loss=0.3748, pruned_loss=0.137, over 28680.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3799, pruned_loss=0.1294, over 5651887.74 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.353, pruned_loss=0.09759, over 5737944.11 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3828, pruned_loss=0.1327, over 5639179.37 frames. ], batch size: 92, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:24:12,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.785e+02 1.524e+03 2.004e+03 2.775e+03 5.845e+03, threshold=4.009e+03, percent-clipped=4.0
+2023-03-06 17:24:37,989 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-06 17:24:46,584 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:24:46,944 INFO [train.py:968] (0/2) Epoch 13, batch 12350, libri_loss[loss=0.3184, simple_loss=0.3899, pruned_loss=0.1234, over 29294.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3797, pruned_loss=0.1284, over 5661997.83 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3532, pruned_loss=0.09772, over 5744196.91 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3828, pruned_loss=0.1321, over 5642880.34 frames. ], batch size: 94, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:24:48,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:25:16,602 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:25:35,562 INFO [train.py:968] (0/2) Epoch 13, batch 12400, libri_loss[loss=0.2867, simple_loss=0.3683, pruned_loss=0.1025, over 29485.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3783, pruned_loss=0.1274, over 5664518.80 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3532, pruned_loss=0.09766, over 5748156.84 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3814, pruned_loss=0.1312, over 5643811.03 frames. ], batch size: 85, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:25:43,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.542e+02 1.697e+03 2.232e+03 3.267e+03 9.350e+03, threshold=4.464e+03, percent-clipped=14.0
+2023-03-06 17:25:50,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4869, 4.3382, 4.0880, 2.2198], device='cuda:0'), covar=tensor([0.0511, 0.0651, 0.0738, 0.1786], device='cuda:0'), in_proj_covar=tensor([0.1112, 0.1038, 0.0900, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-06 17:26:20,988 INFO [train.py:968] (0/2) Epoch 13, batch 12450, giga_loss[loss=0.314, simple_loss=0.3752, pruned_loss=0.1265, over 28681.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3761, pruned_loss=0.1257, over 5668651.16 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3538, pruned_loss=0.09794, over 5748940.23 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3789, pruned_loss=0.1294, over 5648845.27 frames. ], batch size: 262, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:26:45,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4462, 1.7234, 1.3859, 1.5830], device='cuda:0'), covar=tensor([0.2227, 0.2259, 0.2430, 0.2057], device='cuda:0'), in_proj_covar=tensor([0.1337, 0.0991, 0.1182, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 17:27:08,487 INFO [train.py:968] (0/2) Epoch 13, batch 12500, libri_loss[loss=0.2727, simple_loss=0.357, pruned_loss=0.09414, over 29773.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3718, pruned_loss=0.1226, over 5679098.75 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3534, pruned_loss=0.09765, over 5749583.75 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.375, pruned_loss=0.1267, over 5659726.01 frames. ], batch size: 87, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:27:17,054 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.637e+03 2.159e+03 2.811e+03 7.391e+03, threshold=4.318e+03, percent-clipped=2.0
+2023-03-06 17:27:37,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:27:40,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:27:40,644 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-560000.pt
+2023-03-06 17:27:55,899 INFO [train.py:968] (0/2) Epoch 13, batch 12550, giga_loss[loss=0.2857, simple_loss=0.354, pruned_loss=0.1087, over 28636.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3683, pruned_loss=0.1214, over 5662098.17 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3535, pruned_loss=0.09779, over 5742754.51 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3711, pruned_loss=0.1249, over 5651979.14 frames. ], batch size: 307, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:28:41,309 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-06 17:28:44,000 INFO [train.py:968] (0/2) Epoch 13, batch 12600, giga_loss[loss=0.2737, simple_loss=0.3492, pruned_loss=0.09903, over 28907.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.368, pruned_loss=0.1224, over 5659715.84 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3535, pruned_loss=0.09773, over 5744009.82 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3704, pruned_loss=0.1257, over 5648937.53 frames. ], batch size: 174, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:28:53,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.769e+02 1.622e+03 2.261e+03 3.318e+03 1.329e+04, threshold=4.522e+03, percent-clipped=14.0
+2023-03-06 17:29:33,248 INFO [train.py:968] (0/2) Epoch 13, batch 12650, giga_loss[loss=0.296, simple_loss=0.3529, pruned_loss=0.1196, over 28936.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.367, pruned_loss=0.1222, over 5649117.00 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3531, pruned_loss=0.09746, over 5744313.12 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3698, pruned_loss=0.1259, over 5637556.97 frames. ], batch size: 213, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:29:55,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:29:56,611 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:29:58,024 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:29:59,832 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=560143.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 17:29:59,871 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:30:03,411 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:30:10,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8220, 3.0727, 1.8923, 0.9776], device='cuda:0'), covar=tensor([0.6014, 0.2571, 0.3123, 0.5253], device='cuda:0'), in_proj_covar=tensor([0.1601, 0.1526, 0.1513, 0.1311], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 17:30:21,357 INFO [train.py:968] (0/2) Epoch 13, batch 12700, giga_loss[loss=0.2874, simple_loss=0.3514, pruned_loss=0.1116, over 28481.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3657, pruned_loss=0.1205, over 5649737.92 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.353, pruned_loss=0.09738, over 5738137.61 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3683, pruned_loss=0.124, over 5643836.67 frames. ], batch size: 71, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:30:29,322 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=560172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:30:31,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.346e+02 1.538e+03 2.167e+03 3.224e+03 1.515e+04, threshold=4.333e+03, percent-clipped=13.0
+2023-03-06 17:30:31,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=560174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:31:12,703 INFO [train.py:968] (0/2) Epoch 13, batch 12750, giga_loss[loss=0.3039, simple_loss=0.3744, pruned_loss=0.1167, over 28361.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3644, pruned_loss=0.1176, over 5651638.56 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3527, pruned_loss=0.09723, over 5740497.59 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.367, pruned_loss=0.1209, over 5643373.60 frames. ], batch size: 368, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:32:09,022 INFO [train.py:968] (0/2) Epoch 13, batch 12800, giga_loss[loss=0.27, simple_loss=0.3432, pruned_loss=0.09844, over 28895.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3629, pruned_loss=0.1149, over 5648546.44 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3525, pruned_loss=0.09713, over 5741765.45 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3651, pruned_loss=0.1178, over 5640251.83 frames. ], batch size: 213, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:32:17,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.609e+02 1.562e+03 2.156e+03 3.006e+03 9.530e+03, threshold=4.312e+03, percent-clipped=13.0
+2023-03-06 17:32:25,132 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:32:28,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:32:46,759 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=560299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:32:58,916 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=560313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:33:00,518 INFO [train.py:968] (0/2) Epoch 13, batch 12850, giga_loss[loss=0.2722, simple_loss=0.3487, pruned_loss=0.09786, over 28917.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3597, pruned_loss=0.1116, over 5651887.47 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3521, pruned_loss=0.09695, over 5745035.65 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3621, pruned_loss=0.1143, over 5640612.46 frames. ], batch size: 227, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:33:21,760 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=560334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:33:50,837 INFO [train.py:968] (0/2) Epoch 13, batch 12900, libri_loss[loss=0.2757, simple_loss=0.3515, pruned_loss=0.09995, over 29660.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3567, pruned_loss=0.1093, over 5651224.00 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3521, pruned_loss=0.09734, over 5751253.53 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3589, pruned_loss=0.1118, over 5631628.06 frames. ], batch size: 88, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:33:58,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.861e+02 1.379e+03 1.680e+03 2.204e+03 5.800e+03, threshold=3.361e+03, percent-clipped=2.0
+2023-03-06 17:34:39,166 INFO [train.py:968] (0/2) Epoch 13, batch 12950, giga_loss[loss=0.286, simple_loss=0.3698, pruned_loss=0.1011, over 28581.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3557, pruned_loss=0.106, over 5663184.28 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3518, pruned_loss=0.09721, over 5753541.41 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3577, pruned_loss=0.1082, over 5644229.11 frames. ], batch size: 336, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:35:33,433 INFO [train.py:968] (0/2) Epoch 13, batch 13000, giga_loss[loss=0.2725, simple_loss=0.3552, pruned_loss=0.09491, over 28944.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3557, pruned_loss=0.1058, over 5659876.06 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.351, pruned_loss=0.09695, over 5755412.74 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3581, pruned_loss=0.1079, over 5641548.48 frames. ], batch size: 164, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:35:40,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.864e+02 1.319e+03 1.798e+03 2.717e+03 5.038e+03, threshold=3.597e+03, percent-clipped=9.0
+2023-03-06 17:35:45,470 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=560478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:36:21,120 INFO [train.py:968] (0/2) Epoch 13, batch 13050, giga_loss[loss=0.3383, simple_loss=0.3847, pruned_loss=0.146, over 26786.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3535, pruned_loss=0.1041, over 5662506.74 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3498, pruned_loss=0.09626, over 5759751.47 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3568, pruned_loss=0.1067, over 5640588.00 frames. ], batch size: 555, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:36:24,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560518.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 17:36:36,789 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3934, 1.6474, 1.6408, 1.4569], device='cuda:0'), covar=tensor([0.1396, 0.1708, 0.1541, 0.1616], device='cuda:0'), in_proj_covar=tensor([0.0437, 0.0735, 0.0675, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 17:37:12,203 INFO [train.py:968] (0/2) Epoch 13, batch 13100, giga_loss[loss=0.2959, simple_loss=0.3556, pruned_loss=0.1181, over 27980.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3509, pruned_loss=0.1022, over 5665413.54 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.349, pruned_loss=0.09603, over 5765225.43 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3543, pruned_loss=0.1048, over 5639960.94 frames. ], batch size: 412, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:37:20,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.673e+02 1.271e+03 1.590e+03 2.329e+03 5.789e+03, threshold=3.180e+03, percent-clipped=7.0
+2023-03-06 17:37:29,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9189, 1.1209, 1.1086, 0.8625], device='cuda:0'), covar=tensor([0.1670, 0.1774, 0.0986, 0.1462], device='cuda:0'), in_proj_covar=tensor([0.1731, 0.1639, 0.1592, 0.1692], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 17:37:29,319 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.38 vs. limit=5.0
+2023-03-06 17:37:58,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4475, 1.7184, 1.8078, 1.2972], device='cuda:0'), covar=tensor([0.1745, 0.2409, 0.1391, 0.1689], device='cuda:0'), in_proj_covar=tensor([0.0840, 0.0690, 0.0879, 0.0786], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 17:38:02,043 INFO [train.py:968] (0/2) Epoch 13, batch 13150, giga_loss[loss=0.3221, simple_loss=0.3803, pruned_loss=0.132, over 27618.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3496, pruned_loss=0.102, over 5660421.02 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3488, pruned_loss=0.09597, over 5768578.97 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3526, pruned_loss=0.1042, over 5634853.69 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:38:37,481 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=560651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:38:47,572 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560661.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 17:38:50,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4274, 1.1711, 3.9029, 3.3419], device='cuda:0'), covar=tensor([0.1498, 0.2715, 0.0419, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0684, 0.0603, 0.0877, 0.0789], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 17:38:52,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560664.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 17:38:52,904 INFO [train.py:968] (0/2) Epoch 13, batch 13200, giga_loss[loss=0.3521, simple_loss=0.3903, pruned_loss=0.1569, over 26744.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3496, pruned_loss=0.1017, over 5652435.83 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3486, pruned_loss=0.09592, over 5766534.17 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3521, pruned_loss=0.1036, over 5632676.05 frames. ], batch size: 555, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:39:01,753 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:39:02,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.556e+02 1.488e+03 2.048e+03 2.948e+03 6.696e+03, threshold=4.097e+03, percent-clipped=18.0
+2023-03-06 17:39:22,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=560693.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 17:39:29,351 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4915, 2.1765, 1.7013, 1.8498], device='cuda:0'), covar=tensor([0.0750, 0.0238, 0.0300, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0055, 0.0093], device='cuda:0')
+2023-03-06 17:39:39,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:39:43,472 INFO [train.py:968] (0/2) Epoch 13, batch 13250, giga_loss[loss=0.2466, simple_loss=0.3353, pruned_loss=0.07894, over 28857.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3482, pruned_loss=0.1, over 5660068.88 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3486, pruned_loss=0.0959, over 5767238.50 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3501, pruned_loss=0.1015, over 5643523.32 frames. ], batch size: 112, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:40:38,571 INFO [train.py:968] (0/2) Epoch 13, batch 13300, giga_loss[loss=0.2599, simple_loss=0.3338, pruned_loss=0.09301, over 28988.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3449, pruned_loss=0.09741, over 5652973.10 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3483, pruned_loss=0.09577, over 5767310.28 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3467, pruned_loss=0.09875, over 5638499.45 frames. ], batch size: 128, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:40:47,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.688e+02 1.213e+03 1.558e+03 2.188e+03 4.360e+03, threshold=3.117e+03, percent-clipped=2.0
+2023-03-06 17:41:29,120 INFO [train.py:968] (0/2) Epoch 13, batch 13350, giga_loss[loss=0.2533, simple_loss=0.3321, pruned_loss=0.08727, over 28866.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3412, pruned_loss=0.09523, over 5654407.77 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.348, pruned_loss=0.09578, over 5770941.28 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3427, pruned_loss=0.09633, over 5636273.50 frames. ], batch size: 227, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:41:32,412 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:41:34,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:04,530 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=560849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:09,164 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:09,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:12,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:21,848 INFO [train.py:968] (0/2) Epoch 13, batch 13400, libri_loss[loss=0.2658, simple_loss=0.3421, pruned_loss=0.09476, over 29529.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3397, pruned_loss=0.09471, over 5661095.10 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3474, pruned_loss=0.09551, over 5766418.65 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3412, pruned_loss=0.09581, over 5646938.79 frames. ], batch size: 81, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 17:42:36,493 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.378e+02 1.453e+03 1.817e+03 2.534e+03 6.489e+03, threshold=3.633e+03, percent-clipped=12.0
+2023-03-06 17:42:44,202 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=560884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:47,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5269, 2.1729, 1.5753, 0.8495], device='cuda:0'), covar=tensor([0.4101, 0.2267, 0.2979, 0.4345], device='cuda:0'), in_proj_covar=tensor([0.1596, 0.1512, 0.1504, 0.1305], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 17:43:16,154 INFO [train.py:968] (0/2) Epoch 13, batch 13450, giga_loss[loss=0.288, simple_loss=0.3588, pruned_loss=0.1086, over 27511.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3393, pruned_loss=0.0955, over 5647773.36 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3475, pruned_loss=0.09577, over 5766992.83 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3403, pruned_loss=0.09612, over 5634443.47 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 17:44:14,691 INFO [train.py:968] (0/2) Epoch 13, batch 13500, libri_loss[loss=0.257, simple_loss=0.3172, pruned_loss=0.0984, over 29461.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3411, pruned_loss=0.09669, over 5646173.86 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3472, pruned_loss=0.09591, over 5760956.02 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3421, pruned_loss=0.09707, over 5638619.99 frames. ], batch size: 70, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 17:44:24,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.205e+02 1.521e+03 2.267e+03 3.542e+03 1.060e+04, threshold=4.535e+03, percent-clipped=23.0
+2023-03-06 17:44:48,796 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:44:52,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:45:07,268 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=561012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:45:09,212 INFO [train.py:968] (0/2) Epoch 13, batch 13550, giga_loss[loss=0.2494, simple_loss=0.3362, pruned_loss=0.08134, over 28867.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3432, pruned_loss=0.09639, over 5646740.33 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3468, pruned_loss=0.09564, over 5762216.96 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3442, pruned_loss=0.09695, over 5637361.52 frames. ], batch size: 186, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 17:45:15,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-06 17:45:23,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=561026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:45:25,240 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=561028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:46:02,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7826, 1.4114, 4.8752, 3.5595], device='cuda:0'), covar=tensor([0.1495, 0.2642, 0.0346, 0.0773], device='cuda:0'), in_proj_covar=tensor([0.0681, 0.0600, 0.0868, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 17:46:09,692 INFO [train.py:968] (0/2) Epoch 13, batch 13600, giga_loss[loss=0.278, simple_loss=0.3625, pruned_loss=0.09675, over 28680.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3428, pruned_loss=0.09554, over 5661496.05 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3466, pruned_loss=0.09574, over 5765424.85 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3438, pruned_loss=0.09591, over 5649305.30 frames. ], batch size: 243, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:46:23,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.535e+02 1.354e+03 1.661e+03 2.493e+03 8.856e+03, threshold=3.322e+03, percent-clipped=8.0
+2023-03-06 17:46:36,469 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3403, 1.5502, 1.4253, 1.3296], device='cuda:0'), covar=tensor([0.1895, 0.1586, 0.1149, 0.1412], device='cuda:0'), in_proj_covar=tensor([0.1738, 0.1634, 0.1587, 0.1690], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 17:46:58,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3885, 1.4252, 1.5214, 1.1156], device='cuda:0'), covar=tensor([0.1750, 0.2820, 0.1440, 0.1545], device='cuda:0'), in_proj_covar=tensor([0.0840, 0.0685, 0.0877, 0.0785], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 17:47:10,840 INFO [train.py:968] (0/2) Epoch 13, batch 13650, giga_loss[loss=0.3037, simple_loss=0.3795, pruned_loss=0.114, over 28748.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3419, pruned_loss=0.09502, over 5657798.60 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3458, pruned_loss=0.09537, over 5758725.20 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3433, pruned_loss=0.09564, over 5652504.14 frames. ], batch size: 243, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:48:09,276 INFO [train.py:968] (0/2) Epoch 13, batch 13700, libri_loss[loss=0.2674, simple_loss=0.337, pruned_loss=0.09886, over 29656.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3399, pruned_loss=0.0933, over 5665384.42 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3454, pruned_loss=0.09522, over 5759275.16 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3413, pruned_loss=0.09388, over 5657864.07 frames. ], batch size: 91, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:48:14,653 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=561169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:48:17,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=561172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:48:21,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.261e+02 1.305e+03 1.852e+03 2.338e+03 5.938e+03, threshold=3.704e+03, percent-clipped=7.0
+2023-03-06 17:48:48,065 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=561201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:49:03,910 INFO [train.py:968] (0/2) Epoch 13, batch 13750, giga_loss[loss=0.2663, simple_loss=0.3368, pruned_loss=0.09793, over 27738.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3387, pruned_loss=0.09143, over 5668744.02 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3453, pruned_loss=0.09544, over 5760843.70 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3398, pruned_loss=0.09164, over 5658118.20 frames. ], batch size: 474, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:50:03,906 INFO [train.py:968] (0/2) Epoch 13, batch 13800, giga_loss[loss=0.2398, simple_loss=0.3142, pruned_loss=0.08274, over 28938.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3367, pruned_loss=0.09098, over 5667546.08 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3447, pruned_loss=0.09503, over 5765271.11 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3378, pruned_loss=0.09135, over 5651416.64 frames. ], batch size: 213, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:50:18,225 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.590e+02 1.325e+03 1.671e+03 2.354e+03 7.648e+03, threshold=3.342e+03, percent-clipped=8.0
+2023-03-06 17:50:59,098 INFO [train.py:968] (0/2) Epoch 13, batch 13850, giga_loss[loss=0.2987, simple_loss=0.378, pruned_loss=0.1098, over 29061.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3357, pruned_loss=0.09098, over 5666445.59 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3446, pruned_loss=0.09494, over 5754008.04 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3364, pruned_loss=0.09123, over 5660421.50 frames. ], batch size: 285, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:51:53,113 INFO [train.py:968] (0/2) Epoch 13, batch 13900, libri_loss[loss=0.2137, simple_loss=0.2879, pruned_loss=0.06973, over 29629.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3339, pruned_loss=0.09039, over 5669372.85 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3439, pruned_loss=0.09481, over 5750979.92 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3346, pruned_loss=0.09053, over 5662007.98 frames. ], batch size: 73, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:52:07,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.506e+02 1.344e+03 1.856e+03 2.387e+03 6.562e+03, threshold=3.712e+03, percent-clipped=7.0
+2023-03-06 17:52:12,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3914, 5.1791, 4.9375, 2.3062], device='cuda:0'), covar=tensor([0.0430, 0.0582, 0.0743, 0.1841], device='cuda:0'), in_proj_covar=tensor([0.1064, 0.0991, 0.0857, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 17:52:19,116 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=561387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:52:30,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9604, 4.7262, 1.9933, 1.9581], device='cuda:0'), covar=tensor([0.0797, 0.0234, 0.0801, 0.1193], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0516, 0.0349, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 17:52:51,330 INFO [train.py:968] (0/2) Epoch 13, batch 13950, giga_loss[loss=0.2653, simple_loss=0.3444, pruned_loss=0.09311, over 27635.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.335, pruned_loss=0.09078, over 5658804.47 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3438, pruned_loss=0.0948, over 5752468.51 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3355, pruned_loss=0.0908, over 5649708.99 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:53:14,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4908, 1.4940, 1.1813, 1.2292], device='cuda:0'), covar=tensor([0.0631, 0.0308, 0.0782, 0.0767], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0438, 0.0501, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 17:53:55,358 INFO [train.py:968] (0/2) Epoch 13, batch 14000, giga_loss[loss=0.2481, simple_loss=0.3275, pruned_loss=0.08429, over 28912.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3386, pruned_loss=0.0921, over 5661722.37 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3439, pruned_loss=0.09499, over 5754130.01 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3388, pruned_loss=0.09192, over 5652199.15 frames. ], batch size: 186, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:54:10,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.397e+02 1.380e+03 1.691e+03 2.481e+03 4.800e+03, threshold=3.382e+03, percent-clipped=7.0
+2023-03-06 17:54:23,844 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=561487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:54:58,178 INFO [train.py:968] (0/2) Epoch 13, batch 14050, giga_loss[loss=0.2846, simple_loss=0.3518, pruned_loss=0.1087, over 28211.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3353, pruned_loss=0.09011, over 5660446.73 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3437, pruned_loss=0.09494, over 5746453.68 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3355, pruned_loss=0.08987, over 5656236.81 frames. ], batch size: 412, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:55:15,543 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=561530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:55:17,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8793, 2.8715, 1.8458, 0.8654], device='cuda:0'), covar=tensor([0.6188, 0.2618, 0.3550, 0.5813], device='cuda:0'), in_proj_covar=tensor([0.1590, 0.1512, 0.1504, 0.1300], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 17:55:18,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=561533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:55:53,393 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=561562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:55:55,405 INFO [train.py:968] (0/2) Epoch 13, batch 14100, giga_loss[loss=0.2844, simple_loss=0.3593, pruned_loss=0.1048, over 28983.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3352, pruned_loss=0.09019, over 5678671.22 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3426, pruned_loss=0.09439, over 5753175.06 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.336, pruned_loss=0.09027, over 5665544.73 frames. ], batch size: 285, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:56:06,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2898, 1.5571, 1.3962, 1.4805], device='cuda:0'), covar=tensor([0.0802, 0.0344, 0.0316, 0.0985], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0060, 0.0054, 0.0093], device='cuda:0')
+2023-03-06 17:56:10,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.676e+02 1.432e+03 1.840e+03 3.188e+03 7.494e+03, threshold=3.680e+03, percent-clipped=18.0
+2023-03-06 17:57:01,116 INFO [train.py:968] (0/2) Epoch 13, batch 14150, giga_loss[loss=0.2481, simple_loss=0.3411, pruned_loss=0.07755, over 28607.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3376, pruned_loss=0.09094, over 5679035.96 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3432, pruned_loss=0.09483, over 5752404.56 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3375, pruned_loss=0.09045, over 5667193.20 frames. ], batch size: 242, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:58:00,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3868, 2.0535, 1.3856, 0.5953], device='cuda:0'), covar=tensor([0.4646, 0.2292, 0.3803, 0.4933], device='cuda:0'), in_proj_covar=tensor([0.1596, 0.1522, 0.1508, 0.1301], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 17:58:03,198 INFO [train.py:968] (0/2) Epoch 13, batch 14200, giga_loss[loss=0.2626, simple_loss=0.3588, pruned_loss=0.08322, over 28638.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.341, pruned_loss=0.09034, over 5677004.93 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3429, pruned_loss=0.0947, over 5755773.11 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3412, pruned_loss=0.09003, over 5663160.49 frames. ], batch size: 242, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:58:10,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3234, 1.5546, 1.4956, 1.4603], device='cuda:0'), covar=tensor([0.1532, 0.1929, 0.1998, 0.1782], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0723, 0.0669, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 17:58:12,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7114, 1.8325, 1.4076, 1.4535], device='cuda:0'), covar=tensor([0.0785, 0.0489, 0.0960, 0.0908], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0437, 0.0499, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 17:58:19,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.077e+02 1.419e+03 1.828e+03 2.522e+03 6.892e+03, threshold=3.657e+03, percent-clipped=10.0
+2023-03-06 17:59:04,252 INFO [train.py:968] (0/2) Epoch 13, batch 14250, giga_loss[loss=0.2638, simple_loss=0.3443, pruned_loss=0.0916, over 27555.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3428, pruned_loss=0.08989, over 5683988.20 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3422, pruned_loss=0.09441, over 5757956.21 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3435, pruned_loss=0.08976, over 5668524.47 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:59:12,426 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=561722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:59:51,252 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=561754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:00:05,005 INFO [train.py:968] (0/2) Epoch 13, batch 14300, giga_loss[loss=0.2444, simple_loss=0.334, pruned_loss=0.07737, over 28744.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3417, pruned_loss=0.08838, over 5674531.39 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3421, pruned_loss=0.09435, over 5756186.52 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3424, pruned_loss=0.08825, over 5662866.68 frames. ], batch size: 243, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 18:00:06,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5098, 4.3196, 4.0762, 1.9564], device='cuda:0'), covar=tensor([0.0529, 0.0692, 0.0751, 0.2038], device='cuda:0'), in_proj_covar=tensor([0.1058, 0.0982, 0.0849, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 18:00:19,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.887e+02 1.174e+03 1.441e+03 2.012e+03 4.411e+03, threshold=2.881e+03, percent-clipped=2.0
+2023-03-06 18:01:02,672 INFO [train.py:968] (0/2) Epoch 13, batch 14350, giga_loss[loss=0.2531, simple_loss=0.3364, pruned_loss=0.08493, over 28889.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3416, pruned_loss=0.08878, over 5675439.10 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.342, pruned_loss=0.09437, over 5753499.12 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3421, pruned_loss=0.0884, over 5664709.80 frames. ], batch size: 136, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 18:01:59,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=561862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:02:02,482 INFO [train.py:968] (0/2) Epoch 13, batch 14400, libri_loss[loss=0.2668, simple_loss=0.3423, pruned_loss=0.09567, over 29489.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3399, pruned_loss=0.08873, over 5684135.42 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3416, pruned_loss=0.09409, over 5758946.84 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3406, pruned_loss=0.08846, over 5668054.58 frames. ], batch size: 85, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 18:02:06,225 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-06 18:02:19,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.371e+02 1.258e+03 1.749e+03 2.535e+03 5.180e+03, threshold=3.499e+03, percent-clipped=15.0
+2023-03-06 18:03:10,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2922, 1.1697, 3.7277, 3.2022], device='cuda:0'), covar=tensor([0.1500, 0.2708, 0.0410, 0.1751], device='cuda:0'), in_proj_covar=tensor([0.0674, 0.0597, 0.0861, 0.0774], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 18:03:18,832 INFO [train.py:968] (0/2) Epoch 13, batch 14450, giga_loss[loss=0.2816, simple_loss=0.355, pruned_loss=0.1041, over 28898.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3412, pruned_loss=0.09055, over 5693683.84 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3413, pruned_loss=0.09392, over 5760534.70 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3422, pruned_loss=0.09048, over 5678896.25 frames. ], batch size: 284, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 18:03:55,478 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4509, 1.8265, 1.7487, 1.2846], device='cuda:0'), covar=tensor([0.1882, 0.2437, 0.1532, 0.1815], device='cuda:0'), in_proj_covar=tensor([0.0840, 0.0683, 0.0880, 0.0787], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 18:04:38,915 INFO [train.py:968] (0/2) Epoch 13, batch 14500, libri_loss[loss=0.2575, simple_loss=0.3404, pruned_loss=0.08724, over 29108.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3386, pruned_loss=0.08963, over 5691116.86 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3411, pruned_loss=0.09378, over 5762987.45 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3395, pruned_loss=0.08963, over 5675905.22 frames. ], batch size: 101, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 18:04:58,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.114e+02 1.322e+03 1.740e+03 2.643e+03 7.123e+03, threshold=3.481e+03, percent-clipped=10.0
+2023-03-06 18:05:26,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4188, 2.0392, 1.6244, 1.5435], device='cuda:0'), covar=tensor([0.0733, 0.0253, 0.0293, 0.0818], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0060, 0.0055, 0.0093], device='cuda:0')
+2023-03-06 18:05:29,986 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-562000.pt
+2023-03-06 18:05:37,909 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=562005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:05:41,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:05:51,841 INFO [train.py:968] (0/2) Epoch 13, batch 14550, giga_loss[loss=0.2087, simple_loss=0.2969, pruned_loss=0.06023, over 28462.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3354, pruned_loss=0.0878, over 5684899.33 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3407, pruned_loss=0.09368, over 5765413.09 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3363, pruned_loss=0.08781, over 5669890.81 frames. ], batch size: 71, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 18:06:21,173 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=562037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:06:59,985 INFO [train.py:968] (0/2) Epoch 13, batch 14600, giga_loss[loss=0.2643, simple_loss=0.346, pruned_loss=0.09126, over 29006.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.333, pruned_loss=0.08677, over 5677206.53 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3409, pruned_loss=0.09378, over 5756107.44 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3335, pruned_loss=0.08661, over 5673438.29 frames. ], batch size: 186, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 18:07:21,269 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.677e+02 1.340e+03 1.668e+03 2.427e+03 7.501e+03, threshold=3.336e+03, percent-clipped=7.0
+2023-03-06 18:07:23,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7496, 3.5570, 3.3913, 1.7421], device='cuda:0'), covar=tensor([0.0799, 0.0879, 0.0954, 0.2149], device='cuda:0'), in_proj_covar=tensor([0.1070, 0.0993, 0.0860, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 18:07:41,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=562097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:07:54,584 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=562108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:08:02,378 INFO [train.py:968] (0/2) Epoch 13, batch 14650, giga_loss[loss=0.2932, simple_loss=0.368, pruned_loss=0.1092, over 28948.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3368, pruned_loss=0.08929, over 5678124.80 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3404, pruned_loss=0.09347, over 5760354.28 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3375, pruned_loss=0.08926, over 5669186.45 frames. ], batch size: 199, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 18:08:21,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=562129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:09:03,616 INFO [train.py:968] (0/2) Epoch 13, batch 14700, giga_loss[loss=0.2626, simple_loss=0.3302, pruned_loss=0.09749, over 26822.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3393, pruned_loss=0.09118, over 5685561.23 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.34, pruned_loss=0.09334, over 5763871.30 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3401, pruned_loss=0.09118, over 5673033.45 frames. ], batch size: 555, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 18:09:23,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.451e+02 1.487e+03 1.912e+03 2.580e+03 8.041e+03, threshold=3.824e+03, percent-clipped=16.0
+2023-03-06 18:10:03,945 INFO [train.py:968] (0/2) Epoch 13, batch 14750, giga_loss[loss=0.2425, simple_loss=0.3284, pruned_loss=0.07833, over 28990.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3379, pruned_loss=0.09196, over 5686549.67 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3394, pruned_loss=0.0931, over 5766348.95 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3392, pruned_loss=0.09215, over 5671839.51 frames. ], batch size: 136, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 18:10:41,198 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=562240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:10:45,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:11:10,333 INFO [train.py:968] (0/2) Epoch 13, batch 14800, giga_loss[loss=0.2557, simple_loss=0.3304, pruned_loss=0.09044, over 28614.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3386, pruned_loss=0.09295, over 5687835.94 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3394, pruned_loss=0.09313, over 5768255.54 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3395, pruned_loss=0.09308, over 5673399.11 frames. ], batch size: 85, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:11:19,704 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=562272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:11:19,747 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=562272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:11:22,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:11:29,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.409e+02 1.361e+03 1.788e+03 2.491e+03 7.892e+03, threshold=3.576e+03, percent-clipped=3.0
+2023-03-06 18:11:59,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=562304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:12:11,921 INFO [train.py:968] (0/2) Epoch 13, batch 14850, giga_loss[loss=0.2474, simple_loss=0.3336, pruned_loss=0.08062, over 28075.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3391, pruned_loss=0.09266, over 5678863.72 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.339, pruned_loss=0.09296, over 5757136.82 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3403, pruned_loss=0.09292, over 5675850.92 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:13:17,482 INFO [train.py:968] (0/2) Epoch 13, batch 14900, giga_loss[loss=0.248, simple_loss=0.3423, pruned_loss=0.07685, over 28895.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3404, pruned_loss=0.09231, over 5682273.94 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3384, pruned_loss=0.09258, over 5761318.67 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3418, pruned_loss=0.09284, over 5674205.01 frames. ], batch size: 174, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:13:44,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.689e+02 1.389e+03 1.840e+03 2.692e+03 6.940e+03, threshold=3.681e+03, percent-clipped=8.0
+2023-03-06 18:14:00,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2812, 1.6139, 1.5458, 1.1470], device='cuda:0'), covar=tensor([0.1676, 0.2416, 0.1441, 0.1682], device='cuda:0'), in_proj_covar=tensor([0.0841, 0.0686, 0.0881, 0.0788], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 18:14:45,142 INFO [train.py:968] (0/2) Epoch 13, batch 14950, giga_loss[loss=0.2759, simple_loss=0.3496, pruned_loss=0.1011, over 27651.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3399, pruned_loss=0.09198, over 5667783.72 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3384, pruned_loss=0.09258, over 5761318.67 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3411, pruned_loss=0.09239, over 5661503.57 frames. ], batch size: 472, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:15:21,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4179, 3.9826, 1.5520, 1.6328], device='cuda:0'), covar=tensor([0.0949, 0.0329, 0.0921, 0.1259], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0513, 0.0348, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 18:15:30,651 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=562442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:15:58,694 INFO [train.py:968] (0/2) Epoch 13, batch 15000, giga_loss[loss=0.2835, simple_loss=0.3383, pruned_loss=0.1143, over 26918.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3371, pruned_loss=0.09194, over 5662856.52 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3384, pruned_loss=0.09269, over 5759920.60 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.338, pruned_loss=0.09215, over 5658064.80 frames. ], batch size: 555, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:15:58,699 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 18:16:07,818 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2036, simple_loss=0.3034, pruned_loss=0.0519, over 944034.00 frames. 
+2023-03-06 18:16:07,818 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 18:16:14,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3639, 4.4881, 1.8195, 1.6587], device='cuda:0'), covar=tensor([0.0967, 0.0224, 0.0786, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0514, 0.0348, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 18:16:20,835 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 18:16:25,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.112e+02 1.380e+03 1.820e+03 2.707e+03 1.166e+04, threshold=3.639e+03, percent-clipped=10.0
+2023-03-06 18:16:30,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=562483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:17:11,729 INFO [train.py:968] (0/2) Epoch 13, batch 15050, giga_loss[loss=0.2288, simple_loss=0.3086, pruned_loss=0.07453, over 28855.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3306, pruned_loss=0.08877, over 5666155.13 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3382, pruned_loss=0.09254, over 5762216.82 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3314, pruned_loss=0.08902, over 5658704.01 frames. ], batch size: 164, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:17:15,354 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=562518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:18:12,583 INFO [train.py:968] (0/2) Epoch 13, batch 15100, giga_loss[loss=0.2532, simple_loss=0.3345, pruned_loss=0.08593, over 29003.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3305, pruned_loss=0.08885, over 5668764.73 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3376, pruned_loss=0.09233, over 5765029.06 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3315, pruned_loss=0.08918, over 5658710.23 frames. ], batch size: 136, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:18:32,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.304e+02 1.406e+03 1.776e+03 2.715e+03 5.770e+03, threshold=3.552e+03, percent-clipped=11.0
+2023-03-06 18:19:09,690 INFO [train.py:968] (0/2) Epoch 13, batch 15150, giga_loss[loss=0.2705, simple_loss=0.3388, pruned_loss=0.1011, over 28955.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3325, pruned_loss=0.0904, over 5670034.92 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3378, pruned_loss=0.09242, over 5768111.69 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3329, pruned_loss=0.09052, over 5656309.97 frames. ], batch size: 106, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:19:20,169 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=562626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:19:24,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:20:02,800 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=562658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:20:08,483 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-06 18:20:09,373 INFO [train.py:968] (0/2) Epoch 13, batch 15200, giga_loss[loss=0.2117, simple_loss=0.3033, pruned_loss=0.06004, over 28841.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3303, pruned_loss=0.08837, over 5666642.05 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3379, pruned_loss=0.09241, over 5764926.87 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3305, pruned_loss=0.08844, over 5657105.06 frames. ], batch size: 174, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:20:29,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.905e+02 1.212e+03 1.540e+03 2.300e+03 1.124e+04, threshold=3.081e+03, percent-clipped=6.0
+2023-03-06 18:20:37,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1970, 0.7799, 0.9033, 1.4996], device='cuda:0'), covar=tensor([0.0757, 0.0348, 0.0347, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0116, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-06 18:21:10,779 INFO [train.py:968] (0/2) Epoch 13, batch 15250, giga_loss[loss=0.2458, simple_loss=0.3246, pruned_loss=0.08351, over 28167.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3295, pruned_loss=0.0872, over 5668366.11 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3381, pruned_loss=0.09263, over 5769193.46 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3293, pruned_loss=0.08693, over 5654090.18 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:22:10,307 INFO [train.py:968] (0/2) Epoch 13, batch 15300, libri_loss[loss=0.2996, simple_loss=0.3708, pruned_loss=0.1142, over 25914.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3289, pruned_loss=0.08756, over 5662844.71 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3377, pruned_loss=0.09256, over 5763467.20 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3287, pruned_loss=0.08715, over 5651352.29 frames. ], batch size: 136, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:22:33,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.588e+02 1.448e+03 1.780e+03 2.803e+03 6.512e+03, threshold=3.560e+03, percent-clipped=18.0
+2023-03-06 18:23:15,706 INFO [train.py:968] (0/2) Epoch 13, batch 15350, giga_loss[loss=0.2275, simple_loss=0.3151, pruned_loss=0.06995, over 28727.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3296, pruned_loss=0.08773, over 5655876.47 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3378, pruned_loss=0.09276, over 5756777.46 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3291, pruned_loss=0.08711, over 5651531.24 frames. ], batch size: 119, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 18:23:17,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1469, 1.4605, 1.5032, 1.2883], device='cuda:0'), covar=tensor([0.1475, 0.1625, 0.1891, 0.1674], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0718, 0.0663, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 18:23:19,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=562817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:24:16,169 INFO [train.py:968] (0/2) Epoch 13, batch 15400, giga_loss[loss=0.2452, simple_loss=0.3229, pruned_loss=0.08375, over 27742.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3288, pruned_loss=0.08701, over 5661226.92 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3373, pruned_loss=0.09245, over 5761065.24 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3286, pruned_loss=0.08659, over 5650439.56 frames. ], batch size: 474, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 18:24:42,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.950e+02 1.250e+03 1.605e+03 2.393e+03 5.707e+03, threshold=3.210e+03, percent-clipped=12.0
+2023-03-06 18:24:44,633 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-06 18:24:57,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=562893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:24:58,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0735, 3.9004, 3.6649, 1.7237], device='cuda:0'), covar=tensor([0.0591, 0.0702, 0.0776, 0.2360], device='cuda:0'), in_proj_covar=tensor([0.1060, 0.0983, 0.0851, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 18:25:10,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-06 18:25:25,458 INFO [train.py:968] (0/2) Epoch 13, batch 15450, giga_loss[loss=0.2583, simple_loss=0.3243, pruned_loss=0.09614, over 26767.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3296, pruned_loss=0.0884, over 5659252.00 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3371, pruned_loss=0.09244, over 5762990.07 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3296, pruned_loss=0.08802, over 5647471.72 frames. ], batch size: 555, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 18:25:29,220 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=562917.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 18:26:20,924 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=562960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:26:25,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:26:26,914 INFO [train.py:968] (0/2) Epoch 13, batch 15500, giga_loss[loss=0.2163, simple_loss=0.3091, pruned_loss=0.06178, over 29024.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3289, pruned_loss=0.08761, over 5654388.29 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3371, pruned_loss=0.09239, over 5753285.38 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3286, pruned_loss=0.08724, over 5650971.07 frames. ], batch size: 155, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 18:26:37,299 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=562974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:26:44,995 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.359e+02 1.245e+03 1.677e+03 2.183e+03 6.964e+03, threshold=3.354e+03, percent-clipped=9.0
+2023-03-06 18:26:56,180 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=562992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:27:20,959 INFO [train.py:968] (0/2) Epoch 13, batch 15550, giga_loss[loss=0.2463, simple_loss=0.3429, pruned_loss=0.07488, over 28741.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3304, pruned_loss=0.08682, over 5670837.67 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.337, pruned_loss=0.09252, over 5758426.23 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.33, pruned_loss=0.08623, over 5660495.40 frames. ], batch size: 119, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 18:27:49,982 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:27:54,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:28:25,918 INFO [train.py:968] (0/2) Epoch 13, batch 15600, giga_loss[loss=0.2445, simple_loss=0.3267, pruned_loss=0.0811, over 28082.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3332, pruned_loss=0.08814, over 5659271.21 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.337, pruned_loss=0.09258, over 5754207.32 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3328, pruned_loss=0.08755, over 5653291.96 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:28:29,372 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:28:44,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.582e+02 1.372e+03 2.009e+03 2.797e+03 6.624e+03, threshold=4.019e+03, percent-clipped=15.0
+2023-03-06 18:29:16,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6540, 2.0377, 1.9711, 1.4882], device='cuda:0'), covar=tensor([0.1830, 0.2282, 0.1435, 0.1762], device='cuda:0'), in_proj_covar=tensor([0.0845, 0.0684, 0.0883, 0.0790], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 18:29:25,469 INFO [train.py:968] (0/2) Epoch 13, batch 15650, giga_loss[loss=0.2317, simple_loss=0.3217, pruned_loss=0.07089, over 28980.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3353, pruned_loss=0.08891, over 5663740.46 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3369, pruned_loss=0.0925, over 5755189.01 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.335, pruned_loss=0.08844, over 5656667.59 frames. ], batch size: 100, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:30:12,841 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 18:30:14,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7558, 2.0045, 1.2722, 1.6237], device='cuda:0'), covar=tensor([0.0906, 0.0625, 0.1058, 0.1105], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0435, 0.0501, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 18:30:15,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4585, 1.8755, 1.7275, 1.2903], device='cuda:0'), covar=tensor([0.1567, 0.2083, 0.1277, 0.1540], device='cuda:0'), in_proj_covar=tensor([0.0844, 0.0684, 0.0881, 0.0789], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 18:30:28,853 INFO [train.py:968] (0/2) Epoch 13, batch 15700, giga_loss[loss=0.2956, simple_loss=0.3667, pruned_loss=0.1123, over 28122.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3351, pruned_loss=0.08887, over 5677203.95 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3363, pruned_loss=0.09218, over 5757321.44 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3354, pruned_loss=0.08874, over 5668650.59 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:30:37,944 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3498, 1.6672, 1.2827, 1.3752], device='cuda:0'), covar=tensor([0.2435, 0.2229, 0.2646, 0.2000], device='cuda:0'), in_proj_covar=tensor([0.1333, 0.0975, 0.1181, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 18:30:47,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.104e+02 1.456e+03 1.922e+03 2.885e+03 1.418e+04, threshold=3.844e+03, percent-clipped=13.0
+2023-03-06 18:31:26,129 INFO [train.py:968] (0/2) Epoch 13, batch 15750, giga_loss[loss=0.2045, simple_loss=0.2943, pruned_loss=0.05732, over 28826.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3328, pruned_loss=0.08697, over 5688277.75 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3363, pruned_loss=0.09208, over 5760825.23 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.333, pruned_loss=0.08683, over 5675515.64 frames. ], batch size: 174, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:32:13,444 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:32:30,939 INFO [train.py:968] (0/2) Epoch 13, batch 15800, giga_loss[loss=0.2491, simple_loss=0.3323, pruned_loss=0.08293, over 29043.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3309, pruned_loss=0.08579, over 5687716.69 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3361, pruned_loss=0.09202, over 5762011.54 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3312, pruned_loss=0.08563, over 5675616.23 frames. ], batch size: 155, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:32:51,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.501e+02 1.122e+03 1.601e+03 2.195e+03 3.606e+03, threshold=3.201e+03, percent-clipped=0.0
+2023-03-06 18:33:02,773 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-06 18:33:06,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=563292.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 18:33:31,499 INFO [train.py:968] (0/2) Epoch 13, batch 15850, libri_loss[loss=0.2098, simple_loss=0.2817, pruned_loss=0.0689, over 29462.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3282, pruned_loss=0.08507, over 5688825.95 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3356, pruned_loss=0.09182, over 5764820.16 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3288, pruned_loss=0.08501, over 5675062.35 frames. ], batch size: 70, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:33:47,337 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:34:00,852 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-06 18:34:14,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=563349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:34:24,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:34:32,468 INFO [train.py:968] (0/2) Epoch 13, batch 15900, giga_loss[loss=0.2531, simple_loss=0.3368, pruned_loss=0.08466, over 28685.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3299, pruned_loss=0.08601, over 5684113.41 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3355, pruned_loss=0.09183, over 5767811.98 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3303, pruned_loss=0.08584, over 5668907.15 frames. ], batch size: 307, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:34:41,583 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-06 18:34:55,760 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.031e+02 1.448e+03 1.915e+03 2.546e+03 8.038e+03, threshold=3.830e+03, percent-clipped=10.0
+2023-03-06 18:35:28,338 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.88 vs. limit=2.0
+2023-03-06 18:35:38,619 INFO [train.py:968] (0/2) Epoch 13, batch 15950, giga_loss[loss=0.2452, simple_loss=0.3295, pruned_loss=0.08045, over 28592.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3318, pruned_loss=0.08696, over 5685187.35 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3353, pruned_loss=0.09176, over 5769578.53 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3323, pruned_loss=0.08681, over 5670565.29 frames. ], batch size: 242, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:36:05,692 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563435.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 18:36:09,862 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563438.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 18:36:44,922 INFO [train.py:968] (0/2) Epoch 13, batch 16000, giga_loss[loss=0.2447, simple_loss=0.3321, pruned_loss=0.07865, over 28999.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3339, pruned_loss=0.08867, over 5678775.39 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3357, pruned_loss=0.09191, over 5767932.10 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3339, pruned_loss=0.08834, over 5667014.28 frames. ], batch size: 165, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:36:49,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563467.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 18:36:54,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4067, 1.6847, 1.6678, 1.2522], device='cuda:0'), covar=tensor([0.1659, 0.2385, 0.1398, 0.1664], device='cuda:0'), in_proj_covar=tensor([0.0840, 0.0682, 0.0879, 0.0788], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 18:37:05,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.485e+02 1.374e+03 1.688e+03 2.569e+03 8.721e+03, threshold=3.376e+03, percent-clipped=5.0
+2023-03-06 18:37:11,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5508, 1.7149, 1.4260, 1.8030], device='cuda:0'), covar=tensor([0.2354, 0.2275, 0.2446, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.1333, 0.0975, 0.1180, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 18:37:17,699 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:37:21,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:37:41,870 INFO [train.py:968] (0/2) Epoch 13, batch 16050, giga_loss[loss=0.2761, simple_loss=0.3593, pruned_loss=0.09644, over 28898.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3375, pruned_loss=0.09035, over 5675870.93 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3353, pruned_loss=0.09168, over 5761390.25 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3378, pruned_loss=0.09025, over 5670458.34 frames. ], batch size: 164, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:37:52,388 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:38:41,236 INFO [train.py:968] (0/2) Epoch 13, batch 16100, giga_loss[loss=0.2429, simple_loss=0.3361, pruned_loss=0.07485, over 28529.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3387, pruned_loss=0.08989, over 5680301.60 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3348, pruned_loss=0.09145, over 5762246.66 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3394, pruned_loss=0.08998, over 5674262.71 frames. ], batch size: 370, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:39:02,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.743e+02 1.274e+03 1.651e+03 2.088e+03 5.094e+03, threshold=3.302e+03, percent-clipped=4.0
+2023-03-06 18:39:43,609 INFO [train.py:968] (0/2) Epoch 13, batch 16150, giga_loss[loss=0.2265, simple_loss=0.3201, pruned_loss=0.06641, over 29062.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3397, pruned_loss=0.09089, over 5689601.68 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3346, pruned_loss=0.09124, over 5766247.06 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3406, pruned_loss=0.09114, over 5678663.76 frames. ], batch size: 128, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:40:02,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=563625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:40:10,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9271, 2.5287, 1.9747, 1.7301], device='cuda:0'), covar=tensor([0.2549, 0.1466, 0.1826, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.1749, 0.1624, 0.1576, 0.1682], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 18:40:53,858 INFO [train.py:968] (0/2) Epoch 13, batch 16200, giga_loss[loss=0.2481, simple_loss=0.324, pruned_loss=0.08613, over 28714.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3371, pruned_loss=0.0896, over 5691158.67 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3342, pruned_loss=0.09105, over 5765577.35 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3382, pruned_loss=0.08997, over 5681782.35 frames. ], batch size: 307, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:41:12,503 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 18:41:17,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.304e+02 1.389e+03 1.839e+03 2.832e+03 6.457e+03, threshold=3.679e+03, percent-clipped=15.0
+2023-03-06 18:41:44,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=563701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:41:51,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0920, 1.0679, 3.5556, 3.0144], device='cuda:0'), covar=tensor([0.1708, 0.2774, 0.0502, 0.1000], device='cuda:0'), in_proj_covar=tensor([0.0676, 0.0597, 0.0862, 0.0777], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 18:42:03,448 INFO [train.py:968] (0/2) Epoch 13, batch 16250, giga_loss[loss=0.2628, simple_loss=0.3437, pruned_loss=0.09093, over 28109.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3377, pruned_loss=0.09032, over 5685020.81 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3345, pruned_loss=0.09124, over 5764166.98 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3384, pruned_loss=0.09043, over 5678260.60 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:42:26,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=563732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:43:05,908 INFO [train.py:968] (0/2) Epoch 13, batch 16300, giga_loss[loss=0.2362, simple_loss=0.3129, pruned_loss=0.07973, over 29096.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3365, pruned_loss=0.09049, over 5671241.30 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3347, pruned_loss=0.09125, over 5763278.11 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3369, pruned_loss=0.09055, over 5665515.28 frames. ], batch size: 128, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:43:09,575 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:43:13,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:43:29,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.108e+02 1.364e+03 1.775e+03 2.236e+03 6.677e+03, threshold=3.551e+03, percent-clipped=5.0
+2023-03-06 18:43:51,528 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:44:05,828 INFO [train.py:968] (0/2) Epoch 13, batch 16350, giga_loss[loss=0.2741, simple_loss=0.3457, pruned_loss=0.1012, over 27623.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3339, pruned_loss=0.09007, over 5677649.10 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3343, pruned_loss=0.09092, over 5767043.29 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3347, pruned_loss=0.09041, over 5667552.70 frames. ], batch size: 472, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:44:42,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:44:43,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8554, 2.2735, 2.2068, 1.6618], device='cuda:0'), covar=tensor([0.1819, 0.2111, 0.1434, 0.1607], device='cuda:0'), in_proj_covar=tensor([0.0840, 0.0681, 0.0879, 0.0787], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 18:44:46,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:45:08,584 INFO [train.py:968] (0/2) Epoch 13, batch 16400, giga_loss[loss=0.249, simple_loss=0.3363, pruned_loss=0.08085, over 28322.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3341, pruned_loss=0.08925, over 5677277.25 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3344, pruned_loss=0.09106, over 5759289.55 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3345, pruned_loss=0.08939, over 5674498.70 frames. ], batch size: 368, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:45:22,486 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:45:22,958 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:45:27,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:45:31,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.704e+02 1.245e+03 1.734e+03 2.681e+03 7.770e+03, threshold=3.469e+03, percent-clipped=12.0
+2023-03-06 18:45:44,701 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:45:55,830 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:46:03,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 18:46:04,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:46:05,130 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-06 18:46:12,038 INFO [train.py:968] (0/2) Epoch 13, batch 16450, giga_loss[loss=0.2527, simple_loss=0.3313, pruned_loss=0.08707, over 29103.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3328, pruned_loss=0.0878, over 5668932.45 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3347, pruned_loss=0.09116, over 5759203.40 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3329, pruned_loss=0.0878, over 5665717.31 frames. ], batch size: 128, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:46:35,994 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:46:40,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1477, 1.1437, 3.7250, 3.1163], device='cuda:0'), covar=tensor([0.1694, 0.2718, 0.0375, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0678, 0.0598, 0.0861, 0.0779], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 18:47:09,574 INFO [train.py:968] (0/2) Epoch 13, batch 16500, giga_loss[loss=0.2523, simple_loss=0.3475, pruned_loss=0.07857, over 28874.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3346, pruned_loss=0.08703, over 5675844.16 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3342, pruned_loss=0.09097, over 5758982.07 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3351, pruned_loss=0.08713, over 5671977.70 frames. ], batch size: 227, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:47:29,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.754e+02 1.267e+03 1.714e+03 2.297e+03 5.723e+03, threshold=3.429e+03, percent-clipped=4.0
+2023-03-06 18:47:39,891 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:47:51,441 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-564000.pt
+2023-03-06 18:48:12,221 INFO [train.py:968] (0/2) Epoch 13, batch 16550, giga_loss[loss=0.2258, simple_loss=0.3144, pruned_loss=0.06862, over 29050.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3358, pruned_loss=0.08669, over 5666494.97 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3343, pruned_loss=0.09108, over 5752724.10 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3361, pruned_loss=0.08659, over 5667635.85 frames. ], batch size: 128, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:49:13,811 INFO [train.py:968] (0/2) Epoch 13, batch 16600, giga_loss[loss=0.2152, simple_loss=0.286, pruned_loss=0.0722, over 24429.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3349, pruned_loss=0.08576, over 5670430.37 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3341, pruned_loss=0.09101, over 5747794.84 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3353, pruned_loss=0.08568, over 5674691.59 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:49:34,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.689e+02 1.271e+03 1.639e+03 2.434e+03 5.156e+03, threshold=3.279e+03, percent-clipped=7.0
+2023-03-06 18:50:18,645 INFO [train.py:968] (0/2) Epoch 13, batch 16650, giga_loss[loss=0.2365, simple_loss=0.32, pruned_loss=0.07657, over 29027.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3344, pruned_loss=0.08558, over 5665822.11 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3337, pruned_loss=0.09075, over 5747555.77 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.335, pruned_loss=0.08562, over 5667469.59 frames. ], batch size: 128, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:51:16,837 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 18:51:26,447 INFO [train.py:968] (0/2) Epoch 13, batch 16700, giga_loss[loss=0.2364, simple_loss=0.3216, pruned_loss=0.07558, over 28139.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3348, pruned_loss=0.08576, over 5671586.53 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3339, pruned_loss=0.09094, over 5749995.46 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3351, pruned_loss=0.08548, over 5669016.88 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:51:50,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.621e+02 1.369e+03 1.774e+03 2.450e+03 4.931e+03, threshold=3.547e+03, percent-clipped=10.0
+2023-03-06 18:52:29,050 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:52:34,441 INFO [train.py:968] (0/2) Epoch 13, batch 16750, giga_loss[loss=0.2834, simple_loss=0.3589, pruned_loss=0.104, over 28975.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3355, pruned_loss=0.08573, over 5674911.89 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3338, pruned_loss=0.09094, over 5751620.00 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3359, pruned_loss=0.08536, over 5669631.16 frames. ], batch size: 213, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:53:45,191 INFO [train.py:968] (0/2) Epoch 13, batch 16800, giga_loss[loss=0.2406, simple_loss=0.3376, pruned_loss=0.0718, over 28999.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3379, pruned_loss=0.08742, over 5682168.62 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.334, pruned_loss=0.09118, over 5755613.96 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3381, pruned_loss=0.08682, over 5672669.86 frames. ], batch size: 136, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:53:47,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:54:00,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:54:04,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 18:54:07,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.217e+02 1.417e+03 2.017e+03 2.821e+03 7.521e+03, threshold=4.033e+03, percent-clipped=12.0
+2023-03-06 18:54:24,968 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-06 18:54:33,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8203, 2.6973, 1.7526, 1.1268], device='cuda:0'), covar=tensor([0.5277, 0.2698, 0.3058, 0.4403], device='cuda:0'), in_proj_covar=tensor([0.1593, 0.1516, 0.1507, 0.1311], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 18:54:47,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:54:53,253 INFO [train.py:968] (0/2) Epoch 13, batch 16850, giga_loss[loss=0.2517, simple_loss=0.3448, pruned_loss=0.07928, over 28881.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3402, pruned_loss=0.08825, over 5678390.15 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3333, pruned_loss=0.09093, over 5748275.57 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3411, pruned_loss=0.08792, over 5675272.26 frames. ], batch size: 119, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:55:06,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8592, 4.6511, 4.4030, 2.1740], device='cuda:0'), covar=tensor([0.0446, 0.0578, 0.0693, 0.2045], device='cuda:0'), in_proj_covar=tensor([0.1054, 0.0980, 0.0852, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 18:55:24,588 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:55:54,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:55:54,838 INFO [train.py:968] (0/2) Epoch 13, batch 16900, giga_loss[loss=0.2513, simple_loss=0.3341, pruned_loss=0.08427, over 28750.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3403, pruned_loss=0.08871, over 5681318.28 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3335, pruned_loss=0.0908, over 5748765.90 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.341, pruned_loss=0.08846, over 5676124.14 frames. ], batch size: 262, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:56:07,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:56:23,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.055e+02 1.280e+03 1.594e+03 2.384e+03 9.027e+03, threshold=3.188e+03, percent-clipped=6.0
+2023-03-06 18:56:58,189 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:57:02,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:57:05,362 INFO [train.py:968] (0/2) Epoch 13, batch 16950, giga_loss[loss=0.2431, simple_loss=0.3208, pruned_loss=0.08265, over 29024.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3376, pruned_loss=0.08799, over 5693513.98 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.333, pruned_loss=0.0906, over 5752063.97 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3388, pruned_loss=0.08789, over 5684803.41 frames. ], batch size: 155, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:57:13,338 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:57:18,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:57:46,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:57:47,216 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 18:58:00,188 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:58:02,275 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:58:06,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:58:24,591 INFO [train.py:968] (0/2) Epoch 13, batch 17000, giga_loss[loss=0.2498, simple_loss=0.3358, pruned_loss=0.08187, over 28419.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3356, pruned_loss=0.08618, over 5692329.89 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3331, pruned_loss=0.09075, over 5750335.79 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3365, pruned_loss=0.08593, over 5686060.16 frames. ], batch size: 368, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:58:47,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.385e+02 1.261e+03 1.655e+03 2.212e+03 6.415e+03, threshold=3.310e+03, percent-clipped=10.0
+2023-03-06 18:58:48,312 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:58:53,459 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 18:59:09,791 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:59:16,080 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:59:19,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:59:27,305 INFO [train.py:968] (0/2) Epoch 13, batch 17050, giga_loss[loss=0.2458, simple_loss=0.3148, pruned_loss=0.08846, over 24220.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3339, pruned_loss=0.08507, over 5694615.48 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3333, pruned_loss=0.09084, over 5750509.43 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3345, pruned_loss=0.08462, over 5688017.32 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:59:44,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4051, 3.7040, 1.5417, 1.5980], device='cuda:0'), covar=tensor([0.0953, 0.0287, 0.0880, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0514, 0.0349, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 18:59:55,929 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:00:31,968 INFO [train.py:968] (0/2) Epoch 13, batch 17100, giga_loss[loss=0.2725, simple_loss=0.3551, pruned_loss=0.09493, over 28676.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.335, pruned_loss=0.08589, over 5689988.27 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3331, pruned_loss=0.09081, over 5753167.81 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3357, pruned_loss=0.08548, over 5681921.65 frames. ], batch size: 307, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:00:55,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.215e+02 1.331e+03 1.732e+03 2.374e+03 4.757e+03, threshold=3.464e+03, percent-clipped=8.0
+2023-03-06 19:00:57,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:01:06,773 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564592.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 19:01:36,465 INFO [train.py:968] (0/2) Epoch 13, batch 17150, giga_loss[loss=0.2338, simple_loss=0.3023, pruned_loss=0.0826, over 24489.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3376, pruned_loss=0.08756, over 5682684.26 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3326, pruned_loss=0.09056, over 5753729.34 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3385, pruned_loss=0.08741, over 5675378.90 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:02:34,398 INFO [train.py:968] (0/2) Epoch 13, batch 17200, giga_loss[loss=0.2234, simple_loss=0.3058, pruned_loss=0.07044, over 28667.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3365, pruned_loss=0.08769, over 5682080.54 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3323, pruned_loss=0.09042, over 5754133.29 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3376, pruned_loss=0.08764, over 5674162.43 frames. ], batch size: 262, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:03:00,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.769e+02 1.423e+03 2.019e+03 3.014e+03 1.165e+04, threshold=4.037e+03, percent-clipped=18.0
+2023-03-06 19:03:09,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5977, 1.9224, 1.7302, 1.6897], device='cuda:0'), covar=tensor([0.1488, 0.1849, 0.1969, 0.1856], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0717, 0.0662, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 19:03:34,966 INFO [train.py:968] (0/2) Epoch 13, batch 17250, giga_loss[loss=0.2565, simple_loss=0.3395, pruned_loss=0.08678, over 28903.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3356, pruned_loss=0.08841, over 5670229.09 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3327, pruned_loss=0.0907, over 5745264.50 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3362, pruned_loss=0.08804, over 5670731.95 frames. ], batch size: 186, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:03:36,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:03:53,350 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:03:56,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:04:18,035 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:04:34,222 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:04:37,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5951, 1.8236, 1.5637, 1.6127], device='cuda:0'), covar=tensor([0.2175, 0.1879, 0.1955, 0.1704], device='cuda:0'), in_proj_covar=tensor([0.1327, 0.0972, 0.1176, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 19:04:39,526 INFO [train.py:968] (0/2) Epoch 13, batch 17300, giga_loss[loss=0.2621, simple_loss=0.3231, pruned_loss=0.1005, over 24455.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3363, pruned_loss=0.08909, over 5681690.19 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3326, pruned_loss=0.09066, over 5749357.54 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3369, pruned_loss=0.08877, over 5676840.87 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:04:40,008 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-06 19:05:00,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.453e+02 1.483e+03 1.787e+03 2.472e+03 6.476e+03, threshold=3.575e+03, percent-clipped=7.0
+2023-03-06 19:05:29,737 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=5.22 vs. limit=5.0
+2023-03-06 19:05:39,441 INFO [train.py:968] (0/2) Epoch 13, batch 17350, giga_loss[loss=0.2835, simple_loss=0.3703, pruned_loss=0.09837, over 28770.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3451, pruned_loss=0.09475, over 5678577.70 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3324, pruned_loss=0.09055, over 5750456.21 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3458, pruned_loss=0.09459, over 5673316.96 frames. ], batch size: 243, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:06:13,681 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7725, 1.7488, 1.1876, 1.3235], device='cuda:0'), covar=tensor([0.0795, 0.0634, 0.1086, 0.1080], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0436, 0.0504, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 19:06:23,314 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:06:25,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:06:26,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3823, 3.2688, 1.5617, 1.4121], device='cuda:0'), covar=tensor([0.0963, 0.0284, 0.0892, 0.1364], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0510, 0.0348, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 19:06:29,004 INFO [train.py:968] (0/2) Epoch 13, batch 17400, giga_loss[loss=0.3082, simple_loss=0.3848, pruned_loss=0.1158, over 29055.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3543, pruned_loss=0.09973, over 5686202.82 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3325, pruned_loss=0.09059, over 5750962.93 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3548, pruned_loss=0.09962, over 5681272.57 frames. ], batch size: 136, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:06:40,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 19:06:40,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:06:46,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.096e+02 1.328e+03 1.785e+03 2.581e+03 7.709e+03, threshold=3.569e+03, percent-clipped=11.0
+2023-03-06 19:06:51,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:06:53,010 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:06:57,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:07:14,663 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-06 19:07:18,407 INFO [train.py:968] (0/2) Epoch 13, batch 17450, giga_loss[loss=0.2739, simple_loss=0.3501, pruned_loss=0.09891, over 28707.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3548, pruned_loss=0.1007, over 5693752.43 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3322, pruned_loss=0.09044, over 5755152.35 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.356, pruned_loss=0.1009, over 5684890.17 frames. ], batch size: 99, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:07:26,487 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:08:01,136 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:08:08,109 INFO [train.py:968] (0/2) Epoch 13, batch 17500, giga_loss[loss=0.26, simple_loss=0.3303, pruned_loss=0.09488, over 28631.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3485, pruned_loss=0.0984, over 5690386.42 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3327, pruned_loss=0.09063, over 5755424.47 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3492, pruned_loss=0.09851, over 5682598.78 frames. ], batch size: 307, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:08:10,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564967.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 19:08:28,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.141e+02 1.130e+03 1.479e+03 2.244e+03 7.732e+03, threshold=2.958e+03, percent-clipped=6.0
+2023-03-06 19:08:30,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3475, 1.2408, 4.1310, 3.2934], device='cuda:0'), covar=tensor([0.1654, 0.2726, 0.0391, 0.1111], device='cuda:0'), in_proj_covar=tensor([0.0676, 0.0600, 0.0864, 0.0781], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 19:08:42,478 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:08:52,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9190, 1.8644, 1.4440, 1.4060], device='cuda:0'), covar=tensor([0.0855, 0.0760, 0.1022, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0437, 0.0506, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 19:08:55,398 INFO [train.py:968] (0/2) Epoch 13, batch 17550, giga_loss[loss=0.2108, simple_loss=0.2876, pruned_loss=0.06699, over 28762.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3409, pruned_loss=0.09515, over 5685322.03 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3324, pruned_loss=0.09032, over 5755681.30 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.342, pruned_loss=0.09566, over 5677260.35 frames. ], batch size: 119, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:09:01,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:09:02,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4580, 1.7549, 1.7361, 1.2955], device='cuda:0'), covar=tensor([0.1719, 0.2399, 0.1437, 0.1592], device='cuda:0'), in_proj_covar=tensor([0.0844, 0.0683, 0.0882, 0.0790], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 19:09:03,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:09:13,529 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:09:28,943 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:09:42,268 INFO [train.py:968] (0/2) Epoch 13, batch 17600, giga_loss[loss=0.2182, simple_loss=0.2911, pruned_loss=0.07266, over 28848.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3336, pruned_loss=0.09222, over 5684467.11 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3326, pruned_loss=0.09029, over 5757744.47 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3344, pruned_loss=0.09268, over 5675363.46 frames. ], batch size: 199, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:10:01,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.434e+02 1.097e+03 1.413e+03 1.950e+03 5.648e+03, threshold=2.826e+03, percent-clipped=6.0
+2023-03-06 19:10:21,641 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 19:10:24,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565113.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 19:10:25,137 INFO [train.py:968] (0/2) Epoch 13, batch 17650, giga_loss[loss=0.2056, simple_loss=0.2823, pruned_loss=0.06441, over 29040.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3263, pruned_loss=0.0889, over 5692856.44 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.333, pruned_loss=0.09046, over 5759251.37 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3264, pruned_loss=0.08911, over 5683297.42 frames. ], batch size: 136, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:10:50,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565142.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 19:10:56,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:11:08,173 INFO [train.py:968] (0/2) Epoch 13, batch 17700, giga_loss[loss=0.2222, simple_loss=0.2962, pruned_loss=0.07405, over 28766.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.32, pruned_loss=0.08581, over 5698742.23 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3333, pruned_loss=0.09055, over 5761598.54 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3197, pruned_loss=0.08585, over 5688226.08 frames. ], batch size: 199, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:11:26,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.175e+02 1.041e+03 1.315e+03 1.825e+03 6.121e+03, threshold=2.630e+03, percent-clipped=4.0
+2023-03-06 19:11:49,568 INFO [train.py:968] (0/2) Epoch 13, batch 17750, giga_loss[loss=0.2021, simple_loss=0.2738, pruned_loss=0.06516, over 28470.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3185, pruned_loss=0.08536, over 5700822.01 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3346, pruned_loss=0.09129, over 5763645.86 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3167, pruned_loss=0.08461, over 5689437.46 frames. ], batch size: 71, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:12:23,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4314, 1.7330, 1.6952, 1.2978], device='cuda:0'), covar=tensor([0.1602, 0.2192, 0.1334, 0.1519], device='cuda:0'), in_proj_covar=tensor([0.0850, 0.0689, 0.0889, 0.0796], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 19:12:33,167 INFO [train.py:968] (0/2) Epoch 13, batch 17800, giga_loss[loss=0.2167, simple_loss=0.2994, pruned_loss=0.06705, over 28854.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3148, pruned_loss=0.0836, over 5700522.20 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3348, pruned_loss=0.09123, over 5765001.74 frames. ], giga_tot_loss[loss=0.2392, simple_loss=0.3127, pruned_loss=0.08284, over 5687952.03 frames. ], batch size: 174, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:12:49,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.730e+02 9.965e+02 1.296e+03 1.738e+03 4.964e+03, threshold=2.592e+03, percent-clipped=6.0
+2023-03-06 19:12:50,671 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:13:14,536 INFO [train.py:968] (0/2) Epoch 13, batch 17850, giga_loss[loss=0.2131, simple_loss=0.2923, pruned_loss=0.06694, over 28932.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3123, pruned_loss=0.08226, over 5710960.45 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3351, pruned_loss=0.09116, over 5770485.98 frames. ], giga_tot_loss[loss=0.2361, simple_loss=0.3095, pruned_loss=0.0814, over 5693491.18 frames. ], batch size: 213, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:13:27,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=565331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:13:34,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6183, 1.7698, 1.4926, 1.7800], device='cuda:0'), covar=tensor([0.2301, 0.2393, 0.2446, 0.2592], device='cuda:0'), in_proj_covar=tensor([0.1330, 0.0978, 0.1177, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 19:13:55,769 INFO [train.py:968] (0/2) Epoch 13, batch 17900, giga_loss[loss=0.2178, simple_loss=0.2876, pruned_loss=0.074, over 28428.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3082, pruned_loss=0.08013, over 5706116.38 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3349, pruned_loss=0.09094, over 5772743.89 frames. ], giga_tot_loss[loss=0.2322, simple_loss=0.3056, pruned_loss=0.07941, over 5688872.74 frames. ], batch size: 71, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:14:08,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=565377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:14:17,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.535e+02 9.896e+02 1.258e+03 1.814e+03 7.272e+03, threshold=2.516e+03, percent-clipped=9.0
+2023-03-06 19:14:36,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=565408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:14:40,737 INFO [train.py:968] (0/2) Epoch 13, batch 17950, giga_loss[loss=0.2366, simple_loss=0.3066, pruned_loss=0.08331, over 28536.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3059, pruned_loss=0.07912, over 5692328.99 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3356, pruned_loss=0.09128, over 5761436.98 frames. ], giga_tot_loss[loss=0.2295, simple_loss=0.3028, pruned_loss=0.07806, over 5688547.38 frames. ], batch size: 336, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:15:25,149 INFO [train.py:968] (0/2) Epoch 13, batch 18000, giga_loss[loss=0.1951, simple_loss=0.2754, pruned_loss=0.05744, over 28664.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.303, pruned_loss=0.07819, over 5689582.68 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3358, pruned_loss=0.09128, over 5761734.21 frames. ], giga_tot_loss[loss=0.2272, simple_loss=0.3, pruned_loss=0.07719, over 5685428.24 frames. ], batch size: 119, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:15:25,154 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 19:15:33,540 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2143, simple_loss=0.32, pruned_loss=0.05427, over 944034.00 frames. 
+2023-03-06 19:15:33,541 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 19:15:40,970 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:15:43,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:15:51,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.129e+02 9.387e+02 1.309e+03 1.851e+03 4.208e+03, threshold=2.618e+03, percent-clipped=6.0
+2023-03-06 19:16:06,636 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:11,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3879, 1.5389, 1.4090, 1.3519], device='cuda:0'), covar=tensor([0.2309, 0.2055, 0.1408, 0.1855], device='cuda:0'), in_proj_covar=tensor([0.1779, 0.1647, 0.1601, 0.1709], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 19:16:14,370 INFO [train.py:968] (0/2) Epoch 13, batch 18050, giga_loss[loss=0.2651, simple_loss=0.3297, pruned_loss=0.1002, over 28802.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3005, pruned_loss=0.07688, over 5692679.10 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3357, pruned_loss=0.09126, over 5765049.58 frames. ], giga_tot_loss[loss=0.2242, simple_loss=0.297, pruned_loss=0.07564, over 5684016.96 frames. ], batch size: 262, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:16:19,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:21,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:25,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=565527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:47,408 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:47,905 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:49,079 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:58,371 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 19:17:00,017 INFO [train.py:968] (0/2) Epoch 13, batch 18100, giga_loss[loss=0.2441, simple_loss=0.3079, pruned_loss=0.09012, over 28562.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2977, pruned_loss=0.07595, over 5680643.75 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.336, pruned_loss=0.09142, over 5756333.99 frames. ], giga_tot_loss[loss=0.2215, simple_loss=0.2939, pruned_loss=0.07453, over 5679581.36 frames. ], batch size: 336, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:17:10,360 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-06 19:17:13,203 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:17:15,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8721, 2.0814, 1.4289, 1.5897], device='cuda:0'), covar=tensor([0.0788, 0.0512, 0.0990, 0.0965], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0435, 0.0503, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 19:17:16,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 4.686e+02 9.463e+02 1.226e+03 1.939e+03 3.728e+03, threshold=2.451e+03, percent-clipped=10.0
+2023-03-06 19:17:31,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8017, 1.9916, 1.5172, 1.5276], device='cuda:0'), covar=tensor([0.0795, 0.0568, 0.0931, 0.1043], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0434, 0.0503, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 19:17:46,161 INFO [train.py:968] (0/2) Epoch 13, batch 18150, giga_loss[loss=0.3115, simple_loss=0.3835, pruned_loss=0.1197, over 28870.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3028, pruned_loss=0.07927, over 5676393.66 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.336, pruned_loss=0.09131, over 5761682.73 frames. ], giga_tot_loss[loss=0.227, simple_loss=0.2986, pruned_loss=0.07775, over 5668860.58 frames. ], batch size: 199, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:18:13,179 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3991, 1.8918, 1.4032, 0.6876], device='cuda:0'), covar=tensor([0.4229, 0.2250, 0.2662, 0.4808], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1510, 0.1487, 0.1296], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 19:18:31,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=565662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:18:33,568 INFO [train.py:968] (0/2) Epoch 13, batch 18200, giga_loss[loss=0.3311, simple_loss=0.3965, pruned_loss=0.1329, over 28716.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3151, pruned_loss=0.08569, over 5682627.46 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3361, pruned_loss=0.09141, over 5765827.56 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3109, pruned_loss=0.0842, over 5670391.92 frames. ], batch size: 242, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:18:37,895 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:18:39,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:18:52,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.428e+02 1.283e+03 1.655e+03 2.166e+03 1.044e+04, threshold=3.310e+03, percent-clipped=18.0
+2023-03-06 19:19:03,494 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:19:14,535 INFO [train.py:968] (0/2) Epoch 13, batch 18250, giga_loss[loss=0.3063, simple_loss=0.3774, pruned_loss=0.1176, over 28939.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3285, pruned_loss=0.09264, over 5695174.02 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3365, pruned_loss=0.09149, over 5768980.33 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3245, pruned_loss=0.09132, over 5680500.31 frames. ], batch size: 145, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:19:45,159 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:19:54,039 INFO [train.py:968] (0/2) Epoch 13, batch 18300, giga_loss[loss=0.2721, simple_loss=0.3456, pruned_loss=0.09933, over 28728.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3371, pruned_loss=0.09676, over 5693773.20 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3373, pruned_loss=0.09209, over 5766833.76 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.333, pruned_loss=0.09526, over 5681140.91 frames. ], batch size: 99, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:20:00,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7851, 2.3719, 1.6340, 0.8540], device='cuda:0'), covar=tensor([0.4076, 0.2222, 0.3303, 0.4493], device='cuda:0'), in_proj_covar=tensor([0.1593, 0.1519, 0.1500, 0.1301], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 19:20:15,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.127e+02 1.410e+03 1.742e+03 2.680e+03 5.753e+03, threshold=3.485e+03, percent-clipped=15.0
+2023-03-06 19:20:31,187 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:20:33,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:20:41,258 INFO [train.py:968] (0/2) Epoch 13, batch 18350, giga_loss[loss=0.2943, simple_loss=0.3771, pruned_loss=0.1057, over 28757.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3429, pruned_loss=0.09853, over 5687191.57 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3373, pruned_loss=0.09209, over 5766833.76 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3397, pruned_loss=0.09737, over 5677359.70 frames. ], batch size: 284, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:20:58,334 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:21:17,844 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:21:21,385 INFO [train.py:968] (0/2) Epoch 13, batch 18400, giga_loss[loss=0.2596, simple_loss=0.3429, pruned_loss=0.08815, over 28867.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3447, pruned_loss=0.09826, over 5684772.56 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.338, pruned_loss=0.09231, over 5767104.14 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3417, pruned_loss=0.0973, over 5674765.47 frames. ], batch size: 145, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:21:43,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.460e+02 1.107e+03 1.402e+03 1.878e+03 1.057e+04, threshold=2.804e+03, percent-clipped=6.0
+2023-03-06 19:21:55,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5058, 1.5779, 1.4679, 1.4102], device='cuda:0'), covar=tensor([0.1825, 0.1874, 0.1554, 0.1771], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1655, 0.1613, 0.1716], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 19:22:06,561 INFO [train.py:968] (0/2) Epoch 13, batch 18450, giga_loss[loss=0.3274, simple_loss=0.3965, pruned_loss=0.1292, over 28930.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3468, pruned_loss=0.09908, over 5680643.90 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3382, pruned_loss=0.09232, over 5768239.66 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3444, pruned_loss=0.09842, over 5669899.17 frames. ], batch size: 213, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:22:16,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3719, 1.2537, 4.1256, 3.1231], device='cuda:0'), covar=tensor([0.1599, 0.2662, 0.0359, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0668, 0.0591, 0.0851, 0.0771], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 19:22:21,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6465, 4.4505, 4.2230, 2.0888], device='cuda:0'), covar=tensor([0.0531, 0.0705, 0.0738, 0.2073], device='cuda:0'), in_proj_covar=tensor([0.1047, 0.0978, 0.0851, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 19:22:47,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7783, 3.5724, 3.3835, 1.7042], device='cuda:0'), covar=tensor([0.0676, 0.0779, 0.0736, 0.2435], device='cuda:0'), in_proj_covar=tensor([0.1049, 0.0980, 0.0852, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 19:22:49,183 INFO [train.py:968] (0/2) Epoch 13, batch 18500, giga_loss[loss=0.3052, simple_loss=0.3712, pruned_loss=0.1196, over 28751.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3492, pruned_loss=0.1012, over 5684673.54 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3381, pruned_loss=0.09215, over 5772159.33 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3475, pruned_loss=0.101, over 5670298.59 frames. ], batch size: 92, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:23:12,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.994e+02 1.157e+03 1.461e+03 2.010e+03 5.831e+03, threshold=2.922e+03, percent-clipped=8.0
+2023-03-06 19:23:12,580 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-06 19:23:21,122 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-566000.pt
+2023-03-06 19:23:34,886 INFO [train.py:968] (0/2) Epoch 13, batch 18550, giga_loss[loss=0.2677, simple_loss=0.3515, pruned_loss=0.09197, over 28956.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3531, pruned_loss=0.1042, over 5669740.74 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3383, pruned_loss=0.09232, over 5755098.96 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3517, pruned_loss=0.104, over 5672680.80 frames. ], batch size: 136, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:24:15,594 INFO [train.py:968] (0/2) Epoch 13, batch 18600, giga_loss[loss=0.2757, simple_loss=0.3524, pruned_loss=0.0995, over 28372.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3554, pruned_loss=0.1047, over 5677287.08 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3387, pruned_loss=0.09235, over 5756219.65 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3545, pruned_loss=0.105, over 5675538.87 frames. ], batch size: 77, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:24:28,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9181, 1.1300, 0.9947, 0.8257], device='cuda:0'), covar=tensor([0.1777, 0.2083, 0.1293, 0.1740], device='cuda:0'), in_proj_covar=tensor([0.1770, 0.1645, 0.1604, 0.1708], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 19:24:36,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.292e+02 1.197e+03 1.420e+03 1.786e+03 4.331e+03, threshold=2.840e+03, percent-clipped=4.0
+2023-03-06 19:24:55,874 INFO [train.py:968] (0/2) Epoch 13, batch 18650, giga_loss[loss=0.2931, simple_loss=0.3709, pruned_loss=0.1077, over 27896.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3569, pruned_loss=0.1046, over 5671324.87 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3388, pruned_loss=0.09242, over 5747428.04 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3565, pruned_loss=0.1051, over 5674834.30 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:25:05,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=566128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:25:15,280 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-06 19:25:37,192 INFO [train.py:968] (0/2) Epoch 13, batch 18700, giga_loss[loss=0.268, simple_loss=0.3499, pruned_loss=0.09305, over 29040.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3588, pruned_loss=0.1053, over 5682854.53 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3386, pruned_loss=0.09215, over 5751674.23 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3593, pruned_loss=0.1063, over 5679686.00 frames. ], batch size: 155, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:25:48,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3564, 0.9833, 4.7494, 3.4709], device='cuda:0'), covar=tensor([0.1795, 0.2996, 0.0328, 0.0792], device='cuda:0'), in_proj_covar=tensor([0.0672, 0.0595, 0.0855, 0.0778], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 19:25:56,025 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.242e+02 1.229e+03 1.596e+03 2.086e+03 7.174e+03, threshold=3.193e+03, percent-clipped=13.0
+2023-03-06 19:26:19,173 INFO [train.py:968] (0/2) Epoch 13, batch 18750, giga_loss[loss=0.2742, simple_loss=0.3581, pruned_loss=0.0951, over 28745.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3596, pruned_loss=0.1047, over 5676358.28 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.339, pruned_loss=0.09227, over 5741404.53 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3599, pruned_loss=0.1055, over 5682963.84 frames. ], batch size: 119, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:26:36,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=566236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:26:52,207 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7620, 1.8694, 1.5475, 2.1003], device='cuda:0'), covar=tensor([0.2418, 0.2458, 0.2615, 0.2282], device='cuda:0'), in_proj_covar=tensor([0.1329, 0.0978, 0.1176, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 19:27:02,976 INFO [train.py:968] (0/2) Epoch 13, batch 18800, giga_loss[loss=0.2689, simple_loss=0.3523, pruned_loss=0.09268, over 28533.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3582, pruned_loss=0.1022, over 5692482.52 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.339, pruned_loss=0.09227, over 5741404.53 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3584, pruned_loss=0.1028, over 5697623.72 frames. ], batch size: 60, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:27:07,599 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=566271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:27:09,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=566274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:27:21,114 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.234e+02 1.044e+03 1.298e+03 1.625e+03 3.855e+03, threshold=2.596e+03, percent-clipped=3.0
+2023-03-06 19:27:31,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=566303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:27:40,601 INFO [train.py:968] (0/2) Epoch 13, batch 18850, giga_loss[loss=0.2772, simple_loss=0.3567, pruned_loss=0.09887, over 28605.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3559, pruned_loss=0.1002, over 5691016.25 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.339, pruned_loss=0.09204, over 5736555.44 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3566, pruned_loss=0.1011, over 5698511.80 frames. ], batch size: 85, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:28:10,448 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5482, 1.7253, 1.3864, 1.2530], device='cuda:0'), covar=tensor([0.1975, 0.1856, 0.1888, 0.2030], device='cuda:0'), in_proj_covar=tensor([0.1771, 0.1646, 0.1615, 0.1712], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 19:28:21,607 INFO [train.py:968] (0/2) Epoch 13, batch 18900, giga_loss[loss=0.2956, simple_loss=0.3625, pruned_loss=0.1144, over 28859.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3576, pruned_loss=0.1021, over 5700304.30 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3395, pruned_loss=0.09214, over 5739751.17 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.358, pruned_loss=0.103, over 5702589.71 frames. ], batch size: 199, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:28:36,618 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=566379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:28:38,577 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=566382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:28:38,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4340, 1.7582, 1.3778, 1.5451], device='cuda:0'), covar=tensor([0.2303, 0.2214, 0.2422, 0.1934], device='cuda:0'), in_proj_covar=tensor([0.1329, 0.0979, 0.1178, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 19:28:43,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.454e+02 1.108e+03 1.507e+03 1.974e+03 4.939e+03, threshold=3.015e+03, percent-clipped=11.0
+2023-03-06 19:28:45,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3905, 3.3403, 1.4554, 1.4909], device='cuda:0'), covar=tensor([0.0919, 0.0247, 0.0878, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0507, 0.0346, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-06 19:29:01,479 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=566411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:29:03,870 INFO [train.py:968] (0/2) Epoch 13, batch 18950, giga_loss[loss=0.3197, simple_loss=0.3684, pruned_loss=0.1355, over 27667.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3595, pruned_loss=0.1056, over 5693995.22 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3401, pruned_loss=0.09248, over 5727184.99 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.36, pruned_loss=0.1064, over 5705515.47 frames. ], batch size: 472, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:29:42,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=566459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:29:48,349 INFO [train.py:968] (0/2) Epoch 13, batch 19000, giga_loss[loss=0.3059, simple_loss=0.3726, pruned_loss=0.1196, over 28647.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3606, pruned_loss=0.1084, over 5688274.23 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3405, pruned_loss=0.09264, over 5722314.14 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3608, pruned_loss=0.1091, over 5701351.17 frames. ], batch size: 242, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:30:05,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.493e+02 1.318e+03 1.739e+03 2.578e+03 6.339e+03, threshold=3.479e+03, percent-clipped=14.0
+2023-03-06 19:30:29,658 INFO [train.py:968] (0/2) Epoch 13, batch 19050, libri_loss[loss=0.2924, simple_loss=0.3757, pruned_loss=0.1045, over 26114.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.359, pruned_loss=0.1086, over 5686697.02 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3407, pruned_loss=0.09271, over 5720884.13 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3593, pruned_loss=0.1092, over 5698213.71 frames. ], batch size: 136, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:31:10,953 INFO [train.py:968] (0/2) Epoch 13, batch 19100, giga_loss[loss=0.2958, simple_loss=0.3639, pruned_loss=0.1138, over 28848.00 frames. ], tot_loss[loss=0.285, simple_loss=0.356, pruned_loss=0.107, over 5688894.71 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3408, pruned_loss=0.09254, over 5724629.30 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3566, pruned_loss=0.108, over 5693933.76 frames. ], batch size: 284, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:31:27,145 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-06 19:31:30,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.542e+02 1.233e+03 1.574e+03 2.099e+03 6.161e+03, threshold=3.148e+03, percent-clipped=4.0
+2023-03-06 19:31:46,450 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 19:31:53,256 INFO [train.py:968] (0/2) Epoch 13, batch 19150, giga_loss[loss=0.313, simple_loss=0.3737, pruned_loss=0.1261, over 27992.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3548, pruned_loss=0.1053, over 5698981.95 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3414, pruned_loss=0.09266, over 5721077.75 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3552, pruned_loss=0.1065, over 5704817.31 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:32:36,248 INFO [train.py:968] (0/2) Epoch 13, batch 19200, giga_loss[loss=0.3903, simple_loss=0.4178, pruned_loss=0.1813, over 26623.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3541, pruned_loss=0.1045, over 5701893.94 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3414, pruned_loss=0.09266, over 5721077.75 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3545, pruned_loss=0.1054, over 5706435.67 frames. ], batch size: 555, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:33:00,474 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.310e+02 1.179e+03 1.500e+03 2.753e+03 8.589e+03, threshold=3.001e+03, percent-clipped=16.0
+2023-03-06 19:33:21,883 INFO [train.py:968] (0/2) Epoch 13, batch 19250, giga_loss[loss=0.2525, simple_loss=0.3218, pruned_loss=0.09163, over 27506.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3507, pruned_loss=0.1027, over 5688097.52 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3413, pruned_loss=0.09256, over 5724711.66 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3513, pruned_loss=0.1037, over 5687993.78 frames. ], batch size: 472, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:34:08,650 INFO [train.py:968] (0/2) Epoch 13, batch 19300, giga_loss[loss=0.2349, simple_loss=0.3165, pruned_loss=0.07669, over 28614.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3446, pruned_loss=0.09911, over 5686212.84 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3416, pruned_loss=0.0926, over 5727722.38 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3447, pruned_loss=0.1, over 5682419.57 frames. ], batch size: 85, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:34:30,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.873e+02 9.617e+02 1.375e+03 1.977e+03 4.619e+03, threshold=2.751e+03, percent-clipped=3.0
+2023-03-06 19:34:51,928 INFO [train.py:968] (0/2) Epoch 13, batch 19350, giga_loss[loss=0.2219, simple_loss=0.3024, pruned_loss=0.07076, over 28743.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3397, pruned_loss=0.0962, over 5683824.98 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3424, pruned_loss=0.09289, over 5724448.63 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3392, pruned_loss=0.09685, over 5681667.32 frames. ], batch size: 119, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:35:12,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=566834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:35:41,637 INFO [train.py:968] (0/2) Epoch 13, batch 19400, giga_loss[loss=0.2539, simple_loss=0.333, pruned_loss=0.08742, over 28779.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3361, pruned_loss=0.09497, over 5657650.90 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3425, pruned_loss=0.09294, over 5728331.53 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3354, pruned_loss=0.09549, over 5651583.69 frames. ], batch size: 66, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:36:04,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.009e+02 9.906e+02 1.231e+03 1.665e+03 4.927e+03, threshold=2.462e+03, percent-clipped=5.0
+2023-03-06 19:36:24,192 INFO [train.py:968] (0/2) Epoch 13, batch 19450, giga_loss[loss=0.2655, simple_loss=0.335, pruned_loss=0.09795, over 28864.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3374, pruned_loss=0.09545, over 5666022.40 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3429, pruned_loss=0.09303, over 5731884.49 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3364, pruned_loss=0.09584, over 5656726.41 frames. ], batch size: 112, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:37:03,068 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=566956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:37:08,885 INFO [train.py:968] (0/2) Epoch 13, batch 19500, giga_loss[loss=0.2435, simple_loss=0.3157, pruned_loss=0.08571, over 28740.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3375, pruned_loss=0.09553, over 5661339.62 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3431, pruned_loss=0.09305, over 5725865.16 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3365, pruned_loss=0.09585, over 5658673.76 frames. ], batch size: 92, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:37:19,051 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=566977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:37:21,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=566980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:37:27,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.143e+02 1.034e+03 1.302e+03 1.662e+03 3.978e+03, threshold=2.605e+03, percent-clipped=7.0
+2023-03-06 19:37:43,607 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=567009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:37:48,300 INFO [train.py:968] (0/2) Epoch 13, batch 19550, libri_loss[loss=0.2764, simple_loss=0.3627, pruned_loss=0.09508, over 29650.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3358, pruned_loss=0.09423, over 5674692.97 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3433, pruned_loss=0.09288, over 5727493.25 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3346, pruned_loss=0.0947, over 5669332.42 frames. ], batch size: 88, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:38:28,024 INFO [train.py:968] (0/2) Epoch 13, batch 19600, giga_loss[loss=0.2392, simple_loss=0.3134, pruned_loss=0.08249, over 28083.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3346, pruned_loss=0.09367, over 5681803.38 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3441, pruned_loss=0.09306, over 5731716.98 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3326, pruned_loss=0.09389, over 5672329.86 frames. ], batch size: 77, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:38:48,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.318e+02 1.079e+03 1.356e+03 1.711e+03 4.745e+03, threshold=2.713e+03, percent-clipped=7.0
+2023-03-06 19:39:08,364 INFO [train.py:968] (0/2) Epoch 13, batch 19650, giga_loss[loss=0.2333, simple_loss=0.3113, pruned_loss=0.07765, over 29010.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3319, pruned_loss=0.09226, over 5686982.59 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3443, pruned_loss=0.09315, over 5724645.62 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3301, pruned_loss=0.09239, over 5684284.10 frames. ], batch size: 136, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:39:13,905 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-06 19:39:48,755 INFO [train.py:968] (0/2) Epoch 13, batch 19700, giga_loss[loss=0.2223, simple_loss=0.2958, pruned_loss=0.07433, over 28540.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3301, pruned_loss=0.09109, over 5691105.71 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3455, pruned_loss=0.09355, over 5726576.71 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3273, pruned_loss=0.09081, over 5686611.68 frames. ], batch size: 78, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:39:51,839 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:39:57,412 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-06 19:40:08,646 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.884e+02 9.997e+02 1.212e+03 1.794e+03 1.295e+04, threshold=2.424e+03, percent-clipped=10.0
+2023-03-06 19:40:27,276 INFO [train.py:968] (0/2) Epoch 13, batch 19750, giga_loss[loss=0.2382, simple_loss=0.3151, pruned_loss=0.08061, over 28746.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3293, pruned_loss=0.09077, over 5704142.74 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3462, pruned_loss=0.09375, over 5732541.05 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3256, pruned_loss=0.09026, over 5694134.89 frames. ], batch size: 262, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 19:41:05,393 INFO [train.py:968] (0/2) Epoch 13, batch 19800, libri_loss[loss=0.2501, simple_loss=0.3411, pruned_loss=0.07951, over 29580.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3262, pruned_loss=0.08905, over 5718249.71 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3467, pruned_loss=0.09367, over 5739586.85 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3222, pruned_loss=0.08856, over 5702960.00 frames. ], batch size: 78, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 19:41:26,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.531e+02 9.829e+02 1.286e+03 1.577e+03 6.054e+03, threshold=2.573e+03, percent-clipped=10.0
+2023-03-06 19:41:26,310 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:41:32,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6906, 1.7689, 1.7577, 1.6070], device='cuda:0'), covar=tensor([0.1742, 0.2275, 0.2255, 0.2127], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0734, 0.0678, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 19:41:45,311 INFO [train.py:968] (0/2) Epoch 13, batch 19850, giga_loss[loss=0.2275, simple_loss=0.3039, pruned_loss=0.07554, over 28998.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3246, pruned_loss=0.08793, over 5724805.01 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3473, pruned_loss=0.09383, over 5744857.76 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.32, pruned_loss=0.08723, over 5707226.56 frames. ], batch size: 164, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 19:41:57,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=567331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:42:25,505 INFO [train.py:968] (0/2) Epoch 13, batch 19900, giga_loss[loss=0.2152, simple_loss=0.2904, pruned_loss=0.06999, over 28439.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.324, pruned_loss=0.08772, over 5726660.51 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3479, pruned_loss=0.09395, over 5750985.25 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3189, pruned_loss=0.08682, over 5706366.14 frames. ], batch size: 65, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 19:42:46,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.101e+02 9.582e+02 1.151e+03 1.609e+03 5.798e+03, threshold=2.302e+03, percent-clipped=8.0
+2023-03-06 19:43:06,445 INFO [train.py:968] (0/2) Epoch 13, batch 19950, libri_loss[loss=0.2968, simple_loss=0.3794, pruned_loss=0.1071, over 29536.00 frames. ], tot_loss[loss=0.248, simple_loss=0.322, pruned_loss=0.08705, over 5727040.73 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3481, pruned_loss=0.09402, over 5751766.33 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3177, pruned_loss=0.08625, over 5710260.67 frames. ], batch size: 83, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 19:43:10,322 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-06 19:43:29,364 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5520, 1.7989, 1.4227, 1.7758], device='cuda:0'), covar=tensor([0.2414, 0.2474, 0.2739, 0.2308], device='cuda:0'), in_proj_covar=tensor([0.1340, 0.0984, 0.1183, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 19:43:45,325 INFO [train.py:968] (0/2) Epoch 13, batch 20000, giga_loss[loss=0.2726, simple_loss=0.3482, pruned_loss=0.09849, over 28707.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3226, pruned_loss=0.08736, over 5721502.18 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3483, pruned_loss=0.09412, over 5745889.55 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3183, pruned_loss=0.08645, over 5712450.21 frames. ], batch size: 307, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:43:50,107 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:43:53,595 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=567474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:43:55,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=567477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:44:07,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.308e+02 1.111e+03 1.299e+03 1.663e+03 4.284e+03, threshold=2.598e+03, percent-clipped=8.0
+2023-03-06 19:44:22,928 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=567506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:44:24,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3098, 1.1948, 4.0124, 3.1916], device='cuda:0'), covar=tensor([0.1592, 0.2631, 0.0403, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0672, 0.0592, 0.0854, 0.0774], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 19:44:30,150 INFO [train.py:968] (0/2) Epoch 13, batch 20050, giga_loss[loss=0.286, simple_loss=0.3575, pruned_loss=0.1072, over 28355.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3277, pruned_loss=0.09091, over 5718738.85 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3486, pruned_loss=0.09417, over 5747138.30 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3239, pruned_loss=0.09011, over 5710373.16 frames. ], batch size: 368, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:44:58,261 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=567543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:45:12,078 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-06 19:45:20,190 INFO [train.py:968] (0/2) Epoch 13, batch 20100, giga_loss[loss=0.3318, simple_loss=0.3983, pruned_loss=0.1327, over 28826.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.334, pruned_loss=0.09526, over 5702282.26 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3489, pruned_loss=0.09434, over 5749397.87 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3305, pruned_loss=0.09446, over 5693167.57 frames. ], batch size: 284, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:45:48,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.393e+02 1.308e+03 1.685e+03 2.256e+03 5.205e+03, threshold=3.370e+03, percent-clipped=15.0
+2023-03-06 19:46:08,047 INFO [train.py:968] (0/2) Epoch 13, batch 20150, libri_loss[loss=0.2628, simple_loss=0.3538, pruned_loss=0.08591, over 29231.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.343, pruned_loss=0.1013, over 5702611.64 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.349, pruned_loss=0.09419, over 5752067.29 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3399, pruned_loss=0.1009, over 5691570.37 frames. ], batch size: 94, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:46:50,939 INFO [train.py:968] (0/2) Epoch 13, batch 20200, giga_loss[loss=0.3383, simple_loss=0.3985, pruned_loss=0.139, over 27582.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3481, pruned_loss=0.1033, over 5686308.80 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3493, pruned_loss=0.09437, over 5737627.48 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3451, pruned_loss=0.1031, over 5687736.05 frames. ], batch size: 472, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:46:54,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=567667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:46:55,325 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:47:12,641 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=567686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:47:15,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=567689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:47:17,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.185e+02 1.183e+03 1.563e+03 2.204e+03 4.195e+03, threshold=3.127e+03, percent-clipped=7.0
+2023-03-06 19:47:35,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7175, 1.9294, 1.8174, 1.5082], device='cuda:0'), covar=tensor([0.2168, 0.1600, 0.1463, 0.1805], device='cuda:0'), in_proj_covar=tensor([0.1758, 0.1640, 0.1616, 0.1719], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 19:47:35,834 INFO [train.py:968] (0/2) Epoch 13, batch 20250, libri_loss[loss=0.2904, simple_loss=0.3762, pruned_loss=0.1023, over 29229.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3525, pruned_loss=0.1046, over 5689605.07 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3494, pruned_loss=0.09437, over 5741063.29 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3501, pruned_loss=0.1047, over 5685996.75 frames. ], batch size: 97, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:47:38,842 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=567718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:10,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9811, 1.3149, 1.1296, 0.1269], device='cuda:0'), covar=tensor([0.2386, 0.1921, 0.2827, 0.4240], device='cuda:0'), in_proj_covar=tensor([0.1588, 0.1505, 0.1497, 0.1297], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 19:48:19,588 INFO [train.py:968] (0/2) Epoch 13, batch 20300, giga_loss[loss=0.3025, simple_loss=0.3721, pruned_loss=0.1165, over 28536.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3571, pruned_loss=0.1069, over 5692000.86 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.349, pruned_loss=0.09418, over 5744847.74 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3557, pruned_loss=0.1076, over 5683725.84 frames. ], batch size: 71, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:48:23,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:32,191 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:40,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.291e+02 1.117e+03 1.374e+03 1.689e+03 6.599e+03, threshold=2.748e+03, percent-clipped=3.0
+2023-03-06 19:48:44,476 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:56,773 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=567810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:58,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=567813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:59,772 INFO [train.py:968] (0/2) Epoch 13, batch 20350, giga_loss[loss=0.2373, simple_loss=0.3212, pruned_loss=0.07668, over 28815.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3599, pruned_loss=0.1088, over 5699938.11 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3489, pruned_loss=0.09411, over 5748293.67 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3591, pruned_loss=0.1097, over 5689515.38 frames. ], batch size: 112, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:49:24,365 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=567842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:49:27,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=567846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:49:29,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5686, 1.8105, 1.4432, 1.8483], device='cuda:0'), covar=tensor([0.2558, 0.2472, 0.2616, 0.2191], device='cuda:0'), in_proj_covar=tensor([0.1330, 0.0977, 0.1174, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 19:49:44,631 INFO [train.py:968] (0/2) Epoch 13, batch 20400, giga_loss[loss=0.2478, simple_loss=0.339, pruned_loss=0.07828, over 28919.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.353, pruned_loss=0.1036, over 5700307.03 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3487, pruned_loss=0.09397, over 5751076.98 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3528, pruned_loss=0.1047, over 5688461.34 frames. ], batch size: 199, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:50:03,225 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:50:06,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.328e+02 1.036e+03 1.365e+03 1.837e+03 6.021e+03, threshold=2.731e+03, percent-clipped=7.0
+2023-03-06 19:50:25,577 INFO [train.py:968] (0/2) Epoch 13, batch 20450, giga_loss[loss=0.2563, simple_loss=0.3413, pruned_loss=0.08565, over 29053.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3521, pruned_loss=0.1025, over 5698845.15 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3487, pruned_loss=0.09408, over 5752160.19 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.352, pruned_loss=0.1036, over 5687040.40 frames. ], batch size: 128, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:50:38,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5129, 2.1762, 1.6124, 0.6840], device='cuda:0'), covar=tensor([0.3696, 0.1883, 0.2974, 0.3795], device='cuda:0'), in_proj_covar=tensor([0.1592, 0.1514, 0.1507, 0.1304], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 19:51:08,341 INFO [train.py:968] (0/2) Epoch 13, batch 20500, giga_loss[loss=0.2603, simple_loss=0.3413, pruned_loss=0.08959, over 29019.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3514, pruned_loss=0.1014, over 5698380.17 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.349, pruned_loss=0.09415, over 5753811.25 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3511, pruned_loss=0.1022, over 5687025.06 frames. ], batch size: 136, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:51:29,781 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=567989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:51:31,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.244e+02 1.216e+03 1.477e+03 1.837e+03 4.201e+03, threshold=2.955e+03, percent-clipped=4.0
+2023-03-06 19:51:31,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=567992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:51:37,345 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-568000.pt
+2023-03-06 19:51:48,457 INFO [train.py:968] (0/2) Epoch 13, batch 20550, giga_loss[loss=0.304, simple_loss=0.3827, pruned_loss=0.1127, over 28866.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.353, pruned_loss=0.1018, over 5707273.04 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3489, pruned_loss=0.09415, over 5759936.50 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3529, pruned_loss=0.1028, over 5690667.99 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:51:55,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:52:13,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:52:29,821 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:52:31,075 INFO [train.py:968] (0/2) Epoch 13, batch 20600, giga_loss[loss=0.2809, simple_loss=0.3556, pruned_loss=0.1031, over 28688.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3558, pruned_loss=0.1038, over 5708010.82 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3492, pruned_loss=0.09421, over 5764240.83 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3557, pruned_loss=0.1048, over 5689343.67 frames. ], batch size: 92, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:52:55,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.653e+02 1.217e+03 1.599e+03 2.053e+03 5.761e+03, threshold=3.199e+03, percent-clipped=9.0
+2023-03-06 19:53:12,145 INFO [train.py:968] (0/2) Epoch 13, batch 20650, giga_loss[loss=0.303, simple_loss=0.3704, pruned_loss=0.1178, over 28734.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3576, pruned_loss=0.1056, over 5712469.81 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3494, pruned_loss=0.09428, over 5768693.90 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3576, pruned_loss=0.1066, over 5692075.01 frames. ], batch size: 284, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:53:40,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:53:48,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:53:49,351 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 19:53:56,619 INFO [train.py:968] (0/2) Epoch 13, batch 20700, giga_loss[loss=0.2916, simple_loss=0.3664, pruned_loss=0.1084, over 28056.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3585, pruned_loss=0.1063, over 5721343.14 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3493, pruned_loss=0.09419, over 5770881.49 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3589, pruned_loss=0.1076, over 5700920.57 frames. ], batch size: 412, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:54:03,423 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:54:15,437 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:54:18,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:54:20,801 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.625e+02 1.282e+03 1.688e+03 2.445e+03 6.704e+03, threshold=3.376e+03, percent-clipped=12.0
+2023-03-06 19:54:40,568 INFO [train.py:968] (0/2) Epoch 13, batch 20750, giga_loss[loss=0.3763, simple_loss=0.4087, pruned_loss=0.1719, over 26605.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3587, pruned_loss=0.1073, over 5716152.12 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3492, pruned_loss=0.09422, over 5772752.75 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3593, pruned_loss=0.1085, over 5697392.60 frames. ], batch size: 555, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:54:44,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:19,185 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:20,156 INFO [train.py:968] (0/2) Epoch 13, batch 20800, giga_loss[loss=0.335, simple_loss=0.3945, pruned_loss=0.1377, over 27822.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3582, pruned_loss=0.1064, over 5721456.66 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3492, pruned_loss=0.09413, over 5774119.87 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3588, pruned_loss=0.1076, over 5704971.10 frames. ], batch size: 412, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:55:33,337 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:38,227 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:41,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:42,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.336e+02 1.156e+03 1.481e+03 2.114e+03 4.820e+03, threshold=2.962e+03, percent-clipped=6.0
+2023-03-06 19:55:48,128 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:50,056 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:56,571 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:59,917 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:56:00,268 INFO [train.py:968] (0/2) Epoch 13, batch 20850, giga_loss[loss=0.3154, simple_loss=0.3824, pruned_loss=0.1242, over 28248.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3581, pruned_loss=0.1053, over 5718807.66 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3491, pruned_loss=0.09423, over 5775682.00 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3588, pruned_loss=0.1064, over 5702973.04 frames. ], batch size: 368, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:56:02,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:56:05,635 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:56:13,074 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:56:25,489 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:56:39,234 INFO [train.py:968] (0/2) Epoch 13, batch 20900, libri_loss[loss=0.3107, simple_loss=0.3818, pruned_loss=0.1198, over 29534.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3592, pruned_loss=0.1052, over 5703418.17 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3496, pruned_loss=0.0945, over 5756090.00 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3596, pruned_loss=0.1062, over 5707446.67 frames. ], batch size: 82, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:57:01,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2077, 1.3228, 1.1067, 0.9915], device='cuda:0'), covar=tensor([0.0862, 0.0490, 0.1026, 0.0954], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0430, 0.0500, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 19:57:02,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.447e+02 1.033e+03 1.301e+03 1.765e+03 3.794e+03, threshold=2.601e+03, percent-clipped=3.0
+2023-03-06 19:57:12,761 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:57:14,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5875, 1.1511, 4.7626, 3.5250], device='cuda:0'), covar=tensor([0.1639, 0.2874, 0.0322, 0.0807], device='cuda:0'), in_proj_covar=tensor([0.0673, 0.0590, 0.0854, 0.0779], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 19:57:14,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:57:18,752 INFO [train.py:968] (0/2) Epoch 13, batch 20950, libri_loss[loss=0.2954, simple_loss=0.3743, pruned_loss=0.1082, over 29525.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3582, pruned_loss=0.1044, over 5705509.40 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3499, pruned_loss=0.09479, over 5749846.44 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3585, pruned_loss=0.1053, over 5711828.22 frames. ], batch size: 82, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:57:35,801 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:57:35,814 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:57:38,910 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-06 19:57:44,095 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-06 19:57:56,927 INFO [train.py:968] (0/2) Epoch 13, batch 21000, giga_loss[loss=0.2426, simple_loss=0.3197, pruned_loss=0.08281, over 28646.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.355, pruned_loss=0.1028, over 5705077.05 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3497, pruned_loss=0.09463, over 5752135.87 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3557, pruned_loss=0.1038, over 5707110.51 frames. ], batch size: 85, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:57:56,931 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 19:58:05,934 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2147, simple_loss=0.3211, pruned_loss=0.05416, over 944034.00 frames. 
+2023-03-06 19:58:05,935 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 19:58:28,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.035e+02 1.034e+03 1.280e+03 1.751e+03 8.454e+03, threshold=2.559e+03, percent-clipped=12.0
+2023-03-06 19:58:43,717 INFO [train.py:968] (0/2) Epoch 13, batch 21050, libri_loss[loss=0.3319, simple_loss=0.4032, pruned_loss=0.1303, over 29234.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3521, pruned_loss=0.1012, over 5710617.51 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3496, pruned_loss=0.09459, over 5754355.77 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3527, pruned_loss=0.1022, over 5709296.15 frames. ], batch size: 94, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:59:15,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2889, 5.0713, 4.8212, 2.5003], device='cuda:0'), covar=tensor([0.0413, 0.0628, 0.0672, 0.1758], device='cuda:0'), in_proj_covar=tensor([0.1069, 0.0994, 0.0864, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 19:59:21,728 INFO [train.py:968] (0/2) Epoch 13, batch 21100, giga_loss[loss=0.262, simple_loss=0.342, pruned_loss=0.09099, over 28969.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3501, pruned_loss=0.09984, over 5717657.60 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3495, pruned_loss=0.09451, over 5757454.37 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3507, pruned_loss=0.1008, over 5712969.76 frames. ], batch size: 128, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:59:35,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:59:38,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:59:46,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.085e+02 1.111e+03 1.432e+03 1.824e+03 3.969e+03, threshold=2.864e+03, percent-clipped=9.0
+2023-03-06 19:59:47,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2003, 1.2123, 3.8398, 3.1772], device='cuda:0'), covar=tensor([0.1698, 0.2714, 0.0417, 0.0963], device='cuda:0'), in_proj_covar=tensor([0.0677, 0.0593, 0.0859, 0.0782], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 20:00:00,935 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:00:01,886 INFO [train.py:968] (0/2) Epoch 13, batch 21150, giga_loss[loss=0.2898, simple_loss=0.3674, pruned_loss=0.1062, over 29086.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3516, pruned_loss=0.1016, over 5709822.87 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3495, pruned_loss=0.09464, over 5750573.63 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3521, pruned_loss=0.1024, over 5710862.13 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:00:35,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:00:43,289 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:00:44,241 INFO [train.py:968] (0/2) Epoch 13, batch 21200, giga_loss[loss=0.2656, simple_loss=0.3465, pruned_loss=0.09237, over 28762.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3513, pruned_loss=0.1012, over 5708973.12 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3491, pruned_loss=0.09446, over 5753755.57 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3521, pruned_loss=0.1021, over 5705779.35 frames. ], batch size: 284, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:01:02,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:01:09,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.759e+02 1.026e+03 1.221e+03 1.508e+03 3.904e+03, threshold=2.442e+03, percent-clipped=3.0
+2023-03-06 20:01:26,158 INFO [train.py:968] (0/2) Epoch 13, batch 21250, giga_loss[loss=0.2704, simple_loss=0.3497, pruned_loss=0.0956, over 28890.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3511, pruned_loss=0.1004, over 5706659.67 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3493, pruned_loss=0.09468, over 5745485.57 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3516, pruned_loss=0.101, over 5711364.27 frames. ], batch size: 199, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:02:10,905 INFO [train.py:968] (0/2) Epoch 13, batch 21300, giga_loss[loss=0.2641, simple_loss=0.3361, pruned_loss=0.09603, over 28470.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3507, pruned_loss=0.1001, over 5700985.46 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3498, pruned_loss=0.09497, over 5749564.71 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3507, pruned_loss=0.1004, over 5700082.57 frames. ], batch size: 60, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:02:35,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.281e+02 1.003e+03 1.175e+03 1.536e+03 7.234e+03, threshold=2.350e+03, percent-clipped=8.0
+2023-03-06 20:02:37,867 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:02:40,004 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:02:52,786 INFO [train.py:968] (0/2) Epoch 13, batch 21350, giga_loss[loss=0.2737, simple_loss=0.3509, pruned_loss=0.0983, over 28617.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3503, pruned_loss=0.1005, over 5696336.24 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3502, pruned_loss=0.09524, over 5750949.99 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.35, pruned_loss=0.1006, over 5694068.98 frames. ], batch size: 242, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:03:03,014 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:03:04,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:03:05,653 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:03:27,777 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:03:32,532 INFO [train.py:968] (0/2) Epoch 13, batch 21400, giga_loss[loss=0.2353, simple_loss=0.3118, pruned_loss=0.07942, over 28524.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3486, pruned_loss=0.1, over 5703103.07 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3502, pruned_loss=0.09524, over 5752500.64 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3483, pruned_loss=0.1002, over 5699493.76 frames. ], batch size: 71, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:03:39,957 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:03:50,288 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568887.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:03:55,955 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568893.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:03:57,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.494e+02 1.169e+03 1.549e+03 2.024e+03 8.085e+03, threshold=3.098e+03, percent-clipped=16.0
+2023-03-06 20:04:11,399 INFO [train.py:968] (0/2) Epoch 13, batch 21450, giga_loss[loss=0.3694, simple_loss=0.4072, pruned_loss=0.1658, over 26588.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3458, pruned_loss=0.09891, over 5701217.28 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3504, pruned_loss=0.09542, over 5756019.21 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3453, pruned_loss=0.09894, over 5693874.02 frames. ], batch size: 555, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:04:49,267 INFO [train.py:968] (0/2) Epoch 13, batch 21500, giga_loss[loss=0.2529, simple_loss=0.3188, pruned_loss=0.09351, over 28359.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3462, pruned_loss=0.09954, over 5699893.94 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.351, pruned_loss=0.09591, over 5755628.65 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3451, pruned_loss=0.09924, over 5692951.11 frames. ], batch size: 71, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:05:13,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.579e+02 1.004e+03 1.369e+03 1.738e+03 5.139e+03, threshold=2.737e+03, percent-clipped=3.0
+2023-03-06 20:05:20,204 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-06 20:05:30,752 INFO [train.py:968] (0/2) Epoch 13, batch 21550, giga_loss[loss=0.2435, simple_loss=0.3237, pruned_loss=0.08167, over 29070.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3444, pruned_loss=0.09863, over 5700242.81 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3509, pruned_loss=0.09579, over 5757216.34 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3436, pruned_loss=0.09853, over 5692972.82 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:05:37,014 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=569022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:05:50,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:06:11,027 INFO [train.py:968] (0/2) Epoch 13, batch 21600, giga_loss[loss=0.2125, simple_loss=0.2901, pruned_loss=0.06746, over 28603.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3416, pruned_loss=0.09751, over 5707033.02 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3508, pruned_loss=0.09582, over 5760545.57 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3408, pruned_loss=0.09746, over 5696726.47 frames. ], batch size: 60, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:06:34,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.452e+02 1.070e+03 1.444e+03 1.875e+03 3.644e+03, threshold=2.889e+03, percent-clipped=5.0
+2023-03-06 20:06:49,194 INFO [train.py:968] (0/2) Epoch 13, batch 21650, giga_loss[loss=0.271, simple_loss=0.3428, pruned_loss=0.09956, over 28265.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3414, pruned_loss=0.09795, over 5696432.50 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3515, pruned_loss=0.0964, over 5746295.52 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3398, pruned_loss=0.09741, over 5699101.96 frames. ], batch size: 368, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:07:12,649 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=569147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:07:25,994 INFO [train.py:968] (0/2) Epoch 13, batch 21700, giga_loss[loss=0.2571, simple_loss=0.3278, pruned_loss=0.0932, over 28746.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3389, pruned_loss=0.09648, over 5711044.11 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3516, pruned_loss=0.09651, over 5751031.92 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3372, pruned_loss=0.09597, over 5707308.43 frames. ], batch size: 119, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:07:37,574 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:07:39,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:07:47,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.898e+02 1.117e+03 1.569e+03 2.395e+03 6.794e+03, threshold=3.137e+03, percent-clipped=17.0
+2023-03-06 20:07:57,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 20:07:57,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0275, 2.0621, 1.8653, 1.8459], device='cuda:0'), covar=tensor([0.1502, 0.2286, 0.1986, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0724, 0.0672, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 20:08:00,582 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:08:01,915 INFO [train.py:968] (0/2) Epoch 13, batch 21750, giga_loss[loss=0.2936, simple_loss=0.3742, pruned_loss=0.1065, over 28878.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3376, pruned_loss=0.09558, over 5714265.91 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3515, pruned_loss=0.09655, over 5755180.61 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.336, pruned_loss=0.09514, over 5706715.09 frames. ], batch size: 227, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:08:32,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:08:42,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569262.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 20:08:44,701 INFO [train.py:968] (0/2) Epoch 13, batch 21800, giga_loss[loss=0.2436, simple_loss=0.3333, pruned_loss=0.07692, over 28984.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3403, pruned_loss=0.09689, over 5716298.40 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3516, pruned_loss=0.09656, over 5757994.37 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3387, pruned_loss=0.09653, over 5707290.93 frames. ], batch size: 164, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:08:48,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569268.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:09:14,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.051e+02 1.012e+03 1.226e+03 1.530e+03 3.847e+03, threshold=2.451e+03, percent-clipped=2.0
+2023-03-06 20:09:31,259 INFO [train.py:968] (0/2) Epoch 13, batch 21850, giga_loss[loss=0.2922, simple_loss=0.3733, pruned_loss=0.1055, over 28647.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3444, pruned_loss=0.0993, over 5703182.29 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3518, pruned_loss=0.09675, over 5758763.40 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.343, pruned_loss=0.09886, over 5695284.28 frames. ], batch size: 307, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:09:32,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2073, 1.5567, 1.5198, 1.0907], device='cuda:0'), covar=tensor([0.1597, 0.2333, 0.1383, 0.1585], device='cuda:0'), in_proj_covar=tensor([0.0842, 0.0686, 0.0883, 0.0789], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 20:10:10,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2380, 1.7418, 1.4533, 1.6190], device='cuda:0'), covar=tensor([0.0760, 0.0274, 0.0317, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0093], device='cuda:0')
+2023-03-06 20:10:12,106 INFO [train.py:968] (0/2) Epoch 13, batch 21900, giga_loss[loss=0.3055, simple_loss=0.376, pruned_loss=0.1175, over 28357.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3465, pruned_loss=0.09997, over 5704851.49 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3517, pruned_loss=0.09698, over 5759776.21 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3452, pruned_loss=0.0995, over 5695512.13 frames. ], batch size: 368, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:10:34,701 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:10:37,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:10:39,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.194e+02 1.028e+03 1.361e+03 2.111e+03 9.946e+03, threshold=2.722e+03, percent-clipped=17.0
+2023-03-06 20:10:39,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:10:42,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3276, 4.1473, 3.9130, 1.9491], device='cuda:0'), covar=tensor([0.0487, 0.0649, 0.0651, 0.2119], device='cuda:0'), in_proj_covar=tensor([0.1072, 0.0993, 0.0863, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 20:10:44,981 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569405.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:10:48,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569408.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:10:50,009 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569411.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:10:51,932 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569414.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 20:10:52,266 INFO [train.py:968] (0/2) Epoch 13, batch 21950, giga_loss[loss=0.3833, simple_loss=0.4172, pruned_loss=0.1747, over 26560.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3478, pruned_loss=0.1001, over 5700061.79 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3521, pruned_loss=0.09749, over 5753323.18 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3462, pruned_loss=0.09928, over 5696660.63 frames. ], batch size: 555, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:10:58,531 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:11:10,010 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569437.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 20:11:15,226 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569443.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:11:27,137 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-06 20:11:35,777 INFO [train.py:968] (0/2) Epoch 13, batch 22000, giga_loss[loss=0.2606, simple_loss=0.337, pruned_loss=0.09207, over 28925.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3461, pruned_loss=0.09834, over 5702095.76 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3526, pruned_loss=0.09805, over 5756579.32 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3443, pruned_loss=0.0973, over 5695591.78 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:11:53,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2374, 1.5085, 1.2010, 1.0242], device='cuda:0'), covar=tensor([0.2253, 0.2199, 0.2469, 0.2041], device='cuda:0'), in_proj_covar=tensor([0.1335, 0.0983, 0.1178, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 20:12:00,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.144e+02 1.085e+03 1.337e+03 1.756e+03 3.119e+03, threshold=2.674e+03, percent-clipped=5.0
+2023-03-06 20:12:16,440 INFO [train.py:968] (0/2) Epoch 13, batch 22050, giga_loss[loss=0.2847, simple_loss=0.3586, pruned_loss=0.1054, over 27710.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3465, pruned_loss=0.09859, over 5699029.09 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3535, pruned_loss=0.09858, over 5755971.20 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3441, pruned_loss=0.09728, over 5692677.00 frames. ], batch size: 472, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:12:17,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9143, 1.1116, 1.0431, 0.7957], device='cuda:0'), covar=tensor([0.1953, 0.2061, 0.1120, 0.1862], device='cuda:0'), in_proj_covar=tensor([0.1747, 0.1655, 0.1620, 0.1709], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 20:12:22,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:12:27,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2791, 1.3907, 4.1172, 3.2496], device='cuda:0'), covar=tensor([0.1614, 0.2476, 0.0386, 0.0766], device='cuda:0'), in_proj_covar=tensor([0.0674, 0.0594, 0.0863, 0.0780], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 20:12:37,062 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:12:39,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:12:56,708 INFO [train.py:968] (0/2) Epoch 13, batch 22100, giga_loss[loss=0.2524, simple_loss=0.3334, pruned_loss=0.08574, over 28872.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3465, pruned_loss=0.09898, over 5704790.94 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3538, pruned_loss=0.09879, over 5757742.45 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3442, pruned_loss=0.09775, over 5697571.33 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:13:01,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:13:09,506 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=569581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:13:11,683 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 20:13:23,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.544e+02 1.183e+03 1.434e+03 1.886e+03 8.973e+03, threshold=2.868e+03, percent-clipped=10.0
+2023-03-06 20:13:38,687 INFO [train.py:968] (0/2) Epoch 13, batch 22150, giga_loss[loss=0.341, simple_loss=0.3948, pruned_loss=0.1436, over 26588.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.349, pruned_loss=0.1009, over 5701555.20 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3543, pruned_loss=0.09928, over 5759402.26 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3466, pruned_loss=0.09955, over 5692869.80 frames. ], batch size: 555, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:14:19,200 INFO [train.py:968] (0/2) Epoch 13, batch 22200, libri_loss[loss=0.2905, simple_loss=0.3652, pruned_loss=0.1079, over 29557.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3531, pruned_loss=0.1034, over 5712433.11 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3547, pruned_loss=0.09955, over 5761947.04 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3508, pruned_loss=0.1021, over 5702460.99 frames. ], batch size: 78, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:14:19,581 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:14:21,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:14:38,788 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=569688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:14:45,326 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.362e+03 1.710e+03 2.410e+03 6.495e+03, threshold=3.419e+03, percent-clipped=17.0
+2023-03-06 20:14:45,517 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:15:00,505 INFO [train.py:968] (0/2) Epoch 13, batch 22250, giga_loss[loss=0.2976, simple_loss=0.3647, pruned_loss=0.1153, over 28534.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3552, pruned_loss=0.1039, over 5711799.23 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3557, pruned_loss=0.09996, over 5758238.71 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3524, pruned_loss=0.1026, over 5705390.79 frames. ], batch size: 60, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:15:38,364 INFO [train.py:968] (0/2) Epoch 13, batch 22300, giga_loss[loss=0.2728, simple_loss=0.3503, pruned_loss=0.09767, over 28954.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3564, pruned_loss=0.1043, over 5721042.05 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3561, pruned_loss=0.1003, over 5762604.05 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3538, pruned_loss=0.103, over 5710286.33 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:15:56,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=569789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:16:02,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.834e+02 1.219e+03 1.632e+03 2.375e+03 7.465e+03, threshold=3.264e+03, percent-clipped=8.0
+2023-03-06 20:16:14,462 INFO [train.py:968] (0/2) Epoch 13, batch 22350, giga_loss[loss=0.3073, simple_loss=0.3723, pruned_loss=0.1211, over 28717.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3573, pruned_loss=0.1049, over 5726140.01 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3566, pruned_loss=0.1007, over 5762196.35 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3548, pruned_loss=0.1037, over 5716646.09 frames. ], batch size: 92, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:16:58,705 INFO [train.py:968] (0/2) Epoch 13, batch 22400, giga_loss[loss=0.2789, simple_loss=0.3426, pruned_loss=0.1076, over 28731.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3584, pruned_loss=0.1058, over 5720053.87 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.357, pruned_loss=0.1008, over 5764849.57 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3561, pruned_loss=0.1048, over 5709370.08 frames. ], batch size: 66, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:17:12,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2223, 1.9302, 1.5032, 1.6573], device='cuda:0'), covar=tensor([0.0803, 0.0785, 0.1018, 0.1183], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0438, 0.0503, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 20:17:25,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.977e+02 1.188e+03 1.525e+03 1.988e+03 4.727e+03, threshold=3.050e+03, percent-clipped=4.0
+2023-03-06 20:17:34,566 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-06 20:17:36,160 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 20:17:40,728 INFO [train.py:968] (0/2) Epoch 13, batch 22450, giga_loss[loss=0.2895, simple_loss=0.3508, pruned_loss=0.1142, over 28774.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3571, pruned_loss=0.1055, over 5722260.30 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3575, pruned_loss=0.1012, over 5768152.74 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3547, pruned_loss=0.1044, over 5709783.94 frames. ], batch size: 92, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:18:15,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:18:21,965 INFO [train.py:968] (0/2) Epoch 13, batch 22500, giga_loss[loss=0.2904, simple_loss=0.3552, pruned_loss=0.1128, over 28919.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3539, pruned_loss=0.1038, over 5718424.72 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3577, pruned_loss=0.1014, over 5760241.46 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3519, pruned_loss=0.1028, over 5715217.17 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:18:25,321 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 20:18:30,641 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=569975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:18:48,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.395e+02 1.035e+03 1.277e+03 1.671e+03 4.060e+03, threshold=2.554e+03, percent-clipped=6.0
+2023-03-06 20:18:49,982 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-570000.pt
+2023-03-06 20:19:02,070 INFO [train.py:968] (0/2) Epoch 13, batch 22550, giga_loss[loss=0.2341, simple_loss=0.3177, pruned_loss=0.07522, over 28854.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3504, pruned_loss=0.1022, over 5718117.05 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.358, pruned_loss=0.1017, over 5763267.42 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3484, pruned_loss=0.1012, over 5711971.50 frames. ], batch size: 174, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:19:10,061 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:19:40,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:19:41,822 INFO [train.py:968] (0/2) Epoch 13, batch 22600, giga_loss[loss=0.2401, simple_loss=0.3277, pruned_loss=0.07623, over 28980.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.349, pruned_loss=0.1008, over 5718141.19 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3578, pruned_loss=0.1018, over 5763504.49 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3474, pruned_loss=0.09984, over 5712308.89 frames. ], batch size: 164, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:20:02,008 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-06 20:20:12,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.936e+02 1.051e+03 1.324e+03 1.905e+03 4.514e+03, threshold=2.649e+03, percent-clipped=11.0
+2023-03-06 20:20:13,269 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:20:16,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=570102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:20:26,823 INFO [train.py:968] (0/2) Epoch 13, batch 22650, giga_loss[loss=0.255, simple_loss=0.3303, pruned_loss=0.08986, over 28885.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3502, pruned_loss=0.1001, over 5714445.32 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3579, pruned_loss=0.1018, over 5763704.17 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3489, pruned_loss=0.09936, over 5709176.92 frames. ], batch size: 106, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:20:38,873 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:20:52,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4799, 5.2675, 5.0131, 2.3942], device='cuda:0'), covar=tensor([0.0394, 0.0582, 0.0613, 0.1873], device='cuda:0'), in_proj_covar=tensor([0.1069, 0.0988, 0.0864, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 20:21:04,362 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:21:05,668 INFO [train.py:968] (0/2) Epoch 13, batch 22700, giga_loss[loss=0.2806, simple_loss=0.3542, pruned_loss=0.1036, over 29048.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3502, pruned_loss=0.1, over 5725081.34 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3581, pruned_loss=0.1023, over 5767204.14 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3487, pruned_loss=0.09895, over 5716579.59 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:21:25,655 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4426, 1.6549, 1.3160, 1.7061], device='cuda:0'), covar=tensor([0.2346, 0.2337, 0.2645, 0.2176], device='cuda:0'), in_proj_covar=tensor([0.1335, 0.0984, 0.1179, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 20:21:32,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.420e+02 1.157e+03 1.415e+03 1.889e+03 4.068e+03, threshold=2.830e+03, percent-clipped=9.0
+2023-03-06 20:21:41,307 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:21:43,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=570209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:21:48,399 INFO [train.py:968] (0/2) Epoch 13, batch 22750, giga_loss[loss=0.26, simple_loss=0.3243, pruned_loss=0.09783, over 28058.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3487, pruned_loss=0.1002, over 5726002.67 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3581, pruned_loss=0.1023, over 5769514.22 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3474, pruned_loss=0.09932, over 5716540.80 frames. ], batch size: 77, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:22:06,592 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:22:07,861 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:22:27,423 INFO [train.py:968] (0/2) Epoch 13, batch 22800, giga_loss[loss=0.2716, simple_loss=0.3402, pruned_loss=0.1015, over 28809.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3464, pruned_loss=0.1003, over 5721452.03 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3584, pruned_loss=0.1026, over 5762769.58 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3449, pruned_loss=0.0993, over 5719386.01 frames. ], batch size: 199, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:22:42,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5794, 1.9717, 1.7323, 1.4284], device='cuda:0'), covar=tensor([0.2997, 0.2007, 0.2221, 0.2458], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1687, 0.1648, 0.1732], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 20:22:55,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.683e+02 1.127e+03 1.337e+03 1.742e+03 4.761e+03, threshold=2.675e+03, percent-clipped=6.0
+2023-03-06 20:23:02,944 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:23:04,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=570310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:23:09,607 INFO [train.py:968] (0/2) Epoch 13, batch 22850, giga_loss[loss=0.2918, simple_loss=0.3505, pruned_loss=0.1166, over 28706.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3464, pruned_loss=0.1021, over 5713892.91 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3586, pruned_loss=0.103, over 5756559.09 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3448, pruned_loss=0.1009, over 5716660.39 frames. ], batch size: 92, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:23:10,588 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5126, 1.1677, 4.5653, 3.7043], device='cuda:0'), covar=tensor([0.1603, 0.2801, 0.0356, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0678, 0.0600, 0.0873, 0.0794], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 20:23:30,295 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:23:40,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:23:50,598 INFO [train.py:968] (0/2) Epoch 13, batch 22900, libri_loss[loss=0.2412, simple_loss=0.3224, pruned_loss=0.08, over 29609.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.345, pruned_loss=0.1019, over 5720436.66 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3585, pruned_loss=0.103, over 5759363.77 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3435, pruned_loss=0.1009, over 5719076.99 frames. ], batch size: 75, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:23:55,477 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-06 20:24:17,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.129e+02 1.155e+03 1.512e+03 1.942e+03 7.611e+03, threshold=3.024e+03, percent-clipped=12.0
+2023-03-06 20:24:18,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:24:31,516 INFO [train.py:968] (0/2) Epoch 13, batch 22950, giga_loss[loss=0.3214, simple_loss=0.3717, pruned_loss=0.1356, over 26724.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3432, pruned_loss=0.1015, over 5707465.89 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3592, pruned_loss=0.1037, over 5751738.43 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3413, pruned_loss=0.1001, over 5712513.10 frames. ], batch size: 555, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:24:40,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6050, 1.7925, 1.4870, 1.7962], device='cuda:0'), covar=tensor([0.2282, 0.2350, 0.2640, 0.2432], device='cuda:0'), in_proj_covar=tensor([0.1328, 0.0977, 0.1172, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 20:25:06,888 INFO [train.py:968] (0/2) Epoch 13, batch 23000, giga_loss[loss=0.2248, simple_loss=0.3019, pruned_loss=0.07383, over 29039.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3402, pruned_loss=0.0999, over 5707748.47 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3595, pruned_loss=0.1042, over 5743436.42 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3377, pruned_loss=0.09815, over 5718693.01 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:25:19,870 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:25:29,157 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:25:32,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=570496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:25:34,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.737e+02 1.142e+03 1.480e+03 2.049e+03 4.403e+03, threshold=2.959e+03, percent-clipped=6.0
+2023-03-06 20:25:46,595 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:25:46,989 INFO [train.py:968] (0/2) Epoch 13, batch 23050, giga_loss[loss=0.2335, simple_loss=0.3122, pruned_loss=0.07737, over 29014.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3367, pruned_loss=0.098, over 5706556.09 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3602, pruned_loss=0.1047, over 5741260.68 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3336, pruned_loss=0.09601, over 5716487.46 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:25:53,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5790, 2.3121, 1.8069, 0.8284], device='cuda:0'), covar=tensor([0.4992, 0.2222, 0.3468, 0.5280], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1503, 0.1497, 0.1305], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 20:25:54,631 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:10,456 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:12,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=570547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:18,686 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-06 20:26:26,188 INFO [train.py:968] (0/2) Epoch 13, batch 23100, giga_loss[loss=0.3136, simple_loss=0.3639, pruned_loss=0.1317, over 23679.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3363, pruned_loss=0.09731, over 5709300.28 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3601, pruned_loss=0.1048, over 5744532.94 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3334, pruned_loss=0.09549, over 5713787.23 frames. ], batch size: 705, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:26:30,339 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-06 20:26:36,550 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:39,063 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:44,993 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:56,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.644e+02 1.246e+03 1.837e+03 2.798e+03 7.390e+03, threshold=3.675e+03, percent-clipped=20.0
+2023-03-06 20:27:07,240 INFO [train.py:968] (0/2) Epoch 13, batch 23150, libri_loss[loss=0.2658, simple_loss=0.3405, pruned_loss=0.09551, over 29558.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3397, pruned_loss=0.09893, over 5709616.24 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3601, pruned_loss=0.1051, over 5745640.11 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3366, pruned_loss=0.09699, over 5710997.88 frames. ], batch size: 76, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:27:07,451 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:27:48,038 INFO [train.py:968] (0/2) Epoch 13, batch 23200, giga_loss[loss=0.2424, simple_loss=0.3191, pruned_loss=0.08288, over 28984.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3423, pruned_loss=0.09991, over 5712552.52 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3604, pruned_loss=0.1056, over 5748862.81 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3393, pruned_loss=0.09783, over 5709859.81 frames. ], batch size: 106, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:28:12,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.523e+02 1.204e+03 1.583e+03 2.157e+03 6.522e+03, threshold=3.166e+03, percent-clipped=5.0
+2023-03-06 20:28:23,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1305, 1.1461, 3.7281, 3.2654], device='cuda:0'), covar=tensor([0.2082, 0.3150, 0.0749, 0.1258], device='cuda:0'), in_proj_covar=tensor([0.0682, 0.0603, 0.0877, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 20:28:25,465 INFO [train.py:968] (0/2) Epoch 13, batch 23250, giga_loss[loss=0.2936, simple_loss=0.3692, pruned_loss=0.109, over 28930.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.346, pruned_loss=0.1013, over 5715908.96 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3604, pruned_loss=0.106, over 5753558.40 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3428, pruned_loss=0.099, over 5708010.06 frames. ], batch size: 227, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:28:30,870 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-06 20:28:33,832 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:28:49,452 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:28:59,481 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:29:02,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=570761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:29:04,515 INFO [train.py:968] (0/2) Epoch 13, batch 23300, giga_loss[loss=0.271, simple_loss=0.3438, pruned_loss=0.09912, over 28718.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3487, pruned_loss=0.1024, over 5715413.44 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3602, pruned_loss=0.1061, over 5746805.78 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3459, pruned_loss=0.1004, over 5713229.10 frames. ], batch size: 92, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:29:27,177 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:29:35,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.710e+02 1.183e+03 1.591e+03 2.249e+03 1.066e+04, threshold=3.181e+03, percent-clipped=12.0
+2023-03-06 20:29:45,137 INFO [train.py:968] (0/2) Epoch 13, batch 23350, giga_loss[loss=0.2947, simple_loss=0.3667, pruned_loss=0.1113, over 28445.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3505, pruned_loss=0.1036, over 5716527.95 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3599, pruned_loss=0.1062, over 5742718.93 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3483, pruned_loss=0.1018, over 5716912.62 frames. ], batch size: 60, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:30:24,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:30:34,399 INFO [train.py:968] (0/2) Epoch 13, batch 23400, giga_loss[loss=0.2853, simple_loss=0.3555, pruned_loss=0.1076, over 28946.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3574, pruned_loss=0.1096, over 5702494.36 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3602, pruned_loss=0.1066, over 5743605.08 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3552, pruned_loss=0.1078, over 5701375.01 frames. ], batch size: 199, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:30:52,479 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:30:57,301 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:31:11,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.122e+02 1.544e+03 1.804e+03 2.215e+03 7.340e+03, threshold=3.608e+03, percent-clipped=7.0
+2023-03-06 20:31:23,073 INFO [train.py:968] (0/2) Epoch 13, batch 23450, libri_loss[loss=0.2841, simple_loss=0.3548, pruned_loss=0.1067, over 29507.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3637, pruned_loss=0.1146, over 5698293.85 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3604, pruned_loss=0.1069, over 5746596.51 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3618, pruned_loss=0.1129, over 5693895.10 frames. ], batch size: 81, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:31:40,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 20:32:07,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:32:13,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:32:13,179 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3865, 1.9161, 1.4304, 0.7357], device='cuda:0'), covar=tensor([0.2808, 0.1646, 0.2152, 0.3668], device='cuda:0'), in_proj_covar=tensor([0.1590, 0.1508, 0.1503, 0.1311], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 20:32:15,535 INFO [train.py:968] (0/2) Epoch 13, batch 23500, giga_loss[loss=0.301, simple_loss=0.3762, pruned_loss=0.1129, over 28940.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3713, pruned_loss=0.1201, over 5689624.27 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3604, pruned_loss=0.1069, over 5748089.20 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3699, pruned_loss=0.119, over 5684217.45 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:32:47,022 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:32:49,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.606e+03 2.103e+03 2.682e+03 5.871e+03, threshold=4.207e+03, percent-clipped=8.0
+2023-03-06 20:32:50,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:33:02,174 INFO [train.py:968] (0/2) Epoch 13, batch 23550, giga_loss[loss=0.3625, simple_loss=0.3885, pruned_loss=0.1683, over 23438.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3762, pruned_loss=0.1242, over 5686427.03 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3611, pruned_loss=0.1077, over 5748536.92 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3749, pruned_loss=0.1231, over 5679624.21 frames. ], batch size: 705, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:33:16,626 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:33:17,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:33:19,587 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:33:41,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5869, 1.4236, 4.6141, 3.4991], device='cuda:0'), covar=tensor([0.1520, 0.2546, 0.0354, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0680, 0.0599, 0.0870, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 20:33:49,226 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:33:49,665 INFO [train.py:968] (0/2) Epoch 13, batch 23600, giga_loss[loss=0.3149, simple_loss=0.3806, pruned_loss=0.1246, over 28483.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3827, pruned_loss=0.1299, over 5688997.26 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3609, pruned_loss=0.1078, over 5751133.33 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3822, pruned_loss=0.1294, over 5679836.35 frames. ], batch size: 71, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:34:22,090 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:23,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=571100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:23,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.664e+03 2.282e+03 3.238e+03 1.143e+04, threshold=4.564e+03, percent-clipped=13.0
+2023-03-06 20:34:23,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:26,630 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:29,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:33,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:35,143 INFO [train.py:968] (0/2) Epoch 13, batch 23650, libri_loss[loss=0.2887, simple_loss=0.3636, pruned_loss=0.1069, over 19507.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3862, pruned_loss=0.1331, over 5676868.64 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3613, pruned_loss=0.1083, over 5747702.86 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3865, pruned_loss=0.1332, over 5670341.16 frames. ], batch size: 186, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:34:39,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=571121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:48,077 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:54,455 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:35:21,242 INFO [train.py:968] (0/2) Epoch 13, batch 23700, giga_loss[loss=0.3175, simple_loss=0.3896, pruned_loss=0.1227, over 28972.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3872, pruned_loss=0.1345, over 5664409.40 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3613, pruned_loss=0.1084, over 5740620.37 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.388, pruned_loss=0.1351, over 5664448.76 frames. ], batch size: 164, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:35:56,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.638e+03 2.141e+03 2.927e+03 9.845e+03, threshold=4.281e+03, percent-clipped=2.0
+2023-03-06 20:36:11,950 INFO [train.py:968] (0/2) Epoch 13, batch 23750, libri_loss[loss=0.3173, simple_loss=0.3778, pruned_loss=0.1285, over 29559.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3917, pruned_loss=0.1402, over 5655741.49 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.361, pruned_loss=0.1086, over 5744976.03 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3934, pruned_loss=0.1413, over 5649676.27 frames. ], batch size: 79, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:36:39,772 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:36:42,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:36:58,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=571259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:03,963 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:04,362 INFO [train.py:968] (0/2) Epoch 13, batch 23800, libri_loss[loss=0.3122, simple_loss=0.3782, pruned_loss=0.1232, over 29738.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3957, pruned_loss=0.145, over 5650416.02 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.361, pruned_loss=0.1087, over 5748664.06 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3977, pruned_loss=0.1463, over 5640725.31 frames. ], batch size: 87, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:37:05,815 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:05,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:12,433 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:32,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:39,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.870e+03 2.490e+03 3.734e+03 7.963e+03, threshold=4.979e+03, percent-clipped=18.0
+2023-03-06 20:37:53,990 INFO [train.py:968] (0/2) Epoch 13, batch 23850, giga_loss[loss=0.3331, simple_loss=0.3985, pruned_loss=0.1338, over 28905.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3963, pruned_loss=0.1441, over 5654125.55 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3615, pruned_loss=0.1093, over 5744170.35 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.399, pruned_loss=0.1462, over 5645694.17 frames. ], batch size: 174, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:38:44,819 INFO [train.py:968] (0/2) Epoch 13, batch 23900, giga_loss[loss=0.4525, simple_loss=0.4631, pruned_loss=0.2209, over 26565.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3971, pruned_loss=0.1455, over 5651793.75 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3619, pruned_loss=0.1096, over 5745565.36 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4, pruned_loss=0.1481, over 5640709.88 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:39:20,695 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.794e+03 2.299e+03 2.977e+03 6.036e+03, threshold=4.598e+03, percent-clipped=5.0
+2023-03-06 20:39:21,706 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:39:27,579 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:39:36,399 INFO [train.py:968] (0/2) Epoch 13, batch 23950, giga_loss[loss=0.3799, simple_loss=0.4183, pruned_loss=0.1707, over 27465.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3956, pruned_loss=0.1452, over 5642415.03 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3621, pruned_loss=0.1098, over 5742590.53 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.3979, pruned_loss=0.1472, over 5635821.46 frames. ], batch size: 472, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:39:44,654 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1541, 1.1909, 3.5626, 3.0509], device='cuda:0'), covar=tensor([0.1566, 0.2515, 0.0467, 0.1079], device='cuda:0'), in_proj_covar=tensor([0.0690, 0.0605, 0.0885, 0.0804], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 20:39:52,535 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:40:11,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1115, 1.2537, 3.5869, 3.0497], device='cuda:0'), covar=tensor([0.1661, 0.2635, 0.0465, 0.1017], device='cuda:0'), in_proj_covar=tensor([0.0689, 0.0605, 0.0883, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 20:40:21,757 INFO [train.py:968] (0/2) Epoch 13, batch 24000, giga_loss[loss=0.3453, simple_loss=0.4049, pruned_loss=0.1429, over 28837.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.3958, pruned_loss=0.1448, over 5647240.60 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3625, pruned_loss=0.1104, over 5742945.97 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3981, pruned_loss=0.1469, over 5638962.37 frames. ], batch size: 186, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 20:40:21,763 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 20:40:30,434 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2129, simple_loss=0.3197, pruned_loss=0.05301, over 944034.00 frames. 
+2023-03-06 20:40:30,435 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 20:40:49,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=571487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:41:04,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.842e+03 2.462e+03 3.968e+03 9.252e+03, threshold=4.925e+03, percent-clipped=14.0
+2023-03-06 20:41:19,207 INFO [train.py:968] (0/2) Epoch 13, batch 24050, giga_loss[loss=0.3151, simple_loss=0.3849, pruned_loss=0.1227, over 29000.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3955, pruned_loss=0.1438, over 5625071.43 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3629, pruned_loss=0.1108, over 5718658.90 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3977, pruned_loss=0.1458, over 5637254.91 frames. ], batch size: 213, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:41:26,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2970, 1.5121, 1.5355, 1.3651], device='cuda:0'), covar=tensor([0.1511, 0.1517, 0.1851, 0.1492], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0739, 0.0683, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 20:41:40,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6442, 1.9407, 1.5400, 1.5886], device='cuda:0'), covar=tensor([0.2400, 0.2337, 0.2507, 0.2089], device='cuda:0'), in_proj_covar=tensor([0.1334, 0.0986, 0.1180, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 20:41:49,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-06 20:42:10,582 INFO [train.py:968] (0/2) Epoch 13, batch 24100, giga_loss[loss=0.3491, simple_loss=0.4083, pruned_loss=0.1449, over 28977.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3962, pruned_loss=0.144, over 5608981.70 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.363, pruned_loss=0.1111, over 5711306.37 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.3982, pruned_loss=0.1458, over 5623181.62 frames. ], batch size: 227, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:42:48,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.934e+02 1.650e+03 2.032e+03 2.707e+03 6.451e+03, threshold=4.065e+03, percent-clipped=5.0
+2023-03-06 20:43:01,653 INFO [train.py:968] (0/2) Epoch 13, batch 24150, giga_loss[loss=0.3559, simple_loss=0.3952, pruned_loss=0.1583, over 24035.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3935, pruned_loss=0.1417, over 5607987.06 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3631, pruned_loss=0.1114, over 5704531.38 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3955, pruned_loss=0.1433, over 5623476.20 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:43:18,799 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:43:20,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:43:29,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=571642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:43:46,640 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:43:50,589 INFO [train.py:968] (0/2) Epoch 13, batch 24200, giga_loss[loss=0.2628, simple_loss=0.3489, pruned_loss=0.08836, over 28764.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3892, pruned_loss=0.1367, over 5619900.51 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.363, pruned_loss=0.1115, over 5705470.24 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3916, pruned_loss=0.1385, over 5629212.36 frames. ], batch size: 92, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:44:24,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.322e+02 1.453e+03 1.983e+03 2.702e+03 6.135e+03, threshold=3.965e+03, percent-clipped=13.0
+2023-03-06 20:44:30,190 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:44:39,509 INFO [train.py:968] (0/2) Epoch 13, batch 24250, giga_loss[loss=0.3106, simple_loss=0.3808, pruned_loss=0.1202, over 28843.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3858, pruned_loss=0.133, over 5636602.00 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3632, pruned_loss=0.1118, over 5708901.56 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3881, pruned_loss=0.1347, over 5638942.51 frames. ], batch size: 145, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:45:26,106 INFO [train.py:968] (0/2) Epoch 13, batch 24300, giga_loss[loss=0.2885, simple_loss=0.3607, pruned_loss=0.1081, over 28618.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3831, pruned_loss=0.1301, over 5645862.43 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3635, pruned_loss=0.1121, over 5699475.29 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.385, pruned_loss=0.1315, over 5655370.15 frames. ], batch size: 242, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:45:26,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5589, 1.6442, 1.7907, 1.3613], device='cuda:0'), covar=tensor([0.1679, 0.2318, 0.1350, 0.1586], device='cuda:0'), in_proj_covar=tensor([0.0838, 0.0689, 0.0879, 0.0785], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 20:45:34,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7614, 2.0116, 1.6188, 2.0326], device='cuda:0'), covar=tensor([0.2336, 0.2318, 0.2636, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.1337, 0.0984, 0.1179, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 20:45:47,118 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:45:49,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:46:01,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.199e+02 1.589e+03 2.057e+03 2.934e+03 7.410e+03, threshold=4.115e+03, percent-clipped=12.0
+2023-03-06 20:46:04,331 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 20:46:14,895 INFO [train.py:968] (0/2) Epoch 13, batch 24350, giga_loss[loss=0.379, simple_loss=0.4085, pruned_loss=0.1748, over 26676.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3802, pruned_loss=0.1286, over 5644314.21 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3637, pruned_loss=0.1124, over 5702749.26 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3818, pruned_loss=0.1297, over 5647957.27 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:46:16,398 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:46:38,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1564, 1.2207, 3.4008, 2.9825], device='cuda:0'), covar=tensor([0.1545, 0.2463, 0.0482, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0603, 0.0878, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 20:46:57,929 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2727, 2.4742, 1.3051, 1.3783], device='cuda:0'), covar=tensor([0.0912, 0.0396, 0.0844, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0518, 0.0348, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 20:46:58,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:46:59,108 INFO [train.py:968] (0/2) Epoch 13, batch 24400, giga_loss[loss=0.34, simple_loss=0.3976, pruned_loss=0.1412, over 28826.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3803, pruned_loss=0.1284, over 5649903.33 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3639, pruned_loss=0.1127, over 5687873.56 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3818, pruned_loss=0.1294, over 5665170.38 frames. ], batch size: 199, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:47:35,736 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:47:42,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.482e+03 1.893e+03 2.549e+03 5.141e+03, threshold=3.787e+03, percent-clipped=2.0
+2023-03-06 20:47:52,242 INFO [train.py:968] (0/2) Epoch 13, batch 24450, giga_loss[loss=0.3038, simple_loss=0.3695, pruned_loss=0.119, over 28606.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3811, pruned_loss=0.1289, over 5654275.03 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3641, pruned_loss=0.1131, over 5690651.67 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3824, pruned_loss=0.1297, over 5663295.34 frames. ], batch size: 336, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:48:47,860 INFO [train.py:968] (0/2) Epoch 13, batch 24500, giga_loss[loss=0.3173, simple_loss=0.3891, pruned_loss=0.1227, over 28677.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3793, pruned_loss=0.1263, over 5657498.02 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3642, pruned_loss=0.1132, over 5691714.41 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3803, pruned_loss=0.1268, over 5663475.79 frames. ], batch size: 92, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:48:49,429 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3278, 1.5088, 1.5215, 1.2332], device='cuda:0'), covar=tensor([0.1989, 0.1808, 0.1254, 0.1625], device='cuda:0'), in_proj_covar=tensor([0.1779, 0.1683, 0.1650, 0.1741], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 20:49:14,400 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:49:24,257 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-572000.pt
+2023-03-06 20:49:27,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.235e+02 1.562e+03 1.883e+03 2.807e+03 8.036e+03, threshold=3.766e+03, percent-clipped=16.0
+2023-03-06 20:49:42,216 INFO [train.py:968] (0/2) Epoch 13, batch 24550, giga_loss[loss=0.3154, simple_loss=0.3906, pruned_loss=0.1201, over 28915.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3802, pruned_loss=0.1245, over 5664026.86 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3642, pruned_loss=0.1134, over 5695049.40 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3812, pruned_loss=0.125, over 5665441.98 frames. ], batch size: 213, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:50:32,468 INFO [train.py:968] (0/2) Epoch 13, batch 24600, giga_loss[loss=0.3169, simple_loss=0.3788, pruned_loss=0.1275, over 28774.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3815, pruned_loss=0.1255, over 5656182.85 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3648, pruned_loss=0.1139, over 5692663.64 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3821, pruned_loss=0.1256, over 5658738.84 frames. ], batch size: 199, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:50:52,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=572082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:51:13,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.268e+02 1.682e+03 2.058e+03 2.959e+03 6.541e+03, threshold=4.115e+03, percent-clipped=11.0
+2023-03-06 20:51:23,553 INFO [train.py:968] (0/2) Epoch 13, batch 24650, giga_loss[loss=0.3337, simple_loss=0.3902, pruned_loss=0.1386, over 27903.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3809, pruned_loss=0.1257, over 5655335.63 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3646, pruned_loss=0.1138, over 5694645.17 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3819, pruned_loss=0.1261, over 5655098.01 frames. ], batch size: 412, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:51:28,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4988, 1.5735, 1.7290, 1.3200], device='cuda:0'), covar=tensor([0.1509, 0.2233, 0.1255, 0.1492], device='cuda:0'), in_proj_covar=tensor([0.0838, 0.0689, 0.0880, 0.0786], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 20:51:34,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6473, 5.4656, 5.1763, 2.4864], device='cuda:0'), covar=tensor([0.0419, 0.0575, 0.0632, 0.1840], device='cuda:0'), in_proj_covar=tensor([0.1100, 0.1025, 0.0891, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-06 20:52:13,425 INFO [train.py:968] (0/2) Epoch 13, batch 24700, giga_loss[loss=0.2644, simple_loss=0.343, pruned_loss=0.09296, over 28894.00 frames. ], tot_loss[loss=0.315, simple_loss=0.379, pruned_loss=0.1255, over 5647203.43 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3642, pruned_loss=0.1137, over 5698076.72 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3804, pruned_loss=0.1262, over 5643240.01 frames. ], batch size: 174, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:52:51,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.766e+03 2.279e+03 3.084e+03 5.382e+03, threshold=4.557e+03, percent-clipped=9.0
+2023-03-06 20:53:00,461 INFO [train.py:968] (0/2) Epoch 13, batch 24750, giga_loss[loss=0.3063, simple_loss=0.3686, pruned_loss=0.122, over 28671.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.377, pruned_loss=0.1253, over 5663040.89 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3644, pruned_loss=0.114, over 5701407.77 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3782, pruned_loss=0.1257, over 5656122.53 frames. ], batch size: 262, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:53:10,262 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=572225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:53:13,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=572228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:53:17,454 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 20:53:23,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=572239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:53:24,263 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-06 20:53:41,883 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=572257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:53:48,132 INFO [train.py:968] (0/2) Epoch 13, batch 24800, giga_loss[loss=0.3361, simple_loss=0.3946, pruned_loss=0.1388, over 28963.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3757, pruned_loss=0.1244, over 5665779.43 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3645, pruned_loss=0.114, over 5703640.97 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3767, pruned_loss=0.125, over 5657822.69 frames. ], batch size: 213, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 20:53:53,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=572271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:54:19,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.905e+02 1.552e+03 2.054e+03 2.511e+03 6.385e+03, threshold=4.107e+03, percent-clipped=2.0
+2023-03-06 20:54:28,688 INFO [train.py:968] (0/2) Epoch 13, batch 24850, giga_loss[loss=0.3085, simple_loss=0.3779, pruned_loss=0.1196, over 28725.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3757, pruned_loss=0.1232, over 5683436.62 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3648, pruned_loss=0.1144, over 5708508.90 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3765, pruned_loss=0.1236, over 5671963.21 frames. ], batch size: 243, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:55:17,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=572363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:55:18,462 INFO [train.py:968] (0/2) Epoch 13, batch 24900, giga_loss[loss=0.3098, simple_loss=0.379, pruned_loss=0.1203, over 29022.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3759, pruned_loss=0.1235, over 5673670.65 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3648, pruned_loss=0.1146, over 5713410.94 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3768, pruned_loss=0.1238, over 5659416.32 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:55:36,354 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=572382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:55:38,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=572385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:55:55,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.974e+02 1.646e+03 2.328e+03 3.529e+03 1.163e+04, threshold=4.657e+03, percent-clipped=22.0
+2023-03-06 20:55:56,506 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1475, 1.0625, 3.9400, 3.2963], device='cuda:0'), covar=tensor([0.1769, 0.2846, 0.0447, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0686, 0.0606, 0.0883, 0.0802], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 20:56:05,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=572414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:56:05,243 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=572414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:56:05,573 INFO [train.py:968] (0/2) Epoch 13, batch 24950, giga_loss[loss=0.3391, simple_loss=0.3933, pruned_loss=0.1425, over 27977.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3747, pruned_loss=0.1219, over 5684148.87 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3649, pruned_loss=0.1148, over 5717000.25 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3757, pruned_loss=0.1222, over 5668381.35 frames. ], batch size: 412, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:56:07,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=572417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:56:35,649 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=572446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:56:50,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7365, 2.2216, 1.8760, 1.4631], device='cuda:0'), covar=tensor([0.2967, 0.1998, 0.1986, 0.2369], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1681, 0.1644, 0.1744], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 20:56:52,190 INFO [train.py:968] (0/2) Epoch 13, batch 25000, giga_loss[loss=0.2774, simple_loss=0.3487, pruned_loss=0.103, over 28378.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3739, pruned_loss=0.1218, over 5689846.37 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3652, pruned_loss=0.1152, over 5718940.48 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3746, pruned_loss=0.1218, over 5674815.11 frames. ], batch size: 65, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:57:33,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.630e+02 1.521e+03 1.977e+03 2.541e+03 6.746e+03, threshold=3.954e+03, percent-clipped=4.0
+2023-03-06 20:57:36,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=572506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:57:39,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=572509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:57:44,168 INFO [train.py:968] (0/2) Epoch 13, batch 25050, giga_loss[loss=0.3157, simple_loss=0.3723, pruned_loss=0.1295, over 28208.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3729, pruned_loss=0.1218, over 5687436.96 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3652, pruned_loss=0.1152, over 5720831.51 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3736, pruned_loss=0.1219, over 5673506.46 frames. ], batch size: 77, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:57:57,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6018, 1.7014, 1.3494, 1.6580], device='cuda:0'), covar=tensor([0.0664, 0.0284, 0.0316, 0.0758], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-06 20:58:05,914 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=572538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:58:31,797 INFO [train.py:968] (0/2) Epoch 13, batch 25100, giga_loss[loss=0.3033, simple_loss=0.366, pruned_loss=0.1203, over 28991.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3712, pruned_loss=0.1211, over 5697138.29 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3652, pruned_loss=0.1153, over 5722570.46 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3719, pruned_loss=0.1212, over 5683736.68 frames. ], batch size: 155, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:58:41,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4635, 2.9789, 2.2942, 1.8461], device='cuda:0'), covar=tensor([0.2184, 0.1338, 0.1672, 0.2133], device='cuda:0'), in_proj_covar=tensor([0.1783, 0.1683, 0.1653, 0.1750], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 20:58:51,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9578, 1.0475, 3.3175, 3.0106], device='cuda:0'), covar=tensor([0.1682, 0.2659, 0.0494, 0.0962], device='cuda:0'), in_proj_covar=tensor([0.0688, 0.0607, 0.0884, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 20:59:07,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.705e+03 2.178e+03 2.755e+03 6.148e+03, threshold=4.357e+03, percent-clipped=5.0
+2023-03-06 20:59:15,816 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=572613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:59:17,650 INFO [train.py:968] (0/2) Epoch 13, batch 25150, giga_loss[loss=0.3841, simple_loss=0.4166, pruned_loss=0.1758, over 26697.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3709, pruned_loss=0.1215, over 5694719.76 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3659, pruned_loss=0.1159, over 5716530.54 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.371, pruned_loss=0.1212, over 5688345.98 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:00:05,982 INFO [train.py:968] (0/2) Epoch 13, batch 25200, giga_loss[loss=0.2712, simple_loss=0.3505, pruned_loss=0.096, over 29024.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3693, pruned_loss=0.1211, over 5690534.08 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3662, pruned_loss=0.1161, over 5719689.21 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3691, pruned_loss=0.1208, over 5682330.45 frames. ], batch size: 155, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 21:00:08,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=572668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:00:44,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.790e+02 1.731e+03 2.350e+03 3.358e+03 6.962e+03, threshold=4.700e+03, percent-clipped=9.0
+2023-03-06 21:00:55,710 INFO [train.py:968] (0/2) Epoch 13, batch 25250, libri_loss[loss=0.3702, simple_loss=0.4181, pruned_loss=0.1612, over 29266.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3693, pruned_loss=0.1218, over 5687409.14 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3665, pruned_loss=0.1164, over 5723848.25 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3689, pruned_loss=0.1213, over 5676715.57 frames. ], batch size: 94, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:01:42,508 INFO [train.py:968] (0/2) Epoch 13, batch 25300, giga_loss[loss=0.2811, simple_loss=0.3577, pruned_loss=0.1022, over 28436.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3706, pruned_loss=0.1222, over 5677607.02 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3667, pruned_loss=0.1167, over 5709168.48 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3702, pruned_loss=0.1217, over 5679996.36 frames. ], batch size: 65, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:02:11,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7329, 1.9460, 1.2352, 1.5754], device='cuda:0'), covar=tensor([0.0767, 0.0523, 0.1035, 0.1001], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0443, 0.0503, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 21:02:16,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.568e+03 2.065e+03 2.721e+03 6.005e+03, threshold=4.130e+03, percent-clipped=3.0
+2023-03-06 21:02:19,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=572807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:02:24,991 INFO [train.py:968] (0/2) Epoch 13, batch 25350, giga_loss[loss=0.3072, simple_loss=0.3811, pruned_loss=0.1167, over 28879.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3709, pruned_loss=0.1211, over 5677037.58 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3671, pruned_loss=0.117, over 5704983.73 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3703, pruned_loss=0.1206, over 5681024.51 frames. ], batch size: 199, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:02:27,749 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-06 21:03:13,279 INFO [train.py:968] (0/2) Epoch 13, batch 25400, giga_loss[loss=0.3189, simple_loss=0.3914, pruned_loss=0.1232, over 28598.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3712, pruned_loss=0.1208, over 5673524.74 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3674, pruned_loss=0.1174, over 5698339.20 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3705, pruned_loss=0.1201, over 5681306.71 frames. ], batch size: 242, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:03:35,784 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-06 21:03:50,576 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.540e+02 1.769e+03 2.500e+03 3.692e+03 1.167e+04, threshold=5.000e+03, percent-clipped=20.0
+2023-03-06 21:04:00,387 INFO [train.py:968] (0/2) Epoch 13, batch 25450, giga_loss[loss=0.3927, simple_loss=0.4342, pruned_loss=0.1756, over 27582.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3718, pruned_loss=0.1218, over 5664244.41 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3676, pruned_loss=0.1176, over 5690927.01 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3711, pruned_loss=0.1211, over 5676246.04 frames. ], batch size: 472, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:04:00,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3875, 3.2641, 1.4881, 1.4623], device='cuda:0'), covar=tensor([0.0915, 0.0330, 0.0845, 0.1289], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0524, 0.0351, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-06 21:04:08,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=572923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:04:13,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3803, 1.5269, 1.4840, 1.3165], device='cuda:0'), covar=tensor([0.1383, 0.1665, 0.1908, 0.1736], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0741, 0.0684, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 21:04:48,148 INFO [train.py:968] (0/2) Epoch 13, batch 25500, giga_loss[loss=0.3261, simple_loss=0.3833, pruned_loss=0.1345, over 28731.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3734, pruned_loss=0.1232, over 5672590.86 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3677, pruned_loss=0.1178, over 5695426.38 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3729, pruned_loss=0.1226, over 5677545.59 frames. ], batch size: 262, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:05:10,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=572988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:05:29,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.845e+02 1.678e+03 2.102e+03 3.087e+03 1.042e+04, threshold=4.204e+03, percent-clipped=7.0
+2023-03-06 21:05:37,653 INFO [train.py:968] (0/2) Epoch 13, batch 25550, giga_loss[loss=0.277, simple_loss=0.3434, pruned_loss=0.1053, over 28429.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3753, pruned_loss=0.1261, over 5675316.96 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3681, pruned_loss=0.1181, over 5699542.97 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3747, pruned_loss=0.1255, over 5675273.94 frames. ], batch size: 65, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:05:44,066 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-06 21:06:05,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=573043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:06:26,395 INFO [train.py:968] (0/2) Epoch 13, batch 25600, giga_loss[loss=0.299, simple_loss=0.3693, pruned_loss=0.1143, over 28894.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.376, pruned_loss=0.1278, over 5675822.73 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3684, pruned_loss=0.1184, over 5701405.34 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3755, pruned_loss=0.1273, over 5673411.55 frames. ], batch size: 174, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:07:05,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.234e+03 1.848e+03 2.360e+03 3.610e+03 9.717e+03, threshold=4.720e+03, percent-clipped=15.0
+2023-03-06 21:07:08,314 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:07:11,302 INFO [train.py:968] (0/2) Epoch 13, batch 25650, giga_loss[loss=0.2843, simple_loss=0.3502, pruned_loss=0.1092, over 28723.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3767, pruned_loss=0.1286, over 5688518.21 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3683, pruned_loss=0.1185, over 5706274.23 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3765, pruned_loss=0.1284, over 5681418.05 frames. ], batch size: 99, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:07:13,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9479, 1.9306, 1.7439, 1.7160], device='cuda:0'), covar=tensor([0.1521, 0.2426, 0.2054, 0.2107], device='cuda:0'), in_proj_covar=tensor([0.0442, 0.0739, 0.0683, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 21:07:26,973 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:07:30,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:07:31,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5860, 1.6672, 1.9165, 1.4440], device='cuda:0'), covar=tensor([0.1501, 0.1975, 0.1227, 0.1607], device='cuda:0'), in_proj_covar=tensor([0.0840, 0.0689, 0.0879, 0.0784], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 21:07:33,387 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-06 21:07:40,220 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.00 vs. limit=2.0
+2023-03-06 21:07:45,612 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-06 21:07:55,244 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=573163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:07:56,421 INFO [train.py:968] (0/2) Epoch 13, batch 25700, libri_loss[loss=0.3131, simple_loss=0.3773, pruned_loss=0.1244, over 26268.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3757, pruned_loss=0.1284, over 5669356.20 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3687, pruned_loss=0.1188, over 5703944.80 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3755, pruned_loss=0.1282, over 5665042.20 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:08:12,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=573182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:08:17,648 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:08:19,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:08:31,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.400e+02 1.508e+03 2.040e+03 2.884e+03 9.471e+03, threshold=4.080e+03, percent-clipped=7.0
+2023-03-06 21:08:38,810 INFO [train.py:968] (0/2) Epoch 13, batch 25750, giga_loss[loss=0.3161, simple_loss=0.3919, pruned_loss=0.1201, over 29095.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3752, pruned_loss=0.1266, over 5674811.54 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3689, pruned_loss=0.119, over 5699906.44 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3751, pruned_loss=0.1266, over 5673408.14 frames. ], batch size: 164, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:08:39,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6761, 1.8878, 1.7674, 1.6832], device='cuda:0'), covar=tensor([0.1588, 0.1963, 0.1951, 0.1986], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0740, 0.0684, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 21:08:41,115 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=573218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:09:03,618 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6779, 1.7923, 1.1486, 1.3497], device='cuda:0'), covar=tensor([0.0788, 0.0543, 0.1056, 0.1044], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0441, 0.0502, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 21:09:23,643 INFO [train.py:968] (0/2) Epoch 13, batch 25800, giga_loss[loss=0.303, simple_loss=0.3723, pruned_loss=0.1168, over 28891.00 frames. ], tot_loss[loss=0.313, simple_loss=0.375, pruned_loss=0.1255, over 5672638.78 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3693, pruned_loss=0.1193, over 5702825.29 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3747, pruned_loss=0.1255, over 5668124.31 frames. ], batch size: 174, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:09:53,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=573298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:10:00,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.140e+03 1.702e+03 2.223e+03 2.946e+03 5.803e+03, threshold=4.445e+03, percent-clipped=6.0
+2023-03-06 21:10:06,761 INFO [train.py:968] (0/2) Epoch 13, batch 25850, libri_loss[loss=0.2553, simple_loss=0.3299, pruned_loss=0.09033, over 29650.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.374, pruned_loss=0.1249, over 5676775.31 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3694, pruned_loss=0.1192, over 5710330.02 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3739, pruned_loss=0.1251, over 5664679.81 frames. ], batch size: 73, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:10:17,802 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:10:20,643 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:10:45,602 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=573357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:10:52,178 INFO [train.py:968] (0/2) Epoch 13, batch 25900, giga_loss[loss=0.2749, simple_loss=0.3452, pruned_loss=0.1023, over 28937.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3728, pruned_loss=0.1246, over 5671699.32 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3701, pruned_loss=0.1197, over 5703262.38 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3722, pruned_loss=0.1245, over 5668205.91 frames. ], batch size: 164, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:11:32,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.593e+02 1.738e+03 2.239e+03 3.141e+03 7.912e+03, threshold=4.479e+03, percent-clipped=7.0
+2023-03-06 21:11:35,057 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-06 21:11:40,293 INFO [train.py:968] (0/2) Epoch 13, batch 25950, giga_loss[loss=0.2611, simple_loss=0.3424, pruned_loss=0.08992, over 28975.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3742, pruned_loss=0.1269, over 5653438.23 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3701, pruned_loss=0.1199, over 5702216.26 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3738, pruned_loss=0.1269, over 5650036.01 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:11:58,463 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:12:06,148 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:12:08,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:12:28,369 INFO [train.py:968] (0/2) Epoch 13, batch 26000, giga_loss[loss=0.2899, simple_loss=0.3698, pruned_loss=0.105, over 28814.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3761, pruned_loss=0.1284, over 5660665.88 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3703, pruned_loss=0.1202, over 5707196.89 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3758, pruned_loss=0.1282, over 5652254.64 frames. ], batch size: 119, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:12:35,243 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=573473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:12:46,147 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=573485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:13:02,495 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:13:04,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.584e+03 2.187e+03 3.580e+03 1.203e+04, threshold=4.373e+03, percent-clipped=18.0
+2023-03-06 21:13:10,327 INFO [train.py:968] (0/2) Epoch 13, batch 26050, libri_loss[loss=0.4087, simple_loss=0.4359, pruned_loss=0.1907, over 19580.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.38, pruned_loss=0.1288, over 5664139.04 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3707, pruned_loss=0.1208, over 5704268.90 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3796, pruned_loss=0.1285, over 5658987.56 frames. ], batch size: 187, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:13:37,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5109, 1.6953, 1.4543, 1.6721], device='cuda:0'), covar=tensor([0.2539, 0.2599, 0.2849, 0.2269], device='cuda:0'), in_proj_covar=tensor([0.1339, 0.0987, 0.1179, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-06 21:13:56,428 INFO [train.py:968] (0/2) Epoch 13, batch 26100, giga_loss[loss=0.3817, simple_loss=0.4263, pruned_loss=0.1685, over 28893.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3815, pruned_loss=0.1278, over 5668525.05 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3705, pruned_loss=0.1208, over 5708040.75 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3815, pruned_loss=0.1277, over 5660159.27 frames. ], batch size: 106, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:14:35,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.119e+02 1.393e+03 1.839e+03 2.630e+03 5.403e+03, threshold=3.679e+03, percent-clipped=4.0
+2023-03-06 21:14:43,740 INFO [train.py:968] (0/2) Epoch 13, batch 26150, giga_loss[loss=0.326, simple_loss=0.3835, pruned_loss=0.1342, over 27912.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3812, pruned_loss=0.1279, over 5666986.05 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3699, pruned_loss=0.1204, over 5711828.22 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3821, pruned_loss=0.1283, over 5656014.56 frames. ], batch size: 412, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:14:57,607 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:15:01,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:15:19,436 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:15:28,643 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=573660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:15:28,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2383, 1.2113, 1.0603, 0.8718], device='cuda:0'), covar=tensor([0.0754, 0.0499, 0.1000, 0.0993], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0442, 0.0501, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 21:15:31,534 INFO [train.py:968] (0/2) Epoch 13, batch 26200, giga_loss[loss=0.3012, simple_loss=0.3663, pruned_loss=0.118, over 28862.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3828, pruned_loss=0.1295, over 5661569.88 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3697, pruned_loss=0.1202, over 5713417.70 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3837, pruned_loss=0.13, over 5651132.42 frames. ], batch size: 119, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:16:11,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.962e+02 1.523e+03 2.025e+03 2.968e+03 8.183e+03, threshold=4.051e+03, percent-clipped=10.0
+2023-03-06 21:16:20,716 INFO [train.py:968] (0/2) Epoch 13, batch 26250, giga_loss[loss=0.379, simple_loss=0.4205, pruned_loss=0.1688, over 28349.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3824, pruned_loss=0.1303, over 5655792.42 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3696, pruned_loss=0.1202, over 5717593.57 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3835, pruned_loss=0.1309, over 5642910.90 frames. ], batch size: 368, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:16:28,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5349, 3.3162, 1.5734, 1.6947], device='cuda:0'), covar=tensor([0.0901, 0.0338, 0.0836, 0.1213], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0521, 0.0349, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 21:17:07,558 INFO [train.py:968] (0/2) Epoch 13, batch 26300, giga_loss[loss=0.2563, simple_loss=0.3269, pruned_loss=0.09286, over 28609.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3807, pruned_loss=0.1298, over 5651362.33 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3695, pruned_loss=0.1201, over 5717327.87 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3819, pruned_loss=0.1305, over 5640400.89 frames. ], batch size: 78, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:17:15,560 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:17:47,402 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=573806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:17:47,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.361e+03 1.825e+03 3.122e+03 1.703e+04, threshold=3.650e+03, percent-clipped=13.0
+2023-03-06 21:17:50,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1522, 1.2465, 3.8720, 3.0617], device='cuda:0'), covar=tensor([0.1729, 0.2680, 0.0424, 0.1031], device='cuda:0'), in_proj_covar=tensor([0.0686, 0.0605, 0.0881, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 21:17:55,017 INFO [train.py:968] (0/2) Epoch 13, batch 26350, giga_loss[loss=0.2811, simple_loss=0.3506, pruned_loss=0.1057, over 29068.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3784, pruned_loss=0.1286, over 5660822.64 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3692, pruned_loss=0.1198, over 5721246.12 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3798, pruned_loss=0.1296, over 5647805.43 frames. ], batch size: 155, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:18:46,428 INFO [train.py:968] (0/2) Epoch 13, batch 26400, giga_loss[loss=0.284, simple_loss=0.3473, pruned_loss=0.1103, over 28750.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3777, pruned_loss=0.1291, over 5655529.08 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3691, pruned_loss=0.1198, over 5723767.61 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3791, pruned_loss=0.1302, over 5641362.42 frames. ], batch size: 99, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 21:18:52,987 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-06 21:18:58,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=573879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:19:11,325 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:19:22,858 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:19:23,521 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:19:23,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.603e+03 2.017e+03 3.185e+03 1.135e+04, threshold=4.034e+03, percent-clipped=17.0
+2023-03-06 21:19:31,504 INFO [train.py:968] (0/2) Epoch 13, batch 26450, giga_loss[loss=0.4158, simple_loss=0.4412, pruned_loss=0.1952, over 26579.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3789, pruned_loss=0.1302, over 5649137.75 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 5719087.65 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3796, pruned_loss=0.1308, over 5639468.12 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:20:01,545 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:20:03,186 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-06 21:20:03,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:20:15,086 INFO [train.py:968] (0/2) Epoch 13, batch 26500, giga_loss[loss=0.3403, simple_loss=0.3997, pruned_loss=0.1405, over 28753.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3777, pruned_loss=0.1293, over 5660085.73 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3698, pruned_loss=0.1204, over 5721967.78 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3785, pruned_loss=0.13, over 5649042.77 frames. ], batch size: 262, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:20:29,769 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=573981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:20:47,424 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3285, 1.5140, 1.2244, 1.4642], device='cuda:0'), covar=tensor([0.0722, 0.0340, 0.0321, 0.0777], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-06 21:20:47,771 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-574000.pt
+2023-03-06 21:20:53,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.658e+03 2.022e+03 2.971e+03 9.759e+03, threshold=4.045e+03, percent-clipped=11.0
+2023-03-06 21:20:59,537 INFO [train.py:968] (0/2) Epoch 13, batch 26550, libri_loss[loss=0.2574, simple_loss=0.322, pruned_loss=0.09643, over 29466.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3746, pruned_loss=0.1273, over 5671510.66 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3699, pruned_loss=0.1204, over 5722357.49 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3754, pruned_loss=0.128, over 5660943.06 frames. ], batch size: 70, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:21:06,196 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:21:08,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:21:10,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:21:22,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1234, 1.0829, 4.0789, 3.1300], device='cuda:0'), covar=tensor([0.1796, 0.2847, 0.0429, 0.1409], device='cuda:0'), in_proj_covar=tensor([0.0685, 0.0602, 0.0882, 0.0801], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 21:21:39,735 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:21:48,018 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-06 21:21:50,256 INFO [train.py:968] (0/2) Epoch 13, batch 26600, giga_loss[loss=0.2864, simple_loss=0.354, pruned_loss=0.1094, over 28724.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3743, pruned_loss=0.127, over 5667577.46 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3703, pruned_loss=0.1207, over 5719750.98 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3746, pruned_loss=0.1274, over 5660921.75 frames. ], batch size: 284, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:22:27,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.268e+02 1.492e+03 2.073e+03 2.853e+03 9.192e+03, threshold=4.147e+03, percent-clipped=9.0
+2023-03-06 21:22:34,576 INFO [train.py:968] (0/2) Epoch 13, batch 26650, giga_loss[loss=0.3034, simple_loss=0.3764, pruned_loss=0.1152, over 28843.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3765, pruned_loss=0.1275, over 5670464.51 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3704, pruned_loss=0.1207, over 5724210.73 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3767, pruned_loss=0.128, over 5659932.57 frames. ], batch size: 199, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:22:41,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4318, 1.8889, 1.3209, 0.7588], device='cuda:0'), covar=tensor([0.3547, 0.1938, 0.2581, 0.4383], device='cuda:0'), in_proj_covar=tensor([0.1613, 0.1534, 0.1510, 0.1322], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-06 21:23:07,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:23:29,193 INFO [train.py:968] (0/2) Epoch 13, batch 26700, giga_loss[loss=0.3553, simple_loss=0.3859, pruned_loss=0.1624, over 23804.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3782, pruned_loss=0.1289, over 5652873.92 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3706, pruned_loss=0.1209, over 5714832.67 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3782, pruned_loss=0.1291, over 5652425.26 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:23:34,073 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:23:35,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:24:04,247 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:24:09,721 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.556e+03 2.193e+03 3.208e+03 2.261e+04, threshold=4.386e+03, percent-clipped=13.0
+2023-03-06 21:24:14,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:24:16,558 INFO [train.py:968] (0/2) Epoch 13, batch 26750, giga_loss[loss=0.2947, simple_loss=0.3788, pruned_loss=0.1053, over 28992.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3779, pruned_loss=0.1288, over 5656476.36 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3706, pruned_loss=0.1209, over 5715796.51 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.378, pruned_loss=0.129, over 5655084.59 frames. ], batch size: 155, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:24:28,249 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:25:00,467 INFO [train.py:968] (0/2) Epoch 13, batch 26800, giga_loss[loss=0.2998, simple_loss=0.3813, pruned_loss=0.1092, over 28481.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3789, pruned_loss=0.1264, over 5672681.63 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3707, pruned_loss=0.121, over 5716553.66 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3791, pruned_loss=0.1267, over 5669641.57 frames. ], batch size: 71, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:25:01,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-06 21:25:01,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:25:15,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:25:15,962 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:25:26,128 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:25:28,818 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:25:31,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4277, 1.7068, 1.2746, 1.8358], device='cuda:0'), covar=tensor([0.2541, 0.2481, 0.2905, 0.2358], device='cuda:0'), in_proj_covar=tensor([0.1349, 0.0991, 0.1189, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 21:25:42,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.970e+02 1.393e+03 1.944e+03 2.558e+03 5.962e+03, threshold=3.889e+03, percent-clipped=4.0
+2023-03-06 21:25:43,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5700, 1.8221, 1.8733, 1.3564], device='cuda:0'), covar=tensor([0.1908, 0.2338, 0.1571, 0.1798], device='cuda:0'), in_proj_covar=tensor([0.0842, 0.0693, 0.0883, 0.0786], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 21:25:47,197 INFO [train.py:968] (0/2) Epoch 13, batch 26850, giga_loss[loss=0.2992, simple_loss=0.3877, pruned_loss=0.1053, over 28858.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3798, pruned_loss=0.1251, over 5670820.03 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3707, pruned_loss=0.1211, over 5718465.17 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3801, pruned_loss=0.1253, over 5666060.64 frames. ], batch size: 145, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:25:56,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:26:03,030 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.56 vs. limit=5.0
+2023-03-06 21:26:34,685 INFO [train.py:968] (0/2) Epoch 13, batch 26900, giga_loss[loss=0.3357, simple_loss=0.3951, pruned_loss=0.1382, over 28610.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3825, pruned_loss=0.1267, over 5673832.14 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3704, pruned_loss=0.121, over 5719277.85 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.383, pruned_loss=0.1271, over 5669153.04 frames. ], batch size: 307, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:26:58,362 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-06 21:27:15,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.806e+02 1.500e+03 2.095e+03 2.688e+03 4.128e+03, threshold=4.189e+03, percent-clipped=1.0
+2023-03-06 21:27:15,659 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:19,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:21,145 INFO [train.py:968] (0/2) Epoch 13, batch 26950, giga_loss[loss=0.312, simple_loss=0.3851, pruned_loss=0.1195, over 28710.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3852, pruned_loss=0.1301, over 5674963.19 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3702, pruned_loss=0.1208, over 5723483.70 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3862, pruned_loss=0.1308, over 5666223.11 frames. ], batch size: 99, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:27:32,712 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:33,423 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:35,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:36,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:47,940 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:28:01,706 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:28:02,253 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:28:11,037 INFO [train.py:968] (0/2) Epoch 13, batch 27000, giga_loss[loss=0.3377, simple_loss=0.3926, pruned_loss=0.1415, over 28716.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3863, pruned_loss=0.1316, over 5677980.03 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.37, pruned_loss=0.1207, over 5717891.87 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3878, pruned_loss=0.1325, over 5675131.93 frames. ], batch size: 242, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:28:11,041 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 21:28:19,820 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2111, simple_loss=0.3169, pruned_loss=0.05268, over 944034.00 frames. 
+2023-03-06 21:28:19,821 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 21:28:41,846 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:29:00,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.729e+03 2.497e+03 3.742e+03 9.346e+03, threshold=4.993e+03, percent-clipped=17.0
+2023-03-06 21:29:08,439 INFO [train.py:968] (0/2) Epoch 13, batch 27050, giga_loss[loss=0.3422, simple_loss=0.3961, pruned_loss=0.1442, over 27526.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3862, pruned_loss=0.1327, over 5656248.11 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3698, pruned_loss=0.1208, over 5703294.03 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3879, pruned_loss=0.1335, over 5665364.24 frames. ], batch size: 472, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:29:54,494 INFO [train.py:968] (0/2) Epoch 13, batch 27100, giga_loss[loss=0.296, simple_loss=0.3693, pruned_loss=0.1114, over 28959.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3835, pruned_loss=0.1299, over 5671792.47 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3697, pruned_loss=0.1208, over 5707301.36 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3853, pruned_loss=0.1308, over 5674559.27 frames. ], batch size: 213, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:30:00,864 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:30:15,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:30:30,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:30:35,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.718e+02 1.515e+03 1.994e+03 2.616e+03 8.804e+03, threshold=3.988e+03, percent-clipped=2.0
+2023-03-06 21:30:40,785 INFO [train.py:968] (0/2) Epoch 13, batch 27150, giga_loss[loss=0.3384, simple_loss=0.394, pruned_loss=0.1414, over 27691.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3831, pruned_loss=0.128, over 5659025.79 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.37, pruned_loss=0.121, over 5701228.86 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3845, pruned_loss=0.1286, over 5666168.53 frames. ], batch size: 472, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:31:27,249 INFO [train.py:968] (0/2) Epoch 13, batch 27200, giga_loss[loss=0.3009, simple_loss=0.3771, pruned_loss=0.1124, over 28982.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3839, pruned_loss=0.1283, over 5662637.15 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5707284.39 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3858, pruned_loss=0.1293, over 5661453.42 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:31:38,580 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:32:12,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.727e+03 2.481e+03 3.349e+03 8.122e+03, threshold=4.962e+03, percent-clipped=20.0
+2023-03-06 21:32:17,081 INFO [train.py:968] (0/2) Epoch 13, batch 27250, giga_loss[loss=0.3154, simple_loss=0.3817, pruned_loss=0.1245, over 28681.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3857, pruned_loss=0.1302, over 5656279.33 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3691, pruned_loss=0.1205, over 5712982.99 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3881, pruned_loss=0.1314, over 5648895.67 frames. ], batch size: 262, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:32:32,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:32:36,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:32:46,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:32:50,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:33:01,364 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:33:03,185 INFO [train.py:968] (0/2) Epoch 13, batch 27300, giga_loss[loss=0.2888, simple_loss=0.3644, pruned_loss=0.1066, over 28964.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3844, pruned_loss=0.1293, over 5668539.94 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3691, pruned_loss=0.1204, over 5718833.31 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3869, pruned_loss=0.1307, over 5655697.51 frames. ], batch size: 164, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:33:15,501 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:33:29,313 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2940, 2.2101, 1.3508, 1.4602], device='cuda:0'), covar=tensor([0.0747, 0.0371, 0.0671, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0523, 0.0350, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 21:33:46,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.452e+02 1.718e+03 2.229e+03 2.637e+03 1.123e+04, threshold=4.459e+03, percent-clipped=7.0
+2023-03-06 21:33:51,951 INFO [train.py:968] (0/2) Epoch 13, batch 27350, libri_loss[loss=0.3194, simple_loss=0.3862, pruned_loss=0.1263, over 29645.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3843, pruned_loss=0.1305, over 5674710.92 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3696, pruned_loss=0.1209, over 5725517.76 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3863, pruned_loss=0.1316, over 5656281.67 frames. ], batch size: 88, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:34:31,400 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:34:37,854 INFO [train.py:968] (0/2) Epoch 13, batch 27400, giga_loss[loss=0.3141, simple_loss=0.3731, pruned_loss=0.1276, over 28932.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.383, pruned_loss=0.1301, over 5672343.03 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3703, pruned_loss=0.1213, over 5719634.61 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3845, pruned_loss=0.1308, over 5662155.32 frames. ], batch size: 227, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:34:41,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:35:26,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.815e+02 1.558e+03 2.018e+03 2.710e+03 8.680e+03, threshold=4.037e+03, percent-clipped=9.0
+2023-03-06 21:35:31,651 INFO [train.py:968] (0/2) Epoch 13, batch 27450, giga_loss[loss=0.3213, simple_loss=0.381, pruned_loss=0.1308, over 28728.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3794, pruned_loss=0.1279, over 5667567.81 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3702, pruned_loss=0.1212, over 5718684.68 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3807, pruned_loss=0.1287, over 5659673.35 frames. ], batch size: 284, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:35:59,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:36:17,913 INFO [train.py:968] (0/2) Epoch 13, batch 27500, giga_loss[loss=0.3094, simple_loss=0.3508, pruned_loss=0.134, over 23273.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3791, pruned_loss=0.1285, over 5658327.08 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3705, pruned_loss=0.1212, over 5711910.23 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3801, pruned_loss=0.1292, over 5656256.86 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:36:37,816 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:36:57,962 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:36:58,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.808e+03 2.370e+03 3.292e+03 9.488e+03, threshold=4.739e+03, percent-clipped=14.0
+2023-03-06 21:36:59,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=575013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:37:02,230 INFO [train.py:968] (0/2) Epoch 13, batch 27550, giga_loss[loss=0.3434, simple_loss=0.3904, pruned_loss=0.1482, over 26557.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3789, pruned_loss=0.1289, over 5653162.88 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3711, pruned_loss=0.1216, over 5706805.30 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3794, pruned_loss=0.1293, over 5654139.72 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:37:25,174 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:37:29,141 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 21:37:34,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:37:43,860 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-06 21:37:46,188 INFO [train.py:968] (0/2) Epoch 13, batch 27600, giga_loss[loss=0.2964, simple_loss=0.3668, pruned_loss=0.113, over 28950.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.376, pruned_loss=0.1254, over 5660804.41 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3709, pruned_loss=0.1214, over 5706441.08 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3766, pruned_loss=0.1261, over 5660815.98 frames. ], batch size: 227, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:37:50,973 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-06 21:38:08,657 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:38:11,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=575092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:38:30,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.831e+02 1.303e+03 1.656e+03 2.124e+03 6.705e+03, threshold=3.312e+03, percent-clipped=4.0
+2023-03-06 21:38:35,286 INFO [train.py:968] (0/2) Epoch 13, batch 27650, giga_loss[loss=0.3614, simple_loss=0.3974, pruned_loss=0.1627, over 26576.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3741, pruned_loss=0.1238, over 5649134.17 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3713, pruned_loss=0.1219, over 5699573.62 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3743, pruned_loss=0.1239, over 5655431.53 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:38:41,190 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:39:02,827 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6368, 1.5489, 1.2176, 1.1967], device='cuda:0'), covar=tensor([0.0691, 0.0530, 0.0916, 0.1238], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0443, 0.0503, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 21:39:28,804 INFO [train.py:968] (0/2) Epoch 13, batch 27700, giga_loss[loss=0.3151, simple_loss=0.3832, pruned_loss=0.1235, over 27806.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.373, pruned_loss=0.1229, over 5652259.40 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3713, pruned_loss=0.1219, over 5700647.76 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3732, pruned_loss=0.123, over 5656131.03 frames. ], batch size: 412, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:39:59,185 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:39:59,736 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=575195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:40:02,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=575197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:40:15,736 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.169e+02 1.583e+03 2.258e+03 2.999e+03 8.811e+03, threshold=4.517e+03, percent-clipped=20.0
+2023-03-06 21:40:21,725 INFO [train.py:968] (0/2) Epoch 13, batch 27750, giga_loss[loss=0.2803, simple_loss=0.3444, pruned_loss=0.1081, over 28810.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3703, pruned_loss=0.1224, over 5651817.83 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3712, pruned_loss=0.1218, over 5705332.92 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3706, pruned_loss=0.1225, over 5649647.48 frames. ], batch size: 284, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:40:35,165 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:40:40,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-06 21:40:42,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:41:12,669 INFO [train.py:968] (0/2) Epoch 13, batch 27800, giga_loss[loss=0.2837, simple_loss=0.3596, pruned_loss=0.1038, over 28940.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3683, pruned_loss=0.1217, over 5648824.30 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3711, pruned_loss=0.1219, over 5704929.25 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3685, pruned_loss=0.1218, over 5646338.73 frames. ], batch size: 213, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:41:53,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.000e+02 1.739e+03 2.353e+03 3.499e+03 1.023e+04, threshold=4.706e+03, percent-clipped=13.0
+2023-03-06 21:41:56,799 INFO [train.py:968] (0/2) Epoch 13, batch 27850, giga_loss[loss=0.2809, simple_loss=0.3612, pruned_loss=0.1003, over 28521.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3698, pruned_loss=0.1217, over 5657181.61 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.371, pruned_loss=0.1219, over 5706840.00 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.37, pruned_loss=0.1218, over 5651831.64 frames. ], batch size: 85, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:42:46,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:42:47,204 INFO [train.py:968] (0/2) Epoch 13, batch 27900, giga_loss[loss=0.2733, simple_loss=0.3531, pruned_loss=0.09674, over 28873.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3711, pruned_loss=0.1218, over 5653820.64 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3711, pruned_loss=0.1219, over 5708940.02 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3711, pruned_loss=0.1218, over 5647221.69 frames. ], batch size: 145, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:42:58,287 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:43:01,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=575379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:43:30,330 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:43:32,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.977e+02 1.392e+03 1.730e+03 2.501e+03 1.038e+04, threshold=3.461e+03, percent-clipped=4.0
+2023-03-06 21:43:35,247 INFO [train.py:968] (0/2) Epoch 13, batch 27950, giga_loss[loss=0.2817, simple_loss=0.3574, pruned_loss=0.103, over 28845.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3708, pruned_loss=0.1218, over 5653419.91 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3712, pruned_loss=0.122, over 5711049.78 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3708, pruned_loss=0.1217, over 5645596.31 frames. ], batch size: 186, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:43:49,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2946, 1.6225, 1.5742, 1.1207], device='cuda:0'), covar=tensor([0.1552, 0.2574, 0.1410, 0.1634], device='cuda:0'), in_proj_covar=tensor([0.0847, 0.0698, 0.0892, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 21:44:23,052 INFO [train.py:968] (0/2) Epoch 13, batch 28000, giga_loss[loss=0.3158, simple_loss=0.38, pruned_loss=0.1259, over 27932.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.372, pruned_loss=0.1234, over 5643726.83 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3717, pruned_loss=0.1222, over 5704855.34 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.1231, over 5641858.95 frames. ], batch size: 412, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 21:44:27,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8431, 1.8078, 1.7504, 1.6361], device='cuda:0'), covar=tensor([0.1481, 0.1986, 0.2119, 0.1894], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0739, 0.0687, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 21:44:47,371 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-06 21:44:49,693 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=575496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:01,058 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:03,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=575510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:06,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.584e+03 2.092e+03 2.898e+03 8.352e+03, threshold=4.184e+03, percent-clipped=13.0
+2023-03-06 21:45:09,054 INFO [train.py:968] (0/2) Epoch 13, batch 28050, giga_loss[loss=0.3072, simple_loss=0.3749, pruned_loss=0.1198, over 29008.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3734, pruned_loss=0.1245, over 5646106.17 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3719, pruned_loss=0.1224, over 5710127.42 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3728, pruned_loss=0.1242, over 5638208.72 frames. ], batch size: 155, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:45:09,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=575516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:12,758 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=575520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:29,882 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:51,661 INFO [train.py:968] (0/2) Epoch 13, batch 28100, giga_loss[loss=0.2987, simple_loss=0.3695, pruned_loss=0.1139, over 28910.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.375, pruned_loss=0.1251, over 5655313.52 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3717, pruned_loss=0.1224, over 5711406.36 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3749, pruned_loss=0.125, over 5645872.67 frames. ], batch size: 186, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:45:57,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:46:24,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8341, 1.9266, 1.3244, 1.5344], device='cuda:0'), covar=tensor([0.0808, 0.0629, 0.1037, 0.1050], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0445, 0.0505, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 21:46:32,907 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-06 21:46:36,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.510e+03 2.112e+03 2.906e+03 8.234e+03, threshold=4.223e+03, percent-clipped=6.0
+2023-03-06 21:46:40,110 INFO [train.py:968] (0/2) Epoch 13, batch 28150, giga_loss[loss=0.2816, simple_loss=0.3578, pruned_loss=0.1027, over 28680.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3779, pruned_loss=0.1278, over 5651393.91 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3724, pruned_loss=0.1228, over 5712162.56 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3773, pruned_loss=0.1274, over 5641562.71 frames. ], batch size: 60, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:47:23,812 INFO [train.py:968] (0/2) Epoch 13, batch 28200, giga_loss[loss=0.3407, simple_loss=0.3798, pruned_loss=0.1508, over 28577.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3797, pruned_loss=0.1294, over 5659028.64 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3727, pruned_loss=0.1231, over 5714693.89 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3792, pruned_loss=0.1291, over 5647289.35 frames. ], batch size: 85, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:48:14,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.395e+02 1.762e+03 2.170e+03 2.816e+03 5.521e+03, threshold=4.339e+03, percent-clipped=3.0
+2023-03-06 21:48:16,192 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:48:18,534 INFO [train.py:968] (0/2) Epoch 13, batch 28250, giga_loss[loss=0.2911, simple_loss=0.3695, pruned_loss=0.1063, over 28028.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3805, pruned_loss=0.1294, over 5648691.47 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3729, pruned_loss=0.1233, over 5708447.91 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.38, pruned_loss=0.129, over 5644521.64 frames. ], batch size: 412, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:48:19,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=575716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:48:49,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:49:12,177 INFO [train.py:968] (0/2) Epoch 13, batch 28300, giga_loss[loss=0.2542, simple_loss=0.3287, pruned_loss=0.08988, over 28998.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3801, pruned_loss=0.1284, over 5649126.20 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3727, pruned_loss=0.1232, over 5709567.67 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3799, pruned_loss=0.1282, over 5644668.37 frames. ], batch size: 120, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:49:59,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.692e+03 2.264e+03 2.870e+03 7.955e+03, threshold=4.527e+03, percent-clipped=9.0
+2023-03-06 21:50:03,109 INFO [train.py:968] (0/2) Epoch 13, batch 28350, giga_loss[loss=0.4219, simple_loss=0.4303, pruned_loss=0.2068, over 23563.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3815, pruned_loss=0.1306, over 5634953.20 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3726, pruned_loss=0.1232, over 5713636.70 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3815, pruned_loss=0.1305, over 5626781.04 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:50:36,959 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=575848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:50:55,682 INFO [train.py:968] (0/2) Epoch 13, batch 28400, giga_loss[loss=0.3233, simple_loss=0.3677, pruned_loss=0.1395, over 28116.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.381, pruned_loss=0.1311, over 5639888.21 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3729, pruned_loss=0.1236, over 5718242.80 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.381, pruned_loss=0.1309, over 5627500.04 frames. ], batch size: 77, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 21:51:03,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:51:04,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-06 21:51:28,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:51:32,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:51:50,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.686e+03 2.186e+03 3.323e+03 8.386e+03, threshold=4.371e+03, percent-clipped=10.0
+2023-03-06 21:51:53,924 INFO [train.py:968] (0/2) Epoch 13, batch 28450, giga_loss[loss=0.3046, simple_loss=0.3515, pruned_loss=0.1289, over 23590.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3786, pruned_loss=0.1304, over 5622433.28 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3726, pruned_loss=0.1234, over 5715365.83 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3792, pruned_loss=0.1307, over 5611391.82 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:52:38,724 INFO [train.py:968] (0/2) Epoch 13, batch 28500, giga_loss[loss=0.3017, simple_loss=0.3613, pruned_loss=0.121, over 28971.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3777, pruned_loss=0.1297, over 5645756.92 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3731, pruned_loss=0.1238, over 5717728.01 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3779, pruned_loss=0.1297, over 5632219.35 frames. ], batch size: 106, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:53:11,105 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-576000.pt
+2023-03-06 21:53:11,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3560, 1.6288, 1.5028, 1.4681], device='cuda:0'), covar=tensor([0.0782, 0.0324, 0.0306, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0115, 0.0115, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-06 21:53:20,494 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:24,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9127, 2.0173, 1.7070, 2.1008], device='cuda:0'), covar=tensor([0.2355, 0.2445, 0.2636, 0.2244], device='cuda:0'), in_proj_covar=tensor([0.1350, 0.0991, 0.1188, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 21:53:24,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.597e+03 2.119e+03 2.762e+03 8.043e+03, threshold=4.239e+03, percent-clipped=5.0
+2023-03-06 21:53:25,280 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:25,651 INFO [train.py:968] (0/2) Epoch 13, batch 28550, giga_loss[loss=0.2981, simple_loss=0.3586, pruned_loss=0.1188, over 28762.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3768, pruned_loss=0.1292, over 5647517.38 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3731, pruned_loss=0.1238, over 5710517.04 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.377, pruned_loss=0.1293, over 5641656.48 frames. ], batch size: 119, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 21:53:28,442 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:44,133 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:47,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:49,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:49,191 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-06 21:53:51,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:57,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:54:13,065 INFO [train.py:968] (0/2) Epoch 13, batch 28600, giga_loss[loss=0.2709, simple_loss=0.3428, pruned_loss=0.09954, over 28851.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3753, pruned_loss=0.1283, over 5642786.59 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3727, pruned_loss=0.1237, over 5704973.81 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3759, pruned_loss=0.1286, over 5641686.30 frames. ], batch size: 199, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 21:54:13,961 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:54:15,872 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-06 21:54:16,934 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:54:36,783 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:54:55,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.487e+03 1.990e+03 3.029e+03 7.685e+03, threshold=3.979e+03, percent-clipped=9.0
+2023-03-06 21:54:55,918 INFO [train.py:968] (0/2) Epoch 13, batch 28650, giga_loss[loss=0.3684, simple_loss=0.4204, pruned_loss=0.1582, over 28632.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3757, pruned_loss=0.1279, over 5658504.20 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3729, pruned_loss=0.1236, over 5711209.21 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3761, pruned_loss=0.1283, over 5650102.47 frames. ], batch size: 307, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 21:55:19,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4010, 2.0520, 1.5446, 0.6163], device='cuda:0'), covar=tensor([0.4117, 0.2109, 0.2909, 0.4656], device='cuda:0'), in_proj_covar=tensor([0.1614, 0.1529, 0.1518, 0.1333], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-06 21:55:39,515 INFO [train.py:968] (0/2) Epoch 13, batch 28700, giga_loss[loss=0.3303, simple_loss=0.382, pruned_loss=0.1393, over 28471.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3784, pruned_loss=0.1303, over 5666391.59 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.373, pruned_loss=0.1238, over 5715142.06 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3788, pruned_loss=0.1307, over 5653925.09 frames. ], batch size: 78, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 21:56:05,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5422, 1.7122, 1.5343, 1.3306], device='cuda:0'), covar=tensor([0.2144, 0.1788, 0.1566, 0.1879], device='cuda:0'), in_proj_covar=tensor([0.1789, 0.1703, 0.1670, 0.1757], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 21:56:22,229 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-06 21:56:26,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.602e+03 2.208e+03 3.011e+03 8.105e+03, threshold=4.416e+03, percent-clipped=14.0
+2023-03-06 21:56:27,219 INFO [train.py:968] (0/2) Epoch 13, batch 28750, giga_loss[loss=0.3475, simple_loss=0.3994, pruned_loss=0.1478, over 28934.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3787, pruned_loss=0.1305, over 5666133.86 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3727, pruned_loss=0.1237, over 5710500.13 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3795, pruned_loss=0.1311, over 5659693.04 frames. ], batch size: 213, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 21:56:36,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=576223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:57:11,569 INFO [train.py:968] (0/2) Epoch 13, batch 28800, giga_loss[loss=0.3224, simple_loss=0.3831, pruned_loss=0.1309, over 28450.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3792, pruned_loss=0.131, over 5668212.76 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.373, pruned_loss=0.1241, over 5707261.56 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3797, pruned_loss=0.1313, over 5665732.89 frames. ], batch size: 65, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 21:57:55,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.189e+03 1.722e+03 2.085e+03 2.824e+03 5.517e+03, threshold=4.169e+03, percent-clipped=6.0
+2023-03-06 21:57:57,860 INFO [train.py:968] (0/2) Epoch 13, batch 28850, giga_loss[loss=0.328, simple_loss=0.3815, pruned_loss=0.1372, over 27676.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3794, pruned_loss=0.1315, over 5654506.30 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3734, pruned_loss=0.1245, over 5693181.75 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3795, pruned_loss=0.1316, over 5663160.15 frames. ], batch size: 472, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 21:58:13,640 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:58:50,382 INFO [train.py:968] (0/2) Epoch 13, batch 28900, giga_loss[loss=0.3112, simple_loss=0.3778, pruned_loss=0.1223, over 28684.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3804, pruned_loss=0.1315, over 5656371.39 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3734, pruned_loss=0.1244, over 5693528.71 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3806, pruned_loss=0.1316, over 5662600.45 frames. ], batch size: 92, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 21:58:51,357 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:58:54,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:59:06,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=576384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:59:20,243 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:59:33,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.784e+02 1.497e+03 2.027e+03 2.768e+03 9.150e+03, threshold=4.054e+03, percent-clipped=10.0
+2023-03-06 21:59:34,734 INFO [train.py:968] (0/2) Epoch 13, batch 28950, giga_loss[loss=0.3195, simple_loss=0.3894, pruned_loss=0.1249, over 28340.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3807, pruned_loss=0.1309, over 5660159.90 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3736, pruned_loss=0.1245, over 5690869.72 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3809, pruned_loss=0.1312, over 5666764.14 frames. ], batch size: 65, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 21:59:40,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0209, 4.8463, 4.5771, 2.2287], device='cuda:0'), covar=tensor([0.0485, 0.0657, 0.0698, 0.1969], device='cuda:0'), in_proj_covar=tensor([0.1117, 0.1038, 0.0905, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-06 21:59:45,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3070, 3.1284, 2.9752, 1.4144], device='cuda:0'), covar=tensor([0.0937, 0.1070, 0.0944, 0.2268], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.1038, 0.0905, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-06 21:59:48,381 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-06 22:00:19,953 INFO [train.py:968] (0/2) Epoch 13, batch 29000, giga_loss[loss=0.2971, simple_loss=0.3646, pruned_loss=0.1148, over 28802.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3814, pruned_loss=0.1319, over 5665546.84 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3733, pruned_loss=0.1244, over 5694012.96 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3819, pruned_loss=0.1323, over 5667536.25 frames. ], batch size: 119, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:00:22,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=576468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:00:33,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9090, 3.6780, 3.5399, 1.8017], device='cuda:0'), covar=tensor([0.0851, 0.1159, 0.1122, 0.2154], device='cuda:0'), in_proj_covar=tensor([0.1115, 0.1038, 0.0904, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-06 22:00:48,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3437, 2.0182, 1.5176, 0.6235], device='cuda:0'), covar=tensor([0.4293, 0.2240, 0.2938, 0.4771], device='cuda:0'), in_proj_covar=tensor([0.1616, 0.1528, 0.1519, 0.1335], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-06 22:00:58,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.776e+02 1.603e+03 1.957e+03 2.485e+03 5.481e+03, threshold=3.914e+03, percent-clipped=3.0
+2023-03-06 22:00:59,417 INFO [train.py:968] (0/2) Epoch 13, batch 29050, giga_loss[loss=0.3318, simple_loss=0.3915, pruned_loss=0.1361, over 28837.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3825, pruned_loss=0.133, over 5669244.40 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.373, pruned_loss=0.1241, over 5702866.60 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3837, pruned_loss=0.1341, over 5661442.57 frames. ], batch size: 186, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:01:08,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4470, 3.3745, 1.5672, 1.5763], device='cuda:0'), covar=tensor([0.0947, 0.0435, 0.0907, 0.1323], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0523, 0.0351, 0.0400], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-06 22:01:10,483 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:01:12,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:01:27,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5608, 4.3795, 4.1541, 2.0964], device='cuda:0'), covar=tensor([0.0567, 0.0777, 0.0744, 0.1942], device='cuda:0'), in_proj_covar=tensor([0.1120, 0.1043, 0.0908, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-06 22:01:37,777 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:01:43,972 INFO [train.py:968] (0/2) Epoch 13, batch 29100, giga_loss[loss=0.3626, simple_loss=0.3865, pruned_loss=0.1694, over 23423.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3815, pruned_loss=0.1323, over 5664568.22 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3732, pruned_loss=0.1241, over 5705604.36 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3826, pruned_loss=0.1333, over 5655106.93 frames. ], batch size: 705, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:02:20,999 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-06 22:02:23,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4134, 1.6576, 1.5940, 1.4253], device='cuda:0'), covar=tensor([0.1590, 0.1661, 0.1978, 0.1899], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0742, 0.0687, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 22:02:28,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:02:30,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.719e+02 1.626e+03 2.300e+03 3.214e+03 7.997e+03, threshold=4.599e+03, percent-clipped=16.0
+2023-03-06 22:02:30,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:02:30,841 INFO [train.py:968] (0/2) Epoch 13, batch 29150, giga_loss[loss=0.3057, simple_loss=0.3741, pruned_loss=0.1186, over 28952.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3802, pruned_loss=0.1302, over 5653355.52 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3725, pruned_loss=0.1237, over 5704980.93 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3818, pruned_loss=0.1315, over 5645379.05 frames. ], batch size: 213, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:03:00,280 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:03:23,064 INFO [train.py:968] (0/2) Epoch 13, batch 29200, giga_loss[loss=0.3204, simple_loss=0.3824, pruned_loss=0.1292, over 28265.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.381, pruned_loss=0.1302, over 5648797.73 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3724, pruned_loss=0.1236, over 5707912.11 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3826, pruned_loss=0.1316, over 5638781.30 frames. ], batch size: 77, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:03:24,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7900, 1.7901, 1.8556, 1.6807], device='cuda:0'), covar=tensor([0.1641, 0.2103, 0.2028, 0.1962], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0746, 0.0690, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 22:03:40,279 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:03:59,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=576707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:04:01,440 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:04:06,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.035e+02 1.475e+03 1.868e+03 2.555e+03 7.879e+03, threshold=3.736e+03, percent-clipped=4.0
+2023-03-06 22:04:06,835 INFO [train.py:968] (0/2) Epoch 13, batch 29250, giga_loss[loss=0.3036, simple_loss=0.3672, pruned_loss=0.12, over 28907.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3795, pruned_loss=0.1285, over 5657409.74 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1236, over 5709904.72 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3809, pruned_loss=0.1296, over 5647034.40 frames. ], batch size: 186, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:04:39,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3205, 1.5527, 1.4823, 1.5826], device='cuda:0'), covar=tensor([0.0776, 0.0308, 0.0300, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-06 22:04:52,698 INFO [train.py:968] (0/2) Epoch 13, batch 29300, giga_loss[loss=0.3896, simple_loss=0.4237, pruned_loss=0.1778, over 26578.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3779, pruned_loss=0.1279, over 5656522.82 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3719, pruned_loss=0.1234, over 5704037.98 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3796, pruned_loss=0.1291, over 5652666.28 frames. ], batch size: 555, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:05:25,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0531, 1.2431, 3.3621, 2.9898], device='cuda:0'), covar=tensor([0.1618, 0.2464, 0.0474, 0.0870], device='cuda:0'), in_proj_covar=tensor([0.0687, 0.0603, 0.0881, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 22:05:31,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5422, 1.8266, 1.4716, 1.7736], device='cuda:0'), covar=tensor([0.2321, 0.2330, 0.2611, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.1345, 0.0989, 0.1188, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 22:05:31,946 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0356, 1.3442, 5.2646, 3.6727], device='cuda:0'), covar=tensor([0.1375, 0.2594, 0.0377, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0687, 0.0603, 0.0881, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 22:05:38,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.639e+02 1.453e+03 1.867e+03 2.535e+03 4.861e+03, threshold=3.735e+03, percent-clipped=7.0
+2023-03-06 22:05:39,011 INFO [train.py:968] (0/2) Epoch 13, batch 29350, giga_loss[loss=0.3318, simple_loss=0.3896, pruned_loss=0.137, over 28590.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3785, pruned_loss=0.1282, over 5654889.71 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3722, pruned_loss=0.1236, over 5707709.85 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3797, pruned_loss=0.1291, over 5647594.15 frames. ], batch size: 78, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:05:43,817 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:06:13,889 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:06:16,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:06:26,541 INFO [train.py:968] (0/2) Epoch 13, batch 29400, giga_loss[loss=0.3116, simple_loss=0.3761, pruned_loss=0.1235, over 28883.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3803, pruned_loss=0.1297, over 5641827.54 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3727, pruned_loss=0.124, over 5688738.57 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.381, pruned_loss=0.1302, over 5650948.27 frames. ], batch size: 174, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:06:36,611 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:06:44,067 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:07:16,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.744e+03 2.530e+03 3.747e+03 1.124e+04, threshold=5.060e+03, percent-clipped=24.0
+2023-03-06 22:07:16,657 INFO [train.py:968] (0/2) Epoch 13, batch 29450, libri_loss[loss=0.3269, simple_loss=0.3914, pruned_loss=0.1312, over 27811.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3795, pruned_loss=0.1299, over 5647126.09 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3728, pruned_loss=0.124, over 5689978.73 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.38, pruned_loss=0.1303, over 5652806.59 frames. ], batch size: 116, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:07:39,080 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 22:07:56,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1340, 2.5949, 2.2365, 1.7644], device='cuda:0'), covar=tensor([0.2309, 0.1514, 0.1656, 0.2190], device='cuda:0'), in_proj_covar=tensor([0.1769, 0.1679, 0.1640, 0.1735], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 22:08:03,216 INFO [train.py:968] (0/2) Epoch 13, batch 29500, giga_loss[loss=0.3361, simple_loss=0.401, pruned_loss=0.1356, over 28608.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3799, pruned_loss=0.1302, over 5662276.33 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3729, pruned_loss=0.124, over 5693522.48 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3804, pruned_loss=0.1308, over 5662706.72 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:08:23,482 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:08:51,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.457e+02 1.595e+03 2.081e+03 2.662e+03 8.920e+03, threshold=4.161e+03, percent-clipped=3.0
+2023-03-06 22:08:51,533 INFO [train.py:968] (0/2) Epoch 13, batch 29550, giga_loss[loss=0.2974, simple_loss=0.3708, pruned_loss=0.1121, over 28807.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3817, pruned_loss=0.1317, over 5658135.95 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.373, pruned_loss=0.1239, over 5695396.18 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3821, pruned_loss=0.1323, over 5656592.69 frames. ], batch size: 174, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:09:34,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:09:39,734 INFO [train.py:968] (0/2) Epoch 13, batch 29600, giga_loss[loss=0.3554, simple_loss=0.4093, pruned_loss=0.1507, over 29058.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3824, pruned_loss=0.1326, over 5647752.30 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3733, pruned_loss=0.124, over 5698469.05 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3826, pruned_loss=0.1331, over 5642697.72 frames. ], batch size: 155, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:09:56,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:10:08,967 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=577099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:10:24,046 INFO [train.py:968] (0/2) Epoch 13, batch 29650, giga_loss[loss=0.3331, simple_loss=0.3958, pruned_loss=0.1352, over 29035.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3823, pruned_loss=0.1322, over 5654327.58 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3735, pruned_loss=0.124, over 5701061.68 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3826, pruned_loss=0.1328, over 5646881.52 frames. ], batch size: 155, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:10:25,549 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.638e+02 1.655e+03 2.235e+03 2.991e+03 8.786e+03, threshold=4.470e+03, percent-clipped=7.0
+2023-03-06 22:10:29,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0016, 1.2201, 0.9513, 0.2391], device='cuda:0'), covar=tensor([0.2022, 0.1578, 0.2200, 0.4006], device='cuda:0'), in_proj_covar=tensor([0.1598, 0.1515, 0.1505, 0.1320], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-06 22:10:59,476 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=577150.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:11:12,356 INFO [train.py:968] (0/2) Epoch 13, batch 29700, libri_loss[loss=0.3349, simple_loss=0.3952, pruned_loss=0.1373, over 29287.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3824, pruned_loss=0.132, over 5643547.92 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3736, pruned_loss=0.1241, over 5686823.35 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3827, pruned_loss=0.1325, over 5649645.19 frames. ], batch size: 97, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:11:44,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:11:51,880 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:11:54,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:02,104 INFO [train.py:968] (0/2) Epoch 13, batch 29750, giga_loss[loss=0.3078, simple_loss=0.3744, pruned_loss=0.1206, over 28679.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3814, pruned_loss=0.1304, over 5645488.70 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.374, pruned_loss=0.1244, over 5685540.41 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3813, pruned_loss=0.1307, over 5650676.61 frames. ], batch size: 242, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:12:04,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.124e+02 1.544e+03 2.048e+03 3.014e+03 7.688e+03, threshold=4.096e+03, percent-clipped=10.0
+2023-03-06 22:12:09,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=577220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:11,474 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=577222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:16,388 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=577227.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:12:18,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:20,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:24,547 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:38,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:47,003 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:51,622 INFO [train.py:968] (0/2) Epoch 13, batch 29800, giga_loss[loss=0.2829, simple_loss=0.359, pruned_loss=0.1034, over 28765.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3797, pruned_loss=0.1294, over 5651302.83 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.374, pruned_loss=0.1244, over 5687781.99 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3798, pruned_loss=0.1297, over 5652476.07 frames. ], batch size: 119, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:12:56,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2950, 3.0835, 1.3782, 1.4672], device='cuda:0'), covar=tensor([0.1006, 0.0351, 0.0905, 0.1377], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0522, 0.0352, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-06 22:13:34,952 INFO [train.py:968] (0/2) Epoch 13, batch 29850, libri_loss[loss=0.2338, simple_loss=0.3064, pruned_loss=0.08065, over 29386.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3768, pruned_loss=0.1271, over 5658378.20 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3727, pruned_loss=0.1235, over 5686224.78 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3783, pruned_loss=0.1284, over 5658804.21 frames. ], batch size: 67, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:13:35,550 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.151e+03 1.699e+03 2.185e+03 2.858e+03 8.540e+03, threshold=4.370e+03, percent-clipped=10.0
+2023-03-06 22:13:43,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4326, 3.3646, 1.6254, 1.5087], device='cuda:0'), covar=tensor([0.0934, 0.0315, 0.0860, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0523, 0.0352, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-06 22:13:55,588 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:14:00,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:14:16,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:14:19,739 INFO [train.py:968] (0/2) Epoch 13, batch 29900, giga_loss[loss=0.2697, simple_loss=0.3377, pruned_loss=0.1008, over 28840.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3735, pruned_loss=0.1255, over 5657808.92 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3728, pruned_loss=0.1237, over 5681050.06 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3747, pruned_loss=0.1264, over 5661372.28 frames. ], batch size: 199, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:14:25,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:14:26,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2913, 1.2118, 1.1575, 1.4588], device='cuda:0'), covar=tensor([0.0768, 0.0352, 0.0342, 0.0824], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-06 22:14:49,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7971, 2.1683, 1.9515, 1.5546], device='cuda:0'), covar=tensor([0.2467, 0.1950, 0.1932, 0.2169], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1691, 0.1659, 0.1757], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 22:14:49,427 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:14:52,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:15:06,710 INFO [train.py:968] (0/2) Epoch 13, batch 29950, giga_loss[loss=0.2769, simple_loss=0.3321, pruned_loss=0.1108, over 28690.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3705, pruned_loss=0.1244, over 5638791.60 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3733, pruned_loss=0.1241, over 5668695.10 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.371, pruned_loss=0.1249, over 5651138.40 frames. ], batch size: 92, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 22:15:07,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6037, 1.7441, 1.2395, 1.3506], device='cuda:0'), covar=tensor([0.0777, 0.0579, 0.0908, 0.1285], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0443, 0.0503, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 22:15:08,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.708e+03 2.409e+03 3.334e+03 9.815e+03, threshold=4.818e+03, percent-clipped=14.0
+2023-03-06 22:15:19,000 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:15:53,636 INFO [train.py:968] (0/2) Epoch 13, batch 30000, giga_loss[loss=0.284, simple_loss=0.3531, pruned_loss=0.1074, over 28726.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3694, pruned_loss=0.1245, over 5645116.96 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3734, pruned_loss=0.1242, over 5663034.61 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3697, pruned_loss=0.1247, over 5659132.40 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:15:53,640 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 22:15:58,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3720, 1.7708, 1.3559, 1.3822], device='cuda:0'), covar=tensor([0.3012, 0.2646, 0.3148, 0.2337], device='cuda:0'), in_proj_covar=tensor([0.1346, 0.0986, 0.1184, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 22:16:02,156 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2139, simple_loss=0.3215, pruned_loss=0.05312, over 944034.00 frames. 
+2023-03-06 22:16:02,157 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 22:16:08,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:16:38,130 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:16:42,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:16:48,160 INFO [train.py:968] (0/2) Epoch 13, batch 30050, giga_loss[loss=0.3091, simple_loss=0.3678, pruned_loss=0.1252, over 28780.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3681, pruned_loss=0.1242, over 5627518.19 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3732, pruned_loss=0.1239, over 5663477.81 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3684, pruned_loss=0.1247, over 5637640.15 frames. ], batch size: 119, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:16:49,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.118e+02 1.650e+03 1.988e+03 2.659e+03 6.650e+03, threshold=3.977e+03, percent-clipped=4.0
+2023-03-06 22:16:57,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577525.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:17:09,248 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:17:32,070 INFO [train.py:968] (0/2) Epoch 13, batch 30100, giga_loss[loss=0.3276, simple_loss=0.3878, pruned_loss=0.1337, over 28608.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3673, pruned_loss=0.122, over 5640882.20 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3729, pruned_loss=0.1236, over 5667345.34 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3676, pruned_loss=0.1226, over 5644444.28 frames. ], batch size: 242, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:17:51,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8786, 1.2785, 1.2671, 1.0576], device='cuda:0'), covar=tensor([0.1513, 0.1078, 0.1782, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0742, 0.0686, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 22:18:05,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:18:07,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:18:13,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577602.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:18:27,078 INFO [train.py:968] (0/2) Epoch 13, batch 30150, giga_loss[loss=0.3081, simple_loss=0.3831, pruned_loss=0.1165, over 28294.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3652, pruned_loss=0.1187, over 5634463.36 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3727, pruned_loss=0.1236, over 5670471.40 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3655, pruned_loss=0.1191, over 5633830.61 frames. ], batch size: 368, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:18:28,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.499e+03 2.061e+03 2.802e+03 9.526e+03, threshold=4.123e+03, percent-clipped=14.0
+2023-03-06 22:18:30,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:18:33,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:18:51,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5178, 1.4753, 1.6978, 1.2099], device='cuda:0'), covar=tensor([0.1961, 0.3242, 0.1601, 0.1752], device='cuda:0'), in_proj_covar=tensor([0.0848, 0.0698, 0.0889, 0.0794], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 22:19:04,152 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:19:17,191 INFO [train.py:968] (0/2) Epoch 13, batch 30200, giga_loss[loss=0.2584, simple_loss=0.3361, pruned_loss=0.09042, over 28881.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3632, pruned_loss=0.1157, over 5648121.83 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3731, pruned_loss=0.1241, over 5672423.33 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3629, pruned_loss=0.1154, over 5645183.67 frames. ], batch size: 199, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:19:21,559 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577668.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:19:25,347 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577671.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:19:53,201 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577700.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:20:04,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4714, 1.3070, 4.6648, 3.5097], device='cuda:0'), covar=tensor([0.1679, 0.2730, 0.0376, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0689, 0.0608, 0.0891, 0.0807], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 22:20:04,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9553, 2.7932, 1.7686, 0.9167], device='cuda:0'), covar=tensor([0.5659, 0.2868, 0.3364, 0.5375], device='cuda:0'), in_proj_covar=tensor([0.1620, 0.1541, 0.1521, 0.1338], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-06 22:20:09,306 INFO [train.py:968] (0/2) Epoch 13, batch 30250, giga_loss[loss=0.3008, simple_loss=0.3658, pruned_loss=0.1179, over 28897.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3604, pruned_loss=0.1129, over 5649974.89 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3726, pruned_loss=0.124, over 5676877.46 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3604, pruned_loss=0.1126, over 5643431.64 frames. ], batch size: 213, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:20:11,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.202e+02 1.704e+03 2.007e+03 3.256e+03 7.280e+03, threshold=4.014e+03, percent-clipped=10.0
+2023-03-06 22:20:31,769 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:20:33,182 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:20:35,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:20:37,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:20:38,521 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577745.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:20:41,245 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577748.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:20:48,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4806, 3.8123, 1.6312, 1.6493], device='cuda:0'), covar=tensor([0.0975, 0.0255, 0.0946, 0.1365], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0521, 0.0350, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-06 22:20:58,293 INFO [train.py:968] (0/2) Epoch 13, batch 30300, giga_loss[loss=0.2475, simple_loss=0.3349, pruned_loss=0.08002, over 28655.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3566, pruned_loss=0.1094, over 5655115.01 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3718, pruned_loss=0.1237, over 5681734.29 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3571, pruned_loss=0.1092, over 5645093.18 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:21:02,130 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:21:04,180 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:21:08,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577777.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:21:46,301 INFO [train.py:968] (0/2) Epoch 13, batch 30350, giga_loss[loss=0.2691, simple_loss=0.3544, pruned_loss=0.09197, over 28680.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3558, pruned_loss=0.1062, over 5675834.08 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3712, pruned_loss=0.1235, over 5686822.05 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3563, pruned_loss=0.1058, over 5662826.22 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:21:48,757 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.447e+02 1.365e+03 1.855e+03 2.424e+03 5.916e+03, threshold=3.711e+03, percent-clipped=8.0
+2023-03-06 22:22:36,471 INFO [train.py:968] (0/2) Epoch 13, batch 30400, giga_loss[loss=0.2621, simple_loss=0.3457, pruned_loss=0.08924, over 28729.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3564, pruned_loss=0.1067, over 5677843.57 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3709, pruned_loss=0.1236, over 5692127.50 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3567, pruned_loss=0.1058, over 5662286.37 frames. ], batch size: 243, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:22:58,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4710, 1.7395, 1.7469, 1.3190], device='cuda:0'), covar=tensor([0.1822, 0.2364, 0.1466, 0.1746], device='cuda:0'), in_proj_covar=tensor([0.0842, 0.0689, 0.0881, 0.0789], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 22:22:59,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3857, 1.9414, 1.4826, 1.6602], device='cuda:0'), covar=tensor([0.0761, 0.0257, 0.0319, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0060, 0.0055, 0.0093], device='cuda:0')
+2023-03-06 22:23:27,403 INFO [train.py:968] (0/2) Epoch 13, batch 30450, libri_loss[loss=0.3011, simple_loss=0.3536, pruned_loss=0.1243, over 29525.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3549, pruned_loss=0.1054, over 5676870.25 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3706, pruned_loss=0.1236, over 5697647.79 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3551, pruned_loss=0.1043, over 5659006.71 frames. ], batch size: 80, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:23:31,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.316e+02 1.398e+03 1.833e+03 2.638e+03 6.541e+03, threshold=3.666e+03, percent-clipped=9.0
+2023-03-06 22:24:16,411 INFO [train.py:968] (0/2) Epoch 13, batch 30500, libri_loss[loss=0.2953, simple_loss=0.3609, pruned_loss=0.1148, over 29668.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3522, pruned_loss=0.1037, over 5681960.46 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3699, pruned_loss=0.1235, over 5704598.97 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3523, pruned_loss=0.1021, over 5660492.10 frames. ], batch size: 88, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:24:51,144 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-578000.pt
+2023-03-06 22:25:05,437 INFO [train.py:968] (0/2) Epoch 13, batch 30550, giga_loss[loss=0.2798, simple_loss=0.359, pruned_loss=0.1003, over 28894.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3499, pruned_loss=0.1025, over 5670272.27 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3691, pruned_loss=0.1231, over 5704837.82 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3502, pruned_loss=0.101, over 5652012.22 frames. ], batch size: 164, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:25:07,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.050e+02 1.376e+03 1.930e+03 2.714e+03 1.143e+04, threshold=3.860e+03, percent-clipped=13.0
+2023-03-06 22:25:52,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3525, 1.3421, 3.7520, 3.3310], device='cuda:0'), covar=tensor([0.1500, 0.2716, 0.0406, 0.1690], device='cuda:0'), in_proj_covar=tensor([0.0684, 0.0603, 0.0879, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 22:25:53,902 INFO [train.py:968] (0/2) Epoch 13, batch 30600, libri_loss[loss=0.3574, simple_loss=0.3962, pruned_loss=0.1593, over 19107.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3501, pruned_loss=0.1024, over 5665089.24 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3688, pruned_loss=0.1232, over 5696383.96 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3501, pruned_loss=0.1005, over 5658172.22 frames. ], batch size: 187, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:25:54,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1819, 2.8268, 1.3661, 1.3604], device='cuda:0'), covar=tensor([0.0963, 0.0364, 0.0915, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0519, 0.0350, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-06 22:26:07,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5434, 1.7509, 1.4150, 1.6533], device='cuda:0'), covar=tensor([0.2547, 0.2242, 0.2548, 0.2064], device='cuda:0'), in_proj_covar=tensor([0.1351, 0.0987, 0.1192, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 22:26:43,188 INFO [train.py:968] (0/2) Epoch 13, batch 30650, giga_loss[loss=0.285, simple_loss=0.3589, pruned_loss=0.1055, over 28958.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3487, pruned_loss=0.101, over 5659697.61 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3685, pruned_loss=0.1231, over 5695309.13 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3486, pruned_loss=0.09924, over 5654857.59 frames. ], batch size: 199, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:26:45,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.935e+02 1.424e+03 1.752e+03 2.203e+03 6.496e+03, threshold=3.503e+03, percent-clipped=3.0
+2023-03-06 22:27:13,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5543, 1.8449, 1.8549, 1.6116], device='cuda:0'), covar=tensor([0.1525, 0.1815, 0.1534, 0.1680], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0725, 0.0672, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 22:27:30,414 INFO [train.py:968] (0/2) Epoch 13, batch 30700, giga_loss[loss=0.2773, simple_loss=0.3541, pruned_loss=0.1003, over 28223.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3454, pruned_loss=0.09849, over 5667334.71 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3679, pruned_loss=0.123, over 5702362.41 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3452, pruned_loss=0.09639, over 5656119.09 frames. ], batch size: 368, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:28:19,619 INFO [train.py:968] (0/2) Epoch 13, batch 30750, giga_loss[loss=0.2295, simple_loss=0.3132, pruned_loss=0.07294, over 28744.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3423, pruned_loss=0.09633, over 5677992.40 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3677, pruned_loss=0.123, over 5706291.37 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3418, pruned_loss=0.09422, over 5664983.77 frames. ], batch size: 284, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 22:28:22,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.273e+02 1.309e+03 1.622e+03 2.413e+03 5.557e+03, threshold=3.243e+03, percent-clipped=12.0
+2023-03-06 22:28:48,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3556, 1.2279, 3.8727, 3.2728], device='cuda:0'), covar=tensor([0.1562, 0.2732, 0.0426, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0602, 0.0877, 0.0792], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 22:29:06,886 INFO [train.py:968] (0/2) Epoch 13, batch 30800, giga_loss[loss=0.3308, simple_loss=0.3849, pruned_loss=0.1384, over 28961.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3405, pruned_loss=0.09609, over 5678938.37 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3669, pruned_loss=0.1226, over 5711102.93 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3402, pruned_loss=0.09403, over 5663594.55 frames. ], batch size: 164, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:29:33,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:29:57,008 INFO [train.py:968] (0/2) Epoch 13, batch 30850, libri_loss[loss=0.3031, simple_loss=0.3613, pruned_loss=0.1224, over 29585.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3409, pruned_loss=0.09733, over 5663365.08 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3666, pruned_loss=0.1227, over 5705138.89 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3403, pruned_loss=0.09498, over 5655051.11 frames. ], batch size: 76, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:30:01,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.128e+02 1.467e+03 2.030e+03 2.748e+03 6.301e+03, threshold=4.059e+03, percent-clipped=17.0
+2023-03-06 22:30:40,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6811, 2.0153, 1.9858, 1.5196], device='cuda:0'), covar=tensor([0.1937, 0.2403, 0.1520, 0.1841], device='cuda:0'), in_proj_covar=tensor([0.0845, 0.0689, 0.0884, 0.0791], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 22:30:51,642 INFO [train.py:968] (0/2) Epoch 13, batch 30900, giga_loss[loss=0.2586, simple_loss=0.3346, pruned_loss=0.09134, over 28855.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3412, pruned_loss=0.09745, over 5652997.37 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3664, pruned_loss=0.1227, over 5707194.78 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3407, pruned_loss=0.09541, over 5644254.31 frames. ], batch size: 99, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:31:15,444 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-06 22:31:44,405 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-06 22:31:48,396 INFO [train.py:968] (0/2) Epoch 13, batch 30950, giga_loss[loss=0.2598, simple_loss=0.3461, pruned_loss=0.08677, over 28920.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3435, pruned_loss=0.09779, over 5651559.04 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3659, pruned_loss=0.1223, over 5711640.47 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.343, pruned_loss=0.09586, over 5639077.91 frames. ], batch size: 227, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:31:49,376 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3822, 4.1754, 4.0038, 1.9217], device='cuda:0'), covar=tensor([0.0520, 0.0706, 0.0716, 0.2096], device='cuda:0'), in_proj_covar=tensor([0.1087, 0.1009, 0.0873, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 22:31:52,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.957e+02 1.478e+03 1.870e+03 2.554e+03 7.750e+03, threshold=3.741e+03, percent-clipped=6.0
+2023-03-06 22:32:49,519 INFO [train.py:968] (0/2) Epoch 13, batch 31000, libri_loss[loss=0.245, simple_loss=0.3123, pruned_loss=0.0888, over 29598.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3431, pruned_loss=0.09687, over 5643165.88 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3657, pruned_loss=0.1222, over 5714388.42 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3426, pruned_loss=0.09513, over 5629846.00 frames. ], batch size: 74, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:33:14,151 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:33:54,329 INFO [train.py:968] (0/2) Epoch 13, batch 31050, giga_loss[loss=0.2436, simple_loss=0.3089, pruned_loss=0.08913, over 24415.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3419, pruned_loss=0.09618, over 5646146.86 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.365, pruned_loss=0.1219, over 5717377.29 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3417, pruned_loss=0.09462, over 5631660.72 frames. ], batch size: 705, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:33:57,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.927e+02 1.442e+03 1.763e+03 2.216e+03 4.946e+03, threshold=3.526e+03, percent-clipped=5.0
+2023-03-06 22:34:38,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 22:34:43,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3901, 1.4414, 1.1115, 1.5085], device='cuda:0'), covar=tensor([0.0757, 0.0333, 0.0361, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-06 22:34:52,057 INFO [train.py:968] (0/2) Epoch 13, batch 31100, giga_loss[loss=0.242, simple_loss=0.3306, pruned_loss=0.07669, over 28992.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3398, pruned_loss=0.09424, over 5645338.66 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3643, pruned_loss=0.1215, over 5710035.21 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3396, pruned_loss=0.09266, over 5637750.69 frames. ], batch size: 227, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:34:53,661 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 22:35:12,307 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:35:36,610 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:35:53,477 INFO [train.py:968] (0/2) Epoch 13, batch 31150, giga_loss[loss=0.259, simple_loss=0.3361, pruned_loss=0.0909, over 28641.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3384, pruned_loss=0.09232, over 5633707.33 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3646, pruned_loss=0.1219, over 5702504.27 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3376, pruned_loss=0.09012, over 5632616.99 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:35:57,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.246e+02 1.211e+03 1.585e+03 2.740e+03 9.420e+03, threshold=3.170e+03, percent-clipped=8.0
+2023-03-06 22:36:46,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1140, 2.3135, 1.5438, 2.0295], device='cuda:0'), covar=tensor([0.0864, 0.0630, 0.0968, 0.1066], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0436, 0.0500, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 22:36:47,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3219, 1.5315, 1.3144, 1.4684], device='cuda:0'), covar=tensor([0.0726, 0.0310, 0.0322, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-06 22:36:51,642 INFO [train.py:968] (0/2) Epoch 13, batch 31200, giga_loss[loss=0.2314, simple_loss=0.305, pruned_loss=0.0789, over 29078.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3368, pruned_loss=0.09233, over 5631253.30 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3648, pruned_loss=0.1223, over 5687853.95 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3351, pruned_loss=0.08938, over 5641704.83 frames. ], batch size: 165, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:36:53,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=578666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:37:03,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3317, 1.2012, 3.9800, 3.3658], device='cuda:0'), covar=tensor([0.1618, 0.2815, 0.0421, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0676, 0.0598, 0.0868, 0.0784], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 22:37:49,828 INFO [train.py:968] (0/2) Epoch 13, batch 31250, giga_loss[loss=0.2728, simple_loss=0.3511, pruned_loss=0.09722, over 27742.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3364, pruned_loss=0.0926, over 5640624.19 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3648, pruned_loss=0.1224, over 5682607.59 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3343, pruned_loss=0.08939, over 5652209.58 frames. ], batch size: 472, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:37:55,168 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.129e+02 1.218e+03 1.678e+03 3.171e+03 8.868e+03, threshold=3.355e+03, percent-clipped=25.0
+2023-03-06 22:38:23,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2090, 1.5235, 1.4310, 1.5026], device='cuda:0'), covar=tensor([0.0791, 0.0290, 0.0291, 0.0993], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-06 22:38:46,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:38:47,850 INFO [train.py:968] (0/2) Epoch 13, batch 31300, giga_loss[loss=0.2666, simple_loss=0.3377, pruned_loss=0.09771, over 28890.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3358, pruned_loss=0.09224, over 5655663.25 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3643, pruned_loss=0.1221, over 5686989.08 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3341, pruned_loss=0.08942, over 5660529.60 frames. ], batch size: 227, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:39:05,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5775, 1.9854, 1.7114, 1.3791], device='cuda:0'), covar=tensor([0.2392, 0.1634, 0.1832, 0.2026], device='cuda:0'), in_proj_covar=tensor([0.1734, 0.1640, 0.1592, 0.1692], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 22:39:09,029 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:39:39,459 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=578809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:39:42,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=578812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:39:45,398 INFO [train.py:968] (0/2) Epoch 13, batch 31350, giga_loss[loss=0.2829, simple_loss=0.3625, pruned_loss=0.1016, over 28996.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3371, pruned_loss=0.09204, over 5650789.77 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3638, pruned_loss=0.1218, over 5681209.16 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3357, pruned_loss=0.08956, over 5659736.58 frames. ], batch size: 285, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:39:53,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.366e+02 1.379e+03 1.754e+03 2.327e+03 4.993e+03, threshold=3.509e+03, percent-clipped=5.0
+2023-03-06 22:39:58,385 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 22:40:19,249 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=578841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:40:31,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9464, 1.3857, 1.1441, 0.1752], device='cuda:0'), covar=tensor([0.2605, 0.2115, 0.3300, 0.3932], device='cuda:0'), in_proj_covar=tensor([0.1588, 0.1508, 0.1496, 0.1310], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 22:40:40,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=578857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:40:47,649 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-06 22:40:47,706 INFO [train.py:968] (0/2) Epoch 13, batch 31400, giga_loss[loss=0.2671, simple_loss=0.346, pruned_loss=0.09413, over 28953.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3392, pruned_loss=0.09324, over 5641312.70 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3637, pruned_loss=0.1218, over 5678520.05 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3373, pruned_loss=0.09015, over 5648697.41 frames. ], batch size: 199, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:40:55,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4291, 4.2665, 4.0614, 2.0434], device='cuda:0'), covar=tensor([0.0524, 0.0679, 0.0724, 0.1998], device='cuda:0'), in_proj_covar=tensor([0.1083, 0.1008, 0.0874, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 22:41:16,156 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:41:47,977 INFO [train.py:968] (0/2) Epoch 13, batch 31450, giga_loss[loss=0.2446, simple_loss=0.3284, pruned_loss=0.08043, over 28774.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3358, pruned_loss=0.09118, over 5647162.23 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3635, pruned_loss=0.1219, over 5665507.88 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3337, pruned_loss=0.0879, over 5664244.00 frames. ], batch size: 60, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:41:49,542 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:41:52,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.671e+02 1.376e+03 1.774e+03 2.454e+03 8.164e+03, threshold=3.547e+03, percent-clipped=8.0
+2023-03-06 22:41:55,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5998, 1.8365, 1.4935, 2.1325], device='cuda:0'), covar=tensor([0.2611, 0.2409, 0.2665, 0.2027], device='cuda:0'), in_proj_covar=tensor([0.1353, 0.0989, 0.1194, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 22:42:40,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=578955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:42:53,638 INFO [train.py:968] (0/2) Epoch 13, batch 31500, giga_loss[loss=0.264, simple_loss=0.3466, pruned_loss=0.09075, over 28624.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3374, pruned_loss=0.09251, over 5653752.97 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.363, pruned_loss=0.1215, over 5671210.29 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3356, pruned_loss=0.08955, over 5662015.94 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:43:05,119 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=578974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:43:09,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:43:10,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0179, 1.2050, 3.1739, 2.8223], device='cuda:0'), covar=tensor([0.1654, 0.2743, 0.0454, 0.1874], device='cuda:0'), in_proj_covar=tensor([0.0680, 0.0604, 0.0872, 0.0789], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 22:43:35,208 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:43:38,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:43:46,189 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 22:43:51,911 INFO [train.py:968] (0/2) Epoch 13, batch 31550, giga_loss[loss=0.2555, simple_loss=0.3563, pruned_loss=0.07735, over 28914.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3411, pruned_loss=0.09374, over 5657707.54 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3631, pruned_loss=0.1217, over 5670846.20 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3387, pruned_loss=0.09023, over 5664566.80 frames. ], batch size: 227, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:43:58,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.444e+02 1.524e+03 1.923e+03 2.779e+03 1.032e+04, threshold=3.845e+03, percent-clipped=13.0
+2023-03-06 22:44:12,242 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:44:54,761 INFO [train.py:968] (0/2) Epoch 13, batch 31600, giga_loss[loss=0.2567, simple_loss=0.3495, pruned_loss=0.08195, over 28422.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3427, pruned_loss=0.09206, over 5657927.25 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3631, pruned_loss=0.1218, over 5675599.30 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3405, pruned_loss=0.08869, over 5659010.47 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:44:55,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5346, 3.9575, 1.6417, 1.7534], device='cuda:0'), covar=tensor([0.0877, 0.0365, 0.0857, 0.1202], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0519, 0.0353, 0.0402], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-06 22:45:20,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1985, 2.7280, 1.3515, 1.3314], device='cuda:0'), covar=tensor([0.0947, 0.0329, 0.0937, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0519, 0.0353, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-06 22:45:33,584 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:45:38,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:45:53,803 INFO [train.py:968] (0/2) Epoch 13, batch 31650, giga_loss[loss=0.2433, simple_loss=0.3416, pruned_loss=0.07251, over 28962.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3433, pruned_loss=0.09112, over 5660657.23 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3627, pruned_loss=0.1216, over 5680352.89 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3413, pruned_loss=0.08787, over 5657040.38 frames. ], batch size: 155, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:45:56,830 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:45:59,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.957e+02 1.229e+03 1.627e+03 2.243e+03 4.384e+03, threshold=3.253e+03, percent-clipped=6.0
+2023-03-06 22:46:00,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:46:02,050 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=579122.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:46:09,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2344, 1.2082, 3.9790, 3.3870], device='cuda:0'), covar=tensor([0.2199, 0.3370, 0.0732, 0.1562], device='cuda:0'), in_proj_covar=tensor([0.0677, 0.0602, 0.0869, 0.0787], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 22:46:12,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:46:20,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:46:36,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:46:48,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:46:52,217 INFO [train.py:968] (0/2) Epoch 13, batch 31700, giga_loss[loss=0.242, simple_loss=0.3312, pruned_loss=0.07644, over 28972.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3431, pruned_loss=0.09034, over 5664071.30 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3622, pruned_loss=0.1214, over 5674588.27 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3416, pruned_loss=0.08729, over 5666797.89 frames. ], batch size: 145, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:46:54,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9206, 3.7246, 3.5407, 1.8284], device='cuda:0'), covar=tensor([0.0653, 0.0813, 0.0779, 0.2030], device='cuda:0'), in_proj_covar=tensor([0.1068, 0.0997, 0.0862, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 22:47:05,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3489, 1.5910, 1.5881, 1.2188], device='cuda:0'), covar=tensor([0.1688, 0.2477, 0.1388, 0.1601], device='cuda:0'), in_proj_covar=tensor([0.0850, 0.0687, 0.0889, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-06 22:47:54,615 INFO [train.py:968] (0/2) Epoch 13, batch 31750, giga_loss[loss=0.2196, simple_loss=0.3079, pruned_loss=0.06562, over 28860.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3446, pruned_loss=0.09235, over 5663882.42 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3625, pruned_loss=0.1217, over 5667139.94 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.08914, over 5672913.93 frames. ], batch size: 164, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:48:03,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.387e+02 1.343e+03 1.928e+03 2.719e+03 7.874e+03, threshold=3.856e+03, percent-clipped=16.0
+2023-03-06 22:48:21,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=579233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:48:37,149 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=579246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:49:02,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:49:02,623 INFO [train.py:968] (0/2) Epoch 13, batch 31800, giga_loss[loss=0.2782, simple_loss=0.3489, pruned_loss=0.1037, over 28420.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3434, pruned_loss=0.09307, over 5667944.96 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3623, pruned_loss=0.1215, over 5669612.81 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3417, pruned_loss=0.09003, over 5672978.59 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:49:28,982 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:49:32,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:49:36,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8743, 1.8498, 1.3462, 1.4827], device='cuda:0'), covar=tensor([0.0723, 0.0550, 0.0991, 0.1008], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0435, 0.0502, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 22:49:39,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:49:59,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:50:06,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:50:18,212 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:50:21,603 INFO [train.py:968] (0/2) Epoch 13, batch 31850, giga_loss[loss=0.2281, simple_loss=0.3127, pruned_loss=0.07178, over 28792.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3434, pruned_loss=0.094, over 5670626.74 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3617, pruned_loss=0.1213, over 5673954.00 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3419, pruned_loss=0.09083, over 5670876.97 frames. ], batch size: 243, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:50:27,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.908e+02 1.159e+03 1.609e+03 2.097e+03 4.067e+03, threshold=3.217e+03, percent-clipped=2.0
+2023-03-06 22:50:47,026 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:51:13,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:51:28,811 INFO [train.py:968] (0/2) Epoch 13, batch 31900, giga_loss[loss=0.2452, simple_loss=0.3193, pruned_loss=0.08551, over 28947.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3388, pruned_loss=0.09131, over 5678268.48 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3615, pruned_loss=0.1211, over 5679743.06 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3374, pruned_loss=0.08828, over 5673287.49 frames. ], batch size: 106, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:52:19,434 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1560, 1.6025, 1.3952, 1.4067], device='cuda:0'), covar=tensor([0.1750, 0.1675, 0.2040, 0.1880], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0717, 0.0664, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 22:52:19,453 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:52:22,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:52:28,371 INFO [train.py:968] (0/2) Epoch 13, batch 31950, giga_loss[loss=0.2257, simple_loss=0.3117, pruned_loss=0.06984, over 29096.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3364, pruned_loss=0.08999, over 5683145.19 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3604, pruned_loss=0.1205, over 5687031.32 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3355, pruned_loss=0.08701, over 5672515.44 frames. ], batch size: 285, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:52:38,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.143e+02 1.258e+03 1.620e+03 2.166e+03 7.677e+03, threshold=3.240e+03, percent-clipped=5.0
+2023-03-06 22:52:58,146 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:53:01,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:53:03,526 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:53:17,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5294, 1.7493, 1.4486, 1.6548], device='cuda:0'), covar=tensor([0.0743, 0.0295, 0.0311, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0116, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-06 22:53:27,197 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 22:53:33,723 INFO [train.py:968] (0/2) Epoch 13, batch 32000, giga_loss[loss=0.2572, simple_loss=0.3359, pruned_loss=0.0893, over 28418.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3355, pruned_loss=0.09035, over 5672945.77 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3605, pruned_loss=0.1208, over 5674654.82 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3342, pruned_loss=0.0872, over 5675789.63 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:53:35,737 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:54:12,996 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:54:14,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579497.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:54:15,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:54:37,232 INFO [train.py:968] (0/2) Epoch 13, batch 32050, giga_loss[loss=0.2472, simple_loss=0.3317, pruned_loss=0.08128, over 28865.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3392, pruned_loss=0.09208, over 5679691.36 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3607, pruned_loss=0.121, over 5680461.85 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3375, pruned_loss=0.08867, over 5676709.48 frames. ], batch size: 186, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:54:44,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.738e+02 1.491e+03 2.097e+03 3.353e+03 8.849e+03, threshold=4.194e+03, percent-clipped=25.0
+2023-03-06 22:54:51,317 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:55:32,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4510, 3.6950, 1.6354, 1.6125], device='cuda:0'), covar=tensor([0.0965, 0.0377, 0.0921, 0.1275], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0519, 0.0354, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-06 22:55:39,745 INFO [train.py:968] (0/2) Epoch 13, batch 32100, giga_loss[loss=0.2403, simple_loss=0.3136, pruned_loss=0.08351, over 28754.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3391, pruned_loss=0.09243, over 5688284.04 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3603, pruned_loss=0.1208, over 5682791.40 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3379, pruned_loss=0.08968, over 5683936.14 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:56:38,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:56:46,386 INFO [train.py:968] (0/2) Epoch 13, batch 32150, giga_loss[loss=0.2799, simple_loss=0.3539, pruned_loss=0.1029, over 28926.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3381, pruned_loss=0.09282, over 5668111.19 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3607, pruned_loss=0.1212, over 5666031.66 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3366, pruned_loss=0.09001, over 5680528.26 frames. ], batch size: 284, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:56:54,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:56:54,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.287e+02 1.332e+03 1.743e+03 2.408e+03 5.202e+03, threshold=3.486e+03, percent-clipped=2.0
+2023-03-06 22:57:18,198 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579640.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:57:21,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579643.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:57:47,697 INFO [train.py:968] (0/2) Epoch 13, batch 32200, giga_loss[loss=0.2652, simple_loss=0.3387, pruned_loss=0.09581, over 28924.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3383, pruned_loss=0.09336, over 5670296.11 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3602, pruned_loss=0.1208, over 5668489.89 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3372, pruned_loss=0.09103, over 5677754.12 frames. ], batch size: 199, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:57:56,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579672.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:58:37,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5356, 1.3293, 4.8648, 3.3975], device='cuda:0'), covar=tensor([0.1596, 0.2762, 0.0378, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0682, 0.0604, 0.0875, 0.0790], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 22:58:53,923 INFO [train.py:968] (0/2) Epoch 13, batch 32250, libri_loss[loss=0.2799, simple_loss=0.3393, pruned_loss=0.1103, over 29550.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.34, pruned_loss=0.09365, over 5673311.75 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3603, pruned_loss=0.121, over 5673875.60 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3385, pruned_loss=0.09087, over 5674310.71 frames. ], batch size: 80, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:59:06,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.921e+02 1.416e+03 1.778e+03 2.497e+03 1.104e+04, threshold=3.556e+03, percent-clipped=12.0
+2023-03-06 22:59:48,656 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:59:52,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:00:04,626 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:00:05,001 INFO [train.py:968] (0/2) Epoch 13, batch 32300, giga_loss[loss=0.2861, simple_loss=0.3574, pruned_loss=0.1074, over 27460.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.342, pruned_loss=0.09457, over 5667617.75 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3604, pruned_loss=0.1211, over 5675923.33 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3402, pruned_loss=0.09149, over 5666530.58 frames. ], batch size: 472, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:00:11,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:00:37,981 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:00:56,304 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:00:57,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=579797.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:01:18,878 INFO [train.py:968] (0/2) Epoch 13, batch 32350, giga_loss[loss=0.2646, simple_loss=0.3341, pruned_loss=0.09755, over 28867.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3408, pruned_loss=0.09414, over 5671773.17 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3597, pruned_loss=0.1207, over 5680565.46 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3396, pruned_loss=0.09157, over 5666342.73 frames. ], batch size: 164, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:01:26,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2079, 3.3983, 1.3777, 1.4414], device='cuda:0'), covar=tensor([0.1048, 0.0405, 0.0969, 0.1412], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0516, 0.0351, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-06 23:01:27,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.617e+02 1.511e+03 2.035e+03 2.631e+03 6.195e+03, threshold=4.070e+03, percent-clipped=9.0
+2023-03-06 23:02:15,201 INFO [train.py:968] (0/2) Epoch 13, batch 32400, giga_loss[loss=0.2166, simple_loss=0.3029, pruned_loss=0.06514, over 29023.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3374, pruned_loss=0.0939, over 5677844.44 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3594, pruned_loss=0.1205, over 5684625.06 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3357, pruned_loss=0.09056, over 5669768.23 frames. ], batch size: 155, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 23:02:50,672 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-06 23:03:15,371 INFO [train.py:968] (0/2) Epoch 13, batch 32450, giga_loss[loss=0.2491, simple_loss=0.3264, pruned_loss=0.08589, over 28823.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3325, pruned_loss=0.09178, over 5673240.29 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3591, pruned_loss=0.1204, over 5678739.09 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3306, pruned_loss=0.0883, over 5671272.86 frames. ], batch size: 263, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 23:03:22,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.343e+02 1.432e+03 1.867e+03 2.693e+03 1.014e+04, threshold=3.734e+03, percent-clipped=15.0
+2023-03-06 23:04:11,456 INFO [train.py:968] (0/2) Epoch 13, batch 32500, giga_loss[loss=0.2942, simple_loss=0.3614, pruned_loss=0.1135, over 28747.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3331, pruned_loss=0.0926, over 5676948.66 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3582, pruned_loss=0.1198, over 5686265.71 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3312, pruned_loss=0.08906, over 5668061.92 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:04:44,655 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-580000.pt
+2023-03-06 23:05:03,007 INFO [train.py:968] (0/2) Epoch 13, batch 32550, giga_loss[loss=0.2584, simple_loss=0.3403, pruned_loss=0.08821, over 28669.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3344, pruned_loss=0.09375, over 5680950.78 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3574, pruned_loss=0.1194, over 5694625.24 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3325, pruned_loss=0.09013, over 5666104.94 frames. ], batch size: 307, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:05:09,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.541e+02 1.580e+03 2.031e+03 2.946e+03 7.696e+03, threshold=4.062e+03, percent-clipped=16.0
+2023-03-06 23:06:01,665 INFO [train.py:968] (0/2) Epoch 13, batch 32600, giga_loss[loss=0.2237, simple_loss=0.3121, pruned_loss=0.06767, over 28714.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3324, pruned_loss=0.09158, over 5684455.53 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3573, pruned_loss=0.1192, over 5701255.56 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3302, pruned_loss=0.08797, over 5666438.66 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:07:03,325 INFO [train.py:968] (0/2) Epoch 13, batch 32650, giga_loss[loss=0.2342, simple_loss=0.3167, pruned_loss=0.0759, over 27538.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3306, pruned_loss=0.08954, over 5671200.99 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3572, pruned_loss=0.1192, over 5696140.87 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3284, pruned_loss=0.08603, over 5660213.96 frames. ], batch size: 472, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:07:12,858 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.965e+02 1.232e+03 1.666e+03 2.237e+03 4.450e+03, threshold=3.331e+03, percent-clipped=4.0
+2023-03-06 23:07:28,903 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=580133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:08:09,744 INFO [train.py:968] (0/2) Epoch 13, batch 32700, giga_loss[loss=0.2513, simple_loss=0.3284, pruned_loss=0.08712, over 28802.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3287, pruned_loss=0.08872, over 5667950.41 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3567, pruned_loss=0.1189, over 5698147.65 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.327, pruned_loss=0.08579, over 5657165.47 frames. ], batch size: 243, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:08:18,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=580172.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:09:20,786 INFO [train.py:968] (0/2) Epoch 13, batch 32750, giga_loss[loss=0.2387, simple_loss=0.3266, pruned_loss=0.07537, over 28764.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3277, pruned_loss=0.08708, over 5672994.17 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3565, pruned_loss=0.1189, over 5691246.80 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3261, pruned_loss=0.08445, over 5670325.11 frames. ], batch size: 119, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:09:30,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.800e+02 1.296e+03 1.749e+03 2.333e+03 9.863e+03, threshold=3.498e+03, percent-clipped=10.0
+2023-03-06 23:10:25,856 INFO [train.py:968] (0/2) Epoch 13, batch 32800, giga_loss[loss=0.3045, simple_loss=0.3711, pruned_loss=0.119, over 28100.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3291, pruned_loss=0.08772, over 5679118.58 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3565, pruned_loss=0.1188, over 5694639.17 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3275, pruned_loss=0.08529, over 5673980.90 frames. ], batch size: 412, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 23:11:25,115 INFO [train.py:968] (0/2) Epoch 13, batch 32850, giga_loss[loss=0.2857, simple_loss=0.3572, pruned_loss=0.1071, over 28691.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3303, pruned_loss=0.08921, over 5686763.26 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3563, pruned_loss=0.1188, over 5698259.89 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3286, pruned_loss=0.08654, over 5679072.25 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 23:11:25,712 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=580315.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:11:30,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=580318.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:11:36,639 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.745e+02 1.278e+03 1.616e+03 2.173e+03 4.736e+03, threshold=3.231e+03, percent-clipped=4.0
+2023-03-06 23:12:09,479 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=580347.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:12:25,782 INFO [train.py:968] (0/2) Epoch 13, batch 32900, libri_loss[loss=0.3406, simple_loss=0.3789, pruned_loss=0.1512, over 19772.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3297, pruned_loss=0.08875, over 5667220.47 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3563, pruned_loss=0.1189, over 5688542.53 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3278, pruned_loss=0.08584, over 5670142.50 frames. ], batch size: 186, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 23:13:28,384 INFO [train.py:968] (0/2) Epoch 13, batch 32950, giga_loss[loss=0.2838, simple_loss=0.359, pruned_loss=0.1042, over 29027.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3303, pruned_loss=0.08733, over 5659323.81 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3561, pruned_loss=0.1188, over 5689071.04 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3287, pruned_loss=0.0848, over 5660874.58 frames. ], batch size: 285, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:13:35,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.595e+02 1.219e+03 1.703e+03 2.357e+03 6.922e+03, threshold=3.406e+03, percent-clipped=8.0
+2023-03-06 23:13:56,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1069, 1.5906, 1.3840, 1.0404], device='cuda:0'), covar=tensor([0.1564, 0.2124, 0.1270, 0.1559], device='cuda:0'), in_proj_covar=tensor([0.0847, 0.0682, 0.0885, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 23:14:24,126 INFO [train.py:968] (0/2) Epoch 13, batch 33000, libri_loss[loss=0.2811, simple_loss=0.3479, pruned_loss=0.1072, over 29213.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3345, pruned_loss=0.08931, over 5658032.83 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3562, pruned_loss=0.1188, over 5687394.83 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3325, pruned_loss=0.08654, over 5659719.77 frames. ], batch size: 97, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:14:24,130 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-06 23:14:32,735 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.1997, simple_loss=0.3004, pruned_loss=0.04946, over 944034.00 frames. 
+2023-03-06 23:14:32,736 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-06 23:14:44,548 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=580476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:14:49,069 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 23:15:12,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=580499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:15:23,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=580508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:15:34,625 INFO [train.py:968] (0/2) Epoch 13, batch 33050, giga_loss[loss=0.2886, simple_loss=0.3626, pruned_loss=0.1073, over 28644.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3357, pruned_loss=0.08957, over 5659355.05 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3562, pruned_loss=0.1187, over 5681682.23 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3339, pruned_loss=0.08693, over 5665131.99 frames. ], batch size: 307, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:15:46,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.820e+02 1.332e+03 1.762e+03 2.586e+03 8.651e+03, threshold=3.523e+03, percent-clipped=13.0
+2023-03-06 23:16:39,123 INFO [train.py:968] (0/2) Epoch 13, batch 33100, giga_loss[loss=0.3013, simple_loss=0.3768, pruned_loss=0.1129, over 28703.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3355, pruned_loss=0.08964, over 5659236.05 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3557, pruned_loss=0.1184, over 5681468.28 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3341, pruned_loss=0.08745, over 5663379.21 frames. ], batch size: 307, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 23:17:17,322 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=580596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:17:37,951 INFO [train.py:968] (0/2) Epoch 13, batch 33150, giga_loss[loss=0.2764, simple_loss=0.3542, pruned_loss=0.09925, over 28613.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3317, pruned_loss=0.08705, over 5669664.42 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3551, pruned_loss=0.118, over 5685448.94 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3307, pruned_loss=0.08504, over 5668865.93 frames. ], batch size: 307, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 23:17:49,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.192e+02 1.290e+03 1.635e+03 2.245e+03 1.669e+04, threshold=3.271e+03, percent-clipped=9.0
+2023-03-06 23:18:18,815 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=580651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:18:22,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=580654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:18:33,907 INFO [train.py:968] (0/2) Epoch 13, batch 33200, giga_loss[loss=0.2357, simple_loss=0.3208, pruned_loss=0.07533, over 28604.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3315, pruned_loss=0.08751, over 5671018.13 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3547, pruned_loss=0.1179, over 5683629.11 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3302, pruned_loss=0.08478, over 5671950.23 frames. ], batch size: 307, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:18:53,210 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=580683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:19:01,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1896, 1.5805, 1.4907, 1.0996], device='cuda:0'), covar=tensor([0.1642, 0.2464, 0.1377, 0.1578], device='cuda:0'), in_proj_covar=tensor([0.0845, 0.0682, 0.0883, 0.0792], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-06 23:19:30,530 INFO [train.py:968] (0/2) Epoch 13, batch 33250, giga_loss[loss=0.24, simple_loss=0.3227, pruned_loss=0.07864, over 28061.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3294, pruned_loss=0.08737, over 5674398.20 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3541, pruned_loss=0.1177, over 5688518.67 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3283, pruned_loss=0.08459, over 5670500.80 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:19:45,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.312e+02 1.258e+03 1.708e+03 2.398e+03 1.571e+04, threshold=3.416e+03, percent-clipped=12.0
+2023-03-06 23:19:52,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5978, 1.9249, 1.5398, 1.6772], device='cuda:0'), covar=tensor([0.2303, 0.1898, 0.2140, 0.1848], device='cuda:0'), in_proj_covar=tensor([0.1342, 0.0978, 0.1187, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 23:20:07,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7615, 1.9605, 1.5276, 2.1956], device='cuda:0'), covar=tensor([0.2440, 0.2453, 0.2869, 0.2219], device='cuda:0'), in_proj_covar=tensor([0.1343, 0.0977, 0.1187, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 23:20:31,588 INFO [train.py:968] (0/2) Epoch 13, batch 33300, giga_loss[loss=0.2983, simple_loss=0.3769, pruned_loss=0.1098, over 28141.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3307, pruned_loss=0.08777, over 5673750.16 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3537, pruned_loss=0.1175, over 5690183.00 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3298, pruned_loss=0.08534, over 5668830.51 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:21:00,706 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0955, 3.9035, 3.7252, 1.7320], device='cuda:0'), covar=tensor([0.0649, 0.0814, 0.0775, 0.2322], device='cuda:0'), in_proj_covar=tensor([0.1071, 0.0989, 0.0859, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 23:21:33,716 INFO [train.py:968] (0/2) Epoch 13, batch 33350, giga_loss[loss=0.2587, simple_loss=0.335, pruned_loss=0.09115, over 28685.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3327, pruned_loss=0.08905, over 5674687.74 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3534, pruned_loss=0.1174, over 5691369.32 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3316, pruned_loss=0.08636, over 5669188.87 frames. ], batch size: 242, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:21:42,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5114, 1.7789, 1.3676, 1.8812], device='cuda:0'), covar=tensor([0.2809, 0.2579, 0.2934, 0.2342], device='cuda:0'), in_proj_covar=tensor([0.1348, 0.0980, 0.1190, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 23:21:44,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.457e+02 1.325e+03 1.665e+03 2.281e+03 6.567e+03, threshold=3.331e+03, percent-clipped=6.0
+2023-03-06 23:22:01,149 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-06 23:22:18,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=580851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:22:40,517 INFO [train.py:968] (0/2) Epoch 13, batch 33400, giga_loss[loss=0.227, simple_loss=0.3055, pruned_loss=0.07422, over 28700.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3343, pruned_loss=0.09088, over 5669558.20 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3536, pruned_loss=0.1177, over 5694345.91 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.333, pruned_loss=0.08815, over 5662392.61 frames. ], batch size: 92, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:22:48,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3870, 5.2069, 4.9163, 2.0513], device='cuda:0'), covar=tensor([0.0397, 0.0581, 0.0660, 0.2142], device='cuda:0'), in_proj_covar=tensor([0.1072, 0.0992, 0.0861, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 23:22:51,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=580874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:23:15,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5491, 1.8566, 1.4922, 1.8516], device='cuda:0'), covar=tensor([0.2309, 0.2051, 0.2277, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.1347, 0.0980, 0.1188, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 23:23:41,669 INFO [train.py:968] (0/2) Epoch 13, batch 33450, giga_loss[loss=0.299, simple_loss=0.3767, pruned_loss=0.1106, over 28997.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3373, pruned_loss=0.0925, over 5662459.90 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3529, pruned_loss=0.1172, over 5698803.71 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3365, pruned_loss=0.09003, over 5651709.06 frames. ], batch size: 199, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:23:54,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.518e+02 1.528e+03 2.000e+03 2.948e+03 1.751e+04, threshold=4.001e+03, percent-clipped=18.0
+2023-03-06 23:24:37,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6236, 1.7037, 1.2017, 1.2730], device='cuda:0'), covar=tensor([0.0744, 0.0514, 0.1002, 0.1101], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0432, 0.0502, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 23:24:39,010 INFO [train.py:968] (0/2) Epoch 13, batch 33500, giga_loss[loss=0.271, simple_loss=0.3465, pruned_loss=0.09774, over 27622.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3405, pruned_loss=0.09316, over 5660596.37 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3533, pruned_loss=0.1176, over 5691419.73 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3392, pruned_loss=0.09051, over 5658569.65 frames. ], batch size: 472, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:24:46,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=580971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:25:16,220 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=580994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:25:17,669 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=580995.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:25:21,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=580997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:25:44,337 INFO [train.py:968] (0/2) Epoch 13, batch 33550, giga_loss[loss=0.255, simple_loss=0.3396, pruned_loss=0.08516, over 28689.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3422, pruned_loss=0.09436, over 5663373.27 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3535, pruned_loss=0.1177, over 5692007.61 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3409, pruned_loss=0.09177, over 5660832.03 frames. ], batch size: 262, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:25:48,118 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:25:56,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:26:05,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.285e+02 1.523e+03 1.950e+03 2.794e+03 1.253e+04, threshold=3.899e+03, percent-clipped=17.0
+2023-03-06 23:26:05,730 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:26:26,821 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:26:48,276 INFO [train.py:968] (0/2) Epoch 13, batch 33600, giga_loss[loss=0.2316, simple_loss=0.313, pruned_loss=0.07505, over 28137.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3391, pruned_loss=0.09311, over 5662927.49 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3532, pruned_loss=0.1176, over 5687987.26 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3377, pruned_loss=0.0899, over 5663218.38 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:27:46,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:27:48,096 INFO [train.py:968] (0/2) Epoch 13, batch 33650, giga_loss[loss=0.2663, simple_loss=0.3375, pruned_loss=0.09752, over 28767.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3381, pruned_loss=0.09247, over 5679176.07 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3529, pruned_loss=0.1173, over 5694586.92 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3368, pruned_loss=0.08946, over 5673102.99 frames. ], batch size: 99, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:27:51,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:28:04,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.708e+02 1.381e+03 1.770e+03 2.658e+03 5.444e+03, threshold=3.540e+03, percent-clipped=5.0
+2023-03-06 23:28:30,480 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:28:31,697 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:28:47,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 23:28:54,195 INFO [train.py:968] (0/2) Epoch 13, batch 33700, giga_loss[loss=0.2861, simple_loss=0.3542, pruned_loss=0.109, over 28531.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3373, pruned_loss=0.09229, over 5677657.14 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3528, pruned_loss=0.1173, over 5697423.96 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3361, pruned_loss=0.08946, over 5669692.59 frames. ], batch size: 336, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:30:01,498 INFO [train.py:968] (0/2) Epoch 13, batch 33750, giga_loss[loss=0.2501, simple_loss=0.3265, pruned_loss=0.08686, over 28647.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3353, pruned_loss=0.09211, over 5670552.64 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3527, pruned_loss=0.1171, over 5690475.32 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3342, pruned_loss=0.08965, over 5669850.33 frames. ], batch size: 307, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:30:13,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.326e+02 1.144e+03 1.636e+03 2.212e+03 4.565e+03, threshold=3.272e+03, percent-clipped=4.0
+2023-03-06 23:30:38,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581245.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:31:00,172 INFO [train.py:968] (0/2) Epoch 13, batch 33800, giga_loss[loss=0.2279, simple_loss=0.3223, pruned_loss=0.06674, over 28759.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3348, pruned_loss=0.09176, over 5670769.88 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3528, pruned_loss=0.1173, over 5683865.63 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3335, pruned_loss=0.0891, over 5676082.82 frames. ], batch size: 243, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:31:34,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6105, 2.1396, 1.4838, 0.8875], device='cuda:0'), covar=tensor([0.4429, 0.2508, 0.3693, 0.4776], device='cuda:0'), in_proj_covar=tensor([0.1582, 0.1522, 0.1502, 0.1314], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 23:32:03,342 INFO [train.py:968] (0/2) Epoch 13, batch 33850, giga_loss[loss=0.2388, simple_loss=0.3199, pruned_loss=0.07884, over 29034.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3331, pruned_loss=0.08956, over 5657378.59 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3526, pruned_loss=0.1172, over 5676434.31 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.332, pruned_loss=0.08732, over 5668666.22 frames. ], batch size: 128, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:32:18,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.552e+02 1.312e+03 1.799e+03 2.406e+03 9.691e+03, threshold=3.598e+03, percent-clipped=14.0
+2023-03-06 23:32:58,581 INFO [train.py:968] (0/2) Epoch 13, batch 33900, giga_loss[loss=0.2289, simple_loss=0.3276, pruned_loss=0.06514, over 29067.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3338, pruned_loss=0.08837, over 5662635.14 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3523, pruned_loss=0.1171, over 5673580.60 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3328, pruned_loss=0.08598, over 5673849.59 frames. ], batch size: 155, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:33:05,293 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=581370.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:33:17,132 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.08 vs. limit=2.0
+2023-03-06 23:33:49,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3654, 1.5698, 1.4045, 1.5392], device='cuda:0'), covar=tensor([0.0777, 0.0317, 0.0333, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0095], device='cuda:0')
+2023-03-06 23:33:58,087 INFO [train.py:968] (0/2) Epoch 13, batch 33950, giga_loss[loss=0.2439, simple_loss=0.3326, pruned_loss=0.07762, over 28537.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3361, pruned_loss=0.08823, over 5668285.51 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3521, pruned_loss=0.1169, over 5677007.24 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3352, pruned_loss=0.0861, over 5673876.00 frames. ], batch size: 336, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:34:10,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.090e+02 1.270e+03 1.764e+03 2.537e+03 5.972e+03, threshold=3.529e+03, percent-clipped=11.0
+2023-03-06 23:34:34,300 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 23:34:47,297 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:34:57,275 INFO [train.py:968] (0/2) Epoch 13, batch 34000, giga_loss[loss=0.2464, simple_loss=0.3325, pruned_loss=0.08011, over 28900.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3374, pruned_loss=0.0893, over 5662114.05 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3519, pruned_loss=0.1169, over 5667872.69 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3366, pruned_loss=0.08695, over 5674789.57 frames. ], batch size: 227, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:35:00,639 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:35:11,798 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8154, 4.5911, 4.3722, 1.9309], device='cuda:0'), covar=tensor([0.0555, 0.0825, 0.0847, 0.2089], device='cuda:0'), in_proj_covar=tensor([0.1069, 0.0991, 0.0861, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-06 23:36:02,809 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581513.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:36:04,233 INFO [train.py:968] (0/2) Epoch 13, batch 34050, giga_loss[loss=0.2736, simple_loss=0.3479, pruned_loss=0.09964, over 28370.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3369, pruned_loss=0.08908, over 5665705.24 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3522, pruned_loss=0.1171, over 5672626.50 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3356, pruned_loss=0.08637, over 5671403.52 frames. ], batch size: 368, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:36:05,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581516.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:36:12,046 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=581520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:36:21,576 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.735e+02 1.443e+03 2.022e+03 2.484e+03 6.217e+03, threshold=4.043e+03, percent-clipped=7.0
+2023-03-06 23:36:42,069 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581545.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:36:57,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6344, 2.0050, 1.4726, 1.8577], device='cuda:0'), covar=tensor([0.2540, 0.2284, 0.2770, 0.2086], device='cuda:0'), in_proj_covar=tensor([0.1342, 0.0984, 0.1187, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-06 23:37:07,039 INFO [train.py:968] (0/2) Epoch 13, batch 34100, giga_loss[loss=0.2268, simple_loss=0.3202, pruned_loss=0.06668, over 28948.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3375, pruned_loss=0.0898, over 5661989.78 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3518, pruned_loss=0.117, over 5670105.95 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3363, pruned_loss=0.08692, over 5668213.60 frames. ], batch size: 155, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:38:19,633 INFO [train.py:968] (0/2) Epoch 13, batch 34150, giga_loss[loss=0.2472, simple_loss=0.3388, pruned_loss=0.07777, over 29033.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3368, pruned_loss=0.08853, over 5668722.70 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3514, pruned_loss=0.1167, over 5674926.06 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3359, pruned_loss=0.08594, over 5669102.18 frames. ], batch size: 285, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:38:27,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=581620.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:38:39,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.711e+02 1.535e+03 2.069e+03 2.821e+03 7.968e+03, threshold=4.138e+03, percent-clipped=13.0
+2023-03-06 23:39:29,046 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:39:31,063 INFO [train.py:968] (0/2) Epoch 13, batch 34200, giga_loss[loss=0.2395, simple_loss=0.3243, pruned_loss=0.07735, over 28075.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3367, pruned_loss=0.0883, over 5665234.02 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.351, pruned_loss=0.1164, over 5675193.85 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3361, pruned_loss=0.08589, over 5665176.38 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:39:33,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:40:09,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:40:24,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7575, 1.8253, 1.2954, 1.4294], device='cuda:0'), covar=tensor([0.0816, 0.0568, 0.1044, 0.1038], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0433, 0.0503, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 23:40:35,225 INFO [train.py:968] (0/2) Epoch 13, batch 34250, giga_loss[loss=0.2886, simple_loss=0.3644, pruned_loss=0.1064, over 29050.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3403, pruned_loss=0.08988, over 5663839.67 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3509, pruned_loss=0.1164, over 5671114.95 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3396, pruned_loss=0.08725, over 5667231.85 frames. ], batch size: 100, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:40:47,638 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:40:49,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.205e+02 1.349e+03 1.702e+03 2.350e+03 6.418e+03, threshold=3.405e+03, percent-clipped=4.0
+2023-03-06 23:41:33,349 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:41:37,388 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581763.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:41:38,314 INFO [train.py:968] (0/2) Epoch 13, batch 34300, libri_loss[loss=0.3099, simple_loss=0.3691, pruned_loss=0.1254, over 25905.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3397, pruned_loss=0.08948, over 5674121.43 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3504, pruned_loss=0.1159, over 5676317.92 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3391, pruned_loss=0.08678, over 5672184.23 frames. ], batch size: 136, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:41:41,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581766.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:42:01,270 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:42:20,898 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581795.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:42:48,784 INFO [train.py:968] (0/2) Epoch 13, batch 34350, giga_loss[loss=0.2597, simple_loss=0.3354, pruned_loss=0.09197, over 28948.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3387, pruned_loss=0.08996, over 5672435.22 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.35, pruned_loss=0.1157, over 5677107.49 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3383, pruned_loss=0.08743, over 5670280.37 frames. ], batch size: 186, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:43:03,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.943e+02 1.335e+03 1.779e+03 2.650e+03 1.087e+04, threshold=3.558e+03, percent-clipped=11.0
+2023-03-06 23:43:11,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=581831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:43:29,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=581842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:43:45,574 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 23:44:01,770 INFO [train.py:968] (0/2) Epoch 13, batch 34400, giga_loss[loss=0.239, simple_loss=0.3247, pruned_loss=0.07661, over 28671.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3361, pruned_loss=0.08787, over 5679455.27 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3499, pruned_loss=0.1155, over 5680982.34 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3357, pruned_loss=0.08563, over 5674401.47 frames. ], batch size: 307, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:44:38,760 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581889.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:45:09,606 INFO [train.py:968] (0/2) Epoch 13, batch 34450, giga_loss[loss=0.2446, simple_loss=0.3286, pruned_loss=0.08036, over 28909.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3349, pruned_loss=0.08673, over 5686920.77 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3497, pruned_loss=0.1154, over 5685496.77 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3345, pruned_loss=0.0845, over 5678829.59 frames. ], batch size: 227, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:45:28,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.075e+02 1.179e+03 1.446e+03 2.006e+03 6.196e+03, threshold=2.891e+03, percent-clipped=4.0
+2023-03-06 23:46:15,738 INFO [train.py:968] (0/2) Epoch 13, batch 34500, giga_loss[loss=0.3449, simple_loss=0.3934, pruned_loss=0.1481, over 26831.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3365, pruned_loss=0.08793, over 5672290.91 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3496, pruned_loss=0.1154, over 5683711.57 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3362, pruned_loss=0.08596, over 5667371.98 frames. ], batch size: 555, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:46:26,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:46:30,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:46:39,567 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:46:43,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:46:57,278 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-582000.pt
+2023-03-06 23:47:05,121 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:47:16,396 INFO [train.py:968] (0/2) Epoch 13, batch 34550, giga_loss[loss=0.2387, simple_loss=0.3185, pruned_loss=0.07949, over 24587.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.34, pruned_loss=0.08994, over 5660814.45 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3499, pruned_loss=0.1157, over 5672017.17 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3393, pruned_loss=0.0878, over 5666431.34 frames. ], batch size: 705, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:47:18,641 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:47:31,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.103e+02 1.441e+03 1.902e+03 3.001e+03 7.654e+03, threshold=3.805e+03, percent-clipped=24.0
+2023-03-06 23:47:38,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4040, 1.6100, 1.2988, 1.4803], device='cuda:0'), covar=tensor([0.0726, 0.0376, 0.0349, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-06 23:47:53,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4847, 1.7678, 1.6704, 1.6220], device='cuda:0'), covar=tensor([0.1596, 0.1863, 0.2049, 0.1808], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0711, 0.0659, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 23:48:14,149 INFO [train.py:968] (0/2) Epoch 13, batch 34600, giga_loss[loss=0.2453, simple_loss=0.3273, pruned_loss=0.08169, over 28859.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3398, pruned_loss=0.09057, over 5677537.11 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3502, pruned_loss=0.1159, over 5678629.64 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3388, pruned_loss=0.08798, over 5675807.17 frames. ], batch size: 164, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:48:45,634 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-06 23:48:54,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:49:10,088 INFO [train.py:968] (0/2) Epoch 13, batch 34650, giga_loss[loss=0.2645, simple_loss=0.3408, pruned_loss=0.09412, over 28675.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.337, pruned_loss=0.09023, over 5669644.85 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3501, pruned_loss=0.1158, over 5680274.11 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3359, pruned_loss=0.08759, over 5666306.96 frames. ], batch size: 307, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:49:22,925 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.523e+02 1.446e+03 2.008e+03 2.627e+03 9.432e+03, threshold=4.016e+03, percent-clipped=12.0
+2023-03-06 23:49:24,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3928, 1.5652, 1.4336, 1.4069], device='cuda:0'), covar=tensor([0.1776, 0.1451, 0.1367, 0.1364], device='cuda:0'), in_proj_covar=tensor([0.1752, 0.1631, 0.1590, 0.1689], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-06 23:49:31,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:49:54,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:50:04,991 INFO [train.py:968] (0/2) Epoch 13, batch 34700, giga_loss[loss=0.2569, simple_loss=0.3379, pruned_loss=0.08792, over 28102.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3369, pruned_loss=0.09088, over 5664623.96 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3505, pruned_loss=0.116, over 5676556.69 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3354, pruned_loss=0.08803, over 5664912.32 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:50:56,428 INFO [train.py:968] (0/2) Epoch 13, batch 34750, giga_loss[loss=0.2536, simple_loss=0.336, pruned_loss=0.08558, over 28071.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3418, pruned_loss=0.09407, over 5670707.97 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3498, pruned_loss=0.1155, over 5681493.09 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3409, pruned_loss=0.09146, over 5665822.37 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:50:59,689 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2659, 1.5816, 1.4494, 1.4219], device='cuda:0'), covar=tensor([0.1417, 0.1356, 0.1934, 0.1543], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0713, 0.0661, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 23:51:09,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.368e+02 1.457e+03 1.841e+03 2.730e+03 6.321e+03, threshold=3.681e+03, percent-clipped=7.0
+2023-03-06 23:51:18,233 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=582242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:51:20,919 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=582245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:51:25,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.21 vs. limit=5.0
+2023-03-06 23:51:39,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582264.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:51:39,904 INFO [train.py:968] (0/2) Epoch 13, batch 34800, giga_loss[loss=0.3024, simple_loss=0.3838, pruned_loss=0.1104, over 28301.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3497, pruned_loss=0.09908, over 5659915.87 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3496, pruned_loss=0.1153, over 5672238.70 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.349, pruned_loss=0.09647, over 5663779.52 frames. ], batch size: 368, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:51:46,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:51:52,424 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=582279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:51:55,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=582282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:52:06,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=582293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:52:10,676 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=582299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:52:12,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=582302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:52:20,745 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:52:24,283 INFO [train.py:968] (0/2) Epoch 13, batch 34850, giga_loss[loss=0.2483, simple_loss=0.3308, pruned_loss=0.08292, over 29041.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3548, pruned_loss=0.1022, over 5655288.82 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.35, pruned_loss=0.1156, over 5664644.67 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3539, pruned_loss=0.09952, over 5664568.13 frames. ], batch size: 155, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:52:36,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.034e+02 1.300e+03 1.721e+03 2.327e+03 9.619e+03, threshold=3.441e+03, percent-clipped=6.0
+2023-03-06 23:52:38,246 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:53:05,333 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-06 23:53:07,967 INFO [train.py:968] (0/2) Epoch 13, batch 34900, giga_loss[loss=0.2432, simple_loss=0.3166, pruned_loss=0.08493, over 28887.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3516, pruned_loss=0.1016, over 5663020.97 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3496, pruned_loss=0.1153, over 5659404.93 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3513, pruned_loss=0.09953, over 5675596.83 frames. ], batch size: 199, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:53:37,411 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5880, 1.6215, 1.1954, 1.2125], device='cuda:0'), covar=tensor([0.0700, 0.0444, 0.0929, 0.1110], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0430, 0.0499, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-06 23:53:42,254 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=582407.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:53:45,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=582410.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:53:48,149 INFO [train.py:968] (0/2) Epoch 13, batch 34950, giga_loss[loss=0.2165, simple_loss=0.2991, pruned_loss=0.06694, over 28989.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3451, pruned_loss=0.09895, over 5676777.61 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3492, pruned_loss=0.115, over 5665300.89 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3452, pruned_loss=0.09722, over 5681888.28 frames. ], batch size: 164, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:54:02,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.276e+02 1.026e+03 1.245e+03 1.571e+03 3.147e+03, threshold=2.491e+03, percent-clipped=0.0
+2023-03-06 23:54:04,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3672, 1.3029, 3.7622, 3.2194], device='cuda:0'), covar=tensor([0.1447, 0.2687, 0.0406, 0.1726], device='cuda:0'), in_proj_covar=tensor([0.0681, 0.0600, 0.0878, 0.0789], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 23:54:09,968 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582439.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:54:28,308 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-06 23:54:31,693 INFO [train.py:968] (0/2) Epoch 13, batch 35000, giga_loss[loss=0.2436, simple_loss=0.3174, pruned_loss=0.08487, over 28575.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3384, pruned_loss=0.09605, over 5675760.49 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3498, pruned_loss=0.1152, over 5668887.44 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3378, pruned_loss=0.09419, over 5676887.18 frames. ], batch size: 336, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:55:13,863 INFO [train.py:968] (0/2) Epoch 13, batch 35050, libri_loss[loss=0.4117, simple_loss=0.4445, pruned_loss=0.1895, over 29502.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3329, pruned_loss=0.09443, over 5683710.65 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3506, pruned_loss=0.1156, over 5675437.86 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3313, pruned_loss=0.09194, over 5679100.24 frames. ], batch size: 85, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:55:23,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.673e+02 9.747e+02 1.210e+03 1.574e+03 4.766e+03, threshold=2.420e+03, percent-clipped=6.0
+2023-03-06 23:55:42,842 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=582554.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:55:43,691 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-06 23:55:52,059 INFO [train.py:968] (0/2) Epoch 13, batch 35100, giga_loss[loss=0.2209, simple_loss=0.2966, pruned_loss=0.07265, over 28461.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3261, pruned_loss=0.09127, over 5695433.75 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3506, pruned_loss=0.1155, over 5681986.71 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3241, pruned_loss=0.0887, over 5686074.00 frames. ], batch size: 71, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:56:17,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4374, 1.6013, 1.6142, 1.4892], device='cuda:0'), covar=tensor([0.1735, 0.1922, 0.2238, 0.1914], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0725, 0.0670, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-06 23:56:35,348 INFO [train.py:968] (0/2) Epoch 13, batch 35150, giga_loss[loss=0.2446, simple_loss=0.2971, pruned_loss=0.09611, over 23919.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3217, pruned_loss=0.08975, over 5689680.90 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3505, pruned_loss=0.1155, over 5687973.17 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3195, pruned_loss=0.08706, over 5677245.26 frames. ], batch size: 705, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:56:46,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.475e+02 1.078e+03 1.324e+03 1.963e+03 4.436e+03, threshold=2.648e+03, percent-clipped=12.0
+2023-03-06 23:57:17,162 INFO [train.py:968] (0/2) Epoch 13, batch 35200, giga_loss[loss=0.2267, simple_loss=0.3004, pruned_loss=0.07653, over 28805.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3179, pruned_loss=0.08749, over 5695784.24 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3504, pruned_loss=0.1153, over 5691273.88 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3159, pruned_loss=0.08517, over 5683023.59 frames. ], batch size: 199, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:57:20,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:57:54,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0631, 1.1125, 3.7198, 3.0971], device='cuda:0'), covar=tensor([0.1760, 0.2891, 0.0455, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0679, 0.0598, 0.0873, 0.0783], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-06 23:57:58,986 INFO [train.py:968] (0/2) Epoch 13, batch 35250, giga_loss[loss=0.2191, simple_loss=0.2992, pruned_loss=0.06954, over 28833.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.316, pruned_loss=0.08643, over 5697070.89 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3519, pruned_loss=0.1162, over 5687190.84 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.3118, pruned_loss=0.08278, over 5691382.66 frames. ], batch size: 199, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:58:11,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.951e+02 1.093e+03 1.415e+03 2.006e+03 4.685e+03, threshold=2.830e+03, percent-clipped=11.0
+2023-03-06 23:58:42,376 INFO [train.py:968] (0/2) Epoch 13, batch 35300, giga_loss[loss=0.2027, simple_loss=0.2761, pruned_loss=0.06465, over 28738.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3144, pruned_loss=0.08553, over 5707099.94 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3519, pruned_loss=0.116, over 5692013.34 frames. ], giga_tot_loss[loss=0.236, simple_loss=0.3093, pruned_loss=0.08138, over 5698929.91 frames. ], batch size: 99, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:59:00,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9637, 2.9075, 1.9428, 1.1390], device='cuda:0'), covar=tensor([0.5763, 0.2355, 0.3104, 0.5322], device='cuda:0'), in_proj_covar=tensor([0.1591, 0.1528, 0.1498, 0.1316], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 23:59:19,540 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=582811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:59:22,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=582814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:59:22,385 INFO [train.py:968] (0/2) Epoch 13, batch 35350, giga_loss[loss=0.2135, simple_loss=0.2841, pruned_loss=0.07145, over 28991.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3125, pruned_loss=0.08501, over 5699752.53 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3521, pruned_loss=0.1159, over 5689319.64 frames. ], giga_tot_loss[loss=0.234, simple_loss=0.3068, pruned_loss=0.0806, over 5695525.82 frames. ], batch size: 136, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:59:27,478 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5551, 2.2846, 1.6739, 0.7619], device='cuda:0'), covar=tensor([0.4812, 0.2264, 0.3104, 0.4330], device='cuda:0'), in_proj_covar=tensor([0.1594, 0.1530, 0.1502, 0.1319], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-06 23:59:33,080 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.572e+02 1.009e+03 1.352e+03 2.257e+03 9.857e+03, threshold=2.703e+03, percent-clipped=14.0
+2023-03-06 23:59:46,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:00:00,417 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=582861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:00:02,917 INFO [train.py:968] (0/2) Epoch 13, batch 35400, giga_loss[loss=0.2009, simple_loss=0.2817, pruned_loss=0.06004, over 29016.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3097, pruned_loss=0.08361, over 5699385.14 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3522, pruned_loss=0.1159, over 5686322.13 frames. ], giga_tot_loss[loss=0.231, simple_loss=0.3038, pruned_loss=0.07914, over 5699438.55 frames. ], batch size: 164, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:00:13,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2843, 3.1321, 2.9328, 1.3777], device='cuda:0'), covar=tensor([0.0865, 0.1039, 0.0899, 0.2270], device='cuda:0'), in_proj_covar=tensor([0.1071, 0.0997, 0.0866, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 00:00:25,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5207, 1.6550, 1.7123, 1.4818], device='cuda:0'), covar=tensor([0.1704, 0.1897, 0.2149, 0.2010], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0725, 0.0670, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 00:00:41,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2855, 1.5035, 1.6008, 1.3228], device='cuda:0'), covar=tensor([0.1559, 0.1439, 0.1986, 0.1683], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0724, 0.0670, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 00:00:43,269 INFO [train.py:968] (0/2) Epoch 13, batch 35450, giga_loss[loss=0.1991, simple_loss=0.2784, pruned_loss=0.05984, over 29009.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3074, pruned_loss=0.08252, over 5688738.85 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3527, pruned_loss=0.1161, over 5676548.74 frames. ], giga_tot_loss[loss=0.2287, simple_loss=0.3014, pruned_loss=0.07806, over 5698248.89 frames. ], batch size: 164, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:00:56,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582929.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:00:56,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.445e+02 1.051e+03 1.341e+03 1.757e+03 3.995e+03, threshold=2.683e+03, percent-clipped=5.0
+2023-03-07 00:01:22,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2773, 1.4403, 3.3773, 3.1597], device='cuda:0'), covar=tensor([0.1364, 0.2377, 0.0421, 0.0931], device='cuda:0'), in_proj_covar=tensor([0.0680, 0.0599, 0.0875, 0.0786], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-07 00:01:22,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3251, 1.6918, 1.2310, 0.8113], device='cuda:0'), covar=tensor([0.4160, 0.2275, 0.2472, 0.4497], device='cuda:0'), in_proj_covar=tensor([0.1597, 0.1532, 0.1505, 0.1321], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 00:01:24,238 INFO [train.py:968] (0/2) Epoch 13, batch 35500, giga_loss[loss=0.2013, simple_loss=0.2695, pruned_loss=0.06652, over 28892.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3045, pruned_loss=0.0811, over 5688890.59 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3526, pruned_loss=0.1158, over 5685581.46 frames. ], giga_tot_loss[loss=0.2255, simple_loss=0.298, pruned_loss=0.0765, over 5688613.71 frames. ], batch size: 93, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:01:41,149 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 00:02:10,001 INFO [train.py:968] (0/2) Epoch 13, batch 35550, giga_loss[loss=0.2224, simple_loss=0.2858, pruned_loss=0.0795, over 28573.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3014, pruned_loss=0.07958, over 5698506.13 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3524, pruned_loss=0.1155, over 5690244.60 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2949, pruned_loss=0.07513, over 5694235.09 frames. ], batch size: 85, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:02:22,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.401e+02 1.130e+03 1.371e+03 1.912e+03 8.870e+03, threshold=2.743e+03, percent-clipped=9.0
+2023-03-07 00:02:54,186 INFO [train.py:968] (0/2) Epoch 13, batch 35600, giga_loss[loss=0.2867, simple_loss=0.3535, pruned_loss=0.11, over 28835.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3064, pruned_loss=0.08266, over 5686571.59 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.353, pruned_loss=0.1158, over 5683690.14 frames. ], giga_tot_loss[loss=0.2283, simple_loss=0.2999, pruned_loss=0.07829, over 5688636.87 frames. ], batch size: 199, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:03:00,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=583072.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:03:02,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4129, 1.5350, 1.3029, 1.5748], device='cuda:0'), covar=tensor([0.0784, 0.0336, 0.0341, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 00:03:03,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=583075.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:03:23,069 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-07 00:03:30,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=583104.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:03:43,494 INFO [train.py:968] (0/2) Epoch 13, batch 35650, giga_loss[loss=0.3226, simple_loss=0.3888, pruned_loss=0.1282, over 28307.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3193, pruned_loss=0.0899, over 5678867.31 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3534, pruned_loss=0.1162, over 5678851.06 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.3131, pruned_loss=0.08559, over 5684246.46 frames. ], batch size: 368, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:03:58,527 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.142e+02 1.281e+03 1.608e+03 2.180e+03 3.944e+03, threshold=3.217e+03, percent-clipped=15.0
+2023-03-07 00:04:20,686 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=583156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:04:28,543 INFO [train.py:968] (0/2) Epoch 13, batch 35700, giga_loss[loss=0.3639, simple_loss=0.4181, pruned_loss=0.1549, over 27947.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3333, pruned_loss=0.09732, over 5683589.40 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3537, pruned_loss=0.1162, over 5678136.49 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3278, pruned_loss=0.09358, over 5688825.00 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:04:30,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0161, 1.9211, 1.5115, 1.6303], device='cuda:0'), covar=tensor([0.0823, 0.0735, 0.0960, 0.1141], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0433, 0.0503, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 00:04:41,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6462, 1.8100, 1.8882, 1.4316], device='cuda:0'), covar=tensor([0.1579, 0.2324, 0.1276, 0.1650], device='cuda:0'), in_proj_covar=tensor([0.0848, 0.0687, 0.0888, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 00:05:13,115 INFO [train.py:968] (0/2) Epoch 13, batch 35750, giga_loss[loss=0.3023, simple_loss=0.3779, pruned_loss=0.1133, over 27913.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3413, pruned_loss=0.1007, over 5684437.70 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3533, pruned_loss=0.116, over 5680888.21 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3369, pruned_loss=0.09765, over 5686212.24 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:05:27,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.821e+02 1.303e+03 1.537e+03 1.897e+03 3.953e+03, threshold=3.075e+03, percent-clipped=4.0
+2023-03-07 00:05:32,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=583236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:05:32,996 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=583237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:05:41,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3478, 3.3415, 1.5576, 1.4036], device='cuda:0'), covar=tensor([0.0984, 0.0301, 0.0911, 0.1416], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0512, 0.0348, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 00:05:57,569 INFO [train.py:968] (0/2) Epoch 13, batch 35800, giga_loss[loss=0.2483, simple_loss=0.3318, pruned_loss=0.08236, over 29036.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3446, pruned_loss=0.1006, over 5672734.97 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3536, pruned_loss=0.1161, over 5671935.77 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3409, pruned_loss=0.09799, over 5682135.88 frames. ], batch size: 128, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:06:24,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9330, 1.1644, 1.2388, 1.0331], device='cuda:0'), covar=tensor([0.1393, 0.1226, 0.1907, 0.1428], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0725, 0.0673, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 00:06:29,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2893, 1.6177, 1.2731, 0.9452], device='cuda:0'), covar=tensor([0.2485, 0.2412, 0.2741, 0.2195], device='cuda:0'), in_proj_covar=tensor([0.1337, 0.0987, 0.1184, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-07 00:06:42,330 INFO [train.py:968] (0/2) Epoch 13, batch 35850, giga_loss[loss=0.2599, simple_loss=0.3424, pruned_loss=0.08873, over 28951.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3465, pruned_loss=0.1005, over 5675755.28 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.354, pruned_loss=0.1163, over 5668042.85 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3429, pruned_loss=0.09778, over 5686783.54 frames. ], batch size: 164, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:06:58,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.902e+02 1.143e+03 1.442e+03 1.999e+03 5.171e+03, threshold=2.883e+03, percent-clipped=7.0
+2023-03-07 00:07:03,552 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-07 00:07:24,567 INFO [train.py:968] (0/2) Epoch 13, batch 35900, giga_loss[loss=0.3038, simple_loss=0.3721, pruned_loss=0.1177, over 28934.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3494, pruned_loss=0.1021, over 5673337.79 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3545, pruned_loss=0.1165, over 5659190.49 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3459, pruned_loss=0.09939, over 5690208.39 frames. ], batch size: 213, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:07:36,740 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=583379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:07:39,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=583382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:08:06,102 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=583411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:08:08,670 INFO [train.py:968] (0/2) Epoch 13, batch 35950, giga_loss[loss=0.2766, simple_loss=0.3497, pruned_loss=0.1017, over 28864.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3523, pruned_loss=0.1049, over 5668899.46 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3549, pruned_loss=0.1166, over 5659647.11 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3491, pruned_loss=0.1024, over 5682056.95 frames. ], batch size: 112, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:08:21,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3600, 1.4977, 1.2727, 1.4680], device='cuda:0'), covar=tensor([0.0816, 0.0340, 0.0333, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 00:08:23,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.745e+02 1.231e+03 1.553e+03 2.208e+03 6.178e+03, threshold=3.106e+03, percent-clipped=16.0
+2023-03-07 00:08:32,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3284, 2.9947, 1.4775, 1.5029], device='cuda:0'), covar=tensor([0.0992, 0.0280, 0.0846, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0511, 0.0348, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 00:08:40,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4206, 1.6144, 1.1135, 1.1827], device='cuda:0'), covar=tensor([0.0872, 0.0493, 0.1051, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0434, 0.0505, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 00:08:51,862 INFO [train.py:968] (0/2) Epoch 13, batch 36000, giga_loss[loss=0.2795, simple_loss=0.3538, pruned_loss=0.1026, over 28808.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3552, pruned_loss=0.1065, over 5674949.56 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3548, pruned_loss=0.1166, over 5664415.28 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3526, pruned_loss=0.1043, over 5681234.66 frames. ], batch size: 199, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:08:51,866 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 00:08:57,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6297, 3.4486, 3.3451, 1.5142], device='cuda:0'), covar=tensor([0.0684, 0.0783, 0.0765, 0.2279], device='cuda:0'), in_proj_covar=tensor([0.1065, 0.0992, 0.0864, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 00:09:00,168 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2109, simple_loss=0.3184, pruned_loss=0.0517, over 944034.00 frames. 
+2023-03-07 00:09:00,168 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 00:09:38,296 INFO [train.py:968] (0/2) Epoch 13, batch 36050, giga_loss[loss=0.2731, simple_loss=0.3563, pruned_loss=0.09489, over 28773.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.358, pruned_loss=0.1074, over 5682981.02 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3553, pruned_loss=0.1168, over 5665191.83 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3556, pruned_loss=0.1052, over 5687597.17 frames. ], batch size: 119, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:09:52,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=583531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:09:52,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.703e+02 1.135e+03 1.543e+03 2.104e+03 6.363e+03, threshold=3.086e+03, percent-clipped=9.0
+2023-03-07 00:10:17,744 INFO [train.py:968] (0/2) Epoch 13, batch 36100, giga_loss[loss=0.2777, simple_loss=0.3523, pruned_loss=0.1016, over 28910.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3604, pruned_loss=0.1088, over 5687256.84 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3562, pruned_loss=0.1173, over 5676139.78 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3578, pruned_loss=0.1061, over 5681892.09 frames. ], batch size: 213, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:10:18,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-07 00:10:25,265 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3448, 1.4521, 1.3508, 1.5282], device='cuda:0'), covar=tensor([0.0712, 0.0432, 0.0324, 0.0763], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 00:10:56,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=583612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:10:58,283 INFO [train.py:968] (0/2) Epoch 13, batch 36150, giga_loss[loss=0.3062, simple_loss=0.3784, pruned_loss=0.117, over 28256.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3629, pruned_loss=0.11, over 5687463.20 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.357, pruned_loss=0.1178, over 5682619.25 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3603, pruned_loss=0.1072, over 5677816.89 frames. ], batch size: 368, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:11:10,143 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.512e+02 1.231e+03 1.625e+03 2.274e+03 9.090e+03, threshold=3.249e+03, percent-clipped=11.0
+2023-03-07 00:11:34,807 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=583659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:11:39,398 INFO [train.py:968] (0/2) Epoch 13, batch 36200, giga_loss[loss=0.2637, simple_loss=0.3445, pruned_loss=0.09146, over 27973.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.363, pruned_loss=0.1081, over 5696534.74 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3571, pruned_loss=0.1178, over 5683683.16 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3609, pruned_loss=0.1058, over 5688130.38 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:11:40,953 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6571, 1.8653, 1.4708, 2.0914], device='cuda:0'), covar=tensor([0.2443, 0.2475, 0.2732, 0.2269], device='cuda:0'), in_proj_covar=tensor([0.1345, 0.0992, 0.1188, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 00:11:46,701 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=583674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:11:48,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=583677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:12:11,206 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=583706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:12:18,156 INFO [train.py:968] (0/2) Epoch 13, batch 36250, giga_loss[loss=0.3022, simple_loss=0.3826, pruned_loss=0.1109, over 28837.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3612, pruned_loss=0.1058, over 5701027.20 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3575, pruned_loss=0.1179, over 5688473.52 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3593, pruned_loss=0.1036, over 5690271.61 frames. ], batch size: 119, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:12:32,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.395e+02 1.063e+03 1.425e+03 2.064e+03 6.652e+03, threshold=2.850e+03, percent-clipped=6.0
+2023-03-07 00:12:42,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5906, 2.3780, 1.5981, 0.8169], device='cuda:0'), covar=tensor([0.3800, 0.2041, 0.2765, 0.3574], device='cuda:0'), in_proj_covar=tensor([0.1601, 0.1527, 0.1507, 0.1319], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 00:12:51,511 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=583755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:12:53,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=583758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:12:58,384 INFO [train.py:968] (0/2) Epoch 13, batch 36300, giga_loss[loss=0.2651, simple_loss=0.3464, pruned_loss=0.09195, over 28715.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3588, pruned_loss=0.1035, over 5712655.66 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3575, pruned_loss=0.1176, over 5692952.22 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3573, pruned_loss=0.1017, over 5700243.31 frames. ], batch size: 78, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:13:16,649 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=583787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:13:26,276 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=583801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:13:37,521 INFO [train.py:968] (0/2) Epoch 13, batch 36350, giga_loss[loss=0.3017, simple_loss=0.3671, pruned_loss=0.1181, over 29096.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3585, pruned_loss=0.1035, over 5722159.06 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3579, pruned_loss=0.1177, over 5698753.47 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3569, pruned_loss=0.1016, over 5707749.46 frames. ], batch size: 128, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:13:49,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=583827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:13:52,344 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.991e+02 1.124e+03 1.499e+03 2.017e+03 7.961e+03, threshold=2.998e+03, percent-clipped=10.0
+2023-03-07 00:14:23,340 INFO [train.py:968] (0/2) Epoch 13, batch 36400, giga_loss[loss=0.2509, simple_loss=0.3354, pruned_loss=0.08322, over 28742.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3608, pruned_loss=0.1079, over 5706080.27 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3577, pruned_loss=0.1175, over 5693374.81 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3598, pruned_loss=0.1062, over 5700520.93 frames. ], batch size: 71, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:15:06,776 INFO [train.py:968] (0/2) Epoch 13, batch 36450, libri_loss[loss=0.3492, simple_loss=0.4089, pruned_loss=0.1447, over 29521.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3629, pruned_loss=0.1112, over 5702813.47 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3585, pruned_loss=0.1179, over 5697401.89 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3615, pruned_loss=0.1094, over 5694798.03 frames. ], batch size: 82, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:15:21,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.233e+02 1.323e+03 1.629e+03 2.119e+03 6.920e+03, threshold=3.257e+03, percent-clipped=12.0
+2023-03-07 00:15:50,701 INFO [train.py:968] (0/2) Epoch 13, batch 36500, giga_loss[loss=0.2411, simple_loss=0.318, pruned_loss=0.08214, over 28609.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3616, pruned_loss=0.1113, over 5709811.14 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3586, pruned_loss=0.1178, over 5701286.93 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3605, pruned_loss=0.1098, over 5700073.65 frames. ], batch size: 71, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:16:16,099 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4817, 2.7557, 1.7896, 2.1689], device='cuda:0'), covar=tensor([0.0706, 0.0494, 0.0951, 0.0925], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0432, 0.0504, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 00:16:16,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8979, 2.0805, 1.7164, 2.2999], device='cuda:0'), covar=tensor([0.2201, 0.2259, 0.2476, 0.2089], device='cuda:0'), in_proj_covar=tensor([0.1339, 0.0985, 0.1181, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-07 00:16:18,666 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-584000.pt
+2023-03-07 00:16:30,225 INFO [train.py:968] (0/2) Epoch 13, batch 36550, giga_loss[loss=0.2928, simple_loss=0.3537, pruned_loss=0.1159, over 28645.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3589, pruned_loss=0.1102, over 5707727.83 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.359, pruned_loss=0.118, over 5703913.96 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3577, pruned_loss=0.1086, over 5697772.40 frames. ], batch size: 92, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:16:30,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5049, 3.3384, 1.7393, 1.6303], device='cuda:0'), covar=tensor([0.0801, 0.0407, 0.0733, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0515, 0.0348, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 00:16:46,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.694e+02 1.226e+03 1.437e+03 1.948e+03 5.012e+03, threshold=2.874e+03, percent-clipped=5.0
+2023-03-07 00:16:47,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:17:01,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3294, 1.6520, 1.3258, 1.2866], device='cuda:0'), covar=tensor([0.2104, 0.1954, 0.2126, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.1340, 0.0984, 0.1181, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:0')
+2023-03-07 00:17:11,814 INFO [train.py:968] (0/2) Epoch 13, batch 36600, libri_loss[loss=0.2435, simple_loss=0.31, pruned_loss=0.08854, over 28181.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3581, pruned_loss=0.1095, over 5697626.94 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3592, pruned_loss=0.1183, over 5694888.26 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3569, pruned_loss=0.1077, over 5697875.40 frames. ], batch size: 62, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:17:55,693 INFO [train.py:968] (0/2) Epoch 13, batch 36650, giga_loss[loss=0.2598, simple_loss=0.3384, pruned_loss=0.09065, over 28969.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3567, pruned_loss=0.1077, over 5694313.26 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3595, pruned_loss=0.1185, over 5698943.47 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3555, pruned_loss=0.106, over 5690995.23 frames. ], batch size: 145, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:18:13,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.670e+02 1.243e+03 1.551e+03 2.031e+03 7.667e+03, threshold=3.101e+03, percent-clipped=8.0
+2023-03-07 00:18:33,446 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9577, 1.3086, 1.0258, 0.1974], device='cuda:0'), covar=tensor([0.3233, 0.2574, 0.3991, 0.5158], device='cuda:0'), in_proj_covar=tensor([0.1602, 0.1529, 0.1511, 0.1319], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 00:18:34,388 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-07 00:18:38,581 INFO [train.py:968] (0/2) Epoch 13, batch 36700, giga_loss[loss=0.245, simple_loss=0.328, pruned_loss=0.08101, over 28949.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3558, pruned_loss=0.107, over 5692754.37 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3598, pruned_loss=0.1182, over 5702971.12 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3543, pruned_loss=0.1053, over 5686182.98 frames. ], batch size: 213, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:18:48,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:18:49,471 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:18:51,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:19:09,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:19:17,371 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:19:21,878 INFO [train.py:968] (0/2) Epoch 13, batch 36750, giga_loss[loss=0.2468, simple_loss=0.325, pruned_loss=0.08436, over 29087.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3492, pruned_loss=0.1034, over 5672387.93 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3595, pruned_loss=0.1179, over 5695838.25 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3482, pruned_loss=0.1021, over 5672532.38 frames. ], batch size: 136, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:19:28,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5307, 1.5466, 1.2534, 1.1052], device='cuda:0'), covar=tensor([0.0902, 0.0576, 0.1107, 0.1101], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0430, 0.0502, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 00:19:38,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.530e+02 1.107e+03 1.545e+03 2.236e+03 1.634e+04, threshold=3.090e+03, percent-clipped=15.0
+2023-03-07 00:20:06,672 INFO [train.py:968] (0/2) Epoch 13, batch 36800, giga_loss[loss=0.2193, simple_loss=0.3023, pruned_loss=0.06813, over 28899.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3442, pruned_loss=0.1011, over 5670684.25 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3593, pruned_loss=0.1175, over 5700949.07 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3431, pruned_loss=0.09963, over 5664881.39 frames. ], batch size: 227, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:20:31,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4924, 4.3453, 4.1386, 2.0660], device='cuda:0'), covar=tensor([0.0519, 0.0631, 0.0628, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.1069, 0.0995, 0.0866, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 00:20:35,875 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4257, 3.2198, 1.6010, 1.5635], device='cuda:0'), covar=tensor([0.0957, 0.0268, 0.0855, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0511, 0.0346, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 00:20:58,109 INFO [train.py:968] (0/2) Epoch 13, batch 36850, giga_loss[loss=0.2568, simple_loss=0.3362, pruned_loss=0.08871, over 27965.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3405, pruned_loss=0.09945, over 5660230.93 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3602, pruned_loss=0.1181, over 5701433.72 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3382, pruned_loss=0.0973, over 5654052.33 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:21:03,328 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:06,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2664, 1.5259, 1.6411, 1.3292], device='cuda:0'), covar=tensor([0.1347, 0.1327, 0.1713, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0732, 0.0676, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 00:21:06,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:16,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.037e+02 9.399e+02 1.194e+03 1.609e+03 5.081e+03, threshold=2.389e+03, percent-clipped=4.0
+2023-03-07 00:21:21,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5319, 1.5814, 1.5640, 1.3901], device='cuda:0'), covar=tensor([0.2159, 0.1943, 0.1712, 0.2055], device='cuda:0'), in_proj_covar=tensor([0.1749, 0.1646, 0.1613, 0.1712], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 00:21:27,086 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:29,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:31,408 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:34,992 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-07 00:21:40,560 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:43,922 INFO [train.py:968] (0/2) Epoch 13, batch 36900, giga_loss[loss=0.2331, simple_loss=0.3188, pruned_loss=0.07373, over 29011.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3386, pruned_loss=0.09725, over 5671015.60 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3597, pruned_loss=0.1176, over 5706139.06 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3366, pruned_loss=0.09542, over 5660549.94 frames. ], batch size: 155, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:21:49,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:21:53,463 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:53,534 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:22:25,325 INFO [train.py:968] (0/2) Epoch 13, batch 36950, libri_loss[loss=0.2737, simple_loss=0.3464, pruned_loss=0.1005, over 29585.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.338, pruned_loss=0.09659, over 5675834.65 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3595, pruned_loss=0.1174, over 5709261.73 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3362, pruned_loss=0.09491, over 5664003.90 frames. ], batch size: 74, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:22:40,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.096e+02 1.042e+03 1.287e+03 1.884e+03 4.027e+03, threshold=2.573e+03, percent-clipped=10.0
+2023-03-07 00:23:08,148 INFO [train.py:968] (0/2) Epoch 13, batch 37000, giga_loss[loss=0.2366, simple_loss=0.3092, pruned_loss=0.08204, over 28596.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3363, pruned_loss=0.0956, over 5683450.16 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3596, pruned_loss=0.1174, over 5702223.97 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3346, pruned_loss=0.09404, over 5679637.02 frames. ], batch size: 60, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:23:13,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2893, 1.2538, 3.9358, 3.2164], device='cuda:0'), covar=tensor([0.1666, 0.2676, 0.0404, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0679, 0.0597, 0.0871, 0.0786], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-07 00:23:22,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3901, 1.6331, 1.6461, 1.2496], device='cuda:0'), covar=tensor([0.1685, 0.2341, 0.1343, 0.1510], device='cuda:0'), in_proj_covar=tensor([0.0846, 0.0686, 0.0885, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 00:23:30,743 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3799, 1.6619, 1.3220, 1.2232], device='cuda:0'), covar=tensor([0.2432, 0.2437, 0.2748, 0.2295], device='cuda:0'), in_proj_covar=tensor([0.1347, 0.0990, 0.1189, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 00:23:48,055 INFO [train.py:968] (0/2) Epoch 13, batch 37050, giga_loss[loss=0.2554, simple_loss=0.3323, pruned_loss=0.08927, over 28837.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3351, pruned_loss=0.09533, over 5700316.64 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3605, pruned_loss=0.1178, over 5706645.75 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3323, pruned_loss=0.09322, over 5693113.19 frames. ], batch size: 174, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:24:00,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.391e+02 1.075e+03 1.396e+03 1.954e+03 7.158e+03, threshold=2.792e+03, percent-clipped=13.0
+2023-03-07 00:24:28,296 INFO [train.py:968] (0/2) Epoch 13, batch 37100, giga_loss[loss=0.233, simple_loss=0.3171, pruned_loss=0.07446, over 28894.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3332, pruned_loss=0.0942, over 5704173.53 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3611, pruned_loss=0.1178, over 5710478.87 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.33, pruned_loss=0.09207, over 5695104.50 frames. ], batch size: 174, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:24:53,698 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:25:07,673 INFO [train.py:968] (0/2) Epoch 13, batch 37150, giga_loss[loss=0.2872, simple_loss=0.3501, pruned_loss=0.1122, over 28862.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3299, pruned_loss=0.09275, over 5710356.93 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3608, pruned_loss=0.1175, over 5712328.21 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3272, pruned_loss=0.09099, over 5701566.14 frames. ], batch size: 99, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:25:23,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.418e+02 9.556e+02 1.189e+03 1.560e+03 5.178e+03, threshold=2.377e+03, percent-clipped=2.0
+2023-03-07 00:25:50,626 INFO [train.py:968] (0/2) Epoch 13, batch 37200, giga_loss[loss=0.2793, simple_loss=0.3408, pruned_loss=0.1089, over 28858.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3276, pruned_loss=0.09175, over 5714800.28 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.361, pruned_loss=0.1176, over 5713283.57 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3252, pruned_loss=0.0902, over 5707162.70 frames. ], batch size: 186, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:25:54,185 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5907, 1.6965, 1.6283, 1.5103], device='cuda:0'), covar=tensor([0.1685, 0.2008, 0.2253, 0.2076], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0736, 0.0683, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 00:26:30,847 INFO [train.py:968] (0/2) Epoch 13, batch 37250, giga_loss[loss=0.2841, simple_loss=0.3348, pruned_loss=0.1167, over 28726.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3254, pruned_loss=0.09064, over 5721027.73 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3618, pruned_loss=0.1178, over 5717068.91 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3222, pruned_loss=0.08872, over 5711783.00 frames. ], batch size: 92, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:26:37,814 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:26:46,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.319e+02 1.034e+03 1.272e+03 1.780e+03 4.373e+03, threshold=2.544e+03, percent-clipped=8.0
+2023-03-07 00:26:49,277 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:26:56,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584747.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:26:59,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:27:08,969 INFO [train.py:968] (0/2) Epoch 13, batch 37300, giga_loss[loss=0.2491, simple_loss=0.3195, pruned_loss=0.08939, over 28819.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3257, pruned_loss=0.09094, over 5713868.67 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3638, pruned_loss=0.119, over 5711412.39 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3204, pruned_loss=0.08769, over 5711311.54 frames. ], batch size: 119, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:27:44,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4633, 1.4044, 1.1803, 1.1185], device='cuda:0'), covar=tensor([0.0657, 0.0370, 0.0857, 0.0990], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0435, 0.0506, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 00:27:49,114 INFO [train.py:968] (0/2) Epoch 13, batch 37350, giga_loss[loss=0.2476, simple_loss=0.3174, pruned_loss=0.08893, over 28851.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3227, pruned_loss=0.08925, over 5715607.62 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3633, pruned_loss=0.1186, over 5714314.48 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3185, pruned_loss=0.08663, over 5711275.76 frames. ], batch size: 199, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:27:54,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:28:05,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.088e+02 9.681e+02 1.205e+03 1.559e+03 6.114e+03, threshold=2.411e+03, percent-clipped=7.0
+2023-03-07 00:28:27,886 INFO [train.py:968] (0/2) Epoch 13, batch 37400, libri_loss[loss=0.3759, simple_loss=0.428, pruned_loss=0.1619, over 29526.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3234, pruned_loss=0.08986, over 5704488.76 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3647, pruned_loss=0.1194, over 5708725.83 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3178, pruned_loss=0.08635, over 5706377.19 frames. ], batch size: 80, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:28:40,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:28:43,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:28:48,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584890.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:28:50,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584893.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:28:51,685 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:28:53,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:29:04,384 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:29:07,273 INFO [train.py:968] (0/2) Epoch 13, batch 37450, giga_loss[loss=0.234, simple_loss=0.3127, pruned_loss=0.07768, over 28887.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3246, pruned_loss=0.09037, over 5708631.41 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3653, pruned_loss=0.1195, over 5705166.95 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3183, pruned_loss=0.08662, over 5712716.78 frames. ], batch size: 174, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:29:12,582 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584922.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:29:16,719 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:29:23,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.791e+02 1.051e+03 1.433e+03 1.903e+03 1.162e+04, threshold=2.865e+03, percent-clipped=16.0
+2023-03-07 00:29:26,540 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-07 00:29:42,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584955.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:29:49,738 INFO [train.py:968] (0/2) Epoch 13, batch 37500, libri_loss[loss=0.2616, simple_loss=0.3414, pruned_loss=0.09089, over 29556.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3299, pruned_loss=0.09392, over 5711403.08 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3651, pruned_loss=0.1193, over 5704839.07 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3246, pruned_loss=0.09077, over 5714808.11 frames. ], batch size: 75, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:29:55,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:30:16,454 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 00:30:18,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4097, 3.6796, 1.5663, 1.5868], device='cuda:0'), covar=tensor([0.0872, 0.0297, 0.0815, 0.1216], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0512, 0.0347, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 00:30:33,026 INFO [train.py:968] (0/2) Epoch 13, batch 37550, giga_loss[loss=0.2911, simple_loss=0.353, pruned_loss=0.1146, over 28586.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3373, pruned_loss=0.09876, over 5702758.07 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3653, pruned_loss=0.1192, over 5702343.19 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3316, pruned_loss=0.09533, over 5707785.07 frames. ], batch size: 60, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:30:43,964 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=585025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:30:53,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.734e+02 1.245e+03 1.523e+03 2.173e+03 8.851e+03, threshold=3.046e+03, percent-clipped=14.0
+2023-03-07 00:31:19,969 INFO [train.py:968] (0/2) Epoch 13, batch 37600, libri_loss[loss=0.2777, simple_loss=0.3408, pruned_loss=0.1073, over 29664.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3467, pruned_loss=0.1052, over 5699277.78 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3661, pruned_loss=0.1198, over 5708317.27 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3407, pruned_loss=0.1015, over 5697800.81 frames. ], batch size: 73, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:31:20,364 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-07 00:31:52,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=585098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:32:10,851 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=585114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:32:11,269 INFO [train.py:968] (0/2) Epoch 13, batch 37650, libri_loss[loss=0.2663, simple_loss=0.3294, pruned_loss=0.1016, over 29659.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3507, pruned_loss=0.107, over 5689291.81 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.366, pruned_loss=0.1197, over 5710393.31 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3458, pruned_loss=0.1038, over 5685898.43 frames. ], batch size: 69, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:32:12,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=585117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:32:13,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3069, 1.5533, 1.2733, 1.4793], device='cuda:0'), covar=tensor([0.0761, 0.0398, 0.0346, 0.0793], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 00:32:28,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.234e+02 1.247e+03 1.642e+03 2.290e+03 4.797e+03, threshold=3.285e+03, percent-clipped=7.0
+2023-03-07 00:32:28,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6805, 3.4906, 3.2992, 1.7689], device='cuda:0'), covar=tensor([0.0716, 0.0839, 0.0765, 0.2506], device='cuda:0'), in_proj_covar=tensor([0.1076, 0.0997, 0.0869, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 00:32:37,109 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=585146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:32:54,759 INFO [train.py:968] (0/2) Epoch 13, batch 37700, giga_loss[loss=0.3963, simple_loss=0.424, pruned_loss=0.1843, over 23722.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3553, pruned_loss=0.1085, over 5685986.95 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.366, pruned_loss=0.1199, over 5706730.43 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3509, pruned_loss=0.1056, over 5685444.50 frames. ], batch size: 705, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:33:22,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=585196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:33:39,121 INFO [train.py:968] (0/2) Epoch 13, batch 37750, giga_loss[loss=0.2922, simple_loss=0.3693, pruned_loss=0.1075, over 28720.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3602, pruned_loss=0.1109, over 5674771.79 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3659, pruned_loss=0.1197, over 5699018.94 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3567, pruned_loss=0.1084, over 5679965.08 frames. ], batch size: 119, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:33:56,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.501e+02 1.115e+03 1.394e+03 1.846e+03 5.973e+03, threshold=2.789e+03, percent-clipped=3.0
+2023-03-07 00:34:02,778 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=585241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:34:05,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=585244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:34:22,006 INFO [train.py:968] (0/2) Epoch 13, batch 37800, giga_loss[loss=0.3086, simple_loss=0.3667, pruned_loss=0.1252, over 28900.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3605, pruned_loss=0.1104, over 5680918.58 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3662, pruned_loss=0.1199, over 5701365.25 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3574, pruned_loss=0.1082, over 5682638.11 frames. ], batch size: 213, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:34:29,053 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=585273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:35:04,018 INFO [train.py:968] (0/2) Epoch 13, batch 37850, giga_loss[loss=0.2527, simple_loss=0.3374, pruned_loss=0.08393, over 28875.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3562, pruned_loss=0.1068, over 5690496.11 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3662, pruned_loss=0.1199, over 5704139.52 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3536, pruned_loss=0.1049, over 5689294.13 frames. ], batch size: 174, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:35:17,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=585330.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:35:24,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.376e+02 1.079e+03 1.424e+03 2.060e+03 6.210e+03, threshold=2.849e+03, percent-clipped=16.0
+2023-03-07 00:35:27,103 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=585339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:35:28,917 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=585342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:35:45,929 INFO [train.py:968] (0/2) Epoch 13, batch 37900, giga_loss[loss=0.216, simple_loss=0.3082, pruned_loss=0.06189, over 28961.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3541, pruned_loss=0.105, over 5682909.03 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3668, pruned_loss=0.1206, over 5693420.34 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3512, pruned_loss=0.1025, over 5690661.85 frames. ], batch size: 164, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:35:51,392 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=585371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:36:16,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=585400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:36:27,551 INFO [train.py:968] (0/2) Epoch 13, batch 37950, giga_loss[loss=0.3018, simple_loss=0.3724, pruned_loss=0.1156, over 28690.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3547, pruned_loss=0.105, over 5692505.36 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3667, pruned_loss=0.1205, over 5695793.40 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3522, pruned_loss=0.1028, over 5696387.99 frames. ], batch size: 307, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:36:45,999 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-07 00:36:46,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.424e+02 1.299e+03 1.663e+03 2.320e+03 7.500e+03, threshold=3.325e+03, percent-clipped=13.0
+2023-03-07 00:37:10,119 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-07 00:37:13,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-07 00:37:13,707 INFO [train.py:968] (0/2) Epoch 13, batch 38000, giga_loss[loss=0.2634, simple_loss=0.3502, pruned_loss=0.08829, over 28704.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3573, pruned_loss=0.1063, over 5691143.83 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3667, pruned_loss=0.1205, over 5695793.40 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3554, pruned_loss=0.1046, over 5694165.73 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:37:20,518 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=585473.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:37:22,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=585476.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:37:47,276 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=585505.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:37:55,417 INFO [train.py:968] (0/2) Epoch 13, batch 38050, giga_loss[loss=0.3132, simple_loss=0.3706, pruned_loss=0.1279, over 23645.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3607, pruned_loss=0.109, over 5692349.26 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3677, pruned_loss=0.1211, over 5694705.69 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3579, pruned_loss=0.1065, over 5694878.90 frames. ], batch size: 705, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:38:15,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.769e+02 1.276e+03 1.577e+03 2.092e+03 5.388e+03, threshold=3.155e+03, percent-clipped=4.0
+2023-03-07 00:38:20,967 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=585543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:38:24,323 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=585546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:38:40,136 INFO [train.py:968] (0/2) Epoch 13, batch 38100, giga_loss[loss=0.3017, simple_loss=0.3704, pruned_loss=0.1165, over 28930.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3623, pruned_loss=0.1103, over 5692119.26 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3681, pruned_loss=0.1212, over 5697098.17 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3597, pruned_loss=0.108, over 5691984.40 frames. ], batch size: 227, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:38:49,468 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=585575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:39:23,438 INFO [train.py:968] (0/2) Epoch 13, batch 38150, giga_loss[loss=0.2711, simple_loss=0.3503, pruned_loss=0.09598, over 28951.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.362, pruned_loss=0.1106, over 5694422.61 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3678, pruned_loss=0.121, over 5696309.29 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3601, pruned_loss=0.1088, over 5695096.70 frames. ], batch size: 164, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:39:43,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.710e+02 1.243e+03 1.565e+03 2.157e+03 4.414e+03, threshold=3.131e+03, percent-clipped=8.0
+2023-03-07 00:40:05,487 INFO [train.py:968] (0/2) Epoch 13, batch 38200, giga_loss[loss=0.2919, simple_loss=0.3472, pruned_loss=0.1183, over 23824.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3624, pruned_loss=0.1114, over 5680400.05 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.368, pruned_loss=0.1212, over 5688661.26 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3606, pruned_loss=0.1097, over 5687858.68 frames. ], batch size: 705, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:40:46,598 INFO [train.py:968] (0/2) Epoch 13, batch 38250, giga_loss[loss=0.2958, simple_loss=0.3625, pruned_loss=0.1145, over 28334.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3623, pruned_loss=0.1107, over 5688294.35 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3683, pruned_loss=0.1213, over 5689012.06 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3605, pruned_loss=0.1091, over 5693593.02 frames. ], batch size: 77, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:41:03,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.888e+02 1.077e+03 1.299e+03 1.659e+03 3.264e+03, threshold=2.598e+03, percent-clipped=1.0
+2023-03-07 00:41:26,138 INFO [train.py:968] (0/2) Epoch 13, batch 38300, giga_loss[loss=0.2459, simple_loss=0.3296, pruned_loss=0.08104, over 28348.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.361, pruned_loss=0.1083, over 5694529.89 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3685, pruned_loss=0.1211, over 5692002.43 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3594, pruned_loss=0.107, over 5696099.39 frames. ], batch size: 77, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:42:05,586 INFO [train.py:968] (0/2) Epoch 13, batch 38350, giga_loss[loss=0.3045, simple_loss=0.3672, pruned_loss=0.1208, over 28758.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3618, pruned_loss=0.1081, over 5707080.03 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3689, pruned_loss=0.1214, over 5696922.42 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3599, pruned_loss=0.1065, over 5704263.18 frames. ], batch size: 92, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:42:24,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.323e+02 1.079e+03 1.312e+03 1.741e+03 7.074e+03, threshold=2.624e+03, percent-clipped=11.0
+2023-03-07 00:42:41,294 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-07 00:42:46,204 INFO [train.py:968] (0/2) Epoch 13, batch 38400, giga_loss[loss=0.2745, simple_loss=0.3471, pruned_loss=0.101, over 28848.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3597, pruned_loss=0.1074, over 5707926.14 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1214, over 5703186.00 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3581, pruned_loss=0.1056, over 5700207.24 frames. ], batch size: 99, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:43:05,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3783, 1.5548, 1.3643, 1.2004], device='cuda:0'), covar=tensor([0.1941, 0.1867, 0.1425, 0.1899], device='cuda:0'), in_proj_covar=tensor([0.1740, 0.1638, 0.1620, 0.1708], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 00:43:26,210 INFO [train.py:968] (0/2) Epoch 13, batch 38450, giga_loss[loss=0.2351, simple_loss=0.3157, pruned_loss=0.07728, over 28604.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3573, pruned_loss=0.106, over 5708953.14 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3681, pruned_loss=0.1209, over 5707035.55 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3562, pruned_loss=0.1047, over 5699606.42 frames. ], batch size: 66, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:43:44,958 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.627e+02 1.072e+03 1.382e+03 1.988e+03 6.314e+03, threshold=2.764e+03, percent-clipped=13.0
+2023-03-07 00:44:06,276 INFO [train.py:968] (0/2) Epoch 13, batch 38500, libri_loss[loss=0.2968, simple_loss=0.3637, pruned_loss=0.115, over 29363.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3547, pruned_loss=0.1044, over 5711457.66 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3682, pruned_loss=0.121, over 5709879.77 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3533, pruned_loss=0.1027, over 5701486.10 frames. ], batch size: 92, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:44:34,706 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-586000.pt
+2023-03-07 00:44:47,207 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=586013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:44:48,276 INFO [train.py:968] (0/2) Epoch 13, batch 38550, giga_loss[loss=0.2361, simple_loss=0.3138, pruned_loss=0.07919, over 28304.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3541, pruned_loss=0.1046, over 5709169.74 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3686, pruned_loss=0.1213, over 5712256.76 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3524, pruned_loss=0.1026, over 5699311.89 frames. ], batch size: 77, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:45:02,272 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=586031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:45:06,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.826e+02 1.048e+03 1.367e+03 1.833e+03 7.218e+03, threshold=2.734e+03, percent-clipped=10.0
+2023-03-07 00:45:26,492 INFO [train.py:968] (0/2) Epoch 13, batch 38600, giga_loss[loss=0.2769, simple_loss=0.3537, pruned_loss=0.1001, over 28898.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3542, pruned_loss=0.1046, over 5713358.13 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3682, pruned_loss=0.1212, over 5717051.02 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3529, pruned_loss=0.1028, over 5700991.87 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:45:47,161 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=586091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:46:04,947 INFO [train.py:968] (0/2) Epoch 13, batch 38650, giga_loss[loss=0.2738, simple_loss=0.3549, pruned_loss=0.0964, over 28538.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3542, pruned_loss=0.1041, over 5708087.87 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3685, pruned_loss=0.1215, over 5709873.49 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3528, pruned_loss=0.1022, over 5704054.70 frames. ], batch size: 78, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:46:22,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.192e+02 1.031e+03 1.306e+03 1.868e+03 6.251e+03, threshold=2.611e+03, percent-clipped=9.0
+2023-03-07 00:46:36,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9279, 2.8378, 1.9328, 1.0761], device='cuda:0'), covar=tensor([0.5762, 0.2126, 0.3076, 0.5303], device='cuda:0'), in_proj_covar=tensor([0.1587, 0.1493, 0.1493, 0.1310], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 00:46:37,908 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=586159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:46:39,002 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-07 00:46:43,129 INFO [train.py:968] (0/2) Epoch 13, batch 38700, giga_loss[loss=0.2678, simple_loss=0.3437, pruned_loss=0.09593, over 28645.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3533, pruned_loss=0.103, over 5714203.71 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3687, pruned_loss=0.1215, over 5711330.65 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3518, pruned_loss=0.1011, over 5709750.51 frames. ], batch size: 85, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:47:21,224 INFO [train.py:968] (0/2) Epoch 13, batch 38750, giga_loss[loss=0.2477, simple_loss=0.3319, pruned_loss=0.08177, over 28596.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3534, pruned_loss=0.1031, over 5717925.68 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3681, pruned_loss=0.1211, over 5715346.88 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3524, pruned_loss=0.1017, over 5710802.83 frames. ], batch size: 60, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:47:42,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.804e+02 9.902e+02 1.255e+03 1.781e+03 8.703e+03, threshold=2.509e+03, percent-clipped=12.0
+2023-03-07 00:48:06,176 INFO [train.py:968] (0/2) Epoch 13, batch 38800, giga_loss[loss=0.2675, simple_loss=0.3385, pruned_loss=0.09822, over 28428.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3511, pruned_loss=0.1023, over 5708579.91 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3679, pruned_loss=0.121, over 5716305.14 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3504, pruned_loss=0.1012, over 5702195.45 frames. ], batch size: 65, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:48:47,100 INFO [train.py:968] (0/2) Epoch 13, batch 38850, giga_loss[loss=0.2702, simple_loss=0.3447, pruned_loss=0.09785, over 28575.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3481, pruned_loss=0.1009, over 5710713.53 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3677, pruned_loss=0.1208, over 5719110.21 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3475, pruned_loss=0.09996, over 5703198.48 frames. ], batch size: 307, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:48:51,039 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-07 00:48:58,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1069, 1.0668, 3.9356, 3.1428], device='cuda:0'), covar=tensor([0.1702, 0.2814, 0.0402, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0674, 0.0593, 0.0864, 0.0785], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-07 00:49:05,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.044e+02 1.100e+03 1.283e+03 1.548e+03 3.826e+03, threshold=2.567e+03, percent-clipped=5.0
+2023-03-07 00:49:17,195 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2347, 3.0612, 1.3568, 1.3933], device='cuda:0'), covar=tensor([0.0990, 0.0261, 0.0867, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0507, 0.0344, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:0')
+2023-03-07 00:49:24,751 INFO [train.py:968] (0/2) Epoch 13, batch 38900, libri_loss[loss=0.2947, simple_loss=0.367, pruned_loss=0.1112, over 29560.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3456, pruned_loss=0.0997, over 5713805.39 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3679, pruned_loss=0.121, over 5721588.32 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3447, pruned_loss=0.0985, over 5705399.61 frames. ], batch size: 89, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:49:43,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=586388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:49:58,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=586406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:50:07,132 INFO [train.py:968] (0/2) Epoch 13, batch 38950, giga_loss[loss=0.2429, simple_loss=0.3174, pruned_loss=0.08418, over 28567.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3449, pruned_loss=0.09975, over 5708302.57 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3678, pruned_loss=0.1209, over 5719945.02 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3438, pruned_loss=0.09842, over 5702674.53 frames. ], batch size: 85, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:50:25,811 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.477e+02 1.129e+03 1.529e+03 2.557e+03 1.311e+04, threshold=3.059e+03, percent-clipped=24.0
+2023-03-07 00:50:46,454 INFO [train.py:968] (0/2) Epoch 13, batch 39000, giga_loss[loss=0.2617, simple_loss=0.3365, pruned_loss=0.09344, over 29071.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3439, pruned_loss=0.09956, over 5712909.27 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.368, pruned_loss=0.121, over 5721622.13 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3424, pruned_loss=0.09809, over 5706885.13 frames. ], batch size: 128, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:50:46,458 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 00:50:54,792 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2159, simple_loss=0.3232, pruned_loss=0.05425, over 944034.00 frames. 
+2023-03-07 00:50:54,793 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 00:50:55,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=586466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:51:34,706 INFO [train.py:968] (0/2) Epoch 13, batch 39050, giga_loss[loss=0.2708, simple_loss=0.3394, pruned_loss=0.1011, over 28444.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3425, pruned_loss=0.09948, over 5712622.66 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.368, pruned_loss=0.1209, over 5725309.63 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3409, pruned_loss=0.09803, over 5704692.44 frames. ], batch size: 71, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:51:47,951 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=586531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:51:49,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=586534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:51:49,634 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=586534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:51:52,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.266e+02 1.050e+03 1.349e+03 1.928e+03 5.040e+03, threshold=2.697e+03, percent-clipped=8.0
+2023-03-07 00:52:00,795 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=586549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:02,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=586552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:06,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5255, 3.2560, 1.5294, 1.6887], device='cuda:0'), covar=tensor([0.0822, 0.0328, 0.0872, 0.1137], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0511, 0.0344, 0.0393], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0021, 0.0026], device='cuda:0')
+2023-03-07 00:52:10,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:12,175 INFO [train.py:968] (0/2) Epoch 13, batch 39100, giga_loss[loss=0.2335, simple_loss=0.3086, pruned_loss=0.07923, over 28270.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3412, pruned_loss=0.09948, over 5711884.39 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1213, over 5727121.77 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3389, pruned_loss=0.09749, over 5703796.09 frames. ], batch size: 77, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:52:27,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:36,452 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-07 00:52:49,682 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=586609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:53,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=586612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:54,710 INFO [train.py:968] (0/2) Epoch 13, batch 39150, giga_loss[loss=0.2655, simple_loss=0.3382, pruned_loss=0.09641, over 28684.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3403, pruned_loss=0.09911, over 5710781.76 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.369, pruned_loss=0.1214, over 5726589.83 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3371, pruned_loss=0.0967, over 5704282.05 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:53:15,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.515e+02 1.112e+03 1.355e+03 1.928e+03 9.840e+03, threshold=2.710e+03, percent-clipped=10.0
+2023-03-07 00:53:17,312 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:53:30,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3005, 1.5617, 1.3170, 1.5118], device='cuda:0'), covar=tensor([0.0723, 0.0296, 0.0321, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0113, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-07 00:53:35,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1926, 1.1091, 3.8659, 3.0869], device='cuda:0'), covar=tensor([0.1734, 0.2897, 0.0405, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0674, 0.0594, 0.0864, 0.0784], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-07 00:53:37,765 INFO [train.py:968] (0/2) Epoch 13, batch 39200, giga_loss[loss=0.2503, simple_loss=0.3239, pruned_loss=0.08832, over 28860.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3411, pruned_loss=0.09999, over 5713084.55 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3691, pruned_loss=0.1215, over 5728423.26 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3383, pruned_loss=0.09784, over 5706286.71 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:53:48,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=586677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:53:49,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=586680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:54:14,030 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=586707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:54:15,681 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:54:20,267 INFO [train.py:968] (0/2) Epoch 13, batch 39250, giga_loss[loss=0.2525, simple_loss=0.3225, pruned_loss=0.09127, over 28762.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3433, pruned_loss=0.1001, over 5711573.19 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3695, pruned_loss=0.1217, over 5728524.96 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3402, pruned_loss=0.09777, over 5705902.33 frames. ], batch size: 66, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:54:42,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.872e+02 1.058e+03 1.307e+03 1.905e+03 5.103e+03, threshold=2.615e+03, percent-clipped=11.0
+2023-03-07 00:55:06,364 INFO [train.py:968] (0/2) Epoch 13, batch 39300, giga_loss[loss=0.2867, simple_loss=0.3647, pruned_loss=0.1044, over 28682.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3467, pruned_loss=0.1017, over 5712920.18 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3692, pruned_loss=0.1217, over 5731866.63 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3438, pruned_loss=0.09934, over 5705150.57 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:55:28,582 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-07 00:55:36,480 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-07 00:55:45,511 INFO [train.py:968] (0/2) Epoch 13, batch 39350, giga_loss[loss=0.2609, simple_loss=0.3469, pruned_loss=0.08745, over 28641.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3495, pruned_loss=0.1028, over 5704236.41 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3697, pruned_loss=0.1221, over 5725278.31 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3459, pruned_loss=0.09973, over 5702717.84 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:56:07,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.925e+02 1.030e+03 1.233e+03 1.693e+03 4.401e+03, threshold=2.466e+03, percent-clipped=6.0
+2023-03-07 00:56:08,275 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-07 00:56:29,687 INFO [train.py:968] (0/2) Epoch 13, batch 39400, giga_loss[loss=0.2244, simple_loss=0.3122, pruned_loss=0.06837, over 28940.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3486, pruned_loss=0.1017, over 5697905.67 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3698, pruned_loss=0.1223, over 5729055.29 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3452, pruned_loss=0.09864, over 5692549.95 frames. ], batch size: 145, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:57:10,720 INFO [train.py:968] (0/2) Epoch 13, batch 39450, giga_loss[loss=0.2832, simple_loss=0.3642, pruned_loss=0.1011, over 29038.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3484, pruned_loss=0.101, over 5703309.42 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3698, pruned_loss=0.1225, over 5733772.26 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3452, pruned_loss=0.0979, over 5694088.84 frames. ], batch size: 164, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:57:13,417 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-07 00:57:30,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.201e+02 1.096e+03 1.465e+03 2.178e+03 8.384e+03, threshold=2.930e+03, percent-clipped=17.0
+2023-03-07 00:57:46,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4041, 2.0902, 1.5586, 0.7412], device='cuda:0'), covar=tensor([0.4370, 0.2020, 0.3326, 0.4734], device='cuda:0'), in_proj_covar=tensor([0.1590, 0.1496, 0.1494, 0.1312], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 00:57:52,201 INFO [train.py:968] (0/2) Epoch 13, batch 39500, giga_loss[loss=0.2458, simple_loss=0.3246, pruned_loss=0.08351, over 28384.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3485, pruned_loss=0.1012, over 5687453.66 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3701, pruned_loss=0.1226, over 5718010.73 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3452, pruned_loss=0.09821, over 5693524.66 frames. ], batch size: 78, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:58:35,833 INFO [train.py:968] (0/2) Epoch 13, batch 39550, giga_loss[loss=0.2402, simple_loss=0.3179, pruned_loss=0.08125, over 28456.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3486, pruned_loss=0.1016, over 5679019.47 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3703, pruned_loss=0.1226, over 5709590.76 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3457, pruned_loss=0.09899, over 5689946.83 frames. ], batch size: 60, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:58:45,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5102, 1.4245, 1.6100, 1.1640], device='cuda:0'), covar=tensor([0.1995, 0.3081, 0.1657, 0.1796], device='cuda:0'), in_proj_covar=tensor([0.0845, 0.0687, 0.0884, 0.0790], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-07 00:58:56,930 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.996e+02 1.228e+03 1.593e+03 2.128e+03 8.635e+03, threshold=3.186e+03, percent-clipped=7.0
+2023-03-07 00:59:07,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2496, 1.4099, 1.4322, 1.2177], device='cuda:0'), covar=tensor([0.1214, 0.1293, 0.1631, 0.1494], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0736, 0.0681, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 00:59:16,730 INFO [train.py:968] (0/2) Epoch 13, batch 39600, giga_loss[loss=0.2886, simple_loss=0.3553, pruned_loss=0.111, over 28698.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3526, pruned_loss=0.1038, over 5680944.76 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.371, pruned_loss=0.123, over 5703328.22 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3493, pruned_loss=0.1011, over 5695477.19 frames. ], batch size: 66, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:59:31,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=587082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:59:42,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1588, 1.3084, 1.1481, 1.1227], device='cuda:0'), covar=tensor([0.1736, 0.1489, 0.1301, 0.1519], device='cuda:0'), in_proj_covar=tensor([0.1762, 0.1672, 0.1650, 0.1734], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 00:59:57,067 INFO [train.py:968] (0/2) Epoch 13, batch 39650, giga_loss[loss=0.2999, simple_loss=0.3705, pruned_loss=0.1146, over 28779.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3557, pruned_loss=0.1059, over 5684129.02 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3712, pruned_loss=0.1231, over 5697070.45 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3526, pruned_loss=0.1032, over 5700765.11 frames. ], batch size: 199, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:59:59,342 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:00:19,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.056e+02 1.260e+03 1.562e+03 2.257e+03 7.208e+03, threshold=3.123e+03, percent-clipped=9.0
+2023-03-07 01:00:34,051 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-07 01:00:38,003 INFO [train.py:968] (0/2) Epoch 13, batch 39700, giga_loss[loss=0.3208, simple_loss=0.3929, pruned_loss=0.1244, over 29014.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3568, pruned_loss=0.1057, over 5697988.44 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3714, pruned_loss=0.1232, over 5696873.70 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.354, pruned_loss=0.1032, over 5711032.23 frames. ], batch size: 136, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:00:50,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6369, 1.5954, 1.1842, 1.2261], device='cuda:0'), covar=tensor([0.0768, 0.0621, 0.1063, 0.1145], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0437, 0.0506, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 01:01:21,472 INFO [train.py:968] (0/2) Epoch 13, batch 39750, giga_loss[loss=0.2524, simple_loss=0.3289, pruned_loss=0.08791, over 28359.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3589, pruned_loss=0.1073, over 5700419.83 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3721, pruned_loss=0.1238, over 5697252.07 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3557, pruned_loss=0.1047, over 5710424.41 frames. ], batch size: 60, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:01:24,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4609, 4.5001, 1.6588, 1.7006], device='cuda:0'), covar=tensor([0.0952, 0.0276, 0.0896, 0.1229], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0512, 0.0346, 0.0392], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 01:01:28,684 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-07 01:01:29,109 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=587225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:01:31,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=587228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:01:42,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.746e+02 1.331e+03 1.627e+03 2.216e+03 5.907e+03, threshold=3.254e+03, percent-clipped=6.0
+2023-03-07 01:01:52,939 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=587257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:01:59,537 INFO [train.py:968] (0/2) Epoch 13, batch 39800, giga_loss[loss=0.2994, simple_loss=0.3676, pruned_loss=0.1156, over 28862.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3594, pruned_loss=0.108, over 5702963.90 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3729, pruned_loss=0.1242, over 5699144.07 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3558, pruned_loss=0.105, over 5709289.03 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:02:12,417 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1731, 1.8041, 1.4134, 0.3517], device='cuda:0'), covar=tensor([0.3525, 0.2037, 0.3230, 0.4968], device='cuda:0'), in_proj_covar=tensor([0.1594, 0.1500, 0.1501, 0.1315], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 01:02:37,571 INFO [train.py:968] (0/2) Epoch 13, batch 39850, giga_loss[loss=0.2576, simple_loss=0.3388, pruned_loss=0.0882, over 28703.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3578, pruned_loss=0.1068, over 5705887.91 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3724, pruned_loss=0.1237, over 5702713.59 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.355, pruned_loss=0.1044, over 5707945.90 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:02:57,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.071e+02 1.150e+03 1.514e+03 1.964e+03 8.438e+03, threshold=3.028e+03, percent-clipped=6.0
+2023-03-07 01:03:18,604 INFO [train.py:968] (0/2) Epoch 13, batch 39900, giga_loss[loss=0.2717, simple_loss=0.3463, pruned_loss=0.09858, over 28998.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3554, pruned_loss=0.1055, over 5709250.86 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3723, pruned_loss=0.1236, over 5704758.33 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3531, pruned_loss=0.1035, over 5708985.87 frames. ], batch size: 128, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:03:57,548 INFO [train.py:968] (0/2) Epoch 13, batch 39950, giga_loss[loss=0.2447, simple_loss=0.3266, pruned_loss=0.08135, over 28648.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3521, pruned_loss=0.1038, over 5718243.00 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3732, pruned_loss=0.1241, over 5710201.35 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.349, pruned_loss=0.1013, over 5713379.96 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:03:59,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4420, 1.5656, 1.4790, 1.3902], device='cuda:0'), covar=tensor([0.2595, 0.2129, 0.1590, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.1775, 0.1686, 0.1667, 0.1745], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 01:04:16,257 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-07 01:04:22,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.459e+02 1.024e+03 1.276e+03 1.809e+03 4.258e+03, threshold=2.551e+03, percent-clipped=5.0
+2023-03-07 01:04:40,735 INFO [train.py:968] (0/2) Epoch 13, batch 40000, giga_loss[loss=0.275, simple_loss=0.353, pruned_loss=0.09855, over 28831.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.349, pruned_loss=0.1018, over 5716145.76 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.373, pruned_loss=0.1239, over 5714404.20 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3458, pruned_loss=0.09921, over 5708572.97 frames. ], batch size: 99, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:05:02,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=587493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:05:14,191 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:05:19,147 INFO [train.py:968] (0/2) Epoch 13, batch 40050, giga_loss[loss=0.3613, simple_loss=0.4084, pruned_loss=0.1571, over 26686.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3508, pruned_loss=0.1014, over 5707996.40 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3736, pruned_loss=0.1242, over 5705362.66 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3475, pruned_loss=0.09862, over 5709673.33 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:05:42,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.714e+02 1.246e+03 1.539e+03 2.261e+03 6.194e+03, threshold=3.078e+03, percent-clipped=17.0
+2023-03-07 01:06:01,522 INFO [train.py:968] (0/2) Epoch 13, batch 40100, giga_loss[loss=0.2878, simple_loss=0.3711, pruned_loss=0.1022, over 28288.00 frames. ], tot_loss[loss=0.279, simple_loss=0.353, pruned_loss=0.1026, over 5702214.10 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3742, pruned_loss=0.1248, over 5706805.74 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3492, pruned_loss=0.09927, over 5702585.47 frames. ], batch size: 368, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:06:30,454 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:06:34,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:06:40,239 INFO [train.py:968] (0/2) Epoch 13, batch 40150, giga_loss[loss=0.3368, simple_loss=0.3848, pruned_loss=0.1444, over 26635.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3511, pruned_loss=0.102, over 5707693.88 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3738, pruned_loss=0.1245, over 5706057.62 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3482, pruned_loss=0.09937, over 5708543.92 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:06:59,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=587636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:07:02,094 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=587639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:07:04,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.467e+02 1.137e+03 1.399e+03 1.898e+03 5.193e+03, threshold=2.798e+03, percent-clipped=9.0
+2023-03-07 01:07:15,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-07 01:07:22,407 INFO [train.py:968] (0/2) Epoch 13, batch 40200, giga_loss[loss=0.3311, simple_loss=0.3832, pruned_loss=0.1395, over 26796.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3493, pruned_loss=0.1022, over 5704490.37 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3739, pruned_loss=0.1246, over 5697520.83 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3467, pruned_loss=0.09984, over 5713265.10 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:07:24,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2752, 1.5661, 1.2585, 1.0192], device='cuda:0'), covar=tensor([0.2592, 0.2390, 0.2850, 0.2125], device='cuda:0'), in_proj_covar=tensor([0.1341, 0.0989, 0.1184, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 01:07:24,538 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=587668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:07:36,528 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:08:04,755 INFO [train.py:968] (0/2) Epoch 13, batch 40250, giga_loss[loss=0.2562, simple_loss=0.3297, pruned_loss=0.09131, over 28668.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3481, pruned_loss=0.103, over 5698569.17 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3742, pruned_loss=0.1248, over 5692848.95 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3451, pruned_loss=0.1006, over 5709676.97 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:08:19,914 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:08:26,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5286, 4.3443, 4.0919, 1.8669], device='cuda:0'), covar=tensor([0.0618, 0.0782, 0.0754, 0.2267], device='cuda:0'), in_proj_covar=tensor([0.1080, 0.1004, 0.0875, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 01:08:27,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.263e+02 1.131e+03 1.300e+03 1.837e+03 1.311e+04, threshold=2.600e+03, percent-clipped=9.0
+2023-03-07 01:08:39,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2721, 1.4249, 1.2848, 1.4438], device='cuda:0'), covar=tensor([0.0731, 0.0358, 0.0330, 0.0797], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0060, 0.0055, 0.0093], device='cuda:0')
+2023-03-07 01:08:48,032 INFO [train.py:968] (0/2) Epoch 13, batch 40300, giga_loss[loss=0.2233, simple_loss=0.296, pruned_loss=0.07533, over 28243.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3459, pruned_loss=0.1027, over 5698160.30 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3739, pruned_loss=0.1245, over 5697362.97 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3433, pruned_loss=0.1005, over 5703206.53 frames. ], batch size: 77, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:08:49,837 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-07 01:08:53,671 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 01:09:27,403 INFO [train.py:968] (0/2) Epoch 13, batch 40350, giga_loss[loss=0.2953, simple_loss=0.3607, pruned_loss=0.1149, over 28015.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3459, pruned_loss=0.1029, over 5696665.73 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3739, pruned_loss=0.1244, over 5692355.58 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3431, pruned_loss=0.1007, over 5704690.21 frames. ], batch size: 412, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:09:47,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.544e+02 1.121e+03 1.427e+03 2.033e+03 1.692e+04, threshold=2.855e+03, percent-clipped=17.0
+2023-03-07 01:09:55,329 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587851.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:10:04,830 INFO [train.py:968] (0/2) Epoch 13, batch 40400, giga_loss[loss=0.3225, simple_loss=0.3752, pruned_loss=0.1349, over 26854.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3439, pruned_loss=0.1026, over 5701416.26 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3743, pruned_loss=0.1249, over 5695788.30 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3405, pruned_loss=0.09979, over 5704895.71 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:10:19,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=587882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:10:37,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0148, 2.3236, 2.0097, 2.2101], device='cuda:0'), covar=tensor([0.2002, 0.1831, 0.2086, 0.1652], device='cuda:0'), in_proj_covar=tensor([0.1346, 0.0993, 0.1189, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 01:10:44,791 INFO [train.py:968] (0/2) Epoch 13, batch 40450, giga_loss[loss=0.218, simple_loss=0.2905, pruned_loss=0.07271, over 28661.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3395, pruned_loss=0.1002, over 5697109.23 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3744, pruned_loss=0.1249, over 5691360.97 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3358, pruned_loss=0.09729, over 5704532.04 frames. ], batch size: 60, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:11:07,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.038e+02 1.211e+03 1.581e+03 2.033e+03 4.228e+03, threshold=3.163e+03, percent-clipped=11.0
+2023-03-07 01:11:24,921 INFO [train.py:968] (0/2) Epoch 13, batch 40500, giga_loss[loss=0.2639, simple_loss=0.3383, pruned_loss=0.09478, over 28589.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3363, pruned_loss=0.09824, over 5699198.08 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3741, pruned_loss=0.1247, over 5687571.91 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3327, pruned_loss=0.09549, over 5708274.94 frames. ], batch size: 336, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:11:35,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=587978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:11:38,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=587982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:11:54,674 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-588000.pt
+2023-03-07 01:11:57,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4008, 1.5978, 1.5849, 1.4177], device='cuda:0'), covar=tensor([0.1675, 0.1855, 0.2176, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.0442, 0.0735, 0.0680, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 01:12:07,342 INFO [train.py:968] (0/2) Epoch 13, batch 40550, giga_loss[loss=0.3357, simple_loss=0.3957, pruned_loss=0.1378, over 27502.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3362, pruned_loss=0.09725, over 5706135.57 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3741, pruned_loss=0.1246, over 5689996.65 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3331, pruned_loss=0.09492, over 5711363.31 frames. ], batch size: 472, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:12:11,135 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-07 01:12:14,237 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:12:16,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:12:30,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.324e+02 1.197e+03 1.460e+03 1.982e+03 6.297e+03, threshold=2.919e+03, percent-clipped=8.0
+2023-03-07 01:12:37,640 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-07 01:12:40,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=588055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:12:41,705 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:12:47,953 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9930, 1.0425, 3.3455, 2.8499], device='cuda:0'), covar=tensor([0.1683, 0.2801, 0.0487, 0.1067], device='cuda:0'), in_proj_covar=tensor([0.0677, 0.0598, 0.0873, 0.0791], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-07 01:12:48,273 INFO [train.py:968] (0/2) Epoch 13, batch 40600, giga_loss[loss=0.2986, simple_loss=0.3665, pruned_loss=0.1153, over 28982.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3423, pruned_loss=0.1002, over 5708703.81 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.374, pruned_loss=0.1245, over 5695231.41 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3389, pruned_loss=0.09779, over 5708725.25 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:13:20,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=588106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:13:27,399 INFO [train.py:968] (0/2) Epoch 13, batch 40650, giga_loss[loss=0.2626, simple_loss=0.3414, pruned_loss=0.09194, over 29018.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3458, pruned_loss=0.1024, over 5714768.07 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3732, pruned_loss=0.1241, over 5700075.15 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3426, pruned_loss=0.09983, over 5711164.08 frames. ], batch size: 128, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:13:32,786 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:13:34,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:13:35,522 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:13:37,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:13:49,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.181e+02 1.143e+03 1.488e+03 1.779e+03 3.807e+03, threshold=2.976e+03, percent-clipped=2.0
+2023-03-07 01:13:50,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4893, 1.5629, 1.1268, 1.2468], device='cuda:0'), covar=tensor([0.0654, 0.0476, 0.0920, 0.1218], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0437, 0.0500, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 01:13:57,892 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:14:02,433 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:14:07,806 INFO [train.py:968] (0/2) Epoch 13, batch 40700, giga_loss[loss=0.2452, simple_loss=0.3242, pruned_loss=0.08315, over 28441.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3485, pruned_loss=0.1033, over 5706524.65 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3727, pruned_loss=0.1237, over 5704321.23 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3458, pruned_loss=0.1011, over 5699979.51 frames. ], batch size: 78, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:14:36,382 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:14:37,668 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=588200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:14:39,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:14:49,887 INFO [train.py:968] (0/2) Epoch 13, batch 40750, giga_loss[loss=0.2951, simple_loss=0.371, pruned_loss=0.1096, over 28554.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3514, pruned_loss=0.1044, over 5714972.03 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3727, pruned_loss=0.1236, over 5707026.89 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3489, pruned_loss=0.1024, over 5707343.30 frames. ], batch size: 336, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:14:58,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=588226.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 01:15:03,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:15:13,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.122e+02 1.148e+03 1.466e+03 1.786e+03 6.146e+03, threshold=2.932e+03, percent-clipped=5.0
+2023-03-07 01:15:18,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:15:21,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:15:33,602 INFO [train.py:968] (0/2) Epoch 13, batch 40800, giga_loss[loss=0.4299, simple_loss=0.4401, pruned_loss=0.2099, over 23598.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3551, pruned_loss=0.1077, over 5710853.82 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1234, over 5713094.86 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3529, pruned_loss=0.1058, over 5699366.57 frames. ], batch size: 705, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:15:51,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:16:22,242 INFO [train.py:968] (0/2) Epoch 13, batch 40850, giga_loss[loss=0.3085, simple_loss=0.3775, pruned_loss=0.1198, over 28943.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3615, pruned_loss=0.1136, over 5702417.33 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3725, pruned_loss=0.1237, over 5704340.29 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3592, pruned_loss=0.1115, over 5701204.82 frames. ], batch size: 227, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:16:46,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2434, 2.1600, 1.4995, 1.7525], device='cuda:0'), covar=tensor([0.0739, 0.0669, 0.0996, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0437, 0.0500, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 01:16:48,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.982e+02 1.614e+03 2.198e+03 3.011e+03 5.931e+03, threshold=4.395e+03, percent-clipped=26.0
+2023-03-07 01:17:06,670 INFO [train.py:968] (0/2) Epoch 13, batch 40900, giga_loss[loss=0.2969, simple_loss=0.3731, pruned_loss=0.1103, over 28628.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3678, pruned_loss=0.1183, over 5697678.56 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3722, pruned_loss=0.1236, over 5705342.89 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.366, pruned_loss=0.1165, over 5695694.59 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:17:11,662 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588369.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 01:17:14,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588372.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 01:17:40,358 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588401.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:17:52,946 INFO [train.py:968] (0/2) Epoch 13, batch 40950, giga_loss[loss=0.3461, simple_loss=0.404, pruned_loss=0.1441, over 28545.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3736, pruned_loss=0.1225, over 5694416.97 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.372, pruned_loss=0.1233, over 5709103.48 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3724, pruned_loss=0.1213, over 5689035.57 frames. ], batch size: 336, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:18:04,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7966, 1.0095, 2.8517, 2.7192], device='cuda:0'), covar=tensor([0.1686, 0.2544, 0.0599, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0679, 0.0601, 0.0875, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-07 01:18:06,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2572, 1.1847, 3.9612, 3.0194], device='cuda:0'), covar=tensor([0.1670, 0.2676, 0.0433, 0.0923], device='cuda:0'), in_proj_covar=tensor([0.0679, 0.0601, 0.0875, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-07 01:18:17,802 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.618e+03 2.126e+03 2.799e+03 8.069e+03, threshold=4.253e+03, percent-clipped=6.0
+2023-03-07 01:18:35,642 INFO [train.py:968] (0/2) Epoch 13, batch 41000, giga_loss[loss=0.4011, simple_loss=0.4422, pruned_loss=0.18, over 28326.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3797, pruned_loss=0.1279, over 5697487.73 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3719, pruned_loss=0.1232, over 5711956.19 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3788, pruned_loss=0.127, over 5690165.46 frames. ], batch size: 368, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:18:39,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8489, 2.7324, 1.8478, 1.1542], device='cuda:0'), covar=tensor([0.5485, 0.2613, 0.3026, 0.4801], device='cuda:0'), in_proj_covar=tensor([0.1609, 0.1524, 0.1515, 0.1329], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 01:18:46,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2178, 1.5841, 1.2961, 1.4515], device='cuda:0'), covar=tensor([0.0748, 0.0335, 0.0324, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-07 01:18:55,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=588488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:19:20,979 INFO [train.py:968] (0/2) Epoch 13, batch 41050, giga_loss[loss=0.3183, simple_loss=0.3829, pruned_loss=0.1269, over 28896.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3831, pruned_loss=0.1308, over 5686860.23 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.123, over 5705154.71 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.383, pruned_loss=0.1305, over 5686523.35 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:19:48,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5528, 1.5815, 1.8430, 1.4113], device='cuda:0'), covar=tensor([0.1202, 0.1741, 0.0971, 0.1299], device='cuda:0'), in_proj_covar=tensor([0.0842, 0.0687, 0.0881, 0.0788], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-07 01:19:52,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.775e+03 2.369e+03 3.312e+03 1.114e+04, threshold=4.738e+03, percent-clipped=13.0
+2023-03-07 01:20:13,016 INFO [train.py:968] (0/2) Epoch 13, batch 41100, libri_loss[loss=0.2659, simple_loss=0.3397, pruned_loss=0.0961, over 29569.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3869, pruned_loss=0.1351, over 5658662.60 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3714, pruned_loss=0.123, over 5699428.91 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3873, pruned_loss=0.1352, over 5662331.40 frames. ], batch size: 76, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:20:22,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=588575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:21:11,010 INFO [train.py:968] (0/2) Epoch 13, batch 41150, giga_loss[loss=0.4155, simple_loss=0.4442, pruned_loss=0.1934, over 28011.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3895, pruned_loss=0.1384, over 5661399.33 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3714, pruned_loss=0.1229, over 5700400.77 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3899, pruned_loss=0.1386, over 5663349.50 frames. ], batch size: 412, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:21:20,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4077, 3.3118, 1.5491, 1.4798], device='cuda:0'), covar=tensor([0.0900, 0.0310, 0.0832, 0.1307], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0517, 0.0347, 0.0396], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 01:21:40,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.657e+03 2.082e+03 2.875e+03 7.413e+03, threshold=4.164e+03, percent-clipped=5.0
+2023-03-07 01:22:02,180 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=588663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:22:03,796 INFO [train.py:968] (0/2) Epoch 13, batch 41200, giga_loss[loss=0.3402, simple_loss=0.3928, pruned_loss=0.1439, over 28868.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3928, pruned_loss=0.1423, over 5636172.64 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3714, pruned_loss=0.1228, over 5702378.82 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3935, pruned_loss=0.1428, over 5635121.06 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:22:22,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9858, 2.0944, 1.7363, 1.8672], device='cuda:0'), covar=tensor([0.1356, 0.1909, 0.1905, 0.1708], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0740, 0.0688, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 01:22:46,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5246, 1.6254, 1.8467, 1.4047], device='cuda:0'), covar=tensor([0.1190, 0.1763, 0.0989, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0839, 0.0687, 0.0878, 0.0785], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-07 01:22:51,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4995, 1.6635, 1.4957, 1.3787], device='cuda:0'), covar=tensor([0.2171, 0.1838, 0.1618, 0.1809], device='cuda:0'), in_proj_covar=tensor([0.1777, 0.1687, 0.1657, 0.1748], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 01:22:53,990 INFO [train.py:968] (0/2) Epoch 13, batch 41250, giga_loss[loss=0.3448, simple_loss=0.4034, pruned_loss=0.1431, over 28932.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3955, pruned_loss=0.1444, over 5622913.46 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3716, pruned_loss=0.123, over 5693030.18 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3964, pruned_loss=0.145, over 5629401.56 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:22:56,280 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:22:59,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:23:23,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.126e+03 1.836e+03 2.661e+03 4.662e+03 9.227e+03, threshold=5.322e+03, percent-clipped=25.0
+2023-03-07 01:23:28,152 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:23:47,230 INFO [train.py:968] (0/2) Epoch 13, batch 41300, giga_loss[loss=0.4101, simple_loss=0.4376, pruned_loss=0.1913, over 27962.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3955, pruned_loss=0.1453, over 5611604.14 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3713, pruned_loss=0.1229, over 5686643.78 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3969, pruned_loss=0.1462, over 5620982.84 frames. ], batch size: 412, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:24:31,582 INFO [train.py:968] (0/2) Epoch 13, batch 41350, giga_loss[loss=0.3508, simple_loss=0.4038, pruned_loss=0.1489, over 28558.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3941, pruned_loss=0.1448, over 5627267.51 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3705, pruned_loss=0.1224, over 5695714.30 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.3969, pruned_loss=0.1471, over 5623609.95 frames. ], batch size: 78, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:25:01,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.872e+03 2.519e+03 3.429e+03 8.374e+03, threshold=5.039e+03, percent-clipped=6.0
+2023-03-07 01:25:20,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=588863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:25:21,462 INFO [train.py:968] (0/2) Epoch 13, batch 41400, giga_loss[loss=0.3101, simple_loss=0.3709, pruned_loss=0.1246, over 28552.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3927, pruned_loss=0.1434, over 5646304.20 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1222, over 5699990.20 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3958, pruned_loss=0.1459, over 5638351.27 frames. ], batch size: 336, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:26:11,184 INFO [train.py:968] (0/2) Epoch 13, batch 41450, giga_loss[loss=0.3391, simple_loss=0.4019, pruned_loss=0.1381, over 28534.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3925, pruned_loss=0.1417, over 5650113.52 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3702, pruned_loss=0.1221, over 5701693.01 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3954, pruned_loss=0.1441, over 5641234.82 frames. ], batch size: 307, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:26:42,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.970e+02 1.532e+03 1.970e+03 2.728e+03 6.842e+03, threshold=3.940e+03, percent-clipped=3.0
+2023-03-07 01:26:58,574 INFO [train.py:968] (0/2) Epoch 13, batch 41500, giga_loss[loss=0.2969, simple_loss=0.3707, pruned_loss=0.1116, over 28896.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3932, pruned_loss=0.1409, over 5665891.20 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3708, pruned_loss=0.1225, over 5703559.66 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3956, pruned_loss=0.143, over 5655810.41 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:27:03,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 01:27:18,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4347, 1.7756, 1.5485, 1.5886], device='cuda:0'), covar=tensor([0.0639, 0.0265, 0.0267, 0.0632], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-07 01:27:41,167 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=589006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:27:46,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=589009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:27:51,688 INFO [train.py:968] (0/2) Epoch 13, batch 41550, giga_loss[loss=0.2778, simple_loss=0.3562, pruned_loss=0.09968, over 28979.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3943, pruned_loss=0.1423, over 5652002.07 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5709198.08 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3973, pruned_loss=0.1447, over 5637527.00 frames. ], batch size: 164, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:28:16,643 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=589038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:28:16,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=589038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:28:23,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.661e+02 1.503e+03 1.982e+03 2.801e+03 7.651e+03, threshold=3.964e+03, percent-clipped=5.0
+2023-03-07 01:28:43,191 INFO [train.py:968] (0/2) Epoch 13, batch 41600, giga_loss[loss=0.2973, simple_loss=0.3689, pruned_loss=0.1128, over 28938.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3905, pruned_loss=0.1379, over 5655868.83 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3705, pruned_loss=0.1224, over 5712454.65 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3931, pruned_loss=0.14, over 5640761.25 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:29:33,317 INFO [train.py:968] (0/2) Epoch 13, batch 41650, giga_loss[loss=0.319, simple_loss=0.3902, pruned_loss=0.1239, over 28710.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3877, pruned_loss=0.1346, over 5651386.92 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3703, pruned_loss=0.1223, over 5703973.41 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3901, pruned_loss=0.1365, over 5646999.88 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:30:01,326 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.005e+02 1.524e+03 1.881e+03 2.852e+03 6.229e+03, threshold=3.763e+03, percent-clipped=11.0
+2023-03-07 01:30:16,637 INFO [train.py:968] (0/2) Epoch 13, batch 41700, giga_loss[loss=0.3268, simple_loss=0.3866, pruned_loss=0.1335, over 27613.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3841, pruned_loss=0.1315, over 5649901.25 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3703, pruned_loss=0.1225, over 5694371.13 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3867, pruned_loss=0.1332, over 5651616.22 frames. ], batch size: 474, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:30:20,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=589171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:30:27,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5414, 1.9532, 1.4509, 1.7054], device='cuda:0'), covar=tensor([0.2484, 0.2372, 0.2747, 0.2219], device='cuda:0'), in_proj_covar=tensor([0.1344, 0.0989, 0.1191, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 01:30:29,742 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=589181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:30:32,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=589184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:30:36,926 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1911, 2.5713, 1.2264, 1.3401], device='cuda:0'), covar=tensor([0.0936, 0.0387, 0.0891, 0.1307], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0520, 0.0349, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 01:30:46,647 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-07 01:30:58,185 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=589213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:31:01,761 INFO [train.py:968] (0/2) Epoch 13, batch 41750, giga_loss[loss=0.3537, simple_loss=0.3853, pruned_loss=0.161, over 23658.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3809, pruned_loss=0.1293, over 5646887.56 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3699, pruned_loss=0.1222, over 5693175.58 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3839, pruned_loss=0.1313, over 5648129.90 frames. ], batch size: 705, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:31:02,645 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=589216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:31:19,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4524, 2.5123, 2.1654, 2.2014], device='cuda:0'), covar=tensor([0.1495, 0.2025, 0.2007, 0.1986], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0733, 0.0680, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 01:31:31,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.964e+02 1.747e+03 2.137e+03 3.721e+03 9.740e+03, threshold=4.275e+03, percent-clipped=23.0
+2023-03-07 01:31:34,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=589247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:31:48,887 INFO [train.py:968] (0/2) Epoch 13, batch 41800, libri_loss[loss=0.2862, simple_loss=0.3596, pruned_loss=0.1064, over 29523.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3799, pruned_loss=0.1289, over 5645685.28 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3699, pruned_loss=0.1223, over 5695995.61 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3826, pruned_loss=0.1307, over 5642206.41 frames. ], batch size: 81, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:32:08,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9888, 1.0106, 3.4236, 2.9710], device='cuda:0'), covar=tensor([0.1753, 0.2735, 0.0502, 0.0876], device='cuda:0'), in_proj_covar=tensor([0.0684, 0.0605, 0.0881, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 01:32:34,318 INFO [train.py:968] (0/2) Epoch 13, batch 41850, giga_loss[loss=0.3122, simple_loss=0.3707, pruned_loss=0.1269, over 28636.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3802, pruned_loss=0.1287, over 5666321.37 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.37, pruned_loss=0.1224, over 5698232.47 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3824, pruned_loss=0.1301, over 5660800.06 frames. ], batch size: 78, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:32:55,681 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-07 01:33:06,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.831e+02 1.627e+03 2.128e+03 3.059e+03 7.511e+03, threshold=4.257e+03, percent-clipped=7.0
+2023-03-07 01:33:27,966 INFO [train.py:968] (0/2) Epoch 13, batch 41900, giga_loss[loss=0.3182, simple_loss=0.3732, pruned_loss=0.1316, over 28732.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3788, pruned_loss=0.1271, over 5669185.40 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3698, pruned_loss=0.1223, over 5697870.82 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3809, pruned_loss=0.1284, over 5664791.33 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:34:23,093 INFO [train.py:968] (0/2) Epoch 13, batch 41950, giga_loss[loss=0.2984, simple_loss=0.3784, pruned_loss=0.1092, over 28289.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3785, pruned_loss=0.1249, over 5669516.67 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3698, pruned_loss=0.1224, over 5698915.46 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3801, pruned_loss=0.1259, over 5664930.82 frames. ], batch size: 368, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:34:58,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.776e+02 1.387e+03 1.852e+03 2.587e+03 9.224e+03, threshold=3.704e+03, percent-clipped=12.0
+2023-03-07 01:35:14,863 INFO [train.py:968] (0/2) Epoch 13, batch 42000, giga_loss[loss=0.337, simple_loss=0.3892, pruned_loss=0.1424, over 27581.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3791, pruned_loss=0.1236, over 5673877.14 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3695, pruned_loss=0.1222, over 5702004.46 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3808, pruned_loss=0.1246, over 5667090.90 frames. ], batch size: 472, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:35:14,867 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 01:35:23,302 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2105, simple_loss=0.3156, pruned_loss=0.05275, over 944034.00 frames. 
+2023-03-07 01:35:23,303 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 01:36:08,393 INFO [train.py:968] (0/2) Epoch 13, batch 42050, giga_loss[loss=0.3154, simple_loss=0.3838, pruned_loss=0.1236, over 29033.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3796, pruned_loss=0.1246, over 5673610.44 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3694, pruned_loss=0.1221, over 5702714.55 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3814, pruned_loss=0.1255, over 5667131.17 frames. ], batch size: 106, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:36:36,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=589546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:36:37,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.144e+03 1.797e+03 2.291e+03 3.112e+03 9.527e+03, threshold=4.581e+03, percent-clipped=17.0
+2023-03-07 01:36:52,858 INFO [train.py:968] (0/2) Epoch 13, batch 42100, giga_loss[loss=0.3189, simple_loss=0.3803, pruned_loss=0.1287, over 28509.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3801, pruned_loss=0.1257, over 5669531.30 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3694, pruned_loss=0.1221, over 5697835.27 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3817, pruned_loss=0.1265, over 5668070.19 frames. ], batch size: 65, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:37:01,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1280, 2.5650, 2.0367, 1.6020], device='cuda:0'), covar=tensor([0.2500, 0.1881, 0.1971, 0.2448], device='cuda:0'), in_proj_covar=tensor([0.1781, 0.1695, 0.1661, 0.1753], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 01:37:03,622 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 01:37:16,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=589591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:37:18,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5183, 1.4818, 1.1699, 1.1111], device='cuda:0'), covar=tensor([0.0642, 0.0444, 0.0883, 0.0908], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0439, 0.0503, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 01:37:38,111 INFO [train.py:968] (0/2) Epoch 13, batch 42150, giga_loss[loss=0.3337, simple_loss=0.384, pruned_loss=0.1418, over 27606.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3777, pruned_loss=0.125, over 5673117.03 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.369, pruned_loss=0.1219, over 5699167.10 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3794, pruned_loss=0.1258, over 5670527.90 frames. ], batch size: 472, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:37:44,201 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=589622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:38:06,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.403e+02 1.694e+03 2.123e+03 2.715e+03 1.069e+04, threshold=4.247e+03, percent-clipped=6.0
+2023-03-07 01:38:20,059 INFO [train.py:968] (0/2) Epoch 13, batch 42200, giga_loss[loss=0.3124, simple_loss=0.3758, pruned_loss=0.1245, over 28926.00 frames. ], tot_loss[loss=0.313, simple_loss=0.376, pruned_loss=0.125, over 5662351.36 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3685, pruned_loss=0.1215, over 5699670.76 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3783, pruned_loss=0.1263, over 5658382.50 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:38:21,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5018, 1.6337, 1.5155, 1.4137], device='cuda:0'), covar=tensor([0.2133, 0.1845, 0.1653, 0.1816], device='cuda:0'), in_proj_covar=tensor([0.1792, 0.1703, 0.1668, 0.1759], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 01:38:44,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=589689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:38:44,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5500, 1.7846, 1.4857, 1.5630], device='cuda:0'), covar=tensor([0.1988, 0.1852, 0.1872, 0.1702], device='cuda:0'), in_proj_covar=tensor([0.1346, 0.0990, 0.1193, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 01:38:46,453 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=589692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:07,736 INFO [train.py:968] (0/2) Epoch 13, batch 42250, giga_loss[loss=0.3935, simple_loss=0.4257, pruned_loss=0.1807, over 26587.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3756, pruned_loss=0.1247, over 5673641.99 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3688, pruned_loss=0.1217, over 5705888.04 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3773, pruned_loss=0.1256, over 5663586.41 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:39:13,634 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=589721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:27,036 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=589734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:29,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=589737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:37,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.972e+02 1.683e+03 2.316e+03 2.954e+03 7.701e+03, threshold=4.632e+03, percent-clipped=12.0
+2023-03-07 01:39:52,738 INFO [train.py:968] (0/2) Epoch 13, batch 42300, giga_loss[loss=0.3499, simple_loss=0.4051, pruned_loss=0.1473, over 28565.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3752, pruned_loss=0.1229, over 5678800.53 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3687, pruned_loss=0.1216, over 5700650.98 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3769, pruned_loss=0.1238, over 5673763.21 frames. ], batch size: 78, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:39:53,052 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=589765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:53,603 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=589766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:54,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=589768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:40:23,694 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=589797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:40:41,866 INFO [train.py:968] (0/2) Epoch 13, batch 42350, giga_loss[loss=0.28, simple_loss=0.3504, pruned_loss=0.1048, over 28845.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3747, pruned_loss=0.1218, over 5672230.80 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3687, pruned_loss=0.1216, over 5693771.52 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3762, pruned_loss=0.1225, over 5674211.96 frames. ], batch size: 112, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:41:14,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.853e+02 1.528e+03 1.979e+03 2.716e+03 9.992e+03, threshold=3.959e+03, percent-clipped=9.0
+2023-03-07 01:41:30,104 INFO [train.py:968] (0/2) Epoch 13, batch 42400, giga_loss[loss=0.2831, simple_loss=0.3614, pruned_loss=0.1024, over 28897.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3738, pruned_loss=0.1211, over 5679128.81 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3687, pruned_loss=0.1217, over 5689228.36 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3751, pruned_loss=0.1215, over 5685072.92 frames. ], batch size: 145, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:41:54,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4002, 1.0569, 4.1498, 3.3179], device='cuda:0'), covar=tensor([0.1612, 0.2849, 0.0471, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0686, 0.0606, 0.0884, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 01:42:17,206 INFO [train.py:968] (0/2) Epoch 13, batch 42450, libri_loss[loss=0.391, simple_loss=0.4341, pruned_loss=0.1739, over 25814.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3728, pruned_loss=0.1213, over 5668262.51 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3689, pruned_loss=0.1217, over 5687005.91 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3737, pruned_loss=0.1216, over 5674826.88 frames. ], batch size: 136, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:42:47,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.038e+02 1.512e+03 1.950e+03 2.491e+03 4.923e+03, threshold=3.900e+03, percent-clipped=6.0
+2023-03-07 01:43:04,518 INFO [train.py:968] (0/2) Epoch 13, batch 42500, giga_loss[loss=0.2653, simple_loss=0.3344, pruned_loss=0.09806, over 28906.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.373, pruned_loss=0.1223, over 5668712.34 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.369, pruned_loss=0.1217, over 5689557.66 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3737, pruned_loss=0.1225, over 5671568.24 frames. ], batch size: 112, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:43:43,049 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-590000.pt
+2023-03-07 01:43:55,784 INFO [train.py:968] (0/2) Epoch 13, batch 42550, giga_loss[loss=0.2942, simple_loss=0.3636, pruned_loss=0.1124, over 28629.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3727, pruned_loss=0.1232, over 5658030.59 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3692, pruned_loss=0.1219, over 5682361.08 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3731, pruned_loss=0.1233, over 5667082.72 frames. ], batch size: 336, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:44:09,941 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3607, 1.3010, 1.2345, 1.5234], device='cuda:0'), covar=tensor([0.0764, 0.0347, 0.0314, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0113, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-07 01:44:14,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4753, 1.8450, 1.4273, 1.6084], device='cuda:0'), covar=tensor([0.2372, 0.2357, 0.2688, 0.2290], device='cuda:0'), in_proj_covar=tensor([0.1350, 0.0991, 0.1195, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 01:44:27,728 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.705e+03 2.258e+03 3.040e+03 6.982e+03, threshold=4.516e+03, percent-clipped=13.0
+2023-03-07 01:44:43,433 INFO [train.py:968] (0/2) Epoch 13, batch 42600, libri_loss[loss=0.2961, simple_loss=0.3505, pruned_loss=0.1208, over 29690.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3697, pruned_loss=0.1215, over 5670542.99 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3684, pruned_loss=0.1212, over 5689545.77 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3709, pruned_loss=0.1222, over 5670223.21 frames. ], batch size: 73, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:45:00,271 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:45:28,085 INFO [train.py:968] (0/2) Epoch 13, batch 42650, giga_loss[loss=0.3557, simple_loss=0.4041, pruned_loss=0.1536, over 28332.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.37, pruned_loss=0.1217, over 5677765.51 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3684, pruned_loss=0.1211, over 5691488.15 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.371, pruned_loss=0.1224, over 5675493.75 frames. ], batch size: 369, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:45:59,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:45:59,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.641e+03 2.175e+03 3.396e+03 1.490e+04, threshold=4.351e+03, percent-clipped=14.0
+2023-03-07 01:46:07,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0345, 5.0041, 2.1241, 2.3195], device='cuda:0'), covar=tensor([0.0832, 0.0219, 0.0752, 0.1053], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0522, 0.0351, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 01:46:08,251 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 01:46:16,462 INFO [train.py:968] (0/2) Epoch 13, batch 42700, giga_loss[loss=0.2757, simple_loss=0.3534, pruned_loss=0.09895, over 28785.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5675606.18 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3684, pruned_loss=0.1211, over 5687365.14 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3712, pruned_loss=0.1229, over 5677935.41 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:47:00,874 INFO [train.py:968] (0/2) Epoch 13, batch 42750, giga_loss[loss=0.3529, simple_loss=0.4041, pruned_loss=0.1509, over 29005.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3708, pruned_loss=0.122, over 5682803.31 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3675, pruned_loss=0.1208, over 5693829.59 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3722, pruned_loss=0.1228, over 5678677.52 frames. ], batch size: 136, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:47:31,719 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.711e+03 2.420e+03 3.342e+03 1.291e+04, threshold=4.840e+03, percent-clipped=14.0
+2023-03-07 01:47:46,682 INFO [train.py:968] (0/2) Epoch 13, batch 42800, giga_loss[loss=0.2831, simple_loss=0.3522, pruned_loss=0.107, over 28626.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3724, pruned_loss=0.1222, over 5686306.13 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3676, pruned_loss=0.1207, over 5696101.94 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3736, pruned_loss=0.1229, over 5680843.53 frames. ], batch size: 92, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:48:00,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7602, 2.0688, 1.7685, 1.6817], device='cuda:0'), covar=tensor([0.1629, 0.2152, 0.1996, 0.2204], device='cuda:0'), in_proj_covar=tensor([0.0442, 0.0738, 0.0685, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 01:48:03,617 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-07 01:48:31,710 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:48:32,154 INFO [train.py:968] (0/2) Epoch 13, batch 42850, giga_loss[loss=0.4002, simple_loss=0.4219, pruned_loss=0.1892, over 26662.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3733, pruned_loss=0.1223, over 5680151.09 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3679, pruned_loss=0.1209, over 5698690.85 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.374, pruned_loss=0.1227, over 5673079.53 frames. ], batch size: 555, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:49:08,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.034e+02 1.457e+03 1.944e+03 2.482e+03 4.840e+03, threshold=3.887e+03, percent-clipped=0.0
+2023-03-07 01:49:24,061 INFO [train.py:968] (0/2) Epoch 13, batch 42900, giga_loss[loss=0.3264, simple_loss=0.3888, pruned_loss=0.132, over 29004.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3741, pruned_loss=0.1236, over 5676915.39 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3676, pruned_loss=0.1207, over 5704653.27 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3751, pruned_loss=0.1242, over 5665061.23 frames. ], batch size: 213, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:49:39,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590383.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 01:50:10,090 INFO [train.py:968] (0/2) Epoch 13, batch 42950, giga_loss[loss=0.3398, simple_loss=0.3878, pruned_loss=0.146, over 28527.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3771, pruned_loss=0.1269, over 5672729.17 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3676, pruned_loss=0.1207, over 5705077.09 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.378, pruned_loss=0.1274, over 5662321.12 frames. ], batch size: 336, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:50:39,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3392, 3.1354, 2.9650, 1.6451], device='cuda:0'), covar=tensor([0.1025, 0.1225, 0.1190, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.1123, 0.1041, 0.0907, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 01:50:43,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.867e+03 2.498e+03 3.835e+03 1.257e+04, threshold=4.996e+03, percent-clipped=23.0
+2023-03-07 01:50:55,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=590460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:50:58,403 INFO [train.py:968] (0/2) Epoch 13, batch 43000, giga_loss[loss=0.3918, simple_loss=0.4263, pruned_loss=0.1787, over 27854.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3777, pruned_loss=0.1288, over 5666960.18 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3674, pruned_loss=0.1205, over 5707915.68 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.379, pruned_loss=0.1297, over 5654588.63 frames. ], batch size: 412, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:51:25,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2832, 1.7130, 1.2649, 0.6157], device='cuda:0'), covar=tensor([0.3059, 0.1939, 0.2066, 0.4246], device='cuda:0'), in_proj_covar=tensor([0.1616, 0.1544, 0.1526, 0.1337], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 01:51:52,221 INFO [train.py:968] (0/2) Epoch 13, batch 43050, giga_loss[loss=0.3089, simple_loss=0.3649, pruned_loss=0.1264, over 28794.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3795, pruned_loss=0.1313, over 5663635.27 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3675, pruned_loss=0.1205, over 5708282.67 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3806, pruned_loss=0.1321, over 5653015.85 frames. ], batch size: 119, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:51:58,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=590522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:52:19,850 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-07 01:52:25,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.225e+03 1.819e+03 2.286e+03 3.299e+03 6.555e+03, threshold=4.572e+03, percent-clipped=5.0
+2023-03-07 01:52:38,690 INFO [train.py:968] (0/2) Epoch 13, batch 43100, giga_loss[loss=0.3133, simple_loss=0.3768, pruned_loss=0.1249, over 28967.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3796, pruned_loss=0.1317, over 5668783.46 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3676, pruned_loss=0.1208, over 5708510.42 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3806, pruned_loss=0.1323, over 5659534.58 frames. ], batch size: 213, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:53:10,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=590603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:53:13,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:53:20,206 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9216, 3.0542, 2.0055, 1.1718], device='cuda:0'), covar=tensor([0.5986, 0.2324, 0.3028, 0.5408], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1545, 0.1522, 0.1336], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 01:53:21,866 INFO [train.py:968] (0/2) Epoch 13, batch 43150, giga_loss[loss=0.4209, simple_loss=0.4457, pruned_loss=0.1981, over 28305.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3769, pruned_loss=0.1293, over 5676627.15 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3673, pruned_loss=0.1204, over 5712237.13 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3783, pruned_loss=0.1305, over 5665012.98 frames. ], batch size: 368, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 01:53:42,546 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=590635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:53:55,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.751e+02 1.511e+03 2.026e+03 2.697e+03 7.642e+03, threshold=4.052e+03, percent-clipped=6.0
+2023-03-07 01:54:06,891 INFO [train.py:968] (0/2) Epoch 13, batch 43200, libri_loss[loss=0.3176, simple_loss=0.3846, pruned_loss=0.1253, over 25713.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3767, pruned_loss=0.1276, over 5681639.22 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3677, pruned_loss=0.1206, over 5712462.56 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3776, pruned_loss=0.1284, over 5671679.97 frames. ], batch size: 136, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:54:07,246 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=590665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:54:10,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:54:30,409 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=590689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:54:36,058 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=590697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:54:52,248 INFO [train.py:968] (0/2) Epoch 13, batch 43250, giga_loss[loss=0.2875, simple_loss=0.363, pruned_loss=0.1059, over 29021.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3745, pruned_loss=0.1257, over 5668943.72 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.368, pruned_loss=0.1209, over 5707064.31 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3751, pruned_loss=0.1262, over 5665823.08 frames. ], batch size: 106, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:55:15,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3293, 1.7759, 1.2694, 0.6631], device='cuda:0'), covar=tensor([0.3739, 0.2266, 0.2338, 0.4596], device='cuda:0'), in_proj_covar=tensor([0.1607, 0.1542, 0.1516, 0.1332], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 01:55:23,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6225, 1.6769, 1.6719, 1.4573], device='cuda:0'), covar=tensor([0.2424, 0.2152, 0.1731, 0.2061], device='cuda:0'), in_proj_covar=tensor([0.1800, 0.1704, 0.1669, 0.1762], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 01:55:24,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.900e+02 1.676e+03 2.193e+03 3.060e+03 1.065e+04, threshold=4.385e+03, percent-clipped=12.0
+2023-03-07 01:55:32,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=590758.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:55:37,954 INFO [train.py:968] (0/2) Epoch 13, batch 43300, giga_loss[loss=0.2972, simple_loss=0.3579, pruned_loss=0.1183, over 28495.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3734, pruned_loss=0.1253, over 5659247.91 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3688, pruned_loss=0.1213, over 5700205.72 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3732, pruned_loss=0.1253, over 5661833.40 frames. ], batch size: 336, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:55:42,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6629, 1.6661, 1.2828, 1.2528], device='cuda:0'), covar=tensor([0.0763, 0.0569, 0.0974, 0.1157], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0443, 0.0505, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 01:56:16,733 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:56:22,283 INFO [train.py:968] (0/2) Epoch 13, batch 43350, libri_loss[loss=0.3845, simple_loss=0.4241, pruned_loss=0.1725, over 19579.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3716, pruned_loss=0.1247, over 5656902.66 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.369, pruned_loss=0.1215, over 5693140.04 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3713, pruned_loss=0.1247, over 5665421.20 frames. ], batch size: 187, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:56:38,541 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=590832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:56:40,434 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:56:51,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.559e+02 1.710e+03 2.194e+03 3.206e+03 8.758e+03, threshold=4.388e+03, percent-clipped=9.0
+2023-03-07 01:57:02,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1864, 1.3735, 1.3070, 1.0589], device='cuda:0'), covar=tensor([0.1916, 0.1736, 0.1195, 0.1633], device='cuda:0'), in_proj_covar=tensor([0.1800, 0.1705, 0.1672, 0.1766], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 01:57:04,792 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=590864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:57:05,218 INFO [train.py:968] (0/2) Epoch 13, batch 43400, giga_loss[loss=0.2878, simple_loss=0.3597, pruned_loss=0.1079, over 29042.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.372, pruned_loss=0.1252, over 5665003.94 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3683, pruned_loss=0.121, over 5701126.08 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3725, pruned_loss=0.1258, over 5662814.45 frames. ], batch size: 128, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:57:38,481 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=590901.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:57:41,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590904.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:57:50,490 INFO [train.py:968] (0/2) Epoch 13, batch 43450, giga_loss[loss=0.2858, simple_loss=0.3652, pruned_loss=0.1032, over 28743.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3747, pruned_loss=0.126, over 5665247.91 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3687, pruned_loss=0.1212, over 5701727.05 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.375, pruned_loss=0.1265, over 5661698.01 frames. ], batch size: 284, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:57:59,898 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0014, 2.8338, 2.7041, 1.7282], device='cuda:0'), covar=tensor([0.1064, 0.1184, 0.1046, 0.1882], device='cuda:0'), in_proj_covar=tensor([0.1117, 0.1040, 0.0907, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 01:58:07,781 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=590933.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:58:15,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590942.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:58:22,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.579e+03 2.358e+03 3.200e+03 7.552e+03, threshold=4.716e+03, percent-clipped=9.0
+2023-03-07 01:58:37,115 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-07 01:58:38,339 INFO [train.py:968] (0/2) Epoch 13, batch 43500, giga_loss[loss=0.2804, simple_loss=0.3639, pruned_loss=0.09844, over 28856.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3777, pruned_loss=0.1255, over 5661947.86 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.369, pruned_loss=0.1215, over 5702701.35 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3779, pruned_loss=0.1256, over 5657557.42 frames. ], batch size: 227, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:58:54,794 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-07 01:59:18,119 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 01:59:19,425 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3474, 1.8677, 1.3583, 0.6667], device='cuda:0'), covar=tensor([0.4132, 0.2390, 0.2752, 0.4832], device='cuda:0'), in_proj_covar=tensor([0.1608, 0.1539, 0.1513, 0.1330], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 01:59:28,235 INFO [train.py:968] (0/2) Epoch 13, batch 43550, giga_loss[loss=0.2642, simple_loss=0.3341, pruned_loss=0.09719, over 28608.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3808, pruned_loss=0.1271, over 5665452.50 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3695, pruned_loss=0.122, over 5704445.45 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3806, pruned_loss=0.1268, over 5659926.58 frames. ], batch size: 60, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:00:02,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.587e+03 2.051e+03 2.962e+03 1.600e+04, threshold=4.102e+03, percent-clipped=11.0
+2023-03-07 02:00:14,543 INFO [train.py:968] (0/2) Epoch 13, batch 43600, giga_loss[loss=0.4026, simple_loss=0.4429, pruned_loss=0.1811, over 27688.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3829, pruned_loss=0.1288, over 5669335.51 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3694, pruned_loss=0.1221, over 5701748.51 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3832, pruned_loss=0.1288, over 5665888.22 frames. ], batch size: 472, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:00:53,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2958, 2.0057, 1.4868, 0.5191], device='cuda:0'), covar=tensor([0.3965, 0.2293, 0.3546, 0.4861], device='cuda:0'), in_proj_covar=tensor([0.1617, 0.1547, 0.1522, 0.1335], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 02:01:03,104 INFO [train.py:968] (0/2) Epoch 13, batch 43650, giga_loss[loss=0.2797, simple_loss=0.3572, pruned_loss=0.1011, over 28459.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3831, pruned_loss=0.1295, over 5659496.95 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3697, pruned_loss=0.1223, over 5693909.28 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3833, pruned_loss=0.1293, over 5662427.88 frames. ], batch size: 60, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:01:19,812 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=591134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:01:28,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9358, 1.1519, 1.0869, 0.7631], device='cuda:0'), covar=tensor([0.1873, 0.2015, 0.1207, 0.1868], device='cuda:0'), in_proj_covar=tensor([0.1800, 0.1704, 0.1673, 0.1771], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 02:01:30,422 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5805, 1.8067, 1.5417, 1.7204], device='cuda:0'), covar=tensor([0.1818, 0.1677, 0.1715, 0.1487], device='cuda:0'), in_proj_covar=tensor([0.1350, 0.0993, 0.1192, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 02:01:34,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.168e+03 1.574e+03 1.970e+03 2.990e+03 9.261e+03, threshold=3.940e+03, percent-clipped=15.0
+2023-03-07 02:01:44,923 INFO [train.py:968] (0/2) Epoch 13, batch 43700, giga_loss[loss=0.3122, simple_loss=0.3745, pruned_loss=0.125, over 28863.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3826, pruned_loss=0.1296, over 5674922.38 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3696, pruned_loss=0.1222, over 5700192.58 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3832, pruned_loss=0.1298, over 5670880.08 frames. ], batch size: 145, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:01:59,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5657, 1.5717, 1.1412, 1.1544], device='cuda:0'), covar=tensor([0.0707, 0.0514, 0.0962, 0.1053], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0446, 0.0510, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 02:02:01,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=591181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:02:32,887 INFO [train.py:968] (0/2) Epoch 13, batch 43750, giga_loss[loss=0.264, simple_loss=0.3298, pruned_loss=0.09903, over 28427.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3816, pruned_loss=0.1299, over 5668949.50 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3693, pruned_loss=0.1219, over 5704565.06 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3827, pruned_loss=0.1306, over 5660961.14 frames. ], batch size: 85, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:03:05,569 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.612e+03 2.221e+03 3.114e+03 1.035e+04, threshold=4.442e+03, percent-clipped=13.0
+2023-03-07 02:03:17,891 INFO [train.py:968] (0/2) Epoch 13, batch 43800, giga_loss[loss=0.2528, simple_loss=0.3212, pruned_loss=0.09226, over 28667.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3798, pruned_loss=0.1291, over 5661806.89 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1222, over 5699137.32 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3804, pruned_loss=0.1297, over 5659544.28 frames. ], batch size: 71, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:04:06,919 INFO [train.py:968] (0/2) Epoch 13, batch 43850, giga_loss[loss=0.3815, simple_loss=0.4042, pruned_loss=0.1794, over 23515.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3789, pruned_loss=0.1295, over 5656348.00 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3701, pruned_loss=0.1222, over 5700516.24 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3795, pruned_loss=0.13, over 5652577.52 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:04:10,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=591317.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 02:04:13,985 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7466, 1.7135, 1.8124, 1.5717], device='cuda:0'), covar=tensor([0.1439, 0.2072, 0.1881, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0735, 0.0683, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 02:04:16,915 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=591324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:04:19,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=591327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:04:41,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.881e+02 1.775e+03 2.440e+03 3.993e+03 1.282e+04, threshold=4.881e+03, percent-clipped=18.0
+2023-03-07 02:04:47,192 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=591356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:04:55,325 INFO [train.py:968] (0/2) Epoch 13, batch 43900, giga_loss[loss=0.3456, simple_loss=0.4017, pruned_loss=0.1447, over 28660.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3796, pruned_loss=0.1304, over 5657120.97 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3697, pruned_loss=0.1219, over 5706196.81 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3808, pruned_loss=0.1314, over 5647563.11 frames. ], batch size: 262, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:05:44,224 INFO [train.py:968] (0/2) Epoch 13, batch 43950, giga_loss[loss=0.325, simple_loss=0.3827, pruned_loss=0.1337, over 28881.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3769, pruned_loss=0.129, over 5656657.57 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3695, pruned_loss=0.1218, over 5704849.05 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3782, pruned_loss=0.13, over 5649426.67 frames. ], batch size: 199, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:06:16,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.162e+03 1.670e+03 1.998e+03 2.907e+03 7.082e+03, threshold=3.996e+03, percent-clipped=5.0
+2023-03-07 02:06:25,682 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=591460.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 02:06:27,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=591463.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 02:06:29,277 INFO [train.py:968] (0/2) Epoch 13, batch 44000, giga_loss[loss=0.3517, simple_loss=0.3936, pruned_loss=0.1549, over 26622.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3753, pruned_loss=0.1279, over 5666329.15 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3697, pruned_loss=0.122, over 5706999.49 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3762, pruned_loss=0.1287, over 5658247.93 frames. ], batch size: 555, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:06:54,600 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=591492.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 02:07:10,818 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=591509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:07:15,188 INFO [train.py:968] (0/2) Epoch 13, batch 44050, giga_loss[loss=0.2836, simple_loss=0.3563, pruned_loss=0.1054, over 28851.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3759, pruned_loss=0.1282, over 5665090.76 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3701, pruned_loss=0.1221, over 5710066.92 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3764, pruned_loss=0.1287, over 5655322.95 frames. ], batch size: 199, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:07:55,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.836e+02 1.572e+03 2.079e+03 3.007e+03 6.898e+03, threshold=4.158e+03, percent-clipped=9.0
+2023-03-07 02:08:06,028 INFO [train.py:968] (0/2) Epoch 13, batch 44100, giga_loss[loss=0.2856, simple_loss=0.3634, pruned_loss=0.1039, over 28857.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3793, pruned_loss=0.1303, over 5657026.00 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1223, over 5713125.38 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3798, pruned_loss=0.1307, over 5645751.80 frames. ], batch size: 186, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:08:40,168 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=591604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:08:47,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6191, 1.6191, 1.9395, 1.4590], device='cuda:0'), covar=tensor([0.1385, 0.1848, 0.1096, 0.1401], device='cuda:0'), in_proj_covar=tensor([0.0849, 0.0700, 0.0889, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 02:08:51,917 INFO [train.py:968] (0/2) Epoch 13, batch 44150, giga_loss[loss=0.3366, simple_loss=0.3947, pruned_loss=0.1393, over 28821.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3796, pruned_loss=0.1303, over 5652657.13 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3702, pruned_loss=0.1223, over 5707974.71 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3802, pruned_loss=0.1308, over 5647521.41 frames. ], batch size: 199, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:09:24,437 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=591648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:09:29,010 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=591652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:09:29,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.475e+02 1.600e+03 2.333e+03 3.695e+03 1.314e+04, threshold=4.667e+03, percent-clipped=17.0
+2023-03-07 02:09:31,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=591655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:09:36,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2666, 1.5831, 1.4463, 1.1756], device='cuda:0'), covar=tensor([0.2927, 0.2128, 0.1540, 0.2165], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1683, 0.1656, 0.1753], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 02:09:37,972 INFO [train.py:968] (0/2) Epoch 13, batch 44200, giga_loss[loss=0.3407, simple_loss=0.3934, pruned_loss=0.144, over 28596.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3789, pruned_loss=0.1292, over 5668723.88 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1223, over 5711727.00 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3796, pruned_loss=0.1298, over 5659941.60 frames. ], batch size: 307, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:09:56,775 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=591684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:10:23,320 INFO [train.py:968] (0/2) Epoch 13, batch 44250, giga_loss[loss=0.2718, simple_loss=0.372, pruned_loss=0.08584, over 28973.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3793, pruned_loss=0.1264, over 5676344.00 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5716751.52 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3798, pruned_loss=0.127, over 5663719.79 frames. ], batch size: 155, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:10:53,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.773e+02 1.316e+03 1.732e+03 2.141e+03 4.657e+03, threshold=3.465e+03, percent-clipped=0.0
+2023-03-07 02:11:04,195 INFO [train.py:968] (0/2) Epoch 13, batch 44300, giga_loss[loss=0.3135, simple_loss=0.3909, pruned_loss=0.118, over 29135.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3802, pruned_loss=0.1254, over 5675426.35 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3701, pruned_loss=0.1222, over 5718177.66 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3811, pruned_loss=0.126, over 5662991.83 frames. ], batch size: 128, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:11:47,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6754, 2.5496, 1.9274, 2.3903], device='cuda:0'), covar=tensor([0.0696, 0.0583, 0.0882, 0.0925], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0441, 0.0505, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 02:11:54,522 INFO [train.py:968] (0/2) Epoch 13, batch 44350, giga_loss[loss=0.3734, simple_loss=0.4223, pruned_loss=0.1622, over 27880.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3823, pruned_loss=0.1276, over 5655338.12 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3701, pruned_loss=0.1223, over 5710231.60 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3835, pruned_loss=0.1282, over 5651146.38 frames. ], batch size: 412, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:12:31,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.658e+03 2.212e+03 3.295e+03 1.909e+04, threshold=4.423e+03, percent-clipped=23.0
+2023-03-07 02:12:41,569 INFO [train.py:968] (0/2) Epoch 13, batch 44400, giga_loss[loss=0.3166, simple_loss=0.3756, pruned_loss=0.1289, over 28940.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3851, pruned_loss=0.1308, over 5658697.83 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3699, pruned_loss=0.1222, over 5702810.95 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3867, pruned_loss=0.1316, over 5659494.84 frames. ], batch size: 227, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:13:34,555 INFO [train.py:968] (0/2) Epoch 13, batch 44450, giga_loss[loss=0.3355, simple_loss=0.3905, pruned_loss=0.1402, over 27978.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3865, pruned_loss=0.1326, over 5658014.95 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3699, pruned_loss=0.1221, over 5702204.90 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3879, pruned_loss=0.1334, over 5658961.47 frames. ], batch size: 412, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:14:06,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.600e+03 2.429e+03 3.927e+03 1.111e+04, threshold=4.858e+03, percent-clipped=17.0
+2023-03-07 02:14:10,854 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 02:14:15,896 INFO [train.py:968] (0/2) Epoch 13, batch 44500, giga_loss[loss=0.3889, simple_loss=0.4231, pruned_loss=0.1774, over 26591.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3838, pruned_loss=0.1307, over 5672625.41 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3694, pruned_loss=0.1219, over 5707016.17 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.386, pruned_loss=0.132, over 5667781.49 frames. ], batch size: 555, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:14:27,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=591979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:14:47,003 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-592000.pt
+2023-03-07 02:15:01,081 INFO [train.py:968] (0/2) Epoch 13, batch 44550, giga_loss[loss=0.2777, simple_loss=0.3593, pruned_loss=0.09805, over 28567.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3813, pruned_loss=0.1287, over 5668834.52 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3688, pruned_loss=0.1216, over 5711761.21 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3839, pruned_loss=0.1301, over 5660200.79 frames. ], batch size: 78, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:15:01,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-07 02:15:08,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=592023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:15:13,505 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-07 02:15:25,308 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1587, 1.3524, 1.2774, 1.0946], device='cuda:0'), covar=tensor([0.2001, 0.1820, 0.1279, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.1784, 0.1694, 0.1657, 0.1760], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 02:15:34,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.781e+02 1.530e+03 1.949e+03 2.770e+03 5.480e+03, threshold=3.898e+03, percent-clipped=4.0
+2023-03-07 02:15:47,039 INFO [train.py:968] (0/2) Epoch 13, batch 44600, giga_loss[loss=0.2786, simple_loss=0.3631, pruned_loss=0.09702, over 28992.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3815, pruned_loss=0.1264, over 5677635.86 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1214, over 5713657.06 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3838, pruned_loss=0.1278, over 5668916.42 frames. ], batch size: 128, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:16:33,516 INFO [train.py:968] (0/2) Epoch 13, batch 44650, giga_loss[loss=0.3511, simple_loss=0.4182, pruned_loss=0.142, over 28563.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3815, pruned_loss=0.1253, over 5689037.98 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1214, over 5716599.13 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3835, pruned_loss=0.1264, over 5679228.48 frames. ], batch size: 307, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:16:40,534 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=592122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:16:43,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=592125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:16:50,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:17:08,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.199e+02 1.539e+03 1.989e+03 2.743e+03 5.243e+03, threshold=3.978e+03, percent-clipped=9.0
+2023-03-07 02:17:11,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=592154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:17:21,192 INFO [train.py:968] (0/2) Epoch 13, batch 44700, giga_loss[loss=0.3431, simple_loss=0.3983, pruned_loss=0.1439, over 28332.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3825, pruned_loss=0.1273, over 5674014.98 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3682, pruned_loss=0.1213, over 5722283.08 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.385, pruned_loss=0.1285, over 5660032.69 frames. ], batch size: 368, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:17:22,761 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=592166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:17:26,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=592169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:17:50,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4922, 1.8313, 1.7441, 1.2883], device='cuda:0'), covar=tensor([0.1632, 0.2672, 0.1482, 0.1778], device='cuda:0'), in_proj_covar=tensor([0.0846, 0.0700, 0.0887, 0.0792], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-07 02:17:52,940 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=592198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:18:06,328 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-07 02:18:07,189 INFO [train.py:968] (0/2) Epoch 13, batch 44750, giga_loss[loss=0.2687, simple_loss=0.3407, pruned_loss=0.09833, over 28818.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3811, pruned_loss=0.1267, over 5672076.99 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3679, pruned_loss=0.121, over 5726418.05 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3837, pruned_loss=0.1281, over 5655619.19 frames. ], batch size: 112, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:18:38,269 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.803e+02 1.631e+03 2.102e+03 2.887e+03 1.417e+04, threshold=4.204e+03, percent-clipped=10.0
+2023-03-07 02:18:50,505 INFO [train.py:968] (0/2) Epoch 13, batch 44800, giga_loss[loss=0.384, simple_loss=0.4199, pruned_loss=0.1741, over 27530.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3784, pruned_loss=0.1258, over 5678168.18 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3678, pruned_loss=0.1209, over 5729211.38 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.381, pruned_loss=0.1271, over 5661064.70 frames. ], batch size: 472, lr: 2.43e-03, grad_scale: 8.0
+2023-03-07 02:19:19,673 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:19:28,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4251, 1.2290, 4.8662, 3.3924], device='cuda:0'), covar=tensor([0.1745, 0.2679, 0.0385, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0692, 0.0610, 0.0894, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 02:19:35,093 INFO [train.py:968] (0/2) Epoch 13, batch 44850, giga_loss[loss=0.2688, simple_loss=0.3385, pruned_loss=0.09951, over 28651.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3774, pruned_loss=0.126, over 5669859.62 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3679, pruned_loss=0.121, over 5725050.08 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3796, pruned_loss=0.1271, over 5658501.72 frames. ], batch size: 60, lr: 2.43e-03, grad_scale: 8.0
+2023-03-07 02:20:10,374 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:20:12,062 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.198e+02 1.705e+03 2.026e+03 2.634e+03 9.522e+03, threshold=4.052e+03, percent-clipped=8.0
+2023-03-07 02:20:20,938 INFO [train.py:968] (0/2) Epoch 13, batch 44900, giga_loss[loss=0.2678, simple_loss=0.3426, pruned_loss=0.09651, over 28843.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3754, pruned_loss=0.1253, over 5674043.60 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3677, pruned_loss=0.1209, over 5727921.10 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3775, pruned_loss=0.1263, over 5661127.31 frames. ], batch size: 66, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:21:07,445 INFO [train.py:968] (0/2) Epoch 13, batch 44950, giga_loss[loss=0.2876, simple_loss=0.361, pruned_loss=0.1071, over 29048.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3761, pruned_loss=0.127, over 5670962.50 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3679, pruned_loss=0.1209, over 5730073.62 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3777, pruned_loss=0.1279, over 5658069.93 frames. ], batch size: 155, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:21:42,125 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.561e+03 2.003e+03 2.742e+03 5.765e+03, threshold=4.005e+03, percent-clipped=6.0
+2023-03-07 02:21:51,573 INFO [train.py:968] (0/2) Epoch 13, batch 45000, giga_loss[loss=0.2921, simple_loss=0.3617, pruned_loss=0.1113, over 28868.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3767, pruned_loss=0.1288, over 5657861.48 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3673, pruned_loss=0.1207, over 5735789.31 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.379, pruned_loss=0.13, over 5639853.14 frames. ], batch size: 186, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:21:51,577 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 02:22:00,265 INFO [train.py:1012] (0/2) Epoch 13, validation: loss=0.2132, simple_loss=0.3207, pruned_loss=0.05289, over 944034.00 frames. 
+2023-03-07 02:22:00,266 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 02:22:36,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=592506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:22:40,769 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-07 02:22:45,280 INFO [train.py:968] (0/2) Epoch 13, batch 45050, giga_loss[loss=0.3274, simple_loss=0.3944, pruned_loss=0.1302, over 29014.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3736, pruned_loss=0.1245, over 5669056.24 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3671, pruned_loss=0.1205, over 5739971.67 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3757, pruned_loss=0.1258, over 5649172.77 frames. ], batch size: 155, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:23:18,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.891e+02 1.352e+03 1.843e+03 2.469e+03 9.408e+03, threshold=3.686e+03, percent-clipped=12.0
+2023-03-07 02:23:26,874 INFO [train.py:968] (0/2) Epoch 13, batch 45100, libri_loss[loss=0.3362, simple_loss=0.3984, pruned_loss=0.137, over 29535.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3697, pruned_loss=0.1207, over 5659744.40 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3672, pruned_loss=0.1205, over 5733945.17 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3714, pruned_loss=0.1218, over 5647193.85 frames. ], batch size: 84, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:24:16,188 INFO [train.py:968] (0/2) Epoch 13, batch 45150, giga_loss[loss=0.3568, simple_loss=0.4018, pruned_loss=0.1559, over 27729.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3694, pruned_loss=0.1208, over 5658668.36 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3669, pruned_loss=0.1203, over 5737289.95 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.371, pruned_loss=0.1219, over 5643712.09 frames. ], batch size: 474, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:24:43,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=592649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:24:46,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=592652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:24:49,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.263e+02 1.436e+03 2.045e+03 2.785e+03 8.255e+03, threshold=4.090e+03, percent-clipped=9.0
+2023-03-07 02:24:56,854 INFO [train.py:968] (0/2) Epoch 13, batch 45200, giga_loss[loss=0.2611, simple_loss=0.3362, pruned_loss=0.09299, over 28509.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3673, pruned_loss=0.1195, over 5680452.65 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3668, pruned_loss=0.12, over 5742596.33 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3688, pruned_loss=0.1205, over 5661055.89 frames. ], batch size: 71, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:24:59,618 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3103, 1.8282, 1.3348, 0.5449], device='cuda:0'), covar=tensor([0.2913, 0.1585, 0.2379, 0.4021], device='cuda:0'), in_proj_covar=tensor([0.1610, 0.1533, 0.1514, 0.1327], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 02:25:03,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=592672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:25:12,941 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=592681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:25:44,844 INFO [train.py:968] (0/2) Epoch 13, batch 45250, giga_loss[loss=0.3404, simple_loss=0.3957, pruned_loss=0.1425, over 28969.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3651, pruned_loss=0.1187, over 5684324.74 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3666, pruned_loss=0.1199, over 5743302.89 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3665, pruned_loss=0.1196, over 5666629.85 frames. ], batch size: 136, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:25:45,155 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2803, 1.7145, 1.3482, 0.4677], device='cuda:0'), covar=tensor([0.2982, 0.2094, 0.3066, 0.4573], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1533, 0.1517, 0.1326], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 02:25:54,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7018, 3.5306, 3.3488, 1.7911], device='cuda:0'), covar=tensor([0.0686, 0.0798, 0.0759, 0.2286], device='cuda:0'), in_proj_covar=tensor([0.1115, 0.1038, 0.0902, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 02:25:56,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=592726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:26:22,266 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.556e+02 1.558e+03 2.071e+03 2.646e+03 9.086e+03, threshold=4.141e+03, percent-clipped=5.0
+2023-03-07 02:26:30,555 INFO [train.py:968] (0/2) Epoch 13, batch 45300, giga_loss[loss=0.3245, simple_loss=0.3842, pruned_loss=0.1324, over 28792.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3661, pruned_loss=0.1184, over 5695857.27 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3666, pruned_loss=0.1198, over 5745574.73 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3671, pruned_loss=0.1192, over 5679015.91 frames. ], batch size: 199, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:26:32,154 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 02:26:41,439 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:26:45,699 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:27:14,249 INFO [train.py:968] (0/2) Epoch 13, batch 45350, giga_loss[loss=0.2896, simple_loss=0.3634, pruned_loss=0.1079, over 28699.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3688, pruned_loss=0.1198, over 5683492.69 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3671, pruned_loss=0.12, over 5745313.22 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3692, pruned_loss=0.1202, over 5669003.15 frames. ], batch size: 262, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:27:14,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=592815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:27:17,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=592818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:27:43,404 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=592847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:27:51,619 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.295e+02 1.553e+03 2.139e+03 3.048e+03 7.461e+03, threshold=4.279e+03, percent-clipped=9.0
+2023-03-07 02:27:57,575 INFO [train.py:968] (0/2) Epoch 13, batch 45400, giga_loss[loss=0.2531, simple_loss=0.3334, pruned_loss=0.08642, over 28923.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3703, pruned_loss=0.1209, over 5670728.22 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3677, pruned_loss=0.1203, over 5735666.86 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3702, pruned_loss=0.121, over 5665878.92 frames. ], batch size: 145, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:27:58,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6331, 1.8465, 1.5094, 1.7528], device='cuda:0'), covar=tensor([0.2456, 0.2420, 0.2624, 0.2247], device='cuda:0'), in_proj_covar=tensor([0.1367, 0.1003, 0.1206, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 02:28:00,507 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=592869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:28:02,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=592872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:28:27,715 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=592901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:28:39,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1293, 1.3726, 1.3973, 1.0698], device='cuda:0'), covar=tensor([0.1164, 0.1853, 0.0967, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0849, 0.0699, 0.0888, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-07 02:28:40,218 INFO [train.py:968] (0/2) Epoch 13, batch 45450, giga_loss[loss=0.3683, simple_loss=0.3973, pruned_loss=0.1696, over 23475.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.373, pruned_loss=0.1235, over 5668918.86 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3684, pruned_loss=0.1207, over 5736904.27 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3724, pruned_loss=0.1233, over 5662230.18 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:28:56,725 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:29:06,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4945, 1.7249, 1.3372, 1.8600], device='cuda:0'), covar=tensor([0.2354, 0.2382, 0.2621, 0.2038], device='cuda:0'), in_proj_covar=tensor([0.1361, 0.1000, 0.1201, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 02:29:17,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.869e+02 1.713e+03 2.195e+03 3.140e+03 7.632e+03, threshold=4.391e+03, percent-clipped=6.0
+2023-03-07 02:29:25,121 INFO [train.py:968] (0/2) Epoch 13, batch 45500, giga_loss[loss=0.3284, simple_loss=0.4007, pruned_loss=0.128, over 28936.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3753, pruned_loss=0.1252, over 5668818.52 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3687, pruned_loss=0.1206, over 5740078.57 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3747, pruned_loss=0.1253, over 5658376.09 frames. ], batch size: 213, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:29:37,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:30:12,053 INFO [train.py:968] (0/2) Epoch 13, batch 45550, giga_loss[loss=0.3141, simple_loss=0.3845, pruned_loss=0.1219, over 28892.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3778, pruned_loss=0.1273, over 5635277.17 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3693, pruned_loss=0.1212, over 5724556.70 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3769, pruned_loss=0.1269, over 5638337.93 frames. ], batch size: 145, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:30:45,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.602e+03 2.375e+03 3.312e+03 8.387e+03, threshold=4.750e+03, percent-clipped=13.0
+2023-03-07 02:30:52,667 INFO [train.py:968] (0/2) Epoch 13, batch 45600, giga_loss[loss=0.3589, simple_loss=0.4098, pruned_loss=0.154, over 28256.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3783, pruned_loss=0.1273, over 5654241.49 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3696, pruned_loss=0.1215, over 5726283.58 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3778, pruned_loss=0.1271, over 5651311.32 frames. ], batch size: 368, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:31:14,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1531, 1.2867, 1.1396, 0.9260], device='cuda:0'), covar=tensor([0.0930, 0.0512, 0.1012, 0.1014], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0445, 0.0508, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 02:31:26,333 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 02:31:37,497 INFO [train.py:968] (0/2) Epoch 13, batch 45650, libri_loss[loss=0.3476, simple_loss=0.3968, pruned_loss=0.1492, over 19116.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3796, pruned_loss=0.1288, over 5643293.01 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3691, pruned_loss=0.1212, over 5718624.91 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3798, pruned_loss=0.129, over 5646330.09 frames. ], batch size: 187, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:32:15,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:32:17,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.702e+03 2.113e+03 3.124e+03 6.589e+03, threshold=4.227e+03, percent-clipped=4.0
+2023-03-07 02:32:18,842 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:32:25,847 INFO [train.py:968] (0/2) Epoch 13, batch 45700, giga_loss[loss=0.2643, simple_loss=0.3481, pruned_loss=0.09023, over 28972.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3792, pruned_loss=0.1284, over 5619973.39 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3696, pruned_loss=0.1217, over 5681307.24 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.379, pruned_loss=0.1282, over 5655382.57 frames. ], batch size: 164, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:33:20,419 INFO [train.py:968] (0/2) Epoch 13, batch 45750, giga_loss[loss=0.3053, simple_loss=0.3843, pruned_loss=0.1131, over 28593.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3801, pruned_loss=0.1281, over 5586426.68 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1221, over 5645789.65 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3797, pruned_loss=0.1276, over 5646029.59 frames. ], batch size: 336, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:33:56,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 2.012e+03 2.421e+03 3.574e+03 7.592e+03, threshold=4.843e+03, percent-clipped=19.0
+2023-03-07 02:34:07,227 INFO [train.py:968] (0/2) Epoch 13, batch 45800, giga_loss[loss=0.3842, simple_loss=0.421, pruned_loss=0.1737, over 27603.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3796, pruned_loss=0.1276, over 5555569.68 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3706, pruned_loss=0.1227, over 5586971.23 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3791, pruned_loss=0.1268, over 5653600.17 frames. ], batch size: 472, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:34:14,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6466, 3.4607, 3.2884, 1.8299], device='cuda:0'), covar=tensor([0.0791, 0.0897, 0.0865, 0.2197], device='cuda:0'), in_proj_covar=tensor([0.1117, 0.1039, 0.0907, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 02:34:16,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3123, 3.3927, 1.5182, 1.4667], device='cuda:0'), covar=tensor([0.0959, 0.0321, 0.0849, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0527, 0.0353, 0.0402], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 02:34:27,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8996, 1.2304, 1.1861, 1.0387], device='cuda:0'), covar=tensor([0.1300, 0.0996, 0.1703, 0.1155], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0735, 0.0680, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 02:34:37,418 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:34:40,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:34:42,251 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-07 02:34:45,873 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-13.pt
+2023-03-07 02:35:13,685 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:35:16,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:35:21,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:35:37,876 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=593323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:35:40,837 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:35:46,970 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:36:00,079 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=593348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:36:01,630 INFO [train.py:968] (0/2) Epoch 14, batch 50, giga_loss[loss=0.296, simple_loss=0.3722, pruned_loss=0.1098, over 27904.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3776, pruned_loss=0.1124, over 1267794.71 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3466, pruned_loss=0.09358, over 259809.50 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3848, pruned_loss=0.1167, over 1055457.99 frames. ], batch size: 412, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:36:04,967 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:36:07,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.717e+02 1.409e+03 1.876e+03 2.196e+03 5.369e+03, threshold=3.752e+03, percent-clipped=1.0
+2023-03-07 02:36:47,192 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=593395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:36:50,566 INFO [train.py:968] (0/2) Epoch 14, batch 100, giga_loss[loss=0.3197, simple_loss=0.3904, pruned_loss=0.1245, over 28789.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3688, pruned_loss=0.1075, over 2252377.53 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3437, pruned_loss=0.09158, over 426436.34 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3737, pruned_loss=0.1105, over 1971744.12 frames. ], batch size: 99, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:37:12,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3156, 1.6027, 1.6169, 1.1931], device='cuda:0'), covar=tensor([0.1781, 0.2449, 0.1448, 0.1701], device='cuda:0'), in_proj_covar=tensor([0.0848, 0.0697, 0.0890, 0.0794], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 02:37:36,284 INFO [train.py:968] (0/2) Epoch 14, batch 150, giga_loss[loss=0.2259, simple_loss=0.3005, pruned_loss=0.07565, over 28779.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3551, pruned_loss=0.1019, over 3016888.52 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3404, pruned_loss=0.09049, over 561539.62 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3582, pruned_loss=0.104, over 2720626.99 frames. ], batch size: 119, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 02:37:36,627 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:37:39,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:37:42,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.953e+02 1.209e+03 1.572e+03 2.423e+03 1.050e+04, threshold=3.145e+03, percent-clipped=10.0
+2023-03-07 02:38:04,384 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:38:14,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4384, 3.2510, 1.6012, 1.4904], device='cuda:0'), covar=tensor([0.0962, 0.0273, 0.0902, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0523, 0.0352, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 02:38:16,167 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:38:18,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:38:18,859 INFO [train.py:968] (0/2) Epoch 14, batch 200, giga_loss[loss=0.2414, simple_loss=0.3091, pruned_loss=0.08682, over 28959.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3403, pruned_loss=0.09471, over 3617274.21 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3399, pruned_loss=0.0903, over 588255.93 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3414, pruned_loss=0.09563, over 3374213.48 frames. ], batch size: 106, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 02:38:42,717 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:38:47,146 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-07 02:39:02,159 INFO [train.py:968] (0/2) Epoch 14, batch 250, giga_loss[loss=0.2268, simple_loss=0.3047, pruned_loss=0.07448, over 28664.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3283, pruned_loss=0.08859, over 4082059.22 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3399, pruned_loss=0.09057, over 667533.12 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3283, pruned_loss=0.0889, over 3862979.73 frames. ], batch size: 262, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 02:39:06,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.373e+02 8.984e+02 1.235e+03 1.795e+03 5.164e+03, threshold=2.469e+03, percent-clipped=4.0
+2023-03-07 02:39:41,425 INFO [train.py:968] (0/2) Epoch 14, batch 300, libri_loss[loss=0.2211, simple_loss=0.304, pruned_loss=0.06904, over 29632.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3208, pruned_loss=0.08498, over 4445774.21 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3408, pruned_loss=0.08921, over 872865.76 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3196, pruned_loss=0.08512, over 4210086.49 frames. ], batch size: 69, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 02:40:01,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4304, 2.9168, 1.5330, 1.4321], device='cuda:0'), covar=tensor([0.0925, 0.0349, 0.0867, 0.1328], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0523, 0.0353, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 02:40:27,262 INFO [train.py:968] (0/2) Epoch 14, batch 350, giga_loss[loss=0.2133, simple_loss=0.2903, pruned_loss=0.06818, over 29003.00 frames. ], tot_loss[loss=0.2381, simple_loss=0.313, pruned_loss=0.08167, over 4726444.53 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3397, pruned_loss=0.0881, over 948076.81 frames. ], giga_tot_loss[loss=0.2375, simple_loss=0.3115, pruned_loss=0.08173, over 4522798.64 frames. ], batch size: 136, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 02:40:31,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.612e+02 9.708e+02 1.256e+03 2.096e+03 6.011e+03, threshold=2.513e+03, percent-clipped=17.0
+2023-03-07 02:41:03,215 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2906, 0.7920, 0.8599, 1.4397], device='cuda:0'), covar=tensor([0.0773, 0.0371, 0.0358, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 02:41:05,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:41:06,825 INFO [train.py:968] (0/2) Epoch 14, batch 400, giga_loss[loss=0.2264, simple_loss=0.2966, pruned_loss=0.07813, over 27650.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3092, pruned_loss=0.08016, over 4949728.82 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3406, pruned_loss=0.08846, over 1070320.49 frames. ], giga_tot_loss[loss=0.2333, simple_loss=0.307, pruned_loss=0.07985, over 4763918.23 frames. ], batch size: 472, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:41:28,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:41:34,088 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=593732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:41:48,824 INFO [train.py:968] (0/2) Epoch 14, batch 450, giga_loss[loss=0.2399, simple_loss=0.2918, pruned_loss=0.09397, over 24070.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3073, pruned_loss=0.07937, over 5120217.02 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3417, pruned_loss=0.08915, over 1189903.89 frames. ], giga_tot_loss[loss=0.2309, simple_loss=0.3044, pruned_loss=0.07873, over 4952628.80 frames. ], batch size: 705, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:41:55,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.821e+02 1.100e+03 1.428e+03 1.899e+03 5.779e+03, threshold=2.857e+03, percent-clipped=14.0
+2023-03-07 02:42:06,849 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 02:42:07,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:42:28,406 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3153, 1.1951, 1.1764, 1.5431], device='cuda:0'), covar=tensor([0.0759, 0.0359, 0.0341, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 02:42:31,527 INFO [train.py:968] (0/2) Epoch 14, batch 500, giga_loss[loss=0.2167, simple_loss=0.2929, pruned_loss=0.07025, over 27964.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3045, pruned_loss=0.07789, over 5259185.93 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3434, pruned_loss=0.09044, over 1306218.10 frames. ], giga_tot_loss[loss=0.2273, simple_loss=0.3009, pruned_loss=0.07679, over 5109988.79 frames. ], batch size: 412, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:43:06,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:10,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:14,745 INFO [train.py:968] (0/2) Epoch 14, batch 550, giga_loss[loss=0.2274, simple_loss=0.3024, pruned_loss=0.07621, over 28965.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3025, pruned_loss=0.0771, over 5358550.95 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3434, pruned_loss=0.09068, over 1397448.79 frames. ], giga_tot_loss[loss=0.2253, simple_loss=0.2988, pruned_loss=0.07589, over 5229442.12 frames. ], batch size: 227, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:43:19,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.159e+02 9.623e+02 1.234e+03 1.687e+03 4.154e+03, threshold=2.467e+03, percent-clipped=6.0
+2023-03-07 02:43:30,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0474, 1.9772, 1.5286, 1.5832], device='cuda:0'), covar=tensor([0.0844, 0.0744, 0.1024, 0.1258], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0441, 0.0504, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 02:43:30,848 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:32,892 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:36,621 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:57,518 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:58,721 INFO [train.py:968] (0/2) Epoch 14, batch 600, giga_loss[loss=0.2108, simple_loss=0.285, pruned_loss=0.06832, over 28269.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3018, pruned_loss=0.07686, over 5433252.68 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.344, pruned_loss=0.09087, over 1548766.41 frames. ], giga_tot_loss[loss=0.224, simple_loss=0.2973, pruned_loss=0.07536, over 5316845.77 frames. ], batch size: 368, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:44:08,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6821, 1.7750, 1.3716, 1.3666], device='cuda:0'), covar=tensor([0.0884, 0.0651, 0.1101, 0.1140], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0442, 0.0506, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 02:44:10,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:44:12,835 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:44:37,361 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:44:43,256 INFO [train.py:968] (0/2) Epoch 14, batch 650, giga_loss[loss=0.2429, simple_loss=0.3092, pruned_loss=0.08834, over 28285.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.2998, pruned_loss=0.07578, over 5486365.24 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3424, pruned_loss=0.09029, over 1677784.34 frames. ], giga_tot_loss[loss=0.222, simple_loss=0.2953, pruned_loss=0.07433, over 5381250.19 frames. ], batch size: 369, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:44:49,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.098e+02 9.294e+02 1.386e+03 1.718e+03 7.711e+03, threshold=2.772e+03, percent-clipped=12.0
+2023-03-07 02:45:18,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4059, 1.3629, 1.5688, 1.1631], device='cuda:0'), covar=tensor([0.1593, 0.2695, 0.1254, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0704, 0.0904, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 02:45:29,087 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-594000.pt
+2023-03-07 02:45:29,381 INFO [train.py:968] (0/2) Epoch 14, batch 700, giga_loss[loss=0.2139, simple_loss=0.2945, pruned_loss=0.06667, over 28658.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.2962, pruned_loss=0.07416, over 5532837.03 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3423, pruned_loss=0.09035, over 1698968.42 frames. ], giga_tot_loss[loss=0.2192, simple_loss=0.2925, pruned_loss=0.07293, over 5448879.52 frames. ], batch size: 242, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:46:07,295 INFO [train.py:968] (0/2) Epoch 14, batch 750, giga_loss[loss=0.2012, simple_loss=0.2779, pruned_loss=0.0622, over 28904.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2957, pruned_loss=0.07357, over 5587820.98 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.342, pruned_loss=0.0902, over 1925476.50 frames. ], giga_tot_loss[loss=0.2168, simple_loss=0.2902, pruned_loss=0.07168, over 5500644.07 frames. ], batch size: 145, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:46:15,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.674e+02 1.012e+03 1.263e+03 1.812e+03 5.466e+03, threshold=2.525e+03, percent-clipped=5.0
+2023-03-07 02:46:49,394 INFO [train.py:968] (0/2) Epoch 14, batch 800, libri_loss[loss=0.2382, simple_loss=0.3203, pruned_loss=0.0781, over 29520.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2946, pruned_loss=0.07327, over 5622643.42 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3422, pruned_loss=0.09052, over 2118307.80 frames. ], giga_tot_loss[loss=0.2147, simple_loss=0.2878, pruned_loss=0.07079, over 5538958.58 frames. ], batch size: 80, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:46:54,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=594107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:47:35,230 INFO [train.py:968] (0/2) Epoch 14, batch 850, giga_loss[loss=0.2609, simple_loss=0.3313, pruned_loss=0.09524, over 28929.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3018, pruned_loss=0.07796, over 5628500.74 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3432, pruned_loss=0.09149, over 2229670.01 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2947, pruned_loss=0.07524, over 5554361.48 frames. ], batch size: 227, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:47:41,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.997e+02 1.053e+03 1.504e+03 2.241e+03 4.472e+03, threshold=3.007e+03, percent-clipped=13.0
+2023-03-07 02:48:15,406 INFO [train.py:968] (0/2) Epoch 14, batch 900, giga_loss[loss=0.3455, simple_loss=0.4009, pruned_loss=0.145, over 27705.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3142, pruned_loss=0.08424, over 5648348.00 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.344, pruned_loss=0.09196, over 2453641.31 frames. ], giga_tot_loss[loss=0.2342, simple_loss=0.306, pruned_loss=0.08119, over 5578452.16 frames. ], batch size: 472, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:49:00,762 INFO [train.py:968] (0/2) Epoch 14, batch 950, giga_loss[loss=0.2589, simple_loss=0.3378, pruned_loss=0.09002, over 29094.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3268, pruned_loss=0.09092, over 5662502.62 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3439, pruned_loss=0.0919, over 2522474.94 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.32, pruned_loss=0.08851, over 5602340.31 frames. ], batch size: 155, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:49:01,061 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=594250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:49:04,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=594253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:49:04,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3604, 1.6172, 1.3319, 1.3267], device='cuda:0'), covar=tensor([0.2568, 0.2418, 0.2773, 0.2106], device='cuda:0'), in_proj_covar=tensor([0.1358, 0.0990, 0.1200, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 02:49:07,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.887e+02 1.302e+03 1.625e+03 2.561e+03 6.238e+03, threshold=3.251e+03, percent-clipped=17.0
+2023-03-07 02:49:21,405 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=594273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:49:29,795 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=594282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:49:43,199 INFO [train.py:968] (0/2) Epoch 14, batch 1000, giga_loss[loss=0.2962, simple_loss=0.3761, pruned_loss=0.1082, over 28603.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3359, pruned_loss=0.09505, over 5674285.56 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3432, pruned_loss=0.09148, over 2589522.39 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3306, pruned_loss=0.09336, over 5622370.44 frames. ], batch size: 307, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:49:54,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3430, 1.5102, 1.3735, 1.2230], device='cuda:0'), covar=tensor([0.2196, 0.2032, 0.1497, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.1803, 0.1711, 0.1666, 0.1763], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 02:50:21,224 INFO [train.py:968] (0/2) Epoch 14, batch 1050, giga_loss[loss=0.2373, simple_loss=0.3232, pruned_loss=0.07574, over 28885.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3408, pruned_loss=0.0963, over 5673773.81 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.344, pruned_loss=0.09192, over 2688561.47 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3362, pruned_loss=0.0949, over 5639126.46 frames. ], batch size: 213, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:50:27,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.893e+02 1.225e+03 1.410e+03 1.977e+03 6.369e+03, threshold=2.821e+03, percent-clipped=6.0
+2023-03-07 02:51:06,476 INFO [train.py:968] (0/2) Epoch 14, batch 1100, giga_loss[loss=0.2558, simple_loss=0.3435, pruned_loss=0.08404, over 29102.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3442, pruned_loss=0.09712, over 5666891.02 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3453, pruned_loss=0.09284, over 2748732.99 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3399, pruned_loss=0.09572, over 5638763.36 frames. ], batch size: 155, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:51:49,896 INFO [train.py:968] (0/2) Epoch 14, batch 1150, giga_loss[loss=0.2922, simple_loss=0.3699, pruned_loss=0.1072, over 28896.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3457, pruned_loss=0.09773, over 5683270.11 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3455, pruned_loss=0.09289, over 2779813.56 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3423, pruned_loss=0.09665, over 5659442.30 frames. ], batch size: 186, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:51:57,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-07 02:51:57,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.123e+02 1.067e+03 1.319e+03 1.840e+03 4.730e+03, threshold=2.637e+03, percent-clipped=7.0
+2023-03-07 02:52:34,571 INFO [train.py:968] (0/2) Epoch 14, batch 1200, giga_loss[loss=0.2725, simple_loss=0.3515, pruned_loss=0.09672, over 28442.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3469, pruned_loss=0.09869, over 5673626.00 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3463, pruned_loss=0.09329, over 2865542.01 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3437, pruned_loss=0.0978, over 5654583.60 frames. ], batch size: 71, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:52:56,898 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3888, 1.1613, 4.0355, 3.1166], device='cuda:0'), covar=tensor([0.1613, 0.2835, 0.0412, 0.0934], device='cuda:0'), in_proj_covar=tensor([0.0679, 0.0600, 0.0877, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 02:53:16,891 INFO [train.py:968] (0/2) Epoch 14, batch 1250, giga_loss[loss=0.2866, simple_loss=0.3616, pruned_loss=0.1058, over 28976.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3491, pruned_loss=0.1001, over 5681445.48 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.346, pruned_loss=0.09304, over 2951535.82 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3468, pruned_loss=0.09967, over 5663932.39 frames. ], batch size: 186, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:53:24,761 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.305e+02 1.092e+03 1.369e+03 1.760e+03 4.266e+03, threshold=2.737e+03, percent-clipped=8.0
+2023-03-07 02:53:25,626 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=594558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:54:00,453 INFO [train.py:968] (0/2) Epoch 14, batch 1300, giga_loss[loss=0.2886, simple_loss=0.3661, pruned_loss=0.1055, over 28827.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3519, pruned_loss=0.1015, over 5674900.67 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3465, pruned_loss=0.09319, over 3009692.19 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3498, pruned_loss=0.1013, over 5657127.74 frames. ], batch size: 199, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:54:11,525 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=594614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:54:37,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=594648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:54:38,543 INFO [train.py:968] (0/2) Epoch 14, batch 1350, libri_loss[loss=0.2776, simple_loss=0.3624, pruned_loss=0.0964, over 29527.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3542, pruned_loss=0.1016, over 5692185.52 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3476, pruned_loss=0.09379, over 3104619.05 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3522, pruned_loss=0.1013, over 5675543.46 frames. ], batch size: 89, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:54:43,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.066e+02 1.193e+03 1.524e+03 2.272e+03 4.021e+03, threshold=3.049e+03, percent-clipped=14.0
+2023-03-07 02:54:50,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 02:55:18,685 INFO [train.py:968] (0/2) Epoch 14, batch 1400, giga_loss[loss=0.306, simple_loss=0.3748, pruned_loss=0.1186, over 28771.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3554, pruned_loss=0.1018, over 5695102.45 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3474, pruned_loss=0.09363, over 3145394.39 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3541, pruned_loss=0.1017, over 5680108.06 frames. ], batch size: 119, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:55:40,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1298, 1.2661, 1.0710, 0.8997], device='cuda:0'), covar=tensor([0.1001, 0.0545, 0.1118, 0.1091], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0444, 0.0509, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 02:55:45,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=594733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:55:50,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5411, 1.6233, 1.8167, 1.3686], device='cuda:0'), covar=tensor([0.1676, 0.2248, 0.1346, 0.1597], device='cuda:0'), in_proj_covar=tensor([0.0855, 0.0697, 0.0897, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 02:55:59,064 INFO [train.py:968] (0/2) Epoch 14, batch 1450, giga_loss[loss=0.2712, simple_loss=0.3675, pruned_loss=0.08744, over 29026.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3562, pruned_loss=0.1017, over 5696105.79 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3472, pruned_loss=0.09348, over 3240136.02 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3555, pruned_loss=0.102, over 5677943.20 frames. ], batch size: 164, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:56:05,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.741e+02 1.144e+03 1.328e+03 1.707e+03 3.728e+03, threshold=2.656e+03, percent-clipped=5.0
+2023-03-07 02:56:21,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2308, 0.9072, 0.9713, 1.3741], device='cuda:0'), covar=tensor([0.0786, 0.0370, 0.0339, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 02:56:32,581 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=594791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:56:35,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=594794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:56:38,896 INFO [train.py:968] (0/2) Epoch 14, batch 1500, libri_loss[loss=0.2986, simple_loss=0.3833, pruned_loss=0.1069, over 25833.00 frames. ], tot_loss[loss=0.277, simple_loss=0.355, pruned_loss=0.09948, over 5705700.19 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3483, pruned_loss=0.09394, over 3300424.30 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.354, pruned_loss=0.09962, over 5692182.97 frames. ], batch size: 136, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:56:56,533 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=594823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:56:56,583 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=594823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:57:11,301 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 02:57:15,738 INFO [train.py:968] (0/2) Epoch 14, batch 1550, giga_loss[loss=0.284, simple_loss=0.3678, pruned_loss=0.1001, over 28760.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3521, pruned_loss=0.09677, over 5701445.43 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3478, pruned_loss=0.09347, over 3392655.05 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3517, pruned_loss=0.09729, over 5692214.21 frames. ], batch size: 284, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:57:18,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2138, 1.1693, 1.2157, 1.4486], device='cuda:0'), covar=tensor([0.0807, 0.0363, 0.0339, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 02:57:23,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.175e+02 1.112e+03 1.394e+03 1.784e+03 3.694e+03, threshold=2.788e+03, percent-clipped=1.0
+2023-03-07 02:57:41,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4596, 1.7439, 1.7225, 1.2669], device='cuda:0'), covar=tensor([0.1741, 0.2378, 0.1392, 0.1670], device='cuda:0'), in_proj_covar=tensor([0.0861, 0.0701, 0.0903, 0.0804], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 02:57:55,228 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 02:57:58,098 INFO [train.py:968] (0/2) Epoch 14, batch 1600, giga_loss[loss=0.2614, simple_loss=0.3424, pruned_loss=0.09017, over 28994.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3512, pruned_loss=0.09651, over 5708171.25 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3469, pruned_loss=0.09297, over 3454475.97 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3515, pruned_loss=0.09727, over 5696680.03 frames. ], batch size: 145, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:58:27,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=594933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:58:42,580 INFO [train.py:968] (0/2) Epoch 14, batch 1650, giga_loss[loss=0.2889, simple_loss=0.3655, pruned_loss=0.1062, over 28796.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.354, pruned_loss=0.1009, over 5711923.80 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3472, pruned_loss=0.09303, over 3490828.64 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3541, pruned_loss=0.1016, over 5700486.97 frames. ], batch size: 119, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:58:52,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.195e+02 1.168e+03 1.495e+03 2.187e+03 7.628e+03, threshold=2.991e+03, percent-clipped=11.0
+2023-03-07 02:59:19,692 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=594989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:59:28,793 INFO [train.py:968] (0/2) Epoch 14, batch 1700, giga_loss[loss=0.2877, simple_loss=0.349, pruned_loss=0.1132, over 28966.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3558, pruned_loss=0.1044, over 5691245.11 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3475, pruned_loss=0.09325, over 3520268.74 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3559, pruned_loss=0.105, over 5693870.11 frames. ], batch size: 106, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:00:09,897 INFO [train.py:968] (0/2) Epoch 14, batch 1750, libri_loss[loss=0.271, simple_loss=0.3555, pruned_loss=0.09326, over 29545.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 5687290.06 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.347, pruned_loss=0.09271, over 3589844.87 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3552, pruned_loss=0.1061, over 5684419.75 frames. ], batch size: 78, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:00:12,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6487, 2.3853, 1.6492, 0.7945], device='cuda:0'), covar=tensor([0.3983, 0.2195, 0.3350, 0.4780], device='cuda:0'), in_proj_covar=tensor([0.1608, 0.1518, 0.1509, 0.1322], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 03:00:17,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.220e+02 1.186e+03 1.547e+03 1.990e+03 3.386e+03, threshold=3.094e+03, percent-clipped=5.0
+2023-03-07 03:00:24,739 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=595066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:00:32,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:00:34,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:00:51,825 INFO [train.py:968] (0/2) Epoch 14, batch 1800, giga_loss[loss=0.2958, simple_loss=0.3671, pruned_loss=0.1122, over 28691.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3522, pruned_loss=0.1036, over 5704631.79 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3469, pruned_loss=0.09253, over 3646887.32 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3529, pruned_loss=0.1048, over 5697900.43 frames. ], batch size: 262, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:00:57,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:00:57,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=595108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:00:59,799 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5999, 1.7252, 1.8511, 1.4060], device='cuda:0'), covar=tensor([0.1767, 0.2383, 0.1375, 0.1609], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0697, 0.0896, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 03:01:16,669 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:01:18,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:01:25,623 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=595144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:01:29,391 INFO [train.py:968] (0/2) Epoch 14, batch 1850, giga_loss[loss=0.2593, simple_loss=0.3346, pruned_loss=0.09198, over 28606.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3519, pruned_loss=0.1034, over 5698638.63 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3465, pruned_loss=0.0923, over 3741563.16 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3529, pruned_loss=0.105, over 5698393.07 frames. ], batch size: 92, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 03:01:37,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2031, 1.2670, 3.7858, 3.1374], device='cuda:0'), covar=tensor([0.1610, 0.2504, 0.0440, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0604, 0.0883, 0.0801], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 03:01:38,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.292e+02 1.220e+03 1.539e+03 2.297e+03 5.300e+03, threshold=3.077e+03, percent-clipped=9.0
+2023-03-07 03:01:40,630 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:02:08,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=595198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:02:09,582 INFO [train.py:968] (0/2) Epoch 14, batch 1900, giga_loss[loss=0.2667, simple_loss=0.345, pruned_loss=0.09423, over 29055.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3507, pruned_loss=0.1015, over 5712873.85 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3464, pruned_loss=0.09238, over 3805703.13 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3517, pruned_loss=0.103, over 5707278.71 frames. ], batch size: 128, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 03:02:38,167 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4369, 1.9334, 1.3566, 0.7701], device='cuda:0'), covar=tensor([0.4963, 0.2212, 0.3097, 0.5113], device='cuda:0'), in_proj_covar=tensor([0.1599, 0.1510, 0.1499, 0.1313], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 03:02:52,460 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=595245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:02:58,757 INFO [train.py:968] (0/2) Epoch 14, batch 1950, giga_loss[loss=0.2664, simple_loss=0.3474, pruned_loss=0.09271, over 28245.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3481, pruned_loss=0.0998, over 5693191.35 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3464, pruned_loss=0.09232, over 3837236.05 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3489, pruned_loss=0.1012, over 5686095.29 frames. ], batch size: 368, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 03:03:00,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:03:02,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:03:08,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.712e+02 1.157e+03 1.521e+03 2.422e+03 1.083e+04, threshold=3.043e+03, percent-clipped=15.0
+2023-03-07 03:03:09,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9818, 1.1947, 3.3802, 2.8641], device='cuda:0'), covar=tensor([0.1728, 0.2663, 0.0467, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0680, 0.0603, 0.0879, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 03:03:27,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:03:27,305 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-07 03:03:41,512 INFO [train.py:968] (0/2) Epoch 14, batch 2000, giga_loss[loss=0.2402, simple_loss=0.3173, pruned_loss=0.08149, over 28528.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3436, pruned_loss=0.09714, over 5697266.00 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3463, pruned_loss=0.09213, over 3898572.89 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3442, pruned_loss=0.09854, over 5685853.78 frames. ], batch size: 307, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:04:21,018 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:04:23,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:04:28,524 INFO [train.py:968] (0/2) Epoch 14, batch 2050, giga_loss[loss=0.2749, simple_loss=0.3292, pruned_loss=0.1103, over 27680.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3376, pruned_loss=0.09413, over 5685932.48 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3462, pruned_loss=0.09213, over 3919204.89 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3381, pruned_loss=0.09528, over 5681880.06 frames. ], batch size: 472, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:04:38,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3857, 1.7964, 1.3998, 1.5694], device='cuda:0'), covar=tensor([0.0727, 0.0300, 0.0303, 0.0776], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0113, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 03:04:39,280 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.161e+02 9.119e+02 1.105e+03 1.508e+03 3.699e+03, threshold=2.210e+03, percent-clipped=1.0
+2023-03-07 03:04:48,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:05:14,255 INFO [train.py:968] (0/2) Epoch 14, batch 2100, giga_loss[loss=0.2095, simple_loss=0.2906, pruned_loss=0.06418, over 28744.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3329, pruned_loss=0.09188, over 5671394.89 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3468, pruned_loss=0.09259, over 3974355.79 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3326, pruned_loss=0.09256, over 5673280.31 frames. ], batch size: 119, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:05:41,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5531, 1.6541, 1.3520, 1.6792], device='cuda:0'), covar=tensor([0.2352, 0.2458, 0.2632, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.1355, 0.0994, 0.1197, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 03:05:48,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6346, 1.8025, 1.5408, 1.8191], device='cuda:0'), covar=tensor([0.2448, 0.2588, 0.2704, 0.2233], device='cuda:0'), in_proj_covar=tensor([0.1357, 0.0996, 0.1199, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 03:05:49,923 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=595441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:05:57,076 INFO [train.py:968] (0/2) Epoch 14, batch 2150, giga_loss[loss=0.2587, simple_loss=0.3343, pruned_loss=0.0916, over 28895.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3332, pruned_loss=0.09173, over 5682902.72 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3468, pruned_loss=0.0926, over 3984039.01 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3329, pruned_loss=0.09224, over 5683289.19 frames. ], batch size: 186, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:06:06,321 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.895e+02 9.333e+02 1.190e+03 1.509e+03 4.301e+03, threshold=2.380e+03, percent-clipped=7.0
+2023-03-07 03:06:38,053 INFO [train.py:968] (0/2) Epoch 14, batch 2200, giga_loss[loss=0.3039, simple_loss=0.3643, pruned_loss=0.1218, over 28739.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3333, pruned_loss=0.09154, over 5690865.69 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3474, pruned_loss=0.09296, over 4002875.25 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3327, pruned_loss=0.09172, over 5689483.83 frames. ], batch size: 242, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:06:54,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=595519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:07:00,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7682, 1.9205, 2.0576, 1.5435], device='cuda:0'), covar=tensor([0.1882, 0.2243, 0.1421, 0.1649], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0697, 0.0897, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 03:07:17,213 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.38 vs. limit=5.0
+2023-03-07 03:07:20,484 INFO [train.py:968] (0/2) Epoch 14, batch 2250, giga_loss[loss=0.2069, simple_loss=0.2885, pruned_loss=0.06264, over 28704.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3317, pruned_loss=0.09088, over 5690739.05 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3475, pruned_loss=0.09303, over 4029844.14 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3309, pruned_loss=0.09097, over 5688159.85 frames. ], batch size: 92, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:07:29,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.043e+02 9.743e+02 1.296e+03 1.584e+03 3.815e+03, threshold=2.591e+03, percent-clipped=9.0
+2023-03-07 03:07:48,739 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:07:51,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:08:00,616 INFO [train.py:968] (0/2) Epoch 14, batch 2300, libri_loss[loss=0.2621, simple_loss=0.3613, pruned_loss=0.08146, over 26037.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3289, pruned_loss=0.08958, over 5699539.01 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3474, pruned_loss=0.09272, over 4063333.59 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.328, pruned_loss=0.08981, over 5697356.41 frames. ], batch size: 136, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:08:15,120 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:08:19,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=595620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:08:43,116 INFO [train.py:968] (0/2) Epoch 14, batch 2350, giga_loss[loss=0.257, simple_loss=0.3242, pruned_loss=0.09484, over 28396.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3269, pruned_loss=0.08909, over 5706508.04 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3474, pruned_loss=0.09272, over 4063333.59 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3262, pruned_loss=0.08926, over 5704809.29 frames. ], batch size: 71, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:08:53,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.635e+02 1.076e+03 1.244e+03 1.569e+03 1.048e+04, threshold=2.488e+03, percent-clipped=7.0
+2023-03-07 03:08:54,135 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:08:56,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:09:22,075 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:09:25,747 INFO [train.py:968] (0/2) Epoch 14, batch 2400, giga_loss[loss=0.2276, simple_loss=0.307, pruned_loss=0.07406, over 28877.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3252, pruned_loss=0.08821, over 5716478.42 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3477, pruned_loss=0.09264, over 4117876.16 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3238, pruned_loss=0.08828, over 5709926.68 frames. ], batch size: 227, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 03:09:32,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0984, 3.9373, 3.7207, 1.7472], device='cuda:0'), covar=tensor([0.0553, 0.0641, 0.0636, 0.2318], device='cuda:0'), in_proj_covar=tensor([0.1077, 0.0993, 0.0870, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 03:09:55,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7604, 2.8958, 1.9023, 0.9453], device='cuda:0'), covar=tensor([0.6959, 0.2538, 0.3508, 0.5839], device='cuda:0'), in_proj_covar=tensor([0.1596, 0.1493, 0.1488, 0.1305], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 03:10:03,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3685, 1.5233, 1.3637, 1.2778], device='cuda:0'), covar=tensor([0.2419, 0.1942, 0.1784, 0.2204], device='cuda:0'), in_proj_covar=tensor([0.1771, 0.1671, 0.1640, 0.1746], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 03:10:04,845 INFO [train.py:968] (0/2) Epoch 14, batch 2450, giga_loss[loss=0.2047, simple_loss=0.2811, pruned_loss=0.06419, over 28401.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3226, pruned_loss=0.08722, over 5722350.54 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3478, pruned_loss=0.09268, over 4135769.61 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3212, pruned_loss=0.0872, over 5715472.17 frames. ], batch size: 65, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 03:10:05,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0203, 1.3029, 1.0872, 0.2400], device='cuda:0'), covar=tensor([0.3043, 0.2428, 0.3964, 0.5252], device='cuda:0'), in_proj_covar=tensor([0.1596, 0.1493, 0.1489, 0.1305], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 03:10:12,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.052e+02 9.200e+02 1.285e+03 1.937e+03 5.227e+03, threshold=2.570e+03, percent-clipped=14.0
+2023-03-07 03:10:14,168 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:10:16,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:10:19,639 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 03:10:36,821 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:10:37,091 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-07 03:10:40,289 INFO [train.py:968] (0/2) Epoch 14, batch 2500, giga_loss[loss=0.2086, simple_loss=0.2837, pruned_loss=0.06673, over 28885.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.321, pruned_loss=0.08602, over 5731991.43 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.349, pruned_loss=0.09296, over 4187506.07 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3184, pruned_loss=0.08566, over 5722431.16 frames. ], batch size: 106, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 03:11:08,960 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3463, 1.7006, 1.2792, 1.4621], device='cuda:0'), covar=tensor([0.2766, 0.2714, 0.3101, 0.2319], device='cuda:0'), in_proj_covar=tensor([0.1361, 0.0993, 0.1200, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 03:11:19,195 INFO [train.py:968] (0/2) Epoch 14, batch 2550, giga_loss[loss=0.2501, simple_loss=0.3168, pruned_loss=0.09175, over 28420.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3208, pruned_loss=0.086, over 5724152.53 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3497, pruned_loss=0.09313, over 4245433.80 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3174, pruned_loss=0.08536, over 5711979.67 frames. ], batch size: 71, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:11:27,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.473e+02 1.023e+03 1.241e+03 1.828e+03 9.414e+03, threshold=2.482e+03, percent-clipped=13.0
+2023-03-07 03:11:39,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0250, 2.1121, 1.8042, 2.4403], device='cuda:0'), covar=tensor([0.2172, 0.2356, 0.2525, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.1362, 0.0993, 0.1200, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 03:11:56,535 INFO [train.py:968] (0/2) Epoch 14, batch 2600, giga_loss[loss=0.2205, simple_loss=0.2967, pruned_loss=0.0722, over 28957.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3199, pruned_loss=0.08523, over 5729912.97 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3505, pruned_loss=0.09337, over 4291747.33 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3159, pruned_loss=0.08435, over 5717838.42 frames. ], batch size: 145, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:12:34,175 INFO [train.py:968] (0/2) Epoch 14, batch 2650, giga_loss[loss=0.2137, simple_loss=0.2945, pruned_loss=0.06644, over 28899.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3205, pruned_loss=0.08527, over 5735068.98 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3518, pruned_loss=0.09406, over 4360355.41 frames. ], giga_tot_loss[loss=0.2413, simple_loss=0.315, pruned_loss=0.08374, over 5720392.33 frames. ], batch size: 174, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:12:34,348 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=595950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:12:35,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-07 03:12:43,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.189e+02 1.021e+03 1.365e+03 1.772e+03 5.043e+03, threshold=2.730e+03, percent-clipped=10.0
+2023-03-07 03:13:15,080 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-596000.pt
+2023-03-07 03:13:15,369 INFO [train.py:968] (0/2) Epoch 14, batch 2700, giga_loss[loss=0.2577, simple_loss=0.3327, pruned_loss=0.09132, over 28956.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3211, pruned_loss=0.08592, over 5718652.53 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3524, pruned_loss=0.09446, over 4378373.44 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3158, pruned_loss=0.08429, over 5714385.15 frames. ], batch size: 164, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:13:52,744 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:14:00,098 INFO [train.py:968] (0/2) Epoch 14, batch 2750, giga_loss[loss=0.3407, simple_loss=0.3785, pruned_loss=0.1515, over 23687.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3253, pruned_loss=0.08897, over 5714718.71 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.352, pruned_loss=0.0942, over 4401117.78 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.321, pruned_loss=0.08779, over 5708789.68 frames. ], batch size: 710, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:14:11,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.447e+02 1.141e+03 1.410e+03 1.902e+03 3.636e+03, threshold=2.821e+03, percent-clipped=9.0
+2023-03-07 03:14:47,456 INFO [train.py:968] (0/2) Epoch 14, batch 2800, giga_loss[loss=0.2782, simple_loss=0.3471, pruned_loss=0.1047, over 28839.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3317, pruned_loss=0.09364, over 5702320.15 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3517, pruned_loss=0.09403, over 4408757.20 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3285, pruned_loss=0.09283, over 5696572.48 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:14:53,577 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:15:33,692 INFO [train.py:968] (0/2) Epoch 14, batch 2850, giga_loss[loss=0.2994, simple_loss=0.3716, pruned_loss=0.1136, over 28973.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3409, pruned_loss=0.09948, over 5696309.98 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.352, pruned_loss=0.09426, over 4449701.09 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3377, pruned_loss=0.09876, over 5689169.01 frames. ], batch size: 213, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:15:43,388 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.081e+02 1.170e+03 1.442e+03 1.758e+03 3.796e+03, threshold=2.885e+03, percent-clipped=4.0
+2023-03-07 03:15:46,396 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:15:54,637 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-07 03:16:23,393 INFO [train.py:968] (0/2) Epoch 14, batch 2900, giga_loss[loss=0.3001, simple_loss=0.3769, pruned_loss=0.1117, over 28591.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3472, pruned_loss=0.1026, over 5685706.07 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.352, pruned_loss=0.09435, over 4456015.58 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3446, pruned_loss=0.102, over 5680004.85 frames. ], batch size: 336, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:16:57,730 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:17:07,859 INFO [train.py:968] (0/2) Epoch 14, batch 2950, giga_loss[loss=0.2802, simple_loss=0.3523, pruned_loss=0.1041, over 28770.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.352, pruned_loss=0.1046, over 5679771.70 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3515, pruned_loss=0.0941, over 4484527.15 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3503, pruned_loss=0.1044, over 5671294.26 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:17:21,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.772e+02 1.189e+03 1.442e+03 2.094e+03 1.189e+04, threshold=2.884e+03, percent-clipped=13.0
+2023-03-07 03:17:25,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3591, 2.9749, 1.3022, 1.4789], device='cuda:0'), covar=tensor([0.0985, 0.0293, 0.0928, 0.1342], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0516, 0.0350, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 03:17:57,053 INFO [train.py:968] (0/2) Epoch 14, batch 3000, giga_loss[loss=0.3466, simple_loss=0.3842, pruned_loss=0.1545, over 23470.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3574, pruned_loss=0.1073, over 5684861.96 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3515, pruned_loss=0.09404, over 4499036.82 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3561, pruned_loss=0.1074, over 5675882.23 frames. ], batch size: 705, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:17:57,057 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 03:18:05,437 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2215, simple_loss=0.3267, pruned_loss=0.05814, over 944034.00 frames. 
+2023-03-07 03:18:05,437 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 03:18:27,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=596325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:18:48,550 INFO [train.py:968] (0/2) Epoch 14, batch 3050, giga_loss[loss=0.2601, simple_loss=0.3397, pruned_loss=0.09024, over 28716.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3547, pruned_loss=0.1052, over 5682597.88 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3512, pruned_loss=0.09391, over 4520099.19 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.354, pruned_loss=0.1055, over 5672442.44 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:19:00,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.025e+02 1.150e+03 1.442e+03 2.239e+03 5.557e+03, threshold=2.884e+03, percent-clipped=13.0
+2023-03-07 03:19:33,041 INFO [train.py:968] (0/2) Epoch 14, batch 3100, giga_loss[loss=0.2906, simple_loss=0.3641, pruned_loss=0.1086, over 29067.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3512, pruned_loss=0.1023, over 5686964.52 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3512, pruned_loss=0.09401, over 4532945.46 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3506, pruned_loss=0.1026, over 5677870.46 frames. ], batch size: 128, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:19:38,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2449, 1.2765, 1.1127, 0.9541], device='cuda:0'), covar=tensor([0.0826, 0.0421, 0.0927, 0.0933], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0441, 0.0508, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 03:19:45,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=596417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:20:14,520 INFO [train.py:968] (0/2) Epoch 14, batch 3150, giga_loss[loss=0.2541, simple_loss=0.3352, pruned_loss=0.08651, over 28681.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3504, pruned_loss=0.1017, over 5676678.80 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3513, pruned_loss=0.09434, over 4567072.32 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3499, pruned_loss=0.1019, over 5669245.78 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:20:23,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.384e+02 1.059e+03 1.301e+03 1.823e+03 5.848e+03, threshold=2.601e+03, percent-clipped=6.0
+2023-03-07 03:20:28,899 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=596468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:20:31,764 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=596471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:20:40,740 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=596482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:20:54,289 INFO [train.py:968] (0/2) Epoch 14, batch 3200, giga_loss[loss=0.2439, simple_loss=0.3293, pruned_loss=0.07925, over 28910.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3505, pruned_loss=0.1014, over 5678309.65 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3505, pruned_loss=0.09401, over 4616478.05 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3506, pruned_loss=0.102, over 5667078.39 frames. ], batch size: 145, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:20:54,494 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=596500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:21:26,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=596538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:21:35,735 INFO [train.py:968] (0/2) Epoch 14, batch 3250, giga_loss[loss=0.2815, simple_loss=0.3572, pruned_loss=0.1029, over 28924.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3527, pruned_loss=0.1027, over 5682681.77 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3503, pruned_loss=0.09387, over 4634485.80 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.353, pruned_loss=0.1035, over 5671997.15 frames. ], batch size: 106, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:21:44,294 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=596560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:21:46,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.172e+02 1.258e+03 1.667e+03 2.301e+03 6.134e+03, threshold=3.333e+03, percent-clipped=17.0
+2023-03-07 03:21:46,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=596563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:22:09,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=596592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:22:15,745 INFO [train.py:968] (0/2) Epoch 14, batch 3300, giga_loss[loss=0.3386, simple_loss=0.3868, pruned_loss=0.1452, over 26603.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3543, pruned_loss=0.1036, over 5679398.72 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3508, pruned_loss=0.09419, over 4663448.52 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3542, pruned_loss=0.1042, over 5682213.90 frames. ], batch size: 555, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:22:26,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=596612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:22:30,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5771, 5.3902, 5.0386, 2.6719], device='cuda:0'), covar=tensor([0.0394, 0.0528, 0.0550, 0.1721], device='cuda:0'), in_proj_covar=tensor([0.1088, 0.1008, 0.0877, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 03:22:40,522 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=596625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:22:42,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=596628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:23:00,283 INFO [train.py:968] (0/2) Epoch 14, batch 3350, giga_loss[loss=0.2633, simple_loss=0.3433, pruned_loss=0.09161, over 28949.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3553, pruned_loss=0.1047, over 5681748.03 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3509, pruned_loss=0.0942, over 4675555.46 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3552, pruned_loss=0.1053, over 5682198.66 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:23:07,029 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=596657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:23:10,724 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 03:23:11,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.225e+02 1.127e+03 1.441e+03 1.987e+03 4.491e+03, threshold=2.883e+03, percent-clipped=1.0
+2023-03-07 03:23:16,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2126, 1.6047, 1.5413, 1.1285], device='cuda:0'), covar=tensor([0.1510, 0.2118, 0.1196, 0.1423], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0695, 0.0895, 0.0797], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 03:23:28,851 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=596681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:23:31,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=596684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:23:45,004 INFO [train.py:968] (0/2) Epoch 14, batch 3400, giga_loss[loss=0.2655, simple_loss=0.3439, pruned_loss=0.09353, over 28782.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3563, pruned_loss=0.1059, over 5680682.37 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3513, pruned_loss=0.09428, over 4699542.05 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3561, pruned_loss=0.1065, over 5677224.12 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:23:56,101 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=596713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:24:25,817 INFO [train.py:968] (0/2) Epoch 14, batch 3450, libri_loss[loss=0.2661, simple_loss=0.3374, pruned_loss=0.09737, over 29597.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3564, pruned_loss=0.1061, over 5680772.17 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3513, pruned_loss=0.09428, over 4723304.15 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3563, pruned_loss=0.1069, over 5674064.81 frames. ], batch size: 74, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:24:29,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=596755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:24:31,512 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=596758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:24:37,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.154e+02 1.177e+03 1.556e+03 2.042e+03 5.806e+03, threshold=3.112e+03, percent-clipped=13.0
+2023-03-07 03:24:38,026 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:24:54,228 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=596787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:25:05,414 INFO [train.py:968] (0/2) Epoch 14, batch 3500, giga_loss[loss=0.3206, simple_loss=0.3914, pruned_loss=0.1249, over 28714.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3557, pruned_loss=0.1048, over 5680916.76 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3509, pruned_loss=0.09421, over 4753167.77 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.356, pruned_loss=0.1058, over 5679177.94 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:25:44,894 INFO [train.py:968] (0/2) Epoch 14, batch 3550, giga_loss[loss=0.2465, simple_loss=0.3267, pruned_loss=0.0832, over 28699.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3553, pruned_loss=0.1033, over 5688946.27 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3505, pruned_loss=0.09396, over 4776254.31 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3559, pruned_loss=0.1044, over 5683637.72 frames. ], batch size: 85, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:25:56,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.112e+02 1.043e+03 1.342e+03 1.743e+03 4.884e+03, threshold=2.683e+03, percent-clipped=3.0
+2023-03-07 03:26:09,863 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2051, 1.2140, 1.1819, 1.5682], device='cuda:0'), covar=tensor([0.0815, 0.0356, 0.0334, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0113, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 03:26:18,898 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4273, 1.6000, 1.7235, 1.2895], device='cuda:0'), covar=tensor([0.1705, 0.2486, 0.1407, 0.1758], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0698, 0.0896, 0.0797], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 03:26:28,681 INFO [train.py:968] (0/2) Epoch 14, batch 3600, giga_loss[loss=0.2705, simple_loss=0.348, pruned_loss=0.09653, over 28885.00 frames. ], tot_loss[loss=0.28, simple_loss=0.355, pruned_loss=0.1025, over 5690690.27 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3505, pruned_loss=0.09394, over 4789953.82 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3556, pruned_loss=0.1035, over 5686330.61 frames. ], batch size: 112, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:27:05,823 INFO [train.py:968] (0/2) Epoch 14, batch 3650, giga_loss[loss=0.2542, simple_loss=0.3358, pruned_loss=0.08633, over 28998.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3535, pruned_loss=0.1018, over 5698813.20 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.35, pruned_loss=0.09366, over 4817294.06 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3543, pruned_loss=0.103, over 5691156.03 frames. ], batch size: 145, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:27:19,044 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.810e+02 1.065e+03 1.274e+03 1.570e+03 5.792e+03, threshold=2.549e+03, percent-clipped=8.0
+2023-03-07 03:27:40,395 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596989.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 03:27:50,764 INFO [train.py:968] (0/2) Epoch 14, batch 3700, giga_loss[loss=0.2903, simple_loss=0.356, pruned_loss=0.1123, over 28857.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3514, pruned_loss=0.1012, over 5699350.62 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3502, pruned_loss=0.09382, over 4828022.72 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.352, pruned_loss=0.1021, over 5691518.07 frames. ], batch size: 112, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:28:27,656 INFO [train.py:968] (0/2) Epoch 14, batch 3750, giga_loss[loss=0.3398, simple_loss=0.3779, pruned_loss=0.1508, over 26594.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3499, pruned_loss=0.1002, over 5704096.12 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3506, pruned_loss=0.09397, over 4879148.41 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3501, pruned_loss=0.1012, over 5693471.63 frames. ], batch size: 555, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:28:37,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.142e+02 1.020e+03 1.320e+03 1.723e+03 4.014e+03, threshold=2.639e+03, percent-clipped=6.0
+2023-03-07 03:29:11,416 INFO [train.py:968] (0/2) Epoch 14, batch 3800, giga_loss[loss=0.2518, simple_loss=0.3307, pruned_loss=0.08648, over 28541.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3508, pruned_loss=0.101, over 5695090.25 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3504, pruned_loss=0.09393, over 4898003.42 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.351, pruned_loss=0.102, over 5692094.10 frames. ], batch size: 71, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:29:44,329 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=597139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:29:44,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=597139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:29:48,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3059, 3.2692, 1.5169, 1.4613], device='cuda:0'), covar=tensor([0.0990, 0.0278, 0.0879, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0511, 0.0347, 0.0395], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 03:29:51,431 INFO [train.py:968] (0/2) Epoch 14, batch 3850, giga_loss[loss=0.2604, simple_loss=0.3391, pruned_loss=0.09086, over 28660.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3509, pruned_loss=0.1009, over 5703292.40 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.35, pruned_loss=0.09376, over 4926611.00 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3514, pruned_loss=0.102, over 5696375.00 frames. ], batch size: 78, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:30:00,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.580e+02 1.064e+03 1.433e+03 1.974e+03 9.413e+03, threshold=2.867e+03, percent-clipped=14.0
+2023-03-07 03:30:27,854 INFO [train.py:968] (0/2) Epoch 14, batch 3900, giga_loss[loss=0.2666, simple_loss=0.348, pruned_loss=0.09265, over 28599.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3507, pruned_loss=0.09999, over 5710294.95 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3501, pruned_loss=0.09385, over 4966199.74 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3511, pruned_loss=0.1011, over 5699868.31 frames. ], batch size: 307, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:31:07,165 INFO [train.py:968] (0/2) Epoch 14, batch 3950, giga_loss[loss=0.2895, simple_loss=0.372, pruned_loss=0.1035, over 28922.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3501, pruned_loss=0.09918, over 5707738.70 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3498, pruned_loss=0.09395, over 4998509.95 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3507, pruned_loss=0.1002, over 5703001.27 frames. ], batch size: 186, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:31:17,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.056e+02 1.006e+03 1.196e+03 1.608e+03 5.146e+03, threshold=2.391e+03, percent-clipped=6.0
+2023-03-07 03:31:33,448 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=597282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:31:35,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=597285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:31:46,304 INFO [train.py:968] (0/2) Epoch 14, batch 4000, giga_loss[loss=0.2917, simple_loss=0.3595, pruned_loss=0.112, over 28760.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3499, pruned_loss=0.09947, over 5704990.47 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3502, pruned_loss=0.09432, over 5012300.68 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3501, pruned_loss=0.1001, over 5698061.17 frames. ], batch size: 99, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:31:58,587 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=597314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:32:27,259 INFO [train.py:968] (0/2) Epoch 14, batch 4050, giga_loss[loss=0.2389, simple_loss=0.3159, pruned_loss=0.08094, over 29042.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3481, pruned_loss=0.09891, over 5710295.00 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3501, pruned_loss=0.09425, over 5023215.95 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3483, pruned_loss=0.09953, over 5704195.29 frames. ], batch size: 128, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:32:38,313 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.018e+02 9.192e+02 1.172e+03 1.543e+03 1.049e+04, threshold=2.344e+03, percent-clipped=10.0
+2023-03-07 03:32:38,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=597364.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 03:32:51,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6623, 1.8917, 1.9850, 1.5059], device='cuda:0'), covar=tensor([0.1692, 0.2281, 0.1334, 0.1588], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0700, 0.0898, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 03:33:06,359 INFO [train.py:968] (0/2) Epoch 14, batch 4100, giga_loss[loss=0.2536, simple_loss=0.3313, pruned_loss=0.08795, over 27946.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3458, pruned_loss=0.09767, over 5705559.69 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3499, pruned_loss=0.09414, over 5035598.03 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3461, pruned_loss=0.09836, over 5705219.51 frames. ], batch size: 412, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:33:11,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2922, 1.7055, 0.9360, 1.2257], device='cuda:0'), covar=tensor([0.1148, 0.0742, 0.1593, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0435, 0.0504, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 03:33:45,592 INFO [train.py:968] (0/2) Epoch 14, batch 4150, libri_loss[loss=0.2198, simple_loss=0.3043, pruned_loss=0.06764, over 29466.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3426, pruned_loss=0.09606, over 5708633.23 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3494, pruned_loss=0.09394, over 5050889.65 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3432, pruned_loss=0.09682, over 5706524.37 frames. ], batch size: 70, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:33:55,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.112e+02 1.116e+03 1.378e+03 2.036e+03 4.868e+03, threshold=2.756e+03, percent-clipped=17.0
+2023-03-07 03:34:09,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3545, 3.3694, 1.5435, 1.4234], device='cuda:0'), covar=tensor([0.0949, 0.0380, 0.0897, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0512, 0.0348, 0.0394], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 03:34:23,485 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=597498.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 03:34:25,678 INFO [train.py:968] (0/2) Epoch 14, batch 4200, giga_loss[loss=0.2821, simple_loss=0.3507, pruned_loss=0.1068, over 28558.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3433, pruned_loss=0.09726, over 5710149.22 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3494, pruned_loss=0.09406, over 5058999.71 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3437, pruned_loss=0.09779, over 5707260.72 frames. ], batch size: 78, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:34:30,798 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=597507.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 03:34:32,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=597510.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 03:34:33,370 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=597511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:34:36,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=597514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:34:56,286 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=597539.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 03:35:04,088 INFO [train.py:968] (0/2) Epoch 14, batch 4250, libri_loss[loss=0.2044, simple_loss=0.2961, pruned_loss=0.05637, over 29631.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3422, pruned_loss=0.09718, over 5714385.81 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3487, pruned_loss=0.09381, over 5080347.95 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3429, pruned_loss=0.09789, over 5707308.77 frames. ], batch size: 73, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:35:14,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.504e+02 1.094e+03 1.345e+03 1.943e+03 3.442e+03, threshold=2.689e+03, percent-clipped=7.0
+2023-03-07 03:35:43,329 INFO [train.py:968] (0/2) Epoch 14, batch 4300, giga_loss[loss=0.2875, simple_loss=0.3433, pruned_loss=0.1159, over 28733.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3419, pruned_loss=0.09745, over 5713090.72 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.09389, over 5104170.38 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3422, pruned_loss=0.09805, over 5702899.73 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:36:11,237 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3987, 2.0840, 1.6284, 0.6445], device='cuda:0'), covar=tensor([0.3828, 0.2102, 0.3414, 0.4484], device='cuda:0'), in_proj_covar=tensor([0.1595, 0.1495, 0.1502, 0.1305], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 03:36:23,251 INFO [train.py:968] (0/2) Epoch 14, batch 4350, giga_loss[loss=0.2457, simple_loss=0.3201, pruned_loss=0.08565, over 28685.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3385, pruned_loss=0.09597, over 5715590.04 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3487, pruned_loss=0.09377, over 5111677.41 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3388, pruned_loss=0.09655, over 5706335.39 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:36:29,676 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=597657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:36:31,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=597660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:36:34,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.927e+02 1.001e+03 1.207e+03 1.555e+03 3.523e+03, threshold=2.414e+03, percent-clipped=4.0
+2023-03-07 03:36:54,389 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=597689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:37:03,030 INFO [train.py:968] (0/2) Epoch 14, batch 4400, giga_loss[loss=0.2431, simple_loss=0.3152, pruned_loss=0.08553, over 28727.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.337, pruned_loss=0.09576, over 5696137.84 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3484, pruned_loss=0.09368, over 5116613.72 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3372, pruned_loss=0.09634, over 5700195.23 frames. ], batch size: 99, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:37:19,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3207, 1.5986, 1.6031, 1.2339], device='cuda:0'), covar=tensor([0.1388, 0.1849, 0.1146, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0698, 0.0896, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 03:37:39,537 INFO [train.py:968] (0/2) Epoch 14, batch 4450, giga_loss[loss=0.264, simple_loss=0.3432, pruned_loss=0.09244, over 28894.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3386, pruned_loss=0.09625, over 5697869.45 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.349, pruned_loss=0.09418, over 5126048.96 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3379, pruned_loss=0.09632, over 5706099.50 frames. ], batch size: 145, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:37:53,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.407e+02 1.038e+03 1.320e+03 1.759e+03 4.628e+03, threshold=2.641e+03, percent-clipped=14.0
+2023-03-07 03:38:20,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1843, 1.4326, 1.3407, 1.1014], device='cuda:0'), covar=tensor([0.2428, 0.2042, 0.1270, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.1771, 0.1686, 0.1653, 0.1754], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 03:38:21,265 INFO [train.py:968] (0/2) Epoch 14, batch 4500, giga_loss[loss=0.2704, simple_loss=0.3449, pruned_loss=0.0979, over 28633.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3403, pruned_loss=0.09671, over 5698042.85 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3492, pruned_loss=0.09437, over 5148590.40 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3393, pruned_loss=0.09672, over 5704302.10 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:38:52,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5871, 1.9848, 1.5290, 1.7244], device='cuda:0'), covar=tensor([0.0711, 0.0248, 0.0305, 0.0793], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 03:39:02,568 INFO [train.py:968] (0/2) Epoch 14, batch 4550, giga_loss[loss=0.3257, simple_loss=0.3752, pruned_loss=0.1381, over 23645.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3412, pruned_loss=0.09619, over 5704636.38 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3491, pruned_loss=0.09444, over 5161526.87 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3403, pruned_loss=0.09619, over 5707985.73 frames. ], batch size: 705, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:39:04,510 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=597852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:39:06,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3304, 1.6167, 1.5729, 1.4608], device='cuda:0'), covar=tensor([0.1514, 0.1762, 0.1769, 0.1743], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0731, 0.0682, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 03:39:12,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.589e+02 1.007e+03 1.302e+03 1.690e+03 8.526e+03, threshold=2.604e+03, percent-clipped=6.0
+2023-03-07 03:39:20,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=597873.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 03:39:26,667 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=597879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:39:31,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=597886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:39:38,154 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 03:39:45,556 INFO [train.py:968] (0/2) Epoch 14, batch 4600, giga_loss[loss=0.259, simple_loss=0.3472, pruned_loss=0.08545, over 28869.00 frames. ], tot_loss[loss=0.268, simple_loss=0.343, pruned_loss=0.0965, over 5695237.03 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3493, pruned_loss=0.0946, over 5163366.96 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3421, pruned_loss=0.0964, over 5703136.85 frames. ], batch size: 145, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:40:29,613 INFO [train.py:968] (0/2) Epoch 14, batch 4650, giga_loss[loss=0.2672, simple_loss=0.3447, pruned_loss=0.0949, over 27579.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3421, pruned_loss=0.09515, over 5691128.76 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3489, pruned_loss=0.0943, over 5175554.98 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3416, pruned_loss=0.09534, over 5695810.01 frames. ], batch size: 472, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:40:40,427 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.983e+02 1.000e+03 1.259e+03 1.709e+03 3.640e+03, threshold=2.518e+03, percent-clipped=6.0
+2023-03-07 03:41:06,708 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-598000.pt
+2023-03-07 03:41:07,002 INFO [train.py:968] (0/2) Epoch 14, batch 4700, giga_loss[loss=0.2559, simple_loss=0.3383, pruned_loss=0.08675, over 28652.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3421, pruned_loss=0.09459, over 5698250.52 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.349, pruned_loss=0.0941, over 5210107.15 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3413, pruned_loss=0.09494, over 5693622.02 frames. ], batch size: 242, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:41:20,797 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598016.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 03:41:22,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598019.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 03:41:30,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:41:30,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9719, 1.1439, 1.3074, 0.9744], device='cuda:0'), covar=tensor([0.1630, 0.1472, 0.2195, 0.1771], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0733, 0.0683, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 03:41:32,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:41:44,306 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=598048.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 03:41:46,595 INFO [train.py:968] (0/2) Epoch 14, batch 4750, libri_loss[loss=0.2501, simple_loss=0.3226, pruned_loss=0.08877, over 29636.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3439, pruned_loss=0.09606, over 5707743.54 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3491, pruned_loss=0.09437, over 5230144.28 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.343, pruned_loss=0.09616, over 5698889.43 frames. ], batch size: 69, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:41:55,326 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=598061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:41:58,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.390e+02 1.185e+03 1.476e+03 2.022e+03 5.379e+03, threshold=2.951e+03, percent-clipped=13.0
+2023-03-07 03:42:16,292 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-07 03:42:29,033 INFO [train.py:968] (0/2) Epoch 14, batch 4800, giga_loss[loss=0.2858, simple_loss=0.3582, pruned_loss=0.1067, over 28600.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3439, pruned_loss=0.09621, over 5709796.61 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3494, pruned_loss=0.09457, over 5235197.93 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.343, pruned_loss=0.09616, over 5702656.70 frames. ], batch size: 307, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:43:12,668 INFO [train.py:968] (0/2) Epoch 14, batch 4850, giga_loss[loss=0.2752, simple_loss=0.3545, pruned_loss=0.09797, over 28844.00 frames. ], tot_loss[loss=0.271, simple_loss=0.346, pruned_loss=0.09795, over 5712209.27 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3495, pruned_loss=0.09461, over 5247996.14 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3451, pruned_loss=0.09792, over 5703399.40 frames. ], batch size: 145, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:43:23,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.905e+02 1.116e+03 1.447e+03 1.964e+03 4.266e+03, threshold=2.893e+03, percent-clipped=6.0
+2023-03-07 03:43:31,816 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-07 03:43:52,922 INFO [train.py:968] (0/2) Epoch 14, batch 4900, giga_loss[loss=0.2763, simple_loss=0.3429, pruned_loss=0.1049, over 28691.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3489, pruned_loss=0.09977, over 5706411.58 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3496, pruned_loss=0.09475, over 5248139.08 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3481, pruned_loss=0.09969, over 5705196.17 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:44:03,799 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=598212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:44:16,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=598227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:44:35,273 INFO [train.py:968] (0/2) Epoch 14, batch 4950, giga_loss[loss=0.2591, simple_loss=0.3415, pruned_loss=0.08841, over 28993.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3511, pruned_loss=0.1009, over 5711494.60 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3497, pruned_loss=0.0948, over 5254774.95 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3504, pruned_loss=0.1008, over 5708758.74 frames. ], batch size: 155, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:44:41,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=598254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:44:49,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.234e+02 1.201e+03 1.443e+03 1.790e+03 2.969e+03, threshold=2.886e+03, percent-clipped=1.0
+2023-03-07 03:44:57,835 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-07 03:44:58,971 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2280, 1.6510, 1.2677, 1.3123], device='cuda:0'), covar=tensor([0.2508, 0.2418, 0.2817, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.1361, 0.0995, 0.1200, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 03:45:16,498 INFO [train.py:968] (0/2) Epoch 14, batch 5000, giga_loss[loss=0.2418, simple_loss=0.324, pruned_loss=0.07981, over 28710.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3518, pruned_loss=0.1011, over 5709317.28 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3495, pruned_loss=0.09468, over 5262618.10 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3514, pruned_loss=0.1013, over 5708098.53 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:45:33,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5271, 1.7480, 1.5391, 1.6187], device='cuda:0'), covar=tensor([0.1589, 0.1832, 0.2282, 0.1823], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0732, 0.0684, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 03:45:57,513 INFO [train.py:968] (0/2) Epoch 14, batch 5050, libri_loss[loss=0.2979, simple_loss=0.3787, pruned_loss=0.1086, over 29491.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3535, pruned_loss=0.1025, over 5696485.84 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3502, pruned_loss=0.09506, over 5270952.62 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3526, pruned_loss=0.1025, over 5698927.98 frames. ], batch size: 85, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:46:04,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1770, 3.9734, 3.7890, 1.7539], device='cuda:0'), covar=tensor([0.0587, 0.0736, 0.0695, 0.2138], device='cuda:0'), in_proj_covar=tensor([0.1074, 0.0998, 0.0871, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 03:46:09,004 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.424e+02 1.250e+03 1.666e+03 2.134e+03 4.327e+03, threshold=3.332e+03, percent-clipped=10.0
+2023-03-07 03:46:12,666 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:46:15,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:46:34,145 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:46:35,854 INFO [train.py:968] (0/2) Epoch 14, batch 5100, libri_loss[loss=0.3369, simple_loss=0.3948, pruned_loss=0.1395, over 19748.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3523, pruned_loss=0.1014, over 5697382.92 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3503, pruned_loss=0.09507, over 5276864.03 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3517, pruned_loss=0.1016, over 5704475.83 frames. ], batch size: 187, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:46:36,095 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:46:37,736 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=598402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:47:00,281 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=598429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:47:18,507 INFO [train.py:968] (0/2) Epoch 14, batch 5150, libri_loss[loss=0.2508, simple_loss=0.3361, pruned_loss=0.08274, over 29549.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3509, pruned_loss=0.101, over 5699553.06 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3507, pruned_loss=0.09532, over 5295550.24 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3501, pruned_loss=0.1011, over 5701357.27 frames. ], batch size: 79, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:47:18,684 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=598450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:47:32,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.622e+02 1.082e+03 1.410e+03 1.734e+03 3.982e+03, threshold=2.819e+03, percent-clipped=3.0
+2023-03-07 03:47:51,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3033, 1.5682, 1.6051, 1.1601], device='cuda:0'), covar=tensor([0.1689, 0.2526, 0.1444, 0.1665], device='cuda:0'), in_proj_covar=tensor([0.0846, 0.0690, 0.0889, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 03:48:00,510 INFO [train.py:968] (0/2) Epoch 14, batch 5200, giga_loss[loss=0.3591, simple_loss=0.4033, pruned_loss=0.1574, over 26699.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3473, pruned_loss=0.09952, over 5703434.77 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3509, pruned_loss=0.09547, over 5298843.49 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3465, pruned_loss=0.09948, over 5703676.30 frames. ], batch size: 555, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:48:40,594 INFO [train.py:968] (0/2) Epoch 14, batch 5250, giga_loss[loss=0.2451, simple_loss=0.3268, pruned_loss=0.08169, over 28567.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3455, pruned_loss=0.09794, over 5701414.85 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3512, pruned_loss=0.09566, over 5294950.45 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3445, pruned_loss=0.09781, over 5707812.33 frames. ], batch size: 336, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:48:47,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7154, 1.6304, 1.2827, 1.3131], device='cuda:0'), covar=tensor([0.0833, 0.0701, 0.1050, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0436, 0.0498, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 03:48:53,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.440e+02 1.137e+03 1.568e+03 2.207e+03 5.093e+03, threshold=3.136e+03, percent-clipped=13.0
+2023-03-07 03:49:11,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=598587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:49:22,350 INFO [train.py:968] (0/2) Epoch 14, batch 5300, giga_loss[loss=0.2658, simple_loss=0.3631, pruned_loss=0.08425, over 28830.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3468, pruned_loss=0.09722, over 5708738.43 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3516, pruned_loss=0.09573, over 5321852.04 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3455, pruned_loss=0.09712, over 5706002.06 frames. ], batch size: 174, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:50:04,982 INFO [train.py:968] (0/2) Epoch 14, batch 5350, giga_loss[loss=0.2549, simple_loss=0.3419, pruned_loss=0.08396, over 28728.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3488, pruned_loss=0.09752, over 5714013.82 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3517, pruned_loss=0.09578, over 5327883.87 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3477, pruned_loss=0.09743, over 5710002.79 frames. ], batch size: 284, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:50:15,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.798e+02 1.156e+03 1.494e+03 1.925e+03 5.729e+03, threshold=2.987e+03, percent-clipped=5.0
+2023-03-07 03:50:19,409 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4293, 3.2830, 1.5129, 1.5645], device='cuda:0'), covar=tensor([0.0868, 0.0299, 0.0910, 0.1197], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0518, 0.0351, 0.0397], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 03:50:42,077 INFO [train.py:968] (0/2) Epoch 14, batch 5400, giga_loss[loss=0.2729, simple_loss=0.3501, pruned_loss=0.09787, over 28550.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3482, pruned_loss=0.09837, over 5717205.98 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.352, pruned_loss=0.096, over 5349564.49 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3469, pruned_loss=0.09819, over 5709602.93 frames. ], batch size: 336, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:51:08,210 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:51:11,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:51:15,983 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=598739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:51:16,912 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-07 03:51:25,961 INFO [train.py:968] (0/2) Epoch 14, batch 5450, giga_loss[loss=0.2389, simple_loss=0.3163, pruned_loss=0.08075, over 28811.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3462, pruned_loss=0.09871, over 5721649.81 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.352, pruned_loss=0.09594, over 5352388.67 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3453, pruned_loss=0.09864, over 5714784.49 frames. ], batch size: 186, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:51:37,654 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=598762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:51:41,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.440e+02 1.192e+03 1.508e+03 2.029e+03 4.282e+03, threshold=3.016e+03, percent-clipped=5.0
+2023-03-07 03:52:00,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1655, 1.3441, 0.9423, 0.9744], device='cuda:0'), covar=tensor([0.0929, 0.0600, 0.1375, 0.1312], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0438, 0.0500, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 03:52:08,405 INFO [train.py:968] (0/2) Epoch 14, batch 5500, giga_loss[loss=0.2463, simple_loss=0.3125, pruned_loss=0.09006, over 28837.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3433, pruned_loss=0.09848, over 5727093.83 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3518, pruned_loss=0.09583, over 5357403.68 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3426, pruned_loss=0.09855, over 5720495.58 frames. ], batch size: 99, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:52:27,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=598825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:52:49,040 INFO [train.py:968] (0/2) Epoch 14, batch 5550, giga_loss[loss=0.2414, simple_loss=0.3204, pruned_loss=0.08116, over 29033.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3413, pruned_loss=0.09814, over 5730013.05 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.352, pruned_loss=0.09598, over 5368924.14 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3404, pruned_loss=0.09814, over 5722727.66 frames. ], batch size: 155, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:52:57,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5954, 1.6461, 1.2039, 1.2796], device='cuda:0'), covar=tensor([0.0821, 0.0607, 0.1027, 0.1103], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0439, 0.0502, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 03:53:03,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.275e+02 1.086e+03 1.308e+03 1.643e+03 4.652e+03, threshold=2.615e+03, percent-clipped=6.0
+2023-03-07 03:53:33,183 INFO [train.py:968] (0/2) Epoch 14, batch 5600, giga_loss[loss=0.2945, simple_loss=0.3587, pruned_loss=0.1152, over 28365.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3413, pruned_loss=0.09837, over 5720227.33 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3517, pruned_loss=0.0958, over 5384000.10 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3406, pruned_loss=0.0986, over 5710120.86 frames. ], batch size: 368, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:53:38,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5521, 1.7732, 1.4526, 1.4546], device='cuda:0'), covar=tensor([0.2348, 0.2344, 0.2663, 0.2200], device='cuda:0'), in_proj_covar=tensor([0.1354, 0.0989, 0.1194, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 03:53:52,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9531, 1.0681, 3.7781, 3.0186], device='cuda:0'), covar=tensor([0.1746, 0.2685, 0.0467, 0.0974], device='cuda:0'), in_proj_covar=tensor([0.0684, 0.0603, 0.0877, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 03:53:54,873 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 03:54:15,657 INFO [train.py:968] (0/2) Epoch 14, batch 5650, giga_loss[loss=0.2302, simple_loss=0.3052, pruned_loss=0.07759, over 28910.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3373, pruned_loss=0.09632, over 5722472.79 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3512, pruned_loss=0.09565, over 5398907.44 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3368, pruned_loss=0.09669, over 5710089.33 frames. ], batch size: 186, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:54:19,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3751, 1.3626, 1.2854, 1.5624], device='cuda:0'), covar=tensor([0.0727, 0.0335, 0.0327, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-07 03:54:28,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.314e+02 1.234e+03 1.541e+03 2.019e+03 5.608e+03, threshold=3.083e+03, percent-clipped=13.0
+2023-03-07 03:54:30,106 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:54:32,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:54:54,932 INFO [train.py:968] (0/2) Epoch 14, batch 5700, giga_loss[loss=0.2571, simple_loss=0.3343, pruned_loss=0.08995, over 28337.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3339, pruned_loss=0.0947, over 5719019.10 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3513, pruned_loss=0.09575, over 5407086.96 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.333, pruned_loss=0.09488, over 5707422.28 frames. ], batch size: 368, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:54:55,131 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=599000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:55:00,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2780, 1.5642, 1.3774, 1.5383], device='cuda:0'), covar=tensor([0.0736, 0.0314, 0.0324, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-07 03:55:35,199 INFO [train.py:968] (0/2) Epoch 14, batch 5750, giga_loss[loss=0.2356, simple_loss=0.3121, pruned_loss=0.07953, over 28803.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3325, pruned_loss=0.09397, over 5724826.15 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3514, pruned_loss=0.09588, over 5422008.05 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3312, pruned_loss=0.09396, over 5710343.96 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:55:48,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.891e+02 1.249e+03 1.691e+03 2.937e+03 8.064e+03, threshold=3.381e+03, percent-clipped=21.0
+2023-03-07 03:56:12,191 INFO [train.py:968] (0/2) Epoch 14, batch 5800, giga_loss[loss=0.2908, simple_loss=0.36, pruned_loss=0.1108, over 28851.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3346, pruned_loss=0.09521, over 5712188.78 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3515, pruned_loss=0.09592, over 5424845.81 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.333, pruned_loss=0.09511, over 5706655.40 frames. ], batch size: 112, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:56:23,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=599114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:56:28,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1419, 1.8102, 1.4233, 0.2977], device='cuda:0'), covar=tensor([0.3771, 0.2382, 0.3581, 0.4930], device='cuda:0'), in_proj_covar=tensor([0.1614, 0.1520, 0.1512, 0.1319], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 03:56:33,570 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-07 03:56:49,164 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-07 03:56:52,155 INFO [train.py:968] (0/2) Epoch 14, batch 5850, giga_loss[loss=0.2879, simple_loss=0.3499, pruned_loss=0.1129, over 23931.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3382, pruned_loss=0.09677, over 5715023.21 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.352, pruned_loss=0.09635, over 5436751.71 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3362, pruned_loss=0.09634, over 5706651.71 frames. ], batch size: 705, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:56:58,349 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 03:57:06,078 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.710e+02 1.183e+03 1.463e+03 1.957e+03 6.714e+03, threshold=2.926e+03, percent-clipped=3.0
+2023-03-07 03:57:32,842 INFO [train.py:968] (0/2) Epoch 14, batch 5900, giga_loss[loss=0.2757, simple_loss=0.3543, pruned_loss=0.09851, over 28960.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3432, pruned_loss=0.09909, over 5715441.03 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3525, pruned_loss=0.0966, over 5443370.72 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.341, pruned_loss=0.09857, over 5708940.05 frames. ], batch size: 213, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:57:42,039 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-07 03:57:46,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8175, 3.6233, 3.4199, 1.9456], device='cuda:0'), covar=tensor([0.0602, 0.0728, 0.0666, 0.2139], device='cuda:0'), in_proj_covar=tensor([0.1087, 0.1005, 0.0879, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 03:57:48,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5516, 2.0159, 1.2927, 0.8957], device='cuda:0'), covar=tensor([0.5270, 0.2818, 0.2720, 0.5041], device='cuda:0'), in_proj_covar=tensor([0.1607, 0.1513, 0.1512, 0.1314], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 03:57:56,929 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2294, 1.4170, 1.3911, 1.2955], device='cuda:0'), covar=tensor([0.1406, 0.1475, 0.2002, 0.1551], device='cuda:0'), in_proj_covar=tensor([0.0450, 0.0741, 0.0691, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 03:58:09,517 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7195, 3.3431, 1.8203, 1.9684], device='cuda:0'), covar=tensor([0.0866, 0.0452, 0.0815, 0.1199], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0521, 0.0351, 0.0398], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:0')
+2023-03-07 03:58:17,975 INFO [train.py:968] (0/2) Epoch 14, batch 5950, giga_loss[loss=0.3252, simple_loss=0.3841, pruned_loss=0.1332, over 26766.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3466, pruned_loss=0.1005, over 5707042.34 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3525, pruned_loss=0.09667, over 5443561.85 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3446, pruned_loss=0.1001, over 5708429.04 frames. ], batch size: 555, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:58:24,760 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=599257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:58:26,745 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=599260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:58:33,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.530e+02 1.169e+03 1.452e+03 2.080e+03 1.923e+04, threshold=2.904e+03, percent-clipped=15.0
+2023-03-07 03:58:52,669 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 03:58:53,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=599289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:59:01,859 INFO [train.py:968] (0/2) Epoch 14, batch 6000, giga_loss[loss=0.2932, simple_loss=0.3696, pruned_loss=0.1084, over 28670.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3488, pruned_loss=0.1013, over 5704736.89 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.353, pruned_loss=0.09701, over 5450048.24 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3467, pruned_loss=0.1008, over 5703637.34 frames. ], batch size: 262, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:59:01,864 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 03:59:09,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2912, 1.2911, 1.0918, 1.5049], device='cuda:0'), covar=tensor([0.0799, 0.0340, 0.0348, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-07 03:59:10,162 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2184, simple_loss=0.3243, pruned_loss=0.05621, over 944034.00 frames. 
+2023-03-07 03:59:10,163 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 03:59:56,783 INFO [train.py:968] (0/2) Epoch 14, batch 6050, giga_loss[loss=0.4168, simple_loss=0.4567, pruned_loss=0.1885, over 27651.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3537, pruned_loss=0.1057, over 5697779.02 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3532, pruned_loss=0.09711, over 5452441.00 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.352, pruned_loss=0.1052, over 5695953.82 frames. ], batch size: 472, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:00:11,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.876e+02 1.224e+03 1.423e+03 1.838e+03 6.291e+03, threshold=2.846e+03, percent-clipped=6.0
+2023-03-07 04:00:42,745 INFO [train.py:968] (0/2) Epoch 14, batch 6100, giga_loss[loss=0.3392, simple_loss=0.3984, pruned_loss=0.14, over 29031.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.36, pruned_loss=0.1109, over 5688264.25 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3532, pruned_loss=0.09721, over 5459764.82 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3587, pruned_loss=0.1108, over 5688500.38 frames. ], batch size: 155, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:01:31,535 INFO [train.py:968] (0/2) Epoch 14, batch 6150, giga_loss[loss=0.3662, simple_loss=0.4186, pruned_loss=0.1569, over 28817.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3663, pruned_loss=0.1149, over 5693493.79 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3531, pruned_loss=0.09707, over 5463908.32 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3655, pruned_loss=0.1151, over 5692097.57 frames. ], batch size: 284, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:01:47,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.883e+02 1.470e+03 1.962e+03 3.026e+03 8.772e+03, threshold=3.923e+03, percent-clipped=29.0
+2023-03-07 04:02:20,051 INFO [train.py:968] (0/2) Epoch 14, batch 6200, giga_loss[loss=0.2954, simple_loss=0.3657, pruned_loss=0.1125, over 29041.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.372, pruned_loss=0.1195, over 5694207.88 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3536, pruned_loss=0.09745, over 5471918.31 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3714, pruned_loss=0.1198, over 5692012.01 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:03:08,111 INFO [train.py:968] (0/2) Epoch 14, batch 6250, giga_loss[loss=0.3361, simple_loss=0.4037, pruned_loss=0.1342, over 28849.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.381, pruned_loss=0.1272, over 5692143.98 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3539, pruned_loss=0.09765, over 5477055.16 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3804, pruned_loss=0.1276, over 5688784.37 frames. ], batch size: 186, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:03:22,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.852e+02 1.732e+03 2.136e+03 2.946e+03 8.511e+03, threshold=4.273e+03, percent-clipped=11.0
+2023-03-07 04:03:28,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5060, 1.7003, 1.5552, 1.3098], device='cuda:0'), covar=tensor([0.2243, 0.2045, 0.1684, 0.2098], device='cuda:0'), in_proj_covar=tensor([0.1790, 0.1709, 0.1677, 0.1777], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 04:03:32,901 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-07 04:03:49,615 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1686, 1.1813, 3.8465, 3.1812], device='cuda:0'), covar=tensor([0.1636, 0.2663, 0.0412, 0.0986], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0601, 0.0875, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 04:03:52,648 INFO [train.py:968] (0/2) Epoch 14, batch 6300, giga_loss[loss=0.399, simple_loss=0.4363, pruned_loss=0.1808, over 27950.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3838, pruned_loss=0.1294, over 5688044.86 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3536, pruned_loss=0.09747, over 5493261.15 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3849, pruned_loss=0.1313, over 5679581.19 frames. ], batch size: 412, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:04:44,598 INFO [train.py:968] (0/2) Epoch 14, batch 6350, giga_loss[loss=0.4245, simple_loss=0.4455, pruned_loss=0.2017, over 27520.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3865, pruned_loss=0.1325, over 5671841.28 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.354, pruned_loss=0.09765, over 5499837.30 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3878, pruned_loss=0.1346, over 5663315.71 frames. ], batch size: 472, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:05:01,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.144e+03 1.670e+03 2.041e+03 2.978e+03 8.046e+03, threshold=4.082e+03, percent-clipped=10.0
+2023-03-07 04:05:35,556 INFO [train.py:968] (0/2) Epoch 14, batch 6400, libri_loss[loss=0.2833, simple_loss=0.3722, pruned_loss=0.09721, over 29510.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3885, pruned_loss=0.135, over 5674860.04 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3535, pruned_loss=0.09732, over 5506903.58 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3905, pruned_loss=0.1377, over 5664756.96 frames. ], batch size: 84, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:05:39,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7284, 3.5594, 3.4273, 1.8129], device='cuda:0'), covar=tensor([0.0646, 0.0749, 0.0699, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.1100, 0.1017, 0.0893, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 04:06:30,442 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-07 04:06:30,646 INFO [train.py:968] (0/2) Epoch 14, batch 6450, giga_loss[loss=0.3374, simple_loss=0.3914, pruned_loss=0.1417, over 29117.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3912, pruned_loss=0.1388, over 5662162.56 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3533, pruned_loss=0.09724, over 5511218.34 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3935, pruned_loss=0.1415, over 5652373.79 frames. ], batch size: 155, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:06:51,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.652e+03 2.220e+03 2.827e+03 5.507e+03, threshold=4.440e+03, percent-clipped=4.0
+2023-03-07 04:06:51,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4874, 2.3652, 1.7456, 2.2337], device='cuda:0'), covar=tensor([0.0755, 0.0630, 0.0934, 0.1002], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0442, 0.0505, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 04:07:22,426 INFO [train.py:968] (0/2) Epoch 14, batch 6500, giga_loss[loss=0.3604, simple_loss=0.4062, pruned_loss=0.1573, over 27546.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3956, pruned_loss=0.1429, over 5653946.31 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3534, pruned_loss=0.09745, over 5517202.69 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.398, pruned_loss=0.1457, over 5643487.91 frames. ], batch size: 472, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:08:12,798 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-07 04:08:13,055 INFO [train.py:968] (0/2) Epoch 14, batch 6550, giga_loss[loss=0.3352, simple_loss=0.3905, pruned_loss=0.14, over 28590.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.395, pruned_loss=0.1434, over 5655328.47 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3533, pruned_loss=0.09733, over 5523281.18 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3977, pruned_loss=0.1464, over 5643187.15 frames. ], batch size: 307, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:08:31,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.847e+03 2.656e+03 3.620e+03 1.484e+04, threshold=5.311e+03, percent-clipped=17.0
+2023-03-07 04:09:01,778 INFO [train.py:968] (0/2) Epoch 14, batch 6600, giga_loss[loss=0.3149, simple_loss=0.379, pruned_loss=0.1255, over 29008.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3939, pruned_loss=0.1431, over 5644622.12 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3536, pruned_loss=0.09752, over 5522920.07 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.3968, pruned_loss=0.1466, over 5638674.55 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:09:06,701 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 04:09:51,893 INFO [train.py:968] (0/2) Epoch 14, batch 6650, giga_loss[loss=0.2858, simple_loss=0.365, pruned_loss=0.1033, over 28832.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3921, pruned_loss=0.1412, over 5636729.98 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3541, pruned_loss=0.09786, over 5527074.95 frames. ], giga_tot_loss[loss=0.3429, simple_loss=0.3954, pruned_loss=0.1452, over 5631268.88 frames. ], batch size: 174, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:09:58,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2488, 4.0804, 3.8508, 2.1503], device='cuda:0'), covar=tensor([0.0614, 0.0729, 0.0749, 0.1940], device='cuda:0'), in_proj_covar=tensor([0.1107, 0.1026, 0.0899, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 04:10:03,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3913, 2.0720, 1.5937, 0.5157], device='cuda:0'), covar=tensor([0.4328, 0.2752, 0.3536, 0.5055], device='cuda:0'), in_proj_covar=tensor([0.1614, 0.1530, 0.1522, 0.1325], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 04:10:08,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.612e+02 1.846e+03 2.325e+03 2.922e+03 8.021e+03, threshold=4.651e+03, percent-clipped=3.0
+2023-03-07 04:10:38,718 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-600000.pt
+2023-03-07 04:10:39,010 INFO [train.py:968] (0/2) Epoch 14, batch 6700, giga_loss[loss=0.3704, simple_loss=0.4222, pruned_loss=0.1593, over 28934.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3913, pruned_loss=0.139, over 5649903.29 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3543, pruned_loss=0.09798, over 5536747.29 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3945, pruned_loss=0.143, over 5639372.86 frames. ], batch size: 213, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:11:08,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4036, 1.7190, 1.3290, 1.5845], device='cuda:0'), covar=tensor([0.3006, 0.2885, 0.3240, 0.2503], device='cuda:0'), in_proj_covar=tensor([0.1351, 0.0992, 0.1194, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 04:11:17,099 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4132, 3.1442, 1.4975, 1.6309], device='cuda:0'), covar=tensor([0.0879, 0.0243, 0.0880, 0.1231], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0521, 0.0352, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 04:11:22,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6471, 1.8942, 1.7163, 1.4037], device='cuda:0'), covar=tensor([0.2022, 0.1829, 0.1688, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.1785, 0.1703, 0.1661, 0.1771], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 04:11:27,725 INFO [train.py:968] (0/2) Epoch 14, batch 6750, giga_loss[loss=0.3879, simple_loss=0.4171, pruned_loss=0.1794, over 26578.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3927, pruned_loss=0.1399, over 5635377.43 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3541, pruned_loss=0.09784, over 5537037.99 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3967, pruned_loss=0.1444, over 5629373.73 frames. ], batch size: 555, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:11:45,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.832e+03 2.357e+03 3.398e+03 6.696e+03, threshold=4.714e+03, percent-clipped=10.0
+2023-03-07 04:12:06,462 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:12:15,553 INFO [train.py:968] (0/2) Epoch 14, batch 6800, giga_loss[loss=0.4449, simple_loss=0.4644, pruned_loss=0.2128, over 26562.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3897, pruned_loss=0.1374, over 5621578.73 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3537, pruned_loss=0.09767, over 5534181.53 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3938, pruned_loss=0.1419, over 5620940.98 frames. ], batch size: 555, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:13:08,547 INFO [train.py:968] (0/2) Epoch 14, batch 6850, giga_loss[loss=0.2961, simple_loss=0.3683, pruned_loss=0.1119, over 29050.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3875, pruned_loss=0.1343, over 5635320.91 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3537, pruned_loss=0.09775, over 5541891.86 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3914, pruned_loss=0.1385, over 5629534.57 frames. ], batch size: 128, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:13:13,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:13:26,460 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.368e+02 1.470e+03 1.786e+03 2.338e+03 5.064e+03, threshold=3.573e+03, percent-clipped=1.0
+2023-03-07 04:13:29,096 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:13:55,282 INFO [train.py:968] (0/2) Epoch 14, batch 6900, giga_loss[loss=0.2848, simple_loss=0.3596, pruned_loss=0.105, over 28775.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3844, pruned_loss=0.1307, over 5646522.09 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.354, pruned_loss=0.09789, over 5550185.60 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3881, pruned_loss=0.1349, over 5637309.07 frames. ], batch size: 243, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:14:15,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8000, 1.7826, 1.3696, 1.3511], device='cuda:0'), covar=tensor([0.0826, 0.0647, 0.1020, 0.1099], device='cuda:0'), in_proj_covar=tensor([0.0363, 0.0438, 0.0499, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 04:14:43,928 INFO [train.py:968] (0/2) Epoch 14, batch 6950, giga_loss[loss=0.2426, simple_loss=0.3311, pruned_loss=0.07705, over 28942.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.38, pruned_loss=0.1271, over 5653505.75 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3541, pruned_loss=0.09796, over 5557135.83 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3835, pruned_loss=0.1311, over 5642194.56 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:15:01,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.719e+03 2.190e+03 2.981e+03 9.672e+03, threshold=4.379e+03, percent-clipped=16.0
+2023-03-07 04:15:30,586 INFO [train.py:968] (0/2) Epoch 14, batch 7000, giga_loss[loss=0.2946, simple_loss=0.3711, pruned_loss=0.109, over 28682.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3769, pruned_loss=0.1251, over 5654540.50 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3533, pruned_loss=0.09751, over 5567278.01 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.381, pruned_loss=0.1295, over 5638917.22 frames. ], batch size: 242, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:15:58,661 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600327.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 04:15:58,705 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4659, 1.6385, 1.3929, 1.4212], device='cuda:0'), covar=tensor([0.1227, 0.1476, 0.1547, 0.1517], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0745, 0.0690, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 04:16:17,745 INFO [train.py:968] (0/2) Epoch 14, batch 7050, giga_loss[loss=0.316, simple_loss=0.3867, pruned_loss=0.1227, over 28966.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3764, pruned_loss=0.1247, over 5653767.23 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3534, pruned_loss=0.09756, over 5570488.95 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3805, pruned_loss=0.1291, over 5640667.47 frames. ], batch size: 136, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:16:20,173 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:16:38,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.250e+02 1.445e+03 2.068e+03 2.639e+03 6.445e+03, threshold=4.137e+03, percent-clipped=3.0
+2023-03-07 04:17:04,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2093, 1.0992, 3.6089, 3.1229], device='cuda:0'), covar=tensor([0.1694, 0.2813, 0.0484, 0.0901], device='cuda:0'), in_proj_covar=tensor([0.0693, 0.0609, 0.0887, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 04:17:10,845 INFO [train.py:968] (0/2) Epoch 14, batch 7100, giga_loss[loss=0.289, simple_loss=0.3566, pruned_loss=0.1107, over 28640.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3765, pruned_loss=0.1248, over 5653908.69 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3534, pruned_loss=0.09749, over 5576993.74 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3802, pruned_loss=0.1288, over 5639245.21 frames. ], batch size: 242, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:17:34,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1856, 0.7870, 0.8203, 1.3502], device='cuda:0'), covar=tensor([0.0787, 0.0382, 0.0365, 0.0874], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 04:17:53,286 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:17:53,854 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:18:00,738 INFO [train.py:968] (0/2) Epoch 14, batch 7150, giga_loss[loss=0.3364, simple_loss=0.3958, pruned_loss=0.1384, over 28889.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3738, pruned_loss=0.1223, over 5663545.49 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3529, pruned_loss=0.09711, over 5590280.80 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3783, pruned_loss=0.1272, over 5642806.45 frames. ], batch size: 186, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:18:14,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:18:19,482 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.458e+02 1.384e+03 1.762e+03 2.650e+03 5.588e+03, threshold=3.524e+03, percent-clipped=5.0
+2023-03-07 04:18:36,341 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:18:53,729 INFO [train.py:968] (0/2) Epoch 14, batch 7200, giga_loss[loss=0.3116, simple_loss=0.3927, pruned_loss=0.1152, over 28939.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3745, pruned_loss=0.1208, over 5672189.34 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3526, pruned_loss=0.09696, over 5595607.64 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3786, pruned_loss=0.1252, over 5652156.52 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:19:18,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0675, 1.2644, 1.0401, 0.8616], device='cuda:0'), covar=tensor([0.0936, 0.0502, 0.1095, 0.1011], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0437, 0.0497, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:0')
+2023-03-07 04:19:28,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:19:34,752 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.32 vs. limit=5.0
+2023-03-07 04:19:35,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3285, 1.5849, 1.3136, 1.4649], device='cuda:0'), covar=tensor([0.0755, 0.0315, 0.0312, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0062, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 04:19:43,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:19:44,187 INFO [train.py:968] (0/2) Epoch 14, batch 7250, giga_loss[loss=0.3229, simple_loss=0.3894, pruned_loss=0.1282, over 28875.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3769, pruned_loss=0.121, over 5678864.87 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3523, pruned_loss=0.09688, over 5600205.73 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3809, pruned_loss=0.125, over 5659709.71 frames. ], batch size: 186, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:20:00,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:20:08,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.171e+03 1.807e+03 2.510e+03 3.554e+03 8.777e+03, threshold=5.021e+03, percent-clipped=25.0
+2023-03-07 04:20:10,055 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-07 04:20:40,713 INFO [train.py:968] (0/2) Epoch 14, batch 7300, giga_loss[loss=0.3923, simple_loss=0.4325, pruned_loss=0.176, over 28265.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3781, pruned_loss=0.1227, over 5668567.91 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3523, pruned_loss=0.09691, over 5603228.00 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3814, pruned_loss=0.126, over 5651479.88 frames. ], batch size: 368, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:20:46,691 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:20:48,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:21:15,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:21:25,521 INFO [train.py:968] (0/2) Epoch 14, batch 7350, giga_loss[loss=0.28, simple_loss=0.354, pruned_loss=0.103, over 28951.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3775, pruned_loss=0.1227, over 5672065.15 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3522, pruned_loss=0.09679, over 5608159.32 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3811, pruned_loss=0.1264, over 5655695.29 frames. ], batch size: 106, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:21:42,835 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 04:21:45,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.749e+02 1.651e+03 2.099e+03 3.024e+03 8.836e+03, threshold=4.198e+03, percent-clipped=5.0
+2023-03-07 04:21:49,509 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:21:51,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:00,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3179, 3.4035, 1.5989, 1.4681], device='cuda:0'), covar=tensor([0.0972, 0.0332, 0.0824, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0525, 0.0352, 0.0400], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 04:22:04,348 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-07 04:22:05,783 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:07,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:12,764 INFO [train.py:968] (0/2) Epoch 14, batch 7400, giga_loss[loss=0.247, simple_loss=0.3322, pruned_loss=0.0809, over 29034.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3752, pruned_loss=0.1223, over 5680606.35 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3516, pruned_loss=0.0965, over 5615430.34 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3792, pruned_loss=0.1261, over 5662367.29 frames. ], batch size: 155, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:22:15,579 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600702.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 04:22:18,362 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:34,512 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:39,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:59,642 INFO [train.py:968] (0/2) Epoch 14, batch 7450, giga_loss[loss=0.3522, simple_loss=0.4082, pruned_loss=0.1481, over 28987.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3749, pruned_loss=0.1236, over 5677073.06 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3515, pruned_loss=0.09634, over 5620859.83 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3786, pruned_loss=0.1272, over 5658963.70 frames. ], batch size: 128, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:23:18,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.401e+02 1.598e+03 2.210e+03 3.176e+03 9.005e+03, threshold=4.421e+03, percent-clipped=15.0
+2023-03-07 04:23:48,516 INFO [train.py:968] (0/2) Epoch 14, batch 7500, giga_loss[loss=0.2748, simple_loss=0.3516, pruned_loss=0.09905, over 28783.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3735, pruned_loss=0.1218, over 5676931.14 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3513, pruned_loss=0.09631, over 5626185.65 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3773, pruned_loss=0.1255, over 5658843.56 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:24:03,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:24:04,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:24:29,633 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600845.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 04:24:34,002 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600848.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 04:24:36,326 INFO [train.py:968] (0/2) Epoch 14, batch 7550, giga_loss[loss=0.2758, simple_loss=0.3465, pruned_loss=0.1025, over 28642.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3731, pruned_loss=0.1207, over 5675714.34 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3513, pruned_loss=0.09641, over 5633019.97 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3767, pruned_loss=0.1241, over 5656110.51 frames. ], batch size: 262, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:24:44,061 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-07 04:24:45,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:24:52,851 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.975e+02 1.502e+03 1.946e+03 2.659e+03 6.070e+03, threshold=3.891e+03, percent-clipped=3.0
+2023-03-07 04:24:53,122 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:24:55,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:24:58,778 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600877.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 04:25:19,433 INFO [train.py:968] (0/2) Epoch 14, batch 7600, giga_loss[loss=0.2984, simple_loss=0.3711, pruned_loss=0.1129, over 29075.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.372, pruned_loss=0.1194, over 5684831.15 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3503, pruned_loss=0.09595, over 5643916.37 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3768, pruned_loss=0.1236, over 5660866.31 frames. ], batch size: 155, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:25:21,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:25:52,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:02,817 INFO [train.py:968] (0/2) Epoch 14, batch 7650, giga_loss[loss=0.3119, simple_loss=0.3751, pruned_loss=0.1243, over 28975.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3703, pruned_loss=0.118, over 5699090.10 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3501, pruned_loss=0.09588, over 5650240.54 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3747, pruned_loss=0.122, over 5675343.44 frames. ], batch size: 213, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:26:13,051 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:14,815 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:16,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:17,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:22,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4298, 1.7065, 1.7658, 1.6210], device='cuda:0'), covar=tensor([0.1563, 0.1563, 0.1742, 0.1528], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0739, 0.0688, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 04:26:24,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.524e+03 1.954e+03 2.806e+03 7.389e+03, threshold=3.907e+03, percent-clipped=11.0
+2023-03-07 04:26:43,267 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:43,974 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:51,268 INFO [train.py:968] (0/2) Epoch 14, batch 7700, giga_loss[loss=0.3062, simple_loss=0.363, pruned_loss=0.1247, over 28353.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3697, pruned_loss=0.1182, over 5688202.61 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3502, pruned_loss=0.09574, over 5654585.32 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3738, pruned_loss=0.1223, over 5666304.12 frames. ], batch size: 78, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:26:52,782 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=601002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:55,673 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=601005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:57,843 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3823, 3.0772, 1.6176, 1.4949], device='cuda:0'), covar=tensor([0.0908, 0.0315, 0.0780, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0526, 0.0354, 0.0402], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 04:27:26,065 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=601034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:27:39,195 INFO [train.py:968] (0/2) Epoch 14, batch 7750, giga_loss[loss=0.3081, simple_loss=0.3688, pruned_loss=0.1238, over 28808.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3697, pruned_loss=0.1192, over 5679249.78 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3502, pruned_loss=0.0958, over 5656836.07 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3735, pruned_loss=0.1231, over 5660407.30 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:27:43,753 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=601055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:27:58,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.676e+03 1.986e+03 2.720e+03 9.893e+03, threshold=3.972e+03, percent-clipped=8.0
+2023-03-07 04:28:12,661 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=601083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:28:14,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=601086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:28:27,019 INFO [train.py:968] (0/2) Epoch 14, batch 7800, giga_loss[loss=0.2786, simple_loss=0.3535, pruned_loss=0.1019, over 28890.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.37, pruned_loss=0.1207, over 5662199.89 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3508, pruned_loss=0.09637, over 5652019.52 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3729, pruned_loss=0.1236, over 5652660.59 frames. ], batch size: 174, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:28:33,081 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7912, 1.8441, 1.7576, 1.5910], device='cuda:0'), covar=tensor([0.1521, 0.2293, 0.1925, 0.2185], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0737, 0.0686, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 04:28:43,731 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=601115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:29:16,489 INFO [train.py:968] (0/2) Epoch 14, batch 7850, libri_loss[loss=0.3331, simple_loss=0.3973, pruned_loss=0.1345, over 29775.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3699, pruned_loss=0.1214, over 5664756.29 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3512, pruned_loss=0.09654, over 5653568.14 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3722, pruned_loss=0.124, over 5655949.42 frames. ], batch size: 87, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:29:38,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.759e+02 1.741e+03 2.094e+03 2.872e+03 7.481e+03, threshold=4.189e+03, percent-clipped=10.0
+2023-03-07 04:29:55,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6798, 1.5006, 1.8328, 1.3819], device='cuda:0'), covar=tensor([0.1452, 0.2314, 0.1220, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.0844, 0.0693, 0.0887, 0.0791], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-07 04:30:04,061 INFO [train.py:968] (0/2) Epoch 14, batch 7900, giga_loss[loss=0.2958, simple_loss=0.3612, pruned_loss=0.1152, over 28751.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3699, pruned_loss=0.1217, over 5658480.27 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3511, pruned_loss=0.09653, over 5657239.99 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3719, pruned_loss=0.1241, over 5648535.96 frames. ], batch size: 99, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:30:08,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6051, 1.8334, 1.5274, 1.7256], device='cuda:0'), covar=tensor([0.2301, 0.2363, 0.2627, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.1351, 0.0993, 0.1191, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 04:30:50,734 INFO [train.py:968] (0/2) Epoch 14, batch 7950, giga_loss[loss=0.3867, simple_loss=0.4167, pruned_loss=0.1784, over 26750.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3711, pruned_loss=0.1222, over 5668154.23 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3514, pruned_loss=0.09677, over 5661514.01 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.373, pruned_loss=0.1245, over 5656487.11 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:31:13,687 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.682e+03 2.043e+03 2.871e+03 4.742e+03, threshold=4.085e+03, percent-clipped=1.0
+2023-03-07 04:31:35,124 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 04:31:37,974 INFO [train.py:968] (0/2) Epoch 14, batch 8000, libri_loss[loss=0.2888, simple_loss=0.3663, pruned_loss=0.1056, over 29652.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3708, pruned_loss=0.1208, over 5670049.86 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3515, pruned_loss=0.09673, over 5668078.29 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3727, pruned_loss=0.1233, over 5654621.72 frames. ], batch size: 88, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 04:31:49,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6853, 2.0707, 1.8161, 2.0310], device='cuda:0'), covar=tensor([0.0713, 0.0258, 0.0280, 0.0738], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0113, 0.0210], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:0')
+2023-03-07 04:32:23,619 INFO [train.py:968] (0/2) Epoch 14, batch 8050, giga_loss[loss=0.2771, simple_loss=0.3552, pruned_loss=0.09948, over 28774.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3712, pruned_loss=0.1201, over 5683051.50 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3515, pruned_loss=0.09676, over 5673094.54 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.373, pruned_loss=0.1224, over 5666520.66 frames. ], batch size: 242, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:32:46,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.481e+02 1.526e+03 2.003e+03 2.910e+03 7.059e+03, threshold=4.006e+03, percent-clipped=8.0
+2023-03-07 04:33:12,502 INFO [train.py:968] (0/2) Epoch 14, batch 8100, libri_loss[loss=0.2372, simple_loss=0.3192, pruned_loss=0.07763, over 29535.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3718, pruned_loss=0.1201, over 5688833.25 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3514, pruned_loss=0.09666, over 5678201.48 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3738, pruned_loss=0.1226, over 5671240.55 frames. ], batch size: 77, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:33:40,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3764, 1.4085, 4.1374, 3.2395], device='cuda:0'), covar=tensor([0.1611, 0.2537, 0.0416, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0691, 0.0610, 0.0888, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 04:33:40,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=601430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:33:49,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6774, 1.5469, 1.7716, 1.2876], device='cuda:0'), covar=tensor([0.1970, 0.3040, 0.1517, 0.1679], device='cuda:0'), in_proj_covar=tensor([0.0846, 0.0695, 0.0888, 0.0792], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:0')
+2023-03-07 04:33:59,168 INFO [train.py:968] (0/2) Epoch 14, batch 8150, giga_loss[loss=0.3613, simple_loss=0.3903, pruned_loss=0.1661, over 23447.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3737, pruned_loss=0.1223, over 5688767.57 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3517, pruned_loss=0.09681, over 5683256.43 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3755, pruned_loss=0.1247, over 5670660.88 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:34:25,918 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.082e+02 1.611e+03 2.150e+03 3.485e+03 9.951e+03, threshold=4.301e+03, percent-clipped=16.0
+2023-03-07 04:34:36,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7151, 2.1294, 1.8923, 1.6205], device='cuda:0'), covar=tensor([0.2714, 0.1834, 0.2019, 0.2150], device='cuda:0'), in_proj_covar=tensor([0.1805, 0.1723, 0.1684, 0.1792], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 04:34:55,522 INFO [train.py:968] (0/2) Epoch 14, batch 8200, giga_loss[loss=0.3368, simple_loss=0.3873, pruned_loss=0.1431, over 28877.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3764, pruned_loss=0.1259, over 5672613.45 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3516, pruned_loss=0.09673, over 5684373.15 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3781, pruned_loss=0.128, over 5657349.40 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:35:48,327 INFO [train.py:968] (0/2) Epoch 14, batch 8250, giga_loss[loss=0.3001, simple_loss=0.3622, pruned_loss=0.119, over 28969.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3777, pruned_loss=0.1285, over 5664443.78 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3515, pruned_loss=0.09683, over 5685712.61 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3794, pruned_loss=0.1305, over 5650959.64 frames. ], batch size: 164, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:35:55,789 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9758, 3.7806, 3.6054, 1.9396], device='cuda:0'), covar=tensor([0.0763, 0.0930, 0.0936, 0.1858], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.1033, 0.0907, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 04:36:12,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.150e+03 1.805e+03 2.424e+03 3.253e+03 1.027e+04, threshold=4.849e+03, percent-clipped=13.0
+2023-03-07 04:36:12,856 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=601573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:36:15,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=601576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:36:19,638 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2940, 1.0485, 4.1273, 3.3382], device='cuda:0'), covar=tensor([0.1637, 0.2875, 0.0408, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0688, 0.0607, 0.0884, 0.0807], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 04:36:23,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1314, 3.9535, 3.7309, 1.6257], device='cuda:0'), covar=tensor([0.0678, 0.0804, 0.0824, 0.2321], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.1033, 0.0908, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 04:36:36,915 INFO [train.py:968] (0/2) Epoch 14, batch 8300, giga_loss[loss=0.3333, simple_loss=0.3799, pruned_loss=0.1434, over 28703.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3786, pruned_loss=0.1301, over 5669834.12 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3513, pruned_loss=0.09672, over 5691337.65 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3808, pruned_loss=0.1326, over 5653724.18 frames. ], batch size: 242, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:36:41,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=601605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:36:53,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3710, 1.6894, 1.3408, 1.5521], device='cuda:0'), covar=tensor([0.0735, 0.0313, 0.0317, 0.0773], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 04:37:27,776 INFO [train.py:968] (0/2) Epoch 14, batch 8350, giga_loss[loss=0.2945, simple_loss=0.3518, pruned_loss=0.1186, over 28623.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.378, pruned_loss=0.1298, over 5668189.56 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3512, pruned_loss=0.09666, over 5693261.03 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3802, pruned_loss=0.1322, over 5653599.59 frames. ], batch size: 71, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:37:47,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.839e+03 2.474e+03 3.745e+03 9.382e+03, threshold=4.949e+03, percent-clipped=12.0
+2023-03-07 04:38:08,633 INFO [train.py:968] (0/2) Epoch 14, batch 8400, giga_loss[loss=0.277, simple_loss=0.3495, pruned_loss=0.1023, over 28843.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3774, pruned_loss=0.1281, over 5684227.44 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3518, pruned_loss=0.09685, over 5701150.35 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3797, pruned_loss=0.1312, over 5664167.07 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 04:38:20,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5186, 1.6898, 1.3949, 1.3960], device='cuda:0'), covar=tensor([0.2819, 0.2710, 0.3149, 0.2369], device='cuda:0'), in_proj_covar=tensor([0.1353, 0.0993, 0.1191, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 04:38:45,341 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=601739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:38:54,925 INFO [train.py:968] (0/2) Epoch 14, batch 8450, giga_loss[loss=0.2948, simple_loss=0.3646, pruned_loss=0.1125, over 28729.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3762, pruned_loss=0.1253, over 5689816.21 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3514, pruned_loss=0.09669, over 5703091.46 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3785, pruned_loss=0.1282, over 5672164.13 frames. ], batch size: 99, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:39:17,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.730e+02 1.496e+03 1.964e+03 2.621e+03 6.028e+03, threshold=3.927e+03, percent-clipped=4.0
+2023-03-07 04:39:38,590 INFO [train.py:968] (0/2) Epoch 14, batch 8500, libri_loss[loss=0.2431, simple_loss=0.3222, pruned_loss=0.08197, over 29568.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3735, pruned_loss=0.1232, over 5685469.42 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3518, pruned_loss=0.09717, over 5706903.92 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3758, pruned_loss=0.126, over 5667060.55 frames. ], batch size: 77, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:40:11,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3369, 3.1410, 1.4763, 1.4390], device='cuda:0'), covar=tensor([0.0970, 0.0289, 0.0840, 0.1344], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0526, 0.0354, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 04:40:23,972 INFO [train.py:968] (0/2) Epoch 14, batch 8550, giga_loss[loss=0.3675, simple_loss=0.4013, pruned_loss=0.1669, over 26612.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3715, pruned_loss=0.1226, over 5682316.42 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3517, pruned_loss=0.09703, over 5709856.25 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3737, pruned_loss=0.1252, over 5665028.86 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:40:44,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4494, 1.6296, 1.5495, 1.4583], device='cuda:0'), covar=tensor([0.1593, 0.1884, 0.2267, 0.1880], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0738, 0.0687, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 04:40:44,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.893e+02 1.588e+03 2.165e+03 3.026e+03 8.186e+03, threshold=4.331e+03, percent-clipped=7.0
+2023-03-07 04:41:09,986 INFO [train.py:968] (0/2) Epoch 14, batch 8600, libri_loss[loss=0.2549, simple_loss=0.3415, pruned_loss=0.08409, over 29528.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3699, pruned_loss=0.1216, over 5684397.73 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3522, pruned_loss=0.0975, over 5712504.50 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.372, pruned_loss=0.1243, over 5666840.82 frames. ], batch size: 81, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:41:57,921 INFO [train.py:968] (0/2) Epoch 14, batch 8650, giga_loss[loss=0.2918, simple_loss=0.3621, pruned_loss=0.1107, over 28830.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3718, pruned_loss=0.1227, over 5685683.50 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3524, pruned_loss=0.09749, over 5715959.54 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3736, pruned_loss=0.1252, over 5668411.94 frames. ], batch size: 227, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:42:03,699 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=601953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:42:23,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.498e+02 1.704e+03 2.301e+03 3.336e+03 1.054e+04, threshold=4.601e+03, percent-clipped=13.0
+2023-03-07 04:42:47,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4856, 1.8756, 1.6599, 1.5754], device='cuda:0'), covar=tensor([0.1614, 0.1842, 0.2043, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0738, 0.0686, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 04:42:49,850 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-602000.pt
+2023-03-07 04:42:50,142 INFO [train.py:968] (0/2) Epoch 14, batch 8700, giga_loss[loss=0.3384, simple_loss=0.4038, pruned_loss=0.1365, over 28886.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3759, pruned_loss=0.1237, over 5682077.88 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3524, pruned_loss=0.09749, over 5715959.54 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3773, pruned_loss=0.1257, over 5668635.20 frames. ], batch size: 227, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:43:06,020 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-07 04:43:19,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-07 04:43:41,377 INFO [train.py:968] (0/2) Epoch 14, batch 8750, giga_loss[loss=0.3217, simple_loss=0.3916, pruned_loss=0.1258, over 29039.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3782, pruned_loss=0.1231, over 5678152.17 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3524, pruned_loss=0.09746, over 5717757.61 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3797, pruned_loss=0.1251, over 5665451.75 frames. ], batch size: 155, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:44:04,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.217e+02 1.399e+03 1.739e+03 2.497e+03 6.623e+03, threshold=3.479e+03, percent-clipped=2.0
+2023-03-07 04:44:26,275 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 04:44:28,436 INFO [train.py:968] (0/2) Epoch 14, batch 8800, giga_loss[loss=0.3339, simple_loss=0.3938, pruned_loss=0.137, over 28913.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3802, pruned_loss=0.1249, over 5681294.23 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3521, pruned_loss=0.0973, over 5720225.20 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3824, pruned_loss=0.1273, over 5667637.63 frames. ], batch size: 174, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:44:31,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4785, 1.6865, 1.5269, 1.3545], device='cuda:0'), covar=tensor([0.2076, 0.1675, 0.1529, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.1792, 0.1707, 0.1675, 0.1778], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 04:44:40,091 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=602114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:45:11,922 INFO [train.py:968] (0/2) Epoch 14, batch 8850, giga_loss[loss=0.3473, simple_loss=0.3955, pruned_loss=0.1496, over 28279.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3808, pruned_loss=0.1257, over 5692144.18 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3521, pruned_loss=0.09727, over 5725972.72 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3834, pruned_loss=0.1285, over 5675088.42 frames. ], batch size: 368, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:45:35,241 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.705e+03 2.359e+03 3.230e+03 2.776e+04, threshold=4.719e+03, percent-clipped=22.0
+2023-03-07 04:45:55,850 INFO [train.py:968] (0/2) Epoch 14, batch 8900, libri_loss[loss=0.227, simple_loss=0.3063, pruned_loss=0.07387, over 29652.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3783, pruned_loss=0.1243, over 5699138.72 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.351, pruned_loss=0.0965, over 5734491.23 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3827, pruned_loss=0.1285, over 5675757.00 frames. ], batch size: 73, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:45:59,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2398, 1.3850, 1.2770, 1.4834], device='cuda:0'), covar=tensor([0.0727, 0.0354, 0.0318, 0.0780], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 04:46:45,816 INFO [train.py:968] (0/2) Epoch 14, batch 8950, giga_loss[loss=0.3297, simple_loss=0.3886, pruned_loss=0.1354, over 28673.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3788, pruned_loss=0.1257, over 5699971.95 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.351, pruned_loss=0.09655, over 5736369.29 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3824, pruned_loss=0.1293, over 5679531.74 frames. ], batch size: 262, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:46:52,252 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=602257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:46:55,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=602260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:46:55,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=602260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:47:12,271 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.516e+03 1.953e+03 2.712e+03 6.842e+03, threshold=3.907e+03, percent-clipped=5.0
+2023-03-07 04:47:24,144 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=602289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:47:34,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5344, 1.6100, 1.8153, 1.3833], device='cuda:0'), covar=tensor([0.1322, 0.1871, 0.1116, 0.1430], device='cuda:0'), in_proj_covar=tensor([0.0848, 0.0695, 0.0890, 0.0794], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 04:47:34,891 INFO [train.py:968] (0/2) Epoch 14, batch 9000, giga_loss[loss=0.4372, simple_loss=0.4542, pruned_loss=0.2101, over 27904.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3763, pruned_loss=0.1246, over 5693789.80 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.351, pruned_loss=0.09654, over 5736864.15 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3795, pruned_loss=0.1278, over 5676756.67 frames. ], batch size: 412, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:47:34,896 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 04:47:43,432 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2149, simple_loss=0.3222, pruned_loss=0.05378, over 944034.00 frames. 
+2023-03-07 04:47:43,432 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 04:48:09,115 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=602328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:48:16,990 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4088, 1.4039, 4.0506, 3.3146], device='cuda:0'), covar=tensor([0.1566, 0.2446, 0.0415, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0691, 0.0610, 0.0886, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 04:48:27,456 INFO [train.py:968] (0/2) Epoch 14, batch 9050, libri_loss[loss=0.2679, simple_loss=0.3549, pruned_loss=0.0905, over 29746.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3743, pruned_loss=0.1237, over 5692230.20 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3511, pruned_loss=0.09667, over 5743975.03 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3779, pruned_loss=0.1273, over 5669971.78 frames. ], batch size: 87, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:48:40,113 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 04:48:49,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0012, 2.3115, 2.0760, 1.9264], device='cuda:0'), covar=tensor([0.1943, 0.2259, 0.2004, 0.2103], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0739, 0.0686, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 04:48:52,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.270e+02 1.491e+03 1.897e+03 2.497e+03 5.812e+03, threshold=3.794e+03, percent-clipped=6.0
+2023-03-07 04:49:16,669 INFO [train.py:968] (0/2) Epoch 14, batch 9100, giga_loss[loss=0.3216, simple_loss=0.3806, pruned_loss=0.1313, over 28276.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3731, pruned_loss=0.1234, over 5693060.03 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.351, pruned_loss=0.09667, over 5745840.83 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3765, pruned_loss=0.1269, over 5672438.98 frames. ], batch size: 368, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:50:07,161 INFO [train.py:968] (0/2) Epoch 14, batch 9150, giga_loss[loss=0.2831, simple_loss=0.3531, pruned_loss=0.1065, over 28831.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3743, pruned_loss=0.125, over 5690940.21 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3509, pruned_loss=0.09668, over 5748592.48 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3777, pruned_loss=0.1285, over 5670994.44 frames. ], batch size: 227, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:50:12,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3862, 1.4891, 1.5038, 1.3951], device='cuda:0'), covar=tensor([0.1290, 0.1551, 0.1759, 0.1469], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0738, 0.0685, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 04:50:26,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2052, 4.0242, 3.8057, 1.8046], device='cuda:0'), covar=tensor([0.0611, 0.0758, 0.0736, 0.2183], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.1035, 0.0907, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 04:50:29,923 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=602471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:50:32,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=602474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:50:33,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.565e+02 1.827e+03 2.231e+03 3.132e+03 1.741e+04, threshold=4.462e+03, percent-clipped=17.0
+2023-03-07 04:50:54,415 INFO [train.py:968] (0/2) Epoch 14, batch 9200, giga_loss[loss=0.3005, simple_loss=0.3674, pruned_loss=0.1168, over 28826.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3722, pruned_loss=0.1239, over 5686731.31 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3509, pruned_loss=0.09659, over 5748981.87 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3752, pruned_loss=0.1271, over 5669579.04 frames. ], batch size: 119, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:50:58,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=602503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:51:21,616 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1530, 1.2659, 3.2790, 2.9398], device='cuda:0'), covar=tensor([0.1504, 0.2462, 0.0483, 0.1204], device='cuda:0'), in_proj_covar=tensor([0.0690, 0.0608, 0.0883, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 04:51:44,574 INFO [train.py:968] (0/2) Epoch 14, batch 9250, giga_loss[loss=0.3806, simple_loss=0.4243, pruned_loss=0.1684, over 27993.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.371, pruned_loss=0.123, over 5691854.09 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.351, pruned_loss=0.09651, over 5751949.79 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3737, pruned_loss=0.1261, over 5674309.27 frames. ], batch size: 412, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:52:07,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.195e+02 1.403e+03 1.810e+03 2.313e+03 6.510e+03, threshold=3.620e+03, percent-clipped=3.0
+2023-03-07 04:52:20,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=602591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:52:32,387 INFO [train.py:968] (0/2) Epoch 14, batch 9300, giga_loss[loss=0.3901, simple_loss=0.4201, pruned_loss=0.1801, over 26455.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3723, pruned_loss=0.1233, over 5685253.83 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3511, pruned_loss=0.09664, over 5754306.69 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3748, pruned_loss=0.1262, over 5667947.85 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:53:02,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=602635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:53:17,443 INFO [train.py:968] (0/2) Epoch 14, batch 9350, giga_loss[loss=0.2881, simple_loss=0.3572, pruned_loss=0.1095, over 28779.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3741, pruned_loss=0.124, over 5677937.01 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3508, pruned_loss=0.09627, over 5746023.67 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3769, pruned_loss=0.1273, over 5669099.43 frames. ], batch size: 243, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:53:32,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9478, 1.1380, 1.0999, 0.8372], device='cuda:0'), covar=tensor([0.1839, 0.2087, 0.1260, 0.1736], device='cuda:0'), in_proj_covar=tensor([0.1801, 0.1702, 0.1674, 0.1777], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 04:53:43,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.905e+02 1.516e+03 1.917e+03 2.827e+03 6.983e+03, threshold=3.834e+03, percent-clipped=10.0
+2023-03-07 04:54:07,315 INFO [train.py:968] (0/2) Epoch 14, batch 9400, giga_loss[loss=0.4191, simple_loss=0.4444, pruned_loss=0.1969, over 26623.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3746, pruned_loss=0.1252, over 5675433.06 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3508, pruned_loss=0.09627, over 5746023.67 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3768, pruned_loss=0.1278, over 5668554.66 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:54:28,917 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0153, 1.3160, 1.0883, 0.2614], device='cuda:0'), covar=tensor([0.2984, 0.2667, 0.3910, 0.5322], device='cuda:0'), in_proj_covar=tensor([0.1616, 0.1544, 0.1518, 0.1331], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 04:54:49,280 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-07 04:54:54,849 INFO [train.py:968] (0/2) Epoch 14, batch 9450, giga_loss[loss=0.2497, simple_loss=0.3517, pruned_loss=0.07389, over 28714.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3753, pruned_loss=0.1233, over 5681151.85 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3514, pruned_loss=0.09668, over 5749978.90 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3772, pruned_loss=0.1259, over 5669765.26 frames. ], batch size: 60, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:55:14,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.561e+02 1.493e+03 1.948e+03 2.778e+03 6.963e+03, threshold=3.897e+03, percent-clipped=7.0
+2023-03-07 04:55:16,522 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=602778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:55:18,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=602781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:55:25,489 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=602788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:55:35,146 INFO [train.py:968] (0/2) Epoch 14, batch 9500, giga_loss[loss=0.2646, simple_loss=0.3525, pruned_loss=0.08832, over 28807.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3756, pruned_loss=0.1215, over 5694072.55 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3513, pruned_loss=0.09667, over 5758014.46 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3786, pruned_loss=0.125, over 5674325.20 frames. ], batch size: 119, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:55:40,167 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4766, 1.7769, 1.3955, 1.7711], device='cuda:0'), covar=tensor([0.2612, 0.2572, 0.2872, 0.2482], device='cuda:0'), in_proj_covar=tensor([0.1357, 0.0997, 0.1197, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 04:55:44,345 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=602810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:56:23,462 INFO [train.py:968] (0/2) Epoch 14, batch 9550, giga_loss[loss=0.3975, simple_loss=0.4401, pruned_loss=0.1775, over 28743.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.379, pruned_loss=0.123, over 5689302.67 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3515, pruned_loss=0.09665, over 5759295.43 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3815, pruned_loss=0.126, over 5672014.43 frames. ], batch size: 262, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:56:24,911 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-07 04:56:46,180 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=602873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:56:47,867 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.662e+02 1.475e+03 1.932e+03 2.708e+03 1.082e+04, threshold=3.863e+03, percent-clipped=8.0
+2023-03-07 04:57:05,722 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 04:57:13,273 INFO [train.py:968] (0/2) Epoch 14, batch 9600, giga_loss[loss=0.2867, simple_loss=0.3574, pruned_loss=0.108, over 28324.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3809, pruned_loss=0.1249, over 5684477.35 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3511, pruned_loss=0.09643, over 5758344.97 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3839, pruned_loss=0.1281, over 5669856.44 frames. ], batch size: 65, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 04:57:57,266 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-07 04:57:58,770 INFO [train.py:968] (0/2) Epoch 14, batch 9650, giga_loss[loss=0.4273, simple_loss=0.4553, pruned_loss=0.1996, over 28624.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3834, pruned_loss=0.1281, over 5673759.33 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3511, pruned_loss=0.09648, over 5748886.26 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3863, pruned_loss=0.1312, over 5669750.50 frames. ], batch size: 336, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:58:17,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=602966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:58:26,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.704e+03 2.429e+03 3.035e+03 1.288e+04, threshold=4.859e+03, percent-clipped=18.0
+2023-03-07 04:58:32,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7389, 1.8828, 1.6113, 1.9954], device='cuda:0'), covar=tensor([0.2368, 0.2527, 0.2654, 0.2322], device='cuda:0'), in_proj_covar=tensor([0.1356, 0.0998, 0.1197, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 04:58:47,738 INFO [train.py:968] (0/2) Epoch 14, batch 9700, giga_loss[loss=0.2624, simple_loss=0.346, pruned_loss=0.08943, over 28821.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3831, pruned_loss=0.1291, over 5669068.54 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3512, pruned_loss=0.09647, over 5752102.31 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.386, pruned_loss=0.1321, over 5661737.32 frames. ], batch size: 119, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:59:33,968 INFO [train.py:968] (0/2) Epoch 14, batch 9750, giga_loss[loss=0.3391, simple_loss=0.3856, pruned_loss=0.1463, over 27944.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3813, pruned_loss=0.128, over 5658957.32 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3509, pruned_loss=0.09629, over 5755909.53 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3846, pruned_loss=0.1314, over 5647753.95 frames. ], batch size: 412, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:59:57,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.942e+02 1.451e+03 1.863e+03 2.859e+03 9.454e+03, threshold=3.726e+03, percent-clipped=8.0
+2023-03-07 05:00:19,159 INFO [train.py:968] (0/2) Epoch 14, batch 9800, giga_loss[loss=0.2781, simple_loss=0.3639, pruned_loss=0.09619, over 28675.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3798, pruned_loss=0.1249, over 5666993.14 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3507, pruned_loss=0.09622, over 5756409.38 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3828, pruned_loss=0.1279, over 5656954.03 frames. ], batch size: 92, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:00:27,731 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=603109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:00:30,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=603112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:00:58,383 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=603141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:01:05,982 INFO [train.py:968] (0/2) Epoch 14, batch 9850, giga_loss[loss=0.3474, simple_loss=0.3874, pruned_loss=0.1537, over 23463.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3802, pruned_loss=0.1241, over 5669840.91 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3508, pruned_loss=0.09634, over 5758182.42 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3827, pruned_loss=0.1265, over 5659483.55 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:01:15,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=603163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:01:29,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.574e+02 1.555e+03 1.943e+03 2.890e+03 5.389e+03, threshold=3.886e+03, percent-clipped=9.0
+2023-03-07 05:01:32,771 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:01:50,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2234, 1.4603, 1.1993, 1.1676], device='cuda:0'), covar=tensor([0.1785, 0.1781, 0.1838, 0.1512], device='cuda:0'), in_proj_covar=tensor([0.1355, 0.0996, 0.1195, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 05:01:53,105 INFO [train.py:968] (0/2) Epoch 14, batch 9900, giga_loss[loss=0.3347, simple_loss=0.3992, pruned_loss=0.1351, over 28838.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3818, pruned_loss=0.1253, over 5671828.93 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3507, pruned_loss=0.09624, over 5761490.81 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3848, pruned_loss=0.1282, over 5658402.17 frames. ], batch size: 284, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:02:38,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2773, 1.4750, 1.5117, 1.3331], device='cuda:0'), covar=tensor([0.1682, 0.1767, 0.2275, 0.1817], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0736, 0.0685, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 05:02:40,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=603248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:02:42,905 INFO [train.py:968] (0/2) Epoch 14, batch 9950, giga_loss[loss=0.3239, simple_loss=0.3884, pruned_loss=0.1298, over 29063.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3819, pruned_loss=0.1261, over 5670826.01 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3507, pruned_loss=0.09624, over 5764923.07 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3853, pruned_loss=0.1293, over 5654083.56 frames. ], batch size: 155, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:03:04,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.661e+03 2.167e+03 3.572e+03 1.400e+04, threshold=4.333e+03, percent-clipped=20.0
+2023-03-07 05:03:23,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2105, 1.2752, 3.5993, 3.2744], device='cuda:0'), covar=tensor([0.1577, 0.2592, 0.0470, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.0695, 0.0610, 0.0889, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 05:03:26,494 INFO [train.py:968] (0/2) Epoch 14, batch 10000, giga_loss[loss=0.3018, simple_loss=0.368, pruned_loss=0.1178, over 28882.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3808, pruned_loss=0.1257, over 5678648.37 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3514, pruned_loss=0.09669, over 5759243.85 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3838, pruned_loss=0.1288, over 5667223.05 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:03:34,525 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=603306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:03:37,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=603309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:04:08,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=603338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:04:17,763 INFO [train.py:968] (0/2) Epoch 14, batch 10050, giga_loss[loss=0.3067, simple_loss=0.3708, pruned_loss=0.1213, over 28738.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3809, pruned_loss=0.1271, over 5667871.74 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3517, pruned_loss=0.09675, over 5760952.19 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3834, pruned_loss=0.1298, over 5656294.87 frames. ], batch size: 284, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:04:44,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.762e+03 2.383e+03 3.311e+03 1.048e+04, threshold=4.765e+03, percent-clipped=12.0
+2023-03-07 05:05:00,813 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=603391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:05:03,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=603394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:05:08,484 INFO [train.py:968] (0/2) Epoch 14, batch 10100, giga_loss[loss=0.2794, simple_loss=0.3496, pruned_loss=0.1046, over 28756.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3775, pruned_loss=0.1251, over 5674889.05 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3515, pruned_loss=0.09669, over 5762699.82 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.38, pruned_loss=0.1278, over 5662525.13 frames. ], batch size: 119, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:05:33,261 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=603423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:05:56,589 INFO [train.py:968] (0/2) Epoch 14, batch 10150, giga_loss[loss=0.2778, simple_loss=0.3457, pruned_loss=0.1049, over 28986.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3754, pruned_loss=0.1245, over 5678910.85 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3512, pruned_loss=0.09656, over 5765350.53 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3784, pruned_loss=0.1276, over 5664225.93 frames. ], batch size: 136, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:06:09,161 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:06:24,427 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.332e+02 1.719e+03 2.237e+03 2.968e+03 7.758e+03, threshold=4.474e+03, percent-clipped=5.0
+2023-03-07 05:06:43,945 INFO [train.py:968] (0/2) Epoch 14, batch 10200, giga_loss[loss=0.2907, simple_loss=0.3591, pruned_loss=0.1111, over 28717.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3759, pruned_loss=0.1254, over 5679996.39 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.351, pruned_loss=0.09641, over 5768425.30 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3789, pruned_loss=0.1286, over 5663846.59 frames. ], batch size: 242, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:07:30,494 INFO [train.py:968] (0/2) Epoch 14, batch 10250, giga_loss[loss=0.2645, simple_loss=0.344, pruned_loss=0.09254, over 28673.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3731, pruned_loss=0.1224, over 5678875.07 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3514, pruned_loss=0.09668, over 5772377.37 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3759, pruned_loss=0.1256, over 5659206.33 frames. ], batch size: 262, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:07:36,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=603557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:07:40,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6939, 2.5551, 1.6707, 0.9914], device='cuda:0'), covar=tensor([0.6635, 0.3212, 0.3266, 0.5428], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1534, 0.1510, 0.1327], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 05:07:54,616 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:07:55,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.339e+02 1.430e+03 1.827e+03 2.402e+03 4.925e+03, threshold=3.654e+03, percent-clipped=4.0
+2023-03-07 05:08:13,167 INFO [train.py:968] (0/2) Epoch 14, batch 10300, giga_loss[loss=0.3348, simple_loss=0.3893, pruned_loss=0.1401, over 28896.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.37, pruned_loss=0.1191, over 5684178.10 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3517, pruned_loss=0.09676, over 5776648.66 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3729, pruned_loss=0.1225, over 5660670.38 frames. ], batch size: 213, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:08:36,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3119, 1.6037, 1.3692, 1.5252], device='cuda:0'), covar=tensor([0.0751, 0.0333, 0.0321, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0113, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 05:08:45,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9828, 2.1722, 1.7878, 2.1823], device='cuda:0'), covar=tensor([0.2247, 0.2341, 0.2541, 0.2236], device='cuda:0'), in_proj_covar=tensor([0.1365, 0.1005, 0.1204, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 05:09:02,212 INFO [train.py:968] (0/2) Epoch 14, batch 10350, giga_loss[loss=0.2697, simple_loss=0.3417, pruned_loss=0.09888, over 28883.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3687, pruned_loss=0.1181, over 5677197.95 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.351, pruned_loss=0.09634, over 5779436.63 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3721, pruned_loss=0.1219, over 5653210.17 frames. ], batch size: 145, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:09:13,342 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:09:22,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3846, 1.7668, 1.4630, 1.5616], device='cuda:0'), covar=tensor([0.0728, 0.0314, 0.0301, 0.0777], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0113, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:0')
+2023-03-07 05:09:28,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.563e+02 1.318e+03 1.719e+03 2.249e+03 4.737e+03, threshold=3.438e+03, percent-clipped=3.0
+2023-03-07 05:09:49,819 INFO [train.py:968] (0/2) Epoch 14, batch 10400, giga_loss[loss=0.3314, simple_loss=0.3767, pruned_loss=0.1431, over 28225.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3673, pruned_loss=0.1177, over 5682692.45 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3519, pruned_loss=0.09683, over 5780251.12 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5660711.95 frames. ], batch size: 368, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:09:50,146 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=603700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:09:53,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=603703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:10:19,603 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=603732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:10:27,054 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.12 vs. limit=5.0
+2023-03-07 05:10:34,289 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:10:38,377 INFO [train.py:968] (0/2) Epoch 14, batch 10450, giga_loss[loss=0.3567, simple_loss=0.4079, pruned_loss=0.1527, over 29051.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3652, pruned_loss=0.1171, over 5682020.49 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.352, pruned_loss=0.09683, over 5783767.63 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3674, pruned_loss=0.1202, over 5658459.81 frames. ], batch size: 128, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:11:06,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.315e+02 1.739e+03 2.357e+03 3.650e+03 1.136e+04, threshold=4.713e+03, percent-clipped=29.0
+2023-03-07 05:11:22,607 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-07 05:11:24,310 INFO [train.py:968] (0/2) Epoch 14, batch 10500, giga_loss[loss=0.3647, simple_loss=0.3957, pruned_loss=0.1669, over 23575.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3686, pruned_loss=0.1198, over 5682093.64 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.352, pruned_loss=0.09679, over 5786713.58 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3707, pruned_loss=0.1228, over 5658371.32 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:11:56,568 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=603839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:12:06,329 INFO [train.py:968] (0/2) Epoch 14, batch 10550, giga_loss[loss=0.2894, simple_loss=0.3583, pruned_loss=0.1102, over 28978.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3706, pruned_loss=0.1199, over 5676918.92 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3532, pruned_loss=0.09752, over 5772046.31 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.372, pruned_loss=0.1227, over 5665889.41 frames. ], batch size: 227, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:12:27,028 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 05:12:32,396 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 05:12:37,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.275e+02 1.400e+03 2.030e+03 2.705e+03 8.999e+03, threshold=4.060e+03, percent-clipped=6.0
+2023-03-07 05:12:54,350 INFO [train.py:968] (0/2) Epoch 14, batch 10600, giga_loss[loss=0.2913, simple_loss=0.3501, pruned_loss=0.1162, over 28921.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3713, pruned_loss=0.1206, over 5654063.94 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3533, pruned_loss=0.09748, over 5775101.33 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3726, pruned_loss=0.1234, over 5639943.06 frames. ], batch size: 106, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:12:59,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:13:26,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:13:40,824 INFO [train.py:968] (0/2) Epoch 14, batch 10650, giga_loss[loss=0.2917, simple_loss=0.368, pruned_loss=0.1077, over 28391.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3707, pruned_loss=0.1205, over 5646564.14 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3534, pruned_loss=0.09758, over 5777276.11 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3724, pruned_loss=0.1235, over 5628726.99 frames. ], batch size: 60, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:13:44,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=603952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:14:10,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.675e+02 1.392e+03 1.918e+03 2.575e+03 1.419e+04, threshold=3.836e+03, percent-clipped=7.0
+2023-03-07 05:14:13,434 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=603982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:14:15,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=603985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:14:27,892 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-604000.pt
+2023-03-07 05:14:28,179 INFO [train.py:968] (0/2) Epoch 14, batch 10700, libri_loss[loss=0.2999, simple_loss=0.3779, pruned_loss=0.111, over 29232.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3718, pruned_loss=0.1216, over 5658091.76 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3536, pruned_loss=0.09752, over 5778966.34 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3734, pruned_loss=0.1245, over 5640024.93 frames. ], batch size: 94, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:14:43,384 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:15:05,634 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:15:19,976 INFO [train.py:968] (0/2) Epoch 14, batch 10750, giga_loss[loss=0.3079, simple_loss=0.3801, pruned_loss=0.1178, over 28984.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3731, pruned_loss=0.1222, over 5650027.55 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.354, pruned_loss=0.09771, over 5773605.49 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3743, pruned_loss=0.1249, over 5636841.58 frames. ], batch size: 106, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:15:28,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2851, 1.1824, 3.9227, 3.2467], device='cuda:0'), covar=tensor([0.1628, 0.2731, 0.0438, 0.0912], device='cuda:0'), in_proj_covar=tensor([0.0693, 0.0611, 0.0891, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 05:15:49,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.751e+02 1.788e+03 2.268e+03 3.404e+03 8.809e+03, threshold=4.535e+03, percent-clipped=19.0
+2023-03-07 05:16:02,465 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:16:04,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:16:05,589 INFO [train.py:968] (0/2) Epoch 14, batch 10800, libri_loss[loss=0.2757, simple_loss=0.3493, pruned_loss=0.1011, over 29646.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3729, pruned_loss=0.122, over 5650093.10 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3538, pruned_loss=0.09781, over 5769103.57 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3747, pruned_loss=0.1249, over 5639474.02 frames. ], batch size: 73, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:16:21,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:16:28,700 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:16:50,562 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-07 05:16:50,743 INFO [train.py:968] (0/2) Epoch 14, batch 10850, giga_loss[loss=0.2747, simple_loss=0.3505, pruned_loss=0.09941, over 28955.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3749, pruned_loss=0.1236, over 5652191.52 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3542, pruned_loss=0.09786, over 5768698.50 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3765, pruned_loss=0.1264, over 5641626.30 frames. ], batch size: 213, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:17:19,489 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:17:20,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.219e+02 1.524e+03 2.015e+03 2.621e+03 5.877e+03, threshold=4.030e+03, percent-clipped=1.0
+2023-03-07 05:17:21,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:17:40,853 INFO [train.py:968] (0/2) Epoch 14, batch 10900, giga_loss[loss=0.2725, simple_loss=0.3403, pruned_loss=0.1024, over 28543.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3755, pruned_loss=0.1245, over 5648364.78 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3541, pruned_loss=0.09777, over 5762157.88 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3772, pruned_loss=0.1273, over 5644149.67 frames. ], batch size: 85, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:17:52,157 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=604211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:17:52,173 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:18:31,181 INFO [train.py:968] (0/2) Epoch 14, batch 10950, giga_loss[loss=0.3547, simple_loss=0.4071, pruned_loss=0.1511, over 27497.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3763, pruned_loss=0.1233, over 5654044.55 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3539, pruned_loss=0.09762, over 5761353.74 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3782, pruned_loss=0.1261, over 5649263.72 frames. ], batch size: 472, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:18:33,438 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-07 05:18:47,714 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:18:51,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:19:02,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:19:03,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.367e+02 1.583e+03 1.971e+03 2.846e+03 7.471e+03, threshold=3.942e+03, percent-clipped=9.0
+2023-03-07 05:19:18,194 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:19:24,659 INFO [train.py:968] (0/2) Epoch 14, batch 11000, giga_loss[loss=0.4851, simple_loss=0.4739, pruned_loss=0.2482, over 23654.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3755, pruned_loss=0.1235, over 5643710.38 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3537, pruned_loss=0.09752, over 5764304.66 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3777, pruned_loss=0.1263, over 5635162.24 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:19:32,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6898, 2.4225, 1.8772, 2.1438], device='cuda:0'), covar=tensor([0.0687, 0.0680, 0.0921, 0.1021], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0443, 0.0502, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 05:19:33,415 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:20:12,725 INFO [train.py:968] (0/2) Epoch 14, batch 11050, giga_loss[loss=0.2621, simple_loss=0.3355, pruned_loss=0.09437, over 28890.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3741, pruned_loss=0.1232, over 5647049.38 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3538, pruned_loss=0.0976, over 5754420.43 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3761, pruned_loss=0.1259, over 5647236.68 frames. ], batch size: 112, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:20:51,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.653e+02 1.672e+03 2.102e+03 2.713e+03 7.387e+03, threshold=4.204e+03, percent-clipped=10.0
+2023-03-07 05:21:15,395 INFO [train.py:968] (0/2) Epoch 14, batch 11100, libri_loss[loss=0.2303, simple_loss=0.3177, pruned_loss=0.07148, over 29530.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3744, pruned_loss=0.1246, over 5643989.71 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3536, pruned_loss=0.09751, over 5755870.74 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3764, pruned_loss=0.1271, over 5641647.34 frames. ], batch size: 80, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:21:29,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0452, 2.2780, 2.2911, 1.8686], device='cuda:0'), covar=tensor([0.1637, 0.2047, 0.1270, 0.1454], device='cuda:0'), in_proj_covar=tensor([0.0846, 0.0695, 0.0891, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 05:21:30,771 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-07 05:21:32,817 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=604421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:21:32,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:21:36,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:21:59,413 INFO [train.py:968] (0/2) Epoch 14, batch 11150, giga_loss[loss=0.3391, simple_loss=0.3916, pruned_loss=0.1433, over 28288.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.373, pruned_loss=0.1231, over 5665411.00 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3534, pruned_loss=0.09727, over 5759272.84 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3754, pruned_loss=0.1262, over 5657199.31 frames. ], batch size: 368, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:22:02,528 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:22:02,563 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:22:04,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:22:29,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.635e+03 2.302e+03 3.269e+03 8.346e+03, threshold=4.604e+03, percent-clipped=12.0
+2023-03-07 05:22:33,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:22:47,734 INFO [train.py:968] (0/2) Epoch 14, batch 11200, giga_loss[loss=0.3121, simple_loss=0.3766, pruned_loss=0.1238, over 28700.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3721, pruned_loss=0.1236, over 5665982.80 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3534, pruned_loss=0.09727, over 5759272.84 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3741, pruned_loss=0.126, over 5659591.55 frames. ], batch size: 242, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:23:35,653 INFO [train.py:968] (0/2) Epoch 14, batch 11250, giga_loss[loss=0.3148, simple_loss=0.3772, pruned_loss=0.1262, over 28892.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3728, pruned_loss=0.1243, over 5667509.03 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3536, pruned_loss=0.09741, over 5762286.73 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3745, pruned_loss=0.1266, over 5658244.74 frames. ], batch size: 285, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:24:05,574 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=604578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:24:07,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.559e+02 1.540e+03 2.008e+03 2.554e+03 1.171e+04, threshold=4.015e+03, percent-clipped=7.0
+2023-03-07 05:24:12,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:24:24,302 INFO [train.py:968] (0/2) Epoch 14, batch 11300, giga_loss[loss=0.2741, simple_loss=0.3431, pruned_loss=0.1025, over 28937.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3727, pruned_loss=0.1243, over 5669572.95 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3532, pruned_loss=0.09707, over 5764890.93 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.375, pruned_loss=0.1272, over 5657293.45 frames. ], batch size: 106, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:25:09,268 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 05:25:12,561 INFO [train.py:968] (0/2) Epoch 14, batch 11350, giga_loss[loss=0.31, simple_loss=0.3738, pruned_loss=0.1231, over 28614.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3735, pruned_loss=0.125, over 5673278.11 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3527, pruned_loss=0.09684, over 5767568.96 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3761, pruned_loss=0.128, over 5659416.07 frames. ], batch size: 307, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:25:38,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.592e+02 1.764e+03 2.398e+03 3.252e+03 8.329e+03, threshold=4.796e+03, percent-clipped=12.0
+2023-03-07 05:25:58,752 INFO [train.py:968] (0/2) Epoch 14, batch 11400, libri_loss[loss=0.256, simple_loss=0.3305, pruned_loss=0.09072, over 29572.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3738, pruned_loss=0.1245, over 5676781.31 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3526, pruned_loss=0.09672, over 5769592.56 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3766, pruned_loss=0.1279, over 5660806.65 frames. ], batch size: 74, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:26:00,634 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-07 05:26:04,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1975, 1.2109, 3.4365, 3.0106], device='cuda:0'), covar=tensor([0.1543, 0.2656, 0.0469, 0.1495], device='cuda:0'), in_proj_covar=tensor([0.0696, 0.0613, 0.0894, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 05:26:23,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3659, 2.9399, 2.1230, 1.8863], device='cuda:0'), covar=tensor([0.2309, 0.1326, 0.1970, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.1805, 0.1719, 0.1690, 0.1789], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 05:26:27,391 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:26:30,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:26:49,607 INFO [train.py:968] (0/2) Epoch 14, batch 11450, giga_loss[loss=0.3234, simple_loss=0.3763, pruned_loss=0.1353, over 28908.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3751, pruned_loss=0.1267, over 5671416.29 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3527, pruned_loss=0.09674, over 5770859.06 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3775, pruned_loss=0.1296, over 5656899.67 frames. ], batch size: 136, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:27:00,752 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:27:09,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3005, 1.7554, 1.6131, 1.1620], device='cuda:0'), covar=tensor([0.1795, 0.2608, 0.1548, 0.1824], device='cuda:0'), in_proj_covar=tensor([0.0846, 0.0696, 0.0890, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 05:27:16,522 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-07 05:27:17,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.689e+03 2.288e+03 3.443e+03 9.136e+03, threshold=4.575e+03, percent-clipped=9.0
+2023-03-07 05:27:29,935 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:27:32,909 INFO [train.py:968] (0/2) Epoch 14, batch 11500, giga_loss[loss=0.2732, simple_loss=0.3449, pruned_loss=0.1007, over 28888.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3742, pruned_loss=0.1259, over 5669825.88 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3523, pruned_loss=0.09636, over 5773604.93 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3773, pruned_loss=0.1298, over 5651791.05 frames. ], batch size: 186, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:28:20,606 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9257, 3.7530, 3.5650, 1.7404], device='cuda:0'), covar=tensor([0.0676, 0.0795, 0.0788, 0.1997], device='cuda:0'), in_proj_covar=tensor([0.1127, 0.1050, 0.0917, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 05:28:23,308 INFO [train.py:968] (0/2) Epoch 14, batch 11550, giga_loss[loss=0.3845, simple_loss=0.4086, pruned_loss=0.1802, over 23499.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3734, pruned_loss=0.1246, over 5677522.62 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3523, pruned_loss=0.09645, over 5774618.20 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3763, pruned_loss=0.1283, over 5660062.78 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:28:54,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.808e+02 1.423e+03 1.901e+03 2.532e+03 4.503e+03, threshold=3.803e+03, percent-clipped=0.0
+2023-03-07 05:29:10,933 INFO [train.py:968] (0/2) Epoch 14, batch 11600, giga_loss[loss=0.3327, simple_loss=0.3972, pruned_loss=0.1341, over 28960.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3738, pruned_loss=0.1248, over 5675165.74 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3521, pruned_loss=0.09639, over 5776470.14 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3766, pruned_loss=0.1281, over 5658123.21 frames. ], batch size: 213, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:29:33,198 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-07 05:29:35,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5515, 1.5811, 1.8033, 1.3802], device='cuda:0'), covar=tensor([0.1475, 0.2144, 0.1199, 0.1471], device='cuda:0'), in_proj_covar=tensor([0.0852, 0.0699, 0.0894, 0.0797], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 05:29:50,206 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:29:53,129 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:29:58,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5777, 1.7375, 1.8562, 1.3665], device='cuda:0'), covar=tensor([0.1636, 0.2254, 0.1298, 0.1581], device='cuda:0'), in_proj_covar=tensor([0.0852, 0.0699, 0.0894, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 05:30:04,110 INFO [train.py:968] (0/2) Epoch 14, batch 11650, giga_loss[loss=0.3483, simple_loss=0.395, pruned_loss=0.1508, over 28637.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3753, pruned_loss=0.1255, over 5685003.25 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.352, pruned_loss=0.09639, over 5779263.71 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3781, pruned_loss=0.1286, over 5667249.99 frames. ], batch size: 336, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:30:06,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:30:22,977 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:30:33,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.760e+03 2.263e+03 2.920e+03 5.685e+03, threshold=4.526e+03, percent-clipped=7.0
+2023-03-07 05:30:52,262 INFO [train.py:968] (0/2) Epoch 14, batch 11700, giga_loss[loss=0.4118, simple_loss=0.43, pruned_loss=0.1968, over 26562.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3788, pruned_loss=0.1287, over 5681825.31 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3519, pruned_loss=0.09623, over 5782287.32 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3818, pruned_loss=0.1323, over 5662304.03 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:30:54,106 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=605002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:31:02,328 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 05:31:38,328 INFO [train.py:968] (0/2) Epoch 14, batch 11750, giga_loss[loss=0.3011, simple_loss=0.3728, pruned_loss=0.1147, over 28809.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3781, pruned_loss=0.1278, over 5690992.99 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3518, pruned_loss=0.09617, over 5780823.41 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3811, pruned_loss=0.1313, over 5674761.33 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:31:50,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=605064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:32:07,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.911e+02 1.785e+03 2.311e+03 2.975e+03 8.887e+03, threshold=4.623e+03, percent-clipped=5.0
+2023-03-07 05:32:20,997 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=605096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:32:23,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=605099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:32:24,322 INFO [train.py:968] (0/2) Epoch 14, batch 11800, giga_loss[loss=0.2827, simple_loss=0.362, pruned_loss=0.1017, over 28864.00 frames. ], tot_loss[loss=0.317, simple_loss=0.379, pruned_loss=0.1275, over 5692564.38 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3526, pruned_loss=0.09684, over 5783572.74 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3814, pruned_loss=0.1305, over 5674814.51 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:32:50,839 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=605128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:33:12,814 INFO [train.py:968] (0/2) Epoch 14, batch 11850, giga_loss[loss=0.2945, simple_loss=0.34, pruned_loss=0.1245, over 23424.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3782, pruned_loss=0.1261, over 5679187.28 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3527, pruned_loss=0.09688, over 5778685.38 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3804, pruned_loss=0.129, over 5667757.93 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:33:42,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.279e+02 1.558e+03 2.093e+03 2.914e+03 5.817e+03, threshold=4.185e+03, percent-clipped=6.0
+2023-03-07 05:34:01,273 INFO [train.py:968] (0/2) Epoch 14, batch 11900, giga_loss[loss=0.3275, simple_loss=0.3887, pruned_loss=0.1331, over 28594.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3763, pruned_loss=0.1248, over 5680211.08 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3526, pruned_loss=0.09681, over 5781049.47 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3787, pruned_loss=0.1277, over 5667195.30 frames. ], batch size: 336, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:34:19,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4215, 1.5690, 1.4264, 1.3248], device='cuda:0'), covar=tensor([0.1882, 0.1847, 0.1520, 0.1707], device='cuda:0'), in_proj_covar=tensor([0.1794, 0.1710, 0.1677, 0.1777], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 05:34:24,792 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=605227.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:34:46,199 INFO [train.py:968] (0/2) Epoch 14, batch 11950, giga_loss[loss=0.3128, simple_loss=0.3811, pruned_loss=0.1223, over 28839.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3754, pruned_loss=0.1242, over 5691380.38 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3523, pruned_loss=0.09652, over 5783087.74 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3778, pruned_loss=0.1271, over 5678231.75 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:34:54,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3294, 1.5507, 3.1046, 2.9695], device='cuda:0'), covar=tensor([0.1256, 0.2155, 0.0475, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0696, 0.0611, 0.0895, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 05:35:20,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.568e+03 2.115e+03 2.583e+03 5.008e+03, threshold=4.230e+03, percent-clipped=4.0
+2023-03-07 05:35:39,160 INFO [train.py:968] (0/2) Epoch 14, batch 12000, giga_loss[loss=0.3459, simple_loss=0.401, pruned_loss=0.1454, over 28744.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3779, pruned_loss=0.127, over 5668000.24 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3523, pruned_loss=0.09648, over 5780514.92 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.38, pruned_loss=0.1296, over 5658832.11 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:35:39,164 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 05:35:47,545 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2159, simple_loss=0.3232, pruned_loss=0.0543, over 944034.00 frames. 
+2023-03-07 05:35:47,546 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 05:36:00,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6555, 1.8205, 1.7667, 1.7410], device='cuda:0'), covar=tensor([0.1578, 0.1964, 0.2021, 0.1732], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0739, 0.0691, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 05:36:33,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2642, 1.3723, 3.9178, 3.1492], device='cuda:0'), covar=tensor([0.1626, 0.2508, 0.0418, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0699, 0.0614, 0.0899, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 05:36:34,014 INFO [train.py:968] (0/2) Epoch 14, batch 12050, giga_loss[loss=0.3008, simple_loss=0.3578, pruned_loss=0.1219, over 28769.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3784, pruned_loss=0.1272, over 5672062.68 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3524, pruned_loss=0.09644, over 5782778.56 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3805, pruned_loss=0.1299, over 5660785.43 frames. ], batch size: 99, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:36:50,057 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=605367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:37:00,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=605377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:37:06,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.624e+02 1.437e+03 1.827e+03 2.366e+03 1.199e+04, threshold=3.653e+03, percent-clipped=4.0
+2023-03-07 05:37:22,898 INFO [train.py:968] (0/2) Epoch 14, batch 12100, libri_loss[loss=0.2631, simple_loss=0.3521, pruned_loss=0.08705, over 29518.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3779, pruned_loss=0.1276, over 5671575.60 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3522, pruned_loss=0.09626, over 5785276.72 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3803, pruned_loss=0.1306, over 5657859.28 frames. ], batch size: 82, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:37:46,193 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7224, 0.9685, 2.9063, 2.9213], device='cuda:0'), covar=tensor([0.1818, 0.2636, 0.0610, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0699, 0.0614, 0.0898, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 05:37:56,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=605439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:38:06,696 INFO [train.py:968] (0/2) Epoch 14, batch 12150, giga_loss[loss=0.3197, simple_loss=0.3812, pruned_loss=0.1291, over 28587.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3777, pruned_loss=0.1276, over 5674790.38 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3523, pruned_loss=0.09642, over 5788780.57 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3808, pruned_loss=0.1314, over 5655152.94 frames. ], batch size: 336, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:38:38,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5942, 1.5347, 1.2758, 1.1978], device='cuda:0'), covar=tensor([0.0684, 0.0445, 0.0776, 0.1058], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0443, 0.0503, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 05:38:39,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.780e+02 1.575e+03 2.032e+03 2.608e+03 4.768e+03, threshold=4.065e+03, percent-clipped=7.0
+2023-03-07 05:38:40,238 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 05:38:55,674 INFO [train.py:968] (0/2) Epoch 14, batch 12200, giga_loss[loss=0.3386, simple_loss=0.3969, pruned_loss=0.1402, over 28356.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3781, pruned_loss=0.1279, over 5659210.26 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3526, pruned_loss=0.09655, over 5779006.99 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3807, pruned_loss=0.1313, over 5651165.11 frames. ], batch size: 368, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:39:07,069 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-07 05:39:14,892 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=605520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:39:19,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=605523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:39:34,467 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-07 05:39:45,135 INFO [train.py:968] (0/2) Epoch 14, batch 12250, giga_loss[loss=0.3476, simple_loss=0.3807, pruned_loss=0.1573, over 23503.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3786, pruned_loss=0.1282, over 5656965.57 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3525, pruned_loss=0.09649, over 5776605.25 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.381, pruned_loss=0.1312, over 5651608.79 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:39:47,886 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=605552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:40:15,832 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=605582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:40:16,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.388e+02 1.523e+03 1.886e+03 2.568e+03 6.036e+03, threshold=3.771e+03, percent-clipped=9.0
+2023-03-07 05:40:18,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=605585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:40:22,710 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-07 05:40:32,500 INFO [train.py:968] (0/2) Epoch 14, batch 12300, giga_loss[loss=0.2718, simple_loss=0.3514, pruned_loss=0.09615, over 28975.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3779, pruned_loss=0.128, over 5642307.87 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3526, pruned_loss=0.09652, over 5780365.15 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3804, pruned_loss=0.1313, over 5631499.32 frames. ], batch size: 213, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:40:35,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=605602.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:40:45,551 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=605614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:41:21,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5583, 1.5878, 1.8106, 1.3756], device='cuda:0'), covar=tensor([0.1382, 0.2121, 0.1134, 0.1422], device='cuda:0'), in_proj_covar=tensor([0.0850, 0.0700, 0.0896, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 05:41:22,187 INFO [train.py:968] (0/2) Epoch 14, batch 12350, giga_loss[loss=0.3471, simple_loss=0.4019, pruned_loss=0.1462, over 27829.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3791, pruned_loss=0.1288, over 5647359.26 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3527, pruned_loss=0.09648, over 5782399.95 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3814, pruned_loss=0.1318, over 5635473.22 frames. ], batch size: 412, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:41:50,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.173e+02 1.590e+03 2.017e+03 2.601e+03 8.616e+03, threshold=4.033e+03, percent-clipped=7.0
+2023-03-07 05:42:03,602 INFO [train.py:968] (0/2) Epoch 14, batch 12400, giga_loss[loss=0.3489, simple_loss=0.4072, pruned_loss=0.1453, over 27982.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3784, pruned_loss=0.1272, over 5656765.57 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3525, pruned_loss=0.09629, over 5779722.57 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3813, pruned_loss=0.1309, over 5644845.18 frames. ], batch size: 412, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:42:45,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=605742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:42:49,762 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=605745.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:42:52,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=605748.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:42:54,844 INFO [train.py:968] (0/2) Epoch 14, batch 12450, giga_loss[loss=0.3208, simple_loss=0.3786, pruned_loss=0.1315, over 28550.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.377, pruned_loss=0.1266, over 5658877.59 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3523, pruned_loss=0.09618, over 5782292.27 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.38, pruned_loss=0.1302, over 5645124.61 frames. ], batch size: 307, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:43:10,596 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8242, 2.1218, 1.7226, 2.0308], device='cuda:0'), covar=tensor([0.0712, 0.0263, 0.0289, 0.0770], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 05:43:21,518 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=605777.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:43:29,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.672e+03 2.140e+03 3.077e+03 1.483e+04, threshold=4.280e+03, percent-clipped=15.0
+2023-03-07 05:43:43,643 INFO [train.py:968] (0/2) Epoch 14, batch 12500, giga_loss[loss=0.2738, simple_loss=0.3399, pruned_loss=0.1039, over 28707.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3752, pruned_loss=0.1254, over 5662354.44 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3527, pruned_loss=0.09626, over 5783532.87 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3775, pruned_loss=0.1287, over 5648732.72 frames. ], batch size: 92, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:44:31,337 INFO [train.py:968] (0/2) Epoch 14, batch 12550, giga_loss[loss=0.33, simple_loss=0.3866, pruned_loss=0.1367, over 28842.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3732, pruned_loss=0.1244, over 5677269.20 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3529, pruned_loss=0.09639, over 5782804.90 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3753, pruned_loss=0.1275, over 5664522.81 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:44:57,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1517, 1.2452, 0.9927, 1.0892], device='cuda:0'), covar=tensor([0.1596, 0.1566, 0.1263, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.1802, 0.1716, 0.1678, 0.1785], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 05:45:06,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.793e+02 1.572e+03 1.976e+03 2.945e+03 7.661e+03, threshold=3.952e+03, percent-clipped=10.0
+2023-03-07 05:45:06,445 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=605885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:45:09,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=605888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:45:20,420 INFO [train.py:968] (0/2) Epoch 14, batch 12600, giga_loss[loss=0.3419, simple_loss=0.3934, pruned_loss=0.1452, over 27465.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3699, pruned_loss=0.1234, over 5657354.90 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3533, pruned_loss=0.0967, over 5783306.50 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3715, pruned_loss=0.1259, over 5645463.18 frames. ], batch size: 472, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:45:39,889 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=605917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:45:50,568 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=605930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:46:12,024 INFO [train.py:968] (0/2) Epoch 14, batch 12650, giga_loss[loss=0.2865, simple_loss=0.3573, pruned_loss=0.1079, over 29019.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3687, pruned_loss=0.1233, over 5656619.91 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.353, pruned_loss=0.09652, over 5783739.96 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3703, pruned_loss=0.1256, over 5646247.32 frames. ], batch size: 155, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:46:45,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.634e+03 1.998e+03 2.742e+03 1.146e+04, threshold=3.995e+03, percent-clipped=8.0
+2023-03-07 05:46:56,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3972, 1.6822, 1.6463, 1.2393], device='cuda:0'), covar=tensor([0.1664, 0.2458, 0.1408, 0.1691], device='cuda:0'), in_proj_covar=tensor([0.0851, 0.0698, 0.0895, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 05:47:01,149 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-606000.pt
+2023-03-07 05:47:01,434 INFO [train.py:968] (0/2) Epoch 14, batch 12700, giga_loss[loss=0.285, simple_loss=0.3584, pruned_loss=0.1058, over 28857.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3678, pruned_loss=0.1228, over 5655784.77 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3532, pruned_loss=0.09656, over 5783286.57 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3692, pruned_loss=0.1252, over 5644982.97 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:47:51,562 INFO [train.py:968] (0/2) Epoch 14, batch 12750, giga_loss[loss=0.284, simple_loss=0.3548, pruned_loss=0.1066, over 29076.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3667, pruned_loss=0.1204, over 5656113.67 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3533, pruned_loss=0.0967, over 5785131.85 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.368, pruned_loss=0.1226, over 5644110.37 frames. ], batch size: 128, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:47:51,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606050.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:48:24,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.458e+02 1.522e+03 1.983e+03 2.801e+03 8.503e+03, threshold=3.965e+03, percent-clipped=13.0
+2023-03-07 05:48:38,942 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-07 05:48:39,087 INFO [train.py:968] (0/2) Epoch 14, batch 12800, giga_loss[loss=0.2873, simple_loss=0.3645, pruned_loss=0.1051, over 28662.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3639, pruned_loss=0.1163, over 5657768.78 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3533, pruned_loss=0.09684, over 5786013.03 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3654, pruned_loss=0.1186, over 5643239.06 frames. ], batch size: 262, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:49:03,288 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606123.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:49:17,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-07 05:49:29,031 INFO [train.py:968] (0/2) Epoch 14, batch 12850, giga_loss[loss=0.2962, simple_loss=0.3728, pruned_loss=0.1099, over 28626.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3611, pruned_loss=0.113, over 5656264.47 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3532, pruned_loss=0.09688, over 5787847.09 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3626, pruned_loss=0.1152, over 5640243.82 frames. ], batch size: 307, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:50:00,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:50:03,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.702e+02 1.356e+03 1.745e+03 2.315e+03 6.560e+03, threshold=3.491e+03, percent-clipped=2.0
+2023-03-07 05:50:12,627 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7612, 4.5694, 4.2973, 2.2073], device='cuda:0'), covar=tensor([0.0607, 0.0798, 0.0937, 0.1984], device='cuda:0'), in_proj_covar=tensor([0.1115, 0.1036, 0.0904, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 05:50:19,991 INFO [train.py:968] (0/2) Epoch 14, batch 12900, giga_loss[loss=0.3169, simple_loss=0.3836, pruned_loss=0.1251, over 28894.00 frames. ], tot_loss[loss=0.289, simple_loss=0.358, pruned_loss=0.11, over 5661003.80 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3523, pruned_loss=0.09653, over 5790661.32 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3601, pruned_loss=0.1124, over 5642109.82 frames. ], batch size: 227, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:50:44,398 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:51:09,378 INFO [train.py:968] (0/2) Epoch 14, batch 12950, giga_loss[loss=0.2838, simple_loss=0.3558, pruned_loss=0.1059, over 28308.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3546, pruned_loss=0.1067, over 5660281.04 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3518, pruned_loss=0.0963, over 5793914.53 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3569, pruned_loss=0.1092, over 5638560.49 frames. ], batch size: 368, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:51:43,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.203e+02 1.257e+03 1.710e+03 2.190e+03 4.179e+03, threshold=3.421e+03, percent-clipped=4.0
+2023-03-07 05:52:00,269 INFO [train.py:968] (0/2) Epoch 14, batch 13000, giga_loss[loss=0.2595, simple_loss=0.343, pruned_loss=0.08793, over 28509.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3537, pruned_loss=0.1036, over 5669460.49 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3508, pruned_loss=0.09584, over 5790214.11 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3565, pruned_loss=0.1063, over 5651588.27 frames. ], batch size: 336, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:52:04,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:52:39,333 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606341.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:52:46,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4794, 1.7378, 1.3749, 1.6794], device='cuda:0'), covar=tensor([0.2558, 0.2335, 0.2625, 0.2198], device='cuda:0'), in_proj_covar=tensor([0.1363, 0.0997, 0.1205, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 05:52:47,123 INFO [train.py:968] (0/2) Epoch 14, batch 13050, giga_loss[loss=0.27, simple_loss=0.3439, pruned_loss=0.09807, over 28010.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3542, pruned_loss=0.1044, over 5656656.85 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3498, pruned_loss=0.09543, over 5785615.61 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3575, pruned_loss=0.1072, over 5641662.44 frames. ], batch size: 412, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 05:53:20,925 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.956e+02 1.553e+03 2.226e+03 3.237e+03 1.080e+04, threshold=4.453e+03, percent-clipped=22.0
+2023-03-07 05:53:24,859 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-07 05:53:34,455 INFO [train.py:968] (0/2) Epoch 14, batch 13100, giga_loss[loss=0.2798, simple_loss=0.3568, pruned_loss=0.1014, over 28312.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3536, pruned_loss=0.1037, over 5651370.50 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3497, pruned_loss=0.09559, over 5776969.20 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3564, pruned_loss=0.1061, over 5643627.26 frames. ], batch size: 368, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 05:54:01,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606425.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:54:24,393 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:54:25,371 INFO [train.py:968] (0/2) Epoch 14, batch 13150, giga_loss[loss=0.2393, simple_loss=0.3253, pruned_loss=0.07669, over 29087.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3499, pruned_loss=0.1013, over 5646003.14 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3489, pruned_loss=0.09519, over 5779943.32 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3529, pruned_loss=0.1037, over 5634300.37 frames. ], batch size: 155, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 05:54:26,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:54:27,754 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-07 05:54:49,625 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 05:54:57,075 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:55:02,130 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.967e+02 1.292e+03 1.731e+03 2.547e+03 6.018e+03, threshold=3.463e+03, percent-clipped=2.0
+2023-03-07 05:55:14,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606498.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:55:16,563 INFO [train.py:968] (0/2) Epoch 14, batch 13200, giga_loss[loss=0.2502, simple_loss=0.3086, pruned_loss=0.09592, over 24250.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3489, pruned_loss=0.1008, over 5644529.13 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3489, pruned_loss=0.09516, over 5781052.65 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3513, pruned_loss=0.1029, over 5632323.39 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 05:55:35,488 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:55:51,863 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6409, 4.7292, 1.7254, 1.7291], device='cuda:0'), covar=tensor([0.0919, 0.0264, 0.0914, 0.1323], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0527, 0.0355, 0.0402], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 05:55:53,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606537.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:56:00,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3855, 1.6374, 1.5023, 1.3710], device='cuda:0'), covar=tensor([0.2256, 0.1620, 0.1497, 0.1743], device='cuda:0'), in_proj_covar=tensor([0.1754, 0.1669, 0.1614, 0.1731], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 05:56:05,793 INFO [train.py:968] (0/2) Epoch 14, batch 13250, giga_loss[loss=0.2455, simple_loss=0.3242, pruned_loss=0.08339, over 28896.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3482, pruned_loss=0.1004, over 5648347.49 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3483, pruned_loss=0.09498, over 5779851.12 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3507, pruned_loss=0.1023, over 5637297.51 frames. ], batch size: 227, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 05:56:13,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:56:25,622 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606568.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:56:27,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606571.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:56:28,580 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-07 05:56:41,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.184e+02 1.402e+03 1.792e+03 2.632e+03 9.588e+03, threshold=3.584e+03, percent-clipped=12.0
+2023-03-07 05:56:49,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:56:50,357 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-07 05:56:53,713 INFO [train.py:968] (0/2) Epoch 14, batch 13300, giga_loss[loss=0.2212, simple_loss=0.288, pruned_loss=0.07716, over 24197.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3452, pruned_loss=0.09776, over 5654657.00 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3473, pruned_loss=0.09453, over 5779349.85 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3481, pruned_loss=0.09985, over 5641713.31 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 05:56:53,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606600.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:57:34,114 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606641.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:57:38,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606644.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:57:43,725 INFO [train.py:968] (0/2) Epoch 14, batch 13350, giga_loss[loss=0.2415, simple_loss=0.3258, pruned_loss=0.07862, over 28671.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3426, pruned_loss=0.09553, over 5651286.04 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3474, pruned_loss=0.09473, over 5776137.15 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3448, pruned_loss=0.09709, over 5640398.78 frames. ], batch size: 242, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 05:58:08,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606673.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:58:20,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.661e+02 1.318e+03 1.736e+03 2.472e+03 4.584e+03, threshold=3.472e+03, percent-clipped=6.0
+2023-03-07 05:58:30,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:58:35,216 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:58:36,088 INFO [train.py:968] (0/2) Epoch 14, batch 13400, giga_loss[loss=0.2269, simple_loss=0.3117, pruned_loss=0.07108, over 28788.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.338, pruned_loss=0.09282, over 5654465.26 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3468, pruned_loss=0.09442, over 5779461.40 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3402, pruned_loss=0.09433, over 5639993.61 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 05:58:37,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:58:52,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606716.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:59:06,880 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:59:07,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2602, 1.4598, 1.4085, 1.3649], device='cuda:0'), covar=tensor([0.0800, 0.0314, 0.0306, 0.1031], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 05:59:16,642 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:59:19,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:59:26,709 INFO [train.py:968] (0/2) Epoch 14, batch 13450, giga_loss[loss=0.2715, simple_loss=0.3479, pruned_loss=0.09759, over 28703.00 frames. ], tot_loss[loss=0.261, simple_loss=0.337, pruned_loss=0.09246, over 5651397.06 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3465, pruned_loss=0.09444, over 5763063.03 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3387, pruned_loss=0.09362, over 5648485.54 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 05:59:47,530 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:00:03,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.270e+02 1.283e+03 1.662e+03 2.765e+03 8.116e+03, threshold=3.324e+03, percent-clipped=13.0
+2023-03-07 06:00:12,719 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 06:00:17,974 INFO [train.py:968] (0/2) Epoch 14, batch 13500, giga_loss[loss=0.286, simple_loss=0.3599, pruned_loss=0.106, over 28596.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3364, pruned_loss=0.09293, over 5642250.46 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3459, pruned_loss=0.09422, over 5765048.87 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3381, pruned_loss=0.09401, over 5636703.30 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:01:18,590 INFO [train.py:968] (0/2) Epoch 14, batch 13550, giga_loss[loss=0.2984, simple_loss=0.3671, pruned_loss=0.1149, over 27647.00 frames. ], tot_loss[loss=0.264, simple_loss=0.339, pruned_loss=0.09448, over 5644925.14 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3459, pruned_loss=0.09434, over 5766507.29 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3401, pruned_loss=0.09521, over 5636264.13 frames. ], batch size: 472, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:01:23,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5174, 2.3150, 1.5557, 0.6021], device='cuda:0'), covar=tensor([0.4680, 0.2318, 0.3390, 0.4614], device='cuda:0'), in_proj_covar=tensor([0.1619, 0.1540, 0.1520, 0.1335], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 06:01:26,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606859.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:01:30,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606862.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:02:01,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.891e+02 1.358e+03 1.810e+03 2.428e+03 6.624e+03, threshold=3.619e+03, percent-clipped=9.0
+2023-03-07 06:02:06,479 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606891.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:02:07,675 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:02:15,792 INFO [train.py:968] (0/2) Epoch 14, batch 13600, giga_loss[loss=0.278, simple_loss=0.3574, pruned_loss=0.09928, over 28339.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3422, pruned_loss=0.09492, over 5649367.46 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3457, pruned_loss=0.09418, over 5768137.73 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3432, pruned_loss=0.09565, over 5639393.66 frames. ], batch size: 368, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:02:33,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606912.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:03:19,073 INFO [train.py:968] (0/2) Epoch 14, batch 13650, giga_loss[loss=0.3216, simple_loss=0.371, pruned_loss=0.1361, over 26859.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3436, pruned_loss=0.09577, over 5659738.49 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3458, pruned_loss=0.09428, over 5769489.12 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3443, pruned_loss=0.09628, over 5649665.30 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:04:06,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.552e+02 1.543e+03 2.006e+03 2.801e+03 8.129e+03, threshold=4.013e+03, percent-clipped=12.0
+2023-03-07 06:04:20,256 INFO [train.py:968] (0/2) Epoch 14, batch 13700, giga_loss[loss=0.2376, simple_loss=0.3225, pruned_loss=0.07635, over 28746.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3413, pruned_loss=0.09427, over 5668094.78 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3453, pruned_loss=0.09408, over 5772383.83 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3423, pruned_loss=0.09488, over 5655681.70 frames. ], batch size: 243, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:05:01,906 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=607032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:05:07,236 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:05:10,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:05:23,303 INFO [train.py:968] (0/2) Epoch 14, batch 13750, giga_loss[loss=0.2454, simple_loss=0.3386, pruned_loss=0.07604, over 28682.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3402, pruned_loss=0.09277, over 5668266.86 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3454, pruned_loss=0.09414, over 5773545.98 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3409, pruned_loss=0.09318, over 5656549.32 frames. ], batch size: 242, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:05:29,292 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607055.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:05:33,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607058.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:05:45,572 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=607067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:05:47,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=607070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:06:04,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.922e+02 1.138e+03 1.484e+03 2.106e+03 6.134e+03, threshold=2.969e+03, percent-clipped=5.0
+2023-03-07 06:06:04,937 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=607087.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:06:17,903 INFO [train.py:968] (0/2) Epoch 14, batch 13800, giga_loss[loss=0.2615, simple_loss=0.3392, pruned_loss=0.09195, over 28597.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.339, pruned_loss=0.09082, over 5677663.45 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3454, pruned_loss=0.09421, over 5776717.93 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3392, pruned_loss=0.09096, over 5661121.81 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:06:26,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4897, 1.6203, 1.7799, 1.3637], device='cuda:0'), covar=tensor([0.1743, 0.2322, 0.1444, 0.1728], device='cuda:0'), in_proj_covar=tensor([0.0845, 0.0683, 0.0888, 0.0791], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-07 06:06:37,321 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=2.00 vs. limit=2.0
+2023-03-07 06:06:42,903 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2969, 1.2483, 3.4051, 3.0323], device='cuda:0'), covar=tensor([0.1452, 0.2649, 0.0424, 0.1423], device='cuda:0'), in_proj_covar=tensor([0.0688, 0.0605, 0.0883, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 06:06:46,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2212, 1.0719, 3.9612, 3.2640], device='cuda:0'), covar=tensor([0.1682, 0.2933, 0.0405, 0.0876], device='cuda:0'), in_proj_covar=tensor([0.0688, 0.0605, 0.0883, 0.0796], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 06:07:23,736 INFO [train.py:968] (0/2) Epoch 14, batch 13850, giga_loss[loss=0.249, simple_loss=0.3254, pruned_loss=0.08632, over 29152.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3357, pruned_loss=0.09024, over 5672888.43 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3448, pruned_loss=0.09387, over 5778822.25 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3363, pruned_loss=0.09061, over 5656500.45 frames. ], batch size: 113, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:08:06,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.224e+02 1.369e+03 1.864e+03 2.813e+03 6.864e+03, threshold=3.728e+03, percent-clipped=21.0
+2023-03-07 06:08:16,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4925, 1.6320, 1.5745, 1.5219], device='cuda:0'), covar=tensor([0.1970, 0.1545, 0.1393, 0.1513], device='cuda:0'), in_proj_covar=tensor([0.1738, 0.1647, 0.1602, 0.1721], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 06:08:21,924 INFO [train.py:968] (0/2) Epoch 14, batch 13900, libri_loss[loss=0.2429, simple_loss=0.3293, pruned_loss=0.07828, over 26176.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3347, pruned_loss=0.09022, over 5669540.69 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3447, pruned_loss=0.09385, over 5775608.33 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3351, pruned_loss=0.09046, over 5657197.51 frames. ], batch size: 136, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:08:23,903 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=607202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:08:35,840 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:08:38,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:09:09,583 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=607245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:09:15,419 INFO [train.py:968] (0/2) Epoch 14, batch 13950, giga_loss[loss=0.2987, simple_loss=0.3681, pruned_loss=0.1147, over 27587.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3343, pruned_loss=0.09027, over 5671288.19 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3441, pruned_loss=0.0936, over 5778580.52 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3348, pruned_loss=0.09057, over 5654930.67 frames. ], batch size: 472, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:09:25,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7071, 1.7562, 1.3004, 1.3570], device='cuda:0'), covar=tensor([0.0830, 0.0584, 0.0990, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0436, 0.0500, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 06:10:00,811 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.355e+02 1.445e+03 2.079e+03 2.868e+03 9.141e+03, threshold=4.158e+03, percent-clipped=13.0
+2023-03-07 06:10:17,040 INFO [train.py:968] (0/2) Epoch 14, batch 14000, giga_loss[loss=0.2315, simple_loss=0.3249, pruned_loss=0.06909, over 28701.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3362, pruned_loss=0.09068, over 5662441.85 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3437, pruned_loss=0.09339, over 5777829.02 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3368, pruned_loss=0.09105, over 5648454.99 frames. ], batch size: 242, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:10:54,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5113, 1.8418, 1.7150, 1.6679], device='cuda:0'), covar=tensor([0.1656, 0.1874, 0.1912, 0.1735], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0713, 0.0671, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 06:11:20,325 INFO [train.py:968] (0/2) Epoch 14, batch 14050, giga_loss[loss=0.2028, simple_loss=0.297, pruned_loss=0.05434, over 28946.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3359, pruned_loss=0.08991, over 5663839.00 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3434, pruned_loss=0.0932, over 5777621.57 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3365, pruned_loss=0.09031, over 5650316.29 frames. ], batch size: 164, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:11:39,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8372, 2.6016, 1.6992, 0.6948], device='cuda:0'), covar=tensor([0.4624, 0.2352, 0.3006, 0.5265], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1532, 0.1514, 0.1333], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 06:11:51,728 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 06:12:02,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3467, 1.6878, 1.4128, 1.6462], device='cuda:0'), covar=tensor([0.0724, 0.0314, 0.0315, 0.0778], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0061, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 06:12:10,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.296e+02 1.343e+03 1.650e+03 2.275e+03 9.142e+03, threshold=3.300e+03, percent-clipped=3.0
+2023-03-07 06:12:27,388 INFO [train.py:968] (0/2) Epoch 14, batch 14100, giga_loss[loss=0.1878, simple_loss=0.2611, pruned_loss=0.05726, over 24715.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3337, pruned_loss=0.08882, over 5671655.49 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3427, pruned_loss=0.09296, over 5777833.54 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3346, pruned_loss=0.08926, over 5658185.92 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:12:36,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=607407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:13:04,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=607432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:13:05,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7006, 1.7315, 1.3866, 1.3192], device='cuda:0'), covar=tensor([0.0849, 0.0550, 0.0956, 0.1026], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0436, 0.0501, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 06:13:24,473 INFO [train.py:968] (0/2) Epoch 14, batch 14150, giga_loss[loss=0.2619, simple_loss=0.3535, pruned_loss=0.08515, over 28455.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3356, pruned_loss=0.08977, over 5689950.82 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3427, pruned_loss=0.09308, over 5782345.98 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.336, pruned_loss=0.08987, over 5670915.57 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:13:37,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9587, 2.4810, 2.0940, 1.7908], device='cuda:0'), covar=tensor([0.1927, 0.1346, 0.1675, 0.1696], device='cuda:0'), in_proj_covar=tensor([0.1746, 0.1657, 0.1608, 0.1730], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 06:13:39,022 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0651, 1.1537, 3.4077, 2.8952], device='cuda:0'), covar=tensor([0.1704, 0.2876, 0.0386, 0.1980], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0606, 0.0878, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 06:13:45,438 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 06:14:12,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.643e+02 1.285e+03 1.597e+03 2.340e+03 4.492e+03, threshold=3.194e+03, percent-clipped=10.0
+2023-03-07 06:14:28,160 INFO [train.py:968] (0/2) Epoch 14, batch 14200, giga_loss[loss=0.2579, simple_loss=0.352, pruned_loss=0.08185, over 28957.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3394, pruned_loss=0.09038, over 5685965.55 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3421, pruned_loss=0.0927, over 5784425.51 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3402, pruned_loss=0.09072, over 5665841.52 frames. ], batch size: 199, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:15:28,748 INFO [train.py:968] (0/2) Epoch 14, batch 14250, libri_loss[loss=0.2717, simple_loss=0.3505, pruned_loss=0.09642, over 29501.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3428, pruned_loss=0.09005, over 5684337.89 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3421, pruned_loss=0.09274, over 5783434.08 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3433, pruned_loss=0.09023, over 5666797.52 frames. ], batch size: 89, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:15:29,339 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:15:31,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:16:02,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=607577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:16:09,055 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=607582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:16:09,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6892, 1.7925, 1.2736, 1.3764], device='cuda:0'), covar=tensor([0.0894, 0.0625, 0.1024, 0.1090], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0436, 0.0501, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 06:16:15,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.431e+02 1.351e+03 1.802e+03 2.411e+03 8.618e+03, threshold=3.604e+03, percent-clipped=8.0
+2023-03-07 06:16:30,693 INFO [train.py:968] (0/2) Epoch 14, batch 14300, giga_loss[loss=0.2308, simple_loss=0.3291, pruned_loss=0.06628, over 29027.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3438, pruned_loss=0.08974, over 5682783.08 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3414, pruned_loss=0.09241, over 5785653.17 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3449, pruned_loss=0.09012, over 5665317.44 frames. ], batch size: 199, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:17:17,893 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-07 06:17:30,126 INFO [train.py:968] (0/2) Epoch 14, batch 14350, giga_loss[loss=0.258, simple_loss=0.3423, pruned_loss=0.08681, over 28918.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3417, pruned_loss=0.0885, over 5681316.16 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3407, pruned_loss=0.09217, over 5788789.95 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3432, pruned_loss=0.08893, over 5662486.88 frames. ], batch size: 227, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:18:16,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2119, 1.2314, 1.0623, 0.8329], device='cuda:0'), covar=tensor([0.0839, 0.0469, 0.1102, 0.1038], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0435, 0.0500, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 06:18:16,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.589e+02 1.272e+03 1.639e+03 2.163e+03 7.567e+03, threshold=3.278e+03, percent-clipped=4.0
+2023-03-07 06:18:34,289 INFO [train.py:968] (0/2) Epoch 14, batch 14400, giga_loss[loss=0.2722, simple_loss=0.3413, pruned_loss=0.1016, over 28654.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3423, pruned_loss=0.09014, over 5682869.64 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3407, pruned_loss=0.09218, over 5790891.88 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3436, pruned_loss=0.09043, over 5664080.00 frames. ], batch size: 99, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:19:00,485 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:19:03,753 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:19:39,605 INFO [train.py:968] (0/2) Epoch 14, batch 14450, giga_loss[loss=0.2735, simple_loss=0.352, pruned_loss=0.0975, over 28885.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3415, pruned_loss=0.09024, over 5695740.84 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3407, pruned_loss=0.09215, over 5792691.47 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3425, pruned_loss=0.09043, over 5675747.81 frames. ], batch size: 227, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:19:43,026 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=607752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:20:42,159 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.880e+02 1.292e+03 1.784e+03 2.600e+03 8.099e+03, threshold=3.567e+03, percent-clipped=14.0
+2023-03-07 06:20:58,826 INFO [train.py:968] (0/2) Epoch 14, batch 14500, giga_loss[loss=0.2638, simple_loss=0.3289, pruned_loss=0.0994, over 27131.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3414, pruned_loss=0.09113, over 5694511.83 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3411, pruned_loss=0.09247, over 5794191.84 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3418, pruned_loss=0.09096, over 5675915.17 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:21:12,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=607807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:22:14,318 INFO [train.py:968] (0/2) Epoch 14, batch 14550, giga_loss[loss=0.2329, simple_loss=0.3206, pruned_loss=0.07264, over 29042.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3373, pruned_loss=0.08885, over 5690618.85 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3411, pruned_loss=0.09247, over 5793317.97 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3376, pruned_loss=0.08866, over 5674112.44 frames. ], batch size: 285, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:23:05,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.067e+02 1.316e+03 1.827e+03 2.998e+03 8.539e+03, threshold=3.654e+03, percent-clipped=16.0
+2023-03-07 06:23:16,768 INFO [train.py:968] (0/2) Epoch 14, batch 14600, giga_loss[loss=0.2499, simple_loss=0.329, pruned_loss=0.0854, over 28790.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3348, pruned_loss=0.08736, over 5695633.88 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3402, pruned_loss=0.09201, over 5795509.93 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3357, pruned_loss=0.08749, over 5676954.76 frames. ], batch size: 243, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:24:00,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3561, 3.4461, 1.5753, 1.4535], device='cuda:0'), covar=tensor([0.0988, 0.0306, 0.0873, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0522, 0.0354, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 06:24:06,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.3023, 6.1359, 5.7766, 3.1779], device='cuda:0'), covar=tensor([0.0454, 0.0575, 0.0733, 0.1507], device='cuda:0'), in_proj_covar=tensor([0.1094, 0.1007, 0.0881, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 06:24:23,934 INFO [train.py:968] (0/2) Epoch 14, batch 14650, giga_loss[loss=0.2821, simple_loss=0.3617, pruned_loss=0.1013, over 28734.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3347, pruned_loss=0.08825, over 5668032.39 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.34, pruned_loss=0.092, over 5775774.82 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3355, pruned_loss=0.08832, over 5669018.21 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:24:24,346 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:24:29,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:25:02,978 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=607982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:25:12,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.723e+02 1.316e+03 1.776e+03 2.308e+03 9.275e+03, threshold=3.552e+03, percent-clipped=8.0
+2023-03-07 06:25:25,828 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-608000.pt
+2023-03-07 06:25:26,107 INFO [train.py:968] (0/2) Epoch 14, batch 14700, libri_loss[loss=0.2357, simple_loss=0.3166, pruned_loss=0.0774, over 29532.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3381, pruned_loss=0.08977, over 5675325.34 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3392, pruned_loss=0.09161, over 5779503.21 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3394, pruned_loss=0.0901, over 5669916.79 frames. ], batch size: 81, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:26:26,907 INFO [train.py:968] (0/2) Epoch 14, batch 14750, giga_loss[loss=0.2842, simple_loss=0.3476, pruned_loss=0.1104, over 28963.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3376, pruned_loss=0.09062, over 5678623.74 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3393, pruned_loss=0.09174, over 5777406.83 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3385, pruned_loss=0.09075, over 5674690.40 frames. ], batch size: 186, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:27:18,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.668e+02 1.355e+03 1.813e+03 2.567e+03 1.074e+04, threshold=3.626e+03, percent-clipped=9.0
+2023-03-07 06:27:33,534 INFO [train.py:968] (0/2) Epoch 14, batch 14800, giga_loss[loss=0.254, simple_loss=0.337, pruned_loss=0.08551, over 28839.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3377, pruned_loss=0.09162, over 5684640.57 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3391, pruned_loss=0.09159, over 5779715.41 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3387, pruned_loss=0.09184, over 5677071.20 frames. ], batch size: 174, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:28:28,916 INFO [train.py:968] (0/2) Epoch 14, batch 14850, giga_loss[loss=0.2811, simple_loss=0.35, pruned_loss=0.1061, over 27035.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3382, pruned_loss=0.0923, over 5679572.67 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3389, pruned_loss=0.09158, over 5774551.86 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.339, pruned_loss=0.09251, over 5675231.83 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:29:06,560 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-07 06:29:22,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.026e+02 1.476e+03 1.982e+03 2.887e+03 1.082e+04, threshold=3.964e+03, percent-clipped=16.0
+2023-03-07 06:29:35,173 INFO [train.py:968] (0/2) Epoch 14, batch 14900, giga_loss[loss=0.2984, simple_loss=0.3761, pruned_loss=0.1103, over 28747.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3391, pruned_loss=0.09154, over 5683274.63 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3385, pruned_loss=0.09132, over 5776138.97 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3403, pruned_loss=0.09197, over 5676184.86 frames. ], batch size: 243, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:29:39,079 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-07 06:30:10,695 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:30:24,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2967, 1.3949, 1.1662, 1.2620], device='cuda:0'), covar=tensor([0.1572, 0.1373, 0.1455, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.1747, 0.1644, 0.1587, 0.1721], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 06:30:27,273 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:30:49,243 INFO [train.py:968] (0/2) Epoch 14, batch 14950, giga_loss[loss=0.2507, simple_loss=0.3316, pruned_loss=0.08492, over 28662.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3397, pruned_loss=0.09121, over 5663009.24 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3386, pruned_loss=0.09143, over 5759962.98 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3405, pruned_loss=0.09146, over 5670510.83 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:31:51,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.044e+02 1.339e+03 1.694e+03 2.491e+03 5.524e+03, threshold=3.388e+03, percent-clipped=4.0
+2023-03-07 06:32:06,095 INFO [train.py:968] (0/2) Epoch 14, batch 15000, giga_loss[loss=0.2183, simple_loss=0.3071, pruned_loss=0.0647, over 28688.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3379, pruned_loss=0.09063, over 5658538.54 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3386, pruned_loss=0.09142, over 5763327.82 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3386, pruned_loss=0.09082, over 5659269.37 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:32:06,100 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 06:32:14,462 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2017, simple_loss=0.3019, pruned_loss=0.05081, over 944034.00 frames. 
+2023-03-07 06:32:14,463 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 06:32:40,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7384, 1.8610, 1.2526, 1.4508], device='cuda:0'), covar=tensor([0.0820, 0.0579, 0.0988, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0433, 0.0498, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 06:33:17,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5245, 1.7071, 1.5910, 1.4189], device='cuda:0'), covar=tensor([0.1961, 0.1604, 0.1101, 0.1467], device='cuda:0'), in_proj_covar=tensor([0.1745, 0.1639, 0.1585, 0.1720], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 06:33:17,681 INFO [train.py:968] (0/2) Epoch 14, batch 15050, giga_loss[loss=0.2043, simple_loss=0.2888, pruned_loss=0.05986, over 28812.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3338, pruned_loss=0.08969, over 5653784.55 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3382, pruned_loss=0.09138, over 5757683.33 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3346, pruned_loss=0.08985, over 5655666.51 frames. ], batch size: 164, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:34:09,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.973e+02 1.390e+03 1.838e+03 2.967e+03 5.922e+03, threshold=3.676e+03, percent-clipped=19.0
+2023-03-07 06:34:20,703 INFO [train.py:968] (0/2) Epoch 14, batch 15100, giga_loss[loss=0.267, simple_loss=0.3429, pruned_loss=0.09555, over 27522.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3301, pruned_loss=0.08794, over 5653960.70 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3386, pruned_loss=0.09159, over 5749370.70 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3302, pruned_loss=0.0878, over 5661580.54 frames. ], batch size: 472, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:35:20,510 INFO [train.py:968] (0/2) Epoch 14, batch 15150, giga_loss[loss=0.2766, simple_loss=0.3513, pruned_loss=0.101, over 28928.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3314, pruned_loss=0.08927, over 5643206.94 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3385, pruned_loss=0.09165, over 5742389.74 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3314, pruned_loss=0.08902, over 5653547.45 frames. ], batch size: 145, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:35:51,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2425, 1.1776, 3.8162, 3.1537], device='cuda:0'), covar=tensor([0.1670, 0.2794, 0.0415, 0.1043], device='cuda:0'), in_proj_covar=tensor([0.0678, 0.0601, 0.0873, 0.0784], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-07 06:36:02,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.261e+02 1.620e+03 2.043e+03 2.738e+03 7.604e+03, threshold=4.087e+03, percent-clipped=13.0
+2023-03-07 06:36:09,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:36:14,405 INFO [train.py:968] (0/2) Epoch 14, batch 15200, libri_loss[loss=0.2492, simple_loss=0.3298, pruned_loss=0.08432, over 29670.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3323, pruned_loss=0.08959, over 5656149.91 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3384, pruned_loss=0.09154, over 5743049.59 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3322, pruned_loss=0.08947, over 5661385.23 frames. ], batch size: 91, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:36:15,372 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5097, 2.1476, 1.5524, 0.6861], device='cuda:0'), covar=tensor([0.4635, 0.2339, 0.3580, 0.4979], device='cuda:0'), in_proj_covar=tensor([0.1617, 0.1539, 0.1517, 0.1333], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 06:36:27,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:36:39,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4680, 1.6516, 1.7784, 1.3678], device='cuda:0'), covar=tensor([0.1631, 0.2287, 0.1339, 0.1717], device='cuda:0'), in_proj_covar=tensor([0.0845, 0.0680, 0.0888, 0.0791], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:0')
+2023-03-07 06:36:42,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3785, 1.6398, 1.2917, 1.2859], device='cuda:0'), covar=tensor([0.2626, 0.2289, 0.2651, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.1360, 0.0997, 0.1205, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 06:37:14,299 INFO [train.py:968] (0/2) Epoch 14, batch 15250, giga_loss[loss=0.2536, simple_loss=0.3416, pruned_loss=0.08283, over 28667.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3297, pruned_loss=0.08764, over 5649791.86 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3382, pruned_loss=0.09141, over 5746796.42 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3296, pruned_loss=0.08757, over 5648101.47 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:38:02,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.086e+02 1.206e+03 1.554e+03 2.018e+03 4.438e+03, threshold=3.108e+03, percent-clipped=3.0
+2023-03-07 06:38:10,072 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:38:11,764 INFO [train.py:968] (0/2) Epoch 14, batch 15300, giga_loss[loss=0.2711, simple_loss=0.3415, pruned_loss=0.1003, over 28745.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3287, pruned_loss=0.08659, over 5664924.98 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3378, pruned_loss=0.09124, over 5749021.45 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3288, pruned_loss=0.08655, over 5658879.71 frames. ], batch size: 243, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:38:12,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=608601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:38:28,344 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=608609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:39:23,051 INFO [train.py:968] (0/2) Epoch 14, batch 15350, giga_loss[loss=0.2126, simple_loss=0.2982, pruned_loss=0.06344, over 28820.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3271, pruned_loss=0.08624, over 5654993.92 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3372, pruned_loss=0.09092, over 5748446.78 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3273, pruned_loss=0.08637, over 5648300.04 frames. ], batch size: 243, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:40:16,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.565e+02 1.284e+03 1.745e+03 2.721e+03 1.188e+04, threshold=3.491e+03, percent-clipped=21.0
+2023-03-07 06:40:30,260 INFO [train.py:968] (0/2) Epoch 14, batch 15400, giga_loss[loss=0.2425, simple_loss=0.325, pruned_loss=0.08002, over 29059.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3277, pruned_loss=0.08619, over 5651148.92 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.337, pruned_loss=0.09084, over 5749584.35 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3279, pruned_loss=0.08625, over 5643361.16 frames. ], batch size: 199, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:40:40,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2276, 1.8461, 1.4141, 0.3204], device='cuda:0'), covar=tensor([0.3715, 0.2651, 0.3566, 0.5040], device='cuda:0'), in_proj_covar=tensor([0.1613, 0.1541, 0.1516, 0.1332], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 06:41:24,204 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=608744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:41:27,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=608747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:41:30,310 INFO [train.py:968] (0/2) Epoch 14, batch 15450, giga_loss[loss=0.2494, simple_loss=0.3267, pruned_loss=0.08601, over 29125.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3272, pruned_loss=0.08582, over 5659607.23 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3368, pruned_loss=0.0907, over 5753632.82 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3273, pruned_loss=0.08586, over 5647118.28 frames. ], batch size: 200, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:41:32,767 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=608752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:41:36,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=608755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:41:36,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3832, 3.3504, 1.4587, 1.4731], device='cuda:0'), covar=tensor([0.0951, 0.0407, 0.0926, 0.1323], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0516, 0.0352, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 06:42:02,318 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608776.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:42:02,334 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=608776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:42:13,419 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=608784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:42:24,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.829e+02 1.278e+03 1.533e+03 2.119e+03 4.012e+03, threshold=3.065e+03, percent-clipped=2.0
+2023-03-07 06:42:33,892 INFO [train.py:968] (0/2) Epoch 14, batch 15500, giga_loss[loss=0.2269, simple_loss=0.3102, pruned_loss=0.07175, over 28099.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3274, pruned_loss=0.08656, over 5660309.06 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3361, pruned_loss=0.0903, over 5753677.00 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3278, pruned_loss=0.08686, over 5648291.70 frames. ], batch size: 412, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:43:37,077 INFO [train.py:968] (0/2) Epoch 14, batch 15550, giga_loss[loss=0.2454, simple_loss=0.3311, pruned_loss=0.07985, over 28427.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3271, pruned_loss=0.08543, over 5668477.62 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3357, pruned_loss=0.09019, over 5754982.72 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3276, pruned_loss=0.08568, over 5656057.98 frames. ], batch size: 368, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:44:01,934 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=608871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:44:04,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:44:20,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=608887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:44:26,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.172e+02 1.305e+03 1.844e+03 2.793e+03 6.782e+03, threshold=3.688e+03, percent-clipped=19.0
+2023-03-07 06:44:36,179 INFO [train.py:968] (0/2) Epoch 14, batch 15600, giga_loss[loss=0.2719, simple_loss=0.357, pruned_loss=0.09337, over 28878.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3294, pruned_loss=0.08528, over 5665519.38 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3354, pruned_loss=0.09008, over 5748337.33 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.33, pruned_loss=0.08549, over 5660359.31 frames. ], batch size: 227, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:45:15,400 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4118, 1.4229, 1.4133, 1.5446], device='cuda:0'), covar=tensor([0.0769, 0.0326, 0.0313, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 06:45:34,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3035, 1.2847, 1.1699, 1.4993], device='cuda:0'), covar=tensor([0.0780, 0.0367, 0.0351, 0.0842], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 06:45:39,234 INFO [train.py:968] (0/2) Epoch 14, batch 15650, giga_loss[loss=0.2378, simple_loss=0.3291, pruned_loss=0.0732, over 28679.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3313, pruned_loss=0.0859, over 5653926.93 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3354, pruned_loss=0.09014, over 5741034.71 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3317, pruned_loss=0.08594, over 5654368.97 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:46:04,123 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608971.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:46:06,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=608974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:46:27,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.015e+02 1.463e+03 1.923e+03 2.700e+03 4.769e+03, threshold=3.846e+03, percent-clipped=10.0
+2023-03-07 06:46:36,488 INFO [train.py:968] (0/2) Epoch 14, batch 15700, giga_loss[loss=0.2659, simple_loss=0.3429, pruned_loss=0.09446, over 28644.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3331, pruned_loss=0.08704, over 5665162.54 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.335, pruned_loss=0.09012, over 5743267.94 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3337, pruned_loss=0.08705, over 5661936.29 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:46:36,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3635, 1.6348, 1.5448, 1.5040], device='cuda:0'), covar=tensor([0.1480, 0.1685, 0.1761, 0.1608], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0714, 0.0666, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 06:46:54,040 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:46:57,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:15,404 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:18,362 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:33,187 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:35,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:36,342 INFO [train.py:968] (0/2) Epoch 14, batch 15750, giga_loss[loss=0.2597, simple_loss=0.343, pruned_loss=0.08817, over 28509.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3327, pruned_loss=0.08671, over 5677542.90 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3346, pruned_loss=0.08984, over 5744139.00 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3335, pruned_loss=0.08687, over 5671674.48 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:47:50,368 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:48:27,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.131e+02 1.274e+03 1.616e+03 2.366e+03 6.190e+03, threshold=3.232e+03, percent-clipped=5.0
+2023-03-07 06:48:32,997 INFO [train.py:968] (0/2) Epoch 14, batch 15800, giga_loss[loss=0.2177, simple_loss=0.3095, pruned_loss=0.06298, over 28464.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3289, pruned_loss=0.08378, over 5687450.64 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3341, pruned_loss=0.08957, over 5747800.63 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3298, pruned_loss=0.084, over 5677738.11 frames. ], batch size: 369, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:48:55,841 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:48:59,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:49:33,572 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:49:33,954 INFO [train.py:968] (0/2) Epoch 14, batch 15850, giga_loss[loss=0.2244, simple_loss=0.2904, pruned_loss=0.07925, over 24463.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3281, pruned_loss=0.08418, over 5684025.98 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3337, pruned_loss=0.08931, over 5749225.57 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3291, pruned_loss=0.08448, over 5673975.38 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:49:38,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609151.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:49:59,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:50:27,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.527e+02 1.305e+03 1.635e+03 2.065e+03 6.643e+03, threshold=3.270e+03, percent-clipped=6.0
+2023-03-07 06:50:35,593 INFO [train.py:968] (0/2) Epoch 14, batch 15900, giga_loss[loss=0.2521, simple_loss=0.3347, pruned_loss=0.08477, over 28793.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3267, pruned_loss=0.08403, over 5681368.80 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3335, pruned_loss=0.08917, over 5751548.81 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3276, pruned_loss=0.08433, over 5670383.32 frames. ], batch size: 243, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:50:47,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8828, 1.2035, 1.2767, 1.0339], device='cuda:0'), covar=tensor([0.1598, 0.1302, 0.2041, 0.1519], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0710, 0.0661, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 06:51:32,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5891, 1.9824, 1.8490, 1.7587], device='cuda:0'), covar=tensor([0.1634, 0.1744, 0.1949, 0.1690], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0714, 0.0665, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 06:51:36,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:51:37,519 INFO [train.py:968] (0/2) Epoch 14, batch 15950, libri_loss[loss=0.2051, simple_loss=0.2835, pruned_loss=0.06337, over 28510.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3294, pruned_loss=0.08558, over 5675441.61 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.333, pruned_loss=0.08893, over 5750913.16 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3305, pruned_loss=0.08594, over 5665492.11 frames. ], batch size: 63, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:52:13,886 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-07 06:52:23,642 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:52:30,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.841e+02 1.436e+03 2.110e+03 2.919e+03 7.688e+03, threshold=4.221e+03, percent-clipped=17.0
+2023-03-07 06:52:34,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609294.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:52:36,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609297.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:52:41,158 INFO [train.py:968] (0/2) Epoch 14, batch 16000, giga_loss[loss=0.272, simple_loss=0.3377, pruned_loss=0.1031, over 26812.00 frames. ], tot_loss[loss=0.252, simple_loss=0.331, pruned_loss=0.08653, over 5682301.94 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.333, pruned_loss=0.08889, over 5752350.21 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3318, pruned_loss=0.08678, over 5671498.31 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:53:00,309 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-07 06:53:13,554 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609326.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:53:37,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609346.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:53:42,791 INFO [train.py:968] (0/2) Epoch 14, batch 16050, giga_loss[loss=0.2698, simple_loss=0.3455, pruned_loss=0.09708, over 27420.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.333, pruned_loss=0.08805, over 5677806.97 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3333, pruned_loss=0.08909, over 5753427.33 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3334, pruned_loss=0.08803, over 5667031.04 frames. ], batch size: 472, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:54:14,022 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:54:30,614 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:54:30,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.373e+02 1.403e+03 1.834e+03 2.347e+03 5.279e+03, threshold=3.668e+03, percent-clipped=5.0
+2023-03-07 06:54:34,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:54:38,452 INFO [train.py:968] (0/2) Epoch 14, batch 16100, giga_loss[loss=0.2653, simple_loss=0.3483, pruned_loss=0.09121, over 28838.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3367, pruned_loss=0.08925, over 5689725.48 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3332, pruned_loss=0.08907, over 5758169.23 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3371, pruned_loss=0.08926, over 5674839.65 frames. ], batch size: 227, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:54:59,435 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1443, 1.4675, 1.4886, 1.3313], device='cuda:0'), covar=tensor([0.1611, 0.1665, 0.1989, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0709, 0.0659, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 06:55:03,752 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:55:03,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:55:35,385 INFO [train.py:968] (0/2) Epoch 14, batch 16150, giga_loss[loss=0.2666, simple_loss=0.3406, pruned_loss=0.09634, over 27540.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3375, pruned_loss=0.08922, over 5670126.88 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3332, pruned_loss=0.08913, over 5742389.53 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.338, pruned_loss=0.08916, over 5670766.47 frames. ], batch size: 472, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:56:29,269 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609489.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:56:34,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609492.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:56:35,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.497e+02 1.309e+03 1.795e+03 2.391e+03 9.796e+03, threshold=3.591e+03, percent-clipped=11.0
+2023-03-07 06:56:44,029 INFO [train.py:968] (0/2) Epoch 14, batch 16200, giga_loss[loss=0.2738, simple_loss=0.3468, pruned_loss=0.1004, over 28965.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3376, pruned_loss=0.08941, over 5680262.32 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3331, pruned_loss=0.08903, over 5745414.22 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3383, pruned_loss=0.08948, over 5676193.69 frames. ], batch size: 199, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:57:08,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3283, 1.2868, 1.2027, 1.5110], device='cuda:0'), covar=tensor([0.0751, 0.0347, 0.0338, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0113, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 06:57:10,217 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609521.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:57:39,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:57:44,893 INFO [train.py:968] (0/2) Epoch 14, batch 16250, giga_loss[loss=0.2664, simple_loss=0.3432, pruned_loss=0.0948, over 29052.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3354, pruned_loss=0.08841, over 5691924.49 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3326, pruned_loss=0.08879, over 5746635.61 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3364, pruned_loss=0.08867, over 5686096.18 frames. ], batch size: 285, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:58:07,432 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:58:13,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:58:36,023 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-07 06:58:43,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.990e+02 1.312e+03 1.594e+03 2.098e+03 5.791e+03, threshold=3.188e+03, percent-clipped=5.0
+2023-03-07 06:58:53,500 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:58:54,663 INFO [train.py:968] (0/2) Epoch 14, batch 16300, giga_loss[loss=0.256, simple_loss=0.3391, pruned_loss=0.08647, over 28802.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.336, pruned_loss=0.08933, over 5679783.24 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3327, pruned_loss=0.0888, over 5747252.71 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3367, pruned_loss=0.08953, over 5673524.89 frames. ], batch size: 174, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:58:55,850 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-07 06:59:58,127 INFO [train.py:968] (0/2) Epoch 14, batch 16350, giga_loss[loss=0.2337, simple_loss=0.3211, pruned_loss=0.07315, over 28686.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3347, pruned_loss=0.08959, over 5672144.88 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3324, pruned_loss=0.08866, over 5745255.03 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3356, pruned_loss=0.08988, over 5668109.66 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:00:11,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:00:27,999 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-07 07:00:41,883 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:00:45,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:00:47,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.835e+02 1.342e+03 1.681e+03 2.330e+03 4.948e+03, threshold=3.362e+03, percent-clipped=8.0
+2023-03-07 07:00:56,336 INFO [train.py:968] (0/2) Epoch 14, batch 16400, giga_loss[loss=0.265, simple_loss=0.3419, pruned_loss=0.09404, over 28658.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3317, pruned_loss=0.08881, over 5675225.79 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3313, pruned_loss=0.08817, over 5751177.46 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3335, pruned_loss=0.08954, over 5664074.36 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 07:01:19,345 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:01:52,912 INFO [train.py:968] (0/2) Epoch 14, batch 16450, giga_loss[loss=0.246, simple_loss=0.3119, pruned_loss=0.09007, over 24557.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3323, pruned_loss=0.08832, over 5678200.24 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3313, pruned_loss=0.0881, over 5748352.26 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3338, pruned_loss=0.08899, over 5669142.45 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:01:55,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:02:43,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.908e+02 1.350e+03 1.603e+03 2.538e+03 9.472e+03, threshold=3.206e+03, percent-clipped=15.0
+2023-03-07 07:02:51,373 INFO [train.py:968] (0/2) Epoch 14, batch 16500, giga_loss[loss=0.2319, simple_loss=0.3134, pruned_loss=0.07522, over 28749.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3318, pruned_loss=0.08768, over 5676739.04 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3313, pruned_loss=0.08821, over 5751861.96 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3331, pruned_loss=0.08813, over 5664343.72 frames. ], batch size: 99, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:02:54,204 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:02:57,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:03:19,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8108, 4.6596, 4.3914, 2.0030], device='cuda:0'), covar=tensor([0.0390, 0.0546, 0.0649, 0.1891], device='cuda:0'), in_proj_covar=tensor([0.1087, 0.1000, 0.0874, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 07:03:32,466 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:03:33,870 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:03:49,428 INFO [train.py:968] (0/2) Epoch 14, batch 16550, giga_loss[loss=0.2152, simple_loss=0.3137, pruned_loss=0.05831, over 29006.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3331, pruned_loss=0.08637, over 5675046.36 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.331, pruned_loss=0.08811, over 5744760.94 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3344, pruned_loss=0.08679, over 5670849.25 frames. ], batch size: 128, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:04:11,038 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609870.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:04:41,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.332e+02 1.226e+03 1.577e+03 2.264e+03 4.887e+03, threshold=3.154e+03, percent-clipped=10.0
+2023-03-07 07:04:44,258 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:04:47,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:04:49,003 INFO [train.py:968] (0/2) Epoch 14, batch 16600, giga_loss[loss=0.2353, simple_loss=0.3253, pruned_loss=0.07264, over 28922.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3342, pruned_loss=0.08555, over 5670519.60 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3308, pruned_loss=0.08803, over 5744190.93 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3354, pruned_loss=0.08591, over 5666350.20 frames. ], batch size: 213, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:05:19,110 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:05:47,378 INFO [train.py:968] (0/2) Epoch 14, batch 16650, giga_loss[loss=0.2461, simple_loss=0.333, pruned_loss=0.07954, over 28953.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3339, pruned_loss=0.08482, over 5684106.12 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3308, pruned_loss=0.08804, over 5745861.84 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3349, pruned_loss=0.08507, over 5678655.87 frames. ], batch size: 284, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:06:46,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.275e+02 1.373e+03 1.747e+03 2.542e+03 1.027e+04, threshold=3.493e+03, percent-clipped=13.0
+2023-03-07 07:06:54,377 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-610000.pt
+2023-03-07 07:06:54,666 INFO [train.py:968] (0/2) Epoch 14, batch 16700, giga_loss[loss=0.2171, simple_loss=0.3067, pruned_loss=0.06379, over 29027.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3335, pruned_loss=0.085, over 5667989.42 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3308, pruned_loss=0.08813, over 5740010.32 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3344, pruned_loss=0.08504, over 5667469.59 frames. ], batch size: 128, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:07:58,967 INFO [train.py:968] (0/2) Epoch 14, batch 16750, giga_loss[loss=0.3084, simple_loss=0.3691, pruned_loss=0.1239, over 28139.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3349, pruned_loss=0.08603, over 5674219.02 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.331, pruned_loss=0.08838, over 5743652.16 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3354, pruned_loss=0.08576, over 5669016.88 frames. ], batch size: 412, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:08:22,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8544, 1.9765, 1.3978, 1.6446], device='cuda:0'), covar=tensor([0.0900, 0.0683, 0.1017, 0.1183], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0434, 0.0500, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 07:09:00,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.958e+02 1.282e+03 1.744e+03 2.502e+03 4.525e+03, threshold=3.488e+03, percent-clipped=8.0
+2023-03-07 07:09:06,746 INFO [train.py:968] (0/2) Epoch 14, batch 16800, libri_loss[loss=0.313, simple_loss=0.3736, pruned_loss=0.1262, over 27683.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3352, pruned_loss=0.0855, over 5659934.25 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3314, pruned_loss=0.08874, over 5726934.38 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3353, pruned_loss=0.08484, over 5668756.95 frames. ], batch size: 115, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 07:09:21,589 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=610112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:09:30,574 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=610120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:10:10,573 INFO [train.py:968] (0/2) Epoch 14, batch 16850, giga_loss[loss=0.311, simple_loss=0.373, pruned_loss=0.1245, over 27606.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3377, pruned_loss=0.08751, over 5663037.72 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3317, pruned_loss=0.08916, over 5723394.19 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3377, pruned_loss=0.08649, over 5670823.59 frames. ], batch size: 472, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:10:18,939 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-07 07:10:40,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4632, 1.6546, 1.3114, 1.6270], device='cuda:0'), covar=tensor([0.2767, 0.2702, 0.3245, 0.2204], device='cuda:0'), in_proj_covar=tensor([0.1346, 0.0987, 0.1195, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 07:10:54,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0807, 1.0832, 3.7073, 3.0592], device='cuda:0'), covar=tensor([0.1693, 0.2746, 0.0406, 0.0891], device='cuda:0'), in_proj_covar=tensor([0.0686, 0.0605, 0.0875, 0.0792], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 07:11:15,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.557e+02 1.481e+03 1.906e+03 2.718e+03 4.252e+03, threshold=3.812e+03, percent-clipped=9.0
+2023-03-07 07:11:23,906 INFO [train.py:968] (0/2) Epoch 14, batch 16900, giga_loss[loss=0.2713, simple_loss=0.3574, pruned_loss=0.09254, over 28757.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3408, pruned_loss=0.08852, over 5662452.54 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3317, pruned_loss=0.08919, over 5716369.88 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3409, pruned_loss=0.08769, over 5674184.00 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:11:37,009 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=610212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:12:19,689 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-07 07:12:20,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=610245.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 07:12:25,463 INFO [train.py:968] (0/2) Epoch 14, batch 16950, libri_loss[loss=0.2621, simple_loss=0.3368, pruned_loss=0.09367, over 28536.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.339, pruned_loss=0.08761, over 5671884.04 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3316, pruned_loss=0.08906, over 5720868.77 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3394, pruned_loss=0.08701, over 5675752.90 frames. ], batch size: 106, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:12:34,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-07 07:13:10,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5603, 4.4089, 4.1835, 2.0029], device='cuda:0'), covar=tensor([0.0502, 0.0642, 0.0763, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.1094, 0.1001, 0.0877, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 07:13:14,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1723, 1.4137, 1.5010, 1.2578], device='cuda:0'), covar=tensor([0.1511, 0.1658, 0.2121, 0.1886], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0710, 0.0661, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 07:13:31,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.453e+02 1.354e+03 1.694e+03 2.219e+03 7.497e+03, threshold=3.388e+03, percent-clipped=6.0
+2023-03-07 07:13:38,885 INFO [train.py:968] (0/2) Epoch 14, batch 17000, giga_loss[loss=0.2687, simple_loss=0.3441, pruned_loss=0.09666, over 27628.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3369, pruned_loss=0.08697, over 5683402.66 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3315, pruned_loss=0.08897, over 5722771.42 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3374, pruned_loss=0.08658, over 5684336.98 frames. ], batch size: 474, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:14:26,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2754, 1.4704, 1.3009, 1.5239], device='cuda:0'), covar=tensor([0.0715, 0.0438, 0.0342, 0.0743], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0061, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 07:14:50,506 INFO [train.py:968] (0/2) Epoch 14, batch 17050, giga_loss[loss=0.2151, simple_loss=0.3083, pruned_loss=0.06092, over 28986.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3342, pruned_loss=0.08468, over 5681419.86 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.331, pruned_loss=0.08864, over 5716009.33 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3351, pruned_loss=0.0846, over 5686615.86 frames. ], batch size: 136, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:14:55,314 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=610355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:14:58,405 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=610358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:15:39,395 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=610387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:15:40,764 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=610388.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:15:43,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=610391.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:15:49,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.525e+02 1.330e+03 1.701e+03 2.865e+03 5.870e+03, threshold=3.403e+03, percent-clipped=14.0
+2023-03-07 07:15:53,290 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-07 07:15:55,886 INFO [train.py:968] (0/2) Epoch 14, batch 17100, giga_loss[loss=0.2588, simple_loss=0.3415, pruned_loss=0.08808, over 28510.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3337, pruned_loss=0.0844, over 5682534.25 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3311, pruned_loss=0.08867, over 5713395.28 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3343, pruned_loss=0.0843, over 5688515.18 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:16:22,407 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=610420.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:16:44,283 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=610442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:16:54,401 INFO [train.py:968] (0/2) Epoch 14, batch 17150, giga_loss[loss=0.2494, simple_loss=0.3346, pruned_loss=0.08214, over 28962.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3344, pruned_loss=0.08508, over 5683898.73 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3307, pruned_loss=0.08865, over 5719133.19 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3353, pruned_loss=0.08491, over 5682610.02 frames. ], batch size: 136, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:16:57,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3273, 3.1334, 2.9785, 1.5633], device='cuda:0'), covar=tensor([0.0900, 0.1023, 0.1006, 0.2230], device='cuda:0'), in_proj_covar=tensor([0.1091, 0.0998, 0.0878, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 07:17:03,034 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 07:17:38,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=610487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:17:46,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.768e+02 1.261e+03 1.641e+03 2.450e+03 6.496e+03, threshold=3.282e+03, percent-clipped=9.0
+2023-03-07 07:17:49,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=610495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:17:53,300 INFO [train.py:968] (0/2) Epoch 14, batch 17200, libri_loss[loss=0.2544, simple_loss=0.3363, pruned_loss=0.0863, over 25654.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3375, pruned_loss=0.08726, over 5668340.92 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3307, pruned_loss=0.0886, over 5708336.08 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3383, pruned_loss=0.08713, over 5676494.40 frames. ], batch size: 136, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 07:18:07,782 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=610513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:18:43,030 INFO [train.py:968] (0/2) Epoch 14, batch 17250, giga_loss[loss=0.2422, simple_loss=0.3229, pruned_loss=0.08079, over 28332.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3362, pruned_loss=0.08731, over 5665695.87 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3304, pruned_loss=0.08861, over 5706049.58 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3374, pruned_loss=0.08713, over 5673300.25 frames. ], batch size: 368, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:19:17,703 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 07:19:34,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.269e+02 1.404e+03 1.806e+03 2.333e+03 4.699e+03, threshold=3.612e+03, percent-clipped=4.0
+2023-03-07 07:19:38,406 INFO [train.py:968] (0/2) Epoch 14, batch 17300, giga_loss[loss=0.2604, simple_loss=0.3386, pruned_loss=0.09107, over 28758.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3348, pruned_loss=0.08781, over 5659599.07 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3304, pruned_loss=0.08865, over 5700795.50 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3359, pruned_loss=0.08759, over 5670055.12 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:20:15,217 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=610630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:20:18,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=610633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:20:23,302 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=610638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:20:26,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=610641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:20:37,809 INFO [train.py:968] (0/2) Epoch 14, batch 17350, giga_loss[loss=0.28, simple_loss=0.3601, pruned_loss=0.09993, over 28886.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3359, pruned_loss=0.08871, over 5669647.03 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3305, pruned_loss=0.08868, over 5700981.84 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3367, pruned_loss=0.0885, over 5677342.66 frames. ], batch size: 106, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:20:51,075 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=610662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:20:57,664 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=610670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:21:29,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.890e+02 1.407e+03 1.809e+03 2.181e+03 5.491e+03, threshold=3.618e+03, percent-clipped=4.0
+2023-03-07 07:21:34,357 INFO [train.py:968] (0/2) Epoch 14, batch 17400, giga_loss[loss=0.3099, simple_loss=0.3856, pruned_loss=0.1171, over 28807.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3443, pruned_loss=0.09425, over 5670374.22 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3301, pruned_loss=0.0885, over 5704696.82 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3455, pruned_loss=0.09427, over 5672911.52 frames. ], batch size: 199, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:21:51,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5680, 4.4363, 1.7053, 1.7511], device='cuda:0'), covar=tensor([0.0942, 0.0197, 0.0870, 0.1223], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0512, 0.0353, 0.0400], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 07:22:03,571 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 07:22:17,555 INFO [train.py:968] (0/2) Epoch 14, batch 17450, giga_loss[loss=0.295, simple_loss=0.3792, pruned_loss=0.1055, over 28713.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3532, pruned_loss=0.09931, over 5682726.04 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3303, pruned_loss=0.08857, over 5709643.34 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3543, pruned_loss=0.09944, over 5679861.98 frames. ], batch size: 242, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:22:57,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.010e+02 1.147e+03 1.633e+03 2.563e+03 8.276e+03, threshold=3.267e+03, percent-clipped=7.0
+2023-03-07 07:23:02,084 INFO [train.py:968] (0/2) Epoch 14, batch 17500, giga_loss[loss=0.245, simple_loss=0.3231, pruned_loss=0.08339, over 29073.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3538, pruned_loss=0.1001, over 5687644.89 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3304, pruned_loss=0.08873, over 5711914.45 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.355, pruned_loss=0.1002, over 5682878.60 frames. ], batch size: 136, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:23:17,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=610817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:23:48,347 INFO [train.py:968] (0/2) Epoch 14, batch 17550, giga_loss[loss=0.2075, simple_loss=0.2826, pruned_loss=0.0662, over 28410.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3473, pruned_loss=0.09758, over 5687436.58 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3304, pruned_loss=0.08875, over 5713954.87 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3484, pruned_loss=0.09777, over 5681777.11 frames. ], batch size: 65, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:24:18,970 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=610888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:24:26,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.547e+02 1.124e+03 1.350e+03 1.647e+03 4.736e+03, threshold=2.701e+03, percent-clipped=2.0
+2023-03-07 07:24:30,552 INFO [train.py:968] (0/2) Epoch 14, batch 17600, giga_loss[loss=0.2395, simple_loss=0.3084, pruned_loss=0.08535, over 28451.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3412, pruned_loss=0.09511, over 5672717.24 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3305, pruned_loss=0.08881, over 5703606.03 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3424, pruned_loss=0.09548, over 5676160.57 frames. ], batch size: 78, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 07:25:12,078 INFO [train.py:968] (0/2) Epoch 14, batch 17650, giga_loss[loss=0.2139, simple_loss=0.28, pruned_loss=0.07385, over 28558.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3338, pruned_loss=0.0918, over 5679512.57 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3309, pruned_loss=0.08903, over 5702269.37 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3345, pruned_loss=0.09197, over 5683131.04 frames. ], batch size: 85, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:25:21,988 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=610960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:25:24,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=610963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:25:51,119 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=610992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:25:55,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.277e+02 1.049e+03 1.303e+03 1.946e+03 1.422e+04, threshold=2.607e+03, percent-clipped=16.0
+2023-03-07 07:25:58,609 INFO [train.py:968] (0/2) Epoch 14, batch 17700, giga_loss[loss=0.3077, simple_loss=0.3477, pruned_loss=0.1339, over 26549.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3255, pruned_loss=0.08809, over 5676354.05 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3308, pruned_loss=0.08884, over 5702894.33 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3262, pruned_loss=0.08842, over 5678614.41 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:26:24,651 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=611031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:26:26,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=611034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:26:36,967 INFO [train.py:968] (0/2) Epoch 14, batch 17750, giga_loss[loss=0.2238, simple_loss=0.2948, pruned_loss=0.07637, over 28480.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3203, pruned_loss=0.08561, over 5689671.14 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3315, pruned_loss=0.08912, over 5709277.98 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3198, pruned_loss=0.0855, over 5684922.50 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:26:48,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=611063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:27:17,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.704e+02 1.048e+03 1.336e+03 1.863e+03 8.772e+03, threshold=2.673e+03, percent-clipped=7.0
+2023-03-07 07:27:18,492 INFO [train.py:968] (0/2) Epoch 14, batch 17800, libri_loss[loss=0.2376, simple_loss=0.3097, pruned_loss=0.08276, over 29637.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3162, pruned_loss=0.08367, over 5695674.03 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3313, pruned_loss=0.08894, over 5713332.53 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3158, pruned_loss=0.08365, over 5687907.90 frames. ], batch size: 69, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:27:26,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1738, 1.5235, 1.5614, 1.3257], device='cuda:0'), covar=tensor([0.1635, 0.1427, 0.2058, 0.1603], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0729, 0.0676, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 07:27:32,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9587, 1.0603, 1.0843, 0.9336], device='cuda:0'), covar=tensor([0.1741, 0.1901, 0.1139, 0.1585], device='cuda:0'), in_proj_covar=tensor([0.1806, 0.1682, 0.1629, 0.1775], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 07:27:59,841 INFO [train.py:968] (0/2) Epoch 14, batch 17850, libri_loss[loss=0.2322, simple_loss=0.3235, pruned_loss=0.07048, over 29352.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3139, pruned_loss=0.08245, over 5701552.15 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3316, pruned_loss=0.08883, over 5720464.78 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.3127, pruned_loss=0.08232, over 5688218.38 frames. ], batch size: 92, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:28:38,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.208e+02 1.060e+03 1.437e+03 2.042e+03 4.556e+03, threshold=2.873e+03, percent-clipped=11.0
+2023-03-07 07:28:40,596 INFO [train.py:968] (0/2) Epoch 14, batch 17900, giga_loss[loss=0.2153, simple_loss=0.2886, pruned_loss=0.07094, over 28598.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3113, pruned_loss=0.08142, over 5699872.51 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3319, pruned_loss=0.08884, over 5719441.25 frames. ], giga_tot_loss[loss=0.2359, simple_loss=0.3096, pruned_loss=0.08115, over 5689657.43 frames. ], batch size: 92, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:28:50,518 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=611209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:28:59,401 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-07 07:29:22,365 INFO [train.py:968] (0/2) Epoch 14, batch 17950, giga_loss[loss=0.2122, simple_loss=0.2849, pruned_loss=0.06978, over 28600.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3087, pruned_loss=0.08016, over 5695824.42 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3321, pruned_loss=0.08884, over 5715960.59 frames. ], giga_tot_loss[loss=0.2329, simple_loss=0.3065, pruned_loss=0.07968, over 5690473.70 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:30:04,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.501e+02 9.872e+02 1.228e+03 1.520e+03 4.713e+03, threshold=2.455e+03, percent-clipped=6.0
+2023-03-07 07:30:06,020 INFO [train.py:968] (0/2) Epoch 14, batch 18000, giga_loss[loss=0.1851, simple_loss=0.2695, pruned_loss=0.05035, over 28766.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3066, pruned_loss=0.0793, over 5693179.48 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3324, pruned_loss=0.08899, over 5717809.24 frames. ], giga_tot_loss[loss=0.2306, simple_loss=0.304, pruned_loss=0.07859, over 5686392.98 frames. ], batch size: 243, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:30:06,023 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 07:30:14,509 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2099, simple_loss=0.3147, pruned_loss=0.05255, over 944034.00 frames. 
+2023-03-07 07:30:14,510 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 07:30:37,883 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 07:30:55,346 INFO [train.py:968] (0/2) Epoch 14, batch 18050, giga_loss[loss=0.2202, simple_loss=0.2932, pruned_loss=0.07356, over 28612.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3049, pruned_loss=0.07854, over 5675276.44 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3328, pruned_loss=0.08927, over 5697466.26 frames. ], giga_tot_loss[loss=0.2279, simple_loss=0.3012, pruned_loss=0.07728, over 5687164.96 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:31:08,047 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 07:31:32,894 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=611389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:31:40,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.785e+02 1.033e+03 1.382e+03 1.939e+03 6.736e+03, threshold=2.763e+03, percent-clipped=11.0
+2023-03-07 07:31:42,090 INFO [train.py:968] (0/2) Epoch 14, batch 18100, giga_loss[loss=0.2146, simple_loss=0.2958, pruned_loss=0.06666, over 29088.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3013, pruned_loss=0.07694, over 5680748.45 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3333, pruned_loss=0.08937, over 5700390.67 frames. ], giga_tot_loss[loss=0.2245, simple_loss=0.2976, pruned_loss=0.07569, over 5687171.11 frames. ], batch size: 155, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:32:31,050 INFO [train.py:968] (0/2) Epoch 14, batch 18150, giga_loss[loss=0.192, simple_loss=0.2635, pruned_loss=0.06019, over 28443.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.2983, pruned_loss=0.07578, over 5667731.32 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3335, pruned_loss=0.08938, over 5699479.74 frames. ], giga_tot_loss[loss=0.2218, simple_loss=0.2946, pruned_loss=0.07456, over 5673319.45 frames. ], batch size: 78, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:32:51,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3358, 1.3774, 1.2714, 1.2549], device='cuda:0'), covar=tensor([0.1878, 0.1694, 0.1479, 0.1623], device='cuda:0'), in_proj_covar=tensor([0.1806, 0.1686, 0.1630, 0.1777], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 07:33:13,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.548e+02 9.220e+02 1.174e+03 1.872e+03 4.395e+03, threshold=2.348e+03, percent-clipped=5.0
+2023-03-07 07:33:15,127 INFO [train.py:968] (0/2) Epoch 14, batch 18200, giga_loss[loss=0.2521, simple_loss=0.3266, pruned_loss=0.08878, over 28786.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.2976, pruned_loss=0.07577, over 5660724.64 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3337, pruned_loss=0.08944, over 5690307.89 frames. ], giga_tot_loss[loss=0.2214, simple_loss=0.2938, pruned_loss=0.07451, over 5672464.66 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:34:06,174 INFO [train.py:968] (0/2) Epoch 14, batch 18250, giga_loss[loss=0.3599, simple_loss=0.4131, pruned_loss=0.1533, over 28683.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3094, pruned_loss=0.08212, over 5665623.24 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3335, pruned_loss=0.08923, over 5692753.44 frames. ], giga_tot_loss[loss=0.2342, simple_loss=0.3062, pruned_loss=0.08114, over 5672400.23 frames. ], batch size: 242, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:34:38,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=611584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:34:41,311 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 07:34:47,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.241e+02 1.282e+03 1.572e+03 2.277e+03 7.240e+03, threshold=3.144e+03, percent-clipped=23.0
+2023-03-07 07:34:49,247 INFO [train.py:968] (0/2) Epoch 14, batch 18300, libri_loss[loss=0.3152, simple_loss=0.3761, pruned_loss=0.1272, over 29493.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3233, pruned_loss=0.08935, over 5676524.91 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3342, pruned_loss=0.08959, over 5698196.47 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3195, pruned_loss=0.08811, over 5676239.37 frames. ], batch size: 85, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:35:32,034 INFO [train.py:968] (0/2) Epoch 14, batch 18350, giga_loss[loss=0.2827, simple_loss=0.3412, pruned_loss=0.1121, over 23688.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3338, pruned_loss=0.09472, over 5681378.63 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3341, pruned_loss=0.08942, over 5702456.61 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3308, pruned_loss=0.09397, over 5676865.94 frames. ], batch size: 705, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:35:59,467 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=611685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:36:10,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.410e+02 1.187e+03 1.472e+03 1.866e+03 4.046e+03, threshold=2.944e+03, percent-clipped=1.0
+2023-03-07 07:36:11,733 INFO [train.py:968] (0/2) Epoch 14, batch 18400, giga_loss[loss=0.2745, simple_loss=0.3574, pruned_loss=0.09575, over 28930.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3399, pruned_loss=0.09687, over 5676776.25 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3341, pruned_loss=0.0893, over 5698957.92 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3377, pruned_loss=0.09662, over 5675901.22 frames. ], batch size: 227, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 07:36:32,577 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=611727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:36:34,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=611730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:36:50,884 INFO [train.py:968] (0/2) Epoch 14, batch 18450, giga_loss[loss=0.241, simple_loss=0.3326, pruned_loss=0.07472, over 28743.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.342, pruned_loss=0.09636, over 5689154.93 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3344, pruned_loss=0.08927, over 5703771.56 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3402, pruned_loss=0.09646, over 5683077.76 frames. ], batch size: 262, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 07:36:58,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=611759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:37:03,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=611764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:37:37,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.626e+02 1.113e+03 1.344e+03 1.841e+03 4.781e+03, threshold=2.688e+03, percent-clipped=6.0
+2023-03-07 07:37:38,673 INFO [train.py:968] (0/2) Epoch 14, batch 18500, giga_loss[loss=0.293, simple_loss=0.3699, pruned_loss=0.108, over 28675.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3434, pruned_loss=0.09638, over 5673936.42 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3343, pruned_loss=0.08921, over 5707047.16 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3422, pruned_loss=0.09661, over 5665962.00 frames. ], batch size: 242, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:38:04,970 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-07 07:38:25,229 INFO [train.py:968] (0/2) Epoch 14, batch 18550, giga_loss[loss=0.2604, simple_loss=0.3316, pruned_loss=0.09464, over 28650.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3458, pruned_loss=0.09844, over 5676226.51 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3342, pruned_loss=0.08913, over 5708415.13 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3449, pruned_loss=0.09875, over 5668604.96 frames. ], batch size: 92, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:38:46,255 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-07 07:39:07,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.387e+02 1.225e+03 1.514e+03 2.091e+03 6.120e+03, threshold=3.028e+03, percent-clipped=18.0
+2023-03-07 07:39:07,936 INFO [train.py:968] (0/2) Epoch 14, batch 18600, giga_loss[loss=0.2847, simple_loss=0.3646, pruned_loss=0.1024, over 28727.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3491, pruned_loss=0.1008, over 5672810.07 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.335, pruned_loss=0.0893, over 5705645.58 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3482, pruned_loss=0.1012, over 5667339.36 frames. ], batch size: 284, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:39:12,938 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=611907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:39:16,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=611910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:39:39,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6471, 1.9232, 1.7929, 1.4490], device='cuda:0'), covar=tensor([0.2694, 0.2246, 0.2039, 0.2495], device='cuda:0'), in_proj_covar=tensor([0.1785, 0.1678, 0.1622, 0.1767], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 07:39:39,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 07:39:41,608 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=611939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:39:48,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3934, 1.6005, 1.3950, 1.2956], device='cuda:0'), covar=tensor([0.2099, 0.1905, 0.2008, 0.2097], device='cuda:0'), in_proj_covar=tensor([0.1784, 0.1679, 0.1623, 0.1767], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 07:39:49,151 INFO [train.py:968] (0/2) Epoch 14, batch 18650, giga_loss[loss=0.3302, simple_loss=0.3826, pruned_loss=0.1389, over 26467.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3521, pruned_loss=0.102, over 5674768.03 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3353, pruned_loss=0.0893, over 5706907.25 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3515, pruned_loss=0.1027, over 5668480.75 frames. ], batch size: 555, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:40:30,007 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.272e+02 1.160e+03 1.360e+03 1.848e+03 4.259e+03, threshold=2.721e+03, percent-clipped=5.0
+2023-03-07 07:40:30,088 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-612000.pt
+2023-03-07 07:40:30,374 INFO [train.py:968] (0/2) Epoch 14, batch 18700, giga_loss[loss=0.2668, simple_loss=0.3481, pruned_loss=0.09271, over 28865.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3549, pruned_loss=0.1027, over 5683358.38 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3357, pruned_loss=0.08945, over 5709847.66 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3546, pruned_loss=0.1035, over 5675315.30 frames. ], batch size: 199, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:40:48,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6824, 1.8720, 1.4901, 1.9418], device='cuda:0'), covar=tensor([0.2556, 0.2648, 0.2804, 0.2428], device='cuda:0'), in_proj_covar=tensor([0.1357, 0.1002, 0.1201, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 07:41:09,944 INFO [train.py:968] (0/2) Epoch 14, batch 18750, libri_loss[loss=0.2239, simple_loss=0.3087, pruned_loss=0.06953, over 29605.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3569, pruned_loss=0.1032, over 5680363.29 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3365, pruned_loss=0.08963, over 5710275.09 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3568, pruned_loss=0.1042, over 5672396.07 frames. ], batch size: 74, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:41:17,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=612060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:41:41,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4091, 1.6009, 1.6434, 1.2371], device='cuda:0'), covar=tensor([0.1693, 0.2404, 0.1406, 0.1642], device='cuda:0'), in_proj_covar=tensor([0.0850, 0.0682, 0.0891, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 07:41:48,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.868e+02 1.162e+03 1.454e+03 1.982e+03 4.582e+03, threshold=2.908e+03, percent-clipped=9.0
+2023-03-07 07:41:48,902 INFO [train.py:968] (0/2) Epoch 14, batch 18800, giga_loss[loss=0.2894, simple_loss=0.3627, pruned_loss=0.108, over 28642.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3579, pruned_loss=0.103, over 5694406.80 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3369, pruned_loss=0.08998, over 5714083.13 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3579, pruned_loss=0.1039, over 5684025.63 frames. ], batch size: 85, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:42:29,874 INFO [train.py:968] (0/2) Epoch 14, batch 18850, giga_loss[loss=0.2831, simple_loss=0.3596, pruned_loss=0.1033, over 28610.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3579, pruned_loss=0.1019, over 5701568.81 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3373, pruned_loss=0.09004, over 5718220.14 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3581, pruned_loss=0.1028, over 5689273.54 frames. ], batch size: 85, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:42:31,331 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:43:11,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.300e+02 1.008e+03 1.328e+03 1.641e+03 3.913e+03, threshold=2.655e+03, percent-clipped=8.0
+2023-03-07 07:43:11,287 INFO [train.py:968] (0/2) Epoch 14, batch 18900, giga_loss[loss=0.2502, simple_loss=0.3399, pruned_loss=0.0802, over 29026.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3558, pruned_loss=0.09929, over 5707617.64 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3376, pruned_loss=0.09028, over 5717714.36 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3558, pruned_loss=0.09989, over 5697888.51 frames. ], batch size: 136, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:43:13,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=612203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:43:15,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=612206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:43:22,332 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:43:39,547 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=612235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:43:42,364 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4365, 1.6445, 1.3473, 1.5473], device='cuda:0'), covar=tensor([0.2543, 0.2488, 0.2739, 0.2178], device='cuda:0'), in_proj_covar=tensor([0.1358, 0.1001, 0.1200, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 07:43:43,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 07:43:50,550 INFO [train.py:968] (0/2) Epoch 14, batch 18950, giga_loss[loss=0.3061, simple_loss=0.372, pruned_loss=0.1201, over 28951.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3546, pruned_loss=0.09869, over 5702345.04 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3379, pruned_loss=0.09035, over 5713321.77 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3548, pruned_loss=0.09933, over 5698309.92 frames. ], batch size: 136, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:44:16,199 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0835, 1.4545, 1.5426, 1.2161], device='cuda:0'), covar=tensor([0.1786, 0.1505, 0.2003, 0.1680], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0727, 0.0676, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 07:44:27,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0853, 2.3512, 2.0736, 1.7761], device='cuda:0'), covar=tensor([0.2284, 0.1734, 0.1755, 0.2065], device='cuda:0'), in_proj_covar=tensor([0.1784, 0.1682, 0.1623, 0.1770], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 07:44:32,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.495e+02 1.161e+03 1.461e+03 2.190e+03 7.088e+03, threshold=2.921e+03, percent-clipped=13.0
+2023-03-07 07:44:32,094 INFO [train.py:968] (0/2) Epoch 14, batch 19000, giga_loss[loss=0.2811, simple_loss=0.3539, pruned_loss=0.1041, over 28867.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3564, pruned_loss=0.1018, over 5702054.49 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3379, pruned_loss=0.09039, over 5707562.88 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3569, pruned_loss=0.1024, over 5703484.62 frames. ], batch size: 213, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:45:21,461 INFO [train.py:968] (0/2) Epoch 14, batch 19050, giga_loss[loss=0.2895, simple_loss=0.3619, pruned_loss=0.1086, over 28924.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3601, pruned_loss=0.1076, over 5701830.00 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3382, pruned_loss=0.09055, over 5706077.05 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3604, pruned_loss=0.1081, over 5704221.70 frames. ], batch size: 227, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:45:59,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.605e+02 1.266e+03 1.563e+03 2.032e+03 5.569e+03, threshold=3.125e+03, percent-clipped=7.0
+2023-03-07 07:45:59,209 INFO [train.py:968] (0/2) Epoch 14, batch 19100, giga_loss[loss=0.2749, simple_loss=0.3556, pruned_loss=0.0971, over 28385.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3582, pruned_loss=0.1069, over 5708179.22 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3382, pruned_loss=0.09034, over 5710629.02 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3591, pruned_loss=0.108, over 5705766.91 frames. ], batch size: 65, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:46:26,604 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612435.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 07:46:40,628 INFO [train.py:968] (0/2) Epoch 14, batch 19150, giga_loss[loss=0.2515, simple_loss=0.3323, pruned_loss=0.0854, over 28937.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3563, pruned_loss=0.107, over 5695375.96 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3385, pruned_loss=0.09056, over 5705864.09 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3573, pruned_loss=0.1081, over 5696951.30 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:47:22,376 INFO [train.py:968] (0/2) Epoch 14, batch 19200, giga_loss[loss=0.2966, simple_loss=0.3649, pruned_loss=0.1142, over 28982.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3544, pruned_loss=0.1058, over 5693431.43 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3387, pruned_loss=0.09061, over 5699017.06 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3554, pruned_loss=0.107, over 5700517.29 frames. ], batch size: 136, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:47:23,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.546e+02 1.245e+03 1.764e+03 2.425e+03 6.654e+03, threshold=3.527e+03, percent-clipped=8.0
+2023-03-07 07:47:50,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=612527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:47:58,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:48:05,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1470, 1.2232, 3.8200, 3.1604], device='cuda:0'), covar=tensor([0.1748, 0.2781, 0.0418, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0688, 0.0605, 0.0884, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 07:48:06,281 INFO [train.py:968] (0/2) Epoch 14, batch 19250, giga_loss[loss=0.2666, simple_loss=0.3541, pruned_loss=0.08959, over 28873.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3543, pruned_loss=0.1048, over 5703002.39 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3391, pruned_loss=0.09081, over 5701466.94 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.355, pruned_loss=0.1059, over 5706358.28 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:48:40,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=612589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:48:49,788 INFO [train.py:968] (0/2) Epoch 14, batch 19300, giga_loss[loss=0.2683, simple_loss=0.3423, pruned_loss=0.09717, over 28819.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3523, pruned_loss=0.1031, over 5699412.75 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3393, pruned_loss=0.09092, over 5705873.92 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3531, pruned_loss=0.1042, over 5698091.74 frames. ], batch size: 99, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:48:52,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.547e+02 1.149e+03 1.519e+03 2.054e+03 7.838e+03, threshold=3.039e+03, percent-clipped=5.0
+2023-03-07 07:49:34,451 INFO [train.py:968] (0/2) Epoch 14, batch 19350, giga_loss[loss=0.2313, simple_loss=0.3086, pruned_loss=0.07699, over 28700.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3478, pruned_loss=0.1002, over 5694164.31 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3399, pruned_loss=0.09111, over 5708390.46 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.348, pruned_loss=0.1011, over 5690492.96 frames. ], batch size: 262, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:49:49,466 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-07 07:49:54,618 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=612670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:49:55,457 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-07 07:49:57,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=612673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:50:01,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6022, 1.7417, 1.4997, 1.8501], device='cuda:0'), covar=tensor([0.2219, 0.2166, 0.2202, 0.2089], device='cuda:0'), in_proj_covar=tensor([0.1356, 0.1000, 0.1197, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 07:50:20,404 INFO [train.py:968] (0/2) Epoch 14, batch 19400, giga_loss[loss=0.2459, simple_loss=0.3024, pruned_loss=0.09466, over 23421.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3427, pruned_loss=0.09801, over 5683542.89 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3405, pruned_loss=0.0914, over 5712984.12 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3424, pruned_loss=0.09863, over 5675845.27 frames. ], batch size: 705, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:50:21,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.916e+02 9.388e+02 1.184e+03 1.546e+03 5.512e+03, threshold=2.367e+03, percent-clipped=3.0
+2023-03-07 07:50:21,945 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=612702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:50:25,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:50:49,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6301, 1.8097, 1.5180, 1.8058], device='cuda:0'), covar=tensor([0.2317, 0.2230, 0.2351, 0.2306], device='cuda:0'), in_proj_covar=tensor([0.1352, 0.0996, 0.1195, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 07:50:49,615 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=612732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:50:53,512 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=612735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:51:09,998 INFO [train.py:968] (0/2) Epoch 14, batch 19450, giga_loss[loss=0.2395, simple_loss=0.313, pruned_loss=0.08305, over 28732.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3367, pruned_loss=0.09519, over 5672438.66 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3406, pruned_loss=0.09147, over 5715065.91 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3364, pruned_loss=0.09567, over 5664499.46 frames. ], batch size: 119, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:51:24,240 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=612764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:51:41,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3271, 1.4455, 3.5115, 3.3163], device='cuda:0'), covar=tensor([0.1318, 0.2434, 0.0391, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0602, 0.0877, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 07:51:48,420 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3403, 1.4561, 1.2510, 1.5645], device='cuda:0'), covar=tensor([0.0788, 0.0335, 0.0339, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 07:51:57,829 INFO [train.py:968] (0/2) Epoch 14, batch 19500, giga_loss[loss=0.2791, simple_loss=0.3528, pruned_loss=0.1027, over 28553.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3355, pruned_loss=0.09425, over 5660402.92 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3409, pruned_loss=0.09156, over 5716503.25 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3349, pruned_loss=0.09459, over 5652215.74 frames. ], batch size: 307, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:51:58,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.508e+02 9.458e+02 1.257e+03 1.583e+03 6.385e+03, threshold=2.515e+03, percent-clipped=7.0
+2023-03-07 07:51:59,563 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:52:07,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=612810.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:52:42,082 INFO [train.py:968] (0/2) Epoch 14, batch 19550, giga_loss[loss=0.2369, simple_loss=0.3192, pruned_loss=0.0773, over 28794.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3363, pruned_loss=0.09474, over 5660840.33 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3414, pruned_loss=0.09177, over 5708865.36 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3353, pruned_loss=0.09487, over 5659568.31 frames. ], batch size: 119, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:53:25,200 INFO [train.py:968] (0/2) Epoch 14, batch 19600, giga_loss[loss=0.2606, simple_loss=0.3317, pruned_loss=0.09474, over 29044.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3353, pruned_loss=0.09425, over 5666439.98 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3413, pruned_loss=0.09164, over 5708119.26 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3346, pruned_loss=0.09451, over 5665247.80 frames. ], batch size: 155, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 07:53:25,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.393e+02 1.050e+03 1.449e+03 2.283e+03 5.197e+03, threshold=2.898e+03, percent-clipped=17.0
+2023-03-07 07:53:35,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=612913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:53:38,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8280, 0.9849, 2.9887, 2.8503], device='cuda:0'), covar=tensor([0.1559, 0.2403, 0.0486, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0679, 0.0598, 0.0868, 0.0788], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:0')
+2023-03-07 07:54:05,535 INFO [train.py:968] (0/2) Epoch 14, batch 19650, giga_loss[loss=0.2532, simple_loss=0.3246, pruned_loss=0.09085, over 28880.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3338, pruned_loss=0.0933, over 5680011.53 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3414, pruned_loss=0.09157, over 5712802.05 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3329, pruned_loss=0.09362, over 5674060.01 frames. ], batch size: 112, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 07:54:08,348 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=612953.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 07:54:11,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=612956.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:54:14,648 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:54:33,163 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=612985.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:54:37,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:54:44,059 INFO [train.py:968] (0/2) Epoch 14, batch 19700, giga_loss[loss=0.2308, simple_loss=0.3124, pruned_loss=0.07458, over 28953.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3321, pruned_loss=0.09252, over 5686962.86 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3422, pruned_loss=0.09195, over 5714780.68 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3306, pruned_loss=0.09247, over 5679853.24 frames. ], batch size: 164, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:54:45,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.605e+02 9.873e+02 1.270e+03 1.684e+03 5.003e+03, threshold=2.540e+03, percent-clipped=6.0
+2023-03-07 07:55:27,281 INFO [train.py:968] (0/2) Epoch 14, batch 19750, giga_loss[loss=0.2494, simple_loss=0.322, pruned_loss=0.08837, over 29017.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3295, pruned_loss=0.09153, over 5695583.68 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3428, pruned_loss=0.09236, over 5715951.78 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3278, pruned_loss=0.09115, over 5688941.43 frames. ], batch size: 164, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:55:32,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=613056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:55:35,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=613059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:55:54,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:55:58,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=613088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:56:08,603 INFO [train.py:968] (0/2) Epoch 14, batch 19800, libri_loss[loss=0.3443, simple_loss=0.4132, pruned_loss=0.1377, over 19190.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3276, pruned_loss=0.09067, over 5692883.62 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3435, pruned_loss=0.09269, over 5708579.97 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3254, pruned_loss=0.09006, over 5695082.11 frames. ], batch size: 186, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:56:10,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.975e+02 1.031e+03 1.190e+03 1.512e+03 8.743e+03, threshold=2.379e+03, percent-clipped=3.0
+2023-03-07 07:56:24,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4357, 1.5242, 1.1364, 1.1573], device='cuda:0'), covar=tensor([0.0902, 0.0555, 0.1090, 0.1094], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0438, 0.0505, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 07:56:46,340 INFO [train.py:968] (0/2) Epoch 14, batch 19850, giga_loss[loss=0.2073, simple_loss=0.2902, pruned_loss=0.06222, over 29025.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3253, pruned_loss=0.08943, over 5706143.65 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3445, pruned_loss=0.09304, over 5713907.34 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3221, pruned_loss=0.08855, over 5702619.22 frames. ], batch size: 213, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:57:07,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:57:25,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2181, 1.3001, 1.0934, 1.1729], device='cuda:0'), covar=tensor([0.1999, 0.1672, 0.1464, 0.1745], device='cuda:0'), in_proj_covar=tensor([0.1772, 0.1665, 0.1618, 0.1758], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 07:57:27,507 INFO [train.py:968] (0/2) Epoch 14, batch 19900, libri_loss[loss=0.2438, simple_loss=0.3276, pruned_loss=0.08002, over 28106.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3232, pruned_loss=0.08812, over 5702796.46 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3443, pruned_loss=0.09286, over 5705836.33 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3204, pruned_loss=0.08749, over 5706762.38 frames. ], batch size: 62, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:57:29,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.446e+02 1.085e+03 1.343e+03 1.887e+03 9.760e+03, threshold=2.686e+03, percent-clipped=17.0
+2023-03-07 07:57:47,257 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=613226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:57:52,259 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=613229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:58:08,177 INFO [train.py:968] (0/2) Epoch 14, batch 19950, giga_loss[loss=0.2465, simple_loss=0.3256, pruned_loss=0.08366, over 28608.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3217, pruned_loss=0.08727, over 5708728.17 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3447, pruned_loss=0.093, over 5711864.14 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3184, pruned_loss=0.08648, over 5706442.31 frames. ], batch size: 336, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:58:13,612 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=613258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:58:47,301 INFO [train.py:968] (0/2) Epoch 14, batch 20000, giga_loss[loss=0.2218, simple_loss=0.2931, pruned_loss=0.07527, over 29002.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3192, pruned_loss=0.08557, over 5713599.65 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3451, pruned_loss=0.09313, over 5713634.47 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3161, pruned_loss=0.08478, over 5710260.67 frames. ], batch size: 106, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:58:49,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.548e+02 9.243e+02 1.186e+03 1.622e+03 4.782e+03, threshold=2.372e+03, percent-clipped=7.0
+2023-03-07 07:59:01,144 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=613319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:59:01,961 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-07 07:59:04,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=613322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:59:16,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:59:25,929 INFO [train.py:968] (0/2) Epoch 14, batch 20050, giga_loss[loss=0.2656, simple_loss=0.3431, pruned_loss=0.09399, over 28707.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3197, pruned_loss=0.08552, over 5720152.34 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3455, pruned_loss=0.09326, over 5719176.04 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3161, pruned_loss=0.08453, over 5712450.21 frames. ], batch size: 307, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:59:27,048 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=613351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:59:39,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:00:08,793 INFO [train.py:968] (0/2) Epoch 14, batch 20100, giga_loss[loss=0.2784, simple_loss=0.3521, pruned_loss=0.1024, over 28355.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3256, pruned_loss=0.08984, over 5718154.91 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3454, pruned_loss=0.09326, over 5721053.59 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3226, pruned_loss=0.08902, over 5710373.16 frames. ], batch size: 368, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:00:10,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.810e+02 1.057e+03 1.300e+03 1.672e+03 3.999e+03, threshold=2.601e+03, percent-clipped=6.0
+2023-03-07 08:00:33,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2913, 3.0919, 2.9114, 1.4714], device='cuda:0'), covar=tensor([0.0864, 0.0968, 0.0963, 0.2205], device='cuda:0'), in_proj_covar=tensor([0.1085, 0.1001, 0.0876, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 08:00:53,810 INFO [train.py:968] (0/2) Epoch 14, batch 20150, giga_loss[loss=0.3151, simple_loss=0.3675, pruned_loss=0.1313, over 23692.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3315, pruned_loss=0.0933, over 5702512.18 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3456, pruned_loss=0.09315, over 5724387.29 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3283, pruned_loss=0.09272, over 5692821.19 frames. ], batch size: 705, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:01:23,357 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=613480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:01:26,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=613483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:01:39,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0267, 3.8193, 3.6164, 2.0660], device='cuda:0'), covar=tensor([0.0582, 0.0771, 0.0734, 0.2027], device='cuda:0'), in_proj_covar=tensor([0.1092, 0.1009, 0.0881, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 08:01:41,315 INFO [train.py:968] (0/2) Epoch 14, batch 20200, giga_loss[loss=0.2746, simple_loss=0.3476, pruned_loss=0.1008, over 28798.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3399, pruned_loss=0.09868, over 5700877.32 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3457, pruned_loss=0.09311, over 5727172.63 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3371, pruned_loss=0.0983, over 5690202.19 frames. ], batch size: 186, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:01:44,347 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.709e+02 1.211e+03 1.628e+03 2.163e+03 6.226e+03, threshold=3.256e+03, percent-clipped=12.0
+2023-03-07 08:01:44,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=613503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:01:50,452 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=613508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:01:52,764 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=613511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:01:53,238 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=613512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:02:19,134 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=613540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:02:28,669 INFO [train.py:968] (0/2) Epoch 14, batch 20250, giga_loss[loss=0.3901, simple_loss=0.4364, pruned_loss=0.1719, over 26506.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3451, pruned_loss=0.1009, over 5698040.82 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3459, pruned_loss=0.09322, over 5727689.90 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3426, pruned_loss=0.1006, over 5688597.03 frames. ], batch size: 555, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:02:49,281 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=613569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:02:52,309 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-07 08:03:14,819 INFO [train.py:968] (0/2) Epoch 14, batch 20300, giga_loss[loss=0.2845, simple_loss=0.3668, pruned_loss=0.1011, over 28988.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3508, pruned_loss=0.1033, over 5693617.05 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3462, pruned_loss=0.0934, over 5727680.76 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3485, pruned_loss=0.1031, over 5685687.14 frames. ], batch size: 164, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:03:17,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.467e+02 1.123e+03 1.392e+03 1.803e+03 5.176e+03, threshold=2.784e+03, percent-clipped=5.0
+2023-03-07 08:04:02,403 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=613647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:04:04,324 INFO [train.py:968] (0/2) Epoch 14, batch 20350, giga_loss[loss=0.3881, simple_loss=0.4241, pruned_loss=0.176, over 26560.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.357, pruned_loss=0.1069, over 5692432.66 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3461, pruned_loss=0.09336, over 5728407.35 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3553, pruned_loss=0.1068, over 5685373.39 frames. ], batch size: 555, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:04:13,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-07 08:04:44,163 INFO [train.py:968] (0/2) Epoch 14, batch 20400, libri_loss[loss=0.2867, simple_loss=0.3668, pruned_loss=0.1034, over 29521.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3566, pruned_loss=0.1062, over 5701635.36 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3462, pruned_loss=0.09349, over 5733486.70 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3555, pruned_loss=0.1064, over 5690546.95 frames. ], batch size: 83, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:04:46,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.508e+02 1.147e+03 1.435e+03 1.970e+03 5.933e+03, threshold=2.869e+03, percent-clipped=9.0
+2023-03-07 08:05:03,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6136, 2.0405, 1.6339, 1.9879], device='cuda:0'), covar=tensor([0.0784, 0.0256, 0.0300, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 08:05:27,662 INFO [train.py:968] (0/2) Epoch 14, batch 20450, giga_loss[loss=0.247, simple_loss=0.3374, pruned_loss=0.07826, over 28964.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3516, pruned_loss=0.1021, over 5703946.72 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3466, pruned_loss=0.09361, over 5738059.17 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3506, pruned_loss=0.1024, over 5690114.88 frames. ], batch size: 164, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:05:46,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=613773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:06:08,976 INFO [train.py:968] (0/2) Epoch 14, batch 20500, giga_loss[loss=0.2525, simple_loss=0.33, pruned_loss=0.08749, over 28896.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3513, pruned_loss=0.1016, over 5700255.82 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3469, pruned_loss=0.09373, over 5737062.61 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3502, pruned_loss=0.1018, over 5689227.53 frames. ], batch size: 112, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:06:11,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.257e+02 1.121e+03 1.392e+03 1.946e+03 6.110e+03, threshold=2.784e+03, percent-clipped=11.0
+2023-03-07 08:06:49,247 INFO [train.py:968] (0/2) Epoch 14, batch 20550, libri_loss[loss=0.2754, simple_loss=0.3574, pruned_loss=0.09669, over 29529.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3501, pruned_loss=0.1, over 5704099.18 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3469, pruned_loss=0.09368, over 5742616.43 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3494, pruned_loss=0.1006, over 5688647.72 frames. ], batch size: 89, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:07:13,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:07:32,539 INFO [train.py:968] (0/2) Epoch 14, batch 20600, giga_loss[loss=0.2787, simple_loss=0.3561, pruned_loss=0.1006, over 28919.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3534, pruned_loss=0.1019, over 5700432.51 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3471, pruned_loss=0.09372, over 5743123.80 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3527, pruned_loss=0.1024, over 5687541.04 frames. ], batch size: 227, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:07:37,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.386e+02 1.212e+03 1.625e+03 2.483e+03 5.825e+03, threshold=3.249e+03, percent-clipped=20.0
+2023-03-07 08:08:12,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:08:18,702 INFO [train.py:968] (0/2) Epoch 14, batch 20650, giga_loss[loss=0.3037, simple_loss=0.3761, pruned_loss=0.1156, over 28597.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3571, pruned_loss=0.1049, over 5697612.65 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3474, pruned_loss=0.09381, over 5746752.67 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3565, pruned_loss=0.1055, over 5682922.39 frames. ], batch size: 336, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:08:28,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8851, 1.0896, 1.0545, 0.7595], device='cuda:0'), covar=tensor([0.1708, 0.2021, 0.1207, 0.1793], device='cuda:0'), in_proj_covar=tensor([0.1776, 0.1678, 0.1636, 0.1761], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 08:08:43,994 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-07 08:09:02,763 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-614000.pt
+2023-03-07 08:09:03,072 INFO [train.py:968] (0/2) Epoch 14, batch 20700, giga_loss[loss=0.2885, simple_loss=0.3582, pruned_loss=0.1094, over 28654.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3581, pruned_loss=0.1057, over 5710176.78 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3484, pruned_loss=0.09432, over 5750118.09 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.357, pruned_loss=0.106, over 5694189.29 frames. ], batch size: 262, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:09:06,474 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.852e+02 1.197e+03 1.472e+03 2.017e+03 3.504e+03, threshold=2.943e+03, percent-clipped=3.0
+2023-03-07 08:09:20,368 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:09:21,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:09:23,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:09:33,832 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-07 08:09:42,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0240, 1.2426, 3.6141, 2.8707], device='cuda:0'), covar=tensor([0.1688, 0.2594, 0.0441, 0.1009], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0598, 0.0870, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 08:09:47,603 INFO [train.py:968] (0/2) Epoch 14, batch 20750, giga_loss[loss=0.3005, simple_loss=0.3702, pruned_loss=0.1155, over 29067.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3589, pruned_loss=0.1065, over 5711213.03 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3486, pruned_loss=0.09444, over 5742470.25 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3579, pruned_loss=0.1067, over 5704368.12 frames. ], batch size: 155, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:09:50,345 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:09:55,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4959, 1.6908, 1.7079, 1.3304], device='cuda:0'), covar=tensor([0.1431, 0.1980, 0.1165, 0.1375], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0688, 0.0894, 0.0796], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 08:10:21,382 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:10:23,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:10:31,269 INFO [train.py:968] (0/2) Epoch 14, batch 20800, giga_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.123, over 28718.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3592, pruned_loss=0.1074, over 5707595.82 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3486, pruned_loss=0.09433, over 5744047.39 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3585, pruned_loss=0.1078, over 5700439.03 frames. ], batch size: 92, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:10:36,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.444e+02 1.222e+03 1.512e+03 2.068e+03 4.088e+03, threshold=3.023e+03, percent-clipped=7.0
+2023-03-07 08:10:45,875 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:08,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:10,092 INFO [train.py:968] (0/2) Epoch 14, batch 20850, giga_loss[loss=0.2697, simple_loss=0.3503, pruned_loss=0.09457, over 28638.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3588, pruned_loss=0.1067, over 5702546.06 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3486, pruned_loss=0.09433, over 5735952.71 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3584, pruned_loss=0.1071, over 5703535.13 frames. ], batch size: 85, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:11:20,254 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=614162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:20,938 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-07 08:11:23,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:25,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:48,298 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:50,144 INFO [train.py:968] (0/2) Epoch 14, batch 20900, giga_loss[loss=0.3028, simple_loss=0.3691, pruned_loss=0.1182, over 27605.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.359, pruned_loss=0.1057, over 5705509.08 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3487, pruned_loss=0.09433, over 5739282.39 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3589, pruned_loss=0.1064, over 5702107.61 frames. ], batch size: 472, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:11:53,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.365e+02 1.090e+03 1.420e+03 2.139e+03 7.289e+03, threshold=2.839e+03, percent-clipped=12.0
+2023-03-07 08:12:12,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3268, 1.4020, 1.2950, 1.4367], device='cuda:0'), covar=tensor([0.0787, 0.0340, 0.0330, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0113, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0055, 0.0095], device='cuda:0')
+2023-03-07 08:12:28,294 INFO [train.py:968] (0/2) Epoch 14, batch 20950, giga_loss[loss=0.3162, simple_loss=0.3893, pruned_loss=0.1216, over 28661.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3595, pruned_loss=0.105, over 5717869.23 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3491, pruned_loss=0.09443, over 5743244.96 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3595, pruned_loss=0.1059, over 5710109.52 frames. ], batch size: 307, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:12:43,254 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8695, 1.9534, 1.4385, 1.4613], device='cuda:0'), covar=tensor([0.0784, 0.0553, 0.0948, 0.1010], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0431, 0.0498, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 08:13:00,481 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:13:02,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:13:06,033 INFO [train.py:968] (0/2) Epoch 14, batch 21000, giga_loss[loss=0.286, simple_loss=0.3557, pruned_loss=0.1081, over 29002.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3571, pruned_loss=0.1034, over 5725683.01 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3489, pruned_loss=0.09427, over 5747513.59 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3575, pruned_loss=0.1046, over 5714896.37 frames. ], batch size: 136, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:13:06,040 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 08:13:14,937 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2131, simple_loss=0.3189, pruned_loss=0.05359, over 944034.00 frames. 
+2023-03-07 08:13:14,938 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 08:13:18,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.933e+02 9.900e+02 1.304e+03 1.658e+03 3.276e+03, threshold=2.608e+03, percent-clipped=3.0
+2023-03-07 08:13:33,673 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:13:43,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0540, 1.3363, 1.2813, 1.1975], device='cuda:0'), covar=tensor([0.1468, 0.1319, 0.1742, 0.1369], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0730, 0.0682, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 08:13:45,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=614340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:13:52,289 INFO [train.py:968] (0/2) Epoch 14, batch 21050, giga_loss[loss=0.3008, simple_loss=0.358, pruned_loss=0.1218, over 28699.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3553, pruned_loss=0.1032, over 5719265.04 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.349, pruned_loss=0.09435, over 5749814.97 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3557, pruned_loss=0.1043, over 5707985.40 frames. ], batch size: 92, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:14:32,426 INFO [train.py:968] (0/2) Epoch 14, batch 21100, giga_loss[loss=0.2695, simple_loss=0.3417, pruned_loss=0.09868, over 28944.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.353, pruned_loss=0.1019, over 5717687.99 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3487, pruned_loss=0.09419, over 5752799.23 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3537, pruned_loss=0.1031, over 5705493.91 frames. ], batch size: 186, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:14:35,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.219e+02 1.018e+03 1.224e+03 1.744e+03 5.572e+03, threshold=2.448e+03, percent-clipped=8.0
+2023-03-07 08:15:08,207 INFO [train.py:968] (0/2) Epoch 14, batch 21150, giga_loss[loss=0.2786, simple_loss=0.3498, pruned_loss=0.1037, over 28900.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3528, pruned_loss=0.1022, over 5706693.23 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3493, pruned_loss=0.09465, over 5735501.53 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3531, pruned_loss=0.103, over 5711077.16 frames. ], batch size: 86, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:15:10,570 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=614453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:15:37,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0269, 1.0496, 3.7383, 3.1837], device='cuda:0'), covar=tensor([0.1836, 0.2964, 0.0424, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.0688, 0.0604, 0.0877, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 08:15:43,642 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=614494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:15:48,229 INFO [train.py:968] (0/2) Epoch 14, batch 21200, giga_loss[loss=0.2963, simple_loss=0.3672, pruned_loss=0.1127, over 29044.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3536, pruned_loss=0.1026, over 5704289.49 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3499, pruned_loss=0.0951, over 5735122.94 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3533, pruned_loss=0.103, over 5707466.72 frames. ], batch size: 155, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:15:54,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.062e+02 9.661e+02 1.298e+03 1.918e+03 6.519e+03, threshold=2.596e+03, percent-clipped=12.0
+2023-03-07 08:16:20,793 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:16:30,622 INFO [train.py:968] (0/2) Epoch 14, batch 21250, giga_loss[loss=0.2582, simple_loss=0.341, pruned_loss=0.08772, over 29057.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3529, pruned_loss=0.1018, over 5708243.14 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3502, pruned_loss=0.09524, over 5735960.33 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3525, pruned_loss=0.1021, over 5709865.55 frames. ], batch size: 155, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:16:42,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5618, 1.7065, 1.8450, 1.3687], device='cuda:0'), covar=tensor([0.1939, 0.2624, 0.1580, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.0855, 0.0693, 0.0900, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 08:17:11,505 INFO [train.py:968] (0/2) Epoch 14, batch 21300, giga_loss[loss=0.2543, simple_loss=0.3375, pruned_loss=0.08553, over 28852.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.352, pruned_loss=0.1006, over 5716848.27 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3502, pruned_loss=0.09541, over 5741222.59 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3518, pruned_loss=0.1008, over 5712543.49 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:17:16,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.762e+02 9.243e+02 1.114e+03 1.490e+03 4.561e+03, threshold=2.228e+03, percent-clipped=7.0
+2023-03-07 08:17:54,904 INFO [train.py:968] (0/2) Epoch 14, batch 21350, giga_loss[loss=0.2849, simple_loss=0.3533, pruned_loss=0.1082, over 28957.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3519, pruned_loss=0.1011, over 5706868.39 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3505, pruned_loss=0.09559, over 5742530.25 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1012, over 5702006.86 frames. ], batch size: 227, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:18:18,511 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:18:20,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:18:34,825 INFO [train.py:968] (0/2) Epoch 14, batch 21400, giga_loss[loss=0.258, simple_loss=0.3354, pruned_loss=0.09035, over 28856.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3512, pruned_loss=0.1012, over 5704965.83 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3511, pruned_loss=0.09615, over 5746381.89 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3503, pruned_loss=0.101, over 5695259.15 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:18:40,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.505e+02 9.915e+02 1.314e+03 1.716e+03 6.182e+03, threshold=2.629e+03, percent-clipped=13.0
+2023-03-07 08:18:44,685 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:18:46,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:19:14,032 INFO [train.py:968] (0/2) Epoch 14, batch 21450, giga_loss[loss=0.2362, simple_loss=0.3212, pruned_loss=0.07562, over 28761.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3493, pruned_loss=0.1005, over 5712181.88 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3515, pruned_loss=0.09662, over 5750600.47 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3481, pruned_loss=0.1, over 5699122.71 frames. ], batch size: 66, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:19:39,664 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-07 08:19:52,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1995, 3.9732, 3.7695, 1.7447], device='cuda:0'), covar=tensor([0.0560, 0.0757, 0.0700, 0.2225], device='cuda:0'), in_proj_covar=tensor([0.1082, 0.1002, 0.0871, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 08:19:55,362 INFO [train.py:968] (0/2) Epoch 14, batch 21500, giga_loss[loss=0.2446, simple_loss=0.317, pruned_loss=0.08605, over 28771.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3461, pruned_loss=0.09888, over 5709008.70 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3519, pruned_loss=0.09689, over 5752800.01 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3447, pruned_loss=0.09838, over 5694928.59 frames. ], batch size: 99, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:19:58,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2613, 1.4521, 1.3642, 1.3627], device='cuda:0'), covar=tensor([0.1491, 0.1524, 0.2014, 0.1532], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0732, 0.0682, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 08:20:00,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.326e+02 1.126e+03 1.450e+03 2.043e+03 6.519e+03, threshold=2.900e+03, percent-clipped=9.0
+2023-03-07 08:20:04,428 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 08:20:18,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:20:18,242 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-07 08:20:33,014 INFO [train.py:968] (0/2) Epoch 14, batch 21550, giga_loss[loss=0.2521, simple_loss=0.3273, pruned_loss=0.0885, over 28914.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3459, pruned_loss=0.0989, over 5706010.42 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.352, pruned_loss=0.09723, over 5752274.13 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3444, pruned_loss=0.09825, over 5693859.33 frames. ], batch size: 213, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:20:39,223 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:20:41,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:20:47,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:21:03,449 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:21:10,066 INFO [train.py:968] (0/2) Epoch 14, batch 21600, giga_loss[loss=0.2438, simple_loss=0.3193, pruned_loss=0.08418, over 28258.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3461, pruned_loss=0.09962, over 5700566.89 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3527, pruned_loss=0.09784, over 5746138.60 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3441, pruned_loss=0.09863, over 5694023.63 frames. ], batch size: 77, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:21:15,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.147e+02 1.200e+03 1.457e+03 2.099e+03 7.615e+03, threshold=2.914e+03, percent-clipped=15.0
+2023-03-07 08:21:54,380 INFO [train.py:968] (0/2) Epoch 14, batch 21650, giga_loss[loss=0.2665, simple_loss=0.3318, pruned_loss=0.1006, over 28712.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3434, pruned_loss=0.09869, over 5702177.92 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3528, pruned_loss=0.098, over 5748657.03 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3416, pruned_loss=0.09778, over 5694136.47 frames. ], batch size: 92, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:22:02,835 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-07 08:22:11,058 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:22:13,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:22:32,700 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-07 08:22:33,016 INFO [train.py:968] (0/2) Epoch 14, batch 21700, giga_loss[loss=0.2809, simple_loss=0.3501, pruned_loss=0.1058, over 28510.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3412, pruned_loss=0.09778, over 5707219.61 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3534, pruned_loss=0.09847, over 5752016.05 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.339, pruned_loss=0.09665, over 5696790.70 frames. ], batch size: 336, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:22:35,072 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=615003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:22:39,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.896e+02 1.043e+03 1.264e+03 1.722e+03 1.311e+04, threshold=2.528e+03, percent-clipped=12.0
+2023-03-07 08:22:41,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=615012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:22:44,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=615015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:22:57,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2294, 0.8134, 0.8519, 1.3846], device='cuda:0'), covar=tensor([0.0720, 0.0361, 0.0354, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0113, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0095], device='cuda:0')
+2023-03-07 08:23:08,031 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=615044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:23:13,500 INFO [train.py:968] (0/2) Epoch 14, batch 21750, giga_loss[loss=0.2368, simple_loss=0.3161, pruned_loss=0.07877, over 28922.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3378, pruned_loss=0.09606, over 5718028.03 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3528, pruned_loss=0.0983, over 5755157.87 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3364, pruned_loss=0.09529, over 5706426.89 frames. ], batch size: 199, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:23:52,609 INFO [train.py:968] (0/2) Epoch 14, batch 21800, giga_loss[loss=0.2618, simple_loss=0.3338, pruned_loss=0.0949, over 28710.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3368, pruned_loss=0.09557, over 5717633.42 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.353, pruned_loss=0.09844, over 5756674.49 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3354, pruned_loss=0.09483, over 5706891.52 frames. ], batch size: 92, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:23:56,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 08:23:59,375 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.793e+02 1.043e+03 1.346e+03 1.927e+03 6.242e+03, threshold=2.692e+03, percent-clipped=10.0
+2023-03-07 08:24:32,752 INFO [train.py:968] (0/2) Epoch 14, batch 21850, libri_loss[loss=0.2381, simple_loss=0.3139, pruned_loss=0.08118, over 29349.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.34, pruned_loss=0.0972, over 5722194.41 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3537, pruned_loss=0.09926, over 5761319.87 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3375, pruned_loss=0.09573, over 5707428.24 frames. ], batch size: 71, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:25:13,763 INFO [train.py:968] (0/2) Epoch 14, batch 21900, giga_loss[loss=0.2758, simple_loss=0.3521, pruned_loss=0.09973, over 28892.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3436, pruned_loss=0.099, over 5714388.49 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3537, pruned_loss=0.09953, over 5763899.85 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3412, pruned_loss=0.09756, over 5698968.48 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:25:20,319 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.054e+02 1.083e+03 1.297e+03 1.776e+03 6.495e+03, threshold=2.595e+03, percent-clipped=7.0
+2023-03-07 08:25:35,075 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=615223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:25:58,082 INFO [train.py:968] (0/2) Epoch 14, batch 21950, giga_loss[loss=0.2562, simple_loss=0.3425, pruned_loss=0.08501, over 28783.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3464, pruned_loss=0.09983, over 5706419.12 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3542, pruned_loss=0.09997, over 5763996.14 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.344, pruned_loss=0.09828, over 5693122.65 frames. ], batch size: 243, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:26:38,778 INFO [train.py:968] (0/2) Epoch 14, batch 22000, giga_loss[loss=0.3054, simple_loss=0.3783, pruned_loss=0.1163, over 28814.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3474, pruned_loss=0.09959, over 5699799.81 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3544, pruned_loss=0.1002, over 5749665.41 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3451, pruned_loss=0.09807, over 5700733.07 frames. ], batch size: 285, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:26:44,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.955e+02 1.027e+03 1.383e+03 1.966e+03 9.843e+03, threshold=2.767e+03, percent-clipped=16.0
+2023-03-07 08:26:51,643 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-07 08:27:17,792 INFO [train.py:968] (0/2) Epoch 14, batch 22050, giga_loss[loss=0.2364, simple_loss=0.3213, pruned_loss=0.07581, over 27982.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3469, pruned_loss=0.09895, over 5701597.29 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3548, pruned_loss=0.1009, over 5747693.96 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3443, pruned_loss=0.0971, over 5701646.21 frames. ], batch size: 412, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:27:19,477 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-07 08:28:02,847 INFO [train.py:968] (0/2) Epoch 14, batch 22100, giga_loss[loss=0.2806, simple_loss=0.3535, pruned_loss=0.1038, over 28585.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3469, pruned_loss=0.09944, over 5692750.38 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3552, pruned_loss=0.1012, over 5748449.49 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3445, pruned_loss=0.09769, over 5692017.36 frames. ], batch size: 307, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:28:09,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.067e+02 1.161e+03 1.418e+03 1.967e+03 4.955e+03, threshold=2.837e+03, percent-clipped=11.0
+2023-03-07 08:28:45,887 INFO [train.py:968] (0/2) Epoch 14, batch 22150, libri_loss[loss=0.3079, simple_loss=0.3769, pruned_loss=0.1195, over 29545.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3469, pruned_loss=0.09979, over 5699540.19 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3553, pruned_loss=0.1014, over 5751058.63 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3448, pruned_loss=0.09824, over 5695833.09 frames. ], batch size: 83, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:28:46,928 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-07 08:29:27,611 INFO [train.py:968] (0/2) Epoch 14, batch 22200, libri_loss[loss=0.289, simple_loss=0.3467, pruned_loss=0.1157, over 29659.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3474, pruned_loss=0.1001, over 5701450.01 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3553, pruned_loss=0.1016, over 5754324.68 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3456, pruned_loss=0.09866, over 5694738.55 frames. ], batch size: 69, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:29:33,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.400e+02 1.131e+03 1.424e+03 2.071e+03 4.902e+03, threshold=2.849e+03, percent-clipped=8.0
+2023-03-07 08:29:54,144 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-07 08:30:08,829 INFO [train.py:968] (0/2) Epoch 14, batch 22250, giga_loss[loss=0.3327, simple_loss=0.3903, pruned_loss=0.1375, over 28542.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3512, pruned_loss=0.102, over 5706491.83 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3556, pruned_loss=0.1018, over 5752625.14 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3494, pruned_loss=0.1007, over 5702057.54 frames. ], batch size: 78, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:30:47,794 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=615598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:30:48,956 INFO [train.py:968] (0/2) Epoch 14, batch 22300, giga_loss[loss=0.2738, simple_loss=0.3533, pruned_loss=0.09712, over 28881.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3531, pruned_loss=0.1027, over 5705303.01 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3564, pruned_loss=0.1023, over 5747454.15 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3509, pruned_loss=0.1012, over 5704658.48 frames. ], batch size: 199, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:30:54,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6125, 1.8463, 1.6397, 1.5287], device='cuda:0'), covar=tensor([0.1493, 0.1727, 0.1914, 0.1867], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0730, 0.0681, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 08:30:54,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.460e+02 1.249e+03 1.615e+03 1.943e+03 5.566e+03, threshold=3.231e+03, percent-clipped=9.0
+2023-03-07 08:31:22,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8958, 3.7033, 3.4639, 1.8419], device='cuda:0'), covar=tensor([0.0650, 0.0786, 0.0756, 0.1934], device='cuda:0'), in_proj_covar=tensor([0.1079, 0.0995, 0.0868, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 08:31:28,698 INFO [train.py:968] (0/2) Epoch 14, batch 22350, giga_loss[loss=0.2943, simple_loss=0.3654, pruned_loss=0.1116, over 28954.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.355, pruned_loss=0.1039, over 5705146.45 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3568, pruned_loss=0.1026, over 5740066.40 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3528, pruned_loss=0.1025, over 5710286.33 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:31:44,446 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0909, 1.9279, 1.4450, 1.7112], device='cuda:0'), covar=tensor([0.0789, 0.0766, 0.1048, 0.1179], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0436, 0.0503, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 08:31:56,827 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6981, 1.7085, 1.2491, 1.4178], device='cuda:0'), covar=tensor([0.0820, 0.0617, 0.1056, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0435, 0.0502, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 08:32:07,115 INFO [train.py:968] (0/2) Epoch 14, batch 22400, giga_loss[loss=0.3107, simple_loss=0.3643, pruned_loss=0.1286, over 23995.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3564, pruned_loss=0.1052, over 5710962.78 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3575, pruned_loss=0.1032, over 5742736.48 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.354, pruned_loss=0.1035, over 5712057.86 frames. ], batch size: 705, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:32:07,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4401, 3.5552, 1.5663, 1.5798], device='cuda:0'), covar=tensor([0.0936, 0.0333, 0.0880, 0.1250], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0518, 0.0351, 0.0400], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 08:32:11,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1943, 1.1793, 3.6714, 3.0823], device='cuda:0'), covar=tensor([0.1659, 0.2811, 0.0448, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0691, 0.0608, 0.0883, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 08:32:14,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.504e+02 1.262e+03 1.557e+03 2.019e+03 5.515e+03, threshold=3.114e+03, percent-clipped=7.0
+2023-03-07 08:32:31,542 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 08:32:42,910 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=615741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:32:44,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=615744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:32:49,656 INFO [train.py:968] (0/2) Epoch 14, batch 22450, giga_loss[loss=0.2735, simple_loss=0.3531, pruned_loss=0.0969, over 28758.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3565, pruned_loss=0.1051, over 5711619.44 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3578, pruned_loss=0.1033, over 5745247.00 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3543, pruned_loss=0.1037, over 5709581.23 frames. ], batch size: 284, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:33:08,939 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=615773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:33:15,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3643, 1.6195, 1.4681, 1.4742], device='cuda:0'), covar=tensor([0.1576, 0.1872, 0.2110, 0.1818], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0730, 0.0683, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 08:33:30,427 INFO [train.py:968] (0/2) Epoch 14, batch 22500, giga_loss[loss=0.2605, simple_loss=0.3367, pruned_loss=0.09217, over 28684.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3546, pruned_loss=0.1041, over 5715920.22 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3583, pruned_loss=0.1038, over 5749144.68 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3524, pruned_loss=0.1025, over 5709919.02 frames. ], batch size: 307, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:33:35,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.311e+02 1.092e+03 1.437e+03 2.089e+03 4.621e+03, threshold=2.873e+03, percent-clipped=7.0
+2023-03-07 08:33:51,003 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=615826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:34:03,611 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-07 08:34:11,966 INFO [train.py:968] (0/2) Epoch 14, batch 22550, giga_loss[loss=0.2566, simple_loss=0.329, pruned_loss=0.09211, over 28652.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.351, pruned_loss=0.1022, over 5721478.73 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3582, pruned_loss=0.104, over 5750391.59 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3492, pruned_loss=0.1008, over 5715293.08 frames. ], batch size: 92, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:34:53,676 INFO [train.py:968] (0/2) Epoch 14, batch 22600, giga_loss[loss=0.2666, simple_loss=0.3357, pruned_loss=0.0987, over 28899.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3491, pruned_loss=0.1018, over 5720729.32 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3583, pruned_loss=0.1041, over 5753811.80 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3474, pruned_loss=0.1005, over 5712310.64 frames. ], batch size: 227, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:34:59,033 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.402e+02 1.166e+03 1.368e+03 1.775e+03 7.716e+03, threshold=2.735e+03, percent-clipped=2.0
+2023-03-07 08:35:28,995 INFO [train.py:968] (0/2) Epoch 14, batch 22650, giga_loss[loss=0.302, simple_loss=0.3771, pruned_loss=0.1135, over 27610.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3489, pruned_loss=0.1015, over 5713592.77 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3589, pruned_loss=0.1051, over 5746253.26 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3464, pruned_loss=0.09936, over 5711563.84 frames. ], batch size: 472, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:36:14,285 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-616000.pt
+2023-03-07 08:36:14,594 INFO [train.py:968] (0/2) Epoch 14, batch 22700, giga_loss[loss=0.2762, simple_loss=0.3618, pruned_loss=0.09524, over 28575.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3509, pruned_loss=0.1008, over 5702291.35 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3592, pruned_loss=0.1053, over 5738332.08 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3486, pruned_loss=0.09892, over 5707431.22 frames. ], batch size: 307, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:36:21,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.232e+02 1.230e+03 1.545e+03 2.264e+03 6.717e+03, threshold=3.091e+03, percent-clipped=15.0
+2023-03-07 08:36:24,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-07 08:36:52,661 INFO [train.py:968] (0/2) Epoch 14, batch 22750, giga_loss[loss=0.2763, simple_loss=0.3488, pruned_loss=0.1019, over 28908.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3501, pruned_loss=0.1005, over 5712222.83 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3586, pruned_loss=0.1053, over 5739395.60 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3486, pruned_loss=0.09877, over 5714683.47 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:37:22,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2800, 1.7614, 1.4114, 1.4852], device='cuda:0'), covar=tensor([0.0755, 0.0304, 0.0323, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0179, 0.0113, 0.0113, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0095], device='cuda:0')
+2023-03-07 08:37:33,042 INFO [train.py:968] (0/2) Epoch 14, batch 22800, giga_loss[loss=0.2733, simple_loss=0.3453, pruned_loss=0.1007, over 28786.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3488, pruned_loss=0.1002, over 5720075.38 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3588, pruned_loss=0.1056, over 5740845.78 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3472, pruned_loss=0.09856, over 5720486.19 frames. ], batch size: 284, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:37:43,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.644e+02 1.103e+03 1.460e+03 1.881e+03 5.605e+03, threshold=2.921e+03, percent-clipped=7.0
+2023-03-07 08:37:54,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3312, 1.6153, 1.4959, 1.2420], device='cuda:0'), covar=tensor([0.2705, 0.2110, 0.1458, 0.2019], device='cuda:0'), in_proj_covar=tensor([0.1798, 0.1710, 0.1659, 0.1773], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 08:38:02,474 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=616131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:38:15,427 INFO [train.py:968] (0/2) Epoch 14, batch 22850, giga_loss[loss=0.2203, simple_loss=0.2934, pruned_loss=0.07362, over 28411.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3469, pruned_loss=0.1005, over 5722329.62 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3583, pruned_loss=0.1054, over 5745205.12 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3458, pruned_loss=0.09923, over 5717959.18 frames. ], batch size: 71, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:38:36,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=616176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:38:57,098 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-07 08:38:57,257 INFO [train.py:968] (0/2) Epoch 14, batch 22900, giga_loss[loss=0.3016, simple_loss=0.3673, pruned_loss=0.1179, over 28813.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3465, pruned_loss=0.1019, over 5718911.27 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3588, pruned_loss=0.1057, over 5746354.70 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3451, pruned_loss=0.1005, over 5714207.04 frames. ], batch size: 262, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:38:58,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=616201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:39:04,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.206e+02 1.104e+03 1.430e+03 2.032e+03 5.636e+03, threshold=2.860e+03, percent-clipped=11.0
+2023-03-07 08:39:23,158 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-07 08:39:33,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4899, 1.7303, 1.3480, 1.7441], device='cuda:0'), covar=tensor([0.2623, 0.2482, 0.2972, 0.2100], device='cuda:0'), in_proj_covar=tensor([0.1355, 0.0996, 0.1199, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 08:39:36,050 INFO [train.py:968] (0/2) Epoch 14, batch 22950, giga_loss[loss=0.2388, simple_loss=0.3153, pruned_loss=0.08114, over 28940.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3463, pruned_loss=0.1028, over 5716620.61 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3589, pruned_loss=0.106, over 5739592.99 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3447, pruned_loss=0.1014, over 5718212.15 frames. ], batch size: 213, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:40:14,904 INFO [train.py:968] (0/2) Epoch 14, batch 23000, giga_loss[loss=0.2718, simple_loss=0.344, pruned_loss=0.09983, over 28748.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.345, pruned_loss=0.1021, over 5691211.36 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3597, pruned_loss=0.1066, over 5716880.93 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3428, pruned_loss=0.1003, over 5712042.41 frames. ], batch size: 243, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:40:21,629 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-07 08:40:23,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.378e+02 1.161e+03 1.504e+03 1.897e+03 5.731e+03, threshold=3.008e+03, percent-clipped=5.0
+2023-03-07 08:40:51,094 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=616344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:40:51,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8087, 2.0564, 2.1289, 1.6473], device='cuda:0'), covar=tensor([0.1849, 0.2128, 0.1441, 0.1568], device='cuda:0'), in_proj_covar=tensor([0.0851, 0.0687, 0.0891, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 08:40:52,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=616347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:40:54,731 INFO [train.py:968] (0/2) Epoch 14, batch 23050, giga_loss[loss=0.2813, simple_loss=0.3544, pruned_loss=0.104, over 27907.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3403, pruned_loss=0.09963, over 5702703.42 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3596, pruned_loss=0.1067, over 5718413.73 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3383, pruned_loss=0.09809, over 5717545.97 frames. ], batch size: 412, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:41:16,457 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=616376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:41:19,250 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 08:41:36,556 INFO [train.py:968] (0/2) Epoch 14, batch 23100, giga_loss[loss=0.2189, simple_loss=0.2889, pruned_loss=0.07446, over 28688.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3366, pruned_loss=0.09764, over 5699524.64 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3602, pruned_loss=0.107, over 5712715.84 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3341, pruned_loss=0.09591, over 5715559.70 frames. ], batch size: 99, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:41:42,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.067e+02 1.180e+03 1.488e+03 2.212e+03 6.765e+03, threshold=2.976e+03, percent-clipped=8.0
+2023-03-07 08:42:13,398 INFO [train.py:968] (0/2) Epoch 14, batch 23150, giga_loss[loss=0.2588, simple_loss=0.3307, pruned_loss=0.09344, over 28905.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3353, pruned_loss=0.09689, over 5712321.20 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3605, pruned_loss=0.1075, over 5717191.65 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3324, pruned_loss=0.09486, over 5721046.88 frames. ], batch size: 199, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:42:16,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5282, 1.5785, 1.8319, 1.3945], device='cuda:0'), covar=tensor([0.1507, 0.1956, 0.1234, 0.1450], device='cuda:0'), in_proj_covar=tensor([0.0851, 0.0688, 0.0893, 0.0797], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 08:42:35,926 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5319, 1.7862, 1.5479, 1.5104], device='cuda:0'), covar=tensor([0.1969, 0.2460, 0.2324, 0.2442], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0736, 0.0687, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 08:42:56,491 INFO [train.py:968] (0/2) Epoch 14, batch 23200, giga_loss[loss=0.2498, simple_loss=0.3235, pruned_loss=0.08804, over 28516.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3381, pruned_loss=0.09775, over 5709015.34 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.361, pruned_loss=0.1078, over 5719633.07 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3351, pruned_loss=0.09574, over 5713516.66 frames. ], batch size: 71, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:43:01,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=616506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:43:06,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.736e+02 1.122e+03 1.357e+03 1.823e+03 4.982e+03, threshold=2.715e+03, percent-clipped=5.0
+2023-03-07 08:43:26,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9004, 3.7028, 3.5237, 1.6554], device='cuda:0'), covar=tensor([0.0671, 0.0789, 0.0708, 0.2126], device='cuda:0'), in_proj_covar=tensor([0.1091, 0.1012, 0.0880, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 08:43:39,381 INFO [train.py:968] (0/2) Epoch 14, batch 23250, giga_loss[loss=0.2919, simple_loss=0.3695, pruned_loss=0.1072, over 28929.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3428, pruned_loss=0.1004, over 5708545.69 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3609, pruned_loss=0.108, over 5724098.84 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3397, pruned_loss=0.09831, over 5707611.38 frames. ], batch size: 199, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:43:40,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=616551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:43:49,193 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9028, 1.7110, 1.4398, 1.4329], device='cuda:0'), covar=tensor([0.0785, 0.0706, 0.0957, 0.1146], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0438, 0.0503, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 08:44:20,926 INFO [train.py:968] (0/2) Epoch 14, batch 23300, giga_loss[loss=0.271, simple_loss=0.3512, pruned_loss=0.09538, over 28749.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3452, pruned_loss=0.1008, over 5700085.07 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3612, pruned_loss=0.1082, over 5715415.34 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3425, pruned_loss=0.09891, over 5707618.15 frames. ], batch size: 284, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:44:30,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.240e+02 1.210e+03 1.511e+03 2.025e+03 8.129e+03, threshold=3.023e+03, percent-clipped=8.0
+2023-03-07 08:45:03,221 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=616649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:45:03,580 INFO [train.py:968] (0/2) Epoch 14, batch 23350, giga_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.08912, over 28806.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3478, pruned_loss=0.1015, over 5711847.07 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3616, pruned_loss=0.1088, over 5719953.77 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3449, pruned_loss=0.09937, over 5713468.96 frames. ], batch size: 284, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:45:05,184 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=616652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:45:30,260 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=616681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:45:39,707 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=616694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:45:41,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=616697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:45:44,427 INFO [train.py:968] (0/2) Epoch 14, batch 23400, giga_loss[loss=0.2856, simple_loss=0.3645, pruned_loss=0.1034, over 28698.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3492, pruned_loss=0.1021, over 5718564.52 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3621, pruned_loss=0.1093, over 5722895.65 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3462, pruned_loss=0.09976, over 5717026.06 frames. ], batch size: 262, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:45:52,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.544e+02 1.231e+03 1.494e+03 2.074e+03 7.075e+03, threshold=2.989e+03, percent-clipped=14.0
+2023-03-07 08:45:54,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3966, 4.1977, 3.9720, 2.0502], device='cuda:0'), covar=tensor([0.0492, 0.0684, 0.0664, 0.2193], device='cuda:0'), in_proj_covar=tensor([0.1095, 0.1012, 0.0880, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 08:46:06,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=616726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:46:14,585 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-07 08:46:27,580 INFO [train.py:968] (0/2) Epoch 14, batch 23450, giga_loss[loss=0.3348, simple_loss=0.3951, pruned_loss=0.1372, over 28919.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3551, pruned_loss=0.1073, over 5702306.63 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3624, pruned_loss=0.1096, over 5716292.13 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.352, pruned_loss=0.1049, over 5705962.22 frames. ], batch size: 227, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:47:19,784 INFO [train.py:968] (0/2) Epoch 14, batch 23500, giga_loss[loss=0.3505, simple_loss=0.4084, pruned_loss=0.1463, over 28610.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3619, pruned_loss=0.1127, over 5694144.65 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3627, pruned_loss=0.11, over 5719311.85 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3591, pruned_loss=0.1105, over 5694134.65 frames. ], batch size: 85, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:47:32,484 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-07 08:47:32,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.038e+02 1.554e+03 1.977e+03 2.784e+03 6.668e+03, threshold=3.954e+03, percent-clipped=21.0
+2023-03-07 08:48:07,275 INFO [train.py:968] (0/2) Epoch 14, batch 23550, giga_loss[loss=0.318, simple_loss=0.3828, pruned_loss=0.1266, over 28807.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3674, pruned_loss=0.117, over 5679167.25 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3626, pruned_loss=0.1102, over 5713433.44 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3653, pruned_loss=0.1153, over 5683067.07 frames. ], batch size: 284, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:48:44,464 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-07 08:48:55,824 INFO [train.py:968] (0/2) Epoch 14, batch 23600, giga_loss[loss=0.3103, simple_loss=0.3772, pruned_loss=0.1217, over 28639.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3729, pruned_loss=0.1216, over 5667316.55 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.363, pruned_loss=0.1107, over 5698426.35 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.12, over 5682414.33 frames. ], batch size: 60, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:49:05,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.523e+02 1.732e+03 2.204e+03 2.832e+03 2.469e+04, threshold=4.407e+03, percent-clipped=17.0
+2023-03-07 08:49:18,321 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.12 vs. limit=5.0
+2023-03-07 08:49:42,571 INFO [train.py:968] (0/2) Epoch 14, batch 23650, libri_loss[loss=0.3456, simple_loss=0.4013, pruned_loss=0.145, over 29638.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3791, pruned_loss=0.127, over 5674871.63 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.363, pruned_loss=0.1109, over 5705821.63 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3782, pruned_loss=0.126, over 5679209.24 frames. ], batch size: 88, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:50:17,679 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3830, 1.5518, 1.4544, 1.2244], device='cuda:0'), covar=tensor([0.2217, 0.2028, 0.1447, 0.1844], device='cuda:0'), in_proj_covar=tensor([0.1793, 0.1709, 0.1650, 0.1769], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 08:50:29,776 INFO [train.py:968] (0/2) Epoch 14, batch 23700, giga_loss[loss=0.4277, simple_loss=0.4382, pruned_loss=0.2085, over 23176.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3844, pruned_loss=0.1317, over 5664346.18 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3638, pruned_loss=0.1117, over 5702336.72 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3836, pruned_loss=0.1309, over 5670001.31 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:50:39,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.756e+03 2.204e+03 3.321e+03 1.295e+04, threshold=4.408e+03, percent-clipped=13.0
+2023-03-07 08:51:16,153 INFO [train.py:968] (0/2) Epoch 14, batch 23750, giga_loss[loss=0.2764, simple_loss=0.3472, pruned_loss=0.1028, over 28734.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3855, pruned_loss=0.1337, over 5660449.67 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3633, pruned_loss=0.1114, over 5706329.66 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3857, pruned_loss=0.1337, over 5660719.63 frames. ], batch size: 99, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:52:09,221 INFO [train.py:968] (0/2) Epoch 14, batch 23800, libri_loss[loss=0.2683, simple_loss=0.3248, pruned_loss=0.1059, over 29685.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3885, pruned_loss=0.1373, over 5650405.03 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3635, pruned_loss=0.1117, over 5698531.24 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3891, pruned_loss=0.1376, over 5656632.13 frames. ], batch size: 69, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:52:19,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.630e+03 2.148e+03 2.756e+03 6.904e+03, threshold=4.295e+03, percent-clipped=3.0
+2023-03-07 08:52:54,450 INFO [train.py:968] (0/2) Epoch 14, batch 23850, giga_loss[loss=0.3424, simple_loss=0.395, pruned_loss=0.1449, over 28637.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3909, pruned_loss=0.1404, over 5647447.63 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3634, pruned_loss=0.112, over 5701982.33 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3924, pruned_loss=0.1413, over 5647507.89 frames. ], batch size: 262, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:53:38,518 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=617187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:53:51,073 INFO [train.py:968] (0/2) Epoch 14, batch 23900, giga_loss[loss=0.3313, simple_loss=0.3914, pruned_loss=0.1356, over 28913.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3951, pruned_loss=0.1433, over 5630858.64 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3638, pruned_loss=0.1123, over 5685371.96 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3966, pruned_loss=0.1445, over 5644268.34 frames. ], batch size: 136, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:54:06,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.897e+02 2.059e+03 2.514e+03 3.989e+03 1.137e+04, threshold=5.028e+03, percent-clipped=19.0
+2023-03-07 08:54:45,744 INFO [train.py:968] (0/2) Epoch 14, batch 23950, giga_loss[loss=0.4392, simple_loss=0.4503, pruned_loss=0.2141, over 26553.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3947, pruned_loss=0.1435, over 5634914.42 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3636, pruned_loss=0.1124, over 5689711.97 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3968, pruned_loss=0.1451, over 5640586.52 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:55:32,371 INFO [train.py:968] (0/2) Epoch 14, batch 24000, giga_loss[loss=0.2718, simple_loss=0.339, pruned_loss=0.1023, over 28801.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3927, pruned_loss=0.1424, over 5642886.27 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3636, pruned_loss=0.1124, over 5695557.21 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3952, pruned_loss=0.1445, over 5640731.48 frames. ], batch size: 119, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 08:55:32,377 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 08:55:41,146 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2117, simple_loss=0.3187, pruned_loss=0.05229, over 944034.00 frames. 
+2023-03-07 08:55:41,147 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 08:55:51,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.225e+03 1.799e+03 2.241e+03 3.199e+03 9.343e+03, threshold=4.481e+03, percent-clipped=10.0
+2023-03-07 08:56:29,829 INFO [train.py:968] (0/2) Epoch 14, batch 24050, giga_loss[loss=0.3461, simple_loss=0.4121, pruned_loss=0.14, over 28954.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3934, pruned_loss=0.1426, over 5640363.99 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3641, pruned_loss=0.1128, over 5694361.35 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3953, pruned_loss=0.1444, over 5638841.52 frames. ], batch size: 164, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:57:16,963 INFO [train.py:968] (0/2) Epoch 14, batch 24100, giga_loss[loss=0.3663, simple_loss=0.4136, pruned_loss=0.1595, over 28298.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3936, pruned_loss=0.142, over 5639323.48 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3641, pruned_loss=0.1128, over 5697458.88 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3958, pruned_loss=0.144, over 5634216.71 frames. ], batch size: 368, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:57:30,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.149e+03 1.697e+03 2.181e+03 3.258e+03 9.350e+03, threshold=4.361e+03, percent-clipped=14.0
+2023-03-07 08:57:51,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=617433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:58:07,295 INFO [train.py:968] (0/2) Epoch 14, batch 24150, giga_loss[loss=0.3953, simple_loss=0.413, pruned_loss=0.1888, over 23570.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3951, pruned_loss=0.1433, over 5624743.01 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3638, pruned_loss=0.1128, over 5692437.16 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.3978, pruned_loss=0.1456, over 5623637.40 frames. ], batch size: 710, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:58:57,265 INFO [train.py:968] (0/2) Epoch 14, batch 24200, libri_loss[loss=0.3293, simple_loss=0.3854, pruned_loss=0.1365, over 29546.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3931, pruned_loss=0.1418, over 5636060.42 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3633, pruned_loss=0.1126, over 5700000.70 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3969, pruned_loss=0.1448, over 5626007.35 frames. ], batch size: 80, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 08:59:12,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.694e+03 2.265e+03 3.388e+03 6.829e+03, threshold=4.530e+03, percent-clipped=11.0
+2023-03-07 08:59:45,207 INFO [train.py:968] (0/2) Epoch 14, batch 24250, giga_loss[loss=0.3334, simple_loss=0.3936, pruned_loss=0.1366, over 28555.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3897, pruned_loss=0.1375, over 5630015.10 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3635, pruned_loss=0.1129, over 5690718.41 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3935, pruned_loss=0.1406, over 5627697.57 frames. ], batch size: 336, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 08:59:59,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=617562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:00:34,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2797, 1.1797, 1.0753, 1.4773], device='cuda:0'), covar=tensor([0.0736, 0.0352, 0.0348, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 09:00:34,483 INFO [train.py:968] (0/2) Epoch 14, batch 24300, libri_loss[loss=0.307, simple_loss=0.3728, pruned_loss=0.1206, over 29638.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3872, pruned_loss=0.1345, over 5641380.36 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3637, pruned_loss=0.1131, over 5694388.74 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3905, pruned_loss=0.1372, over 5635134.24 frames. ], batch size: 91, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 09:00:50,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.883e+02 1.559e+03 2.197e+03 2.902e+03 8.329e+03, threshold=4.393e+03, percent-clipped=6.0
+2023-03-07 09:01:05,170 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-07 09:01:21,333 INFO [train.py:968] (0/2) Epoch 14, batch 24350, giga_loss[loss=0.2703, simple_loss=0.3478, pruned_loss=0.09643, over 28864.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3828, pruned_loss=0.1304, over 5658610.13 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3637, pruned_loss=0.1133, over 5694610.88 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.386, pruned_loss=0.1329, over 5652185.33 frames. ], batch size: 145, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 09:01:39,506 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=617667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:02:00,032 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 09:02:09,116 INFO [train.py:968] (0/2) Epoch 14, batch 24400, giga_loss[loss=0.3601, simple_loss=0.4073, pruned_loss=0.1565, over 28643.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3809, pruned_loss=0.1292, over 5648889.37 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.364, pruned_loss=0.1136, over 5688491.49 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3836, pruned_loss=0.1313, over 5648483.46 frames. ], batch size: 262, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:02:14,019 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=617705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:02:17,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=617708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:02:25,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.612e+02 1.696e+03 2.151e+03 3.068e+03 8.063e+03, threshold=4.302e+03, percent-clipped=8.0
+2023-03-07 09:02:46,882 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=617737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:02:55,048 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1451, 1.2175, 3.7241, 3.1136], device='cuda:0'), covar=tensor([0.2069, 0.2873, 0.0735, 0.1031], device='cuda:0'), in_proj_covar=tensor([0.0693, 0.0607, 0.0887, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:02:57,727 INFO [train.py:968] (0/2) Epoch 14, batch 24450, giga_loss[loss=0.3681, simple_loss=0.4111, pruned_loss=0.1626, over 27553.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3803, pruned_loss=0.1288, over 5663171.69 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3641, pruned_loss=0.1138, over 5691173.69 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3826, pruned_loss=0.1306, over 5659763.84 frames. ], batch size: 472, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:03:35,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3956, 1.6991, 1.2857, 1.5343], device='cuda:0'), covar=tensor([0.0729, 0.0361, 0.0334, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 09:03:42,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=617791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:03:51,729 INFO [train.py:968] (0/2) Epoch 14, batch 24500, giga_loss[loss=0.2881, simple_loss=0.3544, pruned_loss=0.1109, over 28522.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3792, pruned_loss=0.1276, over 5665055.78 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3641, pruned_loss=0.1139, over 5695185.18 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3815, pruned_loss=0.1293, over 5657981.06 frames. ], batch size: 85, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:04:01,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=617808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:04:06,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.496e+03 1.954e+03 2.827e+03 6.507e+03, threshold=3.908e+03, percent-clipped=8.0
+2023-03-07 09:04:39,596 INFO [train.py:968] (0/2) Epoch 14, batch 24550, giga_loss[loss=0.2829, simple_loss=0.3546, pruned_loss=0.1056, over 28548.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3763, pruned_loss=0.125, over 5662737.69 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.364, pruned_loss=0.114, over 5691790.95 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3787, pruned_loss=0.1267, over 5659494.57 frames. ], batch size: 92, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:05:32,243 INFO [train.py:968] (0/2) Epoch 14, batch 24600, giga_loss[loss=0.2856, simple_loss=0.3696, pruned_loss=0.1008, over 29138.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3774, pruned_loss=0.1231, over 5679249.50 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.364, pruned_loss=0.1142, over 5694988.43 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3795, pruned_loss=0.1244, over 5673500.45 frames. ], batch size: 128, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:05:34,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5490, 1.5574, 1.1465, 1.1943], device='cuda:0'), covar=tensor([0.0802, 0.0576, 0.1011, 0.1168], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0438, 0.0501, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:05:34,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0934, 4.9903, 2.0092, 2.3641], device='cuda:0'), covar=tensor([0.0826, 0.0321, 0.0843, 0.1094], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0527, 0.0354, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 09:05:35,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4069, 2.0325, 1.4618, 1.7400], device='cuda:0'), covar=tensor([0.0720, 0.0258, 0.0307, 0.0782], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 09:05:44,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.371e+02 1.603e+03 1.847e+03 2.394e+03 7.001e+03, threshold=3.695e+03, percent-clipped=6.0
+2023-03-07 09:05:55,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5319, 4.3478, 4.1368, 1.9574], device='cuda:0'), covar=tensor([0.0520, 0.0715, 0.0822, 0.2313], device='cuda:0'), in_proj_covar=tensor([0.1112, 0.1036, 0.0900, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 09:06:01,023 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=617929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:06:22,935 INFO [train.py:968] (0/2) Epoch 14, batch 24650, giga_loss[loss=0.285, simple_loss=0.3566, pruned_loss=0.1067, over 28616.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3788, pruned_loss=0.1234, over 5653995.49 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3641, pruned_loss=0.1143, over 5693452.72 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3807, pruned_loss=0.1246, over 5650205.38 frames. ], batch size: 92, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:06:24,305 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=617951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:06:26,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=617954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:06:54,384 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=617983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:06:57,119 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-07 09:07:08,891 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-618000.pt
+2023-03-07 09:07:09,188 INFO [train.py:968] (0/2) Epoch 14, batch 24700, giga_loss[loss=0.3199, simple_loss=0.3812, pruned_loss=0.1293, over 28415.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3806, pruned_loss=0.1248, over 5666176.73 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3643, pruned_loss=0.1146, over 5698104.40 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3822, pruned_loss=0.1258, over 5658311.08 frames. ], batch size: 78, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:07:23,443 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.765e+03 2.325e+03 3.027e+03 8.211e+03, threshold=4.649e+03, percent-clipped=11.0
+2023-03-07 09:07:24,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 09:07:47,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:07:54,888 INFO [train.py:968] (0/2) Epoch 14, batch 24750, giga_loss[loss=0.2823, simple_loss=0.3522, pruned_loss=0.1062, over 28305.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3819, pruned_loss=0.1271, over 5663960.58 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3643, pruned_loss=0.1149, over 5702215.30 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3837, pruned_loss=0.1278, over 5653253.88 frames. ], batch size: 77, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:08:15,410 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-07 09:08:46,760 INFO [train.py:968] (0/2) Epoch 14, batch 24800, giga_loss[loss=0.2953, simple_loss=0.3667, pruned_loss=0.1119, over 29063.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3799, pruned_loss=0.127, over 5648712.62 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3643, pruned_loss=0.1148, over 5693821.12 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3815, pruned_loss=0.1278, over 5647397.81 frames. ], batch size: 128, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:08:48,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7640, 1.7412, 1.3176, 1.3762], device='cuda:0'), covar=tensor([0.0794, 0.0675, 0.0946, 0.1254], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0437, 0.0501, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:09:01,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.576e+03 2.109e+03 2.926e+03 5.523e+03, threshold=4.217e+03, percent-clipped=5.0
+2023-03-07 09:09:02,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3160, 1.6492, 0.9288, 1.3165], device='cuda:0'), covar=tensor([0.1070, 0.0687, 0.1599, 0.1198], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0436, 0.0500, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:09:18,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0181, 1.9450, 1.3916, 1.6509], device='cuda:0'), covar=tensor([0.0812, 0.0720, 0.1021, 0.1143], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0436, 0.0501, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:09:29,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6699, 1.7976, 1.0368, 1.5897], device='cuda:0'), covar=tensor([0.1114, 0.0878, 0.1663, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.0366, 0.0437, 0.0501, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:09:30,698 INFO [train.py:968] (0/2) Epoch 14, batch 24850, giga_loss[loss=0.3216, simple_loss=0.3778, pruned_loss=0.1327, over 28910.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3781, pruned_loss=0.1261, over 5666641.97 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3645, pruned_loss=0.1149, over 5697735.63 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3795, pruned_loss=0.1269, over 5661600.57 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:09:47,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:10:03,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:10:07,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:10:18,085 INFO [train.py:968] (0/2) Epoch 14, batch 24900, giga_loss[loss=0.2811, simple_loss=0.3594, pruned_loss=0.1014, over 28940.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3779, pruned_loss=0.1254, over 5671259.41 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3643, pruned_loss=0.1147, over 5700739.58 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3793, pruned_loss=0.1263, over 5664251.83 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:10:20,904 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:10:29,621 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.338e+02 1.541e+03 2.210e+03 3.326e+03 7.755e+03, threshold=4.420e+03, percent-clipped=17.0
+2023-03-07 09:10:31,544 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:10:45,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4637, 2.8325, 1.5397, 1.6705], device='cuda:0'), covar=tensor([0.0759, 0.0312, 0.0696, 0.1024], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0528, 0.0356, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 09:11:04,604 INFO [train.py:968] (0/2) Epoch 14, batch 24950, giga_loss[loss=0.4523, simple_loss=0.4632, pruned_loss=0.2208, over 26672.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3777, pruned_loss=0.1246, over 5673141.53 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3645, pruned_loss=0.1151, over 5703519.70 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3788, pruned_loss=0.1251, over 5664904.70 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:11:11,760 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:11:28,718 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:11:30,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5173, 1.7271, 1.7398, 1.3040], device='cuda:0'), covar=tensor([0.1611, 0.2374, 0.1385, 0.1649], device='cuda:0'), in_proj_covar=tensor([0.0850, 0.0693, 0.0893, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 09:11:48,697 INFO [train.py:968] (0/2) Epoch 14, batch 25000, giga_loss[loss=0.2878, simple_loss=0.3644, pruned_loss=0.1056, over 29062.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3768, pruned_loss=0.1238, over 5670390.02 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3646, pruned_loss=0.1152, over 5700656.30 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3781, pruned_loss=0.1244, over 5664543.25 frames. ], batch size: 155, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:11:55,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:11:58,858 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:12:02,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:12:03,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.345e+02 1.484e+03 1.777e+03 2.309e+03 3.760e+03, threshold=3.554e+03, percent-clipped=0.0
+2023-03-07 09:12:30,377 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:12:38,006 INFO [train.py:968] (0/2) Epoch 14, batch 25050, giga_loss[loss=0.3625, simple_loss=0.4073, pruned_loss=0.1588, over 28606.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3761, pruned_loss=0.1238, over 5675839.76 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3646, pruned_loss=0.1152, over 5705801.56 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3773, pruned_loss=0.1245, over 5665780.17 frames. ], batch size: 307, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:13:04,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2540, 1.2422, 3.5565, 3.1816], device='cuda:0'), covar=tensor([0.1551, 0.2637, 0.0465, 0.1218], device='cuda:0'), in_proj_covar=tensor([0.0694, 0.0611, 0.0894, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:13:29,768 INFO [train.py:968] (0/2) Epoch 14, batch 25100, giga_loss[loss=0.2794, simple_loss=0.3508, pruned_loss=0.1041, over 29095.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3758, pruned_loss=0.1242, over 5678768.75 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3647, pruned_loss=0.1153, over 5702232.74 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3767, pruned_loss=0.1248, over 5674044.23 frames. ], batch size: 136, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:13:40,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2571, 1.1559, 3.9938, 3.2061], device='cuda:0'), covar=tensor([0.1647, 0.2737, 0.0428, 0.0941], device='cuda:0'), in_proj_covar=tensor([0.0694, 0.0610, 0.0893, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:13:44,523 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.698e+03 2.217e+03 2.822e+03 1.106e+04, threshold=4.433e+03, percent-clipped=17.0
+2023-03-07 09:14:11,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4240, 3.5760, 1.5455, 1.6041], device='cuda:0'), covar=tensor([0.0925, 0.0384, 0.0873, 0.1271], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0528, 0.0355, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 09:14:13,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:14:14,145 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:14:14,201 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8340, 2.2067, 1.6382, 2.2551], device='cuda:0'), covar=tensor([0.2303, 0.2343, 0.2684, 0.2188], device='cuda:0'), in_proj_covar=tensor([0.1370, 0.1007, 0.1213, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 09:14:17,128 INFO [train.py:968] (0/2) Epoch 14, batch 25150, giga_loss[loss=0.3316, simple_loss=0.3895, pruned_loss=0.1369, over 28536.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3748, pruned_loss=0.1241, over 5680695.78 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.365, pruned_loss=0.1153, over 5703685.79 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3754, pruned_loss=0.1246, over 5675316.57 frames. ], batch size: 336, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:14:17,415 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:14:42,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:15:01,316 INFO [train.py:968] (0/2) Epoch 14, batch 25200, giga_loss[loss=0.3056, simple_loss=0.3715, pruned_loss=0.1199, over 28648.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3736, pruned_loss=0.1239, over 5694379.50 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3655, pruned_loss=0.1158, over 5709051.98 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.374, pruned_loss=0.1242, over 5684650.56 frames. ], batch size: 262, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:15:16,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.714e+03 2.394e+03 3.401e+03 9.768e+03, threshold=4.789e+03, percent-clipped=12.0
+2023-03-07 09:15:29,819 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-07 09:15:47,509 INFO [train.py:968] (0/2) Epoch 14, batch 25250, giga_loss[loss=0.2946, simple_loss=0.3608, pruned_loss=0.1142, over 28734.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3715, pruned_loss=0.123, over 5683228.07 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3657, pruned_loss=0.1159, over 5702981.94 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3719, pruned_loss=0.1234, over 5680027.50 frames. ], batch size: 242, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:16:11,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:16:33,052 INFO [train.py:968] (0/2) Epoch 14, batch 25300, giga_loss[loss=0.2777, simple_loss=0.3524, pruned_loss=0.1015, over 28658.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3698, pruned_loss=0.1219, over 5690416.81 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3656, pruned_loss=0.1159, over 5707934.19 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3703, pruned_loss=0.1224, over 5683020.26 frames. ], batch size: 262, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:16:44,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:16:51,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.778e+02 1.611e+03 1.955e+03 2.524e+03 7.109e+03, threshold=3.911e+03, percent-clipped=3.0
+2023-03-07 09:17:05,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:17:16,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3781, 3.4058, 1.4343, 1.4521], device='cuda:0'), covar=tensor([0.0992, 0.0383, 0.0907, 0.1425], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0529, 0.0355, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 09:17:20,112 INFO [train.py:968] (0/2) Epoch 14, batch 25350, giga_loss[loss=0.3394, simple_loss=0.3934, pruned_loss=0.1427, over 27971.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3689, pruned_loss=0.1211, over 5687216.84 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3656, pruned_loss=0.116, over 5710171.12 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3694, pruned_loss=0.1215, over 5678791.61 frames. ], batch size: 412, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:17:20,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:18:09,267 INFO [train.py:968] (0/2) Epoch 14, batch 25400, giga_loss[loss=0.3386, simple_loss=0.3995, pruned_loss=0.1389, over 28883.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3701, pruned_loss=0.1209, over 5680071.49 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.366, pruned_loss=0.1164, over 5701552.23 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3702, pruned_loss=0.121, over 5680558.63 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:18:23,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.906e+02 1.609e+03 2.075e+03 2.831e+03 8.802e+03, threshold=4.149e+03, percent-clipped=10.0
+2023-03-07 09:18:26,889 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:18:31,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:18:51,667 INFO [train.py:968] (0/2) Epoch 14, batch 25450, giga_loss[loss=0.2593, simple_loss=0.3425, pruned_loss=0.088, over 29047.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3703, pruned_loss=0.1206, over 5666594.62 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3658, pruned_loss=0.1163, over 5687062.95 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3707, pruned_loss=0.1208, over 5679632.25 frames. ], batch size: 155, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:18:56,974 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:06,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3236, 3.0960, 2.9198, 1.4420], device='cuda:0'), covar=tensor([0.0900, 0.1076, 0.0959, 0.2259], device='cuda:0'), in_proj_covar=tensor([0.1119, 0.1039, 0.0902, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 09:19:14,390 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:18,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:30,662 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 09:19:32,408 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:34,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:37,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3374, 3.1118, 1.3715, 1.5215], device='cuda:0'), covar=tensor([0.0948, 0.0340, 0.0902, 0.1261], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0527, 0.0354, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 09:19:37,506 INFO [train.py:968] (0/2) Epoch 14, batch 25500, giga_loss[loss=0.314, simple_loss=0.384, pruned_loss=0.122, over 29050.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.371, pruned_loss=0.1208, over 5677367.47 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3661, pruned_loss=0.1168, over 5695915.62 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3714, pruned_loss=0.1208, over 5679065.44 frames. ], batch size: 164, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:19:42,291 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:47,896 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:50,231 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.391e+02 1.647e+03 2.314e+03 3.392e+03 1.023e+04, threshold=4.627e+03, percent-clipped=15.0
+2023-03-07 09:19:51,103 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:53,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:57,704 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:20:03,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6028, 1.6665, 1.7857, 1.3372], device='cuda:0'), covar=tensor([0.1713, 0.2413, 0.1369, 0.1651], device='cuda:0'), in_proj_covar=tensor([0.0848, 0.0691, 0.0890, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 09:20:18,256 INFO [train.py:968] (0/2) Epoch 14, batch 25550, giga_loss[loss=0.3752, simple_loss=0.4275, pruned_loss=0.1615, over 28563.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3737, pruned_loss=0.123, over 5673491.72 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3673, pruned_loss=0.1178, over 5691851.16 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.373, pruned_loss=0.1223, over 5678458.55 frames. ], batch size: 336, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:20:33,931 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5127, 1.6827, 1.4111, 1.6177], device='cuda:0'), covar=tensor([0.2465, 0.2562, 0.2862, 0.2240], device='cuda:0'), in_proj_covar=tensor([0.1363, 0.1003, 0.1206, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 09:20:54,077 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618888.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 09:21:03,376 INFO [train.py:968] (0/2) Epoch 14, batch 25600, giga_loss[loss=0.3197, simple_loss=0.3776, pruned_loss=0.1309, over 29014.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3774, pruned_loss=0.1266, over 5677067.16 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3678, pruned_loss=0.1183, over 5694240.91 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3767, pruned_loss=0.1257, over 5678355.02 frames. ], batch size: 128, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:21:13,510 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:21:20,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.584e+03 2.087e+03 2.732e+03 8.808e+03, threshold=4.174e+03, percent-clipped=6.0
+2023-03-07 09:21:54,907 INFO [train.py:968] (0/2) Epoch 14, batch 25650, giga_loss[loss=0.3169, simple_loss=0.3858, pruned_loss=0.1239, over 28961.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3787, pruned_loss=0.1295, over 5678114.41 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.368, pruned_loss=0.1187, over 5697522.18 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3782, pruned_loss=0.1286, over 5675845.64 frames. ], batch size: 136, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:22:09,079 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:22:13,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:22:31,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:22:44,592 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:22:47,883 INFO [train.py:968] (0/2) Epoch 14, batch 25700, giga_loss[loss=0.3311, simple_loss=0.3786, pruned_loss=0.1418, over 28931.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3795, pruned_loss=0.1312, over 5670837.64 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.368, pruned_loss=0.1187, over 5690163.09 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3791, pruned_loss=0.1306, over 5676355.37 frames. ], batch size: 213, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:22:49,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1494, 1.1147, 3.8486, 3.2472], device='cuda:0'), covar=tensor([0.1721, 0.2873, 0.0443, 0.1590], device='cuda:0'), in_proj_covar=tensor([0.0699, 0.0614, 0.0902, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:22:50,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3236, 1.5758, 1.1701, 1.5277], device='cuda:0'), covar=tensor([0.2610, 0.2565, 0.2984, 0.2277], device='cuda:0'), in_proj_covar=tensor([0.1362, 0.1003, 0.1207, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 09:23:02,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.757e+03 2.415e+03 3.076e+03 7.980e+03, threshold=4.830e+03, percent-clipped=11.0
+2023-03-07 09:23:29,850 INFO [train.py:968] (0/2) Epoch 14, batch 25750, giga_loss[loss=0.3509, simple_loss=0.3936, pruned_loss=0.1541, over 28995.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3787, pruned_loss=0.1306, over 5671247.52 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3681, pruned_loss=0.119, over 5684046.43 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3787, pruned_loss=0.1304, over 5679875.47 frames. ], batch size: 128, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:24:11,108 INFO [train.py:968] (0/2) Epoch 14, batch 25800, giga_loss[loss=0.2818, simple_loss=0.3553, pruned_loss=0.1042, over 29012.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3774, pruned_loss=0.1301, over 5659218.67 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3682, pruned_loss=0.1194, over 5682048.49 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3779, pruned_loss=0.1301, over 5667257.37 frames. ], batch size: 155, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:24:27,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.676e+03 2.382e+03 3.372e+03 8.728e+03, threshold=4.764e+03, percent-clipped=11.0
+2023-03-07 09:24:35,057 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:24:38,259 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:24:47,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=619143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:24:50,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2563, 1.7343, 1.3156, 0.6163], device='cuda:0'), covar=tensor([0.3206, 0.2242, 0.1813, 0.4315], device='cuda:0'), in_proj_covar=tensor([0.1620, 0.1554, 0.1530, 0.1336], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 09:24:53,501 INFO [train.py:968] (0/2) Epoch 14, batch 25850, giga_loss[loss=0.3577, simple_loss=0.4094, pruned_loss=0.1529, over 28330.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3777, pruned_loss=0.1285, over 5673012.17 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3684, pruned_loss=0.1194, over 5687480.39 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3782, pruned_loss=0.1288, over 5674090.37 frames. ], batch size: 368, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:25:03,598 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:25:32,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:25:35,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:25:43,123 INFO [train.py:968] (0/2) Epoch 14, batch 25900, giga_loss[loss=0.3335, simple_loss=0.389, pruned_loss=0.139, over 28652.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3744, pruned_loss=0.1256, over 5665482.82 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3682, pruned_loss=0.1192, over 5691196.15 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3752, pruned_loss=0.1261, over 5662808.69 frames. ], batch size: 85, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:25:57,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.086e+03 1.566e+03 2.339e+03 2.965e+03 5.679e+03, threshold=4.677e+03, percent-clipped=4.0
+2023-03-07 09:26:12,554 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5886, 1.6332, 1.5672, 1.5042], device='cuda:0'), covar=tensor([0.1488, 0.2068, 0.2149, 0.1908], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0734, 0.0687, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 09:26:20,922 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=619242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:26:28,940 INFO [train.py:968] (0/2) Epoch 14, batch 25950, giga_loss[loss=0.2849, simple_loss=0.3511, pruned_loss=0.1094, over 28547.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3715, pruned_loss=0.1234, over 5666250.31 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3683, pruned_loss=0.1192, over 5692416.40 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3721, pruned_loss=0.1239, over 5662558.28 frames. ], batch size: 85, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:26:42,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619263.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 09:27:00,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:27:10,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3252, 1.5880, 1.3438, 1.0624], device='cuda:0'), covar=tensor([0.1731, 0.1635, 0.1673, 0.1592], device='cuda:0'), in_proj_covar=tensor([0.1361, 0.0999, 0.1204, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 09:27:13,703 INFO [train.py:968] (0/2) Epoch 14, batch 26000, giga_loss[loss=0.3204, simple_loss=0.3746, pruned_loss=0.1331, over 27976.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3717, pruned_loss=0.1244, over 5668507.25 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3687, pruned_loss=0.1195, over 5695631.44 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3719, pruned_loss=0.1247, over 5661877.55 frames. ], batch size: 412, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:27:30,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.673e+03 2.195e+03 2.980e+03 7.621e+03, threshold=4.390e+03, percent-clipped=4.0
+2023-03-07 09:27:38,735 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-07 09:27:48,069 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:27:49,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:27:50,598 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:27:53,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:28:04,806 INFO [train.py:968] (0/2) Epoch 14, batch 26050, giga_loss[loss=0.2956, simple_loss=0.3573, pruned_loss=0.117, over 28719.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3732, pruned_loss=0.1264, over 5662627.86 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3688, pruned_loss=0.1199, over 5700283.80 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3733, pruned_loss=0.1265, over 5652456.59 frames. ], batch size: 99, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:28:16,452 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:28:20,751 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:28:23,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4095, 1.8184, 1.5401, 1.5461], device='cuda:0'), covar=tensor([0.0636, 0.0262, 0.0268, 0.0660], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0113, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0055, 0.0095], device='cuda:0')
+2023-03-07 09:28:34,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8274, 2.6741, 1.6470, 0.9892], device='cuda:0'), covar=tensor([0.6158, 0.3174, 0.3555, 0.5461], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1546, 0.1517, 0.1330], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 09:28:50,261 INFO [train.py:968] (0/2) Epoch 14, batch 26100, libri_loss[loss=0.2819, simple_loss=0.3506, pruned_loss=0.1066, over 29278.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3759, pruned_loss=0.1275, over 5659450.12 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3687, pruned_loss=0.1198, over 5694689.15 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3762, pruned_loss=0.1278, over 5654426.35 frames. ], batch size: 94, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:28:56,101 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619406.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 09:28:58,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619409.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 09:29:02,022 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6164, 1.6234, 1.2238, 1.2399], device='cuda:0'), covar=tensor([0.0787, 0.0567, 0.0939, 0.1156], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0442, 0.0503, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:29:07,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.329e+02 1.563e+03 1.782e+03 2.493e+03 6.605e+03, threshold=3.564e+03, percent-clipped=4.0
+2023-03-07 09:29:17,185 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:29:19,264 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:29:26,065 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619438.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 09:29:37,565 INFO [train.py:968] (0/2) Epoch 14, batch 26150, giga_loss[loss=0.3816, simple_loss=0.4235, pruned_loss=0.1699, over 26516.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.379, pruned_loss=0.1267, over 5666500.19 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3683, pruned_loss=0.1196, over 5699711.95 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3797, pruned_loss=0.1273, over 5657576.15 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:29:47,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9573, 3.7552, 3.5605, 1.7131], device='cuda:0'), covar=tensor([0.0762, 0.0976, 0.0979, 0.2107], device='cuda:0'), in_proj_covar=tensor([0.1125, 0.1040, 0.0907, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 09:29:48,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:30:24,813 INFO [train.py:968] (0/2) Epoch 14, batch 26200, giga_loss[loss=0.3052, simple_loss=0.3752, pruned_loss=0.1176, over 28876.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3805, pruned_loss=0.1266, over 5659102.66 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.369, pruned_loss=0.1202, over 5692500.89 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3807, pruned_loss=0.1267, over 5657227.57 frames. ], batch size: 186, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:30:40,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:30:40,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.479e+03 1.932e+03 2.525e+03 1.534e+04, threshold=3.863e+03, percent-clipped=9.0
+2023-03-07 09:31:12,149 INFO [train.py:968] (0/2) Epoch 14, batch 26250, libri_loss[loss=0.4127, simple_loss=0.4421, pruned_loss=0.1916, over 29527.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3828, pruned_loss=0.129, over 5642516.79 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3693, pruned_loss=0.1206, over 5686221.01 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3829, pruned_loss=0.1287, over 5645893.74 frames. ], batch size: 82, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:31:55,528 INFO [train.py:968] (0/2) Epoch 14, batch 26300, giga_loss[loss=0.307, simple_loss=0.37, pruned_loss=0.122, over 28924.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3829, pruned_loss=0.1292, over 5654692.98 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3694, pruned_loss=0.1206, over 5687909.02 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3831, pruned_loss=0.1292, over 5655322.75 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:32:14,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:32:16,597 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.736e+03 2.315e+03 4.005e+03 1.169e+04, threshold=4.629e+03, percent-clipped=27.0
+2023-03-07 09:32:45,070 INFO [train.py:968] (0/2) Epoch 14, batch 26350, giga_loss[loss=0.2755, simple_loss=0.351, pruned_loss=0.09999, over 28354.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3824, pruned_loss=0.13, over 5650835.94 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3691, pruned_loss=0.1205, over 5691379.15 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3831, pruned_loss=0.1303, over 5647459.96 frames. ], batch size: 65, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:32:55,828 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:32:58,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:33:00,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=619666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:33:01,663 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 09:33:27,345 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:33:31,896 INFO [train.py:968] (0/2) Epoch 14, batch 26400, giga_loss[loss=0.2801, simple_loss=0.3496, pruned_loss=0.1053, over 28904.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3808, pruned_loss=0.1299, over 5653461.85 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3682, pruned_loss=0.1199, over 5696876.04 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3825, pruned_loss=0.1309, over 5644781.33 frames. ], batch size: 145, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:33:49,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 2.041e+03 2.581e+03 3.900e+03 1.000e+04, threshold=5.161e+03, percent-clipped=17.0
+2023-03-07 09:34:12,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7786, 1.9383, 2.0245, 1.5966], device='cuda:0'), covar=tensor([0.1818, 0.2301, 0.1405, 0.1585], device='cuda:0'), in_proj_covar=tensor([0.0851, 0.0694, 0.0895, 0.0796], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 09:34:20,482 INFO [train.py:968] (0/2) Epoch 14, batch 26450, giga_loss[loss=0.4216, simple_loss=0.4627, pruned_loss=0.1903, over 28920.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3783, pruned_loss=0.1286, over 5657833.73 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3682, pruned_loss=0.1198, over 5697531.91 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3798, pruned_loss=0.1296, over 5649512.81 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:34:28,602 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:34:31,924 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:34:39,502 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=619772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:35:00,197 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:35:08,330 INFO [train.py:968] (0/2) Epoch 14, batch 26500, giga_loss[loss=0.2974, simple_loss=0.3634, pruned_loss=0.1157, over 28844.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3784, pruned_loss=0.1299, over 5653377.01 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1204, over 5703129.46 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3794, pruned_loss=0.1304, over 5640838.49 frames. ], batch size: 112, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:35:24,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 2.030e+03 2.819e+03 3.897e+03 8.025e+03, threshold=5.637e+03, percent-clipped=10.0
+2023-03-07 09:35:53,853 INFO [train.py:968] (0/2) Epoch 14, batch 26550, giga_loss[loss=0.2716, simple_loss=0.3436, pruned_loss=0.09986, over 29010.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3796, pruned_loss=0.1311, over 5658229.85 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1204, over 5707212.17 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3806, pruned_loss=0.1317, over 5643617.38 frames. ], batch size: 128, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:36:40,829 INFO [train.py:968] (0/2) Epoch 14, batch 26600, giga_loss[loss=0.2855, simple_loss=0.3535, pruned_loss=0.1088, over 28645.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3772, pruned_loss=0.1298, over 5661183.88 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3683, pruned_loss=0.1202, over 5708776.73 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3785, pruned_loss=0.1306, over 5647671.84 frames. ], batch size: 242, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:36:57,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.742e+03 2.045e+03 2.817e+03 6.233e+03, threshold=4.091e+03, percent-clipped=1.0
+2023-03-07 09:37:09,478 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3742, 1.7532, 1.3861, 1.3263], device='cuda:0'), covar=tensor([0.2355, 0.2263, 0.2569, 0.2012], device='cuda:0'), in_proj_covar=tensor([0.1364, 0.1004, 0.1207, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 09:37:21,744 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 09:37:21,881 INFO [train.py:968] (0/2) Epoch 14, batch 26650, giga_loss[loss=0.2939, simple_loss=0.3627, pruned_loss=0.1125, over 28616.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3754, pruned_loss=0.1284, over 5670294.84 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1206, over 5703623.49 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3761, pruned_loss=0.1289, over 5663287.54 frames. ], batch size: 242, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:38:12,111 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-620000.pt
+2023-03-07 09:38:12,413 INFO [train.py:968] (0/2) Epoch 14, batch 26700, giga_loss[loss=0.2914, simple_loss=0.3545, pruned_loss=0.1142, over 28914.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3751, pruned_loss=0.1279, over 5672444.77 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1206, over 5706614.58 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3758, pruned_loss=0.1286, over 5663240.75 frames. ], batch size: 99, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:38:18,565 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 09:38:29,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.227e+02 1.637e+03 1.953e+03 2.691e+03 6.817e+03, threshold=3.907e+03, percent-clipped=7.0
+2023-03-07 09:38:40,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:38:41,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9417, 1.9318, 1.4662, 1.6637], device='cuda:0'), covar=tensor([0.0744, 0.0505, 0.0930, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0445, 0.0506, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:38:50,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:38:59,053 INFO [train.py:968] (0/2) Epoch 14, batch 26750, giga_loss[loss=0.342, simple_loss=0.3936, pruned_loss=0.1452, over 28642.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3768, pruned_loss=0.1282, over 5669852.65 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3687, pruned_loss=0.1205, over 5706789.32 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3778, pruned_loss=0.129, over 5661869.17 frames. ], batch size: 92, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:39:41,841 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:39:51,301 INFO [train.py:968] (0/2) Epoch 14, batch 26800, giga_loss[loss=0.343, simple_loss=0.3765, pruned_loss=0.1547, over 23404.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3774, pruned_loss=0.1287, over 5660330.74 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.369, pruned_loss=0.1208, over 5711740.15 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3782, pruned_loss=0.1293, over 5648956.07 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:39:58,666 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:40:09,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.622e+02 1.567e+03 1.899e+03 2.515e+03 5.077e+03, threshold=3.799e+03, percent-clipped=5.0
+2023-03-07 09:40:36,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:40:38,126 INFO [train.py:968] (0/2) Epoch 14, batch 26850, giga_loss[loss=0.3321, simple_loss=0.3961, pruned_loss=0.1341, over 28697.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3784, pruned_loss=0.129, over 5664691.33 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1206, over 5710074.96 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3791, pruned_loss=0.1297, over 5656507.84 frames. ], batch size: 262, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:40:54,105 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:41:06,672 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:41:09,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:41:20,701 INFO [train.py:968] (0/2) Epoch 14, batch 26900, giga_loss[loss=0.3638, simple_loss=0.3873, pruned_loss=0.1701, over 23737.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3792, pruned_loss=0.1268, over 5675859.62 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3692, pruned_loss=0.1208, over 5712748.67 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3798, pruned_loss=0.1274, over 5665521.69 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:41:33,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2257, 1.6800, 1.4714, 1.6002], device='cuda:0'), covar=tensor([0.0773, 0.0328, 0.0322, 0.0813], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 09:41:36,193 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:41:38,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.264e+02 1.431e+03 1.776e+03 2.349e+03 5.351e+03, threshold=3.552e+03, percent-clipped=7.0
+2023-03-07 09:41:49,912 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:41:55,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:42:06,581 INFO [train.py:968] (0/2) Epoch 14, batch 26950, giga_loss[loss=0.3174, simple_loss=0.3999, pruned_loss=0.1174, over 28858.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.382, pruned_loss=0.1272, over 5672435.70 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1212, over 5709064.44 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3825, pruned_loss=0.1275, over 5666060.64 frames. ], batch size: 145, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:42:29,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 09:42:34,421 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 09:42:41,252 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:42:43,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:42:49,866 INFO [train.py:968] (0/2) Epoch 14, batch 27000, giga_loss[loss=0.3171, simple_loss=0.3839, pruned_loss=0.1252, over 28879.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3846, pruned_loss=0.1286, over 5676309.89 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3695, pruned_loss=0.1212, over 5709828.29 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3855, pruned_loss=0.1291, over 5669612.50 frames. ], batch size: 119, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:42:49,871 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 09:42:57,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8875, 3.6020, 3.4716, 1.7560], device='cuda:0'), covar=tensor([0.0826, 0.0985, 0.0928, 0.2315], device='cuda:0'), in_proj_covar=tensor([0.1125, 0.1042, 0.0904, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 09:42:58,437 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2106, simple_loss=0.3165, pruned_loss=0.05238, over 944034.00 frames. 
+2023-03-07 09:42:58,438 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 09:43:15,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.227e+02 1.498e+03 1.956e+03 3.243e+03 6.651e+03, threshold=3.912e+03, percent-clipped=20.0
+2023-03-07 09:43:16,490 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:43:17,216 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:43:44,899 INFO [train.py:968] (0/2) Epoch 14, batch 27050, giga_loss[loss=0.3533, simple_loss=0.4048, pruned_loss=0.1509, over 28021.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3878, pruned_loss=0.1322, over 5662570.58 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3697, pruned_loss=0.1212, over 5699060.84 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3887, pruned_loss=0.1327, over 5665444.55 frames. ], batch size: 412, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:44:37,381 INFO [train.py:968] (0/2) Epoch 14, batch 27100, giga_loss[loss=0.3161, simple_loss=0.377, pruned_loss=0.1276, over 28893.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3889, pruned_loss=0.1336, over 5671190.03 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3702, pruned_loss=0.1217, over 5696639.63 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3894, pruned_loss=0.1337, over 5675170.02 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:44:40,144 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-07 09:44:41,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:44:53,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2275, 1.1338, 1.1369, 1.5034], device='cuda:0'), covar=tensor([0.0782, 0.0357, 0.0339, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 09:44:55,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.448e+02 1.596e+03 1.914e+03 2.493e+03 5.514e+03, threshold=3.828e+03, percent-clipped=5.0
+2023-03-07 09:45:13,320 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:45:27,648 INFO [train.py:968] (0/2) Epoch 14, batch 27150, giga_loss[loss=0.2986, simple_loss=0.366, pruned_loss=0.1156, over 28801.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3877, pruned_loss=0.1335, over 5666524.45 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.37, pruned_loss=0.1216, over 5698772.97 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3884, pruned_loss=0.1338, over 5667636.23 frames. ], batch size: 92, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:45:28,456 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:45:45,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:45:52,528 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.30 vs. limit=5.0
+2023-03-07 09:45:58,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:46:10,146 INFO [train.py:968] (0/2) Epoch 14, batch 27200, giga_loss[loss=0.3461, simple_loss=0.4036, pruned_loss=0.1443, over 28988.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3848, pruned_loss=0.1307, over 5680966.16 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3704, pruned_loss=0.1221, over 5704869.61 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3856, pruned_loss=0.1309, over 5675174.48 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:46:30,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.519e+03 2.150e+03 2.910e+03 6.290e+03, threshold=4.299e+03, percent-clipped=13.0
+2023-03-07 09:46:39,864 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:46:49,673 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:46:51,500 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:46:54,943 INFO [train.py:968] (0/2) Epoch 14, batch 27250, giga_loss[loss=0.3447, simple_loss=0.402, pruned_loss=0.1437, over 27691.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3839, pruned_loss=0.1284, over 5667159.08 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3706, pruned_loss=0.1223, over 5699432.53 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3847, pruned_loss=0.1285, over 5666168.53 frames. ], batch size: 472, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:46:55,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:47:15,235 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 09:47:22,601 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:47:39,767 INFO [train.py:968] (0/2) Epoch 14, batch 27300, giga_loss[loss=0.4137, simple_loss=0.4387, pruned_loss=0.1944, over 27591.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3835, pruned_loss=0.1278, over 5668378.07 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3702, pruned_loss=0.1221, over 5705919.28 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.385, pruned_loss=0.1283, over 5660435.47 frames. ], batch size: 472, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:47:47,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:47:50,280 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:47:51,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:47:52,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:48:01,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.678e+02 1.461e+03 1.918e+03 2.486e+03 4.925e+03, threshold=3.837e+03, percent-clipped=4.0
+2023-03-07 09:48:07,179 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:48:09,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:48:19,804 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:48:28,925 INFO [train.py:968] (0/2) Epoch 14, batch 27350, giga_loss[loss=0.3071, simple_loss=0.3743, pruned_loss=0.12, over 28252.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3839, pruned_loss=0.1284, over 5657485.18 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3697, pruned_loss=0.1219, over 5703140.78 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.386, pruned_loss=0.1292, over 5652652.87 frames. ], batch size: 65, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:48:36,031 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:48:56,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5662, 1.1219, 4.6061, 3.5754], device='cuda:0'), covar=tensor([0.1631, 0.2775, 0.0378, 0.0867], device='cuda:0'), in_proj_covar=tensor([0.0695, 0.0612, 0.0895, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:49:05,457 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:49:08,399 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:49:12,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3266, 1.6311, 1.3400, 1.4804], device='cuda:0'), covar=tensor([0.0733, 0.0327, 0.0317, 0.0782], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0113, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 09:49:12,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:49:14,935 INFO [train.py:968] (0/2) Epoch 14, batch 27400, giga_loss[loss=0.3025, simple_loss=0.3722, pruned_loss=0.1164, over 28964.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3833, pruned_loss=0.1283, over 5661874.49 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.37, pruned_loss=0.122, over 5705023.68 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.385, pruned_loss=0.129, over 5655697.51 frames. ], batch size: 164, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:49:16,254 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:49:34,584 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:49:34,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.785e+03 2.404e+03 3.515e+03 1.086e+04, threshold=4.807e+03, percent-clipped=17.0
+2023-03-07 09:49:42,090 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:03,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:04,578 INFO [train.py:968] (0/2) Epoch 14, batch 27450, giga_loss[loss=0.2861, simple_loss=0.3535, pruned_loss=0.1094, over 29064.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.381, pruned_loss=0.1277, over 5665664.58 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3692, pruned_loss=0.1214, over 5706755.34 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3833, pruned_loss=0.129, over 5658004.26 frames. ], batch size: 128, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:50:06,344 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:07,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:10,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:34,932 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:41,830 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:55,385 INFO [train.py:968] (0/2) Epoch 14, batch 27500, giga_loss[loss=0.292, simple_loss=0.3659, pruned_loss=0.109, over 29087.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3791, pruned_loss=0.1269, over 5673058.55 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3691, pruned_loss=0.1213, over 5708561.62 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3812, pruned_loss=0.128, over 5665227.47 frames. ], batch size: 128, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:51:11,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:51:17,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.614e+02 1.816e+03 2.427e+03 3.173e+03 9.655e+03, threshold=4.853e+03, percent-clipped=5.0
+2023-03-07 09:51:20,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3809, 4.2057, 3.9929, 1.9612], device='cuda:0'), covar=tensor([0.0591, 0.0712, 0.0713, 0.1984], device='cuda:0'), in_proj_covar=tensor([0.1133, 0.1055, 0.0916, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 09:51:23,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:51:34,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3556, 3.6107, 1.4320, 1.5111], device='cuda:0'), covar=tensor([0.1028, 0.0326, 0.0940, 0.1403], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0531, 0.0356, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 09:51:34,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:51:39,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:51:45,728 INFO [train.py:968] (0/2) Epoch 14, batch 27550, giga_loss[loss=0.2906, simple_loss=0.3612, pruned_loss=0.1099, over 28724.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3765, pruned_loss=0.1257, over 5670437.66 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3685, pruned_loss=0.1208, over 5711652.77 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3788, pruned_loss=0.1271, over 5660645.71 frames. ], batch size: 242, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:52:07,145 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:52:31,168 INFO [train.py:968] (0/2) Epoch 14, batch 27600, libri_loss[loss=0.3155, simple_loss=0.3816, pruned_loss=0.1247, over 29241.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3766, pruned_loss=0.1269, over 5641896.72 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3691, pruned_loss=0.1212, over 5684884.63 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3781, pruned_loss=0.1279, over 5658337.24 frames. ], batch size: 97, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:52:35,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:52:37,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3299, 1.4296, 1.4548, 1.4996], device='cuda:0'), covar=tensor([0.0680, 0.0359, 0.0293, 0.0727], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 09:52:47,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.429e+02 1.555e+03 2.155e+03 2.940e+03 8.882e+03, threshold=4.309e+03, percent-clipped=6.0
+2023-03-07 09:53:12,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1426, 1.7382, 5.3460, 3.8064], device='cuda:0'), covar=tensor([0.1413, 0.2425, 0.0361, 0.0748], device='cuda:0'), in_proj_covar=tensor([0.0696, 0.0611, 0.0896, 0.0818], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:53:14,900 INFO [train.py:968] (0/2) Epoch 14, batch 27650, giga_loss[loss=0.2867, simple_loss=0.3565, pruned_loss=0.1084, over 28912.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3761, pruned_loss=0.127, over 5649011.59 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1213, over 5690793.20 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3774, pruned_loss=0.1278, over 5656030.52 frames. ], batch size: 186, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:53:18,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:22,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:22,629 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:31,312 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:33,561 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:50,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:59,856 INFO [train.py:968] (0/2) Epoch 14, batch 27700, giga_loss[loss=0.2545, simple_loss=0.3357, pruned_loss=0.08667, over 28926.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3731, pruned_loss=0.1232, over 5657864.64 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3692, pruned_loss=0.1211, over 5692194.85 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3743, pruned_loss=0.124, over 5661487.27 frames. ], batch size: 213, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:54:02,361 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:54:20,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.721e+02 1.412e+03 1.809e+03 3.038e+03 1.011e+04, threshold=3.618e+03, percent-clipped=14.0
+2023-03-07 09:54:47,100 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:54:47,961 INFO [train.py:968] (0/2) Epoch 14, batch 27750, giga_loss[loss=0.3536, simple_loss=0.4069, pruned_loss=0.1501, over 28753.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3703, pruned_loss=0.1209, over 5660790.74 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3687, pruned_loss=0.121, over 5697854.63 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3717, pruned_loss=0.1217, over 5657681.54 frames. ], batch size: 262, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:54:49,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:54:54,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2970, 1.2932, 3.4538, 3.2294], device='cuda:0'), covar=tensor([0.1371, 0.2589, 0.0395, 0.1100], device='cuda:0'), in_proj_covar=tensor([0.0692, 0.0610, 0.0892, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:55:13,133 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=621076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:55:19,583 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:55:36,267 INFO [train.py:968] (0/2) Epoch 14, batch 27800, giga_loss[loss=0.3191, simple_loss=0.3821, pruned_loss=0.128, over 28599.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3682, pruned_loss=0.1194, over 5662493.61 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3679, pruned_loss=0.1205, over 5701298.19 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.37, pruned_loss=0.1205, over 5656449.37 frames. ], batch size: 336, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:55:39,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=621102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:55:55,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.603e+02 1.446e+03 2.221e+03 3.838e+03 1.204e+04, threshold=4.443e+03, percent-clipped=24.0
+2023-03-07 09:56:10,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4085, 3.2476, 3.1092, 1.9967], device='cuda:0'), covar=tensor([0.0743, 0.0853, 0.0790, 0.1660], device='cuda:0'), in_proj_covar=tensor([0.1131, 0.1048, 0.0909, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 09:56:22,914 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=621148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:56:24,435 INFO [train.py:968] (0/2) Epoch 14, batch 27850, giga_loss[loss=0.268, simple_loss=0.3486, pruned_loss=0.09367, over 28864.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3675, pruned_loss=0.1201, over 5661606.92 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3682, pruned_loss=0.1206, over 5708988.79 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3688, pruned_loss=0.1209, over 5648361.49 frames. ], batch size: 174, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:56:52,995 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2560, 1.3282, 1.1475, 1.0133], device='cuda:0'), covar=tensor([0.0875, 0.0473, 0.1033, 0.0924], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0439, 0.0502, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 09:57:16,169 INFO [train.py:968] (0/2) Epoch 14, batch 27900, giga_loss[loss=0.2631, simple_loss=0.3383, pruned_loss=0.09393, over 29030.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3657, pruned_loss=0.1196, over 5656954.52 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3686, pruned_loss=0.121, over 5703981.74 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3663, pruned_loss=0.1198, over 5650080.11 frames. ], batch size: 155, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:57:33,601 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:57:35,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:57:35,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.710e+02 1.647e+03 2.172e+03 3.275e+03 8.281e+03, threshold=4.345e+03, percent-clipped=11.0
+2023-03-07 09:57:38,849 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1423, 3.9329, 3.7409, 2.1892], device='cuda:0'), covar=tensor([0.0703, 0.0945, 0.0944, 0.1877], device='cuda:0'), in_proj_covar=tensor([0.1131, 0.1047, 0.0909, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 09:57:53,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:57:56,830 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:57:57,914 INFO [train.py:968] (0/2) Epoch 14, batch 27950, libri_loss[loss=0.3545, simple_loss=0.4133, pruned_loss=0.1478, over 29200.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3683, pruned_loss=0.1206, over 5665746.06 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3692, pruned_loss=0.1214, over 5708055.62 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3682, pruned_loss=0.1203, over 5654579.49 frames. ], batch size: 101, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:57:58,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:58:24,439 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:58:45,951 INFO [train.py:968] (0/2) Epoch 14, batch 28000, giga_loss[loss=0.3035, simple_loss=0.3729, pruned_loss=0.1171, over 28757.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3713, pruned_loss=0.1223, over 5643924.29 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3697, pruned_loss=0.1218, over 5698328.62 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3707, pruned_loss=0.1218, over 5642364.66 frames. ], batch size: 242, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:59:09,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.501e+03 1.882e+03 2.534e+03 4.566e+03, threshold=3.765e+03, percent-clipped=2.0
+2023-03-07 09:59:20,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=621334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:59:35,801 INFO [train.py:968] (0/2) Epoch 14, batch 28050, giga_loss[loss=0.3029, simple_loss=0.369, pruned_loss=0.1184, over 28897.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3718, pruned_loss=0.1225, over 5646678.38 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3703, pruned_loss=0.1222, over 5691350.47 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3708, pruned_loss=0.1216, over 5651223.35 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:00:24,184 INFO [train.py:968] (0/2) Epoch 14, batch 28100, giga_loss[loss=0.2898, simple_loss=0.3571, pruned_loss=0.1112, over 28947.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3719, pruned_loss=0.1234, over 5639333.71 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3701, pruned_loss=0.122, over 5693701.39 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3714, pruned_loss=0.1229, over 5640280.89 frames. ], batch size: 136, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:00:44,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.206e+02 1.629e+03 2.075e+03 2.615e+03 8.848e+03, threshold=4.149e+03, percent-clipped=12.0
+2023-03-07 10:01:05,980 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=621449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:01:06,570 INFO [train.py:968] (0/2) Epoch 14, batch 28150, giga_loss[loss=0.3065, simple_loss=0.3759, pruned_loss=0.1186, over 28961.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3729, pruned_loss=0.1241, over 5634994.31 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3705, pruned_loss=0.1222, over 5690954.88 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3722, pruned_loss=0.1236, over 5636270.68 frames. ], batch size: 213, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:01:28,175 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-07 10:01:32,280 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:01:34,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:01:54,947 INFO [train.py:968] (0/2) Epoch 14, batch 28200, giga_loss[loss=0.3089, simple_loss=0.3785, pruned_loss=0.1197, over 28797.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3754, pruned_loss=0.1256, over 5643586.33 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3706, pruned_loss=0.1222, over 5691858.05 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3749, pruned_loss=0.1253, over 5643172.93 frames. ], batch size: 119, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:02:02,844 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:02:15,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=621523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:02:15,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.590e+02 1.709e+03 2.177e+03 3.269e+03 1.406e+04, threshold=4.353e+03, percent-clipped=12.0
+2023-03-07 10:02:43,671 INFO [train.py:968] (0/2) Epoch 14, batch 28250, giga_loss[loss=0.3511, simple_loss=0.4114, pruned_loss=0.1454, over 29005.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.377, pruned_loss=0.1271, over 5644872.15 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3704, pruned_loss=0.1219, over 5695482.56 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3768, pruned_loss=0.1272, over 5640326.79 frames. ], batch size: 106, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:03:14,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2860, 1.5903, 1.2972, 0.9841], device='cuda:0'), covar=tensor([0.2235, 0.2184, 0.2392, 0.1961], device='cuda:0'), in_proj_covar=tensor([0.1369, 0.1009, 0.1212, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 10:03:29,080 INFO [train.py:968] (0/2) Epoch 14, batch 28300, giga_loss[loss=0.2998, simple_loss=0.3563, pruned_loss=0.1217, over 28577.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3779, pruned_loss=0.1278, over 5655443.71 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3708, pruned_loss=0.1221, over 5699533.03 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3775, pruned_loss=0.1278, over 5647289.35 frames. ], batch size: 85, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:03:54,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.890e+02 1.546e+03 2.196e+03 3.136e+03 1.049e+04, threshold=4.392e+03, percent-clipped=9.0
+2023-03-07 10:04:19,163 INFO [train.py:968] (0/2) Epoch 14, batch 28350, giga_loss[loss=0.3123, simple_loss=0.3745, pruned_loss=0.125, over 28398.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3777, pruned_loss=0.127, over 5657583.45 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3711, pruned_loss=0.122, over 5705699.81 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3774, pruned_loss=0.1272, over 5644253.38 frames. ], batch size: 71, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:04:37,424 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:04:39,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:05:10,560 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:05:11,591 INFO [train.py:968] (0/2) Epoch 14, batch 28400, giga_loss[loss=0.356, simple_loss=0.4025, pruned_loss=0.1547, over 26769.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3773, pruned_loss=0.1254, over 5651021.67 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3709, pruned_loss=0.1219, over 5701370.79 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3773, pruned_loss=0.1258, over 5642571.16 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 10:05:30,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.613e+03 2.045e+03 2.638e+03 8.956e+03, threshold=4.091e+03, percent-clipped=5.0
+2023-03-07 10:05:53,842 INFO [train.py:968] (0/2) Epoch 14, batch 28450, giga_loss[loss=0.3294, simple_loss=0.391, pruned_loss=0.1339, over 28740.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3771, pruned_loss=0.1262, over 5650137.88 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3706, pruned_loss=0.122, over 5705115.40 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3777, pruned_loss=0.1267, over 5637758.44 frames. ], batch size: 284, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 10:06:07,605 WARNING [optim.py:389] (0/2) Scaling gradients by 0.09737467765808105, model_norm_threshold=4090.94189453125
+2023-03-07 10:06:07,687 INFO [optim.py:451] (0/2) Parameter Dominanting tot_sumsq module.encoder.encoder_embed.conv.0.weight with proportion 1.00, where dominant_sumsq=(grad_sumsq*orig_rms_sq)=1.760e+09, grad_sumsq = 4.135e+10, orig_rms_sq=4.257e-02
+2023-03-07 10:06:37,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4655, 3.0233, 1.5789, 1.5914], device='cuda:0'), covar=tensor([0.0888, 0.0332, 0.0819, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0531, 0.0355, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 10:06:43,496 INFO [train.py:968] (0/2) Epoch 14, batch 28500, giga_loss[loss=0.3551, simple_loss=0.3846, pruned_loss=0.1628, over 23746.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3775, pruned_loss=0.1274, over 5612003.84 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1227, over 5675026.31 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3777, pruned_loss=0.1273, over 5625639.97 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:07:12,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=621824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:07:14,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.392e+02 1.627e+03 2.592e+03 4.301e+03 4.201e+04, threshold=5.185e+03, percent-clipped=28.0
+2023-03-07 10:07:41,375 INFO [train.py:968] (0/2) Epoch 14, batch 28550, giga_loss[loss=0.2539, simple_loss=0.3274, pruned_loss=0.09023, over 28435.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3768, pruned_loss=0.1278, over 5616165.05 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3709, pruned_loss=0.1226, over 5679354.94 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3773, pruned_loss=0.1279, over 5622046.94 frames. ], batch size: 65, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:07:49,919 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=621859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:08:30,436 INFO [train.py:968] (0/2) Epoch 14, batch 28600, giga_loss[loss=0.3116, simple_loss=0.373, pruned_loss=0.1251, over 28294.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3764, pruned_loss=0.128, over 5620619.54 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3714, pruned_loss=0.1229, over 5677486.16 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3766, pruned_loss=0.1279, over 5625105.56 frames. ], batch size: 368, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:08:52,457 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.304e+02 1.538e+03 1.873e+03 2.291e+03 5.667e+03, threshold=3.747e+03, percent-clipped=1.0
+2023-03-07 10:08:57,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1465, 2.3391, 2.3141, 1.8828], device='cuda:0'), covar=tensor([0.1589, 0.2022, 0.1206, 0.1460], device='cuda:0'), in_proj_covar=tensor([0.0849, 0.0696, 0.0893, 0.0794], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 10:09:05,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4947, 1.9043, 1.8228, 1.5116], device='cuda:0'), covar=tensor([0.2033, 0.1567, 0.1550, 0.1709], device='cuda:0'), in_proj_covar=tensor([0.1804, 0.1733, 0.1662, 0.1785], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 10:09:13,125 INFO [train.py:968] (0/2) Epoch 14, batch 28650, giga_loss[loss=0.301, simple_loss=0.3649, pruned_loss=0.1185, over 28897.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3752, pruned_loss=0.1271, over 5632481.94 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.1229, over 5668228.35 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3752, pruned_loss=0.1272, over 5641903.05 frames. ], batch size: 112, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:09:24,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7146, 2.6626, 1.6424, 0.9131], device='cuda:0'), covar=tensor([0.7029, 0.2917, 0.3629, 0.6031], device='cuda:0'), in_proj_covar=tensor([0.1627, 0.1555, 0.1529, 0.1332], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 10:09:24,934 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-07 10:09:30,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4254, 1.6367, 1.3116, 1.2904], device='cuda:0'), covar=tensor([0.2481, 0.2530, 0.2799, 0.2241], device='cuda:0'), in_proj_covar=tensor([0.1368, 0.1008, 0.1212, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 10:09:31,579 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:09:35,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:10:01,154 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:10:01,587 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-622000.pt
+2023-03-07 10:10:01,876 INFO [train.py:968] (0/2) Epoch 14, batch 28700, giga_loss[loss=0.2718, simple_loss=0.3498, pruned_loss=0.09688, over 28923.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3753, pruned_loss=0.1277, over 5634819.47 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3723, pruned_loss=0.1234, over 5669534.98 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3748, pruned_loss=0.1274, over 5640148.25 frames. ], batch size: 174, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:10:26,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.491e+02 1.536e+03 1.994e+03 2.676e+03 6.863e+03, threshold=3.989e+03, percent-clipped=6.0
+2023-03-07 10:10:47,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3456, 3.3131, 1.4131, 1.6445], device='cuda:0'), covar=tensor([0.0980, 0.0360, 0.0885, 0.1264], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0527, 0.0353, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 10:10:49,693 INFO [train.py:968] (0/2) Epoch 14, batch 28750, giga_loss[loss=0.3387, simple_loss=0.3928, pruned_loss=0.1422, over 28804.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3764, pruned_loss=0.1286, over 5646253.33 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3721, pruned_loss=0.1233, over 5671818.20 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3762, pruned_loss=0.1285, over 5647819.28 frames. ], batch size: 112, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:11:34,635 INFO [train.py:968] (0/2) Epoch 14, batch 28800, giga_loss[loss=0.3587, simple_loss=0.4095, pruned_loss=0.154, over 28245.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3762, pruned_loss=0.1283, over 5653117.30 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.123, over 5674717.66 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3765, pruned_loss=0.1287, over 5651409.20 frames. ], batch size: 368, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:11:57,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.677e+03 2.357e+03 3.486e+03 9.508e+03, threshold=4.714e+03, percent-clipped=17.0
+2023-03-07 10:12:22,368 INFO [train.py:968] (0/2) Epoch 14, batch 28850, giga_loss[loss=0.3051, simple_loss=0.3696, pruned_loss=0.1203, over 28988.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3772, pruned_loss=0.1291, over 5657650.27 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.372, pruned_loss=0.1233, over 5677190.68 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3773, pruned_loss=0.1293, over 5653816.48 frames. ], batch size: 186, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:12:34,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2218, 1.3178, 1.1418, 0.9803], device='cuda:0'), covar=tensor([0.0803, 0.0449, 0.0994, 0.0941], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0441, 0.0504, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 10:13:08,631 INFO [train.py:968] (0/2) Epoch 14, batch 28900, giga_loss[loss=0.3404, simple_loss=0.3943, pruned_loss=0.1433, over 28581.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3782, pruned_loss=0.13, over 5658777.08 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3721, pruned_loss=0.1233, over 5670618.19 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3783, pruned_loss=0.1303, over 5660858.42 frames. ], batch size: 336, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:13:33,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.641e+03 2.114e+03 3.405e+03 7.462e+03, threshold=4.227e+03, percent-clipped=7.0
+2023-03-07 10:13:33,269 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=622226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:13:38,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=622234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:13:53,588 INFO [train.py:968] (0/2) Epoch 14, batch 28950, giga_loss[loss=0.2743, simple_loss=0.3509, pruned_loss=0.09885, over 28816.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3778, pruned_loss=0.1297, over 5652652.38 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3722, pruned_loss=0.1234, over 5662313.25 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3778, pruned_loss=0.1299, over 5661373.40 frames. ], batch size: 242, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:14:41,966 INFO [train.py:968] (0/2) Epoch 14, batch 29000, giga_loss[loss=0.3185, simple_loss=0.3819, pruned_loss=0.1276, over 28946.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3786, pruned_loss=0.1296, over 5666108.11 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3721, pruned_loss=0.1233, over 5666205.94 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3789, pruned_loss=0.13, over 5669468.92 frames. ], batch size: 136, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:15:07,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1358, 3.9604, 3.7454, 1.8035], device='cuda:0'), covar=tensor([0.0608, 0.0745, 0.0724, 0.2156], device='cuda:0'), in_proj_covar=tensor([0.1134, 0.1053, 0.0912, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 10:15:08,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.830e+02 1.601e+03 2.128e+03 3.169e+03 9.125e+03, threshold=4.255e+03, percent-clipped=11.0
+2023-03-07 10:15:31,170 INFO [train.py:968] (0/2) Epoch 14, batch 29050, giga_loss[loss=0.3541, simple_loss=0.4041, pruned_loss=0.152, over 28759.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3795, pruned_loss=0.1306, over 5664989.28 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3715, pruned_loss=0.123, over 5668778.01 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3803, pruned_loss=0.1313, over 5665295.27 frames. ], batch size: 284, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:15:57,431 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=622377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:15:59,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=622380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:16:01,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7893, 3.6169, 3.4421, 1.7814], device='cuda:0'), covar=tensor([0.0703, 0.0836, 0.0802, 0.2415], device='cuda:0'), in_proj_covar=tensor([0.1138, 0.1056, 0.0916, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 10:16:03,678 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-07 10:16:18,806 INFO [train.py:968] (0/2) Epoch 14, batch 29100, giga_loss[loss=0.3614, simple_loss=0.4063, pruned_loss=0.1582, over 28689.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3807, pruned_loss=0.1315, over 5669639.58 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.123, over 5672238.86 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3814, pruned_loss=0.1321, over 5666777.12 frames. ], batch size: 262, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:16:27,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=622409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:16:41,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.897e+02 1.584e+03 2.011e+03 2.897e+03 8.472e+03, threshold=4.021e+03, percent-clipped=13.0
+2023-03-07 10:17:06,072 INFO [train.py:968] (0/2) Epoch 14, batch 29150, giga_loss[loss=0.3298, simple_loss=0.3828, pruned_loss=0.1384, over 28977.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3823, pruned_loss=0.1329, over 5657565.37 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3719, pruned_loss=0.1232, over 5663912.63 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3826, pruned_loss=0.1333, over 5662112.36 frames. ], batch size: 213, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:17:28,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3826, 1.7282, 1.3731, 1.3547], device='cuda:0'), covar=tensor([0.2562, 0.2387, 0.2716, 0.2072], device='cuda:0'), in_proj_covar=tensor([0.1369, 0.1008, 0.1211, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 10:17:32,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4110, 1.4846, 1.5858, 1.4143], device='cuda:0'), covar=tensor([0.1212, 0.1077, 0.1568, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0746, 0.0696, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 10:17:40,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2866, 1.5141, 1.2960, 1.4008], device='cuda:0'), covar=tensor([0.1600, 0.1665, 0.2060, 0.1640], device='cuda:0'), in_proj_covar=tensor([0.0452, 0.0746, 0.0696, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 10:17:52,567 INFO [train.py:968] (0/2) Epoch 14, batch 29200, libri_loss[loss=0.259, simple_loss=0.324, pruned_loss=0.097, over 29668.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3821, pruned_loss=0.1327, over 5655159.81 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3719, pruned_loss=0.1231, over 5665607.17 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3826, pruned_loss=0.1332, over 5657005.61 frames. ], batch size: 69, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:18:18,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.581e+03 1.920e+03 2.672e+03 6.531e+03, threshold=3.840e+03, percent-clipped=7.0
+2023-03-07 10:18:47,735 INFO [train.py:968] (0/2) Epoch 14, batch 29250, giga_loss[loss=0.3504, simple_loss=0.4212, pruned_loss=0.1397, over 29009.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3826, pruned_loss=0.1322, over 5641284.94 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.372, pruned_loss=0.1232, over 5666785.13 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3829, pruned_loss=0.1326, over 5641731.98 frames. ], batch size: 145, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:19:10,137 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=622573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:19:36,766 INFO [train.py:968] (0/2) Epoch 14, batch 29300, giga_loss[loss=0.2958, simple_loss=0.3541, pruned_loss=0.1187, over 28619.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3817, pruned_loss=0.131, over 5644871.18 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1233, over 5670627.36 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3825, pruned_loss=0.1315, over 5640971.67 frames. ], batch size: 85, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:19:37,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=622601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:20:00,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.077e+02 1.681e+03 1.985e+03 2.797e+03 6.763e+03, threshold=3.970e+03, percent-clipped=10.0
+2023-03-07 10:20:22,771 INFO [train.py:968] (0/2) Epoch 14, batch 29350, giga_loss[loss=0.306, simple_loss=0.3718, pruned_loss=0.1201, over 28953.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3804, pruned_loss=0.1297, over 5646113.46 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1234, over 5663766.03 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3812, pruned_loss=0.1302, over 5649228.22 frames. ], batch size: 227, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:20:25,819 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-07 10:20:38,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6558, 2.0426, 1.7900, 1.4904], device='cuda:0'), covar=tensor([0.2809, 0.2043, 0.2242, 0.2434], device='cuda:0'), in_proj_covar=tensor([0.1804, 0.1728, 0.1661, 0.1792], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 10:21:09,369 INFO [train.py:968] (0/2) Epoch 14, batch 29400, giga_loss[loss=0.2927, simple_loss=0.3643, pruned_loss=0.1106, over 29065.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3792, pruned_loss=0.1292, over 5652281.54 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3716, pruned_loss=0.1232, over 5667408.17 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.38, pruned_loss=0.1297, over 5651397.84 frames. ], batch size: 128, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:21:35,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.088e+03 1.647e+03 1.997e+03 2.964e+03 6.165e+03, threshold=3.994e+03, percent-clipped=8.0
+2023-03-07 10:21:53,345 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=622744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:21:55,914 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=622747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:21:57,587 INFO [train.py:968] (0/2) Epoch 14, batch 29450, libri_loss[loss=0.3208, simple_loss=0.3841, pruned_loss=0.1288, over 29514.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.381, pruned_loss=0.1306, over 5638896.31 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1237, over 5658204.76 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3812, pruned_loss=0.1308, over 5646462.29 frames. ], batch size: 89, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:22:26,236 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=622776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:22:46,890 INFO [train.py:968] (0/2) Epoch 14, batch 29500, giga_loss[loss=0.2789, simple_loss=0.3388, pruned_loss=0.1095, over 28543.00 frames. ], tot_loss[loss=0.321, simple_loss=0.381, pruned_loss=0.1305, over 5652492.57 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1237, over 5661823.30 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3813, pruned_loss=0.1307, over 5655082.26 frames. ], batch size: 85, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:23:15,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.726e+02 1.477e+03 2.010e+03 2.691e+03 7.805e+03, threshold=4.021e+03, percent-clipped=12.0
+2023-03-07 10:23:39,770 INFO [train.py:968] (0/2) Epoch 14, batch 29550, giga_loss[loss=0.3492, simple_loss=0.4006, pruned_loss=0.1489, over 28596.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3798, pruned_loss=0.1306, over 5651113.45 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1237, over 5661823.30 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.38, pruned_loss=0.1308, over 5653129.03 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:24:04,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-07 10:24:22,430 INFO [train.py:968] (0/2) Epoch 14, batch 29600, giga_loss[loss=0.3973, simple_loss=0.435, pruned_loss=0.1798, over 28597.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3796, pruned_loss=0.13, over 5667904.76 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3722, pruned_loss=0.1234, over 5672609.51 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3802, pruned_loss=0.1307, over 5659759.76 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 10:24:27,865 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-07 10:24:48,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.019e+02 1.465e+03 1.730e+03 2.961e+03 1.131e+04, threshold=3.461e+03, percent-clipped=7.0
+2023-03-07 10:24:49,276 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=622927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:25:10,399 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=622948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:25:11,360 INFO [train.py:968] (0/2) Epoch 14, batch 29650, giga_loss[loss=0.3093, simple_loss=0.3729, pruned_loss=0.1229, over 28855.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3809, pruned_loss=0.1311, over 5671455.01 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3718, pruned_loss=0.1232, over 5676221.35 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3819, pruned_loss=0.1319, over 5661645.64 frames. ], batch size: 174, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 10:26:01,054 INFO [train.py:968] (0/2) Epoch 14, batch 29700, giga_loss[loss=0.3264, simple_loss=0.3807, pruned_loss=0.136, over 28552.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3799, pruned_loss=0.1307, over 5659218.77 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3712, pruned_loss=0.1228, over 5682227.51 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3813, pruned_loss=0.1319, over 5645970.09 frames. ], batch size: 336, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 10:26:19,371 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 10:26:26,991 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.973e+02 1.548e+03 1.991e+03 2.709e+03 6.243e+03, threshold=3.982e+03, percent-clipped=13.0
+2023-03-07 10:26:46,265 INFO [train.py:968] (0/2) Epoch 14, batch 29750, giga_loss[loss=0.3358, simple_loss=0.3942, pruned_loss=0.1387, over 28769.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.381, pruned_loss=0.1314, over 5648637.35 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.1231, over 5676158.64 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.382, pruned_loss=0.1322, over 5643444.54 frames. ], batch size: 284, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:26:57,407 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-07 10:27:27,050 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:27:29,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=623094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:27:36,564 INFO [train.py:968] (0/2) Epoch 14, batch 29800, giga_loss[loss=0.3188, simple_loss=0.3889, pruned_loss=0.1244, over 28824.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3809, pruned_loss=0.1305, over 5658340.96 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3715, pruned_loss=0.1229, over 5681945.04 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.382, pruned_loss=0.1315, over 5648303.45 frames. ], batch size: 174, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:27:48,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5500, 1.5898, 1.2590, 1.2429], device='cuda:0'), covar=tensor([0.0773, 0.0499, 0.0956, 0.0970], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0438, 0.0504, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 10:28:00,362 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=623123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:28:04,168 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.585e+03 2.116e+03 3.083e+03 5.286e+03, threshold=4.233e+03, percent-clipped=9.0
+2023-03-07 10:28:25,906 INFO [train.py:968] (0/2) Epoch 14, batch 29850, giga_loss[loss=0.284, simple_loss=0.358, pruned_loss=0.105, over 29101.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3792, pruned_loss=0.1286, over 5663142.96 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3716, pruned_loss=0.1229, over 5684657.56 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.38, pruned_loss=0.1295, over 5652417.36 frames. ], batch size: 136, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:28:46,999 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-07 10:28:57,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=623181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:29:15,156 INFO [train.py:968] (0/2) Epoch 14, batch 29900, giga_loss[loss=0.3186, simple_loss=0.3807, pruned_loss=0.1282, over 28697.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3782, pruned_loss=0.1278, over 5661066.82 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3714, pruned_loss=0.1228, over 5677145.98 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3791, pruned_loss=0.1287, over 5658581.18 frames. ], batch size: 92, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:29:40,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.625e+03 1.950e+03 2.713e+03 9.813e+03, threshold=3.900e+03, percent-clipped=6.0
+2023-03-07 10:29:59,153 INFO [train.py:968] (0/2) Epoch 14, batch 29950, giga_loss[loss=0.3029, simple_loss=0.3667, pruned_loss=0.1196, over 27809.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3769, pruned_loss=0.1274, over 5669442.39 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1226, over 5683646.75 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3781, pruned_loss=0.1284, over 5660983.63 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:30:25,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9467, 1.0614, 3.4310, 3.0203], device='cuda:0'), covar=tensor([0.1802, 0.2696, 0.0503, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0701, 0.0615, 0.0896, 0.0820], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 10:30:48,305 INFO [train.py:968] (0/2) Epoch 14, batch 30000, giga_loss[loss=0.3077, simple_loss=0.3493, pruned_loss=0.1331, over 23435.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3719, pruned_loss=0.1245, over 5667010.77 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3706, pruned_loss=0.1223, over 5685328.06 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3734, pruned_loss=0.1257, over 5658561.97 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 10:30:48,310 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 10:30:53,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8402, 3.6078, 3.4412, 1.7348], device='cuda:0'), covar=tensor([0.0758, 0.0949, 0.0877, 0.2497], device='cuda:0'), in_proj_covar=tensor([0.1132, 0.1055, 0.0913, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 10:30:56,784 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2118, simple_loss=0.3195, pruned_loss=0.05203, over 944034.00 frames. 
+2023-03-07 10:30:56,785 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 10:30:58,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=623302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:31:19,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.837e+03 2.256e+03 3.555e+03 9.860e+03, threshold=4.512e+03, percent-clipped=19.0
+2023-03-07 10:31:26,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9436, 2.1958, 2.0160, 1.9895], device='cuda:0'), covar=tensor([0.0679, 0.0252, 0.0260, 0.0743], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 10:31:40,298 INFO [train.py:968] (0/2) Epoch 14, batch 30050, giga_loss[loss=0.2527, simple_loss=0.3219, pruned_loss=0.09174, over 28649.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3689, pruned_loss=0.1238, over 5648229.13 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3706, pruned_loss=0.1223, over 5675258.17 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3701, pruned_loss=0.1247, over 5650724.87 frames. ], batch size: 92, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:32:16,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4874, 1.7725, 1.4020, 1.7312], device='cuda:0'), covar=tensor([0.2279, 0.2357, 0.2659, 0.2070], device='cuda:0'), in_proj_covar=tensor([0.1370, 0.1006, 0.1213, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 10:32:26,422 INFO [train.py:968] (0/2) Epoch 14, batch 30100, giga_loss[loss=0.3806, simple_loss=0.4091, pruned_loss=0.1761, over 26771.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3673, pruned_loss=0.1233, over 5656845.01 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3703, pruned_loss=0.1221, over 5680658.96 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3685, pruned_loss=0.1243, over 5653666.02 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:32:56,576 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.721e+02 1.759e+03 2.396e+03 3.359e+03 1.332e+04, threshold=4.791e+03, percent-clipped=15.0
+2023-03-07 10:33:14,297 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:33:17,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=623448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:33:18,775 INFO [train.py:968] (0/2) Epoch 14, batch 30150, giga_loss[loss=0.3485, simple_loss=0.4017, pruned_loss=0.1476, over 28883.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3675, pruned_loss=0.1233, over 5641213.12 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1223, over 5682372.04 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3681, pruned_loss=0.1238, over 5636626.79 frames. ], batch size: 199, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:33:42,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=623477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:34:06,289 INFO [train.py:968] (0/2) Epoch 14, batch 30200, giga_loss[loss=0.2727, simple_loss=0.3501, pruned_loss=0.09769, over 28990.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3657, pruned_loss=0.1199, over 5649763.72 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1224, over 5687939.46 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.366, pruned_loss=0.1202, over 5639988.57 frames. ], batch size: 164, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:34:32,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.792e+02 1.429e+03 2.065e+03 2.895e+03 1.445e+04, threshold=4.129e+03, percent-clipped=4.0
+2023-03-07 10:34:35,095 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-07 10:34:53,135 INFO [train.py:968] (0/2) Epoch 14, batch 30250, libri_loss[loss=0.2891, simple_loss=0.3532, pruned_loss=0.1126, over 27811.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3647, pruned_loss=0.118, over 5653216.82 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3704, pruned_loss=0.1225, over 5696457.89 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3649, pruned_loss=0.118, over 5635408.68 frames. ], batch size: 116, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:34:54,711 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-07 10:34:59,132 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=623556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:35:39,375 INFO [train.py:968] (0/2) Epoch 14, batch 30300, giga_loss[loss=0.2648, simple_loss=0.347, pruned_loss=0.09125, over 28796.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3622, pruned_loss=0.1151, over 5658897.88 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3703, pruned_loss=0.1229, over 5692945.20 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3623, pruned_loss=0.1145, over 5647272.21 frames. ], batch size: 284, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:36:11,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.616e+02 1.332e+03 2.009e+03 2.829e+03 9.907e+03, threshold=4.018e+03, percent-clipped=8.0
+2023-03-07 10:36:30,548 INFO [train.py:968] (0/2) Epoch 14, batch 30350, giga_loss[loss=0.2616, simple_loss=0.3434, pruned_loss=0.08983, over 28698.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3591, pruned_loss=0.1119, over 5658906.63 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3697, pruned_loss=0.1225, over 5694879.44 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3596, pruned_loss=0.1117, over 5647700.13 frames. ], batch size: 242, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:36:48,985 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-07 10:37:17,478 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:37:17,867 INFO [train.py:968] (0/2) Epoch 14, batch 30400, giga_loss[loss=0.2458, simple_loss=0.3263, pruned_loss=0.08268, over 28701.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3564, pruned_loss=0.1088, over 5660173.03 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3694, pruned_loss=0.1227, over 5694684.01 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3568, pruned_loss=0.1082, over 5650523.14 frames. ], batch size: 92, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:37:20,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=623702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:37:33,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8682, 1.9921, 1.3634, 1.6971], device='cuda:0'), covar=tensor([0.0700, 0.0458, 0.0929, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0435, 0.0501, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 10:37:48,854 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.734e+02 1.325e+03 2.274e+03 3.057e+03 8.767e+03, threshold=4.548e+03, percent-clipped=12.0
+2023-03-07 10:37:50,698 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=623731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:37:57,455 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4113, 1.5720, 1.1636, 1.1953], device='cuda:0'), covar=tensor([0.0823, 0.0433, 0.0950, 0.1033], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0436, 0.0502, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 10:38:04,674 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=623745.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 10:38:10,735 INFO [train.py:968] (0/2) Epoch 14, batch 30450, giga_loss[loss=0.2515, simple_loss=0.3394, pruned_loss=0.08176, over 28905.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3551, pruned_loss=0.1055, over 5673136.04 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3687, pruned_loss=0.1223, over 5698108.20 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3557, pruned_loss=0.1049, over 5661887.15 frames. ], batch size: 186, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:38:20,439 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-07 10:38:44,942 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 10:39:01,441 INFO [train.py:968] (0/2) Epoch 14, batch 30500, giga_loss[loss=0.2698, simple_loss=0.3498, pruned_loss=0.09489, over 28662.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3544, pruned_loss=0.1052, over 5673752.14 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3673, pruned_loss=0.1216, over 5701757.09 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3557, pruned_loss=0.105, over 5661055.49 frames. ], batch size: 262, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:39:30,858 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.692e+02 1.460e+03 1.920e+03 2.654e+03 1.240e+04, threshold=3.840e+03, percent-clipped=10.0
+2023-03-07 10:39:35,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4129, 1.7823, 1.3837, 1.7201], device='cuda:0'), covar=tensor([0.2526, 0.2195, 0.2669, 0.1915], device='cuda:0'), in_proj_covar=tensor([0.1374, 0.1005, 0.1218, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 10:39:52,765 INFO [train.py:968] (0/2) Epoch 14, batch 30550, giga_loss[loss=0.2877, simple_loss=0.346, pruned_loss=0.1147, over 26686.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3519, pruned_loss=0.1029, over 5673646.60 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3669, pruned_loss=0.1213, over 5703573.30 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3531, pruned_loss=0.1027, over 5661841.25 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:39:57,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=623852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:40:42,515 INFO [train.py:968] (0/2) Epoch 14, batch 30600, giga_loss[loss=0.2579, simple_loss=0.3128, pruned_loss=0.1015, over 24101.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3502, pruned_loss=0.1019, over 5673692.38 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3667, pruned_loss=0.1213, over 5707976.61 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3509, pruned_loss=0.1014, over 5659570.57 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:41:06,805 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-07 10:41:11,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.230e+02 1.337e+03 1.786e+03 2.854e+03 4.803e+03, threshold=3.572e+03, percent-clipped=8.0
+2023-03-07 10:41:29,726 INFO [train.py:968] (0/2) Epoch 14, batch 30650, giga_loss[loss=0.2932, simple_loss=0.3741, pruned_loss=0.1062, over 29018.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3496, pruned_loss=0.1021, over 5670643.79 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.366, pruned_loss=0.1212, over 5712272.56 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3502, pruned_loss=0.1011, over 5654320.77 frames. ], batch size: 164, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:41:39,533 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-07 10:42:06,120 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-07 10:42:17,265 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-624000.pt
+2023-03-07 10:42:17,556 INFO [train.py:968] (0/2) Epoch 14, batch 30700, giga_loss[loss=0.2506, simple_loss=0.3342, pruned_loss=0.08348, over 28853.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3493, pruned_loss=0.1016, over 5677656.04 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3658, pruned_loss=0.1212, over 5715679.40 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3496, pruned_loss=0.1004, over 5660935.54 frames. ], batch size: 199, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:42:45,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.662e+02 1.323e+03 1.833e+03 2.877e+03 6.869e+03, threshold=3.666e+03, percent-clipped=11.0
+2023-03-07 10:42:47,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4665, 1.7785, 1.3896, 1.6256], device='cuda:0'), covar=tensor([0.2614, 0.2297, 0.2630, 0.2058], device='cuda:0'), in_proj_covar=tensor([0.1368, 0.1002, 0.1212, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 10:43:04,761 INFO [train.py:968] (0/2) Epoch 14, batch 30750, giga_loss[loss=0.2457, simple_loss=0.3273, pruned_loss=0.08205, over 28505.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3472, pruned_loss=0.09998, over 5661911.31 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3653, pruned_loss=0.121, over 5706846.63 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3474, pruned_loss=0.09869, over 5655212.48 frames. ], batch size: 336, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:43:27,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4117, 1.4780, 1.3327, 1.6541], device='cuda:0'), covar=tensor([0.0646, 0.0284, 0.0307, 0.0697], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 10:43:56,231 INFO [train.py:968] (0/2) Epoch 14, batch 30800, giga_loss[loss=0.2565, simple_loss=0.3352, pruned_loss=0.08891, over 28548.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3447, pruned_loss=0.09807, over 5666434.03 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3654, pruned_loss=0.1211, over 5709474.33 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3445, pruned_loss=0.09643, over 5658005.87 frames. ], batch size: 336, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:44:17,934 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=624120.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 10:44:25,688 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.262e+02 1.216e+03 1.627e+03 2.220e+03 5.052e+03, threshold=3.254e+03, percent-clipped=6.0
+2023-03-07 10:44:45,697 INFO [train.py:968] (0/2) Epoch 14, batch 30850, giga_loss[loss=0.2601, simple_loss=0.3321, pruned_loss=0.09399, over 28962.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3413, pruned_loss=0.09618, over 5672153.88 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3655, pruned_loss=0.1213, over 5709272.69 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3405, pruned_loss=0.09412, over 5664792.45 frames. ], batch size: 213, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:45:32,910 INFO [train.py:968] (0/2) Epoch 14, batch 30900, giga_loss[loss=0.2827, simple_loss=0.3482, pruned_loss=0.1086, over 26662.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3401, pruned_loss=0.09606, over 5665132.21 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3652, pruned_loss=0.1213, over 5705180.21 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.339, pruned_loss=0.09379, over 5661381.82 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:46:03,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=624227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:46:08,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.962e+02 1.358e+03 1.816e+03 2.629e+03 2.799e+04, threshold=3.632e+03, percent-clipped=14.0
+2023-03-07 10:46:08,396 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 10:46:28,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2172, 2.9653, 1.3927, 1.3332], device='cuda:0'), covar=tensor([0.1143, 0.0440, 0.1053, 0.1521], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0520, 0.0351, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 10:46:29,184 INFO [train.py:968] (0/2) Epoch 14, batch 30950, giga_loss[loss=0.2786, simple_loss=0.3495, pruned_loss=0.1038, over 28342.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3393, pruned_loss=0.09575, over 5648600.36 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3652, pruned_loss=0.1213, over 5705180.21 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3385, pruned_loss=0.09397, over 5645681.39 frames. ], batch size: 368, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:46:41,510 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=624263.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 10:46:45,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=624266.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 10:46:56,406 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-07 10:47:15,559 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=624295.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 10:47:21,372 INFO [train.py:968] (0/2) Epoch 14, batch 31000, giga_loss[loss=0.3099, simple_loss=0.3808, pruned_loss=0.1195, over 28355.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3424, pruned_loss=0.09698, over 5652064.45 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3648, pruned_loss=0.121, over 5710266.72 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3412, pruned_loss=0.09497, over 5643242.99 frames. ], batch size: 368, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:47:58,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.440e+02 1.320e+03 1.768e+03 2.857e+03 7.308e+03, threshold=3.536e+03, percent-clipped=18.0
+2023-03-07 10:48:18,985 INFO [train.py:968] (0/2) Epoch 14, batch 31050, giga_loss[loss=0.3113, simple_loss=0.3886, pruned_loss=0.117, over 28884.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3438, pruned_loss=0.09725, over 5638613.89 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.364, pruned_loss=0.1206, over 5706511.48 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.343, pruned_loss=0.09534, over 5633390.99 frames. ], batch size: 284, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:48:43,805 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=624370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:48:45,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=624373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:49:13,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7123, 3.5205, 3.3418, 1.9175], device='cuda:0'), covar=tensor([0.0737, 0.0911, 0.0938, 0.2439], device='cuda:0'), in_proj_covar=tensor([0.1110, 0.1029, 0.0888, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 10:49:21,933 INFO [train.py:968] (0/2) Epoch 14, batch 31100, libri_loss[loss=0.2627, simple_loss=0.3234, pruned_loss=0.1009, over 29618.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.343, pruned_loss=0.09679, over 5636733.78 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3632, pruned_loss=0.1203, over 5711239.43 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3426, pruned_loss=0.09502, over 5626653.39 frames. ], batch size: 69, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:49:24,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=624402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:49:29,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6476, 3.2008, 2.6090, 2.3390], device='cuda:0'), covar=tensor([0.1607, 0.0944, 0.1253, 0.1186], device='cuda:0'), in_proj_covar=tensor([0.1779, 0.1691, 0.1614, 0.1747], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 10:49:42,359 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-07 10:49:56,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.644e+02 1.428e+03 1.816e+03 2.470e+03 5.325e+03, threshold=3.632e+03, percent-clipped=5.0
+2023-03-07 10:50:21,797 INFO [train.py:968] (0/2) Epoch 14, batch 31150, giga_loss[loss=0.2644, simple_loss=0.3406, pruned_loss=0.09414, over 28927.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3421, pruned_loss=0.09678, over 5642936.77 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3623, pruned_loss=0.1197, over 5707460.47 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3421, pruned_loss=0.09513, over 5635386.55 frames. ], batch size: 213, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:51:21,041 INFO [train.py:968] (0/2) Epoch 14, batch 31200, libri_loss[loss=0.2897, simple_loss=0.3423, pruned_loss=0.1185, over 29568.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3405, pruned_loss=0.09481, over 5645155.75 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3621, pruned_loss=0.1197, over 5712512.86 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.34, pruned_loss=0.09274, over 5632387.75 frames. ], batch size: 77, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:51:56,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.960e+02 1.347e+03 1.780e+03 2.751e+03 9.125e+03, threshold=3.560e+03, percent-clipped=13.0
+2023-03-07 10:52:14,608 INFO [train.py:968] (0/2) Epoch 14, batch 31250, giga_loss[loss=0.2551, simple_loss=0.3358, pruned_loss=0.08723, over 28089.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3375, pruned_loss=0.09225, over 5658506.98 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3605, pruned_loss=0.1188, over 5721917.27 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3375, pruned_loss=0.09004, over 5636134.40 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:53:13,675 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=624596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:53:16,711 INFO [train.py:968] (0/2) Epoch 14, batch 31300, giga_loss[loss=0.2266, simple_loss=0.3022, pruned_loss=0.07549, over 29024.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3349, pruned_loss=0.09136, over 5662173.18 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3601, pruned_loss=0.1187, over 5720876.42 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3348, pruned_loss=0.08936, over 5644797.58 frames. ], batch size: 128, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:53:56,299 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.007e+02 1.420e+03 1.937e+03 2.747e+03 5.923e+03, threshold=3.874e+03, percent-clipped=11.0
+2023-03-07 10:54:16,668 INFO [train.py:968] (0/2) Epoch 14, batch 31350, giga_loss[loss=0.2446, simple_loss=0.3254, pruned_loss=0.08193, over 28076.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3349, pruned_loss=0.0918, over 5673019.84 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3601, pruned_loss=0.119, over 5724330.35 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3341, pruned_loss=0.08911, over 5654514.34 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:54:16,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7546, 1.0877, 2.8670, 2.7825], device='cuda:0'), covar=tensor([0.1788, 0.2601, 0.0567, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0691, 0.0606, 0.0884, 0.0801], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 10:54:22,630 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-07 10:55:09,830 INFO [train.py:968] (0/2) Epoch 14, batch 31400, giga_loss[loss=0.283, simple_loss=0.3474, pruned_loss=0.1093, over 26821.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3343, pruned_loss=0.09161, over 5683389.89 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3588, pruned_loss=0.1183, over 5730728.66 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3338, pruned_loss=0.08896, over 5660617.60 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:55:33,303 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 10:55:46,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.227e+02 1.256e+03 1.752e+03 2.187e+03 4.720e+03, threshold=3.505e+03, percent-clipped=3.0
+2023-03-07 10:55:55,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8377, 4.6801, 4.4255, 2.2490], device='cuda:0'), covar=tensor([0.0513, 0.0661, 0.0757, 0.1883], device='cuda:0'), in_proj_covar=tensor([0.1103, 0.1023, 0.0885, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 10:56:08,620 INFO [train.py:968] (0/2) Epoch 14, batch 31450, giga_loss[loss=0.2265, simple_loss=0.2993, pruned_loss=0.07682, over 24342.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3363, pruned_loss=0.09207, over 5661169.51 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3589, pruned_loss=0.1185, over 5713850.97 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3355, pruned_loss=0.0894, over 5656719.78 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:56:13,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5491, 2.0645, 1.3136, 0.6552], device='cuda:0'), covar=tensor([0.5507, 0.3038, 0.2959, 0.5497], device='cuda:0'), in_proj_covar=tensor([0.1613, 0.1543, 0.1519, 0.1323], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 10:57:16,153 INFO [train.py:968] (0/2) Epoch 14, batch 31500, giga_loss[loss=0.2938, simple_loss=0.3674, pruned_loss=0.1101, over 29182.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3378, pruned_loss=0.09247, over 5647235.42 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3587, pruned_loss=0.1184, over 5702928.68 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3371, pruned_loss=0.0901, over 5652783.06 frames. ], batch size: 113, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:57:32,924 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-07 10:57:49,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.795e+02 1.377e+03 1.984e+03 2.913e+03 7.861e+03, threshold=3.969e+03, percent-clipped=11.0
+2023-03-07 10:57:49,908 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=624831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:58:17,211 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-07 10:58:19,227 INFO [train.py:968] (0/2) Epoch 14, batch 31550, giga_loss[loss=0.3341, simple_loss=0.3837, pruned_loss=0.1423, over 26751.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3344, pruned_loss=0.09013, over 5665348.96 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.358, pruned_loss=0.1179, over 5705607.86 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3338, pruned_loss=0.08802, over 5666236.79 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:59:17,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 10:59:27,227 INFO [train.py:968] (0/2) Epoch 14, batch 31600, giga_loss[loss=0.2544, simple_loss=0.3397, pruned_loss=0.0845, over 28984.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3363, pruned_loss=0.09147, over 5666068.30 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3576, pruned_loss=0.1177, over 5708576.40 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3358, pruned_loss=0.0895, over 5663455.72 frames. ], batch size: 128, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:00:07,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.777e+02 1.350e+03 1.724e+03 2.202e+03 5.731e+03, threshold=3.448e+03, percent-clipped=2.0
+2023-03-07 11:00:27,984 INFO [train.py:968] (0/2) Epoch 14, batch 31650, giga_loss[loss=0.2653, simple_loss=0.3531, pruned_loss=0.08871, over 28100.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3409, pruned_loss=0.09211, over 5671105.73 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3578, pruned_loss=0.1179, over 5710362.08 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3397, pruned_loss=0.08952, over 5666201.36 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:00:51,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=624971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:01:29,055 INFO [train.py:968] (0/2) Epoch 14, batch 31700, giga_loss[loss=0.2609, simple_loss=0.3508, pruned_loss=0.08546, over 28612.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3427, pruned_loss=0.0915, over 5663894.34 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3581, pruned_loss=0.1181, over 5714242.10 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3412, pruned_loss=0.08866, over 5655738.71 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:01:49,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2631, 1.5617, 1.5086, 1.4325], device='cuda:0'), covar=tensor([0.1664, 0.1748, 0.1990, 0.1716], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0723, 0.0676, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 11:02:02,321 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.357e+02 1.547e+03 2.287e+03 3.313e+03 7.568e+03, threshold=4.573e+03, percent-clipped=23.0
+2023-03-07 11:02:24,703 INFO [train.py:968] (0/2) Epoch 14, batch 31750, giga_loss[loss=0.2442, simple_loss=0.3369, pruned_loss=0.07576, over 28471.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3431, pruned_loss=0.09083, over 5671398.56 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3572, pruned_loss=0.1176, over 5719484.36 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3421, pruned_loss=0.08801, over 5658407.89 frames. ], batch size: 336, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:02:38,431 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=625062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:03:23,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0032, 2.8268, 2.6821, 1.5395], device='cuda:0'), covar=tensor([0.1050, 0.1133, 0.0976, 0.2125], device='cuda:0'), in_proj_covar=tensor([0.1098, 0.1018, 0.0879, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 11:03:26,661 INFO [train.py:968] (0/2) Epoch 14, batch 31800, giga_loss[loss=0.2768, simple_loss=0.3571, pruned_loss=0.09823, over 28709.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3424, pruned_loss=0.08971, over 5674282.75 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.357, pruned_loss=0.1174, over 5719438.76 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3417, pruned_loss=0.08743, over 5663729.39 frames. ], batch size: 243, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:03:44,348 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=625114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:03:49,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=625117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:03:56,288 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 11:04:07,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.849e+02 1.251e+03 1.652e+03 2.254e+03 6.796e+03, threshold=3.305e+03, percent-clipped=4.0
+2023-03-07 11:04:24,880 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=625146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:04:30,010 INFO [train.py:968] (0/2) Epoch 14, batch 31850, giga_loss[loss=0.2567, simple_loss=0.3381, pruned_loss=0.08767, over 28900.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3425, pruned_loss=0.09092, over 5678808.61 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3568, pruned_loss=0.1175, over 5714528.00 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3418, pruned_loss=0.08848, over 5674708.96 frames. ], batch size: 186, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:05:37,830 INFO [train.py:968] (0/2) Epoch 14, batch 31900, giga_loss[loss=0.2676, simple_loss=0.3606, pruned_loss=0.08732, over 28837.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3413, pruned_loss=0.09202, over 5680168.70 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3558, pruned_loss=0.117, over 5718629.39 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.341, pruned_loss=0.08937, over 5671706.46 frames. ], batch size: 174, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:05:50,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=625206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:06:23,511 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.611e+02 1.326e+03 1.786e+03 2.696e+03 7.701e+03, threshold=3.571e+03, percent-clipped=12.0
+2023-03-07 11:06:57,104 INFO [train.py:968] (0/2) Epoch 14, batch 31950, giga_loss[loss=0.3153, simple_loss=0.3538, pruned_loss=0.1384, over 26962.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3421, pruned_loss=0.09311, over 5672558.91 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3558, pruned_loss=0.1171, over 5709110.37 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3416, pruned_loss=0.0904, over 5672953.87 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:07:31,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4733, 1.6694, 1.3012, 1.6001], device='cuda:0'), covar=tensor([0.0768, 0.0305, 0.0341, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 11:08:09,132 INFO [train.py:968] (0/2) Epoch 14, batch 32000, giga_loss[loss=0.2504, simple_loss=0.3323, pruned_loss=0.08421, over 28864.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3377, pruned_loss=0.09052, over 5675740.56 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3558, pruned_loss=0.1172, over 5711945.39 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3371, pruned_loss=0.08794, over 5673103.39 frames. ], batch size: 227, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:08:25,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3971, 1.6031, 1.6321, 1.2903], device='cuda:0'), covar=tensor([0.1457, 0.2090, 0.1207, 0.1538], device='cuda:0'), in_proj_covar=tensor([0.0849, 0.0680, 0.0890, 0.0797], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 11:08:52,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.358e+02 1.206e+03 1.506e+03 2.419e+03 7.607e+03, threshold=3.011e+03, percent-clipped=11.0
+2023-03-07 11:09:13,980 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=625349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:09:16,466 INFO [train.py:968] (0/2) Epoch 14, batch 32050, giga_loss[loss=0.2567, simple_loss=0.3282, pruned_loss=0.09259, over 28868.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3345, pruned_loss=0.08845, over 5675906.14 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3552, pruned_loss=0.1169, over 5713817.83 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3342, pruned_loss=0.08618, over 5671413.42 frames. ], batch size: 227, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:09:19,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=625352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:09:46,129 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-07 11:09:57,037 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=625381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:10:20,916 INFO [train.py:968] (0/2) Epoch 14, batch 32100, giga_loss[loss=0.2459, simple_loss=0.3214, pruned_loss=0.08523, over 27662.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3344, pruned_loss=0.08915, over 5675120.87 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3551, pruned_loss=0.1169, over 5707613.10 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3338, pruned_loss=0.08672, over 5676339.76 frames. ], batch size: 472, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:10:57,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5077, 1.8078, 1.5931, 1.3413], device='cuda:0'), covar=tensor([0.2439, 0.1505, 0.1536, 0.1914], device='cuda:0'), in_proj_covar=tensor([0.1776, 0.1672, 0.1604, 0.1746], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 11:10:58,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.235e+02 1.399e+03 1.738e+03 2.255e+03 5.064e+03, threshold=3.476e+03, percent-clipped=6.0
+2023-03-07 11:11:03,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=625437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:11:17,667 INFO [train.py:968] (0/2) Epoch 14, batch 32150, giga_loss[loss=0.2876, simple_loss=0.3628, pruned_loss=0.1062, over 29039.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3389, pruned_loss=0.09154, over 5673439.34 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3552, pruned_loss=0.1171, over 5704411.31 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3378, pruned_loss=0.0886, over 5676386.08 frames. ], batch size: 285, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:11:24,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2327, 4.0329, 3.8310, 1.8414], device='cuda:0'), covar=tensor([0.0623, 0.0761, 0.0871, 0.2104], device='cuda:0'), in_proj_covar=tensor([0.1102, 0.1019, 0.0883, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 11:11:58,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5626, 1.8140, 1.8514, 1.3775], device='cuda:0'), covar=tensor([0.1815, 0.2471, 0.1466, 0.1738], device='cuda:0'), in_proj_covar=tensor([0.0844, 0.0679, 0.0886, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 11:12:15,762 INFO [train.py:968] (0/2) Epoch 14, batch 32200, giga_loss[loss=0.2567, simple_loss=0.3316, pruned_loss=0.09091, over 28964.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3387, pruned_loss=0.09252, over 5676045.49 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3551, pruned_loss=0.1171, over 5696163.34 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3375, pruned_loss=0.08949, over 5685328.85 frames. ], batch size: 213, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:12:37,546 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=625515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:12:59,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.695e+02 1.275e+03 1.796e+03 2.751e+03 8.572e+03, threshold=3.591e+03, percent-clipped=14.0
+2023-03-07 11:13:21,076 INFO [train.py:968] (0/2) Epoch 14, batch 32250, giga_loss[loss=0.3162, simple_loss=0.3725, pruned_loss=0.13, over 26980.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3385, pruned_loss=0.09317, over 5675420.59 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3547, pruned_loss=0.1168, over 5699193.87 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3376, pruned_loss=0.09069, over 5680013.85 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:13:57,175 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=625580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:14:00,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=625583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:14:20,740 INFO [train.py:968] (0/2) Epoch 14, batch 32300, giga_loss[loss=0.246, simple_loss=0.3226, pruned_loss=0.0847, over 28604.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3383, pruned_loss=0.09321, over 5681207.61 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3545, pruned_loss=0.1167, over 5704748.85 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3374, pruned_loss=0.09078, over 5679498.24 frames. ], batch size: 85, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:14:37,248 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=625612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:15:09,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.899e+02 1.323e+03 1.634e+03 2.460e+03 7.328e+03, threshold=3.268e+03, percent-clipped=6.0
+2023-03-07 11:15:35,079 INFO [train.py:968] (0/2) Epoch 14, batch 32350, giga_loss[loss=0.2683, simple_loss=0.3502, pruned_loss=0.09317, over 28453.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3405, pruned_loss=0.0931, over 5676387.81 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3544, pruned_loss=0.1166, over 5702303.73 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3397, pruned_loss=0.09108, over 5676745.07 frames. ], batch size: 78, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:16:45,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3101, 1.5861, 1.6035, 1.3542], device='cuda:0'), covar=tensor([0.2148, 0.1595, 0.1165, 0.1605], device='cuda:0'), in_proj_covar=tensor([0.1776, 0.1675, 0.1600, 0.1741], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 11:16:47,093 INFO [train.py:968] (0/2) Epoch 14, batch 32400, giga_loss[loss=0.2437, simple_loss=0.3325, pruned_loss=0.07746, over 29009.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3419, pruned_loss=0.09397, over 5657267.96 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3546, pruned_loss=0.1168, over 5688038.81 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3407, pruned_loss=0.09152, over 5668485.65 frames. ], batch size: 145, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:16:51,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4393, 1.7632, 1.4257, 1.4077], device='cuda:0'), covar=tensor([0.2487, 0.2475, 0.2762, 0.2076], device='cuda:0'), in_proj_covar=tensor([0.1370, 0.1002, 0.1217, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 11:17:34,597 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.240e+02 1.539e+03 2.114e+03 3.057e+03 1.191e+04, threshold=4.228e+03, percent-clipped=24.0
+2023-03-07 11:17:53,776 INFO [train.py:968] (0/2) Epoch 14, batch 32450, giga_loss[loss=0.228, simple_loss=0.3117, pruned_loss=0.07211, over 28828.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3395, pruned_loss=0.09342, over 5664148.19 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3545, pruned_loss=0.1169, over 5690880.95 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3383, pruned_loss=0.09077, over 5669928.78 frames. ], batch size: 243, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:18:32,462 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-07 11:18:42,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4253, 2.8742, 1.5730, 1.7032], device='cuda:0'), covar=tensor([0.0804, 0.0302, 0.0726, 0.1058], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0521, 0.0355, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 11:18:49,556 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-07 11:18:55,637 INFO [train.py:968] (0/2) Epoch 14, batch 32500, libri_loss[loss=0.3033, simple_loss=0.3633, pruned_loss=0.1216, over 29653.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3341, pruned_loss=0.09129, over 5662906.60 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3542, pruned_loss=0.1166, over 5684956.81 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3329, pruned_loss=0.08852, over 5671521.00 frames. ], batch size: 88, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:19:39,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.241e+02 1.319e+03 1.741e+03 2.281e+03 6.107e+03, threshold=3.483e+03, percent-clipped=4.0
+2023-03-07 11:20:00,696 INFO [train.py:968] (0/2) Epoch 14, batch 32550, libri_loss[loss=0.3278, simple_loss=0.3712, pruned_loss=0.1422, over 19272.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3301, pruned_loss=0.08972, over 5653964.43 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.354, pruned_loss=0.1166, over 5678803.18 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3285, pruned_loss=0.08666, over 5666558.12 frames. ], batch size: 187, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:20:47,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=625890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:20:55,978 INFO [train.py:968] (0/2) Epoch 14, batch 32600, giga_loss[loss=0.2803, simple_loss=0.3518, pruned_loss=0.1044, over 28768.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3321, pruned_loss=0.09153, over 5662238.65 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3538, pruned_loss=0.1165, over 5684342.72 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3302, pruned_loss=0.08824, over 5666413.75 frames. ], batch size: 99, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:21:36,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.324e+02 1.521e+03 2.047e+03 2.964e+03 7.501e+03, threshold=4.093e+03, percent-clipped=14.0
+2023-03-07 11:21:55,532 INFO [train.py:968] (0/2) Epoch 14, batch 32650, giga_loss[loss=0.2517, simple_loss=0.3299, pruned_loss=0.08671, over 28414.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3335, pruned_loss=0.0919, over 5670316.50 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3538, pruned_loss=0.1166, over 5686545.82 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3318, pruned_loss=0.08896, over 5671393.54 frames. ], batch size: 368, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:22:56,438 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-626000.pt
+2023-03-07 11:22:56,772 INFO [train.py:968] (0/2) Epoch 14, batch 32700, giga_loss[loss=0.2216, simple_loss=0.3146, pruned_loss=0.06431, over 28947.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3302, pruned_loss=0.08915, over 5667496.89 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3534, pruned_loss=0.1164, over 5690442.78 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3287, pruned_loss=0.08649, over 5664592.53 frames. ], batch size: 213, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:23:25,940 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:23:38,878 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=626033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:23:39,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.731e+02 1.275e+03 1.707e+03 2.489e+03 7.916e+03, threshold=3.414e+03, percent-clipped=7.0
+2023-03-07 11:23:43,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=626036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:24:00,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:24:00,723 INFO [train.py:968] (0/2) Epoch 14, batch 32750, libri_loss[loss=0.2422, simple_loss=0.2989, pruned_loss=0.09276, over 29364.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3285, pruned_loss=0.08794, over 5659178.11 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.353, pruned_loss=0.1162, over 5683828.63 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3273, pruned_loss=0.08549, over 5661479.60 frames. ], batch size: 71, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:24:20,753 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=626065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:25:06,636 INFO [train.py:968] (0/2) Epoch 14, batch 32800, giga_loss[loss=0.2711, simple_loss=0.3494, pruned_loss=0.09637, over 27625.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.328, pruned_loss=0.08814, over 5667990.40 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3531, pruned_loss=0.1164, over 5690160.62 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3263, pruned_loss=0.08527, over 5663593.38 frames. ], batch size: 472, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:25:53,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.714e+02 1.274e+03 1.638e+03 2.021e+03 4.193e+03, threshold=3.276e+03, percent-clipped=3.0
+2023-03-07 11:26:16,829 INFO [train.py:968] (0/2) Epoch 14, batch 32850, giga_loss[loss=0.228, simple_loss=0.3114, pruned_loss=0.0723, over 28878.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3292, pruned_loss=0.08778, over 5670570.06 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3527, pruned_loss=0.1161, over 5684480.25 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3278, pruned_loss=0.08527, over 5671539.46 frames. ], batch size: 106, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:26:21,598 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626154.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 11:27:19,053 INFO [train.py:968] (0/2) Epoch 14, batch 32900, giga_loss[loss=0.244, simple_loss=0.3227, pruned_loss=0.08261, over 28960.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.33, pruned_loss=0.08858, over 5676594.86 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3529, pruned_loss=0.1164, over 5685366.15 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3283, pruned_loss=0.08583, over 5676373.23 frames. ], batch size: 145, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:28:00,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.590e+02 1.285e+03 1.656e+03 2.295e+03 8.403e+03, threshold=3.311e+03, percent-clipped=8.0
+2023-03-07 11:28:21,106 INFO [train.py:968] (0/2) Epoch 14, batch 32950, giga_loss[loss=0.2405, simple_loss=0.3243, pruned_loss=0.07829, over 28425.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.33, pruned_loss=0.08933, over 5676747.41 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3525, pruned_loss=0.1162, over 5687809.77 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3286, pruned_loss=0.08681, over 5674330.07 frames. ], batch size: 368, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:29:16,168 INFO [train.py:968] (0/2) Epoch 14, batch 33000, giga_loss[loss=0.2577, simple_loss=0.3456, pruned_loss=0.08493, over 28941.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3302, pruned_loss=0.08879, over 5666217.66 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3528, pruned_loss=0.1165, over 5680799.55 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3278, pruned_loss=0.08539, over 5670276.76 frames. ], batch size: 199, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:29:16,172 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 11:29:24,643 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.1991, simple_loss=0.2997, pruned_loss=0.04929, over 944034.00 frames. 
+2023-03-07 11:29:24,644 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 11:30:05,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.116e+02 1.273e+03 1.762e+03 2.301e+03 7.032e+03, threshold=3.525e+03, percent-clipped=11.0
+2023-03-07 11:30:20,627 INFO [train.py:968] (0/2) Epoch 14, batch 33050, giga_loss[loss=0.287, simple_loss=0.3653, pruned_loss=0.1043, over 28508.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3323, pruned_loss=0.08912, over 5652329.30 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3533, pruned_loss=0.1169, over 5673552.76 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3292, pruned_loss=0.08502, over 5662011.82 frames. ], batch size: 369, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:30:31,177 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:31:16,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=626398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:31:18,316 INFO [train.py:968] (0/2) Epoch 14, batch 33100, giga_loss[loss=0.2615, simple_loss=0.3421, pruned_loss=0.09051, over 27607.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3362, pruned_loss=0.09087, over 5648438.64 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3536, pruned_loss=0.1171, over 5666719.12 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3332, pruned_loss=0.087, over 5661831.69 frames. ], batch size: 474, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:31:49,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=626424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:32:04,177 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.837e+02 1.270e+03 1.902e+03 2.664e+03 7.644e+03, threshold=3.804e+03, percent-clipped=14.0
+2023-03-07 11:32:21,636 INFO [train.py:968] (0/2) Epoch 14, batch 33150, giga_loss[loss=0.2694, simple_loss=0.3458, pruned_loss=0.0965, over 28741.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3369, pruned_loss=0.09106, over 5658574.91 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3529, pruned_loss=0.1167, over 5672591.79 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3347, pruned_loss=0.08777, over 5663701.24 frames. ], batch size: 262, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:33:20,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-07 11:33:24,160 INFO [train.py:968] (0/2) Epoch 14, batch 33200, libri_loss[loss=0.3169, simple_loss=0.3714, pruned_loss=0.1312, over 29461.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3363, pruned_loss=0.09084, over 5662180.99 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3529, pruned_loss=0.1167, over 5676458.89 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3341, pruned_loss=0.0876, over 5662537.38 frames. ], batch size: 85, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:33:58,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=626529.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 11:34:06,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.329e+02 1.342e+03 1.789e+03 2.831e+03 8.441e+03, threshold=3.579e+03, percent-clipped=13.0
+2023-03-07 11:34:13,292 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=626541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:34:16,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=626544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:34:22,832 INFO [train.py:968] (0/2) Epoch 14, batch 33250, giga_loss[loss=0.2839, simple_loss=0.3451, pruned_loss=0.1113, over 26744.00 frames. ], tot_loss[loss=0.254, simple_loss=0.332, pruned_loss=0.08797, over 5672222.10 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3529, pruned_loss=0.1168, over 5680960.47 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3299, pruned_loss=0.08479, over 5668176.78 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:34:40,037 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-07 11:34:43,459 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=626567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:34:46,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=626570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:34:51,082 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=626573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:34:58,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4836, 1.7581, 1.6106, 1.4943], device='cuda:0'), covar=tensor([0.2098, 0.1691, 0.1825, 0.1700], device='cuda:0'), in_proj_covar=tensor([0.1768, 0.1668, 0.1594, 0.1738], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 11:35:13,195 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5064, 1.8123, 1.6492, 1.5308], device='cuda:0'), covar=tensor([0.1750, 0.2269, 0.2066, 0.2114], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0711, 0.0664, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 11:35:25,529 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=626599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:35:25,995 INFO [train.py:968] (0/2) Epoch 14, batch 33300, giga_loss[loss=0.2449, simple_loss=0.3225, pruned_loss=0.08366, over 28878.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.33, pruned_loss=0.08668, over 5675637.49 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3527, pruned_loss=0.1166, over 5683982.75 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3282, pruned_loss=0.08391, over 5669752.65 frames. ], batch size: 174, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:36:01,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.485e+02 1.296e+03 1.818e+03 2.410e+03 6.201e+03, threshold=3.636e+03, percent-clipped=8.0
+2023-03-07 11:36:16,752 INFO [train.py:968] (0/2) Epoch 14, batch 33350, giga_loss[loss=0.3017, simple_loss=0.3674, pruned_loss=0.118, over 28349.00 frames. ], tot_loss[loss=0.253, simple_loss=0.33, pruned_loss=0.08802, over 5676970.80 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3524, pruned_loss=0.1167, over 5685532.84 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3277, pruned_loss=0.08421, over 5670878.85 frames. ], batch size: 369, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:36:22,097 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-07 11:36:44,634 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=626672.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 11:36:48,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=626675.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 11:37:09,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8978, 2.2375, 2.2124, 1.6718], device='cuda:0'), covar=tensor([0.1523, 0.2050, 0.1224, 0.1472], device='cuda:0'), in_proj_covar=tensor([0.0845, 0.0676, 0.0887, 0.0792], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 11:37:17,586 INFO [train.py:968] (0/2) Epoch 14, batch 33400, giga_loss[loss=0.2706, simple_loss=0.3527, pruned_loss=0.09422, over 28537.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3311, pruned_loss=0.08795, over 5677925.63 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3518, pruned_loss=0.1163, over 5688786.59 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3294, pruned_loss=0.08472, over 5669763.02 frames. ], batch size: 370, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:37:23,525 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=626704.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 11:38:00,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=626734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:38:02,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.644e+02 1.315e+03 1.710e+03 2.304e+03 5.450e+03, threshold=3.419e+03, percent-clipped=6.0
+2023-03-07 11:38:16,613 INFO [train.py:968] (0/2) Epoch 14, batch 33450, giga_loss[loss=0.2846, simple_loss=0.3606, pruned_loss=0.1043, over 28641.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3331, pruned_loss=0.08943, over 5683311.02 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.352, pruned_loss=0.1165, over 5694635.03 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3309, pruned_loss=0.08559, over 5670867.97 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:38:27,423 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:39:21,211 INFO [train.py:968] (0/2) Epoch 14, batch 33500, giga_loss[loss=0.2628, simple_loss=0.3443, pruned_loss=0.09069, over 28802.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3345, pruned_loss=0.09095, over 5672901.69 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3517, pruned_loss=0.1162, over 5693206.19 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3323, pruned_loss=0.08722, over 5662882.65 frames. ], batch size: 243, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:40:10,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.538e+02 1.429e+03 1.855e+03 2.449e+03 6.175e+03, threshold=3.709e+03, percent-clipped=9.0
+2023-03-07 11:40:26,340 INFO [train.py:968] (0/2) Epoch 14, batch 33550, libri_loss[loss=0.225, simple_loss=0.2945, pruned_loss=0.07772, over 29503.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3383, pruned_loss=0.09329, over 5666204.38 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3509, pruned_loss=0.1156, over 5698288.92 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3369, pruned_loss=0.09017, over 5652345.51 frames. ], batch size: 70, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:40:39,537 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:40:54,231 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:40:54,293 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=626877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:40:57,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=626880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:41:19,112 INFO [train.py:968] (0/2) Epoch 14, batch 33600, libri_loss[loss=0.327, simple_loss=0.3761, pruned_loss=0.1389, over 19198.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3406, pruned_loss=0.09371, over 5660596.20 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3505, pruned_loss=0.1154, over 5688571.38 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3395, pruned_loss=0.09066, over 5658451.64 frames. ], batch size: 186, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:41:30,123 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=626909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:42:10,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.805e+02 1.369e+03 1.939e+03 2.723e+03 7.063e+03, threshold=3.878e+03, percent-clipped=10.0
+2023-03-07 11:42:24,251 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:42:25,373 INFO [train.py:968] (0/2) Epoch 14, batch 33650, giga_loss[loss=0.226, simple_loss=0.3091, pruned_loss=0.07145, over 29027.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3419, pruned_loss=0.09468, over 5666663.35 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3501, pruned_loss=0.1152, over 5694435.29 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3411, pruned_loss=0.09186, over 5658665.90 frames. ], batch size: 100, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:43:35,976 INFO [train.py:968] (0/2) Epoch 14, batch 33700, libri_loss[loss=0.2926, simple_loss=0.3575, pruned_loss=0.1138, over 25758.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3389, pruned_loss=0.09296, over 5669463.13 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3503, pruned_loss=0.1153, over 5694331.58 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3379, pruned_loss=0.09018, over 5663105.28 frames. ], batch size: 136, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:44:21,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.144e+02 1.370e+03 1.903e+03 2.680e+03 7.189e+03, threshold=3.807e+03, percent-clipped=8.0
+2023-03-07 11:44:23,467 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-07 11:44:35,097 INFO [train.py:968] (0/2) Epoch 14, batch 33750, giga_loss[loss=0.2721, simple_loss=0.3477, pruned_loss=0.09825, over 28919.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3378, pruned_loss=0.0923, over 5682236.88 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.35, pruned_loss=0.1151, over 5697742.80 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.337, pruned_loss=0.08961, over 5673656.48 frames. ], batch size: 186, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:45:40,492 INFO [train.py:968] (0/2) Epoch 14, batch 33800, giga_loss[loss=0.3043, simple_loss=0.3664, pruned_loss=0.1211, over 28531.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3373, pruned_loss=0.09255, over 5659438.93 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3504, pruned_loss=0.1154, over 5677214.59 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3361, pruned_loss=0.08966, over 5669692.59 frames. ], batch size: 336, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:46:17,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=627134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:46:20,347 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.525e+02 1.334e+03 1.924e+03 2.640e+03 1.218e+04, threshold=3.848e+03, percent-clipped=11.0
+2023-03-07 11:46:34,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2755, 0.8554, 0.9106, 1.4674], device='cuda:0'), covar=tensor([0.0758, 0.0349, 0.0373, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0057, 0.0096], device='cuda:0')
+2023-03-07 11:46:39,276 INFO [train.py:968] (0/2) Epoch 14, batch 33850, giga_loss[loss=0.2416, simple_loss=0.3115, pruned_loss=0.08589, over 28952.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.335, pruned_loss=0.09261, over 5662178.58 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3496, pruned_loss=0.1151, over 5679977.38 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.334, pruned_loss=0.08948, over 5667196.40 frames. ], batch size: 136, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:47:45,801 INFO [train.py:968] (0/2) Epoch 14, batch 33900, giga_loss[loss=0.2758, simple_loss=0.3517, pruned_loss=0.09999, over 27617.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3324, pruned_loss=0.09044, over 5664539.14 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3493, pruned_loss=0.1149, over 5671925.74 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3317, pruned_loss=0.08785, over 5674718.56 frames. ], batch size: 472, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:48:25,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.282e+02 1.483e+03 1.853e+03 2.543e+03 4.551e+03, threshold=3.706e+03, percent-clipped=4.0
+2023-03-07 11:48:27,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=627238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:48:42,985 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3727, 3.8205, 1.6311, 1.4766], device='cuda:0'), covar=tensor([0.0961, 0.0250, 0.0898, 0.1347], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0520, 0.0358, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 11:48:43,298 INFO [train.py:968] (0/2) Epoch 14, batch 33950, giga_loss[loss=0.2437, simple_loss=0.3356, pruned_loss=0.07588, over 28383.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3321, pruned_loss=0.08929, over 5668249.70 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3489, pruned_loss=0.1147, over 5679021.80 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3314, pruned_loss=0.08675, over 5670109.69 frames. ], batch size: 368, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:48:48,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=627252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:49:18,385 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=627277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:49:21,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=627280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:49:42,265 INFO [train.py:968] (0/2) Epoch 14, batch 34000, giga_loss[loss=0.2525, simple_loss=0.3502, pruned_loss=0.07739, over 28989.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3325, pruned_loss=0.08828, over 5675159.55 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3497, pruned_loss=0.1154, over 5680415.90 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3309, pruned_loss=0.08494, over 5674883.91 frames. ], batch size: 136, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:49:52,349 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=627309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:49:55,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5637, 1.8619, 1.9123, 1.3881], device='cuda:0'), covar=tensor([0.1905, 0.2666, 0.1634, 0.1903], device='cuda:0'), in_proj_covar=tensor([0.0848, 0.0676, 0.0891, 0.0795], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 11:50:04,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=627323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:50:23,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.237e+02 1.330e+03 1.877e+03 2.566e+03 6.006e+03, threshold=3.754e+03, percent-clipped=7.0
+2023-03-07 11:50:37,422 INFO [train.py:968] (0/2) Epoch 14, batch 34050, giga_loss[loss=0.268, simple_loss=0.3616, pruned_loss=0.08717, over 28423.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3347, pruned_loss=0.08813, over 5684552.19 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3491, pruned_loss=0.115, over 5689745.84 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3333, pruned_loss=0.08484, over 5676065.04 frames. ], batch size: 369, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:51:06,124 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=627374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:51:16,473 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=627381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:51:19,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=627384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:51:24,663 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-07 11:51:30,484 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=627395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:51:33,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=627398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:51:35,463 INFO [train.py:968] (0/2) Epoch 14, batch 34100, giga_loss[loss=0.2427, simple_loss=0.3332, pruned_loss=0.07609, over 28879.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3353, pruned_loss=0.08774, over 5688569.42 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3489, pruned_loss=0.1148, over 5693660.99 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3341, pruned_loss=0.08475, over 5678140.62 frames. ], batch size: 145, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:51:54,695 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=627413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:52:15,880 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=627427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:52:31,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.976e+02 1.347e+03 1.641e+03 2.073e+03 4.497e+03, threshold=3.282e+03, percent-clipped=1.0
+2023-03-07 11:52:49,818 INFO [train.py:968] (0/2) Epoch 14, batch 34150, giga_loss[loss=0.2185, simple_loss=0.3109, pruned_loss=0.06304, over 28865.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3351, pruned_loss=0.08769, over 5672912.85 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3488, pruned_loss=0.1149, over 5687473.00 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.334, pruned_loss=0.08495, over 5669435.83 frames. ], batch size: 174, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:53:11,857 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=627466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:53:16,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=627469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:53:33,087 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-07 11:53:52,455 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=627498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:53:55,237 INFO [train.py:968] (0/2) Epoch 14, batch 34200, giga_loss[loss=0.237, simple_loss=0.3234, pruned_loss=0.07532, over 28859.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3356, pruned_loss=0.08825, over 5672055.58 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.349, pruned_loss=0.115, over 5689074.47 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3343, pruned_loss=0.08516, over 5667161.73 frames. ], batch size: 174, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:54:13,995 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.24 vs. limit=5.0
+2023-03-07 11:54:50,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.423e+02 1.330e+03 1.735e+03 2.524e+03 8.943e+03, threshold=3.469e+03, percent-clipped=8.0
+2023-03-07 11:54:59,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3078, 3.1008, 2.9655, 1.4100], device='cuda:0'), covar=tensor([0.0850, 0.0949, 0.0871, 0.2214], device='cuda:0'), in_proj_covar=tensor([0.1084, 0.1004, 0.0864, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 11:55:08,113 INFO [train.py:968] (0/2) Epoch 14, batch 34250, giga_loss[loss=0.2474, simple_loss=0.3383, pruned_loss=0.07831, over 28761.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3362, pruned_loss=0.08815, over 5667640.33 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.349, pruned_loss=0.115, over 5687248.36 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.335, pruned_loss=0.08537, over 5665247.46 frames. ], batch size: 243, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:56:18,055 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-07 11:56:21,374 INFO [train.py:968] (0/2) Epoch 14, batch 34300, giga_loss[loss=0.2501, simple_loss=0.3304, pruned_loss=0.08494, over 28125.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3365, pruned_loss=0.08792, over 5665468.54 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3484, pruned_loss=0.1146, over 5686570.63 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3357, pruned_loss=0.08539, over 5663428.53 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:57:12,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.186e+02 1.438e+03 1.876e+03 2.923e+03 9.823e+03, threshold=3.752e+03, percent-clipped=14.0
+2023-03-07 11:57:28,033 INFO [train.py:968] (0/2) Epoch 14, batch 34350, giga_loss[loss=0.3362, simple_loss=0.4026, pruned_loss=0.1349, over 28658.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3411, pruned_loss=0.09034, over 5674406.83 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3482, pruned_loss=0.1143, over 5691781.04 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3405, pruned_loss=0.08794, over 5667712.56 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:58:34,801 INFO [train.py:968] (0/2) Epoch 14, batch 34400, giga_loss[loss=0.2113, simple_loss=0.2979, pruned_loss=0.06237, over 28701.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3409, pruned_loss=0.08965, over 5682158.14 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3484, pruned_loss=0.1144, over 5696792.09 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.34, pruned_loss=0.08709, over 5671761.48 frames. ], batch size: 85, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:59:12,812 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4457, 1.6274, 1.6783, 1.3155], device='cuda:0'), covar=tensor([0.1690, 0.2296, 0.1424, 0.1601], device='cuda:0'), in_proj_covar=tensor([0.0845, 0.0674, 0.0887, 0.0794], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 11:59:30,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.603e+02 1.511e+03 1.984e+03 2.526e+03 5.757e+03, threshold=3.968e+03, percent-clipped=6.0
+2023-03-07 11:59:45,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=627749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:59:46,118 INFO [train.py:968] (0/2) Epoch 14, batch 34450, giga_loss[loss=0.2222, simple_loss=0.3076, pruned_loss=0.06837, over 28905.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3397, pruned_loss=0.08978, over 5681315.09 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3483, pruned_loss=0.114, over 5700962.11 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3389, pruned_loss=0.08739, over 5668822.78 frames. ], batch size: 227, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 12:00:14,876 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 12:00:40,600 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-07 12:00:54,595 INFO [train.py:968] (0/2) Epoch 14, batch 34500, giga_loss[loss=0.2266, simple_loss=0.3235, pruned_loss=0.06485, over 28615.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3373, pruned_loss=0.08868, over 5690564.08 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.348, pruned_loss=0.1137, over 5706072.11 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3366, pruned_loss=0.08638, over 5675781.81 frames. ], batch size: 338, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 12:01:07,428 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=627807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:01:07,669 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 12:01:50,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.559e+02 1.328e+03 1.700e+03 2.194e+03 5.919e+03, threshold=3.400e+03, percent-clipped=5.0
+2023-03-07 12:02:07,025 INFO [train.py:968] (0/2) Epoch 14, batch 34550, giga_loss[loss=0.2386, simple_loss=0.3265, pruned_loss=0.07534, over 28669.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3357, pruned_loss=0.08684, over 5692376.54 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3481, pruned_loss=0.1137, over 5708279.41 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3349, pruned_loss=0.08477, over 5678556.10 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 12:03:02,911 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=627892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:03:06,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=627895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:03:13,084 INFO [train.py:968] (0/2) Epoch 14, batch 34600, libri_loss[loss=0.2623, simple_loss=0.3384, pruned_loss=0.09314, over 29377.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3359, pruned_loss=0.08718, over 5684864.80 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3478, pruned_loss=0.1134, over 5712176.46 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3353, pruned_loss=0.08522, over 5669765.73 frames. ], batch size: 92, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:03:41,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=627924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:03:58,660 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.723e+02 1.309e+03 1.809e+03 2.605e+03 5.899e+03, threshold=3.618e+03, percent-clipped=15.0
+2023-03-07 12:04:07,021 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3789, 1.9229, 1.5328, 1.5235], device='cuda:0'), covar=tensor([0.0797, 0.0308, 0.0326, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0057, 0.0096], device='cuda:0')
+2023-03-07 12:04:12,813 INFO [train.py:968] (0/2) Epoch 14, batch 34650, giga_loss[loss=0.2553, simple_loss=0.3435, pruned_loss=0.08354, over 28335.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3395, pruned_loss=0.0898, over 5679900.90 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3476, pruned_loss=0.1132, over 5712867.97 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.339, pruned_loss=0.08786, over 5666697.21 frames. ], batch size: 368, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:04:43,613 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-07 12:05:11,224 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-628000.pt
+2023-03-07 12:05:11,517 INFO [train.py:968] (0/2) Epoch 14, batch 34700, giga_loss[loss=0.2426, simple_loss=0.3239, pruned_loss=0.08065, over 28650.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3392, pruned_loss=0.08951, over 5686324.45 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3476, pruned_loss=0.1133, over 5711891.72 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3385, pruned_loss=0.08724, over 5676043.76 frames. ], batch size: 262, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:05:27,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=628014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:06:02,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.556e+02 1.542e+03 2.000e+03 2.687e+03 7.808e+03, threshold=4.000e+03, percent-clipped=11.0
+2023-03-07 12:06:14,505 INFO [train.py:968] (0/2) Epoch 14, batch 34750, giga_loss[loss=0.2262, simple_loss=0.3157, pruned_loss=0.06838, over 28774.00 frames. ], tot_loss[loss=0.257, simple_loss=0.336, pruned_loss=0.08901, over 5673519.51 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3474, pruned_loss=0.1132, over 5713041.83 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3356, pruned_loss=0.08712, over 5664217.37 frames. ], batch size: 174, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:07:13,572 INFO [train.py:968] (0/2) Epoch 14, batch 34800, giga_loss[loss=0.2493, simple_loss=0.3304, pruned_loss=0.08409, over 28735.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3363, pruned_loss=0.08998, over 5668806.19 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3475, pruned_loss=0.1132, over 5712887.33 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3357, pruned_loss=0.08806, over 5660832.60 frames. ], batch size: 262, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:07:59,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.884e+02 1.417e+03 1.911e+03 3.004e+03 7.397e+03, threshold=3.821e+03, percent-clipped=11.0
+2023-03-07 12:08:09,402 INFO [train.py:968] (0/2) Epoch 14, batch 34850, giga_loss[loss=0.3205, simple_loss=0.3909, pruned_loss=0.125, over 28289.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3405, pruned_loss=0.09281, over 5664100.20 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3474, pruned_loss=0.1133, over 5704019.88 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3399, pruned_loss=0.09066, over 5665243.09 frames. ], batch size: 368, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:08:39,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=628182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:08:58,200 INFO [train.py:968] (0/2) Epoch 14, batch 34900, giga_loss[loss=0.2921, simple_loss=0.3755, pruned_loss=0.1043, over 28820.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3494, pruned_loss=0.09812, over 5666295.14 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3473, pruned_loss=0.1133, over 5707659.18 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3489, pruned_loss=0.09614, over 5663329.16 frames. ], batch size: 227, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:09:37,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.586e+02 1.275e+03 1.530e+03 2.351e+03 7.404e+03, threshold=3.060e+03, percent-clipped=7.0
+2023-03-07 12:09:45,861 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-07 12:09:46,135 INFO [train.py:968] (0/2) Epoch 14, batch 34950, giga_loss[loss=0.2474, simple_loss=0.3293, pruned_loss=0.08279, over 29041.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.355, pruned_loss=0.1017, over 5669478.80 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3478, pruned_loss=0.1137, over 5709682.94 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3542, pruned_loss=0.09954, over 5664568.13 frames. ], batch size: 155, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:10:07,414 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=628274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:10:12,085 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=628279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:10:30,813 INFO [train.py:968] (0/2) Epoch 14, batch 35000, giga_loss[loss=0.2913, simple_loss=0.3369, pruned_loss=0.1229, over 24034.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3511, pruned_loss=0.1006, over 5678325.15 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3476, pruned_loss=0.1136, over 5712752.91 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3508, pruned_loss=0.09878, over 5671252.85 frames. ], batch size: 705, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:10:47,492 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-07 12:10:50,588 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=628325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:10:52,523 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=628328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:11:02,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.256e+02 1.167e+03 1.374e+03 1.787e+03 4.772e+03, threshold=2.748e+03, percent-clipped=6.0
+2023-03-07 12:11:11,804 INFO [train.py:968] (0/2) Epoch 14, batch 35050, giga_loss[loss=0.2395, simple_loss=0.3176, pruned_loss=0.08072, over 28989.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3454, pruned_loss=0.09848, over 5691137.55 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.348, pruned_loss=0.1138, over 5715925.25 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3447, pruned_loss=0.09642, over 5681888.28 frames. ], batch size: 164, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:11:18,051 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=628357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:11:43,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=628389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:11:52,706 INFO [train.py:968] (0/2) Epoch 14, batch 35100, libri_loss[loss=0.2668, simple_loss=0.3401, pruned_loss=0.09679, over 29375.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.339, pruned_loss=0.09593, over 5695733.25 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3479, pruned_loss=0.1134, over 5724309.13 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3383, pruned_loss=0.09384, over 5679592.02 frames. ], batch size: 92, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:11:55,115 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4978, 1.7128, 1.5866, 1.4777], device='cuda:0'), covar=tensor([0.1577, 0.1774, 0.2022, 0.1813], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0727, 0.0678, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 12:12:28,653 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.146e+02 1.061e+03 1.527e+03 2.194e+03 6.192e+03, threshold=3.054e+03, percent-clipped=15.0
+2023-03-07 12:12:35,881 INFO [train.py:968] (0/2) Epoch 14, batch 35150, giga_loss[loss=0.2351, simple_loss=0.2992, pruned_loss=0.08547, over 27720.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3311, pruned_loss=0.09247, over 5695110.80 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3478, pruned_loss=0.1134, over 5725742.37 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3303, pruned_loss=0.09051, over 5680540.02 frames. ], batch size: 472, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:13:17,197 INFO [train.py:968] (0/2) Epoch 14, batch 35200, giga_loss[loss=0.1958, simple_loss=0.272, pruned_loss=0.05983, over 28993.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3247, pruned_loss=0.08973, over 5697515.00 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3484, pruned_loss=0.1135, over 5725348.01 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3232, pruned_loss=0.0875, over 5685539.56 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:13:31,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9537, 1.0838, 3.3419, 2.8219], device='cuda:0'), covar=tensor([0.1815, 0.2854, 0.0502, 0.0992], device='cuda:0'), in_proj_covar=tensor([0.0695, 0.0613, 0.0893, 0.0807], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:13:49,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=628532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:13:51,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=628535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:13:55,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.671e+02 9.634e+02 1.192e+03 1.641e+03 7.003e+03, threshold=2.385e+03, percent-clipped=7.0
+2023-03-07 12:14:04,363 INFO [train.py:968] (0/2) Epoch 14, batch 35250, giga_loss[loss=0.2431, simple_loss=0.2944, pruned_loss=0.0959, over 23886.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3188, pruned_loss=0.08696, over 5694080.60 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3484, pruned_loss=0.1134, over 5728137.91 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3173, pruned_loss=0.08494, over 5681734.94 frames. ], batch size: 705, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:14:10,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9936, 2.1383, 2.2654, 1.7861], device='cuda:0'), covar=tensor([0.1727, 0.2127, 0.1331, 0.1560], device='cuda:0'), in_proj_covar=tensor([0.0850, 0.0679, 0.0892, 0.0797], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 12:14:16,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=628564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:14:45,724 INFO [train.py:968] (0/2) Epoch 14, batch 35300, giga_loss[loss=0.2157, simple_loss=0.289, pruned_loss=0.07116, over 29048.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3162, pruned_loss=0.08585, over 5685791.62 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3489, pruned_loss=0.1137, over 5720505.88 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3141, pruned_loss=0.08369, over 5682631.75 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:15:21,763 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.454e+02 1.097e+03 1.403e+03 2.034e+03 1.338e+04, threshold=2.805e+03, percent-clipped=13.0
+2023-03-07 12:15:28,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=628649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:15:29,295 INFO [train.py:968] (0/2) Epoch 14, batch 35350, giga_loss[loss=0.2262, simple_loss=0.305, pruned_loss=0.07369, over 29054.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3143, pruned_loss=0.08497, over 5699461.95 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3498, pruned_loss=0.114, over 5726575.69 frames. ], giga_tot_loss[loss=0.2372, simple_loss=0.3106, pruned_loss=0.0819, over 5690529.33 frames. ], batch size: 155, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:15:32,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=628654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:16:13,749 INFO [train.py:968] (0/2) Epoch 14, batch 35400, giga_loss[loss=0.1924, simple_loss=0.266, pruned_loss=0.05946, over 28591.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3115, pruned_loss=0.08366, over 5709226.68 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3501, pruned_loss=0.1141, over 5729214.37 frames. ], giga_tot_loss[loss=0.2347, simple_loss=0.3078, pruned_loss=0.08078, over 5699636.83 frames. ], batch size: 78, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:16:28,333 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6972, 1.8198, 1.3807, 1.3493], device='cuda:0'), covar=tensor([0.0902, 0.0641, 0.1122, 0.1231], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0435, 0.0502, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:16:50,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.558e+02 1.018e+03 1.248e+03 1.603e+03 3.380e+03, threshold=2.495e+03, percent-clipped=3.0
+2023-03-07 12:16:56,251 INFO [train.py:968] (0/2) Epoch 14, batch 35450, giga_loss[loss=0.2115, simple_loss=0.2876, pruned_loss=0.06774, over 28846.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3082, pruned_loss=0.08214, over 5701623.14 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3502, pruned_loss=0.1142, over 5723652.10 frames. ], giga_tot_loss[loss=0.2312, simple_loss=0.3043, pruned_loss=0.07908, over 5698165.02 frames. ], batch size: 186, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:17:15,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4943, 4.3321, 4.0676, 1.9905], device='cuda:0'), covar=tensor([0.0538, 0.0708, 0.0703, 0.2044], device='cuda:0'), in_proj_covar=tensor([0.1080, 0.0999, 0.0861, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 12:17:22,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6272, 1.9305, 1.5589, 1.5833], device='cuda:0'), covar=tensor([0.2435, 0.2465, 0.2801, 0.2411], device='cuda:0'), in_proj_covar=tensor([0.1362, 0.0997, 0.1209, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 12:17:32,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=628792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:17:34,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=628795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:17:35,995 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=628797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:17:38,205 INFO [train.py:968] (0/2) Epoch 14, batch 35500, giga_loss[loss=0.2353, simple_loss=0.3054, pruned_loss=0.08264, over 27896.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3058, pruned_loss=0.08121, over 5694646.26 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3508, pruned_loss=0.1146, over 5717564.96 frames. ], giga_tot_loss[loss=0.2281, simple_loss=0.301, pruned_loss=0.07754, over 5696191.32 frames. ], batch size: 412, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:17:38,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=628800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:17:41,858 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-07 12:17:58,634 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=628824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:18:01,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=628829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:18:11,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.620e+02 1.075e+03 1.535e+03 2.467e+03 9.386e+03, threshold=3.070e+03, percent-clipped=22.0
+2023-03-07 12:18:18,672 INFO [train.py:968] (0/2) Epoch 14, batch 35550, giga_loss[loss=0.199, simple_loss=0.2809, pruned_loss=0.05848, over 28992.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3039, pruned_loss=0.08028, over 5691162.71 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3514, pruned_loss=0.1149, over 5714726.20 frames. ], giga_tot_loss[loss=0.2247, simple_loss=0.2978, pruned_loss=0.07578, over 5694944.36 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:19:01,077 INFO [train.py:968] (0/2) Epoch 14, batch 35600, giga_loss[loss=0.2146, simple_loss=0.2895, pruned_loss=0.0699, over 28659.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.302, pruned_loss=0.07966, over 5680705.33 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3516, pruned_loss=0.1149, over 5707211.58 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2954, pruned_loss=0.07496, over 5688881.50 frames. ], batch size: 307, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:19:27,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5400, 3.8524, 1.6342, 1.6950], device='cuda:0'), covar=tensor([0.0872, 0.0374, 0.0837, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0515, 0.0354, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 12:19:29,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5461, 1.5931, 1.2284, 1.1503], device='cuda:0'), covar=tensor([0.0836, 0.0551, 0.0996, 0.1216], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0434, 0.0501, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:19:36,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.292e+02 1.027e+03 1.385e+03 2.080e+03 9.390e+03, threshold=2.770e+03, percent-clipped=6.0
+2023-03-07 12:19:42,634 INFO [train.py:968] (0/2) Epoch 14, batch 35650, giga_loss[loss=0.2302, simple_loss=0.305, pruned_loss=0.07765, over 28814.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3017, pruned_loss=0.07946, over 5694279.85 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3525, pruned_loss=0.115, over 5713021.46 frames. ], giga_tot_loss[loss=0.2197, simple_loss=0.2925, pruned_loss=0.07346, over 5694577.92 frames. ], batch size: 112, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:20:24,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=628998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:20:25,726 INFO [train.py:968] (0/2) Epoch 14, batch 35700, giga_loss[loss=0.2874, simple_loss=0.36, pruned_loss=0.1074, over 28835.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3065, pruned_loss=0.0824, over 5684702.56 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3526, pruned_loss=0.115, over 5707792.59 frames. ], giga_tot_loss[loss=0.2259, simple_loss=0.298, pruned_loss=0.07688, over 5688636.87 frames. ], batch size: 199, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:21:02,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.648e+02 1.256e+03 1.616e+03 2.249e+03 7.913e+03, threshold=3.231e+03, percent-clipped=8.0
+2023-03-07 12:21:10,609 INFO [train.py:968] (0/2) Epoch 14, batch 35750, giga_loss[loss=0.3112, simple_loss=0.3829, pruned_loss=0.1198, over 28908.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3177, pruned_loss=0.08828, over 5687855.73 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3525, pruned_loss=0.115, over 5708495.81 frames. ], giga_tot_loss[loss=0.2375, simple_loss=0.3095, pruned_loss=0.08278, over 5689420.56 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:21:20,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5758, 1.5120, 1.5973, 1.4832], device='cuda:0'), covar=tensor([0.2458, 0.2238, 0.1726, 0.2053], device='cuda:0'), in_proj_covar=tensor([0.1805, 0.1695, 0.1622, 0.1770], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 12:21:57,064 INFO [train.py:968] (0/2) Epoch 14, batch 35800, giga_loss[loss=0.3124, simple_loss=0.3848, pruned_loss=0.12, over 28270.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.331, pruned_loss=0.09541, over 5690013.95 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3524, pruned_loss=0.1149, over 5711328.53 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.324, pruned_loss=0.09069, over 5688374.34 frames. ], batch size: 77, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:22:04,054 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:22:32,912 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629139.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 12:22:34,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.228e+02 1.299e+03 1.570e+03 2.219e+03 4.392e+03, threshold=3.140e+03, percent-clipped=4.0
+2023-03-07 12:22:38,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9734, 1.4203, 5.1719, 3.4825], device='cuda:0'), covar=tensor([0.1547, 0.2896, 0.0332, 0.0874], device='cuda:0'), in_proj_covar=tensor([0.0693, 0.0613, 0.0892, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:22:41,254 INFO [train.py:968] (0/2) Epoch 14, batch 35850, libri_loss[loss=0.3328, simple_loss=0.3762, pruned_loss=0.1447, over 29542.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3406, pruned_loss=0.1003, over 5679804.52 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3529, pruned_loss=0.1152, over 5703426.44 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3344, pruned_loss=0.09595, over 5685766.01 frames. ], batch size: 77, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:22:48,424 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5823, 1.6354, 1.4515, 1.4112], device='cuda:0'), covar=tensor([0.2117, 0.2115, 0.1725, 0.1875], device='cuda:0'), in_proj_covar=tensor([0.1803, 0.1695, 0.1621, 0.1767], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 12:23:23,468 INFO [train.py:968] (0/2) Epoch 14, batch 35900, giga_loss[loss=0.2545, simple_loss=0.3451, pruned_loss=0.08195, over 28750.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3444, pruned_loss=0.1004, over 5685242.19 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3538, pruned_loss=0.1157, over 5703589.45 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3382, pruned_loss=0.09607, over 5688832.30 frames. ], batch size: 262, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:23:55,584 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-07 12:23:59,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.991e+02 1.186e+03 1.485e+03 2.064e+03 6.767e+03, threshold=2.971e+03, percent-clipped=11.0
+2023-03-07 12:24:05,690 INFO [train.py:968] (0/2) Epoch 14, batch 35950, giga_loss[loss=0.2833, simple_loss=0.3601, pruned_loss=0.1033, over 28926.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3463, pruned_loss=0.1004, over 5686647.58 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3545, pruned_loss=0.1161, over 5701472.30 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3403, pruned_loss=0.09594, over 5690573.64 frames. ], batch size: 145, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:24:53,898 INFO [train.py:968] (0/2) Epoch 14, batch 36000, giga_loss[loss=0.2771, simple_loss=0.3463, pruned_loss=0.1039, over 28963.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3479, pruned_loss=0.101, over 5688542.57 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3542, pruned_loss=0.116, over 5703655.31 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3434, pruned_loss=0.09742, over 5689639.77 frames. ], batch size: 145, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 12:24:53,902 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 12:25:03,381 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2087, simple_loss=0.3148, pruned_loss=0.0513, over 944034.00 frames. 
+2023-03-07 12:25:03,381 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 12:25:25,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9008, 1.0520, 1.0522, 0.8510], device='cuda:0'), covar=tensor([0.1566, 0.1899, 0.1006, 0.1572], device='cuda:0'), in_proj_covar=tensor([0.1811, 0.1701, 0.1626, 0.1775], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 12:25:38,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.637e+02 1.145e+03 1.453e+03 1.782e+03 4.916e+03, threshold=2.905e+03, percent-clipped=3.0
+2023-03-07 12:25:45,627 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2399, 1.2593, 3.8500, 3.1485], device='cuda:0'), covar=tensor([0.1663, 0.2694, 0.0410, 0.0914], device='cuda:0'), in_proj_covar=tensor([0.0690, 0.0611, 0.0885, 0.0805], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:25:45,991 INFO [train.py:968] (0/2) Epoch 14, batch 36050, giga_loss[loss=0.2649, simple_loss=0.3415, pruned_loss=0.09411, over 28918.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3534, pruned_loss=0.1058, over 5685046.84 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3548, pruned_loss=0.1162, over 5709040.37 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3492, pruned_loss=0.1024, over 5680563.91 frames. ], batch size: 164, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:26:04,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:26:25,476 INFO [train.py:968] (0/2) Epoch 14, batch 36100, giga_loss[loss=0.292, simple_loss=0.3659, pruned_loss=0.1091, over 28756.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3561, pruned_loss=0.1075, over 5692688.97 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3546, pruned_loss=0.1158, over 5717129.88 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3528, pruned_loss=0.1047, over 5681071.62 frames. ], batch size: 284, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:26:44,427 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:27:01,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.047e+02 1.150e+03 1.509e+03 1.872e+03 6.996e+03, threshold=3.018e+03, percent-clipped=11.0
+2023-03-07 12:27:08,414 INFO [train.py:968] (0/2) Epoch 14, batch 36150, giga_loss[loss=0.2732, simple_loss=0.3434, pruned_loss=0.1015, over 28586.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3575, pruned_loss=0.1081, over 5696821.58 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3539, pruned_loss=0.1152, over 5722430.92 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3555, pruned_loss=0.106, over 5681818.89 frames. ], batch size: 85, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:27:10,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:27:34,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:27:47,523 INFO [train.py:968] (0/2) Epoch 14, batch 36200, giga_loss[loss=0.3042, simple_loss=0.3751, pruned_loss=0.1166, over 28576.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3592, pruned_loss=0.1077, over 5705424.24 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.354, pruned_loss=0.1151, over 5726374.77 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3576, pruned_loss=0.106, over 5689520.33 frames. ], batch size: 336, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:28:00,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629514.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 12:28:02,355 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:28:04,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=629519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:28:25,717 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.785e+02 1.181e+03 1.420e+03 1.913e+03 6.492e+03, threshold=2.841e+03, percent-clipped=5.0
+2023-03-07 12:28:33,218 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=629548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:28:34,171 INFO [train.py:968] (0/2) Epoch 14, batch 36250, libri_loss[loss=0.2946, simple_loss=0.364, pruned_loss=0.1126, over 29527.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3611, pruned_loss=0.1084, over 5689800.33 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3542, pruned_loss=0.1151, over 5727434.64 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3597, pruned_loss=0.107, over 5675409.66 frames. ], batch size: 84, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:28:36,239 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:28:50,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5034, 1.6778, 1.7291, 1.3061], device='cuda:0'), covar=tensor([0.1863, 0.2474, 0.1493, 0.1751], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0686, 0.0896, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 12:29:13,841 INFO [train.py:968] (0/2) Epoch 14, batch 36300, giga_loss[loss=0.2565, simple_loss=0.3406, pruned_loss=0.08613, over 28446.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3622, pruned_loss=0.1078, over 5700570.34 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3543, pruned_loss=0.1151, over 5728439.90 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3612, pruned_loss=0.1066, over 5688031.38 frames. ], batch size: 65, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:29:35,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:29:37,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=629628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:29:48,239 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.599e+02 1.098e+03 1.353e+03 1.757e+03 3.370e+03, threshold=2.706e+03, percent-clipped=2.0
+2023-03-07 12:29:54,507 INFO [train.py:968] (0/2) Epoch 14, batch 36350, giga_loss[loss=0.2636, simple_loss=0.3225, pruned_loss=0.1023, over 23530.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3601, pruned_loss=0.105, over 5702546.72 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3547, pruned_loss=0.1153, over 5728539.10 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.359, pruned_loss=0.1037, over 5692218.14 frames. ], batch size: 705, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:30:01,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=629657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:30:01,550 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629657.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 12:30:04,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=629660.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 12:30:29,412 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=629689.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 12:30:37,876 INFO [train.py:968] (0/2) Epoch 14, batch 36400, giga_loss[loss=0.2739, simple_loss=0.3512, pruned_loss=0.09831, over 28957.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3581, pruned_loss=0.1027, over 5710331.69 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3547, pruned_loss=0.1152, over 5729134.62 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3573, pruned_loss=0.1016, over 5701533.02 frames. ], batch size: 106, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 12:31:13,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.803e+02 1.111e+03 1.533e+03 2.425e+03 9.751e+03, threshold=3.065e+03, percent-clipped=22.0
+2023-03-07 12:31:17,893 INFO [train.py:968] (0/2) Epoch 14, batch 36450, giga_loss[loss=0.2799, simple_loss=0.3563, pruned_loss=0.1017, over 29096.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3576, pruned_loss=0.1028, over 5717145.55 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3543, pruned_loss=0.1148, over 5731780.23 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3574, pruned_loss=0.1022, over 5707749.46 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:31:41,485 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3033, 3.1874, 1.5005, 1.3752], device='cuda:0'), covar=tensor([0.1004, 0.0299, 0.0845, 0.1393], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0512, 0.0352, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 12:31:48,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2204, 1.2276, 3.7634, 3.0648], device='cuda:0'), covar=tensor([0.1698, 0.2703, 0.0479, 0.1014], device='cuda:0'), in_proj_covar=tensor([0.0686, 0.0604, 0.0879, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:32:01,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:32:04,840 INFO [train.py:968] (0/2) Epoch 14, batch 36500, giga_loss[loss=0.3557, simple_loss=0.4015, pruned_loss=0.155, over 28780.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3618, pruned_loss=0.1087, over 5704193.92 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3551, pruned_loss=0.1153, over 5726572.85 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.361, pruned_loss=0.1075, over 5700798.33 frames. ], batch size: 119, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:32:28,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:32:43,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.089e+02 1.366e+03 1.684e+03 2.127e+03 6.911e+03, threshold=3.367e+03, percent-clipped=8.0
+2023-03-07 12:32:48,271 INFO [train.py:968] (0/2) Epoch 14, batch 36550, giga_loss[loss=0.2736, simple_loss=0.347, pruned_loss=0.1001, over 28829.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3629, pruned_loss=0.1111, over 5700965.19 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3557, pruned_loss=0.1158, over 5728692.09 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3618, pruned_loss=0.1098, over 5696168.84 frames. ], batch size: 145, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:33:34,560 INFO [train.py:968] (0/2) Epoch 14, batch 36600, giga_loss[loss=0.2855, simple_loss=0.3546, pruned_loss=0.1083, over 28918.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3615, pruned_loss=0.1109, over 5706741.97 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.356, pruned_loss=0.1158, over 5730392.70 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3604, pruned_loss=0.1097, over 5701292.29 frames. ], batch size: 213, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:33:52,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3076, 2.1405, 1.9623, 1.9279], device='cuda:0'), covar=tensor([0.1613, 0.2586, 0.2270, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0728, 0.0682, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 12:33:52,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 12:33:55,570 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:33:58,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:07,337 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:09,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=629942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:10,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.938e+02 1.153e+03 1.495e+03 2.191e+03 8.606e+03, threshold=2.989e+03, percent-clipped=6.0
+2023-03-07 12:34:15,984 INFO [train.py:968] (0/2) Epoch 14, batch 36650, giga_loss[loss=0.337, simple_loss=0.3815, pruned_loss=0.1462, over 26594.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3591, pruned_loss=0.11, over 5704472.95 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3564, pruned_loss=0.1161, over 5732746.70 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3579, pruned_loss=0.1087, over 5697110.84 frames. ], batch size: 555, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:34:31,862 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:33,948 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=629971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:34,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=629972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:49,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9009, 1.2658, 5.2601, 3.6630], device='cuda:0'), covar=tensor([0.1573, 0.2893, 0.0333, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.0689, 0.0607, 0.0886, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:34:58,907 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-630000.pt
+2023-03-07 12:34:59,213 INFO [train.py:968] (0/2) Epoch 14, batch 36700, libri_loss[loss=0.3342, simple_loss=0.3941, pruned_loss=0.1371, over 25847.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3578, pruned_loss=0.109, over 5705318.69 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3567, pruned_loss=0.1163, over 5733470.83 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3567, pruned_loss=0.1075, over 5698117.24 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:35:00,986 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=630001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:35:34,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.984e+02 1.203e+03 1.606e+03 2.351e+03 1.744e+04, threshold=3.213e+03, percent-clipped=15.0
+2023-03-07 12:35:41,999 INFO [train.py:968] (0/2) Epoch 14, batch 36750, giga_loss[loss=0.2558, simple_loss=0.3308, pruned_loss=0.09043, over 28969.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3563, pruned_loss=0.1074, over 5699646.47 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3573, pruned_loss=0.1164, over 5737433.36 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.355, pruned_loss=0.1058, over 5689703.35 frames. ], batch size: 213, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:36:00,960 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=630071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:36:04,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=630074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:36:04,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4622, 1.6354, 1.5049, 1.3090], device='cuda:0'), covar=tensor([0.2322, 0.1838, 0.1564, 0.1983], device='cuda:0'), in_proj_covar=tensor([0.1807, 0.1695, 0.1632, 0.1775], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 12:36:27,835 INFO [train.py:968] (0/2) Epoch 14, batch 36800, giga_loss[loss=0.2491, simple_loss=0.3247, pruned_loss=0.08678, over 28461.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3531, pruned_loss=0.105, over 5678696.15 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3573, pruned_loss=0.1164, over 5722166.12 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3519, pruned_loss=0.1035, over 5682759.90 frames. ], batch size: 85, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:36:30,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=630103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:37:10,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.440e+02 1.006e+03 1.229e+03 1.627e+03 4.782e+03, threshold=2.458e+03, percent-clipped=4.0
+2023-03-07 12:37:17,355 INFO [train.py:968] (0/2) Epoch 14, batch 36850, giga_loss[loss=0.2493, simple_loss=0.3233, pruned_loss=0.08769, over 28930.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3471, pruned_loss=0.1019, over 5659338.91 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3577, pruned_loss=0.1167, over 5714332.39 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3456, pruned_loss=0.1004, over 5668732.37 frames. ], batch size: 227, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:37:20,502 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-07 12:38:09,494 INFO [train.py:968] (0/2) Epoch 14, batch 36900, giga_loss[loss=0.2503, simple_loss=0.3206, pruned_loss=0.08996, over 28592.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3409, pruned_loss=0.09896, over 5652797.00 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3579, pruned_loss=0.1167, over 5715641.15 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3395, pruned_loss=0.09745, over 5658241.93 frames. ], batch size: 307, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:38:28,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3495, 1.2343, 1.2582, 1.5356], device='cuda:0'), covar=tensor([0.0783, 0.0362, 0.0332, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 12:38:54,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1936, 1.2778, 1.0767, 0.9470], device='cuda:0'), covar=tensor([0.0920, 0.0561, 0.1111, 0.1072], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0435, 0.0504, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:38:54,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.022e+02 9.404e+02 1.211e+03 1.753e+03 8.701e+03, threshold=2.422e+03, percent-clipped=7.0
+2023-03-07 12:38:59,510 INFO [train.py:968] (0/2) Epoch 14, batch 36950, giga_loss[loss=0.2844, simple_loss=0.3438, pruned_loss=0.1125, over 26509.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3383, pruned_loss=0.09808, over 5652952.51 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3582, pruned_loss=0.1171, over 5720208.00 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3365, pruned_loss=0.09608, over 5651903.08 frames. ], batch size: 555, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:39:37,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7103, 2.1229, 1.7962, 1.4933], device='cuda:0'), covar=tensor([0.2946, 0.2039, 0.2151, 0.2544], device='cuda:0'), in_proj_covar=tensor([0.1807, 0.1693, 0.1637, 0.1783], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 12:39:45,184 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=630298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:39:46,175 INFO [train.py:968] (0/2) Epoch 14, batch 37000, giga_loss[loss=0.26, simple_loss=0.3374, pruned_loss=0.0913, over 28757.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3383, pruned_loss=0.09693, over 5665761.74 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3584, pruned_loss=0.1172, over 5721270.96 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3367, pruned_loss=0.09523, over 5663837.81 frames. ], batch size: 92, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:40:12,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2432, 1.2973, 1.0441, 0.9793], device='cuda:0'), covar=tensor([0.0758, 0.0406, 0.0970, 0.1021], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0436, 0.0506, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:40:24,282 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.469e+02 1.003e+03 1.204e+03 1.411e+03 3.367e+03, threshold=2.407e+03, percent-clipped=2.0
+2023-03-07 12:40:28,020 INFO [train.py:968] (0/2) Epoch 14, batch 37050, giga_loss[loss=0.2865, simple_loss=0.3497, pruned_loss=0.1117, over 28701.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3377, pruned_loss=0.09618, over 5675452.48 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3586, pruned_loss=0.1173, over 5723479.56 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3359, pruned_loss=0.09448, over 5671129.48 frames. ], batch size: 242, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:40:54,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3146, 3.1364, 2.9626, 1.5480], device='cuda:0'), covar=tensor([0.0911, 0.1011, 0.0857, 0.2221], device='cuda:0'), in_proj_covar=tensor([0.1080, 0.1000, 0.0864, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 12:41:12,134 INFO [train.py:968] (0/2) Epoch 14, batch 37100, giga_loss[loss=0.2797, simple_loss=0.3521, pruned_loss=0.1036, over 28966.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3359, pruned_loss=0.0953, over 5691688.83 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3591, pruned_loss=0.1174, over 5726025.55 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3337, pruned_loss=0.09349, over 5685347.76 frames. ], batch size: 213, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:41:30,093 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-07 12:41:31,764 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=630425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:41:45,630 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=630441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:41:47,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=630444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:41:48,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.469e+02 9.990e+02 1.247e+03 1.470e+03 5.412e+03, threshold=2.493e+03, percent-clipped=6.0
+2023-03-07 12:41:52,057 INFO [train.py:968] (0/2) Epoch 14, batch 37150, giga_loss[loss=0.2419, simple_loss=0.3178, pruned_loss=0.08299, over 28969.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3334, pruned_loss=0.09394, over 5708364.48 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3595, pruned_loss=0.1176, over 5727933.79 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.331, pruned_loss=0.09216, over 5701441.21 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:41:57,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9298, 1.1683, 3.6897, 3.1109], device='cuda:0'), covar=tensor([0.1913, 0.2770, 0.0427, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0687, 0.0603, 0.0876, 0.0796], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:42:10,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=630473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:42:29,439 INFO [train.py:968] (0/2) Epoch 14, batch 37200, libri_loss[loss=0.272, simple_loss=0.3466, pruned_loss=0.0987, over 29573.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3325, pruned_loss=0.09388, over 5705220.16 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3602, pruned_loss=0.1177, over 5726133.26 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3285, pruned_loss=0.09115, over 5699665.83 frames. ], batch size: 78, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:42:43,152 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-07 12:42:59,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5667, 1.7283, 1.3877, 1.9143], device='cuda:0'), covar=tensor([0.0759, 0.0292, 0.0321, 0.0750], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 12:43:05,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.565e+02 1.036e+03 1.231e+03 1.718e+03 2.053e+04, threshold=2.462e+03, percent-clipped=14.0
+2023-03-07 12:43:07,898 INFO [train.py:968] (0/2) Epoch 14, batch 37250, giga_loss[loss=0.2618, simple_loss=0.3304, pruned_loss=0.09655, over 28885.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3312, pruned_loss=0.09321, over 5709774.72 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3611, pruned_loss=0.1178, over 5730207.16 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3264, pruned_loss=0.0902, over 5701091.96 frames. ], batch size: 199, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:43:29,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6076, 1.8269, 1.6985, 1.4875], device='cuda:0'), covar=tensor([0.1881, 0.1542, 0.1279, 0.1478], device='cuda:0'), in_proj_covar=tensor([0.1818, 0.1697, 0.1646, 0.1800], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 12:43:30,632 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6418, 1.6237, 1.2786, 1.2296], device='cuda:0'), covar=tensor([0.0783, 0.0521, 0.0953, 0.1127], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0437, 0.0508, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:43:39,053 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=630588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:43:48,301 INFO [train.py:968] (0/2) Epoch 14, batch 37300, giga_loss[loss=0.2393, simple_loss=0.3052, pruned_loss=0.08674, over 28685.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3297, pruned_loss=0.09272, over 5714787.57 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3617, pruned_loss=0.1182, over 5725254.96 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3247, pruned_loss=0.08943, over 5710942.09 frames. ], batch size: 78, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:44:24,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.613e+02 9.514e+02 1.073e+03 1.437e+03 4.817e+03, threshold=2.145e+03, percent-clipped=8.0
+2023-03-07 12:44:26,680 INFO [train.py:968] (0/2) Epoch 14, batch 37350, giga_loss[loss=0.2281, simple_loss=0.3024, pruned_loss=0.07687, over 28673.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3279, pruned_loss=0.0918, over 5709720.42 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3623, pruned_loss=0.1186, over 5718267.21 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3222, pruned_loss=0.08799, over 5712862.17 frames. ], batch size: 60, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:45:07,848 INFO [train.py:968] (0/2) Epoch 14, batch 37400, giga_loss[loss=0.2287, simple_loss=0.3008, pruned_loss=0.07829, over 28777.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3257, pruned_loss=0.09071, over 5711924.17 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.363, pruned_loss=0.1189, over 5719494.33 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3197, pruned_loss=0.08679, over 5713158.86 frames. ], batch size: 66, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:45:10,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1555, 0.9861, 4.2140, 3.4202], device='cuda:0'), covar=tensor([0.1883, 0.2848, 0.0761, 0.1022], device='cuda:0'), in_proj_covar=tensor([0.0691, 0.0605, 0.0882, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:45:21,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9370, 1.2895, 1.0822, 0.1592], device='cuda:0'), covar=tensor([0.3068, 0.2127, 0.3801, 0.4993], device='cuda:0'), in_proj_covar=tensor([0.1607, 0.1525, 0.1512, 0.1305], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 12:45:28,174 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 12:45:43,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.942e+02 1.068e+03 1.413e+03 2.360e+03 1.142e+04, threshold=2.825e+03, percent-clipped=30.0
+2023-03-07 12:45:45,635 INFO [train.py:968] (0/2) Epoch 14, batch 37450, giga_loss[loss=0.2669, simple_loss=0.3374, pruned_loss=0.09815, over 29077.00 frames. ], tot_loss[loss=0.254, simple_loss=0.326, pruned_loss=0.09098, over 5714672.18 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3639, pruned_loss=0.1195, over 5723713.26 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3197, pruned_loss=0.08673, over 5711705.32 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:45:49,532 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-07 12:46:27,467 INFO [train.py:968] (0/2) Epoch 14, batch 37500, giga_loss[loss=0.2523, simple_loss=0.3236, pruned_loss=0.09047, over 28816.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3252, pruned_loss=0.09052, over 5702708.45 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.365, pruned_loss=0.12, over 5717151.49 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3185, pruned_loss=0.08618, over 5706576.14 frames. ], batch size: 199, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:46:27,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=630800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:46:32,922 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 12:47:05,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2085, 2.5836, 1.3469, 1.3527], device='cuda:0'), covar=tensor([0.1000, 0.0379, 0.0870, 0.1406], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0515, 0.0353, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 12:47:05,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.908e+02 1.040e+03 1.254e+03 1.753e+03 1.090e+04, threshold=2.508e+03, percent-clipped=7.0
+2023-03-07 12:47:08,440 INFO [train.py:968] (0/2) Epoch 14, batch 37550, giga_loss[loss=0.2966, simple_loss=0.366, pruned_loss=0.1137, over 28749.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3257, pruned_loss=0.09044, over 5712739.43 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3652, pruned_loss=0.1198, over 5717619.77 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3191, pruned_loss=0.08627, over 5714946.93 frames. ], batch size: 284, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:47:19,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4681, 1.7577, 1.4819, 1.6657], device='cuda:0'), covar=tensor([0.0739, 0.0299, 0.0303, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 12:47:53,111 INFO [train.py:968] (0/2) Epoch 14, batch 37600, giga_loss[loss=0.257, simple_loss=0.332, pruned_loss=0.09102, over 28794.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3304, pruned_loss=0.09327, over 5711552.21 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3653, pruned_loss=0.1197, over 5722111.70 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3242, pruned_loss=0.08942, over 5709385.07 frames. ], batch size: 119, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:48:11,888 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=630921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:48:34,766 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=630943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:48:38,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=630946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:48:38,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.751e+02 1.337e+03 2.019e+03 2.914e+03 6.698e+03, threshold=4.037e+03, percent-clipped=36.0
+2023-03-07 12:48:41,795 INFO [train.py:968] (0/2) Epoch 14, batch 37650, giga_loss[loss=0.3204, simple_loss=0.3887, pruned_loss=0.1261, over 28841.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3379, pruned_loss=0.09817, over 5706231.25 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3651, pruned_loss=0.1196, over 5723839.70 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3328, pruned_loss=0.09501, over 5702964.42 frames. ], batch size: 186, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:48:52,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5384, 1.6572, 1.8342, 1.3626], device='cuda:0'), covar=tensor([0.1437, 0.2215, 0.1194, 0.1538], device='cuda:0'), in_proj_covar=tensor([0.0856, 0.0690, 0.0900, 0.0802], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 12:48:55,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=630963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:49:05,738 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=630975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:49:28,446 INFO [train.py:968] (0/2) Epoch 14, batch 37700, libri_loss[loss=0.2987, simple_loss=0.3657, pruned_loss=0.1159, over 29552.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.346, pruned_loss=0.1035, over 5702118.62 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3655, pruned_loss=0.1196, over 5727761.66 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3409, pruned_loss=0.1004, over 5695188.16 frames. ], batch size: 77, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:49:54,460 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-07 12:50:18,873 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.676e+02 1.267e+03 1.721e+03 2.317e+03 5.432e+03, threshold=3.443e+03, percent-clipped=3.0
+2023-03-07 12:50:21,848 INFO [train.py:968] (0/2) Epoch 14, batch 37750, giga_loss[loss=0.2916, simple_loss=0.3687, pruned_loss=0.1073, over 28991.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3494, pruned_loss=0.1047, over 5680663.78 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3652, pruned_loss=0.1194, over 5718105.02 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3454, pruned_loss=0.1022, over 5683025.96 frames. ], batch size: 112, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:51:08,585 INFO [train.py:968] (0/2) Epoch 14, batch 37800, giga_loss[loss=0.2751, simple_loss=0.358, pruned_loss=0.09611, over 28748.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3556, pruned_loss=0.1074, over 5681416.62 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3653, pruned_loss=0.1194, over 5719179.49 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3524, pruned_loss=0.1054, over 5682191.63 frames. ], batch size: 262, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:51:14,928 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=631106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:51:17,737 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=631109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:51:34,169 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=631127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:51:42,738 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=631138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:51:49,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.402e+02 1.185e+03 1.472e+03 1.849e+03 7.237e+03, threshold=2.944e+03, percent-clipped=7.0
+2023-03-07 12:51:51,851 INFO [train.py:968] (0/2) Epoch 14, batch 37850, giga_loss[loss=0.3099, simple_loss=0.3805, pruned_loss=0.1197, over 28561.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3604, pruned_loss=0.1104, over 5680559.61 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3651, pruned_loss=0.1193, over 5715649.48 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3577, pruned_loss=0.1085, over 5682091.29 frames. ], batch size: 336, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:52:02,281 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 12:52:02,981 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-07 12:52:18,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6603, 1.8684, 1.7070, 1.5276], device='cuda:0'), covar=tensor([0.2039, 0.1724, 0.1549, 0.1631], device='cuda:0'), in_proj_covar=tensor([0.1805, 0.1702, 0.1653, 0.1795], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 12:52:25,756 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=631191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:52:33,237 INFO [train.py:968] (0/2) Epoch 14, batch 37900, libri_loss[loss=0.2422, simple_loss=0.3115, pruned_loss=0.08641, over 29477.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3589, pruned_loss=0.1091, over 5689446.88 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3653, pruned_loss=0.1194, over 5721698.99 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3563, pruned_loss=0.1071, over 5683836.33 frames. ], batch size: 70, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:53:03,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2738, 1.1499, 4.1154, 3.1424], device='cuda:0'), covar=tensor([0.1584, 0.2660, 0.0440, 0.0999], device='cuda:0'), in_proj_covar=tensor([0.0693, 0.0606, 0.0883, 0.0802], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 12:53:10,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.944e+02 1.150e+03 1.412e+03 1.838e+03 8.327e+03, threshold=2.824e+03, percent-clipped=5.0
+2023-03-07 12:53:12,755 INFO [train.py:968] (0/2) Epoch 14, batch 37950, giga_loss[loss=0.2238, simple_loss=0.3097, pruned_loss=0.06894, over 28343.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3545, pruned_loss=0.1055, over 5695545.91 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3653, pruned_loss=0.1195, over 5721697.51 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3522, pruned_loss=0.1035, over 5690188.46 frames. ], batch size: 65, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:53:43,797 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-07 12:53:44,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2832, 3.3403, 1.4709, 1.3647], device='cuda:0'), covar=tensor([0.1073, 0.0287, 0.0925, 0.1498], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0515, 0.0352, 0.0402], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 12:53:55,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=631296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:53:57,489 INFO [train.py:968] (0/2) Epoch 14, batch 38000, giga_loss[loss=0.2635, simple_loss=0.3476, pruned_loss=0.08974, over 28815.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3531, pruned_loss=0.1039, over 5698378.44 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3652, pruned_loss=0.1194, over 5723581.57 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3512, pruned_loss=0.1021, over 5692140.43 frames. ], batch size: 174, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 12:54:17,227 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=631321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:54:38,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.511e+02 1.120e+03 1.423e+03 2.197e+03 6.242e+03, threshold=2.846e+03, percent-clipped=11.0
+2023-03-07 12:54:39,789 INFO [train.py:968] (0/2) Epoch 14, batch 38050, giga_loss[loss=0.2958, simple_loss=0.3658, pruned_loss=0.1129, over 28665.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3536, pruned_loss=0.1043, over 5694395.48 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3657, pruned_loss=0.1198, over 5714408.61 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3513, pruned_loss=0.1021, over 5696900.66 frames. ], batch size: 262, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:55:23,400 INFO [train.py:968] (0/2) Epoch 14, batch 38100, giga_loss[loss=0.328, simple_loss=0.3977, pruned_loss=0.1292, over 28561.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3568, pruned_loss=0.1062, over 5682753.57 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3661, pruned_loss=0.12, over 5701537.48 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3544, pruned_loss=0.104, over 5695893.79 frames. ], batch size: 336, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:55:57,015 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=631439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:55:59,922 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=631442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:56:03,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.747e+02 1.363e+03 1.739e+03 2.462e+03 9.222e+03, threshold=3.478e+03, percent-clipped=15.0
+2023-03-07 12:56:05,965 INFO [train.py:968] (0/2) Epoch 14, batch 38150, giga_loss[loss=0.2717, simple_loss=0.3463, pruned_loss=0.09853, over 28456.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3574, pruned_loss=0.1067, over 5691750.50 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3659, pruned_loss=0.1199, over 5707646.95 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3555, pruned_loss=0.1047, over 5696363.02 frames. ], batch size: 60, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:56:27,142 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=631471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:56:38,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8092, 3.1284, 1.9549, 0.8970], device='cuda:0'), covar=tensor([0.4675, 0.1539, 0.2486, 0.3970], device='cuda:0'), in_proj_covar=tensor([0.1611, 0.1524, 0.1517, 0.1313], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 12:56:52,512 INFO [train.py:968] (0/2) Epoch 14, batch 38200, giga_loss[loss=0.2582, simple_loss=0.3377, pruned_loss=0.08941, over 28753.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3596, pruned_loss=0.1085, over 5686713.82 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3662, pruned_loss=0.12, over 5707344.41 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3576, pruned_loss=0.1066, over 5690319.08 frames. ], batch size: 284, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:56:55,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=631502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:57:05,017 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 12:57:32,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5119, 1.5286, 1.4765, 1.3208], device='cuda:0'), covar=tensor([0.2087, 0.2011, 0.1688, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.1800, 0.1691, 0.1649, 0.1783], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 12:57:35,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.372e+02 1.262e+03 1.562e+03 2.183e+03 6.631e+03, threshold=3.124e+03, percent-clipped=7.0
+2023-03-07 12:57:36,598 INFO [train.py:968] (0/2) Epoch 14, batch 38250, giga_loss[loss=0.3246, simple_loss=0.384, pruned_loss=0.1326, over 28623.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3615, pruned_loss=0.1104, over 5689030.58 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3667, pruned_loss=0.1203, over 5702192.35 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3593, pruned_loss=0.1083, over 5695414.04 frames. ], batch size: 307, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:57:43,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2479, 1.5548, 1.2606, 1.1258], device='cuda:0'), covar=tensor([0.2431, 0.2402, 0.2639, 0.2068], device='cuda:0'), in_proj_covar=tensor([0.1372, 0.1009, 0.1212, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 12:57:51,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=631566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:58:08,818 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-07 12:58:18,931 INFO [train.py:968] (0/2) Epoch 14, batch 38300, libri_loss[loss=0.2494, simple_loss=0.3243, pruned_loss=0.08727, over 29456.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3608, pruned_loss=0.1103, over 5683664.16 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3666, pruned_loss=0.1203, over 5701778.69 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3591, pruned_loss=0.1085, over 5688944.93 frames. ], batch size: 70, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:58:28,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6439, 1.8514, 1.6956, 1.8178], device='cuda:0'), covar=tensor([0.1863, 0.2091, 0.2292, 0.1901], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0735, 0.0687, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 12:58:35,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7150, 1.9335, 2.0095, 1.5662], device='cuda:0'), covar=tensor([0.1732, 0.2446, 0.1439, 0.1715], device='cuda:0'), in_proj_covar=tensor([0.0855, 0.0689, 0.0897, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 12:58:36,771 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=631621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:58:55,823 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=631645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:58:57,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.348e+02 1.264e+03 1.546e+03 2.461e+03 8.780e+03, threshold=3.092e+03, percent-clipped=12.0
+2023-03-07 12:58:57,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=631648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:58:58,778 INFO [train.py:968] (0/2) Epoch 14, batch 38350, giga_loss[loss=0.2777, simple_loss=0.3543, pruned_loss=0.1005, over 28893.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3614, pruned_loss=0.1101, over 5687923.25 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3668, pruned_loss=0.1203, over 5698417.62 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3597, pruned_loss=0.1084, over 5694951.83 frames. ], batch size: 227, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:59:22,606 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=631677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:59:41,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=631696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:59:43,695 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.21 vs. limit=5.0
+2023-03-07 12:59:43,875 INFO [train.py:968] (0/2) Epoch 14, batch 38400, giga_loss[loss=0.222, simple_loss=0.3162, pruned_loss=0.0639, over 28556.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.36, pruned_loss=0.1078, over 5690167.78 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3668, pruned_loss=0.1203, over 5698417.62 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3586, pruned_loss=0.1065, over 5695638.21 frames. ], batch size: 60, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 12:59:52,978 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=631709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:59:55,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=631712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:00:18,236 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=631741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:00:23,905 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.242e+02 9.328e+02 1.148e+03 1.573e+03 4.012e+03, threshold=2.295e+03, percent-clipped=1.0
+2023-03-07 13:00:26,580 INFO [train.py:968] (0/2) Epoch 14, batch 38450, giga_loss[loss=0.3178, simple_loss=0.3652, pruned_loss=0.1353, over 26542.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3606, pruned_loss=0.1075, over 5692556.41 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3667, pruned_loss=0.1203, over 5690875.65 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3596, pruned_loss=0.1064, over 5703371.73 frames. ], batch size: 555, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:01:09,860 INFO [train.py:968] (0/2) Epoch 14, batch 38500, giga_loss[loss=0.2866, simple_loss=0.345, pruned_loss=0.1141, over 27636.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.358, pruned_loss=0.1065, over 5691135.33 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3667, pruned_loss=0.1203, over 5692315.87 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3571, pruned_loss=0.1055, over 5698300.35 frames. ], batch size: 472, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:01:44,009 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=631839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:01:46,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=631842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:01:52,379 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.861e+02 1.053e+03 1.393e+03 1.885e+03 4.204e+03, threshold=2.786e+03, percent-clipped=14.0
+2023-03-07 13:01:53,631 INFO [train.py:968] (0/2) Epoch 14, batch 38550, giga_loss[loss=0.2937, simple_loss=0.3592, pruned_loss=0.1141, over 28462.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3554, pruned_loss=0.105, over 5694279.66 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3667, pruned_loss=0.1202, over 5694225.71 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3547, pruned_loss=0.1042, over 5698261.82 frames. ], batch size: 85, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:02:11,879 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=631871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:02:32,031 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-07 13:02:34,275 INFO [train.py:968] (0/2) Epoch 14, batch 38600, giga_loss[loss=0.2792, simple_loss=0.3463, pruned_loss=0.106, over 27635.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3542, pruned_loss=0.1043, over 5701624.96 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3667, pruned_loss=0.1202, over 5696321.94 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3535, pruned_loss=0.1035, over 5702979.45 frames. ], batch size: 472, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:03:16,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.724e+02 1.028e+03 1.197e+03 1.699e+03 3.691e+03, threshold=2.394e+03, percent-clipped=3.0
+2023-03-07 13:03:17,663 INFO [train.py:968] (0/2) Epoch 14, batch 38650, giga_loss[loss=0.2795, simple_loss=0.356, pruned_loss=0.1015, over 28624.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3548, pruned_loss=0.1048, over 5702539.49 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3669, pruned_loss=0.1203, over 5698341.69 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.354, pruned_loss=0.104, over 5701818.48 frames. ], batch size: 262, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:03:25,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2205, 1.6111, 1.1808, 0.9445], device='cuda:0'), covar=tensor([0.2385, 0.2313, 0.2621, 0.2043], device='cuda:0'), in_proj_covar=tensor([0.1366, 0.1006, 0.1207, 0.0960], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 13:03:42,251 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=631981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:03:53,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=631996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:03:57,077 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-632000.pt
+2023-03-07 13:03:57,382 INFO [train.py:968] (0/2) Epoch 14, batch 38700, giga_loss[loss=0.2948, simple_loss=0.3732, pruned_loss=0.1083, over 28506.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.355, pruned_loss=0.1051, over 5691536.72 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3673, pruned_loss=0.1207, over 5685433.01 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3537, pruned_loss=0.1037, over 5701571.30 frames. ], batch size: 78, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:04:34,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.346e+02 1.092e+03 1.460e+03 2.003e+03 1.124e+04, threshold=2.920e+03, percent-clipped=15.0
+2023-03-07 13:04:35,640 INFO [train.py:968] (0/2) Epoch 14, batch 38750, giga_loss[loss=0.295, simple_loss=0.3681, pruned_loss=0.1109, over 28872.00 frames. ], tot_loss[loss=0.282, simple_loss=0.355, pruned_loss=0.1046, over 5687725.32 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3671, pruned_loss=0.1205, over 5676401.92 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3538, pruned_loss=0.1032, over 5704354.38 frames. ], batch size: 186, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:05:12,715 INFO [train.py:968] (0/2) Epoch 14, batch 38800, libri_loss[loss=0.2854, simple_loss=0.3588, pruned_loss=0.106, over 29510.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3541, pruned_loss=0.1032, over 5693307.05 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3671, pruned_loss=0.1204, over 5671585.13 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3528, pruned_loss=0.1018, over 5710965.84 frames. ], batch size: 84, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:05:46,629 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=632139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:05:48,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=632142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:05:54,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.901e+02 1.020e+03 1.256e+03 1.718e+03 5.693e+03, threshold=2.512e+03, percent-clipped=6.0
+2023-03-07 13:05:55,354 INFO [train.py:968] (0/2) Epoch 14, batch 38850, giga_loss[loss=0.2571, simple_loss=0.3392, pruned_loss=0.0875, over 28813.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3537, pruned_loss=0.1035, over 5691964.35 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3668, pruned_loss=0.1202, over 5674992.11 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3527, pruned_loss=0.1023, over 5703015.53 frames. ], batch size: 174, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:05:57,668 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:06:13,087 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=632171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:06:23,195 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5523, 1.7696, 1.7912, 1.3044], device='cuda:0'), covar=tensor([0.1502, 0.2506, 0.1361, 0.1644], device='cuda:0'), in_proj_covar=tensor([0.0855, 0.0690, 0.0899, 0.0801], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 13:06:33,344 INFO [train.py:968] (0/2) Epoch 14, batch 38900, giga_loss[loss=0.2892, simple_loss=0.3641, pruned_loss=0.1071, over 28877.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3514, pruned_loss=0.1025, over 5692848.41 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3667, pruned_loss=0.1201, over 5672327.20 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3503, pruned_loss=0.1011, over 5704954.70 frames. ], batch size: 186, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:07:02,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5232, 1.6951, 1.7974, 1.3231], device='cuda:0'), covar=tensor([0.1702, 0.2526, 0.1375, 0.1714], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0690, 0.0898, 0.0801], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 13:07:15,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.424e+02 1.217e+03 1.550e+03 2.502e+03 6.029e+03, threshold=3.100e+03, percent-clipped=23.0
+2023-03-07 13:07:15,725 INFO [train.py:968] (0/2) Epoch 14, batch 38950, giga_loss[loss=0.2669, simple_loss=0.3383, pruned_loss=0.09777, over 28607.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3489, pruned_loss=0.1015, over 5691544.94 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3669, pruned_loss=0.1201, over 5676510.36 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3474, pruned_loss=0.0999, over 5697792.39 frames. ], batch size: 85, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:07:22,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3381, 1.9720, 1.5256, 0.5573], device='cuda:0'), covar=tensor([0.4397, 0.2590, 0.3725, 0.5021], device='cuda:0'), in_proj_covar=tensor([0.1604, 0.1511, 0.1501, 0.1303], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 13:07:31,450 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:07:53,207 INFO [train.py:968] (0/2) Epoch 14, batch 39000, giga_loss[loss=0.2365, simple_loss=0.313, pruned_loss=0.07998, over 28975.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.346, pruned_loss=0.09955, over 5702769.46 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3669, pruned_loss=0.12, over 5676492.01 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3445, pruned_loss=0.09803, over 5708091.80 frames. ], batch size: 106, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:07:53,212 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 13:08:02,125 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2129, simple_loss=0.3203, pruned_loss=0.05275, over 944034.00 frames. 
+2023-03-07 13:08:02,125 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 13:08:43,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.986e+02 1.128e+03 1.426e+03 1.900e+03 8.202e+03, threshold=2.852e+03, percent-clipped=4.0
+2023-03-07 13:08:43,086 INFO [train.py:968] (0/2) Epoch 14, batch 39050, libri_loss[loss=0.3269, simple_loss=0.3819, pruned_loss=0.1359, over 19161.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3459, pruned_loss=0.1004, over 5687006.40 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3669, pruned_loss=0.1201, over 5664115.93 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.344, pruned_loss=0.09838, over 5704477.55 frames. ], batch size: 187, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:08:48,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=632356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:09:13,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8640, 2.0471, 2.1141, 1.6582], device='cuda:0'), covar=tensor([0.1775, 0.2142, 0.1398, 0.1567], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0689, 0.0898, 0.0801], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 13:09:22,166 INFO [train.py:968] (0/2) Epoch 14, batch 39100, giga_loss[loss=0.2361, simple_loss=0.3168, pruned_loss=0.0777, over 28526.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3443, pruned_loss=0.09971, over 5689212.33 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3673, pruned_loss=0.1204, over 5657333.12 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3421, pruned_loss=0.09766, over 5709612.56 frames. ], batch size: 71, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:09:22,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2377, 1.4692, 1.4569, 1.3056], device='cuda:0'), covar=tensor([0.1703, 0.1678, 0.2252, 0.1865], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0728, 0.0686, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 13:09:59,012 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632441.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 13:10:05,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.385e+02 1.058e+03 1.314e+03 1.888e+03 6.106e+03, threshold=2.628e+03, percent-clipped=7.0
+2023-03-07 13:10:05,240 INFO [train.py:968] (0/2) Epoch 14, batch 39150, giga_loss[loss=0.2344, simple_loss=0.3151, pruned_loss=0.07682, over 28418.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.343, pruned_loss=0.09975, over 5681752.62 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3679, pruned_loss=0.1208, over 5649309.39 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3404, pruned_loss=0.09737, over 5705057.96 frames. ], batch size: 71, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:10:43,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9331, 2.0376, 1.3594, 1.7656], device='cuda:0'), covar=tensor([0.0887, 0.0744, 0.1141, 0.1151], device='cuda:0'), in_proj_covar=tensor([0.0367, 0.0434, 0.0505, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 13:10:43,667 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=632499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:10:43,959 INFO [train.py:968] (0/2) Epoch 14, batch 39200, giga_loss[loss=0.2271, simple_loss=0.3117, pruned_loss=0.07123, over 28857.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3395, pruned_loss=0.0983, over 5691010.41 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3684, pruned_loss=0.1213, over 5653864.67 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3366, pruned_loss=0.09566, over 5706188.60 frames. ], batch size: 174, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:10:47,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=632502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:11:09,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=632528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:11:15,030 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=632531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:11:26,510 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-07 13:11:30,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.230e+02 9.685e+02 1.263e+03 1.772e+03 5.770e+03, threshold=2.525e+03, percent-clipped=4.0
+2023-03-07 13:11:30,789 INFO [train.py:968] (0/2) Epoch 14, batch 39250, giga_loss[loss=0.2548, simple_loss=0.3248, pruned_loss=0.09237, over 29025.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3376, pruned_loss=0.09742, over 5692658.96 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3685, pruned_loss=0.1213, over 5653324.35 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3351, pruned_loss=0.09526, over 5705140.70 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:11:39,443 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 13:12:06,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632593.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:12:11,904 INFO [train.py:968] (0/2) Epoch 14, batch 39300, giga_loss[loss=0.2728, simple_loss=0.3554, pruned_loss=0.09513, over 28246.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3388, pruned_loss=0.0977, over 5703457.98 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3681, pruned_loss=0.1211, over 5659618.38 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3364, pruned_loss=0.09565, over 5708851.36 frames. ], batch size: 368, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:12:52,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=632646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:12:55,469 INFO [train.py:968] (0/2) Epoch 14, batch 39350, giga_loss[loss=0.2986, simple_loss=0.3702, pruned_loss=0.1135, over 27740.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.342, pruned_loss=0.09886, over 5702455.95 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3688, pruned_loss=0.1215, over 5660235.92 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3388, pruned_loss=0.09621, over 5707069.24 frames. ], batch size: 472, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:12:56,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.135e+02 1.075e+03 1.333e+03 1.791e+03 8.242e+03, threshold=2.667e+03, percent-clipped=15.0
+2023-03-07 13:13:15,597 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=632671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:13:18,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=632674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:13:41,799 INFO [train.py:968] (0/2) Epoch 14, batch 39400, giga_loss[loss=0.3028, simple_loss=0.3676, pruned_loss=0.119, over 29000.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3448, pruned_loss=0.09966, over 5702402.90 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3687, pruned_loss=0.1214, over 5663980.50 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.342, pruned_loss=0.09742, over 5703281.86 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:13:45,500 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=632703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:04,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:08,102 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:25,163 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:26,303 INFO [train.py:968] (0/2) Epoch 14, batch 39450, giga_loss[loss=0.2827, simple_loss=0.3508, pruned_loss=0.1073, over 28796.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3468, pruned_loss=0.09975, over 5698635.75 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3685, pruned_loss=0.1212, over 5668488.06 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3444, pruned_loss=0.09774, over 5696000.65 frames. ], batch size: 99, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:14:26,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.086e+02 1.018e+03 1.262e+03 1.694e+03 4.766e+03, threshold=2.524e+03, percent-clipped=8.0
+2023-03-07 13:14:37,789 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:56,287 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:59,720 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=632789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:15:03,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=632792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:15:08,986 INFO [train.py:968] (0/2) Epoch 14, batch 39500, giga_loss[loss=0.2455, simple_loss=0.3324, pruned_loss=0.0793, over 28271.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3462, pruned_loss=0.09911, over 5686877.30 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3689, pruned_loss=0.1215, over 5662984.06 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3434, pruned_loss=0.09669, over 5690111.89 frames. ], batch size: 368, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 13:15:19,866 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 13:15:23,171 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=632816.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:15:26,603 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=632821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:15:39,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6320, 1.7678, 1.4079, 1.8342], device='cuda:0'), covar=tensor([0.2402, 0.2534, 0.2814, 0.2397], device='cuda:0'), in_proj_covar=tensor([0.1365, 0.1005, 0.1206, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 13:15:52,816 INFO [train.py:968] (0/2) Epoch 14, batch 39550, giga_loss[loss=0.2958, simple_loss=0.3573, pruned_loss=0.1171, over 24244.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.345, pruned_loss=0.09771, over 5695486.75 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3689, pruned_loss=0.1216, over 5667014.13 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3424, pruned_loss=0.09549, over 5694939.03 frames. ], batch size: 705, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 13:15:54,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.242e+02 1.132e+03 1.654e+03 2.729e+03 6.886e+03, threshold=3.307e+03, percent-clipped=29.0
+2023-03-07 13:16:37,109 INFO [train.py:968] (0/2) Epoch 14, batch 39600, giga_loss[loss=0.291, simple_loss=0.3603, pruned_loss=0.1108, over 28942.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3468, pruned_loss=0.09954, over 5695198.99 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3697, pruned_loss=0.122, over 5671134.06 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3437, pruned_loss=0.09701, over 5691699.08 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:17:21,180 INFO [train.py:968] (0/2) Epoch 14, batch 39650, giga_loss[loss=0.259, simple_loss=0.3451, pruned_loss=0.08647, over 28957.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3486, pruned_loss=0.1008, over 5691940.08 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3697, pruned_loss=0.1219, over 5672396.52 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3458, pruned_loss=0.09848, over 5688558.10 frames. ], batch size: 174, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:17:23,614 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.028e+02 1.121e+03 1.348e+03 1.819e+03 4.215e+03, threshold=2.695e+03, percent-clipped=2.0
+2023-03-07 13:17:29,647 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=632959.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:17:31,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=632962.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 13:17:37,579 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=632968.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:17:56,772 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=632991.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 13:18:04,647 INFO [train.py:968] (0/2) Epoch 14, batch 39700, giga_loss[loss=0.2972, simple_loss=0.3625, pruned_loss=0.1159, over 28459.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3522, pruned_loss=0.1032, over 5693591.81 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3698, pruned_loss=0.1219, over 5666234.09 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3495, pruned_loss=0.101, over 5697750.45 frames. ], batch size: 65, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:18:26,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5772, 1.6272, 1.8093, 1.3315], device='cuda:0'), covar=tensor([0.1706, 0.2269, 0.1395, 0.1604], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0685, 0.0896, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 13:18:38,168 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=633039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:18:45,495 INFO [train.py:968] (0/2) Epoch 14, batch 39750, giga_loss[loss=0.2729, simple_loss=0.3584, pruned_loss=0.09369, over 28962.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.355, pruned_loss=0.1044, over 5702638.05 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3699, pruned_loss=0.1219, over 5668344.23 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3526, pruned_loss=0.1024, over 5704453.13 frames. ], batch size: 164, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:18:46,831 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.656e+02 1.261e+03 1.585e+03 2.197e+03 6.347e+03, threshold=3.170e+03, percent-clipped=19.0
+2023-03-07 13:19:15,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5886, 1.6254, 1.8542, 1.3944], device='cuda:0'), covar=tensor([0.1710, 0.2180, 0.1350, 0.1595], device='cuda:0'), in_proj_covar=tensor([0.0855, 0.0687, 0.0896, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 13:19:25,500 INFO [train.py:968] (0/2) Epoch 14, batch 39800, giga_loss[loss=0.2717, simple_loss=0.3484, pruned_loss=0.09749, over 28969.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3559, pruned_loss=0.1044, over 5709585.91 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3702, pruned_loss=0.1219, over 5673479.94 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3533, pruned_loss=0.1025, over 5707408.16 frames. ], batch size: 213, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:19:25,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:19:27,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 13:19:29,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:19:36,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3612, 3.3510, 1.4712, 1.4528], device='cuda:0'), covar=tensor([0.0856, 0.0302, 0.0866, 0.1177], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0516, 0.0352, 0.0399], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 13:19:37,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633111.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:19:38,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633114.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:19:44,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:20:00,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:20:02,670 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633143.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:20:07,030 INFO [train.py:968] (0/2) Epoch 14, batch 39850, giga_loss[loss=0.2522, simple_loss=0.3332, pruned_loss=0.08556, over 29006.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3569, pruned_loss=0.1051, over 5706735.16 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3709, pruned_loss=0.1222, over 5666599.30 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3538, pruned_loss=0.1027, over 5712222.03 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:20:08,263 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.750e+02 1.331e+03 1.824e+03 2.318e+03 7.324e+03, threshold=3.647e+03, percent-clipped=10.0
+2023-03-07 13:20:18,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:20:46,411 INFO [train.py:968] (0/2) Epoch 14, batch 39900, giga_loss[loss=0.2915, simple_loss=0.3626, pruned_loss=0.1102, over 28826.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3562, pruned_loss=0.1049, over 5713206.78 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3707, pruned_loss=0.122, over 5672390.09 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3536, pruned_loss=0.1028, over 5713468.45 frames. ], batch size: 112, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:21:19,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:21,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:22,460 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:24,576 INFO [train.py:968] (0/2) Epoch 14, batch 39950, giga_loss[loss=0.2542, simple_loss=0.328, pruned_loss=0.09017, over 28726.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3539, pruned_loss=0.1035, over 5713138.61 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3703, pruned_loss=0.1217, over 5676624.25 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3518, pruned_loss=0.1017, over 5710683.53 frames. ], batch size: 92, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 13:21:24,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:27,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.524e+02 1.230e+03 1.514e+03 1.969e+03 5.421e+03, threshold=3.029e+03, percent-clipped=4.0
+2023-03-07 13:21:36,496 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:38,824 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:43,048 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:48,213 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:50,483 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:53,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:56,023 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2377, 1.5385, 1.2376, 0.9735], device='cuda:0'), covar=tensor([0.2399, 0.2356, 0.2721, 0.2129], device='cuda:0'), in_proj_covar=tensor([0.1358, 0.1002, 0.1201, 0.0958], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 13:22:03,027 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:22:04,852 INFO [train.py:968] (0/2) Epoch 14, batch 40000, giga_loss[loss=0.2662, simple_loss=0.3385, pruned_loss=0.09689, over 28851.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3517, pruned_loss=0.1027, over 5709963.15 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3706, pruned_loss=0.1217, over 5677864.62 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3494, pruned_loss=0.1008, over 5707632.45 frames. ], batch size: 199, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:22:08,806 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:22:12,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:22:16,232 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:22:22,707 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 13:22:30,615 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-07 13:22:35,479 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:22:46,504 INFO [train.py:968] (0/2) Epoch 14, batch 40050, giga_loss[loss=0.232, simple_loss=0.3076, pruned_loss=0.07821, over 28759.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3469, pruned_loss=0.1001, over 5716478.99 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3702, pruned_loss=0.1215, over 5682630.42 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3451, pruned_loss=0.09846, over 5711019.92 frames. ], batch size: 92, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:22:49,291 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.706e+02 1.029e+03 1.263e+03 1.722e+03 3.738e+03, threshold=2.525e+03, percent-clipped=3.0
+2023-03-07 13:23:18,280 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9991, 3.2905, 1.9475, 1.0497], device='cuda:0'), covar=tensor([0.5639, 0.2568, 0.3421, 0.5177], device='cuda:0'), in_proj_covar=tensor([0.1622, 0.1532, 0.1524, 0.1319], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 13:23:24,170 INFO [train.py:968] (0/2) Epoch 14, batch 40100, giga_loss[loss=0.2723, simple_loss=0.3567, pruned_loss=0.09399, over 28742.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3464, pruned_loss=0.09893, over 5715138.57 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3702, pruned_loss=0.1214, over 5684499.82 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3443, pruned_loss=0.09714, over 5709934.94 frames. ], batch size: 262, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:23:35,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:24:03,677 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.87 vs. limit=2.0
+2023-03-07 13:24:06,457 INFO [train.py:968] (0/2) Epoch 14, batch 40150, giga_loss[loss=0.2503, simple_loss=0.3363, pruned_loss=0.08213, over 29058.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3478, pruned_loss=0.09866, over 5707251.62 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3699, pruned_loss=0.1212, over 5686458.49 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3462, pruned_loss=0.09714, over 5701710.29 frames. ], batch size: 128, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:24:08,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.787e+02 1.138e+03 1.376e+03 1.823e+03 4.635e+03, threshold=2.752e+03, percent-clipped=9.0
+2023-03-07 13:24:08,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2204, 1.5548, 1.4999, 1.0850], device='cuda:0'), covar=tensor([0.1627, 0.2346, 0.1412, 0.1554], device='cuda:0'), in_proj_covar=tensor([0.0856, 0.0689, 0.0899, 0.0802], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 13:24:45,898 INFO [train.py:968] (0/2) Epoch 14, batch 40200, libri_loss[loss=0.3086, simple_loss=0.3696, pruned_loss=0.1238, over 29559.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3486, pruned_loss=0.09918, over 5713066.30 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3699, pruned_loss=0.1213, over 5689983.86 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3469, pruned_loss=0.09745, over 5705939.64 frames. ], batch size: 80, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:25:17,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7295, 2.5461, 1.6822, 0.7130], device='cuda:0'), covar=tensor([0.7224, 0.3528, 0.3410, 0.6819], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1529, 0.1520, 0.1318], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 13:25:24,409 INFO [train.py:968] (0/2) Epoch 14, batch 40250, giga_loss[loss=0.2773, simple_loss=0.3504, pruned_loss=0.1021, over 27955.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3471, pruned_loss=0.09939, over 5715355.83 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.37, pruned_loss=0.1214, over 5693665.09 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3452, pruned_loss=0.09757, over 5706831.77 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:25:24,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1834, 0.9307, 0.9789, 1.2734], device='cuda:0'), covar=tensor([0.0775, 0.0328, 0.0348, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 13:25:26,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.716e+02 1.121e+03 1.491e+03 1.996e+03 9.380e+03, threshold=2.982e+03, percent-clipped=13.0
+2023-03-07 13:25:31,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:25:32,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:25:53,370 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:26:03,378 INFO [train.py:968] (0/2) Epoch 14, batch 40300, giga_loss[loss=0.2891, simple_loss=0.356, pruned_loss=0.1111, over 28564.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3458, pruned_loss=0.09972, over 5717621.32 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3704, pruned_loss=0.1216, over 5689207.97 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3434, pruned_loss=0.0976, over 5715379.67 frames. ], batch size: 307, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:26:41,840 INFO [train.py:968] (0/2) Epoch 14, batch 40350, giga_loss[loss=0.3128, simple_loss=0.3769, pruned_loss=0.1244, over 28315.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3443, pruned_loss=0.1005, over 5710203.19 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3702, pruned_loss=0.1215, over 5691385.33 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3416, pruned_loss=0.09816, over 5707644.52 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:26:43,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.576e+02 1.150e+03 1.489e+03 2.168e+03 5.380e+03, threshold=2.978e+03, percent-clipped=12.0
+2023-03-07 13:26:48,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4364, 1.8131, 1.6999, 1.2622], device='cuda:0'), covar=tensor([0.1512, 0.2315, 0.1360, 0.1632], device='cuda:0'), in_proj_covar=tensor([0.0852, 0.0687, 0.0894, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 13:27:12,979 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 13:27:20,451 INFO [train.py:968] (0/2) Epoch 14, batch 40400, giga_loss[loss=0.2502, simple_loss=0.3264, pruned_loss=0.08707, over 28882.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3431, pruned_loss=0.1005, over 5710329.13 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.37, pruned_loss=0.1213, over 5694463.18 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3408, pruned_loss=0.09851, over 5705734.11 frames. ], batch size: 186, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:27:41,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1811, 1.1094, 3.9219, 3.1122], device='cuda:0'), covar=tensor([0.1687, 0.2739, 0.0392, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0687, 0.0600, 0.0879, 0.0801], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 13:28:00,353 INFO [train.py:968] (0/2) Epoch 14, batch 40450, giga_loss[loss=0.2627, simple_loss=0.3385, pruned_loss=0.09347, over 28766.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3413, pruned_loss=0.1001, over 5708722.91 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3698, pruned_loss=0.1212, over 5698503.99 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3392, pruned_loss=0.09814, over 5701741.83 frames. ], batch size: 284, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:28:02,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.511e+02 1.099e+03 1.415e+03 1.793e+03 4.715e+03, threshold=2.829e+03, percent-clipped=4.0
+2023-03-07 13:28:35,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1561, 1.6803, 1.2129, 0.4235], device='cuda:0'), covar=tensor([0.2868, 0.1426, 0.1979, 0.3849], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1517, 0.1513, 0.1309], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:0')
+2023-03-07 13:28:39,920 INFO [train.py:968] (0/2) Epoch 14, batch 40500, giga_loss[loss=0.2686, simple_loss=0.3357, pruned_loss=0.1008, over 28769.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3376, pruned_loss=0.09828, over 5714467.36 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3696, pruned_loss=0.1211, over 5701903.45 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3355, pruned_loss=0.09645, over 5706194.37 frames. ], batch size: 284, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:29:12,295 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=633838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:29:20,338 INFO [train.py:968] (0/2) Epoch 14, batch 40550, giga_loss[loss=0.2355, simple_loss=0.311, pruned_loss=0.07997, over 28867.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3327, pruned_loss=0.09555, over 5714664.85 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3698, pruned_loss=0.1212, over 5703435.01 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3303, pruned_loss=0.09364, over 5706664.98 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:29:22,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.015e+02 1.015e+03 1.298e+03 1.861e+03 4.158e+03, threshold=2.596e+03, percent-clipped=12.0
+2023-03-07 13:29:54,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5158, 1.6394, 1.5554, 1.4174], device='cuda:0'), covar=tensor([0.2319, 0.1964, 0.1639, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.1803, 0.1709, 0.1659, 0.1784], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 13:29:57,084 INFO [train.py:968] (0/2) Epoch 14, batch 40600, giga_loss[loss=0.2551, simple_loss=0.3388, pruned_loss=0.08573, over 29006.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3325, pruned_loss=0.09487, over 5723872.27 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3701, pruned_loss=0.1215, over 5708910.87 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3288, pruned_loss=0.09198, over 5713054.57 frames. ], batch size: 164, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:30:15,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5950, 1.7930, 1.8380, 1.3519], device='cuda:0'), covar=tensor([0.1816, 0.2437, 0.1504, 0.1668], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0687, 0.0897, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 13:30:40,948 INFO [train.py:968] (0/2) Epoch 14, batch 40650, giga_loss[loss=0.2843, simple_loss=0.3608, pruned_loss=0.1039, over 28986.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3357, pruned_loss=0.09617, over 5717987.84 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3698, pruned_loss=0.1213, over 5711719.06 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3323, pruned_loss=0.09358, over 5706746.76 frames. ], batch size: 164, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:30:43,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.684e+02 1.141e+03 1.438e+03 2.019e+03 4.163e+03, threshold=2.877e+03, percent-clipped=11.0
+2023-03-07 13:31:03,662 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-07 13:31:04,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1518, 1.3905, 1.3957, 1.2459], device='cuda:0'), covar=tensor([0.1559, 0.1529, 0.1976, 0.1650], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0730, 0.0688, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 13:31:22,255 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-634000.pt
+2023-03-07 13:31:22,543 INFO [train.py:968] (0/2) Epoch 14, batch 40700, giga_loss[loss=0.2701, simple_loss=0.3483, pruned_loss=0.09595, over 28860.00 frames. ], tot_loss[loss=0.267, simple_loss=0.339, pruned_loss=0.09748, over 5720677.86 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3696, pruned_loss=0.1211, over 5713459.22 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3362, pruned_loss=0.09537, over 5710254.31 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:31:33,252 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:32:01,977 INFO [train.py:968] (0/2) Epoch 14, batch 40750, giga_loss[loss=0.2735, simple_loss=0.3477, pruned_loss=0.09971, over 28817.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3418, pruned_loss=0.09831, over 5712402.58 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3693, pruned_loss=0.121, over 5714096.79 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3394, pruned_loss=0.09636, over 5703493.74 frames. ], batch size: 106, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:32:04,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.792e+02 1.118e+03 1.334e+03 1.820e+03 4.767e+03, threshold=2.668e+03, percent-clipped=6.0
+2023-03-07 13:32:45,977 INFO [train.py:968] (0/2) Epoch 14, batch 40800, giga_loss[loss=0.287, simple_loss=0.3492, pruned_loss=0.1124, over 28459.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3459, pruned_loss=0.1005, over 5702523.01 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3695, pruned_loss=0.1211, over 5705146.30 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3437, pruned_loss=0.09872, over 5704173.84 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:33:05,473 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:33:28,902 INFO [train.py:968] (0/2) Epoch 14, batch 40850, giga_loss[loss=0.4503, simple_loss=0.4592, pruned_loss=0.2207, over 26634.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3496, pruned_loss=0.103, over 5705483.89 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1212, over 5708268.48 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3475, pruned_loss=0.1012, over 5703793.53 frames. ], batch size: 555, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:33:31,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.725e+02 1.224e+03 1.561e+03 2.155e+03 4.735e+03, threshold=3.122e+03, percent-clipped=14.0
+2023-03-07 13:33:34,083 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:34:13,810 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.38 vs. limit=5.0
+2023-03-07 13:34:20,286 INFO [train.py:968] (0/2) Epoch 14, batch 40900, giga_loss[loss=0.3675, simple_loss=0.428, pruned_loss=0.1535, over 28687.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3557, pruned_loss=0.1084, over 5700451.04 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3697, pruned_loss=0.1213, over 5707736.83 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3537, pruned_loss=0.1067, over 5699523.67 frames. ], batch size: 262, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:34:31,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:35:04,060 INFO [train.py:968] (0/2) Epoch 14, batch 40950, libri_loss[loss=0.3363, simple_loss=0.4036, pruned_loss=0.1345, over 29094.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3631, pruned_loss=0.1138, over 5699719.03 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3697, pruned_loss=0.1212, over 5710516.68 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3612, pruned_loss=0.1123, over 5696349.64 frames. ], batch size: 101, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:35:08,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.952e+02 1.649e+03 2.222e+03 2.850e+03 7.056e+03, threshold=4.443e+03, percent-clipped=20.0
+2023-03-07 13:35:23,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1700, 3.9838, 3.7742, 1.7994], device='cuda:0'), covar=tensor([0.0596, 0.0741, 0.0747, 0.2068], device='cuda:0'), in_proj_covar=tensor([0.1111, 0.1031, 0.0886, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 13:35:50,663 INFO [train.py:968] (0/2) Epoch 14, batch 41000, giga_loss[loss=0.3121, simple_loss=0.3759, pruned_loss=0.1241, over 28554.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.37, pruned_loss=0.1193, over 5694878.11 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3694, pruned_loss=0.121, over 5714354.96 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3688, pruned_loss=0.1182, over 5688704.07 frames. ], batch size: 60, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:36:07,763 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-07 13:36:16,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3505, 1.2505, 1.2002, 1.3495], device='cuda:0'), covar=tensor([0.0735, 0.0328, 0.0320, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 13:36:21,307 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:36:30,442 INFO [train.py:968] (0/2) Epoch 14, batch 41050, giga_loss[loss=0.3053, simple_loss=0.3717, pruned_loss=0.1194, over 28533.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3752, pruned_loss=0.1238, over 5677126.66 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3695, pruned_loss=0.1211, over 5691154.87 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3742, pruned_loss=0.1228, over 5692914.61 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:36:36,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.193e+03 1.738e+03 2.441e+03 3.536e+03 9.211e+03, threshold=4.883e+03, percent-clipped=11.0
+2023-03-07 13:36:36,887 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:36:37,859 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:36:39,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:37:04,312 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:37:05,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:37:15,678 INFO [train.py:968] (0/2) Epoch 14, batch 41100, giga_loss[loss=0.3515, simple_loss=0.4046, pruned_loss=0.1492, over 27914.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3815, pruned_loss=0.1291, over 5680394.95 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3695, pruned_loss=0.1211, over 5695220.48 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3808, pruned_loss=0.1284, over 5688986.58 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:37:33,914 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=2.00 vs. limit=2.0
+2023-03-07 13:37:43,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3051, 3.1245, 3.0055, 1.4377], device='cuda:0'), covar=tensor([0.0921, 0.1000, 0.0926, 0.2235], device='cuda:0'), in_proj_covar=tensor([0.1108, 0.1025, 0.0882, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 13:38:08,789 INFO [train.py:968] (0/2) Epoch 14, batch 41150, giga_loss[loss=0.3285, simple_loss=0.3851, pruned_loss=0.136, over 28973.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3855, pruned_loss=0.1337, over 5653979.62 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3699, pruned_loss=0.1215, over 5693228.15 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3848, pruned_loss=0.133, over 5662202.69 frames. ], batch size: 136, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:38:14,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.836e+03 2.417e+03 3.336e+03 1.321e+04, threshold=4.835e+03, percent-clipped=7.0
+2023-03-07 13:38:54,271 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:38:58,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:39:00,489 INFO [train.py:968] (0/2) Epoch 14, batch 41200, giga_loss[loss=0.3178, simple_loss=0.3705, pruned_loss=0.1325, over 28817.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3868, pruned_loss=0.1357, over 5661283.99 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3699, pruned_loss=0.1216, over 5698205.32 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3866, pruned_loss=0.1354, over 5662523.20 frames. ], batch size: 99, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:39:31,060 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2092, 1.3101, 3.2544, 2.9087], device='cuda:0'), covar=tensor([0.1458, 0.2365, 0.0479, 0.1518], device='cuda:0'), in_proj_covar=tensor([0.0693, 0.0603, 0.0884, 0.0807], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 13:39:34,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:39:35,040 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:39:37,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:39:50,285 INFO [train.py:968] (0/2) Epoch 14, batch 41250, giga_loss[loss=0.3087, simple_loss=0.3705, pruned_loss=0.1234, over 28921.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3904, pruned_loss=0.14, over 5642446.76 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3697, pruned_loss=0.1213, over 5698662.30 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3911, pruned_loss=0.1405, over 5641677.70 frames. ], batch size: 106, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:39:56,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.230e+02 1.572e+03 2.121e+03 3.059e+03 6.397e+03, threshold=4.242e+03, percent-clipped=5.0
+2023-03-07 13:39:56,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4700, 3.0570, 1.6254, 1.6113], device='cuda:0'), covar=tensor([0.0770, 0.0322, 0.0688, 0.1065], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0526, 0.0356, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 13:40:05,432 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:40:41,732 INFO [train.py:968] (0/2) Epoch 14, batch 41300, giga_loss[loss=0.4276, simple_loss=0.4472, pruned_loss=0.2039, over 27952.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3939, pruned_loss=0.1433, over 5645111.55 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3693, pruned_loss=0.1211, over 5703487.90 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3953, pruned_loss=0.1444, over 5638798.78 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:40:51,526 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:41:20,671 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:41:22,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9055, 1.1713, 2.8447, 2.7158], device='cuda:0'), covar=tensor([0.1512, 0.2300, 0.0583, 0.1375], device='cuda:0'), in_proj_covar=tensor([0.0698, 0.0607, 0.0892, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 13:41:23,437 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:41:28,065 INFO [train.py:968] (0/2) Epoch 14, batch 41350, giga_loss[loss=0.2923, simple_loss=0.3587, pruned_loss=0.1129, over 28640.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3938, pruned_loss=0.1436, over 5629057.14 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3689, pruned_loss=0.121, over 5698124.26 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3962, pruned_loss=0.1454, over 5625584.08 frames. ], batch size: 242, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:41:38,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.803e+03 2.375e+03 3.068e+03 9.118e+03, threshold=4.750e+03, percent-clipped=8.0
+2023-03-07 13:41:53,694 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:41:55,747 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:41:57,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:42:18,533 INFO [train.py:968] (0/2) Epoch 14, batch 41400, giga_loss[loss=0.2878, simple_loss=0.351, pruned_loss=0.1122, over 28796.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3939, pruned_loss=0.1445, over 5632651.92 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3695, pruned_loss=0.1215, over 5703764.45 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.3963, pruned_loss=0.1464, over 5622189.86 frames. ], batch size: 99, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:42:26,021 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:42:31,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:42:39,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-07 13:42:50,840 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:43:09,329 INFO [train.py:968] (0/2) Epoch 14, batch 41450, giga_loss[loss=0.3191, simple_loss=0.3801, pruned_loss=0.1291, over 28976.00 frames. ], tot_loss[loss=0.339, simple_loss=0.392, pruned_loss=0.143, over 5649060.41 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3693, pruned_loss=0.1214, over 5705597.93 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3943, pruned_loss=0.1448, over 5638577.50 frames. ], batch size: 213, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:43:16,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.863e+03 2.384e+03 3.394e+03 8.660e+03, threshold=4.768e+03, percent-clipped=11.0
+2023-03-07 13:43:16,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:43:21,275 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-07 13:43:36,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6104, 1.7366, 1.3379, 1.4265], device='cuda:0'), covar=tensor([0.0714, 0.0429, 0.0921, 0.0829], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0442, 0.0506, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 13:43:39,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6477, 2.2250, 1.5361, 0.8888], device='cuda:0'), covar=tensor([0.4434, 0.2290, 0.3371, 0.4852], device='cuda:0'), in_proj_covar=tensor([0.1639, 0.1557, 0.1534, 0.1335], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 13:43:54,904 INFO [train.py:968] (0/2) Epoch 14, batch 41500, giga_loss[loss=0.2834, simple_loss=0.353, pruned_loss=0.1069, over 28478.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3915, pruned_loss=0.1408, over 5656138.47 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3695, pruned_loss=0.1214, over 5709264.86 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3941, pruned_loss=0.1431, over 5642240.71 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:44:04,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4536, 1.5526, 1.6443, 1.3360], device='cuda:0'), covar=tensor([0.2220, 0.2144, 0.1257, 0.1824], device='cuda:0'), in_proj_covar=tensor([0.1826, 0.1732, 0.1679, 0.1800], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 13:44:38,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7597, 5.5915, 5.2869, 2.9604], device='cuda:0'), covar=tensor([0.0415, 0.0531, 0.0683, 0.1512], device='cuda:0'), in_proj_covar=tensor([0.1123, 0.1039, 0.0899, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 13:44:42,392 INFO [train.py:968] (0/2) Epoch 14, batch 41550, giga_loss[loss=0.3156, simple_loss=0.3841, pruned_loss=0.1236, over 28949.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3919, pruned_loss=0.1402, over 5650828.89 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3697, pruned_loss=0.1215, over 5695314.68 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3944, pruned_loss=0.1424, over 5650420.18 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:44:48,879 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:44:50,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.380e+02 1.544e+03 1.997e+03 2.864e+03 6.759e+03, threshold=3.994e+03, percent-clipped=6.0
+2023-03-07 13:44:50,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:44:53,378 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634861.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:44:59,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:45:09,305 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:45:14,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:45:22,658 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:45:33,173 INFO [train.py:968] (0/2) Epoch 14, batch 41600, giga_loss[loss=0.338, simple_loss=0.4061, pruned_loss=0.135, over 28959.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3939, pruned_loss=0.1416, over 5639814.36 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3699, pruned_loss=0.1218, over 5688715.94 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3961, pruned_loss=0.1434, over 5643661.40 frames. ], batch size: 136, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:45:42,456 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:46:21,912 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.13 vs. limit=2.0
+2023-03-07 13:46:22,822 INFO [train.py:968] (0/2) Epoch 14, batch 41650, giga_loss[loss=0.3113, simple_loss=0.3835, pruned_loss=0.1195, over 28917.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3914, pruned_loss=0.139, over 5642202.96 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3702, pruned_loss=0.1219, over 5694998.48 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3935, pruned_loss=0.1409, over 5638159.97 frames. ], batch size: 227, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:46:29,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.722e+02 1.535e+03 2.098e+03 2.854e+03 6.551e+03, threshold=4.195e+03, percent-clipped=9.0
+2023-03-07 13:46:51,606 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6910, 2.0009, 1.7061, 1.6827], device='cuda:0'), covar=tensor([0.0723, 0.0264, 0.0291, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 13:46:54,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:47:06,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2404, 1.5994, 1.4479, 1.2183], device='cuda:0'), covar=tensor([0.2456, 0.1857, 0.1365, 0.1785], device='cuda:0'), in_proj_covar=tensor([0.1830, 0.1735, 0.1684, 0.1800], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 13:47:09,758 INFO [train.py:968] (0/2) Epoch 14, batch 41700, giga_loss[loss=0.3272, simple_loss=0.3832, pruned_loss=0.1356, over 27749.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3883, pruned_loss=0.1352, over 5651533.03 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3696, pruned_loss=0.1218, over 5698435.59 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3908, pruned_loss=0.1371, over 5644223.29 frames. ], batch size: 474, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:47:21,086 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:47:23,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:47:39,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2838, 1.7834, 1.5071, 1.5342], device='cuda:0'), covar=tensor([0.0766, 0.0326, 0.0313, 0.0830], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 13:47:49,018 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:47:55,960 INFO [train.py:968] (0/2) Epoch 14, batch 41750, giga_loss[loss=0.3095, simple_loss=0.3799, pruned_loss=0.1196, over 28546.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3849, pruned_loss=0.1321, over 5664514.30 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3697, pruned_loss=0.122, over 5704523.85 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3876, pruned_loss=0.1339, over 5651342.31 frames. ], batch size: 307, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:47:56,797 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=635051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:48:01,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.601e+03 2.180e+03 3.038e+03 9.645e+03, threshold=4.360e+03, percent-clipped=11.0
+2023-03-07 13:48:44,732 INFO [train.py:968] (0/2) Epoch 14, batch 41800, giga_loss[loss=0.3256, simple_loss=0.3945, pruned_loss=0.1283, over 28315.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3804, pruned_loss=0.1283, over 5667402.01 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3693, pruned_loss=0.1217, over 5707651.13 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.383, pruned_loss=0.1301, over 5653523.90 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:49:07,528 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3493, 1.5867, 1.6229, 1.2114], device='cuda:0'), covar=tensor([0.1628, 0.2414, 0.1340, 0.1588], device='cuda:0'), in_proj_covar=tensor([0.0849, 0.0687, 0.0890, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 13:49:11,147 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:49:11,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4829, 1.8052, 1.7309, 1.2889], device='cuda:0'), covar=tensor([0.1471, 0.2451, 0.1323, 0.1597], device='cuda:0'), in_proj_covar=tensor([0.0849, 0.0687, 0.0890, 0.0793], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:0')
+2023-03-07 13:49:13,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:49:15,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=635132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:49:33,908 INFO [train.py:968] (0/2) Epoch 14, batch 41850, giga_loss[loss=0.3041, simple_loss=0.3681, pruned_loss=0.12, over 28588.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3784, pruned_loss=0.1276, over 5655164.16 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3691, pruned_loss=0.1216, over 5711795.48 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3809, pruned_loss=0.1292, over 5639594.74 frames. ], batch size: 307, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:49:41,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.778e+02 1.509e+03 2.006e+03 2.879e+03 5.824e+03, threshold=4.011e+03, percent-clipped=6.0
+2023-03-07 13:49:42,473 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:49:43,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2339, 1.5385, 1.0072, 1.0853], device='cuda:0'), covar=tensor([0.0934, 0.0519, 0.1153, 0.1003], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0445, 0.0509, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 13:50:17,701 INFO [train.py:968] (0/2) Epoch 14, batch 41900, giga_loss[loss=0.323, simple_loss=0.383, pruned_loss=0.1315, over 28987.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3792, pruned_loss=0.128, over 5674275.53 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3688, pruned_loss=0.1214, over 5714840.19 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3816, pruned_loss=0.1296, over 5658063.50 frames. ], batch size: 136, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:50:27,059 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 13:50:40,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=635221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:50:56,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=635236.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:50:58,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4413, 3.3175, 1.5158, 1.4744], device='cuda:0'), covar=tensor([0.0946, 0.0295, 0.0867, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0526, 0.0355, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 13:51:07,871 INFO [train.py:968] (0/2) Epoch 14, batch 41950, giga_loss[loss=0.248, simple_loss=0.3318, pruned_loss=0.08211, over 28575.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3783, pruned_loss=0.1269, over 5678522.42 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.369, pruned_loss=0.1215, over 5719143.70 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3804, pruned_loss=0.1282, over 5660801.03 frames. ], batch size: 60, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:51:14,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.999e+02 1.575e+03 2.113e+03 2.936e+03 7.818e+03, threshold=4.225e+03, percent-clipped=8.0
+2023-03-07 13:51:35,539 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:51:38,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:51:59,266 INFO [train.py:968] (0/2) Epoch 14, batch 42000, giga_loss[loss=0.2756, simple_loss=0.3644, pruned_loss=0.09343, over 28953.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3771, pruned_loss=0.1244, over 5666736.94 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3695, pruned_loss=0.1219, over 5699594.25 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3785, pruned_loss=0.1252, over 5668549.60 frames. ], batch size: 145, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:51:59,270 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 13:52:08,189 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2101, simple_loss=0.3155, pruned_loss=0.05234, over 944034.00 frames. 
+2023-03-07 13:52:08,190 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 13:52:15,441 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:52:38,195 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 13:52:55,745 INFO [train.py:968] (0/2) Epoch 14, batch 42050, libri_loss[loss=0.301, simple_loss=0.357, pruned_loss=0.1225, over 29354.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3772, pruned_loss=0.1225, over 5676600.27 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1214, over 5707144.15 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3794, pruned_loss=0.1237, over 5670020.67 frames. ], batch size: 71, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:53:03,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.659e+02 1.549e+03 2.045e+03 2.867e+03 5.968e+03, threshold=4.091e+03, percent-clipped=11.0
+2023-03-07 13:53:20,105 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 13:53:23,284 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635379.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:53:26,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635382.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 13:53:40,465 INFO [train.py:968] (0/2) Epoch 14, batch 42100, libri_loss[loss=0.2933, simple_loss=0.3602, pruned_loss=0.1132, over 29531.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3789, pruned_loss=0.1236, over 5677947.06 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3688, pruned_loss=0.1215, over 5709720.95 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3808, pruned_loss=0.1246, over 5669493.61 frames. ], batch size: 81, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:53:50,620 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635411.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:54:06,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=635426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:54:27,356 INFO [train.py:968] (0/2) Epoch 14, batch 42150, libri_loss[loss=0.3058, simple_loss=0.3718, pruned_loss=0.1199, over 29209.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3798, pruned_loss=0.1253, over 5678060.15 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3681, pruned_loss=0.121, over 5712470.23 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3823, pruned_loss=0.1266, over 5667725.51 frames. ], batch size: 97, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:54:34,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.767e+03 2.187e+03 3.056e+03 8.254e+03, threshold=4.375e+03, percent-clipped=16.0
+2023-03-07 13:55:09,230 INFO [train.py:968] (0/2) Epoch 14, batch 42200, libri_loss[loss=0.3079, simple_loss=0.3649, pruned_loss=0.1254, over 29771.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3767, pruned_loss=0.1237, over 5678267.56 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3674, pruned_loss=0.1207, over 5709531.29 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.38, pruned_loss=0.1253, over 5670575.96 frames. ], batch size: 87, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:55:15,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1231, 2.7753, 1.2709, 1.3002], device='cuda:0'), covar=tensor([0.1137, 0.0470, 0.0956, 0.1473], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0530, 0.0357, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 13:55:39,249 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 13:55:50,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2994, 3.3069, 1.4889, 1.3723], device='cuda:0'), covar=tensor([0.1044, 0.0417, 0.0883, 0.1442], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0530, 0.0357, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 13:55:54,046 INFO [train.py:968] (0/2) Epoch 14, batch 42250, giga_loss[loss=0.3189, simple_loss=0.3772, pruned_loss=0.1303, over 28550.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3763, pruned_loss=0.1248, over 5667507.17 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3675, pruned_loss=0.1208, over 5703573.23 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3789, pruned_loss=0.126, over 5666328.83 frames. ], batch size: 336, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:56:01,004 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.681e+02 1.795e+03 2.385e+03 3.159e+03 9.230e+03, threshold=4.770e+03, percent-clipped=11.0
+2023-03-07 13:56:13,050 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:56:16,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:56:19,304 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 13:56:38,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=635596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:56:42,702 INFO [train.py:968] (0/2) Epoch 14, batch 42300, giga_loss[loss=0.3063, simple_loss=0.3783, pruned_loss=0.1171, over 28926.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3747, pruned_loss=0.1243, over 5654410.65 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3675, pruned_loss=0.1206, over 5696130.93 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3769, pruned_loss=0.1254, over 5658745.57 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:56:43,521 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:57:05,586 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-07 13:57:32,356 INFO [train.py:968] (0/2) Epoch 14, batch 42350, giga_loss[loss=0.2874, simple_loss=0.3676, pruned_loss=0.1036, over 28827.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3737, pruned_loss=0.1225, over 5659672.69 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3673, pruned_loss=0.1205, over 5688379.08 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3758, pruned_loss=0.1236, over 5668732.18 frames. ], batch size: 119, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:57:39,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.813e+02 1.552e+03 2.082e+03 2.755e+03 9.030e+03, threshold=4.164e+03, percent-clipped=2.0
+2023-03-07 13:58:20,105 INFO [train.py:968] (0/2) Epoch 14, batch 42400, giga_loss[loss=0.277, simple_loss=0.3671, pruned_loss=0.09347, over 28956.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3734, pruned_loss=0.1207, over 5675584.37 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3673, pruned_loss=0.1206, over 5691967.78 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.375, pruned_loss=0.1216, over 5679270.22 frames. ], batch size: 145, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:58:23,375 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=635702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:58:53,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5677, 2.0309, 1.7811, 1.3457], device='cuda:0'), covar=tensor([0.3113, 0.1989, 0.2145, 0.2574], device='cuda:0'), in_proj_covar=tensor([0.1817, 0.1714, 0.1669, 0.1791], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 13:58:57,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:58:59,094 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:59:03,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3063, 1.4741, 1.5158, 1.2010], device='cuda:0'), covar=tensor([0.2691, 0.2121, 0.1460, 0.2106], device='cuda:0'), in_proj_covar=tensor([0.1820, 0.1716, 0.1672, 0.1794], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 13:59:06,542 INFO [train.py:968] (0/2) Epoch 14, batch 42450, giga_loss[loss=0.2763, simple_loss=0.3483, pruned_loss=0.1021, over 28595.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3734, pruned_loss=0.1204, over 5683240.16 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3676, pruned_loss=0.1208, over 5695448.13 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3747, pruned_loss=0.1209, over 5682554.75 frames. ], batch size: 71, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:59:14,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.597e+03 2.061e+03 2.749e+03 9.647e+03, threshold=4.123e+03, percent-clipped=8.0
+2023-03-07 13:59:25,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:59:49,963 INFO [train.py:968] (0/2) Epoch 14, batch 42500, giga_loss[loss=0.3212, simple_loss=0.3877, pruned_loss=0.1273, over 28680.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3723, pruned_loss=0.1205, over 5672454.30 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3674, pruned_loss=0.1208, over 5688270.30 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3737, pruned_loss=0.1209, over 5677960.94 frames. ], batch size: 242, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:59:56,207 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6697, 4.9311, 1.9382, 2.1026], device='cuda:0'), covar=tensor([0.0967, 0.0275, 0.0819, 0.1179], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0531, 0.0358, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 14:00:37,322 INFO [train.py:968] (0/2) Epoch 14, batch 42550, giga_loss[loss=0.3244, simple_loss=0.3848, pruned_loss=0.1321, over 28308.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3724, pruned_loss=0.1214, over 5668985.97 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3679, pruned_loss=0.1212, over 5687700.55 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3731, pruned_loss=0.1213, over 5673705.84 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:00:40,014 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=635854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:00:42,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.663e+03 2.305e+03 3.224e+03 5.873e+03, threshold=4.610e+03, percent-clipped=16.0
+2023-03-07 14:00:52,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3087, 3.1232, 2.9789, 1.4010], device='cuda:0'), covar=tensor([0.0911, 0.0981, 0.0948, 0.2268], device='cuda:0'), in_proj_covar=tensor([0.1129, 0.1046, 0.0902, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 14:01:22,312 INFO [train.py:968] (0/2) Epoch 14, batch 42600, giga_loss[loss=0.2965, simple_loss=0.3564, pruned_loss=0.1183, over 28840.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3719, pruned_loss=0.1221, over 5667796.27 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3673, pruned_loss=0.1208, over 5688322.22 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.373, pruned_loss=0.1224, over 5670797.52 frames. ], batch size: 99, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:02:10,110 INFO [train.py:968] (0/2) Epoch 14, batch 42650, giga_loss[loss=0.3596, simple_loss=0.4031, pruned_loss=0.1581, over 27867.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3725, pruned_loss=0.1231, over 5660935.23 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3675, pruned_loss=0.1207, over 5685431.38 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3733, pruned_loss=0.1235, over 5665612.41 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:02:12,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4379, 1.5818, 1.3683, 1.5220], device='cuda:0'), covar=tensor([0.0749, 0.0320, 0.0322, 0.0808], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 14:02:16,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.751e+03 2.311e+03 3.424e+03 8.820e+03, threshold=4.621e+03, percent-clipped=7.0
+2023-03-07 14:02:30,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1591, 2.0677, 1.9860, 1.7807], device='cuda:0'), covar=tensor([0.1566, 0.2547, 0.2048, 0.2266], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0739, 0.0690, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 14:02:57,949 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-636000.pt
+2023-03-07 14:02:58,229 INFO [train.py:968] (0/2) Epoch 14, batch 42700, giga_loss[loss=0.2456, simple_loss=0.3276, pruned_loss=0.08182, over 28905.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.37, pruned_loss=0.1217, over 5673989.63 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.367, pruned_loss=0.1203, over 5689530.57 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3712, pruned_loss=0.1224, over 5673891.24 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:03:42,119 INFO [train.py:968] (0/2) Epoch 14, batch 42750, giga_loss[loss=0.2861, simple_loss=0.3602, pruned_loss=0.106, over 28956.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3703, pruned_loss=0.1219, over 5672123.18 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3676, pruned_loss=0.1206, over 5682838.83 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3708, pruned_loss=0.1222, over 5677707.83 frames. ], batch size: 227, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:03:51,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.068e+02 1.593e+03 2.163e+03 3.115e+03 9.264e+03, threshold=4.326e+03, percent-clipped=8.0
+2023-03-07 14:04:09,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:04:10,696 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-07 14:04:27,038 INFO [train.py:968] (0/2) Epoch 14, batch 42800, giga_loss[loss=0.2802, simple_loss=0.3625, pruned_loss=0.09896, over 28831.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3728, pruned_loss=0.1234, over 5671973.80 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3681, pruned_loss=0.1211, over 5677976.45 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3728, pruned_loss=0.1232, over 5680124.76 frames. ], batch size: 112, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:05:10,883 INFO [train.py:968] (0/2) Epoch 14, batch 42850, giga_loss[loss=0.293, simple_loss=0.3603, pruned_loss=0.1129, over 29001.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3724, pruned_loss=0.1224, over 5672792.25 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3678, pruned_loss=0.1209, over 5684771.98 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3729, pruned_loss=0.1225, over 5673132.02 frames. ], batch size: 106, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:05:20,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.580e+02 1.509e+03 1.907e+03 2.606e+03 6.614e+03, threshold=3.814e+03, percent-clipped=2.0
+2023-03-07 14:05:39,660 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636180.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:05:57,331 INFO [train.py:968] (0/2) Epoch 14, batch 42900, giga_loss[loss=0.2806, simple_loss=0.3543, pruned_loss=0.1035, over 28558.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3723, pruned_loss=0.1217, over 5673237.07 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3677, pruned_loss=0.121, over 5685901.12 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3727, pruned_loss=0.1217, over 5672512.19 frames. ], batch size: 336, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:06:15,644 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=636220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:06:19,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=636223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:06:26,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:06:46,782 INFO [train.py:968] (0/2) Epoch 14, batch 42950, giga_loss[loss=0.3325, simple_loss=0.3853, pruned_loss=0.1399, over 27954.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3725, pruned_loss=0.122, over 5669657.40 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3675, pruned_loss=0.1209, over 5685628.96 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3731, pruned_loss=0.1221, over 5669288.45 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:06:51,302 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=636252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:06:59,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.594e+02 1.579e+03 2.002e+03 2.706e+03 5.877e+03, threshold=4.005e+03, percent-clipped=8.0
+2023-03-07 14:07:23,432 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-07 14:07:37,297 INFO [train.py:968] (0/2) Epoch 14, batch 43000, giga_loss[loss=0.3147, simple_loss=0.3689, pruned_loss=0.1302, over 28979.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3745, pruned_loss=0.1244, over 5662255.26 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3673, pruned_loss=0.1207, over 5686765.81 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3752, pruned_loss=0.1246, over 5660980.66 frames. ], batch size: 136, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:08:27,848 INFO [train.py:968] (0/2) Epoch 14, batch 43050, giga_loss[loss=0.324, simple_loss=0.3804, pruned_loss=0.1338, over 28868.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3781, pruned_loss=0.1286, over 5659202.06 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3674, pruned_loss=0.1207, over 5687561.93 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3787, pruned_loss=0.129, over 5657136.90 frames. ], batch size: 186, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:08:32,290 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-07 14:08:38,945 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 14:08:39,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.990e+02 1.871e+03 2.748e+03 4.403e+03 1.215e+04, threshold=5.495e+03, percent-clipped=28.0
+2023-03-07 14:08:50,982 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=636372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:08:56,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=636375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:09:18,504 INFO [train.py:968] (0/2) Epoch 14, batch 43100, libri_loss[loss=0.3243, simple_loss=0.3854, pruned_loss=0.1316, over 28548.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3796, pruned_loss=0.1311, over 5660392.47 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3676, pruned_loss=0.1208, over 5694186.59 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3803, pruned_loss=0.1316, over 5651569.95 frames. ], batch size: 106, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:09:21,881 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=636404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:09:29,390 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:10:05,002 INFO [train.py:968] (0/2) Epoch 14, batch 43150, giga_loss[loss=0.3427, simple_loss=0.394, pruned_loss=0.1457, over 28731.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3815, pruned_loss=0.133, over 5669227.68 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3681, pruned_loss=0.121, over 5697622.42 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3819, pruned_loss=0.1334, over 5658405.77 frames. ], batch size: 243, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:10:15,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.088e+03 1.911e+03 2.673e+03 4.012e+03 8.713e+03, threshold=5.346e+03, percent-clipped=7.0
+2023-03-07 14:10:30,314 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636477.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:10:34,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4721, 4.3113, 4.0714, 1.8589], device='cuda:0'), covar=tensor([0.0580, 0.0689, 0.0737, 0.2112], device='cuda:0'), in_proj_covar=tensor([0.1137, 0.1054, 0.0911, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 14:10:41,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5147, 2.2360, 1.7246, 0.7962], device='cuda:0'), covar=tensor([0.4752, 0.2342, 0.3245, 0.5241], device='cuda:0'), in_proj_covar=tensor([0.1642, 0.1566, 0.1538, 0.1341], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 14:10:45,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4862, 2.8910, 1.5723, 1.6197], device='cuda:0'), covar=tensor([0.0783, 0.0323, 0.0726, 0.1108], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0532, 0.0359, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 14:10:47,662 INFO [train.py:968] (0/2) Epoch 14, batch 43200, libri_loss[loss=0.3026, simple_loss=0.3727, pruned_loss=0.1163, over 29525.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.378, pruned_loss=0.1301, over 5676257.67 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3684, pruned_loss=0.1212, over 5699718.70 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3783, pruned_loss=0.1305, over 5664902.53 frames. ], batch size: 84, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:11:10,796 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:11:32,594 INFO [train.py:968] (0/2) Epoch 14, batch 43250, giga_loss[loss=0.3698, simple_loss=0.4177, pruned_loss=0.161, over 28305.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3757, pruned_loss=0.1275, over 5680001.63 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.368, pruned_loss=0.1211, over 5705560.46 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3765, pruned_loss=0.1282, over 5664804.98 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:11:36,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636555.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:11:39,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.571e+03 2.093e+03 3.001e+03 9.438e+03, threshold=4.187e+03, percent-clipped=2.0
+2023-03-07 14:12:03,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3529, 4.1594, 1.5615, 1.6623], device='cuda:0'), covar=tensor([0.1022, 0.0267, 0.0895, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0532, 0.0358, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 14:12:15,908 INFO [train.py:968] (0/2) Epoch 14, batch 43300, giga_loss[loss=0.276, simple_loss=0.3507, pruned_loss=0.1007, over 28916.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3735, pruned_loss=0.1244, over 5690365.93 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3679, pruned_loss=0.121, over 5706434.90 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3743, pruned_loss=0.1251, over 5677309.96 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:13:03,018 INFO [train.py:968] (0/2) Epoch 14, batch 43350, giga_loss[loss=0.3232, simple_loss=0.3802, pruned_loss=0.1331, over 28947.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3713, pruned_loss=0.1236, over 5673758.23 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3678, pruned_loss=0.1209, over 5709347.64 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3721, pruned_loss=0.1243, over 5660547.61 frames. ], batch size: 213, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:13:12,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.165e+02 1.571e+03 2.069e+03 2.692e+03 4.804e+03, threshold=4.138e+03, percent-clipped=4.0
+2023-03-07 14:13:47,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=636698.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:13:48,926 INFO [train.py:968] (0/2) Epoch 14, batch 43400, giga_loss[loss=0.2645, simple_loss=0.3378, pruned_loss=0.09564, over 28488.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.37, pruned_loss=0.1231, over 5677910.16 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.368, pruned_loss=0.1208, over 5712061.33 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3706, pruned_loss=0.1238, over 5664162.13 frames. ], batch size: 71, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:13:49,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=636701.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:13:50,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 14:14:14,343 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-07 14:14:17,039 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=636730.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:14:34,600 INFO [train.py:968] (0/2) Epoch 14, batch 43450, giga_loss[loss=0.3258, simple_loss=0.3817, pruned_loss=0.1349, over 28713.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3707, pruned_loss=0.1245, over 5670704.90 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3679, pruned_loss=0.1209, over 5713767.49 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3713, pruned_loss=0.125, over 5657746.76 frames. ], batch size: 119, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:14:45,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.644e+03 2.279e+03 3.422e+03 1.174e+04, threshold=4.558e+03, percent-clipped=19.0
+2023-03-07 14:14:51,548 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:14:51,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9498, 3.7785, 3.5782, 1.7289], device='cuda:0'), covar=tensor([0.0654, 0.0775, 0.0737, 0.2372], device='cuda:0'), in_proj_covar=tensor([0.1132, 0.1050, 0.0906, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 14:15:00,215 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 14:15:06,933 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:15:20,385 INFO [train.py:968] (0/2) Epoch 14, batch 43500, giga_loss[loss=0.2917, simple_loss=0.3663, pruned_loss=0.1086, over 28888.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3736, pruned_loss=0.1255, over 5683040.73 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3681, pruned_loss=0.1208, over 5718127.89 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.374, pruned_loss=0.1261, over 5667653.37 frames. ], batch size: 186, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:15:35,458 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:15:40,556 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-07 14:16:05,864 INFO [train.py:968] (0/2) Epoch 14, batch 43550, giga_loss[loss=0.3517, simple_loss=0.4159, pruned_loss=0.1438, over 28597.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3771, pruned_loss=0.1256, over 5679881.08 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3673, pruned_loss=0.1203, over 5722836.73 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3783, pruned_loss=0.1266, over 5662392.76 frames. ], batch size: 336, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:16:09,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636852.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:16:19,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.711e+02 1.513e+03 2.010e+03 3.256e+03 9.581e+03, threshold=4.019e+03, percent-clipped=7.0
+2023-03-07 14:16:22,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9746, 2.0267, 1.4302, 1.6291], device='cuda:0'), covar=tensor([0.0843, 0.0636, 0.1034, 0.1062], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0446, 0.0505, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 14:16:41,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4594, 1.6586, 1.5965, 1.5574], device='cuda:0'), covar=tensor([0.1555, 0.1743, 0.1918, 0.1696], device='cuda:0'), in_proj_covar=tensor([0.0443, 0.0730, 0.0684, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 14:16:58,467 INFO [train.py:968] (0/2) Epoch 14, batch 43600, giga_loss[loss=0.3248, simple_loss=0.3933, pruned_loss=0.1282, over 28099.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3788, pruned_loss=0.1253, over 5674371.16 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3674, pruned_loss=0.1204, over 5725700.59 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3799, pruned_loss=0.1261, over 5657266.51 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:16:58,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:17:25,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=636930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:17:29,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=636933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:17:44,377 INFO [train.py:968] (0/2) Epoch 14, batch 43650, giga_loss[loss=0.3481, simple_loss=0.4078, pruned_loss=0.1442, over 27930.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3808, pruned_loss=0.1268, over 5676958.93 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3678, pruned_loss=0.1207, over 5720560.83 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3816, pruned_loss=0.1273, over 5665710.12 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:17:54,073 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:17:55,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.730e+02 1.643e+03 2.095e+03 3.276e+03 7.199e+03, threshold=4.190e+03, percent-clipped=18.0
+2023-03-07 14:17:56,578 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=636962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:18:31,003 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=636995.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:18:34,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=636998.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:18:35,105 INFO [train.py:968] (0/2) Epoch 14, batch 43700, giga_loss[loss=0.2944, simple_loss=0.3636, pruned_loss=0.1126, over 29070.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3836, pruned_loss=0.1297, over 5675421.25 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3679, pruned_loss=0.1208, over 5721125.03 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3842, pruned_loss=0.1301, over 5665828.79 frames. ], batch size: 128, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:18:46,531 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 14:18:59,880 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637027.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:19:12,977 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:19:15,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:19:18,956 INFO [train.py:968] (0/2) Epoch 14, batch 43750, giga_loss[loss=0.2644, simple_loss=0.3436, pruned_loss=0.09258, over 28863.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3817, pruned_loss=0.129, over 5679770.61 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.368, pruned_loss=0.121, over 5722169.84 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3821, pruned_loss=0.1292, over 5670880.08 frames. ], batch size: 145, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:19:30,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.614e+03 2.281e+03 3.262e+03 8.763e+03, threshold=4.563e+03, percent-clipped=12.0
+2023-03-07 14:19:44,279 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:20:02,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4556, 2.0904, 1.4267, 0.6745], device='cuda:0'), covar=tensor([0.4954, 0.2301, 0.3543, 0.5425], device='cuda:0'), in_proj_covar=tensor([0.1632, 0.1557, 0.1526, 0.1329], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 14:20:08,590 INFO [train.py:968] (0/2) Epoch 14, batch 43800, giga_loss[loss=0.3755, simple_loss=0.4201, pruned_loss=0.1655, over 27567.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3807, pruned_loss=0.1295, over 5671120.46 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3679, pruned_loss=0.1209, over 5725480.74 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3814, pruned_loss=0.1299, over 5660223.34 frames. ], batch size: 472, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:20:13,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2686, 1.5669, 1.6526, 1.3266], device='cuda:0'), covar=tensor([0.1505, 0.1296, 0.1803, 0.1442], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0733, 0.0688, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 14:20:47,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:20:52,953 INFO [train.py:968] (0/2) Epoch 14, batch 43850, giga_loss[loss=0.2885, simple_loss=0.3468, pruned_loss=0.115, over 28905.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3797, pruned_loss=0.1299, over 5656616.77 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3687, pruned_loss=0.1215, over 5710763.37 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3798, pruned_loss=0.1298, over 5660600.96 frames. ], batch size: 106, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:21:04,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.658e+03 2.229e+03 2.690e+03 1.001e+04, threshold=4.458e+03, percent-clipped=5.0
+2023-03-07 14:21:12,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1590, 2.2176, 2.0088, 1.9678], device='cuda:0'), covar=tensor([0.1570, 0.2167, 0.1882, 0.1975], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0733, 0.0688, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 14:21:24,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5194, 1.6985, 1.7583, 1.3456], device='cuda:0'), covar=tensor([0.1588, 0.2296, 0.1275, 0.1519], device='cuda:0'), in_proj_covar=tensor([0.0855, 0.0695, 0.0895, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 14:21:30,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:21:42,050 INFO [train.py:968] (0/2) Epoch 14, batch 43900, giga_loss[loss=0.3222, simple_loss=0.359, pruned_loss=0.1427, over 23613.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3776, pruned_loss=0.1289, over 5651245.03 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3687, pruned_loss=0.1214, over 5715605.20 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3779, pruned_loss=0.129, over 5648632.21 frames. ], batch size: 705, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:22:27,547 INFO [train.py:968] (0/2) Epoch 14, batch 43950, giga_loss[loss=0.2747, simple_loss=0.3494, pruned_loss=0.09998, over 29084.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3774, pruned_loss=0.1284, over 5655454.21 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3692, pruned_loss=0.1218, over 5716638.60 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3774, pruned_loss=0.1284, over 5650791.06 frames. ], batch size: 155, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:22:35,232 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=637256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:22:41,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.457e+02 1.705e+03 2.244e+03 3.045e+03 6.843e+03, threshold=4.487e+03, percent-clipped=10.0
+2023-03-07 14:22:49,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5769, 3.2918, 2.4497, 2.1035], device='cuda:0'), covar=tensor([0.2026, 0.1080, 0.1580, 0.1579], device='cuda:0'), in_proj_covar=tensor([0.1816, 0.1722, 0.1683, 0.1798], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 14:23:03,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:05,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:14,192 INFO [train.py:968] (0/2) Epoch 14, batch 44000, giga_loss[loss=0.3122, simple_loss=0.3655, pruned_loss=0.1295, over 28777.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3757, pruned_loss=0.1279, over 5649193.63 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3689, pruned_loss=0.1216, over 5710264.10 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3761, pruned_loss=0.1282, over 5649841.75 frames. ], batch size: 99, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:23:28,583 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:44,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:44,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:48,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:58,989 INFO [train.py:968] (0/2) Epoch 14, batch 44050, giga_loss[loss=0.2932, simple_loss=0.3635, pruned_loss=0.1115, over 28945.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3751, pruned_loss=0.1276, over 5654679.55 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3691, pruned_loss=0.1217, over 5707064.95 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3754, pruned_loss=0.1278, over 5657180.10 frames. ], batch size: 106, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:24:01,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6746, 1.8438, 1.9190, 1.4496], device='cuda:0'), covar=tensor([0.1893, 0.2154, 0.1426, 0.1651], device='cuda:0'), in_proj_covar=tensor([0.0856, 0.0694, 0.0896, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 14:24:09,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.411e+02 1.649e+03 2.365e+03 3.626e+03 9.089e+03, threshold=4.729e+03, percent-clipped=14.0
+2023-03-07 14:24:12,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:24:28,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=637385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:24:40,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=637399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:24:41,401 INFO [train.py:968] (0/2) Epoch 14, batch 44100, giga_loss[loss=0.2823, simple_loss=0.3513, pruned_loss=0.1066, over 28851.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3735, pruned_loss=0.126, over 5659103.13 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3693, pruned_loss=0.1217, over 5711782.68 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3736, pruned_loss=0.1263, over 5655322.95 frames. ], batch size: 199, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:25:30,107 INFO [train.py:968] (0/2) Epoch 14, batch 44150, giga_loss[loss=0.3142, simple_loss=0.3806, pruned_loss=0.1239, over 28704.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3762, pruned_loss=0.1274, over 5643098.88 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3696, pruned_loss=0.122, over 5705787.10 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3762, pruned_loss=0.1275, over 5644394.37 frames. ], batch size: 262, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:25:41,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.502e+02 1.799e+03 2.269e+03 3.827e+03 1.046e+04, threshold=4.538e+03, percent-clipped=13.0
+2023-03-07 14:25:53,740 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:25:56,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:25:57,315 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-07 14:26:15,201 INFO [train.py:968] (0/2) Epoch 14, batch 44200, giga_loss[loss=0.3134, simple_loss=0.3776, pruned_loss=0.1246, over 28584.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3789, pruned_loss=0.1292, over 5633484.38 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.37, pruned_loss=0.1224, over 5692188.54 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3786, pruned_loss=0.1291, over 5646279.48 frames. ], batch size: 336, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:26:23,061 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:26:31,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4950, 3.8029, 1.5137, 1.7803], device='cuda:0'), covar=tensor([0.0915, 0.0353, 0.0902, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0532, 0.0358, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 14:26:36,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1993, 1.2081, 3.4496, 3.0516], device='cuda:0'), covar=tensor([0.1600, 0.2619, 0.0506, 0.1407], device='cuda:0'), in_proj_covar=tensor([0.0705, 0.0614, 0.0903, 0.0820], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 14:27:02,753 INFO [train.py:968] (0/2) Epoch 14, batch 44250, giga_loss[loss=0.3035, simple_loss=0.3878, pruned_loss=0.1096, over 28463.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3778, pruned_loss=0.1278, over 5654042.66 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3698, pruned_loss=0.1221, over 5696257.44 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3779, pruned_loss=0.128, over 5659643.81 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:27:14,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.818e+02 1.638e+03 2.120e+03 2.796e+03 4.857e+03, threshold=4.239e+03, percent-clipped=1.0
+2023-03-07 14:27:46,844 INFO [train.py:968] (0/2) Epoch 14, batch 44300, giga_loss[loss=0.3442, simple_loss=0.3969, pruned_loss=0.1458, over 27926.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3786, pruned_loss=0.1253, over 5654813.90 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3699, pruned_loss=0.1219, over 5691597.93 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3788, pruned_loss=0.1258, over 5663062.10 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:27:49,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=637603.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:28:10,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4811, 1.7225, 1.4403, 1.5845], device='cuda:0'), covar=tensor([0.2375, 0.2297, 0.2447, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.1371, 0.1009, 0.1213, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 14:28:14,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:28:31,549 INFO [train.py:968] (0/2) Epoch 14, batch 44350, giga_loss[loss=0.2833, simple_loss=0.3688, pruned_loss=0.09886, over 28883.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3811, pruned_loss=0.125, over 5659874.24 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3701, pruned_loss=0.122, over 5692957.76 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3812, pruned_loss=0.1254, over 5664744.52 frames. ], batch size: 112, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:28:42,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.866e+02 1.474e+03 1.850e+03 2.560e+03 7.036e+03, threshold=3.699e+03, percent-clipped=7.0
+2023-03-07 14:29:16,935 INFO [train.py:968] (0/2) Epoch 14, batch 44400, giga_loss[loss=0.4358, simple_loss=0.4503, pruned_loss=0.2106, over 26628.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3829, pruned_loss=0.1273, over 5644778.53 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3696, pruned_loss=0.1217, over 5689116.30 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3838, pruned_loss=0.128, over 5650564.76 frames. ], batch size: 555, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:30:03,644 INFO [train.py:968] (0/2) Epoch 14, batch 44450, giga_loss[loss=0.2996, simple_loss=0.3672, pruned_loss=0.116, over 28278.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3852, pruned_loss=0.1302, over 5650833.48 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3696, pruned_loss=0.1217, over 5692404.03 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3861, pruned_loss=0.1308, over 5651982.23 frames. ], batch size: 65, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:30:15,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:15,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1339, 1.4393, 1.3953, 1.0472], device='cuda:0'), covar=tensor([0.1401, 0.2196, 0.1189, 0.1416], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0694, 0.0895, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 14:30:16,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5708, 1.5869, 1.2000, 1.1824], device='cuda:0'), covar=tensor([0.0741, 0.0503, 0.0927, 0.1088], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0445, 0.0505, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 14:30:18,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.754e+03 2.416e+03 3.420e+03 7.534e+03, threshold=4.832e+03, percent-clipped=19.0
+2023-03-07 14:30:19,800 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=637766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:26,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:27,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:32,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:50,722 INFO [train.py:968] (0/2) Epoch 14, batch 44500, giga_loss[loss=0.3332, simple_loss=0.39, pruned_loss=0.1382, over 27978.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3843, pruned_loss=0.1302, over 5656572.34 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3688, pruned_loss=0.1211, over 5689589.77 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3864, pruned_loss=0.1316, over 5658961.47 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:30:56,021 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:31:01,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3143, 3.1302, 3.0079, 1.3843], device='cuda:0'), covar=tensor([0.0874, 0.0987, 0.0914, 0.2204], device='cuda:0'), in_proj_covar=tensor([0.1135, 0.1061, 0.0914, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 14:31:27,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8875, 2.4889, 1.0518, 1.1878], device='cuda:0'), covar=tensor([0.1330, 0.0507, 0.1098, 0.1703], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0531, 0.0357, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 14:31:33,106 INFO [train.py:968] (0/2) Epoch 14, batch 44550, giga_loss[loss=0.3005, simple_loss=0.3696, pruned_loss=0.1157, over 28657.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3834, pruned_loss=0.1303, over 5659017.99 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3689, pruned_loss=0.1213, over 5686478.28 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3854, pruned_loss=0.1315, over 5663672.53 frames. ], batch size: 262, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:31:44,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.657e+03 2.343e+03 3.481e+03 9.015e+03, threshold=4.686e+03, percent-clipped=9.0
+2023-03-07 14:32:17,228 INFO [train.py:968] (0/2) Epoch 14, batch 44600, giga_loss[loss=0.2703, simple_loss=0.3539, pruned_loss=0.09339, over 28574.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3821, pruned_loss=0.1285, over 5661679.96 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3691, pruned_loss=0.1213, over 5688628.82 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3837, pruned_loss=0.1295, over 5663277.76 frames. ], batch size: 60, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:32:20,412 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:32:23,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:32:30,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:32:33,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:32:39,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1513, 1.7002, 1.2935, 0.3236], device='cuda:0'), covar=tensor([0.3224, 0.2002, 0.2768, 0.4703], device='cuda:0'), in_proj_covar=tensor([0.1637, 0.1559, 0.1525, 0.1335], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 14:32:48,776 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:33:01,372 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:33:01,798 INFO [train.py:968] (0/2) Epoch 14, batch 44650, libri_loss[loss=0.3061, simple_loss=0.3733, pruned_loss=0.1194, over 29281.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3825, pruned_loss=0.1267, over 5672795.28 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3693, pruned_loss=0.1214, over 5691454.39 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3837, pruned_loss=0.1276, over 5671115.78 frames. ], batch size: 97, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:33:17,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.469e+02 1.361e+03 1.744e+03 2.374e+03 5.412e+03, threshold=3.488e+03, percent-clipped=1.0
+2023-03-07 14:33:27,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637978.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:33:31,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4630, 1.5040, 1.0867, 1.0839], device='cuda:0'), covar=tensor([0.0702, 0.0470, 0.0998, 0.1103], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0446, 0.0505, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 14:33:48,670 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-638000.pt
+2023-03-07 14:33:48,955 INFO [train.py:968] (0/2) Epoch 14, batch 44700, giga_loss[loss=0.3126, simple_loss=0.3865, pruned_loss=0.1194, over 29044.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3825, pruned_loss=0.1263, over 5676644.23 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3691, pruned_loss=0.1213, over 5692186.55 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3838, pruned_loss=0.127, over 5674529.63 frames. ], batch size: 155, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:34:30,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7128, 2.4237, 1.5112, 0.6987], device='cuda:0'), covar=tensor([0.6055, 0.3118, 0.3441, 0.6091], device='cuda:0'), in_proj_covar=tensor([0.1634, 0.1554, 0.1523, 0.1332], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 14:34:41,436 INFO [train.py:968] (0/2) Epoch 14, batch 44750, giga_loss[loss=0.3004, simple_loss=0.3719, pruned_loss=0.1144, over 28939.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3834, pruned_loss=0.1282, over 5661631.87 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3689, pruned_loss=0.1213, over 5694186.22 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3848, pruned_loss=0.129, over 5657864.39 frames. ], batch size: 145, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:34:53,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.633e+03 2.332e+03 3.099e+03 6.966e+03, threshold=4.665e+03, percent-clipped=17.0
+2023-03-07 14:34:54,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7151, 1.8120, 1.5194, 1.8035], device='cuda:0'), covar=tensor([0.2371, 0.2495, 0.2786, 0.2247], device='cuda:0'), in_proj_covar=tensor([0.1372, 0.1006, 0.1214, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 14:35:10,935 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=638084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:35:25,348 INFO [train.py:968] (0/2) Epoch 14, batch 44800, giga_loss[loss=0.3414, simple_loss=0.4012, pruned_loss=0.1408, over 28607.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3812, pruned_loss=0.1271, over 5655223.87 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3691, pruned_loss=0.1215, over 5685644.48 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3822, pruned_loss=0.1276, over 5659711.79 frames. ], batch size: 336, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:35:46,642 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=638121.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:35:48,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=638124.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:36:03,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=638141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:36:10,451 INFO [train.py:968] (0/2) Epoch 14, batch 44850, giga_loss[loss=0.3545, simple_loss=0.3793, pruned_loss=0.1649, over 23417.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3796, pruned_loss=0.1275, over 5653662.91 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3691, pruned_loss=0.1215, over 5687204.88 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3809, pruned_loss=0.128, over 5655199.61 frames. ], batch size: 705, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:36:14,550 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=638153.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:36:26,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.242e+02 1.586e+03 2.256e+03 3.317e+03 2.087e+04, threshold=4.512e+03, percent-clipped=12.0
+2023-03-07 14:36:46,403 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-07 14:36:54,561 INFO [train.py:968] (0/2) Epoch 14, batch 44900, libri_loss[loss=0.3279, simple_loss=0.3737, pruned_loss=0.141, over 29611.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3785, pruned_loss=0.1278, over 5652996.45 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3693, pruned_loss=0.1219, over 5683634.87 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3795, pruned_loss=0.128, over 5656209.43 frames. ], batch size: 74, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:37:02,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 14:37:02,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2636, 0.7487, 0.8450, 1.3965], device='cuda:0'), covar=tensor([0.0753, 0.0378, 0.0346, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 14:37:38,106 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6736, 1.9591, 1.9040, 1.4549], device='cuda:0'), covar=tensor([0.1516, 0.2354, 0.1352, 0.1673], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0697, 0.0898, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 14:37:39,921 INFO [train.py:968] (0/2) Epoch 14, batch 44950, giga_loss[loss=0.3255, simple_loss=0.3782, pruned_loss=0.1364, over 28948.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3757, pruned_loss=0.1263, over 5656234.42 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3687, pruned_loss=0.1216, over 5685607.97 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3772, pruned_loss=0.1268, over 5656681.24 frames. ], batch size: 128, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:37:53,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.956e+02 1.604e+03 2.189e+03 3.264e+03 2.271e+04, threshold=4.379e+03, percent-clipped=13.0
+2023-03-07 14:38:07,230 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=638284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:38:09,408 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=638287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:38:11,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=638289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:38:23,197 INFO [train.py:968] (0/2) Epoch 14, batch 45000, giga_loss[loss=0.2911, simple_loss=0.3613, pruned_loss=0.1105, over 28999.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.376, pruned_loss=0.1273, over 5659244.41 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3694, pruned_loss=0.1221, over 5691829.70 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3768, pruned_loss=0.1275, over 5653245.30 frames. ], batch size: 128, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:38:23,202 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 14:38:30,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3208, 3.1401, 1.4218, 1.5102], device='cuda:0'), covar=tensor([0.1113, 0.0413, 0.1029, 0.1538], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0532, 0.0358, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 14:38:32,169 INFO [train.py:1012] (0/2) Epoch 14, validation: loss=0.2117, simple_loss=0.3198, pruned_loss=0.05174, over 944034.00 frames. 
+2023-03-07 14:38:32,170 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 14:38:46,595 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=638316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:39:14,701 INFO [train.py:968] (0/2) Epoch 14, batch 45050, giga_loss[loss=0.3052, simple_loss=0.3758, pruned_loss=0.1173, over 28275.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3752, pruned_loss=0.1264, over 5653745.05 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3694, pruned_loss=0.1218, over 5697096.27 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.376, pruned_loss=0.1269, over 5642986.66 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:39:26,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3949, 1.5779, 1.2879, 1.4806], device='cuda:0'), covar=tensor([0.0787, 0.0330, 0.0344, 0.0874], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 14:39:26,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.747e+02 1.472e+03 1.820e+03 2.620e+03 6.230e+03, threshold=3.640e+03, percent-clipped=3.0
+2023-03-07 14:39:56,873 INFO [train.py:968] (0/2) Epoch 14, batch 45100, giga_loss[loss=0.295, simple_loss=0.3653, pruned_loss=0.1124, over 28243.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.373, pruned_loss=0.1231, over 5664399.50 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3695, pruned_loss=0.1219, over 5700589.12 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3738, pruned_loss=0.1235, over 5651178.50 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:40:11,525 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=638419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:40:36,678 INFO [train.py:968] (0/2) Epoch 14, batch 45150, giga_loss[loss=0.2986, simple_loss=0.3604, pruned_loss=0.1184, over 28691.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3708, pruned_loss=0.1213, over 5653558.86 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3705, pruned_loss=0.1228, over 5699175.81 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3706, pruned_loss=0.1209, over 5642864.82 frames. ], batch size: 92, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:40:45,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=638459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:40:53,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.309e+02 1.412e+03 1.818e+03 2.593e+03 1.437e+04, threshold=3.636e+03, percent-clipped=14.0
+2023-03-07 14:41:25,930 INFO [train.py:968] (0/2) Epoch 14, batch 45200, giga_loss[loss=0.2577, simple_loss=0.3313, pruned_loss=0.09205, over 28833.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3713, pruned_loss=0.1219, over 5651437.34 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3706, pruned_loss=0.1229, over 5692051.83 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3711, pruned_loss=0.1215, over 5649087.82 frames. ], batch size: 199, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:42:15,649 INFO [train.py:968] (0/2) Epoch 14, batch 45250, giga_loss[loss=0.2509, simple_loss=0.328, pruned_loss=0.08689, over 29051.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3681, pruned_loss=0.1202, over 5666348.88 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3706, pruned_loss=0.1228, over 5693781.30 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3679, pruned_loss=0.1199, over 5662377.00 frames. ], batch size: 155, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:42:21,027 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-07 14:42:30,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.295e+02 1.864e+03 2.350e+03 3.182e+03 1.271e+04, threshold=4.700e+03, percent-clipped=18.0
+2023-03-07 14:43:02,651 INFO [train.py:968] (0/2) Epoch 14, batch 45300, giga_loss[loss=0.2926, simple_loss=0.3731, pruned_loss=0.106, over 29035.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3671, pruned_loss=0.1199, over 5675610.38 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3704, pruned_loss=0.1225, over 5697568.32 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.367, pruned_loss=0.1198, over 5668927.12 frames. ], batch size: 136, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:43:04,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=638602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:43:07,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=638605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:43:32,883 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=638634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:43:44,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5125, 1.5240, 1.6414, 1.4814], device='cuda:0'), covar=tensor([0.2322, 0.2417, 0.1525, 0.1860], device='cuda:0'), in_proj_covar=tensor([0.1819, 0.1730, 0.1690, 0.1805], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 14:43:45,356 INFO [train.py:968] (0/2) Epoch 14, batch 45350, giga_loss[loss=0.3053, simple_loss=0.3788, pruned_loss=0.1159, over 29108.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3689, pruned_loss=0.1202, over 5683053.97 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3704, pruned_loss=0.1224, over 5694244.25 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3688, pruned_loss=0.1202, over 5679887.33 frames. ], batch size: 155, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:43:55,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3673, 1.0479, 4.5689, 3.5769], device='cuda:0'), covar=tensor([0.1787, 0.3069, 0.0356, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0705, 0.0614, 0.0901, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 14:43:58,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=638664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:43:59,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.483e+03 2.000e+03 2.572e+03 7.206e+03, threshold=4.001e+03, percent-clipped=2.0
+2023-03-07 14:44:14,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0951, 1.1063, 3.4446, 3.0148], device='cuda:0'), covar=tensor([0.1718, 0.2739, 0.0584, 0.0984], device='cuda:0'), in_proj_covar=tensor([0.0705, 0.0613, 0.0900, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 14:44:29,272 INFO [train.py:968] (0/2) Epoch 14, batch 45400, giga_loss[loss=0.3198, simple_loss=0.3762, pruned_loss=0.1317, over 28267.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3712, pruned_loss=0.1216, over 5668851.09 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1224, over 5694614.93 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3709, pruned_loss=0.1215, over 5665758.55 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:44:58,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1075, 1.2583, 3.7259, 3.1671], device='cuda:0'), covar=tensor([0.1687, 0.2656, 0.0413, 0.0993], device='cuda:0'), in_proj_covar=tensor([0.0704, 0.0613, 0.0899, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 14:45:15,564 INFO [train.py:968] (0/2) Epoch 14, batch 45450, giga_loss[loss=0.3081, simple_loss=0.3756, pruned_loss=0.1203, over 28995.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3716, pruned_loss=0.122, over 5665958.30 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3706, pruned_loss=0.1224, over 5689960.55 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3714, pruned_loss=0.1219, over 5667817.36 frames. ], batch size: 164, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:45:26,835 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-07 14:45:27,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=638764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:45:29,106 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.955e+02 1.411e+03 2.010e+03 2.872e+03 9.158e+03, threshold=4.020e+03, percent-clipped=15.0
+2023-03-07 14:45:55,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=638794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:46:00,487 INFO [train.py:968] (0/2) Epoch 14, batch 45500, giga_loss[loss=0.3363, simple_loss=0.3922, pruned_loss=0.1402, over 28536.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3738, pruned_loss=0.1242, over 5635887.52 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.1229, over 5663191.92 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3731, pruned_loss=0.1236, over 5660734.05 frames. ], batch size: 336, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:46:07,396 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=638807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:46:10,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=638810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:46:32,867 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=638839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:46:41,287 INFO [train.py:968] (0/2) Epoch 14, batch 45550, giga_loss[loss=0.3221, simple_loss=0.3889, pruned_loss=0.1276, over 29025.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.376, pruned_loss=0.1263, over 5567201.15 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3726, pruned_loss=0.1241, over 5592138.04 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3743, pruned_loss=0.1249, over 5649717.64 frames. ], batch size: 155, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:46:59,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.255e+02 1.536e+03 1.859e+03 2.530e+03 7.638e+03, threshold=3.717e+03, percent-clipped=6.0
+2023-03-07 14:47:33,174 INFO [train.py:968] (0/2) Epoch 14, batch 45600, giga_loss[loss=0.3276, simple_loss=0.3878, pruned_loss=0.1337, over 28818.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3777, pruned_loss=0.1274, over 5562273.11 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3728, pruned_loss=0.1243, over 5575510.22 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3761, pruned_loss=0.1261, over 5642402.94 frames. ], batch size: 284, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:47:49,965 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-07 14:47:51,449 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-14.pt
+2023-03-07 14:48:29,106 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5310, 1.7559, 1.3150, 1.6211], device='cuda:0'), covar=tensor([0.2616, 0.2693, 0.3087, 0.2337], device='cuda:0'), in_proj_covar=tensor([0.1379, 0.1014, 0.1219, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 14:48:33,922 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 14:48:38,832 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=638937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:48:40,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=638940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:49:02,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.373e+02 1.375e+03 1.785e+03 2.249e+03 8.346e+03, threshold=3.571e+03, percent-clipped=5.0
+2023-03-07 14:49:07,059 INFO [train.py:968] (0/2) Epoch 15, batch 50, giga_loss[loss=0.2948, simple_loss=0.3751, pruned_loss=0.1073, over 29018.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3823, pruned_loss=0.1158, over 1265164.70 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3644, pruned_loss=0.1062, over 229139.05 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3857, pruned_loss=0.1177, over 1079198.70 frames. ], batch size: 136, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:49:07,295 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=638969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:49:46,625 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-07 14:49:54,881 INFO [train.py:968] (0/2) Epoch 15, batch 100, giga_loss[loss=0.2579, simple_loss=0.3316, pruned_loss=0.09204, over 28868.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3701, pruned_loss=0.1086, over 2247736.18 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3523, pruned_loss=0.09762, over 368381.27 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3731, pruned_loss=0.1105, over 2008358.21 frames. ], batch size: 99, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:49:55,941 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1347, 2.4171, 2.2735, 1.8424], device='cuda:0'), covar=tensor([0.2115, 0.1819, 0.1596, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.1806, 0.1720, 0.1686, 0.1796], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 14:49:58,297 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 14:50:36,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.032e+02 1.055e+03 1.211e+03 1.570e+03 5.399e+03, threshold=2.422e+03, percent-clipped=1.0
+2023-03-07 14:50:38,199 INFO [train.py:968] (0/2) Epoch 15, batch 150, giga_loss[loss=0.2445, simple_loss=0.3268, pruned_loss=0.08112, over 28623.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3551, pruned_loss=0.1017, over 3015350.57 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3502, pruned_loss=0.09759, over 532454.82 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3565, pruned_loss=0.1027, over 2735646.86 frames. ], batch size: 78, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:51:08,266 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=639102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:51:14,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0781, 3.1046, 2.0271, 1.2915], device='cuda:0'), covar=tensor([0.6178, 0.2579, 0.3343, 0.5426], device='cuda:0'), in_proj_covar=tensor([0.1639, 0.1559, 0.1527, 0.1335], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 14:51:20,786 INFO [train.py:968] (0/2) Epoch 15, batch 200, giga_loss[loss=0.2281, simple_loss=0.3119, pruned_loss=0.07215, over 29098.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3396, pruned_loss=0.09363, over 3617274.70 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3516, pruned_loss=0.09798, over 612407.56 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3391, pruned_loss=0.0936, over 3362064.81 frames. ], batch size: 155, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:51:21,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3374, 3.1256, 3.0047, 1.5151], device='cuda:0'), covar=tensor([0.0875, 0.1135, 0.0964, 0.2214], device='cuda:0'), in_proj_covar=tensor([0.1122, 0.1046, 0.0903, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 14:51:26,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=639125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:51:36,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=639139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:51:42,968 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.33 vs. limit=5.0
+2023-03-07 14:52:02,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.346e+02 1.024e+03 1.285e+03 1.811e+03 4.767e+03, threshold=2.569e+03, percent-clipped=10.0
+2023-03-07 14:52:02,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9986, 1.1567, 1.0883, 0.9313], device='cuda:0'), covar=tensor([0.2119, 0.2342, 0.1328, 0.1869], device='cuda:0'), in_proj_covar=tensor([0.1815, 0.1727, 0.1690, 0.1800], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 14:52:03,948 INFO [train.py:968] (0/2) Epoch 15, batch 250, giga_loss[loss=0.221, simple_loss=0.299, pruned_loss=0.07151, over 28664.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3289, pruned_loss=0.08883, over 4071791.95 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3543, pruned_loss=0.1001, over 652862.95 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3274, pruned_loss=0.08819, over 3862979.73 frames. ], batch size: 262, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:52:42,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=639215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:52:45,631 INFO [train.py:968] (0/2) Epoch 15, batch 300, giga_loss[loss=0.2223, simple_loss=0.2979, pruned_loss=0.07338, over 28541.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3203, pruned_loss=0.08481, over 4428908.00 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3544, pruned_loss=0.09975, over 773621.84 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3179, pruned_loss=0.08383, over 4233580.60 frames. ], batch size: 336, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:53:00,956 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 14:53:26,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.859e+02 9.518e+02 1.182e+03 1.712e+03 4.554e+03, threshold=2.365e+03, percent-clipped=6.0
+2023-03-07 14:53:28,298 INFO [train.py:968] (0/2) Epoch 15, batch 350, giga_loss[loss=0.2083, simple_loss=0.2798, pruned_loss=0.06843, over 28381.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3144, pruned_loss=0.08273, over 4715336.35 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3508, pruned_loss=0.09733, over 950339.99 frames. ], giga_tot_loss[loss=0.2375, simple_loss=0.3115, pruned_loss=0.0817, over 4516377.53 frames. ], batch size: 78, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:53:39,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=639282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:53:42,567 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=639285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:54:01,318 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 14:54:06,515 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=639314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:54:11,871 INFO [train.py:968] (0/2) Epoch 15, batch 400, giga_loss[loss=0.2389, simple_loss=0.2908, pruned_loss=0.09348, over 24135.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.31, pruned_loss=0.08087, over 4934955.88 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3519, pruned_loss=0.09789, over 999340.89 frames. ], giga_tot_loss[loss=0.2333, simple_loss=0.307, pruned_loss=0.07974, over 4769289.64 frames. ], batch size: 705, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:54:50,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.191e+02 1.107e+03 1.492e+03 1.946e+03 4.483e+03, threshold=2.985e+03, percent-clipped=11.0
+2023-03-07 14:54:55,579 INFO [train.py:968] (0/2) Epoch 15, batch 450, giga_loss[loss=0.2282, simple_loss=0.2999, pruned_loss=0.07826, over 29019.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.307, pruned_loss=0.07933, over 5110981.38 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3517, pruned_loss=0.09759, over 1048447.64 frames. ], giga_tot_loss[loss=0.2304, simple_loss=0.3043, pruned_loss=0.07831, over 4972756.16 frames. ], batch size: 155, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:54:57,894 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-07 14:55:11,928 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-07 14:55:21,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7277, 1.8830, 1.7009, 1.5091], device='cuda:0'), covar=tensor([0.2415, 0.1965, 0.1563, 0.2046], device='cuda:0'), in_proj_covar=tensor([0.1814, 0.1717, 0.1680, 0.1795], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 14:55:37,287 INFO [train.py:968] (0/2) Epoch 15, batch 500, giga_loss[loss=0.2014, simple_loss=0.2773, pruned_loss=0.06281, over 28382.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3057, pruned_loss=0.07852, over 5247204.72 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3527, pruned_loss=0.09759, over 1215338.18 frames. ], giga_tot_loss[loss=0.2278, simple_loss=0.3016, pruned_loss=0.07702, over 5113873.22 frames. ], batch size: 60, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:56:19,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.485e+02 9.560e+02 1.148e+03 1.524e+03 4.948e+03, threshold=2.295e+03, percent-clipped=5.0
+2023-03-07 14:56:21,711 INFO [train.py:968] (0/2) Epoch 15, batch 550, libri_loss[loss=0.2793, simple_loss=0.3596, pruned_loss=0.09952, over 29521.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3034, pruned_loss=0.0775, over 5351466.43 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3537, pruned_loss=0.09824, over 1261830.98 frames. ], giga_tot_loss[loss=0.2257, simple_loss=0.2995, pruned_loss=0.07594, over 5241369.17 frames. ], batch size: 82, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:56:27,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=639477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:56:47,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=639500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:57:03,524 INFO [train.py:968] (0/2) Epoch 15, batch 600, giga_loss[loss=0.2198, simple_loss=0.2953, pruned_loss=0.0721, over 28930.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3016, pruned_loss=0.07669, over 5429743.25 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3524, pruned_loss=0.09763, over 1422104.27 frames. ], giga_tot_loss[loss=0.2234, simple_loss=0.297, pruned_loss=0.07487, over 5324805.97 frames. ], batch size: 227, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:57:03,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=639519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:57:44,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9010, 2.2018, 1.4541, 1.7781], device='cuda:0'), covar=tensor([0.0925, 0.0659, 0.1081, 0.1207], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0443, 0.0504, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 14:57:48,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.755e+02 9.583e+02 1.245e+03 1.727e+03 9.193e+03, threshold=2.491e+03, percent-clipped=12.0
+2023-03-07 14:57:49,205 INFO [train.py:968] (0/2) Epoch 15, batch 650, giga_loss[loss=0.208, simple_loss=0.2845, pruned_loss=0.06575, over 28878.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.2992, pruned_loss=0.07531, over 5485177.26 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3522, pruned_loss=0.09717, over 1510579.26 frames. ], giga_tot_loss[loss=0.2209, simple_loss=0.2946, pruned_loss=0.07357, over 5394392.33 frames. ], batch size: 112, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:58:09,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=639590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:58:34,280 INFO [train.py:968] (0/2) Epoch 15, batch 700, libri_loss[loss=0.2903, simple_loss=0.3693, pruned_loss=0.1057, over 29211.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2961, pruned_loss=0.07382, over 5532090.79 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3517, pruned_loss=0.09655, over 1595941.87 frames. ], giga_tot_loss[loss=0.2179, simple_loss=0.2916, pruned_loss=0.07215, over 5454594.58 frames. ], batch size: 97, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:58:36,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=639620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:58:39,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=639623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:58:55,244 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=639643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:58:57,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=639646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:59:04,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=639652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:59:20,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.244e+02 9.067e+02 1.177e+03 1.677e+03 5.865e+03, threshold=2.355e+03, percent-clipped=7.0
+2023-03-07 14:59:21,292 INFO [train.py:968] (0/2) Epoch 15, batch 750, giga_loss[loss=0.2257, simple_loss=0.2927, pruned_loss=0.07939, over 28842.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2929, pruned_loss=0.07216, over 5569567.38 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3505, pruned_loss=0.096, over 1671941.72 frames. ], giga_tot_loss[loss=0.2149, simple_loss=0.2886, pruned_loss=0.07059, over 5509940.38 frames. ], batch size: 199, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:59:24,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3383, 1.1137, 4.0498, 3.2735], device='cuda:0'), covar=tensor([0.1638, 0.2915, 0.0439, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0699, 0.0610, 0.0899, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 14:59:26,827 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=639675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:59:55,014 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5103, 1.9390, 1.5866, 1.6486], device='cuda:0'), covar=tensor([0.0649, 0.0251, 0.0278, 0.0661], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0057, 0.0096], device='cuda:0')
+2023-03-07 14:59:58,502 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 15:00:03,512 INFO [train.py:968] (0/2) Epoch 15, batch 800, giga_loss[loss=0.2387, simple_loss=0.3066, pruned_loss=0.08545, over 28780.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.291, pruned_loss=0.07188, over 5585986.94 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3501, pruned_loss=0.09607, over 1703978.25 frames. ], giga_tot_loss[loss=0.2141, simple_loss=0.2873, pruned_loss=0.07045, over 5545090.33 frames. ], batch size: 99, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 15:00:16,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=639733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:00:19,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=639736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:00:41,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4934, 1.7285, 1.6183, 1.4003], device='cuda:0'), covar=tensor([0.2258, 0.1786, 0.1388, 0.1738], device='cuda:0'), in_proj_covar=tensor([0.1811, 0.1716, 0.1678, 0.1794], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 15:00:44,993 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=639765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:00:50,474 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.105e+02 1.024e+03 1.403e+03 1.973e+03 7.668e+03, threshold=2.806e+03, percent-clipped=18.0
+2023-03-07 15:00:50,487 INFO [train.py:968] (0/2) Epoch 15, batch 850, giga_loss[loss=0.2877, simple_loss=0.3654, pruned_loss=0.105, over 28268.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3022, pruned_loss=0.07795, over 5604347.86 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3502, pruned_loss=0.09574, over 1828738.85 frames. ], giga_tot_loss[loss=0.2252, simple_loss=0.2978, pruned_loss=0.07633, over 5563002.38 frames. ], batch size: 368, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:00:51,647 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-07 15:01:35,679 INFO [train.py:968] (0/2) Epoch 15, batch 900, giga_loss[loss=0.2892, simple_loss=0.3548, pruned_loss=0.1118, over 28668.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3174, pruned_loss=0.08587, over 5623526.85 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3508, pruned_loss=0.09611, over 1886575.38 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3133, pruned_loss=0.08433, over 5589930.16 frames. ], batch size: 99, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:01:50,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7260, 1.9039, 1.4864, 2.0117], device='cuda:0'), covar=tensor([0.2386, 0.2557, 0.2795, 0.2324], device='cuda:0'), in_proj_covar=tensor([0.1383, 0.1015, 0.1224, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 15:02:14,094 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-07 15:02:18,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.098e+02 1.264e+03 1.544e+03 2.118e+03 5.828e+03, threshold=3.089e+03, percent-clipped=12.0
+2023-03-07 15:02:18,776 INFO [train.py:968] (0/2) Epoch 15, batch 950, giga_loss[loss=0.323, simple_loss=0.3871, pruned_loss=0.1295, over 28503.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3292, pruned_loss=0.09163, over 5642746.06 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3508, pruned_loss=0.09596, over 1995754.09 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3252, pruned_loss=0.09029, over 5617363.21 frames. ], batch size: 85, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:02:39,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=639894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:03:00,058 INFO [train.py:968] (0/2) Epoch 15, batch 1000, giga_loss[loss=0.2971, simple_loss=0.3672, pruned_loss=0.1135, over 28738.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.337, pruned_loss=0.09521, over 5654864.06 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3517, pruned_loss=0.09647, over 2053653.11 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3334, pruned_loss=0.09397, over 5631752.59 frames. ], batch size: 92, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:03:38,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.267e+02 1.183e+03 1.469e+03 1.912e+03 4.721e+03, threshold=2.937e+03, percent-clipped=3.0
+2023-03-07 15:03:38,375 INFO [train.py:968] (0/2) Epoch 15, batch 1050, giga_loss[loss=0.2577, simple_loss=0.3452, pruned_loss=0.08512, over 28742.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3401, pruned_loss=0.09542, over 5667603.85 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3507, pruned_loss=0.09604, over 2201319.43 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3371, pruned_loss=0.09453, over 5643539.74 frames. ], batch size: 262, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:04:08,729 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-640000.pt
+2023-03-07 15:04:23,757 INFO [train.py:968] (0/2) Epoch 15, batch 1100, giga_loss[loss=0.2615, simple_loss=0.3425, pruned_loss=0.0903, over 28930.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3417, pruned_loss=0.09554, over 5671863.31 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3502, pruned_loss=0.09569, over 2237849.92 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3394, pruned_loss=0.09496, over 5651154.01 frames. ], batch size: 174, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:04:31,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5378, 1.9451, 1.5242, 1.6306], device='cuda:0'), covar=tensor([0.0760, 0.0283, 0.0309, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 15:04:39,524 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=640037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:04:42,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=640040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:05:06,805 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.212e+02 1.116e+03 1.389e+03 1.761e+03 3.571e+03, threshold=2.778e+03, percent-clipped=4.0
+2023-03-07 15:05:06,817 INFO [train.py:968] (0/2) Epoch 15, batch 1150, giga_loss[loss=0.2407, simple_loss=0.3207, pruned_loss=0.0803, over 28791.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3434, pruned_loss=0.09663, over 5684673.98 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.35, pruned_loss=0.09549, over 2328292.07 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3415, pruned_loss=0.09625, over 5663549.11 frames. ], batch size: 119, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:05:06,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=640069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:05:18,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4825, 1.6285, 1.7281, 1.3138], device='cuda:0'), covar=tensor([0.1733, 0.2565, 0.1396, 0.1745], device='cuda:0'), in_proj_covar=tensor([0.0864, 0.0695, 0.0906, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 15:05:21,414 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 15:05:43,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7383, 1.9977, 1.7260, 1.4960], device='cuda:0'), covar=tensor([0.2214, 0.1965, 0.2031, 0.2190], device='cuda:0'), in_proj_covar=tensor([0.1790, 0.1705, 0.1659, 0.1776], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 15:05:49,324 INFO [train.py:968] (0/2) Epoch 15, batch 1200, libri_loss[loss=0.2985, simple_loss=0.38, pruned_loss=0.1085, over 29624.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.346, pruned_loss=0.09846, over 5680384.23 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3504, pruned_loss=0.09556, over 2448169.54 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3441, pruned_loss=0.09818, over 5659292.29 frames. ], batch size: 91, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 15:06:29,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.766e+02 1.157e+03 1.410e+03 1.763e+03 3.928e+03, threshold=2.820e+03, percent-clipped=5.0
+2023-03-07 15:06:29,419 INFO [train.py:968] (0/2) Epoch 15, batch 1250, giga_loss[loss=0.2832, simple_loss=0.3606, pruned_loss=0.1029, over 28964.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3487, pruned_loss=0.09987, over 5681910.34 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3506, pruned_loss=0.09571, over 2532102.94 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.347, pruned_loss=0.09967, over 5661699.91 frames. ], batch size: 136, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 15:06:32,169 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:07:09,805 INFO [train.py:968] (0/2) Epoch 15, batch 1300, giga_loss[loss=0.3122, simple_loss=0.3818, pruned_loss=0.1213, over 28925.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3503, pruned_loss=0.09985, over 5686745.21 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3502, pruned_loss=0.09545, over 2581138.17 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3492, pruned_loss=0.09986, over 5668970.80 frames. ], batch size: 213, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:07:48,428 INFO [train.py:968] (0/2) Epoch 15, batch 1350, giga_loss[loss=0.2801, simple_loss=0.3635, pruned_loss=0.09836, over 28772.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.352, pruned_loss=0.09946, over 5703162.45 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3506, pruned_loss=0.09542, over 2679301.46 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3509, pruned_loss=0.0996, over 5683541.66 frames. ], batch size: 119, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:07:48,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.295e+02 1.145e+03 1.459e+03 1.881e+03 7.176e+03, threshold=2.918e+03, percent-clipped=5.0
+2023-03-07 15:08:27,749 INFO [train.py:968] (0/2) Epoch 15, batch 1400, giga_loss[loss=0.2848, simple_loss=0.3681, pruned_loss=0.1007, over 29064.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3543, pruned_loss=0.1005, over 5691555.73 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.351, pruned_loss=0.09563, over 2701407.18 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3533, pruned_loss=0.1006, over 5682874.91 frames. ], batch size: 155, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:09:11,707 INFO [train.py:968] (0/2) Epoch 15, batch 1450, libri_loss[loss=0.2409, simple_loss=0.326, pruned_loss=0.07792, over 29593.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3536, pruned_loss=0.09913, over 5698867.91 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3501, pruned_loss=0.09515, over 2763798.13 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3533, pruned_loss=0.09947, over 5689245.24 frames. ], batch size: 75, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:09:12,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.817e+02 1.060e+03 1.359e+03 1.793e+03 4.149e+03, threshold=2.718e+03, percent-clipped=5.0
+2023-03-07 15:09:52,101 INFO [train.py:968] (0/2) Epoch 15, batch 1500, giga_loss[loss=0.2681, simple_loss=0.358, pruned_loss=0.08908, over 28938.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3515, pruned_loss=0.09689, over 5699405.87 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3492, pruned_loss=0.09467, over 2794795.38 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3517, pruned_loss=0.09739, over 5690313.35 frames. ], batch size: 227, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:10:34,142 INFO [train.py:968] (0/2) Epoch 15, batch 1550, giga_loss[loss=0.2562, simple_loss=0.3321, pruned_loss=0.09015, over 28803.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3507, pruned_loss=0.0964, over 5703398.18 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3495, pruned_loss=0.09466, over 2854873.37 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3507, pruned_loss=0.09685, over 5694204.56 frames. ], batch size: 99, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:10:34,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.181e+02 1.147e+03 1.388e+03 1.691e+03 4.730e+03, threshold=2.775e+03, percent-clipped=7.0
+2023-03-07 15:10:41,343 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0321, 1.3864, 1.2554, 1.2016], device='cuda:0'), covar=tensor([0.1633, 0.1342, 0.1941, 0.1516], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0732, 0.0687, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 15:11:15,999 INFO [train.py:968] (0/2) Epoch 15, batch 1600, giga_loss[loss=0.3894, simple_loss=0.4114, pruned_loss=0.1837, over 26459.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3544, pruned_loss=0.1011, over 5696421.68 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3505, pruned_loss=0.09537, over 2955442.60 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3541, pruned_loss=0.1013, over 5699961.78 frames. ], batch size: 555, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:11:34,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-07 15:11:38,744 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-07 15:11:42,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=640548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:11:59,264 INFO [train.py:968] (0/2) Epoch 15, batch 1650, giga_loss[loss=0.282, simple_loss=0.3481, pruned_loss=0.1079, over 28541.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3564, pruned_loss=0.105, over 5687525.11 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3495, pruned_loss=0.09495, over 3057398.13 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3568, pruned_loss=0.1056, over 5695518.43 frames. ], batch size: 92, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:12:01,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.349e+02 1.236e+03 1.607e+03 2.225e+03 3.992e+03, threshold=3.215e+03, percent-clipped=7.0
+2023-03-07 15:12:28,591 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:12:32,299 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.99 vs. limit=5.0
+2023-03-07 15:12:43,194 INFO [train.py:968] (0/2) Epoch 15, batch 1700, giga_loss[loss=0.2706, simple_loss=0.3462, pruned_loss=0.09755, over 28926.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3569, pruned_loss=0.1069, over 5676257.01 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09486, over 3117297.62 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3575, pruned_loss=0.1077, over 5686656.56 frames. ], batch size: 227, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:13:07,499 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-07 15:13:14,116 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:13:16,348 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 15:13:25,082 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:13:26,181 INFO [train.py:968] (0/2) Epoch 15, batch 1750, giga_loss[loss=0.2744, simple_loss=0.3421, pruned_loss=0.1034, over 28857.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3539, pruned_loss=0.1054, over 5682315.31 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3492, pruned_loss=0.0951, over 3174610.27 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3545, pruned_loss=0.1061, over 5695268.93 frames. ], batch size: 199, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:13:28,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.222e+02 1.281e+03 1.521e+03 2.019e+03 5.211e+03, threshold=3.043e+03, percent-clipped=6.0
+2023-03-07 15:13:42,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9575, 1.3051, 1.3388, 1.1184], device='cuda:0'), covar=tensor([0.1717, 0.1295, 0.2114, 0.1500], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0733, 0.0686, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 15:13:44,580 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=640691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:13:46,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=640694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:14:08,250 INFO [train.py:968] (0/2) Epoch 15, batch 1800, giga_loss[loss=0.2869, simple_loss=0.3519, pruned_loss=0.1109, over 28300.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3534, pruned_loss=0.1053, over 5683899.75 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3495, pruned_loss=0.09519, over 3231525.70 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3539, pruned_loss=0.1061, over 5699096.41 frames. ], batch size: 77, lr: 2.18e-03, grad_scale: 2.0
+2023-03-07 15:14:10,919 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=640723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:14:41,256 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:14:48,288 INFO [train.py:968] (0/2) Epoch 15, batch 1850, giga_loss[loss=0.2992, simple_loss=0.3597, pruned_loss=0.1194, over 28557.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3526, pruned_loss=0.1041, over 5696347.95 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3491, pruned_loss=0.09488, over 3271620.67 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3532, pruned_loss=0.105, over 5705546.10 frames. ], batch size: 78, lr: 2.18e-03, grad_scale: 2.0
+2023-03-07 15:14:51,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.531e+02 1.143e+03 1.496e+03 2.137e+03 5.831e+03, threshold=2.992e+03, percent-clipped=6.0
+2023-03-07 15:15:19,454 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-07 15:15:33,441 INFO [train.py:968] (0/2) Epoch 15, batch 1900, giga_loss[loss=0.2502, simple_loss=0.3325, pruned_loss=0.08394, over 28897.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3492, pruned_loss=0.1013, over 5694384.80 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3486, pruned_loss=0.09449, over 3336965.84 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3499, pruned_loss=0.1024, over 5696601.95 frames. ], batch size: 199, lr: 2.18e-03, grad_scale: 2.0
+2023-03-07 15:15:35,316 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:16:15,974 INFO [train.py:968] (0/2) Epoch 15, batch 1950, giga_loss[loss=0.2704, simple_loss=0.3405, pruned_loss=0.1002, over 28699.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3446, pruned_loss=0.09835, over 5693703.24 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3483, pruned_loss=0.09437, over 3425455.59 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3454, pruned_loss=0.09946, over 5688808.10 frames. ], batch size: 284, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 15:16:18,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.657e+02 1.078e+03 1.329e+03 1.799e+03 5.850e+03, threshold=2.659e+03, percent-clipped=4.0
+2023-03-07 15:16:30,960 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2911, 1.3711, 1.2961, 1.2574], device='cuda:0'), covar=tensor([0.1942, 0.2014, 0.1517, 0.1780], device='cuda:0'), in_proj_covar=tensor([0.1791, 0.1708, 0.1665, 0.1776], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 15:17:04,546 INFO [train.py:968] (0/2) Epoch 15, batch 2000, giga_loss[loss=0.2301, simple_loss=0.3065, pruned_loss=0.07685, over 28889.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3383, pruned_loss=0.09524, over 5685547.30 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3482, pruned_loss=0.0944, over 3450122.98 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3389, pruned_loss=0.09611, over 5679727.14 frames. ], batch size: 213, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:17:17,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4267, 1.8393, 1.2882, 0.9374], device='cuda:0'), covar=tensor([0.4595, 0.2450, 0.2466, 0.4172], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1540, 0.1511, 0.1319], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 15:17:25,580 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.47 vs. limit=5.0
+2023-03-07 15:17:36,678 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-07 15:17:48,578 INFO [train.py:968] (0/2) Epoch 15, batch 2050, giga_loss[loss=0.2639, simple_loss=0.3329, pruned_loss=0.09743, over 28606.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3337, pruned_loss=0.09274, over 5680808.04 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3483, pruned_loss=0.09439, over 3543261.28 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3336, pruned_loss=0.09343, over 5671025.13 frames. ], batch size: 85, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:17:52,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.255e+02 1.011e+03 1.301e+03 1.726e+03 9.529e+03, threshold=2.602e+03, percent-clipped=9.0
+2023-03-07 15:17:56,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=640975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:18:33,302 INFO [train.py:968] (0/2) Epoch 15, batch 2100, giga_loss[loss=0.2648, simple_loss=0.3391, pruned_loss=0.0952, over 28802.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3337, pruned_loss=0.09228, over 5683475.02 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3485, pruned_loss=0.0947, over 3568649.43 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3333, pruned_loss=0.09262, over 5681210.77 frames. ], batch size: 119, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:18:35,795 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-07 15:18:40,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:18:43,447 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-07 15:18:50,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-07 15:18:53,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:19:15,300 INFO [train.py:968] (0/2) Epoch 15, batch 2150, giga_loss[loss=0.2678, simple_loss=0.3325, pruned_loss=0.1015, over 28718.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3339, pruned_loss=0.09201, over 5692871.91 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.348, pruned_loss=0.09432, over 3637837.12 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3335, pruned_loss=0.09242, over 5685080.95 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:19:17,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.924e+02 1.087e+03 1.281e+03 1.610e+03 4.070e+03, threshold=2.563e+03, percent-clipped=9.0
+2023-03-07 15:19:53,351 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:19:53,759 INFO [train.py:968] (0/2) Epoch 15, batch 2200, libri_loss[loss=0.2559, simple_loss=0.3425, pruned_loss=0.08466, over 29529.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3339, pruned_loss=0.09185, over 5701316.50 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3484, pruned_loss=0.09427, over 3715887.64 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3328, pruned_loss=0.09214, over 5688439.22 frames. ], batch size: 81, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:19:55,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:08,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:19,829 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=641150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:34,830 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 15:20:35,064 INFO [train.py:968] (0/2) Epoch 15, batch 2250, giga_loss[loss=0.229, simple_loss=0.3041, pruned_loss=0.07698, over 28677.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3312, pruned_loss=0.09065, over 5709872.37 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3483, pruned_loss=0.09408, over 3737633.27 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3303, pruned_loss=0.09095, over 5698010.09 frames. ], batch size: 242, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:20:37,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:37,468 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.30 vs. limit=5.0
+2023-03-07 15:20:37,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.286e+02 1.073e+03 1.288e+03 1.788e+03 5.823e+03, threshold=2.575e+03, percent-clipped=4.0
+2023-03-07 15:20:40,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:48,256 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:50,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:59,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:21:05,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=641203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:21:05,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5288, 1.7658, 1.6317, 1.5789], device='cuda:0'), covar=tensor([0.1704, 0.1952, 0.2296, 0.2055], device='cuda:0'), in_proj_covar=tensor([0.0451, 0.0736, 0.0689, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 15:21:17,894 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=641217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:21:19,074 INFO [train.py:968] (0/2) Epoch 15, batch 2300, giga_loss[loss=0.229, simple_loss=0.3051, pruned_loss=0.07645, over 28797.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3288, pruned_loss=0.08969, over 5709078.55 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3479, pruned_loss=0.09367, over 3777340.84 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.328, pruned_loss=0.09009, over 5699120.39 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:21:58,446 INFO [train.py:968] (0/2) Epoch 15, batch 2350, giga_loss[loss=0.2252, simple_loss=0.3062, pruned_loss=0.07212, over 29016.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3271, pruned_loss=0.08857, over 5718620.86 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3486, pruned_loss=0.09376, over 3848963.60 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3255, pruned_loss=0.0887, over 5706250.51 frames. ], batch size: 155, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:21:59,615 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-07 15:22:00,927 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.332e+02 1.003e+03 1.279e+03 1.729e+03 3.737e+03, threshold=2.557e+03, percent-clipped=7.0
+2023-03-07 15:22:06,060 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:22:08,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:22:31,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7399, 2.0960, 1.8739, 2.0411], device='cuda:0'), covar=tensor([0.0739, 0.0273, 0.0278, 0.0733], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 15:22:32,080 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=641310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:22:38,692 INFO [train.py:968] (0/2) Epoch 15, batch 2400, giga_loss[loss=0.3032, simple_loss=0.3548, pruned_loss=0.1258, over 28889.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3252, pruned_loss=0.08771, over 5727790.91 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3486, pruned_loss=0.09357, over 3899583.85 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3233, pruned_loss=0.0878, over 5714179.05 frames. ], batch size: 145, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:22:54,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:22:57,421 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:23:15,291 INFO [train.py:968] (0/2) Epoch 15, batch 2450, giga_loss[loss=0.2255, simple_loss=0.2998, pruned_loss=0.07561, over 29158.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3232, pruned_loss=0.08673, over 5723462.19 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3489, pruned_loss=0.09349, over 3935716.61 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.321, pruned_loss=0.08672, over 5720061.35 frames. ], batch size: 113, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:23:17,522 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=641371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:23:17,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.960e+02 9.274e+02 1.079e+03 1.434e+03 3.848e+03, threshold=2.158e+03, percent-clipped=6.0
+2023-03-07 15:23:53,185 INFO [train.py:968] (0/2) Epoch 15, batch 2500, giga_loss[loss=0.2277, simple_loss=0.3073, pruned_loss=0.07407, over 28561.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3204, pruned_loss=0.08543, over 5720983.38 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3487, pruned_loss=0.09346, over 3973603.71 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3183, pruned_loss=0.0853, over 5716062.11 frames. ], batch size: 85, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:24:27,941 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=641462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:24:32,981 INFO [train.py:968] (0/2) Epoch 15, batch 2550, giga_loss[loss=0.2439, simple_loss=0.3158, pruned_loss=0.08599, over 28897.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3199, pruned_loss=0.08536, over 5719687.38 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3494, pruned_loss=0.09387, over 4001866.61 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3174, pruned_loss=0.0849, over 5713956.35 frames. ], batch size: 213, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:24:36,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.350e+02 9.199e+02 1.207e+03 1.651e+03 5.824e+03, threshold=2.415e+03, percent-clipped=7.0
+2023-03-07 15:25:06,185 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-07 15:25:11,683 INFO [train.py:968] (0/2) Epoch 15, batch 2600, giga_loss[loss=0.2376, simple_loss=0.3096, pruned_loss=0.08284, over 29049.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3193, pruned_loss=0.0848, over 5720075.91 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3506, pruned_loss=0.09441, over 4048933.72 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3157, pruned_loss=0.08384, over 5719045.54 frames. ], batch size: 128, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:25:17,237 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=641525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:25:20,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5241, 1.6140, 1.7132, 1.3011], device='cuda:0'), covar=tensor([0.1816, 0.2404, 0.1480, 0.1695], device='cuda:0'), in_proj_covar=tensor([0.0867, 0.0698, 0.0909, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 15:25:50,917 INFO [train.py:968] (0/2) Epoch 15, batch 2650, giga_loss[loss=0.2288, simple_loss=0.2998, pruned_loss=0.07888, over 28776.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3185, pruned_loss=0.08453, over 5719531.89 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3509, pruned_loss=0.09445, over 4076434.46 frames. ], giga_tot_loss[loss=0.2411, simple_loss=0.3151, pruned_loss=0.08361, over 5716513.38 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:25:53,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.567e+02 9.859e+02 1.314e+03 1.648e+03 4.074e+03, threshold=2.628e+03, percent-clipped=10.0
+2023-03-07 15:26:32,734 INFO [train.py:968] (0/2) Epoch 15, batch 2700, giga_loss[loss=0.2374, simple_loss=0.3211, pruned_loss=0.07681, over 28850.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3223, pruned_loss=0.08694, over 5718491.16 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3508, pruned_loss=0.09426, over 4129488.90 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3188, pruned_loss=0.08611, over 5711907.78 frames. ], batch size: 174, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:27:17,525 INFO [train.py:968] (0/2) Epoch 15, batch 2750, giga_loss[loss=0.297, simple_loss=0.3683, pruned_loss=0.1128, over 28939.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.328, pruned_loss=0.09058, over 5709857.03 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3504, pruned_loss=0.09396, over 4172761.57 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3249, pruned_loss=0.09, over 5700966.45 frames. ], batch size: 136, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:27:21,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.327e+02 1.287e+03 1.617e+03 2.362e+03 4.300e+03, threshold=3.233e+03, percent-clipped=18.0
+2023-03-07 15:28:01,818 INFO [train.py:968] (0/2) Epoch 15, batch 2800, giga_loss[loss=0.3061, simple_loss=0.367, pruned_loss=0.1226, over 28687.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3364, pruned_loss=0.09623, over 5704681.45 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3504, pruned_loss=0.09383, over 4213816.79 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3335, pruned_loss=0.09583, over 5695054.38 frames. ], batch size: 242, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:28:03,735 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-07 15:28:36,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1863, 0.7883, 0.8362, 1.3928], device='cuda:0'), covar=tensor([0.0801, 0.0400, 0.0365, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 15:28:43,438 INFO [train.py:968] (0/2) Epoch 15, batch 2850, giga_loss[loss=0.3456, simple_loss=0.3919, pruned_loss=0.1496, over 26616.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3418, pruned_loss=0.09888, over 5696235.39 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3498, pruned_loss=0.09329, over 4277684.33 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3394, pruned_loss=0.09908, over 5683327.89 frames. ], batch size: 555, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:28:46,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6251, 1.7520, 1.4730, 1.6615], device='cuda:0'), covar=tensor([0.2502, 0.2546, 0.2706, 0.2588], device='cuda:0'), in_proj_covar=tensor([0.1378, 0.1011, 0.1220, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 15:28:47,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.303e+02 1.319e+03 1.596e+03 2.084e+03 5.377e+03, threshold=3.192e+03, percent-clipped=7.0
+2023-03-07 15:29:25,926 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5767, 1.8908, 1.6944, 1.3997], device='cuda:0'), covar=tensor([0.2167, 0.1832, 0.1962, 0.2067], device='cuda:0'), in_proj_covar=tensor([0.1783, 0.1698, 0.1662, 0.1777], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 15:29:30,191 INFO [train.py:968] (0/2) Epoch 15, batch 2900, giga_loss[loss=0.2852, simple_loss=0.3596, pruned_loss=0.1054, over 28483.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3483, pruned_loss=0.1024, over 5681239.00 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.35, pruned_loss=0.09356, over 4326115.54 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3461, pruned_loss=0.1026, over 5664804.57 frames. ], batch size: 65, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:29:47,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:30:12,565 INFO [train.py:968] (0/2) Epoch 15, batch 2950, giga_loss[loss=0.3151, simple_loss=0.3792, pruned_loss=0.1255, over 28941.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3529, pruned_loss=0.1042, over 5692342.88 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3497, pruned_loss=0.09353, over 4369169.75 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3514, pruned_loss=0.1047, over 5676775.53 frames. ], batch size: 106, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:30:15,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.933e+02 1.094e+03 1.407e+03 1.880e+03 8.941e+03, threshold=2.814e+03, percent-clipped=7.0
+2023-03-07 15:30:25,717 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=641885.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:30:31,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1303, 1.5464, 1.4385, 1.0323], device='cuda:0'), covar=tensor([0.1630, 0.2491, 0.1444, 0.1629], device='cuda:0'), in_proj_covar=tensor([0.0860, 0.0694, 0.0904, 0.0804], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 15:30:41,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:30:55,071 INFO [train.py:968] (0/2) Epoch 15, batch 3000, giga_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08909, over 28911.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3564, pruned_loss=0.1062, over 5676610.28 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3499, pruned_loss=0.09378, over 4425292.19 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3553, pruned_loss=0.1069, over 5671094.32 frames. ], batch size: 145, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:30:55,075 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 15:31:03,255 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2169, simple_loss=0.3212, pruned_loss=0.05625, over 944034.00 frames. 
+2023-03-07 15:31:03,256 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 15:31:33,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1899, 1.4256, 1.2750, 1.1059], device='cuda:0'), covar=tensor([0.2181, 0.1944, 0.1224, 0.1806], device='cuda:0'), in_proj_covar=tensor([0.1788, 0.1700, 0.1663, 0.1780], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 15:31:46,304 INFO [train.py:968] (0/2) Epoch 15, batch 3050, giga_loss[loss=0.2728, simple_loss=0.347, pruned_loss=0.09926, over 28558.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3531, pruned_loss=0.1036, over 5680898.38 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.35, pruned_loss=0.0939, over 4439999.41 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3522, pruned_loss=0.1042, over 5674661.92 frames. ], batch size: 71, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:31:48,387 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=641972.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:31:50,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.376e+02 1.203e+03 1.558e+03 2.433e+03 5.546e+03, threshold=3.115e+03, percent-clipped=18.0
+2023-03-07 15:31:54,790 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:31:56,964 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:32:09,844 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-642000.pt
+2023-03-07 15:32:13,318 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-07 15:32:18,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4220, 3.3313, 1.5082, 1.6264], device='cuda:0'), covar=tensor([0.0990, 0.0295, 0.0904, 0.1344], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0518, 0.0354, 0.0402], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 15:32:19,740 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:32:25,193 INFO [train.py:968] (0/2) Epoch 15, batch 3100, giga_loss[loss=0.2877, simple_loss=0.3609, pruned_loss=0.1073, over 29067.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3507, pruned_loss=0.1015, over 5693180.80 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3501, pruned_loss=0.09422, over 4511858.01 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3499, pruned_loss=0.1022, over 5677870.46 frames. ], batch size: 128, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:32:25,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3006, 1.7916, 1.3209, 0.4883], device='cuda:0'), covar=tensor([0.3776, 0.1903, 0.3220, 0.5063], device='cuda:0'), in_proj_covar=tensor([0.1626, 0.1546, 0.1529, 0.1333], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 15:32:46,797 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:32:49,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:33:05,688 INFO [train.py:968] (0/2) Epoch 15, batch 3150, libri_loss[loss=0.3019, simple_loss=0.3809, pruned_loss=0.1114, over 29664.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3493, pruned_loss=0.1001, over 5685475.56 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3501, pruned_loss=0.09432, over 4562562.63 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3487, pruned_loss=0.1008, over 5668609.37 frames. ], batch size: 88, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:33:10,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.213e+02 1.172e+03 1.600e+03 2.157e+03 5.438e+03, threshold=3.200e+03, percent-clipped=7.0
+2023-03-07 15:33:11,643 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:33:34,758 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6076, 1.8110, 1.9082, 1.4235], device='cuda:0'), covar=tensor([0.1868, 0.2407, 0.1478, 0.1727], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0692, 0.0903, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 15:33:45,771 INFO [train.py:968] (0/2) Epoch 15, batch 3200, giga_loss[loss=0.3027, simple_loss=0.3754, pruned_loss=0.1151, over 28606.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3492, pruned_loss=0.09979, over 5678695.73 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3499, pruned_loss=0.09414, over 4600235.65 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3488, pruned_loss=0.1006, over 5666500.89 frames. ], batch size: 307, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:34:24,643 INFO [train.py:968] (0/2) Epoch 15, batch 3250, giga_loss[loss=0.2728, simple_loss=0.3478, pruned_loss=0.09892, over 28882.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3518, pruned_loss=0.1013, over 5686245.13 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3499, pruned_loss=0.09428, over 4638890.55 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3516, pruned_loss=0.102, over 5670564.01 frames. ], batch size: 119, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:34:29,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.272e+02 1.214e+03 1.571e+03 2.011e+03 6.100e+03, threshold=3.142e+03, percent-clipped=6.0
+2023-03-07 15:34:47,423 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:35:04,340 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:35:08,473 INFO [train.py:968] (0/2) Epoch 15, batch 3300, giga_loss[loss=0.2957, simple_loss=0.3648, pruned_loss=0.1133, over 28896.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3538, pruned_loss=0.1027, over 5696763.56 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3499, pruned_loss=0.09424, over 4645331.10 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3537, pruned_loss=0.1034, over 5683398.66 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:35:13,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0622, 2.9118, 1.8890, 1.3160], device='cuda:0'), covar=tensor([0.5789, 0.2285, 0.3348, 0.4681], device='cuda:0'), in_proj_covar=tensor([0.1622, 0.1539, 0.1523, 0.1326], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 15:35:41,005 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3118, 1.3978, 1.4582, 1.2330], device='cuda:0'), covar=tensor([0.1943, 0.1924, 0.1399, 0.1707], device='cuda:0'), in_proj_covar=tensor([0.1783, 0.1700, 0.1661, 0.1774], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 15:35:44,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=642260.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 15:35:50,777 INFO [train.py:968] (0/2) Epoch 15, batch 3350, libri_loss[loss=0.3217, simple_loss=0.39, pruned_loss=0.1267, over 19622.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3546, pruned_loss=0.104, over 5687247.29 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3495, pruned_loss=0.09397, over 4666004.16 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3549, pruned_loss=0.1049, over 5682198.66 frames. ], batch size: 186, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:35:56,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.908e+02 1.206e+03 1.539e+03 1.923e+03 4.026e+03, threshold=3.077e+03, percent-clipped=4.0
+2023-03-07 15:36:23,799 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:36:34,613 INFO [train.py:968] (0/2) Epoch 15, batch 3400, giga_loss[loss=0.2764, simple_loss=0.349, pruned_loss=0.1018, over 28782.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3569, pruned_loss=0.1063, over 5685175.22 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.35, pruned_loss=0.09419, over 4690793.84 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3569, pruned_loss=0.1071, over 5677224.12 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:36:58,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=642347.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:37:08,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6286, 1.8829, 1.5333, 1.7230], device='cuda:0'), covar=tensor([0.2174, 0.2019, 0.2086, 0.2108], device='cuda:0'), in_proj_covar=tensor([0.1374, 0.1009, 0.1216, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 15:37:14,099 INFO [train.py:968] (0/2) Epoch 15, batch 3450, giga_loss[loss=0.2846, simple_loss=0.3532, pruned_loss=0.1081, over 28526.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3572, pruned_loss=0.1066, over 5684500.27 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3505, pruned_loss=0.09423, over 4730953.88 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.357, pruned_loss=0.1076, over 5672285.84 frames. ], batch size: 85, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:37:18,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.242e+02 1.216e+03 1.509e+03 2.098e+03 6.056e+03, threshold=3.017e+03, percent-clipped=5.0
+2023-03-07 15:37:22,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5541, 1.8716, 1.8141, 1.3792], device='cuda:0'), covar=tensor([0.1592, 0.2532, 0.1401, 0.1730], device='cuda:0'), in_proj_covar=tensor([0.0860, 0.0694, 0.0903, 0.0805], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 15:37:31,767 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:37:38,445 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642403.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:37:40,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642406.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:37:51,627 INFO [train.py:968] (0/2) Epoch 15, batch 3500, giga_loss[loss=0.267, simple_loss=0.3452, pruned_loss=0.09434, over 28744.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3564, pruned_loss=0.1053, over 5692598.38 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3497, pruned_loss=0.09417, over 4777267.55 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3571, pruned_loss=0.1067, over 5682627.65 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:38:02,339 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642435.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:38:22,223 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-07 15:38:28,251 INFO [train.py:968] (0/2) Epoch 15, batch 3550, giga_loss[loss=0.2971, simple_loss=0.3772, pruned_loss=0.1085, over 28898.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3562, pruned_loss=0.1042, over 5700401.93 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3497, pruned_loss=0.09436, over 4816409.21 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3571, pruned_loss=0.1055, over 5685450.01 frames. ], batch size: 227, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:38:33,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.019e+02 1.139e+03 1.544e+03 2.289e+03 7.677e+03, threshold=3.088e+03, percent-clipped=12.0
+2023-03-07 15:38:46,723 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642490.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:38:49,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642493.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 15:39:10,166 INFO [train.py:968] (0/2) Epoch 15, batch 3600, giga_loss[loss=0.2934, simple_loss=0.3607, pruned_loss=0.113, over 28748.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3563, pruned_loss=0.1035, over 5697354.07 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3495, pruned_loss=0.09412, over 4831953.40 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3571, pruned_loss=0.1048, over 5683385.75 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:39:11,152 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 15:39:12,265 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642522.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:39:26,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2405, 2.4069, 1.3364, 1.3781], device='cuda:0'), covar=tensor([0.0917, 0.0353, 0.0862, 0.1280], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0517, 0.0354, 0.0402], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 15:39:31,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9817, 2.1407, 2.2301, 1.7981], device='cuda:0'), covar=tensor([0.1595, 0.2217, 0.1237, 0.1545], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0695, 0.0904, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 15:39:47,173 INFO [train.py:968] (0/2) Epoch 15, batch 3650, giga_loss[loss=0.2731, simple_loss=0.3464, pruned_loss=0.09992, over 28861.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3543, pruned_loss=0.1022, over 5708160.86 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.349, pruned_loss=0.09389, over 4847964.29 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3554, pruned_loss=0.1036, over 5694597.62 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:39:48,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=642570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:39:51,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.853e+02 1.024e+03 1.247e+03 1.562e+03 4.913e+03, threshold=2.495e+03, percent-clipped=4.0
+2023-03-07 15:40:05,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=642589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:40:17,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3367, 3.0989, 1.5487, 1.4567], device='cuda:0'), covar=tensor([0.0971, 0.0292, 0.0841, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0516, 0.0353, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 15:40:28,438 INFO [train.py:968] (0/2) Epoch 15, batch 3700, giga_loss[loss=0.2865, simple_loss=0.3672, pruned_loss=0.1029, over 29073.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3518, pruned_loss=0.1012, over 5692602.04 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3494, pruned_loss=0.09423, over 4847920.79 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3524, pruned_loss=0.1021, over 5689392.07 frames. ], batch size: 128, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:40:31,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4206, 3.1990, 1.4495, 1.6002], device='cuda:0'), covar=tensor([0.0996, 0.0264, 0.0874, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0516, 0.0353, 0.0401], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 15:40:43,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:40:49,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2710, 1.2720, 4.0671, 3.3261], device='cuda:0'), covar=tensor([0.1644, 0.2635, 0.0388, 0.0765], device='cuda:0'), in_proj_covar=tensor([0.0696, 0.0606, 0.0886, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 15:41:06,746 INFO [train.py:968] (0/2) Epoch 15, batch 3750, giga_loss[loss=0.2373, simple_loss=0.318, pruned_loss=0.07833, over 28309.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3493, pruned_loss=0.09935, over 5697367.34 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3497, pruned_loss=0.09415, over 4882007.19 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3496, pruned_loss=0.1003, over 5694457.04 frames. ], batch size: 65, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:41:10,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.284e+02 1.051e+03 1.229e+03 1.430e+03 3.477e+03, threshold=2.457e+03, percent-clipped=5.0
+2023-03-07 15:41:18,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=642683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:41:31,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:41:47,384 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:41:49,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:41:51,399 INFO [train.py:968] (0/2) Epoch 15, batch 3800, giga_loss[loss=0.2725, simple_loss=0.3507, pruned_loss=0.09719, over 28761.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3499, pruned_loss=0.1003, over 5695476.96 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3496, pruned_loss=0.09405, over 4887114.15 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3502, pruned_loss=0.1012, over 5692394.63 frames. ], batch size: 119, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:42:01,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5624, 2.1726, 1.5248, 0.7481], device='cuda:0'), covar=tensor([0.4646, 0.2228, 0.3432, 0.5059], device='cuda:0'), in_proj_covar=tensor([0.1627, 0.1536, 0.1521, 0.1325], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 15:42:01,199 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:42:03,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:42:13,919 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:42:16,875 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-07 15:42:27,289 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:42:30,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=642768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:42:30,938 INFO [train.py:968] (0/2) Epoch 15, batch 3850, giga_loss[loss=0.3042, simple_loss=0.3747, pruned_loss=0.1169, over 28840.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3506, pruned_loss=0.1008, over 5686708.62 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3496, pruned_loss=0.09415, over 4892581.87 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3509, pruned_loss=0.1015, over 5697198.90 frames. ], batch size: 66, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:42:35,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.302e+02 1.012e+03 1.299e+03 1.703e+03 5.001e+03, threshold=2.598e+03, percent-clipped=6.0
+2023-03-07 15:43:10,792 INFO [train.py:968] (0/2) Epoch 15, batch 3900, giga_loss[loss=0.2884, simple_loss=0.3607, pruned_loss=0.108, over 28918.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3508, pruned_loss=0.09992, over 5691876.65 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3498, pruned_loss=0.09424, over 4909579.64 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3508, pruned_loss=0.1005, over 5699201.42 frames. ], batch size: 186, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:43:16,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:43:18,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:43:43,173 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:43:43,960 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:43:46,146 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-07 15:43:52,538 INFO [train.py:968] (0/2) Epoch 15, batch 3950, giga_loss[loss=0.3032, simple_loss=0.3612, pruned_loss=0.1226, over 28919.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3507, pruned_loss=0.09969, over 5695052.49 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3502, pruned_loss=0.09456, over 4925198.36 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3505, pruned_loss=0.1, over 5704513.25 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:43:58,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.928e+02 9.670e+02 1.204e+03 1.593e+03 5.317e+03, threshold=2.407e+03, percent-clipped=3.0
+2023-03-07 15:44:18,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2903, 1.2587, 1.2388, 1.5209], device='cuda:0'), covar=tensor([0.0816, 0.0353, 0.0340, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 15:44:25,471 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:44:28,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:44:32,451 INFO [train.py:968] (0/2) Epoch 15, batch 4000, giga_loss[loss=0.3356, simple_loss=0.3821, pruned_loss=0.1445, over 26632.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3514, pruned_loss=0.1006, over 5677622.10 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3499, pruned_loss=0.09442, over 4930592.73 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3514, pruned_loss=0.1012, over 5695933.69 frames. ], batch size: 555, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:44:46,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4292, 1.5990, 1.2441, 1.1812], device='cuda:0'), covar=tensor([0.0926, 0.0542, 0.1023, 0.1061], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0439, 0.0506, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 15:44:50,986 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:45:12,985 INFO [train.py:968] (0/2) Epoch 15, batch 4050, giga_loss[loss=0.2397, simple_loss=0.3173, pruned_loss=0.08104, over 28975.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3486, pruned_loss=0.09932, over 5693337.66 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3497, pruned_loss=0.09432, over 4935559.77 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3488, pruned_loss=0.09985, over 5706522.97 frames. ], batch size: 106, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:45:17,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.150e+02 9.796e+02 1.233e+03 1.657e+03 4.170e+03, threshold=2.467e+03, percent-clipped=9.0
+2023-03-07 15:45:31,432 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-07 15:45:40,968 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=643007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:45:44,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:45:47,253 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-07 15:45:50,295 INFO [train.py:968] (0/2) Epoch 15, batch 4100, giga_loss[loss=0.2677, simple_loss=0.3514, pruned_loss=0.09194, over 28583.00 frames. ], tot_loss[loss=0.271, simple_loss=0.346, pruned_loss=0.09806, over 5693479.39 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.35, pruned_loss=0.09452, over 4954367.20 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3459, pruned_loss=0.09846, over 5706888.16 frames. ], batch size: 336, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:46:16,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4903, 1.7353, 1.4034, 1.5194], device='cuda:0'), covar=tensor([0.0757, 0.0296, 0.0316, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 15:46:20,684 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=643059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:46:27,692 INFO [train.py:968] (0/2) Epoch 15, batch 4150, giga_loss[loss=0.3004, simple_loss=0.3726, pruned_loss=0.1141, over 27916.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3437, pruned_loss=0.09687, over 5700572.33 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3503, pruned_loss=0.09463, over 4973511.69 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3433, pruned_loss=0.09715, over 5707489.32 frames. ], batch size: 412, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:46:31,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:46:33,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.779e+02 1.155e+03 1.522e+03 2.052e+03 6.667e+03, threshold=3.043e+03, percent-clipped=15.0
+2023-03-07 15:47:06,794 INFO [train.py:968] (0/2) Epoch 15, batch 4200, giga_loss[loss=0.2613, simple_loss=0.326, pruned_loss=0.09828, over 28767.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3423, pruned_loss=0.09608, over 5707460.02 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3502, pruned_loss=0.09468, over 5007589.83 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3418, pruned_loss=0.09634, over 5708101.29 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:47:27,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4809, 4.3116, 4.1104, 2.1663], device='cuda:0'), covar=tensor([0.0602, 0.0790, 0.0802, 0.1860], device='cuda:0'), in_proj_covar=tensor([0.1101, 0.1026, 0.0888, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 15:47:35,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9603, 1.1964, 1.3288, 1.0676], device='cuda:0'), covar=tensor([0.1561, 0.1283, 0.1978, 0.1555], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0733, 0.0690, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 15:47:36,845 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=643155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:47:38,839 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=643158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:47:40,291 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6435, 1.6914, 1.8852, 1.4340], device='cuda:0'), covar=tensor([0.1786, 0.2244, 0.1432, 0.1664], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0696, 0.0904, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 15:47:48,721 INFO [train.py:968] (0/2) Epoch 15, batch 4250, giga_loss[loss=0.2741, simple_loss=0.3405, pruned_loss=0.1039, over 28672.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3411, pruned_loss=0.09595, over 5709417.69 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3505, pruned_loss=0.09485, over 5018971.67 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3403, pruned_loss=0.09602, over 5708960.01 frames. ], batch size: 262, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:47:57,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.817e+02 1.136e+03 1.360e+03 1.836e+03 5.078e+03, threshold=2.720e+03, percent-clipped=3.0
+2023-03-07 15:48:05,203 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=643187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:48:27,827 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=643216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:48:29,408 INFO [train.py:968] (0/2) Epoch 15, batch 4300, giga_loss[loss=0.2155, simple_loss=0.2952, pruned_loss=0.06792, over 28754.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3402, pruned_loss=0.0962, over 5700962.01 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3503, pruned_loss=0.09474, over 5035480.38 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3395, pruned_loss=0.09639, over 5704992.22 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:48:29,673 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=643219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:48:29,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5138, 2.1920, 1.6370, 0.7238], device='cuda:0'), covar=tensor([0.5609, 0.2658, 0.3627, 0.5737], device='cuda:0'), in_proj_covar=tensor([0.1625, 0.1531, 0.1520, 0.1323], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 15:48:39,185 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:48:52,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=643248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:49:08,134 INFO [train.py:968] (0/2) Epoch 15, batch 4350, giga_loss[loss=0.2573, simple_loss=0.3155, pruned_loss=0.09956, over 28794.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.338, pruned_loss=0.09529, over 5703192.12 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3502, pruned_loss=0.09463, over 5056060.90 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3373, pruned_loss=0.09555, over 5705935.27 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:49:14,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.581e+02 1.060e+03 1.302e+03 1.736e+03 1.082e+04, threshold=2.603e+03, percent-clipped=11.0
+2023-03-07 15:49:28,640 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-07 15:49:47,139 INFO [train.py:968] (0/2) Epoch 15, batch 4400, giga_loss[loss=0.3008, simple_loss=0.364, pruned_loss=0.1188, over 28905.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3374, pruned_loss=0.09525, over 5702620.31 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3503, pruned_loss=0.09461, over 5075730.72 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3364, pruned_loss=0.09547, over 5701618.49 frames. ], batch size: 174, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:50:05,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3418, 3.2659, 1.4440, 1.4912], device='cuda:0'), covar=tensor([0.0942, 0.0323, 0.0945, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0516, 0.0352, 0.0400], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 15:50:29,129 INFO [train.py:968] (0/2) Epoch 15, batch 4450, giga_loss[loss=0.2535, simple_loss=0.3355, pruned_loss=0.08576, over 28696.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3381, pruned_loss=0.0952, over 5704768.82 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3505, pruned_loss=0.09462, over 5083347.62 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3371, pruned_loss=0.09537, over 5702870.47 frames. ], batch size: 242, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:50:34,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=643375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:50:35,679 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.788e+02 1.014e+03 1.191e+03 1.411e+03 2.814e+03, threshold=2.383e+03, percent-clipped=2.0
+2023-03-07 15:50:36,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=643378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:50:40,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:50:51,264 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-07 15:51:01,868 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=643407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:51:11,892 INFO [train.py:968] (0/2) Epoch 15, batch 4500, giga_loss[loss=0.3073, simple_loss=0.3777, pruned_loss=0.1184, over 27926.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3397, pruned_loss=0.09587, over 5709920.70 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3502, pruned_loss=0.09451, over 5099915.01 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3389, pruned_loss=0.09611, over 5704778.97 frames. ], batch size: 412, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:51:25,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:51:52,888 INFO [train.py:968] (0/2) Epoch 15, batch 4550, giga_loss[loss=0.2462, simple_loss=0.32, pruned_loss=0.08617, over 28739.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3417, pruned_loss=0.0962, over 5716845.16 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3501, pruned_loss=0.09455, over 5108083.37 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.341, pruned_loss=0.09638, over 5710941.72 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:51:59,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.044e+02 9.937e+02 1.127e+03 1.538e+03 3.359e+03, threshold=2.255e+03, percent-clipped=3.0
+2023-03-07 15:52:07,072 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 15:52:09,396 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=643488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:52:32,299 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=643511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:52:38,897 INFO [train.py:968] (0/2) Epoch 15, batch 4600, giga_loss[loss=0.2612, simple_loss=0.3311, pruned_loss=0.09559, over 28774.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3437, pruned_loss=0.09699, over 5707079.73 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3502, pruned_loss=0.09467, over 5124195.85 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.343, pruned_loss=0.09709, over 5698611.44 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:52:42,907 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=643525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:52:44,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=643528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:53:08,854 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=643557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:53:17,844 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-07 15:53:18,611 INFO [train.py:968] (0/2) Epoch 15, batch 4650, giga_loss[loss=0.2663, simple_loss=0.347, pruned_loss=0.09278, over 28662.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3428, pruned_loss=0.09573, over 5702795.92 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3501, pruned_loss=0.09455, over 5143978.71 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3422, pruned_loss=0.09595, over 5693601.40 frames. ], batch size: 242, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:53:25,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.323e+02 1.027e+03 1.203e+03 1.560e+03 3.660e+03, threshold=2.406e+03, percent-clipped=10.0
+2023-03-07 15:53:25,967 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=643577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:53:28,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=643580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:53:36,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5962, 1.7268, 1.8922, 1.3953], device='cuda:0'), covar=tensor([0.1915, 0.2286, 0.1561, 0.1753], device='cuda:0'), in_proj_covar=tensor([0.0857, 0.0692, 0.0900, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 15:53:53,181 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=643609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:54:00,235 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-07 15:54:01,135 INFO [train.py:968] (0/2) Epoch 15, batch 4700, giga_loss[loss=0.2539, simple_loss=0.3359, pruned_loss=0.08601, over 28897.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3427, pruned_loss=0.09523, over 5705311.01 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3502, pruned_loss=0.09456, over 5147716.82 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3421, pruned_loss=0.09539, over 5697276.71 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:54:39,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4895, 1.5956, 1.5985, 1.3678], device='cuda:0'), covar=tensor([0.2996, 0.2295, 0.1813, 0.2349], device='cuda:0'), in_proj_covar=tensor([0.1790, 0.1709, 0.1665, 0.1771], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 15:54:44,487 INFO [train.py:968] (0/2) Epoch 15, batch 4750, libri_loss[loss=0.3028, simple_loss=0.368, pruned_loss=0.1188, over 29342.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3437, pruned_loss=0.0961, over 5712954.26 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3504, pruned_loss=0.09475, over 5162809.10 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3429, pruned_loss=0.0961, over 5702817.73 frames. ], batch size: 67, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:54:50,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.982e+02 1.218e+03 1.490e+03 1.966e+03 4.766e+03, threshold=2.979e+03, percent-clipped=8.0
+2023-03-07 15:55:02,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2353, 1.7455, 1.3201, 0.3702], device='cuda:0'), covar=tensor([0.3722, 0.2392, 0.3808, 0.5479], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1529, 0.1523, 0.1322], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 15:55:21,051 INFO [train.py:968] (0/2) Epoch 15, batch 4800, giga_loss[loss=0.2748, simple_loss=0.3542, pruned_loss=0.09769, over 28661.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3444, pruned_loss=0.09668, over 5718663.36 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3502, pruned_loss=0.09464, over 5181039.49 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3437, pruned_loss=0.09681, over 5706366.72 frames. ], batch size: 307, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:55:52,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3002, 1.5533, 1.2264, 1.3075], device='cuda:0'), covar=tensor([0.2684, 0.2667, 0.3135, 0.2355], device='cuda:0'), in_proj_covar=tensor([0.1374, 0.1010, 0.1218, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 15:56:03,268 INFO [train.py:968] (0/2) Epoch 15, batch 4850, giga_loss[loss=0.3, simple_loss=0.3703, pruned_loss=0.1148, over 28886.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3478, pruned_loss=0.09898, over 5720183.81 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3502, pruned_loss=0.09461, over 5202238.52 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3472, pruned_loss=0.09923, over 5705244.02 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:56:08,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.725e+02 1.301e+03 1.864e+03 2.287e+03 6.098e+03, threshold=3.728e+03, percent-clipped=12.0
+2023-03-07 15:56:42,155 INFO [train.py:968] (0/2) Epoch 15, batch 4900, giga_loss[loss=0.267, simple_loss=0.348, pruned_loss=0.09302, over 28734.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3495, pruned_loss=0.09941, over 5718138.80 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3507, pruned_loss=0.09488, over 5217896.61 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3485, pruned_loss=0.0995, over 5703386.62 frames. ], batch size: 262, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:57:19,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:57:24,234 INFO [train.py:968] (0/2) Epoch 15, batch 4950, giga_loss[loss=0.2765, simple_loss=0.3617, pruned_loss=0.09565, over 28942.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3505, pruned_loss=0.09955, over 5718565.68 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3507, pruned_loss=0.09476, over 5231260.78 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3497, pruned_loss=0.0998, over 5703767.35 frames. ], batch size: 164, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:57:30,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.733e+02 1.135e+03 1.401e+03 1.848e+03 3.974e+03, threshold=2.803e+03, percent-clipped=2.0
+2023-03-07 15:57:36,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:58:03,595 INFO [train.py:968] (0/2) Epoch 15, batch 5000, giga_loss[loss=0.3098, simple_loss=0.3804, pruned_loss=0.1196, over 28807.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3517, pruned_loss=0.1006, over 5715430.75 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3507, pruned_loss=0.09474, over 5237954.76 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3511, pruned_loss=0.1009, over 5702243.44 frames. ], batch size: 284, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:58:07,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=643922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:58:38,777 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=643959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:58:45,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-07 15:58:45,842 INFO [train.py:968] (0/2) Epoch 15, batch 5050, giga_loss[loss=0.2524, simple_loss=0.338, pruned_loss=0.08341, over 28726.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3515, pruned_loss=0.1004, over 5716328.86 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3503, pruned_loss=0.0945, over 5252366.67 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3514, pruned_loss=0.101, over 5703223.00 frames. ], batch size: 262, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:58:53,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.255e+02 1.143e+03 1.444e+03 1.833e+03 4.663e+03, threshold=2.887e+03, percent-clipped=8.0
+2023-03-07 15:59:11,601 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-644000.pt
+2023-03-07 15:59:14,474 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3810, 1.3037, 1.1221, 1.4953], device='cuda:0'), covar=tensor([0.0712, 0.0343, 0.0354, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 15:59:16,464 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:59:18,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:59:26,794 INFO [train.py:968] (0/2) Epoch 15, batch 5100, giga_loss[loss=0.2898, simple_loss=0.3705, pruned_loss=0.1046, over 28980.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3505, pruned_loss=0.09977, over 5719642.37 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3502, pruned_loss=0.09433, over 5264495.80 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3505, pruned_loss=0.1005, over 5706587.28 frames. ], batch size: 136, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:59:35,830 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:59:37,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:59:41,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=644038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:00:01,488 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=644061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:00:08,085 INFO [train.py:968] (0/2) Epoch 15, batch 5150, giga_loss[loss=0.2647, simple_loss=0.3282, pruned_loss=0.1006, over 28701.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3481, pruned_loss=0.09897, over 5704542.08 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.35, pruned_loss=0.09417, over 5274215.94 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3483, pruned_loss=0.0998, over 5697697.25 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:00:14,930 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.119e+02 1.026e+03 1.356e+03 1.842e+03 6.161e+03, threshold=2.711e+03, percent-clipped=6.0
+2023-03-07 16:00:19,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9443, 1.3078, 1.0760, 0.2027], device='cuda:0'), covar=tensor([0.3529, 0.2554, 0.4420, 0.5983], device='cuda:0'), in_proj_covar=tensor([0.1620, 0.1529, 0.1517, 0.1320], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 16:00:47,634 INFO [train.py:968] (0/2) Epoch 15, batch 5200, giga_loss[loss=0.2645, simple_loss=0.3367, pruned_loss=0.09609, over 29084.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3445, pruned_loss=0.09728, over 5714960.77 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3495, pruned_loss=0.0939, over 5286231.25 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.345, pruned_loss=0.09822, over 5706479.06 frames. ], batch size: 155, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:01:17,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2399, 1.9049, 1.3670, 0.4100], device='cuda:0'), covar=tensor([0.4242, 0.2064, 0.3164, 0.5350], device='cuda:0'), in_proj_covar=tensor([0.1617, 0.1525, 0.1514, 0.1317], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 16:01:27,563 INFO [train.py:968] (0/2) Epoch 15, batch 5250, giga_loss[loss=0.263, simple_loss=0.3358, pruned_loss=0.09512, over 23623.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3449, pruned_loss=0.09674, over 5715151.25 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3497, pruned_loss=0.09388, over 5303173.11 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3449, pruned_loss=0.0976, over 5704496.69 frames. ], batch size: 705, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:01:36,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.619e+02 1.142e+03 1.587e+03 2.584e+03 1.133e+04, threshold=3.175e+03, percent-clipped=20.0
+2023-03-07 16:02:00,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8615, 1.0430, 1.0534, 0.8320], device='cuda:0'), covar=tensor([0.1845, 0.1979, 0.1173, 0.1671], device='cuda:0'), in_proj_covar=tensor([0.1794, 0.1722, 0.1675, 0.1785], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 16:02:07,400 INFO [train.py:968] (0/2) Epoch 15, batch 5300, giga_loss[loss=0.3, simple_loss=0.3856, pruned_loss=0.1072, over 28567.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3474, pruned_loss=0.09713, over 5722965.02 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3499, pruned_loss=0.09397, over 5326215.55 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3471, pruned_loss=0.09787, over 5708222.51 frames. ], batch size: 307, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:02:49,980 INFO [train.py:968] (0/2) Epoch 15, batch 5350, giga_loss[loss=0.2769, simple_loss=0.3511, pruned_loss=0.1013, over 28902.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3478, pruned_loss=0.09717, over 5721994.11 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3495, pruned_loss=0.094, over 5333031.19 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3479, pruned_loss=0.09777, over 5711032.16 frames. ], batch size: 213, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:02:58,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.400e+02 1.084e+03 1.309e+03 1.686e+03 6.725e+03, threshold=2.619e+03, percent-clipped=7.0
+2023-03-07 16:03:13,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=644297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:03:25,979 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=644312.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:03:31,226 INFO [train.py:968] (0/2) Epoch 15, batch 5400, giga_loss[loss=0.2771, simple_loss=0.3458, pruned_loss=0.1042, over 28972.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3474, pruned_loss=0.09851, over 5723632.91 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3497, pruned_loss=0.09405, over 5341571.18 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3473, pruned_loss=0.09902, over 5712398.24 frames. ], batch size: 213, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:03:43,155 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5234, 1.6745, 1.6759, 1.4570], device='cuda:0'), covar=tensor([0.2279, 0.2080, 0.1493, 0.1842], device='cuda:0'), in_proj_covar=tensor([0.1800, 0.1722, 0.1674, 0.1784], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 16:03:45,754 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=644334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:04:14,805 INFO [train.py:968] (0/2) Epoch 15, batch 5450, giga_loss[loss=0.2348, simple_loss=0.303, pruned_loss=0.0833, over 28963.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.346, pruned_loss=0.09922, over 5716858.54 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3499, pruned_loss=0.09441, over 5341760.40 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3456, pruned_loss=0.09943, over 5716976.76 frames. ], batch size: 106, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:04:23,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.489e+02 1.359e+03 1.666e+03 2.502e+03 1.123e+04, threshold=3.332e+03, percent-clipped=19.0
+2023-03-07 16:04:57,574 INFO [train.py:968] (0/2) Epoch 15, batch 5500, giga_loss[loss=0.2406, simple_loss=0.3138, pruned_loss=0.08371, over 28902.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3434, pruned_loss=0.09874, over 5724437.28 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3504, pruned_loss=0.0946, over 5349356.80 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3427, pruned_loss=0.09881, over 5722885.96 frames. ], batch size: 145, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:05:12,109 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0627, 1.2720, 3.3848, 3.0943], device='cuda:0'), covar=tensor([0.1669, 0.2577, 0.0485, 0.0769], device='cuda:0'), in_proj_covar=tensor([0.0690, 0.0605, 0.0884, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 16:05:16,404 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:05:20,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:05:42,435 INFO [train.py:968] (0/2) Epoch 15, batch 5550, giga_loss[loss=0.2762, simple_loss=0.3417, pruned_loss=0.1053, over 28959.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3423, pruned_loss=0.09878, over 5724525.97 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3505, pruned_loss=0.0946, over 5352271.02 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3416, pruned_loss=0.09884, over 5722445.36 frames. ], batch size: 213, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:05:42,647 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=644469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:05:44,755 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=644472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:05:48,344 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:05:50,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.402e+02 1.091e+03 1.298e+03 1.520e+03 2.877e+03, threshold=2.596e+03, percent-clipped=0.0
+2023-03-07 16:05:50,737 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:06:16,926 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=644509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:06:24,145 INFO [train.py:968] (0/2) Epoch 15, batch 5600, giga_loss[loss=0.2704, simple_loss=0.3424, pruned_loss=0.0992, over 28855.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3402, pruned_loss=0.09769, over 5718328.76 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3507, pruned_loss=0.09478, over 5365575.20 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3393, pruned_loss=0.09769, over 5713209.04 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:06:29,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4786, 1.9352, 1.4529, 1.7539], device='cuda:0'), covar=tensor([0.0742, 0.0278, 0.0323, 0.0813], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:0'), out_proj_covar=tensor([0.0087, 0.0063, 0.0056, 0.0095], device='cuda:0')
+2023-03-07 16:06:55,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4372, 2.8752, 1.4904, 1.4684], device='cuda:0'), covar=tensor([0.0830, 0.0349, 0.0869, 0.1194], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0523, 0.0355, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 16:07:04,643 INFO [train.py:968] (0/2) Epoch 15, batch 5650, libri_loss[loss=0.2698, simple_loss=0.3528, pruned_loss=0.09338, over 29543.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3372, pruned_loss=0.09569, over 5716766.39 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3504, pruned_loss=0.09458, over 5385757.66 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3363, pruned_loss=0.09592, over 5708667.62 frames. ], batch size: 80, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:07:12,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.569e+02 1.095e+03 1.364e+03 1.792e+03 4.493e+03, threshold=2.727e+03, percent-clipped=6.0
+2023-03-07 16:07:43,268 INFO [train.py:968] (0/2) Epoch 15, batch 5700, giga_loss[loss=0.2431, simple_loss=0.3156, pruned_loss=0.08533, over 28890.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3335, pruned_loss=0.09412, over 5709398.86 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3506, pruned_loss=0.09481, over 5384613.20 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3324, pruned_loss=0.09413, over 5709948.44 frames. ], batch size: 119, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:08:09,570 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 16:08:24,526 INFO [train.py:968] (0/2) Epoch 15, batch 5750, giga_loss[loss=0.2749, simple_loss=0.3465, pruned_loss=0.1016, over 28784.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3322, pruned_loss=0.09349, over 5707578.88 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3505, pruned_loss=0.09483, over 5392344.73 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3311, pruned_loss=0.09347, over 5705537.30 frames. ], batch size: 262, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:08:32,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.302e+02 1.164e+03 1.498e+03 2.119e+03 4.035e+03, threshold=2.997e+03, percent-clipped=13.0
+2023-03-07 16:08:36,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=644687.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:08:51,888 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-07 16:09:03,110 INFO [train.py:968] (0/2) Epoch 15, batch 5800, giga_loss[loss=0.2684, simple_loss=0.3447, pruned_loss=0.09604, over 28703.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3356, pruned_loss=0.09512, over 5712228.40 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3509, pruned_loss=0.09516, over 5402115.69 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3341, pruned_loss=0.09481, over 5707464.37 frames. ], batch size: 242, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:09:43,401 INFO [train.py:968] (0/2) Epoch 15, batch 5850, giga_loss[loss=0.3046, simple_loss=0.3705, pruned_loss=0.1193, over 28970.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3399, pruned_loss=0.09719, over 5695953.46 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3513, pruned_loss=0.09562, over 5401045.70 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3379, pruned_loss=0.09658, over 5702206.39 frames. ], batch size: 128, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:09:51,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.795e+02 1.210e+03 1.513e+03 2.089e+03 5.662e+03, threshold=3.027e+03, percent-clipped=10.0
+2023-03-07 16:10:22,953 INFO [train.py:968] (0/2) Epoch 15, batch 5900, giga_loss[loss=0.2617, simple_loss=0.3425, pruned_loss=0.09042, over 28688.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3434, pruned_loss=0.09804, over 5707307.71 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3516, pruned_loss=0.09578, over 5413172.03 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3412, pruned_loss=0.09746, over 5710526.55 frames. ], batch size: 242, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:10:33,855 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644830.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:10:35,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644833.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:10:47,163 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=644844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:11:02,466 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=644862.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:11:07,194 INFO [train.py:968] (0/2) Epoch 15, batch 5950, giga_loss[loss=0.3044, simple_loss=0.3695, pruned_loss=0.1197, over 27941.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3466, pruned_loss=0.09965, over 5705898.23 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3513, pruned_loss=0.09574, over 5428207.51 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3449, pruned_loss=0.09933, over 5703920.81 frames. ], batch size: 412, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:11:13,746 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 16:11:17,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.686e+02 1.191e+03 1.657e+03 2.543e+03 6.956e+03, threshold=3.314e+03, percent-clipped=15.0
+2023-03-07 16:11:47,267 INFO [train.py:968] (0/2) Epoch 15, batch 6000, giga_loss[loss=0.3303, simple_loss=0.3911, pruned_loss=0.1348, over 28746.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3478, pruned_loss=0.09985, over 5712504.03 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3507, pruned_loss=0.09531, over 5448368.37 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3467, pruned_loss=0.1001, over 5703952.41 frames. ], batch size: 284, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:11:47,271 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 16:11:55,558 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2147, simple_loss=0.3199, pruned_loss=0.05469, over 944034.00 frames. 
+2023-03-07 16:11:55,559 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 16:12:18,957 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=644945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:12:21,226 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=644947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:12:40,077 INFO [train.py:968] (0/2) Epoch 15, batch 6050, giga_loss[loss=0.2607, simple_loss=0.3422, pruned_loss=0.08957, over 28901.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3517, pruned_loss=0.1033, over 5707313.26 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3505, pruned_loss=0.09529, over 5459242.66 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.351, pruned_loss=0.1037, over 5696375.05 frames. ], batch size: 174, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:12:45,302 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-07 16:12:47,204 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7824, 4.5849, 4.3748, 1.9459], device='cuda:0'), covar=tensor([0.0525, 0.0690, 0.0721, 0.2062], device='cuda:0'), in_proj_covar=tensor([0.1114, 0.1037, 0.0896, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 16:12:52,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.161e+02 1.291e+03 1.718e+03 2.439e+03 8.877e+03, threshold=3.436e+03, percent-clipped=13.0
+2023-03-07 16:12:58,627 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:13:02,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:13:12,879 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=645003.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:13:27,792 INFO [train.py:968] (0/2) Epoch 15, batch 6100, giga_loss[loss=0.2762, simple_loss=0.3508, pruned_loss=0.1008, over 28849.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3565, pruned_loss=0.1078, over 5701617.04 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.35, pruned_loss=0.09508, over 5469933.89 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3566, pruned_loss=0.1086, over 5689634.62 frames. ], batch size: 145, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:13:28,051 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:13:34,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-07 16:13:53,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6862, 1.5755, 1.2427, 1.2337], device='cuda:0'), covar=tensor([0.0663, 0.0548, 0.0898, 0.1147], device='cuda:0'), in_proj_covar=tensor([0.0365, 0.0438, 0.0500, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 16:14:13,780 INFO [train.py:968] (0/2) Epoch 15, batch 6150, libri_loss[loss=0.2952, simple_loss=0.37, pruned_loss=0.1102, over 29102.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3625, pruned_loss=0.1115, over 5705640.66 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3498, pruned_loss=0.09507, over 5482889.24 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.363, pruned_loss=0.1127, over 5691382.18 frames. ], batch size: 101, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:14:23,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.283e+02 1.552e+03 2.092e+03 3.316e+03 6.771e+03, threshold=4.183e+03, percent-clipped=22.0
+2023-03-07 16:14:59,144 INFO [train.py:968] (0/2) Epoch 15, batch 6200, libri_loss[loss=0.2816, simple_loss=0.3641, pruned_loss=0.09953, over 29105.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3671, pruned_loss=0.115, over 5705268.76 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3501, pruned_loss=0.09526, over 5497341.47 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.368, pruned_loss=0.1167, over 5690082.76 frames. ], batch size: 101, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:15:26,350 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 16:15:35,230 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 16:15:41,534 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=645166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:15:43,553 INFO [train.py:968] (0/2) Epoch 15, batch 6250, giga_loss[loss=0.3111, simple_loss=0.3778, pruned_loss=0.1223, over 28972.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.374, pruned_loss=0.1211, over 5700275.60 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3504, pruned_loss=0.0954, over 5498121.97 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.375, pruned_loss=0.1228, over 5692500.14 frames. ], batch size: 213, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:15:57,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.478e+02 1.604e+03 1.963e+03 2.829e+03 6.737e+03, threshold=3.926e+03, percent-clipped=6.0
+2023-03-07 16:16:07,847 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-07 16:16:31,035 INFO [train.py:968] (0/2) Epoch 15, batch 6300, giga_loss[loss=0.4019, simple_loss=0.4184, pruned_loss=0.1927, over 23350.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3785, pruned_loss=0.1252, over 5682418.91 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3503, pruned_loss=0.09535, over 5500641.01 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3801, pruned_loss=0.1273, over 5679545.80 frames. ], batch size: 705, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:17:21,464 INFO [train.py:968] (0/2) Epoch 15, batch 6350, giga_loss[loss=0.3036, simple_loss=0.367, pruned_loss=0.1201, over 29090.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3793, pruned_loss=0.1261, over 5678030.96 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3502, pruned_loss=0.09519, over 5510535.07 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3816, pruned_loss=0.129, over 5672217.96 frames. ], batch size: 128, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:17:34,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.672e+03 2.279e+03 3.155e+03 1.005e+04, threshold=4.558e+03, percent-clipped=14.0
+2023-03-07 16:18:06,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3485, 5.1573, 4.9030, 2.4228], device='cuda:0'), covar=tensor([0.0408, 0.0538, 0.0621, 0.1815], device='cuda:0'), in_proj_covar=tensor([0.1126, 0.1047, 0.0906, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 16:18:15,860 INFO [train.py:968] (0/2) Epoch 15, batch 6400, giga_loss[loss=0.367, simple_loss=0.4153, pruned_loss=0.1593, over 27970.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3831, pruned_loss=0.1307, over 5668874.47 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3501, pruned_loss=0.09519, over 5514606.82 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3853, pruned_loss=0.1334, over 5661803.50 frames. ], batch size: 412, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:18:17,354 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=645320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:18:19,754 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=645322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:18:29,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3864, 2.0266, 1.5016, 0.4942], device='cuda:0'), covar=tensor([0.3770, 0.2125, 0.3059, 0.4940], device='cuda:0'), in_proj_covar=tensor([0.1642, 0.1557, 0.1534, 0.1342], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 16:18:42,859 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-07 16:19:08,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=645368.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:19:08,636 INFO [train.py:968] (0/2) Epoch 15, batch 6450, giga_loss[loss=0.3277, simple_loss=0.3794, pruned_loss=0.138, over 28383.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3851, pruned_loss=0.1335, over 5666185.14 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3494, pruned_loss=0.09475, over 5525195.14 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3886, pruned_loss=0.1371, over 5654975.33 frames. ], batch size: 65, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:19:18,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=645378.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:19:23,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.772e+03 2.359e+03 3.087e+03 8.721e+03, threshold=4.717e+03, percent-clipped=11.0
+2023-03-07 16:19:25,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3224, 3.1260, 3.0253, 1.3658], device='cuda:0'), covar=tensor([0.0933, 0.1076, 0.0952, 0.2344], device='cuda:0'), in_proj_covar=tensor([0.1127, 0.1052, 0.0910, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 16:19:50,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4348, 1.8347, 1.2709, 0.7878], device='cuda:0'), covar=tensor([0.4678, 0.2870, 0.2237, 0.4468], device='cuda:0'), in_proj_covar=tensor([0.1643, 0.1557, 0.1537, 0.1342], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 16:19:53,766 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-07 16:20:01,596 INFO [train.py:968] (0/2) Epoch 15, batch 6500, giga_loss[loss=0.4226, simple_loss=0.4507, pruned_loss=0.1973, over 28569.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3901, pruned_loss=0.1381, over 5655544.62 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3493, pruned_loss=0.09464, over 5532857.64 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3935, pruned_loss=0.1418, over 5642487.07 frames. ], batch size: 336, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:20:49,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=645463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:20:51,489 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=645465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:20:52,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=645466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:20:53,839 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=645468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:20:54,270 INFO [train.py:968] (0/2) Epoch 15, batch 6550, giga_loss[loss=0.3806, simple_loss=0.4205, pruned_loss=0.1703, over 28344.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3911, pruned_loss=0.1396, over 5652928.70 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3495, pruned_loss=0.09474, over 5537055.89 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3942, pruned_loss=0.143, over 5640559.47 frames. ], batch size: 368, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:21:08,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.232e+03 1.754e+03 2.347e+03 3.531e+03 7.833e+03, threshold=4.693e+03, percent-clipped=14.0
+2023-03-07 16:21:20,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:21:21,962 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:21:43,795 INFO [train.py:968] (0/2) Epoch 15, batch 6600, giga_loss[loss=0.3301, simple_loss=0.3875, pruned_loss=0.1364, over 28958.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3899, pruned_loss=0.1393, over 5636434.77 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3497, pruned_loss=0.09478, over 5530558.00 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3934, pruned_loss=0.1433, over 5635234.84 frames. ], batch size: 164, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:21:47,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=645521.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:21:49,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=645524.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:21:55,807 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-07 16:22:07,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=645541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:22:20,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645553.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:22:36,187 INFO [train.py:968] (0/2) Epoch 15, batch 6650, giga_loss[loss=0.3263, simple_loss=0.4024, pruned_loss=0.1251, over 28984.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3897, pruned_loss=0.1391, over 5627480.44 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3501, pruned_loss=0.09515, over 5528285.59 frames. ], giga_tot_loss[loss=0.3391, simple_loss=0.3928, pruned_loss=0.1428, over 5630229.65 frames. ], batch size: 164, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:22:50,816 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-07 16:22:51,016 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.359e+02 1.554e+03 2.028e+03 3.112e+03 8.870e+03, threshold=4.056e+03, percent-clipped=6.0
+2023-03-07 16:23:27,050 INFO [train.py:968] (0/2) Epoch 15, batch 6700, giga_loss[loss=0.3402, simple_loss=0.4014, pruned_loss=0.1396, over 28980.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3897, pruned_loss=0.1379, over 5638757.28 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3501, pruned_loss=0.09512, over 5532333.68 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3926, pruned_loss=0.1412, over 5638113.59 frames. ], batch size: 227, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:23:53,510 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=645645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:24:18,713 INFO [train.py:968] (0/2) Epoch 15, batch 6750, giga_loss[loss=0.2874, simple_loss=0.3649, pruned_loss=0.1049, over 28881.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3914, pruned_loss=0.139, over 5633230.30 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3498, pruned_loss=0.09494, over 5536740.44 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3945, pruned_loss=0.1424, over 5630107.86 frames. ], batch size: 145, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:24:33,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.301e+02 1.433e+03 1.913e+03 2.707e+03 6.577e+03, threshold=3.826e+03, percent-clipped=8.0
+2023-03-07 16:24:34,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=645684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:24:37,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=645687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:24:46,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8857, 1.0345, 1.0003, 0.8152], device='cuda:0'), covar=tensor([0.2145, 0.2213, 0.1256, 0.1832], device='cuda:0'), in_proj_covar=tensor([0.1824, 0.1748, 0.1694, 0.1797], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 16:25:05,288 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:25:07,123 INFO [train.py:968] (0/2) Epoch 15, batch 6800, libri_loss[loss=0.2334, simple_loss=0.3111, pruned_loss=0.07783, over 28476.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3887, pruned_loss=0.1367, over 5630129.46 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3496, pruned_loss=0.0947, over 5545021.12 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3926, pruned_loss=0.1409, over 5622491.44 frames. ], batch size: 63, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:25:35,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=645743.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:25:58,915 INFO [train.py:968] (0/2) Epoch 15, batch 6850, giga_loss[loss=0.3329, simple_loss=0.3984, pruned_loss=0.1337, over 28633.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3863, pruned_loss=0.1335, over 5637139.46 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3496, pruned_loss=0.09476, over 5551075.21 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3905, pruned_loss=0.1379, over 5627592.76 frames. ], batch size: 262, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:26:14,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.599e+03 2.087e+03 2.638e+03 8.956e+03, threshold=4.175e+03, percent-clipped=10.0
+2023-03-07 16:26:48,541 INFO [train.py:968] (0/2) Epoch 15, batch 6900, giga_loss[loss=0.2719, simple_loss=0.3527, pruned_loss=0.09557, over 28913.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3836, pruned_loss=0.1302, over 5640091.91 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3498, pruned_loss=0.09505, over 5549692.68 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3875, pruned_loss=0.1343, over 5636096.14 frames. ], batch size: 227, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:27:38,610 INFO [train.py:968] (0/2) Epoch 15, batch 6950, giga_loss[loss=0.2884, simple_loss=0.3579, pruned_loss=0.1094, over 28835.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3801, pruned_loss=0.1276, over 5646198.63 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3496, pruned_loss=0.09507, over 5553215.53 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3839, pruned_loss=0.1314, over 5641459.86 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:27:42,734 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=645873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:27:53,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.989e+02 1.694e+03 2.161e+03 3.554e+03 1.131e+04, threshold=4.321e+03, percent-clipped=19.0
+2023-03-07 16:27:54,906 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=645886.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:27:55,718 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-07 16:27:57,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=645889.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:28:20,876 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645918.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:28:21,279 INFO [train.py:968] (0/2) Epoch 15, batch 7000, giga_loss[loss=0.3733, simple_loss=0.4127, pruned_loss=0.1669, over 28025.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3766, pruned_loss=0.1248, over 5655284.38 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3494, pruned_loss=0.09507, over 5565311.70 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3812, pruned_loss=0.1292, over 5644200.35 frames. ], batch size: 412, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:28:29,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7596, 5.2581, 1.8218, 2.1275], device='cuda:0'), covar=tensor([0.0927, 0.0194, 0.0882, 0.1190], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0528, 0.0356, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 16:28:41,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5598, 2.0255, 1.7681, 1.4217], device='cuda:0'), covar=tensor([0.2973, 0.2043, 0.2350, 0.2450], device='cuda:0'), in_proj_covar=tensor([0.1828, 0.1748, 0.1691, 0.1799], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 16:29:11,859 INFO [train.py:968] (0/2) Epoch 15, batch 7050, giga_loss[loss=0.3427, simple_loss=0.4052, pruned_loss=0.1401, over 28966.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3764, pruned_loss=0.1249, over 5645309.35 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3494, pruned_loss=0.09508, over 5561121.66 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3803, pruned_loss=0.1288, over 5640667.47 frames. ], batch size: 136, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:29:26,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 16:29:26,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.329e+02 1.521e+03 2.085e+03 2.570e+03 6.883e+03, threshold=4.169e+03, percent-clipped=5.0
+2023-03-07 16:29:33,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1717, 1.5685, 1.1758, 0.3252], device='cuda:0'), covar=tensor([0.2415, 0.1479, 0.2047, 0.4336], device='cuda:0'), in_proj_covar=tensor([0.1636, 0.1552, 0.1534, 0.1340], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 16:29:42,475 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-646000.pt
+2023-03-07 16:30:04,545 INFO [train.py:968] (0/2) Epoch 15, batch 7100, giga_loss[loss=0.263, simple_loss=0.3427, pruned_loss=0.09172, over 28997.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3765, pruned_loss=0.1251, over 5644937.26 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.35, pruned_loss=0.0954, over 5566693.71 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3799, pruned_loss=0.1287, over 5638414.28 frames. ], batch size: 164, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:30:06,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:31:00,065 INFO [train.py:968] (0/2) Epoch 15, batch 7150, giga_loss[loss=0.2669, simple_loss=0.3413, pruned_loss=0.09626, over 28834.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3745, pruned_loss=0.1231, over 5652866.53 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3494, pruned_loss=0.0951, over 5573345.95 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3781, pruned_loss=0.1268, over 5643262.35 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:31:15,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.801e+02 1.409e+03 1.790e+03 2.417e+03 5.488e+03, threshold=3.580e+03, percent-clipped=1.0
+2023-03-07 16:31:33,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1492, 1.2572, 1.0677, 0.8850], device='cuda:0'), covar=tensor([0.0945, 0.0545, 0.1101, 0.1104], device='cuda:0'), in_proj_covar=tensor([0.0368, 0.0441, 0.0503, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 16:31:56,991 INFO [train.py:968] (0/2) Epoch 15, batch 7200, giga_loss[loss=0.2826, simple_loss=0.34, pruned_loss=0.1126, over 23793.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3747, pruned_loss=0.1211, over 5659407.33 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3489, pruned_loss=0.09484, over 5581128.44 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3787, pruned_loss=0.1249, over 5646441.78 frames. ], batch size: 705, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:32:30,505 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-07 16:32:38,478 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=646163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:32:40,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=646166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:32:43,441 INFO [train.py:968] (0/2) Epoch 15, batch 7250, libri_loss[loss=0.2665, simple_loss=0.3568, pruned_loss=0.08815, over 29660.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3758, pruned_loss=0.12, over 5677898.29 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3485, pruned_loss=0.09454, over 5594101.16 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3804, pruned_loss=0.1243, over 5658396.34 frames. ], batch size: 91, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:33:00,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.182e+02 1.584e+03 2.092e+03 3.054e+03 6.770e+03, threshold=4.184e+03, percent-clipped=15.0
+2023-03-07 16:33:12,978 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=646195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:33:41,760 INFO [train.py:968] (0/2) Epoch 15, batch 7300, giga_loss[loss=0.3426, simple_loss=0.4009, pruned_loss=0.1421, over 28265.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3779, pruned_loss=0.1222, over 5666658.49 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3485, pruned_loss=0.09454, over 5594101.16 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3815, pruned_loss=0.1256, over 5651479.88 frames. ], batch size: 368, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:33:46,262 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:34:11,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:34:27,594 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:34:28,844 INFO [train.py:968] (0/2) Epoch 15, batch 7350, giga_loss[loss=0.2791, simple_loss=0.3551, pruned_loss=0.1016, over 28959.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3767, pruned_loss=0.122, over 5673544.05 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3483, pruned_loss=0.09468, over 5601587.74 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3806, pruned_loss=0.1254, over 5656375.82 frames. ], batch size: 136, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:34:32,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-07 16:34:42,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.619e+03 2.181e+03 3.048e+03 6.533e+03, threshold=4.362e+03, percent-clipped=8.0
+2023-03-07 16:35:18,540 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 16:35:20,549 INFO [train.py:968] (0/2) Epoch 15, batch 7400, giga_loss[loss=0.3109, simple_loss=0.3742, pruned_loss=0.1238, over 28649.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3748, pruned_loss=0.122, over 5679537.88 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3481, pruned_loss=0.09454, over 5604519.01 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3783, pruned_loss=0.125, over 5664235.98 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:35:42,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3500, 1.4246, 1.3354, 1.5187], device='cuda:0'), covar=tensor([0.0673, 0.0436, 0.0318, 0.0690], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0057, 0.0096], device='cuda:0')
+2023-03-07 16:36:05,950 INFO [train.py:968] (0/2) Epoch 15, batch 7450, giga_loss[loss=0.3417, simple_loss=0.395, pruned_loss=0.1442, over 27589.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3747, pruned_loss=0.1233, over 5674225.28 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3479, pruned_loss=0.09436, over 5609841.02 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3782, pruned_loss=0.1265, over 5658851.02 frames. ], batch size: 472, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:36:20,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.579e+03 1.942e+03 2.503e+03 5.056e+03, threshold=3.884e+03, percent-clipped=3.0
+2023-03-07 16:36:28,846 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=646391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:36:30,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=646394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:37:00,324 INFO [train.py:968] (0/2) Epoch 15, batch 7500, giga_loss[loss=0.2946, simple_loss=0.3734, pruned_loss=0.1079, over 29011.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3746, pruned_loss=0.1227, over 5671657.51 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.348, pruned_loss=0.09436, over 5614205.28 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3777, pruned_loss=0.1256, over 5656597.40 frames. ], batch size: 155, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:37:03,192 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=646423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:37:36,647 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 16:37:50,092 INFO [train.py:968] (0/2) Epoch 15, batch 7550, giga_loss[loss=0.288, simple_loss=0.3615, pruned_loss=0.1073, over 28953.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3743, pruned_loss=0.1216, over 5667798.82 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3481, pruned_loss=0.09437, over 5616095.67 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3772, pruned_loss=0.1244, over 5655175.92 frames. ], batch size: 106, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:38:00,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.721e+02 1.557e+03 2.218e+03 2.949e+03 7.605e+03, threshold=4.435e+03, percent-clipped=8.0
+2023-03-07 16:38:01,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9513, 1.3226, 1.0992, 0.1649], device='cuda:0'), covar=tensor([0.3049, 0.2476, 0.3624, 0.5279], device='cuda:0'), in_proj_covar=tensor([0.1629, 0.1545, 0.1530, 0.1334], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 16:38:02,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1977, 1.5792, 1.1772, 0.3869], device='cuda:0'), covar=tensor([0.2444, 0.1561, 0.2187, 0.4422], device='cuda:0'), in_proj_covar=tensor([0.1629, 0.1544, 0.1530, 0.1334], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 16:38:34,792 INFO [train.py:968] (0/2) Epoch 15, batch 7600, giga_loss[loss=0.3385, simple_loss=0.4034, pruned_loss=0.1368, over 28746.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3755, pruned_loss=0.1219, over 5678310.05 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3488, pruned_loss=0.09483, over 5622808.65 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3778, pruned_loss=0.1244, over 5663540.81 frames. ], batch size: 284, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:39:15,676 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:39:19,522 INFO [train.py:968] (0/2) Epoch 15, batch 7650, giga_loss[loss=0.2659, simple_loss=0.3448, pruned_loss=0.09344, over 29052.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3735, pruned_loss=0.1202, over 5697908.86 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3489, pruned_loss=0.09486, over 5632825.71 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3761, pruned_loss=0.123, over 5679041.41 frames. ], batch size: 128, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:39:32,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.870e+02 1.527e+03 1.940e+03 3.175e+03 5.985e+03, threshold=3.881e+03, percent-clipped=10.0
+2023-03-07 16:39:34,863 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-07 16:39:39,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:39:46,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:40:05,174 INFO [train.py:968] (0/2) Epoch 15, batch 7700, giga_loss[loss=0.2643, simple_loss=0.331, pruned_loss=0.09875, over 28645.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3713, pruned_loss=0.1194, over 5685846.23 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.349, pruned_loss=0.09497, over 5635697.53 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.374, pruned_loss=0.1223, over 5669914.77 frames. ], batch size: 85, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:40:28,551 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:40:56,361 INFO [train.py:968] (0/2) Epoch 15, batch 7750, giga_loss[loss=0.3005, simple_loss=0.3674, pruned_loss=0.1168, over 28950.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3709, pruned_loss=0.1203, over 5670097.54 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3495, pruned_loss=0.09522, over 5632813.48 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3731, pruned_loss=0.1229, over 5660413.92 frames. ], batch size: 213, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:41:15,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.547e+03 2.120e+03 3.106e+03 7.304e+03, threshold=4.240e+03, percent-clipped=10.0
+2023-03-07 16:41:20,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-07 16:41:31,400 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3644, 3.1915, 3.0414, 1.4579], device='cuda:0'), covar=tensor([0.0807, 0.0955, 0.0837, 0.2114], device='cuda:0'), in_proj_covar=tensor([0.1128, 0.1049, 0.0906, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 16:41:35,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4733, 1.7272, 1.6998, 1.4113], device='cuda:0'), covar=tensor([0.2097, 0.1755, 0.1166, 0.1611], device='cuda:0'), in_proj_covar=tensor([0.1821, 0.1747, 0.1684, 0.1801], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 16:41:46,527 INFO [train.py:968] (0/2) Epoch 15, batch 7800, giga_loss[loss=0.3259, simple_loss=0.38, pruned_loss=0.136, over 28742.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3694, pruned_loss=0.1196, over 5670489.99 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3495, pruned_loss=0.09531, over 5641022.42 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3717, pruned_loss=0.1224, over 5655987.70 frames. ], batch size: 119, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:42:11,683 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=646742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:42:15,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=646745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:42:37,251 INFO [train.py:968] (0/2) Epoch 15, batch 7850, giga_loss[loss=0.3911, simple_loss=0.4185, pruned_loss=0.1818, over 26671.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.37, pruned_loss=0.1213, over 5656673.66 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3498, pruned_loss=0.09546, over 5639888.25 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3718, pruned_loss=0.1237, over 5646611.67 frames. ], batch size: 555, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:42:41,262 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=646774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:42:51,325 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=646785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:42:51,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.418e+02 1.630e+03 2.042e+03 3.093e+03 1.012e+04, threshold=4.085e+03, percent-clipped=15.0
+2023-03-07 16:42:54,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=646788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:43:19,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=646817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:43:22,598 INFO [train.py:968] (0/2) Epoch 15, batch 7900, giga_loss[loss=0.3191, simple_loss=0.3768, pruned_loss=0.1307, over 27648.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3688, pruned_loss=0.1206, over 5663420.61 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3497, pruned_loss=0.0955, over 5645135.37 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3708, pruned_loss=0.123, over 5651149.69 frames. ], batch size: 472, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:44:02,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2125, 1.2525, 1.0587, 0.8796], device='cuda:0'), covar=tensor([0.0687, 0.0400, 0.0878, 0.1067], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0444, 0.0504, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 16:44:13,310 INFO [train.py:968] (0/2) Epoch 15, batch 7950, giga_loss[loss=0.3261, simple_loss=0.3889, pruned_loss=0.1317, over 28520.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3702, pruned_loss=0.1212, over 5665563.08 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3498, pruned_loss=0.0955, over 5647964.50 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.372, pruned_loss=0.1235, over 5653718.94 frames. ], batch size: 336, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:44:18,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:44:32,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.708e+03 2.056e+03 3.132e+03 1.498e+04, threshold=4.111e+03, percent-clipped=12.0
+2023-03-07 16:44:41,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2122, 1.3599, 3.9988, 3.1367], device='cuda:0'), covar=tensor([0.1735, 0.2586, 0.0466, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0701, 0.0614, 0.0903, 0.0826], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 16:45:00,980 INFO [train.py:968] (0/2) Epoch 15, batch 8000, giga_loss[loss=0.2612, simple_loss=0.3424, pruned_loss=0.09, over 28328.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3699, pruned_loss=0.1198, over 5671408.99 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3491, pruned_loss=0.09512, over 5652087.22 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3722, pruned_loss=0.1224, over 5658586.08 frames. ], batch size: 65, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:45:22,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:45:47,319 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:45:47,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:45:49,737 INFO [train.py:968] (0/2) Epoch 15, batch 8050, giga_loss[loss=0.3025, simple_loss=0.3688, pruned_loss=0.1181, over 28855.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3708, pruned_loss=0.1195, over 5681853.48 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3491, pruned_loss=0.09502, over 5654536.34 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3729, pruned_loss=0.1219, over 5669755.16 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:46:09,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.790e+02 1.411e+03 1.809e+03 2.235e+03 5.379e+03, threshold=3.619e+03, percent-clipped=1.0
+2023-03-07 16:46:19,596 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9629, 1.2084, 1.3260, 1.0341], device='cuda:0'), covar=tensor([0.1554, 0.1333, 0.1983, 0.1616], device='cuda:0'), in_proj_covar=tensor([0.0452, 0.0736, 0.0693, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 16:46:36,512 INFO [train.py:968] (0/2) Epoch 15, batch 8100, giga_loss[loss=0.2849, simple_loss=0.3637, pruned_loss=0.103, over 28888.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.371, pruned_loss=0.1192, over 5685462.71 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3495, pruned_loss=0.09521, over 5659430.58 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3731, pruned_loss=0.1219, over 5672098.24 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:47:30,562 INFO [train.py:968] (0/2) Epoch 15, batch 8150, giga_loss[loss=0.3678, simple_loss=0.4111, pruned_loss=0.1623, over 28808.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3727, pruned_loss=0.1216, over 5673397.95 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3496, pruned_loss=0.09535, over 5660774.43 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3745, pruned_loss=0.1236, over 5661915.18 frames. ], batch size: 112, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:47:44,661 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=647083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:47:46,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:47:48,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.761e+03 2.309e+03 3.211e+03 1.367e+04, threshold=4.619e+03, percent-clipped=16.0
+2023-03-07 16:48:11,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5533, 1.1188, 4.5289, 3.4267], device='cuda:0'), covar=tensor([0.1572, 0.2826, 0.0414, 0.0979], device='cuda:0'), in_proj_covar=tensor([0.0704, 0.0617, 0.0904, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 16:48:14,194 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=647110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:48:15,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8797, 2.6815, 1.7368, 1.0901], device='cuda:0'), covar=tensor([0.6327, 0.3351, 0.3316, 0.5414], device='cuda:0'), in_proj_covar=tensor([0.1639, 0.1561, 0.1535, 0.1338], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 16:48:16,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:48:19,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:48:25,445 INFO [train.py:968] (0/2) Epoch 15, batch 8200, giga_loss[loss=0.4438, simple_loss=0.4522, pruned_loss=0.2177, over 27537.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3747, pruned_loss=0.1247, over 5663960.50 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3495, pruned_loss=0.09528, over 5663876.95 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3766, pruned_loss=0.127, over 5652461.72 frames. ], batch size: 472, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:48:43,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6446, 1.8499, 1.8834, 1.4145], device='cuda:0'), covar=tensor([0.1731, 0.2282, 0.1391, 0.1604], device='cuda:0'), in_proj_covar=tensor([0.0855, 0.0694, 0.0899, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 16:48:47,204 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:48:57,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7609, 1.7473, 1.3243, 1.3882], device='cuda:0'), covar=tensor([0.0777, 0.0625, 0.0944, 0.1229], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0443, 0.0504, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 16:49:14,371 INFO [train.py:968] (0/2) Epoch 15, batch 8250, libri_loss[loss=0.2633, simple_loss=0.3526, pruned_loss=0.08699, over 29532.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3755, pruned_loss=0.1264, over 5664348.48 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3498, pruned_loss=0.09544, over 5665974.39 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3774, pruned_loss=0.1287, over 5653501.69 frames. ], batch size: 83, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:49:33,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.183e+02 1.834e+03 2.288e+03 3.041e+03 9.576e+03, threshold=4.577e+03, percent-clipped=11.0
+2023-03-07 16:49:59,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4915, 2.0072, 1.3849, 2.1295], device='cuda:0'), covar=tensor([0.2782, 0.2602, 0.3027, 0.2380], device='cuda:0'), in_proj_covar=tensor([0.1375, 0.1011, 0.1217, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 16:50:03,376 INFO [train.py:968] (0/2) Epoch 15, batch 8300, giga_loss[loss=0.3157, simple_loss=0.3753, pruned_loss=0.1281, over 28784.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3763, pruned_loss=0.1276, over 5663631.23 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3501, pruned_loss=0.09561, over 5667712.35 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3781, pruned_loss=0.13, over 5653474.54 frames. ], batch size: 284, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:50:34,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=647248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:50:50,069 INFO [train.py:968] (0/2) Epoch 15, batch 8350, giga_loss[loss=0.3134, simple_loss=0.3822, pruned_loss=0.1223, over 28950.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3742, pruned_loss=0.1258, over 5671541.68 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3498, pruned_loss=0.0956, over 5675095.74 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3768, pruned_loss=0.1288, over 5656229.78 frames. ], batch size: 145, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:51:08,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.683e+02 1.527e+03 2.070e+03 3.106e+03 9.181e+03, threshold=4.140e+03, percent-clipped=7.0
+2023-03-07 16:51:17,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6430, 1.6553, 1.2232, 1.2689], device='cuda:0'), covar=tensor([0.0765, 0.0575, 0.0957, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0443, 0.0504, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 16:51:31,434 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:51:34,802 INFO [train.py:968] (0/2) Epoch 15, batch 8400, giga_loss[loss=0.2964, simple_loss=0.3781, pruned_loss=0.1074, over 28893.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3728, pruned_loss=0.1239, over 5672705.31 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3501, pruned_loss=0.09577, over 5669134.61 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3751, pruned_loss=0.1268, over 5665811.68 frames. ], batch size: 145, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:51:53,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=647342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:52:17,338 INFO [train.py:968] (0/2) Epoch 15, batch 8450, giga_loss[loss=0.2786, simple_loss=0.3495, pruned_loss=0.1039, over 27937.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3711, pruned_loss=0.1206, over 5684648.10 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3504, pruned_loss=0.096, over 5677221.62 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3734, pruned_loss=0.1235, over 5672281.77 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:52:36,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.558e+03 1.973e+03 2.976e+03 1.022e+04, threshold=3.947e+03, percent-clipped=6.0
+2023-03-07 16:52:38,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=647391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:52:39,022 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:52:40,968 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:53:02,243 INFO [train.py:968] (0/2) Epoch 15, batch 8500, giga_loss[loss=0.3299, simple_loss=0.3845, pruned_loss=0.1376, over 28678.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3696, pruned_loss=0.1198, over 5677873.06 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3504, pruned_loss=0.09607, over 5679125.40 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3718, pruned_loss=0.1226, over 5666244.34 frames. ], batch size: 262, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:53:04,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3035, 1.5153, 1.4591, 1.2977], device='cuda:0'), covar=tensor([0.1591, 0.1643, 0.2237, 0.1868], device='cuda:0'), in_proj_covar=tensor([0.0451, 0.0733, 0.0689, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 16:53:06,092 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:53:49,643 INFO [train.py:968] (0/2) Epoch 15, batch 8550, giga_loss[loss=0.2792, simple_loss=0.3448, pruned_loss=0.1068, over 28877.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3678, pruned_loss=0.1192, over 5670034.49 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3507, pruned_loss=0.09627, over 5671652.11 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3696, pruned_loss=0.1216, over 5666780.41 frames. ], batch size: 112, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:53:56,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0291, 1.9288, 1.7725, 1.6925], device='cuda:0'), covar=tensor([0.1665, 0.2559, 0.2253, 0.2374], device='cuda:0'), in_proj_covar=tensor([0.0451, 0.0733, 0.0690, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 16:54:06,004 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=647485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:54:08,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:54:09,231 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.684e+02 1.660e+03 2.364e+03 3.458e+03 1.878e+04, threshold=4.729e+03, percent-clipped=18.0
+2023-03-07 16:54:36,781 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:54:38,162 INFO [train.py:968] (0/2) Epoch 15, batch 8600, giga_loss[loss=0.3053, simple_loss=0.373, pruned_loss=0.1188, over 28914.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3679, pruned_loss=0.1202, over 5670146.87 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3512, pruned_loss=0.09665, over 5674464.28 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3692, pruned_loss=0.1222, over 5664980.18 frames. ], batch size: 145, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:54:41,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3717, 3.2088, 3.0078, 1.8561], device='cuda:0'), covar=tensor([0.0771, 0.0914, 0.0860, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.1132, 0.1053, 0.0910, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 16:55:31,066 INFO [train.py:968] (0/2) Epoch 15, batch 8650, giga_loss[loss=0.3488, simple_loss=0.4049, pruned_loss=0.1463, over 28557.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3707, pruned_loss=0.1221, over 5672738.85 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.351, pruned_loss=0.09657, over 5677177.87 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3722, pruned_loss=0.1243, over 5666212.46 frames. ], batch size: 336, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:55:50,305 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.540e+03 2.039e+03 2.629e+03 7.362e+03, threshold=4.077e+03, percent-clipped=2.0
+2023-03-07 16:56:19,153 INFO [train.py:968] (0/2) Epoch 15, batch 8700, giga_loss[loss=0.4314, simple_loss=0.4615, pruned_loss=0.2006, over 27531.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.374, pruned_loss=0.1219, over 5675501.69 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3511, pruned_loss=0.09652, over 5679060.89 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3757, pruned_loss=0.1242, over 5668796.31 frames. ], batch size: 472, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:56:37,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2790, 3.0929, 2.9025, 1.4142], device='cuda:0'), covar=tensor([0.0984, 0.1153, 0.1087, 0.2406], device='cuda:0'), in_proj_covar=tensor([0.1137, 0.1061, 0.0915, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 16:57:04,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4541, 1.3899, 1.6574, 1.1848], device='cuda:0'), covar=tensor([0.1688, 0.2841, 0.1362, 0.1562], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0693, 0.0898, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 16:57:09,517 INFO [train.py:968] (0/2) Epoch 15, batch 8750, giga_loss[loss=0.3484, simple_loss=0.4071, pruned_loss=0.1449, over 28942.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3772, pruned_loss=0.1224, over 5674725.90 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.351, pruned_loss=0.09645, over 5681565.16 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.379, pruned_loss=0.1247, over 5666948.77 frames. ], batch size: 227, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:57:29,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.914e+02 1.497e+03 2.098e+03 3.057e+03 6.967e+03, threshold=4.195e+03, percent-clipped=13.0
+2023-03-07 16:57:30,165 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=647690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:57:58,899 INFO [train.py:968] (0/2) Epoch 15, batch 8800, giga_loss[loss=0.2892, simple_loss=0.3683, pruned_loss=0.1051, over 28938.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3783, pruned_loss=0.1231, over 5678131.91 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3507, pruned_loss=0.09619, over 5684198.32 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3805, pruned_loss=0.1256, over 5669493.38 frames. ], batch size: 164, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:58:10,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5057, 1.7829, 1.4111, 1.7047], device='cuda:0'), covar=tensor([0.2392, 0.2426, 0.2752, 0.2157], device='cuda:0'), in_proj_covar=tensor([0.1379, 0.1013, 0.1222, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 16:58:44,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3784, 1.5051, 1.3972, 1.2698], device='cuda:0'), covar=tensor([0.2078, 0.1960, 0.1374, 0.1851], device='cuda:0'), in_proj_covar=tensor([0.1823, 0.1751, 0.1689, 0.1802], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 16:58:45,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=647767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:58:46,733 INFO [train.py:968] (0/2) Epoch 15, batch 8850, giga_loss[loss=0.2588, simple_loss=0.344, pruned_loss=0.08685, over 28779.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3788, pruned_loss=0.1239, over 5686589.01 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3507, pruned_loss=0.09627, over 5687490.02 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.381, pruned_loss=0.1263, over 5676658.06 frames. ], batch size: 60, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:59:02,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.708e+02 1.613e+03 2.125e+03 3.284e+03 1.193e+04, threshold=4.250e+03, percent-clipped=15.0
+2023-03-07 16:59:02,841 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:59:08,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0289, 2.1415, 2.2398, 1.8166], device='cuda:0'), covar=tensor([0.1653, 0.2043, 0.1291, 0.1518], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0693, 0.0899, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 16:59:20,114 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:59:32,715 INFO [train.py:968] (0/2) Epoch 15, batch 8900, giga_loss[loss=0.3452, simple_loss=0.3944, pruned_loss=0.1479, over 27919.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.379, pruned_loss=0.1253, over 5687677.27 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3502, pruned_loss=0.09604, over 5690971.66 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3815, pruned_loss=0.1277, over 5676966.39 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:59:36,935 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 16:59:47,591 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=647833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:59:49,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:00:02,289 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:00:14,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3736, 1.8686, 1.5576, 1.5573], device='cuda:0'), covar=tensor([0.0767, 0.0296, 0.0299, 0.0840], device='cuda:0'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 17:00:17,877 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:00:21,314 INFO [train.py:968] (0/2) Epoch 15, batch 8950, giga_loss[loss=0.3352, simple_loss=0.3808, pruned_loss=0.1449, over 28471.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3776, pruned_loss=0.1249, over 5691199.77 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3503, pruned_loss=0.09594, over 5691401.66 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3803, pruned_loss=0.1277, over 5682135.00 frames. ], batch size: 85, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:00:42,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.591e+03 2.286e+03 3.365e+03 1.034e+04, threshold=4.573e+03, percent-clipped=12.0
+2023-03-07 17:00:51,229 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:00:59,684 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=647910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:01:01,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:01:04,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4621, 1.4872, 1.1821, 1.0519], device='cuda:0'), covar=tensor([0.0734, 0.0483, 0.0900, 0.1036], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0441, 0.0500, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 17:01:07,632 INFO [train.py:968] (0/2) Epoch 15, batch 9000, giga_loss[loss=0.2538, simple_loss=0.3252, pruned_loss=0.09127, over 28528.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3743, pruned_loss=0.1225, over 5681577.10 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3505, pruned_loss=0.09588, over 5686882.25 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3774, pruned_loss=0.126, over 5677414.52 frames. ], batch size: 85, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:01:07,638 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 17:01:15,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7381, 3.4592, 3.3422, 1.7384], device='cuda:0'), covar=tensor([0.0851, 0.1077, 0.0984, 0.2483], device='cuda:0'), in_proj_covar=tensor([0.1140, 0.1060, 0.0914, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 17:01:16,528 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2116, simple_loss=0.3182, pruned_loss=0.05252, over 944034.00 frames. 
+2023-03-07 17:01:16,528 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 17:01:19,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1995, 1.4920, 1.5230, 1.0903], device='cuda:0'), covar=tensor([0.1488, 0.2333, 0.1227, 0.1507], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0692, 0.0899, 0.0799], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 17:01:30,574 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-07 17:01:38,485 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:01:50,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6037, 2.2461, 1.6426, 0.7425], device='cuda:0'), covar=tensor([0.4018, 0.2506, 0.3514, 0.5056], device='cuda:0'), in_proj_covar=tensor([0.1646, 0.1565, 0.1539, 0.1345], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 17:02:02,275 INFO [train.py:968] (0/2) Epoch 15, batch 9050, giga_loss[loss=0.3233, simple_loss=0.3741, pruned_loss=0.1362, over 28704.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3738, pruned_loss=0.123, over 5677607.00 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3504, pruned_loss=0.09567, over 5689935.74 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3773, pruned_loss=0.127, over 5671173.11 frames. ], batch size: 99, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:02:25,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.602e+03 2.096e+03 3.033e+03 6.359e+03, threshold=4.191e+03, percent-clipped=9.0
+2023-03-07 17:02:35,116 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-648000.pt
+2023-03-07 17:02:55,605 INFO [train.py:968] (0/2) Epoch 15, batch 9100, giga_loss[loss=0.3171, simple_loss=0.3828, pruned_loss=0.1256, over 28988.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3726, pruned_loss=0.1225, over 5682078.18 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3499, pruned_loss=0.09538, over 5691555.92 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.376, pruned_loss=0.1263, over 5675311.07 frames. ], batch size: 155, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:03:44,114 INFO [train.py:968] (0/2) Epoch 15, batch 9150, giga_loss[loss=0.3213, simple_loss=0.3611, pruned_loss=0.1407, over 23684.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3729, pruned_loss=0.1234, over 5668402.21 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.35, pruned_loss=0.09541, over 5687158.87 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3763, pruned_loss=0.1273, over 5667190.95 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:03:50,180 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7779, 3.6112, 3.4220, 1.6511], device='cuda:0'), covar=tensor([0.0726, 0.0810, 0.0791, 0.2197], device='cuda:0'), in_proj_covar=tensor([0.1144, 0.1062, 0.0918, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 17:04:02,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.219e+02 1.665e+03 1.963e+03 2.434e+03 4.534e+03, threshold=3.926e+03, percent-clipped=1.0
+2023-03-07 17:04:28,662 INFO [train.py:968] (0/2) Epoch 15, batch 9200, giga_loss[loss=0.3105, simple_loss=0.3637, pruned_loss=0.1287, over 28915.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3706, pruned_loss=0.1221, over 5677403.48 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3491, pruned_loss=0.09484, over 5694934.46 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.375, pruned_loss=0.1268, over 5668609.58 frames. ], batch size: 112, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 17:04:42,036 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-07 17:05:13,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3455, 1.6389, 1.3588, 0.9958], device='cuda:0'), covar=tensor([0.2324, 0.2465, 0.2675, 0.2128], device='cuda:0'), in_proj_covar=tensor([0.1375, 0.1013, 0.1222, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 17:05:14,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:05:15,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6218, 2.2442, 1.5874, 0.6936], device='cuda:0'), covar=tensor([0.5045, 0.2277, 0.3633, 0.5442], device='cuda:0'), in_proj_covar=tensor([0.1646, 0.1565, 0.1539, 0.1347], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 17:05:17,487 INFO [train.py:968] (0/2) Epoch 15, batch 9250, libri_loss[loss=0.269, simple_loss=0.3456, pruned_loss=0.09626, over 29542.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3703, pruned_loss=0.1222, over 5684491.66 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3492, pruned_loss=0.09486, over 5697846.89 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.374, pruned_loss=0.1263, over 5674689.72 frames. ], batch size: 77, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:05:20,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2933, 1.9002, 1.4214, 0.3502], device='cuda:0'), covar=tensor([0.3408, 0.2474, 0.3637, 0.5204], device='cuda:0'), in_proj_covar=tensor([0.1646, 0.1564, 0.1538, 0.1347], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 17:05:26,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3448, 1.4837, 1.4821, 1.0146], device='cuda:0'), covar=tensor([0.1906, 0.3630, 0.1670, 0.1876], device='cuda:0'), in_proj_covar=tensor([0.0858, 0.0697, 0.0902, 0.0802], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 17:05:27,592 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:05:30,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=648182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:05:37,831 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.946e+02 1.666e+03 2.191e+03 2.904e+03 5.847e+03, threshold=4.382e+03, percent-clipped=10.0
+2023-03-07 17:05:43,997 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=648197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:06:06,514 INFO [train.py:968] (0/2) Epoch 15, batch 9300, libri_loss[loss=0.263, simple_loss=0.3509, pruned_loss=0.0875, over 29540.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3717, pruned_loss=0.1227, over 5680253.53 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3492, pruned_loss=0.09494, over 5700437.83 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3754, pruned_loss=0.1268, over 5669626.24 frames. ], batch size: 89, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:06:16,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1805, 1.3357, 3.9644, 3.3718], device='cuda:0'), covar=tensor([0.1787, 0.2686, 0.0447, 0.0750], device='cuda:0'), in_proj_covar=tensor([0.0705, 0.0615, 0.0908, 0.0830], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 17:06:16,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:06:56,528 INFO [train.py:968] (0/2) Epoch 15, batch 9350, giga_loss[loss=0.325, simple_loss=0.3816, pruned_loss=0.1342, over 28796.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3736, pruned_loss=0.1237, over 5662784.65 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.349, pruned_loss=0.09491, over 5685926.13 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.377, pruned_loss=0.1274, over 5668163.56 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:07:01,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:07:18,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.506e+02 1.347e+03 1.848e+03 2.603e+03 1.004e+04, threshold=3.697e+03, percent-clipped=7.0
+2023-03-07 17:07:31,620 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:07:34,222 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:07:43,525 INFO [train.py:968] (0/2) Epoch 15, batch 9400, giga_loss[loss=0.2655, simple_loss=0.3397, pruned_loss=0.09571, over 28919.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3738, pruned_loss=0.124, over 5667232.41 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3498, pruned_loss=0.09521, over 5689756.46 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3765, pruned_loss=0.1274, over 5667506.88 frames. ], batch size: 174, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:07:46,215 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:07:49,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:08:02,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:08:09,081 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=648346.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:08:15,890 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:08:28,227 INFO [train.py:968] (0/2) Epoch 15, batch 9450, giga_loss[loss=0.3587, simple_loss=0.3875, pruned_loss=0.165, over 23906.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3749, pruned_loss=0.123, over 5671198.46 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.35, pruned_loss=0.09523, over 5693155.25 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3777, pruned_loss=0.1267, over 5667452.52 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:08:30,486 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:08:33,009 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:08:46,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.169e+02 1.486e+03 2.000e+03 2.775e+03 1.025e+04, threshold=4.001e+03, percent-clipped=10.0
+2023-03-07 17:08:57,499 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:09:05,273 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=648411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:09:09,771 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:09:11,093 INFO [train.py:968] (0/2) Epoch 15, batch 9500, giga_loss[loss=0.3164, simple_loss=0.3928, pruned_loss=0.12, over 28900.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3751, pruned_loss=0.1209, over 5672476.27 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3502, pruned_loss=0.09534, over 5688602.51 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.378, pruned_loss=0.1247, over 5672605.19 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:09:12,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:09:42,141 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:09:55,851 INFO [train.py:968] (0/2) Epoch 15, batch 9550, giga_loss[loss=0.3093, simple_loss=0.3851, pruned_loss=0.1168, over 28845.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3762, pruned_loss=0.1201, over 5675767.41 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3503, pruned_loss=0.09538, over 5693645.36 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3792, pruned_loss=0.1238, over 5670625.45 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:10:19,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.078e+02 1.305e+03 1.740e+03 2.404e+03 5.346e+03, threshold=3.480e+03, percent-clipped=7.0
+2023-03-07 17:10:27,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1629, 1.4278, 1.3246, 1.0890], device='cuda:0'), covar=tensor([0.2959, 0.2313, 0.1493, 0.2126], device='cuda:0'), in_proj_covar=tensor([0.1816, 0.1741, 0.1683, 0.1794], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 17:10:49,251 INFO [train.py:968] (0/2) Epoch 15, batch 9600, giga_loss[loss=0.3219, simple_loss=0.3869, pruned_loss=0.1285, over 28912.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3799, pruned_loss=0.1234, over 5681275.92 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3501, pruned_loss=0.09532, over 5694728.14 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3825, pruned_loss=0.1265, over 5676152.05 frames. ], batch size: 227, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 17:11:21,362 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:11:32,880 INFO [train.py:968] (0/2) Epoch 15, batch 9650, giga_loss[loss=0.3089, simple_loss=0.3763, pruned_loss=0.1208, over 28856.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.38, pruned_loss=0.1249, over 5685238.03 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.349, pruned_loss=0.09473, over 5701417.53 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3841, pruned_loss=0.129, over 5674640.90 frames. ], batch size: 112, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:11:36,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:11:56,535 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.804e+03 2.378e+03 3.263e+03 1.043e+04, threshold=4.756e+03, percent-clipped=21.0
+2023-03-07 17:12:22,043 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-07 17:12:22,926 INFO [train.py:968] (0/2) Epoch 15, batch 9700, giga_loss[loss=0.3033, simple_loss=0.3682, pruned_loss=0.1192, over 28937.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3802, pruned_loss=0.1262, over 5678976.07 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3491, pruned_loss=0.09474, over 5704343.34 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3841, pruned_loss=0.13, over 5667438.70 frames. ], batch size: 227, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:12:34,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4096, 1.9442, 1.3773, 0.6759], device='cuda:0'), covar=tensor([0.3729, 0.2054, 0.2778, 0.4634], device='cuda:0'), in_proj_covar=tensor([0.1638, 0.1561, 0.1530, 0.1340], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 17:13:10,041 INFO [train.py:968] (0/2) Epoch 15, batch 9750, giga_loss[loss=0.3064, simple_loss=0.3647, pruned_loss=0.124, over 29073.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3788, pruned_loss=0.1258, over 5658828.29 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3482, pruned_loss=0.09431, over 5704470.95 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3837, pruned_loss=0.1303, over 5648082.97 frames. ], batch size: 128, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:13:28,045 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.768e+03 2.153e+03 3.080e+03 6.797e+03, threshold=4.306e+03, percent-clipped=6.0
+2023-03-07 17:13:37,012 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:13:39,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:13:47,451 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:13:50,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:13:50,424 INFO [train.py:968] (0/2) Epoch 15, batch 9800, giga_loss[loss=0.3172, simple_loss=0.3574, pruned_loss=0.1385, over 23542.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3774, pruned_loss=0.1236, over 5657938.26 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3488, pruned_loss=0.0947, over 5700677.47 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3822, pruned_loss=0.1283, over 5650596.85 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:13:54,324 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648721.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:14:03,734 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:14:17,091 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:14:36,647 INFO [train.py:968] (0/2) Epoch 15, batch 9850, giga_loss[loss=0.347, simple_loss=0.4125, pruned_loss=0.1407, over 28693.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3775, pruned_loss=0.1219, over 5670739.08 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3487, pruned_loss=0.09464, over 5705823.90 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3821, pruned_loss=0.1263, over 5659628.92 frames. ], batch size: 284, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:14:48,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3030, 3.1040, 2.9642, 1.4450], device='cuda:0'), covar=tensor([0.0963, 0.1129, 0.0988, 0.2338], device='cuda:0'), in_proj_covar=tensor([0.1144, 0.1065, 0.0915, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 17:14:54,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:14:57,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.098e+02 1.335e+03 1.738e+03 2.405e+03 4.950e+03, threshold=3.476e+03, percent-clipped=3.0
+2023-03-07 17:15:22,853 INFO [train.py:968] (0/2) Epoch 15, batch 9900, giga_loss[loss=0.3154, simple_loss=0.3805, pruned_loss=0.1251, over 28029.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3773, pruned_loss=0.1211, over 5680488.88 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3484, pruned_loss=0.09453, over 5708363.97 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3823, pruned_loss=0.1258, over 5668339.18 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:16:11,079 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648864.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:16:15,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648867.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:16:16,855 INFO [train.py:968] (0/2) Epoch 15, batch 9950, giga_loss[loss=0.3209, simple_loss=0.3805, pruned_loss=0.1306, over 27623.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3771, pruned_loss=0.1217, over 5668039.24 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3477, pruned_loss=0.09413, over 5712547.66 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3823, pruned_loss=0.1263, over 5653914.21 frames. ], batch size: 472, lr: 2.16e-03, grad_scale: 1.0
+2023-03-07 17:16:43,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.320e+02 1.597e+03 2.578e+03 4.047e+03 1.564e+04, threshold=5.156e+03, percent-clipped=30.0
+2023-03-07 17:16:46,098 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648896.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:17:05,746 INFO [train.py:968] (0/2) Epoch 15, batch 10000, giga_loss[loss=0.2806, simple_loss=0.3579, pruned_loss=0.1016, over 29039.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3774, pruned_loss=0.1225, over 5674830.44 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3477, pruned_loss=0.09412, over 5714435.88 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3819, pruned_loss=0.1266, over 5661328.44 frames. ], batch size: 155, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:17:14,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:17:18,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:17:54,656 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:18:00,678 INFO [train.py:968] (0/2) Epoch 15, batch 10050, giga_loss[loss=0.2902, simple_loss=0.3584, pruned_loss=0.111, over 28764.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3766, pruned_loss=0.1234, over 5667629.34 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3474, pruned_loss=0.09384, over 5717453.16 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3809, pruned_loss=0.1272, over 5653811.22 frames. ], batch size: 119, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:18:10,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1763, 1.2629, 1.1098, 0.9064], device='cuda:0'), covar=tensor([0.0917, 0.0545, 0.1059, 0.1001], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0449, 0.0510, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 17:18:23,527 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.777e+02 1.664e+03 2.104e+03 3.166e+03 8.849e+03, threshold=4.208e+03, percent-clipped=3.0
+2023-03-07 17:18:47,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 17:18:49,212 INFO [train.py:968] (0/2) Epoch 15, batch 10100, giga_loss[loss=0.3127, simple_loss=0.3845, pruned_loss=0.1205, over 28708.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3753, pruned_loss=0.123, over 5666593.05 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3478, pruned_loss=0.09404, over 5710450.19 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3786, pruned_loss=0.1262, over 5661950.06 frames. ], batch size: 262, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:19:44,870 INFO [train.py:968] (0/2) Epoch 15, batch 10150, giga_loss[loss=0.3545, simple_loss=0.3807, pruned_loss=0.1642, over 23433.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3739, pruned_loss=0.123, over 5668082.57 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3483, pruned_loss=0.09428, over 5716060.16 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3769, pruned_loss=0.1262, over 5658347.66 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:20:05,097 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.357e+02 1.567e+03 1.974e+03 2.716e+03 7.497e+03, threshold=3.947e+03, percent-clipped=7.0
+2023-03-07 17:20:31,924 INFO [train.py:968] (0/2) Epoch 15, batch 10200, libri_loss[loss=0.279, simple_loss=0.366, pruned_loss=0.09606, over 29764.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3733, pruned_loss=0.1232, over 5667359.33 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3483, pruned_loss=0.09421, over 5713139.01 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3765, pruned_loss=0.1267, over 5660197.23 frames. ], batch size: 87, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:21:17,739 INFO [train.py:968] (0/2) Epoch 15, batch 10250, giga_loss[loss=0.2634, simple_loss=0.3424, pruned_loss=0.09222, over 28911.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3705, pruned_loss=0.1209, over 5664444.40 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.348, pruned_loss=0.09394, over 5713994.45 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3736, pruned_loss=0.1243, over 5657127.80 frames. ], batch size: 174, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:21:19,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3787, 3.0016, 1.4714, 1.4370], device='cuda:0'), covar=tensor([0.0924, 0.0346, 0.0849, 0.1311], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0532, 0.0358, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 17:21:40,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.439e+02 1.583e+03 2.215e+03 3.420e+03 1.136e+04, threshold=4.430e+03, percent-clipped=17.0
+2023-03-07 17:22:08,729 INFO [train.py:968] (0/2) Epoch 15, batch 10300, giga_loss[loss=0.387, simple_loss=0.4237, pruned_loss=0.1752, over 27976.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3683, pruned_loss=0.1185, over 5663843.89 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.348, pruned_loss=0.09399, over 5715905.85 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.371, pruned_loss=0.1214, over 5656015.24 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:22:42,853 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-07 17:22:59,533 INFO [train.py:968] (0/2) Epoch 15, batch 10350, giga_loss[loss=0.3014, simple_loss=0.3692, pruned_loss=0.1168, over 28729.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3659, pruned_loss=0.1163, over 5664033.15 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3478, pruned_loss=0.09392, over 5718801.21 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3684, pruned_loss=0.1189, over 5654619.14 frames. ], batch size: 284, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:23:23,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.618e+02 1.370e+03 1.785e+03 2.408e+03 1.719e+04, threshold=3.570e+03, percent-clipped=3.0
+2023-03-07 17:23:41,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8142, 1.9892, 1.3349, 1.5911], device='cuda:0'), covar=tensor([0.0817, 0.0531, 0.1018, 0.0999], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0448, 0.0509, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 17:23:50,624 INFO [train.py:968] (0/2) Epoch 15, batch 10400, giga_loss[loss=0.2957, simple_loss=0.3522, pruned_loss=0.1196, over 28703.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3652, pruned_loss=0.1165, over 5658854.79 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3479, pruned_loss=0.09398, over 5708455.91 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3672, pruned_loss=0.1187, over 5659307.82 frames. ], batch size: 99, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:24:41,973 INFO [train.py:968] (0/2) Epoch 15, batch 10450, libri_loss[loss=0.2775, simple_loss=0.36, pruned_loss=0.09755, over 29533.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3633, pruned_loss=0.1162, over 5661415.20 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3482, pruned_loss=0.09412, over 5712073.52 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3649, pruned_loss=0.1183, over 5657697.29 frames. ], batch size: 80, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:24:48,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1853, 1.7907, 1.4084, 0.3300], device='cuda:0'), covar=tensor([0.3486, 0.2250, 0.3173, 0.4859], device='cuda:0'), in_proj_covar=tensor([0.1643, 0.1569, 0.1540, 0.1345], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 17:25:06,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.740e+03 2.310e+03 3.055e+03 7.099e+03, threshold=4.619e+03, percent-clipped=17.0
+2023-03-07 17:25:23,183 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 17:25:31,822 INFO [train.py:968] (0/2) Epoch 15, batch 10500, giga_loss[loss=0.3106, simple_loss=0.3804, pruned_loss=0.1204, over 28737.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3659, pruned_loss=0.118, over 5664439.65 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3483, pruned_loss=0.09409, over 5714025.92 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3673, pruned_loss=0.1199, over 5659227.55 frames. ], batch size: 284, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:26:16,946 INFO [train.py:968] (0/2) Epoch 15, batch 10550, giga_loss[loss=0.2779, simple_loss=0.3477, pruned_loss=0.104, over 28799.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3668, pruned_loss=0.1174, over 5672529.61 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3486, pruned_loss=0.09399, over 5720101.70 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3683, pruned_loss=0.1196, over 5661580.97 frames. ], batch size: 119, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:26:39,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.770e+02 1.392e+03 1.763e+03 2.476e+03 5.017e+03, threshold=3.527e+03, percent-clipped=2.0
+2023-03-07 17:26:59,854 INFO [train.py:968] (0/2) Epoch 15, batch 10600, giga_loss[loss=0.2973, simple_loss=0.3532, pruned_loss=0.1207, over 28921.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3667, pruned_loss=0.1169, over 5641981.08 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3492, pruned_loss=0.09442, over 5708518.71 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3681, pruned_loss=0.1195, over 5639943.06 frames. ], batch size: 106, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:27:18,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3064, 1.3992, 1.3333, 1.4839], device='cuda:0'), covar=tensor([0.0728, 0.0388, 0.0316, 0.0773], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 17:27:26,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9924, 3.8333, 3.6140, 1.9330], device='cuda:0'), covar=tensor([0.0629, 0.0746, 0.0724, 0.2168], device='cuda:0'), in_proj_covar=tensor([0.1138, 0.1060, 0.0916, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 17:27:28,450 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 17:27:37,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5382, 1.5547, 1.2513, 1.1654], device='cuda:0'), covar=tensor([0.0840, 0.0555, 0.1018, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0444, 0.0504, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 17:27:46,672 INFO [train.py:968] (0/2) Epoch 15, batch 10650, giga_loss[loss=0.2869, simple_loss=0.3569, pruned_loss=0.1084, over 28391.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3671, pruned_loss=0.1174, over 5638507.74 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3496, pruned_loss=0.0948, over 5714634.91 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3685, pruned_loss=0.1201, over 5628726.99 frames. ], batch size: 60, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:28:07,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.222e+02 1.522e+03 2.169e+03 3.424e+03 1.061e+04, threshold=4.338e+03, percent-clipped=23.0
+2023-03-07 17:28:32,644 INFO [train.py:968] (0/2) Epoch 15, batch 10700, giga_loss[loss=0.2876, simple_loss=0.3652, pruned_loss=0.105, over 28767.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3688, pruned_loss=0.119, over 5649476.60 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.35, pruned_loss=0.09503, over 5715478.81 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.37, pruned_loss=0.1214, over 5639436.11 frames. ], batch size: 284, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:29:12,509 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=649659.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:29:24,457 INFO [train.py:968] (0/2) Epoch 15, batch 10750, giga_loss[loss=0.303, simple_loss=0.3706, pruned_loss=0.1177, over 28938.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3707, pruned_loss=0.1201, over 5648826.92 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3506, pruned_loss=0.09514, over 5718751.68 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3719, pruned_loss=0.1227, over 5635552.27 frames. ], batch size: 106, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:29:44,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3085, 1.5161, 3.1151, 2.9465], device='cuda:0'), covar=tensor([0.1279, 0.2161, 0.0475, 0.0950], device='cuda:0'), in_proj_covar=tensor([0.0711, 0.0618, 0.0911, 0.0834], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 17:29:48,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.177e+02 1.518e+03 1.859e+03 2.379e+03 6.563e+03, threshold=3.719e+03, percent-clipped=7.0
+2023-03-07 17:29:51,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5093, 1.8927, 1.5633, 1.6983], device='cuda:0'), covar=tensor([0.0751, 0.0280, 0.0305, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0057, 0.0096], device='cuda:0')
+2023-03-07 17:30:16,610 INFO [train.py:968] (0/2) Epoch 15, batch 10800, libri_loss[loss=0.28, simple_loss=0.3523, pruned_loss=0.1039, over 29576.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3721, pruned_loss=0.121, over 5651917.50 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3506, pruned_loss=0.0952, over 5718883.44 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3731, pruned_loss=0.1232, over 5640605.60 frames. ], batch size: 77, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:30:38,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2493, 1.5247, 1.1961, 0.9695], device='cuda:0'), covar=tensor([0.2491, 0.2429, 0.2759, 0.2108], device='cuda:0'), in_proj_covar=tensor([0.1379, 0.1010, 0.1224, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 17:30:53,818 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-07 17:31:02,113 INFO [train.py:968] (0/2) Epoch 15, batch 10850, giga_loss[loss=0.4251, simple_loss=0.4501, pruned_loss=0.2, over 28211.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3741, pruned_loss=0.1229, over 5653469.26 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.351, pruned_loss=0.09534, over 5720171.76 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3753, pruned_loss=0.1253, over 5641629.17 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:31:28,973 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.992e+02 1.584e+03 2.099e+03 2.863e+03 7.885e+03, threshold=4.197e+03, percent-clipped=12.0
+2023-03-07 17:31:40,495 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=649806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:31:52,655 INFO [train.py:968] (0/2) Epoch 15, batch 10900, giga_loss[loss=0.3117, simple_loss=0.3823, pruned_loss=0.1205, over 28929.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3761, pruned_loss=0.125, over 5658206.63 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3515, pruned_loss=0.09565, over 5723154.61 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3769, pruned_loss=0.1271, over 5644857.92 frames. ], batch size: 136, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:32:42,342 INFO [train.py:968] (0/2) Epoch 15, batch 10950, giga_loss[loss=0.333, simple_loss=0.3975, pruned_loss=0.1342, over 28908.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3763, pruned_loss=0.1232, over 5666991.60 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3514, pruned_loss=0.09559, over 5728174.98 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3777, pruned_loss=0.1258, over 5650016.81 frames. ], batch size: 227, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:32:57,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2722, 1.1866, 1.1201, 1.5457], device='cuda:0'), covar=tensor([0.0769, 0.0351, 0.0356, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:0')
+2023-03-07 17:33:07,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.735e+03 2.164e+03 3.305e+03 1.248e+04, threshold=4.328e+03, percent-clipped=12.0
+2023-03-07 17:33:22,812 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=649910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:33:25,063 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=649912.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:33:32,149 INFO [train.py:968] (0/2) Epoch 15, batch 11000, giga_loss[loss=0.3719, simple_loss=0.4217, pruned_loss=0.161, over 28466.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3758, pruned_loss=0.1237, over 5658011.78 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3516, pruned_loss=0.0959, over 5733354.67 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3776, pruned_loss=0.1263, over 5637425.48 frames. ], batch size: 336, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:34:18,609 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=649966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:34:22,644 INFO [train.py:968] (0/2) Epoch 15, batch 11050, giga_loss[loss=0.254, simple_loss=0.3303, pruned_loss=0.08891, over 28966.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3739, pruned_loss=0.1229, over 5671411.67 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3518, pruned_loss=0.09599, over 5736044.02 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3757, pruned_loss=0.1256, over 5650976.66 frames. ], batch size: 136, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:34:51,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.781e+02 1.444e+03 1.821e+03 2.451e+03 5.369e+03, threshold=3.642e+03, percent-clipped=3.0
+2023-03-07 17:35:00,476 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-650000.pt
+2023-03-07 17:35:23,073 INFO [train.py:968] (0/2) Epoch 15, batch 11100, giga_loss[loss=0.2612, simple_loss=0.3346, pruned_loss=0.09395, over 28665.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.373, pruned_loss=0.1229, over 5651349.25 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3518, pruned_loss=0.09606, over 5728245.67 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3745, pruned_loss=0.1252, over 5641985.42 frames. ], batch size: 262, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:35:38,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650034.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:36:10,195 INFO [train.py:968] (0/2) Epoch 15, batch 11150, libri_loss[loss=0.3544, simple_loss=0.4114, pruned_loss=0.1487, over 29378.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3719, pruned_loss=0.1221, over 5669608.13 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3518, pruned_loss=0.09614, over 5732590.15 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3736, pruned_loss=0.1244, over 5656630.45 frames. ], batch size: 92, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:36:28,417 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8164, 4.6250, 4.3506, 2.1737], device='cuda:0'), covar=tensor([0.0658, 0.0836, 0.1011, 0.2059], device='cuda:0'), in_proj_covar=tensor([0.1150, 0.1063, 0.0921, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 17:36:37,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.765e+02 1.594e+03 1.922e+03 2.650e+03 7.730e+03, threshold=3.844e+03, percent-clipped=8.0
+2023-03-07 17:37:00,644 INFO [train.py:968] (0/2) Epoch 15, batch 11200, giga_loss[loss=0.2676, simple_loss=0.3406, pruned_loss=0.09732, over 28938.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3719, pruned_loss=0.123, over 5670243.85 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3522, pruned_loss=0.0963, over 5734573.31 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3732, pruned_loss=0.125, over 5657388.65 frames. ], batch size: 136, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 17:37:39,680 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0246, 1.2427, 1.3083, 1.0711], device='cuda:0'), covar=tensor([0.1398, 0.1206, 0.1920, 0.1446], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0740, 0.0694, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 17:37:50,346 INFO [train.py:968] (0/2) Epoch 15, batch 11250, giga_loss[loss=0.2994, simple_loss=0.3599, pruned_loss=0.1194, over 28924.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3727, pruned_loss=0.1241, over 5667878.22 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3522, pruned_loss=0.09625, over 5735090.44 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3738, pruned_loss=0.1259, over 5657053.21 frames. ], batch size: 106, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:37:59,567 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=650177.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:38:01,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=650180.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:38:02,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:38:18,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.404e+02 1.421e+03 1.933e+03 2.735e+03 7.673e+03, threshold=3.865e+03, percent-clipped=5.0
+2023-03-07 17:38:31,433 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=650209.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:38:40,851 INFO [train.py:968] (0/2) Epoch 15, batch 11300, giga_loss[loss=0.2959, simple_loss=0.3594, pruned_loss=0.1162, over 28816.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3727, pruned_loss=0.124, over 5666694.18 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3522, pruned_loss=0.09618, over 5734807.88 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3741, pruned_loss=0.1261, over 5656639.65 frames. ], batch size: 119, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:38:58,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7056, 1.3724, 1.7694, 1.3466], device='cuda:0'), covar=tensor([0.1806, 0.2879, 0.1386, 0.1598], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0697, 0.0901, 0.0802], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 17:39:26,298 INFO [train.py:968] (0/2) Epoch 15, batch 11350, giga_loss[loss=0.3593, simple_loss=0.4122, pruned_loss=0.1532, over 28585.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3742, pruned_loss=0.1251, over 5670229.98 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.352, pruned_loss=0.09603, over 5737239.18 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3762, pruned_loss=0.128, over 5657426.13 frames. ], batch size: 307, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:39:33,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2047, 1.2480, 1.1273, 0.8991], device='cuda:0'), covar=tensor([0.0857, 0.0501, 0.1000, 0.0960], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0445, 0.0504, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 17:39:42,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:39:44,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650287.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:39:52,177 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.512e+02 1.543e+03 2.286e+03 3.518e+03 1.031e+04, threshold=4.572e+03, percent-clipped=21.0
+2023-03-07 17:40:16,580 INFO [train.py:968] (0/2) Epoch 15, batch 11400, giga_loss[loss=0.3078, simple_loss=0.3697, pruned_loss=0.1229, over 28915.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3743, pruned_loss=0.1246, over 5674259.72 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3517, pruned_loss=0.09579, over 5740109.36 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3767, pruned_loss=0.1277, over 5660166.92 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:40:20,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=650324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:40:24,801 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=650327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:40:41,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:40:53,751 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=650356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:41:09,349 INFO [train.py:968] (0/2) Epoch 15, batch 11450, giga_loss[loss=0.3125, simple_loss=0.3623, pruned_loss=0.1313, over 28869.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3763, pruned_loss=0.1273, over 5668420.86 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3517, pruned_loss=0.0958, over 5740945.81 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3782, pruned_loss=0.1299, over 5656273.03 frames. ], batch size: 99, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:41:35,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.672e+02 1.526e+03 2.011e+03 2.508e+03 6.340e+03, threshold=4.022e+03, percent-clipped=3.0
+2023-03-07 17:41:57,083 INFO [train.py:968] (0/2) Epoch 15, batch 11500, giga_loss[loss=0.2963, simple_loss=0.3612, pruned_loss=0.1157, over 28923.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.375, pruned_loss=0.127, over 5665240.43 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3512, pruned_loss=0.09563, over 5744050.47 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3776, pruned_loss=0.13, over 5650960.65 frames. ], batch size: 213, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:42:06,475 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=650428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:42:08,019 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=650430.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:42:08,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=650431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:42:09,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=650433.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:42:39,046 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=650460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:42:40,679 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=650462.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:42:45,230 INFO [train.py:968] (0/2) Epoch 15, batch 11550, giga_loss[loss=0.3221, simple_loss=0.3805, pruned_loss=0.1319, over 28874.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3746, pruned_loss=0.1261, over 5676433.75 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3511, pruned_loss=0.09561, over 5746420.38 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3774, pruned_loss=0.1292, over 5661014.26 frames. ], batch size: 227, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:43:00,999 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=650484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:43:03,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=650487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:43:06,472 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:43:08,847 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.683e+03 2.145e+03 3.105e+03 9.789e+03, threshold=4.290e+03, percent-clipped=17.0
+2023-03-07 17:43:30,737 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=650516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:43:32,259 INFO [train.py:968] (0/2) Epoch 15, batch 11600, giga_loss[loss=0.3056, simple_loss=0.3764, pruned_loss=0.1174, over 28727.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3744, pruned_loss=0.1254, over 5676837.22 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3509, pruned_loss=0.09567, over 5752554.05 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3777, pruned_loss=0.129, over 5656286.24 frames. ], batch size: 284, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 17:43:39,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3984, 1.7160, 1.3504, 1.5457], device='cuda:0'), covar=tensor([0.2693, 0.2648, 0.3025, 0.2216], device='cuda:0'), in_proj_covar=tensor([0.1387, 0.1018, 0.1230, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 17:43:57,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4156, 1.8927, 1.3178, 0.7515], device='cuda:0'), covar=tensor([0.4242, 0.2183, 0.2612, 0.4732], device='cuda:0'), in_proj_covar=tensor([0.1638, 0.1565, 0.1541, 0.1337], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 17:43:58,092 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:44:00,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4940, 1.6424, 1.5633, 1.3547], device='cuda:0'), covar=tensor([0.2286, 0.1991, 0.1607, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.1830, 0.1753, 0.1696, 0.1808], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 17:44:05,926 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:44:23,267 INFO [train.py:968] (0/2) Epoch 15, batch 11650, giga_loss[loss=0.3503, simple_loss=0.391, pruned_loss=0.1548, over 27628.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.375, pruned_loss=0.1253, over 5686904.46 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3511, pruned_loss=0.0957, over 5755967.74 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3779, pruned_loss=0.1287, over 5666052.85 frames. ], batch size: 472, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:44:50,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.501e+03 1.938e+03 2.466e+03 4.574e+03, threshold=3.875e+03, percent-clipped=1.0
+2023-03-07 17:45:12,872 INFO [train.py:968] (0/2) Epoch 15, batch 11700, giga_loss[loss=0.3051, simple_loss=0.3724, pruned_loss=0.1189, over 28904.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3773, pruned_loss=0.1274, over 5684715.69 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.351, pruned_loss=0.09569, over 5759409.45 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3803, pruned_loss=0.1309, over 5663431.50 frames. ], batch size: 145, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:45:47,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4655, 1.7152, 1.3638, 1.5904], device='cuda:0'), covar=tensor([0.2468, 0.2450, 0.2727, 0.2165], device='cuda:0'), in_proj_covar=tensor([0.1382, 0.1015, 0.1225, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 17:45:59,946 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2871, 4.0801, 3.8569, 2.0448], device='cuda:0'), covar=tensor([0.0629, 0.0791, 0.0769, 0.1855], device='cuda:0'), in_proj_covar=tensor([0.1146, 0.1064, 0.0917, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 17:46:00,400 INFO [train.py:968] (0/2) Epoch 15, batch 11750, giga_loss[loss=0.2778, simple_loss=0.3549, pruned_loss=0.1004, over 28778.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3769, pruned_loss=0.1267, over 5694073.97 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3509, pruned_loss=0.09556, over 5761274.91 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.38, pruned_loss=0.1303, over 5673533.53 frames. ], batch size: 92, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:46:25,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.675e+03 2.323e+03 3.569e+03 7.848e+03, threshold=4.646e+03, percent-clipped=19.0
+2023-03-07 17:46:48,459 INFO [train.py:968] (0/2) Epoch 15, batch 11800, giga_loss[loss=0.3049, simple_loss=0.3844, pruned_loss=0.1127, over 28713.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3775, pruned_loss=0.1263, over 5694140.27 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3507, pruned_loss=0.09544, over 5764066.36 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3806, pruned_loss=0.1298, over 5674322.13 frames. ], batch size: 262, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:47:00,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4136, 1.7474, 1.4007, 1.5679], device='cuda:0'), covar=tensor([0.2425, 0.2345, 0.2678, 0.2062], device='cuda:0'), in_proj_covar=tensor([0.1383, 0.1016, 0.1229, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 17:47:16,223 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:47:35,376 INFO [train.py:968] (0/2) Epoch 15, batch 11850, giga_loss[loss=0.3132, simple_loss=0.3794, pruned_loss=0.1236, over 28769.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3769, pruned_loss=0.1248, over 5691592.20 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3507, pruned_loss=0.09547, over 5766100.56 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3799, pruned_loss=0.1282, over 5672697.42 frames. ], batch size: 99, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:48:02,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.692e+02 1.482e+03 1.941e+03 2.681e+03 5.345e+03, threshold=3.882e+03, percent-clipped=4.0
+2023-03-07 17:48:11,568 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:48:24,088 INFO [train.py:968] (0/2) Epoch 15, batch 11900, giga_loss[loss=0.3057, simple_loss=0.3792, pruned_loss=0.1161, over 28925.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3754, pruned_loss=0.1238, over 5680293.70 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3507, pruned_loss=0.09547, over 5762049.54 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3787, pruned_loss=0.1276, over 5665969.47 frames. ], batch size: 227, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:49:11,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:49:14,192 INFO [train.py:968] (0/2) Epoch 15, batch 11950, giga_loss[loss=0.3128, simple_loss=0.3744, pruned_loss=0.1256, over 28307.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3744, pruned_loss=0.1231, over 5691717.09 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3512, pruned_loss=0.09561, over 5764450.19 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3771, pruned_loss=0.1265, over 5676833.36 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:49:24,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.05 vs. limit=5.0
+2023-03-07 17:49:42,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.964e+02 1.484e+03 1.796e+03 2.510e+03 1.103e+04, threshold=3.592e+03, percent-clipped=9.0
+2023-03-07 17:50:01,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8163, 4.7467, 1.9688, 1.7649], device='cuda:0'), covar=tensor([0.0912, 0.0332, 0.0814, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0532, 0.0359, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 17:50:02,688 INFO [train.py:968] (0/2) Epoch 15, batch 12000, giga_loss[loss=0.4173, simple_loss=0.4391, pruned_loss=0.1977, over 26600.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3754, pruned_loss=0.1243, over 5671669.81 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3511, pruned_loss=0.0956, over 5764815.57 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3781, pruned_loss=0.1276, over 5657574.11 frames. ], batch size: 555, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:50:02,694 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 17:50:11,132 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.213, simple_loss=0.3197, pruned_loss=0.05318, over 944034.00 frames. 
+2023-03-07 17:50:11,133 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 17:50:12,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:50:19,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:50:24,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4201, 1.5188, 1.5271, 1.3408], device='cuda:0'), covar=tensor([0.1512, 0.1819, 0.2094, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.0451, 0.0737, 0.0690, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 17:50:43,659 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:50:45,796 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650955.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:50:57,308 INFO [train.py:968] (0/2) Epoch 15, batch 12050, giga_loss[loss=0.2948, simple_loss=0.365, pruned_loss=0.1122, over 29102.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3779, pruned_loss=0.1257, over 5678612.72 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3512, pruned_loss=0.09559, over 5765877.30 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3804, pruned_loss=0.1289, over 5665221.17 frames. ], batch size: 155, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:51:25,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.664e+03 2.254e+03 3.812e+03 1.152e+04, threshold=4.507e+03, percent-clipped=26.0
+2023-03-07 17:51:38,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:51:43,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:51:48,233 INFO [train.py:968] (0/2) Epoch 15, batch 12100, giga_loss[loss=0.3034, simple_loss=0.3705, pruned_loss=0.1181, over 28911.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3777, pruned_loss=0.127, over 5668465.58 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3511, pruned_loss=0.09563, over 5759638.46 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3802, pruned_loss=0.13, over 5662130.33 frames. ], batch size: 174, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:52:13,933 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:21,903 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:33,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:36,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:38,040 INFO [train.py:968] (0/2) Epoch 15, batch 12150, giga_loss[loss=0.2806, simple_loss=0.3545, pruned_loss=0.1034, over 28530.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3782, pruned_loss=0.1281, over 5664418.88 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3511, pruned_loss=0.09556, over 5762539.67 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3807, pruned_loss=0.1311, over 5654840.14 frames. ], batch size: 71, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:52:39,627 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:41,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:49,531 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:53:04,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:53:06,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.913e+02 1.504e+03 1.926e+03 2.705e+03 8.619e+03, threshold=3.852e+03, percent-clipped=4.0
+2023-03-07 17:53:13,433 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:53:15,760 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651106.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:53:28,369 INFO [train.py:968] (0/2) Epoch 15, batch 12200, giga_loss[loss=0.3374, simple_loss=0.3942, pruned_loss=0.1403, over 28356.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3796, pruned_loss=0.1298, over 5661994.67 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3507, pruned_loss=0.0954, over 5764848.91 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3822, pruned_loss=0.1328, over 5651165.11 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:53:33,643 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:54:18,995 INFO [train.py:968] (0/2) Epoch 15, batch 12250, giga_loss[loss=0.3179, simple_loss=0.3789, pruned_loss=0.1284, over 28738.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3805, pruned_loss=0.13, over 5668337.52 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3508, pruned_loss=0.09539, over 5767349.90 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3831, pruned_loss=0.133, over 5655758.52 frames. ], batch size: 242, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:54:29,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:54:44,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.669e+03 2.117e+03 3.144e+03 1.006e+04, threshold=4.234e+03, percent-clipped=13.0
+2023-03-07 17:55:07,042 INFO [train.py:968] (0/2) Epoch 15, batch 12300, giga_loss[loss=0.3242, simple_loss=0.3874, pruned_loss=0.1305, over 28975.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3799, pruned_loss=0.1299, over 5645567.90 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.351, pruned_loss=0.09542, over 5770133.90 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3823, pruned_loss=0.1329, over 5631499.32 frames. ], batch size: 213, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:55:18,281 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:55:58,057 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:00,060 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7082, 2.4278, 1.5979, 0.9263], device='cuda:0'), covar=tensor([0.6343, 0.3333, 0.3089, 0.5441], device='cuda:0'), in_proj_covar=tensor([0.1637, 0.1566, 0.1541, 0.1337], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 17:56:00,355 INFO [train.py:968] (0/2) Epoch 15, batch 12350, giga_loss[loss=0.2926, simple_loss=0.3675, pruned_loss=0.1089, over 28589.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3797, pruned_loss=0.1297, over 5648240.00 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3511, pruned_loss=0.09543, over 5770957.23 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3816, pruned_loss=0.1322, over 5636002.90 frames. ], batch size: 307, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:56:00,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:08,702 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651277.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:56:26,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.917e+02 1.439e+03 1.776e+03 2.444e+03 9.916e+03, threshold=3.552e+03, percent-clipped=6.0
+2023-03-07 17:56:27,174 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:46,000 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:46,990 INFO [train.py:968] (0/2) Epoch 15, batch 12400, giga_loss[loss=0.2918, simple_loss=0.361, pruned_loss=0.1113, over 28827.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3794, pruned_loss=0.1289, over 5651706.43 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3512, pruned_loss=0.09558, over 5773764.56 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3815, pruned_loss=0.1315, over 5636689.22 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:56:52,120 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:54,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:56,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:56,834 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:56,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651330.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:57:25,924 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:57:37,985 INFO [train.py:968] (0/2) Epoch 15, batch 12450, giga_loss[loss=0.3103, simple_loss=0.3768, pruned_loss=0.1219, over 28855.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3774, pruned_loss=0.1275, over 5658745.90 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3511, pruned_loss=0.09547, over 5775799.95 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3795, pruned_loss=0.1301, over 5643633.09 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:58:04,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.590e+03 2.144e+03 2.818e+03 7.245e+03, threshold=4.288e+03, percent-clipped=12.0
+2023-03-07 17:58:26,669 INFO [train.py:968] (0/2) Epoch 15, batch 12500, giga_loss[loss=0.2975, simple_loss=0.3623, pruned_loss=0.1163, over 28598.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3747, pruned_loss=0.1257, over 5657987.72 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3512, pruned_loss=0.09549, over 5768054.12 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3768, pruned_loss=0.1283, over 5651305.60 frames. ], batch size: 85, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:58:34,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:58:57,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:59:12,289 INFO [train.py:968] (0/2) Epoch 15, batch 12550, giga_loss[loss=0.2953, simple_loss=0.3595, pruned_loss=0.1155, over 28870.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3712, pruned_loss=0.1231, over 5662966.83 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3513, pruned_loss=0.09561, over 5754939.66 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3731, pruned_loss=0.1257, over 5666931.59 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:59:14,448 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:59:16,447 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651473.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:59:17,745 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:59:19,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651476.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:59:24,132 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651481.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:59:40,897 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.826e+03 2.459e+03 3.560e+03 1.053e+04, threshold=4.919e+03, percent-clipped=18.0
+2023-03-07 17:59:47,877 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:59:50,018 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651505.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:00:05,359 INFO [train.py:968] (0/2) Epoch 15, batch 12600, giga_loss[loss=0.3075, simple_loss=0.3689, pruned_loss=0.1231, over 28631.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3691, pruned_loss=0.1234, over 5632832.65 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3516, pruned_loss=0.0959, over 5747969.27 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3706, pruned_loss=0.1254, over 5640203.80 frames. ], batch size: 307, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:00:32,053 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-07 18:00:52,567 INFO [train.py:968] (0/2) Epoch 15, batch 12650, giga_loss[loss=0.2844, simple_loss=0.3497, pruned_loss=0.1096, over 28997.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3682, pruned_loss=0.1231, over 5643253.14 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3517, pruned_loss=0.0959, over 5747924.45 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3696, pruned_loss=0.1253, over 5647013.09 frames. ], batch size: 155, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:00:53,285 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:00:57,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:01:23,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:01:25,205 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.706e+02 1.821e+03 2.239e+03 3.149e+03 8.993e+03, threshold=4.478e+03, percent-clipped=7.0
+2023-03-07 18:01:26,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:01:26,850 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:01:28,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:01:44,416 INFO [train.py:968] (0/2) Epoch 15, batch 12700, giga_loss[loss=0.3516, simple_loss=0.3928, pruned_loss=0.1552, over 26518.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3683, pruned_loss=0.1236, over 5642265.63 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3521, pruned_loss=0.09609, over 5749889.95 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3694, pruned_loss=0.1256, over 5641991.12 frames. ], batch size: 555, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:01:50,160 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651624.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:01:52,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651627.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:01:53,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:02:14,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651652.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:02:18,010 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651656.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:02:28,955 INFO [train.py:968] (0/2) Epoch 15, batch 12750, giga_loss[loss=0.2562, simple_loss=0.34, pruned_loss=0.0862, over 28247.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3667, pruned_loss=0.1204, over 5654240.00 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3522, pruned_loss=0.09631, over 5756723.69 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3684, pruned_loss=0.1233, over 5642029.69 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:02:53,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:02:57,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.498e+02 1.507e+03 1.951e+03 2.718e+03 1.203e+04, threshold=3.902e+03, percent-clipped=5.0
+2023-03-07 18:03:04,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:03:21,542 INFO [train.py:968] (0/2) Epoch 15, batch 12800, giga_loss[loss=0.2475, simple_loss=0.3316, pruned_loss=0.08167, over 28614.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.364, pruned_loss=0.1166, over 5654557.39 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3514, pruned_loss=0.09597, over 5758036.92 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3663, pruned_loss=0.1196, over 5642497.50 frames. ], batch size: 336, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:03:51,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1998, 3.9976, 3.7646, 1.8403], device='cuda:0'), covar=tensor([0.0645, 0.0862, 0.0926, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.1146, 0.1065, 0.0916, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 18:03:53,045 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:03:55,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:04:04,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5132, 2.0569, 1.8493, 1.4894], device='cuda:0'), covar=tensor([0.2538, 0.1496, 0.1669, 0.1983], device='cuda:0'), in_proj_covar=tensor([0.1818, 0.1746, 0.1680, 0.1798], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 18:04:14,084 INFO [train.py:968] (0/2) Epoch 15, batch 12850, giga_loss[loss=0.2699, simple_loss=0.3528, pruned_loss=0.09357, over 28626.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3612, pruned_loss=0.1136, over 5653373.30 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.351, pruned_loss=0.09589, over 5760167.68 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3634, pruned_loss=0.1162, over 5640243.82 frames. ], batch size: 307, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:04:15,739 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:04:24,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:04:32,655 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1772, 1.4564, 1.3180, 1.1016], device='cuda:0'), covar=tensor([0.2114, 0.1825, 0.1266, 0.1765], device='cuda:0'), in_proj_covar=tensor([0.1814, 0.1743, 0.1676, 0.1794], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 18:04:42,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651795.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:04:45,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651798.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:04:45,897 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.756e+02 1.436e+03 2.061e+03 3.223e+03 6.150e+03, threshold=4.121e+03, percent-clipped=14.0
+2023-03-07 18:05:07,297 INFO [train.py:968] (0/2) Epoch 15, batch 12900, giga_loss[loss=0.2458, simple_loss=0.3313, pruned_loss=0.08014, over 28948.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3581, pruned_loss=0.1102, over 5658370.39 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3507, pruned_loss=0.09583, over 5762835.04 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3604, pruned_loss=0.1126, over 5643659.64 frames. ], batch size: 164, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:05:15,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651827.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:05:27,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:05:30,908 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:05:32,091 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2597, 1.6778, 1.6141, 1.5378], device='cuda:0'), covar=tensor([0.1737, 0.1577, 0.1810, 0.1598], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0729, 0.0685, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 18:05:41,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:05:43,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:05:59,709 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:06:01,821 INFO [train.py:968] (0/2) Epoch 15, batch 12950, giga_loss[loss=0.2051, simple_loss=0.2834, pruned_loss=0.0634, over 24217.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3538, pruned_loss=0.1058, over 5649368.70 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3502, pruned_loss=0.09561, over 5763426.59 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.356, pruned_loss=0.1081, over 5635979.21 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:06:14,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:06:28,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.467e+02 1.349e+03 1.655e+03 2.307e+03 6.035e+03, threshold=3.310e+03, percent-clipped=3.0
+2023-03-07 18:06:50,116 INFO [train.py:968] (0/2) Epoch 15, batch 13000, giga_loss[loss=0.3457, simple_loss=0.3888, pruned_loss=0.1512, over 26712.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3538, pruned_loss=0.1035, over 5667801.20 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3502, pruned_loss=0.09586, over 5767062.59 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3557, pruned_loss=0.1054, over 5650477.08 frames. ], batch size: 555, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:07:14,054 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-07 18:07:24,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651951.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:07:39,521 INFO [train.py:968] (0/2) Epoch 15, batch 13050, giga_loss[loss=0.287, simple_loss=0.3628, pruned_loss=0.1056, over 28011.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3553, pruned_loss=0.1052, over 5656522.12 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3498, pruned_loss=0.09585, over 5761145.29 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3573, pruned_loss=0.1069, over 5644894.04 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:08:09,174 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.453e+02 1.525e+03 1.987e+03 3.086e+03 8.452e+03, threshold=3.974e+03, percent-clipped=20.0
+2023-03-07 18:08:09,896 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-652000.pt
+2023-03-07 18:08:24,914 INFO [train.py:968] (0/2) Epoch 15, batch 13100, giga_loss[loss=0.2565, simple_loss=0.3377, pruned_loss=0.08769, over 28924.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3538, pruned_loss=0.1039, over 5659539.31 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3497, pruned_loss=0.09583, over 5761992.72 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3557, pruned_loss=0.1056, over 5646021.88 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:08:32,192 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-07 18:08:42,608 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:08:49,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4198, 1.0614, 4.0404, 3.2606], device='cuda:0'), covar=tensor([0.1536, 0.2855, 0.0395, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0702, 0.0612, 0.0898, 0.0821], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 18:09:15,077 INFO [train.py:968] (0/2) Epoch 15, batch 13150, giga_loss[loss=0.2656, simple_loss=0.3396, pruned_loss=0.09578, over 28919.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3503, pruned_loss=0.1016, over 5649480.54 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3495, pruned_loss=0.09575, over 5762255.21 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.352, pruned_loss=0.1031, over 5637627.66 frames. ], batch size: 213, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:09:43,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.162e+02 1.267e+03 1.590e+03 2.091e+03 5.984e+03, threshold=3.179e+03, percent-clipped=7.0
+2023-03-07 18:09:51,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5306, 1.2970, 4.8073, 3.4006], device='cuda:0'), covar=tensor([0.1615, 0.2723, 0.0380, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0700, 0.0610, 0.0897, 0.0818], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 18:10:02,058 INFO [train.py:968] (0/2) Epoch 15, batch 13200, giga_loss[loss=0.3078, simple_loss=0.3651, pruned_loss=0.1252, over 26538.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3499, pruned_loss=0.1016, over 5641736.24 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3493, pruned_loss=0.09572, over 5763000.22 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3515, pruned_loss=0.103, over 5628309.11 frames. ], batch size: 555, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 18:10:26,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:10:37,786 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-07 18:10:48,783 INFO [train.py:968] (0/2) Epoch 15, batch 13250, giga_loss[loss=0.2527, simple_loss=0.3308, pruned_loss=0.08729, over 28997.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3499, pruned_loss=0.1014, over 5649584.04 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3493, pruned_loss=0.09585, over 5764811.52 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3512, pruned_loss=0.1025, over 5634930.46 frames. ], batch size: 128, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 18:11:11,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6217, 1.7880, 1.4605, 1.7923], device='cuda:0'), covar=tensor([0.2588, 0.2395, 0.2586, 0.2279], device='cuda:0'), in_proj_covar=tensor([0.1388, 0.1017, 0.1235, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 18:11:18,760 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.047e+02 1.288e+03 1.765e+03 2.371e+03 5.316e+03, threshold=3.530e+03, percent-clipped=9.0
+2023-03-07 18:11:29,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:11:39,454 INFO [train.py:968] (0/2) Epoch 15, batch 13300, giga_loss[loss=0.2153, simple_loss=0.3066, pruned_loss=0.06201, over 28932.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3467, pruned_loss=0.09851, over 5653326.17 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3488, pruned_loss=0.0956, over 5765611.89 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3481, pruned_loss=0.09967, over 5639564.26 frames. ], batch size: 164, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 18:12:33,439 INFO [train.py:968] (0/2) Epoch 15, batch 13350, giga_loss[loss=0.2612, simple_loss=0.3485, pruned_loss=0.08698, over 28944.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3429, pruned_loss=0.09529, over 5650056.82 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3487, pruned_loss=0.09558, over 5766307.83 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.344, pruned_loss=0.09623, over 5638193.25 frames. ], batch size: 164, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:12:35,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7315, 1.8153, 1.4389, 2.2054], device='cuda:0'), covar=tensor([0.2799, 0.2741, 0.3097, 0.2171], device='cuda:0'), in_proj_covar=tensor([0.1387, 0.1015, 0.1233, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 18:12:53,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=652289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:12:56,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=652292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:13:06,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.864e+02 1.434e+03 2.016e+03 3.002e+03 7.683e+03, threshold=4.032e+03, percent-clipped=12.0
+2023-03-07 18:13:25,457 INFO [train.py:968] (0/2) Epoch 15, batch 13400, giga_loss[loss=0.2286, simple_loss=0.3126, pruned_loss=0.07233, over 29059.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3392, pruned_loss=0.09336, over 5655989.65 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3482, pruned_loss=0.0953, over 5766858.82 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3405, pruned_loss=0.09435, over 5644164.77 frames. ], batch size: 128, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:13:28,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:13:31,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652326.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:13:47,494 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:14:15,578 INFO [train.py:968] (0/2) Epoch 15, batch 13450, giga_loss[loss=0.2845, simple_loss=0.3509, pruned_loss=0.1091, over 28238.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3382, pruned_loss=0.09342, over 5660869.25 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3479, pruned_loss=0.09529, over 5770403.42 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3393, pruned_loss=0.09418, over 5645328.40 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:14:47,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.920e+02 1.253e+03 1.641e+03 2.444e+03 4.556e+03, threshold=3.281e+03, percent-clipped=2.0
+2023-03-07 18:14:54,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:14:54,967 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 18:15:05,176 INFO [train.py:968] (0/2) Epoch 15, batch 13500, giga_loss[loss=0.2645, simple_loss=0.3441, pruned_loss=0.09244, over 28618.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3375, pruned_loss=0.09338, over 5648328.06 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3474, pruned_loss=0.09505, over 5770112.92 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3386, pruned_loss=0.09417, over 5633679.38 frames. ], batch size: 307, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:15:57,722 INFO [train.py:968] (0/2) Epoch 15, batch 13550, giga_loss[loss=0.2999, simple_loss=0.3753, pruned_loss=0.1123, over 28876.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3391, pruned_loss=0.09367, over 5659279.08 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3458, pruned_loss=0.09431, over 5771352.31 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3411, pruned_loss=0.09495, over 5641435.78 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:15:58,077 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=652469.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:16:01,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=652472.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:16:25,568 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:16:34,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.361e+02 1.467e+03 1.989e+03 2.885e+03 5.979e+03, threshold=3.978e+03, percent-clipped=20.0
+2023-03-07 18:16:34,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652501.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:16:54,885 INFO [train.py:968] (0/2) Epoch 15, batch 13600, giga_loss[loss=0.2681, simple_loss=0.3495, pruned_loss=0.09338, over 28498.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3415, pruned_loss=0.09378, over 5659729.84 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3454, pruned_loss=0.09421, over 5770788.50 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3434, pruned_loss=0.09491, over 5643004.69 frames. ], batch size: 336, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:17:35,134 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=652552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:17:37,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=652555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:17:52,869 INFO [train.py:968] (0/2) Epoch 15, batch 13650, giga_loss[loss=0.2531, simple_loss=0.3323, pruned_loss=0.08694, over 29020.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3421, pruned_loss=0.09412, over 5669694.10 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3456, pruned_loss=0.09433, over 5771152.27 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3434, pruned_loss=0.09488, over 5654088.05 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:18:13,356 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:18:14,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:18:19,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3335, 1.5168, 1.4323, 1.5931], device='cuda:0'), covar=tensor([0.0675, 0.0280, 0.0306, 0.0687], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:0')
+2023-03-07 18:18:33,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.374e+02 1.367e+03 1.740e+03 2.815e+03 6.816e+03, threshold=3.480e+03, percent-clipped=8.0
+2023-03-07 18:18:34,782 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-07 18:18:52,233 INFO [train.py:968] (0/2) Epoch 15, batch 13700, libri_loss[loss=0.2119, simple_loss=0.2868, pruned_loss=0.0685, over 29651.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3393, pruned_loss=0.09253, over 5672163.81 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3447, pruned_loss=0.0939, over 5774201.59 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3411, pruned_loss=0.09351, over 5654562.81 frames. ], batch size: 69, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:19:48,675 INFO [train.py:968] (0/2) Epoch 15, batch 13750, libri_loss[loss=0.3008, simple_loss=0.3678, pruned_loss=0.1169, over 27660.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3377, pruned_loss=0.09081, over 5666402.04 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3443, pruned_loss=0.09379, over 5770424.91 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3392, pruned_loss=0.09163, over 5652294.57 frames. ], batch size: 115, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:20:03,853 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:20:05,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2295, 1.1847, 3.5279, 3.0385], device='cuda:0'), covar=tensor([0.1567, 0.2807, 0.0435, 0.1201], device='cuda:0'), in_proj_covar=tensor([0.0697, 0.0610, 0.0893, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 18:20:25,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.654e+02 1.231e+03 1.639e+03 2.393e+03 5.815e+03, threshold=3.278e+03, percent-clipped=6.0
+2023-03-07 18:20:45,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:20:48,375 INFO [train.py:968] (0/2) Epoch 15, batch 13800, giga_loss[loss=0.2532, simple_loss=0.3331, pruned_loss=0.08667, over 28831.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3368, pruned_loss=0.08929, over 5670133.89 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3443, pruned_loss=0.09376, over 5771726.24 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.338, pruned_loss=0.08992, over 5656994.79 frames. ], batch size: 174, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:20:59,943 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:21:00,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=652728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:21:02,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2559, 0.9200, 0.9749, 1.3674], device='cuda:0'), covar=tensor([0.0768, 0.0332, 0.0349, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:0')
+2023-03-07 18:21:03,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=652731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:21:38,844 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:21:39,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5326, 1.9406, 1.4878, 1.6700], device='cuda:0'), covar=tensor([0.2694, 0.2389, 0.2832, 0.2125], device='cuda:0'), in_proj_covar=tensor([0.1385, 0.1014, 0.1232, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 18:21:48,329 INFO [train.py:968] (0/2) Epoch 15, batch 13850, giga_loss[loss=0.2112, simple_loss=0.2925, pruned_loss=0.06499, over 28770.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3346, pruned_loss=0.08937, over 5673694.38 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3442, pruned_loss=0.09379, over 5774617.73 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3355, pruned_loss=0.08978, over 5658922.42 frames. ], batch size: 119, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:22:24,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.451e+02 1.277e+03 1.768e+03 2.410e+03 7.618e+03, threshold=3.536e+03, percent-clipped=13.0
+2023-03-07 18:22:32,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-07 18:22:46,462 INFO [train.py:968] (0/2) Epoch 15, batch 13900, giga_loss[loss=0.2849, simple_loss=0.3483, pruned_loss=0.1108, over 28953.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3331, pruned_loss=0.08915, over 5673752.92 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3437, pruned_loss=0.09358, over 5776383.77 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3341, pruned_loss=0.08961, over 5659108.89 frames. ], batch size: 213, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:23:32,824 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=652858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:23:36,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=652861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:23:42,579 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:23:43,726 INFO [train.py:968] (0/2) Epoch 15, batch 13950, giga_loss[loss=0.2686, simple_loss=0.3495, pruned_loss=0.09384, over 28480.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3334, pruned_loss=0.08919, over 5671283.94 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3434, pruned_loss=0.09349, over 5778169.52 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3344, pruned_loss=0.08958, over 5656623.71 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:24:06,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3964, 1.8703, 1.6803, 1.6340], device='cuda:0'), covar=tensor([0.1776, 0.1805, 0.1955, 0.1829], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0722, 0.0677, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 18:24:11,980 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:24:26,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.984e+02 1.318e+03 1.860e+03 2.971e+03 9.596e+03, threshold=3.719e+03, percent-clipped=16.0
+2023-03-07 18:24:45,986 INFO [train.py:968] (0/2) Epoch 15, batch 14000, libri_loss[loss=0.2313, simple_loss=0.3088, pruned_loss=0.07695, over 29532.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3355, pruned_loss=0.08965, over 5661108.13 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.343, pruned_loss=0.09327, over 5777467.73 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3365, pruned_loss=0.09011, over 5647872.57 frames. ], batch size: 79, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 18:25:45,756 INFO [train.py:968] (0/2) Epoch 15, batch 14050, giga_loss[loss=0.228, simple_loss=0.3092, pruned_loss=0.07337, over 29221.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3341, pruned_loss=0.08859, over 5670373.19 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3432, pruned_loss=0.0936, over 5779227.07 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3345, pruned_loss=0.08854, over 5655215.43 frames. ], batch size: 113, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:26:34,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.277e+02 1.414e+03 1.819e+03 2.412e+03 5.953e+03, threshold=3.637e+03, percent-clipped=9.0
+2023-03-07 18:26:43,092 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=653010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:26:45,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=653013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:26:51,759 INFO [train.py:968] (0/2) Epoch 15, batch 14100, giga_loss[loss=0.2648, simple_loss=0.3436, pruned_loss=0.09294, over 28856.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3333, pruned_loss=0.08846, over 5680553.80 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3428, pruned_loss=0.09345, over 5781721.31 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3339, pruned_loss=0.08846, over 5664117.19 frames. ], batch size: 164, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:26:52,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2807, 1.5697, 1.2962, 0.9973], device='cuda:0'), covar=tensor([0.2702, 0.2599, 0.2980, 0.2290], device='cuda:0'), in_proj_covar=tensor([0.1386, 0.1017, 0.1236, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 18:27:16,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=653042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:27:33,627 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-07 18:27:36,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=653057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:27:54,887 INFO [train.py:968] (0/2) Epoch 15, batch 14150, giga_loss[loss=0.2721, simple_loss=0.349, pruned_loss=0.09764, over 28860.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.335, pruned_loss=0.08916, over 5689055.78 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3418, pruned_loss=0.09296, over 5784886.43 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.336, pruned_loss=0.08945, over 5670090.81 frames. ], batch size: 106, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:28:29,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=653096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:28:37,016 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.715e+02 1.378e+03 1.876e+03 2.391e+03 8.000e+03, threshold=3.751e+03, percent-clipped=10.0
+2023-03-07 18:28:39,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=653103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:28:58,722 INFO [train.py:968] (0/2) Epoch 15, batch 14200, libri_loss[loss=0.2426, simple_loss=0.3085, pruned_loss=0.08834, over 29393.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.339, pruned_loss=0.08914, over 5671836.14 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3417, pruned_loss=0.09308, over 5775377.93 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3399, pruned_loss=0.08919, over 5663258.82 frames. ], batch size: 67, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:29:12,559 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-07 18:29:53,884 INFO [train.py:968] (0/2) Epoch 15, batch 14250, giga_loss[loss=0.2803, simple_loss=0.3727, pruned_loss=0.09392, over 28597.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3407, pruned_loss=0.08826, over 5679646.79 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3417, pruned_loss=0.09304, over 5777270.68 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3413, pruned_loss=0.0882, over 5667959.38 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:30:32,520 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=653200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:30:33,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.844e+02 1.271e+03 1.550e+03 2.245e+03 6.736e+03, threshold=3.100e+03, percent-clipped=4.0
+2023-03-07 18:30:34,829 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=653203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:30:52,290 INFO [train.py:968] (0/2) Epoch 15, batch 14300, giga_loss[loss=0.2183, simple_loss=0.3118, pruned_loss=0.06244, over 28948.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3407, pruned_loss=0.0873, over 5679583.69 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3419, pruned_loss=0.09317, over 5780323.21 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3409, pruned_loss=0.08701, over 5665542.50 frames. ], batch size: 145, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:31:08,000 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=653232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:31:23,592 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=653244.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:31:24,962 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=653246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:31:28,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=653249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:31:50,259 INFO [train.py:968] (0/2) Epoch 15, batch 14350, giga_loss[loss=0.269, simple_loss=0.3465, pruned_loss=0.09576, over 29114.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3407, pruned_loss=0.0881, over 5680522.91 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3416, pruned_loss=0.09309, over 5782969.39 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3412, pruned_loss=0.08774, over 5663379.03 frames. ], batch size: 214, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:31:59,703 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=653278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:32:13,156 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-07 18:32:28,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.528e+02 1.284e+03 1.797e+03 2.447e+03 6.997e+03, threshold=3.595e+03, percent-clipped=12.0
+2023-03-07 18:32:47,431 INFO [train.py:968] (0/2) Epoch 15, batch 14400, giga_loss[loss=0.2568, simple_loss=0.3376, pruned_loss=0.08804, over 28236.00 frames. ], tot_loss[loss=0.259, simple_loss=0.34, pruned_loss=0.08896, over 5688631.99 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3415, pruned_loss=0.09325, over 5787016.93 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3404, pruned_loss=0.08839, over 5668249.44 frames. ], batch size: 412, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 18:33:05,105 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 18:33:53,870 INFO [train.py:968] (0/2) Epoch 15, batch 14450, libri_loss[loss=0.2401, simple_loss=0.3137, pruned_loss=0.08328, over 29572.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3402, pruned_loss=0.09001, over 5700212.71 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3404, pruned_loss=0.0927, over 5789400.86 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3416, pruned_loss=0.08996, over 5678552.00 frames. ], batch size: 78, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:34:43,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.096e+02 1.424e+03 1.809e+03 2.282e+03 5.034e+03, threshold=3.618e+03, percent-clipped=9.0
+2023-03-07 18:35:16,775 INFO [train.py:968] (0/2) Epoch 15, batch 14500, giga_loss[loss=0.2099, simple_loss=0.2964, pruned_loss=0.06173, over 28911.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3387, pruned_loss=0.09012, over 5689796.52 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3405, pruned_loss=0.09277, over 5787749.74 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3397, pruned_loss=0.08998, over 5673000.07 frames. ], batch size: 213, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:36:24,166 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-07 18:36:25,977 INFO [train.py:968] (0/2) Epoch 15, batch 14550, libri_loss[loss=0.2202, simple_loss=0.2936, pruned_loss=0.07338, over 29322.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3355, pruned_loss=0.08792, over 5689530.58 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.34, pruned_loss=0.09254, over 5789831.50 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3367, pruned_loss=0.08794, over 5672399.56 frames. ], batch size: 67, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:36:28,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=653471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:37:12,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.155e+02 1.206e+03 1.748e+03 2.562e+03 1.264e+04, threshold=3.496e+03, percent-clipped=12.0
+2023-03-07 18:37:31,334 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-07 18:37:32,977 INFO [train.py:968] (0/2) Epoch 15, batch 14600, giga_loss[loss=0.258, simple_loss=0.3265, pruned_loss=0.09479, over 29030.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3328, pruned_loss=0.08674, over 5684799.59 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3397, pruned_loss=0.09242, over 5790291.34 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3339, pruned_loss=0.08683, over 5670647.31 frames. ], batch size: 120, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:38:09,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3649, 1.8982, 1.3524, 0.5629], device='cuda:0'), covar=tensor([0.4775, 0.2550, 0.3398, 0.5168], device='cuda:0'), in_proj_covar=tensor([0.1626, 0.1550, 0.1531, 0.1334], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 18:38:22,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2494, 1.4247, 3.2291, 2.8915], device='cuda:0'), covar=tensor([0.1241, 0.2089, 0.0493, 0.1128], device='cuda:0'), in_proj_covar=tensor([0.0690, 0.0605, 0.0886, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 18:38:31,613 INFO [train.py:968] (0/2) Epoch 15, batch 14650, giga_loss[loss=0.2509, simple_loss=0.3358, pruned_loss=0.08301, over 27643.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3359, pruned_loss=0.0887, over 5675582.02 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3401, pruned_loss=0.09269, over 5780602.03 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3363, pruned_loss=0.08836, over 5669619.82 frames. ], batch size: 472, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:39:13,858 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.296e+02 1.344e+03 1.778e+03 2.322e+03 8.342e+03, threshold=3.556e+03, percent-clipped=6.0
+2023-03-07 18:39:27,809 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=653614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:39:30,794 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=653617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:39:32,288 INFO [train.py:968] (0/2) Epoch 15, batch 14700, giga_loss[loss=0.2745, simple_loss=0.3478, pruned_loss=0.1007, over 28972.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3388, pruned_loss=0.09036, over 5674505.59 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3399, pruned_loss=0.0927, over 5774453.13 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3392, pruned_loss=0.09, over 5672893.76 frames. ], batch size: 199, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:39:32,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=653619.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:40:05,784 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=653646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:40:35,013 INFO [train.py:968] (0/2) Epoch 15, batch 14750, giga_loss[loss=0.226, simple_loss=0.2993, pruned_loss=0.07629, over 28612.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3368, pruned_loss=0.0905, over 5677465.08 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3397, pruned_loss=0.09271, over 5776575.72 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3372, pruned_loss=0.09017, over 5672553.95 frames. ], batch size: 85, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:41:16,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.563e+02 1.298e+03 1.703e+03 2.287e+03 4.516e+03, threshold=3.406e+03, percent-clipped=4.0
+2023-03-07 18:41:32,981 INFO [train.py:968] (0/2) Epoch 15, batch 14800, giga_loss[loss=0.2516, simple_loss=0.3307, pruned_loss=0.08627, over 28904.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3362, pruned_loss=0.09065, over 5685135.06 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.339, pruned_loss=0.09232, over 5781156.22 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3371, pruned_loss=0.09066, over 5674086.87 frames. ], batch size: 213, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 18:42:07,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0379, 1.1420, 3.3637, 3.0057], device='cuda:0'), covar=tensor([0.1678, 0.2635, 0.0481, 0.0808], device='cuda:0'), in_proj_covar=tensor([0.0695, 0.0608, 0.0889, 0.0804], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 18:42:27,371 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=653762.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:42:32,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=653765.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:42:37,250 INFO [train.py:968] (0/2) Epoch 15, batch 14850, giga_loss[loss=0.2904, simple_loss=0.3693, pruned_loss=0.1058, over 28445.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3382, pruned_loss=0.09196, over 5685386.50 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3389, pruned_loss=0.09231, over 5781837.44 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.339, pruned_loss=0.09198, over 5675712.88 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:42:43,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6684, 1.7276, 1.4614, 1.8023], device='cuda:0'), covar=tensor([0.2618, 0.2680, 0.2976, 0.2565], device='cuda:0'), in_proj_covar=tensor([0.1367, 0.1000, 0.1219, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 18:43:11,558 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=653794.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:43:27,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.944e+02 1.464e+03 1.845e+03 2.544e+03 6.441e+03, threshold=3.691e+03, percent-clipped=11.0
+2023-03-07 18:43:46,638 INFO [train.py:968] (0/2) Epoch 15, batch 14900, giga_loss[loss=0.2991, simple_loss=0.3695, pruned_loss=0.1144, over 28490.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3402, pruned_loss=0.09211, over 5677758.81 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3389, pruned_loss=0.09231, over 5775237.74 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.341, pruned_loss=0.09213, over 5673554.26 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:45:01,818 INFO [train.py:968] (0/2) Epoch 15, batch 14950, giga_loss[loss=0.2262, simple_loss=0.3137, pruned_loss=0.06933, over 28955.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.34, pruned_loss=0.09159, over 5674702.33 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3389, pruned_loss=0.09238, over 5778620.43 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3405, pruned_loss=0.09152, over 5665583.99 frames. ], batch size: 145, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:45:20,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8830, 2.4422, 2.1211, 1.7587], device='cuda:0'), covar=tensor([0.2482, 0.1471, 0.1665, 0.1961], device='cuda:0'), in_proj_covar=tensor([0.1784, 0.1699, 0.1625, 0.1752], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 18:45:59,902 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.589e+02 1.415e+03 1.889e+03 2.709e+03 8.345e+03, threshold=3.777e+03, percent-clipped=12.0
+2023-03-07 18:46:18,596 INFO [train.py:968] (0/2) Epoch 15, batch 15000, giga_loss[loss=0.2396, simple_loss=0.3195, pruned_loss=0.07984, over 28465.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3369, pruned_loss=0.09114, over 5664196.39 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3385, pruned_loss=0.09221, over 5772573.50 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3378, pruned_loss=0.09119, over 5660461.59 frames. ], batch size: 369, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:46:18,600 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 18:46:28,732 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2021, simple_loss=0.302, pruned_loss=0.0511, over 944034.00 frames. 
+2023-03-07 18:46:28,732 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 18:46:40,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2323, 0.8236, 0.8826, 1.4787], device='cuda:0'), covar=tensor([0.0769, 0.0374, 0.0359, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:0')
+2023-03-07 18:47:07,309 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=653947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:47:38,550 INFO [train.py:968] (0/2) Epoch 15, batch 15050, giga_loss[loss=0.2111, simple_loss=0.2916, pruned_loss=0.06532, over 28714.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3303, pruned_loss=0.08793, over 5663860.56 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3382, pruned_loss=0.09195, over 5772791.31 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.331, pruned_loss=0.08814, over 5657613.57 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:48:14,934 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-654000.pt
+2023-03-07 18:48:18,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.489e+02 1.327e+03 1.869e+03 2.801e+03 9.116e+03, threshold=3.738e+03, percent-clipped=11.0
+2023-03-07 18:48:35,967 INFO [train.py:968] (0/2) Epoch 15, batch 15100, giga_loss[loss=0.275, simple_loss=0.3505, pruned_loss=0.0997, over 28666.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3283, pruned_loss=0.08695, over 5665009.02 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3374, pruned_loss=0.09166, over 5765935.91 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3293, pruned_loss=0.08724, over 5663054.71 frames. ], batch size: 242, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:48:47,188 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=654027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:48:54,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2730, 1.3926, 1.2335, 1.2586], device='cuda:0'), covar=tensor([0.1599, 0.1340, 0.1208, 0.1300], device='cuda:0'), in_proj_covar=tensor([0.1778, 0.1686, 0.1617, 0.1747], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 18:49:36,356 INFO [train.py:968] (0/2) Epoch 15, batch 15150, giga_loss[loss=0.2701, simple_loss=0.342, pruned_loss=0.09904, over 28879.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3301, pruned_loss=0.08835, over 5659839.17 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3372, pruned_loss=0.09152, over 5768555.63 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.331, pruned_loss=0.08864, over 5654025.52 frames. ], batch size: 112, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:50:16,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.844e+02 1.330e+03 1.893e+03 2.978e+03 9.135e+03, threshold=3.786e+03, percent-clipped=17.0
+2023-03-07 18:50:37,450 INFO [train.py:968] (0/2) Epoch 15, batch 15200, giga_loss[loss=0.2937, simple_loss=0.3573, pruned_loss=0.1151, over 28485.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3292, pruned_loss=0.08748, over 5665065.39 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.337, pruned_loss=0.09148, over 5768215.34 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3299, pruned_loss=0.08768, over 5658358.96 frames. ], batch size: 369, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:51:37,999 INFO [train.py:968] (0/2) Epoch 15, batch 15250, giga_loss[loss=0.2161, simple_loss=0.3059, pruned_loss=0.06316, over 28849.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3285, pruned_loss=0.08654, over 5658360.56 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3374, pruned_loss=0.09182, over 5768369.94 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3285, pruned_loss=0.08621, over 5649757.06 frames. ], batch size: 145, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:52:27,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.988e+02 1.377e+03 1.832e+03 2.614e+03 7.057e+03, threshold=3.665e+03, percent-clipped=8.0
+2023-03-07 18:52:47,406 INFO [train.py:968] (0/2) Epoch 15, batch 15300, giga_loss[loss=0.2228, simple_loss=0.2864, pruned_loss=0.07954, over 24392.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3271, pruned_loss=0.08574, over 5662172.69 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3375, pruned_loss=0.09183, over 5766648.37 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3269, pruned_loss=0.08541, over 5655642.47 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:54:00,460 INFO [train.py:968] (0/2) Epoch 15, batch 15350, giga_loss[loss=0.2171, simple_loss=0.2997, pruned_loss=0.06719, over 29000.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3263, pruned_loss=0.08506, over 5657436.88 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3372, pruned_loss=0.09164, over 5767437.13 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3263, pruned_loss=0.08495, over 5651011.44 frames. ], batch size: 93, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:54:47,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2288, 1.1993, 3.8306, 3.2049], device='cuda:0'), covar=tensor([0.1608, 0.2741, 0.0429, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0693, 0.0608, 0.0885, 0.0805], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 18:54:49,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.408e+02 1.371e+03 1.718e+03 2.203e+03 1.345e+04, threshold=3.436e+03, percent-clipped=7.0
+2023-03-07 18:54:51,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=654307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:55:06,155 INFO [train.py:968] (0/2) Epoch 15, batch 15400, giga_loss[loss=0.256, simple_loss=0.3326, pruned_loss=0.08974, over 28659.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3271, pruned_loss=0.08546, over 5657019.56 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3369, pruned_loss=0.09156, over 5766583.68 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3273, pruned_loss=0.08534, over 5650952.32 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:55:12,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=654322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:55:35,838 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=654339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:56:05,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4065, 1.6222, 1.6702, 1.2671], device='cuda:0'), covar=tensor([0.1565, 0.2326, 0.1314, 0.1583], device='cuda:0'), in_proj_covar=tensor([0.0855, 0.0683, 0.0900, 0.0802], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 18:56:11,919 INFO [train.py:968] (0/2) Epoch 15, batch 15450, libri_loss[loss=0.2334, simple_loss=0.3056, pruned_loss=0.08059, over 29521.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3274, pruned_loss=0.08595, over 5665899.57 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3364, pruned_loss=0.09121, over 5768819.81 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3278, pruned_loss=0.08604, over 5656489.57 frames. ], batch size: 79, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:56:57,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=654402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:57:05,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.575e+02 1.312e+03 1.774e+03 2.362e+03 5.533e+03, threshold=3.548e+03, percent-clipped=7.0
+2023-03-07 18:57:05,627 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=654406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:57:22,374 INFO [train.py:968] (0/2) Epoch 15, batch 15500, giga_loss[loss=0.2657, simple_loss=0.3524, pruned_loss=0.08949, over 29108.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3258, pruned_loss=0.08484, over 5661451.70 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3362, pruned_loss=0.0911, over 5770206.19 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3261, pruned_loss=0.08496, over 5651824.21 frames. ], batch size: 200, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:58:09,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3482, 1.5952, 1.5245, 1.2653], device='cuda:0'), covar=tensor([0.2242, 0.1932, 0.1459, 0.1842], device='cuda:0'), in_proj_covar=tensor([0.1781, 0.1683, 0.1616, 0.1747], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 18:58:17,555 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=654465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:58:22,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=654468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:58:22,431 INFO [train.py:968] (0/2) Epoch 15, batch 15550, giga_loss[loss=0.2405, simple_loss=0.3275, pruned_loss=0.07677, over 28111.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3282, pruned_loss=0.08462, over 5673306.58 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3359, pruned_loss=0.09095, over 5773074.94 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3286, pruned_loss=0.08473, over 5661253.32 frames. ], batch size: 412, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:58:58,684 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=654497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:59:11,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.040e+02 1.317e+03 1.635e+03 2.598e+03 9.096e+03, threshold=3.271e+03, percent-clipped=16.0
+2023-03-07 18:59:24,017 INFO [train.py:968] (0/2) Epoch 15, batch 15600, giga_loss[loss=0.2743, simple_loss=0.367, pruned_loss=0.09082, over 28691.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3306, pruned_loss=0.08532, over 5669545.21 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3359, pruned_loss=0.09085, over 5775862.51 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3308, pruned_loss=0.08533, over 5654239.42 frames. ], batch size: 242, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:59:56,010 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=654545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:00:00,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=654548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:00:24,414 INFO [train.py:968] (0/2) Epoch 15, batch 15650, giga_loss[loss=0.2647, simple_loss=0.3406, pruned_loss=0.09433, over 27564.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3332, pruned_loss=0.08682, over 5671395.07 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3356, pruned_loss=0.0907, over 5777942.61 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3336, pruned_loss=0.08691, over 5656118.06 frames. ], batch size: 472, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:00:35,534 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=654577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:01:09,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.784e+02 1.445e+03 1.976e+03 2.625e+03 1.450e+04, threshold=3.951e+03, percent-clipped=14.0
+2023-03-07 19:01:27,689 INFO [train.py:968] (0/2) Epoch 15, batch 15700, giga_loss[loss=0.2774, simple_loss=0.3527, pruned_loss=0.1011, over 28454.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3327, pruned_loss=0.08642, over 5684349.64 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3356, pruned_loss=0.09076, over 5779674.37 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3329, pruned_loss=0.08637, over 5669269.53 frames. ], batch size: 369, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:02:26,431 INFO [train.py:968] (0/2) Epoch 15, batch 15750, giga_loss[loss=0.2969, simple_loss=0.3805, pruned_loss=0.1066, over 28684.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3306, pruned_loss=0.08476, over 5694098.05 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3354, pruned_loss=0.09059, over 5781853.28 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3308, pruned_loss=0.08473, over 5677350.03 frames. ], batch size: 242, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:02:42,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=654682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:03:14,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.352e+02 1.330e+03 1.740e+03 2.397e+03 1.174e+04, threshold=3.481e+03, percent-clipped=6.0
+2023-03-07 19:03:26,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=654714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:03:30,821 INFO [train.py:968] (0/2) Epoch 15, batch 15800, giga_loss[loss=0.2467, simple_loss=0.3302, pruned_loss=0.08163, over 28925.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3296, pruned_loss=0.08438, over 5684663.95 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3354, pruned_loss=0.09067, over 5774377.72 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3296, pruned_loss=0.08416, over 5676780.16 frames. ], batch size: 145, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:04:28,598 INFO [train.py:968] (0/2) Epoch 15, batch 15850, giga_loss[loss=0.2295, simple_loss=0.3153, pruned_loss=0.07187, over 28158.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3272, pruned_loss=0.08399, over 5682169.16 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3351, pruned_loss=0.09057, over 5772426.10 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3273, pruned_loss=0.08373, over 5674845.04 frames. ], batch size: 412, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:04:43,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=654781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:05:13,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.689e+02 1.301e+03 1.694e+03 2.191e+03 4.778e+03, threshold=3.387e+03, percent-clipped=2.0
+2023-03-07 19:05:30,257 INFO [train.py:968] (0/2) Epoch 15, batch 15900, giga_loss[loss=0.2505, simple_loss=0.336, pruned_loss=0.08248, over 28685.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3282, pruned_loss=0.08441, over 5678229.79 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3347, pruned_loss=0.09024, over 5774189.91 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3285, pruned_loss=0.08434, over 5668907.15 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:05:40,254 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=654825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:05:43,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=654828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:06:20,586 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=654857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:06:20,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=654857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:06:25,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=654860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:06:37,053 INFO [train.py:968] (0/2) Epoch 15, batch 15950, giga_loss[loss=0.2599, simple_loss=0.341, pruned_loss=0.08942, over 29022.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3304, pruned_loss=0.08577, over 5686272.27 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3339, pruned_loss=0.08976, over 5777315.17 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3312, pruned_loss=0.08606, over 5673808.96 frames. ], batch size: 285, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:07:03,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=654889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:07:24,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.493e+02 1.285e+03 1.692e+03 2.500e+03 8.685e+03, threshold=3.383e+03, percent-clipped=12.0
+2023-03-07 19:07:39,527 INFO [train.py:968] (0/2) Epoch 15, batch 16000, libri_loss[loss=0.1956, simple_loss=0.2756, pruned_loss=0.05779, over 29378.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3311, pruned_loss=0.08668, over 5678941.01 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3336, pruned_loss=0.08962, over 5776173.80 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3321, pruned_loss=0.08696, over 5666318.93 frames. ], batch size: 67, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:07:47,346 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=654924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:07:49,563 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=654927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:08:20,615 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=654956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:08:36,396 INFO [train.py:968] (0/2) Epoch 15, batch 16050, giga_loss[loss=0.2845, simple_loss=0.3414, pruned_loss=0.1138, over 24401.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3343, pruned_loss=0.08836, over 5681544.57 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3332, pruned_loss=0.08945, over 5777270.65 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3354, pruned_loss=0.08869, over 5668326.63 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:09:18,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.613e+02 1.324e+03 1.649e+03 2.426e+03 6.372e+03, threshold=3.298e+03, percent-clipped=9.0
+2023-03-07 19:09:21,851 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-07 19:09:30,540 INFO [train.py:968] (0/2) Epoch 15, batch 16100, giga_loss[loss=0.2511, simple_loss=0.3385, pruned_loss=0.0819, over 29021.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3369, pruned_loss=0.08895, over 5684136.55 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3331, pruned_loss=0.08945, over 5769330.83 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.338, pruned_loss=0.0892, over 5677147.30 frames. ], batch size: 136, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:10:12,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3014, 1.8709, 1.4342, 1.4704], device='cuda:0'), covar=tensor([0.0758, 0.0317, 0.0330, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0063, 0.0057, 0.0098], device='cuda:0')
+2023-03-07 19:10:33,813 INFO [train.py:968] (0/2) Epoch 15, batch 16150, giga_loss[loss=0.2389, simple_loss=0.3277, pruned_loss=0.07506, over 28413.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3381, pruned_loss=0.08962, over 5675673.84 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3333, pruned_loss=0.08955, over 5761251.26 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3389, pruned_loss=0.08973, over 5675810.23 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:10:53,492 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:11:29,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.033e+02 1.372e+03 1.860e+03 2.392e+03 9.099e+03, threshold=3.720e+03, percent-clipped=11.0
+2023-03-07 19:11:45,936 INFO [train.py:968] (0/2) Epoch 15, batch 16200, giga_loss[loss=0.2374, simple_loss=0.3183, pruned_loss=0.07825, over 28496.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3355, pruned_loss=0.08803, over 5685802.25 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3327, pruned_loss=0.08917, over 5764784.29 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3367, pruned_loss=0.08846, over 5680861.63 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:12:19,367 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6398, 4.1669, 1.6708, 1.7863], device='cuda:0'), covar=tensor([0.0886, 0.0326, 0.0897, 0.1234], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0521, 0.0357, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 19:12:35,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9072, 3.7333, 3.4852, 1.9945], device='cuda:0'), covar=tensor([0.0635, 0.0859, 0.0904, 0.2367], device='cuda:0'), in_proj_covar=tensor([0.1101, 0.1009, 0.0873, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 19:12:49,202 INFO [train.py:968] (0/2) Epoch 15, batch 16250, giga_loss[loss=0.2311, simple_loss=0.3176, pruned_loss=0.07229, over 28943.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3352, pruned_loss=0.08845, over 5677708.06 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.333, pruned_loss=0.08932, over 5758264.43 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.336, pruned_loss=0.08863, over 5678628.29 frames. ], batch size: 186, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:13:16,837 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655190.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:13:35,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.637e+02 1.414e+03 1.758e+03 2.791e+03 1.181e+04, threshold=3.515e+03, percent-clipped=13.0
+2023-03-07 19:13:51,551 INFO [train.py:968] (0/2) Epoch 15, batch 16300, libri_loss[loss=0.2211, simple_loss=0.2992, pruned_loss=0.07152, over 29475.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.334, pruned_loss=0.08833, over 5670722.11 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3325, pruned_loss=0.08923, over 5762638.86 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3352, pruned_loss=0.08853, over 5664703.01 frames. ], batch size: 70, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:14:05,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1805, 1.4419, 1.3542, 1.0939], device='cuda:0'), covar=tensor([0.2359, 0.2094, 0.1451, 0.1947], device='cuda:0'), in_proj_covar=tensor([0.1788, 0.1687, 0.1609, 0.1751], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 19:14:54,307 INFO [train.py:968] (0/2) Epoch 15, batch 16350, giga_loss[loss=0.2419, simple_loss=0.3185, pruned_loss=0.08265, over 28681.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3327, pruned_loss=0.08896, over 5671165.50 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3323, pruned_loss=0.08916, over 5763171.03 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3339, pruned_loss=0.08919, over 5664602.24 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:15:38,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.998e+02 1.314e+03 1.691e+03 2.714e+03 7.919e+03, threshold=3.382e+03, percent-clipped=10.0
+2023-03-07 19:15:44,014 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655311.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:15:53,216 INFO [train.py:968] (0/2) Epoch 15, batch 16400, giga_loss[loss=0.2606, simple_loss=0.3439, pruned_loss=0.08861, over 28887.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3326, pruned_loss=0.08893, over 5678395.31 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3322, pruned_loss=0.08911, over 5765538.72 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3336, pruned_loss=0.08916, over 5669545.80 frames. ], batch size: 284, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:16:46,173 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:16:54,024 INFO [train.py:968] (0/2) Epoch 15, batch 16450, giga_loss[loss=0.2041, simple_loss=0.2758, pruned_loss=0.06623, over 24482.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3319, pruned_loss=0.08745, over 5674194.77 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.332, pruned_loss=0.08901, over 5767280.00 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3328, pruned_loss=0.0877, over 5663508.20 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:17:19,848 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-07 19:17:39,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.245e+02 1.268e+03 1.620e+03 2.191e+03 4.805e+03, threshold=3.239e+03, percent-clipped=5.0
+2023-03-07 19:17:40,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1299, 1.3387, 1.2613, 1.0981], device='cuda:0'), covar=tensor([0.1997, 0.1868, 0.1282, 0.1640], device='cuda:0'), in_proj_covar=tensor([0.1797, 0.1688, 0.1611, 0.1754], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 19:17:51,536 INFO [train.py:968] (0/2) Epoch 15, batch 16500, libri_loss[loss=0.2167, simple_loss=0.2942, pruned_loss=0.06958, over 29388.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3323, pruned_loss=0.08644, over 5683415.32 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3317, pruned_loss=0.08889, over 5770556.90 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3334, pruned_loss=0.08671, over 5669203.42 frames. ], batch size: 67, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:18:32,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:18:36,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1976, 1.4448, 1.5236, 1.3818], device='cuda:0'), covar=tensor([0.1295, 0.0965, 0.1379, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0713, 0.0669, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 19:18:50,501 INFO [train.py:968] (0/2) Epoch 15, batch 16550, giga_loss[loss=0.2221, simple_loss=0.3128, pruned_loss=0.06567, over 28936.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3336, pruned_loss=0.08546, over 5669715.11 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3315, pruned_loss=0.08877, over 5763413.05 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3347, pruned_loss=0.08572, over 5663769.34 frames. ], batch size: 93, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:19:10,292 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2305, 5.0661, 4.8131, 2.5176], device='cuda:0'), covar=tensor([0.0387, 0.0565, 0.0614, 0.1668], device='cuda:0'), in_proj_covar=tensor([0.1101, 0.1015, 0.0876, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 19:19:26,663 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:19:30,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.347e+02 1.371e+03 1.988e+03 3.048e+03 1.928e+04, threshold=3.976e+03, percent-clipped=21.0
+2023-03-07 19:19:41,963 INFO [train.py:968] (0/2) Epoch 15, batch 16600, giga_loss[loss=0.2516, simple_loss=0.3314, pruned_loss=0.08596, over 29030.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3345, pruned_loss=0.08561, over 5666930.68 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3314, pruned_loss=0.08883, over 5744959.07 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3356, pruned_loss=0.08567, over 5675018.38 frames. ], batch size: 136, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:20:40,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:20:43,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655565.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:20:48,076 INFO [train.py:968] (0/2) Epoch 15, batch 16650, giga_loss[loss=0.2493, simple_loss=0.3353, pruned_loss=0.08171, over 28616.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3351, pruned_loss=0.0859, over 5668766.06 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3314, pruned_loss=0.08885, over 5737381.28 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3359, pruned_loss=0.0859, over 5680108.06 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:21:10,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4365, 1.6304, 1.3401, 1.5041], device='cuda:0'), covar=tensor([0.2625, 0.2620, 0.2975, 0.2185], device='cuda:0'), in_proj_covar=tensor([0.1369, 0.1000, 0.1222, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 19:21:28,072 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=655597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:21:30,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=655600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:21:30,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-07 19:21:38,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.167e+02 1.254e+03 1.570e+03 2.123e+03 4.424e+03, threshold=3.141e+03, percent-clipped=3.0
+2023-03-07 19:21:43,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9626, 1.1737, 1.1022, 0.9456], device='cuda:0'), covar=tensor([0.1964, 0.1964, 0.1132, 0.1539], device='cuda:0'), in_proj_covar=tensor([0.1786, 0.1679, 0.1606, 0.1744], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 19:21:53,915 INFO [train.py:968] (0/2) Epoch 15, batch 16700, giga_loss[loss=0.2275, simple_loss=0.3168, pruned_loss=0.06913, over 28882.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3348, pruned_loss=0.08559, over 5662658.84 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3316, pruned_loss=0.08894, over 5733040.68 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3353, pruned_loss=0.08541, over 5672787.31 frames. ], batch size: 145, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:22:04,827 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:22:11,496 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=655629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:22:25,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655639.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:23:04,858 INFO [train.py:968] (0/2) Epoch 15, batch 16750, giga_loss[loss=0.2517, simple_loss=0.3306, pruned_loss=0.08644, over 28051.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3343, pruned_loss=0.08497, over 5663464.06 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3315, pruned_loss=0.08884, over 5735530.49 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.335, pruned_loss=0.08486, over 5667804.76 frames. ], batch size: 412, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:23:26,984 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:23:27,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655686.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:23:58,076 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=655708.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:23:58,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.972e+02 1.363e+03 1.795e+03 2.431e+03 4.490e+03, threshold=3.590e+03, percent-clipped=9.0
+2023-03-07 19:24:01,695 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=655711.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:24:12,619 INFO [train.py:968] (0/2) Epoch 15, batch 16800, libri_loss[loss=0.244, simple_loss=0.3194, pruned_loss=0.08426, over 29533.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.334, pruned_loss=0.08432, over 5672001.13 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3315, pruned_loss=0.08877, over 5736547.64 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3347, pruned_loss=0.08416, over 5672345.50 frames. ], batch size: 81, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:24:38,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:24:42,325 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=655740.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:25:18,211 INFO [train.py:968] (0/2) Epoch 15, batch 16850, giga_loss[loss=0.2983, simple_loss=0.3708, pruned_loss=0.1129, over 28131.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.338, pruned_loss=0.08666, over 5670183.64 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3314, pruned_loss=0.08893, over 5731249.66 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3387, pruned_loss=0.08629, over 5673905.34 frames. ], batch size: 412, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:25:20,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3585, 1.7047, 1.4371, 1.4921], device='cuda:0'), covar=tensor([0.0708, 0.0396, 0.0332, 0.0765], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0063, 0.0057, 0.0098], device='cuda:0')
+2023-03-07 19:25:56,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3952, 1.5239, 1.3716, 1.5556], device='cuda:0'), covar=tensor([0.0731, 0.0354, 0.0331, 0.0808], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0113, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0063, 0.0057, 0.0098], device='cuda:0')
+2023-03-07 19:26:08,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:26:10,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.603e+02 1.647e+03 2.350e+03 3.526e+03 8.063e+03, threshold=4.701e+03, percent-clipped=23.0
+2023-03-07 19:26:27,085 INFO [train.py:968] (0/2) Epoch 15, batch 16900, giga_loss[loss=0.2835, simple_loss=0.3546, pruned_loss=0.1062, over 29060.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3377, pruned_loss=0.08623, over 5677290.86 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3314, pruned_loss=0.08887, over 5735481.19 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3384, pruned_loss=0.08593, over 5675105.61 frames. ], batch size: 199, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:26:41,336 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-07 19:26:42,235 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=655829.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:26:45,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=655832.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:27:28,116 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=655861.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:27:37,726 INFO [train.py:968] (0/2) Epoch 15, batch 16950, libri_loss[loss=0.2548, simple_loss=0.3389, pruned_loss=0.08539, over 29314.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3364, pruned_loss=0.08629, over 5677870.38 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3316, pruned_loss=0.08887, over 5727415.27 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3369, pruned_loss=0.08601, over 5681177.01 frames. ], batch size: 94, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:27:50,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:27:50,820 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=655880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:27:57,965 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=655883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:28:05,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0620, 3.1084, 2.1687, 0.8862], device='cuda:0'), covar=tensor([0.6337, 0.2681, 0.3383, 0.6522], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1540, 0.1527, 0.1334], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 19:28:35,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.929e+02 1.323e+03 1.757e+03 2.813e+03 8.070e+03, threshold=3.514e+03, percent-clipped=5.0
+2023-03-07 19:28:37,387 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=655912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:28:45,894 INFO [train.py:968] (0/2) Epoch 15, batch 17000, giga_loss[loss=0.2099, simple_loss=0.3117, pruned_loss=0.05401, over 28892.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3335, pruned_loss=0.08474, over 5679800.07 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3315, pruned_loss=0.08891, over 5722298.77 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3341, pruned_loss=0.08441, over 5684893.96 frames. ], batch size: 174, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:29:10,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:29:18,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1522, 1.6284, 1.1906, 0.4392], device='cuda:0'), covar=tensor([0.3303, 0.2199, 0.3323, 0.4829], device='cuda:0'), in_proj_covar=tensor([0.1607, 0.1535, 0.1520, 0.1328], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 19:29:54,806 INFO [train.py:968] (0/2) Epoch 15, batch 17050, giga_loss[loss=0.2589, simple_loss=0.3407, pruned_loss=0.08861, over 28783.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3315, pruned_loss=0.08279, over 5689456.43 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3308, pruned_loss=0.08847, over 5723366.66 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3326, pruned_loss=0.08282, over 5691550.85 frames. ], batch size: 243, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:30:06,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3347, 1.2595, 3.7107, 3.2915], device='cuda:0'), covar=tensor([0.1532, 0.2841, 0.0403, 0.1753], device='cuda:0'), in_proj_covar=tensor([0.0694, 0.0608, 0.0885, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 19:30:31,300 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-656000.pt
+2023-03-07 19:30:32,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=656000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:30:42,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.601e+02 1.165e+03 1.532e+03 2.120e+03 1.532e+04, threshold=3.063e+03, percent-clipped=9.0
+2023-03-07 19:30:49,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=656014.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:30:55,627 INFO [train.py:968] (0/2) Epoch 15, batch 17100, giga_loss[loss=0.2353, simple_loss=0.3096, pruned_loss=0.08055, over 24254.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3317, pruned_loss=0.08322, over 5684120.47 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3305, pruned_loss=0.08825, over 5727885.36 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3328, pruned_loss=0.08331, over 5681112.72 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:30:59,221 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:31:01,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:31:37,180 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:31:43,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=656061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:31:53,347 INFO [train.py:968] (0/2) Epoch 15, batch 17150, giga_loss[loss=0.3638, simple_loss=0.4211, pruned_loss=0.1532, over 27760.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3348, pruned_loss=0.08541, over 5690235.09 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3303, pruned_loss=0.0882, over 5732041.16 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.336, pruned_loss=0.0854, over 5683375.13 frames. ], batch size: 472, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:32:08,583 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:32:12,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:32:40,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.745e+02 1.490e+03 1.906e+03 2.598e+03 7.816e+03, threshold=3.812e+03, percent-clipped=19.0
+2023-03-07 19:32:43,261 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:32:47,561 INFO [train.py:968] (0/2) Epoch 15, batch 17200, libri_loss[loss=0.3062, simple_loss=0.3709, pruned_loss=0.1208, over 19490.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3363, pruned_loss=0.08685, over 5669811.02 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3305, pruned_loss=0.08836, over 5716629.41 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3372, pruned_loss=0.08662, over 5678282.80 frames. ], batch size: 187, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:33:17,811 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:33:20,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:33:32,349 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656157.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:33:36,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656160.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:33:45,696 INFO [train.py:968] (0/2) Epoch 15, batch 17250, giga_loss[loss=0.2595, simple_loss=0.3372, pruned_loss=0.09092, over 28475.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3341, pruned_loss=0.08647, over 5672167.44 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3304, pruned_loss=0.0883, over 5718383.21 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.335, pruned_loss=0.08634, over 5676893.98 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:33:54,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:34:00,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=656181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:34:12,810 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656189.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:34:28,937 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:34:31,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:34:34,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.553e+02 1.438e+03 1.942e+03 2.783e+03 7.109e+03, threshold=3.884e+03, percent-clipped=12.0
+2023-03-07 19:34:41,003 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-07 19:34:48,144 INFO [train.py:968] (0/2) Epoch 15, batch 17300, giga_loss[loss=0.254, simple_loss=0.3269, pruned_loss=0.09052, over 28979.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3346, pruned_loss=0.0877, over 5667695.81 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3303, pruned_loss=0.08826, over 5711077.27 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3354, pruned_loss=0.08762, over 5676588.25 frames. ], batch size: 136, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:35:04,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5126, 4.2900, 4.1062, 1.8741], device='cuda:0'), covar=tensor([0.0658, 0.0878, 0.0948, 0.2255], device='cuda:0'), in_proj_covar=tensor([0.1089, 0.0995, 0.0864, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 19:35:05,836 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:35:34,751 INFO [train.py:968] (0/2) Epoch 15, batch 17350, giga_loss[loss=0.2578, simple_loss=0.3389, pruned_loss=0.08829, over 28577.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3377, pruned_loss=0.08984, over 5677947.76 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3303, pruned_loss=0.08806, over 5713899.63 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3386, pruned_loss=0.08997, over 5680858.25 frames. ], batch size: 71, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:36:20,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.780e+02 1.381e+03 1.738e+03 2.694e+03 9.868e+03, threshold=3.475e+03, percent-clipped=6.0
+2023-03-07 19:36:24,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 19:36:26,488 INFO [train.py:968] (0/2) Epoch 15, batch 17400, giga_loss[loss=0.3053, simple_loss=0.3832, pruned_loss=0.1137, over 28942.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3474, pruned_loss=0.09592, over 5681079.78 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3309, pruned_loss=0.08846, over 5718576.01 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3479, pruned_loss=0.09578, over 5678320.60 frames. ], batch size: 136, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:36:31,671 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:36:35,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9346, 1.1732, 1.2015, 1.0590], device='cuda:0'), covar=tensor([0.1424, 0.1249, 0.1991, 0.1336], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0722, 0.0674, 0.0658], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 19:36:35,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:37:01,732 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:37:08,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3427, 3.0381, 1.5394, 1.5170], device='cuda:0'), covar=tensor([0.1015, 0.0322, 0.0887, 0.1334], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0523, 0.0358, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 19:37:12,675 INFO [train.py:968] (0/2) Epoch 15, batch 17450, giga_loss[loss=0.2817, simple_loss=0.367, pruned_loss=0.09819, over 28663.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3545, pruned_loss=0.09954, over 5689907.00 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3308, pruned_loss=0.08835, over 5720502.45 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3553, pruned_loss=0.09966, over 5685580.76 frames. ], batch size: 60, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:37:47,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.632e+02 1.168e+03 1.407e+03 1.876e+03 4.542e+03, threshold=2.813e+03, percent-clipped=3.0
+2023-03-07 19:37:56,793 INFO [train.py:968] (0/2) Epoch 15, batch 17500, giga_loss[loss=0.2608, simple_loss=0.3153, pruned_loss=0.1032, over 23696.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3523, pruned_loss=0.09935, over 5687310.30 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3311, pruned_loss=0.08851, over 5720638.58 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3532, pruned_loss=0.09954, over 5683413.32 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:38:13,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7607, 1.6596, 1.3885, 1.2945], device='cuda:0'), covar=tensor([0.0661, 0.0492, 0.0808, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0432, 0.0500, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 19:38:38,983 INFO [train.py:968] (0/2) Epoch 15, batch 17550, giga_loss[loss=0.2334, simple_loss=0.3033, pruned_loss=0.08169, over 28687.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3458, pruned_loss=0.09694, over 5688653.08 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3314, pruned_loss=0.08847, over 5725131.11 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3468, pruned_loss=0.0974, over 5680810.05 frames. ], batch size: 92, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:39:15,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.686e+02 1.175e+03 1.465e+03 1.903e+03 3.858e+03, threshold=2.930e+03, percent-clipped=2.0
+2023-03-07 19:39:23,516 INFO [train.py:968] (0/2) Epoch 15, batch 17600, giga_loss[loss=0.2333, simple_loss=0.3143, pruned_loss=0.07616, over 27956.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3384, pruned_loss=0.09364, over 5677172.90 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3313, pruned_loss=0.08839, over 5719725.07 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3395, pruned_loss=0.09422, over 5674643.12 frames. ], batch size: 412, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:40:06,483 INFO [train.py:968] (0/2) Epoch 15, batch 17650, giga_loss[loss=0.2468, simple_loss=0.3173, pruned_loss=0.08816, over 28962.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3298, pruned_loss=0.08964, over 5686442.78 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3312, pruned_loss=0.08831, over 5724707.42 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3308, pruned_loss=0.09026, over 5678915.29 frames. ], batch size: 145, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:40:42,597 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.980e+02 1.005e+03 1.260e+03 1.716e+03 4.537e+03, threshold=2.520e+03, percent-clipped=4.0
+2023-03-07 19:40:49,248 INFO [train.py:968] (0/2) Epoch 15, batch 17700, libri_loss[loss=0.322, simple_loss=0.3868, pruned_loss=0.1286, over 29487.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3243, pruned_loss=0.08757, over 5690778.57 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3324, pruned_loss=0.08904, over 5724194.94 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3239, pruned_loss=0.08738, over 5684656.01 frames. ], batch size: 85, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:41:18,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3320, 1.2192, 1.1068, 1.4770], device='cuda:0'), covar=tensor([0.0764, 0.0363, 0.0353, 0.0870], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0063, 0.0058, 0.0098], device='cuda:0')
+2023-03-07 19:41:29,274 INFO [train.py:968] (0/2) Epoch 15, batch 17750, giga_loss[loss=0.198, simple_loss=0.2658, pruned_loss=0.06507, over 23941.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3181, pruned_loss=0.08433, over 5691667.27 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3321, pruned_loss=0.08877, over 5726169.95 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3177, pruned_loss=0.08433, over 5684175.23 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:42:01,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.075e+02 1.138e+03 1.408e+03 2.107e+03 1.357e+04, threshold=2.815e+03, percent-clipped=15.0
+2023-03-07 19:42:07,368 INFO [train.py:968] (0/2) Epoch 15, batch 17800, giga_loss[loss=0.299, simple_loss=0.3602, pruned_loss=0.1189, over 29025.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3164, pruned_loss=0.08364, over 5698610.89 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.332, pruned_loss=0.08862, over 5728589.43 frames. ], giga_tot_loss[loss=0.2413, simple_loss=0.3155, pruned_loss=0.08358, over 5689414.53 frames. ], batch size: 155, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:42:25,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2066, 1.0920, 3.9175, 3.1920], device='cuda:0'), covar=tensor([0.2117, 0.3177, 0.0702, 0.0975], device='cuda:0'), in_proj_covar=tensor([0.0696, 0.0608, 0.0890, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 19:42:42,418 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=656759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:42:48,793 INFO [train.py:968] (0/2) Epoch 15, batch 17850, giga_loss[loss=0.2021, simple_loss=0.2876, pruned_loss=0.05827, over 28912.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.313, pruned_loss=0.08206, over 5701901.78 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.332, pruned_loss=0.08848, over 5735474.93 frames. ], giga_tot_loss[loss=0.2378, simple_loss=0.3118, pruned_loss=0.08191, over 5687135.39 frames. ], batch size: 174, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:43:24,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.942e+02 9.680e+02 1.363e+03 1.785e+03 3.620e+03, threshold=2.726e+03, percent-clipped=0.0
+2023-03-07 19:43:31,325 INFO [train.py:968] (0/2) Epoch 15, batch 17900, giga_loss[loss=0.2207, simple_loss=0.2951, pruned_loss=0.0731, over 29008.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3114, pruned_loss=0.08144, over 5706647.18 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3329, pruned_loss=0.08891, over 5736338.03 frames. ], giga_tot_loss[loss=0.2351, simple_loss=0.3088, pruned_loss=0.08072, over 5693024.30 frames. ], batch size: 128, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:43:44,101 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-07 19:44:12,350 INFO [train.py:968] (0/2) Epoch 15, batch 17950, giga_loss[loss=0.2396, simple_loss=0.3078, pruned_loss=0.08564, over 28993.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3079, pruned_loss=0.07973, over 5702804.16 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3332, pruned_loss=0.08897, over 5738708.36 frames. ], giga_tot_loss[loss=0.2316, simple_loss=0.3053, pruned_loss=0.07897, over 5689421.38 frames. ], batch size: 106, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:44:15,420 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-07 19:44:34,088 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=656891.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:44:46,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0880, 3.2363, 2.1163, 1.2911], device='cuda:0'), covar=tensor([0.6480, 0.2379, 0.3641, 0.5626], device='cuda:0'), in_proj_covar=tensor([0.1619, 0.1552, 0.1526, 0.1329], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 19:44:49,132 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.103e+02 1.051e+03 1.304e+03 1.721e+03 4.138e+03, threshold=2.607e+03, percent-clipped=9.0
+2023-03-07 19:44:54,717 INFO [train.py:968] (0/2) Epoch 15, batch 18000, giga_loss[loss=0.2005, simple_loss=0.2726, pruned_loss=0.06419, over 28960.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3061, pruned_loss=0.07881, over 5704700.57 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3334, pruned_loss=0.08888, over 5743307.01 frames. ], giga_tot_loss[loss=0.2293, simple_loss=0.3027, pruned_loss=0.0779, over 5687980.56 frames. ], batch size: 106, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:44:54,721 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 19:45:03,168 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2074, simple_loss=0.3121, pruned_loss=0.05131, over 944034.00 frames. 
+2023-03-07 19:45:03,168 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 19:45:11,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3604, 2.9193, 1.5476, 1.4615], device='cuda:0'), covar=tensor([0.0885, 0.0354, 0.0822, 0.1235], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0520, 0.0356, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 19:45:36,450 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=656958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:45:45,541 INFO [train.py:968] (0/2) Epoch 15, batch 18050, giga_loss[loss=0.2389, simple_loss=0.289, pruned_loss=0.09443, over 23833.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3025, pruned_loss=0.07712, over 5703230.88 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3328, pruned_loss=0.08846, over 5748069.23 frames. ], giga_tot_loss[loss=0.2261, simple_loss=0.2994, pruned_loss=0.07642, over 5684102.23 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:45:57,128 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6638, 1.8458, 1.9362, 1.4875], device='cuda:0'), covar=tensor([0.1813, 0.2339, 0.1431, 0.1654], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0687, 0.0910, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 19:46:22,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.540e+02 1.004e+03 1.348e+03 1.755e+03 7.588e+03, threshold=2.695e+03, percent-clipped=9.0
+2023-03-07 19:46:27,311 INFO [train.py:968] (0/2) Epoch 15, batch 18100, giga_loss[loss=0.222, simple_loss=0.2877, pruned_loss=0.07809, over 27684.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3012, pruned_loss=0.0768, over 5699025.79 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3332, pruned_loss=0.08877, over 5748048.64 frames. ], giga_tot_loss[loss=0.2242, simple_loss=0.2972, pruned_loss=0.07555, over 5682102.89 frames. ], batch size: 472, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:46:36,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5410, 1.7554, 1.5992, 1.4876], device='cuda:0'), covar=tensor([0.2529, 0.1962, 0.1656, 0.1962], device='cuda:0'), in_proj_covar=tensor([0.1808, 0.1697, 0.1629, 0.1773], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 19:47:10,853 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-07 19:47:13,171 INFO [train.py:968] (0/2) Epoch 15, batch 18150, giga_loss[loss=0.1771, simple_loss=0.2572, pruned_loss=0.04853, over 29077.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2978, pruned_loss=0.07506, over 5696004.24 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3334, pruned_loss=0.08866, over 5753160.21 frames. ], giga_tot_loss[loss=0.2205, simple_loss=0.2935, pruned_loss=0.07378, over 5676201.74 frames. ], batch size: 128, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:47:49,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.300e+02 9.448e+02 1.257e+03 1.812e+03 4.603e+03, threshold=2.514e+03, percent-clipped=7.0
+2023-03-07 19:47:54,915 INFO [train.py:968] (0/2) Epoch 15, batch 18200, giga_loss[loss=0.3237, simple_loss=0.3766, pruned_loss=0.1355, over 27637.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3001, pruned_loss=0.07668, over 5695618.13 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.334, pruned_loss=0.08888, over 5757952.80 frames. ], giga_tot_loss[loss=0.2223, simple_loss=0.2947, pruned_loss=0.07497, over 5672981.05 frames. ], batch size: 472, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:48:11,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:48:45,895 INFO [train.py:968] (0/2) Epoch 15, batch 18250, giga_loss[loss=0.3009, simple_loss=0.3688, pruned_loss=0.1165, over 28259.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.312, pruned_loss=0.08318, over 5686349.86 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3338, pruned_loss=0.08871, over 5752174.65 frames. ], giga_tot_loss[loss=0.2355, simple_loss=0.3073, pruned_loss=0.0818, over 5671778.05 frames. ], batch size: 65, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:49:21,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.714e+02 1.366e+03 1.708e+03 2.446e+03 1.095e+04, threshold=3.416e+03, percent-clipped=22.0
+2023-03-07 19:49:26,215 INFO [train.py:968] (0/2) Epoch 15, batch 18300, giga_loss[loss=0.3021, simple_loss=0.3768, pruned_loss=0.1137, over 28557.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3233, pruned_loss=0.08889, over 5697471.97 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3329, pruned_loss=0.08822, over 5758630.35 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3195, pruned_loss=0.08811, over 5676239.37 frames. ], batch size: 60, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:49:27,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4862, 1.5944, 1.7242, 1.3295], device='cuda:0'), covar=tensor([0.1578, 0.2308, 0.1307, 0.1655], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0689, 0.0909, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 19:49:57,541 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:50:03,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657266.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:50:04,872 INFO [train.py:968] (0/2) Epoch 15, batch 18350, giga_loss[loss=0.3021, simple_loss=0.3724, pruned_loss=0.1159, over 28115.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3336, pruned_loss=0.09409, over 5700106.08 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3331, pruned_loss=0.08809, over 5758760.45 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3302, pruned_loss=0.0937, over 5680766.98 frames. ], batch size: 77, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:50:05,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2636, 1.7680, 1.4597, 0.4010], device='cuda:0'), covar=tensor([0.3703, 0.2397, 0.3421, 0.4961], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1544, 0.1524, 0.1327], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 19:50:08,450 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-07 19:50:10,250 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:50:12,259 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:50:37,486 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:50:40,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.619e+02 1.193e+03 1.497e+03 1.954e+03 3.430e+03, threshold=2.995e+03, percent-clipped=1.0
+2023-03-07 19:50:46,685 INFO [train.py:968] (0/2) Epoch 15, batch 18400, giga_loss[loss=0.2736, simple_loss=0.3565, pruned_loss=0.09532, over 28930.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3389, pruned_loss=0.09578, over 5692347.08 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3328, pruned_loss=0.08784, over 5758760.82 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3366, pruned_loss=0.09583, over 5675901.22 frames. ], batch size: 227, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:50:59,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:51:26,549 INFO [train.py:968] (0/2) Epoch 15, batch 18450, giga_loss[loss=0.2443, simple_loss=0.3277, pruned_loss=0.08042, over 28840.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3409, pruned_loss=0.09525, over 5698909.35 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.333, pruned_loss=0.08776, over 5760101.90 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3391, pruned_loss=0.09557, over 5682940.80 frames. ], batch size: 119, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:51:53,047 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657396.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:51:56,378 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:52:00,961 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657407.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:52:02,590 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657409.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:52:04,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.798e+02 1.145e+03 1.433e+03 1.923e+03 3.561e+03, threshold=2.865e+03, percent-clipped=3.0
+2023-03-07 19:52:05,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657412.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:52:11,706 INFO [train.py:968] (0/2) Epoch 15, batch 18500, giga_loss[loss=0.2869, simple_loss=0.3662, pruned_loss=0.1038, over 28675.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3444, pruned_loss=0.09677, over 5680015.28 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3336, pruned_loss=0.08802, over 5759849.46 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3426, pruned_loss=0.09697, over 5665962.00 frames. ], batch size: 242, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:52:28,584 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.11 vs. limit=5.0
+2023-03-07 19:52:34,239 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657441.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:52:55,384 INFO [train.py:968] (0/2) Epoch 15, batch 18550, giga_loss[loss=0.3029, simple_loss=0.37, pruned_loss=0.1179, over 28582.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3466, pruned_loss=0.09849, over 5682198.03 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3342, pruned_loss=0.08819, over 5761881.14 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.345, pruned_loss=0.09877, over 5666991.07 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:53:01,204 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:53:02,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:53:28,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:53:31,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.074e+02 1.236e+03 1.495e+03 1.971e+03 5.241e+03, threshold=2.991e+03, percent-clipped=8.0
+2023-03-07 19:53:38,026 INFO [train.py:968] (0/2) Epoch 15, batch 18600, giga_loss[loss=0.2794, simple_loss=0.3583, pruned_loss=0.1002, over 28625.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3505, pruned_loss=0.1019, over 5680291.85 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3345, pruned_loss=0.08832, over 5761386.71 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3493, pruned_loss=0.1023, over 5666685.52 frames. ], batch size: 242, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:54:21,581 INFO [train.py:968] (0/2) Epoch 15, batch 18650, giga_loss[loss=0.298, simple_loss=0.3703, pruned_loss=0.1129, over 29045.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3544, pruned_loss=0.1042, over 5682642.87 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3351, pruned_loss=0.0888, over 5763433.79 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.353, pruned_loss=0.1042, over 5669183.34 frames. ], batch size: 128, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:54:58,351 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.556e+02 1.185e+03 1.426e+03 1.896e+03 5.681e+03, threshold=2.852e+03, percent-clipped=7.0
+2023-03-07 19:55:03,395 INFO [train.py:968] (0/2) Epoch 15, batch 18700, giga_loss[loss=0.2725, simple_loss=0.3534, pruned_loss=0.09583, over 28900.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3564, pruned_loss=0.1043, over 5689323.17 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3351, pruned_loss=0.0888, over 5764774.72 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3555, pruned_loss=0.1045, over 5676951.46 frames. ], batch size: 227, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:55:16,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:55:43,488 INFO [train.py:968] (0/2) Epoch 15, batch 18750, giga_loss[loss=0.3548, simple_loss=0.419, pruned_loss=0.1453, over 28818.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3588, pruned_loss=0.1052, over 5678901.74 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3356, pruned_loss=0.08899, over 5755006.22 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3583, pruned_loss=0.1056, over 5675404.22 frames. ], batch size: 119, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:55:57,911 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 19:56:19,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.175e+02 1.261e+03 1.586e+03 2.097e+03 4.154e+03, threshold=3.171e+03, percent-clipped=2.0
+2023-03-07 19:56:25,300 INFO [train.py:968] (0/2) Epoch 15, batch 18800, giga_loss[loss=0.2473, simple_loss=0.3368, pruned_loss=0.07886, over 28519.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3595, pruned_loss=0.1049, over 5687420.17 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3356, pruned_loss=0.08887, over 5758970.98 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3596, pruned_loss=0.1057, over 5679647.34 frames. ], batch size: 71, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:57:02,826 INFO [train.py:968] (0/2) Epoch 15, batch 18850, giga_loss[loss=0.2793, simple_loss=0.3585, pruned_loss=0.1001, over 28705.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.359, pruned_loss=0.1031, over 5699821.22 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3362, pruned_loss=0.08903, over 5759141.02 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3591, pruned_loss=0.1039, over 5691789.87 frames. ], batch size: 92, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:57:04,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657771.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:57:11,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:57:12,233 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:57:15,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:57:16,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657782.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:57:37,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7657, 1.9211, 1.4978, 2.4491], device='cuda:0'), covar=tensor([0.2438, 0.2474, 0.2828, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.1374, 0.1010, 0.1224, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 19:57:38,963 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:57:40,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.150e+02 1.120e+03 1.319e+03 1.858e+03 4.601e+03, threshold=2.638e+03, percent-clipped=5.0
+2023-03-07 19:57:44,864 INFO [train.py:968] (0/2) Epoch 15, batch 18900, giga_loss[loss=0.2953, simple_loss=0.3703, pruned_loss=0.1101, over 28527.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3575, pruned_loss=0.1011, over 5704477.40 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3367, pruned_loss=0.08921, over 5759306.82 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3575, pruned_loss=0.1018, over 5697267.07 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:57:51,487 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:58:24,455 INFO [train.py:968] (0/2) Epoch 15, batch 18950, giga_loss[loss=0.3069, simple_loss=0.3756, pruned_loss=0.1192, over 27919.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.356, pruned_loss=0.1, over 5708384.95 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3369, pruned_loss=0.08925, over 5761767.78 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3563, pruned_loss=0.1008, over 5698878.00 frames. ], batch size: 412, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:58:27,355 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:58:38,826 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-07 19:58:45,844 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-07 19:59:00,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.639e+02 1.159e+03 1.421e+03 1.926e+03 5.578e+03, threshold=2.842e+03, percent-clipped=8.0
+2023-03-07 19:59:01,250 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657914.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:59:04,745 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657917.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:59:06,175 INFO [train.py:968] (0/2) Epoch 15, batch 19000, giga_loss[loss=0.2711, simple_loss=0.3519, pruned_loss=0.09519, over 29081.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3582, pruned_loss=0.1036, over 5705655.81 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3373, pruned_loss=0.08938, over 5756008.82 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3586, pruned_loss=0.1044, over 5702723.67 frames. ], batch size: 164, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:59:06,467 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:59:08,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:59:13,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657925.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:59:15,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657928.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:59:29,829 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657946.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:59:35,391 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:59:40,697 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657957.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:59:48,817 INFO [train.py:968] (0/2) Epoch 15, batch 19050, giga_loss[loss=0.3344, simple_loss=0.3815, pruned_loss=0.1436, over 28872.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3603, pruned_loss=0.1072, over 5711203.14 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3377, pruned_loss=0.08954, over 5759731.50 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.361, pruned_loss=0.1083, over 5703816.78 frames. ], batch size: 99, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:00:13,682 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-658000.pt
+2023-03-07 20:00:22,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4893, 1.7764, 1.4195, 1.5010], device='cuda:0'), covar=tensor([0.2419, 0.2371, 0.2697, 0.2221], device='cuda:0'), in_proj_covar=tensor([0.1368, 0.1005, 0.1222, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 20:00:22,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.379e+02 1.307e+03 1.657e+03 2.448e+03 1.018e+04, threshold=3.315e+03, percent-clipped=16.0
+2023-03-07 20:00:27,279 INFO [train.py:968] (0/2) Epoch 15, batch 19100, giga_loss[loss=0.2983, simple_loss=0.3586, pruned_loss=0.119, over 27555.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3591, pruned_loss=0.1073, over 5714206.51 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3379, pruned_loss=0.08959, over 5762774.39 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3599, pruned_loss=0.1084, over 5704793.07 frames. ], batch size: 472, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:01:11,621 INFO [train.py:968] (0/2) Epoch 15, batch 19150, giga_loss[loss=0.3308, simple_loss=0.3756, pruned_loss=0.1431, over 26523.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3579, pruned_loss=0.1077, over 5705604.68 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3383, pruned_loss=0.08969, over 5764355.84 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3586, pruned_loss=0.1089, over 5695192.76 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:01:49,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.328e+02 1.274e+03 1.654e+03 2.066e+03 8.356e+03, threshold=3.308e+03, percent-clipped=5.0
+2023-03-07 20:01:55,593 INFO [train.py:968] (0/2) Epoch 15, batch 19200, giga_loss[loss=0.2892, simple_loss=0.3595, pruned_loss=0.1094, over 28567.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3562, pruned_loss=0.1067, over 5711385.85 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3384, pruned_loss=0.08971, over 5765939.60 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3568, pruned_loss=0.1079, over 5701233.16 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 20:02:06,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7761, 1.8604, 1.9975, 1.5658], device='cuda:0'), covar=tensor([0.1879, 0.2281, 0.1428, 0.1663], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0687, 0.0904, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 20:02:34,815 INFO [train.py:968] (0/2) Epoch 15, batch 19250, giga_loss[loss=0.2687, simple_loss=0.3507, pruned_loss=0.09337, over 28524.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3541, pruned_loss=0.1044, over 5708554.22 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3384, pruned_loss=0.08966, over 5757429.88 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3551, pruned_loss=0.1058, over 5706164.88 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:02:55,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3133, 1.5667, 1.2867, 1.4992], device='cuda:0'), covar=tensor([0.0773, 0.0342, 0.0342, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:0')
+2023-03-07 20:03:01,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=658202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:03:09,466 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8943, 2.8681, 1.8885, 1.0231], device='cuda:0'), covar=tensor([0.6624, 0.2749, 0.3360, 0.5507], device='cuda:0'), in_proj_covar=tensor([0.1602, 0.1536, 0.1516, 0.1321], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 20:03:12,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.724e+02 1.125e+03 1.391e+03 1.668e+03 8.237e+03, threshold=2.782e+03, percent-clipped=4.0
+2023-03-07 20:03:16,776 INFO [train.py:968] (0/2) Epoch 15, batch 19300, giga_loss[loss=0.2625, simple_loss=0.3314, pruned_loss=0.09682, over 28714.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3519, pruned_loss=0.1027, over 5697494.66 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3383, pruned_loss=0.08962, over 5752928.72 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3533, pruned_loss=0.1044, over 5698315.28 frames. ], batch size: 92, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:03:41,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=658248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:03:59,569 INFO [train.py:968] (0/2) Epoch 15, batch 19350, giga_loss[loss=0.2234, simple_loss=0.2971, pruned_loss=0.07492, over 28848.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3469, pruned_loss=0.09956, over 5693599.00 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3388, pruned_loss=0.08984, over 5755647.33 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3479, pruned_loss=0.101, over 5690244.18 frames. ], batch size: 106, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:04:41,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.386e+02 1.074e+03 1.337e+03 1.715e+03 3.956e+03, threshold=2.674e+03, percent-clipped=5.0
+2023-03-07 20:04:44,385 INFO [train.py:968] (0/2) Epoch 15, batch 19400, libri_loss[loss=0.2803, simple_loss=0.367, pruned_loss=0.09682, over 29310.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3421, pruned_loss=0.0973, over 5679477.03 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3392, pruned_loss=0.09, over 5750103.28 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3426, pruned_loss=0.09844, over 5680828.41 frames. ], batch size: 94, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:05:08,806 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=658345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:05:11,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=658348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:05:31,846 INFO [train.py:968] (0/2) Epoch 15, batch 19450, giga_loss[loss=0.2188, simple_loss=0.305, pruned_loss=0.06628, over 28865.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3366, pruned_loss=0.09483, over 5666478.74 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3397, pruned_loss=0.09041, over 5753332.70 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3364, pruned_loss=0.09546, over 5663275.12 frames. ], batch size: 174, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:05:36,391 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-07 20:05:40,553 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=658377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:05:54,233 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=658391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:05:56,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=658394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:06:12,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.145e+02 9.789e+02 1.195e+03 1.675e+03 5.901e+03, threshold=2.389e+03, percent-clipped=10.0
+2023-03-07 20:06:17,937 INFO [train.py:968] (0/2) Epoch 15, batch 19500, giga_loss[loss=0.2549, simple_loss=0.3384, pruned_loss=0.08575, over 29118.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3352, pruned_loss=0.09386, over 5655421.49 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.34, pruned_loss=0.09048, over 5754600.48 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3347, pruned_loss=0.09439, over 5649522.73 frames. ], batch size: 155, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:06:20,933 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=658423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:06:20,959 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=658423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:06:21,483 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=658424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:07:00,388 INFO [train.py:968] (0/2) Epoch 15, batch 19550, giga_loss[loss=0.3219, simple_loss=0.3814, pruned_loss=0.1312, over 28083.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3359, pruned_loss=0.09426, over 5666012.97 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3397, pruned_loss=0.09029, over 5756835.90 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3357, pruned_loss=0.09491, over 5657956.56 frames. ], batch size: 412, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:07:40,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.785e+02 1.074e+03 1.376e+03 1.945e+03 7.596e+03, threshold=2.752e+03, percent-clipped=18.0
+2023-03-07 20:07:43,978 INFO [train.py:968] (0/2) Epoch 15, batch 19600, giga_loss[loss=0.2604, simple_loss=0.3319, pruned_loss=0.09448, over 28935.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3348, pruned_loss=0.0933, over 5676268.88 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.34, pruned_loss=0.09018, over 5758816.13 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3344, pruned_loss=0.09402, over 5665922.66 frames. ], batch size: 227, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 20:08:04,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6062, 1.7586, 1.8971, 1.4219], device='cuda:0'), covar=tensor([0.1932, 0.2276, 0.1463, 0.1661], device='cuda:0'), in_proj_covar=tensor([0.0860, 0.0690, 0.0906, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 20:08:13,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4047, 2.8507, 2.4408, 1.9625], device='cuda:0'), covar=tensor([0.2068, 0.1160, 0.1318, 0.1649], device='cuda:0'), in_proj_covar=tensor([0.1803, 0.1708, 0.1650, 0.1783], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 20:08:26,528 INFO [train.py:968] (0/2) Epoch 15, batch 19650, giga_loss[loss=0.2199, simple_loss=0.2972, pruned_loss=0.07133, over 28368.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3335, pruned_loss=0.09253, over 5684101.42 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3402, pruned_loss=0.09022, over 5761877.63 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3328, pruned_loss=0.09313, over 5671648.50 frames. ], batch size: 77, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 20:09:00,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.152e+02 1.057e+03 1.308e+03 1.806e+03 6.492e+03, threshold=2.616e+03, percent-clipped=5.0
+2023-03-07 20:09:04,065 INFO [train.py:968] (0/2) Epoch 15, batch 19700, giga_loss[loss=0.2709, simple_loss=0.3391, pruned_loss=0.1014, over 28938.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3321, pruned_loss=0.09175, over 5689387.41 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3407, pruned_loss=0.0903, over 5758246.21 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3309, pruned_loss=0.09224, over 5679923.48 frames. ], batch size: 145, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 20:09:07,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9261, 1.2958, 1.0973, 0.2367], device='cuda:0'), covar=tensor([0.3429, 0.2550, 0.4312, 0.5318], device='cuda:0'), in_proj_covar=tensor([0.1603, 0.1535, 0.1514, 0.1321], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 20:09:46,373 INFO [train.py:968] (0/2) Epoch 15, batch 19750, giga_loss[loss=0.2529, simple_loss=0.3294, pruned_loss=0.08815, over 28744.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.33, pruned_loss=0.09059, over 5700581.48 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3413, pruned_loss=0.09054, over 5761569.79 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3283, pruned_loss=0.09078, over 5689055.09 frames. ], batch size: 284, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:10:25,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.889e+02 9.733e+02 1.165e+03 1.493e+03 3.115e+03, threshold=2.330e+03, percent-clipped=6.0
+2023-03-07 20:10:28,111 INFO [train.py:968] (0/2) Epoch 15, batch 19800, giga_loss[loss=0.258, simple_loss=0.321, pruned_loss=0.09745, over 28646.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3271, pruned_loss=0.08956, over 5702595.18 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3413, pruned_loss=0.09053, over 5763759.35 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3256, pruned_loss=0.0897, over 5690615.41 frames. ], batch size: 78, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:10:49,808 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-07 20:11:09,466 INFO [train.py:968] (0/2) Epoch 15, batch 19850, giga_loss[loss=0.3103, simple_loss=0.363, pruned_loss=0.1288, over 26773.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3238, pruned_loss=0.08843, over 5711767.94 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3416, pruned_loss=0.09063, over 5764450.55 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3223, pruned_loss=0.08844, over 5701485.31 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:11:32,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=658798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:11:33,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=658799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:11:44,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.535e+02 9.580e+02 1.214e+03 1.619e+03 6.912e+03, threshold=2.427e+03, percent-clipped=13.0
+2023-03-07 20:11:48,205 INFO [train.py:968] (0/2) Epoch 15, batch 19900, giga_loss[loss=0.2559, simple_loss=0.3176, pruned_loss=0.09707, over 28822.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3237, pruned_loss=0.08815, over 5720584.28 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3432, pruned_loss=0.09128, over 5766012.39 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3204, pruned_loss=0.08751, over 5709380.35 frames. ], batch size: 99, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:12:21,679 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3299, 1.5502, 1.5199, 1.4723], device='cuda:0'), covar=tensor([0.1715, 0.1932, 0.1880, 0.1811], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0733, 0.0687, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 20:12:30,688 INFO [train.py:968] (0/2) Epoch 15, batch 19950, giga_loss[loss=0.2609, simple_loss=0.3213, pruned_loss=0.1002, over 28733.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3229, pruned_loss=0.08749, over 5718108.54 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3441, pruned_loss=0.09162, over 5769776.38 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3189, pruned_loss=0.08656, over 5704563.89 frames. ], batch size: 119, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:13:00,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3612, 3.5668, 1.6494, 1.4458], device='cuda:0'), covar=tensor([0.1041, 0.0286, 0.0842, 0.1411], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0521, 0.0355, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 20:13:08,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.709e+02 9.959e+02 1.183e+03 1.569e+03 7.203e+03, threshold=2.365e+03, percent-clipped=13.0
+2023-03-07 20:13:11,721 INFO [train.py:968] (0/2) Epoch 15, batch 20000, giga_loss[loss=0.2342, simple_loss=0.3142, pruned_loss=0.07712, over 28904.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3208, pruned_loss=0.08632, over 5721111.13 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3442, pruned_loss=0.0915, over 5772090.93 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3173, pruned_loss=0.08563, over 5707910.78 frames. ], batch size: 174, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:13:28,098 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=658941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:13:28,760 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=658942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:13:29,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=658944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:13:30,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=658945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:13:48,623 INFO [train.py:968] (0/2) Epoch 15, batch 20050, giga_loss[loss=0.3128, simple_loss=0.3669, pruned_loss=0.1294, over 26701.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3215, pruned_loss=0.08644, over 5727699.22 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.345, pruned_loss=0.09181, over 5775393.69 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3174, pruned_loss=0.08546, over 5713094.90 frames. ], batch size: 555, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:13:51,689 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=658973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:13:52,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=658974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:14:24,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.493e+02 1.009e+03 1.316e+03 1.986e+03 4.399e+03, threshold=2.632e+03, percent-clipped=18.0
+2023-03-07 20:14:29,351 INFO [train.py:968] (0/2) Epoch 15, batch 20100, giga_loss[loss=0.2989, simple_loss=0.3617, pruned_loss=0.1181, over 28961.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3263, pruned_loss=0.08935, over 5718898.42 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3458, pruned_loss=0.09208, over 5773550.38 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3215, pruned_loss=0.08817, over 5706993.25 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:14:45,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=659038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:15:16,026 INFO [train.py:968] (0/2) Epoch 15, batch 20150, giga_loss[loss=0.3104, simple_loss=0.379, pruned_loss=0.1209, over 28919.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3323, pruned_loss=0.09324, over 5708574.84 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3461, pruned_loss=0.09209, over 5773870.85 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3278, pruned_loss=0.09226, over 5697402.79 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:15:25,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2463, 1.2756, 3.6556, 2.9658], device='cuda:0'), covar=tensor([0.1564, 0.2524, 0.0418, 0.0988], device='cuda:0'), in_proj_covar=tensor([0.0687, 0.0599, 0.0877, 0.0798], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 20:15:34,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2702, 1.5152, 1.2604, 1.0034], device='cuda:0'), covar=tensor([0.2350, 0.2350, 0.2534, 0.2064], device='cuda:0'), in_proj_covar=tensor([0.1374, 0.1008, 0.1223, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 20:15:56,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.837e+02 1.244e+03 1.478e+03 2.025e+03 6.532e+03, threshold=2.956e+03, percent-clipped=15.0
+2023-03-07 20:15:59,475 INFO [train.py:968] (0/2) Epoch 15, batch 20200, giga_loss[loss=0.2714, simple_loss=0.34, pruned_loss=0.1014, over 28629.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3396, pruned_loss=0.09775, over 5704990.73 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3466, pruned_loss=0.09227, over 5770442.65 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3349, pruned_loss=0.09698, over 5695681.25 frames. ], batch size: 85, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:16:36,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=659160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:16:43,796 INFO [train.py:968] (0/2) Epoch 15, batch 20250, giga_loss[loss=0.3146, simple_loss=0.3789, pruned_loss=0.1252, over 28796.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3458, pruned_loss=0.1017, over 5689822.89 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3468, pruned_loss=0.09232, over 5763048.23 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3419, pruned_loss=0.1011, over 5687787.02 frames. ], batch size: 119, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:16:53,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6997, 5.4492, 5.1330, 2.6120], device='cuda:0'), covar=tensor([0.0427, 0.0598, 0.0603, 0.1818], device='cuda:0'), in_proj_covar=tensor([0.1111, 0.1019, 0.0882, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 20:17:19,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5879, 1.7139, 1.4149, 1.5955], device='cuda:0'), covar=tensor([0.2435, 0.2443, 0.2738, 0.2235], device='cuda:0'), in_proj_covar=tensor([0.1375, 0.1008, 0.1222, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 20:17:28,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.097e+02 1.239e+03 1.497e+03 1.947e+03 4.248e+03, threshold=2.995e+03, percent-clipped=4.0
+2023-03-07 20:17:30,152 INFO [train.py:968] (0/2) Epoch 15, batch 20300, giga_loss[loss=0.3008, simple_loss=0.3768, pruned_loss=0.1123, over 28504.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3508, pruned_loss=0.1037, over 5693432.10 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3468, pruned_loss=0.0924, over 5765907.76 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3476, pruned_loss=0.1034, over 5687349.96 frames. ], batch size: 336, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:17:42,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1659, 2.1657, 1.9840, 1.9622], device='cuda:0'), covar=tensor([0.1578, 0.2182, 0.2014, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0730, 0.0684, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 20:18:16,449 INFO [train.py:968] (0/2) Epoch 15, batch 20350, giga_loss[loss=0.3328, simple_loss=0.397, pruned_loss=0.1343, over 27621.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3552, pruned_loss=0.1054, over 5692265.53 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3471, pruned_loss=0.09271, over 5767818.07 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3525, pruned_loss=0.1052, over 5684887.52 frames. ], batch size: 472, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:18:55,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.137e+02 1.191e+03 1.455e+03 1.961e+03 4.293e+03, threshold=2.909e+03, percent-clipped=5.0
+2023-03-07 20:18:58,601 INFO [train.py:968] (0/2) Epoch 15, batch 20400, giga_loss[loss=0.3688, simple_loss=0.407, pruned_loss=0.1653, over 26687.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3591, pruned_loss=0.1075, over 5695965.56 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3475, pruned_loss=0.09293, over 5763769.35 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.357, pruned_loss=0.1075, over 5690975.34 frames. ], batch size: 555, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:19:41,870 INFO [train.py:968] (0/2) Epoch 15, batch 20450, libri_loss[loss=0.3115, simple_loss=0.3723, pruned_loss=0.1253, over 27154.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3548, pruned_loss=0.1046, over 5688937.46 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3481, pruned_loss=0.09353, over 5764538.34 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3526, pruned_loss=0.1044, over 5682787.87 frames. ], batch size: 60, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:19:42,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8884, 1.1573, 1.3541, 0.9568], device='cuda:0'), covar=tensor([0.1793, 0.1304, 0.2041, 0.1607], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0726, 0.0681, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 20:19:54,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6832, 1.7664, 1.4549, 1.7014], device='cuda:0'), covar=tensor([0.2547, 0.2557, 0.2754, 0.2342], device='cuda:0'), in_proj_covar=tensor([0.1370, 0.1004, 0.1218, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 20:20:21,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=659413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:20:24,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.690e+02 1.169e+03 1.448e+03 2.202e+03 1.092e+04, threshold=2.897e+03, percent-clipped=10.0
+2023-03-07 20:20:26,164 INFO [train.py:968] (0/2) Epoch 15, batch 20500, giga_loss[loss=0.3001, simple_loss=0.3692, pruned_loss=0.1155, over 28346.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.351, pruned_loss=0.1011, over 5700534.57 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3481, pruned_loss=0.09354, over 5765362.65 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3493, pruned_loss=0.1009, over 5694749.56 frames. ], batch size: 368, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:20:50,266 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6165, 1.6871, 1.9041, 1.4028], device='cuda:0'), covar=tensor([0.1761, 0.2264, 0.1382, 0.1662], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0693, 0.0911, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 20:21:10,315 INFO [train.py:968] (0/2) Epoch 15, batch 20550, giga_loss[loss=0.2363, simple_loss=0.323, pruned_loss=0.07478, over 28214.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3504, pruned_loss=0.1004, over 5690408.36 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3481, pruned_loss=0.09362, over 5767110.81 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3491, pruned_loss=0.1003, over 5683360.34 frames. ], batch size: 65, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:21:23,908 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 20:21:53,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.432e+02 1.267e+03 1.502e+03 1.888e+03 5.689e+03, threshold=3.004e+03, percent-clipped=4.0
+2023-03-07 20:21:54,624 INFO [train.py:968] (0/2) Epoch 15, batch 20600, libri_loss[loss=0.2715, simple_loss=0.3507, pruned_loss=0.09615, over 29544.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.351, pruned_loss=0.1003, over 5697336.90 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3481, pruned_loss=0.09351, over 5769600.04 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3501, pruned_loss=0.1005, over 5688203.69 frames. ], batch size: 76, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:22:08,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=659535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:22:17,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7504, 1.7905, 1.7451, 1.5758], device='cuda:0'), covar=tensor([0.1643, 0.2154, 0.2074, 0.2038], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0730, 0.0683, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 20:22:26,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=659556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:22:30,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=659559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:22:38,379 INFO [train.py:968] (0/2) Epoch 15, batch 20650, giga_loss[loss=0.2762, simple_loss=0.3482, pruned_loss=0.1021, over 28892.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3536, pruned_loss=0.1022, over 5697998.35 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3481, pruned_loss=0.09346, over 5771532.21 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.353, pruned_loss=0.1026, over 5687369.24 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:22:44,830 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 20:22:55,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=659588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:23:10,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5514, 4.3989, 4.0817, 1.9721], device='cuda:0'), covar=tensor([0.0538, 0.0632, 0.0658, 0.2031], device='cuda:0'), in_proj_covar=tensor([0.1105, 0.1016, 0.0878, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 20:23:17,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.806e+02 1.358e+03 1.830e+03 2.494e+03 6.475e+03, threshold=3.660e+03, percent-clipped=18.0
+2023-03-07 20:23:20,122 INFO [train.py:968] (0/2) Epoch 15, batch 20700, giga_loss[loss=0.2684, simple_loss=0.3506, pruned_loss=0.09309, over 28970.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3555, pruned_loss=0.1037, over 5693024.02 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3485, pruned_loss=0.09371, over 5762989.05 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3548, pruned_loss=0.104, over 5689664.43 frames. ], batch size: 174, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:23:26,324 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=659626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:23:53,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5224, 5.3413, 5.0423, 2.5694], device='cuda:0'), covar=tensor([0.0406, 0.0520, 0.0569, 0.1690], device='cuda:0'), in_proj_covar=tensor([0.1106, 0.1018, 0.0881, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 20:24:07,105 INFO [train.py:968] (0/2) Epoch 15, batch 20750, giga_loss[loss=0.3125, simple_loss=0.3761, pruned_loss=0.1244, over 28667.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3565, pruned_loss=0.1048, over 5706944.98 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3482, pruned_loss=0.09356, over 5765753.57 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3563, pruned_loss=0.1054, over 5700625.32 frames. ], batch size: 92, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:24:15,920 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=659678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:24:18,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=659681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:24:44,699 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=659710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:24:50,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.303e+02 1.185e+03 1.510e+03 2.076e+03 3.853e+03, threshold=3.020e+03, percent-clipped=1.0
+2023-03-07 20:24:51,545 INFO [train.py:968] (0/2) Epoch 15, batch 20800, giga_loss[loss=0.2748, simple_loss=0.3478, pruned_loss=0.1009, over 28920.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.357, pruned_loss=0.1057, over 5705964.09 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.348, pruned_loss=0.09332, over 5767081.77 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3572, pruned_loss=0.1066, over 5698896.50 frames. ], batch size: 213, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:25:29,212 INFO [train.py:968] (0/2) Epoch 15, batch 20850, libri_loss[loss=0.2522, simple_loss=0.3318, pruned_loss=0.08633, over 29576.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3576, pruned_loss=0.1055, over 5715774.88 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3487, pruned_loss=0.0937, over 5772315.96 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3575, pruned_loss=0.1063, over 5703140.34 frames. ], batch size: 77, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:26:07,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.655e+02 1.049e+03 1.331e+03 1.637e+03 3.400e+03, threshold=2.661e+03, percent-clipped=3.0
+2023-03-07 20:26:08,672 INFO [train.py:968] (0/2) Epoch 15, batch 20900, giga_loss[loss=0.2346, simple_loss=0.3231, pruned_loss=0.07307, over 28397.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3568, pruned_loss=0.104, over 5714778.50 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3491, pruned_loss=0.0939, over 5775347.92 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3566, pruned_loss=0.1047, over 5700809.58 frames. ], batch size: 60, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:26:17,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2327, 1.3824, 1.3761, 1.1515], device='cuda:0'), covar=tensor([0.2149, 0.2142, 0.1342, 0.1863], device='cuda:0'), in_proj_covar=tensor([0.1833, 0.1736, 0.1684, 0.1806], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 20:26:23,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2551, 5.0728, 4.7760, 2.2930], device='cuda:0'), covar=tensor([0.0362, 0.0527, 0.0543, 0.1944], device='cuda:0'), in_proj_covar=tensor([0.1096, 0.1007, 0.0872, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 20:26:24,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=659839.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:26:47,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3194, 3.1250, 1.4150, 1.4411], device='cuda:0'), covar=tensor([0.0972, 0.0254, 0.0889, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0520, 0.0354, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 20:26:48,256 INFO [train.py:968] (0/2) Epoch 15, batch 20950, giga_loss[loss=0.2638, simple_loss=0.3531, pruned_loss=0.08728, over 29021.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3577, pruned_loss=0.1036, over 5717324.21 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3495, pruned_loss=0.09403, over 5773079.97 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3573, pruned_loss=0.1043, over 5706433.16 frames. ], batch size: 164, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:27:21,991 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=659910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:27:27,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.240e+02 1.127e+03 1.456e+03 2.022e+03 6.251e+03, threshold=2.911e+03, percent-clipped=10.0
+2023-03-07 20:27:27,919 INFO [train.py:968] (0/2) Epoch 15, batch 21000, giga_loss[loss=0.2696, simple_loss=0.3395, pruned_loss=0.09979, over 28907.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3576, pruned_loss=0.1035, over 5722908.95 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3496, pruned_loss=0.09422, over 5775100.06 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3575, pruned_loss=0.1041, over 5710781.73 frames. ], batch size: 106, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:27:27,923 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 20:27:36,926 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2119, simple_loss=0.3186, pruned_loss=0.0526, over 944034.00 frames. 
+2023-03-07 20:27:36,927 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 20:28:08,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6253, 1.7098, 1.6294, 1.4882], device='cuda:0'), covar=tensor([0.2190, 0.2178, 0.1788, 0.2035], device='cuda:0'), in_proj_covar=tensor([0.1824, 0.1730, 0.1675, 0.1802], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 20:28:16,351 INFO [train.py:968] (0/2) Epoch 15, batch 21050, giga_loss[loss=0.2469, simple_loss=0.3269, pruned_loss=0.08341, over 28327.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3544, pruned_loss=0.102, over 5714517.49 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3498, pruned_loss=0.0944, over 5772840.55 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3541, pruned_loss=0.1024, over 5706821.61 frames. ], batch size: 65, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:28:40,929 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-660000.pt
+2023-03-07 20:28:42,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:28:54,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.859e+02 1.025e+03 1.219e+03 1.584e+03 4.079e+03, threshold=2.438e+03, percent-clipped=3.0
+2023-03-07 20:28:55,115 INFO [train.py:968] (0/2) Epoch 15, batch 21100, giga_loss[loss=0.2528, simple_loss=0.3376, pruned_loss=0.08404, over 29037.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.353, pruned_loss=0.1017, over 5716653.41 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3497, pruned_loss=0.09445, over 5774155.96 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3529, pruned_loss=0.1021, over 5708749.90 frames. ], batch size: 155, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:28:55,400 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:29:18,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7568, 1.8940, 1.5840, 2.2414], device='cuda:0'), covar=tensor([0.2408, 0.2480, 0.2703, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.1375, 0.1012, 0.1222, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 20:29:26,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4093, 1.7013, 1.7016, 1.2645], device='cuda:0'), covar=tensor([0.1690, 0.2318, 0.1342, 0.1635], device='cuda:0'), in_proj_covar=tensor([0.0862, 0.0692, 0.0908, 0.0807], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 20:29:26,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-07 20:29:34,921 INFO [train.py:968] (0/2) Epoch 15, batch 21150, giga_loss[loss=0.2655, simple_loss=0.3505, pruned_loss=0.0902, over 28933.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3507, pruned_loss=0.1001, over 5723560.99 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3498, pruned_loss=0.09446, over 5777883.20 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3506, pruned_loss=0.1006, over 5712166.41 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:30:16,441 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-07 20:30:17,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.831e+02 1.103e+03 1.294e+03 1.616e+03 5.411e+03, threshold=2.587e+03, percent-clipped=8.0
+2023-03-07 20:30:17,935 INFO [train.py:968] (0/2) Epoch 15, batch 21200, giga_loss[loss=0.3225, simple_loss=0.3851, pruned_loss=0.13, over 28856.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3517, pruned_loss=0.1013, over 5723588.84 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3497, pruned_loss=0.09439, over 5781135.31 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3518, pruned_loss=0.1019, over 5710306.22 frames. ], batch size: 199, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:30:32,267 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-07 20:30:39,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660143.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:30:39,876 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:30:41,894 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:30:59,744 INFO [train.py:968] (0/2) Epoch 15, batch 21250, giga_loss[loss=0.268, simple_loss=0.352, pruned_loss=0.09194, over 28256.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3522, pruned_loss=0.1017, over 5718633.50 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3499, pruned_loss=0.09454, over 5782807.03 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3521, pruned_loss=0.1021, over 5705545.07 frames. ], batch size: 368, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:31:04,364 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:31:35,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660214.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:31:37,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.896e+02 1.047e+03 1.336e+03 1.805e+03 4.497e+03, threshold=2.671e+03, percent-clipped=4.0
+2023-03-07 20:31:38,597 INFO [train.py:968] (0/2) Epoch 15, batch 21300, giga_loss[loss=0.2675, simple_loss=0.3469, pruned_loss=0.094, over 28565.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3517, pruned_loss=0.1009, over 5705858.40 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3499, pruned_loss=0.09475, over 5766009.47 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3517, pruned_loss=0.1013, over 5709839.09 frames. ], batch size: 336, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:31:46,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3393, 1.4683, 1.2733, 1.4548], device='cuda:0'), covar=tensor([0.0794, 0.0351, 0.0344, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0056, 0.0097], device='cuda:0')
+2023-03-07 20:32:08,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5638, 1.6737, 1.9226, 1.4195], device='cuda:0'), covar=tensor([0.1455, 0.2041, 0.1205, 0.1653], device='cuda:0'), in_proj_covar=tensor([0.0861, 0.0693, 0.0908, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 20:32:19,204 INFO [train.py:968] (0/2) Epoch 15, batch 21350, giga_loss[loss=0.3136, simple_loss=0.3748, pruned_loss=0.1262, over 27593.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3514, pruned_loss=0.1005, over 5706430.01 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3501, pruned_loss=0.09509, over 5770036.78 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3512, pruned_loss=0.1007, over 5704145.50 frames. ], batch size: 472, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:32:31,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:33:01,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.076e+02 1.027e+03 1.303e+03 1.705e+03 4.581e+03, threshold=2.605e+03, percent-clipped=7.0
+2023-03-07 20:33:01,161 INFO [train.py:968] (0/2) Epoch 15, batch 21400, giga_loss[loss=0.286, simple_loss=0.3668, pruned_loss=0.1027, over 28461.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.35, pruned_loss=0.1001, over 5700362.03 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3501, pruned_loss=0.09508, over 5771363.08 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3499, pruned_loss=0.1002, over 5696810.42 frames. ], batch size: 65, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:33:29,907 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660357.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:33:31,835 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660360.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:33:38,126 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:33:38,544 INFO [train.py:968] (0/2) Epoch 15, batch 21450, giga_loss[loss=0.2472, simple_loss=0.3217, pruned_loss=0.08638, over 28885.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3487, pruned_loss=0.09994, over 5704420.16 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3503, pruned_loss=0.09528, over 5774655.45 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3483, pruned_loss=0.1, over 5697126.42 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:33:55,898 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660389.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:33:59,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:34:15,832 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-07 20:34:18,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.915e+02 1.131e+03 1.486e+03 1.997e+03 7.087e+03, threshold=2.972e+03, percent-clipped=11.0
+2023-03-07 20:34:18,713 INFO [train.py:968] (0/2) Epoch 15, batch 21500, libri_loss[loss=0.2722, simple_loss=0.3384, pruned_loss=0.103, over 29320.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3458, pruned_loss=0.09886, over 5682738.60 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3505, pruned_loss=0.09552, over 5755324.95 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3453, pruned_loss=0.09876, over 5692654.95 frames. ], batch size: 71, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:34:25,964 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:34:28,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:34:51,527 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:34:58,163 INFO [train.py:968] (0/2) Epoch 15, batch 21550, libri_loss[loss=0.3087, simple_loss=0.3789, pruned_loss=0.1193, over 19600.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3454, pruned_loss=0.09903, over 5663549.89 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3508, pruned_loss=0.09588, over 5730936.25 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3446, pruned_loss=0.09872, over 5691869.42 frames. ], batch size: 187, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:35:01,564 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:35:13,923 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-07 20:35:37,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660518.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:35:38,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.908e+02 1.235e+03 1.647e+03 2.343e+03 9.460e+03, threshold=3.294e+03, percent-clipped=14.0
+2023-03-07 20:35:38,315 INFO [train.py:968] (0/2) Epoch 15, batch 21600, libri_loss[loss=0.3165, simple_loss=0.3868, pruned_loss=0.1231, over 29381.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3457, pruned_loss=0.09947, over 5672893.20 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3516, pruned_loss=0.09649, over 5731288.38 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3442, pruned_loss=0.09873, over 5694106.34 frames. ], batch size: 92, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:35:53,867 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:35:55,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:36:19,792 INFO [train.py:968] (0/2) Epoch 15, batch 21650, giga_loss[loss=0.2648, simple_loss=0.3317, pruned_loss=0.09896, over 28817.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3427, pruned_loss=0.09837, over 5679355.22 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3516, pruned_loss=0.09656, over 5733025.98 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3415, pruned_loss=0.09774, over 5693893.94 frames. ], batch size: 119, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:36:19,959 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:36:58,599 INFO [train.py:968] (0/2) Epoch 15, batch 21700, giga_loss[loss=0.2923, simple_loss=0.357, pruned_loss=0.1138, over 27542.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3395, pruned_loss=0.09681, over 5688461.46 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3512, pruned_loss=0.09659, over 5735393.59 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3385, pruned_loss=0.0963, over 5696386.61 frames. ], batch size: 472, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:36:59,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.204e+02 1.024e+03 1.430e+03 1.920e+03 7.089e+03, threshold=2.860e+03, percent-clipped=12.0
+2023-03-07 20:37:33,111 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660661.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:37:34,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660664.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:37:39,035 INFO [train.py:968] (0/2) Epoch 15, batch 21750, giga_loss[loss=0.2486, simple_loss=0.3264, pruned_loss=0.08536, over 29062.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3377, pruned_loss=0.09611, over 5701960.50 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3512, pruned_loss=0.09667, over 5737446.69 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3368, pruned_loss=0.09564, over 5706035.07 frames. ], batch size: 155, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:37:56,085 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660693.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:38:15,997 INFO [train.py:968] (0/2) Epoch 15, batch 21800, giga_loss[loss=0.2836, simple_loss=0.3602, pruned_loss=0.1035, over 27931.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3368, pruned_loss=0.0956, over 5705038.85 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3515, pruned_loss=0.09695, over 5739685.11 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3355, pruned_loss=0.09494, over 5705452.48 frames. ], batch size: 412, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:38:16,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.902e+02 9.914e+02 1.283e+03 1.566e+03 6.516e+03, threshold=2.565e+03, percent-clipped=5.0
+2023-03-07 20:38:38,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:38:58,800 INFO [train.py:968] (0/2) Epoch 15, batch 21850, giga_loss[loss=0.3421, simple_loss=0.4043, pruned_loss=0.14, over 27583.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3392, pruned_loss=0.09634, over 5707388.05 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3519, pruned_loss=0.09729, over 5739648.58 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3376, pruned_loss=0.09551, over 5707375.01 frames. ], batch size: 472, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 20:39:38,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3834, 1.6005, 1.4615, 1.3555], device='cuda:0'), covar=tensor([0.2462, 0.2013, 0.1859, 0.2141], device='cuda:0'), in_proj_covar=tensor([0.1835, 0.1743, 0.1688, 0.1809], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 20:39:43,072 INFO [train.py:968] (0/2) Epoch 15, batch 21900, giga_loss[loss=0.2647, simple_loss=0.3485, pruned_loss=0.09042, over 28927.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3418, pruned_loss=0.09728, over 5698464.38 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.352, pruned_loss=0.09748, over 5742008.99 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3404, pruned_loss=0.09644, over 5695940.67 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 20:39:45,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.438e+02 1.093e+03 1.364e+03 1.873e+03 1.015e+04, threshold=2.728e+03, percent-clipped=10.0
+2023-03-07 20:40:10,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:40:14,884 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:40:25,941 INFO [train.py:968] (0/2) Epoch 15, batch 21950, libri_loss[loss=0.3764, simple_loss=0.4243, pruned_loss=0.1643, over 29674.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3441, pruned_loss=0.09783, over 5699341.86 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3523, pruned_loss=0.098, over 5742985.07 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3425, pruned_loss=0.09665, over 5695192.26 frames. ], batch size: 88, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 20:40:38,693 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:40:42,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:41:03,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4253, 1.5119, 1.1540, 1.1323], device='cuda:0'), covar=tensor([0.0766, 0.0455, 0.0966, 0.1119], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0434, 0.0503, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 20:41:05,638 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:41:05,985 INFO [train.py:968] (0/2) Epoch 15, batch 22000, giga_loss[loss=0.287, simple_loss=0.3731, pruned_loss=0.1004, over 28696.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3456, pruned_loss=0.09809, over 5703650.10 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3526, pruned_loss=0.09853, over 5744074.55 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3438, pruned_loss=0.09665, over 5698593.60 frames. ], batch size: 262, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:41:08,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.002e+02 1.040e+03 1.323e+03 2.304e+03 1.563e+04, threshold=2.645e+03, percent-clipped=16.0
+2023-03-07 20:41:19,585 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660934.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:41:22,015 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 20:41:51,146 INFO [train.py:968] (0/2) Epoch 15, batch 22050, libri_loss[loss=0.2588, simple_loss=0.3412, pruned_loss=0.08819, over 29543.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3444, pruned_loss=0.0968, over 5697640.18 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3527, pruned_loss=0.09865, over 5741051.30 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3428, pruned_loss=0.09553, over 5695736.82 frames. ], batch size: 80, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:42:09,946 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:42:11,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:42:32,050 INFO [train.py:968] (0/2) Epoch 15, batch 22100, giga_loss[loss=0.2851, simple_loss=0.355, pruned_loss=0.1076, over 28936.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3458, pruned_loss=0.09829, over 5697077.72 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3528, pruned_loss=0.09894, over 5742233.22 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3442, pruned_loss=0.09697, over 5693543.25 frames. ], batch size: 106, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:42:33,375 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.470e+02 1.127e+03 1.326e+03 1.910e+03 4.769e+03, threshold=2.653e+03, percent-clipped=7.0
+2023-03-07 20:42:36,476 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:43:02,090 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:43:11,952 INFO [train.py:968] (0/2) Epoch 15, batch 22150, giga_loss[loss=0.2496, simple_loss=0.3242, pruned_loss=0.08748, over 28668.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3474, pruned_loss=0.0996, over 5704291.09 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.354, pruned_loss=0.09986, over 5745865.59 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3448, pruned_loss=0.09768, over 5696770.99 frames. ], batch size: 85, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:43:53,600 INFO [train.py:968] (0/2) Epoch 15, batch 22200, giga_loss[loss=0.2796, simple_loss=0.3527, pruned_loss=0.1033, over 28284.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3481, pruned_loss=0.1001, over 5704132.47 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3543, pruned_loss=0.1002, over 5748866.22 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3456, pruned_loss=0.09829, over 5694755.58 frames. ], batch size: 368, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:43:57,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.964e+02 1.163e+03 1.517e+03 2.001e+03 6.471e+03, threshold=3.034e+03, percent-clipped=16.0
+2023-03-07 20:44:15,543 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:44:37,559 INFO [train.py:968] (0/2) Epoch 15, batch 22250, giga_loss[loss=0.2536, simple_loss=0.3326, pruned_loss=0.08733, over 28059.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3515, pruned_loss=0.102, over 5711753.56 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.355, pruned_loss=0.1007, over 5750549.75 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3489, pruned_loss=0.1002, over 5702460.99 frames. ], batch size: 77, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:45:06,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8498, 1.1436, 1.0778, 0.7888], device='cuda:0'), covar=tensor([0.2054, 0.2231, 0.1320, 0.1961], device='cuda:0'), in_proj_covar=tensor([0.1852, 0.1759, 0.1706, 0.1823], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 20:45:18,496 INFO [train.py:968] (0/2) Epoch 15, batch 22300, giga_loss[loss=0.2686, simple_loss=0.3516, pruned_loss=0.09277, over 28942.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3535, pruned_loss=0.1029, over 5703638.58 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3551, pruned_loss=0.1007, over 5741627.00 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3514, pruned_loss=0.1014, over 5704546.17 frames. ], batch size: 112, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:45:19,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.552e+02 1.176e+03 1.435e+03 1.991e+03 6.957e+03, threshold=2.870e+03, percent-clipped=5.0
+2023-03-07 20:45:23,495 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:45:27,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:45:57,277 INFO [train.py:968] (0/2) Epoch 15, batch 22350, giga_loss[loss=0.2551, simple_loss=0.3328, pruned_loss=0.08867, over 28830.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3548, pruned_loss=0.1033, over 5704860.58 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3559, pruned_loss=0.1013, over 5736028.11 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3523, pruned_loss=0.1016, over 5710323.33 frames. ], batch size: 119, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:46:32,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661309.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:46:41,075 INFO [train.py:968] (0/2) Epoch 15, batch 22400, giga_loss[loss=0.2763, simple_loss=0.3499, pruned_loss=0.1014, over 28903.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3539, pruned_loss=0.1025, over 5697650.92 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3561, pruned_loss=0.1014, over 5726784.97 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3518, pruned_loss=0.1012, over 5709368.81 frames. ], batch size: 112, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:46:42,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.375e+02 1.174e+03 1.394e+03 1.785e+03 3.900e+03, threshold=2.789e+03, percent-clipped=3.0
+2023-03-07 20:47:05,583 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.39 vs. limit=5.0
+2023-03-07 20:47:19,628 INFO [train.py:968] (0/2) Epoch 15, batch 22450, giga_loss[loss=0.266, simple_loss=0.3457, pruned_loss=0.0931, over 28956.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3546, pruned_loss=0.1031, over 5706337.70 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3572, pruned_loss=0.1021, over 5730055.35 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3519, pruned_loss=0.1014, over 5712075.06 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:47:22,807 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:47:24,392 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-07 20:47:24,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:47:47,834 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:47:58,273 INFO [train.py:968] (0/2) Epoch 15, batch 22500, libri_loss[loss=0.3029, simple_loss=0.3731, pruned_loss=0.1164, over 29543.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3531, pruned_loss=0.1024, over 5712987.98 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3577, pruned_loss=0.1027, over 5735030.18 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3503, pruned_loss=0.1004, over 5712330.67 frames. ], batch size: 83, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:48:00,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.854e+02 1.253e+03 1.637e+03 2.282e+03 9.829e+03, threshold=3.275e+03, percent-clipped=13.0
+2023-03-07 20:48:10,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:48:23,061 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:48:27,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661452.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:48:29,154 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:48:29,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661455.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:48:31,578 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5349, 1.6509, 1.7917, 1.3582], device='cuda:0'), covar=tensor([0.1761, 0.2305, 0.1471, 0.1630], device='cuda:0'), in_proj_covar=tensor([0.0861, 0.0691, 0.0906, 0.0805], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 20:48:40,543 INFO [train.py:968] (0/2) Epoch 15, batch 22550, libri_loss[loss=0.2872, simple_loss=0.3577, pruned_loss=0.1083, over 29536.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3504, pruned_loss=0.1015, over 5713705.84 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3582, pruned_loss=0.1034, over 5730261.50 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3475, pruned_loss=0.09925, over 5716696.51 frames. ], batch size: 80, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:48:52,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661484.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:49:05,140 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:49:21,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:49:21,673 INFO [train.py:968] (0/2) Epoch 15, batch 22600, giga_loss[loss=0.2911, simple_loss=0.3699, pruned_loss=0.1062, over 28576.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3481, pruned_loss=0.1004, over 5710137.62 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3585, pruned_loss=0.1038, over 5729326.75 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3455, pruned_loss=0.09823, over 5713337.67 frames. ], batch size: 307, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:49:23,639 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.118e+02 1.085e+03 1.288e+03 1.736e+03 5.493e+03, threshold=2.577e+03, percent-clipped=4.0
+2023-03-07 20:49:44,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:49:50,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-07 20:50:02,445 INFO [train.py:968] (0/2) Epoch 15, batch 22650, giga_loss[loss=0.2407, simple_loss=0.3345, pruned_loss=0.07343, over 28897.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.348, pruned_loss=0.09943, over 5705807.04 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3586, pruned_loss=0.104, over 5722512.81 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3455, pruned_loss=0.09741, over 5713987.96 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:50:06,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:50:11,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:50:16,232 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-07 20:50:26,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:50:32,327 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:50:40,621 INFO [train.py:968] (0/2) Epoch 15, batch 22700, giga_loss[loss=0.2485, simple_loss=0.3415, pruned_loss=0.07778, over 29060.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3499, pruned_loss=0.1, over 5709872.61 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3588, pruned_loss=0.1046, over 5729269.84 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3474, pruned_loss=0.09766, over 5709691.03 frames. ], batch size: 164, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:50:44,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.053e+02 1.181e+03 1.434e+03 2.139e+03 8.406e+03, threshold=2.869e+03, percent-clipped=14.0
+2023-03-07 20:51:15,378 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:51:18,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:51:21,209 INFO [train.py:968] (0/2) Epoch 15, batch 22750, giga_loss[loss=0.2881, simple_loss=0.349, pruned_loss=0.1136, over 28927.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3499, pruned_loss=0.09999, over 5710002.18 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3591, pruned_loss=0.1048, over 5721429.42 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3474, pruned_loss=0.09783, over 5716720.82 frames. ], batch size: 106, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:51:42,429 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:52:02,612 INFO [train.py:968] (0/2) Epoch 15, batch 22800, giga_loss[loss=0.2635, simple_loss=0.3258, pruned_loss=0.1006, over 28058.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3487, pruned_loss=0.1001, over 5705313.42 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3594, pruned_loss=0.1052, over 5715107.34 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3463, pruned_loss=0.09799, over 5716540.80 frames. ], batch size: 77, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:52:05,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.379e+02 1.102e+03 1.462e+03 1.894e+03 5.710e+03, threshold=2.924e+03, percent-clipped=10.0
+2023-03-07 20:52:22,254 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:52:24,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:52:43,216 INFO [train.py:968] (0/2) Epoch 15, batch 22850, giga_loss[loss=0.247, simple_loss=0.3163, pruned_loss=0.08884, over 28845.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3467, pruned_loss=0.1004, over 5705335.33 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.36, pruned_loss=0.1056, over 5709047.66 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.344, pruned_loss=0.09822, over 5719172.88 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:52:47,825 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:52:56,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8810, 3.0111, 2.0942, 1.0295], device='cuda:0'), covar=tensor([0.7207, 0.2627, 0.3426, 0.6557], device='cuda:0'), in_proj_covar=tensor([0.1623, 0.1530, 0.1531, 0.1339], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 20:53:14,418 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:53:22,910 INFO [train.py:968] (0/2) Epoch 15, batch 22900, libri_loss[loss=0.3114, simple_loss=0.3799, pruned_loss=0.1214, over 29691.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3466, pruned_loss=0.102, over 5710717.17 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3606, pruned_loss=0.1064, over 5714907.15 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3435, pruned_loss=0.09941, over 5716184.00 frames. ], batch size: 88, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:53:23,801 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:53:24,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.097e+02 1.199e+03 1.440e+03 1.821e+03 6.126e+03, threshold=2.879e+03, percent-clipped=6.0
+2023-03-07 20:53:31,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:53:52,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4022, 3.4927, 1.4270, 1.5533], device='cuda:0'), covar=tensor([0.0950, 0.0354, 0.0929, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0526, 0.0356, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 20:54:03,925 INFO [train.py:968] (0/2) Epoch 15, batch 22950, giga_loss[loss=0.2976, simple_loss=0.3569, pruned_loss=0.1192, over 28940.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3443, pruned_loss=0.1012, over 5712585.16 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3602, pruned_loss=0.1064, over 5712327.82 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3417, pruned_loss=0.09893, over 5718471.57 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:54:06,610 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:54:07,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2254, 1.1693, 3.8876, 3.1350], device='cuda:0'), covar=tensor([0.1697, 0.2757, 0.0443, 0.0890], device='cuda:0'), in_proj_covar=tensor([0.0695, 0.0607, 0.0888, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 20:54:07,874 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:54:11,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2577, 2.1925, 1.6078, 2.0552], device='cuda:0'), covar=tensor([0.0738, 0.0597, 0.0894, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.0369, 0.0434, 0.0502, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 20:54:12,971 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3120, 1.6324, 1.3264, 1.0319], device='cuda:0'), covar=tensor([0.2460, 0.2428, 0.2899, 0.2223], device='cuda:0'), in_proj_covar=tensor([0.1374, 0.1008, 0.1219, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-07 20:54:42,507 INFO [train.py:968] (0/2) Epoch 15, batch 23000, libri_loss[loss=0.317, simple_loss=0.3854, pruned_loss=0.1243, over 20140.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3426, pruned_loss=0.1003, over 5704183.61 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3608, pruned_loss=0.107, over 5708297.49 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3394, pruned_loss=0.09777, over 5713426.23 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:54:44,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:54:45,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.970e+02 1.145e+03 1.445e+03 1.988e+03 5.086e+03, threshold=2.890e+03, percent-clipped=11.0
+2023-03-07 20:54:52,677 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-07 20:55:15,194 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:18,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:19,991 INFO [train.py:968] (0/2) Epoch 15, batch 23050, giga_loss[loss=0.2554, simple_loss=0.318, pruned_loss=0.09636, over 28832.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3384, pruned_loss=0.09812, over 5711600.54 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3605, pruned_loss=0.107, over 5709611.09 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3356, pruned_loss=0.0959, over 5717624.67 frames. ], batch size: 119, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:55:22,076 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:24,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:41,610 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:46,418 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-662000.pt
+2023-03-07 20:55:49,343 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:52,361 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:59,886 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:56:01,408 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-07 20:56:01,527 INFO [train.py:968] (0/2) Epoch 15, batch 23100, giga_loss[loss=0.2354, simple_loss=0.3072, pruned_loss=0.08179, over 28123.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3346, pruned_loss=0.09635, over 5703830.24 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3606, pruned_loss=0.1072, over 5702357.11 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.332, pruned_loss=0.09435, over 5716093.15 frames. ], batch size: 77, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:56:01,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:56:04,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.521e+02 1.189e+03 1.559e+03 2.088e+03 6.916e+03, threshold=3.117e+03, percent-clipped=10.0
+2023-03-07 20:56:06,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3607, 3.5595, 1.5352, 1.4842], device='cuda:0'), covar=tensor([0.0944, 0.0291, 0.0922, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0528, 0.0357, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 20:56:10,396 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 20:56:24,419 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:56:36,679 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:56:39,434 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:56:39,844 INFO [train.py:968] (0/2) Epoch 15, batch 23150, giga_loss[loss=0.2367, simple_loss=0.3233, pruned_loss=0.07504, over 28880.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3343, pruned_loss=0.09605, over 5710514.63 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3609, pruned_loss=0.1074, over 5704791.50 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3317, pruned_loss=0.09412, over 5718099.07 frames. ], batch size: 174, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:57:01,445 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:57:17,341 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:57:18,457 INFO [train.py:968] (0/2) Epoch 15, batch 23200, giga_loss[loss=0.2963, simple_loss=0.3757, pruned_loss=0.1084, over 28871.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3383, pruned_loss=0.09771, over 5710225.92 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3618, pruned_loss=0.1082, over 5710586.93 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3345, pruned_loss=0.09506, over 5711494.35 frames. ], batch size: 136, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:57:22,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.230e+02 1.246e+03 1.474e+03 1.920e+03 4.415e+03, threshold=2.947e+03, percent-clipped=5.0
+2023-03-07 20:57:41,438 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:57:48,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5390, 1.6908, 1.6834, 1.5579], device='cuda:0'), covar=tensor([0.1670, 0.2170, 0.2151, 0.2136], device='cuda:0'), in_proj_covar=tensor([0.0450, 0.0737, 0.0690, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 20:57:58,389 INFO [train.py:968] (0/2) Epoch 15, batch 23250, giga_loss[loss=0.2258, simple_loss=0.3076, pruned_loss=0.07196, over 28487.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3429, pruned_loss=0.09991, over 5700898.11 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3619, pruned_loss=0.1083, over 5704510.75 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3388, pruned_loss=0.09722, over 5707560.32 frames. ], batch size: 71, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:58:01,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5649, 1.7408, 1.4613, 1.5238], device='cuda:0'), covar=tensor([0.1943, 0.2198, 0.2135, 0.2248], device='cuda:0'), in_proj_covar=tensor([0.0450, 0.0737, 0.0690, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 20:58:07,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:58:37,819 INFO [train.py:968] (0/2) Epoch 15, batch 23300, giga_loss[loss=0.2818, simple_loss=0.3611, pruned_loss=0.1012, over 28667.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3452, pruned_loss=0.1002, over 5705055.18 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.362, pruned_loss=0.1084, over 5708237.33 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3416, pruned_loss=0.09789, over 5706993.15 frames. ], batch size: 262, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:58:39,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1138, 3.9070, 3.6847, 1.7832], device='cuda:0'), covar=tensor([0.0587, 0.0732, 0.0710, 0.2097], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.1026, 0.0885, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 20:58:41,886 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:58:42,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.282e+02 1.154e+03 1.412e+03 1.770e+03 4.202e+03, threshold=2.824e+03, percent-clipped=3.0
+2023-03-07 20:58:48,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0981, 3.8964, 3.6686, 1.8891], device='cuda:0'), covar=tensor([0.0585, 0.0726, 0.0676, 0.2133], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.1025, 0.0885, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 20:59:02,781 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:59:18,742 INFO [train.py:968] (0/2) Epoch 15, batch 23350, giga_loss[loss=0.2601, simple_loss=0.3417, pruned_loss=0.08929, over 28949.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3487, pruned_loss=0.1021, over 5716457.41 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3628, pruned_loss=0.1094, over 5715340.83 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3443, pruned_loss=0.09899, over 5711627.31 frames. ], batch size: 106, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:59:53,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-07 21:00:01,553 INFO [train.py:968] (0/2) Epoch 15, batch 23400, giga_loss[loss=0.3229, simple_loss=0.3909, pruned_loss=0.1274, over 28517.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3497, pruned_loss=0.1024, over 5721780.57 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3626, pruned_loss=0.1093, over 5717519.14 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.346, pruned_loss=0.09973, over 5716110.74 frames. ], batch size: 336, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:00:04,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.717e+02 1.181e+03 1.505e+03 2.406e+03 1.034e+04, threshold=3.010e+03, percent-clipped=15.0
+2023-03-07 21:00:06,197 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:00:09,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:00:36,752 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:00:46,791 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:00:47,896 INFO [train.py:968] (0/2) Epoch 15, batch 23450, giga_loss[loss=0.2891, simple_loss=0.363, pruned_loss=0.1076, over 28950.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3531, pruned_loss=0.1057, over 5720585.89 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3622, pruned_loss=0.1093, over 5724286.45 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3502, pruned_loss=0.1034, over 5709814.39 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:01:00,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:01:09,166 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:01:12,371 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:01:14,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:01:38,750 INFO [train.py:968] (0/2) Epoch 15, batch 23500, giga_loss[loss=0.4286, simple_loss=0.4362, pruned_loss=0.2105, over 23773.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3601, pruned_loss=0.1113, over 5697876.05 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.362, pruned_loss=0.1094, over 5718926.01 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3579, pruned_loss=0.1095, over 5693613.19 frames. ], batch size: 705, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:01:42,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.877e+02 1.577e+03 2.108e+03 2.675e+03 8.124e+03, threshold=4.217e+03, percent-clipped=15.0
+2023-03-07 21:01:43,028 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:02:26,112 INFO [train.py:968] (0/2) Epoch 15, batch 23550, giga_loss[loss=0.2925, simple_loss=0.369, pruned_loss=0.108, over 28928.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3663, pruned_loss=0.1157, over 5674527.11 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3621, pruned_loss=0.1096, over 5705012.42 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3644, pruned_loss=0.1141, over 5682129.62 frames. ], batch size: 106, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:02:34,419 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:02:36,712 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-07 21:02:39,886 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:02:48,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:03:02,977 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-07 21:03:05,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2581, 1.3337, 1.1941, 1.5336], device='cuda:0'), covar=tensor([0.0768, 0.0341, 0.0329, 0.0789], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:0')
+2023-03-07 21:03:13,028 INFO [train.py:968] (0/2) Epoch 15, batch 23600, giga_loss[loss=0.3409, simple_loss=0.4044, pruned_loss=0.1387, over 28894.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3721, pruned_loss=0.1202, over 5679500.20 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3623, pruned_loss=0.1099, over 5708957.13 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3705, pruned_loss=0.1189, over 5681552.02 frames. ], batch size: 199, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:03:17,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:03:19,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.018e+02 1.675e+03 2.115e+03 3.080e+03 7.292e+03, threshold=4.229e+03, percent-clipped=11.0
+2023-03-07 21:03:20,001 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:03:23,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:03:53,315 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:04:02,405 INFO [train.py:968] (0/2) Epoch 15, batch 23650, giga_loss[loss=0.4845, simple_loss=0.4806, pruned_loss=0.2442, over 26655.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3795, pruned_loss=0.1268, over 5679008.59 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3623, pruned_loss=0.1099, over 5709493.59 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3786, pruned_loss=0.1261, over 5679209.24 frames. ], batch size: 555, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:04:27,139 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-07 21:04:33,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:04:44,071 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-07 21:04:53,723 INFO [train.py:968] (0/2) Epoch 15, batch 23700, giga_loss[loss=0.2715, simple_loss=0.3499, pruned_loss=0.09657, over 29015.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3839, pruned_loss=0.1307, over 5672078.80 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3621, pruned_loss=0.1097, over 5711448.72 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3835, pruned_loss=0.1304, over 5670206.19 frames. ], batch size: 164, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:04:58,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.980e+02 1.678e+03 2.119e+03 2.737e+03 7.084e+03, threshold=4.238e+03, percent-clipped=10.0
+2023-03-07 21:05:08,349 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:05:11,925 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:05:36,269 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 21:05:37,379 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:05:38,871 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:05:39,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:05:40,357 INFO [train.py:968] (0/2) Epoch 15, batch 23750, giga_loss[loss=0.2974, simple_loss=0.3704, pruned_loss=0.1122, over 28972.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3852, pruned_loss=0.1327, over 5662059.20 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3623, pruned_loss=0.1099, over 5705615.35 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3852, pruned_loss=0.1326, over 5664448.76 frames. ], batch size: 164, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:06:00,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-07 21:06:10,723 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:06:33,183 INFO [train.py:968] (0/2) Epoch 15, batch 23800, giga_loss[loss=0.3946, simple_loss=0.4315, pruned_loss=0.1789, over 27925.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3897, pruned_loss=0.138, over 5653163.47 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3623, pruned_loss=0.1101, over 5710066.93 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3901, pruned_loss=0.1383, over 5650077.87 frames. ], batch size: 412, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:06:38,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.893e+02 1.580e+03 1.968e+03 2.633e+03 7.825e+03, threshold=3.936e+03, percent-clipped=6.0
+2023-03-07 21:06:55,122 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:06:55,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:06:57,813 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:07:21,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:07:24,035 INFO [train.py:968] (0/2) Epoch 15, batch 23850, giga_loss[loss=0.3348, simple_loss=0.3987, pruned_loss=0.1355, over 28982.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3927, pruned_loss=0.1414, over 5644062.96 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3626, pruned_loss=0.1103, over 5708084.62 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3933, pruned_loss=0.142, over 5642081.16 frames. ], batch size: 213, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:07:29,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:08:16,393 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-07 21:08:22,583 INFO [train.py:968] (0/2) Epoch 15, batch 23900, giga_loss[loss=0.3239, simple_loss=0.3795, pruned_loss=0.1342, over 28533.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3948, pruned_loss=0.1424, over 5644255.29 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3629, pruned_loss=0.1106, over 5702537.24 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3957, pruned_loss=0.1433, over 5646194.76 frames. ], batch size: 307, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:08:28,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.409e+02 1.987e+03 2.382e+03 3.895e+03 1.177e+04, threshold=4.764e+03, percent-clipped=23.0
+2023-03-07 21:08:34,180 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:08:56,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:05,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:14,788 INFO [train.py:968] (0/2) Epoch 15, batch 23950, giga_loss[loss=0.361, simple_loss=0.4093, pruned_loss=0.1564, over 28616.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.394, pruned_loss=0.1427, over 5642509.01 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3629, pruned_loss=0.1107, over 5707100.67 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3955, pruned_loss=0.144, over 5638761.48 frames. ], batch size: 242, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:09:30,514 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:33,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:50,391 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:53,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:59,164 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:10:00,203 INFO [train.py:968] (0/2) Epoch 15, batch 24000, giga_loss[loss=0.3308, simple_loss=0.3939, pruned_loss=0.1339, over 28664.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3924, pruned_loss=0.1425, over 5642262.34 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.363, pruned_loss=0.1111, over 5706813.69 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3946, pruned_loss=0.1442, over 5637440.96 frames. ], batch size: 262, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:10:00,208 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 21:10:07,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2892, 3.0272, 1.3876, 1.4821], device='cuda:0'), covar=tensor([0.1097, 0.0404, 0.1015, 0.1478], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0530, 0.0356, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 21:10:09,291 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2103, simple_loss=0.3175, pruned_loss=0.05152, over 944034.00 frames. 
+2023-03-07 21:10:09,292 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 21:10:14,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.765e+02 1.721e+03 2.104e+03 2.787e+03 5.315e+03, threshold=4.208e+03, percent-clipped=5.0
+2023-03-07 21:10:30,261 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:10:54,836 INFO [train.py:968] (0/2) Epoch 15, batch 24050, giga_loss[loss=0.3077, simple_loss=0.382, pruned_loss=0.1167, over 28774.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3922, pruned_loss=0.1413, over 5647261.11 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3629, pruned_loss=0.1111, over 5707694.97 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3945, pruned_loss=0.1433, over 5641462.29 frames. ], batch size: 284, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:10:57,217 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-07 21:11:22,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:11:24,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:11:29,713 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=663002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:11:33,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=663005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:11:47,010 INFO [train.py:968] (0/2) Epoch 15, batch 24100, libri_loss[loss=0.323, simple_loss=0.3827, pruned_loss=0.1316, over 29535.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3911, pruned_loss=0.1396, over 5642602.60 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3632, pruned_loss=0.1113, over 5708930.78 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3934, pruned_loss=0.1416, over 5635302.24 frames. ], batch size: 84, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:11:53,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.928e+02 1.677e+03 1.990e+03 3.010e+03 7.919e+03, threshold=3.979e+03, percent-clipped=10.0
+2023-03-07 21:11:53,749 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:12:00,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:12:31,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=663063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:12:37,095 INFO [train.py:968] (0/2) Epoch 15, batch 24150, giga_loss[loss=0.3165, simple_loss=0.3755, pruned_loss=0.1287, over 27962.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3924, pruned_loss=0.1406, over 5631028.78 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3631, pruned_loss=0.1113, over 5711262.38 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3945, pruned_loss=0.1425, over 5622708.03 frames. ], batch size: 412, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:13:28,942 INFO [train.py:968] (0/2) Epoch 15, batch 24200, giga_loss[loss=0.3171, simple_loss=0.3913, pruned_loss=0.1215, over 28774.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3902, pruned_loss=0.1381, over 5633022.53 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3637, pruned_loss=0.1118, over 5714855.92 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3922, pruned_loss=0.14, over 5621203.14 frames. ], batch size: 119, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:13:35,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.738e+03 2.261e+03 2.838e+03 5.182e+03, threshold=4.521e+03, percent-clipped=7.0
+2023-03-07 21:13:42,651 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=663133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:14:19,165 INFO [train.py:968] (0/2) Epoch 15, batch 24250, giga_loss[loss=0.337, simple_loss=0.378, pruned_loss=0.148, over 23866.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3877, pruned_loss=0.1348, over 5631272.72 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3638, pruned_loss=0.1121, over 5708956.07 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3896, pruned_loss=0.1364, over 5625196.94 frames. ], batch size: 705, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:14:55,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=663206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:15:07,197 INFO [train.py:968] (0/2) Epoch 15, batch 24300, giga_loss[loss=0.2932, simple_loss=0.371, pruned_loss=0.1077, over 28637.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3844, pruned_loss=0.1314, over 5651294.65 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3639, pruned_loss=0.1123, over 5711316.02 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3863, pruned_loss=0.1329, over 5642893.92 frames. ], batch size: 307, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:15:12,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.190e+03 1.805e+03 2.797e+03 4.359e+03 1.569e+04, threshold=5.594e+03, percent-clipped=21.0
+2023-03-07 21:15:35,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4439, 3.5910, 1.5619, 1.7832], device='cuda:0'), covar=tensor([0.1001, 0.0349, 0.0876, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0531, 0.0358, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 21:15:52,138 INFO [train.py:968] (0/2) Epoch 15, batch 24350, giga_loss[loss=0.3041, simple_loss=0.3731, pruned_loss=0.1175, over 28598.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3811, pruned_loss=0.1284, over 5661315.14 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3637, pruned_loss=0.1125, over 5707351.55 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3833, pruned_loss=0.13, over 5657529.18 frames. ], batch size: 307, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:16:24,968 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-07 21:16:40,699 INFO [train.py:968] (0/2) Epoch 15, batch 24400, giga_loss[loss=0.2802, simple_loss=0.3524, pruned_loss=0.104, over 28695.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3777, pruned_loss=0.1263, over 5648380.32 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3637, pruned_loss=0.1125, over 5701789.07 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3797, pruned_loss=0.1278, over 5649609.69 frames. ], batch size: 262, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:16:44,684 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=663323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:16:47,825 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 1.657e+03 2.299e+03 3.434e+03 1.141e+04, threshold=4.598e+03, percent-clipped=6.0
+2023-03-07 21:17:12,824 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=663349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:17:15,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=663352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:17:16,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3639, 1.5591, 1.6559, 1.2328], device='cuda:0'), covar=tensor([0.1336, 0.2085, 0.1106, 0.1345], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0691, 0.0901, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 21:17:30,588 INFO [train.py:968] (0/2) Epoch 15, batch 24450, giga_loss[loss=0.3692, simple_loss=0.424, pruned_loss=0.1573, over 28963.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3785, pruned_loss=0.1269, over 5655873.74 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3639, pruned_loss=0.1128, over 5701090.08 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.38, pruned_loss=0.128, over 5656525.42 frames. ], batch size: 164, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:17:45,487 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:18:13,800 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 21:18:24,680 INFO [train.py:968] (0/2) Epoch 15, batch 24500, libri_loss[loss=0.2315, simple_loss=0.3009, pruned_loss=0.08108, over 29676.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3779, pruned_loss=0.1261, over 5665160.38 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3639, pruned_loss=0.113, over 5704509.76 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3795, pruned_loss=0.1271, over 5661694.26 frames. ], batch size: 69, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:18:31,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.975e+02 1.364e+03 1.730e+03 2.295e+03 7.139e+03, threshold=3.460e+03, percent-clipped=3.0
+2023-03-07 21:18:43,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=663438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:18:54,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5629, 1.7839, 1.4284, 1.5622], device='cuda:0'), covar=tensor([0.2656, 0.2675, 0.3030, 0.2485], device='cuda:0'), in_proj_covar=tensor([0.1390, 0.1022, 0.1234, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 21:19:01,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4489, 1.6839, 1.3885, 1.4901], device='cuda:0'), covar=tensor([0.2127, 0.2116, 0.2240, 0.2032], device='cuda:0'), in_proj_covar=tensor([0.1388, 0.1021, 0.1233, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 21:19:12,493 INFO [train.py:968] (0/2) Epoch 15, batch 24550, giga_loss[loss=0.319, simple_loss=0.3923, pruned_loss=0.1229, over 28878.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3774, pruned_loss=0.1248, over 5665915.25 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3646, pruned_loss=0.1137, over 5703290.27 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3786, pruned_loss=0.1254, over 5663114.36 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:19:50,916 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=663508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:20:03,844 INFO [train.py:968] (0/2) Epoch 15, batch 24600, libri_loss[loss=0.2904, simple_loss=0.3553, pruned_loss=0.1128, over 29584.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3769, pruned_loss=0.1223, over 5671463.02 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3639, pruned_loss=0.1133, over 5708313.59 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3789, pruned_loss=0.1234, over 5663644.98 frames. ], batch size: 76, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:20:10,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.094e+02 1.517e+03 1.840e+03 2.355e+03 5.616e+03, threshold=3.680e+03, percent-clipped=11.0
+2023-03-07 21:20:56,044 INFO [train.py:968] (0/2) Epoch 15, batch 24650, giga_loss[loss=0.2979, simple_loss=0.3695, pruned_loss=0.1131, over 28716.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3777, pruned_loss=0.1227, over 5659802.37 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3638, pruned_loss=0.1134, over 5702198.84 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3795, pruned_loss=0.1237, over 5657704.64 frames. ], batch size: 242, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:20:57,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6171, 1.6419, 1.6928, 1.3591], device='cuda:0'), covar=tensor([0.2897, 0.2396, 0.1709, 0.2347], device='cuda:0'), in_proj_covar=tensor([0.1830, 0.1746, 0.1684, 0.1805], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 21:21:10,805 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=663581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:21:12,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=663584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:21:42,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:21:46,705 INFO [train.py:968] (0/2) Epoch 15, batch 24700, giga_loss[loss=0.3063, simple_loss=0.3794, pruned_loss=0.1166, over 28838.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3779, pruned_loss=0.1233, over 5650062.75 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3637, pruned_loss=0.1133, over 5693638.21 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3796, pruned_loss=0.1242, over 5655472.37 frames. ], batch size: 174, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:21:56,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 1.604e+03 2.042e+03 3.145e+03 7.750e+03, threshold=4.085e+03, percent-clipped=16.0
+2023-03-07 21:22:12,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6204, 1.7450, 1.3003, 1.3715], device='cuda:0'), covar=tensor([0.0840, 0.0569, 0.1023, 0.1071], device='cuda:0'), in_proj_covar=tensor([0.0370, 0.0437, 0.0503, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 21:22:21,559 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=663651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:22:22,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3806, 4.2025, 4.0010, 1.7450], device='cuda:0'), covar=tensor([0.0579, 0.0724, 0.0734, 0.2069], device='cuda:0'), in_proj_covar=tensor([0.1135, 0.1049, 0.0903, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 21:22:23,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=663654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:22:36,961 INFO [train.py:968] (0/2) Epoch 15, batch 24750, giga_loss[loss=0.3078, simple_loss=0.3734, pruned_loss=0.1211, over 28889.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3773, pruned_loss=0.124, over 5634616.10 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.364, pruned_loss=0.1136, over 5688959.22 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3786, pruned_loss=0.1247, over 5642558.81 frames. ], batch size: 213, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:22:37,224 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=663669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:22:50,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:23:03,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=663698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:23:22,073 INFO [train.py:968] (0/2) Epoch 15, batch 24800, giga_loss[loss=0.353, simple_loss=0.4001, pruned_loss=0.153, over 28585.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3751, pruned_loss=0.1236, over 5654051.38 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3644, pruned_loss=0.114, over 5693191.93 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.376, pruned_loss=0.1239, over 5655730.18 frames. ], batch size: 307, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:23:29,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.919e+02 1.557e+03 2.004e+03 2.596e+03 7.102e+03, threshold=4.008e+03, percent-clipped=6.0
+2023-03-07 21:23:58,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3313, 1.7634, 1.2511, 0.7319], device='cuda:0'), covar=tensor([0.3449, 0.1959, 0.2291, 0.4651], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1551, 0.1539, 0.1340], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 21:24:02,713 INFO [train.py:968] (0/2) Epoch 15, batch 24850, giga_loss[loss=0.2615, simple_loss=0.3343, pruned_loss=0.09437, over 28817.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3753, pruned_loss=0.1242, over 5663831.79 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3649, pruned_loss=0.1145, over 5702550.81 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.376, pruned_loss=0.1244, over 5655333.52 frames. ], batch size: 112, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:24:46,386 INFO [train.py:968] (0/2) Epoch 15, batch 24900, giga_loss[loss=0.3119, simple_loss=0.3855, pruned_loss=0.1191, over 28674.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3744, pruned_loss=0.1221, over 5677131.33 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3651, pruned_loss=0.1148, over 5701371.26 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.375, pruned_loss=0.1222, over 5670798.31 frames. ], batch size: 307, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:24:56,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.604e+02 1.554e+03 2.061e+03 2.831e+03 6.785e+03, threshold=4.121e+03, percent-clipped=9.0
+2023-03-07 21:25:08,949 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=663841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:25:13,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=663844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:25:34,673 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=663864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:25:39,113 INFO [train.py:968] (0/2) Epoch 15, batch 24950, giga_loss[loss=0.2925, simple_loss=0.3672, pruned_loss=0.1089, over 28945.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3737, pruned_loss=0.1216, over 5664678.95 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3648, pruned_loss=0.1146, over 5702520.07 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3744, pruned_loss=0.1219, over 5658510.31 frames. ], batch size: 213, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:25:42,691 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:26:21,593 INFO [train.py:968] (0/2) Epoch 15, batch 25000, giga_loss[loss=0.3049, simple_loss=0.3769, pruned_loss=0.1164, over 28969.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3733, pruned_loss=0.1206, over 5678857.91 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3643, pruned_loss=0.1142, over 5708600.32 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3746, pruned_loss=0.1213, over 5667411.33 frames. ], batch size: 213, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:26:29,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.706e+02 1.569e+03 2.102e+03 2.994e+03 7.077e+03, threshold=4.204e+03, percent-clipped=11.0
+2023-03-07 21:26:59,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9226, 4.7336, 4.4948, 2.2976], device='cuda:0'), covar=tensor([0.0446, 0.0581, 0.0626, 0.1859], device='cuda:0'), in_proj_covar=tensor([0.1148, 0.1060, 0.0912, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 21:27:08,137 INFO [train.py:968] (0/2) Epoch 15, batch 25050, giga_loss[loss=0.2887, simple_loss=0.3576, pruned_loss=0.11, over 28951.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3728, pruned_loss=0.121, over 5672311.55 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3645, pruned_loss=0.1146, over 5695925.84 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3739, pruned_loss=0.1215, over 5673934.90 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:27:39,713 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-664000.pt
+2023-03-07 21:27:57,998 INFO [train.py:968] (0/2) Epoch 15, batch 25100, giga_loss[loss=0.2565, simple_loss=0.334, pruned_loss=0.08951, over 28954.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3723, pruned_loss=0.1215, over 5676437.66 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.365, pruned_loss=0.1152, over 5700678.88 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.373, pruned_loss=0.1214, over 5672881.54 frames. ], batch size: 164, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:28:05,245 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.148e+03 1.636e+03 2.221e+03 2.701e+03 7.957e+03, threshold=4.443e+03, percent-clipped=13.0
+2023-03-07 21:28:18,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:28:31,625 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-07 21:28:41,042 INFO [train.py:968] (0/2) Epoch 15, batch 25150, giga_loss[loss=0.3224, simple_loss=0.379, pruned_loss=0.1329, over 28917.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3732, pruned_loss=0.1229, over 5687205.00 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.366, pruned_loss=0.1159, over 5706128.71 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3732, pruned_loss=0.1226, over 5678668.30 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:29:03,871 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 21:29:04,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2114, 1.1647, 4.0126, 3.1630], device='cuda:0'), covar=tensor([0.1630, 0.2690, 0.0439, 0.0867], device='cuda:0'), in_proj_covar=tensor([0.0704, 0.0614, 0.0904, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 21:29:29,162 INFO [train.py:968] (0/2) Epoch 15, batch 25200, giga_loss[loss=0.2959, simple_loss=0.3665, pruned_loss=0.1127, over 28857.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3719, pruned_loss=0.1224, over 5697375.32 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.366, pruned_loss=0.116, over 5709894.26 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3719, pruned_loss=0.1221, over 5687046.99 frames. ], batch size: 174, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:29:36,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.568e+02 1.619e+03 2.058e+03 2.853e+03 7.898e+03, threshold=4.115e+03, percent-clipped=7.0
+2023-03-07 21:30:15,058 INFO [train.py:968] (0/2) Epoch 15, batch 25250, giga_loss[loss=0.3153, simple_loss=0.3767, pruned_loss=0.1269, over 28983.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3695, pruned_loss=0.1213, over 5687753.10 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3661, pruned_loss=0.1161, over 5703907.09 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3695, pruned_loss=0.1211, over 5684777.39 frames. ], batch size: 213, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:30:27,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3462, 3.5903, 1.5347, 1.4326], device='cuda:0'), covar=tensor([0.1008, 0.0358, 0.0877, 0.1384], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0533, 0.0358, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 21:30:33,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=664187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:30:36,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=664190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:31:04,757 INFO [train.py:968] (0/2) Epoch 15, batch 25300, giga_loss[loss=0.2862, simple_loss=0.3557, pruned_loss=0.1083, over 28817.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3694, pruned_loss=0.1221, over 5687855.09 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3659, pruned_loss=0.116, over 5708409.29 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3696, pruned_loss=0.1221, over 5680824.34 frames. ], batch size: 174, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:31:04,964 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=664219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:31:15,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.671e+03 2.332e+03 3.487e+03 8.575e+03, threshold=4.665e+03, percent-clipped=14.0
+2023-03-07 21:31:28,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:31:57,360 INFO [train.py:968] (0/2) Epoch 15, batch 25350, giga_loss[loss=0.3232, simple_loss=0.3841, pruned_loss=0.1312, over 28696.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3708, pruned_loss=0.1227, over 5684073.96 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3658, pruned_loss=0.116, over 5709695.04 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3712, pruned_loss=0.1228, over 5677005.65 frames. ], batch size: 262, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:32:01,622 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664275.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:32:29,428 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:32:40,081 INFO [train.py:968] (0/2) Epoch 15, batch 25400, giga_loss[loss=0.28, simple_loss=0.3547, pruned_loss=0.1027, over 28902.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3704, pruned_loss=0.1212, over 5693446.09 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3656, pruned_loss=0.1159, over 5713194.37 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3709, pruned_loss=0.1215, over 5684593.51 frames. ], batch size: 213, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:32:51,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.710e+02 1.613e+03 2.092e+03 2.721e+03 7.944e+03, threshold=4.184e+03, percent-clipped=3.0
+2023-03-07 21:33:22,318 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9312, 3.7145, 3.5000, 1.8163], device='cuda:0'), covar=tensor([0.0729, 0.0895, 0.0935, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.1150, 0.1062, 0.0914, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 21:33:28,480 INFO [train.py:968] (0/2) Epoch 15, batch 25450, giga_loss[loss=0.2797, simple_loss=0.3587, pruned_loss=0.1004, over 28906.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3705, pruned_loss=0.1203, over 5684792.47 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3659, pruned_loss=0.1161, over 5705168.94 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3707, pruned_loss=0.1204, over 5684706.16 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:33:44,592 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=664382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:33:47,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=664385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:33:57,653 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 21:34:12,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=664414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:34:15,214 INFO [train.py:968] (0/2) Epoch 15, batch 25500, giga_loss[loss=0.2935, simple_loss=0.3598, pruned_loss=0.1136, over 28622.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3709, pruned_loss=0.1208, over 5682350.38 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3661, pruned_loss=0.1163, over 5707082.12 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.371, pruned_loss=0.1208, over 5679865.55 frames. ], batch size: 92, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:34:25,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.990e+02 1.588e+03 2.089e+03 2.638e+03 6.504e+03, threshold=4.177e+03, percent-clipped=5.0
+2023-03-07 21:35:08,090 INFO [train.py:968] (0/2) Epoch 15, batch 25550, giga_loss[loss=0.3583, simple_loss=0.384, pruned_loss=0.1663, over 23837.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3748, pruned_loss=0.1243, over 5677566.89 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.366, pruned_loss=0.1163, over 5706766.23 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.375, pruned_loss=0.1245, over 5675794.40 frames. ], batch size: 705, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:35:47,002 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5689, 1.5193, 1.6751, 1.2531], device='cuda:0'), covar=tensor([0.1679, 0.2641, 0.1347, 0.1548], device='cuda:0'), in_proj_covar=tensor([0.0861, 0.0696, 0.0905, 0.0804], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 21:35:47,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-07 21:35:57,753 INFO [train.py:968] (0/2) Epoch 15, batch 25600, giga_loss[loss=0.2887, simple_loss=0.3448, pruned_loss=0.1163, over 28807.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3763, pruned_loss=0.1269, over 5670108.47 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3661, pruned_loss=0.1164, over 5698667.04 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3765, pruned_loss=0.127, over 5675059.47 frames. ], batch size: 99, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:36:08,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.066e+02 1.903e+03 2.516e+03 3.809e+03 1.911e+04, threshold=5.031e+03, percent-clipped=23.0
+2023-03-07 21:36:38,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:36:44,840 INFO [train.py:968] (0/2) Epoch 15, batch 25650, giga_loss[loss=0.2978, simple_loss=0.3662, pruned_loss=0.1147, over 28887.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3773, pruned_loss=0.1288, over 5670607.20 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3661, pruned_loss=0.1163, over 5701489.48 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3778, pruned_loss=0.1293, over 5670727.10 frames. ], batch size: 213, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:36:47,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:37:18,847 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-07 21:37:30,473 INFO [train.py:968] (0/2) Epoch 15, batch 25700, giga_loss[loss=0.303, simple_loss=0.3648, pruned_loss=0.1206, over 28859.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3781, pruned_loss=0.1296, over 5685233.94 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3659, pruned_loss=0.1164, over 5706548.51 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.379, pruned_loss=0.1303, over 5679956.63 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:37:32,786 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:37:39,492 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.228e+03 1.555e+03 2.095e+03 2.930e+03 6.975e+03, threshold=4.190e+03, percent-clipped=1.0
+2023-03-07 21:37:51,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6091, 1.6455, 1.9115, 1.4430], device='cuda:0'), covar=tensor([0.1575, 0.1950, 0.1241, 0.1723], device='cuda:0'), in_proj_covar=tensor([0.0858, 0.0695, 0.0904, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 21:37:57,184 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664650.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:38:13,627 INFO [train.py:968] (0/2) Epoch 15, batch 25750, giga_loss[loss=0.3293, simple_loss=0.3865, pruned_loss=0.136, over 28748.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3768, pruned_loss=0.1292, over 5679567.82 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3657, pruned_loss=0.1163, over 5712597.94 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.378, pruned_loss=0.1302, over 5669345.13 frames. ], batch size: 262, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:38:23,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6031, 2.1160, 1.7709, 1.3741], device='cuda:0'), covar=tensor([0.2994, 0.2015, 0.2398, 0.2859], device='cuda:0'), in_proj_covar=tensor([0.1837, 0.1767, 0.1690, 0.1822], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 21:38:26,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:38:57,336 INFO [train.py:968] (0/2) Epoch 15, batch 25800, giga_loss[loss=0.3811, simple_loss=0.4221, pruned_loss=0.1701, over 27603.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3772, pruned_loss=0.1286, over 5679888.36 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3661, pruned_loss=0.1168, over 5710184.57 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3782, pruned_loss=0.1293, over 5672676.52 frames. ], batch size: 472, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:39:07,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.958e+02 1.672e+03 2.596e+03 3.716e+03 9.000e+03, threshold=5.191e+03, percent-clipped=22.0
+2023-03-07 21:39:13,507 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:39:32,099 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5019, 1.6860, 1.7260, 1.2989], device='cuda:0'), covar=tensor([0.1866, 0.2303, 0.1482, 0.1688], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0696, 0.0904, 0.0804], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 21:39:43,360 INFO [train.py:968] (0/2) Epoch 15, batch 25850, giga_loss[loss=0.3445, simple_loss=0.3993, pruned_loss=0.1449, over 28712.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3766, pruned_loss=0.1271, over 5677236.34 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3666, pruned_loss=0.1172, over 5713033.05 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3772, pruned_loss=0.1276, over 5668495.69 frames. ], batch size: 262, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:40:07,063 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=664793.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:40:09,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=664796.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:40:30,813 INFO [train.py:968] (0/2) Epoch 15, batch 25900, giga_loss[loss=0.329, simple_loss=0.3626, pruned_loss=0.1477, over 24058.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.374, pruned_loss=0.1255, over 5661083.13 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3668, pruned_loss=0.1174, over 5706117.32 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3743, pruned_loss=0.1258, over 5660296.97 frames. ], batch size: 705, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:40:36,073 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=664824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:40:36,766 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=664825.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:40:38,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=664827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:40:41,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.508e+03 2.052e+03 2.551e+03 1.208e+04, threshold=4.105e+03, percent-clipped=4.0
+2023-03-07 21:41:06,900 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=664856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:41:19,556 INFO [train.py:968] (0/2) Epoch 15, batch 25950, giga_loss[loss=0.3214, simple_loss=0.3561, pruned_loss=0.1433, over 23553.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3714, pruned_loss=0.1243, over 5663769.67 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3668, pruned_loss=0.1175, over 5707206.73 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3717, pruned_loss=0.1245, over 5661950.37 frames. ], batch size: 705, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:41:22,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4117, 1.6639, 1.6577, 1.2498], device='cuda:0'), covar=tensor([0.1621, 0.2358, 0.1288, 0.1540], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0696, 0.0904, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 21:41:27,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6387, 1.6415, 1.2453, 1.2399], device='cuda:0'), covar=tensor([0.0763, 0.0583, 0.0988, 0.1156], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0443, 0.0506, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 21:42:13,195 INFO [train.py:968] (0/2) Epoch 15, batch 26000, giga_loss[loss=0.3202, simple_loss=0.3843, pruned_loss=0.1281, over 28917.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.373, pruned_loss=0.1263, over 5652001.96 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3667, pruned_loss=0.1175, over 5709263.38 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3733, pruned_loss=0.1265, over 5648397.71 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:42:20,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.573e+03 2.092e+03 2.965e+03 8.434e+03, threshold=4.185e+03, percent-clipped=12.0
+2023-03-07 21:42:24,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:42:39,843 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:42:55,047 INFO [train.py:968] (0/2) Epoch 15, batch 26050, giga_loss[loss=0.3217, simple_loss=0.3878, pruned_loss=0.1278, over 28526.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3755, pruned_loss=0.127, over 5653218.43 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3672, pruned_loss=0.1179, over 5703507.93 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3755, pruned_loss=0.1271, over 5654426.35 frames. ], batch size: 336, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:43:23,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:43:31,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3106, 2.7653, 1.4227, 1.4510], device='cuda:0'), covar=tensor([0.0916, 0.0412, 0.0913, 0.1262], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0531, 0.0358, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 21:43:43,280 INFO [train.py:968] (0/2) Epoch 15, batch 26100, giga_loss[loss=0.364, simple_loss=0.4203, pruned_loss=0.1539, over 28044.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3794, pruned_loss=0.1269, over 5660748.23 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3671, pruned_loss=0.1179, over 5706706.75 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3798, pruned_loss=0.1271, over 5657959.24 frames. ], batch size: 412, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:43:53,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.552e+03 2.152e+03 2.950e+03 6.441e+03, threshold=4.303e+03, percent-clipped=9.0
+2023-03-07 21:44:05,392 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:44:11,028 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3319, 1.6177, 1.4429, 1.5518], device='cuda:0'), covar=tensor([0.0796, 0.0324, 0.0332, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:0')
+2023-03-07 21:44:12,331 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:44:29,725 INFO [train.py:968] (0/2) Epoch 15, batch 26150, giga_loss[loss=0.3433, simple_loss=0.3844, pruned_loss=0.1512, over 23687.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3799, pruned_loss=0.1264, over 5661111.34 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3673, pruned_loss=0.1182, over 5712078.54 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3804, pruned_loss=0.1266, over 5652628.44 frames. ], batch size: 705, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:44:38,318 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:44:40,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:44:42,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5078, 2.1861, 1.5022, 0.6250], device='cuda:0'), covar=tensor([0.5074, 0.2462, 0.3717, 0.6073], device='cuda:0'), in_proj_covar=tensor([0.1617, 0.1551, 0.1520, 0.1331], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 21:44:50,429 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:44:52,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:45:11,478 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:45:12,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=665112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:45:13,470 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2222, 1.2318, 3.5471, 3.0503], device='cuda:0'), covar=tensor([0.1601, 0.2652, 0.0514, 0.1092], device='cuda:0'), in_proj_covar=tensor([0.0706, 0.0617, 0.0906, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 21:45:19,620 INFO [train.py:968] (0/2) Epoch 15, batch 26200, giga_loss[loss=0.313, simple_loss=0.3808, pruned_loss=0.1226, over 28822.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3813, pruned_loss=0.1278, over 5653981.58 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3673, pruned_loss=0.1182, over 5710618.92 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3818, pruned_loss=0.1279, over 5648000.43 frames. ], batch size: 119, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:45:23,660 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:45:30,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.426e+02 1.603e+03 2.201e+03 2.797e+03 6.918e+03, threshold=4.402e+03, percent-clipped=7.0
+2023-03-07 21:45:38,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:45:41,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:46:00,555 INFO [train.py:968] (0/2) Epoch 15, batch 26250, giga_loss[loss=0.3206, simple_loss=0.3822, pruned_loss=0.1295, over 28721.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3813, pruned_loss=0.128, over 5666073.52 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3672, pruned_loss=0.1182, over 5713661.60 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.382, pruned_loss=0.1283, over 5657707.58 frames. ], batch size: 284, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:46:05,823 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:46:49,295 INFO [train.py:968] (0/2) Epoch 15, batch 26300, giga_loss[loss=0.2713, simple_loss=0.3422, pruned_loss=0.1002, over 29040.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.381, pruned_loss=0.1289, over 5660049.07 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3677, pruned_loss=0.1186, over 5716913.17 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3814, pruned_loss=0.129, over 5649387.97 frames. ], batch size: 155, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:47:00,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.859e+02 1.615e+03 2.076e+03 3.148e+03 9.447e+03, threshold=4.151e+03, percent-clipped=12.0
+2023-03-07 21:47:18,602 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:47:19,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6154, 1.5031, 1.2652, 1.1886], device='cuda:0'), covar=tensor([0.0879, 0.0620, 0.1091, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0443, 0.0506, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 21:47:26,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:47:27,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:47:36,549 INFO [train.py:968] (0/2) Epoch 15, batch 26350, giga_loss[loss=0.2635, simple_loss=0.3425, pruned_loss=0.09224, over 28803.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3783, pruned_loss=0.1277, over 5646512.12 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3676, pruned_loss=0.1186, over 5711176.56 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3791, pruned_loss=0.128, over 5641924.21 frames. ], batch size: 174, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:47:40,291 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665273.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:47:52,459 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:48:21,220 INFO [train.py:968] (0/2) Epoch 15, batch 26400, giga_loss[loss=0.3634, simple_loss=0.4054, pruned_loss=0.1607, over 29047.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3771, pruned_loss=0.1275, over 5650275.97 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3684, pruned_loss=0.1192, over 5706001.95 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3772, pruned_loss=0.1275, over 5650312.24 frames. ], batch size: 155, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 21:48:34,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.173e+03 1.705e+03 2.122e+03 3.178e+03 1.111e+04, threshold=4.243e+03, percent-clipped=16.0
+2023-03-07 21:48:45,878 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 21:49:12,911 INFO [train.py:968] (0/2) Epoch 15, batch 26450, giga_loss[loss=0.3083, simple_loss=0.3738, pruned_loss=0.1214, over 28722.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3757, pruned_loss=0.1275, over 5646842.34 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3677, pruned_loss=0.1187, over 5710158.51 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3766, pruned_loss=0.1281, over 5641944.41 frames. ], batch size: 262, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 21:49:55,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=665418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:49:56,153 INFO [train.py:968] (0/2) Epoch 15, batch 26500, giga_loss[loss=0.3114, simple_loss=0.3822, pruned_loss=0.1203, over 28992.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3778, pruned_loss=0.1297, over 5655319.78 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3673, pruned_loss=0.1186, over 5715971.09 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3791, pruned_loss=0.1305, over 5644391.29 frames. ], batch size: 145, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:50:02,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=665425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:50:08,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.866e+03 2.277e+03 3.055e+03 6.817e+03, threshold=4.554e+03, percent-clipped=11.0
+2023-03-07 21:50:17,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5628, 1.7605, 1.5683, 1.5694], device='cuda:0'), covar=tensor([0.1696, 0.2060, 0.2272, 0.1992], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0742, 0.0692, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 21:50:41,807 INFO [train.py:968] (0/2) Epoch 15, batch 26550, giga_loss[loss=0.3284, simple_loss=0.3669, pruned_loss=0.1449, over 23643.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3759, pruned_loss=0.1286, over 5663717.21 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3671, pruned_loss=0.1185, over 5722129.05 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3774, pruned_loss=0.1297, over 5647888.60 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:50:58,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4951, 4.3182, 4.1169, 1.8128], device='cuda:0'), covar=tensor([0.0578, 0.0689, 0.0769, 0.2098], device='cuda:0'), in_proj_covar=tensor([0.1149, 0.1060, 0.0915, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 21:51:24,540 INFO [train.py:968] (0/2) Epoch 15, batch 26600, giga_loss[loss=0.299, simple_loss=0.3625, pruned_loss=0.1178, over 28616.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3738, pruned_loss=0.1272, over 5677759.32 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3671, pruned_loss=0.1188, over 5722475.70 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3752, pruned_loss=0.1281, over 5663287.54 frames. ], batch size: 242, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:51:39,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.486e+02 1.698e+03 2.253e+03 2.757e+03 7.202e+03, threshold=4.507e+03, percent-clipped=2.0
+2023-03-07 21:52:08,614 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:52:11,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:52:14,315 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:52:14,604 INFO [train.py:968] (0/2) Epoch 15, batch 26650, giga_loss[loss=0.2909, simple_loss=0.37, pruned_loss=0.1059, over 28963.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.374, pruned_loss=0.1268, over 5679162.74 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3673, pruned_loss=0.1189, over 5724424.27 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3751, pruned_loss=0.1275, over 5665640.79 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:52:16,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:52:38,382 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:52:42,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5524, 4.3748, 4.1310, 2.0596], device='cuda:0'), covar=tensor([0.0485, 0.0628, 0.0661, 0.2151], device='cuda:0'), in_proj_covar=tensor([0.1152, 0.1062, 0.0916, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 21:52:44,955 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:53:00,083 INFO [train.py:968] (0/2) Epoch 15, batch 26700, giga_loss[loss=0.2937, simple_loss=0.3627, pruned_loss=0.1123, over 28633.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3761, pruned_loss=0.1272, over 5670013.44 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3674, pruned_loss=0.119, over 5718642.41 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1278, over 5663450.62 frames. ], batch size: 284, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:53:02,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=665620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:53:03,980 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-07 21:53:12,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.027e+02 1.474e+03 1.860e+03 2.547e+03 1.198e+04, threshold=3.720e+03, percent-clipped=4.0
+2023-03-07 21:53:28,551 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=665648.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:53:51,800 INFO [train.py:968] (0/2) Epoch 15, batch 26750, libri_loss[loss=0.2844, simple_loss=0.3618, pruned_loss=0.1035, over 29123.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3764, pruned_loss=0.1278, over 5650999.12 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3677, pruned_loss=0.1192, over 5710584.73 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.377, pruned_loss=0.1282, over 5651332.73 frames. ], batch size: 101, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:54:36,638 INFO [train.py:968] (0/2) Epoch 15, batch 26800, giga_loss[loss=0.2976, simple_loss=0.3894, pruned_loss=0.1029, over 29000.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3773, pruned_loss=0.1276, over 5652775.68 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3677, pruned_loss=0.1192, over 5704217.48 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3779, pruned_loss=0.1281, over 5658488.08 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 21:54:48,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.742e+03 2.318e+03 3.368e+03 1.013e+04, threshold=4.637e+03, percent-clipped=18.0
+2023-03-07 21:55:15,751 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:55:16,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9440, 2.5957, 0.9980, 1.2401], device='cuda:0'), covar=tensor([0.1328, 0.0479, 0.1196, 0.1658], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0533, 0.0358, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 21:55:18,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:55:20,420 INFO [train.py:968] (0/2) Epoch 15, batch 26850, giga_loss[loss=0.2921, simple_loss=0.3797, pruned_loss=0.1022, over 28839.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3781, pruned_loss=0.1252, over 5665852.31 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3675, pruned_loss=0.1191, over 5705553.61 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3789, pruned_loss=0.1257, over 5668501.87 frames. ], batch size: 119, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:55:24,564 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-07 21:55:41,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665791.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:55:44,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665794.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:55:44,659 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:56:04,883 INFO [train.py:968] (0/2) Epoch 15, batch 26900, giga_loss[loss=0.3149, simple_loss=0.3819, pruned_loss=0.1239, over 28629.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3801, pruned_loss=0.1251, over 5664066.41 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3677, pruned_loss=0.1193, over 5702568.30 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3809, pruned_loss=0.1256, over 5667761.24 frames. ], batch size: 307, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:56:10,717 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665823.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:56:19,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.192e+02 1.528e+03 1.935e+03 2.539e+03 9.816e+03, threshold=3.871e+03, percent-clipped=4.0
+2023-03-07 21:56:51,162 INFO [train.py:968] (0/2) Epoch 15, batch 26950, giga_loss[loss=0.3823, simple_loss=0.4206, pruned_loss=0.172, over 28947.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3829, pruned_loss=0.127, over 5666974.69 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.368, pruned_loss=0.1195, over 5701093.12 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3834, pruned_loss=0.1272, over 5670835.55 frames. ], batch size: 106, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:57:22,369 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665899.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:57:41,432 INFO [train.py:968] (0/2) Epoch 15, batch 27000, giga_loss[loss=0.3117, simple_loss=0.3685, pruned_loss=0.1275, over 28908.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3868, pruned_loss=0.1313, over 5660562.70 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3685, pruned_loss=0.12, over 5694722.61 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3871, pruned_loss=0.1312, over 5669164.03 frames. ], batch size: 112, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:57:41,437 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 21:57:48,319 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1160, 1.5796, 1.6363, 1.3454], device='cuda:0'), covar=tensor([0.1944, 0.1447, 0.1953, 0.1744], device='cuda:0'), in_proj_covar=tensor([0.0452, 0.0742, 0.0691, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 21:57:49,831 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.21, simple_loss=0.3163, pruned_loss=0.05185, over 944034.00 frames. 
+2023-03-07 21:57:49,831 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 21:58:03,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.646e+03 2.182e+03 2.904e+03 7.174e+03, threshold=4.364e+03, percent-clipped=11.0
+2023-03-07 21:58:17,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-07 21:58:39,115 INFO [train.py:968] (0/2) Epoch 15, batch 27050, giga_loss[loss=0.2991, simple_loss=0.3639, pruned_loss=0.1171, over 28731.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3873, pruned_loss=0.1327, over 5672669.61 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3684, pruned_loss=0.1199, over 5698428.50 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3879, pruned_loss=0.1329, over 5675535.67 frames. ], batch size: 284, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:58:41,949 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:59:08,017 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-666000.pt
+2023-03-07 21:59:26,149 INFO [train.py:968] (0/2) Epoch 15, batch 27100, giga_loss[loss=0.3078, simple_loss=0.3861, pruned_loss=0.1148, over 28930.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3866, pruned_loss=0.1326, over 5664725.85 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3684, pruned_loss=0.12, over 5696568.06 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3877, pruned_loss=0.1332, over 5668228.04 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:59:40,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.662e+03 2.133e+03 3.159e+03 7.693e+03, threshold=4.266e+03, percent-clipped=9.0
+2023-03-07 22:00:05,133 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 22:00:14,787 INFO [train.py:968] (0/2) Epoch 15, batch 27150, giga_loss[loss=0.2892, simple_loss=0.3662, pruned_loss=0.1061, over 28744.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3825, pruned_loss=0.1287, over 5670946.83 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3677, pruned_loss=0.1197, over 5699893.94 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3843, pruned_loss=0.1296, over 5670389.67 frames. ], batch size: 262, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:00:33,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=666088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:01:04,222 INFO [train.py:968] (0/2) Epoch 15, batch 27200, giga_loss[loss=0.3022, simple_loss=0.3773, pruned_loss=0.1136, over 28978.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.382, pruned_loss=0.1268, over 5658297.78 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3678, pruned_loss=0.1197, over 5699948.00 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3835, pruned_loss=0.1275, over 5657469.35 frames. ], batch size: 136, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:01:17,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.899e+02 1.418e+03 1.920e+03 2.508e+03 5.695e+03, threshold=3.839e+03, percent-clipped=5.0
+2023-03-07 22:01:18,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2319, 1.8016, 1.3908, 0.4190], device='cuda:0'), covar=tensor([0.3967, 0.2603, 0.3958, 0.5290], device='cuda:0'), in_proj_covar=tensor([0.1635, 0.1555, 0.1532, 0.1336], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 22:01:49,584 INFO [train.py:968] (0/2) Epoch 15, batch 27250, giga_loss[loss=0.3122, simple_loss=0.38, pruned_loss=0.1222, over 28617.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.382, pruned_loss=0.1261, over 5661858.12 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3676, pruned_loss=0.1198, over 5693929.07 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3837, pruned_loss=0.1268, over 5665886.33 frames. ], batch size: 307, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:02:37,091 INFO [train.py:968] (0/2) Epoch 15, batch 27300, giga_loss[loss=0.3425, simple_loss=0.3956, pruned_loss=0.1447, over 27529.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3834, pruned_loss=0.1279, over 5649933.51 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3679, pruned_loss=0.12, over 5692213.49 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3849, pruned_loss=0.1285, over 5653444.79 frames. ], batch size: 472, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:02:53,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.295e+02 1.567e+03 2.201e+03 3.316e+03 9.293e+03, threshold=4.402e+03, percent-clipped=19.0
+2023-03-07 22:02:54,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6349, 4.7394, 1.9448, 1.7838], device='cuda:0'), covar=tensor([0.0987, 0.0287, 0.0818, 0.1323], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0532, 0.0358, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 22:03:23,216 INFO [train.py:968] (0/2) Epoch 15, batch 27350, libri_loss[loss=0.3292, simple_loss=0.3821, pruned_loss=0.1382, over 19170.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3819, pruned_loss=0.1275, over 5645163.76 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.368, pruned_loss=0.1203, over 5681447.68 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3834, pruned_loss=0.1279, over 5657746.66 frames. ], batch size: 186, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:03:26,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=666274.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 22:04:08,941 INFO [train.py:968] (0/2) Epoch 15, batch 27400, giga_loss[loss=0.3403, simple_loss=0.3927, pruned_loss=0.1439, over 28625.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3814, pruned_loss=0.1278, over 5658184.18 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3687, pruned_loss=0.1205, over 5689503.77 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3825, pruned_loss=0.1282, over 5659836.33 frames. ], batch size: 336, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:04:22,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.177e+03 1.697e+03 2.080e+03 2.424e+03 6.910e+03, threshold=4.160e+03, percent-clipped=6.0
+2023-03-07 22:04:35,308 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=666346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:04:49,622 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-07 22:04:57,332 INFO [train.py:968] (0/2) Epoch 15, batch 27450, giga_loss[loss=0.274, simple_loss=0.3385, pruned_loss=0.1047, over 28543.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3795, pruned_loss=0.1271, over 5659750.31 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3691, pruned_loss=0.1208, over 5682506.07 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3803, pruned_loss=0.1273, over 5666780.01 frames. ], batch size: 85, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:05:47,629 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=666417.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:05:49,564 INFO [train.py:968] (0/2) Epoch 15, batch 27500, giga_loss[loss=0.304, simple_loss=0.3658, pruned_loss=0.1211, over 28332.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3771, pruned_loss=0.1257, over 5662055.57 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3692, pruned_loss=0.1207, over 5687214.30 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3778, pruned_loss=0.1259, over 5663029.83 frames. ], batch size: 78, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:05:50,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=666420.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 22:06:01,787 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-07 22:06:01,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.488e+02 1.671e+03 2.169e+03 3.343e+03 1.025e+04, threshold=4.337e+03, percent-clipped=16.0
+2023-03-07 22:06:16,329 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=666449.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:06:29,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=666463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:06:34,139 INFO [train.py:968] (0/2) Epoch 15, batch 27550, giga_loss[loss=0.4409, simple_loss=0.4508, pruned_loss=0.2155, over 26684.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.377, pruned_loss=0.1267, over 5666955.27 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3692, pruned_loss=0.1208, over 5694307.27 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3778, pruned_loss=0.127, over 5660374.61 frames. ], batch size: 555, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:06:51,499 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=666489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:06:53,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=666492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:07:17,356 INFO [train.py:968] (0/2) Epoch 15, batch 27600, giga_loss[loss=0.2513, simple_loss=0.3272, pruned_loss=0.08765, over 28928.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3757, pruned_loss=0.1262, over 5650649.29 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3693, pruned_loss=0.1208, over 5680299.68 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3764, pruned_loss=0.1267, over 5657945.20 frames. ], batch size: 112, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:07:18,835 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=666521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:07:27,597 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.810e+03 2.397e+03 3.263e+03 1.249e+04, threshold=4.794e+03, percent-clipped=13.0
+2023-03-07 22:07:56,421 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 22:08:01,789 INFO [train.py:968] (0/2) Epoch 15, batch 27650, giga_loss[loss=0.3045, simple_loss=0.3745, pruned_loss=0.1173, over 27966.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3728, pruned_loss=0.1225, over 5652179.54 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3696, pruned_loss=0.1209, over 5680550.36 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3733, pruned_loss=0.1229, over 5657278.01 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:08:30,914 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 22:08:33,141 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=666606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:08:36,574 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=666609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:08:45,573 INFO [train.py:968] (0/2) Epoch 15, batch 27700, libri_loss[loss=0.3627, simple_loss=0.4121, pruned_loss=0.1566, over 29561.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3706, pruned_loss=0.1204, over 5645329.11 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3699, pruned_loss=0.1212, over 5669463.30 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3707, pruned_loss=0.1204, over 5658376.13 frames. ], batch size: 89, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:09:00,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-07 22:09:02,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.485e+02 1.323e+03 1.888e+03 2.974e+03 8.465e+03, threshold=3.775e+03, percent-clipped=6.0
+2023-03-07 22:09:04,863 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=666638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:09:28,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3930, 3.4834, 1.6669, 1.5641], device='cuda:0'), covar=tensor([0.0904, 0.0340, 0.0838, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0528, 0.0356, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:0')
+2023-03-07 22:09:31,699 INFO [train.py:968] (0/2) Epoch 15, batch 27750, giga_loss[loss=0.2753, simple_loss=0.3477, pruned_loss=0.1014, over 28802.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3693, pruned_loss=0.1194, over 5639745.88 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3698, pruned_loss=0.1211, over 5662975.70 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3694, pruned_loss=0.1194, over 5655365.95 frames. ], batch size: 99, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:10:12,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=666708.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 22:10:23,643 INFO [train.py:968] (0/2) Epoch 15, batch 27800, giga_loss[loss=0.2669, simple_loss=0.3492, pruned_loss=0.0923, over 28983.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3674, pruned_loss=0.1192, over 5641562.83 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3697, pruned_loss=0.121, over 5669385.39 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5647736.17 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:10:43,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.717e+03 2.324e+03 3.365e+03 5.990e+03, threshold=4.648e+03, percent-clipped=15.0
+2023-03-07 22:10:55,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2293, 1.4786, 1.4984, 1.2938], device='cuda:0'), covar=tensor([0.1708, 0.1666, 0.2168, 0.1800], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0739, 0.0690, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 22:11:15,539 INFO [train.py:968] (0/2) Epoch 15, batch 27850, giga_loss[loss=0.2874, simple_loss=0.3614, pruned_loss=0.1067, over 29030.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3661, pruned_loss=0.1189, over 5650703.40 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3705, pruned_loss=0.1213, over 5674886.09 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3655, pruned_loss=0.1186, over 5650080.11 frames. ], batch size: 155, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:11:20,820 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=666773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:12:00,613 INFO [train.py:968] (0/2) Epoch 15, batch 27900, giga_loss[loss=0.2666, simple_loss=0.3559, pruned_loss=0.08867, over 29001.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3689, pruned_loss=0.1201, over 5651502.38 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3705, pruned_loss=0.1213, over 5669431.72 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3683, pruned_loss=0.1199, over 5655877.14 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:12:05,417 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 22:12:08,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4628, 1.6474, 1.6935, 1.2799], device='cuda:0'), covar=tensor([0.1788, 0.2493, 0.1419, 0.1659], device='cuda:0'), in_proj_covar=tensor([0.0861, 0.0700, 0.0910, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 22:12:17,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.487e+02 1.648e+03 2.095e+03 2.667e+03 7.631e+03, threshold=4.190e+03, percent-clipped=2.0
+2023-03-07 22:12:40,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5835, 1.8001, 1.2077, 1.4127], device='cuda:0'), covar=tensor([0.0866, 0.0529, 0.1040, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0441, 0.0507, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 22:12:47,114 INFO [train.py:968] (0/2) Epoch 15, batch 27950, giga_loss[loss=0.2797, simple_loss=0.3552, pruned_loss=0.1021, over 29117.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3702, pruned_loss=0.121, over 5647902.37 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.37, pruned_loss=0.1208, over 5677360.48 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3701, pruned_loss=0.1212, over 5643708.81 frames. ], batch size: 155, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:13:30,138 INFO [train.py:968] (0/2) Epoch 15, batch 28000, giga_loss[loss=0.3164, simple_loss=0.3823, pruned_loss=0.1253, over 28529.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3699, pruned_loss=0.1204, over 5645964.41 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3699, pruned_loss=0.1208, over 5669845.67 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3699, pruned_loss=0.1206, over 5648714.45 frames. ], batch size: 336, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:13:40,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6579, 3.4970, 3.3050, 2.0174], device='cuda:0'), covar=tensor([0.0584, 0.0694, 0.0676, 0.1683], device='cuda:0'), in_proj_covar=tensor([0.1149, 0.1060, 0.0915, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 22:13:45,106 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.118e+02 1.609e+03 2.001e+03 2.785e+03 1.091e+04, threshold=4.003e+03, percent-clipped=7.0
+2023-03-07 22:14:17,799 INFO [train.py:968] (0/2) Epoch 15, batch 28050, giga_loss[loss=0.2898, simple_loss=0.3585, pruned_loss=0.1106, over 28906.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3696, pruned_loss=0.1208, over 5643128.98 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3696, pruned_loss=0.1206, over 5673585.08 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3699, pruned_loss=0.1211, over 5641583.97 frames. ], batch size: 112, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:14:42,190 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-07 22:14:58,982 INFO [train.py:968] (0/2) Epoch 15, batch 28100, giga_loss[loss=0.2846, simple_loss=0.356, pruned_loss=0.1066, over 29116.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3708, pruned_loss=0.1218, over 5647212.98 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3704, pruned_loss=0.121, over 5676157.86 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3703, pruned_loss=0.1217, over 5643246.71 frames. ], batch size: 128, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:15:09,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7634, 1.8819, 1.3603, 1.3862], device='cuda:0'), covar=tensor([0.0842, 0.0587, 0.0985, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0440, 0.0506, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 22:15:15,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.199e+02 1.534e+03 1.917e+03 2.451e+03 5.763e+03, threshold=3.835e+03, percent-clipped=4.0
+2023-03-07 22:15:18,728 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=667038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:15:32,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7596, 1.9532, 1.9919, 1.5396], device='cuda:0'), covar=tensor([0.1902, 0.2417, 0.1551, 0.1762], device='cuda:0'), in_proj_covar=tensor([0.0862, 0.0702, 0.0911, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 22:15:47,807 INFO [train.py:968] (0/2) Epoch 15, batch 28150, giga_loss[loss=0.2845, simple_loss=0.3501, pruned_loss=0.1095, over 28463.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3729, pruned_loss=0.123, over 5642070.42 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3707, pruned_loss=0.1212, over 5668971.37 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3722, pruned_loss=0.1227, over 5644757.19 frames. ], batch size: 78, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:15:53,789 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7585, 1.8998, 1.8092, 1.6133], device='cuda:0'), covar=tensor([0.2290, 0.2170, 0.1763, 0.2013], device='cuda:0'), in_proj_covar=tensor([0.1824, 0.1758, 0.1680, 0.1812], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 22:16:01,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=667083.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 22:16:13,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3633, 1.8216, 1.4138, 1.4776], device='cuda:0'), covar=tensor([0.0729, 0.0316, 0.0313, 0.0791], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:0')
+2023-03-07 22:16:34,676 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2902, 1.5496, 1.3939, 1.5202], device='cuda:0'), covar=tensor([0.0742, 0.0373, 0.0322, 0.0776], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:0')
+2023-03-07 22:16:35,793 INFO [train.py:968] (0/2) Epoch 15, batch 28200, giga_loss[loss=0.3373, simple_loss=0.3977, pruned_loss=0.1385, over 28664.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3754, pruned_loss=0.1254, over 5647651.85 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3708, pruned_loss=0.1214, over 5674791.03 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3749, pruned_loss=0.125, over 5643998.98 frames. ], batch size: 85, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:16:52,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.769e+03 2.139e+03 2.925e+03 8.251e+03, threshold=4.278e+03, percent-clipped=13.0
+2023-03-07 22:17:04,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=667148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:17:22,415 INFO [train.py:968] (0/2) Epoch 15, batch 28250, giga_loss[loss=0.3252, simple_loss=0.3862, pruned_loss=0.132, over 28721.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3772, pruned_loss=0.1275, over 5644911.65 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.371, pruned_loss=0.1216, over 5671691.81 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3767, pruned_loss=0.1271, over 5644189.20 frames. ], batch size: 242, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:18:04,975 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=667208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:18:14,921 INFO [train.py:968] (0/2) Epoch 15, batch 28300, giga_loss[loss=0.2997, simple_loss=0.3569, pruned_loss=0.1213, over 23955.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3772, pruned_loss=0.1277, over 5641606.28 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1218, over 5671623.91 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.377, pruned_loss=0.1275, over 5639934.47 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:18:22,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=667226.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:18:25,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=667229.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 22:18:31,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.707e+03 2.064e+03 2.580e+03 6.535e+03, threshold=4.128e+03, percent-clipped=6.0
+2023-03-07 22:18:41,761 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-07 22:18:55,199 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=667258.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 22:19:07,616 INFO [train.py:968] (0/2) Epoch 15, batch 28350, giga_loss[loss=0.3188, simple_loss=0.3778, pruned_loss=0.1299, over 28823.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3773, pruned_loss=0.1261, over 5643193.02 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.371, pruned_loss=0.122, over 5663817.32 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3771, pruned_loss=0.1258, over 5648601.38 frames. ], batch size: 106, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:19:15,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2217, 1.4315, 1.4808, 1.2424], device='cuda:0'), covar=tensor([0.1502, 0.1569, 0.1878, 0.1693], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0738, 0.0688, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 22:19:28,232 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=667291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:19:30,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=667294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:19:32,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0404, 1.2482, 1.2193, 0.9939], device='cuda:0'), covar=tensor([0.1746, 0.1918, 0.1041, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.1831, 0.1759, 0.1680, 0.1810], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 22:19:39,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4128, 1.5191, 1.5792, 1.4332], device='cuda:0'), covar=tensor([0.1255, 0.1285, 0.1481, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0738, 0.0688, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 22:19:51,283 INFO [train.py:968] (0/2) Epoch 15, batch 28400, giga_loss[loss=0.3135, simple_loss=0.3813, pruned_loss=0.1228, over 28988.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3778, pruned_loss=0.1271, over 5637920.86 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3709, pruned_loss=0.1219, over 5670357.14 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3781, pruned_loss=0.1271, over 5635350.46 frames. ], batch size: 136, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:19:53,856 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 22:19:55,370 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=667323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:20:11,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.953e+02 1.588e+03 2.282e+03 2.951e+03 1.067e+04, threshold=4.563e+03, percent-clipped=9.0
+2023-03-07 22:20:43,249 INFO [train.py:968] (0/2) Epoch 15, batch 28450, giga_loss[loss=0.327, simple_loss=0.3658, pruned_loss=0.1441, over 23746.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3776, pruned_loss=0.1275, over 5624891.47 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3708, pruned_loss=0.1219, over 5666135.13 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.378, pruned_loss=0.1276, over 5625639.97 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:20:52,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=667376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:21:09,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 22:21:19,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5118, 1.6447, 1.2242, 1.2699], device='cuda:0'), covar=tensor([0.0822, 0.0539, 0.1017, 0.1067], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0442, 0.0507, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 22:21:32,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=667413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:21:38,190 INFO [train.py:968] (0/2) Epoch 15, batch 28500, giga_loss[loss=0.3178, simple_loss=0.3779, pruned_loss=0.1289, over 28571.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3769, pruned_loss=0.1278, over 5629289.14 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3709, pruned_loss=0.1221, over 5669624.78 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3774, pruned_loss=0.1279, over 5625743.55 frames. ], batch size: 336, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:21:59,492 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.887e+02 1.583e+03 2.052e+03 2.947e+03 6.130e+03, threshold=4.103e+03, percent-clipped=7.0
+2023-03-07 22:22:31,466 INFO [train.py:968] (0/2) Epoch 15, batch 28550, giga_loss[loss=0.2743, simple_loss=0.3411, pruned_loss=0.1038, over 28714.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3752, pruned_loss=0.1274, over 5620963.61 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3708, pruned_loss=0.1221, over 5661651.06 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3757, pruned_loss=0.1276, over 5625530.32 frames. ], batch size: 71, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:23:13,406 INFO [train.py:968] (0/2) Epoch 15, batch 28600, giga_loss[loss=0.307, simple_loss=0.3782, pruned_loss=0.118, over 29080.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3747, pruned_loss=0.1272, over 5628569.93 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3707, pruned_loss=0.1221, over 5648187.44 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3754, pruned_loss=0.1275, over 5643051.50 frames. ], batch size: 155, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:23:25,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2641, 1.1737, 1.1896, 1.4025], device='cuda:0'), covar=tensor([0.0766, 0.0366, 0.0338, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:0')
+2023-03-07 22:23:34,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.578e+02 1.628e+03 2.460e+03 3.425e+03 1.120e+04, threshold=4.921e+03, percent-clipped=16.0
+2023-03-07 22:23:50,498 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=667556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:23:54,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=667559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:24:03,217 INFO [train.py:968] (0/2) Epoch 15, batch 28650, giga_loss[loss=0.3008, simple_loss=0.3575, pruned_loss=0.1221, over 28576.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3735, pruned_loss=0.1268, over 5627582.23 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3704, pruned_loss=0.1219, over 5649863.26 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3743, pruned_loss=0.1273, over 5637134.27 frames. ], batch size: 85, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:24:14,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=667583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:24:19,471 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=667588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:24:32,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4341, 1.5524, 1.5721, 1.3982], device='cuda:0'), covar=tensor([0.2215, 0.1946, 0.1488, 0.1834], device='cuda:0'), in_proj_covar=tensor([0.1831, 0.1763, 0.1683, 0.1816], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 22:24:48,098 INFO [train.py:968] (0/2) Epoch 15, batch 28700, libri_loss[loss=0.3079, simple_loss=0.375, pruned_loss=0.1204, over 27888.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3742, pruned_loss=0.1268, over 5641378.92 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3704, pruned_loss=0.1218, over 5649617.07 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.375, pruned_loss=0.1274, over 5648817.84 frames. ], batch size: 116, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:25:06,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.578e+03 1.934e+03 2.771e+03 6.288e+03, threshold=3.868e+03, percent-clipped=4.0
+2023-03-07 22:25:31,186 INFO [train.py:968] (0/2) Epoch 15, batch 28750, giga_loss[loss=0.3147, simple_loss=0.3792, pruned_loss=0.1251, over 28916.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3757, pruned_loss=0.1278, over 5652904.46 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3699, pruned_loss=0.1216, over 5655839.94 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3769, pruned_loss=0.1286, over 5652952.37 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:25:51,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2948, 1.4598, 3.1032, 2.9289], device='cuda:0'), covar=tensor([0.1276, 0.2258, 0.0484, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.0706, 0.0615, 0.0904, 0.0822], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 22:26:18,593 INFO [train.py:968] (0/2) Epoch 15, batch 28800, giga_loss[loss=0.2946, simple_loss=0.3604, pruned_loss=0.1144, over 28967.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.377, pruned_loss=0.1289, over 5642915.87 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3695, pruned_loss=0.1215, over 5642147.56 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3785, pruned_loss=0.1298, over 5654593.75 frames. ], batch size: 145, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:26:26,470 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=667726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:26:28,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=667729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:26:37,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.179e+03 1.945e+03 2.494e+03 3.517e+03 8.654e+03, threshold=4.988e+03, percent-clipped=20.0
+2023-03-07 22:26:50,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=667751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:26:54,885 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=667758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:26:56,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1662, 4.9613, 4.7207, 2.2990], device='cuda:0'), covar=tensor([0.0482, 0.0610, 0.0655, 0.1891], device='cuda:0'), in_proj_covar=tensor([0.1146, 0.1061, 0.0915, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 22:27:05,513 INFO [train.py:968] (0/2) Epoch 15, batch 28850, giga_loss[loss=0.3421, simple_loss=0.3975, pruned_loss=0.1434, over 28575.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3769, pruned_loss=0.1289, over 5655333.51 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3693, pruned_loss=0.1213, over 5644181.13 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3784, pruned_loss=0.1298, over 5663007.00 frames. ], batch size: 307, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:27:27,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4255, 1.5412, 1.5783, 1.5100], device='cuda:0'), covar=tensor([0.1352, 0.1476, 0.1606, 0.1454], device='cuda:0'), in_proj_covar=tensor([0.0454, 0.0747, 0.0696, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 22:27:49,177 INFO [train.py:968] (0/2) Epoch 15, batch 28900, giga_loss[loss=0.3099, simple_loss=0.3755, pruned_loss=0.1221, over 28945.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3761, pruned_loss=0.1282, over 5660594.18 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3698, pruned_loss=0.1217, over 5647814.52 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3771, pruned_loss=0.1289, over 5663608.10 frames. ], batch size: 136, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:28:05,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.104e+02 1.730e+03 2.216e+03 3.024e+03 7.899e+03, threshold=4.431e+03, percent-clipped=2.0
+2023-03-07 22:28:33,801 INFO [train.py:968] (0/2) Epoch 15, batch 28950, libri_loss[loss=0.2664, simple_loss=0.3383, pruned_loss=0.09721, over 29554.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3769, pruned_loss=0.128, over 5670199.03 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3698, pruned_loss=0.1217, over 5652301.77 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3779, pruned_loss=0.1288, over 5668867.25 frames. ], batch size: 77, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:28:57,048 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=667894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:28:59,793 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=667897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:29:12,027 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.49 vs. limit=5.0
+2023-03-07 22:29:22,818 INFO [train.py:968] (0/2) Epoch 15, batch 29000, giga_loss[loss=0.3354, simple_loss=0.3883, pruned_loss=0.1413, over 27903.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.377, pruned_loss=0.1283, over 5660836.55 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3693, pruned_loss=0.1214, over 5647395.22 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3783, pruned_loss=0.1293, over 5664117.15 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:29:23,783 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=667920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:29:28,291 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=667926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:29:38,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.123e+02 1.705e+03 2.167e+03 2.843e+03 8.361e+03, threshold=4.334e+03, percent-clipped=4.0
+2023-03-07 22:30:05,483 INFO [train.py:968] (0/2) Epoch 15, batch 29050, giga_loss[loss=0.3086, simple_loss=0.3726, pruned_loss=0.1223, over 28799.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3773, pruned_loss=0.1286, over 5673924.47 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1213, over 5655318.57 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3789, pruned_loss=0.1297, over 5669856.30 frames. ], batch size: 119, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:30:34,831 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-668000.pt
+2023-03-07 22:30:50,030 INFO [train.py:968] (0/2) Epoch 15, batch 29100, giga_loss[loss=0.2917, simple_loss=0.3636, pruned_loss=0.1099, over 29007.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3804, pruned_loss=0.1312, over 5667893.82 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1213, over 5657145.36 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3818, pruned_loss=0.1323, over 5663266.24 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:30:53,094 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=668023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:30:59,722 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-07 22:31:07,007 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.563e+03 1.873e+03 2.649e+03 5.140e+03, threshold=3.746e+03, percent-clipped=3.0
+2023-03-07 22:31:19,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6676, 1.9405, 1.3161, 1.5098], device='cuda:0'), covar=tensor([0.0846, 0.0539, 0.1007, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0446, 0.0510, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-07 22:31:32,867 INFO [train.py:968] (0/2) Epoch 15, batch 29150, giga_loss[loss=0.2905, simple_loss=0.3611, pruned_loss=0.11, over 29030.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3792, pruned_loss=0.1301, over 5664074.68 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3693, pruned_loss=0.1213, over 5654015.04 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3806, pruned_loss=0.1312, over 5662648.25 frames. ], batch size: 106, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:32:06,198 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=668107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:32:08,202 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.85 vs. limit=2.0
+2023-03-07 22:32:18,503 INFO [train.py:968] (0/2) Epoch 15, batch 29200, giga_loss[loss=0.3313, simple_loss=0.3859, pruned_loss=0.1383, over 27540.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3787, pruned_loss=0.1289, over 5650622.27 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3685, pruned_loss=0.1207, over 5660137.37 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3808, pruned_loss=0.1306, over 5644135.22 frames. ], batch size: 472, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:32:40,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.907e+02 1.470e+03 2.007e+03 2.615e+03 4.707e+03, threshold=4.014e+03, percent-clipped=11.0
+2023-03-07 22:33:12,058 INFO [train.py:968] (0/2) Epoch 15, batch 29250, giga_loss[loss=0.3652, simple_loss=0.4117, pruned_loss=0.1593, over 27590.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3797, pruned_loss=0.1291, over 5642812.13 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3686, pruned_loss=0.1209, over 5652848.84 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3813, pruned_loss=0.1303, over 5644550.59 frames. ], batch size: 472, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:33:23,801 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 22:33:55,164 INFO [train.py:968] (0/2) Epoch 15, batch 29300, giga_loss[loss=0.3426, simple_loss=0.3743, pruned_loss=0.1555, over 23651.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.378, pruned_loss=0.1277, over 5650664.13 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3687, pruned_loss=0.1211, over 5660370.97 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3794, pruned_loss=0.1287, over 5645178.99 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:34:10,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.965e+02 1.527e+03 1.998e+03 2.615e+03 9.887e+03, threshold=3.996e+03, percent-clipped=6.0
+2023-03-07 22:34:36,495 INFO [train.py:968] (0/2) Epoch 15, batch 29350, libri_loss[loss=0.3068, simple_loss=0.3723, pruned_loss=0.1207, over 29529.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3764, pruned_loss=0.1262, over 5653527.69 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3691, pruned_loss=0.1214, over 5654295.57 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3775, pruned_loss=0.127, over 5654392.45 frames. ], batch size: 83, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:35:00,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=668295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:35:24,336 INFO [train.py:968] (0/2) Epoch 15, batch 29400, giga_loss[loss=0.3421, simple_loss=0.3932, pruned_loss=0.1455, over 27680.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.376, pruned_loss=0.1262, over 5650151.12 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3691, pruned_loss=0.1214, over 5656390.39 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.377, pruned_loss=0.1269, over 5648824.94 frames. ], batch size: 472, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:35:42,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.686e+02 1.559e+03 1.907e+03 2.670e+03 6.912e+03, threshold=3.814e+03, percent-clipped=8.0
+2023-03-07 22:36:10,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5408, 1.8117, 1.4039, 1.7295], device='cuda:0'), covar=tensor([0.2436, 0.2440, 0.2824, 0.2149], device='cuda:0'), in_proj_covar=tensor([0.1399, 0.1025, 0.1236, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 22:36:14,487 INFO [train.py:968] (0/2) Epoch 15, batch 29450, giga_loss[loss=0.3056, simple_loss=0.3579, pruned_loss=0.1267, over 28591.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3767, pruned_loss=0.1268, over 5655402.74 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3686, pruned_loss=0.1209, over 5661830.15 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3782, pruned_loss=0.1279, over 5649146.11 frames. ], batch size: 85, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:36:33,935 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6658, 1.9531, 1.9019, 1.4503], device='cuda:0'), covar=tensor([0.1515, 0.2361, 0.1368, 0.1653], device='cuda:0'), in_proj_covar=tensor([0.0857, 0.0698, 0.0907, 0.0804], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 22:36:41,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=668398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:37:02,396 INFO [train.py:968] (0/2) Epoch 15, batch 29500, giga_loss[loss=0.3622, simple_loss=0.4033, pruned_loss=0.1606, over 27577.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.377, pruned_loss=0.1278, over 5637734.80 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3686, pruned_loss=0.1209, over 5643691.55 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3782, pruned_loss=0.1287, over 5649235.53 frames. ], batch size: 472, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:37:20,394 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=668438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:37:21,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.482e+03 1.990e+03 2.790e+03 7.407e+03, threshold=3.980e+03, percent-clipped=12.0
+2023-03-07 22:37:23,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=668441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:37:23,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5026, 5.3268, 5.0514, 2.3441], device='cuda:0'), covar=tensor([0.0378, 0.0502, 0.0578, 0.1840], device='cuda:0'), in_proj_covar=tensor([0.1159, 0.1073, 0.0926, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 22:37:47,200 INFO [train.py:968] (0/2) Epoch 15, batch 29550, giga_loss[loss=0.2949, simple_loss=0.3639, pruned_loss=0.113, over 28815.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3758, pruned_loss=0.1271, over 5640093.92 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3686, pruned_loss=0.1211, over 5629704.19 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1279, over 5662035.04 frames. ], batch size: 119, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:37:50,282 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=668470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:37:58,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=668482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:38:05,884 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=668491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:38:09,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4226, 1.5695, 1.6848, 1.2800], device='cuda:0'), covar=tensor([0.1430, 0.2329, 0.1206, 0.1588], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0694, 0.0903, 0.0801], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 22:38:31,217 INFO [train.py:968] (0/2) Epoch 15, batch 29600, giga_loss[loss=0.3219, simple_loss=0.3855, pruned_loss=0.1291, over 28742.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3776, pruned_loss=0.1287, over 5643390.76 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3683, pruned_loss=0.1208, over 5636348.16 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3791, pruned_loss=0.1298, over 5655487.00 frames. ], batch size: 99, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:38:50,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.768e+02 1.471e+03 1.919e+03 2.540e+03 6.702e+03, threshold=3.838e+03, percent-clipped=6.0
+2023-03-07 22:38:51,827 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=668541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:38:54,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4065, 1.8526, 1.3856, 1.5242], device='cuda:0'), covar=tensor([0.2540, 0.2427, 0.2800, 0.2122], device='cuda:0'), in_proj_covar=tensor([0.1400, 0.1026, 0.1237, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 22:38:54,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=668544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:39:17,846 INFO [train.py:968] (0/2) Epoch 15, batch 29650, libri_loss[loss=0.321, simple_loss=0.388, pruned_loss=0.127, over 29297.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3797, pruned_loss=0.1302, over 5643468.50 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.369, pruned_loss=0.1211, over 5642422.52 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3807, pruned_loss=0.1311, over 5647889.10 frames. ], batch size: 94, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:39:20,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=668573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:39:58,464 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-07 22:40:03,540 INFO [train.py:968] (0/2) Epoch 15, batch 29700, giga_loss[loss=0.2857, simple_loss=0.3561, pruned_loss=0.1076, over 28838.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3786, pruned_loss=0.1293, over 5646261.57 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3688, pruned_loss=0.121, over 5647615.59 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3798, pruned_loss=0.1303, over 5645118.95 frames. ], batch size: 186, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:40:09,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=668625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:40:12,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=668628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:40:22,420 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5479, 2.4269, 2.3376, 2.1262], device='cuda:0'), covar=tensor([0.1640, 0.2419, 0.1945, 0.2174], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0749, 0.0698, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 22:40:25,016 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.229e+03 1.636e+03 2.232e+03 3.199e+03 9.348e+03, threshold=4.463e+03, percent-clipped=14.0
+2023-03-07 22:40:42,462 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=668657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:40:48,804 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-07 22:40:52,813 INFO [train.py:968] (0/2) Epoch 15, batch 29750, giga_loss[loss=0.2766, simple_loss=0.3565, pruned_loss=0.09831, over 28943.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.378, pruned_loss=0.1284, over 5643549.49 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3688, pruned_loss=0.1209, over 5643924.45 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3792, pruned_loss=0.1296, over 5645401.71 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:41:41,220 INFO [train.py:968] (0/2) Epoch 15, batch 29800, giga_loss[loss=0.2837, simple_loss=0.3461, pruned_loss=0.1106, over 28716.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3794, pruned_loss=0.1285, over 5651870.33 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3692, pruned_loss=0.1212, over 5646223.37 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.38, pruned_loss=0.1292, over 5651214.01 frames. ], batch size: 92, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:41:53,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=668731.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:42:00,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.825e+02 1.520e+03 1.998e+03 2.661e+03 6.113e+03, threshold=3.996e+03, percent-clipped=5.0
+2023-03-07 22:42:04,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2912, 1.4189, 1.2655, 1.4287], device='cuda:0'), covar=tensor([0.0734, 0.0359, 0.0323, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-07 22:42:29,508 INFO [train.py:968] (0/2) Epoch 15, batch 29850, libri_loss[loss=0.2875, simple_loss=0.3601, pruned_loss=0.1075, over 29235.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3789, pruned_loss=0.1288, over 5650617.00 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3693, pruned_loss=0.1214, over 5648049.41 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3795, pruned_loss=0.1293, over 5648537.84 frames. ], batch size: 97, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:42:37,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3716, 4.2335, 3.9821, 1.8304], device='cuda:0'), covar=tensor([0.0569, 0.0706, 0.0768, 0.2121], device='cuda:0'), in_proj_covar=tensor([0.1161, 0.1077, 0.0928, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 22:42:51,400 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=668795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:43:11,956 INFO [train.py:968] (0/2) Epoch 15, batch 29900, giga_loss[loss=0.2922, simple_loss=0.3662, pruned_loss=0.1091, over 28833.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3772, pruned_loss=0.1272, over 5671641.24 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3689, pruned_loss=0.121, over 5654554.73 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3783, pruned_loss=0.1281, over 5664355.35 frames. ], batch size: 199, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:43:21,014 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4828, 1.6314, 1.5981, 1.4333], device='cuda:0'), covar=tensor([0.1702, 0.1897, 0.2236, 0.2005], device='cuda:0'), in_proj_covar=tensor([0.0452, 0.0744, 0.0694, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 22:43:29,215 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 22:43:32,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.677e+03 2.262e+03 2.995e+03 8.135e+03, threshold=4.523e+03, percent-clipped=13.0
+2023-03-07 22:43:54,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=668866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:43:57,576 INFO [train.py:968] (0/2) Epoch 15, batch 29950, giga_loss[loss=0.3089, simple_loss=0.3532, pruned_loss=0.1323, over 23375.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3754, pruned_loss=0.1263, over 5665428.08 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3695, pruned_loss=0.1214, over 5655052.49 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3758, pruned_loss=0.1268, over 5659445.86 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:44:44,466 INFO [train.py:968] (0/2) Epoch 15, batch 30000, giga_loss[loss=0.3618, simple_loss=0.3967, pruned_loss=0.1634, over 26640.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3717, pruned_loss=0.1247, over 5663171.14 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3696, pruned_loss=0.1215, over 5663676.57 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3721, pruned_loss=0.1251, over 5651023.13 frames. ], batch size: 555, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:44:44,470 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 22:44:51,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8592, 3.6378, 3.4622, 1.7652], device='cuda:0'), covar=tensor([0.0883, 0.1066, 0.1080, 0.2526], device='cuda:0'), in_proj_covar=tensor([0.1155, 0.1071, 0.0921, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 22:44:52,857 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2104, simple_loss=0.3176, pruned_loss=0.05162, over 944034.00 frames. 
+2023-03-07 22:44:52,858 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 22:44:59,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0098, 2.2188, 1.5430, 1.9269], device='cuda:0'), covar=tensor([0.0877, 0.0656, 0.0993, 0.1068], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0446, 0.0507, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 22:45:09,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.684e+03 2.346e+03 3.622e+03 9.108e+03, threshold=4.691e+03, percent-clipped=15.0
+2023-03-07 22:45:33,011 INFO [train.py:968] (0/2) Epoch 15, batch 30050, giga_loss[loss=0.3518, simple_loss=0.3984, pruned_loss=0.1526, over 27602.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3703, pruned_loss=0.1251, over 5657879.87 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3697, pruned_loss=0.1218, over 5660030.51 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3706, pruned_loss=0.1253, over 5651895.81 frames. ], batch size: 472, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:46:14,513 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=669009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:46:17,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=669012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:46:23,008 INFO [train.py:968] (0/2) Epoch 15, batch 30100, giga_loss[loss=0.342, simple_loss=0.3967, pruned_loss=0.1436, over 27918.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3686, pruned_loss=0.1246, over 5654385.89 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3697, pruned_loss=0.1218, over 5661393.36 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3687, pruned_loss=0.1247, over 5648416.59 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:46:34,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2674, 1.8119, 5.7044, 4.3252], device='cuda:0'), covar=tensor([0.1618, 0.2562, 0.0593, 0.0678], device='cuda:0'), in_proj_covar=tensor([0.0707, 0.0617, 0.0907, 0.0822], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 22:46:47,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.149e+03 1.672e+03 2.238e+03 3.445e+03 8.614e+03, threshold=4.476e+03, percent-clipped=14.0
+2023-03-07 22:46:48,477 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=669041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:47:03,613 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=669059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:47:12,640 INFO [train.py:968] (0/2) Epoch 15, batch 30150, giga_loss[loss=0.3153, simple_loss=0.3878, pruned_loss=0.1214, over 27976.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3675, pruned_loss=0.1227, over 5650171.36 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3698, pruned_loss=0.122, over 5664738.43 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3676, pruned_loss=0.1227, over 5642259.19 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:47:18,474 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 22:47:31,696 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=669086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:47:31,924 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 22:47:53,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=669106.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 22:48:08,841 INFO [train.py:968] (0/2) Epoch 15, batch 30200, giga_loss[loss=0.2836, simple_loss=0.3486, pruned_loss=0.1092, over 26586.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3648, pruned_loss=0.1186, over 5640677.71 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.37, pruned_loss=0.1221, over 5666977.32 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3647, pruned_loss=0.1185, over 5632522.49 frames. ], batch size: 555, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:48:13,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4250, 1.6808, 1.4992, 1.4069], device='cuda:0'), covar=tensor([0.2057, 0.1601, 0.1480, 0.1575], device='cuda:0'), in_proj_covar=tensor([0.1833, 0.1768, 0.1686, 0.1822], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 22:48:32,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.201e+02 1.495e+03 1.979e+03 2.688e+03 5.379e+03, threshold=3.958e+03, percent-clipped=4.0
+2023-03-07 22:49:01,307 INFO [train.py:968] (0/2) Epoch 15, batch 30250, giga_loss[loss=0.26, simple_loss=0.3447, pruned_loss=0.0877, over 28949.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3632, pruned_loss=0.1157, over 5636027.68 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3698, pruned_loss=0.1221, over 5650227.34 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3631, pruned_loss=0.1156, over 5644525.30 frames. ], batch size: 145, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:49:03,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=669170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:49:27,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2973, 1.6881, 1.5733, 1.4649], device='cuda:0'), covar=tensor([0.1448, 0.1212, 0.1792, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0736, 0.0686, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 22:49:50,773 INFO [train.py:968] (0/2) Epoch 15, batch 30300, libri_loss[loss=0.342, simple_loss=0.3921, pruned_loss=0.146, over 29533.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3613, pruned_loss=0.1138, over 5636205.28 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3703, pruned_loss=0.123, over 5650428.70 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3605, pruned_loss=0.1126, over 5641793.44 frames. ], batch size: 83, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:50:09,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.527e+02 1.381e+03 1.943e+03 2.733e+03 1.074e+04, threshold=3.887e+03, percent-clipped=12.0
+2023-03-07 22:50:13,177 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-07 22:50:20,276 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=669249.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:50:22,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=669252.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:50:35,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9553, 4.7627, 4.5643, 2.9374], device='cuda:0'), covar=tensor([0.0581, 0.0846, 0.0867, 0.1643], device='cuda:0'), in_proj_covar=tensor([0.1148, 0.1065, 0.0917, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 22:50:37,470 INFO [train.py:968] (0/2) Epoch 15, batch 30350, giga_loss[loss=0.2644, simple_loss=0.3399, pruned_loss=0.0944, over 28910.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3584, pruned_loss=0.111, over 5634107.24 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3704, pruned_loss=0.1233, over 5643328.73 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3575, pruned_loss=0.1095, over 5645432.56 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:50:50,597 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=669281.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:51:20,617 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=669313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:51:24,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=669316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:51:26,341 INFO [train.py:968] (0/2) Epoch 15, batch 30400, giga_loss[loss=0.2634, simple_loss=0.3503, pruned_loss=0.08826, over 28341.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3575, pruned_loss=0.1077, over 5644990.42 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3702, pruned_loss=0.1232, over 5636458.45 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3567, pruned_loss=0.1063, over 5660088.16 frames. ], batch size: 368, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:51:51,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.572e+02 1.429e+03 1.821e+03 2.468e+03 4.946e+03, threshold=3.643e+03, percent-clipped=5.0
+2023-03-07 22:51:55,476 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=669345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:52:13,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4366, 1.6464, 1.7316, 1.2507], device='cuda:0'), covar=tensor([0.1818, 0.2735, 0.1548, 0.1925], device='cuda:0'), in_proj_covar=tensor([0.0854, 0.0690, 0.0901, 0.0801], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 22:52:17,270 INFO [train.py:968] (0/2) Epoch 15, batch 30450, giga_loss[loss=0.282, simple_loss=0.3609, pruned_loss=0.1015, over 28308.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3572, pruned_loss=0.1068, over 5653612.49 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3698, pruned_loss=0.123, over 5641404.31 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3566, pruned_loss=0.1054, over 5661303.70 frames. ], batch size: 369, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:52:42,834 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.02 vs. limit=2.0
+2023-03-07 22:53:09,185 INFO [train.py:968] (0/2) Epoch 15, batch 30500, giga_loss[loss=0.2486, simple_loss=0.3338, pruned_loss=0.08163, over 28965.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.355, pruned_loss=0.1049, over 5659155.17 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3696, pruned_loss=0.123, over 5644226.63 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3545, pruned_loss=0.1035, over 5662866.69 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:53:26,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=669434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:53:34,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.607e+02 1.343e+03 1.721e+03 2.402e+03 5.292e+03, threshold=3.441e+03, percent-clipped=7.0
+2023-03-07 22:53:54,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=669461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:54:01,724 INFO [train.py:968] (0/2) Epoch 15, batch 30550, giga_loss[loss=0.2985, simple_loss=0.362, pruned_loss=0.1175, over 27874.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3516, pruned_loss=0.1021, over 5659830.52 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3695, pruned_loss=0.1229, over 5645514.50 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3512, pruned_loss=0.101, over 5661745.34 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:54:46,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4932, 1.5214, 1.2107, 1.1134], device='cuda:0'), covar=tensor([0.0691, 0.0403, 0.0845, 0.1056], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0437, 0.0500, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 22:54:46,757 INFO [train.py:968] (0/2) Epoch 15, batch 30600, giga_loss[loss=0.3162, simple_loss=0.3632, pruned_loss=0.1346, over 26518.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3504, pruned_loss=0.1027, over 5661989.41 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3683, pruned_loss=0.1227, over 5659545.39 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3501, pruned_loss=0.1007, over 5651059.39 frames. ], batch size: 555, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:55:00,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9389, 2.2095, 1.7144, 2.2612], device='cuda:0'), covar=tensor([0.2501, 0.2469, 0.2791, 0.2258], device='cuda:0'), in_proj_covar=tensor([0.1404, 0.1025, 0.1245, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 22:55:06,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.752e+02 1.439e+03 1.933e+03 2.685e+03 7.423e+03, threshold=3.866e+03, percent-clipped=14.0
+2023-03-07 22:55:33,035 INFO [train.py:968] (0/2) Epoch 15, batch 30650, giga_loss[loss=0.2587, simple_loss=0.3399, pruned_loss=0.08875, over 28865.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3493, pruned_loss=0.1016, over 5664144.83 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3667, pruned_loss=0.1219, over 5656098.73 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.35, pruned_loss=0.1001, over 5659095.01 frames. ], batch size: 199, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:55:39,657 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=669577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:55:42,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=669580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:56:09,039 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=669604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:56:11,002 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=669607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:56:12,943 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=669609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:56:14,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9617, 1.2039, 1.1352, 0.9434], device='cuda:0'), covar=tensor([0.1918, 0.1986, 0.1150, 0.1560], device='cuda:0'), in_proj_covar=tensor([0.1802, 0.1734, 0.1651, 0.1786], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 22:56:25,169 INFO [train.py:968] (0/2) Epoch 15, batch 30700, giga_loss[loss=0.265, simple_loss=0.3497, pruned_loss=0.09018, over 28745.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3475, pruned_loss=0.09978, over 5658428.95 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3666, pruned_loss=0.1219, over 5656614.67 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3479, pruned_loss=0.09849, over 5654170.44 frames. ], batch size: 262, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:56:41,033 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=669636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:56:46,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.159e+02 1.350e+03 1.851e+03 2.636e+03 6.364e+03, threshold=3.703e+03, percent-clipped=10.0
+2023-03-07 22:57:15,042 INFO [train.py:968] (0/2) Epoch 15, batch 30750, giga_loss[loss=0.2421, simple_loss=0.3318, pruned_loss=0.07616, over 28718.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3442, pruned_loss=0.0971, over 5664004.73 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3659, pruned_loss=0.1216, over 5661818.40 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3447, pruned_loss=0.09583, over 5655819.60 frames. ], batch size: 242, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:58:02,770 INFO [train.py:968] (0/2) Epoch 15, batch 30800, giga_loss[loss=0.2447, simple_loss=0.3206, pruned_loss=0.08445, over 28679.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3414, pruned_loss=0.09542, over 5676526.69 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3659, pruned_loss=0.1217, over 5667221.60 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3413, pruned_loss=0.09374, over 5665337.85 frames. ], batch size: 262, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:58:30,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.519e+02 1.318e+03 1.701e+03 2.650e+03 6.230e+03, threshold=3.402e+03, percent-clipped=11.0
+2023-03-07 22:58:56,030 INFO [train.py:968] (0/2) Epoch 15, batch 30850, giga_loss[loss=0.2854, simple_loss=0.352, pruned_loss=0.1094, over 28382.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3394, pruned_loss=0.09481, over 5671313.09 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3661, pruned_loss=0.1219, over 5665781.33 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3389, pruned_loss=0.0931, over 5663873.90 frames. ], batch size: 368, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:59:49,331 INFO [train.py:968] (0/2) Epoch 15, batch 30900, giga_loss[loss=0.2368, simple_loss=0.314, pruned_loss=0.07979, over 28782.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3394, pruned_loss=0.0954, over 5662620.13 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.366, pruned_loss=0.1218, over 5668371.39 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.339, pruned_loss=0.09395, over 5654386.65 frames. ], batch size: 119, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:00:16,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.038e+02 1.311e+03 1.718e+03 2.243e+03 5.062e+03, threshold=3.435e+03, percent-clipped=9.0
+2023-03-07 23:00:16,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8311, 2.8322, 2.6836, 2.4562], device='cuda:0'), covar=tensor([0.1330, 0.1781, 0.1499, 0.1659], device='cuda:0'), in_proj_covar=tensor([0.0439, 0.0725, 0.0676, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-07 23:00:40,194 INFO [train.py:968] (0/2) Epoch 15, batch 30950, giga_loss[loss=0.2974, simple_loss=0.3762, pruned_loss=0.1093, over 28823.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3408, pruned_loss=0.09641, over 5650995.86 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3652, pruned_loss=0.1215, over 5667515.93 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3404, pruned_loss=0.09479, over 5644824.07 frames. ], batch size: 186, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:00:50,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4610, 1.5937, 1.7210, 1.3045], device='cuda:0'), covar=tensor([0.1784, 0.2459, 0.1476, 0.1818], device='cuda:0'), in_proj_covar=tensor([0.0857, 0.0688, 0.0903, 0.0804], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 23:01:32,674 INFO [train.py:968] (0/2) Epoch 15, batch 31000, giga_loss[loss=0.259, simple_loss=0.3449, pruned_loss=0.08657, over 28802.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3428, pruned_loss=0.0968, over 5643725.97 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3649, pruned_loss=0.1215, over 5665071.67 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.342, pruned_loss=0.09461, over 5639684.52 frames. ], batch size: 243, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:02:00,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.459e+03 1.921e+03 2.938e+03 7.371e+03, threshold=3.842e+03, percent-clipped=21.0
+2023-03-07 23:02:15,758 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-07 23:02:31,648 INFO [train.py:968] (0/2) Epoch 15, batch 31050, giga_loss[loss=0.237, simple_loss=0.3232, pruned_loss=0.07534, over 28370.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3418, pruned_loss=0.09559, over 5635661.31 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.365, pruned_loss=0.1215, over 5667361.34 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3408, pruned_loss=0.0935, over 5630066.77 frames. ], batch size: 368, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:03:11,013 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-670000.pt
+2023-03-07 23:03:13,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5608, 3.3852, 3.2031, 1.6983], device='cuda:0'), covar=tensor([0.0780, 0.0895, 0.0921, 0.2443], device='cuda:0'), in_proj_covar=tensor([0.1121, 0.1039, 0.0890, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-07 23:03:24,367 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:03:33,382 INFO [train.py:968] (0/2) Epoch 15, batch 31100, giga_loss[loss=0.2356, simple_loss=0.3175, pruned_loss=0.07681, over 28837.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3416, pruned_loss=0.09578, over 5644726.41 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.364, pruned_loss=0.1211, over 5672212.20 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.341, pruned_loss=0.09364, over 5634681.24 frames. ], batch size: 112, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:04:02,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.687e+02 1.315e+03 1.650e+03 2.767e+03 5.686e+03, threshold=3.300e+03, percent-clipped=8.0
+2023-03-07 23:04:15,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0696, 1.1915, 3.5056, 3.0378], device='cuda:0'), covar=tensor([0.1696, 0.2826, 0.0469, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.0698, 0.0609, 0.0893, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 23:04:35,236 INFO [train.py:968] (0/2) Epoch 15, batch 31150, giga_loss[loss=0.2466, simple_loss=0.3319, pruned_loss=0.0807, over 28992.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.34, pruned_loss=0.09434, over 5648994.39 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3634, pruned_loss=0.1208, over 5675317.53 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3396, pruned_loss=0.09244, over 5637750.69 frames. ], batch size: 227, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:04:44,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4163, 1.5938, 1.4324, 1.5848], device='cuda:0'), covar=tensor([0.0689, 0.0293, 0.0304, 0.0710], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:0')
+2023-03-07 23:05:09,305 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:05:35,844 INFO [train.py:968] (0/2) Epoch 15, batch 31200, giga_loss[loss=0.2524, simple_loss=0.3335, pruned_loss=0.08558, over 29029.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3374, pruned_loss=0.09166, over 5647300.45 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3626, pruned_loss=0.1204, over 5680937.66 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3371, pruned_loss=0.08971, over 5632136.67 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:06:02,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3445, 1.9145, 1.4438, 0.6916], device='cuda:0'), covar=tensor([0.4184, 0.2029, 0.2734, 0.4274], device='cuda:0'), in_proj_covar=tensor([0.1614, 0.1532, 0.1517, 0.1328], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 23:06:02,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.711e+02 1.374e+03 1.892e+03 2.817e+03 1.169e+04, threshold=3.784e+03, percent-clipped=19.0
+2023-03-07 23:06:33,607 INFO [train.py:968] (0/2) Epoch 15, batch 31250, giga_loss[loss=0.2768, simple_loss=0.346, pruned_loss=0.1038, over 29029.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3362, pruned_loss=0.09204, over 5656960.06 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3615, pruned_loss=0.1199, over 5684485.95 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3361, pruned_loss=0.09006, over 5640830.98 frames. ], batch size: 186, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:07:28,011 INFO [train.py:968] (0/2) Epoch 15, batch 31300, giga_loss[loss=0.2551, simple_loss=0.3395, pruned_loss=0.08531, over 28936.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3361, pruned_loss=0.09289, over 5664861.62 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3612, pruned_loss=0.1199, over 5682665.16 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3353, pruned_loss=0.09006, over 5652268.04 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:07:50,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.065e+02 1.339e+03 1.740e+03 2.305e+03 5.298e+03, threshold=3.479e+03, percent-clipped=2.0
+2023-03-07 23:08:19,850 INFO [train.py:968] (0/2) Epoch 15, batch 31350, libri_loss[loss=0.2706, simple_loss=0.3348, pruned_loss=0.1032, over 29530.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3351, pruned_loss=0.09288, over 5676106.21 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3604, pruned_loss=0.1197, over 5691030.57 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3339, pruned_loss=0.08959, over 5657454.38 frames. ], batch size: 79, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:08:30,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0813, 2.4624, 2.4662, 1.8502], device='cuda:0'), covar=tensor([0.1463, 0.2205, 0.1212, 0.1615], device='cuda:0'), in_proj_covar=tensor([0.0858, 0.0688, 0.0905, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 23:09:14,131 INFO [train.py:968] (0/2) Epoch 15, batch 31400, giga_loss[loss=0.2226, simple_loss=0.3189, pruned_loss=0.06313, over 29003.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3362, pruned_loss=0.09289, over 5676035.68 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3601, pruned_loss=0.1196, over 5695791.07 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3349, pruned_loss=0.08954, over 5656426.06 frames. ], batch size: 136, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:09:42,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.872e+02 1.443e+03 2.029e+03 2.859e+03 1.181e+04, threshold=4.058e+03, percent-clipped=15.0
+2023-03-07 23:10:13,933 INFO [train.py:968] (0/2) Epoch 15, batch 31450, giga_loss[loss=0.2502, simple_loss=0.3344, pruned_loss=0.083, over 29045.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3392, pruned_loss=0.09403, over 5670628.14 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3601, pruned_loss=0.1196, over 5699350.51 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3376, pruned_loss=0.0906, over 5651131.76 frames. ], batch size: 106, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:10:22,651 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:10:36,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=670385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:11:17,920 INFO [train.py:968] (0/2) Epoch 15, batch 31500, giga_loss[loss=0.23, simple_loss=0.3036, pruned_loss=0.07826, over 27832.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3354, pruned_loss=0.0911, over 5681107.60 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3599, pruned_loss=0.1197, over 5703614.89 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3338, pruned_loss=0.0879, over 5661461.17 frames. ], batch size: 476, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:11:26,742 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 23:11:49,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.685e+02 1.235e+03 1.618e+03 2.282e+03 4.327e+03, threshold=3.235e+03, percent-clipped=1.0
+2023-03-07 23:12:22,189 INFO [train.py:968] (0/2) Epoch 15, batch 31550, giga_loss[loss=0.263, simple_loss=0.3344, pruned_loss=0.09576, over 27777.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3365, pruned_loss=0.09247, over 5672779.41 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3599, pruned_loss=0.1199, over 5692584.95 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3341, pruned_loss=0.08843, over 5665289.84 frames. ], batch size: 474, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:12:24,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=670470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:12:49,292 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:13:10,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4708, 1.6342, 1.6345, 1.4940], device='cuda:0'), covar=tensor([0.2153, 0.1922, 0.1862, 0.1886], device='cuda:0'), in_proj_covar=tensor([0.1803, 0.1737, 0.1646, 0.1792], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-07 23:13:22,108 INFO [train.py:968] (0/2) Epoch 15, batch 31600, giga_loss[loss=0.2955, simple_loss=0.3674, pruned_loss=0.1118, over 27587.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3384, pruned_loss=0.09297, over 5674944.98 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3594, pruned_loss=0.1197, over 5698654.90 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3364, pruned_loss=0.08919, over 5662945.34 frames. ], batch size: 472, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:13:32,076 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=670528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:13:36,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=670531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:13:54,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.590e+02 1.477e+03 1.844e+03 2.459e+03 5.402e+03, threshold=3.689e+03, percent-clipped=9.0
+2023-03-07 23:14:02,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2630, 1.8283, 1.3305, 0.4250], device='cuda:0'), covar=tensor([0.3889, 0.2767, 0.4249, 0.4987], device='cuda:0'), in_proj_covar=tensor([0.1627, 0.1553, 0.1529, 0.1333], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-07 23:14:05,164 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 23:14:15,508 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=670560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:14:25,985 INFO [train.py:968] (0/2) Epoch 15, batch 31650, giga_loss[loss=0.2642, simple_loss=0.3584, pruned_loss=0.08495, over 28971.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3424, pruned_loss=0.09215, over 5673090.48 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3594, pruned_loss=0.1197, over 5700009.58 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3406, pruned_loss=0.08879, over 5661900.02 frames. ], batch size: 284, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:14:59,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2718, 1.8629, 1.6487, 1.3387], device='cuda:0'), covar=tensor([0.0861, 0.0288, 0.0291, 0.1115], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0063, 0.0058, 0.0098], device='cuda:0')
+2023-03-07 23:15:20,145 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=670613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:15:23,323 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=670616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:15:25,568 INFO [train.py:968] (0/2) Epoch 15, batch 31700, giga_loss[loss=0.1964, simple_loss=0.2688, pruned_loss=0.062, over 24493.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3437, pruned_loss=0.0918, over 5657187.49 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3594, pruned_loss=0.1197, over 5692976.70 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.08838, over 5652975.97 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:15:56,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.470e+02 1.480e+03 1.914e+03 2.787e+03 7.445e+03, threshold=3.828e+03, percent-clipped=10.0
+2023-03-07 23:15:56,695 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=670645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:16:10,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6203, 1.7095, 1.2143, 1.3643], device='cuda:0'), covar=tensor([0.0863, 0.0618, 0.1012, 0.1094], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0439, 0.0504, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 23:16:22,057 INFO [train.py:968] (0/2) Epoch 15, batch 31750, libri_loss[loss=0.2998, simple_loss=0.3568, pruned_loss=0.1215, over 29515.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3435, pruned_loss=0.09069, over 5671219.24 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3585, pruned_loss=0.1191, over 5700175.51 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3423, pruned_loss=0.08751, over 5659866.32 frames. ], batch size: 84, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:16:34,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7165, 4.5494, 4.3087, 2.1971], device='cuda:0'), covar=tensor([0.0469, 0.0675, 0.0667, 0.1841], device='cuda:0'), in_proj_covar=tensor([0.1110, 0.1030, 0.0881, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 23:17:17,865 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:17:19,148 INFO [train.py:968] (0/2) Epoch 15, batch 31800, giga_loss[loss=0.2946, simple_loss=0.363, pruned_loss=0.1131, over 28962.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3446, pruned_loss=0.09223, over 5674041.04 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3583, pruned_loss=0.119, over 5695639.85 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3433, pruned_loss=0.08877, over 5668015.22 frames. ], batch size: 199, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:17:27,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5996, 4.4403, 1.6352, 1.6807], device='cuda:0'), covar=tensor([0.0931, 0.0250, 0.0875, 0.1295], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0528, 0.0358, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 23:17:52,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.882e+02 1.292e+03 1.826e+03 2.624e+03 1.127e+04, threshold=3.652e+03, percent-clipped=9.0
+2023-03-07 23:17:58,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=670749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:18:21,887 INFO [train.py:968] (0/2) Epoch 15, batch 31850, giga_loss[loss=0.2082, simple_loss=0.2885, pruned_loss=0.06393, over 28469.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3421, pruned_loss=0.09186, over 5674544.77 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3581, pruned_loss=0.119, over 5687848.40 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3408, pruned_loss=0.08837, over 5675823.88 frames. ], batch size: 78, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:19:42,011 INFO [train.py:968] (0/2) Epoch 15, batch 31900, libri_loss[loss=0.3456, simple_loss=0.3943, pruned_loss=0.1484, over 29543.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3434, pruned_loss=0.09353, over 5677107.79 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.358, pruned_loss=0.1189, over 5692634.00 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3421, pruned_loss=0.0902, over 5673138.23 frames. ], batch size: 89, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:20:09,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5797, 5.4210, 5.0920, 2.2608], device='cuda:0'), covar=tensor([0.0397, 0.0560, 0.0699, 0.2002], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.1032, 0.0886, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-07 23:20:29,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.292e+02 1.452e+03 1.939e+03 2.489e+03 5.071e+03, threshold=3.879e+03, percent-clipped=4.0
+2023-03-07 23:20:45,425 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 23:20:47,980 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670856.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:21:04,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=670865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:21:08,889 INFO [train.py:968] (0/2) Epoch 15, batch 31950, giga_loss[loss=0.2436, simple_loss=0.3192, pruned_loss=0.084, over 28237.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3393, pruned_loss=0.0912, over 5672617.73 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.358, pruned_loss=0.1189, over 5692634.00 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3382, pruned_loss=0.0886, over 5669528.17 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:21:29,489 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:21:42,261 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=670892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:21:46,990 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=670895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:22:01,023 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:22:17,149 INFO [train.py:968] (0/2) Epoch 15, batch 32000, giga_loss[loss=0.2344, simple_loss=0.3003, pruned_loss=0.08423, over 24091.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3361, pruned_loss=0.08921, over 5675186.14 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3576, pruned_loss=0.1186, over 5695285.49 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3351, pruned_loss=0.08657, over 5669673.78 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:22:24,099 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=670924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:22:48,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.848e+02 1.405e+03 1.910e+03 2.629e+03 7.382e+03, threshold=3.820e+03, percent-clipped=7.0
+2023-03-07 23:23:25,087 INFO [train.py:968] (0/2) Epoch 15, batch 32050, giga_loss[loss=0.2516, simple_loss=0.3351, pruned_loss=0.08409, over 28913.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3325, pruned_loss=0.08744, over 5675958.80 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3577, pruned_loss=0.1188, over 5693823.13 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3312, pruned_loss=0.08467, over 5672510.54 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:24:19,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:24:22,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:24:34,469 INFO [train.py:968] (0/2) Epoch 15, batch 32100, giga_loss[loss=0.2893, simple_loss=0.3725, pruned_loss=0.1031, over 29023.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3364, pruned_loss=0.08952, over 5684761.70 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3578, pruned_loss=0.119, over 5696896.85 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3349, pruned_loss=0.08659, over 5679113.12 frames. ], batch size: 136, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:24:59,819 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:25:04,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.551e+02 1.386e+03 1.702e+03 2.405e+03 6.158e+03, threshold=3.404e+03, percent-clipped=5.0
+2023-03-07 23:25:33,021 INFO [train.py:968] (0/2) Epoch 15, batch 32150, giga_loss[loss=0.262, simple_loss=0.333, pruned_loss=0.09553, over 28846.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3384, pruned_loss=0.09115, over 5693576.64 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.357, pruned_loss=0.1185, over 5703368.33 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.337, pruned_loss=0.08818, over 5682921.58 frames. ], batch size: 174, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:25:39,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3992, 1.9088, 1.4841, 1.6021], device='cuda:0'), covar=tensor([0.0697, 0.0332, 0.0322, 0.0766], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0063, 0.0058, 0.0098], device='cuda:0')
+2023-03-07 23:26:07,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:26:42,313 INFO [train.py:968] (0/2) Epoch 15, batch 32200, giga_loss[loss=0.2765, simple_loss=0.3494, pruned_loss=0.1018, over 28352.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3368, pruned_loss=0.09131, over 5685382.70 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3572, pruned_loss=0.1186, over 5696544.25 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3354, pruned_loss=0.08858, over 5683445.36 frames. ], batch size: 368, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:26:45,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4417, 1.6571, 1.3005, 1.5120], device='cuda:0'), covar=tensor([0.2641, 0.2595, 0.3056, 0.2040], device='cuda:0'), in_proj_covar=tensor([0.1394, 0.1019, 0.1239, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 23:27:14,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.606e+02 1.361e+03 1.742e+03 2.252e+03 5.114e+03, threshold=3.484e+03, percent-clipped=6.0
+2023-03-07 23:27:42,692 INFO [train.py:968] (0/2) Epoch 15, batch 32250, giga_loss[loss=0.2488, simple_loss=0.3323, pruned_loss=0.0827, over 28456.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3379, pruned_loss=0.09302, over 5668812.99 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3571, pruned_loss=0.1186, over 5685251.02 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3362, pruned_loss=0.08996, over 5676810.49 frames. ], batch size: 369, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:28:54,210 INFO [train.py:968] (0/2) Epoch 15, batch 32300, giga_loss[loss=0.245, simple_loss=0.3355, pruned_loss=0.0772, over 28935.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3378, pruned_loss=0.09227, over 5669956.11 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3573, pruned_loss=0.1186, over 5686627.91 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.336, pruned_loss=0.08946, over 5674731.88 frames. ], batch size: 227, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:29:10,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671231.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:29:13,736 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671233.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:29:16,294 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:29:20,548 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:29:33,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.569e+02 1.420e+03 2.049e+03 2.862e+03 1.112e+04, threshold=4.098e+03, percent-clipped=16.0
+2023-03-07 23:29:51,580 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:30:01,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9550, 1.0269, 3.3598, 2.9114], device='cuda:0'), covar=tensor([0.1687, 0.2816, 0.0492, 0.0984], device='cuda:0'), in_proj_covar=tensor([0.0694, 0.0610, 0.0888, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-07 23:30:05,511 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:30:06,496 INFO [train.py:968] (0/2) Epoch 15, batch 32350, libri_loss[loss=0.2852, simple_loss=0.3395, pruned_loss=0.1154, over 29557.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3403, pruned_loss=0.09263, over 5667874.17 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3572, pruned_loss=0.1187, over 5685302.97 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3387, pruned_loss=0.08993, over 5672188.27 frames. ], batch size: 76, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:30:29,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:31:29,209 INFO [train.py:968] (0/2) Epoch 15, batch 32400, giga_loss[loss=0.2371, simple_loss=0.3182, pruned_loss=0.07799, over 28865.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3393, pruned_loss=0.09244, over 5662257.63 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3562, pruned_loss=0.1182, over 5689743.43 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3386, pruned_loss=0.09025, over 5661584.14 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:31:48,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4229, 1.6727, 1.6958, 1.3119], device='cuda:0'), covar=tensor([0.1634, 0.2316, 0.1374, 0.1648], device='cuda:0'), in_proj_covar=tensor([0.0858, 0.0687, 0.0906, 0.0805], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 23:32:10,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.128e+02 1.257e+03 1.621e+03 2.242e+03 4.219e+03, threshold=3.241e+03, percent-clipped=1.0
+2023-03-07 23:32:39,820 INFO [train.py:968] (0/2) Epoch 15, batch 32450, giga_loss[loss=0.238, simple_loss=0.3116, pruned_loss=0.08224, over 28944.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.336, pruned_loss=0.09164, over 5677330.45 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3562, pruned_loss=0.1181, over 5693190.53 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3351, pruned_loss=0.08951, over 5673214.53 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:32:48,592 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671374.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:32:52,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671377.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:32:55,107 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:33:22,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:33:25,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:33:27,758 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671406.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:33:27,922 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-07 23:33:44,682 INFO [train.py:968] (0/2) Epoch 15, batch 32500, giga_loss[loss=0.2678, simple_loss=0.3347, pruned_loss=0.1004, over 28714.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3294, pruned_loss=0.08884, over 5666487.33 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3555, pruned_loss=0.1179, over 5681576.88 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3285, pruned_loss=0.08637, over 5672789.23 frames. ], batch size: 243, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:33:50,753 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:33:55,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:34:06,473 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:34:24,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.846e+02 1.381e+03 1.826e+03 2.612e+03 6.265e+03, threshold=3.651e+03, percent-clipped=10.0
+2023-03-07 23:34:36,641 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:34:38,215 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4779, 1.8464, 1.3750, 1.5537], device='cuda:0'), covar=tensor([0.0730, 0.0269, 0.0328, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0063, 0.0058, 0.0098], device='cuda:0')
+2023-03-07 23:34:50,183 INFO [train.py:968] (0/2) Epoch 15, batch 32550, giga_loss[loss=0.2585, simple_loss=0.3352, pruned_loss=0.09088, over 28961.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3296, pruned_loss=0.08928, over 5659633.64 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3553, pruned_loss=0.1177, over 5678419.05 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3284, pruned_loss=0.08669, over 5667297.39 frames. ], batch size: 213, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:35:08,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 23:35:51,449 INFO [train.py:968] (0/2) Epoch 15, batch 32600, giga_loss[loss=0.2795, simple_loss=0.3463, pruned_loss=0.1063, over 28961.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3317, pruned_loss=0.09059, over 5665014.71 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.355, pruned_loss=0.1174, over 5683053.49 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3305, pruned_loss=0.08826, over 5666735.42 frames. ], batch size: 199, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:36:29,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.687e+02 1.591e+03 2.094e+03 3.111e+03 1.010e+04, threshold=4.188e+03, percent-clipped=20.0
+2023-03-07 23:36:57,529 INFO [train.py:968] (0/2) Epoch 15, batch 32650, giga_loss[loss=0.2299, simple_loss=0.3219, pruned_loss=0.06899, over 28919.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.329, pruned_loss=0.08813, over 5666655.96 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3546, pruned_loss=0.1172, over 5684905.93 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3281, pruned_loss=0.08607, over 5666079.06 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:37:48,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671608.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:38:02,911 INFO [train.py:968] (0/2) Epoch 15, batch 32700, giga_loss[loss=0.2642, simple_loss=0.3359, pruned_loss=0.09619, over 28108.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3278, pruned_loss=0.08675, over 5664427.59 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3544, pruned_loss=0.1171, over 5687632.50 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3268, pruned_loss=0.08469, over 5661163.49 frames. ], batch size: 412, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:38:40,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5119, 1.7535, 1.4272, 1.5434], device='cuda:0'), covar=tensor([0.2742, 0.2434, 0.2765, 0.2208], device='cuda:0'), in_proj_covar=tensor([0.1390, 0.1015, 0.1238, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-07 23:38:43,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.218e+02 1.214e+03 1.573e+03 2.156e+03 8.129e+03, threshold=3.146e+03, percent-clipped=7.0
+2023-03-07 23:39:15,120 INFO [train.py:968] (0/2) Epoch 15, batch 32750, giga_loss[loss=0.2324, simple_loss=0.3135, pruned_loss=0.07568, over 28931.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3266, pruned_loss=0.0863, over 5663740.07 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3544, pruned_loss=0.117, over 5689481.78 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3255, pruned_loss=0.08447, over 5659299.67 frames. ], batch size: 227, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:40:26,657 INFO [train.py:968] (0/2) Epoch 15, batch 32800, giga_loss[loss=0.2496, simple_loss=0.3301, pruned_loss=0.08454, over 29079.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3277, pruned_loss=0.08633, over 5676017.78 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3545, pruned_loss=0.117, over 5689734.71 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3263, pruned_loss=0.0843, over 5671768.84 frames. ], batch size: 136, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:41:06,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.171e+02 1.374e+03 1.762e+03 3.039e+03 1.283e+04, threshold=3.525e+03, percent-clipped=23.0
+2023-03-07 23:41:11,056 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671751.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:41:14,585 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671754.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:41:18,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:41:35,797 INFO [train.py:968] (0/2) Epoch 15, batch 32850, giga_loss[loss=0.241, simple_loss=0.322, pruned_loss=0.07998, over 28962.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.329, pruned_loss=0.08759, over 5681959.21 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3546, pruned_loss=0.1173, over 5694911.46 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3272, pruned_loss=0.08494, over 5673785.15 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:41:56,741 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671783.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:41:57,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:42:27,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6933, 1.7776, 1.8982, 1.4741], device='cuda:0'), covar=tensor([0.1917, 0.2471, 0.1527, 0.1773], device='cuda:0'), in_proj_covar=tensor([0.0857, 0.0686, 0.0905, 0.0805], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 23:42:40,099 INFO [train.py:968] (0/2) Epoch 15, batch 32900, giga_loss[loss=0.241, simple_loss=0.3194, pruned_loss=0.08132, over 28817.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3297, pruned_loss=0.0883, over 5685636.93 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3539, pruned_loss=0.1168, over 5698092.76 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3283, pruned_loss=0.08606, over 5676015.26 frames. ], batch size: 263, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:43:00,686 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:43:03,727 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3391, 1.2033, 1.1427, 1.5059], device='cuda:0'), covar=tensor([0.0782, 0.0349, 0.0361, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0063, 0.0058, 0.0098], device='cuda:0')
+2023-03-07 23:43:21,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.073e+02 1.150e+03 1.412e+03 1.832e+03 7.283e+03, threshold=2.824e+03, percent-clipped=2.0
+2023-03-07 23:43:44,962 INFO [train.py:968] (0/2) Epoch 15, batch 32950, giga_loss[loss=0.2866, simple_loss=0.3503, pruned_loss=0.1115, over 27580.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3289, pruned_loss=0.0876, over 5682428.17 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3532, pruned_loss=0.1164, over 5702487.60 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3279, pruned_loss=0.08552, over 5670247.72 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:44:27,959 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:44:32,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:44:52,298 INFO [train.py:968] (0/2) Epoch 15, batch 33000, giga_loss[loss=0.2895, simple_loss=0.3507, pruned_loss=0.1141, over 26914.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3301, pruned_loss=0.08678, over 5671304.22 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.353, pruned_loss=0.1163, over 5703518.16 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3293, pruned_loss=0.08513, over 5660616.05 frames. ], batch size: 555, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:44:52,302 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-07 23:45:01,953 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.1999, simple_loss=0.3005, pruned_loss=0.04966, over 944034.00 frames. 
+2023-03-07 23:45:01,954 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-07 23:45:18,149 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:45:31,605 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671944.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:45:35,523 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.706e+02 1.376e+03 1.801e+03 2.674e+03 8.575e+03, threshold=3.601e+03, percent-clipped=18.0
+2023-03-07 23:45:40,007 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-07 23:45:59,165 INFO [train.py:968] (0/2) Epoch 15, batch 33050, giga_loss[loss=0.2726, simple_loss=0.3487, pruned_loss=0.09821, over 27677.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3329, pruned_loss=0.08823, over 5678776.51 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3523, pruned_loss=0.1157, over 5713548.19 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.332, pruned_loss=0.086, over 5659098.17 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:46:03,023 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 23:46:05,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-07 23:46:22,720 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:46:39,750 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-672000.pt
+2023-03-07 23:47:06,712 INFO [train.py:968] (0/2) Epoch 15, batch 33100, giga_loss[loss=0.266, simple_loss=0.341, pruned_loss=0.09553, over 29004.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.334, pruned_loss=0.08826, over 5673969.53 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3523, pruned_loss=0.1158, over 5706083.40 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.333, pruned_loss=0.08611, over 5664470.40 frames. ], batch size: 199, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:47:45,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.839e+02 1.309e+03 1.721e+03 2.468e+03 5.000e+03, threshold=3.441e+03, percent-clipped=8.0
+2023-03-07 23:48:05,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4311, 1.7417, 1.6869, 1.2590], device='cuda:0'), covar=tensor([0.1675, 0.2379, 0.1411, 0.1737], device='cuda:0'), in_proj_covar=tensor([0.0857, 0.0685, 0.0903, 0.0805], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-07 23:48:11,596 INFO [train.py:968] (0/2) Epoch 15, batch 33150, giga_loss[loss=0.2533, simple_loss=0.3263, pruned_loss=0.09015, over 28930.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3342, pruned_loss=0.0886, over 5675003.56 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.352, pruned_loss=0.1155, over 5709370.43 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3333, pruned_loss=0.08662, over 5663992.32 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:48:41,890 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-07 23:49:06,783 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=672115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:49:12,472 INFO [train.py:968] (0/2) Epoch 15, batch 33200, libri_loss[loss=0.2908, simple_loss=0.3526, pruned_loss=0.1145, over 29523.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3305, pruned_loss=0.08606, over 5680002.62 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3514, pruned_loss=0.1151, over 5710267.79 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3299, pruned_loss=0.08418, over 5669286.60 frames. ], batch size: 89, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:49:43,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.167e+02 1.202e+03 1.467e+03 2.066e+03 8.187e+03, threshold=2.935e+03, percent-clipped=11.0
+2023-03-07 23:50:02,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:50:09,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3299, 1.1962, 1.1685, 1.5129], device='cuda:0'), covar=tensor([0.0711, 0.0454, 0.0358, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0063, 0.0058, 0.0098], device='cuda:0')
+2023-03-07 23:50:09,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3973, 3.2143, 1.4970, 1.5397], device='cuda:0'), covar=tensor([0.0965, 0.0297, 0.0935, 0.1336], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0526, 0.0359, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 23:50:13,279 INFO [train.py:968] (0/2) Epoch 15, batch 33250, giga_loss[loss=0.2561, simple_loss=0.3207, pruned_loss=0.09576, over 24531.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3287, pruned_loss=0.08537, over 5683927.29 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3502, pruned_loss=0.1144, over 5717464.27 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3285, pruned_loss=0.08344, over 5668121.47 frames. ], batch size: 705, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:50:50,431 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=672201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:51:00,426 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:51:16,211 INFO [train.py:968] (0/2) Epoch 15, batch 33300, libri_loss[loss=0.301, simple_loss=0.3662, pruned_loss=0.1179, over 29144.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3279, pruned_loss=0.0856, over 5682386.66 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3504, pruned_loss=0.1145, over 5716208.39 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3269, pruned_loss=0.0832, over 5669973.30 frames. ], batch size: 101, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:51:48,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.480e+02 1.397e+03 1.840e+03 2.637e+03 9.865e+03, threshold=3.679e+03, percent-clipped=22.0
+2023-03-07 23:52:02,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4255, 3.1496, 1.5032, 1.5216], device='cuda:0'), covar=tensor([0.0936, 0.0242, 0.0917, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0525, 0.0358, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-07 23:52:16,486 INFO [train.py:968] (0/2) Epoch 15, batch 33350, giga_loss[loss=0.2798, simple_loss=0.3544, pruned_loss=0.1026, over 27654.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3303, pruned_loss=0.0871, over 5684349.23 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3503, pruned_loss=0.1145, over 5722103.94 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3291, pruned_loss=0.08447, over 5668298.18 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:53:04,323 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:53:10,986 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:53:27,618 INFO [train.py:968] (0/2) Epoch 15, batch 33400, giga_loss[loss=0.2481, simple_loss=0.3285, pruned_loss=0.08385, over 28685.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3318, pruned_loss=0.08752, over 5682080.22 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3505, pruned_loss=0.1145, over 5721653.18 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3303, pruned_loss=0.08494, over 5669188.87 frames. ], batch size: 242, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:53:27,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672319.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:53:49,310 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:53:55,529 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 23:54:08,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.423e+02 1.350e+03 1.860e+03 2.403e+03 4.356e+03, threshold=3.719e+03, percent-clipped=3.0
+2023-03-07 23:54:12,574 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:54:14,954 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:54:25,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:54:37,622 INFO [train.py:968] (0/2) Epoch 15, batch 33450, giga_loss[loss=0.2258, simple_loss=0.2892, pruned_loss=0.08124, over 24497.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3316, pruned_loss=0.088, over 5676208.02 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3499, pruned_loss=0.1141, over 5725873.91 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3305, pruned_loss=0.08568, over 5661220.73 frames. ], batch size: 705, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:54:54,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:55:41,875 INFO [train.py:968] (0/2) Epoch 15, batch 33500, giga_loss[loss=0.2665, simple_loss=0.3525, pruned_loss=0.09025, over 28332.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.337, pruned_loss=0.09192, over 5657282.55 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3502, pruned_loss=0.1144, over 5716155.45 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3355, pruned_loss=0.08919, over 5651662.20 frames. ], batch size: 368, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:56:12,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.821e+02 1.334e+03 1.597e+03 2.564e+03 1.121e+04, threshold=3.195e+03, percent-clipped=10.0
+2023-03-07 23:56:29,402 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672462.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:56:31,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672465.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:56:35,779 INFO [train.py:968] (0/2) Epoch 15, batch 33550, giga_loss[loss=0.2605, simple_loss=0.3511, pruned_loss=0.08496, over 28519.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.339, pruned_loss=0.09228, over 5662334.10 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3495, pruned_loss=0.114, over 5713473.95 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3379, pruned_loss=0.08955, over 5658715.13 frames. ], batch size: 336, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:56:59,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:57:04,549 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672494.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:57:23,066 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:57:26,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:57:42,657 INFO [train.py:968] (0/2) Epoch 15, batch 33600, giga_loss[loss=0.2142, simple_loss=0.3003, pruned_loss=0.06399, over 28882.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3397, pruned_loss=0.09281, over 5669259.99 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3492, pruned_loss=0.114, over 5719935.85 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3387, pruned_loss=0.08982, over 5658866.21 frames. ], batch size: 119, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:58:08,785 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:58:25,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.021e+02 1.344e+03 1.603e+03 2.213e+03 7.601e+03, threshold=3.207e+03, percent-clipped=8.0
+2023-03-07 23:58:53,639 INFO [train.py:968] (0/2) Epoch 15, batch 33650, libri_loss[loss=0.2502, simple_loss=0.3283, pruned_loss=0.0861, over 29497.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3368, pruned_loss=0.09122, over 5672350.42 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3488, pruned_loss=0.1138, over 5721928.65 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3359, pruned_loss=0.08819, over 5660247.71 frames. ], batch size: 85, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:58:59,531 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:00:00,140 INFO [train.py:968] (0/2) Epoch 15, batch 33700, giga_loss[loss=0.2596, simple_loss=0.3403, pruned_loss=0.08942, over 28960.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3356, pruned_loss=0.09067, over 5676161.04 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3487, pruned_loss=0.1138, over 5715053.30 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3348, pruned_loss=0.08774, over 5672044.00 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:00:19,281 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:00:22,745 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:00:44,853 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.262e+02 1.369e+03 1.769e+03 2.556e+03 4.829e+03, threshold=3.539e+03, percent-clipped=13.0
+2023-03-08 00:00:57,894 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-08 00:01:04,363 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:01:07,904 INFO [train.py:968] (0/2) Epoch 15, batch 33750, giga_loss[loss=0.2378, simple_loss=0.3202, pruned_loss=0.07764, over 28919.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3351, pruned_loss=0.09064, over 5670774.92 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3487, pruned_loss=0.114, over 5711031.51 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.334, pruned_loss=0.08744, over 5669874.64 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:02:11,279 INFO [train.py:968] (0/2) Epoch 15, batch 33800, giga_loss[loss=0.2356, simple_loss=0.3133, pruned_loss=0.07897, over 28759.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3333, pruned_loss=0.09058, over 5675025.43 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3486, pruned_loss=0.114, over 5717082.89 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3321, pruned_loss=0.08726, over 5667662.59 frames. ], batch size: 243, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:02:13,773 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:02:15,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:02:22,395 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=672726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:02:53,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.154e+02 1.431e+03 1.918e+03 2.411e+03 6.115e+03, threshold=3.836e+03, percent-clipped=6.0
+2023-03-08 00:02:55,630 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:03:19,436 INFO [train.py:968] (0/2) Epoch 15, batch 33850, giga_loss[loss=0.2668, simple_loss=0.3406, pruned_loss=0.09651, over 27515.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3317, pruned_loss=0.08984, over 5673290.81 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3491, pruned_loss=0.1143, over 5709281.05 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.33, pruned_loss=0.08664, over 5673449.84 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:03:22,684 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 00:03:42,372 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3289, 1.3233, 1.4872, 1.1020], device='cuda:0'), covar=tensor([0.1810, 0.3087, 0.1484, 0.1613], device='cuda:0'), in_proj_covar=tensor([0.0856, 0.0684, 0.0902, 0.0803], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 00:04:19,554 INFO [train.py:968] (0/2) Epoch 15, batch 33900, giga_loss[loss=0.2377, simple_loss=0.3249, pruned_loss=0.07527, over 28389.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3319, pruned_loss=0.0888, over 5668188.29 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3486, pruned_loss=0.1141, over 5703035.41 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3307, pruned_loss=0.08578, over 5673579.77 frames. ], batch size: 369, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:05:00,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.982e+02 1.353e+03 1.663e+03 2.352e+03 6.954e+03, threshold=3.326e+03, percent-clipped=8.0
+2023-03-08 00:05:22,880 INFO [train.py:968] (0/2) Epoch 15, batch 33950, giga_loss[loss=0.2651, simple_loss=0.354, pruned_loss=0.08806, over 28922.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3306, pruned_loss=0.08665, over 5667721.99 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3491, pruned_loss=0.1143, over 5701557.17 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3289, pruned_loss=0.08359, over 5672367.48 frames. ], batch size: 227, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:06:07,912 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=672905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:06:21,653 INFO [train.py:968] (0/2) Epoch 15, batch 34000, giga_loss[loss=0.2291, simple_loss=0.3197, pruned_loss=0.06928, over 28829.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3329, pruned_loss=0.08613, over 5676419.59 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3493, pruned_loss=0.1144, over 5706009.64 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3309, pruned_loss=0.08293, over 5675456.43 frames. ], batch size: 164, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:06:31,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3230, 1.5907, 1.5710, 1.1574], device='cuda:0'), covar=tensor([0.1689, 0.2539, 0.1472, 0.1723], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0683, 0.0904, 0.0804], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 00:06:54,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6746, 1.9965, 1.5677, 1.8228], device='cuda:0'), covar=tensor([0.2430, 0.2273, 0.2753, 0.2064], device='cuda:0'), in_proj_covar=tensor([0.1390, 0.1013, 0.1235, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 00:07:01,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.168e+02 1.353e+03 1.668e+03 2.477e+03 9.306e+03, threshold=3.337e+03, percent-clipped=8.0
+2023-03-08 00:07:24,499 INFO [train.py:968] (0/2) Epoch 15, batch 34050, libri_loss[loss=0.269, simple_loss=0.3296, pruned_loss=0.1043, over 29399.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3333, pruned_loss=0.08539, over 5680879.69 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3493, pruned_loss=0.1144, over 5707870.56 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3316, pruned_loss=0.08258, over 5678140.62 frames. ], batch size: 71, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:08:11,944 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6907, 3.5361, 3.3107, 1.6646], device='cuda:0'), covar=tensor([0.0759, 0.0875, 0.0871, 0.2446], device='cuda:0'), in_proj_covar=tensor([0.1110, 0.1017, 0.0879, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-08 00:08:33,338 INFO [train.py:968] (0/2) Epoch 15, batch 34100, giga_loss[loss=0.2836, simple_loss=0.3532, pruned_loss=0.107, over 28642.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3333, pruned_loss=0.086, over 5679709.28 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3488, pruned_loss=0.1141, over 5711679.91 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3317, pruned_loss=0.08275, over 5672830.99 frames. ], batch size: 262, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:09:15,942 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.312e+02 1.229e+03 1.567e+03 2.237e+03 4.864e+03, threshold=3.135e+03, percent-clipped=4.0
+2023-03-08 00:09:41,139 INFO [train.py:968] (0/2) Epoch 15, batch 34150, giga_loss[loss=0.235, simple_loss=0.3238, pruned_loss=0.07307, over 28623.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3333, pruned_loss=0.08624, over 5664930.81 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3485, pruned_loss=0.1139, over 5704456.65 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3318, pruned_loss=0.08316, over 5665105.50 frames. ], batch size: 307, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:10:21,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=673101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:10:43,919 INFO [train.py:968] (0/2) Epoch 15, batch 34200, libri_loss[loss=0.2628, simple_loss=0.3228, pruned_loss=0.1014, over 29489.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3335, pruned_loss=0.08689, over 5639248.46 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3485, pruned_loss=0.1141, over 5678027.65 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3317, pruned_loss=0.08302, over 5660763.00 frames. ], batch size: 70, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:11:28,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.400e+02 1.454e+03 1.966e+03 2.573e+03 1.287e+04, threshold=3.932e+03, percent-clipped=12.0
+2023-03-08 00:12:00,663 INFO [train.py:968] (0/2) Epoch 15, batch 34250, giga_loss[loss=0.2347, simple_loss=0.3217, pruned_loss=0.07387, over 28508.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3332, pruned_loss=0.08598, over 5649457.62 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3476, pruned_loss=0.1136, over 5682952.29 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3322, pruned_loss=0.08268, over 5661704.62 frames. ], batch size: 242, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:13:09,094 INFO [train.py:968] (0/2) Epoch 15, batch 34300, giga_loss[loss=0.2603, simple_loss=0.3472, pruned_loss=0.0867, over 28456.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3365, pruned_loss=0.08761, over 5658603.66 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3477, pruned_loss=0.1137, over 5687129.84 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3354, pruned_loss=0.08442, over 5664167.93 frames. ], batch size: 336, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:13:23,930 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 00:13:42,414 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=673244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:13:46,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=673247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:13:50,805 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.807e+02 1.366e+03 1.894e+03 2.563e+03 5.882e+03, threshold=3.787e+03, percent-clipped=5.0
+2023-03-08 00:14:18,665 INFO [train.py:968] (0/2) Epoch 15, batch 34350, giga_loss[loss=0.2149, simple_loss=0.3091, pruned_loss=0.06038, over 28930.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3385, pruned_loss=0.0882, over 5667324.10 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3477, pruned_loss=0.1137, over 5691177.43 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3374, pruned_loss=0.0852, over 5667642.04 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:14:20,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4354, 3.8252, 1.5999, 1.7294], device='cuda:0'), covar=tensor([0.0952, 0.0294, 0.0897, 0.1224], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0526, 0.0360, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 00:14:25,993 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2043, 1.1665, 3.7827, 3.0319], device='cuda:0'), covar=tensor([0.1678, 0.2802, 0.0390, 0.1212], device='cuda:0'), in_proj_covar=tensor([0.0690, 0.0606, 0.0881, 0.0801], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 00:14:29,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=673276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:14:35,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=673280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:14:35,354 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-08 00:15:35,637 INFO [train.py:968] (0/2) Epoch 15, batch 34400, giga_loss[loss=0.2354, simple_loss=0.3151, pruned_loss=0.07791, over 28502.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3368, pruned_loss=0.08767, over 5668337.73 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3477, pruned_loss=0.1136, over 5693329.26 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3359, pruned_loss=0.08516, over 5666521.23 frames. ], batch size: 78, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:16:16,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.897e+02 1.348e+03 1.838e+03 2.526e+03 6.844e+03, threshold=3.676e+03, percent-clipped=8.0
+2023-03-08 00:16:42,658 INFO [train.py:968] (0/2) Epoch 15, batch 34450, giga_loss[loss=0.2557, simple_loss=0.3306, pruned_loss=0.09042, over 28876.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3344, pruned_loss=0.08693, over 5690024.03 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3474, pruned_loss=0.1134, over 5699895.66 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3334, pruned_loss=0.08389, over 5681963.27 frames. ], batch size: 213, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:17:50,871 INFO [train.py:968] (0/2) Epoch 15, batch 34500, giga_loss[loss=0.2266, simple_loss=0.3206, pruned_loss=0.06629, over 28710.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.333, pruned_loss=0.0859, over 5688370.88 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3471, pruned_loss=0.1131, over 5705323.00 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3319, pruned_loss=0.08259, over 5676176.15 frames. ], batch size: 307, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:17:57,064 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=673423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:18:02,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=673426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:18:39,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.538e+02 1.215e+03 1.639e+03 2.192e+03 7.672e+03, threshold=3.278e+03, percent-clipped=4.0
+2023-03-08 00:18:42,905 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=673455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:19:01,667 INFO [train.py:968] (0/2) Epoch 15, batch 34550, giga_loss[loss=0.282, simple_loss=0.354, pruned_loss=0.105, over 27037.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3332, pruned_loss=0.08637, over 5667447.96 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3478, pruned_loss=0.1137, over 5695891.24 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3313, pruned_loss=0.08249, over 5666246.44 frames. ], batch size: 555, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:20:04,412 INFO [train.py:968] (0/2) Epoch 15, batch 34600, giga_loss[loss=0.2818, simple_loss=0.3628, pruned_loss=0.1005, over 28626.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.336, pruned_loss=0.08817, over 5668540.44 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3471, pruned_loss=0.1132, over 5699621.69 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3349, pruned_loss=0.08503, over 5663760.30 frames. ], batch size: 242, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:20:45,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.880e+02 1.556e+03 2.012e+03 2.521e+03 6.675e+03, threshold=4.025e+03, percent-clipped=14.0
+2023-03-08 00:21:07,292 INFO [train.py:968] (0/2) Epoch 15, batch 34650, giga_loss[loss=0.241, simple_loss=0.3285, pruned_loss=0.07681, over 28358.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3381, pruned_loss=0.08907, over 5682069.31 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3469, pruned_loss=0.1131, over 5704921.02 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3372, pruned_loss=0.08613, over 5672819.65 frames. ], batch size: 65, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:22:10,402 INFO [train.py:968] (0/2) Epoch 15, batch 34700, giga_loss[loss=0.2278, simple_loss=0.3121, pruned_loss=0.07181, over 28632.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3356, pruned_loss=0.08884, over 5664511.08 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3467, pruned_loss=0.1131, over 5696984.70 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3348, pruned_loss=0.08601, over 5664122.92 frames. ], batch size: 307, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:22:47,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.394e+02 1.551e+03 2.258e+03 3.806e+03 7.294e+03, threshold=4.516e+03, percent-clipped=22.0
+2023-03-08 00:22:50,911 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=673654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:23:05,677 INFO [train.py:968] (0/2) Epoch 15, batch 34750, giga_loss[loss=0.2764, simple_loss=0.3538, pruned_loss=0.09949, over 28368.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3343, pruned_loss=0.08915, over 5671042.99 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3462, pruned_loss=0.1126, over 5705888.94 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3337, pruned_loss=0.08629, over 5661023.71 frames. ], batch size: 368, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:24:06,537 INFO [train.py:968] (0/2) Epoch 15, batch 34800, giga_loss[loss=0.204, simple_loss=0.279, pruned_loss=0.06449, over 24105.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3352, pruned_loss=0.08975, over 5674999.75 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3462, pruned_loss=0.1126, over 5708557.29 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3345, pruned_loss=0.08712, over 5664228.24 frames. ], batch size: 705, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:24:37,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.264e+02 1.409e+03 1.878e+03 2.609e+03 6.613e+03, threshold=3.756e+03, percent-clipped=7.0
+2023-03-08 00:24:54,696 INFO [train.py:968] (0/2) Epoch 15, batch 34850, giga_loss[loss=0.3408, simple_loss=0.4115, pruned_loss=0.1351, over 28607.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3445, pruned_loss=0.09528, over 5670510.86 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3463, pruned_loss=0.1127, over 5702356.74 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3436, pruned_loss=0.09263, over 5666270.41 frames. ], batch size: 85, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:25:12,541 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=673789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:25:43,538 INFO [train.py:968] (0/2) Epoch 15, batch 34900, giga_loss[loss=0.2645, simple_loss=0.3555, pruned_loss=0.08673, over 28660.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3526, pruned_loss=0.1003, over 5675245.32 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3464, pruned_loss=0.1129, over 5707169.67 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3518, pruned_loss=0.09749, over 5666833.30 frames. ], batch size: 262, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:26:12,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.176e+02 1.250e+03 1.573e+03 2.217e+03 4.844e+03, threshold=3.146e+03, percent-clipped=3.0
+2023-03-08 00:26:23,051 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-08 00:26:26,815 INFO [train.py:968] (0/2) Epoch 15, batch 34950, giga_loss[loss=0.2913, simple_loss=0.3683, pruned_loss=0.1071, over 28953.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3548, pruned_loss=0.1021, over 5680270.35 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3467, pruned_loss=0.1128, over 5712448.11 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3542, pruned_loss=0.09965, over 5668105.52 frames. ], batch size: 106, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:26:54,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3195, 4.1442, 3.8908, 1.8411], device='cuda:0'), covar=tensor([0.0539, 0.0704, 0.0694, 0.2173], device='cuda:0'), in_proj_covar=tensor([0.1108, 0.1018, 0.0877, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0013, 0.0011, 0.0011], device='cuda:0')
+2023-03-08 00:27:10,293 INFO [train.py:968] (0/2) Epoch 15, batch 35000, giga_loss[loss=0.2516, simple_loss=0.3271, pruned_loss=0.08807, over 28966.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3517, pruned_loss=0.1024, over 5673421.08 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3476, pruned_loss=0.1137, over 5697956.90 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3505, pruned_loss=0.0993, over 5676926.62 frames. ], batch size: 136, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:27:32,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5961, 1.9205, 1.5521, 1.7963], device='cuda:0'), covar=tensor([0.2485, 0.2429, 0.2778, 0.2244], device='cuda:0'), in_proj_covar=tensor([0.1383, 0.1012, 0.1231, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-08 00:27:38,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.790e+02 1.235e+03 1.723e+03 2.464e+03 6.240e+03, threshold=3.446e+03, percent-clipped=13.0
+2023-03-08 00:27:54,987 INFO [train.py:968] (0/2) Epoch 15, batch 35050, giga_loss[loss=0.2757, simple_loss=0.3331, pruned_loss=0.1091, over 26631.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3458, pruned_loss=0.1001, over 5679289.77 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3479, pruned_loss=0.1137, over 5699649.59 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3446, pruned_loss=0.09738, over 5680283.25 frames. ], batch size: 555, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:28:22,244 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-674000.pt
+2023-03-08 00:28:37,746 INFO [train.py:968] (0/2) Epoch 15, batch 35100, giga_loss[loss=0.2187, simple_loss=0.2879, pruned_loss=0.07475, over 28886.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3388, pruned_loss=0.0972, over 5679682.99 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3488, pruned_loss=0.1143, over 5698827.22 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3369, pruned_loss=0.09424, over 5680844.00 frames. ], batch size: 112, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:28:47,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:29:06,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.782e+02 1.106e+03 1.496e+03 2.280e+03 4.204e+03, threshold=2.992e+03, percent-clipped=6.0
+2023-03-08 00:29:19,739 INFO [train.py:968] (0/2) Epoch 15, batch 35150, giga_loss[loss=0.2284, simple_loss=0.2913, pruned_loss=0.08273, over 28272.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.33, pruned_loss=0.09271, over 5686435.54 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3488, pruned_loss=0.114, over 5701491.52 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3282, pruned_loss=0.09027, over 5684653.97 frames. ], batch size: 77, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:29:20,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:29:56,087 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:30:05,678 INFO [train.py:968] (0/2) Epoch 15, batch 35200, libri_loss[loss=0.2363, simple_loss=0.3087, pruned_loss=0.08196, over 29382.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3232, pruned_loss=0.08954, over 5679736.80 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3491, pruned_loss=0.1141, over 5694919.05 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3211, pruned_loss=0.0871, over 5684572.98 frames. ], batch size: 67, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:30:34,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.944e+02 9.199e+02 1.138e+03 1.503e+03 4.223e+03, threshold=2.276e+03, percent-clipped=3.0
+2023-03-08 00:30:43,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:30:48,187 INFO [train.py:968] (0/2) Epoch 15, batch 35250, giga_loss[loss=0.2242, simple_loss=0.2969, pruned_loss=0.07572, over 28515.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.319, pruned_loss=0.08768, over 5674355.54 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3489, pruned_loss=0.1139, over 5687937.41 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3171, pruned_loss=0.08555, over 5684678.22 frames. ], batch size: 78, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:30:50,837 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:30:56,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:31:04,456 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:31:21,531 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:31:31,544 INFO [train.py:968] (0/2) Epoch 15, batch 35300, libri_loss[loss=0.3338, simple_loss=0.393, pruned_loss=0.1372, over 25756.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3168, pruned_loss=0.08657, over 5681078.00 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3492, pruned_loss=0.1139, over 5687838.10 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3139, pruned_loss=0.08383, over 5689118.98 frames. ], batch size: 136, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:31:50,155 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4859, 3.3690, 1.6147, 1.5132], device='cuda:0'), covar=tensor([0.0933, 0.0319, 0.0852, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0524, 0.0359, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 00:31:57,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9390, 1.1195, 0.9784, 0.8574], device='cuda:0'), covar=tensor([0.1911, 0.2064, 0.1362, 0.1790], device='cuda:0'), in_proj_covar=tensor([0.1806, 0.1714, 0.1631, 0.1793], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 00:32:00,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.539e+02 1.100e+03 1.531e+03 2.227e+03 4.921e+03, threshold=3.063e+03, percent-clipped=24.0
+2023-03-08 00:32:12,560 INFO [train.py:968] (0/2) Epoch 15, batch 35350, giga_loss[loss=0.2154, simple_loss=0.2889, pruned_loss=0.07098, over 28861.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3146, pruned_loss=0.08546, over 5696413.99 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3494, pruned_loss=0.1138, over 5694985.96 frames. ], giga_tot_loss[loss=0.2379, simple_loss=0.3109, pruned_loss=0.08248, over 5696247.02 frames. ], batch size: 112, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:32:27,096 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:32:50,994 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:32:53,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:33:00,920 INFO [train.py:968] (0/2) Epoch 15, batch 35400, giga_loss[loss=0.2348, simple_loss=0.3027, pruned_loss=0.08347, over 27670.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3111, pruned_loss=0.08365, over 5704684.75 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3494, pruned_loss=0.1138, over 5698444.98 frames. ], giga_tot_loss[loss=0.2346, simple_loss=0.3076, pruned_loss=0.08083, over 5701446.53 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:33:18,994 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:33:19,010 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:33:29,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.707e+02 1.088e+03 1.427e+03 2.066e+03 5.639e+03, threshold=2.854e+03, percent-clipped=14.0
+2023-03-08 00:33:43,719 INFO [train.py:968] (0/2) Epoch 15, batch 35450, giga_loss[loss=0.2231, simple_loss=0.2994, pruned_loss=0.07343, over 27859.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3069, pruned_loss=0.08146, over 5700338.27 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3487, pruned_loss=0.1133, over 5702261.33 frames. ], giga_tot_loss[loss=0.2308, simple_loss=0.3038, pruned_loss=0.07892, over 5694535.79 frames. ], batch size: 412, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:34:27,862 INFO [train.py:968] (0/2) Epoch 15, batch 35500, giga_loss[loss=0.188, simple_loss=0.2682, pruned_loss=0.05392, over 28877.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3046, pruned_loss=0.08036, over 5700698.45 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3489, pruned_loss=0.1133, over 5704324.76 frames. ], giga_tot_loss[loss=0.2289, simple_loss=0.3016, pruned_loss=0.07806, over 5694295.22 frames. ], batch size: 174, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:34:51,862 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:34:56,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.047e+02 9.589e+02 1.281e+03 1.863e+03 6.773e+03, threshold=2.562e+03, percent-clipped=8.0
+2023-03-08 00:35:11,247 INFO [train.py:968] (0/2) Epoch 15, batch 35550, giga_loss[loss=0.2329, simple_loss=0.3033, pruned_loss=0.08123, over 28733.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3041, pruned_loss=0.08037, over 5702785.95 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3492, pruned_loss=0.1131, over 5711786.23 frames. ], giga_tot_loss[loss=0.2273, simple_loss=0.2997, pruned_loss=0.07746, over 5690527.95 frames. ], batch size: 262, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:35:19,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:35:54,321 INFO [train.py:968] (0/2) Epoch 15, batch 35600, giga_loss[loss=0.164, simple_loss=0.2388, pruned_loss=0.04465, over 28297.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3022, pruned_loss=0.07992, over 5706859.38 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3501, pruned_loss=0.1137, over 5712780.50 frames. ], giga_tot_loss[loss=0.2241, simple_loss=0.2963, pruned_loss=0.076, over 5696029.42 frames. ], batch size: 60, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:36:21,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.128e+02 1.026e+03 1.328e+03 1.905e+03 5.113e+03, threshold=2.657e+03, percent-clipped=16.0
+2023-03-08 00:36:28,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:36:35,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5087, 2.9230, 2.4456, 2.0197], device='cuda:0'), covar=tensor([0.2558, 0.1659, 0.1841, 0.2499], device='cuda:0'), in_proj_covar=tensor([0.1830, 0.1738, 0.1648, 0.1814], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 00:36:37,284 INFO [train.py:968] (0/2) Epoch 15, batch 35650, giga_loss[loss=0.2703, simple_loss=0.3403, pruned_loss=0.1002, over 28944.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3037, pruned_loss=0.08127, over 5700577.93 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3503, pruned_loss=0.1136, over 5719717.97 frames. ], giga_tot_loss[loss=0.2258, simple_loss=0.2973, pruned_loss=0.07712, over 5685658.53 frames. ], batch size: 213, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:36:55,590 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:36:58,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:37:07,918 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4898, 2.4360, 2.3956, 2.3136], device='cuda:0'), covar=tensor([0.0817, 0.1141, 0.1067, 0.0949], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0728, 0.0680, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 00:37:10,134 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 00:37:23,237 INFO [train.py:968] (0/2) Epoch 15, batch 35700, giga_loss[loss=0.2791, simple_loss=0.3516, pruned_loss=0.1033, over 28487.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3137, pruned_loss=0.08619, over 5691278.32 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3503, pruned_loss=0.1136, over 5712608.79 frames. ], giga_tot_loss[loss=0.2359, simple_loss=0.3076, pruned_loss=0.08213, over 5685496.03 frames. ], batch size: 85, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:37:26,169 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:37:28,195 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:37:31,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:00,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.330e+02 1.203e+03 1.618e+03 2.260e+03 5.183e+03, threshold=3.236e+03, percent-clipped=20.0
+2023-03-08 00:38:00,903 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:05,409 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:13,650 INFO [train.py:968] (0/2) Epoch 15, batch 35750, giga_loss[loss=0.2679, simple_loss=0.347, pruned_loss=0.09433, over 28614.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3255, pruned_loss=0.0919, over 5682004.58 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3505, pruned_loss=0.1137, over 5703946.67 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3203, pruned_loss=0.08838, over 5685496.46 frames. ], batch size: 307, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:38:41,575 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:43,559 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:43,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:51,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:55,970 INFO [train.py:968] (0/2) Epoch 15, batch 35800, giga_loss[loss=0.3074, simple_loss=0.3845, pruned_loss=0.1151, over 28906.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3361, pruned_loss=0.09688, over 5691047.67 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3507, pruned_loss=0.1137, over 5711049.89 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.331, pruned_loss=0.09346, over 5686972.05 frames. ], batch size: 136, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:39:10,042 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:39:27,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.551e+02 1.223e+03 1.504e+03 2.099e+03 4.922e+03, threshold=3.008e+03, percent-clipped=7.0
+2023-03-08 00:39:41,651 INFO [train.py:968] (0/2) Epoch 15, batch 35850, giga_loss[loss=0.2944, simple_loss=0.3681, pruned_loss=0.1103, over 29066.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3415, pruned_loss=0.09844, over 5684421.23 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3512, pruned_loss=0.114, over 5700048.09 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3369, pruned_loss=0.0953, over 5690965.70 frames. ], batch size: 128, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:40:13,239 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:40:17,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:40:27,723 INFO [train.py:968] (0/2) Epoch 15, batch 35900, giga_loss[loss=0.2613, simple_loss=0.3457, pruned_loss=0.08845, over 28677.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3431, pruned_loss=0.09798, over 5673047.65 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3515, pruned_loss=0.1141, over 5687969.90 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3389, pruned_loss=0.09486, over 5688214.64 frames. ], batch size: 242, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:40:29,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674821.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:40:43,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:40:51,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6467, 4.2232, 1.8624, 1.8146], device='cuda:0'), covar=tensor([0.0967, 0.0228, 0.0810, 0.1264], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0522, 0.0358, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 00:40:57,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.306e+02 1.212e+03 1.520e+03 2.359e+03 1.047e+04, threshold=3.040e+03, percent-clipped=12.0
+2023-03-08 00:41:02,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:41:04,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:41:14,762 INFO [train.py:968] (0/2) Epoch 15, batch 35950, giga_loss[loss=0.2541, simple_loss=0.3337, pruned_loss=0.08721, over 28908.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.346, pruned_loss=0.09973, over 5670744.14 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3517, pruned_loss=0.1143, over 5686629.34 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3421, pruned_loss=0.09635, over 5683536.88 frames. ], batch size: 119, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:41:21,992 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674878.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:41:32,835 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:42:01,737 INFO [train.py:968] (0/2) Epoch 15, batch 36000, giga_loss[loss=0.2739, simple_loss=0.3462, pruned_loss=0.1008, over 28698.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3487, pruned_loss=0.1014, over 5676422.16 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3518, pruned_loss=0.1143, over 5687893.20 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3455, pruned_loss=0.09859, over 5685119.76 frames. ], batch size: 92, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:42:01,742 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 00:42:09,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4078, 1.7430, 1.3667, 1.3432], device='cuda:0'), covar=tensor([0.2756, 0.2605, 0.2870, 0.2242], device='cuda:0'), in_proj_covar=tensor([0.1384, 0.1014, 0.1230, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-08 00:42:09,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2127, 1.6533, 1.5270, 1.0959], device='cuda:0'), covar=tensor([0.1725, 0.2574, 0.1537, 0.1742], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0690, 0.0908, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 00:42:10,747 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2079, simple_loss=0.3149, pruned_loss=0.0505, over 944034.00 frames. 
+2023-03-08 00:42:10,748 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 00:42:42,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.257e+02 1.197e+03 1.525e+03 1.863e+03 4.112e+03, threshold=3.051e+03, percent-clipped=3.0
+2023-03-08 00:42:50,058 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-08 00:42:53,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8639, 2.0093, 1.8772, 1.7333], device='cuda:0'), covar=tensor([0.1895, 0.2428, 0.2339, 0.2295], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0733, 0.0685, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 00:42:56,186 INFO [train.py:968] (0/2) Epoch 15, batch 36050, giga_loss[loss=0.273, simple_loss=0.3455, pruned_loss=0.1003, over 28418.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3517, pruned_loss=0.1036, over 5674675.54 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.352, pruned_loss=0.1143, over 5691179.63 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.349, pruned_loss=0.1013, over 5678500.70 frames. ], batch size: 78, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:43:06,993 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-08 00:43:07,642 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 00:43:26,813 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-08 00:43:40,062 INFO [train.py:968] (0/2) Epoch 15, batch 36100, giga_loss[loss=0.2615, simple_loss=0.3584, pruned_loss=0.0823, over 28455.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3544, pruned_loss=0.1048, over 5675758.82 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3522, pruned_loss=0.1143, over 5685932.38 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.352, pruned_loss=0.1027, over 5683451.06 frames. ], batch size: 71, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:44:07,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.429e+02 1.247e+03 1.499e+03 2.079e+03 6.398e+03, threshold=2.997e+03, percent-clipped=5.0
+2023-03-08 00:44:17,599 INFO [train.py:968] (0/2) Epoch 15, batch 36150, giga_loss[loss=0.2657, simple_loss=0.3469, pruned_loss=0.09228, over 28784.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3578, pruned_loss=0.1066, over 5672663.03 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3528, pruned_loss=0.1145, over 5675150.51 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3555, pruned_loss=0.1043, over 5688845.87 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:44:18,077 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-08 00:44:27,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:45:05,194 INFO [train.py:968] (0/2) Epoch 15, batch 36200, giga_loss[loss=0.2647, simple_loss=0.3493, pruned_loss=0.09003, over 28835.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3588, pruned_loss=0.1063, over 5670516.56 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3529, pruned_loss=0.1145, over 5676450.75 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3569, pruned_loss=0.1046, over 5681982.62 frames. ], batch size: 199, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:45:34,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.886e+02 1.270e+03 1.642e+03 2.306e+03 7.310e+03, threshold=3.285e+03, percent-clipped=13.0
+2023-03-08 00:45:43,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=675165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:45:47,477 INFO [train.py:968] (0/2) Epoch 15, batch 36250, giga_loss[loss=0.2703, simple_loss=0.3511, pruned_loss=0.09479, over 28993.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3596, pruned_loss=0.106, over 5680878.56 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3533, pruned_loss=0.1147, over 5680670.76 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3579, pruned_loss=0.1042, over 5686188.97 frames. ], batch size: 128, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:46:10,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675196.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:46:28,464 INFO [train.py:968] (0/2) Epoch 15, batch 36300, giga_loss[loss=0.3014, simple_loss=0.3762, pruned_loss=0.1133, over 28772.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3589, pruned_loss=0.1044, over 5695407.09 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3536, pruned_loss=0.1148, over 5687935.03 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3574, pruned_loss=0.1026, over 5693464.71 frames. ], batch size: 242, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:46:31,431 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=675223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:46:31,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-08 00:46:34,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=675226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:46:35,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3756, 1.3688, 1.3091, 1.5262], device='cuda:0'), covar=tensor([0.0739, 0.0373, 0.0326, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 00:46:57,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675253.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:46:58,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.440e+02 1.159e+03 1.493e+03 1.829e+03 4.090e+03, threshold=2.987e+03, percent-clipped=2.0
+2023-03-08 00:46:58,992 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=675255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:47:11,321 INFO [train.py:968] (0/2) Epoch 15, batch 36350, giga_loss[loss=0.2496, simple_loss=0.3352, pruned_loss=0.08203, over 28346.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3571, pruned_loss=0.1022, over 5704745.22 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3543, pruned_loss=0.1151, over 5693764.28 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3554, pruned_loss=0.1002, over 5698081.93 frames. ], batch size: 65, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:47:11,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2260, 2.2467, 2.1175, 1.9371], device='cuda:0'), covar=tensor([0.1595, 0.2193, 0.1960, 0.2033], device='cuda:0'), in_proj_covar=tensor([0.0440, 0.0729, 0.0684, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 00:47:52,031 INFO [train.py:968] (0/2) Epoch 15, batch 36400, giga_loss[loss=0.2612, simple_loss=0.3488, pruned_loss=0.08683, over 28242.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3561, pruned_loss=0.1012, over 5714669.62 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3548, pruned_loss=0.1152, over 5699811.34 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3544, pruned_loss=0.09917, over 5704424.45 frames. ], batch size: 77, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:48:09,953 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=675339.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:48:10,520 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=675340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:48:12,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=675342.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:48:19,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4316, 3.2749, 1.4872, 1.4989], device='cuda:0'), covar=tensor([0.0988, 0.0281, 0.0880, 0.1379], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0523, 0.0357, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 00:48:24,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.775e+02 1.077e+03 1.411e+03 1.924e+03 7.511e+03, threshold=2.822e+03, percent-clipped=11.0
+2023-03-08 00:48:37,346 INFO [train.py:968] (0/2) Epoch 15, batch 36450, giga_loss[loss=0.295, simple_loss=0.3593, pruned_loss=0.1153, over 28796.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3581, pruned_loss=0.1045, over 5712488.78 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3552, pruned_loss=0.1154, over 5702414.89 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3564, pruned_loss=0.1025, over 5702081.78 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:48:40,094 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=675371.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:49:05,019 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=675396.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:49:08,002 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=675399.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:49:22,771 INFO [train.py:968] (0/2) Epoch 15, batch 36500, giga_loss[loss=0.2552, simple_loss=0.3332, pruned_loss=0.08863, over 28572.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3626, pruned_loss=0.1105, over 5708970.41 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3564, pruned_loss=0.1163, over 5707095.03 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3603, pruned_loss=0.1078, over 5696062.83 frames. ], batch size: 71, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:49:30,202 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=675428.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:49:56,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.135e+02 1.526e+03 2.162e+03 2.966e+03 7.968e+03, threshold=4.325e+03, percent-clipped=26.0
+2023-03-08 00:50:06,974 INFO [train.py:968] (0/2) Epoch 15, batch 36550, giga_loss[loss=0.2835, simple_loss=0.3553, pruned_loss=0.1059, over 28189.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3622, pruned_loss=0.1115, over 5710813.49 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3563, pruned_loss=0.1162, over 5712225.67 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3606, pruned_loss=0.1093, over 5695988.03 frames. ], batch size: 368, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:50:40,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-08 00:50:42,656 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=675505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:50:55,220 INFO [train.py:968] (0/2) Epoch 15, batch 36600, giga_loss[loss=0.2751, simple_loss=0.3501, pruned_loss=0.1001, over 28796.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3598, pruned_loss=0.1104, over 5693766.35 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3569, pruned_loss=0.1165, over 5695830.35 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3581, pruned_loss=0.1083, over 5697282.10 frames. ], batch size: 119, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:50:59,339 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=675524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:51:11,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:51:14,995 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=675543.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:51:26,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.059e+02 1.250e+03 1.733e+03 2.533e+03 7.197e+03, threshold=3.466e+03, percent-clipped=6.0
+2023-03-08 00:51:30,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4107, 1.7713, 1.4089, 1.4973], device='cuda:0'), covar=tensor([0.0776, 0.0306, 0.0321, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0063, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 00:51:38,330 INFO [train.py:968] (0/2) Epoch 15, batch 36650, giga_loss[loss=0.3245, simple_loss=0.3797, pruned_loss=0.1346, over 28616.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3582, pruned_loss=0.1098, over 5699337.47 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3571, pruned_loss=0.1167, over 5697658.20 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3567, pruned_loss=0.1078, over 5700525.00 frames. ], batch size: 92, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:52:18,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5516, 1.7565, 1.7826, 1.3355], device='cuda:0'), covar=tensor([0.1435, 0.2297, 0.1313, 0.1547], device='cuda:0'), in_proj_covar=tensor([0.0862, 0.0690, 0.0908, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 00:52:20,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3040, 1.4729, 1.3318, 1.2309], device='cuda:0'), covar=tensor([0.2403, 0.1944, 0.1842, 0.2090], device='cuda:0'), in_proj_covar=tensor([0.1835, 0.1751, 0.1666, 0.1825], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 00:52:21,184 INFO [train.py:968] (0/2) Epoch 15, batch 36700, libri_loss[loss=0.2546, simple_loss=0.3249, pruned_loss=0.09216, over 29670.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3583, pruned_loss=0.1095, over 5695146.84 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3577, pruned_loss=0.1171, over 5694473.59 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3566, pruned_loss=0.1073, over 5698554.22 frames. ], batch size: 73, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:52:55,763 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.771e+02 1.138e+03 1.401e+03 2.010e+03 5.758e+03, threshold=2.802e+03, percent-clipped=4.0
+2023-03-08 00:53:07,025 INFO [train.py:968] (0/2) Epoch 15, batch 36750, giga_loss[loss=0.2416, simple_loss=0.3277, pruned_loss=0.07772, over 28979.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3567, pruned_loss=0.1077, over 5678891.91 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3586, pruned_loss=0.1175, over 5687431.71 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3545, pruned_loss=0.1053, over 5687556.73 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:53:18,836 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=675683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:53:19,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5354, 2.0235, 1.3268, 0.8921], device='cuda:0'), covar=tensor([0.5714, 0.2943, 0.2891, 0.5261], device='cuda:0'), in_proj_covar=tensor([0.1631, 0.1563, 0.1532, 0.1337], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 00:53:21,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=675686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:53:51,240 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=675715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:53:51,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:53:53,576 INFO [train.py:968] (0/2) Epoch 15, batch 36800, giga_loss[loss=0.2214, simple_loss=0.3028, pruned_loss=0.06995, over 28966.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.351, pruned_loss=0.104, over 5674023.33 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3586, pruned_loss=0.1174, over 5691482.44 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3492, pruned_loss=0.102, over 5676907.17 frames. ], batch size: 136, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:54:32,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.378e+02 9.333e+02 1.185e+03 1.412e+03 9.239e+03, threshold=2.370e+03, percent-clipped=5.0
+2023-03-08 00:54:45,486 INFO [train.py:968] (0/2) Epoch 15, batch 36850, giga_loss[loss=0.2189, simple_loss=0.2991, pruned_loss=0.06933, over 28228.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3442, pruned_loss=0.1005, over 5667728.79 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3588, pruned_loss=0.1174, over 5695098.62 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3424, pruned_loss=0.09869, over 5666530.08 frames. ], batch size: 368, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:55:41,086 INFO [train.py:968] (0/2) Epoch 15, batch 36900, giga_loss[loss=0.2262, simple_loss=0.3118, pruned_loss=0.07026, over 28981.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3398, pruned_loss=0.09877, over 5660009.23 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3587, pruned_loss=0.1172, over 5699149.00 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3382, pruned_loss=0.09709, over 5654802.87 frames. ], batch size: 164, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:56:16,340 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.171e+02 1.102e+03 1.527e+03 2.706e+03 1.136e+04, threshold=3.054e+03, percent-clipped=29.0
+2023-03-08 00:56:16,631 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=675858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:56:19,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=675861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:56:26,535 INFO [train.py:968] (0/2) Epoch 15, batch 36950, giga_loss[loss=0.2617, simple_loss=0.3386, pruned_loss=0.09238, over 28943.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3386, pruned_loss=0.09772, over 5658423.08 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.359, pruned_loss=0.1174, over 5695357.22 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3364, pruned_loss=0.09567, over 5656867.12 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:56:36,512 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:56:45,380 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=675890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:56:53,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:57:09,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675918.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:57:09,460 INFO [train.py:968] (0/2) Epoch 15, batch 37000, giga_loss[loss=0.2591, simple_loss=0.3357, pruned_loss=0.09125, over 28734.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3398, pruned_loss=0.09781, over 5656389.55 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3596, pruned_loss=0.1177, over 5685461.73 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3369, pruned_loss=0.09533, over 5663677.55 frames. ], batch size: 119, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:57:35,841 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5105, 2.1746, 1.6230, 0.7871], device='cuda:0'), covar=tensor([0.5537, 0.2608, 0.3956, 0.5607], device='cuda:0'), in_proj_covar=tensor([0.1632, 0.1557, 0.1530, 0.1335], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 00:57:42,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.436e+02 1.057e+03 1.453e+03 2.297e+03 1.273e+04, threshold=2.907e+03, percent-clipped=12.0
+2023-03-08 00:57:53,315 INFO [train.py:968] (0/2) Epoch 15, batch 37050, giga_loss[loss=0.2382, simple_loss=0.3145, pruned_loss=0.08091, over 28579.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3389, pruned_loss=0.09707, over 5673509.87 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3602, pruned_loss=0.1181, over 5688699.10 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3358, pruned_loss=0.09452, over 5676041.87 frames. ], batch size: 60, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:58:20,748 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-676000.pt
+2023-03-08 00:58:31,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8473, 2.0098, 1.3675, 1.6282], device='cuda:0'), covar=tensor([0.0883, 0.0588, 0.1058, 0.1095], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0438, 0.0509, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 00:58:36,907 INFO [train.py:968] (0/2) Epoch 15, batch 37100, giga_loss[loss=0.2264, simple_loss=0.3111, pruned_loss=0.07084, over 28583.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3349, pruned_loss=0.09472, over 5690089.12 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3601, pruned_loss=0.1179, over 5689903.60 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3325, pruned_loss=0.09272, over 5690993.46 frames. ], batch size: 336, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:58:40,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:58:42,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:58:54,888 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-08 00:58:56,616 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:58:59,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:59:08,134 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=676055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:59:09,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.238e+02 9.670e+02 1.181e+03 1.629e+03 5.111e+03, threshold=2.362e+03, percent-clipped=4.0
+2023-03-08 00:59:13,505 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676061.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:59:16,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676064.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:59:20,242 INFO [train.py:968] (0/2) Epoch 15, batch 37150, giga_loss[loss=0.2531, simple_loss=0.3209, pruned_loss=0.09263, over 28640.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3321, pruned_loss=0.09339, over 5699520.57 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.36, pruned_loss=0.1177, over 5692208.55 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3301, pruned_loss=0.09181, over 5698308.58 frames. ], batch size: 85, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:59:25,185 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=676074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:59:39,484 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=676093.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:00:01,585 INFO [train.py:968] (0/2) Epoch 15, batch 37200, giga_loss[loss=0.204, simple_loss=0.2828, pruned_loss=0.06265, over 28629.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3307, pruned_loss=0.09294, over 5704368.02 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3611, pruned_loss=0.1183, over 5696179.92 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3275, pruned_loss=0.09074, over 5699924.64 frames. ], batch size: 60, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:00:06,621 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3534, 3.4837, 1.5130, 1.4644], device='cuda:0'), covar=tensor([0.1026, 0.0372, 0.0877, 0.1382], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0523, 0.0358, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 01:00:32,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.893e+02 1.040e+03 1.321e+03 1.854e+03 5.035e+03, threshold=2.642e+03, percent-clipped=14.0
+2023-03-08 01:00:44,681 INFO [train.py:968] (0/2) Epoch 15, batch 37250, giga_loss[loss=0.2492, simple_loss=0.3258, pruned_loss=0.08633, over 28613.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3283, pruned_loss=0.09201, over 5702099.59 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3613, pruned_loss=0.1183, over 5688276.54 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3254, pruned_loss=0.09003, over 5705645.49 frames. ], batch size: 307, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:01:01,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2153, 2.4342, 1.8466, 2.0525], device='cuda:0'), covar=tensor([0.1817, 0.1907, 0.2187, 0.2044], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0738, 0.0692, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 01:01:28,087 INFO [train.py:968] (0/2) Epoch 15, batch 37300, giga_loss[loss=0.2656, simple_loss=0.3332, pruned_loss=0.09897, over 28825.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3258, pruned_loss=0.09067, over 5708883.30 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3619, pruned_loss=0.1186, over 5689488.06 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3225, pruned_loss=0.08848, over 5710879.68 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:01:29,208 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-08 01:01:53,405 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=676248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:02:02,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.818e+02 9.745e+02 1.147e+03 1.608e+03 3.994e+03, threshold=2.295e+03, percent-clipped=3.0
+2023-03-08 01:02:10,308 INFO [train.py:968] (0/2) Epoch 15, batch 37350, giga_loss[loss=0.2382, simple_loss=0.3088, pruned_loss=0.08376, over 28819.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3227, pruned_loss=0.08902, over 5709381.44 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.362, pruned_loss=0.1187, over 5689176.62 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3199, pruned_loss=0.08715, over 5711311.54 frames. ], batch size: 119, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:02:23,978 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=676283.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:02:53,232 INFO [train.py:968] (0/2) Epoch 15, batch 37400, giga_loss[loss=0.2534, simple_loss=0.3313, pruned_loss=0.08771, over 28548.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3208, pruned_loss=0.08776, over 5711669.61 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3622, pruned_loss=0.1186, over 5691501.85 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.318, pruned_loss=0.08611, over 5711267.38 frames. ], batch size: 336, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:03:24,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.721e+02 1.034e+03 1.381e+03 1.873e+03 7.290e+03, threshold=2.762e+03, percent-clipped=15.0
+2023-03-08 01:03:35,817 INFO [train.py:968] (0/2) Epoch 15, batch 37450, giga_loss[loss=0.2588, simple_loss=0.3298, pruned_loss=0.09393, over 28985.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3211, pruned_loss=0.08799, over 5702785.04 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3627, pruned_loss=0.1187, over 5685582.71 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.318, pruned_loss=0.08619, over 5707231.00 frames. ], batch size: 227, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:04:09,498 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-08 01:04:17,132 INFO [train.py:968] (0/2) Epoch 15, batch 37500, giga_loss[loss=0.2678, simple_loss=0.3401, pruned_loss=0.09775, over 28895.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3228, pruned_loss=0.08873, over 5714493.31 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3633, pruned_loss=0.1187, over 5691510.95 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3187, pruned_loss=0.08647, over 5713469.94 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:04:39,101 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-08 01:04:51,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.139e+02 1.072e+03 1.466e+03 2.110e+03 7.075e+03, threshold=2.932e+03, percent-clipped=15.0
+2023-03-08 01:05:00,321 INFO [train.py:968] (0/2) Epoch 15, batch 37550, libri_loss[loss=0.334, simple_loss=0.4035, pruned_loss=0.1323, over 29636.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3286, pruned_loss=0.09244, over 5720396.62 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3636, pruned_loss=0.1186, over 5697391.68 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3237, pruned_loss=0.08982, over 5714808.11 frames. ], batch size: 91, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:05:07,507 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=676476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:05:13,398 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=676482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:05:21,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2227, 1.1053, 4.0244, 3.2608], device='cuda:0'), covar=tensor([0.1726, 0.2815, 0.0389, 0.0830], device='cuda:0'), in_proj_covar=tensor([0.0695, 0.0609, 0.0890, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 01:05:44,568 INFO [train.py:968] (0/2) Epoch 15, batch 37600, libri_loss[loss=0.2969, simple_loss=0.3556, pruned_loss=0.1191, over 29603.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3351, pruned_loss=0.0965, over 5720312.10 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3641, pruned_loss=0.1186, over 5706924.67 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3292, pruned_loss=0.09312, over 5707737.76 frames. ], batch size: 74, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 01:06:19,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6696, 4.4928, 4.2523, 1.7786], device='cuda:0'), covar=tensor([0.0483, 0.0630, 0.0643, 0.2170], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.1027, 0.0883, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-08 01:06:23,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.463e+02 1.379e+03 1.831e+03 2.596e+03 1.092e+04, threshold=3.662e+03, percent-clipped=16.0
+2023-03-08 01:06:32,970 INFO [train.py:968] (0/2) Epoch 15, batch 37650, giga_loss[loss=0.3053, simple_loss=0.3732, pruned_loss=0.1187, over 28906.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3436, pruned_loss=0.1023, over 5704596.49 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3643, pruned_loss=0.1188, over 5702791.04 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.338, pruned_loss=0.09901, over 5697593.78 frames. ], batch size: 106, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:06:48,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4743, 1.7071, 1.7224, 1.2607], device='cuda:0'), covar=tensor([0.1455, 0.2138, 0.1223, 0.1486], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0693, 0.0910, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 01:07:25,202 INFO [train.py:968] (0/2) Epoch 15, batch 37700, giga_loss[loss=0.3393, simple_loss=0.3966, pruned_loss=0.141, over 27627.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.348, pruned_loss=0.1045, over 5692236.45 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.364, pruned_loss=0.1185, over 5704316.32 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3434, pruned_loss=0.1017, over 5685042.05 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:07:28,916 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=676623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:07:59,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=676658.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:07:59,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.878e+02 1.260e+03 1.735e+03 2.273e+03 4.663e+03, threshold=3.469e+03, percent-clipped=6.0
+2023-03-08 01:08:10,901 INFO [train.py:968] (0/2) Epoch 15, batch 37750, giga_loss[loss=0.2654, simple_loss=0.3546, pruned_loss=0.08813, over 28959.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3515, pruned_loss=0.1052, over 5697062.67 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3638, pruned_loss=0.1184, over 5705320.93 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3478, pruned_loss=0.1028, over 5690173.36 frames. ], batch size: 174, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:08:14,237 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7285, 1.7607, 1.3619, 1.2806], device='cuda:0'), covar=tensor([0.0910, 0.0635, 0.0997, 0.1264], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0439, 0.0508, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 01:08:56,654 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-08 01:08:58,804 INFO [train.py:968] (0/2) Epoch 15, batch 37800, giga_loss[loss=0.2889, simple_loss=0.3624, pruned_loss=0.1077, over 28926.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3566, pruned_loss=0.1077, over 5681225.31 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3635, pruned_loss=0.1182, over 5700689.64 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3536, pruned_loss=0.1057, over 5679643.30 frames. ], batch size: 106, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:09:33,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.001e+02 1.167e+03 1.399e+03 1.786e+03 3.734e+03, threshold=2.799e+03, percent-clipped=1.0
+2023-03-08 01:09:40,614 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:09:42,171 INFO [train.py:968] (0/2) Epoch 15, batch 37850, libri_loss[loss=0.3273, simple_loss=0.3828, pruned_loss=0.1359, over 25987.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3574, pruned_loss=0.1077, over 5685033.72 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3634, pruned_loss=0.1181, over 5702256.18 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3549, pruned_loss=0.1059, over 5682153.41 frames. ], batch size: 136, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:09:42,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:10:05,654 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=676798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:10:07,517 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676801.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:10:08,081 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5519, 4.4199, 4.2018, 1.5985], device='cuda:0'), covar=tensor([0.0619, 0.0724, 0.0873, 0.2023], device='cuda:0'), in_proj_covar=tensor([0.1124, 0.1037, 0.0889, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 01:10:09,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676804.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:10:16,139 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 01:10:22,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5045, 2.3391, 1.7009, 0.5971], device='cuda:0'), covar=tensor([0.4589, 0.2449, 0.3028, 0.4865], device='cuda:0'), in_proj_covar=tensor([0.1629, 0.1557, 0.1537, 0.1332], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 01:10:23,730 INFO [train.py:968] (0/2) Epoch 15, batch 37900, giga_loss[loss=0.2707, simple_loss=0.351, pruned_loss=0.09519, over 28855.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3543, pruned_loss=0.1053, over 5694795.82 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3638, pruned_loss=0.1185, over 5706323.18 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3518, pruned_loss=0.1033, over 5688441.86 frames. ], batch size: 199, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:10:33,311 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=676833.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:10:40,280 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 01:10:51,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=676851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:10:56,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=676857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:10:57,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.769e+02 1.105e+03 1.402e+03 1.900e+03 5.111e+03, threshold=2.805e+03, percent-clipped=9.0
+2023-03-08 01:11:09,192 INFO [train.py:968] (0/2) Epoch 15, batch 37950, giga_loss[loss=0.3267, simple_loss=0.3785, pruned_loss=0.1374, over 23571.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3511, pruned_loss=0.1027, over 5690073.08 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3635, pruned_loss=0.1184, over 5702122.85 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3491, pruned_loss=0.1007, over 5687986.08 frames. ], batch size: 705, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:11:51,878 INFO [train.py:968] (0/2) Epoch 15, batch 38000, giga_loss[loss=0.2645, simple_loss=0.3403, pruned_loss=0.09434, over 28846.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3515, pruned_loss=0.1027, over 5696636.76 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3634, pruned_loss=0.1184, over 5702660.25 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3496, pruned_loss=0.1009, over 5694172.98 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 01:12:29,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.025e+02 1.165e+03 1.519e+03 2.164e+03 5.014e+03, threshold=3.038e+03, percent-clipped=9.0
+2023-03-08 01:12:37,261 INFO [train.py:968] (0/2) Epoch 15, batch 38050, giga_loss[loss=0.2765, simple_loss=0.3567, pruned_loss=0.09819, over 29114.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.354, pruned_loss=0.1041, over 5690708.99 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3634, pruned_loss=0.1184, over 5696222.94 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3524, pruned_loss=0.1024, over 5693699.97 frames. ], batch size: 128, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:12:59,564 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:01,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:03,372 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=677000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:05,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=677003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:20,725 INFO [train.py:968] (0/2) Epoch 15, batch 38100, giga_loss[loss=0.2812, simple_loss=0.3537, pruned_loss=0.1044, over 28996.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3555, pruned_loss=0.1053, over 5702615.22 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3631, pruned_loss=0.1182, over 5703089.82 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3543, pruned_loss=0.1036, over 5698933.51 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:13:26,347 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=677026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:31,397 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=677032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:45,505 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3517, 1.5197, 1.4043, 1.4561], device='cuda:0'), covar=tensor([0.0763, 0.0328, 0.0305, 0.0827], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 01:13:47,469 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8822, 2.1432, 1.7585, 1.9466], device='cuda:0'), covar=tensor([0.2439, 0.2399, 0.2680, 0.2205], device='cuda:0'), in_proj_covar=tensor([0.1390, 0.1020, 0.1230, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 01:13:52,807 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:57,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.926e+02 1.507e+03 1.949e+03 2.829e+03 6.993e+03, threshold=3.898e+03, percent-clipped=21.0
+2023-03-08 01:14:03,596 INFO [train.py:968] (0/2) Epoch 15, batch 38150, giga_loss[loss=0.2884, simple_loss=0.3591, pruned_loss=0.1089, over 28911.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3589, pruned_loss=0.1079, over 5699104.82 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3636, pruned_loss=0.1184, over 5704046.03 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3573, pruned_loss=0.1061, over 5695538.72 frames. ], batch size: 213, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:14:44,877 INFO [train.py:968] (0/2) Epoch 15, batch 38200, libri_loss[loss=0.2904, simple_loss=0.3479, pruned_loss=0.1164, over 27260.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3598, pruned_loss=0.1088, over 5696226.03 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3638, pruned_loss=0.1184, over 5706223.71 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3581, pruned_loss=0.1069, over 5691182.72 frames. ], batch size: 60, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:15:19,988 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:15:24,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.121e+02 1.302e+03 1.714e+03 2.456e+03 5.786e+03, threshold=3.427e+03, percent-clipped=3.0
+2023-03-08 01:15:30,983 INFO [train.py:968] (0/2) Epoch 15, batch 38250, giga_loss[loss=0.2764, simple_loss=0.3443, pruned_loss=0.1043, over 28689.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3608, pruned_loss=0.1099, over 5703377.50 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3644, pruned_loss=0.1187, over 5710155.75 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3588, pruned_loss=0.108, over 5695429.66 frames. ], batch size: 92, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:16:12,228 INFO [train.py:968] (0/2) Epoch 15, batch 38300, giga_loss[loss=0.2743, simple_loss=0.3521, pruned_loss=0.09828, over 28963.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3613, pruned_loss=0.1104, over 5698402.69 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3646, pruned_loss=0.1188, over 5711289.68 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3595, pruned_loss=0.1086, over 5690814.95 frames. ], batch size: 136, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:16:12,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9968, 1.2843, 3.3518, 2.9064], device='cuda:0'), covar=tensor([0.1719, 0.2561, 0.0467, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0695, 0.0606, 0.0890, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 01:16:47,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.656e+02 1.076e+03 1.311e+03 1.797e+03 2.960e+03, threshold=2.623e+03, percent-clipped=0.0
+2023-03-08 01:16:55,518 INFO [train.py:968] (0/2) Epoch 15, batch 38350, giga_loss[loss=0.3202, simple_loss=0.3835, pruned_loss=0.1284, over 27690.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3609, pruned_loss=0.109, over 5703097.55 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.365, pruned_loss=0.1191, over 5710694.86 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.359, pruned_loss=0.1072, over 5697214.24 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:17:38,274 INFO [train.py:968] (0/2) Epoch 15, batch 38400, giga_loss[loss=0.293, simple_loss=0.3763, pruned_loss=0.1049, over 28668.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.36, pruned_loss=0.1074, over 5710017.89 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3646, pruned_loss=0.1188, over 5714470.60 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3588, pruned_loss=0.1059, over 5701633.85 frames. ], batch size: 307, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:18:04,030 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.38 vs. limit=5.0
+2023-03-08 01:18:10,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.208e+02 1.081e+03 1.327e+03 1.913e+03 8.205e+03, threshold=2.654e+03, percent-clipped=12.0
+2023-03-08 01:18:17,288 INFO [train.py:968] (0/2) Epoch 15, batch 38450, giga_loss[loss=0.2633, simple_loss=0.3389, pruned_loss=0.09386, over 28741.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3598, pruned_loss=0.1075, over 5713969.26 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3646, pruned_loss=0.1187, over 5719101.97 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3586, pruned_loss=0.1061, over 5702862.70 frames. ], batch size: 119, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:18:21,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5620, 1.7765, 1.4617, 1.7535], device='cuda:0'), covar=tensor([0.2458, 0.2345, 0.2464, 0.2284], device='cuda:0'), in_proj_covar=tensor([0.1391, 0.1020, 0.1231, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 01:18:58,501 INFO [train.py:968] (0/2) Epoch 15, batch 38500, giga_loss[loss=0.2676, simple_loss=0.3487, pruned_loss=0.09323, over 28848.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3587, pruned_loss=0.1073, over 5707105.74 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3652, pruned_loss=0.119, over 5714572.81 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.357, pruned_loss=0.1054, over 5701171.19 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:19:10,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=677431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:19:17,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3894, 1.7105, 1.6619, 1.2238], device='cuda:0'), covar=tensor([0.1710, 0.2460, 0.1389, 0.1671], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0695, 0.0910, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 01:19:21,608 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-08 01:19:35,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.722e+02 1.185e+03 1.489e+03 1.761e+03 8.849e+03, threshold=2.979e+03, percent-clipped=9.0
+2023-03-08 01:19:41,591 INFO [train.py:968] (0/2) Epoch 15, batch 38550, giga_loss[loss=0.3034, simple_loss=0.3719, pruned_loss=0.1174, over 28251.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3562, pruned_loss=0.106, over 5691403.58 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3654, pruned_loss=0.1193, over 5701108.54 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3545, pruned_loss=0.1039, over 5699030.37 frames. ], batch size: 368, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:19:58,665 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:20:21,349 INFO [train.py:968] (0/2) Epoch 15, batch 38600, giga_loss[loss=0.2873, simple_loss=0.3616, pruned_loss=0.1065, over 28777.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3539, pruned_loss=0.1046, over 5687788.48 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.365, pruned_loss=0.1191, over 5693281.76 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3525, pruned_loss=0.1028, over 5701438.71 frames. ], batch size: 119, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:20:31,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=677531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:20:57,602 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 01:20:57,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.926e+02 9.851e+02 1.172e+03 1.548e+03 7.240e+03, threshold=2.344e+03, percent-clipped=5.0
+2023-03-08 01:20:58,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677561.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:21:04,372 INFO [train.py:968] (0/2) Epoch 15, batch 38650, giga_loss[loss=0.2767, simple_loss=0.3588, pruned_loss=0.09726, over 28901.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3548, pruned_loss=0.1051, over 5678258.43 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3653, pruned_loss=0.1193, over 5679220.90 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3533, pruned_loss=0.1033, over 5702210.39 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:21:08,187 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=677574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:21:10,287 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=677577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:21:15,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6823, 1.8970, 1.6251, 1.7554], device='cuda:0'), covar=tensor([0.2136, 0.1958, 0.2015, 0.1972], device='cuda:0'), in_proj_covar=tensor([0.1387, 0.1017, 0.1228, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-08 01:21:35,456 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=677606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:21:45,199 INFO [train.py:968] (0/2) Epoch 15, batch 38700, giga_loss[loss=0.2684, simple_loss=0.3487, pruned_loss=0.0941, over 28914.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3546, pruned_loss=0.1049, over 5670417.78 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3657, pruned_loss=0.1196, over 5663532.02 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3529, pruned_loss=0.1029, over 5702269.18 frames. ], batch size: 213, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:21:51,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6148, 1.8681, 1.8414, 1.4312], device='cuda:0'), covar=tensor([0.1804, 0.2445, 0.1460, 0.1680], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0696, 0.0913, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 01:22:18,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.905e+02 1.134e+03 1.407e+03 1.986e+03 7.159e+03, threshold=2.813e+03, percent-clipped=14.0
+2023-03-08 01:22:24,263 INFO [train.py:968] (0/2) Epoch 15, batch 38750, giga_loss[loss=0.2858, simple_loss=0.3573, pruned_loss=0.1072, over 28951.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3542, pruned_loss=0.1036, over 5680042.84 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3657, pruned_loss=0.1196, over 5663475.55 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3527, pruned_loss=0.1018, over 5705236.10 frames. ], batch size: 213, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:22:28,937 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=677674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:22:31,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=677677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:22:42,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4371, 1.6589, 1.6940, 1.2897], device='cuda:0'), covar=tensor([0.1851, 0.2507, 0.1511, 0.1772], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0695, 0.0912, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 01:22:52,811 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=677706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:23:04,634 INFO [train.py:968] (0/2) Epoch 15, batch 38800, giga_loss[loss=0.291, simple_loss=0.3607, pruned_loss=0.1107, over 28327.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3545, pruned_loss=0.1035, over 5696752.89 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3656, pruned_loss=0.1194, over 5673950.73 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.353, pruned_loss=0.1016, over 5709145.16 frames. ], batch size: 369, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 01:23:40,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.631e+02 9.895e+02 1.253e+03 1.750e+03 6.281e+03, threshold=2.506e+03, percent-clipped=5.0
+2023-03-08 01:23:44,464 INFO [train.py:968] (0/2) Epoch 15, batch 38850, giga_loss[loss=0.274, simple_loss=0.3508, pruned_loss=0.09862, over 28438.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3548, pruned_loss=0.1043, over 5687962.57 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3658, pruned_loss=0.1196, over 5668545.52 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3532, pruned_loss=0.1023, over 5702938.45 frames. ], batch size: 71, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:23:50,883 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:24:08,409 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:24:24,938 INFO [train.py:968] (0/2) Epoch 15, batch 38900, giga_loss[loss=0.2673, simple_loss=0.3423, pruned_loss=0.09616, over 27949.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3526, pruned_loss=0.1034, over 5697297.40 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3657, pruned_loss=0.1194, over 5674397.79 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3511, pruned_loss=0.1016, over 5704661.07 frames. ], batch size: 412, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:25:00,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.653e+02 1.162e+03 1.542e+03 2.150e+03 8.943e+03, threshold=3.083e+03, percent-clipped=19.0
+2023-03-08 01:25:02,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=677864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:25:03,470 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677865.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:25:04,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2715, 4.0987, 3.8774, 1.9294], device='cuda:0'), covar=tensor([0.0566, 0.0720, 0.0708, 0.2019], device='cuda:0'), in_proj_covar=tensor([0.1121, 0.1031, 0.0885, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-08 01:25:06,039 INFO [train.py:968] (0/2) Epoch 15, batch 38950, giga_loss[loss=0.2555, simple_loss=0.3268, pruned_loss=0.09213, over 28786.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3497, pruned_loss=0.1025, over 5696426.97 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3656, pruned_loss=0.1193, over 5678872.00 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3483, pruned_loss=0.1009, over 5698628.93 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:25:14,358 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:25:47,458 INFO [train.py:968] (0/2) Epoch 15, batch 39000, giga_loss[loss=0.2813, simple_loss=0.3427, pruned_loss=0.1099, over 28797.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3474, pruned_loss=0.1009, over 5701230.57 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3657, pruned_loss=0.1193, over 5677234.40 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3461, pruned_loss=0.09943, over 5704515.73 frames. ], batch size: 99, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:25:47,463 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 01:25:52,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2449, 1.5276, 1.4324, 1.2156], device='cuda:0'), covar=tensor([0.2447, 0.2139, 0.1446, 0.1812], device='cuda:0'), in_proj_covar=tensor([0.1842, 0.1758, 0.1690, 0.1836], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 01:25:56,540 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2145, simple_loss=0.3218, pruned_loss=0.05359, over 944034.00 frames. 
+2023-03-08 01:25:56,541 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 01:26:02,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8468, 3.6597, 3.4570, 1.9725], device='cuda:0'), covar=tensor([0.0538, 0.0689, 0.0671, 0.1865], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.1029, 0.0883, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-08 01:26:12,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=677936.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:26:16,143 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-08 01:26:35,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.749e+02 1.086e+03 1.332e+03 1.902e+03 6.344e+03, threshold=2.664e+03, percent-clipped=6.0
+2023-03-08 01:26:40,805 INFO [train.py:968] (0/2) Epoch 15, batch 39050, giga_loss[loss=0.2519, simple_loss=0.3342, pruned_loss=0.08476, over 28827.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.346, pruned_loss=0.1003, over 5707164.08 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3658, pruned_loss=0.1193, over 5680808.52 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3446, pruned_loss=0.09892, over 5706909.71 frames. ], batch size: 243, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:26:46,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4634, 1.8038, 1.3843, 1.6683], device='cuda:0'), covar=tensor([0.2598, 0.2453, 0.2942, 0.2176], device='cuda:0'), in_proj_covar=tensor([0.1390, 0.1019, 0.1229, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:0')
+2023-03-08 01:27:06,345 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-678000.pt
+2023-03-08 01:27:11,696 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:27:13,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:27:17,836 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:27:18,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4113, 1.6336, 1.6425, 1.2036], device='cuda:0'), covar=tensor([0.1522, 0.2535, 0.1339, 0.1613], device='cuda:0'), in_proj_covar=tensor([0.0861, 0.0694, 0.0909, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 01:27:20,813 INFO [train.py:968] (0/2) Epoch 15, batch 39100, giga_loss[loss=0.2884, simple_loss=0.3524, pruned_loss=0.1122, over 28928.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3445, pruned_loss=0.1001, over 5705750.82 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.366, pruned_loss=0.1195, over 5674048.88 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3427, pruned_loss=0.09837, over 5712223.91 frames. ], batch size: 285, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 01:27:40,181 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:27:58,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.125e+02 1.123e+03 1.411e+03 2.013e+03 1.875e+04, threshold=2.823e+03, percent-clipped=15.0
+2023-03-08 01:28:03,569 INFO [train.py:968] (0/2) Epoch 15, batch 39150, libri_loss[loss=0.3133, simple_loss=0.3667, pruned_loss=0.1299, over 29648.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3424, pruned_loss=0.09962, over 5702045.41 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3662, pruned_loss=0.1197, over 5675121.80 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3403, pruned_loss=0.09757, over 5706831.62 frames. ], batch size: 73, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 01:28:11,312 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678079.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:28:14,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678082.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:28:21,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2725, 1.0913, 3.6487, 3.1175], device='cuda:0'), covar=tensor([0.1525, 0.2755, 0.0396, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0693, 0.0607, 0.0887, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 01:28:35,672 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678111.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:28:42,231 INFO [train.py:968] (0/2) Epoch 15, batch 39200, giga_loss[loss=0.2461, simple_loss=0.3167, pruned_loss=0.08769, over 28864.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3392, pruned_loss=0.09803, over 5706959.81 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.366, pruned_loss=0.1195, over 5681480.51 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.337, pruned_loss=0.096, over 5705898.90 frames. ], batch size: 99, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:29:13,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:29:21,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.087e+02 1.048e+03 1.301e+03 1.663e+03 6.565e+03, threshold=2.601e+03, percent-clipped=10.0
+2023-03-08 01:29:26,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2981, 3.0681, 1.4331, 1.4726], device='cuda:0'), covar=tensor([0.0904, 0.0319, 0.0893, 0.1228], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0522, 0.0356, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-08 01:29:29,164 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-08 01:29:29,399 INFO [train.py:968] (0/2) Epoch 15, batch 39250, giga_loss[loss=0.28, simple_loss=0.3549, pruned_loss=0.1025, over 28378.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3374, pruned_loss=0.09718, over 5699478.99 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3663, pruned_loss=0.1197, over 5675407.26 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3351, pruned_loss=0.09511, over 5703781.21 frames. ], batch size: 368, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:29:33,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:29:58,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2273, 1.2288, 1.1068, 1.5732], device='cuda:0'), covar=tensor([0.0760, 0.0330, 0.0355, 0.0842], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 01:30:08,370 INFO [train.py:968] (0/2) Epoch 15, batch 39300, giga_loss[loss=0.3447, simple_loss=0.4076, pruned_loss=0.1409, over 28685.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3395, pruned_loss=0.09828, over 5713090.21 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3665, pruned_loss=0.12, over 5685138.55 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3361, pruned_loss=0.09535, over 5709005.15 frames. ], batch size: 242, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:30:27,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678240.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:30:27,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6400, 1.8578, 1.5706, 1.4364], device='cuda:0'), covar=tensor([0.1959, 0.2241, 0.2265, 0.2509], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0733, 0.0685, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 01:30:28,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:30:39,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:30:47,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.479e+02 1.042e+03 1.341e+03 1.968e+03 7.668e+03, threshold=2.682e+03, percent-clipped=13.0
+2023-03-08 01:30:52,191 INFO [train.py:968] (0/2) Epoch 15, batch 39350, giga_loss[loss=0.2747, simple_loss=0.3539, pruned_loss=0.09771, over 28895.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3417, pruned_loss=0.09872, over 5705320.21 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3665, pruned_loss=0.1199, over 5679094.60 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3384, pruned_loss=0.09592, over 5707381.49 frames. ], batch size: 213, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:31:18,866 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:31:18,896 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6163, 1.7729, 1.4905, 1.9372], device='cuda:0'), covar=tensor([0.2201, 0.2251, 0.2365, 0.2122], device='cuda:0'), in_proj_covar=tensor([0.1391, 0.1019, 0.1228, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 01:31:21,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:31:26,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3130, 5.1285, 4.8385, 2.5151], device='cuda:0'), covar=tensor([0.0366, 0.0514, 0.0554, 0.1663], device='cuda:0'), in_proj_covar=tensor([0.1122, 0.1032, 0.0886, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 01:31:35,898 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:31:38,622 INFO [train.py:968] (0/2) Epoch 15, batch 39400, giga_loss[loss=0.3274, simple_loss=0.3972, pruned_loss=0.1288, over 27663.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3447, pruned_loss=0.09972, over 5706634.45 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3665, pruned_loss=0.12, over 5684125.28 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3414, pruned_loss=0.09686, over 5704315.71 frames. ], batch size: 472, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:31:38,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:31:46,951 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:31:52,885 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:32:04,145 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:32:18,897 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.406e+02 1.049e+03 1.248e+03 1.666e+03 5.292e+03, threshold=2.496e+03, percent-clipped=8.0
+2023-03-08 01:32:23,917 INFO [train.py:968] (0/2) Epoch 15, batch 39450, giga_loss[loss=0.2931, simple_loss=0.3695, pruned_loss=0.1084, over 28554.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3479, pruned_loss=0.1012, over 5688103.66 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3669, pruned_loss=0.1204, over 5671821.15 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3442, pruned_loss=0.09791, over 5699148.63 frames. ], batch size: 336, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:32:36,660 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678383.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:32:39,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678386.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:32:42,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:32:47,324 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:32:51,163 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:33:03,215 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678415.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:33:06,670 INFO [train.py:968] (0/2) Epoch 15, batch 39500, giga_loss[loss=0.2651, simple_loss=0.3342, pruned_loss=0.09803, over 23852.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3474, pruned_loss=0.1004, over 5686866.76 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3668, pruned_loss=0.1201, over 5678669.48 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.344, pruned_loss=0.09741, over 5689865.67 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:33:16,036 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:33:32,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-08 01:33:34,555 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-08 01:33:45,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.673e+02 1.087e+03 1.419e+03 2.062e+03 5.361e+03, threshold=2.838e+03, percent-clipped=11.0
+2023-03-08 01:33:49,594 INFO [train.py:968] (0/2) Epoch 15, batch 39550, giga_loss[loss=0.2771, simple_loss=0.3568, pruned_loss=0.0987, over 28960.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.346, pruned_loss=0.09864, over 5680910.32 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3667, pruned_loss=0.1202, over 5663323.94 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3431, pruned_loss=0.09593, over 5697659.20 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:34:05,789 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3672, 1.6012, 1.2166, 1.5654], device='cuda:0'), covar=tensor([0.0709, 0.0291, 0.0329, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 01:34:13,839 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678498.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:34:24,163 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:34:33,301 INFO [train.py:968] (0/2) Epoch 15, batch 39600, giga_loss[loss=0.2674, simple_loss=0.3468, pruned_loss=0.09399, over 28647.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3475, pruned_loss=0.1003, over 5679192.13 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3674, pruned_loss=0.1206, over 5667516.32 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3442, pruned_loss=0.0973, over 5689326.29 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 01:34:42,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6148, 2.3078, 2.2543, 2.2006], device='cuda:0'), covar=tensor([0.1533, 0.2373, 0.1993, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0733, 0.0684, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 01:34:45,550 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:34:47,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:35:11,271 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.436e+02 1.331e+03 1.610e+03 2.457e+03 8.084e+03, threshold=3.220e+03, percent-clipped=21.0
+2023-03-08 01:35:12,130 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:35:15,600 INFO [train.py:968] (0/2) Epoch 15, batch 39650, giga_loss[loss=0.2671, simple_loss=0.3437, pruned_loss=0.09519, over 28616.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3475, pruned_loss=0.1005, over 5675247.63 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3674, pruned_loss=0.1206, over 5664487.66 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3442, pruned_loss=0.09753, over 5686471.20 frames. ], batch size: 78, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:35:16,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2647, 1.7837, 1.3138, 0.4553], device='cuda:0'), covar=tensor([0.3736, 0.2444, 0.4034, 0.5079], device='cuda:0'), in_proj_covar=tensor([0.1624, 0.1542, 0.1528, 0.1326], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 01:35:33,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5242, 4.3883, 1.6820, 1.7437], device='cuda:0'), covar=tensor([0.0940, 0.0273, 0.0894, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0520, 0.0355, 0.0402], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:0')
+2023-03-08 01:35:46,229 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:35:57,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:35:59,972 INFO [train.py:968] (0/2) Epoch 15, batch 39700, giga_loss[loss=0.3261, simple_loss=0.3941, pruned_loss=0.1291, over 29073.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3513, pruned_loss=0.1027, over 5685330.72 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3674, pruned_loss=0.1204, over 5664564.54 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3485, pruned_loss=0.1002, over 5694448.55 frames. ], batch size: 128, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:36:38,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.933e+02 1.220e+03 1.521e+03 1.836e+03 4.279e+03, threshold=3.041e+03, percent-clipped=3.0
+2023-03-08 01:36:42,883 INFO [train.py:968] (0/2) Epoch 15, batch 39750, giga_loss[loss=0.2828, simple_loss=0.3533, pruned_loss=0.1062, over 28594.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3535, pruned_loss=0.1034, over 5687580.79 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3679, pruned_loss=0.1207, over 5656426.49 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3506, pruned_loss=0.101, over 5701759.42 frames. ], batch size: 85, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:37:14,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:37:20,660 INFO [train.py:968] (0/2) Epoch 15, batch 39800, giga_loss[loss=0.2592, simple_loss=0.3404, pruned_loss=0.08898, over 29014.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3551, pruned_loss=0.1039, over 5704332.56 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3687, pruned_loss=0.1209, over 5663569.97 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3513, pruned_loss=0.1009, over 5711032.23 frames. ], batch size: 136, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:37:39,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3375, 1.4834, 1.4241, 1.2633], device='cuda:0'), covar=tensor([0.2556, 0.2079, 0.1881, 0.2348], device='cuda:0'), in_proj_covar=tensor([0.1849, 0.1764, 0.1698, 0.1841], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 01:37:55,625 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:37:57,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:37:58,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.530e+02 1.365e+03 1.642e+03 2.462e+03 9.401e+03, threshold=3.284e+03, percent-clipped=18.0
+2023-03-08 01:38:05,490 INFO [train.py:968] (0/2) Epoch 15, batch 39850, giga_loss[loss=0.2987, simple_loss=0.3759, pruned_loss=0.1108, over 28842.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3561, pruned_loss=0.1048, over 5707482.65 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3691, pruned_loss=0.1213, over 5667349.21 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3525, pruned_loss=0.1017, over 5710520.10 frames. ], batch size: 145, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:38:23,237 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:38:48,544 INFO [train.py:968] (0/2) Epoch 15, batch 39900, giga_loss[loss=0.264, simple_loss=0.3459, pruned_loss=0.09098, over 28862.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3559, pruned_loss=0.1048, over 5709350.54 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3693, pruned_loss=0.1214, over 5670867.56 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3526, pruned_loss=0.1019, over 5709289.03 frames. ], batch size: 186, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:39:12,173 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 01:39:14,117 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:39:16,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:39:23,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.400e+02 1.147e+03 1.513e+03 1.938e+03 5.960e+03, threshold=3.027e+03, percent-clipped=4.0
+2023-03-08 01:39:28,731 INFO [train.py:968] (0/2) Epoch 15, batch 39950, giga_loss[loss=0.2634, simple_loss=0.3383, pruned_loss=0.09429, over 28852.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3551, pruned_loss=0.1043, over 5711463.98 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3696, pruned_loss=0.1215, over 5675608.17 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3519, pruned_loss=0.1015, over 5708258.17 frames. ], batch size: 186, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:39:32,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678873.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:39:40,871 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:39:41,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:40:07,946 INFO [train.py:968] (0/2) Epoch 15, batch 40000, giga_loss[loss=0.3598, simple_loss=0.4014, pruned_loss=0.1591, over 26801.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.353, pruned_loss=0.1034, over 5695475.36 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3697, pruned_loss=0.1218, over 5656617.18 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3495, pruned_loss=0.1, over 5713065.90 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:40:44,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.328e+02 1.265e+03 1.627e+03 2.274e+03 9.107e+03, threshold=3.253e+03, percent-clipped=15.0
+2023-03-08 01:40:48,021 INFO [train.py:968] (0/2) Epoch 15, batch 40050, giga_loss[loss=0.2411, simple_loss=0.3167, pruned_loss=0.08273, over 28910.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3498, pruned_loss=0.1019, over 5703915.63 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3698, pruned_loss=0.1218, over 5664106.18 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3463, pruned_loss=0.09847, over 5712656.36 frames. ], batch size: 186, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:40:56,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:41:27,996 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:41:30,351 INFO [train.py:968] (0/2) Epoch 15, batch 40100, libri_loss[loss=0.2806, simple_loss=0.3591, pruned_loss=0.1011, over 29533.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3472, pruned_loss=0.09998, over 5709103.09 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3697, pruned_loss=0.1215, over 5667340.09 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3439, pruned_loss=0.09694, over 5714217.47 frames. ], batch size: 81, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:41:31,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679019.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:41:39,482 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:41:41,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:41:55,355 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=679048.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:41:59,999 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679054.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:42:04,130 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=679060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:42:07,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.890e+02 1.042e+03 1.367e+03 1.846e+03 5.229e+03, threshold=2.734e+03, percent-clipped=5.0
+2023-03-08 01:42:10,791 INFO [train.py:968] (0/2) Epoch 15, batch 40150, giga_loss[loss=0.2871, simple_loss=0.3759, pruned_loss=0.09915, over 28651.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3486, pruned_loss=0.09988, over 5698913.64 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3698, pruned_loss=0.1216, over 5659863.48 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3453, pruned_loss=0.09689, over 5711385.42 frames. ], batch size: 242, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:42:53,058 INFO [train.py:968] (0/2) Epoch 15, batch 40200, giga_loss[loss=0.2737, simple_loss=0.3457, pruned_loss=0.1008, over 28839.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3498, pruned_loss=0.1003, over 5701395.99 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3699, pruned_loss=0.1218, over 5667574.32 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3464, pruned_loss=0.09682, over 5705418.31 frames. ], batch size: 199, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:42:55,507 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:42:57,968 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:43:22,903 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=679154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:43:30,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3627, 1.7583, 1.3511, 1.3016], device='cuda:0'), covar=tensor([0.2554, 0.2436, 0.2958, 0.2215], device='cuda:0'), in_proj_covar=tensor([0.1396, 0.1021, 0.1234, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 01:43:31,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.162e+02 1.172e+03 1.544e+03 2.018e+03 7.996e+03, threshold=3.087e+03, percent-clipped=12.0
+2023-03-08 01:43:35,995 INFO [train.py:968] (0/2) Epoch 15, batch 40250, giga_loss[loss=0.2275, simple_loss=0.3133, pruned_loss=0.07084, over 28907.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.35, pruned_loss=0.1009, over 5696978.29 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.37, pruned_loss=0.1219, over 5658342.00 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.347, pruned_loss=0.09784, over 5707637.59 frames. ], batch size: 174, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:43:48,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 01:43:58,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2996, 1.5721, 1.3529, 1.5091], device='cuda:0'), covar=tensor([0.0754, 0.0323, 0.0335, 0.0850], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 01:44:16,398 INFO [train.py:968] (0/2) Epoch 15, batch 40300, giga_loss[loss=0.257, simple_loss=0.3272, pruned_loss=0.09339, over 29042.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3477, pruned_loss=0.1006, over 5705899.53 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3699, pruned_loss=0.1218, over 5662162.23 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3451, pruned_loss=0.09806, over 5711476.25 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:44:58,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.194e+02 1.080e+03 1.335e+03 1.649e+03 4.068e+03, threshold=2.670e+03, percent-clipped=1.0
+2023-03-08 01:45:02,536 INFO [train.py:968] (0/2) Epoch 15, batch 40350, giga_loss[loss=0.2838, simple_loss=0.3493, pruned_loss=0.1092, over 27587.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3456, pruned_loss=0.1008, over 5705259.50 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3701, pruned_loss=0.1219, over 5655539.15 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3432, pruned_loss=0.09847, over 5715080.80 frames. ], batch size: 472, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:45:07,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3413, 4.1382, 3.9119, 1.8705], device='cuda:0'), covar=tensor([0.0573, 0.0763, 0.0758, 0.2273], device='cuda:0'), in_proj_covar=tensor([0.1129, 0.1038, 0.0894, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 01:45:11,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3430, 1.4919, 1.2842, 1.5351], device='cuda:0'), covar=tensor([0.0741, 0.0335, 0.0341, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 01:45:43,777 INFO [train.py:968] (0/2) Epoch 15, batch 40400, libri_loss[loss=0.2992, simple_loss=0.3681, pruned_loss=0.1152, over 29534.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3442, pruned_loss=0.1012, over 5705348.19 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3697, pruned_loss=0.1215, over 5661891.55 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3416, pruned_loss=0.09885, over 5709241.49 frames. ], batch size: 89, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 01:46:23,340 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.046e+02 1.109e+03 1.352e+03 2.197e+03 1.127e+04, threshold=2.704e+03, percent-clipped=13.0
+2023-03-08 01:46:26,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8672, 1.2246, 2.8623, 2.6780], device='cuda:0'), covar=tensor([0.1567, 0.2417, 0.0549, 0.1067], device='cuda:0'), in_proj_covar=tensor([0.0700, 0.0611, 0.0899, 0.0817], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 01:46:27,150 INFO [train.py:968] (0/2) Epoch 15, batch 40450, libri_loss[loss=0.325, simple_loss=0.3916, pruned_loss=0.1292, over 29500.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3448, pruned_loss=0.1022, over 5698775.52 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3699, pruned_loss=0.1216, over 5662940.26 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.342, pruned_loss=0.09986, over 5701596.65 frames. ], batch size: 85, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:47:02,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7580, 1.9169, 1.7997, 1.8165], device='cuda:0'), covar=tensor([0.1831, 0.2265, 0.2301, 0.2069], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0735, 0.0686, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 01:47:08,382 INFO [train.py:968] (0/2) Epoch 15, batch 40500, giga_loss[loss=0.2464, simple_loss=0.3161, pruned_loss=0.08839, over 28636.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3408, pruned_loss=0.09987, over 5705464.80 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3698, pruned_loss=0.1214, over 5666346.75 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3383, pruned_loss=0.09792, over 5705120.62 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:47:12,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6753, 1.7524, 1.2675, 1.2960], device='cuda:0'), covar=tensor([0.0835, 0.0625, 0.1008, 0.1190], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0440, 0.0506, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 01:47:15,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=679429.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:47:23,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:47:46,067 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679465.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:47:46,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.442e+02 1.145e+03 1.490e+03 2.286e+03 7.408e+03, threshold=2.981e+03, percent-clipped=10.0
+2023-03-08 01:47:50,952 INFO [train.py:968] (0/2) Epoch 15, batch 40550, giga_loss[loss=0.2456, simple_loss=0.3086, pruned_loss=0.09131, over 28719.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3366, pruned_loss=0.09773, over 5705671.75 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3699, pruned_loss=0.1213, over 5667203.41 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3339, pruned_loss=0.09575, over 5705392.73 frames. ], batch size: 99, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:48:33,706 INFO [train.py:968] (0/2) Epoch 15, batch 40600, giga_loss[loss=0.2552, simple_loss=0.3315, pruned_loss=0.08942, over 28548.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3329, pruned_loss=0.09558, over 5700556.94 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3701, pruned_loss=0.1214, over 5660517.95 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.33, pruned_loss=0.09351, over 5707122.48 frames. ], batch size: 336, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:48:46,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679534.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:49:12,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.282e+02 1.142e+03 1.463e+03 1.956e+03 4.649e+03, threshold=2.926e+03, percent-clipped=9.0
+2023-03-08 01:49:15,886 INFO [train.py:968] (0/2) Epoch 15, batch 40650, libri_loss[loss=0.3812, simple_loss=0.4207, pruned_loss=0.1708, over 28708.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3341, pruned_loss=0.09581, over 5700100.71 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3703, pruned_loss=0.1217, over 5656322.39 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3306, pruned_loss=0.09321, over 5710406.74 frames. ], batch size: 106, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:49:19,339 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679572.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:49:21,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679575.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:49:37,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:49:47,102 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=679604.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:50:00,182 INFO [train.py:968] (0/2) Epoch 15, batch 40700, giga_loss[loss=0.2966, simple_loss=0.3632, pruned_loss=0.115, over 28900.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3387, pruned_loss=0.09782, over 5700459.00 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3703, pruned_loss=0.1217, over 5657589.77 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3358, pruned_loss=0.09568, over 5707668.39 frames. ], batch size: 112, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:50:38,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.310e+02 1.129e+03 1.448e+03 1.854e+03 4.893e+03, threshold=2.896e+03, percent-clipped=9.0
+2023-03-08 01:50:41,470 INFO [train.py:968] (0/2) Epoch 15, batch 40750, giga_loss[loss=0.295, simple_loss=0.3574, pruned_loss=0.1163, over 23818.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3414, pruned_loss=0.0988, over 5704045.60 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3704, pruned_loss=0.1218, over 5662213.55 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3384, pruned_loss=0.09656, over 5706426.26 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:51:15,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:51:22,739 INFO [train.py:968] (0/2) Epoch 15, batch 40800, giga_loss[loss=0.2539, simple_loss=0.3282, pruned_loss=0.08973, over 28441.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3447, pruned_loss=0.1001, over 5700823.89 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.37, pruned_loss=0.1213, over 5666443.40 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3419, pruned_loss=0.09802, over 5699979.51 frames. ], batch size: 78, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 01:51:58,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4751, 1.6675, 1.3524, 1.6646], device='cuda:0'), covar=tensor([0.2699, 0.2589, 0.2964, 0.2347], device='cuda:0'), in_proj_covar=tensor([0.1397, 0.1020, 0.1233, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 01:52:06,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.252e+02 1.240e+03 1.723e+03 2.231e+03 4.993e+03, threshold=3.445e+03, percent-clipped=16.0
+2023-03-08 01:52:08,138 INFO [train.py:968] (0/2) Epoch 15, batch 40850, giga_loss[loss=0.2699, simple_loss=0.3496, pruned_loss=0.09505, over 28767.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3485, pruned_loss=0.102, over 5708782.58 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3698, pruned_loss=0.1213, over 5668540.71 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3458, pruned_loss=0.09994, over 5706966.76 frames. ], batch size: 284, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:52:45,077 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:52:45,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=679812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:52:52,526 INFO [train.py:968] (0/2) Epoch 15, batch 40900, giga_loss[loss=0.345, simple_loss=0.4036, pruned_loss=0.1432, over 28654.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3525, pruned_loss=0.1053, over 5699200.34 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3696, pruned_loss=0.1211, over 5662564.34 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3502, pruned_loss=0.1034, over 5704290.02 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:53:19,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=679840.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:53:44,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.978e+02 1.404e+03 2.085e+03 2.953e+03 1.089e+04, threshold=4.169e+03, percent-clipped=15.0
+2023-03-08 01:53:46,680 INFO [train.py:968] (0/2) Epoch 15, batch 40950, libri_loss[loss=0.3354, simple_loss=0.3967, pruned_loss=0.137, over 25900.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3593, pruned_loss=0.1112, over 5698358.84 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1212, over 5665040.20 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3571, pruned_loss=0.1093, over 5701204.82 frames. ], batch size: 136, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:54:09,563 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:54:15,929 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3391, 1.2346, 1.1009, 1.5369], device='cuda:0'), covar=tensor([0.0732, 0.0355, 0.0351, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 01:54:28,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=679909.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:54:35,384 INFO [train.py:968] (0/2) Epoch 15, batch 41000, giga_loss[loss=0.3159, simple_loss=0.3893, pruned_loss=0.1212, over 28975.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3665, pruned_loss=0.1162, over 5686907.61 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3697, pruned_loss=0.1212, over 5658176.23 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3645, pruned_loss=0.1146, over 5696288.16 frames. ], batch size: 136, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:55:00,292 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1932, 4.0199, 3.8200, 2.0325], device='cuda:0'), covar=tensor([0.0523, 0.0689, 0.0674, 0.2316], device='cuda:0'), in_proj_covar=tensor([0.1133, 0.1044, 0.0896, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 01:55:10,925 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:55:14,777 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-08 01:55:15,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:55:23,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.572e+03 1.978e+03 3.053e+03 7.936e+03, threshold=3.956e+03, percent-clipped=4.0
+2023-03-08 01:55:25,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=679968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:55:25,669 INFO [train.py:968] (0/2) Epoch 15, batch 41050, giga_loss[loss=0.3148, simple_loss=0.3722, pruned_loss=0.1287, over 28530.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3723, pruned_loss=0.121, over 5685902.41 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3693, pruned_loss=0.121, over 5662558.23 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.371, pruned_loss=0.1198, over 5689962.59 frames. ], batch size: 85, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:55:35,423 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679983.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:55:37,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679986.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:55:38,226 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=679987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:55:47,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2574, 1.4073, 3.2884, 3.0150], device='cuda:0'), covar=tensor([0.1349, 0.2334, 0.0466, 0.1080], device='cuda:0'), in_proj_covar=tensor([0.0701, 0.0612, 0.0901, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 01:55:51,608 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-680000.pt
+2023-03-08 01:56:05,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3280, 3.1379, 2.9804, 1.6001], device='cuda:0'), covar=tensor([0.0843, 0.1016, 0.0849, 0.1984], device='cuda:0'), in_proj_covar=tensor([0.1133, 0.1047, 0.0899, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 01:56:05,273 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680015.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:56:10,164 INFO [train.py:968] (0/2) Epoch 15, batch 41100, giga_loss[loss=0.3974, simple_loss=0.436, pruned_loss=0.1795, over 28326.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3774, pruned_loss=0.1252, over 5694253.40 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3687, pruned_loss=0.1204, over 5672257.28 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3771, pruned_loss=0.1248, over 5690165.46 frames. ], batch size: 368, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:56:42,792 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680052.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:56:45,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680055.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:56:50,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.47 vs. limit=5.0
+2023-03-08 01:56:58,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.654e+03 2.080e+03 2.502e+03 6.061e+03, threshold=4.159e+03, percent-clipped=13.0
+2023-03-08 01:57:00,868 INFO [train.py:968] (0/2) Epoch 15, batch 41150, giga_loss[loss=0.3298, simple_loss=0.3934, pruned_loss=0.1331, over 28945.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3833, pruned_loss=0.1305, over 5682830.09 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1206, over 5674158.79 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.383, pruned_loss=0.1301, over 5678123.82 frames. ], batch size: 213, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:57:17,122 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680084.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:57:17,891 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:57:23,189 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.18 vs. limit=5.0
+2023-03-08 01:57:43,910 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:57:48,598 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:57:52,849 INFO [train.py:968] (0/2) Epoch 15, batch 41200, giga_loss[loss=0.4466, simple_loss=0.4765, pruned_loss=0.2084, over 28660.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3865, pruned_loss=0.1341, over 5668783.69 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.369, pruned_loss=0.1205, over 5675950.42 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3866, pruned_loss=0.1342, over 5663343.22 frames. ], batch size: 307, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:58:17,340 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:58:48,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.151e+03 1.819e+03 2.510e+03 3.810e+03 1.313e+04, threshold=5.020e+03, percent-clipped=22.0
+2023-03-08 01:58:48,309 INFO [train.py:968] (0/2) Epoch 15, batch 41250, giga_loss[loss=0.3947, simple_loss=0.4278, pruned_loss=0.1807, over 28011.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3885, pruned_loss=0.1366, over 5669318.21 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3691, pruned_loss=0.1205, over 5677387.88 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3889, pruned_loss=0.1371, over 5663349.50 frames. ], batch size: 412, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 01:58:48,518 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=680169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:59:08,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:59:42,231 INFO [train.py:968] (0/2) Epoch 15, batch 41300, giga_loss[loss=0.2891, simple_loss=0.3608, pruned_loss=0.1087, over 28879.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3902, pruned_loss=0.1394, over 5650856.45 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1202, over 5684339.63 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3919, pruned_loss=0.1407, over 5638459.57 frames. ], batch size: 119, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 01:59:51,575 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:59:54,165 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:00:21,510 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:00:31,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:00:34,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.612e+03 2.023e+03 2.691e+03 5.444e+03, threshold=4.045e+03, percent-clipped=2.0
+2023-03-08 02:00:34,432 INFO [train.py:968] (0/2) Epoch 15, batch 41350, giga_loss[loss=0.351, simple_loss=0.408, pruned_loss=0.147, over 28823.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3942, pruned_loss=0.1431, over 5641836.85 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3684, pruned_loss=0.12, over 5688761.85 frames. ], giga_tot_loss[loss=0.3429, simple_loss=0.3962, pruned_loss=0.1447, over 5627160.60 frames. ], batch size: 199, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:01:27,530 INFO [train.py:968] (0/2) Epoch 15, batch 41400, giga_loss[loss=0.2803, simple_loss=0.3514, pruned_loss=0.1046, over 28895.00 frames. ], tot_loss[loss=0.343, simple_loss=0.396, pruned_loss=0.145, over 5636375.03 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3683, pruned_loss=0.1199, over 5691339.19 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.3984, pruned_loss=0.1471, over 5621070.36 frames. ], batch size: 227, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:01:39,693 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:01:42,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:02:10,338 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:02:18,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.939e+02 1.678e+03 2.409e+03 3.561e+03 1.140e+04, threshold=4.817e+03, percent-clipped=18.0
+2023-03-08 02:02:18,329 INFO [train.py:968] (0/2) Epoch 15, batch 41450, giga_loss[loss=0.2767, simple_loss=0.3407, pruned_loss=0.1063, over 28496.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3948, pruned_loss=0.1453, over 5642156.34 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3679, pruned_loss=0.1197, over 5696083.36 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.3976, pruned_loss=0.1477, over 5624743.06 frames. ], batch size: 65, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:02:21,347 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=680372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:02:26,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1572, 2.9354, 1.3157, 1.3643], device='cuda:0'), covar=tensor([0.1090, 0.0472, 0.0894, 0.1423], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0529, 0.0358, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 02:02:52,963 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=680403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:02:58,542 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:03:00,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:03:02,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4432, 1.6697, 1.3160, 1.7634], device='cuda:0'), covar=tensor([0.2334, 0.2411, 0.2656, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.1389, 0.1017, 0.1232, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 02:03:08,234 INFO [train.py:968] (0/2) Epoch 15, batch 41500, giga_loss[loss=0.3532, simple_loss=0.4009, pruned_loss=0.1528, over 28552.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3918, pruned_loss=0.1426, over 5658644.52 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3672, pruned_loss=0.1194, over 5701587.01 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.3954, pruned_loss=0.1454, over 5638351.27 frames. ], batch size: 336, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:03:27,465 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:03:39,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2339, 1.2310, 3.8733, 3.1646], device='cuda:0'), covar=tensor([0.1690, 0.2733, 0.0437, 0.0896], device='cuda:0'), in_proj_covar=tensor([0.0702, 0.0614, 0.0901, 0.0822], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 02:03:56,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.620e+03 2.145e+03 3.099e+03 5.497e+03, threshold=4.290e+03, percent-clipped=3.0
+2023-03-08 02:03:56,569 INFO [train.py:968] (0/2) Epoch 15, batch 41550, giga_loss[loss=0.3282, simple_loss=0.389, pruned_loss=0.1337, over 28293.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3916, pruned_loss=0.1416, over 5654194.27 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3667, pruned_loss=0.1194, over 5698367.01 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3961, pruned_loss=0.1449, over 5638223.32 frames. ], batch size: 368, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:04:03,586 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-08 02:04:33,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3607, 1.5428, 1.3314, 1.6286], device='cuda:0'), covar=tensor([0.0795, 0.0328, 0.0326, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 02:04:48,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2504, 2.8561, 1.4234, 1.3911], device='cuda:0'), covar=tensor([0.1013, 0.0317, 0.0902, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0529, 0.0359, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 02:04:48,912 INFO [train.py:968] (0/2) Epoch 15, batch 41600, giga_loss[loss=0.3314, simple_loss=0.3915, pruned_loss=0.1356, over 28786.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3922, pruned_loss=0.1408, over 5670788.78 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3668, pruned_loss=0.1194, over 5701578.16 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3961, pruned_loss=0.1437, over 5654781.26 frames. ], batch size: 284, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:05:20,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:05:35,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3420, 3.1355, 1.5067, 1.4616], device='cuda:0'), covar=tensor([0.0940, 0.0299, 0.0845, 0.1290], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0527, 0.0357, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 02:05:43,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.174e+03 1.521e+03 1.935e+03 2.794e+03 4.613e+03, threshold=3.869e+03, percent-clipped=3.0
+2023-03-08 02:05:43,069 INFO [train.py:968] (0/2) Epoch 15, batch 41650, giga_loss[loss=0.2767, simple_loss=0.3498, pruned_loss=0.1018, over 29045.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3926, pruned_loss=0.1407, over 5654645.04 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3671, pruned_loss=0.1196, over 5705899.12 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3962, pruned_loss=0.1434, over 5636337.35 frames. ], batch size: 128, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:06:26,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3604, 1.2177, 4.3355, 3.3942], device='cuda:0'), covar=tensor([0.1698, 0.2834, 0.0400, 0.1050], device='cuda:0'), in_proj_covar=tensor([0.0705, 0.0613, 0.0902, 0.0822], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 02:06:35,625 INFO [train.py:968] (0/2) Epoch 15, batch 41700, giga_loss[loss=0.2688, simple_loss=0.3555, pruned_loss=0.09099, over 28921.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3893, pruned_loss=0.1366, over 5658193.07 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3673, pruned_loss=0.1198, over 5709595.88 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3927, pruned_loss=0.1393, over 5639026.83 frames. ], batch size: 174, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:06:49,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=680634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:07:23,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.483e+02 1.478e+03 1.850e+03 2.539e+03 9.007e+03, threshold=3.700e+03, percent-clipped=8.0
+2023-03-08 02:07:23,274 INFO [train.py:968] (0/2) Epoch 15, batch 41750, giga_loss[loss=0.3761, simple_loss=0.4011, pruned_loss=0.1755, over 23787.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3867, pruned_loss=0.1338, over 5662344.47 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3676, pruned_loss=0.1202, over 5711916.71 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3896, pruned_loss=0.1358, over 5643882.72 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:07:41,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:07:43,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:07:57,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4706, 1.7387, 1.7414, 1.2903], device='cuda:0'), covar=tensor([0.1751, 0.2533, 0.1467, 0.1719], device='cuda:0'), in_proj_covar=tensor([0.0853, 0.0688, 0.0897, 0.0800], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 02:08:12,823 INFO [train.py:968] (0/2) Epoch 15, batch 41800, giga_loss[loss=0.3055, simple_loss=0.3723, pruned_loss=0.1194, over 27960.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.382, pruned_loss=0.1296, over 5673602.74 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3668, pruned_loss=0.1199, over 5713303.64 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3855, pruned_loss=0.1318, over 5656426.94 frames. ], batch size: 412, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:08:13,109 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:08:46,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:09:05,687 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.868e+03 2.589e+03 4.007e+03 8.806e+03, threshold=5.177e+03, percent-clipped=27.0
+2023-03-08 02:09:05,701 INFO [train.py:968] (0/2) Epoch 15, batch 41850, giga_loss[loss=0.3642, simple_loss=0.389, pruned_loss=0.1697, over 23658.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3788, pruned_loss=0.1273, over 5663596.62 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3668, pruned_loss=0.12, over 5713726.37 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3818, pruned_loss=0.1292, over 5648129.90 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:09:13,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:09:56,597 INFO [train.py:968] (0/2) Epoch 15, batch 41900, giga_loss[loss=0.3365, simple_loss=0.3949, pruned_loss=0.139, over 28832.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3791, pruned_loss=0.1283, over 5649567.62 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3667, pruned_loss=0.12, over 5705511.43 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3819, pruned_loss=0.13, over 5643446.23 frames. ], batch size: 243, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:10:22,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4933, 4.3198, 4.0946, 2.0879], device='cuda:0'), covar=tensor([0.0555, 0.0721, 0.0718, 0.1888], device='cuda:0'), in_proj_covar=tensor([0.1144, 0.1056, 0.0908, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 02:10:47,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.983e+02 1.521e+03 1.922e+03 2.624e+03 7.262e+03, threshold=3.844e+03, percent-clipped=2.0
+2023-03-08 02:10:47,372 INFO [train.py:968] (0/2) Epoch 15, batch 41950, giga_loss[loss=0.4117, simple_loss=0.4317, pruned_loss=0.1958, over 26742.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3808, pruned_loss=0.1294, over 5659838.85 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3668, pruned_loss=0.12, over 5706609.04 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.383, pruned_loss=0.1308, over 5653548.38 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:11:11,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:11:13,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:11:41,989 INFO [train.py:968] (0/2) Epoch 15, batch 42000, giga_loss[loss=0.2554, simple_loss=0.3307, pruned_loss=0.09005, over 28868.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3772, pruned_loss=0.1255, over 5675411.66 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3669, pruned_loss=0.12, over 5710071.91 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3791, pruned_loss=0.1267, over 5666649.55 frames. ], batch size: 99, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 02:11:41,993 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 02:11:45,792 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3339, 3.1276, 1.4151, 1.5537], device='cuda:0'), covar=tensor([0.1097, 0.0357, 0.1033, 0.1490], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0531, 0.0359, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 02:11:50,697 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.209, simple_loss=0.3159, pruned_loss=0.05104, over 944034.00 frames. 
+2023-03-08 02:11:50,697 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 02:11:53,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:11:53,740 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:11:56,574 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:12:28,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:12:46,254 INFO [train.py:968] (0/2) Epoch 15, batch 42050, giga_loss[loss=0.3135, simple_loss=0.3941, pruned_loss=0.1164, over 28914.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3781, pruned_loss=0.1242, over 5675755.22 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.367, pruned_loss=0.12, over 5713081.62 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3797, pruned_loss=0.1252, over 5665520.17 frames. ], batch size: 136, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:12:46,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.300e+02 1.383e+03 1.911e+03 2.750e+03 8.494e+03, threshold=3.822e+03, percent-clipped=14.0
+2023-03-08 02:12:56,317 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-08 02:13:27,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=681009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:13:36,572 INFO [train.py:968] (0/2) Epoch 15, batch 42100, giga_loss[loss=0.3422, simple_loss=0.3986, pruned_loss=0.1429, over 29155.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3798, pruned_loss=0.1238, over 5680305.01 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3675, pruned_loss=0.1204, over 5715313.22 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3809, pruned_loss=0.1244, over 5669508.69 frames. ], batch size: 113, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:14:01,443 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1810, 1.1214, 3.6069, 3.0874], device='cuda:0'), covar=tensor([0.1638, 0.2698, 0.0492, 0.1009], device='cuda:0'), in_proj_covar=tensor([0.0704, 0.0613, 0.0901, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 02:14:23,897 INFO [train.py:968] (0/2) Epoch 15, batch 42150, giga_loss[loss=0.3568, simple_loss=0.3966, pruned_loss=0.1585, over 26536.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3793, pruned_loss=0.1244, over 5670889.42 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3671, pruned_loss=0.1201, over 5710623.92 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3809, pruned_loss=0.1252, over 5666021.43 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:14:24,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 1.784e+03 2.522e+03 3.737e+03 1.213e+04, threshold=5.043e+03, percent-clipped=23.0
+2023-03-08 02:14:52,720 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:14:59,782 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5507, 1.7767, 1.4132, 1.6674], device='cuda:0'), covar=tensor([0.2508, 0.2620, 0.2878, 0.2374], device='cuda:0'), in_proj_covar=tensor([0.1394, 0.1021, 0.1237, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 02:15:01,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:15:12,086 INFO [train.py:968] (0/2) Epoch 15, batch 42200, giga_loss[loss=0.2766, simple_loss=0.3527, pruned_loss=0.1002, over 28509.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3792, pruned_loss=0.1248, over 5672908.61 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3671, pruned_loss=0.1201, over 5710675.62 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3806, pruned_loss=0.1256, over 5668070.19 frames. ], batch size: 65, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:15:43,372 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=681152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:15:46,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=681155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:15:58,992 INFO [train.py:968] (0/2) Epoch 15, batch 42250, giga_loss[loss=0.3613, simple_loss=0.3976, pruned_loss=0.1624, over 23494.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3772, pruned_loss=0.1243, over 5678702.30 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3667, pruned_loss=0.1199, over 5714141.79 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.379, pruned_loss=0.1252, over 5670964.11 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:15:59,610 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.615e+03 1.978e+03 2.666e+03 8.325e+03, threshold=3.956e+03, percent-clipped=2.0
+2023-03-08 02:16:14,649 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=681184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:16:51,457 INFO [train.py:968] (0/2) Epoch 15, batch 42300, giga_loss[loss=0.359, simple_loss=0.4109, pruned_loss=0.1535, over 28590.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3759, pruned_loss=0.1247, over 5669227.23 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3667, pruned_loss=0.1199, over 5716848.30 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3774, pruned_loss=0.1255, over 5660008.47 frames. ], batch size: 307, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:17:12,057 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 02:17:44,972 INFO [train.py:968] (0/2) Epoch 15, batch 42350, giga_loss[loss=0.2835, simple_loss=0.3576, pruned_loss=0.1047, over 28864.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3741, pruned_loss=0.123, over 5670913.71 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3663, pruned_loss=0.1195, over 5717917.25 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3758, pruned_loss=0.124, over 5661850.67 frames. ], batch size: 66, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:17:45,878 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.660e+03 2.056e+03 2.781e+03 5.841e+03, threshold=4.112e+03, percent-clipped=10.0
+2023-03-08 02:18:13,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4920, 1.6196, 1.5911, 1.3784], device='cuda:0'), covar=tensor([0.2166, 0.2002, 0.1642, 0.1892], device='cuda:0'), in_proj_covar=tensor([0.1856, 0.1773, 0.1701, 0.1845], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 02:18:29,847 INFO [train.py:968] (0/2) Epoch 15, batch 42400, giga_loss[loss=0.3145, simple_loss=0.3773, pruned_loss=0.1259, over 28062.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3728, pruned_loss=0.1204, over 5671526.20 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3663, pruned_loss=0.1195, over 5703701.90 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3743, pruned_loss=0.1212, over 5676203.72 frames. ], batch size: 412, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 02:18:42,588 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-08 02:18:44,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9389, 3.7256, 3.5324, 1.8233], device='cuda:0'), covar=tensor([0.0708, 0.0807, 0.0811, 0.2299], device='cuda:0'), in_proj_covar=tensor([0.1155, 0.1065, 0.0914, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 02:19:23,816 INFO [train.py:968] (0/2) Epoch 15, batch 42450, giga_loss[loss=0.3136, simple_loss=0.3842, pruned_loss=0.1215, over 28726.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3738, pruned_loss=0.1204, over 5672995.32 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3663, pruned_loss=0.1195, over 5703261.39 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3751, pruned_loss=0.1211, over 5676902.31 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:19:25,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.875e+02 1.603e+03 2.085e+03 2.964e+03 7.439e+03, threshold=4.170e+03, percent-clipped=7.0
+2023-03-08 02:19:48,288 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681397.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:19:52,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9188, 3.7312, 3.5360, 1.7044], device='cuda:0'), covar=tensor([0.0731, 0.0851, 0.0853, 0.2227], device='cuda:0'), in_proj_covar=tensor([0.1154, 0.1065, 0.0915, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 02:20:02,516 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 02:20:12,242 INFO [train.py:968] (0/2) Epoch 15, batch 42500, giga_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.1141, over 28303.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3727, pruned_loss=0.1202, over 5679228.71 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3664, pruned_loss=0.1195, over 5707175.80 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3738, pruned_loss=0.1207, over 5678444.43 frames. ], batch size: 368, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:20:59,085 INFO [train.py:968] (0/2) Epoch 15, batch 42550, giga_loss[loss=0.2787, simple_loss=0.3503, pruned_loss=0.1035, over 28843.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3712, pruned_loss=0.1196, over 5680621.66 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3666, pruned_loss=0.1195, over 5708962.51 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.372, pruned_loss=0.1201, over 5677887.39 frames. ], batch size: 186, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:21:01,847 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.602e+03 2.115e+03 2.954e+03 6.513e+03, threshold=4.229e+03, percent-clipped=10.0
+2023-03-08 02:21:04,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=681474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:21:10,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=681482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:21:11,438 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-08 02:21:27,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5251, 1.6962, 1.6647, 1.5555], device='cuda:0'), covar=tensor([0.1716, 0.1958, 0.2143, 0.1885], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0739, 0.0692, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 02:21:46,530 INFO [train.py:968] (0/2) Epoch 15, batch 42600, giga_loss[loss=0.3771, simple_loss=0.4073, pruned_loss=0.1735, over 26643.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3712, pruned_loss=0.1208, over 5676937.44 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3668, pruned_loss=0.1197, over 5716640.75 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3718, pruned_loss=0.1209, over 5666894.53 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:22:35,616 INFO [train.py:968] (0/2) Epoch 15, batch 42650, giga_loss[loss=0.2802, simple_loss=0.3542, pruned_loss=0.1031, over 28721.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3707, pruned_loss=0.1212, over 5673817.72 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3668, pruned_loss=0.1197, over 5718514.89 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3713, pruned_loss=0.1214, over 5663244.56 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:22:37,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.945e+03 2.724e+03 4.085e+03 1.056e+04, threshold=5.448e+03, percent-clipped=22.0
+2023-03-08 02:22:57,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:21,991 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=681617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:23,460 INFO [train.py:968] (0/2) Epoch 15, batch 42700, giga_loss[loss=0.3618, simple_loss=0.3998, pruned_loss=0.1619, over 26674.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3687, pruned_loss=0.1204, over 5674850.25 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3665, pruned_loss=0.1196, over 5712485.87 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3695, pruned_loss=0.1208, over 5669630.30 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:23:24,474 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=681620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:29,329 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-08 02:23:29,870 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=681625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:32,752 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:32,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=681628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:51,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2701, 1.6368, 1.5850, 1.4096], device='cuda:0'), covar=tensor([0.1572, 0.1283, 0.1887, 0.1495], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0737, 0.0689, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 02:23:53,639 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=681649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:24:00,246 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=681657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:24:11,696 INFO [train.py:968] (0/2) Epoch 15, batch 42750, giga_loss[loss=0.301, simple_loss=0.3643, pruned_loss=0.1189, over 28383.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3681, pruned_loss=0.1201, over 5685730.30 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3661, pruned_loss=0.1192, over 5717593.38 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3692, pruned_loss=0.1208, over 5675499.28 frames. ], batch size: 71, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:24:15,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.834e+02 1.753e+03 2.154e+03 3.057e+03 8.416e+03, threshold=4.307e+03, percent-clipped=2.0
+2023-03-08 02:24:29,772 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681686.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:25:00,882 INFO [train.py:968] (0/2) Epoch 15, batch 42800, giga_loss[loss=0.305, simple_loss=0.3746, pruned_loss=0.1178, over 28785.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3688, pruned_loss=0.1211, over 5690454.10 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3665, pruned_loss=0.1197, over 5721092.97 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3695, pruned_loss=0.1213, over 5677935.41 frames. ], batch size: 186, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:25:27,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:25:50,423 INFO [train.py:968] (0/2) Epoch 15, batch 42850, giga_loss[loss=0.3122, simple_loss=0.3791, pruned_loss=0.1226, over 28657.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5685716.22 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3665, pruned_loss=0.1196, over 5715475.92 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3702, pruned_loss=0.121, over 5679860.51 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:25:52,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.158e+03 1.650e+03 1.968e+03 2.820e+03 9.481e+03, threshold=3.936e+03, percent-clipped=13.0
+2023-03-08 02:25:53,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=681772.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:26:35,829 INFO [train.py:968] (0/2) Epoch 15, batch 42900, giga_loss[loss=0.2906, simple_loss=0.3607, pruned_loss=0.1102, over 28626.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3695, pruned_loss=0.1198, over 5690114.96 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3662, pruned_loss=0.1194, over 5719882.28 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3702, pruned_loss=0.1201, over 5680843.53 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:26:45,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4593, 1.8012, 1.6692, 1.5767], device='cuda:0'), covar=tensor([0.1753, 0.1821, 0.2088, 0.1960], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0735, 0.0687, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 02:27:25,446 INFO [train.py:968] (0/2) Epoch 15, batch 42950, giga_loss[loss=0.3816, simple_loss=0.4131, pruned_loss=0.1751, over 26662.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3691, pruned_loss=0.1188, over 5683438.75 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3658, pruned_loss=0.1192, over 5722656.26 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3702, pruned_loss=0.1194, over 5673079.53 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:27:29,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.470e+02 1.544e+03 1.984e+03 2.729e+03 8.583e+03, threshold=3.969e+03, percent-clipped=12.0
+2023-03-08 02:28:14,633 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=681915.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:28:16,678 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=681918.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:28:17,037 INFO [train.py:968] (0/2) Epoch 15, batch 43000, giga_loss[loss=0.3186, simple_loss=0.3857, pruned_loss=0.1257, over 28956.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3707, pruned_loss=0.1208, over 5665828.57 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3656, pruned_loss=0.119, over 5712176.69 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3719, pruned_loss=0.1214, over 5664697.61 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:28:43,935 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=681947.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:29:04,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=681968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:29:05,151 INFO [train.py:968] (0/2) Epoch 15, batch 43050, giga_loss[loss=0.4508, simple_loss=0.4714, pruned_loss=0.2151, over 26589.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3744, pruned_loss=0.1247, over 5660259.04 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3655, pruned_loss=0.1191, over 5708559.56 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3756, pruned_loss=0.1253, over 5660772.57 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:29:08,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.809e+02 1.655e+03 2.083e+03 2.757e+03 1.010e+04, threshold=4.166e+03, percent-clipped=9.0
+2023-03-08 02:29:17,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0094, 2.5250, 1.0926, 1.2761], device='cuda:0'), covar=tensor([0.1235, 0.0563, 0.1020, 0.1576], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0532, 0.0359, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 02:29:32,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-08 02:29:36,010 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-682000.pt
+2023-03-08 02:29:39,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=682003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:29:48,746 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5010, 1.5057, 1.2169, 1.1471], device='cuda:0'), covar=tensor([0.0749, 0.0509, 0.0983, 0.1004], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0447, 0.0506, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 02:29:49,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2816, 1.1841, 1.1491, 1.4727], device='cuda:0'), covar=tensor([0.0748, 0.0343, 0.0326, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 02:29:56,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2662, 1.2894, 4.0693, 3.4146], device='cuda:0'), covar=tensor([0.1718, 0.2699, 0.0437, 0.0759], device='cuda:0'), in_proj_covar=tensor([0.0707, 0.0617, 0.0907, 0.0826], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 02:29:56,723 INFO [train.py:968] (0/2) Epoch 15, batch 43100, giga_loss[loss=0.2708, simple_loss=0.3447, pruned_loss=0.09847, over 29008.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3766, pruned_loss=0.1279, over 5659856.61 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3656, pruned_loss=0.119, over 5713021.70 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3778, pruned_loss=0.1286, over 5655009.68 frames. ], batch size: 136, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:30:42,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=682061.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:30:49,158 INFO [train.py:968] (0/2) Epoch 15, batch 43150, giga_loss[loss=0.4048, simple_loss=0.4315, pruned_loss=0.189, over 23900.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3763, pruned_loss=0.1287, over 5657585.43 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3652, pruned_loss=0.1188, over 5714847.79 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3779, pruned_loss=0.1297, over 5650604.44 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:30:54,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.184e+02 1.864e+03 2.309e+03 3.353e+03 1.110e+04, threshold=4.619e+03, percent-clipped=13.0
+2023-03-08 02:31:01,557 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-08 02:31:11,486 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-08 02:31:32,249 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:31:34,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:31:37,501 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=682118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:31:37,886 INFO [train.py:968] (0/2) Epoch 15, batch 43200, giga_loss[loss=0.4061, simple_loss=0.439, pruned_loss=0.1866, over 23925.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3764, pruned_loss=0.1287, over 5662336.11 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3648, pruned_loss=0.1184, over 5713553.37 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3786, pruned_loss=0.1303, over 5656193.48 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:31:43,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=682124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:31:44,558 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=682125.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:32:01,415 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:32:02,405 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-08 02:32:05,045 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:32:07,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:32:25,613 INFO [train.py:968] (0/2) Epoch 15, batch 43250, giga_loss[loss=0.3859, simple_loss=0.4196, pruned_loss=0.176, over 26540.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3748, pruned_loss=0.1275, over 5669958.09 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3649, pruned_loss=0.1185, over 5715487.34 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3765, pruned_loss=0.1288, over 5662522.42 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:32:29,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.251e+03 1.600e+03 2.019e+03 2.791e+03 9.170e+03, threshold=4.038e+03, percent-clipped=4.0
+2023-03-08 02:32:34,410 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:32:56,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2006, 2.6381, 1.1991, 1.4143], device='cuda:0'), covar=tensor([0.0988, 0.0401, 0.0916, 0.1351], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0532, 0.0359, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 02:32:58,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682204.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:33:01,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682207.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:33:09,770 INFO [train.py:968] (0/2) Epoch 15, batch 43300, giga_loss[loss=0.2641, simple_loss=0.3474, pruned_loss=0.09044, over 28863.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3729, pruned_loss=0.1245, over 5680045.24 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3648, pruned_loss=0.1183, over 5719514.91 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3746, pruned_loss=0.1259, over 5669720.67 frames. ], batch size: 199, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:33:28,292 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682236.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:33:59,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:34:00,267 INFO [train.py:968] (0/2) Epoch 15, batch 43350, giga_loss[loss=0.2924, simple_loss=0.3587, pruned_loss=0.1131, over 28714.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3702, pruned_loss=0.1218, over 5680371.74 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3646, pruned_loss=0.1181, over 5722479.19 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3719, pruned_loss=0.1231, over 5668816.74 frames. ], batch size: 284, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:34:03,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:34:05,095 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 1.560e+03 1.978e+03 2.718e+03 6.663e+03, threshold=3.957e+03, percent-clipped=7.0
+2023-03-08 02:34:27,914 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:34:45,109 INFO [train.py:968] (0/2) Epoch 15, batch 43400, giga_loss[loss=0.2571, simple_loss=0.3249, pruned_loss=0.09463, over 28829.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3697, pruned_loss=0.1218, over 5682411.96 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.366, pruned_loss=0.1189, over 5731489.74 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3701, pruned_loss=0.1224, over 5662458.49 frames. ], batch size: 99, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:35:31,077 INFO [train.py:968] (0/2) Epoch 15, batch 43450, giga_loss[loss=0.3151, simple_loss=0.3799, pruned_loss=0.1252, over 28488.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3685, pruned_loss=0.1216, over 5681978.70 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3659, pruned_loss=0.1189, over 5731299.72 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.369, pruned_loss=0.1222, over 5664162.13 frames. ], batch size: 71, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:35:33,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.741e+03 2.192e+03 3.030e+03 7.964e+03, threshold=4.385e+03, percent-clipped=14.0
+2023-03-08 02:35:41,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9505, 1.1193, 3.3897, 2.9865], device='cuda:0'), covar=tensor([0.1732, 0.2629, 0.0503, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0710, 0.0618, 0.0908, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 02:35:53,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7801, 2.0554, 1.3942, 1.6240], device='cuda:0'), covar=tensor([0.0739, 0.0434, 0.0968, 0.0899], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0445, 0.0504, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 02:36:19,658 INFO [train.py:968] (0/2) Epoch 15, batch 43500, giga_loss[loss=0.2824, simple_loss=0.3519, pruned_loss=0.1064, over 28713.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3692, pruned_loss=0.1229, over 5675165.64 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3662, pruned_loss=0.1191, over 5733828.91 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3694, pruned_loss=0.1232, over 5657746.76 frames. ], batch size: 119, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:36:37,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 02:37:09,124 INFO [train.py:968] (0/2) Epoch 15, batch 43550, giga_loss[loss=0.3373, simple_loss=0.3709, pruned_loss=0.1518, over 23720.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3714, pruned_loss=0.1233, over 5679044.34 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.366, pruned_loss=0.1188, over 5735076.29 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3718, pruned_loss=0.1239, over 5663035.11 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:37:14,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.577e+03 2.044e+03 2.711e+03 4.906e+03, threshold=4.089e+03, percent-clipped=2.0
+2023-03-08 02:37:32,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=682493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:37:39,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=682500.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:37:56,922 INFO [train.py:968] (0/2) Epoch 15, batch 43600, giga_loss[loss=0.3592, simple_loss=0.419, pruned_loss=0.1497, over 28597.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3754, pruned_loss=0.1236, over 5677632.45 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3659, pruned_loss=0.1187, over 5735930.31 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3761, pruned_loss=0.1244, over 5662392.76 frames. ], batch size: 336, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 02:38:56,145 INFO [train.py:968] (0/2) Epoch 15, batch 43650, giga_loss[loss=0.3014, simple_loss=0.3685, pruned_loss=0.1172, over 28887.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3773, pruned_loss=0.1237, over 5671675.49 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3658, pruned_loss=0.1186, over 5736795.66 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3779, pruned_loss=0.1244, over 5658575.29 frames. ], batch size: 227, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:38:59,344 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.687e+02 1.484e+03 1.976e+03 2.932e+03 8.857e+03, threshold=3.951e+03, percent-clipped=8.0
+2023-03-08 02:39:39,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4322, 1.6458, 1.3191, 1.6202], device='cuda:0'), covar=tensor([0.2448, 0.2535, 0.2817, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.1396, 0.1022, 0.1236, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 02:39:44,025 INFO [train.py:968] (0/2) Epoch 15, batch 43700, giga_loss[loss=0.3576, simple_loss=0.4155, pruned_loss=0.1498, over 28728.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.379, pruned_loss=0.1252, over 5681200.75 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3655, pruned_loss=0.1184, over 5739957.52 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.38, pruned_loss=0.126, over 5666689.80 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:40:02,349 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:40:05,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:40:10,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682643.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:40:13,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682646.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:40:13,677 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=682647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:40:35,037 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:40:35,831 INFO [train.py:968] (0/2) Epoch 15, batch 43750, giga_loss[loss=0.3011, simple_loss=0.3684, pruned_loss=0.1169, over 29070.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3811, pruned_loss=0.1276, over 5673338.00 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3661, pruned_loss=0.1189, over 5734555.94 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3817, pruned_loss=0.128, over 5665828.79 frames. ], batch size: 128, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:40:40,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.601e+03 2.113e+03 2.978e+03 1.224e+04, threshold=4.225e+03, percent-clipped=10.0
+2023-03-08 02:40:43,659 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682675.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:41:06,557 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2010, 1.2222, 1.0861, 0.8738], device='cuda:0'), covar=tensor([0.0804, 0.0493, 0.0990, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0443, 0.0504, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 02:41:20,433 INFO [train.py:968] (0/2) Epoch 15, batch 43800, giga_loss[loss=0.2857, simple_loss=0.3548, pruned_loss=0.1083, over 28825.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3798, pruned_loss=0.1274, over 5675068.45 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3659, pruned_loss=0.1188, over 5729487.84 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3809, pruned_loss=0.1281, over 5671465.00 frames. ], batch size: 99, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:41:27,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4018, 1.2277, 4.5223, 3.5298], device='cuda:0'), covar=tensor([0.1707, 0.2715, 0.0390, 0.0910], device='cuda:0'), in_proj_covar=tensor([0.0713, 0.0618, 0.0912, 0.0830], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 02:42:11,537 INFO [train.py:968] (0/2) Epoch 15, batch 43850, giga_loss[loss=0.2899, simple_loss=0.3549, pruned_loss=0.1124, over 28691.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3779, pruned_loss=0.1271, over 5651752.18 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3658, pruned_loss=0.1188, over 5717839.22 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.379, pruned_loss=0.1277, over 5659434.44 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:42:15,291 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.844e+02 1.541e+03 1.892e+03 2.713e+03 1.157e+04, threshold=3.784e+03, percent-clipped=5.0
+2023-03-08 02:42:35,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=682792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:42:52,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4742, 1.6700, 1.3718, 1.5085], device='cuda:0'), covar=tensor([0.2191, 0.2238, 0.2449, 0.1976], device='cuda:0'), in_proj_covar=tensor([0.1396, 0.1022, 0.1236, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 02:43:01,435 INFO [train.py:968] (0/2) Epoch 15, batch 43900, giga_loss[loss=0.2787, simple_loss=0.3539, pruned_loss=0.1018, over 28999.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3755, pruned_loss=0.1261, over 5657049.36 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3657, pruned_loss=0.1187, over 5720840.82 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3766, pruned_loss=0.1268, over 5659614.01 frames. ], batch size: 155, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:43:48,570 INFO [train.py:968] (0/2) Epoch 15, batch 43950, giga_loss[loss=0.3326, simple_loss=0.3716, pruned_loss=0.1468, over 23681.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3749, pruned_loss=0.1261, over 5655262.66 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.366, pruned_loss=0.1188, over 5717865.27 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3759, pruned_loss=0.1269, over 5657028.19 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:43:58,474 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.943e+02 1.654e+03 2.179e+03 2.752e+03 7.968e+03, threshold=4.357e+03, percent-clipped=13.0
+2023-03-08 02:44:39,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5961, 1.7755, 1.4778, 1.8317], device='cuda:0'), covar=tensor([0.2274, 0.2367, 0.2506, 0.2246], device='cuda:0'), in_proj_covar=tensor([0.1398, 0.1023, 0.1237, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 02:44:41,079 INFO [train.py:968] (0/2) Epoch 15, batch 44000, giga_loss[loss=0.2916, simple_loss=0.3613, pruned_loss=0.111, over 28978.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3753, pruned_loss=0.1267, over 5650897.68 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3659, pruned_loss=0.1188, over 5721477.98 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3764, pruned_loss=0.1276, over 5647138.81 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:44:55,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=682931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:45:03,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7334, 1.8012, 1.2931, 1.3702], device='cuda:0'), covar=tensor([0.0868, 0.0615, 0.1058, 0.1059], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0445, 0.0504, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 02:45:32,875 INFO [train.py:968] (0/2) Epoch 15, batch 44050, giga_loss[loss=0.2983, simple_loss=0.3673, pruned_loss=0.1146, over 28643.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3739, pruned_loss=0.1265, over 5647623.29 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3658, pruned_loss=0.1187, over 5715598.49 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3751, pruned_loss=0.1274, over 5648789.62 frames. ], batch size: 336, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:45:37,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.854e+03 2.422e+03 3.492e+03 8.094e+03, threshold=4.845e+03, percent-clipped=17.0
+2023-03-08 02:45:55,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6232, 1.7494, 1.6379, 1.5087], device='cuda:0'), covar=tensor([0.1680, 0.2220, 0.2284, 0.2187], device='cuda:0'), in_proj_covar=tensor([0.0450, 0.0739, 0.0691, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 02:46:09,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3288, 1.8634, 1.4525, 1.5877], device='cuda:0'), covar=tensor([0.0757, 0.0299, 0.0314, 0.0792], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 02:46:20,140 INFO [train.py:968] (0/2) Epoch 15, batch 44100, giga_loss[loss=0.3701, simple_loss=0.4071, pruned_loss=0.1665, over 26622.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3731, pruned_loss=0.1261, over 5659656.49 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3659, pruned_loss=0.1187, over 5717604.15 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.374, pruned_loss=0.1268, over 5658247.93 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:46:25,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:46:49,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=683047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:47:07,594 INFO [train.py:968] (0/2) Epoch 15, batch 44150, giga_loss[loss=0.358, simple_loss=0.41, pruned_loss=0.153, over 28311.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3725, pruned_loss=0.1254, over 5658998.33 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3658, pruned_loss=0.1186, over 5719741.17 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3734, pruned_loss=0.1262, over 5654745.68 frames. ], batch size: 368, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:47:13,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.324e+02 1.454e+03 1.852e+03 2.749e+03 5.986e+03, threshold=3.704e+03, percent-clipped=3.0
+2023-03-08 02:48:00,383 INFO [train.py:968] (0/2) Epoch 15, batch 44200, giga_loss[loss=0.3095, simple_loss=0.3796, pruned_loss=0.1197, over 28948.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3761, pruned_loss=0.1275, over 5642278.89 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3661, pruned_loss=0.1189, over 5711132.56 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3766, pruned_loss=0.1279, over 5645120.40 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:48:45,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=683165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:48:48,065 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:48:49,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=683168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:48:49,360 INFO [train.py:968] (0/2) Epoch 15, batch 44250, giga_loss[loss=0.2914, simple_loss=0.3608, pruned_loss=0.111, over 28821.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3768, pruned_loss=0.1277, over 5647529.13 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3668, pruned_loss=0.1193, over 5711883.50 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3769, pruned_loss=0.128, over 5647521.41 frames. ], batch size: 199, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:48:54,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.150e+03 1.587e+03 2.054e+03 2.953e+03 1.217e+04, threshold=4.108e+03, percent-clipped=16.0
+2023-03-08 02:48:56,656 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=683179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:49:16,018 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=683197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:49:35,216 INFO [train.py:968] (0/2) Epoch 15, batch 44300, giga_loss[loss=0.3203, simple_loss=0.3999, pruned_loss=0.1203, over 28852.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3751, pruned_loss=0.1262, over 5666104.65 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3657, pruned_loss=0.1186, over 5718188.55 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3765, pruned_loss=0.1274, over 5657947.34 frames. ], batch size: 174, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:50:20,272 INFO [train.py:968] (0/2) Epoch 15, batch 44350, giga_loss[loss=0.3452, simple_loss=0.4028, pruned_loss=0.1438, over 28345.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3762, pruned_loss=0.124, over 5658794.02 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3662, pruned_loss=0.119, over 5704437.99 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3771, pruned_loss=0.1247, over 5661640.23 frames. ], batch size: 368, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:50:21,707 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 02:50:25,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.416e+02 1.505e+03 1.930e+03 2.844e+03 5.755e+03, threshold=3.860e+03, percent-clipped=14.0
+2023-03-08 02:50:49,171 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:50:53,752 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=683310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:50:57,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=683313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:51:02,923 INFO [train.py:968] (0/2) Epoch 15, batch 44400, giga_loss[loss=0.3447, simple_loss=0.4114, pruned_loss=0.139, over 28681.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3778, pruned_loss=0.1231, over 5650066.24 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3662, pruned_loss=0.1189, over 5694222.53 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3789, pruned_loss=0.1239, over 5659354.43 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:51:24,668 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=683342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:51:34,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4307, 3.4726, 1.5691, 1.6505], device='cuda:0'), covar=tensor([0.0967, 0.0421, 0.0917, 0.1280], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0532, 0.0358, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 02:51:51,933 INFO [train.py:968] (0/2) Epoch 15, batch 44450, giga_loss[loss=0.2939, simple_loss=0.3613, pruned_loss=0.1132, over 28620.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3794, pruned_loss=0.1244, over 5644054.19 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.366, pruned_loss=0.1188, over 5691129.18 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3809, pruned_loss=0.1254, over 5652890.79 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:51:58,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.583e+02 1.553e+03 1.975e+03 3.090e+03 7.993e+03, threshold=3.951e+03, percent-clipped=18.0
+2023-03-08 02:52:36,430 INFO [train.py:968] (0/2) Epoch 15, batch 44500, giga_loss[loss=0.2851, simple_loss=0.3597, pruned_loss=0.1052, over 28968.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3827, pruned_loss=0.1279, over 5648633.76 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3666, pruned_loss=0.1195, over 5688024.32 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.384, pruned_loss=0.1284, over 5656239.09 frames. ], batch size: 213, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:52:40,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:52:47,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3306, 1.5270, 3.2014, 3.0125], device='cuda:0'), covar=tensor([0.1247, 0.2101, 0.0483, 0.0982], device='cuda:0'), in_proj_covar=tensor([0.0711, 0.0618, 0.0913, 0.0830], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 02:53:09,273 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=683449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:53:11,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=683452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:53:31,163 INFO [train.py:968] (0/2) Epoch 15, batch 44550, giga_loss[loss=0.3326, simple_loss=0.3958, pruned_loss=0.1347, over 28964.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3849, pruned_loss=0.1305, over 5651434.97 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3668, pruned_loss=0.1196, over 5690447.65 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3858, pruned_loss=0.1308, over 5654972.47 frames. ], batch size: 145, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:53:39,958 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.133e+03 1.841e+03 2.593e+03 3.539e+03 1.096e+04, threshold=5.185e+03, percent-clipped=18.0
+2023-03-08 02:53:43,255 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=683481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:54:18,803 INFO [train.py:968] (0/2) Epoch 15, batch 44600, giga_loss[loss=0.3286, simple_loss=0.3934, pruned_loss=0.1319, over 28647.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3839, pruned_loss=0.1302, over 5668992.26 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3671, pruned_loss=0.1198, over 5694379.22 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3847, pruned_loss=0.1304, over 5667573.86 frames. ], batch size: 336, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:54:51,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:54:59,910 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=683565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:55:01,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=683568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:55:02,183 INFO [train.py:968] (0/2) Epoch 15, batch 44650, giga_loss[loss=0.3169, simple_loss=0.3805, pruned_loss=0.1267, over 28983.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3822, pruned_loss=0.129, over 5663943.60 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3667, pruned_loss=0.1195, over 5696385.09 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3839, pruned_loss=0.1299, over 5660013.37 frames. ], batch size: 136, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:55:09,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.682e+02 1.497e+03 1.807e+03 2.676e+03 9.013e+03, threshold=3.613e+03, percent-clipped=2.0
+2023-03-08 02:55:27,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=683597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:55:48,956 INFO [train.py:968] (0/2) Epoch 15, batch 44700, libri_loss[loss=0.2629, simple_loss=0.325, pruned_loss=0.1004, over 29518.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3802, pruned_loss=0.1252, over 5675659.08 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3662, pruned_loss=0.1192, over 5692530.78 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3824, pruned_loss=0.1264, over 5674951.26 frames. ], batch size: 70, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:55:53,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=683623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:56:13,500 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=683644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:56:29,463 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=683660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:56:37,384 INFO [train.py:968] (0/2) Epoch 15, batch 44750, giga_loss[loss=0.4055, simple_loss=0.4399, pruned_loss=0.1855, over 27553.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3802, pruned_loss=0.1241, over 5671043.42 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3662, pruned_loss=0.1192, over 5686217.56 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3821, pruned_loss=0.1251, over 5675889.51 frames. ], batch size: 472, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:56:44,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.746e+02 1.487e+03 1.908e+03 2.398e+03 5.796e+03, threshold=3.817e+03, percent-clipped=9.0
+2023-03-08 02:57:06,819 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=683697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:57:10,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=683700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:57:26,284 INFO [train.py:968] (0/2) Epoch 15, batch 44800, giga_loss[loss=0.3476, simple_loss=0.399, pruned_loss=0.1481, over 27920.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.38, pruned_loss=0.1247, over 5654836.11 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3657, pruned_loss=0.1189, over 5682128.30 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3823, pruned_loss=0.1259, over 5662515.61 frames. ], batch size: 412, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:57:37,034 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=683729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:58:14,626 INFO [train.py:968] (0/2) Epoch 15, batch 44850, giga_loss[loss=0.277, simple_loss=0.3518, pruned_loss=0.1011, over 28844.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3798, pruned_loss=0.1252, over 5649917.76 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3661, pruned_loss=0.119, over 5680286.64 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3817, pruned_loss=0.1262, over 5656950.07 frames. ], batch size: 66, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:58:21,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.726e+03 2.356e+03 3.237e+03 6.204e+03, threshold=4.712e+03, percent-clipped=14.0
+2023-03-08 02:58:27,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=683784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:59:00,325 INFO [train.py:968] (0/2) Epoch 15, batch 44900, giga_loss[loss=0.3211, simple_loss=0.3714, pruned_loss=0.1354, over 27609.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3783, pruned_loss=0.1252, over 5606090.32 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3668, pruned_loss=0.1198, over 5627875.25 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3795, pruned_loss=0.1255, over 5658918.94 frames. ], batch size: 472, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:59:51,021 INFO [train.py:968] (0/2) Epoch 15, batch 44950, giga_loss[loss=0.3323, simple_loss=0.3926, pruned_loss=0.136, over 28853.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3774, pruned_loss=0.1259, over 5574248.83 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3677, pruned_loss=0.1207, over 5585523.40 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3777, pruned_loss=0.1254, over 5654527.78 frames. ], batch size: 199, lr: 2.11e-03, grad_scale: 1.0
+2023-03-08 02:59:57,015 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-08 02:59:58,719 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 2.022e+03 2.588e+03 4.347e+03 2.833e+04, threshold=5.177e+03, percent-clipped=19.0
+2023-03-08 03:00:37,611 INFO [train.py:968] (0/2) Epoch 15, batch 45000, giga_loss[loss=0.2901, simple_loss=0.3667, pruned_loss=0.1067, over 28945.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3764, pruned_loss=0.1258, over 5558411.21 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1214, over 5544844.82 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.376, pruned_loss=0.1248, over 5659049.83 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 1.0
+2023-03-08 03:00:37,615 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 03:00:41,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9639, 3.7399, 3.5890, 1.7852], device='cuda:0'), covar=tensor([0.0701, 0.0885, 0.0840, 0.2502], device='cuda:0'), in_proj_covar=tensor([0.1164, 0.1078, 0.0924, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 03:00:46,435 INFO [train.py:1012] (0/2) Epoch 15, validation: loss=0.2109, simple_loss=0.3196, pruned_loss=0.05111, over 944034.00 frames. 
+2023-03-08 03:00:46,436 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 03:01:19,751 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-08 03:01:21,239 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-15.pt
+2023-03-08 03:02:07,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9177, 3.7092, 3.5145, 1.7998], device='cuda:0'), covar=tensor([0.0678, 0.0829, 0.0777, 0.2427], device='cuda:0'), in_proj_covar=tensor([0.1163, 0.1076, 0.0922, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 03:02:19,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.412e+02 1.391e+03 2.167e+03 2.976e+03 6.931e+03, threshold=4.334e+03, percent-clipped=4.0
+2023-03-08 03:02:39,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9257, 2.2763, 2.0720, 1.6853], device='cuda:0'), covar=tensor([0.2710, 0.1957, 0.2174, 0.2446], device='cuda:0'), in_proj_covar=tensor([0.1856, 0.1782, 0.1712, 0.1861], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 03:02:41,185 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:02:42,201 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-684000.pt
+2023-03-08 03:02:43,358 INFO [train.py:968] (0/2) Epoch 16, batch 50, giga_loss[loss=0.2657, simple_loss=0.3579, pruned_loss=0.0868, over 28851.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3708, pruned_loss=0.1068, over 1254045.85 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3365, pruned_loss=0.08812, over 188692.11 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3762, pruned_loss=0.1098, over 1102653.71 frames. ], batch size: 174, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:03:02,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:03:03,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5623, 1.6433, 1.7378, 1.4819], device='cuda:0'), covar=tensor([0.2419, 0.2259, 0.1686, 0.2117], device='cuda:0'), in_proj_covar=tensor([0.1857, 0.1783, 0.1714, 0.1863], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 03:03:19,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:03:25,706 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:03:34,324 INFO [train.py:968] (0/2) Epoch 16, batch 100, giga_loss[loss=0.2227, simple_loss=0.3054, pruned_loss=0.07002, over 28422.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3638, pruned_loss=0.1041, over 2239875.44 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3442, pruned_loss=0.09147, over 329841.45 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3664, pruned_loss=0.1058, over 2026738.42 frames. ], batch size: 71, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:03:39,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7505, 2.1528, 1.9318, 1.4885], device='cuda:0'), covar=tensor([0.3011, 0.2175, 0.2331, 0.2802], device='cuda:0'), in_proj_covar=tensor([0.1861, 0.1787, 0.1718, 0.1865], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 03:04:00,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.844e+02 1.125e+03 1.393e+03 1.789e+03 3.236e+03, threshold=2.786e+03, percent-clipped=0.0
+2023-03-08 03:04:22,242 INFO [train.py:968] (0/2) Epoch 16, batch 150, giga_loss[loss=0.275, simple_loss=0.3325, pruned_loss=0.1087, over 26635.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3498, pruned_loss=0.09803, over 3004537.82 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3466, pruned_loss=0.09435, over 411811.52 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3504, pruned_loss=0.09852, over 2793708.84 frames. ], batch size: 555, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:04:50,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4072, 1.3075, 4.3067, 3.4798], device='cuda:0'), covar=tensor([0.1616, 0.2832, 0.0380, 0.0990], device='cuda:0'), in_proj_covar=tensor([0.0708, 0.0617, 0.0914, 0.0832], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 03:04:56,398 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:04:58,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:02,823 INFO [train.py:968] (0/2) Epoch 16, batch 200, libri_loss[loss=0.2194, simple_loss=0.3105, pruned_loss=0.06412, over 29551.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3391, pruned_loss=0.09277, over 3613915.41 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3481, pruned_loss=0.09338, over 679147.36 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3383, pruned_loss=0.093, over 3325428.87 frames. ], batch size: 78, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:05:08,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:11,344 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:13,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:21,084 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:22,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3255, 1.2347, 1.2309, 1.5178], device='cuda:0'), covar=tensor([0.0776, 0.0360, 0.0341, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 03:05:24,936 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:25,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.059e+02 1.076e+03 1.433e+03 1.823e+03 3.970e+03, threshold=2.866e+03, percent-clipped=5.0
+2023-03-08 03:05:28,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:38,861 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:46,524 INFO [train.py:968] (0/2) Epoch 16, batch 250, giga_loss[loss=0.197, simple_loss=0.2801, pruned_loss=0.05695, over 29017.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3266, pruned_loss=0.08658, over 4080511.68 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3465, pruned_loss=0.09228, over 757282.99 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3252, pruned_loss=0.08647, over 3825198.43 frames. ], batch size: 136, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:05:55,400 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-08 03:05:55,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:06:30,795 INFO [train.py:968] (0/2) Epoch 16, batch 300, giga_loss[loss=0.2123, simple_loss=0.2796, pruned_loss=0.07248, over 28531.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3188, pruned_loss=0.08345, over 4434113.37 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3451, pruned_loss=0.09177, over 899268.71 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3167, pruned_loss=0.08302, over 4195342.28 frames. ], batch size: 85, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:06:53,688 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.914e+02 9.455e+02 1.179e+03 1.595e+03 3.189e+03, threshold=2.358e+03, percent-clipped=2.0
+2023-03-08 03:07:16,895 INFO [train.py:968] (0/2) Epoch 16, batch 350, giga_loss[loss=0.2011, simple_loss=0.2831, pruned_loss=0.05955, over 29140.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3129, pruned_loss=0.0811, over 4710190.60 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3446, pruned_loss=0.09095, over 1037387.56 frames. ], giga_tot_loss[loss=0.2356, simple_loss=0.3102, pruned_loss=0.08052, over 4492195.66 frames. ], batch size: 128, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:07:18,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:07:19,461 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684304.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 03:07:20,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:07:43,869 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:07:57,766 INFO [train.py:968] (0/2) Epoch 16, batch 400, giga_loss[loss=0.206, simple_loss=0.2805, pruned_loss=0.06581, over 28883.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3076, pruned_loss=0.0787, over 4935341.39 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3436, pruned_loss=0.09038, over 1155302.32 frames. ], giga_tot_loss[loss=0.2303, simple_loss=0.3046, pruned_loss=0.07796, over 4739610.70 frames. ], batch size: 112, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:08:01,460 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:08:20,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.977e+02 1.017e+03 1.549e+03 1.995e+03 1.032e+04, threshold=3.099e+03, percent-clipped=17.0
+2023-03-08 03:08:35,723 INFO [train.py:968] (0/2) Epoch 16, batch 450, giga_loss[loss=0.2093, simple_loss=0.2868, pruned_loss=0.06587, over 28972.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3074, pruned_loss=0.07833, over 5116759.77 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3415, pruned_loss=0.08903, over 1452553.54 frames. ], giga_tot_loss[loss=0.2288, simple_loss=0.303, pruned_loss=0.07728, over 4904534.03 frames. ], batch size: 213, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:08:37,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5957, 1.8755, 1.5829, 1.5061], device='cuda:0'), covar=tensor([0.2748, 0.2750, 0.3105, 0.2490], device='cuda:0'), in_proj_covar=tensor([0.1408, 0.1031, 0.1250, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 03:08:48,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:09:17,596 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0826, 2.9158, 2.7785, 1.2903], device='cuda:0'), covar=tensor([0.0997, 0.1122, 0.1056, 0.2109], device='cuda:0'), in_proj_covar=tensor([0.1137, 0.1052, 0.0900, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 03:09:18,044 INFO [train.py:968] (0/2) Epoch 16, batch 500, libri_loss[loss=0.3101, simple_loss=0.3841, pruned_loss=0.118, over 26054.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3051, pruned_loss=0.07704, over 5246313.45 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3415, pruned_loss=0.08913, over 1613991.58 frames. ], giga_tot_loss[loss=0.2258, simple_loss=0.3002, pruned_loss=0.07568, over 5061451.65 frames. ], batch size: 136, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:09:43,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.134e+02 9.936e+02 1.303e+03 1.756e+03 4.075e+03, threshold=2.606e+03, percent-clipped=6.0
+2023-03-08 03:10:03,316 INFO [train.py:968] (0/2) Epoch 16, batch 550, giga_loss[loss=0.2044, simple_loss=0.2796, pruned_loss=0.0646, over 28963.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.303, pruned_loss=0.07628, over 5341278.76 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3419, pruned_loss=0.08901, over 1712721.17 frames. ], giga_tot_loss[loss=0.224, simple_loss=0.2981, pruned_loss=0.07494, over 5186756.82 frames. ], batch size: 106, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:10:07,246 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 03:10:50,491 INFO [train.py:968] (0/2) Epoch 16, batch 600, giga_loss[loss=0.1952, simple_loss=0.2754, pruned_loss=0.05745, over 28841.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3013, pruned_loss=0.07548, over 5420476.15 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08942, over 1769894.03 frames. ], giga_tot_loss[loss=0.2224, simple_loss=0.2967, pruned_loss=0.07411, over 5296089.36 frames. ], batch size: 145, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:10:58,160 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:11:00,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:11:19,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.988e+02 1.009e+03 1.197e+03 1.716e+03 5.081e+03, threshold=2.395e+03, percent-clipped=9.0
+2023-03-08 03:11:28,437 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:11:30,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6101, 1.7414, 1.7380, 1.5532], device='cuda:0'), covar=tensor([0.1820, 0.2083, 0.2288, 0.2175], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0738, 0.0688, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 03:11:37,105 INFO [train.py:968] (0/2) Epoch 16, batch 650, giga_loss[loss=0.2025, simple_loss=0.2782, pruned_loss=0.06344, over 28717.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.2993, pruned_loss=0.07491, over 5468070.45 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3424, pruned_loss=0.08957, over 1881228.66 frames. ], giga_tot_loss[loss=0.2206, simple_loss=0.2944, pruned_loss=0.07335, over 5366745.24 frames. ], batch size: 262, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:11:48,017 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 03:12:21,228 INFO [train.py:968] (0/2) Epoch 16, batch 700, giga_loss[loss=0.2215, simple_loss=0.2938, pruned_loss=0.07463, over 28838.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2983, pruned_loss=0.07454, over 5521448.90 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3432, pruned_loss=0.09006, over 2020418.57 frames. ], giga_tot_loss[loss=0.219, simple_loss=0.2927, pruned_loss=0.07263, over 5426539.27 frames. ], batch size: 199, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:12:49,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.598e+02 1.007e+03 1.236e+03 1.809e+03 4.810e+03, threshold=2.472e+03, percent-clipped=5.0
+2023-03-08 03:12:49,424 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684679.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 03:13:06,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1008, 3.8992, 3.6665, 1.9818], device='cuda:0'), covar=tensor([0.0610, 0.0816, 0.0778, 0.2023], device='cuda:0'), in_proj_covar=tensor([0.1130, 0.1047, 0.0896, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 03:13:06,710 INFO [train.py:968] (0/2) Epoch 16, batch 750, giga_loss[loss=0.1991, simple_loss=0.2695, pruned_loss=0.06435, over 28962.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2958, pruned_loss=0.07357, over 5550243.49 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3439, pruned_loss=0.09052, over 2078442.34 frames. ], giga_tot_loss[loss=0.2169, simple_loss=0.2904, pruned_loss=0.07166, over 5484966.99 frames. ], batch size: 213, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:13:37,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:13:55,450 INFO [train.py:968] (0/2) Epoch 16, batch 800, giga_loss[loss=0.2205, simple_loss=0.286, pruned_loss=0.07747, over 28886.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2921, pruned_loss=0.07211, over 5580594.44 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3446, pruned_loss=0.09099, over 2107056.24 frames. ], giga_tot_loss[loss=0.2139, simple_loss=0.2872, pruned_loss=0.07029, over 5535554.85 frames. ], batch size: 186, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:14:18,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.986e+02 1.084e+03 1.382e+03 2.285e+03 8.607e+03, threshold=2.765e+03, percent-clipped=21.0
+2023-03-08 03:14:18,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4539, 1.7756, 1.7739, 1.3107], device='cuda:0'), covar=tensor([0.1719, 0.2427, 0.1380, 0.1600], device='cuda:0'), in_proj_covar=tensor([0.0876, 0.0701, 0.0922, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 03:14:41,536 INFO [train.py:968] (0/2) Epoch 16, batch 850, giga_loss[loss=0.2453, simple_loss=0.322, pruned_loss=0.08429, over 28654.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2976, pruned_loss=0.07549, over 5595649.15 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3456, pruned_loss=0.09161, over 2219748.07 frames. ], giga_tot_loss[loss=0.2193, simple_loss=0.2919, pruned_loss=0.07333, over 5550974.42 frames. ], batch size: 92, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:14:52,752 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:15:06,713 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684822.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 03:15:08,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684825.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 03:15:32,832 INFO [train.py:968] (0/2) Epoch 16, batch 900, giga_loss[loss=0.2397, simple_loss=0.3249, pruned_loss=0.0772, over 28823.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3102, pruned_loss=0.08193, over 5612869.04 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.346, pruned_loss=0.09209, over 2297955.54 frames. ], giga_tot_loss[loss=0.2322, simple_loss=0.3047, pruned_loss=0.07979, over 5582066.11 frames. ], batch size: 199, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:15:36,254 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684854.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 03:15:53,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:15:56,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:15:58,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.845e+02 1.258e+03 1.668e+03 2.389e+03 5.316e+03, threshold=3.336e+03, percent-clipped=17.0
+2023-03-08 03:15:58,578 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:16:11,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3553, 3.2416, 1.5444, 1.4883], device='cuda:0'), covar=tensor([0.1030, 0.0302, 0.0928, 0.1427], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0527, 0.0357, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0024, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 03:16:16,871 INFO [train.py:968] (0/2) Epoch 16, batch 950, giga_loss[loss=0.2882, simple_loss=0.3712, pruned_loss=0.1026, over 29026.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3225, pruned_loss=0.08792, over 5635021.16 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3451, pruned_loss=0.09132, over 2419944.27 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3175, pruned_loss=0.08632, over 5603354.61 frames. ], batch size: 136, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:16:20,037 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:16:49,152 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 03:16:56,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4830, 1.8952, 1.4871, 1.5966], device='cuda:0'), covar=tensor([0.2551, 0.2460, 0.2896, 0.2266], device='cuda:0'), in_proj_covar=tensor([0.1403, 0.1026, 0.1242, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 03:16:58,889 INFO [train.py:968] (0/2) Epoch 16, batch 1000, giga_loss[loss=0.2607, simple_loss=0.3456, pruned_loss=0.08789, over 28660.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3317, pruned_loss=0.09226, over 5651804.19 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3446, pruned_loss=0.09116, over 2488189.23 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3277, pruned_loss=0.09106, over 5622918.59 frames. ], batch size: 92, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:17:24,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.748e+02 1.185e+03 1.483e+03 1.762e+03 5.025e+03, threshold=2.966e+03, percent-clipped=3.0
+2023-03-08 03:17:35,584 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:17:43,264 INFO [train.py:968] (0/2) Epoch 16, batch 1050, giga_loss[loss=0.2693, simple_loss=0.3509, pruned_loss=0.09386, over 28799.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3364, pruned_loss=0.0933, over 5665037.31 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3439, pruned_loss=0.09083, over 2505234.28 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3335, pruned_loss=0.09251, over 5641366.62 frames. ], batch size: 285, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:18:30,735 INFO [train.py:968] (0/2) Epoch 16, batch 1100, giga_loss[loss=0.2599, simple_loss=0.3465, pruned_loss=0.08669, over 28540.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3388, pruned_loss=0.09351, over 5657371.45 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3428, pruned_loss=0.09037, over 2579275.70 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3368, pruned_loss=0.09315, over 5641401.46 frames. ], batch size: 307, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:18:56,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.255e+02 1.064e+03 1.315e+03 1.793e+03 3.603e+03, threshold=2.630e+03, percent-clipped=4.0
+2023-03-08 03:19:17,068 INFO [train.py:968] (0/2) Epoch 16, batch 1150, giga_loss[loss=0.2684, simple_loss=0.3481, pruned_loss=0.09431, over 28561.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3404, pruned_loss=0.09416, over 5675037.36 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3426, pruned_loss=0.09023, over 2595917.32 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.339, pruned_loss=0.09396, over 5661427.20 frames. ], batch size: 307, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:19:20,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6679, 1.6875, 1.6947, 1.5633], device='cuda:0'), covar=tensor([0.2279, 0.2354, 0.1860, 0.2114], device='cuda:0'), in_proj_covar=tensor([0.1836, 0.1762, 0.1699, 0.1845], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 03:20:03,381 INFO [train.py:968] (0/2) Epoch 16, batch 1200, giga_loss[loss=0.2439, simple_loss=0.3229, pruned_loss=0.08251, over 28510.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3436, pruned_loss=0.09681, over 5673380.85 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3426, pruned_loss=0.09012, over 2677749.49 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3425, pruned_loss=0.09682, over 5657526.08 frames. ], batch size: 60, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:20:27,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.113e+02 1.186e+03 1.485e+03 1.992e+03 5.927e+03, threshold=2.969e+03, percent-clipped=11.0
+2023-03-08 03:20:32,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=685184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:20:46,751 INFO [train.py:968] (0/2) Epoch 16, batch 1250, libri_loss[loss=0.2218, simple_loss=0.3052, pruned_loss=0.06918, over 29348.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3461, pruned_loss=0.09832, over 5681937.60 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3423, pruned_loss=0.08997, over 2788674.04 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3454, pruned_loss=0.09867, over 5662180.91 frames. ], batch size: 71, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:21:30,026 INFO [train.py:968] (0/2) Epoch 16, batch 1300, giga_loss[loss=0.2721, simple_loss=0.356, pruned_loss=0.09412, over 28961.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3497, pruned_loss=0.09987, over 5672145.78 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.343, pruned_loss=0.09035, over 2856223.12 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3489, pruned_loss=0.1002, over 5660504.00 frames. ], batch size: 186, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:21:34,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=685254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:21:52,760 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.234e+02 1.136e+03 1.367e+03 1.667e+03 3.897e+03, threshold=2.734e+03, percent-clipped=4.0
+2023-03-08 03:22:00,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1547, 1.1174, 3.6021, 2.9880], device='cuda:0'), covar=tensor([0.1750, 0.2994, 0.0464, 0.1109], device='cuda:0'), in_proj_covar=tensor([0.0701, 0.0612, 0.0900, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 03:22:10,127 INFO [train.py:968] (0/2) Epoch 16, batch 1350, giga_loss[loss=0.2746, simple_loss=0.357, pruned_loss=0.09612, over 28935.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.351, pruned_loss=0.0991, over 5691142.96 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3428, pruned_loss=0.09001, over 2971759.67 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3507, pruned_loss=0.09983, over 5677485.47 frames. ], batch size: 186, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:22:24,874 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6664, 3.4556, 3.2903, 1.6900], device='cuda:0'), covar=tensor([0.0658, 0.0840, 0.0756, 0.2390], device='cuda:0'), in_proj_covar=tensor([0.1123, 0.1033, 0.0886, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 03:22:33,528 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=685327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:22:35,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=685330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:22:54,351 INFO [train.py:968] (0/2) Epoch 16, batch 1400, giga_loss[loss=0.2763, simple_loss=0.3561, pruned_loss=0.09824, over 28244.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3527, pruned_loss=0.09974, over 5689745.71 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3424, pruned_loss=0.08969, over 3029648.37 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3529, pruned_loss=0.1006, over 5675458.51 frames. ], batch size: 368, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:23:01,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=685359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:23:08,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=685367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:23:17,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.166e+02 1.173e+03 1.485e+03 1.832e+03 3.429e+03, threshold=2.971e+03, percent-clipped=2.0
+2023-03-08 03:23:34,383 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=685397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:23:36,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=685400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:23:36,962 INFO [train.py:968] (0/2) Epoch 16, batch 1450, giga_loss[loss=0.2809, simple_loss=0.3596, pruned_loss=0.1011, over 28882.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3526, pruned_loss=0.09865, over 5693383.61 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3423, pruned_loss=0.08961, over 3072653.91 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.353, pruned_loss=0.09954, over 5679170.87 frames. ], batch size: 199, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:24:00,323 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=685429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:24:12,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6214, 2.7236, 1.9082, 2.2458], device='cuda:0'), covar=tensor([0.0801, 0.0595, 0.0996, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0444, 0.0509, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 03:24:18,463 INFO [train.py:968] (0/2) Epoch 16, batch 1500, giga_loss[loss=0.2555, simple_loss=0.3479, pruned_loss=0.08154, over 29004.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3512, pruned_loss=0.09639, over 5703207.06 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3423, pruned_loss=0.0895, over 3114984.43 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3516, pruned_loss=0.09726, over 5689166.38 frames. ], batch size: 136, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:24:32,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6047, 2.4338, 2.2551, 2.1615], device='cuda:0'), covar=tensor([0.1475, 0.2012, 0.1922, 0.1939], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0735, 0.0687, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 03:24:43,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.514e+02 1.065e+03 1.293e+03 1.649e+03 5.658e+03, threshold=2.586e+03, percent-clipped=3.0
+2023-03-08 03:24:50,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5423, 2.2578, 1.6424, 0.7105], device='cuda:0'), covar=tensor([0.5493, 0.2556, 0.4004, 0.6013], device='cuda:0'), in_proj_covar=tensor([0.1636, 0.1557, 0.1536, 0.1338], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 03:25:01,470 INFO [train.py:968] (0/2) Epoch 16, batch 1550, giga_loss[loss=0.2551, simple_loss=0.3371, pruned_loss=0.08661, over 28378.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3493, pruned_loss=0.09468, over 5704081.22 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3422, pruned_loss=0.08936, over 3156669.66 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3498, pruned_loss=0.09551, over 5690379.75 frames. ], batch size: 77, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:25:09,369 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=685510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:25:12,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=685513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:25:37,972 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=685542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:25:47,022 INFO [train.py:968] (0/2) Epoch 16, batch 1600, giga_loss[loss=0.2622, simple_loss=0.3393, pruned_loss=0.09249, over 28796.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3497, pruned_loss=0.09588, over 5713725.31 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3422, pruned_loss=0.08919, over 3224760.36 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3503, pruned_loss=0.09676, over 5698589.90 frames. ], batch size: 112, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:25:49,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4102, 2.1001, 1.5062, 0.5794], device='cuda:0'), covar=tensor([0.4541, 0.2376, 0.3691, 0.5321], device='cuda:0'), in_proj_covar=tensor([0.1641, 0.1561, 0.1541, 0.1340], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 03:26:11,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.855e+02 1.277e+03 1.649e+03 2.290e+03 7.748e+03, threshold=3.298e+03, percent-clipped=18.0
+2023-03-08 03:26:16,591 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-08 03:26:20,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9701, 1.2378, 1.3708, 1.0449], device='cuda:0'), covar=tensor([0.1649, 0.1180, 0.1959, 0.1511], device='cuda:0'), in_proj_covar=tensor([0.0452, 0.0730, 0.0682, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 03:26:30,875 INFO [train.py:968] (0/2) Epoch 16, batch 1650, giga_loss[loss=0.3525, simple_loss=0.4011, pruned_loss=0.1519, over 28230.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3542, pruned_loss=0.1019, over 5708538.41 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08994, over 3267211.46 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3542, pruned_loss=0.1024, over 5703052.07 frames. ], batch size: 77, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:27:19,640 INFO [train.py:968] (0/2) Epoch 16, batch 1700, giga_loss[loss=0.2634, simple_loss=0.3397, pruned_loss=0.0936, over 28871.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3554, pruned_loss=0.1046, over 5695195.91 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.343, pruned_loss=0.08975, over 3302939.04 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3557, pruned_loss=0.1053, over 5691378.12 frames. ], batch size: 119, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:27:19,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3893, 3.3236, 1.6257, 1.5020], device='cuda:0'), covar=tensor([0.0958, 0.0271, 0.0810, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0526, 0.0356, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 03:27:46,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.941e+02 1.380e+03 1.683e+03 2.492e+03 6.845e+03, threshold=3.366e+03, percent-clipped=8.0
+2023-03-08 03:27:52,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6695, 1.4539, 5.0702, 3.6505], device='cuda:0'), covar=tensor([0.1715, 0.2706, 0.0346, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0702, 0.0612, 0.0902, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 03:28:04,732 INFO [train.py:968] (0/2) Epoch 16, batch 1750, giga_loss[loss=0.2994, simple_loss=0.368, pruned_loss=0.1154, over 29042.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.354, pruned_loss=0.1048, over 5690776.32 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3424, pruned_loss=0.08961, over 3371132.24 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3549, pruned_loss=0.1058, over 5690626.99 frames. ], batch size: 128, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:28:13,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8313, 2.0220, 1.7350, 2.1533], device='cuda:0'), covar=tensor([0.2320, 0.2412, 0.2622, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.1402, 0.1026, 0.1239, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 03:28:46,584 INFO [train.py:968] (0/2) Epoch 16, batch 1800, giga_loss[loss=0.2721, simple_loss=0.3454, pruned_loss=0.0994, over 28610.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3522, pruned_loss=0.1043, over 5704289.30 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3419, pruned_loss=0.08936, over 3420641.96 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3533, pruned_loss=0.1055, over 5701180.65 frames. ], batch size: 78, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:29:10,285 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.414e+02 1.322e+03 1.602e+03 2.184e+03 9.802e+03, threshold=3.205e+03, percent-clipped=7.0
+2023-03-08 03:29:30,742 INFO [train.py:968] (0/2) Epoch 16, batch 1850, giga_loss[loss=0.302, simple_loss=0.377, pruned_loss=0.1134, over 28330.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3511, pruned_loss=0.1031, over 5708871.26 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3419, pruned_loss=0.08935, over 3470406.68 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3523, pruned_loss=0.1043, over 5702451.43 frames. ], batch size: 368, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:29:47,389 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-08 03:30:18,190 INFO [train.py:968] (0/2) Epoch 16, batch 1900, giga_loss[loss=0.2386, simple_loss=0.3188, pruned_loss=0.07924, over 28886.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3488, pruned_loss=0.1009, over 5705034.26 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3423, pruned_loss=0.08954, over 3506250.71 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3496, pruned_loss=0.1019, over 5697829.57 frames. ], batch size: 112, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:30:49,060 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=685879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:30:49,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.370e+02 1.087e+03 1.315e+03 1.899e+03 4.865e+03, threshold=2.629e+03, percent-clipped=2.0
+2023-03-08 03:30:58,574 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=685889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:30:58,831 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 03:31:08,331 INFO [train.py:968] (0/2) Epoch 16, batch 1950, giga_loss[loss=0.2467, simple_loss=0.3185, pruned_loss=0.08743, over 28614.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3447, pruned_loss=0.0982, over 5699626.69 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3419, pruned_loss=0.08929, over 3540785.34 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3455, pruned_loss=0.0993, over 5692146.78 frames. ], batch size: 336, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:31:21,809 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-08 03:31:57,744 INFO [train.py:968] (0/2) Epoch 16, batch 2000, giga_loss[loss=0.2584, simple_loss=0.3286, pruned_loss=0.09408, over 27781.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3388, pruned_loss=0.09451, over 5693134.11 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3418, pruned_loss=0.0889, over 3632626.99 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3395, pruned_loss=0.0959, over 5680216.13 frames. ], batch size: 474, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:32:07,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=685960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:32:21,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3437, 3.0770, 1.5363, 1.4616], device='cuda:0'), covar=tensor([0.0951, 0.0267, 0.0869, 0.1336], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0524, 0.0356, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 03:32:23,703 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 03:32:25,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.176e+02 9.403e+02 1.169e+03 1.882e+03 6.155e+03, threshold=2.339e+03, percent-clipped=11.0
+2023-03-08 03:32:29,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4224, 1.5794, 1.5115, 1.3122], device='cuda:0'), covar=tensor([0.2397, 0.2141, 0.1701, 0.2227], device='cuda:0'), in_proj_covar=tensor([0.1829, 0.1746, 0.1688, 0.1835], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 03:32:44,866 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-686000.pt
+2023-03-08 03:32:45,865 INFO [train.py:968] (0/2) Epoch 16, batch 2050, giga_loss[loss=0.2337, simple_loss=0.3096, pruned_loss=0.07887, over 29037.00 frames. ], tot_loss[loss=0.258, simple_loss=0.333, pruned_loss=0.09154, over 5687126.57 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.342, pruned_loss=0.08901, over 3677440.17 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3334, pruned_loss=0.09263, over 5673358.00 frames. ], batch size: 128, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:33:39,496 INFO [train.py:968] (0/2) Epoch 16, batch 2100, giga_loss[loss=0.2399, simple_loss=0.3157, pruned_loss=0.08205, over 28775.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3304, pruned_loss=0.08979, over 5691188.56 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3421, pruned_loss=0.08902, over 3710705.13 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3304, pruned_loss=0.09065, over 5677435.65 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:34:03,867 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.751e+02 9.839e+02 1.139e+03 1.448e+03 3.698e+03, threshold=2.278e+03, percent-clipped=4.0
+2023-03-08 03:34:21,717 INFO [train.py:968] (0/2) Epoch 16, batch 2150, giga_loss[loss=0.2298, simple_loss=0.306, pruned_loss=0.07679, over 28769.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3311, pruned_loss=0.08955, over 5699908.22 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3415, pruned_loss=0.08855, over 3754559.86 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3312, pruned_loss=0.09052, over 5685028.24 frames. ], batch size: 92, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:34:44,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3337, 1.5134, 1.4045, 1.5099], device='cuda:0'), covar=tensor([0.0817, 0.0348, 0.0331, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 03:35:03,732 INFO [train.py:968] (0/2) Epoch 16, batch 2200, giga_loss[loss=0.2879, simple_loss=0.3401, pruned_loss=0.1178, over 23845.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3314, pruned_loss=0.08974, over 5701630.95 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3422, pruned_loss=0.08876, over 3797210.12 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3309, pruned_loss=0.0904, over 5686114.04 frames. ], batch size: 705, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:35:28,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6968, 1.8136, 1.5467, 1.6468], device='cuda:0'), covar=tensor([0.2377, 0.2493, 0.2790, 0.2279], device='cuda:0'), in_proj_covar=tensor([0.1396, 0.1021, 0.1233, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 03:35:29,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.993e+02 1.013e+03 1.201e+03 1.545e+03 4.540e+03, threshold=2.402e+03, percent-clipped=7.0
+2023-03-08 03:35:41,461 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-08 03:35:45,285 INFO [train.py:968] (0/2) Epoch 16, batch 2250, giga_loss[loss=0.2839, simple_loss=0.3488, pruned_loss=0.1095, over 28645.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3315, pruned_loss=0.08982, over 5713170.33 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3434, pruned_loss=0.08925, over 3879222.58 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3299, pruned_loss=0.09009, over 5694300.58 frames. ], batch size: 336, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:36:06,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9695, 1.2173, 1.3088, 1.0483], device='cuda:0'), covar=tensor([0.1892, 0.1483, 0.2442, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0737, 0.0689, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 03:36:32,669 INFO [train.py:968] (0/2) Epoch 16, batch 2300, giga_loss[loss=0.2252, simple_loss=0.3031, pruned_loss=0.07366, over 28879.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3307, pruned_loss=0.09047, over 5712064.82 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3445, pruned_loss=0.09005, over 3899723.68 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3288, pruned_loss=0.09021, over 5695352.49 frames. ], batch size: 66, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:36:34,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:36:42,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:36:56,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.091e+02 1.009e+03 1.403e+03 2.353e+03 1.234e+04, threshold=2.806e+03, percent-clipped=22.0
+2023-03-08 03:37:12,642 INFO [train.py:968] (0/2) Epoch 16, batch 2350, giga_loss[loss=0.2306, simple_loss=0.3, pruned_loss=0.08065, over 28509.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3285, pruned_loss=0.08904, over 5725467.02 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3446, pruned_loss=0.08973, over 3959388.82 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3265, pruned_loss=0.08902, over 5706765.10 frames. ], batch size: 85, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:37:41,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:37:54,452 INFO [train.py:968] (0/2) Epoch 16, batch 2400, giga_loss[loss=0.1902, simple_loss=0.2721, pruned_loss=0.05413, over 28468.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3254, pruned_loss=0.08767, over 5729942.39 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.344, pruned_loss=0.08943, over 3997772.76 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3238, pruned_loss=0.08782, over 5712190.13 frames. ], batch size: 65, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:38:08,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=686369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:18,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8642, 1.0205, 0.9521, 0.8277], device='cuda:0'), covar=tensor([0.1824, 0.1866, 0.1313, 0.1775], device='cuda:0'), in_proj_covar=tensor([0.1821, 0.1738, 0.1684, 0.1828], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 03:38:18,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.719e+02 9.792e+02 1.154e+03 1.495e+03 4.351e+03, threshold=2.309e+03, percent-clipped=7.0
+2023-03-08 03:38:27,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4436, 1.3350, 4.7297, 3.4049], device='cuda:0'), covar=tensor([0.1778, 0.2835, 0.0351, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0702, 0.0612, 0.0896, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 03:38:30,483 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-08 03:38:30,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=686397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:33,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=686400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:33,624 INFO [train.py:968] (0/2) Epoch 16, batch 2450, giga_loss[loss=0.2047, simple_loss=0.281, pruned_loss=0.0642, over 28484.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3244, pruned_loss=0.08736, over 5729923.46 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3449, pruned_loss=0.08963, over 4038098.69 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3221, pruned_loss=0.0873, over 5718267.04 frames. ], batch size: 60, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:38:37,868 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=686407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:40,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=686410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:53,877 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=686429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:55,216 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=686431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:39:03,210 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=686439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:39:13,702 INFO [train.py:968] (0/2) Epoch 16, batch 2500, libri_loss[loss=0.2515, simple_loss=0.3361, pruned_loss=0.08344, over 29549.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3216, pruned_loss=0.08575, over 5732843.48 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3448, pruned_loss=0.08934, over 4084452.11 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3193, pruned_loss=0.08581, over 5719617.54 frames. ], batch size: 79, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:39:36,176 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=686478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:39:38,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=686481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:39:40,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.022e+02 1.117e+03 1.351e+03 1.772e+03 4.435e+03, threshold=2.702e+03, percent-clipped=11.0
+2023-03-08 03:39:54,082 INFO [train.py:968] (0/2) Epoch 16, batch 2550, giga_loss[loss=0.2176, simple_loss=0.2938, pruned_loss=0.07073, over 28884.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3195, pruned_loss=0.08491, over 5721095.35 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3451, pruned_loss=0.08964, over 4119893.70 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.317, pruned_loss=0.08467, over 5715161.76 frames. ], batch size: 112, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 03:40:02,841 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=686510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:40:17,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6137, 2.3609, 1.7036, 0.7818], device='cuda:0'), covar=tensor([0.4853, 0.2668, 0.3574, 0.5334], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1554, 0.1536, 0.1336], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 03:40:21,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4066, 1.5325, 1.5586, 1.3752], device='cuda:0'), covar=tensor([0.1729, 0.1964, 0.2301, 0.2018], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0734, 0.0689, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 03:40:36,716 INFO [train.py:968] (0/2) Epoch 16, batch 2600, giga_loss[loss=0.2417, simple_loss=0.3234, pruned_loss=0.07993, over 29036.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3181, pruned_loss=0.08372, over 5721780.00 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3458, pruned_loss=0.08976, over 4153156.78 frames. ], giga_tot_loss[loss=0.2409, simple_loss=0.3151, pruned_loss=0.08335, over 5715764.57 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 03:40:49,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2268, 1.0843, 4.4629, 3.4180], device='cuda:0'), covar=tensor([0.1902, 0.3092, 0.0365, 0.0988], device='cuda:0'), in_proj_covar=tensor([0.0704, 0.0614, 0.0900, 0.0826], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 03:41:01,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.811e+02 1.016e+03 1.241e+03 1.777e+03 4.905e+03, threshold=2.481e+03, percent-clipped=10.0
+2023-03-08 03:41:16,440 INFO [train.py:968] (0/2) Epoch 16, batch 2650, giga_loss[loss=0.2263, simple_loss=0.3034, pruned_loss=0.07456, over 28683.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3177, pruned_loss=0.08334, over 5725474.26 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3466, pruned_loss=0.08999, over 4208704.75 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3138, pruned_loss=0.08266, over 5719102.24 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 03:41:28,258 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=686616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:41:36,718 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=686624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:42:02,469 INFO [train.py:968] (0/2) Epoch 16, batch 2700, giga_loss[loss=0.2839, simple_loss=0.3523, pruned_loss=0.1078, over 28966.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.32, pruned_loss=0.08512, over 5715610.69 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3468, pruned_loss=0.09004, over 4225225.55 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3165, pruned_loss=0.08449, over 5709249.44 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 03:42:28,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.534e+02 1.010e+03 1.260e+03 1.562e+03 5.361e+03, threshold=2.521e+03, percent-clipped=6.0
+2023-03-08 03:42:36,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2585, 1.4633, 1.5593, 1.3504], device='cuda:0'), covar=tensor([0.1788, 0.1647, 0.2145, 0.1694], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0734, 0.0689, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 03:42:46,758 INFO [train.py:968] (0/2) Epoch 16, batch 2750, giga_loss[loss=0.2955, simple_loss=0.3626, pruned_loss=0.1141, over 28892.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3256, pruned_loss=0.08846, over 5711514.00 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.347, pruned_loss=0.09004, over 4258294.52 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3222, pruned_loss=0.08788, over 5710685.30 frames. ], batch size: 112, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 03:43:28,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:43:35,178 INFO [train.py:968] (0/2) Epoch 16, batch 2800, giga_loss[loss=0.2826, simple_loss=0.3583, pruned_loss=0.1034, over 29045.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3336, pruned_loss=0.0938, over 5696916.82 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.347, pruned_loss=0.09021, over 4297197.52 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3303, pruned_loss=0.09325, over 5699055.67 frames. ], batch size: 155, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:43:41,207 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=686758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:43:55,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5428, 1.7546, 1.4262, 1.4529], device='cuda:0'), covar=tensor([0.2447, 0.2406, 0.2688, 0.2195], device='cuda:0'), in_proj_covar=tensor([0.1402, 0.1026, 0.1240, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 03:44:04,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.995e+02 1.362e+03 1.647e+03 2.287e+03 1.013e+04, threshold=3.294e+03, percent-clipped=19.0
+2023-03-08 03:44:19,569 INFO [train.py:968] (0/2) Epoch 16, batch 2850, giga_loss[loss=0.2902, simple_loss=0.3643, pruned_loss=0.1081, over 27817.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3407, pruned_loss=0.09843, over 5686135.24 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3473, pruned_loss=0.09062, over 4339460.77 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3377, pruned_loss=0.09786, over 5687528.33 frames. ], batch size: 412, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:44:25,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:45:09,692 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-08 03:45:14,418 INFO [train.py:968] (0/2) Epoch 16, batch 2900, giga_loss[loss=0.2879, simple_loss=0.3582, pruned_loss=0.1088, over 28677.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3467, pruned_loss=0.1016, over 5663844.01 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3475, pruned_loss=0.09079, over 4353786.16 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3441, pruned_loss=0.1011, over 5664170.97 frames. ], batch size: 92, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:45:39,429 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0752, 4.8729, 4.6167, 2.2675], device='cuda:0'), covar=tensor([0.0438, 0.0627, 0.0615, 0.2025], device='cuda:0'), in_proj_covar=tensor([0.1125, 0.1038, 0.0894, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 03:45:43,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.049e+02 1.126e+03 1.303e+03 1.722e+03 3.741e+03, threshold=2.606e+03, percent-clipped=1.0
+2023-03-08 03:45:47,040 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=686887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:45:48,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=686890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:45:58,580 INFO [train.py:968] (0/2) Epoch 16, batch 2950, giga_loss[loss=0.3611, simple_loss=0.4214, pruned_loss=0.1504, over 28727.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3513, pruned_loss=0.1029, over 5677434.77 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3473, pruned_loss=0.09065, over 4382569.19 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3494, pruned_loss=0.1027, over 5675425.99 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:46:20,245 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=686919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:46:46,855 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=686949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:46:48,726 INFO [train.py:968] (0/2) Epoch 16, batch 3000, giga_loss[loss=0.324, simple_loss=0.3849, pruned_loss=0.1316, over 27908.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3573, pruned_loss=0.1067, over 5668939.32 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3476, pruned_loss=0.09078, over 4417112.89 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3558, pruned_loss=0.1069, over 5670646.85 frames. ], batch size: 412, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:46:48,730 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 03:46:58,141 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2188, simple_loss=0.3232, pruned_loss=0.05722, over 944034.00 frames. 
+2023-03-08 03:46:58,141 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 03:46:59,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=686952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:47:24,195 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=686981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:47:26,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.392e+02 1.213e+03 1.641e+03 2.280e+03 8.003e+03, threshold=3.283e+03, percent-clipped=15.0
+2023-03-08 03:47:33,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:47:41,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:47:43,072 INFO [train.py:968] (0/2) Epoch 16, batch 3050, libri_loss[loss=0.2774, simple_loss=0.3608, pruned_loss=0.09704, over 29537.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3528, pruned_loss=0.1033, over 5675982.87 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3476, pruned_loss=0.09085, over 4432021.29 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3516, pruned_loss=0.1035, over 5675132.61 frames. ], batch size: 84, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:48:26,001 INFO [train.py:968] (0/2) Epoch 16, batch 3100, giga_loss[loss=0.2552, simple_loss=0.3404, pruned_loss=0.08498, over 28887.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3494, pruned_loss=0.1006, over 5676402.68 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3475, pruned_loss=0.09089, over 4450339.04 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3486, pruned_loss=0.1009, over 5675998.19 frames. ], batch size: 106, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:48:55,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.263e+02 1.144e+03 1.487e+03 1.989e+03 3.848e+03, threshold=2.974e+03, percent-clipped=3.0
+2023-03-08 03:49:11,349 INFO [train.py:968] (0/2) Epoch 16, batch 3150, giga_loss[loss=0.2559, simple_loss=0.3308, pruned_loss=0.09056, over 28968.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3489, pruned_loss=0.1, over 5676251.72 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.348, pruned_loss=0.09128, over 4478524.45 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3479, pruned_loss=0.1002, over 5672367.63 frames. ], batch size: 106, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:49:41,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=687133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:49:42,402 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=687134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:49:44,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=687137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:49:47,944 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=687142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:49:50,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=687145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:49:56,284 INFO [train.py:968] (0/2) Epoch 16, batch 3200, giga_loss[loss=0.2584, simple_loss=0.343, pruned_loss=0.08689, over 28901.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3501, pruned_loss=0.1006, over 5679890.66 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3475, pruned_loss=0.09101, over 4500092.32 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3497, pruned_loss=0.1011, over 5673540.38 frames. ], batch size: 227, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:50:10,641 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=687166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:50:18,649 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=687174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:50:26,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.210e+02 1.184e+03 1.533e+03 2.039e+03 1.090e+04, threshold=3.066e+03, percent-clipped=7.0
+2023-03-08 03:50:41,408 INFO [train.py:968] (0/2) Epoch 16, batch 3250, giga_loss[loss=0.2915, simple_loss=0.3626, pruned_loss=0.1102, over 28986.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3515, pruned_loss=0.1014, over 5685570.64 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3476, pruned_loss=0.0912, over 4520936.38 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3512, pruned_loss=0.1018, over 5677449.15 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:51:11,236 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-08 03:51:26,416 INFO [train.py:968] (0/2) Epoch 16, batch 3300, giga_loss[loss=0.3036, simple_loss=0.3716, pruned_loss=0.1178, over 28615.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.354, pruned_loss=0.1036, over 5679401.17 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3478, pruned_loss=0.09146, over 4534251.03 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3537, pruned_loss=0.1039, over 5681205.28 frames. ], batch size: 85, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:51:47,243 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=687276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:51:49,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=687279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:51:53,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.217e+02 1.245e+03 1.522e+03 2.186e+03 4.217e+03, threshold=3.045e+03, percent-clipped=6.0
+2023-03-08 03:52:07,893 INFO [train.py:968] (0/2) Epoch 16, batch 3350, giga_loss[loss=0.2913, simple_loss=0.3654, pruned_loss=0.1086, over 29020.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3547, pruned_loss=0.1041, over 5688311.36 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3473, pruned_loss=0.09097, over 4573573.67 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.355, pruned_loss=0.1051, over 5684860.66 frames. ], batch size: 128, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:52:13,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4097, 1.8298, 1.2552, 0.8279], device='cuda:0'), covar=tensor([0.4689, 0.2534, 0.2475, 0.4473], device='cuda:0'), in_proj_covar=tensor([0.1635, 0.1556, 0.1529, 0.1337], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 03:52:14,131 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=687308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:52:42,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4695, 4.5383, 1.7281, 1.7509], device='cuda:0'), covar=tensor([0.1014, 0.0237, 0.0815, 0.1314], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0526, 0.0357, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 03:52:53,628 INFO [train.py:968] (0/2) Epoch 16, batch 3400, giga_loss[loss=0.3792, simple_loss=0.4161, pruned_loss=0.1712, over 26608.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3563, pruned_loss=0.1062, over 5681160.94 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3471, pruned_loss=0.09094, over 4606302.85 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3569, pruned_loss=0.1073, over 5678276.86 frames. ], batch size: 555, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:53:21,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.393e+02 1.181e+03 1.488e+03 2.116e+03 6.380e+03, threshold=2.977e+03, percent-clipped=7.0
+2023-03-08 03:53:37,761 INFO [train.py:968] (0/2) Epoch 16, batch 3450, giga_loss[loss=0.2716, simple_loss=0.3486, pruned_loss=0.09728, over 28877.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3562, pruned_loss=0.106, over 5678964.71 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3473, pruned_loss=0.0911, over 4625717.99 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3566, pruned_loss=0.107, over 5673582.47 frames. ], batch size: 112, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:53:42,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6877, 1.8515, 1.6614, 1.7129], device='cuda:0'), covar=tensor([0.1711, 0.2219, 0.2135, 0.1831], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0734, 0.0687, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 03:54:17,486 INFO [train.py:968] (0/2) Epoch 16, batch 3500, giga_loss[loss=0.3056, simple_loss=0.3752, pruned_loss=0.118, over 28903.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3558, pruned_loss=0.1045, over 5691048.33 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3475, pruned_loss=0.09126, over 4655846.73 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3562, pruned_loss=0.1055, over 5683103.61 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:54:45,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.616e+02 1.046e+03 1.303e+03 1.872e+03 4.442e+03, threshold=2.606e+03, percent-clipped=6.0
+2023-03-08 03:55:02,875 INFO [train.py:968] (0/2) Epoch 16, batch 3550, giga_loss[loss=0.2817, simple_loss=0.3609, pruned_loss=0.1012, over 28950.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3556, pruned_loss=0.1033, over 5691012.85 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3476, pruned_loss=0.09129, over 4665644.99 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3559, pruned_loss=0.1042, over 5685225.77 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:55:17,346 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=687517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:55:45,033 INFO [train.py:968] (0/2) Epoch 16, batch 3600, giga_loss[loss=0.2589, simple_loss=0.3293, pruned_loss=0.09429, over 28604.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3543, pruned_loss=0.1022, over 5697328.95 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3478, pruned_loss=0.09145, over 4678168.49 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3544, pruned_loss=0.1028, over 5690635.63 frames. ], batch size: 85, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:56:12,527 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.131e+02 1.106e+03 1.350e+03 1.750e+03 3.693e+03, threshold=2.699e+03, percent-clipped=7.0
+2023-03-08 03:56:27,913 INFO [train.py:968] (0/2) Epoch 16, batch 3650, giga_loss[loss=0.2476, simple_loss=0.3358, pruned_loss=0.07969, over 28474.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3521, pruned_loss=0.1011, over 5694107.07 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3475, pruned_loss=0.09109, over 4720050.51 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3526, pruned_loss=0.1022, over 5682361.42 frames. ], batch size: 60, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:56:51,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6531, 5.4545, 5.1591, 2.5909], device='cuda:0'), covar=tensor([0.0407, 0.0576, 0.0642, 0.1749], device='cuda:0'), in_proj_covar=tensor([0.1123, 0.1039, 0.0893, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 03:57:01,525 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=687642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:57:03,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=687645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:57:08,061 INFO [train.py:968] (0/2) Epoch 16, batch 3700, giga_loss[loss=0.2722, simple_loss=0.3482, pruned_loss=0.09807, over 28964.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3506, pruned_loss=0.1003, over 5692771.62 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.348, pruned_loss=0.09147, over 4734537.13 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3507, pruned_loss=0.1011, over 5695898.09 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:57:33,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.295e+02 1.137e+03 1.444e+03 2.120e+03 6.658e+03, threshold=2.888e+03, percent-clipped=11.0
+2023-03-08 03:57:34,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4129, 2.0631, 1.5618, 0.5939], device='cuda:0'), covar=tensor([0.4767, 0.2331, 0.3540, 0.5276], device='cuda:0'), in_proj_covar=tensor([0.1629, 0.1536, 0.1515, 0.1327], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 03:57:42,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5212, 1.5250, 1.2248, 1.1519], device='cuda:0'), covar=tensor([0.0771, 0.0511, 0.0967, 0.1054], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0439, 0.0505, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 03:57:47,174 INFO [train.py:968] (0/2) Epoch 16, batch 3750, giga_loss[loss=0.2414, simple_loss=0.3265, pruned_loss=0.07819, over 28736.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.349, pruned_loss=0.09974, over 5696352.06 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3476, pruned_loss=0.09126, over 4754139.77 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3495, pruned_loss=0.1007, over 5698574.22 frames. ], batch size: 66, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:58:13,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8416, 1.0684, 3.3715, 2.9688], device='cuda:0'), covar=tensor([0.1838, 0.2787, 0.0473, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0699, 0.0606, 0.0891, 0.0817], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 03:58:25,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2475, 1.4338, 1.2802, 1.4254], device='cuda:0'), covar=tensor([0.0759, 0.0387, 0.0333, 0.0813], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 03:58:33,164 INFO [train.py:968] (0/2) Epoch 16, batch 3800, giga_loss[loss=0.264, simple_loss=0.3476, pruned_loss=0.09016, over 28792.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3513, pruned_loss=0.1015, over 5687372.17 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3475, pruned_loss=0.09121, over 4771827.32 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3518, pruned_loss=0.1025, over 5694284.01 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:58:59,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.007e+02 1.109e+03 1.341e+03 1.868e+03 6.454e+03, threshold=2.682e+03, percent-clipped=8.0
+2023-03-08 03:59:12,885 INFO [train.py:968] (0/2) Epoch 16, batch 3850, giga_loss[loss=0.2868, simple_loss=0.3625, pruned_loss=0.1055, over 28951.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3516, pruned_loss=0.101, over 5694709.83 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3474, pruned_loss=0.09111, over 4805165.89 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3522, pruned_loss=0.1022, over 5694899.19 frames. ], batch size: 99, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:59:39,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3666, 1.5084, 3.8104, 3.2702], device='cuda:0'), covar=tensor([0.1607, 0.2546, 0.0409, 0.0789], device='cuda:0'), in_proj_covar=tensor([0.0699, 0.0606, 0.0890, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 03:59:56,709 INFO [train.py:968] (0/2) Epoch 16, batch 3900, giga_loss[loss=0.2812, simple_loss=0.362, pruned_loss=0.1002, over 28977.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3506, pruned_loss=0.09956, over 5703163.71 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3474, pruned_loss=0.09107, over 4810665.06 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3511, pruned_loss=0.1006, over 5702322.40 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:00:25,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.424e+02 1.011e+03 1.253e+03 1.732e+03 4.847e+03, threshold=2.505e+03, percent-clipped=5.0
+2023-03-08 04:00:30,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=687892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:00:35,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6245, 1.7627, 1.8637, 1.3952], device='cuda:0'), covar=tensor([0.1659, 0.2497, 0.1390, 0.1716], device='cuda:0'), in_proj_covar=tensor([0.0867, 0.0697, 0.0916, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 04:00:38,146 INFO [train.py:968] (0/2) Epoch 16, batch 3950, giga_loss[loss=0.2705, simple_loss=0.3501, pruned_loss=0.09546, over 28949.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3506, pruned_loss=0.09978, over 5706076.37 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3474, pruned_loss=0.0912, over 4843193.07 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3511, pruned_loss=0.1007, over 5699961.89 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:01:19,576 INFO [train.py:968] (0/2) Epoch 16, batch 4000, giga_loss[loss=0.3042, simple_loss=0.3683, pruned_loss=0.12, over 27896.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3499, pruned_loss=0.09973, over 5701640.03 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3474, pruned_loss=0.09131, over 4874002.46 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3503, pruned_loss=0.1007, over 5700027.94 frames. ], batch size: 412, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:01:48,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.896e+02 1.024e+03 1.400e+03 1.897e+03 8.665e+03, threshold=2.800e+03, percent-clipped=12.0
+2023-03-08 04:01:59,449 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-688000.pt
+2023-03-08 04:02:00,265 INFO [train.py:968] (0/2) Epoch 16, batch 4050, giga_loss[loss=0.282, simple_loss=0.352, pruned_loss=0.1061, over 28988.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3467, pruned_loss=0.09795, over 5698800.91 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.347, pruned_loss=0.09109, over 4889252.68 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3474, pruned_loss=0.09906, over 5702820.50 frames. ], batch size: 136, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:02:12,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=688017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:02:15,523 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=688020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:02:26,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6629, 4.4847, 4.2336, 2.0902], device='cuda:0'), covar=tensor([0.0455, 0.0601, 0.0620, 0.2192], device='cuda:0'), in_proj_covar=tensor([0.1129, 0.1039, 0.0894, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 04:02:27,555 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:02:29,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:02:38,471 INFO [train.py:968] (0/2) Epoch 16, batch 4100, giga_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1115, over 28989.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3434, pruned_loss=0.09617, over 5709663.68 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3468, pruned_loss=0.09112, over 4921308.31 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3441, pruned_loss=0.09725, over 5709281.44 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:02:54,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:03:07,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.580e+02 1.151e+03 1.408e+03 1.976e+03 5.415e+03, threshold=2.816e+03, percent-clipped=5.0
+2023-03-08 04:03:20,543 INFO [train.py:968] (0/2) Epoch 16, batch 4150, giga_loss[loss=0.2859, simple_loss=0.3638, pruned_loss=0.104, over 28919.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3417, pruned_loss=0.09559, over 5707363.68 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3467, pruned_loss=0.09126, over 4931235.31 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3422, pruned_loss=0.09638, over 5705232.01 frames. ], batch size: 227, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:03:48,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9920, 1.2770, 1.0716, 0.2871], device='cuda:0'), covar=tensor([0.3126, 0.2425, 0.3637, 0.5077], device='cuda:0'), in_proj_covar=tensor([0.1634, 0.1539, 0.1523, 0.1330], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 04:04:01,425 INFO [train.py:968] (0/2) Epoch 16, batch 4200, giga_loss[loss=0.2949, simple_loss=0.3523, pruned_loss=0.1188, over 23679.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3409, pruned_loss=0.09524, over 5700576.14 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3471, pruned_loss=0.09142, over 4949142.54 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3409, pruned_loss=0.09584, over 5703996.12 frames. ], batch size: 705, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:04:08,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:09,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1930, 1.2293, 1.1481, 0.8319], device='cuda:0'), covar=tensor([0.0816, 0.0546, 0.1059, 0.1023], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0441, 0.0506, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 04:04:10,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:10,756 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:13,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:28,199 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.125e+02 1.096e+03 1.306e+03 1.834e+03 3.687e+03, threshold=2.612e+03, percent-clipped=5.0
+2023-03-08 04:04:33,978 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:38,536 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:43,186 INFO [train.py:968] (0/2) Epoch 16, batch 4250, giga_loss[loss=0.236, simple_loss=0.319, pruned_loss=0.07647, over 28715.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3407, pruned_loss=0.09583, over 5709886.76 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3469, pruned_loss=0.09137, over 4981728.31 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3407, pruned_loss=0.09648, over 5706890.85 frames. ], batch size: 284, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:05:25,128 INFO [train.py:968] (0/2) Epoch 16, batch 4300, giga_loss[loss=0.2866, simple_loss=0.3573, pruned_loss=0.108, over 28751.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3385, pruned_loss=0.09528, over 5711940.92 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3467, pruned_loss=0.09129, over 4990841.82 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3386, pruned_loss=0.0959, over 5707932.97 frames. ], batch size: 119, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:05:45,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3856, 2.3675, 2.4170, 2.0493], device='cuda:0'), covar=tensor([0.1566, 0.2169, 0.1775, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0734, 0.0687, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 04:05:51,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1908, 1.3222, 1.2600, 1.1924], device='cuda:0'), covar=tensor([0.2016, 0.1663, 0.1331, 0.1635], device='cuda:0'), in_proj_covar=tensor([0.1820, 0.1741, 0.1685, 0.1826], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 04:05:53,263 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.867e+02 1.117e+03 1.352e+03 1.806e+03 1.084e+04, threshold=2.704e+03, percent-clipped=8.0
+2023-03-08 04:06:04,636 INFO [train.py:968] (0/2) Epoch 16, batch 4350, giga_loss[loss=0.2273, simple_loss=0.3081, pruned_loss=0.07319, over 28972.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3362, pruned_loss=0.09434, over 5715767.43 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3465, pruned_loss=0.09109, over 5013107.60 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3362, pruned_loss=0.09507, over 5708717.10 frames. ], batch size: 136, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:06:33,507 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=688335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:06:44,634 INFO [train.py:968] (0/2) Epoch 16, batch 4400, giga_loss[loss=0.2633, simple_loss=0.3468, pruned_loss=0.08995, over 28614.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3357, pruned_loss=0.09408, over 5712237.05 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3464, pruned_loss=0.09103, over 5034885.30 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3356, pruned_loss=0.09479, over 5702277.96 frames. ], batch size: 307, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:06:45,716 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=688352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:07:15,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.009e+02 1.065e+03 1.232e+03 1.643e+03 3.875e+03, threshold=2.465e+03, percent-clipped=4.0
+2023-03-08 04:07:16,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=688386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:07:21,610 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-08 04:07:29,392 INFO [train.py:968] (0/2) Epoch 16, batch 4450, giga_loss[loss=0.2552, simple_loss=0.3412, pruned_loss=0.08464, over 28943.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3383, pruned_loss=0.09524, over 5709210.21 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3468, pruned_loss=0.09144, over 5045859.92 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3377, pruned_loss=0.09553, over 5700409.36 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:08:14,884 INFO [train.py:968] (0/2) Epoch 16, batch 4500, libri_loss[loss=0.2697, simple_loss=0.3417, pruned_loss=0.0988, over 29592.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3402, pruned_loss=0.09564, over 5720013.68 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3469, pruned_loss=0.09149, over 5062716.50 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3396, pruned_loss=0.09591, over 5709641.93 frames. ], batch size: 75, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:08:44,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.134e+02 1.003e+03 1.158e+03 1.498e+03 5.456e+03, threshold=2.317e+03, percent-clipped=3.0
+2023-03-08 04:08:59,092 INFO [train.py:968] (0/2) Epoch 16, batch 4550, giga_loss[loss=0.2432, simple_loss=0.3225, pruned_loss=0.08194, over 28266.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3424, pruned_loss=0.09637, over 5720194.02 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3473, pruned_loss=0.09179, over 5070723.99 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3415, pruned_loss=0.09637, over 5710757.28 frames. ], batch size: 77, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:09:41,320 INFO [train.py:968] (0/2) Epoch 16, batch 4600, giga_loss[loss=0.2555, simple_loss=0.3396, pruned_loss=0.08574, over 28749.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3435, pruned_loss=0.09652, over 5714965.80 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3472, pruned_loss=0.09204, over 5104994.60 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3427, pruned_loss=0.09652, over 5701959.40 frames. ], batch size: 284, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:10:06,557 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9529, 3.0549, 2.0913, 0.9132], device='cuda:0'), covar=tensor([0.7085, 0.2560, 0.3425, 0.6637], device='cuda:0'), in_proj_covar=tensor([0.1627, 0.1535, 0.1526, 0.1329], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 04:10:14,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.319e+02 1.090e+03 1.384e+03 1.920e+03 4.032e+03, threshold=2.768e+03, percent-clipped=17.0
+2023-03-08 04:10:28,419 INFO [train.py:968] (0/2) Epoch 16, batch 4650, giga_loss[loss=0.2726, simple_loss=0.3474, pruned_loss=0.09894, over 28990.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3417, pruned_loss=0.09463, over 5708048.37 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3474, pruned_loss=0.09216, over 5121166.10 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3408, pruned_loss=0.09459, over 5694303.23 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:11:08,826 INFO [train.py:968] (0/2) Epoch 16, batch 4700, giga_loss[loss=0.3013, simple_loss=0.3552, pruned_loss=0.1237, over 28814.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3415, pruned_loss=0.0944, over 5710755.80 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3473, pruned_loss=0.09211, over 5136764.72 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3408, pruned_loss=0.09446, over 5696446.27 frames. ], batch size: 66, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:11:41,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.167e+02 1.253e+03 1.498e+03 1.858e+03 4.340e+03, threshold=2.996e+03, percent-clipped=6.0
+2023-03-08 04:11:46,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0712, 1.1860, 3.5407, 3.1804], device='cuda:0'), covar=tensor([0.1549, 0.2401, 0.0449, 0.0776], device='cuda:0'), in_proj_covar=tensor([0.0702, 0.0612, 0.0894, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 04:11:52,755 INFO [train.py:968] (0/2) Epoch 16, batch 4750, giga_loss[loss=0.2589, simple_loss=0.3339, pruned_loss=0.09197, over 28954.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3421, pruned_loss=0.09491, over 5719867.41 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.347, pruned_loss=0.09196, over 5148028.05 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3417, pruned_loss=0.09513, over 5706070.50 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:12:01,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=688710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:12:13,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=688727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:12:36,889 INFO [train.py:968] (0/2) Epoch 16, batch 4800, giga_loss[loss=0.2652, simple_loss=0.3389, pruned_loss=0.09574, over 28813.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3439, pruned_loss=0.09646, over 5716235.05 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3469, pruned_loss=0.09198, over 5166365.69 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3435, pruned_loss=0.0967, over 5701523.53 frames. ], batch size: 199, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:12:47,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=688761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:13:06,958 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.477e+02 1.258e+03 1.667e+03 2.058e+03 4.149e+03, threshold=3.334e+03, percent-clipped=10.0
+2023-03-08 04:13:08,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6664, 2.0558, 1.7717, 1.4009], device='cuda:0'), covar=tensor([0.2812, 0.2065, 0.2408, 0.2869], device='cuda:0'), in_proj_covar=tensor([0.1829, 0.1754, 0.1703, 0.1835], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 04:13:20,649 INFO [train.py:968] (0/2) Epoch 16, batch 4850, giga_loss[loss=0.2809, simple_loss=0.3584, pruned_loss=0.1017, over 28862.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.347, pruned_loss=0.09832, over 5719122.48 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3468, pruned_loss=0.09189, over 5181515.76 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3468, pruned_loss=0.09868, over 5703648.20 frames. ], batch size: 199, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:13:58,138 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4317, 1.6243, 1.3222, 1.5555], device='cuda:0'), covar=tensor([0.2418, 0.2481, 0.2767, 0.2307], device='cuda:0'), in_proj_covar=tensor([0.1396, 0.1021, 0.1233, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 04:14:00,713 INFO [train.py:968] (0/2) Epoch 16, batch 4900, giga_loss[loss=0.2613, simple_loss=0.3405, pruned_loss=0.09108, over 28727.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.349, pruned_loss=0.09873, over 5724009.25 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.347, pruned_loss=0.09191, over 5199068.71 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3487, pruned_loss=0.09916, over 5707885.73 frames. ], batch size: 119, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:14:02,211 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:04,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:16,296 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:18,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:28,268 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:28,639 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.853e+02 1.342e+03 1.784e+03 2.248e+03 7.156e+03, threshold=3.568e+03, percent-clipped=10.0
+2023-03-08 04:14:40,092 INFO [train.py:968] (0/2) Epoch 16, batch 4950, libri_loss[loss=0.2815, simple_loss=0.3595, pruned_loss=0.1017, over 29574.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3511, pruned_loss=0.1, over 5716589.86 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3472, pruned_loss=0.09195, over 5215877.74 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3509, pruned_loss=0.1006, over 5706772.40 frames. ], batch size: 75, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:14:41,518 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:42,884 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:45,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:15:10,607 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:15:24,199 INFO [train.py:968] (0/2) Epoch 16, batch 5000, giga_loss[loss=0.2917, simple_loss=0.3679, pruned_loss=0.1077, over 28927.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.353, pruned_loss=0.1014, over 5714476.18 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3478, pruned_loss=0.09215, over 5231264.68 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3523, pruned_loss=0.1019, over 5703918.65 frames. ], batch size: 227, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:15:46,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6366, 1.9682, 1.8620, 1.6811], device='cuda:0'), covar=tensor([0.1595, 0.1703, 0.1874, 0.1771], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0733, 0.0685, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 04:15:46,810 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-08 04:15:51,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9675, 1.1837, 3.4472, 2.9752], device='cuda:0'), covar=tensor([0.1685, 0.2519, 0.0498, 0.0971], device='cuda:0'), in_proj_covar=tensor([0.0706, 0.0614, 0.0902, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 04:15:52,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.510e+02 1.213e+03 1.498e+03 1.899e+03 3.785e+03, threshold=2.996e+03, percent-clipped=2.0
+2023-03-08 04:16:04,276 INFO [train.py:968] (0/2) Epoch 16, batch 5050, giga_loss[loss=0.2873, simple_loss=0.3746, pruned_loss=0.09995, over 29013.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3527, pruned_loss=0.1013, over 5711864.23 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3477, pruned_loss=0.09224, over 5245944.59 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3524, pruned_loss=0.1018, over 5700841.89 frames. ], batch size: 155, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:16:45,114 INFO [train.py:968] (0/2) Epoch 16, batch 5100, giga_loss[loss=0.281, simple_loss=0.3504, pruned_loss=0.1058, over 28912.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3515, pruned_loss=0.1008, over 5714993.12 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3474, pruned_loss=0.09211, over 5263679.19 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3517, pruned_loss=0.1015, over 5702821.51 frames. ], batch size: 112, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:17:16,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.322e+02 1.148e+03 1.433e+03 1.758e+03 5.338e+03, threshold=2.867e+03, percent-clipped=2.0
+2023-03-08 04:17:27,262 INFO [train.py:968] (0/2) Epoch 16, batch 5150, giga_loss[loss=0.2508, simple_loss=0.3301, pruned_loss=0.08572, over 28986.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3491, pruned_loss=0.09993, over 5713401.35 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3477, pruned_loss=0.09225, over 5279374.52 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.349, pruned_loss=0.1006, over 5699599.09 frames. ], batch size: 136, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:18:10,004 INFO [train.py:968] (0/2) Epoch 16, batch 5200, giga_loss[loss=0.275, simple_loss=0.3536, pruned_loss=0.09816, over 28696.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3456, pruned_loss=0.09809, over 5715450.19 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3481, pruned_loss=0.0925, over 5289107.88 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3452, pruned_loss=0.09856, over 5703497.71 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:18:39,717 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.822e+02 1.046e+03 1.397e+03 2.164e+03 7.882e+03, threshold=2.795e+03, percent-clipped=11.0
+2023-03-08 04:18:50,958 INFO [train.py:968] (0/2) Epoch 16, batch 5250, giga_loss[loss=0.2399, simple_loss=0.3198, pruned_loss=0.07998, over 28812.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3447, pruned_loss=0.09646, over 5711693.19 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3482, pruned_loss=0.09269, over 5300173.16 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3441, pruned_loss=0.09678, over 5705823.21 frames. ], batch size: 119, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:18:57,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-08 04:19:30,439 INFO [train.py:968] (0/2) Epoch 16, batch 5300, giga_loss[loss=0.2291, simple_loss=0.3222, pruned_loss=0.06797, over 28974.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3477, pruned_loss=0.09702, over 5717453.96 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3476, pruned_loss=0.09244, over 5317508.29 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3476, pruned_loss=0.09761, over 5709681.02 frames. ], batch size: 136, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:19:33,068 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:20:02,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.931e+02 1.112e+03 1.335e+03 1.899e+03 5.472e+03, threshold=2.671e+03, percent-clipped=7.0
+2023-03-08 04:20:14,884 INFO [train.py:968] (0/2) Epoch 16, batch 5350, giga_loss[loss=0.3288, simple_loss=0.3938, pruned_loss=0.1319, over 28686.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3492, pruned_loss=0.09805, over 5717802.34 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3476, pruned_loss=0.0925, over 5324100.32 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3492, pruned_loss=0.09856, over 5712639.14 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:20:56,173 INFO [train.py:968] (0/2) Epoch 16, batch 5400, giga_loss[loss=0.262, simple_loss=0.3375, pruned_loss=0.09326, over 28878.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3471, pruned_loss=0.0982, over 5712389.22 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3479, pruned_loss=0.09262, over 5331347.60 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3469, pruned_loss=0.09865, over 5712973.34 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:21:03,034 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-08 04:21:30,761 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.874e+02 1.219e+03 1.580e+03 2.389e+03 7.797e+03, threshold=3.161e+03, percent-clipped=16.0
+2023-03-08 04:21:42,362 INFO [train.py:968] (0/2) Epoch 16, batch 5450, giga_loss[loss=0.2638, simple_loss=0.3328, pruned_loss=0.09735, over 28839.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3463, pruned_loss=0.09928, over 5719577.55 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3483, pruned_loss=0.09291, over 5336735.46 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3458, pruned_loss=0.09946, over 5718661.60 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:21:44,454 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.38 vs. limit=5.0
+2023-03-08 04:22:06,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9985, 4.8108, 4.5813, 2.3061], device='cuda:0'), covar=tensor([0.0477, 0.0650, 0.0654, 0.1995], device='cuda:0'), in_proj_covar=tensor([0.1129, 0.1038, 0.0900, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 04:22:26,171 INFO [train.py:968] (0/2) Epoch 16, batch 5500, giga_loss[loss=0.2553, simple_loss=0.331, pruned_loss=0.08983, over 28881.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3444, pruned_loss=0.09917, over 5727429.26 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3486, pruned_loss=0.09293, over 5352270.00 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3437, pruned_loss=0.0995, over 5722943.10 frames. ], batch size: 174, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:22:36,920 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 04:22:44,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4580, 4.3682, 1.7086, 1.6591], device='cuda:0'), covar=tensor([0.0941, 0.0390, 0.0871, 0.1302], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0529, 0.0359, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 04:22:56,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.747e+02 1.106e+03 1.355e+03 1.962e+03 7.009e+03, threshold=2.710e+03, percent-clipped=7.0
+2023-03-08 04:23:04,767 INFO [train.py:968] (0/2) Epoch 16, batch 5550, giga_loss[loss=0.2181, simple_loss=0.2939, pruned_loss=0.0711, over 28534.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3425, pruned_loss=0.09831, over 5732031.65 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3484, pruned_loss=0.093, over 5371154.77 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3419, pruned_loss=0.09873, over 5722988.90 frames. ], batch size: 60, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:23:42,864 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 04:23:53,170 INFO [train.py:968] (0/2) Epoch 16, batch 5600, giga_loss[loss=0.252, simple_loss=0.3284, pruned_loss=0.08777, over 28855.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3408, pruned_loss=0.09778, over 5722001.99 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3483, pruned_loss=0.09295, over 5376563.85 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3404, pruned_loss=0.0982, over 5713209.04 frames. ], batch size: 199, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:24:25,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.444e+02 1.192e+03 1.640e+03 2.493e+03 5.732e+03, threshold=3.279e+03, percent-clipped=20.0
+2023-03-08 04:24:35,113 INFO [train.py:968] (0/2) Epoch 16, batch 5650, giga_loss[loss=0.2501, simple_loss=0.3187, pruned_loss=0.09077, over 29046.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3369, pruned_loss=0.09564, over 5719445.42 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3487, pruned_loss=0.09321, over 5386106.62 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3361, pruned_loss=0.09582, over 5709931.90 frames. ], batch size: 128, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:24:36,234 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:24:37,490 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:24:54,891 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:24:56,301 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=689627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:25:08,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2054, 1.1067, 3.3247, 2.9788], device='cuda:0'), covar=tensor([0.1533, 0.2718, 0.0481, 0.0954], device='cuda:0'), in_proj_covar=tensor([0.0704, 0.0613, 0.0899, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 04:25:15,479 INFO [train.py:968] (0/2) Epoch 16, batch 5700, giga_loss[loss=0.2347, simple_loss=0.3088, pruned_loss=0.08024, over 28975.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3333, pruned_loss=0.09382, over 5721096.31 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3493, pruned_loss=0.09348, over 5395446.99 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3319, pruned_loss=0.09377, over 5711094.99 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:25:46,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.556e+02 1.095e+03 1.331e+03 1.694e+03 2.879e+03, threshold=2.662e+03, percent-clipped=0.0
+2023-03-08 04:25:56,570 INFO [train.py:968] (0/2) Epoch 16, batch 5750, giga_loss[loss=0.29, simple_loss=0.3627, pruned_loss=0.1086, over 28935.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3329, pruned_loss=0.09378, over 5718253.81 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3493, pruned_loss=0.0935, over 5409490.10 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3313, pruned_loss=0.09374, over 5706213.57 frames. ], batch size: 227, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:26:36,765 INFO [train.py:968] (0/2) Epoch 16, batch 5800, giga_loss[loss=0.2418, simple_loss=0.3196, pruned_loss=0.08198, over 28235.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3349, pruned_loss=0.09457, over 5707725.87 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3493, pruned_loss=0.09355, over 5405044.87 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3335, pruned_loss=0.09449, over 5704854.07 frames. ], batch size: 77, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:26:50,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=689770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:26:53,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=689773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:27:08,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.266e+02 1.235e+03 1.653e+03 2.212e+03 5.863e+03, threshold=3.306e+03, percent-clipped=13.0
+2023-03-08 04:27:18,779 INFO [train.py:968] (0/2) Epoch 16, batch 5850, giga_loss[loss=0.2522, simple_loss=0.3334, pruned_loss=0.08545, over 28833.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3393, pruned_loss=0.09662, over 5700850.21 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3493, pruned_loss=0.09371, over 5404863.51 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.338, pruned_loss=0.09645, over 5703964.61 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:27:19,554 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=689802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:28:02,463 INFO [train.py:968] (0/2) Epoch 16, batch 5900, giga_loss[loss=0.2462, simple_loss=0.3235, pruned_loss=0.08444, over 28404.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3426, pruned_loss=0.09772, over 5706252.64 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3493, pruned_loss=0.09369, over 5407350.19 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3415, pruned_loss=0.09762, over 5707857.39 frames. ], batch size: 71, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:28:02,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4841, 1.7454, 1.3461, 1.7521], device='cuda:0'), covar=tensor([0.2672, 0.2666, 0.3059, 0.2341], device='cuda:0'), in_proj_covar=tensor([0.1397, 0.1020, 0.1236, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 04:28:13,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3865, 1.5563, 1.1337, 1.0808], device='cuda:0'), covar=tensor([0.0827, 0.0502, 0.1017, 0.1294], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0439, 0.0500, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 04:28:13,958 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689865.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 04:28:36,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.003e+02 1.235e+03 1.599e+03 2.006e+03 6.037e+03, threshold=3.197e+03, percent-clipped=4.0
+2023-03-08 04:28:47,127 INFO [train.py:968] (0/2) Epoch 16, batch 5950, giga_loss[loss=0.2503, simple_loss=0.3334, pruned_loss=0.08358, over 28958.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3455, pruned_loss=0.09891, over 5699931.27 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3496, pruned_loss=0.09387, over 5409863.53 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3443, pruned_loss=0.0988, over 5706072.40 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:28:47,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4775, 1.5842, 1.5584, 1.4090], device='cuda:0'), covar=tensor([0.1562, 0.2123, 0.2186, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0735, 0.0689, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 04:29:18,400 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-08 04:29:35,684 INFO [train.py:968] (0/2) Epoch 16, batch 6000, giga_loss[loss=0.2676, simple_loss=0.3345, pruned_loss=0.1003, over 28774.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3479, pruned_loss=0.1007, over 5692519.40 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3494, pruned_loss=0.09376, over 5412191.21 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3471, pruned_loss=0.1007, over 5696617.57 frames. ], batch size: 92, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:29:35,688 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 04:29:44,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2045, 1.5897, 1.4464, 1.1929], device='cuda:0'), covar=tensor([0.2378, 0.1807, 0.1132, 0.1590], device='cuda:0'), in_proj_covar=tensor([0.1841, 0.1772, 0.1715, 0.1828], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 04:29:45,709 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2179, simple_loss=0.3243, pruned_loss=0.05574, over 944034.00 frames. 
+2023-03-08 04:29:45,710 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 04:30:08,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=689977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:30:09,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=689979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:30:19,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.148e+02 1.232e+03 1.511e+03 2.359e+03 5.814e+03, threshold=3.021e+03, percent-clipped=11.0
+2023-03-08 04:30:22,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3333, 1.6726, 1.2546, 1.5921], device='cuda:0'), covar=tensor([0.2394, 0.2330, 0.2743, 0.1979], device='cuda:0'), in_proj_covar=tensor([0.1396, 0.1020, 0.1236, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 04:30:26,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8011, 1.8393, 1.3300, 1.5258], device='cuda:0'), covar=tensor([0.0802, 0.0683, 0.0989, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0439, 0.0502, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 04:30:26,693 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2560, 2.9298, 1.4463, 1.3709], device='cuda:0'), covar=tensor([0.0938, 0.0396, 0.0860, 0.1308], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0530, 0.0359, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 04:30:32,915 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-690000.pt
+2023-03-08 04:30:33,480 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=690000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:30:33,961 INFO [train.py:968] (0/2) Epoch 16, batch 6050, giga_loss[loss=0.3093, simple_loss=0.3793, pruned_loss=0.1197, over 28807.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3541, pruned_loss=0.1059, over 5687825.72 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3494, pruned_loss=0.09387, over 5412479.22 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3534, pruned_loss=0.106, over 5695163.97 frames. ], batch size: 174, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:31:27,609 INFO [train.py:968] (0/2) Epoch 16, batch 6100, giga_loss[loss=0.2965, simple_loss=0.3708, pruned_loss=0.1111, over 28721.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3602, pruned_loss=0.1104, over 5686868.12 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3496, pruned_loss=0.09394, over 5414046.64 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3596, pruned_loss=0.1105, over 5693722.29 frames. ], batch size: 92, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:32:02,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.265e+02 1.750e+03 2.383e+03 3.688e+03 1.048e+04, threshold=4.765e+03, percent-clipped=34.0
+2023-03-08 04:32:13,156 INFO [train.py:968] (0/2) Epoch 16, batch 6150, giga_loss[loss=0.3115, simple_loss=0.3715, pruned_loss=0.1258, over 28556.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3668, pruned_loss=0.1155, over 5683020.85 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3498, pruned_loss=0.09396, over 5426676.57 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3666, pruned_loss=0.1162, over 5685265.87 frames. ], batch size: 60, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:32:31,562 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=690120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:32:32,928 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=690122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:32:33,506 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=690123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:32:35,839 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=690125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:32:55,297 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=690143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:32:57,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=690146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:33:00,385 INFO [train.py:968] (0/2) Epoch 16, batch 6200, giga_loss[loss=0.3479, simple_loss=0.3995, pruned_loss=0.1482, over 28497.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3711, pruned_loss=0.119, over 5679991.44 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3503, pruned_loss=0.09444, over 5428130.60 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.1199, over 5690719.79 frames. ], batch size: 336, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:33:02,217 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=690152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:33:04,697 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=690154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:33:24,523 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=690175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:33:40,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.717e+03 2.049e+03 2.756e+03 1.440e+04, threshold=4.098e+03, percent-clipped=6.0
+2023-03-08 04:33:47,079 INFO [train.py:968] (0/2) Epoch 16, batch 6250, libri_loss[loss=0.3009, simple_loss=0.3754, pruned_loss=0.1132, over 29528.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3773, pruned_loss=0.1241, over 5681656.32 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3508, pruned_loss=0.09478, over 5438639.15 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3776, pruned_loss=0.1252, over 5686132.31 frames. ], batch size: 84, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:34:11,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6607, 1.8093, 1.7987, 1.4409], device='cuda:0'), covar=tensor([0.2502, 0.2217, 0.1721, 0.2305], device='cuda:0'), in_proj_covar=tensor([0.1830, 0.1764, 0.1705, 0.1823], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 04:34:32,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=690240.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 04:34:40,007 INFO [train.py:968] (0/2) Epoch 16, batch 6300, giga_loss[loss=0.3073, simple_loss=0.3738, pruned_loss=0.1204, over 28691.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3806, pruned_loss=0.1274, over 5676603.79 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3503, pruned_loss=0.09451, over 5449234.54 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3818, pruned_loss=0.1293, over 5675733.36 frames. ], batch size: 242, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:34:40,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9997, 5.7863, 5.4204, 2.6942], device='cuda:0'), covar=tensor([0.0498, 0.0742, 0.0873, 0.1825], device='cuda:0'), in_proj_covar=tensor([0.1137, 0.1045, 0.0901, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 04:35:17,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 1.761e+03 2.311e+03 3.485e+03 1.372e+04, threshold=4.621e+03, percent-clipped=17.0
+2023-03-08 04:35:27,978 INFO [train.py:968] (0/2) Epoch 16, batch 6350, giga_loss[loss=0.4356, simple_loss=0.4453, pruned_loss=0.2129, over 23510.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3812, pruned_loss=0.1288, over 5667089.06 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3496, pruned_loss=0.09394, over 5467338.07 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3842, pruned_loss=0.1322, over 5658787.06 frames. ], batch size: 705, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:35:41,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5358, 1.7001, 1.7820, 1.3142], device='cuda:0'), covar=tensor([0.1673, 0.2427, 0.1373, 0.1694], device='cuda:0'), in_proj_covar=tensor([0.0860, 0.0690, 0.0907, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 04:35:48,909 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-08 04:36:21,204 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 04:36:22,063 INFO [train.py:968] (0/2) Epoch 16, batch 6400, giga_loss[loss=0.3161, simple_loss=0.368, pruned_loss=0.1321, over 28607.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3842, pruned_loss=0.1321, over 5674212.68 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3486, pruned_loss=0.09346, over 5482517.32 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3885, pruned_loss=0.1365, over 5660124.75 frames. ], batch size: 92, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:36:25,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3623, 1.8986, 1.4037, 0.4658], device='cuda:0'), covar=tensor([0.3922, 0.2788, 0.4182, 0.5306], device='cuda:0'), in_proj_covar=tensor([0.1646, 0.1560, 0.1542, 0.1346], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 04:36:58,245 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=690383.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 04:37:04,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=690386.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 04:37:08,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.313e+02 1.775e+03 2.101e+03 2.829e+03 9.690e+03, threshold=4.201e+03, percent-clipped=9.0
+2023-03-08 04:37:20,303 INFO [train.py:968] (0/2) Epoch 16, batch 6450, libri_loss[loss=0.2627, simple_loss=0.3484, pruned_loss=0.08846, over 29521.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3886, pruned_loss=0.1371, over 5653794.81 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3489, pruned_loss=0.09366, over 5483659.28 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3927, pruned_loss=0.1414, over 5645110.28 frames. ], batch size: 84, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:37:35,062 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=690415.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 04:37:58,966 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-08 04:38:12,356 INFO [train.py:968] (0/2) Epoch 16, batch 6500, giga_loss[loss=0.3093, simple_loss=0.3776, pruned_loss=0.1205, over 29119.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3907, pruned_loss=0.1388, over 5645151.29 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3486, pruned_loss=0.09347, over 5486488.97 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3947, pruned_loss=0.1428, over 5637149.97 frames. ], batch size: 128, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:38:43,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2353, 4.0628, 3.8779, 1.9161], device='cuda:0'), covar=tensor([0.0633, 0.0767, 0.0763, 0.2103], device='cuda:0'), in_proj_covar=tensor([0.1143, 0.1054, 0.0908, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 04:38:55,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.164e+03 1.938e+03 2.595e+03 4.396e+03 1.066e+04, threshold=5.190e+03, percent-clipped=26.0
+2023-03-08 04:39:02,768 INFO [train.py:968] (0/2) Epoch 16, batch 6550, giga_loss[loss=0.3404, simple_loss=0.3812, pruned_loss=0.1498, over 28960.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3905, pruned_loss=0.1399, over 5644255.82 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3484, pruned_loss=0.09349, over 5493121.86 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3948, pruned_loss=0.1442, over 5635217.71 frames. ], batch size: 100, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:39:09,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2343, 1.7792, 1.3152, 0.3588], device='cuda:0'), covar=tensor([0.3206, 0.1979, 0.3034, 0.4390], device='cuda:0'), in_proj_covar=tensor([0.1655, 0.1566, 0.1550, 0.1349], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 04:39:15,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7420, 1.9726, 1.6498, 1.7800], device='cuda:0'), covar=tensor([0.1955, 0.1800, 0.1848, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.1395, 0.1020, 0.1239, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 04:39:30,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1930, 4.2122, 1.3474, 1.5586], device='cuda:0'), covar=tensor([0.1181, 0.0350, 0.0969, 0.1409], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0531, 0.0359, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 04:39:57,749 INFO [train.py:968] (0/2) Epoch 16, batch 6600, giga_loss[loss=0.3459, simple_loss=0.3982, pruned_loss=0.1468, over 28608.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3888, pruned_loss=0.1391, over 5636891.04 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3484, pruned_loss=0.09345, over 5498725.11 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3928, pruned_loss=0.1433, over 5626493.32 frames. ], batch size: 336, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:40:24,637 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6589, 1.8724, 1.4693, 1.7912], device='cuda:0'), covar=tensor([0.2476, 0.2556, 0.2850, 0.2532], device='cuda:0'), in_proj_covar=tensor([0.1393, 0.1019, 0.1238, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 04:40:39,978 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.718e+03 2.298e+03 2.831e+03 6.088e+03, threshold=4.595e+03, percent-clipped=2.0
+2023-03-08 04:40:46,694 INFO [train.py:968] (0/2) Epoch 16, batch 6650, giga_loss[loss=0.3166, simple_loss=0.3902, pruned_loss=0.1215, over 28912.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3882, pruned_loss=0.1375, over 5645264.89 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3481, pruned_loss=0.0933, over 5505199.47 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3927, pruned_loss=0.142, over 5633745.36 frames. ], batch size: 199, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:41:38,101 INFO [train.py:968] (0/2) Epoch 16, batch 6700, libri_loss[loss=0.2952, simple_loss=0.3746, pruned_loss=0.1079, over 29102.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3899, pruned_loss=0.138, over 5649712.35 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3483, pruned_loss=0.09338, over 5512044.70 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3941, pruned_loss=0.1423, over 5636940.67 frames. ], batch size: 101, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:42:07,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2058, 4.0368, 3.8274, 1.9494], device='cuda:0'), covar=tensor([0.0514, 0.0693, 0.0660, 0.2311], device='cuda:0'), in_proj_covar=tensor([0.1143, 0.1054, 0.0909, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 04:42:17,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7313, 1.8787, 1.5855, 1.7301], device='cuda:0'), covar=tensor([0.0717, 0.0282, 0.0295, 0.0794], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 04:42:21,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.834e+03 2.869e+03 4.214e+03 9.133e+03, threshold=5.738e+03, percent-clipped=21.0
+2023-03-08 04:42:29,524 INFO [train.py:968] (0/2) Epoch 16, batch 6750, libri_loss[loss=0.2354, simple_loss=0.318, pruned_loss=0.07637, over 29537.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3897, pruned_loss=0.1376, over 5637451.49 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.348, pruned_loss=0.09317, over 5521708.07 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3944, pruned_loss=0.1424, over 5621007.17 frames. ], batch size: 76, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:43:23,868 INFO [train.py:968] (0/2) Epoch 16, batch 6800, giga_loss[loss=0.3799, simple_loss=0.4099, pruned_loss=0.1749, over 26529.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.387, pruned_loss=0.135, over 5627117.75 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3485, pruned_loss=0.09354, over 5520031.86 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3914, pruned_loss=0.1397, over 5617618.29 frames. ], batch size: 555, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:44:04,653 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.001e+02 1.511e+03 1.969e+03 2.535e+03 1.169e+04, threshold=3.939e+03, percent-clipped=1.0
+2023-03-08 04:44:12,154 INFO [train.py:968] (0/2) Epoch 16, batch 6850, giga_loss[loss=0.3732, simple_loss=0.4244, pruned_loss=0.161, over 28932.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3853, pruned_loss=0.1322, over 5644514.43 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3483, pruned_loss=0.09355, over 5523471.89 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3896, pruned_loss=0.1366, over 5635669.12 frames. ], batch size: 106, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:45:02,376 INFO [train.py:968] (0/2) Epoch 16, batch 6900, giga_loss[loss=0.2666, simple_loss=0.345, pruned_loss=0.09413, over 29005.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3802, pruned_loss=0.1276, over 5650711.11 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3481, pruned_loss=0.09358, over 5532681.85 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3846, pruned_loss=0.132, over 5637869.55 frames. ], batch size: 128, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:45:40,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.587e+03 1.974e+03 2.805e+03 8.746e+03, threshold=3.949e+03, percent-clipped=10.0
+2023-03-08 04:45:50,610 INFO [train.py:968] (0/2) Epoch 16, batch 6950, giga_loss[loss=0.281, simple_loss=0.359, pruned_loss=0.1015, over 28791.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3778, pruned_loss=0.1254, over 5658550.78 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3482, pruned_loss=0.09358, over 5542158.94 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3823, pruned_loss=0.1299, over 5643504.16 frames. ], batch size: 119, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:46:34,087 INFO [train.py:968] (0/2) Epoch 16, batch 7000, libri_loss[loss=0.2586, simple_loss=0.3438, pruned_loss=0.08677, over 29539.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3739, pruned_loss=0.1227, over 5660579.78 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3474, pruned_loss=0.09318, over 5558172.66 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3795, pruned_loss=0.1282, over 5638653.61 frames. ], batch size: 83, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:46:40,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6117, 3.4188, 3.2414, 1.7141], device='cuda:0'), covar=tensor([0.0792, 0.0883, 0.0814, 0.2332], device='cuda:0'), in_proj_covar=tensor([0.1152, 0.1062, 0.0915, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 04:46:45,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=690962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:47:15,811 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.337e+02 1.514e+03 2.182e+03 3.348e+03 8.130e+03, threshold=4.365e+03, percent-clipped=13.0
+2023-03-08 04:47:24,970 INFO [train.py:968] (0/2) Epoch 16, batch 7050, giga_loss[loss=0.3334, simple_loss=0.3822, pruned_loss=0.1423, over 28737.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.374, pruned_loss=0.1232, over 5658384.96 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.347, pruned_loss=0.09297, over 5563670.51 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3792, pruned_loss=0.1282, over 5637488.25 frames. ], batch size: 284, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:48:24,101 INFO [train.py:968] (0/2) Epoch 16, batch 7100, giga_loss[loss=0.337, simple_loss=0.3921, pruned_loss=0.141, over 28555.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3745, pruned_loss=0.1234, over 5661655.98 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.347, pruned_loss=0.09299, over 5570666.34 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3791, pruned_loss=0.1279, over 5640676.12 frames. ], batch size: 336, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:48:35,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5627, 1.6801, 1.9362, 1.3823], device='cuda:0'), covar=tensor([0.1735, 0.2514, 0.1427, 0.1751], device='cuda:0'), in_proj_covar=tensor([0.0858, 0.0692, 0.0905, 0.0806], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 04:48:55,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8024, 2.2500, 2.0603, 1.6056], device='cuda:0'), covar=tensor([0.3141, 0.2125, 0.2089, 0.2542], device='cuda:0'), in_proj_covar=tensor([0.1841, 0.1772, 0.1721, 0.1832], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 04:49:07,414 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-08 04:49:08,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.606e+02 1.402e+03 1.740e+03 2.377e+03 4.992e+03, threshold=3.481e+03, percent-clipped=4.0
+2023-03-08 04:49:17,698 INFO [train.py:968] (0/2) Epoch 16, batch 7150, giga_loss[loss=0.3174, simple_loss=0.3786, pruned_loss=0.1282, over 28796.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3736, pruned_loss=0.1213, over 5664687.99 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3473, pruned_loss=0.09326, over 5573480.75 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3774, pruned_loss=0.1252, over 5646916.49 frames. ], batch size: 99, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:49:55,141 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-08 04:50:05,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0977, 1.2308, 1.0795, 0.8989], device='cuda:0'), covar=tensor([0.1005, 0.0592, 0.1147, 0.1103], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0445, 0.0510, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 04:50:14,913 INFO [train.py:968] (0/2) Epoch 16, batch 7200, giga_loss[loss=0.3483, simple_loss=0.4053, pruned_loss=0.1457, over 27576.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3738, pruned_loss=0.1192, over 5671080.10 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3472, pruned_loss=0.09329, over 5583378.42 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3776, pruned_loss=0.123, over 5650541.81 frames. ], batch size: 472, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:50:50,778 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=691185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:50:58,973 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.985e+02 1.629e+03 2.289e+03 3.420e+03 1.211e+04, threshold=4.578e+03, percent-clipped=22.0
+2023-03-08 04:51:07,154 INFO [train.py:968] (0/2) Epoch 16, batch 7250, giga_loss[loss=0.3379, simple_loss=0.3937, pruned_loss=0.1411, over 28937.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3757, pruned_loss=0.1199, over 5680864.69 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3471, pruned_loss=0.09325, over 5586635.31 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3791, pruned_loss=0.1231, over 5662616.90 frames. ], batch size: 227, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:51:56,207 INFO [train.py:968] (0/2) Epoch 16, batch 7300, giga_loss[loss=0.3385, simple_loss=0.3955, pruned_loss=0.1407, over 28913.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3756, pruned_loss=0.1203, over 5675799.72 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3473, pruned_loss=0.09325, over 5599374.85 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3794, pruned_loss=0.124, over 5652945.89 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:52:03,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4811, 3.8441, 1.5862, 1.6625], device='cuda:0'), covar=tensor([0.0926, 0.0255, 0.0919, 0.1276], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0536, 0.0362, 0.0410], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 04:52:33,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.566e+02 1.496e+03 2.109e+03 3.017e+03 9.624e+03, threshold=4.219e+03, percent-clipped=8.0
+2023-03-08 04:52:40,587 INFO [train.py:968] (0/2) Epoch 16, batch 7350, giga_loss[loss=0.2973, simple_loss=0.3628, pruned_loss=0.1159, over 28208.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3737, pruned_loss=0.1193, over 5671344.97 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3471, pruned_loss=0.09326, over 5596936.34 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3785, pruned_loss=0.1239, over 5657984.39 frames. ], batch size: 77, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:53:16,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=691337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:53:30,170 INFO [train.py:968] (0/2) Epoch 16, batch 7400, giga_loss[loss=0.3005, simple_loss=0.3643, pruned_loss=0.1183, over 28700.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3723, pruned_loss=0.1194, over 5672910.17 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3472, pruned_loss=0.09343, over 5603957.70 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3768, pruned_loss=0.1237, over 5658169.36 frames. ], batch size: 307, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:53:58,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 04:54:09,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.876e+02 1.617e+03 1.920e+03 2.393e+03 6.205e+03, threshold=3.840e+03, percent-clipped=3.0
+2023-03-08 04:54:14,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1333, 1.3454, 3.6847, 3.0521], device='cuda:0'), covar=tensor([0.1734, 0.2559, 0.0468, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0707, 0.0616, 0.0906, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 04:54:15,617 INFO [train.py:968] (0/2) Epoch 16, batch 7450, giga_loss[loss=0.3143, simple_loss=0.3796, pruned_loss=0.1245, over 28775.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3722, pruned_loss=0.1206, over 5678555.16 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3473, pruned_loss=0.09344, over 5609301.41 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3762, pruned_loss=0.1245, over 5663109.33 frames. ], batch size: 119, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:55:06,545 INFO [train.py:968] (0/2) Epoch 16, batch 7500, giga_loss[loss=0.2802, simple_loss=0.355, pruned_loss=0.1027, over 28917.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3712, pruned_loss=0.1187, over 5674335.68 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3475, pruned_loss=0.09342, over 5614165.42 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3749, pruned_loss=0.1226, over 5659232.90 frames. ], batch size: 227, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:55:35,946 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=691480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:55:40,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=691483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:55:50,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.990e+02 1.425e+03 1.872e+03 2.632e+03 8.679e+03, threshold=3.743e+03, percent-clipped=10.0
+2023-03-08 04:55:56,860 INFO [train.py:968] (0/2) Epoch 16, batch 7550, libri_loss[loss=0.2672, simple_loss=0.3512, pruned_loss=0.09157, over 28612.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3714, pruned_loss=0.1179, over 5672543.56 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3478, pruned_loss=0.09355, over 5617294.05 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3744, pruned_loss=0.1212, over 5658488.46 frames. ], batch size: 106, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:56:07,650 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=691512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:56:20,018 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 04:56:41,856 INFO [train.py:968] (0/2) Epoch 16, batch 7600, giga_loss[loss=0.2759, simple_loss=0.3504, pruned_loss=0.1008, over 28976.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3716, pruned_loss=0.1181, over 5680868.98 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3477, pruned_loss=0.09373, over 5622451.07 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3747, pruned_loss=0.1212, over 5666471.81 frames. ], batch size: 136, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:56:49,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=691560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:57:01,768 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-08 04:57:19,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.545e+02 1.674e+03 2.102e+03 3.002e+03 4.702e+03, threshold=4.205e+03, percent-clipped=10.0
+2023-03-08 04:57:27,738 INFO [train.py:968] (0/2) Epoch 16, batch 7650, giga_loss[loss=0.3086, simple_loss=0.3507, pruned_loss=0.1333, over 23824.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3686, pruned_loss=0.116, over 5694251.65 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3471, pruned_loss=0.09339, over 5631723.99 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3724, pruned_loss=0.1196, over 5676134.60 frames. ], batch size: 705, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:57:34,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6608, 2.0308, 1.8903, 1.4486], device='cuda:0'), covar=tensor([0.1974, 0.2615, 0.1661, 0.1933], device='cuda:0'), in_proj_covar=tensor([0.0864, 0.0697, 0.0911, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 04:58:14,519 INFO [train.py:968] (0/2) Epoch 16, batch 7700, giga_loss[loss=0.373, simple_loss=0.4052, pruned_loss=0.1704, over 23529.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3678, pruned_loss=0.1165, over 5680385.35 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3472, pruned_loss=0.09341, over 5638369.26 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3714, pruned_loss=0.1201, over 5661713.61 frames. ], batch size: 705, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:58:52,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7783, 1.9788, 1.5638, 1.9157], device='cuda:0'), covar=tensor([0.2356, 0.2445, 0.2787, 0.2510], device='cuda:0'), in_proj_covar=tensor([0.1395, 0.1021, 0.1238, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 04:58:58,114 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.943e+02 1.592e+03 1.981e+03 2.692e+03 5.857e+03, threshold=3.962e+03, percent-clipped=5.0
+2023-03-08 04:59:06,391 INFO [train.py:968] (0/2) Epoch 16, batch 7750, giga_loss[loss=0.3071, simple_loss=0.3724, pruned_loss=0.1209, over 28997.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.366, pruned_loss=0.1163, over 5682285.97 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3468, pruned_loss=0.09328, over 5643874.90 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3696, pruned_loss=0.1197, over 5663287.09 frames. ], batch size: 136, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:59:09,401 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=691703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:59:12,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=691706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:59:17,416 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 04:59:31,846 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=691725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:59:40,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=691735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:59:42,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2158, 3.0401, 2.8832, 1.5206], device='cuda:0'), covar=tensor([0.1020, 0.1052, 0.0944, 0.2320], device='cuda:0'), in_proj_covar=tensor([0.1147, 0.1056, 0.0911, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 04:59:56,159 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=691748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:59:58,354 INFO [train.py:968] (0/2) Epoch 16, batch 7800, giga_loss[loss=0.342, simple_loss=0.4017, pruned_loss=0.1412, over 28670.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3661, pruned_loss=0.1175, over 5671937.91 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3466, pruned_loss=0.09314, over 5648372.80 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3697, pruned_loss=0.1209, over 5653640.58 frames. ], batch size: 307, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:00:45,062 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=691792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:00:46,266 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.912e+02 1.605e+03 1.964e+03 2.944e+03 8.859e+03, threshold=3.929e+03, percent-clipped=11.0
+2023-03-08 05:00:51,725 INFO [train.py:968] (0/2) Epoch 16, batch 7850, libri_loss[loss=0.2464, simple_loss=0.3265, pruned_loss=0.08309, over 29571.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3651, pruned_loss=0.1176, over 5667806.89 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3465, pruned_loss=0.09309, over 5649701.94 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.368, pruned_loss=0.1204, over 5652256.12 frames. ], batch size: 75, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:01:40,536 INFO [train.py:968] (0/2) Epoch 16, batch 7900, giga_loss[loss=0.309, simple_loss=0.372, pruned_loss=0.1229, over 28273.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3647, pruned_loss=0.1175, over 5666530.74 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3463, pruned_loss=0.0929, over 5653590.06 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3675, pruned_loss=0.1204, over 5650870.46 frames. ], batch size: 368, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:02:24,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.431e+02 1.667e+03 2.070e+03 2.761e+03 5.152e+03, threshold=4.141e+03, percent-clipped=7.0
+2023-03-08 05:02:32,646 INFO [train.py:968] (0/2) Epoch 16, batch 7950, giga_loss[loss=0.2986, simple_loss=0.3686, pruned_loss=0.1143, over 28836.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3663, pruned_loss=0.1181, over 5667890.46 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3459, pruned_loss=0.09264, over 5657441.00 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.1209, over 5652341.35 frames. ], batch size: 284, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:03:18,939 INFO [train.py:968] (0/2) Epoch 16, batch 8000, giga_loss[loss=0.3057, simple_loss=0.3717, pruned_loss=0.1198, over 28198.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.367, pruned_loss=0.1177, over 5672887.89 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3464, pruned_loss=0.09305, over 5661541.32 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3694, pruned_loss=0.1203, over 5656733.26 frames. ], batch size: 368, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 05:03:51,738 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=691990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:03:56,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.116e+02 1.366e+03 1.794e+03 2.873e+03 1.016e+04, threshold=3.588e+03, percent-clipped=14.0
+2023-03-08 05:04:04,015 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-692000.pt
+2023-03-08 05:04:05,028 INFO [train.py:968] (0/2) Epoch 16, batch 8050, libri_loss[loss=0.2264, simple_loss=0.3111, pruned_loss=0.07088, over 29574.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3666, pruned_loss=0.116, over 5690200.11 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3463, pruned_loss=0.09293, over 5669860.26 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3693, pruned_loss=0.119, over 5670293.00 frames. ], batch size: 76, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:04:55,803 INFO [train.py:968] (0/2) Epoch 16, batch 8100, giga_loss[loss=0.3405, simple_loss=0.3812, pruned_loss=0.1499, over 26506.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3666, pruned_loss=0.1163, over 5688162.22 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.346, pruned_loss=0.09268, over 5674604.63 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3694, pruned_loss=0.1194, over 5668394.47 frames. ], batch size: 555, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:05:23,679 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692079.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:05:41,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.561e+03 2.020e+03 2.716e+03 7.153e+03, threshold=4.039e+03, percent-clipped=11.0
+2023-03-08 05:05:46,908 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:05:47,262 INFO [train.py:968] (0/2) Epoch 16, batch 8150, giga_loss[loss=0.3069, simple_loss=0.3754, pruned_loss=0.1192, over 28948.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3701, pruned_loss=0.1199, over 5682220.01 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3459, pruned_loss=0.09265, over 5677903.26 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3727, pruned_loss=0.1227, over 5663900.17 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:05:48,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9624, 1.1400, 3.2851, 2.9116], device='cuda:0'), covar=tensor([0.1764, 0.2681, 0.0536, 0.1099], device='cuda:0'), in_proj_covar=tensor([0.0714, 0.0620, 0.0915, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 05:06:10,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:06:28,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2244, 1.3391, 1.3714, 0.9825], device='cuda:0'), covar=tensor([0.1770, 0.3228, 0.1509, 0.1575], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0698, 0.0911, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 05:06:31,769 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-08 05:06:39,977 INFO [train.py:968] (0/2) Epoch 16, batch 8200, giga_loss[loss=0.3672, simple_loss=0.401, pruned_loss=0.1667, over 27537.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3716, pruned_loss=0.1222, over 5677143.72 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3462, pruned_loss=0.09284, over 5688190.32 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3745, pruned_loss=0.1256, over 5652461.72 frames. ], batch size: 472, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:06:55,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:07:24,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4736, 1.6624, 1.4551, 1.2958], device='cuda:0'), covar=tensor([0.2233, 0.1886, 0.1687, 0.2011], device='cuda:0'), in_proj_covar=tensor([0.1851, 0.1787, 0.1727, 0.1844], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 05:07:25,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.828e+03 2.309e+03 3.516e+03 8.443e+03, threshold=4.617e+03, percent-clipped=21.0
+2023-03-08 05:07:31,560 INFO [train.py:968] (0/2) Epoch 16, batch 8250, giga_loss[loss=0.3511, simple_loss=0.3975, pruned_loss=0.1523, over 28333.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3722, pruned_loss=0.1238, over 5675025.18 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.346, pruned_loss=0.09264, over 5689344.46 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3751, pruned_loss=0.127, over 5654322.87 frames. ], batch size: 368, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:08:18,470 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:08:24,469 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:08:28,941 INFO [train.py:968] (0/2) Epoch 16, batch 8300, giga_loss[loss=0.3381, simple_loss=0.3883, pruned_loss=0.1439, over 28970.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3751, pruned_loss=0.1272, over 5666513.92 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.346, pruned_loss=0.09264, over 5689344.46 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3773, pruned_loss=0.1297, over 5650401.06 frames. ], batch size: 227, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:08:45,925 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:08:48,611 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:08:49,191 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:08:55,960 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=692275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:09:12,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.701e+02 1.793e+03 2.560e+03 4.684e+03 1.039e+04, threshold=5.120e+03, percent-clipped=25.0
+2023-03-08 05:09:16,131 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=692298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:09:17,985 INFO [train.py:968] (0/2) Epoch 16, batch 8350, libri_loss[loss=0.2935, simple_loss=0.3714, pruned_loss=0.1078, over 29523.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3737, pruned_loss=0.1259, over 5674915.82 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3461, pruned_loss=0.0927, over 5691501.58 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3756, pruned_loss=0.1282, over 5659951.07 frames. ], batch size: 82, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:09:27,374 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:09:29,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:09:54,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=692342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:09:55,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2436, 3.1196, 1.3688, 1.3658], device='cuda:0'), covar=tensor([0.1018, 0.0387, 0.0933, 0.1422], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0534, 0.0359, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 05:10:03,374 INFO [train.py:968] (0/2) Epoch 16, batch 8400, giga_loss[loss=0.2881, simple_loss=0.3632, pruned_loss=0.1066, over 28683.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3726, pruned_loss=0.1236, over 5678721.84 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3464, pruned_loss=0.0929, over 5686831.34 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3744, pruned_loss=0.1259, over 5671127.09 frames. ], batch size: 78, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 05:10:14,423 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:10:22,903 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9745, 2.0597, 1.5023, 1.6709], device='cuda:0'), covar=tensor([0.0896, 0.0704, 0.1040, 0.1210], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0446, 0.0507, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 05:10:44,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.626e+02 1.460e+03 2.183e+03 2.857e+03 9.687e+03, threshold=4.367e+03, percent-clipped=4.0
+2023-03-08 05:10:48,333 INFO [train.py:968] (0/2) Epoch 16, batch 8450, giga_loss[loss=0.2947, simple_loss=0.3653, pruned_loss=0.1121, over 28172.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.371, pruned_loss=0.1208, over 5684826.76 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3469, pruned_loss=0.09306, over 5692207.42 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3726, pruned_loss=0.1232, over 5673836.54 frames. ], batch size: 77, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:11:29,390 INFO [train.py:968] (0/2) Epoch 16, batch 8500, giga_loss[loss=0.2679, simple_loss=0.3462, pruned_loss=0.09481, over 28583.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3687, pruned_loss=0.1192, over 5675447.23 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3469, pruned_loss=0.09327, over 5685778.19 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3709, pruned_loss=0.122, over 5671434.62 frames. ], batch size: 85, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:11:33,200 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692454.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:12:07,702 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:12:16,679 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.739e+03 2.091e+03 3.239e+03 8.876e+03, threshold=4.181e+03, percent-clipped=11.0
+2023-03-08 05:12:20,324 INFO [train.py:968] (0/2) Epoch 16, batch 8550, giga_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1117, over 28966.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3669, pruned_loss=0.1188, over 5665677.90 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3473, pruned_loss=0.09352, over 5678014.89 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3685, pruned_loss=0.121, over 5669844.70 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 05:12:27,568 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:12:30,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:12:58,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4378, 1.6158, 1.6711, 1.3657], device='cuda:0'), covar=tensor([0.2133, 0.1837, 0.1358, 0.1902], device='cuda:0'), in_proj_covar=tensor([0.1850, 0.1783, 0.1724, 0.1846], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 05:13:03,448 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=692540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:13:15,276 INFO [train.py:968] (0/2) Epoch 16, batch 8600, giga_loss[loss=0.3276, simple_loss=0.3913, pruned_loss=0.132, over 28963.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3675, pruned_loss=0.1198, over 5661743.45 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3474, pruned_loss=0.09352, over 5678734.82 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3687, pruned_loss=0.1216, over 5664391.42 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 05:13:22,394 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 05:14:03,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.620e+03 2.203e+03 2.643e+03 7.923e+03, threshold=4.406e+03, percent-clipped=6.0
+2023-03-08 05:14:04,149 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692597.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:14:08,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692600.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:14:09,831 INFO [train.py:968] (0/2) Epoch 16, batch 8650, giga_loss[loss=0.3092, simple_loss=0.388, pruned_loss=0.1152, over 28941.00 frames. ], tot_loss[loss=0.307, simple_loss=0.371, pruned_loss=0.1215, over 5667361.69 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3476, pruned_loss=0.09361, over 5683492.53 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3722, pruned_loss=0.1234, over 5665075.20 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 05:14:23,309 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4772, 1.2910, 4.0006, 3.3074], device='cuda:0'), covar=tensor([0.1529, 0.2683, 0.0433, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0710, 0.0617, 0.0911, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 05:14:37,460 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=692629.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:14:51,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:14:56,925 INFO [train.py:968] (0/2) Epoch 16, batch 8700, giga_loss[loss=0.3128, simple_loss=0.3975, pruned_loss=0.114, over 28912.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3754, pruned_loss=0.1218, over 5674726.31 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3479, pruned_loss=0.09382, over 5689610.11 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3768, pruned_loss=0.1239, over 5666745.58 frames. ], batch size: 164, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:15:22,790 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:15:30,116 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:15:45,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.835e+02 1.581e+03 2.067e+03 2.553e+03 6.753e+03, threshold=4.134e+03, percent-clipped=6.0
+2023-03-08 05:15:47,936 INFO [train.py:968] (0/2) Epoch 16, batch 8750, giga_loss[loss=0.2677, simple_loss=0.3474, pruned_loss=0.094, over 28986.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3774, pruned_loss=0.1222, over 5679685.07 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3477, pruned_loss=0.09376, over 5692765.22 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3791, pruned_loss=0.1243, over 5670393.59 frames. ], batch size: 136, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:16:35,344 INFO [train.py:968] (0/2) Epoch 16, batch 8800, giga_loss[loss=0.3019, simple_loss=0.3637, pruned_loss=0.12, over 28706.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3784, pruned_loss=0.1232, over 5666417.36 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3477, pruned_loss=0.09377, over 5676875.62 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3805, pruned_loss=0.1255, over 5672771.29 frames. ], batch size: 99, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:17:10,940 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:17:13,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:17:14,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5218, 1.6520, 1.5765, 1.3844], device='cuda:0'), covar=tensor([0.2316, 0.2038, 0.1705, 0.2026], device='cuda:0'), in_proj_covar=tensor([0.1849, 0.1784, 0.1721, 0.1845], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 05:17:18,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.704e+03 2.107e+03 3.125e+03 9.001e+03, threshold=4.214e+03, percent-clipped=13.0
+2023-03-08 05:17:23,193 INFO [train.py:968] (0/2) Epoch 16, batch 8850, giga_loss[loss=0.2961, simple_loss=0.3581, pruned_loss=0.117, over 28906.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.379, pruned_loss=0.1239, over 5673910.98 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3481, pruned_loss=0.094, over 5674766.63 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3808, pruned_loss=0.1261, over 5680805.09 frames. ], batch size: 106, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:17:39,012 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 05:17:40,162 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=692818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:17:45,680 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7121, 1.7617, 1.5527, 1.5347], device='cuda:0'), covar=tensor([0.1676, 0.2436, 0.2281, 0.2171], device='cuda:0'), in_proj_covar=tensor([0.0454, 0.0738, 0.0691, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 05:18:11,891 INFO [train.py:968] (0/2) Epoch 16, batch 8900, giga_loss[loss=0.3172, simple_loss=0.3825, pruned_loss=0.1259, over 28996.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3781, pruned_loss=0.1246, over 5675028.99 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3477, pruned_loss=0.09376, over 5681238.90 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3807, pruned_loss=0.1274, over 5674562.74 frames. ], batch size: 227, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:18:23,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:18:59,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.231e+02 1.432e+03 1.745e+03 2.551e+03 8.059e+03, threshold=3.489e+03, percent-clipped=4.0
+2023-03-08 05:19:02,242 INFO [train.py:968] (0/2) Epoch 16, batch 8950, libri_loss[loss=0.2606, simple_loss=0.3475, pruned_loss=0.08691, over 29675.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3753, pruned_loss=0.123, over 5687200.53 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3477, pruned_loss=0.09373, over 5690947.88 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3787, pruned_loss=0.1265, over 5677921.56 frames. ], batch size: 88, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:19:39,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9179, 1.7905, 1.4852, 1.3284], device='cuda:0'), covar=tensor([0.0846, 0.0675, 0.0988, 0.1162], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0444, 0.0505, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 05:19:52,385 INFO [train.py:968] (0/2) Epoch 16, batch 9000, giga_loss[loss=0.2831, simple_loss=0.3488, pruned_loss=0.1087, over 28556.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3743, pruned_loss=0.1238, over 5674721.24 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3478, pruned_loss=0.09382, over 5683629.00 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3771, pruned_loss=0.1267, over 5673930.65 frames. ], batch size: 85, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:19:52,389 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 05:19:59,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3038, 3.1170, 1.3982, 1.4824], device='cuda:0'), covar=tensor([0.1139, 0.0335, 0.1061, 0.1574], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0535, 0.0359, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 05:20:01,272 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2119, simple_loss=0.3202, pruned_loss=0.05187, over 944034.00 frames. 
+2023-03-08 05:20:01,272 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 05:20:07,913 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3239, 1.2991, 3.8370, 3.1973], device='cuda:0'), covar=tensor([0.1603, 0.2688, 0.0447, 0.1000], device='cuda:0'), in_proj_covar=tensor([0.0714, 0.0622, 0.0917, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 05:20:36,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5441, 3.2340, 1.6271, 1.5741], device='cuda:0'), covar=tensor([0.0893, 0.0344, 0.0827, 0.1250], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0535, 0.0359, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 05:20:43,534 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692993.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:20:47,113 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.766e+03 2.233e+03 3.347e+03 1.544e+04, threshold=4.465e+03, percent-clipped=22.0
+2023-03-08 05:20:50,851 INFO [train.py:968] (0/2) Epoch 16, batch 9050, giga_loss[loss=0.3606, simple_loss=0.3808, pruned_loss=0.1702, over 23370.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3725, pruned_loss=0.1232, over 5668697.18 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3478, pruned_loss=0.09381, over 5684509.24 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3752, pruned_loss=0.126, over 5667278.50 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:20:51,783 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693002.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:20:55,489 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:20:57,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:21:29,100 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:21:39,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:21:42,094 INFO [train.py:968] (0/2) Epoch 16, batch 9100, giga_loss[loss=0.2974, simple_loss=0.3689, pruned_loss=0.1129, over 29092.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.372, pruned_loss=0.1225, over 5678548.42 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3481, pruned_loss=0.09389, over 5688914.65 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3743, pruned_loss=0.1252, over 5673474.50 frames. ], batch size: 155, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:21:50,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:22:13,416 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 05:22:27,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1842, 1.4646, 1.4680, 1.1039], device='cuda:0'), covar=tensor([0.1169, 0.1898, 0.0996, 0.1295], device='cuda:0'), in_proj_covar=tensor([0.0862, 0.0695, 0.0908, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 05:22:34,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.781e+02 1.571e+03 1.920e+03 2.782e+03 5.156e+03, threshold=3.839e+03, percent-clipped=4.0
+2023-03-08 05:22:36,132 INFO [train.py:968] (0/2) Epoch 16, batch 9150, giga_loss[loss=0.3177, simple_loss=0.3727, pruned_loss=0.1313, over 28309.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3721, pruned_loss=0.1238, over 5672037.76 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3482, pruned_loss=0.094, over 5689016.07 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3739, pruned_loss=0.126, over 5667882.51 frames. ], batch size: 368, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:23:25,017 INFO [train.py:968] (0/2) Epoch 16, batch 9200, giga_loss[loss=0.3742, simple_loss=0.4121, pruned_loss=0.1681, over 26529.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3707, pruned_loss=0.1227, over 5677707.40 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3488, pruned_loss=0.09438, over 5693232.54 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3719, pruned_loss=0.1246, over 5670464.61 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:23:35,692 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:03,507 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:05,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:07,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.558e+03 2.110e+03 2.926e+03 9.293e+03, threshold=4.219e+03, percent-clipped=14.0
+2023-03-08 05:24:09,728 INFO [train.py:968] (0/2) Epoch 16, batch 9250, giga_loss[loss=0.2711, simple_loss=0.3495, pruned_loss=0.09632, over 28964.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3709, pruned_loss=0.1225, over 5680856.39 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3489, pruned_loss=0.09443, over 5689264.83 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3723, pruned_loss=0.1246, over 5678344.48 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:24:10,000 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:13,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:32,143 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:42,043 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:59,243 INFO [train.py:968] (0/2) Epoch 16, batch 9300, giga_loss[loss=0.2747, simple_loss=0.3456, pruned_loss=0.1019, over 28943.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3718, pruned_loss=0.1223, over 5660864.71 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.349, pruned_loss=0.09447, over 5675046.03 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3735, pruned_loss=0.1248, over 5671102.82 frames. ], batch size: 106, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:25:48,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.166e+02 1.383e+03 1.814e+03 2.464e+03 6.802e+03, threshold=3.629e+03, percent-clipped=3.0
+2023-03-08 05:25:50,891 INFO [train.py:968] (0/2) Epoch 16, batch 9350, giga_loss[loss=0.3058, simple_loss=0.3607, pruned_loss=0.1254, over 28752.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3741, pruned_loss=0.1241, over 5655831.31 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3493, pruned_loss=0.09458, over 5676044.68 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3756, pruned_loss=0.1265, over 5662856.03 frames. ], batch size: 99, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:26:39,608 INFO [train.py:968] (0/2) Epoch 16, batch 9400, giga_loss[loss=0.3498, simple_loss=0.4065, pruned_loss=0.1465, over 28689.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3736, pruned_loss=0.1236, over 5663516.49 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3499, pruned_loss=0.09484, over 5679446.01 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3748, pruned_loss=0.1259, over 5665694.21 frames. ], batch size: 307, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:26:55,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693368.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:27:07,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693377.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:27:26,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.085e+02 1.285e+03 1.704e+03 2.188e+03 5.778e+03, threshold=3.409e+03, percent-clipped=6.0
+2023-03-08 05:27:30,950 INFO [train.py:968] (0/2) Epoch 16, batch 9450, giga_loss[loss=0.3274, simple_loss=0.3904, pruned_loss=0.1322, over 28596.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3751, pruned_loss=0.1216, over 5676139.47 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3498, pruned_loss=0.09477, over 5681647.32 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3763, pruned_loss=0.1237, over 5675933.30 frames. ], batch size: 307, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:28:00,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5987, 1.9350, 1.2369, 1.5338], device='cuda:0'), covar=tensor([0.0954, 0.0602, 0.1098, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0445, 0.0504, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 05:28:13,218 INFO [train.py:968] (0/2) Epoch 16, batch 9500, giga_loss[loss=0.2388, simple_loss=0.3305, pruned_loss=0.07354, over 28278.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3752, pruned_loss=0.12, over 5681070.37 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3496, pruned_loss=0.0949, over 5691689.41 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3776, pruned_loss=0.1228, over 5671352.28 frames. ], batch size: 65, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:28:58,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.104e+02 1.757e+03 2.333e+03 3.127e+03 7.775e+03, threshold=4.666e+03, percent-clipped=17.0
+2023-03-08 05:29:02,646 INFO [train.py:968] (0/2) Epoch 16, batch 9550, giga_loss[loss=0.4248, simple_loss=0.4494, pruned_loss=0.2001, over 26752.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3792, pruned_loss=0.1235, over 5669729.87 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09483, over 5684502.39 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.382, pruned_loss=0.1263, over 5667546.63 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:29:07,348 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693507.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:29:11,259 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693511.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:29:13,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693514.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:29:20,225 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693520.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:29:24,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:29:38,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:29:43,999 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693543.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:29:50,605 INFO [train.py:968] (0/2) Epoch 16, batch 9600, giga_loss[loss=0.3804, simple_loss=0.4214, pruned_loss=0.1697, over 29052.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.382, pruned_loss=0.1264, over 5675724.02 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3494, pruned_loss=0.09481, over 5687896.55 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3845, pruned_loss=0.1291, over 5671004.31 frames. ], batch size: 136, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 05:29:51,510 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693552.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:30:01,094 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:30:10,025 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-08 05:30:36,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4976, 2.2026, 1.6566, 0.6654], device='cuda:0'), covar=tensor([0.4766, 0.2463, 0.3680, 0.5386], device='cuda:0'), in_proj_covar=tensor([0.1661, 0.1576, 0.1544, 0.1356], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 05:30:40,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.783e+02 1.635e+03 2.428e+03 3.125e+03 7.670e+03, threshold=4.856e+03, percent-clipped=6.0
+2023-03-08 05:30:41,355 INFO [train.py:968] (0/2) Epoch 16, batch 9650, giga_loss[loss=0.3672, simple_loss=0.4094, pruned_loss=0.1625, over 29019.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3824, pruned_loss=0.128, over 5675307.55 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3494, pruned_loss=0.09486, over 5691016.02 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.385, pruned_loss=0.1306, over 5668779.86 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:30:58,664 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:31:31,282 INFO [train.py:968] (0/2) Epoch 16, batch 9700, giga_loss[loss=0.3254, simple_loss=0.3747, pruned_loss=0.138, over 23749.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3807, pruned_loss=0.1272, over 5663523.58 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3491, pruned_loss=0.09476, over 5695340.60 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3835, pruned_loss=0.13, over 5654066.56 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:31:56,219 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:31:58,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:32:13,869 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.407e+02 1.881e+03 2.573e+03 3.447e+03 7.765e+03, threshold=5.146e+03, percent-clipped=9.0
+2023-03-08 05:32:14,983 INFO [train.py:968] (0/2) Epoch 16, batch 9750, giga_loss[loss=0.3997, simple_loss=0.4447, pruned_loss=0.1774, over 28587.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3792, pruned_loss=0.1253, over 5670531.72 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.349, pruned_loss=0.09474, over 5700768.66 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3825, pruned_loss=0.1286, over 5656989.44 frames. ], batch size: 336, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:32:17,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-08 05:32:26,996 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:32:43,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5712, 1.6557, 1.6459, 1.4722], device='cuda:0'), covar=tensor([0.2726, 0.2465, 0.1964, 0.2211], device='cuda:0'), in_proj_covar=tensor([0.1842, 0.1780, 0.1720, 0.1835], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 05:33:03,034 INFO [train.py:968] (0/2) Epoch 16, batch 9800, giga_loss[loss=0.3761, simple_loss=0.418, pruned_loss=0.167, over 27622.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.379, pruned_loss=0.1233, over 5670818.82 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3495, pruned_loss=0.09516, over 5700979.28 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3815, pruned_loss=0.1258, over 5659686.08 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:33:30,433 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:33:38,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8269, 1.0870, 2.8093, 2.7801], device='cuda:0'), covar=tensor([0.1732, 0.2655, 0.0654, 0.1035], device='cuda:0'), in_proj_covar=tensor([0.0718, 0.0625, 0.0923, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 05:33:47,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.609e+02 1.512e+03 2.080e+03 3.193e+03 9.701e+03, threshold=4.160e+03, percent-clipped=4.0
+2023-03-08 05:33:48,517 INFO [train.py:968] (0/2) Epoch 16, batch 9850, giga_loss[loss=0.3253, simple_loss=0.3843, pruned_loss=0.1331, over 28570.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3786, pruned_loss=0.1226, over 5682036.20 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3493, pruned_loss=0.09514, over 5706933.27 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3816, pruned_loss=0.1253, over 5666848.41 frames. ], batch size: 336, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:34:06,293 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:34:10,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1711, 1.4498, 1.4541, 1.0974], device='cuda:0'), covar=tensor([0.1326, 0.2062, 0.1109, 0.1354], device='cuda:0'), in_proj_covar=tensor([0.0862, 0.0697, 0.0910, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 05:34:13,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9536, 1.3057, 1.1169, 0.1357], device='cuda:0'), covar=tensor([0.3590, 0.2860, 0.3970, 0.5806], device='cuda:0'), in_proj_covar=tensor([0.1657, 0.1567, 0.1539, 0.1348], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 05:34:33,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7058, 5.2165, 1.9195, 1.9254], device='cuda:0'), covar=tensor([0.1004, 0.0290, 0.0857, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0537, 0.0362, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 05:34:41,429 INFO [train.py:968] (0/2) Epoch 16, batch 9900, giga_loss[loss=0.3668, simple_loss=0.4147, pruned_loss=0.1594, over 28596.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3803, pruned_loss=0.1244, over 5666352.32 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3495, pruned_loss=0.09523, over 5700585.34 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3829, pruned_loss=0.1269, over 5658857.68 frames. ], batch size: 307, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:35:00,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3281, 3.1472, 2.9651, 1.4291], device='cuda:0'), covar=tensor([0.0928, 0.1051, 0.0937, 0.2292], device='cuda:0'), in_proj_covar=tensor([0.1159, 0.1070, 0.0923, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 05:35:07,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-08 05:35:14,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693882.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:35:31,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.290e+02 1.453e+03 1.906e+03 2.651e+03 7.235e+03, threshold=3.813e+03, percent-clipped=7.0
+2023-03-08 05:35:32,290 INFO [train.py:968] (0/2) Epoch 16, batch 9950, giga_loss[loss=0.2861, simple_loss=0.3578, pruned_loss=0.1072, over 28857.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3796, pruned_loss=0.1243, over 5671619.72 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3493, pruned_loss=0.09502, over 5704453.66 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3827, pruned_loss=0.1273, over 5661309.18 frames. ], batch size: 174, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:36:09,967 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:36:20,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2860, 3.5190, 1.5026, 1.3807], device='cuda:0'), covar=tensor([0.0939, 0.0375, 0.0845, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0537, 0.0361, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 05:36:22,901 INFO [train.py:968] (0/2) Epoch 16, batch 10000, giga_loss[loss=0.2641, simple_loss=0.3431, pruned_loss=0.09251, over 28888.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3786, pruned_loss=0.1247, over 5669419.11 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3499, pruned_loss=0.09538, over 5708084.78 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3811, pruned_loss=0.1273, over 5657459.06 frames. ], batch size: 174, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:37:04,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:37:11,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.159e+03 1.688e+03 2.288e+03 3.576e+03 1.041e+04, threshold=4.576e+03, percent-clipped=21.0
+2023-03-08 05:37:11,283 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-694000.pt
+2023-03-08 05:37:12,451 INFO [train.py:968] (0/2) Epoch 16, batch 10050, giga_loss[loss=0.2617, simple_loss=0.342, pruned_loss=0.09072, over 28959.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3774, pruned_loss=0.1249, over 5671768.32 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3503, pruned_loss=0.09562, over 5712888.16 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3796, pruned_loss=0.1274, over 5657237.63 frames. ], batch size: 164, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:37:31,418 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694025.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:37:34,737 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694028.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:37:44,557 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:37:59,045 INFO [train.py:968] (0/2) Epoch 16, batch 10100, giga_loss[loss=0.3112, simple_loss=0.3767, pruned_loss=0.1229, over 28967.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3743, pruned_loss=0.1225, over 5687100.57 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3498, pruned_loss=0.09528, over 5722493.97 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3777, pruned_loss=0.1261, over 5665363.16 frames. ], batch size: 136, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:38:08,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694057.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:38:19,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7120, 1.8681, 1.7611, 1.6656], device='cuda:0'), covar=tensor([0.1689, 0.2074, 0.2160, 0.2015], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0741, 0.0695, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 05:38:31,162 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:38:31,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4180, 1.5280, 1.4084, 1.6052], device='cuda:0'), covar=tensor([0.0734, 0.0317, 0.0299, 0.0784], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:0')
+2023-03-08 05:38:34,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:38:49,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.753e+02 1.617e+03 1.978e+03 2.891e+03 6.470e+03, threshold=3.955e+03, percent-clipped=6.0
+2023-03-08 05:38:52,663 INFO [train.py:968] (0/2) Epoch 16, batch 10150, giga_loss[loss=0.4139, simple_loss=0.4407, pruned_loss=0.1935, over 26637.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.374, pruned_loss=0.1237, over 5680659.44 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3498, pruned_loss=0.09523, over 5724758.50 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.377, pruned_loss=0.127, over 5660989.51 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:39:06,169 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:39:20,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2973, 1.7612, 1.4083, 1.4658], device='cuda:0'), covar=tensor([0.0790, 0.0304, 0.0338, 0.0896], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:0')
+2023-03-08 05:39:26,179 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:39:30,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:39:41,091 INFO [train.py:968] (0/2) Epoch 16, batch 10200, giga_loss[loss=0.315, simple_loss=0.3561, pruned_loss=0.137, over 23648.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3711, pruned_loss=0.1215, over 5679864.92 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3497, pruned_loss=0.09506, over 5728347.26 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3743, pruned_loss=0.125, over 5659591.77 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:39:43,548 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:39:57,678 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:40:17,580 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:40:26,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.211e+02 1.449e+03 1.946e+03 2.910e+03 9.478e+03, threshold=3.891e+03, percent-clipped=11.0
+2023-03-08 05:40:28,240 INFO [train.py:968] (0/2) Epoch 16, batch 10250, giga_loss[loss=0.2713, simple_loss=0.3418, pruned_loss=0.1003, over 28810.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.369, pruned_loss=0.1186, over 5675550.90 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3498, pruned_loss=0.095, over 5733930.75 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3721, pruned_loss=0.1223, over 5652554.13 frames. ], batch size: 112, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:40:55,151 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-08 05:41:19,157 INFO [train.py:968] (0/2) Epoch 16, batch 10300, giga_loss[loss=0.2827, simple_loss=0.3598, pruned_loss=0.1028, over 28731.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3664, pruned_loss=0.1163, over 5675880.38 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3498, pruned_loss=0.095, over 5735893.49 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3694, pruned_loss=0.1198, over 5654272.27 frames. ], batch size: 242, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:42:03,112 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:42:05,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.403e+02 1.431e+03 1.782e+03 2.629e+03 6.546e+03, threshold=3.565e+03, percent-clipped=6.0
+2023-03-08 05:42:06,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:42:06,411 INFO [train.py:968] (0/2) Epoch 16, batch 10350, giga_loss[loss=0.285, simple_loss=0.354, pruned_loss=0.108, over 28710.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3665, pruned_loss=0.1163, over 5670602.24 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3497, pruned_loss=0.09495, over 5727328.03 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3694, pruned_loss=0.1198, over 5658061.93 frames. ], batch size: 85, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:42:36,557 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:42:39,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:42:43,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:43:01,739 INFO [train.py:968] (0/2) Epoch 16, batch 10400, giga_loss[loss=0.2896, simple_loss=0.3496, pruned_loss=0.1149, over 28435.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3635, pruned_loss=0.1155, over 5667739.55 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3493, pruned_loss=0.09466, over 5729861.49 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3665, pruned_loss=0.119, over 5654057.26 frames. ], batch size: 336, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 05:43:05,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:43:11,549 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:43:21,370 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:43:51,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.622e+03 2.234e+03 2.867e+03 8.431e+03, threshold=4.468e+03, percent-clipped=13.0
+2023-03-08 05:43:51,526 INFO [train.py:968] (0/2) Epoch 16, batch 10450, giga_loss[loss=0.2722, simple_loss=0.3579, pruned_loss=0.09328, over 29008.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.363, pruned_loss=0.1159, over 5675599.57 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3488, pruned_loss=0.0944, over 5732450.62 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.366, pruned_loss=0.1192, over 5661947.43 frames. ], batch size: 164, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:44:03,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:44:06,328 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-08 05:44:15,726 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 05:44:24,632 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:44:35,608 INFO [train.py:968] (0/2) Epoch 16, batch 10500, giga_loss[loss=0.442, simple_loss=0.4493, pruned_loss=0.2173, over 26417.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3658, pruned_loss=0.117, over 5681696.05 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3477, pruned_loss=0.09379, over 5737442.20 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3697, pruned_loss=0.1209, over 5664692.05 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:45:26,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.511e+03 1.994e+03 2.753e+03 8.974e+03, threshold=3.988e+03, percent-clipped=6.0
+2023-03-08 05:45:26,035 INFO [train.py:968] (0/2) Epoch 16, batch 10550, giga_loss[loss=0.2845, simple_loss=0.3589, pruned_loss=0.105, over 28901.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3679, pruned_loss=0.1182, over 5657347.67 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.348, pruned_loss=0.09389, over 5728012.76 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3712, pruned_loss=0.1217, over 5651507.31 frames. ], batch size: 145, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:46:15,793 INFO [train.py:968] (0/2) Epoch 16, batch 10600, giga_loss[loss=0.2823, simple_loss=0.3565, pruned_loss=0.104, over 28896.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3676, pruned_loss=0.1179, over 5645534.62 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3483, pruned_loss=0.09395, over 5732581.27 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3703, pruned_loss=0.1213, over 5634935.47 frames. ], batch size: 199, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:46:16,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6416, 1.7609, 1.5889, 1.6120], device='cuda:0'), covar=tensor([0.1483, 0.2110, 0.1982, 0.1850], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0745, 0.0697, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 05:46:22,428 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:46:25,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:46:36,880 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-08 05:46:38,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5591, 3.4140, 3.2091, 1.6231], device='cuda:0'), covar=tensor([0.0756, 0.0790, 0.0777, 0.2418], device='cuda:0'), in_proj_covar=tensor([0.1160, 0.1070, 0.0926, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 05:46:53,555 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:47:04,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.751e+02 1.369e+03 1.807e+03 2.707e+03 9.180e+03, threshold=3.613e+03, percent-clipped=11.0
+2023-03-08 05:47:04,376 INFO [train.py:968] (0/2) Epoch 16, batch 10650, giga_loss[loss=0.2829, simple_loss=0.3616, pruned_loss=0.1022, over 29032.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3679, pruned_loss=0.1188, over 5649622.63 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3477, pruned_loss=0.09348, over 5736291.45 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3713, pruned_loss=0.1225, over 5635439.13 frames. ], batch size: 155, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:47:39,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:47:54,382 INFO [train.py:968] (0/2) Epoch 16, batch 10700, giga_loss[loss=0.3372, simple_loss=0.3928, pruned_loss=0.1409, over 28965.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.369, pruned_loss=0.1197, over 5631340.05 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3477, pruned_loss=0.09344, over 5720387.59 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3724, pruned_loss=0.1237, over 5630551.23 frames. ], batch size: 227, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:48:02,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7497, 2.4392, 1.8138, 0.8854], device='cuda:0'), covar=tensor([0.4522, 0.2226, 0.3260, 0.5114], device='cuda:0'), in_proj_covar=tensor([0.1664, 0.1574, 0.1544, 0.1359], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 05:48:42,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.615e+03 2.219e+03 3.202e+03 6.068e+03, threshold=4.438e+03, percent-clipped=19.0
+2023-03-08 05:48:42,981 INFO [train.py:968] (0/2) Epoch 16, batch 10750, giga_loss[loss=0.428, simple_loss=0.4443, pruned_loss=0.2058, over 27487.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3701, pruned_loss=0.1196, over 5645559.11 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3478, pruned_loss=0.09331, over 5724643.81 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3735, pruned_loss=0.1237, over 5638764.62 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:49:12,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:49:26,819 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-08 05:49:27,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:49:29,164 INFO [train.py:968] (0/2) Epoch 16, batch 10800, giga_loss[loss=0.2667, simple_loss=0.3438, pruned_loss=0.09473, over 28470.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3708, pruned_loss=0.12, over 5657263.09 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3476, pruned_loss=0.09311, over 5728288.23 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3746, pruned_loss=0.1246, over 5645871.71 frames. ], batch size: 71, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 05:50:17,956 INFO [train.py:968] (0/2) Epoch 16, batch 10850, giga_loss[loss=0.3283, simple_loss=0.366, pruned_loss=0.1453, over 23634.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3734, pruned_loss=0.1224, over 5645215.43 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3478, pruned_loss=0.09308, over 5719883.48 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3768, pruned_loss=0.1268, over 5642222.91 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:50:18,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.868e+02 1.697e+03 2.088e+03 2.886e+03 7.770e+03, threshold=4.176e+03, percent-clipped=8.0
+2023-03-08 05:50:28,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:51:08,318 INFO [train.py:968] (0/2) Epoch 16, batch 10900, giga_loss[loss=0.3109, simple_loss=0.3836, pruned_loss=0.1191, over 28915.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.375, pruned_loss=0.1237, over 5643146.13 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3482, pruned_loss=0.09331, over 5714047.76 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.378, pruned_loss=0.1276, over 5643930.13 frames. ], batch size: 227, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:51:09,398 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 05:51:25,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3423, 3.1424, 2.9889, 1.3842], device='cuda:0'), covar=tensor([0.0977, 0.1165, 0.1165, 0.2363], device='cuda:0'), in_proj_covar=tensor([0.1160, 0.1070, 0.0926, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 05:51:32,726 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:51:35,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:51:43,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694884.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:51:53,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:51:56,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:52:03,040 INFO [train.py:968] (0/2) Epoch 16, batch 10950, giga_loss[loss=0.3696, simple_loss=0.4143, pruned_loss=0.1624, over 27513.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3747, pruned_loss=0.1225, over 5644570.43 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.348, pruned_loss=0.09326, over 5717317.70 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3776, pruned_loss=0.1261, over 5641190.53 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:52:04,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.674e+03 2.218e+03 3.237e+03 6.526e+03, threshold=4.436e+03, percent-clipped=16.0
+2023-03-08 05:52:07,452 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:52:29,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:52:57,391 INFO [train.py:968] (0/2) Epoch 16, batch 11000, giga_loss[loss=0.2986, simple_loss=0.3591, pruned_loss=0.1191, over 28871.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3731, pruned_loss=0.1218, over 5650761.29 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3481, pruned_loss=0.09329, over 5719086.51 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3758, pruned_loss=0.1252, over 5645028.42 frames. ], batch size: 112, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:53:01,333 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:53:04,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:53:35,464 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:53:48,319 INFO [train.py:968] (0/2) Epoch 16, batch 11050, giga_loss[loss=0.3299, simple_loss=0.3873, pruned_loss=0.1363, over 28925.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3717, pruned_loss=0.1213, over 5663726.92 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3478, pruned_loss=0.09314, over 5721391.67 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3745, pruned_loss=0.1247, over 5655895.75 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:53:51,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.456e+02 1.771e+03 2.509e+03 3.624e+03 9.665e+03, threshold=5.018e+03, percent-clipped=16.0
+2023-03-08 05:54:01,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=695010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:54:46,352 INFO [train.py:968] (0/2) Epoch 16, batch 11100, giga_loss[loss=0.2677, simple_loss=0.3346, pruned_loss=0.1004, over 28408.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3704, pruned_loss=0.1209, over 5652940.99 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3476, pruned_loss=0.093, over 5718792.05 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3736, pruned_loss=0.1247, over 5647533.34 frames. ], batch size: 78, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:54:48,255 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 05:55:24,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6138, 1.8678, 1.5277, 1.8075], device='cuda:0'), covar=tensor([0.2303, 0.2455, 0.2699, 0.2231], device='cuda:0'), in_proj_covar=tensor([0.1400, 0.1022, 0.1238, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 05:55:32,513 INFO [train.py:968] (0/2) Epoch 16, batch 11150, giga_loss[loss=0.3046, simple_loss=0.3694, pruned_loss=0.1199, over 28647.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3689, pruned_loss=0.1197, over 5665768.83 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.348, pruned_loss=0.0932, over 5716864.38 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3718, pruned_loss=0.1233, over 5661690.18 frames. ], batch size: 242, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:55:34,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.635e+02 1.508e+03 1.969e+03 2.912e+03 8.763e+03, threshold=3.939e+03, percent-clipped=7.0
+2023-03-08 05:56:21,204 INFO [train.py:968] (0/2) Epoch 16, batch 11200, giga_loss[loss=0.2586, simple_loss=0.3444, pruned_loss=0.08641, over 28998.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3697, pruned_loss=0.1211, over 5663975.18 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3477, pruned_loss=0.09312, over 5720276.32 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3727, pruned_loss=0.1245, over 5656594.62 frames. ], batch size: 155, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:56:23,335 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=695153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:56:27,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=695156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:56:53,937 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=695185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:56:56,463 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2306, 2.0291, 1.5542, 0.4511], device='cuda:0'), covar=tensor([0.4011, 0.2019, 0.3011, 0.4172], device='cuda:0'), in_proj_covar=tensor([0.1660, 0.1575, 0.1547, 0.1362], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 05:57:10,834 INFO [train.py:968] (0/2) Epoch 16, batch 11250, giga_loss[loss=0.2881, simple_loss=0.3618, pruned_loss=0.1072, over 29000.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.37, pruned_loss=0.1213, over 5674290.17 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3476, pruned_loss=0.09291, over 5725686.26 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3731, pruned_loss=0.125, over 5662117.22 frames. ], batch size: 136, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:57:12,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.813e+02 1.477e+03 1.851e+03 2.487e+03 5.603e+03, threshold=3.701e+03, percent-clipped=6.0
+2023-03-08 05:57:23,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5391, 1.6004, 1.7821, 1.3953], device='cuda:0'), covar=tensor([0.1398, 0.2064, 0.1145, 0.1457], device='cuda:0'), in_proj_covar=tensor([0.0867, 0.0702, 0.0915, 0.0814], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 05:57:39,807 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-08 05:57:59,351 INFO [train.py:968] (0/2) Epoch 16, batch 11300, giga_loss[loss=0.3446, simple_loss=0.3962, pruned_loss=0.1465, over 28948.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3713, pruned_loss=0.122, over 5670362.88 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3479, pruned_loss=0.09296, over 5723036.21 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3742, pruned_loss=0.1258, over 5661896.51 frames. ], batch size: 227, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:58:08,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=695257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:58:09,871 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=695259.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:58:48,968 INFO [train.py:968] (0/2) Epoch 16, batch 11350, giga_loss[loss=0.3416, simple_loss=0.3998, pruned_loss=0.1418, over 29014.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3736, pruned_loss=0.1241, over 5671238.11 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3475, pruned_loss=0.09276, over 5726709.13 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3767, pruned_loss=0.1279, over 5660079.26 frames. ], batch size: 155, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:58:50,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.212e+03 1.727e+03 2.177e+03 3.188e+03 7.776e+03, threshold=4.354e+03, percent-clipped=18.0
+2023-03-08 05:58:55,378 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2674, 1.2496, 3.7816, 3.1969], device='cuda:0'), covar=tensor([0.1596, 0.2688, 0.0462, 0.1068], device='cuda:0'), in_proj_covar=tensor([0.0715, 0.0624, 0.0917, 0.0837], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 05:59:34,693 INFO [train.py:968] (0/2) Epoch 16, batch 11400, giga_loss[loss=0.3435, simple_loss=0.3975, pruned_loss=0.1447, over 27839.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3738, pruned_loss=0.1241, over 5677056.12 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3476, pruned_loss=0.09279, over 5732920.07 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3774, pruned_loss=0.1284, over 5660030.91 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:59:58,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5702, 1.7788, 1.4469, 1.7995], device='cuda:0'), covar=tensor([0.2125, 0.2224, 0.2313, 0.2072], device='cuda:0'), in_proj_covar=tensor([0.1409, 0.1031, 0.1246, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 06:00:25,319 INFO [train.py:968] (0/2) Epoch 16, batch 11450, giga_loss[loss=0.3383, simple_loss=0.3779, pruned_loss=0.1494, over 23612.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3741, pruned_loss=0.1255, over 5666922.07 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.347, pruned_loss=0.09246, over 5737678.62 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3782, pruned_loss=0.1303, over 5647049.52 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:00:26,332 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=695402.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 06:00:26,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.685e+03 2.202e+03 2.781e+03 1.214e+04, threshold=4.403e+03, percent-clipped=9.0
+2023-03-08 06:00:29,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=695405.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 06:00:34,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3603, 1.6073, 1.6439, 1.2158], device='cuda:0'), covar=tensor([0.1496, 0.2271, 0.1255, 0.1490], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0701, 0.0913, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 06:00:35,619 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 06:00:58,422 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=695434.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 06:01:16,193 INFO [train.py:968] (0/2) Epoch 16, batch 11500, giga_loss[loss=0.3451, simple_loss=0.3922, pruned_loss=0.149, over 27953.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3744, pruned_loss=0.1261, over 5665803.56 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3473, pruned_loss=0.09264, over 5738359.23 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3776, pruned_loss=0.1298, over 5649336.70 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:02:06,121 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 06:02:06,370 INFO [train.py:968] (0/2) Epoch 16, batch 11550, giga_loss[loss=0.3272, simple_loss=0.3876, pruned_loss=0.1334, over 28784.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3735, pruned_loss=0.1247, over 5674519.01 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3474, pruned_loss=0.09256, over 5740567.64 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3764, pruned_loss=0.1283, over 5658399.64 frames. ], batch size: 99, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:02:08,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.637e+03 1.928e+03 2.649e+03 7.648e+03, threshold=3.857e+03, percent-clipped=8.0
+2023-03-08 06:02:48,957 INFO [train.py:968] (0/2) Epoch 16, batch 11600, giga_loss[loss=0.2827, simple_loss=0.3583, pruned_loss=0.1036, over 28775.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3741, pruned_loss=0.1245, over 5671996.31 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3479, pruned_loss=0.09296, over 5736669.55 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3768, pruned_loss=0.128, over 5660466.37 frames. ], batch size: 243, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:03:41,428 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0837, 3.2918, 2.1198, 1.0882], device='cuda:0'), covar=tensor([0.5813, 0.2350, 0.3078, 0.5644], device='cuda:0'), in_proj_covar=tensor([0.1660, 0.1577, 0.1547, 0.1361], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 06:03:45,639 INFO [train.py:968] (0/2) Epoch 16, batch 11650, giga_loss[loss=0.3019, simple_loss=0.3673, pruned_loss=0.1183, over 28776.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3752, pruned_loss=0.1252, over 5680767.77 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.348, pruned_loss=0.09298, over 5737529.20 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3774, pruned_loss=0.1281, over 5670713.37 frames. ], batch size: 284, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:03:46,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.068e+02 1.640e+03 2.028e+03 2.968e+03 6.023e+03, threshold=4.056e+03, percent-clipped=10.0
+2023-03-08 06:03:49,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4148, 3.6796, 1.6093, 1.5640], device='cuda:0'), covar=tensor([0.0959, 0.0349, 0.0838, 0.1308], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0536, 0.0360, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 06:04:12,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5408, 1.2514, 4.4718, 3.4885], device='cuda:0'), covar=tensor([0.1599, 0.2812, 0.0426, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.0712, 0.0619, 0.0912, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 06:04:19,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=695632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:04:26,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6660, 2.2950, 1.9864, 1.8892], device='cuda:0'), covar=tensor([0.0743, 0.0243, 0.0277, 0.0789], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0115, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:0')
+2023-03-08 06:04:38,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0244, 2.0024, 1.7968, 1.7192], device='cuda:0'), covar=tensor([0.1738, 0.2484, 0.2273, 0.2354], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0745, 0.0700, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 06:04:38,478 INFO [train.py:968] (0/2) Epoch 16, batch 11700, giga_loss[loss=0.2783, simple_loss=0.357, pruned_loss=0.09979, over 28879.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3769, pruned_loss=0.1271, over 5676850.30 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3481, pruned_loss=0.09301, over 5738670.52 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3788, pruned_loss=0.1296, over 5667391.98 frames. ], batch size: 174, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:04:47,792 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=695659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:05:23,205 INFO [train.py:968] (0/2) Epoch 16, batch 11750, giga_loss[loss=0.3093, simple_loss=0.3778, pruned_loss=0.1204, over 28790.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3769, pruned_loss=0.1264, over 5690012.39 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3481, pruned_loss=0.09287, over 5744518.28 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3794, pruned_loss=0.1299, over 5674933.89 frames. ], batch size: 112, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 06:05:26,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.906e+02 1.711e+03 2.360e+03 3.922e+03 1.123e+04, threshold=4.719e+03, percent-clipped=22.0
+2023-03-08 06:05:45,583 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=695723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:05:56,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9846, 4.7843, 4.5336, 2.3653], device='cuda:0'), covar=tensor([0.0465, 0.0677, 0.0754, 0.1878], device='cuda:0'), in_proj_covar=tensor([0.1161, 0.1071, 0.0923, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 06:06:15,411 INFO [train.py:968] (0/2) Epoch 16, batch 11800, libri_loss[loss=0.2225, simple_loss=0.31, pruned_loss=0.06752, over 29379.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3766, pruned_loss=0.1254, over 5682461.76 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3477, pruned_loss=0.09267, over 5746726.01 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3794, pruned_loss=0.1288, over 5667807.82 frames. ], batch size: 71, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 06:06:37,071 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=695775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:06:39,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=695778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:07:05,735 INFO [train.py:968] (0/2) Epoch 16, batch 11850, giga_loss[loss=0.2879, simple_loss=0.3658, pruned_loss=0.105, over 28948.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3753, pruned_loss=0.1239, over 5675994.26 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3476, pruned_loss=0.09261, over 5748646.47 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3782, pruned_loss=0.1273, over 5661303.00 frames. ], batch size: 164, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 06:07:08,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.885e+02 1.445e+03 1.770e+03 2.295e+03 7.280e+03, threshold=3.540e+03, percent-clipped=2.0
+2023-03-08 06:07:10,680 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=695807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:07:55,358 INFO [train.py:968] (0/2) Epoch 16, batch 11900, giga_loss[loss=0.2914, simple_loss=0.3599, pruned_loss=0.1114, over 28508.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3735, pruned_loss=0.1225, over 5683170.88 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3478, pruned_loss=0.09274, over 5747603.05 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3759, pruned_loss=0.1254, over 5671325.30 frames. ], batch size: 60, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 06:08:33,291 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4820, 2.2825, 1.5964, 0.6093], device='cuda:0'), covar=tensor([0.4647, 0.2615, 0.3637, 0.5508], device='cuda:0'), in_proj_covar=tensor([0.1656, 0.1573, 0.1543, 0.1358], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 06:08:42,410 INFO [train.py:968] (0/2) Epoch 16, batch 11950, giga_loss[loss=0.2845, simple_loss=0.353, pruned_loss=0.108, over 28774.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3736, pruned_loss=0.123, over 5679574.75 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3476, pruned_loss=0.09267, over 5751709.82 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3763, pruned_loss=0.1261, over 5665210.54 frames. ], batch size: 99, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 06:08:48,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.399e+02 1.553e+03 2.035e+03 2.493e+03 7.071e+03, threshold=4.071e+03, percent-clipped=9.0
+2023-03-08 06:09:35,571 INFO [train.py:968] (0/2) Epoch 16, batch 12000, giga_loss[loss=0.3092, simple_loss=0.3701, pruned_loss=0.1241, over 28939.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.375, pruned_loss=0.1238, over 5675205.22 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3478, pruned_loss=0.09275, over 5750630.68 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3774, pruned_loss=0.1266, over 5663832.43 frames. ], batch size: 106, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:09:35,579 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 06:09:44,048 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9776, 3.7306, 3.5509, 1.6625], device='cuda:0'), covar=tensor([0.0626, 0.0908, 0.0743, 0.2487], device='cuda:0'), in_proj_covar=tensor([0.1165, 0.1074, 0.0926, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 06:09:44,945 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2134, simple_loss=0.3206, pruned_loss=0.05314, over 944034.00 frames. 
+2023-03-08 06:09:44,946 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 06:10:00,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8833, 1.9754, 2.1339, 1.6295], device='cuda:0'), covar=tensor([0.1780, 0.2302, 0.1360, 0.1625], device='cuda:0'), in_proj_covar=tensor([0.0868, 0.0702, 0.0914, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 06:10:19,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-08 06:10:28,337 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-696000.pt
+2023-03-08 06:10:29,227 INFO [train.py:968] (0/2) Epoch 16, batch 12050, giga_loss[loss=0.338, simple_loss=0.3924, pruned_loss=0.1418, over 28928.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3759, pruned_loss=0.125, over 5647104.02 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3481, pruned_loss=0.09317, over 5722200.26 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3782, pruned_loss=0.1277, over 5662959.38 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:10:31,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.526e+03 2.051e+03 2.938e+03 5.950e+03, threshold=4.101e+03, percent-clipped=9.0
+2023-03-08 06:10:47,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6732, 1.7454, 1.6931, 1.5061], device='cuda:0'), covar=tensor([0.1567, 0.2052, 0.2134, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0746, 0.0699, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 06:11:00,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=696034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:11:11,567 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:11:19,061 INFO [train.py:968] (0/2) Epoch 16, batch 12100, giga_loss[loss=0.388, simple_loss=0.4173, pruned_loss=0.1794, over 27635.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3759, pruned_loss=0.1259, over 5654428.57 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.348, pruned_loss=0.09311, over 5728626.08 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.379, pruned_loss=0.1295, over 5658609.62 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:12:09,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=696098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:12:10,902 INFO [train.py:968] (0/2) Epoch 16, batch 12150, libri_loss[loss=0.2952, simple_loss=0.3808, pruned_loss=0.1048, over 27852.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3771, pruned_loss=0.127, over 5659595.33 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3483, pruned_loss=0.09325, over 5729822.10 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3797, pruned_loss=0.1302, over 5660949.03 frames. ], batch size: 116, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:12:13,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.540e+03 1.975e+03 3.144e+03 7.746e+03, threshold=3.950e+03, percent-clipped=9.0
+2023-03-08 06:13:00,544 INFO [train.py:968] (0/2) Epoch 16, batch 12200, giga_loss[loss=0.3123, simple_loss=0.3767, pruned_loss=0.124, over 28668.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3775, pruned_loss=0.1274, over 5653824.62 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3487, pruned_loss=0.09374, over 5732730.86 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3801, pruned_loss=0.1304, over 5650390.15 frames. ], batch size: 284, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:13:17,058 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:13:24,557 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=696177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:13:28,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=696180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:13:34,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6119, 1.8606, 1.4862, 1.6029], device='cuda:0'), covar=tensor([0.2437, 0.2486, 0.2788, 0.2270], device='cuda:0'), in_proj_covar=tensor([0.1410, 0.1029, 0.1248, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 06:13:47,939 INFO [train.py:968] (0/2) Epoch 16, batch 12250, giga_loss[loss=0.3727, simple_loss=0.4103, pruned_loss=0.1676, over 27552.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3785, pruned_loss=0.1281, over 5655210.46 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3489, pruned_loss=0.09379, over 5735336.13 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3809, pruned_loss=0.1311, over 5648689.58 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:13:51,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.677e+02 1.597e+03 1.994e+03 2.549e+03 5.165e+03, threshold=3.988e+03, percent-clipped=7.0
+2023-03-08 06:13:56,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=696209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:14:12,683 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:14:28,126 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=696241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:14:33,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=696244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:14:39,196 INFO [train.py:968] (0/2) Epoch 16, batch 12300, libri_loss[loss=0.3198, simple_loss=0.3931, pruned_loss=0.1233, over 26167.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3793, pruned_loss=0.1292, over 5640075.64 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3491, pruned_loss=0.09385, over 5736071.99 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3817, pruned_loss=0.1323, over 5632487.36 frames. ], batch size: 136, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:15:00,238 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=696273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:15:25,818 INFO [train.py:968] (0/2) Epoch 16, batch 12350, giga_loss[loss=0.2709, simple_loss=0.3517, pruned_loss=0.09501, over 28754.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3792, pruned_loss=0.1284, over 5651273.42 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3494, pruned_loss=0.09384, over 5740573.73 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3817, pruned_loss=0.1319, over 5638643.23 frames. ], batch size: 119, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:15:29,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.145e+02 1.502e+03 1.989e+03 2.567e+03 8.383e+03, threshold=3.978e+03, percent-clipped=4.0
+2023-03-08 06:16:11,941 INFO [train.py:968] (0/2) Epoch 16, batch 12400, giga_loss[loss=0.3226, simple_loss=0.3856, pruned_loss=0.1298, over 28866.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3787, pruned_loss=0.1278, over 5644675.50 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3494, pruned_loss=0.09392, over 5734008.44 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3812, pruned_loss=0.1311, over 5639672.31 frames. ], batch size: 199, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:16:26,043 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-08 06:17:00,936 INFO [train.py:968] (0/2) Epoch 16, batch 12450, giga_loss[loss=0.2568, simple_loss=0.3376, pruned_loss=0.08802, over 28950.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3773, pruned_loss=0.1268, over 5650572.71 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3493, pruned_loss=0.09381, over 5733204.29 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3802, pruned_loss=0.1304, over 5645079.17 frames. ], batch size: 145, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:17:05,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.635e+03 2.120e+03 2.876e+03 7.097e+03, threshold=4.239e+03, percent-clipped=9.0
+2023-03-08 06:17:17,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=696418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:17:39,160 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2587, 2.7336, 2.2476, 1.9187], device='cuda:0'), covar=tensor([0.2527, 0.1721, 0.2159, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.1833, 0.1763, 0.1705, 0.1827], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 06:17:48,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8766, 2.1434, 1.6823, 2.3163], device='cuda:0'), covar=tensor([0.2672, 0.2596, 0.2960, 0.2269], device='cuda:0'), in_proj_covar=tensor([0.1408, 0.1032, 0.1249, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 06:17:51,747 INFO [train.py:968] (0/2) Epoch 16, batch 12500, giga_loss[loss=0.2587, simple_loss=0.3289, pruned_loss=0.09432, over 28417.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3741, pruned_loss=0.1249, over 5661220.70 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.349, pruned_loss=0.09378, over 5736777.41 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3771, pruned_loss=0.1285, over 5652237.67 frames. ], batch size: 60, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:18:06,620 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-08 06:18:34,263 INFO [train.py:968] (0/2) Epoch 16, batch 12550, giga_loss[loss=0.2782, simple_loss=0.3471, pruned_loss=0.1047, over 28938.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3709, pruned_loss=0.1225, over 5674921.67 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3494, pruned_loss=0.094, over 5735206.58 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.374, pruned_loss=0.1264, over 5666096.04 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:18:39,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.511e+02 1.916e+03 2.496e+03 3.805e+03 1.293e+04, threshold=4.991e+03, percent-clipped=19.0
+2023-03-08 06:19:01,394 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.88 vs. limit=5.0
+2023-03-08 06:19:18,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=696543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:19:25,932 INFO [train.py:968] (0/2) Epoch 16, batch 12600, giga_loss[loss=0.3078, simple_loss=0.365, pruned_loss=0.1254, over 28792.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3677, pruned_loss=0.1216, over 5651784.74 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3495, pruned_loss=0.09405, over 5737959.97 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3703, pruned_loss=0.125, over 5641271.01 frames. ], batch size: 186, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:19:37,430 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=696561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:19:40,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=696564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:19:53,614 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-08 06:20:10,214 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=696593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:20:14,191 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=696598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:20:16,050 INFO [train.py:968] (0/2) Epoch 16, batch 12650, libri_loss[loss=0.2823, simple_loss=0.367, pruned_loss=0.09885, over 29249.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3696, pruned_loss=0.1239, over 5658583.68 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3499, pruned_loss=0.09415, over 5740327.13 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3715, pruned_loss=0.127, over 5647013.09 frames. ], batch size: 97, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:20:18,213 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 06:20:21,388 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.645e+03 2.031e+03 2.956e+03 5.337e+03, threshold=4.061e+03, percent-clipped=2.0
+2023-03-08 06:21:01,936 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:21:07,069 INFO [train.py:968] (0/2) Epoch 16, batch 12700, giga_loss[loss=0.3271, simple_loss=0.38, pruned_loss=0.1371, over 27619.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3673, pruned_loss=0.1224, over 5656650.85 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3501, pruned_loss=0.09444, over 5743072.01 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3689, pruned_loss=0.125, over 5643691.58 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:21:23,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4657, 1.6405, 1.2773, 1.2656], device='cuda:0'), covar=tensor([0.0932, 0.0515, 0.1042, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0449, 0.0507, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 06:21:41,089 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=696686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:21:43,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=696689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:21:50,743 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3448, 1.8491, 1.3201, 0.4902], device='cuda:0'), covar=tensor([0.3218, 0.2419, 0.3605, 0.4639], device='cuda:0'), in_proj_covar=tensor([0.1656, 0.1576, 0.1544, 0.1360], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 06:21:54,267 INFO [train.py:968] (0/2) Epoch 16, batch 12750, giga_loss[loss=0.2745, simple_loss=0.3542, pruned_loss=0.09742, over 28716.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3651, pruned_loss=0.1189, over 5653398.84 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3502, pruned_loss=0.09456, over 5739670.38 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3671, pruned_loss=0.1222, over 5642535.54 frames. ], batch size: 262, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:21:58,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.648e+02 1.515e+03 1.904e+03 3.176e+03 8.749e+03, threshold=3.808e+03, percent-clipped=5.0
+2023-03-08 06:22:09,703 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=696718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:22:27,704 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:22:31,503 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=696741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:22:33,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2432, 2.9540, 1.3950, 1.4359], device='cuda:0'), covar=tensor([0.0987, 0.0341, 0.0973, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0540, 0.0363, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 06:22:35,924 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=696744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:22:42,937 INFO [train.py:968] (0/2) Epoch 16, batch 12800, libri_loss[loss=0.2721, simple_loss=0.3576, pruned_loss=0.09326, over 29654.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1157, over 5655775.91 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3503, pruned_loss=0.09474, over 5738549.42 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3657, pruned_loss=0.1189, over 5645172.01 frames. ], batch size: 91, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:23:04,093 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=696773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:23:07,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1983, 1.1563, 3.4269, 3.1438], device='cuda:0'), covar=tensor([0.1605, 0.2725, 0.0510, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0714, 0.0623, 0.0914, 0.0837], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 06:23:31,486 INFO [train.py:968] (0/2) Epoch 16, batch 12850, giga_loss[loss=0.2781, simple_loss=0.3541, pruned_loss=0.101, over 28769.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3602, pruned_loss=0.1122, over 5653272.22 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.349, pruned_loss=0.09428, over 5736645.53 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.116, over 5642790.17 frames. ], batch size: 284, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:23:39,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.111e+02 1.443e+03 1.745e+03 2.083e+03 7.648e+03, threshold=3.490e+03, percent-clipped=4.0
+2023-03-08 06:24:09,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7131, 1.8165, 1.3253, 1.4125], device='cuda:0'), covar=tensor([0.0845, 0.0545, 0.0958, 0.1054], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0445, 0.0507, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 06:24:26,197 INFO [train.py:968] (0/2) Epoch 16, batch 12900, giga_loss[loss=0.2927, simple_loss=0.363, pruned_loss=0.1112, over 28801.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3571, pruned_loss=0.1092, over 5649318.67 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.349, pruned_loss=0.09435, over 5735458.80 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3598, pruned_loss=0.1123, over 5640930.13 frames. ], batch size: 119, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:24:45,895 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6479, 1.6492, 1.3737, 1.3429], device='cuda:0'), covar=tensor([0.0656, 0.0362, 0.0723, 0.0980], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0444, 0.0505, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 06:25:18,595 INFO [train.py:968] (0/2) Epoch 16, batch 12950, giga_loss[loss=0.2526, simple_loss=0.3333, pruned_loss=0.08596, over 28555.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3536, pruned_loss=0.1059, over 5647818.21 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3487, pruned_loss=0.09426, over 5738817.75 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3561, pruned_loss=0.1087, over 5636407.12 frames. ], batch size: 336, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:25:23,781 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-08 06:25:24,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.375e+02 1.378e+03 1.820e+03 2.511e+03 4.399e+03, threshold=3.641e+03, percent-clipped=8.0
+2023-03-08 06:25:42,080 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:25:51,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2522, 1.5519, 1.5643, 1.1283], device='cuda:0'), covar=tensor([0.1795, 0.2682, 0.1499, 0.1778], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0695, 0.0910, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 06:26:07,674 INFO [train.py:968] (0/2) Epoch 16, batch 13000, libri_loss[loss=0.2446, simple_loss=0.3241, pruned_loss=0.08255, over 29571.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3525, pruned_loss=0.1027, over 5663636.77 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3482, pruned_loss=0.09418, over 5741935.20 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3552, pruned_loss=0.1054, over 5648451.75 frames. ], batch size: 79, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:27:03,008 INFO [train.py:968] (0/2) Epoch 16, batch 13050, giga_loss[loss=0.2685, simple_loss=0.3428, pruned_loss=0.09709, over 28010.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3535, pruned_loss=0.1033, over 5655049.82 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3481, pruned_loss=0.09409, over 5742989.82 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3558, pruned_loss=0.1056, over 5641662.44 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:27:10,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.430e+02 1.454e+03 2.157e+03 2.981e+03 8.634e+03, threshold=4.314e+03, percent-clipped=13.0
+2023-03-08 06:27:22,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:27:52,850 INFO [train.py:968] (0/2) Epoch 16, batch 13100, giga_loss[loss=0.285, simple_loss=0.3598, pruned_loss=0.1051, over 28312.00 frames. ], tot_loss[loss=0.278, simple_loss=0.352, pruned_loss=0.102, over 5660398.56 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3479, pruned_loss=0.09413, over 5747288.12 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3541, pruned_loss=0.104, over 5643627.26 frames. ], batch size: 368, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:28:14,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1700, 1.7126, 1.2240, 0.3913], device='cuda:0'), covar=tensor([0.3599, 0.2298, 0.3748, 0.4678], device='cuda:0'), in_proj_covar=tensor([0.1651, 0.1571, 0.1540, 0.1361], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 06:28:40,506 INFO [train.py:968] (0/2) Epoch 16, batch 13150, libri_loss[loss=0.2665, simple_loss=0.3443, pruned_loss=0.09432, over 29541.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3475, pruned_loss=0.09896, over 5664391.80 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3467, pruned_loss=0.09367, over 5754863.20 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3506, pruned_loss=0.1014, over 5638905.43 frames. ], batch size: 89, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:28:45,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.436e+02 1.324e+03 1.647e+03 2.246e+03 5.473e+03, threshold=3.294e+03, percent-clipped=3.0
+2023-03-08 06:28:51,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:29:30,313 INFO [train.py:968] (0/2) Epoch 16, batch 13200, libri_loss[loss=0.2275, simple_loss=0.3035, pruned_loss=0.07572, over 29482.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3468, pruned_loss=0.09893, over 5662631.61 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3462, pruned_loss=0.09361, over 5758390.03 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3496, pruned_loss=0.1011, over 5636329.89 frames. ], batch size: 70, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:29:44,748 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=697164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:29:48,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=697167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:30:14,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=697196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:30:14,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=697196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:30:19,534 INFO [train.py:968] (0/2) Epoch 16, batch 13250, giga_loss[loss=0.2328, simple_loss=0.3169, pruned_loss=0.07441, over 28762.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3459, pruned_loss=0.09827, over 5660189.33 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3457, pruned_loss=0.09344, over 5761205.56 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3488, pruned_loss=0.1003, over 5634340.55 frames. ], batch size: 119, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:30:21,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5927, 3.0297, 2.6121, 2.2670], device='cuda:0'), covar=tensor([0.2003, 0.1245, 0.1378, 0.1629], device='cuda:0'), in_proj_covar=tensor([0.1820, 0.1747, 0.1688, 0.1813], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 06:30:23,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.696e+02 1.260e+03 1.663e+03 2.505e+03 5.722e+03, threshold=3.327e+03, percent-clipped=12.0
+2023-03-08 06:31:06,688 INFO [train.py:968] (0/2) Epoch 16, batch 13300, giga_loss[loss=0.2457, simple_loss=0.3329, pruned_loss=0.07926, over 29184.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.344, pruned_loss=0.09655, over 5670727.58 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3449, pruned_loss=0.09322, over 5765975.04 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.347, pruned_loss=0.09847, over 5642880.72 frames. ], batch size: 128, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:31:12,457 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=697256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:31:16,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=697259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:31:51,239 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=697288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:32:03,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:32:04,027 INFO [train.py:968] (0/2) Epoch 16, batch 13350, giga_loss[loss=0.2253, simple_loss=0.3111, pruned_loss=0.06971, over 28988.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3409, pruned_loss=0.09422, over 5659502.16 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3445, pruned_loss=0.09311, over 5763944.27 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3435, pruned_loss=0.09584, over 5638499.45 frames. ], batch size: 128, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:32:10,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.606e+02 1.255e+03 1.710e+03 2.397e+03 4.984e+03, threshold=3.421e+03, percent-clipped=10.0
+2023-03-08 06:33:02,160 INFO [train.py:968] (0/2) Epoch 16, batch 13400, giga_loss[loss=0.2186, simple_loss=0.305, pruned_loss=0.06611, over 28757.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3377, pruned_loss=0.09243, over 5657918.85 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3445, pruned_loss=0.09314, over 5765843.37 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3396, pruned_loss=0.09367, over 5638046.76 frames. ], batch size: 284, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:33:02,932 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=697352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:34:01,578 INFO [train.py:968] (0/2) Epoch 16, batch 13450, giga_loss[loss=0.2821, simple_loss=0.352, pruned_loss=0.1061, over 28538.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3363, pruned_loss=0.09205, over 5660398.77 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3445, pruned_loss=0.09314, over 5765843.37 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3378, pruned_loss=0.09301, over 5644932.08 frames. ], batch size: 307, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:34:07,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.235e+02 1.186e+03 1.877e+03 2.452e+03 9.262e+03, threshold=3.754e+03, percent-clipped=15.0
+2023-03-08 06:34:38,263 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-08 06:34:46,650 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=697442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:34:50,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=697445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:34:55,294 INFO [train.py:968] (0/2) Epoch 16, batch 13500, libri_loss[loss=0.2053, simple_loss=0.2844, pruned_loss=0.06313, over 29512.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3352, pruned_loss=0.09193, over 5651580.78 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.344, pruned_loss=0.09279, over 5768438.53 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3368, pruned_loss=0.09301, over 5635079.82 frames. ], batch size: 70, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:35:24,452 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=697474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:35:42,378 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-08 06:35:56,536 INFO [train.py:968] (0/2) Epoch 16, batch 13550, giga_loss[loss=0.26, simple_loss=0.3443, pruned_loss=0.08788, over 28626.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3375, pruned_loss=0.09284, over 5657322.09 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3439, pruned_loss=0.09288, over 5770503.73 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3387, pruned_loss=0.09361, over 5640875.58 frames. ], batch size: 336, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:36:04,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.027e+02 1.372e+03 1.894e+03 2.842e+03 6.604e+03, threshold=3.788e+03, percent-clipped=10.0
+2023-03-08 06:36:15,944 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5163, 1.7559, 1.8538, 1.3813], device='cuda:0'), covar=tensor([0.1936, 0.2427, 0.1503, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0688, 0.0906, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 06:36:26,136 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=697524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:36:56,589 INFO [train.py:968] (0/2) Epoch 16, batch 13600, giga_loss[loss=0.2348, simple_loss=0.3211, pruned_loss=0.07422, over 28896.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3399, pruned_loss=0.09306, over 5649518.56 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3437, pruned_loss=0.09285, over 5762995.22 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.341, pruned_loss=0.09371, over 5640780.54 frames. ], batch size: 227, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:36:59,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1611, 2.5208, 1.2589, 1.3770], device='cuda:0'), covar=tensor([0.0986, 0.0330, 0.0944, 0.1351], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0532, 0.0360, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 06:37:23,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:37:58,429 INFO [train.py:968] (0/2) Epoch 16, batch 13650, giga_loss[loss=0.2342, simple_loss=0.3223, pruned_loss=0.07304, over 28898.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3402, pruned_loss=0.09317, over 5657624.60 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3433, pruned_loss=0.09279, over 5760368.64 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3414, pruned_loss=0.09376, over 5651289.10 frames. ], batch size: 106, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:38:07,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4681, 1.6318, 1.7670, 1.3343], device='cuda:0'), covar=tensor([0.1979, 0.2556, 0.1578, 0.1848], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0689, 0.0908, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 06:38:08,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.701e+02 1.517e+03 2.100e+03 2.845e+03 7.091e+03, threshold=4.200e+03, percent-clipped=10.0
+2023-03-08 06:38:26,670 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=697621.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 06:39:01,525 INFO [train.py:968] (0/2) Epoch 16, batch 13700, giga_loss[loss=0.2517, simple_loss=0.3244, pruned_loss=0.08955, over 27651.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3383, pruned_loss=0.09212, over 5669441.39 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3428, pruned_loss=0.09281, over 5766320.73 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3395, pruned_loss=0.09256, over 5655211.76 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:40:05,102 INFO [train.py:968] (0/2) Epoch 16, batch 13750, giga_loss[loss=0.2436, simple_loss=0.3291, pruned_loss=0.07909, over 28374.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3368, pruned_loss=0.09029, over 5670754.48 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3427, pruned_loss=0.09275, over 5769179.13 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3378, pruned_loss=0.09065, over 5655191.65 frames. ], batch size: 368, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:40:16,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.287e+02 1.266e+03 1.568e+03 2.271e+03 6.845e+03, threshold=3.136e+03, percent-clipped=2.0
+2023-03-08 06:40:23,376 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=697714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:40:26,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=697717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:40:36,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:40:59,388 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=697746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:41:05,193 INFO [train.py:968] (0/2) Epoch 16, batch 13800, giga_loss[loss=0.2416, simple_loss=0.322, pruned_loss=0.08064, over 28597.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3355, pruned_loss=0.08859, over 5679704.09 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.342, pruned_loss=0.09243, over 5773132.23 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3368, pruned_loss=0.08907, over 5661121.81 frames. ], batch size: 307, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:42:07,803 INFO [train.py:968] (0/2) Epoch 16, batch 13850, giga_loss[loss=0.2395, simple_loss=0.3131, pruned_loss=0.08298, over 27558.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3334, pruned_loss=0.08878, over 5676212.94 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3418, pruned_loss=0.09248, over 5777241.74 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3343, pruned_loss=0.08901, over 5654436.35 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:42:17,905 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.436e+02 1.507e+03 2.002e+03 2.988e+03 9.317e+03, threshold=4.004e+03, percent-clipped=22.0
+2023-03-08 06:42:26,487 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=697814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:43:04,111 INFO [train.py:968] (0/2) Epoch 16, batch 13900, libri_loss[loss=0.2025, simple_loss=0.2817, pruned_loss=0.0616, over 28473.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3331, pruned_loss=0.08903, over 5682140.01 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3412, pruned_loss=0.09233, over 5780177.47 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3341, pruned_loss=0.08922, over 5657606.20 frames. ], batch size: 63, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:43:26,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=697870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:43:29,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8903, 5.2945, 2.1067, 2.3124], device='cuda:0'), covar=tensor([0.0896, 0.0158, 0.0856, 0.1124], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0532, 0.0362, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 06:43:29,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=697873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:44:01,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:44:03,552 INFO [train.py:968] (0/2) Epoch 16, batch 13950, giga_loss[loss=0.2911, simple_loss=0.367, pruned_loss=0.1076, over 28920.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3323, pruned_loss=0.08893, over 5676828.77 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3407, pruned_loss=0.09216, over 5781231.08 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3334, pruned_loss=0.08918, over 5654642.93 frames. ], batch size: 284, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:44:05,975 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=697902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:44:12,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.153e+02 1.349e+03 1.704e+03 2.128e+03 1.331e+04, threshold=3.407e+03, percent-clipped=8.0
+2023-03-08 06:45:02,811 INFO [train.py:968] (0/2) Epoch 16, batch 14000, giga_loss[loss=0.2583, simple_loss=0.3474, pruned_loss=0.08462, over 28767.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3337, pruned_loss=0.08888, over 5673116.32 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3404, pruned_loss=0.09201, over 5783490.78 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3346, pruned_loss=0.08912, over 5649805.21 frames. ], batch size: 262, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:46:00,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697996.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 06:46:05,052 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-698000.pt
+2023-03-08 06:46:06,615 INFO [train.py:968] (0/2) Epoch 16, batch 14050, giga_loss[loss=0.2414, simple_loss=0.3241, pruned_loss=0.0794, over 28928.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3345, pruned_loss=0.08866, over 5667556.25 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3404, pruned_loss=0.09221, over 5781598.78 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3351, pruned_loss=0.08863, over 5648668.20 frames. ], batch size: 186, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:46:17,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.407e+02 1.329e+03 1.642e+03 2.228e+03 4.780e+03, threshold=3.283e+03, percent-clipped=6.0
+2023-03-08 06:47:03,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=698042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:47:07,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=698045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:47:12,965 INFO [train.py:968] (0/2) Epoch 16, batch 14100, giga_loss[loss=0.2296, simple_loss=0.308, pruned_loss=0.07556, over 28896.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3323, pruned_loss=0.08742, over 5666112.62 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3404, pruned_loss=0.09226, over 5768284.09 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3326, pruned_loss=0.08727, over 5661354.94 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:47:43,801 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=698074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:48:05,536 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=698090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:48:18,251 INFO [train.py:968] (0/2) Epoch 16, batch 14150, giga_loss[loss=0.2759, simple_loss=0.3482, pruned_loss=0.1018, over 28940.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3349, pruned_loss=0.08935, over 5673524.92 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3402, pruned_loss=0.09228, over 5764276.62 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3352, pruned_loss=0.08909, over 5669713.33 frames. ], batch size: 155, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:48:30,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.601e+02 1.344e+03 1.789e+03 2.581e+03 6.243e+03, threshold=3.579e+03, percent-clipped=13.0
+2023-03-08 06:49:15,371 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=698139.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 06:49:17,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=698142.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 06:49:27,968 INFO [train.py:968] (0/2) Epoch 16, batch 14200, giga_loss[loss=0.2326, simple_loss=0.333, pruned_loss=0.06607, over 29055.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3381, pruned_loss=0.08965, over 5675182.49 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3395, pruned_loss=0.09188, over 5767981.40 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3389, pruned_loss=0.08972, over 5665704.19 frames. ], batch size: 120, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:49:34,713 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 06:49:53,558 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=698171.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 06:50:15,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=698189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:50:28,181 INFO [train.py:968] (0/2) Epoch 16, batch 14250, giga_loss[loss=0.2653, simple_loss=0.3502, pruned_loss=0.09021, over 28024.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3406, pruned_loss=0.08895, over 5678095.24 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3394, pruned_loss=0.09181, over 5770986.44 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3414, pruned_loss=0.08901, over 5665097.09 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:50:38,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.858e+02 1.352e+03 1.698e+03 2.463e+03 6.883e+03, threshold=3.397e+03, percent-clipped=8.0
+2023-03-08 06:50:48,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7406, 1.8648, 1.5527, 2.0449], device='cuda:0'), covar=tensor([0.2557, 0.2703, 0.2919, 0.2420], device='cuda:0'), in_proj_covar=tensor([0.1409, 0.1028, 0.1249, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 06:51:31,975 INFO [train.py:968] (0/2) Epoch 16, batch 14300, giga_loss[loss=0.2775, simple_loss=0.3579, pruned_loss=0.09856, over 28146.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.341, pruned_loss=0.08776, over 5677308.44 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.339, pruned_loss=0.09165, over 5773263.47 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.342, pruned_loss=0.08791, over 5663841.56 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:51:40,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8208, 1.1216, 2.8567, 2.5948], device='cuda:0'), covar=tensor([0.1690, 0.2631, 0.0538, 0.1179], device='cuda:0'), in_proj_covar=tensor([0.0699, 0.0611, 0.0892, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 06:51:41,547 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-08 06:52:36,319 INFO [train.py:968] (0/2) Epoch 16, batch 14350, libri_loss[loss=0.248, simple_loss=0.3249, pruned_loss=0.08557, over 29540.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3405, pruned_loss=0.08736, over 5679217.45 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3387, pruned_loss=0.09147, over 5774997.67 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3416, pruned_loss=0.08759, over 5665492.73 frames. ], batch size: 77, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:52:45,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.951e+02 1.344e+03 1.732e+03 2.469e+03 4.725e+03, threshold=3.465e+03, percent-clipped=9.0
+2023-03-08 06:53:15,123 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=698332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:53:17,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=698335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:53:29,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4310, 1.6957, 1.5315, 1.4934], device='cuda:0'), covar=tensor([0.1525, 0.1964, 0.1855, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0722, 0.0679, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 06:53:34,892 INFO [train.py:968] (0/2) Epoch 16, batch 14400, giga_loss[loss=0.2692, simple_loss=0.3415, pruned_loss=0.09846, over 28740.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3408, pruned_loss=0.0886, over 5679710.92 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3387, pruned_loss=0.09152, over 5774233.21 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3417, pruned_loss=0.08862, over 5665804.29 frames. ], batch size: 262, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:53:49,257 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=698364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:54:41,080 INFO [train.py:968] (0/2) Epoch 16, batch 14450, giga_loss[loss=0.2841, simple_loss=0.3603, pruned_loss=0.104, over 28101.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3398, pruned_loss=0.08931, over 5691274.73 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3383, pruned_loss=0.09125, over 5777394.21 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.341, pruned_loss=0.08948, over 5674945.56 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:54:54,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.014e+02 1.424e+03 1.861e+03 2.686e+03 1.148e+04, threshold=3.723e+03, percent-clipped=15.0
+2023-03-08 06:55:58,952 INFO [train.py:968] (0/2) Epoch 16, batch 14500, giga_loss[loss=0.2244, simple_loss=0.3105, pruned_loss=0.06919, over 28152.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.34, pruned_loss=0.09006, over 5691766.69 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3384, pruned_loss=0.09139, over 5777534.74 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3408, pruned_loss=0.09004, over 5677624.40 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:56:29,392 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=698465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:56:43,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.87 vs. limit=2.0
+2023-03-08 06:57:24,483 INFO [train.py:968] (0/2) Epoch 16, batch 14550, giga_loss[loss=0.257, simple_loss=0.3409, pruned_loss=0.08654, over 28814.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3361, pruned_loss=0.08821, over 5688219.21 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3384, pruned_loss=0.09136, over 5779043.99 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3368, pruned_loss=0.08818, over 5674150.71 frames. ], batch size: 174, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:57:35,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.070e+02 1.291e+03 1.799e+03 2.279e+03 6.562e+03, threshold=3.597e+03, percent-clipped=10.0
+2023-03-08 06:58:23,734 INFO [train.py:968] (0/2) Epoch 16, batch 14600, giga_loss[loss=0.2586, simple_loss=0.3344, pruned_loss=0.09146, over 27768.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3342, pruned_loss=0.08717, over 5694634.49 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3374, pruned_loss=0.09092, over 5783050.50 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3355, pruned_loss=0.08736, over 5675019.71 frames. ], batch size: 474, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:58:30,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4880, 1.7857, 1.7491, 1.3051], device='cuda:0'), covar=tensor([0.1928, 0.2610, 0.1544, 0.1883], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0687, 0.0908, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 06:58:42,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6382, 2.3975, 1.8757, 0.8573], device='cuda:0'), covar=tensor([0.5201, 0.2496, 0.3164, 0.5260], device='cuda:0'), in_proj_covar=tensor([0.1641, 0.1555, 0.1538, 0.1350], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 06:59:33,379 INFO [train.py:968] (0/2) Epoch 16, batch 14650, giga_loss[loss=0.328, simple_loss=0.3967, pruned_loss=0.1297, over 28411.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3336, pruned_loss=0.08771, over 5679612.52 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3373, pruned_loss=0.0909, over 5774423.92 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3346, pruned_loss=0.08784, over 5670866.52 frames. ], batch size: 336, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:59:43,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=698608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:59:48,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.647e+02 1.335e+03 2.086e+03 2.674e+03 1.084e+04, threshold=4.172e+03, percent-clipped=14.0
+2023-03-08 06:59:48,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=698611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:59:55,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2948, 1.5249, 1.4913, 1.2930], device='cuda:0'), covar=tensor([0.2011, 0.1659, 0.1231, 0.1569], device='cuda:0'), in_proj_covar=tensor([0.1809, 0.1724, 0.1658, 0.1785], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 07:00:24,276 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=698640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:00:28,953 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=698643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:00:36,102 INFO [train.py:968] (0/2) Epoch 16, batch 14700, giga_loss[loss=0.29, simple_loss=0.3702, pruned_loss=0.1048, over 28792.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.339, pruned_loss=0.0908, over 5666149.77 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3374, pruned_loss=0.09092, over 5763284.43 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3398, pruned_loss=0.09087, over 5667692.06 frames. ], batch size: 243, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:01:39,733 INFO [train.py:968] (0/2) Epoch 16, batch 14750, giga_loss[loss=0.249, simple_loss=0.3217, pruned_loss=0.08814, over 29134.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3376, pruned_loss=0.09083, over 5677226.84 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3367, pruned_loss=0.09051, over 5767139.84 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3389, pruned_loss=0.09127, over 5672958.28 frames. ], batch size: 200, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:01:53,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.557e+02 1.631e+03 2.203e+03 2.930e+03 6.210e+03, threshold=4.407e+03, percent-clipped=6.0
+2023-03-08 07:01:54,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2392, 1.2029, 3.9153, 3.2446], device='cuda:0'), covar=tensor([0.1714, 0.2801, 0.0410, 0.0950], device='cuda:0'), in_proj_covar=tensor([0.0706, 0.0617, 0.0901, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 07:02:04,272 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-08 07:02:43,231 INFO [train.py:968] (0/2) Epoch 16, batch 14800, giga_loss[loss=0.2505, simple_loss=0.3298, pruned_loss=0.08559, over 28503.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3372, pruned_loss=0.09128, over 5684120.89 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3366, pruned_loss=0.09043, over 5769867.55 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3384, pruned_loss=0.09173, over 5675883.06 frames. ], batch size: 369, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 07:03:39,092 INFO [train.py:968] (0/2) Epoch 16, batch 14850, giga_loss[loss=0.2423, simple_loss=0.3233, pruned_loss=0.08064, over 28853.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3362, pruned_loss=0.09078, over 5691586.09 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3361, pruned_loss=0.09018, over 5773287.40 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3377, pruned_loss=0.09142, over 5677356.93 frames. ], batch size: 145, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:03:51,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.839e+02 1.284e+03 1.599e+03 2.197e+03 4.908e+03, threshold=3.197e+03, percent-clipped=2.0
+2023-03-08 07:04:45,626 INFO [train.py:968] (0/2) Epoch 16, batch 14900, giga_loss[loss=0.29, simple_loss=0.37, pruned_loss=0.105, over 28928.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3383, pruned_loss=0.09136, over 5687454.83 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3363, pruned_loss=0.0904, over 5775797.11 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3392, pruned_loss=0.0917, over 5672597.80 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:05:58,011 INFO [train.py:968] (0/2) Epoch 16, batch 14950, giga_loss[loss=0.2851, simple_loss=0.3572, pruned_loss=0.1065, over 28979.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3398, pruned_loss=0.09162, over 5682136.47 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3359, pruned_loss=0.09017, over 5776677.97 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3411, pruned_loss=0.09214, over 5666879.65 frames. ], batch size: 155, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:06:13,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.516e+02 1.543e+03 1.959e+03 2.717e+03 8.325e+03, threshold=3.917e+03, percent-clipped=19.0
+2023-03-08 07:06:28,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4066, 1.8206, 1.4008, 1.5972], device='cuda:0'), covar=tensor([0.0745, 0.0329, 0.0336, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 07:07:18,051 INFO [train.py:968] (0/2) Epoch 16, batch 15000, giga_loss[loss=0.254, simple_loss=0.3373, pruned_loss=0.08533, over 28729.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3395, pruned_loss=0.09168, over 5658777.55 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3363, pruned_loss=0.09046, over 5759263.96 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3403, pruned_loss=0.09183, over 5660565.61 frames. ], batch size: 262, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:07:18,057 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 07:07:27,434 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.1998, simple_loss=0.3003, pruned_loss=0.04969, over 944034.00 frames. 
+2023-03-08 07:07:27,434 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 07:07:41,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2542, 1.0065, 1.0261, 1.5292], device='cuda:0'), covar=tensor([0.0664, 0.0339, 0.0319, 0.0667], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 07:08:34,635 INFO [train.py:968] (0/2) Epoch 16, batch 15050, giga_loss[loss=0.247, simple_loss=0.322, pruned_loss=0.08604, over 29011.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3368, pruned_loss=0.09171, over 5660673.30 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.336, pruned_loss=0.09046, over 5759361.32 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3377, pruned_loss=0.09186, over 5659770.11 frames. ], batch size: 186, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:08:51,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.647e+02 1.286e+03 1.772e+03 2.543e+03 5.106e+03, threshold=3.544e+03, percent-clipped=5.0
+2023-03-08 07:09:01,132 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=699018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:09:44,467 INFO [train.py:968] (0/2) Epoch 16, batch 15100, giga_loss[loss=0.2315, simple_loss=0.3098, pruned_loss=0.07656, over 28878.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3307, pruned_loss=0.08881, over 5663426.36 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.336, pruned_loss=0.09054, over 5761310.48 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3313, pruned_loss=0.08884, over 5659624.22 frames. ], batch size: 174, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:09:46,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9294, 1.1058, 2.8453, 2.7243], device='cuda:0'), covar=tensor([0.1469, 0.2401, 0.0509, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0699, 0.0612, 0.0890, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 07:10:46,707 INFO [train.py:968] (0/2) Epoch 16, batch 15150, giga_loss[loss=0.2404, simple_loss=0.303, pruned_loss=0.08889, over 24363.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3311, pruned_loss=0.08935, over 5658165.85 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3355, pruned_loss=0.09037, over 5764185.43 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.332, pruned_loss=0.08949, over 5650831.25 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:10:58,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.294e+02 1.514e+03 1.838e+03 2.729e+03 9.170e+03, threshold=3.675e+03, percent-clipped=12.0
+2023-03-08 07:11:41,142 INFO [train.py:968] (0/2) Epoch 16, batch 15200, giga_loss[loss=0.2176, simple_loss=0.3046, pruned_loss=0.06529, over 28891.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3333, pruned_loss=0.09068, over 5665784.13 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3355, pruned_loss=0.09033, over 5763287.91 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3339, pruned_loss=0.09083, over 5658347.91 frames. ], batch size: 145, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 07:11:52,402 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=699161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:11:56,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=699164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:12:30,013 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=699193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:12:39,714 INFO [train.py:968] (0/2) Epoch 16, batch 15250, libri_loss[loss=0.2863, simple_loss=0.3549, pruned_loss=0.1088, over 29517.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3304, pruned_loss=0.08878, over 5662233.98 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3353, pruned_loss=0.09054, over 5765336.44 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3308, pruned_loss=0.08866, over 5651214.81 frames. ], batch size: 81, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 07:12:53,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.967e+02 1.391e+03 1.887e+03 2.576e+03 5.043e+03, threshold=3.774e+03, percent-clipped=5.0
+2023-03-08 07:13:41,430 INFO [train.py:968] (0/2) Epoch 16, batch 15300, libri_loss[loss=0.255, simple_loss=0.3363, pruned_loss=0.08684, over 29751.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3294, pruned_loss=0.08741, over 5669707.19 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3351, pruned_loss=0.0904, over 5768381.88 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3299, pruned_loss=0.0874, over 5656386.55 frames. ], batch size: 87, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 07:13:49,317 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 07:14:14,280 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3073, 1.4112, 1.2859, 1.4657], device='cuda:0'), covar=tensor([0.0744, 0.0367, 0.0347, 0.0824], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 07:14:54,172 INFO [train.py:968] (0/2) Epoch 16, batch 15350, giga_loss[loss=0.245, simple_loss=0.3272, pruned_loss=0.08141, over 28987.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3279, pruned_loss=0.08713, over 5656685.91 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3351, pruned_loss=0.09058, over 5761369.56 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3281, pruned_loss=0.08691, over 5650663.28 frames. ], batch size: 285, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:15:11,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.297e+02 1.356e+03 1.722e+03 2.399e+03 6.006e+03, threshold=3.445e+03, percent-clipped=9.0
+2023-03-08 07:16:04,490 INFO [train.py:968] (0/2) Epoch 16, batch 15400, giga_loss[loss=0.2151, simple_loss=0.2877, pruned_loss=0.07124, over 24447.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3273, pruned_loss=0.08594, over 5648969.63 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3353, pruned_loss=0.09068, over 5760429.74 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3272, pruned_loss=0.08558, over 5643131.55 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:16:07,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 07:16:21,467 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.45 vs. limit=5.0
+2023-03-08 07:16:54,864 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-08 07:16:56,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8533, 3.6867, 3.4787, 1.7163], device='cuda:0'), covar=tensor([0.0769, 0.0823, 0.0839, 0.2157], device='cuda:0'), in_proj_covar=tensor([0.1122, 0.1032, 0.0888, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 07:17:06,994 INFO [train.py:968] (0/2) Epoch 16, batch 15450, libri_loss[loss=0.1975, simple_loss=0.2807, pruned_loss=0.05719, over 28068.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3264, pruned_loss=0.08531, over 5644941.22 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3349, pruned_loss=0.09048, over 5749198.02 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3265, pruned_loss=0.08511, over 5647166.10 frames. ], batch size: 62, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:17:21,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.103e+02 1.407e+03 1.832e+03 2.335e+03 6.926e+03, threshold=3.664e+03, percent-clipped=10.0
+2023-03-08 07:17:44,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3909, 2.0913, 1.5017, 0.5173], device='cuda:0'), covar=tensor([0.4127, 0.2456, 0.3971, 0.5330], device='cuda:0'), in_proj_covar=tensor([0.1649, 0.1565, 0.1544, 0.1360], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 07:18:07,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5961, 1.7391, 1.9168, 1.4928], device='cuda:0'), covar=tensor([0.1616, 0.2034, 0.1270, 0.1814], device='cuda:0'), in_proj_covar=tensor([0.0859, 0.0681, 0.0904, 0.0808], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 07:18:13,032 INFO [train.py:968] (0/2) Epoch 16, batch 15500, giga_loss[loss=0.2773, simple_loss=0.3549, pruned_loss=0.09982, over 28472.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3273, pruned_loss=0.08687, over 5650617.29 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3348, pruned_loss=0.09055, over 5752846.39 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3273, pruned_loss=0.08654, over 5646358.45 frames. ], batch size: 336, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:18:45,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8783, 1.1969, 1.3211, 0.9682], device='cuda:0'), covar=tensor([0.1818, 0.1379, 0.2128, 0.1687], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0719, 0.0676, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 07:19:10,635 INFO [train.py:968] (0/2) Epoch 16, batch 15550, giga_loss[loss=0.2329, simple_loss=0.3194, pruned_loss=0.07319, over 28069.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3263, pruned_loss=0.08548, over 5655215.40 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3344, pruned_loss=0.0905, over 5745477.53 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3265, pruned_loss=0.08517, over 5655037.24 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:19:15,031 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 07:19:18,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3974, 1.5159, 1.1732, 1.0965], device='cuda:0'), covar=tensor([0.0863, 0.0503, 0.0972, 0.1140], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0436, 0.0503, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 07:19:22,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.929e+02 1.272e+03 1.563e+03 2.212e+03 4.504e+03, threshold=3.125e+03, percent-clipped=1.0
+2023-03-08 07:20:09,663 INFO [train.py:968] (0/2) Epoch 16, batch 15600, giga_loss[loss=0.2853, simple_loss=0.3609, pruned_loss=0.1048, over 28076.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3285, pruned_loss=0.08496, over 5667442.77 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3343, pruned_loss=0.09051, over 5747768.07 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3285, pruned_loss=0.08456, over 5663144.69 frames. ], batch size: 412, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:21:14,421 INFO [train.py:968] (0/2) Epoch 16, batch 15650, giga_loss[loss=0.2613, simple_loss=0.3503, pruned_loss=0.08611, over 28454.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3311, pruned_loss=0.08605, over 5658587.60 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.334, pruned_loss=0.09036, over 5749058.77 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3313, pruned_loss=0.08581, over 5652727.59 frames. ], batch size: 336, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:21:29,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.444e+02 1.201e+03 1.804e+03 3.156e+03 8.642e+03, threshold=3.608e+03, percent-clipped=26.0
+2023-03-08 07:22:14,734 INFO [train.py:968] (0/2) Epoch 16, batch 15700, libri_loss[loss=0.2143, simple_loss=0.2895, pruned_loss=0.06952, over 29459.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3325, pruned_loss=0.08653, over 5670129.75 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3333, pruned_loss=0.08996, over 5753441.45 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3334, pruned_loss=0.08662, over 5659249.85 frames. ], batch size: 70, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:23:16,008 INFO [train.py:968] (0/2) Epoch 16, batch 15750, giga_loss[loss=0.2242, simple_loss=0.3134, pruned_loss=0.06754, over 28862.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3319, pruned_loss=0.08618, over 5682342.11 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3326, pruned_loss=0.08956, over 5756712.29 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3332, pruned_loss=0.08652, over 5669556.45 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:23:30,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.928e+02 1.317e+03 1.768e+03 2.821e+03 8.168e+03, threshold=3.536e+03, percent-clipped=10.0
+2023-03-08 07:23:59,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5952, 2.0461, 1.9801, 1.5723], device='cuda:0'), covar=tensor([0.2879, 0.1881, 0.1868, 0.2243], device='cuda:0'), in_proj_covar=tensor([0.1816, 0.1730, 0.1658, 0.1795], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 07:24:19,178 INFO [train.py:968] (0/2) Epoch 16, batch 15800, giga_loss[loss=0.2411, simple_loss=0.3269, pruned_loss=0.0777, over 28986.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.33, pruned_loss=0.08431, over 5684429.26 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3327, pruned_loss=0.08958, over 5752387.45 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3308, pruned_loss=0.08449, over 5676741.87 frames. ], batch size: 155, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:24:53,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3084, 2.7081, 1.4457, 1.4336], device='cuda:0'), covar=tensor([0.0899, 0.0370, 0.0895, 0.1328], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0529, 0.0362, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 07:25:16,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1577, 2.5126, 2.0386, 2.3553], device='cuda:0'), covar=tensor([0.0548, 0.0220, 0.0245, 0.0595], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 07:25:23,541 INFO [train.py:968] (0/2) Epoch 16, batch 15850, giga_loss[loss=0.2472, simple_loss=0.3218, pruned_loss=0.08628, over 28935.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3294, pruned_loss=0.08421, over 5677150.48 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3326, pruned_loss=0.0895, over 5744828.73 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3302, pruned_loss=0.08436, over 5677553.76 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:25:41,829 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.750e+02 1.274e+03 1.649e+03 2.243e+03 5.292e+03, threshold=3.297e+03, percent-clipped=7.0
+2023-03-08 07:26:24,814 INFO [train.py:968] (0/2) Epoch 16, batch 15900, giga_loss[loss=0.2118, simple_loss=0.2799, pruned_loss=0.07188, over 28581.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3275, pruned_loss=0.08422, over 5677440.46 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.332, pruned_loss=0.08917, over 5750756.19 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3285, pruned_loss=0.08446, over 5669924.19 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:27:20,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3047, 4.1098, 3.8986, 1.8718], device='cuda:0'), covar=tensor([0.0599, 0.0747, 0.0758, 0.2140], device='cuda:0'), in_proj_covar=tensor([0.1121, 0.1026, 0.0886, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 07:27:28,518 INFO [train.py:968] (0/2) Epoch 16, batch 15950, libri_loss[loss=0.2268, simple_loss=0.3022, pruned_loss=0.07576, over 29511.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3302, pruned_loss=0.08583, over 5676945.69 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3315, pruned_loss=0.0889, over 5754547.08 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3314, pruned_loss=0.08617, over 5665640.96 frames. ], batch size: 70, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:27:43,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.906e+02 1.443e+03 1.981e+03 3.030e+03 6.511e+03, threshold=3.962e+03, percent-clipped=18.0
+2023-03-08 07:28:22,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5031, 1.9421, 1.6679, 1.5421], device='cuda:0'), covar=tensor([0.2200, 0.2454, 0.2380, 0.2403], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0723, 0.0680, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 07:28:30,825 INFO [train.py:968] (0/2) Epoch 16, batch 16000, giga_loss[loss=0.228, simple_loss=0.3167, pruned_loss=0.06968, over 28284.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3318, pruned_loss=0.08649, over 5686673.57 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3314, pruned_loss=0.08883, over 5758796.89 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3328, pruned_loss=0.08673, over 5671393.17 frames. ], batch size: 368, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:29:04,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9911, 1.3433, 1.0814, 0.2364], device='cuda:0'), covar=tensor([0.3009, 0.2591, 0.4155, 0.5221], device='cuda:0'), in_proj_covar=tensor([0.1643, 0.1558, 0.1536, 0.1355], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 07:29:22,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-08 07:29:37,266 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-700000.pt
+2023-03-08 07:29:38,227 INFO [train.py:968] (0/2) Epoch 16, batch 16050, giga_loss[loss=0.2687, simple_loss=0.3455, pruned_loss=0.09599, over 28890.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.333, pruned_loss=0.08794, over 5678801.22 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3311, pruned_loss=0.08867, over 5760142.98 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3342, pruned_loss=0.08828, over 5664899.51 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:29:51,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.325e+02 1.452e+03 1.697e+03 2.176e+03 5.622e+03, threshold=3.394e+03, percent-clipped=2.0
+2023-03-08 07:30:28,349 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:30:33,596 INFO [train.py:968] (0/2) Epoch 16, batch 16100, libri_loss[loss=0.2874, simple_loss=0.3554, pruned_loss=0.1096, over 19953.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3362, pruned_loss=0.08909, over 5673356.53 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3311, pruned_loss=0.08873, over 5746780.62 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3372, pruned_loss=0.08928, over 5672022.01 frames. ], batch size: 187, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:31:20,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6286, 1.6501, 1.3039, 1.2288], device='cuda:0'), covar=tensor([0.0702, 0.0433, 0.0782, 0.1029], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0437, 0.0504, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 07:31:24,922 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:31:29,725 INFO [train.py:968] (0/2) Epoch 16, batch 16150, giga_loss[loss=0.249, simple_loss=0.3352, pruned_loss=0.08139, over 28860.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3376, pruned_loss=0.08889, over 5673873.97 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3308, pruned_loss=0.08855, over 5741498.09 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3388, pruned_loss=0.08922, over 5675292.37 frames. ], batch size: 164, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:31:46,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.465e+02 1.356e+03 1.774e+03 2.645e+03 1.158e+04, threshold=3.549e+03, percent-clipped=8.0
+2023-03-08 07:32:39,729 INFO [train.py:968] (0/2) Epoch 16, batch 16200, giga_loss[loss=0.2383, simple_loss=0.3237, pruned_loss=0.07645, over 28400.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3372, pruned_loss=0.08888, over 5680409.94 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3308, pruned_loss=0.08854, over 5745909.66 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3384, pruned_loss=0.08918, over 5676178.06 frames. ], batch size: 71, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:33:42,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4204, 1.6047, 1.1854, 1.1901], device='cuda:0'), covar=tensor([0.0937, 0.0510, 0.1112, 0.1019], device='cuda:0'), in_proj_covar=tensor([0.0372, 0.0434, 0.0502, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 07:33:46,433 INFO [train.py:968] (0/2) Epoch 16, batch 16250, giga_loss[loss=0.2983, simple_loss=0.3585, pruned_loss=0.1191, over 28881.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3346, pruned_loss=0.08773, over 5691198.62 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3305, pruned_loss=0.08842, over 5749333.34 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3359, pruned_loss=0.08806, over 5683282.68 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:34:03,337 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.341e+02 1.339e+03 1.866e+03 2.681e+03 7.396e+03, threshold=3.732e+03, percent-clipped=13.0
+2023-03-08 07:34:35,008 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700239.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 07:34:51,982 INFO [train.py:968] (0/2) Epoch 16, batch 16300, giga_loss[loss=0.2504, simple_loss=0.3343, pruned_loss=0.08325, over 29030.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.335, pruned_loss=0.08818, over 5683534.71 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3302, pruned_loss=0.08827, over 5746025.00 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3364, pruned_loss=0.08858, over 5678899.30 frames. ], batch size: 155, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:34:53,096 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:35:18,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4532, 1.7463, 1.5273, 1.6772], device='cuda:0'), covar=tensor([0.0680, 0.0266, 0.0303, 0.0695], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 07:35:57,376 INFO [train.py:968] (0/2) Epoch 16, batch 16350, giga_loss[loss=0.2761, simple_loss=0.3467, pruned_loss=0.1027, over 28643.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3338, pruned_loss=0.08839, over 5671702.85 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3299, pruned_loss=0.0881, over 5747465.81 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3352, pruned_loss=0.08885, over 5665830.71 frames. ], batch size: 307, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:36:14,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.175e+02 1.341e+03 1.773e+03 2.414e+03 8.614e+03, threshold=3.546e+03, percent-clipped=3.0
+2023-03-08 07:36:18,915 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-08 07:36:36,027 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-08 07:37:02,255 INFO [train.py:968] (0/2) Epoch 16, batch 16400, giga_loss[loss=0.2492, simple_loss=0.3314, pruned_loss=0.08348, over 28997.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3315, pruned_loss=0.08793, over 5676639.60 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.33, pruned_loss=0.08816, over 5750245.60 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3326, pruned_loss=0.08825, over 5668385.30 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:37:52,488 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:38:05,564 INFO [train.py:968] (0/2) Epoch 16, batch 16450, giga_loss[loss=0.317, simple_loss=0.3761, pruned_loss=0.129, over 26824.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3326, pruned_loss=0.088, over 5672966.03 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3303, pruned_loss=0.08837, over 5741194.03 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3333, pruned_loss=0.08807, over 5672950.20 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:38:22,457 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.361e+02 1.509e+03 1.963e+03 2.811e+03 5.501e+03, threshold=3.926e+03, percent-clipped=9.0
+2023-03-08 07:38:27,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:39:05,239 INFO [train.py:968] (0/2) Epoch 16, batch 16500, giga_loss[loss=0.2364, simple_loss=0.3209, pruned_loss=0.07597, over 28487.00 frames. ], tot_loss[loss=0.252, simple_loss=0.331, pruned_loss=0.08646, over 5670597.35 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3304, pruned_loss=0.08846, over 5746012.17 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3316, pruned_loss=0.08639, over 5663914.29 frames. ], batch size: 336, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:39:27,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:40:02,128 INFO [train.py:968] (0/2) Epoch 16, batch 16550, giga_loss[loss=0.2641, simple_loss=0.3517, pruned_loss=0.08823, over 28966.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3321, pruned_loss=0.08525, over 5680531.29 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3297, pruned_loss=0.08809, over 5750535.19 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3331, pruned_loss=0.08545, over 5669496.33 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:40:17,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.555e+02 1.342e+03 1.730e+03 2.336e+03 8.569e+03, threshold=3.460e+03, percent-clipped=4.0
+2023-03-08 07:40:59,166 INFO [train.py:968] (0/2) Epoch 16, batch 16600, giga_loss[loss=0.274, simple_loss=0.3558, pruned_loss=0.09607, over 28922.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3332, pruned_loss=0.08482, over 5663491.24 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3299, pruned_loss=0.08835, over 5732715.52 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.334, pruned_loss=0.08471, over 5667538.01 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:41:12,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=700562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:41:14,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=700565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:41:44,721 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:41:46,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=700594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:41:54,820 INFO [train.py:968] (0/2) Epoch 16, batch 16650, giga_loss[loss=0.2537, simple_loss=0.34, pruned_loss=0.08373, over 28484.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3335, pruned_loss=0.08452, over 5681522.89 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3296, pruned_loss=0.08809, over 5737023.06 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3345, pruned_loss=0.08458, over 5679221.67 frames. ], batch size: 336, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:42:13,296 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700614.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 07:42:13,323 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=700614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:42:15,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.295e+02 1.413e+03 2.050e+03 2.893e+03 6.943e+03, threshold=4.101e+03, percent-clipped=14.0
+2023-03-08 07:42:17,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=700617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:42:24,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7367, 1.5101, 4.8203, 3.4887], device='cuda:0'), covar=tensor([0.1552, 0.2564, 0.0387, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0702, 0.0611, 0.0891, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 07:42:30,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:42:55,377 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=700646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:42:55,582 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-08 07:43:03,550 INFO [train.py:968] (0/2) Epoch 16, batch 16700, giga_loss[loss=0.2517, simple_loss=0.3172, pruned_loss=0.09312, over 24467.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3339, pruned_loss=0.08507, over 5674021.31 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3298, pruned_loss=0.08808, over 5735632.05 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3346, pruned_loss=0.08502, over 5671802.10 frames. ], batch size: 705, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:43:13,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3033, 1.5415, 1.2584, 1.0955], device='cuda:0'), covar=tensor([0.2645, 0.2593, 0.3007, 0.2243], device='cuda:0'), in_proj_covar=tensor([0.1401, 0.1017, 0.1245, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 07:44:10,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-08 07:44:10,598 INFO [train.py:968] (0/2) Epoch 16, batch 16750, giga_loss[loss=0.2344, simple_loss=0.3165, pruned_loss=0.07617, over 27881.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3328, pruned_loss=0.08411, over 5674531.64 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3294, pruned_loss=0.08791, over 5738928.36 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3338, pruned_loss=0.08413, over 5668165.05 frames. ], batch size: 476, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:44:33,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.383e+02 1.370e+03 1.858e+03 2.433e+03 7.730e+03, threshold=3.717e+03, percent-clipped=9.0
+2023-03-08 07:44:35,792 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2378, 1.3608, 1.1680, 1.0426], device='cuda:0'), covar=tensor([0.0879, 0.0464, 0.1039, 0.0908], device='cuda:0'), in_proj_covar=tensor([0.0371, 0.0432, 0.0499, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 07:45:23,571 INFO [train.py:968] (0/2) Epoch 16, batch 16800, giga_loss[loss=0.2739, simple_loss=0.3559, pruned_loss=0.09599, over 28868.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3334, pruned_loss=0.08368, over 5677994.95 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3293, pruned_loss=0.08776, over 5739974.59 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3344, pruned_loss=0.08377, over 5670887.40 frames. ], batch size: 213, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:45:31,638 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=700757.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 07:45:36,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=700760.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 07:45:37,851 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-08 07:45:45,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:45:50,499 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=700770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:45:53,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=700773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:45:55,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5547, 1.9601, 1.7897, 1.4795], device='cuda:0'), covar=tensor([0.2443, 0.1873, 0.1948, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.1814, 0.1710, 0.1635, 0.1779], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 07:46:19,104 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=700789.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 07:46:37,263 INFO [train.py:968] (0/2) Epoch 16, batch 16850, giga_loss[loss=0.2913, simple_loss=0.3703, pruned_loss=0.1062, over 28946.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3353, pruned_loss=0.08514, over 5673682.68 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3291, pruned_loss=0.08774, over 5733750.89 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3363, pruned_loss=0.08516, over 5672172.67 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:46:39,730 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=700802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:47:05,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.114e+02 1.331e+03 1.890e+03 2.749e+03 1.882e+04, threshold=3.780e+03, percent-clipped=15.0
+2023-03-08 07:47:52,016 INFO [train.py:968] (0/2) Epoch 16, batch 16900, giga_loss[loss=0.2483, simple_loss=0.3366, pruned_loss=0.07998, over 28933.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3393, pruned_loss=0.08702, over 5668288.16 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3291, pruned_loss=0.08772, over 5727004.93 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3402, pruned_loss=0.08702, over 5673356.22 frames. ], batch size: 136, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:48:58,702 INFO [train.py:968] (0/2) Epoch 16, batch 16950, giga_loss[loss=0.2623, simple_loss=0.3394, pruned_loss=0.0926, over 28838.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3391, pruned_loss=0.0873, over 5667925.83 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3292, pruned_loss=0.08777, over 5723699.35 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3401, pruned_loss=0.08724, over 5673621.75 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:49:08,075 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=700908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:49:10,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2160, 1.4608, 1.4486, 1.0747], device='cuda:0'), covar=tensor([0.1554, 0.2382, 0.1360, 0.1515], device='cuda:0'), in_proj_covar=tensor([0.0860, 0.0679, 0.0904, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 07:49:11,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=700911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:49:23,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.972e+02 1.318e+03 1.627e+03 2.363e+03 7.075e+03, threshold=3.253e+03, percent-clipped=12.0
+2023-03-08 07:49:30,682 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 07:49:31,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:49:56,262 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=700940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:50:15,505 INFO [train.py:968] (0/2) Epoch 16, batch 17000, giga_loss[loss=0.2527, simple_loss=0.3263, pruned_loss=0.08958, over 28962.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3368, pruned_loss=0.0867, over 5680477.77 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3292, pruned_loss=0.08776, over 5724595.85 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3376, pruned_loss=0.08666, over 5683952.18 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:50:27,036 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:50:42,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:51:24,132 INFO [train.py:968] (0/2) Epoch 16, batch 17050, giga_loss[loss=0.2289, simple_loss=0.308, pruned_loss=0.07486, over 28687.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3338, pruned_loss=0.0845, over 5687250.35 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.329, pruned_loss=0.08748, over 5728059.11 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3349, pruned_loss=0.08466, over 5685673.92 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:51:51,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.737e+02 1.160e+03 1.564e+03 2.310e+03 6.053e+03, threshold=3.127e+03, percent-clipped=9.0
+2023-03-08 07:52:29,071 INFO [train.py:968] (0/2) Epoch 16, batch 17100, giga_loss[loss=0.2639, simple_loss=0.3389, pruned_loss=0.09451, over 28921.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3321, pruned_loss=0.08301, over 5698446.26 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3291, pruned_loss=0.08757, over 5732485.59 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3331, pruned_loss=0.08293, over 5691752.18 frames. ], batch size: 213, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:53:33,192 INFO [train.py:968] (0/2) Epoch 16, batch 17150, giga_loss[loss=0.2646, simple_loss=0.347, pruned_loss=0.09112, over 28039.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3337, pruned_loss=0.08441, over 5687714.47 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3289, pruned_loss=0.08751, over 5734385.16 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3346, pruned_loss=0.08436, over 5680222.09 frames. ], batch size: 412, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:53:45,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=701111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:53:49,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=701114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:53:54,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.973e+02 1.411e+03 1.834e+03 2.636e+03 5.480e+03, threshold=3.667e+03, percent-clipped=17.0
+2023-03-08 07:54:21,576 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=701143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:54:31,508 INFO [train.py:968] (0/2) Epoch 16, batch 17200, giga_loss[loss=0.2631, simple_loss=0.3506, pruned_loss=0.08782, over 28808.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3362, pruned_loss=0.0859, over 5693064.28 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3292, pruned_loss=0.08767, over 5739319.83 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.337, pruned_loss=0.08564, over 5681207.66 frames. ], batch size: 243, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:55:23,947 INFO [train.py:968] (0/2) Epoch 16, batch 17250, giga_loss[loss=0.2537, simple_loss=0.3378, pruned_loss=0.0848, over 28488.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3354, pruned_loss=0.08589, over 5693682.01 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3287, pruned_loss=0.08749, over 5744611.75 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3366, pruned_loss=0.0858, over 5677526.53 frames. ], batch size: 336, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:55:42,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.490e+02 1.518e+03 2.179e+03 2.860e+03 6.185e+03, threshold=4.358e+03, percent-clipped=15.0
+2023-03-08 07:55:45,972 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701221.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 07:56:19,398 INFO [train.py:968] (0/2) Epoch 16, batch 17300, giga_loss[loss=0.2273, simple_loss=0.3135, pruned_loss=0.07054, over 28282.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.333, pruned_loss=0.08579, over 5692340.55 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3287, pruned_loss=0.08739, over 5748629.00 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3342, pruned_loss=0.08577, over 5674476.01 frames. ], batch size: 368, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:57:18,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=701298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:57:21,005 INFO [train.py:968] (0/2) Epoch 16, batch 17350, giga_loss[loss=0.259, simple_loss=0.3349, pruned_loss=0.0916, over 28453.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3328, pruned_loss=0.08649, over 5692757.25 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3282, pruned_loss=0.08719, over 5750602.88 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3343, pruned_loss=0.08663, over 5675907.42 frames. ], batch size: 336, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:57:41,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.541e+02 1.450e+03 1.964e+03 2.698e+03 1.095e+04, threshold=3.929e+03, percent-clipped=5.0
+2023-03-08 07:57:57,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=701332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:58:18,870 INFO [train.py:968] (0/2) Epoch 16, batch 17400, giga_loss[loss=0.3177, simple_loss=0.3884, pruned_loss=0.1235, over 28296.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3389, pruned_loss=0.09049, over 5692008.79 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3283, pruned_loss=0.08719, over 5753290.15 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3402, pruned_loss=0.09062, over 5675319.72 frames. ], batch size: 368, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:58:48,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1459, 1.1389, 3.5374, 3.0073], device='cuda:0'), covar=tensor([0.1695, 0.2817, 0.0441, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0703, 0.0615, 0.0896, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 07:59:07,102 INFO [train.py:968] (0/2) Epoch 16, batch 17450, giga_loss[loss=0.277, simple_loss=0.3649, pruned_loss=0.09454, over 28841.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3482, pruned_loss=0.09592, over 5698144.48 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3286, pruned_loss=0.08723, over 5756411.36 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3494, pruned_loss=0.09619, over 5679956.35 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:59:22,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.934e+02 1.355e+03 1.752e+03 2.386e+03 6.321e+03, threshold=3.503e+03, percent-clipped=6.0
+2023-03-08 07:59:43,643 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=701441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:59:45,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=701444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:59:50,494 INFO [train.py:968] (0/2) Epoch 16, batch 17500, giga_loss[loss=0.2697, simple_loss=0.3506, pruned_loss=0.09441, over 29096.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3527, pruned_loss=0.09838, over 5704521.66 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3287, pruned_loss=0.08727, over 5756462.21 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3542, pruned_loss=0.09884, over 5688245.12 frames. ], batch size: 136, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:00:14,547 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=701473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:00:15,906 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=701475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:00:17,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=701478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:00:29,643 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-08 08:00:40,332 INFO [train.py:968] (0/2) Epoch 16, batch 17550, giga_loss[loss=0.2715, simple_loss=0.3447, pruned_loss=0.09913, over 29027.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3492, pruned_loss=0.09732, over 5700838.71 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.329, pruned_loss=0.08735, over 5757912.02 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3503, pruned_loss=0.09772, over 5686328.75 frames. ], batch size: 155, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:00:47,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=701507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:00:58,851 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.132e+02 1.118e+03 1.335e+03 1.837e+03 3.575e+03, threshold=2.669e+03, percent-clipped=1.0
+2023-03-08 08:01:17,540 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-08 08:01:20,621 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:01:26,750 INFO [train.py:968] (0/2) Epoch 16, batch 17600, giga_loss[loss=0.2321, simple_loss=0.3089, pruned_loss=0.0777, over 28624.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3413, pruned_loss=0.09398, over 5692969.45 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3291, pruned_loss=0.08737, over 5758228.57 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3423, pruned_loss=0.09437, over 5680278.61 frames. ], batch size: 307, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:01:42,293 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:01:53,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6362, 1.7576, 1.4725, 1.8856], device='cuda:0'), covar=tensor([0.2816, 0.2884, 0.3249, 0.2566], device='cuda:0'), in_proj_covar=tensor([0.1396, 0.1019, 0.1243, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 08:02:06,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6599, 1.9632, 1.5431, 1.8742], device='cuda:0'), covar=tensor([0.2554, 0.2575, 0.2967, 0.2518], device='cuda:0'), in_proj_covar=tensor([0.1395, 0.1018, 0.1241, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 08:02:09,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=701596.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:02:12,307 INFO [train.py:968] (0/2) Epoch 16, batch 17650, giga_loss[loss=0.278, simple_loss=0.3432, pruned_loss=0.1064, over 28753.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3344, pruned_loss=0.09111, over 5694917.24 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3298, pruned_loss=0.08768, over 5761202.53 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3347, pruned_loss=0.09125, over 5681019.35 frames. ], batch size: 262, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:02:28,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.873e+02 1.051e+03 1.442e+03 2.013e+03 5.344e+03, threshold=2.883e+03, percent-clipped=8.0
+2023-03-08 08:02:36,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9499, 2.1011, 1.9040, 1.7624], device='cuda:0'), covar=tensor([0.1798, 0.2345, 0.2272, 0.2357], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0720, 0.0677, 0.0656], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 08:02:41,911 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:02:57,742 INFO [train.py:968] (0/2) Epoch 16, batch 17700, giga_loss[loss=0.2583, simple_loss=0.3127, pruned_loss=0.1019, over 26489.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3264, pruned_loss=0.08727, over 5688649.53 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3301, pruned_loss=0.08768, over 5761033.61 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3264, pruned_loss=0.08742, over 5675534.50 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:03:38,153 INFO [train.py:968] (0/2) Epoch 16, batch 17750, giga_loss[loss=0.2241, simple_loss=0.2966, pruned_loss=0.07574, over 28390.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3206, pruned_loss=0.08479, over 5692350.36 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3302, pruned_loss=0.08762, over 5760365.19 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3202, pruned_loss=0.08488, over 5680151.06 frames. ], batch size: 71, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:03:44,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6333, 4.4456, 4.1832, 1.9414], device='cuda:0'), covar=tensor([0.0538, 0.0752, 0.0787, 0.2065], device='cuda:0'), in_proj_covar=tensor([0.1118, 0.1029, 0.0884, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-08 08:03:52,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.823e+02 1.005e+03 1.177e+03 1.513e+03 5.457e+03, threshold=2.354e+03, percent-clipped=3.0
+2023-03-08 08:04:01,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4176, 1.7652, 1.3796, 1.3065], device='cuda:0'), covar=tensor([0.2535, 0.2420, 0.2803, 0.2335], device='cuda:0'), in_proj_covar=tensor([0.1402, 0.1024, 0.1246, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 08:04:02,397 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 08:04:08,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=701739.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:04:10,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=701742.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:04:16,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2163, 1.1210, 3.8625, 3.0945], device='cuda:0'), covar=tensor([0.1727, 0.2892, 0.0406, 0.1006], device='cuda:0'), in_proj_covar=tensor([0.0703, 0.0614, 0.0899, 0.0817], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 08:04:19,076 INFO [train.py:968] (0/2) Epoch 16, batch 17800, giga_loss[loss=0.3024, simple_loss=0.3489, pruned_loss=0.1279, over 27613.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3156, pruned_loss=0.08237, over 5699842.91 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3302, pruned_loss=0.08745, over 5763640.41 frames. ], giga_tot_loss[loss=0.24, simple_loss=0.315, pruned_loss=0.08248, over 5685805.34 frames. ], batch size: 472, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:04:35,663 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=701771.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:04:47,706 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:04:49,747 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701788.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:05:00,201 INFO [train.py:968] (0/2) Epoch 16, batch 17850, giga_loss[loss=0.2075, simple_loss=0.2865, pruned_loss=0.06427, over 28923.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3133, pruned_loss=0.08162, over 5708527.87 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3301, pruned_loss=0.08739, over 5767643.64 frames. ], giga_tot_loss[loss=0.2379, simple_loss=0.3125, pruned_loss=0.08167, over 5692420.84 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:05:15,683 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 08:05:17,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.935e+02 1.128e+03 1.578e+03 2.154e+03 9.400e+03, threshold=3.157e+03, percent-clipped=19.0
+2023-03-08 08:05:42,212 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 08:05:44,215 INFO [train.py:968] (0/2) Epoch 16, batch 17900, giga_loss[loss=0.2328, simple_loss=0.2963, pruned_loss=0.08465, over 28849.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3113, pruned_loss=0.081, over 5697643.10 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3307, pruned_loss=0.08764, over 5763758.90 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.3095, pruned_loss=0.08058, over 5685642.55 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:06:25,664 INFO [train.py:968] (0/2) Epoch 16, batch 17950, giga_loss[loss=0.1923, simple_loss=0.2714, pruned_loss=0.05661, over 28801.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3075, pruned_loss=0.07931, over 5710849.99 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.33, pruned_loss=0.08735, over 5766072.15 frames. ], giga_tot_loss[loss=0.2323, simple_loss=0.3064, pruned_loss=0.07911, over 5698016.01 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:06:40,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.484e+02 9.910e+02 1.314e+03 2.042e+03 5.541e+03, threshold=2.628e+03, percent-clipped=5.0
+2023-03-08 08:06:40,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=701919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:06:44,288 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:07:01,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=701941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:07:11,777 INFO [train.py:968] (0/2) Epoch 16, batch 18000, giga_loss[loss=0.226, simple_loss=0.3012, pruned_loss=0.07539, over 28969.00 frames. ], tot_loss[loss=0.231, simple_loss=0.305, pruned_loss=0.07849, over 5693646.54 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3303, pruned_loss=0.08731, over 5766757.17 frames. ], giga_tot_loss[loss=0.2301, simple_loss=0.3036, pruned_loss=0.07826, over 5682232.67 frames. ], batch size: 213, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:07:11,782 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 08:07:21,359 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2096, simple_loss=0.3154, pruned_loss=0.0519, over 944034.00 frames. 
+2023-03-08 08:07:21,359 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 08:07:40,078 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:07:43,425 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:08:04,722 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-702000.pt
+2023-03-08 08:08:05,645 INFO [train.py:968] (0/2) Epoch 16, batch 18050, giga_loss[loss=0.2511, simple_loss=0.3096, pruned_loss=0.09635, over 28885.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.3028, pruned_loss=0.07747, over 5699925.18 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3304, pruned_loss=0.08717, over 5770261.78 frames. ], giga_tot_loss[loss=0.2277, simple_loss=0.301, pruned_loss=0.07719, over 5686205.96 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:08:13,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:08:13,943 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 08:08:20,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.133e+02 1.008e+03 1.298e+03 1.876e+03 5.616e+03, threshold=2.596e+03, percent-clipped=6.0
+2023-03-08 08:08:35,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0702, 2.4247, 2.1041, 1.7279], device='cuda:0'), covar=tensor([0.2871, 0.2003, 0.2063, 0.2599], device='cuda:0'), in_proj_covar=tensor([0.1855, 0.1749, 0.1669, 0.1826], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 08:08:45,883 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=702050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:08:46,397 INFO [train.py:968] (0/2) Epoch 16, batch 18100, giga_loss[loss=0.2083, simple_loss=0.2811, pruned_loss=0.06779, over 28485.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.2999, pruned_loss=0.07619, over 5696424.20 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3304, pruned_loss=0.08714, over 5770183.32 frames. ], giga_tot_loss[loss=0.2245, simple_loss=0.2976, pruned_loss=0.07574, over 5683453.93 frames. ], batch size: 60, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:08:58,141 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:08:59,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:09:18,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:09:22,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:09:31,184 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:09:36,469 INFO [train.py:968] (0/2) Epoch 16, batch 18150, giga_loss[loss=0.2003, simple_loss=0.2754, pruned_loss=0.06264, over 28622.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2974, pruned_loss=0.07532, over 5688575.63 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3305, pruned_loss=0.08709, over 5771378.96 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2953, pruned_loss=0.07491, over 5676689.89 frames. ], batch size: 242, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:09:38,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3525, 1.6509, 1.3469, 1.5905], device='cuda:0'), covar=tensor([0.0750, 0.0341, 0.0337, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 08:09:48,526 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:09:52,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.891e+02 1.036e+03 1.262e+03 1.619e+03 3.904e+03, threshold=2.525e+03, percent-clipped=7.0
+2023-03-08 08:10:04,629 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 08:10:19,881 INFO [train.py:968] (0/2) Epoch 16, batch 18200, giga_loss[loss=0.2043, simple_loss=0.2769, pruned_loss=0.06583, over 28751.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2957, pruned_loss=0.07487, over 5691156.03 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3307, pruned_loss=0.08713, over 5774092.23 frames. ], giga_tot_loss[loss=0.2208, simple_loss=0.2931, pruned_loss=0.07426, over 5677909.59 frames. ], batch size: 242, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:10:21,730 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:10:23,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:10:27,418 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:10:29,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702163.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:10:54,090 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:11:11,510 INFO [train.py:968] (0/2) Epoch 16, batch 18250, libri_loss[loss=0.2608, simple_loss=0.3405, pruned_loss=0.09061, over 19796.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3067, pruned_loss=0.08095, over 5673905.38 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3312, pruned_loss=0.08735, over 5765912.52 frames. ], giga_tot_loss[loss=0.232, simple_loss=0.3038, pruned_loss=0.08013, over 5670033.12 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:11:15,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1984, 4.0071, 3.7756, 1.9539], device='cuda:0'), covar=tensor([0.0586, 0.0733, 0.0746, 0.1977], device='cuda:0'), in_proj_covar=tensor([0.1121, 0.1034, 0.0888, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 08:11:18,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3893, 3.1154, 1.4501, 1.4620], device='cuda:0'), covar=tensor([0.0991, 0.0336, 0.0940, 0.1365], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0525, 0.0360, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 08:11:27,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.013e+02 1.305e+03 1.666e+03 2.466e+03 7.915e+03, threshold=3.332e+03, percent-clipped=24.0
+2023-03-08 08:11:31,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2561, 1.5353, 1.2216, 1.1506], device='cuda:0'), covar=tensor([0.2432, 0.2421, 0.2804, 0.1953], device='cuda:0'), in_proj_covar=tensor([0.1401, 0.1022, 0.1245, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 08:11:57,466 INFO [train.py:968] (0/2) Epoch 16, batch 18300, giga_loss[loss=0.2908, simple_loss=0.3719, pruned_loss=0.1049, over 28979.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3215, pruned_loss=0.08894, over 5675435.94 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3317, pruned_loss=0.08765, over 5760575.09 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3182, pruned_loss=0.08791, over 5675036.90 frames. ], batch size: 136, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:12:17,372 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=702273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:12:37,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:12:38,157 INFO [train.py:968] (0/2) Epoch 16, batch 18350, giga_loss[loss=0.2838, simple_loss=0.3657, pruned_loss=0.1009, over 28882.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3328, pruned_loss=0.09446, over 5688085.10 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.332, pruned_loss=0.0876, over 5764552.97 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.33, pruned_loss=0.09383, over 5682180.95 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:12:40,758 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:12:42,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:12:43,006 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702306.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:12:45,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702309.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:12:52,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.270e+03 1.594e+03 2.009e+03 6.462e+03, threshold=3.189e+03, percent-clipped=7.0
+2023-03-08 08:12:54,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5701, 1.8180, 1.4463, 1.7756], device='cuda:0'), covar=tensor([0.2483, 0.2540, 0.2882, 0.2233], device='cuda:0'), in_proj_covar=tensor([0.1400, 0.1022, 0.1246, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 08:12:59,310 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-08 08:13:05,641 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:13:08,845 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702338.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:13:15,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:13:17,820 INFO [train.py:968] (0/2) Epoch 16, batch 18400, giga_loss[loss=0.2706, simple_loss=0.3553, pruned_loss=0.09296, over 28574.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.34, pruned_loss=0.09728, over 5688268.25 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.332, pruned_loss=0.08751, over 5764405.27 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3378, pruned_loss=0.09706, over 5681810.58 frames. ], batch size: 307, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:13:19,534 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:13:22,800 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.33 vs. limit=5.0
+2023-03-08 08:13:33,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=702368.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:13:46,303 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 08:13:59,244 INFO [train.py:968] (0/2) Epoch 16, batch 18450, giga_loss[loss=0.2711, simple_loss=0.3445, pruned_loss=0.09886, over 29043.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3429, pruned_loss=0.09762, over 5694177.72 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3324, pruned_loss=0.08763, over 5769893.55 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3411, pruned_loss=0.09773, over 5681056.45 frames. ], batch size: 128, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:14:16,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.266e+02 1.150e+03 1.450e+03 2.215e+03 4.727e+03, threshold=2.900e+03, percent-clipped=8.0
+2023-03-08 08:14:21,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:14:32,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4768, 1.8088, 1.4361, 1.4753], device='cuda:0'), covar=tensor([0.2596, 0.2510, 0.2816, 0.2230], device='cuda:0'), in_proj_covar=tensor([0.1398, 0.1020, 0.1244, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 08:14:38,579 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:14:40,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:14:43,621 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4867, 1.4648, 1.1237, 1.0897], device='cuda:0'), covar=tensor([0.0639, 0.0339, 0.0846, 0.0959], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0439, 0.0508, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 08:14:45,278 INFO [train.py:968] (0/2) Epoch 16, batch 18500, giga_loss[loss=0.256, simple_loss=0.3377, pruned_loss=0.08715, over 28821.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3444, pruned_loss=0.09732, over 5682126.01 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3325, pruned_loss=0.08758, over 5771295.13 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3431, pruned_loss=0.09754, over 5669757.61 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:15:09,007 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:23,376 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:25,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:26,500 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:28,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:29,979 INFO [train.py:968] (0/2) Epoch 16, batch 18550, giga_loss[loss=0.2849, simple_loss=0.3514, pruned_loss=0.1092, over 28729.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3477, pruned_loss=0.09977, over 5679105.76 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3337, pruned_loss=0.08836, over 5764927.95 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.346, pruned_loss=0.09968, over 5670728.84 frames. ], batch size: 119, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:15:49,998 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.843e+02 1.213e+03 1.601e+03 2.181e+03 7.232e+03, threshold=3.202e+03, percent-clipped=10.0
+2023-03-08 08:15:50,189 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:53,834 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:53,991 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-08 08:16:16,480 INFO [train.py:968] (0/2) Epoch 16, batch 18600, giga_loss[loss=0.2704, simple_loss=0.3399, pruned_loss=0.1005, over 28738.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3503, pruned_loss=0.1017, over 5681121.74 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.334, pruned_loss=0.08841, over 5766149.02 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.349, pruned_loss=0.1018, over 5671977.09 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:16:33,660 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-08 08:16:34,199 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:16:36,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:17:01,633 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:17:02,087 INFO [train.py:968] (0/2) Epoch 16, batch 18650, giga_loss[loss=0.2946, simple_loss=0.3698, pruned_loss=0.1098, over 28627.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3536, pruned_loss=0.1033, over 5675992.67 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3344, pruned_loss=0.08856, over 5764735.43 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3524, pruned_loss=0.1035, over 5668874.76 frames. ], batch size: 85, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:17:20,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.429e+02 1.146e+03 1.396e+03 1.695e+03 3.658e+03, threshold=2.793e+03, percent-clipped=2.0
+2023-03-08 08:17:44,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:17:46,032 INFO [train.py:968] (0/2) Epoch 16, batch 18700, giga_loss[loss=0.3045, simple_loss=0.3887, pruned_loss=0.1102, over 28938.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3568, pruned_loss=0.1045, over 5682173.60 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3345, pruned_loss=0.08868, over 5765954.55 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.356, pruned_loss=0.1047, over 5674824.42 frames. ], batch size: 213, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:18:26,657 INFO [train.py:968] (0/2) Epoch 16, batch 18750, giga_loss[loss=0.3165, simple_loss=0.3828, pruned_loss=0.1251, over 28702.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3588, pruned_loss=0.1049, over 5682419.85 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3351, pruned_loss=0.08876, over 5767113.07 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3582, pruned_loss=0.1053, over 5673700.99 frames. ], batch size: 307, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:18:44,474 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.983e+02 1.227e+03 1.559e+03 2.269e+03 8.426e+03, threshold=3.118e+03, percent-clipped=16.0
+2023-03-08 08:19:01,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702743.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:19:10,942 INFO [train.py:968] (0/2) Epoch 16, batch 18800, giga_loss[loss=0.287, simple_loss=0.3567, pruned_loss=0.1086, over 27592.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3592, pruned_loss=0.1042, over 5688977.91 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.335, pruned_loss=0.08863, over 5762593.37 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3594, pruned_loss=0.105, over 5683734.39 frames. ], batch size: 472, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:19:43,216 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:19:45,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:19:49,669 INFO [train.py:968] (0/2) Epoch 16, batch 18850, giga_loss[loss=0.2781, simple_loss=0.3566, pruned_loss=0.09979, over 28610.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3588, pruned_loss=0.1023, over 5700123.95 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3357, pruned_loss=0.08879, over 5767313.64 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3592, pruned_loss=0.1033, over 5689273.54 frames. ], batch size: 85, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:20:07,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.719e+02 1.101e+03 1.348e+03 1.850e+03 5.482e+03, threshold=2.697e+03, percent-clipped=6.0
+2023-03-08 08:20:08,665 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:20:31,077 INFO [train.py:968] (0/2) Epoch 16, batch 18900, giga_loss[loss=0.2449, simple_loss=0.3347, pruned_loss=0.07751, over 28685.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.357, pruned_loss=0.09989, over 5707850.93 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3364, pruned_loss=0.08898, over 5768688.31 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3576, pruned_loss=0.1011, over 5695284.04 frames. ], batch size: 60, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:21:00,425 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:21:02,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702889.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:21:06,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3359, 1.4799, 1.1262, 1.4976], device='cuda:0'), covar=tensor([0.0785, 0.0331, 0.0349, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:0')
+2023-03-08 08:21:12,182 INFO [train.py:968] (0/2) Epoch 16, batch 18950, giga_loss[loss=0.2631, simple_loss=0.3423, pruned_loss=0.09196, over 28614.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3553, pruned_loss=0.09864, over 5710798.31 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3365, pruned_loss=0.08887, over 5772149.27 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3561, pruned_loss=0.09991, over 5696362.87 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:21:25,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702918.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:21:28,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.461e+02 1.123e+03 1.349e+03 1.891e+03 3.752e+03, threshold=2.697e+03, percent-clipped=2.0
+2023-03-08 08:21:35,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4955, 2.2441, 1.5238, 0.6526], device='cuda:0'), covar=tensor([0.4597, 0.2647, 0.4116, 0.5448], device='cuda:0'), in_proj_covar=tensor([0.1641, 0.1552, 0.1542, 0.1349], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 08:21:55,538 INFO [train.py:968] (0/2) Epoch 16, batch 19000, giga_loss[loss=0.3154, simple_loss=0.3707, pruned_loss=0.1301, over 28899.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3579, pruned_loss=0.1018, over 5717771.21 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3373, pruned_loss=0.08918, over 5775558.52 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3583, pruned_loss=0.1028, over 5701650.85 frames. ], batch size: 112, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:22:10,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2190, 3.9748, 3.7949, 1.7680], device='cuda:0'), covar=tensor([0.0644, 0.0875, 0.0809, 0.2262], device='cuda:0'), in_proj_covar=tensor([0.1111, 0.1025, 0.0882, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:0')
+2023-03-08 08:22:39,893 INFO [train.py:968] (0/2) Epoch 16, batch 19050, giga_loss[loss=0.28, simple_loss=0.3528, pruned_loss=0.1036, over 28831.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3605, pruned_loss=0.1063, over 5720363.88 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3376, pruned_loss=0.08917, over 5779831.76 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3613, pruned_loss=0.1076, over 5701748.06 frames. ], batch size: 66, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:22:55,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.620e+02 1.404e+03 1.723e+03 2.620e+03 7.865e+03, threshold=3.445e+03, percent-clipped=22.0
+2023-03-08 08:23:19,063 INFO [train.py:968] (0/2) Epoch 16, batch 19100, giga_loss[loss=0.2691, simple_loss=0.3442, pruned_loss=0.09705, over 28710.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3602, pruned_loss=0.1077, over 5711682.05 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3382, pruned_loss=0.08961, over 5771064.59 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3607, pruned_loss=0.1087, over 5704470.13 frames. ], batch size: 119, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:23:36,345 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:23:40,267 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-08 08:23:59,554 INFO [train.py:968] (0/2) Epoch 16, batch 19150, giga_loss[loss=0.2656, simple_loss=0.3328, pruned_loss=0.09917, over 28887.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3574, pruned_loss=0.1068, over 5707690.38 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3385, pruned_loss=0.08973, over 5770005.94 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.358, pruned_loss=0.1078, over 5701607.74 frames. ], batch size: 112, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:24:18,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.315e+02 1.257e+03 1.490e+03 1.894e+03 4.157e+03, threshold=2.979e+03, percent-clipped=3.0
+2023-03-08 08:24:35,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:24:41,405 INFO [train.py:968] (0/2) Epoch 16, batch 19200, giga_loss[loss=0.2519, simple_loss=0.3357, pruned_loss=0.08405, over 28792.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3552, pruned_loss=0.1058, over 5706213.89 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3387, pruned_loss=0.08983, over 5772119.47 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3558, pruned_loss=0.1069, over 5697533.29 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:24:42,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7355, 2.4167, 1.6072, 0.9467], device='cuda:0'), covar=tensor([0.6367, 0.2906, 0.3130, 0.5530], device='cuda:0'), in_proj_covar=tensor([0.1639, 0.1555, 0.1538, 0.1346], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 08:25:14,860 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:25:26,930 INFO [train.py:968] (0/2) Epoch 16, batch 19250, giga_loss[loss=0.2831, simple_loss=0.3551, pruned_loss=0.1056, over 28886.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3539, pruned_loss=0.1044, over 5715710.29 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3385, pruned_loss=0.0897, over 5774699.31 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3551, pruned_loss=0.1059, over 5704632.48 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:25:42,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.821e+02 1.217e+03 1.675e+03 2.339e+03 6.631e+03, threshold=3.351e+03, percent-clipped=15.0
+2023-03-08 08:25:48,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:25:59,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:26:01,795 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703243.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:26:06,878 INFO [train.py:968] (0/2) Epoch 16, batch 19300, giga_loss[loss=0.2614, simple_loss=0.3354, pruned_loss=0.09368, over 29030.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3518, pruned_loss=0.1024, over 5713844.99 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3384, pruned_loss=0.08956, over 5775876.89 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3533, pruned_loss=0.1042, over 5701499.81 frames. ], batch size: 106, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:26:54,503 INFO [train.py:968] (0/2) Epoch 16, batch 19350, libri_loss[loss=0.2321, simple_loss=0.3187, pruned_loss=0.0727, over 29575.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3475, pruned_loss=0.1001, over 5698472.72 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3384, pruned_loss=0.08947, over 5777931.42 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.349, pruned_loss=0.1018, over 5685591.73 frames. ], batch size: 76, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:27:10,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.083e+02 1.010e+03 1.222e+03 1.667e+03 4.788e+03, threshold=2.444e+03, percent-clipped=1.0
+2023-03-08 08:27:38,473 INFO [train.py:968] (0/2) Epoch 16, batch 19400, giga_loss[loss=0.2353, simple_loss=0.3181, pruned_loss=0.07624, over 28853.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3414, pruned_loss=0.09679, over 5690587.37 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3385, pruned_loss=0.08945, over 5778825.74 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3425, pruned_loss=0.09839, over 5677311.56 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:28:29,119 INFO [train.py:968] (0/2) Epoch 16, batch 19450, giga_loss[loss=0.2761, simple_loss=0.3306, pruned_loss=0.1107, over 26474.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3359, pruned_loss=0.09422, over 5686568.51 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3385, pruned_loss=0.08945, over 5778825.74 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3368, pruned_loss=0.09546, over 5676235.79 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:28:49,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.385e+02 9.399e+02 1.274e+03 1.686e+03 5.603e+03, threshold=2.549e+03, percent-clipped=10.0
+2023-03-08 08:28:58,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7952, 1.8160, 1.9922, 1.5372], device='cuda:0'), covar=tensor([0.1704, 0.2202, 0.1305, 0.1570], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0685, 0.0910, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 08:29:10,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:29:14,761 INFO [train.py:968] (0/2) Epoch 16, batch 19500, giga_loss[loss=0.2683, simple_loss=0.3458, pruned_loss=0.09541, over 28558.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.333, pruned_loss=0.0929, over 5658575.45 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3386, pruned_loss=0.08957, over 5780541.25 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3334, pruned_loss=0.09395, over 5645081.33 frames. ], batch size: 85, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:29:59,764 INFO [train.py:968] (0/2) Epoch 16, batch 19550, giga_loss[loss=0.2403, simple_loss=0.3146, pruned_loss=0.08294, over 28607.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3335, pruned_loss=0.09259, over 5668444.59 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3384, pruned_loss=0.08933, over 5782903.43 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3339, pruned_loss=0.09371, over 5653523.04 frames. ], batch size: 78, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:30:14,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:30:19,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.388e+02 1.005e+03 1.200e+03 1.623e+03 4.559e+03, threshold=2.401e+03, percent-clipped=6.0
+2023-03-08 08:30:20,018 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:30:46,155 INFO [train.py:968] (0/2) Epoch 16, batch 19600, giga_loss[loss=0.3561, simple_loss=0.3939, pruned_loss=0.1592, over 26664.00 frames. ], tot_loss[loss=0.259, simple_loss=0.333, pruned_loss=0.09249, over 5675024.15 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3382, pruned_loss=0.0892, over 5783505.91 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3335, pruned_loss=0.09348, over 5662372.04 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:30:54,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:02,833 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 08:31:19,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:21,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:29,268 INFO [train.py:968] (0/2) Epoch 16, batch 19650, giga_loss[loss=0.3381, simple_loss=0.3911, pruned_loss=0.1425, over 26700.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3328, pruned_loss=0.09265, over 5677744.29 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3387, pruned_loss=0.08936, over 5780975.19 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3327, pruned_loss=0.09332, over 5668774.04 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:31:31,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:39,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:40,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703618.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:31:42,993 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:45,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.473e+02 1.120e+03 1.442e+03 1.873e+03 7.099e+03, threshold=2.884e+03, percent-clipped=15.0
+2023-03-08 08:32:00,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5645, 4.4158, 4.1696, 2.1590], device='cuda:0'), covar=tensor([0.0493, 0.0621, 0.0617, 0.2018], device='cuda:0'), in_proj_covar=tensor([0.1117, 0.1026, 0.0885, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 08:32:07,751 INFO [train.py:968] (0/2) Epoch 16, batch 19700, giga_loss[loss=0.2505, simple_loss=0.3223, pruned_loss=0.0894, over 28951.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3315, pruned_loss=0.09188, over 5690002.68 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3396, pruned_loss=0.08971, over 5783362.62 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3304, pruned_loss=0.09216, over 5678482.24 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:32:16,599 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:19,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:39,243 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:42,814 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:43,925 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 08:32:48,678 INFO [train.py:968] (0/2) Epoch 16, batch 19750, giga_loss[loss=0.2421, simple_loss=0.3163, pruned_loss=0.08397, over 28931.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.329, pruned_loss=0.09055, over 5699723.74 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3398, pruned_loss=0.08972, over 5784498.66 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3279, pruned_loss=0.09077, over 5688754.86 frames. ], batch size: 112, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:32:51,345 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:54,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:54,640 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-08 08:33:09,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.346e+02 9.818e+02 1.263e+03 1.736e+03 7.297e+03, threshold=2.526e+03, percent-clipped=6.0
+2023-03-08 08:33:19,074 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:33:22,728 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-08 08:33:29,551 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:33:31,421 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:33:31,851 INFO [train.py:968] (0/2) Epoch 16, batch 19800, libri_loss[loss=0.1984, simple_loss=0.2909, pruned_loss=0.05298, over 28575.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3264, pruned_loss=0.08898, over 5694284.56 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3397, pruned_loss=0.08937, over 5776288.30 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3254, pruned_loss=0.08952, over 5689566.49 frames. ], batch size: 63, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:33:37,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:33:38,909 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703761.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:33:39,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:33:40,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5992, 1.6431, 1.3064, 1.2429], device='cuda:0'), covar=tensor([0.0901, 0.0544, 0.1007, 0.1125], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0440, 0.0507, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 08:33:40,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703764.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:33:52,924 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:34:03,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:34:04,359 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703793.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:34:09,969 INFO [train.py:968] (0/2) Epoch 16, batch 19850, libri_loss[loss=0.2763, simple_loss=0.3715, pruned_loss=0.09058, over 29538.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3253, pruned_loss=0.08826, over 5708391.81 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3407, pruned_loss=0.08952, over 5777807.79 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3229, pruned_loss=0.08852, over 5700341.84 frames. ], batch size: 83, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:34:28,549 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.606e+02 1.097e+03 1.412e+03 2.033e+03 6.170e+03, threshold=2.823e+03, percent-clipped=18.0
+2023-03-08 08:34:50,955 INFO [train.py:968] (0/2) Epoch 16, batch 19900, libri_loss[loss=0.2258, simple_loss=0.3207, pruned_loss=0.06539, over 29559.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3233, pruned_loss=0.08733, over 5718191.83 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3412, pruned_loss=0.08965, over 5780411.30 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3207, pruned_loss=0.0874, over 5708259.04 frames. ], batch size: 76, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:35:29,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:35:32,092 INFO [train.py:968] (0/2) Epoch 16, batch 19950, giga_loss[loss=0.252, simple_loss=0.3222, pruned_loss=0.09091, over 28627.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3218, pruned_loss=0.08705, over 5715738.04 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3417, pruned_loss=0.08994, over 5782368.60 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3191, pruned_loss=0.08684, over 5705446.96 frames. ], batch size: 242, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:35:48,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.191e+02 1.046e+03 1.228e+03 1.607e+03 6.305e+03, threshold=2.457e+03, percent-clipped=3.0
+2023-03-08 08:36:11,288 INFO [train.py:968] (0/2) Epoch 16, batch 20000, giga_loss[loss=0.2285, simple_loss=0.3069, pruned_loss=0.07505, over 28963.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3209, pruned_loss=0.08664, over 5717285.56 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3419, pruned_loss=0.08982, over 5784521.53 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3181, pruned_loss=0.08652, over 5706100.47 frames. ], batch size: 213, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:36:12,671 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:36:46,758 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-704000.pt
+2023-03-08 08:36:47,809 INFO [train.py:968] (0/2) Epoch 16, batch 20050, giga_loss[loss=0.2328, simple_loss=0.2975, pruned_loss=0.08408, over 28616.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3197, pruned_loss=0.08554, over 5722347.14 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3423, pruned_loss=0.08993, over 5782177.69 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3167, pruned_loss=0.08527, over 5714490.83 frames. ], batch size: 85, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:37:07,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.232e+02 1.009e+03 1.268e+03 1.800e+03 6.526e+03, threshold=2.535e+03, percent-clipped=8.0
+2023-03-08 08:37:21,404 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=704041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:37:24,323 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=704044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:37:31,224 INFO [train.py:968] (0/2) Epoch 16, batch 20100, giga_loss[loss=0.3257, simple_loss=0.3759, pruned_loss=0.1378, over 23817.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3249, pruned_loss=0.08909, over 5714470.84 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3428, pruned_loss=0.09006, over 5782755.10 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3217, pruned_loss=0.08871, over 5706565.08 frames. ], batch size: 705, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:37:44,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=704064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:37:51,948 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=704073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:37:56,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=704078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:38:11,940 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 08:38:20,875 INFO [train.py:968] (0/2) Epoch 16, batch 20150, giga_loss[loss=0.2722, simple_loss=0.3403, pruned_loss=0.102, over 28632.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3309, pruned_loss=0.09307, over 5705613.38 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3429, pruned_loss=0.0901, over 5783988.47 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3281, pruned_loss=0.09276, over 5697575.75 frames. ], batch size: 85, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:38:42,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.807e+02 1.262e+03 1.596e+03 1.964e+03 5.713e+03, threshold=3.192e+03, percent-clipped=12.0
+2023-03-08 08:38:46,901 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 08:38:54,852 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-08 08:39:10,107 INFO [train.py:968] (0/2) Epoch 16, batch 20200, giga_loss[loss=0.328, simple_loss=0.3813, pruned_loss=0.1374, over 28893.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3391, pruned_loss=0.09842, over 5702810.09 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3429, pruned_loss=0.08997, over 5783412.81 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3368, pruned_loss=0.09843, over 5695357.14 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:39:56,478 INFO [train.py:968] (0/2) Epoch 16, batch 20250, giga_loss[loss=0.2652, simple_loss=0.3404, pruned_loss=0.09496, over 28868.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3445, pruned_loss=0.1012, over 5699429.93 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3431, pruned_loss=0.09001, over 5785887.08 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3425, pruned_loss=0.1014, over 5689610.39 frames. ], batch size: 112, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:40:02,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=704207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:40:06,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=704210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:40:07,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5375, 1.7267, 1.6392, 1.4111], device='cuda:0'), covar=tensor([0.2611, 0.2125, 0.1749, 0.2258], device='cuda:0'), in_proj_covar=tensor([0.1840, 0.1753, 0.1693, 0.1842], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 08:40:20,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.249e+02 1.105e+03 1.374e+03 1.973e+03 5.448e+03, threshold=2.749e+03, percent-clipped=4.0
+2023-03-08 08:40:32,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=704239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:40:42,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3772, 1.5212, 1.4228, 1.5120], device='cuda:0'), covar=tensor([0.0800, 0.0339, 0.0325, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 08:40:43,167 INFO [train.py:968] (0/2) Epoch 16, batch 20300, giga_loss[loss=0.2565, simple_loss=0.3371, pruned_loss=0.0879, over 28548.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3498, pruned_loss=0.1036, over 5695548.87 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3434, pruned_loss=0.09027, over 5787630.73 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3481, pruned_loss=0.1039, over 5682975.39 frames. ], batch size: 60, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:41:03,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4365, 1.5190, 1.1436, 1.1649], device='cuda:0'), covar=tensor([0.0866, 0.0507, 0.1046, 0.1147], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0442, 0.0510, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 08:41:23,910 INFO [train.py:968] (0/2) Epoch 16, batch 20350, giga_loss[loss=0.3088, simple_loss=0.3796, pruned_loss=0.119, over 29039.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3536, pruned_loss=0.1048, over 5702505.07 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3433, pruned_loss=0.09035, over 5791892.92 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3526, pruned_loss=0.1056, over 5684736.49 frames. ], batch size: 155, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:41:47,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.245e+02 1.234e+03 1.665e+03 2.306e+03 4.317e+03, threshold=3.331e+03, percent-clipped=15.0
+2023-03-08 08:41:49,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=704328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:42:08,570 INFO [train.py:968] (0/2) Epoch 16, batch 20400, giga_loss[loss=0.2423, simple_loss=0.3317, pruned_loss=0.07646, over 28950.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3586, pruned_loss=0.1077, over 5700543.69 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3435, pruned_loss=0.09047, over 5788707.64 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3578, pruned_loss=0.1085, over 5687673.49 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:42:28,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3819, 2.5922, 2.5955, 2.0565], device='cuda:0'), covar=tensor([0.1571, 0.1855, 0.1559, 0.1886], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0730, 0.0688, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 08:42:50,928 INFO [train.py:968] (0/2) Epoch 16, batch 20450, giga_loss[loss=0.2195, simple_loss=0.3137, pruned_loss=0.06263, over 28911.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3544, pruned_loss=0.1047, over 5701125.22 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3442, pruned_loss=0.09093, over 5793128.95 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3538, pruned_loss=0.1056, over 5683284.93 frames. ], batch size: 164, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:43:10,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.632e+02 1.115e+03 1.434e+03 2.124e+03 4.520e+03, threshold=2.868e+03, percent-clipped=5.0
+2023-03-08 08:43:31,376 INFO [train.py:968] (0/2) Epoch 16, batch 20500, giga_loss[loss=0.2714, simple_loss=0.3476, pruned_loss=0.09761, over 28612.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3509, pruned_loss=0.1014, over 5713887.69 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3438, pruned_loss=0.09078, over 5796146.17 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.351, pruned_loss=0.1028, over 5692644.57 frames. ], batch size: 307, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:43:33,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=704453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:43:50,371 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=704471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:43:52,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=704474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:44:15,549 INFO [train.py:968] (0/2) Epoch 16, batch 20550, giga_loss[loss=0.2923, simple_loss=0.3638, pruned_loss=0.1104, over 27522.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3504, pruned_loss=0.1008, over 5697701.15 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3441, pruned_loss=0.09107, over 5787759.00 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3503, pruned_loss=0.1017, over 5686871.66 frames. ], batch size: 472, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:44:16,975 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=704503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:44:33,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.767e+02 1.188e+03 1.577e+03 1.936e+03 6.568e+03, threshold=3.154e+03, percent-clipped=11.0
+2023-03-08 08:44:55,918 INFO [train.py:968] (0/2) Epoch 16, batch 20600, libri_loss[loss=0.2398, simple_loss=0.3303, pruned_loss=0.07463, over 29555.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3505, pruned_loss=0.09992, over 5678383.62 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3447, pruned_loss=0.09144, over 5765245.37 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3502, pruned_loss=0.1007, over 5685443.64 frames. ], batch size: 80, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:45:24,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1674, 1.2584, 3.1717, 2.8735], device='cuda:0'), covar=tensor([0.1467, 0.2544, 0.0453, 0.1890], device='cuda:0'), in_proj_covar=tensor([0.0707, 0.0615, 0.0902, 0.0826], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 08:45:38,463 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=704596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:45:39,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=704597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:45:40,354 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=704599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:45:41,497 INFO [train.py:968] (0/2) Epoch 16, batch 20650, giga_loss[loss=0.2873, simple_loss=0.3589, pruned_loss=0.1079, over 28858.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3529, pruned_loss=0.1016, over 5671676.99 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3448, pruned_loss=0.09149, over 5753398.14 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3526, pruned_loss=0.1022, over 5686263.43 frames. ], batch size: 106, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:45:45,099 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=704605.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:46:01,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.098e+02 1.166e+03 1.408e+03 1.775e+03 3.186e+03, threshold=2.816e+03, percent-clipped=2.0
+2023-03-08 08:46:06,863 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=704628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:46:26,427 INFO [train.py:968] (0/2) Epoch 16, batch 20700, giga_loss[loss=0.2793, simple_loss=0.3558, pruned_loss=0.1014, over 28684.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3547, pruned_loss=0.1031, over 5678321.26 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3452, pruned_loss=0.09169, over 5755298.57 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3542, pruned_loss=0.1035, over 5687446.05 frames. ], batch size: 262, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:46:58,497 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-08 08:47:13,542 INFO [train.py:968] (0/2) Epoch 16, batch 20750, giga_loss[loss=0.3205, simple_loss=0.3856, pruned_loss=0.1277, over 28533.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.356, pruned_loss=0.1041, over 5696221.72 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3454, pruned_loss=0.09181, over 5758539.39 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3556, pruned_loss=0.1045, over 5699365.39 frames. ], batch size: 336, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:47:33,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.151e+02 1.261e+03 1.565e+03 2.434e+03 6.988e+03, threshold=3.131e+03, percent-clipped=18.0
+2023-03-08 08:47:47,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6077, 2.3638, 1.5495, 0.7767], device='cuda:0'), covar=tensor([0.6358, 0.2934, 0.3385, 0.5971], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1553, 0.1535, 0.1346], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 08:47:57,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3570, 1.4359, 1.3921, 1.3706], device='cuda:0'), covar=tensor([0.2248, 0.2159, 0.2024, 0.2088], device='cuda:0'), in_proj_covar=tensor([0.1852, 0.1766, 0.1699, 0.1849], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 08:47:58,737 INFO [train.py:968] (0/2) Epoch 16, batch 20800, giga_loss[loss=0.2757, simple_loss=0.3477, pruned_loss=0.1019, over 28751.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3566, pruned_loss=0.1051, over 5695214.68 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3453, pruned_loss=0.09179, over 5759126.73 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3566, pruned_loss=0.1058, over 5695661.48 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:48:23,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4123, 3.5709, 1.4594, 1.5037], device='cuda:0'), covar=tensor([0.1061, 0.0285, 0.0983, 0.1470], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0528, 0.0362, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 08:48:38,172 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9116, 3.7540, 3.5239, 1.6854], device='cuda:0'), covar=tensor([0.0638, 0.0744, 0.0708, 0.2330], device='cuda:0'), in_proj_covar=tensor([0.1124, 0.1041, 0.0893, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 08:48:38,713 INFO [train.py:968] (0/2) Epoch 16, batch 20850, giga_loss[loss=0.2881, simple_loss=0.3562, pruned_loss=0.11, over 28342.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3572, pruned_loss=0.1054, over 5704073.20 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3455, pruned_loss=0.09177, over 5761363.37 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3573, pruned_loss=0.1063, over 5701216.38 frames. ], batch size: 368, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:48:39,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-08 08:48:57,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.477e+02 1.124e+03 1.490e+03 1.967e+03 5.603e+03, threshold=2.980e+03, percent-clipped=6.0
+2023-03-08 08:49:05,166 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=704835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:49:17,487 INFO [train.py:968] (0/2) Epoch 16, batch 20900, giga_loss[loss=0.3849, simple_loss=0.435, pruned_loss=0.1674, over 28752.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.358, pruned_loss=0.1054, over 5711976.47 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3456, pruned_loss=0.0918, over 5765581.79 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3584, pruned_loss=0.1064, over 5704400.49 frames. ], batch size: 284, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:49:55,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6667, 1.9072, 1.3064, 1.5760], device='cuda:0'), covar=tensor([0.0803, 0.0472, 0.0984, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0441, 0.0509, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 08:49:58,424 INFO [train.py:968] (0/2) Epoch 16, batch 20950, giga_loss[loss=0.2823, simple_loss=0.3601, pruned_loss=0.1023, over 28976.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3585, pruned_loss=0.1047, over 5704768.66 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3458, pruned_loss=0.09199, over 5757030.58 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3587, pruned_loss=0.1055, over 5705768.73 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:50:19,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.502e+02 1.093e+03 1.360e+03 2.055e+03 4.478e+03, threshold=2.719e+03, percent-clipped=8.0
+2023-03-08 08:50:21,175 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-08 08:50:30,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3492, 1.0391, 3.8266, 3.3566], device='cuda:0'), covar=tensor([0.1975, 0.3205, 0.0744, 0.1028], device='cuda:0'), in_proj_covar=tensor([0.0704, 0.0612, 0.0899, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 08:50:39,218 INFO [train.py:968] (0/2) Epoch 16, batch 21000, giga_loss[loss=0.2783, simple_loss=0.3541, pruned_loss=0.1012, over 28675.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3579, pruned_loss=0.1037, over 5711902.67 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3461, pruned_loss=0.09218, over 5759388.42 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3582, pruned_loss=0.1045, over 5709189.14 frames. ], batch size: 78, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:50:39,223 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 08:50:48,174 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2109, simple_loss=0.3179, pruned_loss=0.05197, over 944034.00 frames. 
+2023-03-08 08:50:48,174 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 08:51:04,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=704972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:51:10,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=704980.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:51:27,973 INFO [train.py:968] (0/2) Epoch 16, batch 21050, giga_loss[loss=0.2711, simple_loss=0.3474, pruned_loss=0.09736, over 28817.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3549, pruned_loss=0.1023, over 5702983.72 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.346, pruned_loss=0.09229, over 5752717.81 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3555, pruned_loss=0.1031, over 5705549.52 frames. ], batch size: 99, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:51:47,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.243e+02 1.168e+03 1.496e+03 2.024e+03 4.973e+03, threshold=2.991e+03, percent-clipped=10.0
+2023-03-08 08:52:09,129 INFO [train.py:968] (0/2) Epoch 16, batch 21100, giga_loss[loss=0.2595, simple_loss=0.3364, pruned_loss=0.09134, over 28908.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3528, pruned_loss=0.1016, over 5707685.99 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3461, pruned_loss=0.09238, over 5754806.34 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3532, pruned_loss=0.1023, over 5707102.25 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:52:48,806 INFO [train.py:968] (0/2) Epoch 16, batch 21150, giga_loss[loss=0.2613, simple_loss=0.3341, pruned_loss=0.09427, over 28808.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3511, pruned_loss=0.1005, over 5703648.33 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3465, pruned_loss=0.09265, over 5746586.39 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3511, pruned_loss=0.101, over 5709419.18 frames. ], batch size: 99, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:52:52,613 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.34 vs. limit=5.0
+2023-03-08 08:52:58,942 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3812, 1.8332, 1.4520, 1.5120], device='cuda:0'), covar=tensor([0.0679, 0.0371, 0.0305, 0.0706], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 08:52:59,611 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:53:02,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:53:06,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705123.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:53:09,266 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.783e+02 1.101e+03 1.260e+03 1.735e+03 3.998e+03, threshold=2.520e+03, percent-clipped=3.0
+2023-03-08 08:53:09,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705126.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:53:27,425 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:53:29,990 INFO [train.py:968] (0/2) Epoch 16, batch 21200, libri_loss[loss=0.2633, simple_loss=0.3396, pruned_loss=0.09348, over 29541.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3522, pruned_loss=0.1016, over 5708564.16 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3471, pruned_loss=0.09302, over 5749437.12 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3519, pruned_loss=0.102, over 5708816.44 frames. ], batch size: 79, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:53:33,087 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705155.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:53:36,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3454, 1.5917, 1.3080, 1.4152], device='cuda:0'), covar=tensor([0.2842, 0.2682, 0.3030, 0.2356], device='cuda:0'), in_proj_covar=tensor([0.1398, 0.1024, 0.1241, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 08:54:09,469 INFO [train.py:968] (0/2) Epoch 16, batch 21250, giga_loss[loss=0.2565, simple_loss=0.3386, pruned_loss=0.08724, over 28467.00 frames. ], tot_loss[loss=0.278, simple_loss=0.352, pruned_loss=0.102, over 5706545.60 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3465, pruned_loss=0.093, over 5752186.33 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3524, pruned_loss=0.1025, over 5703078.98 frames. ], batch size: 65, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:54:17,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=705210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:54:29,237 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=705224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:54:31,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.979e+02 1.053e+03 1.374e+03 1.815e+03 6.097e+03, threshold=2.749e+03, percent-clipped=14.0
+2023-03-08 08:54:54,505 INFO [train.py:968] (0/2) Epoch 16, batch 21300, giga_loss[loss=0.2477, simple_loss=0.332, pruned_loss=0.08172, over 28853.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3514, pruned_loss=0.1007, over 5714554.64 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.347, pruned_loss=0.0933, over 5753307.41 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3514, pruned_loss=0.101, over 5710299.30 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:55:07,389 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-08 08:55:19,143 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=705279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:55:35,781 INFO [train.py:968] (0/2) Epoch 16, batch 21350, giga_loss[loss=0.29, simple_loss=0.3594, pruned_loss=0.1103, over 28604.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3505, pruned_loss=0.09974, over 5707725.56 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3469, pruned_loss=0.0933, over 5754114.87 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3506, pruned_loss=0.1, over 5703450.82 frames. ], batch size: 336, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:55:47,363 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=705312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:55:56,292 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=705323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:55:58,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.419e+02 1.075e+03 1.376e+03 1.907e+03 8.123e+03, threshold=2.752e+03, percent-clipped=8.0
+2023-03-08 08:56:18,838 INFO [train.py:968] (0/2) Epoch 16, batch 21400, giga_loss[loss=0.2821, simple_loss=0.3483, pruned_loss=0.108, over 29144.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3507, pruned_loss=0.1006, over 5706720.08 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3474, pruned_loss=0.09371, over 5755645.82 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3503, pruned_loss=0.1005, over 5701537.54 frames. ], batch size: 128, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:56:20,569 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:56:23,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:56:47,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:56:58,681 INFO [train.py:968] (0/2) Epoch 16, batch 21450, giga_loss[loss=0.2692, simple_loss=0.3406, pruned_loss=0.09895, over 28618.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3492, pruned_loss=0.1004, over 5705157.71 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3477, pruned_loss=0.09399, over 5757942.51 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3486, pruned_loss=0.1002, over 5697786.32 frames. ], batch size: 85, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:57:19,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.630e+02 1.042e+03 1.352e+03 2.098e+03 6.310e+03, threshold=2.703e+03, percent-clipped=8.0
+2023-03-08 08:57:38,697 INFO [train.py:968] (0/2) Epoch 16, batch 21500, giga_loss[loss=0.2426, simple_loss=0.3166, pruned_loss=0.0843, over 28543.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3451, pruned_loss=0.09812, over 5706311.77 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3475, pruned_loss=0.09401, over 5761231.24 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3449, pruned_loss=0.0981, over 5696145.74 frames. ], batch size: 85, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:58:19,964 INFO [train.py:968] (0/2) Epoch 16, batch 21550, giga_loss[loss=0.3125, simple_loss=0.3849, pruned_loss=0.1201, over 28876.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3445, pruned_loss=0.09824, over 5693988.30 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3478, pruned_loss=0.09435, over 5754962.59 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.344, pruned_loss=0.09797, over 5690668.69 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:58:29,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4779, 1.6791, 1.6466, 1.4413], device='cuda:0'), covar=tensor([0.2718, 0.2307, 0.1547, 0.2214], device='cuda:0'), in_proj_covar=tensor([0.1839, 0.1767, 0.1691, 0.1838], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 08:58:31,645 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-08 08:58:41,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.004e+02 1.148e+03 1.378e+03 1.743e+03 4.171e+03, threshold=2.757e+03, percent-clipped=5.0
+2023-03-08 08:59:01,021 INFO [train.py:968] (0/2) Epoch 16, batch 21600, libri_loss[loss=0.2377, simple_loss=0.3159, pruned_loss=0.07972, over 29631.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3445, pruned_loss=0.09887, over 5702582.35 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.348, pruned_loss=0.09469, over 5760440.78 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3438, pruned_loss=0.0985, over 5692803.72 frames. ], batch size: 73, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:59:43,874 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3363, 1.5787, 1.3526, 1.5114], device='cuda:0'), covar=tensor([0.0731, 0.0313, 0.0329, 0.0818], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 08:59:44,418 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=705599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:59:45,544 INFO [train.py:968] (0/2) Epoch 16, batch 21650, giga_loss[loss=0.2478, simple_loss=0.326, pruned_loss=0.08482, over 29063.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3412, pruned_loss=0.09728, over 5702057.59 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3479, pruned_loss=0.09475, over 5761766.57 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3406, pruned_loss=0.09696, over 5692518.15 frames. ], batch size: 155, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 09:00:06,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.378e+02 1.124e+03 1.428e+03 1.839e+03 8.063e+03, threshold=2.856e+03, percent-clipped=12.0
+2023-03-08 09:00:19,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5049, 3.5257, 1.7564, 1.5676], device='cuda:0'), covar=tensor([0.0934, 0.0395, 0.0863, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0524, 0.0359, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 09:00:25,191 INFO [train.py:968] (0/2) Epoch 16, batch 21700, giga_loss[loss=0.2754, simple_loss=0.3423, pruned_loss=0.1042, over 28258.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3386, pruned_loss=0.09629, over 5708574.44 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.348, pruned_loss=0.09487, over 5765219.39 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.338, pruned_loss=0.09597, over 5696493.35 frames. ], batch size: 65, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 09:00:27,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=705654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:00:53,634 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=705687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:01:01,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=705698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:01:04,008 INFO [train.py:968] (0/2) Epoch 16, batch 21750, giga_loss[loss=0.2254, simple_loss=0.3092, pruned_loss=0.07076, over 28941.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3373, pruned_loss=0.09576, over 5718498.43 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3484, pruned_loss=0.09528, over 5768752.15 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3361, pruned_loss=0.09516, over 5704559.85 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 09:01:25,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.086e+02 1.009e+03 1.257e+03 1.697e+03 6.162e+03, threshold=2.514e+03, percent-clipped=3.0
+2023-03-08 09:01:35,774 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:01:38,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:01:44,167 INFO [train.py:968] (0/2) Epoch 16, batch 21800, giga_loss[loss=0.2534, simple_loss=0.3374, pruned_loss=0.08472, over 28982.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3368, pruned_loss=0.09613, over 5719131.06 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3488, pruned_loss=0.09578, over 5770462.78 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3353, pruned_loss=0.09519, over 5705402.73 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:02:05,724 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:02:05,827 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5739, 1.8021, 1.4493, 1.6419], device='cuda:0'), covar=tensor([0.2547, 0.2547, 0.2912, 0.2309], device='cuda:0'), in_proj_covar=tensor([0.1396, 0.1019, 0.1236, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 09:02:20,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2896, 3.2619, 1.3921, 1.4746], device='cuda:0'), covar=tensor([0.0983, 0.0310, 0.0974, 0.1351], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0525, 0.0359, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 09:02:25,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:02:28,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:02:29,030 INFO [train.py:968] (0/2) Epoch 16, batch 21850, giga_loss[loss=0.2455, simple_loss=0.3299, pruned_loss=0.0806, over 28927.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3381, pruned_loss=0.09616, over 5718748.51 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3491, pruned_loss=0.09606, over 5771236.20 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3364, pruned_loss=0.09517, over 5706596.24 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:02:51,894 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.860e+02 1.085e+03 1.321e+03 1.588e+03 6.501e+03, threshold=2.641e+03, percent-clipped=8.0
+2023-03-08 09:02:52,652 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:02:53,571 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:02:55,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:03:03,522 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:03:07,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:03:12,223 INFO [train.py:968] (0/2) Epoch 16, batch 21900, giga_loss[loss=0.2906, simple_loss=0.3669, pruned_loss=0.1071, over 28952.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3411, pruned_loss=0.09743, over 5711737.37 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.349, pruned_loss=0.09626, over 5774022.04 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3397, pruned_loss=0.09648, over 5698569.33 frames. ], batch size: 213, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:03:20,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:03:31,814 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:03:54,282 INFO [train.py:968] (0/2) Epoch 16, batch 21950, giga_loss[loss=0.2937, simple_loss=0.3615, pruned_loss=0.1129, over 28820.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3448, pruned_loss=0.09877, over 5707456.21 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3491, pruned_loss=0.09646, over 5777512.02 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3434, pruned_loss=0.09788, over 5692484.09 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:04:01,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3156, 1.8842, 1.5453, 1.7933], device='cuda:0'), covar=tensor([0.0738, 0.0281, 0.0307, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:0')
+2023-03-08 09:04:06,859 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=705915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:04:17,239 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.143e+02 1.083e+03 1.367e+03 1.963e+03 6.815e+03, threshold=2.735e+03, percent-clipped=10.0
+2023-03-08 09:04:22,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9012, 2.8698, 1.8901, 1.0372], device='cuda:0'), covar=tensor([0.7407, 0.3009, 0.3628, 0.6499], device='cuda:0'), in_proj_covar=tensor([0.1640, 0.1549, 0.1540, 0.1353], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 09:04:39,790 INFO [train.py:968] (0/2) Epoch 16, batch 22000, giga_loss[loss=0.2515, simple_loss=0.3374, pruned_loss=0.0828, over 28725.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3452, pruned_loss=0.09816, over 5704140.93 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3492, pruned_loss=0.09658, over 5766246.70 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.344, pruned_loss=0.09737, over 5700954.41 frames. ], batch size: 262, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 09:05:19,862 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-706000.pt
+2023-03-08 09:05:20,743 INFO [train.py:968] (0/2) Epoch 16, batch 22050, giga_loss[loss=0.2275, simple_loss=0.312, pruned_loss=0.07152, over 28952.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3446, pruned_loss=0.09743, over 5699248.21 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3503, pruned_loss=0.0979, over 5756562.11 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3425, pruned_loss=0.0956, over 5702858.85 frames. ], batch size: 136, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 09:05:45,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.713e+02 1.084e+03 1.376e+03 1.886e+03 8.524e+03, threshold=2.753e+03, percent-clipped=6.0
+2023-03-08 09:06:04,916 INFO [train.py:968] (0/2) Epoch 16, batch 22100, giga_loss[loss=0.3189, simple_loss=0.3669, pruned_loss=0.1354, over 23946.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3446, pruned_loss=0.09797, over 5692253.84 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3503, pruned_loss=0.0983, over 5759208.82 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3426, pruned_loss=0.09608, over 5690636.80 frames. ], batch size: 705, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:06:22,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3877, 1.6846, 1.3265, 1.4452], device='cuda:0'), covar=tensor([0.2493, 0.2414, 0.2833, 0.2273], device='cuda:0'), in_proj_covar=tensor([0.1398, 0.1018, 0.1240, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 09:06:44,212 INFO [train.py:968] (0/2) Epoch 16, batch 22150, libri_loss[loss=0.2547, simple_loss=0.3183, pruned_loss=0.09554, over 29425.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3451, pruned_loss=0.09849, over 5693986.73 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3507, pruned_loss=0.09872, over 5754126.69 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3429, pruned_loss=0.09657, over 5695860.50 frames. ], batch size: 67, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 09:06:50,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5428, 1.6469, 1.5697, 1.4879], device='cuda:0'), covar=tensor([0.1565, 0.2029, 0.2147, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.0452, 0.0730, 0.0686, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 09:06:51,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2376, 1.2999, 1.3197, 1.2484], device='cuda:0'), covar=tensor([0.2101, 0.2016, 0.1751, 0.1911], device='cuda:0'), in_proj_covar=tensor([0.1832, 0.1762, 0.1697, 0.1835], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 09:07:07,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.016e+02 1.274e+03 1.678e+03 2.274e+03 8.160e+03, threshold=3.355e+03, percent-clipped=13.0
+2023-03-08 09:07:21,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0437, 2.2175, 1.5400, 1.7492], device='cuda:0'), covar=tensor([0.0854, 0.0645, 0.1008, 0.1148], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0439, 0.0505, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 09:07:25,658 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=706150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:07:26,122 INFO [train.py:968] (0/2) Epoch 16, batch 22200, giga_loss[loss=0.265, simple_loss=0.3434, pruned_loss=0.09331, over 28834.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3458, pruned_loss=0.09906, over 5692855.76 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3506, pruned_loss=0.09873, over 5753508.06 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3442, pruned_loss=0.09755, over 5694609.23 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 09:07:35,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-08 09:07:50,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0209, 1.2189, 3.3489, 2.9732], device='cuda:0'), covar=tensor([0.1706, 0.2732, 0.0474, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0701, 0.0614, 0.0897, 0.0821], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 09:08:10,699 INFO [train.py:968] (0/2) Epoch 16, batch 22250, giga_loss[loss=0.2902, simple_loss=0.3593, pruned_loss=0.1105, over 28884.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3497, pruned_loss=0.101, over 5702917.56 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3511, pruned_loss=0.09898, over 5755156.51 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3481, pruned_loss=0.09965, over 5702338.59 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 09:08:28,468 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 09:08:30,713 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-08 09:08:33,063 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.707e+02 1.213e+03 1.415e+03 1.828e+03 3.564e+03, threshold=2.830e+03, percent-clipped=1.0
+2023-03-08 09:08:52,909 INFO [train.py:968] (0/2) Epoch 16, batch 22300, giga_loss[loss=0.2768, simple_loss=0.3525, pruned_loss=0.1005, over 28881.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3522, pruned_loss=0.1023, over 5707021.40 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.351, pruned_loss=0.09905, over 5756766.02 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3509, pruned_loss=0.1012, over 5704658.48 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 09:09:02,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5342, 4.6595, 1.8000, 1.7480], device='cuda:0'), covar=tensor([0.0964, 0.0306, 0.0876, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0529, 0.0361, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 09:09:23,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=706290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:09:32,203 INFO [train.py:968] (0/2) Epoch 16, batch 22350, giga_loss[loss=0.2689, simple_loss=0.3401, pruned_loss=0.09888, over 28712.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3539, pruned_loss=0.1033, over 5714433.68 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3518, pruned_loss=0.0996, over 5758761.89 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3522, pruned_loss=0.1019, over 5709881.74 frames. ], batch size: 99, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 09:09:42,304 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 09:09:54,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.295e+02 1.325e+03 1.661e+03 2.388e+03 4.933e+03, threshold=3.321e+03, percent-clipped=14.0
+2023-03-08 09:10:11,637 INFO [train.py:968] (0/2) Epoch 16, batch 22400, giga_loss[loss=0.2654, simple_loss=0.3414, pruned_loss=0.09471, over 28717.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3536, pruned_loss=0.1028, over 5723673.65 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3525, pruned_loss=0.1001, over 5761629.53 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3517, pruned_loss=0.1013, over 5716646.09 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:10:15,636 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=706353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:10:16,266 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5217, 1.6440, 1.0807, 1.4550], device='cuda:0'), covar=tensor([0.1110, 0.0926, 0.1583, 0.1407], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0441, 0.0506, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 09:10:53,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5422, 1.5250, 1.2019, 1.2023], device='cuda:0'), covar=tensor([0.0673, 0.0499, 0.0896, 0.1175], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0440, 0.0505, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 09:10:58,603 INFO [train.py:968] (0/2) Epoch 16, batch 22450, giga_loss[loss=0.2812, simple_loss=0.3603, pruned_loss=0.1011, over 28758.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3539, pruned_loss=0.1029, over 5717200.91 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3524, pruned_loss=0.1001, over 5763415.86 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3524, pruned_loss=0.1018, over 5709581.23 frames. ], batch size: 284, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:11:00,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2242, 4.0355, 3.8015, 1.9347], device='cuda:0'), covar=tensor([0.0554, 0.0684, 0.0678, 0.2097], device='cuda:0'), in_proj_covar=tensor([0.1124, 0.1037, 0.0893, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 09:11:24,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.006e+02 1.166e+03 1.489e+03 2.049e+03 6.142e+03, threshold=2.978e+03, percent-clipped=7.0
+2023-03-08 09:11:26,552 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=706433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:11:29,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=706436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:11:40,901 INFO [train.py:968] (0/2) Epoch 16, batch 22500, giga_loss[loss=0.2538, simple_loss=0.3322, pruned_loss=0.08776, over 28684.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3523, pruned_loss=0.1022, over 5710018.00 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3529, pruned_loss=0.1007, over 5756059.51 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3506, pruned_loss=0.1008, over 5709919.02 frames. ], batch size: 307, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:11:51,009 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=706465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:12:22,244 INFO [train.py:968] (0/2) Epoch 16, batch 22550, giga_loss[loss=0.2238, simple_loss=0.313, pruned_loss=0.06735, over 28893.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3504, pruned_loss=0.1014, over 5709591.24 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3536, pruned_loss=0.1014, over 5749055.93 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3484, pruned_loss=0.09972, over 5714226.49 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:12:27,636 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=706507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:12:41,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=706525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:12:46,187 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.859e+02 1.193e+03 1.525e+03 2.324e+03 1.427e+04, threshold=3.050e+03, percent-clipped=13.0
+2023-03-08 09:13:01,523 INFO [train.py:968] (0/2) Epoch 16, batch 22600, giga_loss[loss=0.2207, simple_loss=0.299, pruned_loss=0.07114, over 28347.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3476, pruned_loss=0.1001, over 5713819.24 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3542, pruned_loss=0.1021, over 5754439.51 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3453, pruned_loss=0.09805, over 5711601.07 frames. ], batch size: 77, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:13:32,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6437, 1.4799, 4.5871, 3.6911], device='cuda:0'), covar=tensor([0.1509, 0.2577, 0.0349, 0.0791], device='cuda:0'), in_proj_covar=tensor([0.0703, 0.0614, 0.0898, 0.0822], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 09:13:38,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6931, 1.3653, 4.7702, 3.5527], device='cuda:0'), covar=tensor([0.1515, 0.2746, 0.0345, 0.0842], device='cuda:0'), in_proj_covar=tensor([0.0704, 0.0614, 0.0898, 0.0822], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 09:13:44,022 INFO [train.py:968] (0/2) Epoch 16, batch 22650, giga_loss[loss=0.312, simple_loss=0.3845, pruned_loss=0.1198, over 28659.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3464, pruned_loss=0.099, over 5713628.00 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3542, pruned_loss=0.1023, over 5754315.68 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3446, pruned_loss=0.09724, over 5711665.02 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:14:09,927 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.613e+02 1.186e+03 1.482e+03 1.951e+03 8.208e+03, threshold=2.964e+03, percent-clipped=8.0
+2023-03-08 09:14:21,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2681, 1.1804, 1.1139, 1.3937], device='cuda:0'), covar=tensor([0.0742, 0.0338, 0.0351, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:0')
+2023-03-08 09:14:28,328 INFO [train.py:968] (0/2) Epoch 16, batch 22700, giga_loss[loss=0.3144, simple_loss=0.3915, pruned_loss=0.1187, over 28225.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3494, pruned_loss=0.09939, over 5712001.31 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3546, pruned_loss=0.1026, over 5756151.17 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3474, pruned_loss=0.09762, over 5707990.14 frames. ], batch size: 368, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:14:44,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=706668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:14:47,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=706671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:15:11,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=706700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:15:12,242 INFO [train.py:968] (0/2) Epoch 16, batch 22750, libri_loss[loss=0.3384, simple_loss=0.4038, pruned_loss=0.1366, over 29285.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3491, pruned_loss=0.09895, over 5721232.01 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3547, pruned_loss=0.1028, over 5758994.46 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3473, pruned_loss=0.09725, over 5714645.06 frames. ], batch size: 94, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:15:32,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=706728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:15:34,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.759e+02 1.154e+03 1.359e+03 1.809e+03 3.090e+03, threshold=2.718e+03, percent-clipped=2.0
+2023-03-08 09:15:39,713 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=706735.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:15:53,502 INFO [train.py:968] (0/2) Epoch 16, batch 22800, giga_loss[loss=0.2362, simple_loss=0.3162, pruned_loss=0.07808, over 29021.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3482, pruned_loss=0.09904, over 5728195.42 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3545, pruned_loss=0.1029, over 5759796.64 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3467, pruned_loss=0.0975, over 5720537.08 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:16:34,065 INFO [train.py:968] (0/2) Epoch 16, batch 22850, giga_loss[loss=0.2499, simple_loss=0.3241, pruned_loss=0.08786, over 28902.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3481, pruned_loss=0.1006, over 5725731.28 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3549, pruned_loss=0.1034, over 5760378.41 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3464, pruned_loss=0.09886, over 5718099.53 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:16:40,152 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-08 09:16:59,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.602e+02 1.155e+03 1.386e+03 1.816e+03 6.886e+03, threshold=2.773e+03, percent-clipped=7.0
+2023-03-08 09:17:02,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2650, 1.0523, 4.6448, 3.4967], device='cuda:0'), covar=tensor([0.1805, 0.3026, 0.0348, 0.0900], device='cuda:0'), in_proj_covar=tensor([0.0707, 0.0617, 0.0903, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 09:17:16,167 INFO [train.py:968] (0/2) Epoch 16, batch 22900, giga_loss[loss=0.2556, simple_loss=0.3248, pruned_loss=0.09319, over 28940.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3463, pruned_loss=0.1009, over 5721779.76 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3557, pruned_loss=0.1041, over 5759545.99 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.344, pruned_loss=0.09879, over 5715984.54 frames. ], batch size: 186, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:17:16,589 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-08 09:17:32,338 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=706871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:17:35,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=706874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:17:41,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=706882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:17:45,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-08 09:17:49,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3811, 1.6178, 1.6335, 1.2244], device='cuda:0'), covar=tensor([0.1608, 0.2361, 0.1417, 0.1595], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0690, 0.0909, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 09:17:58,265 INFO [train.py:968] (0/2) Epoch 16, batch 22950, giga_loss[loss=0.3399, simple_loss=0.3875, pruned_loss=0.1462, over 26792.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3442, pruned_loss=0.1004, over 5723982.47 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3561, pruned_loss=0.1044, over 5761123.06 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3418, pruned_loss=0.09836, over 5717298.23 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:17:59,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=706903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:18:19,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9527, 3.7779, 3.5561, 1.8137], device='cuda:0'), covar=tensor([0.0702, 0.0885, 0.0818, 0.2067], device='cuda:0'), in_proj_covar=tensor([0.1129, 0.1046, 0.0901, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 09:18:20,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.554e+02 1.236e+03 1.434e+03 1.990e+03 5.267e+03, threshold=2.867e+03, percent-clipped=14.0
+2023-03-08 09:18:33,463 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6557, 1.7377, 1.4459, 1.8848], device='cuda:0'), covar=tensor([0.2573, 0.2782, 0.3083, 0.2590], device='cuda:0'), in_proj_covar=tensor([0.1398, 0.1021, 0.1238, 0.0967], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 09:18:37,307 INFO [train.py:968] (0/2) Epoch 16, batch 23000, giga_loss[loss=0.2645, simple_loss=0.3372, pruned_loss=0.09591, over 29053.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3435, pruned_loss=0.1005, over 5717342.48 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3561, pruned_loss=0.1046, over 5761823.19 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3413, pruned_loss=0.09862, over 5710377.76 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:18:51,288 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 09:18:55,343 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-08 09:19:06,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3685, 1.6132, 1.5971, 1.4660], device='cuda:0'), covar=tensor([0.1719, 0.1868, 0.2120, 0.1942], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0739, 0.0693, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 09:19:16,943 INFO [train.py:968] (0/2) Epoch 16, batch 23050, giga_loss[loss=0.2433, simple_loss=0.3121, pruned_loss=0.08731, over 28983.00 frames. ], tot_loss[loss=0.269, simple_loss=0.34, pruned_loss=0.09901, over 5716641.59 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3567, pruned_loss=0.1052, over 5754489.89 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3374, pruned_loss=0.09688, over 5717047.11 frames. ], batch size: 136, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:19:33,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2424, 4.0623, 3.8176, 1.8983], device='cuda:0'), covar=tensor([0.0555, 0.0675, 0.0722, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.1130, 0.1044, 0.0900, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 09:19:35,883 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=707025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:19:37,233 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-08 09:19:38,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=707028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:19:40,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.934e+02 1.207e+03 1.609e+03 2.211e+03 6.287e+03, threshold=3.218e+03, percent-clipped=11.0
+2023-03-08 09:19:58,335 INFO [train.py:968] (0/2) Epoch 16, batch 23100, libri_loss[loss=0.2872, simple_loss=0.363, pruned_loss=0.1057, over 29521.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3366, pruned_loss=0.09727, over 5710363.19 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3568, pruned_loss=0.1054, over 5749417.68 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3338, pruned_loss=0.09512, over 5713949.15 frames. ], batch size: 82, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:19:58,548 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:20:02,782 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=707057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:20:22,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1609, 2.3093, 2.4257, 1.9140], device='cuda:0'), covar=tensor([0.1708, 0.2028, 0.1293, 0.1485], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0689, 0.0908, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 09:20:35,416 INFO [train.py:968] (0/2) Epoch 16, batch 23150, giga_loss[loss=0.2483, simple_loss=0.3254, pruned_loss=0.08556, over 28555.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3333, pruned_loss=0.09549, over 5710735.54 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3569, pruned_loss=0.1055, over 5741889.58 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3305, pruned_loss=0.09343, over 5719081.19 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:20:41,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=707110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:20:59,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.682e+02 1.279e+03 1.707e+03 2.277e+03 6.229e+03, threshold=3.413e+03, percent-clipped=11.0
+2023-03-08 09:21:16,634 INFO [train.py:968] (0/2) Epoch 16, batch 23200, libri_loss[loss=0.2804, simple_loss=0.3442, pruned_loss=0.1083, over 29569.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3368, pruned_loss=0.09709, over 5708698.32 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3569, pruned_loss=0.1058, over 5746310.09 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3337, pruned_loss=0.09476, over 5710266.98 frames. ], batch size: 77, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:21:59,158 INFO [train.py:968] (0/2) Epoch 16, batch 23250, libri_loss[loss=0.3466, simple_loss=0.4079, pruned_loss=0.1426, over 29266.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3411, pruned_loss=0.0993, over 5708040.56 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3573, pruned_loss=0.1061, over 5747624.07 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3381, pruned_loss=0.09714, over 5707576.97 frames. ], batch size: 94, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:22:24,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.582e+02 1.278e+03 1.684e+03 2.231e+03 5.416e+03, threshold=3.368e+03, percent-clipped=8.0
+2023-03-08 09:22:40,857 INFO [train.py:968] (0/2) Epoch 16, batch 23300, giga_loss[loss=0.2533, simple_loss=0.3278, pruned_loss=0.08943, over 28699.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3439, pruned_loss=0.0998, over 5710531.92 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3576, pruned_loss=0.1063, over 5745390.82 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3411, pruned_loss=0.0978, over 5711697.53 frames. ], batch size: 85, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:22:43,174 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=707253.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 09:22:45,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=707256.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:23:08,000 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=707285.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 09:23:15,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9859, 3.7869, 3.5580, 1.8305], device='cuda:0'), covar=tensor([0.0581, 0.0741, 0.0689, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.1128, 0.1042, 0.0896, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 09:23:19,652 INFO [train.py:968] (0/2) Epoch 16, batch 23350, giga_loss[loss=0.3394, simple_loss=0.402, pruned_loss=0.1384, over 28702.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3469, pruned_loss=0.1009, over 5706846.81 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3577, pruned_loss=0.1065, over 5739950.73 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.344, pruned_loss=0.09891, over 5711936.26 frames. ], batch size: 242, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:23:47,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.193e+02 1.185e+03 1.397e+03 1.998e+03 8.064e+03, threshold=2.793e+03, percent-clipped=6.0
+2023-03-08 09:23:55,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9568, 1.8316, 1.3907, 1.5719], device='cuda:0'), covar=tensor([0.0824, 0.0781, 0.1052, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0441, 0.0507, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 09:24:03,612 INFO [train.py:968] (0/2) Epoch 16, batch 23400, giga_loss[loss=0.3079, simple_loss=0.3644, pruned_loss=0.1257, over 28800.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3478, pruned_loss=0.101, over 5715740.81 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3578, pruned_loss=0.1068, over 5743703.61 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3451, pruned_loss=0.09892, over 5715740.25 frames. ], batch size: 99, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:24:07,070 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:24:51,549 INFO [train.py:968] (0/2) Epoch 16, batch 23450, giga_loss[loss=0.2629, simple_loss=0.3449, pruned_loss=0.0904, over 29034.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3517, pruned_loss=0.1042, over 5697530.38 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.358, pruned_loss=0.107, over 5732109.79 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3492, pruned_loss=0.1023, over 5707806.93 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:25:15,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=707426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:25:21,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.758e+02 1.579e+03 2.098e+03 2.912e+03 6.058e+03, threshold=4.197e+03, percent-clipped=29.0
+2023-03-08 09:25:37,259 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:25:39,359 INFO [train.py:968] (0/2) Epoch 16, batch 23500, giga_loss[loss=0.3001, simple_loss=0.3732, pruned_loss=0.1135, over 28931.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3575, pruned_loss=0.1089, over 5697947.19 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3582, pruned_loss=0.1072, over 5735425.01 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3554, pruned_loss=0.1073, over 5702519.41 frames. ], batch size: 199, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:26:14,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1998, 1.3243, 1.1647, 1.0849], device='cuda:0'), covar=tensor([0.0723, 0.0391, 0.0872, 0.0850], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0441, 0.0506, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 09:26:31,964 INFO [train.py:968] (0/2) Epoch 16, batch 23550, giga_loss[loss=0.3236, simple_loss=0.3892, pruned_loss=0.129, over 28877.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3648, pruned_loss=0.1144, over 5683248.35 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3587, pruned_loss=0.1077, over 5733557.74 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3627, pruned_loss=0.1126, over 5687859.15 frames. ], batch size: 145, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:27:02,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.781e+02 1.794e+03 2.513e+03 3.834e+03 8.713e+03, threshold=5.025e+03, percent-clipped=17.0
+2023-03-08 09:27:06,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2988, 1.3970, 1.0768, 1.0930], device='cuda:0'), covar=tensor([0.0605, 0.0358, 0.0730, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0442, 0.0508, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 09:27:20,632 INFO [train.py:968] (0/2) Epoch 16, batch 23600, giga_loss[loss=0.3397, simple_loss=0.3937, pruned_loss=0.1428, over 28960.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3709, pruned_loss=0.1195, over 5671310.74 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3589, pruned_loss=0.1079, over 5728302.26 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3692, pruned_loss=0.1181, over 5678615.70 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:27:37,516 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:27:39,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=707569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:27:43,934 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=707572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:27:57,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6166, 1.6671, 1.8037, 1.3843], device='cuda:0'), covar=tensor([0.1308, 0.1981, 0.1109, 0.1401], device='cuda:0'), in_proj_covar=tensor([0.0858, 0.0685, 0.0903, 0.0805], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 09:28:10,666 INFO [train.py:968] (0/2) Epoch 16, batch 23650, giga_loss[loss=0.3172, simple_loss=0.3845, pruned_loss=0.1249, over 28905.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.378, pruned_loss=0.1258, over 5677313.92 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3585, pruned_loss=0.1077, over 5730907.27 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3774, pruned_loss=0.1253, over 5679741.94 frames. ], batch size: 199, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:28:10,873 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=707601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:28:37,280 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6982, 2.7376, 2.5089, 2.4180], device='cuda:0'), covar=tensor([0.1478, 0.1776, 0.1656, 0.1603], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0742, 0.0696, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 09:28:40,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.734e+03 2.135e+03 3.006e+03 6.723e+03, threshold=4.270e+03, percent-clipped=4.0
+2023-03-08 09:28:51,040 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707644.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 09:28:57,288 INFO [train.py:968] (0/2) Epoch 16, batch 23700, giga_loss[loss=0.3569, simple_loss=0.4056, pruned_loss=0.1541, over 27478.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3812, pruned_loss=0.1285, over 5678119.09 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3589, pruned_loss=0.1082, over 5729598.62 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3812, pruned_loss=0.1285, over 5679131.72 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:29:10,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4223, 1.7934, 1.3873, 1.4012], device='cuda:0'), covar=tensor([0.2435, 0.2332, 0.2770, 0.2084], device='cuda:0'), in_proj_covar=tensor([0.1398, 0.1024, 0.1240, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 09:29:20,160 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:29:45,249 INFO [train.py:968] (0/2) Epoch 16, batch 23750, giga_loss[loss=0.2705, simple_loss=0.3488, pruned_loss=0.09607, over 28979.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3841, pruned_loss=0.1322, over 5671675.62 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3587, pruned_loss=0.1083, over 5733143.79 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3849, pruned_loss=0.1326, over 5667972.28 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:30:10,675 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=707730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:30:14,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.952e+02 1.651e+03 2.350e+03 3.281e+03 8.422e+03, threshold=4.701e+03, percent-clipped=14.0
+2023-03-08 09:30:30,369 INFO [train.py:968] (0/2) Epoch 16, batch 23800, giga_loss[loss=0.4636, simple_loss=0.4688, pruned_loss=0.2292, over 27481.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3884, pruned_loss=0.1371, over 5655213.90 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.359, pruned_loss=0.1086, over 5728641.02 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3896, pruned_loss=0.138, over 5653513.03 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:31:06,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5056, 1.6584, 1.5422, 1.4137], device='cuda:0'), covar=tensor([0.2011, 0.1889, 0.1598, 0.1828], device='cuda:0'), in_proj_covar=tensor([0.1853, 0.1790, 0.1719, 0.1857], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 09:31:07,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3229, 1.6857, 1.3154, 1.3476], device='cuda:0'), covar=tensor([0.2190, 0.2161, 0.2435, 0.1863], device='cuda:0'), in_proj_covar=tensor([0.1399, 0.1024, 0.1239, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 09:31:24,352 INFO [train.py:968] (0/2) Epoch 16, batch 23850, giga_loss[loss=0.359, simple_loss=0.4033, pruned_loss=0.1573, over 28448.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3912, pruned_loss=0.1406, over 5649311.57 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3589, pruned_loss=0.1086, over 5730325.07 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3925, pruned_loss=0.1415, over 5645784.92 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:31:52,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=707824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:32:05,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.696e+03 2.307e+03 3.429e+03 1.169e+04, threshold=4.614e+03, percent-clipped=14.0
+2023-03-08 09:32:18,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7519, 1.0368, 2.8933, 2.6883], device='cuda:0'), covar=tensor([0.1664, 0.2464, 0.0552, 0.0992], device='cuda:0'), in_proj_covar=tensor([0.0708, 0.0616, 0.0908, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 09:32:25,223 INFO [train.py:968] (0/2) Epoch 16, batch 23900, giga_loss[loss=0.3596, simple_loss=0.4135, pruned_loss=0.1529, over 28775.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3945, pruned_loss=0.1427, over 5650139.93 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3586, pruned_loss=0.1085, over 5732095.28 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3962, pruned_loss=0.1439, over 5644822.00 frames. ], batch size: 145, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:32:51,723 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=707873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:32:58,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=707876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:33:22,281 INFO [train.py:968] (0/2) Epoch 16, batch 23950, giga_loss[loss=0.3193, simple_loss=0.3781, pruned_loss=0.1302, over 28491.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3946, pruned_loss=0.1431, over 5648253.34 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3589, pruned_loss=0.1087, over 5732766.60 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3962, pruned_loss=0.1444, over 5642356.66 frames. ], batch size: 60, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:33:27,663 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=707905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:33:53,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.631e+02 1.744e+03 2.294e+03 3.170e+03 7.699e+03, threshold=4.587e+03, percent-clipped=11.0
+2023-03-08 09:34:00,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=707942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:34:09,042 INFO [train.py:968] (0/2) Epoch 16, batch 24000, giga_loss[loss=0.3162, simple_loss=0.3782, pruned_loss=0.1271, over 28896.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3931, pruned_loss=0.1426, over 5642283.82 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3593, pruned_loss=0.109, over 5724666.08 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3951, pruned_loss=0.1443, over 5641772.42 frames. ], batch size: 199, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:34:09,046 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 09:34:17,808 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2099, simple_loss=0.3175, pruned_loss=0.05122, over 944034.00 frames. 
+2023-03-08 09:34:17,809 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 09:34:33,475 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=707967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:34:36,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=707970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:34:53,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9677, 1.2480, 1.2591, 1.0691], device='cuda:0'), covar=tensor([0.1783, 0.1446, 0.2365, 0.1721], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0746, 0.0698, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 09:35:02,975 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=707999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:35:03,704 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-708000.pt
+2023-03-08 09:35:05,629 INFO [train.py:968] (0/2) Epoch 16, batch 24050, giga_loss[loss=0.3146, simple_loss=0.3765, pruned_loss=0.1263, over 28847.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3927, pruned_loss=0.1422, over 5634755.11 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3591, pruned_loss=0.109, over 5718536.94 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3953, pruned_loss=0.1445, over 5638077.91 frames. ], batch size: 99, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:35:15,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 09:35:20,100 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.44 vs. limit=5.0
+2023-03-08 09:35:20,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=708019.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:35:35,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.707e+03 2.224e+03 3.025e+03 6.916e+03, threshold=4.447e+03, percent-clipped=5.0
+2023-03-08 09:35:56,476 INFO [train.py:968] (0/2) Epoch 16, batch 24100, giga_loss[loss=0.3303, simple_loss=0.3912, pruned_loss=0.1347, over 28298.00 frames. ], tot_loss[loss=0.338, simple_loss=0.393, pruned_loss=0.1415, over 5617808.15 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3596, pruned_loss=0.1095, over 5701370.06 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3952, pruned_loss=0.1433, over 5634216.71 frames. ], batch size: 368, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:35:59,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=708053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:36:25,457 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=708081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:36:29,624 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=708085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:36:31,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=708088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:36:45,610 INFO [train.py:968] (0/2) Epoch 16, batch 24150, giga_loss[loss=0.3294, simple_loss=0.3911, pruned_loss=0.1339, over 28504.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3936, pruned_loss=0.1415, over 5622188.66 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3599, pruned_loss=0.1097, over 5706956.02 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.396, pruned_loss=0.1436, over 5628208.44 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:37:04,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=708117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:37:22,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.662e+02 1.536e+03 1.953e+03 2.634e+03 4.777e+03, threshold=3.906e+03, percent-clipped=3.0
+2023-03-08 09:37:38,652 INFO [train.py:968] (0/2) Epoch 16, batch 24200, giga_loss[loss=0.2839, simple_loss=0.3605, pruned_loss=0.1037, over 28829.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3909, pruned_loss=0.1388, over 5617412.99 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3598, pruned_loss=0.1097, over 5698774.20 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3931, pruned_loss=0.1407, over 5629001.79 frames. ], batch size: 174, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:37:54,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=708162.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 09:37:57,323 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=708165.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 09:38:21,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3121, 3.6928, 1.5521, 1.4791], device='cuda:0'), covar=tensor([0.1016, 0.0351, 0.0940, 0.1440], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0537, 0.0364, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 09:38:25,385 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=708194.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:38:28,091 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=708196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:38:32,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=708199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:38:33,061 INFO [train.py:968] (0/2) Epoch 16, batch 24250, giga_loss[loss=0.3291, simple_loss=0.3906, pruned_loss=0.1339, over 27866.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.388, pruned_loss=0.1349, over 5610147.18 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3602, pruned_loss=0.11, over 5690910.31 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3897, pruned_loss=0.1364, over 5625992.49 frames. ], batch size: 412, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:39:02,849 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=708228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:39:08,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.609e+03 2.198e+03 3.715e+03 1.205e+04, threshold=4.396e+03, percent-clipped=19.0
+2023-03-08 09:39:27,064 INFO [train.py:968] (0/2) Epoch 16, batch 24300, giga_loss[loss=0.2872, simple_loss=0.36, pruned_loss=0.1072, over 28339.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3848, pruned_loss=0.1315, over 5621519.90 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3603, pruned_loss=0.1102, over 5684520.64 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3863, pruned_loss=0.1328, over 5638290.97 frames. ], batch size: 369, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:40:00,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5120, 4.3195, 4.0557, 1.9985], device='cuda:0'), covar=tensor([0.0709, 0.0853, 0.1047, 0.2047], device='cuda:0'), in_proj_covar=tensor([0.1156, 0.1072, 0.0922, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 09:40:16,256 INFO [train.py:968] (0/2) Epoch 16, batch 24350, giga_loss[loss=0.3038, simple_loss=0.3714, pruned_loss=0.1181, over 28618.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3817, pruned_loss=0.1285, over 5644166.00 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3603, pruned_loss=0.1103, over 5686022.32 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3832, pruned_loss=0.1297, over 5655370.15 frames. ], batch size: 242, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:40:21,876 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-08 09:40:48,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.366e+02 1.756e+03 2.462e+03 3.481e+03 7.264e+03, threshold=4.923e+03, percent-clipped=9.0
+2023-03-08 09:41:07,091 INFO [train.py:968] (0/2) Epoch 16, batch 24400, giga_loss[loss=0.3686, simple_loss=0.4094, pruned_loss=0.164, over 26676.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3793, pruned_loss=0.1275, over 5642946.33 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3599, pruned_loss=0.1101, over 5689739.01 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3812, pruned_loss=0.1289, over 5647957.27 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:41:24,975 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=708369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:41:55,437 INFO [train.py:968] (0/2) Epoch 16, batch 24450, giga_loss[loss=0.3375, simple_loss=0.3922, pruned_loss=0.1414, over 28717.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3786, pruned_loss=0.1265, over 5657787.10 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3599, pruned_loss=0.1102, over 5684187.19 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3804, pruned_loss=0.1277, over 5666630.44 frames. ], batch size: 307, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:42:35,728 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.377e+02 1.569e+03 1.941e+03 2.497e+03 5.157e+03, threshold=3.881e+03, percent-clipped=2.0
+2023-03-08 09:42:50,757 INFO [train.py:968] (0/2) Epoch 16, batch 24500, giga_loss[loss=0.3148, simple_loss=0.3809, pruned_loss=0.1243, over 28879.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.379, pruned_loss=0.1268, over 5657948.22 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3608, pruned_loss=0.1109, over 5684688.10 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3801, pruned_loss=0.1276, over 5663857.87 frames. ], batch size: 112, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:42:55,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=708456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:43:28,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3948, 1.9478, 1.4526, 1.5131], device='cuda:0'), covar=tensor([0.0755, 0.0289, 0.0314, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0213], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0098], device='cuda:0')
+2023-03-08 09:43:39,852 INFO [train.py:968] (0/2) Epoch 16, batch 24550, giga_loss[loss=0.3072, simple_loss=0.3871, pruned_loss=0.1136, over 28835.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3763, pruned_loss=0.1241, over 5666691.70 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3612, pruned_loss=0.1116, over 5694873.87 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3776, pruned_loss=0.1249, over 5661199.39 frames. ], batch size: 186, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:44:02,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4614, 1.6288, 1.2154, 1.2617], device='cuda:0'), covar=tensor([0.0912, 0.0571, 0.1097, 0.1068], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0446, 0.0511, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 09:44:06,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5137, 1.8250, 1.4443, 1.8102], device='cuda:0'), covar=tensor([0.2535, 0.2572, 0.2848, 0.2380], device='cuda:0'), in_proj_covar=tensor([0.1394, 0.1021, 0.1238, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 09:44:17,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.017e+02 1.435e+03 1.935e+03 2.743e+03 6.614e+03, threshold=3.871e+03, percent-clipped=9.0
+2023-03-08 09:44:30,767 INFO [train.py:968] (0/2) Epoch 16, batch 24600, giga_loss[loss=0.3517, simple_loss=0.4126, pruned_loss=0.1454, over 28846.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3763, pruned_loss=0.1212, over 5674838.87 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3615, pruned_loss=0.1119, over 5686535.90 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3775, pruned_loss=0.1218, over 5677869.69 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:45:26,661 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=708599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:45:27,660 INFO [train.py:968] (0/2) Epoch 16, batch 24650, giga_loss[loss=0.3229, simple_loss=0.3897, pruned_loss=0.1281, over 28552.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3773, pruned_loss=0.1215, over 5658942.34 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3616, pruned_loss=0.112, over 5691071.57 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3783, pruned_loss=0.122, over 5656656.46 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:45:29,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=708602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:45:58,049 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=708631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:46:01,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.631e+02 1.670e+03 2.268e+03 3.083e+03 8.895e+03, threshold=4.535e+03, percent-clipped=10.0
+2023-03-08 09:46:07,141 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=708639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:46:16,949 INFO [train.py:968] (0/2) Epoch 16, batch 24700, giga_loss[loss=0.3566, simple_loss=0.3919, pruned_loss=0.1607, over 23496.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3774, pruned_loss=0.1227, over 5660981.11 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3617, pruned_loss=0.1122, over 5697647.62 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3786, pruned_loss=0.1232, over 5652335.45 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:46:24,942 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3247, 3.0290, 1.4920, 1.4418], device='cuda:0'), covar=tensor([0.0973, 0.0397, 0.0869, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0535, 0.0363, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 09:46:30,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5985, 1.9449, 1.5599, 1.6285], device='cuda:0'), covar=tensor([0.2383, 0.2357, 0.2702, 0.2147], device='cuda:0'), in_proj_covar=tensor([0.1400, 0.1024, 0.1242, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 09:47:03,102 INFO [train.py:968] (0/2) Epoch 16, batch 24750, giga_loss[loss=0.2949, simple_loss=0.3645, pruned_loss=0.1127, over 28513.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3763, pruned_loss=0.1228, over 5647391.26 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3619, pruned_loss=0.1126, over 5693291.86 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3775, pruned_loss=0.1231, over 5643152.33 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:47:25,149 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-08 09:47:37,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.870e+03 2.545e+03 3.643e+03 1.470e+04, threshold=5.091e+03, percent-clipped=16.0
+2023-03-08 09:47:45,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=708744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:47:52,216 INFO [train.py:968] (0/2) Epoch 16, batch 24800, giga_loss[loss=0.3016, simple_loss=0.3669, pruned_loss=0.1182, over 28881.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3756, pruned_loss=0.1234, over 5650544.59 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3629, pruned_loss=0.1134, over 5688378.97 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3761, pruned_loss=0.1233, over 5650239.63 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:48:21,121 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-08 09:48:37,407 INFO [train.py:968] (0/2) Epoch 16, batch 24850, giga_loss[loss=0.2944, simple_loss=0.3593, pruned_loss=0.1148, over 28791.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3756, pruned_loss=0.1244, over 5661325.84 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3632, pruned_loss=0.1136, over 5691747.99 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3759, pruned_loss=0.1242, over 5657835.88 frames. ], batch size: 243, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:49:09,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.143e+02 1.573e+03 2.042e+03 3.013e+03 5.935e+03, threshold=4.084e+03, percent-clipped=2.0
+2023-03-08 09:49:14,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5235, 1.6638, 1.7614, 1.3288], device='cuda:0'), covar=tensor([0.1874, 0.2443, 0.1530, 0.1720], device='cuda:0'), in_proj_covar=tensor([0.0861, 0.0690, 0.0907, 0.0807], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 09:49:20,865 INFO [train.py:968] (0/2) Epoch 16, batch 24900, giga_loss[loss=0.2927, simple_loss=0.37, pruned_loss=0.1077, over 29007.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3748, pruned_loss=0.1228, over 5674971.49 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3634, pruned_loss=0.1138, over 5697352.62 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3752, pruned_loss=0.1228, over 5666519.12 frames. ], batch size: 136, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:49:56,093 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=708887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:49:59,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=708890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:50:10,192 INFO [train.py:968] (0/2) Epoch 16, batch 24950, giga_loss[loss=0.269, simple_loss=0.3442, pruned_loss=0.0969, over 29095.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.374, pruned_loss=0.1215, over 5662207.75 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3635, pruned_loss=0.114, over 5688176.98 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3742, pruned_loss=0.1214, over 5663067.57 frames. ], batch size: 113, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:50:27,349 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=708919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:50:32,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6965, 1.8054, 1.3459, 1.4167], device='cuda:0'), covar=tensor([0.0884, 0.0590, 0.1063, 0.1071], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0446, 0.0512, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 09:50:42,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.377e+02 1.435e+03 1.805e+03 2.311e+03 4.832e+03, threshold=3.609e+03, percent-clipped=4.0
+2023-03-08 09:50:57,280 INFO [train.py:968] (0/2) Epoch 16, batch 25000, giga_loss[loss=0.2798, simple_loss=0.3488, pruned_loss=0.1054, over 28632.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3738, pruned_loss=0.1212, over 5667297.21 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3634, pruned_loss=0.114, over 5693259.02 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3744, pruned_loss=0.1213, over 5662939.93 frames. ], batch size: 85, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:51:46,474 INFO [train.py:968] (0/2) Epoch 16, batch 25050, giga_loss[loss=0.2815, simple_loss=0.3541, pruned_loss=0.1044, over 28664.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.373, pruned_loss=0.1207, over 5667333.08 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3639, pruned_loss=0.1144, over 5684649.83 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3731, pruned_loss=0.1205, over 5671291.63 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:51:59,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=709014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:52:25,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.481e+02 1.508e+03 2.005e+03 2.723e+03 7.556e+03, threshold=4.009e+03, percent-clipped=10.0
+2023-03-08 09:52:26,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7070, 2.1752, 1.9284, 1.4692], device='cuda:0'), covar=tensor([0.2797, 0.2128, 0.2206, 0.2661], device='cuda:0'), in_proj_covar=tensor([0.1855, 0.1785, 0.1718, 0.1854], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 09:52:30,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=709041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:52:38,870 INFO [train.py:968] (0/2) Epoch 16, batch 25100, libri_loss[loss=0.2608, simple_loss=0.3284, pruned_loss=0.09659, over 29580.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.371, pruned_loss=0.12, over 5669450.94 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3643, pruned_loss=0.1147, over 5684667.65 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.371, pruned_loss=0.1197, over 5671923.58 frames. ], batch size: 75, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:53:22,452 INFO [train.py:968] (0/2) Epoch 16, batch 25150, giga_loss[loss=0.3115, simple_loss=0.3786, pruned_loss=0.1222, over 28536.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3696, pruned_loss=0.1196, over 5682468.35 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3639, pruned_loss=0.1144, over 5694396.22 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3703, pruned_loss=0.1199, over 5675316.57 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:53:54,973 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.625e+02 1.730e+03 2.127e+03 2.973e+03 7.515e+03, threshold=4.254e+03, percent-clipped=10.0
+2023-03-08 09:53:57,211 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-08 09:54:07,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6853, 2.6766, 2.6735, 2.5764], device='cuda:0'), covar=tensor([0.1455, 0.1851, 0.1414, 0.1499], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0745, 0.0698, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 09:54:11,440 INFO [train.py:968] (0/2) Epoch 16, batch 25200, giga_loss[loss=0.3022, simple_loss=0.3664, pruned_loss=0.119, over 28648.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3687, pruned_loss=0.1194, over 5680996.35 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3637, pruned_loss=0.1143, over 5683023.53 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3696, pruned_loss=0.1198, over 5684650.56 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:54:17,718 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=709157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:54:19,563 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=709160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:54:23,851 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-08 09:54:49,488 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=709189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:54:57,549 INFO [train.py:968] (0/2) Epoch 16, batch 25250, giga_loss[loss=0.3078, simple_loss=0.3637, pruned_loss=0.1259, over 28755.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3672, pruned_loss=0.119, over 5671082.07 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3633, pruned_loss=0.1142, over 5675159.03 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3685, pruned_loss=0.1196, over 5681012.07 frames. ], batch size: 85, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:55:16,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8429, 2.8380, 1.8759, 1.0387], device='cuda:0'), covar=tensor([0.6537, 0.2695, 0.3221, 0.5857], device='cuda:0'), in_proj_covar=tensor([0.1660, 0.1575, 0.1552, 0.1367], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 09:55:30,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.651e+03 2.023e+03 2.813e+03 5.849e+03, threshold=4.046e+03, percent-clipped=5.0
+2023-03-08 09:55:46,726 INFO [train.py:968] (0/2) Epoch 16, batch 25300, giga_loss[loss=0.3091, simple_loss=0.3726, pruned_loss=0.1228, over 28658.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3671, pruned_loss=0.1195, over 5678798.09 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3637, pruned_loss=0.1146, over 5679353.74 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3679, pruned_loss=0.1198, over 5683020.26 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:56:32,361 INFO [train.py:968] (0/2) Epoch 16, batch 25350, libri_loss[loss=0.2633, simple_loss=0.3298, pruned_loss=0.09844, over 29599.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3659, pruned_loss=0.1187, over 5681112.31 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3631, pruned_loss=0.1145, over 5687104.46 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3672, pruned_loss=0.1192, over 5677322.74 frames. ], batch size: 74, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:56:58,293 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 09:57:08,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.668e+03 2.487e+03 3.274e+03 9.621e+03, threshold=4.974e+03, percent-clipped=15.0
+2023-03-08 09:57:12,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2425, 1.5726, 1.4655, 1.3786], device='cuda:0'), covar=tensor([0.1657, 0.1540, 0.2229, 0.1798], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0747, 0.0700, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 09:57:19,988 INFO [train.py:968] (0/2) Epoch 16, batch 25400, giga_loss[loss=0.2929, simple_loss=0.3696, pruned_loss=0.108, over 29031.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3675, pruned_loss=0.119, over 5678742.25 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.364, pruned_loss=0.1153, over 5682810.24 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3678, pruned_loss=0.1189, over 5679781.49 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:58:01,115 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 09:58:01,208 INFO [train.py:968] (0/2) Epoch 16, batch 25450, giga_loss[loss=0.2799, simple_loss=0.3601, pruned_loss=0.09989, over 28970.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3681, pruned_loss=0.1191, over 5674522.94 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3643, pruned_loss=0.1158, over 5680936.88 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3681, pruned_loss=0.1186, over 5676723.24 frames. ], batch size: 164, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:58:16,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=709416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:58:37,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.247e+02 1.779e+03 2.522e+03 4.068e+03 2.216e+04, threshold=5.044e+03, percent-clipped=15.0
+2023-03-08 09:58:45,080 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=709445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:58:50,964 INFO [train.py:968] (0/2) Epoch 16, batch 25500, giga_loss[loss=0.312, simple_loss=0.3791, pruned_loss=0.1224, over 28239.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3683, pruned_loss=0.1185, over 5676430.44 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3643, pruned_loss=0.1157, over 5681907.74 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3684, pruned_loss=0.1182, over 5677395.85 frames. ], batch size: 368, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:58:51,423 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=709451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:59:19,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3333, 2.7506, 2.3718, 1.9089], device='cuda:0'), covar=tensor([0.2560, 0.1747, 0.1952, 0.2327], device='cuda:0'), in_proj_covar=tensor([0.1850, 0.1791, 0.1716, 0.1853], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 09:59:28,993 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2460, 1.4128, 1.3740, 1.2800], device='cuda:0'), covar=tensor([0.2210, 0.1877, 0.1718, 0.1897], device='cuda:0'), in_proj_covar=tensor([0.1850, 0.1791, 0.1716, 0.1853], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 09:59:37,948 INFO [train.py:968] (0/2) Epoch 16, batch 25550, giga_loss[loss=0.3286, simple_loss=0.3925, pruned_loss=0.1324, over 28856.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3698, pruned_loss=0.1199, over 5681102.81 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3645, pruned_loss=0.1158, over 5686059.26 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3699, pruned_loss=0.1197, over 5678286.98 frames. ], batch size: 199, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:00:15,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.646e+03 1.883e+03 2.693e+03 6.080e+03, threshold=3.765e+03, percent-clipped=2.0
+2023-03-08 10:00:26,669 INFO [train.py:968] (0/2) Epoch 16, batch 25600, giga_loss[loss=0.2948, simple_loss=0.3635, pruned_loss=0.1131, over 28649.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3719, pruned_loss=0.1221, over 5685926.95 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3641, pruned_loss=0.1157, over 5691138.42 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3723, pruned_loss=0.1222, over 5679201.51 frames. ], batch size: 307, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:00:35,355 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=709559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:00:38,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=709562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:00:43,978 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3976, 1.4558, 3.3292, 3.1773], device='cuda:0'), covar=tensor([0.1268, 0.2296, 0.0475, 0.1022], device='cuda:0'), in_proj_covar=tensor([0.0715, 0.0622, 0.0915, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 10:01:09,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=709591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:01:19,953 INFO [train.py:968] (0/2) Epoch 16, batch 25650, giga_loss[loss=0.3007, simple_loss=0.3674, pruned_loss=0.117, over 28870.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3734, pruned_loss=0.1248, over 5685286.48 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3642, pruned_loss=0.1157, over 5694400.81 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.1249, over 5676876.47 frames. ], batch size: 243, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:02:00,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.154e+03 1.831e+03 2.526e+03 3.428e+03 9.497e+03, threshold=5.051e+03, percent-clipped=21.0
+2023-03-08 10:02:16,565 INFO [train.py:968] (0/2) Epoch 16, batch 25700, giga_loss[loss=0.3145, simple_loss=0.3735, pruned_loss=0.1278, over 28931.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3746, pruned_loss=0.1267, over 5686648.54 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3643, pruned_loss=0.1159, over 5698207.50 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3749, pruned_loss=0.1268, over 5676355.37 frames. ], batch size: 213, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:02:57,754 INFO [train.py:968] (0/2) Epoch 16, batch 25750, giga_loss[loss=0.2867, simple_loss=0.3453, pruned_loss=0.114, over 28487.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3748, pruned_loss=0.1272, over 5683387.08 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1156, over 5692109.42 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3762, pruned_loss=0.1281, over 5679099.53 frames. ], batch size: 65, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:03:31,851 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.744e+03 2.388e+03 4.173e+03 1.645e+04, threshold=4.775e+03, percent-clipped=17.0
+2023-03-08 10:03:46,185 INFO [train.py:968] (0/2) Epoch 16, batch 25800, giga_loss[loss=0.3986, simple_loss=0.4323, pruned_loss=0.1824, over 27649.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3732, pruned_loss=0.1265, over 5667649.41 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3638, pruned_loss=0.1157, over 5688263.12 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3746, pruned_loss=0.1274, over 5667024.51 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:04:30,657 INFO [train.py:968] (0/2) Epoch 16, batch 25850, giga_loss[loss=0.2603, simple_loss=0.3447, pruned_loss=0.08791, over 29112.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3733, pruned_loss=0.1249, over 5673807.90 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3634, pruned_loss=0.1155, over 5689812.87 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3748, pruned_loss=0.126, over 5671904.47 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:04:31,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5352, 1.8459, 1.5084, 1.5655], device='cuda:0'), covar=tensor([0.2347, 0.2211, 0.2441, 0.2030], device='cuda:0'), in_proj_covar=tensor([0.1405, 0.1024, 0.1245, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 10:04:47,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=709820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:04:53,144 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=709826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:05:06,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.846e+02 1.606e+03 2.112e+03 3.053e+03 9.570e+03, threshold=4.224e+03, percent-clipped=11.0
+2023-03-08 10:05:20,353 INFO [train.py:968] (0/2) Epoch 16, batch 25900, giga_loss[loss=0.3072, simple_loss=0.3506, pruned_loss=0.1319, over 23818.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3718, pruned_loss=0.1233, over 5652921.46 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3638, pruned_loss=0.1157, over 5682023.96 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3728, pruned_loss=0.1241, over 5657594.71 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 1.0
+2023-03-08 10:05:46,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0455, 5.8967, 5.6027, 2.8282], device='cuda:0'), covar=tensor([0.0420, 0.0570, 0.0628, 0.1606], device='cuda:0'), in_proj_covar=tensor([0.1168, 0.1075, 0.0926, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 10:06:08,529 INFO [train.py:968] (0/2) Epoch 16, batch 25950, giga_loss[loss=0.3632, simple_loss=0.3917, pruned_loss=0.1673, over 26645.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3703, pruned_loss=0.1228, over 5650147.74 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.364, pruned_loss=0.1159, over 5675637.93 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3711, pruned_loss=0.1234, over 5658282.40 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 1.0
+2023-03-08 10:06:41,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.682e+02 1.571e+03 2.037e+03 2.835e+03 9.186e+03, threshold=4.075e+03, percent-clipped=6.0
+2023-03-08 10:06:52,394 INFO [train.py:968] (0/2) Epoch 16, batch 26000, giga_loss[loss=0.3238, simple_loss=0.3737, pruned_loss=0.137, over 27572.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3692, pruned_loss=0.1225, over 5652048.04 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3645, pruned_loss=0.1161, over 5671208.71 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3695, pruned_loss=0.123, over 5662855.53 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:07:00,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=709959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:05,287 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=709963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:08,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=709966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:12,946 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=709969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:16,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=709972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:39,625 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=709995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:43,809 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-710000.pt
+2023-03-08 10:07:44,835 INFO [train.py:968] (0/2) Epoch 16, batch 26050, giga_loss[loss=0.3281, simple_loss=0.3854, pruned_loss=0.1354, over 28952.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3712, pruned_loss=0.1247, over 5647362.26 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3647, pruned_loss=0.1163, over 5675905.32 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3715, pruned_loss=0.1251, over 5651170.52 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:07:45,034 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=710001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:08:07,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4321, 1.5374, 1.6236, 1.2566], device='cuda:0'), covar=tensor([0.1667, 0.2479, 0.1401, 0.1634], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0695, 0.0911, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 10:08:14,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5659, 4.3934, 4.1075, 1.9065], device='cuda:0'), covar=tensor([0.0544, 0.0701, 0.0850, 0.2109], device='cuda:0'), in_proj_covar=tensor([0.1167, 0.1077, 0.0927, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 10:08:20,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.639e+03 2.150e+03 3.302e+03 7.597e+03, threshold=4.301e+03, percent-clipped=18.0
+2023-03-08 10:08:28,781 INFO [train.py:968] (0/2) Epoch 16, batch 26100, giga_loss[loss=0.2707, simple_loss=0.3532, pruned_loss=0.09404, over 28517.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3732, pruned_loss=0.1253, over 5656886.02 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3645, pruned_loss=0.1163, over 5679346.68 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3738, pruned_loss=0.1259, over 5656198.74 frames. ], batch size: 71, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:08:42,199 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=710064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:08:58,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=710083.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:09:16,180 INFO [train.py:968] (0/2) Epoch 16, batch 26150, giga_loss[loss=0.3907, simple_loss=0.4436, pruned_loss=0.169, over 28044.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3773, pruned_loss=0.1254, over 5652084.29 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3645, pruned_loss=0.1164, over 5671801.84 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.378, pruned_loss=0.1261, over 5657959.24 frames. ], batch size: 412, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:09:54,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.727e+02 1.538e+03 2.075e+03 2.763e+03 1.749e+04, threshold=4.150e+03, percent-clipped=8.0
+2023-03-08 10:10:05,892 INFO [train.py:968] (0/2) Epoch 16, batch 26200, giga_loss[loss=0.2935, simple_loss=0.3735, pruned_loss=0.1068, over 29084.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3786, pruned_loss=0.1253, over 5641083.61 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3646, pruned_loss=0.1165, over 5661567.09 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3793, pruned_loss=0.1259, over 5654437.63 frames. ], batch size: 136, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:10:53,461 INFO [train.py:968] (0/2) Epoch 16, batch 26250, giga_loss[loss=0.3322, simple_loss=0.3922, pruned_loss=0.1361, over 28508.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3799, pruned_loss=0.1264, over 5639288.12 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3645, pruned_loss=0.1164, over 5663388.55 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.381, pruned_loss=0.1273, over 5647415.51 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:11:26,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4730, 1.6406, 1.2876, 1.1994], device='cuda:0'), covar=tensor([0.0930, 0.0532, 0.1029, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0446, 0.0510, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 10:11:27,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.943e+03 2.519e+03 3.250e+03 7.778e+03, threshold=5.038e+03, percent-clipped=12.0
+2023-03-08 10:11:28,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6381, 1.7376, 1.8662, 1.3939], device='cuda:0'), covar=tensor([0.1715, 0.2342, 0.1370, 0.1651], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0696, 0.0913, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 10:11:36,700 INFO [train.py:968] (0/2) Epoch 16, batch 26300, giga_loss[loss=0.286, simple_loss=0.3566, pruned_loss=0.1077, over 28924.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3795, pruned_loss=0.1265, over 5647868.15 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3645, pruned_loss=0.1165, over 5661971.82 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3809, pruned_loss=0.1274, over 5655322.75 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:11:48,699 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6042, 1.6112, 1.8990, 1.4382], device='cuda:0'), covar=tensor([0.1312, 0.1876, 0.1062, 0.1426], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0696, 0.0913, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 10:12:24,165 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3107, 1.5170, 1.4581, 1.3904], device='cuda:0'), covar=tensor([0.1146, 0.1114, 0.1664, 0.1177], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0738, 0.0691, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 10:12:25,752 INFO [train.py:968] (0/2) Epoch 16, batch 26350, giga_loss[loss=0.3188, simple_loss=0.3835, pruned_loss=0.127, over 28725.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3812, pruned_loss=0.1289, over 5642169.31 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3648, pruned_loss=0.1167, over 5661708.34 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3823, pruned_loss=0.1296, over 5647947.66 frames. ], batch size: 242, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:12:35,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-08 10:12:51,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1134, 1.0810, 3.5153, 3.1243], device='cuda:0'), covar=tensor([0.1622, 0.2769, 0.0449, 0.1561], device='cuda:0'), in_proj_covar=tensor([0.0713, 0.0619, 0.0911, 0.0832], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 10:13:01,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=710334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:13:06,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.324e+02 1.605e+03 2.032e+03 2.897e+03 6.146e+03, threshold=4.064e+03, percent-clipped=2.0
+2023-03-08 10:13:14,709 INFO [train.py:968] (0/2) Epoch 16, batch 26400, giga_loss[loss=0.2645, simple_loss=0.3406, pruned_loss=0.09422, over 28803.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3794, pruned_loss=0.1283, over 5641124.39 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3648, pruned_loss=0.1166, over 5665506.22 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3805, pruned_loss=0.1291, over 5641924.21 frames. ], batch size: 174, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:14:02,765 INFO [train.py:968] (0/2) Epoch 16, batch 26450, giga_loss[loss=0.2997, simple_loss=0.3665, pruned_loss=0.1165, over 29047.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3766, pruned_loss=0.1268, over 5653610.91 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3648, pruned_loss=0.1165, over 5669781.54 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3779, pruned_loss=0.1279, over 5650312.24 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:14:31,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5786, 3.4143, 3.2410, 2.1337], device='cuda:0'), covar=tensor([0.0684, 0.0841, 0.0788, 0.1696], device='cuda:0'), in_proj_covar=tensor([0.1163, 0.1077, 0.0926, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 10:14:39,902 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-08 10:14:43,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=710439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:14:44,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.144e+03 1.608e+03 2.306e+03 3.502e+03 9.444e+03, threshold=4.612e+03, percent-clipped=16.0
+2023-03-08 10:14:53,276 INFO [train.py:968] (0/2) Epoch 16, batch 26500, giga_loss[loss=0.3335, simple_loss=0.3759, pruned_loss=0.1455, over 23588.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3757, pruned_loss=0.1266, over 5641463.41 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.365, pruned_loss=0.1164, over 5668077.06 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.377, pruned_loss=0.1279, over 5640195.47 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:15:00,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=710458.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:15:19,790 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=710477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:15:22,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=710480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:15:33,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4140, 1.7890, 1.4335, 1.5002], device='cuda:0'), covar=tensor([0.2042, 0.1978, 0.2135, 0.1795], device='cuda:0'), in_proj_covar=tensor([0.1408, 0.1028, 0.1246, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 10:15:38,624 INFO [train.py:968] (0/2) Epoch 16, batch 26550, giga_loss[loss=0.2667, simple_loss=0.3362, pruned_loss=0.09859, over 28808.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.377, pruned_loss=0.128, over 5650452.71 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3653, pruned_loss=0.1168, over 5676246.98 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3782, pruned_loss=0.1292, over 5641019.70 frames. ], batch size: 99, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:15:44,139 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=710509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:16:14,840 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.814e+03 2.398e+03 3.314e+03 1.274e+04, threshold=4.795e+03, percent-clipped=7.0
+2023-03-08 10:16:23,371 INFO [train.py:968] (0/2) Epoch 16, batch 26600, giga_loss[loss=0.2773, simple_loss=0.3331, pruned_loss=0.1107, over 28645.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3759, pruned_loss=0.1276, over 5663188.30 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.365, pruned_loss=0.1166, over 5680655.52 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3773, pruned_loss=0.1288, over 5651517.58 frames. ], batch size: 92, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:16:34,987 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=710564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:16:52,284 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=710582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:16:54,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.26 vs. limit=5.0
+2023-03-08 10:16:55,405 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=710585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:17:10,867 INFO [train.py:968] (0/2) Epoch 16, batch 26650, giga_loss[loss=0.2611, simple_loss=0.3371, pruned_loss=0.09259, over 28953.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3742, pruned_loss=0.1264, over 5674267.34 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.365, pruned_loss=0.1166, over 5681919.38 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3753, pruned_loss=0.1274, over 5663924.11 frames. ], batch size: 164, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:17:12,583 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=710601.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:17:14,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=710604.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:17:16,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2256, 1.2951, 1.1228, 0.9508], device='cuda:0'), covar=tensor([0.0911, 0.0510, 0.1064, 0.1028], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0445, 0.0510, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 10:17:24,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=710614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:17:24,588 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-08 10:17:43,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6769, 1.8557, 1.7046, 1.4951], device='cuda:0'), covar=tensor([0.2489, 0.2162, 0.1803, 0.2250], device='cuda:0'), in_proj_covar=tensor([0.1850, 0.1791, 0.1714, 0.1855], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 10:17:45,695 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=710633.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:17:51,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.913e+02 1.722e+03 2.194e+03 3.051e+03 1.301e+04, threshold=4.388e+03, percent-clipped=6.0
+2023-03-08 10:17:58,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8532, 1.8761, 1.3935, 1.4914], device='cuda:0'), covar=tensor([0.0840, 0.0595, 0.1017, 0.1036], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0445, 0.0510, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 10:18:00,744 INFO [train.py:968] (0/2) Epoch 16, batch 26700, giga_loss[loss=0.28, simple_loss=0.3591, pruned_loss=0.1004, over 28963.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3742, pruned_loss=0.1264, over 5676951.26 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3657, pruned_loss=0.1172, over 5685493.84 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3747, pruned_loss=0.1268, over 5665640.79 frames. ], batch size: 213, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:18:20,962 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=710673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:18:27,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6873, 2.0826, 1.8367, 1.5475], device='cuda:0'), covar=tensor([0.2369, 0.1663, 0.1420, 0.1848], device='cuda:0'), in_proj_covar=tensor([0.1845, 0.1786, 0.1709, 0.1850], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 10:18:47,455 INFO [train.py:968] (0/2) Epoch 16, batch 26750, libri_loss[loss=0.2576, simple_loss=0.3265, pruned_loss=0.09439, over 28532.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.376, pruned_loss=0.1269, over 5677803.42 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3654, pruned_loss=0.1171, over 5691971.60 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.377, pruned_loss=0.1277, over 5662508.83 frames. ], batch size: 63, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:19:29,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.994e+02 1.667e+03 2.241e+03 2.993e+03 9.070e+03, threshold=4.481e+03, percent-clipped=10.0
+2023-03-08 10:19:39,902 INFO [train.py:968] (0/2) Epoch 16, batch 26800, giga_loss[loss=0.2484, simple_loss=0.3254, pruned_loss=0.08565, over 28301.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3755, pruned_loss=0.1268, over 5664889.08 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3655, pruned_loss=0.1171, over 5694551.56 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3764, pruned_loss=0.1275, over 5650274.95 frames. ], batch size: 65, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:20:12,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3585, 1.4891, 1.3875, 1.4873], device='cuda:0'), covar=tensor([0.0737, 0.0349, 0.0312, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 10:20:27,996 INFO [train.py:968] (0/2) Epoch 16, batch 26850, giga_loss[loss=0.324, simple_loss=0.3969, pruned_loss=0.1256, over 28994.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3761, pruned_loss=0.1267, over 5661765.18 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3658, pruned_loss=0.1175, over 5686712.72 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3766, pruned_loss=0.127, over 5657776.96 frames. ], batch size: 136, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:20:29,546 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 10:20:39,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8722, 1.9596, 2.1002, 1.7349], device='cuda:0'), covar=tensor([0.1530, 0.1769, 0.1626, 0.1682], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0744, 0.0697, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 10:21:04,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.274e+02 1.537e+03 2.010e+03 2.864e+03 6.170e+03, threshold=4.020e+03, percent-clipped=3.0
+2023-03-08 10:21:13,653 INFO [train.py:968] (0/2) Epoch 16, batch 26900, giga_loss[loss=0.2915, simple_loss=0.3758, pruned_loss=0.1036, over 28442.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.377, pruned_loss=0.1242, over 5674573.95 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3658, pruned_loss=0.1174, over 5690456.68 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3775, pruned_loss=0.1247, over 5667767.72 frames. ], batch size: 85, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:21:27,777 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=710864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:21:50,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5166, 1.7709, 1.7523, 1.2763], device='cuda:0'), covar=tensor([0.2077, 0.2582, 0.1743, 0.1972], device='cuda:0'), in_proj_covar=tensor([0.0864, 0.0695, 0.0912, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 10:22:05,934 INFO [train.py:968] (0/2) Epoch 16, batch 26950, giga_loss[loss=0.2945, simple_loss=0.366, pruned_loss=0.1115, over 28865.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3779, pruned_loss=0.123, over 5678468.26 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3655, pruned_loss=0.1173, over 5694708.53 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3789, pruned_loss=0.1236, over 5669019.94 frames. ], batch size: 119, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:22:34,227 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-08 10:22:38,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=710939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:22:39,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.427e+02 1.462e+03 1.959e+03 2.856e+03 6.006e+03, threshold=3.918e+03, percent-clipped=6.0
+2023-03-08 10:22:47,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3267, 3.2162, 1.5128, 1.5305], device='cuda:0'), covar=tensor([0.0945, 0.0373, 0.0896, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0536, 0.0363, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 10:22:49,552 INFO [train.py:968] (0/2) Epoch 16, batch 27000, giga_loss[loss=0.302, simple_loss=0.369, pruned_loss=0.1175, over 28947.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3798, pruned_loss=0.1244, over 5682891.40 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3656, pruned_loss=0.1174, over 5699460.33 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3808, pruned_loss=0.1249, over 5670835.55 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:22:49,556 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 10:22:59,222 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2093, simple_loss=0.3155, pruned_loss=0.05157, over 944034.00 frames. 
+2023-03-08 10:22:59,222 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 10:23:47,474 INFO [train.py:968] (0/2) Epoch 16, batch 27050, libri_loss[loss=0.262, simple_loss=0.3207, pruned_loss=0.1016, over 29484.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3823, pruned_loss=0.1277, over 5681863.07 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3649, pruned_loss=0.1169, over 5703121.56 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3843, pruned_loss=0.1289, over 5667904.51 frames. ], batch size: 70, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:23:57,306 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=711010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:24:35,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.172e+03 1.819e+03 2.638e+03 4.102e+03 1.110e+04, threshold=5.276e+03, percent-clipped=28.0
+2023-03-08 10:24:40,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=711048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:24:41,990 INFO [train.py:968] (0/2) Epoch 16, batch 27100, giga_loss[loss=0.4378, simple_loss=0.4501, pruned_loss=0.2127, over 26377.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.383, pruned_loss=0.1291, over 5685182.54 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3647, pruned_loss=0.1169, over 5702556.29 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.385, pruned_loss=0.1302, over 5674503.09 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:24:56,572 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-08 10:25:13,642 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=711082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:25:16,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=711085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:25:31,531 INFO [train.py:968] (0/2) Epoch 16, batch 27150, giga_loss[loss=0.3239, simple_loss=0.3919, pruned_loss=0.128, over 28864.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3827, pruned_loss=0.1294, over 5681902.85 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3647, pruned_loss=0.1169, over 5707025.33 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3848, pruned_loss=0.1307, over 5668750.90 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:25:38,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8132, 5.1736, 2.1413, 2.2749], device='cuda:0'), covar=tensor([0.0968, 0.0243, 0.0834, 0.1197], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0534, 0.0362, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 10:25:47,077 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=711114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:26:14,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.709e+03 2.328e+03 3.011e+03 8.105e+03, threshold=4.655e+03, percent-clipped=3.0
+2023-03-08 10:26:23,685 INFO [train.py:968] (0/2) Epoch 16, batch 27200, giga_loss[loss=0.2632, simple_loss=0.3501, pruned_loss=0.08817, over 28525.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3805, pruned_loss=0.1266, over 5682090.43 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3647, pruned_loss=0.1168, over 5707895.20 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3823, pruned_loss=0.1277, over 5671023.98 frames. ], batch size: 85, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:27:04,173 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=711191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:27:06,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=711194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:27:13,828 INFO [train.py:968] (0/2) Epoch 16, batch 27250, giga_loss[loss=0.2955, simple_loss=0.3754, pruned_loss=0.1078, over 28961.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3819, pruned_loss=0.1263, over 5666384.13 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3649, pruned_loss=0.117, over 5706267.40 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3833, pruned_loss=0.1271, over 5658789.01 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:27:21,166 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-08 10:27:22,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5697, 1.5758, 1.8244, 1.3922], device='cuda:0'), covar=tensor([0.1571, 0.2201, 0.1275, 0.1538], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0698, 0.0915, 0.0814], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 10:27:36,518 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=711223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:27:53,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=711239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:27:55,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.463e+03 1.956e+03 2.425e+03 6.038e+03, threshold=3.912e+03, percent-clipped=1.0
+2023-03-08 10:28:01,782 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-08 10:28:04,032 INFO [train.py:968] (0/2) Epoch 16, batch 27300, giga_loss[loss=0.3503, simple_loss=0.4095, pruned_loss=0.1456, over 28704.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3831, pruned_loss=0.1268, over 5672723.74 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3652, pruned_loss=0.1172, over 5706692.53 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3842, pruned_loss=0.1273, over 5665533.82 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:28:28,879 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-08 10:28:59,763 INFO [train.py:968] (0/2) Epoch 16, batch 27350, giga_loss[loss=0.3247, simple_loss=0.3845, pruned_loss=0.1325, over 28772.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3832, pruned_loss=0.1278, over 5657270.90 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3652, pruned_loss=0.1173, over 5707728.06 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3841, pruned_loss=0.1282, over 5650480.21 frames. ], batch size: 284, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:29:27,969 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-08 10:29:36,379 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.705e+03 2.138e+03 2.979e+03 7.336e+03, threshold=4.276e+03, percent-clipped=11.0
+2023-03-08 10:29:46,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4211, 1.8590, 1.5324, 1.4662], device='cuda:0'), covar=tensor([0.0639, 0.0258, 0.0261, 0.0768], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 10:29:47,011 INFO [train.py:968] (0/2) Epoch 16, batch 27400, giga_loss[loss=0.3381, simple_loss=0.3725, pruned_loss=0.1519, over 23309.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3829, pruned_loss=0.1284, over 5647992.33 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3652, pruned_loss=0.1173, over 5699215.86 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3839, pruned_loss=0.1289, over 5649461.42 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:30:12,316 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=711373.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:30:21,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=711382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:30:24,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=711385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:30:24,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=711385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:30:32,398 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-08 10:30:37,176 INFO [train.py:968] (0/2) Epoch 16, batch 27450, giga_loss[loss=0.3388, simple_loss=0.4034, pruned_loss=0.1371, over 28978.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3808, pruned_loss=0.1276, over 5670026.32 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.365, pruned_loss=0.1171, over 5702712.48 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3822, pruned_loss=0.1284, over 5667054.63 frames. ], batch size: 213, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:30:46,865 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-08 10:30:53,195 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=711414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:31:20,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5501, 1.7465, 1.8072, 1.3431], device='cuda:0'), covar=tensor([0.1567, 0.2221, 0.1296, 0.1517], device='cuda:0'), in_proj_covar=tensor([0.0867, 0.0698, 0.0916, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 10:31:25,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.140e+03 1.537e+03 2.202e+03 2.970e+03 5.555e+03, threshold=4.404e+03, percent-clipped=7.0
+2023-03-08 10:31:36,617 INFO [train.py:968] (0/2) Epoch 16, batch 27500, giga_loss[loss=0.3653, simple_loss=0.3885, pruned_loss=0.171, over 23446.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3782, pruned_loss=0.1264, over 5662657.54 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3651, pruned_loss=0.1172, over 5703569.92 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3792, pruned_loss=0.1271, over 5659359.12 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:31:39,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0435, 5.8736, 5.5358, 3.0127], device='cuda:0'), covar=tensor([0.0398, 0.0529, 0.0623, 0.1564], device='cuda:0'), in_proj_covar=tensor([0.1171, 0.1088, 0.0935, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 10:32:23,235 INFO [train.py:968] (0/2) Epoch 16, batch 27550, libri_loss[loss=0.3259, simple_loss=0.3926, pruned_loss=0.1296, over 29555.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3763, pruned_loss=0.1255, over 5659712.29 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3657, pruned_loss=0.1173, over 5696830.98 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.377, pruned_loss=0.1262, over 5661627.61 frames. ], batch size: 84, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:32:47,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-08 10:32:49,817 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=711528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:32:54,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=711531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:33:03,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.786e+03 2.177e+03 3.180e+03 8.174e+03, threshold=4.353e+03, percent-clipped=14.0
+2023-03-08 10:33:11,386 INFO [train.py:968] (0/2) Epoch 16, batch 27600, giga_loss[loss=0.3148, simple_loss=0.3801, pruned_loss=0.1247, over 28708.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3767, pruned_loss=0.127, over 5656361.20 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3653, pruned_loss=0.117, over 5692000.03 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3779, pruned_loss=0.128, over 5660869.13 frames. ], batch size: 284, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:33:18,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=711560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:33:45,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=711591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:33:54,068 INFO [train.py:968] (0/2) Epoch 16, batch 27650, giga_loss[loss=0.2985, simple_loss=0.3649, pruned_loss=0.116, over 28737.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3758, pruned_loss=0.1259, over 5664977.31 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3654, pruned_loss=0.1168, over 5698826.78 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3771, pruned_loss=0.1272, over 5661281.50 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:34:13,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2121, 1.3673, 1.4009, 1.1633], device='cuda:0'), covar=tensor([0.2551, 0.2298, 0.1551, 0.2110], device='cuda:0'), in_proj_covar=tensor([0.1855, 0.1793, 0.1722, 0.1858], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 10:34:32,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.226e+02 1.478e+03 1.942e+03 2.777e+03 4.933e+03, threshold=3.884e+03, percent-clipped=1.0
+2023-03-08 10:34:41,757 INFO [train.py:968] (0/2) Epoch 16, batch 27700, giga_loss[loss=0.2957, simple_loss=0.3603, pruned_loss=0.1155, over 28746.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3723, pruned_loss=0.1221, over 5660068.36 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3655, pruned_loss=0.117, over 5692603.00 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3733, pruned_loss=0.1231, over 5661627.14 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:35:06,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6403, 1.7209, 1.3208, 1.3669], device='cuda:0'), covar=tensor([0.0846, 0.0579, 0.0995, 0.1086], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0445, 0.0508, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 10:35:33,437 INFO [train.py:968] (0/2) Epoch 16, batch 27750, giga_loss[loss=0.3505, simple_loss=0.3937, pruned_loss=0.1537, over 26583.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3707, pruned_loss=0.1209, over 5658670.56 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3658, pruned_loss=0.1173, over 5695685.72 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3714, pruned_loss=0.1216, over 5656263.55 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:36:16,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7855, 1.9114, 1.3862, 1.5430], device='cuda:0'), covar=tensor([0.0865, 0.0634, 0.1052, 0.1073], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0445, 0.0508, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 10:36:17,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.499e+03 2.091e+03 3.364e+03 7.830e+03, threshold=4.182e+03, percent-clipped=17.0
+2023-03-08 10:36:22,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=711748.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:36:26,013 INFO [train.py:968] (0/2) Epoch 16, batch 27800, libri_loss[loss=0.2674, simple_loss=0.3437, pruned_loss=0.09556, over 29506.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3695, pruned_loss=0.1204, over 5655876.46 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3657, pruned_loss=0.1171, over 5697311.40 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3701, pruned_loss=0.1212, over 5651496.49 frames. ], batch size: 84, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:36:36,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5134, 1.6808, 1.6945, 1.3642], device='cuda:0'), covar=tensor([0.2430, 0.2196, 0.1613, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.1858, 0.1792, 0.1723, 0.1858], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 10:36:45,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3097, 1.8197, 1.2095, 0.6802], device='cuda:0'), covar=tensor([0.3928, 0.2036, 0.2598, 0.4225], device='cuda:0'), in_proj_covar=tensor([0.1656, 0.1578, 0.1548, 0.1358], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 10:37:21,313 INFO [train.py:968] (0/2) Epoch 16, batch 27850, giga_loss[loss=0.2601, simple_loss=0.334, pruned_loss=0.09308, over 28567.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3658, pruned_loss=0.119, over 5652335.20 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3661, pruned_loss=0.1173, over 5696236.69 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3659, pruned_loss=0.1194, over 5649407.47 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:38:06,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.794e+02 1.827e+03 2.401e+03 3.213e+03 8.729e+03, threshold=4.801e+03, percent-clipped=19.0
+2023-03-08 10:38:11,961 INFO [train.py:968] (0/2) Epoch 16, batch 27900, giga_loss[loss=0.353, simple_loss=0.3995, pruned_loss=0.1532, over 27509.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3675, pruned_loss=0.1198, over 5645607.84 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3671, pruned_loss=0.1179, over 5688720.90 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3667, pruned_loss=0.1196, over 5648771.20 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:38:54,569 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=711891.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:38:58,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=711894.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:39:05,175 INFO [train.py:968] (0/2) Epoch 16, batch 27950, giga_loss[loss=0.2702, simple_loss=0.3464, pruned_loss=0.09696, over 28786.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3697, pruned_loss=0.1208, over 5637911.66 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3669, pruned_loss=0.118, over 5689939.30 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3691, pruned_loss=0.1206, over 5639213.99 frames. ], batch size: 119, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:39:07,420 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=711902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:39:17,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3601, 3.1779, 1.4964, 1.4719], device='cuda:0'), covar=tensor([0.0937, 0.0271, 0.0856, 0.1329], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0535, 0.0364, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 10:39:19,459 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-08 10:39:25,741 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=711923.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:39:47,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.854e+02 1.392e+03 1.859e+03 2.736e+03 5.812e+03, threshold=3.718e+03, percent-clipped=5.0
+2023-03-08 10:39:52,425 INFO [train.py:968] (0/2) Epoch 16, batch 28000, giga_loss[loss=0.2664, simple_loss=0.3432, pruned_loss=0.09477, over 29070.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3703, pruned_loss=0.1207, over 5645324.81 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3673, pruned_loss=0.1182, over 5688037.44 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3696, pruned_loss=0.1204, over 5646803.49 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:40:07,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=711966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:40:08,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-08 10:40:41,447 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-712000.pt
+2023-03-08 10:40:43,238 INFO [train.py:968] (0/2) Epoch 16, batch 28050, giga_loss[loss=0.3148, simple_loss=0.3803, pruned_loss=0.1247, over 28764.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3698, pruned_loss=0.121, over 5648967.00 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3676, pruned_loss=0.1185, over 5693747.80 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3691, pruned_loss=0.1206, over 5644117.57 frames. ], batch size: 284, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:40:52,229 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=712011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:41:20,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.912e+02 1.537e+03 1.899e+03 2.848e+03 7.069e+03, threshold=3.799e+03, percent-clipped=7.0
+2023-03-08 10:41:26,948 INFO [train.py:968] (0/2) Epoch 16, batch 28100, giga_loss[loss=0.3446, simple_loss=0.39, pruned_loss=0.1496, over 27477.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3691, pruned_loss=0.1209, over 5654993.03 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.367, pruned_loss=0.1181, over 5699365.08 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3691, pruned_loss=0.121, over 5644610.30 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:42:17,637 INFO [train.py:968] (0/2) Epoch 16, batch 28150, giga_loss[loss=0.3003, simple_loss=0.3695, pruned_loss=0.1156, over 28739.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3719, pruned_loss=0.1225, over 5653997.63 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3673, pruned_loss=0.1183, over 5698035.24 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3717, pruned_loss=0.1224, over 5646702.04 frames. ], batch size: 284, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:42:27,068 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=712109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:42:29,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=712112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:42:36,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1572, 2.6625, 1.1582, 1.4536], device='cuda:0'), covar=tensor([0.1034, 0.0371, 0.0923, 0.1320], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0536, 0.0364, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 10:42:43,919 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=712129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:42:57,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=712141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:42:58,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.801e+03 2.569e+03 3.620e+03 1.163e+04, threshold=5.138e+03, percent-clipped=23.0
+2023-03-08 10:43:05,652 INFO [train.py:968] (0/2) Epoch 16, batch 28200, giga_loss[loss=0.389, simple_loss=0.4084, pruned_loss=0.1848, over 23652.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3735, pruned_loss=0.1241, over 5642128.18 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3673, pruned_loss=0.1184, over 5691105.71 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3735, pruned_loss=0.1241, over 5641915.51 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:43:07,139 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-08 10:43:46,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5227, 1.6582, 1.5755, 1.3087], device='cuda:0'), covar=tensor([0.2551, 0.2143, 0.1899, 0.2346], device='cuda:0'), in_proj_covar=tensor([0.1859, 0.1798, 0.1724, 0.1860], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 10:43:58,259 INFO [train.py:968] (0/2) Epoch 16, batch 28250, giga_loss[loss=0.3595, simple_loss=0.4101, pruned_loss=0.1545, over 27836.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3761, pruned_loss=0.1265, over 5637885.32 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3677, pruned_loss=0.1187, over 5685523.97 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3759, pruned_loss=0.1263, over 5641073.53 frames. ], batch size: 412, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:44:35,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.639e+03 2.141e+03 2.741e+03 6.060e+03, threshold=4.281e+03, percent-clipped=2.0
+2023-03-08 10:44:37,400 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 10:44:43,810 INFO [train.py:968] (0/2) Epoch 16, batch 28300, giga_loss[loss=0.3017, simple_loss=0.3775, pruned_loss=0.1129, over 29040.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3759, pruned_loss=0.1268, over 5652711.42 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3671, pruned_loss=0.1186, over 5695404.38 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3768, pruned_loss=0.1272, over 5644233.19 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:45:13,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:45:37,099 INFO [train.py:968] (0/2) Epoch 16, batch 28350, giga_loss[loss=0.3003, simple_loss=0.3719, pruned_loss=0.1143, over 28891.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3757, pruned_loss=0.1247, over 5655130.98 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3674, pruned_loss=0.1188, over 5698283.30 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3764, pruned_loss=0.1251, over 5644588.74 frames. ], batch size: 186, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:46:22,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.420e+02 1.703e+03 2.197e+03 2.988e+03 1.106e+04, threshold=4.393e+03, percent-clipped=9.0
+2023-03-08 10:46:28,786 INFO [train.py:968] (0/2) Epoch 16, batch 28400, giga_loss[loss=0.2761, simple_loss=0.3508, pruned_loss=0.1007, over 29060.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3746, pruned_loss=0.1241, over 5655723.41 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3667, pruned_loss=0.1186, over 5700039.46 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.376, pruned_loss=0.1247, over 5644868.19 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 10:46:34,956 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-08 10:47:07,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:47:20,655 INFO [train.py:968] (0/2) Epoch 16, batch 28450, giga_loss[loss=0.3885, simple_loss=0.4161, pruned_loss=0.1804, over 26642.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3738, pruned_loss=0.124, over 5642722.67 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3664, pruned_loss=0.1184, over 5701343.07 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3754, pruned_loss=0.1249, over 5631656.40 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:47:45,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=712420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:47:46,970 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=712423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:48:10,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.144e+03 1.800e+03 2.260e+03 3.108e+03 8.585e+03, threshold=4.520e+03, percent-clipped=13.0
+2023-03-08 10:48:14,609 INFO [train.py:968] (0/2) Epoch 16, batch 28500, giga_loss[loss=0.2988, simple_loss=0.3642, pruned_loss=0.1167, over 28886.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3729, pruned_loss=0.124, over 5650845.47 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3653, pruned_loss=0.1175, over 5710382.04 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3756, pruned_loss=0.1257, over 5630810.84 frames. ], batch size: 199, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:48:15,418 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=712452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:48:24,119 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=712457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:48:29,992 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 10:49:16,030 INFO [train.py:968] (0/2) Epoch 16, batch 28550, giga_loss[loss=0.2781, simple_loss=0.3504, pruned_loss=0.1029, over 29026.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3711, pruned_loss=0.1237, over 5638181.73 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.365, pruned_loss=0.1174, over 5711458.13 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3735, pruned_loss=0.1253, over 5621287.47 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:49:18,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:49:26,606 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=712512.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:49:32,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6709, 2.4882, 1.7201, 0.7204], device='cuda:0'), covar=tensor([0.4892, 0.2863, 0.3687, 0.5515], device='cuda:0'), in_proj_covar=tensor([0.1662, 0.1582, 0.1554, 0.1364], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 10:49:40,669 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=712529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:49:43,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=712532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:49:55,245 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.572e+03 2.177e+03 2.968e+03 9.015e+03, threshold=4.354e+03, percent-clipped=6.0
+2023-03-08 10:50:01,681 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 10:50:01,868 INFO [train.py:968] (0/2) Epoch 16, batch 28600, giga_loss[loss=0.2959, simple_loss=0.3664, pruned_loss=0.1127, over 28613.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3721, pruned_loss=0.1248, over 5656045.85 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3656, pruned_loss=0.1177, over 5715617.74 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3738, pruned_loss=0.1259, over 5636573.38 frames. ], batch size: 307, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:50:11,432 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=712561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:50:22,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 10:50:52,712 INFO [train.py:968] (0/2) Epoch 16, batch 28650, giga_loss[loss=0.3664, simple_loss=0.3908, pruned_loss=0.171, over 23307.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3718, pruned_loss=0.1248, over 5649365.69 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3656, pruned_loss=0.1179, over 5708523.97 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3731, pruned_loss=0.1256, over 5639998.70 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:51:34,091 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9862, 4.8173, 4.5952, 2.2356], device='cuda:0'), covar=tensor([0.0481, 0.0617, 0.0660, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.1172, 0.1092, 0.0939, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 10:51:35,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.600e+03 2.326e+03 3.248e+03 9.095e+03, threshold=4.651e+03, percent-clipped=7.0
+2023-03-08 10:51:37,723 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=712647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:51:40,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=712650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:51:41,944 INFO [train.py:968] (0/2) Epoch 16, batch 28700, giga_loss[loss=0.3505, simple_loss=0.3861, pruned_loss=0.1575, over 23439.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3715, pruned_loss=0.1244, over 5652342.36 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3658, pruned_loss=0.118, over 5709149.02 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3725, pruned_loss=0.1252, over 5643240.75 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:52:08,979 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=712679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:52:31,891 INFO [train.py:968] (0/2) Epoch 16, batch 28750, giga_loss[loss=0.3071, simple_loss=0.3732, pruned_loss=0.1205, over 28926.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3719, pruned_loss=0.1245, over 5663330.96 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3662, pruned_loss=0.1182, over 5712022.74 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3725, pruned_loss=0.1249, over 5652609.49 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:53:12,094 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-08 10:53:17,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.540e+03 1.958e+03 2.821e+03 5.559e+03, threshold=3.916e+03, percent-clipped=7.0
+2023-03-08 10:53:21,311 INFO [train.py:968] (0/2) Epoch 16, batch 28800, giga_loss[loss=0.2889, simple_loss=0.3671, pruned_loss=0.1054, over 29041.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3751, pruned_loss=0.1273, over 5665526.31 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3657, pruned_loss=0.1179, over 5716126.25 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3763, pruned_loss=0.1282, over 5651734.69 frames. ], batch size: 164, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:53:55,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=712780.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:54:13,862 INFO [train.py:968] (0/2) Epoch 16, batch 28850, giga_loss[loss=0.2887, simple_loss=0.3628, pruned_loss=0.1073, over 28941.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3748, pruned_loss=0.1272, over 5672893.54 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3655, pruned_loss=0.1178, over 5719144.58 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3761, pruned_loss=0.1282, over 5658561.78 frames. ], batch size: 164, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:54:19,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-08 10:54:44,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:54:58,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.740e+02 1.684e+03 2.040e+03 2.765e+03 8.351e+03, threshold=4.079e+03, percent-clipped=9.0
+2023-03-08 10:55:04,137 INFO [train.py:968] (0/2) Epoch 16, batch 28900, giga_loss[loss=0.2785, simple_loss=0.3563, pruned_loss=0.1003, over 28845.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3743, pruned_loss=0.1267, over 5678154.44 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3655, pruned_loss=0.1178, over 5720221.85 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3753, pruned_loss=0.1275, over 5665712.27 frames. ], batch size: 199, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:55:24,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5478, 1.7325, 1.6832, 1.5383], device='cuda:0'), covar=tensor([0.1577, 0.1971, 0.2164, 0.1992], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0742, 0.0696, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 10:55:37,962 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712887.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:55:51,270 INFO [train.py:968] (0/2) Epoch 16, batch 28950, giga_loss[loss=0.3007, simple_loss=0.3685, pruned_loss=0.1165, over 28719.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3739, pruned_loss=0.1259, over 5686053.75 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3651, pruned_loss=0.1175, over 5722956.42 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3753, pruned_loss=0.1271, over 5672571.68 frames. ], batch size: 284, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:56:05,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3528, 1.4226, 1.2960, 1.5834], device='cuda:0'), covar=tensor([0.0737, 0.0347, 0.0327, 0.0786], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 10:56:41,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.367e+02 1.723e+03 2.187e+03 3.194e+03 1.014e+04, threshold=4.374e+03, percent-clipped=10.0
+2023-03-08 10:56:44,621 INFO [train.py:968] (0/2) Epoch 16, batch 29000, giga_loss[loss=0.3327, simple_loss=0.3811, pruned_loss=0.1422, over 29100.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3746, pruned_loss=0.1266, over 5676245.02 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.365, pruned_loss=0.1175, over 5723702.99 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3758, pruned_loss=0.1275, over 5664536.84 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:56:58,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5418, 1.6401, 1.6923, 1.5557], device='cuda:0'), covar=tensor([0.1863, 0.2276, 0.2336, 0.2115], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0741, 0.0696, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 10:57:08,113 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=712975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:57:12,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=712978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:57:33,241 INFO [train.py:968] (0/2) Epoch 16, batch 29050, libri_loss[loss=0.2577, simple_loss=0.3297, pruned_loss=0.09287, over 29567.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3752, pruned_loss=0.127, over 5664556.79 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.365, pruned_loss=0.1177, over 5705443.20 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3764, pruned_loss=0.1278, over 5670942.14 frames. ], batch size: 75, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:57:38,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=713007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:58:00,970 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=713030.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:58:03,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=713033.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:58:16,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.736e+02 1.611e+03 1.962e+03 2.660e+03 1.382e+04, threshold=3.924e+03, percent-clipped=9.0
+2023-03-08 10:58:19,489 INFO [train.py:968] (0/2) Epoch 16, batch 29100, giga_loss[loss=0.3306, simple_loss=0.4034, pruned_loss=0.1289, over 28986.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3773, pruned_loss=0.1286, over 5662906.47 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3651, pruned_loss=0.1177, over 5708113.33 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3784, pruned_loss=0.1294, over 5664823.17 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:58:29,683 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=713062.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:59:07,493 INFO [train.py:968] (0/2) Epoch 16, batch 29150, giga_loss[loss=0.2803, simple_loss=0.3585, pruned_loss=0.1011, over 28963.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3784, pruned_loss=0.1295, over 5661575.13 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3653, pruned_loss=0.1178, over 5710812.28 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3792, pruned_loss=0.1302, over 5659983.85 frames. ], batch size: 136, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:59:51,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.585e+03 1.896e+03 2.330e+03 6.324e+03, threshold=3.792e+03, percent-clipped=5.0
+2023-03-08 10:59:55,539 INFO [train.py:968] (0/2) Epoch 16, batch 29200, giga_loss[loss=0.3617, simple_loss=0.4033, pruned_loss=0.16, over 26610.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3792, pruned_loss=0.1294, over 5652028.71 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.366, pruned_loss=0.1182, over 5716623.28 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3796, pruned_loss=0.1299, over 5643915.88 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:00:01,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=713155.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:00:50,674 INFO [train.py:968] (0/2) Epoch 16, batch 29250, giga_loss[loss=0.339, simple_loss=0.3895, pruned_loss=0.1443, over 28265.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3791, pruned_loss=0.1287, over 5645502.56 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3663, pruned_loss=0.1183, over 5716400.99 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3793, pruned_loss=0.1291, over 5638781.30 frames. ], batch size: 77, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:00:52,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3385, 2.9595, 1.4096, 1.4029], device='cuda:0'), covar=tensor([0.0969, 0.0371, 0.0914, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0539, 0.0364, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 11:01:30,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.549e+03 2.035e+03 2.753e+03 6.819e+03, threshold=4.071e+03, percent-clipped=12.0
+2023-03-08 11:01:33,871 INFO [train.py:968] (0/2) Epoch 16, batch 29300, giga_loss[loss=0.2849, simple_loss=0.3587, pruned_loss=0.1056, over 28615.00 frames. ], tot_loss[loss=0.315, simple_loss=0.377, pruned_loss=0.1265, over 5659552.51 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3661, pruned_loss=0.1182, over 5721855.59 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3777, pruned_loss=0.1272, over 5647511.54 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:01:50,159 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=713264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:01:52,849 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3456, 1.5453, 1.6055, 1.2161], device='cuda:0'), covar=tensor([0.1438, 0.2206, 0.1177, 0.1470], device='cuda:0'), in_proj_covar=tensor([0.0869, 0.0701, 0.0917, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 11:02:19,770 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=713298.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:02:21,813 INFO [train.py:968] (0/2) Epoch 16, batch 29350, libri_loss[loss=0.3259, simple_loss=0.3908, pruned_loss=0.1305, over 29640.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3766, pruned_loss=0.1265, over 5668434.67 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3663, pruned_loss=0.1183, over 5725512.52 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3771, pruned_loss=0.1271, over 5654360.08 frames. ], batch size: 91, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:02:22,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=713301.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:02:46,494 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=713330.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:03:07,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.716e+03 2.253e+03 3.012e+03 5.211e+03, threshold=4.505e+03, percent-clipped=10.0
+2023-03-08 11:03:10,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2386, 1.5084, 1.5811, 1.3532], device='cuda:0'), covar=tensor([0.1706, 0.1520, 0.1968, 0.1700], device='cuda:0'), in_proj_covar=tensor([0.0454, 0.0743, 0.0697, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 11:03:11,398 INFO [train.py:968] (0/2) Epoch 16, batch 29400, giga_loss[loss=0.2985, simple_loss=0.3701, pruned_loss=0.1134, over 28923.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3759, pruned_loss=0.1258, over 5654218.52 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3664, pruned_loss=0.1184, over 5719087.06 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3764, pruned_loss=0.1263, over 5647240.35 frames. ], batch size: 145, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:04:05,313 INFO [train.py:968] (0/2) Epoch 16, batch 29450, giga_loss[loss=0.2727, simple_loss=0.3457, pruned_loss=0.09981, over 28909.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3771, pruned_loss=0.1268, over 5655415.58 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3665, pruned_loss=0.1186, over 5712410.25 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3776, pruned_loss=0.1271, over 5654464.03 frames. ], batch size: 186, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:04:52,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.369e+02 1.480e+03 2.069e+03 2.630e+03 6.099e+03, threshold=4.138e+03, percent-clipped=4.0
+2023-03-08 11:04:55,528 INFO [train.py:968] (0/2) Epoch 16, batch 29500, giga_loss[loss=0.2687, simple_loss=0.3395, pruned_loss=0.09897, over 28969.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3763, pruned_loss=0.1277, over 5650909.12 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3665, pruned_loss=0.1186, over 5712886.22 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3768, pruned_loss=0.1281, over 5649126.95 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:05:42,780 INFO [train.py:968] (0/2) Epoch 16, batch 29550, giga_loss[loss=0.3275, simple_loss=0.3808, pruned_loss=0.1371, over 28785.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3761, pruned_loss=0.1271, over 5668452.63 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3665, pruned_loss=0.1186, over 5714711.73 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3767, pruned_loss=0.1277, over 5663864.46 frames. ], batch size: 99, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:06:26,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.762e+03 2.312e+03 3.151e+03 8.710e+03, threshold=4.625e+03, percent-clipped=13.0
+2023-03-08 11:06:29,462 INFO [train.py:968] (0/2) Epoch 16, batch 29600, giga_loss[loss=0.328, simple_loss=0.3877, pruned_loss=0.1342, over 28725.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3783, pruned_loss=0.1292, over 5664932.39 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3658, pruned_loss=0.1181, over 5717284.35 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3797, pruned_loss=0.1302, over 5657720.31 frames. ], batch size: 284, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 11:06:41,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6914, 1.8394, 1.7748, 1.6220], device='cuda:0'), covar=tensor([0.2362, 0.2131, 0.1740, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.1879, 0.1810, 0.1734, 0.1878], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 11:07:21,783 INFO [train.py:968] (0/2) Epoch 16, batch 29650, giga_loss[loss=0.2996, simple_loss=0.3668, pruned_loss=0.1162, over 28729.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3784, pruned_loss=0.1295, over 5650550.87 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3658, pruned_loss=0.1181, over 5719315.18 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3796, pruned_loss=0.1305, over 5642646.45 frames. ], batch size: 119, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:07:52,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5880, 1.8006, 1.5922, 1.6655], device='cuda:0'), covar=tensor([0.1592, 0.1956, 0.2124, 0.1742], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0744, 0.0699, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 11:07:55,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=713639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:08:03,474 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.631e+03 1.970e+03 2.661e+03 8.586e+03, threshold=3.941e+03, percent-clipped=7.0
+2023-03-08 11:08:06,810 INFO [train.py:968] (0/2) Epoch 16, batch 29700, giga_loss[loss=0.3671, simple_loss=0.411, pruned_loss=0.1616, over 28561.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.378, pruned_loss=0.1289, over 5657550.42 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3658, pruned_loss=0.1181, over 5724730.99 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3793, pruned_loss=0.13, over 5644456.54 frames. ], batch size: 78, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:08:55,609 INFO [train.py:968] (0/2) Epoch 16, batch 29750, giga_loss[loss=0.3047, simple_loss=0.3718, pruned_loss=0.1188, over 28736.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.38, pruned_loss=0.1298, over 5642640.09 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3664, pruned_loss=0.1186, over 5699794.95 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3808, pruned_loss=0.1305, over 5651717.22 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:09:23,855 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=713727.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:09:39,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.293e+02 1.476e+03 2.058e+03 2.957e+03 6.864e+03, threshold=4.116e+03, percent-clipped=8.0
+2023-03-08 11:09:41,752 INFO [train.py:968] (0/2) Epoch 16, batch 29800, giga_loss[loss=0.3204, simple_loss=0.3855, pruned_loss=0.1277, over 28679.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3801, pruned_loss=0.1292, over 5632855.30 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3665, pruned_loss=0.1186, over 5686703.97 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3811, pruned_loss=0.1301, over 5650676.61 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:09:46,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4669, 1.6435, 1.3448, 1.5282], device='cuda:0'), covar=tensor([0.2643, 0.2742, 0.3008, 0.2357], device='cuda:0'), in_proj_covar=tensor([0.1413, 0.1037, 0.1252, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 11:10:05,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5025, 3.8893, 1.6106, 1.6169], device='cuda:0'), covar=tensor([0.0913, 0.0367, 0.0899, 0.1283], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0542, 0.0367, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 11:10:15,049 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=713782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:10:16,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2770, 1.2349, 1.1311, 0.9567], device='cuda:0'), covar=tensor([0.0880, 0.0552, 0.1106, 0.0998], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0445, 0.0509, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 11:10:17,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=713785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:10:30,933 INFO [train.py:968] (0/2) Epoch 16, batch 29850, giga_loss[loss=0.3226, simple_loss=0.3851, pruned_loss=0.1301, over 28765.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3786, pruned_loss=0.1284, over 5643526.78 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3667, pruned_loss=0.119, over 5691279.81 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3796, pruned_loss=0.1291, over 5652476.07 frames. ], batch size: 119, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:10:43,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=713814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:11:15,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.748e+03 2.197e+03 3.141e+03 9.041e+03, threshold=4.394e+03, percent-clipped=9.0
+2023-03-08 11:11:19,727 INFO [train.py:968] (0/2) Epoch 16, batch 29900, giga_loss[loss=0.3481, simple_loss=0.3993, pruned_loss=0.1485, over 28734.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3782, pruned_loss=0.1285, over 5656480.08 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.367, pruned_loss=0.1192, over 5697117.25 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.379, pruned_loss=0.129, over 5657400.60 frames. ], batch size: 284, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:11:20,181 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 11:11:21,990 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=713854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:11:45,422 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5855, 2.4022, 2.4459, 2.2568], device='cuda:0'), covar=tensor([0.1629, 0.2578, 0.1941, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0739, 0.0695, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 11:12:06,067 INFO [train.py:968] (0/2) Epoch 16, batch 29950, giga_loss[loss=0.2636, simple_loss=0.3386, pruned_loss=0.09428, over 28910.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3736, pruned_loss=0.1255, over 5664321.01 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3669, pruned_loss=0.1193, over 5698810.07 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3745, pruned_loss=0.1261, over 5663132.27 frames. ], batch size: 136, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:12:11,111 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=713907.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:12:33,665 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-08 11:12:54,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.593e+02 1.650e+03 2.190e+03 3.214e+03 6.373e+03, threshold=4.381e+03, percent-clipped=9.0
+2023-03-08 11:12:57,349 INFO [train.py:968] (0/2) Epoch 16, batch 30000, giga_loss[loss=0.3056, simple_loss=0.3698, pruned_loss=0.1207, over 28907.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3693, pruned_loss=0.124, over 5643199.51 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.367, pruned_loss=0.1194, over 5691295.22 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.37, pruned_loss=0.1244, over 5648137.16 frames. ], batch size: 199, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:12:57,354 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 11:13:05,726 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2111, simple_loss=0.3194, pruned_loss=0.05137, over 944034.00 frames. 
+2023-03-08 11:13:05,726 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 11:13:25,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=713976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:13:49,202 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-714000.pt
+2023-03-08 11:13:50,823 INFO [train.py:968] (0/2) Epoch 16, batch 30050, giga_loss[loss=0.2712, simple_loss=0.346, pruned_loss=0.09821, over 29046.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3684, pruned_loss=0.1239, over 5655115.90 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3671, pruned_loss=0.1194, over 5690999.43 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3689, pruned_loss=0.1242, over 5658799.14 frames. ], batch size: 155, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:14:39,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.743e+02 1.624e+03 2.064e+03 2.749e+03 5.211e+03, threshold=4.129e+03, percent-clipped=3.0
+2023-03-08 11:14:44,379 INFO [train.py:968] (0/2) Epoch 16, batch 30100, giga_loss[loss=0.2994, simple_loss=0.3534, pruned_loss=0.1227, over 24296.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3673, pruned_loss=0.1233, over 5637415.08 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3671, pruned_loss=0.1194, over 5694133.27 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3677, pruned_loss=0.1236, over 5636797.38 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:15:33,751 INFO [train.py:968] (0/2) Epoch 16, batch 30150, giga_loss[loss=0.2899, simple_loss=0.3644, pruned_loss=0.1077, over 27998.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3667, pruned_loss=0.1211, over 5642590.85 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3674, pruned_loss=0.1197, over 5698564.65 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3667, pruned_loss=0.1211, over 5637076.24 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:15:35,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714102.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:16:25,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.549e+02 1.523e+03 1.962e+03 3.036e+03 1.031e+04, threshold=3.923e+03, percent-clipped=16.0
+2023-03-08 11:16:28,514 INFO [train.py:968] (0/2) Epoch 16, batch 30200, giga_loss[loss=0.2843, simple_loss=0.3547, pruned_loss=0.1069, over 28944.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3643, pruned_loss=0.1173, over 5634225.14 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3676, pruned_loss=0.12, over 5689377.18 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3641, pruned_loss=0.1171, over 5637063.13 frames. ], batch size: 199, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:17:19,097 INFO [train.py:968] (0/2) Epoch 16, batch 30250, giga_loss[loss=0.2772, simple_loss=0.3542, pruned_loss=0.1001, over 28690.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3614, pruned_loss=0.1136, over 5647894.18 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3673, pruned_loss=0.1199, over 5691400.24 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3614, pruned_loss=0.1134, over 5647523.26 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:17:50,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:18:05,525 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=714245.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:18:08,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=714248.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:18:09,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.293e+02 1.408e+03 1.770e+03 2.427e+03 7.621e+03, threshold=3.541e+03, percent-clipped=7.0
+2023-03-08 11:18:11,465 INFO [train.py:968] (0/2) Epoch 16, batch 30300, giga_loss[loss=0.2441, simple_loss=0.3262, pruned_loss=0.08094, over 28678.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3583, pruned_loss=0.1103, over 5633578.01 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3675, pruned_loss=0.1201, over 5674442.70 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3581, pruned_loss=0.1099, over 5648234.34 frames. ], batch size: 99, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:18:37,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4673, 1.6903, 1.3321, 1.6094], device='cuda:0'), covar=tensor([0.2658, 0.2577, 0.2987, 0.2312], device='cuda:0'), in_proj_covar=tensor([0.1415, 0.1032, 0.1256, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 11:18:38,591 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=714277.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:18:43,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714282.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:18:51,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7441, 2.1490, 1.3766, 1.6194], device='cuda:0'), covar=tensor([0.0957, 0.0545, 0.1027, 0.1080], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0441, 0.0506, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 11:19:01,754 INFO [train.py:968] (0/2) Epoch 16, batch 30350, giga_loss[loss=0.2698, simple_loss=0.3513, pruned_loss=0.09419, over 28943.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.356, pruned_loss=0.1074, over 5640257.89 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3676, pruned_loss=0.1204, over 5673067.21 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3556, pruned_loss=0.1066, over 5652677.32 frames. ], batch size: 213, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:19:54,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.916e+02 1.337e+03 1.639e+03 2.324e+03 4.337e+03, threshold=3.279e+03, percent-clipped=3.0
+2023-03-08 11:19:55,780 INFO [train.py:968] (0/2) Epoch 16, batch 30400, giga_loss[loss=0.3312, simple_loss=0.3744, pruned_loss=0.1439, over 26515.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3556, pruned_loss=0.1054, over 5658022.82 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.367, pruned_loss=0.1201, over 5680581.03 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3554, pruned_loss=0.1045, over 5660461.75 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:19:56,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:20:16,202 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=714372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:20:18,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=714375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:20:19,690 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=714376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:20:36,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3907, 4.2557, 1.6223, 1.5085], device='cuda:0'), covar=tensor([0.1020, 0.0386, 0.0964, 0.1416], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0536, 0.0365, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 11:20:44,988 INFO [train.py:968] (0/2) Epoch 16, batch 30450, giga_loss[loss=0.2797, simple_loss=0.3541, pruned_loss=0.1026, over 27568.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3549, pruned_loss=0.1052, over 5657020.01 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3661, pruned_loss=0.1199, over 5677835.70 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3552, pruned_loss=0.1042, over 5660653.25 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:20:47,749 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=714404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:21:08,741 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=714425.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:21:10,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=714428.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:21:27,965 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2686, 2.6491, 1.2891, 1.3927], device='cuda:0'), covar=tensor([0.0977, 0.0329, 0.0941, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0535, 0.0364, 0.0410], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 11:21:33,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.381e+02 1.518e+03 1.906e+03 2.473e+03 6.674e+03, threshold=3.812e+03, percent-clipped=10.0
+2023-03-08 11:21:35,708 INFO [train.py:968] (0/2) Epoch 16, batch 30500, giga_loss[loss=0.256, simple_loss=0.3253, pruned_loss=0.09338, over 26686.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3524, pruned_loss=0.103, over 5653188.52 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3658, pruned_loss=0.1198, over 5670342.59 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3527, pruned_loss=0.102, over 5661841.25 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:21:43,093 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=714457.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:22:19,325 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=714494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:22:21,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=714497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:22:25,580 INFO [train.py:968] (0/2) Epoch 16, batch 30550, giga_loss[loss=0.2628, simple_loss=0.3432, pruned_loss=0.09124, over 28752.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3496, pruned_loss=0.1008, over 5656300.16 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3656, pruned_loss=0.1197, over 5672938.60 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3498, pruned_loss=0.09987, over 5660652.71 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:22:54,039 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=714526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:23:18,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.415e+02 1.382e+03 1.771e+03 2.523e+03 5.325e+03, threshold=3.542e+03, percent-clipped=10.0
+2023-03-08 11:23:18,910 INFO [train.py:968] (0/2) Epoch 16, batch 30600, giga_loss[loss=0.2638, simple_loss=0.3478, pruned_loss=0.08985, over 28782.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3489, pruned_loss=0.1004, over 5654540.66 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3654, pruned_loss=0.1197, over 5676194.14 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.349, pruned_loss=0.09938, over 5654793.04 frames. ], batch size: 284, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:24:09,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=714600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:24:10,181 INFO [train.py:968] (0/2) Epoch 16, batch 30650, libri_loss[loss=0.2878, simple_loss=0.3529, pruned_loss=0.1113, over 27887.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3472, pruned_loss=0.09883, over 5661504.13 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3647, pruned_loss=0.1194, over 5679093.51 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3476, pruned_loss=0.09797, over 5658784.48 frames. ], batch size: 116, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:24:20,997 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-08 11:25:04,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.719e+02 1.331e+03 1.893e+03 2.530e+03 1.205e+04, threshold=3.786e+03, percent-clipped=12.0
+2023-03-08 11:25:06,717 INFO [train.py:968] (0/2) Epoch 16, batch 30700, giga_loss[loss=0.2484, simple_loss=0.3311, pruned_loss=0.08284, over 28915.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3446, pruned_loss=0.09663, over 5657989.41 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3645, pruned_loss=0.1193, over 5680070.04 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.345, pruned_loss=0.0959, over 5654914.19 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:25:59,177 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 11:26:01,884 INFO [train.py:968] (0/2) Epoch 16, batch 30750, giga_loss[loss=0.248, simple_loss=0.3142, pruned_loss=0.09091, over 26570.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3423, pruned_loss=0.09537, over 5666301.98 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3646, pruned_loss=0.1196, over 5683398.15 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3422, pruned_loss=0.09413, over 5660598.85 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:26:08,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7883, 1.1318, 2.6641, 2.5630], device='cuda:0'), covar=tensor([0.1404, 0.2182, 0.0619, 0.1054], device='cuda:0'), in_proj_covar=tensor([0.0711, 0.0618, 0.0909, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 11:26:50,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.857e+02 1.410e+03 1.853e+03 3.042e+03 8.480e+03, threshold=3.707e+03, percent-clipped=13.0
+2023-03-08 11:26:52,662 INFO [train.py:968] (0/2) Epoch 16, batch 30800, giga_loss[loss=0.2334, simple_loss=0.3134, pruned_loss=0.07668, over 28034.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3386, pruned_loss=0.09368, over 5661547.27 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.364, pruned_loss=0.1196, over 5675212.93 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3384, pruned_loss=0.09208, over 5663357.89 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:26:52,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:27:23,991 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 11:27:43,326 INFO [train.py:968] (0/2) Epoch 16, batch 30850, libri_loss[loss=0.2933, simple_loss=0.3612, pruned_loss=0.1127, over 27994.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.338, pruned_loss=0.09428, over 5653356.71 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3637, pruned_loss=0.1196, over 5669027.29 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3374, pruned_loss=0.09229, over 5660026.32 frames. ], batch size: 116, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:28:02,333 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2175, 3.0691, 2.9311, 1.6775], device='cuda:0'), covar=tensor([0.0757, 0.0907, 0.0873, 0.1745], device='cuda:0'), in_proj_covar=tensor([0.1133, 0.1056, 0.0905, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 11:28:35,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.678e+02 1.639e+03 2.001e+03 2.453e+03 5.728e+03, threshold=4.003e+03, percent-clipped=11.0
+2023-03-08 11:28:35,047 INFO [train.py:968] (0/2) Epoch 16, batch 30900, giga_loss[loss=0.2704, simple_loss=0.3328, pruned_loss=0.104, over 26690.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3391, pruned_loss=0.09567, over 5631994.25 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3633, pruned_loss=0.1194, over 5660184.46 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.338, pruned_loss=0.09325, over 5644419.48 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:28:39,031 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=714855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:29:24,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=714894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:29:26,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=714897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:29:30,298 INFO [train.py:968] (0/2) Epoch 16, batch 30950, giga_loss[loss=0.3296, simple_loss=0.3821, pruned_loss=0.1385, over 26650.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3424, pruned_loss=0.09753, over 5637721.62 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3631, pruned_loss=0.1195, over 5666955.94 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.09478, over 5640361.82 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:29:57,776 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=714926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:30:30,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.672e+02 1.581e+03 2.070e+03 2.786e+03 4.711e+03, threshold=4.139e+03, percent-clipped=8.0
+2023-03-08 11:30:30,925 INFO [train.py:968] (0/2) Epoch 16, batch 31000, giga_loss[loss=0.257, simple_loss=0.3304, pruned_loss=0.09181, over 27638.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3426, pruned_loss=0.09658, over 5631264.44 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3625, pruned_loss=0.1192, over 5669970.14 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3416, pruned_loss=0.09427, over 5630283.65 frames. ], batch size: 474, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:30:42,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3273, 1.6348, 1.6521, 1.4046], device='cuda:0'), covar=tensor([0.1634, 0.1768, 0.1818, 0.1844], device='cuda:0'), in_proj_covar=tensor([0.0445, 0.0726, 0.0688, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 11:31:02,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:31:41,127 INFO [train.py:968] (0/2) Epoch 16, batch 31050, giga_loss[loss=0.2941, simple_loss=0.3641, pruned_loss=0.1121, over 28997.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3422, pruned_loss=0.09595, over 5629947.91 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3625, pruned_loss=0.1192, over 5670389.37 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3413, pruned_loss=0.09394, over 5628248.14 frames. ], batch size: 285, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:32:14,876 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-08 11:32:14,887 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-08 11:32:25,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:32:30,316 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4163, 1.6788, 1.3648, 1.5505], device='cuda:0'), covar=tensor([0.0773, 0.0313, 0.0340, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 11:32:30,932 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:32:33,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:32:47,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.678e+02 1.418e+03 1.915e+03 2.690e+03 5.733e+03, threshold=3.831e+03, percent-clipped=4.0
+2023-03-08 11:32:47,624 INFO [train.py:968] (0/2) Epoch 16, batch 31100, giga_loss[loss=0.2943, simple_loss=0.3672, pruned_loss=0.1107, over 28876.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3392, pruned_loss=0.09357, over 5643477.38 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3623, pruned_loss=0.1191, over 5673342.74 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3384, pruned_loss=0.09173, over 5638965.04 frames. ], batch size: 284, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:33:03,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9970, 4.8334, 4.5433, 2.1279], device='cuda:0'), covar=tensor([0.0443, 0.0607, 0.0725, 0.2058], device='cuda:0'), in_proj_covar=tensor([0.1139, 0.1056, 0.0906, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 11:33:23,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0916, 1.3618, 1.0154, 1.0367], device='cuda:0'), covar=tensor([0.0861, 0.0394, 0.1052, 0.1016], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0440, 0.0507, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 11:33:56,053 INFO [train.py:968] (0/2) Epoch 16, batch 31150, giga_loss[loss=0.2448, simple_loss=0.3314, pruned_loss=0.07915, over 28678.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3377, pruned_loss=0.0916, over 5637720.50 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3624, pruned_loss=0.1193, over 5676665.71 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3367, pruned_loss=0.08961, over 5631110.32 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:33:56,873 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:34:15,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5721, 1.6966, 1.4168, 1.5789], device='cuda:0'), covar=tensor([0.2930, 0.2755, 0.3235, 0.2528], device='cuda:0'), in_proj_covar=tensor([0.1416, 0.1031, 0.1257, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 11:34:17,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:34:21,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:34:59,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:34:59,421 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.784e+02 1.214e+03 1.675e+03 2.434e+03 4.224e+03, threshold=3.350e+03, percent-clipped=2.0
+2023-03-08 11:34:59,434 INFO [train.py:968] (0/2) Epoch 16, batch 31200, giga_loss[loss=0.2309, simple_loss=0.3123, pruned_loss=0.07473, over 28579.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3345, pruned_loss=0.08924, over 5643917.10 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3618, pruned_loss=0.1189, over 5679144.39 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3339, pruned_loss=0.08759, over 5635915.45 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:35:02,536 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:36:06,088 INFO [train.py:968] (0/2) Epoch 16, batch 31250, giga_loss[loss=0.2913, simple_loss=0.3549, pruned_loss=0.1139, over 27002.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3341, pruned_loss=0.09023, over 5651857.02 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3616, pruned_loss=0.119, over 5672823.85 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3331, pruned_loss=0.08821, over 5650001.69 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:36:39,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:37:04,035 INFO [train.py:968] (0/2) Epoch 16, batch 31300, giga_loss[loss=0.2409, simple_loss=0.3168, pruned_loss=0.08247, over 28894.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3343, pruned_loss=0.09116, over 5668146.43 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3611, pruned_loss=0.1188, over 5679697.08 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3333, pruned_loss=0.08887, over 5660026.02 frames. ], batch size: 186, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:37:04,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.400e+02 1.510e+03 1.862e+03 2.593e+03 6.919e+03, threshold=3.725e+03, percent-clipped=14.0
+2023-03-08 11:37:18,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4599, 1.6558, 1.7608, 1.3023], device='cuda:0'), covar=tensor([0.1902, 0.2693, 0.1536, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0689, 0.0912, 0.0814], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 11:37:37,231 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:37:39,486 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-08 11:37:43,155 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:38:01,603 INFO [train.py:968] (0/2) Epoch 16, batch 31350, giga_loss[loss=0.2263, simple_loss=0.312, pruned_loss=0.07031, over 28988.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.335, pruned_loss=0.09103, over 5667522.00 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3605, pruned_loss=0.1185, over 5681934.24 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3342, pruned_loss=0.08892, over 5658760.55 frames. ], batch size: 93, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:39:05,831 INFO [train.py:968] (0/2) Epoch 16, batch 31400, giga_loss[loss=0.2289, simple_loss=0.3148, pruned_loss=0.07151, over 28321.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3382, pruned_loss=0.09225, over 5652114.32 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.361, pruned_loss=0.1189, over 5676089.08 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3366, pruned_loss=0.08959, over 5649673.83 frames. ], batch size: 78, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:39:07,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.210e+02 1.388e+03 2.034e+03 3.077e+03 9.789e+03, threshold=4.069e+03, percent-clipped=12.0
+2023-03-08 11:39:36,069 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:39:40,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:40:13,513 INFO [train.py:968] (0/2) Epoch 16, batch 31450, giga_loss[loss=0.221, simple_loss=0.3055, pruned_loss=0.06827, over 28675.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3353, pruned_loss=0.09001, over 5650323.71 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.361, pruned_loss=0.1191, over 5659641.91 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3337, pruned_loss=0.08749, over 5661938.86 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:40:22,121 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:40:22,287 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-08 11:40:28,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:40:33,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:40:33,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2967, 1.2992, 3.7755, 3.1571], device='cuda:0'), covar=tensor([0.1557, 0.2674, 0.0401, 0.0947], device='cuda:0'), in_proj_covar=tensor([0.0709, 0.0619, 0.0905, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 11:40:34,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:40:57,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6915, 5.0385, 1.9821, 1.9324], device='cuda:0'), covar=tensor([0.0936, 0.0227, 0.0863, 0.1284], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0533, 0.0365, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 11:41:27,558 INFO [train.py:968] (0/2) Epoch 16, batch 31500, giga_loss[loss=0.2554, simple_loss=0.3394, pruned_loss=0.08576, over 28921.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3364, pruned_loss=0.09091, over 5658631.39 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3607, pruned_loss=0.1191, over 5659694.44 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.335, pruned_loss=0.08858, over 5667517.91 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:41:29,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.350e+02 1.387e+03 1.810e+03 2.501e+03 5.507e+03, threshold=3.620e+03, percent-clipped=6.0
+2023-03-08 11:42:01,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:42:32,272 INFO [train.py:968] (0/2) Epoch 16, batch 31550, giga_loss[loss=0.2643, simple_loss=0.3604, pruned_loss=0.08404, over 28431.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3389, pruned_loss=0.09141, over 5661180.59 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3604, pruned_loss=0.1189, over 5663276.06 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3378, pruned_loss=0.08932, over 5664958.56 frames. ], batch size: 368, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:42:36,693 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:12,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:41,502 INFO [train.py:968] (0/2) Epoch 16, batch 31600, giga_loss[loss=0.2604, simple_loss=0.3469, pruned_loss=0.08693, over 28702.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3412, pruned_loss=0.08996, over 5658050.47 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3606, pruned_loss=0.119, over 5665612.46 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3401, pruned_loss=0.08797, over 5658882.89 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:43:42,022 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:42,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.363e+02 1.478e+03 2.293e+03 3.158e+03 1.209e+04, threshold=4.587e+03, percent-clipped=23.0
+2023-03-08 11:43:46,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:47,571 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:51,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:52,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:56,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:44:25,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:44:29,509 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:44:31,305 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:44:45,654 INFO [train.py:968] (0/2) Epoch 16, batch 31650, giga_loss[loss=0.3053, simple_loss=0.3861, pruned_loss=0.1123, over 28962.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3431, pruned_loss=0.08992, over 5660926.61 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3606, pruned_loss=0.119, over 5670122.33 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3418, pruned_loss=0.08777, over 5657040.38 frames. ], batch size: 155, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:45:06,103 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:45:09,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:45:32,677 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:45:43,234 INFO [train.py:968] (0/2) Epoch 16, batch 31700, libri_loss[loss=0.2825, simple_loss=0.3564, pruned_loss=0.1043, over 29666.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3426, pruned_loss=0.08931, over 5675173.09 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3606, pruned_loss=0.119, over 5677381.07 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3408, pruned_loss=0.08646, over 5664971.32 frames. ], batch size: 91, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:45:43,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:45:44,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.358e+02 1.358e+03 1.924e+03 2.647e+03 5.412e+03, threshold=3.848e+03, percent-clipped=1.0
+2023-03-08 11:45:46,686 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:45:53,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:46:10,459 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:46:13,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:46:50,942 INFO [train.py:968] (0/2) Epoch 16, batch 31750, giga_loss[loss=0.2228, simple_loss=0.3121, pruned_loss=0.06674, over 28860.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3432, pruned_loss=0.09031, over 5673465.95 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3608, pruned_loss=0.1192, over 5668737.17 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3416, pruned_loss=0.08781, over 5672913.93 frames. ], batch size: 164, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:46:55,080 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:48:06,472 INFO [train.py:968] (0/2) Epoch 16, batch 31800, giga_loss[loss=0.2511, simple_loss=0.3363, pruned_loss=0.08296, over 29021.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3412, pruned_loss=0.09093, over 5676601.56 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3601, pruned_loss=0.1189, over 5672218.39 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3401, pruned_loss=0.08846, over 5672923.20 frames. ], batch size: 155, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:48:09,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.409e+02 1.314e+03 1.746e+03 2.400e+03 6.816e+03, threshold=3.492e+03, percent-clipped=5.0
+2023-03-08 11:49:29,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:49:32,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:49:34,187 INFO [train.py:968] (0/2) Epoch 16, batch 31850, giga_loss[loss=0.2399, simple_loss=0.3211, pruned_loss=0.07935, over 29140.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3427, pruned_loss=0.0925, over 5677198.67 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.36, pruned_loss=0.1188, over 5673368.30 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3418, pruned_loss=0.09045, over 5673319.67 frames. ], batch size: 120, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:49:36,489 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:49:42,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:50:18,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:50:18,712 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-08 11:50:25,775 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:50:25,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3530, 1.6364, 1.5703, 1.3265], device='cuda:0'), covar=tensor([0.2827, 0.1921, 0.1666, 0.2171], device='cuda:0'), in_proj_covar=tensor([0.1843, 0.1759, 0.1675, 0.1821], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 11:50:48,996 INFO [train.py:968] (0/2) Epoch 16, batch 31900, giga_loss[loss=0.2648, simple_loss=0.3418, pruned_loss=0.09394, over 28399.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3374, pruned_loss=0.08934, over 5676801.11 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3601, pruned_loss=0.1189, over 5675729.80 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3362, pruned_loss=0.08725, over 5671564.81 frames. ], batch size: 368, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:50:51,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.568e+02 1.367e+03 1.743e+03 2.637e+03 5.536e+03, threshold=3.487e+03, percent-clipped=10.0
+2023-03-08 11:51:06,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:51:17,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:51:34,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3914, 3.2317, 1.4491, 1.5489], device='cuda:0'), covar=tensor([0.0954, 0.0398, 0.0954, 0.1303], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0534, 0.0364, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 11:51:45,767 INFO [train.py:968] (0/2) Epoch 16, batch 31950, giga_loss[loss=0.2286, simple_loss=0.315, pruned_loss=0.07112, over 28480.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.336, pruned_loss=0.08939, over 5679510.63 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3601, pruned_loss=0.1193, over 5682659.64 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3339, pruned_loss=0.08585, over 5668961.42 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:52:03,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-08 11:52:44,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4112, 4.2367, 4.0336, 1.9292], device='cuda:0'), covar=tensor([0.0578, 0.0691, 0.0826, 0.2134], device='cuda:0'), in_proj_covar=tensor([0.1131, 0.1046, 0.0903, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 11:52:56,182 INFO [train.py:968] (0/2) Epoch 16, batch 32000, giga_loss[loss=0.2696, simple_loss=0.3233, pruned_loss=0.1079, over 24435.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3335, pruned_loss=0.08823, over 5685050.98 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3597, pruned_loss=0.1191, over 5685064.26 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3321, pruned_loss=0.08536, over 5674472.26 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:52:57,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.006e+02 1.368e+03 1.752e+03 2.558e+03 7.096e+03, threshold=3.504e+03, percent-clipped=14.0
+2023-03-08 11:53:59,741 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-716000.pt
+2023-03-08 11:54:00,960 INFO [train.py:968] (0/2) Epoch 16, batch 32050, giga_loss[loss=0.2789, simple_loss=0.3587, pruned_loss=0.09955, over 29106.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3381, pruned_loss=0.09049, over 5678786.97 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3595, pruned_loss=0.119, over 5678432.30 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3369, pruned_loss=0.08793, over 5676020.72 frames. ], batch size: 200, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:54:19,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=716016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:54:26,433 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=716021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:54:29,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=716024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:55:08,667 INFO [train.py:968] (0/2) Epoch 16, batch 32100, giga_loss[loss=0.2357, simple_loss=0.3011, pruned_loss=0.08517, over 24302.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.338, pruned_loss=0.09144, over 5685186.11 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3597, pruned_loss=0.1192, over 5680677.25 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3367, pruned_loss=0.08901, over 5681204.21 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:55:13,897 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=716053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:55:14,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.966e+02 1.414e+03 1.836e+03 2.761e+03 6.701e+03, threshold=3.673e+03, percent-clipped=16.0
+2023-03-08 11:56:09,623 INFO [train.py:968] (0/2) Epoch 16, batch 32150, giga_loss[loss=0.2555, simple_loss=0.3343, pruned_loss=0.08831, over 28188.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3383, pruned_loss=0.09271, over 5679820.35 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3596, pruned_loss=0.1193, over 5676211.25 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3367, pruned_loss=0.08995, over 5681145.79 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:56:14,324 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=716103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:56:29,677 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-08 11:57:15,039 INFO [train.py:968] (0/2) Epoch 16, batch 32200, giga_loss[loss=0.2638, simple_loss=0.3361, pruned_loss=0.09574, over 28604.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3372, pruned_loss=0.09207, over 5681162.44 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3591, pruned_loss=0.119, over 5679314.17 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3361, pruned_loss=0.08979, over 5679498.24 frames. ], batch size: 85, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:57:21,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.805e+02 1.566e+03 2.319e+03 3.219e+03 1.301e+04, threshold=4.638e+03, percent-clipped=19.0
+2023-03-08 11:57:28,606 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=716159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:57:31,434 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=716162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:58:09,135 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=716191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:58:23,972 INFO [train.py:968] (0/2) Epoch 16, batch 32250, giga_loss[loss=0.2572, simple_loss=0.3411, pruned_loss=0.08664, over 28886.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3393, pruned_loss=0.09265, over 5682150.37 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3583, pruned_loss=0.1186, over 5685453.09 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3384, pruned_loss=0.09031, over 5675373.28 frames. ], batch size: 99, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:58:46,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6676, 2.4581, 1.5356, 0.7974], device='cuda:0'), covar=tensor([0.7768, 0.3537, 0.4096, 0.6467], device='cuda:0'), in_proj_covar=tensor([0.1657, 0.1565, 0.1549, 0.1362], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 11:59:29,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=716242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:59:45,387 INFO [train.py:968] (0/2) Epoch 16, batch 32300, giga_loss[loss=0.2526, simple_loss=0.339, pruned_loss=0.0831, over 29038.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3403, pruned_loss=0.0926, over 5676026.73 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3583, pruned_loss=0.1186, over 5688730.45 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3394, pruned_loss=0.09044, over 5667815.73 frames. ], batch size: 285, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:59:47,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.048e+02 1.384e+03 1.733e+03 2.238e+03 4.751e+03, threshold=3.466e+03, percent-clipped=2.0
+2023-03-08 12:00:41,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4407, 2.1779, 1.5453, 0.5652], device='cuda:0'), covar=tensor([0.5326, 0.2694, 0.4584, 0.5975], device='cuda:0'), in_proj_covar=tensor([0.1662, 0.1570, 0.1553, 0.1367], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 12:00:57,230 INFO [train.py:968] (0/2) Epoch 16, batch 32350, giga_loss[loss=0.2423, simple_loss=0.3174, pruned_loss=0.08363, over 28651.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3371, pruned_loss=0.09104, over 5680115.94 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3586, pruned_loss=0.1189, over 5692626.25 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3359, pruned_loss=0.08864, over 5670230.14 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:01:00,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-08 12:01:41,361 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.42 vs. limit=5.0
+2023-03-08 12:02:09,454 INFO [train.py:968] (0/2) Epoch 16, batch 32400, giga_loss[loss=0.2148, simple_loss=0.2927, pruned_loss=0.06839, over 29027.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3314, pruned_loss=0.08887, over 5680516.22 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3584, pruned_loss=0.1188, over 5691523.88 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3303, pruned_loss=0.08682, over 5673567.81 frames. ], batch size: 93, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 12:02:15,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.628e+02 1.412e+03 1.935e+03 2.638e+03 8.184e+03, threshold=3.870e+03, percent-clipped=10.0
+2023-03-08 12:02:40,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5819, 4.1944, 1.8162, 1.6897], device='cuda:0'), covar=tensor([0.0908, 0.0285, 0.0865, 0.1245], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0531, 0.0363, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 12:02:58,887 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=716385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:03:02,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=716388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:03:18,926 INFO [train.py:968] (0/2) Epoch 16, batch 32450, giga_loss[loss=0.2371, simple_loss=0.3096, pruned_loss=0.08229, over 28595.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.329, pruned_loss=0.08796, over 5670006.98 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3582, pruned_loss=0.1186, over 5690971.36 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3275, pruned_loss=0.08562, over 5664594.05 frames. ], batch size: 85, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:03:35,912 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=716417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:04:13,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5921, 1.6550, 1.7780, 1.3758], device='cuda:0'), covar=tensor([0.1668, 0.2449, 0.1421, 0.1760], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0687, 0.0912, 0.0814], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 12:04:17,257 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=716449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:04:19,879 INFO [train.py:968] (0/2) Epoch 16, batch 32500, giga_loss[loss=0.2581, simple_loss=0.3364, pruned_loss=0.0899, over 28704.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3309, pruned_loss=0.08947, over 5667500.76 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3582, pruned_loss=0.1188, over 5684729.75 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3291, pruned_loss=0.08688, over 5668044.00 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:04:23,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.777e+02 1.485e+03 1.825e+03 2.538e+03 5.190e+03, threshold=3.649e+03, percent-clipped=2.0
+2023-03-08 12:04:55,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=716478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:05:22,317 INFO [train.py:968] (0/2) Epoch 16, batch 32550, libri_loss[loss=0.3313, simple_loss=0.3847, pruned_loss=0.139, over 27461.00 frames. ], tot_loss[loss=0.256, simple_loss=0.332, pruned_loss=0.09002, over 5674969.60 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.358, pruned_loss=0.1187, over 5686709.80 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3303, pruned_loss=0.08749, over 5673326.66 frames. ], batch size: 115, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:05:34,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1793, 1.2756, 1.0687, 0.9870], device='cuda:0'), covar=tensor([0.0902, 0.0475, 0.1066, 0.0940], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0437, 0.0506, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 12:05:57,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5714, 1.9667, 1.8327, 1.3657], device='cuda:0'), covar=tensor([0.1580, 0.2399, 0.1396, 0.1702], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0687, 0.0913, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 12:06:25,869 INFO [train.py:968] (0/2) Epoch 16, batch 32600, giga_loss[loss=0.2142, simple_loss=0.306, pruned_loss=0.06119, over 28910.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3295, pruned_loss=0.08791, over 5659375.70 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3576, pruned_loss=0.1184, over 5682235.58 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3279, pruned_loss=0.08548, over 5661745.25 frames. ], batch size: 136, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:06:29,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.553e+02 1.520e+03 1.890e+03 2.621e+03 1.336e+04, threshold=3.779e+03, percent-clipped=6.0
+2023-03-08 12:07:02,627 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=716581.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:07:30,157 INFO [train.py:968] (0/2) Epoch 16, batch 32650, giga_loss[loss=0.2349, simple_loss=0.3092, pruned_loss=0.08033, over 29080.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3293, pruned_loss=0.08809, over 5658065.02 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3576, pruned_loss=0.1185, over 5679318.11 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3273, pruned_loss=0.08526, over 5662829.06 frames. ], batch size: 113, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:07:58,559 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=716621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:08:04,830 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=716624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:08:07,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-08 12:08:08,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4238, 1.7345, 1.4152, 1.3183], device='cuda:0'), covar=tensor([0.2836, 0.2637, 0.3108, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.1412, 0.1028, 0.1255, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 12:08:40,222 INFO [train.py:968] (0/2) Epoch 16, batch 32700, giga_loss[loss=0.2113, simple_loss=0.2959, pruned_loss=0.06335, over 28863.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.328, pruned_loss=0.08732, over 5664224.23 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3575, pruned_loss=0.1184, over 5683798.03 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.326, pruned_loss=0.08456, over 5663561.24 frames. ], batch size: 99, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:08:42,551 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=716653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:08:45,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.486e+02 1.311e+03 1.760e+03 2.310e+03 1.046e+04, threshold=3.519e+03, percent-clipped=9.0
+2023-03-08 12:09:51,811 INFO [train.py:968] (0/2) Epoch 16, batch 32750, giga_loss[loss=0.2325, simple_loss=0.3152, pruned_loss=0.07486, over 28630.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3288, pruned_loss=0.0869, over 5678569.25 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3571, pruned_loss=0.1183, over 5686547.54 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3272, pruned_loss=0.08444, over 5675428.99 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:10:57,161 INFO [train.py:968] (0/2) Epoch 16, batch 32800, giga_loss[loss=0.2682, simple_loss=0.3426, pruned_loss=0.09687, over 28549.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.33, pruned_loss=0.08815, over 5677453.67 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3569, pruned_loss=0.1182, over 5687623.28 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3284, pruned_loss=0.08572, over 5674006.43 frames. ], batch size: 370, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 12:11:01,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.334e+02 1.316e+03 1.854e+03 2.465e+03 8.347e+03, threshold=3.707e+03, percent-clipped=13.0
+2023-03-08 12:12:01,285 INFO [train.py:968] (0/2) Epoch 16, batch 32850, giga_loss[loss=0.2857, simple_loss=0.362, pruned_loss=0.1047, over 28495.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3295, pruned_loss=0.08822, over 5678251.66 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3564, pruned_loss=0.118, over 5683885.83 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.328, pruned_loss=0.08585, over 5678009.54 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:12:31,700 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=716824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:12:47,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6456, 2.2645, 1.6148, 0.7820], device='cuda:0'), covar=tensor([0.5026, 0.2469, 0.3960, 0.5492], device='cuda:0'), in_proj_covar=tensor([0.1660, 0.1572, 0.1551, 0.1364], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 12:13:03,366 INFO [train.py:968] (0/2) Epoch 16, batch 32900, giga_loss[loss=0.2732, simple_loss=0.3609, pruned_loss=0.09274, over 28964.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3291, pruned_loss=0.0874, over 5670865.85 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3553, pruned_loss=0.1174, over 5688441.41 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3283, pruned_loss=0.08537, over 5666374.05 frames. ], batch size: 213, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:13:04,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.12 vs. limit=5.0
+2023-03-08 12:13:10,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.588e+02 1.301e+03 1.778e+03 2.266e+03 5.080e+03, threshold=3.557e+03, percent-clipped=3.0
+2023-03-08 12:13:12,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=716859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:14:06,217 INFO [train.py:968] (0/2) Epoch 16, batch 32950, giga_loss[loss=0.2299, simple_loss=0.2961, pruned_loss=0.08186, over 24488.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3327, pruned_loss=0.08841, over 5664397.56 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3551, pruned_loss=0.1172, over 5691148.46 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.332, pruned_loss=0.08652, over 5658298.22 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:15:05,968 INFO [train.py:968] (0/2) Epoch 16, batch 33000, giga_loss[loss=0.2122, simple_loss=0.3072, pruned_loss=0.05861, over 28735.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3347, pruned_loss=0.08901, over 5669597.38 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3551, pruned_loss=0.1173, over 5695469.95 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3336, pruned_loss=0.08688, over 5660489.36 frames. ], batch size: 119, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:15:05,973 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 12:15:14,318 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.1984, simple_loss=0.2997, pruned_loss=0.04852, over 944034.00 frames. 
+2023-03-08 12:15:14,319 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 12:15:19,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=716956.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 12:15:20,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.438e+02 1.461e+03 1.908e+03 2.890e+03 1.081e+04, threshold=3.816e+03, percent-clipped=15.0
+2023-03-08 12:15:25,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3932, 1.6017, 1.1983, 1.1359], device='cuda:0'), covar=tensor([0.0832, 0.0404, 0.0894, 0.1045], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0435, 0.0505, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 12:15:32,954 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=716967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:15:37,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=716970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:16:09,653 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 12:16:10,108 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=716999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:16:11,690 INFO [train.py:968] (0/2) Epoch 16, batch 33050, giga_loss[loss=0.2378, simple_loss=0.323, pruned_loss=0.07629, over 28778.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3354, pruned_loss=0.09019, over 5656770.07 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3545, pruned_loss=0.1171, over 5680603.64 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3342, pruned_loss=0.0873, over 5661804.21 frames. ], batch size: 243, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:17:03,541 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=717042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:17:11,162 INFO [train.py:968] (0/2) Epoch 16, batch 33100, giga_loss[loss=0.1971, simple_loss=0.2907, pruned_loss=0.0518, over 28827.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3359, pruned_loss=0.09048, over 5668410.17 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3546, pruned_loss=0.1171, over 5687747.79 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3342, pruned_loss=0.0874, over 5665496.61 frames. ], batch size: 164, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:17:20,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.651e+02 1.446e+03 1.776e+03 2.316e+03 8.941e+03, threshold=3.553e+03, percent-clipped=9.0
+2023-03-08 12:17:21,808 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=717057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:18:11,841 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717099.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:18:13,614 INFO [train.py:968] (0/2) Epoch 16, batch 33150, giga_loss[loss=0.2438, simple_loss=0.3367, pruned_loss=0.07544, over 28843.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3328, pruned_loss=0.08827, over 5669279.53 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3554, pruned_loss=0.1179, over 5683228.32 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3303, pruned_loss=0.08446, over 5670617.83 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:18:15,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717102.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:18:19,877 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 12:18:54,265 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717131.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:18:55,147 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=717132.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 12:19:19,425 INFO [train.py:968] (0/2) Epoch 16, batch 33200, giga_loss[loss=0.1998, simple_loss=0.2849, pruned_loss=0.0573, over 28938.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3307, pruned_loss=0.08709, over 5674580.43 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3554, pruned_loss=0.1178, over 5685430.45 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3285, pruned_loss=0.0839, over 5673526.88 frames. ], batch size: 164, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:19:21,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3717, 0.8115, 0.8581, 1.4446], device='cuda:0'), covar=tensor([0.0683, 0.0333, 0.0347, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 12:19:26,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.980e+02 1.270e+03 1.564e+03 2.424e+03 5.775e+03, threshold=3.129e+03, percent-clipped=5.0
+2023-03-08 12:20:05,373 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-08 12:20:07,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=717188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:20:20,177 INFO [train.py:968] (0/2) Epoch 16, batch 33250, giga_loss[loss=0.2594, simple_loss=0.3395, pruned_loss=0.08966, over 27627.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.331, pruned_loss=0.08804, over 5672927.96 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3551, pruned_loss=0.1178, over 5690189.28 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3289, pruned_loss=0.08482, over 5667565.54 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:20:37,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4074, 3.1367, 1.5671, 1.4770], device='cuda:0'), covar=tensor([0.0967, 0.0255, 0.0930, 0.1384], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0530, 0.0364, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 12:21:08,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:21:30,794 INFO [train.py:968] (0/2) Epoch 16, batch 33300, giga_loss[loss=0.2739, simple_loss=0.3304, pruned_loss=0.1087, over 24500.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3337, pruned_loss=0.08898, over 5663223.91 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3554, pruned_loss=0.118, over 5681050.90 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3316, pruned_loss=0.08586, over 5667431.28 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:21:38,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.066e+02 1.367e+03 1.929e+03 3.347e+03 1.171e+04, threshold=3.857e+03, percent-clipped=27.0
+2023-03-08 12:22:24,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5571, 1.6610, 1.1727, 1.1806], device='cuda:0'), covar=tensor([0.0814, 0.0470, 0.0971, 0.1199], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0438, 0.0509, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 12:22:37,207 INFO [train.py:968] (0/2) Epoch 16, batch 33350, giga_loss[loss=0.2508, simple_loss=0.3308, pruned_loss=0.08539, over 28138.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3346, pruned_loss=0.09006, over 5671928.34 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3552, pruned_loss=0.118, over 5685591.86 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3326, pruned_loss=0.08694, over 5671028.95 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:23:42,252 INFO [train.py:968] (0/2) Epoch 16, batch 33400, giga_loss[loss=0.2978, simple_loss=0.3836, pruned_loss=0.106, over 28814.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3379, pruned_loss=0.09225, over 5658422.54 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3552, pruned_loss=0.1179, over 5682127.30 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3357, pruned_loss=0.08906, over 5660624.50 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:23:49,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.603e+02 1.408e+03 1.870e+03 2.811e+03 1.060e+04, threshold=3.740e+03, percent-clipped=12.0
+2023-03-08 12:24:11,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4237, 3.2451, 1.5149, 1.5520], device='cuda:0'), covar=tensor([0.0931, 0.0334, 0.0958, 0.1305], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0530, 0.0365, 0.0410], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 12:24:13,582 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:24:17,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:24:38,485 INFO [train.py:968] (0/2) Epoch 16, batch 33450, giga_loss[loss=0.2472, simple_loss=0.3375, pruned_loss=0.07847, over 28879.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3408, pruned_loss=0.09393, over 5659139.52 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3548, pruned_loss=0.1178, over 5688354.76 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3389, pruned_loss=0.09071, over 5654815.15 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:24:49,458 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:25:02,871 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:25:19,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:25:37,580 INFO [train.py:968] (0/2) Epoch 16, batch 33500, libri_loss[loss=0.27, simple_loss=0.3256, pruned_loss=0.1072, over 29569.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3425, pruned_loss=0.09437, over 5663270.07 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3543, pruned_loss=0.1174, over 5689288.19 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3409, pruned_loss=0.09114, over 5657842.77 frames. ], batch size: 76, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:25:46,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.014e+02 1.357e+03 1.748e+03 2.684e+03 9.776e+03, threshold=3.495e+03, percent-clipped=8.0
+2023-03-08 12:26:08,901 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 12:26:54,654 INFO [train.py:968] (0/2) Epoch 16, batch 33550, giga_loss[loss=0.259, simple_loss=0.345, pruned_loss=0.08654, over 28625.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3418, pruned_loss=0.09378, over 5669101.71 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3543, pruned_loss=0.1173, over 5688682.75 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3406, pruned_loss=0.09122, over 5665226.77 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:27:02,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717507.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 12:27:32,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2102, 1.3531, 1.1032, 1.0532], device='cuda:0'), covar=tensor([0.0894, 0.0446, 0.1044, 0.0940], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0436, 0.0507, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 12:27:48,415 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=717538.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 12:28:00,691 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.41 vs. limit=5.0
+2023-03-08 12:28:06,246 INFO [train.py:968] (0/2) Epoch 16, batch 33600, giga_loss[loss=0.2627, simple_loss=0.3426, pruned_loss=0.0914, over 28733.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3388, pruned_loss=0.09202, over 5679761.47 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3544, pruned_loss=0.1174, over 5692378.65 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3374, pruned_loss=0.08943, over 5673066.43 frames. ], batch size: 243, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 12:28:12,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.340e+02 1.379e+03 1.812e+03 2.421e+03 5.074e+03, threshold=3.623e+03, percent-clipped=7.0
+2023-03-08 12:28:17,283 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:28:22,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:28:22,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:28:32,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:28:34,595 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-08 12:28:37,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:28:56,218 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:29:08,452 INFO [train.py:968] (0/2) Epoch 16, batch 33650, giga_loss[loss=0.2158, simple_loss=0.301, pruned_loss=0.06533, over 29106.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3372, pruned_loss=0.09113, over 5682706.06 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3535, pruned_loss=0.1169, over 5696932.67 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3364, pruned_loss=0.0887, over 5672699.94 frames. ], batch size: 155, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:29:17,663 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:29:23,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5388, 2.6356, 1.5568, 1.7229], device='cuda:0'), covar=tensor([0.0733, 0.0294, 0.0741, 0.0993], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0528, 0.0363, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 12:29:24,220 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 12:29:33,020 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-08 12:30:14,169 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717650.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:30:14,506 INFO [train.py:968] (0/2) Epoch 16, batch 33700, giga_loss[loss=0.2333, simple_loss=0.3127, pruned_loss=0.07691, over 28944.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3364, pruned_loss=0.09167, over 5679085.13 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3535, pruned_loss=0.1168, over 5696582.89 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3354, pruned_loss=0.08923, over 5671078.99 frames. ], batch size: 106, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:30:17,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717653.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:30:22,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.977e+02 1.492e+03 1.898e+03 2.675e+03 7.447e+03, threshold=3.796e+03, percent-clipped=7.0
+2023-03-08 12:30:54,691 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717682.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:31:15,904 INFO [train.py:968] (0/2) Epoch 16, batch 33750, giga_loss[loss=0.2853, simple_loss=0.352, pruned_loss=0.1093, over 28089.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3346, pruned_loss=0.09175, over 5689021.22 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3529, pruned_loss=0.1165, over 5702809.45 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3337, pruned_loss=0.08911, over 5676376.55 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:31:24,952 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:31:26,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:31:31,620 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 12:32:00,864 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:32:17,130 INFO [train.py:968] (0/2) Epoch 16, batch 33800, libri_loss[loss=0.218, simple_loss=0.2779, pruned_loss=0.07901, over 29500.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3343, pruned_loss=0.09087, over 5695840.61 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3519, pruned_loss=0.116, over 5707294.48 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3338, pruned_loss=0.08834, over 5680983.72 frames. ], batch size: 70, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:32:25,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.946e+02 1.475e+03 1.932e+03 2.783e+03 1.132e+04, threshold=3.864e+03, percent-clipped=13.0
+2023-03-08 12:33:16,931 INFO [train.py:968] (0/2) Epoch 16, batch 33850, giga_loss[loss=0.2335, simple_loss=0.2979, pruned_loss=0.08449, over 24476.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3324, pruned_loss=0.08945, over 5669347.18 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3522, pruned_loss=0.1162, over 5697445.80 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3312, pruned_loss=0.08646, over 5665822.93 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:34:12,455 INFO [train.py:968] (0/2) Epoch 16, batch 33900, giga_loss[loss=0.2554, simple_loss=0.3453, pruned_loss=0.08271, over 28673.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3339, pruned_loss=0.08822, over 5678036.03 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3519, pruned_loss=0.1161, over 5702667.70 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3327, pruned_loss=0.08526, over 5670262.80 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:34:21,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.659e+02 1.313e+03 1.709e+03 2.561e+03 6.436e+03, threshold=3.418e+03, percent-clipped=9.0
+2023-03-08 12:35:12,706 INFO [train.py:968] (0/2) Epoch 16, batch 33950, giga_loss[loss=0.2599, simple_loss=0.3413, pruned_loss=0.08929, over 28789.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3352, pruned_loss=0.08761, over 5681588.89 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3519, pruned_loss=0.1162, over 5702105.73 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.334, pruned_loss=0.08481, over 5675802.02 frames. ], batch size: 243, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:35:28,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717913.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:35:43,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8560, 1.2140, 1.3016, 1.0466], device='cuda:0'), covar=tensor([0.1830, 0.1308, 0.2093, 0.1627], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0720, 0.0680, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 12:35:49,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7724, 1.1074, 2.8619, 2.6636], device='cuda:0'), covar=tensor([0.1686, 0.2579, 0.0535, 0.1236], device='cuda:0'), in_proj_covar=tensor([0.0701, 0.0614, 0.0898, 0.0817], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 12:36:10,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 12:36:17,505 INFO [train.py:968] (0/2) Epoch 16, batch 34000, giga_loss[loss=0.2217, simple_loss=0.3125, pruned_loss=0.0655, over 29037.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3349, pruned_loss=0.08744, over 5677748.28 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3516, pruned_loss=0.1161, over 5704311.16 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3338, pruned_loss=0.08453, over 5670654.35 frames. ], batch size: 155, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:36:31,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.336e+02 1.451e+03 1.915e+03 2.672e+03 8.455e+03, threshold=3.830e+03, percent-clipped=13.0
+2023-03-08 12:36:48,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0250, 3.8326, 3.6622, 1.7749], device='cuda:0'), covar=tensor([0.0633, 0.0792, 0.0767, 0.2307], device='cuda:0'), in_proj_covar=tensor([0.1125, 0.1039, 0.0898, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 12:37:30,302 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-718000.pt
+2023-03-08 12:37:31,310 INFO [train.py:968] (0/2) Epoch 16, batch 34050, giga_loss[loss=0.2562, simple_loss=0.3384, pruned_loss=0.08696, over 28464.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3347, pruned_loss=0.08739, over 5665442.78 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3519, pruned_loss=0.1163, over 5696620.28 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3333, pruned_loss=0.08437, over 5665404.65 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:38:40,454 INFO [train.py:968] (0/2) Epoch 16, batch 34100, giga_loss[loss=0.2384, simple_loss=0.3311, pruned_loss=0.07282, over 28800.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3345, pruned_loss=0.0873, over 5663539.13 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3515, pruned_loss=0.1162, over 5700102.99 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3335, pruned_loss=0.08466, over 5660055.47 frames. ], batch size: 174, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:38:46,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2205, 0.8928, 0.9105, 1.3962], device='cuda:0'), covar=tensor([0.0736, 0.0417, 0.0371, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0114, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0064, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 12:38:49,662 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=718056.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:38:53,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=718059.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:38:54,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.062e+02 1.362e+03 1.835e+03 2.869e+03 1.090e+04, threshold=3.670e+03, percent-clipped=9.0
+2023-03-08 12:39:29,230 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 12:39:39,354 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=718088.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:39:59,152 INFO [train.py:968] (0/2) Epoch 16, batch 34150, giga_loss[loss=0.2326, simple_loss=0.3256, pruned_loss=0.06982, over 28888.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3346, pruned_loss=0.08617, over 5668397.12 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3513, pruned_loss=0.116, over 5703453.60 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3337, pruned_loss=0.08375, over 5662161.91 frames. ], batch size: 164, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:40:44,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.30 vs. limit=5.0
+2023-03-08 12:41:04,243 INFO [train.py:968] (0/2) Epoch 16, batch 34200, libri_loss[loss=0.2504, simple_loss=0.3123, pruned_loss=0.09421, over 29578.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3359, pruned_loss=0.08699, over 5667769.43 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3508, pruned_loss=0.1158, over 5698220.93 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3352, pruned_loss=0.08436, over 5666074.99 frames. ], batch size: 74, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:41:14,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.769e+02 1.336e+03 1.794e+03 2.431e+03 5.182e+03, threshold=3.588e+03, percent-clipped=8.0
+2023-03-08 12:42:12,431 INFO [train.py:968] (0/2) Epoch 16, batch 34250, libri_loss[loss=0.2623, simple_loss=0.3365, pruned_loss=0.09406, over 29538.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3386, pruned_loss=0.08789, over 5667683.84 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3505, pruned_loss=0.1157, over 5692276.65 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3381, pruned_loss=0.08538, over 5670271.03 frames. ], batch size: 89, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:43:24,580 INFO [train.py:968] (0/2) Epoch 16, batch 34300, giga_loss[loss=0.2172, simple_loss=0.3024, pruned_loss=0.06603, over 28618.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3376, pruned_loss=0.08813, over 5668310.24 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.35, pruned_loss=0.1154, over 5696807.01 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3375, pruned_loss=0.08594, over 5665895.57 frames. ], batch size: 85, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:43:37,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.102e+02 1.527e+03 2.281e+03 3.288e+03 1.372e+04, threshold=4.563e+03, percent-clipped=18.0
+2023-03-08 12:44:32,352 INFO [train.py:968] (0/2) Epoch 16, batch 34350, giga_loss[loss=0.2413, simple_loss=0.3202, pruned_loss=0.08122, over 28987.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3362, pruned_loss=0.08787, over 5676512.05 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.35, pruned_loss=0.1153, over 5690657.27 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3356, pruned_loss=0.08533, over 5679841.40 frames. ], batch size: 136, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:45:25,480 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 12:45:50,633 INFO [train.py:968] (0/2) Epoch 16, batch 34400, giga_loss[loss=0.2891, simple_loss=0.3652, pruned_loss=0.1065, over 28710.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3341, pruned_loss=0.08565, over 5674692.52 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3499, pruned_loss=0.1152, over 5691728.99 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3337, pruned_loss=0.0836, over 5676176.15 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:46:01,853 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.549e+02 1.132e+03 1.350e+03 1.803e+03 8.831e+03, threshold=2.700e+03, percent-clipped=1.0
+2023-03-08 12:46:59,296 INFO [train.py:968] (0/2) Epoch 16, batch 34450, giga_loss[loss=0.246, simple_loss=0.3282, pruned_loss=0.08194, over 27532.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3334, pruned_loss=0.08543, over 5666940.45 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3497, pruned_loss=0.1151, over 5694083.83 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.333, pruned_loss=0.08338, over 5665447.21 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:47:37,025 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=718431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:48:00,820 INFO [train.py:968] (0/2) Epoch 16, batch 34500, giga_loss[loss=0.2434, simple_loss=0.3344, pruned_loss=0.07623, over 28941.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3365, pruned_loss=0.08759, over 5667502.73 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3494, pruned_loss=0.1149, over 5695787.95 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3362, pruned_loss=0.0856, over 5664382.50 frames. ], batch size: 199, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:48:10,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.425e+02 1.343e+03 1.723e+03 2.242e+03 3.407e+03, threshold=3.447e+03, percent-clipped=14.0
+2023-03-08 12:49:06,772 INFO [train.py:968] (0/2) Epoch 16, batch 34550, giga_loss[loss=0.2723, simple_loss=0.3353, pruned_loss=0.1047, over 26916.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3381, pruned_loss=0.08823, over 5676294.77 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3493, pruned_loss=0.1149, over 5698016.56 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3379, pruned_loss=0.08649, over 5671686.47 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:50:07,721 INFO [train.py:968] (0/2) Epoch 16, batch 34600, giga_loss[loss=0.2718, simple_loss=0.3304, pruned_loss=0.1066, over 26760.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3358, pruned_loss=0.08849, over 5667929.27 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3491, pruned_loss=0.1146, over 5699817.46 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3355, pruned_loss=0.08657, over 5661811.47 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:50:18,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.333e+02 1.596e+03 2.142e+03 3.047e+03 6.030e+03, threshold=4.283e+03, percent-clipped=18.0
+2023-03-08 12:50:36,988 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-08 12:51:06,065 INFO [train.py:968] (0/2) Epoch 16, batch 34650, giga_loss[loss=0.2525, simple_loss=0.3272, pruned_loss=0.08893, over 27621.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3363, pruned_loss=0.0897, over 5662095.79 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3495, pruned_loss=0.115, over 5692990.02 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3354, pruned_loss=0.0874, over 5663451.51 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:51:16,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1127, 1.0863, 3.7217, 3.0582], device='cuda:0'), covar=tensor([0.1728, 0.2789, 0.0449, 0.0927], device='cuda:0'), in_proj_covar=tensor([0.0705, 0.0619, 0.0904, 0.0820], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 12:52:06,638 INFO [train.py:968] (0/2) Epoch 16, batch 34700, giga_loss[loss=0.2703, simple_loss=0.3609, pruned_loss=0.08989, over 28621.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3371, pruned_loss=0.09067, over 5661533.42 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3494, pruned_loss=0.1147, over 5696012.06 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3362, pruned_loss=0.08851, over 5659247.97 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:52:14,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.064e+02 1.422e+03 1.915e+03 2.595e+03 8.588e+03, threshold=3.829e+03, percent-clipped=6.0
+2023-03-08 12:52:15,133 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 12:52:22,958 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=718669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:52:30,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3570, 2.8879, 1.5116, 1.4498], device='cuda:0'), covar=tensor([0.0929, 0.0350, 0.0886, 0.1290], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0527, 0.0364, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 12:52:46,841 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5611, 1.9058, 1.5485, 1.7016], device='cuda:0'), covar=tensor([0.2456, 0.2335, 0.2743, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.1407, 0.1025, 0.1251, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 12:52:49,931 INFO [train.py:968] (0/2) Epoch 16, batch 34750, giga_loss[loss=0.2912, simple_loss=0.3692, pruned_loss=0.1066, over 28589.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3447, pruned_loss=0.09524, over 5664504.94 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3492, pruned_loss=0.1145, over 5690094.72 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3438, pruned_loss=0.0927, over 5666528.06 frames. ], batch size: 71, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:52:56,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5010, 3.1793, 1.5648, 1.5780], device='cuda:0'), covar=tensor([0.0956, 0.0306, 0.0918, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0527, 0.0364, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 12:53:37,940 INFO [train.py:968] (0/2) Epoch 16, batch 34800, giga_loss[loss=0.3545, simple_loss=0.392, pruned_loss=0.1585, over 23961.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3527, pruned_loss=0.09975, over 5666030.02 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3487, pruned_loss=0.1141, over 5694318.40 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3523, pruned_loss=0.09777, over 5663480.30 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:53:45,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.855e+02 1.315e+03 1.804e+03 2.293e+03 5.924e+03, threshold=3.607e+03, percent-clipped=5.0
+2023-03-08 12:54:00,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7621, 1.8763, 1.4125, 1.4480], device='cuda:0'), covar=tensor([0.0870, 0.0599, 0.1031, 0.1107], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0437, 0.0509, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 12:54:17,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4727, 3.2889, 1.5651, 1.5196], device='cuda:0'), covar=tensor([0.0972, 0.0302, 0.0922, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0528, 0.0363, 0.0410], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 12:54:22,044 INFO [train.py:968] (0/2) Epoch 16, batch 34850, giga_loss[loss=0.2797, simple_loss=0.3514, pruned_loss=0.1041, over 28329.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3524, pruned_loss=0.1001, over 5676519.18 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.348, pruned_loss=0.1136, over 5699400.69 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3529, pruned_loss=0.09863, over 5669512.82 frames. ], batch size: 77, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:54:25,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=718806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:55:02,973 INFO [train.py:968] (0/2) Epoch 16, batch 34900, giga_loss[loss=0.2439, simple_loss=0.3277, pruned_loss=0.08007, over 28943.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3469, pruned_loss=0.09814, over 5678136.08 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3481, pruned_loss=0.1136, over 5692841.76 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3471, pruned_loss=0.09648, over 5678389.03 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:55:03,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8305, 1.9303, 2.0526, 1.6132], device='cuda:0'), covar=tensor([0.1758, 0.2190, 0.1417, 0.1569], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0683, 0.0912, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 12:55:12,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.197e+02 1.163e+03 1.462e+03 2.231e+03 7.815e+03, threshold=2.924e+03, percent-clipped=11.0
+2023-03-08 12:55:45,485 INFO [train.py:968] (0/2) Epoch 16, batch 34950, giga_loss[loss=0.2553, simple_loss=0.3293, pruned_loss=0.09066, over 28919.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3412, pruned_loss=0.09593, over 5669988.42 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3484, pruned_loss=0.1137, over 5679981.50 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3411, pruned_loss=0.09412, over 5681015.34 frames. ], batch size: 174, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:56:26,126 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=718949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:56:27,260 INFO [train.py:968] (0/2) Epoch 16, batch 35000, giga_loss[loss=0.2297, simple_loss=0.3037, pruned_loss=0.07781, over 28886.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3344, pruned_loss=0.09323, over 5677148.52 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3486, pruned_loss=0.1137, over 5685511.54 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3338, pruned_loss=0.09121, over 5680844.00 frames. ], batch size: 112, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:56:28,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=718952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:56:36,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.751e+02 1.100e+03 1.334e+03 1.756e+03 6.102e+03, threshold=2.669e+03, percent-clipped=7.0
+2023-03-08 12:56:55,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=718981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:57:03,741 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-08 12:57:10,457 INFO [train.py:968] (0/2) Epoch 16, batch 35050, giga_loss[loss=0.2324, simple_loss=0.3007, pruned_loss=0.08206, over 28846.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3277, pruned_loss=0.09061, over 5686099.03 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3487, pruned_loss=0.1137, over 5690090.56 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3267, pruned_loss=0.08853, over 5684806.00 frames. ], batch size: 284, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:57:15,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8350, 1.9713, 2.0773, 1.6305], device='cuda:0'), covar=tensor([0.1874, 0.2286, 0.1462, 0.1674], device='cuda:0'), in_proj_covar=tensor([0.0869, 0.0687, 0.0916, 0.0818], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 12:57:48,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=719044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:57:53,459 INFO [train.py:968] (0/2) Epoch 16, batch 35100, giga_loss[loss=0.2254, simple_loss=0.2914, pruned_loss=0.07975, over 28521.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3215, pruned_loss=0.08808, over 5686515.56 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3487, pruned_loss=0.1135, over 5691601.26 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3203, pruned_loss=0.08609, over 5683930.63 frames. ], batch size: 71, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:58:01,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.988e+02 1.132e+03 1.356e+03 1.770e+03 5.369e+03, threshold=2.711e+03, percent-clipped=8.0
+2023-03-08 12:58:34,344 INFO [train.py:968] (0/2) Epoch 16, batch 35150, giga_loss[loss=0.1967, simple_loss=0.2684, pruned_loss=0.06252, over 28582.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.319, pruned_loss=0.08704, over 5691357.36 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3492, pruned_loss=0.1135, over 5699123.08 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3165, pruned_loss=0.08456, over 5682308.85 frames. ], batch size: 85, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:59:18,861 INFO [train.py:968] (0/2) Epoch 16, batch 35200, giga_loss[loss=0.2181, simple_loss=0.2933, pruned_loss=0.07144, over 28670.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3162, pruned_loss=0.0856, over 5700187.04 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3497, pruned_loss=0.1136, over 5704046.90 frames. ], giga_tot_loss[loss=0.2393, simple_loss=0.313, pruned_loss=0.08282, over 5688301.78 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:59:27,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.752e+02 1.036e+03 1.292e+03 1.738e+03 3.836e+03, threshold=2.585e+03, percent-clipped=4.0
+2023-03-08 12:59:47,906 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=719187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:59:50,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=719190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:59:59,848 INFO [train.py:968] (0/2) Epoch 16, batch 35250, giga_loss[loss=0.2098, simple_loss=0.2829, pruned_loss=0.06838, over 28942.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3138, pruned_loss=0.08466, over 5707845.34 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3502, pruned_loss=0.1138, over 5707882.10 frames. ], giga_tot_loss[loss=0.2359, simple_loss=0.3094, pruned_loss=0.08118, over 5694856.84 frames. ], batch size: 136, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:00:04,277 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=719206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:00:06,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=719207.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:00:16,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=719219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:00:35,835 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=719240.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:00:45,463 INFO [train.py:968] (0/2) Epoch 16, batch 35300, giga_loss[loss=0.1962, simple_loss=0.2773, pruned_loss=0.05762, over 28851.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3108, pruned_loss=0.083, over 5696239.73 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3505, pruned_loss=0.1139, over 5692582.67 frames. ], giga_tot_loss[loss=0.2327, simple_loss=0.3063, pruned_loss=0.07955, over 5700693.49 frames. ], batch size: 145, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:00:54,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.377e+02 1.038e+03 1.353e+03 2.037e+03 1.044e+04, threshold=2.707e+03, percent-clipped=12.0
+2023-03-08 13:01:10,491 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-08 13:01:28,461 INFO [train.py:968] (0/2) Epoch 16, batch 35350, giga_loss[loss=0.1911, simple_loss=0.2643, pruned_loss=0.05895, over 28491.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3073, pruned_loss=0.08133, over 5691774.77 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3502, pruned_loss=0.1136, over 5694787.43 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.3028, pruned_loss=0.07798, over 5693354.91 frames. ], batch size: 60, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:01:35,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2568, 0.8736, 0.9852, 1.4650], device='cuda:0'), covar=tensor([0.0777, 0.0389, 0.0360, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 13:01:56,396 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=719332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:02:12,755 INFO [train.py:968] (0/2) Epoch 16, batch 35400, giga_loss[loss=0.2404, simple_loss=0.3018, pruned_loss=0.08951, over 26661.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.304, pruned_loss=0.07959, over 5691868.16 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3504, pruned_loss=0.1136, over 5694725.51 frames. ], giga_tot_loss[loss=0.226, simple_loss=0.2994, pruned_loss=0.07625, over 5693078.53 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:02:21,650 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.572e+02 1.004e+03 1.331e+03 1.788e+03 6.604e+03, threshold=2.662e+03, percent-clipped=9.0
+2023-03-08 13:02:53,108 INFO [train.py:968] (0/2) Epoch 16, batch 35450, giga_loss[loss=0.2241, simple_loss=0.2867, pruned_loss=0.08076, over 28805.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3022, pruned_loss=0.07877, over 5694588.38 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.351, pruned_loss=0.1139, over 5700494.52 frames. ], giga_tot_loss[loss=0.2236, simple_loss=0.297, pruned_loss=0.07509, over 5690471.39 frames. ], batch size: 99, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:03:35,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7210, 1.9874, 1.5860, 1.9860], device='cuda:0'), covar=tensor([0.2441, 0.2598, 0.2891, 0.2561], device='cuda:0'), in_proj_covar=tensor([0.1413, 0.1029, 0.1252, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 13:03:39,842 INFO [train.py:968] (0/2) Epoch 16, batch 35500, giga_loss[loss=0.1906, simple_loss=0.276, pruned_loss=0.05264, over 28896.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.2991, pruned_loss=0.0772, over 5702103.61 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3509, pruned_loss=0.1136, over 5704007.28 frames. ], giga_tot_loss[loss=0.2206, simple_loss=0.2938, pruned_loss=0.07371, over 5695391.71 frames. ], batch size: 174, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:03:49,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.019e+02 9.974e+02 1.170e+03 1.843e+03 6.064e+03, threshold=2.341e+03, percent-clipped=6.0
+2023-03-08 13:03:59,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2081, 2.6340, 2.2365, 1.8808], device='cuda:0'), covar=tensor([0.2758, 0.1859, 0.1881, 0.2406], device='cuda:0'), in_proj_covar=tensor([0.1861, 0.1763, 0.1683, 0.1845], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 13:04:26,330 INFO [train.py:968] (0/2) Epoch 16, batch 35550, giga_loss[loss=0.2748, simple_loss=0.3468, pruned_loss=0.1014, over 28690.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.2987, pruned_loss=0.07745, over 5685867.08 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.351, pruned_loss=0.1135, over 5698073.16 frames. ], giga_tot_loss[loss=0.2209, simple_loss=0.2936, pruned_loss=0.07412, over 5685140.23 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:04:36,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7344, 1.8751, 1.5902, 1.9886], device='cuda:0'), covar=tensor([0.2416, 0.2614, 0.2822, 0.2392], device='cuda:0'), in_proj_covar=tensor([0.1413, 0.1027, 0.1253, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 13:05:09,025 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 13:05:10,997 INFO [train.py:968] (0/2) Epoch 16, batch 35600, libri_loss[loss=0.3246, simple_loss=0.3843, pruned_loss=0.1325, over 18716.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3105, pruned_loss=0.08363, over 5671524.92 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3511, pruned_loss=0.1136, over 5683126.44 frames. ], giga_tot_loss[loss=0.2326, simple_loss=0.3051, pruned_loss=0.08007, over 5685436.21 frames. ], batch size: 186, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:05:20,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.455e+02 1.220e+03 1.460e+03 2.136e+03 1.011e+04, threshold=2.921e+03, percent-clipped=22.0
+2023-03-08 13:05:43,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=719581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:05:43,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=719582.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:05:45,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2062, 1.2530, 1.1349, 0.9336], device='cuda:0'), covar=tensor([0.0937, 0.0539, 0.1108, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0437, 0.0508, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 13:06:00,241 INFO [train.py:968] (0/2) Epoch 16, batch 35650, giga_loss[loss=0.2795, simple_loss=0.3526, pruned_loss=0.1032, over 28295.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.324, pruned_loss=0.09084, over 5669956.90 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3512, pruned_loss=0.1136, over 5677731.24 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.319, pruned_loss=0.08753, over 5684948.10 frames. ], batch size: 65, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:06:07,792 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-08 13:06:14,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=719615.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:06:24,258 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-08 13:06:40,632 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-08 13:06:46,230 INFO [train.py:968] (0/2) Epoch 16, batch 35700, giga_loss[loss=0.296, simple_loss=0.3737, pruned_loss=0.1091, over 28700.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3359, pruned_loss=0.0968, over 5674569.83 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3514, pruned_loss=0.1138, over 5676286.04 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3315, pruned_loss=0.09384, over 5687360.71 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:06:53,137 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 13:06:57,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.960e+02 1.297e+03 1.631e+03 2.335e+03 7.490e+03, threshold=3.262e+03, percent-clipped=17.0
+2023-03-08 13:07:32,661 INFO [train.py:968] (0/2) Epoch 16, batch 35750, giga_loss[loss=0.2654, simple_loss=0.3417, pruned_loss=0.09454, over 28813.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3408, pruned_loss=0.09807, over 5675241.44 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3516, pruned_loss=0.114, over 5669637.69 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.337, pruned_loss=0.09541, over 5691648.41 frames. ], batch size: 99, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:07:36,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=719707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:07:51,915 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=719724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:07:52,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=719725.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:07:53,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=719727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:07:54,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=719728.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:08:16,823 INFO [train.py:968] (0/2) Epoch 16, batch 35800, libri_loss[loss=0.3267, simple_loss=0.3841, pruned_loss=0.1347, over 29473.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3434, pruned_loss=0.09816, over 5674059.65 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3523, pruned_loss=0.1141, over 5668223.11 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3394, pruned_loss=0.0953, over 5688379.99 frames. ], batch size: 85, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:08:19,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3055, 1.4576, 1.3442, 1.2120], device='cuda:0'), covar=tensor([0.2306, 0.1975, 0.1482, 0.2019], device='cuda:0'), in_proj_covar=tensor([0.1860, 0.1773, 0.1693, 0.1856], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 13:08:20,758 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=719756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:08:21,415 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=719757.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:08:23,342 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=719758.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:08:25,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=719761.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:08:26,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.043e+02 1.273e+03 1.569e+03 2.028e+03 4.343e+03, threshold=3.137e+03, percent-clipped=5.0
+2023-03-08 13:08:34,544 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=719770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:08:48,978 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-08 13:08:53,838 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=719790.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:09:04,281 INFO [train.py:968] (0/2) Epoch 16, batch 35850, giga_loss[loss=0.2695, simple_loss=0.3539, pruned_loss=0.0926, over 28913.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3466, pruned_loss=0.09973, over 5669669.49 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3526, pruned_loss=0.1143, over 5663993.21 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3429, pruned_loss=0.09687, over 5684862.67 frames. ], batch size: 213, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:09:06,509 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-08 13:09:29,903 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4622, 2.4657, 2.3894, 2.3689], device='cuda:0'), covar=tensor([0.1719, 0.2158, 0.1876, 0.1797], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0734, 0.0689, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 13:09:45,961 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=719850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:09:46,344 INFO [train.py:968] (0/2) Epoch 16, batch 35900, giga_loss[loss=0.2724, simple_loss=0.3422, pruned_loss=0.1014, over 28870.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3477, pruned_loss=0.1007, over 5668725.89 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3525, pruned_loss=0.1141, over 5657894.50 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3447, pruned_loss=0.09825, over 5685564.16 frames. ], batch size: 99, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 13:09:49,506 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=719853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:09:59,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.455e+02 1.174e+03 1.488e+03 2.195e+03 1.157e+04, threshold=2.977e+03, percent-clipped=12.0
+2023-03-08 13:10:14,360 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=719882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:10:25,812 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6285, 1.7055, 1.8373, 1.4359], device='cuda:0'), covar=tensor([0.1812, 0.2493, 0.1424, 0.1727], device='cuda:0'), in_proj_covar=tensor([0.0872, 0.0692, 0.0920, 0.0820], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 13:10:30,332 INFO [train.py:968] (0/2) Epoch 16, batch 35950, giga_loss[loss=0.2866, simple_loss=0.3652, pruned_loss=0.104, over 28664.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3518, pruned_loss=0.1037, over 5667587.22 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3527, pruned_loss=0.1142, over 5659245.02 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3493, pruned_loss=0.1016, over 5679636.87 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 13:10:33,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8189, 1.8694, 1.7138, 1.7182], device='cuda:0'), covar=tensor([0.1763, 0.2128, 0.2251, 0.2027], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0734, 0.0690, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 13:10:56,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1979, 0.8516, 0.9114, 1.3381], device='cuda:0'), covar=tensor([0.0841, 0.0401, 0.0376, 0.0907], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 13:11:06,809 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 13:11:14,245 INFO [train.py:968] (0/2) Epoch 16, batch 36000, giga_loss[loss=0.27, simple_loss=0.3412, pruned_loss=0.09938, over 28586.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3542, pruned_loss=0.1049, over 5675516.91 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3529, pruned_loss=0.1144, over 5662985.00 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3521, pruned_loss=0.103, over 5681818.89 frames. ], batch size: 85, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:11:14,249 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 13:11:19,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3340, 1.6884, 1.6257, 1.1733], device='cuda:0'), covar=tensor([0.1993, 0.2911, 0.1765, 0.2014], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0693, 0.0920, 0.0820], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 13:11:23,452 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2132, simple_loss=0.3197, pruned_loss=0.05332, over 944034.00 frames. 
+2023-03-08 13:11:23,452 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 13:11:32,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.633e+02 1.184e+03 1.437e+03 1.904e+03 4.436e+03, threshold=2.874e+03, percent-clipped=7.0
+2023-03-08 13:12:02,110 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-720000.pt
+2023-03-08 13:12:03,010 INFO [train.py:968] (0/2) Epoch 16, batch 36050, giga_loss[loss=0.2858, simple_loss=0.3648, pruned_loss=0.1034, over 28916.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3564, pruned_loss=0.105, over 5684751.93 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3531, pruned_loss=0.1145, over 5663037.21 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3545, pruned_loss=0.1033, over 5689988.73 frames. ], batch size: 199, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:12:24,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4059, 1.4954, 1.3564, 1.2613], device='cuda:0'), covar=tensor([0.2321, 0.2486, 0.1702, 0.2054], device='cuda:0'), in_proj_covar=tensor([0.1853, 0.1771, 0.1693, 0.1850], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 13:12:30,827 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2349, 1.4143, 1.2810, 1.0584], device='cuda:0'), covar=tensor([0.2394, 0.2435, 0.1572, 0.2201], device='cuda:0'), in_proj_covar=tensor([0.1852, 0.1770, 0.1693, 0.1849], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 13:12:48,171 INFO [train.py:968] (0/2) Epoch 16, batch 36100, giga_loss[loss=0.304, simple_loss=0.3684, pruned_loss=0.1198, over 27599.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3576, pruned_loss=0.1052, over 5665114.88 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3535, pruned_loss=0.1146, over 5656844.81 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3559, pruned_loss=0.1036, over 5675195.73 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:12:57,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.348e+02 1.174e+03 1.414e+03 1.805e+03 5.518e+03, threshold=2.827e+03, percent-clipped=6.0
+2023-03-08 13:13:29,133 INFO [train.py:968] (0/2) Epoch 16, batch 36150, giga_loss[loss=0.254, simple_loss=0.3382, pruned_loss=0.08489, over 28666.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3572, pruned_loss=0.1032, over 5674317.99 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3536, pruned_loss=0.1146, over 5651137.35 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3557, pruned_loss=0.1018, over 5688257.22 frames. ], batch size: 78, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:14:05,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=720145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:14:09,279 INFO [train.py:968] (0/2) Epoch 16, batch 36200, giga_loss[loss=0.2647, simple_loss=0.3458, pruned_loss=0.09179, over 28820.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3562, pruned_loss=0.1017, over 5690118.59 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.354, pruned_loss=0.1149, over 5656617.32 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3548, pruned_loss=0.1001, over 5696949.75 frames. ], batch size: 99, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:14:19,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.351e+02 1.204e+03 1.574e+03 2.507e+03 1.200e+04, threshold=3.148e+03, percent-clipped=18.0
+2023-03-08 13:14:52,339 INFO [train.py:968] (0/2) Epoch 16, batch 36250, giga_loss[loss=0.2399, simple_loss=0.3296, pruned_loss=0.07513, over 28507.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3548, pruned_loss=0.09985, over 5699094.15 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.355, pruned_loss=0.1154, over 5660703.15 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3528, pruned_loss=0.09772, over 5701575.94 frames. ], batch size: 60, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:15:36,186 INFO [train.py:968] (0/2) Epoch 16, batch 36300, giga_loss[loss=0.297, simple_loss=0.356, pruned_loss=0.119, over 28670.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3551, pruned_loss=0.1004, over 5696923.53 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3551, pruned_loss=0.1154, over 5654151.28 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3534, pruned_loss=0.09854, over 5706213.18 frames. ], batch size: 92, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:15:48,383 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.878e+02 1.100e+03 1.394e+03 1.932e+03 6.227e+03, threshold=2.788e+03, percent-clipped=3.0
+2023-03-08 13:16:12,439 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=720288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:16:16,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=720291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:16:26,137 INFO [train.py:968] (0/2) Epoch 16, batch 36350, giga_loss[loss=0.2517, simple_loss=0.3304, pruned_loss=0.08649, over 28923.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3583, pruned_loss=0.1056, over 5692395.89 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3553, pruned_loss=0.1154, over 5653958.86 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3569, pruned_loss=0.104, over 5700195.31 frames. ], batch size: 66, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:16:38,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4126, 1.7008, 1.4447, 1.6304], device='cuda:0'), covar=tensor([0.0802, 0.0316, 0.0316, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0100], device='cuda:0')
+2023-03-08 13:16:43,185 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=720320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:17:10,726 INFO [train.py:968] (0/2) Epoch 16, batch 36400, giga_loss[loss=0.3304, simple_loss=0.3887, pruned_loss=0.136, over 29000.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3594, pruned_loss=0.1081, over 5690628.74 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3551, pruned_loss=0.1152, over 5656566.72 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3585, pruned_loss=0.1069, over 5694798.03 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:17:22,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=720362.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:17:23,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.354e+02 1.269e+03 1.533e+03 1.900e+03 4.056e+03, threshold=3.066e+03, percent-clipped=3.0
+2023-03-08 13:17:57,641 INFO [train.py:968] (0/2) Epoch 16, batch 36450, giga_loss[loss=0.2568, simple_loss=0.3353, pruned_loss=0.08917, over 28609.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3585, pruned_loss=0.1085, over 5690936.09 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3554, pruned_loss=0.1154, over 5651517.94 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3575, pruned_loss=0.1073, over 5700073.65 frames. ], batch size: 71, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:18:40,989 INFO [train.py:968] (0/2) Epoch 16, batch 36500, giga_loss[loss=0.2855, simple_loss=0.3542, pruned_loss=0.1084, over 28945.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3563, pruned_loss=0.1078, over 5691812.80 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.356, pruned_loss=0.1157, over 5653157.28 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.355, pruned_loss=0.1063, over 5698344.09 frames. ], batch size: 213, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:18:52,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.573e+02 1.247e+03 1.527e+03 2.361e+03 5.287e+03, threshold=3.053e+03, percent-clipped=13.0
+2023-03-08 13:19:05,632 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=720479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:19:24,543 INFO [train.py:968] (0/2) Epoch 16, batch 36550, giga_loss[loss=0.3026, simple_loss=0.3664, pruned_loss=0.1195, over 27972.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3558, pruned_loss=0.1072, over 5697055.06 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.356, pruned_loss=0.1156, over 5659429.10 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3548, pruned_loss=0.1059, over 5697661.54 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:20:09,216 INFO [train.py:968] (0/2) Epoch 16, batch 36600, giga_loss[loss=0.2709, simple_loss=0.343, pruned_loss=0.09944, over 28969.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3545, pruned_loss=0.1058, over 5692409.40 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3566, pruned_loss=0.1159, over 5663876.77 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3531, pruned_loss=0.1043, over 5689703.35 frames. ], batch size: 213, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:20:22,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.531e+02 1.134e+03 1.447e+03 2.109e+03 8.283e+03, threshold=2.894e+03, percent-clipped=6.0
+2023-03-08 13:20:25,434 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 13:20:49,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3943, 1.5818, 1.5564, 1.5125], device='cuda:0'), covar=tensor([0.1667, 0.1716, 0.1833, 0.1581], device='cuda:0'), in_proj_covar=tensor([0.0451, 0.0739, 0.0695, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 13:20:55,269 INFO [train.py:968] (0/2) Epoch 16, batch 36650, giga_loss[loss=0.2581, simple_loss=0.3284, pruned_loss=0.0939, over 28461.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3521, pruned_loss=0.104, over 5684964.62 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3572, pruned_loss=0.1162, over 5665061.62 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3503, pruned_loss=0.1022, over 5682759.90 frames. ], batch size: 85, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:21:42,272 INFO [train.py:968] (0/2) Epoch 16, batch 36700, libri_loss[loss=0.2951, simple_loss=0.3703, pruned_loss=0.11, over 29526.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3464, pruned_loss=0.1011, over 5668213.17 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3573, pruned_loss=0.1162, over 5661655.31 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3446, pruned_loss=0.09926, over 5669923.70 frames. ], batch size: 81, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:21:54,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.393e+02 1.043e+03 1.285e+03 1.621e+03 4.572e+03, threshold=2.570e+03, percent-clipped=9.0
+2023-03-08 13:22:31,257 INFO [train.py:968] (0/2) Epoch 16, batch 36750, giga_loss[loss=0.2334, simple_loss=0.3079, pruned_loss=0.07939, over 28551.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3416, pruned_loss=0.09933, over 5652719.52 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.358, pruned_loss=0.1167, over 5657866.37 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3392, pruned_loss=0.09694, over 5657217.67 frames. ], batch size: 71, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:23:07,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=720734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:23:12,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=720737.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:23:25,112 INFO [train.py:968] (0/2) Epoch 16, batch 36800, giga_loss[loss=0.2279, simple_loss=0.3167, pruned_loss=0.06953, over 28873.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3387, pruned_loss=0.09791, over 5652121.93 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3584, pruned_loss=0.117, over 5660714.84 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3363, pruned_loss=0.09566, over 5653282.81 frames. ], batch size: 174, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:23:35,832 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-08 13:23:38,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.356e+02 9.850e+02 1.141e+03 1.649e+03 5.299e+03, threshold=2.283e+03, percent-clipped=10.0
+2023-03-08 13:24:06,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6123, 1.6966, 1.7322, 1.5142], device='cuda:0'), covar=tensor([0.1804, 0.2117, 0.2194, 0.2245], device='cuda:0'), in_proj_covar=tensor([0.0450, 0.0738, 0.0695, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 13:24:13,167 INFO [train.py:968] (0/2) Epoch 16, batch 36850, giga_loss[loss=0.2868, simple_loss=0.3513, pruned_loss=0.1111, over 28778.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3383, pruned_loss=0.0967, over 5663412.15 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3585, pruned_loss=0.117, over 5661989.26 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3363, pruned_loss=0.09482, over 5663094.27 frames. ], batch size: 99, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:24:55,418 INFO [train.py:968] (0/2) Epoch 16, batch 36900, giga_loss[loss=0.2395, simple_loss=0.3171, pruned_loss=0.08097, over 28870.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3371, pruned_loss=0.09573, over 5666009.85 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3588, pruned_loss=0.1172, over 5656438.94 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3349, pruned_loss=0.09375, over 5670136.61 frames. ], batch size: 199, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:24:59,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=720854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:25:08,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.373e+02 1.029e+03 1.270e+03 1.874e+03 6.225e+03, threshold=2.541e+03, percent-clipped=17.0
+2023-03-08 13:25:22,098 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=720880.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:25:24,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=720883.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:25:38,085 INFO [train.py:968] (0/2) Epoch 16, batch 36950, giga_loss[loss=0.2361, simple_loss=0.3139, pruned_loss=0.07912, over 29076.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3356, pruned_loss=0.09465, over 5684955.62 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.359, pruned_loss=0.1169, over 5663184.73 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3329, pruned_loss=0.09257, over 5682906.61 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:25:47,523 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=720912.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:25:59,304 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-08 13:26:10,632 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2229, 4.0297, 3.8131, 2.0145], device='cuda:0'), covar=tensor([0.0591, 0.0728, 0.0678, 0.2143], device='cuda:0'), in_proj_covar=tensor([0.1116, 0.1031, 0.0887, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 13:26:19,577 INFO [train.py:968] (0/2) Epoch 16, batch 37000, giga_loss[loss=0.2344, simple_loss=0.3042, pruned_loss=0.08224, over 28686.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.334, pruned_loss=0.09408, over 5703780.11 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3596, pruned_loss=0.1173, over 5667589.30 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3309, pruned_loss=0.09175, over 5698766.87 frames. ], batch size: 92, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:26:33,925 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.040e+02 1.104e+03 1.633e+03 2.419e+03 1.504e+04, threshold=3.267e+03, percent-clipped=22.0
+2023-03-08 13:26:39,318 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 13:27:00,044 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=720997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:27:02,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=721000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:27:02,361 INFO [train.py:968] (0/2) Epoch 16, batch 37050, giga_loss[loss=0.2657, simple_loss=0.3375, pruned_loss=0.09698, over 29036.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3311, pruned_loss=0.09267, over 5697918.98 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3602, pruned_loss=0.1174, over 5662251.65 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3277, pruned_loss=0.09026, over 5699438.64 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:27:23,149 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=721029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:27:41,598 INFO [train.py:968] (0/2) Epoch 16, batch 37100, giga_loss[loss=0.2545, simple_loss=0.3166, pruned_loss=0.09624, over 28724.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3285, pruned_loss=0.09152, over 5703632.91 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3601, pruned_loss=0.1172, over 5667969.26 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3252, pruned_loss=0.0893, over 5700710.51 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:27:41,827 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3725, 1.0998, 1.0794, 1.5079], device='cuda:0'), covar=tensor([0.0761, 0.0385, 0.0365, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 13:27:54,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.673e+02 9.426e+02 1.142e+03 1.417e+03 3.505e+03, threshold=2.284e+03, percent-clipped=2.0
+2023-03-08 13:28:23,053 INFO [train.py:968] (0/2) Epoch 16, batch 37150, giga_loss[loss=0.2174, simple_loss=0.3023, pruned_loss=0.06623, over 28903.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3264, pruned_loss=0.09043, over 5716376.89 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.36, pruned_loss=0.117, over 5671675.81 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3235, pruned_loss=0.08847, over 5711290.38 frames. ], batch size: 174, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:28:30,408 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=721109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:28:37,055 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:29:03,242 INFO [train.py:968] (0/2) Epoch 16, batch 37200, giga_loss[loss=0.2491, simple_loss=0.323, pruned_loss=0.08764, over 28669.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3254, pruned_loss=0.08994, over 5721787.61 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3607, pruned_loss=0.1171, over 5677109.59 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3216, pruned_loss=0.08765, over 5713778.41 frames. ], batch size: 336, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:29:16,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.710e+02 1.089e+03 1.356e+03 1.916e+03 8.120e+03, threshold=2.712e+03, percent-clipped=11.0
+2023-03-08 13:29:19,706 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-08 13:29:43,693 INFO [train.py:968] (0/2) Epoch 16, batch 37250, giga_loss[loss=0.2111, simple_loss=0.2929, pruned_loss=0.0646, over 28981.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3239, pruned_loss=0.08928, over 5722665.11 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3613, pruned_loss=0.1172, over 5681720.29 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3192, pruned_loss=0.08652, over 5713574.07 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:30:02,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:30:21,831 INFO [train.py:968] (0/2) Epoch 16, batch 37300, giga_loss[loss=0.2376, simple_loss=0.3124, pruned_loss=0.08141, over 29044.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.324, pruned_loss=0.0892, over 5711104.46 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.362, pruned_loss=0.1174, over 5672912.68 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3188, pruned_loss=0.08624, over 5712190.79 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:30:22,772 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=721252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:30:25,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=721255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:30:35,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.442e+02 9.953e+02 1.151e+03 1.593e+03 4.920e+03, threshold=2.303e+03, percent-clipped=3.0
+2023-03-08 13:30:43,690 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:30:47,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=721284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:31:02,539 INFO [train.py:968] (0/2) Epoch 16, batch 37350, giga_loss[loss=0.245, simple_loss=0.3143, pruned_loss=0.08786, over 28826.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3236, pruned_loss=0.08883, over 5709852.41 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3627, pruned_loss=0.1175, over 5677851.54 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3181, pruned_loss=0.08584, over 5706869.26 frames. ], batch size: 112, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:31:44,122 INFO [train.py:968] (0/2) Epoch 16, batch 37400, giga_loss[loss=0.2446, simple_loss=0.3251, pruned_loss=0.08204, over 28886.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3249, pruned_loss=0.08979, over 5719590.98 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3629, pruned_loss=0.1174, over 5680505.29 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3197, pruned_loss=0.08694, over 5715639.33 frames. ], batch size: 227, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:31:57,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.582e+02 1.054e+03 1.399e+03 1.897e+03 5.929e+03, threshold=2.798e+03, percent-clipped=19.0
+2023-03-08 13:32:17,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0932, 3.9220, 3.6937, 1.7253], device='cuda:0'), covar=tensor([0.0580, 0.0674, 0.0655, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.1125, 0.1040, 0.0894, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 13:32:28,707 INFO [train.py:968] (0/2) Epoch 16, batch 37450, giga_loss[loss=0.254, simple_loss=0.3271, pruned_loss=0.09042, over 28889.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3303, pruned_loss=0.09331, over 5707444.80 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3634, pruned_loss=0.1177, over 5674607.87 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3252, pruned_loss=0.09043, over 5710635.08 frames. ], batch size: 106, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:33:22,012 INFO [train.py:968] (0/2) Epoch 16, batch 37500, giga_loss[loss=0.3135, simple_loss=0.3787, pruned_loss=0.1242, over 28637.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3387, pruned_loss=0.09891, over 5698600.71 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3635, pruned_loss=0.1177, over 5675799.83 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3345, pruned_loss=0.09656, over 5700147.12 frames. ], batch size: 336, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:33:31,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-08 13:33:35,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.970e+02 1.209e+03 1.578e+03 2.089e+03 1.510e+04, threshold=3.157e+03, percent-clipped=10.0
+2023-03-08 13:33:51,365 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.33 vs. limit=5.0
+2023-03-08 13:34:01,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=721494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:34:08,975 INFO [train.py:968] (0/2) Epoch 16, batch 37550, giga_loss[loss=0.2537, simple_loss=0.3362, pruned_loss=0.08562, over 28478.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.346, pruned_loss=0.1035, over 5685504.01 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3637, pruned_loss=0.1178, over 5670101.10 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3422, pruned_loss=0.1013, over 5692131.71 frames. ], batch size: 71, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:34:38,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1971, 1.7217, 1.3050, 1.0964], device='cuda:0'), covar=tensor([0.2327, 0.2347, 0.2617, 0.1947], device='cuda:0'), in_proj_covar=tensor([0.1409, 0.1029, 0.1250, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 13:34:56,374 INFO [train.py:968] (0/2) Epoch 16, batch 37600, giga_loss[loss=0.3357, simple_loss=0.3934, pruned_loss=0.139, over 28719.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3498, pruned_loss=0.1047, over 5682308.62 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3635, pruned_loss=0.1178, over 5673717.61 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3466, pruned_loss=0.1027, over 5684380.00 frames. ], batch size: 92, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:35:11,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.369e+02 1.098e+03 1.322e+03 1.685e+03 2.806e+03, threshold=2.644e+03, percent-clipped=0.0
+2023-03-08 13:35:43,045 INFO [train.py:968] (0/2) Epoch 16, batch 37650, giga_loss[loss=0.3045, simple_loss=0.3813, pruned_loss=0.1138, over 28993.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3547, pruned_loss=0.1065, over 5684669.15 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3631, pruned_loss=0.1175, over 5677325.33 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3524, pruned_loss=0.105, over 5683230.38 frames. ], batch size: 227, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:35:43,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=721601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:36:09,368 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-08 13:36:14,450 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=721637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:36:16,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=721640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:36:27,000 INFO [train.py:968] (0/2) Epoch 16, batch 37700, giga_loss[loss=0.2552, simple_loss=0.3387, pruned_loss=0.08584, over 28870.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.36, pruned_loss=0.1096, over 5680248.05 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3633, pruned_loss=0.1175, over 5671503.82 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3579, pruned_loss=0.1082, over 5684920.86 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:36:30,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=721654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:36:30,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8726, 3.7047, 3.4781, 1.6901], device='cuda:0'), covar=tensor([0.0691, 0.0755, 0.0729, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1122, 0.1037, 0.0891, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 13:36:39,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.772e+02 1.138e+03 1.446e+03 2.057e+03 4.360e+03, threshold=2.891e+03, percent-clipped=16.0
+2023-03-08 13:36:40,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=721669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:37:06,809 INFO [train.py:968] (0/2) Epoch 16, batch 37750, giga_loss[loss=0.2678, simple_loss=0.3548, pruned_loss=0.0904, over 29074.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3563, pruned_loss=0.1067, over 5675035.33 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3635, pruned_loss=0.1177, over 5667547.65 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3542, pruned_loss=0.1052, over 5683136.78 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:37:33,268 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:37:41,542 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=721744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:37:43,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=721747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:37:48,345 INFO [train.py:968] (0/2) Epoch 16, batch 37800, giga_loss[loss=0.2582, simple_loss=0.342, pruned_loss=0.08716, over 28625.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3539, pruned_loss=0.1045, over 5684579.21 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3644, pruned_loss=0.1184, over 5669905.52 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3513, pruned_loss=0.1024, over 5689082.14 frames. ], batch size: 242, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:37:53,571 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 13:37:55,226 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:37:57,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1195, 1.2742, 3.9210, 3.2634], device='cuda:0'), covar=tensor([0.2318, 0.3252, 0.0799, 0.1002], device='cuda:0'), in_proj_covar=tensor([0.0708, 0.0617, 0.0900, 0.0825], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 13:38:00,562 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.403e+02 1.236e+03 1.523e+03 2.082e+03 5.278e+03, threshold=3.047e+03, percent-clipped=11.0
+2023-03-08 13:38:09,224 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=721776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:38:27,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=721797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:38:30,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=721800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:38:30,667 INFO [train.py:968] (0/2) Epoch 16, batch 37850, giga_loss[loss=0.2419, simple_loss=0.3272, pruned_loss=0.07831, over 28848.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3526, pruned_loss=0.1031, over 5682303.91 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3649, pruned_loss=0.1186, over 5664397.82 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3498, pruned_loss=0.1009, over 5691221.72 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:38:47,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:38:54,824 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=721829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:39:12,726 INFO [train.py:968] (0/2) Epoch 16, batch 37900, giga_loss[loss=0.2486, simple_loss=0.3359, pruned_loss=0.08065, over 28737.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3537, pruned_loss=0.1035, over 5681219.88 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3649, pruned_loss=0.1187, over 5659210.26 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3513, pruned_loss=0.1014, over 5693725.59 frames. ], batch size: 284, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:39:25,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.084e+02 1.202e+03 1.581e+03 2.426e+03 5.588e+03, threshold=3.161e+03, percent-clipped=13.0
+2023-03-08 13:39:56,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:39:56,412 INFO [train.py:968] (0/2) Epoch 16, batch 37950, giga_loss[loss=0.2764, simple_loss=0.3564, pruned_loss=0.09816, over 28582.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3568, pruned_loss=0.1054, over 5680143.80 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3654, pruned_loss=0.1191, over 5647569.16 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3544, pruned_loss=0.1033, over 5699622.82 frames. ], batch size: 307, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:40:01,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7252, 1.8032, 1.3050, 1.2864], device='cuda:0'), covar=tensor([0.0798, 0.0529, 0.0954, 0.1097], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0440, 0.0512, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 13:40:04,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 13:40:24,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6163, 1.8195, 1.8822, 1.4448], device='cuda:0'), covar=tensor([0.1742, 0.2292, 0.1379, 0.1590], device='cuda:0'), in_proj_covar=tensor([0.0867, 0.0689, 0.0914, 0.0814], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 13:40:35,316 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9456, 1.1374, 1.2627, 0.9910], device='cuda:0'), covar=tensor([0.1579, 0.1310, 0.1987, 0.1474], device='cuda:0'), in_proj_covar=tensor([0.0454, 0.0734, 0.0691, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 13:40:41,179 INFO [train.py:968] (0/2) Epoch 16, batch 38000, giga_loss[loss=0.3014, simple_loss=0.3809, pruned_loss=0.111, over 28952.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3591, pruned_loss=0.1076, over 5682011.94 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3654, pruned_loss=0.1191, over 5652680.72 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.357, pruned_loss=0.1056, over 5693532.21 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:40:41,529 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:40:56,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.262e+02 1.271e+03 1.544e+03 2.175e+03 6.473e+03, threshold=3.087e+03, percent-clipped=4.0
+2023-03-08 13:41:23,758 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-722000.pt
+2023-03-08 13:41:25,344 INFO [train.py:968] (0/2) Epoch 16, batch 38050, giga_loss[loss=0.3154, simple_loss=0.3863, pruned_loss=0.1223, over 29015.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3597, pruned_loss=0.1085, over 5691907.83 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3658, pruned_loss=0.1194, over 5660590.48 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3575, pruned_loss=0.1063, over 5694787.69 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:42:09,243 INFO [train.py:968] (0/2) Epoch 16, batch 38100, giga_loss[loss=0.2707, simple_loss=0.3504, pruned_loss=0.09546, over 28949.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3601, pruned_loss=0.1093, over 5684353.13 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.366, pruned_loss=0.1194, over 5654717.32 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3581, pruned_loss=0.1074, over 5693490.90 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:42:13,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5341, 1.6839, 1.7686, 1.3476], device='cuda:0'), covar=tensor([0.1703, 0.2433, 0.1331, 0.1599], device='cuda:0'), in_proj_covar=tensor([0.0868, 0.0691, 0.0915, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 13:42:23,402 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.095e+02 1.253e+03 1.669e+03 2.513e+03 6.315e+03, threshold=3.338e+03, percent-clipped=13.0
+2023-03-08 13:42:52,057 INFO [train.py:968] (0/2) Epoch 16, batch 38150, giga_loss[loss=0.3514, simple_loss=0.4077, pruned_loss=0.1475, over 27637.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3593, pruned_loss=0.1087, over 5675718.32 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3656, pruned_loss=0.1191, over 5648720.12 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3579, pruned_loss=0.1073, over 5689626.98 frames. ], batch size: 472, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:42:58,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:43:03,074 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-08 13:43:18,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:43:32,630 INFO [train.py:968] (0/2) Epoch 16, batch 38200, giga_loss[loss=0.2741, simple_loss=0.3518, pruned_loss=0.09822, over 28822.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3593, pruned_loss=0.1076, over 5676869.09 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3664, pruned_loss=0.1197, over 5640359.59 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3574, pruned_loss=0.1058, over 5695623.32 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:43:46,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.707e+02 1.173e+03 1.637e+03 2.320e+03 8.750e+03, threshold=3.275e+03, percent-clipped=13.0
+2023-03-08 13:43:46,620 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-08 13:44:08,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:44:12,493 INFO [train.py:968] (0/2) Epoch 16, batch 38250, giga_loss[loss=0.2605, simple_loss=0.3483, pruned_loss=0.08634, over 29096.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3591, pruned_loss=0.1068, over 5688647.70 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3663, pruned_loss=0.1195, over 5647912.53 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3574, pruned_loss=0.105, over 5698115.57 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:44:18,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0340, 1.0351, 3.8839, 3.1166], device='cuda:0'), covar=tensor([0.1853, 0.3000, 0.0409, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0704, 0.0613, 0.0898, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 13:44:22,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:44:33,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4118, 3.4890, 1.6249, 1.5764], device='cuda:0'), covar=tensor([0.0998, 0.0254, 0.0893, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0520, 0.0357, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 13:44:55,649 INFO [train.py:968] (0/2) Epoch 16, batch 38300, giga_loss[loss=0.2746, simple_loss=0.3471, pruned_loss=0.1011, over 28918.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3587, pruned_loss=0.1061, over 5696201.36 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3659, pruned_loss=0.1193, over 5650632.21 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3576, pruned_loss=0.1049, over 5701405.22 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:44:56,065 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:44:58,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:45:12,206 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.555e+02 1.139e+03 1.388e+03 1.990e+03 5.947e+03, threshold=2.776e+03, percent-clipped=6.0
+2023-03-08 13:45:12,415 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722269.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:45:17,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:45:18,584 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:45:20,576 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:45:22,993 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:45:32,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5637, 1.7152, 1.4148, 1.7183], device='cuda:0'), covar=tensor([0.2505, 0.2476, 0.2739, 0.2186], device='cuda:0'), in_proj_covar=tensor([0.1407, 0.1027, 0.1246, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 13:45:37,766 INFO [train.py:968] (0/2) Epoch 16, batch 38350, giga_loss[loss=0.2653, simple_loss=0.3443, pruned_loss=0.09316, over 29085.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3564, pruned_loss=0.1053, over 5690285.33 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3659, pruned_loss=0.1194, over 5644770.15 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3555, pruned_loss=0.104, over 5700827.32 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:45:45,432 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:45:51,586 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 13:46:00,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:10,890 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:12,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:20,070 INFO [train.py:968] (0/2) Epoch 16, batch 38400, giga_loss[loss=0.2408, simple_loss=0.3258, pruned_loss=0.0779, over 28967.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3547, pruned_loss=0.1046, over 5693313.88 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3667, pruned_loss=0.1199, over 5648751.50 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.353, pruned_loss=0.1027, over 5699275.75 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:46:34,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.119e+02 1.081e+03 1.399e+03 2.010e+03 7.845e+03, threshold=2.798e+03, percent-clipped=13.0
+2023-03-08 13:46:34,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:35,707 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:53,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0502, 1.9988, 1.8687, 1.6639], device='cuda:0'), covar=tensor([0.1736, 0.2420, 0.2193, 0.2324], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0732, 0.0691, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 13:46:59,325 INFO [train.py:968] (0/2) Epoch 16, batch 38450, giga_loss[loss=0.3035, simple_loss=0.363, pruned_loss=0.122, over 27635.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3532, pruned_loss=0.1039, over 5694062.30 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3663, pruned_loss=0.1197, over 5644951.82 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3518, pruned_loss=0.1021, over 5702979.45 frames. ], batch size: 472, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:47:14,820 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:47:15,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1464, 2.1998, 2.1788, 1.8470], device='cuda:0'), covar=tensor([0.1770, 0.2204, 0.1898, 0.2094], device='cuda:0'), in_proj_covar=tensor([0.0454, 0.0732, 0.0691, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 13:47:17,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:47:17,929 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6625, 1.8486, 1.8410, 1.6231], device='cuda:0'), covar=tensor([0.1885, 0.1629, 0.1136, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.1830, 0.1759, 0.1674, 0.1836], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 13:47:42,105 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:47:42,642 INFO [train.py:968] (0/2) Epoch 16, batch 38500, giga_loss[loss=0.264, simple_loss=0.3458, pruned_loss=0.09111, over 29010.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3533, pruned_loss=0.104, over 5696498.66 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3663, pruned_loss=0.1196, over 5648265.89 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.352, pruned_loss=0.1024, over 5701607.03 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:47:53,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5981, 2.3141, 1.7635, 0.7917], device='cuda:0'), covar=tensor([0.5445, 0.2539, 0.3816, 0.5596], device='cuda:0'), in_proj_covar=tensor([0.1648, 0.1550, 0.1537, 0.1344], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 13:47:57,805 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.349e+02 1.028e+03 1.264e+03 1.540e+03 4.556e+03, threshold=2.529e+03, percent-clipped=3.0
+2023-03-08 13:47:58,109 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:48:00,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:48:23,434 INFO [train.py:968] (0/2) Epoch 16, batch 38550, giga_loss[loss=0.2773, simple_loss=0.352, pruned_loss=0.1013, over 28925.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.353, pruned_loss=0.1036, over 5698553.05 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3663, pruned_loss=0.1195, over 5649623.85 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3517, pruned_loss=0.1022, over 5701988.44 frames. ], batch size: 199, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:48:23,678 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:48:24,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9100, 1.0877, 1.0916, 0.8797], device='cuda:0'), covar=tensor([0.2053, 0.2357, 0.1326, 0.1911], device='cuda:0'), in_proj_covar=tensor([0.1841, 0.1770, 0.1686, 0.1846], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 13:48:38,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:48:47,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9550, 1.2458, 1.2926, 1.0895], device='cuda:0'), covar=tensor([0.1942, 0.1367, 0.2331, 0.1642], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0737, 0.0695, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 13:49:00,327 INFO [train.py:968] (0/2) Epoch 16, batch 38600, giga_loss[loss=0.2665, simple_loss=0.3542, pruned_loss=0.08938, over 28382.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3532, pruned_loss=0.1033, over 5694980.12 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3663, pruned_loss=0.1197, over 5643892.14 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3519, pruned_loss=0.1017, over 5704021.81 frames. ], batch size: 65, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:49:11,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5457, 5.3552, 5.0256, 2.5726], device='cuda:0'), covar=tensor([0.0356, 0.0500, 0.0558, 0.1796], device='cuda:0'), in_proj_covar=tensor([0.1128, 0.1045, 0.0897, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 13:49:11,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0499, 2.3375, 2.1698, 1.7646], device='cuda:0'), covar=tensor([0.2654, 0.2020, 0.1866, 0.2313], device='cuda:0'), in_proj_covar=tensor([0.1844, 0.1773, 0.1687, 0.1845], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 13:49:15,263 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.357e+02 1.007e+03 1.287e+03 1.718e+03 4.825e+03, threshold=2.575e+03, percent-clipped=6.0
+2023-03-08 13:49:27,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:49:40,157 INFO [train.py:968] (0/2) Epoch 16, batch 38650, giga_loss[loss=0.2849, simple_loss=0.3466, pruned_loss=0.1116, over 28636.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3523, pruned_loss=0.1022, over 5702537.52 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3665, pruned_loss=0.1199, over 5645439.15 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3509, pruned_loss=0.1005, over 5709432.72 frames. ], batch size: 92, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:50:17,121 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722644.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:50:21,368 INFO [train.py:968] (0/2) Epoch 16, batch 38700, giga_loss[loss=0.2461, simple_loss=0.3301, pruned_loss=0.08105, over 28438.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3521, pruned_loss=0.1024, over 5703435.00 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3664, pruned_loss=0.1197, over 5654287.64 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3506, pruned_loss=0.1007, over 5702938.45 frames. ], batch size: 71, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:50:36,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.872e+02 1.020e+03 1.330e+03 1.863e+03 7.187e+03, threshold=2.660e+03, percent-clipped=12.0
+2023-03-08 13:50:55,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:00,937 INFO [train.py:968] (0/2) Epoch 16, batch 38750, giga_loss[loss=0.2466, simple_loss=0.3231, pruned_loss=0.08508, over 28913.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3499, pruned_loss=0.1015, over 5698460.64 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.367, pruned_loss=0.1201, over 5647905.80 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.348, pruned_loss=0.09945, over 5704911.16 frames. ], batch size: 119, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:51:01,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:16,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3464, 3.1398, 1.5105, 1.4664], device='cuda:0'), covar=tensor([0.1000, 0.0258, 0.0889, 0.1345], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0518, 0.0357, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 13:51:17,525 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:20,077 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:24,251 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:26,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:33,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:39,619 INFO [train.py:968] (0/2) Epoch 16, batch 38800, giga_loss[loss=0.2915, simple_loss=0.3521, pruned_loss=0.1154, over 28626.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3485, pruned_loss=0.1012, over 5697376.68 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3675, pruned_loss=0.1203, over 5655262.16 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3457, pruned_loss=0.09863, over 5697929.43 frames. ], batch size: 78, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:51:40,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4941, 1.6911, 1.3002, 1.3168], device='cuda:0'), covar=tensor([0.0810, 0.0444, 0.0948, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0435, 0.0508, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 13:51:49,256 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:53,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2337, 1.2488, 4.3034, 3.3770], device='cuda:0'), covar=tensor([0.1730, 0.2909, 0.0383, 0.1055], device='cuda:0'), in_proj_covar=tensor([0.0706, 0.0614, 0.0900, 0.0826], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 13:51:54,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.006e+02 1.143e+03 1.456e+03 1.938e+03 7.032e+03, threshold=2.912e+03, percent-clipped=10.0
+2023-03-08 13:52:07,246 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722787.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:52:09,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722790.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:52:17,223 INFO [train.py:968] (0/2) Epoch 16, batch 38850, giga_loss[loss=0.2483, simple_loss=0.3247, pruned_loss=0.08594, over 28459.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3465, pruned_loss=0.09988, over 5710676.93 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3671, pruned_loss=0.1199, over 5662390.63 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3436, pruned_loss=0.09715, over 5706758.92 frames. ], batch size: 71, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:52:19,076 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:52:19,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7100, 4.5276, 1.8231, 1.8755], device='cuda:0'), covar=tensor([0.0934, 0.0254, 0.0877, 0.1227], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0519, 0.0357, 0.0403], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 13:52:25,542 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:52:30,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7737, 2.1208, 1.8709, 1.5903], device='cuda:0'), covar=tensor([0.2340, 0.1904, 0.2025, 0.2148], device='cuda:0'), in_proj_covar=tensor([0.1841, 0.1772, 0.1683, 0.1840], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 13:52:31,451 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722819.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:52:51,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-08 13:53:00,540 INFO [train.py:968] (0/2) Epoch 16, batch 38900, giga_loss[loss=0.3012, simple_loss=0.3689, pruned_loss=0.1168, over 28273.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3469, pruned_loss=0.1008, over 5709334.07 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3674, pruned_loss=0.12, over 5664724.94 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3442, pruned_loss=0.09842, over 5704477.55 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:53:09,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5866, 4.4042, 4.1844, 1.8870], device='cuda:0'), covar=tensor([0.0562, 0.0735, 0.0736, 0.2191], device='cuda:0'), in_proj_covar=tensor([0.1129, 0.1050, 0.0898, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 13:53:10,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6043, 1.8143, 1.4982, 1.8064], device='cuda:0'), covar=tensor([0.2492, 0.2574, 0.2896, 0.2523], device='cuda:0'), in_proj_covar=tensor([0.1411, 0.1032, 0.1250, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 13:53:14,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.726e+02 1.156e+03 1.524e+03 1.911e+03 3.443e+03, threshold=3.047e+03, percent-clipped=7.0
+2023-03-08 13:53:30,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:53:32,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:53:37,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:53:40,365 INFO [train.py:968] (0/2) Epoch 16, batch 38950, giga_loss[loss=0.2293, simple_loss=0.3152, pruned_loss=0.07174, over 28941.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3457, pruned_loss=0.1008, over 5716578.87 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.368, pruned_loss=0.1203, over 5669156.26 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3426, pruned_loss=0.09819, over 5709634.74 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:53:56,040 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:54:20,213 INFO [train.py:968] (0/2) Epoch 16, batch 39000, giga_loss[loss=0.2506, simple_loss=0.323, pruned_loss=0.0891, over 28775.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3445, pruned_loss=0.1009, over 5710680.59 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3681, pruned_loss=0.1206, over 5670510.98 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3413, pruned_loss=0.09815, over 5704961.60 frames. ], batch size: 66, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:54:20,217 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 13:54:25,713 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2273, 2.9364, 1.2980, 1.3919], device='cuda:0'), covar=tensor([0.0986, 0.0290, 0.0967, 0.1388], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0522, 0.0358, 0.0404], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 13:54:25,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0052, 3.7972, 3.7127, 1.6647], device='cuda:0'), covar=tensor([0.0648, 0.0811, 0.0721, 0.2344], device='cuda:0'), in_proj_covar=tensor([0.1129, 0.1049, 0.0899, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 13:54:28,571 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2114, simple_loss=0.3188, pruned_loss=0.05206, over 944034.00 frames. 
+2023-03-08 13:54:28,572 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 13:54:35,849 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4920, 1.6516, 1.7601, 1.3363], device='cuda:0'), covar=tensor([0.1751, 0.2356, 0.1406, 0.1613], device='cuda:0'), in_proj_covar=tensor([0.0872, 0.0697, 0.0920, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 13:54:38,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5733, 1.7499, 1.4650, 1.7223], device='cuda:0'), covar=tensor([0.2539, 0.2640, 0.2968, 0.2354], device='cuda:0'), in_proj_covar=tensor([0.1409, 0.1031, 0.1248, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 13:54:40,803 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722968.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:54:44,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.604e+02 1.095e+03 1.419e+03 2.044e+03 4.620e+03, threshold=2.838e+03, percent-clipped=6.0
+2023-03-08 13:54:50,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3973, 1.6561, 1.4339, 1.6099], device='cuda:0'), covar=tensor([0.0747, 0.0306, 0.0323, 0.0830], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 13:54:57,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9621, 1.1586, 1.3624, 0.9740], device='cuda:0'), covar=tensor([0.1724, 0.1459, 0.2153, 0.1814], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0738, 0.0696, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 13:55:06,735 INFO [train.py:968] (0/2) Epoch 16, batch 39050, giga_loss[loss=0.2204, simple_loss=0.2879, pruned_loss=0.07651, over 28827.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3405, pruned_loss=0.09889, over 5704468.18 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.368, pruned_loss=0.1205, over 5663934.54 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3378, pruned_loss=0.09654, over 5705860.90 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:55:23,526 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:55:42,290 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:55:44,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:55:50,086 INFO [train.py:968] (0/2) Epoch 16, batch 39100, giga_loss[loss=0.2376, simple_loss=0.3203, pruned_loss=0.07742, over 28688.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3386, pruned_loss=0.09777, over 5700104.36 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3684, pruned_loss=0.1207, over 5659602.07 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3354, pruned_loss=0.09524, over 5705745.78 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 1.0
+2023-03-08 13:56:04,801 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:56:08,384 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:56:08,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.027e+02 1.100e+03 1.392e+03 2.115e+03 1.091e+04, threshold=2.784e+03, percent-clipped=16.0
+2023-03-08 13:56:11,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:56:27,399 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:56:30,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:56:31,339 INFO [train.py:968] (0/2) Epoch 16, batch 39150, giga_loss[loss=0.2442, simple_loss=0.3245, pruned_loss=0.08197, over 28860.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3402, pruned_loss=0.09908, over 5710921.66 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3685, pruned_loss=0.1209, over 5669513.84 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3365, pruned_loss=0.09604, over 5707897.17 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 1.0
+2023-03-08 13:56:39,233 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723111.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:56:39,819 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:57:15,716 INFO [train.py:968] (0/2) Epoch 16, batch 39200, giga_loss[loss=0.2684, simple_loss=0.3337, pruned_loss=0.1016, over 28611.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3415, pruned_loss=0.09893, over 5710816.97 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3678, pruned_loss=0.1205, over 5673479.25 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3386, pruned_loss=0.09638, over 5705747.26 frames. ], batch size: 85, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:57:37,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.966e+02 1.047e+03 1.348e+03 1.922e+03 5.739e+03, threshold=2.697e+03, percent-clipped=10.0
+2023-03-08 13:57:40,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:57:49,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:57:52,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1989, 1.4818, 1.5847, 1.3262], device='cuda:0'), covar=tensor([0.1767, 0.1567, 0.1981, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0736, 0.0695, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 13:57:52,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4595, 2.0230, 1.6275, 1.6961], device='cuda:0'), covar=tensor([0.0729, 0.0248, 0.0306, 0.0820], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 13:57:54,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2363, 1.1638, 3.9462, 3.2623], device='cuda:0'), covar=tensor([0.1724, 0.2859, 0.0403, 0.0830], device='cuda:0'), in_proj_covar=tensor([0.0706, 0.0614, 0.0900, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 13:58:02,303 INFO [train.py:968] (0/2) Epoch 16, batch 39250, giga_loss[loss=0.2848, simple_loss=0.3602, pruned_loss=0.1047, over 27605.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3438, pruned_loss=0.09899, over 5710620.18 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.368, pruned_loss=0.1205, over 5675442.15 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3411, pruned_loss=0.09674, over 5705178.61 frames. ], batch size: 472, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:58:11,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:14,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:19,559 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:22,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:35,831 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:39,167 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:39,857 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:40,370 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:41,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:48,186 INFO [train.py:968] (0/2) Epoch 16, batch 39300, giga_loss[loss=0.2468, simple_loss=0.3361, pruned_loss=0.07873, over 28809.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.345, pruned_loss=0.0984, over 5705520.71 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.368, pruned_loss=0.1205, over 5676358.94 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3428, pruned_loss=0.09654, over 5700561.58 frames. ], batch size: 284, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:58:49,016 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:02,505 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:04,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.849e+02 1.031e+03 1.363e+03 1.930e+03 5.239e+03, threshold=2.726e+03, percent-clipped=10.0
+2023-03-08 13:59:05,696 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:29,991 INFO [train.py:968] (0/2) Epoch 16, batch 39350, giga_loss[loss=0.273, simple_loss=0.3487, pruned_loss=0.0987, over 28876.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3448, pruned_loss=0.09789, over 5687851.17 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.368, pruned_loss=0.1205, over 5669683.64 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3426, pruned_loss=0.09604, over 5690141.05 frames. ], batch size: 227, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:59:37,511 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:45,724 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:47,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:50,634 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:52,573 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:00:01,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723343.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:00:07,812 INFO [train.py:968] (0/2) Epoch 16, batch 39400, libri_loss[loss=0.3221, simple_loss=0.3851, pruned_loss=0.1295, over 29379.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3446, pruned_loss=0.09747, over 5696894.27 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.368, pruned_loss=0.1204, over 5672073.16 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3422, pruned_loss=0.09532, over 5696799.31 frames. ], batch size: 92, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:00:08,700 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:00:09,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8925, 1.1477, 1.0556, 0.8007], device='cuda:0'), covar=tensor([0.1939, 0.1913, 0.1228, 0.1867], device='cuda:0'), in_proj_covar=tensor([0.1846, 0.1776, 0.1687, 0.1840], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 14:00:14,605 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:00:24,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.487e+02 1.178e+03 1.485e+03 1.941e+03 6.708e+03, threshold=2.969e+03, percent-clipped=9.0
+2023-03-08 14:00:34,921 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:00:40,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:00:45,051 INFO [train.py:968] (0/2) Epoch 16, batch 39450, giga_loss[loss=0.2613, simple_loss=0.3401, pruned_loss=0.09128, over 29017.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3469, pruned_loss=0.09941, over 5695290.91 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3679, pruned_loss=0.1203, over 5679091.99 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3441, pruned_loss=0.097, over 5689896.65 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:00:55,347 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723414.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:01:27,008 INFO [train.py:968] (0/2) Epoch 16, batch 39500, giga_loss[loss=0.2892, simple_loss=0.3587, pruned_loss=0.1099, over 27997.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3488, pruned_loss=0.1011, over 5694267.25 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3682, pruned_loss=0.1204, over 5682433.67 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3461, pruned_loss=0.09879, over 5687244.18 frames. ], batch size: 412, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:01:27,220 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:01:47,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.909e+02 1.247e+03 1.455e+03 1.987e+03 4.932e+03, threshold=2.910e+03, percent-clipped=6.0
+2023-03-08 14:01:58,675 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723486.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:01:58,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723486.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:01:59,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:02:00,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723489.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:02:11,422 INFO [train.py:968] (0/2) Epoch 16, batch 39550, giga_loss[loss=0.2913, simple_loss=0.3563, pruned_loss=0.1132, over 28744.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3525, pruned_loss=0.1033, over 5703362.44 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3681, pruned_loss=0.1204, over 5684340.60 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3502, pruned_loss=0.1012, over 5696189.27 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:02:25,836 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723518.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:02:25,874 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2666, 1.5510, 1.3934, 1.3597], device='cuda:0'), covar=tensor([0.1494, 0.1647, 0.1822, 0.1762], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0740, 0.0698, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 14:02:41,388 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:02:44,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:02:52,619 INFO [train.py:968] (0/2) Epoch 16, batch 39600, libri_loss[loss=0.2966, simple_loss=0.3681, pruned_loss=0.1126, over 29517.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3547, pruned_loss=0.1044, over 5705282.74 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3686, pruned_loss=0.1206, over 5682889.97 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3523, pruned_loss=0.1023, over 5701380.65 frames. ], batch size: 83, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:03:08,191 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:03:09,990 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.840e+02 1.257e+03 1.613e+03 2.252e+03 5.805e+03, threshold=3.227e+03, percent-clipped=12.0
+2023-03-08 14:03:30,859 INFO [train.py:968] (0/2) Epoch 16, batch 39650, giga_loss[loss=0.2859, simple_loss=0.3611, pruned_loss=0.1054, over 28814.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3558, pruned_loss=0.1044, over 5715071.13 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.121, over 5687807.05 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.353, pruned_loss=0.1021, over 5708064.69 frames. ], batch size: 119, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:03:38,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:03:39,398 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-08 14:03:52,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723629.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:03:52,956 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:03:54,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723632.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:03:54,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:04:09,081 INFO [train.py:968] (0/2) Epoch 16, batch 39700, giga_loss[loss=0.2595, simple_loss=0.344, pruned_loss=0.08753, over 28984.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3565, pruned_loss=0.1051, over 5720708.25 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3693, pruned_loss=0.121, over 5692742.26 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3537, pruned_loss=0.1025, over 5711351.26 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:04:15,624 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723661.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:04:17,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:04:26,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.464e+02 1.270e+03 1.721e+03 2.381e+03 1.050e+04, threshold=3.442e+03, percent-clipped=11.0
+2023-03-08 14:04:34,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:04:48,332 INFO [train.py:968] (0/2) Epoch 16, batch 39750, giga_loss[loss=0.2732, simple_loss=0.3412, pruned_loss=0.1026, over 28934.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3565, pruned_loss=0.1052, over 5718348.00 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1212, over 5692521.23 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3538, pruned_loss=0.1028, over 5711206.62 frames. ], batch size: 112, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:05:25,583 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:05:29,503 INFO [train.py:968] (0/2) Epoch 16, batch 39800, giga_loss[loss=0.2573, simple_loss=0.3416, pruned_loss=0.08654, over 28990.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3543, pruned_loss=0.1034, over 5717729.87 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3699, pruned_loss=0.1214, over 5693830.62 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3518, pruned_loss=0.1013, over 5711008.05 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:05:38,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:05:43,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5537, 1.7415, 1.6260, 1.4378], device='cuda:0'), covar=tensor([0.2861, 0.2301, 0.1635, 0.2290], device='cuda:0'), in_proj_covar=tensor([0.1854, 0.1789, 0.1701, 0.1848], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 14:05:44,612 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6455, 1.7479, 1.2421, 1.2998], device='cuda:0'), covar=tensor([0.0864, 0.0643, 0.1143, 0.1226], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0438, 0.0509, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 14:05:45,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6495, 1.1618, 4.6506, 3.5004], device='cuda:0'), covar=tensor([0.1503, 0.2922, 0.0359, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0706, 0.0614, 0.0901, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 14:05:47,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.585e+02 1.102e+03 1.300e+03 1.771e+03 7.568e+03, threshold=2.599e+03, percent-clipped=3.0
+2023-03-08 14:05:48,221 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:05:59,427 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:06:02,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723789.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:06:11,705 INFO [train.py:968] (0/2) Epoch 16, batch 39850, giga_loss[loss=0.2317, simple_loss=0.3083, pruned_loss=0.07762, over 28493.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3519, pruned_loss=0.1025, over 5713333.69 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3703, pruned_loss=0.1217, over 5695000.18 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3495, pruned_loss=0.1004, over 5707318.04 frames. ], batch size: 71, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:06:32,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:06:33,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:06:35,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:06:53,552 INFO [train.py:968] (0/2) Epoch 16, batch 39900, giga_loss[loss=0.2208, simple_loss=0.3011, pruned_loss=0.0702, over 28552.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3484, pruned_loss=0.1005, over 5710962.70 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3707, pruned_loss=0.1219, over 5686246.44 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3455, pruned_loss=0.09812, over 5714895.72 frames. ], batch size: 85, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:06:59,062 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:07:05,456 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 14:07:09,365 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.899e+02 1.116e+03 1.408e+03 1.982e+03 5.324e+03, threshold=2.816e+03, percent-clipped=12.0
+2023-03-08 14:07:30,192 INFO [train.py:968] (0/2) Epoch 16, batch 39950, giga_loss[loss=0.2955, simple_loss=0.3731, pruned_loss=0.1089, over 28832.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3492, pruned_loss=0.1013, over 5713942.96 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3711, pruned_loss=0.1221, over 5696166.56 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3454, pruned_loss=0.09804, over 5709194.81 frames. ], batch size: 174, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:07:31,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6200, 2.4133, 1.6771, 0.8942], device='cuda:0'), covar=tensor([0.4769, 0.2693, 0.3419, 0.4572], device='cuda:0'), in_proj_covar=tensor([0.1642, 0.1548, 0.1536, 0.1343], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 14:07:33,102 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:07:35,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:07:53,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5273, 5.2064, 1.8586, 2.0263], device='cuda:0'), covar=tensor([0.0894, 0.0332, 0.0848, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0526, 0.0361, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 14:07:55,541 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723932.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:07:57,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723935.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:07:58,667 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:08:11,780 INFO [train.py:968] (0/2) Epoch 16, batch 40000, giga_loss[loss=0.2706, simple_loss=0.3388, pruned_loss=0.1012, over 28485.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3498, pruned_loss=0.1001, over 5703256.42 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3712, pruned_loss=0.1223, over 5688859.09 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3461, pruned_loss=0.09689, over 5706628.02 frames. ], batch size: 71, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:08:21,828 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723964.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:08:25,239 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:08:27,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:08:28,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.489e+02 1.179e+03 1.482e+03 2.217e+03 1.027e+04, threshold=2.965e+03, percent-clipped=17.0
+2023-03-08 14:08:31,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5609, 1.7397, 1.7203, 1.5730], device='cuda:0'), covar=tensor([0.1676, 0.2174, 0.2210, 0.2205], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0741, 0.0698, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 14:08:39,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:08:49,122 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-724000.pt
+2023-03-08 14:08:50,032 INFO [train.py:968] (0/2) Epoch 16, batch 40050, giga_loss[loss=0.2372, simple_loss=0.3194, pruned_loss=0.07752, over 28805.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.351, pruned_loss=0.1007, over 5698583.32 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3707, pruned_loss=0.1221, over 5687269.98 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3478, pruned_loss=0.09749, over 5703387.44 frames. ], batch size: 174, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:08:50,745 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:09:30,389 INFO [train.py:968] (0/2) Epoch 16, batch 40100, giga_loss[loss=0.2593, simple_loss=0.33, pruned_loss=0.09435, over 29036.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3494, pruned_loss=0.1006, over 5705785.09 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3708, pruned_loss=0.1221, over 5687010.00 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3465, pruned_loss=0.09759, over 5710105.48 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:09:30,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=724051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:09:47,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.559e+02 1.180e+03 1.608e+03 2.254e+03 5.660e+03, threshold=3.217e+03, percent-clipped=9.0
+2023-03-08 14:10:01,930 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4925, 1.7089, 1.7305, 1.2984], device='cuda:0'), covar=tensor([0.1757, 0.2348, 0.1489, 0.1675], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0693, 0.0913, 0.0814], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 14:10:08,445 INFO [train.py:968] (0/2) Epoch 16, batch 40150, giga_loss[loss=0.362, simple_loss=0.4014, pruned_loss=0.1613, over 26796.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3481, pruned_loss=0.1011, over 5713332.86 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3702, pruned_loss=0.1218, over 5691582.95 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3453, pruned_loss=0.09824, over 5713265.10 frames. ], batch size: 555, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:10:26,577 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=724122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:10:32,176 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=724129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:10:35,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=724132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:10:51,830 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=724149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:10:52,817 INFO [train.py:968] (0/2) Epoch 16, batch 40200, libri_loss[loss=0.2993, simple_loss=0.3706, pruned_loss=0.114, over 29541.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3462, pruned_loss=0.1016, over 5713104.09 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3704, pruned_loss=0.1218, over 5694759.48 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3435, pruned_loss=0.09899, over 5710431.38 frames. ], batch size: 89, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:11:01,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=724160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:11:01,637 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:11:10,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.619e+02 1.169e+03 1.592e+03 1.991e+03 4.753e+03, threshold=3.185e+03, percent-clipped=9.0
+2023-03-08 14:11:35,969 INFO [train.py:968] (0/2) Epoch 16, batch 40250, giga_loss[loss=0.3534, simple_loss=0.4001, pruned_loss=0.1533, over 26765.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3434, pruned_loss=0.1011, over 5706184.59 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3704, pruned_loss=0.1218, over 5696034.68 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3413, pruned_loss=0.099, over 5703088.21 frames. ], batch size: 555, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:11:58,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3643, 1.4118, 3.6104, 3.0730], device='cuda:0'), covar=tensor([0.1514, 0.2631, 0.0423, 0.1290], device='cuda:0'), in_proj_covar=tensor([0.0708, 0.0616, 0.0905, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 14:12:16,540 INFO [train.py:968] (0/2) Epoch 16, batch 40300, giga_loss[loss=0.2232, simple_loss=0.3113, pruned_loss=0.06757, over 28942.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3417, pruned_loss=0.09967, over 5708410.25 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3708, pruned_loss=0.1219, over 5701559.57 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3389, pruned_loss=0.09733, over 5701176.41 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:12:28,125 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=724265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:12:30,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=724268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:12:34,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.056e+02 1.141e+03 1.405e+03 1.971e+03 4.110e+03, threshold=2.810e+03, percent-clipped=4.0
+2023-03-08 14:12:48,553 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=724292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:12:50,287 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=724295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:12:51,766 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:12:55,017 INFO [train.py:968] (0/2) Epoch 16, batch 40350, giga_loss[loss=0.2562, simple_loss=0.3292, pruned_loss=0.09166, over 28571.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3377, pruned_loss=0.09739, over 5716113.87 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.371, pruned_loss=0.1219, over 5705383.89 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3348, pruned_loss=0.09507, over 5707148.97 frames. ], batch size: 336, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:12:56,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=724303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:13:00,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=724306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:13:00,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8520, 1.9800, 1.6816, 2.0776], device='cuda:0'), covar=tensor([0.2444, 0.2626, 0.2912, 0.2484], device='cuda:0'), in_proj_covar=tensor([0.1413, 0.1026, 0.1245, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 14:13:15,418 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:13:23,223 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:13:35,878 INFO [train.py:968] (0/2) Epoch 16, batch 40400, giga_loss[loss=0.241, simple_loss=0.3135, pruned_loss=0.08429, over 28811.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3324, pruned_loss=0.09451, over 5708606.58 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3706, pruned_loss=0.1217, over 5700743.96 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3295, pruned_loss=0.09225, over 5705408.85 frames. ], batch size: 119, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:13:47,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3696, 1.6508, 1.4883, 1.5157], device='cuda:0'), covar=tensor([0.0724, 0.0318, 0.0318, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0214], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 14:13:54,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.768e+02 1.092e+03 1.431e+03 2.282e+03 6.550e+03, threshold=2.861e+03, percent-clipped=17.0
+2023-03-08 14:14:16,425 INFO [train.py:968] (0/2) Epoch 16, batch 40450, giga_loss[loss=0.2284, simple_loss=0.306, pruned_loss=0.07537, over 28859.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3308, pruned_loss=0.09321, over 5713523.33 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3709, pruned_loss=0.1219, over 5701044.56 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3275, pruned_loss=0.09075, over 5710980.94 frames. ], batch size: 112, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:14:20,887 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 14:14:32,061 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-08 14:14:38,786 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=724426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:15:01,454 INFO [train.py:968] (0/2) Epoch 16, batch 40500, giga_loss[loss=0.303, simple_loss=0.3595, pruned_loss=0.1232, over 23873.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3335, pruned_loss=0.09452, over 5701161.77 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3711, pruned_loss=0.1221, over 5692498.34 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3307, pruned_loss=0.09237, over 5706956.81 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:15:19,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.667e+02 1.186e+03 1.501e+03 2.101e+03 5.748e+03, threshold=3.001e+03, percent-clipped=12.0
+2023-03-08 14:15:40,666 INFO [train.py:968] (0/2) Epoch 16, batch 40550, giga_loss[loss=0.2765, simple_loss=0.3452, pruned_loss=0.1039, over 28684.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3371, pruned_loss=0.0959, over 5692690.75 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.371, pruned_loss=0.122, over 5681588.10 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3337, pruned_loss=0.09333, over 5707972.71 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:15:47,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=724509.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:16:21,382 INFO [train.py:968] (0/2) Epoch 16, batch 40600, giga_loss[loss=0.271, simple_loss=0.3495, pruned_loss=0.0962, over 28885.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3403, pruned_loss=0.09711, over 5697847.63 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.371, pruned_loss=0.122, over 5683883.45 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3373, pruned_loss=0.09485, over 5707931.66 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:16:34,389 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-08 14:16:38,773 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=724569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:16:43,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=724572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:16:44,991 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.389e+02 1.174e+03 1.591e+03 2.217e+03 5.871e+03, threshold=3.181e+03, percent-clipped=11.0
+2023-03-08 14:16:51,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2690, 1.7518, 1.2606, 0.6607], device='cuda:0'), covar=tensor([0.5498, 0.2637, 0.2925, 0.5960], device='cuda:0'), in_proj_covar=tensor([0.1653, 0.1557, 0.1543, 0.1353], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 14:17:06,347 INFO [train.py:968] (0/2) Epoch 16, batch 40650, giga_loss[loss=0.3114, simple_loss=0.3764, pruned_loss=0.1232, over 28841.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3443, pruned_loss=0.09919, over 5695942.20 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3708, pruned_loss=0.122, over 5686848.81 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3418, pruned_loss=0.09708, over 5701268.49 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:17:06,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:17:49,981 INFO [train.py:968] (0/2) Epoch 16, batch 40700, giga_loss[loss=0.2775, simple_loss=0.3412, pruned_loss=0.1069, over 28548.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3473, pruned_loss=0.1008, over 5701088.72 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3711, pruned_loss=0.1222, over 5688249.86 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3448, pruned_loss=0.09873, over 5704094.30 frames. ], batch size: 85, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:18:12,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.294e+02 1.163e+03 1.433e+03 2.093e+03 4.774e+03, threshold=2.865e+03, percent-clipped=6.0
+2023-03-08 14:18:40,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-08 14:18:41,885 INFO [train.py:968] (0/2) Epoch 16, batch 40750, giga_loss[loss=0.2593, simple_loss=0.3319, pruned_loss=0.09336, over 28883.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3537, pruned_loss=0.1065, over 5701081.87 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3714, pruned_loss=0.1225, over 5691516.03 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3511, pruned_loss=0.1043, over 5700601.43 frames. ], batch size: 106, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:19:29,302 INFO [train.py:968] (0/2) Epoch 16, batch 40800, giga_loss[loss=0.322, simple_loss=0.3903, pruned_loss=0.1268, over 29030.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3593, pruned_loss=0.1109, over 5692431.30 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3705, pruned_loss=0.1219, over 5687100.39 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3577, pruned_loss=0.1094, over 5696119.89 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:19:44,543 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-08 14:19:55,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.126e+03 1.634e+03 2.150e+03 2.962e+03 1.124e+04, threshold=4.299e+03, percent-clipped=27.0
+2023-03-08 14:20:14,515 INFO [train.py:968] (0/2) Epoch 16, batch 40850, giga_loss[loss=0.3449, simple_loss=0.4062, pruned_loss=0.1418, over 28893.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.367, pruned_loss=0.1163, over 5688671.52 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3712, pruned_loss=0.1224, over 5683482.92 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3648, pruned_loss=0.1144, over 5695001.74 frames. ], batch size: 174, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:20:55,596 INFO [train.py:968] (0/2) Epoch 16, batch 40900, giga_loss[loss=0.3631, simple_loss=0.4109, pruned_loss=0.1577, over 28961.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3718, pruned_loss=0.1204, over 5697256.52 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3707, pruned_loss=0.122, over 5690534.13 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3705, pruned_loss=0.1191, over 5696191.64 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:21:18,810 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.42 vs. limit=5.0
+2023-03-08 14:21:22,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.985e+02 1.668e+03 2.192e+03 2.764e+03 6.731e+03, threshold=4.385e+03, percent-clipped=4.0
+2023-03-08 14:21:28,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=724884.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:21:44,207 INFO [train.py:968] (0/2) Epoch 16, batch 40950, giga_loss[loss=0.3457, simple_loss=0.4044, pruned_loss=0.1434, over 28868.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3801, pruned_loss=0.1273, over 5694228.29 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3709, pruned_loss=0.122, over 5692689.50 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.379, pruned_loss=0.1264, over 5691466.43 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:21:57,180 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=724913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:22:24,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8632, 2.3156, 1.9460, 1.5469], device='cuda:0'), covar=tensor([0.2498, 0.1990, 0.2276, 0.2506], device='cuda:0'), in_proj_covar=tensor([0.1845, 0.1783, 0.1697, 0.1838], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 14:22:38,530 INFO [train.py:968] (0/2) Epoch 16, batch 41000, giga_loss[loss=0.2979, simple_loss=0.3737, pruned_loss=0.1111, over 28950.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3838, pruned_loss=0.1314, over 5653565.68 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3712, pruned_loss=0.1222, over 5674883.61 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3829, pruned_loss=0.1305, over 5666995.11 frames. ], batch size: 174, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:23:05,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.825e+03 2.305e+03 3.015e+03 1.096e+04, threshold=4.610e+03, percent-clipped=12.0
+2023-03-08 14:23:32,912 INFO [train.py:968] (0/2) Epoch 16, batch 41050, giga_loss[loss=0.2787, simple_loss=0.3507, pruned_loss=0.1033, over 29002.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3856, pruned_loss=0.1337, over 5655481.08 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3711, pruned_loss=0.1221, over 5678931.56 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3852, pruned_loss=0.1333, over 5662020.97 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:23:34,433 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=725003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:23:57,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4228, 1.6829, 1.3068, 1.5890], device='cuda:0'), covar=tensor([0.0711, 0.0347, 0.0334, 0.0777], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:0')
+2023-03-08 14:24:00,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=725027.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:24:02,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4387, 1.9300, 1.4236, 0.7206], device='cuda:0'), covar=tensor([0.4331, 0.2960, 0.2775, 0.5161], device='cuda:0'), in_proj_covar=tensor([0.1651, 0.1561, 0.1548, 0.1352], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 14:24:03,011 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=725030.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:24:26,008 INFO [train.py:968] (0/2) Epoch 16, batch 41100, giga_loss[loss=0.2974, simple_loss=0.3646, pruned_loss=0.1151, over 28930.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3896, pruned_loss=0.1384, over 5643100.30 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3713, pruned_loss=0.1221, over 5681981.49 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3894, pruned_loss=0.1383, over 5645394.08 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:24:33,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=725059.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:24:51,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.725e+03 2.280e+03 3.159e+03 9.694e+03, threshold=4.560e+03, percent-clipped=11.0
+2023-03-08 14:25:16,795 INFO [train.py:968] (0/2) Epoch 16, batch 41150, giga_loss[loss=0.3724, simple_loss=0.4189, pruned_loss=0.163, over 28731.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3924, pruned_loss=0.1416, over 5639516.22 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3709, pruned_loss=0.122, over 5687281.14 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3931, pruned_loss=0.1421, over 5635756.02 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:25:18,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8207, 1.7709, 1.3739, 1.3721], device='cuda:0'), covar=tensor([0.0789, 0.0643, 0.0960, 0.1284], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0446, 0.0514, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 14:25:36,731 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=725118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:26:08,792 INFO [train.py:968] (0/2) Epoch 16, batch 41200, giga_loss[loss=0.3032, simple_loss=0.3745, pruned_loss=0.116, over 28853.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3938, pruned_loss=0.1427, over 5634967.29 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3704, pruned_loss=0.1217, over 5690479.12 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3951, pruned_loss=0.1438, over 5628161.69 frames. ], batch size: 112, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:26:34,551 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-08 14:26:38,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.785e+03 2.346e+03 3.560e+03 1.206e+04, threshold=4.692e+03, percent-clipped=15.0
+2023-03-08 14:27:02,930 INFO [train.py:968] (0/2) Epoch 16, batch 41250, giga_loss[loss=0.312, simple_loss=0.3779, pruned_loss=0.1231, over 29072.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.394, pruned_loss=0.1437, over 5624644.34 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3707, pruned_loss=0.1217, over 5686655.54 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.3958, pruned_loss=0.1453, over 5621116.51 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:27:21,374 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=725220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:27:52,710 INFO [train.py:968] (0/2) Epoch 16, batch 41300, giga_loss[loss=0.282, simple_loss=0.3563, pruned_loss=0.1038, over 28531.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3929, pruned_loss=0.1433, over 5640009.96 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3706, pruned_loss=0.1217, over 5688519.12 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3946, pruned_loss=0.1448, over 5635076.08 frames. ], batch size: 85, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:27:58,231 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-08 14:28:20,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.976e+03 2.436e+03 3.198e+03 6.978e+03, threshold=4.871e+03, percent-clipped=7.0
+2023-03-08 14:28:29,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=725288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:28:41,781 INFO [train.py:968] (0/2) Epoch 16, batch 41350, giga_loss[loss=0.3078, simple_loss=0.3803, pruned_loss=0.1177, over 28931.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3927, pruned_loss=0.142, over 5647207.21 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3702, pruned_loss=0.1215, over 5684993.13 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.395, pruned_loss=0.1439, over 5645375.83 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:29:16,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9379, 1.2517, 5.0808, 3.4510], device='cuda:0'), covar=tensor([0.1567, 0.2986, 0.0400, 0.0928], device='cuda:0'), in_proj_covar=tensor([0.0715, 0.0621, 0.0913, 0.0837], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 14:29:35,227 INFO [train.py:968] (0/2) Epoch 16, batch 41400, libri_loss[loss=0.2591, simple_loss=0.3313, pruned_loss=0.09343, over 29665.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3913, pruned_loss=0.1396, over 5653897.64 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.37, pruned_loss=0.1213, over 5685639.26 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3937, pruned_loss=0.1417, over 5650844.56 frames. ], batch size: 73, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:29:57,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.581e+03 2.148e+03 2.763e+03 7.837e+03, threshold=4.297e+03, percent-clipped=5.0
+2023-03-08 14:29:59,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=725378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:30:23,872 INFO [train.py:968] (0/2) Epoch 16, batch 41450, giga_loss[loss=0.3701, simple_loss=0.3977, pruned_loss=0.1712, over 23950.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3931, pruned_loss=0.1406, over 5640351.91 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3701, pruned_loss=0.1214, over 5680128.17 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3955, pruned_loss=0.1428, over 5642824.65 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:30:58,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=725431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:31:01,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=725434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:31:18,776 INFO [train.py:968] (0/2) Epoch 16, batch 41500, giga_loss[loss=0.347, simple_loss=0.3872, pruned_loss=0.1534, over 23788.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3915, pruned_loss=0.1392, over 5637702.94 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1211, over 5682001.54 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3942, pruned_loss=0.1413, over 5637430.12 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:31:32,630 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=725463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:31:44,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.546e+03 2.015e+03 2.986e+03 1.241e+04, threshold=4.029e+03, percent-clipped=8.0
+2023-03-08 14:32:01,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=725493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:32:08,134 INFO [train.py:968] (0/2) Epoch 16, batch 41550, giga_loss[loss=0.3382, simple_loss=0.3953, pruned_loss=0.1405, over 28371.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3889, pruned_loss=0.1357, over 5637072.02 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3695, pruned_loss=0.1212, over 5672139.32 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3915, pruned_loss=0.1376, over 5644697.78 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:32:28,752 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=725521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:32:30,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=725524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:33:00,132 INFO [train.py:968] (0/2) Epoch 16, batch 41600, giga_loss[loss=0.2894, simple_loss=0.3608, pruned_loss=0.109, over 28454.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3839, pruned_loss=0.131, over 5650050.34 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3691, pruned_loss=0.1211, over 5674565.26 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3865, pruned_loss=0.1328, over 5653554.99 frames. ], batch size: 60, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:33:01,673 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=725553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:33:26,033 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.701e+02 1.771e+03 2.461e+03 3.695e+03 1.413e+04, threshold=4.923e+03, percent-clipped=18.0
+2023-03-08 14:33:43,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=725595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:33:47,230 INFO [train.py:968] (0/2) Epoch 16, batch 41650, giga_loss[loss=0.2799, simple_loss=0.3451, pruned_loss=0.1073, over 28748.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3798, pruned_loss=0.1277, over 5657791.05 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3689, pruned_loss=0.1211, over 5681550.40 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3826, pruned_loss=0.1295, over 5653476.28 frames. ], batch size: 85, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:34:00,023 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-08 14:34:21,323 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=725636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:34:26,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=725639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:34:35,222 INFO [train.py:968] (0/2) Epoch 16, batch 41700, giga_loss[loss=0.33, simple_loss=0.3898, pruned_loss=0.1351, over 28878.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3784, pruned_loss=0.1276, over 5644285.89 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3689, pruned_loss=0.1213, over 5683633.09 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3809, pruned_loss=0.129, over 5637993.07 frames. ], batch size: 112, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:34:53,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=725668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:35:03,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.201e+03 1.661e+03 2.053e+03 3.050e+03 8.655e+03, threshold=4.105e+03, percent-clipped=9.0
+2023-03-08 14:35:25,216 INFO [train.py:968] (0/2) Epoch 16, batch 41750, giga_loss[loss=0.3045, simple_loss=0.3721, pruned_loss=0.1184, over 28987.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3776, pruned_loss=0.1265, over 5656935.65 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3687, pruned_loss=0.1212, over 5676148.60 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3798, pruned_loss=0.1277, over 5658063.50 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:36:07,118 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=725738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:36:08,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4611, 1.7176, 1.3918, 1.6290], device='cuda:0'), covar=tensor([0.2438, 0.2442, 0.2716, 0.2102], device='cuda:0'), in_proj_covar=tensor([0.1412, 0.1029, 0.1248, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 14:36:09,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=725741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:36:19,113 INFO [train.py:968] (0/2) Epoch 16, batch 41800, giga_loss[loss=0.2925, simple_loss=0.3671, pruned_loss=0.1089, over 28983.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3762, pruned_loss=0.1249, over 5663027.87 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3689, pruned_loss=0.1214, over 5679080.85 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3779, pruned_loss=0.1257, over 5661178.54 frames. ], batch size: 106, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:36:38,285 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=725770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:36:46,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.691e+02 1.528e+03 1.984e+03 2.985e+03 9.246e+03, threshold=3.968e+03, percent-clipped=11.0
+2023-03-08 14:36:54,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3780, 3.8856, 1.6046, 1.5267], device='cuda:0'), covar=tensor([0.1036, 0.0300, 0.0953, 0.1455], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0537, 0.0365, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 14:37:08,526 INFO [train.py:968] (0/2) Epoch 16, batch 41850, giga_loss[loss=0.3546, simple_loss=0.3944, pruned_loss=0.1574, over 26743.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3747, pruned_loss=0.1223, over 5678074.66 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3686, pruned_loss=0.1212, over 5688519.63 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3766, pruned_loss=0.1233, over 5667936.29 frames. ], batch size: 555, lr: 1.98e-03, grad_scale: 1.0
+2023-03-08 14:37:26,824 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-08 14:38:05,681 INFO [train.py:968] (0/2) Epoch 16, batch 41900, giga_loss[loss=0.3137, simple_loss=0.3719, pruned_loss=0.1277, over 28974.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3756, pruned_loss=0.1207, over 5666175.80 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3687, pruned_loss=0.1213, over 5678235.34 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3771, pruned_loss=0.1214, over 5666490.13 frames. ], batch size: 66, lr: 1.98e-03, grad_scale: 1.0
+2023-03-08 14:38:32,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.591e+03 2.041e+03 3.395e+03 2.394e+04, threshold=4.081e+03, percent-clipped=20.0
+2023-03-08 14:38:38,242 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=725883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:38:51,476 INFO [train.py:968] (0/2) Epoch 16, batch 41950, giga_loss[loss=0.291, simple_loss=0.3598, pruned_loss=0.111, over 28411.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3749, pruned_loss=0.12, over 5657387.80 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3688, pruned_loss=0.1214, over 5663476.97 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3762, pruned_loss=0.1204, over 5670041.98 frames. ], batch size: 85, lr: 1.98e-03, grad_scale: 1.0
+2023-03-08 14:39:36,718 INFO [train.py:968] (0/2) Epoch 16, batch 42000, giga_loss[loss=0.3127, simple_loss=0.3794, pruned_loss=0.123, over 28654.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3761, pruned_loss=0.1219, over 5655818.74 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3688, pruned_loss=0.1215, over 5660005.62 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3773, pruned_loss=0.1222, over 5669699.89 frames. ], batch size: 307, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:39:36,722 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 14:39:45,356 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2062, simple_loss=0.313, pruned_loss=0.04967, over 944034.00 frames. 
+2023-03-08 14:39:45,357 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 14:39:46,629 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=725951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:40:10,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.670e+03 2.258e+03 3.158e+03 9.677e+03, threshold=4.517e+03, percent-clipped=10.0
+2023-03-08 14:40:30,937 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-726000.pt
+2023-03-08 14:40:31,720 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-08 14:40:31,897 INFO [train.py:968] (0/2) Epoch 16, batch 42050, giga_loss[loss=0.3036, simple_loss=0.3724, pruned_loss=0.1174, over 28887.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3748, pruned_loss=0.1217, over 5656814.63 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3683, pruned_loss=0.1211, over 5654913.93 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3763, pruned_loss=0.1222, over 5671830.52 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:40:42,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5027, 1.8402, 1.4576, 1.6994], device='cuda:0'), covar=tensor([0.2424, 0.2519, 0.2798, 0.2486], device='cuda:0'), in_proj_covar=tensor([0.1413, 0.1030, 0.1251, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 14:40:59,915 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:41:20,890 INFO [train.py:968] (0/2) Epoch 16, batch 42100, giga_loss[loss=0.3204, simple_loss=0.3799, pruned_loss=0.1304, over 28812.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3738, pruned_loss=0.1224, over 5659969.99 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3682, pruned_loss=0.121, over 5657335.33 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3752, pruned_loss=0.1229, over 5669898.02 frames. ], batch size: 284, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:41:48,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.738e+03 2.266e+03 2.913e+03 1.395e+04, threshold=4.532e+03, percent-clipped=6.0
+2023-03-08 14:42:09,203 INFO [train.py:968] (0/2) Epoch 16, batch 42150, giga_loss[loss=0.3041, simple_loss=0.3805, pruned_loss=0.1139, over 28907.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.373, pruned_loss=0.1224, over 5659758.56 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3682, pruned_loss=0.121, over 5662334.77 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3741, pruned_loss=0.1229, over 5663149.25 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:42:22,616 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3299, 1.2127, 1.1457, 1.4582], device='cuda:0'), covar=tensor([0.0776, 0.0356, 0.0339, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0100], device='cuda:0')
+2023-03-08 14:42:59,862 INFO [train.py:968] (0/2) Epoch 16, batch 42200, giga_loss[loss=0.2717, simple_loss=0.3594, pruned_loss=0.092, over 28982.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3725, pruned_loss=0.1207, over 5672699.71 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3681, pruned_loss=0.1209, over 5665946.16 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3736, pruned_loss=0.1212, over 5672087.48 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:43:27,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.084e+02 1.636e+03 1.946e+03 2.711e+03 6.695e+03, threshold=3.891e+03, percent-clipped=8.0
+2023-03-08 14:43:47,676 INFO [train.py:968] (0/2) Epoch 16, batch 42250, libri_loss[loss=0.2801, simple_loss=0.3525, pruned_loss=0.1039, over 29655.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.373, pruned_loss=0.1204, over 5669494.74 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3682, pruned_loss=0.1208, over 5662117.87 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.374, pruned_loss=0.1208, over 5672238.79 frames. ], batch size: 88, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:44:35,825 INFO [train.py:968] (0/2) Epoch 16, batch 42300, giga_loss[loss=0.262, simple_loss=0.3336, pruned_loss=0.09523, over 28397.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3717, pruned_loss=0.1193, over 5684982.30 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3677, pruned_loss=0.1206, over 5666810.56 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.373, pruned_loss=0.1198, over 5683113.68 frames. ], batch size: 65, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:44:44,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=726258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:44:46,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7925, 1.9951, 2.0303, 1.5880], device='cuda:0'), covar=tensor([0.1836, 0.2186, 0.1412, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0698, 0.0914, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 14:45:04,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.629e+03 2.127e+03 2.759e+03 6.938e+03, threshold=4.254e+03, percent-clipped=12.0
+2023-03-08 14:45:24,951 INFO [train.py:968] (0/2) Epoch 16, batch 42350, giga_loss[loss=0.3328, simple_loss=0.3901, pruned_loss=0.1377, over 28307.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3699, pruned_loss=0.1186, over 5684566.15 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3679, pruned_loss=0.1207, over 5670733.23 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3708, pruned_loss=0.119, over 5679847.28 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:45:48,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=726326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:46:09,492 INFO [train.py:968] (0/2) Epoch 16, batch 42400, giga_loss[loss=0.2661, simple_loss=0.3367, pruned_loss=0.09774, over 29155.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3688, pruned_loss=0.1185, over 5687028.79 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3675, pruned_loss=0.1204, over 5676331.74 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3699, pruned_loss=0.119, over 5678066.50 frames. ], batch size: 113, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:46:38,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.580e+03 1.968e+03 2.686e+03 8.138e+03, threshold=3.936e+03, percent-clipped=7.0
+2023-03-08 14:46:41,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5410, 1.5272, 1.2645, 1.1071], device='cuda:0'), covar=tensor([0.0871, 0.0577, 0.1076, 0.1057], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0445, 0.0512, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 14:46:57,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:46:59,931 INFO [train.py:968] (0/2) Epoch 16, batch 42450, giga_loss[loss=0.355, simple_loss=0.3837, pruned_loss=0.1632, over 23764.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3684, pruned_loss=0.1195, over 5675189.38 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3675, pruned_loss=0.1203, over 5678186.65 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3694, pruned_loss=0.1199, over 5666757.21 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:47:00,272 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=726401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:47:05,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=726404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:47:06,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=726407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:47:30,369 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=726433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:47:47,089 INFO [train.py:968] (0/2) Epoch 16, batch 42500, giga_loss[loss=0.2496, simple_loss=0.3292, pruned_loss=0.08502, over 29057.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3682, pruned_loss=0.1203, over 5668862.08 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3676, pruned_loss=0.1205, over 5673780.10 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1204, over 5666026.45 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:48:04,646 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=726469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:48:07,891 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=726472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:48:13,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:48:14,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.706e+03 2.220e+03 2.983e+03 6.635e+03, threshold=4.441e+03, percent-clipped=14.0
+2023-03-08 14:48:35,256 INFO [train.py:968] (0/2) Epoch 16, batch 42550, libri_loss[loss=0.3161, simple_loss=0.3855, pruned_loss=0.1233, over 29574.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3672, pruned_loss=0.1197, over 5679948.42 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3675, pruned_loss=0.1203, over 5680261.90 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3678, pruned_loss=0.1199, over 5671856.32 frames. ], batch size: 83, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:48:35,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=726501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:49:23,816 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=726550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:49:24,314 INFO [train.py:968] (0/2) Epoch 16, batch 42600, giga_loss[loss=0.3051, simple_loss=0.3687, pruned_loss=0.1208, over 28036.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3679, pruned_loss=0.1205, over 5678290.86 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3675, pruned_loss=0.1201, over 5675349.87 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3684, pruned_loss=0.1209, over 5676180.10 frames. ], batch size: 412, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:49:27,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=726553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:49:53,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.538e+02 1.669e+03 1.979e+03 2.727e+03 6.840e+03, threshold=3.958e+03, percent-clipped=4.0
+2023-03-08 14:49:54,684 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=726582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:50:12,550 INFO [train.py:968] (0/2) Epoch 16, batch 42650, giga_loss[loss=0.3201, simple_loss=0.3856, pruned_loss=0.1273, over 28864.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3686, pruned_loss=0.1198, over 5684710.71 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3677, pruned_loss=0.1202, over 5680862.85 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3688, pruned_loss=0.12, over 5678053.25 frames. ], batch size: 199, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:50:21,080 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-08 14:50:59,507 INFO [train.py:968] (0/2) Epoch 16, batch 42700, libri_loss[loss=0.3156, simple_loss=0.3853, pruned_loss=0.123, over 29527.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3686, pruned_loss=0.1189, over 5688274.91 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3678, pruned_loss=0.1201, over 5685309.34 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3687, pruned_loss=0.1191, over 5679019.47 frames. ], batch size: 84, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:51:26,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.071e+02 1.498e+03 1.911e+03 2.539e+03 6.192e+03, threshold=3.821e+03, percent-clipped=6.0
+2023-03-08 14:51:44,564 INFO [train.py:968] (0/2) Epoch 16, batch 42750, giga_loss[loss=0.2831, simple_loss=0.3579, pruned_loss=0.1042, over 28873.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3685, pruned_loss=0.1181, over 5687723.29 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3672, pruned_loss=0.1198, over 5690762.43 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3691, pruned_loss=0.1186, over 5675319.70 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:52:12,290 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:52:36,959 INFO [train.py:968] (0/2) Epoch 16, batch 42800, giga_loss[loss=0.3721, simple_loss=0.4044, pruned_loss=0.1699, over 23792.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3692, pruned_loss=0.1193, over 5674056.32 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3669, pruned_loss=0.1196, over 5694254.92 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.37, pruned_loss=0.1199, over 5660857.74 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:52:59,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=726774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:53:02,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.609e+03 2.079e+03 2.778e+03 4.928e+03, threshold=4.158e+03, percent-clipped=6.0
+2023-03-08 14:53:05,587 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:53:21,140 INFO [train.py:968] (0/2) Epoch 16, batch 42850, giga_loss[loss=0.344, simple_loss=0.3971, pruned_loss=0.1454, over 28416.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3716, pruned_loss=0.1219, over 5673164.17 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3671, pruned_loss=0.1195, over 5694375.08 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3722, pruned_loss=0.1224, over 5661646.81 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:54:10,163 INFO [train.py:968] (0/2) Epoch 16, batch 42900, giga_loss[loss=0.2912, simple_loss=0.3569, pruned_loss=0.1127, over 28868.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3756, pruned_loss=0.1265, over 5673436.87 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3679, pruned_loss=0.1201, over 5701513.90 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3756, pruned_loss=0.1266, over 5657136.90 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:54:14,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=726854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:54:48,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.225e+03 2.089e+03 2.770e+03 3.965e+03 1.373e+04, threshold=5.541e+03, percent-clipped=22.0
+2023-03-08 14:54:59,680 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3554, 1.9100, 1.4353, 0.6852], device='cuda:0'), covar=tensor([0.4258, 0.2342, 0.2839, 0.4948], device='cuda:0'), in_proj_covar=tensor([0.1666, 0.1574, 0.1551, 0.1361], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 14:55:07,546 INFO [train.py:968] (0/2) Epoch 16, batch 42950, giga_loss[loss=0.3008, simple_loss=0.3644, pruned_loss=0.1186, over 28719.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3775, pruned_loss=0.1297, over 5666680.45 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3679, pruned_loss=0.1201, over 5703554.64 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3776, pruned_loss=0.1299, over 5651808.32 frames. ], batch size: 242, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:55:23,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=726917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:55:26,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=726920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:55:35,233 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:55:45,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2869, 1.5497, 1.4893, 1.3393], device='cuda:0'), covar=tensor([0.1744, 0.1671, 0.2381, 0.1900], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0741, 0.0701, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 14:55:54,814 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=726949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:55:55,976 INFO [train.py:968] (0/2) Epoch 16, batch 43000, libri_loss[loss=0.3466, simple_loss=0.3943, pruned_loss=0.1494, over 19083.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3783, pruned_loss=0.1301, over 5665228.34 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3685, pruned_loss=0.1203, over 5696280.00 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3782, pruned_loss=0.1304, over 5658832.74 frames. ], batch size: 187, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:56:19,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5853, 1.6646, 1.7744, 1.3585], device='cuda:0'), covar=tensor([0.1784, 0.2481, 0.1455, 0.1658], device='cuda:0'), in_proj_covar=tensor([0.0868, 0.0698, 0.0913, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 14:56:23,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.654e+03 1.943e+03 2.546e+03 8.011e+03, threshold=3.885e+03, percent-clipped=3.0
+2023-03-08 14:56:37,859 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=726997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:56:40,954 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:56:41,281 INFO [train.py:968] (0/2) Epoch 16, batch 43050, giga_loss[loss=0.268, simple_loss=0.338, pruned_loss=0.09902, over 29076.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3756, pruned_loss=0.1279, over 5667394.47 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3684, pruned_loss=0.1202, over 5695889.97 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3757, pruned_loss=0.1284, over 5662213.74 frames. ], batch size: 128, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:56:59,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6389, 2.3652, 1.7293, 0.8537], device='cuda:0'), covar=tensor([0.5380, 0.2625, 0.3431, 0.5664], device='cuda:0'), in_proj_covar=tensor([0.1669, 0.1579, 0.1553, 0.1360], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 14:57:06,441 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:57:25,807 INFO [train.py:968] (0/2) Epoch 16, batch 43100, giga_loss[loss=0.2876, simple_loss=0.3622, pruned_loss=0.1065, over 28674.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3743, pruned_loss=0.1258, over 5668935.07 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3685, pruned_loss=0.1203, over 5692773.29 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3745, pruned_loss=0.1263, over 5666114.08 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:57:52,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.070e+02 1.393e+03 1.814e+03 2.489e+03 8.211e+03, threshold=3.628e+03, percent-clipped=9.0
+2023-03-08 14:58:10,547 INFO [train.py:968] (0/2) Epoch 16, batch 43150, giga_loss[loss=0.2735, simple_loss=0.3238, pruned_loss=0.1116, over 23909.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3708, pruned_loss=0.1217, over 5678903.55 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3681, pruned_loss=0.12, over 5695764.09 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3713, pruned_loss=0.1224, over 5673754.45 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:58:13,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:58:13,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4289, 1.7532, 1.4017, 1.5867], device='cuda:0'), covar=tensor([0.2401, 0.2371, 0.2514, 0.2141], device='cuda:0'), in_proj_covar=tensor([0.1414, 0.1032, 0.1251, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 14:58:14,251 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:58:17,069 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 14:58:58,292 INFO [train.py:968] (0/2) Epoch 16, batch 43200, giga_loss[loss=0.2755, simple_loss=0.3393, pruned_loss=0.1058, over 28829.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3691, pruned_loss=0.1213, over 5666421.67 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3683, pruned_loss=0.1203, over 5695318.35 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3694, pruned_loss=0.1217, over 5662458.49 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:59:05,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:59:17,631 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-08 14:59:21,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7263, 1.9910, 1.6093, 2.3036], device='cuda:0'), covar=tensor([0.2266, 0.2387, 0.2626, 0.2047], device='cuda:0'), in_proj_covar=tensor([0.1414, 0.1032, 0.1251, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 14:59:24,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.745e+03 2.330e+03 3.572e+03 1.413e+04, threshold=4.659e+03, percent-clipped=23.0
+2023-03-08 14:59:30,569 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-08 14:59:44,067 INFO [train.py:968] (0/2) Epoch 16, batch 43250, giga_loss[loss=0.2893, simple_loss=0.3506, pruned_loss=0.1141, over 28997.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3676, pruned_loss=0.1209, over 5673071.11 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3684, pruned_loss=0.1204, over 5699289.21 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3678, pruned_loss=0.1211, over 5665851.72 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:59:55,678 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-08 15:00:19,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6112, 1.7622, 1.4544, 1.6626], device='cuda:0'), covar=tensor([0.2456, 0.2541, 0.2692, 0.2402], device='cuda:0'), in_proj_covar=tensor([0.1414, 0.1030, 0.1250, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 15:00:23,830 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:00:26,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:00:28,267 INFO [train.py:968] (0/2) Epoch 16, batch 43300, giga_loss[loss=0.4323, simple_loss=0.4414, pruned_loss=0.2116, over 26722.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3678, pruned_loss=0.1219, over 5670433.27 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3678, pruned_loss=0.12, over 5706021.77 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3684, pruned_loss=0.1225, over 5657320.37 frames. ], batch size: 555, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:00:53,467 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:00:54,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6429, 1.6857, 1.8505, 1.4153], device='cuda:0'), covar=tensor([0.1687, 0.2381, 0.1378, 0.1621], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0699, 0.0916, 0.0814], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 15:00:56,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.554e+02 1.634e+03 2.069e+03 2.701e+03 1.156e+04, threshold=4.137e+03, percent-clipped=6.0
+2023-03-08 15:01:17,902 INFO [train.py:968] (0/2) Epoch 16, batch 43350, giga_loss[loss=0.3168, simple_loss=0.3842, pruned_loss=0.1247, over 28926.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3704, pruned_loss=0.123, over 5675987.44 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3674, pruned_loss=0.1197, over 5707805.29 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3713, pruned_loss=0.1237, over 5663645.93 frames. ], batch size: 227, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:01:18,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:01:20,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:01:20,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:01:39,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3988, 1.7248, 1.6650, 1.2381], device='cuda:0'), covar=tensor([0.1322, 0.2253, 0.1183, 0.1527], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0699, 0.0916, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 15:01:46,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:01:46,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5814, 1.6616, 1.8274, 1.3391], device='cuda:0'), covar=tensor([0.1998, 0.2517, 0.1578, 0.1844], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0699, 0.0916, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 15:02:03,076 INFO [train.py:968] (0/2) Epoch 16, batch 43400, giga_loss[loss=0.3327, simple_loss=0.4052, pruned_loss=0.1301, over 28600.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3748, pruned_loss=0.1236, over 5676622.45 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3672, pruned_loss=0.1194, over 5711609.59 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3759, pruned_loss=0.1246, over 5662106.29 frames. ], batch size: 85, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:02:12,309 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:02:37,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.159e+02 1.658e+03 2.157e+03 3.114e+03 7.320e+03, threshold=4.313e+03, percent-clipped=8.0
+2023-03-08 15:02:56,760 INFO [train.py:968] (0/2) Epoch 16, batch 43450, giga_loss[loss=0.3112, simple_loss=0.3767, pruned_loss=0.1228, over 28716.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3772, pruned_loss=0.1238, over 5663410.86 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3671, pruned_loss=0.1193, over 5705901.36 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3784, pruned_loss=0.1247, over 5656182.42 frames. ], batch size: 242, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:03:08,436 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-08 15:03:15,867 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 15:03:37,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:03:39,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:03:40,072 INFO [train.py:968] (0/2) Epoch 16, batch 43500, giga_loss[loss=0.3447, simple_loss=0.4058, pruned_loss=0.1418, over 28952.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3783, pruned_loss=0.1249, over 5672358.20 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3666, pruned_loss=0.1192, over 5704651.37 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3801, pruned_loss=0.126, over 5665659.70 frames. ], batch size: 227, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:04:02,682 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:04:08,396 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:04:09,908 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:04:13,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.904e+03 2.654e+03 3.861e+03 1.174e+04, threshold=5.309e+03, percent-clipped=18.0
+2023-03-08 15:04:30,509 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:04:30,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4548, 1.5462, 1.6769, 1.3289], device='cuda:0'), covar=tensor([0.1375, 0.2050, 0.1123, 0.1410], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0699, 0.0916, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 15:04:32,374 INFO [train.py:968] (0/2) Epoch 16, batch 43550, giga_loss[loss=0.2564, simple_loss=0.3264, pruned_loss=0.09322, over 28383.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3802, pruned_loss=0.1268, over 5671596.13 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3665, pruned_loss=0.1191, over 5705805.11 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3817, pruned_loss=0.1278, over 5665075.37 frames. ], batch size: 77, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:04:39,736 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:04:58,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4152, 1.4496, 1.2908, 1.4845], device='cuda:0'), covar=tensor([0.0762, 0.0338, 0.0326, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0100], device='cuda:0')
+2023-03-08 15:05:18,308 INFO [train.py:968] (0/2) Epoch 16, batch 43600, giga_loss[loss=0.3101, simple_loss=0.3695, pruned_loss=0.1253, over 28880.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3798, pruned_loss=0.1274, over 5670400.41 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3669, pruned_loss=0.1194, over 5699699.79 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3811, pruned_loss=0.1281, over 5668933.85 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:05:30,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727564.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:05:50,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.555e+03 1.948e+03 2.652e+03 5.317e+03, threshold=3.896e+03, percent-clipped=1.0
+2023-03-08 15:06:08,126 INFO [train.py:968] (0/2) Epoch 16, batch 43650, giga_loss[loss=0.2602, simple_loss=0.3369, pruned_loss=0.09174, over 28939.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3792, pruned_loss=0.1279, over 5652321.37 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3671, pruned_loss=0.1195, over 5692088.16 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3802, pruned_loss=0.1285, over 5658032.43 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:06:27,713 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:06:32,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:06:55,877 INFO [train.py:968] (0/2) Epoch 16, batch 43700, giga_loss[loss=0.2884, simple_loss=0.355, pruned_loss=0.1109, over 28953.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3766, pruned_loss=0.1268, over 5659439.49 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3673, pruned_loss=0.1197, over 5691976.28 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3773, pruned_loss=0.1272, over 5663424.12 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:06:59,082 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:07:27,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.606e+03 1.955e+03 2.630e+03 7.162e+03, threshold=3.911e+03, percent-clipped=9.0
+2023-03-08 15:07:43,116 INFO [train.py:968] (0/2) Epoch 16, batch 43750, giga_loss[loss=0.3458, simple_loss=0.3812, pruned_loss=0.1552, over 23437.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3758, pruned_loss=0.1267, over 5663920.40 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3678, pruned_loss=0.1199, over 5697367.25 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3762, pruned_loss=0.127, over 5661252.01 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:07:56,027 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-08 15:08:17,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:08:24,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3297, 1.7097, 1.4128, 1.5653], device='cuda:0'), covar=tensor([0.0776, 0.0316, 0.0314, 0.0820], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 15:08:36,126 INFO [train.py:968] (0/2) Epoch 16, batch 43800, giga_loss[loss=0.4315, simple_loss=0.455, pruned_loss=0.204, over 27909.00 frames. ], tot_loss[loss=0.314, simple_loss=0.375, pruned_loss=0.1265, over 5653722.64 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3674, pruned_loss=0.1197, over 5702670.81 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3759, pruned_loss=0.1271, over 5645801.78 frames. ], batch size: 412, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:08:56,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3990, 1.3852, 3.8189, 3.1978], device='cuda:0'), covar=tensor([0.1548, 0.2519, 0.0463, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0722, 0.0624, 0.0921, 0.0844], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 15:09:06,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.980e+02 1.806e+03 2.298e+03 2.989e+03 8.393e+03, threshold=4.595e+03, percent-clipped=8.0
+2023-03-08 15:09:24,672 INFO [train.py:968] (0/2) Epoch 16, batch 43850, giga_loss[loss=0.3549, simple_loss=0.4117, pruned_loss=0.149, over 28621.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3737, pruned_loss=0.1262, over 5658456.05 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3673, pruned_loss=0.1196, over 5706848.81 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3747, pruned_loss=0.127, over 5647272.25 frames. ], batch size: 307, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:10:06,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:10:08,142 INFO [train.py:968] (0/2) Epoch 16, batch 43900, giga_loss[loss=0.2886, simple_loss=0.3586, pruned_loss=0.1093, over 28798.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3713, pruned_loss=0.1243, over 5674666.95 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3668, pruned_loss=0.1191, over 5712588.66 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3727, pruned_loss=0.1254, over 5659091.56 frames. ], batch size: 285, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:10:22,410 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-08 15:10:27,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:10:30,018 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:10:32,990 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:10:37,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.355e+02 1.593e+03 2.084e+03 3.187e+03 7.856e+03, threshold=4.169e+03, percent-clipped=5.0
+2023-03-08 15:10:37,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:10:52,762 INFO [train.py:968] (0/2) Epoch 16, batch 43950, giga_loss[loss=0.3281, simple_loss=0.3898, pruned_loss=0.1333, over 28755.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.37, pruned_loss=0.1232, over 5673680.96 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3666, pruned_loss=0.1189, over 5716858.37 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3714, pruned_loss=0.1245, over 5655986.35 frames. ], batch size: 284, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:10:59,169 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:11:18,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4449, 4.4076, 1.7020, 1.6058], device='cuda:0'), covar=tensor([0.1023, 0.0266, 0.0933, 0.1411], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0540, 0.0366, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 15:11:32,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727939.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:11:42,808 INFO [train.py:968] (0/2) Epoch 16, batch 44000, giga_loss[loss=0.2838, simple_loss=0.3581, pruned_loss=0.1048, over 28996.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3723, pruned_loss=0.1238, over 5636121.18 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1192, over 5682827.95 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3731, pruned_loss=0.1246, over 5652644.88 frames. ], batch size: 128, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 15:11:54,679 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:15,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.183e+02 1.581e+03 2.034e+03 2.624e+03 5.731e+03, threshold=4.069e+03, percent-clipped=5.0
+2023-03-08 15:12:22,885 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:24,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:31,283 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-728000.pt
+2023-03-08 15:12:32,258 INFO [train.py:968] (0/2) Epoch 16, batch 44050, giga_loss[loss=0.3563, simple_loss=0.4, pruned_loss=0.1564, over 27452.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3761, pruned_loss=0.1267, over 5635953.15 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3674, pruned_loss=0.1195, over 5681672.93 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3764, pruned_loss=0.1272, over 5649307.35 frames. ], batch size: 472, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 15:12:46,403 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=728016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:49,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=728019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:52,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:52,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3591, 1.9471, 1.5010, 0.6881], device='cuda:0'), covar=tensor([0.4155, 0.2375, 0.3154, 0.4753], device='cuda:0'), in_proj_covar=tensor([0.1674, 0.1583, 0.1551, 0.1363], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 15:12:52,883 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-08 15:12:55,706 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=728026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:13:01,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=728029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:13:20,281 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:13:21,885 INFO [train.py:968] (0/2) Epoch 16, batch 44100, giga_loss[loss=0.3511, simple_loss=0.3942, pruned_loss=0.154, over 27542.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3753, pruned_loss=0.1266, over 5637228.84 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3675, pruned_loss=0.1196, over 5673470.88 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3756, pruned_loss=0.127, over 5655050.77 frames. ], batch size: 472, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:13:27,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4739, 1.7770, 1.6894, 1.5749], device='cuda:0'), covar=tensor([0.1677, 0.1733, 0.1949, 0.1826], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0747, 0.0703, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 15:13:27,667 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=728057.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:13:28,467 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:13:51,030 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=728082.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:13:52,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.545e+02 1.628e+03 2.084e+03 2.889e+03 4.540e+03, threshold=4.168e+03, percent-clipped=6.0
+2023-03-08 15:13:52,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=728085.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:14:05,284 INFO [train.py:968] (0/2) Epoch 16, batch 44150, giga_loss[loss=0.2728, simple_loss=0.3582, pruned_loss=0.09365, over 28914.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3758, pruned_loss=0.1236, over 5653538.74 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3676, pruned_loss=0.1196, over 5674760.96 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3763, pruned_loss=0.1241, over 5666108.51 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:14:18,059 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728114.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:14:52,432 INFO [train.py:968] (0/2) Epoch 16, batch 44200, giga_loss[loss=0.342, simple_loss=0.413, pruned_loss=0.1356, over 29023.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3776, pruned_loss=0.123, over 5644203.31 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3674, pruned_loss=0.1195, over 5667342.61 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3782, pruned_loss=0.1235, over 5660080.66 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:15:28,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.503e+02 1.468e+03 1.696e+03 2.321e+03 5.222e+03, threshold=3.392e+03, percent-clipped=4.0
+2023-03-08 15:15:43,084 INFO [train.py:968] (0/2) Epoch 16, batch 44250, giga_loss[loss=0.3632, simple_loss=0.4126, pruned_loss=0.1569, over 28786.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3804, pruned_loss=0.1252, over 5638415.90 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3676, pruned_loss=0.1197, over 5662013.41 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3809, pruned_loss=0.1256, over 5655117.04 frames. ], batch size: 284, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:16:27,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0477, 1.3063, 0.9744, 0.9743], device='cuda:0'), covar=tensor([0.1054, 0.0516, 0.1263, 0.1043], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0449, 0.0511, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 15:16:31,833 INFO [train.py:968] (0/2) Epoch 16, batch 44300, giga_loss[loss=0.2933, simple_loss=0.366, pruned_loss=0.1103, over 29051.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3833, pruned_loss=0.1288, over 5649681.23 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3674, pruned_loss=0.1195, over 5666248.15 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3842, pruned_loss=0.1293, over 5658849.09 frames. ], batch size: 128, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:17:04,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.290e+02 1.878e+03 2.292e+03 3.190e+03 1.070e+04, threshold=4.584e+03, percent-clipped=22.0
+2023-03-08 15:17:22,339 INFO [train.py:968] (0/2) Epoch 16, batch 44350, giga_loss[loss=0.3744, simple_loss=0.42, pruned_loss=0.1644, over 29084.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3835, pruned_loss=0.1297, over 5656010.05 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1191, over 5670297.36 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.385, pruned_loss=0.1308, over 5659071.96 frames. ], batch size: 128, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:17:53,021 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3448, 1.3180, 3.1687, 3.0114], device='cuda:0'), covar=tensor([0.1244, 0.2405, 0.0492, 0.0975], device='cuda:0'), in_proj_covar=tensor([0.0721, 0.0621, 0.0920, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 15:17:55,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=728336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:18:08,534 INFO [train.py:968] (0/2) Epoch 16, batch 44400, libri_loss[loss=0.3657, simple_loss=0.4143, pruned_loss=0.1585, over 29242.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3816, pruned_loss=0.1287, over 5669635.74 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3668, pruned_loss=0.1192, over 5674758.16 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3831, pruned_loss=0.1296, over 5668077.58 frames. ], batch size: 94, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:18:26,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6765, 2.0892, 2.0235, 1.6060], device='cuda:0'), covar=tensor([0.2944, 0.2123, 0.2091, 0.2432], device='cuda:0'), in_proj_covar=tensor([0.1859, 0.1802, 0.1725, 0.1864], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 15:18:36,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.655e+03 2.414e+03 3.281e+03 8.250e+03, threshold=4.827e+03, percent-clipped=7.0
+2023-03-08 15:18:53,254 INFO [train.py:968] (0/2) Epoch 16, batch 44450, giga_loss[loss=0.3069, simple_loss=0.3775, pruned_loss=0.1182, over 28570.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3792, pruned_loss=0.1267, over 5662507.35 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3665, pruned_loss=0.119, over 5675855.98 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3811, pruned_loss=0.1279, over 5659933.46 frames. ], batch size: 307, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:19:18,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=728432.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:19:37,931 INFO [train.py:968] (0/2) Epoch 16, batch 44500, libri_loss[loss=0.2732, simple_loss=0.3502, pruned_loss=0.09807, over 29521.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3778, pruned_loss=0.1236, over 5672049.48 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3658, pruned_loss=0.1187, over 5672621.59 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3804, pruned_loss=0.125, over 5671804.54 frames. ], batch size: 84, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:20:05,250 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=728479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:20:07,392 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=728482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:20:09,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.984e+02 1.520e+03 2.015e+03 2.700e+03 4.980e+03, threshold=4.031e+03, percent-clipped=2.0
+2023-03-08 15:20:19,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-08 15:20:22,204 INFO [train.py:968] (0/2) Epoch 16, batch 44550, giga_loss[loss=0.3106, simple_loss=0.3827, pruned_loss=0.1193, over 28644.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3776, pruned_loss=0.1223, over 5681485.75 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3657, pruned_loss=0.1185, over 5677654.87 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3801, pruned_loss=0.1237, over 5676727.73 frames. ], batch size: 307, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:20:31,857 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:21:12,884 INFO [train.py:968] (0/2) Epoch 16, batch 44600, giga_loss[loss=0.3171, simple_loss=0.3843, pruned_loss=0.125, over 28829.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3796, pruned_loss=0.1248, over 5663809.61 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3661, pruned_loss=0.1188, over 5679132.66 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3815, pruned_loss=0.1258, over 5658478.12 frames. ], batch size: 99, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:21:31,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3133, 1.3858, 1.2472, 1.4850], device='cuda:0'), covar=tensor([0.0762, 0.0339, 0.0320, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 15:21:39,318 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=728575.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:21:42,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=728578.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:21:47,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.746e+03 2.376e+03 3.378e+03 9.655e+03, threshold=4.751e+03, percent-clipped=16.0
+2023-03-08 15:22:01,668 INFO [train.py:968] (0/2) Epoch 16, batch 44650, giga_loss[loss=0.2704, simple_loss=0.3499, pruned_loss=0.09547, over 28982.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3792, pruned_loss=0.125, over 5652015.79 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3662, pruned_loss=0.1188, over 5672358.62 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3809, pruned_loss=0.1259, over 5653754.41 frames. ], batch size: 145, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:22:07,441 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728607.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:22:50,167 INFO [train.py:968] (0/2) Epoch 16, batch 44700, giga_loss[loss=0.2714, simple_loss=0.3516, pruned_loss=0.09555, over 28442.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3768, pruned_loss=0.1241, over 5666490.85 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3661, pruned_loss=0.1187, over 5676999.91 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3785, pruned_loss=0.125, over 5663600.92 frames. ], batch size: 60, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:22:57,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9144, 2.1518, 1.9921, 1.8136], device='cuda:0'), covar=tensor([0.1942, 0.1640, 0.1499, 0.1567], device='cuda:0'), in_proj_covar=tensor([0.1851, 0.1790, 0.1717, 0.1855], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 15:23:22,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.610e+03 2.244e+03 3.090e+03 5.544e+03, threshold=4.487e+03, percent-clipped=6.0
+2023-03-08 15:23:33,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8873, 2.2359, 2.0925, 1.6693], device='cuda:0'), covar=tensor([0.2347, 0.1805, 0.1837, 0.2170], device='cuda:0'), in_proj_covar=tensor([0.1846, 0.1788, 0.1716, 0.1855], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 15:23:36,481 INFO [train.py:968] (0/2) Epoch 16, batch 44750, giga_loss[loss=0.3207, simple_loss=0.3779, pruned_loss=0.1318, over 28017.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3764, pruned_loss=0.1253, over 5656743.04 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3661, pruned_loss=0.1187, over 5673064.60 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3781, pruned_loss=0.1262, over 5657509.98 frames. ], batch size: 412, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:24:02,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0868, 2.4185, 2.1210, 1.7287], device='cuda:0'), covar=tensor([0.2761, 0.1858, 0.2115, 0.2430], device='cuda:0'), in_proj_covar=tensor([0.1849, 0.1789, 0.1716, 0.1857], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 15:24:02,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=728727.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:24:26,146 INFO [train.py:968] (0/2) Epoch 16, batch 44800, giga_loss[loss=0.3126, simple_loss=0.3796, pruned_loss=0.1228, over 28831.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3745, pruned_loss=0.1243, over 5663532.03 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3663, pruned_loss=0.119, over 5675645.98 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3756, pruned_loss=0.1248, over 5661652.68 frames. ], batch size: 174, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:24:54,180 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=728782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:24:56,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.618e+03 2.188e+03 2.898e+03 8.100e+03, threshold=4.376e+03, percent-clipped=4.0
+2023-03-08 15:25:09,281 INFO [train.py:968] (0/2) Epoch 16, batch 44850, libri_loss[loss=0.2572, simple_loss=0.3293, pruned_loss=0.0926, over 29592.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3741, pruned_loss=0.1251, over 5664960.50 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3662, pruned_loss=0.1188, over 5682365.75 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3754, pruned_loss=0.1258, over 5656874.84 frames. ], batch size: 74, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:25:17,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4944, 1.6066, 1.6240, 1.4914], device='cuda:0'), covar=tensor([0.1388, 0.1633, 0.1657, 0.1553], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0742, 0.0698, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 15:25:43,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3778, 2.8273, 1.4313, 1.4947], device='cuda:0'), covar=tensor([0.0907, 0.0349, 0.0881, 0.1277], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0540, 0.0366, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 15:25:57,701 INFO [train.py:968] (0/2) Epoch 16, batch 44900, giga_loss[loss=0.2946, simple_loss=0.3648, pruned_loss=0.1122, over 28573.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3742, pruned_loss=0.1256, over 5649683.62 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3664, pruned_loss=0.1187, over 5684840.20 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3751, pruned_loss=0.1264, over 5640610.97 frames. ], batch size: 336, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:26:00,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2222, 1.1925, 3.6700, 3.1184], device='cuda:0'), covar=tensor([0.1638, 0.2700, 0.0447, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0720, 0.0621, 0.0919, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 15:26:17,435 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6869, 1.8614, 1.3541, 1.5163], device='cuda:0'), covar=tensor([0.0802, 0.0459, 0.0948, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0446, 0.0510, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 15:26:27,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.850e+02 1.672e+03 2.257e+03 3.371e+03 8.840e+03, threshold=4.515e+03, percent-clipped=11.0
+2023-03-08 15:26:43,638 INFO [train.py:968] (0/2) Epoch 16, batch 44950, giga_loss[loss=0.3035, simple_loss=0.3748, pruned_loss=0.1161, over 28940.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3709, pruned_loss=0.1213, over 5662772.23 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3663, pruned_loss=0.1185, over 5690646.09 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3719, pruned_loss=0.1222, over 5649564.56 frames. ], batch size: 227, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:27:10,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9047, 1.1406, 3.3043, 2.9638], device='cuda:0'), covar=tensor([0.2177, 0.3123, 0.0907, 0.1373], device='cuda:0'), in_proj_covar=tensor([0.0721, 0.0621, 0.0918, 0.0841], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 15:27:21,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6646, 1.7827, 1.2800, 1.4083], device='cuda:0'), covar=tensor([0.0929, 0.0610, 0.1095, 0.1092], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0445, 0.0508, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 15:27:26,139 INFO [train.py:968] (0/2) Epoch 16, batch 45000, giga_loss[loss=0.2689, simple_loss=0.3521, pruned_loss=0.09283, over 29089.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.367, pruned_loss=0.1174, over 5658384.04 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3666, pruned_loss=0.1186, over 5688742.12 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3676, pruned_loss=0.118, over 5648046.88 frames. ], batch size: 155, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:27:26,145 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 15:27:33,287 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2538, 1.5518, 1.4641, 1.2115], device='cuda:0'), covar=tensor([0.2425, 0.2002, 0.1381, 0.1868], device='cuda:0'), in_proj_covar=tensor([0.1845, 0.1790, 0.1717, 0.1855], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 15:27:34,847 INFO [train.py:1012] (0/2) Epoch 16, validation: loss=0.2086, simple_loss=0.3155, pruned_loss=0.05086, over 944034.00 frames. 
+2023-03-08 15:27:34,847 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 15:28:02,270 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-08 15:28:11,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.860e+02 1.398e+03 1.925e+03 3.071e+03 8.933e+03, threshold=3.851e+03, percent-clipped=7.0
+2023-03-08 15:28:29,335 INFO [train.py:968] (0/2) Epoch 16, batch 45050, giga_loss[loss=0.3001, simple_loss=0.3677, pruned_loss=0.1163, over 29052.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3671, pruned_loss=0.1179, over 5655923.19 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3667, pruned_loss=0.1187, over 5689113.41 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3675, pruned_loss=0.1184, over 5647321.76 frames. ], batch size: 128, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:29:18,460 INFO [train.py:968] (0/2) Epoch 16, batch 45100, giga_loss[loss=0.2857, simple_loss=0.3485, pruned_loss=0.1115, over 28611.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3647, pruned_loss=0.1171, over 5669653.66 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3668, pruned_loss=0.1188, over 5690320.39 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3649, pruned_loss=0.1173, over 5661357.23 frames. ], batch size: 307, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:29:23,269 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.36 vs. limit=5.0
+2023-03-08 15:29:46,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-08 15:29:54,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.790e+02 1.637e+03 2.046e+03 2.665e+03 7.646e+03, threshold=4.092e+03, percent-clipped=10.0
+2023-03-08 15:30:06,113 INFO [train.py:968] (0/2) Epoch 16, batch 45150, giga_loss[loss=0.2738, simple_loss=0.342, pruned_loss=0.1028, over 28804.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3643, pruned_loss=0.1174, over 5669296.51 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3669, pruned_loss=0.1188, over 5683061.08 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3644, pruned_loss=0.1176, over 5667934.62 frames. ], batch size: 99, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:30:07,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=729102.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:30:51,392 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-08 15:30:52,971 INFO [train.py:968] (0/2) Epoch 16, batch 45200, giga_loss[loss=0.2892, simple_loss=0.3687, pruned_loss=0.1049, over 29108.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3663, pruned_loss=0.1179, over 5684002.47 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3673, pruned_loss=0.119, over 5686406.95 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.366, pruned_loss=0.1178, over 5679887.33 frames. ], batch size: 155, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:31:00,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=729157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:31:29,187 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.856e+02 1.599e+03 1.930e+03 2.577e+03 8.276e+03, threshold=3.860e+03, percent-clipped=4.0
+2023-03-08 15:31:37,429 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-08 15:31:42,116 INFO [train.py:968] (0/2) Epoch 16, batch 45250, giga_loss[loss=0.298, simple_loss=0.3697, pruned_loss=0.1132, over 28700.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3687, pruned_loss=0.1194, over 5668753.63 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3675, pruned_loss=0.1192, over 5684338.12 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3683, pruned_loss=0.1192, over 5667209.10 frames. ], batch size: 284, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:32:01,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2668, 1.3552, 1.3402, 1.5061], device='cuda:0'), covar=tensor([0.0803, 0.0366, 0.0333, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 15:32:04,613 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=729223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:32:25,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=729245.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:32:28,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=729248.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:32:30,938 INFO [train.py:968] (0/2) Epoch 16, batch 45300, giga_loss[loss=0.3008, simple_loss=0.3689, pruned_loss=0.1163, over 28846.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3689, pruned_loss=0.1198, over 5669948.84 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3671, pruned_loss=0.1189, over 5684761.97 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.369, pruned_loss=0.12, over 5667984.45 frames. ], batch size: 119, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:32:35,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3463, 3.1482, 3.0282, 1.4343], device='cuda:0'), covar=tensor([0.0894, 0.1063, 0.0950, 0.2259], device='cuda:0'), in_proj_covar=tensor([0.1167, 0.1091, 0.0936, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 15:32:55,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=729277.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:33:03,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.887e+02 1.749e+03 2.340e+03 3.124e+03 1.190e+04, threshold=4.679e+03, percent-clipped=12.0
+2023-03-08 15:33:15,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=729300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:33:17,283 INFO [train.py:968] (0/2) Epoch 16, batch 45350, giga_loss[loss=0.3179, simple_loss=0.3755, pruned_loss=0.1302, over 27867.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3705, pruned_loss=0.1215, over 5656472.28 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3677, pruned_loss=0.1192, over 5678884.79 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3701, pruned_loss=0.1213, over 5660582.28 frames. ], batch size: 412, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:33:19,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=729303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:33:19,689 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5990, 1.7655, 1.5783, 1.4888], device='cuda:0'), covar=tensor([0.1961, 0.2488, 0.2368, 0.2375], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0743, 0.0697, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 15:33:31,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4274, 2.3077, 2.2266, 2.0118], device='cuda:0'), covar=tensor([0.1626, 0.2395, 0.2000, 0.2196], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0742, 0.0697, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 15:33:45,077 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=729332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:34:04,423 INFO [train.py:968] (0/2) Epoch 16, batch 45400, giga_loss[loss=0.2948, simple_loss=0.3682, pruned_loss=0.1107, over 28571.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3725, pruned_loss=0.1233, over 5638852.37 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3681, pruned_loss=0.1195, over 5672244.26 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3719, pruned_loss=0.1229, over 5647280.13 frames. ], batch size: 78, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:34:41,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.735e+02 1.617e+03 2.092e+03 2.993e+03 6.833e+03, threshold=4.185e+03, percent-clipped=5.0
+2023-03-08 15:34:54,759 INFO [train.py:968] (0/2) Epoch 16, batch 45450, libri_loss[loss=0.2964, simple_loss=0.3692, pruned_loss=0.1119, over 29516.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3737, pruned_loss=0.1239, over 5634947.22 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3678, pruned_loss=0.1194, over 5667453.74 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3737, pruned_loss=0.1239, over 5645072.00 frames. ], batch size: 82, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:35:05,210 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6615, 1.9094, 1.8839, 1.6054], device='cuda:0'), covar=tensor([0.2413, 0.1864, 0.1379, 0.1679], device='cuda:0'), in_proj_covar=tensor([0.1866, 0.1808, 0.1735, 0.1871], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 15:35:40,035 INFO [train.py:968] (0/2) Epoch 16, batch 45500, giga_loss[loss=0.302, simple_loss=0.3704, pruned_loss=0.1168, over 28737.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3739, pruned_loss=0.1242, over 5636359.71 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3679, pruned_loss=0.1195, over 5663856.88 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3739, pruned_loss=0.1242, over 5647217.78 frames. ], batch size: 262, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:36:22,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.749e+03 2.295e+03 3.609e+03 1.299e+04, threshold=4.591e+03, percent-clipped=17.0
+2023-03-08 15:36:35,371 INFO [train.py:968] (0/2) Epoch 16, batch 45550, giga_loss[loss=0.3429, simple_loss=0.3796, pruned_loss=0.1531, over 23618.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3756, pruned_loss=0.1256, over 5641518.63 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.368, pruned_loss=0.1196, over 5664895.60 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3755, pruned_loss=0.1255, over 5648908.49 frames. ], batch size: 705, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:36:58,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2534, 1.3645, 1.1827, 1.4805], device='cuda:0'), covar=tensor([0.0810, 0.0356, 0.0357, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 15:37:27,813 INFO [train.py:968] (0/2) Epoch 16, batch 45600, giga_loss[loss=0.2903, simple_loss=0.3713, pruned_loss=0.1046, over 28679.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3761, pruned_loss=0.1247, over 5642546.68 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3679, pruned_loss=0.1196, over 5668665.68 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3762, pruned_loss=0.1247, over 5644702.68 frames. ], batch size: 307, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:37:42,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5438, 1.8789, 1.4458, 1.7271], device='cuda:0'), covar=tensor([0.2409, 0.2384, 0.2785, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.1422, 0.1034, 0.1258, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 15:38:05,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.667e+03 2.095e+03 3.269e+03 7.551e+03, threshold=4.190e+03, percent-clipped=9.0
+2023-03-08 15:38:15,824 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=729598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:38:19,562 INFO [train.py:968] (0/2) Epoch 16, batch 45650, giga_loss[loss=0.3115, simple_loss=0.3777, pruned_loss=0.1226, over 28502.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3756, pruned_loss=0.124, over 5639558.74 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1197, over 5655766.84 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3759, pruned_loss=0.1241, over 5651835.70 frames. ], batch size: 336, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:38:41,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8213, 1.8989, 1.3476, 1.4582], device='cuda:0'), covar=tensor([0.0847, 0.0654, 0.1042, 0.1155], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0448, 0.0509, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 15:39:09,221 INFO [train.py:968] (0/2) Epoch 16, batch 45700, giga_loss[loss=0.2855, simple_loss=0.3551, pruned_loss=0.1079, over 28953.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.375, pruned_loss=0.1246, over 5651255.89 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3678, pruned_loss=0.1196, over 5658314.64 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3754, pruned_loss=0.1247, over 5658409.57 frames. ], batch size: 164, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:39:38,627 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2497, 1.4733, 1.4158, 1.1477], device='cuda:0'), covar=tensor([0.2342, 0.2202, 0.1359, 0.1972], device='cuda:0'), in_proj_covar=tensor([0.1863, 0.1804, 0.1728, 0.1863], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 15:39:49,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.631e+03 2.130e+03 2.645e+03 6.869e+03, threshold=4.260e+03, percent-clipped=7.0
+2023-03-08 15:40:03,787 INFO [train.py:968] (0/2) Epoch 16, batch 45750, giga_loss[loss=0.4295, simple_loss=0.443, pruned_loss=0.208, over 26461.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3741, pruned_loss=0.1246, over 5649953.13 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3678, pruned_loss=0.1196, over 5660282.37 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3746, pruned_loss=0.1248, over 5653775.42 frames. ], batch size: 555, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:40:09,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4430, 1.6423, 1.5066, 1.5459], device='cuda:0'), covar=tensor([0.0791, 0.0324, 0.0310, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 15:40:41,394 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=729741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:40:44,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=729744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:40:51,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5417, 1.7569, 1.4030, 1.8426], device='cuda:0'), covar=tensor([0.2306, 0.2381, 0.2650, 0.2026], device='cuda:0'), in_proj_covar=tensor([0.1417, 0.1029, 0.1254, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 15:40:52,444 INFO [train.py:968] (0/2) Epoch 16, batch 45800, giga_loss[loss=0.3399, simple_loss=0.3947, pruned_loss=0.1425, over 28855.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3722, pruned_loss=0.1235, over 5664424.25 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3679, pruned_loss=0.1197, over 5663354.38 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3727, pruned_loss=0.1237, over 5664775.46 frames. ], batch size: 285, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:41:16,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=729773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:41:29,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2639, 1.5591, 1.2267, 0.9486], device='cuda:0'), covar=tensor([0.2443, 0.2395, 0.2709, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.1413, 0.1026, 0.1251, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 15:41:30,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.111e+03 1.651e+03 2.168e+03 3.110e+03 1.059e+04, threshold=4.336e+03, percent-clipped=12.0
+2023-03-08 15:41:42,837 INFO [train.py:968] (0/2) Epoch 16, batch 45850, giga_loss[loss=0.3035, simple_loss=0.3691, pruned_loss=0.119, over 28308.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3713, pruned_loss=0.1234, over 5658366.42 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3677, pruned_loss=0.1195, over 5669145.56 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.372, pruned_loss=0.1238, over 5653282.11 frames. ], batch size: 368, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:42:26,345 INFO [train.py:968] (0/2) Epoch 16, batch 45900, giga_loss[loss=0.3573, simple_loss=0.4036, pruned_loss=0.1555, over 27996.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3722, pruned_loss=0.1242, over 5666652.01 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3677, pruned_loss=0.1195, over 5673195.83 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3728, pruned_loss=0.1247, over 5658759.85 frames. ], batch size: 412, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:42:59,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.690e+03 2.001e+03 2.835e+03 5.842e+03, threshold=4.001e+03, percent-clipped=11.0
+2023-03-08 15:43:00,141 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8108, 2.2025, 1.9662, 1.6204], device='cuda:0'), covar=tensor([0.3028, 0.2191, 0.2424, 0.2660], device='cuda:0'), in_proj_covar=tensor([0.1874, 0.1814, 0.1740, 0.1871], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 15:43:10,283 INFO [train.py:968] (0/2) Epoch 16, batch 45950, giga_loss[loss=0.28, simple_loss=0.3529, pruned_loss=0.1036, over 28640.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3733, pruned_loss=0.125, over 5674368.49 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3677, pruned_loss=0.1194, over 5678560.84 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3739, pruned_loss=0.1257, over 5662780.88 frames. ], batch size: 336, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:43:37,209 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=729926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:44:04,047 INFO [train.py:968] (0/2) Epoch 16, batch 46000, giga_loss[loss=0.3544, simple_loss=0.399, pruned_loss=0.1549, over 27532.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3739, pruned_loss=0.126, over 5670665.75 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3678, pruned_loss=0.1197, over 5680658.49 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3744, pruned_loss=0.1263, over 5659682.43 frames. ], batch size: 472, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:44:35,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.391e+02 1.873e+03 2.330e+03 3.950e+03 1.361e+04, threshold=4.660e+03, percent-clipped=23.0
+2023-03-08 15:44:46,112 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-730000.pt
+2023-03-08 15:44:47,051 INFO [train.py:968] (0/2) Epoch 16, batch 46050, giga_loss[loss=0.3035, simple_loss=0.3643, pruned_loss=0.1213, over 28927.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3758, pruned_loss=0.1275, over 5670358.18 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1198, over 5685516.77 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3763, pruned_loss=0.1279, over 5656728.07 frames. ], batch size: 112, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:45:30,606 INFO [train.py:968] (0/2) Epoch 16, batch 46100, giga_loss[loss=0.3294, simple_loss=0.3853, pruned_loss=0.1368, over 28706.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3763, pruned_loss=0.128, over 5665471.13 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3682, pruned_loss=0.1199, over 5680163.60 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3767, pruned_loss=0.1284, over 5659939.89 frames. ], batch size: 242, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:46:09,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.656e+03 2.219e+03 3.103e+03 6.272e+03, threshold=4.438e+03, percent-clipped=6.0
+2023-03-08 15:46:20,595 INFO [train.py:968] (0/2) Epoch 16, batch 46150, giga_loss[loss=0.2827, simple_loss=0.3525, pruned_loss=0.1065, over 28471.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3774, pruned_loss=0.1294, over 5666334.54 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3681, pruned_loss=0.1198, over 5683980.41 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3779, pruned_loss=0.13, over 5658074.09 frames. ], batch size: 65, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:46:36,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4852, 3.5146, 1.5609, 1.6096], device='cuda:0'), covar=tensor([0.0968, 0.0417, 0.0908, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0538, 0.0364, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 15:47:03,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2619, 1.9103, 1.5518, 1.5294], device='cuda:0'), covar=tensor([0.0758, 0.0305, 0.0301, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 15:47:07,387 INFO [train.py:968] (0/2) Epoch 16, batch 46200, giga_loss[loss=0.3272, simple_loss=0.3665, pruned_loss=0.1439, over 23484.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3761, pruned_loss=0.1282, over 5658063.22 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3677, pruned_loss=0.1196, over 5686906.35 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3771, pruned_loss=0.129, over 5648531.23 frames. ], batch size: 705, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:47:18,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5270, 1.7341, 1.4150, 1.8583], device='cuda:0'), covar=tensor([0.2505, 0.2605, 0.2799, 0.2358], device='cuda:0'), in_proj_covar=tensor([0.1420, 0.1035, 0.1257, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 15:47:41,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.807e+02 1.529e+03 1.994e+03 3.048e+03 8.393e+03, threshold=3.988e+03, percent-clipped=6.0
+2023-03-08 15:47:51,506 INFO [train.py:968] (0/2) Epoch 16, batch 46250, giga_loss[loss=0.2969, simple_loss=0.367, pruned_loss=0.1134, over 28959.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3765, pruned_loss=0.1272, over 5674063.02 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3679, pruned_loss=0.1197, over 5693683.62 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3774, pruned_loss=0.128, over 5659507.41 frames. ], batch size: 112, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:48:14,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7401, 1.7669, 1.3129, 1.4199], device='cuda:0'), covar=tensor([0.0885, 0.0673, 0.1092, 0.1104], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0451, 0.0513, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 15:48:41,734 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=730249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:48:42,710 INFO [train.py:968] (0/2) Epoch 16, batch 46300, giga_loss[loss=0.3707, simple_loss=0.4151, pruned_loss=0.1632, over 28919.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3776, pruned_loss=0.128, over 5667104.94 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3681, pruned_loss=0.1201, over 5694030.97 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3783, pruned_loss=0.1286, over 5654876.96 frames. ], batch size: 213, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:49:07,862 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=730278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:49:19,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.482e+02 1.695e+03 2.524e+03 3.422e+03 9.029e+03, threshold=5.047e+03, percent-clipped=18.0
+2023-03-08 15:49:31,818 INFO [train.py:968] (0/2) Epoch 16, batch 46350, libri_loss[loss=0.3013, simple_loss=0.3677, pruned_loss=0.1175, over 29580.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3787, pruned_loss=0.1295, over 5660432.29 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3684, pruned_loss=0.1203, over 5701962.85 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3797, pruned_loss=0.1302, over 5641555.54 frames. ], batch size: 74, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:49:32,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=730301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:50:13,392 INFO [train.py:968] (0/2) Epoch 16, batch 46400, giga_loss[loss=0.3088, simple_loss=0.3721, pruned_loss=0.1228, over 28791.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.38, pruned_loss=0.1308, over 5663640.71 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3684, pruned_loss=0.1202, over 5709201.61 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3811, pruned_loss=0.1318, over 5640519.58 frames. ], batch size: 284, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:50:48,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.599e+03 2.137e+03 2.807e+03 9.004e+03, threshold=4.275e+03, percent-clipped=9.0
+2023-03-08 15:50:58,840 INFO [train.py:968] (0/2) Epoch 16, batch 46450, giga_loss[loss=0.2874, simple_loss=0.3605, pruned_loss=0.1071, over 28856.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3781, pruned_loss=0.1294, over 5656297.47 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3674, pruned_loss=0.1195, over 5717408.47 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3804, pruned_loss=0.1312, over 5627701.24 frames. ], batch size: 145, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:51:07,621 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-08 15:51:30,230 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=730433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:51:39,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=730444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:51:41,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=730447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:51:45,836 INFO [train.py:968] (0/2) Epoch 16, batch 46500, giga_loss[loss=0.3143, simple_loss=0.3752, pruned_loss=0.1267, over 27529.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.378, pruned_loss=0.1276, over 5651766.58 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3673, pruned_loss=0.1196, over 5702309.67 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3801, pruned_loss=0.1292, over 5641545.97 frames. ], batch size: 472, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:52:02,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4513, 1.8903, 1.3132, 0.7867], device='cuda:0'), covar=tensor([0.3467, 0.2153, 0.2507, 0.4745], device='cuda:0'), in_proj_covar=tensor([0.1679, 0.1592, 0.1553, 0.1370], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 15:52:12,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=730476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:52:25,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.165e+02 1.476e+03 1.885e+03 2.915e+03 1.104e+04, threshold=3.771e+03, percent-clipped=7.0
+2023-03-08 15:52:34,002 INFO [train.py:968] (0/2) Epoch 16, batch 46550, giga_loss[loss=0.2846, simple_loss=0.3558, pruned_loss=0.1067, over 28863.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.377, pruned_loss=0.1267, over 5634845.46 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3671, pruned_loss=0.1195, over 5699275.88 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3794, pruned_loss=0.1285, over 5626676.26 frames. ], batch size: 186, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:52:46,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4474, 2.0884, 1.4800, 0.6738], device='cuda:0'), covar=tensor([0.5464, 0.2736, 0.3608, 0.5839], device='cuda:0'), in_proj_covar=tensor([0.1681, 0.1590, 0.1551, 0.1370], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 15:52:58,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6400, 2.1561, 1.4734, 2.1568], device='cuda:0'), covar=tensor([0.2607, 0.2478, 0.2901, 0.2237], device='cuda:0'), in_proj_covar=tensor([0.1419, 0.1035, 0.1257, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 15:53:06,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3568, 1.4270, 1.3420, 1.5870], device='cuda:0'), covar=tensor([0.0782, 0.0331, 0.0314, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 15:53:25,022 INFO [train.py:968] (0/2) Epoch 16, batch 46600, giga_loss[loss=0.2918, simple_loss=0.359, pruned_loss=0.1123, over 28679.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3744, pruned_loss=0.1245, over 5646853.65 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3666, pruned_loss=0.1191, over 5699703.49 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3769, pruned_loss=0.1263, over 5638921.42 frames. ], batch size: 92, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:53:50,803 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-08 15:54:04,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.565e+03 2.143e+03 3.245e+03 9.822e+03, threshold=4.287e+03, percent-clipped=17.0
+2023-03-08 15:54:12,919 INFO [train.py:968] (0/2) Epoch 16, batch 46650, giga_loss[loss=0.3683, simple_loss=0.4128, pruned_loss=0.1619, over 27645.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3741, pruned_loss=0.1246, over 5630348.90 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3669, pruned_loss=0.1192, over 5685059.61 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.376, pruned_loss=0.1261, over 5635838.50 frames. ], batch size: 472, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:54:34,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=730624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:54:35,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3510, 1.1549, 3.9925, 3.2551], device='cuda:0'), covar=tensor([0.1667, 0.2831, 0.0464, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0722, 0.0621, 0.0921, 0.0844], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 15:55:00,326 INFO [train.py:968] (0/2) Epoch 16, batch 46700, giga_loss[loss=0.4404, simple_loss=0.446, pruned_loss=0.2174, over 26620.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3761, pruned_loss=0.126, over 5632587.85 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3672, pruned_loss=0.1197, over 5678999.42 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3775, pruned_loss=0.127, over 5640816.35 frames. ], batch size: 555, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:55:02,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=730653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:55:03,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=730654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:55:38,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.847e+03 2.387e+03 3.468e+03 6.448e+03, threshold=4.774e+03, percent-clipped=10.0
+2023-03-08 15:55:48,475 INFO [train.py:968] (0/2) Epoch 16, batch 46750, giga_loss[loss=0.291, simple_loss=0.3537, pruned_loss=0.1141, over 28858.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3765, pruned_loss=0.1262, over 5621020.01 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3676, pruned_loss=0.12, over 5656452.81 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3775, pruned_loss=0.1267, over 5647557.85 frames. ], batch size: 112, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:56:33,351 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 15:56:38,807 INFO [train.py:968] (0/2) Epoch 16, batch 46800, giga_loss[loss=0.2754, simple_loss=0.3456, pruned_loss=0.1026, over 28751.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3778, pruned_loss=0.1287, over 5589698.70 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3682, pruned_loss=0.1206, over 5619934.84 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3781, pruned_loss=0.1286, over 5643793.58 frames. ], batch size: 284, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:56:57,150 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=730767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:00,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=730770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:01,825 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=730772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:18,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.617e+02 1.629e+03 2.143e+03 2.810e+03 1.127e+04, threshold=4.286e+03, percent-clipped=5.0
+2023-03-08 15:57:23,984 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=730796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:27,461 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=730799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:27,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=730799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:29,215 INFO [train.py:968] (0/2) Epoch 16, batch 46850, giga_loss[loss=0.457, simple_loss=0.4632, pruned_loss=0.2254, over 26630.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3787, pruned_loss=0.1294, over 5569635.48 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3687, pruned_loss=0.121, over 5576256.73 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3786, pruned_loss=0.1291, over 5649980.85 frames. ], batch size: 555, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:57:37,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=730808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:44,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4138, 4.2348, 4.0358, 2.0172], device='cuda:0'), covar=tensor([0.0555, 0.0668, 0.0721, 0.1913], device='cuda:0'), in_proj_covar=tensor([0.1170, 0.1090, 0.0939, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 15:57:55,249 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=730828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:58:17,788 INFO [train.py:968] (0/2) Epoch 16, batch 46900, giga_loss[loss=0.3467, simple_loss=0.41, pruned_loss=0.1417, over 29082.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3807, pruned_loss=0.1309, over 5548234.73 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3692, pruned_loss=0.1214, over 5535987.36 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3803, pruned_loss=0.1304, over 5649371.53 frames. ], batch size: 155, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:58:33,280 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-08 15:58:45,478 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-08 15:58:47,926 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-16.pt
+2023-03-08 15:59:33,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.922e+02 1.609e+03 1.974e+03 2.877e+03 1.091e+04, threshold=3.948e+03, percent-clipped=8.0
+2023-03-08 16:00:07,036 INFO [train.py:968] (0/2) Epoch 17, batch 50, giga_loss[loss=0.2915, simple_loss=0.3776, pruned_loss=0.1027, over 28851.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3786, pruned_loss=0.1118, over 1268170.11 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3514, pruned_loss=0.09364, over 204004.82 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3836, pruned_loss=0.1151, over 1102653.71 frames. ], batch size: 174, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:00:25,734 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=730951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:00:28,559 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=730954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:00:54,498 INFO [train.py:968] (0/2) Epoch 17, batch 100, giga_loss[loss=0.2577, simple_loss=0.3313, pruned_loss=0.09205, over 28868.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3696, pruned_loss=0.1082, over 2243743.09 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3497, pruned_loss=0.09335, over 363486.26 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3729, pruned_loss=0.1107, over 2008358.21 frames. ], batch size: 99, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:00:57,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=730983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:01:04,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.242e+02 1.234e+03 1.474e+03 1.900e+03 4.144e+03, threshold=2.948e+03, percent-clipped=1.0
+2023-03-08 16:01:31,845 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=731023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:01:36,818 INFO [train.py:968] (0/2) Epoch 17, batch 150, giga_loss[loss=0.2384, simple_loss=0.3152, pruned_loss=0.08084, over 28839.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3545, pruned_loss=0.1006, over 3006265.61 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.346, pruned_loss=0.09184, over 598508.65 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.357, pruned_loss=0.1027, over 2690038.06 frames. ], batch size: 186, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:01:37,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=731029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:01:54,373 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-08 16:02:16,369 INFO [train.py:968] (0/2) Epoch 17, batch 200, giga_loss[loss=0.2591, simple_loss=0.3288, pruned_loss=0.09471, over 28848.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3415, pruned_loss=0.09428, over 3615547.22 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3449, pruned_loss=0.09093, over 858489.26 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3422, pruned_loss=0.09572, over 3238910.74 frames. ], batch size: 186, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:02:27,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.939e+02 1.043e+03 1.280e+03 1.736e+03 4.450e+03, threshold=2.561e+03, percent-clipped=6.0
+2023-03-08 16:02:38,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2203, 1.4861, 1.6337, 1.2847], device='cuda:0'), covar=tensor([0.1939, 0.1784, 0.2319, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0744, 0.0703, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 16:02:57,242 INFO [train.py:968] (0/2) Epoch 17, batch 250, giga_loss[loss=0.1936, simple_loss=0.2683, pruned_loss=0.05948, over 28533.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3326, pruned_loss=0.09025, over 4072951.23 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3464, pruned_loss=0.09301, over 973402.41 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3316, pruned_loss=0.09039, over 3735400.62 frames. ], batch size: 71, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:03:14,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=731147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:03:14,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9081, 2.1277, 1.4615, 1.6369], device='cuda:0'), covar=tensor([0.1006, 0.0733, 0.1102, 0.1294], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0446, 0.0509, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 16:03:20,520 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=731154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:03:25,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3927, 4.2171, 3.9874, 1.8801], device='cuda:0'), covar=tensor([0.0530, 0.0760, 0.0736, 0.2070], device='cuda:0'), in_proj_covar=tensor([0.1158, 0.1078, 0.0929, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 16:03:35,325 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=731172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:03:39,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=731175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:03:42,138 INFO [train.py:968] (0/2) Epoch 17, batch 300, giga_loss[loss=0.2165, simple_loss=0.2873, pruned_loss=0.07282, over 28667.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3229, pruned_loss=0.08535, over 4438513.03 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.344, pruned_loss=0.0913, over 1140862.51 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3212, pruned_loss=0.08528, over 4117110.13 frames. ], batch size: 92, lr: 1.91e-03, grad_scale: 2.0
+2023-03-08 16:03:51,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.273e+02 1.029e+03 1.311e+03 1.951e+03 5.102e+03, threshold=2.622e+03, percent-clipped=5.0
+2023-03-08 16:04:01,470 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=731204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:04:26,255 INFO [train.py:968] (0/2) Epoch 17, batch 350, giga_loss[loss=0.2238, simple_loss=0.3045, pruned_loss=0.07156, over 28853.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3155, pruned_loss=0.08227, over 4716538.64 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3426, pruned_loss=0.09086, over 1293160.73 frames. ], giga_tot_loss[loss=0.2385, simple_loss=0.3133, pruned_loss=0.08189, over 4426295.24 frames. ], batch size: 199, lr: 1.91e-03, grad_scale: 2.0
+2023-03-08 16:05:09,509 INFO [train.py:968] (0/2) Epoch 17, batch 400, giga_loss[loss=0.2501, simple_loss=0.318, pruned_loss=0.09116, over 28873.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3116, pruned_loss=0.08093, over 4936153.18 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3427, pruned_loss=0.09063, over 1360993.39 frames. ], giga_tot_loss[loss=0.2351, simple_loss=0.3093, pruned_loss=0.08048, over 4693304.44 frames. ], batch size: 112, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:05:10,768 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=731280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:05:18,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=731290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:05:20,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.798e+02 1.091e+03 1.388e+03 2.352e+03 8.946e+03, threshold=2.775e+03, percent-clipped=18.0
+2023-03-08 16:05:21,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=731293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:05:47,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=731322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:05:51,564 INFO [train.py:968] (0/2) Epoch 17, batch 450, giga_loss[loss=0.1999, simple_loss=0.2776, pruned_loss=0.06106, over 28972.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3085, pruned_loss=0.07941, over 5107290.04 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3429, pruned_loss=0.09058, over 1444293.85 frames. ], giga_tot_loss[loss=0.2318, simple_loss=0.3059, pruned_loss=0.07883, over 4904534.03 frames. ], batch size: 213, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:06:25,859 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 16:06:28,839 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.15 vs. limit=5.0
+2023-03-08 16:06:33,683 INFO [train.py:968] (0/2) Epoch 17, batch 500, giga_loss[loss=0.2475, simple_loss=0.3121, pruned_loss=0.09146, over 28898.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3057, pruned_loss=0.07793, over 5249589.41 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3431, pruned_loss=0.09111, over 1596436.22 frames. ], giga_tot_loss[loss=0.228, simple_loss=0.3023, pruned_loss=0.07686, over 5065042.39 frames. ], batch size: 186, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:06:38,292 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-08 16:06:42,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5383, 1.6868, 1.7840, 1.3741], device='cuda:0'), covar=tensor([0.1832, 0.2623, 0.1528, 0.1778], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0700, 0.0925, 0.0822], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 16:06:43,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.183e+02 1.019e+03 1.298e+03 1.850e+03 5.843e+03, threshold=2.596e+03, percent-clipped=5.0
+2023-03-08 16:06:49,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=731398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:07:11,218 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-08 16:07:18,493 INFO [train.py:968] (0/2) Epoch 17, batch 550, libri_loss[loss=0.2885, simple_loss=0.365, pruned_loss=0.106, over 25989.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3042, pruned_loss=0.07741, over 5337190.46 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3422, pruned_loss=0.09063, over 1728539.97 frames. ], giga_tot_loss[loss=0.2265, simple_loss=0.3005, pruned_loss=0.07627, over 5183712.38 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:07:50,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5013, 1.6781, 1.6278, 1.4739], device='cuda:0'), covar=tensor([0.2539, 0.2239, 0.1576, 0.2038], device='cuda:0'), in_proj_covar=tensor([0.1860, 0.1794, 0.1713, 0.1850], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 16:07:56,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3451, 1.5899, 1.6251, 1.2138], device='cuda:0'), covar=tensor([0.1796, 0.2487, 0.1521, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.0881, 0.0702, 0.0926, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 16:08:03,823 INFO [train.py:968] (0/2) Epoch 17, batch 600, libri_loss[loss=0.2359, simple_loss=0.324, pruned_loss=0.07392, over 29536.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3024, pruned_loss=0.07653, over 5423295.12 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3408, pruned_loss=0.08958, over 1831544.18 frames. ], giga_tot_loss[loss=0.225, simple_loss=0.2989, pruned_loss=0.0756, over 5289452.79 frames. ], batch size: 79, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:08:15,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.698e+02 9.804e+02 1.327e+03 1.660e+03 2.968e+03, threshold=2.653e+03, percent-clipped=3.0
+2023-03-08 16:08:52,494 INFO [train.py:968] (0/2) Epoch 17, batch 650, giga_loss[loss=0.201, simple_loss=0.2772, pruned_loss=0.06237, over 28717.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3004, pruned_loss=0.07607, over 5477489.38 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3427, pruned_loss=0.09078, over 1891289.31 frames. ], giga_tot_loss[loss=0.223, simple_loss=0.2965, pruned_loss=0.07477, over 5366745.24 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 2.0
+2023-03-08 16:08:52,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=731529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:09:06,436 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=731541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:09:08,986 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=731544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:09:33,306 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-08 16:09:35,761 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=731573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:09:40,210 INFO [train.py:968] (0/2) Epoch 17, batch 700, giga_loss[loss=0.1989, simple_loss=0.2768, pruned_loss=0.06045, over 29052.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.2973, pruned_loss=0.07483, over 5524609.48 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3437, pruned_loss=0.09145, over 1910977.86 frames. ], giga_tot_loss[loss=0.2204, simple_loss=0.2938, pruned_loss=0.07353, over 5436538.98 frames. ], batch size: 128, lr: 1.91e-03, grad_scale: 2.0
+2023-03-08 16:09:53,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.897e+02 1.172e+03 1.757e+03 2.780e+03 6.889e+03, threshold=3.515e+03, percent-clipped=28.0
+2023-03-08 16:09:55,826 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 16:10:26,371 INFO [train.py:968] (0/2) Epoch 17, batch 750, giga_loss[loss=0.1825, simple_loss=0.2596, pruned_loss=0.05269, over 29053.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2952, pruned_loss=0.0737, over 5567996.33 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3442, pruned_loss=0.09152, over 2006436.49 frames. ], giga_tot_loss[loss=0.2178, simple_loss=0.2911, pruned_loss=0.07226, over 5492464.87 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 2.0
+2023-03-08 16:10:51,065 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=731655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:11:03,758 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=731672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:11:06,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=731675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:11:09,393 INFO [train.py:968] (0/2) Epoch 17, batch 800, giga_loss[loss=0.258, simple_loss=0.3216, pruned_loss=0.09721, over 28700.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2928, pruned_loss=0.07259, over 5601889.08 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3447, pruned_loss=0.0917, over 2099166.40 frames. ], giga_tot_loss[loss=0.2151, simple_loss=0.2882, pruned_loss=0.07094, over 5537616.08 frames. ], batch size: 284, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:11:21,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.008e+02 9.854e+02 1.295e+03 1.669e+03 8.197e+03, threshold=2.589e+03, percent-clipped=5.0
+2023-03-08 16:11:33,288 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=731704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:11:57,828 INFO [train.py:968] (0/2) Epoch 17, batch 850, giga_loss[loss=0.2857, simple_loss=0.3458, pruned_loss=0.1128, over 28714.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.2985, pruned_loss=0.07612, over 5609708.41 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.344, pruned_loss=0.09125, over 2191379.64 frames. ], giga_tot_loss[loss=0.2215, simple_loss=0.2939, pruned_loss=0.07458, over 5553198.47 frames. ], batch size: 242, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:12:50,588 INFO [train.py:968] (0/2) Epoch 17, batch 900, giga_loss[loss=0.3525, simple_loss=0.3941, pruned_loss=0.1555, over 23666.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3125, pruned_loss=0.08362, over 5625976.25 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3442, pruned_loss=0.09126, over 2228178.30 frames. ], giga_tot_loss[loss=0.2367, simple_loss=0.3086, pruned_loss=0.08236, over 5579212.21 frames. ], batch size: 705, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:13:02,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.564e+02 1.202e+03 1.528e+03 2.256e+03 5.144e+03, threshold=3.056e+03, percent-clipped=19.0
+2023-03-08 16:13:06,571 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=731798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:13:08,913 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=731801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:13:34,037 INFO [train.py:968] (0/2) Epoch 17, batch 950, giga_loss[loss=0.296, simple_loss=0.3766, pruned_loss=0.1077, over 29019.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3242, pruned_loss=0.08894, over 5650578.43 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3444, pruned_loss=0.09138, over 2300894.75 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3206, pruned_loss=0.08784, over 5609066.00 frames. ], batch size: 164, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:13:36,965 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=731830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:13:48,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4778, 2.0486, 1.4038, 0.6374], device='cuda:0'), covar=tensor([0.5822, 0.2790, 0.4014, 0.6262], device='cuda:0'), in_proj_covar=tensor([0.1667, 0.1573, 0.1544, 0.1358], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 16:14:17,016 INFO [train.py:968] (0/2) Epoch 17, batch 1000, giga_loss[loss=0.2967, simple_loss=0.3692, pruned_loss=0.1121, over 28863.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3332, pruned_loss=0.09315, over 5665392.37 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3448, pruned_loss=0.09134, over 2407119.14 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3299, pruned_loss=0.09234, over 5625629.61 frames. ], batch size: 186, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:14:18,416 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=731881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:14:27,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.001e+02 1.239e+03 1.572e+03 2.207e+03 6.868e+03, threshold=3.143e+03, percent-clipped=13.0
+2023-03-08 16:14:48,927 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-08 16:14:55,689 INFO [train.py:968] (0/2) Epoch 17, batch 1050, giga_loss[loss=0.2546, simple_loss=0.3434, pruned_loss=0.08295, over 29031.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3379, pruned_loss=0.09424, over 5681382.41 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3448, pruned_loss=0.09188, over 2560380.99 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3348, pruned_loss=0.09346, over 5640771.48 frames. ], batch size: 164, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:15:31,099 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 16:15:44,716 INFO [train.py:968] (0/2) Epoch 17, batch 1100, libri_loss[loss=0.2348, simple_loss=0.3102, pruned_loss=0.07964, over 29657.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3397, pruned_loss=0.09404, over 5678382.21 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3435, pruned_loss=0.09139, over 2627252.37 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3378, pruned_loss=0.09369, over 5641401.46 frames. ], batch size: 69, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:15:54,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.849e+02 1.167e+03 1.462e+03 1.850e+03 6.845e+03, threshold=2.923e+03, percent-clipped=3.0
+2023-03-08 16:16:01,032 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-732000.pt
+2023-03-08 16:16:23,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5560, 2.2090, 1.8397, 1.7490], device='cuda:0'), covar=tensor([0.0765, 0.0260, 0.0285, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 16:16:27,113 INFO [train.py:968] (0/2) Epoch 17, batch 1150, giga_loss[loss=0.2788, simple_loss=0.3517, pruned_loss=0.103, over 28687.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3407, pruned_loss=0.09423, over 5692578.09 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3432, pruned_loss=0.09112, over 2674729.38 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3393, pruned_loss=0.09411, over 5661815.14 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:17:10,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5455, 1.6325, 1.2042, 1.2271], device='cuda:0'), covar=tensor([0.0902, 0.0542, 0.0992, 0.1181], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0445, 0.0511, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 16:17:13,244 INFO [train.py:968] (0/2) Epoch 17, batch 1200, giga_loss[loss=0.2721, simple_loss=0.3434, pruned_loss=0.1005, over 28510.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3448, pruned_loss=0.09744, over 5681697.14 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.343, pruned_loss=0.09079, over 2719535.17 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3438, pruned_loss=0.09759, over 5657526.08 frames. ], batch size: 60, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:17:24,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5933, 2.6179, 2.3598, 2.4384], device='cuda:0'), covar=tensor([0.1705, 0.2090, 0.1969, 0.1729], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0741, 0.0699, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 16:17:25,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.714e+02 1.242e+03 1.589e+03 2.092e+03 5.718e+03, threshold=3.178e+03, percent-clipped=9.0
+2023-03-08 16:17:56,217 INFO [train.py:968] (0/2) Epoch 17, batch 1250, libri_loss[loss=0.2211, simple_loss=0.3012, pruned_loss=0.07045, over 29318.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3475, pruned_loss=0.09899, over 5688700.37 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3429, pruned_loss=0.09068, over 2829208.20 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3469, pruned_loss=0.09945, over 5662180.91 frames. ], batch size: 71, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:18:33,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7924, 1.9929, 1.6549, 1.8275], device='cuda:0'), covar=tensor([0.0752, 0.0280, 0.0299, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 16:18:34,584 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 16:18:40,021 INFO [train.py:968] (0/2) Epoch 17, batch 1300, giga_loss[loss=0.2759, simple_loss=0.3636, pruned_loss=0.09409, over 28961.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3505, pruned_loss=0.1004, over 5683387.37 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3428, pruned_loss=0.09034, over 2902093.48 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3502, pruned_loss=0.1012, over 5660504.00 frames. ], batch size: 186, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:18:52,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.460e+02 1.287e+03 1.617e+03 2.262e+03 4.717e+03, threshold=3.234e+03, percent-clipped=11.0
+2023-03-08 16:19:22,149 INFO [train.py:968] (0/2) Epoch 17, batch 1350, libri_loss[loss=0.2656, simple_loss=0.3529, pruned_loss=0.08913, over 29533.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.352, pruned_loss=0.1001, over 5700159.75 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3426, pruned_loss=0.09024, over 2932113.19 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3519, pruned_loss=0.1008, over 5680133.46 frames. ], batch size: 82, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:19:41,162 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7825, 5.0431, 1.8997, 2.1100], device='cuda:0'), covar=tensor([0.0870, 0.0184, 0.0859, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0532, 0.0363, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 16:19:44,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=732256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:20:04,956 INFO [train.py:968] (0/2) Epoch 17, batch 1400, giga_loss[loss=0.2601, simple_loss=0.3531, pruned_loss=0.08354, over 29013.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3539, pruned_loss=0.1008, over 5695279.05 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3424, pruned_loss=0.08996, over 3016700.59 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3542, pruned_loss=0.1018, over 5676660.30 frames. ], batch size: 164, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:20:15,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.232e+02 1.173e+03 1.511e+03 2.067e+03 4.256e+03, threshold=3.021e+03, percent-clipped=4.0
+2023-03-08 16:20:35,128 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 16:20:45,541 INFO [train.py:968] (0/2) Epoch 17, batch 1450, giga_loss[loss=0.2533, simple_loss=0.3365, pruned_loss=0.08502, over 28890.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3531, pruned_loss=0.09938, over 5699589.03 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3424, pruned_loss=0.08989, over 3074214.48 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3536, pruned_loss=0.1004, over 5680722.27 frames. ], batch size: 186, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:21:08,367 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=732356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:21:29,473 INFO [train.py:968] (0/2) Epoch 17, batch 1500, giga_loss[loss=0.2528, simple_loss=0.338, pruned_loss=0.08379, over 28643.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3511, pruned_loss=0.09668, over 5697646.16 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3424, pruned_loss=0.08991, over 3092150.49 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3515, pruned_loss=0.09753, over 5689509.17 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:21:42,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.639e+02 1.060e+03 1.280e+03 1.729e+03 3.186e+03, threshold=2.561e+03, percent-clipped=2.0
+2023-03-08 16:21:46,098 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=732399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:21:47,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=732402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:21:52,075 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=732408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:22:08,972 INFO [train.py:968] (0/2) Epoch 17, batch 1550, libri_loss[loss=0.2872, simple_loss=0.3644, pruned_loss=0.105, over 29226.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3491, pruned_loss=0.0948, over 5703285.75 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3423, pruned_loss=0.08975, over 3188902.35 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3498, pruned_loss=0.09575, over 5690379.75 frames. ], batch size: 97, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:22:10,756 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=732431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:22:52,609 INFO [train.py:968] (0/2) Epoch 17, batch 1600, giga_loss[loss=0.2657, simple_loss=0.3414, pruned_loss=0.09498, over 28796.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3497, pruned_loss=0.09635, over 5712015.74 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3422, pruned_loss=0.08967, over 3254506.86 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3505, pruned_loss=0.0973, over 5698589.90 frames. ], batch size: 112, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:23:06,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.009e+02 1.246e+03 1.560e+03 2.612e+03 1.466e+04, threshold=3.120e+03, percent-clipped=26.0
+2023-03-08 16:23:13,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5246, 1.6376, 1.7040, 1.4955], device='cuda:0'), covar=tensor([0.1642, 0.1898, 0.1943, 0.1829], device='cuda:0'), in_proj_covar=tensor([0.0454, 0.0734, 0.0692, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 16:23:36,017 INFO [train.py:968] (0/2) Epoch 17, batch 1650, giga_loss[loss=0.371, simple_loss=0.3985, pruned_loss=0.1717, over 27995.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3525, pruned_loss=0.1008, over 5719506.49 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3422, pruned_loss=0.08991, over 3357399.24 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3534, pruned_loss=0.1018, over 5702960.99 frames. ], batch size: 412, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:24:23,850 INFO [train.py:968] (0/2) Epoch 17, batch 1700, giga_loss[loss=0.3634, simple_loss=0.3957, pruned_loss=0.1656, over 26535.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3539, pruned_loss=0.1037, over 5704514.18 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3422, pruned_loss=0.08985, over 3369863.25 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3548, pruned_loss=0.1046, over 5690914.51 frames. ], batch size: 555, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:24:37,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.987e+02 1.299e+03 1.544e+03 1.993e+03 3.539e+03, threshold=3.087e+03, percent-clipped=4.0
+2023-03-08 16:25:07,689 INFO [train.py:968] (0/2) Epoch 17, batch 1750, giga_loss[loss=0.2629, simple_loss=0.3404, pruned_loss=0.09269, over 29013.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3524, pruned_loss=0.1033, over 5704536.18 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3422, pruned_loss=0.08978, over 3455455.69 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3534, pruned_loss=0.1044, over 5689108.69 frames. ], batch size: 164, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:25:47,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0033, 1.1489, 1.0896, 0.9493], device='cuda:0'), covar=tensor([0.1993, 0.2422, 0.1354, 0.1810], device='cuda:0'), in_proj_covar=tensor([0.1842, 0.1789, 0.1714, 0.1854], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 16:25:51,748 INFO [train.py:968] (0/2) Epoch 17, batch 1800, giga_loss[loss=0.2945, simple_loss=0.3655, pruned_loss=0.1118, over 28937.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.351, pruned_loss=0.1027, over 5713206.73 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3433, pruned_loss=0.09044, over 3512300.78 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3513, pruned_loss=0.1035, over 5699775.85 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:26:03,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.089e+02 1.255e+03 1.627e+03 2.120e+03 8.682e+03, threshold=3.253e+03, percent-clipped=11.0
+2023-03-08 16:26:18,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3309, 1.1799, 4.0802, 3.2105], device='cuda:0'), covar=tensor([0.1701, 0.2839, 0.0416, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0714, 0.0616, 0.0905, 0.0832], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 16:26:32,320 INFO [train.py:968] (0/2) Epoch 17, batch 1850, giga_loss[loss=0.2607, simple_loss=0.3467, pruned_loss=0.08736, over 28704.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3506, pruned_loss=0.1022, over 5707853.39 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.343, pruned_loss=0.09019, over 3562066.54 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3512, pruned_loss=0.1032, over 5701746.16 frames. ], batch size: 65, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:26:34,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=732731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:27:16,780 INFO [train.py:968] (0/2) Epoch 17, batch 1900, giga_loss[loss=0.2906, simple_loss=0.3527, pruned_loss=0.1142, over 26558.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3498, pruned_loss=0.1008, over 5707880.03 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.343, pruned_loss=0.09007, over 3618043.50 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3505, pruned_loss=0.1019, over 5700532.32 frames. ], batch size: 555, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:27:22,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=732783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:27:30,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0013, 2.2623, 2.0189, 1.9185], device='cuda:0'), covar=tensor([0.2059, 0.2319, 0.2294, 0.2318], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0738, 0.0694, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 16:27:31,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.608e+02 1.137e+03 1.430e+03 1.955e+03 3.255e+03, threshold=2.860e+03, percent-clipped=1.0
+2023-03-08 16:27:55,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.7641, 1.1248, 1.1230, 0.9706], device='cuda:0'), covar=tensor([0.2084, 0.1500, 0.2379, 0.1769], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0738, 0.0694, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 16:28:04,035 INFO [train.py:968] (0/2) Epoch 17, batch 1950, giga_loss[loss=0.275, simple_loss=0.3514, pruned_loss=0.09927, over 29028.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3475, pruned_loss=0.09947, over 5697931.87 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3434, pruned_loss=0.09032, over 3671315.62 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3479, pruned_loss=0.1005, over 5690490.26 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:28:09,935 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=732835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:28:47,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2100, 1.5255, 1.5478, 1.1136], device='cuda:0'), covar=tensor([0.1705, 0.2478, 0.1372, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.0876, 0.0698, 0.0924, 0.0821], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 16:28:48,818 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=732874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:28:51,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=732877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:28:53,118 INFO [train.py:968] (0/2) Epoch 17, batch 2000, giga_loss[loss=0.2327, simple_loss=0.3095, pruned_loss=0.07795, over 28926.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3413, pruned_loss=0.09599, over 5679911.84 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3438, pruned_loss=0.09057, over 3683365.14 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3414, pruned_loss=0.09671, over 5680749.86 frames. ], batch size: 213, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:28:56,267 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-08 16:29:07,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.962e+02 9.809e+02 1.189e+03 1.797e+03 4.693e+03, threshold=2.378e+03, percent-clipped=9.0
+2023-03-08 16:29:17,149 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=732906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:29:36,669 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=732926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:29:38,347 INFO [train.py:968] (0/2) Epoch 17, batch 2050, giga_loss[loss=0.208, simple_loss=0.2806, pruned_loss=0.06767, over 28337.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3359, pruned_loss=0.09334, over 5675278.68 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3442, pruned_loss=0.0907, over 3741878.97 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3357, pruned_loss=0.09394, over 5676556.21 frames. ], batch size: 65, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:29:39,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=732929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:29:43,433 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=732934.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 16:29:51,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2620, 4.1042, 3.8468, 1.8387], device='cuda:0'), covar=tensor([0.0514, 0.0631, 0.0663, 0.2245], device='cuda:0'), in_proj_covar=tensor([0.1135, 0.1054, 0.0906, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 16:30:05,341 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=732958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:30:25,724 INFO [train.py:968] (0/2) Epoch 17, batch 2100, giga_loss[loss=0.2631, simple_loss=0.3409, pruned_loss=0.09268, over 28787.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3329, pruned_loss=0.09161, over 5673338.16 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.344, pruned_loss=0.09071, over 3794855.72 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3326, pruned_loss=0.09214, over 5677807.75 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:30:38,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.137e+02 1.098e+03 1.400e+03 2.039e+03 5.378e+03, threshold=2.799e+03, percent-clipped=17.0
+2023-03-08 16:31:06,666 INFO [train.py:968] (0/2) Epoch 17, batch 2150, giga_loss[loss=0.2663, simple_loss=0.3412, pruned_loss=0.09572, over 28727.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3341, pruned_loss=0.09184, over 5672717.83 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3448, pruned_loss=0.09099, over 3835062.46 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3332, pruned_loss=0.09211, over 5681996.00 frames. ], batch size: 119, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:31:46,780 INFO [train.py:968] (0/2) Epoch 17, batch 2200, giga_loss[loss=0.26, simple_loss=0.337, pruned_loss=0.09153, over 28317.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3343, pruned_loss=0.09164, over 5684563.30 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3456, pruned_loss=0.0914, over 3885937.85 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3328, pruned_loss=0.0916, over 5687276.18 frames. ], batch size: 368, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:31:50,069 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-08 16:32:02,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.686e+02 9.582e+02 1.234e+03 1.808e+03 5.505e+03, threshold=2.469e+03, percent-clipped=5.0
+2023-03-08 16:32:28,942 INFO [train.py:968] (0/2) Epoch 17, batch 2250, giga_loss[loss=0.2377, simple_loss=0.3143, pruned_loss=0.08053, over 29024.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3317, pruned_loss=0.09035, over 5694214.62 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.345, pruned_loss=0.09107, over 3925889.71 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3306, pruned_loss=0.09052, over 5692584.01 frames. ], batch size: 155, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:33:11,305 INFO [train.py:968] (0/2) Epoch 17, batch 2300, libri_loss[loss=0.2579, simple_loss=0.3544, pruned_loss=0.08066, over 28610.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3291, pruned_loss=0.08905, over 5697462.64 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3451, pruned_loss=0.09074, over 3980015.37 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3278, pruned_loss=0.08936, over 5694550.88 frames. ], batch size: 106, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:33:25,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.431e+02 1.001e+03 1.292e+03 1.705e+03 3.309e+03, threshold=2.584e+03, percent-clipped=5.0
+2023-03-08 16:33:35,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=733210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:33:52,310 INFO [train.py:968] (0/2) Epoch 17, batch 2350, giga_loss[loss=0.2479, simple_loss=0.3119, pruned_loss=0.09196, over 28345.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3277, pruned_loss=0.08828, over 5712089.23 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3457, pruned_loss=0.09079, over 4036542.12 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3258, pruned_loss=0.08845, over 5704880.94 frames. ], batch size: 65, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:33:53,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4873, 1.5128, 1.2812, 1.1215], device='cuda:0'), covar=tensor([0.0890, 0.0554, 0.1003, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0441, 0.0506, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 16:34:15,727 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-08 16:34:33,788 INFO [train.py:968] (0/2) Epoch 17, batch 2400, giga_loss[loss=0.2271, simple_loss=0.3028, pruned_loss=0.07564, over 28660.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.326, pruned_loss=0.0877, over 5710991.97 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3465, pruned_loss=0.09108, over 4072802.92 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3237, pruned_loss=0.0876, over 5710732.61 frames. ], batch size: 60, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:34:48,212 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.634e+02 1.001e+03 1.281e+03 1.717e+03 4.849e+03, threshold=2.562e+03, percent-clipped=8.0
+2023-03-08 16:34:49,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=733297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:34:58,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=733309.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 16:35:00,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-08 16:35:14,321 INFO [train.py:968] (0/2) Epoch 17, batch 2450, giga_loss[loss=0.2188, simple_loss=0.2991, pruned_loss=0.06923, over 28918.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3245, pruned_loss=0.08692, over 5720574.36 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3471, pruned_loss=0.09115, over 4134050.31 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3215, pruned_loss=0.08665, over 5716205.44 frames. ], batch size: 164, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:35:16,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7689, 1.8245, 1.4048, 1.3399], device='cuda:0'), covar=tensor([0.0879, 0.0587, 0.0965, 0.1185], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0440, 0.0505, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 16:35:19,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0848, 1.1264, 3.8286, 3.1683], device='cuda:0'), covar=tensor([0.1694, 0.2668, 0.0392, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0715, 0.0615, 0.0903, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 16:35:31,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=733353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:35:33,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=733356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:35:49,928 INFO [train.py:968] (0/2) Epoch 17, batch 2500, giga_loss[loss=0.2614, simple_loss=0.3299, pruned_loss=0.09649, over 28591.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3226, pruned_loss=0.08578, over 5729850.09 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3472, pruned_loss=0.09088, over 4193947.95 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3193, pruned_loss=0.08557, over 5721688.27 frames. ], batch size: 307, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:35:55,329 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=733385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:36:04,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.603e+02 1.066e+03 1.330e+03 2.136e+03 5.572e+03, threshold=2.659e+03, percent-clipped=10.0
+2023-03-08 16:36:30,276 INFO [train.py:968] (0/2) Epoch 17, batch 2550, giga_loss[loss=0.2441, simple_loss=0.3189, pruned_loss=0.08469, over 28875.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3208, pruned_loss=0.08446, over 5727790.90 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3469, pruned_loss=0.09042, over 4269744.12 frames. ], giga_tot_loss[loss=0.2429, simple_loss=0.3172, pruned_loss=0.08435, over 5713411.50 frames. ], batch size: 145, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:36:47,936 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=733452.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 16:36:51,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=733455.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 16:36:57,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2856, 1.7280, 1.6437, 1.1483], device='cuda:0'), covar=tensor([0.1819, 0.2878, 0.1587, 0.1956], device='cuda:0'), in_proj_covar=tensor([0.0878, 0.0699, 0.0926, 0.0822], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 16:37:09,073 INFO [train.py:968] (0/2) Epoch 17, batch 2600, giga_loss[loss=0.2408, simple_loss=0.3151, pruned_loss=0.0833, over 28808.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3197, pruned_loss=0.08393, over 5723836.18 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3469, pruned_loss=0.09045, over 4314980.51 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.316, pruned_loss=0.08366, over 5716518.65 frames. ], batch size: 119, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:37:12,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=733484.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 16:37:23,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.291e+02 1.020e+03 1.283e+03 1.754e+03 7.888e+03, threshold=2.566e+03, percent-clipped=10.0
+2023-03-08 16:37:47,774 INFO [train.py:968] (0/2) Epoch 17, batch 2650, giga_loss[loss=0.2322, simple_loss=0.3032, pruned_loss=0.08062, over 28669.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3185, pruned_loss=0.08342, over 5729751.53 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3473, pruned_loss=0.09047, over 4354622.10 frames. ], giga_tot_loss[loss=0.2403, simple_loss=0.3146, pruned_loss=0.08303, over 5719657.08 frames. ], batch size: 92, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:38:30,014 INFO [train.py:968] (0/2) Epoch 17, batch 2700, giga_loss[loss=0.3269, simple_loss=0.3776, pruned_loss=0.1381, over 28997.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3185, pruned_loss=0.08382, over 5724283.70 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3477, pruned_loss=0.0907, over 4377659.81 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3148, pruned_loss=0.08326, over 5713999.15 frames. ], batch size: 128, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:38:47,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.445e+02 9.533e+02 1.213e+03 1.804e+03 4.329e+03, threshold=2.426e+03, percent-clipped=10.0
+2023-03-08 16:38:48,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=733597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:39:16,166 INFO [train.py:968] (0/2) Epoch 17, batch 2750, giga_loss[loss=0.2832, simple_loss=0.3568, pruned_loss=0.1048, over 28898.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3223, pruned_loss=0.08625, over 5723904.39 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3482, pruned_loss=0.09081, over 4400091.78 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3187, pruned_loss=0.08569, over 5713671.04 frames. ], batch size: 227, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:39:23,967 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 16:40:00,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=733672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:40:06,933 INFO [train.py:968] (0/2) Epoch 17, batch 2800, giga_loss[loss=0.3047, simple_loss=0.3704, pruned_loss=0.1194, over 27587.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3291, pruned_loss=0.09122, over 5705142.06 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3479, pruned_loss=0.09064, over 4407720.32 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3263, pruned_loss=0.09088, over 5696214.55 frames. ], batch size: 472, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:40:19,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-08 16:40:23,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.306e+02 1.327e+03 1.581e+03 2.092e+03 5.552e+03, threshold=3.163e+03, percent-clipped=20.0
+2023-03-08 16:40:54,052 INFO [train.py:968] (0/2) Epoch 17, batch 2850, libri_loss[loss=0.2884, simple_loss=0.3698, pruned_loss=0.1034, over 29774.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3371, pruned_loss=0.09642, over 5700283.53 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.348, pruned_loss=0.0907, over 4450786.21 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3344, pruned_loss=0.09619, over 5688639.21 frames. ], batch size: 87, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:41:41,757 INFO [train.py:968] (0/2) Epoch 17, batch 2900, giga_loss[loss=0.2808, simple_loss=0.3587, pruned_loss=0.1014, over 29067.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3426, pruned_loss=0.09885, over 5695490.30 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3474, pruned_loss=0.0904, over 4471961.34 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3407, pruned_loss=0.09896, over 5684006.60 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:42:03,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.254e+02 1.197e+03 1.425e+03 1.673e+03 4.610e+03, threshold=2.849e+03, percent-clipped=4.0
+2023-03-08 16:42:17,112 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=733815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:42:21,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=733818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:42:30,384 INFO [train.py:968] (0/2) Epoch 17, batch 2950, giga_loss[loss=0.2979, simple_loss=0.3715, pruned_loss=0.1122, over 28899.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3484, pruned_loss=0.1015, over 5663887.70 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3478, pruned_loss=0.09061, over 4474029.55 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3467, pruned_loss=0.1016, over 5670204.67 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:42:46,391 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=733847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:43:19,320 INFO [train.py:968] (0/2) Epoch 17, batch 3000, giga_loss[loss=0.2995, simple_loss=0.3701, pruned_loss=0.1144, over 28717.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3544, pruned_loss=0.1049, over 5676234.74 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3475, pruned_loss=0.0905, over 4499125.41 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3533, pruned_loss=0.1052, over 5679987.14 frames. ], batch size: 284, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:43:19,325 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 16:43:28,116 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2169, simple_loss=0.323, pruned_loss=0.05543, over 944034.00 frames. 
+2023-03-08 16:43:28,116 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 16:43:42,060 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7298, 1.9664, 1.2972, 1.4851], device='cuda:0'), covar=tensor([0.0901, 0.0482, 0.0947, 0.0980], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0441, 0.0506, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 16:43:43,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.364e+02 1.336e+03 1.610e+03 2.285e+03 8.354e+03, threshold=3.221e+03, percent-clipped=15.0
+2023-03-08 16:44:10,493 INFO [train.py:968] (0/2) Epoch 17, batch 3050, giga_loss[loss=0.2625, simple_loss=0.3379, pruned_loss=0.0936, over 28592.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3529, pruned_loss=0.1035, over 5661807.58 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3472, pruned_loss=0.09043, over 4523066.33 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3523, pruned_loss=0.1041, over 5669348.67 frames. ], batch size: 307, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:44:48,185 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=733972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:44:53,891 INFO [train.py:968] (0/2) Epoch 17, batch 3100, giga_loss[loss=0.2243, simple_loss=0.3098, pruned_loss=0.06938, over 28906.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3485, pruned_loss=0.09998, over 5673639.38 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3477, pruned_loss=0.09082, over 4562008.08 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3478, pruned_loss=0.1005, over 5675126.04 frames. ], batch size: 213, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:45:09,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.456e+02 1.105e+03 1.352e+03 1.791e+03 4.903e+03, threshold=2.705e+03, percent-clipped=3.0
+2023-03-08 16:45:10,972 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-734000.pt
+2023-03-08 16:45:18,129 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:45:35,282 INFO [train.py:968] (0/2) Epoch 17, batch 3150, giga_loss[loss=0.3305, simple_loss=0.3837, pruned_loss=0.1386, over 28667.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3474, pruned_loss=0.09863, over 5677868.63 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3473, pruned_loss=0.09079, over 4614636.59 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3471, pruned_loss=0.09937, over 5671150.34 frames. ], batch size: 242, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:45:38,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5648, 1.7945, 1.6794, 1.6221], device='cuda:0'), covar=tensor([0.1854, 0.2140, 0.2199, 0.1974], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0738, 0.0694, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 16:45:55,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6045, 1.6864, 1.6409, 1.4923], device='cuda:0'), covar=tensor([0.2545, 0.2328, 0.1883, 0.2225], device='cuda:0'), in_proj_covar=tensor([0.1840, 0.1778, 0.1713, 0.1860], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 16:46:18,590 INFO [train.py:968] (0/2) Epoch 17, batch 3200, giga_loss[loss=0.3222, simple_loss=0.3855, pruned_loss=0.1294, over 28641.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3478, pruned_loss=0.09883, over 5673687.52 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3471, pruned_loss=0.09069, over 4639429.74 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3477, pruned_loss=0.09966, over 5669734.97 frames. ], batch size: 307, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:46:28,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7665, 1.8874, 1.8274, 1.7706], device='cuda:0'), covar=tensor([0.1784, 0.2172, 0.2199, 0.1967], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0739, 0.0693, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 16:46:34,174 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.427e+02 1.201e+03 1.426e+03 1.874e+03 3.862e+03, threshold=2.853e+03, percent-clipped=5.0
+2023-03-08 16:46:50,709 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:46:52,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:47:01,373 INFO [train.py:968] (0/2) Epoch 17, batch 3250, giga_loss[loss=0.2886, simple_loss=0.3585, pruned_loss=0.1093, over 28959.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3512, pruned_loss=0.1011, over 5679631.22 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3477, pruned_loss=0.09109, over 4658167.67 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3508, pruned_loss=0.1016, over 5673396.90 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:47:16,691 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:47:44,501 INFO [train.py:968] (0/2) Epoch 17, batch 3300, giga_loss[loss=0.2628, simple_loss=0.3343, pruned_loss=0.09558, over 28649.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3527, pruned_loss=0.1017, over 5691139.43 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3477, pruned_loss=0.09117, over 4682666.95 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3524, pruned_loss=0.1022, over 5682411.97 frames. ], batch size: 66, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:48:00,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.825e+02 1.134e+03 1.353e+03 1.881e+03 5.088e+03, threshold=2.706e+03, percent-clipped=7.0
+2023-03-08 16:48:27,424 INFO [train.py:968] (0/2) Epoch 17, batch 3350, giga_loss[loss=0.3016, simple_loss=0.3673, pruned_loss=0.1179, over 27663.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3538, pruned_loss=0.1028, over 5687261.53 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3478, pruned_loss=0.09139, over 4729898.75 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3536, pruned_loss=0.1035, over 5677977.49 frames. ], batch size: 472, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:49:11,312 INFO [train.py:968] (0/2) Epoch 17, batch 3400, libri_loss[loss=0.2589, simple_loss=0.3513, pruned_loss=0.0832, over 29771.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3542, pruned_loss=0.1032, over 5690667.57 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3473, pruned_loss=0.09106, over 4757611.18 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3545, pruned_loss=0.1043, over 5679717.52 frames. ], batch size: 87, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:49:28,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.559e+02 1.289e+03 1.662e+03 2.133e+03 5.617e+03, threshold=3.324e+03, percent-clipped=18.0
+2023-03-08 16:49:40,258 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:49:54,005 INFO [train.py:968] (0/2) Epoch 17, batch 3450, giga_loss[loss=0.2717, simple_loss=0.3464, pruned_loss=0.0985, over 28759.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3556, pruned_loss=0.1048, over 5685342.17 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3481, pruned_loss=0.09147, over 4786257.29 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3555, pruned_loss=0.1057, over 5675397.97 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:50:07,468 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:50:22,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1783, 1.0992, 3.6430, 3.1045], device='cuda:0'), covar=tensor([0.1646, 0.2779, 0.0451, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0713, 0.0610, 0.0902, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 16:50:37,850 INFO [train.py:968] (0/2) Epoch 17, batch 3500, giga_loss[loss=0.2775, simple_loss=0.3562, pruned_loss=0.09935, over 28970.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3561, pruned_loss=0.1046, over 5690466.54 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3479, pruned_loss=0.09136, over 4791884.01 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3561, pruned_loss=0.1054, over 5681603.85 frames. ], batch size: 112, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:50:42,024 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:50:43,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=734384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:50:48,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8606, 1.2712, 5.1593, 3.7081], device='cuda:0'), covar=tensor([0.1558, 0.2864, 0.0329, 0.0762], device='cuda:0'), in_proj_covar=tensor([0.0714, 0.0611, 0.0902, 0.0832], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 16:50:54,688 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.313e+02 1.128e+03 1.555e+03 2.303e+03 9.442e+03, threshold=3.109e+03, percent-clipped=7.0
+2023-03-08 16:51:18,700 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-08 16:51:18,974 INFO [train.py:968] (0/2) Epoch 17, batch 3550, giga_loss[loss=0.2725, simple_loss=0.3623, pruned_loss=0.09134, over 28898.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3557, pruned_loss=0.1034, over 5695691.57 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3477, pruned_loss=0.09128, over 4813239.99 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.356, pruned_loss=0.1044, over 5685450.01 frames. ], batch size: 227, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:51:21,092 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:51:21,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2866, 1.2249, 3.7172, 3.1538], device='cuda:0'), covar=tensor([0.2080, 0.3218, 0.0767, 0.1168], device='cuda:0'), in_proj_covar=tensor([0.0711, 0.0610, 0.0899, 0.0830], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 16:51:36,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:52:02,619 INFO [train.py:968] (0/2) Epoch 17, batch 3600, giga_loss[loss=0.2743, simple_loss=0.3522, pruned_loss=0.09818, over 28944.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3553, pruned_loss=0.1025, over 5700713.41 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3474, pruned_loss=0.09113, over 4845141.46 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3561, pruned_loss=0.1037, over 5687306.72 frames. ], batch size: 227, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:52:08,654 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-08 16:52:21,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.166e+02 1.026e+03 1.428e+03 1.994e+03 5.584e+03, threshold=2.856e+03, percent-clipped=8.0
+2023-03-08 16:52:43,872 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:52:44,898 INFO [train.py:968] (0/2) Epoch 17, batch 3650, giga_loss[loss=0.2523, simple_loss=0.3279, pruned_loss=0.0884, over 29010.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.353, pruned_loss=0.101, over 5706977.92 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3469, pruned_loss=0.09093, over 4860950.20 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3541, pruned_loss=0.1023, over 5693573.52 frames. ], batch size: 128, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:52:45,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:53:10,821 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:53:22,021 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0919, 2.2565, 1.9035, 2.4259], device='cuda:0'), covar=tensor([0.2317, 0.2403, 0.2633, 0.2165], device='cuda:0'), in_proj_covar=tensor([0.1426, 0.1039, 0.1262, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 16:53:28,335 INFO [train.py:968] (0/2) Epoch 17, batch 3700, giga_loss[loss=0.2905, simple_loss=0.3639, pruned_loss=0.1085, over 28948.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3502, pruned_loss=0.09987, over 5701080.64 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.347, pruned_loss=0.09091, over 4881011.32 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3511, pruned_loss=0.1011, over 5687618.37 frames. ], batch size: 199, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:53:47,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.172e+02 1.050e+03 1.264e+03 1.586e+03 2.512e+03, threshold=2.528e+03, percent-clipped=0.0
+2023-03-08 16:53:56,690 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 16:54:07,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4567, 2.0104, 1.4575, 1.7566], device='cuda:0'), covar=tensor([0.0706, 0.0250, 0.0306, 0.0793], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 16:54:08,175 INFO [train.py:968] (0/2) Epoch 17, batch 3750, giga_loss[loss=0.2799, simple_loss=0.354, pruned_loss=0.1029, over 28995.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.348, pruned_loss=0.09879, over 5707523.40 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3471, pruned_loss=0.09123, over 4907439.74 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3486, pruned_loss=0.09971, over 5694675.54 frames. ], batch size: 128, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:54:28,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7589, 3.5872, 3.3731, 1.8067], device='cuda:0'), covar=tensor([0.0747, 0.0860, 0.0852, 0.2174], device='cuda:0'), in_proj_covar=tensor([0.1135, 0.1048, 0.0903, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 16:54:52,582 INFO [train.py:968] (0/2) Epoch 17, batch 3800, giga_loss[loss=0.2711, simple_loss=0.3509, pruned_loss=0.09569, over 28724.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3486, pruned_loss=0.09961, over 5699596.42 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3469, pruned_loss=0.09114, over 4929833.99 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3493, pruned_loss=0.1006, over 5690542.89 frames. ], batch size: 60, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:54:58,921 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=734687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:55:07,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.977e+02 1.212e+03 1.449e+03 1.767e+03 8.885e+03, threshold=2.898e+03, percent-clipped=8.0
+2023-03-08 16:55:07,510 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:55:24,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=734718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:55:33,057 INFO [train.py:968] (0/2) Epoch 17, batch 3850, giga_loss[loss=0.2584, simple_loss=0.3327, pruned_loss=0.09208, over 28766.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3493, pruned_loss=0.09998, over 5703039.14 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3469, pruned_loss=0.09114, over 4958406.84 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.35, pruned_loss=0.1011, over 5693828.87 frames. ], batch size: 78, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:55:54,688 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=734757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:56:11,623 INFO [train.py:968] (0/2) Epoch 17, batch 3900, giga_loss[loss=0.2473, simple_loss=0.3316, pruned_loss=0.08153, over 29145.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3485, pruned_loss=0.09846, over 5711115.76 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3464, pruned_loss=0.0909, over 4985898.64 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3495, pruned_loss=0.09976, over 5698949.39 frames. ], batch size: 128, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:56:28,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.831e+02 1.020e+03 1.237e+03 1.526e+03 5.315e+03, threshold=2.475e+03, percent-clipped=4.0
+2023-03-08 16:56:32,523 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=734806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:56:47,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=734822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:56:53,184 INFO [train.py:968] (0/2) Epoch 17, batch 3950, giga_loss[loss=0.2581, simple_loss=0.3415, pruned_loss=0.08731, over 28638.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3484, pruned_loss=0.09789, over 5717039.17 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3466, pruned_loss=0.09096, over 5016826.90 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3491, pruned_loss=0.0991, over 5702063.55 frames. ], batch size: 336, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:56:55,414 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:56:58,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:57:00,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9474, 5.7498, 5.5221, 2.8566], device='cuda:0'), covar=tensor([0.0480, 0.0658, 0.0844, 0.1621], device='cuda:0'), in_proj_covar=tensor([0.1135, 0.1046, 0.0902, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 16:57:15,104 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734856.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 16:57:18,540 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:57:19,272 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:57:20,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:57:32,990 INFO [train.py:968] (0/2) Epoch 17, batch 4000, giga_loss[loss=0.268, simple_loss=0.3444, pruned_loss=0.09578, over 28918.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3478, pruned_loss=0.09766, over 5712550.96 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3459, pruned_loss=0.09062, over 5041140.94 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3489, pruned_loss=0.09908, over 5697432.50 frames. ], batch size: 112, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:57:40,464 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4778, 3.3430, 1.5666, 1.5631], device='cuda:0'), covar=tensor([0.0941, 0.0257, 0.0852, 0.1299], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0525, 0.0361, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 16:57:46,025 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:57:50,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.248e+02 1.022e+03 1.182e+03 1.659e+03 1.268e+04, threshold=2.363e+03, percent-clipped=7.0
+2023-03-08 16:57:50,585 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:57:53,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:12,392 INFO [train.py:968] (0/2) Epoch 17, batch 4050, giga_loss[loss=0.2396, simple_loss=0.3188, pruned_loss=0.08024, over 28866.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3471, pruned_loss=0.09786, over 5711516.33 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3457, pruned_loss=0.09049, over 5056738.66 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3482, pruned_loss=0.0993, over 5703739.39 frames. ], batch size: 112, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:58:15,694 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:24,714 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 16:58:31,101 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:32,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:42,735 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:44,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:52,506 INFO [train.py:968] (0/2) Epoch 17, batch 4100, giga_loss[loss=0.2681, simple_loss=0.3411, pruned_loss=0.09756, over 29025.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3443, pruned_loss=0.09631, over 5715125.68 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3455, pruned_loss=0.09033, over 5073686.11 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3454, pruned_loss=0.09771, over 5705303.42 frames. ], batch size: 155, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:58:54,887 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:59:07,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:59:10,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.824e+02 1.164e+03 1.500e+03 1.968e+03 6.202e+03, threshold=3.000e+03, percent-clipped=18.0
+2023-03-08 16:59:33,324 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:59:33,758 INFO [train.py:968] (0/2) Epoch 17, batch 4150, giga_loss[loss=0.2389, simple_loss=0.3228, pruned_loss=0.07744, over 28988.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3415, pruned_loss=0.09502, over 5715031.57 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3457, pruned_loss=0.09042, over 5083826.16 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3422, pruned_loss=0.09613, over 5706922.35 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:00:08,908 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:00:11,754 INFO [train.py:968] (0/2) Epoch 17, batch 4200, giga_loss[loss=0.2716, simple_loss=0.3417, pruned_loss=0.1007, over 28844.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3413, pruned_loss=0.09498, over 5713783.70 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3456, pruned_loss=0.09044, over 5104310.61 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3417, pruned_loss=0.09597, over 5705802.96 frames. ], batch size: 199, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:00:31,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.904e+02 1.098e+03 1.294e+03 1.704e+03 4.320e+03, threshold=2.587e+03, percent-clipped=3.0
+2023-03-08 17:00:55,549 INFO [train.py:968] (0/2) Epoch 17, batch 4250, giga_loss[loss=0.2748, simple_loss=0.3458, pruned_loss=0.1019, over 28840.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3399, pruned_loss=0.09464, over 5714350.82 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3457, pruned_loss=0.09047, over 5116057.31 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3401, pruned_loss=0.09546, over 5705600.43 frames. ], batch size: 243, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:01:05,479 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:01:26,633 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:01:40,257 INFO [train.py:968] (0/2) Epoch 17, batch 4300, giga_loss[loss=0.2438, simple_loss=0.3271, pruned_loss=0.08028, over 28916.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3395, pruned_loss=0.09555, over 5708343.38 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3458, pruned_loss=0.0906, over 5121843.79 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3396, pruned_loss=0.09613, over 5701694.32 frames. ], batch size: 174, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:01:58,340 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.173e+02 1.039e+03 1.259e+03 1.641e+03 6.301e+03, threshold=2.518e+03, percent-clipped=4.0
+2023-03-08 17:01:58,615 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3227, 1.7117, 1.3888, 1.5774], device='cuda:0'), covar=tensor([0.0743, 0.0280, 0.0327, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 17:02:12,091 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:02:14,046 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:02:19,439 INFO [train.py:968] (0/2) Epoch 17, batch 4350, libri_loss[loss=0.3166, simple_loss=0.3891, pruned_loss=0.122, over 29491.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3357, pruned_loss=0.09352, over 5712433.88 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3458, pruned_loss=0.09066, over 5136461.88 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3355, pruned_loss=0.094, over 5704371.17 frames. ], batch size: 85, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:02:21,740 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735231.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:02:32,763 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 17:02:35,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:02:56,302 INFO [train.py:968] (0/2) Epoch 17, batch 4400, giga_loss[loss=0.2813, simple_loss=0.3479, pruned_loss=0.1074, over 28828.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.335, pruned_loss=0.09337, over 5711861.08 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.346, pruned_loss=0.09096, over 5174425.42 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3341, pruned_loss=0.0937, over 5703420.12 frames. ], batch size: 243, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 17:03:13,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.290e+02 1.247e+03 1.723e+03 2.424e+03 8.036e+03, threshold=3.446e+03, percent-clipped=21.0
+2023-03-08 17:03:15,565 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-08 17:03:35,151 INFO [train.py:968] (0/2) Epoch 17, batch 4450, giga_loss[loss=0.2538, simple_loss=0.3306, pruned_loss=0.08851, over 28977.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3344, pruned_loss=0.09263, over 5714105.42 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3458, pruned_loss=0.09091, over 5185174.68 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3336, pruned_loss=0.09295, over 5705044.13 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:03:57,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:04:04,249 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:04:15,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735374.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:04:17,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735377.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:04:18,457 INFO [train.py:968] (0/2) Epoch 17, batch 4500, giga_loss[loss=0.2845, simple_loss=0.3591, pruned_loss=0.1049, over 28779.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3373, pruned_loss=0.09422, over 5712959.77 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.346, pruned_loss=0.09109, over 5206910.37 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3362, pruned_loss=0.09443, over 5702361.43 frames. ], batch size: 119, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:04:36,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.697e+02 1.019e+03 1.320e+03 1.693e+03 4.229e+03, threshold=2.639e+03, percent-clipped=1.0
+2023-03-08 17:04:38,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:04:41,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735406.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:04:53,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1830, 2.5827, 1.2494, 1.3374], device='cuda:0'), covar=tensor([0.0977, 0.0331, 0.0984, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0530, 0.0363, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 17:04:58,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2954, 1.4893, 1.3992, 1.2698], device='cuda:0'), covar=tensor([0.2597, 0.2434, 0.1647, 0.2231], device='cuda:0'), in_proj_covar=tensor([0.1841, 0.1781, 0.1706, 0.1852], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 17:05:00,611 INFO [train.py:968] (0/2) Epoch 17, batch 4550, giga_loss[loss=0.2756, simple_loss=0.3572, pruned_loss=0.09701, over 28805.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3399, pruned_loss=0.09509, over 5722045.13 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3457, pruned_loss=0.09097, over 5222637.84 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3391, pruned_loss=0.09541, over 5710930.44 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:05:31,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-08 17:05:41,979 INFO [train.py:968] (0/2) Epoch 17, batch 4600, giga_loss[loss=0.2607, simple_loss=0.3389, pruned_loss=0.09129, over 29089.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3414, pruned_loss=0.09529, over 5717138.46 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3455, pruned_loss=0.09096, over 5230466.06 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3408, pruned_loss=0.09566, over 5711811.37 frames. ], batch size: 128, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:06:03,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.544e+02 1.000e+03 1.184e+03 1.632e+03 5.065e+03, threshold=2.368e+03, percent-clipped=5.0
+2023-03-08 17:06:16,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:06:27,006 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:06:28,093 INFO [train.py:968] (0/2) Epoch 17, batch 4650, giga_loss[loss=0.2734, simple_loss=0.339, pruned_loss=0.1039, over 28792.00 frames. ], tot_loss[loss=0.266, simple_loss=0.342, pruned_loss=0.09499, over 5711317.32 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3454, pruned_loss=0.09101, over 5248208.88 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3415, pruned_loss=0.09538, over 5703980.66 frames. ], batch size: 99, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:06:39,002 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:06:39,616 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3692, 1.6141, 1.4768, 1.2973], device='cuda:0'), covar=tensor([0.3093, 0.2160, 0.1810, 0.2394], device='cuda:0'), in_proj_covar=tensor([0.1837, 0.1777, 0.1702, 0.1846], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 17:06:45,164 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:06:47,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:06:50,038 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 17:07:11,146 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:07:11,568 INFO [train.py:968] (0/2) Epoch 17, batch 4700, giga_loss[loss=0.2452, simple_loss=0.3337, pruned_loss=0.0783, over 28742.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3414, pruned_loss=0.09401, over 5700696.69 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3454, pruned_loss=0.09098, over 5262784.96 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3409, pruned_loss=0.09442, over 5691844.24 frames. ], batch size: 284, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:07:28,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.001e+02 1.141e+03 1.408e+03 1.912e+03 8.424e+03, threshold=2.817e+03, percent-clipped=15.0
+2023-03-08 17:07:50,738 INFO [train.py:968] (0/2) Epoch 17, batch 4750, giga_loss[loss=0.3287, simple_loss=0.3929, pruned_loss=0.1322, over 27969.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3417, pruned_loss=0.094, over 5709538.42 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.345, pruned_loss=0.09076, over 5281625.30 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3415, pruned_loss=0.0946, over 5697280.28 frames. ], batch size: 412, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:08:09,833 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:17,402 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:19,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:36,315 INFO [train.py:968] (0/2) Epoch 17, batch 4800, giga_loss[loss=0.2687, simple_loss=0.3437, pruned_loss=0.09682, over 28546.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3432, pruned_loss=0.09515, over 5714045.37 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3455, pruned_loss=0.09105, over 5290813.46 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3427, pruned_loss=0.09546, over 5701911.26 frames. ], batch size: 336, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 17:08:39,961 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:42,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:46,467 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:55,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.551e+02 1.282e+03 1.520e+03 1.999e+03 3.721e+03, threshold=3.039e+03, percent-clipped=7.0
+2023-03-08 17:09:02,230 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-08 17:09:06,922 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:09:20,981 INFO [train.py:968] (0/2) Epoch 17, batch 4850, giga_loss[loss=0.2717, simple_loss=0.3464, pruned_loss=0.09856, over 28869.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3449, pruned_loss=0.09656, over 5713472.94 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3462, pruned_loss=0.09146, over 5302903.46 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3438, pruned_loss=0.09655, over 5700694.41 frames. ], batch size: 145, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:09:21,814 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:09:27,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:10:02,554 INFO [train.py:968] (0/2) Epoch 17, batch 4900, giga_loss[loss=0.3005, simple_loss=0.3738, pruned_loss=0.1137, over 28956.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3482, pruned_loss=0.09856, over 5710469.41 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.347, pruned_loss=0.09194, over 5312676.37 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3468, pruned_loss=0.09834, over 5704338.45 frames. ], batch size: 155, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:10:22,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.654e+02 1.230e+03 1.577e+03 2.493e+03 4.131e+03, threshold=3.153e+03, percent-clipped=12.0
+2023-03-08 17:10:45,000 INFO [train.py:968] (0/2) Epoch 17, batch 4950, giga_loss[loss=0.2736, simple_loss=0.3523, pruned_loss=0.09742, over 28737.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3506, pruned_loss=0.0998, over 5713364.16 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3473, pruned_loss=0.09207, over 5321279.18 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3493, pruned_loss=0.09963, over 5706140.67 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:10:58,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-08 17:11:21,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:24,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:26,907 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:27,222 INFO [train.py:968] (0/2) Epoch 17, batch 5000, giga_loss[loss=0.3041, simple_loss=0.3722, pruned_loss=0.118, over 28922.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3519, pruned_loss=0.1003, over 5716419.56 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3483, pruned_loss=0.09258, over 5333748.86 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.35, pruned_loss=0.09994, over 5708201.70 frames. ], batch size: 227, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:11:29,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:43,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:44,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.213e+02 1.307e+03 1.507e+03 1.935e+03 3.989e+03, threshold=3.014e+03, percent-clipped=6.0
+2023-03-08 17:11:47,658 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:53,469 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:12:07,456 INFO [train.py:968] (0/2) Epoch 17, batch 5050, giga_loss[loss=0.2891, simple_loss=0.361, pruned_loss=0.1086, over 28987.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3531, pruned_loss=0.1013, over 5706573.46 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3485, pruned_loss=0.09269, over 5346037.45 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3516, pruned_loss=0.1011, over 5700567.01 frames. ], batch size: 227, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:12:42,390 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-08 17:12:44,699 INFO [train.py:968] (0/2) Epoch 17, batch 5100, giga_loss[loss=0.2437, simple_loss=0.3223, pruned_loss=0.08261, over 28793.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3523, pruned_loss=0.1007, over 5712370.46 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3481, pruned_loss=0.09245, over 5368823.85 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3516, pruned_loss=0.1011, over 5701130.68 frames. ], batch size: 99, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:13:00,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5405, 1.6801, 1.7881, 1.3347], device='cuda:0'), covar=tensor([0.1887, 0.2440, 0.1522, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.0872, 0.0694, 0.0918, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 17:13:02,089 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-08 17:13:02,302 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-736000.pt
+2023-03-08 17:13:04,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.180e+02 1.215e+03 1.449e+03 1.813e+03 7.197e+03, threshold=2.899e+03, percent-clipped=6.0
+2023-03-08 17:13:24,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=736026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:13:26,113 INFO [train.py:968] (0/2) Epoch 17, batch 5150, giga_loss[loss=0.269, simple_loss=0.352, pruned_loss=0.09295, over 29065.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3512, pruned_loss=0.1002, over 5707825.22 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3483, pruned_loss=0.09273, over 5373479.05 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3506, pruned_loss=0.1006, over 5703372.41 frames. ], batch size: 164, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:13:38,780 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=736045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:13:40,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=736048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:14:07,397 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=736077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:14:09,134 INFO [train.py:968] (0/2) Epoch 17, batch 5200, giga_loss[loss=0.2606, simple_loss=0.3266, pruned_loss=0.09729, over 28411.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3485, pruned_loss=0.09934, over 5705094.24 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3485, pruned_loss=0.0929, over 5382289.67 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3478, pruned_loss=0.09956, over 5699649.41 frames. ], batch size: 60, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:14:30,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.306e+02 1.084e+03 1.314e+03 1.784e+03 3.989e+03, threshold=2.628e+03, percent-clipped=5.0
+2023-03-08 17:14:45,648 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-08 17:14:51,627 INFO [train.py:968] (0/2) Epoch 17, batch 5250, giga_loss[loss=0.2381, simple_loss=0.3242, pruned_loss=0.07595, over 28891.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3446, pruned_loss=0.097, over 5707676.48 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3487, pruned_loss=0.09297, over 5386033.30 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3439, pruned_loss=0.09719, over 5704619.00 frames. ], batch size: 174, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:15:25,690 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=736169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:15:28,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=736172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:15:33,987 INFO [train.py:968] (0/2) Epoch 17, batch 5300, giga_loss[loss=0.2566, simple_loss=0.3332, pruned_loss=0.08999, over 28658.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3445, pruned_loss=0.09603, over 5709025.17 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3487, pruned_loss=0.09298, over 5392992.32 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3439, pruned_loss=0.09622, over 5704678.55 frames. ], batch size: 92, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:15:53,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=736201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:15:55,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.765e+02 1.097e+03 1.372e+03 1.820e+03 5.861e+03, threshold=2.744e+03, percent-clipped=8.0
+2023-03-08 17:16:17,399 INFO [train.py:968] (0/2) Epoch 17, batch 5350, giga_loss[loss=0.2684, simple_loss=0.3405, pruned_loss=0.09815, over 28928.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3465, pruned_loss=0.09593, over 5712239.28 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.349, pruned_loss=0.09309, over 5401705.52 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3456, pruned_loss=0.09605, over 5706618.56 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:16:30,632 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=736246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:16:58,627 INFO [train.py:968] (0/2) Epoch 17, batch 5400, giga_loss[loss=0.2771, simple_loss=0.3531, pruned_loss=0.1005, over 28876.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3476, pruned_loss=0.09671, over 5715686.98 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3493, pruned_loss=0.09327, over 5414577.15 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3466, pruned_loss=0.09677, over 5709237.32 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:17:17,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.195e+02 1.217e+03 1.462e+03 1.955e+03 4.713e+03, threshold=2.924e+03, percent-clipped=10.0
+2023-03-08 17:17:38,939 INFO [train.py:968] (0/2) Epoch 17, batch 5450, giga_loss[loss=0.2707, simple_loss=0.3436, pruned_loss=0.09895, over 28609.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3448, pruned_loss=0.0967, over 5714179.98 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3495, pruned_loss=0.09346, over 5410096.25 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3439, pruned_loss=0.09662, over 5715245.73 frames. ], batch size: 262, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:18:22,713 INFO [train.py:968] (0/2) Epoch 17, batch 5500, giga_loss[loss=0.2799, simple_loss=0.3486, pruned_loss=0.1056, over 28850.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3439, pruned_loss=0.09741, over 5715108.48 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3499, pruned_loss=0.09368, over 5412043.30 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3427, pruned_loss=0.09724, over 5720308.88 frames. ], batch size: 227, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:18:41,300 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.409e+02 1.289e+03 1.566e+03 1.941e+03 6.271e+03, threshold=3.132e+03, percent-clipped=4.0
+2023-03-08 17:18:51,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7201, 1.9591, 1.5767, 1.9465], device='cuda:0'), covar=tensor([0.2463, 0.2611, 0.2928, 0.2516], device='cuda:0'), in_proj_covar=tensor([0.1418, 0.1030, 0.1253, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 17:19:02,429 INFO [train.py:968] (0/2) Epoch 17, batch 5550, giga_loss[loss=0.2843, simple_loss=0.3544, pruned_loss=0.1071, over 28697.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3423, pruned_loss=0.09773, over 5721289.54 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3499, pruned_loss=0.0937, over 5424739.17 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3412, pruned_loss=0.09767, over 5722209.87 frames. ], batch size: 307, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:19:15,854 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=736442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:19:48,332 INFO [train.py:968] (0/2) Epoch 17, batch 5600, giga_loss[loss=0.3081, simple_loss=0.3719, pruned_loss=0.1221, over 28980.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3421, pruned_loss=0.09845, over 5721201.15 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3495, pruned_loss=0.09349, over 5429547.84 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3414, pruned_loss=0.09862, over 5720314.56 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:20:12,337 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.468e+02 1.259e+03 1.510e+03 2.054e+03 5.705e+03, threshold=3.019e+03, percent-clipped=6.0
+2023-03-08 17:20:31,226 INFO [train.py:968] (0/2) Epoch 17, batch 5650, giga_loss[loss=0.2274, simple_loss=0.3072, pruned_loss=0.0738, over 29126.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3396, pruned_loss=0.09758, over 5714170.76 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.349, pruned_loss=0.09327, over 5439881.19 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3393, pruned_loss=0.09802, over 5710428.42 frames. ], batch size: 155, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:20:59,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2367, 1.5930, 1.5663, 1.1346], device='cuda:0'), covar=tensor([0.1685, 0.2486, 0.1410, 0.1590], device='cuda:0'), in_proj_covar=tensor([0.0872, 0.0694, 0.0917, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 17:21:12,828 INFO [train.py:968] (0/2) Epoch 17, batch 5700, giga_loss[loss=0.1993, simple_loss=0.2739, pruned_loss=0.06229, over 28613.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3353, pruned_loss=0.09524, over 5714838.44 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.349, pruned_loss=0.09333, over 5445369.90 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3349, pruned_loss=0.09559, over 5711604.25 frames. ], batch size: 78, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:21:34,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.246e+02 1.146e+03 1.693e+03 2.150e+03 6.099e+03, threshold=3.385e+03, percent-clipped=5.0
+2023-03-08 17:21:48,407 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=736621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:21:53,395 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 17:21:53,548 INFO [train.py:968] (0/2) Epoch 17, batch 5750, giga_loss[loss=0.241, simple_loss=0.3183, pruned_loss=0.08191, over 29041.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.332, pruned_loss=0.09322, over 5715674.19 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3488, pruned_loss=0.09328, over 5461102.57 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3313, pruned_loss=0.09356, over 5706975.61 frames. ], batch size: 164, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:22:34,926 INFO [train.py:968] (0/2) Epoch 17, batch 5800, giga_loss[loss=0.2852, simple_loss=0.356, pruned_loss=0.1072, over 28729.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3311, pruned_loss=0.0923, over 5719513.56 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.349, pruned_loss=0.09337, over 5474565.34 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3299, pruned_loss=0.09248, over 5707286.50 frames. ], batch size: 262, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:22:55,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.592e+02 1.157e+03 1.657e+03 2.153e+03 5.804e+03, threshold=3.314e+03, percent-clipped=4.0
+2023-03-08 17:23:15,728 INFO [train.py:968] (0/2) Epoch 17, batch 5850, giga_loss[loss=0.3015, simple_loss=0.3732, pruned_loss=0.1148, over 28750.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3336, pruned_loss=0.09334, over 5718369.91 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3485, pruned_loss=0.09321, over 5480958.68 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3328, pruned_loss=0.09361, over 5706353.44 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:23:31,967 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 17:23:44,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=736764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:23:46,488 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=736767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:23:57,876 INFO [train.py:968] (0/2) Epoch 17, batch 5900, giga_loss[loss=0.2441, simple_loss=0.3291, pruned_loss=0.0795, over 29026.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3369, pruned_loss=0.09413, over 5716577.25 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3485, pruned_loss=0.09313, over 5488176.30 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.336, pruned_loss=0.09442, over 5704899.03 frames. ], batch size: 145, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:23:59,595 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-08 17:24:00,163 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4981, 1.9366, 1.6724, 1.3169], device='cuda:0'), covar=tensor([0.2864, 0.2051, 0.2408, 0.2725], device='cuda:0'), in_proj_covar=tensor([0.1850, 0.1791, 0.1718, 0.1858], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 17:24:01,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=736785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:24:11,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=736796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:24:18,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.026e+02 1.115e+03 1.301e+03 1.576e+03 3.614e+03, threshold=2.602e+03, percent-clipped=1.0
+2023-03-08 17:24:24,937 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-08 17:24:29,840 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=736817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:24:40,059 INFO [train.py:968] (0/2) Epoch 17, batch 5950, giga_loss[loss=0.276, simple_loss=0.3533, pruned_loss=0.09931, over 28997.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3404, pruned_loss=0.09538, over 5721218.63 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3485, pruned_loss=0.09316, over 5496368.96 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3395, pruned_loss=0.0956, over 5708830.00 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:25:22,283 INFO [train.py:968] (0/2) Epoch 17, batch 6000, giga_loss[loss=0.3304, simple_loss=0.3949, pruned_loss=0.1329, over 28661.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.344, pruned_loss=0.09736, over 5716891.77 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3487, pruned_loss=0.09316, over 5503958.31 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.343, pruned_loss=0.0976, over 5706545.34 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:25:22,288 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 17:25:30,917 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2158, simple_loss=0.322, pruned_loss=0.05475, over 944034.00 frames. 
+2023-03-08 17:25:30,918 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 17:25:53,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.496e+02 1.194e+03 1.439e+03 1.856e+03 6.365e+03, threshold=2.878e+03, percent-clipped=9.0
+2023-03-08 17:26:18,174 INFO [train.py:968] (0/2) Epoch 17, batch 6050, giga_loss[loss=0.2793, simple_loss=0.3587, pruned_loss=0.09998, over 28961.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3468, pruned_loss=0.09962, over 5706508.09 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3484, pruned_loss=0.09313, over 5507849.49 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3462, pruned_loss=0.09987, over 5696879.48 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:26:25,081 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3238, 1.1333, 4.3836, 3.4132], device='cuda:0'), covar=tensor([0.1706, 0.2968, 0.0388, 0.0993], device='cuda:0'), in_proj_covar=tensor([0.0715, 0.0614, 0.0904, 0.0832], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 17:26:48,097 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=736960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:26:50,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=736963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:27:07,947 INFO [train.py:968] (0/2) Epoch 17, batch 6100, giga_loss[loss=0.3017, simple_loss=0.3678, pruned_loss=0.1178, over 28807.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3542, pruned_loss=0.1061, over 5703183.48 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3486, pruned_loss=0.09328, over 5510079.88 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3536, pruned_loss=0.1063, over 5696474.32 frames. ], batch size: 112, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:27:18,232 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=736992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:27:31,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.398e+02 1.563e+03 1.917e+03 2.929e+03 7.840e+03, threshold=3.833e+03, percent-clipped=26.0
+2023-03-08 17:27:53,372 INFO [train.py:968] (0/2) Epoch 17, batch 6150, giga_loss[loss=0.3593, simple_loss=0.4145, pruned_loss=0.1521, over 27686.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3609, pruned_loss=0.1111, over 5696464.38 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3488, pruned_loss=0.09336, over 5513839.46 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3605, pruned_loss=0.1116, over 5693065.96 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:27:55,633 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:28:42,671 INFO [train.py:968] (0/2) Epoch 17, batch 6200, giga_loss[loss=0.2962, simple_loss=0.3677, pruned_loss=0.1123, over 28556.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3674, pruned_loss=0.116, over 5688881.47 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3488, pruned_loss=0.09344, over 5519402.58 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3674, pruned_loss=0.1168, over 5685265.87 frames. ], batch size: 60, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:29:04,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.645e+03 2.117e+03 2.644e+03 5.398e+03, threshold=4.234e+03, percent-clipped=9.0
+2023-03-08 17:29:08,578 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:29:21,346 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 17:29:27,038 INFO [train.py:968] (0/2) Epoch 17, batch 6250, giga_loss[loss=0.3349, simple_loss=0.3928, pruned_loss=0.1385, over 28497.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3709, pruned_loss=0.1188, over 5696408.34 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3487, pruned_loss=0.09339, over 5530099.77 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3716, pruned_loss=0.1203, over 5690719.79 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:29:55,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=737160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:30:11,126 INFO [train.py:968] (0/2) Epoch 17, batch 6300, giga_loss[loss=0.2935, simple_loss=0.3603, pruned_loss=0.1133, over 28895.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3758, pruned_loss=0.1227, over 5690600.31 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3487, pruned_loss=0.09325, over 5535256.54 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3771, pruned_loss=0.1249, over 5685640.51 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:30:38,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.315e+02 1.915e+03 2.267e+03 2.991e+03 6.931e+03, threshold=4.533e+03, percent-clipped=5.0
+2023-03-08 17:31:06,091 INFO [train.py:968] (0/2) Epoch 17, batch 6350, giga_loss[loss=0.3688, simple_loss=0.4156, pruned_loss=0.161, over 28338.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3795, pruned_loss=0.1263, over 5682026.35 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3489, pruned_loss=0.09336, over 5540493.79 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3807, pruned_loss=0.1283, over 5675692.69 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:31:14,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2758, 2.2554, 1.3808, 1.3975], device='cuda:0'), covar=tensor([0.0772, 0.0398, 0.0695, 0.1101], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0534, 0.0364, 0.0410], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 17:31:42,373 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-08 17:31:59,249 INFO [train.py:968] (0/2) Epoch 17, batch 6400, giga_loss[loss=0.4091, simple_loss=0.4384, pruned_loss=0.1899, over 27459.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3816, pruned_loss=0.129, over 5661741.62 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3491, pruned_loss=0.09352, over 5540142.75 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3831, pruned_loss=0.1312, over 5659219.18 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:32:02,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:32:20,827 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3316, 3.6100, 1.5703, 1.4046], device='cuda:0'), covar=tensor([0.1005, 0.0337, 0.0869, 0.1409], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0535, 0.0364, 0.0410], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 17:32:28,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=737303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:32:32,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=737306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:32:32,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.819e+03 2.407e+03 3.789e+03 1.038e+04, threshold=4.814e+03, percent-clipped=15.0
+2023-03-08 17:32:55,289 INFO [train.py:968] (0/2) Epoch 17, batch 6450, giga_loss[loss=0.2686, simple_loss=0.3321, pruned_loss=0.1026, over 28555.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3846, pruned_loss=0.1326, over 5661330.15 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3489, pruned_loss=0.09342, over 5541931.03 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3861, pruned_loss=0.1346, over 5658260.32 frames. ], batch size: 85, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:32:58,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5762, 1.7651, 1.5052, 1.5406], device='cuda:0'), covar=tensor([0.2057, 0.1912, 0.1972, 0.1656], device='cuda:0'), in_proj_covar=tensor([0.1416, 0.1032, 0.1255, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 17:33:04,306 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=737335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:33:48,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2029, 1.4816, 1.4901, 1.0842], device='cuda:0'), covar=tensor([0.1583, 0.2423, 0.1306, 0.1557], device='cuda:0'), in_proj_covar=tensor([0.0864, 0.0692, 0.0912, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 17:33:55,327 INFO [train.py:968] (0/2) Epoch 17, batch 6500, giga_loss[loss=0.291, simple_loss=0.3664, pruned_loss=0.1078, over 28938.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3903, pruned_loss=0.1385, over 5642178.30 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3489, pruned_loss=0.09345, over 5543798.38 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3917, pruned_loss=0.1403, over 5638633.91 frames. ], batch size: 145, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:34:25,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.888e+03 2.650e+03 3.147e+03 9.513e+03, threshold=5.299e+03, percent-clipped=7.0
+2023-03-08 17:34:25,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=737407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:34:38,731 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:34:48,395 INFO [train.py:968] (0/2) Epoch 17, batch 6550, giga_loss[loss=0.2968, simple_loss=0.3594, pruned_loss=0.1171, over 28902.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3903, pruned_loss=0.1387, over 5642642.60 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.349, pruned_loss=0.09344, over 5549090.10 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.392, pruned_loss=0.1407, over 5636338.23 frames. ], batch size: 112, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:35:39,364 INFO [train.py:968] (0/2) Epoch 17, batch 6600, giga_loss[loss=0.3556, simple_loss=0.4147, pruned_loss=0.1483, over 28864.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3891, pruned_loss=0.1387, over 5640981.16 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3488, pruned_loss=0.09339, over 5554203.08 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3911, pruned_loss=0.141, over 5632485.55 frames. ], batch size: 227, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:35:45,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=737485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:35:55,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6069, 1.6955, 1.2461, 1.2982], device='cuda:0'), covar=tensor([0.0939, 0.0643, 0.1101, 0.1103], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0442, 0.0509, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 17:36:05,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3139, 1.2045, 3.6282, 3.1216], device='cuda:0'), covar=tensor([0.1526, 0.2662, 0.0476, 0.0930], device='cuda:0'), in_proj_covar=tensor([0.0718, 0.0618, 0.0908, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 17:36:08,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.753e+03 2.439e+03 3.432e+03 9.305e+03, threshold=4.879e+03, percent-clipped=7.0
+2023-03-08 17:36:31,452 INFO [train.py:968] (0/2) Epoch 17, batch 6650, libri_loss[loss=0.2371, simple_loss=0.3178, pruned_loss=0.07821, over 29555.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3869, pruned_loss=0.1371, over 5641333.70 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3486, pruned_loss=0.09332, over 5565936.80 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3901, pruned_loss=0.1405, over 5626682.25 frames. ], batch size: 76, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:36:53,740 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=737550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:36:57,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=737553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:37:05,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8960, 1.0640, 1.0804, 0.8887], device='cuda:0'), covar=tensor([0.1954, 0.2344, 0.1378, 0.1739], device='cuda:0'), in_proj_covar=tensor([0.1857, 0.1799, 0.1718, 0.1855], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 17:37:22,505 INFO [train.py:968] (0/2) Epoch 17, batch 6700, giga_loss[loss=0.3217, simple_loss=0.3936, pruned_loss=0.1249, over 28880.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3861, pruned_loss=0.1349, over 5654326.88 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3483, pruned_loss=0.09318, over 5571121.56 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3894, pruned_loss=0.1383, over 5639122.05 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:37:26,985 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=737582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:37:50,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.605e+03 2.033e+03 2.669e+03 5.818e+03, threshold=4.066e+03, percent-clipped=4.0
+2023-03-08 17:38:11,207 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737625.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:38:14,233 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:38:14,336 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=737628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:38:14,645 INFO [train.py:968] (0/2) Epoch 17, batch 6750, giga_loss[loss=0.286, simple_loss=0.3675, pruned_loss=0.1023, over 28920.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.387, pruned_loss=0.135, over 5650446.46 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3479, pruned_loss=0.09301, over 5577962.32 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3907, pruned_loss=0.1387, over 5633583.56 frames. ], batch size: 136, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:38:16,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=737631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:38:33,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3652, 1.5581, 1.6354, 1.4687], device='cuda:0'), covar=tensor([0.1566, 0.1219, 0.1445, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0741, 0.0697, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 17:38:41,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=737657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:38:44,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=737660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:38:57,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5695, 1.7353, 1.7489, 1.5570], device='cuda:0'), covar=tensor([0.1799, 0.2031, 0.2067, 0.2006], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0739, 0.0695, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 17:39:04,607 INFO [train.py:968] (0/2) Epoch 17, batch 6800, giga_loss[loss=0.2904, simple_loss=0.3598, pruned_loss=0.1105, over 28652.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3856, pruned_loss=0.1342, over 5635625.74 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3477, pruned_loss=0.093, over 5585906.50 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3895, pruned_loss=0.1381, over 5616771.40 frames. ], batch size: 242, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:39:05,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:39:30,351 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.628e+03 2.012e+03 2.573e+03 7.856e+03, threshold=4.023e+03, percent-clipped=10.0
+2023-03-08 17:39:56,356 INFO [train.py:968] (0/2) Epoch 17, batch 6850, giga_loss[loss=0.3193, simple_loss=0.3868, pruned_loss=0.126, over 28517.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3827, pruned_loss=0.1304, over 5645083.60 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.348, pruned_loss=0.0931, over 5595619.75 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3868, pruned_loss=0.1346, over 5623193.70 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:40:10,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2992, 5.1108, 4.8394, 2.4109], device='cuda:0'), covar=tensor([0.0450, 0.0643, 0.0765, 0.1746], device='cuda:0'), in_proj_covar=tensor([0.1168, 0.1077, 0.0929, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 17:40:29,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3360, 1.5434, 1.5778, 1.1639], device='cuda:0'), covar=tensor([0.1623, 0.2386, 0.1369, 0.1616], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0692, 0.0911, 0.0810], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 17:40:48,613 INFO [train.py:968] (0/2) Epoch 17, batch 6900, giga_loss[loss=0.3292, simple_loss=0.3883, pruned_loss=0.135, over 27929.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3808, pruned_loss=0.1277, over 5652852.41 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3483, pruned_loss=0.09325, over 5597187.65 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3839, pruned_loss=0.1311, over 5634633.74 frames. ], batch size: 412, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:41:02,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:41:03,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=737794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:41:10,985 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=737800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:41:16,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=737803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:41:19,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.383e+02 1.636e+03 2.137e+03 3.019e+03 7.635e+03, threshold=4.273e+03, percent-clipped=7.0
+2023-03-08 17:41:26,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2257, 1.3785, 1.4638, 1.2201], device='cuda:0'), covar=tensor([0.2768, 0.2325, 0.1557, 0.2231], device='cuda:0'), in_proj_covar=tensor([0.1853, 0.1792, 0.1717, 0.1854], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 17:41:41,000 INFO [train.py:968] (0/2) Epoch 17, batch 6950, giga_loss[loss=0.2803, simple_loss=0.3465, pruned_loss=0.1071, over 28874.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3782, pruned_loss=0.126, over 5648338.21 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.09342, over 5592346.44 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.381, pruned_loss=0.1289, over 5639987.16 frames. ], batch size: 112, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:41:43,861 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=737832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:41:49,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737839.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:42:26,019 INFO [train.py:968] (0/2) Epoch 17, batch 7000, giga_loss[loss=0.3398, simple_loss=0.3925, pruned_loss=0.1435, over 28934.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3756, pruned_loss=0.1241, over 5659769.14 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3485, pruned_loss=0.09409, over 5605948.65 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3788, pruned_loss=0.1272, over 5642607.83 frames. ], batch size: 145, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:42:50,191 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-08 17:42:56,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.603e+02 1.665e+03 2.083e+03 3.072e+03 8.096e+03, threshold=4.165e+03, percent-clipped=5.0
+2023-03-08 17:43:15,844 INFO [train.py:968] (0/2) Epoch 17, batch 7050, giga_loss[loss=0.3544, simple_loss=0.3847, pruned_loss=0.162, over 23452.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3745, pruned_loss=0.1237, over 5653979.73 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3481, pruned_loss=0.09392, over 5610190.89 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3779, pruned_loss=0.1268, over 5637325.17 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:43:22,375 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=737937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:43:27,144 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=737940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:43:47,237 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-08 17:43:50,491 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737963.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:43:59,173 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=737969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:44:09,551 INFO [train.py:968] (0/2) Epoch 17, batch 7100, libri_loss[loss=0.2957, simple_loss=0.3693, pruned_loss=0.111, over 29528.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3737, pruned_loss=0.1233, over 5655811.87 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3478, pruned_loss=0.09374, over 5619146.73 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3775, pruned_loss=0.1268, over 5635600.56 frames. ], batch size: 82, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 17:44:24,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3049, 1.3216, 3.5936, 3.1680], device='cuda:0'), covar=tensor([0.1867, 0.2925, 0.0785, 0.1583], device='cuda:0'), in_proj_covar=tensor([0.0719, 0.0620, 0.0912, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 17:44:26,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5010, 2.2215, 1.6687, 0.7157], device='cuda:0'), covar=tensor([0.5267, 0.2561, 0.3475, 0.5746], device='cuda:0'), in_proj_covar=tensor([0.1672, 0.1577, 0.1549, 0.1369], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 17:44:27,833 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-738000.pt
+2023-03-08 17:44:29,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738000.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:44:31,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:44:34,584 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=738004.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:44:39,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.419e+02 1.541e+03 2.005e+03 2.653e+03 8.206e+03, threshold=4.010e+03, percent-clipped=7.0
+2023-03-08 17:44:47,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4652, 1.9603, 1.8278, 1.6540], device='cuda:0'), covar=tensor([0.1888, 0.1641, 0.1954, 0.1773], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0740, 0.0696, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 17:44:59,633 INFO [train.py:968] (0/2) Epoch 17, batch 7150, giga_loss[loss=0.2772, simple_loss=0.3577, pruned_loss=0.0984, over 28842.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.372, pruned_loss=0.1217, over 5664534.94 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3474, pruned_loss=0.09349, over 5629862.90 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3763, pruned_loss=0.1258, over 5640081.05 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 17:45:17,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2854, 4.0156, 3.7897, 1.7540], device='cuda:0'), covar=tensor([0.0649, 0.0867, 0.0914, 0.2366], device='cuda:0'), in_proj_covar=tensor([0.1172, 0.1084, 0.0933, 0.0710], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 17:45:26,185 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:45:49,074 INFO [train.py:968] (0/2) Epoch 17, batch 7200, libri_loss[loss=0.2864, simple_loss=0.3551, pruned_loss=0.1088, over 29576.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3705, pruned_loss=0.1183, over 5676002.79 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3471, pruned_loss=0.09338, over 5641871.76 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3754, pruned_loss=0.123, over 5646399.99 frames. ], batch size: 77, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:46:21,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.859e+02 1.453e+03 2.057e+03 2.850e+03 6.390e+03, threshold=4.114e+03, percent-clipped=11.0
+2023-03-08 17:46:41,622 INFO [train.py:968] (0/2) Epoch 17, batch 7250, giga_loss[loss=0.3298, simple_loss=0.3892, pruned_loss=0.1352, over 28544.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3725, pruned_loss=0.1176, over 5678837.95 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3471, pruned_loss=0.09331, over 5643437.65 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.377, pruned_loss=0.1219, over 5654556.21 frames. ], batch size: 307, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:46:53,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2355, 1.5542, 1.5515, 1.3336], device='cuda:0'), covar=tensor([0.1716, 0.1520, 0.2030, 0.1767], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0735, 0.0693, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 17:46:55,476 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738143.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:46:59,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738146.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:46:59,027 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:47:03,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:47:24,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:47:33,296 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738175.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:47:36,313 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:47:36,719 INFO [train.py:968] (0/2) Epoch 17, batch 7300, giga_loss[loss=0.3308, simple_loss=0.389, pruned_loss=0.1363, over 28290.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3748, pruned_loss=0.1196, over 5677435.27 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3471, pruned_loss=0.09338, over 5647333.23 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3788, pruned_loss=0.1233, over 5655148.70 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:47:44,294 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-08 17:47:54,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:47:57,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:48:04,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.822e+03 2.317e+03 3.519e+03 1.587e+04, threshold=4.635e+03, percent-clipped=19.0
+2023-03-08 17:48:10,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738214.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:48:27,181 INFO [train.py:968] (0/2) Epoch 17, batch 7350, giga_loss[loss=0.3012, simple_loss=0.3721, pruned_loss=0.1152, over 28762.00 frames. ], tot_loss[loss=0.308, simple_loss=0.375, pruned_loss=0.1205, over 5667615.97 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3473, pruned_loss=0.0935, over 5648328.30 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3784, pruned_loss=0.1238, over 5649564.15 frames. ], batch size: 262, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:48:27,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:48:31,040 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=738233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:49:16,988 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-08 17:49:17,284 INFO [train.py:968] (0/2) Epoch 17, batch 7400, giga_loss[loss=0.4844, simple_loss=0.4784, pruned_loss=0.2453, over 26524.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3741, pruned_loss=0.121, over 5675251.39 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3473, pruned_loss=0.09351, over 5652313.30 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3772, pruned_loss=0.1239, over 5657931.83 frames. ], batch size: 555, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:49:46,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.681e+03 2.181e+03 3.287e+03 8.261e+03, threshold=4.361e+03, percent-clipped=12.0
+2023-03-08 17:49:48,264 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:49:50,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:50:06,219 INFO [train.py:968] (0/2) Epoch 17, batch 7450, giga_loss[loss=0.3603, simple_loss=0.3953, pruned_loss=0.1626, over 26572.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.374, pruned_loss=0.1223, over 5672489.45 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3476, pruned_loss=0.09357, over 5656969.41 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3767, pruned_loss=0.1251, over 5654984.15 frames. ], batch size: 555, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:50:12,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738338.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:50:17,178 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:50:19,463 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 17:50:29,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738357.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:50:32,009 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738360.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:50:50,644 INFO [train.py:968] (0/2) Epoch 17, batch 7500, giga_loss[loss=0.3203, simple_loss=0.3677, pruned_loss=0.1364, over 23670.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3723, pruned_loss=0.1205, over 5667321.55 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3479, pruned_loss=0.09379, over 5656771.32 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3754, pruned_loss=0.1239, over 5652911.60 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:50:51,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738379.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:50:59,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738389.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:51:18,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.505e+02 1.519e+03 1.856e+03 2.479e+03 4.983e+03, threshold=3.713e+03, percent-clipped=3.0
+2023-03-08 17:51:37,416 INFO [train.py:968] (0/2) Epoch 17, batch 7550, giga_loss[loss=0.3554, simple_loss=0.4082, pruned_loss=0.1513, over 28964.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3725, pruned_loss=0.1195, over 5664783.80 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3484, pruned_loss=0.09417, over 5653675.22 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.375, pruned_loss=0.1224, over 5655727.10 frames. ], batch size: 128, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:52:09,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2222, 3.0493, 2.8731, 1.3812], device='cuda:0'), covar=tensor([0.0981, 0.1013, 0.0961, 0.2520], device='cuda:0'), in_proj_covar=tensor([0.1171, 0.1081, 0.0930, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 17:52:24,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3640, 1.5814, 1.4946, 1.2166], device='cuda:0'), covar=tensor([0.2653, 0.2409, 0.1651, 0.2269], device='cuda:0'), in_proj_covar=tensor([0.1861, 0.1803, 0.1721, 0.1864], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 17:52:26,878 INFO [train.py:968] (0/2) Epoch 17, batch 7600, giga_loss[loss=0.3027, simple_loss=0.37, pruned_loss=0.1177, over 28962.00 frames. ], tot_loss[loss=0.306, simple_loss=0.373, pruned_loss=0.1195, over 5666822.22 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3485, pruned_loss=0.09423, over 5657465.91 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3753, pruned_loss=0.1221, over 5656519.13 frames. ], batch size: 227, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:52:29,375 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738481.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:52:31,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738484.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:52:54,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.686e+03 2.372e+03 3.134e+03 7.399e+03, threshold=4.745e+03, percent-clipped=12.0
+2023-03-08 17:52:57,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738513.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:53:01,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7142, 1.7823, 1.3689, 1.3784], device='cuda:0'), covar=tensor([0.0865, 0.0617, 0.0990, 0.1146], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0440, 0.0505, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 17:53:04,096 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738522.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:53:06,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738525.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:53:09,232 INFO [train.py:968] (0/2) Epoch 17, batch 7650, giga_loss[loss=0.3065, simple_loss=0.3685, pruned_loss=0.1223, over 28791.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3716, pruned_loss=0.118, over 5684446.87 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3486, pruned_loss=0.09419, over 5661308.25 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3737, pruned_loss=0.1204, over 5673127.06 frames. ], batch size: 99, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:53:37,647 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=738554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:53:37,667 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738554.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:54:01,574 INFO [train.py:968] (0/2) Epoch 17, batch 7700, giga_loss[loss=0.3554, simple_loss=0.3877, pruned_loss=0.1616, over 23707.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3699, pruned_loss=0.1178, over 5679027.79 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3483, pruned_loss=0.094, over 5667243.21 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3723, pruned_loss=0.1204, over 5665310.99 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:54:32,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:54:34,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.489e+03 1.943e+03 2.754e+03 7.096e+03, threshold=3.886e+03, percent-clipped=3.0
+2023-03-08 17:54:51,645 INFO [train.py:968] (0/2) Epoch 17, batch 7750, giga_loss[loss=0.3171, simple_loss=0.3824, pruned_loss=0.126, over 28724.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3686, pruned_loss=0.118, over 5666688.54 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3483, pruned_loss=0.09404, over 5662064.93 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3709, pruned_loss=0.1206, over 5659898.80 frames. ], batch size: 262, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:55:43,909 INFO [train.py:968] (0/2) Epoch 17, batch 7800, libri_loss[loss=0.2843, simple_loss=0.3652, pruned_loss=0.1017, over 28797.00 frames. ], tot_loss[loss=0.304, simple_loss=0.369, pruned_loss=0.1196, over 5650863.44 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3486, pruned_loss=0.09424, over 5653114.70 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3707, pruned_loss=0.1216, over 5653399.28 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:56:16,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.784e+03 2.300e+03 2.885e+03 5.477e+03, threshold=4.600e+03, percent-clipped=7.0
+2023-03-08 17:56:36,196 INFO [train.py:968] (0/2) Epoch 17, batch 7850, giga_loss[loss=0.3431, simple_loss=0.3792, pruned_loss=0.1536, over 23697.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3703, pruned_loss=0.1218, over 5631495.25 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3489, pruned_loss=0.09445, over 5638246.85 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3716, pruned_loss=0.1235, over 5646684.46 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:56:58,223 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:57:00,695 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:57:22,902 INFO [train.py:968] (0/2) Epoch 17, batch 7900, libri_loss[loss=0.257, simple_loss=0.3448, pruned_loss=0.0846, over 29540.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3683, pruned_loss=0.1203, over 5641242.43 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3488, pruned_loss=0.09432, over 5640992.36 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3697, pruned_loss=0.122, over 5650451.16 frames. ], batch size: 82, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:57:27,400 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:57:52,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.625e+03 2.216e+03 3.019e+03 8.933e+03, threshold=4.432e+03, percent-clipped=5.0
+2023-03-08 17:58:12,030 INFO [train.py:968] (0/2) Epoch 17, batch 7950, giga_loss[loss=0.2841, simple_loss=0.3591, pruned_loss=0.1046, over 28505.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3693, pruned_loss=0.1203, over 5650999.00 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3491, pruned_loss=0.09454, over 5647074.23 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3706, pruned_loss=0.1221, over 5653005.89 frames. ], batch size: 71, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:58:25,414 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-08 17:58:56,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4773, 3.8797, 1.6058, 1.5688], device='cuda:0'), covar=tensor([0.0935, 0.0380, 0.0876, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0539, 0.0366, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 17:59:00,105 INFO [train.py:968] (0/2) Epoch 17, batch 8000, giga_loss[loss=0.3659, simple_loss=0.4077, pruned_loss=0.1621, over 27966.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3691, pruned_loss=0.1192, over 5657429.10 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09484, over 5650602.28 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3704, pruned_loss=0.1209, over 5655900.28 frames. ], batch size: 412, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:59:28,125 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.033e+02 1.602e+03 2.139e+03 3.393e+03 9.496e+03, threshold=4.278e+03, percent-clipped=12.0
+2023-03-08 17:59:45,590 INFO [train.py:968] (0/2) Epoch 17, batch 8050, giga_loss[loss=0.295, simple_loss=0.3673, pruned_loss=0.1113, over 28801.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3682, pruned_loss=0.1177, over 5671412.81 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3486, pruned_loss=0.09448, over 5652294.14 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3702, pruned_loss=0.1198, over 5668787.17 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:59:46,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:00:12,081 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=738955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:00:34,659 INFO [train.py:968] (0/2) Epoch 17, batch 8100, giga_loss[loss=0.4315, simple_loss=0.4603, pruned_loss=0.2013, over 28619.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3686, pruned_loss=0.1175, over 5679554.54 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3487, pruned_loss=0.09439, over 5658570.69 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3708, pruned_loss=0.1199, over 5672199.46 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:01:02,523 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.645e+03 2.160e+03 2.868e+03 5.131e+03, threshold=4.320e+03, percent-clipped=5.0
+2023-03-08 18:01:05,941 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-08 18:01:18,682 INFO [train.py:968] (0/2) Epoch 17, batch 8150, giga_loss[loss=0.2778, simple_loss=0.3506, pruned_loss=0.1025, over 28977.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3693, pruned_loss=0.1187, over 5683091.11 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3483, pruned_loss=0.09435, over 5665990.01 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.372, pruned_loss=0.1215, over 5671284.57 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:02:05,884 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=739072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:02:10,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=739075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:02:14,735 INFO [train.py:968] (0/2) Epoch 17, batch 8200, giga_loss[loss=0.329, simple_loss=0.3859, pruned_loss=0.136, over 28801.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.373, pruned_loss=0.1229, over 5660921.03 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3485, pruned_loss=0.09459, over 5660853.60 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3753, pruned_loss=0.1252, over 5656756.18 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:02:28,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5077, 1.6313, 1.5807, 1.4933], device='cuda:0'), covar=tensor([0.1247, 0.1477, 0.1741, 0.1520], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0737, 0.0696, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 18:02:40,104 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=739104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:02:47,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.415e+02 1.939e+03 2.516e+03 3.342e+03 6.909e+03, threshold=5.032e+03, percent-clipped=10.0
+2023-03-08 18:03:08,489 INFO [train.py:968] (0/2) Epoch 17, batch 8250, giga_loss[loss=0.3323, simple_loss=0.3865, pruned_loss=0.1391, over 28859.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3746, pruned_loss=0.1257, over 5656399.85 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3484, pruned_loss=0.09455, over 5664144.35 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3768, pruned_loss=0.128, over 5650241.85 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:03:32,304 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-08 18:03:57,930 INFO [train.py:968] (0/2) Epoch 17, batch 8300, giga_loss[loss=0.3144, simple_loss=0.3742, pruned_loss=0.1273, over 29010.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3752, pruned_loss=0.1268, over 5654671.43 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3486, pruned_loss=0.09477, over 5661619.98 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3775, pruned_loss=0.1293, over 5652572.48 frames. ], batch size: 155, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:04:28,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.776e+02 1.847e+03 2.432e+03 3.330e+03 8.786e+03, threshold=4.863e+03, percent-clipped=11.0
+2023-03-08 18:04:45,118 INFO [train.py:968] (0/2) Epoch 17, batch 8350, giga_loss[loss=0.298, simple_loss=0.3642, pruned_loss=0.1158, over 28968.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3759, pruned_loss=0.1272, over 5653792.72 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3492, pruned_loss=0.09508, over 5662846.51 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3781, pruned_loss=0.1299, over 5650776.52 frames. ], batch size: 227, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:05:27,081 INFO [train.py:968] (0/2) Epoch 17, batch 8400, giga_loss[loss=0.2685, simple_loss=0.3454, pruned_loss=0.09583, over 28287.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3729, pruned_loss=0.1244, over 5668778.89 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3491, pruned_loss=0.09494, over 5667753.43 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3756, pruned_loss=0.1278, over 5661743.13 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:05:54,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.858e+02 1.534e+03 1.924e+03 3.105e+03 6.553e+03, threshold=3.848e+03, percent-clipped=4.0
+2023-03-08 18:06:11,952 INFO [train.py:968] (0/2) Epoch 17, batch 8450, giga_loss[loss=0.2767, simple_loss=0.3545, pruned_loss=0.09946, over 28773.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3708, pruned_loss=0.1209, over 5684987.77 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3486, pruned_loss=0.09471, over 5674494.46 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3739, pruned_loss=0.1245, over 5673406.80 frames. ], batch size: 99, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:06:12,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=739330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:06:26,821 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 18:06:39,245 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-08 18:06:54,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4773, 1.7765, 1.4272, 1.5469], device='cuda:0'), covar=tensor([0.2487, 0.2483, 0.2847, 0.2201], device='cuda:0'), in_proj_covar=tensor([0.1425, 0.1035, 0.1264, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 18:06:57,336 INFO [train.py:968] (0/2) Epoch 17, batch 8500, giga_loss[loss=0.3156, simple_loss=0.3758, pruned_loss=0.1277, over 28914.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3693, pruned_loss=0.1196, over 5682375.41 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3488, pruned_loss=0.09483, over 5679058.77 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.372, pruned_loss=0.1229, over 5669295.60 frames. ], batch size: 174, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:07:29,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.779e+03 2.233e+03 3.658e+03 8.071e+03, threshold=4.466e+03, percent-clipped=22.0
+2023-03-08 18:07:32,672 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3190, 1.5408, 1.5536, 1.1820], device='cuda:0'), covar=tensor([0.1483, 0.2338, 0.1251, 0.1531], device='cuda:0'), in_proj_covar=tensor([0.0869, 0.0697, 0.0915, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 18:07:44,611 INFO [train.py:968] (0/2) Epoch 17, batch 8550, giga_loss[loss=0.2987, simple_loss=0.3617, pruned_loss=0.1179, over 28882.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3674, pruned_loss=0.1187, over 5679010.58 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3494, pruned_loss=0.0952, over 5685127.01 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3697, pruned_loss=0.1218, over 5662733.53 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:07:55,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0347, 2.3269, 1.8575, 2.4728], device='cuda:0'), covar=tensor([0.2273, 0.2341, 0.2670, 0.2135], device='cuda:0'), in_proj_covar=tensor([0.1427, 0.1036, 0.1265, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 18:08:27,390 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=739473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:08:30,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=739476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:08:34,102 INFO [train.py:968] (0/2) Epoch 17, batch 8600, giga_loss[loss=0.2961, simple_loss=0.3628, pruned_loss=0.1147, over 28587.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3667, pruned_loss=0.1187, over 5682772.48 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3497, pruned_loss=0.09526, over 5684441.32 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3684, pruned_loss=0.1213, over 5670296.87 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:09:00,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=739505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:09:07,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.570e+02 1.518e+03 2.097e+03 3.204e+03 7.978e+03, threshold=4.193e+03, percent-clipped=13.0
+2023-03-08 18:09:25,177 INFO [train.py:968] (0/2) Epoch 17, batch 8650, giga_loss[loss=0.3603, simple_loss=0.408, pruned_loss=0.1563, over 28557.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3685, pruned_loss=0.12, over 5680591.15 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3493, pruned_loss=0.09504, over 5686419.00 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3707, pruned_loss=0.1229, over 5668581.86 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:09:44,446 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=739547.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:10:16,264 INFO [train.py:968] (0/2) Epoch 17, batch 8700, giga_loss[loss=0.3079, simple_loss=0.3929, pruned_loss=0.1114, over 28976.00 frames. ], tot_loss[loss=0.307, simple_loss=0.372, pruned_loss=0.121, over 5684360.66 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3488, pruned_loss=0.09477, over 5690901.55 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3746, pruned_loss=0.1241, over 5670566.47 frames. ], batch size: 112, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:10:24,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3424, 3.0433, 1.3853, 1.5087], device='cuda:0'), covar=tensor([0.0981, 0.0464, 0.0981, 0.1347], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0541, 0.0367, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 18:10:49,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.078e+02 1.588e+03 2.115e+03 2.822e+03 5.163e+03, threshold=4.230e+03, percent-clipped=6.0
+2023-03-08 18:11:04,575 INFO [train.py:968] (0/2) Epoch 17, batch 8750, libri_loss[loss=0.2989, simple_loss=0.3719, pruned_loss=0.113, over 29529.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3743, pruned_loss=0.1204, over 5676293.68 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3489, pruned_loss=0.09492, over 5692191.99 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.377, pruned_loss=0.1234, over 5663499.55 frames. ], batch size: 81, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:11:43,016 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=739672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:11:51,616 INFO [train.py:968] (0/2) Epoch 17, batch 8800, giga_loss[loss=0.3292, simple_loss=0.3719, pruned_loss=0.1433, over 23512.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3771, pruned_loss=0.1222, over 5677454.74 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3493, pruned_loss=0.09514, over 5694915.89 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3794, pruned_loss=0.1249, over 5664655.31 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:12:17,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-08 18:12:18,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.580e+03 2.001e+03 3.059e+03 1.197e+04, threshold=4.002e+03, percent-clipped=8.0
+2023-03-08 18:12:27,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 18:12:31,992 INFO [train.py:968] (0/2) Epoch 17, batch 8850, giga_loss[loss=0.352, simple_loss=0.4076, pruned_loss=0.1482, over 28285.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3776, pruned_loss=0.1229, over 5693956.64 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3491, pruned_loss=0.09525, over 5700928.08 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3805, pruned_loss=0.1258, over 5677995.13 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:12:38,202 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=739736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:13:20,179 INFO [train.py:968] (0/2) Epoch 17, batch 8900, giga_loss[loss=0.3338, simple_loss=0.3951, pruned_loss=0.1363, over 28691.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3778, pruned_loss=0.1238, over 5693531.73 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3488, pruned_loss=0.09501, over 5706874.50 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3813, pruned_loss=0.1272, over 5675055.95 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:13:52,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.602e+03 2.193e+03 2.864e+03 6.499e+03, threshold=4.386e+03, percent-clipped=10.0
+2023-03-08 18:14:08,252 INFO [train.py:968] (0/2) Epoch 17, batch 8950, giga_loss[loss=0.3102, simple_loss=0.3749, pruned_loss=0.1228, over 28689.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3783, pruned_loss=0.1253, over 5694449.11 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.349, pruned_loss=0.09505, over 5708484.40 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3811, pruned_loss=0.1282, over 5678510.63 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:14:13,908 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6266, 1.8134, 1.4896, 1.7455], device='cuda:0'), covar=tensor([0.2416, 0.2537, 0.2777, 0.2273], device='cuda:0'), in_proj_covar=tensor([0.1420, 0.1036, 0.1262, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 18:14:40,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3215, 1.7328, 1.4682, 1.4711], device='cuda:0'), covar=tensor([0.0734, 0.0330, 0.0304, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 18:14:59,537 INFO [train.py:968] (0/2) Epoch 17, batch 9000, libri_loss[loss=0.2389, simple_loss=0.3168, pruned_loss=0.08047, over 28531.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3743, pruned_loss=0.123, over 5693363.48 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3487, pruned_loss=0.09483, over 5709420.72 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3772, pruned_loss=0.1259, over 5679711.06 frames. ], batch size: 63, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:14:59,543 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 18:15:08,262 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2118, simple_loss=0.3196, pruned_loss=0.05194, over 944034.00 frames. 
+2023-03-08 18:15:08,262 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 18:15:42,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.461e+02 1.517e+03 1.927e+03 2.555e+03 5.760e+03, threshold=3.854e+03, percent-clipped=4.0
+2023-03-08 18:15:45,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4093, 1.5812, 1.3642, 1.6219], device='cuda:0'), covar=tensor([0.0744, 0.0317, 0.0310, 0.0794], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 18:15:49,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=739922.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:15:54,871 INFO [train.py:968] (0/2) Epoch 17, batch 9050, giga_loss[loss=0.4075, simple_loss=0.431, pruned_loss=0.192, over 27867.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3736, pruned_loss=0.1235, over 5681746.49 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3491, pruned_loss=0.09505, over 5713273.11 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3763, pruned_loss=0.1264, over 5666860.61 frames. ], batch size: 412, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:16:10,959 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-08 18:16:41,592 INFO [train.py:968] (0/2) Epoch 17, batch 9100, giga_loss[loss=0.2713, simple_loss=0.3366, pruned_loss=0.103, over 28931.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3713, pruned_loss=0.1222, over 5688366.27 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.349, pruned_loss=0.09488, over 5720321.94 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3743, pruned_loss=0.1256, over 5669280.59 frames. ], batch size: 112, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:17:04,312 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-740000.pt
+2023-03-08 18:17:16,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.568e+02 1.557e+03 2.245e+03 3.160e+03 8.809e+03, threshold=4.490e+03, percent-clipped=14.0
+2023-03-08 18:17:23,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-08 18:17:32,307 INFO [train.py:968] (0/2) Epoch 17, batch 9150, giga_loss[loss=0.3527, simple_loss=0.3927, pruned_loss=0.1563, over 28773.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3728, pruned_loss=0.1233, over 5690750.46 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3493, pruned_loss=0.09491, over 5726482.99 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3755, pruned_loss=0.1268, over 5668820.38 frames. ], batch size: 99, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:17:49,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=740047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:17:59,792 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4519, 1.5530, 1.5308, 1.4536], device='cuda:0'), covar=tensor([0.1487, 0.1859, 0.1911, 0.1707], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0744, 0.0702, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 18:18:06,773 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740065.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:18:08,883 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740068.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:18:20,129 INFO [train.py:968] (0/2) Epoch 17, batch 9200, giga_loss[loss=0.3446, simple_loss=0.3999, pruned_loss=0.1447, over 27591.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3715, pruned_loss=0.1231, over 5686855.98 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.349, pruned_loss=0.09472, over 5728672.42 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3742, pruned_loss=0.1264, over 5667111.50 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:18:24,100 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=740083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:18:36,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=740097.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:18:50,970 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=740111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:18:53,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.729e+02 1.789e+03 2.310e+03 3.289e+03 9.361e+03, threshold=4.621e+03, percent-clipped=10.0
+2023-03-08 18:19:07,249 INFO [train.py:968] (0/2) Epoch 17, batch 9250, libri_loss[loss=0.2537, simple_loss=0.3464, pruned_loss=0.08052, over 25623.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3709, pruned_loss=0.1227, over 5687637.91 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3494, pruned_loss=0.09489, over 5727053.15 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3733, pruned_loss=0.1259, over 5672577.10 frames. ], batch size: 136, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:19:51,734 INFO [train.py:968] (0/2) Epoch 17, batch 9300, giga_loss[loss=0.3268, simple_loss=0.3931, pruned_loss=0.1302, over 28720.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3713, pruned_loss=0.1221, over 5696866.64 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3495, pruned_loss=0.09486, over 5730860.21 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3736, pruned_loss=0.1253, over 5680857.11 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:20:08,719 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:20:12,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:20:32,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.395e+02 1.479e+03 1.834e+03 2.343e+03 6.929e+03, threshold=3.669e+03, percent-clipped=2.0
+2023-03-08 18:20:41,471 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=740222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:20:43,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3035, 1.1454, 4.1780, 3.3996], device='cuda:0'), covar=tensor([0.1751, 0.2854, 0.0452, 0.0782], device='cuda:0'), in_proj_covar=tensor([0.0724, 0.0621, 0.0919, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 18:20:46,631 INFO [train.py:968] (0/2) Epoch 17, batch 9350, giga_loss[loss=0.2976, simple_loss=0.3683, pruned_loss=0.1134, over 28925.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3742, pruned_loss=0.1241, over 5673052.10 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3496, pruned_loss=0.09497, over 5722680.47 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3761, pruned_loss=0.1268, over 5668006.44 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:21:13,117 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:21:17,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:21:38,190 INFO [train.py:968] (0/2) Epoch 17, batch 9400, giga_loss[loss=0.334, simple_loss=0.3925, pruned_loss=0.1378, over 28984.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3742, pruned_loss=0.1247, over 5672280.35 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3496, pruned_loss=0.09499, over 5724768.90 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.376, pruned_loss=0.1271, over 5665787.47 frames. ], batch size: 164, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:21:44,740 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=740286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:22:09,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3699, 1.7749, 1.6077, 1.5510], device='cuda:0'), covar=tensor([0.2053, 0.1993, 0.2395, 0.2276], device='cuda:0'), in_proj_covar=tensor([0.0462, 0.0745, 0.0701, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 18:22:11,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.063e+02 1.732e+03 2.376e+03 3.519e+03 8.233e+03, threshold=4.752e+03, percent-clipped=24.0
+2023-03-08 18:22:22,878 INFO [train.py:968] (0/2) Epoch 17, batch 9450, libri_loss[loss=0.3158, simple_loss=0.3829, pruned_loss=0.1243, over 29516.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3753, pruned_loss=0.1237, over 5679916.61 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3501, pruned_loss=0.09526, over 5726664.99 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.377, pruned_loss=0.1262, over 5671383.97 frames. ], batch size: 81, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:22:39,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8905, 1.1895, 2.8467, 2.6743], device='cuda:0'), covar=tensor([0.1553, 0.2383, 0.0603, 0.1342], device='cuda:0'), in_proj_covar=tensor([0.0723, 0.0621, 0.0918, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 18:23:07,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4122, 2.2602, 2.4641, 2.0237], device='cuda:0'), covar=tensor([0.1800, 0.2539, 0.1739, 0.2145], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0744, 0.0700, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 18:23:10,095 INFO [train.py:968] (0/2) Epoch 17, batch 9500, giga_loss[loss=0.3293, simple_loss=0.3987, pruned_loss=0.13, over 28648.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3752, pruned_loss=0.1212, over 5681794.01 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.35, pruned_loss=0.09515, over 5727617.81 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3772, pruned_loss=0.124, over 5672890.17 frames. ], batch size: 307, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:23:18,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.6882, 1.3205, 1.2220, 1.0547], device='cuda:0'), covar=tensor([0.2219, 0.1204, 0.2317, 0.1862], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0744, 0.0699, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 18:23:39,096 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=740409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:23:42,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.531e+02 1.250e+03 1.571e+03 2.125e+03 5.622e+03, threshold=3.142e+03, percent-clipped=2.0
+2023-03-08 18:23:56,973 INFO [train.py:968] (0/2) Epoch 17, batch 9550, giga_loss[loss=0.2981, simple_loss=0.3752, pruned_loss=0.1105, over 28962.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3767, pruned_loss=0.1206, over 5674151.96 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.35, pruned_loss=0.09517, over 5721175.07 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3787, pruned_loss=0.1232, over 5672158.26 frames. ], batch size: 164, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:24:21,260 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=740453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:24:27,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=740458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:24:28,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3943, 1.8796, 1.4679, 1.5476], device='cuda:0'), covar=tensor([0.0771, 0.0294, 0.0314, 0.0837], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 18:24:47,739 INFO [train.py:968] (0/2) Epoch 17, batch 9600, giga_loss[loss=0.4217, simple_loss=0.4389, pruned_loss=0.2023, over 23625.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3802, pruned_loss=0.1238, over 5675922.97 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3501, pruned_loss=0.09523, over 5723978.37 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3821, pruned_loss=0.1261, over 5671396.28 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:25:21,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.587e+03 2.002e+03 2.710e+03 1.104e+04, threshold=4.004e+03, percent-clipped=14.0
+2023-03-08 18:25:32,873 INFO [train.py:968] (0/2) Epoch 17, batch 9650, giga_loss[loss=0.2758, simple_loss=0.3511, pruned_loss=0.1003, over 28771.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3807, pruned_loss=0.1254, over 5677377.85 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3501, pruned_loss=0.09539, over 5725841.63 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3831, pruned_loss=0.128, over 5670555.88 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:26:23,345 INFO [train.py:968] (0/2) Epoch 17, batch 9700, giga_loss[loss=0.3471, simple_loss=0.4024, pruned_loss=0.1459, over 28505.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3806, pruned_loss=0.1262, over 5677933.07 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3498, pruned_loss=0.0952, over 5729671.04 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3834, pruned_loss=0.1291, over 5668187.40 frames. ], batch size: 71, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:26:45,676 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:26:48,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:26:57,095 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.711e+03 2.129e+03 2.833e+03 8.006e+03, threshold=4.258e+03, percent-clipped=8.0
+2023-03-08 18:27:09,504 INFO [train.py:968] (0/2) Epoch 17, batch 9750, libri_loss[loss=0.2512, simple_loss=0.3417, pruned_loss=0.08031, over 29341.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3793, pruned_loss=0.1258, over 5657033.64 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3496, pruned_loss=0.09497, over 5731552.38 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3825, pruned_loss=0.1291, over 5645734.47 frames. ], batch size: 92, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:27:12,766 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=740633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:27:49,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-08 18:27:52,958 INFO [train.py:968] (0/2) Epoch 17, batch 9800, libri_loss[loss=0.272, simple_loss=0.3553, pruned_loss=0.0944, over 29641.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3775, pruned_loss=0.1232, over 5659026.16 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3495, pruned_loss=0.09497, over 5727640.17 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3813, pruned_loss=0.127, over 5651538.46 frames. ], batch size: 88, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:28:03,541 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-08 18:28:24,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.756e+03 2.588e+03 3.736e+03 9.546e+03, threshold=5.176e+03, percent-clipped=19.0
+2023-03-08 18:28:38,754 INFO [train.py:968] (0/2) Epoch 17, batch 9850, giga_loss[loss=0.3016, simple_loss=0.3771, pruned_loss=0.1131, over 29084.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3773, pruned_loss=0.1216, over 5667249.61 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3494, pruned_loss=0.09503, over 5729715.99 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3808, pruned_loss=0.1249, over 5658558.19 frames. ], batch size: 128, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:28:48,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5791, 1.7629, 1.5031, 1.5427], device='cuda:0'), covar=tensor([0.2510, 0.2602, 0.2946, 0.2448], device='cuda:0'), in_proj_covar=tensor([0.1423, 0.1035, 0.1263, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 18:29:26,423 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=740775.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:29:28,788 INFO [train.py:968] (0/2) Epoch 17, batch 9900, giga_loss[loss=0.3304, simple_loss=0.3898, pruned_loss=0.1355, over 28297.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3785, pruned_loss=0.1222, over 5671322.10 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3495, pruned_loss=0.09506, over 5732950.89 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3817, pruned_loss=0.1252, over 5660676.75 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:29:32,453 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=740784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:30:03,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.497e+03 2.065e+03 3.040e+03 8.206e+03, threshold=4.131e+03, percent-clipped=2.0
+2023-03-08 18:30:15,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=740828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:30:15,719 INFO [train.py:968] (0/2) Epoch 17, batch 9950, giga_loss[loss=0.3429, simple_loss=0.3998, pruned_loss=0.143, over 28284.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3776, pruned_loss=0.1219, over 5668808.91 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3495, pruned_loss=0.09498, over 5734709.60 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3811, pruned_loss=0.1254, over 5656262.64 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:30:22,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2697, 1.2140, 1.1965, 1.4048], device='cuda:0'), covar=tensor([0.0783, 0.0371, 0.0340, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0114, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0058, 0.0100], device='cuda:0')
+2023-03-08 18:30:27,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=740839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:30:27,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9201, 3.7439, 3.5528, 1.7238], device='cuda:0'), covar=tensor([0.0696, 0.0811, 0.0769, 0.2095], device='cuda:0'), in_proj_covar=tensor([0.1178, 0.1091, 0.0937, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 18:31:07,774 INFO [train.py:968] (0/2) Epoch 17, batch 10000, giga_loss[loss=0.3406, simple_loss=0.387, pruned_loss=0.1471, over 27473.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3758, pruned_loss=0.1213, over 5676778.30 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09489, over 5736600.08 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3789, pruned_loss=0.1243, over 5664611.87 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:31:39,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.115e+02 1.571e+03 1.991e+03 2.760e+03 6.929e+03, threshold=3.983e+03, percent-clipped=9.0
+2023-03-08 18:31:52,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:31:53,742 INFO [train.py:968] (0/2) Epoch 17, batch 10050, giga_loss[loss=0.3199, simple_loss=0.3822, pruned_loss=0.1288, over 28858.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3746, pruned_loss=0.1215, over 5650646.04 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3494, pruned_loss=0.09496, over 5716462.36 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3782, pruned_loss=0.125, over 5655633.01 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:31:55,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:32:23,454 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=740959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:32:35,723 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:32:39,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:32:46,613 INFO [train.py:968] (0/2) Epoch 17, batch 10100, giga_loss[loss=0.2662, simple_loss=0.3402, pruned_loss=0.09608, over 28549.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3718, pruned_loss=0.1201, over 5657994.64 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.0949, over 5720339.13 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3753, pruned_loss=0.1235, over 5657383.71 frames. ], batch size: 71, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:33:11,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=741003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:33:27,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.543e+02 1.653e+03 2.085e+03 2.888e+03 7.379e+03, threshold=4.169e+03, percent-clipped=11.0
+2023-03-08 18:33:40,809 INFO [train.py:968] (0/2) Epoch 17, batch 10150, giga_loss[loss=0.2647, simple_loss=0.338, pruned_loss=0.09566, over 28961.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3704, pruned_loss=0.1201, over 5664101.28 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09484, over 5721894.95 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3732, pruned_loss=0.1231, over 5661764.86 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:34:03,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3504, 1.4906, 1.5668, 1.2808], device='cuda:0'), covar=tensor([0.1631, 0.1696, 0.2137, 0.1844], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0741, 0.0696, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 18:34:15,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3485, 2.9989, 1.4888, 1.4256], device='cuda:0'), covar=tensor([0.0980, 0.0356, 0.0884, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0541, 0.0367, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 18:34:18,516 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-08 18:34:28,815 INFO [train.py:968] (0/2) Epoch 17, batch 10200, giga_loss[loss=0.2764, simple_loss=0.3567, pruned_loss=0.09803, over 28592.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3719, pruned_loss=0.122, over 5667682.43 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3492, pruned_loss=0.0946, over 5726711.04 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.375, pruned_loss=0.1253, over 5659760.14 frames. ], batch size: 60, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:34:59,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.517e+03 1.977e+03 2.687e+03 7.203e+03, threshold=3.954e+03, percent-clipped=7.0
+2023-03-08 18:35:12,806 INFO [train.py:968] (0/2) Epoch 17, batch 10250, giga_loss[loss=0.3248, simple_loss=0.3916, pruned_loss=0.129, over 28278.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.37, pruned_loss=0.1205, over 5667533.25 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3493, pruned_loss=0.09481, over 5728786.15 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3731, pruned_loss=0.1239, over 5657624.34 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:35:32,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=741150.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:36:01,441 INFO [train.py:968] (0/2) Epoch 17, batch 10300, giga_loss[loss=0.2657, simple_loss=0.342, pruned_loss=0.09468, over 28950.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3668, pruned_loss=0.1171, over 5668367.42 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09485, over 5731578.90 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3695, pruned_loss=0.12, over 5657259.20 frames. ], batch size: 227, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:36:35,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=741214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:36:37,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.339e+02 1.366e+03 1.783e+03 2.433e+03 7.477e+03, threshold=3.565e+03, percent-clipped=6.0
+2023-03-08 18:36:49,367 INFO [train.py:968] (0/2) Epoch 17, batch 10350, giga_loss[loss=0.3017, simple_loss=0.3702, pruned_loss=0.1166, over 28836.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3657, pruned_loss=0.1157, over 5668189.94 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3493, pruned_loss=0.09478, over 5735642.61 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3683, pruned_loss=0.1188, over 5653460.25 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:37:23,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5010, 4.0986, 1.6287, 1.5341], device='cuda:0'), covar=tensor([0.0965, 0.0326, 0.0897, 0.1390], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0540, 0.0367, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 18:37:33,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=741273.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:37:38,914 INFO [train.py:968] (0/2) Epoch 17, batch 10400, giga_loss[loss=0.2658, simple_loss=0.3327, pruned_loss=0.09939, over 28691.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3646, pruned_loss=0.1157, over 5673478.00 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.349, pruned_loss=0.09462, over 5738196.98 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3673, pruned_loss=0.1187, over 5658148.70 frames. ], batch size: 71, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:37:43,520 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-08 18:37:53,656 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=741293.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:37:58,184 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=741296.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:38:16,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.769e+02 1.752e+03 2.138e+03 3.420e+03 2.007e+04, threshold=4.276e+03, percent-clipped=21.0
+2023-03-08 18:38:18,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8152, 1.9034, 1.4067, 1.4602], device='cuda:0'), covar=tensor([0.0896, 0.0603, 0.1019, 0.1118], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0443, 0.0508, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 18:38:23,249 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=741325.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:38:26,065 INFO [train.py:968] (0/2) Epoch 17, batch 10450, giga_loss[loss=0.248, simple_loss=0.3279, pruned_loss=0.08405, over 28993.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3606, pruned_loss=0.1134, over 5677948.43 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3488, pruned_loss=0.09436, over 5740564.74 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3633, pruned_loss=0.1165, over 5662045.96 frames. ], batch size: 145, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:38:51,405 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=741353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:38:54,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=741357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:38:56,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=741360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:39:13,405 INFO [train.py:968] (0/2) Epoch 17, batch 10500, giga_loss[loss=0.3531, simple_loss=0.4018, pruned_loss=0.1522, over 27575.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1146, over 5674654.20 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3484, pruned_loss=0.09428, over 5734875.61 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3649, pruned_loss=0.1175, over 5665529.90 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:39:23,315 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=741389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:39:49,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.464e+03 1.971e+03 2.796e+03 6.684e+03, threshold=3.942e+03, percent-clipped=8.0
+2023-03-08 18:40:01,682 INFO [train.py:968] (0/2) Epoch 17, batch 10550, giga_loss[loss=0.3306, simple_loss=0.3731, pruned_loss=0.1441, over 23294.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3643, pruned_loss=0.1154, over 5669628.58 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3485, pruned_loss=0.09432, over 5735768.04 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3666, pruned_loss=0.118, over 5660853.96 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:40:48,594 INFO [train.py:968] (0/2) Epoch 17, batch 10600, giga_loss[loss=0.275, simple_loss=0.339, pruned_loss=0.1055, over 28921.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3648, pruned_loss=0.1155, over 5651909.04 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3487, pruned_loss=0.09428, over 5739220.17 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3669, pruned_loss=0.1182, over 5639943.06 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:41:00,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-08 18:41:09,799 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-08 18:41:13,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6550, 1.7728, 1.5505, 1.7397], device='cuda:0'), covar=tensor([0.2289, 0.2220, 0.2208, 0.2008], device='cuda:0'), in_proj_covar=tensor([0.1429, 0.1035, 0.1265, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 18:41:26,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.487e+03 1.926e+03 2.379e+03 8.477e+03, threshold=3.852e+03, percent-clipped=9.0
+2023-03-08 18:41:39,139 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=741528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:41:39,655 INFO [train.py:968] (0/2) Epoch 17, batch 10650, giga_loss[loss=0.2791, simple_loss=0.3502, pruned_loss=0.1041, over 28983.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3663, pruned_loss=0.1172, over 5638652.83 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3487, pruned_loss=0.09425, over 5737668.23 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3683, pruned_loss=0.1198, over 5628933.51 frames. ], batch size: 136, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:42:23,399 INFO [train.py:968] (0/2) Epoch 17, batch 10700, libri_loss[loss=0.2121, simple_loss=0.2988, pruned_loss=0.06267, over 28471.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3678, pruned_loss=0.1184, over 5647771.19 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3489, pruned_loss=0.09459, over 5732954.06 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3698, pruned_loss=0.121, over 5640816.80 frames. ], batch size: 63, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:42:40,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3962, 1.7777, 1.3746, 1.6391], device='cuda:0'), covar=tensor([0.2535, 0.2571, 0.2888, 0.2212], device='cuda:0'), in_proj_covar=tensor([0.1428, 0.1035, 0.1265, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 18:43:04,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.935e+03 2.539e+03 3.836e+03 1.649e+04, threshold=5.078e+03, percent-clipped=24.0
+2023-03-08 18:43:11,193 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7524, 1.8207, 1.3613, 1.3181], device='cuda:0'), covar=tensor([0.0898, 0.0621, 0.1038, 0.1140], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0445, 0.0510, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 18:43:15,800 INFO [train.py:968] (0/2) Epoch 17, batch 10750, giga_loss[loss=0.3464, simple_loss=0.4085, pruned_loss=0.1422, over 28490.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3697, pruned_loss=0.1197, over 5646038.34 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3488, pruned_loss=0.09451, over 5734515.79 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3719, pruned_loss=0.1224, over 5637147.37 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:43:35,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=741648.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:44:05,543 INFO [train.py:968] (0/2) Epoch 17, batch 10800, giga_loss[loss=0.285, simple_loss=0.3565, pruned_loss=0.1067, over 29036.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3717, pruned_loss=0.1213, over 5652796.05 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3483, pruned_loss=0.09418, over 5737080.66 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3743, pruned_loss=0.1241, over 5642049.55 frames. ], batch size: 155, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:44:40,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.216e+02 1.533e+03 2.036e+03 2.890e+03 5.521e+03, threshold=4.071e+03, percent-clipped=1.0
+2023-03-08 18:44:53,145 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=741728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:44:53,484 INFO [train.py:968] (0/2) Epoch 17, batch 10850, giga_loss[loss=0.3468, simple_loss=0.3956, pruned_loss=0.1491, over 27624.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3742, pruned_loss=0.1236, over 5652465.44 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3488, pruned_loss=0.09452, over 5736440.72 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3762, pruned_loss=0.1261, over 5642986.02 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:45:40,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 18:45:41,814 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=741777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:45:43,917 INFO [train.py:968] (0/2) Epoch 17, batch 10900, libri_loss[loss=0.2909, simple_loss=0.3706, pruned_loss=0.1056, over 29358.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.375, pruned_loss=0.1244, over 5657101.81 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3488, pruned_loss=0.09449, over 5739138.34 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3771, pruned_loss=0.127, over 5645476.34 frames. ], batch size: 92, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:45:57,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=741791.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:46:00,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=741794.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:46:02,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5039, 1.6399, 1.7387, 1.2902], device='cuda:0'), covar=tensor([0.1698, 0.2556, 0.1441, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.0870, 0.0699, 0.0917, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 18:46:27,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.676e+03 1.937e+03 2.582e+03 9.374e+03, threshold=3.875e+03, percent-clipped=6.0
+2023-03-08 18:46:30,240 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=741823.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:46:36,601 INFO [train.py:968] (0/2) Epoch 17, batch 10950, giga_loss[loss=0.3029, simple_loss=0.3543, pruned_loss=0.1258, over 23831.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3759, pruned_loss=0.1236, over 5653308.49 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3489, pruned_loss=0.09462, over 5738950.80 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3779, pruned_loss=0.1259, over 5643129.38 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:47:14,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5986, 1.5408, 1.2182, 1.1567], device='cuda:0'), covar=tensor([0.0640, 0.0360, 0.0813, 0.1060], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0446, 0.0511, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 18:47:20,018 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=741871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:47:22,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6690, 1.6411, 1.2359, 1.2274], device='cuda:0'), covar=tensor([0.0620, 0.0488, 0.0808, 0.1152], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0446, 0.0510, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 18:47:23,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=741874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:47:28,492 INFO [train.py:968] (0/2) Epoch 17, batch 11000, giga_loss[loss=0.3162, simple_loss=0.3795, pruned_loss=0.1265, over 28976.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3753, pruned_loss=0.1239, over 5641411.78 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3488, pruned_loss=0.09463, over 5733035.21 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3775, pruned_loss=0.1263, over 5637140.36 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:47:54,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=741903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:47:54,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=741903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:48:07,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.691e+02 1.729e+03 2.407e+03 3.285e+03 9.361e+03, threshold=4.813e+03, percent-clipped=19.0
+2023-03-08 18:48:18,427 INFO [train.py:968] (0/2) Epoch 17, batch 11050, giga_loss[loss=0.2914, simple_loss=0.3609, pruned_loss=0.111, over 28506.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3722, pruned_loss=0.1219, over 5658493.81 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.349, pruned_loss=0.09469, over 5736536.89 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3743, pruned_loss=0.1244, over 5650075.15 frames. ], batch size: 307, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:48:49,821 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=741953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:49:17,525 INFO [train.py:968] (0/2) Epoch 17, batch 11100, libri_loss[loss=0.2917, simple_loss=0.3564, pruned_loss=0.1135, over 29354.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3726, pruned_loss=0.1228, over 5647399.05 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3494, pruned_loss=0.09504, over 5730709.23 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3745, pruned_loss=0.1253, over 5642563.41 frames. ], batch size: 71, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:49:36,020 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-742000.pt
+2023-03-08 18:49:52,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.613e+03 1.912e+03 2.475e+03 4.211e+03, threshold=3.824e+03, percent-clipped=0.0
+2023-03-08 18:50:01,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1203, 1.2576, 3.6632, 3.1652], device='cuda:0'), covar=tensor([0.1691, 0.2587, 0.0467, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0724, 0.0618, 0.0914, 0.0843], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 18:50:02,641 INFO [train.py:968] (0/2) Epoch 17, batch 11150, giga_loss[loss=0.3877, simple_loss=0.418, pruned_loss=0.1788, over 26580.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3701, pruned_loss=0.1207, over 5666717.39 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3492, pruned_loss=0.09491, over 5735536.34 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3723, pruned_loss=0.1233, over 5656961.83 frames. ], batch size: 555, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:50:21,425 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:50:23,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=742049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:50:45,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=742073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:50:49,650 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=742078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:50:50,630 INFO [train.py:968] (0/2) Epoch 17, batch 11200, giga_loss[loss=0.2942, simple_loss=0.3656, pruned_loss=0.1114, over 28920.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3707, pruned_loss=0.1218, over 5671086.60 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3493, pruned_loss=0.09488, over 5738206.09 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3727, pruned_loss=0.1243, over 5660213.59 frames. ], batch size: 227, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:50:53,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4636, 1.6559, 1.3383, 1.7352], device='cuda:0'), covar=tensor([0.2441, 0.2571, 0.2845, 0.2168], device='cuda:0'), in_proj_covar=tensor([0.1425, 0.1034, 0.1264, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 18:51:29,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2904, 3.4332, 1.4344, 1.4291], device='cuda:0'), covar=tensor([0.1039, 0.0319, 0.0908, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0540, 0.0366, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 18:51:30,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.926e+02 1.629e+03 1.996e+03 2.619e+03 6.806e+03, threshold=3.991e+03, percent-clipped=12.0
+2023-03-08 18:51:40,105 INFO [train.py:968] (0/2) Epoch 17, batch 11250, giga_loss[loss=0.3244, simple_loss=0.3908, pruned_loss=0.129, over 28950.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3709, pruned_loss=0.1219, over 5671588.27 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3494, pruned_loss=0.09482, over 5741286.60 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3728, pruned_loss=0.1245, over 5658903.52 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:52:03,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=742152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:52:26,360 INFO [train.py:968] (0/2) Epoch 17, batch 11300, giga_loss[loss=0.2944, simple_loss=0.3653, pruned_loss=0.1118, over 28816.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3723, pruned_loss=0.1229, over 5672354.33 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3499, pruned_loss=0.09486, over 5744190.73 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3743, pruned_loss=0.126, over 5656639.65 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:52:47,363 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2191, 0.8494, 0.9221, 1.3716], device='cuda:0'), covar=tensor([0.0707, 0.0406, 0.0338, 0.0765], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 18:53:04,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.784e+02 1.683e+03 2.339e+03 3.406e+03 1.573e+04, threshold=4.678e+03, percent-clipped=20.0
+2023-03-08 18:53:13,132 INFO [train.py:968] (0/2) Epoch 17, batch 11350, giga_loss[loss=0.3092, simple_loss=0.3786, pruned_loss=0.1199, over 28688.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3746, pruned_loss=0.1247, over 5676959.28 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3502, pruned_loss=0.09485, over 5748642.48 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3766, pruned_loss=0.1282, over 5657827.00 frames. ], batch size: 262, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:54:03,673 INFO [train.py:968] (0/2) Epoch 17, batch 11400, giga_loss[loss=0.3659, simple_loss=0.3996, pruned_loss=0.166, over 28636.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3753, pruned_loss=0.1251, over 5678162.56 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3504, pruned_loss=0.09495, over 5750253.45 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3769, pruned_loss=0.1281, over 5661138.61 frames. ], batch size: 92, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:54:17,747 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:54:24,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=742298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:54:33,975 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=742309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:54:43,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.505e+02 1.724e+03 2.313e+03 3.041e+03 5.469e+03, threshold=4.626e+03, percent-clipped=4.0
+2023-03-08 18:54:51,927 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=742327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:54:55,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=742328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:54:55,700 INFO [train.py:968] (0/2) Epoch 17, batch 11450, giga_loss[loss=0.3042, simple_loss=0.3578, pruned_loss=0.1253, over 28869.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3769, pruned_loss=0.1273, over 5665110.09 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3506, pruned_loss=0.09502, over 5743367.04 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3784, pruned_loss=0.1301, over 5656273.03 frames. ], batch size: 99, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:55:07,118 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=742339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:55:07,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3930, 1.1594, 4.3481, 3.4430], device='cuda:0'), covar=tensor([0.1653, 0.2866, 0.0402, 0.1095], device='cuda:0'), in_proj_covar=tensor([0.0724, 0.0619, 0.0916, 0.0843], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 18:55:44,585 INFO [train.py:968] (0/2) Epoch 17, batch 11500, giga_loss[loss=0.282, simple_loss=0.353, pruned_loss=0.1055, over 28609.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3765, pruned_loss=0.1274, over 5661162.15 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3509, pruned_loss=0.09517, over 5744823.92 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3779, pruned_loss=0.1301, over 5651314.85 frames. ], batch size: 60, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:56:25,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.576e+03 2.086e+03 2.984e+03 5.440e+03, threshold=4.172e+03, percent-clipped=5.0
+2023-03-08 18:56:34,819 INFO [train.py:968] (0/2) Epoch 17, batch 11550, giga_loss[loss=0.296, simple_loss=0.3632, pruned_loss=0.1144, over 28617.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3742, pruned_loss=0.125, over 5674262.82 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3507, pruned_loss=0.09512, over 5748244.08 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3759, pruned_loss=0.1277, over 5662035.55 frames. ], batch size: 307, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:56:53,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=742448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:57:15,955 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:57:17,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=742474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:57:22,603 INFO [train.py:968] (0/2) Epoch 17, batch 11600, giga_loss[loss=0.2842, simple_loss=0.3598, pruned_loss=0.1043, over 28932.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3751, pruned_loss=0.1252, over 5671809.74 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3508, pruned_loss=0.09517, over 5750473.33 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3767, pruned_loss=0.1278, over 5658983.55 frames. ], batch size: 174, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:57:47,882 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=742503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:57:59,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2444, 1.4679, 1.3437, 1.1264], device='cuda:0'), covar=tensor([0.2213, 0.2215, 0.1462, 0.2083], device='cuda:0'), in_proj_covar=tensor([0.1872, 0.1832, 0.1744, 0.1884], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 18:57:59,736 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-08 18:58:08,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.668e+03 2.582e+03 3.974e+03 9.693e+03, threshold=5.165e+03, percent-clipped=22.0
+2023-03-08 18:58:16,882 INFO [train.py:968] (0/2) Epoch 17, batch 11650, giga_loss[loss=0.3917, simple_loss=0.4359, pruned_loss=0.1737, over 28288.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3757, pruned_loss=0.1254, over 5682193.54 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3509, pruned_loss=0.09519, over 5752423.01 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3772, pruned_loss=0.1278, over 5669304.33 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:58:39,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2884, 1.4815, 1.3677, 1.2651], device='cuda:0'), covar=tensor([0.2068, 0.1709, 0.1788, 0.1742], device='cuda:0'), in_proj_covar=tensor([0.1865, 0.1828, 0.1738, 0.1879], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 18:59:05,917 INFO [train.py:968] (0/2) Epoch 17, batch 11700, giga_loss[loss=0.3773, simple_loss=0.435, pruned_loss=0.1598, over 29031.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3772, pruned_loss=0.127, over 5675004.16 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3506, pruned_loss=0.09497, over 5751775.38 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3795, pruned_loss=0.13, over 5663251.36 frames. ], batch size: 155, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:59:17,052 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:59:18,447 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-08 18:59:18,934 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=742594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:59:41,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.624e+03 1.999e+03 2.831e+03 8.821e+03, threshold=3.998e+03, percent-clipped=2.0
+2023-03-08 18:59:45,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=742623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:59:45,996 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=742623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:59:52,000 INFO [train.py:968] (0/2) Epoch 17, batch 11750, giga_loss[loss=0.3018, simple_loss=0.3704, pruned_loss=0.1165, over 28711.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3758, pruned_loss=0.1253, over 5691898.51 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3504, pruned_loss=0.09485, over 5757663.98 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3787, pruned_loss=0.1289, over 5675006.18 frames. ], batch size: 242, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:00:01,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6616, 4.5643, 1.6438, 1.8999], device='cuda:0'), covar=tensor([0.0920, 0.0387, 0.0909, 0.1213], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0542, 0.0367, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 19:00:36,928 INFO [train.py:968] (0/2) Epoch 17, batch 11800, giga_loss[loss=0.3255, simple_loss=0.3862, pruned_loss=0.1324, over 28904.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3759, pruned_loss=0.1245, over 5691858.53 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3503, pruned_loss=0.09485, over 5759874.07 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3788, pruned_loss=0.128, over 5675067.98 frames. ], batch size: 227, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:00:41,077 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-08 19:00:44,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=742684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:01:00,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3380, 1.6312, 1.5778, 1.4371], device='cuda:0'), covar=tensor([0.1931, 0.2023, 0.2366, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0748, 0.0703, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 19:01:13,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=742714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:01:17,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.988e+02 1.716e+03 2.229e+03 3.128e+03 1.305e+04, threshold=4.459e+03, percent-clipped=10.0
+2023-03-08 19:01:27,493 INFO [train.py:968] (0/2) Epoch 17, batch 11850, giga_loss[loss=0.3299, simple_loss=0.3852, pruned_loss=0.1373, over 28596.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3746, pruned_loss=0.1231, over 5680758.75 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3494, pruned_loss=0.09434, over 5763037.55 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3783, pruned_loss=0.1269, over 5663098.50 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 19:01:54,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8015, 1.8350, 1.4038, 1.3751], device='cuda:0'), covar=tensor([0.0902, 0.0625, 0.1072, 0.1111], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0447, 0.0512, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:02:16,294 INFO [train.py:968] (0/2) Epoch 17, batch 11900, giga_loss[loss=0.3682, simple_loss=0.4106, pruned_loss=0.1629, over 27659.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3737, pruned_loss=0.1224, over 5682781.09 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3497, pruned_loss=0.09461, over 5762554.24 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3768, pruned_loss=0.1258, over 5667758.14 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 19:02:53,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.719e+02 1.489e+03 1.972e+03 2.572e+03 5.982e+03, threshold=3.944e+03, percent-clipped=2.0
+2023-03-08 19:02:58,303 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:02:59,405 INFO [train.py:968] (0/2) Epoch 17, batch 11950, giga_loss[loss=0.2908, simple_loss=0.3618, pruned_loss=0.1099, over 28874.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3726, pruned_loss=0.1217, over 5695898.53 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3496, pruned_loss=0.09464, over 5766382.11 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3759, pruned_loss=0.1252, over 5677781.66 frames. ], batch size: 174, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 19:03:00,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=742830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:03:27,646 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:03:28,749 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=742859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:03:31,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=742860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:03:48,687 INFO [train.py:968] (0/2) Epoch 17, batch 12000, giga_loss[loss=0.3307, simple_loss=0.3942, pruned_loss=0.1336, over 28794.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3742, pruned_loss=0.1235, over 5675783.59 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3496, pruned_loss=0.09464, over 5769445.99 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3775, pruned_loss=0.127, over 5656885.70 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:03:48,692 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 19:03:54,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5212, 1.7738, 1.2951, 1.3753], device='cuda:0'), covar=tensor([0.0943, 0.0490, 0.1013, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0445, 0.0510, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:03:57,324 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2118, simple_loss=0.3186, pruned_loss=0.0525, over 944034.00 frames. 
+2023-03-08 19:03:57,325 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 19:03:59,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0201, 3.8380, 3.6389, 1.9824], device='cuda:0'), covar=tensor([0.0653, 0.0790, 0.0751, 0.2104], device='cuda:0'), in_proj_covar=tensor([0.1179, 0.1091, 0.0938, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 19:03:59,890 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-08 19:04:06,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=742889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:04:35,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.776e+02 1.584e+03 1.970e+03 2.435e+03 5.772e+03, threshold=3.940e+03, percent-clipped=4.0
+2023-03-08 19:04:44,186 INFO [train.py:968] (0/2) Epoch 17, batch 12050, giga_loss[loss=0.2812, simple_loss=0.3506, pruned_loss=0.1059, over 28530.00 frames. ], tot_loss[loss=0.31, simple_loss=0.374, pruned_loss=0.1229, over 5678024.84 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3491, pruned_loss=0.09444, over 5772357.36 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3776, pruned_loss=0.1265, over 5658695.90 frames. ], batch size: 85, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:04:48,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5534, 1.7015, 1.1853, 1.2843], device='cuda:0'), covar=tensor([0.0878, 0.0552, 0.1036, 0.1087], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0447, 0.0511, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:05:32,089 INFO [train.py:968] (0/2) Epoch 17, batch 12100, giga_loss[loss=0.3354, simple_loss=0.3895, pruned_loss=0.1407, over 27958.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.375, pruned_loss=0.1247, over 5678044.56 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3495, pruned_loss=0.09466, over 5775813.54 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3781, pruned_loss=0.128, over 5657487.26 frames. ], batch size: 412, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:05:49,367 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8073, 1.9005, 1.3648, 1.4668], device='cuda:0'), covar=tensor([0.0866, 0.0595, 0.0972, 0.1059], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0446, 0.0510, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:05:51,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=742998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:06:04,543 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-08 19:06:13,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.654e+03 1.965e+03 2.592e+03 7.491e+03, threshold=3.931e+03, percent-clipped=7.0
+2023-03-08 19:06:20,105 INFO [train.py:968] (0/2) Epoch 17, batch 12150, giga_loss[loss=0.3424, simple_loss=0.3965, pruned_loss=0.1441, over 28587.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3757, pruned_loss=0.1258, over 5672667.74 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3496, pruned_loss=0.0947, over 5773981.27 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3787, pruned_loss=0.1292, over 5655152.94 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:07:06,747 INFO [train.py:968] (0/2) Epoch 17, batch 12200, giga_loss[loss=0.3153, simple_loss=0.3708, pruned_loss=0.1299, over 28939.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3758, pruned_loss=0.1256, over 5673530.93 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3496, pruned_loss=0.0947, over 5768976.00 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3791, pruned_loss=0.1295, over 5660185.93 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:07:30,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3296, 1.5832, 1.4481, 1.5287], device='cuda:0'), covar=tensor([0.0788, 0.0328, 0.0311, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 19:07:45,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.564e+02 1.540e+03 2.158e+03 2.877e+03 6.786e+03, threshold=4.316e+03, percent-clipped=10.0
+2023-03-08 19:07:52,945 INFO [train.py:968] (0/2) Epoch 17, batch 12250, libri_loss[loss=0.258, simple_loss=0.3422, pruned_loss=0.0869, over 29518.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3762, pruned_loss=0.1253, over 5678678.75 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3498, pruned_loss=0.09475, over 5775978.78 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3802, pruned_loss=0.1301, over 5656090.90 frames. ], batch size: 82, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:07:58,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=743136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:08:04,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=743141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:08:06,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=743144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:08:33,471 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:08:38,441 INFO [train.py:968] (0/2) Epoch 17, batch 12300, giga_loss[loss=0.3157, simple_loss=0.373, pruned_loss=0.1291, over 27959.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.375, pruned_loss=0.1249, over 5666825.46 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3495, pruned_loss=0.09458, over 5777605.77 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3794, pruned_loss=0.1298, over 5644308.85 frames. ], batch size: 412, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:09:00,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4507, 1.6329, 1.6791, 1.2944], device='cuda:0'), covar=tensor([0.1604, 0.2331, 0.1337, 0.1591], device='cuda:0'), in_proj_covar=tensor([0.0869, 0.0698, 0.0917, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 19:09:21,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.600e+03 1.988e+03 3.211e+03 8.732e+03, threshold=3.977e+03, percent-clipped=9.0
+2023-03-08 19:09:28,730 INFO [train.py:968] (0/2) Epoch 17, batch 12350, giga_loss[loss=0.2818, simple_loss=0.3522, pruned_loss=0.1057, over 28820.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3752, pruned_loss=0.1248, over 5652671.66 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3498, pruned_loss=0.09477, over 5770559.14 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3788, pruned_loss=0.129, over 5639333.88 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:10:15,905 INFO [train.py:968] (0/2) Epoch 17, batch 12400, giga_loss[loss=0.3254, simple_loss=0.3914, pruned_loss=0.1297, over 28888.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3761, pruned_loss=0.1255, over 5653681.30 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3495, pruned_loss=0.09463, over 5771052.34 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3792, pruned_loss=0.1291, over 5642339.04 frames. ], batch size: 174, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 19:10:20,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5279, 1.1906, 4.3770, 3.3923], device='cuda:0'), covar=tensor([0.1631, 0.2937, 0.0417, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.0731, 0.0624, 0.0923, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:10:58,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.238e+02 1.579e+03 1.963e+03 2.575e+03 6.424e+03, threshold=3.925e+03, percent-clipped=9.0
+2023-03-08 19:11:05,023 INFO [train.py:968] (0/2) Epoch 17, batch 12450, libri_loss[loss=0.2899, simple_loss=0.3745, pruned_loss=0.1027, over 26067.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3755, pruned_loss=0.1251, over 5645558.33 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3497, pruned_loss=0.09481, over 5760428.98 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3784, pruned_loss=0.1285, over 5643498.52 frames. ], batch size: 136, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:11:18,788 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=743342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:11:46,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5506, 1.5650, 1.1811, 1.2181], device='cuda:0'), covar=tensor([0.0694, 0.0361, 0.0889, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0444, 0.0507, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:11:50,305 INFO [train.py:968] (0/2) Epoch 17, batch 12500, giga_loss[loss=0.282, simple_loss=0.3498, pruned_loss=0.1071, over 28866.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3739, pruned_loss=0.1241, over 5638982.14 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3501, pruned_loss=0.09494, over 5744106.83 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3766, pruned_loss=0.1276, over 5648406.30 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:12:26,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3661, 4.2100, 4.0083, 1.8998], device='cuda:0'), covar=tensor([0.0578, 0.0707, 0.0704, 0.2043], device='cuda:0'), in_proj_covar=tensor([0.1177, 0.1090, 0.0934, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 19:12:28,927 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4302, 1.5245, 1.2834, 1.1342], device='cuda:0'), covar=tensor([0.0860, 0.0543, 0.0963, 0.1046], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0445, 0.0509, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:12:31,326 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.821e+03 2.326e+03 3.252e+03 6.704e+03, threshold=4.652e+03, percent-clipped=10.0
+2023-03-08 19:12:33,093 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.29 vs. limit=5.0
+2023-03-08 19:12:37,657 INFO [train.py:968] (0/2) Epoch 17, batch 12550, giga_loss[loss=0.2985, simple_loss=0.3615, pruned_loss=0.1177, over 28718.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3715, pruned_loss=0.1229, over 5658921.73 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3497, pruned_loss=0.09476, over 5748923.69 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3746, pruned_loss=0.1266, over 5659726.01 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:13:26,300 INFO [train.py:968] (0/2) Epoch 17, batch 12600, giga_loss[loss=0.3333, simple_loss=0.3753, pruned_loss=0.1456, over 28935.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3678, pruned_loss=0.1213, over 5656530.35 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3494, pruned_loss=0.09451, over 5752751.46 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3711, pruned_loss=0.1251, over 5651601.14 frames. ], batch size: 136, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:13:58,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=743511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:14:09,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.164e+02 1.529e+03 1.974e+03 2.668e+03 1.061e+04, threshold=3.949e+03, percent-clipped=5.0
+2023-03-08 19:14:13,799 INFO [train.py:968] (0/2) Epoch 17, batch 12650, giga_loss[loss=0.3334, simple_loss=0.3709, pruned_loss=0.148, over 28490.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3681, pruned_loss=0.1226, over 5655463.82 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3494, pruned_loss=0.09442, over 5754469.89 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.371, pruned_loss=0.1262, over 5648271.89 frames. ], batch size: 85, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:14:29,911 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-08 19:14:31,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 19:15:05,621 INFO [train.py:968] (0/2) Epoch 17, batch 12700, libri_loss[loss=0.2244, simple_loss=0.3034, pruned_loss=0.07269, over 29632.00 frames. ], tot_loss[loss=0.306, simple_loss=0.367, pruned_loss=0.1225, over 5647611.60 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3488, pruned_loss=0.09405, over 5757347.52 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3703, pruned_loss=0.1263, over 5637335.19 frames. ], batch size: 69, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 19:15:50,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.606e+02 1.576e+03 2.519e+03 3.845e+03 9.746e+03, threshold=5.038e+03, percent-clipped=20.0
+2023-03-08 19:15:55,801 INFO [train.py:968] (0/2) Epoch 17, batch 12750, libri_loss[loss=0.257, simple_loss=0.3262, pruned_loss=0.09394, over 29364.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3666, pruned_loss=0.1212, over 5656840.22 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.349, pruned_loss=0.09425, over 5759248.80 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3694, pruned_loss=0.1246, over 5644949.27 frames. ], batch size: 71, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 19:16:00,292 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 19:16:03,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4642, 1.9421, 1.4227, 0.7716], device='cuda:0'), covar=tensor([0.4328, 0.2545, 0.3169, 0.5013], device='cuda:0'), in_proj_covar=tensor([0.1673, 0.1590, 0.1555, 0.1376], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 19:16:21,628 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=743654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:16:23,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=743657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:16:26,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4433, 1.5739, 1.4085, 1.6214], device='cuda:0'), covar=tensor([0.0679, 0.0301, 0.0302, 0.0680], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 19:16:31,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5428, 2.2158, 1.5741, 0.7726], device='cuda:0'), covar=tensor([0.5315, 0.2867, 0.3923, 0.5632], device='cuda:0'), in_proj_covar=tensor([0.1669, 0.1586, 0.1553, 0.1374], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 19:16:45,369 INFO [train.py:968] (0/2) Epoch 17, batch 12800, giga_loss[loss=0.288, simple_loss=0.3763, pruned_loss=0.09987, over 28876.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3656, pruned_loss=0.1182, over 5655440.42 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.349, pruned_loss=0.09438, over 5758589.43 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3683, pruned_loss=0.1214, over 5644032.74 frames. ], batch size: 174, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:16:52,407 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:17:09,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5210, 1.2897, 4.0881, 3.5510], device='cuda:0'), covar=tensor([0.1554, 0.2922, 0.0433, 0.1167], device='cuda:0'), in_proj_covar=tensor([0.0730, 0.0625, 0.0923, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:17:26,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=743717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:17:31,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.084e+02 1.432e+03 1.865e+03 2.229e+03 4.940e+03, threshold=3.731e+03, percent-clipped=0.0
+2023-03-08 19:17:38,351 INFO [train.py:968] (0/2) Epoch 17, batch 12850, giga_loss[loss=0.2984, simple_loss=0.3675, pruned_loss=0.1147, over 28877.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3621, pruned_loss=0.1143, over 5654020.50 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3483, pruned_loss=0.09409, over 5762413.86 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3651, pruned_loss=0.1175, over 5639440.80 frames. ], batch size: 227, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:18:30,064 INFO [train.py:968] (0/2) Epoch 17, batch 12900, giga_loss[loss=0.2889, simple_loss=0.3554, pruned_loss=0.1112, over 27523.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3591, pruned_loss=0.1104, over 5653730.66 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3482, pruned_loss=0.09395, over 5763201.60 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3618, pruned_loss=0.1135, over 5639558.73 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:19:09,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4904, 1.6109, 1.7367, 1.3237], device='cuda:0'), covar=tensor([0.1604, 0.2384, 0.1375, 0.1671], device='cuda:0'), in_proj_covar=tensor([0.0870, 0.0695, 0.0916, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 19:19:19,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.650e+02 1.447e+03 1.943e+03 2.765e+03 6.150e+03, threshold=3.885e+03, percent-clipped=11.0
+2023-03-08 19:19:24,938 INFO [train.py:968] (0/2) Epoch 17, batch 12950, giga_loss[loss=0.2567, simple_loss=0.337, pruned_loss=0.0882, over 28645.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.356, pruned_loss=0.1077, over 5649199.83 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.348, pruned_loss=0.0939, over 5766267.74 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3586, pruned_loss=0.1105, over 5633528.72 frames. ], batch size: 307, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:19:34,466 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-08 19:19:45,702 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-08 19:19:54,887 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=743860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:19:58,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=743863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:20:13,495 INFO [train.py:968] (0/2) Epoch 17, batch 13000, giga_loss[loss=0.288, simple_loss=0.3689, pruned_loss=0.1035, over 28932.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3549, pruned_loss=0.1046, over 5662387.65 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3481, pruned_loss=0.09417, over 5768602.07 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3571, pruned_loss=0.1069, over 5644939.96 frames. ], batch size: 227, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:20:24,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3045, 1.4703, 1.3034, 1.5251], device='cuda:0'), covar=tensor([0.0792, 0.0357, 0.0355, 0.0890], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 19:20:29,137 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:21:03,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.144e+02 1.398e+03 1.969e+03 2.847e+03 7.200e+03, threshold=3.937e+03, percent-clipped=10.0
+2023-03-08 19:21:06,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4177, 1.6854, 1.5783, 1.4571], device='cuda:0'), covar=tensor([0.2116, 0.1801, 0.1306, 0.1539], device='cuda:0'), in_proj_covar=tensor([0.1850, 0.1805, 0.1722, 0.1861], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 19:21:09,842 INFO [train.py:968] (0/2) Epoch 17, batch 13050, giga_loss[loss=0.2448, simple_loss=0.3334, pruned_loss=0.07811, over 28889.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3559, pruned_loss=0.105, over 5656094.40 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.348, pruned_loss=0.0942, over 5766188.13 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3578, pruned_loss=0.1069, over 5642229.74 frames. ], batch size: 145, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:21:36,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=743955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:22:02,126 INFO [train.py:968] (0/2) Epoch 17, batch 13100, giga_loss[loss=0.3085, simple_loss=0.3616, pruned_loss=0.1277, over 26599.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3554, pruned_loss=0.1047, over 5652885.36 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3479, pruned_loss=0.09423, over 5764754.40 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3571, pruned_loss=0.1064, over 5641652.11 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:22:22,151 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-744000.pt
+2023-03-08 19:22:45,921 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.824e+02 1.433e+03 1.803e+03 2.472e+03 6.903e+03, threshold=3.607e+03, percent-clipped=6.0
+2023-03-08 19:22:47,834 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:22:52,410 INFO [train.py:968] (0/2) Epoch 17, batch 13150, giga_loss[loss=0.2571, simple_loss=0.3177, pruned_loss=0.09825, over 24106.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3516, pruned_loss=0.1021, over 5639310.88 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3477, pruned_loss=0.09426, over 5758294.30 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3533, pruned_loss=0.1036, over 5633407.46 frames. ], batch size: 705, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:23:42,350 INFO [train.py:968] (0/2) Epoch 17, batch 13200, giga_loss[loss=0.264, simple_loss=0.3432, pruned_loss=0.09241, over 29022.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3505, pruned_loss=0.1014, over 5647026.53 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3475, pruned_loss=0.09411, over 5759538.55 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.352, pruned_loss=0.1028, over 5640276.65 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 19:23:44,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3401, 1.5336, 1.4323, 1.3280], device='cuda:0'), covar=tensor([0.2081, 0.1651, 0.1309, 0.1570], device='cuda:0'), in_proj_covar=tensor([0.1845, 0.1795, 0.1711, 0.1850], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 19:24:28,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.296e+02 1.353e+03 1.724e+03 2.499e+03 7.128e+03, threshold=3.448e+03, percent-clipped=10.0
+2023-03-08 19:24:31,952 INFO [train.py:968] (0/2) Epoch 17, batch 13250, giga_loss[loss=0.2567, simple_loss=0.3375, pruned_loss=0.088, over 28649.00 frames. ], tot_loss[loss=0.276, simple_loss=0.35, pruned_loss=0.101, over 5628876.03 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3471, pruned_loss=0.09404, over 5742047.39 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3516, pruned_loss=0.1023, over 5637436.04 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:25:21,234 INFO [train.py:968] (0/2) Epoch 17, batch 13300, giga_loss[loss=0.2568, simple_loss=0.3296, pruned_loss=0.09203, over 27548.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3477, pruned_loss=0.09881, over 5643518.61 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3471, pruned_loss=0.09407, over 5744905.56 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.349, pruned_loss=0.09992, over 5645745.03 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:25:38,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0996, 5.8943, 5.5858, 3.2123], device='cuda:0'), covar=tensor([0.0470, 0.0629, 0.0807, 0.1509], device='cuda:0'), in_proj_covar=tensor([0.1148, 0.1064, 0.0911, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 19:25:38,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7152, 2.0978, 1.5828, 1.9616], device='cuda:0'), covar=tensor([0.0715, 0.0253, 0.0312, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 19:26:08,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.090e+02 1.417e+03 1.761e+03 2.845e+03 5.724e+03, threshold=3.522e+03, percent-clipped=17.0
+2023-03-08 19:26:14,125 INFO [train.py:968] (0/2) Epoch 17, batch 13350, giga_loss[loss=0.2473, simple_loss=0.3262, pruned_loss=0.0842, over 28256.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3439, pruned_loss=0.0959, over 5645224.75 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3461, pruned_loss=0.09357, over 5749318.58 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3458, pruned_loss=0.09735, over 5639860.72 frames. ], batch size: 368, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:26:49,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2432, 1.3146, 1.2048, 1.0189], device='cuda:0'), covar=tensor([0.0921, 0.0455, 0.0957, 0.0964], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0440, 0.0506, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 19:26:49,955 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:27:08,222 INFO [train.py:968] (0/2) Epoch 17, batch 13400, giga_loss[loss=0.2313, simple_loss=0.3132, pruned_loss=0.07471, over 28757.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3396, pruned_loss=0.0934, over 5649119.31 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3456, pruned_loss=0.09336, over 5753389.77 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3415, pruned_loss=0.09481, over 5638046.76 frames. ], batch size: 284, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:28:01,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.833e+02 1.414e+03 2.006e+03 3.462e+03 1.250e+04, threshold=4.013e+03, percent-clipped=24.0
+2023-03-08 19:28:03,840 INFO [train.py:968] (0/2) Epoch 17, batch 13450, giga_loss[loss=0.2453, simple_loss=0.3243, pruned_loss=0.08314, over 27886.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3388, pruned_loss=0.09334, over 5651182.12 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3457, pruned_loss=0.09352, over 5745167.64 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3401, pruned_loss=0.0943, over 5648668.97 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 19:28:05,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=744330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:28:12,514 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 19:28:52,017 INFO [train.py:968] (0/2) Epoch 17, batch 13500, giga_loss[loss=0.2547, simple_loss=0.3358, pruned_loss=0.08682, over 28592.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3372, pruned_loss=0.09331, over 5644427.76 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3443, pruned_loss=0.0929, over 5748937.09 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3393, pruned_loss=0.0947, over 5634863.25 frames. ], batch size: 242, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 19:29:10,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=744398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:29:13,192 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:29:35,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0507, 1.1869, 3.2995, 2.9116], device='cuda:0'), covar=tensor([0.1663, 0.2784, 0.0540, 0.1094], device='cuda:0'), in_proj_covar=tensor([0.0722, 0.0619, 0.0912, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:29:42,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.652e+02 1.351e+03 1.662e+03 2.150e+03 6.357e+03, threshold=3.324e+03, percent-clipped=3.0
+2023-03-08 19:29:45,130 INFO [train.py:968] (0/2) Epoch 17, batch 13550, giga_loss[loss=0.286, simple_loss=0.3612, pruned_loss=0.1054, over 28976.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.337, pruned_loss=0.09282, over 5650662.93 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3435, pruned_loss=0.0927, over 5746273.42 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3392, pruned_loss=0.09417, over 5639754.42 frames. ], batch size: 213, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 19:30:37,087 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=744473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:30:40,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=744476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:30:42,633 INFO [train.py:968] (0/2) Epoch 17, batch 13600, giga_loss[loss=0.2631, simple_loss=0.3516, pruned_loss=0.0873, over 28554.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3402, pruned_loss=0.09375, over 5648132.98 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3432, pruned_loss=0.09268, over 5748615.99 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.342, pruned_loss=0.09486, over 5635122.61 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:30:57,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4079, 1.7848, 1.4456, 1.5603], device='cuda:0'), covar=tensor([0.0774, 0.0281, 0.0334, 0.0874], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 19:31:16,177 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=744505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:31:39,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.783e+02 1.449e+03 1.883e+03 2.832e+03 7.968e+03, threshold=3.766e+03, percent-clipped=16.0
+2023-03-08 19:31:44,292 INFO [train.py:968] (0/2) Epoch 17, batch 13650, giga_loss[loss=0.2507, simple_loss=0.3308, pruned_loss=0.08529, over 28925.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3407, pruned_loss=0.09337, over 5655115.15 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3432, pruned_loss=0.09289, over 5744838.91 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3421, pruned_loss=0.09403, over 5647037.29 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:31:58,040 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=744541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:32:01,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=744544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:32:35,863 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=744573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:32:42,246 INFO [train.py:968] (0/2) Epoch 17, batch 13700, giga_loss[loss=0.266, simple_loss=0.3182, pruned_loss=0.1069, over 24453.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3401, pruned_loss=0.09297, over 5671124.98 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3424, pruned_loss=0.09253, over 5750139.33 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3419, pruned_loss=0.09388, over 5655665.42 frames. ], batch size: 705, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:33:21,146 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-08 19:33:35,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.809e+02 1.317e+03 1.724e+03 2.271e+03 5.493e+03, threshold=3.448e+03, percent-clipped=6.0
+2023-03-08 19:33:42,650 INFO [train.py:968] (0/2) Epoch 17, batch 13750, giga_loss[loss=0.235, simple_loss=0.3224, pruned_loss=0.07378, over 28033.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3378, pruned_loss=0.09148, over 5671327.19 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3422, pruned_loss=0.09256, over 5749522.44 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3393, pruned_loss=0.09216, over 5658228.68 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:33:49,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=744635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:33:56,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744640.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 19:34:43,040 INFO [train.py:968] (0/2) Epoch 17, batch 13800, giga_loss[loss=0.2832, simple_loss=0.3616, pruned_loss=0.1024, over 28458.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3374, pruned_loss=0.08977, over 5672361.62 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.342, pruned_loss=0.09244, over 5751983.82 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3387, pruned_loss=0.09039, over 5658934.17 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:35:43,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.279e+02 1.257e+03 1.565e+03 2.245e+03 1.125e+04, threshold=3.130e+03, percent-clipped=9.0
+2023-03-08 19:35:50,510 INFO [train.py:968] (0/2) Epoch 17, batch 13850, giga_loss[loss=0.2431, simple_loss=0.3144, pruned_loss=0.0859, over 28938.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3344, pruned_loss=0.08867, over 5664392.24 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3419, pruned_loss=0.09247, over 5753393.33 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3354, pruned_loss=0.08908, over 5651416.64 frames. ], batch size: 213, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:36:02,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9355, 5.1937, 2.2749, 2.3882], device='cuda:0'), covar=tensor([0.0917, 0.0371, 0.0820, 0.1096], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0536, 0.0367, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 19:36:49,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=744776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:36:51,589 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=744778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:36:51,895 INFO [train.py:968] (0/2) Epoch 17, batch 13900, giga_loss[loss=0.2506, simple_loss=0.3276, pruned_loss=0.08679, over 27761.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3325, pruned_loss=0.08823, over 5672608.32 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3414, pruned_loss=0.09229, over 5755710.92 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3336, pruned_loss=0.08865, over 5659193.73 frames. ], batch size: 474, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:36:55,753 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=744781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:37:29,406 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=744810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:37:49,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.376e+02 1.346e+03 1.702e+03 2.392e+03 6.787e+03, threshold=3.405e+03, percent-clipped=16.0
+2023-03-08 19:37:52,132 INFO [train.py:968] (0/2) Epoch 17, batch 13950, giga_loss[loss=0.2372, simple_loss=0.3022, pruned_loss=0.08613, over 24458.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3313, pruned_loss=0.08785, over 5670166.13 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3415, pruned_loss=0.09241, over 5757623.00 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3319, pruned_loss=0.08798, over 5656296.59 frames. ], batch size: 705, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:38:51,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2857, 1.1613, 3.9803, 3.3615], device='cuda:0'), covar=tensor([0.1738, 0.3023, 0.0437, 0.1413], device='cuda:0'), in_proj_covar=tensor([0.0725, 0.0622, 0.0914, 0.0836], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:38:52,228 INFO [train.py:968] (0/2) Epoch 17, batch 14000, giga_loss[loss=0.2637, simple_loss=0.3395, pruned_loss=0.0939, over 27461.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3341, pruned_loss=0.08881, over 5663668.20 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3417, pruned_loss=0.09259, over 5759013.12 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3342, pruned_loss=0.08866, over 5649864.55 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 19:38:59,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:39:00,022 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:39:23,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2538, 2.7729, 1.4567, 1.3416], device='cuda:0'), covar=tensor([0.0927, 0.0332, 0.0877, 0.1348], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0535, 0.0367, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 19:39:42,810 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=744919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:39:43,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6303, 3.4879, 3.2691, 1.9368], device='cuda:0'), covar=tensor([0.0719, 0.0878, 0.0920, 0.1815], device='cuda:0'), in_proj_covar=tensor([0.1145, 0.1058, 0.0906, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 19:39:46,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=744922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:39:50,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.741e+02 1.320e+03 1.689e+03 2.398e+03 7.865e+03, threshold=3.378e+03, percent-clipped=8.0
+2023-03-08 19:39:55,327 INFO [train.py:968] (0/2) Epoch 17, batch 14050, giga_loss[loss=0.2268, simple_loss=0.3164, pruned_loss=0.06857, over 28427.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3349, pruned_loss=0.08853, over 5657879.39 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3412, pruned_loss=0.09242, over 5751715.48 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3353, pruned_loss=0.08848, over 5651873.42 frames. ], batch size: 71, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:40:28,173 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=744951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:41:04,075 INFO [train.py:968] (0/2) Epoch 17, batch 14100, giga_loss[loss=0.245, simple_loss=0.3251, pruned_loss=0.08246, over 28413.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3314, pruned_loss=0.0863, over 5669559.93 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3411, pruned_loss=0.0924, over 5754738.60 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3316, pruned_loss=0.08619, over 5660656.66 frames. ], batch size: 368, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:41:54,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745015.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 19:42:06,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.035e+02 1.517e+03 2.066e+03 2.749e+03 9.893e+03, threshold=4.132e+03, percent-clipped=15.0
+2023-03-08 19:42:13,034 INFO [train.py:968] (0/2) Epoch 17, batch 14150, giga_loss[loss=0.2301, simple_loss=0.3051, pruned_loss=0.0775, over 28540.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3331, pruned_loss=0.08768, over 5680044.46 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3408, pruned_loss=0.09219, over 5756980.70 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3334, pruned_loss=0.08771, over 5669904.77 frames. ], batch size: 78, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:42:32,168 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4164, 4.2389, 4.0331, 1.7885], device='cuda:0'), covar=tensor([0.0494, 0.0692, 0.0671, 0.2244], device='cuda:0'), in_proj_covar=tensor([0.1147, 0.1057, 0.0904, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 19:43:24,213 INFO [train.py:968] (0/2) Epoch 17, batch 14200, giga_loss[loss=0.2453, simple_loss=0.3462, pruned_loss=0.07215, over 28792.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3378, pruned_loss=0.0886, over 5675560.05 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3407, pruned_loss=0.09217, over 5757944.85 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3381, pruned_loss=0.08862, over 5666114.06 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:44:14,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3017, 1.5159, 1.4220, 1.2474], device='cuda:0'), covar=tensor([0.2066, 0.1961, 0.1497, 0.1996], device='cuda:0'), in_proj_covar=tensor([0.1835, 0.1778, 0.1691, 0.1838], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 19:44:25,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.710e+02 1.200e+03 1.565e+03 2.169e+03 5.730e+03, threshold=3.129e+03, percent-clipped=4.0
+2023-03-08 19:44:30,846 INFO [train.py:968] (0/2) Epoch 17, batch 14250, giga_loss[loss=0.3435, simple_loss=0.4012, pruned_loss=0.1429, over 27760.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3414, pruned_loss=0.08853, over 5673560.46 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3407, pruned_loss=0.09217, over 5757944.85 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3417, pruned_loss=0.08854, over 5666208.53 frames. ], batch size: 474, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:44:32,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3600, 1.5216, 1.4706, 1.3730], device='cuda:0'), covar=tensor([0.1764, 0.1560, 0.1594, 0.1660], device='cuda:0'), in_proj_covar=tensor([0.1834, 0.1777, 0.1690, 0.1837], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 19:45:05,680 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2657, 1.4963, 1.5131, 1.1270], device='cuda:0'), covar=tensor([0.1570, 0.2500, 0.1332, 0.1697], device='cuda:0'), in_proj_covar=tensor([0.0875, 0.0694, 0.0921, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 19:45:06,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=745158.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 19:45:10,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=745161.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 19:45:18,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6192, 1.7186, 1.2233, 1.2865], device='cuda:0'), covar=tensor([0.0906, 0.0608, 0.1031, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0438, 0.0506, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 19:45:30,814 INFO [train.py:968] (0/2) Epoch 17, batch 14300, giga_loss[loss=0.2795, simple_loss=0.3512, pruned_loss=0.1039, over 27702.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3423, pruned_loss=0.08817, over 5671697.31 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3408, pruned_loss=0.09217, over 5758796.53 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3424, pruned_loss=0.08811, over 5663205.76 frames. ], batch size: 474, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:45:46,766 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=745190.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 19:46:35,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.153e+02 1.393e+03 1.693e+03 2.281e+03 8.675e+03, threshold=3.385e+03, percent-clipped=10.0
+2023-03-08 19:46:40,175 INFO [train.py:968] (0/2) Epoch 17, batch 14350, giga_loss[loss=0.2419, simple_loss=0.3276, pruned_loss=0.07807, over 29163.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3411, pruned_loss=0.08747, over 5670379.74 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3405, pruned_loss=0.09197, over 5759653.55 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3415, pruned_loss=0.08758, over 5662594.16 frames. ], batch size: 113, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:47:14,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:47:16,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:47:40,464 INFO [train.py:968] (0/2) Epoch 17, batch 14400, libri_loss[loss=0.2834, simple_loss=0.3594, pruned_loss=0.1037, over 29203.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3396, pruned_loss=0.08766, over 5670358.46 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3405, pruned_loss=0.09206, over 5752619.43 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3399, pruned_loss=0.08755, over 5667341.97 frames. ], batch size: 94, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 19:47:44,742 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 19:48:42,595 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.714e+02 1.421e+03 1.997e+03 2.879e+03 5.737e+03, threshold=3.995e+03, percent-clipped=13.0
+2023-03-08 19:48:45,891 INFO [train.py:968] (0/2) Epoch 17, batch 14450, giga_loss[loss=0.265, simple_loss=0.3465, pruned_loss=0.09181, over 28958.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3398, pruned_loss=0.08856, over 5686459.64 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3404, pruned_loss=0.09204, over 5755927.58 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3401, pruned_loss=0.08836, over 5678956.94 frames. ], batch size: 213, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 19:50:07,045 INFO [train.py:968] (0/2) Epoch 17, batch 14500, giga_loss[loss=0.2594, simple_loss=0.3365, pruned_loss=0.09114, over 28941.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.34, pruned_loss=0.09017, over 5675405.93 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3404, pruned_loss=0.09209, over 5750074.59 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3402, pruned_loss=0.08989, over 5672415.15 frames. ], batch size: 285, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:50:45,607 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=745401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:50:46,906 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=745402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:50:49,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=745404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:50:51,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=745405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:51:25,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.228e+02 1.258e+03 1.679e+03 2.529e+03 8.332e+03, threshold=3.358e+03, percent-clipped=8.0
+2023-03-08 19:51:26,140 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=745427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:51:27,620 INFO [train.py:968] (0/2) Epoch 17, batch 14550, giga_loss[loss=0.2526, simple_loss=0.3319, pruned_loss=0.08664, over 28786.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3355, pruned_loss=0.08747, over 5677132.29 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3401, pruned_loss=0.09199, over 5751792.87 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3359, pruned_loss=0.0873, over 5672554.61 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:51:33,386 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=745433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:51:34,201 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=745434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:52:35,769 INFO [train.py:968] (0/2) Epoch 17, batch 14600, libri_loss[loss=0.2639, simple_loss=0.3407, pruned_loss=0.09352, over 25492.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3328, pruned_loss=0.08601, over 5676973.72 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.34, pruned_loss=0.0919, over 5749577.74 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3331, pruned_loss=0.08583, over 5674038.26 frames. ], batch size: 136, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:53:22,101 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=745516.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 19:53:34,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.805e+02 1.266e+03 1.524e+03 1.963e+03 5.008e+03, threshold=3.049e+03, percent-clipped=5.0
+2023-03-08 19:53:37,046 INFO [train.py:968] (0/2) Epoch 17, batch 14650, giga_loss[loss=0.2922, simple_loss=0.3738, pruned_loss=0.1053, over 28893.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3347, pruned_loss=0.0877, over 5664940.96 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.34, pruned_loss=0.09193, over 5739756.65 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3348, pruned_loss=0.08743, over 5669828.23 frames. ], batch size: 227, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:54:03,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4936, 1.6422, 1.3800, 1.5023], device='cuda:0'), covar=tensor([0.2917, 0.2612, 0.3080, 0.2281], device='cuda:0'), in_proj_covar=tensor([0.1422, 0.1026, 0.1263, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 19:54:38,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4359, 1.7370, 1.4744, 1.5276], device='cuda:0'), covar=tensor([0.0702, 0.0385, 0.0326, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 19:54:41,777 INFO [train.py:968] (0/2) Epoch 17, batch 14700, giga_loss[loss=0.2684, simple_loss=0.3457, pruned_loss=0.0956, over 29052.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3395, pruned_loss=0.09077, over 5672225.49 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3394, pruned_loss=0.09165, over 5744034.93 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3401, pruned_loss=0.09075, over 5670712.33 frames. ], batch size: 128, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:55:19,912 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=745611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:55:39,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.904e+02 1.485e+03 1.940e+03 3.126e+03 1.076e+04, threshold=3.880e+03, percent-clipped=25.0
+2023-03-08 19:55:42,860 INFO [train.py:968] (0/2) Epoch 17, batch 14750, libri_loss[loss=0.2418, simple_loss=0.3218, pruned_loss=0.08093, over 29505.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3373, pruned_loss=0.09067, over 5675863.99 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3394, pruned_loss=0.09172, over 5746586.54 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3378, pruned_loss=0.09058, over 5670871.53 frames. ], batch size: 84, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:55:56,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6363, 1.8712, 1.9434, 1.4512], device='cuda:0'), covar=tensor([0.1760, 0.2433, 0.1436, 0.1777], device='cuda:0'), in_proj_covar=tensor([0.0867, 0.0689, 0.0912, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 19:56:46,211 INFO [train.py:968] (0/2) Epoch 17, batch 14800, giga_loss[loss=0.2647, simple_loss=0.3445, pruned_loss=0.0925, over 28597.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3371, pruned_loss=0.09105, over 5672237.25 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3389, pruned_loss=0.0915, over 5740374.80 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3379, pruned_loss=0.09115, over 5672485.24 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 19:57:46,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.044e+02 1.539e+03 2.123e+03 3.023e+03 7.508e+03, threshold=4.245e+03, percent-clipped=17.0
+2023-03-08 19:57:47,253 INFO [train.py:968] (0/2) Epoch 17, batch 14850, giga_loss[loss=0.3233, simple_loss=0.3886, pruned_loss=0.129, over 28646.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3389, pruned_loss=0.09228, over 5673320.66 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3388, pruned_loss=0.09152, over 5734551.52 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3395, pruned_loss=0.09234, over 5677262.95 frames. ], batch size: 85, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:58:59,548 INFO [train.py:968] (0/2) Epoch 17, batch 14900, giga_loss[loss=0.2484, simple_loss=0.3377, pruned_loss=0.07962, over 28690.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3407, pruned_loss=0.09224, over 5677272.05 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3385, pruned_loss=0.09141, over 5736354.43 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3414, pruned_loss=0.09238, over 5678210.46 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:59:31,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4572, 5.2837, 4.9695, 2.4892], device='cuda:0'), covar=tensor([0.0400, 0.0546, 0.0660, 0.1775], device='cuda:0'), in_proj_covar=tensor([0.1142, 0.1054, 0.0903, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 19:59:34,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:00:14,522 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=745826.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:00:17,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.464e+03 2.010e+03 2.537e+03 1.506e+04, threshold=4.020e+03, percent-clipped=7.0
+2023-03-08 20:00:18,016 INFO [train.py:968] (0/2) Epoch 17, batch 14950, giga_loss[loss=0.2476, simple_loss=0.3243, pruned_loss=0.08539, over 28857.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3403, pruned_loss=0.09169, over 5667895.88 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3383, pruned_loss=0.09132, over 5739895.42 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3411, pruned_loss=0.0919, over 5664647.34 frames. ], batch size: 227, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:00:59,868 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 20:01:32,763 INFO [train.py:968] (0/2) Epoch 17, batch 15000, giga_loss[loss=0.258, simple_loss=0.337, pruned_loss=0.08947, over 28662.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3378, pruned_loss=0.09112, over 5667487.34 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.338, pruned_loss=0.09107, over 5744913.90 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3388, pruned_loss=0.09152, over 5658223.53 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:01:32,768 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 20:01:41,315 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.1993, simple_loss=0.2998, pruned_loss=0.04937, over 944034.00 frames. 
+2023-03-08 20:01:41,315 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 20:01:55,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745891.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:02:48,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.975e+02 1.561e+03 2.131e+03 3.203e+03 7.367e+03, threshold=4.261e+03, percent-clipped=15.0
+2023-03-08 20:02:49,588 INFO [train.py:968] (0/2) Epoch 17, batch 15050, giga_loss[loss=0.2341, simple_loss=0.3094, pruned_loss=0.07941, over 28872.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3324, pruned_loss=0.08913, over 5667994.99 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3381, pruned_loss=0.09124, over 5747672.56 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.333, pruned_loss=0.08926, over 5656359.31 frames. ], batch size: 227, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:03:10,169 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=745945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:03:13,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=745948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:03:50,964 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=745977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:03:52,109 INFO [train.py:968] (0/2) Epoch 17, batch 15100, giga_loss[loss=0.261, simple_loss=0.3401, pruned_loss=0.09093, over 28525.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3309, pruned_loss=0.08848, over 5674150.72 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3383, pruned_loss=0.09145, over 5748267.00 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.331, pruned_loss=0.08835, over 5663264.44 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:04:02,481 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:04:15,678 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-746000.pt
+2023-03-08 20:04:26,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5134, 1.4241, 4.2515, 3.1960], device='cuda:0'), covar=tensor([0.1510, 0.2602, 0.0386, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0721, 0.0626, 0.0911, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 20:04:47,798 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-08 20:04:48,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.882e+02 1.564e+03 2.093e+03 2.699e+03 6.902e+03, threshold=4.187e+03, percent-clipped=5.0
+2023-03-08 20:04:49,009 INFO [train.py:968] (0/2) Epoch 17, batch 15150, giga_loss[loss=0.246, simple_loss=0.3316, pruned_loss=0.08025, over 28881.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3323, pruned_loss=0.08992, over 5666019.50 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.338, pruned_loss=0.09139, over 5750410.25 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3325, pruned_loss=0.0898, over 5652673.96 frames. ], batch size: 174, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:04:53,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=746034.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:04:58,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=746037.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:05:32,535 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=746066.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:05:32,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4922, 1.6107, 1.2326, 1.2364], device='cuda:0'), covar=tensor([0.0838, 0.0425, 0.0877, 0.1123], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0437, 0.0506, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 20:05:35,622 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=746069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:05:47,035 INFO [train.py:968] (0/2) Epoch 17, batch 15200, libri_loss[loss=0.2404, simple_loss=0.3221, pruned_loss=0.07933, over 29515.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3313, pruned_loss=0.08894, over 5679064.81 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3376, pruned_loss=0.09132, over 5754734.88 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3316, pruned_loss=0.08885, over 5662037.99 frames. ], batch size: 82, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:06:27,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3903, 3.2224, 2.9934, 1.8203], device='cuda:0'), covar=tensor([0.0773, 0.0983, 0.1039, 0.2056], device='cuda:0'), in_proj_covar=tensor([0.1139, 0.1048, 0.0898, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 20:06:45,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.405e+02 1.271e+03 1.755e+03 2.413e+03 8.181e+03, threshold=3.511e+03, percent-clipped=6.0
+2023-03-08 20:06:45,698 INFO [train.py:968] (0/2) Epoch 17, batch 15250, libri_loss[loss=0.218, simple_loss=0.3009, pruned_loss=0.06753, over 29579.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3289, pruned_loss=0.08698, over 5668609.37 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3367, pruned_loss=0.09067, over 5761113.62 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3298, pruned_loss=0.08738, over 5645883.53 frames. ], batch size: 74, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:06:46,215 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=746129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:06:49,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=746132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:06:51,374 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-08 20:06:57,898 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-08 20:07:23,297 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=746161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:07:25,363 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2977, 1.5668, 1.6142, 1.4414], device='cuda:0'), covar=tensor([0.1801, 0.1788, 0.1995, 0.1749], device='cuda:0'), in_proj_covar=tensor([0.0439, 0.0720, 0.0680, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 20:07:48,988 INFO [train.py:968] (0/2) Epoch 17, batch 15300, giga_loss[loss=0.2206, simple_loss=0.3057, pruned_loss=0.06776, over 29107.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3276, pruned_loss=0.08581, over 5673843.18 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3364, pruned_loss=0.0906, over 5755524.55 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3283, pruned_loss=0.08609, over 5657720.71 frames. ], batch size: 146, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:08:22,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=746201.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:08:32,505 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-08 20:08:58,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.331e+03 1.653e+03 2.522e+03 5.032e+03, threshold=3.305e+03, percent-clipped=6.0
+2023-03-08 20:08:58,826 INFO [train.py:968] (0/2) Epoch 17, batch 15350, libri_loss[loss=0.2027, simple_loss=0.2782, pruned_loss=0.06361, over 29360.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3263, pruned_loss=0.08537, over 5665153.99 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3361, pruned_loss=0.09055, over 5755105.08 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.327, pruned_loss=0.08553, over 5650543.75 frames. ], batch size: 71, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:09:17,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4069, 1.6953, 1.6612, 1.2166], device='cuda:0'), covar=tensor([0.1781, 0.2606, 0.1521, 0.1857], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0684, 0.0911, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 20:09:19,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2015, 4.0340, 3.7893, 1.7606], device='cuda:0'), covar=tensor([0.0637, 0.0750, 0.0770, 0.2095], device='cuda:0'), in_proj_covar=tensor([0.1134, 0.1047, 0.0897, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 20:09:55,453 INFO [train.py:968] (0/2) Epoch 17, batch 15400, giga_loss[loss=0.2465, simple_loss=0.3328, pruned_loss=0.08008, over 29092.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3263, pruned_loss=0.08492, over 5666015.34 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.335, pruned_loss=0.08996, over 5758106.24 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3273, pruned_loss=0.08529, over 5645955.29 frames. ], batch size: 165, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:10:59,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.340e+02 1.275e+03 1.622e+03 2.449e+03 5.843e+03, threshold=3.245e+03, percent-clipped=12.0
+2023-03-08 20:11:00,160 INFO [train.py:968] (0/2) Epoch 17, batch 15450, giga_loss[loss=0.3203, simple_loss=0.3785, pruned_loss=0.131, over 28011.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3269, pruned_loss=0.08567, over 5669805.22 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3348, pruned_loss=0.08996, over 5761289.25 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3277, pruned_loss=0.08585, over 5648808.41 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:11:18,772 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=746344.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:11:23,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=746347.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:12:01,317 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=746376.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:12:05,831 INFO [train.py:968] (0/2) Epoch 17, batch 15500, giga_loss[loss=0.2444, simple_loss=0.3263, pruned_loss=0.0813, over 28748.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3268, pruned_loss=0.08623, over 5666154.78 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3346, pruned_loss=0.08975, over 5763954.50 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3274, pruned_loss=0.08645, over 5644291.09 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:13:01,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.943e+02 1.304e+03 1.592e+03 2.055e+03 4.901e+03, threshold=3.184e+03, percent-clipped=8.0
+2023-03-08 20:13:01,591 INFO [train.py:968] (0/2) Epoch 17, batch 15550, giga_loss[loss=0.2454, simple_loss=0.334, pruned_loss=0.07839, over 28427.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3267, pruned_loss=0.08486, over 5681384.02 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3345, pruned_loss=0.08977, over 5768756.48 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3271, pruned_loss=0.08488, over 5656057.98 frames. ], batch size: 368, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:13:19,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=746444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:14:01,452 INFO [train.py:968] (0/2) Epoch 17, batch 15600, giga_loss[loss=0.2432, simple_loss=0.3344, pruned_loss=0.07602, over 28606.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3299, pruned_loss=0.08541, over 5683918.22 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3346, pruned_loss=0.08985, over 5770926.68 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3299, pruned_loss=0.08521, over 5659780.21 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:14:02,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4873, 3.8164, 1.6200, 1.6715], device='cuda:0'), covar=tensor([0.0919, 0.0259, 0.0871, 0.1235], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0532, 0.0366, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 20:14:09,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5912, 1.7321, 1.1916, 1.3875], device='cuda:0'), covar=tensor([0.0865, 0.0509, 0.0978, 0.1127], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0439, 0.0507, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 20:14:32,702 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-08 20:15:01,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4423, 1.7337, 1.0391, 1.3063], device='cuda:0'), covar=tensor([0.1064, 0.0612, 0.1229, 0.1216], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0437, 0.0505, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 20:15:04,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.986e+02 1.349e+03 1.724e+03 2.286e+03 5.573e+03, threshold=3.449e+03, percent-clipped=9.0
+2023-03-08 20:15:04,534 INFO [train.py:968] (0/2) Epoch 17, batch 15650, giga_loss[loss=0.2389, simple_loss=0.3314, pruned_loss=0.07325, over 28887.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3316, pruned_loss=0.08591, over 5667602.55 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3344, pruned_loss=0.08975, over 5763514.09 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3318, pruned_loss=0.08579, over 5653959.77 frames. ], batch size: 145, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:15:22,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2353, 3.0590, 1.3490, 1.4016], device='cuda:0'), covar=tensor([0.0979, 0.0323, 0.0930, 0.1412], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0531, 0.0365, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 20:15:39,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2456, 1.7383, 1.3029, 0.3927], device='cuda:0'), covar=tensor([0.3725, 0.2385, 0.3804, 0.5004], device='cuda:0'), in_proj_covar=tensor([0.1672, 0.1577, 0.1548, 0.1371], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 20:15:49,161 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5424, 3.6079, 1.7392, 1.6639], device='cuda:0'), covar=tensor([0.0946, 0.0306, 0.0886, 0.1302], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0531, 0.0366, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 20:15:56,247 INFO [train.py:968] (0/2) Epoch 17, batch 15700, libri_loss[loss=0.2667, simple_loss=0.3441, pruned_loss=0.09471, over 29671.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3325, pruned_loss=0.086, over 5676360.65 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3341, pruned_loss=0.08951, over 5765618.84 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3329, pruned_loss=0.08596, over 5659075.16 frames. ], batch size: 91, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:16:05,644 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=746587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:16:09,921 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=746590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:16:49,196 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=746619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:17:00,597 INFO [train.py:968] (0/2) Epoch 17, batch 15750, giga_loss[loss=0.2692, simple_loss=0.3429, pruned_loss=0.09773, over 28940.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3326, pruned_loss=0.08612, over 5682680.45 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3342, pruned_loss=0.08952, over 5762639.06 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3328, pruned_loss=0.08605, over 5670817.18 frames. ], batch size: 213, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:17:01,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.915e+02 1.355e+03 1.932e+03 2.830e+03 4.123e+03, threshold=3.864e+03, percent-clipped=9.0
+2023-03-08 20:17:57,445 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=746678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:17:57,876 INFO [train.py:968] (0/2) Epoch 17, batch 15800, giga_loss[loss=0.2258, simple_loss=0.3148, pruned_loss=0.06838, over 28961.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3287, pruned_loss=0.08333, over 5690231.66 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.334, pruned_loss=0.08955, over 5763584.98 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3288, pruned_loss=0.0831, over 5677960.23 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:18:53,257 INFO [train.py:968] (0/2) Epoch 17, batch 15850, giga_loss[loss=0.2343, simple_loss=0.3093, pruned_loss=0.07962, over 26948.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3286, pruned_loss=0.0839, over 5690442.57 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3335, pruned_loss=0.08931, over 5764533.71 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3289, pruned_loss=0.08357, over 5675229.30 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:18:54,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.285e+02 1.325e+03 1.739e+03 2.333e+03 8.494e+03, threshold=3.479e+03, percent-clipped=8.0
+2023-03-08 20:19:47,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9085, 3.7205, 3.5419, 1.7363], device='cuda:0'), covar=tensor([0.0730, 0.0869, 0.0861, 0.2220], device='cuda:0'), in_proj_covar=tensor([0.1144, 0.1051, 0.0904, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 20:19:53,762 INFO [train.py:968] (0/2) Epoch 17, batch 15900, giga_loss[loss=0.2445, simple_loss=0.3163, pruned_loss=0.08635, over 26938.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3265, pruned_loss=0.08351, over 5685643.20 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3331, pruned_loss=0.08914, over 5766846.96 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3269, pruned_loss=0.08331, over 5670710.77 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:20:26,632 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-08 20:20:57,648 INFO [train.py:968] (0/2) Epoch 17, batch 15950, giga_loss[loss=0.2327, simple_loss=0.3172, pruned_loss=0.07411, over 28910.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3294, pruned_loss=0.08534, over 5679653.30 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3329, pruned_loss=0.08898, over 5767837.36 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.33, pruned_loss=0.08527, over 5665771.30 frames. ], batch size: 136, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:21:00,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.677e+02 1.489e+03 2.255e+03 3.877e+03 1.556e+04, threshold=4.510e+03, percent-clipped=28.0
+2023-03-08 20:21:07,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5708, 1.6874, 1.8854, 1.4236], device='cuda:0'), covar=tensor([0.1808, 0.2491, 0.1425, 0.1822], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0684, 0.0911, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 20:21:37,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3517, 1.5761, 1.3452, 1.5158], device='cuda:0'), covar=tensor([0.0780, 0.0308, 0.0339, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0114, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 20:21:50,963 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=746870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:22:00,621 INFO [train.py:968] (0/2) Epoch 17, batch 16000, giga_loss[loss=0.251, simple_loss=0.3345, pruned_loss=0.08381, over 29075.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3319, pruned_loss=0.08674, over 5687216.69 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3328, pruned_loss=0.08895, over 5770260.74 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3323, pruned_loss=0.08663, over 5671466.50 frames. ], batch size: 200, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:22:50,925 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-08 20:23:05,950 INFO [train.py:968] (0/2) Epoch 17, batch 16050, giga_loss[loss=0.2759, simple_loss=0.3538, pruned_loss=0.09896, over 27584.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3324, pruned_loss=0.08763, over 5680052.61 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3321, pruned_loss=0.08852, over 5773117.80 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3333, pruned_loss=0.08785, over 5662356.77 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:23:07,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.038e+02 1.256e+03 1.476e+03 2.099e+03 6.446e+03, threshold=2.953e+03, percent-clipped=3.0
+2023-03-08 20:23:25,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 20:24:01,587 INFO [train.py:968] (0/2) Epoch 17, batch 16100, giga_loss[loss=0.2923, simple_loss=0.3815, pruned_loss=0.1016, over 28821.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3361, pruned_loss=0.0892, over 5688798.92 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3318, pruned_loss=0.08846, over 5774658.04 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3371, pruned_loss=0.08944, over 5671388.44 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:24:14,712 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=746989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:24:36,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5341, 1.1396, 4.6702, 3.4680], device='cuda:0'), covar=tensor([0.1656, 0.2924, 0.0384, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0712, 0.0616, 0.0899, 0.0825], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 20:25:01,891 INFO [train.py:968] (0/2) Epoch 17, batch 16150, giga_loss[loss=0.2527, simple_loss=0.347, pruned_loss=0.07915, over 28860.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3384, pruned_loss=0.08965, over 5691862.00 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3315, pruned_loss=0.08827, over 5776635.92 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3396, pruned_loss=0.09001, over 5675292.37 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:25:03,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.253e+02 1.329e+03 1.792e+03 2.611e+03 9.627e+03, threshold=3.583e+03, percent-clipped=14.0
+2023-03-08 20:25:08,789 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3831, 1.6659, 1.3585, 1.2557], device='cuda:0'), covar=tensor([0.2318, 0.2134, 0.2405, 0.2048], device='cuda:0'), in_proj_covar=tensor([0.1420, 0.1028, 0.1263, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 20:25:34,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:26:12,236 INFO [train.py:968] (0/2) Epoch 17, batch 16200, giga_loss[loss=0.2692, simple_loss=0.335, pruned_loss=0.1017, over 26998.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3382, pruned_loss=0.08971, over 5691794.82 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3312, pruned_loss=0.08815, over 5779230.94 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3395, pruned_loss=0.09015, over 5674795.17 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:26:47,188 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=747105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:27:18,123 INFO [train.py:968] (0/2) Epoch 17, batch 16250, libri_loss[loss=0.2949, simple_loss=0.3688, pruned_loss=0.1105, over 29772.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.336, pruned_loss=0.08866, over 5701059.11 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3313, pruned_loss=0.08819, over 5781727.22 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.337, pruned_loss=0.08898, over 5683631.53 frames. ], batch size: 87, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:27:20,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.161e+02 1.385e+03 1.675e+03 2.176e+03 4.922e+03, threshold=3.350e+03, percent-clipped=6.0
+2023-03-08 20:27:31,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5174, 1.9897, 1.8172, 1.5028], device='cuda:0'), covar=tensor([0.2861, 0.1837, 0.2116, 0.2214], device='cuda:0'), in_proj_covar=tensor([0.1854, 0.1781, 0.1689, 0.1845], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 20:28:12,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=747169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:28:24,034 INFO [train.py:968] (0/2) Epoch 17, batch 16300, giga_loss[loss=0.2811, simple_loss=0.3492, pruned_loss=0.1065, over 26846.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3361, pruned_loss=0.08896, over 5690942.80 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.331, pruned_loss=0.08797, over 5779054.45 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3373, pruned_loss=0.08942, over 5677868.70 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:28:45,855 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:28:47,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:29:25,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:29:26,343 INFO [train.py:968] (0/2) Epoch 17, batch 16350, giga_loss[loss=0.2343, simple_loss=0.3149, pruned_loss=0.07683, over 28985.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3355, pruned_loss=0.08951, over 5673386.84 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3314, pruned_loss=0.08825, over 5773410.83 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3363, pruned_loss=0.08967, over 5664743.00 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:29:29,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.299e+02 1.486e+03 1.923e+03 2.784e+03 6.479e+03, threshold=3.846e+03, percent-clipped=16.0
+2023-03-08 20:29:38,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2453, 0.8235, 0.9087, 1.5001], device='cuda:0'), covar=tensor([0.0755, 0.0363, 0.0374, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0114, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 20:29:43,975 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6181, 3.4666, 1.6704, 1.6782], device='cuda:0'), covar=tensor([0.0898, 0.0312, 0.0916, 0.1258], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0534, 0.0368, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 20:29:46,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:30:25,698 INFO [train.py:968] (0/2) Epoch 17, batch 16400, giga_loss[loss=0.2708, simple_loss=0.3501, pruned_loss=0.09579, over 28730.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3323, pruned_loss=0.08866, over 5677390.88 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3308, pruned_loss=0.08809, over 5773479.12 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3335, pruned_loss=0.08898, over 5667119.17 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:31:25,485 INFO [train.py:968] (0/2) Epoch 17, batch 16450, giga_loss[loss=0.2456, simple_loss=0.3258, pruned_loss=0.08266, over 28030.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3325, pruned_loss=0.08835, over 5676711.61 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3305, pruned_loss=0.08796, over 5766339.76 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3338, pruned_loss=0.08875, over 5671764.15 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:31:25,765 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=747329.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:31:28,717 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.797e+02 1.429e+03 1.982e+03 2.891e+03 9.492e+03, threshold=3.963e+03, percent-clipped=14.0
+2023-03-08 20:32:09,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:32:26,998 INFO [train.py:968] (0/2) Epoch 17, batch 16500, giga_loss[loss=0.2325, simple_loss=0.3185, pruned_loss=0.07324, over 28894.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3305, pruned_loss=0.08638, over 5668166.70 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.33, pruned_loss=0.0877, over 5766870.22 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3321, pruned_loss=0.08693, over 5662492.63 frames. ], batch size: 284, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:32:37,576 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:32:41,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:33:13,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6519, 1.9357, 1.5192, 2.0804], device='cuda:0'), covar=tensor([0.2632, 0.2579, 0.3105, 0.2322], device='cuda:0'), in_proj_covar=tensor([0.1421, 0.1029, 0.1265, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 20:33:15,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:33:25,806 INFO [train.py:968] (0/2) Epoch 17, batch 16550, giga_loss[loss=0.2974, simple_loss=0.3659, pruned_loss=0.1144, over 24525.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3322, pruned_loss=0.08575, over 5668361.56 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3297, pruned_loss=0.08762, over 5760222.95 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3337, pruned_loss=0.08622, over 5668279.10 frames. ], batch size: 705, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:33:27,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.968e+02 1.390e+03 1.707e+03 2.466e+03 6.391e+03, threshold=3.414e+03, percent-clipped=5.0
+2023-03-08 20:33:30,621 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-08 20:34:02,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8234, 1.9390, 1.4804, 1.5370], device='cuda:0'), covar=tensor([0.0967, 0.0673, 0.0953, 0.1186], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0434, 0.0505, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 20:34:20,942 INFO [train.py:968] (0/2) Epoch 17, batch 16600, giga_loss[loss=0.259, simple_loss=0.3437, pruned_loss=0.08721, over 28868.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3337, pruned_loss=0.0854, over 5662373.15 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3302, pruned_loss=0.08798, over 5750537.70 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3347, pruned_loss=0.08541, over 5667431.94 frames. ], batch size: 284, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:34:23,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:34:54,412 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:34:59,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:35:23,083 INFO [train.py:968] (0/2) Epoch 17, batch 16650, giga_loss[loss=0.268, simple_loss=0.3559, pruned_loss=0.09009, over 28954.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3347, pruned_loss=0.08538, over 5677741.50 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.33, pruned_loss=0.08789, over 5751954.23 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3356, pruned_loss=0.08545, over 5679779.56 frames. ], batch size: 227, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:35:27,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.465e+02 1.340e+03 1.759e+03 2.477e+03 5.041e+03, threshold=3.517e+03, percent-clipped=7.0
+2023-03-08 20:35:37,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:35:43,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:36:28,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5632, 3.1269, 1.6199, 1.7014], device='cuda:0'), covar=tensor([0.0769, 0.0334, 0.0805, 0.1142], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0531, 0.0367, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 20:36:31,629 INFO [train.py:968] (0/2) Epoch 17, batch 16700, giga_loss[loss=0.2558, simple_loss=0.3415, pruned_loss=0.08506, over 28852.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3359, pruned_loss=0.0869, over 5671138.95 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3298, pruned_loss=0.08778, over 5755269.29 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3369, pruned_loss=0.08701, over 5668543.07 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:36:33,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1396, 1.4715, 1.4548, 1.2120], device='cuda:0'), covar=tensor([0.1693, 0.1586, 0.2068, 0.1767], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0716, 0.0674, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 20:36:40,644 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=747587.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:37:16,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-08 20:37:28,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:37:33,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:37:36,538 INFO [train.py:968] (0/2) Epoch 17, batch 16750, giga_loss[loss=0.2409, simple_loss=0.3305, pruned_loss=0.07563, over 28442.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3356, pruned_loss=0.0866, over 5672387.51 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3298, pruned_loss=0.08777, over 5755509.09 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3365, pruned_loss=0.08668, over 5667938.56 frames. ], batch size: 369, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:37:40,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.209e+02 1.607e+03 2.184e+03 3.288e+03 8.373e+03, threshold=4.368e+03, percent-clipped=22.0
+2023-03-08 20:38:15,089 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:38:30,240 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-08 20:38:46,661 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=747678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:38:47,038 INFO [train.py:968] (0/2) Epoch 17, batch 16800, giga_loss[loss=0.2363, simple_loss=0.3328, pruned_loss=0.06992, over 28602.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3361, pruned_loss=0.08622, over 5675849.53 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3297, pruned_loss=0.08766, over 5758381.28 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3371, pruned_loss=0.08637, over 5668271.07 frames. ], batch size: 242, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:39:02,770 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:39:05,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:39:23,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747704.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:39:44,278 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:39:57,954 INFO [train.py:968] (0/2) Epoch 17, batch 16850, giga_loss[loss=0.2518, simple_loss=0.3322, pruned_loss=0.08566, over 27585.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3371, pruned_loss=0.0868, over 5681689.11 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3295, pruned_loss=0.08758, over 5760616.19 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3383, pruned_loss=0.08696, over 5671396.81 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:40:04,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.642e+02 1.410e+03 1.802e+03 2.271e+03 5.933e+03, threshold=3.604e+03, percent-clipped=3.0
+2023-03-08 20:41:11,723 INFO [train.py:968] (0/2) Epoch 17, batch 16900, giga_loss[loss=0.3114, simple_loss=0.3904, pruned_loss=0.1162, over 29080.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3406, pruned_loss=0.08816, over 5674306.99 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3293, pruned_loss=0.08751, over 5751803.14 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.08836, over 5673304.50 frames. ], batch size: 285, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:42:16,295 INFO [train.py:968] (0/2) Epoch 17, batch 16950, giga_loss[loss=0.2713, simple_loss=0.3502, pruned_loss=0.09622, over 28361.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3388, pruned_loss=0.08721, over 5680981.69 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3292, pruned_loss=0.08727, over 5755133.68 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3402, pruned_loss=0.08759, over 5674721.72 frames. ], batch size: 368, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:42:21,452 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.627e+02 1.310e+03 1.620e+03 2.365e+03 7.896e+03, threshold=3.241e+03, percent-clipped=14.0
+2023-03-08 20:42:41,360 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747847.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:42:48,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747850.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:43:29,495 INFO [train.py:968] (0/2) Epoch 17, batch 17000, giga_loss[loss=0.257, simple_loss=0.3389, pruned_loss=0.0876, over 28668.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3371, pruned_loss=0.08673, over 5691641.59 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3294, pruned_loss=0.08731, over 5757030.94 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3382, pruned_loss=0.08699, over 5684200.42 frames. ], batch size: 370, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:43:30,639 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747879.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:43:49,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5407, 1.8258, 1.7664, 1.3495], device='cuda:0'), covar=tensor([0.1871, 0.2465, 0.1567, 0.1761], device='cuda:0'), in_proj_covar=tensor([0.0865, 0.0683, 0.0910, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 20:44:36,701 INFO [train.py:968] (0/2) Epoch 17, batch 17050, giga_loss[loss=0.2262, simple_loss=0.3201, pruned_loss=0.06613, over 29052.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3343, pruned_loss=0.08464, over 5696931.58 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3291, pruned_loss=0.08709, over 5759578.53 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3356, pruned_loss=0.08499, over 5686297.55 frames. ], batch size: 175, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:44:43,930 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.035e+02 1.437e+03 1.954e+03 2.835e+03 1.207e+04, threshold=3.908e+03, percent-clipped=17.0
+2023-03-08 20:45:23,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747962.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:45:43,555 INFO [train.py:968] (0/2) Epoch 17, batch 17100, giga_loss[loss=0.2337, simple_loss=0.3241, pruned_loss=0.07169, over 28785.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3327, pruned_loss=0.08331, over 5699101.09 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3292, pruned_loss=0.08705, over 5757684.85 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3337, pruned_loss=0.08354, over 5691264.11 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:46:08,652 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-748000.pt
+2023-03-08 20:46:43,473 INFO [train.py:968] (0/2) Epoch 17, batch 17150, giga_loss[loss=0.2522, simple_loss=0.3403, pruned_loss=0.08208, over 28039.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3335, pruned_loss=0.08423, over 5692317.87 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3289, pruned_loss=0.08681, over 5762167.88 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3348, pruned_loss=0.08457, over 5680222.09 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:46:49,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.070e+02 1.301e+03 1.690e+03 2.262e+03 5.079e+03, threshold=3.380e+03, percent-clipped=2.0
+2023-03-08 20:47:11,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:47:20,406 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3987, 1.2935, 1.5069, 1.1527], device='cuda:0'), covar=tensor([0.1726, 0.2881, 0.1424, 0.1624], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0683, 0.0912, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 20:47:37,801 INFO [train.py:968] (0/2) Epoch 17, batch 17200, giga_loss[loss=0.2511, simple_loss=0.3409, pruned_loss=0.0807, over 28808.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3364, pruned_loss=0.08613, over 5685995.33 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3287, pruned_loss=0.0869, over 5754894.86 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3378, pruned_loss=0.08628, over 5681207.66 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:48:11,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=748105.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:48:13,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=748108.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:48:15,356 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-08 20:48:33,593 INFO [train.py:968] (0/2) Epoch 17, batch 17250, giga_loss[loss=0.2259, simple_loss=0.3172, pruned_loss=0.06734, over 28692.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3353, pruned_loss=0.0859, over 5684580.17 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3279, pruned_loss=0.08658, over 5757327.24 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3374, pruned_loss=0.08631, over 5676386.53 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:48:38,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.536e+02 1.392e+03 1.981e+03 3.009e+03 1.049e+04, threshold=3.962e+03, percent-clipped=20.0
+2023-03-08 20:48:43,999 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=748137.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:49:31,069 INFO [train.py:968] (0/2) Epoch 17, batch 17300, libri_loss[loss=0.2261, simple_loss=0.2961, pruned_loss=0.07807, over 29481.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3335, pruned_loss=0.08616, over 5685437.98 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3277, pruned_loss=0.08641, over 5759048.93 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3356, pruned_loss=0.08665, over 5675167.92 frames. ], batch size: 70, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:49:52,453 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=748196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:49:56,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=748199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:50:04,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=748208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:50:26,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2913, 1.8742, 1.3335, 0.4658], device='cuda:0'), covar=tensor([0.4324, 0.2604, 0.4047, 0.5441], device='cuda:0'), in_proj_covar=tensor([0.1681, 0.1593, 0.1561, 0.1385], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 20:50:27,679 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=748228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:50:28,053 INFO [train.py:968] (0/2) Epoch 17, batch 17350, giga_loss[loss=0.2511, simple_loss=0.3335, pruned_loss=0.08437, over 28833.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3333, pruned_loss=0.08661, over 5681698.42 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3275, pruned_loss=0.0863, over 5753129.80 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3353, pruned_loss=0.08711, over 5676360.89 frames. ], batch size: 199, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:50:32,483 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.891e+02 1.485e+03 2.017e+03 3.273e+03 7.546e+03, threshold=4.035e+03, percent-clipped=13.0
+2023-03-08 20:50:43,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3932, 1.1906, 4.1646, 3.3521], device='cuda:0'), covar=tensor([0.1636, 0.2909, 0.0384, 0.1267], device='cuda:0'), in_proj_covar=tensor([0.0718, 0.0619, 0.0905, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 20:51:10,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-08 20:51:19,941 INFO [train.py:968] (0/2) Epoch 17, batch 17400, giga_loss[loss=0.2951, simple_loss=0.3742, pruned_loss=0.108, over 28911.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3372, pruned_loss=0.08934, over 5692550.97 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3267, pruned_loss=0.08589, over 5757977.73 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3398, pruned_loss=0.09014, over 5681041.09 frames. ], batch size: 227, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:52:12,271 INFO [train.py:968] (0/2) Epoch 17, batch 17450, giga_loss[loss=0.3404, simple_loss=0.4034, pruned_loss=0.1387, over 27497.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3463, pruned_loss=0.09495, over 5693848.21 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3266, pruned_loss=0.08582, over 5761854.60 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3489, pruned_loss=0.09584, over 5679503.42 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:52:17,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.448e+02 1.400e+03 1.794e+03 2.333e+03 4.325e+03, threshold=3.589e+03, percent-clipped=1.0
+2023-03-08 20:52:29,743 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 20:52:55,803 INFO [train.py:968] (0/2) Epoch 17, batch 17500, giga_loss[loss=0.3102, simple_loss=0.3882, pruned_loss=0.1161, over 28381.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3523, pruned_loss=0.09824, over 5700393.41 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3266, pruned_loss=0.08579, over 5762260.54 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3549, pruned_loss=0.09925, over 5687046.28 frames. ], batch size: 77, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:52:59,578 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=748383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:53:42,680 INFO [train.py:968] (0/2) Epoch 17, batch 17550, giga_loss[loss=0.2356, simple_loss=0.3185, pruned_loss=0.07634, over 29123.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.349, pruned_loss=0.09709, over 5685480.18 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3272, pruned_loss=0.08607, over 5749467.33 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3511, pruned_loss=0.0979, over 5684675.35 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:53:47,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.465e+02 1.158e+03 1.407e+03 2.035e+03 1.058e+04, threshold=2.814e+03, percent-clipped=3.0
+2023-03-08 20:54:23,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=748474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:54:27,283 INFO [train.py:968] (0/2) Epoch 17, batch 17600, giga_loss[loss=0.2421, simple_loss=0.3201, pruned_loss=0.08209, over 28308.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3409, pruned_loss=0.0938, over 5686015.44 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3273, pruned_loss=0.08622, over 5751114.18 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3427, pruned_loss=0.09444, over 5683204.47 frames. ], batch size: 368, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:55:07,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.1602, 5.9222, 5.6737, 3.0796], device='cuda:0'), covar=tensor([0.0458, 0.0614, 0.0748, 0.1474], device='cuda:0'), in_proj_covar=tensor([0.1133, 0.1047, 0.0900, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 20:55:14,351 INFO [train.py:968] (0/2) Epoch 17, batch 17650, giga_loss[loss=0.2445, simple_loss=0.3181, pruned_loss=0.08544, over 29014.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3334, pruned_loss=0.09033, over 5685799.29 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3268, pruned_loss=0.08585, over 5754209.38 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3355, pruned_loss=0.09131, over 5679166.92 frames. ], batch size: 136, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:55:17,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.366e+02 1.142e+03 1.433e+03 1.917e+03 5.132e+03, threshold=2.865e+03, percent-clipped=8.0
+2023-03-08 20:55:52,231 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-08 20:55:56,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=748575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:56:00,564 INFO [train.py:968] (0/2) Epoch 17, batch 17700, giga_loss[loss=0.2086, simple_loss=0.2876, pruned_loss=0.06477, over 28839.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3265, pruned_loss=0.08749, over 5690999.47 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.327, pruned_loss=0.08579, over 5756927.58 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3281, pruned_loss=0.08838, over 5681918.10 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:56:05,332 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:56:16,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=748598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:56:41,132 INFO [train.py:968] (0/2) Epoch 17, batch 17750, giga_loss[loss=0.2009, simple_loss=0.2817, pruned_loss=0.0601, over 28783.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3202, pruned_loss=0.08484, over 5696256.41 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3269, pruned_loss=0.08566, over 5760241.58 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3214, pruned_loss=0.08568, over 5684106.69 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:56:47,556 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.770e+02 1.039e+03 1.350e+03 2.024e+03 5.435e+03, threshold=2.700e+03, percent-clipped=10.0
+2023-03-08 20:57:22,957 INFO [train.py:968] (0/2) Epoch 17, batch 17800, giga_loss[loss=0.228, simple_loss=0.3001, pruned_loss=0.07801, over 27607.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3144, pruned_loss=0.08164, over 5700055.37 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3267, pruned_loss=0.08536, over 5763934.31 frames. ], giga_tot_loss[loss=0.2402, simple_loss=0.3153, pruned_loss=0.08251, over 5685590.26 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:58:02,105 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=748726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:58:03,924 INFO [train.py:968] (0/2) Epoch 17, batch 17850, giga_loss[loss=0.2036, simple_loss=0.2816, pruned_loss=0.0628, over 28996.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3123, pruned_loss=0.08061, over 5707246.22 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3268, pruned_loss=0.08521, over 5766850.54 frames. ], giga_tot_loss[loss=0.2376, simple_loss=0.3125, pruned_loss=0.08132, over 5691168.51 frames. ], batch size: 106, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:58:04,201 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=748729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:58:04,965 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6244, 1.8151, 1.8876, 1.4016], device='cuda:0'), covar=tensor([0.1723, 0.2386, 0.1397, 0.1630], device='cuda:0'), in_proj_covar=tensor([0.0874, 0.0689, 0.0919, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 20:58:07,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.129e+02 1.005e+03 1.250e+03 1.830e+03 4.787e+03, threshold=2.501e+03, percent-clipped=11.0
+2023-03-08 20:58:30,395 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=748758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:58:30,409 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:58:49,126 INFO [train.py:968] (0/2) Epoch 17, batch 17900, giga_loss[loss=0.2487, simple_loss=0.32, pruned_loss=0.08871, over 28253.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3092, pruned_loss=0.07947, over 5701821.63 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3266, pruned_loss=0.08505, over 5769670.57 frames. ], giga_tot_loss[loss=0.2347, simple_loss=0.3092, pruned_loss=0.08009, over 5685219.65 frames. ], batch size: 368, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:59:29,120 INFO [train.py:968] (0/2) Epoch 17, batch 17950, giga_loss[loss=0.2322, simple_loss=0.3016, pruned_loss=0.08142, over 28821.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3068, pruned_loss=0.07836, over 5702424.44 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3267, pruned_loss=0.08503, over 5759070.78 frames. ], giga_tot_loss[loss=0.2319, simple_loss=0.3063, pruned_loss=0.07871, over 5697014.06 frames. ], batch size: 285, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:59:35,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.359e+02 1.102e+03 1.393e+03 2.284e+03 6.112e+03, threshold=2.786e+03, percent-clipped=21.0
+2023-03-08 20:59:47,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:59:54,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0804, 3.8963, 3.6414, 1.8070], device='cuda:0'), covar=tensor([0.0602, 0.0790, 0.0755, 0.2113], device='cuda:0'), in_proj_covar=tensor([0.1135, 0.1048, 0.0899, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 21:00:01,550 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 21:00:13,240 INFO [train.py:968] (0/2) Epoch 17, batch 18000, giga_loss[loss=0.2174, simple_loss=0.2922, pruned_loss=0.07136, over 28612.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3061, pruned_loss=0.07817, over 5688694.54 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3274, pruned_loss=0.08531, over 5758429.44 frames. ], giga_tot_loss[loss=0.23, simple_loss=0.3042, pruned_loss=0.07792, over 5683001.42 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:00:13,244 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 21:00:21,828 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2064, simple_loss=0.3117, pruned_loss=0.05051, over 944034.00 frames. 
+2023-03-08 21:00:21,829 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 21:00:29,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8093, 1.9647, 1.3353, 1.5593], device='cuda:0'), covar=tensor([0.0860, 0.0513, 0.1046, 0.1007], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0437, 0.0510, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 21:00:41,377 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=748901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:00:43,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=748904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:01:06,764 INFO [train.py:968] (0/2) Epoch 17, batch 18050, giga_loss[loss=0.1942, simple_loss=0.2656, pruned_loss=0.06144, over 28744.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.303, pruned_loss=0.07696, over 5691036.54 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3276, pruned_loss=0.08535, over 5760030.41 frames. ], giga_tot_loss[loss=0.2272, simple_loss=0.3011, pruned_loss=0.07666, over 5684500.01 frames. ], batch size: 92, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:01:07,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-08 21:01:10,680 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=748933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:01:12,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.264e+02 9.769e+02 1.299e+03 1.575e+03 3.769e+03, threshold=2.598e+03, percent-clipped=6.0
+2023-03-08 21:01:26,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:01:45,451 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:01:49,367 INFO [train.py:968] (0/2) Epoch 17, batch 18100, giga_loss[loss=0.1937, simple_loss=0.2673, pruned_loss=0.06009, over 28349.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3005, pruned_loss=0.07588, over 5697768.33 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3283, pruned_loss=0.08568, over 5762286.69 frames. ], giga_tot_loss[loss=0.2238, simple_loss=0.2975, pruned_loss=0.07505, over 5688599.38 frames. ], batch size: 71, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:02:02,911 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=748992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:02:04,814 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=748995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:02:14,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7699, 1.4503, 5.0366, 3.6247], device='cuda:0'), covar=tensor([0.1591, 0.2778, 0.0370, 0.0794], device='cuda:0'), in_proj_covar=tensor([0.0716, 0.0617, 0.0907, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 21:02:31,276 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=749024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:02:34,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3103, 1.6888, 1.1394, 1.2324], device='cuda:0'), covar=tensor([0.1114, 0.0638, 0.1559, 0.1227], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0436, 0.0507, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 21:02:35,927 INFO [train.py:968] (0/2) Epoch 17, batch 18150, giga_loss[loss=0.2261, simple_loss=0.2945, pruned_loss=0.07891, over 28590.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2986, pruned_loss=0.07533, over 5681188.12 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3288, pruned_loss=0.08594, over 5752031.23 frames. ], giga_tot_loss[loss=0.2218, simple_loss=0.2952, pruned_loss=0.07422, over 5680903.91 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:02:44,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.717e+02 1.006e+03 1.233e+03 1.717e+03 4.771e+03, threshold=2.465e+03, percent-clipped=5.0
+2023-03-08 21:03:01,766 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 21:03:03,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3859, 3.3882, 1.6464, 1.4923], device='cuda:0'), covar=tensor([0.0998, 0.0313, 0.0889, 0.1332], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0528, 0.0365, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 21:03:06,304 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-08 21:03:22,801 INFO [train.py:968] (0/2) Epoch 17, batch 18200, giga_loss[loss=0.2315, simple_loss=0.3015, pruned_loss=0.08075, over 29049.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2962, pruned_loss=0.07419, over 5679315.35 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3296, pruned_loss=0.08633, over 5749184.79 frames. ], giga_tot_loss[loss=0.2189, simple_loss=0.2923, pruned_loss=0.0728, over 5680737.84 frames. ], batch size: 128, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:03:36,200 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=749093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:03:39,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=749096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:04:02,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=749116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:04:05,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=749119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:04:12,092 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=749125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:04:15,342 INFO [train.py:968] (0/2) Epoch 17, batch 18250, giga_loss[loss=0.2907, simple_loss=0.3625, pruned_loss=0.1094, over 28621.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3041, pruned_loss=0.07898, over 5670251.59 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3301, pruned_loss=0.08644, over 5752392.39 frames. ], giga_tot_loss[loss=0.2276, simple_loss=0.3, pruned_loss=0.07758, over 5667039.13 frames. ], batch size: 92, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:04:21,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.748e+02 1.133e+03 1.558e+03 2.015e+03 6.654e+03, threshold=3.117e+03, percent-clipped=13.0
+2023-03-08 21:04:32,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=749148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:05:01,533 INFO [train.py:968] (0/2) Epoch 17, batch 18300, giga_loss[loss=0.3051, simple_loss=0.379, pruned_loss=0.1156, over 28822.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3175, pruned_loss=0.08573, over 5680050.07 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3307, pruned_loss=0.08679, over 5755906.96 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3132, pruned_loss=0.08418, over 5672084.39 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:05:10,415 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=749189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:05:12,891 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=749193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:05:41,830 INFO [train.py:968] (0/2) Epoch 17, batch 18350, giga_loss[loss=0.2955, simple_loss=0.3763, pruned_loss=0.1074, over 28851.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3294, pruned_loss=0.09157, over 5691994.30 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3307, pruned_loss=0.08666, over 5756867.91 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3258, pruned_loss=0.09049, over 5683599.83 frames. ], batch size: 285, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:05:47,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.818e+02 1.216e+03 1.601e+03 2.096e+03 4.585e+03, threshold=3.201e+03, percent-clipped=6.0
+2023-03-08 21:06:23,652 INFO [train.py:968] (0/2) Epoch 17, batch 18400, giga_loss[loss=0.352, simple_loss=0.4038, pruned_loss=0.1501, over 26436.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3363, pruned_loss=0.09441, over 5691074.44 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3311, pruned_loss=0.08686, over 5761629.72 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3331, pruned_loss=0.09353, over 5678384.01 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:07:07,072 INFO [train.py:968] (0/2) Epoch 17, batch 18450, giga_loss[loss=0.237, simple_loss=0.3318, pruned_loss=0.07105, over 28921.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3389, pruned_loss=0.09407, over 5692597.89 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3312, pruned_loss=0.08684, over 5762385.57 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3364, pruned_loss=0.09343, over 5681590.86 frames. ], batch size: 174, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:07:13,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.273e+02 1.123e+03 1.460e+03 1.988e+03 4.866e+03, threshold=2.921e+03, percent-clipped=5.0
+2023-03-08 21:07:50,932 INFO [train.py:968] (0/2) Epoch 17, batch 18500, giga_loss[loss=0.2998, simple_loss=0.354, pruned_loss=0.1228, over 23643.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.0947, over 5685137.31 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3315, pruned_loss=0.08721, over 5763750.26 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3391, pruned_loss=0.09412, over 5672492.63 frames. ], batch size: 705, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:08:12,562 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-08 21:08:35,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1661, 1.2386, 3.6042, 3.0778], device='cuda:0'), covar=tensor([0.1656, 0.2763, 0.0417, 0.1322], device='cuda:0'), in_proj_covar=tensor([0.0715, 0.0615, 0.0905, 0.0829], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 21:08:38,494 INFO [train.py:968] (0/2) Epoch 17, batch 18550, giga_loss[loss=0.2799, simple_loss=0.3491, pruned_loss=0.1053, over 28604.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3437, pruned_loss=0.0965, over 5682378.42 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3317, pruned_loss=0.08721, over 5766055.94 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3422, pruned_loss=0.09624, over 5668365.61 frames. ], batch size: 85, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:08:43,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.978e+02 1.170e+03 1.535e+03 1.939e+03 4.478e+03, threshold=3.070e+03, percent-clipped=12.0
+2023-03-08 21:09:02,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4696, 5.2619, 4.9941, 2.3474], device='cuda:0'), covar=tensor([0.0414, 0.0599, 0.0589, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.1125, 0.1046, 0.0894, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 21:09:22,504 INFO [train.py:968] (0/2) Epoch 17, batch 18600, giga_loss[loss=0.271, simple_loss=0.3426, pruned_loss=0.09972, over 28590.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3458, pruned_loss=0.09808, over 5680723.87 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3322, pruned_loss=0.08745, over 5761258.29 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3445, pruned_loss=0.09801, over 5670525.87 frames. ], batch size: 60, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:09:57,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5233, 1.6893, 1.6920, 1.5123], device='cuda:0'), covar=tensor([0.2101, 0.2030, 0.1318, 0.1638], device='cuda:0'), in_proj_covar=tensor([0.1864, 0.1786, 0.1702, 0.1866], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 21:10:08,497 INFO [train.py:968] (0/2) Epoch 17, batch 18650, giga_loss[loss=0.2735, simple_loss=0.3567, pruned_loss=0.09516, over 28954.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3489, pruned_loss=0.09988, over 5674096.98 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3324, pruned_loss=0.08745, over 5753834.05 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3482, pruned_loss=0.1001, over 5671408.88 frames. ], batch size: 145, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:10:15,053 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=749535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:10:15,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.663e+02 1.298e+03 1.779e+03 2.458e+03 5.121e+03, threshold=3.557e+03, percent-clipped=13.0
+2023-03-08 21:10:27,251 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=749550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:10:27,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4756, 1.7410, 1.3696, 1.4782], device='cuda:0'), covar=tensor([0.2626, 0.2662, 0.2951, 0.2317], device='cuda:0'), in_proj_covar=tensor([0.1419, 0.1033, 0.1262, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 21:10:28,106 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3182, 1.5974, 1.3236, 1.1446], device='cuda:0'), covar=tensor([0.2566, 0.2588, 0.2890, 0.2100], device='cuda:0'), in_proj_covar=tensor([0.1419, 0.1033, 0.1262, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 21:10:39,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=749564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:10:42,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=749568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:10:52,183 INFO [train.py:968] (0/2) Epoch 17, batch 18700, giga_loss[loss=0.2758, simple_loss=0.3542, pruned_loss=0.09872, over 28817.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3527, pruned_loss=0.1015, over 5675220.01 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3331, pruned_loss=0.08779, over 5754795.52 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3518, pruned_loss=0.1016, over 5670884.07 frames. ], batch size: 99, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:11:06,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1702, 3.9901, 3.7924, 1.8202], device='cuda:0'), covar=tensor([0.0561, 0.0742, 0.0679, 0.2384], device='cuda:0'), in_proj_covar=tensor([0.1124, 0.1042, 0.0892, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 21:11:24,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6474, 4.4503, 4.2168, 1.9894], device='cuda:0'), covar=tensor([0.0510, 0.0710, 0.0691, 0.2115], device='cuda:0'), in_proj_covar=tensor([0.1121, 0.1042, 0.0892, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 21:11:32,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4881, 3.4106, 1.5902, 1.5244], device='cuda:0'), covar=tensor([0.0986, 0.0254, 0.0912, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0526, 0.0364, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 21:11:35,759 INFO [train.py:968] (0/2) Epoch 17, batch 18750, giga_loss[loss=0.3453, simple_loss=0.3951, pruned_loss=0.1477, over 26528.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3553, pruned_loss=0.1023, over 5676868.51 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3332, pruned_loss=0.0878, over 5755048.34 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3546, pruned_loss=0.1025, over 5672506.33 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:11:42,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.562e+02 1.129e+03 1.395e+03 1.954e+03 6.667e+03, threshold=2.791e+03, percent-clipped=4.0
+2023-03-08 21:12:12,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0805, 1.3348, 3.5494, 3.1253], device='cuda:0'), covar=tensor([0.2161, 0.3004, 0.0771, 0.1195], device='cuda:0'), in_proj_covar=tensor([0.0712, 0.0613, 0.0900, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 21:12:16,011 INFO [train.py:968] (0/2) Epoch 17, batch 18800, libri_loss[loss=0.2652, simple_loss=0.3434, pruned_loss=0.09347, over 29556.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3567, pruned_loss=0.1025, over 5693009.61 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3335, pruned_loss=0.08796, over 5760202.62 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3567, pruned_loss=0.103, over 5682380.42 frames. ], batch size: 78, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:12:40,955 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=749707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:12:43,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=749710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:12:43,521 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=749711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:12:43,595 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=749711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:12:45,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=749714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:12:57,643 INFO [train.py:968] (0/2) Epoch 17, batch 18850, giga_loss[loss=0.2733, simple_loss=0.3608, pruned_loss=0.09293, over 28869.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3563, pruned_loss=0.1013, over 5680930.40 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3334, pruned_loss=0.08785, over 5741921.47 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.357, pruned_loss=0.1022, over 5686125.06 frames. ], batch size: 199, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:13:04,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.418e+02 1.252e+03 1.677e+03 2.150e+03 4.303e+03, threshold=3.354e+03, percent-clipped=11.0
+2023-03-08 21:13:05,675 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=749739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:13:09,095 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=749743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:13:40,808 INFO [train.py:968] (0/2) Epoch 17, batch 18900, giga_loss[loss=0.2471, simple_loss=0.3374, pruned_loss=0.07842, over 28947.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.354, pruned_loss=0.09836, over 5693036.51 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3334, pruned_loss=0.08773, over 5744387.89 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3549, pruned_loss=0.09932, over 5694345.67 frames. ], batch size: 106, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:13:52,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6705, 2.3205, 1.6994, 0.8154], device='cuda:0'), covar=tensor([0.4597, 0.2712, 0.3743, 0.5363], device='cuda:0'), in_proj_covar=tensor([0.1667, 0.1583, 0.1558, 0.1364], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 21:14:20,973 INFO [train.py:968] (0/2) Epoch 17, batch 18950, libri_loss[loss=0.257, simple_loss=0.3463, pruned_loss=0.08383, over 27707.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3518, pruned_loss=0.09671, over 5696317.32 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3336, pruned_loss=0.08773, over 5745456.35 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3528, pruned_loss=0.09775, over 5695327.90 frames. ], batch size: 116, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:14:26,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2359, 1.3971, 1.5376, 1.2225], device='cuda:0'), covar=tensor([0.1795, 0.1823, 0.2193, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0731, 0.0687, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 21:14:27,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.048e+02 1.058e+03 1.248e+03 1.628e+03 4.347e+03, threshold=2.496e+03, percent-clipped=1.0
+2023-03-08 21:14:41,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2842, 1.1770, 1.1310, 1.4768], device='cuda:0'), covar=tensor([0.0777, 0.0362, 0.0345, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 21:14:41,910 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 21:15:05,035 INFO [train.py:968] (0/2) Epoch 17, batch 19000, libri_loss[loss=0.2865, simple_loss=0.359, pruned_loss=0.107, over 19327.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3544, pruned_loss=0.09989, over 5694175.39 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3339, pruned_loss=0.08787, over 5736991.08 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3551, pruned_loss=0.1007, over 5701650.85 frames. ], batch size: 187, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:15:33,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=749910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:15:34,248 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-08 21:15:45,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=749925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:15:50,230 INFO [train.py:968] (0/2) Epoch 17, batch 19050, giga_loss[loss=0.3471, simple_loss=0.4021, pruned_loss=0.146, over 28864.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3572, pruned_loss=0.1045, over 5702044.70 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3343, pruned_loss=0.08797, over 5742456.87 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.358, pruned_loss=0.1054, over 5702103.31 frames. ], batch size: 227, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:15:58,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.949e+02 1.452e+03 1.829e+03 2.582e+03 1.940e+04, threshold=3.659e+03, percent-clipped=25.0
+2023-03-08 21:16:01,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9219, 3.6614, 3.5715, 1.6186], device='cuda:0'), covar=tensor([0.0673, 0.0939, 0.0810, 0.2359], device='cuda:0'), in_proj_covar=tensor([0.1124, 0.1046, 0.0895, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 21:16:29,243 INFO [train.py:968] (0/2) Epoch 17, batch 19100, libri_loss[loss=0.3059, simple_loss=0.3741, pruned_loss=0.1189, over 29498.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3562, pruned_loss=0.105, over 5709326.08 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3346, pruned_loss=0.08795, over 5746906.92 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3572, pruned_loss=0.1062, over 5704281.53 frames. ], batch size: 85, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:16:46,154 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-750000.pt
+2023-03-08 21:16:48,516 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750002.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:17:10,340 INFO [train.py:968] (0/2) Epoch 17, batch 19150, giga_loss[loss=0.2821, simple_loss=0.3521, pruned_loss=0.106, over 28767.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3543, pruned_loss=0.1046, over 5693153.43 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3353, pruned_loss=0.08845, over 5730530.55 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3553, pruned_loss=0.1057, over 5701060.48 frames. ], batch size: 86, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:17:19,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.018e+02 1.200e+03 1.570e+03 2.159e+03 6.192e+03, threshold=3.140e+03, percent-clipped=7.0
+2023-03-08 21:17:34,048 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:17:37,781 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:17:45,670 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:17:48,959 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:17:55,549 INFO [train.py:968] (0/2) Epoch 17, batch 19200, giga_loss[loss=0.2837, simple_loss=0.3662, pruned_loss=0.1006, over 29079.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3537, pruned_loss=0.1045, over 5690795.04 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3362, pruned_loss=0.08888, over 5729588.37 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.354, pruned_loss=0.1054, over 5696852.40 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:18:01,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:18:01,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=750086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:18:07,803 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:18:12,634 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:18:16,257 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:18:41,030 INFO [train.py:968] (0/2) Epoch 17, batch 19250, giga_loss[loss=0.2712, simple_loss=0.3474, pruned_loss=0.09756, over 28938.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.353, pruned_loss=0.1033, over 5701679.42 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3368, pruned_loss=0.08912, over 5731807.40 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.353, pruned_loss=0.1041, over 5703935.50 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:18:49,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.871e+02 1.153e+03 1.392e+03 1.946e+03 4.250e+03, threshold=2.784e+03, percent-clipped=5.0
+2023-03-08 21:19:15,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4037, 3.3036, 1.5019, 1.5761], device='cuda:0'), covar=tensor([0.0990, 0.0277, 0.0910, 0.1337], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0531, 0.0366, 0.0410], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 21:19:26,183 INFO [train.py:968] (0/2) Epoch 17, batch 19300, libri_loss[loss=0.2738, simple_loss=0.3568, pruned_loss=0.09536, over 29190.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3517, pruned_loss=0.1022, over 5701132.89 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3372, pruned_loss=0.08928, over 5734365.66 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3516, pruned_loss=0.1029, over 5699967.59 frames. ], batch size: 101, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:19:32,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8060, 1.9782, 1.4496, 1.6451], device='cuda:0'), covar=tensor([0.1000, 0.0742, 0.1069, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0439, 0.0511, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 21:20:03,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2697, 1.2748, 3.8172, 3.1558], device='cuda:0'), covar=tensor([0.1595, 0.2673, 0.0418, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.0716, 0.0615, 0.0904, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 21:20:09,606 INFO [train.py:968] (0/2) Epoch 17, batch 19350, giga_loss[loss=0.2423, simple_loss=0.3218, pruned_loss=0.08139, over 28694.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3484, pruned_loss=0.09985, over 5696687.59 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3376, pruned_loss=0.08929, over 5738992.70 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3483, pruned_loss=0.1008, over 5690058.41 frames. ], batch size: 284, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:20:09,982 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:20:12,700 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:20:18,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.260e+02 1.097e+03 1.383e+03 1.878e+03 7.874e+03, threshold=2.767e+03, percent-clipped=6.0
+2023-03-08 21:20:40,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:20:58,197 INFO [train.py:968] (0/2) Epoch 17, batch 19400, giga_loss[loss=0.2254, simple_loss=0.3082, pruned_loss=0.0713, over 28853.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3419, pruned_loss=0.09686, over 5685666.68 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3375, pruned_loss=0.08925, over 5741494.80 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3421, pruned_loss=0.09779, over 5677311.56 frames. ], batch size: 174, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:21:44,068 INFO [train.py:968] (0/2) Epoch 17, batch 19450, giga_loss[loss=0.2278, simple_loss=0.3018, pruned_loss=0.07689, over 28008.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3363, pruned_loss=0.09374, over 5691123.41 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3377, pruned_loss=0.0892, over 5743309.14 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3363, pruned_loss=0.09462, over 5682011.82 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:21:52,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.889e+02 1.002e+03 1.268e+03 1.719e+03 5.114e+03, threshold=2.536e+03, percent-clipped=6.0
+2023-03-08 21:22:10,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7053, 3.7267, 1.8194, 1.8026], device='cuda:0'), covar=tensor([0.0927, 0.0244, 0.0827, 0.1277], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0525, 0.0364, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 21:22:26,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:22:28,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=750377.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:22:30,864 INFO [train.py:968] (0/2) Epoch 17, batch 19500, giga_loss[loss=0.2585, simple_loss=0.3398, pruned_loss=0.08859, over 28750.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3335, pruned_loss=0.09232, over 5664927.07 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3381, pruned_loss=0.08933, over 5745116.06 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3331, pruned_loss=0.09303, over 5654074.75 frames. ], batch size: 284, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:23:12,354 INFO [train.py:968] (0/2) Epoch 17, batch 19550, giga_loss[loss=0.2534, simple_loss=0.3319, pruned_loss=0.0875, over 27828.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.334, pruned_loss=0.0921, over 5663588.68 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3383, pruned_loss=0.08933, over 5744189.56 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3333, pruned_loss=0.09271, over 5654392.51 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:23:23,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.641e+02 1.124e+03 1.458e+03 2.144e+03 8.019e+03, threshold=2.917e+03, percent-clipped=18.0
+2023-03-08 21:23:30,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0125, 3.8198, 3.6908, 1.6306], device='cuda:0'), covar=tensor([0.0772, 0.0919, 0.0967, 0.2132], device='cuda:0'), in_proj_covar=tensor([0.1136, 0.1053, 0.0901, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 21:23:31,400 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-08 21:23:47,121 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=750466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:23:50,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=750470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:23:54,425 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7073, 1.8003, 1.8017, 1.6038], device='cuda:0'), covar=tensor([0.1864, 0.2235, 0.2325, 0.2281], device='cuda:0'), in_proj_covar=tensor([0.0454, 0.0736, 0.0690, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 21:23:58,305 INFO [train.py:968] (0/2) Epoch 17, batch 19600, giga_loss[loss=0.244, simple_loss=0.3205, pruned_loss=0.0838, over 28841.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3338, pruned_loss=0.09207, over 5673595.41 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3386, pruned_loss=0.08938, over 5747576.02 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3329, pruned_loss=0.09255, over 5661909.60 frames. ], batch size: 199, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:24:35,563 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750520.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:24:37,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:24:41,848 INFO [train.py:968] (0/2) Epoch 17, batch 19650, giga_loss[loss=0.2835, simple_loss=0.3539, pruned_loss=0.1065, over 28741.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3326, pruned_loss=0.09173, over 5681322.03 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3388, pruned_loss=0.08929, over 5749872.81 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3317, pruned_loss=0.09222, over 5669200.79 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:24:49,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5553, 1.8286, 1.4373, 1.6557], device='cuda:0'), covar=tensor([0.2527, 0.2651, 0.2969, 0.2395], device='cuda:0'), in_proj_covar=tensor([0.1418, 0.1031, 0.1262, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 21:24:50,071 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.149e+02 1.100e+03 1.333e+03 1.874e+03 4.356e+03, threshold=2.666e+03, percent-clipped=5.0
+2023-03-08 21:24:59,859 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750552.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:25:19,652 INFO [train.py:968] (0/2) Epoch 17, batch 19700, giga_loss[loss=0.2537, simple_loss=0.3306, pruned_loss=0.08835, over 28916.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3307, pruned_loss=0.09053, over 5679240.89 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3395, pruned_loss=0.08943, over 5739384.34 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3292, pruned_loss=0.09085, over 5676726.15 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:25:43,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6824, 1.9619, 1.7783, 1.3991], device='cuda:0'), covar=tensor([0.3762, 0.2546, 0.2364, 0.3139], device='cuda:0'), in_proj_covar=tensor([0.1865, 0.1776, 0.1703, 0.1864], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 21:25:44,545 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:25:47,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:25:48,604 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:25:50,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:26:01,326 INFO [train.py:968] (0/2) Epoch 17, batch 19750, giga_loss[loss=0.2286, simple_loss=0.3081, pruned_loss=0.07457, over 28786.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3285, pruned_loss=0.08923, over 5692339.65 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.34, pruned_loss=0.08954, over 5743028.77 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3266, pruned_loss=0.08941, over 5685742.05 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:26:09,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.779e+02 9.852e+02 1.225e+03 1.794e+03 6.875e+03, threshold=2.449e+03, percent-clipped=4.0
+2023-03-08 21:26:10,165 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:26:12,558 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750644.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:26:13,333 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:26:44,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:26:44,691 INFO [train.py:968] (0/2) Epoch 17, batch 19800, giga_loss[loss=0.2284, simple_loss=0.3056, pruned_loss=0.07557, over 29104.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3262, pruned_loss=0.08807, over 5695569.06 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3401, pruned_loss=0.08946, over 5746129.14 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3244, pruned_loss=0.08824, over 5686753.91 frames. ], batch size: 128, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:26:59,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0982, 1.1543, 3.7682, 3.2255], device='cuda:0'), covar=tensor([0.1825, 0.2847, 0.0436, 0.0783], device='cuda:0'), in_proj_covar=tensor([0.0721, 0.0618, 0.0908, 0.0836], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 21:27:24,065 INFO [train.py:968] (0/2) Epoch 17, batch 19850, giga_loss[loss=0.2393, simple_loss=0.3133, pruned_loss=0.08265, over 28935.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3233, pruned_loss=0.08685, over 5698186.55 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3405, pruned_loss=0.08957, over 5737395.48 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3212, pruned_loss=0.08684, over 5698893.95 frames. ], batch size: 227, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:27:33,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.273e+02 9.923e+02 1.220e+03 1.429e+03 6.502e+03, threshold=2.440e+03, percent-clipped=8.0
+2023-03-08 21:27:39,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=750749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:27:49,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4957, 3.6363, 1.6860, 1.6883], device='cuda:0'), covar=tensor([0.0986, 0.0298, 0.0854, 0.1275], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0528, 0.0364, 0.0409], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 21:28:03,948 INFO [train.py:968] (0/2) Epoch 17, batch 19900, giga_loss[loss=0.2394, simple_loss=0.3092, pruned_loss=0.0848, over 28708.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3223, pruned_loss=0.08604, over 5711862.67 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3413, pruned_loss=0.08971, over 5742985.46 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3193, pruned_loss=0.08579, over 5706401.06 frames. ], batch size: 119, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:28:42,367 INFO [train.py:968] (0/2) Epoch 17, batch 19950, giga_loss[loss=0.2387, simple_loss=0.3176, pruned_loss=0.07988, over 28772.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3212, pruned_loss=0.0858, over 5710075.53 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3417, pruned_loss=0.08972, over 5737943.27 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3179, pruned_loss=0.08548, over 5709827.40 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:28:52,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.155e+02 1.146e+03 1.504e+03 2.126e+03 6.992e+03, threshold=3.008e+03, percent-clipped=17.0
+2023-03-08 21:28:58,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3184, 3.1367, 2.9883, 1.5652], device='cuda:0'), covar=tensor([0.0932, 0.1069, 0.0866, 0.2198], device='cuda:0'), in_proj_covar=tensor([0.1144, 0.1058, 0.0909, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 21:29:22,736 INFO [train.py:968] (0/2) Epoch 17, batch 20000, giga_loss[loss=0.2233, simple_loss=0.2978, pruned_loss=0.07436, over 28739.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3202, pruned_loss=0.08521, over 5705280.64 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3429, pruned_loss=0.09029, over 5732164.74 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3159, pruned_loss=0.08432, over 5709154.72 frames. ], batch size: 92, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:29:34,708 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:29:36,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:29:58,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:30:01,951 INFO [train.py:968] (0/2) Epoch 17, batch 20050, giga_loss[loss=0.2423, simple_loss=0.3219, pruned_loss=0.08138, over 28425.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3189, pruned_loss=0.08421, over 5711901.78 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3431, pruned_loss=0.09034, over 5733787.98 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3151, pruned_loss=0.08342, over 5713373.88 frames. ], batch size: 60, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:30:09,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.807e+02 9.493e+02 1.226e+03 1.506e+03 4.121e+03, threshold=2.452e+03, percent-clipped=3.0
+2023-03-08 21:30:44,493 INFO [train.py:968] (0/2) Epoch 17, batch 20100, libri_loss[loss=0.2552, simple_loss=0.334, pruned_loss=0.0882, over 29497.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3215, pruned_loss=0.08601, over 5712565.54 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3432, pruned_loss=0.09027, over 5736333.91 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3182, pruned_loss=0.08536, over 5711179.14 frames. ], batch size: 70, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:31:20,933 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751019.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:31:32,042 INFO [train.py:968] (0/2) Epoch 17, batch 20150, giga_loss[loss=0.282, simple_loss=0.3499, pruned_loss=0.107, over 28588.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3277, pruned_loss=0.08992, over 5703342.93 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3432, pruned_loss=0.09023, over 5735653.01 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3246, pruned_loss=0.0894, over 5702441.17 frames. ], batch size: 78, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:31:37,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6750, 1.5347, 1.7905, 1.3496], device='cuda:0'), covar=tensor([0.2074, 0.2871, 0.1578, 0.1700], device='cuda:0'), in_proj_covar=tensor([0.0877, 0.0693, 0.0922, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 21:31:44,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.849e+02 1.217e+03 1.571e+03 2.172e+03 8.193e+03, threshold=3.143e+03, percent-clipped=18.0
+2023-03-08 21:31:45,176 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 21:31:57,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:32:04,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5729, 1.5524, 1.2593, 1.2053], device='cuda:0'), covar=tensor([0.0835, 0.0530, 0.1001, 0.1024], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0441, 0.0511, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 21:32:27,507 INFO [train.py:968] (0/2) Epoch 17, batch 20200, giga_loss[loss=0.3217, simple_loss=0.3804, pruned_loss=0.1315, over 28069.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3358, pruned_loss=0.09559, over 5697718.35 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3431, pruned_loss=0.0902, over 5737232.03 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3334, pruned_loss=0.09524, over 5695271.85 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:32:34,805 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:33:11,410 INFO [train.py:968] (0/2) Epoch 17, batch 20250, giga_loss[loss=0.2879, simple_loss=0.3605, pruned_loss=0.1076, over 28796.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3419, pruned_loss=0.09909, over 5694388.23 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3428, pruned_loss=0.08984, over 5741517.20 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3402, pruned_loss=0.0993, over 5687787.02 frames. ], batch size: 119, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:33:23,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.414e+02 1.392e+03 1.660e+03 2.077e+03 4.266e+03, threshold=3.319e+03, percent-clipped=2.0
+2023-03-08 21:33:48,705 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751162.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:33:50,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751165.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:34:04,031 INFO [train.py:968] (0/2) Epoch 17, batch 20300, giga_loss[loss=0.2644, simple_loss=0.348, pruned_loss=0.09044, over 28646.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3492, pruned_loss=0.1024, over 5694827.68 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3431, pruned_loss=0.08996, over 5742868.44 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3476, pruned_loss=0.1026, over 5687930.96 frames. ], batch size: 242, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:34:12,473 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:34:17,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751194.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:34:18,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:34:21,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:34:36,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4903, 1.6393, 1.7325, 1.3179], device='cuda:0'), covar=tensor([0.1551, 0.2116, 0.1220, 0.1502], device='cuda:0'), in_proj_covar=tensor([0.0873, 0.0692, 0.0918, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 21:34:41,193 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 21:34:48,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:34:48,477 INFO [train.py:968] (0/2) Epoch 17, batch 20350, giga_loss[loss=0.2921, simple_loss=0.3671, pruned_loss=0.1085, over 29039.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3532, pruned_loss=0.1041, over 5694488.38 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3432, pruned_loss=0.08999, over 5746398.57 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3521, pruned_loss=0.1045, over 5684736.49 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:34:53,584 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:35:00,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.809e+02 1.204e+03 1.501e+03 1.839e+03 4.358e+03, threshold=3.001e+03, percent-clipped=2.0
+2023-03-08 21:35:06,187 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751247.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:35:09,513 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 21:35:35,877 INFO [train.py:968] (0/2) Epoch 17, batch 20400, giga_loss[loss=0.2807, simple_loss=0.3594, pruned_loss=0.1011, over 28931.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3565, pruned_loss=0.1055, over 5696613.82 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3428, pruned_loss=0.08974, over 5746996.80 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3561, pruned_loss=0.1064, over 5687159.28 frames. ], batch size: 136, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:36:17,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2014, 5.0334, 4.7218, 2.2591], device='cuda:0'), covar=tensor([0.0409, 0.0512, 0.0572, 0.2016], device='cuda:0'), in_proj_covar=tensor([0.1139, 0.1054, 0.0902, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 21:36:17,399 INFO [train.py:968] (0/2) Epoch 17, batch 20450, giga_loss[loss=0.2301, simple_loss=0.3102, pruned_loss=0.07502, over 28802.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3524, pruned_loss=0.1024, over 5690997.56 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3432, pruned_loss=0.08998, over 5745054.11 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.352, pruned_loss=0.1033, over 5683670.50 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:36:29,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.205e+02 1.225e+03 1.552e+03 2.039e+03 5.448e+03, threshold=3.104e+03, percent-clipped=11.0
+2023-03-08 21:36:41,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4833, 2.3556, 2.2728, 1.1086], device='cuda:0'), covar=tensor([0.0872, 0.0947, 0.0822, 0.1810], device='cuda:0'), in_proj_covar=tensor([0.1139, 0.1055, 0.0902, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 21:36:57,136 INFO [train.py:968] (0/2) Epoch 17, batch 20500, giga_loss[loss=0.2555, simple_loss=0.3324, pruned_loss=0.08932, over 28616.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3497, pruned_loss=0.09992, over 5696408.57 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3432, pruned_loss=0.09014, over 5739695.63 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3496, pruned_loss=0.1008, over 5693904.66 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:37:10,741 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:37:16,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6923, 4.5043, 4.2974, 1.8847], device='cuda:0'), covar=tensor([0.0571, 0.0772, 0.0881, 0.2141], device='cuda:0'), in_proj_covar=tensor([0.1140, 0.1056, 0.0904, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 21:37:41,240 INFO [train.py:968] (0/2) Epoch 17, batch 20550, giga_loss[loss=0.2806, simple_loss=0.3605, pruned_loss=0.1004, over 28531.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3493, pruned_loss=0.09956, over 5688527.37 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3436, pruned_loss=0.09043, over 5743688.83 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3491, pruned_loss=0.1002, over 5681926.28 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:37:45,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4066, 1.4484, 1.3507, 1.5138], device='cuda:0'), covar=tensor([0.0821, 0.0333, 0.0333, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 21:37:53,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.692e+02 1.147e+03 1.473e+03 2.233e+03 6.650e+03, threshold=2.946e+03, percent-clipped=10.0
+2023-03-08 21:38:11,598 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:38:26,684 INFO [train.py:968] (0/2) Epoch 17, batch 20600, giga_loss[loss=0.302, simple_loss=0.3721, pruned_loss=0.116, over 28819.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3495, pruned_loss=0.09908, over 5695886.21 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3435, pruned_loss=0.09035, over 5745526.46 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3494, pruned_loss=0.09973, over 5688599.46 frames. ], batch size: 284, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:38:54,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9808, 2.2884, 1.9905, 1.8179], device='cuda:0'), covar=tensor([0.3068, 0.2221, 0.2407, 0.2591], device='cuda:0'), in_proj_covar=tensor([0.1869, 0.1786, 0.1724, 0.1872], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 21:39:11,482 INFO [train.py:968] (0/2) Epoch 17, batch 20650, libri_loss[loss=0.2591, simple_loss=0.3462, pruned_loss=0.08602, over 29525.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3527, pruned_loss=0.1014, over 5694584.98 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3435, pruned_loss=0.09036, over 5747091.34 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3526, pruned_loss=0.102, over 5686911.79 frames. ], batch size: 84, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:39:24,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.759e+02 1.199e+03 1.469e+03 2.031e+03 5.640e+03, threshold=2.937e+03, percent-clipped=7.0
+2023-03-08 21:39:40,976 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-08 21:39:44,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:39:56,393 INFO [train.py:968] (0/2) Epoch 17, batch 20700, giga_loss[loss=0.2919, simple_loss=0.3652, pruned_loss=0.1093, over 28681.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3548, pruned_loss=0.1033, over 5696916.35 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3438, pruned_loss=0.09047, over 5747121.49 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3546, pruned_loss=0.1038, over 5690216.18 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:40:23,796 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:40:25,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:40:28,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:40:37,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751622.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:40:45,113 INFO [train.py:968] (0/2) Epoch 17, batch 20750, giga_loss[loss=0.2992, simple_loss=0.3695, pruned_loss=0.1145, over 28923.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3553, pruned_loss=0.1035, over 5700436.15 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3442, pruned_loss=0.09059, over 5739711.54 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.355, pruned_loss=0.104, over 5701371.43 frames. ], batch size: 213, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:40:52,695 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:40:55,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.555e+02 1.270e+03 1.588e+03 2.144e+03 4.684e+03, threshold=3.177e+03, percent-clipped=10.0
+2023-03-08 21:41:27,571 INFO [train.py:968] (0/2) Epoch 17, batch 20800, giga_loss[loss=0.2736, simple_loss=0.3499, pruned_loss=0.09862, over 28882.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3563, pruned_loss=0.1046, over 5701348.81 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3443, pruned_loss=0.0905, over 5742310.73 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3561, pruned_loss=0.1054, over 5699284.02 frames. ], batch size: 112, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:41:47,129 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5539, 1.9799, 1.7951, 1.6924], device='cuda:0'), covar=tensor([0.0777, 0.0276, 0.0289, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 21:41:51,870 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:41:53,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:01,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:08,143 INFO [train.py:968] (0/2) Epoch 17, batch 20850, giga_loss[loss=0.2785, simple_loss=0.3578, pruned_loss=0.09958, over 28892.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3566, pruned_loss=0.1047, over 5702007.71 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3447, pruned_loss=0.09069, over 5737711.34 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3566, pruned_loss=0.1056, over 5703516.64 frames. ], batch size: 199, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:42:16,091 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:16,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9868, 1.0198, 3.4248, 3.0466], device='cuda:0'), covar=tensor([0.1745, 0.2798, 0.0493, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0721, 0.0617, 0.0909, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 21:42:19,208 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.191e+02 1.243e+03 1.649e+03 2.241e+03 4.833e+03, threshold=3.298e+03, percent-clipped=9.0
+2023-03-08 21:42:25,448 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:27,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:29,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:31,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4701, 1.7013, 1.6337, 1.4109], device='cuda:0'), covar=tensor([0.2351, 0.1993, 0.1382, 0.1855], device='cuda:0'), in_proj_covar=tensor([0.1871, 0.1790, 0.1725, 0.1873], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 21:42:35,616 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751765.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:42:37,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:37,493 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751768.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:42:46,758 INFO [train.py:968] (0/2) Epoch 17, batch 20900, giga_loss[loss=0.2553, simple_loss=0.3376, pruned_loss=0.08655, over 28397.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3565, pruned_loss=0.1038, over 5695017.76 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3453, pruned_loss=0.09119, over 5732136.07 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3563, pruned_loss=0.1045, over 5700809.58 frames. ], batch size: 60, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:42:48,423 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751781.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:42:49,868 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:43:00,648 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751797.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:43:27,604 INFO [train.py:968] (0/2) Epoch 17, batch 20950, giga_loss[loss=0.318, simple_loss=0.3858, pruned_loss=0.1251, over 28898.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3563, pruned_loss=0.1025, over 5705116.92 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.345, pruned_loss=0.09107, over 5734482.22 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3566, pruned_loss=0.1033, over 5707076.43 frames. ], batch size: 145, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:43:38,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.119e+02 1.142e+03 1.452e+03 2.058e+03 8.858e+03, threshold=2.903e+03, percent-clipped=10.0
+2023-03-08 21:43:53,617 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:44:05,996 INFO [train.py:968] (0/2) Epoch 17, batch 21000, giga_loss[loss=0.2571, simple_loss=0.3363, pruned_loss=0.08894, over 28864.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3551, pruned_loss=0.1015, over 5714704.49 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3449, pruned_loss=0.09111, over 5739834.57 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3558, pruned_loss=0.1026, over 5710426.87 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:44:06,000 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 21:44:12,810 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3490, 2.9626, 1.4200, 1.5067], device='cuda:0'), covar=tensor([0.1014, 0.0296, 0.0959, 0.1425], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0527, 0.0363, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 21:44:14,794 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.21, simple_loss=0.3167, pruned_loss=0.05164, over 944034.00 frames. 
+2023-03-08 21:44:14,795 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 21:44:38,305 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:44:40,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:44:53,333 INFO [train.py:968] (0/2) Epoch 17, batch 21050, giga_loss[loss=0.2305, simple_loss=0.3123, pruned_loss=0.07441, over 28874.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.353, pruned_loss=0.1006, over 5705572.70 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3458, pruned_loss=0.09177, over 5735531.87 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3533, pruned_loss=0.1014, over 5705258.82 frames. ], batch size: 66, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:45:02,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.879e+02 1.131e+03 1.502e+03 1.851e+03 6.407e+03, threshold=3.003e+03, percent-clipped=7.0
+2023-03-08 21:45:03,187 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:45:13,318 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751957.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:45:30,316 INFO [train.py:968] (0/2) Epoch 17, batch 21100, giga_loss[loss=0.2895, simple_loss=0.3606, pruned_loss=0.1092, over 28918.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3521, pruned_loss=0.1007, over 5711572.48 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3459, pruned_loss=0.0919, over 5740147.49 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3523, pruned_loss=0.1014, over 5706440.11 frames. ], batch size: 213, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:45:47,387 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-752000.pt
+2023-03-08 21:46:09,023 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 21:46:09,381 INFO [train.py:968] (0/2) Epoch 17, batch 21150, giga_loss[loss=0.293, simple_loss=0.3627, pruned_loss=0.1117, over 28708.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3507, pruned_loss=0.1, over 5714631.06 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3465, pruned_loss=0.09229, over 5741052.82 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3505, pruned_loss=0.1004, over 5709116.55 frames. ], batch size: 242, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:46:10,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=752030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:46:15,815 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-08 21:46:22,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.971e+02 1.194e+03 1.432e+03 2.129e+03 5.107e+03, threshold=2.865e+03, percent-clipped=8.0
+2023-03-08 21:46:53,625 INFO [train.py:968] (0/2) Epoch 17, batch 21200, giga_loss[loss=0.3418, simple_loss=0.4053, pruned_loss=0.1392, over 28454.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3524, pruned_loss=0.1016, over 5713213.31 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3473, pruned_loss=0.0928, over 5740060.30 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3516, pruned_loss=0.1015, over 5709038.96 frames. ], batch size: 71, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:47:05,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:47:34,028 INFO [train.py:968] (0/2) Epoch 17, batch 21250, giga_loss[loss=0.2652, simple_loss=0.3429, pruned_loss=0.0937, over 28407.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3522, pruned_loss=0.1016, over 5704772.20 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3472, pruned_loss=0.09292, over 5736257.80 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3517, pruned_loss=0.1017, over 5703264.40 frames. ], batch size: 78, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:47:36,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:47:45,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.120e+02 1.099e+03 1.378e+03 1.738e+03 5.630e+03, threshold=2.757e+03, percent-clipped=9.0
+2023-03-08 21:47:51,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6264, 1.7692, 1.7144, 1.5477], device='cuda:0'), covar=tensor([0.1753, 0.2172, 0.2177, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0731, 0.0689, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 21:47:55,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752156.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:48:13,945 INFO [train.py:968] (0/2) Epoch 17, batch 21300, giga_loss[loss=0.2803, simple_loss=0.3566, pruned_loss=0.102, over 27987.00 frames. ], tot_loss[loss=0.277, simple_loss=0.352, pruned_loss=0.101, over 5706872.06 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3474, pruned_loss=0.09343, over 5729647.07 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3515, pruned_loss=0.1009, over 5709996.28 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:48:44,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5563, 1.7042, 1.4582, 1.5932], device='cuda:0'), covar=tensor([0.2422, 0.2401, 0.2588, 0.2150], device='cuda:0'), in_proj_covar=tensor([0.1420, 0.1036, 0.1257, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 21:48:55,848 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=752228.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:48:56,271 INFO [train.py:968] (0/2) Epoch 17, batch 21350, giga_loss[loss=0.3086, simple_loss=0.3536, pruned_loss=0.1318, over 23443.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3509, pruned_loss=0.0998, over 5706238.38 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3477, pruned_loss=0.09375, over 5733600.92 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3504, pruned_loss=0.09957, over 5704792.10 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:49:02,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:02,564 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:07,194 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:09,214 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 21:49:09,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.598e+02 1.003e+03 1.267e+03 1.805e+03 4.131e+03, threshold=2.535e+03, percent-clipped=11.0
+2023-03-08 21:49:16,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3276, 2.9651, 1.3790, 1.5133], device='cuda:0'), covar=tensor([0.0988, 0.0275, 0.0910, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0524, 0.0362, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:0')
+2023-03-08 21:49:29,390 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:34,280 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:36,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:36,739 INFO [train.py:968] (0/2) Epoch 17, batch 21400, giga_loss[loss=0.2516, simple_loss=0.3282, pruned_loss=0.08746, over 28962.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3506, pruned_loss=0.1004, over 5703367.06 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3478, pruned_loss=0.09379, over 5735191.24 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3501, pruned_loss=0.1003, over 5700393.18 frames. ], batch size: 112, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:49:40,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3434, 3.3199, 1.5376, 1.4332], device='cuda:0'), covar=tensor([0.1029, 0.0242, 0.0938, 0.1461], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0525, 0.0363, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 21:49:53,728 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752299.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:49:56,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752302.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:49:59,470 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:50:08,370 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=752319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:50:15,163 INFO [train.py:968] (0/2) Epoch 17, batch 21450, giga_loss[loss=0.2351, simple_loss=0.3132, pruned_loss=0.07852, over 28329.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3479, pruned_loss=0.09899, over 5705267.78 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3478, pruned_loss=0.0939, over 5739320.15 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3476, pruned_loss=0.099, over 5697947.34 frames. ], batch size: 71, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:50:17,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752331.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:50:17,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752332.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:50:28,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.295e+02 1.052e+03 1.384e+03 1.843e+03 4.582e+03, threshold=2.767e+03, percent-clipped=8.0
+2023-03-08 21:50:58,132 INFO [train.py:968] (0/2) Epoch 17, batch 21500, giga_loss[loss=0.2633, simple_loss=0.3337, pruned_loss=0.09646, over 28674.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3448, pruned_loss=0.09795, over 5704087.05 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3473, pruned_loss=0.09382, over 5739564.49 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3449, pruned_loss=0.09814, over 5697269.35 frames. ], batch size: 71, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:50:58,992 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:51:00,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:51:19,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:51:23,939 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:51:38,575 INFO [train.py:968] (0/2) Epoch 17, batch 21550, giga_loss[loss=0.2356, simple_loss=0.3126, pruned_loss=0.07932, over 28861.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3433, pruned_loss=0.09768, over 5688627.55 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3471, pruned_loss=0.09395, over 5734191.01 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3436, pruned_loss=0.09785, over 5686792.98 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:51:49,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.029e+02 1.185e+03 1.441e+03 2.000e+03 5.536e+03, threshold=2.881e+03, percent-clipped=11.0
+2023-03-08 21:52:13,470 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=752472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:52:15,647 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752475.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:52:18,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752478.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:52:19,652 INFO [train.py:968] (0/2) Epoch 17, batch 21600, giga_loss[loss=0.2561, simple_loss=0.3325, pruned_loss=0.08983, over 28249.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3431, pruned_loss=0.09768, over 5695095.68 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3472, pruned_loss=0.09409, over 5736985.09 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3432, pruned_loss=0.09774, over 5690627.29 frames. ], batch size: 77, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:52:39,899 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752507.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:52:56,413 INFO [train.py:968] (0/2) Epoch 17, batch 21650, giga_loss[loss=0.2617, simple_loss=0.3289, pruned_loss=0.09727, over 28550.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3426, pruned_loss=0.09788, over 5684546.95 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3482, pruned_loss=0.09507, over 5721292.63 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3415, pruned_loss=0.09716, over 5692421.90 frames. ], batch size: 71, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:53:03,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6162, 1.5273, 1.2135, 1.2865], device='cuda:0'), covar=tensor([0.0630, 0.0531, 0.0859, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0440, 0.0509, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 21:53:09,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.595e+02 1.157e+03 1.523e+03 2.356e+03 9.121e+03, threshold=3.046e+03, percent-clipped=16.0
+2023-03-08 21:53:11,969 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:53:17,030 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:53:36,997 INFO [train.py:968] (0/2) Epoch 17, batch 21700, giga_loss[loss=0.2669, simple_loss=0.3363, pruned_loss=0.09869, over 29043.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3397, pruned_loss=0.09653, over 5697979.93 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3491, pruned_loss=0.0958, over 5725436.87 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3378, pruned_loss=0.09531, over 5699703.65 frames. ], batch size: 136, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:53:38,523 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:53:55,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752603.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:54:03,222 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 21:54:16,871 INFO [train.py:968] (0/2) Epoch 17, batch 21750, giga_loss[loss=0.2759, simple_loss=0.3468, pruned_loss=0.1025, over 28851.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3385, pruned_loss=0.09629, over 5702713.83 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3495, pruned_loss=0.09616, over 5728509.02 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3363, pruned_loss=0.09498, over 5700583.25 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:54:26,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4500, 1.6346, 1.6753, 1.2653], device='cuda:0'), covar=tensor([0.1844, 0.2415, 0.1462, 0.1690], device='cuda:0'), in_proj_covar=tensor([0.0872, 0.0692, 0.0919, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 21:54:28,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.698e+02 1.102e+03 1.283e+03 1.582e+03 6.403e+03, threshold=2.565e+03, percent-clipped=6.0
+2023-03-08 21:54:46,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5349, 2.4727, 2.2862, 2.0876], device='cuda:0'), covar=tensor([0.1720, 0.2448, 0.2151, 0.2353], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0733, 0.0690, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 21:54:57,689 INFO [train.py:968] (0/2) Epoch 17, batch 21800, libri_loss[loss=0.3058, simple_loss=0.3703, pruned_loss=0.1206, over 29561.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3362, pruned_loss=0.09492, over 5712994.41 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3498, pruned_loss=0.09654, over 5733100.29 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3338, pruned_loss=0.09352, over 5706690.15 frames. ], batch size: 78, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:55:08,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:55:10,890 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-08 21:55:37,790 INFO [train.py:968] (0/2) Epoch 17, batch 21850, libri_loss[loss=0.2886, simple_loss=0.3605, pruned_loss=0.1083, over 29768.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3366, pruned_loss=0.095, over 5699377.27 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3502, pruned_loss=0.09685, over 5717195.20 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3339, pruned_loss=0.09355, over 5707146.77 frames. ], batch size: 87, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:55:50,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.185e+02 1.042e+03 1.247e+03 1.736e+03 4.882e+03, threshold=2.495e+03, percent-clipped=9.0
+2023-03-08 21:55:53,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752746.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:55:55,559 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752749.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:56:21,199 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752778.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:56:21,646 INFO [train.py:968] (0/2) Epoch 17, batch 21900, libri_loss[loss=0.2579, simple_loss=0.3255, pruned_loss=0.09518, over 29474.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3388, pruned_loss=0.09563, over 5700296.76 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3501, pruned_loss=0.09705, over 5721798.21 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3364, pruned_loss=0.09422, over 5701802.59 frames. ], batch size: 70, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:56:32,538 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 21:56:42,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=752803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:57:01,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1361, 1.1920, 1.0634, 0.9121], device='cuda:0'), covar=tensor([0.0930, 0.0526, 0.1056, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0438, 0.0506, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 21:57:06,960 INFO [train.py:968] (0/2) Epoch 17, batch 21950, giga_loss[loss=0.2869, simple_loss=0.3649, pruned_loss=0.1045, over 27957.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3422, pruned_loss=0.09684, over 5684131.06 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3503, pruned_loss=0.09721, over 5713532.38 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3401, pruned_loss=0.09558, over 5693138.51 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:57:15,176 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:57:17,011 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:57:20,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.256e+02 1.119e+03 1.541e+03 2.089e+03 8.645e+03, threshold=3.082e+03, percent-clipped=15.0
+2023-03-08 21:57:21,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:57:32,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4971, 1.7850, 1.4807, 1.5084], device='cuda:0'), covar=tensor([0.2503, 0.2539, 0.2887, 0.2160], device='cuda:0'), in_proj_covar=tensor([0.1422, 0.1035, 0.1259, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 21:57:40,743 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:57:47,782 INFO [train.py:968] (0/2) Epoch 17, batch 22000, giga_loss[loss=0.2769, simple_loss=0.353, pruned_loss=0.1004, over 28868.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3441, pruned_loss=0.09714, over 5684866.92 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3508, pruned_loss=0.09777, over 5708355.00 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3418, pruned_loss=0.09561, over 5696169.24 frames. ], batch size: 186, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:57:55,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3641, 1.5838, 1.3548, 1.1987], device='cuda:0'), covar=tensor([0.2915, 0.2397, 0.1893, 0.2462], device='cuda:0'), in_proj_covar=tensor([0.1873, 0.1800, 0.1733, 0.1874], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 21:58:30,393 INFO [train.py:968] (0/2) Epoch 17, batch 22050, giga_loss[loss=0.2318, simple_loss=0.3116, pruned_loss=0.07597, over 29079.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3449, pruned_loss=0.09719, over 5685523.89 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3512, pruned_loss=0.09823, over 5702607.62 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3426, pruned_loss=0.09554, over 5699862.62 frames. ], batch size: 155, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:58:43,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.676e+02 1.075e+03 1.351e+03 1.906e+03 5.217e+03, threshold=2.701e+03, percent-clipped=5.0
+2023-03-08 21:58:47,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3119, 3.0408, 1.4464, 1.4259], device='cuda:0'), covar=tensor([0.0935, 0.0304, 0.0948, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0531, 0.0364, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 21:59:10,982 INFO [train.py:968] (0/2) Epoch 17, batch 22100, giga_loss[loss=0.2329, simple_loss=0.318, pruned_loss=0.07391, over 28833.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3457, pruned_loss=0.09797, over 5693092.33 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3521, pruned_loss=0.09918, over 5711465.04 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3427, pruned_loss=0.09568, over 5695969.79 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:59:15,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 21:59:21,048 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:59:23,695 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:59:29,442 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 21:59:46,726 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=753022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:59:50,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=753025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:59:53,165 INFO [train.py:968] (0/2) Epoch 17, batch 22150, giga_loss[loss=0.2958, simple_loss=0.3615, pruned_loss=0.115, over 28948.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3452, pruned_loss=0.09768, over 5697370.55 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.352, pruned_loss=0.09924, over 5713648.04 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3427, pruned_loss=0.09578, over 5697438.11 frames. ], batch size: 213, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:00:06,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.872e+02 1.224e+03 1.696e+03 2.295e+03 6.118e+03, threshold=3.392e+03, percent-clipped=14.0
+2023-03-08 22:00:09,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2273, 3.1146, 1.4423, 1.3186], device='cuda:0'), covar=tensor([0.1007, 0.0376, 0.0919, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0532, 0.0365, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 22:00:19,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4878, 2.1325, 1.5204, 0.7415], device='cuda:0'), covar=tensor([0.5781, 0.2667, 0.4195, 0.6066], device='cuda:0'), in_proj_covar=tensor([0.1656, 0.1562, 0.1538, 0.1349], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 22:00:34,574 INFO [train.py:968] (0/2) Epoch 17, batch 22200, giga_loss[loss=0.2878, simple_loss=0.3551, pruned_loss=0.1103, over 28946.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.346, pruned_loss=0.0986, over 5699094.33 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3524, pruned_loss=0.09957, over 5718487.90 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3436, pruned_loss=0.09677, over 5694613.71 frames. ], batch size: 136, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:01:14,878 INFO [train.py:968] (0/2) Epoch 17, batch 22250, giga_loss[loss=0.3095, simple_loss=0.3682, pruned_loss=0.1254, over 28466.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3478, pruned_loss=0.09966, over 5706444.88 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3528, pruned_loss=0.1, over 5724944.33 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3453, pruned_loss=0.09776, over 5696409.50 frames. ], batch size: 78, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:01:19,332 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 22:01:22,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1421, 1.2013, 1.0440, 0.8391], device='cuda:0'), covar=tensor([0.0839, 0.0507, 0.1036, 0.1084], device='cuda:0'), in_proj_covar=tensor([0.0373, 0.0437, 0.0505, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-08 22:01:30,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.035e+02 1.310e+03 1.625e+03 2.072e+03 5.203e+03, threshold=3.250e+03, percent-clipped=5.0
+2023-03-08 22:01:33,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 22:01:55,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=753178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:01:55,555 INFO [train.py:968] (0/2) Epoch 17, batch 22300, giga_loss[loss=0.2748, simple_loss=0.3502, pruned_loss=0.09976, over 29009.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3521, pruned_loss=0.1023, over 5708735.17 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3535, pruned_loss=0.1008, over 5721853.16 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3493, pruned_loss=0.1001, over 5703415.52 frames. ], batch size: 128, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:02:02,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4405, 1.6629, 1.3496, 1.4722], device='cuda:0'), covar=tensor([0.0738, 0.0303, 0.0323, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0114, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 22:02:36,906 INFO [train.py:968] (0/2) Epoch 17, batch 22350, giga_loss[loss=0.2756, simple_loss=0.3503, pruned_loss=0.1005, over 28915.00 frames. ], tot_loss[loss=0.28, simple_loss=0.354, pruned_loss=0.103, over 5714100.04 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3541, pruned_loss=0.1014, over 5725508.39 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3513, pruned_loss=0.1007, over 5706168.86 frames. ], batch size: 186, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:02:50,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.294e+02 1.398e+03 1.698e+03 2.319e+03 7.817e+03, threshold=3.396e+03, percent-clipped=11.0
+2023-03-08 22:03:07,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=753266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:03:16,421 INFO [train.py:968] (0/2) Epoch 17, batch 22400, giga_loss[loss=0.3068, simple_loss=0.3799, pruned_loss=0.1168, over 28274.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3543, pruned_loss=0.1031, over 5719414.07 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.354, pruned_loss=0.1014, over 5727475.69 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3522, pruned_loss=0.1013, over 5711266.71 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:03:51,241 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=753321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:03:54,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=753324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:03:57,476 INFO [train.py:968] (0/2) Epoch 17, batch 22450, giga_loss[loss=0.2372, simple_loss=0.3297, pruned_loss=0.07242, over 28996.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3549, pruned_loss=0.1037, over 5712899.62 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3551, pruned_loss=0.1022, over 5724656.37 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3523, pruned_loss=0.1016, over 5709010.87 frames. ], batch size: 164, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:04:13,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.052e+02 1.191e+03 1.456e+03 1.856e+03 4.243e+03, threshold=2.913e+03, percent-clipped=2.0
+2023-03-08 22:04:17,458 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=753351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:04:18,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 22:04:19,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=753353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:04:27,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8465, 1.1142, 1.0938, 0.8158], device='cuda:0'), covar=tensor([0.2112, 0.2163, 0.1295, 0.1958], device='cuda:0'), in_proj_covar=tensor([0.1877, 0.1804, 0.1735, 0.1875], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 22:04:33,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6346, 1.7015, 1.2618, 1.3327], device='cuda:0'), covar=tensor([0.0837, 0.0627, 0.1083, 0.1145], device='cuda:0'), in_proj_covar=tensor([0.0375, 0.0440, 0.0507, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 22:04:40,049 INFO [train.py:968] (0/2) Epoch 17, batch 22500, giga_loss[loss=0.2931, simple_loss=0.3667, pruned_loss=0.1098, over 28548.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3553, pruned_loss=0.1043, over 5716164.61 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3553, pruned_loss=0.1024, over 5726639.64 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3531, pruned_loss=0.1025, over 5711205.62 frames. ], batch size: 336, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:05:00,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=753400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:05:23,635 INFO [train.py:968] (0/2) Epoch 17, batch 22550, giga_loss[loss=0.3019, simple_loss=0.3651, pruned_loss=0.1194, over 28682.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3525, pruned_loss=0.1025, over 5720073.89 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3557, pruned_loss=0.1027, over 5728540.15 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3504, pruned_loss=0.1008, over 5714469.85 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:05:40,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.429e+02 1.123e+03 1.503e+03 2.074e+03 8.306e+03, threshold=3.005e+03, percent-clipped=7.0
+2023-03-08 22:06:06,903 INFO [train.py:968] (0/2) Epoch 17, batch 22600, giga_loss[loss=0.256, simple_loss=0.3283, pruned_loss=0.09188, over 28777.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3494, pruned_loss=0.1009, over 5722310.78 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3564, pruned_loss=0.1035, over 5728531.80 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.347, pruned_loss=0.09887, over 5717600.96 frames. ], batch size: 99, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:06:47,128 INFO [train.py:968] (0/2) Epoch 17, batch 22650, giga_loss[loss=0.2675, simple_loss=0.3452, pruned_loss=0.09491, over 27920.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3472, pruned_loss=0.0998, over 5708501.90 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3575, pruned_loss=0.1044, over 5718018.48 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3441, pruned_loss=0.09729, over 5713791.20 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:06:59,756 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=753543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:07:02,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=753546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:07:03,347 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.321e+02 1.130e+03 1.362e+03 1.878e+03 4.729e+03, threshold=2.725e+03, percent-clipped=9.0
+2023-03-08 22:07:23,745 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=753570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:07:28,406 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=753575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:07:32,426 INFO [train.py:968] (0/2) Epoch 17, batch 22700, giga_loss[loss=0.2233, simple_loss=0.3029, pruned_loss=0.07185, over 28383.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3467, pruned_loss=0.09806, over 5707913.78 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3575, pruned_loss=0.1045, over 5718972.39 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3442, pruned_loss=0.09599, over 5711246.99 frames. ], batch size: 78, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:08:15,072 INFO [train.py:968] (0/2) Epoch 17, batch 22750, giga_loss[loss=0.2702, simple_loss=0.3543, pruned_loss=0.09301, over 28973.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3467, pruned_loss=0.09685, over 5709542.09 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3573, pruned_loss=0.1044, over 5721553.70 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3447, pruned_loss=0.09513, over 5709682.79 frames. ], batch size: 213, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:08:22,830 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=753641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:08:30,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.687e+02 1.158e+03 1.487e+03 2.003e+03 5.878e+03, threshold=2.974e+03, percent-clipped=4.0
+2023-03-08 22:08:37,597 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-08 22:08:38,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3159, 1.8522, 1.5453, 1.4872], device='cuda:0'), covar=tensor([0.0732, 0.0309, 0.0312, 0.0813], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 22:08:53,374 INFO [train.py:968] (0/2) Epoch 17, batch 22800, libri_loss[loss=0.3369, simple_loss=0.3838, pruned_loss=0.145, over 29402.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3464, pruned_loss=0.097, over 5725533.38 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3579, pruned_loss=0.105, over 5727974.77 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3438, pruned_loss=0.09475, over 5719501.16 frames. ], batch size: 67, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:09:34,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=753726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:09:36,580 INFO [train.py:968] (0/2) Epoch 17, batch 22850, giga_loss[loss=0.3096, simple_loss=0.3593, pruned_loss=0.1299, over 28919.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3461, pruned_loss=0.09819, over 5722524.54 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3585, pruned_loss=0.1054, over 5727484.61 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3433, pruned_loss=0.09586, over 5717927.06 frames. ], batch size: 106, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:09:48,185 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 22:09:52,897 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.998e+02 1.129e+03 1.420e+03 2.079e+03 8.848e+03, threshold=2.840e+03, percent-clipped=10.0
+2023-03-08 22:09:54,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5344, 3.8302, 1.5045, 1.6794], device='cuda:0'), covar=tensor([0.0889, 0.0357, 0.0900, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0531, 0.0364, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 22:10:13,749 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=753774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:10:17,776 INFO [train.py:968] (0/2) Epoch 17, batch 22900, giga_loss[loss=0.2363, simple_loss=0.3099, pruned_loss=0.08136, over 28971.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3442, pruned_loss=0.09884, over 5716183.28 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3584, pruned_loss=0.1056, over 5722184.26 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3417, pruned_loss=0.0967, over 5716540.88 frames. ], batch size: 136, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:10:21,671 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=753784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:10:23,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=753787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:10:47,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=753816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:10:57,688 INFO [train.py:968] (0/2) Epoch 17, batch 22950, giga_loss[loss=0.2386, simple_loss=0.3098, pruned_loss=0.08363, over 28432.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3446, pruned_loss=0.1007, over 5717094.23 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3588, pruned_loss=0.106, over 5724570.66 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.342, pruned_loss=0.09858, over 5715194.81 frames. ], batch size: 65, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:11:14,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.407e+02 1.202e+03 1.407e+03 1.826e+03 4.357e+03, threshold=2.815e+03, percent-clipped=10.0
+2023-03-08 22:11:24,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9182, 2.0158, 1.4218, 1.6486], device='cuda:0'), covar=tensor([0.0811, 0.0592, 0.1029, 0.1086], device='cuda:0'), in_proj_covar=tensor([0.0374, 0.0441, 0.0506, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 22:11:30,455 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=753869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:11:33,914 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=753872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:11:37,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2161, 1.5584, 1.2055, 0.9578], device='cuda:0'), covar=tensor([0.2466, 0.2470, 0.2869, 0.2268], device='cuda:0'), in_proj_covar=tensor([0.1419, 0.1031, 0.1258, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 22:11:38,840 INFO [train.py:968] (0/2) Epoch 17, batch 23000, giga_loss[loss=0.2555, simple_loss=0.3317, pruned_loss=0.0897, over 28926.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3425, pruned_loss=0.09986, over 5721327.77 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3586, pruned_loss=0.1059, over 5727427.99 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3404, pruned_loss=0.09816, over 5717122.85 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:11:54,632 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=753901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:12:16,500 INFO [train.py:968] (0/2) Epoch 17, batch 23050, giga_loss[loss=0.2569, simple_loss=0.3297, pruned_loss=0.09204, over 28948.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3403, pruned_loss=0.09866, over 5722776.55 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.359, pruned_loss=0.1063, over 5733451.18 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3377, pruned_loss=0.09674, over 5713739.16 frames. ], batch size: 106, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:12:30,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=753945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:12:32,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.227e+02 1.136e+03 1.443e+03 1.814e+03 5.118e+03, threshold=2.887e+03, percent-clipped=2.0
+2023-03-08 22:12:59,086 INFO [train.py:968] (0/2) Epoch 17, batch 23100, giga_loss[loss=0.2529, simple_loss=0.3165, pruned_loss=0.09466, over 23795.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3354, pruned_loss=0.09586, over 5723197.68 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3592, pruned_loss=0.1063, over 5734452.92 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3331, pruned_loss=0.09424, over 5715116.29 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:13:16,462 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-754000.pt
+2023-03-08 22:13:36,910 INFO [train.py:968] (0/2) Epoch 17, batch 23150, giga_loss[loss=0.2182, simple_loss=0.2971, pruned_loss=0.0697, over 28557.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3319, pruned_loss=0.0941, over 5719998.48 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3595, pruned_loss=0.1068, over 5729002.37 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3288, pruned_loss=0.09187, over 5717549.60 frames. ], batch size: 71, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:13:54,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.234e+02 1.287e+03 1.553e+03 1.949e+03 5.952e+03, threshold=3.107e+03, percent-clipped=10.0
+2023-03-08 22:13:55,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3457, 1.4006, 1.1597, 1.5393], device='cuda:0'), covar=tensor([0.0746, 0.0334, 0.0353, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 22:14:19,331 INFO [train.py:968] (0/2) Epoch 17, batch 23200, giga_loss[loss=0.2532, simple_loss=0.3351, pruned_loss=0.0857, over 28920.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3325, pruned_loss=0.09416, over 5719116.65 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3587, pruned_loss=0.1065, over 5733549.40 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3301, pruned_loss=0.09236, over 5712673.99 frames. ], batch size: 186, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:14:27,878 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=754088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:14:29,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=754091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:14:54,538 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=754120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:14:56,716 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=754123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:15:01,080 INFO [train.py:968] (0/2) Epoch 17, batch 23250, giga_loss[loss=0.2778, simple_loss=0.3565, pruned_loss=0.09956, over 28632.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3359, pruned_loss=0.09576, over 5717539.63 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3587, pruned_loss=0.1067, over 5734988.78 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3337, pruned_loss=0.09402, over 5710778.48 frames. ], batch size: 336, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:15:10,679 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6523, 4.5476, 1.8343, 1.7234], device='cuda:0'), covar=tensor([0.0919, 0.0305, 0.0912, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0532, 0.0365, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 22:15:17,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.361e+02 1.151e+03 1.481e+03 1.910e+03 5.767e+03, threshold=2.963e+03, percent-clipped=8.0
+2023-03-08 22:15:18,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=754149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:15:42,875 INFO [train.py:968] (0/2) Epoch 17, batch 23300, giga_loss[loss=0.283, simple_loss=0.3599, pruned_loss=0.1031, over 28782.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3394, pruned_loss=0.09724, over 5710432.44 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.359, pruned_loss=0.1072, over 5731288.92 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3368, pruned_loss=0.0951, over 5707741.31 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:15:47,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4060, 3.0101, 1.4023, 1.5190], device='cuda:0'), covar=tensor([0.0925, 0.0357, 0.0936, 0.1304], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0533, 0.0365, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 22:15:54,826 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4337, 1.7082, 1.6285, 1.5148], device='cuda:0'), covar=tensor([0.1665, 0.2033, 0.2099, 0.2024], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0739, 0.0696, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 22:16:26,635 INFO [train.py:968] (0/2) Epoch 17, batch 23350, giga_loss[loss=0.2799, simple_loss=0.3514, pruned_loss=0.1041, over 28742.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3432, pruned_loss=0.09853, over 5713749.50 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3591, pruned_loss=0.1073, over 5732965.75 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3408, pruned_loss=0.09664, over 5709985.07 frames. ], batch size: 99, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:16:48,995 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.994e+02 1.224e+03 1.480e+03 2.179e+03 4.956e+03, threshold=2.959e+03, percent-clipped=10.0
+2023-03-08 22:17:16,140 INFO [train.py:968] (0/2) Epoch 17, batch 23400, giga_loss[loss=0.2593, simple_loss=0.3397, pruned_loss=0.08941, over 28957.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3454, pruned_loss=0.09949, over 5724725.58 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3589, pruned_loss=0.1074, over 5737779.61 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3433, pruned_loss=0.09774, over 5717170.91 frames. ], batch size: 155, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:17:26,089 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=754292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:17:26,758 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6238, 2.2333, 1.9414, 1.8199], device='cuda:0'), covar=tensor([0.0710, 0.0236, 0.0273, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 22:17:30,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=754295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:17:54,864 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=754324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:17:59,778 INFO [train.py:968] (0/2) Epoch 17, batch 23450, giga_loss[loss=0.3119, simple_loss=0.3784, pruned_loss=0.1228, over 28736.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3484, pruned_loss=0.1021, over 5709597.52 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.359, pruned_loss=0.1075, over 5731298.53 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3464, pruned_loss=0.1004, over 5708452.00 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:18:17,168 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.937e+02 1.263e+03 1.836e+03 2.525e+03 8.413e+03, threshold=3.671e+03, percent-clipped=15.0
+2023-03-08 22:18:21,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1905, 1.7294, 1.2871, 0.3973], device='cuda:0'), covar=tensor([0.3572, 0.2205, 0.3281, 0.4783], device='cuda:0'), in_proj_covar=tensor([0.1670, 0.1574, 0.1547, 0.1359], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 22:18:46,907 INFO [train.py:968] (0/2) Epoch 17, batch 23500, libri_loss[loss=0.2668, simple_loss=0.3355, pruned_loss=0.09899, over 29544.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3549, pruned_loss=0.1075, over 5688603.00 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3592, pruned_loss=0.1079, over 5711637.82 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3529, pruned_loss=0.1058, over 5703461.81 frames. ], batch size: 76, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:18:49,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1886, 1.1843, 3.8397, 3.1302], device='cuda:0'), covar=tensor([0.1724, 0.2856, 0.0434, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.0722, 0.0621, 0.0915, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 22:19:38,260 INFO [train.py:968] (0/2) Epoch 17, batch 23550, libri_loss[loss=0.3196, simple_loss=0.3865, pruned_loss=0.1264, over 29289.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3614, pruned_loss=0.1128, over 5671833.31 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3586, pruned_loss=0.1077, over 5705908.32 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3603, pruned_loss=0.1116, over 5687635.83 frames. ], batch size: 94, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:19:57,717 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.232e+03 1.995e+03 2.606e+03 3.527e+03 1.230e+04, threshold=5.213e+03, percent-clipped=24.0
+2023-03-08 22:20:10,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6982, 3.7045, 1.6695, 1.7545], device='cuda:0'), covar=tensor([0.0870, 0.0379, 0.0814, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0533, 0.0364, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 22:20:20,095 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-08 22:20:29,113 INFO [train.py:968] (0/2) Epoch 17, batch 23600, giga_loss[loss=0.3212, simple_loss=0.3809, pruned_loss=0.1307, over 28483.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3673, pruned_loss=0.117, over 5667598.99 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3586, pruned_loss=0.1079, over 5707828.48 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3665, pruned_loss=0.116, over 5677713.31 frames. ], batch size: 85, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:20:49,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=754498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:21:20,163 INFO [train.py:968] (0/2) Epoch 17, batch 23650, giga_loss[loss=0.2825, simple_loss=0.3562, pruned_loss=0.1043, over 29006.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3726, pruned_loss=0.1215, over 5669649.79 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3587, pruned_loss=0.1079, over 5710167.82 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3722, pruned_loss=0.1209, over 5674692.17 frames. ], batch size: 128, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:21:39,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.730e+03 2.108e+03 2.763e+03 5.602e+03, threshold=4.217e+03, percent-clipped=1.0
+2023-03-08 22:22:06,596 INFO [train.py:968] (0/2) Epoch 17, batch 23700, giga_loss[loss=0.3152, simple_loss=0.3779, pruned_loss=0.1262, over 28637.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3785, pruned_loss=0.1262, over 5672696.26 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3595, pruned_loss=0.1086, over 5708928.32 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3781, pruned_loss=0.1257, over 5676256.41 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:22:43,115 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=754616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:22:56,157 INFO [train.py:968] (0/2) Epoch 17, batch 23750, giga_loss[loss=0.4237, simple_loss=0.4451, pruned_loss=0.2011, over 26672.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3832, pruned_loss=0.1307, over 5670630.88 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3595, pruned_loss=0.1087, over 5711907.52 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3831, pruned_loss=0.1305, over 5670318.86 frames. ], batch size: 555, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:23:08,323 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=754641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:23:11,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=754644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:23:17,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.901e+03 2.605e+03 3.899e+03 9.880e+03, threshold=5.210e+03, percent-clipped=18.0
+2023-03-08 22:23:40,443 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=754673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:23:47,292 INFO [train.py:968] (0/2) Epoch 17, batch 23800, giga_loss[loss=0.3658, simple_loss=0.411, pruned_loss=0.1603, over 28423.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.386, pruned_loss=0.1341, over 5660446.09 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3598, pruned_loss=0.1088, over 5713875.34 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3862, pruned_loss=0.1344, over 5657468.25 frames. ], batch size: 370, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:23:49,114 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=754680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:24:43,141 INFO [train.py:968] (0/2) Epoch 17, batch 23850, libri_loss[loss=0.3223, simple_loss=0.3863, pruned_loss=0.1291, over 29536.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3908, pruned_loss=0.1393, over 5650377.44 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3599, pruned_loss=0.1089, over 5714987.39 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3913, pruned_loss=0.1398, over 5646092.20 frames. ], batch size: 83, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:25:04,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.731e+03 2.175e+03 3.054e+03 5.330e+03, threshold=4.350e+03, percent-clipped=1.0
+2023-03-08 22:25:07,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6294, 1.7851, 1.7707, 1.3773], device='cuda:0'), covar=tensor([0.1622, 0.2310, 0.1347, 0.1635], device='cuda:0'), in_proj_covar=tensor([0.0863, 0.0689, 0.0911, 0.0809], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 22:25:25,465 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-08 22:25:33,552 INFO [train.py:968] (0/2) Epoch 17, batch 23900, giga_loss[loss=0.4458, simple_loss=0.4455, pruned_loss=0.2231, over 23600.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3945, pruned_loss=0.1429, over 5629496.83 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3604, pruned_loss=0.1097, over 5697836.29 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3955, pruned_loss=0.1437, over 5638002.76 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:25:52,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7151, 1.3318, 5.0269, 3.7045], device='cuda:0'), covar=tensor([0.1576, 0.2743, 0.0384, 0.0775], device='cuda:0'), in_proj_covar=tensor([0.0724, 0.0622, 0.0919, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 22:26:32,907 INFO [train.py:968] (0/2) Epoch 17, batch 23950, giga_loss[loss=0.3006, simple_loss=0.3655, pruned_loss=0.1179, over 28945.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.395, pruned_loss=0.1431, over 5644047.82 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3604, pruned_loss=0.1098, over 5701786.48 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3964, pruned_loss=0.1442, over 5646405.37 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:26:52,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.924e+03 2.570e+03 3.481e+03 7.754e+03, threshold=5.141e+03, percent-clipped=15.0
+2023-03-08 22:27:21,711 INFO [train.py:968] (0/2) Epoch 17, batch 24000, giga_loss[loss=0.3715, simple_loss=0.4095, pruned_loss=0.1667, over 27482.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3943, pruned_loss=0.1435, over 5626662.59 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3603, pruned_loss=0.1099, over 5698027.05 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3965, pruned_loss=0.1454, over 5630321.61 frames. ], batch size: 472, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:27:21,716 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 22:27:29,869 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2087, simple_loss=0.3157, pruned_loss=0.05078, over 944034.00 frames. 
+2023-03-08 22:27:29,870 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 22:27:45,701 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=754896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:28:07,563 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4697, 1.7630, 1.4060, 1.5063], device='cuda:0'), covar=tensor([0.2532, 0.2578, 0.2969, 0.2241], device='cuda:0'), in_proj_covar=tensor([0.1420, 0.1033, 0.1259, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 22:28:13,443 INFO [train.py:968] (0/2) Epoch 17, batch 24050, libri_loss[loss=0.2393, simple_loss=0.3046, pruned_loss=0.08694, over 28597.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3908, pruned_loss=0.1408, over 5635158.79 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3599, pruned_loss=0.1098, over 5693138.07 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3941, pruned_loss=0.1434, over 5640381.91 frames. ], batch size: 63, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:28:33,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.731e+03 2.101e+03 2.619e+03 1.989e+04, threshold=4.202e+03, percent-clipped=3.0
+2023-03-08 22:28:59,834 INFO [train.py:968] (0/2) Epoch 17, batch 24100, giga_loss[loss=0.3244, simple_loss=0.3951, pruned_loss=0.1268, over 28905.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3903, pruned_loss=0.1392, over 5648139.48 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3602, pruned_loss=0.1101, over 5701020.83 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3938, pruned_loss=0.1421, over 5643357.60 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:29:16,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=754991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:29:19,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4497, 2.1415, 1.4904, 0.6313], device='cuda:0'), covar=tensor([0.5876, 0.2886, 0.4174, 0.6202], device='cuda:0'), in_proj_covar=tensor([0.1682, 0.1588, 0.1555, 0.1370], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 22:29:30,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5063, 1.8790, 1.5473, 1.7151], device='cuda:0'), covar=tensor([0.0765, 0.0277, 0.0294, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 22:29:32,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4261, 3.4109, 1.5476, 1.6644], device='cuda:0'), covar=tensor([0.0925, 0.0388, 0.0888, 0.1283], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0538, 0.0367, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 22:29:52,885 INFO [train.py:968] (0/2) Epoch 17, batch 24150, giga_loss[loss=0.3775, simple_loss=0.4181, pruned_loss=0.1685, over 27457.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3903, pruned_loss=0.1386, over 5644938.33 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3606, pruned_loss=0.1104, over 5702918.72 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3936, pruned_loss=0.1415, over 5637882.60 frames. ], batch size: 472, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:30:16,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.199e+03 1.653e+03 2.033e+03 2.867e+03 5.768e+03, threshold=4.067e+03, percent-clipped=7.0
+2023-03-08 22:30:18,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:30:21,826 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5313, 1.8727, 1.5023, 1.7647], device='cuda:0'), covar=tensor([0.2438, 0.2505, 0.2743, 0.2013], device='cuda:0'), in_proj_covar=tensor([0.1421, 0.1033, 0.1261, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 22:30:43,467 INFO [train.py:968] (0/2) Epoch 17, batch 24200, giga_loss[loss=0.317, simple_loss=0.3895, pruned_loss=0.1223, over 29027.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3908, pruned_loss=0.1393, over 5638063.73 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.36, pruned_loss=0.1105, over 5711160.95 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3953, pruned_loss=0.1427, over 5621985.17 frames. ], batch size: 164, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:31:04,461 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:31:13,540 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-08 22:31:34,943 INFO [train.py:968] (0/2) Epoch 17, batch 24250, giga_loss[loss=0.361, simple_loss=0.4126, pruned_loss=0.1547, over 27682.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3879, pruned_loss=0.1363, over 5626900.96 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3602, pruned_loss=0.1107, over 5703783.59 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3917, pruned_loss=0.1392, over 5618844.61 frames. ], batch size: 472, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:31:42,102 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:31:45,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:31:58,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.427e+02 1.604e+03 2.041e+03 2.886e+03 8.219e+03, threshold=4.083e+03, percent-clipped=8.0
+2023-03-08 22:32:12,081 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:32:24,701 INFO [train.py:968] (0/2) Epoch 17, batch 24300, giga_loss[loss=0.3007, simple_loss=0.3745, pruned_loss=0.1134, over 28839.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3854, pruned_loss=0.1326, over 5635920.67 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3605, pruned_loss=0.111, over 5703858.60 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3886, pruned_loss=0.1351, over 5628180.43 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:32:41,262 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:32:45,588 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:32:47,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:33:17,069 INFO [train.py:968] (0/2) Epoch 17, batch 24350, giga_loss[loss=0.2968, simple_loss=0.3667, pruned_loss=0.1134, over 28021.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3818, pruned_loss=0.1291, over 5653729.34 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3604, pruned_loss=0.111, over 5704893.30 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3844, pruned_loss=0.1311, over 5646492.60 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:33:18,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:33:38,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.979e+02 1.658e+03 2.179e+03 3.235e+03 7.580e+03, threshold=4.358e+03, percent-clipped=14.0
+2023-03-08 22:33:46,905 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:33:56,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:34:00,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:34:04,662 INFO [train.py:968] (0/2) Epoch 17, batch 24400, giga_loss[loss=0.3128, simple_loss=0.3725, pruned_loss=0.1266, over 27903.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3794, pruned_loss=0.1269, over 5663960.71 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3602, pruned_loss=0.111, over 5706564.36 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3822, pruned_loss=0.1289, over 5655675.62 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:34:50,319 INFO [train.py:968] (0/2) Epoch 17, batch 24450, giga_loss[loss=0.2793, simple_loss=0.3478, pruned_loss=0.1054, over 28526.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3765, pruned_loss=0.1255, over 5655924.25 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3599, pruned_loss=0.1111, over 5703454.17 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3796, pruned_loss=0.1276, over 5650832.57 frames. ], batch size: 65, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:35:14,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.591e+03 2.217e+03 2.796e+03 8.992e+03, threshold=4.435e+03, percent-clipped=13.0
+2023-03-08 22:35:43,748 INFO [train.py:968] (0/2) Epoch 17, batch 24500, giga_loss[loss=0.3071, simple_loss=0.3702, pruned_loss=0.1219, over 28236.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3771, pruned_loss=0.1261, over 5654788.83 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3599, pruned_loss=0.1112, over 5701922.80 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3799, pruned_loss=0.1279, over 5651315.01 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:35:48,719 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7476, 1.7609, 1.9959, 1.5180], device='cuda:0'), covar=tensor([0.1862, 0.2533, 0.1443, 0.1771], device='cuda:0'), in_proj_covar=tensor([0.0867, 0.0694, 0.0913, 0.0812], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 22:36:03,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:36:21,373 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:36:25,145 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:36:36,668 INFO [train.py:968] (0/2) Epoch 17, batch 24550, giga_loss[loss=0.2801, simple_loss=0.3514, pruned_loss=0.1043, over 28903.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3758, pruned_loss=0.1244, over 5669519.43 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3599, pruned_loss=0.1111, over 5703487.08 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3781, pruned_loss=0.1259, over 5664977.36 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:36:56,620 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:37:02,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.625e+02 1.452e+03 1.770e+03 2.499e+03 5.352e+03, threshold=3.539e+03, percent-clipped=2.0
+2023-03-08 22:37:25,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:37:29,590 INFO [train.py:968] (0/2) Epoch 17, batch 24600, giga_loss[loss=0.2683, simple_loss=0.3535, pruned_loss=0.09157, over 28874.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3751, pruned_loss=0.1219, over 5668668.30 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3599, pruned_loss=0.1112, over 5699380.70 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3772, pruned_loss=0.1233, over 5667945.62 frames. ], batch size: 145, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:38:21,951 INFO [train.py:968] (0/2) Epoch 17, batch 24650, giga_loss[loss=0.3244, simple_loss=0.3686, pruned_loss=0.1401, over 23640.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3756, pruned_loss=0.1208, over 5653777.89 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3598, pruned_loss=0.1112, over 5692437.99 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3778, pruned_loss=0.1222, over 5658078.22 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:38:47,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.521e+02 1.615e+03 2.188e+03 3.267e+03 9.512e+03, threshold=4.375e+03, percent-clipped=18.0
+2023-03-08 22:39:03,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:39:05,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6725, 1.8511, 1.8486, 1.4258], device='cuda:0'), covar=tensor([0.1775, 0.2360, 0.1407, 0.1688], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0693, 0.0912, 0.0811], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 22:39:14,080 INFO [train.py:968] (0/2) Epoch 17, batch 24700, giga_loss[loss=0.3057, simple_loss=0.3731, pruned_loss=0.1191, over 29014.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3767, pruned_loss=0.1216, over 5657931.32 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3594, pruned_loss=0.111, over 5694653.10 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3789, pruned_loss=0.123, over 5658990.20 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:39:49,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7888, 1.7885, 1.3517, 1.3875], device='cuda:0'), covar=tensor([0.0853, 0.0643, 0.1020, 0.1146], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0442, 0.0508, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 22:39:50,593 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:39:53,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:40:02,728 INFO [train.py:968] (0/2) Epoch 17, batch 24750, giga_loss[loss=0.3174, simple_loss=0.3794, pruned_loss=0.1277, over 28972.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3773, pruned_loss=0.1228, over 5654808.20 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3594, pruned_loss=0.1109, over 5695253.61 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3796, pruned_loss=0.1241, over 5654593.55 frames. ], batch size: 213, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:40:07,442 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:40:12,908 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:40:24,452 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:40:24,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:40:26,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.818e+03 2.247e+03 2.869e+03 7.978e+03, threshold=4.494e+03, percent-clipped=11.0
+2023-03-08 22:40:55,600 INFO [train.py:968] (0/2) Epoch 17, batch 24800, giga_loss[loss=0.2619, simple_loss=0.3381, pruned_loss=0.09284, over 28903.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3754, pruned_loss=0.1224, over 5648847.52 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3594, pruned_loss=0.1109, over 5695253.61 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3771, pruned_loss=0.1235, over 5648680.46 frames. ], batch size: 145, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:41:29,460 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:41:31,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:41:43,261 INFO [train.py:968] (0/2) Epoch 17, batch 24850, giga_loss[loss=0.2856, simple_loss=0.355, pruned_loss=0.1081, over 28182.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3737, pruned_loss=0.122, over 5664060.82 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3591, pruned_loss=0.1108, over 5698084.44 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3757, pruned_loss=0.1233, over 5660418.02 frames. ], batch size: 77, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:41:56,682 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:06,708 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.746e+02 1.680e+03 2.179e+03 2.990e+03 9.017e+03, threshold=4.358e+03, percent-clipped=7.0
+2023-03-08 22:42:20,781 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:27,148 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:27,532 INFO [train.py:968] (0/2) Epoch 17, batch 24900, libri_loss[loss=0.3427, simple_loss=0.4031, pruned_loss=0.1412, over 29621.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3738, pruned_loss=0.1223, over 5666301.98 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3598, pruned_loss=0.1115, over 5702741.82 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.375, pruned_loss=0.123, over 5657985.54 frames. ], batch size: 91, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:42:29,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:40,778 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:43,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:54,618 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:43:09,511 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:43:13,653 INFO [train.py:968] (0/2) Epoch 17, batch 24950, giga_loss[loss=0.3206, simple_loss=0.3824, pruned_loss=0.1294, over 28322.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3742, pruned_loss=0.1215, over 5675272.98 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3597, pruned_loss=0.1114, over 5701545.55 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3754, pruned_loss=0.1222, over 5669634.03 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:43:37,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.311e+02 1.435e+03 1.777e+03 2.337e+03 8.876e+03, threshold=3.553e+03, percent-clipped=5.0
+2023-03-08 22:44:01,572 INFO [train.py:968] (0/2) Epoch 17, batch 25000, giga_loss[loss=0.302, simple_loss=0.3734, pruned_loss=0.1153, over 28700.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3745, pruned_loss=0.1219, over 5668400.16 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3598, pruned_loss=0.1115, over 5706641.86 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3759, pruned_loss=0.1226, over 5657975.58 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:44:04,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7469, 1.9733, 1.5529, 1.8654], device='cuda:0'), covar=tensor([0.2402, 0.2455, 0.2841, 0.2426], device='cuda:0'), in_proj_covar=tensor([0.1423, 0.1035, 0.1265, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 22:44:33,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:44:35,781 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:44:48,305 INFO [train.py:968] (0/2) Epoch 17, batch 25050, giga_loss[loss=0.3416, simple_loss=0.3849, pruned_loss=0.1491, over 26624.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3734, pruned_loss=0.121, over 5669841.93 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3596, pruned_loss=0.1115, over 5702311.00 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3752, pruned_loss=0.122, over 5665193.24 frames. ], batch size: 555, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:45:05,156 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:45:12,059 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.442e+02 1.626e+03 1.995e+03 2.871e+03 6.539e+03, threshold=3.989e+03, percent-clipped=13.0
+2023-03-08 22:45:25,224 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755965.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 22:45:40,399 INFO [train.py:968] (0/2) Epoch 17, batch 25100, giga_loss[loss=0.3418, simple_loss=0.399, pruned_loss=0.1423, over 28559.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3727, pruned_loss=0.1209, over 5682776.65 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3599, pruned_loss=0.1116, over 5704032.87 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3739, pruned_loss=0.1216, over 5677417.36 frames. ], batch size: 336, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:45:59,004 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-756000.pt
+2023-03-08 22:46:03,502 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=756006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:46:12,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:46:25,865 INFO [train.py:968] (0/2) Epoch 17, batch 25150, giga_loss[loss=0.3172, simple_loss=0.3843, pruned_loss=0.1251, over 28568.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3721, pruned_loss=0.1212, over 5679254.73 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3604, pruned_loss=0.1122, over 5703802.06 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3733, pruned_loss=0.1218, over 5673248.96 frames. ], batch size: 307, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:46:44,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4476, 1.6067, 1.5280, 1.3528], device='cuda:0'), covar=tensor([0.2687, 0.2270, 0.1878, 0.2452], device='cuda:0'), in_proj_covar=tensor([0.1895, 0.1828, 0.1752, 0.1892], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 22:46:49,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.902e+03 2.689e+03 4.198e+03 1.584e+04, threshold=5.377e+03, percent-clipped=28.0
+2023-03-08 22:47:13,272 INFO [train.py:968] (0/2) Epoch 17, batch 25200, giga_loss[loss=0.3118, simple_loss=0.3764, pruned_loss=0.1235, over 28637.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3705, pruned_loss=0.1209, over 5690272.55 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3602, pruned_loss=0.1121, over 5704968.52 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3718, pruned_loss=0.1215, over 5684248.84 frames. ], batch size: 336, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:48:04,167 INFO [train.py:968] (0/2) Epoch 17, batch 25250, giga_loss[loss=0.283, simple_loss=0.3519, pruned_loss=0.1071, over 28907.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3688, pruned_loss=0.1202, over 5690457.45 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3605, pruned_loss=0.1123, over 5703788.19 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3697, pruned_loss=0.1206, over 5686801.27 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:48:26,482 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.612e+03 2.096e+03 2.677e+03 6.622e+03, threshold=4.191e+03, percent-clipped=3.0
+2023-03-08 22:48:29,576 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=756158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:48:31,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=756161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:48:48,929 INFO [train.py:968] (0/2) Epoch 17, batch 25300, giga_loss[loss=0.2706, simple_loss=0.3426, pruned_loss=0.09933, over 28811.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3666, pruned_loss=0.119, over 5685516.10 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3606, pruned_loss=0.1126, over 5701272.16 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3674, pruned_loss=0.1193, over 5684254.11 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:48:57,243 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=756187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:49:00,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=756190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:49:19,159 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=756206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:49:43,851 INFO [train.py:968] (0/2) Epoch 17, batch 25350, giga_loss[loss=0.2902, simple_loss=0.3647, pruned_loss=0.1079, over 28974.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3657, pruned_loss=0.1188, over 5676994.28 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3605, pruned_loss=0.1126, over 5702292.80 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3664, pruned_loss=0.1191, over 5674890.73 frames. ], batch size: 136, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:50:09,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.126e+03 1.742e+03 2.429e+03 3.264e+03 1.040e+04, threshold=4.857e+03, percent-clipped=13.0
+2023-03-08 22:50:29,494 INFO [train.py:968] (0/2) Epoch 17, batch 25400, giga_loss[loss=0.2604, simple_loss=0.347, pruned_loss=0.08692, over 29016.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3662, pruned_loss=0.118, over 5680819.37 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3608, pruned_loss=0.1129, over 5696550.71 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3667, pruned_loss=0.1181, over 5683728.80 frames. ], batch size: 155, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:51:08,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5331, 2.3060, 1.6819, 0.7578], device='cuda:0'), covar=tensor([0.4522, 0.2549, 0.3604, 0.4969], device='cuda:0'), in_proj_covar=tensor([0.1681, 0.1592, 0.1558, 0.1370], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-08 22:51:17,191 INFO [train.py:968] (0/2) Epoch 17, batch 25450, giga_loss[loss=0.2588, simple_loss=0.349, pruned_loss=0.08435, over 28970.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3671, pruned_loss=0.118, over 5676058.01 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3611, pruned_loss=0.1131, over 5697891.08 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3673, pruned_loss=0.118, over 5676723.24 frames. ], batch size: 164, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:51:26,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756340.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 22:51:30,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6109, 4.4222, 4.2136, 2.1313], device='cuda:0'), covar=tensor([0.0586, 0.0731, 0.0770, 0.1951], device='cuda:0'), in_proj_covar=tensor([0.1190, 0.1102, 0.0942, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 22:51:31,781 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 22:51:38,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.739e+02 1.546e+03 2.048e+03 2.664e+03 1.293e+04, threshold=4.097e+03, percent-clipped=7.0
+2023-03-08 22:52:03,145 INFO [train.py:968] (0/2) Epoch 17, batch 25500, giga_loss[loss=0.3265, simple_loss=0.3853, pruned_loss=0.1339, over 28617.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3681, pruned_loss=0.118, over 5682179.26 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3607, pruned_loss=0.1129, over 5702779.88 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3688, pruned_loss=0.1183, over 5678074.94 frames. ], batch size: 307, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:52:06,259 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:52:28,307 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-08 22:52:48,891 INFO [train.py:968] (0/2) Epoch 17, batch 25550, giga_loss[loss=0.3183, simple_loss=0.3908, pruned_loss=0.1229, over 28943.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3697, pruned_loss=0.1195, over 5668817.37 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3613, pruned_loss=0.1133, over 5686639.72 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.37, pruned_loss=0.1195, over 5677973.82 frames. ], batch size: 164, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:52:54,287 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-08 22:53:13,885 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 22:53:16,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.774e+03 2.345e+03 3.466e+03 1.138e+04, threshold=4.690e+03, percent-clipped=14.0
+2023-03-08 22:53:40,677 INFO [train.py:968] (0/2) Epoch 17, batch 25600, giga_loss[loss=0.2958, simple_loss=0.3734, pruned_loss=0.1091, over 28867.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3726, pruned_loss=0.1222, over 5670662.10 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3612, pruned_loss=0.1133, over 5686363.69 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.373, pruned_loss=0.1223, over 5677729.67 frames. ], batch size: 145, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:53:43,456 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=756483.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 22:53:45,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=756486.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 22:54:20,459 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=756515.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 22:54:27,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=756524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:54:27,617 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=756525.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 22:54:29,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=756527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:54:30,765 INFO [train.py:968] (0/2) Epoch 17, batch 25650, giga_loss[loss=0.3209, simple_loss=0.3796, pruned_loss=0.1311, over 28850.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3744, pruned_loss=0.1253, over 5667479.70 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3614, pruned_loss=0.1135, over 5681453.35 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.375, pruned_loss=0.1256, over 5676303.66 frames. ], batch size: 112, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:54:56,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5223, 2.6868, 1.5810, 1.6696], device='cuda:0'), covar=tensor([0.0729, 0.0341, 0.0664, 0.1007], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0543, 0.0370, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 22:54:59,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 2.029e+03 3.107e+03 4.969e+03 1.445e+04, threshold=6.213e+03, percent-clipped=28.0
+2023-03-08 22:54:59,697 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=756556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:55:05,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:55:20,701 INFO [train.py:968] (0/2) Epoch 17, batch 25700, giga_loss[loss=0.2752, simple_loss=0.3422, pruned_loss=0.1041, over 28919.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.375, pruned_loss=0.1266, over 5671347.04 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3617, pruned_loss=0.1136, over 5686574.21 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3754, pruned_loss=0.1269, over 5673559.43 frames. ], batch size: 112, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:55:24,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:56:07,209 INFO [train.py:968] (0/2) Epoch 17, batch 25750, giga_loss[loss=0.3315, simple_loss=0.3879, pruned_loss=0.1375, over 28604.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3746, pruned_loss=0.1266, over 5673771.93 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5680948.04 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3752, pruned_loss=0.1269, over 5679572.21 frames. ], batch size: 336, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:56:31,370 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-08 22:56:33,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.202e+03 1.756e+03 2.107e+03 2.959e+03 1.000e+04, threshold=4.214e+03, percent-clipped=4.0
+2023-03-08 22:56:56,990 INFO [train.py:968] (0/2) Epoch 17, batch 25800, libri_loss[loss=0.2724, simple_loss=0.3421, pruned_loss=0.1013, over 29590.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.374, pruned_loss=0.1269, over 5663567.97 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3614, pruned_loss=0.1137, over 5682894.02 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3751, pruned_loss=0.1276, over 5665892.17 frames. ], batch size: 77, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:57:19,017 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=756705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:57:21,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=756708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:57:35,917 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=756724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:57:38,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=756727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:57:39,099 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-08 22:57:39,278 INFO [train.py:968] (0/2) Epoch 17, batch 25850, giga_loss[loss=0.288, simple_loss=0.3562, pruned_loss=0.1099, over 28280.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3737, pruned_loss=0.1248, over 5675231.11 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3616, pruned_loss=0.1138, over 5688337.41 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3747, pruned_loss=0.1255, over 5671824.95 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:57:45,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=756737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:57:57,046 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-08 22:58:03,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.582e+02 1.527e+03 2.098e+03 2.468e+03 4.604e+03, threshold=4.196e+03, percent-clipped=4.0
+2023-03-08 22:58:03,314 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=756756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:58:26,971 INFO [train.py:968] (0/2) Epoch 17, batch 25900, giga_loss[loss=0.2875, simple_loss=0.3503, pruned_loss=0.1124, over 28279.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.1229, over 5669188.45 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3609, pruned_loss=0.1133, over 5692874.52 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3733, pruned_loss=0.1242, over 5662087.15 frames. ], batch size: 369, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:59:17,158 INFO [train.py:968] (0/2) Epoch 17, batch 25950, giga_loss[loss=0.3456, simple_loss=0.3818, pruned_loss=0.1547, over 26645.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3695, pruned_loss=0.1219, over 5666952.45 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3606, pruned_loss=0.1131, over 5696002.08 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3713, pruned_loss=0.1233, over 5658282.40 frames. ], batch size: 555, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:59:38,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.141e+03 1.656e+03 2.241e+03 3.123e+03 6.286e+03, threshold=4.482e+03, percent-clipped=5.0
+2023-03-08 23:00:02,708 INFO [train.py:968] (0/2) Epoch 17, batch 26000, giga_loss[loss=0.2647, simple_loss=0.3428, pruned_loss=0.09323, over 28927.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3679, pruned_loss=0.1213, over 5665359.53 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3605, pruned_loss=0.1129, over 5697130.10 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3694, pruned_loss=0.1226, over 5657089.57 frames. ], batch size: 145, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:00:20,938 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 23:00:29,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756900.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 23:00:54,285 INFO [train.py:968] (0/2) Epoch 17, batch 26050, libri_loss[loss=0.3038, simple_loss=0.3711, pruned_loss=0.1182, over 29513.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3696, pruned_loss=0.1228, over 5666050.61 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3605, pruned_loss=0.113, over 5703784.18 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3711, pruned_loss=0.1242, over 5651857.67 frames. ], batch size: 84, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:01:19,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.667e+03 2.111e+03 2.901e+03 1.034e+04, threshold=4.221e+03, percent-clipped=11.0
+2023-03-08 23:01:39,827 INFO [train.py:968] (0/2) Epoch 17, batch 26100, libri_loss[loss=0.2356, simple_loss=0.3045, pruned_loss=0.0833, over 28109.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3727, pruned_loss=0.1243, over 5667990.50 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3611, pruned_loss=0.1136, over 5703636.09 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3738, pruned_loss=0.1252, over 5655896.55 frames. ], batch size: 62, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:02:30,158 INFO [train.py:968] (0/2) Epoch 17, batch 26150, giga_loss[loss=0.319, simple_loss=0.3856, pruned_loss=0.1262, over 27574.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.377, pruned_loss=0.1247, over 5668319.61 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3611, pruned_loss=0.1137, over 5703915.14 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3779, pruned_loss=0.1253, over 5658428.47 frames. ], batch size: 472, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:02:44,542 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=757043.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 23:02:48,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=757046.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 23:02:59,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.888e+02 1.535e+03 1.960e+03 2.494e+03 8.254e+03, threshold=3.920e+03, percent-clipped=5.0
+2023-03-08 23:03:18,977 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=757075.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 23:03:21,698 INFO [train.py:968] (0/2) Epoch 17, batch 26200, giga_loss[loss=0.3156, simple_loss=0.3826, pruned_loss=0.1243, over 28967.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3788, pruned_loss=0.1253, over 5665768.40 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3612, pruned_loss=0.1137, over 5704923.56 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3795, pruned_loss=0.1258, over 5656842.25 frames. ], batch size: 213, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:03:31,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8334, 2.0780, 1.3308, 1.6687], device='cuda:0'), covar=tensor([0.0849, 0.0588, 0.1025, 0.1118], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0441, 0.0507, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 23:03:43,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7165, 3.5568, 3.3730, 2.0046], device='cuda:0'), covar=tensor([0.0650, 0.0808, 0.0746, 0.1725], device='cuda:0'), in_proj_covar=tensor([0.1179, 0.1086, 0.0936, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 23:03:46,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5114, 4.1598, 1.6245, 1.5545], device='cuda:0'), covar=tensor([0.0933, 0.0338, 0.0884, 0.1337], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0542, 0.0369, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 23:04:09,280 INFO [train.py:968] (0/2) Epoch 17, batch 26250, giga_loss[loss=0.2537, simple_loss=0.3367, pruned_loss=0.08536, over 28579.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3795, pruned_loss=0.1262, over 5662253.55 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3613, pruned_loss=0.1139, over 5707858.93 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3805, pruned_loss=0.1268, over 5651219.99 frames. ], batch size: 60, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:04:33,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.084e+02 1.778e+03 2.460e+03 3.355e+03 1.051e+04, threshold=4.921e+03, percent-clipped=20.0
+2023-03-08 23:04:56,673 INFO [train.py:968] (0/2) Epoch 17, batch 26300, giga_loss[loss=0.3672, simple_loss=0.4129, pruned_loss=0.1608, over 28265.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3803, pruned_loss=0.1276, over 5658799.20 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1144, over 5711225.55 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3809, pruned_loss=0.1279, over 5646217.01 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:05:14,702 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-08 23:05:43,665 INFO [train.py:968] (0/2) Epoch 17, batch 26350, giga_loss[loss=0.3268, simple_loss=0.3887, pruned_loss=0.1325, over 28685.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3794, pruned_loss=0.1276, over 5659457.90 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3623, pruned_loss=0.1147, over 5715437.47 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3799, pruned_loss=0.1279, over 5644241.99 frames. ], batch size: 242, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:06:07,100 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-08 23:06:07,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4369, 1.6351, 1.1288, 1.2442], device='cuda:0'), covar=tensor([0.0963, 0.0556, 0.1155, 0.1002], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0443, 0.0509, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 23:06:10,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2363, 2.8091, 1.4346, 1.4247], device='cuda:0'), covar=tensor([0.0975, 0.0351, 0.0851, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0542, 0.0369, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 23:06:15,206 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.534e+02 1.627e+03 1.913e+03 2.769e+03 6.875e+03, threshold=3.827e+03, percent-clipped=6.0
+2023-03-08 23:06:31,416 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-08 23:06:34,828 INFO [train.py:968] (0/2) Epoch 17, batch 26400, giga_loss[loss=0.3237, simple_loss=0.3938, pruned_loss=0.1268, over 28848.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.378, pruned_loss=0.1271, over 5658248.85 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3624, pruned_loss=0.1147, over 5716043.28 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3784, pruned_loss=0.1274, over 5645492.66 frames. ], batch size: 145, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:07:23,467 INFO [train.py:968] (0/2) Epoch 17, batch 26450, giga_loss[loss=0.2857, simple_loss=0.357, pruned_loss=0.1072, over 28791.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3773, pruned_loss=0.1277, over 5649947.41 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3629, pruned_loss=0.1152, over 5706059.90 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3774, pruned_loss=0.1277, over 5647958.81 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:07:52,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.109e+03 1.787e+03 2.135e+03 3.242e+03 5.654e+03, threshold=4.270e+03, percent-clipped=12.0
+2023-03-08 23:08:06,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9840, 1.3763, 1.3432, 1.1904], device='cuda:0'), covar=tensor([0.1715, 0.1117, 0.2111, 0.1495], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0740, 0.0699, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 23:08:15,259 INFO [train.py:968] (0/2) Epoch 17, batch 26500, giga_loss[loss=0.3585, simple_loss=0.4057, pruned_loss=0.1557, over 28605.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3769, pruned_loss=0.1283, over 5645792.99 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3629, pruned_loss=0.1153, over 5711022.52 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3772, pruned_loss=0.1285, over 5638672.46 frames. ], batch size: 242, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:08:54,317 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=757422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:08:59,446 INFO [train.py:968] (0/2) Epoch 17, batch 26550, giga_loss[loss=0.3167, simple_loss=0.381, pruned_loss=0.1262, over 28713.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1272, over 5642780.15 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.363, pruned_loss=0.1154, over 5704961.43 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.376, pruned_loss=0.1275, over 5641740.22 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:09:14,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=757446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:09:21,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.741e+03 2.288e+03 3.017e+03 6.240e+03, threshold=4.577e+03, percent-clipped=8.0
+2023-03-08 23:09:44,795 INFO [train.py:968] (0/2) Epoch 17, batch 26600, giga_loss[loss=0.3009, simple_loss=0.3552, pruned_loss=0.1233, over 28968.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3739, pruned_loss=0.1264, over 5663662.75 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3629, pruned_loss=0.1152, over 5710413.39 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3747, pruned_loss=0.127, over 5656697.55 frames. ], batch size: 106, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:09:58,584 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=757495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:10:26,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4737, 1.3454, 4.4105, 3.4658], device='cuda:0'), covar=tensor([0.1583, 0.2619, 0.0382, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0727, 0.0627, 0.0925, 0.0853], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 23:10:31,932 INFO [train.py:968] (0/2) Epoch 17, batch 26650, giga_loss[loss=0.2763, simple_loss=0.3468, pruned_loss=0.1029, over 28909.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3719, pruned_loss=0.1249, over 5673555.92 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3625, pruned_loss=0.115, over 5714109.82 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3733, pruned_loss=0.126, over 5663299.77 frames. ], batch size: 112, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:10:50,421 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-08 23:10:58,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.969e+02 1.625e+03 2.054e+03 2.822e+03 8.851e+03, threshold=4.107e+03, percent-clipped=6.0
+2023-03-08 23:11:17,242 INFO [train.py:968] (0/2) Epoch 17, batch 26700, libri_loss[loss=0.294, simple_loss=0.3647, pruned_loss=0.1117, over 29102.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3714, pruned_loss=0.1238, over 5671564.64 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.362, pruned_loss=0.1148, over 5709338.63 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3732, pruned_loss=0.1251, over 5665671.00 frames. ], batch size: 101, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:12:05,391 INFO [train.py:968] (0/2) Epoch 17, batch 26750, giga_loss[loss=0.2727, simple_loss=0.3461, pruned_loss=0.09968, over 28910.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3727, pruned_loss=0.124, over 5671501.00 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3616, pruned_loss=0.1146, over 5713629.67 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3749, pruned_loss=0.1255, over 5661445.88 frames. ], batch size: 106, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:12:08,117 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-08 23:12:29,223 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3199, 1.9122, 1.5408, 1.4997], device='cuda:0'), covar=tensor([0.0686, 0.0366, 0.0302, 0.0724], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 23:12:38,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.570e+03 2.043e+03 2.820e+03 1.480e+04, threshold=4.086e+03, percent-clipped=8.0
+2023-03-08 23:12:55,927 INFO [train.py:968] (0/2) Epoch 17, batch 26800, libri_loss[loss=0.2692, simple_loss=0.3321, pruned_loss=0.1031, over 29359.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3714, pruned_loss=0.1234, over 5667041.55 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1144, over 5716866.13 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3736, pruned_loss=0.1249, over 5655077.48 frames. ], batch size: 71, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:13:33,649 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 23:13:39,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0823, 3.8620, 3.6224, 1.8315], device='cuda:0'), covar=tensor([0.0745, 0.0927, 0.0933, 0.2245], device='cuda:0'), in_proj_covar=tensor([0.1192, 0.1099, 0.0943, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 23:13:40,187 INFO [train.py:968] (0/2) Epoch 17, batch 26850, giga_loss[loss=0.2978, simple_loss=0.3786, pruned_loss=0.1085, over 28771.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3733, pruned_loss=0.1229, over 5679136.99 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1143, over 5721479.87 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3754, pruned_loss=0.1244, over 5664313.10 frames. ], batch size: 243, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:13:58,328 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=757748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:14:09,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.879e+02 1.548e+03 2.027e+03 2.619e+03 5.641e+03, threshold=4.053e+03, percent-clipped=8.0
+2023-03-08 23:14:26,160 INFO [train.py:968] (0/2) Epoch 17, batch 26900, giga_loss[loss=0.2999, simple_loss=0.3784, pruned_loss=0.1107, over 28864.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3747, pruned_loss=0.1219, over 5673616.35 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1143, over 5712608.56 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3767, pruned_loss=0.1233, over 5669448.36 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:14:45,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=757797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:14:51,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3415, 1.6816, 1.5289, 1.2688], device='cuda:0'), covar=tensor([0.3185, 0.2278, 0.1941, 0.2559], device='cuda:0'), in_proj_covar=tensor([0.1883, 0.1816, 0.1746, 0.1886], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-08 23:14:57,844 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-08 23:15:03,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7426, 2.0454, 1.6034, 2.1214], device='cuda:0'), covar=tensor([0.2570, 0.2603, 0.3086, 0.2296], device='cuda:0'), in_proj_covar=tensor([0.1427, 0.1038, 0.1267, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 23:15:06,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=757821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:15:13,480 INFO [train.py:968] (0/2) Epoch 17, batch 26950, giga_loss[loss=0.3774, simple_loss=0.4223, pruned_loss=0.1662, over 27922.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3762, pruned_loss=0.1212, over 5669390.46 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3613, pruned_loss=0.1143, over 5711651.71 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3778, pruned_loss=0.1224, over 5666317.62 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:15:21,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5781, 1.8752, 1.6876, 1.5480], device='cuda:0'), covar=tensor([0.1794, 0.1966, 0.2139, 0.2133], device='cuda:0'), in_proj_covar=tensor([0.0462, 0.0743, 0.0701, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 23:15:37,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.967e+02 1.500e+03 2.148e+03 2.995e+03 6.143e+03, threshold=4.297e+03, percent-clipped=6.0
+2023-03-08 23:15:48,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=757870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:15:58,924 INFO [train.py:968] (0/2) Epoch 17, batch 27000, giga_loss[loss=0.295, simple_loss=0.3583, pruned_loss=0.1159, over 28669.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3802, pruned_loss=0.1247, over 5669213.29 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3615, pruned_loss=0.1145, over 5706604.75 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3818, pruned_loss=0.1258, over 5669638.24 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:15:58,929 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-08 23:16:06,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1017, 1.1591, 3.2470, 2.9118], device='cuda:0'), covar=tensor([0.1602, 0.2613, 0.0523, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0728, 0.0628, 0.0929, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 23:16:07,088 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.208, simple_loss=0.3145, pruned_loss=0.05078, over 944034.00 frames. 
+2023-03-08 23:16:07,089 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-08 23:16:52,733 INFO [train.py:968] (0/2) Epoch 17, batch 27050, libri_loss[loss=0.34, simple_loss=0.3719, pruned_loss=0.154, over 29364.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3824, pruned_loss=0.1281, over 5672499.31 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3618, pruned_loss=0.115, over 5706591.24 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3844, pruned_loss=0.129, over 5670979.09 frames. ], batch size: 67, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:17:03,679 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=757940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:17:06,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=757943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:17:20,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.155e+03 1.807e+03 2.480e+03 3.727e+03 1.058e+04, threshold=4.961e+03, percent-clipped=19.0
+2023-03-08 23:17:30,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=757964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:17:32,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=757967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:17:36,001 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=757972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:17:43,645 INFO [train.py:968] (0/2) Epoch 17, batch 27100, giga_loss[loss=0.3387, simple_loss=0.4057, pruned_loss=0.1358, over 28942.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3817, pruned_loss=0.1281, over 5683323.23 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3618, pruned_loss=0.115, over 5711558.15 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3837, pruned_loss=0.1291, over 5676932.12 frames. ], batch size: 213, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:18:00,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=757996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:18:05,986 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-758000.pt
+2023-03-08 23:18:19,663 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=758013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:18:22,308 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=758016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:18:36,661 INFO [train.py:968] (0/2) Epoch 17, batch 27150, giga_loss[loss=0.3494, simple_loss=0.3823, pruned_loss=0.1583, over 23827.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3803, pruned_loss=0.1273, over 5669117.59 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3615, pruned_loss=0.1149, over 5712356.36 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3822, pruned_loss=0.1283, over 5663260.35 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:18:53,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=758045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:18:57,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6302, 1.4790, 5.0770, 3.5711], device='cuda:0'), covar=tensor([0.1696, 0.2729, 0.0412, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0632, 0.0933, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 23:19:01,763 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-08 23:19:04,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.602e+03 2.225e+03 3.097e+03 8.244e+03, threshold=4.450e+03, percent-clipped=4.0
+2023-03-08 23:19:25,179 INFO [train.py:968] (0/2) Epoch 17, batch 27200, giga_loss[loss=0.3961, simple_loss=0.4313, pruned_loss=0.1804, over 27953.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3782, pruned_loss=0.1244, over 5675859.49 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3614, pruned_loss=0.1148, over 5715103.64 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3802, pruned_loss=0.1254, over 5668316.15 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 23:19:29,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3769, 1.7207, 1.6249, 1.4877], device='cuda:0'), covar=tensor([0.1908, 0.1931, 0.2313, 0.2037], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0742, 0.0699, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 23:20:06,011 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=758123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:20:13,625 INFO [train.py:968] (0/2) Epoch 17, batch 27250, giga_loss[loss=0.3643, simple_loss=0.3934, pruned_loss=0.1676, over 23445.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3788, pruned_loss=0.124, over 5665641.80 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3617, pruned_loss=0.1152, over 5719508.51 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3805, pruned_loss=0.1248, over 5654578.78 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 23:20:13,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3491, 1.5327, 1.4658, 1.4323], device='cuda:0'), covar=tensor([0.1333, 0.1493, 0.1846, 0.1481], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0739, 0.0695, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-08 23:20:38,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.559e+03 2.004e+03 2.835e+03 7.999e+03, threshold=4.007e+03, percent-clipped=2.0
+2023-03-08 23:20:59,311 INFO [train.py:968] (0/2) Epoch 17, batch 27300, giga_loss[loss=0.2877, simple_loss=0.3661, pruned_loss=0.1046, over 28704.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3801, pruned_loss=0.1244, over 5665756.91 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3624, pruned_loss=0.1156, over 5708998.11 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3814, pruned_loss=0.1248, over 5665533.82 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:21:55,297 INFO [train.py:968] (0/2) Epoch 17, batch 27350, giga_loss[loss=0.3035, simple_loss=0.3689, pruned_loss=0.1191, over 28618.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3811, pruned_loss=0.1263, over 5652281.20 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3622, pruned_loss=0.1155, over 5710912.12 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3824, pruned_loss=0.1268, over 5649958.00 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:22:12,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9111, 4.7383, 4.5027, 2.1063], device='cuda:0'), covar=tensor([0.0455, 0.0614, 0.0641, 0.2000], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.1107, 0.0950, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 23:22:20,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.629e+03 2.086e+03 2.742e+03 7.042e+03, threshold=4.173e+03, percent-clipped=10.0
+2023-03-08 23:22:27,469 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=758266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:22:30,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=758269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:22:40,543 INFO [train.py:968] (0/2) Epoch 17, batch 27400, giga_loss[loss=0.3841, simple_loss=0.4107, pruned_loss=0.1788, over 23476.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3799, pruned_loss=0.1259, over 5650850.21 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3621, pruned_loss=0.1154, over 5702508.66 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3814, pruned_loss=0.1266, over 5654673.55 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:22:59,312 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=758298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:23:09,741 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-08 23:23:26,164 INFO [train.py:968] (0/2) Epoch 17, batch 27450, giga_loss[loss=0.2891, simple_loss=0.3641, pruned_loss=0.1071, over 28969.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3778, pruned_loss=0.1253, over 5664989.14 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3622, pruned_loss=0.1155, over 5705164.43 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3796, pruned_loss=0.1262, over 5664171.07 frames. ], batch size: 145, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:23:57,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.906e+02 1.822e+03 2.486e+03 3.926e+03 1.504e+04, threshold=4.973e+03, percent-clipped=22.0
+2023-03-08 23:24:18,525 INFO [train.py:968] (0/2) Epoch 17, batch 27500, giga_loss[loss=0.2446, simple_loss=0.3208, pruned_loss=0.08416, over 28668.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3758, pruned_loss=0.1246, over 5669538.39 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3621, pruned_loss=0.1155, over 5708619.33 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3776, pruned_loss=0.1255, over 5664844.96 frames. ], batch size: 85, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:24:24,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4443, 1.7947, 1.4438, 1.5300], device='cuda:0'), covar=tensor([0.2507, 0.2395, 0.2731, 0.2192], device='cuda:0'), in_proj_covar=tensor([0.1431, 0.1039, 0.1270, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 23:25:11,958 INFO [train.py:968] (0/2) Epoch 17, batch 27550, giga_loss[loss=0.2942, simple_loss=0.3558, pruned_loss=0.1163, over 28940.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3739, pruned_loss=0.1237, over 5666468.14 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3622, pruned_loss=0.1155, over 5709832.23 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3752, pruned_loss=0.1245, over 5661641.03 frames. ], batch size: 213, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:25:42,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.558e+03 1.920e+03 2.668e+03 6.954e+03, threshold=3.840e+03, percent-clipped=5.0
+2023-03-08 23:25:59,198 INFO [train.py:968] (0/2) Epoch 17, batch 27600, giga_loss[loss=0.3395, simple_loss=0.3968, pruned_loss=0.1411, over 28708.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3743, pruned_loss=0.125, over 5667828.63 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3624, pruned_loss=0.1158, over 5712943.04 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3753, pruned_loss=0.1255, over 5660869.13 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:26:35,213 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-08 23:26:45,226 INFO [train.py:968] (0/2) Epoch 17, batch 27650, libri_loss[loss=0.3006, simple_loss=0.3745, pruned_loss=0.1134, over 29402.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3713, pruned_loss=0.1224, over 5667936.52 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3628, pruned_loss=0.1159, over 5716800.10 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.372, pruned_loss=0.1228, over 5657900.29 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:27:13,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.418e+02 1.529e+03 2.024e+03 2.837e+03 5.895e+03, threshold=4.048e+03, percent-clipped=5.0
+2023-03-08 23:27:29,281 INFO [train.py:968] (0/2) Epoch 17, batch 27700, giga_loss[loss=0.2716, simple_loss=0.3426, pruned_loss=0.1004, over 28746.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3682, pruned_loss=0.1186, over 5672159.13 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3625, pruned_loss=0.1156, over 5717999.80 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3692, pruned_loss=0.1193, over 5661627.14 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:28:21,961 INFO [train.py:968] (0/2) Epoch 17, batch 27750, giga_loss[loss=0.3093, simple_loss=0.3826, pruned_loss=0.118, over 28563.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3671, pruned_loss=0.1176, over 5662744.98 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3627, pruned_loss=0.1157, over 5719673.07 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3678, pruned_loss=0.1181, over 5652120.15 frames. ], batch size: 71, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:28:55,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.479e+02 1.482e+03 1.934e+03 2.589e+03 1.174e+04, threshold=3.867e+03, percent-clipped=4.0
+2023-03-08 23:29:15,814 INFO [train.py:968] (0/2) Epoch 17, batch 27800, giga_loss[loss=0.254, simple_loss=0.3226, pruned_loss=0.09268, over 28896.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3658, pruned_loss=0.1173, over 5665665.87 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3628, pruned_loss=0.1156, over 5723367.17 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3663, pruned_loss=0.1178, over 5652802.17 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:30:04,111 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.36 vs. limit=5.0
+2023-03-08 23:30:08,076 INFO [train.py:968] (0/2) Epoch 17, batch 27850, giga_loss[loss=0.2862, simple_loss=0.3558, pruned_loss=0.1083, over 28396.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.364, pruned_loss=0.1175, over 5664257.14 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3631, pruned_loss=0.1159, over 5727153.99 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3641, pruned_loss=0.1177, over 5649556.44 frames. ], batch size: 65, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:30:09,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=758730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:30:39,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.983e+02 1.767e+03 2.262e+03 3.208e+03 1.008e+04, threshold=4.523e+03, percent-clipped=16.0
+2023-03-08 23:30:57,296 INFO [train.py:968] (0/2) Epoch 17, batch 27900, giga_loss[loss=0.3705, simple_loss=0.4195, pruned_loss=0.1608, over 27973.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3668, pruned_loss=0.1194, over 5664055.14 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3638, pruned_loss=0.1164, over 5729235.41 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3664, pruned_loss=0.1191, over 5649496.05 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:31:46,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7462, 1.8853, 2.0209, 1.5272], device='cuda:0'), covar=tensor([0.1854, 0.2342, 0.1352, 0.1693], device='cuda:0'), in_proj_covar=tensor([0.0873, 0.0701, 0.0920, 0.0817], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 23:31:46,910 INFO [train.py:968] (0/2) Epoch 17, batch 27950, giga_loss[loss=0.2804, simple_loss=0.3599, pruned_loss=0.1005, over 28804.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3683, pruned_loss=0.1198, over 5654337.17 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3636, pruned_loss=0.1162, over 5732524.67 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3682, pruned_loss=0.1198, over 5638621.10 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:31:56,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7668, 1.0859, 2.8714, 2.7331], device='cuda:0'), covar=tensor([0.1754, 0.2615, 0.0599, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0729, 0.0627, 0.0927, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 23:32:14,460 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.796e+02 1.693e+03 2.203e+03 2.656e+03 7.492e+03, threshold=4.405e+03, percent-clipped=4.0
+2023-03-08 23:32:33,548 INFO [train.py:968] (0/2) Epoch 17, batch 28000, giga_loss[loss=0.3235, simple_loss=0.3797, pruned_loss=0.1337, over 27573.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3687, pruned_loss=0.1196, over 5655142.35 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3636, pruned_loss=0.116, over 5725992.66 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3687, pruned_loss=0.1198, over 5647319.50 frames. ], batch size: 472, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 23:32:33,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5406, 1.6693, 1.5214, 1.6737], device='cuda:0'), covar=tensor([0.0771, 0.0309, 0.0291, 0.0810], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 23:33:17,045 INFO [train.py:968] (0/2) Epoch 17, batch 28050, giga_loss[loss=0.3167, simple_loss=0.3847, pruned_loss=0.1243, over 28764.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3694, pruned_loss=0.1207, over 5657737.84 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3639, pruned_loss=0.1162, over 5731176.97 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3693, pruned_loss=0.121, over 5644117.57 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 23:33:44,763 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.537e+02 1.683e+03 2.401e+03 3.589e+03 7.015e+03, threshold=4.803e+03, percent-clipped=14.0
+2023-03-08 23:33:59,984 INFO [train.py:968] (0/2) Epoch 17, batch 28100, giga_loss[loss=0.2691, simple_loss=0.3495, pruned_loss=0.09431, over 28964.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3694, pruned_loss=0.1208, over 5660588.24 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.364, pruned_loss=0.1162, over 5733810.07 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3694, pruned_loss=0.1212, over 5645351.25 frames. ], batch size: 174, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:34:20,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3976, 1.3348, 4.0449, 3.3347], device='cuda:0'), covar=tensor([0.1558, 0.2648, 0.0438, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0728, 0.0626, 0.0925, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 23:34:48,884 INFO [train.py:968] (0/2) Epoch 17, batch 28150, giga_loss[loss=0.274, simple_loss=0.355, pruned_loss=0.09649, over 28818.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3718, pruned_loss=0.1223, over 5663124.87 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3649, pruned_loss=0.1168, over 5737386.70 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3711, pruned_loss=0.122, over 5646194.01 frames. ], batch size: 174, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:35:16,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.096e+02 1.709e+03 2.438e+03 3.528e+03 9.562e+03, threshold=4.876e+03, percent-clipped=14.0
+2023-03-08 23:35:32,018 INFO [train.py:968] (0/2) Epoch 17, batch 28200, giga_loss[loss=0.2915, simple_loss=0.355, pruned_loss=0.114, over 29011.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3737, pruned_loss=0.1239, over 5659803.40 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3654, pruned_loss=0.1174, over 5732349.83 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3729, pruned_loss=0.1233, over 5647564.12 frames. ], batch size: 106, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:36:01,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=759105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:36:10,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-08 23:36:22,329 INFO [train.py:968] (0/2) Epoch 17, batch 28250, giga_loss[loss=0.3781, simple_loss=0.4055, pruned_loss=0.1754, over 23165.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3757, pruned_loss=0.126, over 5630535.55 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3659, pruned_loss=0.118, over 5708413.43 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3749, pruned_loss=0.1252, over 5639452.85 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:36:49,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.738e+03 2.294e+03 3.085e+03 8.135e+03, threshold=4.587e+03, percent-clipped=6.0
+2023-03-08 23:37:09,472 INFO [train.py:968] (0/2) Epoch 17, batch 28300, giga_loss[loss=0.3217, simple_loss=0.386, pruned_loss=0.1287, over 28697.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3758, pruned_loss=0.1266, over 5636161.25 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.366, pruned_loss=0.1182, over 5706794.80 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3754, pruned_loss=0.126, over 5642275.51 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:37:58,160 INFO [train.py:968] (0/2) Epoch 17, batch 28350, libri_loss[loss=0.2927, simple_loss=0.3602, pruned_loss=0.1126, over 25844.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3763, pruned_loss=0.1253, over 5643467.19 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3662, pruned_loss=0.1183, over 5711998.33 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3762, pruned_loss=0.1252, over 5641200.43 frames. ], batch size: 136, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:38:18,813 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=759248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:38:21,923 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=759251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:38:33,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.673e+03 2.203e+03 2.821e+03 9.694e+03, threshold=4.405e+03, percent-clipped=5.0
+2023-03-08 23:38:51,210 INFO [train.py:968] (0/2) Epoch 17, batch 28400, giga_loss[loss=0.3648, simple_loss=0.403, pruned_loss=0.1633, over 26630.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3764, pruned_loss=0.1253, over 5647597.36 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3659, pruned_loss=0.1181, over 5713929.30 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3768, pruned_loss=0.1255, over 5643291.79 frames. ], batch size: 555, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:38:53,058 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=759280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:39:33,673 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-08 23:39:35,754 INFO [train.py:968] (0/2) Epoch 17, batch 28450, giga_loss[loss=0.3594, simple_loss=0.412, pruned_loss=0.1534, over 28713.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.376, pruned_loss=0.1259, over 5637525.76 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3658, pruned_loss=0.1179, over 5715168.47 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.377, pruned_loss=0.1265, over 5628707.09 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:40:12,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.189e+03 1.822e+03 2.347e+03 3.109e+03 8.127e+03, threshold=4.694e+03, percent-clipped=10.0
+2023-03-08 23:40:32,140 INFO [train.py:968] (0/2) Epoch 17, batch 28500, giga_loss[loss=0.2863, simple_loss=0.349, pruned_loss=0.1118, over 28566.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3754, pruned_loss=0.126, over 5637486.72 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3655, pruned_loss=0.1178, over 5715375.85 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3766, pruned_loss=0.1268, over 5628854.28 frames. ], batch size: 85, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:41:00,431 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2995, 1.5147, 1.2674, 1.4697], device='cuda:0'), covar=tensor([0.0699, 0.0375, 0.0326, 0.0745], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 23:41:09,845 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-08 23:41:29,458 INFO [train.py:968] (0/2) Epoch 17, batch 28550, giga_loss[loss=0.2865, simple_loss=0.3535, pruned_loss=0.1098, over 28850.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3741, pruned_loss=0.1259, over 5616999.43 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3657, pruned_loss=0.118, over 5705458.33 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.375, pruned_loss=0.1265, over 5616909.55 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:41:59,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.633e+02 1.669e+03 2.037e+03 3.328e+03 7.166e+03, threshold=4.074e+03, percent-clipped=7.0
+2023-03-08 23:42:06,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3811, 1.8665, 1.4067, 1.6018], device='cuda:0'), covar=tensor([0.0740, 0.0283, 0.0304, 0.0791], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:0')
+2023-03-08 23:42:13,691 INFO [train.py:968] (0/2) Epoch 17, batch 28600, giga_loss[loss=0.3207, simple_loss=0.3858, pruned_loss=0.1278, over 28811.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3721, pruned_loss=0.1244, over 5637852.48 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.365, pruned_loss=0.1175, over 5704742.77 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.374, pruned_loss=0.1258, over 5634099.58 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:43:01,738 INFO [train.py:968] (0/2) Epoch 17, batch 28650, giga_loss[loss=0.3353, simple_loss=0.3941, pruned_loss=0.1383, over 28693.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3728, pruned_loss=0.1251, over 5638986.28 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3651, pruned_loss=0.1176, over 5696174.36 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3743, pruned_loss=0.1263, over 5642555.83 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:43:13,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2987, 1.5932, 1.2422, 0.9339], device='cuda:0'), covar=tensor([0.2362, 0.2398, 0.2762, 0.2148], device='cuda:0'), in_proj_covar=tensor([0.1433, 0.1041, 0.1271, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 23:43:34,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.746e+03 2.376e+03 3.318e+03 7.636e+03, threshold=4.753e+03, percent-clipped=15.0
+2023-03-08 23:43:50,649 INFO [train.py:968] (0/2) Epoch 17, batch 28700, giga_loss[loss=0.2732, simple_loss=0.3428, pruned_loss=0.1018, over 28799.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3717, pruned_loss=0.1244, over 5645051.06 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3649, pruned_loss=0.1174, over 5700531.79 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3733, pruned_loss=0.1257, over 5642627.67 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:44:32,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3814, 1.3008, 3.7524, 3.2159], device='cuda:0'), covar=tensor([0.1487, 0.2700, 0.0461, 0.2000], device='cuda:0'), in_proj_covar=tensor([0.0727, 0.0625, 0.0925, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-08 23:44:40,693 INFO [train.py:968] (0/2) Epoch 17, batch 28750, giga_loss[loss=0.3204, simple_loss=0.3819, pruned_loss=0.1294, over 28264.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3717, pruned_loss=0.1244, over 5651884.97 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3647, pruned_loss=0.1174, over 5703414.16 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3733, pruned_loss=0.1255, over 5646482.13 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:45:12,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.304e+02 1.599e+03 2.127e+03 2.952e+03 1.081e+04, threshold=4.255e+03, percent-clipped=11.0
+2023-03-08 23:45:19,298 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-08 23:45:33,830 INFO [train.py:968] (0/2) Epoch 17, batch 28800, giga_loss[loss=0.3072, simple_loss=0.3729, pruned_loss=0.1208, over 28828.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3743, pruned_loss=0.1263, over 5653794.12 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3647, pruned_loss=0.1174, over 5704575.09 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3755, pruned_loss=0.1272, over 5648360.37 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 23:46:14,306 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=759723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:46:19,993 INFO [train.py:968] (0/2) Epoch 17, batch 28850, giga_loss[loss=0.3476, simple_loss=0.3989, pruned_loss=0.1481, over 27549.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3744, pruned_loss=0.1263, over 5658293.59 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3651, pruned_loss=0.1176, over 5701090.41 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3751, pruned_loss=0.127, over 5655938.90 frames. ], batch size: 472, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:46:20,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5882, 1.8055, 1.8183, 1.4042], device='cuda:0'), covar=tensor([0.1762, 0.2417, 0.1409, 0.1700], device='cuda:0'), in_proj_covar=tensor([0.0870, 0.0700, 0.0919, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-08 23:46:49,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.627e+03 2.012e+03 2.836e+03 6.092e+03, threshold=4.023e+03, percent-clipped=7.0
+2023-03-08 23:47:04,442 INFO [train.py:968] (0/2) Epoch 17, batch 28900, giga_loss[loss=0.3182, simple_loss=0.3656, pruned_loss=0.1354, over 23471.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3751, pruned_loss=0.1269, over 5665134.70 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3652, pruned_loss=0.1176, over 5705595.18 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.376, pruned_loss=0.1278, over 5657942.17 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:47:46,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4654, 1.7205, 1.4078, 1.6072], device='cuda:0'), covar=tensor([0.2278, 0.2190, 0.2435, 0.1911], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1044, 0.1274, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-08 23:47:54,143 INFO [train.py:968] (0/2) Epoch 17, batch 28950, giga_loss[loss=0.2845, simple_loss=0.3633, pruned_loss=0.1029, over 29075.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3743, pruned_loss=0.1258, over 5677719.83 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3655, pruned_loss=0.1178, over 5705807.21 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3748, pruned_loss=0.1264, over 5671771.46 frames. ], batch size: 128, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:48:30,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.842e+03 2.470e+03 3.724e+03 6.877e+03, threshold=4.939e+03, percent-clipped=18.0
+2023-03-08 23:48:43,326 INFO [train.py:968] (0/2) Epoch 17, batch 29000, giga_loss[loss=0.2892, simple_loss=0.3667, pruned_loss=0.1058, over 28961.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3755, pruned_loss=0.1265, over 5671926.31 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3657, pruned_loss=0.118, over 5708124.76 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3759, pruned_loss=0.1269, over 5664398.32 frames. ], batch size: 164, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:49:29,504 INFO [train.py:968] (0/2) Epoch 17, batch 29050, giga_loss[loss=0.3444, simple_loss=0.3936, pruned_loss=0.1476, over 28637.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3762, pruned_loss=0.1269, over 5679877.82 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3653, pruned_loss=0.1178, over 5712044.60 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.377, pruned_loss=0.1277, over 5669493.01 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:49:29,717 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=759929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:50:00,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.126e+03 1.674e+03 2.381e+03 3.257e+03 8.881e+03, threshold=4.763e+03, percent-clipped=2.0
+2023-03-08 23:50:13,867 INFO [train.py:968] (0/2) Epoch 17, batch 29100, giga_loss[loss=0.3253, simple_loss=0.3904, pruned_loss=0.1301, over 28829.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3773, pruned_loss=0.1283, over 5672974.65 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3653, pruned_loss=0.1177, over 5709054.22 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3784, pruned_loss=0.1293, over 5665314.42 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:50:32,003 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-760000.pt
+2023-03-08 23:50:51,259 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 23:50:55,699 INFO [train.py:968] (0/2) Epoch 17, batch 29150, giga_loss[loss=0.3506, simple_loss=0.3827, pruned_loss=0.1593, over 23733.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3779, pruned_loss=0.1291, over 5672532.34 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.365, pruned_loss=0.1175, over 5714041.48 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3798, pruned_loss=0.1307, over 5659599.88 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:51:24,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.746e+02 1.674e+03 2.363e+03 3.290e+03 1.381e+04, threshold=4.727e+03, percent-clipped=8.0
+2023-03-08 23:51:36,946 INFO [train.py:968] (0/2) Epoch 17, batch 29200, giga_loss[loss=0.3048, simple_loss=0.3729, pruned_loss=0.1184, over 28896.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.378, pruned_loss=0.1287, over 5673725.32 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3654, pruned_loss=0.118, over 5716715.36 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3798, pruned_loss=0.1301, over 5658927.47 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 8.0
+2023-03-08 23:51:57,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=760098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:52:20,483 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-08 23:52:26,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3798, 1.8201, 1.4027, 1.5245], device='cuda:0'), covar=tensor([0.0690, 0.0372, 0.0326, 0.0747], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 23:52:30,853 INFO [train.py:968] (0/2) Epoch 17, batch 29250, giga_loss[loss=0.2927, simple_loss=0.3641, pruned_loss=0.1106, over 28626.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3776, pruned_loss=0.1275, over 5656496.19 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3654, pruned_loss=0.1179, over 5718434.63 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3791, pruned_loss=0.1288, over 5642879.04 frames. ], batch size: 78, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:52:52,235 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 23:53:07,493 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.647e+03 2.084e+03 2.699e+03 5.506e+03, threshold=4.168e+03, percent-clipped=4.0
+2023-03-08 23:53:15,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2990, 3.6743, 1.4906, 1.4361], device='cuda:0'), covar=tensor([0.0982, 0.0337, 0.0884, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0544, 0.0370, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-08 23:53:18,913 INFO [train.py:968] (0/2) Epoch 17, batch 29300, giga_loss[loss=0.302, simple_loss=0.3686, pruned_loss=0.1177, over 28920.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3772, pruned_loss=0.127, over 5657856.70 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3648, pruned_loss=0.1176, over 5720991.51 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3792, pruned_loss=0.1285, over 5643663.96 frames. ], batch size: 186, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:53:45,955 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=760207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:54:08,186 INFO [train.py:968] (0/2) Epoch 17, batch 29350, giga_loss[loss=0.3704, simple_loss=0.4134, pruned_loss=0.1637, over 27845.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3766, pruned_loss=0.1264, over 5665219.62 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3651, pruned_loss=0.1178, over 5723561.74 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3781, pruned_loss=0.1275, over 5650942.31 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:54:17,139 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=760241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:54:20,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=760244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:54:37,205 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.135e+02 1.617e+03 2.046e+03 2.881e+03 9.248e+03, threshold=4.091e+03, percent-clipped=15.0
+2023-03-08 23:54:46,269 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=760273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:54:49,714 INFO [train.py:968] (0/2) Epoch 17, batch 29400, giga_loss[loss=0.3092, simple_loss=0.3694, pruned_loss=0.1245, over 28499.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3742, pruned_loss=0.1248, over 5668560.11 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3647, pruned_loss=0.1177, over 5725981.53 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.376, pruned_loss=0.126, over 5653950.67 frames. ], batch size: 78, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:55:13,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=760304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:55:38,362 INFO [train.py:968] (0/2) Epoch 17, batch 29450, giga_loss[loss=0.285, simple_loss=0.3662, pruned_loss=0.1019, over 28873.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3756, pruned_loss=0.1255, over 5650087.09 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1176, over 5710566.40 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3771, pruned_loss=0.1267, over 5649514.36 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 2.0
+2023-03-08 23:56:06,207 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5263, 3.6261, 1.6250, 1.5637], device='cuda:0'), covar=tensor([0.0984, 0.0380, 0.0905, 0.1392], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0547, 0.0372, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-08 23:56:14,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.915e+02 1.616e+03 2.315e+03 3.690e+03 8.746e+03, threshold=4.631e+03, percent-clipped=17.0
+2023-03-08 23:56:26,637 INFO [train.py:968] (0/2) Epoch 17, batch 29500, giga_loss[loss=0.2824, simple_loss=0.3558, pruned_loss=0.1044, over 29048.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.376, pruned_loss=0.1258, over 5660129.60 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3649, pruned_loss=0.1175, over 5717643.95 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3776, pruned_loss=0.1272, over 5651305.30 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 2.0
+2023-03-08 23:57:08,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=760421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:57:15,399 INFO [train.py:968] (0/2) Epoch 17, batch 29550, giga_loss[loss=0.2933, simple_loss=0.3591, pruned_loss=0.1138, over 28835.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3754, pruned_loss=0.1264, over 5643688.01 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.365, pruned_loss=0.1176, over 5699426.10 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3767, pruned_loss=0.1275, over 5652090.47 frames. ], batch size: 243, lr: 1.87e-03, grad_scale: 2.0
+2023-03-08 23:57:28,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8857, 3.7303, 3.5224, 1.8889], device='cuda:0'), covar=tensor([0.0667, 0.0757, 0.0702, 0.2115], device='cuda:0'), in_proj_covar=tensor([0.1199, 0.1105, 0.0951, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-08 23:57:29,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3538, 1.4482, 1.2299, 1.5673], device='cuda:0'), covar=tensor([0.0679, 0.0390, 0.0328, 0.0740], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0092, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-08 23:57:30,971 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=760447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:57:33,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=760450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:57:48,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.275e+02 1.685e+03 1.990e+03 2.860e+03 8.190e+03, threshold=3.979e+03, percent-clipped=11.0
+2023-03-08 23:58:01,306 INFO [train.py:968] (0/2) Epoch 17, batch 29600, libri_loss[loss=0.3472, simple_loss=0.407, pruned_loss=0.1437, over 19213.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3766, pruned_loss=0.127, over 5649505.82 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3654, pruned_loss=0.1176, over 5693900.41 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3776, pruned_loss=0.128, over 5661111.30 frames. ], batch size: 187, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:58:01,570 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=760479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:58:49,877 INFO [train.py:968] (0/2) Epoch 17, batch 29650, giga_loss[loss=0.3286, simple_loss=0.3824, pruned_loss=0.1374, over 28781.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3783, pruned_loss=0.1283, over 5650714.86 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3653, pruned_loss=0.1176, over 5695825.97 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3792, pruned_loss=0.1293, over 5657662.79 frames. ], batch size: 99, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:59:28,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.693e+02 1.679e+03 2.169e+03 3.342e+03 6.844e+03, threshold=4.338e+03, percent-clipped=18.0
+2023-03-08 23:59:33,935 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 23:59:39,814 INFO [train.py:968] (0/2) Epoch 17, batch 29700, giga_loss[loss=0.3252, simple_loss=0.3818, pruned_loss=0.1343, over 27926.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3793, pruned_loss=0.1296, over 5645974.87 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3653, pruned_loss=0.1176, over 5700042.28 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3803, pruned_loss=0.1305, over 5646766.95 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 2.0
+2023-03-08 23:59:43,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=760582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:59:56,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0065, 1.2071, 5.3168, 3.9194], device='cuda:0'), covar=tensor([0.1484, 0.2862, 0.0438, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0630, 0.0936, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:00:16,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5444, 3.3759, 3.1693, 1.7797], device='cuda:0'), covar=tensor([0.0802, 0.0918, 0.0827, 0.2388], device='cuda:0'), in_proj_covar=tensor([0.1206, 0.1113, 0.0957, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 00:00:26,653 INFO [train.py:968] (0/2) Epoch 17, batch 29750, giga_loss[loss=0.3138, simple_loss=0.3826, pruned_loss=0.1225, over 29005.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3787, pruned_loss=0.1291, over 5644880.10 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3648, pruned_loss=0.1174, over 5703260.18 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3802, pruned_loss=0.1303, over 5641506.43 frames. ], batch size: 128, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:00:38,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-09 00:00:59,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.733e+03 2.186e+03 3.064e+03 6.384e+03, threshold=4.372e+03, percent-clipped=6.0
+2023-03-09 00:01:11,456 INFO [train.py:968] (0/2) Epoch 17, batch 29800, giga_loss[loss=0.2735, simple_loss=0.3533, pruned_loss=0.09688, over 28824.00 frames. ], tot_loss[loss=0.318, simple_loss=0.379, pruned_loss=0.1285, over 5660061.04 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.365, pruned_loss=0.1173, over 5710842.90 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3807, pruned_loss=0.13, over 5648303.45 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:01:16,198 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-09 00:01:35,619 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-09 00:01:53,339 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=760725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:01:55,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=760728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:01:55,623 INFO [train.py:968] (0/2) Epoch 17, batch 29850, giga_loss[loss=0.332, simple_loss=0.3913, pruned_loss=0.1363, over 28001.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3792, pruned_loss=0.1284, over 5663490.02 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3659, pruned_loss=0.1181, over 5712235.25 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3804, pruned_loss=0.1293, over 5650656.17 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:02:21,544 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=760757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:02:28,711 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0459, 2.0932, 1.4831, 1.7952], device='cuda:0'), covar=tensor([0.0848, 0.0642, 0.0981, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0447, 0.0514, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:02:29,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.628e+03 1.927e+03 2.498e+03 6.612e+03, threshold=3.854e+03, percent-clipped=3.0
+2023-03-09 00:02:40,827 INFO [train.py:968] (0/2) Epoch 17, batch 29900, giga_loss[loss=0.2996, simple_loss=0.3741, pruned_loss=0.1126, over 28918.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3779, pruned_loss=0.1273, over 5664197.88 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3663, pruned_loss=0.1183, over 5710010.62 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3789, pruned_loss=0.1282, over 5653853.68 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:02:58,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=760796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:03:06,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6248, 1.6701, 1.2715, 1.1797], device='cuda:0'), covar=tensor([0.0871, 0.0554, 0.1005, 0.1085], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0447, 0.0514, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:03:28,290 INFO [train.py:968] (0/2) Epoch 17, batch 29950, giga_loss[loss=0.2856, simple_loss=0.3513, pruned_loss=0.1099, over 28790.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3769, pruned_loss=0.1272, over 5669071.79 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.366, pruned_loss=0.1182, over 5711813.10 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3781, pruned_loss=0.1282, over 5658487.57 frames. ], batch size: 262, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:03:53,349 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=760859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:03:59,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.816e+03 2.383e+03 2.913e+03 7.983e+03, threshold=4.766e+03, percent-clipped=11.0
+2023-03-09 00:04:07,921 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=760875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:04:12,139 INFO [train.py:968] (0/2) Epoch 17, batch 30000, giga_loss[loss=0.2872, simple_loss=0.3517, pruned_loss=0.1114, over 28859.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3729, pruned_loss=0.1245, over 5676621.95 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3665, pruned_loss=0.1184, over 5715381.71 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3737, pruned_loss=0.1253, over 5664279.33 frames. ], batch size: 243, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:04:12,143 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 00:04:17,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2050, 1.5963, 1.4891, 1.0679], device='cuda:0'), covar=tensor([0.1949, 0.2856, 0.1630, 0.2010], device='cuda:0'), in_proj_covar=tensor([0.0872, 0.0700, 0.0919, 0.0817], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 00:04:20,411 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2087, simple_loss=0.3168, pruned_loss=0.05035, over 944034.00 frames. 
+2023-03-09 00:04:20,411 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 00:04:36,463 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=760892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:05:08,679 INFO [train.py:968] (0/2) Epoch 17, batch 30050, libri_loss[loss=0.3459, simple_loss=0.3963, pruned_loss=0.1478, over 29641.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3699, pruned_loss=0.1239, over 5663271.06 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3666, pruned_loss=0.1184, over 5719216.23 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3705, pruned_loss=0.1246, over 5648888.48 frames. ], batch size: 88, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:05:18,573 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=760939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:05:20,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=760942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:05:43,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.699e+03 2.309e+03 3.513e+03 9.242e+03, threshold=4.617e+03, percent-clipped=6.0
+2023-03-09 00:05:49,108 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=760971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:05:49,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5450, 4.3853, 4.1469, 2.0782], device='cuda:0'), covar=tensor([0.0605, 0.0750, 0.0798, 0.1952], device='cuda:0'), in_proj_covar=tensor([0.1199, 0.1108, 0.0951, 0.0712], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 00:05:54,630 INFO [train.py:968] (0/2) Epoch 17, batch 30100, libri_loss[loss=0.2779, simple_loss=0.3449, pruned_loss=0.1055, over 29555.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3671, pruned_loss=0.1224, over 5671343.27 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3663, pruned_loss=0.1184, over 5720161.97 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.368, pruned_loss=0.1232, over 5656902.15 frames. ], batch size: 79, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:06:46,567 INFO [train.py:968] (0/2) Epoch 17, batch 30150, libri_loss[loss=0.3204, simple_loss=0.3809, pruned_loss=0.13, over 27306.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3672, pruned_loss=0.1228, over 5641149.14 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3667, pruned_loss=0.1187, over 5709236.73 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3674, pruned_loss=0.1232, over 5639312.00 frames. ], batch size: 115, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:07:01,309 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2218, 1.3037, 1.1676, 0.9241], device='cuda:0'), covar=tensor([0.0904, 0.0498, 0.1006, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0448, 0.0514, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:07:23,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.506e+03 1.993e+03 3.023e+03 7.870e+03, threshold=3.985e+03, percent-clipped=4.0
+2023-03-09 00:07:37,238 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:07:39,990 INFO [train.py:968] (0/2) Epoch 17, batch 30200, libri_loss[loss=0.3426, simple_loss=0.3921, pruned_loss=0.1466, over 28697.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3664, pruned_loss=0.1203, over 5639017.72 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3669, pruned_loss=0.1188, over 5709387.55 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3665, pruned_loss=0.1205, over 5637076.24 frames. ], batch size: 106, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:07:59,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5436, 1.7863, 1.4095, 1.9857], device='cuda:0'), covar=tensor([0.2792, 0.2694, 0.3048, 0.2326], device='cuda:0'), in_proj_covar=tensor([0.1435, 0.1043, 0.1276, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 00:08:35,190 INFO [train.py:968] (0/2) Epoch 17, batch 30250, giga_loss[loss=0.275, simple_loss=0.3463, pruned_loss=0.1018, over 28944.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3634, pruned_loss=0.1165, over 5634618.44 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3666, pruned_loss=0.1189, over 5702684.61 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3637, pruned_loss=0.1166, over 5637063.13 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:09:00,409 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2144, 1.8062, 1.3475, 0.3956], device='cuda:0'), covar=tensor([0.3550, 0.2239, 0.3466, 0.4948], device='cuda:0'), in_proj_covar=tensor([0.1684, 0.1601, 0.1562, 0.1375], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 00:09:04,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:09:11,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.654e+02 1.535e+03 2.034e+03 3.154e+03 7.258e+03, threshold=4.068e+03, percent-clipped=11.0
+2023-03-09 00:09:22,611 INFO [train.py:968] (0/2) Epoch 17, batch 30300, giga_loss[loss=0.2548, simple_loss=0.3357, pruned_loss=0.08699, over 28619.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3611, pruned_loss=0.1131, over 5652877.24 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3667, pruned_loss=0.119, over 5707929.90 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3612, pruned_loss=0.113, over 5648624.73 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:10:16,150 INFO [train.py:968] (0/2) Epoch 17, batch 30350, giga_loss[loss=0.2525, simple_loss=0.3335, pruned_loss=0.08572, over 28286.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3579, pruned_loss=0.1097, over 5653285.75 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3666, pruned_loss=0.119, over 5709610.23 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3579, pruned_loss=0.1095, over 5647795.32 frames. ], batch size: 369, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:10:23,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:10:40,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:10:55,991 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.203e+02 1.384e+03 1.766e+03 2.602e+03 6.642e+03, threshold=3.533e+03, percent-clipped=8.0
+2023-03-09 00:10:56,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:11:06,250 INFO [train.py:968] (0/2) Epoch 17, batch 30400, giga_loss[loss=0.265, simple_loss=0.3475, pruned_loss=0.09123, over 28943.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3558, pruned_loss=0.1071, over 5649444.50 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3668, pruned_loss=0.1192, over 5700728.90 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3555, pruned_loss=0.1065, over 5652677.32 frames. ], batch size: 213, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:11:12,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0967, 1.0299, 3.6923, 3.0238], device='cuda:0'), covar=tensor([0.1738, 0.2950, 0.0477, 0.0918], device='cuda:0'), in_proj_covar=tensor([0.0731, 0.0629, 0.0932, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:11:57,455 INFO [train.py:968] (0/2) Epoch 17, batch 30450, libri_loss[loss=0.2824, simple_loss=0.3484, pruned_loss=0.1083, over 25559.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3546, pruned_loss=0.1043, over 5662502.48 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3655, pruned_loss=0.1186, over 5702618.63 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.355, pruned_loss=0.1039, over 5661936.72 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:12:25,414 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:12:42,375 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.431e+02 1.419e+03 1.903e+03 2.558e+03 7.101e+03, threshold=3.806e+03, percent-clipped=9.0
+2023-03-09 00:12:52,885 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:12:53,881 INFO [train.py:968] (0/2) Epoch 17, batch 30500, giga_loss[loss=0.2433, simple_loss=0.3281, pruned_loss=0.07927, over 28936.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3546, pruned_loss=0.1043, over 5650698.31 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3654, pruned_loss=0.1186, over 5692715.54 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.355, pruned_loss=0.104, over 5658812.43 frames. ], batch size: 106, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:12:55,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:09,872 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:11,914 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:25,310 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:25,944 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:31,091 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:43,272 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:47,219 INFO [train.py:968] (0/2) Epoch 17, batch 30550, giga_loss[loss=0.2472, simple_loss=0.3227, pruned_loss=0.08589, over 27612.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3521, pruned_loss=0.1023, over 5659304.85 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3651, pruned_loss=0.1185, over 5695673.81 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3526, pruned_loss=0.102, over 5662606.40 frames. ], batch size: 472, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:13:57,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:14:06,916 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:14:20,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.503e+02 1.432e+03 1.967e+03 2.669e+03 1.363e+04, threshold=3.933e+03, percent-clipped=6.0
+2023-03-09 00:14:30,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.30 vs. limit=5.0
+2023-03-09 00:14:30,566 INFO [train.py:968] (0/2) Epoch 17, batch 30600, giga_loss[loss=0.2701, simple_loss=0.3212, pruned_loss=0.1095, over 24101.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3493, pruned_loss=0.1014, over 5666938.06 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3637, pruned_loss=0.1182, over 5704937.35 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3503, pruned_loss=0.1005, over 5659570.57 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:14:36,469 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 00:14:39,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9090, 1.1000, 0.9284, 0.2243], device='cuda:0'), covar=tensor([0.2769, 0.2301, 0.2715, 0.4738], device='cuda:0'), in_proj_covar=tensor([0.1683, 0.1596, 0.1563, 0.1378], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 00:15:14,224 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4147, 4.0206, 1.5633, 1.5160], device='cuda:0'), covar=tensor([0.0952, 0.0313, 0.0943, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0542, 0.0369, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 00:15:16,287 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761526.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:15:19,151 INFO [train.py:968] (0/2) Epoch 17, batch 30650, giga_loss[loss=0.2863, simple_loss=0.3479, pruned_loss=0.1124, over 26828.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3485, pruned_loss=0.1012, over 5664396.69 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3633, pruned_loss=0.1181, over 5707823.82 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.349, pruned_loss=0.09981, over 5653570.62 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:15:24,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:15:44,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:15:55,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.338e+02 1.377e+03 1.809e+03 2.491e+03 6.429e+03, threshold=3.618e+03, percent-clipped=3.0
+2023-03-09 00:16:08,870 INFO [train.py:968] (0/2) Epoch 17, batch 30700, giga_loss[loss=0.2606, simple_loss=0.3219, pruned_loss=0.09963, over 23785.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3478, pruned_loss=0.1001, over 5657959.60 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3632, pruned_loss=0.1182, over 5700058.04 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3482, pruned_loss=0.09883, over 5656415.86 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:16:23,736 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:16:26,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:16:54,169 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:16:56,538 INFO [train.py:968] (0/2) Epoch 17, batch 30750, giga_loss[loss=0.3315, simple_loss=0.3884, pruned_loss=0.1373, over 28932.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3455, pruned_loss=0.09793, over 5660515.18 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3629, pruned_loss=0.118, over 5702016.75 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3458, pruned_loss=0.09664, over 5656890.93 frames. ], batch size: 285, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:17:37,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.588e+02 1.373e+03 1.858e+03 2.801e+03 8.491e+03, threshold=3.716e+03, percent-clipped=10.0
+2023-03-09 00:17:38,009 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:17:47,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:17:48,220 INFO [train.py:968] (0/2) Epoch 17, batch 30800, giga_loss[loss=0.2335, simple_loss=0.3174, pruned_loss=0.07486, over 28269.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3431, pruned_loss=0.09607, over 5655714.29 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.363, pruned_loss=0.1183, over 5694079.50 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3428, pruned_loss=0.09439, over 5658947.16 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:17:49,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:18:20,336 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:18:39,308 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:18:39,736 INFO [train.py:968] (0/2) Epoch 17, batch 30850, giga_loss[loss=0.2539, simple_loss=0.3382, pruned_loss=0.08482, over 28722.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3394, pruned_loss=0.09389, over 5664307.86 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3631, pruned_loss=0.1184, over 5694514.97 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3387, pruned_loss=0.09197, over 5666141.42 frames. ], batch size: 262, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:18:56,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3863, 3.6169, 1.5448, 1.4552], device='cuda:0'), covar=tensor([0.1005, 0.0340, 0.0974, 0.1416], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0543, 0.0370, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 00:19:11,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.103e+02 1.459e+03 1.913e+03 2.551e+03 6.129e+03, threshold=3.825e+03, percent-clipped=8.0
+2023-03-09 00:19:23,393 INFO [train.py:968] (0/2) Epoch 17, batch 30900, giga_loss[loss=0.2637, simple_loss=0.3383, pruned_loss=0.09459, over 27964.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3388, pruned_loss=0.09425, over 5640995.51 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3629, pruned_loss=0.1185, over 5671269.29 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3375, pruned_loss=0.09173, over 5661038.91 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:20:17,325 INFO [train.py:968] (0/2) Epoch 17, batch 30950, giga_loss[loss=0.2615, simple_loss=0.3363, pruned_loss=0.09331, over 27906.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3382, pruned_loss=0.09453, over 5633739.77 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.362, pruned_loss=0.1181, over 5676004.09 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3375, pruned_loss=0.09245, over 5644732.57 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:20:18,822 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761830.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:20:59,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.633e+02 1.391e+03 1.812e+03 2.582e+03 7.305e+03, threshold=3.624e+03, percent-clipped=11.0
+2023-03-09 00:21:03,272 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:21:06,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:21:14,352 INFO [train.py:968] (0/2) Epoch 17, batch 31000, giga_loss[loss=0.2775, simple_loss=0.3522, pruned_loss=0.1014, over 27665.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3404, pruned_loss=0.09513, over 5628322.87 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.362, pruned_loss=0.1182, over 5669440.55 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3396, pruned_loss=0.09305, over 5642691.77 frames. ], batch size: 472, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:21:42,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761901.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:21:44,169 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:22:15,245 INFO [train.py:968] (0/2) Epoch 17, batch 31050, giga_loss[loss=0.2648, simple_loss=0.3213, pruned_loss=0.1041, over 24179.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3417, pruned_loss=0.09516, over 5620834.92 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3618, pruned_loss=0.1182, over 5670416.32 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.341, pruned_loss=0.09338, over 5630799.65 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:22:15,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:23:05,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.859e+02 1.464e+03 1.858e+03 2.307e+03 5.666e+03, threshold=3.716e+03, percent-clipped=7.0
+2023-03-09 00:23:20,601 INFO [train.py:968] (0/2) Epoch 17, batch 31100, giga_loss[loss=0.3074, simple_loss=0.3667, pruned_loss=0.124, over 27776.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3415, pruned_loss=0.09514, over 5621912.48 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.361, pruned_loss=0.1178, over 5673727.86 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3413, pruned_loss=0.09361, over 5626031.33 frames. ], batch size: 474, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:23:45,763 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-762000.pt
+2023-03-09 00:23:59,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4040, 1.7529, 1.3409, 1.4545], device='cuda:0'), covar=tensor([0.2954, 0.2669, 0.3261, 0.2343], device='cuda:0'), in_proj_covar=tensor([0.1438, 0.1040, 0.1281, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 00:24:20,323 INFO [train.py:968] (0/2) Epoch 17, batch 31150, giga_loss[loss=0.2287, simple_loss=0.3149, pruned_loss=0.0712, over 28689.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3394, pruned_loss=0.09397, over 5640915.08 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3603, pruned_loss=0.1175, over 5680438.18 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3393, pruned_loss=0.09241, over 5637235.06 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:24:35,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=762042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:24:36,901 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=762044.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:24:41,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=762047.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:25:02,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.773e+02 1.272e+03 1.879e+03 2.679e+03 9.484e+03, threshold=3.759e+03, percent-clipped=15.0
+2023-03-09 00:25:10,001 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=762072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:25:14,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=762075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:25:14,776 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=762076.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:25:17,311 INFO [train.py:968] (0/2) Epoch 17, batch 31200, giga_loss[loss=0.2154, simple_loss=0.3081, pruned_loss=0.06135, over 28890.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3388, pruned_loss=0.09285, over 5633312.90 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3604, pruned_loss=0.1176, over 5677483.87 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3379, pruned_loss=0.09054, over 5630829.43 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:25:51,468 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=762104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:26:22,365 INFO [train.py:968] (0/2) Epoch 17, batch 31250, giga_loss[loss=0.2368, simple_loss=0.3158, pruned_loss=0.07889, over 28894.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3358, pruned_loss=0.08988, over 5639060.63 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3602, pruned_loss=0.1175, over 5679809.27 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3349, pruned_loss=0.08786, over 5634627.35 frames. ], batch size: 112, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:27:05,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.205e+02 1.314e+03 1.891e+03 2.890e+03 5.981e+03, threshold=3.783e+03, percent-clipped=14.0
+2023-03-09 00:27:17,184 INFO [train.py:968] (0/2) Epoch 17, batch 31300, giga_loss[loss=0.2667, simple_loss=0.3468, pruned_loss=0.09335, over 28430.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3336, pruned_loss=0.08972, over 5659352.39 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3602, pruned_loss=0.1176, over 5687358.80 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.332, pruned_loss=0.08688, over 5647508.12 frames. ], batch size: 369, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:27:27,973 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=762185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:27:30,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=762188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:27:43,902 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 00:27:50,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=762205.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:28:02,657 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=762217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:28:16,526 INFO [train.py:968] (0/2) Epoch 17, batch 31350, giga_loss[loss=0.2518, simple_loss=0.3272, pruned_loss=0.08817, over 28862.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3331, pruned_loss=0.08993, over 5670963.00 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3596, pruned_loss=0.1175, over 5690439.67 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3315, pruned_loss=0.08693, over 5657951.98 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:28:20,734 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-09 00:29:05,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.544e+02 1.350e+03 1.682e+03 2.158e+03 6.337e+03, threshold=3.364e+03, percent-clipped=6.0
+2023-03-09 00:29:16,344 INFO [train.py:968] (0/2) Epoch 17, batch 31400, giga_loss[loss=0.254, simple_loss=0.3368, pruned_loss=0.08561, over 29001.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3339, pruned_loss=0.08991, over 5675551.19 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3594, pruned_loss=0.1174, over 5693162.35 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3325, pruned_loss=0.08733, over 5662752.85 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:30:13,193 INFO [train.py:968] (0/2) Epoch 17, batch 31450, giga_loss[loss=0.2449, simple_loss=0.3381, pruned_loss=0.07588, over 28905.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3358, pruned_loss=0.09101, over 5663168.69 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3594, pruned_loss=0.1177, over 5689992.37 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3338, pruned_loss=0.0876, over 5655009.34 frames. ], batch size: 119, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:30:41,882 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=762348.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:30:44,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=762351.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:31:07,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.261e+02 1.479e+03 1.950e+03 3.171e+03 6.601e+03, threshold=3.900e+03, percent-clipped=22.0
+2023-03-09 00:31:18,736 INFO [train.py:968] (0/2) Epoch 17, batch 31500, giga_loss[loss=0.22, simple_loss=0.3046, pruned_loss=0.06773, over 28361.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3356, pruned_loss=0.09055, over 5665739.88 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3589, pruned_loss=0.1175, over 5694690.98 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3338, pruned_loss=0.08738, over 5654117.03 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:31:21,303 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=762380.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:31:36,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2477, 1.5090, 1.2393, 1.4498], device='cuda:0'), covar=tensor([0.0794, 0.0306, 0.0348, 0.0910], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-09 00:32:16,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4211, 1.9565, 1.4284, 0.7327], device='cuda:0'), covar=tensor([0.5809, 0.2727, 0.3846, 0.5999], device='cuda:0'), in_proj_covar=tensor([0.1683, 0.1597, 0.1560, 0.1376], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 00:32:29,620 INFO [train.py:968] (0/2) Epoch 17, batch 31550, giga_loss[loss=0.2766, simple_loss=0.3601, pruned_loss=0.0966, over 28052.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3335, pruned_loss=0.08892, over 5676446.17 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.359, pruned_loss=0.1176, over 5691815.53 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.332, pruned_loss=0.08626, over 5669555.09 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:32:37,572 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 00:33:23,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.471e+02 1.313e+03 2.035e+03 3.088e+03 9.083e+03, threshold=4.070e+03, percent-clipped=17.0
+2023-03-09 00:33:36,287 INFO [train.py:968] (0/2) Epoch 17, batch 31600, libri_loss[loss=0.315, simple_loss=0.3724, pruned_loss=0.1288, over 29299.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3358, pruned_loss=0.0905, over 5677646.05 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3588, pruned_loss=0.1177, over 5698145.49 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3341, pruned_loss=0.08748, over 5665684.39 frames. ], batch size: 94, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:34:39,157 INFO [train.py:968] (0/2) Epoch 17, batch 31650, giga_loss[loss=0.2004, simple_loss=0.2749, pruned_loss=0.06301, over 24432.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3404, pruned_loss=0.0907, over 5664926.67 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3592, pruned_loss=0.1182, over 5688081.16 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3383, pruned_loss=0.08741, over 5664434.82 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:35:28,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.338e+02 1.409e+03 1.871e+03 2.494e+03 6.770e+03, threshold=3.741e+03, percent-clipped=4.0
+2023-03-09 00:35:41,676 INFO [train.py:968] (0/2) Epoch 17, batch 31700, giga_loss[loss=0.2759, simple_loss=0.3615, pruned_loss=0.09517, over 28507.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3412, pruned_loss=0.09008, over 5658521.88 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3585, pruned_loss=0.1179, over 5692941.14 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3398, pruned_loss=0.0871, over 5653444.90 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:35:44,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4249, 1.4931, 1.1456, 1.1595], device='cuda:0'), covar=tensor([0.0860, 0.0487, 0.0996, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0440, 0.0511, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:35:50,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:36:16,217 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:36:38,482 INFO [train.py:968] (0/2) Epoch 17, batch 31750, giga_loss[loss=0.2648, simple_loss=0.353, pruned_loss=0.08825, over 28812.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3414, pruned_loss=0.08919, over 5655287.36 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3587, pruned_loss=0.1182, over 5685577.91 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3395, pruned_loss=0.08573, over 5656792.30 frames. ], batch size: 243, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:37:04,759 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:37:29,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.263e+02 1.489e+03 2.223e+03 3.874e+03 1.064e+04, threshold=4.445e+03, percent-clipped=27.0
+2023-03-09 00:37:39,066 INFO [train.py:968] (0/2) Epoch 17, batch 31800, giga_loss[loss=0.3153, simple_loss=0.3852, pruned_loss=0.1227, over 28499.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3411, pruned_loss=0.08918, over 5671435.31 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3582, pruned_loss=0.118, over 5689296.60 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3397, pruned_loss=0.08594, over 5668552.48 frames. ], batch size: 369, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:38:39,881 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762724.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:38:44,360 INFO [train.py:968] (0/2) Epoch 17, batch 31850, giga_loss[loss=0.2358, simple_loss=0.3194, pruned_loss=0.07612, over 28760.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3402, pruned_loss=0.08975, over 5675168.26 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3581, pruned_loss=0.118, over 5682119.33 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3389, pruned_loss=0.0867, over 5679272.86 frames. ], batch size: 243, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:39:29,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5415, 2.6796, 2.0399, 2.4994], device='cuda:0'), covar=tensor([0.0690, 0.0455, 0.0782, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0439, 0.0509, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:39:48,363 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:39:48,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.431e+02 1.439e+03 1.922e+03 2.422e+03 6.133e+03, threshold=3.844e+03, percent-clipped=3.0
+2023-03-09 00:39:58,606 INFO [train.py:968] (0/2) Epoch 17, batch 31900, giga_loss[loss=0.2626, simple_loss=0.3501, pruned_loss=0.0876, over 29044.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3402, pruned_loss=0.09075, over 5664190.96 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3577, pruned_loss=0.1179, over 5676533.38 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3392, pruned_loss=0.08794, over 5671401.08 frames. ], batch size: 285, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:40:02,467 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 00:41:03,500 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-09 00:41:18,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3343, 1.8279, 1.7163, 1.2694], device='cuda:0'), covar=tensor([0.3420, 0.1853, 0.2031, 0.2729], device='cuda:0'), in_proj_covar=tensor([0.1830, 0.1772, 0.1684, 0.1831], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 00:41:20,455 INFO [train.py:968] (0/2) Epoch 17, batch 31950, giga_loss[loss=0.2454, simple_loss=0.3329, pruned_loss=0.07892, over 28112.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.339, pruned_loss=0.09074, over 5666523.07 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3579, pruned_loss=0.1181, over 5673136.81 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3376, pruned_loss=0.08761, over 5675673.57 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:42:12,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.416e+02 1.336e+03 1.756e+03 2.398e+03 8.304e+03, threshold=3.513e+03, percent-clipped=2.0
+2023-03-09 00:42:24,201 INFO [train.py:968] (0/2) Epoch 17, batch 32000, giga_loss[loss=0.2556, simple_loss=0.3255, pruned_loss=0.09283, over 26886.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3359, pruned_loss=0.08914, over 5665808.91 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3576, pruned_loss=0.118, over 5676303.39 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3344, pruned_loss=0.08585, over 5670142.31 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:43:02,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5707, 2.2919, 1.6814, 0.6230], device='cuda:0'), covar=tensor([0.4136, 0.2493, 0.3313, 0.5105], device='cuda:0'), in_proj_covar=tensor([0.1683, 0.1599, 0.1563, 0.1376], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 00:43:12,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2083, 1.2429, 1.1457, 0.8966], device='cuda:0'), covar=tensor([0.0947, 0.0491, 0.1046, 0.1032], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0440, 0.0509, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:43:27,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6389, 1.7575, 1.3594, 1.3377], device='cuda:0'), covar=tensor([0.0881, 0.0534, 0.0969, 0.1087], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0439, 0.0508, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:43:28,949 INFO [train.py:968] (0/2) Epoch 17, batch 32050, giga_loss[loss=0.2136, simple_loss=0.299, pruned_loss=0.06409, over 28930.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3329, pruned_loss=0.08722, over 5677538.00 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.357, pruned_loss=0.1175, over 5684341.66 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3315, pruned_loss=0.08408, over 5673677.45 frames. ], batch size: 155, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:43:48,713 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:44:06,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=762961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:44:15,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.213e+02 1.455e+03 1.869e+03 2.454e+03 5.457e+03, threshold=3.738e+03, percent-clipped=10.0
+2023-03-09 00:44:25,846 INFO [train.py:968] (0/2) Epoch 17, batch 32100, giga_loss[loss=0.2618, simple_loss=0.3453, pruned_loss=0.0891, over 28888.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3333, pruned_loss=0.08831, over 5689814.04 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3557, pruned_loss=0.1169, over 5693583.70 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3321, pruned_loss=0.0849, over 5677924.72 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:44:32,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=762984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:45:10,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5614, 1.7015, 1.4283, 1.7848], device='cuda:0'), covar=tensor([0.2637, 0.2543, 0.2869, 0.2202], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1041, 0.1279, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 00:45:21,149 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=763024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:45:25,825 INFO [train.py:968] (0/2) Epoch 17, batch 32150, libri_loss[loss=0.2528, simple_loss=0.3151, pruned_loss=0.09531, over 28616.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.337, pruned_loss=0.08972, over 5692074.36 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3553, pruned_loss=0.1166, over 5697006.39 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3362, pruned_loss=0.08674, over 5679585.38 frames. ], batch size: 63, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:45:41,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5760, 4.4258, 4.1528, 1.9674], device='cuda:0'), covar=tensor([0.0452, 0.0619, 0.0748, 0.2238], device='cuda:0'), in_proj_covar=tensor([0.1160, 0.1071, 0.0922, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 00:46:19,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.294e+02 1.305e+03 1.730e+03 2.311e+03 6.829e+03, threshold=3.459e+03, percent-clipped=4.0
+2023-03-09 00:46:29,727 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=763078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:46:30,229 INFO [train.py:968] (0/2) Epoch 17, batch 32200, libri_loss[loss=0.2283, simple_loss=0.3044, pruned_loss=0.07614, over 29568.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3349, pruned_loss=0.0898, over 5695803.50 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3545, pruned_loss=0.1161, over 5702341.36 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3345, pruned_loss=0.08725, over 5680792.34 frames. ], batch size: 79, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:46:49,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9063, 2.0294, 1.4375, 1.7025], device='cuda:0'), covar=tensor([0.0891, 0.0623, 0.1067, 0.1066], device='cuda:0'), in_proj_covar=tensor([0.0376, 0.0436, 0.0506, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:46:53,577 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=763099.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:46:58,651 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:47:04,408 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:47:25,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4363, 2.1466, 1.5278, 0.6405], device='cuda:0'), covar=tensor([0.4828, 0.2798, 0.3993, 0.5478], device='cuda:0'), in_proj_covar=tensor([0.1687, 0.1604, 0.1564, 0.1377], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 00:47:28,385 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:47:29,568 INFO [train.py:968] (0/2) Epoch 17, batch 32250, giga_loss[loss=0.2529, simple_loss=0.3208, pruned_loss=0.09253, over 28629.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3362, pruned_loss=0.09156, over 5689489.47 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3541, pruned_loss=0.1158, over 5698911.47 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3358, pruned_loss=0.08921, over 5680670.62 frames. ], batch size: 85, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:47:32,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:47:39,769 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:47:49,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=763144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:48:07,835 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:48:18,593 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:48:20,922 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:48:21,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.139e+02 1.533e+03 2.027e+03 2.960e+03 7.192e+03, threshold=4.055e+03, percent-clipped=16.0
+2023-03-09 00:48:32,082 INFO [train.py:968] (0/2) Epoch 17, batch 32300, giga_loss[loss=0.235, simple_loss=0.328, pruned_loss=0.07095, over 28134.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3371, pruned_loss=0.09234, over 5690105.85 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.354, pruned_loss=0.1158, over 5703123.16 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3363, pruned_loss=0.08977, over 5678921.12 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:48:43,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4783, 2.2093, 1.5525, 0.6532], device='cuda:0'), covar=tensor([0.4884, 0.2717, 0.4215, 0.5664], device='cuda:0'), in_proj_covar=tensor([0.1685, 0.1602, 0.1563, 0.1376], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 00:49:01,142 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:49:42,864 INFO [train.py:968] (0/2) Epoch 17, batch 32350, giga_loss[loss=0.2768, simple_loss=0.3563, pruned_loss=0.09871, over 28453.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3386, pruned_loss=0.09224, over 5690893.72 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3538, pruned_loss=0.1158, over 5708415.16 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3378, pruned_loss=0.08953, over 5676745.07 frames. ], batch size: 78, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:50:01,899 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763242.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:50:07,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763245.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:50:45,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.456e+02 1.424e+03 1.928e+03 2.772e+03 5.036e+03, threshold=3.856e+03, percent-clipped=6.0
+2023-03-09 00:50:49,137 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763274.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:50:55,411 INFO [train.py:968] (0/2) Epoch 17, batch 32400, libri_loss[loss=0.2821, simple_loss=0.3423, pruned_loss=0.111, over 25968.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3393, pruned_loss=0.09248, over 5683370.16 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3527, pruned_loss=0.1153, over 5710899.91 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3392, pruned_loss=0.09015, over 5669056.23 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:51:09,940 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:51:13,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:51:18,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6351, 1.8103, 1.4945, 1.7333], device='cuda:0'), covar=tensor([0.2626, 0.2720, 0.3102, 0.2379], device='cuda:0'), in_proj_covar=tensor([0.1438, 0.1039, 0.1277, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 00:51:57,196 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:51:58,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=763321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:52:06,699 INFO [train.py:968] (0/2) Epoch 17, batch 32450, giga_loss[loss=0.2241, simple_loss=0.3013, pruned_loss=0.0734, over 28823.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3363, pruned_loss=0.09116, over 5685022.63 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3527, pruned_loss=0.1152, over 5713787.98 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.336, pruned_loss=0.089, over 5670622.05 frames. ], batch size: 119, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:52:11,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3205, 1.1673, 4.0191, 3.2215], device='cuda:0'), covar=tensor([0.1570, 0.2792, 0.0455, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0720, 0.0625, 0.0917, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 00:53:01,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.484e+03 1.848e+03 2.393e+03 5.288e+03, threshold=3.695e+03, percent-clipped=5.0
+2023-03-09 00:53:10,385 INFO [train.py:968] (0/2) Epoch 17, batch 32500, giga_loss[loss=0.2623, simple_loss=0.3216, pruned_loss=0.1015, over 27731.00 frames. ], tot_loss[loss=0.256, simple_loss=0.332, pruned_loss=0.08996, over 5689312.73 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3525, pruned_loss=0.1151, over 5718400.12 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3314, pruned_loss=0.0877, over 5672905.34 frames. ], batch size: 474, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:54:03,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3439, 1.7475, 1.5695, 1.4885], device='cuda:0'), covar=tensor([0.1798, 0.1848, 0.2048, 0.1952], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0723, 0.0681, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 00:54:19,829 INFO [train.py:968] (0/2) Epoch 17, batch 32550, giga_loss[loss=0.2301, simple_loss=0.3041, pruned_loss=0.07807, over 28694.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3278, pruned_loss=0.08773, over 5682287.50 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3524, pruned_loss=0.115, over 5722286.32 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.327, pruned_loss=0.0855, over 5664965.08 frames. ], batch size: 92, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:54:33,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3691, 1.7951, 1.6759, 1.5372], device='cuda:0'), covar=tensor([0.1787, 0.1668, 0.2080, 0.1848], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0723, 0.0681, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 00:54:40,771 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-09 00:54:45,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=763453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:54:59,396 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:55:04,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:55:07,905 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.374e+02 1.310e+03 1.770e+03 2.310e+03 4.267e+03, threshold=3.540e+03, percent-clipped=5.0
+2023-03-09 00:55:15,717 INFO [train.py:968] (0/2) Epoch 17, batch 32600, giga_loss[loss=0.2456, simple_loss=0.3342, pruned_loss=0.07848, over 28865.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3285, pruned_loss=0.08833, over 5687076.51 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3513, pruned_loss=0.1144, over 5726781.47 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.328, pruned_loss=0.0862, over 5667678.39 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:55:35,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:56:14,745 INFO [train.py:968] (0/2) Epoch 17, batch 32650, libri_loss[loss=0.2343, simple_loss=0.3004, pruned_loss=0.08415, over 29591.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3296, pruned_loss=0.0885, over 5693198.55 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3511, pruned_loss=0.1143, over 5730715.41 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3289, pruned_loss=0.08631, over 5673183.08 frames. ], batch size: 74, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:57:08,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.956e+02 1.369e+03 1.602e+03 2.523e+03 5.501e+03, threshold=3.205e+03, percent-clipped=12.0
+2023-03-09 00:57:16,995 INFO [train.py:968] (0/2) Epoch 17, batch 32700, giga_loss[loss=0.2608, simple_loss=0.3387, pruned_loss=0.09143, over 28790.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3266, pruned_loss=0.08584, over 5679210.87 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3513, pruned_loss=0.1143, over 5732424.53 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3255, pruned_loss=0.08353, over 5661140.96 frames. ], batch size: 243, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:57:18,818 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 00:57:41,377 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:57:45,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:58:25,066 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:58:25,802 INFO [train.py:968] (0/2) Epoch 17, batch 32750, giga_loss[loss=0.2406, simple_loss=0.3183, pruned_loss=0.08151, over 28148.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3256, pruned_loss=0.08508, over 5672813.70 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3512, pruned_loss=0.1143, over 5729635.41 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3247, pruned_loss=0.08319, over 5661084.83 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:58:39,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 00:58:49,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 00:59:25,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.953e+02 1.474e+03 2.135e+03 3.054e+03 1.072e+04, threshold=4.270e+03, percent-clipped=23.0
+2023-03-09 00:59:32,020 INFO [train.py:968] (0/2) Epoch 17, batch 32800, giga_loss[loss=0.2313, simple_loss=0.3159, pruned_loss=0.07335, over 28539.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3258, pruned_loss=0.08554, over 5678046.90 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3512, pruned_loss=0.1143, over 5733133.55 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3244, pruned_loss=0.08335, over 5664192.16 frames. ], batch size: 370, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:00:41,557 INFO [train.py:968] (0/2) Epoch 17, batch 32850, giga_loss[loss=0.2318, simple_loss=0.317, pruned_loss=0.07332, over 28630.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3278, pruned_loss=0.08617, over 5681592.48 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3515, pruned_loss=0.1144, over 5725061.08 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3258, pruned_loss=0.08358, over 5675428.99 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:01:21,312 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-09 01:01:35,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.539e+02 1.550e+03 1.864e+03 2.508e+03 8.874e+03, threshold=3.727e+03, percent-clipped=5.0
+2023-03-09 01:01:44,075 INFO [train.py:968] (0/2) Epoch 17, batch 32900, giga_loss[loss=0.2614, simple_loss=0.3333, pruned_loss=0.0947, over 28063.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3283, pruned_loss=0.08721, over 5684073.05 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3511, pruned_loss=0.1144, over 5729729.90 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3266, pruned_loss=0.08455, over 5673826.57 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:02:41,428 INFO [train.py:968] (0/2) Epoch 17, batch 32950, giga_loss[loss=0.2464, simple_loss=0.3322, pruned_loss=0.0803, over 28584.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3284, pruned_loss=0.08794, over 5679947.87 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3501, pruned_loss=0.1138, over 5723489.21 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3272, pruned_loss=0.08554, over 5676641.77 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:03:39,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.605e+02 1.388e+03 1.796e+03 2.514e+03 5.890e+03, threshold=3.593e+03, percent-clipped=8.0
+2023-03-09 01:03:45,688 INFO [train.py:968] (0/2) Epoch 17, batch 33000, giga_loss[loss=0.247, simple_loss=0.3435, pruned_loss=0.07523, over 28964.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3277, pruned_loss=0.08631, over 5670952.43 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3498, pruned_loss=0.1137, over 5725436.75 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3268, pruned_loss=0.08426, over 5666374.05 frames. ], batch size: 213, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:03:45,692 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 01:03:54,127 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.1987, simple_loss=0.2993, pruned_loss=0.04903, over 944034.00 frames. 
+2023-03-09 01:03:54,128 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 01:03:56,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5454, 1.7893, 1.4726, 1.6041], device='cuda:0'), covar=tensor([0.2635, 0.2442, 0.2728, 0.2283], device='cuda:0'), in_proj_covar=tensor([0.1432, 0.1037, 0.1274, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 01:04:49,062 INFO [train.py:968] (0/2) Epoch 17, batch 33050, giga_loss[loss=0.2607, simple_loss=0.3466, pruned_loss=0.0874, over 28577.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3302, pruned_loss=0.0866, over 5670103.36 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3493, pruned_loss=0.1134, over 5729086.36 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3296, pruned_loss=0.08471, over 5662120.82 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:05:40,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.775e+02 1.479e+03 1.860e+03 2.553e+03 8.288e+03, threshold=3.721e+03, percent-clipped=10.0
+2023-03-09 01:05:49,245 INFO [train.py:968] (0/2) Epoch 17, batch 33100, giga_loss[loss=0.2482, simple_loss=0.3288, pruned_loss=0.08382, over 28086.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3318, pruned_loss=0.08709, over 5669166.28 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3493, pruned_loss=0.1134, over 5730594.03 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3309, pruned_loss=0.08498, over 5660047.87 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:06:09,192 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 01:06:16,803 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-764000.pt
+2023-03-09 01:06:57,517 INFO [train.py:968] (0/2) Epoch 17, batch 33150, giga_loss[loss=0.246, simple_loss=0.327, pruned_loss=0.08255, over 28096.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.332, pruned_loss=0.08685, over 5671970.71 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3488, pruned_loss=0.113, over 5732841.13 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3315, pruned_loss=0.08509, over 5661726.55 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:07:50,427 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.890e+02 1.492e+03 1.977e+03 2.884e+03 7.675e+03, threshold=3.953e+03, percent-clipped=15.0
+2023-03-09 01:07:59,322 INFO [train.py:968] (0/2) Epoch 17, batch 33200, libri_loss[loss=0.2231, simple_loss=0.2871, pruned_loss=0.07957, over 29468.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3302, pruned_loss=0.08587, over 5677868.08 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.348, pruned_loss=0.1124, over 5737204.27 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3301, pruned_loss=0.08434, over 5664319.60 frames. ], batch size: 70, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:08:57,194 INFO [train.py:968] (0/2) Epoch 17, batch 33250, libri_loss[loss=0.2903, simple_loss=0.3514, pruned_loss=0.1146, over 29520.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3281, pruned_loss=0.08449, over 5677088.41 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3485, pruned_loss=0.1128, over 5731096.05 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3269, pruned_loss=0.082, over 5669161.65 frames. ], batch size: 84, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:09:55,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.387e+02 1.288e+03 1.705e+03 2.320e+03 5.743e+03, threshold=3.409e+03, percent-clipped=3.0
+2023-03-09 01:10:03,980 INFO [train.py:968] (0/2) Epoch 17, batch 33300, giga_loss[loss=0.2408, simple_loss=0.3187, pruned_loss=0.08146, over 27559.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3265, pruned_loss=0.08377, over 5677146.81 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3484, pruned_loss=0.1127, over 5727765.57 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3256, pruned_loss=0.0818, over 5673735.87 frames. ], batch size: 472, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:10:59,224 INFO [train.py:968] (0/2) Epoch 17, batch 33350, giga_loss[loss=0.2101, simple_loss=0.3032, pruned_loss=0.05855, over 28877.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3278, pruned_loss=0.08517, over 5669161.48 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3481, pruned_loss=0.1126, over 5722091.88 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3267, pruned_loss=0.08277, over 5669224.41 frames. ], batch size: 164, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:11:49,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5634, 1.6375, 1.2078, 1.2208], device='cuda:0'), covar=tensor([0.0876, 0.0498, 0.0993, 0.1096], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0437, 0.0509, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 01:12:03,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.715e+02 1.388e+03 1.983e+03 2.791e+03 7.026e+03, threshold=3.966e+03, percent-clipped=13.0
+2023-03-09 01:12:07,517 INFO [train.py:968] (0/2) Epoch 17, batch 33400, libri_loss[loss=0.2544, simple_loss=0.3121, pruned_loss=0.09835, over 29355.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3305, pruned_loss=0.0862, over 5671463.32 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3478, pruned_loss=0.1125, over 5723877.73 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3297, pruned_loss=0.08419, over 5669449.08 frames. ], batch size: 67, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:12:40,350 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=764304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:13:15,302 INFO [train.py:968] (0/2) Epoch 17, batch 33450, giga_loss[loss=0.2408, simple_loss=0.3258, pruned_loss=0.07794, over 28657.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3319, pruned_loss=0.08743, over 5671391.76 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3479, pruned_loss=0.1126, over 5722068.83 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3311, pruned_loss=0.08548, over 5670958.05 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:14:14,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.133e+02 1.417e+03 1.795e+03 2.503e+03 6.212e+03, threshold=3.589e+03, percent-clipped=5.0
+2023-03-09 01:14:16,090 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=764375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:14:20,394 INFO [train.py:968] (0/2) Epoch 17, batch 33500, giga_loss[loss=0.2876, simple_loss=0.355, pruned_loss=0.1101, over 26835.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3345, pruned_loss=0.08977, over 5663677.45 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.347, pruned_loss=0.1123, over 5727476.06 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3342, pruned_loss=0.08776, over 5656814.18 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:14:21,241 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 01:15:04,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3091, 3.1396, 2.9743, 1.4058], device='cuda:0'), covar=tensor([0.0920, 0.1035, 0.0990, 0.2294], device='cuda:0'), in_proj_covar=tensor([0.1154, 0.1065, 0.0915, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 01:15:20,841 INFO [train.py:968] (0/2) Epoch 17, batch 33550, giga_loss[loss=0.242, simple_loss=0.3257, pruned_loss=0.07914, over 28929.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3376, pruned_loss=0.09085, over 5665654.86 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3472, pruned_loss=0.1124, over 5729870.53 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3369, pruned_loss=0.08868, over 5656835.90 frames. ], batch size: 106, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:16:15,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.432e+03 1.885e+03 2.429e+03 7.362e+03, threshold=3.769e+03, percent-clipped=9.0
+2023-03-09 01:16:21,583 INFO [train.py:968] (0/2) Epoch 17, batch 33600, giga_loss[loss=0.2682, simple_loss=0.3464, pruned_loss=0.09501, over 28072.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3401, pruned_loss=0.0925, over 5663503.45 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3476, pruned_loss=0.1128, over 5729318.12 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3389, pruned_loss=0.08965, over 5654384.43 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:16:43,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1096, 1.5021, 1.4576, 1.2989], device='cuda:0'), covar=tensor([0.1837, 0.1662, 0.2138, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.0437, 0.0719, 0.0677, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 01:17:29,203 INFO [train.py:968] (0/2) Epoch 17, batch 33650, giga_loss[loss=0.2462, simple_loss=0.3249, pruned_loss=0.08377, over 28996.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3383, pruned_loss=0.09138, over 5679324.53 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3478, pruned_loss=0.1132, over 5732831.89 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.337, pruned_loss=0.08838, over 5667679.90 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:17:44,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7986, 2.6373, 1.6594, 0.9650], device='cuda:0'), covar=tensor([0.7691, 0.3355, 0.4129, 0.6566], device='cuda:0'), in_proj_covar=tensor([0.1678, 0.1600, 0.1552, 0.1373], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 01:18:31,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.872e+02 1.394e+03 1.840e+03 2.473e+03 8.122e+03, threshold=3.680e+03, percent-clipped=8.0
+2023-03-09 01:18:34,932 INFO [train.py:968] (0/2) Epoch 17, batch 33700, giga_loss[loss=0.235, simple_loss=0.3253, pruned_loss=0.07237, over 28833.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3357, pruned_loss=0.09011, over 5687460.72 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3481, pruned_loss=0.1133, over 5734217.21 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3341, pruned_loss=0.08685, over 5674861.42 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:19:37,610 INFO [train.py:968] (0/2) Epoch 17, batch 33750, giga_loss[loss=0.2942, simple_loss=0.366, pruned_loss=0.1112, over 28518.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3349, pruned_loss=0.08982, over 5678569.56 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3476, pruned_loss=0.113, over 5727861.98 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3337, pruned_loss=0.08684, over 5673183.48 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:20:04,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9426, 2.0490, 1.4623, 1.6077], device='cuda:0'), covar=tensor([0.0832, 0.0572, 0.0969, 0.1073], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0437, 0.0509, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 01:20:41,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.426e+02 1.422e+03 1.843e+03 2.488e+03 9.867e+03, threshold=3.686e+03, percent-clipped=13.0
+2023-03-09 01:20:45,582 INFO [train.py:968] (0/2) Epoch 17, batch 33800, giga_loss[loss=0.2791, simple_loss=0.3367, pruned_loss=0.1108, over 26885.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3333, pruned_loss=0.09007, over 5671855.01 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3471, pruned_loss=0.1128, over 5722749.40 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3325, pruned_loss=0.08732, over 5669988.40 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:20:46,362 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=764679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:20:49,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2988, 1.6489, 1.2977, 1.0072], device='cuda:0'), covar=tensor([0.2413, 0.2296, 0.2653, 0.2128], device='cuda:0'), in_proj_covar=tensor([0.1424, 0.1031, 0.1267, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 01:21:18,476 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-09 01:21:51,560 INFO [train.py:968] (0/2) Epoch 17, batch 33850, giga_loss[loss=0.2405, simple_loss=0.3216, pruned_loss=0.07976, over 28972.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3306, pruned_loss=0.0887, over 5677470.89 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3474, pruned_loss=0.113, over 5721705.32 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3295, pruned_loss=0.08606, over 5676232.37 frames. ], batch size: 186, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:22:17,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=764750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:22:45,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.703e+02 1.299e+03 1.860e+03 2.859e+03 1.373e+04, threshold=3.719e+03, percent-clipped=12.0
+2023-03-09 01:22:50,863 INFO [train.py:968] (0/2) Epoch 17, batch 33900, giga_loss[loss=0.2439, simple_loss=0.3299, pruned_loss=0.07893, over 28439.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3313, pruned_loss=0.08802, over 5678923.73 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3475, pruned_loss=0.113, over 5723037.37 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3298, pruned_loss=0.08523, over 5675651.02 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:22:52,693 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8740, 1.2202, 1.2991, 1.0201], device='cuda:0'), covar=tensor([0.1789, 0.1428, 0.2152, 0.1699], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0720, 0.0679, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 01:23:43,958 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=764822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:23:47,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=764825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:23:52,384 INFO [train.py:968] (0/2) Epoch 17, batch 33950, giga_loss[loss=0.2057, simple_loss=0.293, pruned_loss=0.05915, over 28835.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3297, pruned_loss=0.08636, over 5673974.22 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3475, pruned_loss=0.113, over 5724747.93 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3283, pruned_loss=0.08363, over 5668755.16 frames. ], batch size: 119, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:24:21,020 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=764854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:24:45,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.418e+02 1.414e+03 1.737e+03 2.470e+03 6.898e+03, threshold=3.475e+03, percent-clipped=7.0
+2023-03-09 01:24:48,657 INFO [train.py:968] (0/2) Epoch 17, batch 34000, giga_loss[loss=0.2476, simple_loss=0.3339, pruned_loss=0.0806, over 28945.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3315, pruned_loss=0.08501, over 5672737.02 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3474, pruned_loss=0.1129, over 5717525.86 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3303, pruned_loss=0.08265, over 5674658.30 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:25:06,970 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=764893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:25:11,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=764896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:25:43,597 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=764925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:25:48,689 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4139, 4.0878, 1.7046, 1.5804], device='cuda:0'), covar=tensor([0.0985, 0.0268, 0.0903, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0538, 0.0371, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 01:25:49,018 INFO [train.py:968] (0/2) Epoch 17, batch 34050, giga_loss[loss=0.2423, simple_loss=0.3302, pruned_loss=0.07724, over 28961.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3327, pruned_loss=0.08489, over 5677909.11 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3476, pruned_loss=0.1129, over 5720327.79 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3312, pruned_loss=0.08242, over 5676066.16 frames. ], batch size: 285, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:25:58,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2373, 5.0579, 4.7896, 2.1568], device='cuda:0'), covar=tensor([0.0443, 0.0569, 0.0739, 0.2109], device='cuda:0'), in_proj_covar=tensor([0.1155, 0.1063, 0.0913, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 01:26:50,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.091e+02 1.735e+03 2.571e+03 3.837e+03 9.866e+03, threshold=5.143e+03, percent-clipped=35.0
+2023-03-09 01:26:54,680 INFO [train.py:968] (0/2) Epoch 17, batch 34100, giga_loss[loss=0.2496, simple_loss=0.3323, pruned_loss=0.08344, over 28705.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3335, pruned_loss=0.0858, over 5662205.98 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3475, pruned_loss=0.1129, over 5705431.77 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3319, pruned_loss=0.08303, over 5671704.14 frames. ], batch size: 242, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:27:58,781 INFO [train.py:968] (0/2) Epoch 17, batch 34150, giga_loss[loss=0.2496, simple_loss=0.3362, pruned_loss=0.08155, over 28104.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3333, pruned_loss=0.08609, over 5650789.61 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3474, pruned_loss=0.1129, over 5697208.43 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3317, pruned_loss=0.083, over 5663771.29 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:28:10,526 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 01:28:29,747 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-09 01:29:03,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.907e+02 1.377e+03 1.923e+03 2.451e+03 4.639e+03, threshold=3.846e+03, percent-clipped=0.0
+2023-03-09 01:29:07,360 INFO [train.py:968] (0/2) Epoch 17, batch 34200, giga_loss[loss=0.2309, simple_loss=0.3227, pruned_loss=0.06955, over 28803.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3344, pruned_loss=0.08682, over 5635097.78 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3479, pruned_loss=0.1131, over 5680450.92 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3324, pruned_loss=0.08354, over 5660763.00 frames. ], batch size: 243, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:29:08,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5620, 1.8237, 1.8424, 1.5025], device='cuda:0'), covar=tensor([0.2726, 0.2054, 0.1732, 0.2166], device='cuda:0'), in_proj_covar=tensor([0.1836, 0.1777, 0.1686, 0.1819], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 01:29:41,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3010, 1.4639, 1.2703, 1.5207], device='cuda:0'), covar=tensor([0.0793, 0.0348, 0.0353, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0115, 0.0117, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 01:30:23,148 INFO [train.py:968] (0/2) Epoch 17, batch 34250, giga_loss[loss=0.258, simple_loss=0.3484, pruned_loss=0.08378, over 28808.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3347, pruned_loss=0.08609, over 5645446.00 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3481, pruned_loss=0.1134, over 5682635.51 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3328, pruned_loss=0.08304, over 5663255.50 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:31:26,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.623e+02 1.534e+03 2.084e+03 2.608e+03 9.772e+03, threshold=4.168e+03, percent-clipped=11.0
+2023-03-09 01:31:30,923 INFO [train.py:968] (0/2) Epoch 17, batch 34300, giga_loss[loss=0.2819, simple_loss=0.3686, pruned_loss=0.09762, over 29104.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3385, pruned_loss=0.08785, over 5653865.49 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3479, pruned_loss=0.1132, over 5685000.86 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.337, pruned_loss=0.08531, over 5665534.81 frames. ], batch size: 200, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:32:04,372 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765205.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 01:32:36,094 INFO [train.py:968] (0/2) Epoch 17, batch 34350, giga_loss[loss=0.2461, simple_loss=0.3353, pruned_loss=0.07841, over 28945.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3398, pruned_loss=0.08846, over 5663053.38 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3481, pruned_loss=0.1133, over 5688875.34 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3383, pruned_loss=0.08564, over 5668405.99 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:32:43,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2008, 1.1096, 3.6568, 3.0446], device='cuda:0'), covar=tensor([0.1675, 0.2860, 0.0446, 0.1007], device='cuda:0'), in_proj_covar=tensor([0.0717, 0.0622, 0.0907, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-09 01:33:39,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.039e+02 1.400e+03 1.764e+03 2.458e+03 6.431e+03, threshold=3.528e+03, percent-clipped=2.0
+2023-03-09 01:33:42,161 INFO [train.py:968] (0/2) Epoch 17, batch 34400, giga_loss[loss=0.251, simple_loss=0.3317, pruned_loss=0.0851, over 28992.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3383, pruned_loss=0.08848, over 5666302.34 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.348, pruned_loss=0.1132, over 5692413.76 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3368, pruned_loss=0.08548, over 5666428.68 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:34:05,220 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:34:15,716 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:34:53,454 INFO [train.py:968] (0/2) Epoch 17, batch 34450, giga_loss[loss=0.2157, simple_loss=0.3084, pruned_loss=0.06152, over 28975.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3363, pruned_loss=0.08731, over 5673145.67 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3478, pruned_loss=0.113, over 5684414.14 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3352, pruned_loss=0.08486, over 5680512.42 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:35:04,271 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765334.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 01:35:39,579 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:36:01,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.921e+02 1.216e+03 1.663e+03 2.298e+03 5.661e+03, threshold=3.326e+03, percent-clipped=4.0
+2023-03-09 01:36:06,140 INFO [train.py:968] (0/2) Epoch 17, batch 34500, giga_loss[loss=0.2401, simple_loss=0.3254, pruned_loss=0.07735, over 28922.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3355, pruned_loss=0.08629, over 5667353.50 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3478, pruned_loss=0.1131, over 5678172.33 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3345, pruned_loss=0.0839, over 5678105.50 frames. ], batch size: 213, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:36:32,122 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 01:36:48,331 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 01:37:11,350 INFO [train.py:968] (0/2) Epoch 17, batch 34550, libri_loss[loss=0.2867, simple_loss=0.3359, pruned_loss=0.1188, over 29549.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3348, pruned_loss=0.08604, over 5664126.57 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3475, pruned_loss=0.113, over 5682379.12 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.334, pruned_loss=0.08367, over 5668379.13 frames. ], batch size: 77, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:37:32,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.7958, 1.3201, 1.2400, 1.0843], device='cuda:0'), covar=tensor([0.1913, 0.1085, 0.2189, 0.1579], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0718, 0.0677, 0.0657], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 01:38:10,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.662e+02 1.294e+03 1.757e+03 2.397e+03 5.062e+03, threshold=3.515e+03, percent-clipped=8.0
+2023-03-09 01:38:13,447 INFO [train.py:968] (0/2) Epoch 17, batch 34600, giga_loss[loss=0.2362, simple_loss=0.3251, pruned_loss=0.07368, over 28878.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3374, pruned_loss=0.08766, over 5663085.52 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3474, pruned_loss=0.113, over 5682676.58 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3367, pruned_loss=0.08546, over 5665944.37 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:39:11,562 INFO [train.py:968] (0/2) Epoch 17, batch 34650, giga_loss[loss=0.2525, simple_loss=0.331, pruned_loss=0.08703, over 28891.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.338, pruned_loss=0.08823, over 5679870.17 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3466, pruned_loss=0.1123, over 5690652.70 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3379, pruned_loss=0.08623, over 5674335.79 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:39:47,528 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0368, 1.5521, 1.3755, 1.2744], device='cuda:0'), covar=tensor([0.1784, 0.1244, 0.2071, 0.1591], device='cuda:0'), in_proj_covar=tensor([0.0437, 0.0720, 0.0681, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 01:40:00,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0972, 1.4269, 1.1704, 0.2809], device='cuda:0'), covar=tensor([0.3041, 0.2991, 0.4489, 0.5296], device='cuda:0'), in_proj_covar=tensor([0.1669, 0.1588, 0.1549, 0.1368], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 01:40:06,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.439e+02 1.550e+03 2.137e+03 3.331e+03 8.854e+03, threshold=4.275e+03, percent-clipped=21.0
+2023-03-09 01:40:08,977 INFO [train.py:968] (0/2) Epoch 17, batch 34700, giga_loss[loss=0.2526, simple_loss=0.3296, pruned_loss=0.08784, over 28632.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3354, pruned_loss=0.08864, over 5671643.89 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3463, pruned_loss=0.1123, over 5692648.24 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3354, pruned_loss=0.08655, over 5665062.72 frames. ], batch size: 242, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:40:10,009 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=765580.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 01:41:03,146 INFO [train.py:968] (0/2) Epoch 17, batch 34750, giga_loss[loss=0.2548, simple_loss=0.3126, pruned_loss=0.09847, over 24691.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3344, pruned_loss=0.08886, over 5667537.86 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3461, pruned_loss=0.1122, over 5694649.13 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3343, pruned_loss=0.08657, over 5659943.19 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:41:29,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6139, 1.7879, 1.3796, 1.3276], device='cuda:0'), covar=tensor([0.0785, 0.0395, 0.0934, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0436, 0.0508, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 01:41:43,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=765668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:41:54,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.152e+02 1.559e+03 2.050e+03 2.878e+03 7.217e+03, threshold=4.100e+03, percent-clipped=5.0
+2023-03-09 01:41:54,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=765677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:41:55,683 INFO [train.py:968] (0/2) Epoch 17, batch 34800, giga_loss[loss=0.2921, simple_loss=0.3678, pruned_loss=0.1081, over 28862.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3364, pruned_loss=0.09065, over 5673453.70 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3459, pruned_loss=0.1122, over 5701017.90 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3361, pruned_loss=0.08814, over 5660999.28 frames. ], batch size: 119, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:42:25,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=765709.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 01:42:37,066 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=765723.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 01:42:39,116 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=765726.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 01:42:41,165 INFO [train.py:968] (0/2) Epoch 17, batch 34850, giga_loss[loss=0.2866, simple_loss=0.3725, pruned_loss=0.1004, over 29067.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3455, pruned_loss=0.09545, over 5685263.39 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3461, pruned_loss=0.1122, over 5705782.78 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.345, pruned_loss=0.093, over 5670529.23 frames. ], batch size: 155, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:42:44,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=765732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:43:06,219 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=765755.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 01:43:27,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.110e+02 1.364e+03 1.658e+03 2.231e+03 4.977e+03, threshold=3.315e+03, percent-clipped=4.0
+2023-03-09 01:43:28,741 INFO [train.py:968] (0/2) Epoch 17, batch 34900, giga_loss[loss=0.2981, simple_loss=0.3825, pruned_loss=0.1068, over 28854.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3524, pruned_loss=0.09979, over 5683095.77 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3457, pruned_loss=0.1119, over 5710985.86 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3525, pruned_loss=0.09772, over 5665807.91 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:43:43,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2608, 1.4617, 1.3791, 1.1332], device='cuda:0'), covar=tensor([0.2399, 0.2384, 0.1479, 0.2147], device='cuda:0'), in_proj_covar=tensor([0.1839, 0.1771, 0.1683, 0.1826], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 01:43:57,020 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=765811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:00,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=765814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:04,458 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=765820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:08,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=765823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:13,227 INFO [train.py:968] (0/2) Epoch 17, batch 34950, giga_loss[loss=0.2355, simple_loss=0.3209, pruned_loss=0.07507, over 28532.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3517, pruned_loss=0.1001, over 5680527.45 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3455, pruned_loss=0.1117, over 5705990.89 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3521, pruned_loss=0.09838, over 5669874.96 frames. ], batch size: 65, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:44:24,655 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=765843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:31,967 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=765852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:32,050 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=765852.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 01:44:34,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=765855.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 01:44:41,063 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:51,427 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=765875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:53,243 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.806e+02 1.245e+03 1.610e+03 2.184e+03 8.518e+03, threshold=3.221e+03, percent-clipped=9.0
+2023-03-09 01:44:53,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=765878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:53,860 INFO [train.py:968] (0/2) Epoch 17, batch 35000, giga_loss[loss=0.2294, simple_loss=0.3037, pruned_loss=0.07753, over 29032.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3467, pruned_loss=0.09858, over 5686448.03 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3464, pruned_loss=0.1121, over 5702897.98 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3463, pruned_loss=0.09606, over 5679029.09 frames. ], batch size: 106, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:44:57,897 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=765884.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 01:45:17,963 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=765907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:45:35,675 INFO [train.py:968] (0/2) Epoch 17, batch 35050, giga_loss[loss=0.2134, simple_loss=0.2904, pruned_loss=0.06821, over 28750.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3413, pruned_loss=0.09665, over 5691303.64 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3466, pruned_loss=0.1123, over 5706920.25 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3407, pruned_loss=0.09407, over 5681360.26 frames. ], batch size: 119, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:46:02,992 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-09 01:46:12,776 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-09 01:46:14,190 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-09 01:46:16,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.185e+02 1.086e+03 1.507e+03 2.256e+03 9.007e+03, threshold=3.015e+03, percent-clipped=7.0
+2023-03-09 01:46:17,670 INFO [train.py:968] (0/2) Epoch 17, batch 35100, giga_loss[loss=0.237, simple_loss=0.3098, pruned_loss=0.08207, over 28676.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3341, pruned_loss=0.09339, over 5692896.71 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3471, pruned_loss=0.1124, over 5710177.35 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.333, pruned_loss=0.09081, over 5681754.35 frames. ], batch size: 284, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:46:33,796 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-766000.pt
+2023-03-09 01:46:58,210 INFO [train.py:968] (0/2) Epoch 17, batch 35150, giga_loss[loss=0.208, simple_loss=0.2773, pruned_loss=0.06934, over 28272.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3268, pruned_loss=0.09035, over 5692172.18 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3475, pruned_loss=0.1127, over 5708234.03 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3252, pruned_loss=0.08754, over 5684653.97 frames. ], batch size: 77, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:47:14,381 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 01:47:39,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.735e+02 1.095e+03 1.499e+03 2.225e+03 1.095e+04, threshold=2.999e+03, percent-clipped=15.0
+2023-03-09 01:47:40,662 INFO [train.py:968] (0/2) Epoch 17, batch 35200, giga_loss[loss=0.182, simple_loss=0.2631, pruned_loss=0.05047, over 28521.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3219, pruned_loss=0.08823, over 5684557.56 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.348, pruned_loss=0.113, over 5701080.23 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3196, pruned_loss=0.08526, over 5683930.63 frames. ], batch size: 71, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:48:23,228 INFO [train.py:968] (0/2) Epoch 17, batch 35250, giga_loss[loss=0.22, simple_loss=0.2889, pruned_loss=0.07559, over 28786.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3175, pruned_loss=0.08619, over 5687985.39 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3482, pruned_loss=0.113, over 5704238.25 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3152, pruned_loss=0.08345, over 5684586.15 frames. ], batch size: 99, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:49:08,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.425e+02 1.045e+03 1.309e+03 1.684e+03 3.136e+03, threshold=2.618e+03, percent-clipped=1.0
+2023-03-09 01:49:08,991 INFO [train.py:968] (0/2) Epoch 17, batch 35300, giga_loss[loss=0.2223, simple_loss=0.2984, pruned_loss=0.07309, over 28666.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3138, pruned_loss=0.08406, over 5696260.39 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3481, pruned_loss=0.113, over 5707413.95 frames. ], giga_tot_loss[loss=0.2374, simple_loss=0.3116, pruned_loss=0.08158, over 5690430.25 frames. ], batch size: 242, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:49:40,868 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 01:49:52,051 INFO [train.py:968] (0/2) Epoch 17, batch 35350, giga_loss[loss=0.2222, simple_loss=0.3, pruned_loss=0.07219, over 28846.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.3103, pruned_loss=0.08211, over 5704855.19 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3481, pruned_loss=0.1127, over 5710378.26 frames. ], giga_tot_loss[loss=0.2338, simple_loss=0.3079, pruned_loss=0.0798, over 5697375.97 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:49:53,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4516, 4.2792, 4.0241, 1.9222], device='cuda:0'), covar=tensor([0.0535, 0.0740, 0.0705, 0.2182], device='cuda:0'), in_proj_covar=tensor([0.1148, 0.1058, 0.0908, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 01:49:59,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=766237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:50:36,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.460e+02 9.835e+02 1.243e+03 1.612e+03 5.816e+03, threshold=2.487e+03, percent-clipped=8.0
+2023-03-09 01:50:36,447 INFO [train.py:968] (0/2) Epoch 17, batch 35400, giga_loss[loss=0.2093, simple_loss=0.2859, pruned_loss=0.06639, over 29009.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3076, pruned_loss=0.0809, over 5703580.57 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3486, pruned_loss=0.113, over 5708891.40 frames. ], giga_tot_loss[loss=0.2305, simple_loss=0.3045, pruned_loss=0.07826, over 5698836.96 frames. ], batch size: 128, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:51:03,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4531, 2.6882, 2.3852, 2.4062], device='cuda:0'), covar=tensor([0.1801, 0.2198, 0.2121, 0.2062], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0731, 0.0687, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 01:51:08,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4513, 1.8640, 1.4962, 1.6463], device='cuda:0'), covar=tensor([0.0766, 0.0308, 0.0323, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0115, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 01:51:18,865 INFO [train.py:968] (0/2) Epoch 17, batch 35450, giga_loss[loss=0.2136, simple_loss=0.2856, pruned_loss=0.0708, over 28535.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3046, pruned_loss=0.07942, over 5691331.03 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.349, pruned_loss=0.1132, over 5699918.31 frames. ], giga_tot_loss[loss=0.2266, simple_loss=0.3007, pruned_loss=0.07628, over 5696067.10 frames. ], batch size: 85, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:51:40,215 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-09 01:51:42,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6634, 1.9830, 1.5946, 1.7860], device='cuda:0'), covar=tensor([0.2646, 0.2636, 0.2983, 0.2458], device='cuda:0'), in_proj_covar=tensor([0.1430, 0.1039, 0.1272, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 01:51:58,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.461e+02 1.141e+03 1.553e+03 2.143e+03 7.345e+03, threshold=3.105e+03, percent-clipped=18.0
+2023-03-09 01:51:58,367 INFO [train.py:968] (0/2) Epoch 17, batch 35500, giga_loss[loss=0.2018, simple_loss=0.2805, pruned_loss=0.06153, over 28955.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3031, pruned_loss=0.07885, over 5697950.83 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.349, pruned_loss=0.1129, over 5707979.11 frames. ], giga_tot_loss[loss=0.2243, simple_loss=0.2981, pruned_loss=0.07526, over 5693887.66 frames. ], batch size: 164, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:51:59,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=766380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:52:01,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=766383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:52:25,828 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=766412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:52:41,209 INFO [train.py:968] (0/2) Epoch 17, batch 35550, giga_loss[loss=0.2224, simple_loss=0.2974, pruned_loss=0.07367, over 28712.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.2996, pruned_loss=0.07704, over 5698955.53 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3484, pruned_loss=0.1124, over 5712901.36 frames. ], giga_tot_loss[loss=0.2213, simple_loss=0.295, pruned_loss=0.07376, over 5690991.64 frames. ], batch size: 242, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:53:25,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.519e+02 9.795e+02 1.367e+03 2.023e+03 5.287e+03, threshold=2.735e+03, percent-clipped=10.0
+2023-03-09 01:53:25,486 INFO [train.py:968] (0/2) Epoch 17, batch 35600, libri_loss[loss=0.3161, simple_loss=0.3831, pruned_loss=0.1246, over 29742.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.2963, pruned_loss=0.07537, over 5707609.31 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3488, pruned_loss=0.1125, over 5718183.38 frames. ], giga_tot_loss[loss=0.2171, simple_loss=0.2909, pruned_loss=0.0717, over 5695885.80 frames. ], batch size: 87, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:54:04,050 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 01:54:06,153 INFO [train.py:968] (0/2) Epoch 17, batch 35650, giga_loss[loss=0.2686, simple_loss=0.3435, pruned_loss=0.09683, over 28604.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.2998, pruned_loss=0.07818, over 5693652.23 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3493, pruned_loss=0.1128, over 5714600.69 frames. ], giga_tot_loss[loss=0.2205, simple_loss=0.2934, pruned_loss=0.07383, over 5687059.57 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:54:36,783 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=766562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:54:51,301 INFO [train.py:968] (0/2) Epoch 17, batch 35700, giga_loss[loss=0.2939, simple_loss=0.3702, pruned_loss=0.1088, over 28955.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3115, pruned_loss=0.08412, over 5696520.61 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3494, pruned_loss=0.1125, over 5718041.11 frames. ], giga_tot_loss[loss=0.2326, simple_loss=0.3051, pruned_loss=0.08002, over 5687385.28 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:54:51,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.851e+02 1.155e+03 1.441e+03 1.978e+03 4.598e+03, threshold=2.883e+03, percent-clipped=14.0
+2023-03-09 01:55:39,460 INFO [train.py:968] (0/2) Epoch 17, batch 35750, giga_loss[loss=0.3412, simple_loss=0.3948, pruned_loss=0.1438, over 27606.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3249, pruned_loss=0.09134, over 5687015.88 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3496, pruned_loss=0.1127, over 5712325.44 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3191, pruned_loss=0.08755, over 5684799.87 frames. ], batch size: 472, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:56:11,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4623, 1.6113, 1.4859, 1.2813], device='cuda:0'), covar=tensor([0.2603, 0.2336, 0.1863, 0.2356], device='cuda:0'), in_proj_covar=tensor([0.1858, 0.1781, 0.1704, 0.1845], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 01:56:23,107 INFO [train.py:968] (0/2) Epoch 17, batch 35800, giga_loss[loss=0.2956, simple_loss=0.3721, pruned_loss=0.1095, over 28679.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3369, pruned_loss=0.09727, over 5694644.69 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3504, pruned_loss=0.1131, over 5717136.00 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3312, pruned_loss=0.09344, over 5687861.80 frames. ], batch size: 262, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:56:23,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.849e+02 1.363e+03 1.739e+03 2.419e+03 5.480e+03, threshold=3.479e+03, percent-clipped=16.0
+2023-03-09 01:56:50,116 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 01:57:07,384 INFO [train.py:968] (0/2) Epoch 17, batch 35850, giga_loss[loss=0.3188, simple_loss=0.3955, pruned_loss=0.1211, over 28589.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3414, pruned_loss=0.09847, over 5699440.35 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3503, pruned_loss=0.1132, over 5718924.41 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3368, pruned_loss=0.09512, over 5692229.66 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:57:33,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3539, 1.5756, 1.5262, 1.4182], device='cuda:0'), covar=tensor([0.1792, 0.1937, 0.2350, 0.1948], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0735, 0.0693, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 01:57:55,686 INFO [train.py:968] (0/2) Epoch 17, batch 35900, libri_loss[loss=0.3445, simple_loss=0.3932, pruned_loss=0.1479, over 19701.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.343, pruned_loss=0.09776, over 5679956.31 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3507, pruned_loss=0.1135, over 5702400.63 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3389, pruned_loss=0.09469, over 5689308.05 frames. ], batch size: 187, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:57:56,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.960e+02 1.176e+03 1.532e+03 2.164e+03 6.246e+03, threshold=3.064e+03, percent-clipped=3.0
+2023-03-09 01:58:41,645 INFO [train.py:968] (0/2) Epoch 17, batch 35950, giga_loss[loss=0.3255, simple_loss=0.3836, pruned_loss=0.1337, over 27635.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3453, pruned_loss=0.0988, over 5675308.36 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3507, pruned_loss=0.1135, over 5695610.04 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3419, pruned_loss=0.0961, over 5687558.83 frames. ], batch size: 472, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:58:58,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-09 01:59:26,850 INFO [train.py:968] (0/2) Epoch 17, batch 36000, giga_loss[loss=0.3157, simple_loss=0.3799, pruned_loss=0.1258, over 28929.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.348, pruned_loss=0.1014, over 5667692.34 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3505, pruned_loss=0.1133, over 5691707.43 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3453, pruned_loss=0.09904, over 5679953.18 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:59:26,854 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 01:59:35,323 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2051, simple_loss=0.312, pruned_loss=0.04908, over 944034.00 frames. 
+2023-03-09 01:59:35,324 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 01:59:36,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.941e+02 1.250e+03 1.559e+03 2.177e+03 1.041e+04, threshold=3.118e+03, percent-clipped=15.0
+2023-03-09 01:59:53,627 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-09 02:00:16,107 INFO [train.py:968] (0/2) Epoch 17, batch 36050, giga_loss[loss=0.3105, simple_loss=0.3801, pruned_loss=0.1205, over 28756.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.352, pruned_loss=0.1039, over 5671167.62 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3512, pruned_loss=0.1137, over 5690910.61 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3492, pruned_loss=0.1015, over 5681071.62 frames. ], batch size: 284, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:00:23,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5405, 3.3755, 3.1718, 1.6697], device='cuda:0'), covar=tensor([0.0770, 0.0870, 0.0816, 0.2458], device='cuda:0'), in_proj_covar=tensor([0.1140, 0.1053, 0.0902, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 02:00:25,586 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=766937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:00:53,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5019, 1.7026, 1.5222, 1.5255], device='cuda:0'), covar=tensor([0.1816, 0.2328, 0.2216, 0.2119], device='cuda:0'), in_proj_covar=tensor([0.0451, 0.0738, 0.0695, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 02:00:58,201 INFO [train.py:968] (0/2) Epoch 17, batch 36100, giga_loss[loss=0.2786, simple_loss=0.3581, pruned_loss=0.09959, over 29081.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3548, pruned_loss=0.1054, over 5670173.42 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3515, pruned_loss=0.1137, over 5685789.28 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3524, pruned_loss=0.1032, over 5682490.80 frames. ], batch size: 128, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:00:58,732 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.554e+02 1.120e+03 1.423e+03 1.912e+03 4.724e+03, threshold=2.846e+03, percent-clipped=3.0
+2023-03-09 02:01:41,123 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767027.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 02:01:42,073 INFO [train.py:968] (0/2) Epoch 17, batch 36150, giga_loss[loss=0.2969, simple_loss=0.3704, pruned_loss=0.1117, over 28641.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3576, pruned_loss=0.1061, over 5676526.08 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3515, pruned_loss=0.1137, over 5685789.28 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3558, pruned_loss=0.1043, over 5686112.86 frames. ], batch size: 242, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:02:24,054 INFO [train.py:968] (0/2) Epoch 17, batch 36200, giga_loss[loss=0.2748, simple_loss=0.3491, pruned_loss=0.1002, over 28482.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3596, pruned_loss=0.1071, over 5668802.56 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3521, pruned_loss=0.1139, over 5686756.20 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3579, pruned_loss=0.1053, over 5675301.75 frames. ], batch size: 85, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:02:24,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.884e+02 1.281e+03 1.500e+03 2.195e+03 4.968e+03, threshold=3.000e+03, percent-clipped=12.0
+2023-03-09 02:02:25,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:02:26,506 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:02:27,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:02:48,887 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:03:03,185 INFO [train.py:968] (0/2) Epoch 17, batch 36250, giga_loss[loss=0.3061, simple_loss=0.3792, pruned_loss=0.1166, over 28446.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.36, pruned_loss=0.1056, over 5684453.21 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3526, pruned_loss=0.1143, over 5687959.46 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3582, pruned_loss=0.1037, over 5688031.38 frames. ], batch size: 65, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:03:33,307 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-09 02:03:41,583 INFO [train.py:968] (0/2) Epoch 17, batch 36300, giga_loss[loss=0.2896, simple_loss=0.3638, pruned_loss=0.1077, over 27928.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3587, pruned_loss=0.1035, over 5697334.75 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.353, pruned_loss=0.1143, over 5692102.55 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.357, pruned_loss=0.1017, over 5696612.82 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:03:43,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.262e+02 1.102e+03 1.349e+03 1.703e+03 5.417e+03, threshold=2.698e+03, percent-clipped=4.0
+2023-03-09 02:03:46,716 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:03:59,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7974, 4.3592, 2.0173, 1.8699], device='cuda:0'), covar=tensor([0.0966, 0.0177, 0.0854, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0533, 0.0368, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 02:04:22,200 INFO [train.py:968] (0/2) Epoch 17, batch 36350, giga_loss[loss=0.2652, simple_loss=0.3427, pruned_loss=0.09387, over 28895.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3576, pruned_loss=0.102, over 5703848.45 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3533, pruned_loss=0.1144, over 5695259.63 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3562, pruned_loss=0.1001, over 5700832.73 frames. ], batch size: 112, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:04:41,076 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:05:05,472 INFO [train.py:968] (0/2) Epoch 17, batch 36400, giga_loss[loss=0.2774, simple_loss=0.357, pruned_loss=0.09886, over 28935.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3566, pruned_loss=0.1012, over 5709295.87 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3535, pruned_loss=0.1145, over 5696343.34 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3553, pruned_loss=0.09957, over 5706073.55 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:05:06,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.157e+02 1.086e+03 1.367e+03 1.924e+03 5.806e+03, threshold=2.735e+03, percent-clipped=10.0
+2023-03-09 02:05:20,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4544, 4.2477, 4.0630, 2.3890], device='cuda:0'), covar=tensor([0.0521, 0.0737, 0.0707, 0.1781], device='cuda:0'), in_proj_covar=tensor([0.1137, 0.1046, 0.0898, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 02:05:21,364 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-09 02:05:53,774 INFO [train.py:968] (0/2) Epoch 17, batch 36450, giga_loss[loss=0.3648, simple_loss=0.4125, pruned_loss=0.1586, over 29118.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3594, pruned_loss=0.1059, over 5703047.76 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3537, pruned_loss=0.1145, over 5697946.92 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3583, pruned_loss=0.1044, over 5699386.10 frames. ], batch size: 128, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:06:08,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2926, 1.3782, 1.2354, 1.5147], device='cuda:0'), covar=tensor([0.0816, 0.0352, 0.0341, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-09 02:06:37,503 INFO [train.py:968] (0/2) Epoch 17, batch 36500, giga_loss[loss=0.2887, simple_loss=0.3565, pruned_loss=0.1105, over 28590.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3612, pruned_loss=0.1092, over 5699768.21 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3541, pruned_loss=0.1146, over 5701663.46 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3602, pruned_loss=0.1077, over 5693567.38 frames. ], batch size: 78, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:06:40,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.103e+02 1.339e+03 1.747e+03 2.256e+03 3.879e+03, threshold=3.493e+03, percent-clipped=14.0
+2023-03-09 02:06:55,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7407, 1.8350, 1.7508, 1.6505], device='cuda:0'), covar=tensor([0.1744, 0.1985, 0.2203, 0.1911], device='cuda:0'), in_proj_covar=tensor([0.0452, 0.0739, 0.0693, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 02:06:56,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767402.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 02:07:21,775 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:07:22,837 INFO [train.py:968] (0/2) Epoch 17, batch 36550, giga_loss[loss=0.2917, simple_loss=0.3705, pruned_loss=0.1064, over 28279.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3591, pruned_loss=0.1084, over 5707097.20 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3538, pruned_loss=0.1143, over 5703845.16 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3585, pruned_loss=0.1075, over 5700251.74 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:07:47,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:08:03,370 INFO [train.py:968] (0/2) Epoch 17, batch 36600, giga_loss[loss=0.2679, simple_loss=0.3406, pruned_loss=0.09759, over 28848.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3569, pruned_loss=0.1078, over 5705577.95 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3549, pruned_loss=0.1148, over 5707732.74 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3557, pruned_loss=0.1064, over 5696603.62 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:08:06,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.276e+02 1.267e+03 1.732e+03 2.429e+03 1.140e+04, threshold=3.464e+03, percent-clipped=9.0
+2023-03-09 02:08:46,573 INFO [train.py:968] (0/2) Epoch 17, batch 36650, giga_loss[loss=0.2702, simple_loss=0.3484, pruned_loss=0.09606, over 28801.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3564, pruned_loss=0.1075, over 5699804.56 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3549, pruned_loss=0.1147, over 5701353.76 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3554, pruned_loss=0.1063, over 5697733.22 frames. ], batch size: 99, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:08:56,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5768, 1.6466, 1.8070, 1.4005], device='cuda:0'), covar=tensor([0.1736, 0.2325, 0.1380, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.0873, 0.0691, 0.0918, 0.0818], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 02:09:00,668 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767545.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 02:09:03,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767548.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 02:09:13,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:09:14,358 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:09:27,553 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767577.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 02:09:28,692 INFO [train.py:968] (0/2) Epoch 17, batch 36700, giga_loss[loss=0.2554, simple_loss=0.329, pruned_loss=0.09086, over 28508.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3565, pruned_loss=0.1072, over 5696037.80 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3562, pruned_loss=0.1153, over 5701513.27 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3546, pruned_loss=0.1054, over 5693627.14 frames. ], batch size: 85, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:09:32,492 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.275e+02 1.233e+03 1.593e+03 2.261e+03 6.509e+03, threshold=3.186e+03, percent-clipped=7.0
+2023-03-09 02:09:32,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2801, 3.1206, 2.9533, 1.4716], device='cuda:0'), covar=tensor([0.0923, 0.0971, 0.0835, 0.2266], device='cuda:0'), in_proj_covar=tensor([0.1147, 0.1060, 0.0909, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 02:09:49,533 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:09:52,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:10:12,489 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:10:14,950 INFO [train.py:968] (0/2) Epoch 17, batch 36750, giga_loss[loss=0.2944, simple_loss=0.3515, pruned_loss=0.1186, over 26437.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3537, pruned_loss=0.1051, over 5673125.96 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3564, pruned_loss=0.1156, over 5686410.77 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3519, pruned_loss=0.1032, over 5685302.80 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:10:19,513 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:11:02,101 INFO [train.py:968] (0/2) Epoch 17, batch 36800, giga_loss[loss=0.252, simple_loss=0.323, pruned_loss=0.09052, over 28225.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3478, pruned_loss=0.1019, over 5657568.59 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3566, pruned_loss=0.1156, over 5680663.47 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3461, pruned_loss=0.1002, over 5671804.40 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:11:05,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.592e+02 1.101e+03 1.408e+03 1.953e+03 7.954e+03, threshold=2.817e+03, percent-clipped=6.0
+2023-03-09 02:11:19,029 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-09 02:11:25,305 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:11:28,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:11:35,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2523, 1.3421, 1.0648, 0.9233], device='cuda:0'), covar=tensor([0.0905, 0.0471, 0.1152, 0.1090], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0442, 0.0513, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 02:11:51,639 INFO [train.py:968] (0/2) Epoch 17, batch 36850, libri_loss[loss=0.2962, simple_loss=0.3689, pruned_loss=0.1118, over 29529.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3417, pruned_loss=0.0988, over 5655328.49 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3565, pruned_loss=0.1154, over 5687511.78 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.34, pruned_loss=0.09714, over 5659773.02 frames. ], batch size: 82, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:11:58,717 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:12:36,262 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:12:40,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:12:45,124 INFO [train.py:968] (0/2) Epoch 17, batch 36900, giga_loss[loss=0.2139, simple_loss=0.2933, pruned_loss=0.06722, over 28813.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3373, pruned_loss=0.09693, over 5650440.38 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.357, pruned_loss=0.1156, over 5689975.15 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3353, pruned_loss=0.09506, over 5651118.29 frames. ], batch size: 66, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:12:48,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.095e+02 9.682e+02 1.278e+03 1.839e+03 1.035e+04, threshold=2.555e+03, percent-clipped=16.0
+2023-03-09 02:13:00,571 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-09 02:13:07,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:13:07,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:13:30,254 INFO [train.py:968] (0/2) Epoch 17, batch 36950, giga_loss[loss=0.2642, simple_loss=0.3393, pruned_loss=0.09455, over 28925.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3376, pruned_loss=0.09588, over 5662365.14 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3574, pruned_loss=0.1158, over 5688884.22 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3354, pruned_loss=0.09403, over 5663592.00 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:14:11,174 INFO [train.py:968] (0/2) Epoch 17, batch 37000, giga_loss[loss=0.283, simple_loss=0.3487, pruned_loss=0.1087, over 28768.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3375, pruned_loss=0.0957, over 5668272.49 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3575, pruned_loss=0.1157, over 5693200.83 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3353, pruned_loss=0.09387, over 5665059.64 frames. ], batch size: 119, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:14:14,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.131e+02 1.056e+03 1.315e+03 1.650e+03 4.873e+03, threshold=2.630e+03, percent-clipped=10.0
+2023-03-09 02:14:16,048 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:14:22,918 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2261, 2.5573, 1.2429, 1.4498], device='cuda:0'), covar=tensor([0.1017, 0.0353, 0.0899, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0534, 0.0369, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 02:14:43,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:14:53,514 INFO [train.py:968] (0/2) Epoch 17, batch 37050, giga_loss[loss=0.2121, simple_loss=0.291, pruned_loss=0.06656, over 28218.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3357, pruned_loss=0.0943, over 5684401.20 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3583, pruned_loss=0.1161, over 5693751.76 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3328, pruned_loss=0.0921, over 5681225.70 frames. ], batch size: 77, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:14:59,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:15:06,817 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:15:09,395 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:15:28,590 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 02:15:33,294 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:15:34,373 INFO [train.py:968] (0/2) Epoch 17, batch 37100, libri_loss[loss=0.283, simple_loss=0.3438, pruned_loss=0.1111, over 29367.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3338, pruned_loss=0.09355, over 5702147.45 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3589, pruned_loss=0.1163, over 5698433.27 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3304, pruned_loss=0.0911, over 5695304.12 frames. ], batch size: 71, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:15:37,033 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.964e+02 1.129e+03 1.457e+03 2.292e+03 6.784e+03, threshold=2.914e+03, percent-clipped=20.0
+2023-03-09 02:15:50,800 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-768000.pt
+2023-03-09 02:16:15,046 INFO [train.py:968] (0/2) Epoch 17, batch 37150, giga_loss[loss=0.2716, simple_loss=0.3339, pruned_loss=0.1047, over 28826.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3319, pruned_loss=0.09285, over 5705141.79 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3594, pruned_loss=0.1164, over 5700770.45 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3283, pruned_loss=0.09044, over 5697838.15 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:16:53,650 INFO [train.py:968] (0/2) Epoch 17, batch 37200, giga_loss[loss=0.241, simple_loss=0.3183, pruned_loss=0.0819, over 28800.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3297, pruned_loss=0.09191, over 5701396.68 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3599, pruned_loss=0.1164, over 5692790.54 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3259, pruned_loss=0.08951, over 5703296.03 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:16:53,912 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=768079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:16:57,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=768082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:16:57,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.143e+02 9.810e+02 1.222e+03 1.784e+03 7.878e+03, threshold=2.444e+03, percent-clipped=10.0
+2023-03-09 02:17:20,257 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=768111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:17:34,687 INFO [train.py:968] (0/2) Epoch 17, batch 37250, giga_loss[loss=0.2212, simple_loss=0.2978, pruned_loss=0.07224, over 28929.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3279, pruned_loss=0.09126, over 5710609.35 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.36, pruned_loss=0.1163, over 5697054.10 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3242, pruned_loss=0.089, over 5708573.42 frames. ], batch size: 213, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:18:15,187 INFO [train.py:968] (0/2) Epoch 17, batch 37300, giga_loss[loss=0.2583, simple_loss=0.3282, pruned_loss=0.09424, over 28577.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3264, pruned_loss=0.09051, over 5714875.05 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3612, pruned_loss=0.1168, over 5698860.81 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3216, pruned_loss=0.08765, over 5711908.94 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:18:18,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.379e+02 9.788e+02 1.266e+03 1.758e+03 3.800e+03, threshold=2.532e+03, percent-clipped=9.0
+2023-03-09 02:18:56,408 INFO [train.py:968] (0/2) Epoch 17, batch 37350, giga_loss[loss=0.2576, simple_loss=0.3339, pruned_loss=0.0907, over 28609.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3243, pruned_loss=0.08945, over 5709860.06 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3618, pruned_loss=0.1171, over 5693810.98 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3194, pruned_loss=0.0866, over 5712300.42 frames. ], batch size: 336, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:19:11,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2694, 3.0999, 2.9306, 1.4390], device='cuda:0'), covar=tensor([0.0925, 0.1066, 0.0837, 0.2340], device='cuda:0'), in_proj_covar=tensor([0.1136, 0.1052, 0.0902, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 02:19:19,194 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=768260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:19:34,783 INFO [train.py:968] (0/2) Epoch 17, batch 37400, libri_loss[loss=0.2858, simple_loss=0.3677, pruned_loss=0.1019, over 29545.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3238, pruned_loss=0.08912, over 5715540.96 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3624, pruned_loss=0.1172, over 5701028.33 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3181, pruned_loss=0.08579, over 5711458.04 frames. ], batch size: 80, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:19:39,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.321e+02 1.038e+03 1.417e+03 1.929e+03 8.904e+03, threshold=2.834e+03, percent-clipped=17.0
+2023-03-09 02:19:44,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=768291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:19:59,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6365, 1.9492, 1.4944, 1.9165], device='cuda:0'), covar=tensor([0.2547, 0.2585, 0.3038, 0.2212], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1045, 0.1276, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 02:20:15,394 INFO [train.py:968] (0/2) Epoch 17, batch 37450, giga_loss[loss=0.2174, simple_loss=0.3024, pruned_loss=0.06622, over 28985.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3236, pruned_loss=0.0892, over 5711800.14 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.363, pruned_loss=0.1174, over 5702770.02 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3179, pruned_loss=0.0859, over 5707231.00 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:20:55,927 INFO [train.py:968] (0/2) Epoch 17, batch 37500, giga_loss[loss=0.2987, simple_loss=0.3612, pruned_loss=0.1181, over 28895.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3239, pruned_loss=0.08909, over 5712634.36 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3634, pruned_loss=0.1175, over 5697360.47 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3179, pruned_loss=0.08559, over 5713469.94 frames. ], batch size: 99, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:21:00,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.423e+02 1.060e+03 1.406e+03 2.121e+03 2.428e+04, threshold=2.813e+03, percent-clipped=16.0
+2023-03-09 02:21:11,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9131, 1.9482, 2.0998, 1.6991], device='cuda:0'), covar=tensor([0.1753, 0.2302, 0.1346, 0.1587], device='cuda:0'), in_proj_covar=tensor([0.0874, 0.0693, 0.0920, 0.0818], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 02:21:14,772 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=768403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:21:16,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=768406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:21:19,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5057, 3.4364, 1.6535, 1.5704], device='cuda:0'), covar=tensor([0.0978, 0.0326, 0.0889, 0.1352], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0530, 0.0366, 0.0410], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 02:21:40,448 INFO [train.py:968] (0/2) Epoch 17, batch 37550, giga_loss[loss=0.2728, simple_loss=0.3493, pruned_loss=0.09816, over 28650.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3295, pruned_loss=0.09245, over 5717551.50 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.364, pruned_loss=0.1177, over 5701376.78 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3236, pruned_loss=0.08909, over 5715180.97 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:21:45,690 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=768434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:21:46,184 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=768435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:21:47,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=768437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:22:13,370 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=768466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:22:26,873 INFO [train.py:968] (0/2) Epoch 17, batch 37600, giga_loss[loss=0.2751, simple_loss=0.3559, pruned_loss=0.09714, over 28542.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.335, pruned_loss=0.09588, over 5714519.17 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3641, pruned_loss=0.1177, over 5705061.62 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3297, pruned_loss=0.09278, over 5709629.82 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:22:34,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.090e+02 1.258e+03 1.602e+03 2.200e+03 5.574e+03, threshold=3.204e+03, percent-clipped=12.0
+2023-03-09 02:23:08,853 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=768523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:23:14,478 INFO [train.py:968] (0/2) Epoch 17, batch 37650, giga_loss[loss=0.3565, simple_loss=0.4037, pruned_loss=0.1547, over 27577.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3432, pruned_loss=0.101, over 5707074.40 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3646, pruned_loss=0.1178, over 5709145.78 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3378, pruned_loss=0.09791, over 5699430.19 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:23:17,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2460, 0.8257, 0.9143, 1.4366], device='cuda:0'), covar=tensor([0.0801, 0.0377, 0.0352, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:0')
+2023-03-09 02:24:03,732 INFO [train.py:968] (0/2) Epoch 17, batch 37700, giga_loss[loss=0.2824, simple_loss=0.3593, pruned_loss=0.1028, over 28814.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3481, pruned_loss=0.1034, over 5688525.67 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3647, pruned_loss=0.1179, over 5704342.07 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3433, pruned_loss=0.1005, over 5685869.94 frames. ], batch size: 99, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:24:09,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.476e+02 1.276e+03 1.655e+03 2.305e+03 5.870e+03, threshold=3.310e+03, percent-clipped=9.0
+2023-03-09 02:24:48,567 INFO [train.py:968] (0/2) Epoch 17, batch 37750, libri_loss[loss=0.3389, simple_loss=0.3989, pruned_loss=0.1394, over 29540.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3526, pruned_loss=0.1051, over 5682400.42 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3646, pruned_loss=0.1178, over 5697095.40 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3486, pruned_loss=0.1026, over 5686339.89 frames. ], batch size: 84, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:24:50,400 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1722, 1.3807, 1.1138, 0.9706], device='cuda:0'), covar=tensor([0.0975, 0.0481, 0.1082, 0.1054], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0438, 0.0509, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 02:25:20,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3945, 1.8232, 1.2199, 0.8358], device='cuda:0'), covar=tensor([0.5078, 0.2845, 0.2602, 0.4712], device='cuda:0'), in_proj_covar=tensor([0.1665, 0.1590, 0.1553, 0.1367], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 02:25:33,611 INFO [train.py:968] (0/2) Epoch 17, batch 37800, giga_loss[loss=0.3409, simple_loss=0.4052, pruned_loss=0.1383, over 28258.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3589, pruned_loss=0.109, over 5674292.90 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3647, pruned_loss=0.1179, over 5693388.09 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3553, pruned_loss=0.1066, over 5680775.63 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:25:38,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.050e+02 1.229e+03 1.629e+03 2.259e+03 1.014e+04, threshold=3.257e+03, percent-clipped=10.0
+2023-03-09 02:26:07,787 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=768718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:26:08,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5407, 1.7151, 1.2754, 1.2811], device='cuda:0'), covar=tensor([0.0872, 0.0403, 0.0887, 0.0899], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0438, 0.0510, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 02:26:15,283 INFO [train.py:968] (0/2) Epoch 17, batch 37850, giga_loss[loss=0.2246, simple_loss=0.3119, pruned_loss=0.06863, over 28648.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3581, pruned_loss=0.108, over 5674142.96 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3648, pruned_loss=0.118, over 5688414.09 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3551, pruned_loss=0.1058, over 5683437.95 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:26:35,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8499, 2.1677, 1.3813, 1.6690], device='cuda:0'), covar=tensor([0.0947, 0.0504, 0.1035, 0.0975], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0438, 0.0509, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 02:26:54,620 INFO [train.py:968] (0/2) Epoch 17, batch 37900, giga_loss[loss=0.2339, simple_loss=0.3139, pruned_loss=0.077, over 28596.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3553, pruned_loss=0.1056, over 5676264.54 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.365, pruned_loss=0.1181, over 5681835.51 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3524, pruned_loss=0.1034, over 5689443.66 frames. ], batch size: 336, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:27:00,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.383e+02 1.236e+03 1.672e+03 2.321e+03 1.034e+04, threshold=3.344e+03, percent-clipped=11.0
+2023-03-09 02:27:36,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6063, 1.5282, 4.9137, 3.7818], device='cuda:0'), covar=tensor([0.1672, 0.2702, 0.0357, 0.0706], device='cuda:0'), in_proj_covar=tensor([0.0721, 0.0621, 0.0909, 0.0841], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-09 02:27:37,334 INFO [train.py:968] (0/2) Epoch 17, batch 37950, giga_loss[loss=0.2371, simple_loss=0.3225, pruned_loss=0.0759, over 29035.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3524, pruned_loss=0.1031, over 5685763.35 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3651, pruned_loss=0.1183, over 5687272.80 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3498, pruned_loss=0.101, over 5691666.11 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:28:21,441 INFO [train.py:968] (0/2) Epoch 17, batch 38000, giga_loss[loss=0.2828, simple_loss=0.3574, pruned_loss=0.1041, over 28067.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3521, pruned_loss=0.1026, over 5695164.99 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3655, pruned_loss=0.1187, over 5691751.23 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3494, pruned_loss=0.1002, over 5695766.15 frames. ], batch size: 412, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:28:27,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.649e+02 1.254e+03 1.585e+03 2.535e+03 1.170e+04, threshold=3.169e+03, percent-clipped=16.0
+2023-03-09 02:28:39,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=768898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:28:53,720 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 02:29:04,592 INFO [train.py:968] (0/2) Epoch 17, batch 38050, giga_loss[loss=0.3789, simple_loss=0.4154, pruned_loss=0.1712, over 26526.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.355, pruned_loss=0.1044, over 5695099.15 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3658, pruned_loss=0.1189, over 5694128.24 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3523, pruned_loss=0.102, over 5693624.41 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:29:47,722 INFO [train.py:968] (0/2) Epoch 17, batch 38100, giga_loss[loss=0.308, simple_loss=0.3853, pruned_loss=0.1153, over 28458.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3569, pruned_loss=0.1057, over 5698548.12 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.366, pruned_loss=0.1189, over 5696853.95 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3542, pruned_loss=0.1034, over 5694862.50 frames. ], batch size: 71, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:29:54,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.848e+02 1.212e+03 1.503e+03 1.953e+03 6.541e+03, threshold=3.005e+03, percent-clipped=9.0
+2023-03-09 02:30:35,174 INFO [train.py:968] (0/2) Epoch 17, batch 38150, giga_loss[loss=0.2837, simple_loss=0.3573, pruned_loss=0.105, over 28930.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.359, pruned_loss=0.1074, over 5690435.78 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3665, pruned_loss=0.1191, over 5691832.39 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3563, pruned_loss=0.1051, over 5691984.40 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:30:43,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=769039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:30:45,717 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:30:48,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:30:49,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7547, 1.0533, 2.8792, 2.6163], device='cuda:0'), covar=tensor([0.1755, 0.2648, 0.0592, 0.1034], device='cuda:0'), in_proj_covar=tensor([0.0720, 0.0619, 0.0911, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-09 02:31:01,706 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3451, 2.7963, 1.4547, 1.4243], device='cuda:0'), covar=tensor([0.0926, 0.0333, 0.0814, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0533, 0.0368, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 02:31:11,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-09 02:31:13,759 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:31:19,771 INFO [train.py:968] (0/2) Epoch 17, batch 38200, libri_loss[loss=0.2772, simple_loss=0.3481, pruned_loss=0.1031, over 29584.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3588, pruned_loss=0.1076, over 5697414.20 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3664, pruned_loss=0.1189, over 5696251.32 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3566, pruned_loss=0.1057, over 5694618.80 frames. ], batch size: 76, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:31:23,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=769083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:31:25,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.215e+02 1.264e+03 1.546e+03 2.116e+03 5.688e+03, threshold=3.091e+03, percent-clipped=8.0
+2023-03-09 02:31:30,126 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=769091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:31:31,563 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:32:00,267 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 02:32:01,317 INFO [train.py:968] (0/2) Epoch 17, batch 38250, giga_loss[loss=0.3207, simple_loss=0.376, pruned_loss=0.1327, over 27573.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3597, pruned_loss=0.1086, over 5689375.89 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.367, pruned_loss=0.1194, over 5684911.91 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3572, pruned_loss=0.1065, over 5696818.62 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:32:07,741 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=769137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:32:08,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1437, 2.1743, 1.9691, 1.8748], device='cuda:0'), covar=tensor([0.1776, 0.2295, 0.2298, 0.2102], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0742, 0.0698, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 02:32:33,273 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=769169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:32:42,928 INFO [train.py:968] (0/2) Epoch 17, batch 38300, giga_loss[loss=0.2977, simple_loss=0.3779, pruned_loss=0.1087, over 29005.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3602, pruned_loss=0.1088, over 5691244.93 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3673, pruned_loss=0.1195, over 5690628.90 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3578, pruned_loss=0.1067, over 5692152.40 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:32:47,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.251e+02 1.171e+03 1.400e+03 1.964e+03 7.472e+03, threshold=2.799e+03, percent-clipped=9.0
+2023-03-09 02:33:05,012 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-09 02:33:24,758 INFO [train.py:968] (0/2) Epoch 17, batch 38350, giga_loss[loss=0.3084, simple_loss=0.3827, pruned_loss=0.1171, over 28974.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3594, pruned_loss=0.1069, over 5696523.78 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3671, pruned_loss=0.1194, over 5691843.76 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3576, pruned_loss=0.1053, over 5696153.32 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:33:31,898 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:33:35,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:33:59,677 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:34:07,803 INFO [train.py:968] (0/2) Epoch 17, batch 38400, giga_loss[loss=0.3027, simple_loss=0.367, pruned_loss=0.1192, over 28685.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3583, pruned_loss=0.1052, over 5707697.31 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.367, pruned_loss=0.1192, over 5695356.42 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3569, pruned_loss=0.1038, over 5704426.87 frames. ], batch size: 92, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:34:10,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5550, 1.8952, 1.7223, 1.8240], device='cuda:0'), covar=tensor([0.1591, 0.1469, 0.1982, 0.1507], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0738, 0.0696, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 02:34:12,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.448e+02 1.081e+03 1.277e+03 1.653e+03 4.106e+03, threshold=2.554e+03, percent-clipped=3.0
+2023-03-09 02:34:48,973 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0266, 2.2122, 1.5647, 1.7830], device='cuda:0'), covar=tensor([0.0906, 0.0678, 0.0981, 0.1081], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0437, 0.0508, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 02:34:49,927 INFO [train.py:968] (0/2) Epoch 17, batch 38450, libri_loss[loss=0.3095, simple_loss=0.369, pruned_loss=0.125, over 25627.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3561, pruned_loss=0.1042, over 5704284.91 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1191, over 5696434.44 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3548, pruned_loss=0.1029, over 5701397.34 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:35:01,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3395, 1.8999, 1.4658, 1.4388], device='cuda:0'), covar=tensor([0.0754, 0.0367, 0.0321, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:0')
+2023-03-09 02:35:30,651 INFO [train.py:968] (0/2) Epoch 17, batch 38500, giga_loss[loss=0.2739, simple_loss=0.3435, pruned_loss=0.1021, over 28973.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.354, pruned_loss=0.1035, over 5707236.05 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3667, pruned_loss=0.1189, over 5700739.71 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.353, pruned_loss=0.1024, over 5701102.66 frames. ], batch size: 106, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:35:36,300 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.036e+02 1.053e+03 1.372e+03 1.841e+03 4.862e+03, threshold=2.745e+03, percent-clipped=13.0
+2023-03-09 02:36:00,965 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:36:11,950 INFO [train.py:968] (0/2) Epoch 17, batch 38550, giga_loss[loss=0.315, simple_loss=0.3751, pruned_loss=0.1275, over 28955.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3529, pruned_loss=0.1032, over 5708455.00 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3666, pruned_loss=0.1189, over 5703344.78 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.352, pruned_loss=0.1022, over 5701480.45 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:36:34,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:36:42,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:36:53,231 INFO [train.py:968] (0/2) Epoch 17, batch 38600, giga_loss[loss=0.2602, simple_loss=0.3387, pruned_loss=0.09083, over 28580.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3524, pruned_loss=0.1033, over 5702282.51 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3668, pruned_loss=0.119, over 5696140.89 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3514, pruned_loss=0.1021, over 5702607.64 frames. ], batch size: 78, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:37:01,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.009e+02 1.055e+03 1.216e+03 1.652e+03 4.531e+03, threshold=2.431e+03, percent-clipped=9.0
+2023-03-09 02:37:21,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:37:33,544 INFO [train.py:968] (0/2) Epoch 17, batch 38650, libri_loss[loss=0.3639, simple_loss=0.4089, pruned_loss=0.1595, over 18784.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3529, pruned_loss=0.1035, over 5684451.60 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3673, pruned_loss=0.1194, over 5678405.96 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3512, pruned_loss=0.1017, over 5702027.49 frames. ], batch size: 187, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:37:41,972 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-09 02:37:46,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:37:56,193 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:37:58,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:37:59,891 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-09 02:38:12,048 INFO [train.py:968] (0/2) Epoch 17, batch 38700, giga_loss[loss=0.2568, simple_loss=0.3371, pruned_loss=0.08825, over 28770.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3536, pruned_loss=0.1035, over 5693632.33 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3674, pruned_loss=0.1194, over 5681381.80 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3517, pruned_loss=0.1016, over 5705191.77 frames. ], batch size: 99, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:38:18,669 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.065e+02 1.110e+03 1.328e+03 1.697e+03 7.506e+03, threshold=2.656e+03, percent-clipped=10.0
+2023-03-09 02:38:20,383 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:29,687 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:31,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:35,366 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:37,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:50,160 INFO [train.py:968] (0/2) Epoch 17, batch 38750, giga_loss[loss=0.2314, simple_loss=0.3179, pruned_loss=0.07243, over 28779.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3515, pruned_loss=0.1014, over 5706845.50 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3663, pruned_loss=0.1187, over 5689535.99 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3504, pruned_loss=0.09991, over 5709653.78 frames. ], batch size: 60, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:38:53,292 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:59,909 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:05,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4660, 1.2976, 4.4437, 3.4105], device='cuda:0'), covar=tensor([0.1635, 0.2810, 0.0364, 0.0975], device='cuda:0'), in_proj_covar=tensor([0.0719, 0.0619, 0.0908, 0.0843], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-09 02:39:12,135 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:14,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:30,556 INFO [train.py:968] (0/2) Epoch 17, batch 38800, giga_loss[loss=0.2652, simple_loss=0.3444, pruned_loss=0.09295, over 28874.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3512, pruned_loss=0.1011, over 5707409.20 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3663, pruned_loss=0.1186, over 5688387.91 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.09965, over 5710760.64 frames. ], batch size: 174, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:39:39,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.691e+02 1.016e+03 1.264e+03 1.648e+03 4.671e+03, threshold=2.528e+03, percent-clipped=6.0
+2023-03-09 02:39:39,928 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:39,972 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:42,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:40:06,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:40:14,352 INFO [train.py:968] (0/2) Epoch 17, batch 38850, giga_loss[loss=0.264, simple_loss=0.341, pruned_loss=0.0935, over 28562.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3501, pruned_loss=0.1011, over 5702680.03 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3661, pruned_loss=0.1185, over 5692910.10 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.349, pruned_loss=0.09969, over 5701539.56 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:40:18,167 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-09 02:40:21,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7875, 2.1129, 1.9520, 1.5444], device='cuda:0'), covar=tensor([0.1877, 0.2384, 0.1548, 0.1746], device='cuda:0'), in_proj_covar=tensor([0.0873, 0.0696, 0.0919, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 02:40:54,792 INFO [train.py:968] (0/2) Epoch 17, batch 38900, giga_loss[loss=0.249, simple_loss=0.3263, pruned_loss=0.08585, over 28855.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3457, pruned_loss=0.09876, over 5704913.12 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3657, pruned_loss=0.1182, over 5695153.55 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.345, pruned_loss=0.09771, over 5702172.78 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:41:01,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.041e+02 1.157e+03 1.513e+03 1.963e+03 6.228e+03, threshold=3.027e+03, percent-clipped=9.0
+2023-03-09 02:41:10,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5682, 1.8734, 1.4713, 1.8143], device='cuda:0'), covar=tensor([0.2661, 0.2499, 0.2924, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.1433, 0.1040, 0.1272, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 02:41:35,427 INFO [train.py:968] (0/2) Epoch 17, batch 38950, giga_loss[loss=0.2494, simple_loss=0.3274, pruned_loss=0.08572, over 28707.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.342, pruned_loss=0.09644, over 5709018.98 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3656, pruned_loss=0.118, over 5697427.45 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3414, pruned_loss=0.0956, over 5704997.29 frames. ], batch size: 60, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:41:40,421 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 02:42:16,997 INFO [train.py:968] (0/2) Epoch 17, batch 39000, giga_loss[loss=0.2658, simple_loss=0.34, pruned_loss=0.09584, over 29073.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3435, pruned_loss=0.098, over 5701681.48 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3662, pruned_loss=0.1184, over 5694250.12 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3417, pruned_loss=0.09628, over 5701954.69 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:42:17,002 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 02:42:25,447 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2094, simple_loss=0.3168, pruned_loss=0.05107, over 944034.00 frames. 
+2023-03-09 02:42:25,447 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 02:42:32,758 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.950e+02 1.166e+03 1.541e+03 2.123e+03 1.044e+04, threshold=3.081e+03, percent-clipped=15.0
+2023-03-09 02:43:07,938 INFO [train.py:968] (0/2) Epoch 17, batch 39050, giga_loss[loss=0.2729, simple_loss=0.3482, pruned_loss=0.09873, over 28848.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3433, pruned_loss=0.09849, over 5708119.74 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3665, pruned_loss=0.1185, over 5697875.47 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3412, pruned_loss=0.09663, over 5705282.11 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:43:25,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7393, 2.6893, 1.8438, 1.0054], device='cuda:0'), covar=tensor([0.8082, 0.3400, 0.3552, 0.6178], device='cuda:0'), in_proj_covar=tensor([0.1669, 0.1576, 0.1554, 0.1367], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 02:43:47,146 INFO [train.py:968] (0/2) Epoch 17, batch 39100, giga_loss[loss=0.2818, simple_loss=0.3414, pruned_loss=0.1111, over 28671.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3408, pruned_loss=0.09777, over 5706374.77 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3657, pruned_loss=0.1181, over 5699414.05 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3389, pruned_loss=0.09591, over 5702973.29 frames. ], batch size: 92, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:43:54,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.598e+02 1.128e+03 1.436e+03 1.834e+03 4.604e+03, threshold=2.872e+03, percent-clipped=3.0
+2023-03-09 02:44:02,926 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-770000.pt
+2023-03-09 02:44:04,946 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-09 02:44:21,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5174, 1.7927, 1.4739, 1.2834], device='cuda:0'), covar=tensor([0.2526, 0.2538, 0.2940, 0.2364], device='cuda:0'), in_proj_covar=tensor([0.1437, 0.1043, 0.1272, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 02:44:25,585 INFO [train.py:968] (0/2) Epoch 17, batch 39150, giga_loss[loss=0.2471, simple_loss=0.316, pruned_loss=0.0891, over 28750.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3399, pruned_loss=0.09773, over 5700994.70 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3664, pruned_loss=0.1185, over 5694413.62 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3372, pruned_loss=0.09545, over 5702658.98 frames. ], batch size: 92, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:45:05,007 INFO [train.py:968] (0/2) Epoch 17, batch 39200, giga_loss[loss=0.2437, simple_loss=0.3156, pruned_loss=0.08589, over 28665.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3377, pruned_loss=0.09648, over 5710477.52 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3664, pruned_loss=0.1185, over 5697958.69 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3348, pruned_loss=0.0941, over 5708956.92 frames. ], batch size: 78, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:45:11,602 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=770085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:45:13,847 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.640e+02 1.112e+03 1.384e+03 1.865e+03 5.845e+03, threshold=2.768e+03, percent-clipped=7.0
+2023-03-09 02:45:15,254 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6453, 1.7767, 1.2604, 1.3768], device='cuda:0'), covar=tensor([0.0900, 0.0658, 0.1087, 0.1201], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0438, 0.0509, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 02:45:46,891 INFO [train.py:968] (0/2) Epoch 17, batch 39250, giga_loss[loss=0.2387, simple_loss=0.3166, pruned_loss=0.08042, over 28626.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3373, pruned_loss=0.09662, over 5710211.64 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3664, pruned_loss=0.1186, over 5701847.58 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3344, pruned_loss=0.09425, over 5705828.65 frames. ], batch size: 92, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:46:29,599 INFO [train.py:968] (0/2) Epoch 17, batch 39300, giga_loss[loss=0.2316, simple_loss=0.3193, pruned_loss=0.07199, over 28551.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3395, pruned_loss=0.09722, over 5715436.35 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3665, pruned_loss=0.1186, over 5707306.62 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3363, pruned_loss=0.09468, over 5707317.85 frames. ], batch size: 60, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:46:36,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.267e+02 1.076e+03 1.374e+03 2.013e+03 7.729e+03, threshold=2.747e+03, percent-clipped=12.0
+2023-03-09 02:47:13,943 INFO [train.py:968] (0/2) Epoch 17, batch 39350, giga_loss[loss=0.3182, simple_loss=0.3755, pruned_loss=0.1304, over 26723.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3419, pruned_loss=0.09746, over 5713819.87 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3666, pruned_loss=0.1187, over 5707476.84 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3388, pruned_loss=0.09502, over 5707004.93 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:47:25,009 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=770241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:47:27,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3811, 3.4829, 1.5357, 1.5375], device='cuda:0'), covar=tensor([0.0989, 0.0231, 0.0967, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0535, 0.0367, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 02:47:55,034 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=770277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:47:57,041 INFO [train.py:968] (0/2) Epoch 17, batch 39400, giga_loss[loss=0.3315, simple_loss=0.3933, pruned_loss=0.1348, over 28632.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3445, pruned_loss=0.09843, over 5710620.82 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.366, pruned_loss=0.1182, over 5710926.29 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3422, pruned_loss=0.09651, over 5701988.53 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:48:05,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.820e+02 1.075e+03 1.356e+03 1.833e+03 4.798e+03, threshold=2.712e+03, percent-clipped=7.0
+2023-03-09 02:48:41,663 INFO [train.py:968] (0/2) Epoch 17, batch 39450, giga_loss[loss=0.2209, simple_loss=0.3027, pruned_loss=0.06959, over 29035.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3432, pruned_loss=0.09686, over 5700668.08 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3659, pruned_loss=0.1181, over 5713378.60 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3413, pruned_loss=0.09517, over 5691760.85 frames. ], batch size: 106, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:48:49,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2880, 3.5409, 1.3943, 1.5769], device='cuda:0'), covar=tensor([0.0995, 0.0284, 0.0998, 0.1310], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0535, 0.0367, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 02:49:23,611 INFO [train.py:968] (0/2) Epoch 17, batch 39500, giga_loss[loss=0.244, simple_loss=0.3221, pruned_loss=0.08292, over 29018.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3426, pruned_loss=0.09583, over 5702894.53 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3656, pruned_loss=0.1178, over 5716356.83 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3407, pruned_loss=0.09427, over 5693007.60 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:49:32,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.987e+02 1.053e+03 1.313e+03 1.684e+03 5.116e+03, threshold=2.625e+03, percent-clipped=3.0
+2023-03-09 02:50:06,598 INFO [train.py:968] (0/2) Epoch 17, batch 39550, giga_loss[loss=0.2817, simple_loss=0.3624, pruned_loss=0.1005, over 28955.00 frames. ], tot_loss[loss=0.269, simple_loss=0.344, pruned_loss=0.09696, over 5699691.92 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3659, pruned_loss=0.118, over 5715928.51 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3418, pruned_loss=0.09501, over 5691907.28 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:50:12,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3621, 1.9409, 1.4289, 0.7547], device='cuda:0'), covar=tensor([0.6149, 0.3065, 0.3775, 0.6376], device='cuda:0'), in_proj_covar=tensor([0.1674, 0.1583, 0.1560, 0.1376], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 02:50:14,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4461, 1.6126, 1.4722, 1.6352], device='cuda:0'), covar=tensor([0.0636, 0.0289, 0.0297, 0.0681], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:0')
+2023-03-09 02:50:22,779 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=770449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:50:28,750 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-09 02:50:31,934 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=770460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:50:50,318 INFO [train.py:968] (0/2) Epoch 17, batch 39600, giga_loss[loss=0.2797, simple_loss=0.3541, pruned_loss=0.1026, over 28255.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3436, pruned_loss=0.0969, over 5697015.98 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3659, pruned_loss=0.118, over 5718691.09 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3415, pruned_loss=0.09507, over 5688214.53 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:50:57,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.473e+02 1.252e+03 1.611e+03 2.044e+03 6.302e+03, threshold=3.221e+03, percent-clipped=11.0
+2023-03-09 02:51:09,728 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-09 02:51:20,962 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=770518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:51:30,955 INFO [train.py:968] (0/2) Epoch 17, batch 39650, giga_loss[loss=0.2828, simple_loss=0.3607, pruned_loss=0.1024, over 28711.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3464, pruned_loss=0.09861, over 5702876.20 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3657, pruned_loss=0.1179, over 5720434.75 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3447, pruned_loss=0.097, over 5694214.89 frames. ], batch size: 284, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:51:33,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=770532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:52:11,374 INFO [train.py:968] (0/2) Epoch 17, batch 39700, giga_loss[loss=0.2809, simple_loss=0.3609, pruned_loss=0.1004, over 28802.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3507, pruned_loss=0.1018, over 5710389.25 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3657, pruned_loss=0.118, over 5725590.92 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3488, pruned_loss=0.1, over 5698584.11 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:52:12,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4860, 5.3177, 5.1076, 2.4279], device='cuda:0'), covar=tensor([0.0480, 0.0700, 0.0774, 0.1693], device='cuda:0'), in_proj_covar=tensor([0.1137, 0.1054, 0.0904, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 02:52:19,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.732e+02 1.329e+03 1.686e+03 2.326e+03 5.611e+03, threshold=3.373e+03, percent-clipped=9.0
+2023-03-09 02:52:31,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=770603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:52:33,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=770606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:52:41,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=770616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:52:47,470 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-09 02:52:50,984 INFO [train.py:968] (0/2) Epoch 17, batch 39750, giga_loss[loss=0.2634, simple_loss=0.3547, pruned_loss=0.08607, over 28910.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3521, pruned_loss=0.1019, over 5718853.23 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3658, pruned_loss=0.1181, over 5724894.54 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3504, pruned_loss=0.1002, over 5709688.93 frames. ], batch size: 145, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:52:56,247 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=770635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:52:57,136 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-09 02:53:11,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=770652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:53:15,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1452, 3.2424, 2.0632, 1.0953], device='cuda:0'), covar=tensor([0.5680, 0.2615, 0.3436, 0.5490], device='cuda:0'), in_proj_covar=tensor([0.1670, 0.1581, 0.1559, 0.1374], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 02:53:32,185 INFO [train.py:968] (0/2) Epoch 17, batch 39800, giga_loss[loss=0.2576, simple_loss=0.3397, pruned_loss=0.08777, over 28947.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3543, pruned_loss=0.1032, over 5719702.29 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3666, pruned_loss=0.1184, over 5727122.20 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.352, pruned_loss=0.1012, over 5710256.83 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:53:42,497 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.155e+02 1.182e+03 1.520e+03 2.065e+03 5.142e+03, threshold=3.040e+03, percent-clipped=6.0
+2023-03-09 02:54:13,896 INFO [train.py:968] (0/2) Epoch 17, batch 39850, giga_loss[loss=0.2783, simple_loss=0.3479, pruned_loss=0.1044, over 28831.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3538, pruned_loss=0.103, over 5720824.74 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3667, pruned_loss=0.1184, over 5729753.58 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3517, pruned_loss=0.1012, over 5710925.93 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:54:22,812 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.20 vs. limit=5.0
+2023-03-09 02:54:39,980 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=770759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:54:41,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=770762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:54:44,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5400, 1.7542, 1.8539, 1.5051], device='cuda:0'), covar=tensor([0.1523, 0.1662, 0.1773, 0.1713], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0744, 0.0699, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 02:54:55,560 INFO [train.py:968] (0/2) Epoch 17, batch 39900, giga_loss[loss=0.2472, simple_loss=0.3273, pruned_loss=0.08352, over 28852.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3529, pruned_loss=0.1024, over 5716635.28 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3667, pruned_loss=0.1184, over 5730380.81 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3511, pruned_loss=0.1009, over 5708258.17 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:55:03,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.887e+02 1.161e+03 1.354e+03 1.596e+03 9.521e+03, threshold=2.707e+03, percent-clipped=4.0
+2023-03-09 02:55:04,517 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=770791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:55:07,189 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=770795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:55:09,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=770798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:55:31,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=770824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:55:33,806 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=770827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:55:34,800 INFO [train.py:968] (0/2) Epoch 17, batch 39950, giga_loss[loss=0.2833, simple_loss=0.3552, pruned_loss=0.1057, over 28998.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3512, pruned_loss=0.1017, over 5717307.80 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3669, pruned_loss=0.1183, over 5732153.73 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3494, pruned_loss=0.1004, over 5708985.87 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:55:47,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4965, 1.6323, 1.2321, 1.2419], device='cuda:0'), covar=tensor([0.0848, 0.0532, 0.1056, 0.1009], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0439, 0.0508, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 02:56:16,767 INFO [train.py:968] (0/2) Epoch 17, batch 40000, giga_loss[loss=0.2174, simple_loss=0.2993, pruned_loss=0.06773, over 29031.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3466, pruned_loss=0.09918, over 5721141.57 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3667, pruned_loss=0.1182, over 5732532.96 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3452, pruned_loss=0.098, over 5714061.34 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:56:26,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.450e+02 1.100e+03 1.448e+03 2.003e+03 4.976e+03, threshold=2.896e+03, percent-clipped=11.0
+2023-03-09 02:56:29,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=770893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:56:40,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=770907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:56:58,966 INFO [train.py:968] (0/2) Epoch 17, batch 40050, giga_loss[loss=0.2443, simple_loss=0.3222, pruned_loss=0.08322, over 28523.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3449, pruned_loss=0.0984, over 5714802.73 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3672, pruned_loss=0.1185, over 5731581.33 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.343, pruned_loss=0.09692, over 5710185.28 frames. ], batch size: 85, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:57:01,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4799, 1.3528, 4.6262, 3.4269], device='cuda:0'), covar=tensor([0.1653, 0.2834, 0.0382, 0.0957], device='cuda:0'), in_proj_covar=tensor([0.0722, 0.0620, 0.0913, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 02:57:27,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=770967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:57:30,692 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=770970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:57:37,505 INFO [train.py:968] (0/2) Epoch 17, batch 40100, giga_loss[loss=0.2336, simple_loss=0.3141, pruned_loss=0.07657, over 28790.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3468, pruned_loss=0.0984, over 5716773.73 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3677, pruned_loss=0.1189, over 5733313.99 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3444, pruned_loss=0.09639, over 5711226.87 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:57:48,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.190e+02 1.154e+03 1.444e+03 2.032e+03 5.987e+03, threshold=2.889e+03, percent-clipped=12.0
+2023-03-09 02:57:56,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=770999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:19,559 INFO [train.py:968] (0/2) Epoch 17, batch 40150, giga_loss[loss=0.2748, simple_loss=0.3446, pruned_loss=0.1025, over 28986.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3492, pruned_loss=0.09964, over 5691858.03 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3684, pruned_loss=0.1195, over 5716811.87 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3459, pruned_loss=0.09688, over 5702552.70 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:58:25,069 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=771036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:27,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=771039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:36,410 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=771050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:38,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=771053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:52,689 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=771068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:59,572 INFO [train.py:968] (0/2) Epoch 17, batch 40200, giga_loss[loss=0.2658, simple_loss=0.3363, pruned_loss=0.09761, over 28813.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3484, pruned_loss=0.1001, over 5691274.20 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3689, pruned_loss=0.1197, over 5706359.68 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3451, pruned_loss=0.09732, over 5709661.88 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:59:01,978 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=771082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:59:08,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.741e+02 1.192e+03 1.640e+03 2.242e+03 8.161e+03, threshold=3.280e+03, percent-clipped=19.0
+2023-03-09 02:59:37,915 INFO [train.py:968] (0/2) Epoch 17, batch 40250, giga_loss[loss=0.285, simple_loss=0.3483, pruned_loss=0.1109, over 28707.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3477, pruned_loss=0.1007, over 5704619.97 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3692, pruned_loss=0.1198, over 5711298.75 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3439, pruned_loss=0.09765, over 5714597.31 frames. ], batch size: 284, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:00:22,473 INFO [train.py:968] (0/2) Epoch 17, batch 40300, giga_loss[loss=0.246, simple_loss=0.3256, pruned_loss=0.08321, over 29113.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3449, pruned_loss=0.1004, over 5699143.44 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3694, pruned_loss=0.1199, over 5707719.03 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3416, pruned_loss=0.09778, over 5710241.58 frames. ], batch size: 155, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:00:32,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3843, 1.5863, 1.4079, 1.2510], device='cuda:0'), covar=tensor([0.2967, 0.2390, 0.1887, 0.2433], device='cuda:0'), in_proj_covar=tensor([0.1871, 0.1819, 0.1734, 0.1875], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 03:00:34,097 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.104e+02 1.127e+03 1.335e+03 1.640e+03 6.029e+03, threshold=2.670e+03, percent-clipped=4.0
+2023-03-09 03:00:42,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5160, 1.6339, 1.7307, 1.3127], device='cuda:0'), covar=tensor([0.1586, 0.2252, 0.1334, 0.1535], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0693, 0.0916, 0.0815], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 03:01:04,707 INFO [train.py:968] (0/2) Epoch 17, batch 40350, giga_loss[loss=0.2852, simple_loss=0.3475, pruned_loss=0.1115, over 28853.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.343, pruned_loss=0.1001, over 5698464.07 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3695, pruned_loss=0.1199, over 5711799.23 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3397, pruned_loss=0.09764, over 5703543.50 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:01:45,378 INFO [train.py:968] (0/2) Epoch 17, batch 40400, giga_loss[loss=0.2617, simple_loss=0.3276, pruned_loss=0.09788, over 29079.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3434, pruned_loss=0.1007, over 5697504.33 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3697, pruned_loss=0.12, over 5707581.97 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3398, pruned_loss=0.09807, over 5705245.76 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:01:55,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.669e+02 1.150e+03 1.445e+03 1.869e+03 4.649e+03, threshold=2.891e+03, percent-clipped=7.0
+2023-03-09 03:01:56,901 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 03:02:23,569 INFO [train.py:968] (0/2) Epoch 17, batch 40450, giga_loss[loss=0.2457, simple_loss=0.3167, pruned_loss=0.08733, over 28715.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3401, pruned_loss=0.09906, over 5704467.75 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3697, pruned_loss=0.1199, over 5713494.64 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3365, pruned_loss=0.09644, over 5705086.23 frames. ], batch size: 99, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:02:38,756 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7037, 2.0413, 1.5743, 1.9362], device='cuda:0'), covar=tensor([0.2658, 0.2636, 0.2921, 0.2348], device='cuda:0'), in_proj_covar=tensor([0.1434, 0.1039, 0.1270, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 03:02:55,370 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=771368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:03:03,528 INFO [train.py:968] (0/2) Epoch 17, batch 40500, giga_loss[loss=0.2308, simple_loss=0.3015, pruned_loss=0.08005, over 28660.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3357, pruned_loss=0.09706, over 5707868.55 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3688, pruned_loss=0.1194, over 5717342.64 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3328, pruned_loss=0.0948, over 5704669.38 frames. ], batch size: 92, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:03:14,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6081, 1.6844, 1.2997, 1.3482], device='cuda:0'), covar=tensor([0.0801, 0.0636, 0.0945, 0.1302], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0441, 0.0509, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 03:03:14,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.571e+02 1.122e+03 1.595e+03 2.087e+03 4.975e+03, threshold=3.190e+03, percent-clipped=14.0
+2023-03-09 03:03:45,806 INFO [train.py:968] (0/2) Epoch 17, batch 40550, giga_loss[loss=0.2921, simple_loss=0.3631, pruned_loss=0.1106, over 28302.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3327, pruned_loss=0.09506, over 5713193.33 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3689, pruned_loss=0.1195, over 5719061.33 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.33, pruned_loss=0.09303, over 5709087.93 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:04:28,144 INFO [train.py:968] (0/2) Epoch 17, batch 40600, giga_loss[loss=0.2547, simple_loss=0.337, pruned_loss=0.08617, over 28723.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3337, pruned_loss=0.09495, over 5700930.13 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.369, pruned_loss=0.1194, over 5701953.84 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3308, pruned_loss=0.09291, over 5711830.16 frames. ], batch size: 284, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:04:39,973 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.505e+02 1.184e+03 1.456e+03 2.010e+03 4.947e+03, threshold=2.912e+03, percent-clipped=2.0
+2023-03-09 03:04:46,429 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-09 03:05:09,260 INFO [train.py:968] (0/2) Epoch 17, batch 40650, giga_loss[loss=0.2856, simple_loss=0.3595, pruned_loss=0.1059, over 28739.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3386, pruned_loss=0.09705, over 5694137.26 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3692, pruned_loss=0.1196, over 5695554.82 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3357, pruned_loss=0.09496, over 5709267.55 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:05:51,804 INFO [train.py:968] (0/2) Epoch 17, batch 40700, giga_loss[loss=0.2643, simple_loss=0.3406, pruned_loss=0.094, over 28960.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3411, pruned_loss=0.09803, over 5698356.20 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3691, pruned_loss=0.1195, over 5698641.58 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3385, pruned_loss=0.09615, over 5707448.24 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:06:02,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.077e+02 1.188e+03 1.699e+03 2.636e+03 1.008e+04, threshold=3.398e+03, percent-clipped=22.0
+2023-03-09 03:06:16,712 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-09 03:06:28,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=771620.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 03:06:35,541 INFO [train.py:968] (0/2) Epoch 17, batch 40750, giga_loss[loss=0.2917, simple_loss=0.3662, pruned_loss=0.1086, over 28253.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3457, pruned_loss=0.1004, over 5694568.39 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3694, pruned_loss=0.1198, over 5701371.94 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.343, pruned_loss=0.09842, over 5699330.58 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:06:40,988 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=771636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:07:16,237 INFO [train.py:968] (0/2) Epoch 17, batch 40800, giga_loss[loss=0.2775, simple_loss=0.3566, pruned_loss=0.09923, over 28939.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3473, pruned_loss=0.1008, over 5705946.28 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3691, pruned_loss=0.1195, over 5706062.52 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3448, pruned_loss=0.09906, over 5705529.05 frames. ], batch size: 174, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:07:29,927 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.757e+02 1.136e+03 1.436e+03 2.067e+03 4.923e+03, threshold=2.872e+03, percent-clipped=6.0
+2023-03-09 03:08:07,988 INFO [train.py:968] (0/2) Epoch 17, batch 40850, giga_loss[loss=0.2697, simple_loss=0.343, pruned_loss=0.09818, over 28377.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.353, pruned_loss=0.1061, over 5697075.50 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3692, pruned_loss=0.1197, over 5705326.95 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3506, pruned_loss=0.1042, over 5697262.76 frames. ], batch size: 65, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:08:22,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=771743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:08:54,074 INFO [train.py:968] (0/2) Epoch 17, batch 40900, giga_loss[loss=0.2871, simple_loss=0.364, pruned_loss=0.1051, over 28978.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3576, pruned_loss=0.1094, over 5700922.93 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3697, pruned_loss=0.12, over 5706813.12 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3549, pruned_loss=0.1073, over 5699865.77 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:09:07,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.171e+03 1.800e+03 2.384e+03 3.629e+03 1.106e+04, threshold=4.768e+03, percent-clipped=39.0
+2023-03-09 03:09:40,510 INFO [train.py:968] (0/2) Epoch 17, batch 40950, giga_loss[loss=0.413, simple_loss=0.4462, pruned_loss=0.1899, over 28725.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.367, pruned_loss=0.1164, over 5697705.42 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3697, pruned_loss=0.1201, over 5710936.49 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3647, pruned_loss=0.1145, over 5692891.90 frames. ], batch size: 262, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:09:51,198 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=771841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:10:25,721 INFO [train.py:968] (0/2) Epoch 17, batch 41000, giga_loss[loss=0.3548, simple_loss=0.4084, pruned_loss=0.1506, over 28864.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.372, pruned_loss=0.1208, over 5699272.63 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3696, pruned_loss=0.12, over 5712107.96 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3703, pruned_loss=0.1193, over 5694349.13 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:10:31,863 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=771886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:10:34,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=771889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:10:37,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.208e+03 1.705e+03 2.200e+03 3.024e+03 5.745e+03, threshold=4.400e+03, percent-clipped=3.0
+2023-03-09 03:10:58,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5350, 1.6584, 1.7675, 1.3184], device='cuda:0'), covar=tensor([0.1567, 0.2331, 0.1278, 0.1582], device='cuda:0'), in_proj_covar=tensor([0.0868, 0.0691, 0.0915, 0.0813], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 03:10:58,632 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4825, 1.7138, 1.6594, 1.3645], device='cuda:0'), covar=tensor([0.1995, 0.1843, 0.1328, 0.1696], device='cuda:0'), in_proj_covar=tensor([0.1874, 0.1821, 0.1732, 0.1876], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 03:11:00,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=771918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:11:08,693 INFO [train.py:968] (0/2) Epoch 17, batch 41050, giga_loss[loss=0.3369, simple_loss=0.3928, pruned_loss=0.1405, over 28689.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3785, pruned_loss=0.1263, over 5689960.42 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3695, pruned_loss=0.1199, over 5705488.10 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3774, pruned_loss=0.1253, over 5692339.46 frames. ], batch size: 92, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:11:17,767 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=771937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:11:59,989 INFO [train.py:968] (0/2) Epoch 17, batch 41100, giga_loss[loss=0.2923, simple_loss=0.3599, pruned_loss=0.1123, over 28883.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3831, pruned_loss=0.1306, over 5670337.51 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3693, pruned_loss=0.1197, over 5709277.46 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3826, pruned_loss=0.1303, over 5668000.72 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:12:14,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 1.635e+03 2.325e+03 2.806e+03 5.499e+03, threshold=4.651e+03, percent-clipped=3.0
+2023-03-09 03:12:15,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=771995.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 03:12:19,213 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-772000.pt
+2023-03-09 03:12:32,374 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=772011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:12:49,061 INFO [train.py:968] (0/2) Epoch 17, batch 41150, giga_loss[loss=0.3914, simple_loss=0.4274, pruned_loss=0.1777, over 27543.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3844, pruned_loss=0.1327, over 5671393.61 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3686, pruned_loss=0.1192, over 5716400.04 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3851, pruned_loss=0.1333, over 5661743.17 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:13:42,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3933, 1.9675, 1.6240, 1.6329], device='cuda:0'), covar=tensor([0.0638, 0.0245, 0.0260, 0.0642], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-09 03:13:46,585 INFO [train.py:968] (0/2) Epoch 17, batch 41200, giga_loss[loss=0.2821, simple_loss=0.3497, pruned_loss=0.1073, over 28577.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.386, pruned_loss=0.1354, over 5655842.58 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3684, pruned_loss=0.1191, over 5708816.25 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3871, pruned_loss=0.1362, over 5654062.85 frames. ], batch size: 65, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:14:02,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.722e+03 2.194e+03 2.878e+03 7.509e+03, threshold=4.387e+03, percent-clipped=8.0
+2023-03-09 03:14:40,420 INFO [train.py:968] (0/2) Epoch 17, batch 41250, giga_loss[loss=0.3447, simple_loss=0.408, pruned_loss=0.1407, over 28939.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3893, pruned_loss=0.139, over 5632414.17 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3682, pruned_loss=0.119, over 5701463.38 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3905, pruned_loss=0.1398, over 5637850.98 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:14:52,705 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772138.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 03:14:55,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772141.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 03:15:06,044 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:15:09,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:15:22,898 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=772170.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 03:15:32,036 INFO [train.py:968] (0/2) Epoch 17, batch 41300, giga_loss[loss=0.2808, simple_loss=0.3548, pruned_loss=0.1034, over 28599.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3926, pruned_loss=0.1416, over 5619231.77 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3684, pruned_loss=0.1193, over 5693744.09 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3939, pruned_loss=0.1427, over 5628972.94 frames. ], batch size: 85, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:15:38,298 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=772186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:15:48,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.254e+03 2.183e+03 2.771e+03 4.537e+03 1.422e+04, threshold=5.542e+03, percent-clipped=26.0
+2023-03-09 03:16:14,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=772216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:16:29,757 INFO [train.py:968] (0/2) Epoch 17, batch 41350, giga_loss[loss=0.4061, simple_loss=0.4301, pruned_loss=0.1911, over 26580.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3939, pruned_loss=0.1439, over 5611097.62 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3682, pruned_loss=0.1192, over 5694061.05 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3954, pruned_loss=0.1449, over 5617889.75 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:16:35,813 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=772236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:16:55,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4303, 1.2633, 1.1420, 1.5753], device='cuda:0'), covar=tensor([0.0761, 0.0339, 0.0329, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-09 03:17:15,187 INFO [train.py:968] (0/2) Epoch 17, batch 41400, giga_loss[loss=0.3072, simple_loss=0.3768, pruned_loss=0.1188, over 28863.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3914, pruned_loss=0.1418, over 5627917.13 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3686, pruned_loss=0.1196, over 5693653.83 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3929, pruned_loss=0.1431, over 5631736.93 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:17:27,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.639e+03 1.978e+03 2.629e+03 5.407e+03, threshold=3.957e+03, percent-clipped=0.0
+2023-03-09 03:17:42,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=772312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:17:59,469 INFO [train.py:968] (0/2) Epoch 17, batch 41450, giga_loss[loss=0.3287, simple_loss=0.3923, pruned_loss=0.1326, over 28746.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3887, pruned_loss=0.139, over 5651595.33 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3679, pruned_loss=0.1194, over 5702980.46 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3917, pruned_loss=0.1412, over 5643584.56 frames. ], batch size: 284, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:18:07,929 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-09 03:18:11,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2137, 1.4002, 1.3357, 1.1370], device='cuda:0'), covar=tensor([0.2281, 0.2254, 0.1437, 0.1964], device='cuda:0'), in_proj_covar=tensor([0.1900, 0.1840, 0.1748, 0.1895], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 03:18:29,911 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:18:35,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:18:37,085 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=772364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:18:50,448 INFO [train.py:968] (0/2) Epoch 17, batch 41500, giga_loss[loss=0.4018, simple_loss=0.4349, pruned_loss=0.1844, over 27431.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3875, pruned_loss=0.1368, over 5653886.33 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5704094.44 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3908, pruned_loss=0.1392, over 5644823.28 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:19:04,300 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=772391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:19:06,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.846e+02 1.691e+03 2.049e+03 2.860e+03 5.758e+03, threshold=4.097e+03, percent-clipped=9.0
+2023-03-09 03:19:47,512 INFO [train.py:968] (0/2) Epoch 17, batch 41550, giga_loss[loss=0.3628, simple_loss=0.4146, pruned_loss=0.1554, over 28576.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3901, pruned_loss=0.1378, over 5667498.09 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5707209.24 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.393, pruned_loss=0.1399, over 5657132.89 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:20:13,359 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:20:15,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:20:37,039 INFO [train.py:968] (0/2) Epoch 17, batch 41600, giga_loss[loss=0.4281, simple_loss=0.4475, pruned_loss=0.2043, over 26521.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3893, pruned_loss=0.1376, over 5648779.02 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3672, pruned_loss=0.1193, over 5705201.40 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3928, pruned_loss=0.14, over 5640374.49 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:20:44,429 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=772487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:20:50,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.214e+03 1.762e+03 2.214e+03 3.166e+03 1.006e+04, threshold=4.427e+03, percent-clipped=10.0
+2023-03-09 03:21:25,334 INFO [train.py:968] (0/2) Epoch 17, batch 41650, giga_loss[loss=0.2909, simple_loss=0.3624, pruned_loss=0.1097, over 28619.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3878, pruned_loss=0.1351, over 5652405.30 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3675, pruned_loss=0.1194, over 5707904.44 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3908, pruned_loss=0.1373, over 5642369.23 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:21:52,290 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.30 vs. limit=5.0
+2023-03-09 03:22:17,991 INFO [train.py:968] (0/2) Epoch 17, batch 41700, giga_loss[loss=0.3498, simple_loss=0.3989, pruned_loss=0.1503, over 28270.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3839, pruned_loss=0.1309, over 5661238.89 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3673, pruned_loss=0.1192, over 5709662.23 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3867, pruned_loss=0.133, over 5651308.79 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:22:39,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.598e+03 2.090e+03 2.781e+03 6.601e+03, threshold=4.180e+03, percent-clipped=3.0
+2023-03-09 03:22:54,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=772611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:23:14,354 INFO [train.py:968] (0/2) Epoch 17, batch 41750, giga_loss[loss=0.3005, simple_loss=0.3678, pruned_loss=0.1166, over 28935.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3798, pruned_loss=0.1276, over 5659410.98 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3671, pruned_loss=0.1192, over 5712697.06 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3823, pruned_loss=0.1294, over 5648157.67 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:23:14,675 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6308, 1.8326, 1.8688, 1.6186], device='cuda:0'), covar=tensor([0.1994, 0.1775, 0.1289, 0.1567], device='cuda:0'), in_proj_covar=tensor([0.1890, 0.1823, 0.1737, 0.1883], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 03:24:02,466 INFO [train.py:968] (0/2) Epoch 17, batch 41800, giga_loss[loss=0.257, simple_loss=0.3329, pruned_loss=0.09053, over 28882.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3773, pruned_loss=0.1262, over 5656909.60 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3672, pruned_loss=0.1192, over 5715079.40 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3797, pruned_loss=0.1279, over 5644010.40 frames. ], batch size: 99, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:24:20,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.907e+02 1.837e+03 2.444e+03 3.659e+03 1.480e+04, threshold=4.888e+03, percent-clipped=18.0
+2023-03-09 03:24:53,377 INFO [train.py:968] (0/2) Epoch 17, batch 41850, libri_loss[loss=0.3632, simple_loss=0.4012, pruned_loss=0.1626, over 29753.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3773, pruned_loss=0.1262, over 5663504.08 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3673, pruned_loss=0.1193, over 5717485.27 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3792, pruned_loss=0.1275, over 5649645.27 frames. ], batch size: 87, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:25:01,686 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=772739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:25:15,264 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:25:18,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:25:41,626 INFO [train.py:968] (0/2) Epoch 17, batch 41900, giga_loss[loss=0.2774, simple_loss=0.3577, pruned_loss=0.09857, over 28995.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3767, pruned_loss=0.1253, over 5665776.27 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3674, pruned_loss=0.1194, over 5711858.41 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3783, pruned_loss=0.1264, over 5659584.66 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:25:48,866 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=772786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:25:56,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.408e+02 1.599e+03 1.905e+03 2.530e+03 5.909e+03, threshold=3.811e+03, percent-clipped=2.0
+2023-03-09 03:26:19,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-09 03:26:20,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9677, 4.7864, 4.5535, 2.1630], device='cuda:0'), covar=tensor([0.0461, 0.0613, 0.0669, 0.1929], device='cuda:0'), in_proj_covar=tensor([0.1168, 0.1085, 0.0927, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 03:26:32,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5232, 1.6895, 1.7446, 1.3276], device='cuda:0'), covar=tensor([0.1778, 0.2399, 0.1448, 0.1739], device='cuda:0'), in_proj_covar=tensor([0.0868, 0.0696, 0.0914, 0.0814], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 03:26:33,988 INFO [train.py:968] (0/2) Epoch 17, batch 41950, giga_loss[loss=0.2933, simple_loss=0.3652, pruned_loss=0.1107, over 28789.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3729, pruned_loss=0.1218, over 5666815.98 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.367, pruned_loss=0.1193, over 5706711.80 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3747, pruned_loss=0.1228, over 5665780.96 frames. ], batch size: 243, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:27:24,118 INFO [train.py:968] (0/2) Epoch 17, batch 42000, giga_loss[loss=0.272, simple_loss=0.3553, pruned_loss=0.09435, over 28875.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3742, pruned_loss=0.1205, over 5674912.76 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3669, pruned_loss=0.1193, over 5708920.12 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3759, pruned_loss=0.1214, over 5671242.11 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:27:24,122 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 03:27:33,055 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2056, simple_loss=0.3104, pruned_loss=0.05038, over 944034.00 frames. 
+2023-03-09 03:27:33,056 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 03:27:35,714 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:27:39,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:27:44,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4104, 1.2203, 3.7421, 3.1933], device='cuda:0'), covar=tensor([0.1552, 0.2746, 0.0484, 0.1098], device='cuda:0'), in_proj_covar=tensor([0.0728, 0.0628, 0.0926, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 03:27:50,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.591e+02 1.418e+03 1.760e+03 2.252e+03 5.838e+03, threshold=3.520e+03, percent-clipped=4.0
+2023-03-09 03:28:10,242 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=772914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:28:22,482 INFO [train.py:968] (0/2) Epoch 17, batch 42050, giga_loss[loss=0.3662, simple_loss=0.4094, pruned_loss=0.1616, over 26531.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3749, pruned_loss=0.1202, over 5674264.74 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3665, pruned_loss=0.1192, over 5710929.55 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3768, pruned_loss=0.1211, over 5668527.40 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:28:24,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4362, 1.7229, 1.3695, 1.5720], device='cuda:0'), covar=tensor([0.2497, 0.2476, 0.2730, 0.2247], device='cuda:0'), in_proj_covar=tensor([0.1432, 0.1039, 0.1271, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 03:29:08,773 INFO [train.py:968] (0/2) Epoch 17, batch 42100, giga_loss[loss=0.299, simple_loss=0.37, pruned_loss=0.114, over 28960.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3754, pruned_loss=0.1216, over 5667638.54 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3661, pruned_loss=0.1188, over 5708382.16 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3777, pruned_loss=0.1227, over 5663672.95 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:29:23,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.145e+03 1.965e+03 2.524e+03 3.155e+03 9.309e+03, threshold=5.048e+03, percent-clipped=17.0
+2023-03-09 03:29:26,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=772999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:29:28,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7933, 2.2616, 1.6802, 2.1668], device='cuda:0'), covar=tensor([0.2406, 0.2323, 0.2779, 0.2183], device='cuda:0'), in_proj_covar=tensor([0.1430, 0.1036, 0.1269, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 03:29:54,723 INFO [train.py:968] (0/2) Epoch 17, batch 42150, giga_loss[loss=0.2988, simple_loss=0.37, pruned_loss=0.1138, over 28649.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3753, pruned_loss=0.1219, over 5674617.65 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3658, pruned_loss=0.1184, over 5713259.97 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3777, pruned_loss=0.1231, over 5666237.14 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:29:56,179 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=773031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:30:36,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=773075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:30:40,612 INFO [train.py:968] (0/2) Epoch 17, batch 42200, giga_loss[loss=0.27, simple_loss=0.3397, pruned_loss=0.1001, over 28575.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3733, pruned_loss=0.1215, over 5678474.32 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3655, pruned_loss=0.1183, over 5716403.70 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3755, pruned_loss=0.1226, over 5668623.52 frames. ], batch size: 85, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:30:58,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.068e+03 1.689e+03 2.006e+03 2.622e+03 9.428e+03, threshold=4.011e+03, percent-clipped=7.0
+2023-03-09 03:31:27,683 INFO [train.py:968] (0/2) Epoch 17, batch 42250, giga_loss[loss=0.3146, simple_loss=0.3844, pruned_loss=0.1223, over 28719.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3717, pruned_loss=0.121, over 5672708.60 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3653, pruned_loss=0.1181, over 5717867.30 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3739, pruned_loss=0.1223, over 5662444.25 frames. ], batch size: 66, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:31:42,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3495, 1.2478, 3.8852, 3.2922], device='cuda:0'), covar=tensor([0.1606, 0.2740, 0.0456, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0727, 0.0627, 0.0922, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 03:32:00,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1599, 1.4334, 1.4092, 1.2902], device='cuda:0'), covar=tensor([0.1963, 0.1868, 0.2530, 0.2063], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0744, 0.0698, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 03:32:16,731 INFO [train.py:968] (0/2) Epoch 17, batch 42300, giga_loss[loss=0.2952, simple_loss=0.3598, pruned_loss=0.1153, over 28866.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3709, pruned_loss=0.1198, over 5664668.77 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3655, pruned_loss=0.1182, over 5709350.33 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3726, pruned_loss=0.1207, over 5663444.88 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:32:31,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.252e+02 1.595e+03 2.097e+03 2.769e+03 9.175e+03, threshold=4.193e+03, percent-clipped=9.0
+2023-03-09 03:32:51,324 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 03:32:59,141 INFO [train.py:968] (0/2) Epoch 17, batch 42350, giga_loss[loss=0.2818, simple_loss=0.3654, pruned_loss=0.09907, over 28927.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3707, pruned_loss=0.1185, over 5685446.74 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3648, pruned_loss=0.1177, over 5715817.83 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3729, pruned_loss=0.1197, over 5677487.03 frames. ], batch size: 174, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:33:33,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3155, 1.6496, 1.2822, 0.9822], device='cuda:0'), covar=tensor([0.2507, 0.2503, 0.2947, 0.2201], device='cuda:0'), in_proj_covar=tensor([0.1434, 0.1040, 0.1273, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 03:33:48,928 INFO [train.py:968] (0/2) Epoch 17, batch 42400, giga_loss[loss=0.3306, simple_loss=0.3861, pruned_loss=0.1375, over 27932.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3719, pruned_loss=0.1188, over 5685718.12 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3645, pruned_loss=0.1175, over 5718556.86 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.374, pruned_loss=0.12, over 5676449.80 frames. ], batch size: 412, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:34:07,125 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.472e+02 1.689e+03 2.186e+03 2.880e+03 9.496e+03, threshold=4.371e+03, percent-clipped=7.0
+2023-03-09 03:34:12,975 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-09 03:34:39,650 INFO [train.py:968] (0/2) Epoch 17, batch 42450, giga_loss[loss=0.2729, simple_loss=0.3416, pruned_loss=0.1021, over 28412.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3709, pruned_loss=0.1185, over 5687137.79 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3647, pruned_loss=0.1175, over 5719833.17 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3725, pruned_loss=0.1195, over 5678429.19 frames. ], batch size: 71, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:35:19,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=773374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:35:24,933 INFO [train.py:968] (0/2) Epoch 17, batch 42500, giga_loss[loss=0.3438, simple_loss=0.3987, pruned_loss=0.1445, over 28672.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3701, pruned_loss=0.1187, over 5682292.11 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3642, pruned_loss=0.1171, over 5720553.74 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3719, pruned_loss=0.1199, over 5673580.22 frames. ], batch size: 284, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:35:43,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.493e+02 1.631e+03 1.899e+03 2.759e+03 8.365e+03, threshold=3.797e+03, percent-clipped=6.0
+2023-03-09 03:35:52,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=773406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:36:17,616 INFO [train.py:968] (0/2) Epoch 17, batch 42550, giga_loss[loss=0.2571, simple_loss=0.3367, pruned_loss=0.08872, over 29024.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3696, pruned_loss=0.1195, over 5680497.58 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3646, pruned_loss=0.1176, over 5724369.13 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3708, pruned_loss=0.1201, over 5669608.26 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:36:27,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6384, 1.8036, 1.5478, 1.8115], device='cuda:0'), covar=tensor([0.2061, 0.2054, 0.2047, 0.1935], device='cuda:0'), in_proj_covar=tensor([0.1431, 0.1038, 0.1269, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 03:36:36,362 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=773450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:37:03,002 INFO [train.py:968] (0/2) Epoch 17, batch 42600, giga_loss[loss=0.2521, simple_loss=0.3292, pruned_loss=0.08744, over 28465.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3698, pruned_loss=0.1205, over 5664069.30 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3649, pruned_loss=0.1177, over 5713377.59 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3706, pruned_loss=0.1209, over 5663557.01 frames. ], batch size: 60, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:37:20,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.678e+03 2.228e+03 3.214e+03 7.413e+03, threshold=4.457e+03, percent-clipped=19.0
+2023-03-09 03:37:38,109 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=773517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:37:42,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=773520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:37:50,420 INFO [train.py:968] (0/2) Epoch 17, batch 42650, giga_loss[loss=0.2773, simple_loss=0.3501, pruned_loss=0.1023, over 28666.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3673, pruned_loss=0.1193, over 5669784.61 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3642, pruned_loss=0.1173, over 5708067.45 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3688, pruned_loss=0.1201, over 5672772.88 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:37:54,953 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-09 03:38:11,065 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=773549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:11,122 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=773549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:13,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=773552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:39,827 INFO [train.py:968] (0/2) Epoch 17, batch 42700, giga_loss[loss=0.3171, simple_loss=0.3776, pruned_loss=0.1283, over 28823.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3674, pruned_loss=0.1196, over 5678523.48 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1177, over 5714043.18 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.1199, over 5674523.53 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:38:42,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=773581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:43,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2943, 1.8118, 1.4601, 1.5172], device='cuda:0'), covar=tensor([0.0783, 0.0296, 0.0319, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:0')
+2023-03-09 03:38:52,352 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=773593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:55,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=773596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:58,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.677e+02 1.676e+03 2.070e+03 2.867e+03 7.585e+03, threshold=4.141e+03, percent-clipped=8.0
+2023-03-09 03:39:14,650 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-09 03:39:24,546 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=773625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:39:24,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8225, 2.0474, 1.4733, 1.5364], device='cuda:0'), covar=tensor([0.0946, 0.0637, 0.1013, 0.1157], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0445, 0.0509, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 03:39:28,056 INFO [train.py:968] (0/2) Epoch 17, batch 42750, giga_loss[loss=0.3013, simple_loss=0.3721, pruned_loss=0.1153, over 29022.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.368, pruned_loss=0.1198, over 5673837.88 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3646, pruned_loss=0.1175, over 5706611.76 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3688, pruned_loss=0.1203, over 5677066.20 frames. ], batch size: 106, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 03:40:10,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=773672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:40:17,436 INFO [train.py:968] (0/2) Epoch 17, batch 42800, libri_loss[loss=0.3268, simple_loss=0.3847, pruned_loss=0.1344, over 27554.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3683, pruned_loss=0.1192, over 5670941.48 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3646, pruned_loss=0.1174, over 5707579.75 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.369, pruned_loss=0.1196, over 5672050.23 frames. ], batch size: 115, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:40:32,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.197e+02 1.544e+03 2.108e+03 3.169e+03 1.678e+04, threshold=4.217e+03, percent-clipped=16.0
+2023-03-09 03:41:01,647 INFO [train.py:968] (0/2) Epoch 17, batch 42850, giga_loss[loss=0.3, simple_loss=0.3663, pruned_loss=0.1169, over 28629.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3684, pruned_loss=0.1184, over 5675209.10 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3649, pruned_loss=0.1176, over 5710642.87 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3688, pruned_loss=0.1187, over 5672677.97 frames. ], batch size: 92, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:41:40,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3528, 1.5420, 1.6137, 1.1917], device='cuda:0'), covar=tensor([0.1610, 0.2263, 0.1288, 0.1553], device='cuda:0'), in_proj_covar=tensor([0.0872, 0.0698, 0.0918, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 03:41:47,938 INFO [train.py:968] (0/2) Epoch 17, batch 42900, giga_loss[loss=0.2856, simple_loss=0.3582, pruned_loss=0.1065, over 28836.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3693, pruned_loss=0.119, over 5664337.02 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3645, pruned_loss=0.1173, over 5702006.73 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3701, pruned_loss=0.1195, over 5669748.77 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:41:48,100 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=773779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:42:05,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.284e+02 1.529e+03 2.200e+03 2.759e+03 1.294e+04, threshold=4.400e+03, percent-clipped=10.0
+2023-03-09 03:42:37,270 INFO [train.py:968] (0/2) Epoch 17, batch 42950, giga_loss[loss=0.3594, simple_loss=0.4099, pruned_loss=0.1545, over 27891.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3706, pruned_loss=0.1206, over 5661602.04 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3646, pruned_loss=0.1174, over 5703630.77 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3713, pruned_loss=0.121, over 5663695.33 frames. ], batch size: 412, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:43:18,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4392, 1.5825, 1.6556, 1.3936], device='cuda:0'), covar=tensor([0.1793, 0.2010, 0.2344, 0.2201], device='cuda:0'), in_proj_covar=tensor([0.0462, 0.0745, 0.0700, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 03:43:27,991 INFO [train.py:968] (0/2) Epoch 17, batch 43000, giga_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.121, over 28924.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3742, pruned_loss=0.1242, over 5666994.22 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3647, pruned_loss=0.1172, over 5708116.78 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3748, pruned_loss=0.1248, over 5663875.56 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:43:47,973 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.699e+03 2.249e+03 2.999e+03 9.913e+03, threshold=4.498e+03, percent-clipped=8.0
+2023-03-09 03:44:20,247 INFO [train.py:968] (0/2) Epoch 17, batch 43050, giga_loss[loss=0.2662, simple_loss=0.3369, pruned_loss=0.09781, over 28449.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3758, pruned_loss=0.1272, over 5651151.23 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3649, pruned_loss=0.1175, over 5702328.61 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3763, pruned_loss=0.1276, over 5652839.37 frames. ], batch size: 65, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 03:45:12,472 INFO [train.py:968] (0/2) Epoch 17, batch 43100, giga_loss[loss=0.2953, simple_loss=0.3571, pruned_loss=0.1167, over 28910.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3767, pruned_loss=0.1288, over 5655039.11 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1176, over 5705345.65 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3774, pruned_loss=0.1292, over 5653121.25 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 03:45:32,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.113e+03 1.786e+03 2.626e+03 3.670e+03 9.508e+03, threshold=5.252e+03, percent-clipped=18.0
+2023-03-09 03:45:33,536 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-774000.pt
+2023-03-09 03:46:01,681 INFO [train.py:968] (0/2) Epoch 17, batch 43150, giga_loss[loss=0.3063, simple_loss=0.3706, pruned_loss=0.121, over 28303.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3748, pruned_loss=0.1272, over 5668638.39 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3646, pruned_loss=0.1174, over 5706221.93 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3756, pruned_loss=0.1277, over 5666119.09 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 03:46:18,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=774047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:46:32,275 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-09 03:46:47,985 INFO [train.py:968] (0/2) Epoch 17, batch 43200, giga_loss[loss=0.3222, simple_loss=0.3915, pruned_loss=0.1265, over 29043.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3727, pruned_loss=0.1252, over 5669985.87 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1176, over 5706309.88 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3732, pruned_loss=0.1256, over 5666855.67 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:47:05,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.502e+03 1.871e+03 2.385e+03 5.479e+03, threshold=3.742e+03, percent-clipped=2.0
+2023-03-09 03:47:23,670 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-09 03:47:31,317 INFO [train.py:968] (0/2) Epoch 17, batch 43250, giga_loss[loss=0.3357, simple_loss=0.3959, pruned_loss=0.1378, over 28686.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1218, over 5672958.02 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3648, pruned_loss=0.1175, over 5699134.99 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3716, pruned_loss=0.1224, over 5676420.99 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:47:53,259 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 03:47:58,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=774154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:48:20,884 INFO [train.py:968] (0/2) Epoch 17, batch 43300, giga_loss[loss=0.3158, simple_loss=0.3652, pruned_loss=0.1332, over 28642.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3681, pruned_loss=0.1202, over 5666160.02 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.365, pruned_loss=0.1176, over 5701192.52 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3686, pruned_loss=0.1206, over 5666994.92 frames. ], batch size: 92, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:48:29,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5410, 1.7855, 1.3989, 1.7035], device='cuda:0'), covar=tensor([0.2783, 0.2811, 0.3027, 0.2697], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1042, 0.1276, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 03:48:33,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=774190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:48:36,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=774193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:48:40,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.504e+03 1.855e+03 2.707e+03 6.379e+03, threshold=3.709e+03, percent-clipped=7.0
+2023-03-09 03:48:46,709 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-09 03:49:01,689 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=774222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:49:07,991 INFO [train.py:968] (0/2) Epoch 17, batch 43350, giga_loss[loss=0.296, simple_loss=0.3595, pruned_loss=0.1163, over 28815.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3677, pruned_loss=0.1209, over 5662195.65 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3655, pruned_loss=0.1179, over 5698067.89 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3676, pruned_loss=0.121, over 5665383.37 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:49:10,457 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=774231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:49:21,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6875, 1.6366, 1.2447, 1.2525], device='cuda:0'), covar=tensor([0.0836, 0.0615, 0.1065, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0379, 0.0445, 0.0509, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 03:49:58,044 INFO [train.py:968] (0/2) Epoch 17, batch 43400, giga_loss[loss=0.3074, simple_loss=0.3757, pruned_loss=0.1195, over 28825.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3673, pruned_loss=0.1212, over 5657550.22 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3659, pruned_loss=0.1181, over 5701689.03 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3669, pruned_loss=0.1211, over 5656162.83 frames. ], batch size: 284, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:50:13,049 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=774297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:50:14,027 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.959e+02 1.664e+03 2.170e+03 2.730e+03 5.367e+03, threshold=4.341e+03, percent-clipped=7.0
+2023-03-09 03:50:15,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=774300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:50:25,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6234, 1.8214, 1.4727, 1.7061], device='cuda:0'), covar=tensor([0.2484, 0.2498, 0.2757, 0.2368], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1042, 0.1276, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 03:50:40,407 INFO [train.py:968] (0/2) Epoch 17, batch 43450, giga_loss[loss=0.4223, simple_loss=0.4412, pruned_loss=0.2017, over 26453.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3695, pruned_loss=0.1226, over 5662391.21 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3655, pruned_loss=0.118, over 5698644.44 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3696, pruned_loss=0.1227, over 5663670.27 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:50:40,647 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=774329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:51:12,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1648, 1.3085, 1.0809, 0.9668], device='cuda:0'), covar=tensor([0.0974, 0.0524, 0.1108, 0.1009], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0447, 0.0510, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 03:51:26,134 INFO [train.py:968] (0/2) Epoch 17, batch 43500, giga_loss[loss=0.3156, simple_loss=0.391, pruned_loss=0.1201, over 28566.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3745, pruned_loss=0.1242, over 5668532.50 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3654, pruned_loss=0.118, over 5703579.85 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3749, pruned_loss=0.1244, over 5664465.79 frames. ], batch size: 336, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:51:47,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.585e+03 2.041e+03 2.819e+03 8.917e+03, threshold=4.082e+03, percent-clipped=4.0
+2023-03-09 03:52:22,407 INFO [train.py:968] (0/2) Epoch 17, batch 43550, giga_loss[loss=0.2798, simple_loss=0.3579, pruned_loss=0.1009, over 29094.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3767, pruned_loss=0.1233, over 5662252.95 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3653, pruned_loss=0.118, over 5703690.95 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3771, pruned_loss=0.1235, over 5658794.20 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:53:10,621 INFO [train.py:968] (0/2) Epoch 17, batch 43600, giga_loss[loss=0.3178, simple_loss=0.3792, pruned_loss=0.1282, over 28520.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3782, pruned_loss=0.1245, over 5669897.75 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3652, pruned_loss=0.1178, over 5705395.70 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3788, pruned_loss=0.125, over 5665032.86 frames. ], batch size: 85, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:53:27,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.658e+02 1.759e+03 2.342e+03 3.386e+03 8.910e+03, threshold=4.683e+03, percent-clipped=19.0
+2023-03-09 03:53:56,746 INFO [train.py:968] (0/2) Epoch 17, batch 43650, giga_loss[loss=0.3841, simple_loss=0.4332, pruned_loss=0.1674, over 27546.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3794, pruned_loss=0.1261, over 5675596.91 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1177, over 5712483.07 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3806, pruned_loss=0.1268, over 5664208.23 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:54:39,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7709, 2.0105, 1.6127, 2.0390], device='cuda:0'), covar=tensor([0.2444, 0.2545, 0.2845, 0.2366], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1044, 0.1277, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 03:54:43,682 INFO [train.py:968] (0/2) Epoch 17, batch 43700, giga_loss[loss=0.3896, simple_loss=0.4155, pruned_loss=0.1818, over 23425.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3796, pruned_loss=0.127, over 5671456.60 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3653, pruned_loss=0.1179, over 5706923.64 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3806, pruned_loss=0.1276, over 5666349.54 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:55:01,482 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.185e+02 1.685e+03 2.206e+03 3.235e+03 9.146e+03, threshold=4.413e+03, percent-clipped=6.0
+2023-03-09 03:55:09,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=774606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:55:32,555 INFO [train.py:968] (0/2) Epoch 17, batch 43750, giga_loss[loss=0.3907, simple_loss=0.4089, pruned_loss=0.1863, over 23653.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3789, pruned_loss=0.1276, over 5665459.40 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3652, pruned_loss=0.1178, over 5707656.34 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.38, pruned_loss=0.1283, over 5660255.71 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:55:39,948 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=774637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:56:04,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2076, 1.2503, 1.0942, 0.9006], device='cuda:0'), covar=tensor([0.0927, 0.0523, 0.1069, 0.1010], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0445, 0.0508, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 03:56:16,927 INFO [train.py:968] (0/2) Epoch 17, batch 43800, giga_loss[loss=0.3262, simple_loss=0.3803, pruned_loss=0.1361, over 29046.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3756, pruned_loss=0.1258, over 5677806.77 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3654, pruned_loss=0.1181, over 5715594.88 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3768, pruned_loss=0.1265, over 5664842.65 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:56:35,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.745e+03 2.367e+03 3.027e+03 7.410e+03, threshold=4.735e+03, percent-clipped=10.0
+2023-03-09 03:57:03,543 INFO [train.py:968] (0/2) Epoch 17, batch 43850, giga_loss[loss=0.3346, simple_loss=0.3939, pruned_loss=0.1377, over 28562.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3747, pruned_loss=0.1258, over 5678039.24 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3652, pruned_loss=0.1179, over 5719970.80 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3761, pruned_loss=0.1268, over 5662370.30 frames. ], batch size: 71, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:57:11,718 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=774738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:57:22,630 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=774749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:57:27,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=774752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:57:53,298 INFO [train.py:968] (0/2) Epoch 17, batch 43900, giga_loss[loss=0.3, simple_loss=0.3734, pruned_loss=0.1133, over 28973.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3752, pruned_loss=0.1268, over 5668288.67 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3648, pruned_loss=0.1175, over 5725913.78 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.377, pruned_loss=0.1281, over 5648910.60 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:57:55,176 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=774781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:58:14,719 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.091e+02 1.582e+03 2.042e+03 2.930e+03 8.367e+03, threshold=4.085e+03, percent-clipped=8.0
+2023-03-09 03:58:15,048 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8552, 2.2712, 1.9473, 1.5297], device='cuda:0'), covar=tensor([0.3185, 0.2131, 0.2363, 0.2884], device='cuda:0'), in_proj_covar=tensor([0.1896, 0.1824, 0.1753, 0.1888], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 03:58:43,542 INFO [train.py:968] (0/2) Epoch 17, batch 43950, giga_loss[loss=0.2738, simple_loss=0.3455, pruned_loss=0.101, over 28774.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3736, pruned_loss=0.1257, over 5666413.68 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3648, pruned_loss=0.1175, over 5728470.21 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3751, pruned_loss=0.127, over 5647699.09 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:59:08,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2632, 1.4161, 1.5179, 1.2927], device='cuda:0'), covar=tensor([0.1700, 0.1712, 0.2184, 0.1810], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0744, 0.0699, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 03:59:21,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2205, 1.8039, 1.4047, 0.4018], device='cuda:0'), covar=tensor([0.4196, 0.2518, 0.3817, 0.5789], device='cuda:0'), in_proj_covar=tensor([0.1687, 0.1603, 0.1569, 0.1384], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 03:59:27,540 INFO [train.py:968] (0/2) Epoch 17, batch 44000, giga_loss[loss=0.3805, simple_loss=0.408, pruned_loss=0.1765, over 26661.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3732, pruned_loss=0.126, over 5680749.76 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3649, pruned_loss=0.1175, over 5734164.00 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3747, pruned_loss=0.1273, over 5658617.20 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:59:42,503 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=774894.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 03:59:48,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.831e+03 2.333e+03 3.042e+03 6.166e+03, threshold=4.666e+03, percent-clipped=9.0
+2023-03-09 04:00:13,924 INFO [train.py:968] (0/2) Epoch 17, batch 44050, giga_loss[loss=0.3105, simple_loss=0.378, pruned_loss=0.1215, over 28847.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3732, pruned_loss=0.126, over 5676317.81 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3651, pruned_loss=0.1175, over 5737735.35 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3743, pruned_loss=0.1271, over 5654595.28 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:01:00,056 INFO [train.py:968] (0/2) Epoch 17, batch 44100, giga_loss[loss=0.2744, simple_loss=0.3517, pruned_loss=0.09858, over 28582.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3729, pruned_loss=0.1244, over 5679529.89 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3651, pruned_loss=0.1175, over 5741413.97 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3739, pruned_loss=0.1256, over 5657525.21 frames. ], batch size: 85, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:01:27,806 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.016e+02 1.571e+03 1.969e+03 2.551e+03 5.185e+03, threshold=3.939e+03, percent-clipped=1.0
+2023-03-09 04:01:37,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=775012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:01:45,000 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-09 04:01:50,850 INFO [train.py:968] (0/2) Epoch 17, batch 44150, giga_loss[loss=0.2826, simple_loss=0.364, pruned_loss=0.1007, over 28935.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3751, pruned_loss=0.1256, over 5664567.89 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3651, pruned_loss=0.1176, over 5735474.18 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3763, pruned_loss=0.1266, over 5650152.35 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:01:51,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3594, 1.6285, 1.3339, 1.2217], device='cuda:0'), covar=tensor([0.2242, 0.2287, 0.2443, 0.2123], device='cuda:0'), in_proj_covar=tensor([0.1437, 0.1045, 0.1276, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 04:02:09,256 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3893, 1.8661, 1.4776, 1.5468], device='cuda:0'), covar=tensor([0.0673, 0.0364, 0.0311, 0.0727], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-09 04:02:43,194 INFO [train.py:968] (0/2) Epoch 17, batch 44200, giga_loss[loss=0.3251, simple_loss=0.3852, pruned_loss=0.1325, over 29018.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3747, pruned_loss=0.1258, over 5663926.46 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.365, pruned_loss=0.1175, over 5731638.01 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3758, pruned_loss=0.1267, over 5655575.72 frames. ], batch size: 155, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:03:06,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.608e+03 2.007e+03 2.871e+03 6.100e+03, threshold=4.013e+03, percent-clipped=12.0
+2023-03-09 04:03:16,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=775113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:03:30,857 INFO [train.py:968] (0/2) Epoch 17, batch 44250, giga_loss[loss=0.3091, simple_loss=0.3913, pruned_loss=0.1135, over 28914.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3758, pruned_loss=0.1243, over 5665080.65 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3652, pruned_loss=0.1176, over 5725376.23 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3766, pruned_loss=0.1251, over 5663075.55 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:03:58,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=775155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:03:59,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=775158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:04:16,432 INFO [train.py:968] (0/2) Epoch 17, batch 44300, giga_loss[loss=0.2984, simple_loss=0.3839, pruned_loss=0.1065, over 29117.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3777, pruned_loss=0.123, over 5665301.24 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3655, pruned_loss=0.1177, over 5727402.79 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3782, pruned_loss=0.1235, over 5661194.76 frames. ], batch size: 155, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:04:25,939 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=775187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:04:38,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.922e+02 1.358e+03 1.848e+03 2.505e+03 4.336e+03, threshold=3.696e+03, percent-clipped=2.0
+2023-03-09 04:05:07,237 INFO [train.py:968] (0/2) Epoch 17, batch 44350, giga_loss[loss=0.3091, simple_loss=0.3798, pruned_loss=0.1191, over 29037.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3802, pruned_loss=0.1247, over 5661543.89 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3655, pruned_loss=0.1178, over 5729527.61 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3809, pruned_loss=0.1252, over 5655202.12 frames. ], batch size: 155, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:05:36,072 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=775256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:05:41,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=775259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:05:41,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5171, 1.6010, 1.7443, 1.3106], device='cuda:0'), covar=tensor([0.1579, 0.2381, 0.1292, 0.1552], device='cuda:0'), in_proj_covar=tensor([0.0872, 0.0701, 0.0918, 0.0818], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 04:05:51,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=775269.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 04:05:59,956 INFO [train.py:968] (0/2) Epoch 17, batch 44400, giga_loss[loss=0.3835, simple_loss=0.4282, pruned_loss=0.1694, over 28234.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3834, pruned_loss=0.1281, over 5658232.87 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3656, pruned_loss=0.1179, over 5726701.97 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3839, pruned_loss=0.1284, over 5655551.46 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:06:07,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=775288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:06:20,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.740e+03 2.304e+03 3.540e+03 1.043e+04, threshold=4.608e+03, percent-clipped=21.0
+2023-03-09 04:06:47,998 INFO [train.py:968] (0/2) Epoch 17, batch 44450, giga_loss[loss=0.2993, simple_loss=0.3706, pruned_loss=0.114, over 29103.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3828, pruned_loss=0.1284, over 5657879.70 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3655, pruned_loss=0.1177, over 5725822.19 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3838, pruned_loss=0.1291, over 5654445.10 frames. ], batch size: 155, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:07:20,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6604, 1.8412, 1.5480, 1.7060], device='cuda:0'), covar=tensor([0.2375, 0.2452, 0.2761, 0.2329], device='cuda:0'), in_proj_covar=tensor([0.1429, 0.1039, 0.1269, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 04:07:33,367 INFO [train.py:968] (0/2) Epoch 17, batch 44500, giga_loss[loss=0.2989, simple_loss=0.3735, pruned_loss=0.1122, over 28927.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3815, pruned_loss=0.1281, over 5674825.38 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3654, pruned_loss=0.1177, over 5730679.37 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3828, pruned_loss=0.129, over 5666454.30 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:07:52,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.780e+02 1.706e+03 2.239e+03 3.189e+03 1.097e+04, threshold=4.478e+03, percent-clipped=8.0
+2023-03-09 04:08:02,228 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=775412.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 04:08:05,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=775415.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 04:08:17,335 INFO [train.py:968] (0/2) Epoch 17, batch 44550, giga_loss[loss=0.2702, simple_loss=0.3493, pruned_loss=0.0955, over 28983.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3808, pruned_loss=0.1277, over 5656092.87 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3655, pruned_loss=0.118, over 5716250.65 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.382, pruned_loss=0.1284, over 5660013.37 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:08:30,182 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=775444.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 04:09:04,148 INFO [train.py:968] (0/2) Epoch 17, batch 44600, giga_loss[loss=0.3228, simple_loss=0.366, pruned_loss=0.1398, over 23603.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3794, pruned_loss=0.1247, over 5671567.93 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3654, pruned_loss=0.1178, over 5720071.81 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3807, pruned_loss=0.1255, over 5670179.50 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:09:21,730 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=775497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:09:22,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3272, 1.4105, 5.3273, 3.8431], device='cuda:0'), covar=tensor([0.1366, 0.2831, 0.0421, 0.0730], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0630, 0.0934, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 04:09:25,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.222e+02 1.570e+03 2.080e+03 3.002e+03 9.847e+03, threshold=4.160e+03, percent-clipped=8.0
+2023-03-09 04:09:41,629 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=775519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:09:50,679 INFO [train.py:968] (0/2) Epoch 17, batch 44650, giga_loss[loss=0.3531, simple_loss=0.403, pruned_loss=0.1516, over 27553.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.379, pruned_loss=0.1231, over 5680898.35 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.365, pruned_loss=0.1175, over 5723570.19 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3806, pruned_loss=0.1242, over 5675889.51 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:10:10,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5226, 2.6982, 1.6631, 1.6583], device='cuda:0'), covar=tensor([0.0767, 0.0326, 0.0678, 0.1052], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0548, 0.0373, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 04:10:40,674 INFO [train.py:968] (0/2) Epoch 17, batch 44700, giga_loss[loss=0.3796, simple_loss=0.4024, pruned_loss=0.1784, over 23334.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3796, pruned_loss=0.1245, over 5659192.53 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3651, pruned_loss=0.1176, over 5718291.53 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3813, pruned_loss=0.1255, over 5657537.03 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:11:06,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.997e+02 1.548e+03 2.075e+03 3.104e+03 7.601e+03, threshold=4.150e+03, percent-clipped=7.0
+2023-03-09 04:11:17,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9618, 1.9821, 1.4851, 1.6487], device='cuda:0'), covar=tensor([0.0817, 0.0522, 0.0929, 0.0981], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0445, 0.0509, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 04:11:27,812 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1733, 3.2410, 1.4224, 1.3956], device='cuda:0'), covar=tensor([0.1085, 0.0382, 0.0958, 0.1436], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0546, 0.0373, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 04:11:28,714 INFO [train.py:968] (0/2) Epoch 17, batch 44750, libri_loss[loss=0.2949, simple_loss=0.3402, pruned_loss=0.1248, over 29384.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3793, pruned_loss=0.1246, over 5664356.06 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3651, pruned_loss=0.1175, over 5722349.85 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.381, pruned_loss=0.1257, over 5657565.32 frames. ], batch size: 67, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:11:30,136 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-09 04:12:14,403 INFO [train.py:968] (0/2) Epoch 17, batch 44800, giga_loss[loss=0.2965, simple_loss=0.365, pruned_loss=0.114, over 28904.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3776, pruned_loss=0.1243, over 5668668.62 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3653, pruned_loss=0.1176, over 5722604.62 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3791, pruned_loss=0.1252, over 5661464.79 frames. ], batch size: 174, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 04:12:31,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0267, 1.3384, 1.0833, 0.1635], device='cuda:0'), covar=tensor([0.3390, 0.2729, 0.4006, 0.6017], device='cuda:0'), in_proj_covar=tensor([0.1695, 0.1611, 0.1571, 0.1387], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 04:12:38,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.625e+03 2.060e+03 2.711e+03 6.850e+03, threshold=4.120e+03, percent-clipped=9.0
+2023-03-09 04:13:02,460 INFO [train.py:968] (0/2) Epoch 17, batch 44850, giga_loss[loss=0.2791, simple_loss=0.3568, pruned_loss=0.1007, over 28927.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3769, pruned_loss=0.1247, over 5664888.60 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3656, pruned_loss=0.1175, over 5724649.69 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3781, pruned_loss=0.1257, over 5655825.10 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 04:13:49,576 INFO [train.py:968] (0/2) Epoch 17, batch 44900, giga_loss[loss=0.2895, simple_loss=0.356, pruned_loss=0.1115, over 28744.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3761, pruned_loss=0.1248, over 5662426.30 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3659, pruned_loss=0.1179, over 5717509.13 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.377, pruned_loss=0.1254, over 5660190.05 frames. ], batch size: 262, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:14:12,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.209e+03 1.745e+03 2.257e+03 2.822e+03 6.204e+03, threshold=4.514e+03, percent-clipped=11.0
+2023-03-09 04:14:36,540 INFO [train.py:968] (0/2) Epoch 17, batch 44950, giga_loss[loss=0.3165, simple_loss=0.3821, pruned_loss=0.1254, over 28636.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3752, pruned_loss=0.1257, over 5656959.38 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3658, pruned_loss=0.1179, over 5717528.67 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3762, pruned_loss=0.1263, over 5653961.94 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:15:14,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=775872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:15:20,398 INFO [train.py:968] (0/2) Epoch 17, batch 45000, giga_loss[loss=0.3199, simple_loss=0.3804, pruned_loss=0.1297, over 28335.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3741, pruned_loss=0.1256, over 5647748.32 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3657, pruned_loss=0.1179, over 5715703.26 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3755, pruned_loss=0.1266, over 5643296.55 frames. ], batch size: 369, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:15:20,402 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 04:15:28,821 INFO [train.py:1012] (0/2) Epoch 17, validation: loss=0.2079, simple_loss=0.3159, pruned_loss=0.05001, over 944034.00 frames. 
+2023-03-09 04:15:28,822 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 04:15:42,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=775894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:15:50,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.544e+03 2.242e+03 3.497e+03 9.300e+03, threshold=4.484e+03, percent-clipped=17.0
+2023-03-09 04:15:53,438 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=775904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:16:11,403 INFO [train.py:968] (0/2) Epoch 17, batch 45050, giga_loss[loss=0.3122, simple_loss=0.3703, pruned_loss=0.1271, over 26782.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1234, over 5651035.65 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.366, pruned_loss=0.1181, over 5718205.45 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3734, pruned_loss=0.1242, over 5642673.22 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:16:54,449 INFO [train.py:968] (0/2) Epoch 17, batch 45100, giga_loss[loss=0.2971, simple_loss=0.3644, pruned_loss=0.1148, over 28785.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3676, pruned_loss=0.1184, over 5663204.29 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3654, pruned_loss=0.1178, over 5722470.93 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3691, pruned_loss=0.1193, over 5651024.96 frames. ], batch size: 284, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:17:16,313 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-776000.pt
+2023-03-09 04:17:18,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.334e+02 1.406e+03 1.738e+03 2.483e+03 9.291e+03, threshold=3.476e+03, percent-clipped=3.0
+2023-03-09 04:17:33,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-09 04:17:35,018 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=776015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:17:37,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=776018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:17:45,833 INFO [train.py:968] (0/2) Epoch 17, batch 45150, giga_loss[loss=0.3358, simple_loss=0.391, pruned_loss=0.1403, over 27531.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3663, pruned_loss=0.1176, over 5655319.14 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3651, pruned_loss=0.1175, over 5725329.46 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3678, pruned_loss=0.1186, over 5641854.04 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:17:54,771 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=776037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:17:57,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=776040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:18:04,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=776047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:18:05,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2962, 2.2592, 2.1179, 2.1074], device='cuda:0'), covar=tensor([0.1828, 0.2557, 0.2247, 0.2133], device='cuda:0'), in_proj_covar=tensor([0.0463, 0.0749, 0.0705, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 04:18:25,946 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=776069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:18:35,631 INFO [train.py:968] (0/2) Epoch 17, batch 45200, giga_loss[loss=0.2918, simple_loss=0.3551, pruned_loss=0.1143, over 28723.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.365, pruned_loss=0.117, over 5672193.28 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3651, pruned_loss=0.1174, over 5726725.42 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3662, pruned_loss=0.1179, over 5659930.68 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 04:18:43,611 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-09 04:18:55,002 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=776097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:19:04,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.750e+02 1.581e+03 2.129e+03 2.961e+03 9.352e+03, threshold=4.258e+03, percent-clipped=10.0
+2023-03-09 04:19:27,248 INFO [train.py:968] (0/2) Epoch 17, batch 45250, giga_loss[loss=0.2883, simple_loss=0.3656, pruned_loss=0.1055, over 28960.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3625, pruned_loss=0.1162, over 5671719.12 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3649, pruned_loss=0.1171, over 5720301.31 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3637, pruned_loss=0.1171, over 5665875.78 frames. ], batch size: 145, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:19:32,717 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=776135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:20:09,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9273, 3.2059, 1.8229, 1.2154], device='cuda:0'), covar=tensor([0.6264, 0.2907, 0.4332, 0.5598], device='cuda:0'), in_proj_covar=tensor([0.1689, 0.1603, 0.1567, 0.1379], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 04:20:11,147 INFO [train.py:968] (0/2) Epoch 17, batch 45300, giga_loss[loss=0.3072, simple_loss=0.3681, pruned_loss=0.1232, over 29067.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3651, pruned_loss=0.1177, over 5669614.38 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3652, pruned_loss=0.1174, over 5705117.60 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3657, pruned_loss=0.1183, over 5677247.76 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:20:32,894 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.849e+02 1.692e+03 2.118e+03 3.018e+03 8.891e+03, threshold=4.236e+03, percent-clipped=11.0
+2023-03-09 04:20:54,977 INFO [train.py:968] (0/2) Epoch 17, batch 45350, giga_loss[loss=0.3249, simple_loss=0.384, pruned_loss=0.1329, over 28638.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3679, pruned_loss=0.1195, over 5665740.54 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.366, pruned_loss=0.1181, over 5706353.44 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3677, pruned_loss=0.1192, over 5669689.51 frames. ], batch size: 60, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:21:45,288 INFO [train.py:968] (0/2) Epoch 17, batch 45400, giga_loss[loss=0.2631, simple_loss=0.3384, pruned_loss=0.09388, over 28923.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3694, pruned_loss=0.1204, over 5664115.24 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3663, pruned_loss=0.1183, over 5707225.67 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.369, pruned_loss=0.1201, over 5665878.92 frames. ], batch size: 145, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:21:45,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=776279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:22:07,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.725e+03 2.361e+03 3.691e+03 6.469e+03, threshold=4.722e+03, percent-clipped=16.0
+2023-03-09 04:22:10,310 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-09 04:22:27,342 INFO [train.py:968] (0/2) Epoch 17, batch 45450, giga_loss[loss=0.3592, simple_loss=0.4165, pruned_loss=0.1509, over 28902.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3705, pruned_loss=0.1214, over 5667389.02 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3667, pruned_loss=0.1184, over 5705427.56 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.37, pruned_loss=0.1212, over 5667839.83 frames. ], batch size: 186, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:22:46,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3464, 1.5246, 1.6094, 1.1937], device='cuda:0'), covar=tensor([0.1543, 0.2404, 0.1240, 0.1555], device='cuda:0'), in_proj_covar=tensor([0.0877, 0.0703, 0.0924, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 04:22:49,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-09 04:23:10,886 INFO [train.py:968] (0/2) Epoch 17, batch 45500, giga_loss[loss=0.2646, simple_loss=0.346, pruned_loss=0.09154, over 29038.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.372, pruned_loss=0.1229, over 5661265.36 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3668, pruned_loss=0.1183, over 5709349.91 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.1229, over 5656676.55 frames. ], batch size: 155, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:23:36,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.085e+02 1.762e+03 2.169e+03 3.113e+03 7.738e+03, threshold=4.337e+03, percent-clipped=3.0
+2023-03-09 04:23:56,307 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=776422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:23:59,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=776425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:24:01,608 INFO [train.py:968] (0/2) Epoch 17, batch 45550, giga_loss[loss=0.302, simple_loss=0.3789, pruned_loss=0.1125, over 28892.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3745, pruned_loss=0.1249, over 5642124.50 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3669, pruned_loss=0.1184, over 5708813.89 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3742, pruned_loss=0.1249, over 5638337.93 frames. ], batch size: 145, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:24:23,364 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=776454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:24:28,355 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 04:24:39,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=776472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:24:46,611 INFO [train.py:968] (0/2) Epoch 17, batch 45600, libri_loss[loss=0.3027, simple_loss=0.3775, pruned_loss=0.1139, over 29526.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3762, pruned_loss=0.1259, over 5642709.66 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3673, pruned_loss=0.1187, over 5695215.47 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3759, pruned_loss=0.1259, over 5650266.69 frames. ], batch size: 81, lr: 1.85e-03, grad_scale: 4.0
+2023-03-09 04:24:53,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3070, 1.2278, 3.9250, 3.3139], device='cuda:0'), covar=tensor([0.1679, 0.2670, 0.0452, 0.0769], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0629, 0.0930, 0.0861], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 04:25:09,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.662e+03 2.106e+03 2.773e+03 6.335e+03, threshold=4.212e+03, percent-clipped=7.0
+2023-03-09 04:25:17,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=776510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:25:38,945 INFO [train.py:968] (0/2) Epoch 17, batch 45650, giga_loss[loss=0.3505, simple_loss=0.3824, pruned_loss=0.1593, over 23634.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3769, pruned_loss=0.1268, over 5633041.51 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3675, pruned_loss=0.1188, over 5690003.58 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3766, pruned_loss=0.1268, over 5642261.09 frames. ], batch size: 705, lr: 1.85e-03, grad_scale: 4.0
+2023-03-09 04:26:04,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4592, 4.3265, 1.7335, 1.6256], device='cuda:0'), covar=tensor([0.0991, 0.0330, 0.0857, 0.1334], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0542, 0.0371, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 04:26:18,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6303, 4.4736, 4.2422, 2.0706], device='cuda:0'), covar=tensor([0.0576, 0.0719, 0.0787, 0.1920], device='cuda:0'), in_proj_covar=tensor([0.1173, 0.1092, 0.0933, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 04:26:29,389 INFO [train.py:968] (0/2) Epoch 17, batch 45700, giga_loss[loss=0.2868, simple_loss=0.3661, pruned_loss=0.1037, over 29034.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3767, pruned_loss=0.1261, over 5646375.25 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3677, pruned_loss=0.1189, over 5693374.06 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3765, pruned_loss=0.1262, over 5650020.84 frames. ], batch size: 136, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:26:58,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 1.725e+03 2.179e+03 3.101e+03 8.362e+03, threshold=4.359e+03, percent-clipped=13.0
+2023-03-09 04:27:11,420 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=776615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:27:13,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=776618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:27:24,882 INFO [train.py:968] (0/2) Epoch 17, batch 45750, giga_loss[loss=0.2876, simple_loss=0.3527, pruned_loss=0.1113, over 28628.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3764, pruned_loss=0.1248, over 5648661.62 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3678, pruned_loss=0.1191, over 5696403.04 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3762, pruned_loss=0.1247, over 5648145.89 frames. ], batch size: 92, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:27:40,095 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=776647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:27:48,201 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=776653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:27:49,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=776656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:28:11,839 INFO [train.py:968] (0/2) Epoch 17, batch 45800, giga_loss[loss=0.3581, simple_loss=0.4034, pruned_loss=0.1565, over 27544.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3778, pruned_loss=0.1262, over 5600140.49 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3685, pruned_loss=0.1199, over 5639753.23 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3772, pruned_loss=0.1256, over 5650072.03 frames. ], batch size: 472, lr: 1.85e-03, grad_scale: 1.0
+2023-03-09 04:28:17,042 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=776685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:28:32,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.809e+02 1.788e+03 2.434e+03 3.316e+03 9.867e+03, threshold=4.869e+03, percent-clipped=12.0
+2023-03-09 04:28:37,396 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-09 04:28:40,449 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-17.pt
+2023-03-09 04:29:54,702 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5072, 1.7777, 1.8160, 1.2997], device='cuda:0'), covar=tensor([0.1923, 0.2698, 0.1612, 0.1903], device='cuda:0'), in_proj_covar=tensor([0.0873, 0.0701, 0.0921, 0.0820], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 04:29:59,694 INFO [train.py:968] (0/2) Epoch 18, batch 50, libri_loss[loss=0.2562, simple_loss=0.3353, pruned_loss=0.08861, over 29577.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3787, pruned_loss=0.1135, over 1264309.54 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.348, pruned_loss=0.09833, over 146000.54 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3825, pruned_loss=0.1155, over 1147001.26 frames. ], batch size: 75, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:30:23,809 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=776788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:30:41,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.996e+02 1.244e+03 1.482e+03 1.941e+03 4.489e+03, threshold=2.964e+03, percent-clipped=1.0
+2023-03-09 04:30:46,681 INFO [train.py:968] (0/2) Epoch 18, batch 100, giga_loss[loss=0.2714, simple_loss=0.3496, pruned_loss=0.09662, over 28857.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3677, pruned_loss=0.1072, over 2249089.13 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3442, pruned_loss=0.09611, over 287458.72 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3708, pruned_loss=0.1087, over 2064280.47 frames. ], batch size: 174, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:31:18,921 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=776847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:31:22,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4449, 1.8005, 1.4324, 1.3294], device='cuda:0'), covar=tensor([0.2807, 0.2801, 0.3113, 0.2434], device='cuda:0'), in_proj_covar=tensor([0.1442, 0.1049, 0.1278, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 04:31:31,120 INFO [train.py:968] (0/2) Epoch 18, batch 150, giga_loss[loss=0.2762, simple_loss=0.332, pruned_loss=0.1102, over 26635.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3515, pruned_loss=0.09951, over 3014193.42 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3347, pruned_loss=0.08917, over 426545.68 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3543, pruned_loss=0.1012, over 2793708.84 frames. ], batch size: 555, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:32:02,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-09 04:32:10,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.866e+02 1.088e+03 1.299e+03 1.806e+03 4.616e+03, threshold=2.599e+03, percent-clipped=4.0
+2023-03-09 04:32:14,773 INFO [train.py:968] (0/2) Epoch 18, batch 200, giga_loss[loss=0.2591, simple_loss=0.3195, pruned_loss=0.09932, over 28767.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3387, pruned_loss=0.09293, over 3618548.54 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3361, pruned_loss=0.08818, over 534710.07 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3399, pruned_loss=0.094, over 3397892.14 frames. ], batch size: 92, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:32:55,615 INFO [train.py:968] (0/2) Epoch 18, batch 250, giga_loss[loss=0.2078, simple_loss=0.2798, pruned_loss=0.06795, over 28707.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.328, pruned_loss=0.08775, over 4085032.95 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3377, pruned_loss=0.09025, over 693161.00 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3279, pruned_loss=0.08799, over 3853583.65 frames. ], batch size: 92, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:33:32,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.895e+02 1.049e+03 1.375e+03 1.869e+03 1.211e+04, threshold=2.750e+03, percent-clipped=12.0
+2023-03-09 04:33:38,626 INFO [train.py:968] (0/2) Epoch 18, batch 300, giga_loss[loss=0.2229, simple_loss=0.296, pruned_loss=0.07492, over 28284.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3199, pruned_loss=0.08413, over 4444560.63 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3379, pruned_loss=0.08942, over 869134.90 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3188, pruned_loss=0.08412, over 4210086.49 frames. ], batch size: 368, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:33:58,557 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=777032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:34:26,521 INFO [train.py:968] (0/2) Epoch 18, batch 350, giga_loss[loss=0.2376, simple_loss=0.3089, pruned_loss=0.08319, over 29003.00 frames. ], tot_loss[loss=0.238, simple_loss=0.3129, pruned_loss=0.08157, over 4725710.11 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3364, pruned_loss=0.08882, over 944386.41 frames. ], giga_tot_loss[loss=0.2373, simple_loss=0.3117, pruned_loss=0.08145, over 4522798.64 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:34:26,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0352, 3.8480, 3.6184, 2.0386], device='cuda:0'), covar=tensor([0.0580, 0.0771, 0.0732, 0.2162], device='cuda:0'), in_proj_covar=tensor([0.1163, 0.1081, 0.0922, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 04:35:03,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.719e+02 9.363e+02 1.128e+03 1.453e+03 2.570e+03, threshold=2.256e+03, percent-clipped=0.0
+2023-03-09 04:35:07,690 INFO [train.py:968] (0/2) Epoch 18, batch 400, giga_loss[loss=0.2175, simple_loss=0.2892, pruned_loss=0.07292, over 27650.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3079, pruned_loss=0.07907, over 4949829.64 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3358, pruned_loss=0.08838, over 1067417.18 frames. ], giga_tot_loss[loss=0.2319, simple_loss=0.3062, pruned_loss=0.07873, over 4763918.23 frames. ], batch size: 472, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:35:47,649 INFO [train.py:968] (0/2) Epoch 18, batch 450, giga_loss[loss=0.218, simple_loss=0.2905, pruned_loss=0.07268, over 29018.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3059, pruned_loss=0.0776, over 5116411.63 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3384, pruned_loss=0.08899, over 1199695.29 frames. ], giga_tot_loss[loss=0.2284, simple_loss=0.3031, pruned_loss=0.07687, over 4953325.42 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:35:51,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=777163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:36:29,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.908e+02 1.073e+03 1.335e+03 1.902e+03 5.529e+03, threshold=2.669e+03, percent-clipped=18.0
+2023-03-09 04:36:34,435 INFO [train.py:968] (0/2) Epoch 18, batch 500, giga_loss[loss=0.2183, simple_loss=0.2921, pruned_loss=0.0723, over 28640.00 frames. ], tot_loss[loss=0.229, simple_loss=0.304, pruned_loss=0.077, over 5247264.38 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3406, pruned_loss=0.09015, over 1264897.81 frames. ], giga_tot_loss[loss=0.2265, simple_loss=0.3009, pruned_loss=0.07602, over 5113078.85 frames. ], batch size: 242, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:36:43,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=777222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:36:52,874 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9942, 5.1967, 2.2552, 2.2170], device='cuda:0'), covar=tensor([0.0942, 0.0243, 0.0824, 0.1236], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0539, 0.0370, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 04:37:19,611 INFO [train.py:968] (0/2) Epoch 18, batch 550, giga_loss[loss=0.2151, simple_loss=0.2929, pruned_loss=0.0686, over 28965.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3024, pruned_loss=0.07662, over 5349754.38 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3394, pruned_loss=0.08972, over 1379188.99 frames. ], giga_tot_loss[loss=0.2252, simple_loss=0.2993, pruned_loss=0.07559, over 5229442.12 frames. ], batch size: 227, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:37:42,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5065, 3.1135, 2.4604, 2.2293], device='cuda:0'), covar=tensor([0.1904, 0.1041, 0.1421, 0.1539], device='cuda:0'), in_proj_covar=tensor([0.1896, 0.1824, 0.1752, 0.1892], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 04:37:47,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1796, 2.3097, 1.8617, 2.0188], device='cuda:0'), covar=tensor([0.0954, 0.0704, 0.1025, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0378, 0.0441, 0.0509, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 04:37:58,774 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=777306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:37:59,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.575e+02 1.112e+03 1.400e+03 1.973e+03 4.002e+03, threshold=2.800e+03, percent-clipped=10.0
+2023-03-09 04:38:00,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=777309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:38:02,849 INFO [train.py:968] (0/2) Epoch 18, batch 600, giga_loss[loss=0.2013, simple_loss=0.2793, pruned_loss=0.06163, over 28269.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3009, pruned_loss=0.07562, over 5429201.15 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3401, pruned_loss=0.08958, over 1534425.21 frames. ], giga_tot_loss[loss=0.2228, simple_loss=0.2969, pruned_loss=0.07435, over 5316845.77 frames. ], batch size: 368, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:38:10,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-09 04:38:31,018 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=777338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:38:52,520 INFO [train.py:968] (0/2) Epoch 18, batch 650, giga_loss[loss=0.2868, simple_loss=0.335, pruned_loss=0.1193, over 28851.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3004, pruned_loss=0.07586, over 5480890.75 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3406, pruned_loss=0.08946, over 1639819.21 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.296, pruned_loss=0.07454, over 5383194.94 frames. ], batch size: 199, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:38:54,665 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=777365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:38:56,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=777368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:39:22,516 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=777397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:39:32,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.724e+02 1.039e+03 1.330e+03 1.768e+03 5.172e+03, threshold=2.659e+03, percent-clipped=5.0
+2023-03-09 04:39:32,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=777407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:39:36,771 INFO [train.py:968] (0/2) Epoch 18, batch 700, giga_loss[loss=0.1997, simple_loss=0.2742, pruned_loss=0.0626, over 28926.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2973, pruned_loss=0.07431, over 5520005.09 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.342, pruned_loss=0.09055, over 1733714.17 frames. ], giga_tot_loss[loss=0.2188, simple_loss=0.2925, pruned_loss=0.07261, over 5444506.27 frames. ], batch size: 145, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:40:19,799 INFO [train.py:968] (0/2) Epoch 18, batch 750, giga_loss[loss=0.2546, simple_loss=0.3088, pruned_loss=0.1002, over 26581.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2944, pruned_loss=0.0725, over 5570644.99 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3411, pruned_loss=0.08994, over 1856794.32 frames. ], giga_tot_loss[loss=0.2155, simple_loss=0.2894, pruned_loss=0.07082, over 5500979.31 frames. ], batch size: 555, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:40:59,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.127e+02 9.637e+02 1.243e+03 1.674e+03 3.878e+03, threshold=2.486e+03, percent-clipped=7.0
+2023-03-09 04:41:02,505 INFO [train.py:968] (0/2) Epoch 18, batch 800, giga_loss[loss=0.2642, simple_loss=0.3282, pruned_loss=0.1001, over 28284.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2936, pruned_loss=0.07219, over 5601906.24 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3406, pruned_loss=0.0893, over 2030373.85 frames. ], giga_tot_loss[loss=0.2141, simple_loss=0.2877, pruned_loss=0.07026, over 5536991.45 frames. ], batch size: 368, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:41:40,827 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=777550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:41:43,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=777553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:41:52,260 INFO [train.py:968] (0/2) Epoch 18, batch 850, libri_loss[loss=0.2425, simple_loss=0.3245, pruned_loss=0.08029, over 29536.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3011, pruned_loss=0.07699, over 5614600.75 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3399, pruned_loss=0.08905, over 2107212.28 frames. ], giga_tot_loss[loss=0.2233, simple_loss=0.2959, pruned_loss=0.07533, over 5557462.13 frames. ], batch size: 79, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:42:10,811 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=777582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:42:16,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0970, 2.7264, 1.2637, 1.2688], device='cuda:0'), covar=tensor([0.1296, 0.0469, 0.1104, 0.1679], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0537, 0.0369, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 04:42:34,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.057e+02 1.188e+03 1.515e+03 1.924e+03 1.219e+04, threshold=3.029e+03, percent-clipped=14.0
+2023-03-09 04:42:37,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=777610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:42:39,185 INFO [train.py:968] (0/2) Epoch 18, batch 900, giga_loss[loss=0.3094, simple_loss=0.378, pruned_loss=0.1204, over 28692.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.313, pruned_loss=0.08286, over 5634125.67 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3393, pruned_loss=0.08878, over 2219561.32 frames. ], giga_tot_loss[loss=0.2356, simple_loss=0.3083, pruned_loss=0.08144, over 5580970.25 frames. ], batch size: 307, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:42:40,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-09 04:42:41,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1022, 2.5442, 2.1815, 1.8063], device='cuda:0'), covar=tensor([0.2620, 0.1785, 0.2109, 0.2431], device='cuda:0'), in_proj_covar=tensor([0.1886, 0.1811, 0.1740, 0.1884], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 04:43:23,019 INFO [train.py:968] (0/2) Epoch 18, batch 950, giga_loss[loss=0.2576, simple_loss=0.3345, pruned_loss=0.09037, over 28515.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3251, pruned_loss=0.08884, over 5653319.24 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3396, pruned_loss=0.08897, over 2289418.50 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3209, pruned_loss=0.08763, over 5609240.82 frames. ], batch size: 71, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:43:36,706 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-09 04:43:59,815 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.774e+02 1.287e+03 1.528e+03 2.038e+03 9.638e+03, threshold=3.056e+03, percent-clipped=7.0
+2023-03-09 04:44:03,687 INFO [train.py:968] (0/2) Epoch 18, batch 1000, giga_loss[loss=0.2715, simple_loss=0.3502, pruned_loss=0.09643, over 28906.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3324, pruned_loss=0.09159, over 5668577.88 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3399, pruned_loss=0.08892, over 2378686.31 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3288, pruned_loss=0.0907, over 5628044.01 frames. ], batch size: 112, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:44:43,751 INFO [train.py:968] (0/2) Epoch 18, batch 1050, giga_loss[loss=0.2613, simple_loss=0.3486, pruned_loss=0.08698, over 29049.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3366, pruned_loss=0.0921, over 5681018.17 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3405, pruned_loss=0.08904, over 2431125.32 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3335, pruned_loss=0.09141, over 5646070.95 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:45:11,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0694, 1.2279, 3.6418, 2.9586], device='cuda:0'), covar=tensor([0.1746, 0.2734, 0.0466, 0.1025], device='cuda:0'), in_proj_covar=tensor([0.0724, 0.0625, 0.0924, 0.0851], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 04:45:28,185 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.948e+02 1.168e+03 1.459e+03 1.974e+03 4.797e+03, threshold=2.919e+03, percent-clipped=5.0
+2023-03-09 04:45:32,023 INFO [train.py:968] (0/2) Epoch 18, batch 1100, giga_loss[loss=0.2685, simple_loss=0.357, pruned_loss=0.09004, over 28841.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.339, pruned_loss=0.09261, over 5676704.31 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3399, pruned_loss=0.08878, over 2532337.05 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3367, pruned_loss=0.09225, over 5644195.68 frames. ], batch size: 145, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:46:04,429 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7327, 1.8210, 1.8601, 1.6949], device='cuda:0'), covar=tensor([0.1830, 0.2084, 0.2046, 0.1895], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0742, 0.0698, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 04:46:16,341 INFO [train.py:968] (0/2) Epoch 18, batch 1150, giga_loss[loss=0.2966, simple_loss=0.3702, pruned_loss=0.1115, over 28521.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3407, pruned_loss=0.09367, over 5691419.68 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3402, pruned_loss=0.08892, over 2566071.15 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3387, pruned_loss=0.09338, over 5663800.27 frames. ], batch size: 336, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:46:20,975 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=777867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:46:30,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6520, 1.8284, 1.5720, 1.7384], device='cuda:0'), covar=tensor([0.2216, 0.2104, 0.2103, 0.1975], device='cuda:0'), in_proj_covar=tensor([0.1448, 0.1050, 0.1283, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 04:46:55,925 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.391e+02 1.286e+03 1.733e+03 2.897e+03 6.721e+03, threshold=3.466e+03, percent-clipped=23.0
+2023-03-09 04:47:00,912 INFO [train.py:968] (0/2) Epoch 18, batch 1200, giga_loss[loss=0.2982, simple_loss=0.3702, pruned_loss=0.1131, over 28591.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3438, pruned_loss=0.09635, over 5680280.01 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3412, pruned_loss=0.08932, over 2631795.79 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3419, pruned_loss=0.09609, over 5654810.10 frames. ], batch size: 336, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:47:29,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8206, 2.6737, 1.7284, 1.0208], device='cuda:0'), covar=tensor([0.7480, 0.3302, 0.3895, 0.6200], device='cuda:0'), in_proj_covar=tensor([0.1685, 0.1595, 0.1569, 0.1375], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 04:47:43,014 INFO [train.py:968] (0/2) Epoch 18, batch 1250, giga_loss[loss=0.276, simple_loss=0.3529, pruned_loss=0.0995, over 28293.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3459, pruned_loss=0.09766, over 5676320.55 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.341, pruned_loss=0.08916, over 2732441.30 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3446, pruned_loss=0.09779, over 5659005.58 frames. ], batch size: 368, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:48:01,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=777985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:48:17,401 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-778000.pt
+2023-03-09 04:48:24,050 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.555e+02 1.273e+03 1.552e+03 1.989e+03 4.431e+03, threshold=3.103e+03, percent-clipped=3.0
+2023-03-09 04:48:28,317 INFO [train.py:968] (0/2) Epoch 18, batch 1300, giga_loss[loss=0.2706, simple_loss=0.3493, pruned_loss=0.09593, over 28832.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3489, pruned_loss=0.09877, over 5680608.14 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3414, pruned_loss=0.08931, over 2764114.39 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3478, pruned_loss=0.09888, over 5665140.42 frames. ], batch size: 199, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:48:41,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2137, 1.4533, 1.5549, 1.3167], device='cuda:0'), covar=tensor([0.1970, 0.1740, 0.2143, 0.1803], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0739, 0.0698, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 04:49:08,133 INFO [train.py:968] (0/2) Epoch 18, batch 1350, libri_loss[loss=0.3318, simple_loss=0.395, pruned_loss=0.1342, over 20166.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3513, pruned_loss=0.09915, over 5681399.53 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3432, pruned_loss=0.09052, over 2822326.10 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3498, pruned_loss=0.09888, over 5682703.96 frames. ], batch size: 187, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:49:44,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.680e+02 1.221e+03 1.473e+03 1.868e+03 4.637e+03, threshold=2.946e+03, percent-clipped=4.0
+2023-03-09 04:49:48,204 INFO [train.py:968] (0/2) Epoch 18, batch 1400, giga_loss[loss=0.2845, simple_loss=0.3655, pruned_loss=0.1018, over 29036.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3524, pruned_loss=0.0992, over 5684131.12 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3428, pruned_loss=0.09001, over 2913752.04 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3516, pruned_loss=0.09944, over 5679253.03 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:49:49,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1480, 1.1169, 3.7533, 3.0727], device='cuda:0'), covar=tensor([0.1752, 0.2995, 0.0426, 0.1130], device='cuda:0'), in_proj_covar=tensor([0.0720, 0.0622, 0.0916, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 04:50:01,770 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=778128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:50:03,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=778131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:50:32,955 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=778160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:50:34,906 INFO [train.py:968] (0/2) Epoch 18, batch 1450, giga_loss[loss=0.2732, simple_loss=0.3568, pruned_loss=0.0948, over 28991.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3518, pruned_loss=0.09791, over 5681290.21 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3429, pruned_loss=0.09013, over 2918796.28 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3512, pruned_loss=0.09807, over 5684689.00 frames. ], batch size: 128, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:50:35,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9837, 1.2980, 3.3411, 2.9265], device='cuda:0'), covar=tensor([0.1777, 0.2732, 0.0482, 0.0945], device='cuda:0'), in_proj_covar=tensor([0.0720, 0.0623, 0.0917, 0.0849], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 04:51:09,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.552e+02 1.095e+03 1.386e+03 1.754e+03 3.344e+03, threshold=2.771e+03, percent-clipped=3.0
+2023-03-09 04:51:12,519 INFO [train.py:968] (0/2) Epoch 18, batch 1500, giga_loss[loss=0.2455, simple_loss=0.336, pruned_loss=0.07756, over 29063.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3491, pruned_loss=0.09518, over 5685604.72 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3422, pruned_loss=0.09003, over 3011587.15 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3491, pruned_loss=0.09555, over 5691579.50 frames. ], batch size: 155, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:51:34,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=778242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:51:52,737 INFO [train.py:968] (0/2) Epoch 18, batch 1550, giga_loss[loss=0.2739, simple_loss=0.3483, pruned_loss=0.09976, over 28868.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3464, pruned_loss=0.09309, over 5691765.13 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3413, pruned_loss=0.0894, over 3083315.10 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.347, pruned_loss=0.09377, over 5691883.50 frames. ], batch size: 186, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:52:33,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.148e+02 1.074e+03 1.422e+03 2.117e+03 4.819e+03, threshold=2.845e+03, percent-clipped=9.0
+2023-03-09 04:52:38,324 INFO [train.py:968] (0/2) Epoch 18, batch 1600, giga_loss[loss=0.2916, simple_loss=0.3647, pruned_loss=0.1093, over 28638.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3473, pruned_loss=0.09484, over 5702106.59 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3413, pruned_loss=0.08962, over 3135551.53 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3478, pruned_loss=0.09534, over 5701977.62 frames. ], batch size: 242, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:53:23,175 INFO [train.py:968] (0/2) Epoch 18, batch 1650, giga_loss[loss=0.3208, simple_loss=0.3754, pruned_loss=0.1331, over 27602.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3511, pruned_loss=0.1002, over 5703152.46 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3421, pruned_loss=0.0899, over 3204210.37 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3514, pruned_loss=0.1006, over 5698512.70 frames. ], batch size: 472, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:53:27,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5045, 1.7249, 1.4822, 1.2842], device='cuda:0'), covar=tensor([0.2614, 0.2244, 0.2150, 0.2400], device='cuda:0'), in_proj_covar=tensor([0.1879, 0.1812, 0.1743, 0.1881], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 04:53:41,053 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=778382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:53:42,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=778385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:53:42,998 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=778385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:53:45,011 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=778388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:53:58,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.734e+02 1.476e+03 1.881e+03 2.801e+03 7.154e+03, threshold=3.762e+03, percent-clipped=23.0
+2023-03-09 04:54:03,925 INFO [train.py:968] (0/2) Epoch 18, batch 1700, giga_loss[loss=0.2706, simple_loss=0.3477, pruned_loss=0.09677, over 28939.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3529, pruned_loss=0.1032, over 5692731.01 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3416, pruned_loss=0.08971, over 3297902.67 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3538, pruned_loss=0.1041, over 5692072.28 frames. ], batch size: 213, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:54:08,186 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=778417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:54:23,199 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9676, 1.3235, 1.1117, 0.1733], device='cuda:0'), covar=tensor([0.3684, 0.2719, 0.4263, 0.5649], device='cuda:0'), in_proj_covar=tensor([0.1672, 0.1579, 0.1560, 0.1362], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 04:54:32,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=778444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:54:48,054 INFO [train.py:968] (0/2) Epoch 18, batch 1750, giga_loss[loss=0.2539, simple_loss=0.3294, pruned_loss=0.08916, over 29118.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3516, pruned_loss=0.1031, over 5695612.74 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.342, pruned_loss=0.08998, over 3372845.65 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3523, pruned_loss=0.104, over 5691489.73 frames. ], batch size: 113, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:55:17,596 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=778500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:55:17,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6439, 1.7641, 1.9187, 1.4624], device='cuda:0'), covar=tensor([0.1767, 0.2386, 0.1399, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.0880, 0.0699, 0.0930, 0.0825], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 04:55:25,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.980e+02 1.202e+03 1.525e+03 1.882e+03 5.091e+03, threshold=3.050e+03, percent-clipped=3.0
+2023-03-09 04:55:28,585 INFO [train.py:968] (0/2) Epoch 18, batch 1800, giga_loss[loss=0.3025, simple_loss=0.3661, pruned_loss=0.1195, over 28937.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.35, pruned_loss=0.1019, over 5709305.49 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3425, pruned_loss=0.08995, over 3458707.33 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3506, pruned_loss=0.1031, over 5701076.03 frames. ], batch size: 106, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:55:38,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2350, 2.5943, 1.3072, 1.3238], device='cuda:0'), covar=tensor([0.1005, 0.0318, 0.0879, 0.1421], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0540, 0.0370, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 04:55:52,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9739, 1.2413, 1.3554, 1.0112], device='cuda:0'), covar=tensor([0.1772, 0.1306, 0.2163, 0.1592], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0737, 0.0697, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 04:56:12,022 INFO [train.py:968] (0/2) Epoch 18, batch 1850, giga_loss[loss=0.282, simple_loss=0.3593, pruned_loss=0.1024, over 28704.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3492, pruned_loss=0.1008, over 5712603.81 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3431, pruned_loss=0.09036, over 3507090.53 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3494, pruned_loss=0.1017, over 5702634.60 frames. ], batch size: 262, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:56:55,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.957e+02 1.106e+03 1.310e+03 1.891e+03 6.941e+03, threshold=2.621e+03, percent-clipped=7.0
+2023-03-09 04:56:59,620 INFO [train.py:968] (0/2) Epoch 18, batch 1900, giga_loss[loss=0.2364, simple_loss=0.3206, pruned_loss=0.07615, over 28929.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3464, pruned_loss=0.09817, over 5707562.75 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3429, pruned_loss=0.0903, over 3542515.96 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3468, pruned_loss=0.09904, over 5697430.72 frames. ], batch size: 213, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:57:05,507 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.58 vs. limit=5.0
+2023-03-09 04:57:45,466 INFO [train.py:968] (0/2) Epoch 18, batch 1950, giga_loss[loss=0.2532, simple_loss=0.3269, pruned_loss=0.08978, over 28863.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3438, pruned_loss=0.09647, over 5703460.27 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3428, pruned_loss=0.09005, over 3600784.77 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3442, pruned_loss=0.09748, over 5690887.42 frames. ], batch size: 227, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:57:53,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4369, 1.6377, 1.4624, 1.2932], device='cuda:0'), covar=tensor([0.2783, 0.2496, 0.2009, 0.2400], device='cuda:0'), in_proj_covar=tensor([0.1883, 0.1819, 0.1746, 0.1886], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 04:58:02,933 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 04:58:26,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.423e+02 1.063e+03 1.323e+03 1.912e+03 6.086e+03, threshold=2.646e+03, percent-clipped=11.0
+2023-03-09 04:58:30,020 INFO [train.py:968] (0/2) Epoch 18, batch 2000, giga_loss[loss=0.2243, simple_loss=0.3011, pruned_loss=0.07378, over 28411.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3382, pruned_loss=0.09349, over 5694625.36 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3421, pruned_loss=0.08958, over 3677488.06 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3389, pruned_loss=0.09475, over 5680757.11 frames. ], batch size: 369, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:59:03,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4701, 1.8229, 1.6191, 1.5683], device='cuda:0'), covar=tensor([0.0655, 0.0266, 0.0275, 0.0679], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-09 04:59:13,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=778757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:59:15,317 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=778760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:59:16,651 INFO [train.py:968] (0/2) Epoch 18, batch 2050, giga_loss[loss=0.2473, simple_loss=0.3209, pruned_loss=0.08688, over 28789.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3325, pruned_loss=0.09089, over 5687006.31 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3423, pruned_loss=0.08966, over 3721753.08 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3328, pruned_loss=0.09188, over 5672115.00 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:00:01,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.806e+02 9.705e+02 1.213e+03 1.914e+03 9.806e+03, threshold=2.427e+03, percent-clipped=9.0
+2023-03-09 05:00:03,685 INFO [train.py:968] (0/2) Epoch 18, batch 2100, libri_loss[loss=0.2753, simple_loss=0.3646, pruned_loss=0.09301, over 29495.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3316, pruned_loss=0.08982, over 5695601.25 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.08988, over 3814484.84 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3308, pruned_loss=0.0905, over 5677807.75 frames. ], batch size: 85, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:00:09,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=778819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:00:44,818 INFO [train.py:968] (0/2) Epoch 18, batch 2150, giga_loss[loss=0.2403, simple_loss=0.3249, pruned_loss=0.07784, over 28953.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3323, pruned_loss=0.08976, over 5699374.44 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3437, pruned_loss=0.09001, over 3854815.96 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3311, pruned_loss=0.09021, over 5682539.02 frames. ], batch size: 145, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:00:56,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=778875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:16,151 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=778900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:19,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=778903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:19,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=778903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:22,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=778906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:24,059 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.275e+02 1.047e+03 1.264e+03 1.653e+03 4.676e+03, threshold=2.527e+03, percent-clipped=5.0
+2023-03-09 05:01:26,602 INFO [train.py:968] (0/2) Epoch 18, batch 2200, giga_loss[loss=0.228, simple_loss=0.3019, pruned_loss=0.07704, over 28811.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3323, pruned_loss=0.08965, over 5705161.71 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3438, pruned_loss=0.0901, over 3895689.96 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.331, pruned_loss=0.08995, over 5688020.56 frames. ], batch size: 85, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:01:44,883 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=778932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:47,227 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=778935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:02:04,824 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-09 05:02:07,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4790, 1.1972, 4.7016, 3.5568], device='cuda:0'), covar=tensor([0.1739, 0.3000, 0.0366, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0719, 0.0622, 0.0916, 0.0849], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 05:02:09,619 INFO [train.py:968] (0/2) Epoch 18, batch 2250, giga_loss[loss=0.2246, simple_loss=0.3029, pruned_loss=0.07321, over 28725.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3309, pruned_loss=0.08947, over 5710041.92 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3437, pruned_loss=0.09012, over 3944281.51 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3297, pruned_loss=0.0897, over 5692846.09 frames. ], batch size: 92, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:02:09,995 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=778962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:02:12,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=778965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:02:17,511 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5046, 1.8024, 1.7546, 1.4498], device='cuda:0'), covar=tensor([0.3676, 0.2651, 0.2730, 0.3148], device='cuda:0'), in_proj_covar=tensor([0.1875, 0.1805, 0.1739, 0.1878], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 05:02:36,611 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=778994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:02:49,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.937e+02 1.070e+03 1.271e+03 1.629e+03 3.062e+03, threshold=2.543e+03, percent-clipped=6.0
+2023-03-09 05:02:53,384 INFO [train.py:968] (0/2) Epoch 18, batch 2300, giga_loss[loss=0.2576, simple_loss=0.3301, pruned_loss=0.09259, over 27698.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3285, pruned_loss=0.08874, over 5708756.32 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.09016, over 3973469.68 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3273, pruned_loss=0.08887, over 5692732.80 frames. ], batch size: 472, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:02:58,879 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=779018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:03:00,919 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=779021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:03:16,414 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 05:03:23,557 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=779050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:03:33,575 INFO [train.py:968] (0/2) Epoch 18, batch 2350, giga_loss[loss=0.2253, simple_loss=0.2991, pruned_loss=0.07573, over 28313.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3265, pruned_loss=0.0879, over 5720106.71 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3446, pruned_loss=0.09054, over 4000412.39 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3248, pruned_loss=0.08774, over 5706555.48 frames. ], batch size: 71, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:03:55,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5702, 1.7905, 1.7310, 1.5650], device='cuda:0'), covar=tensor([0.1947, 0.2199, 0.2376, 0.2218], device='cuda:0'), in_proj_covar=tensor([0.0465, 0.0742, 0.0700, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 05:04:17,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.970e+02 9.483e+02 1.208e+03 1.677e+03 4.504e+03, threshold=2.417e+03, percent-clipped=5.0
+2023-03-09 05:04:19,478 INFO [train.py:968] (0/2) Epoch 18, batch 2400, giga_loss[loss=0.2247, simple_loss=0.2929, pruned_loss=0.07821, over 28653.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.324, pruned_loss=0.08687, over 5714907.69 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3451, pruned_loss=0.0907, over 4016395.19 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3222, pruned_loss=0.08662, over 5712243.77 frames. ], batch size: 92, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:04:57,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2115, 1.4129, 1.3534, 1.1980], device='cuda:0'), covar=tensor([0.3394, 0.2560, 0.2066, 0.2711], device='cuda:0'), in_proj_covar=tensor([0.1868, 0.1795, 0.1731, 0.1872], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 05:04:58,168 INFO [train.py:968] (0/2) Epoch 18, batch 2450, libri_loss[loss=0.2748, simple_loss=0.3637, pruned_loss=0.09292, over 29660.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3219, pruned_loss=0.08564, over 5724737.68 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3456, pruned_loss=0.09076, over 4063408.05 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3196, pruned_loss=0.08531, over 5718168.75 frames. ], batch size: 88, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:05:20,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779194.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:05:32,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.299e+02 1.072e+03 1.399e+03 2.444e+03 9.775e+03, threshold=2.799e+03, percent-clipped=25.0
+2023-03-09 05:05:34,109 INFO [train.py:968] (0/2) Epoch 18, batch 2500, giga_loss[loss=0.2508, simple_loss=0.3212, pruned_loss=0.09023, over 28994.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.321, pruned_loss=0.08504, over 5729926.94 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3458, pruned_loss=0.09058, over 4151464.40 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3179, pruned_loss=0.08461, over 5718590.88 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:05:37,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3684, 1.5989, 1.2984, 1.3319], device='cuda:0'), covar=tensor([0.2410, 0.2505, 0.2764, 0.2213], device='cuda:0'), in_proj_covar=tensor([0.1447, 0.1047, 0.1282, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 05:06:17,052 INFO [train.py:968] (0/2) Epoch 18, batch 2550, giga_loss[loss=0.2428, simple_loss=0.311, pruned_loss=0.08729, over 28889.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3193, pruned_loss=0.08459, over 5725597.64 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3455, pruned_loss=0.09055, over 4186055.78 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3165, pruned_loss=0.08415, over 5713665.35 frames. ], batch size: 186, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:06:55,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.932e+02 9.898e+02 1.293e+03 1.772e+03 6.933e+03, threshold=2.587e+03, percent-clipped=12.0
+2023-03-09 05:06:57,928 INFO [train.py:968] (0/2) Epoch 18, batch 2600, giga_loss[loss=0.2163, simple_loss=0.2966, pruned_loss=0.06806, over 28967.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3181, pruned_loss=0.08363, over 5728922.08 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3459, pruned_loss=0.09036, over 4228294.92 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.315, pruned_loss=0.08327, over 5715620.51 frames. ], batch size: 213, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:07:33,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0847, 2.1417, 2.0965, 1.8758], device='cuda:0'), covar=tensor([0.2040, 0.2763, 0.2219, 0.2488], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0747, 0.0704, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 05:07:37,272 INFO [train.py:968] (0/2) Epoch 18, batch 2650, giga_loss[loss=0.2616, simple_loss=0.3264, pruned_loss=0.09841, over 28737.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3176, pruned_loss=0.08309, over 5723286.93 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3465, pruned_loss=0.09039, over 4260172.41 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.314, pruned_loss=0.08262, over 5718115.31 frames. ], batch size: 99, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:07:52,879 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 05:08:17,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.451e+02 1.039e+03 1.305e+03 1.657e+03 5.346e+03, threshold=2.610e+03, percent-clipped=7.0
+2023-03-09 05:08:19,363 INFO [train.py:968] (0/2) Epoch 18, batch 2700, giga_loss[loss=0.2259, simple_loss=0.3041, pruned_loss=0.07381, over 28849.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3195, pruned_loss=0.08441, over 5715275.11 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3471, pruned_loss=0.09045, over 4306486.23 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3155, pruned_loss=0.0838, over 5708386.10 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:08:41,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:08:45,025 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-09 05:09:02,891 INFO [train.py:968] (0/2) Epoch 18, batch 2750, giga_loss[loss=0.2515, simple_loss=0.3299, pruned_loss=0.0865, over 28987.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3249, pruned_loss=0.08778, over 5715973.78 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3471, pruned_loss=0.09031, over 4350783.54 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.321, pruned_loss=0.08727, over 5708225.82 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:09:49,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.995e+02 1.239e+03 1.480e+03 1.873e+03 3.857e+03, threshold=2.959e+03, percent-clipped=9.0
+2023-03-09 05:09:51,713 INFO [train.py:968] (0/2) Epoch 18, batch 2800, giga_loss[loss=0.316, simple_loss=0.384, pruned_loss=0.124, over 28008.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3319, pruned_loss=0.09241, over 5704826.77 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3475, pruned_loss=0.09043, over 4374067.39 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3283, pruned_loss=0.09192, over 5695918.85 frames. ], batch size: 412, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:10:10,374 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:10:37,968 INFO [train.py:968] (0/2) Epoch 18, batch 2850, giga_loss[loss=0.2882, simple_loss=0.3656, pruned_loss=0.1054, over 28520.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3393, pruned_loss=0.09706, over 5694892.80 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3477, pruned_loss=0.09054, over 4408258.04 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3361, pruned_loss=0.09672, over 5686421.60 frames. ], batch size: 336, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:10:43,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=779569.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:11:24,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0821, 3.8700, 3.6509, 1.6679], device='cuda:0'), covar=tensor([0.0663, 0.0824, 0.0818, 0.2407], device='cuda:0'), in_proj_covar=tensor([0.1147, 0.1058, 0.0910, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 05:11:25,126 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-09 05:11:25,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.444e+02 1.249e+03 1.563e+03 1.983e+03 6.481e+03, threshold=3.126e+03, percent-clipped=5.0
+2023-03-09 05:11:26,753 INFO [train.py:968] (0/2) Epoch 18, batch 2900, giga_loss[loss=0.3492, simple_loss=0.4044, pruned_loss=0.147, over 27909.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3448, pruned_loss=0.09966, over 5690958.68 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3469, pruned_loss=0.09024, over 4457801.72 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3425, pruned_loss=0.09984, over 5679513.82 frames. ], batch size: 412, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:12:13,300 INFO [train.py:968] (0/2) Epoch 18, batch 2950, giga_loss[loss=0.2745, simple_loss=0.3477, pruned_loss=0.1006, over 28770.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3493, pruned_loss=0.1014, over 5685894.30 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3468, pruned_loss=0.09028, over 4493977.18 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3476, pruned_loss=0.1018, over 5671294.26 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:12:34,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8541, 1.9076, 1.5967, 2.0973], device='cuda:0'), covar=tensor([0.2457, 0.2612, 0.2833, 0.2341], device='cuda:0'), in_proj_covar=tensor([0.1436, 0.1038, 0.1273, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 05:12:42,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1740, 1.1919, 3.7229, 3.1145], device='cuda:0'), covar=tensor([0.1665, 0.2848, 0.0431, 0.1249], device='cuda:0'), in_proj_covar=tensor([0.0719, 0.0618, 0.0910, 0.0843], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-09 05:13:02,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.278e+02 1.189e+03 1.478e+03 1.845e+03 2.974e+03, threshold=2.957e+03, percent-clipped=0.0
+2023-03-09 05:13:05,240 INFO [train.py:968] (0/2) Epoch 18, batch 3000, giga_loss[loss=0.3548, simple_loss=0.3913, pruned_loss=0.1591, over 23470.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3546, pruned_loss=0.1045, over 5680925.58 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3466, pruned_loss=0.09028, over 4498092.51 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3535, pruned_loss=0.1049, over 5675882.23 frames. ], batch size: 705, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:13:05,244 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 05:13:13,705 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2163, simple_loss=0.322, pruned_loss=0.0553, over 944034.00 frames. 
+2023-03-09 05:13:13,706 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 05:13:13,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=779712.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:13:16,132 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=779715.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:13:25,333 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3306, 1.8631, 1.3457, 0.5449], device='cuda:0'), covar=tensor([0.4652, 0.2301, 0.3634, 0.5588], device='cuda:0'), in_proj_covar=tensor([0.1676, 0.1585, 0.1566, 0.1373], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 05:13:41,104 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=779744.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:13:55,378 INFO [train.py:968] (0/2) Epoch 18, batch 3050, giga_loss[loss=0.2291, simple_loss=0.3122, pruned_loss=0.07298, over 28979.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3528, pruned_loss=0.103, over 5678834.51 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3466, pruned_loss=0.09049, over 4537271.09 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3522, pruned_loss=0.1036, over 5671085.04 frames. ], batch size: 227, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:14:17,458 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:14:36,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.957e+02 1.214e+03 1.661e+03 2.285e+03 9.580e+03, threshold=3.321e+03, percent-clipped=12.0
+2023-03-09 05:14:37,738 INFO [train.py:968] (0/2) Epoch 18, batch 3100, giga_loss[loss=0.2483, simple_loss=0.3362, pruned_loss=0.08022, over 28952.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.349, pruned_loss=0.09997, over 5683326.82 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.347, pruned_loss=0.09097, over 4573317.97 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3483, pruned_loss=0.1004, over 5675510.69 frames. ], batch size: 164, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:14:38,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=779812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:14:49,872 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:15:24,166 INFO [train.py:968] (0/2) Epoch 18, batch 3150, giga_loss[loss=0.2439, simple_loss=0.324, pruned_loss=0.08191, over 28524.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3471, pruned_loss=0.09832, over 5668862.36 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.347, pruned_loss=0.09106, over 4582662.86 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3466, pruned_loss=0.09868, over 5668540.42 frames. ], batch size: 85, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:15:28,223 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3242, 2.4936, 2.1521, 2.0436], device='cuda:0'), covar=tensor([0.1925, 0.2235, 0.2242, 0.2435], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0742, 0.0698, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 05:16:00,580 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3934, 1.6399, 1.5649, 1.5248], device='cuda:0'), covar=tensor([0.1851, 0.2161, 0.2216, 0.2031], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0744, 0.0699, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 05:16:03,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=779906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:16:07,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.942e+02 1.219e+03 1.597e+03 2.055e+03 4.961e+03, threshold=3.194e+03, percent-clipped=5.0
+2023-03-09 05:16:07,768 INFO [train.py:968] (0/2) Epoch 18, batch 3200, giga_loss[loss=0.3374, simple_loss=0.4007, pruned_loss=0.1371, over 28606.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3477, pruned_loss=0.09875, over 5671675.70 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3471, pruned_loss=0.09128, over 4614896.66 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3472, pruned_loss=0.09905, over 5666500.89 frames. ], batch size: 307, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:16:08,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0008, 3.1231, 2.0763, 1.0977], device='cuda:0'), covar=tensor([0.7711, 0.2762, 0.4040, 0.6675], device='cuda:0'), in_proj_covar=tensor([0.1680, 0.1587, 0.1572, 0.1376], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 05:16:10,927 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779916.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:16:45,142 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=779955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:16:47,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=779958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:16:50,407 INFO [train.py:968] (0/2) Epoch 18, batch 3250, giga_loss[loss=0.2808, simple_loss=0.3569, pruned_loss=0.1023, over 28545.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3503, pruned_loss=0.1003, over 5668309.24 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3474, pruned_loss=0.09152, over 4622119.24 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3497, pruned_loss=0.1004, over 5671430.30 frames. ], batch size: 78, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:17:12,102 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=779987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:17:22,436 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-780000.pt
+2023-03-09 05:17:32,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.215e+02 1.262e+03 1.605e+03 2.280e+03 4.259e+03, threshold=3.211e+03, percent-clipped=6.0
+2023-03-09 05:17:32,772 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:17:33,328 INFO [train.py:968] (0/2) Epoch 18, batch 3300, giga_loss[loss=0.2707, simple_loss=0.3462, pruned_loss=0.09758, over 28913.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3521, pruned_loss=0.1016, over 5677597.81 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.347, pruned_loss=0.09143, over 4649685.41 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.352, pruned_loss=0.102, over 5682715.83 frames. ], batch size: 112, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:18:07,754 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:18:09,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=780052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:18:17,761 INFO [train.py:968] (0/2) Epoch 18, batch 3350, giga_loss[loss=0.2685, simple_loss=0.3437, pruned_loss=0.09665, over 28841.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.353, pruned_loss=0.1026, over 5679166.01 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.347, pruned_loss=0.0914, over 4671619.07 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.353, pruned_loss=0.1031, over 5681657.95 frames. ], batch size: 99, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:18:29,286 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-09 05:18:32,520 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:18:43,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5391, 1.5960, 1.2005, 1.1554], device='cuda:0'), covar=tensor([0.0768, 0.0494, 0.0921, 0.1202], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0441, 0.0510, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 05:18:59,806 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.644e+02 1.283e+03 1.604e+03 2.016e+03 5.529e+03, threshold=3.209e+03, percent-clipped=8.0
+2023-03-09 05:19:02,115 INFO [train.py:968] (0/2) Epoch 18, batch 3400, giga_loss[loss=0.3766, simple_loss=0.4152, pruned_loss=0.169, over 26722.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3542, pruned_loss=0.1038, over 5674062.66 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3474, pruned_loss=0.09154, over 4701791.08 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3541, pruned_loss=0.1045, over 5676007.64 frames. ], batch size: 555, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:19:46,006 INFO [train.py:968] (0/2) Epoch 18, batch 3450, giga_loss[loss=0.2898, simple_loss=0.3602, pruned_loss=0.1097, over 28787.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3547, pruned_loss=0.1042, over 5675484.63 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3477, pruned_loss=0.09152, over 4731271.10 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 5672120.44 frames. ], batch size: 99, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:19:47,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:20:19,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:20:28,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.011e+02 1.101e+03 1.377e+03 1.937e+03 4.802e+03, threshold=2.755e+03, percent-clipped=4.0
+2023-03-09 05:20:29,535 INFO [train.py:968] (0/2) Epoch 18, batch 3500, giga_loss[loss=0.2734, simple_loss=0.3459, pruned_loss=0.1004, over 28714.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3535, pruned_loss=0.1025, over 5683340.21 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3473, pruned_loss=0.09123, over 4746862.82 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3539, pruned_loss=0.1036, over 5679177.94 frames. ], batch size: 92, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:20:38,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3687, 1.5344, 1.4972, 1.2591], device='cuda:0'), covar=tensor([0.2500, 0.2312, 0.1665, 0.2252], device='cuda:0'), in_proj_covar=tensor([0.1879, 0.1808, 0.1742, 0.1884], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 05:21:01,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-09 05:21:04,662 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-09 05:21:07,797 INFO [train.py:968] (0/2) Epoch 18, batch 3550, giga_loss[loss=0.2732, simple_loss=0.3538, pruned_loss=0.0963, over 28942.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3527, pruned_loss=0.1006, over 5695303.53 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3471, pruned_loss=0.09102, over 4793977.08 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3534, pruned_loss=0.1021, over 5686706.68 frames. ], batch size: 199, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:21:24,427 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:21:37,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780291.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:21:50,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:21:52,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=780310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:21:53,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.355e+02 1.143e+03 1.520e+03 2.035e+03 5.697e+03, threshold=3.039e+03, percent-clipped=7.0
+2023-03-09 05:21:53,750 INFO [train.py:968] (0/2) Epoch 18, batch 3600, giga_loss[loss=0.2422, simple_loss=0.3184, pruned_loss=0.08303, over 28860.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3526, pruned_loss=0.1, over 5695308.69 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3471, pruned_loss=0.09093, over 4810924.82 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3533, pruned_loss=0.1014, over 5685258.96 frames. ], batch size: 66, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:22:16,146 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:22:20,333 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:22:22,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=780347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:22:25,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1767, 1.2067, 3.7490, 3.0724], device='cuda:0'), covar=tensor([0.1631, 0.2754, 0.0407, 0.1182], device='cuda:0'), in_proj_covar=tensor([0.0719, 0.0617, 0.0910, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-09 05:22:34,525 INFO [train.py:968] (0/2) Epoch 18, batch 3650, giga_loss[loss=0.2745, simple_loss=0.329, pruned_loss=0.11, over 23521.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3514, pruned_loss=0.0995, over 5701567.92 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3476, pruned_loss=0.09137, over 4837549.58 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3517, pruned_loss=0.1005, over 5689420.78 frames. ], batch size: 705, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:22:46,744 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 05:22:47,319 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:22:55,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:23:17,685 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.612e+02 1.057e+03 1.273e+03 1.626e+03 5.272e+03, threshold=2.545e+03, percent-clipped=3.0
+2023-03-09 05:23:17,698 INFO [train.py:968] (0/2) Epoch 18, batch 3700, libri_loss[loss=0.2383, simple_loss=0.3232, pruned_loss=0.07666, over 29651.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3494, pruned_loss=0.09887, over 5687411.75 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3477, pruned_loss=0.09142, over 4851584.81 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3496, pruned_loss=0.09979, over 5688769.73 frames. ], batch size: 73, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:23:29,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2707, 1.1495, 3.7851, 3.2935], device='cuda:0'), covar=tensor([0.1626, 0.2804, 0.0423, 0.0862], device='cuda:0'), in_proj_covar=tensor([0.0718, 0.0615, 0.0908, 0.0844], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-09 05:23:34,993 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780434.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:23:35,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5211, 1.5252, 1.6267, 1.2057], device='cuda:0'), covar=tensor([0.2069, 0.3545, 0.1656, 0.1846], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0700, 0.0928, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 05:23:37,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=780437.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:23:54,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2207, 1.7213, 1.3860, 0.4365], device='cuda:0'), covar=tensor([0.3902, 0.2309, 0.3615, 0.5426], device='cuda:0'), in_proj_covar=tensor([0.1656, 0.1558, 0.1543, 0.1353], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 05:23:57,308 INFO [train.py:968] (0/2) Epoch 18, batch 3750, giga_loss[loss=0.3318, simple_loss=0.3723, pruned_loss=0.1456, over 26594.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3471, pruned_loss=0.09785, over 5694161.85 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3476, pruned_loss=0.09134, over 4861992.21 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3473, pruned_loss=0.0987, over 5693471.63 frames. ], batch size: 555, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:24:00,144 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780466.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:24:21,136 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-09 05:24:41,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.322e+02 9.813e+02 1.187e+03 1.560e+03 3.604e+03, threshold=2.373e+03, percent-clipped=1.0
+2023-03-09 05:24:41,377 INFO [train.py:968] (0/2) Epoch 18, batch 3800, giga_loss[loss=0.2978, simple_loss=0.3623, pruned_loss=0.1167, over 28740.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.348, pruned_loss=0.09887, over 5685334.86 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3474, pruned_loss=0.09129, over 4870478.56 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3483, pruned_loss=0.09968, over 5692672.66 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:24:46,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:24:49,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5860, 1.7806, 1.6191, 1.4082], device='cuda:0'), covar=tensor([0.2698, 0.2293, 0.2200, 0.2539], device='cuda:0'), in_proj_covar=tensor([0.1883, 0.1814, 0.1745, 0.1882], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 05:24:52,661 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780525.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:24:56,275 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:25:01,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=780532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:25:03,105 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:25:14,911 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:25:22,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:25:23,441 INFO [train.py:968] (0/2) Epoch 18, batch 3850, giga_loss[loss=0.2561, simple_loss=0.333, pruned_loss=0.08965, over 28676.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3483, pruned_loss=0.09902, over 5693442.14 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3474, pruned_loss=0.09132, over 4875681.17 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3486, pruned_loss=0.09967, over 5698287.02 frames. ], batch size: 92, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:26:05,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.217e+02 1.020e+03 1.240e+03 1.933e+03 7.920e+03, threshold=2.480e+03, percent-clipped=14.0
+2023-03-09 05:26:05,021 INFO [train.py:968] (0/2) Epoch 18, batch 3900, giga_loss[loss=0.2543, simple_loss=0.3372, pruned_loss=0.08566, over 28510.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3478, pruned_loss=0.09796, over 5699439.56 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3473, pruned_loss=0.09133, over 4890529.39 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3481, pruned_loss=0.09857, over 5700922.01 frames. ], batch size: 78, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:26:43,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:26:51,232 INFO [train.py:968] (0/2) Epoch 18, batch 3950, giga_loss[loss=0.2624, simple_loss=0.3408, pruned_loss=0.09196, over 28703.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3469, pruned_loss=0.09743, over 5696035.45 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3473, pruned_loss=0.09133, over 4890529.39 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3472, pruned_loss=0.09791, over 5697189.26 frames. ], batch size: 262, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:27:21,837 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 05:27:30,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.594e+02 9.845e+02 1.163e+03 1.549e+03 8.128e+03, threshold=2.326e+03, percent-clipped=5.0
+2023-03-09 05:27:30,598 INFO [train.py:968] (0/2) Epoch 18, batch 4000, giga_loss[loss=0.2279, simple_loss=0.3146, pruned_loss=0.07061, over 28345.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3472, pruned_loss=0.09777, over 5693913.25 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.347, pruned_loss=0.09121, over 4919589.35 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3475, pruned_loss=0.09847, over 5697762.49 frames. ], batch size: 71, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:28:09,267 INFO [train.py:968] (0/2) Epoch 18, batch 4050, giga_loss[loss=0.2516, simple_loss=0.3281, pruned_loss=0.08759, over 28925.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3456, pruned_loss=0.09716, over 5698863.29 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3477, pruned_loss=0.09173, over 4933685.22 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3453, pruned_loss=0.09742, over 5707325.56 frames. ], batch size: 227, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:28:11,490 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:28:38,504 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:28:40,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=780799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:28:50,379 INFO [train.py:968] (0/2) Epoch 18, batch 4100, giga_loss[loss=0.2869, simple_loss=0.3663, pruned_loss=0.1038, over 27860.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3428, pruned_loss=0.09572, over 5703939.51 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3478, pruned_loss=0.09177, over 4952037.49 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3424, pruned_loss=0.09598, over 5707789.80 frames. ], batch size: 412, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:28:50,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.015e+02 1.123e+03 1.349e+03 1.803e+03 9.437e+03, threshold=2.698e+03, percent-clipped=13.0
+2023-03-09 05:29:00,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4996, 1.6469, 1.7596, 1.3329], device='cuda:0'), covar=tensor([0.1845, 0.2617, 0.1463, 0.1718], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0701, 0.0930, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 05:29:02,379 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:29:18,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8209, 1.8606, 1.4248, 1.4623], device='cuda:0'), covar=tensor([0.0871, 0.0649, 0.1016, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0377, 0.0438, 0.0507, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 05:29:28,875 INFO [train.py:968] (0/2) Epoch 18, batch 4150, giga_loss[loss=0.2627, simple_loss=0.3363, pruned_loss=0.09458, over 28523.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3405, pruned_loss=0.09444, over 5702699.65 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3477, pruned_loss=0.09172, over 4974983.76 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3401, pruned_loss=0.09479, over 5708047.36 frames. ], batch size: 78, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:29:54,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:29:58,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780900.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:30:05,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:30:06,929 INFO [train.py:968] (0/2) Epoch 18, batch 4200, giga_loss[loss=0.29, simple_loss=0.364, pruned_loss=0.108, over 28596.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3406, pruned_loss=0.09463, over 5709415.63 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3478, pruned_loss=0.09175, over 4997564.00 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.34, pruned_loss=0.09496, over 5709439.64 frames. ], batch size: 336, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:30:09,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.804e+02 1.275e+03 1.573e+03 2.256e+03 5.324e+03, threshold=3.147e+03, percent-clipped=17.0
+2023-03-09 05:30:20,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:30:49,329 INFO [train.py:968] (0/2) Epoch 18, batch 4250, giga_loss[loss=0.2437, simple_loss=0.3162, pruned_loss=0.08556, over 28552.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3393, pruned_loss=0.0947, over 5711616.38 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.348, pruned_loss=0.09197, over 5015427.12 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3385, pruned_loss=0.09484, over 5708184.27 frames. ], batch size: 71, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:31:00,177 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-09 05:31:28,864 INFO [train.py:968] (0/2) Epoch 18, batch 4300, giga_loss[loss=0.2789, simple_loss=0.3538, pruned_loss=0.1019, over 28583.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3387, pruned_loss=0.09494, over 5704551.87 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3484, pruned_loss=0.09221, over 5031166.74 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3375, pruned_loss=0.09495, over 5705892.91 frames. ], batch size: 307, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:31:29,457 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.715e+02 1.052e+03 1.371e+03 1.812e+03 4.700e+03, threshold=2.742e+03, percent-clipped=5.0
+2023-03-09 05:31:50,390 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:31:52,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:31:55,025 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781043.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:31:56,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781046.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:32:01,562 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:04,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:08,672 INFO [train.py:968] (0/2) Epoch 18, batch 4350, giga_loss[loss=0.2181, simple_loss=0.2928, pruned_loss=0.07167, over 28837.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3373, pruned_loss=0.0947, over 5709128.53 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3485, pruned_loss=0.09239, over 5052810.15 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.336, pruned_loss=0.09462, over 5705574.17 frames. ], batch size: 99, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:32:12,483 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:13,005 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:15,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:19,647 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781075.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:32:27,866 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:29,562 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-09 05:32:39,190 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:48,039 INFO [train.py:968] (0/2) Epoch 18, batch 4400, libri_loss[loss=0.2163, simple_loss=0.3058, pruned_loss=0.06343, over 29563.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3349, pruned_loss=0.09339, over 5712929.62 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.348, pruned_loss=0.09224, over 5090081.91 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3338, pruned_loss=0.0935, over 5702098.19 frames. ], batch size: 76, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:32:48,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.671e+02 1.129e+03 1.526e+03 2.343e+03 6.054e+03, threshold=3.053e+03, percent-clipped=21.0
+2023-03-09 05:33:04,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:33:10,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:33:29,172 INFO [train.py:968] (0/2) Epoch 18, batch 4450, giga_loss[loss=0.2315, simple_loss=0.3198, pruned_loss=0.07163, over 29056.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3351, pruned_loss=0.09297, over 5706442.54 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3477, pruned_loss=0.0922, over 5099757.42 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3341, pruned_loss=0.09311, over 5702687.91 frames. ], batch size: 164, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:33:59,420 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-09 05:34:14,814 INFO [train.py:968] (0/2) Epoch 18, batch 4500, giga_loss[loss=0.2782, simple_loss=0.3541, pruned_loss=0.1012, over 28388.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3367, pruned_loss=0.0936, over 5710473.98 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3475, pruned_loss=0.09208, over 5114056.18 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3358, pruned_loss=0.09383, over 5705379.87 frames. ], batch size: 368, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:34:15,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.780e+02 1.038e+03 1.247e+03 1.817e+03 4.091e+03, threshold=2.493e+03, percent-clipped=2.0
+2023-03-09 05:34:22,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4437, 1.6979, 1.3573, 1.2711], device='cuda:0'), covar=tensor([0.2714, 0.2700, 0.3138, 0.2498], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1041, 0.1273, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 05:34:24,256 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:34:56,455 INFO [train.py:968] (0/2) Epoch 18, batch 4550, giga_loss[loss=0.2582, simple_loss=0.3378, pruned_loss=0.0893, over 28928.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.339, pruned_loss=0.09412, over 5713886.68 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3472, pruned_loss=0.09199, over 5125788.85 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3384, pruned_loss=0.0944, over 5710125.63 frames. ], batch size: 186, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:35:17,270 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:35:19,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:35:39,474 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:35:45,363 INFO [train.py:968] (0/2) Epoch 18, batch 4600, giga_loss[loss=0.2388, simple_loss=0.3248, pruned_loss=0.07642, over 28899.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3413, pruned_loss=0.09479, over 5703129.72 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3476, pruned_loss=0.0922, over 5133642.37 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3404, pruned_loss=0.09486, over 5698329.08 frames. ], batch size: 199, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:35:46,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.780e+02 1.011e+03 1.220e+03 1.652e+03 6.044e+03, threshold=2.439e+03, percent-clipped=11.0
+2023-03-09 05:35:48,126 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:35:57,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3082, 3.4064, 1.4403, 1.4702], device='cuda:0'), covar=tensor([0.0984, 0.0276, 0.0981, 0.1379], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0534, 0.0367, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 05:36:09,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5910, 1.8301, 1.7700, 1.5148], device='cuda:0'), covar=tensor([0.2336, 0.1801, 0.1458, 0.1867], device='cuda:0'), in_proj_covar=tensor([0.1877, 0.1818, 0.1747, 0.1883], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 05:36:27,610 INFO [train.py:968] (0/2) Epoch 18, batch 4650, giga_loss[loss=0.2643, simple_loss=0.3321, pruned_loss=0.09821, over 28553.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3393, pruned_loss=0.09289, over 5702966.95 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3472, pruned_loss=0.09214, over 5160200.19 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3386, pruned_loss=0.09303, over 5692917.02 frames. ], batch size: 85, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:37:09,546 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781411.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:37:09,974 INFO [train.py:968] (0/2) Epoch 18, batch 4700, giga_loss[loss=0.264, simple_loss=0.3407, pruned_loss=0.09359, over 29064.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3401, pruned_loss=0.09336, over 5706494.87 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.347, pruned_loss=0.09213, over 5167760.47 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3397, pruned_loss=0.09349, over 5696584.11 frames. ], batch size: 155, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:37:10,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.503e+02 1.102e+03 1.312e+03 1.943e+03 5.677e+03, threshold=2.623e+03, percent-clipped=15.0
+2023-03-09 05:37:52,491 INFO [train.py:968] (0/2) Epoch 18, batch 4750, giga_loss[loss=0.3048, simple_loss=0.3715, pruned_loss=0.1191, over 28260.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3415, pruned_loss=0.09447, over 5713900.60 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3471, pruned_loss=0.09221, over 5181883.49 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.341, pruned_loss=0.09455, over 5703039.80 frames. ], batch size: 368, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:38:29,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:38:33,380 INFO [train.py:968] (0/2) Epoch 18, batch 4800, giga_loss[loss=0.2721, simple_loss=0.3415, pruned_loss=0.1013, over 28565.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3424, pruned_loss=0.09545, over 5712523.71 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3471, pruned_loss=0.09211, over 5190056.96 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.342, pruned_loss=0.09565, over 5705869.25 frames. ], batch size: 85, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:38:35,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.164e+02 1.288e+03 1.615e+03 2.241e+03 5.111e+03, threshold=3.230e+03, percent-clipped=14.0
+2023-03-09 05:39:14,338 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:39:17,975 INFO [train.py:968] (0/2) Epoch 18, batch 4850, giga_loss[loss=0.2701, simple_loss=0.3503, pruned_loss=0.09492, over 28931.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3452, pruned_loss=0.09701, over 5712444.05 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3472, pruned_loss=0.09218, over 5199463.85 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3447, pruned_loss=0.09717, over 5705648.80 frames. ], batch size: 164, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:39:46,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:39:50,582 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:39:56,047 INFO [train.py:968] (0/2) Epoch 18, batch 4900, giga_loss[loss=0.3149, simple_loss=0.3802, pruned_loss=0.1248, over 28786.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3487, pruned_loss=0.09892, over 5705246.83 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3478, pruned_loss=0.09253, over 5216662.36 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3478, pruned_loss=0.09897, over 5703115.03 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:39:58,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.364e+02 1.378e+03 1.758e+03 2.316e+03 4.637e+03, threshold=3.516e+03, percent-clipped=6.0
+2023-03-09 05:40:17,790 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:40:27,322 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:40:29,434 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:40:38,139 INFO [train.py:968] (0/2) Epoch 18, batch 4950, giga_loss[loss=0.2544, simple_loss=0.3381, pruned_loss=0.08533, over 28915.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3495, pruned_loss=0.09894, over 5710322.64 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3479, pruned_loss=0.09258, over 5235116.30 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3487, pruned_loss=0.09913, over 5705245.92 frames. ], batch size: 145, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:40:53,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:40:54,898 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:41:20,437 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:41:20,887 INFO [train.py:968] (0/2) Epoch 18, batch 5000, giga_loss[loss=0.3238, simple_loss=0.3893, pruned_loss=0.1292, over 27719.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3497, pruned_loss=0.09904, over 5709919.62 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3477, pruned_loss=0.09246, over 5241888.16 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3493, pruned_loss=0.09934, over 5704081.42 frames. ], batch size: 472, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:41:22,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.987e+02 1.186e+03 1.367e+03 1.835e+03 3.278e+03, threshold=2.735e+03, percent-clipped=0.0
+2023-03-09 05:41:45,317 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:41:47,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:41:50,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2389, 4.0448, 3.8367, 2.0694], device='cuda:0'), covar=tensor([0.0598, 0.0715, 0.0741, 0.1849], device='cuda:0'), in_proj_covar=tensor([0.1149, 0.1060, 0.0914, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 05:42:01,600 INFO [train.py:968] (0/2) Epoch 18, batch 5050, giga_loss[loss=0.2617, simple_loss=0.3427, pruned_loss=0.09041, over 28688.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3512, pruned_loss=0.1001, over 5710103.11 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3482, pruned_loss=0.09274, over 5254911.16 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3505, pruned_loss=0.1002, over 5702008.92 frames. ], batch size: 242, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:42:11,353 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:42:20,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781786.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:42:42,900 INFO [train.py:968] (0/2) Epoch 18, batch 5100, giga_loss[loss=0.2729, simple_loss=0.3407, pruned_loss=0.1026, over 28758.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3485, pruned_loss=0.09812, over 5718537.45 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3476, pruned_loss=0.09234, over 5271267.95 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3485, pruned_loss=0.0987, over 5707488.28 frames. ], batch size: 92, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 05:42:45,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.659e+02 1.180e+03 1.464e+03 1.981e+03 7.237e+03, threshold=2.928e+03, percent-clipped=13.0
+2023-03-09 05:42:51,753 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:42:53,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:43:18,413 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:43:24,294 INFO [train.py:968] (0/2) Epoch 18, batch 5150, giga_loss[loss=0.2299, simple_loss=0.309, pruned_loss=0.07545, over 28765.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3462, pruned_loss=0.09719, over 5710430.37 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3478, pruned_loss=0.09244, over 5285560.31 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.346, pruned_loss=0.09769, over 5698734.92 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 05:43:37,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3077, 1.6740, 1.5556, 1.3916], device='cuda:0'), covar=tensor([0.1896, 0.1710, 0.2290, 0.2018], device='cuda:0'), in_proj_covar=tensor([0.0465, 0.0741, 0.0699, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 05:44:06,571 INFO [train.py:968] (0/2) Epoch 18, batch 5200, giga_loss[loss=0.2808, simple_loss=0.3567, pruned_loss=0.1025, over 28290.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3439, pruned_loss=0.09644, over 5714998.48 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3479, pruned_loss=0.0925, over 5300070.15 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3436, pruned_loss=0.09691, over 5704334.61 frames. ], batch size: 368, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:44:09,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.828e+02 1.188e+03 1.585e+03 2.195e+03 7.483e+03, threshold=3.169e+03, percent-clipped=11.0
+2023-03-09 05:44:17,272 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:44:20,345 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781929.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:44:22,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781932.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:44:23,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:44:46,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781961.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:44:46,971 INFO [train.py:968] (0/2) Epoch 18, batch 5250, giga_loss[loss=0.2951, simple_loss=0.3711, pruned_loss=0.1095, over 28798.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3443, pruned_loss=0.09623, over 5707124.25 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3484, pruned_loss=0.09282, over 5294291.33 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3436, pruned_loss=0.09638, over 5709128.43 frames. ], batch size: 112, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:45:03,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:45:21,742 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-782000.pt
+2023-03-09 05:45:29,059 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:45:30,925 INFO [train.py:968] (0/2) Epoch 18, batch 5300, giga_loss[loss=0.371, simple_loss=0.4334, pruned_loss=0.1543, over 28567.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3474, pruned_loss=0.09706, over 5711894.26 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3489, pruned_loss=0.09321, over 5312399.35 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3463, pruned_loss=0.09693, over 5708222.51 frames. ], batch size: 307, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:45:31,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:45:34,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.016e+02 1.113e+03 1.288e+03 1.775e+03 5.719e+03, threshold=2.575e+03, percent-clipped=3.0
+2023-03-09 05:45:41,118 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:46:13,890 INFO [train.py:968] (0/2) Epoch 18, batch 5350, giga_loss[loss=0.3751, simple_loss=0.4228, pruned_loss=0.1637, over 26681.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3475, pruned_loss=0.09661, over 5715084.44 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3489, pruned_loss=0.09316, over 5323872.49 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3466, pruned_loss=0.09663, over 5710683.57 frames. ], batch size: 555, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:46:27,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:46:29,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:46:34,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:46:53,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:46:55,837 INFO [train.py:968] (0/2) Epoch 18, batch 5400, giga_loss[loss=0.3238, simple_loss=0.3806, pruned_loss=0.1335, over 27931.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3454, pruned_loss=0.09644, over 5717684.99 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3489, pruned_loss=0.09317, over 5332250.59 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3446, pruned_loss=0.09648, over 5711986.17 frames. ], batch size: 412, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:46:58,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.379e+02 1.188e+03 1.416e+03 1.925e+03 3.544e+03, threshold=2.833e+03, percent-clipped=6.0
+2023-03-09 05:47:06,485 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:08,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:14,319 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:21,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6051, 0.9489, 5.4139, 3.6005], device='cuda:0'), covar=tensor([0.1628, 0.2964, 0.0335, 0.0945], device='cuda:0'), in_proj_covar=tensor([0.0721, 0.0621, 0.0916, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 05:47:31,137 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:32,830 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:34,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:36,616 INFO [train.py:968] (0/2) Epoch 18, batch 5450, giga_loss[loss=0.2876, simple_loss=0.3526, pruned_loss=0.1113, over 28638.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3457, pruned_loss=0.09797, over 5723384.28 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3492, pruned_loss=0.09334, over 5348571.99 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3446, pruned_loss=0.09798, over 5716107.99 frames. ], batch size: 242, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:47:58,852 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:48:19,268 INFO [train.py:968] (0/2) Epoch 18, batch 5500, giga_loss[loss=0.2273, simple_loss=0.2955, pruned_loss=0.07954, over 28408.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.342, pruned_loss=0.0972, over 5729216.45 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3488, pruned_loss=0.09317, over 5359943.23 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3414, pruned_loss=0.09747, over 5721531.39 frames. ], batch size: 78, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:48:23,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.399e+02 1.183e+03 1.507e+03 1.953e+03 5.298e+03, threshold=3.015e+03, percent-clipped=9.0
+2023-03-09 05:48:28,686 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:48:34,062 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:48:36,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:49:01,239 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:49:01,738 INFO [train.py:968] (0/2) Epoch 18, batch 5550, giga_loss[loss=0.2887, simple_loss=0.3655, pruned_loss=0.1059, over 28611.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3401, pruned_loss=0.09678, over 5732574.57 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3483, pruned_loss=0.09293, over 5370944.97 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3399, pruned_loss=0.09726, over 5723270.27 frames. ], batch size: 336, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:49:22,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5239, 1.8024, 1.4572, 1.5705], device='cuda:0'), covar=tensor([0.2448, 0.2483, 0.2900, 0.2213], device='cuda:0'), in_proj_covar=tensor([0.1438, 0.1039, 0.1273, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 05:49:35,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:49:44,742 INFO [train.py:968] (0/2) Epoch 18, batch 5600, giga_loss[loss=0.2395, simple_loss=0.3082, pruned_loss=0.08539, over 29001.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.34, pruned_loss=0.09744, over 5724002.32 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3478, pruned_loss=0.09288, over 5388620.41 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.34, pruned_loss=0.09802, over 5711606.10 frames. ], batch size: 106, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:49:47,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.022e+02 1.187e+03 1.508e+03 2.020e+03 6.495e+03, threshold=3.017e+03, percent-clipped=10.0
+2023-03-09 05:50:26,085 INFO [train.py:968] (0/2) Epoch 18, batch 5650, giga_loss[loss=0.293, simple_loss=0.351, pruned_loss=0.1175, over 26788.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3366, pruned_loss=0.09545, over 5720231.92 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3481, pruned_loss=0.09301, over 5405423.72 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.336, pruned_loss=0.09593, over 5706750.58 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:50:43,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:50:54,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:51:06,467 INFO [train.py:968] (0/2) Epoch 18, batch 5700, giga_loss[loss=0.2292, simple_loss=0.3049, pruned_loss=0.07678, over 29000.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.333, pruned_loss=0.09365, over 5725096.85 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3476, pruned_loss=0.09288, over 5424955.31 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3324, pruned_loss=0.09419, over 5707623.40 frames. ], batch size: 186, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:51:09,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.577e+02 1.116e+03 1.506e+03 2.020e+03 4.193e+03, threshold=3.011e+03, percent-clipped=4.0
+2023-03-09 05:51:32,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:51:34,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:51:37,445 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:51:44,724 INFO [train.py:968] (0/2) Epoch 18, batch 5750, giga_loss[loss=0.2419, simple_loss=0.3032, pruned_loss=0.09037, over 28715.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3309, pruned_loss=0.09233, over 5729066.06 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3474, pruned_loss=0.09295, over 5443670.51 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3299, pruned_loss=0.09273, over 5709828.83 frames. ], batch size: 92, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:51:59,378 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:12,111 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782495.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:52:20,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:23,372 INFO [train.py:968] (0/2) Epoch 18, batch 5800, libri_loss[loss=0.2745, simple_loss=0.3595, pruned_loss=0.09472, over 27727.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3318, pruned_loss=0.09255, over 5730230.91 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3476, pruned_loss=0.09312, over 5455493.92 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3304, pruned_loss=0.09271, over 5711563.09 frames. ], batch size: 116, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:52:27,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.820e+02 1.221e+03 1.477e+03 1.911e+03 5.848e+03, threshold=2.954e+03, percent-clipped=6.0
+2023-03-09 05:52:36,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:38,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:46,350 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782538.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:52:47,014 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:49,009 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:55,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4925, 2.1629, 1.5281, 0.5436], device='cuda:0'), covar=tensor([0.4391, 0.2772, 0.3999, 0.5755], device='cuda:0'), in_proj_covar=tensor([0.1674, 0.1578, 0.1551, 0.1371], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 05:53:02,941 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:53:05,545 INFO [train.py:968] (0/2) Epoch 18, batch 5850, giga_loss[loss=0.2627, simple_loss=0.3398, pruned_loss=0.09286, over 28946.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3344, pruned_loss=0.09335, over 5728710.47 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3472, pruned_loss=0.0929, over 5466479.53 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3333, pruned_loss=0.09368, over 5709800.05 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:53:10,401 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 05:53:13,259 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:53:17,180 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2658, 1.8868, 1.3959, 0.5311], device='cuda:0'), covar=tensor([0.4779, 0.2486, 0.3595, 0.5660], device='cuda:0'), in_proj_covar=tensor([0.1675, 0.1581, 0.1553, 0.1373], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 05:53:33,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:53:37,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4663, 1.6676, 1.6721, 1.2788], device='cuda:0'), covar=tensor([0.1775, 0.2414, 0.1506, 0.1646], device='cuda:0'), in_proj_covar=tensor([0.0876, 0.0696, 0.0923, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 05:53:46,114 INFO [train.py:968] (0/2) Epoch 18, batch 5900, libri_loss[loss=0.2892, simple_loss=0.3668, pruned_loss=0.1058, over 29546.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3371, pruned_loss=0.09397, over 5727913.81 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3469, pruned_loss=0.09267, over 5482547.70 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3361, pruned_loss=0.0945, over 5707026.02 frames. ], batch size: 84, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:53:48,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.732e+02 1.261e+03 1.715e+03 2.438e+03 6.585e+03, threshold=3.430e+03, percent-clipped=16.0
+2023-03-09 05:54:18,190 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:54:19,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2947, 1.2343, 3.9009, 3.2708], device='cuda:0'), covar=tensor([0.1644, 0.2844, 0.0415, 0.0971], device='cuda:0'), in_proj_covar=tensor([0.0727, 0.0626, 0.0926, 0.0861], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 05:54:20,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:54:28,347 INFO [train.py:968] (0/2) Epoch 18, batch 5950, giga_loss[loss=0.2574, simple_loss=0.3336, pruned_loss=0.09058, over 28479.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3402, pruned_loss=0.09494, over 5728207.03 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3471, pruned_loss=0.09282, over 5489717.59 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3391, pruned_loss=0.09528, over 5709641.30 frames. ], batch size: 78, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:54:46,969 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:55:14,479 INFO [train.py:968] (0/2) Epoch 18, batch 6000, giga_loss[loss=0.2734, simple_loss=0.338, pruned_loss=0.1044, over 23890.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3444, pruned_loss=0.09753, over 5720099.16 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3468, pruned_loss=0.09273, over 5497987.45 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3437, pruned_loss=0.09793, over 5702431.35 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:55:14,484 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 05:55:22,920 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2137, simple_loss=0.3209, pruned_loss=0.05322, over 944034.00 frames. 
+2023-03-09 05:55:22,921 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 05:55:27,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.252e+02 1.113e+03 1.387e+03 1.951e+03 5.898e+03, threshold=2.774e+03, percent-clipped=8.0
+2023-03-09 05:55:47,065 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:55:50,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:56:10,954 INFO [train.py:968] (0/2) Epoch 18, batch 6050, giga_loss[loss=0.2748, simple_loss=0.3566, pruned_loss=0.09643, over 28881.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3477, pruned_loss=0.1001, over 5713001.49 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3468, pruned_loss=0.09267, over 5501806.93 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3471, pruned_loss=0.1005, over 5697590.14 frames. ], batch size: 136, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:56:19,846 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:56:59,959 INFO [train.py:968] (0/2) Epoch 18, batch 6100, giga_loss[loss=0.3764, simple_loss=0.423, pruned_loss=0.1648, over 28501.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3546, pruned_loss=0.106, over 5702824.96 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3468, pruned_loss=0.09267, over 5507780.86 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3542, pruned_loss=0.1064, over 5688015.57 frames. ], batch size: 78, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:57:04,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.338e+02 1.637e+03 2.101e+03 2.956e+03 6.822e+03, threshold=4.202e+03, percent-clipped=29.0
+2023-03-09 05:57:08,168 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6063, 1.9279, 1.8184, 1.5410], device='cuda:0'), covar=tensor([0.2496, 0.2059, 0.2173, 0.2316], device='cuda:0'), in_proj_covar=tensor([0.1876, 0.1821, 0.1745, 0.1873], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 05:57:14,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:57:14,490 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-09 05:57:24,433 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:57:48,510 INFO [train.py:968] (0/2) Epoch 18, batch 6150, giga_loss[loss=0.3098, simple_loss=0.3845, pruned_loss=0.1175, over 28912.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.361, pruned_loss=0.1098, over 5710876.96 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3468, pruned_loss=0.09254, over 5517678.24 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3609, pruned_loss=0.1106, over 5694747.84 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:57:55,191 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782870.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:58:34,767 INFO [train.py:968] (0/2) Epoch 18, batch 6200, giga_loss[loss=0.3392, simple_loss=0.4004, pruned_loss=0.139, over 28657.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3669, pruned_loss=0.1147, over 5704347.86 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3473, pruned_loss=0.09282, over 5526105.85 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.367, pruned_loss=0.1158, over 5689463.07 frames. ], batch size: 92, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 05:58:37,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782913.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:58:43,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.711e+03 2.205e+03 2.762e+03 5.023e+03, threshold=4.410e+03, percent-clipped=4.0
+2023-03-09 05:58:46,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6127, 1.7330, 1.6484, 1.4369], device='cuda:0'), covar=tensor([0.2418, 0.2026, 0.1809, 0.2182], device='cuda:0'), in_proj_covar=tensor([0.1879, 0.1820, 0.1744, 0.1878], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 05:59:23,761 INFO [train.py:968] (0/2) Epoch 18, batch 6250, giga_loss[loss=0.299, simple_loss=0.3699, pruned_loss=0.114, over 28972.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3729, pruned_loss=0.12, over 5706258.57 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3472, pruned_loss=0.09269, over 5531801.76 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3735, pruned_loss=0.1215, over 5692500.14 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 05:59:32,342 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:59:34,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:59:35,640 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-09 05:59:39,393 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:00:00,111 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:00:10,082 INFO [train.py:968] (0/2) Epoch 18, batch 6300, giga_loss[loss=0.3517, simple_loss=0.4134, pruned_loss=0.1449, over 28887.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3783, pruned_loss=0.1242, over 5700586.55 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3476, pruned_loss=0.09287, over 5543047.66 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3793, pruned_loss=0.1262, over 5684618.90 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:00:14,183 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783013.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 06:00:15,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 06:00:17,530 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.582e+03 1.966e+03 2.656e+03 1.118e+04, threshold=3.933e+03, percent-clipped=4.0
+2023-03-09 06:00:17,968 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-09 06:00:45,179 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=783044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:00:45,942 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783045.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 06:00:57,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783056.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 06:01:01,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783059.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 06:01:03,167 INFO [train.py:968] (0/2) Epoch 18, batch 6350, giga_loss[loss=0.3715, simple_loss=0.398, pruned_loss=0.1725, over 23655.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3809, pruned_loss=0.1273, over 5677722.86 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3476, pruned_loss=0.0929, over 5543723.79 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3825, pruned_loss=0.1296, over 5667511.87 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:01:31,985 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783088.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 06:01:55,831 INFO [train.py:968] (0/2) Epoch 18, batch 6400, giga_loss[loss=0.3769, simple_loss=0.4177, pruned_loss=0.168, over 28235.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3832, pruned_loss=0.1302, over 5675290.08 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3477, pruned_loss=0.09298, over 5552615.55 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3852, pruned_loss=0.1329, over 5661925.85 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:02:01,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.800e+03 2.370e+03 3.574e+03 8.396e+03, threshold=4.740e+03, percent-clipped=16.0
+2023-03-09 06:02:10,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3634, 1.4276, 1.3522, 1.6178], device='cuda:0'), covar=tensor([0.0596, 0.0292, 0.0280, 0.0615], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-09 06:02:51,236 INFO [train.py:968] (0/2) Epoch 18, batch 6450, giga_loss[loss=0.29, simple_loss=0.3522, pruned_loss=0.1139, over 28383.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3857, pruned_loss=0.1333, over 5667894.68 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.348, pruned_loss=0.09318, over 5557490.72 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3877, pruned_loss=0.1359, over 5654975.33 frames. ], batch size: 65, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:02:55,520 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=783166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:03:42,289 INFO [train.py:968] (0/2) Epoch 18, batch 6500, giga_loss[loss=0.3503, simple_loss=0.4007, pruned_loss=0.15, over 28325.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3897, pruned_loss=0.1369, over 5656317.37 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3485, pruned_loss=0.09346, over 5567660.42 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3924, pruned_loss=0.1403, over 5640399.62 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:03:43,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=783213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:03:47,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.226e+03 1.773e+03 2.519e+03 3.175e+03 8.569e+03, threshold=5.038e+03, percent-clipped=12.0
+2023-03-09 06:04:29,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3776, 1.6315, 1.6370, 1.2210], device='cuda:0'), covar=tensor([0.1549, 0.2471, 0.1318, 0.1581], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0696, 0.0920, 0.0819], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 06:04:34,820 INFO [train.py:968] (0/2) Epoch 18, batch 6550, giga_loss[loss=0.3377, simple_loss=0.38, pruned_loss=0.1477, over 28453.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3902, pruned_loss=0.1381, over 5653001.09 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3485, pruned_loss=0.09342, over 5569285.12 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3924, pruned_loss=0.141, over 5639617.64 frames. ], batch size: 78, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:05:26,441 INFO [train.py:968] (0/2) Epoch 18, batch 6600, giga_loss[loss=0.3149, simple_loss=0.376, pruned_loss=0.1269, over 28659.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3895, pruned_loss=0.1386, over 5633542.14 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3489, pruned_loss=0.09375, over 5558600.58 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3919, pruned_loss=0.1417, over 5634449.09 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:05:32,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.891e+03 2.374e+03 3.517e+03 1.178e+04, threshold=4.749e+03, percent-clipped=8.0
+2023-03-09 06:06:07,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=783353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:06:10,246 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:06:13,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:06:17,868 INFO [train.py:968] (0/2) Epoch 18, batch 6650, giga_loss[loss=0.3384, simple_loss=0.3978, pruned_loss=0.1395, over 28965.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3878, pruned_loss=0.1376, over 5630786.12 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3489, pruned_loss=0.09372, over 5563754.13 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3907, pruned_loss=0.1412, over 5628516.19 frames. ], batch size: 106, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:06:42,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3396, 1.4775, 1.3433, 1.5228], device='cuda:0'), covar=tensor([0.0709, 0.0360, 0.0312, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-09 06:06:43,529 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:06:59,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3379, 1.2050, 3.6927, 3.1938], device='cuda:0'), covar=tensor([0.1577, 0.2757, 0.0489, 0.1105], device='cuda:0'), in_proj_covar=tensor([0.0729, 0.0627, 0.0926, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:07:09,205 INFO [train.py:968] (0/2) Epoch 18, batch 6700, giga_loss[loss=0.4093, simple_loss=0.4351, pruned_loss=0.1918, over 27578.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3874, pruned_loss=0.1359, over 5639006.38 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3486, pruned_loss=0.0936, over 5563974.36 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.39, pruned_loss=0.139, over 5637320.19 frames. ], batch size: 472, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:07:10,715 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-09 06:07:18,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=783419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:07:18,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.592e+03 2.104e+03 3.051e+03 8.266e+03, threshold=4.207e+03, percent-clipped=7.0
+2023-03-09 06:07:48,112 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=783447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:07:50,411 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 06:08:02,626 INFO [train.py:968] (0/2) Epoch 18, batch 6750, libri_loss[loss=0.2837, simple_loss=0.3566, pruned_loss=0.1054, over 29659.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3891, pruned_loss=0.1369, over 5634611.03 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3484, pruned_loss=0.09349, over 5569057.72 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.392, pruned_loss=0.1401, over 5629373.73 frames. ], batch size: 73, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:08:06,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4465, 1.7204, 1.5773, 1.3222], device='cuda:0'), covar=tensor([0.2350, 0.1932, 0.1486, 0.1928], device='cuda:0'), in_proj_covar=tensor([0.1882, 0.1825, 0.1746, 0.1881], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 06:08:14,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7350, 2.2301, 1.9455, 1.6111], device='cuda:0'), covar=tensor([0.0682, 0.0243, 0.0263, 0.0797], device='cuda:0'), in_proj_covar=tensor([0.0183, 0.0116, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:0')
+2023-03-09 06:08:37,187 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:08:40,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:08:50,767 INFO [train.py:968] (0/2) Epoch 18, batch 6800, giga_loss[loss=0.2569, simple_loss=0.3381, pruned_loss=0.08789, over 28895.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3864, pruned_loss=0.1345, over 5633719.44 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3487, pruned_loss=0.09354, over 5579411.20 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3896, pruned_loss=0.1383, over 5621619.53 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:08:58,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.632e+03 2.161e+03 2.751e+03 6.083e+03, threshold=4.323e+03, percent-clipped=6.0
+2023-03-09 06:09:10,022 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:09:14,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6778, 1.7180, 1.9827, 1.4612], device='cuda:0'), covar=tensor([0.1658, 0.2375, 0.1336, 0.1753], device='cuda:0'), in_proj_covar=tensor([0.0867, 0.0694, 0.0916, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 06:09:27,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=783541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:09:46,987 INFO [train.py:968] (0/2) Epoch 18, batch 6850, giga_loss[loss=0.2912, simple_loss=0.3675, pruned_loss=0.1075, over 28709.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3846, pruned_loss=0.1318, over 5640960.96 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3485, pruned_loss=0.09346, over 5581175.14 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3874, pruned_loss=0.135, over 5630095.90 frames. ], batch size: 92, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:09:47,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:09:51,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:10:08,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9102, 1.1735, 1.0378, 0.8953], device='cuda:0'), covar=tensor([0.2330, 0.2380, 0.1496, 0.1994], device='cuda:0'), in_proj_covar=tensor([0.1879, 0.1826, 0.1747, 0.1878], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 06:10:18,503 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:10:36,388 INFO [train.py:968] (0/2) Epoch 18, batch 6900, giga_loss[loss=0.3132, simple_loss=0.3769, pruned_loss=0.1247, over 28790.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3821, pruned_loss=0.1289, over 5650450.00 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3486, pruned_loss=0.0936, over 5588939.17 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3851, pruned_loss=0.1322, over 5636432.42 frames. ], batch size: 92, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:10:44,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.376e+02 1.703e+03 2.471e+03 3.907e+03 9.350e+03, threshold=4.943e+03, percent-clipped=18.0
+2023-03-09 06:11:16,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1299, 1.3193, 1.3639, 1.0539], device='cuda:0'), covar=tensor([0.1206, 0.1975, 0.1017, 0.1306], device='cuda:0'), in_proj_covar=tensor([0.0868, 0.0694, 0.0916, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 06:11:24,448 INFO [train.py:968] (0/2) Epoch 18, batch 6950, giga_loss[loss=0.3035, simple_loss=0.3677, pruned_loss=0.1197, over 28834.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3783, pruned_loss=0.1257, over 5658686.28 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3487, pruned_loss=0.0937, over 5594874.33 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3814, pruned_loss=0.129, over 5643567.84 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:11:47,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:11:51,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:12:14,450 INFO [train.py:968] (0/2) Epoch 18, batch 7000, giga_loss[loss=0.3376, simple_loss=0.4012, pruned_loss=0.137, over 28863.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3751, pruned_loss=0.1236, over 5655577.13 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3482, pruned_loss=0.09347, over 5601029.16 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3785, pruned_loss=0.127, over 5639088.94 frames. ], batch size: 136, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:12:18,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:12:22,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.603e+03 2.094e+03 3.027e+03 6.795e+03, threshold=4.187e+03, percent-clipped=7.0
+2023-03-09 06:12:48,005 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3201, 2.5889, 2.2975, 2.0654], device='cuda:0'), covar=tensor([0.2470, 0.1986, 0.1971, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.1876, 0.1823, 0.1745, 0.1876], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 06:13:04,540 INFO [train.py:968] (0/2) Epoch 18, batch 7050, giga_loss[loss=0.3671, simple_loss=0.4041, pruned_loss=0.1651, over 26669.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3746, pruned_loss=0.1234, over 5646439.34 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3484, pruned_loss=0.09359, over 5599108.48 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3777, pruned_loss=0.1266, over 5636740.15 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:13:46,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3153, 1.3122, 4.0245, 3.4334], device='cuda:0'), covar=tensor([0.1961, 0.2917, 0.0677, 0.1584], device='cuda:0'), in_proj_covar=tensor([0.0730, 0.0629, 0.0928, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:13:59,716 INFO [train.py:968] (0/2) Epoch 18, batch 7100, giga_loss[loss=0.3513, simple_loss=0.3994, pruned_loss=0.1516, over 27949.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3746, pruned_loss=0.1236, over 5650925.65 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3481, pruned_loss=0.0934, over 5604112.03 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3778, pruned_loss=0.1269, over 5639504.19 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:14:00,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5730, 1.8999, 1.5184, 1.7299], device='cuda:0'), covar=tensor([0.0735, 0.0280, 0.0313, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:0')
+2023-03-09 06:14:02,167 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 06:14:08,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.738e+02 1.391e+03 1.782e+03 2.698e+03 5.861e+03, threshold=3.565e+03, percent-clipped=9.0
+2023-03-09 06:14:13,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=783822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:14:48,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4373, 1.7601, 1.3839, 1.5148], device='cuda:0'), covar=tensor([0.2596, 0.2463, 0.2844, 0.2178], device='cuda:0'), in_proj_covar=tensor([0.1440, 0.1043, 0.1278, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 06:14:53,782 INFO [train.py:968] (0/2) Epoch 18, batch 7150, giga_loss[loss=0.2655, simple_loss=0.3478, pruned_loss=0.09164, over 29009.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3728, pruned_loss=0.1215, over 5656440.54 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3482, pruned_loss=0.0933, over 5611921.92 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.376, pruned_loss=0.1251, over 5641343.94 frames. ], batch size: 136, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:15:49,970 INFO [train.py:968] (0/2) Epoch 18, batch 7200, libri_loss[loss=0.2178, simple_loss=0.3046, pruned_loss=0.06553, over 29583.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3729, pruned_loss=0.119, over 5663083.65 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3476, pruned_loss=0.09302, over 5615549.05 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3766, pruned_loss=0.1228, over 5648978.59 frames. ], batch size: 75, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:15:57,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.558e+03 1.990e+03 2.737e+03 8.606e+03, threshold=3.979e+03, percent-clipped=10.0
+2023-03-09 06:16:37,566 INFO [train.py:968] (0/2) Epoch 18, batch 7250, giga_loss[loss=0.3344, simple_loss=0.391, pruned_loss=0.1389, over 28605.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3744, pruned_loss=0.1185, over 5676347.26 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3475, pruned_loss=0.09298, over 5621231.55 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.378, pruned_loss=0.1221, over 5661152.31 frames. ], batch size: 78, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:16:42,175 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:16:44,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:16:59,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8357, 2.2273, 1.9182, 1.5435], device='cuda:0'), covar=tensor([0.2665, 0.1978, 0.2385, 0.2583], device='cuda:0'), in_proj_covar=tensor([0.1873, 0.1818, 0.1738, 0.1873], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 06:17:18,781 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:17:22,754 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-784000.pt
+2023-03-09 06:17:32,716 INFO [train.py:968] (0/2) Epoch 18, batch 7300, giga_loss[loss=0.4062, simple_loss=0.4299, pruned_loss=0.1912, over 26710.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3752, pruned_loss=0.1197, over 5665488.72 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3478, pruned_loss=0.09309, over 5622452.79 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3782, pruned_loss=0.1229, over 5652880.15 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:17:41,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.599e+03 2.110e+03 2.671e+03 8.545e+03, threshold=4.220e+03, percent-clipped=8.0
+2023-03-09 06:18:17,705 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:18:21,640 INFO [train.py:968] (0/2) Epoch 18, batch 7350, giga_loss[loss=0.3945, simple_loss=0.4248, pruned_loss=0.1821, over 27676.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3761, pruned_loss=0.1217, over 5669604.84 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3476, pruned_loss=0.09295, over 5625333.91 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3789, pruned_loss=0.1246, over 5657445.95 frames. ], batch size: 472, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:19:12,188 INFO [train.py:968] (0/2) Epoch 18, batch 7400, libri_loss[loss=0.2741, simple_loss=0.3618, pruned_loss=0.09318, over 29537.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3742, pruned_loss=0.1212, over 5669772.10 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3479, pruned_loss=0.09306, over 5629034.39 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3769, pruned_loss=0.1241, over 5657758.15 frames. ], batch size: 79, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:19:19,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.673e+03 2.198e+03 2.800e+03 9.046e+03, threshold=4.396e+03, percent-clipped=3.0
+2023-03-09 06:19:41,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3035, 3.8745, 1.4886, 1.5313], device='cuda:0'), covar=tensor([0.1029, 0.0384, 0.0894, 0.1357], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0544, 0.0371, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 06:19:48,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5197, 3.8488, 1.7497, 1.6513], device='cuda:0'), covar=tensor([0.0995, 0.0312, 0.0848, 0.1352], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0544, 0.0370, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 06:19:54,324 INFO [train.py:968] (0/2) Epoch 18, batch 7450, libri_loss[loss=0.2889, simple_loss=0.3652, pruned_loss=0.1063, over 29776.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3721, pruned_loss=0.1205, over 5670355.02 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3478, pruned_loss=0.09311, over 5630595.14 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3753, pruned_loss=0.1238, over 5660194.11 frames. ], batch size: 87, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:20:46,574 INFO [train.py:968] (0/2) Epoch 18, batch 7500, giga_loss[loss=0.3153, simple_loss=0.3868, pruned_loss=0.1219, over 29011.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3716, pruned_loss=0.1194, over 5670761.25 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3481, pruned_loss=0.09314, over 5638157.52 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3746, pruned_loss=0.1228, over 5656597.40 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:20:52,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.882e+02 1.635e+03 2.071e+03 2.478e+03 6.536e+03, threshold=4.141e+03, percent-clipped=5.0
+2023-03-09 06:21:33,569 INFO [train.py:968] (0/2) Epoch 18, batch 7550, giga_loss[loss=0.277, simple_loss=0.3598, pruned_loss=0.09706, over 28997.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3713, pruned_loss=0.1181, over 5669981.46 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3478, pruned_loss=0.09297, over 5644300.96 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3746, pruned_loss=0.1217, over 5653847.63 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:22:13,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6617, 1.5714, 1.2106, 1.1830], device='cuda:0'), covar=tensor([0.0775, 0.0561, 0.0964, 0.1249], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0444, 0.0510, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:22:21,543 INFO [train.py:968] (0/2) Epoch 18, batch 7600, giga_loss[loss=0.2565, simple_loss=0.3347, pruned_loss=0.08913, over 29070.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3724, pruned_loss=0.1188, over 5678702.41 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3479, pruned_loss=0.09304, over 5645732.46 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.375, pruned_loss=0.1217, over 5664973.98 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:22:28,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.516e+03 2.034e+03 2.760e+03 5.066e+03, threshold=4.069e+03, percent-clipped=5.0
+2023-03-09 06:23:07,616 INFO [train.py:968] (0/2) Epoch 18, batch 7650, giga_loss[loss=0.3201, simple_loss=0.3833, pruned_loss=0.1284, over 28916.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3695, pruned_loss=0.1165, over 5696677.88 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3478, pruned_loss=0.09297, over 5652265.36 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3722, pruned_loss=0.1194, over 5680714.18 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:23:53,675 INFO [train.py:968] (0/2) Epoch 18, batch 7700, giga_loss[loss=0.3377, simple_loss=0.3742, pruned_loss=0.1506, over 23475.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3678, pruned_loss=0.1163, over 5683205.14 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3475, pruned_loss=0.09282, over 5658262.94 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.371, pruned_loss=0.1197, over 5666159.00 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:24:08,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.002e+02 1.555e+03 2.385e+03 3.345e+03 8.566e+03, threshold=4.771e+03, percent-clipped=19.0
+2023-03-09 06:24:10,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1018, 4.9313, 4.6755, 2.2859], device='cuda:0'), covar=tensor([0.0447, 0.0574, 0.0628, 0.1844], device='cuda:0'), in_proj_covar=tensor([0.1172, 0.1084, 0.0930, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 06:24:19,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=784432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:24:41,749 INFO [train.py:968] (0/2) Epoch 18, batch 7750, libri_loss[loss=0.2667, simple_loss=0.3418, pruned_loss=0.0958, over 29561.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3659, pruned_loss=0.1156, over 5679486.20 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3474, pruned_loss=0.09295, over 5665047.33 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3694, pruned_loss=0.1192, over 5659762.74 frames. ], batch size: 76, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:25:15,081 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:25:36,432 INFO [train.py:968] (0/2) Epoch 18, batch 7800, giga_loss[loss=0.317, simple_loss=0.3762, pruned_loss=0.1289, over 27912.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3663, pruned_loss=0.1171, over 5671499.35 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3472, pruned_loss=0.0928, over 5667617.18 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3695, pruned_loss=0.1203, over 5653657.04 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:25:47,104 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.692e+03 2.069e+03 2.717e+03 5.179e+03, threshold=4.138e+03, percent-clipped=1.0
+2023-03-09 06:26:25,473 INFO [train.py:968] (0/2) Epoch 18, batch 7850, giga_loss[loss=0.3034, simple_loss=0.3768, pruned_loss=0.115, over 28930.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3668, pruned_loss=0.1184, over 5663036.11 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3476, pruned_loss=0.09294, over 5669433.82 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3694, pruned_loss=0.1214, over 5647308.61 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:26:38,348 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=784575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:26:41,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=784578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:26:49,451 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:27:02,292 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:27:06,835 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=784607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:27:13,260 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:27:13,671 INFO [train.py:968] (0/2) Epoch 18, batch 7900, giga_loss[loss=0.304, simple_loss=0.3703, pruned_loss=0.1188, over 28702.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3661, pruned_loss=0.1182, over 5662947.22 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3473, pruned_loss=0.09289, over 5672958.68 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3686, pruned_loss=0.121, over 5647193.78 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:27:24,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.711e+03 2.152e+03 2.938e+03 1.046e+04, threshold=4.304e+03, percent-clipped=12.0
+2023-03-09 06:28:04,353 INFO [train.py:968] (0/2) Epoch 18, batch 7950, giga_loss[loss=0.2769, simple_loss=0.3508, pruned_loss=0.1015, over 28664.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3674, pruned_loss=0.1188, over 5664404.04 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3479, pruned_loss=0.0933, over 5676392.38 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3691, pruned_loss=0.121, over 5648902.37 frames. ], batch size: 99, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:28:24,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3273, 1.7177, 1.4927, 1.5135], device='cuda:0'), covar=tensor([0.0758, 0.0319, 0.0310, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0067, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 06:28:52,466 INFO [train.py:968] (0/2) Epoch 18, batch 8000, giga_loss[loss=0.3012, simple_loss=0.3721, pruned_loss=0.1151, over 28746.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3671, pruned_loss=0.1174, over 5669821.11 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3476, pruned_loss=0.09309, over 5676220.56 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.369, pruned_loss=0.1196, over 5657548.92 frames. ], batch size: 284, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:29:03,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.942e+02 1.530e+03 1.986e+03 2.969e+03 7.114e+03, threshold=3.972e+03, percent-clipped=9.0
+2023-03-09 06:29:13,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9906, 1.9942, 1.5530, 1.7057], device='cuda:0'), covar=tensor([0.0976, 0.0810, 0.1066, 0.1267], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0444, 0.0512, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:29:31,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7258, 1.7932, 1.9377, 1.4851], device='cuda:0'), covar=tensor([0.1938, 0.2451, 0.1511, 0.1765], device='cuda:0'), in_proj_covar=tensor([0.0874, 0.0701, 0.0921, 0.0820], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 06:29:39,342 INFO [train.py:968] (0/2) Epoch 18, batch 8050, giga_loss[loss=0.3107, simple_loss=0.3723, pruned_loss=0.1246, over 27959.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3663, pruned_loss=0.1156, over 5684370.31 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3475, pruned_loss=0.09297, over 5681710.27 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3684, pruned_loss=0.1181, over 5669365.39 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:30:23,079 INFO [train.py:968] (0/2) Epoch 18, batch 8100, giga_loss[loss=0.3018, simple_loss=0.3657, pruned_loss=0.1189, over 28901.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3663, pruned_loss=0.1155, over 5683083.31 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3477, pruned_loss=0.09316, over 5680682.04 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3688, pruned_loss=0.1184, over 5671568.08 frames. ], batch size: 99, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:30:33,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.482e+02 1.474e+03 1.996e+03 2.718e+03 1.183e+04, threshold=3.993e+03, percent-clipped=7.0
+2023-03-09 06:31:13,790 INFO [train.py:968] (0/2) Epoch 18, batch 8150, giga_loss[loss=0.4558, simple_loss=0.4788, pruned_loss=0.2163, over 26684.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3694, pruned_loss=0.1186, over 5665630.20 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3478, pruned_loss=0.09329, over 5675391.56 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3715, pruned_loss=0.1211, over 5660813.25 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:31:19,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=784868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:32:03,846 INFO [train.py:968] (0/2) Epoch 18, batch 8200, giga_loss[loss=0.286, simple_loss=0.3544, pruned_loss=0.1087, over 28576.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3709, pruned_loss=0.1212, over 5658799.50 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.348, pruned_loss=0.09339, over 5671561.73 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.373, pruned_loss=0.1238, over 5657733.65 frames. ], batch size: 85, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:32:16,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.136e+03 1.806e+03 2.182e+03 2.810e+03 9.908e+03, threshold=4.363e+03, percent-clipped=12.0
+2023-03-09 06:32:27,434 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6350, 1.7331, 1.7220, 1.5253], device='cuda:0'), covar=tensor([0.1666, 0.2102, 0.2072, 0.2134], device='cuda:0'), in_proj_covar=tensor([0.0462, 0.0747, 0.0701, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 06:32:40,039 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:32:53,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=784960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:32:54,077 INFO [train.py:968] (0/2) Epoch 18, batch 8250, giga_loss[loss=0.2901, simple_loss=0.3584, pruned_loss=0.1109, over 28708.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.372, pruned_loss=0.1231, over 5653025.44 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.348, pruned_loss=0.09321, over 5671912.20 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3743, pruned_loss=0.1262, over 5651412.84 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:33:06,961 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:10,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=784976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:20,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=784986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:30,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5957, 1.6914, 1.7098, 1.5499], device='cuda:0'), covar=tensor([0.2429, 0.2373, 0.1813, 0.2097], device='cuda:0'), in_proj_covar=tensor([0.1881, 0.1828, 0.1747, 0.1879], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 06:33:31,312 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 06:33:44,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:44,602 INFO [train.py:968] (0/2) Epoch 18, batch 8300, giga_loss[loss=0.4399, simple_loss=0.4634, pruned_loss=0.2082, over 26409.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3732, pruned_loss=0.1249, over 5656726.35 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3481, pruned_loss=0.09322, over 5675622.01 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3755, pruned_loss=0.128, over 5651864.56 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:33:46,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:54,577 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=785020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:57,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.933e+02 1.777e+03 2.260e+03 2.950e+03 1.048e+04, threshold=4.520e+03, percent-clipped=6.0
+2023-03-09 06:34:15,125 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:34:32,875 INFO [train.py:968] (0/2) Epoch 18, batch 8350, libri_loss[loss=0.2865, simple_loss=0.3684, pruned_loss=0.1023, over 29541.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3734, pruned_loss=0.1248, over 5662056.67 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3483, pruned_loss=0.09327, over 5680681.84 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3757, pruned_loss=0.128, over 5653242.81 frames. ], batch size: 84, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:34:38,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3213, 1.4926, 1.5523, 1.3040], device='cuda:0'), covar=tensor([0.1627, 0.1607, 0.2152, 0.1802], device='cuda:0'), in_proj_covar=tensor([0.0465, 0.0750, 0.0703, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 06:35:10,120 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:12,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:18,956 INFO [train.py:968] (0/2) Epoch 18, batch 8400, giga_loss[loss=0.2855, simple_loss=0.3613, pruned_loss=0.1048, over 28912.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3718, pruned_loss=0.1233, over 5674336.45 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3481, pruned_loss=0.09317, over 5684356.12 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3742, pruned_loss=0.1265, over 5663925.92 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:35:22,922 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-09 06:35:23,848 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:27,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:27,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.618e+03 2.107e+03 3.006e+03 9.364e+03, threshold=4.215e+03, percent-clipped=8.0
+2023-03-09 06:35:32,498 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:35,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:37,551 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:53,705 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:36:02,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:36:02,701 INFO [train.py:968] (0/2) Epoch 18, batch 8450, giga_loss[loss=0.2878, simple_loss=0.3537, pruned_loss=0.111, over 28672.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3699, pruned_loss=0.1199, over 5677547.10 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3479, pruned_loss=0.09317, over 5679657.50 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3724, pruned_loss=0.1229, over 5672708.32 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:36:48,607 INFO [train.py:968] (0/2) Epoch 18, batch 8500, giga_loss[loss=0.3513, simple_loss=0.3828, pruned_loss=0.1599, over 26602.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3689, pruned_loss=0.1193, over 5667352.72 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3485, pruned_loss=0.09353, over 5675895.01 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3707, pruned_loss=0.1219, over 5665895.82 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:36:56,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.577e+02 1.523e+03 1.898e+03 2.362e+03 7.752e+03, threshold=3.797e+03, percent-clipped=3.0
+2023-03-09 06:37:22,314 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-09 06:37:33,630 INFO [train.py:968] (0/2) Epoch 18, batch 8550, giga_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1146, over 28571.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3661, pruned_loss=0.1177, over 5672188.69 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.0934, over 5682021.67 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3682, pruned_loss=0.1205, over 5665596.96 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:37:40,981 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 06:37:58,676 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.45 vs. limit=5.0
+2023-03-09 06:38:25,554 INFO [train.py:968] (0/2) Epoch 18, batch 8600, giga_loss[loss=0.33, simple_loss=0.3969, pruned_loss=0.1316, over 28914.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3657, pruned_loss=0.1181, over 5671970.36 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3485, pruned_loss=0.09346, over 5683949.78 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3673, pruned_loss=0.1205, over 5664980.18 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:38:34,831 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.949e+02 1.582e+03 2.005e+03 2.682e+03 6.382e+03, threshold=4.010e+03, percent-clipped=11.0
+2023-03-09 06:38:35,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=785324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:38:59,754 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=785347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:39:13,117 INFO [train.py:968] (0/2) Epoch 18, batch 8650, giga_loss[loss=0.4193, simple_loss=0.4311, pruned_loss=0.2037, over 23607.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3674, pruned_loss=0.1193, over 5671082.66 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3478, pruned_loss=0.09307, over 5684525.47 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3701, pruned_loss=0.1225, over 5663764.68 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:39:24,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4326, 1.6184, 1.2574, 1.1836], device='cuda:0'), covar=tensor([0.0964, 0.0562, 0.1064, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0443, 0.0510, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:39:45,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=785395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:39:58,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=785409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:40:00,712 INFO [train.py:968] (0/2) Epoch 18, batch 8700, giga_loss[loss=0.3016, simple_loss=0.3798, pruned_loss=0.1117, over 28886.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3716, pruned_loss=0.1197, over 5677375.22 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.348, pruned_loss=0.09305, over 5687707.85 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3738, pruned_loss=0.1227, over 5668635.20 frames. ], batch size: 227, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:40:15,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4626, 1.7858, 1.3997, 1.5759], device='cuda:0'), covar=tensor([0.2691, 0.2619, 0.3045, 0.2248], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1042, 0.1274, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 06:40:15,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.753e+02 1.367e+03 1.840e+03 2.417e+03 7.540e+03, threshold=3.681e+03, percent-clipped=7.0
+2023-03-09 06:40:22,463 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2976, 1.6294, 1.4399, 1.4472], device='cuda:0'), covar=tensor([0.0763, 0.0390, 0.0332, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 06:40:36,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3834, 1.5831, 1.5108, 1.3307], device='cuda:0'), covar=tensor([0.2386, 0.2028, 0.1852, 0.2015], device='cuda:0'), in_proj_covar=tensor([0.1896, 0.1837, 0.1758, 0.1892], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 06:40:51,097 INFO [train.py:968] (0/2) Epoch 18, batch 8750, giga_loss[loss=0.3126, simple_loss=0.3784, pruned_loss=0.1235, over 28315.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3736, pruned_loss=0.1194, over 5676550.95 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3471, pruned_loss=0.0925, over 5693305.55 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3768, pruned_loss=0.123, over 5664180.33 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:40:56,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:40:59,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:41:18,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:41:20,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:41:27,261 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:41:38,712 INFO [train.py:968] (0/2) Epoch 18, batch 8800, giga_loss[loss=0.3211, simple_loss=0.3907, pruned_loss=0.1258, over 28987.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3754, pruned_loss=0.1209, over 5667413.86 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.347, pruned_loss=0.09252, over 5682838.26 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3786, pruned_loss=0.1243, over 5666871.29 frames. ], batch size: 136, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:41:48,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:41:50,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.506e+02 1.758e+03 2.276e+03 3.358e+03 8.166e+03, threshold=4.553e+03, percent-clipped=20.0
+2023-03-09 06:41:59,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:42:03,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:42:21,858 INFO [train.py:968] (0/2) Epoch 18, batch 8850, giga_loss[loss=0.3114, simple_loss=0.3749, pruned_loss=0.124, over 28812.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3757, pruned_loss=0.1212, over 5679552.97 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.347, pruned_loss=0.09252, over 5686999.88 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3789, pruned_loss=0.1246, over 5675185.35 frames. ], batch size: 284, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:42:32,190 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:42:53,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-09 06:43:07,891 INFO [train.py:968] (0/2) Epoch 18, batch 8900, libri_loss[loss=0.2604, simple_loss=0.3464, pruned_loss=0.08726, over 27887.00 frames. ], tot_loss[loss=0.309, simple_loss=0.375, pruned_loss=0.1215, over 5673800.21 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3465, pruned_loss=0.09231, over 5680165.06 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.379, pruned_loss=0.1254, over 5676049.21 frames. ], batch size: 115, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:43:19,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.688e+03 2.267e+03 3.188e+03 8.051e+03, threshold=4.534e+03, percent-clipped=10.0
+2023-03-09 06:43:35,804 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-09 06:43:58,382 INFO [train.py:968] (0/2) Epoch 18, batch 8950, giga_loss[loss=0.2931, simple_loss=0.3714, pruned_loss=0.1074, over 28673.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3745, pruned_loss=0.1221, over 5680854.07 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3469, pruned_loss=0.09252, over 5683762.05 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3778, pruned_loss=0.1254, over 5679531.74 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:44:25,618 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5243, 1.6598, 1.3076, 1.5559], device='cuda:0'), covar=tensor([0.0740, 0.0313, 0.0323, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 06:44:49,162 INFO [train.py:968] (0/2) Epoch 18, batch 9000, giga_loss[loss=0.3005, simple_loss=0.351, pruned_loss=0.1249, over 23655.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3719, pruned_loss=0.1209, over 5681372.68 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3465, pruned_loss=0.09231, over 5687059.64 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3753, pruned_loss=0.1243, over 5677238.86 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:44:49,166 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 06:44:57,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1360, 1.5724, 1.6813, 1.3822], device='cuda:0'), covar=tensor([0.1796, 0.1407, 0.1933, 0.1612], device='cuda:0'), in_proj_covar=tensor([0.0463, 0.0747, 0.0700, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 06:44:57,691 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2106, simple_loss=0.3183, pruned_loss=0.05144, over 944034.00 frames. 
+2023-03-09 06:44:57,691 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 06:45:10,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.577e+02 1.520e+03 1.998e+03 2.948e+03 9.066e+03, threshold=3.995e+03, percent-clipped=8.0
+2023-03-09 06:45:10,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6366, 1.8674, 1.5194, 1.7933], device='cuda:0'), covar=tensor([0.2507, 0.2591, 0.2884, 0.2336], device='cuda:0'), in_proj_covar=tensor([0.1438, 0.1044, 0.1272, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0011, 0.0009], device='cuda:0')
+2023-03-09 06:45:26,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8989, 1.9666, 1.4707, 1.4828], device='cuda:0'), covar=tensor([0.0905, 0.0634, 0.1043, 0.1153], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0446, 0.0511, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:45:45,785 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 06:45:46,914 INFO [train.py:968] (0/2) Epoch 18, batch 9050, giga_loss[loss=0.269, simple_loss=0.3403, pruned_loss=0.09886, over 28898.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3705, pruned_loss=0.1211, over 5666736.88 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3465, pruned_loss=0.0923, over 5677817.57 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3736, pruned_loss=0.1242, over 5671715.24 frames. ], batch size: 199, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:46:09,149 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=785784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:46:37,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2472, 1.4965, 1.2268, 1.1368], device='cuda:0'), covar=tensor([0.2435, 0.2536, 0.2808, 0.2090], device='cuda:0'), in_proj_covar=tensor([0.1437, 0.1043, 0.1272, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 06:46:39,395 INFO [train.py:968] (0/2) Epoch 18, batch 9100, giga_loss[loss=0.3184, simple_loss=0.3751, pruned_loss=0.1309, over 28773.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3691, pruned_loss=0.1201, over 5673967.05 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3468, pruned_loss=0.09234, over 5680140.32 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3717, pruned_loss=0.1229, over 5675707.52 frames. ], batch size: 284, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:46:48,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9764, 1.2335, 1.3030, 1.0908], device='cuda:0'), covar=tensor([0.1511, 0.1254, 0.1955, 0.1464], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0751, 0.0703, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 06:46:49,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-09 06:46:51,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.611e+03 2.038e+03 2.828e+03 1.182e+04, threshold=4.077e+03, percent-clipped=11.0
+2023-03-09 06:46:57,474 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3161, 1.7143, 1.3701, 1.6040], device='cuda:0'), covar=tensor([0.0757, 0.0323, 0.0321, 0.0797], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 06:47:22,685 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=785855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:47:29,314 INFO [train.py:968] (0/2) Epoch 18, batch 9150, giga_loss[loss=0.2725, simple_loss=0.3461, pruned_loss=0.09942, over 28831.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3696, pruned_loss=0.1208, over 5670107.40 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3467, pruned_loss=0.09224, over 5683803.19 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.372, pruned_loss=0.1237, over 5667943.95 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:47:37,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7051, 1.8833, 1.3053, 1.6111], device='cuda:0'), covar=tensor([0.0983, 0.0775, 0.1193, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0444, 0.0508, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:47:55,172 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6908, 1.6689, 1.2491, 1.2750], device='cuda:0'), covar=tensor([0.0840, 0.0636, 0.1062, 0.1179], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0443, 0.0507, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:48:14,585 INFO [train.py:968] (0/2) Epoch 18, batch 9200, giga_loss[loss=0.3032, simple_loss=0.3681, pruned_loss=0.1192, over 28632.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3679, pruned_loss=0.1198, over 5676152.76 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3467, pruned_loss=0.09218, over 5689043.00 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3707, pruned_loss=0.1233, over 5669190.63 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:48:27,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.681e+02 1.608e+03 2.099e+03 2.743e+03 7.786e+03, threshold=4.197e+03, percent-clipped=8.0
+2023-03-09 06:48:28,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:48:32,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:48:49,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7448, 1.8329, 2.0325, 1.5404], device='cuda:0'), covar=tensor([0.1907, 0.2352, 0.1488, 0.1743], device='cuda:0'), in_proj_covar=tensor([0.0876, 0.0704, 0.0924, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 06:49:03,865 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:49:05,692 INFO [train.py:968] (0/2) Epoch 18, batch 9250, giga_loss[loss=0.3074, simple_loss=0.3713, pruned_loss=0.1218, over 28862.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3677, pruned_loss=0.1202, over 5682158.54 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3469, pruned_loss=0.09222, over 5689803.78 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3699, pruned_loss=0.123, over 5676013.39 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:49:42,372 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-786000.pt
+2023-03-09 06:49:44,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2045, 2.5849, 1.2882, 1.3352], device='cuda:0'), covar=tensor([0.1015, 0.0382, 0.0927, 0.1422], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0546, 0.0369, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0033, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 06:49:58,041 INFO [train.py:968] (0/2) Epoch 18, batch 9300, giga_loss[loss=0.4166, simple_loss=0.4212, pruned_loss=0.206, over 23544.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3704, pruned_loss=0.1219, over 5664034.60 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3473, pruned_loss=0.09254, over 5683076.33 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.372, pruned_loss=0.1241, over 5665341.04 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:50:05,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3298, 3.1609, 2.9864, 1.4295], device='cuda:0'), covar=tensor([0.0927, 0.1037, 0.0918, 0.2228], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.1098, 0.0946, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 06:50:11,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.809e+03 2.370e+03 3.522e+03 1.143e+04, threshold=4.740e+03, percent-clipped=17.0
+2023-03-09 06:50:19,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=786034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:50:48,345 INFO [train.py:968] (0/2) Epoch 18, batch 9350, giga_loss[loss=0.2766, simple_loss=0.357, pruned_loss=0.09813, over 28992.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3732, pruned_loss=0.1235, over 5669178.68 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3475, pruned_loss=0.09265, over 5681728.15 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3746, pruned_loss=0.1254, over 5671155.35 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:51:23,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3876, 1.1984, 1.1325, 1.5528], device='cuda:0'), covar=tensor([0.0738, 0.0363, 0.0328, 0.0789], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 06:51:34,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 06:51:37,116 INFO [train.py:968] (0/2) Epoch 18, batch 9400, giga_loss[loss=0.3652, simple_loss=0.4067, pruned_loss=0.1618, over 28560.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.372, pruned_loss=0.1233, over 5668049.31 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3471, pruned_loss=0.09249, over 5685908.87 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3739, pruned_loss=0.1255, over 5665493.62 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:51:51,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 1.637e+03 2.091e+03 2.793e+03 5.528e+03, threshold=4.182e+03, percent-clipped=1.0
+2023-03-09 06:51:51,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3213, 1.5981, 1.2652, 1.3737], device='cuda:0'), covar=tensor([0.2929, 0.2853, 0.3442, 0.2353], device='cuda:0'), in_proj_covar=tensor([0.1436, 0.1041, 0.1274, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 06:52:14,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1187, 3.9319, 3.7125, 1.7884], device='cuda:0'), covar=tensor([0.0655, 0.0807, 0.0806, 0.2260], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.1095, 0.0943, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 06:52:25,280 INFO [train.py:968] (0/2) Epoch 18, batch 9450, libri_loss[loss=0.2682, simple_loss=0.3493, pruned_loss=0.09352, over 29535.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3733, pruned_loss=0.1213, over 5679144.95 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3471, pruned_loss=0.09255, over 5690250.59 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3752, pruned_loss=0.1236, over 5672790.37 frames. ], batch size: 78, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:52:27,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6363, 1.8165, 1.4458, 1.7656], device='cuda:0'), covar=tensor([0.2780, 0.2783, 0.3242, 0.2408], device='cuda:0'), in_proj_covar=tensor([0.1437, 0.1042, 0.1274, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 06:53:05,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6185, 1.5209, 1.2896, 1.1711], device='cuda:0'), covar=tensor([0.0640, 0.0417, 0.0772, 0.1065], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0441, 0.0505, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:53:07,024 INFO [train.py:968] (0/2) Epoch 18, batch 9500, giga_loss[loss=0.2636, simple_loss=0.3561, pruned_loss=0.08554, over 28496.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3731, pruned_loss=0.1194, over 5683294.14 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.347, pruned_loss=0.09263, over 5697787.91 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.376, pruned_loss=0.1224, over 5671024.42 frames. ], batch size: 60, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:53:20,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.496e+02 1.350e+03 2.040e+03 2.797e+03 6.967e+03, threshold=4.080e+03, percent-clipped=8.0
+2023-03-09 06:53:24,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=786230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:53:59,175 INFO [train.py:968] (0/2) Epoch 18, batch 9550, giga_loss[loss=0.4504, simple_loss=0.4645, pruned_loss=0.2182, over 26643.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.377, pruned_loss=0.1219, over 5676537.29 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3468, pruned_loss=0.09255, over 5698787.97 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3796, pruned_loss=0.1245, over 5666016.80 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:54:25,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6052, 1.7965, 1.4556, 1.7401], device='cuda:0'), covar=tensor([0.2587, 0.2635, 0.2995, 0.2235], device='cuda:0'), in_proj_covar=tensor([0.1438, 0.1041, 0.1274, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 06:54:35,167 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0796, 1.3144, 1.0118, 0.8914], device='cuda:0'), covar=tensor([0.1027, 0.0530, 0.1236, 0.1060], device='cuda:0'), in_proj_covar=tensor([0.0380, 0.0442, 0.0505, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:54:47,870 INFO [train.py:968] (0/2) Epoch 18, batch 9600, giga_loss[loss=0.2983, simple_loss=0.3706, pruned_loss=0.113, over 28982.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.38, pruned_loss=0.1247, over 5675430.33 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3469, pruned_loss=0.09255, over 5696871.98 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3823, pruned_loss=0.1272, over 5668504.27 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:55:00,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.014e+02 1.628e+03 2.372e+03 3.537e+03 6.836e+03, threshold=4.744e+03, percent-clipped=17.0
+2023-03-09 06:55:06,031 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-09 06:55:23,021 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4929, 1.6072, 1.0628, 1.2538], device='cuda:0'), covar=tensor([0.1255, 0.0867, 0.1633, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0444, 0.0507, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 06:55:37,552 INFO [train.py:968] (0/2) Epoch 18, batch 9650, giga_loss[loss=0.3266, simple_loss=0.3882, pruned_loss=0.1325, over 28735.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3811, pruned_loss=0.1267, over 5673870.54 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3466, pruned_loss=0.09236, over 5699072.24 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3837, pruned_loss=0.1294, over 5666101.02 frames. ], batch size: 85, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:55:50,372 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=786373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:55:52,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=786376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:56:13,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1996, 1.4379, 1.3442, 1.2819], device='cuda:0'), covar=tensor([0.0756, 0.0372, 0.0298, 0.0968], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0216], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 06:56:20,945 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=786405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:56:23,917 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=786409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:56:26,845 INFO [train.py:968] (0/2) Epoch 18, batch 9700, giga_loss[loss=0.3883, simple_loss=0.4193, pruned_loss=0.1787, over 23505.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3799, pruned_loss=0.1268, over 5663890.52 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3461, pruned_loss=0.09211, over 5701450.74 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.383, pruned_loss=0.1297, over 5655000.42 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:56:43,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.244e+03 1.794e+03 2.514e+03 3.503e+03 1.385e+04, threshold=5.028e+03, percent-clipped=13.0
+2023-03-09 06:56:53,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2266, 1.5664, 1.5296, 1.3394], device='cuda:0'), covar=tensor([0.1792, 0.1619, 0.2151, 0.1846], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0744, 0.0698, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 06:57:12,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2127, 1.5845, 1.2052, 1.0263], device='cuda:0'), covar=tensor([0.2739, 0.2625, 0.3119, 0.2311], device='cuda:0'), in_proj_covar=tensor([0.1440, 0.1042, 0.1276, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 06:57:12,383 INFO [train.py:968] (0/2) Epoch 18, batch 9750, giga_loss[loss=0.3144, simple_loss=0.3921, pruned_loss=0.1184, over 28958.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3774, pruned_loss=0.1241, over 5666710.34 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3461, pruned_loss=0.09203, over 5705240.25 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3806, pruned_loss=0.1271, over 5655251.57 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:57:56,954 INFO [train.py:968] (0/2) Epoch 18, batch 9800, giga_loss[loss=0.3609, simple_loss=0.4086, pruned_loss=0.1566, over 28540.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3775, pruned_loss=0.1224, over 5669983.11 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3463, pruned_loss=0.09218, over 5708530.69 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3806, pruned_loss=0.1255, over 5656856.37 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:58:11,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.549e+02 1.480e+03 1.999e+03 2.972e+03 1.134e+04, threshold=3.999e+03, percent-clipped=7.0
+2023-03-09 06:58:34,999 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=786552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:58:36,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=786555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:58:41,714 INFO [train.py:968] (0/2) Epoch 18, batch 9850, giga_loss[loss=0.3563, simple_loss=0.4129, pruned_loss=0.1498, over 28657.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3787, pruned_loss=0.1224, over 5678555.04 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3461, pruned_loss=0.0921, over 5712537.45 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.382, pruned_loss=0.1255, over 5663905.16 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:59:08,740 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=786584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:59:34,417 INFO [train.py:968] (0/2) Epoch 18, batch 9900, giga_loss[loss=0.3383, simple_loss=0.4017, pruned_loss=0.1374, over 28322.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3792, pruned_loss=0.1233, over 5669887.72 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3459, pruned_loss=0.09211, over 5715316.42 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3827, pruned_loss=0.1264, over 5654996.66 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:59:35,908 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.47 vs. limit=5.0
+2023-03-09 06:59:51,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.641e+03 2.218e+03 3.084e+03 1.597e+04, threshold=4.435e+03, percent-clipped=7.0
+2023-03-09 07:00:24,841 INFO [train.py:968] (0/2) Epoch 18, batch 9950, libri_loss[loss=0.3494, simple_loss=0.4095, pruned_loss=0.1447, over 19895.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3784, pruned_loss=0.1229, over 5665151.38 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3461, pruned_loss=0.09232, over 5709450.14 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3816, pruned_loss=0.1258, over 5658512.24 frames. ], batch size: 187, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:01:09,055 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=786704.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 07:01:16,113 INFO [train.py:968] (0/2) Epoch 18, batch 10000, giga_loss[loss=0.3204, simple_loss=0.3596, pruned_loss=0.1406, over 23468.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3775, pruned_loss=0.1237, over 5663798.40 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3463, pruned_loss=0.09252, over 5712151.97 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3802, pruned_loss=0.1262, over 5655626.61 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:01:32,280 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.734e+03 2.434e+03 3.267e+03 7.307e+03, threshold=4.868e+03, percent-clipped=11.0
+2023-03-09 07:02:06,320 INFO [train.py:968] (0/2) Epoch 18, batch 10050, giga_loss[loss=0.265, simple_loss=0.3364, pruned_loss=0.09681, over 28701.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3745, pruned_loss=0.1222, over 5669765.82 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3461, pruned_loss=0.09234, over 5717290.96 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3776, pruned_loss=0.125, over 5657658.35 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:02:55,135 INFO [train.py:968] (0/2) Epoch 18, batch 10100, giga_loss[loss=0.2707, simple_loss=0.3462, pruned_loss=0.09765, over 28942.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3731, pruned_loss=0.1215, over 5671204.49 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3467, pruned_loss=0.0928, over 5712785.27 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3758, pruned_loss=0.1242, over 5664252.02 frames. ], batch size: 164, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:03:12,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.419e+02 1.606e+03 2.170e+03 2.870e+03 5.457e+03, threshold=4.339e+03, percent-clipped=2.0
+2023-03-09 07:03:42,961 INFO [train.py:968] (0/2) Epoch 18, batch 10150, giga_loss[loss=0.2924, simple_loss=0.3661, pruned_loss=0.1093, over 29047.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3717, pruned_loss=0.1214, over 5671979.95 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3462, pruned_loss=0.09246, over 5717523.69 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.375, pruned_loss=0.1246, over 5661255.41 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:04:32,125 INFO [train.py:968] (0/2) Epoch 18, batch 10200, giga_loss[loss=0.3035, simple_loss=0.3624, pruned_loss=0.1223, over 28830.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3698, pruned_loss=0.1201, over 5675081.15 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3461, pruned_loss=0.09232, over 5720089.11 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3729, pruned_loss=0.1232, over 5663641.30 frames. ], batch size: 199, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:04:47,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.582e+03 1.997e+03 2.681e+03 7.875e+03, threshold=3.995e+03, percent-clipped=3.0
+2023-03-09 07:05:04,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2122, 1.3884, 1.4059, 1.2248], device='cuda:0'), covar=tensor([0.2818, 0.2076, 0.1752, 0.2225], device='cuda:0'), in_proj_covar=tensor([0.1897, 0.1840, 0.1769, 0.1903], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 07:05:09,782 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1119, 2.8186, 1.3533, 1.2811], device='cuda:0'), covar=tensor([0.1091, 0.0441, 0.0975, 0.1507], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0547, 0.0371, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 07:05:19,544 INFO [train.py:968] (0/2) Epoch 18, batch 10250, giga_loss[loss=0.2651, simple_loss=0.3471, pruned_loss=0.09159, over 28900.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3669, pruned_loss=0.1169, over 5664779.93 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3456, pruned_loss=0.09208, over 5722024.29 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3701, pruned_loss=0.12, over 5653191.36 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:06:05,041 INFO [train.py:968] (0/2) Epoch 18, batch 10300, giga_loss[loss=0.2798, simple_loss=0.3541, pruned_loss=0.1028, over 28540.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3646, pruned_loss=0.1147, over 5671223.74 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3455, pruned_loss=0.09189, over 5729463.13 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3681, pruned_loss=0.1182, over 5653237.51 frames. ], batch size: 71, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:06:05,999 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:06:06,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6807, 1.8584, 1.9698, 1.5065], device='cuda:0'), covar=tensor([0.1909, 0.2388, 0.1492, 0.1735], device='cuda:0'), in_proj_covar=tensor([0.0876, 0.0700, 0.0924, 0.0821], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 07:06:21,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.706e+02 1.446e+03 1.899e+03 2.399e+03 7.012e+03, threshold=3.797e+03, percent-clipped=4.0
+2023-03-09 07:06:54,942 INFO [train.py:968] (0/2) Epoch 18, batch 10350, libri_loss[loss=0.286, simple_loss=0.3683, pruned_loss=0.1019, over 29529.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3641, pruned_loss=0.1142, over 5668416.57 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3461, pruned_loss=0.09225, over 5722783.54 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3667, pruned_loss=0.1171, over 5658574.62 frames. ], batch size: 81, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:07:13,340 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:07:36,694 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3135, 1.2359, 1.1895, 1.4304], device='cuda:0'), covar=tensor([0.0758, 0.0345, 0.0334, 0.0830], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 07:07:45,862 INFO [train.py:968] (0/2) Epoch 18, batch 10400, giga_loss[loss=0.3465, simple_loss=0.3904, pruned_loss=0.1513, over 28435.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3629, pruned_loss=0.1148, over 5665124.92 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3463, pruned_loss=0.09239, over 5724280.69 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3653, pruned_loss=0.1177, over 5654057.26 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:07:59,419 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:08:00,639 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.821e+03 2.357e+03 3.283e+03 7.278e+03, threshold=4.713e+03, percent-clipped=14.0
+2023-03-09 07:08:34,944 INFO [train.py:968] (0/2) Epoch 18, batch 10450, giga_loss[loss=0.3185, simple_loss=0.3972, pruned_loss=0.1199, over 28932.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3618, pruned_loss=0.1146, over 5674681.82 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3459, pruned_loss=0.09226, over 5729043.02 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3644, pruned_loss=0.1175, over 5660456.94 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:09:01,350 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:09:16,394 INFO [train.py:968] (0/2) Epoch 18, batch 10500, giga_loss[loss=0.291, simple_loss=0.3675, pruned_loss=0.1073, over 29066.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3643, pruned_loss=0.1151, over 5684488.54 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3462, pruned_loss=0.09242, over 5735272.57 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3667, pruned_loss=0.1181, over 5665666.47 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:09:25,194 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787222.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:09:28,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787225.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:09:33,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.542e+03 2.194e+03 2.787e+03 6.097e+03, threshold=4.388e+03, percent-clipped=3.0
+2023-03-09 07:09:57,440 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=787254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:10:05,117 INFO [train.py:968] (0/2) Epoch 18, batch 10550, giga_loss[loss=0.3769, simple_loss=0.4118, pruned_loss=0.171, over 26513.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3663, pruned_loss=0.1165, over 5661581.48 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3463, pruned_loss=0.09243, over 5730177.25 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3686, pruned_loss=0.1194, over 5649899.34 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:10:45,954 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 07:10:49,999 INFO [train.py:968] (0/2) Epoch 18, batch 10600, giga_loss[loss=0.2613, simple_loss=0.3412, pruned_loss=0.09069, over 28893.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3657, pruned_loss=0.1162, over 5654500.15 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3458, pruned_loss=0.09223, over 5734911.31 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3687, pruned_loss=0.1195, over 5638543.55 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:10:55,046 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-09 07:11:04,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.991e+02 1.535e+03 1.922e+03 2.514e+03 7.115e+03, threshold=3.844e+03, percent-clipped=7.0
+2023-03-09 07:11:36,573 INFO [train.py:968] (0/2) Epoch 18, batch 10650, giga_loss[loss=0.2766, simple_loss=0.3466, pruned_loss=0.1033, over 28767.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3654, pruned_loss=0.1164, over 5652900.16 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3457, pruned_loss=0.09208, over 5739081.86 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3685, pruned_loss=0.12, over 5633706.44 frames. ], batch size: 99, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:12:00,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:12:26,445 INFO [train.py:968] (0/2) Epoch 18, batch 10700, giga_loss[loss=0.2934, simple_loss=0.3634, pruned_loss=0.1117, over 28975.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.367, pruned_loss=0.1177, over 5648669.01 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.346, pruned_loss=0.09217, over 5741418.97 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3696, pruned_loss=0.121, over 5629620.01 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:12:41,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.129e+03 1.666e+03 2.044e+03 2.742e+03 5.279e+03, threshold=4.089e+03, percent-clipped=8.0
+2023-03-09 07:13:04,865 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 07:13:16,859 INFO [train.py:968] (0/2) Epoch 18, batch 10750, giga_loss[loss=0.4073, simple_loss=0.4242, pruned_loss=0.1952, over 23365.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.369, pruned_loss=0.1188, over 5645351.01 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3462, pruned_loss=0.09212, over 5735021.34 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3713, pruned_loss=0.1218, over 5633935.80 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:13:24,209 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:13:30,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:13:36,433 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:13:54,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:14:04,296 INFO [train.py:968] (0/2) Epoch 18, batch 10800, giga_loss[loss=0.2829, simple_loss=0.3632, pruned_loss=0.1013, over 28887.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3712, pruned_loss=0.1205, over 5633868.89 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3465, pruned_loss=0.09232, over 5719310.24 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3732, pruned_loss=0.1232, over 5638151.36 frames. ], batch size: 164, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:14:23,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.786e+03 2.606e+03 4.204e+03 1.384e+04, threshold=5.211e+03, percent-clipped=26.0
+2023-03-09 07:14:23,934 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:14:27,634 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:14:30,284 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:14:45,941 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 07:14:52,702 INFO [train.py:968] (0/2) Epoch 18, batch 10850, giga_loss[loss=0.326, simple_loss=0.3815, pruned_loss=0.1352, over 29069.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3727, pruned_loss=0.122, over 5646736.55 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3462, pruned_loss=0.0922, over 5722353.59 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3751, pruned_loss=0.1249, over 5645731.47 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:14:54,139 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=787563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:15:01,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:15:43,176 INFO [train.py:968] (0/2) Epoch 18, batch 10900, giga_loss[loss=0.322, simple_loss=0.4006, pruned_loss=0.1216, over 28478.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3737, pruned_loss=0.1221, over 5650211.04 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3459, pruned_loss=0.09201, over 5724718.62 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3764, pruned_loss=0.1253, over 5645674.60 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:16:06,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.210e+03 1.788e+03 2.361e+03 3.406e+03 1.227e+04, threshold=4.723e+03, percent-clipped=9.0
+2023-03-09 07:16:20,264 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:16:23,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:16:38,347 INFO [train.py:968] (0/2) Epoch 18, batch 10950, giga_loss[loss=0.3405, simple_loss=0.397, pruned_loss=0.142, over 28940.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3751, pruned_loss=0.1228, over 5649711.74 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3463, pruned_loss=0.09234, over 5724643.85 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3775, pruned_loss=0.1256, over 5644869.58 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:16:52,348 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=787676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:17:08,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3528, 1.2756, 4.0530, 3.3327], device='cuda:0'), covar=tensor([0.1684, 0.2790, 0.0463, 0.0866], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0632, 0.0930, 0.0868], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 07:17:13,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1601, 3.4056, 2.2859, 1.2019], device='cuda:0'), covar=tensor([0.6921, 0.2187, 0.3443, 0.6421], device='cuda:0'), in_proj_covar=tensor([0.1687, 0.1606, 0.1574, 0.1382], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 07:17:28,114 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:17:28,417 INFO [train.py:968] (0/2) Epoch 18, batch 11000, giga_loss[loss=0.2953, simple_loss=0.3621, pruned_loss=0.1143, over 28679.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3726, pruned_loss=0.1213, over 5655690.42 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3462, pruned_loss=0.09239, over 5728840.72 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3754, pruned_loss=0.1244, over 5645728.90 frames. ], batch size: 242, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:17:30,215 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:17:44,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.355e+02 1.720e+03 2.173e+03 3.133e+03 1.293e+04, threshold=4.345e+03, percent-clipped=6.0
+2023-03-09 07:17:58,578 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=787743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:18:23,087 INFO [train.py:968] (0/2) Epoch 18, batch 11050, giga_loss[loss=0.385, simple_loss=0.4077, pruned_loss=0.1812, over 23529.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3715, pruned_loss=0.1215, over 5655805.80 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3462, pruned_loss=0.09243, over 5729925.10 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3739, pruned_loss=0.1242, over 5646513.27 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:19:19,463 INFO [train.py:968] (0/2) Epoch 18, batch 11100, giga_loss[loss=0.2739, simple_loss=0.3524, pruned_loss=0.09773, over 29007.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3711, pruned_loss=0.1214, over 5660193.38 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.346, pruned_loss=0.09233, over 5731554.24 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3735, pruned_loss=0.1239, over 5650739.52 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:19:31,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7372, 4.5563, 4.2879, 2.4678], device='cuda:0'), covar=tensor([0.0661, 0.0850, 0.0999, 0.1815], device='cuda:0'), in_proj_covar=tensor([0.1181, 0.1094, 0.0944, 0.0711], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 07:19:35,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.113e+03 1.590e+03 2.193e+03 3.007e+03 6.254e+03, threshold=4.386e+03, percent-clipped=6.0
+2023-03-09 07:19:47,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:19:51,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:20:00,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:20:04,632 INFO [train.py:968] (0/2) Epoch 18, batch 11150, giga_loss[loss=0.3014, simple_loss=0.3601, pruned_loss=0.1213, over 28569.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3703, pruned_loss=0.1215, over 5659826.96 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3463, pruned_loss=0.09256, over 5725546.22 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3728, pruned_loss=0.1243, over 5655511.02 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:20:46,247 INFO [train.py:968] (0/2) Epoch 18, batch 11200, giga_loss[loss=0.2816, simple_loss=0.3486, pruned_loss=0.1073, over 28482.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3692, pruned_loss=0.1206, over 5666478.65 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3463, pruned_loss=0.09241, over 5730353.15 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3719, pruned_loss=0.1238, over 5656754.83 frames. ], batch size: 85, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:20:47,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:21:00,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9653, 1.2730, 1.0069, 0.1882], device='cuda:0'), covar=tensor([0.2927, 0.2263, 0.3241, 0.5272], device='cuda:0'), in_proj_covar=tensor([0.1691, 0.1609, 0.1577, 0.1386], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 07:21:03,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.036e+02 1.598e+03 1.984e+03 2.866e+03 6.110e+03, threshold=3.967e+03, percent-clipped=6.0
+2023-03-09 07:21:28,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0162, 1.3492, 1.0629, 0.1515], device='cuda:0'), covar=tensor([0.3631, 0.2892, 0.4539, 0.6005], device='cuda:0'), in_proj_covar=tensor([0.1692, 0.1611, 0.1579, 0.1387], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 07:21:32,109 INFO [train.py:968] (0/2) Epoch 18, batch 11250, giga_loss[loss=0.318, simple_loss=0.3555, pruned_loss=0.1403, over 23590.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3694, pruned_loss=0.121, over 5659300.91 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3463, pruned_loss=0.0925, over 5727302.93 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3723, pruned_loss=0.1244, over 5653186.64 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:21:56,366 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:21:59,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:02,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:04,563 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:07,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:08,591 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-788000.pt
+2023-03-09 07:22:10,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=788002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:20,444 INFO [train.py:968] (0/2) Epoch 18, batch 11300, giga_loss[loss=0.3226, simple_loss=0.3786, pruned_loss=0.1333, over 27917.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3707, pruned_loss=0.122, over 5662537.26 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3466, pruned_loss=0.09262, over 5724547.64 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.373, pruned_loss=0.1249, over 5659114.55 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:22:25,356 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=788018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:30,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8098, 1.9362, 2.0447, 1.8485], device='cuda:0'), covar=tensor([0.2391, 0.2069, 0.1533, 0.1740], device='cuda:0'), in_proj_covar=tensor([0.1898, 0.1850, 0.1771, 0.1912], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 07:22:31,497 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=788024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:36,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.843e+03 2.338e+03 3.043e+03 8.086e+03, threshold=4.677e+03, percent-clipped=13.0
+2023-03-09 07:22:36,597 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=788031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:58,320 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=788056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:23:03,474 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=788059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:23:05,207 INFO [train.py:968] (0/2) Epoch 18, batch 11350, libri_loss[loss=0.2596, simple_loss=0.3526, pruned_loss=0.08324, over 29245.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3719, pruned_loss=0.1226, over 5669861.01 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3465, pruned_loss=0.09244, over 5729929.30 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3745, pruned_loss=0.126, over 5660687.46 frames. ], batch size: 97, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:23:24,798 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-09 07:23:30,214 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=788088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:23:39,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6643, 1.7019, 1.8881, 1.4356], device='cuda:0'), covar=tensor([0.1768, 0.2527, 0.1394, 0.1662], device='cuda:0'), in_proj_covar=tensor([0.0874, 0.0701, 0.0923, 0.0820], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 07:23:56,827 INFO [train.py:968] (0/2) Epoch 18, batch 11400, giga_loss[loss=0.4308, simple_loss=0.4456, pruned_loss=0.2079, over 27648.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3732, pruned_loss=0.1244, over 5659076.01 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3464, pruned_loss=0.0924, over 5730850.65 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3754, pruned_loss=0.1272, over 5650897.92 frames. ], batch size: 472, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:24:13,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.754e+02 1.739e+03 2.565e+03 3.332e+03 1.034e+04, threshold=5.129e+03, percent-clipped=10.0
+2023-03-09 07:24:45,152 INFO [train.py:968] (0/2) Epoch 18, batch 11450, giga_loss[loss=0.3266, simple_loss=0.3782, pruned_loss=0.1375, over 28690.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3738, pruned_loss=0.1256, over 5658244.90 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3469, pruned_loss=0.09266, over 5733243.66 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3755, pruned_loss=0.128, over 5648768.35 frames. ], batch size: 92, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:24:52,447 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3664, 1.5621, 1.3920, 1.5429], device='cuda:0'), covar=tensor([0.0784, 0.0334, 0.0334, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0067, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 07:25:29,834 INFO [train.py:968] (0/2) Epoch 18, batch 11500, giga_loss[loss=0.3254, simple_loss=0.3834, pruned_loss=0.1337, over 29036.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3727, pruned_loss=0.1245, over 5668467.85 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3466, pruned_loss=0.09252, over 5737761.06 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.375, pruned_loss=0.1274, over 5654966.20 frames. ], batch size: 136, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:25:48,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.160e+03 1.677e+03 2.082e+03 2.865e+03 9.791e+03, threshold=4.165e+03, percent-clipped=9.0
+2023-03-09 07:25:52,808 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=788235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:26:17,218 INFO [train.py:968] (0/2) Epoch 18, batch 11550, giga_loss[loss=0.326, simple_loss=0.3938, pruned_loss=0.1291, over 28871.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3724, pruned_loss=0.1235, over 5666137.14 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3468, pruned_loss=0.09281, over 5731743.87 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3744, pruned_loss=0.1261, over 5658939.58 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:27:00,793 INFO [train.py:968] (0/2) Epoch 18, batch 11600, giga_loss[loss=0.3024, simple_loss=0.3663, pruned_loss=0.1193, over 28888.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3735, pruned_loss=0.124, over 5649607.77 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3477, pruned_loss=0.09347, over 5712418.46 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3752, pruned_loss=0.1264, over 5658404.85 frames. ], batch size: 112, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:27:18,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.955e+02 1.703e+03 2.294e+03 2.878e+03 6.289e+03, threshold=4.587e+03, percent-clipped=8.0
+2023-03-09 07:27:48,410 INFO [train.py:968] (0/2) Epoch 18, batch 11650, giga_loss[loss=0.2997, simple_loss=0.375, pruned_loss=0.1122, over 28904.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3763, pruned_loss=0.1258, over 5652551.63 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3482, pruned_loss=0.09376, over 5699647.66 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3776, pruned_loss=0.128, over 5669723.64 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:28:36,732 INFO [train.py:968] (0/2) Epoch 18, batch 11700, giga_loss[loss=0.2918, simple_loss=0.3601, pruned_loss=0.1117, over 28968.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3769, pruned_loss=0.1269, over 5660932.87 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.348, pruned_loss=0.09367, over 5703261.37 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3787, pruned_loss=0.1294, over 5670334.93 frames. ], batch size: 227, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:28:53,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.679e+03 2.079e+03 3.270e+03 7.220e+03, threshold=4.158e+03, percent-clipped=8.0
+2023-03-09 07:29:20,043 INFO [train.py:968] (0/2) Epoch 18, batch 11750, libri_loss[loss=0.2493, simple_loss=0.3373, pruned_loss=0.08063, over 29768.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3755, pruned_loss=0.1252, over 5677089.59 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3481, pruned_loss=0.09375, over 5711270.00 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3781, pruned_loss=0.1284, over 5676064.32 frames. ], batch size: 87, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:29:24,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=788467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:30:07,547 INFO [train.py:968] (0/2) Epoch 18, batch 11800, giga_loss[loss=0.309, simple_loss=0.3784, pruned_loss=0.1198, over 28322.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3757, pruned_loss=0.1243, over 5668431.28 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3479, pruned_loss=0.09366, over 5713218.51 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3781, pruned_loss=0.1272, over 5665714.71 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:30:24,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.254e+02 1.531e+03 1.925e+03 2.359e+03 6.892e+03, threshold=3.850e+03, percent-clipped=7.0
+2023-03-09 07:30:27,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5107, 1.7682, 1.2485, 1.2742], device='cuda:0'), covar=tensor([0.0952, 0.0541, 0.1037, 0.1124], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0447, 0.0511, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 07:30:51,910 INFO [train.py:968] (0/2) Epoch 18, batch 11850, giga_loss[loss=0.2597, simple_loss=0.337, pruned_loss=0.09125, over 28794.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3731, pruned_loss=0.1221, over 5650743.41 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.347, pruned_loss=0.09321, over 5698271.51 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3764, pruned_loss=0.1255, over 5660952.22 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:31:00,925 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-09 07:31:05,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7433, 1.8954, 1.9210, 1.5688], device='cuda:0'), covar=tensor([0.1711, 0.1857, 0.1890, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0750, 0.0704, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 07:31:11,183 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.30 vs. limit=5.0
+2023-03-09 07:31:34,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=788610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:31:35,592 INFO [train.py:968] (0/2) Epoch 18, batch 11900, giga_loss[loss=0.3095, simple_loss=0.373, pruned_loss=0.123, over 29019.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3703, pruned_loss=0.1198, over 5672065.60 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3465, pruned_loss=0.0931, over 5705938.45 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3743, pruned_loss=0.1236, over 5671987.67 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:31:50,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.893e+02 1.684e+03 2.034e+03 2.736e+03 5.525e+03, threshold=4.069e+03, percent-clipped=9.0
+2023-03-09 07:32:21,344 INFO [train.py:968] (0/2) Epoch 18, batch 11950, giga_loss[loss=0.3038, simple_loss=0.3765, pruned_loss=0.1156, over 28607.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3705, pruned_loss=0.1202, over 5667963.65 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3467, pruned_loss=0.09322, over 5708123.74 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.374, pruned_loss=0.1237, over 5665491.48 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:33:08,639 INFO [train.py:968] (0/2) Epoch 18, batch 12000, giga_loss[loss=0.321, simple_loss=0.3938, pruned_loss=0.124, over 28859.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3727, pruned_loss=0.1219, over 5664202.59 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3466, pruned_loss=0.0931, over 5705723.64 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3762, pruned_loss=0.1254, over 5663209.48 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:33:08,644 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 07:33:17,636 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2108, simple_loss=0.3183, pruned_loss=0.05168, over 944034.00 frames. 
+2023-03-09 07:33:17,636 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 07:33:30,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2644, 1.2768, 1.3162, 1.4512], device='cuda:0'), covar=tensor([0.0749, 0.0356, 0.0321, 0.0820], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 07:33:34,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.851e+02 1.470e+03 1.846e+03 2.427e+03 4.872e+03, threshold=3.692e+03, percent-clipped=1.0
+2023-03-09 07:33:41,448 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4570, 1.7452, 1.4091, 1.3882], device='cuda:0'), covar=tensor([0.2568, 0.2566, 0.2877, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.1444, 0.1053, 0.1282, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 07:33:56,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=788753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:33:58,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=788756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:34:02,053 INFO [train.py:968] (0/2) Epoch 18, batch 12050, giga_loss[loss=0.3247, simple_loss=0.3859, pruned_loss=0.1318, over 28550.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3745, pruned_loss=0.1235, over 5668141.60 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3467, pruned_loss=0.09308, over 5709276.24 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3779, pruned_loss=0.1271, over 5663194.58 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:34:25,072 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=788785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:34:49,136 INFO [train.py:968] (0/2) Epoch 18, batch 12100, giga_loss[loss=0.2849, simple_loss=0.3554, pruned_loss=0.1072, over 28545.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3745, pruned_loss=0.1243, over 5664998.23 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3466, pruned_loss=0.09286, over 5707757.00 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3779, pruned_loss=0.1279, over 5661426.39 frames. ], batch size: 71, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:35:10,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.611e+03 1.993e+03 2.722e+03 8.613e+03, threshold=3.987e+03, percent-clipped=15.0
+2023-03-09 07:35:18,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=788842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:35:31,969 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 07:35:38,116 INFO [train.py:968] (0/2) Epoch 18, batch 12150, giga_loss[loss=0.3808, simple_loss=0.4191, pruned_loss=0.1713, over 27601.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3752, pruned_loss=0.1253, over 5663042.30 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3464, pruned_loss=0.0927, over 5711315.02 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3786, pruned_loss=0.129, over 5656151.11 frames. ], batch size: 472, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:36:25,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2394, 1.6861, 1.3631, 1.4707], device='cuda:0'), covar=tensor([0.0784, 0.0308, 0.0333, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0093, 0.0067, 0.0059, 0.0102], device='cuda:0')
+2023-03-09 07:36:25,840 INFO [train.py:968] (0/2) Epoch 18, batch 12200, libri_loss[loss=0.2829, simple_loss=0.3698, pruned_loss=0.09802, over 29543.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3769, pruned_loss=0.1266, over 5652951.68 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3469, pruned_loss=0.0929, over 5705161.05 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3796, pruned_loss=0.1298, over 5652718.60 frames. ], batch size: 83, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:36:42,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.320e+02 1.538e+03 2.024e+03 2.691e+03 7.890e+03, threshold=4.049e+03, percent-clipped=3.0
+2023-03-09 07:36:49,190 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 07:37:09,076 INFO [train.py:968] (0/2) Epoch 18, batch 12250, giga_loss[loss=0.348, simple_loss=0.3977, pruned_loss=0.1491, over 28627.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3766, pruned_loss=0.1263, over 5655453.55 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3472, pruned_loss=0.09308, over 5709314.38 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3796, pruned_loss=0.1299, over 5650064.42 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:37:28,017 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-09 07:37:30,039 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=788985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:37:33,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=788988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:37:57,031 INFO [train.py:968] (0/2) Epoch 18, batch 12300, giga_loss[loss=0.2871, simple_loss=0.3614, pruned_loss=0.1064, over 28504.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3762, pruned_loss=0.1261, over 5641010.51 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3473, pruned_loss=0.09307, over 5711982.73 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3792, pruned_loss=0.1299, over 5632619.54 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:38:02,097 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=789017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:38:16,212 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.680e+02 1.482e+03 1.981e+03 2.805e+03 9.735e+03, threshold=3.963e+03, percent-clipped=10.0
+2023-03-09 07:38:44,548 INFO [train.py:968] (0/2) Epoch 18, batch 12350, libri_loss[loss=0.3007, simple_loss=0.377, pruned_loss=0.1122, over 27719.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.377, pruned_loss=0.1263, over 5644795.60 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3475, pruned_loss=0.09313, over 5714656.93 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3799, pruned_loss=0.1301, over 5634127.09 frames. ], batch size: 116, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:39:23,353 INFO [train.py:968] (0/2) Epoch 18, batch 12400, giga_loss[loss=0.2809, simple_loss=0.351, pruned_loss=0.1054, over 28964.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3759, pruned_loss=0.1251, over 5654165.23 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.347, pruned_loss=0.09298, over 5719739.86 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3799, pruned_loss=0.1296, over 5638532.98 frames. ], batch size: 164, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:39:45,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.696e+02 1.448e+03 1.912e+03 2.509e+03 5.653e+03, threshold=3.824e+03, percent-clipped=5.0
+2023-03-09 07:39:53,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5210, 1.6591, 1.6322, 1.4833], device='cuda:0'), covar=tensor([0.1659, 0.1965, 0.2135, 0.1967], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0751, 0.0707, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 07:40:16,879 INFO [train.py:968] (0/2) Epoch 18, batch 12450, giga_loss[loss=0.3171, simple_loss=0.3799, pruned_loss=0.1272, over 28813.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3749, pruned_loss=0.1246, over 5659442.53 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.347, pruned_loss=0.09298, over 5719739.86 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.378, pruned_loss=0.1281, over 5647275.75 frames. ], batch size: 199, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:40:21,799 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=789168.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:40:31,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=789179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:40:31,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4873, 3.5544, 1.5450, 1.5730], device='cuda:0'), covar=tensor([0.0931, 0.0326, 0.0883, 0.1332], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0549, 0.0371, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 07:40:51,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9095, 3.0551, 1.9029, 1.1098], device='cuda:0'), covar=tensor([0.7168, 0.2882, 0.3951, 0.6382], device='cuda:0'), in_proj_covar=tensor([0.1689, 0.1605, 0.1575, 0.1383], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 07:41:04,254 INFO [train.py:968] (0/2) Epoch 18, batch 12500, giga_loss[loss=0.3134, simple_loss=0.3772, pruned_loss=0.1248, over 29008.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3729, pruned_loss=0.1233, over 5654729.14 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3473, pruned_loss=0.09315, over 5709327.23 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3754, pruned_loss=0.1263, over 5653626.17 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:41:22,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.063e+02 1.789e+03 2.366e+03 3.457e+03 1.201e+04, threshold=4.732e+03, percent-clipped=21.0
+2023-03-09 07:41:39,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7198, 2.0850, 2.0260, 1.6007], device='cuda:0'), covar=tensor([0.2707, 0.2271, 0.2396, 0.2706], device='cuda:0'), in_proj_covar=tensor([0.1906, 0.1859, 0.1786, 0.1928], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 07:41:47,984 INFO [train.py:968] (0/2) Epoch 18, batch 12550, giga_loss[loss=0.3242, simple_loss=0.3757, pruned_loss=0.1364, over 28308.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3697, pruned_loss=0.1214, over 5660948.74 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3473, pruned_loss=0.09315, over 5703600.55 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3722, pruned_loss=0.1245, over 5664359.91 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:42:06,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5495, 1.7713, 1.4726, 1.4838], device='cuda:0'), covar=tensor([0.2596, 0.2565, 0.2947, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.1451, 0.1055, 0.1285, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 07:42:15,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6233, 1.7647, 1.5169, 1.5154], device='cuda:0'), covar=tensor([0.2144, 0.2044, 0.2090, 0.1861], device='cuda:0'), in_proj_covar=tensor([0.1452, 0.1055, 0.1286, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 07:42:37,743 INFO [train.py:968] (0/2) Epoch 18, batch 12600, giga_loss[loss=0.2542, simple_loss=0.3267, pruned_loss=0.09088, over 28858.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3671, pruned_loss=0.1208, over 5634605.76 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3475, pruned_loss=0.09313, over 5698204.35 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3695, pruned_loss=0.124, over 5641245.23 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:42:52,755 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-09 07:42:54,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.604e+03 2.245e+03 3.022e+03 1.082e+04, threshold=4.490e+03, percent-clipped=7.0
+2023-03-09 07:43:13,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4396, 1.9761, 1.4735, 0.8067], device='cuda:0'), covar=tensor([0.5064, 0.2632, 0.3221, 0.5618], device='cuda:0'), in_proj_covar=tensor([0.1691, 0.1607, 0.1576, 0.1386], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 07:43:16,363 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=789355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:43:20,895 INFO [train.py:968] (0/2) Epoch 18, batch 12650, giga_loss[loss=0.2589, simple_loss=0.3316, pruned_loss=0.09312, over 28996.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3659, pruned_loss=0.1205, over 5649244.59 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3473, pruned_loss=0.09297, over 5703108.49 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3685, pruned_loss=0.124, over 5648599.68 frames. ], batch size: 164, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:43:38,941 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 07:44:06,328 INFO [train.py:968] (0/2) Epoch 18, batch 12700, giga_loss[loss=0.3384, simple_loss=0.3887, pruned_loss=0.1441, over 27536.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3645, pruned_loss=0.1194, over 5651898.09 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3472, pruned_loss=0.09275, over 5708731.41 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3674, pruned_loss=0.1234, over 5644294.05 frames. ], batch size: 472, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:44:11,268 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 07:44:28,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.710e+02 1.662e+03 2.258e+03 3.420e+03 8.934e+03, threshold=4.517e+03, percent-clipped=9.0
+2023-03-09 07:44:54,914 INFO [train.py:968] (0/2) Epoch 18, batch 12750, giga_loss[loss=0.2814, simple_loss=0.3531, pruned_loss=0.1048, over 29076.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3631, pruned_loss=0.1168, over 5642807.98 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.347, pruned_loss=0.09274, over 5699508.17 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3658, pruned_loss=0.1204, over 5644110.37 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:45:46,870 INFO [train.py:968] (0/2) Epoch 18, batch 12800, giga_loss[loss=0.3103, simple_loss=0.3753, pruned_loss=0.1227, over 27951.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3617, pruned_loss=0.1139, over 5641318.99 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3468, pruned_loss=0.09266, over 5701623.99 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3642, pruned_loss=0.1171, over 5640120.99 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:45:53,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=789518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:45:56,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2845, 2.7153, 1.4238, 1.4321], device='cuda:0'), covar=tensor([0.0911, 0.0372, 0.0908, 0.1323], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0550, 0.0371, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 07:45:57,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7064, 1.8700, 1.5248, 1.7460], device='cuda:0'), covar=tensor([0.2735, 0.2623, 0.3149, 0.2260], device='cuda:0'), in_proj_covar=tensor([0.1447, 0.1052, 0.1287, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 07:46:06,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.451e+02 1.514e+03 1.883e+03 2.688e+03 6.851e+03, threshold=3.765e+03, percent-clipped=5.0
+2023-03-09 07:46:12,340 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-09 07:46:17,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=789543.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:46:18,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2042, 1.2418, 1.1989, 1.5474], device='cuda:0'), covar=tensor([0.0784, 0.0371, 0.0361, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 07:46:28,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=789554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:46:36,696 INFO [train.py:968] (0/2) Epoch 18, batch 12850, giga_loss[loss=0.2913, simple_loss=0.3672, pruned_loss=0.1077, over 28921.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3591, pruned_loss=0.1108, over 5640950.55 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3468, pruned_loss=0.09277, over 5702336.82 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3612, pruned_loss=0.1134, over 5638613.25 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:47:28,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4085, 1.3985, 4.1734, 3.3219], device='cuda:0'), covar=tensor([0.1591, 0.2641, 0.0414, 0.0866], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0633, 0.0934, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 07:47:29,270 INFO [train.py:968] (0/2) Epoch 18, batch 12900, giga_loss[loss=0.3287, simple_loss=0.3849, pruned_loss=0.1363, over 28802.00 frames. ], tot_loss[loss=0.286, simple_loss=0.356, pruned_loss=0.1079, over 5648501.13 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3465, pruned_loss=0.09271, over 5705214.35 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3581, pruned_loss=0.1103, over 5643443.35 frames. ], batch size: 199, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:47:37,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4608, 1.4951, 3.2159, 3.1634], device='cuda:0'), covar=tensor([0.1224, 0.2450, 0.0446, 0.0958], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0634, 0.0934, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 07:47:53,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.492e+02 1.356e+03 1.826e+03 2.476e+03 4.647e+03, threshold=3.651e+03, percent-clipped=5.0
+2023-03-09 07:48:17,283 INFO [train.py:968] (0/2) Epoch 18, batch 12950, giga_loss[loss=0.2777, simple_loss=0.3644, pruned_loss=0.09549, over 28932.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3528, pruned_loss=0.1044, over 5641809.85 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.346, pruned_loss=0.09265, over 5702220.29 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3552, pruned_loss=0.1068, over 5638478.49 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:48:39,917 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=789686.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 07:48:42,190 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=789689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:48:42,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=789689.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 07:48:51,896 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=789697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:48:54,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=789700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:49:06,226 INFO [train.py:968] (0/2) Epoch 18, batch 13000, giga_loss[loss=0.3124, simple_loss=0.378, pruned_loss=0.1234, over 28530.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3537, pruned_loss=0.1027, over 5658459.12 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3461, pruned_loss=0.09279, over 5703428.82 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3556, pruned_loss=0.1046, over 5654009.08 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:49:12,264 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=789718.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:49:17,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2547, 1.5574, 1.2082, 1.0638], device='cuda:0'), covar=tensor([0.2663, 0.2490, 0.2919, 0.2097], device='cuda:0'), in_proj_covar=tensor([0.1449, 0.1050, 0.1288, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 07:49:25,546 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=789729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:49:26,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=789730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:49:30,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.706e+02 1.439e+03 1.851e+03 2.848e+03 9.724e+03, threshold=3.702e+03, percent-clipped=9.0
+2023-03-09 07:49:55,604 INFO [train.py:968] (0/2) Epoch 18, batch 13050, giga_loss[loss=0.2878, simple_loss=0.3589, pruned_loss=0.1083, over 28580.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3539, pruned_loss=0.1032, over 5651026.27 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3461, pruned_loss=0.09296, over 5706598.60 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3557, pruned_loss=0.1048, over 5643256.57 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:50:17,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6414, 4.4543, 4.2149, 2.0344], device='cuda:0'), covar=tensor([0.0676, 0.0909, 0.1082, 0.2194], device='cuda:0'), in_proj_covar=tensor([0.1173, 0.1085, 0.0932, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 07:50:25,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4153, 1.5646, 1.1785, 1.1157], device='cuda:0'), covar=tensor([0.0892, 0.0453, 0.0984, 0.1093], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0442, 0.0507, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 07:50:43,596 INFO [train.py:968] (0/2) Epoch 18, batch 13100, giga_loss[loss=0.2628, simple_loss=0.3392, pruned_loss=0.0932, over 28617.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3516, pruned_loss=0.1015, over 5655772.35 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3457, pruned_loss=0.09274, over 5709728.87 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3535, pruned_loss=0.1032, over 5645860.87 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:51:07,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.637e+02 1.259e+03 1.610e+03 2.255e+03 7.533e+03, threshold=3.221e+03, percent-clipped=7.0
+2023-03-09 07:51:31,705 INFO [train.py:968] (0/2) Epoch 18, batch 13150, giga_loss[loss=0.2828, simple_loss=0.3466, pruned_loss=0.1095, over 27847.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3484, pruned_loss=0.09991, over 5646462.46 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3451, pruned_loss=0.09243, over 5715472.14 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3506, pruned_loss=0.1018, over 5631592.39 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:51:43,315 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=789873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:51:46,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=789876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:52:04,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=789893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:52:13,631 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=789905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:52:20,277 INFO [train.py:968] (0/2) Epoch 18, batch 13200, giga_loss[loss=0.2631, simple_loss=0.3415, pruned_loss=0.09241, over 27893.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3489, pruned_loss=0.1001, over 5647271.59 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3453, pruned_loss=0.09265, over 5717315.73 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3504, pruned_loss=0.1015, over 5632846.12 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:52:29,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5560, 1.8251, 1.2710, 1.3759], device='cuda:0'), covar=tensor([0.0982, 0.0548, 0.1077, 0.1102], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0440, 0.0506, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 07:52:37,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.716e+02 1.402e+03 1.799e+03 2.355e+03 5.369e+03, threshold=3.599e+03, percent-clipped=13.0
+2023-03-09 07:53:00,851 INFO [train.py:968] (0/2) Epoch 18, batch 13250, giga_loss[loss=0.2869, simple_loss=0.3563, pruned_loss=0.1087, over 27589.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.346, pruned_loss=0.09826, over 5654015.20 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3433, pruned_loss=0.09182, over 5723167.90 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3492, pruned_loss=0.1006, over 5632767.66 frames. ], batch size: 472, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:53:34,495 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-790000.pt
+2023-03-09 07:53:46,699 INFO [train.py:968] (0/2) Epoch 18, batch 13300, libri_loss[loss=0.2121, simple_loss=0.2899, pruned_loss=0.06718, over 29477.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3442, pruned_loss=0.09649, over 5663423.01 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3428, pruned_loss=0.09155, over 5728522.06 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3473, pruned_loss=0.09875, over 5639564.26 frames. ], batch size: 70, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:54:12,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.703e+02 1.363e+03 2.008e+03 3.182e+03 5.957e+03, threshold=4.016e+03, percent-clipped=13.0
+2023-03-09 07:54:13,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=790036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:54:14,996 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 07:54:16,248 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=790039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:54:19,995 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9850, 2.8240, 2.6639, 1.5683], device='cuda:0'), covar=tensor([0.1116, 0.1199, 0.1108, 0.2093], device='cuda:0'), in_proj_covar=tensor([0.1163, 0.1074, 0.0922, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 07:54:25,913 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-09 07:54:38,884 INFO [train.py:968] (0/2) Epoch 18, batch 13350, giga_loss[loss=0.2485, simple_loss=0.3286, pruned_loss=0.08418, over 28925.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3423, pruned_loss=0.09502, over 5650205.11 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.343, pruned_loss=0.09177, over 5720569.92 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3446, pruned_loss=0.09665, over 5637979.33 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:54:42,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=790064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:54:46,064 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=790068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:55:07,728 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=790088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:55:32,252 INFO [train.py:968] (0/2) Epoch 18, batch 13400, giga_loss[loss=0.2197, simple_loss=0.3028, pruned_loss=0.06828, over 27899.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3383, pruned_loss=0.09283, over 5656839.38 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3429, pruned_loss=0.09177, over 5723545.79 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3402, pruned_loss=0.09412, over 5643322.38 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:55:56,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.669e+02 1.456e+03 1.739e+03 2.540e+03 4.648e+03, threshold=3.479e+03, percent-clipped=4.0
+2023-03-09 07:55:58,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2302, 1.1527, 3.6029, 3.1017], device='cuda:0'), covar=tensor([0.1643, 0.2842, 0.0482, 0.1038], device='cuda:0'), in_proj_covar=tensor([0.0730, 0.0628, 0.0924, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 07:56:22,247 INFO [train.py:968] (0/2) Epoch 18, batch 13450, libri_loss[loss=0.3097, simple_loss=0.3762, pruned_loss=0.1215, over 27800.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3382, pruned_loss=0.09318, over 5655712.43 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3431, pruned_loss=0.09209, over 5719227.97 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3393, pruned_loss=0.09391, over 5647166.64 frames. ], batch size: 116, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:57:03,972 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=790207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:57:08,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=790210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:57:10,245 INFO [train.py:968] (0/2) Epoch 18, batch 13500, giga_loss[loss=0.2613, simple_loss=0.3308, pruned_loss=0.09591, over 26684.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3374, pruned_loss=0.09308, over 5644555.21 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3425, pruned_loss=0.09196, over 5721280.04 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3386, pruned_loss=0.09381, over 5634348.66 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:57:41,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.524e+02 1.471e+03 1.893e+03 2.542e+03 9.374e+03, threshold=3.786e+03, percent-clipped=9.0
+2023-03-09 07:57:43,683 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=790239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:58:03,825 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=790259.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 07:58:08,528 INFO [train.py:968] (0/2) Epoch 18, batch 13550, giga_loss[loss=0.2627, simple_loss=0.3496, pruned_loss=0.08788, over 28833.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3395, pruned_loss=0.09389, over 5653907.39 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3419, pruned_loss=0.09174, over 5723930.71 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.341, pruned_loss=0.09472, over 5642061.61 frames. ], batch size: 227, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:58:55,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-09 07:59:02,856 INFO [train.py:968] (0/2) Epoch 18, batch 13600, giga_loss[loss=0.2751, simple_loss=0.3616, pruned_loss=0.09429, over 28498.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3406, pruned_loss=0.09301, over 5657319.11 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3412, pruned_loss=0.09148, over 5726763.30 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3425, pruned_loss=0.09398, over 5643004.69 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:59:29,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.829e+02 1.364e+03 1.742e+03 2.225e+03 5.648e+03, threshold=3.484e+03, percent-clipped=5.0
+2023-03-09 07:59:42,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5310, 1.6288, 1.2375, 1.1858], device='cuda:0'), covar=tensor([0.0957, 0.0584, 0.1020, 0.1191], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0441, 0.0506, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 07:59:50,071 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 07:59:57,901 INFO [train.py:968] (0/2) Epoch 18, batch 13650, giga_loss[loss=0.2582, simple_loss=0.3371, pruned_loss=0.08965, over 28896.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3422, pruned_loss=0.09416, over 5670325.67 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.341, pruned_loss=0.09149, over 5732176.73 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3439, pruned_loss=0.09498, over 5651928.65 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:00:58,309 INFO [train.py:968] (0/2) Epoch 18, batch 13700, giga_loss[loss=0.2319, simple_loss=0.2959, pruned_loss=0.08397, over 24417.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.339, pruned_loss=0.09219, over 5672412.81 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3407, pruned_loss=0.09142, over 5736514.95 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3407, pruned_loss=0.09295, over 5652697.92 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:01:30,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.257e+02 1.271e+03 1.814e+03 2.332e+03 5.952e+03, threshold=3.628e+03, percent-clipped=2.0
+2023-03-09 08:01:43,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2144, 1.2608, 3.4030, 3.1149], device='cuda:0'), covar=tensor([0.1556, 0.2960, 0.0463, 0.1914], device='cuda:0'), in_proj_covar=tensor([0.0729, 0.0629, 0.0923, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 08:01:58,916 INFO [train.py:968] (0/2) Epoch 18, batch 13750, giga_loss[loss=0.2467, simple_loss=0.3359, pruned_loss=0.07875, over 28395.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3377, pruned_loss=0.09043, over 5670946.26 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3405, pruned_loss=0.09136, over 5738689.84 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3391, pruned_loss=0.09108, over 5652428.10 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:01:59,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=790463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:02:44,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3187, 1.2853, 1.1772, 1.5227], device='cuda:0'), covar=tensor([0.0763, 0.0342, 0.0363, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 08:02:53,684 INFO [train.py:968] (0/2) Epoch 18, batch 13800, giga_loss[loss=0.2773, simple_loss=0.3488, pruned_loss=0.1029, over 28494.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3362, pruned_loss=0.0888, over 5679504.81 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3399, pruned_loss=0.09106, over 5743631.27 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3378, pruned_loss=0.08953, over 5658036.16 frames. ], batch size: 85, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:03:00,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3977, 3.3710, 1.4265, 1.6001], device='cuda:0'), covar=tensor([0.0973, 0.0313, 0.1003, 0.1285], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0541, 0.0371, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 08:03:22,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.785e+02 1.364e+03 1.753e+03 2.445e+03 7.279e+03, threshold=3.506e+03, percent-clipped=5.0
+2023-03-09 08:03:28,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2337, 1.7960, 1.5559, 1.4273], device='cuda:0'), covar=tensor([0.1762, 0.1512, 0.1783, 0.1780], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0725, 0.0685, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 08:03:31,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3772, 1.7418, 1.6542, 1.4756], device='cuda:0'), covar=tensor([0.1675, 0.1798, 0.1817, 0.1915], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0725, 0.0686, 0.0661], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 08:03:35,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5719, 2.0064, 1.8827, 1.6999], device='cuda:0'), covar=tensor([0.1609, 0.1553, 0.1694, 0.1666], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0725, 0.0686, 0.0662], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 08:03:52,022 INFO [train.py:968] (0/2) Epoch 18, batch 13850, giga_loss[loss=0.2304, simple_loss=0.3116, pruned_loss=0.07459, over 28609.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3339, pruned_loss=0.08893, over 5680649.94 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3394, pruned_loss=0.09083, over 5748914.53 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3354, pruned_loss=0.08966, over 5656826.95 frames. ], batch size: 243, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:04:12,365 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=790577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:04:44,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=790606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:04:48,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=790609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:04:53,244 INFO [train.py:968] (0/2) Epoch 18, batch 13900, giga_loss[loss=0.292, simple_loss=0.3481, pruned_loss=0.1179, over 26984.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3326, pruned_loss=0.0886, over 5678708.99 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3391, pruned_loss=0.09072, over 5751473.16 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3341, pruned_loss=0.08925, over 5656406.97 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:05:19,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=790634.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 08:05:20,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.071e+02 1.291e+03 1.919e+03 2.650e+03 1.017e+04, threshold=3.839e+03, percent-clipped=16.0
+2023-03-09 08:05:22,996 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=790638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:05:47,783 INFO [train.py:968] (0/2) Epoch 18, batch 13950, giga_loss[loss=0.2464, simple_loss=0.3302, pruned_loss=0.08127, over 28916.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3325, pruned_loss=0.08866, over 5677822.67 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3387, pruned_loss=0.09057, over 5754512.73 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3338, pruned_loss=0.08925, over 5655638.66 frames. ], batch size: 186, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:06:13,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5333, 1.6236, 1.8447, 1.3677], device='cuda:0'), covar=tensor([0.1881, 0.2522, 0.1492, 0.1840], device='cuda:0'), in_proj_covar=tensor([0.0870, 0.0687, 0.0916, 0.0817], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 08:06:39,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6339, 2.0369, 1.7636, 1.6075], device='cuda:0'), covar=tensor([0.1876, 0.2119, 0.1986, 0.2169], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0728, 0.0688, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 08:06:45,882 INFO [train.py:968] (0/2) Epoch 18, batch 14000, giga_loss[loss=0.2581, simple_loss=0.3459, pruned_loss=0.08513, over 28914.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3335, pruned_loss=0.08847, over 5666406.53 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3384, pruned_loss=0.09045, over 5755459.19 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3347, pruned_loss=0.089, over 5645680.16 frames. ], batch size: 284, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 08:07:16,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.225e+02 1.440e+03 1.994e+03 2.869e+03 7.649e+03, threshold=3.988e+03, percent-clipped=9.0
+2023-03-09 08:07:46,668 INFO [train.py:968] (0/2) Epoch 18, batch 14050, giga_loss[loss=0.2428, simple_loss=0.3207, pruned_loss=0.0824, over 28963.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3338, pruned_loss=0.08796, over 5673239.06 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3387, pruned_loss=0.09065, over 5758263.47 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3344, pruned_loss=0.08817, over 5652866.57 frames. ], batch size: 199, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:08:07,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=790777.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 08:08:11,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=790780.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 08:08:50,605 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=790809.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 08:08:52,950 INFO [train.py:968] (0/2) Epoch 18, batch 14100, giga_loss[loss=0.208, simple_loss=0.2965, pruned_loss=0.05976, over 28995.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3315, pruned_loss=0.08679, over 5679598.38 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3381, pruned_loss=0.09042, over 5761262.00 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3324, pruned_loss=0.08711, over 5659197.46 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:08:59,016 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-09 08:09:23,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.674e+02 1.363e+03 2.015e+03 2.743e+03 1.174e+04, threshold=4.030e+03, percent-clipped=8.0
+2023-03-09 08:09:31,883 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=790844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:09:53,122 INFO [train.py:968] (0/2) Epoch 18, batch 14150, giga_loss[loss=0.2648, simple_loss=0.3566, pruned_loss=0.08652, over 28886.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3339, pruned_loss=0.08791, over 5694971.36 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3383, pruned_loss=0.09061, over 5765524.06 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3344, pruned_loss=0.08792, over 5672633.86 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:10:34,971 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3686, 1.6745, 1.3757, 1.3340], device='cuda:0'), covar=tensor([0.2360, 0.2238, 0.2510, 0.2201], device='cuda:0'), in_proj_covar=tensor([0.1445, 0.1049, 0.1286, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 08:10:51,156 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=790906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:10:57,776 INFO [train.py:968] (0/2) Epoch 18, batch 14200, giga_loss[loss=0.2745, simple_loss=0.3612, pruned_loss=0.09391, over 28386.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3386, pruned_loss=0.0887, over 5684780.85 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3381, pruned_loss=0.09052, over 5767072.34 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3392, pruned_loss=0.08877, over 5664172.21 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:11:28,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.527e+02 1.376e+03 2.058e+03 3.212e+03 8.147e+03, threshold=4.116e+03, percent-clipped=15.0
+2023-03-09 08:11:39,506 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1768, 1.7717, 1.2434, 0.4402], device='cuda:0'), covar=tensor([0.4805, 0.2741, 0.4522, 0.5789], device='cuda:0'), in_proj_covar=tensor([0.1685, 0.1591, 0.1565, 0.1378], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 08:11:44,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=790952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:11:57,347 INFO [train.py:968] (0/2) Epoch 18, batch 14250, giga_loss[loss=0.2911, simple_loss=0.3691, pruned_loss=0.1065, over 28105.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.08804, over 5677175.04 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3383, pruned_loss=0.09062, over 5759618.90 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.341, pruned_loss=0.08797, over 5665287.16 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:12:01,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-09 08:12:38,252 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=790994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:12:58,169 INFO [train.py:968] (0/2) Epoch 18, batch 14300, giga_loss[loss=0.267, simple_loss=0.3492, pruned_loss=0.09242, over 28915.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3413, pruned_loss=0.08722, over 5674265.02 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3386, pruned_loss=0.09081, over 5759675.10 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3412, pruned_loss=0.08694, over 5663836.31 frames. ], batch size: 284, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:13:27,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.516e+02 1.330e+03 1.738e+03 2.601e+03 6.819e+03, threshold=3.475e+03, percent-clipped=4.0
+2023-03-09 08:13:54,080 INFO [train.py:968] (0/2) Epoch 18, batch 14350, libri_loss[loss=0.2247, simple_loss=0.3015, pruned_loss=0.07398, over 29558.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3405, pruned_loss=0.0874, over 5678213.46 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.338, pruned_loss=0.09066, over 5764736.27 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.341, pruned_loss=0.08714, over 5661739.86 frames. ], batch size: 76, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:14:15,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2895, 1.2065, 3.8933, 3.1803], device='cuda:0'), covar=tensor([0.1572, 0.2745, 0.0422, 0.1008], device='cuda:0'), in_proj_covar=tensor([0.0721, 0.0624, 0.0911, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 08:14:30,549 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=791095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:14:37,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=791098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:14:37,638 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=791099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:14:51,729 INFO [train.py:968] (0/2) Epoch 18, batch 14400, giga_loss[loss=0.2596, simple_loss=0.3384, pruned_loss=0.09045, over 29019.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3391, pruned_loss=0.0875, over 5685114.94 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.338, pruned_loss=0.09067, over 5767666.82 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3396, pruned_loss=0.08725, over 5668024.26 frames. ], batch size: 285, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 08:15:10,676 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=791127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:15:25,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.397e+02 1.285e+03 1.676e+03 2.521e+03 7.246e+03, threshold=3.352e+03, percent-clipped=12.0
+2023-03-09 08:16:00,954 INFO [train.py:968] (0/2) Epoch 18, batch 14450, giga_loss[loss=0.2684, simple_loss=0.3502, pruned_loss=0.0933, over 28938.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3401, pruned_loss=0.08892, over 5694249.85 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.338, pruned_loss=0.09064, over 5768107.39 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3406, pruned_loss=0.08872, over 5679734.28 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 08:17:20,138 INFO [train.py:968] (0/2) Epoch 18, batch 14500, giga_loss[loss=0.2196, simple_loss=0.3056, pruned_loss=0.06674, over 29180.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3375, pruned_loss=0.08816, over 5688305.85 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3378, pruned_loss=0.09053, over 5770746.14 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3381, pruned_loss=0.08804, over 5672451.33 frames. ], batch size: 200, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:17:28,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7894, 3.6074, 3.4213, 1.7412], device='cuda:0'), covar=tensor([0.0730, 0.0826, 0.0824, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.1151, 0.1064, 0.0917, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 08:17:29,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=791219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:17:57,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.996e+02 1.304e+03 1.716e+03 2.206e+03 6.884e+03, threshold=3.432e+03, percent-clipped=5.0
+2023-03-09 08:18:21,159 INFO [train.py:968] (0/2) Epoch 18, batch 14550, giga_loss[loss=0.2746, simple_loss=0.3295, pruned_loss=0.1099, over 23698.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3337, pruned_loss=0.08612, over 5685668.09 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3367, pruned_loss=0.09001, over 5768598.18 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.335, pruned_loss=0.08632, over 5670706.21 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:18:43,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=791281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:19:24,576 INFO [train.py:968] (0/2) Epoch 18, batch 14600, giga_loss[loss=0.3274, simple_loss=0.3906, pruned_loss=0.1321, over 28475.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3315, pruned_loss=0.08487, over 5686386.22 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3367, pruned_loss=0.09002, over 5768052.01 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3325, pruned_loss=0.08493, over 5673057.02 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:19:42,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8018, 2.0319, 1.2836, 1.6762], device='cuda:0'), covar=tensor([0.0951, 0.0643, 0.1045, 0.1145], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0440, 0.0505, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 08:19:49,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7578, 1.0055, 2.9185, 2.7383], device='cuda:0'), covar=tensor([0.1728, 0.2602, 0.0640, 0.1071], device='cuda:0'), in_proj_covar=tensor([0.0722, 0.0624, 0.0916, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 08:19:59,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.731e+02 1.379e+03 1.833e+03 2.384e+03 8.034e+03, threshold=3.667e+03, percent-clipped=8.0
+2023-03-09 08:20:26,576 INFO [train.py:968] (0/2) Epoch 18, batch 14650, giga_loss[loss=0.2268, simple_loss=0.3199, pruned_loss=0.06686, over 29029.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3319, pruned_loss=0.0856, over 5682461.43 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3366, pruned_loss=0.08998, over 5770964.42 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3327, pruned_loss=0.08558, over 5667076.95 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:20:26,957 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=791362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:20:29,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=791365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:20:33,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=791369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:20:59,033 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.36 vs. limit=5.0
+2023-03-09 08:21:02,550 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=791394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:21:21,995 INFO [train.py:968] (0/2) Epoch 18, batch 14700, giga_loss[loss=0.2414, simple_loss=0.3221, pruned_loss=0.08037, over 28393.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3356, pruned_loss=0.08737, over 5682203.82 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3361, pruned_loss=0.08973, over 5766783.04 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3367, pruned_loss=0.08747, over 5670005.40 frames. ], batch size: 78, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:21:25,689 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 08:21:34,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5217, 1.7331, 1.4409, 1.8424], device='cuda:0'), covar=tensor([0.2732, 0.2705, 0.3120, 0.2261], device='cuda:0'), in_proj_covar=tensor([0.1443, 0.1044, 0.1281, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 08:21:37,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=791424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:21:40,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=791427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:21:53,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.375e+02 1.476e+03 1.955e+03 3.035e+03 9.108e+03, threshold=3.909e+03, percent-clipped=17.0
+2023-03-09 08:22:13,215 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=791456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:22:19,632 INFO [train.py:968] (0/2) Epoch 18, batch 14750, giga_loss[loss=0.277, simple_loss=0.3462, pruned_loss=0.1039, over 27625.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3348, pruned_loss=0.08834, over 5688949.20 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3361, pruned_loss=0.08993, over 5770422.82 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3357, pruned_loss=0.0882, over 5673941.95 frames. ], batch size: 472, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:22:36,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=791474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:22:59,333 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3476, 4.1758, 3.9717, 2.1121], device='cuda:0'), covar=tensor([0.0546, 0.0681, 0.0754, 0.1852], device='cuda:0'), in_proj_covar=tensor([0.1144, 0.1061, 0.0912, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 08:23:25,063 INFO [train.py:968] (0/2) Epoch 18, batch 14800, giga_loss[loss=0.2477, simple_loss=0.3244, pruned_loss=0.08549, over 28736.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3348, pruned_loss=0.08935, over 5693346.14 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.336, pruned_loss=0.09, over 5772508.22 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3355, pruned_loss=0.08916, over 5678572.45 frames. ], batch size: 119, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 08:23:25,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=791512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:23:29,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=791515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:23:31,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4117, 2.0187, 1.5222, 0.6245], device='cuda:0'), covar=tensor([0.4617, 0.2408, 0.3252, 0.5321], device='cuda:0'), in_proj_covar=tensor([0.1686, 0.1593, 0.1568, 0.1380], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 08:23:55,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.010e+02 1.453e+03 1.920e+03 2.303e+03 5.754e+03, threshold=3.840e+03, percent-clipped=5.0
+2023-03-09 08:24:01,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=791544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:24:24,196 INFO [train.py:968] (0/2) Epoch 18, batch 14850, giga_loss[loss=0.2938, simple_loss=0.3609, pruned_loss=0.1133, over 28716.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3365, pruned_loss=0.09074, over 5684072.72 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3365, pruned_loss=0.09039, over 5767255.67 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3366, pruned_loss=0.09023, over 5675248.39 frames. ], batch size: 242, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:25:32,155 INFO [train.py:968] (0/2) Epoch 18, batch 14900, giga_loss[loss=0.2565, simple_loss=0.3493, pruned_loss=0.08187, over 28940.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3389, pruned_loss=0.09118, over 5684168.05 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3366, pruned_loss=0.09055, over 5769094.38 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.339, pruned_loss=0.09066, over 5674749.02 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:25:41,145 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=791617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:25:45,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=791620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:26:14,717 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.346e+02 1.464e+03 1.863e+03 2.735e+03 9.410e+03, threshold=3.726e+03, percent-clipped=12.0
+2023-03-09 08:26:25,552 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=791649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:26:43,537 INFO [train.py:968] (0/2) Epoch 18, batch 14950, giga_loss[loss=0.2736, simple_loss=0.3486, pruned_loss=0.09927, over 28274.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3391, pruned_loss=0.09072, over 5665613.74 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3367, pruned_loss=0.09055, over 5753378.60 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.339, pruned_loss=0.09029, over 5671138.85 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:27:07,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2318, 2.9815, 1.3790, 1.3866], device='cuda:0'), covar=tensor([0.0990, 0.0446, 0.0963, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0535, 0.0367, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 08:27:50,607 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-09 08:27:57,001 INFO [train.py:968] (0/2) Epoch 18, batch 15000, giga_loss[loss=0.263, simple_loss=0.3332, pruned_loss=0.09645, over 28363.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3372, pruned_loss=0.09072, over 5644413.21 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3363, pruned_loss=0.09047, over 5737481.83 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3376, pruned_loss=0.09046, over 5660904.75 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:27:57,005 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 08:28:05,680 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.1995, simple_loss=0.2997, pruned_loss=0.04963, over 944034.00 frames. 
+2023-03-09 08:28:05,681 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 08:28:38,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.377e+02 1.602e+03 2.188e+03 3.132e+03 1.116e+04, threshold=4.377e+03, percent-clipped=14.0
+2023-03-09 08:29:03,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4386, 1.7608, 1.6710, 1.5799], device='cuda:0'), covar=tensor([0.1762, 0.2096, 0.2113, 0.1913], device='cuda:0'), in_proj_covar=tensor([0.0441, 0.0719, 0.0680, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 08:29:08,805 INFO [train.py:968] (0/2) Epoch 18, batch 15050, giga_loss[loss=0.238, simple_loss=0.3058, pruned_loss=0.08509, over 28670.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3314, pruned_loss=0.08844, over 5649171.05 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.336, pruned_loss=0.09035, over 5739861.44 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.332, pruned_loss=0.08835, over 5658169.78 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:29:16,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3541, 1.4646, 1.2880, 1.5859], device='cuda:0'), covar=tensor([0.0732, 0.0346, 0.0342, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 08:30:08,084 INFO [train.py:968] (0/2) Epoch 18, batch 15100, libri_loss[loss=0.2444, simple_loss=0.3254, pruned_loss=0.0817, over 29762.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3276, pruned_loss=0.08653, over 5657393.87 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3357, pruned_loss=0.0902, over 5744298.09 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3282, pruned_loss=0.08652, over 5658419.18 frames. ], batch size: 87, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:30:31,753 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-09 08:30:41,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.931e+02 1.481e+03 1.882e+03 2.697e+03 9.781e+03, threshold=3.764e+03, percent-clipped=9.0
+2023-03-09 08:30:53,997 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=791850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:31:06,836 INFO [train.py:968] (0/2) Epoch 18, batch 15150, giga_loss[loss=0.2801, simple_loss=0.3556, pruned_loss=0.1023, over 28879.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3289, pruned_loss=0.08748, over 5656285.21 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3355, pruned_loss=0.08999, over 5745897.80 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3294, pruned_loss=0.08763, over 5654347.18 frames. ], batch size: 284, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:32:04,150 INFO [train.py:968] (0/2) Epoch 18, batch 15200, giga_loss[loss=0.2198, simple_loss=0.3026, pruned_loss=0.06851, over 29075.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3284, pruned_loss=0.08688, over 5658326.36 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3353, pruned_loss=0.08994, over 5739146.60 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.329, pruned_loss=0.08702, over 5660943.27 frames. ], batch size: 285, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 08:32:35,132 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.232e+02 1.389e+03 1.764e+03 2.379e+03 4.718e+03, threshold=3.529e+03, percent-clipped=4.0
+2023-03-09 08:32:40,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4022, 1.0401, 4.0531, 3.4073], device='cuda:0'), covar=tensor([0.1661, 0.3015, 0.0462, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0724, 0.0627, 0.0915, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 08:33:01,726 INFO [train.py:968] (0/2) Epoch 18, batch 15250, giga_loss[loss=0.2498, simple_loss=0.3346, pruned_loss=0.08248, over 28887.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3266, pruned_loss=0.08524, over 5652707.38 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3348, pruned_loss=0.08969, over 5741586.46 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3272, pruned_loss=0.08546, over 5649686.97 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:33:09,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2438, 1.5156, 1.3787, 1.1158], device='cuda:0'), covar=tensor([0.2346, 0.2050, 0.1450, 0.1946], device='cuda:0'), in_proj_covar=tensor([0.1843, 0.1775, 0.1697, 0.1838], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 08:33:42,821 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-792000.pt
+2023-03-09 08:34:03,174 INFO [train.py:968] (0/2) Epoch 18, batch 15300, giga_loss[loss=0.28, simple_loss=0.3489, pruned_loss=0.1056, over 29012.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3259, pruned_loss=0.08474, over 5665327.84 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3348, pruned_loss=0.08988, over 5745068.73 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3263, pruned_loss=0.08465, over 5658444.11 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:34:42,870 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 08:34:43,007 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.798e+02 1.459e+03 1.916e+03 2.550e+03 8.365e+03, threshold=3.831e+03, percent-clipped=14.0
+2023-03-09 08:35:06,943 INFO [train.py:968] (0/2) Epoch 18, batch 15350, giga_loss[loss=0.2213, simple_loss=0.3059, pruned_loss=0.06835, over 28864.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3254, pruned_loss=0.08437, over 5653700.91 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3346, pruned_loss=0.08983, over 5739181.55 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3256, pruned_loss=0.08424, over 5651155.04 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:35:49,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-09 08:36:11,550 INFO [train.py:968] (0/2) Epoch 18, batch 15400, libri_loss[loss=0.2537, simple_loss=0.3374, pruned_loss=0.08498, over 29675.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3269, pruned_loss=0.08495, over 5657025.64 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3344, pruned_loss=0.08964, over 5743650.15 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3271, pruned_loss=0.08491, over 5648952.72 frames. ], batch size: 88, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:36:43,297 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-09 08:36:43,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.427e+02 1.289e+03 1.627e+03 2.100e+03 4.746e+03, threshold=3.254e+03, percent-clipped=1.0
+2023-03-09 08:37:05,861 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-09 08:37:09,149 INFO [train.py:968] (0/2) Epoch 18, batch 15450, libri_loss[loss=0.2168, simple_loss=0.2932, pruned_loss=0.07019, over 29554.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3274, pruned_loss=0.08554, over 5663740.27 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3343, pruned_loss=0.08958, over 5746852.79 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3274, pruned_loss=0.08542, over 5651487.42 frames. ], batch size: 75, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:37:23,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3485, 1.3960, 3.4979, 3.0553], device='cuda:0'), covar=tensor([0.1503, 0.2589, 0.0498, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0724, 0.0626, 0.0914, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 08:37:46,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3180, 1.6595, 1.2827, 0.9511], device='cuda:0'), covar=tensor([0.2635, 0.2550, 0.2963, 0.2321], device='cuda:0'), in_proj_covar=tensor([0.1443, 0.1045, 0.1283, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 08:37:57,841 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3585, 2.1131, 1.5532, 0.5557], device='cuda:0'), covar=tensor([0.3993, 0.2467, 0.3508, 0.4463], device='cuda:0'), in_proj_covar=tensor([0.1689, 0.1597, 0.1573, 0.1385], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 08:38:09,834 INFO [train.py:968] (0/2) Epoch 18, batch 15500, giga_loss[loss=0.228, simple_loss=0.309, pruned_loss=0.0735, over 28908.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3278, pruned_loss=0.08656, over 5662760.61 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3346, pruned_loss=0.08993, over 5750006.84 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3274, pruned_loss=0.08609, over 5648633.38 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:38:26,156 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-09 08:38:27,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=792225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:38:43,124 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-09 08:38:45,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.796e+02 1.355e+03 1.931e+03 2.656e+03 9.143e+03, threshold=3.862e+03, percent-clipped=16.0
+2023-03-09 08:39:11,078 INFO [train.py:968] (0/2) Epoch 18, batch 15550, giga_loss[loss=0.2425, simple_loss=0.3327, pruned_loss=0.07612, over 28592.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3282, pruned_loss=0.08508, over 5672231.13 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3344, pruned_loss=0.08984, over 5750897.81 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3281, pruned_loss=0.08477, over 5660086.00 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 08:39:27,269 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=792276.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 08:40:09,138 INFO [train.py:968] (0/2) Epoch 18, batch 15600, giga_loss[loss=0.2196, simple_loss=0.2926, pruned_loss=0.07333, over 24250.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3306, pruned_loss=0.08536, over 5661401.94 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3342, pruned_loss=0.08966, over 5743826.88 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3305, pruned_loss=0.08516, over 5656344.57 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:40:36,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3358, 1.6052, 1.5882, 1.1605], device='cuda:0'), covar=tensor([0.1870, 0.2653, 0.1582, 0.1837], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0687, 0.0918, 0.0818], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 08:40:45,379 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.838e+02 1.341e+03 1.939e+03 2.862e+03 7.653e+03, threshold=3.878e+03, percent-clipped=15.0
+2023-03-09 08:41:10,443 INFO [train.py:968] (0/2) Epoch 18, batch 15650, giga_loss[loss=0.3092, simple_loss=0.368, pruned_loss=0.1252, over 26781.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3326, pruned_loss=0.08613, over 5659116.72 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3344, pruned_loss=0.08967, over 5745145.84 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3323, pruned_loss=0.08589, over 5652923.44 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:41:18,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=792368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:41:21,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=792371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:41:37,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8338, 2.6436, 1.7175, 0.9494], device='cuda:0'), covar=tensor([0.7061, 0.3400, 0.4056, 0.6312], device='cuda:0'), in_proj_covar=tensor([0.1690, 0.1597, 0.1569, 0.1383], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 08:41:56,128 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=792400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:42:07,575 INFO [train.py:968] (0/2) Epoch 18, batch 15700, giga_loss[loss=0.2305, simple_loss=0.3185, pruned_loss=0.0713, over 28466.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3324, pruned_loss=0.08561, over 5672665.33 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3343, pruned_loss=0.08963, over 5747586.61 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3322, pruned_loss=0.0854, over 5664365.66 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:42:44,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.757e+02 1.332e+03 1.883e+03 2.565e+03 3.972e+03, threshold=3.766e+03, percent-clipped=2.0
+2023-03-09 08:43:01,944 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 08:43:07,856 INFO [train.py:968] (0/2) Epoch 18, batch 15750, giga_loss[loss=0.226, simple_loss=0.3108, pruned_loss=0.07059, over 28761.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.331, pruned_loss=0.08471, over 5681431.70 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3343, pruned_loss=0.08959, over 5749350.15 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3309, pruned_loss=0.08455, over 5672717.41 frames. ], batch size: 262, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:43:19,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3438, 1.7013, 1.5573, 1.4820], device='cuda:0'), covar=tensor([0.1846, 0.2012, 0.2237, 0.2052], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0716, 0.0676, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 08:44:07,576 INFO [train.py:968] (0/2) Epoch 18, batch 15800, giga_loss[loss=0.2419, simple_loss=0.3303, pruned_loss=0.07668, over 28065.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3277, pruned_loss=0.0825, over 5686166.50 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3336, pruned_loss=0.08927, over 5753137.63 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3281, pruned_loss=0.08253, over 5674479.09 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:44:33,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8820, 1.3698, 1.3497, 1.1424], device='cuda:0'), covar=tensor([0.1940, 0.1149, 0.2268, 0.1559], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0718, 0.0679, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 08:44:39,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.309e+02 1.310e+03 1.816e+03 2.346e+03 6.875e+03, threshold=3.632e+03, percent-clipped=6.0
+2023-03-09 08:45:04,051 INFO [train.py:968] (0/2) Epoch 18, batch 15850, giga_loss[loss=0.2554, simple_loss=0.3261, pruned_loss=0.09237, over 28918.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3263, pruned_loss=0.08268, over 5686045.33 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3338, pruned_loss=0.08949, over 5755715.12 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3263, pruned_loss=0.08232, over 5672627.70 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:45:20,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2955, 1.2307, 3.5245, 3.1072], device='cuda:0'), covar=tensor([0.1569, 0.2720, 0.0497, 0.1238], device='cuda:0'), in_proj_covar=tensor([0.0725, 0.0626, 0.0915, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 08:45:20,886 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-09 08:45:32,787 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=792589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:46:02,448 INFO [train.py:968] (0/2) Epoch 18, batch 15900, giga_loss[loss=0.2679, simple_loss=0.3545, pruned_loss=0.09061, over 28901.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3266, pruned_loss=0.08281, over 5672416.60 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.334, pruned_loss=0.08954, over 5745184.43 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3264, pruned_loss=0.08238, over 5669487.85 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:46:40,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.136e+02 1.338e+03 1.689e+03 2.290e+03 7.295e+03, threshold=3.378e+03, percent-clipped=10.0
+2023-03-09 08:46:47,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=792651.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 08:46:59,848 INFO [train.py:968] (0/2) Epoch 18, batch 15950, giga_loss[loss=0.2814, simple_loss=0.3499, pruned_loss=0.1065, over 28017.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3295, pruned_loss=0.08464, over 5676239.33 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3334, pruned_loss=0.08932, over 5747802.40 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3296, pruned_loss=0.08431, over 5669181.17 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 08:47:19,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2179, 1.3605, 1.2132, 1.2364], device='cuda:0'), covar=tensor([0.1838, 0.1615, 0.1285, 0.1441], device='cuda:0'), in_proj_covar=tensor([0.1857, 0.1789, 0.1701, 0.1851], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 08:47:52,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6261, 1.8640, 1.5547, 1.6212], device='cuda:0'), covar=tensor([0.2667, 0.2620, 0.2944, 0.2324], device='cuda:0'), in_proj_covar=tensor([0.1437, 0.1038, 0.1276, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 08:48:02,087 INFO [train.py:968] (0/2) Epoch 18, batch 16000, libri_loss[loss=0.2692, simple_loss=0.3454, pruned_loss=0.09655, over 19320.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.33, pruned_loss=0.0856, over 5660037.11 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3329, pruned_loss=0.08912, over 5733236.41 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3304, pruned_loss=0.08538, over 5665747.89 frames. ], batch size: 186, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:48:35,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.451e+02 1.347e+03 1.802e+03 2.525e+03 5.106e+03, threshold=3.605e+03, percent-clipped=7.0
+2023-03-09 08:48:56,680 INFO [train.py:968] (0/2) Epoch 18, batch 16050, giga_loss[loss=0.2411, simple_loss=0.3256, pruned_loss=0.07836, over 28939.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3325, pruned_loss=0.08685, over 5668113.32 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3326, pruned_loss=0.08896, over 5734154.07 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.333, pruned_loss=0.08676, over 5669999.36 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:49:31,768 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=792794.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 08:49:34,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=792797.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 08:49:47,169 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=792807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:49:52,579 INFO [train.py:968] (0/2) Epoch 18, batch 16100, giga_loss[loss=0.2996, simple_loss=0.371, pruned_loss=0.1141, over 29027.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3359, pruned_loss=0.08804, over 5677279.35 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3329, pruned_loss=0.08921, over 5737410.87 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3361, pruned_loss=0.08771, over 5674549.18 frames. ], batch size: 285, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:50:07,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3104, 1.5632, 1.5975, 1.4604], device='cuda:0'), covar=tensor([0.1497, 0.1360, 0.1521, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0718, 0.0678, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 08:50:09,749 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=792826.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 08:50:27,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.996e+02 1.411e+03 1.988e+03 2.894e+03 9.262e+03, threshold=3.976e+03, percent-clipped=15.0
+2023-03-09 08:50:50,016 INFO [train.py:968] (0/2) Epoch 18, batch 16150, giga_loss[loss=0.2583, simple_loss=0.3423, pruned_loss=0.08711, over 28982.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3369, pruned_loss=0.08869, over 5683104.71 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3324, pruned_loss=0.08898, over 5741814.74 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3377, pruned_loss=0.08863, over 5675375.93 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:51:25,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2033, 2.8772, 1.3639, 1.3810], device='cuda:0'), covar=tensor([0.0987, 0.0382, 0.0939, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0539, 0.0370, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 08:51:56,933 INFO [train.py:968] (0/2) Epoch 18, batch 16200, giga_loss[loss=0.2227, simple_loss=0.3031, pruned_loss=0.07114, over 28467.00 frames. ], tot_loss[loss=0.255, simple_loss=0.335, pruned_loss=0.08745, over 5689145.20 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3322, pruned_loss=0.08886, over 5744446.76 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3359, pruned_loss=0.0875, over 5680141.54 frames. ], batch size: 78, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:52:35,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.888e+02 1.426e+03 1.795e+03 2.709e+03 8.096e+03, threshold=3.591e+03, percent-clipped=8.0
+2023-03-09 08:53:00,602 INFO [train.py:968] (0/2) Epoch 18, batch 16250, libri_loss[loss=0.251, simple_loss=0.3357, pruned_loss=0.08321, over 25845.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3339, pruned_loss=0.08762, over 5682297.17 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3321, pruned_loss=0.08884, over 5741769.38 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3348, pruned_loss=0.08766, over 5676519.61 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:53:03,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=792964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:53:45,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4958, 1.6241, 1.8143, 1.4092], device='cuda:0'), covar=tensor([0.1711, 0.2251, 0.1372, 0.1852], device='cuda:0'), in_proj_covar=tensor([0.0870, 0.0686, 0.0917, 0.0817], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 08:53:56,282 INFO [train.py:968] (0/2) Epoch 18, batch 16300, giga_loss[loss=0.2561, simple_loss=0.3391, pruned_loss=0.0866, over 29051.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3333, pruned_loss=0.08717, over 5679037.54 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3323, pruned_loss=0.08871, over 5749965.70 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3339, pruned_loss=0.08727, over 5664435.74 frames. ], batch size: 285, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:54:30,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.705e+02 1.309e+03 1.639e+03 2.161e+03 5.474e+03, threshold=3.279e+03, percent-clipped=5.0
+2023-03-09 08:54:52,595 INFO [train.py:968] (0/2) Epoch 18, batch 16350, giga_loss[loss=0.2588, simple_loss=0.3324, pruned_loss=0.09261, over 28058.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3322, pruned_loss=0.08732, over 5677973.36 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3322, pruned_loss=0.0886, over 5743587.59 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3328, pruned_loss=0.08746, over 5669631.44 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:54:53,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2658, 0.8390, 0.8826, 1.4559], device='cuda:0'), covar=tensor([0.0757, 0.0379, 0.0376, 0.0842], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0116, 0.0117, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0066, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 08:55:47,615 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=793107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:55:49,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=793110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:55:53,365 INFO [train.py:968] (0/2) Epoch 18, batch 16400, giga_loss[loss=0.2476, simple_loss=0.3257, pruned_loss=0.08476, over 28155.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3308, pruned_loss=0.08738, over 5675301.83 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3318, pruned_loss=0.08841, over 5746396.72 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3315, pruned_loss=0.08764, over 5664834.59 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 08:56:22,827 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=793139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:56:26,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.205e+02 1.314e+03 1.945e+03 3.004e+03 8.393e+03, threshold=3.891e+03, percent-clipped=24.0
+2023-03-09 08:56:29,921 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-09 08:56:50,811 INFO [train.py:968] (0/2) Epoch 18, batch 16450, giga_loss[loss=0.2507, simple_loss=0.332, pruned_loss=0.08469, over 27661.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3314, pruned_loss=0.08707, over 5680542.88 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3318, pruned_loss=0.08849, over 5751482.48 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.332, pruned_loss=0.08719, over 5665971.17 frames. ], batch size: 472, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:57:11,213 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=793180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:57:13,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=793182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:57:49,117 INFO [train.py:968] (0/2) Epoch 18, batch 16500, giga_loss[loss=0.2755, simple_loss=0.3647, pruned_loss=0.09311, over 29159.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3314, pruned_loss=0.08599, over 5679460.51 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3318, pruned_loss=0.08851, over 5750508.06 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3319, pruned_loss=0.08605, over 5667770.97 frames. ], batch size: 200, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:58:19,787 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=793241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:58:22,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.533e+02 1.380e+03 2.043e+03 2.663e+03 7.290e+03, threshold=4.086e+03, percent-clipped=9.0
+2023-03-09 08:58:25,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5000, 2.1811, 1.4879, 0.7159], device='cuda:0'), covar=tensor([0.5410, 0.2787, 0.4804, 0.6012], device='cuda:0'), in_proj_covar=tensor([0.1700, 0.1600, 0.1571, 0.1387], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 08:58:40,845 INFO [train.py:968] (0/2) Epoch 18, batch 16550, giga_loss[loss=0.2702, simple_loss=0.3558, pruned_loss=0.09236, over 28471.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3329, pruned_loss=0.08525, over 5674124.97 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3316, pruned_loss=0.08838, over 5743516.57 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3335, pruned_loss=0.08533, over 5669157.57 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:58:47,261 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-09 08:59:36,066 INFO [train.py:968] (0/2) Epoch 18, batch 16600, giga_loss[loss=0.2581, simple_loss=0.3406, pruned_loss=0.08784, over 28924.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3336, pruned_loss=0.08498, over 5671625.77 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3314, pruned_loss=0.08831, over 5737432.10 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3343, pruned_loss=0.08504, over 5672413.15 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:59:52,870 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=793325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:59:56,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=793328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:00:15,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.358e+02 1.388e+03 1.856e+03 2.462e+03 5.308e+03, threshold=3.713e+03, percent-clipped=5.0
+2023-03-09 09:00:29,971 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=793357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:00:35,712 INFO [train.py:968] (0/2) Epoch 18, batch 16650, giga_loss[loss=0.2353, simple_loss=0.3265, pruned_loss=0.07199, over 28907.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3329, pruned_loss=0.08452, over 5680864.41 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3307, pruned_loss=0.08808, over 5741685.25 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3342, pruned_loss=0.08471, over 5676147.43 frames. ], batch size: 164, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:01:38,221 INFO [train.py:968] (0/2) Epoch 18, batch 16700, giga_loss[loss=0.2115, simple_loss=0.3001, pruned_loss=0.06148, over 28959.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3329, pruned_loss=0.08436, over 5668341.27 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.331, pruned_loss=0.08831, over 5735183.72 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3337, pruned_loss=0.08423, over 5669425.85 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:02:23,610 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.361e+02 1.399e+03 2.074e+03 2.813e+03 7.173e+03, threshold=4.148e+03, percent-clipped=9.0
+2023-03-09 09:02:42,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2404, 1.6662, 1.5686, 1.3677], device='cuda:0'), covar=tensor([0.1912, 0.1839, 0.2180, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0718, 0.0678, 0.0655], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 09:02:48,208 INFO [train.py:968] (0/2) Epoch 18, batch 16750, giga_loss[loss=0.2566, simple_loss=0.3301, pruned_loss=0.09155, over 27019.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3329, pruned_loss=0.08441, over 5667508.11 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3307, pruned_loss=0.08823, over 5739458.31 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3338, pruned_loss=0.08428, over 5662927.90 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:02:48,415 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=793462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:03:04,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 09:03:32,642 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=793495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:03:38,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4320, 1.6699, 1.6879, 1.2456], device='cuda:0'), covar=tensor([0.1839, 0.2775, 0.1610, 0.1945], device='cuda:0'), in_proj_covar=tensor([0.0868, 0.0683, 0.0916, 0.0818], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 09:03:52,479 INFO [train.py:968] (0/2) Epoch 18, batch 16800, giga_loss[loss=0.2801, simple_loss=0.3601, pruned_loss=0.1001, over 28472.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3326, pruned_loss=0.08317, over 5679404.75 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3301, pruned_loss=0.08777, over 5742236.60 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.334, pruned_loss=0.08339, over 5671749.18 frames. ], batch size: 369, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:04:10,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1988, 1.6337, 1.3738, 1.3671], device='cuda:0'), covar=tensor([0.1880, 0.1777, 0.2111, 0.1834], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0717, 0.0676, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 09:04:36,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.668e+02 1.377e+03 1.936e+03 2.719e+03 8.148e+03, threshold=3.871e+03, percent-clipped=7.0
+2023-03-09 09:04:49,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=793555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:04:56,961 INFO [train.py:968] (0/2) Epoch 18, batch 16850, giga_loss[loss=0.3171, simple_loss=0.3863, pruned_loss=0.1239, over 27594.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3355, pruned_loss=0.08523, over 5668029.81 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3299, pruned_loss=0.08766, over 5734654.56 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3369, pruned_loss=0.08539, over 5666361.80 frames. ], batch size: 472, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:05:20,957 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-09 09:05:44,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5159, 1.6473, 1.8055, 1.3086], device='cuda:0'), covar=tensor([0.1991, 0.2696, 0.1658, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.0866, 0.0681, 0.0914, 0.0816], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 09:05:55,125 INFO [train.py:968] (0/2) Epoch 18, batch 16900, giga_loss[loss=0.2283, simple_loss=0.3225, pruned_loss=0.06703, over 29285.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3376, pruned_loss=0.08584, over 5679129.14 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3297, pruned_loss=0.0875, over 5731674.45 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3392, pruned_loss=0.08605, over 5677789.22 frames. ], batch size: 129, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:06:01,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=793616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:06:35,978 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.633e+02 1.371e+03 1.735e+03 2.311e+03 5.982e+03, threshold=3.470e+03, percent-clipped=6.0
+2023-03-09 09:06:54,167 INFO [train.py:968] (0/2) Epoch 18, batch 16950, giga_loss[loss=0.2291, simple_loss=0.3133, pruned_loss=0.07246, over 29073.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3382, pruned_loss=0.08683, over 5674092.48 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3303, pruned_loss=0.0877, over 5725119.98 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3393, pruned_loss=0.08676, over 5676703.65 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:07:47,727 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=793698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:07:50,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=793701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:07:54,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8641, 1.1327, 2.8691, 2.6943], device='cuda:0'), covar=tensor([0.1634, 0.2514, 0.0575, 0.0966], device='cuda:0'), in_proj_covar=tensor([0.0723, 0.0626, 0.0912, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 09:08:06,201 INFO [train.py:968] (0/2) Epoch 18, batch 17000, giga_loss[loss=0.2349, simple_loss=0.322, pruned_loss=0.07391, over 28707.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3361, pruned_loss=0.0861, over 5680742.61 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3305, pruned_loss=0.08784, over 5725182.04 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3368, pruned_loss=0.0859, over 5682322.17 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:08:29,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=793730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:08:50,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.227e+02 1.407e+03 1.937e+03 3.096e+03 9.231e+03, threshold=3.874e+03, percent-clipped=23.0
+2023-03-09 09:09:09,897 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=793759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:09:12,745 INFO [train.py:968] (0/2) Epoch 18, batch 17050, giga_loss[loss=0.2702, simple_loss=0.3614, pruned_loss=0.0895, over 28443.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3341, pruned_loss=0.0842, over 5691403.51 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3303, pruned_loss=0.08779, over 5729988.90 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.335, pruned_loss=0.08402, over 5687532.07 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:09:13,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=793762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:09:43,574 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=793788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:09:47,234 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=793791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:09:52,124 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 09:10:11,702 INFO [train.py:968] (0/2) Epoch 18, batch 17100, giga_loss[loss=0.2797, simple_loss=0.3594, pruned_loss=0.09997, over 28510.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3338, pruned_loss=0.08437, over 5688210.25 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3304, pruned_loss=0.088, over 5724702.90 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3345, pruned_loss=0.08397, over 5688515.18 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:10:39,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=793837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:10:50,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.313e+02 1.144e+03 1.637e+03 2.119e+03 8.766e+03, threshold=3.275e+03, percent-clipped=2.0
+2023-03-09 09:11:07,910 INFO [train.py:968] (0/2) Epoch 18, batch 17150, giga_loss[loss=0.2535, simple_loss=0.3353, pruned_loss=0.08587, over 28920.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3359, pruned_loss=0.08587, over 5678517.06 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3307, pruned_loss=0.08824, over 5718535.92 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3363, pruned_loss=0.08526, over 5683116.97 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:11:17,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=793870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:12:03,563 INFO [train.py:968] (0/2) Epoch 18, batch 17200, giga_loss[loss=0.3093, simple_loss=0.3761, pruned_loss=0.1212, over 28714.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3376, pruned_loss=0.08717, over 5676283.79 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3311, pruned_loss=0.08867, over 5721954.00 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3377, pruned_loss=0.08627, over 5675929.04 frames. ], batch size: 243, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:12:41,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.791e+02 1.483e+03 1.872e+03 2.782e+03 1.006e+04, threshold=3.744e+03, percent-clipped=18.0
+2023-03-09 09:12:59,299 INFO [train.py:968] (0/2) Epoch 18, batch 17250, giga_loss[loss=0.2725, simple_loss=0.3387, pruned_loss=0.1031, over 28829.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3356, pruned_loss=0.08677, over 5679317.78 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3308, pruned_loss=0.08843, over 5724556.48 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3361, pruned_loss=0.08626, over 5676031.72 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:13:18,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3899, 1.4299, 1.3119, 1.5589], device='cuda:0'), covar=tensor([0.0762, 0.0339, 0.0352, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0115, 0.0117, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 09:13:21,148 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=793980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:13:24,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=793983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:13:44,157 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-794000.pt
+2023-03-09 09:13:57,204 INFO [train.py:968] (0/2) Epoch 18, batch 17300, giga_loss[loss=0.246, simple_loss=0.3307, pruned_loss=0.0806, over 28937.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3343, pruned_loss=0.08683, over 5675953.62 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3306, pruned_loss=0.08832, over 5723668.81 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3348, pruned_loss=0.08649, over 5673282.28 frames. ], batch size: 164, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:13:57,842 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=794012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:13:59,897 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=794013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:14:02,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=794016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:14:31,878 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=794045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:14:32,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.708e+02 1.558e+03 2.195e+03 2.877e+03 1.051e+04, threshold=4.390e+03, percent-clipped=9.0
+2023-03-09 09:14:47,960 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:14:53,849 INFO [train.py:968] (0/2) Epoch 18, batch 17350, giga_loss[loss=0.2511, simple_loss=0.3254, pruned_loss=0.08841, over 28800.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3346, pruned_loss=0.08776, over 5682657.75 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3305, pruned_loss=0.08838, over 5724257.81 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3353, pruned_loss=0.08741, over 5678742.88 frames. ], batch size: 99, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:15:43,130 INFO [train.py:968] (0/2) Epoch 18, batch 17400, giga_loss[loss=0.3536, simple_loss=0.4262, pruned_loss=0.1405, over 28807.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3422, pruned_loss=0.09251, over 5680108.55 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3307, pruned_loss=0.08855, over 5727313.21 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3428, pruned_loss=0.09209, over 5672911.52 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:16:12,990 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.403e+02 1.361e+03 1.749e+03 2.301e+03 8.827e+03, threshold=3.497e+03, percent-clipped=4.0
+2023-03-09 09:16:15,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:16:28,216 INFO [train.py:968] (0/2) Epoch 18, batch 17450, giga_loss[loss=0.304, simple_loss=0.3874, pruned_loss=0.1104, over 28992.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3506, pruned_loss=0.09697, over 5687789.65 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3304, pruned_loss=0.08837, over 5729118.73 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3514, pruned_loss=0.09687, over 5680164.50 frames. ], batch size: 164, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:16:29,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=794163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:16:46,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7758, 2.2240, 1.6851, 1.8059], device='cuda:0'), covar=tensor([0.0692, 0.0242, 0.0303, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 09:17:00,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3944, 4.1991, 3.9289, 1.8488], device='cuda:0'), covar=tensor([0.0554, 0.0738, 0.0746, 0.2150], device='cuda:0'), in_proj_covar=tensor([0.1147, 0.1048, 0.0908, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 09:17:12,841 INFO [train.py:968] (0/2) Epoch 18, batch 17500, giga_loss[loss=0.2683, simple_loss=0.346, pruned_loss=0.0953, over 29024.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.35, pruned_loss=0.09676, over 5693698.51 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3301, pruned_loss=0.08801, over 5732772.33 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3515, pruned_loss=0.09724, over 5683488.21 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:17:43,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.368e+02 1.243e+03 1.596e+03 2.156e+03 4.926e+03, threshold=3.192e+03, percent-clipped=2.0
+2023-03-09 09:17:56,896 INFO [train.py:968] (0/2) Epoch 18, batch 17550, giga_loss[loss=0.2297, simple_loss=0.3102, pruned_loss=0.07462, over 28915.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3439, pruned_loss=0.09443, over 5693312.70 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3299, pruned_loss=0.08786, over 5735777.26 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3456, pruned_loss=0.0951, over 5681775.99 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:18:37,144 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=794306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:18:38,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=794309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:18:40,388 INFO [train.py:968] (0/2) Epoch 18, batch 17600, giga_loss[loss=0.2558, simple_loss=0.3292, pruned_loss=0.09117, over 28724.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.337, pruned_loss=0.09179, over 5680309.69 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3299, pruned_loss=0.0878, over 5729454.34 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3385, pruned_loss=0.09244, over 5676128.16 frames. ], batch size: 284, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:18:45,428 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:19:02,042 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=794338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:19:07,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.928e+02 1.045e+03 1.393e+03 1.692e+03 5.670e+03, threshold=2.786e+03, percent-clipped=6.0
+2023-03-09 09:19:21,784 INFO [train.py:968] (0/2) Epoch 18, batch 17650, giga_loss[loss=0.2432, simple_loss=0.2973, pruned_loss=0.09455, over 23782.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3293, pruned_loss=0.08845, over 5672324.21 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3298, pruned_loss=0.08777, over 5723566.36 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3306, pruned_loss=0.08905, over 5672863.29 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:20:04,011 INFO [train.py:968] (0/2) Epoch 18, batch 17700, giga_loss[loss=0.242, simple_loss=0.3146, pruned_loss=0.08473, over 29129.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3228, pruned_loss=0.08569, over 5684319.49 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3298, pruned_loss=0.0877, over 5727889.11 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3237, pruned_loss=0.0862, over 5679895.05 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:20:22,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=794433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:20:25,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2425, 1.3628, 1.1737, 0.9978], device='cuda:0'), covar=tensor([0.0966, 0.0539, 0.1147, 0.1165], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0436, 0.0504, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:0')
+2023-03-09 09:20:33,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.235e+02 1.024e+03 1.198e+03 1.651e+03 3.989e+03, threshold=2.396e+03, percent-clipped=4.0
+2023-03-09 09:20:44,958 INFO [train.py:968] (0/2) Epoch 18, batch 17750, giga_loss[loss=0.3078, simple_loss=0.3621, pruned_loss=0.1268, over 28234.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3188, pruned_loss=0.08418, over 5692389.17 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3306, pruned_loss=0.08803, over 5730546.76 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3186, pruned_loss=0.08419, over 5685647.74 frames. ], batch size: 77, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:21:26,648 INFO [train.py:968] (0/2) Epoch 18, batch 17800, giga_loss[loss=0.2312, simple_loss=0.3033, pruned_loss=0.07959, over 29032.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3158, pruned_loss=0.08319, over 5696966.90 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3306, pruned_loss=0.088, over 5732880.10 frames. ], giga_tot_loss[loss=0.2409, simple_loss=0.3154, pruned_loss=0.08315, over 5689172.32 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:21:31,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0345, 2.3977, 2.0322, 2.2573], device='cuda:0'), covar=tensor([0.2106, 0.2010, 0.2274, 0.1784], device='cuda:0'), in_proj_covar=tensor([0.1434, 0.1041, 0.1274, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 09:21:34,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=794523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:21:43,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2525, 1.5036, 1.3433, 1.1839], device='cuda:0'), covar=tensor([0.2709, 0.2414, 0.1480, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.1877, 0.1792, 0.1717, 0.1874], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 09:21:54,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.316e+02 1.177e+03 1.453e+03 2.053e+03 4.907e+03, threshold=2.907e+03, percent-clipped=11.0
+2023-03-09 09:22:07,324 INFO [train.py:968] (0/2) Epoch 18, batch 17850, giga_loss[loss=0.1956, simple_loss=0.2818, pruned_loss=0.05476, over 28705.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3124, pruned_loss=0.08142, over 5685052.25 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.331, pruned_loss=0.08808, over 5718512.77 frames. ], giga_tot_loss[loss=0.2367, simple_loss=0.3112, pruned_loss=0.08113, over 5690117.64 frames. ], batch size: 242, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:22:12,059 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:22:20,348 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=794576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:22:22,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=794579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:22:46,591 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=794608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:22:50,176 INFO [train.py:968] (0/2) Epoch 18, batch 17900, libri_loss[loss=0.2587, simple_loss=0.3472, pruned_loss=0.08515, over 29376.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3091, pruned_loss=0.07978, over 5690959.44 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.331, pruned_loss=0.08796, over 5721181.78 frames. ], giga_tot_loss[loss=0.2334, simple_loss=0.3078, pruned_loss=0.07955, over 5692110.25 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:23:16,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.531e+02 1.029e+03 1.249e+03 1.530e+03 4.746e+03, threshold=2.498e+03, percent-clipped=5.0
+2023-03-09 09:23:27,605 INFO [train.py:968] (0/2) Epoch 18, batch 17950, giga_loss[loss=0.2114, simple_loss=0.2846, pruned_loss=0.06912, over 29048.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3079, pruned_loss=0.07908, over 5691049.30 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3323, pruned_loss=0.08851, over 5718904.04 frames. ], giga_tot_loss[loss=0.2305, simple_loss=0.3048, pruned_loss=0.07805, over 5692910.71 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:23:31,721 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=794666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:23:33,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=794669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:23:35,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2057, 1.4539, 1.4228, 1.3054], device='cuda:0'), covar=tensor([0.1607, 0.1497, 0.2172, 0.1624], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0734, 0.0691, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 09:23:44,449 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:23:53,892 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=794692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:23:57,943 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=794698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:24:07,866 INFO [train.py:968] (0/2) Epoch 18, batch 18000, giga_loss[loss=0.2142, simple_loss=0.2891, pruned_loss=0.06961, over 28557.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3054, pruned_loss=0.0782, over 5688607.19 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.332, pruned_loss=0.08838, over 5722167.86 frames. ], giga_tot_loss[loss=0.2285, simple_loss=0.3025, pruned_loss=0.07724, over 5686237.79 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:24:07,870 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 09:24:17,119 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2056, simple_loss=0.3113, pruned_loss=0.04996, over 944034.00 frames. 
+2023-03-09 09:24:17,120 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 09:24:36,854 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794738.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:24:44,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.935e+02 1.119e+03 1.430e+03 1.843e+03 3.715e+03, threshold=2.860e+03, percent-clipped=5.0
+2023-03-09 09:24:49,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3451, 1.8610, 1.4330, 1.5732], device='cuda:0'), covar=tensor([0.0787, 0.0328, 0.0344, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 09:24:56,445 INFO [train.py:968] (0/2) Epoch 18, batch 18050, giga_loss[loss=0.2355, simple_loss=0.2967, pruned_loss=0.08718, over 27627.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3044, pruned_loss=0.07768, over 5687769.03 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3328, pruned_loss=0.08862, over 5721121.85 frames. ], giga_tot_loss[loss=0.2262, simple_loss=0.3, pruned_loss=0.07616, over 5685846.21 frames. ], batch size: 472, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:25:36,206 INFO [train.py:968] (0/2) Epoch 18, batch 18100, giga_loss[loss=0.2145, simple_loss=0.2918, pruned_loss=0.06859, over 27856.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3016, pruned_loss=0.07619, over 5693347.06 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3338, pruned_loss=0.08897, over 5725962.26 frames. ], giga_tot_loss[loss=0.2224, simple_loss=0.2963, pruned_loss=0.0743, over 5686542.65 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:25:58,049 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=794835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:26:01,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=794838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:26:08,853 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.096e+02 9.980e+02 1.330e+03 1.824e+03 4.503e+03, threshold=2.659e+03, percent-clipped=7.0
+2023-03-09 09:26:20,164 INFO [train.py:968] (0/2) Epoch 18, batch 18150, giga_loss[loss=0.2028, simple_loss=0.2862, pruned_loss=0.05968, over 29103.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.2986, pruned_loss=0.07483, over 5677859.75 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3341, pruned_loss=0.08903, over 5722846.15 frames. ], giga_tot_loss[loss=0.2193, simple_loss=0.293, pruned_loss=0.07282, over 5673242.66 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:26:25,679 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=794867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:26:44,733 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:27:02,708 INFO [train.py:968] (0/2) Epoch 18, batch 18200, giga_loss[loss=0.2638, simple_loss=0.3185, pruned_loss=0.1046, over 23657.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.2973, pruned_loss=0.07463, over 5676764.75 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3342, pruned_loss=0.089, over 5723112.99 frames. ], giga_tot_loss[loss=0.2188, simple_loss=0.2921, pruned_loss=0.0728, over 5672038.85 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:27:34,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=794941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:27:39,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.396e+02 1.084e+03 1.458e+03 1.778e+03 6.214e+03, threshold=2.915e+03, percent-clipped=10.0
+2023-03-09 09:27:52,587 INFO [train.py:968] (0/2) Epoch 18, batch 18250, giga_loss[loss=0.3085, simple_loss=0.3801, pruned_loss=0.1185, over 28858.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3085, pruned_loss=0.08074, over 5668789.19 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3342, pruned_loss=0.08901, over 5715667.46 frames. ], giga_tot_loss[loss=0.231, simple_loss=0.3038, pruned_loss=0.07909, over 5671484.03 frames. ], batch size: 186, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:28:30,459 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-09 09:28:35,844 INFO [train.py:968] (0/2) Epoch 18, batch 18300, giga_loss[loss=0.3261, simple_loss=0.3895, pruned_loss=0.1314, over 28344.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3219, pruned_loss=0.08752, over 5677671.02 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3343, pruned_loss=0.08903, over 5718444.47 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3178, pruned_loss=0.0861, over 5676707.44 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:29:05,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.513e+02 1.410e+03 1.670e+03 2.236e+03 8.189e+03, threshold=3.340e+03, percent-clipped=7.0
+2023-03-09 09:29:14,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:29:18,310 INFO [train.py:968] (0/2) Epoch 18, batch 18350, giga_loss[loss=0.2761, simple_loss=0.3508, pruned_loss=0.1007, over 28566.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3318, pruned_loss=0.09236, over 5680488.63 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3343, pruned_loss=0.08903, over 5718444.47 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3286, pruned_loss=0.09126, over 5679738.67 frames. ], batch size: 60, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:29:36,066 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:29:38,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:29:57,675 INFO [train.py:968] (0/2) Epoch 18, batch 18400, libri_loss[loss=0.2393, simple_loss=0.3287, pruned_loss=0.07497, over 29519.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3384, pruned_loss=0.09511, over 5675778.63 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3354, pruned_loss=0.08964, over 5712592.62 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3349, pruned_loss=0.0938, over 5679279.03 frames. ], batch size: 81, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:29:58,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795113.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:30:01,233 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:30:07,316 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 09:30:26,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.485e+02 1.328e+03 1.585e+03 2.323e+03 7.488e+03, threshold=3.170e+03, percent-clipped=4.0
+2023-03-09 09:30:38,686 INFO [train.py:968] (0/2) Epoch 18, batch 18450, giga_loss[loss=0.2319, simple_loss=0.311, pruned_loss=0.07636, over 23484.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3404, pruned_loss=0.09487, over 5674275.85 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3356, pruned_loss=0.0897, over 5715648.45 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3375, pruned_loss=0.09388, over 5673852.65 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:31:12,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2885, 2.8381, 1.3693, 1.4656], device='cuda:0'), covar=tensor([0.1023, 0.0326, 0.0897, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0532, 0.0368, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 09:31:13,295 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:31:16,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:31:24,281 INFO [train.py:968] (0/2) Epoch 18, batch 18500, giga_loss[loss=0.2865, simple_loss=0.3588, pruned_loss=0.1071, over 28911.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3419, pruned_loss=0.09534, over 5660873.78 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3355, pruned_loss=0.0897, over 5708255.94 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3398, pruned_loss=0.09464, over 5666514.84 frames. ], batch size: 186, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:31:37,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5006, 1.7723, 1.4296, 1.5784], device='cuda:0'), covar=tensor([0.2530, 0.2493, 0.2849, 0.2137], device='cuda:0'), in_proj_covar=tensor([0.1435, 0.1044, 0.1277, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 09:31:40,850 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:31:49,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4127, 1.7333, 1.4948, 1.6274], device='cuda:0'), covar=tensor([0.0808, 0.0317, 0.0323, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 09:31:55,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.344e+02 1.142e+03 1.352e+03 1.844e+03 4.189e+03, threshold=2.704e+03, percent-clipped=7.0
+2023-03-09 09:32:02,910 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795256.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:32:04,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795259.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:32:07,148 INFO [train.py:968] (0/2) Epoch 18, batch 18550, giga_loss[loss=0.2762, simple_loss=0.3528, pruned_loss=0.09982, over 28971.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3445, pruned_loss=0.09776, over 5665121.16 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3357, pruned_loss=0.08978, over 5707956.78 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3427, pruned_loss=0.09721, over 5669391.45 frames. ], batch size: 164, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:32:11,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:32:28,229 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795288.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:32:50,347 INFO [train.py:968] (0/2) Epoch 18, batch 18600, giga_loss[loss=0.3281, simple_loss=0.381, pruned_loss=0.1376, over 28748.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3477, pruned_loss=0.09985, over 5664183.20 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3359, pruned_loss=0.08986, over 5703195.55 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3463, pruned_loss=0.09952, over 5670966.74 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:32:50,837 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 09:32:51,335 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:32:55,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2467, 1.4718, 1.4946, 1.2993], device='cuda:0'), covar=tensor([0.1873, 0.1793, 0.2394, 0.1923], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0733, 0.0690, 0.0666], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 09:33:20,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.403e+02 1.170e+03 1.467e+03 1.785e+03 4.490e+03, threshold=2.934e+03, percent-clipped=5.0
+2023-03-09 09:33:30,957 INFO [train.py:968] (0/2) Epoch 18, batch 18650, giga_loss[loss=0.2779, simple_loss=0.3573, pruned_loss=0.09929, over 28624.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3499, pruned_loss=0.1005, over 5674300.19 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3357, pruned_loss=0.08957, over 5708502.23 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3493, pruned_loss=0.1008, over 5673861.01 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:33:38,524 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:33:54,668 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795389.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:33:59,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7904, 2.6752, 1.7253, 1.1076], device='cuda:0'), covar=tensor([0.7388, 0.3410, 0.3857, 0.5746], device='cuda:0'), in_proj_covar=tensor([0.1688, 0.1595, 0.1571, 0.1378], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 09:34:11,336 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:34:12,446 INFO [train.py:968] (0/2) Epoch 18, batch 18700, giga_loss[loss=0.2923, simple_loss=0.3686, pruned_loss=0.108, over 27896.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3518, pruned_loss=0.1006, over 5677351.51 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3357, pruned_loss=0.0895, over 5710499.27 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3516, pruned_loss=0.101, over 5674834.30 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:34:14,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:34:36,410 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:34:40,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.577e+02 1.269e+03 1.710e+03 2.295e+03 5.604e+03, threshold=3.419e+03, percent-clipped=12.0
+2023-03-09 09:34:51,496 INFO [train.py:968] (0/2) Epoch 18, batch 18750, giga_loss[loss=0.2748, simple_loss=0.3554, pruned_loss=0.0971, over 28941.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.353, pruned_loss=0.1004, over 5688981.33 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3358, pruned_loss=0.08946, over 5717729.31 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3534, pruned_loss=0.1012, over 5678954.98 frames. ], batch size: 112, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:35:09,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5483, 1.8108, 1.4881, 2.0145], device='cuda:0'), covar=tensor([0.2570, 0.2674, 0.2922, 0.2161], device='cuda:0'), in_proj_covar=tensor([0.1436, 0.1045, 0.1276, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 09:35:30,020 INFO [train.py:968] (0/2) Epoch 18, batch 18800, giga_loss[loss=0.2455, simple_loss=0.3345, pruned_loss=0.0783, over 28709.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3537, pruned_loss=0.09973, over 5684754.47 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3362, pruned_loss=0.08942, over 5712882.76 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3544, pruned_loss=0.1009, over 5679217.09 frames. ], batch size: 242, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:35:30,886 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:35:44,316 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:35:59,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.582e+02 1.074e+03 1.378e+03 1.836e+03 4.280e+03, threshold=2.756e+03, percent-clipped=1.0
+2023-03-09 09:36:07,363 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8858, 1.1241, 3.2970, 2.7902], device='cuda:0'), covar=tensor([0.2391, 0.3318, 0.0935, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0721, 0.0626, 0.0916, 0.0851], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 09:36:10,964 INFO [train.py:968] (0/2) Epoch 18, batch 18850, giga_loss[loss=0.2689, simple_loss=0.3505, pruned_loss=0.09368, over 28650.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3532, pruned_loss=0.09782, over 5699495.09 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3362, pruned_loss=0.08942, over 5712882.76 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3537, pruned_loss=0.09872, over 5695185.28 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:36:17,738 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:36:29,151 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-09 09:36:32,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:36:36,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3248, 1.2429, 1.2373, 1.5475], device='cuda:0'), covar=tensor([0.0816, 0.0360, 0.0347, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 09:36:49,912 INFO [train.py:968] (0/2) Epoch 18, batch 18900, giga_loss[loss=0.263, simple_loss=0.3473, pruned_loss=0.08934, over 28785.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3519, pruned_loss=0.09632, over 5703243.62 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3362, pruned_loss=0.08925, over 5712976.55 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3526, pruned_loss=0.09733, over 5699690.31 frames. ], batch size: 99, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:37:03,081 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795627.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:37:19,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.218e+02 1.112e+03 1.404e+03 1.721e+03 4.653e+03, threshold=2.809e+03, percent-clipped=2.0
+2023-03-09 09:37:28,480 INFO [train.py:968] (0/2) Epoch 18, batch 18950, giga_loss[loss=0.306, simple_loss=0.3833, pruned_loss=0.1143, over 29011.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3544, pruned_loss=0.09831, over 5708417.40 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3367, pruned_loss=0.08944, over 5717137.70 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3549, pruned_loss=0.0991, over 5701427.20 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:37:49,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5322, 1.7018, 1.6348, 1.5402], device='cuda:0'), covar=tensor([0.1677, 0.1715, 0.2226, 0.1806], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0737, 0.0695, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 09:37:51,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:38:13,594 INFO [train.py:968] (0/2) Epoch 18, batch 19000, giga_loss[loss=0.2787, simple_loss=0.342, pruned_loss=0.1077, over 28657.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3575, pruned_loss=0.1034, over 5704555.27 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3367, pruned_loss=0.0895, over 5708755.01 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3579, pruned_loss=0.104, over 5705644.89 frames. ], batch size: 85, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:38:41,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:38:43,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4941, 3.4717, 1.5692, 1.6670], device='cuda:0'), covar=tensor([0.1029, 0.0265, 0.0891, 0.1337], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0535, 0.0369, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 09:38:44,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.339e+02 1.452e+03 1.825e+03 2.590e+03 6.789e+03, threshold=3.649e+03, percent-clipped=20.0
+2023-03-09 09:38:54,413 INFO [train.py:968] (0/2) Epoch 18, batch 19050, libri_loss[loss=0.2161, simple_loss=0.2989, pruned_loss=0.0666, over 29632.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3585, pruned_loss=0.1063, over 5705579.80 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3364, pruned_loss=0.08918, over 5714710.54 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3597, pruned_loss=0.1075, over 5701157.00 frames. ], batch size: 69, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:38:55,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795764.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:39:32,736 INFO [train.py:968] (0/2) Epoch 18, batch 19100, giga_loss[loss=0.2489, simple_loss=0.3282, pruned_loss=0.08475, over 28675.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3569, pruned_loss=0.106, over 5702598.50 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3367, pruned_loss=0.08927, over 5717083.67 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3582, pruned_loss=0.1074, over 5696520.85 frames. ], batch size: 262, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:39:47,112 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:39:50,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:05,687 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.819e+02 1.160e+03 1.454e+03 1.842e+03 5.869e+03, threshold=2.907e+03, percent-clipped=1.0
+2023-03-09 09:40:15,410 INFO [train.py:968] (0/2) Epoch 18, batch 19150, giga_loss[loss=0.2586, simple_loss=0.3373, pruned_loss=0.08989, over 28917.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3537, pruned_loss=0.1047, over 5700886.08 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3369, pruned_loss=0.08928, over 5719899.23 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3548, pruned_loss=0.1059, over 5693316.46 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:40:16,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:38,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:38,742 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:43,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:55,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:57,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795907.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:40:59,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795910.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:41:00,954 INFO [train.py:968] (0/2) Epoch 18, batch 19200, giga_loss[loss=0.2953, simple_loss=0.3655, pruned_loss=0.1125, over 28502.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3521, pruned_loss=0.103, over 5711807.22 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.337, pruned_loss=0.08917, over 5720825.55 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3532, pruned_loss=0.1044, over 5704795.22 frames. ], batch size: 65, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:41:07,778 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:41:21,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795939.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:41:27,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:41:32,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.689e+02 1.167e+03 1.377e+03 1.787e+03 3.239e+03, threshold=2.754e+03, percent-clipped=3.0
+2023-03-09 09:41:40,736 INFO [train.py:968] (0/2) Epoch 18, batch 19250, giga_loss[loss=0.246, simple_loss=0.3292, pruned_loss=0.08135, over 28907.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3518, pruned_loss=0.1021, over 5717423.55 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3376, pruned_loss=0.0894, over 5724691.75 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3524, pruned_loss=0.1033, over 5708330.92 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:41:41,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:42:13,929 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-796000.pt
+2023-03-09 09:42:16,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=796002.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:42:17,512 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 09:42:22,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-09 09:42:23,883 INFO [train.py:968] (0/2) Epoch 18, batch 19300, giga_loss[loss=0.2375, simple_loss=0.3189, pruned_loss=0.078, over 28834.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3491, pruned_loss=0.1003, over 5703073.08 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.338, pruned_loss=0.08941, over 5730466.77 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3497, pruned_loss=0.1017, over 5690137.96 frames. ], batch size: 243, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:42:42,940 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:42:45,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:42:54,881 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:42:55,717 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-09 09:42:57,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:42:57,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.176e+02 1.076e+03 1.514e+03 2.175e+03 6.476e+03, threshold=3.028e+03, percent-clipped=13.0
+2023-03-09 09:43:04,476 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-09 09:43:09,368 INFO [train.py:968] (0/2) Epoch 18, batch 19350, giga_loss[loss=0.2729, simple_loss=0.336, pruned_loss=0.1049, over 27997.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3429, pruned_loss=0.09703, over 5702011.06 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3378, pruned_loss=0.08924, over 5734946.47 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3438, pruned_loss=0.0985, over 5686988.27 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:43:10,222 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:24,506 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:32,556 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:35,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:47,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:49,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:51,232 INFO [train.py:968] (0/2) Epoch 18, batch 19400, giga_loss[loss=0.2528, simple_loss=0.3221, pruned_loss=0.0918, over 28578.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3379, pruned_loss=0.09458, over 5693374.03 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3378, pruned_loss=0.08919, over 5735719.72 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3386, pruned_loss=0.09595, over 5678948.37 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:44:00,528 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:44:06,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3939, 2.3822, 2.4456, 2.2414], device='cuda:0'), covar=tensor([0.2011, 0.2596, 0.2047, 0.2110], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0741, 0.0698, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 09:44:07,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-09 09:44:18,899 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:44:24,766 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796145.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:44:27,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796148.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:44:30,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.337e+02 9.990e+02 1.283e+03 1.726e+03 7.239e+03, threshold=2.565e+03, percent-clipped=4.0
+2023-03-09 09:44:42,135 INFO [train.py:968] (0/2) Epoch 18, batch 19450, giga_loss[loss=0.2465, simple_loss=0.3265, pruned_loss=0.08326, over 28882.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3331, pruned_loss=0.09252, over 5668307.83 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3376, pruned_loss=0.08904, over 5736534.12 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3338, pruned_loss=0.09375, over 5656015.77 frames. ], batch size: 213, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:44:56,698 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796177.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:45:15,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:45:24,542 INFO [train.py:968] (0/2) Epoch 18, batch 19500, giga_loss[loss=0.2545, simple_loss=0.3351, pruned_loss=0.08697, over 28986.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3328, pruned_loss=0.09214, over 5667007.68 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3376, pruned_loss=0.08908, over 5737181.72 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3333, pruned_loss=0.09314, over 5655093.94 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:45:58,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.989e+02 1.004e+03 1.247e+03 1.682e+03 4.847e+03, threshold=2.493e+03, percent-clipped=8.0
+2023-03-09 09:46:06,362 INFO [train.py:968] (0/2) Epoch 18, batch 19550, giga_loss[loss=0.2296, simple_loss=0.3094, pruned_loss=0.07489, over 28944.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3344, pruned_loss=0.09264, over 5657368.24 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3391, pruned_loss=0.08979, over 5726327.24 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3334, pruned_loss=0.09289, over 5655823.94 frames. ], batch size: 213, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:46:45,424 INFO [train.py:968] (0/2) Epoch 18, batch 19600, giga_loss[loss=0.2275, simple_loss=0.3125, pruned_loss=0.07129, over 28927.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3327, pruned_loss=0.09155, over 5665716.82 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3394, pruned_loss=0.0899, over 5720388.19 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3314, pruned_loss=0.09167, over 5668886.14 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:46:59,523 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-09 09:47:18,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.127e+02 1.024e+03 1.241e+03 1.745e+03 6.054e+03, threshold=2.481e+03, percent-clipped=9.0
+2023-03-09 09:47:19,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8995, 2.9552, 2.0084, 0.8414], device='cuda:0'), covar=tensor([0.7476, 0.2610, 0.3753, 0.7119], device='cuda:0'), in_proj_covar=tensor([0.1675, 0.1575, 0.1557, 0.1372], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 09:47:20,832 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:47:27,216 INFO [train.py:968] (0/2) Epoch 18, batch 19650, giga_loss[loss=0.2556, simple_loss=0.3223, pruned_loss=0.09446, over 28086.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3305, pruned_loss=0.09071, over 5672467.90 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3395, pruned_loss=0.08983, over 5722370.63 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3294, pruned_loss=0.09088, over 5672609.91 frames. ], batch size: 77, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:47:44,071 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:48:06,213 INFO [train.py:968] (0/2) Epoch 18, batch 19700, giga_loss[loss=0.2089, simple_loss=0.2846, pruned_loss=0.06661, over 28569.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3282, pruned_loss=0.08934, over 5688287.75 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3397, pruned_loss=0.08979, over 5726771.71 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3268, pruned_loss=0.08952, over 5683396.54 frames. ], batch size: 60, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:48:12,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4239, 1.7202, 1.7254, 1.2636], device='cuda:0'), covar=tensor([0.1814, 0.2427, 0.1423, 0.1616], device='cuda:0'), in_proj_covar=tensor([0.0875, 0.0691, 0.0921, 0.0821], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 09:48:13,556 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-09 09:48:36,206 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.921e+02 1.115e+03 1.557e+03 2.164e+03 3.497e+03, threshold=3.114e+03, percent-clipped=19.0
+2023-03-09 09:48:45,770 INFO [train.py:968] (0/2) Epoch 18, batch 19750, giga_loss[loss=0.2543, simple_loss=0.3327, pruned_loss=0.08801, over 28251.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3257, pruned_loss=0.0877, over 5694614.78 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3398, pruned_loss=0.08984, over 5730997.06 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3243, pruned_loss=0.08779, over 5686504.20 frames. ], batch size: 77, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:49:25,620 INFO [train.py:968] (0/2) Epoch 18, batch 19800, giga_loss[loss=0.211, simple_loss=0.2898, pruned_loss=0.06611, over 28832.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3236, pruned_loss=0.08689, over 5705712.93 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3407, pruned_loss=0.09014, over 5734640.16 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3214, pruned_loss=0.08667, over 5695653.83 frames. ], batch size: 112, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:49:54,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.574e+02 1.020e+03 1.254e+03 1.708e+03 3.652e+03, threshold=2.508e+03, percent-clipped=3.0
+2023-03-09 09:50:02,116 INFO [train.py:968] (0/2) Epoch 18, batch 19850, giga_loss[loss=0.2194, simple_loss=0.2956, pruned_loss=0.07157, over 28635.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3219, pruned_loss=0.08616, over 5713606.84 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3416, pruned_loss=0.09043, over 5734767.21 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3189, pruned_loss=0.08562, over 5704453.00 frames. ], batch size: 60, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:50:14,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=796575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:50:19,604 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-09 09:50:25,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1826, 1.7758, 1.3847, 0.3930], device='cuda:0'), covar=tensor([0.4319, 0.2420, 0.4297, 0.5692], device='cuda:0'), in_proj_covar=tensor([0.1671, 0.1573, 0.1558, 0.1371], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 09:50:43,459 INFO [train.py:968] (0/2) Epoch 18, batch 19900, libri_loss[loss=0.2888, simple_loss=0.3812, pruned_loss=0.09823, over 27804.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.321, pruned_loss=0.08575, over 5717489.07 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3426, pruned_loss=0.0908, over 5736772.65 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3172, pruned_loss=0.08486, over 5707913.87 frames. ], batch size: 116, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:51:12,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.675e+02 1.059e+03 1.323e+03 1.772e+03 1.212e+04, threshold=2.646e+03, percent-clipped=11.0
+2023-03-09 09:51:19,444 INFO [train.py:968] (0/2) Epoch 18, batch 19950, giga_loss[loss=0.2227, simple_loss=0.2958, pruned_loss=0.0748, over 28817.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3208, pruned_loss=0.08518, over 5719021.54 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3435, pruned_loss=0.09107, over 5739187.96 frames. ], giga_tot_loss[loss=0.2418, simple_loss=0.3157, pruned_loss=0.08393, over 5708139.95 frames. ], batch size: 99, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:51:56,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3315, 3.1512, 2.9924, 1.3257], device='cuda:0'), covar=tensor([0.0926, 0.1046, 0.0878, 0.2449], device='cuda:0'), in_proj_covar=tensor([0.1141, 0.1054, 0.0909, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 09:51:56,642 INFO [train.py:968] (0/2) Epoch 18, batch 20000, giga_loss[loss=0.248, simple_loss=0.3125, pruned_loss=0.09173, over 29002.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3197, pruned_loss=0.08434, over 5723928.15 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.344, pruned_loss=0.09118, over 5744026.28 frames. ], giga_tot_loss[loss=0.2403, simple_loss=0.3146, pruned_loss=0.08305, over 5710260.67 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:51:57,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4229, 1.6327, 1.7361, 1.2838], device='cuda:0'), covar=tensor([0.1820, 0.2400, 0.1433, 0.1644], device='cuda:0'), in_proj_covar=tensor([0.0880, 0.0694, 0.0927, 0.0825], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 09:52:01,096 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:04,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:06,303 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:10,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=796729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:26,239 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:27,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.694e+02 1.001e+03 1.219e+03 1.687e+03 7.420e+03, threshold=2.439e+03, percent-clipped=9.0
+2023-03-09 09:52:34,150 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-09 09:52:35,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=796759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:37,196 INFO [train.py:968] (0/2) Epoch 18, batch 20050, giga_loss[loss=0.2504, simple_loss=0.3313, pruned_loss=0.08473, over 28851.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3196, pruned_loss=0.08419, over 5725057.99 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3441, pruned_loss=0.0912, over 5744690.15 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3154, pruned_loss=0.08314, over 5713737.37 frames. ], batch size: 174, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:52:49,534 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2803, 1.4619, 1.5916, 1.2865], device='cuda:0'), covar=tensor([0.1934, 0.1785, 0.2359, 0.1986], device='cuda:0'), in_proj_covar=tensor([0.0463, 0.0745, 0.0704, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 09:53:16,721 INFO [train.py:968] (0/2) Epoch 18, batch 20100, giga_loss[loss=0.2851, simple_loss=0.3536, pruned_loss=0.1083, over 28736.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3261, pruned_loss=0.08884, over 5723326.34 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3443, pruned_loss=0.09135, over 5748296.60 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3219, pruned_loss=0.08774, over 5710557.30 frames. ], batch size: 242, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:53:29,642 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:53:50,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2523, 3.1498, 1.5334, 1.3705], device='cuda:0'), covar=tensor([0.0994, 0.0330, 0.0878, 0.1453], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0535, 0.0369, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 09:53:55,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.778e+02 1.373e+03 1.817e+03 2.629e+03 5.927e+03, threshold=3.634e+03, percent-clipped=28.0
+2023-03-09 09:54:04,841 INFO [train.py:968] (0/2) Epoch 18, batch 20150, libri_loss[loss=0.2755, simple_loss=0.3578, pruned_loss=0.09655, over 29524.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3316, pruned_loss=0.09265, over 5707507.41 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3445, pruned_loss=0.09128, over 5752463.47 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3278, pruned_loss=0.09181, over 5692805.60 frames. ], batch size: 84, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:54:09,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5412, 2.1786, 1.6269, 0.7749], device='cuda:0'), covar=tensor([0.5783, 0.3032, 0.3732, 0.5802], device='cuda:0'), in_proj_covar=tensor([0.1675, 0.1581, 0.1564, 0.1374], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 09:54:12,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:17,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:44,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:46,672 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:47,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:48,708 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:52,748 INFO [train.py:968] (0/2) Epoch 18, batch 20200, giga_loss[loss=0.3135, simple_loss=0.3761, pruned_loss=0.1254, over 27959.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3394, pruned_loss=0.09761, over 5706530.06 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3444, pruned_loss=0.09123, over 5755664.42 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3362, pruned_loss=0.09709, over 5691071.52 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:55:12,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:55:28,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.172e+02 1.220e+03 1.557e+03 2.024e+03 4.660e+03, threshold=3.114e+03, percent-clipped=2.0
+2023-03-09 09:55:38,464 INFO [train.py:968] (0/2) Epoch 18, batch 20250, libri_loss[loss=0.2387, simple_loss=0.3248, pruned_loss=0.0763, over 29661.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3439, pruned_loss=0.09927, over 5698594.30 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3447, pruned_loss=0.09131, over 5756159.06 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3411, pruned_loss=0.09896, over 5684469.34 frames. ], batch size: 73, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:56:18,482 INFO [train.py:968] (0/2) Epoch 18, batch 20300, giga_loss[loss=0.2843, simple_loss=0.3635, pruned_loss=0.1026, over 28988.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3484, pruned_loss=0.1008, over 5704195.02 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3449, pruned_loss=0.09128, over 5761707.23 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3461, pruned_loss=0.1009, over 5685687.14 frames. ], batch size: 164, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:56:54,761 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.489e+02 1.196e+03 1.540e+03 2.221e+03 5.244e+03, threshold=3.081e+03, percent-clipped=7.0
+2023-03-09 09:57:03,413 INFO [train.py:968] (0/2) Epoch 18, batch 20350, giga_loss[loss=0.3355, simple_loss=0.4024, pruned_loss=0.1343, over 28649.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3548, pruned_loss=0.1045, over 5704355.71 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3452, pruned_loss=0.09138, over 5764281.80 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3528, pruned_loss=0.1047, over 5686399.84 frames. ], batch size: 262, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:57:34,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:57:42,524 INFO [train.py:968] (0/2) Epoch 18, batch 20400, giga_loss[loss=0.2563, simple_loss=0.3325, pruned_loss=0.08999, over 28609.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3558, pruned_loss=0.1046, over 5708475.58 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3452, pruned_loss=0.09139, over 5766947.63 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3544, pruned_loss=0.105, over 5690733.62 frames. ], batch size: 85, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:58:18,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.549e+02 1.221e+03 1.505e+03 1.967e+03 4.015e+03, threshold=3.010e+03, percent-clipped=6.0
+2023-03-09 09:58:26,743 INFO [train.py:968] (0/2) Epoch 18, batch 20450, giga_loss[loss=0.2245, simple_loss=0.322, pruned_loss=0.06353, over 28964.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3506, pruned_loss=0.1008, over 5708125.37 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.345, pruned_loss=0.09143, over 5769999.82 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3498, pruned_loss=0.1013, over 5690114.88 frames. ], batch size: 164, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:58:41,291 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-09 09:58:59,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:59:04,529 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 09:59:08,120 INFO [train.py:968] (0/2) Epoch 18, batch 20500, giga_loss[loss=0.2403, simple_loss=0.3207, pruned_loss=0.07998, over 28896.00 frames. ], tot_loss[loss=0.274, simple_loss=0.349, pruned_loss=0.09945, over 5707827.13 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3448, pruned_loss=0.09141, over 5773059.45 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3487, pruned_loss=0.1, over 5689227.53 frames. ], batch size: 112, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:59:34,251 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=797242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:59:38,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=797245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:59:43,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.434e+02 1.149e+03 1.412e+03 2.105e+03 5.374e+03, threshold=2.824e+03, percent-clipped=8.0
+2023-03-09 09:59:47,718 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-09 09:59:50,865 INFO [train.py:968] (0/2) Epoch 18, batch 20550, giga_loss[loss=0.3062, simple_loss=0.3777, pruned_loss=0.1173, over 29032.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3485, pruned_loss=0.09861, over 5707597.02 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3447, pruned_loss=0.09147, over 5773993.99 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3483, pruned_loss=0.09916, over 5690909.14 frames. ], batch size: 128, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:00:00,319 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=797274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:00:07,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:00:10,447 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=797287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:00:31,321 INFO [train.py:968] (0/2) Epoch 18, batch 20600, giga_loss[loss=0.2636, simple_loss=0.3479, pruned_loss=0.0896, over 28919.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3505, pruned_loss=0.09962, over 5704022.49 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3447, pruned_loss=0.09145, over 5775369.63 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3504, pruned_loss=0.1003, over 5687541.04 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:00:59,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=797344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:01:01,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=797347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:01:05,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.939e+02 1.396e+03 2.045e+03 3.034e+03 5.698e+03, threshold=4.090e+03, percent-clipped=27.0
+2023-03-09 10:01:16,228 INFO [train.py:968] (0/2) Epoch 18, batch 20650, giga_loss[loss=0.2752, simple_loss=0.3519, pruned_loss=0.09919, over 28558.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3543, pruned_loss=0.1027, over 5697148.82 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3452, pruned_loss=0.0917, over 5774300.96 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3541, pruned_loss=0.1033, over 5682602.09 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:01:27,053 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=797376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:01:53,205 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-09 10:01:59,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2861, 1.8762, 1.4027, 0.5168], device='cuda:0'), covar=tensor([0.4465, 0.2332, 0.3655, 0.5454], device='cuda:0'), in_proj_covar=tensor([0.1673, 0.1580, 0.1564, 0.1375], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 10:01:59,667 INFO [train.py:968] (0/2) Epoch 18, batch 20700, giga_loss[loss=0.2762, simple_loss=0.3508, pruned_loss=0.1008, over 28616.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3559, pruned_loss=0.1039, over 5707900.90 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3456, pruned_loss=0.09184, over 5775596.20 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3555, pruned_loss=0.1044, over 5694334.34 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:02:09,208 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=797425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:02:13,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=797428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:02:33,295 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-09 10:02:37,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.677e+02 1.250e+03 1.571e+03 2.060e+03 5.024e+03, threshold=3.143e+03, percent-clipped=1.0
+2023-03-09 10:02:39,941 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=797457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:02:40,219 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-09 10:02:43,303 INFO [train.py:968] (0/2) Epoch 18, batch 20750, giga_loss[loss=0.3335, simple_loss=0.3932, pruned_loss=0.1369, over 28873.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.356, pruned_loss=0.1042, over 5709791.31 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3455, pruned_loss=0.09173, over 5769503.08 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.356, pruned_loss=0.105, over 5703820.22 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:03:25,477 INFO [train.py:968] (0/2) Epoch 18, batch 20800, giga_loss[loss=0.256, simple_loss=0.3411, pruned_loss=0.08541, over 28610.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3566, pruned_loss=0.1053, over 5706362.63 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3454, pruned_loss=0.09163, over 5770604.55 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3568, pruned_loss=0.1061, over 5700222.14 frames. ], batch size: 60, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:03:56,619 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.543e+02 1.125e+03 1.326e+03 1.823e+03 4.718e+03, threshold=2.653e+03, percent-clipped=3.0
+2023-03-09 10:03:59,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=797558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:04:02,180 INFO [train.py:968] (0/2) Epoch 18, batch 20850, giga_loss[loss=0.3195, simple_loss=0.3891, pruned_loss=0.1249, over 28786.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.357, pruned_loss=0.1049, over 5712434.00 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3455, pruned_loss=0.09155, over 5774067.45 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3573, pruned_loss=0.106, over 5702723.14 frames. ], batch size: 99, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:04:21,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7074, 1.8365, 1.3267, 1.4075], device='cuda:0'), covar=tensor([0.0944, 0.0637, 0.1050, 0.1127], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0440, 0.0509, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 10:04:27,807 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=797595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:04:40,804 INFO [train.py:968] (0/2) Epoch 18, batch 20900, giga_loss[loss=0.3108, simple_loss=0.3793, pruned_loss=0.1211, over 27959.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3561, pruned_loss=0.1032, over 5717740.67 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3452, pruned_loss=0.09129, over 5779678.29 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.357, pruned_loss=0.1047, over 5702542.34 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:05:13,897 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.943e+02 1.089e+03 1.327e+03 1.816e+03 5.213e+03, threshold=2.653e+03, percent-clipped=8.0
+2023-03-09 10:05:16,070 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=797656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:05:19,975 INFO [train.py:968] (0/2) Epoch 18, batch 20950, giga_loss[loss=0.2606, simple_loss=0.3431, pruned_loss=0.08903, over 28714.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3562, pruned_loss=0.1021, over 5725320.03 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.345, pruned_loss=0.09115, over 5782201.93 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3573, pruned_loss=0.1037, over 5709998.52 frames. ], batch size: 262, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:05:20,200 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:05:57,800 INFO [train.py:968] (0/2) Epoch 18, batch 21000, giga_loss[loss=0.2409, simple_loss=0.3214, pruned_loss=0.08016, over 29038.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3548, pruned_loss=0.1015, over 5720120.71 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3452, pruned_loss=0.09129, over 5771837.25 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3557, pruned_loss=0.1028, over 5715359.89 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:05:57,805 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 10:06:06,440 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2114, simple_loss=0.318, pruned_loss=0.05243, over 944034.00 frames. 
+2023-03-09 10:06:06,441 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 10:06:38,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.779e+02 1.121e+03 1.362e+03 2.042e+03 4.605e+03, threshold=2.725e+03, percent-clipped=11.0
+2023-03-09 10:06:41,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2991, 1.5143, 1.5521, 1.3567], device='cuda:0'), covar=tensor([0.1929, 0.1715, 0.2282, 0.1823], device='cuda:0'), in_proj_covar=tensor([0.0459, 0.0739, 0.0698, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 10:06:44,033 INFO [train.py:968] (0/2) Epoch 18, batch 21050, giga_loss[loss=0.2595, simple_loss=0.3383, pruned_loss=0.09032, over 28919.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3525, pruned_loss=0.1007, over 5712294.78 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3451, pruned_loss=0.09123, over 5770690.76 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3535, pruned_loss=0.1021, over 5708364.47 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:06:51,409 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4177, 1.7651, 1.6852, 1.3069], device='cuda:0'), covar=tensor([0.2933, 0.2166, 0.2317, 0.2767], device='cuda:0'), in_proj_covar=tensor([0.1874, 0.1795, 0.1733, 0.1887], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 10:07:12,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3907, 1.5651, 1.6177, 1.4578], device='cuda:0'), covar=tensor([0.1657, 0.1628, 0.1796, 0.1705], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0737, 0.0696, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 10:07:17,732 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=797805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:07:20,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=797808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:07:23,133 INFO [train.py:968] (0/2) Epoch 18, batch 21100, giga_loss[loss=0.269, simple_loss=0.3485, pruned_loss=0.09479, over 28715.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3508, pruned_loss=0.0999, over 5709552.24 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3452, pruned_loss=0.09124, over 5770879.73 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3517, pruned_loss=0.1012, over 5705075.28 frames. ], batch size: 242, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:07:29,014 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2621, 0.7676, 0.9054, 1.4170], device='cuda:0'), covar=tensor([0.0784, 0.0370, 0.0351, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0115, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0059, 0.0102], device='cuda:0')
+2023-03-09 10:07:41,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=797837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:07:56,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.502e+02 1.074e+03 1.361e+03 2.101e+03 1.125e+04, threshold=2.722e+03, percent-clipped=19.0
+2023-03-09 10:08:03,644 INFO [train.py:968] (0/2) Epoch 18, batch 21150, giga_loss[loss=0.3052, simple_loss=0.375, pruned_loss=0.1177, over 28769.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3504, pruned_loss=0.09997, over 5710476.08 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3455, pruned_loss=0.09156, over 5762733.95 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3509, pruned_loss=0.1008, over 5712647.23 frames. ], batch size: 284, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:08:04,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 10:08:46,847 INFO [train.py:968] (0/2) Epoch 18, batch 21200, giga_loss[loss=0.2495, simple_loss=0.3309, pruned_loss=0.08409, over 28388.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3518, pruned_loss=0.1014, over 5697580.81 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3457, pruned_loss=0.09163, over 5759967.28 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3521, pruned_loss=0.1021, over 5701302.94 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:08:56,919 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.39 vs. limit=5.0
+2023-03-09 10:09:01,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5553, 1.6417, 1.6507, 1.4592], device='cuda:0'), covar=tensor([0.2690, 0.2469, 0.1992, 0.2491], device='cuda:0'), in_proj_covar=tensor([0.1886, 0.1806, 0.1744, 0.1896], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 10:09:01,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:09:19,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.200e+02 1.042e+03 1.270e+03 1.569e+03 4.681e+03, threshold=2.539e+03, percent-clipped=2.0
+2023-03-09 10:09:26,679 INFO [train.py:968] (0/2) Epoch 18, batch 21250, giga_loss[loss=0.3254, simple_loss=0.3802, pruned_loss=0.1353, over 26553.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3504, pruned_loss=0.09978, over 5709934.53 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3455, pruned_loss=0.09148, over 5762545.60 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3509, pruned_loss=0.1006, over 5709630.78 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:09:29,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2489, 1.1777, 1.1340, 1.4876], device='cuda:0'), covar=tensor([0.0814, 0.0369, 0.0361, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0115, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0059, 0.0102], device='cuda:0')
+2023-03-09 10:09:32,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:09:55,408 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-798000.pt
+2023-03-09 10:10:05,804 INFO [train.py:968] (0/2) Epoch 18, batch 21300, giga_loss[loss=0.2536, simple_loss=0.3424, pruned_loss=0.08244, over 29129.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3498, pruned_loss=0.09881, over 5709184.90 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3458, pruned_loss=0.09175, over 5762823.55 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.35, pruned_loss=0.09928, over 5708257.15 frames. ], batch size: 128, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:10:22,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=798031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:10:27,825 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=798038.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:10:39,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.317e+02 1.030e+03 1.300e+03 1.659e+03 6.417e+03, threshold=2.599e+03, percent-clipped=7.0
+2023-03-09 10:10:45,668 INFO [train.py:968] (0/2) Epoch 18, batch 21350, giga_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 28299.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3493, pruned_loss=0.0992, over 5700961.28 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3455, pruned_loss=0.0917, over 5764644.06 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3497, pruned_loss=0.09975, over 5697390.10 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:10:55,864 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=798076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:10:58,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=798079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:11:11,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6932, 1.8527, 1.9224, 1.4537], device='cuda:0'), covar=tensor([0.1783, 0.2514, 0.1436, 0.1702], device='cuda:0'), in_proj_covar=tensor([0.0880, 0.0698, 0.0927, 0.0825], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 10:11:22,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=798108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:11:25,366 INFO [train.py:968] (0/2) Epoch 18, batch 21400, giga_loss[loss=0.263, simple_loss=0.3432, pruned_loss=0.09135, over 28660.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3474, pruned_loss=0.09848, over 5704782.22 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3459, pruned_loss=0.09216, over 5767627.58 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3474, pruned_loss=0.09869, over 5697134.97 frames. ], batch size: 85, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:11:26,280 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=798113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:11:28,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=798116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:11:42,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6317, 1.7507, 1.8630, 1.4331], device='cuda:0'), covar=tensor([0.1794, 0.2412, 0.1453, 0.1690], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0698, 0.0927, 0.0825], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 10:11:49,371 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=798145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:11:57,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3958, 1.6577, 1.6780, 1.2281], device='cuda:0'), covar=tensor([0.1707, 0.2665, 0.1431, 0.1738], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0697, 0.0927, 0.0825], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 10:11:59,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.185e+02 1.140e+03 1.444e+03 1.963e+03 6.431e+03, threshold=2.888e+03, percent-clipped=7.0
+2023-03-09 10:12:04,131 INFO [train.py:968] (0/2) Epoch 18, batch 21450, giga_loss[loss=0.3036, simple_loss=0.3718, pruned_loss=0.1177, over 28766.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3458, pruned_loss=0.09797, over 5698578.59 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3461, pruned_loss=0.09232, over 5757129.40 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3456, pruned_loss=0.09812, over 5700279.57 frames. ], batch size: 284, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:12:13,400 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=798174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:12:15,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=798177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:12:37,555 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=798206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:12:41,892 INFO [train.py:968] (0/2) Epoch 18, batch 21500, giga_loss[loss=0.3381, simple_loss=0.4043, pruned_loss=0.136, over 28220.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3443, pruned_loss=0.09814, over 5697956.42 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3464, pruned_loss=0.09268, over 5759966.14 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3439, pruned_loss=0.09805, over 5695756.00 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:13:02,040 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 10:13:15,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.904e+02 1.133e+03 1.497e+03 1.921e+03 3.980e+03, threshold=2.993e+03, percent-clipped=12.0
+2023-03-09 10:13:20,745 INFO [train.py:968] (0/2) Epoch 18, batch 21550, giga_loss[loss=0.2798, simple_loss=0.3544, pruned_loss=0.1026, over 27880.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3441, pruned_loss=0.09844, over 5695428.22 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3461, pruned_loss=0.09245, over 5763513.58 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.344, pruned_loss=0.0987, over 5689102.46 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:13:33,363 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5188, 4.3427, 4.1203, 1.7618], device='cuda:0'), covar=tensor([0.0513, 0.0702, 0.0698, 0.2283], device='cuda:0'), in_proj_covar=tensor([0.1146, 0.1061, 0.0913, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 10:13:46,147 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-09 10:14:00,661 INFO [train.py:968] (0/2) Epoch 18, batch 21600, giga_loss[loss=0.2285, simple_loss=0.3036, pruned_loss=0.07674, over 28540.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3417, pruned_loss=0.09723, over 5700071.74 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3465, pruned_loss=0.09291, over 5765653.67 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3413, pruned_loss=0.09713, over 5692499.79 frames. ], batch size: 85, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:14:38,319 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.192e+02 1.110e+03 1.343e+03 1.696e+03 3.427e+03, threshold=2.686e+03, percent-clipped=3.0
+2023-03-09 10:14:43,058 INFO [train.py:968] (0/2) Epoch 18, batch 21650, giga_loss[loss=0.2461, simple_loss=0.3225, pruned_loss=0.08488, over 28392.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3397, pruned_loss=0.09648, over 5698946.63 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3469, pruned_loss=0.09325, over 5757242.41 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3388, pruned_loss=0.09614, over 5699597.13 frames. ], batch size: 65, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:14:49,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4777, 1.4192, 4.1963, 3.2466], device='cuda:0'), covar=tensor([0.1555, 0.2567, 0.0392, 0.0866], device='cuda:0'), in_proj_covar=tensor([0.0725, 0.0622, 0.0914, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 10:15:07,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-09 10:15:23,255 INFO [train.py:968] (0/2) Epoch 18, batch 21700, giga_loss[loss=0.2282, simple_loss=0.3069, pruned_loss=0.07474, over 28948.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3373, pruned_loss=0.0954, over 5701566.10 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.347, pruned_loss=0.09336, over 5759119.72 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3364, pruned_loss=0.09509, over 5699345.96 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:15:24,014 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=798413.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:15:26,711 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4015, 1.5460, 1.6477, 1.3310], device='cuda:0'), covar=tensor([0.1833, 0.1994, 0.2264, 0.2178], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0741, 0.0701, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 10:15:39,172 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-09 10:15:56,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.736e+02 1.301e+03 1.790e+03 2.587e+03 9.466e+03, threshold=3.580e+03, percent-clipped=21.0
+2023-03-09 10:15:59,087 INFO [train.py:968] (0/2) Epoch 18, batch 21750, giga_loss[loss=0.2321, simple_loss=0.313, pruned_loss=0.07564, over 29105.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3361, pruned_loss=0.09481, over 5716109.50 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3471, pruned_loss=0.09368, over 5765012.28 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3349, pruned_loss=0.09434, over 5707261.70 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:16:07,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5114, 1.8513, 1.5569, 1.5307], device='cuda:0'), covar=tensor([0.2123, 0.2365, 0.2406, 0.2535], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0740, 0.0700, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 10:16:11,488 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5398, 1.5866, 1.7960, 1.4082], device='cuda:0'), covar=tensor([0.1529, 0.2025, 0.1271, 0.1567], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0696, 0.0927, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 10:16:40,896 INFO [train.py:968] (0/2) Epoch 18, batch 21800, giga_loss[loss=0.2112, simple_loss=0.2972, pruned_loss=0.06259, over 28440.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3359, pruned_loss=0.09447, over 5717754.82 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3473, pruned_loss=0.09386, over 5767820.62 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3345, pruned_loss=0.09396, over 5707478.62 frames. ], batch size: 60, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:16:41,167 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2111, 3.1663, 1.3234, 1.4252], device='cuda:0'), covar=tensor([0.0990, 0.0339, 0.0921, 0.1327], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0535, 0.0370, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 10:16:46,626 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=798519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:16:50,563 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-09 10:17:03,942 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4707, 1.7307, 1.7334, 1.2945], device='cuda:0'), covar=tensor([0.1845, 0.2444, 0.1545, 0.1776], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0695, 0.0926, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 10:17:17,597 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=798556.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:17:18,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.931e+02 1.097e+03 1.379e+03 2.287e+03 4.236e+03, threshold=2.758e+03, percent-clipped=2.0
+2023-03-09 10:17:20,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=798559.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:17:22,142 INFO [train.py:968] (0/2) Epoch 18, batch 21850, libri_loss[loss=0.2897, simple_loss=0.3585, pruned_loss=0.1105, over 29537.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3394, pruned_loss=0.09582, over 5714932.42 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3475, pruned_loss=0.09416, over 5770208.16 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3379, pruned_loss=0.09518, over 5703249.74 frames. ], batch size: 80, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:17:29,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3992, 1.6898, 1.3471, 1.4480], device='cuda:0'), covar=tensor([0.2806, 0.2639, 0.3047, 0.2336], device='cuda:0'), in_proj_covar=tensor([0.1437, 0.1046, 0.1274, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 10:17:44,704 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=798588.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:17:51,654 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-09 10:18:05,526 INFO [train.py:968] (0/2) Epoch 18, batch 21900, giga_loss[loss=0.3022, simple_loss=0.3727, pruned_loss=0.1158, over 28567.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3437, pruned_loss=0.09772, over 5696721.38 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3481, pruned_loss=0.09475, over 5762079.40 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3419, pruned_loss=0.09674, over 5693625.19 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 10:18:11,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4284, 1.6734, 1.3678, 1.3356], device='cuda:0'), covar=tensor([0.2482, 0.2431, 0.2752, 0.2199], device='cuda:0'), in_proj_covar=tensor([0.1436, 0.1045, 0.1273, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 10:18:41,319 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3603, 1.9100, 1.4492, 0.6924], device='cuda:0'), covar=tensor([0.6223, 0.2838, 0.3384, 0.6233], device='cuda:0'), in_proj_covar=tensor([0.1676, 0.1579, 0.1566, 0.1373], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 10:18:42,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.455e+02 1.037e+03 1.189e+03 1.623e+03 3.762e+03, threshold=2.379e+03, percent-clipped=5.0
+2023-03-09 10:18:46,269 INFO [train.py:968] (0/2) Epoch 18, batch 21950, giga_loss[loss=0.2505, simple_loss=0.3377, pruned_loss=0.08162, over 28844.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3451, pruned_loss=0.0977, over 5703886.26 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3485, pruned_loss=0.09515, over 5765244.65 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3432, pruned_loss=0.09662, over 5697048.75 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 10:19:21,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1512, 1.4659, 1.5629, 1.2004], device='cuda:0'), covar=tensor([0.1890, 0.1735, 0.2192, 0.2169], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0738, 0.0697, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 10:19:26,422 INFO [train.py:968] (0/2) Epoch 18, batch 22000, giga_loss[loss=0.2575, simple_loss=0.3342, pruned_loss=0.09037, over 28999.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3447, pruned_loss=0.09672, over 5709711.94 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.349, pruned_loss=0.09567, over 5768702.10 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3427, pruned_loss=0.09541, over 5699790.86 frames. ], batch size: 128, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:20:05,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.409e+02 1.131e+03 1.433e+03 1.892e+03 7.857e+03, threshold=2.866e+03, percent-clipped=18.0
+2023-03-09 10:20:09,235 INFO [train.py:968] (0/2) Epoch 18, batch 22050, giga_loss[loss=0.2584, simple_loss=0.3317, pruned_loss=0.09259, over 28716.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3447, pruned_loss=0.0972, over 5701323.50 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3493, pruned_loss=0.09593, over 5766201.68 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3428, pruned_loss=0.09594, over 5695432.26 frames. ], batch size: 99, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:20:43,673 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=798803.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:20:49,963 INFO [train.py:968] (0/2) Epoch 18, batch 22100, libri_loss[loss=0.3172, simple_loss=0.378, pruned_loss=0.1282, over 26055.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.346, pruned_loss=0.09829, over 5699085.78 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.35, pruned_loss=0.09656, over 5764773.28 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3437, pruned_loss=0.09674, over 5693974.43 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:21:15,610 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=798844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:21:26,829 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.548e+02 1.246e+03 1.456e+03 1.975e+03 5.775e+03, threshold=2.911e+03, percent-clipped=6.0
+2023-03-09 10:21:30,250 INFO [train.py:968] (0/2) Epoch 18, batch 22150, giga_loss[loss=0.3209, simple_loss=0.3734, pruned_loss=0.1342, over 26667.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.345, pruned_loss=0.09812, over 5693514.45 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3502, pruned_loss=0.09682, over 5757215.17 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.343, pruned_loss=0.09669, over 5695882.97 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:21:57,843 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=798894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:22:12,295 INFO [train.py:968] (0/2) Epoch 18, batch 22200, giga_loss[loss=0.3105, simple_loss=0.3792, pruned_loss=0.1209, over 28379.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3476, pruned_loss=0.09939, over 5691357.26 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3504, pruned_loss=0.09696, over 5748272.33 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3458, pruned_loss=0.09814, over 5700691.41 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:22:37,400 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9130, 1.9354, 1.3892, 1.6820], device='cuda:0'), covar=tensor([0.0864, 0.0773, 0.1046, 0.1185], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0441, 0.0507, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 10:22:49,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.183e+02 1.231e+03 1.607e+03 2.170e+03 5.119e+03, threshold=3.213e+03, percent-clipped=10.0
+2023-03-09 10:22:53,520 INFO [train.py:968] (0/2) Epoch 18, batch 22250, giga_loss[loss=0.2793, simple_loss=0.3605, pruned_loss=0.0991, over 28868.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.351, pruned_loss=0.1012, over 5691767.10 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3509, pruned_loss=0.0975, over 5746617.43 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3491, pruned_loss=0.09973, over 5700371.83 frames. ], batch size: 186, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:22:54,754 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 10:23:31,978 INFO [train.py:968] (0/2) Epoch 18, batch 22300, giga_loss[loss=0.2629, simple_loss=0.3439, pruned_loss=0.09096, over 28569.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.353, pruned_loss=0.1023, over 5694235.65 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3512, pruned_loss=0.09762, over 5738399.44 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3512, pruned_loss=0.1011, over 5707600.15 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:23:40,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2839, 1.6547, 1.4675, 1.5261], device='cuda:0'), covar=tensor([0.0699, 0.0385, 0.0322, 0.0775], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 10:23:50,993 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=799037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:23:52,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=799040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:24:06,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.693e+02 1.161e+03 1.457e+03 2.233e+03 4.506e+03, threshold=2.914e+03, percent-clipped=10.0
+2023-03-09 10:24:09,140 INFO [train.py:968] (0/2) Epoch 18, batch 22350, giga_loss[loss=0.2848, simple_loss=0.3646, pruned_loss=0.1024, over 28620.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3536, pruned_loss=0.1024, over 5708548.12 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.352, pruned_loss=0.09821, over 5741814.90 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3516, pruned_loss=0.1011, over 5715263.06 frames. ], batch size: 242, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:24:14,935 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=799069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:24:15,833 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-09 10:24:48,662 INFO [train.py:968] (0/2) Epoch 18, batch 22400, giga_loss[loss=0.2742, simple_loss=0.3566, pruned_loss=0.09586, over 28983.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3545, pruned_loss=0.103, over 5707852.94 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3524, pruned_loss=0.09847, over 5746674.65 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3525, pruned_loss=0.1019, over 5707397.39 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:25:25,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.463e+02 1.263e+03 1.547e+03 2.288e+03 9.035e+03, threshold=3.093e+03, percent-clipped=12.0
+2023-03-09 10:25:27,408 INFO [train.py:968] (0/2) Epoch 18, batch 22450, giga_loss[loss=0.2783, simple_loss=0.3565, pruned_loss=0.1, over 29054.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3531, pruned_loss=0.1022, over 5707724.28 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3526, pruned_loss=0.09867, over 5740131.06 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3513, pruned_loss=0.1012, over 5713055.47 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:25:42,296 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=799178.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:26:02,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3965, 2.0359, 1.4901, 0.6344], device='cuda:0'), covar=tensor([0.5185, 0.2485, 0.3868, 0.6004], device='cuda:0'), in_proj_covar=tensor([0.1679, 0.1577, 0.1565, 0.1373], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 10:26:06,481 INFO [train.py:968] (0/2) Epoch 18, batch 22500, giga_loss[loss=0.2512, simple_loss=0.3259, pruned_loss=0.08822, over 28695.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3507, pruned_loss=0.1009, over 5717224.20 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3524, pruned_loss=0.09875, over 5745905.75 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3494, pruned_loss=0.1002, over 5715090.09 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:26:12,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=799219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:26:20,073 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 10:26:36,039 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=799247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:26:44,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.831e+02 1.159e+03 1.439e+03 1.981e+03 5.037e+03, threshold=2.877e+03, percent-clipped=8.0
+2023-03-09 10:26:46,443 INFO [train.py:968] (0/2) Epoch 18, batch 22550, giga_loss[loss=0.2898, simple_loss=0.3595, pruned_loss=0.11, over 28998.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3478, pruned_loss=0.0996, over 5719682.39 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.353, pruned_loss=0.09935, over 5745483.27 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3462, pruned_loss=0.09844, over 5717848.10 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:26:47,576 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-09 10:27:00,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3380, 3.4769, 1.4829, 1.4714], device='cuda:0'), covar=tensor([0.0985, 0.0293, 0.0942, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0538, 0.0369, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 10:27:11,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1891, 3.9941, 3.7943, 1.9382], device='cuda:0'), covar=tensor([0.0579, 0.0719, 0.0699, 0.1998], device='cuda:0'), in_proj_covar=tensor([0.1153, 0.1066, 0.0917, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 10:27:25,347 INFO [train.py:968] (0/2) Epoch 18, batch 22600, giga_loss[loss=0.2939, simple_loss=0.3685, pruned_loss=0.1096, over 28890.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3456, pruned_loss=0.09838, over 5721445.56 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3532, pruned_loss=0.09958, over 5749483.96 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3439, pruned_loss=0.09722, over 5715690.20 frames. ], batch size: 186, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:27:32,817 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=799321.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:27:34,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=799324.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:27:57,871 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=799353.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:28:02,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.179e+02 1.191e+03 1.370e+03 1.756e+03 5.673e+03, threshold=2.741e+03, percent-clipped=7.0
+2023-03-09 10:28:05,821 INFO [train.py:968] (0/2) Epoch 18, batch 22650, giga_loss[loss=0.2435, simple_loss=0.3238, pruned_loss=0.08156, over 28648.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3455, pruned_loss=0.09707, over 5707407.28 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3532, pruned_loss=0.09974, over 5743508.40 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.344, pruned_loss=0.09597, over 5707180.86 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:28:06,077 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=799362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:28:07,919 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=799365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:28:31,221 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=799394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:28:46,590 INFO [train.py:968] (0/2) Epoch 18, batch 22700, giga_loss[loss=0.2745, simple_loss=0.3586, pruned_loss=0.09517, over 28626.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3468, pruned_loss=0.0969, over 5713470.23 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3537, pruned_loss=0.1002, over 5745988.63 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.345, pruned_loss=0.09554, over 5710343.31 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:29:21,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.498e+02 1.145e+03 1.447e+03 1.764e+03 5.145e+03, threshold=2.894e+03, percent-clipped=9.0
+2023-03-09 10:29:24,740 INFO [train.py:968] (0/2) Epoch 18, batch 22750, giga_loss[loss=0.2477, simple_loss=0.3109, pruned_loss=0.0923, over 28502.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3456, pruned_loss=0.09644, over 5724537.59 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3542, pruned_loss=0.1006, over 5747907.43 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3435, pruned_loss=0.09488, over 5719861.26 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:29:37,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 10:29:41,009 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=799484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:30:03,361 INFO [train.py:968] (0/2) Epoch 18, batch 22800, giga_loss[loss=0.2738, simple_loss=0.3399, pruned_loss=0.1039, over 28979.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3458, pruned_loss=0.09796, over 5721008.82 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.355, pruned_loss=0.1013, over 5744754.69 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3434, pruned_loss=0.09608, over 5719474.55 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:30:08,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3329, 3.5569, 1.4712, 1.4647], device='cuda:0'), covar=tensor([0.0924, 0.0355, 0.0928, 0.1294], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0539, 0.0369, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 10:30:26,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5489, 2.3218, 2.1914, 2.0562], device='cuda:0'), covar=tensor([0.1678, 0.2518, 0.2155, 0.2282], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0736, 0.0698, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 10:30:37,557 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 10:30:42,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.854e+02 1.139e+03 1.504e+03 1.989e+03 6.030e+03, threshold=3.007e+03, percent-clipped=11.0
+2023-03-09 10:30:44,922 INFO [train.py:968] (0/2) Epoch 18, batch 22850, giga_loss[loss=0.2302, simple_loss=0.3079, pruned_loss=0.07624, over 28966.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3443, pruned_loss=0.0988, over 5716801.20 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3551, pruned_loss=0.1015, over 5745356.87 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3421, pruned_loss=0.097, over 5714395.23 frames. ], batch size: 213, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:31:21,915 INFO [train.py:968] (0/2) Epoch 18, batch 22900, giga_loss[loss=0.2709, simple_loss=0.3426, pruned_loss=0.09967, over 28891.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3425, pruned_loss=0.0986, over 5722498.80 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3551, pruned_loss=0.1017, over 5748630.79 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3403, pruned_loss=0.09692, over 5716847.08 frames. ], batch size: 186, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:31:30,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=799622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:31:57,530 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.562e+02 1.143e+03 1.314e+03 2.091e+03 5.161e+03, threshold=2.627e+03, percent-clipped=9.0
+2023-03-09 10:32:00,357 INFO [train.py:968] (0/2) Epoch 18, batch 22950, giga_loss[loss=0.2497, simple_loss=0.3266, pruned_loss=0.08635, over 28926.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3416, pruned_loss=0.09876, over 5715717.13 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3548, pruned_loss=0.1017, over 5741096.10 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3397, pruned_loss=0.09734, over 5717122.85 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:32:13,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4709, 1.4139, 1.2608, 1.0591], device='cuda:0'), covar=tensor([0.0668, 0.0388, 0.0795, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0442, 0.0508, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 10:32:37,027 INFO [train.py:968] (0/2) Epoch 18, batch 23000, giga_loss[loss=0.2622, simple_loss=0.3227, pruned_loss=0.1009, over 28948.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3388, pruned_loss=0.09761, over 5716450.85 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3548, pruned_loss=0.1021, over 5744150.16 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3368, pruned_loss=0.09596, over 5713739.16 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:32:55,260 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=799736.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:32:55,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4633, 2.2083, 1.6777, 0.7827], device='cuda:0'), covar=tensor([0.5698, 0.2639, 0.3772, 0.6403], device='cuda:0'), in_proj_covar=tensor([0.1672, 0.1570, 0.1555, 0.1367], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 10:33:11,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.310e+02 1.278e+03 1.751e+03 2.382e+03 6.418e+03, threshold=3.502e+03, percent-clipped=17.0
+2023-03-09 10:33:14,955 INFO [train.py:968] (0/2) Epoch 18, batch 23050, giga_loss[loss=0.2288, simple_loss=0.3062, pruned_loss=0.07568, over 28868.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3348, pruned_loss=0.09557, over 5723315.32 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.355, pruned_loss=0.1024, over 5745683.68 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3328, pruned_loss=0.09393, over 5719314.89 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:33:15,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3737, 1.7974, 1.7323, 1.5423], device='cuda:0'), covar=tensor([0.1673, 0.1536, 0.1931, 0.1689], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0739, 0.0699, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 10:33:17,237 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=799765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:33:19,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=799768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:33:42,630 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=799797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:33:52,115 INFO [train.py:968] (0/2) Epoch 18, batch 23100, giga_loss[loss=0.236, simple_loss=0.3144, pruned_loss=0.07876, over 28884.00 frames. ], tot_loss[loss=0.259, simple_loss=0.331, pruned_loss=0.09347, over 5717821.81 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.355, pruned_loss=0.1025, over 5740695.97 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3288, pruned_loss=0.09178, over 5717580.51 frames. ], batch size: 174, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:34:11,227 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=799835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:34:31,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2628, 1.1199, 4.0069, 3.2372], device='cuda:0'), covar=tensor([0.1894, 0.3083, 0.0752, 0.0981], device='cuda:0'), in_proj_covar=tensor([0.0724, 0.0626, 0.0919, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 10:34:31,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=799859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:34:32,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.877e+02 1.229e+03 1.586e+03 2.140e+03 5.857e+03, threshold=3.171e+03, percent-clipped=6.0
+2023-03-09 10:34:33,941 INFO [train.py:968] (0/2) Epoch 18, batch 23150, giga_loss[loss=0.3346, simple_loss=0.3917, pruned_loss=0.1388, over 26637.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.333, pruned_loss=0.0944, over 5713101.96 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3555, pruned_loss=0.103, over 5742921.68 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3304, pruned_loss=0.09256, over 5710747.62 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:35:14,264 INFO [train.py:968] (0/2) Epoch 18, batch 23200, giga_loss[loss=0.2774, simple_loss=0.3562, pruned_loss=0.09934, over 28632.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3365, pruned_loss=0.0959, over 5715676.66 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3559, pruned_loss=0.1033, over 5745851.13 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3337, pruned_loss=0.09399, over 5710778.48 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:35:53,465 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-09 10:35:55,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.870e+02 1.195e+03 1.558e+03 2.142e+03 5.822e+03, threshold=3.117e+03, percent-clipped=7.0
+2023-03-09 10:35:56,957 INFO [train.py:968] (0/2) Epoch 18, batch 23250, giga_loss[loss=0.262, simple_loss=0.3398, pruned_loss=0.09212, over 28942.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3406, pruned_loss=0.09797, over 5702479.20 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3561, pruned_loss=0.1036, over 5733827.89 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.338, pruned_loss=0.09615, over 5708634.91 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:36:27,213 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-800000.pt
+2023-03-09 10:36:29,213 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:36:30,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:36:36,242 INFO [train.py:968] (0/2) Epoch 18, batch 23300, libri_loss[loss=0.291, simple_loss=0.3646, pruned_loss=0.1087, over 29088.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.343, pruned_loss=0.09829, over 5710025.36 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3561, pruned_loss=0.1036, over 5738669.68 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3406, pruned_loss=0.09666, over 5709792.03 frames. ], batch size: 101, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:36:55,295 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:37:19,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.901e+02 1.231e+03 1.614e+03 2.167e+03 5.244e+03, threshold=3.227e+03, percent-clipped=6.0
+2023-03-09 10:37:20,401 INFO [train.py:968] (0/2) Epoch 18, batch 23350, giga_loss[loss=0.2672, simple_loss=0.3487, pruned_loss=0.09287, over 28607.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3445, pruned_loss=0.09857, over 5718619.89 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3559, pruned_loss=0.1036, over 5740476.78 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3426, pruned_loss=0.09722, over 5716743.28 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 10:37:39,757 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 10:38:00,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800111.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:38:02,073 INFO [train.py:968] (0/2) Epoch 18, batch 23400, giga_loss[loss=0.3702, simple_loss=0.4223, pruned_loss=0.159, over 28736.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3484, pruned_loss=0.102, over 5715793.68 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3565, pruned_loss=0.1045, over 5745713.07 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3459, pruned_loss=0.09998, over 5708452.00 frames. ], batch size: 284, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:38:47,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.052e+02 1.541e+03 2.001e+03 3.474e+03 1.501e+04, threshold=4.001e+03, percent-clipped=28.0
+2023-03-09 10:38:48,179 INFO [train.py:968] (0/2) Epoch 18, batch 23450, giga_loss[loss=0.2348, simple_loss=0.3204, pruned_loss=0.0746, over 28911.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3552, pruned_loss=0.1076, over 5715371.47 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3572, pruned_loss=0.1054, over 5750955.47 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3526, pruned_loss=0.1052, over 5703461.81 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:38:55,697 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-09 10:39:36,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:39:37,807 INFO [train.py:968] (0/2) Epoch 18, batch 23500, giga_loss[loss=0.3069, simple_loss=0.3815, pruned_loss=0.1161, over 29064.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3641, pruned_loss=0.1149, over 5701700.58 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3577, pruned_loss=0.1059, over 5754443.99 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3615, pruned_loss=0.1126, over 5688261.65 frames. ], batch size: 128, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:40:20,622 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:40:23,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800257.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:40:27,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.760e+03 2.110e+03 2.958e+03 7.796e+03, threshold=4.219e+03, percent-clipped=8.0
+2023-03-09 10:40:28,013 INFO [train.py:968] (0/2) Epoch 18, batch 23550, giga_loss[loss=0.3137, simple_loss=0.3789, pruned_loss=0.1242, over 28508.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3692, pruned_loss=0.1185, over 5692356.48 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3577, pruned_loss=0.1059, over 5756748.57 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3673, pruned_loss=0.1168, over 5678866.38 frames. ], batch size: 71, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:40:47,330 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=800284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:40:48,623 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800286.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:41:02,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4905, 1.7487, 1.3833, 1.6031], device='cuda:0'), covar=tensor([0.2307, 0.2363, 0.2700, 0.1984], device='cuda:0'), in_proj_covar=tensor([0.1433, 0.1045, 0.1272, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 10:41:14,369 INFO [train.py:968] (0/2) Epoch 18, batch 23600, giga_loss[loss=0.3428, simple_loss=0.3957, pruned_loss=0.1449, over 28542.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3753, pruned_loss=0.1237, over 5685863.99 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.358, pruned_loss=0.1063, over 5755259.54 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3737, pruned_loss=0.1223, over 5675297.02 frames. ], batch size: 60, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 10:41:49,194 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=800345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:41:51,037 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-09 10:41:55,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:41:58,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:42:06,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.643e+03 2.213e+03 2.984e+03 8.183e+03, threshold=4.425e+03, percent-clipped=11.0
+2023-03-09 10:42:06,131 INFO [train.py:968] (0/2) Epoch 18, batch 23650, giga_loss[loss=0.33, simple_loss=0.3842, pruned_loss=0.1379, over 28781.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3805, pruned_loss=0.1277, over 5687845.18 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3583, pruned_loss=0.1065, over 5755647.24 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3791, pruned_loss=0.1265, over 5678980.05 frames. ], batch size: 92, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:42:28,699 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:42:32,393 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=800389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:42:51,223 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=800407.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:42:55,212 INFO [train.py:968] (0/2) Epoch 18, batch 23700, giga_loss[loss=0.2787, simple_loss=0.3526, pruned_loss=0.1024, over 28979.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3837, pruned_loss=0.1313, over 5675911.31 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3588, pruned_loss=0.1069, over 5756257.00 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3823, pruned_loss=0.1301, over 5667972.28 frames. ], batch size: 128, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:43:40,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.753e+03 2.321e+03 3.101e+03 1.093e+04, threshold=4.642e+03, percent-clipped=11.0
+2023-03-09 10:43:40,611 INFO [train.py:968] (0/2) Epoch 18, batch 23750, giga_loss[loss=0.3177, simple_loss=0.3757, pruned_loss=0.1299, over 28835.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3876, pruned_loss=0.1359, over 5643149.78 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3595, pruned_loss=0.1077, over 5733925.48 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3867, pruned_loss=0.1352, over 5653514.12 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:44:13,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7147, 1.4094, 4.9596, 3.6602], device='cuda:0'), covar=tensor([0.1569, 0.2793, 0.0430, 0.0947], device='cuda:0'), in_proj_covar=tensor([0.0727, 0.0626, 0.0923, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 10:44:28,477 INFO [train.py:968] (0/2) Epoch 18, batch 23800, giga_loss[loss=0.3612, simple_loss=0.4081, pruned_loss=0.1572, over 28293.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3901, pruned_loss=0.139, over 5638168.57 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3596, pruned_loss=0.1079, over 5735443.68 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3899, pruned_loss=0.1388, over 5643331.44 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:45:22,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.170e+03 1.796e+03 2.203e+03 2.752e+03 6.708e+03, threshold=4.405e+03, percent-clipped=2.0
+2023-03-09 10:45:22,685 INFO [train.py:968] (0/2) Epoch 18, batch 23850, giga_loss[loss=0.492, simple_loss=0.4837, pruned_loss=0.2501, over 26526.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3935, pruned_loss=0.1416, over 5635237.59 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3596, pruned_loss=0.108, over 5731796.04 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3942, pruned_loss=0.1422, over 5639686.01 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:45:58,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9517, 2.2140, 2.1531, 1.7058], device='cuda:0'), covar=tensor([0.2756, 0.2213, 0.2127, 0.2762], device='cuda:0'), in_proj_covar=tensor([0.1893, 0.1817, 0.1758, 0.1899], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 10:46:03,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6533, 4.8569, 1.9631, 1.7719], device='cuda:0'), covar=tensor([0.0950, 0.0262, 0.0799, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0542, 0.0370, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 10:46:12,673 INFO [train.py:968] (0/2) Epoch 18, batch 23900, giga_loss[loss=0.2805, simple_loss=0.3549, pruned_loss=0.1031, over 28879.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3918, pruned_loss=0.14, over 5640867.27 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3598, pruned_loss=0.1083, over 5729526.67 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3936, pruned_loss=0.1416, over 5642967.72 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:46:59,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:47:01,056 INFO [train.py:968] (0/2) Epoch 18, batch 23950, libri_loss[loss=0.2509, simple_loss=0.325, pruned_loss=0.08836, over 29653.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3918, pruned_loss=0.1412, over 5629999.56 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3599, pruned_loss=0.1086, over 5724954.84 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3939, pruned_loss=0.1431, over 5633748.77 frames. ], batch size: 73, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 10:47:01,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.863e+03 2.502e+03 3.506e+03 8.750e+03, threshold=5.005e+03, percent-clipped=11.0
+2023-03-09 10:47:20,819 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-09 10:47:36,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3658, 4.1931, 4.0304, 2.0640], device='cuda:0'), covar=tensor([0.0613, 0.0765, 0.0760, 0.1897], device='cuda:0'), in_proj_covar=tensor([0.1181, 0.1092, 0.0939, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 10:47:46,027 INFO [train.py:968] (0/2) Epoch 18, batch 24000, giga_loss[loss=0.4208, simple_loss=0.4545, pruned_loss=0.1936, over 27520.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.392, pruned_loss=0.1414, over 5636312.60 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3604, pruned_loss=0.109, over 5724551.74 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3943, pruned_loss=0.1435, over 5637067.93 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:47:46,031 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 10:47:55,528 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2069, simple_loss=0.3142, pruned_loss=0.04982, over 944034.00 frames. 
+2023-03-09 10:47:55,529 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 10:47:57,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2407, 1.2695, 3.7864, 3.2735], device='cuda:0'), covar=tensor([0.1636, 0.2661, 0.0453, 0.0918], device='cuda:0'), in_proj_covar=tensor([0.0731, 0.0630, 0.0928, 0.0861], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 10:48:01,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:48:12,757 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-09 10:48:36,674 INFO [train.py:968] (0/2) Epoch 18, batch 24050, giga_loss[loss=0.3328, simple_loss=0.3941, pruned_loss=0.1358, over 28930.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3913, pruned_loss=0.1397, over 5644713.06 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3604, pruned_loss=0.1092, over 5726017.16 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3942, pruned_loss=0.1423, over 5641250.58 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:48:39,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.525e+02 1.777e+03 2.350e+03 3.156e+03 7.379e+03, threshold=4.699e+03, percent-clipped=7.0
+2023-03-09 10:48:40,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:48:59,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800782.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:49:20,035 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:49:22,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:49:28,156 INFO [train.py:968] (0/2) Epoch 18, batch 24100, giga_loss[loss=0.3911, simple_loss=0.4188, pruned_loss=0.1817, over 26668.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3919, pruned_loss=0.1403, over 5625924.93 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3606, pruned_loss=0.1094, over 5719964.73 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3946, pruned_loss=0.1427, over 5627556.67 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:49:52,263 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:50:19,082 INFO [train.py:968] (0/2) Epoch 18, batch 24150, giga_loss[loss=0.2671, simple_loss=0.348, pruned_loss=0.09313, over 28883.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3912, pruned_loss=0.1393, over 5616378.93 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3602, pruned_loss=0.1092, over 5715558.81 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3946, pruned_loss=0.1422, over 5619225.61 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:50:20,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.583e+02 1.671e+03 2.035e+03 2.593e+03 4.929e+03, threshold=4.069e+03, percent-clipped=1.0
+2023-03-09 10:50:20,347 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:50:22,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:50:26,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4784, 1.2300, 4.8970, 3.7229], device='cuda:0'), covar=tensor([0.2067, 0.3066, 0.0668, 0.0981], device='cuda:0'), in_proj_covar=tensor([0.0730, 0.0629, 0.0926, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 10:50:51,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:51:04,646 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:51:06,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:51:07,695 INFO [train.py:968] (0/2) Epoch 18, batch 24200, giga_loss[loss=0.2852, simple_loss=0.3649, pruned_loss=0.1028, over 28870.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3864, pruned_loss=0.1345, over 5629095.55 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3603, pruned_loss=0.1096, over 5721556.88 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.39, pruned_loss=0.1374, over 5623102.37 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:51:19,506 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800925.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:51:22,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800928.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:51:32,852 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:51:49,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800957.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:51:55,373 INFO [train.py:968] (0/2) Epoch 18, batch 24250, giga_loss[loss=0.292, simple_loss=0.3695, pruned_loss=0.1073, over 28999.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3845, pruned_loss=0.1313, over 5637061.97 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3603, pruned_loss=0.1097, over 5721936.09 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3879, pruned_loss=0.134, over 5630339.84 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:51:56,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.341e+02 1.683e+03 2.489e+03 3.524e+03 1.108e+04, threshold=4.978e+03, percent-clipped=18.0
+2023-03-09 10:52:05,808 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.87 vs. limit=5.0
+2023-03-09 10:52:43,281 INFO [train.py:968] (0/2) Epoch 18, batch 24300, giga_loss[loss=0.3468, simple_loss=0.4008, pruned_loss=0.1464, over 27541.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3807, pruned_loss=0.1282, over 5650823.59 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3604, pruned_loss=0.1101, over 5718486.32 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3841, pruned_loss=0.1307, over 5645801.14 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:53:21,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:53:28,892 INFO [train.py:968] (0/2) Epoch 18, batch 24350, giga_loss[loss=0.3006, simple_loss=0.3677, pruned_loss=0.1167, over 29062.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3776, pruned_loss=0.1252, over 5658524.53 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3606, pruned_loss=0.1102, over 5720098.11 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3804, pruned_loss=0.1274, over 5652094.79 frames. ], batch size: 155, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:53:30,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.628e+03 2.167e+03 2.720e+03 6.665e+03, threshold=4.335e+03, percent-clipped=1.0
+2023-03-09 10:53:35,803 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:54:16,418 INFO [train.py:968] (0/2) Epoch 18, batch 24400, giga_loss[loss=0.3249, simple_loss=0.3934, pruned_loss=0.1282, over 29025.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3751, pruned_loss=0.1239, over 5664593.91 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3602, pruned_loss=0.1102, over 5724803.09 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.378, pruned_loss=0.126, over 5653765.52 frames. ], batch size: 155, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 10:55:04,351 INFO [train.py:968] (0/2) Epoch 18, batch 24450, giga_loss[loss=0.2841, simple_loss=0.3631, pruned_loss=0.1025, over 28857.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.375, pruned_loss=0.1241, over 5656317.02 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3602, pruned_loss=0.1104, over 5717061.48 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3779, pruned_loss=0.1259, over 5651880.04 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 10:55:05,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.199e+03 1.786e+03 2.309e+03 2.909e+03 7.620e+03, threshold=4.619e+03, percent-clipped=6.0
+2023-03-09 10:55:16,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2096, 1.5081, 1.4784, 1.0706], device='cuda:0'), covar=tensor([0.1501, 0.2492, 0.1286, 0.1549], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0695, 0.0919, 0.0820], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 10:55:21,908 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4688, 1.6098, 1.5255, 1.4632], device='cuda:0'), covar=tensor([0.1483, 0.1796, 0.2040, 0.1764], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0742, 0.0702, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 10:55:43,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3414, 1.4310, 3.8820, 3.3606], device='cuda:0'), covar=tensor([0.1610, 0.2504, 0.0491, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0633, 0.0933, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 10:55:57,222 INFO [train.py:968] (0/2) Epoch 18, batch 24500, giga_loss[loss=0.3091, simple_loss=0.3514, pruned_loss=0.1334, over 23466.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3746, pruned_loss=0.1234, over 5665425.17 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3605, pruned_loss=0.1106, over 5718438.14 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3768, pruned_loss=0.1249, over 5660118.47 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:56:45,171 INFO [train.py:968] (0/2) Epoch 18, batch 24550, giga_loss[loss=0.2927, simple_loss=0.3732, pruned_loss=0.1061, over 28874.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3744, pruned_loss=0.1215, over 5673496.64 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3603, pruned_loss=0.1107, over 5718058.76 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3767, pruned_loss=0.1229, over 5667945.62 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:56:46,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.884e+02 1.479e+03 1.970e+03 2.646e+03 6.550e+03, threshold=3.939e+03, percent-clipped=4.0
+2023-03-09 10:56:51,168 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:57:36,572 INFO [train.py:968] (0/2) Epoch 18, batch 24600, giga_loss[loss=0.2951, simple_loss=0.3845, pruned_loss=0.1029, over 28921.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3753, pruned_loss=0.1205, over 5666183.23 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3601, pruned_loss=0.1107, over 5720716.07 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3776, pruned_loss=0.1218, over 5658708.83 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:58:26,351 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:58:28,204 INFO [train.py:968] (0/2) Epoch 18, batch 24650, libri_loss[loss=0.3429, simple_loss=0.3846, pruned_loss=0.1507, over 29563.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3758, pruned_loss=0.1211, over 5668605.35 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3599, pruned_loss=0.1108, over 5724450.23 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.378, pruned_loss=0.1222, over 5658267.54 frames. ], batch size: 75, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:58:30,027 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.756e+03 2.257e+03 2.971e+03 7.273e+03, threshold=4.515e+03, percent-clipped=8.0
+2023-03-09 10:59:06,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7176, 1.6867, 1.4230, 1.3211], device='cuda:0'), covar=tensor([0.0720, 0.0452, 0.0947, 0.0890], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0442, 0.0507, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 10:59:11,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6457, 1.5603, 1.8842, 1.4466], device='cuda:0'), covar=tensor([0.1606, 0.2255, 0.1307, 0.1541], device='cuda:0'), in_proj_covar=tensor([0.0871, 0.0694, 0.0918, 0.0820], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 10:59:14,887 INFO [train.py:968] (0/2) Epoch 18, batch 24700, giga_loss[loss=0.3093, simple_loss=0.3744, pruned_loss=0.1221, over 28957.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3773, pruned_loss=0.1231, over 5659317.62 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3602, pruned_loss=0.111, over 5719602.95 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3793, pruned_loss=0.1241, over 5653891.01 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:59:28,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=801428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:59:33,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7171, 1.8859, 1.6315, 1.9067], device='cuda:0'), covar=tensor([0.2019, 0.2080, 0.1998, 0.1881], device='cuda:0'), in_proj_covar=tensor([0.1437, 0.1048, 0.1277, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 10:59:45,921 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=801443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:59:57,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2464, 1.4924, 1.4353, 1.3345], device='cuda:0'), covar=tensor([0.1667, 0.1583, 0.2246, 0.1719], device='cuda:0'), in_proj_covar=tensor([0.0462, 0.0743, 0.0703, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 11:00:00,249 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:00:02,811 INFO [train.py:968] (0/2) Epoch 18, batch 24750, libri_loss[loss=0.3537, simple_loss=0.4017, pruned_loss=0.1528, over 19656.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3761, pruned_loss=0.1235, over 5644230.90 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3602, pruned_loss=0.1112, over 5712552.46 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3778, pruned_loss=0.1243, over 5646317.30 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:00:04,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.129e+03 1.903e+03 2.367e+03 3.063e+03 1.094e+04, threshold=4.734e+03, percent-clipped=5.0
+2023-03-09 11:00:49,236 INFO [train.py:968] (0/2) Epoch 18, batch 24800, giga_loss[loss=0.3256, simple_loss=0.3672, pruned_loss=0.142, over 23553.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3739, pruned_loss=0.1228, over 5655191.84 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3601, pruned_loss=0.1111, over 5713530.70 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3754, pruned_loss=0.1235, over 5655668.93 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:00:49,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2461, 4.0709, 3.8754, 1.8906], device='cuda:0'), covar=tensor([0.0608, 0.0742, 0.0741, 0.2133], device='cuda:0'), in_proj_covar=tensor([0.1177, 0.1092, 0.0937, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 11:01:32,667 INFO [train.py:968] (0/2) Epoch 18, batch 24850, giga_loss[loss=0.2779, simple_loss=0.3552, pruned_loss=0.1003, over 28557.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3731, pruned_loss=0.122, over 5661896.94 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3606, pruned_loss=0.1114, over 5716043.85 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3741, pruned_loss=0.1225, over 5659065.33 frames. ], batch size: 71, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:01:35,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.608e+02 1.627e+03 2.196e+03 2.927e+03 5.749e+03, threshold=4.392e+03, percent-clipped=3.0
+2023-03-09 11:01:40,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=801571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:01:43,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=801574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:01:53,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=801586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:01:55,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=801589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:02:11,271 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=801603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:02:16,974 INFO [train.py:968] (0/2) Epoch 18, batch 24900, giga_loss[loss=0.2803, simple_loss=0.3574, pruned_loss=0.1016, over 28878.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3724, pruned_loss=0.1202, over 5676428.28 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.361, pruned_loss=0.1118, over 5719343.79 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3733, pruned_loss=0.1206, over 5668981.73 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:02:23,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=801618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:02:48,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=801643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:02:50,833 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:02:58,041 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-09 11:03:06,550 INFO [train.py:968] (0/2) Epoch 18, batch 24950, giga_loss[loss=0.3195, simple_loss=0.3858, pruned_loss=0.1266, over 28700.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3724, pruned_loss=0.1204, over 5667293.74 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3611, pruned_loss=0.1118, over 5721934.99 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3733, pruned_loss=0.1208, over 5658385.70 frames. ], batch size: 242, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:03:08,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.248e+02 1.561e+03 2.000e+03 2.837e+03 1.080e+04, threshold=4.000e+03, percent-clipped=10.0
+2023-03-09 11:03:14,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-09 11:03:22,044 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:03:56,037 INFO [train.py:968] (0/2) Epoch 18, batch 25000, giga_loss[loss=0.401, simple_loss=0.4178, pruned_loss=0.1921, over 26505.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3721, pruned_loss=0.12, over 5674346.30 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3611, pruned_loss=0.1119, over 5724199.12 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3729, pruned_loss=0.1204, over 5664592.58 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:04:15,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=801734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:04:41,921 INFO [train.py:968] (0/2) Epoch 18, batch 25050, libri_loss[loss=0.3665, simple_loss=0.4155, pruned_loss=0.1587, over 29650.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3709, pruned_loss=0.1194, over 5691168.99 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3616, pruned_loss=0.1122, over 5728764.27 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3713, pruned_loss=0.1196, over 5678153.18 frames. ], batch size: 88, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:04:43,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.352e+02 1.546e+03 2.036e+03 2.864e+03 7.888e+03, threshold=4.072e+03, percent-clipped=9.0
+2023-03-09 11:05:05,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=801786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:05:07,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=801789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:05:26,167 INFO [train.py:968] (0/2) Epoch 18, batch 25100, giga_loss[loss=0.3284, simple_loss=0.3882, pruned_loss=0.1343, over 29022.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3702, pruned_loss=0.1197, over 5686639.99 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3615, pruned_loss=0.1123, over 5727838.75 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3708, pruned_loss=0.12, over 5675876.57 frames. ], batch size: 155, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:05:31,991 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=801818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:05:46,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=801833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:05:56,052 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:06:12,517 INFO [train.py:968] (0/2) Epoch 18, batch 25150, giga_loss[loss=0.2832, simple_loss=0.3542, pruned_loss=0.1061, over 29004.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3689, pruned_loss=0.1194, over 5695759.18 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.362, pruned_loss=0.1127, over 5728397.30 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.369, pruned_loss=0.1193, over 5686460.81 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:06:16,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.739e+03 2.385e+03 3.387e+03 1.000e+04, threshold=4.770e+03, percent-clipped=15.0
+2023-03-09 11:06:17,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4629, 1.5227, 1.5278, 1.4664], device='cuda:0'), covar=tensor([0.2229, 0.2174, 0.1523, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.1888, 0.1816, 0.1756, 0.1899], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 11:06:20,110 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:06:28,811 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=801877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:06:32,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=801880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:06:35,828 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.81 vs. limit=5.0
+2023-03-09 11:06:57,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=801909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:07:01,303 INFO [train.py:968] (0/2) Epoch 18, batch 25200, giga_loss[loss=0.3252, simple_loss=0.3677, pruned_loss=0.1414, over 23585.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3671, pruned_loss=0.1186, over 5695689.90 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3623, pruned_loss=0.1128, over 5732792.03 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3671, pruned_loss=0.1185, over 5683427.13 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:07:04,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3718, 1.8917, 1.4171, 0.7650], device='cuda:0'), covar=tensor([0.4768, 0.2483, 0.2812, 0.5141], device='cuda:0'), in_proj_covar=tensor([0.1692, 0.1604, 0.1576, 0.1382], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 11:07:23,631 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:07:25,228 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-09 11:07:46,131 INFO [train.py:968] (0/2) Epoch 18, batch 25250, giga_loss[loss=0.2669, simple_loss=0.3448, pruned_loss=0.09448, over 28807.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3675, pruned_loss=0.1195, over 5687034.78 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3627, pruned_loss=0.1131, over 5725557.93 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3673, pruned_loss=0.1194, over 5682191.88 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:07:48,669 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.697e+03 2.144e+03 2.820e+03 7.064e+03, threshold=4.289e+03, percent-clipped=4.0
+2023-03-09 11:07:52,631 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:08:02,958 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=801976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:08:05,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=801979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:08:21,792 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-09 11:08:22,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4142, 2.5742, 1.5350, 1.5803], device='cuda:0'), covar=tensor([0.0811, 0.0329, 0.0713, 0.1087], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0544, 0.0371, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 11:08:23,981 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-802000.pt
+2023-03-09 11:08:32,339 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:08:34,658 INFO [train.py:968] (0/2) Epoch 18, batch 25300, libri_loss[loss=0.2755, simple_loss=0.3544, pruned_loss=0.09829, over 29654.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.12, over 5681228.02 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3628, pruned_loss=0.1131, over 5725852.46 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3679, pruned_loss=0.1201, over 5675856.64 frames. ], batch size: 88, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:08:40,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:08:41,531 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-09 11:09:04,282 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 11:09:09,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:09:14,710 INFO [train.py:968] (0/2) Epoch 18, batch 25350, giga_loss[loss=0.2714, simple_loss=0.3538, pruned_loss=0.09445, over 29016.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3669, pruned_loss=0.1184, over 5696765.64 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3621, pruned_loss=0.1128, over 5733593.83 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3676, pruned_loss=0.119, over 5683728.80 frames. ], batch size: 155, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:09:18,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.849e+03 2.352e+03 3.649e+03 1.035e+04, threshold=4.703e+03, percent-clipped=16.0
+2023-03-09 11:09:59,576 INFO [train.py:968] (0/2) Epoch 18, batch 25400, giga_loss[loss=0.2647, simple_loss=0.3435, pruned_loss=0.09295, over 29002.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3672, pruned_loss=0.1183, over 5677795.55 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3622, pruned_loss=0.1129, over 5723724.62 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3678, pruned_loss=0.1188, over 5675708.45 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:10:42,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6062, 2.5970, 1.6164, 1.7584], device='cuda:0'), covar=tensor([0.0736, 0.0320, 0.0694, 0.0990], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0544, 0.0371, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 11:10:45,527 INFO [train.py:968] (0/2) Epoch 18, batch 25450, giga_loss[loss=0.2994, simple_loss=0.3738, pruned_loss=0.1124, over 28982.00 frames. ], tot_loss[loss=0.302, simple_loss=0.368, pruned_loss=0.118, over 5685433.62 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3623, pruned_loss=0.1131, over 5728940.16 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3685, pruned_loss=0.1184, over 5677967.02 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:10:47,392 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:10:49,922 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.179e+02 1.629e+03 2.130e+03 3.502e+03 1.125e+04, threshold=4.260e+03, percent-clipped=12.0
+2023-03-09 11:10:50,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:11:15,868 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:11:18,362 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:11:20,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:11:30,958 INFO [train.py:968] (0/2) Epoch 18, batch 25500, giga_loss[loss=0.3451, simple_loss=0.3907, pruned_loss=0.1498, over 28800.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3699, pruned_loss=0.12, over 5677419.03 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3622, pruned_loss=0.113, over 5722274.85 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3705, pruned_loss=0.1204, over 5676737.35 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:11:37,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:11:46,576 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:12:01,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:12:19,379 INFO [train.py:968] (0/2) Epoch 18, batch 25550, giga_loss[loss=0.354, simple_loss=0.4119, pruned_loss=0.148, over 28984.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3734, pruned_loss=0.1233, over 5682611.43 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3624, pruned_loss=0.1131, over 5725165.68 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3738, pruned_loss=0.1236, over 5678947.25 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:12:22,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.681e+03 2.125e+03 2.898e+03 7.373e+03, threshold=4.249e+03, percent-clipped=7.0
+2023-03-09 11:13:01,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4231, 4.2313, 4.0251, 2.1438], device='cuda:0'), covar=tensor([0.0561, 0.0706, 0.0783, 0.2118], device='cuda:0'), in_proj_covar=tensor([0.1188, 0.1104, 0.0949, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 11:13:07,031 INFO [train.py:968] (0/2) Epoch 18, batch 25600, giga_loss[loss=0.4724, simple_loss=0.4748, pruned_loss=0.235, over 26605.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3747, pruned_loss=0.1255, over 5680932.98 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.363, pruned_loss=0.1137, over 5726931.98 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3748, pruned_loss=0.1256, over 5675260.95 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:13:08,839 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:13:37,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:13:56,331 INFO [train.py:968] (0/2) Epoch 18, batch 25650, giga_loss[loss=0.3133, simple_loss=0.3733, pruned_loss=0.1266, over 28692.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3743, pruned_loss=0.1262, over 5675926.92 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3632, pruned_loss=0.1139, over 5721505.93 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3744, pruned_loss=0.1263, over 5674904.05 frames. ], batch size: 242, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:13:56,637 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:14:00,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:14:01,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.277e+03 1.834e+03 2.630e+03 3.597e+03 8.669e+03, threshold=5.259e+03, percent-clipped=14.0
+2023-03-09 11:14:19,365 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:14:21,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:14:24,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:14:39,076 INFO [train.py:968] (0/2) Epoch 18, batch 25700, giga_loss[loss=0.3463, simple_loss=0.3956, pruned_loss=0.1484, over 27524.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3741, pruned_loss=0.1262, over 5675796.45 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3633, pruned_loss=0.114, over 5716120.06 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3744, pruned_loss=0.1265, over 5678698.35 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:14:45,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:15:20,235 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:15:24,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:15:25,950 INFO [train.py:968] (0/2) Epoch 18, batch 25750, giga_loss[loss=0.3996, simple_loss=0.4362, pruned_loss=0.1815, over 27581.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3737, pruned_loss=0.1261, over 5668648.57 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3635, pruned_loss=0.1141, over 5720459.56 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.374, pruned_loss=0.1266, over 5665892.17 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:15:29,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.597e+03 2.154e+03 2.983e+03 1.029e+04, threshold=4.307e+03, percent-clipped=4.0
+2023-03-09 11:15:45,672 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:15:47,970 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:15:47,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:16:01,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5814, 1.7533, 1.2301, 1.3074], device='cuda:0'), covar=tensor([0.0872, 0.0535, 0.1029, 0.1194], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0445, 0.0511, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 11:16:06,790 INFO [train.py:968] (0/2) Epoch 18, batch 25800, giga_loss[loss=0.2691, simple_loss=0.3507, pruned_loss=0.09368, over 29112.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.373, pruned_loss=0.1239, over 5670619.95 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3637, pruned_loss=0.1145, over 5715917.70 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3733, pruned_loss=0.1243, over 5671904.47 frames. ], batch size: 128, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:16:12,261 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:16:56,390 INFO [train.py:968] (0/2) Epoch 18, batch 25850, giga_loss[loss=0.295, simple_loss=0.336, pruned_loss=0.1269, over 23907.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3718, pruned_loss=0.1229, over 5654883.56 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3637, pruned_loss=0.1144, over 5717013.11 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3721, pruned_loss=0.1232, over 5654575.86 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:16:59,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.719e+03 2.311e+03 3.394e+03 1.117e+04, threshold=4.622e+03, percent-clipped=14.0
+2023-03-09 11:17:39,744 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=802610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:17:40,748 INFO [train.py:968] (0/2) Epoch 18, batch 25900, giga_loss[loss=0.2822, simple_loss=0.3534, pruned_loss=0.1055, over 28912.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3703, pruned_loss=0.1222, over 5646440.49 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3643, pruned_loss=0.115, over 5702394.29 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3701, pruned_loss=0.1221, over 5658902.99 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:17:48,325 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=802623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:18:24,707 INFO [train.py:968] (0/2) Epoch 18, batch 25950, giga_loss[loss=0.2754, simple_loss=0.3522, pruned_loss=0.09928, over 28927.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3687, pruned_loss=0.1218, over 5651793.87 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3642, pruned_loss=0.1151, over 5706400.59 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3687, pruned_loss=0.1218, over 5657089.57 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:18:30,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.232e+02 1.557e+03 2.032e+03 2.897e+03 7.402e+03, threshold=4.064e+03, percent-clipped=8.0
+2023-03-09 11:19:13,588 INFO [train.py:968] (0/2) Epoch 18, batch 26000, giga_loss[loss=0.3025, simple_loss=0.3714, pruned_loss=0.1168, over 28959.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3708, pruned_loss=0.1233, over 5642530.46 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3648, pruned_loss=0.1155, over 5696476.37 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3704, pruned_loss=0.1231, over 5653820.01 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:19:59,064 INFO [train.py:968] (0/2) Epoch 18, batch 26050, giga_loss[loss=0.3118, simple_loss=0.3875, pruned_loss=0.118, over 28913.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3742, pruned_loss=0.1245, over 5648983.28 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3651, pruned_loss=0.1157, over 5695472.80 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3738, pruned_loss=0.1242, over 5658135.42 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:20:04,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.721e+03 2.222e+03 3.409e+03 6.158e+03, threshold=4.444e+03, percent-clipped=13.0
+2023-03-09 11:20:21,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1523, 1.4338, 1.4126, 1.0251], device='cuda:0'), covar=tensor([0.1681, 0.2468, 0.1439, 0.1670], device='cuda:0'), in_proj_covar=tensor([0.0875, 0.0699, 0.0923, 0.0822], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 11:20:37,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1823, 4.0141, 3.7858, 2.1090], device='cuda:0'), covar=tensor([0.0604, 0.0775, 0.0799, 0.2000], device='cuda:0'), in_proj_covar=tensor([0.1184, 0.1102, 0.0944, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 11:20:43,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9751, 1.3666, 1.0895, 0.1907], device='cuda:0'), covar=tensor([0.3341, 0.2823, 0.3612, 0.5500], device='cuda:0'), in_proj_covar=tensor([0.1689, 0.1609, 0.1571, 0.1384], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 11:20:46,732 INFO [train.py:968] (0/2) Epoch 18, batch 26100, giga_loss[loss=0.328, simple_loss=0.3897, pruned_loss=0.1331, over 28335.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3779, pruned_loss=0.1246, over 5652896.03 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3654, pruned_loss=0.116, over 5694397.75 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3775, pruned_loss=0.1244, over 5659853.31 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:21:00,048 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=802826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:21:30,522 INFO [train.py:968] (0/2) Epoch 18, batch 26150, libri_loss[loss=0.2995, simple_loss=0.3612, pruned_loss=0.1189, over 29515.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3779, pruned_loss=0.1246, over 5654393.44 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3654, pruned_loss=0.116, over 5697772.30 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.378, pruned_loss=0.1246, over 5655660.73 frames. ], batch size: 80, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:21:34,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.799e+02 1.576e+03 2.096e+03 2.971e+03 1.214e+04, threshold=4.193e+03, percent-clipped=10.0
+2023-03-09 11:22:16,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=802910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:22:17,449 INFO [train.py:968] (0/2) Epoch 18, batch 26200, giga_loss[loss=0.3241, simple_loss=0.392, pruned_loss=0.1281, over 28933.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3803, pruned_loss=0.1267, over 5649293.46 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3656, pruned_loss=0.1162, over 5696195.68 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3805, pruned_loss=0.1268, over 5650895.46 frames. ], batch size: 227, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:22:58,451 INFO [train.py:968] (0/2) Epoch 18, batch 26250, giga_loss[loss=0.3583, simple_loss=0.3939, pruned_loss=0.1614, over 23391.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3805, pruned_loss=0.1275, over 5650857.70 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3653, pruned_loss=0.1159, over 5696726.56 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3814, pruned_loss=0.1281, over 5650255.78 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:23:06,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.159e+03 1.623e+03 2.105e+03 2.979e+03 6.081e+03, threshold=4.211e+03, percent-clipped=7.0
+2023-03-09 11:23:22,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:23:33,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:23:37,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0444, 4.8627, 4.5978, 2.2707], device='cuda:0'), covar=tensor([0.0441, 0.0594, 0.0690, 0.1895], device='cuda:0'), in_proj_covar=tensor([0.1184, 0.1104, 0.0947, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 11:23:44,299 INFO [train.py:968] (0/2) Epoch 18, batch 26300, giga_loss[loss=0.3013, simple_loss=0.3722, pruned_loss=0.1152, over 28888.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3783, pruned_loss=0.1264, over 5660450.61 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.365, pruned_loss=0.1157, over 5703443.86 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3797, pruned_loss=0.1275, over 5652267.52 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:24:33,314 INFO [train.py:968] (0/2) Epoch 18, batch 26350, giga_loss[loss=0.3803, simple_loss=0.4144, pruned_loss=0.1731, over 26647.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3758, pruned_loss=0.1253, over 5651178.25 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3648, pruned_loss=0.1156, over 5703718.23 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3771, pruned_loss=0.1263, over 5644103.31 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:24:38,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 1.692e+03 2.276e+03 3.253e+03 9.439e+03, threshold=4.552e+03, percent-clipped=14.0
+2023-03-09 11:25:02,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8040, 1.4318, 4.9278, 3.7066], device='cuda:0'), covar=tensor([0.1794, 0.2821, 0.0683, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0735, 0.0632, 0.0935, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 11:25:17,749 INFO [train.py:968] (0/2) Epoch 18, batch 26400, giga_loss[loss=0.286, simple_loss=0.3572, pruned_loss=0.1074, over 28968.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3741, pruned_loss=0.1248, over 5653198.27 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3652, pruned_loss=0.1158, over 5700333.54 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3752, pruned_loss=0.1257, over 5649264.75 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:25:36,206 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:25:40,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:25:49,587 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:25:52,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:26:00,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3103, 0.8186, 0.9575, 1.4422], device='cuda:0'), covar=tensor([0.0728, 0.0362, 0.0332, 0.0802], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0059, 0.0102], device='cuda:0')
+2023-03-09 11:26:02,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6471, 1.6623, 1.7030, 1.5537], device='cuda:0'), covar=tensor([0.2490, 0.2477, 0.1903, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.1899, 0.1825, 0.1770, 0.1911], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 11:26:07,342 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:26:08,325 INFO [train.py:968] (0/2) Epoch 18, batch 26450, giga_loss[loss=0.3242, simple_loss=0.3801, pruned_loss=0.1342, over 28642.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3737, pruned_loss=0.1255, over 5644579.50 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3652, pruned_loss=0.1157, over 5701495.55 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3747, pruned_loss=0.1265, over 5639561.09 frames. ], batch size: 78, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:26:13,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.707e+03 2.116e+03 3.127e+03 8.066e+03, threshold=4.233e+03, percent-clipped=7.0
+2023-03-09 11:26:17,866 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:26:42,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:26:50,443 INFO [train.py:968] (0/2) Epoch 18, batch 26500, giga_loss[loss=0.3571, simple_loss=0.4014, pruned_loss=0.1564, over 26511.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3742, pruned_loss=0.1261, over 5641437.02 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3649, pruned_loss=0.1157, over 5694801.16 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3755, pruned_loss=0.1272, over 5641642.98 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:27:28,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=803254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:27:33,454 INFO [train.py:968] (0/2) Epoch 18, batch 26550, giga_loss[loss=0.2807, simple_loss=0.3417, pruned_loss=0.1098, over 28165.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3725, pruned_loss=0.1251, over 5654869.99 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.365, pruned_loss=0.1159, over 5698223.80 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3737, pruned_loss=0.1261, over 5650149.07 frames. ], batch size: 77, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:27:40,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.771e+03 2.369e+03 3.257e+03 1.227e+04, threshold=4.737e+03, percent-clipped=12.0
+2023-03-09 11:27:54,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:28:16,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=803310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:28:17,417 INFO [train.py:968] (0/2) Epoch 18, batch 26600, giga_loss[loss=0.3662, simple_loss=0.4145, pruned_loss=0.159, over 28210.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3711, pruned_loss=0.1241, over 5673154.34 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3654, pruned_loss=0.1163, over 5705088.11 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.372, pruned_loss=0.1249, over 5661865.33 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:28:25,137 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-09 11:28:46,515 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:28:49,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:29:01,235 INFO [train.py:968] (0/2) Epoch 18, batch 26650, giga_loss[loss=0.2906, simple_loss=0.3632, pruned_loss=0.109, over 28677.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3722, pruned_loss=0.1247, over 5676025.26 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3655, pruned_loss=0.1165, over 5707273.59 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3729, pruned_loss=0.1253, over 5664612.94 frames. ], batch size: 92, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:29:09,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.645e+03 2.222e+03 3.189e+03 9.784e+03, threshold=4.445e+03, percent-clipped=8.0
+2023-03-09 11:29:14,932 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:29:49,553 INFO [train.py:968] (0/2) Epoch 18, batch 26700, giga_loss[loss=0.3595, simple_loss=0.404, pruned_loss=0.1575, over 27720.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3757, pruned_loss=0.1266, over 5669402.29 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3655, pruned_loss=0.1164, over 5709389.20 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3763, pruned_loss=0.1271, over 5658241.54 frames. ], batch size: 474, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:30:03,576 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:30:06,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:30:36,653 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:30:37,788 INFO [train.py:968] (0/2) Epoch 18, batch 26750, giga_loss[loss=0.2634, simple_loss=0.329, pruned_loss=0.09892, over 28729.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3738, pruned_loss=0.1253, over 5665828.33 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3654, pruned_loss=0.1164, over 5711980.62 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3745, pruned_loss=0.126, over 5653845.80 frames. ], batch size: 92, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:30:45,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.527e+02 1.681e+03 2.256e+03 3.436e+03 1.395e+04, threshold=4.512e+03, percent-clipped=17.0
+2023-03-09 11:30:46,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=803471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:31:02,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5984, 1.8718, 1.8533, 1.5059], device='cuda:0'), covar=tensor([0.1941, 0.2169, 0.2237, 0.2573], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0745, 0.0704, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 11:31:07,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3023, 1.3742, 1.2067, 1.2985], device='cuda:0'), covar=tensor([0.1689, 0.1709, 0.1575, 0.1511], device='cuda:0'), in_proj_covar=tensor([0.1909, 0.1830, 0.1779, 0.1920], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 11:31:19,319 INFO [train.py:968] (0/2) Epoch 18, batch 26800, giga_loss[loss=0.2689, simple_loss=0.3584, pruned_loss=0.08972, over 28541.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3744, pruned_loss=0.1248, over 5671777.00 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3651, pruned_loss=0.1161, over 5712625.89 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3756, pruned_loss=0.1259, over 5659702.48 frames. ], batch size: 65, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:31:34,230 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=803529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:31:41,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4885, 4.2495, 4.0460, 1.9669], device='cuda:0'), covar=tensor([0.0596, 0.0804, 0.0853, 0.2000], device='cuda:0'), in_proj_covar=tensor([0.1187, 0.1105, 0.0947, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 11:32:01,199 INFO [train.py:968] (0/2) Epoch 18, batch 26850, giga_loss[loss=0.3714, simple_loss=0.4304, pruned_loss=0.1562, over 28574.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3763, pruned_loss=0.1238, over 5674964.85 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3654, pruned_loss=0.1165, over 5708908.34 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3773, pruned_loss=0.1246, over 5668002.89 frames. ], batch size: 336, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:32:07,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.044e+02 1.439e+03 1.810e+03 2.369e+03 5.017e+03, threshold=3.619e+03, percent-clipped=3.0
+2023-03-09 11:32:45,940 INFO [train.py:968] (0/2) Epoch 18, batch 26900, giga_loss[loss=0.3082, simple_loss=0.3904, pruned_loss=0.113, over 29075.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3768, pruned_loss=0.1222, over 5679750.33 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3646, pruned_loss=0.116, over 5714235.49 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3786, pruned_loss=0.1235, over 5668497.43 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:32:58,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3455, 1.5282, 1.6266, 1.1805], device='cuda:0'), covar=tensor([0.1820, 0.2553, 0.1519, 0.1762], device='cuda:0'), in_proj_covar=tensor([0.0876, 0.0702, 0.0925, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 11:33:00,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:33:28,044 INFO [train.py:968] (0/2) Epoch 18, batch 26950, giga_loss[loss=0.3092, simple_loss=0.3723, pruned_loss=0.1231, over 28947.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3792, pruned_loss=0.1238, over 5681162.95 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3647, pruned_loss=0.116, over 5715054.36 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3807, pruned_loss=0.1249, over 5670835.55 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:33:36,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.774e+02 1.459e+03 1.924e+03 2.568e+03 9.449e+03, threshold=3.848e+03, percent-clipped=6.0
+2023-03-09 11:33:51,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:34:18,738 INFO [train.py:968] (0/2) Epoch 18, batch 27000, libri_loss[loss=0.32, simple_loss=0.3803, pruned_loss=0.1298, over 29522.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3826, pruned_loss=0.1277, over 5671495.63 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3648, pruned_loss=0.1161, over 5706595.09 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3839, pruned_loss=0.1286, over 5670374.76 frames. ], batch size: 81, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:34:18,743 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 11:34:28,239 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2082, simple_loss=0.3149, pruned_loss=0.05074, over 944034.00 frames. 
+2023-03-09 11:34:28,240 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 11:35:17,847 INFO [train.py:968] (0/2) Epoch 18, batch 27050, giga_loss[loss=0.2981, simple_loss=0.3639, pruned_loss=0.1161, over 28948.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3832, pruned_loss=0.1289, over 5680206.75 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3651, pruned_loss=0.1162, over 5709301.56 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3843, pruned_loss=0.1297, over 5676371.98 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:35:23,313 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.462e+02 1.739e+03 2.268e+03 3.239e+03 8.177e+03, threshold=4.536e+03, percent-clipped=13.0
+2023-03-09 11:35:25,096 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:35:28,092 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:35:54,951 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=803801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:35:56,794 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:36:05,166 INFO [train.py:968] (0/2) Epoch 18, batch 27100, giga_loss[loss=0.3297, simple_loss=0.3958, pruned_loss=0.1318, over 28910.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3832, pruned_loss=0.1294, over 5670221.85 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.365, pruned_loss=0.1162, over 5709339.98 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3845, pruned_loss=0.1303, over 5666488.12 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:36:23,179 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:36:25,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:36:39,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:36:49,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:36:50,963 INFO [train.py:968] (0/2) Epoch 18, batch 27150, libri_loss[loss=0.2371, simple_loss=0.3071, pruned_loss=0.08358, over 29386.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3799, pruned_loss=0.1256, over 5676452.15 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3648, pruned_loss=0.116, over 5710905.10 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3815, pruned_loss=0.1268, over 5671023.98 frames. ], batch size: 67, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:36:58,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.590e+03 2.167e+03 3.319e+03 1.096e+04, threshold=4.334e+03, percent-clipped=15.0
+2023-03-09 11:37:29,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:37:37,117 INFO [train.py:968] (0/2) Epoch 18, batch 27200, giga_loss[loss=0.356, simple_loss=0.4192, pruned_loss=0.1464, over 28329.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3812, pruned_loss=0.1254, over 5664993.34 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3649, pruned_loss=0.1162, over 5713310.11 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3826, pruned_loss=0.1263, over 5658118.60 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:38:23,087 INFO [train.py:968] (0/2) Epoch 18, batch 27250, giga_loss[loss=0.3645, simple_loss=0.4142, pruned_loss=0.1575, over 28617.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3825, pruned_loss=0.1259, over 5674784.15 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3655, pruned_loss=0.1164, over 5715431.37 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3838, pruned_loss=0.1268, over 5665886.33 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:38:31,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.866e+02 1.576e+03 2.006e+03 2.602e+03 9.914e+03, threshold=4.011e+03, percent-clipped=4.0
+2023-03-09 11:38:50,124 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:38:53,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:38:59,921 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-804000.pt
+2023-03-09 11:39:10,520 INFO [train.py:968] (0/2) Epoch 18, batch 27300, giga_loss[loss=0.4146, simple_loss=0.4441, pruned_loss=0.1925, over 27584.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3827, pruned_loss=0.1268, over 5663388.19 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3653, pruned_loss=0.1164, over 5717382.94 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3841, pruned_loss=0.1276, over 5653765.63 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:39:20,496 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 11:39:21,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:39:39,716 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=804045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:39:41,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:39:45,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:39:55,609 INFO [train.py:968] (0/2) Epoch 18, batch 27350, giga_loss[loss=0.314, simple_loss=0.373, pruned_loss=0.1275, over 28764.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3801, pruned_loss=0.1254, over 5668901.71 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3648, pruned_loss=0.116, over 5720425.81 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3821, pruned_loss=0.1266, over 5657352.74 frames. ], batch size: 284, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:40:04,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.985e+02 1.672e+03 2.087e+03 2.735e+03 5.472e+03, threshold=4.174e+03, percent-clipped=9.0
+2023-03-09 11:40:13,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:40:21,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4945, 1.6270, 1.5274, 1.4722], device='cuda:0'), covar=tensor([0.1598, 0.2183, 0.2181, 0.2013], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0740, 0.0700, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 11:40:45,983 INFO [train.py:968] (0/2) Epoch 18, batch 27400, giga_loss[loss=0.291, simple_loss=0.366, pruned_loss=0.108, over 28929.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3782, pruned_loss=0.1251, over 5659122.32 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3646, pruned_loss=0.116, over 5703999.81 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3803, pruned_loss=0.1263, over 5662160.54 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:41:15,152 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 11:41:35,940 INFO [train.py:968] (0/2) Epoch 18, batch 27450, giga_loss[loss=0.2914, simple_loss=0.3606, pruned_loss=0.1111, over 29017.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3757, pruned_loss=0.1237, over 5663178.98 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3648, pruned_loss=0.116, over 5705110.82 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3773, pruned_loss=0.1247, over 5664499.46 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:41:44,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.762e+03 2.281e+03 3.538e+03 7.443e+03, threshold=4.562e+03, percent-clipped=13.0
+2023-03-09 11:41:52,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:42:26,022 INFO [train.py:968] (0/2) Epoch 18, batch 27500, giga_loss[loss=0.2899, simple_loss=0.3547, pruned_loss=0.1126, over 28701.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3734, pruned_loss=0.1229, over 5661105.23 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3647, pruned_loss=0.116, over 5706193.26 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3747, pruned_loss=0.1237, over 5661006.06 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:43:08,316 INFO [train.py:968] (0/2) Epoch 18, batch 27550, giga_loss[loss=0.2832, simple_loss=0.3564, pruned_loss=0.105, over 28835.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3737, pruned_loss=0.124, over 5667038.44 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3648, pruned_loss=0.1161, over 5710410.79 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3749, pruned_loss=0.1248, over 5662056.84 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:43:16,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.716e+03 2.381e+03 3.462e+03 1.145e+04, threshold=4.762e+03, percent-clipped=10.0
+2023-03-09 11:43:27,658 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=804282.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 11:43:47,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6356, 1.7233, 1.2697, 1.3263], device='cuda:0'), covar=tensor([0.0902, 0.0620, 0.1070, 0.1204], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0445, 0.0511, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 11:43:49,615 INFO [train.py:968] (0/2) Epoch 18, batch 27600, giga_loss[loss=0.2741, simple_loss=0.3483, pruned_loss=0.09994, over 28953.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.372, pruned_loss=0.1229, over 5662201.14 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3655, pruned_loss=0.1168, over 5707211.77 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3726, pruned_loss=0.1232, over 5660049.66 frames. ], batch size: 227, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:43:51,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4039, 1.5329, 1.6425, 1.2559], device='cuda:0'), covar=tensor([0.1668, 0.2479, 0.1401, 0.1662], device='cuda:0'), in_proj_covar=tensor([0.0877, 0.0703, 0.0926, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 11:43:55,457 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:43:58,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:44:23,449 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 11:44:23,942 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:44:26,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3899, 1.3419, 4.0357, 3.1691], device='cuda:0'), covar=tensor([0.1594, 0.2713, 0.0483, 0.1195], device='cuda:0'), in_proj_covar=tensor([0.0736, 0.0633, 0.0938, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 11:44:31,071 INFO [train.py:968] (0/2) Epoch 18, batch 27650, giga_loss[loss=0.3091, simple_loss=0.3795, pruned_loss=0.1194, over 27613.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3695, pruned_loss=0.1197, over 5667511.63 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3656, pruned_loss=0.1168, over 5713593.12 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3701, pruned_loss=0.12, over 5658478.51 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:44:35,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7304, 1.1059, 4.9434, 3.7565], device='cuda:0'), covar=tensor([0.1586, 0.3078, 0.0365, 0.0818], device='cuda:0'), in_proj_covar=tensor([0.0736, 0.0635, 0.0939, 0.0873], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 11:44:40,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.261e+02 1.402e+03 2.023e+03 3.043e+03 9.687e+03, threshold=4.046e+03, percent-clipped=7.0
+2023-03-09 11:45:17,480 INFO [train.py:968] (0/2) Epoch 18, batch 27700, giga_loss[loss=0.294, simple_loss=0.3643, pruned_loss=0.1118, over 28883.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.366, pruned_loss=0.1167, over 5666810.85 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3654, pruned_loss=0.1168, over 5716704.33 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3667, pruned_loss=0.117, over 5656009.75 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:45:24,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:45:29,864 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=804423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:45:58,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=804451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:46:06,835 INFO [train.py:968] (0/2) Epoch 18, batch 27750, giga_loss[loss=0.3403, simple_loss=0.3908, pruned_loss=0.1448, over 28331.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3652, pruned_loss=0.1164, over 5665048.59 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3651, pruned_loss=0.1165, over 5716690.14 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.366, pruned_loss=0.1168, over 5655548.66 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:46:17,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.084e+02 1.578e+03 2.117e+03 2.774e+03 8.827e+03, threshold=4.233e+03, percent-clipped=10.0
+2023-03-09 11:46:51,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5308, 1.7756, 1.6692, 1.4901], device='cuda:0'), covar=tensor([0.2675, 0.2307, 0.2404, 0.2426], device='cuda:0'), in_proj_covar=tensor([0.1910, 0.1834, 0.1778, 0.1917], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 11:46:53,042 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-09 11:46:56,269 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=804509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:46:57,848 INFO [train.py:968] (0/2) Epoch 18, batch 27800, giga_loss[loss=0.3189, simple_loss=0.3765, pruned_loss=0.1306, over 28974.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3622, pruned_loss=0.1156, over 5652571.10 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3651, pruned_loss=0.1165, over 5711620.03 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3628, pruned_loss=0.116, over 5647473.00 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:47:17,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3272, 2.6696, 2.3813, 1.9029], device='cuda:0'), covar=tensor([0.2558, 0.1922, 0.2213, 0.2696], device='cuda:0'), in_proj_covar=tensor([0.1905, 0.1830, 0.1773, 0.1912], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 11:47:45,493 INFO [train.py:968] (0/2) Epoch 18, batch 27850, giga_loss[loss=0.322, simple_loss=0.3921, pruned_loss=0.126, over 28619.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3631, pruned_loss=0.1164, over 5645838.28 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3653, pruned_loss=0.1166, over 5705935.01 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3633, pruned_loss=0.1166, over 5645791.53 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:47:46,372 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:47:49,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:47:52,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.161e+03 1.905e+03 2.547e+03 3.334e+03 7.435e+03, threshold=5.094e+03, percent-clipped=10.0
+2023-03-09 11:48:13,745 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:48:28,777 INFO [train.py:968] (0/2) Epoch 18, batch 27900, libri_loss[loss=0.327, simple_loss=0.3921, pruned_loss=0.1309, over 29364.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3657, pruned_loss=0.1173, over 5653272.77 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3654, pruned_loss=0.1167, over 5704123.95 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3657, pruned_loss=0.1173, over 5652119.79 frames. ], batch size: 92, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:49:12,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804657.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 11:49:16,087 INFO [train.py:968] (0/2) Epoch 18, batch 27950, giga_loss[loss=0.2648, simple_loss=0.3422, pruned_loss=0.09373, over 28905.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3676, pruned_loss=0.1184, over 5650435.42 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3655, pruned_loss=0.1168, over 5705855.97 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3676, pruned_loss=0.1184, over 5647753.53 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:49:25,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.776e+02 1.591e+03 2.097e+03 2.962e+03 6.966e+03, threshold=4.194e+03, percent-clipped=5.0
+2023-03-09 11:50:03,792 INFO [train.py:968] (0/2) Epoch 18, batch 28000, giga_loss[loss=0.2971, simple_loss=0.3612, pruned_loss=0.1165, over 28745.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3682, pruned_loss=0.1194, over 5650694.17 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3658, pruned_loss=0.117, over 5708510.94 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3679, pruned_loss=0.1192, over 5645274.37 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:50:44,263 INFO [train.py:968] (0/2) Epoch 18, batch 28050, giga_loss[loss=0.2685, simple_loss=0.343, pruned_loss=0.09701, over 28439.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3692, pruned_loss=0.1207, over 5652598.77 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3667, pruned_loss=0.1175, over 5708012.22 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3683, pruned_loss=0.1202, over 5646708.84 frames. ], batch size: 65, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:50:52,369 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.476e+03 1.962e+03 2.697e+03 6.893e+03, threshold=3.923e+03, percent-clipped=6.0
+2023-03-09 11:50:56,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1822, 1.3709, 1.2610, 1.1064], device='cuda:0'), covar=tensor([0.1777, 0.1892, 0.1346, 0.1811], device='cuda:0'), in_proj_covar=tensor([0.1901, 0.1833, 0.1770, 0.1910], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 11:51:17,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:51:18,608 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804800.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 11:51:21,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804803.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 11:51:30,018 INFO [train.py:968] (0/2) Epoch 18, batch 28100, giga_loss[loss=0.3111, simple_loss=0.3774, pruned_loss=0.1224, over 28672.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.371, pruned_loss=0.1219, over 5652128.63 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3668, pruned_loss=0.1176, over 5711444.67 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3702, pruned_loss=0.1215, over 5643354.24 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:51:35,567 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.33 vs. limit=5.0
+2023-03-09 11:51:41,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:51:46,713 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804832.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 11:51:59,216 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-09 11:52:14,812 INFO [train.py:968] (0/2) Epoch 18, batch 28150, giga_loss[loss=0.276, simple_loss=0.3503, pruned_loss=0.1008, over 28450.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3728, pruned_loss=0.1228, over 5659506.02 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3668, pruned_loss=0.1175, over 5712019.65 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3723, pruned_loss=0.1226, over 5651440.44 frames. ], batch size: 71, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:52:26,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.418e+02 1.655e+03 1.972e+03 2.593e+03 5.370e+03, threshold=3.943e+03, percent-clipped=12.0
+2023-03-09 11:52:38,675 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:53:05,444 INFO [train.py:968] (0/2) Epoch 18, batch 28200, giga_loss[loss=0.3838, simple_loss=0.4113, pruned_loss=0.1782, over 23165.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3753, pruned_loss=0.1255, over 5643234.64 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3671, pruned_loss=0.1178, over 5708046.79 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3747, pruned_loss=0.1252, over 5639452.85 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:53:32,252 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:53:35,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:53:54,268 INFO [train.py:968] (0/2) Epoch 18, batch 28250, giga_loss[loss=0.2849, simple_loss=0.3632, pruned_loss=0.1034, over 28922.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3756, pruned_loss=0.1262, over 5636701.32 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3671, pruned_loss=0.1178, over 5699308.39 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3753, pruned_loss=0.1261, over 5640395.67 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:54:02,874 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:54:06,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:54:07,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.189e+03 1.897e+03 2.620e+03 4.040e+03 1.319e+04, threshold=5.241e+03, percent-clipped=25.0
+2023-03-09 11:54:07,630 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:54:34,739 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=805001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:54:45,343 INFO [train.py:968] (0/2) Epoch 18, batch 28300, giga_loss[loss=0.2961, simple_loss=0.3723, pruned_loss=0.11, over 28616.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3746, pruned_loss=0.1236, over 5644237.11 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3669, pruned_loss=0.1176, over 5703515.88 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3747, pruned_loss=0.1238, over 5642195.77 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:55:00,666 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=805027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:55:03,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=805030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:55:23,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3663, 1.3535, 4.0493, 3.3878], device='cuda:0'), covar=tensor([0.1547, 0.2744, 0.0428, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0633, 0.0932, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 11:55:32,790 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=805059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:55:34,544 INFO [train.py:968] (0/2) Epoch 18, batch 28350, libri_loss[loss=0.2643, simple_loss=0.3322, pruned_loss=0.09817, over 29566.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3742, pruned_loss=0.1239, over 5636480.57 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3665, pruned_loss=0.1177, over 5699813.05 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3749, pruned_loss=0.1241, over 5635734.84 frames. ], batch size: 75, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:55:45,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.713e+03 2.107e+03 3.012e+03 2.016e+04, threshold=4.215e+03, percent-clipped=7.0
+2023-03-09 11:56:23,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5388, 2.5913, 1.6376, 1.6036], device='cuda:0'), covar=tensor([0.0754, 0.0333, 0.0674, 0.1061], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0546, 0.0372, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 11:56:24,133 INFO [train.py:968] (0/2) Epoch 18, batch 28400, giga_loss[loss=0.288, simple_loss=0.3579, pruned_loss=0.1091, over 28899.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3745, pruned_loss=0.1246, over 5630893.97 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3664, pruned_loss=0.1177, over 5700831.99 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3751, pruned_loss=0.1249, over 5629215.88 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:57:06,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6967, 1.8894, 1.5083, 1.8737], device='cuda:0'), covar=tensor([0.2449, 0.2565, 0.2822, 0.2457], device='cuda:0'), in_proj_covar=tensor([0.1451, 0.1054, 0.1286, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 11:57:17,197 INFO [train.py:968] (0/2) Epoch 18, batch 28450, giga_loss[loss=0.2898, simple_loss=0.3556, pruned_loss=0.112, over 28835.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3738, pruned_loss=0.1247, over 5636272.20 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3668, pruned_loss=0.1179, over 5705466.56 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3742, pruned_loss=0.1249, over 5628981.23 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:57:32,413 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.541e+03 2.157e+03 2.932e+03 7.842e+03, threshold=4.315e+03, percent-clipped=9.0
+2023-03-09 11:57:59,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6821, 2.3184, 1.5377, 1.0285], device='cuda:0'), covar=tensor([0.6801, 0.3586, 0.3307, 0.5839], device='cuda:0'), in_proj_covar=tensor([0.1702, 0.1608, 0.1575, 0.1384], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 11:58:09,607 INFO [train.py:968] (0/2) Epoch 18, batch 28500, giga_loss[loss=0.3556, simple_loss=0.405, pruned_loss=0.1531, over 27944.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3733, pruned_loss=0.1254, over 5629222.91 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3668, pruned_loss=0.1179, over 5705531.69 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3738, pruned_loss=0.1257, over 5621784.05 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:58:21,735 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3847, 1.5141, 1.1518, 1.0913], device='cuda:0'), covar=tensor([0.0888, 0.0530, 0.1028, 0.1072], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0448, 0.0514, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 11:58:35,023 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3543, 1.4813, 3.5240, 3.2461], device='cuda:0'), covar=tensor([0.1337, 0.2363, 0.0438, 0.0982], device='cuda:0'), in_proj_covar=tensor([0.0735, 0.0633, 0.0932, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 11:58:52,197 INFO [train.py:968] (0/2) Epoch 18, batch 28550, giga_loss[loss=0.343, simple_loss=0.4024, pruned_loss=0.1418, over 29005.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3726, pruned_loss=0.1246, over 5629457.71 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3673, pruned_loss=0.1182, over 5686113.57 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3727, pruned_loss=0.1248, over 5639610.72 frames. ], batch size: 128, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:59:02,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.099e+03 1.506e+03 2.042e+03 2.660e+03 7.751e+03, threshold=4.083e+03, percent-clipped=5.0
+2023-03-09 11:59:39,968 INFO [train.py:968] (0/2) Epoch 18, batch 28600, giga_loss[loss=0.2998, simple_loss=0.3634, pruned_loss=0.1181, over 28907.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3723, pruned_loss=0.1249, over 5634557.69 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.367, pruned_loss=0.1181, over 5690375.22 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3727, pruned_loss=0.1253, over 5637682.26 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:00:07,758 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3337, 1.2311, 3.7601, 3.3053], device='cuda:0'), covar=tensor([0.1517, 0.2721, 0.0465, 0.1961], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0631, 0.0932, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 12:00:25,821 INFO [train.py:968] (0/2) Epoch 18, batch 28650, giga_loss[loss=0.2804, simple_loss=0.3486, pruned_loss=0.1062, over 28987.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3734, pruned_loss=0.1255, over 5639188.59 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3674, pruned_loss=0.1183, over 5683793.67 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3735, pruned_loss=0.1258, over 5646248.81 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:00:37,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.671e+03 2.187e+03 3.115e+03 1.248e+04, threshold=4.375e+03, percent-clipped=17.0
+2023-03-09 12:00:54,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1337, 2.9867, 1.2736, 1.4112], device='cuda:0'), covar=tensor([0.1172, 0.0428, 0.1014, 0.1477], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0545, 0.0372, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 12:01:11,017 INFO [train.py:968] (0/2) Epoch 18, batch 28700, giga_loss[loss=0.3138, simple_loss=0.3728, pruned_loss=0.1273, over 28791.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3734, pruned_loss=0.1255, over 5640933.81 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3673, pruned_loss=0.1183, over 5676175.89 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3737, pruned_loss=0.1258, over 5652560.29 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:01:31,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7131, 2.7244, 2.7077, 2.3549], device='cuda:0'), covar=tensor([0.1891, 0.2362, 0.1927, 0.2191], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0750, 0.0709, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 12:01:58,704 INFO [train.py:968] (0/2) Epoch 18, batch 28750, giga_loss[loss=0.304, simple_loss=0.3696, pruned_loss=0.1192, over 28988.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3755, pruned_loss=0.127, over 5648373.07 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3673, pruned_loss=0.1183, over 5679662.85 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3758, pruned_loss=0.1274, over 5653816.48 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:02:11,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.870e+02 1.672e+03 2.003e+03 2.560e+03 5.968e+03, threshold=4.007e+03, percent-clipped=3.0
+2023-03-09 12:02:36,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6328, 1.6186, 1.2707, 1.2467], device='cuda:0'), covar=tensor([0.0734, 0.0467, 0.0866, 0.1151], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0448, 0.0515, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 12:02:43,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4488, 1.5593, 1.4958, 1.2855], device='cuda:0'), covar=tensor([0.2046, 0.2151, 0.1606, 0.1985], device='cuda:0'), in_proj_covar=tensor([0.1913, 0.1842, 0.1780, 0.1913], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 12:02:44,110 INFO [train.py:968] (0/2) Epoch 18, batch 28800, giga_loss[loss=0.3121, simple_loss=0.3712, pruned_loss=0.1265, over 28906.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3763, pruned_loss=0.1278, over 5650495.87 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3676, pruned_loss=0.1186, over 5673717.08 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3765, pruned_loss=0.1282, over 5660116.08 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:02:51,930 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-09 12:03:02,776 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 12:03:28,739 INFO [train.py:968] (0/2) Epoch 18, batch 28850, giga_loss[loss=0.3312, simple_loss=0.3853, pruned_loss=0.1386, over 28503.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3761, pruned_loss=0.1279, over 5655795.35 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3679, pruned_loss=0.119, over 5672840.80 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3763, pruned_loss=0.1281, over 5664076.18 frames. ], batch size: 336, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:03:40,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.655e+03 2.031e+03 2.731e+03 7.909e+03, threshold=4.061e+03, percent-clipped=11.0
+2023-03-09 12:04:17,422 INFO [train.py:968] (0/2) Epoch 18, batch 28900, giga_loss[loss=0.344, simple_loss=0.3986, pruned_loss=0.1447, over 28719.00 frames. ], tot_loss[loss=0.315, simple_loss=0.376, pruned_loss=0.127, over 5667818.50 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3679, pruned_loss=0.119, over 5674683.27 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3763, pruned_loss=0.1272, over 5672571.68 frames. ], batch size: 284, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:04:33,635 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 12:05:04,369 INFO [train.py:968] (0/2) Epoch 18, batch 28950, giga_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.116, over 28271.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3768, pruned_loss=0.1278, over 5659987.55 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3682, pruned_loss=0.1191, over 5676271.14 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.377, pruned_loss=0.1281, over 5661883.80 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:05:14,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.649e+03 2.220e+03 2.804e+03 5.438e+03, threshold=4.440e+03, percent-clipped=5.0
+2023-03-09 12:05:48,914 INFO [train.py:968] (0/2) Epoch 18, batch 29000, giga_loss[loss=0.2733, simple_loss=0.3387, pruned_loss=0.104, over 28551.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3773, pruned_loss=0.1284, over 5666949.71 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3684, pruned_loss=0.1196, over 5676997.22 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3776, pruned_loss=0.1285, over 5667406.99 frames. ], batch size: 78, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:06:35,312 INFO [train.py:968] (0/2) Epoch 18, batch 29050, giga_loss[loss=0.3329, simple_loss=0.3876, pruned_loss=0.1391, over 28829.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3789, pruned_loss=0.1294, over 5663676.56 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3685, pruned_loss=0.1195, over 5679083.09 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3791, pruned_loss=0.1296, over 5662054.35 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:06:46,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.347e+03 1.904e+03 2.493e+03 3.969e+03 1.209e+04, threshold=4.986e+03, percent-clipped=18.0
+2023-03-09 12:07:05,985 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 12:07:21,750 INFO [train.py:968] (0/2) Epoch 18, batch 29100, giga_loss[loss=0.3308, simple_loss=0.3973, pruned_loss=0.1322, over 29030.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3792, pruned_loss=0.1296, over 5657355.12 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3689, pruned_loss=0.1198, over 5671862.10 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3792, pruned_loss=0.1296, over 5662648.25 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:07:45,998 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-09 12:08:04,763 INFO [train.py:968] (0/2) Epoch 18, batch 29150, giga_loss[loss=0.2757, simple_loss=0.3614, pruned_loss=0.09495, over 28943.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3782, pruned_loss=0.1281, over 5638075.29 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3687, pruned_loss=0.1198, over 5669408.03 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3789, pruned_loss=0.1286, over 5642766.44 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:08:18,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.893e+02 1.684e+03 2.142e+03 2.796e+03 9.807e+03, threshold=4.285e+03, percent-clipped=8.0
+2023-03-09 12:08:53,871 INFO [train.py:968] (0/2) Epoch 18, batch 29200, libri_loss[loss=0.3432, simple_loss=0.3997, pruned_loss=0.1434, over 26189.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3781, pruned_loss=0.1274, over 5639881.89 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3682, pruned_loss=0.1196, over 5670572.14 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3792, pruned_loss=0.1281, over 5642219.67 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:09:36,990 INFO [train.py:968] (0/2) Epoch 18, batch 29250, giga_loss[loss=0.3284, simple_loss=0.3855, pruned_loss=0.1357, over 28716.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.376, pruned_loss=0.1252, over 5648442.80 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3676, pruned_loss=0.1192, over 5672758.49 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3776, pruned_loss=0.1263, over 5647793.85 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:09:49,194 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.122e+03 1.620e+03 1.879e+03 3.082e+03 8.511e+03, threshold=3.757e+03, percent-clipped=9.0
+2023-03-09 12:10:09,669 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-806000.pt
+2023-03-09 12:10:19,758 INFO [train.py:968] (0/2) Epoch 18, batch 29300, giga_loss[loss=0.3373, simple_loss=0.3919, pruned_loss=0.1413, over 27960.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3744, pruned_loss=0.1245, over 5654609.08 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3672, pruned_loss=0.1191, over 5675217.93 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3763, pruned_loss=0.1257, over 5651353.26 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:11:03,680 INFO [train.py:968] (0/2) Epoch 18, batch 29350, giga_loss[loss=0.3635, simple_loss=0.4027, pruned_loss=0.1621, over 26584.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3743, pruned_loss=0.1246, over 5660333.66 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3673, pruned_loss=0.1194, over 5680505.91 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3759, pruned_loss=0.1253, over 5652615.19 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:11:18,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.620e+03 2.011e+03 2.955e+03 5.451e+03, threshold=4.023e+03, percent-clipped=9.0
+2023-03-09 12:11:29,950 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 12:11:51,252 INFO [train.py:968] (0/2) Epoch 18, batch 29400, giga_loss[loss=0.353, simple_loss=0.4024, pruned_loss=0.1518, over 28878.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3758, pruned_loss=0.1257, over 5657455.75 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3676, pruned_loss=0.1197, over 5677059.77 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3771, pruned_loss=0.1262, over 5654318.27 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:12:39,499 INFO [train.py:968] (0/2) Epoch 18, batch 29450, giga_loss[loss=0.3358, simple_loss=0.3867, pruned_loss=0.1425, over 28835.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3764, pruned_loss=0.127, over 5653967.82 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3676, pruned_loss=0.1197, over 5676610.60 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3776, pruned_loss=0.1276, over 5651482.08 frames. ], batch size: 112, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:12:52,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.564e+03 1.910e+03 2.534e+03 5.131e+03, threshold=3.821e+03, percent-clipped=3.0
+2023-03-09 12:13:23,410 INFO [train.py:968] (0/2) Epoch 18, batch 29500, giga_loss[loss=0.3918, simple_loss=0.4252, pruned_loss=0.1792, over 27868.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3766, pruned_loss=0.1275, over 5665559.37 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3675, pruned_loss=0.1193, over 5682855.20 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3779, pruned_loss=0.1285, over 5657484.38 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:14:06,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1890, 1.2924, 1.2010, 1.1243], device='cuda:0'), covar=tensor([0.1695, 0.1834, 0.1310, 0.1611], device='cuda:0'), in_proj_covar=tensor([0.1911, 0.1841, 0.1767, 0.1911], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 12:14:08,188 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=806261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:14:08,758 INFO [train.py:968] (0/2) Epoch 18, batch 29550, giga_loss[loss=0.2676, simple_loss=0.3408, pruned_loss=0.0972, over 28697.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3771, pruned_loss=0.1277, over 5665425.76 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3675, pruned_loss=0.1193, over 5679267.12 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3783, pruned_loss=0.1287, over 5661863.52 frames. ], batch size: 85, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:14:19,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2393, 3.2710, 1.3813, 1.4232], device='cuda:0'), covar=tensor([0.1073, 0.0382, 0.0958, 0.1426], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0546, 0.0372, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 12:14:21,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.991e+02 1.615e+03 2.225e+03 3.393e+03 6.752e+03, threshold=4.451e+03, percent-clipped=11.0
+2023-03-09 12:14:57,912 INFO [train.py:968] (0/2) Epoch 18, batch 29600, giga_loss[loss=0.2783, simple_loss=0.3501, pruned_loss=0.1032, over 28922.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3787, pruned_loss=0.1292, over 5648510.17 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3679, pruned_loss=0.1195, over 5678977.45 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3795, pruned_loss=0.1299, over 5645820.53 frames. ], batch size: 112, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:15:28,743 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-09 12:15:44,758 INFO [train.py:968] (0/2) Epoch 18, batch 29650, giga_loss[loss=0.3606, simple_loss=0.3899, pruned_loss=0.1656, over 23483.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3781, pruned_loss=0.1288, over 5647712.51 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3677, pruned_loss=0.1194, over 5680040.96 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.379, pruned_loss=0.1295, over 5644503.47 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:15:56,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.134e+03 1.627e+03 2.151e+03 2.757e+03 6.093e+03, threshold=4.303e+03, percent-clipped=4.0
+2023-03-09 12:16:20,194 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=806401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:16:28,225 INFO [train.py:968] (0/2) Epoch 18, batch 29700, giga_loss[loss=0.2935, simple_loss=0.3663, pruned_loss=0.1103, over 28893.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3765, pruned_loss=0.1271, over 5647774.33 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3671, pruned_loss=0.1189, over 5681350.39 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3781, pruned_loss=0.1284, over 5643354.84 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:17:15,250 INFO [train.py:968] (0/2) Epoch 18, batch 29750, giga_loss[loss=0.2836, simple_loss=0.3561, pruned_loss=0.1056, over 28997.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3768, pruned_loss=0.1263, over 5660909.19 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3672, pruned_loss=0.1189, over 5684265.13 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3781, pruned_loss=0.1274, over 5654447.02 frames. ], batch size: 128, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:17:24,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6514, 1.7060, 1.8802, 1.4401], device='cuda:0'), covar=tensor([0.1882, 0.2515, 0.1496, 0.1756], device='cuda:0'), in_proj_covar=tensor([0.0878, 0.0703, 0.0927, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 12:17:29,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.640e+03 2.099e+03 2.866e+03 1.111e+04, threshold=4.197e+03, percent-clipped=12.0
+2023-03-09 12:17:54,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3074, 1.1043, 4.1367, 3.3361], device='cuda:0'), covar=tensor([0.1641, 0.2902, 0.0436, 0.1348], device='cuda:0'), in_proj_covar=tensor([0.0738, 0.0636, 0.0940, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 12:18:02,420 INFO [train.py:968] (0/2) Epoch 18, batch 29800, giga_loss[loss=0.2701, simple_loss=0.3465, pruned_loss=0.09687, over 29026.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3754, pruned_loss=0.1252, over 5650340.70 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.367, pruned_loss=0.1189, over 5678975.15 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3768, pruned_loss=0.1263, over 5649348.86 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:18:18,240 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=806530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:18:34,849 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5493, 1.7026, 1.5995, 1.4143], device='cuda:0'), covar=tensor([0.2845, 0.2259, 0.1909, 0.2439], device='cuda:0'), in_proj_covar=tensor([0.1915, 0.1843, 0.1772, 0.1915], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 12:18:45,519 INFO [train.py:968] (0/2) Epoch 18, batch 29850, giga_loss[loss=0.3684, simple_loss=0.4113, pruned_loss=0.1628, over 28921.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3755, pruned_loss=0.1253, over 5668986.50 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3672, pruned_loss=0.1188, over 5684120.52 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3767, pruned_loss=0.1264, over 5663066.16 frames. ], batch size: 227, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:18:58,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.996e+03 3.008e+03 4.118e+03 1.582e+04, threshold=6.017e+03, percent-clipped=22.0
+2023-03-09 12:19:31,424 INFO [train.py:968] (0/2) Epoch 18, batch 29900, giga_loss[loss=0.2603, simple_loss=0.3288, pruned_loss=0.09593, over 29093.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.373, pruned_loss=0.1243, over 5664255.77 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3667, pruned_loss=0.1186, over 5677979.46 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3744, pruned_loss=0.1254, over 5664392.83 frames. ], batch size: 113, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:19:51,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=806636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:20:03,615 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-09 12:20:06,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 12:20:18,843 INFO [train.py:968] (0/2) Epoch 18, batch 29950, giga_loss[loss=0.2755, simple_loss=0.3401, pruned_loss=0.1054, over 28982.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3686, pruned_loss=0.1223, over 5651720.05 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3664, pruned_loss=0.1183, over 5682644.90 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3701, pruned_loss=0.1236, over 5647218.67 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:20:33,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.643e+03 2.345e+03 3.079e+03 1.540e+04, threshold=4.690e+03, percent-clipped=5.0
+2023-03-09 12:20:34,292 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6897, 1.7384, 1.3435, 1.3344], device='cuda:0'), covar=tensor([0.0890, 0.0579, 0.1000, 0.1098], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0449, 0.0515, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 12:21:01,700 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 12:21:01,870 INFO [train.py:968] (0/2) Epoch 18, batch 30000, giga_loss[loss=0.3599, simple_loss=0.4167, pruned_loss=0.1515, over 28950.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3667, pruned_loss=0.1219, over 5659951.84 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3664, pruned_loss=0.1183, over 5683677.04 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3679, pruned_loss=0.1231, over 5654829.93 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:21:01,874 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 12:21:10,552 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2088, simple_loss=0.3174, pruned_loss=0.05008, over 944034.00 frames. 
+2023-03-09 12:21:10,553 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 12:21:35,835 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=806737.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 12:21:58,820 INFO [train.py:968] (0/2) Epoch 18, batch 30050, libri_loss[loss=0.4175, simple_loss=0.4418, pruned_loss=0.1966, over 25799.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3653, pruned_loss=0.1217, over 5643008.47 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3668, pruned_loss=0.1186, over 5681250.30 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3658, pruned_loss=0.1223, over 5641076.50 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:22:13,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=806776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:22:16,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.951e+02 1.623e+03 2.045e+03 2.789e+03 8.541e+03, threshold=4.090e+03, percent-clipped=7.0
+2023-03-09 12:22:18,474 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=806779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:22:22,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=806782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:22:46,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=806811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:22:46,852 INFO [train.py:968] (0/2) Epoch 18, batch 30100, giga_loss[loss=0.2733, simple_loss=0.3471, pruned_loss=0.09977, over 28749.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3642, pruned_loss=0.1192, over 5650017.61 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3668, pruned_loss=0.1187, over 5683773.07 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3647, pruned_loss=0.1196, over 5645765.84 frames. ], batch size: 119, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:22:53,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6104, 1.7761, 1.8816, 1.4078], device='cuda:0'), covar=tensor([0.2017, 0.2732, 0.1634, 0.1896], device='cuda:0'), in_proj_covar=tensor([0.0878, 0.0701, 0.0926, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 12:23:02,830 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=806828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:23:13,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2909, 1.4527, 1.3194, 1.2537], device='cuda:0'), covar=tensor([0.1869, 0.1556, 0.1289, 0.1571], device='cuda:0'), in_proj_covar=tensor([0.1903, 0.1827, 0.1757, 0.1903], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 12:23:37,030 INFO [train.py:968] (0/2) Epoch 18, batch 30150, giga_loss[loss=0.2635, simple_loss=0.3473, pruned_loss=0.08987, over 28921.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3617, pruned_loss=0.1157, over 5643504.64 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3664, pruned_loss=0.1186, over 5690164.87 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3623, pruned_loss=0.1161, over 5633705.13 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:23:56,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.035e+02 1.474e+03 1.989e+03 2.552e+03 7.051e+03, threshold=3.979e+03, percent-clipped=6.0
+2023-03-09 12:24:22,894 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=806905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:24:27,232 INFO [train.py:968] (0/2) Epoch 18, batch 30200, libri_loss[loss=0.3186, simple_loss=0.3778, pruned_loss=0.1297, over 27803.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3605, pruned_loss=0.1131, over 5653368.31 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3663, pruned_loss=0.1185, over 5690187.60 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.361, pruned_loss=0.1134, over 5644525.30 frames. ], batch size: 116, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:24:34,983 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=806919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:24:37,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=806922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:25:08,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=806951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:25:18,002 INFO [train.py:968] (0/2) Epoch 18, batch 30250, giga_loss[loss=0.2749, simple_loss=0.3511, pruned_loss=0.0994, over 28034.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3581, pruned_loss=0.1103, over 5645956.56 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3659, pruned_loss=0.1184, over 5684978.15 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3588, pruned_loss=0.1105, over 5642748.38 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:25:31,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.808e+02 1.249e+03 1.651e+03 2.200e+03 4.586e+03, threshold=3.303e+03, percent-clipped=1.0
+2023-03-09 12:25:44,690 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 12:26:03,599 INFO [train.py:968] (0/2) Epoch 18, batch 30300, giga_loss[loss=0.2464, simple_loss=0.3302, pruned_loss=0.0813, over 28918.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.355, pruned_loss=0.1072, over 5645155.77 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3657, pruned_loss=0.1185, over 5680102.27 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3555, pruned_loss=0.1071, over 5646746.97 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:26:39,553 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=807048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:26:43,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=807051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:26:53,338 INFO [train.py:968] (0/2) Epoch 18, batch 30350, giga_loss[loss=0.2627, simple_loss=0.3476, pruned_loss=0.08888, over 28895.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3536, pruned_loss=0.1036, over 5664196.80 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3653, pruned_loss=0.1183, over 5683239.98 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.354, pruned_loss=0.1033, over 5662232.68 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:27:10,168 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.124e+02 1.292e+03 1.857e+03 2.454e+03 6.104e+03, threshold=3.714e+03, percent-clipped=10.0
+2023-03-09 12:27:13,706 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=807080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:27:38,425 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7007, 2.2181, 1.4112, 0.8543], device='cuda:0'), covar=tensor([0.6290, 0.3823, 0.3561, 0.5819], device='cuda:0'), in_proj_covar=tensor([0.1703, 0.1609, 0.1577, 0.1386], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 12:27:42,719 INFO [train.py:968] (0/2) Epoch 18, batch 30400, giga_loss[loss=0.3037, simple_loss=0.378, pruned_loss=0.1147, over 28703.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3548, pruned_loss=0.1044, over 5657186.88 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3647, pruned_loss=0.1184, over 5676580.88 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3554, pruned_loss=0.1038, over 5661431.86 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:27:42,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=807112.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 12:28:21,314 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-09 12:28:32,437 INFO [train.py:968] (0/2) Epoch 18, batch 30450, giga_loss[loss=0.2493, simple_loss=0.3339, pruned_loss=0.0824, over 27977.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3533, pruned_loss=0.1034, over 5660086.41 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3639, pruned_loss=0.118, over 5681653.67 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3543, pruned_loss=0.1029, over 5658716.30 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:28:49,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.971e+02 1.396e+03 1.991e+03 2.949e+03 9.210e+03, threshold=3.983e+03, percent-clipped=12.0
+2023-03-09 12:29:01,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5055, 3.3216, 3.1323, 1.7889], device='cuda:0'), covar=tensor([0.0737, 0.0988, 0.0938, 0.2091], device='cuda:0'), in_proj_covar=tensor([0.1177, 0.1094, 0.0939, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 12:29:12,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=807203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:29:19,301 INFO [train.py:968] (0/2) Epoch 18, batch 30500, giga_loss[loss=0.2485, simple_loss=0.3242, pruned_loss=0.08639, over 28743.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3502, pruned_loss=0.1011, over 5666283.60 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3631, pruned_loss=0.1175, over 5684921.09 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3514, pruned_loss=0.1008, over 5661657.54 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:29:26,543 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-09 12:29:26,985 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2174, 1.5750, 1.2092, 0.4563], device='cuda:0'), covar=tensor([0.3004, 0.1864, 0.2892, 0.4799], device='cuda:0'), in_proj_covar=tensor([0.1697, 0.1605, 0.1571, 0.1382], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 12:30:03,404 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=807255.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 12:30:05,692 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=807258.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 12:30:08,268 INFO [train.py:968] (0/2) Epoch 18, batch 30550, giga_loss[loss=0.2903, simple_loss=0.3647, pruned_loss=0.1079, over 28637.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3486, pruned_loss=0.1005, over 5659406.07 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.363, pruned_loss=0.1176, over 5685884.26 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3494, pruned_loss=0.09982, over 5654310.86 frames. ], batch size: 336, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:30:25,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.846e+02 1.478e+03 1.840e+03 3.103e+03 1.052e+04, threshold=3.681e+03, percent-clipped=9.0
+2023-03-09 12:30:32,358 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=807287.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 12:30:55,201 INFO [train.py:968] (0/2) Epoch 18, batch 30600, giga_loss[loss=0.3232, simple_loss=0.3973, pruned_loss=0.1246, over 28943.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3484, pruned_loss=0.1, over 5664790.75 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3624, pruned_loss=0.1174, over 5687159.61 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3492, pruned_loss=0.09927, over 5659178.39 frames. ], batch size: 227, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:31:27,128 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=807346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:31:27,233 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=807346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:31:30,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=807349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:31:42,680 INFO [train.py:968] (0/2) Epoch 18, batch 30650, giga_loss[loss=0.2158, simple_loss=0.3067, pruned_loss=0.06249, over 28838.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3464, pruned_loss=0.09849, over 5665431.98 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3622, pruned_loss=0.1174, over 5692043.33 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3469, pruned_loss=0.09752, over 5656162.76 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:31:48,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2074, 2.5881, 1.2737, 1.3575], device='cuda:0'), covar=tensor([0.1027, 0.0339, 0.0979, 0.1440], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0545, 0.0373, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 12:31:57,708 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=807378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:31:58,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.304e+02 1.375e+03 1.804e+03 2.636e+03 6.049e+03, threshold=3.609e+03, percent-clipped=8.0
+2023-03-09 12:32:14,550 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=807394.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 12:32:33,524 INFO [train.py:968] (0/2) Epoch 18, batch 30700, giga_loss[loss=0.2365, simple_loss=0.3126, pruned_loss=0.08023, over 27627.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3439, pruned_loss=0.09649, over 5648504.53 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.362, pruned_loss=0.1174, over 5676211.18 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3441, pruned_loss=0.09537, over 5654940.38 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:33:19,010 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 12:33:22,272 INFO [train.py:968] (0/2) Epoch 18, batch 30750, giga_loss[loss=0.2816, simple_loss=0.3349, pruned_loss=0.1141, over 26746.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3403, pruned_loss=0.09443, over 5654302.20 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3614, pruned_loss=0.1172, over 5670144.42 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3406, pruned_loss=0.0932, over 5664921.72 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:33:28,842 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 12:33:39,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.608e+02 1.367e+03 1.716e+03 2.647e+03 1.269e+04, threshold=3.431e+03, percent-clipped=10.0
+2023-03-09 12:33:57,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4939, 1.4920, 1.4469, 1.7037], device='cuda:0'), covar=tensor([0.0639, 0.0295, 0.0302, 0.0665], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0115, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:0')
+2023-03-09 12:34:11,658 INFO [train.py:968] (0/2) Epoch 18, batch 30800, giga_loss[loss=0.2319, simple_loss=0.3123, pruned_loss=0.07574, over 28187.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3373, pruned_loss=0.09308, over 5649663.34 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.361, pruned_loss=0.117, over 5663859.95 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3375, pruned_loss=0.09181, over 5664327.21 frames. ], batch size: 77, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:34:33,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4780, 1.7161, 1.4533, 1.2880], device='cuda:0'), covar=tensor([0.2278, 0.1993, 0.2188, 0.1943], device='cuda:0'), in_proj_covar=tensor([0.1450, 0.1051, 0.1288, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 12:34:48,976 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=807550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:34:59,638 INFO [train.py:968] (0/2) Epoch 18, batch 30850, giga_loss[loss=0.2365, simple_loss=0.3135, pruned_loss=0.07968, over 28458.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3376, pruned_loss=0.09423, over 5648153.66 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3608, pruned_loss=0.1171, over 5670495.90 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3373, pruned_loss=0.09245, over 5653292.49 frames. ], batch size: 71, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:35:12,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1893, 1.2521, 3.2617, 2.9493], device='cuda:0'), covar=tensor([0.1565, 0.2689, 0.0533, 0.1413], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0634, 0.0934, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 12:35:20,181 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.344e+02 1.547e+03 2.159e+03 3.048e+03 1.212e+04, threshold=4.319e+03, percent-clipped=17.0
+2023-03-09 12:35:50,155 INFO [train.py:968] (0/2) Epoch 18, batch 30900, giga_loss[loss=0.2733, simple_loss=0.3467, pruned_loss=0.09998, over 28811.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3381, pruned_loss=0.09452, over 5645250.96 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3598, pruned_loss=0.1166, over 5674745.85 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3379, pruned_loss=0.09281, over 5644844.04 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:36:44,845 INFO [train.py:968] (0/2) Epoch 18, batch 30950, giga_loss[loss=0.2629, simple_loss=0.3492, pruned_loss=0.08833, over 28802.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.341, pruned_loss=0.09527, over 5637011.55 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3593, pruned_loss=0.1165, over 5670198.77 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3409, pruned_loss=0.09355, over 5639684.52 frames. ], batch size: 243, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:37:06,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.561e+02 1.511e+03 1.999e+03 2.714e+03 9.044e+03, threshold=3.998e+03, percent-clipped=9.0
+2023-03-09 12:37:36,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4710, 1.5054, 1.3841, 1.5647], device='cuda:0'), covar=tensor([0.0736, 0.0319, 0.0321, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 12:37:39,186 INFO [train.py:968] (0/2) Epoch 18, batch 31000, giga_loss[loss=0.2376, simple_loss=0.3246, pruned_loss=0.07524, over 28811.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3412, pruned_loss=0.09523, over 5635671.28 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3594, pruned_loss=0.1167, over 5677565.95 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3405, pruned_loss=0.09297, over 5629979.95 frames. ], batch size: 119, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:37:52,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=807721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:38:31,033 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=807753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:38:42,142 INFO [train.py:968] (0/2) Epoch 18, batch 31050, giga_loss[loss=0.2744, simple_loss=0.3525, pruned_loss=0.09812, over 28715.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3409, pruned_loss=0.09506, over 5643640.18 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3586, pruned_loss=0.1163, over 5682682.78 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3406, pruned_loss=0.09314, over 5633542.03 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:38:52,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=807769.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 12:39:06,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.060e+02 1.467e+03 1.816e+03 2.539e+03 7.676e+03, threshold=3.633e+03, percent-clipped=7.0
+2023-03-09 12:39:37,514 INFO [train.py:968] (0/2) Epoch 18, batch 31100, giga_loss[loss=0.2733, simple_loss=0.3535, pruned_loss=0.09652, over 28083.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3393, pruned_loss=0.09372, over 5654370.25 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3583, pruned_loss=0.1162, over 5686407.54 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3387, pruned_loss=0.09143, over 5641630.53 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:40:09,788 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=807836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:40:26,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3806, 1.7954, 1.2864, 0.7459], device='cuda:0'), covar=tensor([0.4809, 0.2691, 0.4259, 0.5534], device='cuda:0'), in_proj_covar=tensor([0.1708, 0.1613, 0.1580, 0.1392], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 12:40:35,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4035, 2.6075, 1.5607, 1.6036], device='cuda:0'), covar=tensor([0.0853, 0.0323, 0.0821, 0.1169], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0544, 0.0373, 0.0418], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 12:40:40,407 INFO [train.py:968] (0/2) Epoch 18, batch 31150, giga_loss[loss=0.2373, simple_loss=0.338, pruned_loss=0.06826, over 28934.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3372, pruned_loss=0.0909, over 5643888.96 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3582, pruned_loss=0.1162, over 5689869.36 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3365, pruned_loss=0.0888, over 5630406.02 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:40:43,094 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=807864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:40:47,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=807867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:40:56,047 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-09 12:41:02,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.648e+02 1.419e+03 1.871e+03 2.734e+03 6.231e+03, threshold=3.742e+03, percent-clipped=7.0
+2023-03-09 12:41:19,595 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=807896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:41:30,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8532, 1.9471, 1.5404, 1.5506], device='cuda:0'), covar=tensor([0.0940, 0.0686, 0.0911, 0.1221], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0444, 0.0511, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 12:41:38,765 INFO [train.py:968] (0/2) Epoch 18, batch 31200, giga_loss[loss=0.2414, simple_loss=0.3155, pruned_loss=0.0837, over 29266.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3346, pruned_loss=0.08995, over 5630193.08 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3579, pruned_loss=0.1163, over 5665566.77 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3338, pruned_loss=0.0876, over 5640001.99 frames. ], batch size: 107, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:41:39,250 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=807912.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 12:41:41,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=807915.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 12:41:51,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=807925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:42:19,918 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=807944.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 12:42:19,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5132, 1.7715, 1.4546, 1.6541], device='cuda:0'), covar=tensor([0.2717, 0.2361, 0.2800, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.1450, 0.1051, 0.1289, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 12:42:38,988 INFO [train.py:968] (0/2) Epoch 18, batch 31250, giga_loss[loss=0.3087, simple_loss=0.3756, pruned_loss=0.1209, over 28852.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3338, pruned_loss=0.09027, over 5651396.19 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3575, pruned_loss=0.1163, over 5672648.58 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3328, pruned_loss=0.08764, over 5652142.95 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:42:59,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.629e+02 1.384e+03 1.940e+03 2.818e+03 1.020e+04, threshold=3.879e+03, percent-clipped=9.0
+2023-03-09 12:43:22,824 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-808000.pt
+2023-03-09 12:43:36,341 INFO [train.py:968] (0/2) Epoch 18, batch 31300, giga_loss[loss=0.2586, simple_loss=0.3455, pruned_loss=0.08582, over 28815.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3334, pruned_loss=0.08989, over 5659474.88 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3575, pruned_loss=0.1162, over 5672308.52 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3323, pruned_loss=0.08753, over 5659939.30 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:44:29,320 INFO [train.py:968] (0/2) Epoch 18, batch 31350, giga_loss[loss=0.2485, simple_loss=0.3285, pruned_loss=0.08423, over 28754.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3342, pruned_loss=0.08998, over 5661439.06 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3572, pruned_loss=0.1161, over 5675005.73 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3329, pruned_loss=0.08734, over 5659080.39 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:44:34,945 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:44:38,135 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:44:49,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.437e+02 1.382e+03 1.739e+03 2.300e+03 4.671e+03, threshold=3.478e+03, percent-clipped=3.0
+2023-03-09 12:45:14,264 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=808100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:45:23,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3817, 1.7479, 1.4818, 1.5788], device='cuda:0'), covar=tensor([0.0759, 0.0339, 0.0333, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 12:45:30,673 INFO [train.py:968] (0/2) Epoch 18, batch 31400, giga_loss[loss=0.2336, simple_loss=0.3213, pruned_loss=0.07298, over 29053.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3359, pruned_loss=0.09062, over 5647809.66 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3568, pruned_loss=0.1159, over 5671126.94 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3347, pruned_loss=0.08805, over 5648549.07 frames. ], batch size: 155, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:45:38,371 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=808117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:45:54,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=808128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:46:39,550 INFO [train.py:968] (0/2) Epoch 18, batch 31450, giga_loss[loss=0.2149, simple_loss=0.2987, pruned_loss=0.06552, over 28437.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3326, pruned_loss=0.08845, over 5667530.95 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3567, pruned_loss=0.116, over 5675025.67 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3314, pruned_loss=0.08591, over 5664359.78 frames. ], batch size: 78, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:46:39,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=808162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:47:06,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.804e+02 1.397e+03 1.933e+03 2.935e+03 9.350e+03, threshold=3.865e+03, percent-clipped=17.0
+2023-03-09 12:47:24,880 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-09 12:47:47,293 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=808211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:47:47,613 INFO [train.py:968] (0/2) Epoch 18, batch 31500, giga_loss[loss=0.2822, simple_loss=0.3651, pruned_loss=0.09967, over 28771.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3351, pruned_loss=0.09062, over 5661316.55 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3565, pruned_loss=0.1159, over 5670360.56 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3338, pruned_loss=0.0881, over 5662476.86 frames. ], batch size: 243, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:48:25,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1414, 3.9555, 3.7916, 1.9477], device='cuda:0'), covar=tensor([0.0554, 0.0706, 0.0652, 0.1938], device='cuda:0'), in_proj_covar=tensor([0.1159, 0.1078, 0.0923, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 12:48:38,447 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=808253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:48:47,080 INFO [train.py:968] (0/2) Epoch 18, batch 31550, giga_loss[loss=0.2589, simple_loss=0.3564, pruned_loss=0.0807, over 28679.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3374, pruned_loss=0.0905, over 5660322.56 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3556, pruned_loss=0.1154, over 5664884.45 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3366, pruned_loss=0.08822, over 5664922.96 frames. ], batch size: 242, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:48:56,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:49:00,324 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:49:09,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.797e+02 1.457e+03 1.793e+03 2.690e+03 6.388e+03, threshold=3.586e+03, percent-clipped=9.0
+2023-03-09 12:49:35,338 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=808303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:49:47,711 INFO [train.py:968] (0/2) Epoch 18, batch 31600, giga_loss[loss=0.2463, simple_loss=0.3374, pruned_loss=0.07759, over 28355.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3389, pruned_loss=0.08917, over 5642151.29 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3553, pruned_loss=0.1154, over 5650518.53 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3379, pruned_loss=0.0865, over 5658127.89 frames. ], batch size: 369, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:50:37,853 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:50:42,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:50:47,507 INFO [train.py:968] (0/2) Epoch 18, batch 31650, giga_loss[loss=0.2444, simple_loss=0.3379, pruned_loss=0.07541, over 28972.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3404, pruned_loss=0.08889, over 5648200.25 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3547, pruned_loss=0.115, over 5656872.13 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3398, pruned_loss=0.08646, over 5655376.90 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:50:58,359 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 12:51:09,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.943e+02 1.467e+03 1.818e+03 2.410e+03 6.096e+03, threshold=3.636e+03, percent-clipped=6.0
+2023-03-09 12:51:17,503 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=808386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:51:49,906 INFO [train.py:968] (0/2) Epoch 18, batch 31700, giga_loss[loss=0.2225, simple_loss=0.3158, pruned_loss=0.06459, over 28927.00 frames. ], tot_loss[loss=0.256, simple_loss=0.339, pruned_loss=0.08654, over 5663858.43 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3543, pruned_loss=0.1147, over 5660614.10 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3387, pruned_loss=0.08453, over 5666156.67 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:52:49,134 INFO [train.py:968] (0/2) Epoch 18, batch 31750, giga_loss[loss=0.255, simple_loss=0.33, pruned_loss=0.09003, over 27774.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3405, pruned_loss=0.08876, over 5676722.43 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.354, pruned_loss=0.1146, over 5669035.14 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.34, pruned_loss=0.08633, over 5671127.41 frames. ], batch size: 474, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:53:13,597 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.233e+02 1.252e+03 1.701e+03 2.304e+03 1.203e+04, threshold=3.401e+03, percent-clipped=7.0
+2023-03-09 12:53:29,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=808492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:53:58,791 INFO [train.py:968] (0/2) Epoch 18, batch 31800, giga_loss[loss=0.2635, simple_loss=0.3412, pruned_loss=0.0929, over 28782.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3384, pruned_loss=0.0888, over 5679412.48 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3535, pruned_loss=0.1142, over 5672946.58 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3382, pruned_loss=0.0867, over 5671729.17 frames. ], batch size: 243, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:54:36,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=808537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:54:54,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6519, 1.7963, 1.5213, 1.7176], device='cuda:0'), covar=tensor([0.2734, 0.2680, 0.2992, 0.2713], device='cuda:0'), in_proj_covar=tensor([0.1442, 0.1045, 0.1282, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 12:55:18,292 INFO [train.py:968] (0/2) Epoch 18, batch 31850, giga_loss[loss=0.2178, simple_loss=0.309, pruned_loss=0.06329, over 28973.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.34, pruned_loss=0.09089, over 5672064.62 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3534, pruned_loss=0.1143, over 5669613.55 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3396, pruned_loss=0.08853, over 5669846.39 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:55:18,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7299, 1.9216, 1.6057, 2.0788], device='cuda:0'), covar=tensor([0.2913, 0.2906, 0.3255, 0.2597], device='cuda:0'), in_proj_covar=tensor([0.1439, 0.1042, 0.1279, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 12:55:46,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.154e+02 1.374e+03 1.692e+03 2.227e+03 3.665e+03, threshold=3.385e+03, percent-clipped=2.0
+2023-03-09 12:56:30,155 INFO [train.py:968] (0/2) Epoch 18, batch 31900, giga_loss[loss=0.2378, simple_loss=0.3166, pruned_loss=0.07953, over 28947.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3356, pruned_loss=0.08846, over 5678120.76 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3532, pruned_loss=0.1142, over 5673015.80 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3352, pruned_loss=0.08625, over 5673287.49 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:56:52,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=808628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:57:01,744 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:57:05,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:57:35,117 INFO [train.py:968] (0/2) Epoch 18, batch 31950, giga_loss[loss=0.2233, simple_loss=0.3125, pruned_loss=0.0671, over 28911.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3333, pruned_loss=0.08687, over 5675290.55 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3529, pruned_loss=0.114, over 5675876.32 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3327, pruned_loss=0.08469, over 5668825.55 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:57:42,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=808667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:58:00,223 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:58:01,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.618e+02 1.307e+03 1.645e+03 2.170e+03 7.303e+03, threshold=3.290e+03, percent-clipped=10.0
+2023-03-09 12:58:02,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:58:39,298 INFO [train.py:968] (0/2) Epoch 18, batch 32000, giga_loss[loss=0.2792, simple_loss=0.3516, pruned_loss=0.1033, over 28715.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3322, pruned_loss=0.08707, over 5683853.11 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3526, pruned_loss=0.114, over 5678777.94 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3315, pruned_loss=0.08461, over 5676125.68 frames. ], batch size: 263, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:58:40,264 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=808712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:59:43,885 INFO [train.py:968] (0/2) Epoch 18, batch 32050, giga_loss[loss=0.2805, simple_loss=0.36, pruned_loss=0.1005, over 28993.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3364, pruned_loss=0.08914, over 5683162.30 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3527, pruned_loss=0.1143, over 5678006.81 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3354, pruned_loss=0.08649, over 5677755.19 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:59:53,920 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:59:57,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:00:02,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5831, 4.4167, 4.1899, 1.9539], device='cuda:0'), covar=tensor([0.0499, 0.0668, 0.0728, 0.2005], device='cuda:0'), in_proj_covar=tensor([0.1162, 0.1074, 0.0925, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 13:00:05,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.416e+02 1.542e+03 1.977e+03 2.677e+03 6.101e+03, threshold=3.955e+03, percent-clipped=13.0
+2023-03-09 13:00:20,359 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=808792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:00:32,022 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=808803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:00:41,862 INFO [train.py:968] (0/2) Epoch 18, batch 32100, giga_loss[loss=0.2609, simple_loss=0.3297, pruned_loss=0.09606, over 26909.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3362, pruned_loss=0.0897, over 5689924.37 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3526, pruned_loss=0.1143, over 5680043.32 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3351, pruned_loss=0.08694, over 5684422.25 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 13:01:47,477 INFO [train.py:968] (0/2) Epoch 18, batch 32150, giga_loss[loss=0.2568, simple_loss=0.3376, pruned_loss=0.08804, over 28681.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3348, pruned_loss=0.08978, over 5681516.10 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3518, pruned_loss=0.1139, over 5674280.96 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3342, pruned_loss=0.08728, over 5682846.98 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 13:02:09,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.042e+02 1.392e+03 1.934e+03 2.466e+03 7.356e+03, threshold=3.869e+03, percent-clipped=9.0
+2023-03-09 13:02:45,864 INFO [train.py:968] (0/2) Epoch 18, batch 32200, giga_loss[loss=0.2394, simple_loss=0.3192, pruned_loss=0.07983, over 28552.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3353, pruned_loss=0.09048, over 5682335.37 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3517, pruned_loss=0.1136, over 5677609.14 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3344, pruned_loss=0.08804, over 5680535.39 frames. ], batch size: 65, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 13:02:47,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5372, 1.7560, 1.8053, 1.3306], device='cuda:0'), covar=tensor([0.1883, 0.2542, 0.1517, 0.1820], device='cuda:0'), in_proj_covar=tensor([0.0872, 0.0691, 0.0919, 0.0818], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 13:03:28,627 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 13:03:54,069 INFO [train.py:968] (0/2) Epoch 18, batch 32250, giga_loss[loss=0.3078, simple_loss=0.3823, pruned_loss=0.1167, over 28087.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3361, pruned_loss=0.09068, over 5661619.24 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3518, pruned_loss=0.114, over 5664385.71 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3349, pruned_loss=0.08783, over 5672058.09 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 13:04:25,212 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.385e+02 1.333e+03 1.775e+03 2.558e+03 8.657e+03, threshold=3.550e+03, percent-clipped=5.0
+2023-03-09 13:05:03,886 INFO [train.py:968] (0/2) Epoch 18, batch 32300, giga_loss[loss=0.2634, simple_loss=0.3464, pruned_loss=0.09015, over 28820.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3385, pruned_loss=0.09149, over 5665649.74 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3512, pruned_loss=0.1137, over 5671406.84 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3376, pruned_loss=0.08861, over 5667586.69 frames. ], batch size: 243, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 13:06:17,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=809060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:06:18,604 INFO [train.py:968] (0/2) Epoch 18, batch 32350, giga_loss[loss=0.2284, simple_loss=0.313, pruned_loss=0.07194, over 28790.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3378, pruned_loss=0.09093, over 5665363.38 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.351, pruned_loss=0.1135, over 5676994.34 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3369, pruned_loss=0.08831, over 5662049.75 frames. ], batch size: 119, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 13:06:46,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.966e+02 1.449e+03 1.799e+03 2.663e+03 9.489e+03, threshold=3.597e+03, percent-clipped=6.0
+2023-03-09 13:07:09,828 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 13:07:27,220 INFO [train.py:968] (0/2) Epoch 18, batch 32400, giga_loss[loss=0.2166, simple_loss=0.2954, pruned_loss=0.06889, over 28321.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3329, pruned_loss=0.08903, over 5678580.04 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3504, pruned_loss=0.1131, over 5681398.48 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3325, pruned_loss=0.08684, over 5672038.19 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 13:07:31,577 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-09 13:07:43,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5477, 4.3984, 4.1654, 1.8207], device='cuda:0'), covar=tensor([0.0503, 0.0636, 0.0729, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.1161, 0.1076, 0.0921, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 13:08:35,039 INFO [train.py:968] (0/2) Epoch 18, batch 32450, giga_loss[loss=0.2667, simple_loss=0.3344, pruned_loss=0.09953, over 26779.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3274, pruned_loss=0.08661, over 5670020.37 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3503, pruned_loss=0.1132, over 5675103.94 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3268, pruned_loss=0.08439, over 5670803.88 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:08:41,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=809167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:09:00,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 13:09:04,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.370e+02 1.364e+03 1.854e+03 2.814e+03 5.945e+03, threshold=3.708e+03, percent-clipped=16.0
+2023-03-09 13:09:36,027 INFO [train.py:968] (0/2) Epoch 18, batch 32500, giga_loss[loss=0.3132, simple_loss=0.3574, pruned_loss=0.1345, over 26799.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.328, pruned_loss=0.08748, over 5670058.42 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3502, pruned_loss=0.1132, over 5678600.85 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3268, pruned_loss=0.08484, over 5667212.96 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:09:43,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=809218.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:10:13,574 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.15 vs. limit=5.0
+2023-03-09 13:10:34,771 INFO [train.py:968] (0/2) Epoch 18, batch 32550, giga_loss[loss=0.2727, simple_loss=0.3472, pruned_loss=0.09913, over 28660.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.329, pruned_loss=0.08798, over 5675651.75 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3499, pruned_loss=0.1131, over 5682229.61 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3282, pruned_loss=0.08566, over 5669823.09 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:10:46,485 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-09 13:11:02,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.796e+02 1.696e+03 2.404e+03 3.810e+03 1.215e+04, threshold=4.809e+03, percent-clipped=26.0
+2023-03-09 13:11:23,189 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-09 13:11:33,797 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=809310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:11:34,966 INFO [train.py:968] (0/2) Epoch 18, batch 32600, giga_loss[loss=0.2377, simple_loss=0.3264, pruned_loss=0.07446, over 29131.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3269, pruned_loss=0.086, over 5675781.10 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3501, pruned_loss=0.1134, over 5684879.26 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3256, pruned_loss=0.08342, over 5668712.08 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:11:38,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=809313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:12:14,166 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=809342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:12:39,958 INFO [train.py:968] (0/2) Epoch 18, batch 32650, giga_loss[loss=0.2574, simple_loss=0.3285, pruned_loss=0.09312, over 27647.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3262, pruned_loss=0.08533, over 5673280.65 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3501, pruned_loss=0.1134, over 5690041.26 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3247, pruned_loss=0.08263, over 5662781.16 frames. ], batch size: 472, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:13:07,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.815e+02 1.215e+03 1.521e+03 1.955e+03 5.790e+03, threshold=3.042e+03, percent-clipped=2.0
+2023-03-09 13:13:24,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3458, 1.2290, 1.1121, 1.5082], device='cuda:0'), covar=tensor([0.0750, 0.0358, 0.0355, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 13:13:46,441 INFO [train.py:968] (0/2) Epoch 18, batch 32700, giga_loss[loss=0.2592, simple_loss=0.3323, pruned_loss=0.09308, over 29009.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3258, pruned_loss=0.08589, over 5666223.31 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3497, pruned_loss=0.1132, over 5686876.84 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3241, pruned_loss=0.08314, over 5660655.44 frames. ], batch size: 199, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:14:19,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=809435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:14:45,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2583, 1.5101, 1.4074, 1.2423], device='cuda:0'), covar=tensor([0.2407, 0.1927, 0.1529, 0.1952], device='cuda:0'), in_proj_covar=tensor([0.1867, 0.1770, 0.1705, 0.1857], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 13:14:54,908 INFO [train.py:968] (0/2) Epoch 18, batch 32750, giga_loss[loss=0.2539, simple_loss=0.3342, pruned_loss=0.08679, over 29027.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3255, pruned_loss=0.08476, over 5672227.96 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3496, pruned_loss=0.1132, over 5681117.98 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3239, pruned_loss=0.08216, over 5672147.60 frames. ], batch size: 199, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:15:25,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.938e+02 1.298e+03 1.696e+03 2.525e+03 5.908e+03, threshold=3.393e+03, percent-clipped=14.0
+2023-03-09 13:15:58,528 INFO [train.py:968] (0/2) Epoch 18, batch 32800, giga_loss[loss=0.2558, simple_loss=0.3382, pruned_loss=0.08676, over 28420.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3265, pruned_loss=0.08571, over 5676465.27 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3493, pruned_loss=0.1131, over 5684409.31 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3248, pruned_loss=0.08292, over 5673022.39 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:16:05,237 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=809516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:17:00,141 INFO [train.py:968] (0/2) Epoch 18, batch 32850, giga_loss[loss=0.2576, simple_loss=0.3334, pruned_loss=0.09089, over 29006.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3276, pruned_loss=0.08678, over 5682706.72 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3491, pruned_loss=0.1129, over 5687565.35 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3262, pruned_loss=0.0844, over 5677151.59 frames. ], batch size: 213, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:17:20,942 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=809578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:17:27,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=809581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:17:33,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.914e+02 1.171e+03 1.542e+03 2.517e+03 1.015e+04, threshold=3.084e+03, percent-clipped=13.0
+2023-03-09 13:17:42,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=809593.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:17:51,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4729, 1.7775, 1.4457, 1.6256], device='cuda:0'), covar=tensor([0.0785, 0.0307, 0.0336, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 13:17:58,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=809610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:18:00,013 INFO [train.py:968] (0/2) Epoch 18, batch 32900, libri_loss[loss=0.3095, simple_loss=0.3636, pruned_loss=0.1277, over 29484.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3272, pruned_loss=0.0864, over 5682412.98 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3489, pruned_loss=0.1128, over 5694868.69 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3255, pruned_loss=0.08369, over 5670591.48 frames. ], batch size: 85, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:18:13,201 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=809622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:18:57,767 INFO [train.py:968] (0/2) Epoch 18, batch 32950, giga_loss[loss=0.2279, simple_loss=0.3229, pruned_loss=0.06645, over 29027.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3282, pruned_loss=0.08594, over 5674441.35 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3486, pruned_loss=0.1126, over 5698384.71 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3264, pruned_loss=0.08299, over 5660874.58 frames. ], batch size: 285, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:19:15,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3782, 2.9622, 1.4983, 1.5034], device='cuda:0'), covar=tensor([0.0980, 0.0333, 0.0946, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0539, 0.0371, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 13:19:22,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.442e+02 1.408e+03 2.246e+03 3.716e+03 1.972e+04, threshold=4.492e+03, percent-clipped=26.0
+2023-03-09 13:19:37,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-09 13:19:57,057 INFO [train.py:968] (0/2) Epoch 18, batch 33000, giga_loss[loss=0.2714, simple_loss=0.3494, pruned_loss=0.09668, over 28935.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3313, pruned_loss=0.08679, over 5672816.34 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3483, pruned_loss=0.1124, over 5700581.35 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3298, pruned_loss=0.08417, over 5659737.68 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:19:57,060 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 13:20:05,812 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.1974, simple_loss=0.2982, pruned_loss=0.04836, over 944034.00 frames. 
+2023-03-09 13:20:05,812 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 13:20:33,929 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=809736.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:20:34,528 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=809737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:20:36,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=809739.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:21:08,335 INFO [train.py:968] (0/2) Epoch 18, batch 33050, giga_loss[loss=0.284, simple_loss=0.3428, pruned_loss=0.1126, over 26979.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3331, pruned_loss=0.08754, over 5679303.15 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3485, pruned_loss=0.1125, over 5705792.60 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3315, pruned_loss=0.08481, over 5663785.33 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:21:08,719 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7347, 2.2280, 2.1118, 1.6564], device='cuda:0'), covar=tensor([0.2436, 0.1804, 0.1751, 0.2151], device='cuda:0'), in_proj_covar=tensor([0.1864, 0.1766, 0.1697, 0.1848], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 13:21:16,901 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=809768.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:21:35,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.098e+02 1.305e+03 1.748e+03 2.201e+03 6.339e+03, threshold=3.496e+03, percent-clipped=2.0
+2023-03-09 13:22:04,905 INFO [train.py:968] (0/2) Epoch 18, batch 33100, giga_loss[loss=0.2765, simple_loss=0.3482, pruned_loss=0.1024, over 29145.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3333, pruned_loss=0.08821, over 5667969.17 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3485, pruned_loss=0.1125, over 5697964.36 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3313, pruned_loss=0.08507, over 5661382.82 frames. ], batch size: 200, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:22:11,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7080, 2.3959, 1.8078, 0.8711], device='cuda:0'), covar=tensor([0.5926, 0.2835, 0.3993, 0.6325], device='cuda:0'), in_proj_covar=tensor([0.1688, 0.1599, 0.1562, 0.1379], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 13:22:50,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5486, 2.2780, 1.6449, 0.8159], device='cuda:0'), covar=tensor([0.5669, 0.2583, 0.3968, 0.5919], device='cuda:0'), in_proj_covar=tensor([0.1684, 0.1597, 0.1559, 0.1376], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 13:22:57,331 INFO [train.py:968] (0/2) Epoch 18, batch 33150, giga_loss[loss=0.2253, simple_loss=0.3072, pruned_loss=0.07173, over 28820.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3313, pruned_loss=0.08695, over 5675970.03 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3485, pruned_loss=0.1123, over 5696096.38 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3289, pruned_loss=0.08327, over 5670194.14 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:23:26,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7243, 1.8632, 1.9910, 1.5229], device='cuda:0'), covar=tensor([0.2122, 0.2609, 0.1660, 0.1891], device='cuda:0'), in_proj_covar=tensor([0.0875, 0.0690, 0.0922, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 13:23:29,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.544e+02 1.298e+03 1.825e+03 2.641e+03 6.029e+03, threshold=3.650e+03, percent-clipped=14.0
+2023-03-09 13:23:35,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=809891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:24:00,679 INFO [train.py:968] (0/2) Epoch 18, batch 33200, giga_loss[loss=0.2486, simple_loss=0.3313, pruned_loss=0.08292, over 28645.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3288, pruned_loss=0.08492, over 5675384.51 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3484, pruned_loss=0.1122, over 5694285.87 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3268, pruned_loss=0.08188, over 5672261.65 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:25:00,400 INFO [train.py:968] (0/2) Epoch 18, batch 33250, giga_loss[loss=0.2523, simple_loss=0.3237, pruned_loss=0.09044, over 26948.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3275, pruned_loss=0.08504, over 5663350.20 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3486, pruned_loss=0.1124, over 5678575.91 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3254, pruned_loss=0.08206, over 5674331.44 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:25:16,155 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-09 13:25:31,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.103e+02 1.348e+03 1.699e+03 2.561e+03 7.244e+03, threshold=3.399e+03, percent-clipped=9.0
+2023-03-09 13:25:45,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=809997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:25:48,554 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-810000.pt
+2023-03-09 13:26:02,848 INFO [train.py:968] (0/2) Epoch 18, batch 33300, giga_loss[loss=0.3091, simple_loss=0.3618, pruned_loss=0.1282, over 24604.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3278, pruned_loss=0.085, over 5663643.12 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3483, pruned_loss=0.1122, over 5682818.94 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3257, pruned_loss=0.08194, over 5668013.18 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:26:31,352 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:26:33,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:27:04,856 INFO [train.py:968] (0/2) Epoch 18, batch 33350, giga_loss[loss=0.2403, simple_loss=0.3232, pruned_loss=0.07871, over 28418.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3303, pruned_loss=0.08658, over 5660266.53 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3486, pruned_loss=0.1126, over 5674446.87 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.328, pruned_loss=0.08322, over 5671049.93 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:27:10,719 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:27:34,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.331e+02 1.498e+03 1.822e+03 2.765e+03 5.601e+03, threshold=3.645e+03, percent-clipped=14.0
+2023-03-09 13:27:57,927 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 13:28:02,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5100, 2.0043, 1.3073, 0.8572], device='cuda:0'), covar=tensor([0.7344, 0.4185, 0.3518, 0.6337], device='cuda:0'), in_proj_covar=tensor([0.1693, 0.1606, 0.1565, 0.1384], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 13:28:07,534 INFO [train.py:968] (0/2) Epoch 18, batch 33400, giga_loss[loss=0.2573, simple_loss=0.3425, pruned_loss=0.08609, over 28489.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3312, pruned_loss=0.0876, over 5665203.08 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3483, pruned_loss=0.1125, over 5681359.33 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3292, pruned_loss=0.08433, over 5667496.71 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:28:07,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=810112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:28:50,514 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:28:55,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:29:19,441 INFO [train.py:968] (0/2) Epoch 18, batch 33450, giga_loss[loss=0.2669, simple_loss=0.3516, pruned_loss=0.09113, over 28905.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3346, pruned_loss=0.08965, over 5650943.94 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3485, pruned_loss=0.1127, over 5682584.53 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3328, pruned_loss=0.08684, over 5651588.15 frames. ], batch size: 284, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:29:33,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:29:37,059 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:29:40,039 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:29:47,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.878e+02 1.433e+03 2.015e+03 3.056e+03 6.424e+03, threshold=4.030e+03, percent-clipped=12.0
+2023-03-09 13:30:14,953 INFO [train.py:968] (0/2) Epoch 18, batch 33500, giga_loss[loss=0.2345, simple_loss=0.3308, pruned_loss=0.06911, over 28836.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3373, pruned_loss=0.09052, over 5665107.28 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3478, pruned_loss=0.1121, over 5688416.74 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.336, pruned_loss=0.088, over 5659257.56 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:31:09,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:31:13,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:31:17,289 INFO [train.py:968] (0/2) Epoch 18, batch 33550, libri_loss[loss=0.3426, simple_loss=0.3781, pruned_loss=0.1536, over 19592.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3384, pruned_loss=0.09116, over 5659533.93 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3475, pruned_loss=0.1119, over 5683172.64 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3372, pruned_loss=0.08854, over 5659537.87 frames. ], batch size: 187, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:31:45,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3491, 2.0316, 1.5753, 0.6840], device='cuda:0'), covar=tensor([0.4899, 0.2637, 0.4160, 0.5336], device='cuda:0'), in_proj_covar=tensor([0.1695, 0.1605, 0.1569, 0.1385], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 13:31:49,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.195e+02 1.351e+03 1.710e+03 2.268e+03 7.507e+03, threshold=3.420e+03, percent-clipped=1.0
+2023-03-09 13:31:51,582 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:32:22,987 INFO [train.py:968] (0/2) Epoch 18, batch 33600, giga_loss[loss=0.2238, simple_loss=0.3063, pruned_loss=0.07067, over 28660.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3365, pruned_loss=0.09038, over 5655796.02 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3476, pruned_loss=0.112, over 5675151.83 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3353, pruned_loss=0.08756, over 5661375.37 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:32:37,883 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 13:33:25,968 INFO [train.py:968] (0/2) Epoch 18, batch 33650, libri_loss[loss=0.2853, simple_loss=0.3517, pruned_loss=0.1095, over 29498.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3357, pruned_loss=0.09059, over 5672233.29 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3473, pruned_loss=0.1119, over 5680617.62 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3345, pruned_loss=0.08748, over 5670937.90 frames. ], batch size: 85, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:33:56,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.913e+02 1.585e+03 2.003e+03 2.816e+03 7.549e+03, threshold=4.006e+03, percent-clipped=20.0
+2023-03-09 13:34:29,026 INFO [train.py:968] (0/2) Epoch 18, batch 33700, giga_loss[loss=0.2501, simple_loss=0.332, pruned_loss=0.08412, over 28925.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.335, pruned_loss=0.09045, over 5675594.45 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3474, pruned_loss=0.1122, over 5686177.62 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3336, pruned_loss=0.08724, over 5669302.15 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:34:29,372 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:35:37,460 INFO [train.py:968] (0/2) Epoch 18, batch 33750, giga_loss[loss=0.2552, simple_loss=0.3291, pruned_loss=0.09062, over 28097.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3323, pruned_loss=0.08957, over 5674275.87 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3472, pruned_loss=0.112, over 5689773.44 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3312, pruned_loss=0.08676, over 5665923.49 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:35:47,272 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810470.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:35:48,617 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-09 13:36:11,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.327e+02 1.447e+03 2.170e+03 3.220e+03 7.291e+03, threshold=4.340e+03, percent-clipped=14.0
+2023-03-09 13:36:41,584 INFO [train.py:968] (0/2) Epoch 18, batch 33800, giga_loss[loss=0.1978, simple_loss=0.2699, pruned_loss=0.06288, over 24375.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.33, pruned_loss=0.08849, over 5680327.51 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3468, pruned_loss=0.1118, over 5688891.35 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3291, pruned_loss=0.08581, over 5674306.37 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:37:28,296 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=810551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:37:33,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=810554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:37:40,194 INFO [train.py:968] (0/2) Epoch 18, batch 33850, giga_loss[loss=0.2267, simple_loss=0.2933, pruned_loss=0.08006, over 24677.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3306, pruned_loss=0.08766, over 5672549.50 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3466, pruned_loss=0.1117, over 5683235.35 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3296, pruned_loss=0.08498, over 5673558.56 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:37:50,618 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-09 13:38:14,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.222e+02 1.403e+03 1.877e+03 2.638e+03 8.867e+03, threshold=3.755e+03, percent-clipped=8.0
+2023-03-09 13:38:43,844 INFO [train.py:968] (0/2) Epoch 18, batch 33900, giga_loss[loss=0.233, simple_loss=0.3247, pruned_loss=0.07059, over 28889.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3298, pruned_loss=0.0858, over 5665176.30 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3468, pruned_loss=0.1118, over 5675063.66 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3287, pruned_loss=0.08337, over 5672894.64 frames. ], batch size: 227, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:38:56,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8494, 2.0334, 1.4178, 1.5662], device='cuda:0'), covar=tensor([0.0988, 0.0646, 0.1041, 0.1228], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0443, 0.0513, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 13:39:27,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-09 13:39:40,098 INFO [train.py:968] (0/2) Epoch 18, batch 33950, giga_loss[loss=0.2451, simple_loss=0.3311, pruned_loss=0.07951, over 28829.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3314, pruned_loss=0.08495, over 5671804.66 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3467, pruned_loss=0.1118, over 5677870.81 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3302, pruned_loss=0.08226, over 5675456.43 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:40:08,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4706, 2.1048, 1.5198, 0.7266], device='cuda:0'), covar=tensor([0.5106, 0.2748, 0.4035, 0.5733], device='cuda:0'), in_proj_covar=tensor([0.1699, 0.1608, 0.1569, 0.1389], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 13:40:10,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.832e+02 1.385e+03 1.785e+03 2.613e+03 9.911e+03, threshold=3.570e+03, percent-clipped=14.0
+2023-03-09 13:40:18,667 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:23,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:23,342 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:25,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:27,600 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:38,099 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:40,436 INFO [train.py:968] (0/2) Epoch 18, batch 34000, giga_loss[loss=0.244, simple_loss=0.3352, pruned_loss=0.07637, over 29048.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.332, pruned_loss=0.08475, over 5670422.17 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3463, pruned_loss=0.1116, over 5672153.63 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3311, pruned_loss=0.08237, over 5677647.11 frames. ], batch size: 155, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:40:59,287 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:41:02,634 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:41:45,630 INFO [train.py:968] (0/2) Epoch 18, batch 34050, giga_loss[loss=0.2406, simple_loss=0.3273, pruned_loss=0.07693, over 28642.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3325, pruned_loss=0.08546, over 5673078.42 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3455, pruned_loss=0.1111, over 5678594.82 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.332, pruned_loss=0.08304, over 5672830.99 frames. ], batch size: 262, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:42:06,445 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-09 13:42:15,378 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5412, 1.8095, 1.4402, 1.6099], device='cuda:0'), covar=tensor([0.2713, 0.2572, 0.2899, 0.2418], device='cuda:0'), in_proj_covar=tensor([0.1444, 0.1045, 0.1283, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 13:42:20,611 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=810787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:42:20,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.905e+02 1.276e+03 1.689e+03 2.249e+03 7.400e+03, threshold=3.377e+03, percent-clipped=5.0
+2023-03-09 13:42:32,534 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:42:54,833 INFO [train.py:968] (0/2) Epoch 18, batch 34100, giga_loss[loss=0.2497, simple_loss=0.3438, pruned_loss=0.07782, over 28458.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3324, pruned_loss=0.08519, over 5666529.03 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3456, pruned_loss=0.1111, over 5678825.97 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3318, pruned_loss=0.08299, over 5665890.46 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:43:47,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=810845.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:43:51,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3025, 1.3229, 1.2074, 1.4673], device='cuda:0'), covar=tensor([0.0782, 0.0368, 0.0366, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 13:44:11,691 INFO [train.py:968] (0/2) Epoch 18, batch 34150, giga_loss[loss=0.2287, simple_loss=0.3282, pruned_loss=0.06456, over 28935.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3337, pruned_loss=0.08549, over 5665958.75 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3458, pruned_loss=0.1112, over 5679524.84 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3331, pruned_loss=0.08357, over 5664810.51 frames. ], batch size: 155, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:44:41,044 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810879.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:44:56,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.053e+02 1.408e+03 1.703e+03 2.740e+03 9.389e+03, threshold=3.407e+03, percent-clipped=15.0
+2023-03-09 13:45:28,807 INFO [train.py:968] (0/2) Epoch 18, batch 34200, giga_loss[loss=0.245, simple_loss=0.3367, pruned_loss=0.07662, over 28456.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3352, pruned_loss=0.08592, over 5659126.01 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3459, pruned_loss=0.1114, over 5672331.55 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3344, pruned_loss=0.08397, over 5664276.68 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:45:51,352 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:45:56,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:46:15,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:46:24,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3779, 1.5581, 1.4934, 1.4486], device='cuda:0'), covar=tensor([0.0763, 0.0329, 0.0312, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 13:46:29,350 INFO [train.py:968] (0/2) Epoch 18, batch 34250, libri_loss[loss=0.2637, simple_loss=0.3406, pruned_loss=0.09343, over 29658.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3386, pruned_loss=0.08815, over 5660032.23 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3458, pruned_loss=0.1113, over 5669101.02 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3377, pruned_loss=0.08569, over 5667391.51 frames. ], batch size: 88, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:46:29,560 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:46:46,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4157, 1.6631, 1.3433, 1.2576], device='cuda:0'), covar=tensor([0.2626, 0.2661, 0.3157, 0.2395], device='cuda:0'), in_proj_covar=tensor([0.1438, 0.1043, 0.1280, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 13:47:00,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.545e+02 1.431e+03 1.753e+03 2.326e+03 4.530e+03, threshold=3.506e+03, percent-clipped=9.0
+2023-03-09 13:47:04,704 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810988.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:47:05,332 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:47:08,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810991.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:47:32,126 INFO [train.py:968] (0/2) Epoch 18, batch 34300, libri_loss[loss=0.2524, simple_loss=0.3208, pruned_loss=0.09204, over 29608.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3377, pruned_loss=0.08752, over 5663303.87 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3449, pruned_loss=0.1108, over 5667808.19 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3376, pruned_loss=0.08531, over 5670457.21 frames. ], batch size: 74, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:47:42,852 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811020.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:48:40,642 INFO [train.py:968] (0/2) Epoch 18, batch 34350, giga_loss[loss=0.2701, simple_loss=0.3508, pruned_loss=0.09476, over 28367.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3378, pruned_loss=0.08867, over 5667743.54 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3455, pruned_loss=0.1111, over 5674482.28 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.337, pruned_loss=0.08591, over 5667424.81 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:48:56,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3144, 1.5993, 1.4947, 1.2025], device='cuda:0'), covar=tensor([0.3116, 0.2052, 0.1628, 0.2293], device='cuda:0'), in_proj_covar=tensor([0.1869, 0.1772, 0.1697, 0.1849], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 13:49:01,222 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:49:09,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:49:12,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.343e+02 1.325e+03 1.860e+03 2.516e+03 1.595e+04, threshold=3.721e+03, percent-clipped=9.0
+2023-03-09 13:49:43,546 INFO [train.py:968] (0/2) Epoch 18, batch 34400, giga_loss[loss=0.2306, simple_loss=0.3214, pruned_loss=0.0699, over 28975.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3351, pruned_loss=0.08745, over 5684029.06 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3447, pruned_loss=0.1106, over 5678092.79 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3349, pruned_loss=0.08483, over 5680512.42 frames. ], batch size: 136, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:50:52,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1665, 5.0189, 4.7481, 2.2379], device='cuda:0'), covar=tensor([0.0390, 0.0495, 0.0658, 0.1903], device='cuda:0'), in_proj_covar=tensor([0.1148, 0.1059, 0.0909, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 13:51:00,448 INFO [train.py:968] (0/2) Epoch 18, batch 34450, giga_loss[loss=0.2415, simple_loss=0.3306, pruned_loss=0.07623, over 28470.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3332, pruned_loss=0.08546, over 5683549.38 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3445, pruned_loss=0.1104, over 5680514.05 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3332, pruned_loss=0.08334, over 5678480.96 frames. ], batch size: 369, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:51:09,095 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:51:38,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.344e+02 1.223e+03 1.550e+03 2.103e+03 5.984e+03, threshold=3.100e+03, percent-clipped=3.0
+2023-03-09 13:51:46,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5371, 1.6037, 1.7839, 1.3751], device='cuda:0'), covar=tensor([0.1837, 0.2458, 0.1500, 0.1783], device='cuda:0'), in_proj_covar=tensor([0.0876, 0.0690, 0.0921, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 13:52:05,601 INFO [train.py:968] (0/2) Epoch 18, batch 34500, giga_loss[loss=0.2617, simple_loss=0.347, pruned_loss=0.08815, over 28142.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3334, pruned_loss=0.08553, over 5655675.72 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3448, pruned_loss=0.1107, over 5662386.80 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3329, pruned_loss=0.08323, over 5668752.46 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:52:19,473 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:52:22,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:52:28,004 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:52:30,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:52:53,607 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:52:56,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:53:02,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:53:05,237 INFO [train.py:968] (0/2) Epoch 18, batch 34550, giga_loss[loss=0.335, simple_loss=0.3951, pruned_loss=0.1374, over 27561.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3368, pruned_loss=0.08806, over 5656698.29 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3449, pruned_loss=0.1108, over 5664236.61 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3361, pruned_loss=0.08541, over 5665393.33 frames. ], batch size: 472, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:53:34,978 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:53:36,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.126e+02 1.538e+03 2.091e+03 2.849e+03 9.215e+03, threshold=4.182e+03, percent-clipped=22.0
+2023-03-09 13:53:53,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1090, 1.1420, 3.4145, 2.9662], device='cuda:0'), covar=tensor([0.1603, 0.2702, 0.0506, 0.0951], device='cuda:0'), in_proj_covar=tensor([0.0723, 0.0627, 0.0917, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 13:53:59,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:53:59,748 INFO [train.py:968] (0/2) Epoch 18, batch 34600, libri_loss[loss=0.2785, simple_loss=0.3471, pruned_loss=0.105, over 29532.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3377, pruned_loss=0.08895, over 5667309.85 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3445, pruned_loss=0.1106, over 5665709.95 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3371, pruned_loss=0.08601, over 5672522.42 frames. ], batch size: 89, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:54:01,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2556, 1.3073, 1.1084, 0.9074], device='cuda:0'), covar=tensor([0.0919, 0.0474, 0.1089, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0441, 0.0510, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 13:54:01,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:54:15,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:54:38,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:54:51,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:54:58,753 INFO [train.py:968] (0/2) Epoch 18, batch 34650, giga_loss[loss=0.2154, simple_loss=0.3042, pruned_loss=0.06324, over 29029.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3346, pruned_loss=0.08838, over 5665017.49 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.344, pruned_loss=0.1103, over 5671772.43 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3344, pruned_loss=0.08582, over 5663491.38 frames. ], batch size: 155, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:55:01,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:55:25,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.006e+02 1.455e+03 1.981e+03 2.661e+03 6.922e+03, threshold=3.963e+03, percent-clipped=7.0
+2023-03-09 13:55:34,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:55:36,907 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811397.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:55:40,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811400.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:55:53,885 INFO [train.py:968] (0/2) Epoch 18, batch 34700, libri_loss[loss=0.2573, simple_loss=0.3258, pruned_loss=0.09437, over 27799.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3343, pruned_loss=0.08879, over 5669578.07 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.344, pruned_loss=0.1103, over 5677154.50 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3338, pruned_loss=0.086, over 5663451.51 frames. ], batch size: 116, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:56:14,694 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811429.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:56:26,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3208, 1.4590, 1.2470, 1.4154], device='cuda:0'), covar=tensor([0.0785, 0.0345, 0.0353, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 13:56:49,652 INFO [train.py:968] (0/2) Epoch 18, batch 34750, giga_loss[loss=0.2768, simple_loss=0.36, pruned_loss=0.09676, over 28911.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3357, pruned_loss=0.09014, over 5661291.06 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3441, pruned_loss=0.1103, over 5673997.73 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.335, pruned_loss=0.08734, over 5658921.58 frames. ], batch size: 199, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:56:53,799 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:56:55,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:57:06,061 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-09 13:57:17,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.983e+02 1.597e+03 1.980e+03 2.600e+03 8.892e+03, threshold=3.960e+03, percent-clipped=9.0
+2023-03-09 13:57:24,203 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:57:31,612 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:57:34,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:57:35,933 INFO [train.py:968] (0/2) Epoch 18, batch 34800, giga_loss[loss=0.2814, simple_loss=0.3699, pruned_loss=0.09645, over 28812.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3457, pruned_loss=0.09564, over 5673166.06 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3442, pruned_loss=0.1103, over 5675481.24 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.345, pruned_loss=0.09314, over 5669856.57 frames. ], batch size: 243, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:58:02,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:58:25,446 INFO [train.py:968] (0/2) Epoch 18, batch 34850, giga_loss[loss=0.2687, simple_loss=0.3603, pruned_loss=0.08857, over 28854.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3525, pruned_loss=0.09958, over 5660728.17 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3447, pruned_loss=0.1108, over 5666857.19 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3516, pruned_loss=0.09696, over 5665807.91 frames. ], batch size: 199, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:58:42,085 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811579.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:58:51,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.653e+02 1.332e+03 1.760e+03 2.727e+03 6.156e+03, threshold=3.521e+03, percent-clipped=5.0
+2023-03-09 13:59:02,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6809, 4.8897, 1.9347, 2.1739], device='cuda:0'), covar=tensor([0.1146, 0.0275, 0.0933, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0536, 0.0370, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 13:59:05,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7768, 1.5180, 1.8408, 1.4063], device='cuda:0'), covar=tensor([0.2263, 0.3354, 0.1712, 0.1893], device='cuda:0'), in_proj_covar=tensor([0.0878, 0.0691, 0.0924, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 13:59:08,185 INFO [train.py:968] (0/2) Epoch 18, batch 34900, giga_loss[loss=0.2381, simple_loss=0.3242, pruned_loss=0.07602, over 28963.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3518, pruned_loss=0.1002, over 5662204.80 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3442, pruned_loss=0.1105, over 5662618.86 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3517, pruned_loss=0.09812, over 5669691.42 frames. ], batch size: 136, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:59:13,866 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:59:35,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7391, 1.8838, 1.9608, 1.5388], device='cuda:0'), covar=tensor([0.1750, 0.2387, 0.1437, 0.1646], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0691, 0.0924, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 13:59:50,696 INFO [train.py:968] (0/2) Epoch 18, batch 34950, giga_loss[loss=0.226, simple_loss=0.311, pruned_loss=0.07047, over 28952.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3456, pruned_loss=0.09765, over 5678401.74 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.344, pruned_loss=0.1103, over 5665859.41 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3458, pruned_loss=0.09606, over 5681679.97 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:59:51,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:00:14,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.434e+02 1.101e+03 1.405e+03 1.793e+03 7.018e+03, threshold=2.811e+03, percent-clipped=4.0
+2023-03-09 14:00:36,330 INFO [train.py:968] (0/2) Epoch 18, batch 35000, giga_loss[loss=0.2349, simple_loss=0.3007, pruned_loss=0.08449, over 28689.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3393, pruned_loss=0.09525, over 5662622.13 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3445, pruned_loss=0.1105, over 5650667.36 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.339, pruned_loss=0.09339, over 5678932.19 frames. ], batch size: 85, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:00:48,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:01:15,176 INFO [train.py:968] (0/2) Epoch 18, batch 35050, giga_loss[loss=0.2352, simple_loss=0.291, pruned_loss=0.08969, over 24035.00 frames. ], tot_loss[loss=0.259, simple_loss=0.333, pruned_loss=0.09247, over 5673282.83 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3454, pruned_loss=0.1107, over 5660208.88 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3317, pruned_loss=0.09017, over 5678665.53 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:01:21,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:01:39,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.113e+02 1.025e+03 1.336e+03 1.920e+03 7.207e+03, threshold=2.671e+03, percent-clipped=9.0
+2023-03-09 14:01:50,236 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:01:53,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:01:55,784 INFO [train.py:968] (0/2) Epoch 18, batch 35100, libri_loss[loss=0.3187, simple_loss=0.382, pruned_loss=0.1277, over 29263.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3262, pruned_loss=0.08949, over 5682308.96 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3456, pruned_loss=0.1105, over 5664081.08 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3245, pruned_loss=0.08735, over 5683238.44 frames. ], batch size: 94, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:02:16,932 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:02:27,721 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:02:37,600 INFO [train.py:968] (0/2) Epoch 18, batch 35150, giga_loss[loss=0.2079, simple_loss=0.2848, pruned_loss=0.06554, over 28385.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3207, pruned_loss=0.08688, over 5687172.75 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.346, pruned_loss=0.1104, over 5668684.37 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3181, pruned_loss=0.08435, over 5684535.11 frames. ], batch size: 60, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:02:47,625 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:02:49,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:03:03,340 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.245e+02 1.097e+03 1.402e+03 1.930e+03 5.747e+03, threshold=2.804e+03, percent-clipped=14.0
+2023-03-09 14:03:08,398 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-09 14:03:13,874 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:03:18,722 INFO [train.py:968] (0/2) Epoch 18, batch 35200, giga_loss[loss=0.2121, simple_loss=0.2844, pruned_loss=0.06991, over 28786.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3165, pruned_loss=0.08492, over 5691575.68 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3457, pruned_loss=0.1102, over 5674526.13 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3141, pruned_loss=0.08257, over 5684586.15 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:03:19,210 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:03:22,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:03:49,021 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:03:58,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811954.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:04:04,194 INFO [train.py:968] (0/2) Epoch 18, batch 35250, giga_loss[loss=0.2151, simple_loss=0.2926, pruned_loss=0.06874, over 28666.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3123, pruned_loss=0.08247, over 5698787.55 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3455, pruned_loss=0.11, over 5677981.41 frames. ], giga_tot_loss[loss=0.2355, simple_loss=0.3102, pruned_loss=0.08038, over 5690430.25 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:04:14,850 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-09 14:04:27,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.286e+02 1.032e+03 1.289e+03 1.613e+03 4.447e+03, threshold=2.578e+03, percent-clipped=5.0
+2023-03-09 14:04:28,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:04:34,702 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-812000.pt
+2023-03-09 14:04:44,469 INFO [train.py:968] (0/2) Epoch 18, batch 35300, giga_loss[loss=0.1986, simple_loss=0.2784, pruned_loss=0.05942, over 28845.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3115, pruned_loss=0.08276, over 5704980.54 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3462, pruned_loss=0.1103, over 5679978.74 frames. ], giga_tot_loss[loss=0.2339, simple_loss=0.3079, pruned_loss=0.07997, over 5697115.73 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:05:14,858 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-09 14:05:26,645 INFO [train.py:968] (0/2) Epoch 18, batch 35350, giga_loss[loss=0.2145, simple_loss=0.289, pruned_loss=0.07004, over 27670.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3101, pruned_loss=0.08208, over 5700660.64 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3468, pruned_loss=0.1104, over 5673554.21 frames. ], giga_tot_loss[loss=0.2318, simple_loss=0.3057, pruned_loss=0.07894, over 5701446.53 frames. ], batch size: 472, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:05:51,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.882e+02 1.068e+03 1.502e+03 2.075e+03 1.618e+04, threshold=3.003e+03, percent-clipped=17.0
+2023-03-09 14:05:56,759 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812097.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:05:59,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812100.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:06:09,091 INFO [train.py:968] (0/2) Epoch 18, batch 35400, giga_loss[loss=0.2088, simple_loss=0.293, pruned_loss=0.06232, over 28865.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.306, pruned_loss=0.07992, over 5697533.88 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3467, pruned_loss=0.1102, over 5677100.47 frames. ], giga_tot_loss[loss=0.2282, simple_loss=0.302, pruned_loss=0.07721, over 5695348.32 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:06:23,401 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=812129.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:06:23,551 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-09 14:06:28,977 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:06:31,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:06:43,743 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8329, 1.8835, 4.0017, 3.6147], device='cuda:0'), covar=tensor([0.1212, 0.2208, 0.0427, 0.1044], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0634, 0.0933, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 14:06:49,578 INFO [train.py:968] (0/2) Epoch 18, batch 35450, giga_loss[loss=0.2043, simple_loss=0.278, pruned_loss=0.06533, over 28747.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3044, pruned_loss=0.07933, over 5681372.19 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3475, pruned_loss=0.1105, over 5663598.27 frames. ], giga_tot_loss[loss=0.2258, simple_loss=0.2995, pruned_loss=0.07608, over 5693399.65 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:06:54,012 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=812167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:07:15,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.854e+02 1.015e+03 1.327e+03 1.865e+03 5.591e+03, threshold=2.655e+03, percent-clipped=6.0
+2023-03-09 14:07:31,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5163, 4.3448, 4.0738, 1.8146], device='cuda:0'), covar=tensor([0.0536, 0.0708, 0.0705, 0.2113], device='cuda:0'), in_proj_covar=tensor([0.1157, 0.1076, 0.0919, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 14:07:33,177 INFO [train.py:968] (0/2) Epoch 18, batch 35500, giga_loss[loss=0.2352, simple_loss=0.3061, pruned_loss=0.08214, over 28712.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3014, pruned_loss=0.07797, over 5676581.54 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3478, pruned_loss=0.1107, over 5658553.34 frames. ], giga_tot_loss[loss=0.223, simple_loss=0.2965, pruned_loss=0.07471, over 5690991.64 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:07:43,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=812222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:08:09,591 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 14:08:21,561 INFO [train.py:968] (0/2) Epoch 18, batch 35550, giga_loss[loss=0.2153, simple_loss=0.2909, pruned_loss=0.06983, over 28058.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.2971, pruned_loss=0.07586, over 5687606.94 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3477, pruned_loss=0.1105, over 5662126.10 frames. ], giga_tot_loss[loss=0.2192, simple_loss=0.2926, pruned_loss=0.07293, over 5695963.43 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:08:36,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=812279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:08:47,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.214e+02 1.089e+03 1.406e+03 2.073e+03 6.722e+03, threshold=2.812e+03, percent-clipped=13.0
+2023-03-09 14:09:03,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6296, 1.9565, 1.5898, 1.7185], device='cuda:0'), covar=tensor([0.1780, 0.2051, 0.2192, 0.2015], device='cuda:0'), in_proj_covar=tensor([0.0449, 0.0738, 0.0695, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 14:09:06,406 INFO [train.py:968] (0/2) Epoch 18, batch 35600, giga_loss[loss=0.2831, simple_loss=0.3577, pruned_loss=0.1042, over 28246.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3002, pruned_loss=0.078, over 5665781.77 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3482, pruned_loss=0.1107, over 5646604.01 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2954, pruned_loss=0.07493, over 5686878.82 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:09:10,453 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=812316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:09:20,210 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=812326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:09:55,248 INFO [train.py:968] (0/2) Epoch 18, batch 35650, giga_loss[loss=0.2807, simple_loss=0.3562, pruned_loss=0.1026, over 28805.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3118, pruned_loss=0.08343, over 5675165.13 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3483, pruned_loss=0.1108, over 5650682.64 frames. ], giga_tot_loss[loss=0.2342, simple_loss=0.3073, pruned_loss=0.08058, over 5688478.74 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:09:59,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:10:01,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:10:07,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-09 14:10:24,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.887e+02 1.314e+03 1.546e+03 2.253e+03 5.175e+03, threshold=3.092e+03, percent-clipped=12.0
+2023-03-09 14:10:29,173 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=812397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:10:42,422 INFO [train.py:968] (0/2) Epoch 18, batch 35700, giga_loss[loss=0.313, simple_loss=0.3883, pruned_loss=0.1188, over 28706.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.326, pruned_loss=0.09108, over 5681947.05 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3489, pruned_loss=0.1112, over 5655878.16 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3214, pruned_loss=0.08806, over 5688302.59 frames. ], batch size: 262, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:11:26,441 INFO [train.py:968] (0/2) Epoch 18, batch 35750, giga_loss[loss=0.2806, simple_loss=0.345, pruned_loss=0.1081, over 23737.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3356, pruned_loss=0.09597, over 5676682.72 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.349, pruned_loss=0.1112, over 5652758.86 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3312, pruned_loss=0.09294, over 5684701.84 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:11:50,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.379e+03 1.806e+03 2.627e+03 9.200e+03, threshold=3.612e+03, percent-clipped=14.0
+2023-03-09 14:11:51,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=812495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:12:06,060 INFO [train.py:968] (0/2) Epoch 18, batch 35800, giga_loss[loss=0.2492, simple_loss=0.3361, pruned_loss=0.08115, over 28978.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3405, pruned_loss=0.09695, over 5684075.47 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3495, pruned_loss=0.1115, over 5650338.78 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3361, pruned_loss=0.09383, over 5693190.14 frames. ], batch size: 213, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:12:52,093 INFO [train.py:968] (0/2) Epoch 18, batch 35850, giga_loss[loss=0.2679, simple_loss=0.3425, pruned_loss=0.09667, over 28726.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3427, pruned_loss=0.09691, over 5678294.01 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3503, pruned_loss=0.112, over 5646532.70 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3383, pruned_loss=0.09364, over 5689159.85 frames. ], batch size: 92, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:13:03,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5747, 0.9959, 4.8472, 3.4222], device='cuda:0'), covar=tensor([0.1716, 0.3140, 0.0355, 0.1027], device='cuda:0'), in_proj_covar=tensor([0.0728, 0.0631, 0.0928, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 14:13:23,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.267e+02 1.172e+03 1.440e+03 1.935e+03 9.646e+03, threshold=2.881e+03, percent-clipped=6.0
+2023-03-09 14:13:38,758 INFO [train.py:968] (0/2) Epoch 18, batch 35900, libri_loss[loss=0.3473, simple_loss=0.4066, pruned_loss=0.144, over 29539.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3446, pruned_loss=0.09775, over 5683458.96 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3505, pruned_loss=0.1121, over 5651221.24 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3408, pruned_loss=0.09471, over 5688365.66 frames. ], batch size: 89, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:13:57,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4644, 1.9535, 1.3945, 1.5700], device='cuda:0'), covar=tensor([0.0724, 0.0271, 0.0331, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 14:14:14,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=812654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:14:21,663 INFO [train.py:968] (0/2) Epoch 18, batch 35950, giga_loss[loss=0.2765, simple_loss=0.3583, pruned_loss=0.09739, over 28929.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3478, pruned_loss=0.1007, over 5681352.52 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3505, pruned_loss=0.1119, over 5658193.94 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3446, pruned_loss=0.09802, over 5679953.18 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:14:47,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=812691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:14:47,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5469, 2.1523, 1.5499, 0.6721], device='cuda:0'), covar=tensor([0.5447, 0.2896, 0.4004, 0.6171], device='cuda:0'), in_proj_covar=tensor([0.1702, 0.1617, 0.1582, 0.1387], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 14:14:49,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.648e+02 1.244e+03 1.637e+03 2.311e+03 1.062e+04, threshold=3.274e+03, percent-clipped=14.0
+2023-03-09 14:14:55,442 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=812701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:14:58,660 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-09 14:15:04,608 INFO [train.py:968] (0/2) Epoch 18, batch 36000, libri_loss[loss=0.3465, simple_loss=0.4007, pruned_loss=0.1462, over 29196.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3512, pruned_loss=0.1029, over 5682455.25 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3507, pruned_loss=0.112, over 5659110.79 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3484, pruned_loss=0.1005, over 5680719.21 frames. ], batch size: 97, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:15:04,613 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 14:15:14,356 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2061, simple_loss=0.3137, pruned_loss=0.04927, over 944034.00 frames. 
+2023-03-09 14:15:14,356 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 14:15:57,772 INFO [train.py:968] (0/2) Epoch 18, batch 36050, giga_loss[loss=0.3021, simple_loss=0.3729, pruned_loss=0.1156, over 29081.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3544, pruned_loss=0.1046, over 5680967.09 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3513, pruned_loss=0.1124, over 5655995.33 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3516, pruned_loss=0.1021, over 5682490.80 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:16:20,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.702e+02 1.282e+03 1.562e+03 2.237e+03 6.088e+03, threshold=3.123e+03, percent-clipped=6.0
+2023-03-09 14:16:24,195 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:16:26,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:16:36,209 INFO [train.py:968] (0/2) Epoch 18, batch 36100, giga_loss[loss=0.2789, simple_loss=0.3579, pruned_loss=0.09996, over 28297.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3562, pruned_loss=0.1048, over 5694214.76 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3513, pruned_loss=0.1122, over 5663606.88 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3542, pruned_loss=0.1028, over 5689391.29 frames. ], batch size: 77, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:16:50,522 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=812829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:16:55,309 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:16:58,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:04,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:08,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:09,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5389, 1.5371, 1.7587, 1.3486], device='cuda:0'), covar=tensor([0.1663, 0.2402, 0.1365, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.0882, 0.0694, 0.0928, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 14:17:20,954 INFO [train.py:968] (0/2) Epoch 18, batch 36150, giga_loss[loss=0.3541, simple_loss=0.4013, pruned_loss=0.1535, over 26577.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3573, pruned_loss=0.105, over 5682989.85 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3514, pruned_loss=0.1123, over 5668436.88 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3557, pruned_loss=0.1031, over 5675409.66 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:17:23,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2238, 1.6068, 1.5605, 1.0984], device='cuda:0'), covar=tensor([0.1520, 0.2466, 0.1351, 0.1593], device='cuda:0'), in_proj_covar=tensor([0.0883, 0.0694, 0.0928, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 14:17:24,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=812866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:27,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=812870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:32,029 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=812876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:43,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.920e+02 1.160e+03 1.393e+03 1.965e+03 7.094e+03, threshold=2.786e+03, percent-clipped=6.0
+2023-03-09 14:18:02,008 INFO [train.py:968] (0/2) Epoch 18, batch 36200, giga_loss[loss=0.2722, simple_loss=0.3559, pruned_loss=0.0942, over 28968.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3573, pruned_loss=0.1036, over 5697107.77 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3511, pruned_loss=0.1121, over 5673926.96 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3565, pruned_loss=0.1022, over 5686793.84 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:18:42,382 INFO [train.py:968] (0/2) Epoch 18, batch 36250, giga_loss[loss=0.2751, simple_loss=0.3638, pruned_loss=0.09322, over 29013.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3565, pruned_loss=0.1018, over 5704017.31 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3519, pruned_loss=0.1125, over 5673125.80 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3552, pruned_loss=0.1001, over 5696813.24 frames. ], batch size: 155, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:19:09,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.492e+02 1.067e+03 1.344e+03 1.912e+03 5.828e+03, threshold=2.688e+03, percent-clipped=9.0
+2023-03-09 14:19:25,394 INFO [train.py:968] (0/2) Epoch 18, batch 36300, giga_loss[loss=0.267, simple_loss=0.3494, pruned_loss=0.09229, over 29019.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3549, pruned_loss=0.09955, over 5708358.65 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3524, pruned_loss=0.1127, over 5675007.81 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3534, pruned_loss=0.09778, over 5701347.56 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:19:26,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=813013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:19:28,531 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=813016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:19:50,742 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=813045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:20:01,972 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=813061.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:20:02,362 INFO [train.py:968] (0/2) Epoch 18, batch 36350, giga_loss[loss=0.3348, simple_loss=0.4001, pruned_loss=0.1347, over 28299.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3543, pruned_loss=0.09953, over 5715672.52 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3523, pruned_loss=0.1124, over 5682753.08 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3532, pruned_loss=0.09773, over 5704383.79 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:20:21,982 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-09 14:20:29,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.770e+02 1.366e+03 1.648e+03 2.370e+03 7.727e+03, threshold=3.296e+03, percent-clipped=14.0
+2023-03-09 14:20:47,622 INFO [train.py:968] (0/2) Epoch 18, batch 36400, giga_loss[loss=0.2953, simple_loss=0.3622, pruned_loss=0.1142, over 28872.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3558, pruned_loss=0.1024, over 5717010.63 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3526, pruned_loss=0.1124, over 5689347.26 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3548, pruned_loss=0.1006, over 5702859.72 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:21:04,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9427, 1.9237, 2.1352, 1.6924], device='cuda:0'), covar=tensor([0.1728, 0.2289, 0.1315, 0.1642], device='cuda:0'), in_proj_covar=tensor([0.0881, 0.0693, 0.0927, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 14:21:30,114 INFO [train.py:968] (0/2) Epoch 18, batch 36450, giga_loss[loss=0.27, simple_loss=0.3434, pruned_loss=0.09827, over 28680.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3595, pruned_loss=0.1074, over 5710900.93 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3527, pruned_loss=0.1123, over 5694260.97 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3588, pruned_loss=0.1057, over 5695698.04 frames. ], batch size: 60, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:21:41,574 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 14:21:59,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.781e+02 1.280e+03 1.575e+03 2.379e+03 6.901e+03, threshold=3.151e+03, percent-clipped=7.0
+2023-03-09 14:22:14,929 INFO [train.py:968] (0/2) Epoch 18, batch 36500, giga_loss[loss=0.2458, simple_loss=0.3249, pruned_loss=0.0834, over 29012.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3589, pruned_loss=0.1084, over 5703037.54 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3531, pruned_loss=0.1125, over 5688095.04 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3581, pruned_loss=0.1068, over 5696356.20 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:22:56,650 INFO [train.py:968] (0/2) Epoch 18, batch 36550, giga_loss[loss=0.268, simple_loss=0.3378, pruned_loss=0.09909, over 29047.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3561, pruned_loss=0.1072, over 5699522.99 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3533, pruned_loss=0.1126, over 5684205.56 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3554, pruned_loss=0.1058, over 5698662.14 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:23:23,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.411e+02 1.248e+03 1.517e+03 1.959e+03 8.362e+03, threshold=3.035e+03, percent-clipped=7.0
+2023-03-09 14:23:36,939 INFO [train.py:968] (0/2) Epoch 18, batch 36600, giga_loss[loss=0.3176, simple_loss=0.3798, pruned_loss=0.1277, over 28919.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3562, pruned_loss=0.1076, over 5688115.09 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3538, pruned_loss=0.1129, over 5670412.87 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3551, pruned_loss=0.106, over 5700941.38 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:23:54,843 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1400, 1.7874, 1.3689, 0.3908], device='cuda:0'), covar=tensor([0.4048, 0.2445, 0.3421, 0.4439], device='cuda:0'), in_proj_covar=tensor([0.1697, 0.1613, 0.1576, 0.1386], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 14:24:20,887 INFO [train.py:968] (0/2) Epoch 18, batch 36650, giga_loss[loss=0.2643, simple_loss=0.3412, pruned_loss=0.09364, over 28599.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3547, pruned_loss=0.1061, over 5685234.16 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3538, pruned_loss=0.1128, over 5667061.46 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.354, pruned_loss=0.1048, over 5699111.41 frames. ], batch size: 85, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:24:53,265 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.746e+02 1.181e+03 1.686e+03 2.429e+03 6.981e+03, threshold=3.372e+03, percent-clipped=12.0
+2023-03-09 14:25:08,139 INFO [train.py:968] (0/2) Epoch 18, batch 36700, giga_loss[loss=0.2766, simple_loss=0.3503, pruned_loss=0.1014, over 28573.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3517, pruned_loss=0.1033, over 5681363.07 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.354, pruned_loss=0.113, over 5670767.39 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3509, pruned_loss=0.102, over 5689152.86 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:25:31,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=813436.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:25:39,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-09 14:25:55,403 INFO [train.py:968] (0/2) Epoch 18, batch 36750, giga_loss[loss=0.2402, simple_loss=0.3159, pruned_loss=0.08227, over 29055.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3466, pruned_loss=0.1003, over 5674122.19 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3542, pruned_loss=0.1132, over 5672321.81 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3457, pruned_loss=0.09901, over 5679087.32 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:26:31,078 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.321e+02 1.011e+03 1.227e+03 1.645e+03 6.897e+03, threshold=2.455e+03, percent-clipped=1.0
+2023-03-09 14:26:36,063 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 14:26:49,682 INFO [train.py:968] (0/2) Epoch 18, batch 36800, giga_loss[loss=0.2617, simple_loss=0.325, pruned_loss=0.09921, over 26496.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3396, pruned_loss=0.09701, over 5652628.61 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3545, pruned_loss=0.1134, over 5663961.21 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3387, pruned_loss=0.0957, over 5664303.91 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:27:33,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2528, 1.3043, 3.7821, 3.1382], device='cuda:0'), covar=tensor([0.1615, 0.2753, 0.0402, 0.1046], device='cuda:0'), in_proj_covar=tensor([0.0725, 0.0629, 0.0923, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 14:27:45,184 INFO [train.py:968] (0/2) Epoch 18, batch 36850, giga_loss[loss=0.2508, simple_loss=0.3234, pruned_loss=0.08905, over 28844.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3348, pruned_loss=0.09488, over 5633151.52 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3547, pruned_loss=0.1136, over 5656233.68 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3336, pruned_loss=0.09342, over 5649702.55 frames. ], batch size: 199, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:28:01,340 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=813579.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:28:03,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=813582.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:28:10,243 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 14:28:14,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.062e+02 1.041e+03 1.284e+03 1.681e+03 5.089e+03, threshold=2.568e+03, percent-clipped=4.0
+2023-03-09 14:28:29,439 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=813611.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:28:29,878 INFO [train.py:968] (0/2) Epoch 18, batch 36900, giga_loss[loss=0.2612, simple_loss=0.3356, pruned_loss=0.09338, over 28853.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3349, pruned_loss=0.09413, over 5645088.01 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3551, pruned_loss=0.1137, over 5651300.61 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3332, pruned_loss=0.09256, over 5662676.42 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:29:10,473 INFO [train.py:968] (0/2) Epoch 18, batch 36950, giga_loss[loss=0.2564, simple_loss=0.3371, pruned_loss=0.08785, over 29027.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3357, pruned_loss=0.09397, over 5636739.65 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3558, pruned_loss=0.1139, over 5636481.31 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3332, pruned_loss=0.09205, over 5664348.68 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:29:36,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.125e+02 1.092e+03 1.410e+03 1.706e+03 4.865e+03, threshold=2.820e+03, percent-clipped=11.0
+2023-03-09 14:29:53,319 INFO [train.py:968] (0/2) Epoch 18, batch 37000, giga_loss[loss=0.2335, simple_loss=0.3152, pruned_loss=0.07594, over 29003.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3355, pruned_loss=0.0936, over 5662177.92 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3563, pruned_loss=0.1139, over 5641070.65 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3327, pruned_loss=0.09166, over 5680241.84 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:29:55,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4774, 1.8577, 1.4541, 1.7653], device='cuda:0'), covar=tensor([0.2566, 0.2632, 0.2876, 0.2261], device='cuda:0'), in_proj_covar=tensor([0.1443, 0.1050, 0.1283, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 14:30:32,987 INFO [train.py:968] (0/2) Epoch 18, batch 37050, giga_loss[loss=0.2413, simple_loss=0.3161, pruned_loss=0.08327, over 28833.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.333, pruned_loss=0.09268, over 5674074.30 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3565, pruned_loss=0.1139, over 5636386.05 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3303, pruned_loss=0.09083, over 5693846.22 frames. ], batch size: 92, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:30:42,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-09 14:31:00,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.663e+02 1.042e+03 1.255e+03 1.804e+03 8.135e+03, threshold=2.510e+03, percent-clipped=4.0
+2023-03-09 14:31:15,661 INFO [train.py:968] (0/2) Epoch 18, batch 37100, giga_loss[loss=0.2449, simple_loss=0.3198, pruned_loss=0.08499, over 28810.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3306, pruned_loss=0.09154, over 5683137.11 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3566, pruned_loss=0.1139, over 5639155.18 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3282, pruned_loss=0.08992, over 5696665.84 frames. ], batch size: 199, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:31:54,676 INFO [train.py:968] (0/2) Epoch 18, batch 37150, giga_loss[loss=0.2304, simple_loss=0.3019, pruned_loss=0.07942, over 28617.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3285, pruned_loss=0.09062, over 5684527.75 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3574, pruned_loss=0.1141, over 5633097.86 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3251, pruned_loss=0.08844, over 5701675.31 frames. ], batch size: 85, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:31:55,541 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=813863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:32:21,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.425e+02 1.019e+03 1.223e+03 1.765e+03 4.272e+03, threshold=2.447e+03, percent-clipped=12.0
+2023-03-09 14:32:35,475 INFO [train.py:968] (0/2) Epoch 18, batch 37200, giga_loss[loss=0.2435, simple_loss=0.3229, pruned_loss=0.082, over 28651.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3265, pruned_loss=0.08959, over 5691503.73 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3577, pruned_loss=0.1142, over 5632109.10 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3231, pruned_loss=0.08745, over 5706838.89 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:33:17,178 INFO [train.py:968] (0/2) Epoch 18, batch 37250, giga_loss[loss=0.2276, simple_loss=0.2988, pruned_loss=0.07817, over 28346.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3236, pruned_loss=0.08804, over 5702635.51 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3582, pruned_loss=0.1143, over 5636981.95 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3198, pruned_loss=0.08575, over 5711615.07 frames. ], batch size: 65, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:33:45,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.628e+02 1.108e+03 1.463e+03 2.043e+03 7.764e+03, threshold=2.925e+03, percent-clipped=16.0
+2023-03-09 14:33:46,793 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 14:33:47,663 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-814000.pt
+2023-03-09 14:33:54,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3295, 3.4789, 2.5435, 1.3411], device='cuda:0'), covar=tensor([0.6108, 0.1938, 0.3152, 0.5639], device='cuda:0'), in_proj_covar=tensor([0.1687, 0.1600, 0.1567, 0.1383], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 14:33:56,075 INFO [train.py:968] (0/2) Epoch 18, batch 37300, giga_loss[loss=0.2247, simple_loss=0.3055, pruned_loss=0.07199, over 28738.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3234, pruned_loss=0.08817, over 5687718.23 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3595, pruned_loss=0.115, over 5622368.56 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3182, pruned_loss=0.08497, over 5711636.97 frames. ], batch size: 262, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:34:35,882 INFO [train.py:968] (0/2) Epoch 18, batch 37350, giga_loss[loss=0.2319, simple_loss=0.3097, pruned_loss=0.07699, over 29032.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3219, pruned_loss=0.08739, over 5695041.68 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3598, pruned_loss=0.1149, over 5628091.27 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3167, pruned_loss=0.08436, over 5710124.59 frames. ], batch size: 155, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:34:43,290 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-09 14:35:04,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.754e+02 9.803e+02 1.197e+03 2.070e+03 8.470e+03, threshold=2.394e+03, percent-clipped=9.0
+2023-03-09 14:35:14,332 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:35:15,433 INFO [train.py:968] (0/2) Epoch 18, batch 37400, giga_loss[loss=0.2588, simple_loss=0.3261, pruned_loss=0.09579, over 28774.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3211, pruned_loss=0.08681, over 5701504.08 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.36, pruned_loss=0.1148, over 5633354.53 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3158, pruned_loss=0.08376, over 5710714.29 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:35:30,866 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814129.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:35:57,025 INFO [train.py:968] (0/2) Epoch 18, batch 37450, libri_loss[loss=0.3055, simple_loss=0.3798, pruned_loss=0.1156, over 27806.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3223, pruned_loss=0.08755, over 5709322.61 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3608, pruned_loss=0.1151, over 5641187.83 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.316, pruned_loss=0.08387, over 5711776.75 frames. ], batch size: 116, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:36:26,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.736e+02 1.103e+03 1.380e+03 1.906e+03 6.507e+03, threshold=2.760e+03, percent-clipped=14.0
+2023-03-09 14:36:37,823 INFO [train.py:968] (0/2) Epoch 18, batch 37500, giga_loss[loss=0.2646, simple_loss=0.3296, pruned_loss=0.0998, over 28708.00 frames. ], tot_loss[loss=0.254, simple_loss=0.327, pruned_loss=0.09054, over 5715300.96 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3611, pruned_loss=0.1151, over 5645770.10 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3211, pruned_loss=0.08717, over 5714333.04 frames. ], batch size: 92, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:36:58,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4644, 1.5519, 1.6945, 1.4368], device='cuda:0'), covar=tensor([0.2472, 0.2365, 0.1643, 0.2020], device='cuda:0'), in_proj_covar=tensor([0.1885, 0.1795, 0.1740, 0.1887], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 14:37:00,145 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=814238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:37:02,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2496, 0.7757, 0.8454, 1.4003], device='cuda:0'), covar=tensor([0.0791, 0.0381, 0.0367, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 14:37:11,054 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:37:24,455 INFO [train.py:968] (0/2) Epoch 18, batch 37550, giga_loss[loss=0.2384, simple_loss=0.3172, pruned_loss=0.0798, over 28586.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3332, pruned_loss=0.09458, over 5711566.01 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3614, pruned_loss=0.1152, over 5650674.74 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3277, pruned_loss=0.09136, over 5707785.07 frames. ], batch size: 60, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:37:49,176 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-09 14:37:57,837 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-09 14:37:57,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.326e+02 1.394e+03 1.863e+03 2.720e+03 1.439e+04, threshold=3.726e+03, percent-clipped=23.0
+2023-03-09 14:38:02,308 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:38:10,705 INFO [train.py:968] (0/2) Epoch 18, batch 37600, giga_loss[loss=0.2739, simple_loss=0.3502, pruned_loss=0.09877, over 28695.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3419, pruned_loss=0.1008, over 5705782.20 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3619, pruned_loss=0.1157, over 5657898.88 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3364, pruned_loss=0.09738, over 5697800.81 frames. ], batch size: 284, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:39:00,210 INFO [train.py:968] (0/2) Epoch 18, batch 37650, giga_loss[loss=0.2765, simple_loss=0.3483, pruned_loss=0.1024, over 28645.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3458, pruned_loss=0.1023, over 5689712.16 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3618, pruned_loss=0.1154, over 5655946.81 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3412, pruned_loss=0.09949, over 5685898.43 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:39:16,116 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=814381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:39:19,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=814384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:39:29,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.737e+02 1.173e+03 1.577e+03 1.972e+03 5.451e+03, threshold=3.153e+03, percent-clipped=2.0
+2023-03-09 14:39:45,330 INFO [train.py:968] (0/2) Epoch 18, batch 37700, giga_loss[loss=0.3001, simple_loss=0.3704, pruned_loss=0.1149, over 28717.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3495, pruned_loss=0.1031, over 5682297.88 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3617, pruned_loss=0.1154, over 5648753.45 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3457, pruned_loss=0.1007, over 5686327.07 frames. ], batch size: 262, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:39:46,213 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=814413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:40:27,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5803, 1.6798, 1.8203, 1.4022], device='cuda:0'), covar=tensor([0.1788, 0.2399, 0.1425, 0.1713], device='cuda:0'), in_proj_covar=tensor([0.0883, 0.0696, 0.0928, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 14:40:32,337 INFO [train.py:968] (0/2) Epoch 18, batch 37750, giga_loss[loss=0.2671, simple_loss=0.3488, pruned_loss=0.09267, over 28857.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3554, pruned_loss=0.1063, over 5681674.34 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3619, pruned_loss=0.1155, over 5653712.99 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.352, pruned_loss=0.1041, over 5681162.15 frames. ], batch size: 112, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:40:53,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=814485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:40:59,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-09 14:41:02,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.693e+02 1.357e+03 1.561e+03 2.163e+03 8.277e+03, threshold=3.122e+03, percent-clipped=11.0
+2023-03-09 14:41:08,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=814504.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:41:14,014 INFO [train.py:968] (0/2) Epoch 18, batch 37800, libri_loss[loss=0.3319, simple_loss=0.3912, pruned_loss=0.1363, over 27804.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3559, pruned_loss=0.1062, over 5688191.82 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3619, pruned_loss=0.1155, over 5661121.28 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.353, pruned_loss=0.1042, over 5682106.65 frames. ], batch size: 116, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:41:56,247 INFO [train.py:968] (0/2) Epoch 18, batch 37850, giga_loss[loss=0.2882, simple_loss=0.3699, pruned_loss=0.1032, over 28251.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3522, pruned_loss=0.1033, over 5696555.71 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3615, pruned_loss=0.1154, over 5665607.58 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3501, pruned_loss=0.1016, over 5688069.20 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:41:57,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9973, 2.1315, 1.4338, 1.7424], device='cuda:0'), covar=tensor([0.0751, 0.0452, 0.0921, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0442, 0.0511, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 14:42:26,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.238e+02 1.207e+03 1.393e+03 1.825e+03 5.475e+03, threshold=2.786e+03, percent-clipped=3.0
+2023-03-09 14:42:39,389 INFO [train.py:968] (0/2) Epoch 18, batch 37900, libri_loss[loss=0.2509, simple_loss=0.3267, pruned_loss=0.08754, over 29671.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3504, pruned_loss=0.1017, over 5695624.42 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3617, pruned_loss=0.1155, over 5666339.35 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3483, pruned_loss=0.09991, over 5689180.69 frames. ], batch size: 73, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:42:47,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=814622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:42:51,788 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=814628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:42:56,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=814631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:43:08,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=814647.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:43:11,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=814650.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:43:13,424 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:43:20,003 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=814660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:43:21,241 INFO [train.py:968] (0/2) Epoch 18, batch 37950, giga_loss[loss=0.2264, simple_loss=0.3074, pruned_loss=0.07275, over 28517.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3509, pruned_loss=0.1016, over 5705042.25 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3622, pruned_loss=0.1159, over 5672313.51 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3485, pruned_loss=0.09951, over 5695221.00 frames. ], batch size: 60, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:43:33,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=814677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:43:35,281 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=814679.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:43:39,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3332, 1.4853, 3.2372, 3.1845], device='cuda:0'), covar=tensor([0.1325, 0.2490, 0.0425, 0.0997], device='cuda:0'), in_proj_covar=tensor([0.0728, 0.0629, 0.0924, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 14:43:41,156 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814685.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:43:45,690 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814691.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:43:51,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.831e+02 1.305e+03 1.772e+03 2.406e+03 8.757e+03, threshold=3.544e+03, percent-clipped=17.0
+2023-03-09 14:44:05,514 INFO [train.py:968] (0/2) Epoch 18, batch 38000, giga_loss[loss=0.2929, simple_loss=0.3639, pruned_loss=0.111, over 28192.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3542, pruned_loss=0.1035, over 5690038.80 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3626, pruned_loss=0.1161, over 5660063.66 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3517, pruned_loss=0.1013, over 5694216.66 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:44:47,450 INFO [train.py:968] (0/2) Epoch 18, batch 38050, giga_loss[loss=0.2595, simple_loss=0.3352, pruned_loss=0.09193, over 28920.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3559, pruned_loss=0.1048, over 5701060.63 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3627, pruned_loss=0.1161, over 5666299.52 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3536, pruned_loss=0.1027, over 5699646.96 frames. ], batch size: 112, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:44:51,411 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=814765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:44:51,952 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:44:53,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=814768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:44:59,882 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 14:45:18,778 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=814797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:45:19,313 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.134e+02 1.350e+03 1.619e+03 2.088e+03 5.189e+03, threshold=3.237e+03, percent-clipped=4.0
+2023-03-09 14:45:29,694 INFO [train.py:968] (0/2) Epoch 18, batch 38100, giga_loss[loss=0.2937, simple_loss=0.3707, pruned_loss=0.1083, over 28863.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3584, pruned_loss=0.1069, over 5690697.48 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3632, pruned_loss=0.1166, over 5659761.88 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3559, pruned_loss=0.1045, over 5696943.99 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:45:39,321 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=814820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:45:43,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=814823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:46:07,843 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=814852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:46:16,304 INFO [train.py:968] (0/2) Epoch 18, batch 38150, libri_loss[loss=0.3112, simple_loss=0.3791, pruned_loss=0.1217, over 25642.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3587, pruned_loss=0.1076, over 5688267.30 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3632, pruned_loss=0.1164, over 5659984.30 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3567, pruned_loss=0.1057, over 5693736.26 frames. ], batch size: 136, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:46:31,004 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-09 14:46:47,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.056e+02 1.308e+03 1.647e+03 2.437e+03 9.961e+03, threshold=3.295e+03, percent-clipped=7.0
+2023-03-09 14:46:57,684 INFO [train.py:968] (0/2) Epoch 18, batch 38200, giga_loss[loss=0.3059, simple_loss=0.3784, pruned_loss=0.1168, over 28804.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3587, pruned_loss=0.1079, over 5696120.12 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3631, pruned_loss=0.1163, over 5664631.23 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3572, pruned_loss=0.1064, over 5696693.73 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:47:24,930 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-09 14:47:40,587 INFO [train.py:968] (0/2) Epoch 18, batch 38250, giga_loss[loss=0.2978, simple_loss=0.3851, pruned_loss=0.1053, over 28973.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3596, pruned_loss=0.1083, over 5691994.41 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3634, pruned_loss=0.1163, over 5667281.22 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.358, pruned_loss=0.1069, over 5690939.55 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:47:54,443 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3529, 3.2105, 1.5009, 1.4564], device='cuda:0'), covar=tensor([0.1060, 0.0285, 0.0939, 0.1442], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0537, 0.0369, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 14:47:55,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9881, 1.1851, 1.3303, 1.0156], device='cuda:0'), covar=tensor([0.1758, 0.1496, 0.2238, 0.1714], device='cuda:0'), in_proj_covar=tensor([0.0462, 0.0739, 0.0700, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 14:48:08,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.817e+02 1.113e+03 1.355e+03 1.890e+03 4.743e+03, threshold=2.710e+03, percent-clipped=2.0
+2023-03-09 14:48:19,673 INFO [train.py:968] (0/2) Epoch 18, batch 38300, libri_loss[loss=0.2881, simple_loss=0.3656, pruned_loss=0.1053, over 29527.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3594, pruned_loss=0.1071, over 5700655.70 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3635, pruned_loss=0.1162, over 5668160.96 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.358, pruned_loss=0.1058, over 5699558.03 frames. ], batch size: 81, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:48:31,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:49:00,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815060.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:49:01,200 INFO [train.py:968] (0/2) Epoch 18, batch 38350, giga_loss[loss=0.281, simple_loss=0.3498, pruned_loss=0.1061, over 28706.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3589, pruned_loss=0.1059, over 5698832.08 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3635, pruned_loss=0.1163, over 5664485.26 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3577, pruned_loss=0.1046, over 5701321.85 frames. ], batch size: 85, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:49:02,060 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815063.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:49:04,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815066.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:49:10,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:49:29,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.140e+02 1.208e+03 1.438e+03 2.117e+03 7.224e+03, threshold=2.876e+03, percent-clipped=11.0
+2023-03-09 14:49:42,398 INFO [train.py:968] (0/2) Epoch 18, batch 38400, giga_loss[loss=0.2696, simple_loss=0.3528, pruned_loss=0.09318, over 28400.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3585, pruned_loss=0.1057, over 5698415.59 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.364, pruned_loss=0.1166, over 5667214.90 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.357, pruned_loss=0.1043, over 5698492.25 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:49:58,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:50:04,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:50:23,052 INFO [train.py:968] (0/2) Epoch 18, batch 38450, giga_loss[loss=0.2838, simple_loss=0.3557, pruned_loss=0.1059, over 28360.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3567, pruned_loss=0.1052, over 5686232.63 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3642, pruned_loss=0.1167, over 5654451.02 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3552, pruned_loss=0.1036, over 5698242.91 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:50:28,469 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:50:30,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:50:51,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.587e+02 1.114e+03 1.377e+03 1.942e+03 5.351e+03, threshold=2.753e+03, percent-clipped=11.0
+2023-03-09 14:50:55,341 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:50:57,699 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815203.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:51:00,135 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815206.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:51:02,118 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815209.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:51:04,124 INFO [train.py:968] (0/2) Epoch 18, batch 38500, giga_loss[loss=0.2751, simple_loss=0.3477, pruned_loss=0.1013, over 27878.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3542, pruned_loss=0.1038, over 5686040.95 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3642, pruned_loss=0.1169, over 5652336.46 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3526, pruned_loss=0.1019, over 5699948.80 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:51:04,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:51:20,988 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815235.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:51:25,823 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815241.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:51:41,606 INFO [train.py:968] (0/2) Epoch 18, batch 38550, giga_loss[loss=0.2725, simple_loss=0.3542, pruned_loss=0.09541, over 29009.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3536, pruned_loss=0.1038, over 5693012.83 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3646, pruned_loss=0.1173, over 5654180.32 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3517, pruned_loss=0.1016, over 5703711.32 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:52:01,318 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:52:05,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:52:15,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.524e+02 1.091e+03 1.284e+03 1.838e+03 7.864e+03, threshold=2.567e+03, percent-clipped=7.0
+2023-03-09 14:52:25,369 INFO [train.py:968] (0/2) Epoch 18, batch 38600, giga_loss[loss=0.2608, simple_loss=0.3336, pruned_loss=0.09396, over 28514.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3533, pruned_loss=0.1037, over 5698461.38 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3644, pruned_loss=0.1172, over 5657937.05 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3518, pruned_loss=0.1018, over 5704274.12 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:52:28,407 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:53:04,211 INFO [train.py:968] (0/2) Epoch 18, batch 38650, giga_loss[loss=0.2534, simple_loss=0.3284, pruned_loss=0.08914, over 28435.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3523, pruned_loss=0.1027, over 5703051.75 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3639, pruned_loss=0.1168, over 5663043.64 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3514, pruned_loss=0.1013, over 5703995.02 frames. ], batch size: 60, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:53:08,850 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:53:25,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:53:32,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.808e+02 1.026e+03 1.399e+03 1.929e+03 3.891e+03, threshold=2.798e+03, percent-clipped=10.0
+2023-03-09 14:53:41,889 INFO [train.py:968] (0/2) Epoch 18, batch 38700, giga_loss[loss=0.2783, simple_loss=0.3554, pruned_loss=0.1006, over 28569.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3522, pruned_loss=0.1022, over 5711629.25 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3642, pruned_loss=0.117, over 5671824.39 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3507, pruned_loss=0.1003, over 5705933.29 frames. ], batch size: 85, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:53:42,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3747, 1.1588, 1.1418, 1.6016], device='cuda:0'), covar=tensor([0.0815, 0.0370, 0.0356, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0115, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 14:53:43,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6282, 1.9081, 1.8979, 1.5614], device='cuda:0'), covar=tensor([0.1540, 0.1375, 0.1766, 0.1638], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0738, 0.0698, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 14:54:01,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815438.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:54:08,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:54:19,749 INFO [train.py:968] (0/2) Epoch 18, batch 38750, giga_loss[loss=0.2796, simple_loss=0.361, pruned_loss=0.09912, over 28299.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3516, pruned_loss=0.1013, over 5708684.78 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3642, pruned_loss=0.1171, over 5667419.31 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3501, pruned_loss=0.09932, over 5709636.85 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:54:53,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.448e+02 1.062e+03 1.310e+03 1.888e+03 6.974e+03, threshold=2.620e+03, percent-clipped=12.0
+2023-03-09 14:54:54,000 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815501.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:54:58,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:55:01,834 INFO [train.py:968] (0/2) Epoch 18, batch 38800, giga_loss[loss=0.2766, simple_loss=0.3532, pruned_loss=0.09996, over 28892.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3521, pruned_loss=0.1024, over 5709107.32 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3647, pruned_loss=0.1175, over 5673529.35 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3501, pruned_loss=0.1, over 5704890.74 frames. ], batch size: 227, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:55:04,463 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-09 14:55:18,085 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:55:21,851 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-09 14:55:43,379 INFO [train.py:968] (0/2) Epoch 18, batch 38850, giga_loss[loss=0.2569, simple_loss=0.3367, pruned_loss=0.08848, over 28791.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.349, pruned_loss=0.101, over 5695912.35 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3649, pruned_loss=0.1176, over 5666583.23 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.347, pruned_loss=0.09887, over 5698761.31 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:55:58,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 14:55:58,662 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815581.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:56:01,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815584.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:56:06,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:56:08,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:56:11,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.778e+02 1.117e+03 1.495e+03 1.910e+03 6.056e+03, threshold=2.991e+03, percent-clipped=11.0
+2023-03-09 14:56:21,163 INFO [train.py:968] (0/2) Epoch 18, batch 38900, giga_loss[loss=0.2624, simple_loss=0.3354, pruned_loss=0.09472, over 28757.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3468, pruned_loss=0.1001, over 5698023.83 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3648, pruned_loss=0.1176, over 5668846.91 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3446, pruned_loss=0.09767, over 5699821.85 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:56:22,047 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815613.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:56:30,372 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:56:49,388 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:56:51,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:56:58,338 INFO [train.py:968] (0/2) Epoch 18, batch 38950, giga_loss[loss=0.282, simple_loss=0.3524, pruned_loss=0.1058, over 28880.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3457, pruned_loss=0.09971, over 5695022.57 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3649, pruned_loss=0.1176, over 5661678.28 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3432, pruned_loss=0.09704, over 5703706.85 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:57:05,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3472, 2.4048, 1.7146, 1.9597], device='cuda:0'), covar=tensor([0.0923, 0.0747, 0.1080, 0.1171], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0443, 0.0514, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 14:57:14,283 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:57:29,351 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.464e+02 1.178e+03 1.462e+03 2.036e+03 7.119e+03, threshold=2.924e+03, percent-clipped=14.0
+2023-03-09 14:57:38,261 INFO [train.py:968] (0/2) Epoch 18, batch 39000, giga_loss[loss=0.2553, simple_loss=0.3278, pruned_loss=0.09141, over 28927.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3452, pruned_loss=0.1, over 5697649.88 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3651, pruned_loss=0.1179, over 5661678.13 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3424, pruned_loss=0.09705, over 5705641.82 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:57:38,266 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 14:57:46,799 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2068, simple_loss=0.3137, pruned_loss=0.04999, over 944034.00 frames. 
+2023-03-09 14:57:46,799 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 14:58:11,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:58:25,234 INFO [train.py:968] (0/2) Epoch 18, batch 39050, giga_loss[loss=0.2485, simple_loss=0.3292, pruned_loss=0.08389, over 28891.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3427, pruned_loss=0.09874, over 5698648.57 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3653, pruned_loss=0.1181, over 5657409.54 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3398, pruned_loss=0.0958, over 5710367.37 frames. ], batch size: 112, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:58:28,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:58:55,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.478e+02 1.074e+03 1.460e+03 1.823e+03 6.526e+03, threshold=2.921e+03, percent-clipped=7.0
+2023-03-09 14:59:03,685 INFO [train.py:968] (0/2) Epoch 18, batch 39100, libri_loss[loss=0.289, simple_loss=0.3636, pruned_loss=0.1072, over 29531.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3409, pruned_loss=0.09807, over 5698351.77 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3653, pruned_loss=0.1179, over 5661352.42 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3379, pruned_loss=0.09534, over 5705057.96 frames. ], batch size: 83, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:59:42,982 INFO [train.py:968] (0/2) Epoch 18, batch 39150, libri_loss[loss=0.3528, simple_loss=0.3986, pruned_loss=0.1535, over 19410.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3389, pruned_loss=0.09766, over 5690735.42 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3649, pruned_loss=0.1178, over 5654212.10 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3359, pruned_loss=0.09488, over 5704861.67 frames. ], batch size: 188, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:59:54,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815876.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:00:04,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:05,801 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:15,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.114e+02 1.083e+03 1.446e+03 1.880e+03 4.518e+03, threshold=2.893e+03, percent-clipped=6.0
+2023-03-09 15:00:19,905 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:19,935 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:22,177 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:24,268 INFO [train.py:968] (0/2) Epoch 18, batch 39200, giga_loss[loss=0.2511, simple_loss=0.3297, pruned_loss=0.08626, over 28946.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3374, pruned_loss=0.09707, over 5697426.24 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3652, pruned_loss=0.1179, over 5658102.62 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3344, pruned_loss=0.09444, over 5705585.71 frames. ], batch size: 106, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:00:24,959 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:31,631 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:32,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9013, 3.7155, 3.5095, 1.8783], device='cuda:0'), covar=tensor([0.0585, 0.0767, 0.0726, 0.2292], device='cuda:0'), in_proj_covar=tensor([0.1157, 0.1078, 0.0919, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 15:00:47,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3177, 1.6912, 1.3417, 1.3884], device='cuda:0'), covar=tensor([0.2868, 0.2767, 0.3318, 0.2382], device='cuda:0'), in_proj_covar=tensor([0.1442, 0.1047, 0.1279, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 15:00:50,683 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:01:10,129 INFO [train.py:968] (0/2) Epoch 18, batch 39250, giga_loss[loss=0.2455, simple_loss=0.3159, pruned_loss=0.08752, over 28482.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3379, pruned_loss=0.0974, over 5704360.14 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3651, pruned_loss=0.1179, over 5660596.24 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3353, pruned_loss=0.09513, over 5708864.47 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:01:43,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.756e+02 1.074e+03 1.244e+03 1.831e+03 3.890e+03, threshold=2.487e+03, percent-clipped=4.0
+2023-03-09 15:01:43,138 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-816000.pt
+2023-03-09 15:01:52,681 INFO [train.py:968] (0/2) Epoch 18, batch 39300, giga_loss[loss=0.2661, simple_loss=0.3354, pruned_loss=0.09836, over 28903.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.34, pruned_loss=0.09771, over 5700344.08 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3652, pruned_loss=0.118, over 5659060.85 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3372, pruned_loss=0.09516, over 5706820.70 frames. ], batch size: 106, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:01:57,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5966, 1.7561, 1.3312, 1.3139], device='cuda:0'), covar=tensor([0.0896, 0.0539, 0.1035, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0442, 0.0511, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:01:57,874 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816019.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:02:00,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=816022.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:02:27,384 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:02:28,212 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 15:02:29,591 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=816051.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:02:30,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=816052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:02:36,432 INFO [train.py:968] (0/2) Epoch 18, batch 39350, giga_loss[loss=0.2511, simple_loss=0.3385, pruned_loss=0.08185, over 28892.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3442, pruned_loss=0.09941, over 5699847.92 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3652, pruned_loss=0.1181, over 5658915.15 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3412, pruned_loss=0.09677, over 5706104.13 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:02:54,474 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=816081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:03:11,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.246e+02 1.091e+03 1.359e+03 1.742e+03 5.176e+03, threshold=2.717e+03, percent-clipped=13.0
+2023-03-09 15:03:15,076 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-09 15:03:18,640 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=816105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:03:23,832 INFO [train.py:968] (0/2) Epoch 18, batch 39400, giga_loss[loss=0.2617, simple_loss=0.3461, pruned_loss=0.08862, over 28639.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.346, pruned_loss=0.09929, over 5696554.77 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3653, pruned_loss=0.1182, over 5660843.16 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3434, pruned_loss=0.09701, over 5700167.69 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:03:30,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5030, 3.7154, 1.6331, 1.5647], device='cuda:0'), covar=tensor([0.0921, 0.0270, 0.0905, 0.1347], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0534, 0.0368, 0.0412], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:0')
+2023-03-09 15:04:04,868 INFO [train.py:968] (0/2) Epoch 18, batch 39450, giga_loss[loss=0.2637, simple_loss=0.343, pruned_loss=0.09223, over 28709.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3449, pruned_loss=0.09814, over 5685483.44 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3652, pruned_loss=0.118, over 5660449.20 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3422, pruned_loss=0.09573, over 5690125.34 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:04:36,805 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.982e+02 1.117e+03 1.395e+03 1.817e+03 4.181e+03, threshold=2.791e+03, percent-clipped=8.0
+2023-03-09 15:04:44,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7077, 1.8657, 1.5377, 2.0871], device='cuda:0'), covar=tensor([0.2549, 0.2715, 0.3092, 0.2465], device='cuda:0'), in_proj_covar=tensor([0.1444, 0.1048, 0.1281, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 15:04:45,502 INFO [train.py:968] (0/2) Epoch 18, batch 39500, giga_loss[loss=0.2921, simple_loss=0.3525, pruned_loss=0.1158, over 28849.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.344, pruned_loss=0.09692, over 5689493.28 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3649, pruned_loss=0.1179, over 5655176.15 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3417, pruned_loss=0.09475, over 5698019.90 frames. ], batch size: 112, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:05:28,820 INFO [train.py:968] (0/2) Epoch 18, batch 39550, giga_loss[loss=0.3828, simple_loss=0.4242, pruned_loss=0.1707, over 26821.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3459, pruned_loss=0.09859, over 5687874.02 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3655, pruned_loss=0.1182, over 5662731.41 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3429, pruned_loss=0.09592, over 5689124.91 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:05:42,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=816281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:06:01,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.389e+02 1.291e+03 1.484e+03 1.828e+03 4.564e+03, threshold=2.969e+03, percent-clipped=9.0
+2023-03-09 15:06:09,261 INFO [train.py:968] (0/2) Epoch 18, batch 39600, giga_loss[loss=0.3212, simple_loss=0.3803, pruned_loss=0.131, over 26647.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.346, pruned_loss=0.09889, over 5691125.20 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3654, pruned_loss=0.1181, over 5669775.40 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3431, pruned_loss=0.09631, over 5686286.63 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:06:22,946 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2681, 4.0792, 3.8746, 1.9020], device='cuda:0'), covar=tensor([0.0647, 0.0833, 0.0787, 0.1976], device='cuda:0'), in_proj_covar=tensor([0.1158, 0.1079, 0.0923, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 15:06:23,707 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 15:06:48,824 INFO [train.py:968] (0/2) Epoch 18, batch 39650, giga_loss[loss=0.2583, simple_loss=0.3349, pruned_loss=0.09083, over 28897.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3484, pruned_loss=0.1002, over 5703408.14 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3649, pruned_loss=0.1175, over 5674380.71 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.346, pruned_loss=0.09798, over 5695896.80 frames. ], batch size: 106, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:07:20,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.666e+02 1.277e+03 1.737e+03 2.593e+03 7.393e+03, threshold=3.473e+03, percent-clipped=21.0
+2023-03-09 15:07:26,042 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-09 15:07:28,592 INFO [train.py:968] (0/2) Epoch 18, batch 39700, giga_loss[loss=0.2421, simple_loss=0.326, pruned_loss=0.07911, over 28755.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3514, pruned_loss=0.1019, over 5710052.80 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.365, pruned_loss=0.1176, over 5678704.41 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.349, pruned_loss=0.09969, over 5701016.28 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:07:36,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5003, 1.7945, 1.4351, 1.4662], device='cuda:0'), covar=tensor([0.2355, 0.2423, 0.2768, 0.2238], device='cuda:0'), in_proj_covar=tensor([0.1436, 0.1044, 0.1276, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 15:07:39,351 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:07:41,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=816427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:08:03,414 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=816456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:08:10,406 INFO [train.py:968] (0/2) Epoch 18, batch 39750, giga_loss[loss=0.2446, simple_loss=0.3281, pruned_loss=0.08058, over 28760.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3536, pruned_loss=0.1025, over 5719965.52 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3656, pruned_loss=0.1179, over 5680959.78 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.351, pruned_loss=0.1004, over 5711237.07 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:08:24,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-09 15:08:25,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=816480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:08:39,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2197, 2.5801, 1.2655, 1.3550], device='cuda:0'), covar=tensor([0.0953, 0.0402, 0.0914, 0.1352], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0538, 0.0370, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 15:08:44,750 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.612e+02 1.210e+03 1.712e+03 2.121e+03 5.690e+03, threshold=3.424e+03, percent-clipped=4.0
+2023-03-09 15:08:54,407 INFO [train.py:968] (0/2) Epoch 18, batch 39800, giga_loss[loss=0.2522, simple_loss=0.3272, pruned_loss=0.08864, over 28458.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3549, pruned_loss=0.1035, over 5718178.23 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3659, pruned_loss=0.118, over 5682429.21 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3526, pruned_loss=0.1016, over 5710444.59 frames. ], batch size: 71, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:09:21,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5948, 0.8837, 2.8144, 2.6314], device='cuda:0'), covar=tensor([0.1801, 0.2678, 0.0581, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0729, 0.0629, 0.0926, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:09:25,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6402, 1.6919, 1.8413, 1.4154], device='cuda:0'), covar=tensor([0.1915, 0.2478, 0.1548, 0.1682], device='cuda:0'), in_proj_covar=tensor([0.0882, 0.0695, 0.0925, 0.0826], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 15:09:33,455 INFO [train.py:968] (0/2) Epoch 18, batch 39850, libri_loss[loss=0.3033, simple_loss=0.3758, pruned_loss=0.1154, over 29649.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3553, pruned_loss=0.1039, over 5713357.84 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3663, pruned_loss=0.1182, over 5679813.01 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3527, pruned_loss=0.1018, over 5709854.59 frames. ], batch size: 88, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:09:36,528 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-09 15:10:06,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.916e+02 1.170e+03 1.376e+03 1.838e+03 9.628e+03, threshold=2.752e+03, percent-clipped=5.0
+2023-03-09 15:10:16,120 INFO [train.py:968] (0/2) Epoch 18, batch 39900, giga_loss[loss=0.2884, simple_loss=0.3568, pruned_loss=0.11, over 28932.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3539, pruned_loss=0.1031, over 5706711.08 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3665, pruned_loss=0.1183, over 5674017.15 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3515, pruned_loss=0.101, over 5708878.79 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:10:24,432 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:10:27,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=816626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:10:34,422 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6958, 2.3352, 1.5036, 0.8571], device='cuda:0'), covar=tensor([0.7244, 0.3393, 0.3684, 0.6602], device='cuda:0'), in_proj_covar=tensor([0.1695, 0.1598, 0.1570, 0.1380], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 15:10:41,994 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-09 15:10:50,676 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=816655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:10:55,387 INFO [train.py:968] (0/2) Epoch 18, batch 39950, giga_loss[loss=0.2392, simple_loss=0.3205, pruned_loss=0.07891, over 29118.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3515, pruned_loss=0.102, over 5710851.54 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3667, pruned_loss=0.1184, over 5677593.06 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3492, pruned_loss=0.09999, over 5709634.36 frames. ], batch size: 155, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:11:25,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.048e+02 1.197e+03 1.473e+03 2.073e+03 6.224e+03, threshold=2.946e+03, percent-clipped=15.0
+2023-03-09 15:11:35,644 INFO [train.py:968] (0/2) Epoch 18, batch 40000, giga_loss[loss=0.2576, simple_loss=0.3363, pruned_loss=0.08941, over 28709.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.347, pruned_loss=0.09913, over 5718826.45 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3664, pruned_loss=0.1182, over 5682670.96 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3451, pruned_loss=0.09738, over 5714200.03 frames. ], batch size: 284, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:12:13,819 INFO [train.py:968] (0/2) Epoch 18, batch 40050, giga_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 28853.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3454, pruned_loss=0.09863, over 5697499.20 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3672, pruned_loss=0.1189, over 5664527.52 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3426, pruned_loss=0.09608, over 5709483.81 frames. ], batch size: 112, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:12:43,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.945e+02 1.226e+03 1.527e+03 2.131e+03 7.738e+03, threshold=3.053e+03, percent-clipped=11.0
+2023-03-09 15:12:51,625 INFO [train.py:968] (0/2) Epoch 18, batch 40100, libri_loss[loss=0.2981, simple_loss=0.3685, pruned_loss=0.1139, over 29589.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3467, pruned_loss=0.09795, over 5707117.11 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3673, pruned_loss=0.1189, over 5673312.51 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3437, pruned_loss=0.09521, over 5710017.97 frames. ], batch size: 77, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:13:33,409 INFO [train.py:968] (0/2) Epoch 18, batch 40150, giga_loss[loss=0.3017, simple_loss=0.3707, pruned_loss=0.1164, over 27623.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3487, pruned_loss=0.09899, over 5694642.22 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3682, pruned_loss=0.1196, over 5668300.40 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.345, pruned_loss=0.09572, over 5702167.25 frames. ], batch size: 472, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:14:07,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.529e+02 1.276e+03 1.548e+03 2.149e+03 5.739e+03, threshold=3.097e+03, percent-clipped=5.0
+2023-03-09 15:14:13,961 INFO [train.py:968] (0/2) Epoch 18, batch 40200, giga_loss[loss=0.2224, simple_loss=0.2923, pruned_loss=0.07625, over 28528.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.347, pruned_loss=0.09834, over 5706253.63 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3677, pruned_loss=0.1192, over 5673044.58 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3441, pruned_loss=0.09566, over 5708529.20 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:14:54,239 INFO [train.py:968] (0/2) Epoch 18, batch 40250, giga_loss[loss=0.3615, simple_loss=0.4065, pruned_loss=0.1582, over 26796.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3454, pruned_loss=0.09861, over 5711936.11 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.368, pruned_loss=0.1195, over 5674012.97 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3425, pruned_loss=0.09601, over 5713265.10 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:15:28,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.417e+02 1.176e+03 1.571e+03 2.209e+03 6.596e+03, threshold=3.141e+03, percent-clipped=7.0
+2023-03-09 15:15:36,061 INFO [train.py:968] (0/2) Epoch 18, batch 40300, giga_loss[loss=0.2966, simple_loss=0.3568, pruned_loss=0.1182, over 28944.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3446, pruned_loss=0.09993, over 5694449.12 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3682, pruned_loss=0.1195, over 5659778.36 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3416, pruned_loss=0.09731, over 5709556.75 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 15:16:06,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2356, 1.2293, 3.3897, 3.0804], device='cuda:0'), covar=tensor([0.1507, 0.2762, 0.0440, 0.2033], device='cuda:0'), in_proj_covar=tensor([0.0726, 0.0628, 0.0926, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:16:19,061 INFO [train.py:968] (0/2) Epoch 18, batch 40350, giga_loss[loss=0.275, simple_loss=0.3345, pruned_loss=0.1077, over 28779.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3427, pruned_loss=0.1001, over 5694056.27 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3683, pruned_loss=0.1197, over 5662280.19 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3401, pruned_loss=0.09771, over 5703804.78 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 15:16:48,170 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4231, 1.7767, 1.3597, 1.5926], device='cuda:0'), covar=tensor([0.0750, 0.0300, 0.0354, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 15:16:54,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.900e+02 1.318e+03 1.664e+03 2.171e+03 5.452e+03, threshold=3.329e+03, percent-clipped=6.0
+2023-03-09 15:16:59,986 INFO [train.py:968] (0/2) Epoch 18, batch 40400, giga_loss[loss=0.2342, simple_loss=0.3047, pruned_loss=0.08186, over 28427.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3421, pruned_loss=0.09986, over 5700665.38 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3683, pruned_loss=0.1197, over 5666974.66 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3393, pruned_loss=0.09737, over 5705146.53 frames. ], batch size: 85, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:17:04,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0302, 1.0703, 3.4758, 3.1297], device='cuda:0'), covar=tensor([0.1595, 0.2584, 0.0497, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0727, 0.0629, 0.0929, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:17:11,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=817125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:17:39,312 INFO [train.py:968] (0/2) Epoch 18, batch 40450, giga_loss[loss=0.2592, simple_loss=0.3296, pruned_loss=0.09434, over 28911.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3384, pruned_loss=0.09798, over 5705036.97 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1196, over 5669700.34 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3359, pruned_loss=0.09582, over 5706657.62 frames. ], batch size: 112, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:18:14,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.920e+02 1.226e+03 1.626e+03 2.263e+03 1.002e+04, threshold=3.253e+03, percent-clipped=12.0
+2023-03-09 15:18:19,760 INFO [train.py:968] (0/2) Epoch 18, batch 40500, giga_loss[loss=0.226, simple_loss=0.307, pruned_loss=0.07247, over 28726.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3344, pruned_loss=0.09612, over 5704723.54 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3684, pruned_loss=0.1198, over 5670434.60 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3316, pruned_loss=0.09383, over 5705995.44 frames. ], batch size: 284, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:19:01,088 INFO [train.py:968] (0/2) Epoch 18, batch 40550, libri_loss[loss=0.3213, simple_loss=0.3818, pruned_loss=0.1304, over 19102.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3327, pruned_loss=0.09472, over 5693500.03 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3684, pruned_loss=0.1198, over 5655166.09 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3297, pruned_loss=0.09238, over 5710357.09 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:19:09,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4762, 2.2415, 1.6932, 1.5417], device='cuda:0'), covar=tensor([0.0701, 0.0243, 0.0292, 0.0821], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 15:19:34,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.718e+02 1.221e+03 1.585e+03 2.060e+03 6.265e+03, threshold=3.170e+03, percent-clipped=12.0
+2023-03-09 15:19:40,306 INFO [train.py:968] (0/2) Epoch 18, batch 40600, giga_loss[loss=0.3557, simple_loss=0.3976, pruned_loss=0.1569, over 26599.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3347, pruned_loss=0.09563, over 5693724.44 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3683, pruned_loss=0.1198, over 5654540.83 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3313, pruned_loss=0.0929, over 5709870.01 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:19:46,048 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0057, 2.1405, 1.6326, 1.7862], device='cuda:0'), covar=tensor([0.0951, 0.0754, 0.1079, 0.1213], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0445, 0.0511, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:20:19,589 INFO [train.py:968] (0/2) Epoch 18, batch 40650, giga_loss[loss=0.2414, simple_loss=0.3125, pruned_loss=0.08517, over 28797.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3381, pruned_loss=0.09678, over 5701379.82 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1197, over 5661342.59 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3347, pruned_loss=0.0941, over 5709301.43 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:20:54,757 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.257e+02 1.243e+03 1.614e+03 2.152e+03 6.286e+03, threshold=3.229e+03, percent-clipped=10.0
+2023-03-09 15:21:01,597 INFO [train.py:968] (0/2) Epoch 18, batch 40700, giga_loss[loss=0.2579, simple_loss=0.3454, pruned_loss=0.08521, over 28896.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3418, pruned_loss=0.09862, over 5694574.61 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3683, pruned_loss=0.1199, over 5655845.06 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3381, pruned_loss=0.0956, over 5707045.36 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:21:32,687 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=817448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:21:43,172 INFO [train.py:968] (0/2) Epoch 18, batch 40750, giga_loss[loss=0.3692, simple_loss=0.4174, pruned_loss=0.1605, over 26836.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3455, pruned_loss=0.1006, over 5693852.52 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3679, pruned_loss=0.1198, over 5661924.28 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3422, pruned_loss=0.09768, over 5699283.30 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:22:04,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9561, 2.1210, 1.8173, 2.2095], device='cuda:0'), covar=tensor([0.2287, 0.2541, 0.2777, 0.2372], device='cuda:0'), in_proj_covar=tensor([0.1443, 0.1049, 0.1278, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 15:22:14,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=817500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:22:17,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.202e+02 1.154e+03 1.504e+03 1.775e+03 5.004e+03, threshold=3.008e+03, percent-clipped=2.0
+2023-03-09 15:22:24,829 INFO [train.py:968] (0/2) Epoch 18, batch 40800, giga_loss[loss=0.263, simple_loss=0.3386, pruned_loss=0.09371, over 28692.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3483, pruned_loss=0.102, over 5694150.63 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3677, pruned_loss=0.1196, over 5658153.40 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3453, pruned_loss=0.09936, over 5702922.86 frames. ], batch size: 92, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:23:04,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3856, 1.6086, 1.5145, 1.4651], device='cuda:0'), covar=tensor([0.1648, 0.1759, 0.2239, 0.1856], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0740, 0.0704, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 15:23:13,584 INFO [train.py:968] (0/2) Epoch 18, batch 40850, giga_loss[loss=0.3062, simple_loss=0.3707, pruned_loss=0.1208, over 28757.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.354, pruned_loss=0.1069, over 5694205.68 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3676, pruned_loss=0.1196, over 5660877.03 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3516, pruned_loss=0.1047, over 5698886.62 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:24:01,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.959e+02 1.465e+03 1.857e+03 2.383e+03 5.191e+03, threshold=3.715e+03, percent-clipped=8.0
+2023-03-09 15:24:07,573 INFO [train.py:968] (0/2) Epoch 18, batch 40900, giga_loss[loss=0.337, simple_loss=0.3994, pruned_loss=0.1373, over 28578.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3581, pruned_loss=0.1102, over 5697806.68 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3673, pruned_loss=0.1194, over 5665681.06 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3562, pruned_loss=0.1085, over 5697734.13 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:24:27,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 15:24:37,482 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=817643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:24:40,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=817646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:24:55,664 INFO [train.py:968] (0/2) Epoch 18, batch 40950, giga_loss[loss=0.3076, simple_loss=0.3824, pruned_loss=0.1164, over 28920.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3657, pruned_loss=0.1154, over 5693238.26 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3674, pruned_loss=0.1194, over 5667779.46 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3641, pruned_loss=0.1139, over 5691698.31 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:25:00,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6242, 1.4039, 4.6046, 3.6323], device='cuda:0'), covar=tensor([0.1566, 0.2662, 0.0376, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0727, 0.0629, 0.0930, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:25:09,261 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=817675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:25:10,861 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-09 15:25:37,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.643e+03 2.135e+03 2.727e+03 5.573e+03, threshold=4.271e+03, percent-clipped=9.0
+2023-03-09 15:25:41,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7321, 1.9417, 1.8815, 1.6033], device='cuda:0'), covar=tensor([0.1933, 0.1516, 0.1339, 0.1601], device='cuda:0'), in_proj_covar=tensor([0.1914, 0.1834, 0.1773, 0.1899], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 15:25:42,913 INFO [train.py:968] (0/2) Epoch 18, batch 41000, giga_loss[loss=0.311, simple_loss=0.3794, pruned_loss=0.1213, over 28857.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3703, pruned_loss=0.1189, over 5693445.12 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3674, pruned_loss=0.1193, over 5668557.36 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.369, pruned_loss=0.1177, over 5692327.57 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:26:10,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3961, 3.1064, 1.5687, 1.4240], device='cuda:0'), covar=tensor([0.0910, 0.0364, 0.0836, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0541, 0.0371, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 15:26:29,857 INFO [train.py:968] (0/2) Epoch 18, batch 41050, giga_loss[loss=0.3562, simple_loss=0.4152, pruned_loss=0.1487, over 28605.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3762, pruned_loss=0.124, over 5694653.16 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3672, pruned_loss=0.1193, over 5670852.01 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3755, pruned_loss=0.1232, over 5691928.65 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:26:39,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3853, 1.6322, 1.3911, 1.4373], device='cuda:0'), covar=tensor([0.0774, 0.0315, 0.0329, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 15:27:13,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.895e+03 2.410e+03 3.517e+03 7.159e+03, threshold=4.819e+03, percent-clipped=11.0
+2023-03-09 15:27:21,080 INFO [train.py:968] (0/2) Epoch 18, batch 41100, giga_loss[loss=0.3026, simple_loss=0.3674, pruned_loss=0.1189, over 28883.00 frames. ], tot_loss[loss=0.319, simple_loss=0.381, pruned_loss=0.1285, over 5672471.59 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3674, pruned_loss=0.1193, over 5673851.47 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3805, pruned_loss=0.1279, over 5668000.72 frames. ], batch size: 199, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 15:27:21,277 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=817812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:27:32,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=817823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:28:13,902 INFO [train.py:968] (0/2) Epoch 18, batch 41150, giga_loss[loss=0.3906, simple_loss=0.4138, pruned_loss=0.1837, over 23452.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3827, pruned_loss=0.1311, over 5664828.98 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.367, pruned_loss=0.1191, over 5678148.21 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.383, pruned_loss=0.131, over 5657367.32 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 15:28:27,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5843, 1.7957, 1.7918, 1.5938], device='cuda:0'), covar=tensor([0.1679, 0.1741, 0.2021, 0.1765], device='cuda:0'), in_proj_covar=tensor([0.0463, 0.0745, 0.0707, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 15:28:57,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6208, 1.9737, 1.5234, 1.7844], device='cuda:0'), covar=tensor([0.2341, 0.2379, 0.2742, 0.2186], device='cuda:0'), in_proj_covar=tensor([0.1441, 0.1050, 0.1283, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 15:29:03,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.178e+03 1.833e+03 2.522e+03 3.956e+03 9.967e+03, threshold=5.045e+03, percent-clipped=17.0
+2023-03-09 15:29:08,896 INFO [train.py:968] (0/2) Epoch 18, batch 41200, giga_loss[loss=0.3224, simple_loss=0.3776, pruned_loss=0.1336, over 28577.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3858, pruned_loss=0.1348, over 5664891.39 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3669, pruned_loss=0.119, over 5683341.37 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3864, pruned_loss=0.1351, over 5654062.85 frames. ], batch size: 65, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:29:24,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4306, 3.2042, 1.5179, 1.5436], device='cuda:0'), covar=tensor([0.0936, 0.0316, 0.0848, 0.1267], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0543, 0.0371, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 15:29:29,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6885, 4.5319, 4.3484, 2.0117], device='cuda:0'), covar=tensor([0.0584, 0.0703, 0.0782, 0.2040], device='cuda:0'), in_proj_covar=tensor([0.1165, 0.1086, 0.0926, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 15:29:59,130 INFO [train.py:968] (0/2) Epoch 18, batch 41250, libri_loss[loss=0.3749, simple_loss=0.4117, pruned_loss=0.1691, over 29743.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3899, pruned_loss=0.1392, over 5641002.43 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3671, pruned_loss=0.1192, over 5680751.60 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3912, pruned_loss=0.1401, over 5633363.76 frames. ], batch size: 87, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:30:02,963 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=817966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:30:05,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=817969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:30:05,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4220, 1.6580, 1.7959, 1.4741], device='cuda:0'), covar=tensor([0.1768, 0.1695, 0.1956, 0.1777], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0748, 0.0709, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 15:30:34,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=817998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:30:38,199 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-818000.pt
+2023-03-09 15:30:44,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.217e+03 1.864e+03 2.446e+03 3.546e+03 7.768e+03, threshold=4.892e+03, percent-clipped=4.0
+2023-03-09 15:30:49,816 INFO [train.py:968] (0/2) Epoch 18, batch 41300, giga_loss[loss=0.3566, simple_loss=0.4118, pruned_loss=0.1507, over 28994.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3917, pruned_loss=0.1406, over 5646191.47 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3664, pruned_loss=0.1187, over 5687593.36 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3941, pruned_loss=0.1424, over 5632817.28 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:31:44,929 INFO [train.py:968] (0/2) Epoch 18, batch 41350, giga_loss[loss=0.291, simple_loss=0.3578, pruned_loss=0.1121, over 28953.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3927, pruned_loss=0.1422, over 5637637.99 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3664, pruned_loss=0.1185, over 5692057.12 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3952, pruned_loss=0.1443, over 5621810.10 frames. ], batch size: 106, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:32:22,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6576, 1.9025, 1.6520, 1.6772], device='cuda:0'), covar=tensor([0.1658, 0.1852, 0.1990, 0.1886], device='cuda:0'), in_proj_covar=tensor([0.0461, 0.0743, 0.0704, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 15:32:23,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.756e+03 2.186e+03 2.958e+03 6.388e+03, threshold=4.372e+03, percent-clipped=5.0
+2023-03-09 15:32:30,197 INFO [train.py:968] (0/2) Epoch 18, batch 41400, giga_loss[loss=0.2897, simple_loss=0.3583, pruned_loss=0.1106, over 28812.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3907, pruned_loss=0.1414, over 5643499.17 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3662, pruned_loss=0.1185, over 5694449.77 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3936, pruned_loss=0.1439, over 5626991.96 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:33:21,959 INFO [train.py:968] (0/2) Epoch 18, batch 41450, giga_loss[loss=0.3397, simple_loss=0.4016, pruned_loss=0.1389, over 28586.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3893, pruned_loss=0.1397, over 5659878.49 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.366, pruned_loss=0.1185, over 5698537.93 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3922, pruned_loss=0.142, over 5642295.31 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:33:48,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=818187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:34:06,280 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.800e+03 2.535e+03 3.210e+03 8.048e+03, threshold=5.070e+03, percent-clipped=11.0
+2023-03-09 15:34:14,254 INFO [train.py:968] (0/2) Epoch 18, batch 41500, giga_loss[loss=0.3055, simple_loss=0.3748, pruned_loss=0.1181, over 29045.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3878, pruned_loss=0.137, over 5655780.43 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3655, pruned_loss=0.1182, over 5692575.77 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3911, pruned_loss=0.1396, over 5645620.38 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:35:05,164 INFO [train.py:968] (0/2) Epoch 18, batch 41550, giga_loss[loss=0.2967, simple_loss=0.374, pruned_loss=0.1097, over 28726.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.389, pruned_loss=0.1371, over 5670800.35 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3651, pruned_loss=0.1182, over 5698908.25 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3927, pruned_loss=0.1398, over 5655953.05 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:35:55,025 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.622e+03 2.056e+03 2.531e+03 1.007e+04, threshold=4.113e+03, percent-clipped=3.0
+2023-03-09 15:36:00,903 INFO [train.py:968] (0/2) Epoch 18, batch 41600, giga_loss[loss=0.2525, simple_loss=0.3428, pruned_loss=0.08105, over 28990.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3888, pruned_loss=0.1371, over 5655217.50 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.365, pruned_loss=0.1181, over 5700988.54 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3921, pruned_loss=0.1395, over 5641162.62 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:36:21,321 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=818330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:36:23,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=818333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:36:37,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3374, 1.1708, 4.1772, 3.3437], device='cuda:0'), covar=tensor([0.1682, 0.2895, 0.0414, 0.1268], device='cuda:0'), in_proj_covar=tensor([0.0732, 0.0631, 0.0935, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:36:52,158 INFO [train.py:968] (0/2) Epoch 18, batch 41650, giga_loss[loss=0.2957, simple_loss=0.3662, pruned_loss=0.1126, over 28704.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3865, pruned_loss=0.1341, over 5653773.42 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3647, pruned_loss=0.1179, over 5704800.55 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3898, pruned_loss=0.1365, over 5638424.84 frames. ], batch size: 92, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:36:52,344 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=818362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:36:53,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=818364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:36:57,898 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-09 15:37:36,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.577e+03 2.047e+03 2.644e+03 7.043e+03, threshold=4.093e+03, percent-clipped=9.0
+2023-03-09 15:37:40,325 INFO [train.py:968] (0/2) Epoch 18, batch 41700, giga_loss[loss=0.2629, simple_loss=0.3405, pruned_loss=0.09261, over 28749.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3842, pruned_loss=0.131, over 5668318.58 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3655, pruned_loss=0.1184, over 5709145.46 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3866, pruned_loss=0.1328, over 5651295.27 frames. ], batch size: 99, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:38:32,399 INFO [train.py:968] (0/2) Epoch 18, batch 41750, giga_loss[loss=0.2915, simple_loss=0.3708, pruned_loss=0.106, over 28741.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3797, pruned_loss=0.1272, over 5664520.33 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3654, pruned_loss=0.1183, over 5712943.72 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.382, pruned_loss=0.1289, over 5647097.46 frames. ], batch size: 242, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:38:45,019 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-09 15:39:17,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.785e+03 2.280e+03 3.018e+03 6.979e+03, threshold=4.560e+03, percent-clipped=10.0
+2023-03-09 15:39:22,484 INFO [train.py:968] (0/2) Epoch 18, batch 41800, giga_loss[loss=0.3458, simple_loss=0.3763, pruned_loss=0.1576, over 23685.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3768, pruned_loss=0.1256, over 5652421.39 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3654, pruned_loss=0.1184, over 5711311.74 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3788, pruned_loss=0.1269, over 5639475.35 frames. ], batch size: 705, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:39:35,991 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=818524.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:40:12,979 INFO [train.py:968] (0/2) Epoch 18, batch 41850, libri_loss[loss=0.3092, simple_loss=0.3806, pruned_loss=0.1189, over 27823.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3765, pruned_loss=0.1252, over 5664230.07 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3654, pruned_loss=0.1182, over 5714460.76 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3784, pruned_loss=0.1266, over 5649645.27 frames. ], batch size: 116, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:40:54,475 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.674e+03 2.212e+03 2.952e+03 7.409e+03, threshold=4.423e+03, percent-clipped=5.0
+2023-03-09 15:41:03,734 INFO [train.py:968] (0/2) Epoch 18, batch 41900, giga_loss[loss=0.2973, simple_loss=0.3698, pruned_loss=0.1124, over 28890.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3766, pruned_loss=0.1247, over 5674048.73 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3654, pruned_loss=0.1182, over 5716423.69 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3782, pruned_loss=0.1259, over 5660176.73 frames. ], batch size: 186, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:41:34,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7761, 2.6109, 1.7234, 0.8767], device='cuda:0'), covar=tensor([0.8190, 0.3381, 0.3785, 0.7790], device='cuda:0'), in_proj_covar=tensor([0.1701, 0.1608, 0.1574, 0.1385], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 15:41:39,045 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=818646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:41:54,089 INFO [train.py:968] (0/2) Epoch 18, batch 41950, giga_loss[loss=0.2966, simple_loss=0.3746, pruned_loss=0.1093, over 28859.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3743, pruned_loss=0.1223, over 5668971.69 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3654, pruned_loss=0.1183, over 5705596.19 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3758, pruned_loss=0.1233, over 5666757.90 frames. ], batch size: 199, lr: 1.75e-03, grad_scale: 2.0
+2023-03-09 15:42:37,818 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 15:42:44,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.410e+02 1.548e+03 1.911e+03 2.668e+03 7.866e+03, threshold=3.823e+03, percent-clipped=7.0
+2023-03-09 15:42:48,678 INFO [train.py:968] (0/2) Epoch 18, batch 42000, giga_loss[loss=0.315, simple_loss=0.3951, pruned_loss=0.1175, over 28940.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3748, pruned_loss=0.1201, over 5679042.12 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3654, pruned_loss=0.1183, over 5708722.23 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3762, pruned_loss=0.121, over 5673834.69 frames. ], batch size: 186, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:42:48,682 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 15:42:57,324 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2053, simple_loss=0.3105, pruned_loss=0.05006, over 944034.00 frames. 
+2023-03-09 15:42:57,325 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 15:43:15,176 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-09 15:43:25,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=818739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:43:32,776 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=818746.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:43:46,822 INFO [train.py:968] (0/2) Epoch 18, batch 42050, giga_loss[loss=0.2761, simple_loss=0.3556, pruned_loss=0.0983, over 28900.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3755, pruned_loss=0.1202, over 5676189.95 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3647, pruned_loss=0.1178, over 5714724.71 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3775, pruned_loss=0.1214, over 5665490.59 frames. ], batch size: 119, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:43:55,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6979, 1.7889, 1.3487, 1.3352], device='cuda:0'), covar=tensor([0.0973, 0.0719, 0.1117, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0450, 0.0517, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:44:01,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2311, 4.0424, 3.8653, 1.9092], device='cuda:0'), covar=tensor([0.0552, 0.0712, 0.0769, 0.2426], device='cuda:0'), in_proj_covar=tensor([0.1183, 0.1104, 0.0940, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 15:44:14,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2027, 1.1461, 3.8746, 3.2562], device='cuda:0'), covar=tensor([0.1719, 0.2803, 0.0447, 0.1114], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0633, 0.0935, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:44:28,003 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.551e+02 1.785e+03 2.178e+03 3.195e+03 8.757e+03, threshold=4.356e+03, percent-clipped=18.0
+2023-03-09 15:44:30,827 INFO [train.py:968] (0/2) Epoch 18, batch 42100, giga_loss[loss=0.2819, simple_loss=0.3563, pruned_loss=0.1038, over 28851.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3764, pruned_loss=0.1219, over 5680188.64 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3649, pruned_loss=0.118, over 5718451.16 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3782, pruned_loss=0.1228, over 5666914.66 frames. ], batch size: 119, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:44:32,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3288, 1.6633, 1.6040, 1.1673], device='cuda:0'), covar=tensor([0.1624, 0.2580, 0.1441, 0.1708], device='cuda:0'), in_proj_covar=tensor([0.0877, 0.0696, 0.0922, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 15:45:14,732 INFO [train.py:968] (0/2) Epoch 18, batch 42150, giga_loss[loss=0.3138, simple_loss=0.3821, pruned_loss=0.1228, over 28863.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.376, pruned_loss=0.1225, over 5684339.39 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3647, pruned_loss=0.118, over 5723502.14 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.378, pruned_loss=0.1233, over 5667920.41 frames. ], batch size: 284, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:45:34,435 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=818882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:45:36,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=818885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:45:39,243 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 15:45:47,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=818899.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:45:57,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.160e+03 1.786e+03 2.396e+03 3.327e+03 6.716e+03, threshold=4.792e+03, percent-clipped=10.0
+2023-03-09 15:46:00,764 INFO [train.py:968] (0/2) Epoch 18, batch 42200, giga_loss[loss=0.3364, simple_loss=0.3974, pruned_loss=0.1377, over 29003.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3745, pruned_loss=0.1223, over 5678836.73 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3645, pruned_loss=0.1179, over 5717614.89 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3767, pruned_loss=0.1233, over 5669886.25 frames. ], batch size: 155, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:46:02,441 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=818914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:46:51,229 INFO [train.py:968] (0/2) Epoch 18, batch 42250, giga_loss[loss=0.2766, simple_loss=0.3518, pruned_loss=0.1007, over 28627.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3726, pruned_loss=0.1219, over 5669377.30 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3646, pruned_loss=0.1179, over 5719598.06 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3743, pruned_loss=0.1226, over 5660028.45 frames. ], batch size: 262, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:47:02,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6143, 1.6184, 1.8590, 1.4223], device='cuda:0'), covar=tensor([0.1880, 0.2384, 0.1458, 0.1700], device='cuda:0'), in_proj_covar=tensor([0.0878, 0.0697, 0.0923, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 15:47:40,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.152e+03 1.479e+03 1.834e+03 2.713e+03 5.292e+03, threshold=3.667e+03, percent-clipped=3.0
+2023-03-09 15:47:44,177 INFO [train.py:968] (0/2) Epoch 18, batch 42300, libri_loss[loss=0.2642, simple_loss=0.3261, pruned_loss=0.1012, over 29375.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3716, pruned_loss=0.1203, over 5676284.64 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3643, pruned_loss=0.1177, over 5721299.44 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3733, pruned_loss=0.1211, over 5666921.92 frames. ], batch size: 67, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:47:53,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:48:13,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819042.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:48:14,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:48:16,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819045.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:48:32,060 INFO [train.py:968] (0/2) Epoch 18, batch 42350, giga_loss[loss=0.3201, simple_loss=0.3673, pruned_loss=0.1364, over 24007.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3706, pruned_loss=0.1182, over 5687256.68 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3642, pruned_loss=0.1175, over 5724152.86 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3722, pruned_loss=0.119, over 5676350.15 frames. ], batch size: 705, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:48:43,754 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=819074.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:49:13,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9225, 1.9559, 1.9575, 1.7996], device='cuda:0'), covar=tensor([0.1827, 0.2626, 0.2144, 0.2371], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0745, 0.0704, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 15:49:13,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6987, 2.0604, 1.8815, 1.4522], device='cuda:0'), covar=tensor([0.2979, 0.2303, 0.2629, 0.2888], device='cuda:0'), in_proj_covar=tensor([0.1916, 0.1828, 0.1772, 0.1907], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 15:49:16,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.734e+02 1.472e+03 2.170e+03 2.833e+03 5.964e+03, threshold=4.340e+03, percent-clipped=10.0
+2023-03-09 15:49:19,548 INFO [train.py:968] (0/2) Epoch 18, batch 42400, giga_loss[loss=0.3071, simple_loss=0.3623, pruned_loss=0.126, over 28910.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3706, pruned_loss=0.1179, over 5694840.19 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3637, pruned_loss=0.1173, over 5728442.64 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3725, pruned_loss=0.1188, over 5681351.49 frames. ], batch size: 99, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 15:49:27,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819121.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:49:31,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1643, 1.0905, 3.7298, 3.1986], device='cuda:0'), covar=tensor([0.1717, 0.2906, 0.0508, 0.0970], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0634, 0.0938, 0.0876], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:50:02,052 INFO [train.py:968] (0/2) Epoch 18, batch 42450, giga_loss[loss=0.3222, simple_loss=0.3677, pruned_loss=0.1384, over 23769.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3687, pruned_loss=0.1173, over 5690047.77 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3633, pruned_loss=0.117, over 5732536.68 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3708, pruned_loss=0.1183, over 5674114.35 frames. ], batch size: 705, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 15:50:03,837 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:50:06,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2852, 1.5487, 1.3740, 1.4933], device='cuda:0'), covar=tensor([0.0751, 0.0319, 0.0313, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0117, 0.0115, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 15:50:06,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:50:33,306 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=819196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:50:43,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.252e+02 1.627e+03 2.149e+03 3.033e+03 5.651e+03, threshold=4.299e+03, percent-clipped=6.0
+2023-03-09 15:50:45,972 INFO [train.py:968] (0/2) Epoch 18, batch 42500, libri_loss[loss=0.3031, simple_loss=0.3688, pruned_loss=0.1187, over 29549.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.369, pruned_loss=0.1182, over 5680275.41 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.364, pruned_loss=0.1175, over 5723557.55 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3702, pruned_loss=0.1186, over 5674158.32 frames. ], batch size: 89, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:51:35,893 INFO [train.py:968] (0/2) Epoch 18, batch 42550, giga_loss[loss=0.2907, simple_loss=0.3575, pruned_loss=0.112, over 28840.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3687, pruned_loss=0.1188, over 5660185.93 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3645, pruned_loss=0.1177, over 5706845.32 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3694, pruned_loss=0.1189, over 5668928.91 frames. ], batch size: 199, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:51:38,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819264.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:51:40,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819267.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:51:50,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2352, 2.8569, 1.3702, 1.4716], device='cuda:0'), covar=tensor([0.0993, 0.0377, 0.0867, 0.1286], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0548, 0.0374, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 15:52:06,507 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=819296.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:52:17,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.783e+03 2.275e+03 3.161e+03 9.070e+03, threshold=4.550e+03, percent-clipped=11.0
+2023-03-09 15:52:20,636 INFO [train.py:968] (0/2) Epoch 18, batch 42600, giga_loss[loss=0.2727, simple_loss=0.3454, pruned_loss=0.1, over 28381.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3694, pruned_loss=0.1203, over 5660977.44 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3645, pruned_loss=0.1176, over 5711772.71 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3701, pruned_loss=0.1205, over 5661743.98 frames. ], batch size: 65, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:52:42,877 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:53:05,216 INFO [train.py:968] (0/2) Epoch 18, batch 42650, giga_loss[loss=0.3306, simple_loss=0.3963, pruned_loss=0.1324, over 28680.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3678, pruned_loss=0.1197, over 5668755.65 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3643, pruned_loss=0.1176, over 5708888.88 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3687, pruned_loss=0.12, over 5670261.49 frames. ], batch size: 262, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:53:05,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9327, 2.9614, 1.9343, 1.0270], device='cuda:0'), covar=tensor([0.7142, 0.2845, 0.3585, 0.6389], device='cuda:0'), in_proj_covar=tensor([0.1702, 0.1613, 0.1575, 0.1389], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 15:53:30,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9487, 1.1435, 1.2536, 0.9817], device='cuda:0'), covar=tensor([0.1589, 0.1349, 0.2076, 0.1524], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0750, 0.0710, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 15:53:52,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.396e+02 1.836e+03 2.653e+03 3.701e+03 1.173e+04, threshold=5.306e+03, percent-clipped=13.0
+2023-03-09 15:53:55,302 INFO [train.py:968] (0/2) Epoch 18, batch 42700, giga_loss[loss=0.2684, simple_loss=0.3376, pruned_loss=0.0996, over 29009.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3678, pruned_loss=0.1201, over 5676757.26 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3641, pruned_loss=0.1174, over 5712200.82 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3688, pruned_loss=0.1205, over 5674155.50 frames. ], batch size: 128, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:54:02,215 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:54:15,776 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:54:44,139 INFO [train.py:968] (0/2) Epoch 18, batch 42750, giga_loss[loss=0.3048, simple_loss=0.3753, pruned_loss=0.1172, over 28992.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3672, pruned_loss=0.1193, over 5685155.00 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3636, pruned_loss=0.1171, over 5714641.04 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3685, pruned_loss=0.1201, over 5680056.03 frames. ], batch size: 164, lr: 1.75e-03, grad_scale: 2.0
+2023-03-09 15:54:50,729 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:55:23,854 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.749e+03 2.187e+03 3.241e+03 1.643e+04, threshold=4.374e+03, percent-clipped=11.0
+2023-03-09 15:55:24,265 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 15:55:25,724 INFO [train.py:968] (0/2) Epoch 18, batch 42800, giga_loss[loss=0.3212, simple_loss=0.3626, pruned_loss=0.1398, over 23537.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3671, pruned_loss=0.1185, over 5683716.05 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3633, pruned_loss=0.1167, over 5718073.00 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3687, pruned_loss=0.1195, over 5675260.58 frames. ], batch size: 705, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:55:49,170 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:56:11,719 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:56:11,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5454, 1.5379, 1.7549, 1.3597], device='cuda:0'), covar=tensor([0.1459, 0.2272, 0.1221, 0.1495], device='cuda:0'), in_proj_covar=tensor([0.0881, 0.0700, 0.0925, 0.0826], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 15:56:12,006 INFO [train.py:968] (0/2) Epoch 18, batch 42850, giga_loss[loss=0.3453, simple_loss=0.3962, pruned_loss=0.1472, over 28034.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3683, pruned_loss=0.1184, over 5687629.93 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3631, pruned_loss=0.1166, over 5718947.18 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3697, pruned_loss=0.1193, over 5679701.17 frames. ], batch size: 412, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:56:14,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:56:38,210 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=819593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:56:54,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.718e+02 1.475e+03 1.839e+03 2.587e+03 7.407e+03, threshold=3.678e+03, percent-clipped=4.0
+2023-03-09 15:56:57,421 INFO [train.py:968] (0/2) Epoch 18, batch 42900, giga_loss[loss=0.3563, simple_loss=0.3977, pruned_loss=0.1574, over 26662.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3689, pruned_loss=0.1186, over 5683216.28 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3631, pruned_loss=0.1166, over 5721891.59 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3702, pruned_loss=0.1194, over 5673079.53 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:57:13,996 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-09 15:57:23,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0221, 1.1859, 3.3664, 3.1003], device='cuda:0'), covar=tensor([0.2102, 0.2916, 0.0902, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0635, 0.0940, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 15:57:49,279 INFO [train.py:968] (0/2) Epoch 18, batch 42950, giga_loss[loss=0.3072, simple_loss=0.3788, pruned_loss=0.1178, over 29004.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3708, pruned_loss=0.1206, over 5669189.21 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3634, pruned_loss=0.1167, over 5716352.74 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3718, pruned_loss=0.1213, over 5665061.23 frames. ], batch size: 213, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:58:33,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.159e+03 1.794e+03 2.294e+03 3.574e+03 9.245e+03, threshold=4.588e+03, percent-clipped=23.0
+2023-03-09 15:58:36,064 INFO [train.py:968] (0/2) Epoch 18, batch 43000, giga_loss[loss=0.324, simple_loss=0.3833, pruned_loss=0.1324, over 28807.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3742, pruned_loss=0.1238, over 5669697.83 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3637, pruned_loss=0.1167, over 5719900.54 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.375, pruned_loss=0.1245, over 5662104.65 frames. ], batch size: 199, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:58:37,469 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:59:32,769 INFO [train.py:968] (0/2) Epoch 18, batch 43050, giga_loss[loss=0.358, simple_loss=0.4086, pruned_loss=0.1537, over 28735.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3766, pruned_loss=0.1275, over 5660150.19 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3641, pruned_loss=0.117, over 5721769.47 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1278, over 5651752.81 frames. ], batch size: 262, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:59:42,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:00:09,063 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3407, 1.7080, 1.3561, 1.4697], device='cuda:0'), covar=tensor([0.1691, 0.1558, 0.1901, 0.1780], device='cuda:0'), in_proj_covar=tensor([0.0465, 0.0746, 0.0707, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 16:00:20,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:00:23,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.184e+03 1.976e+03 2.627e+03 3.876e+03 7.897e+03, threshold=5.254e+03, percent-clipped=11.0
+2023-03-09 16:00:25,510 INFO [train.py:968] (0/2) Epoch 18, batch 43100, giga_loss[loss=0.3197, simple_loss=0.3784, pruned_loss=0.1304, over 28969.00 frames. ], tot_loss[loss=0.316, simple_loss=0.376, pruned_loss=0.128, over 5666293.93 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3636, pruned_loss=0.1167, over 5726414.10 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3771, pruned_loss=0.1288, over 5653719.77 frames. ], batch size: 164, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:00:56,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:01:07,090 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:01:10,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:01:12,427 INFO [train.py:968] (0/2) Epoch 18, batch 43150, giga_loss[loss=0.2734, simple_loss=0.3527, pruned_loss=0.09705, over 28892.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.376, pruned_loss=0.1278, over 5663855.73 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.364, pruned_loss=0.117, over 5719876.11 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3768, pruned_loss=0.1286, over 5658861.89 frames. ], batch size: 199, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:01:36,403 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=819888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:01:57,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.721e+03 2.252e+03 2.963e+03 7.383e+03, threshold=4.503e+03, percent-clipped=3.0
+2023-03-09 16:01:59,392 INFO [train.py:968] (0/2) Epoch 18, batch 43200, giga_loss[loss=0.2991, simple_loss=0.3559, pruned_loss=0.1212, over 29108.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3733, pruned_loss=0.1262, over 5673564.72 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3637, pruned_loss=0.1168, over 5721784.98 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3744, pruned_loss=0.1271, over 5667314.72 frames. ], batch size: 128, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:01:59,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:02:34,109 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:02:36,004 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:02:41,855 INFO [train.py:968] (0/2) Epoch 18, batch 43250, giga_loss[loss=0.3046, simple_loss=0.3718, pruned_loss=0.1186, over 28211.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3724, pruned_loss=0.1236, over 5679003.56 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3637, pruned_loss=0.1169, over 5722740.68 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3736, pruned_loss=0.1245, over 5671532.57 frames. ], batch size: 368, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:03:03,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=819983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:03:07,809 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:03:09,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:03:19,994 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-820000.pt
+2023-03-09 16:03:31,995 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.734e+03 2.365e+03 3.202e+03 9.816e+03, threshold=4.730e+03, percent-clipped=11.0
+2023-03-09 16:03:32,608 INFO [train.py:968] (0/2) Epoch 18, batch 43300, giga_loss[loss=0.3205, simple_loss=0.3803, pruned_loss=0.1303, over 28505.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3706, pruned_loss=0.1222, over 5664316.70 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3641, pruned_loss=0.1172, over 5711989.41 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3713, pruned_loss=0.1227, over 5667546.18 frames. ], batch size: 336, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:03:39,222 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=820019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:03:39,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2345, 1.5435, 1.5402, 1.1100], device='cuda:0'), covar=tensor([0.1541, 0.2324, 0.1262, 0.1495], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0696, 0.0923, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 16:04:11,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=820055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:04:13,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=820058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:04:16,745 INFO [train.py:968] (0/2) Epoch 18, batch 43350, giga_loss[loss=0.2752, simple_loss=0.3419, pruned_loss=0.1042, over 28715.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3699, pruned_loss=0.1224, over 5660040.08 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3647, pruned_loss=0.1176, over 5710468.81 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3702, pruned_loss=0.1227, over 5661646.63 frames. ], batch size: 92, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:04:30,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=820077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:04:39,451 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=820087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:05:03,299 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.683e+03 2.175e+03 3.051e+03 6.920e+03, threshold=4.350e+03, percent-clipped=6.0
+2023-03-09 16:05:03,995 INFO [train.py:968] (0/2) Epoch 18, batch 43400, giga_loss[loss=0.2951, simple_loss=0.3641, pruned_loss=0.1131, over 28971.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3682, pruned_loss=0.1219, over 5649894.33 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3646, pruned_loss=0.1177, over 5694157.80 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3685, pruned_loss=0.1221, over 5665421.20 frames. ], batch size: 164, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:05:35,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=820144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:05:52,424 INFO [train.py:968] (0/2) Epoch 18, batch 43450, giga_loss[loss=0.3395, simple_loss=0.3922, pruned_loss=0.1434, over 28591.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3695, pruned_loss=0.123, over 5649244.35 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.365, pruned_loss=0.118, over 5697071.45 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3694, pruned_loss=0.1229, over 5658291.60 frames. ], batch size: 71, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:05:52,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4302, 2.4379, 1.8369, 2.2806], device='cuda:0'), covar=tensor([0.0785, 0.0556, 0.0877, 0.0921], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0448, 0.0514, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 16:06:35,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.931e+02 1.686e+03 2.223e+03 3.345e+03 9.246e+03, threshold=4.446e+03, percent-clipped=13.0
+2023-03-09 16:06:36,207 INFO [train.py:968] (0/2) Epoch 18, batch 43500, giga_loss[loss=0.2924, simple_loss=0.3748, pruned_loss=0.105, over 28770.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3722, pruned_loss=0.1235, over 5662357.41 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3649, pruned_loss=0.118, over 5701544.50 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3724, pruned_loss=0.1237, over 5664425.99 frames. ], batch size: 119, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:06:52,898 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-09 16:07:23,264 INFO [train.py:968] (0/2) Epoch 18, batch 43550, giga_loss[loss=0.3042, simple_loss=0.3894, pruned_loss=0.1095, over 29019.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.375, pruned_loss=0.1226, over 5662493.29 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3651, pruned_loss=0.1182, over 5705852.64 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3752, pruned_loss=0.1227, over 5659419.11 frames. ], batch size: 128, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:07:46,309 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=820287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:07:51,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=820290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:07:54,385 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.52 vs. limit=5.0
+2023-03-09 16:08:11,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.596e+03 1.956e+03 2.695e+03 7.068e+03, threshold=3.912e+03, percent-clipped=5.0
+2023-03-09 16:08:11,757 INFO [train.py:968] (0/2) Epoch 18, batch 43600, giga_loss[loss=0.3696, simple_loss=0.4099, pruned_loss=0.1647, over 26670.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.377, pruned_loss=0.1238, over 5662299.52 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3647, pruned_loss=0.1181, over 5703891.29 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3778, pruned_loss=0.1241, over 5659616.67 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:08:19,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=820319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:09:00,650 INFO [train.py:968] (0/2) Epoch 18, batch 43650, giga_loss[loss=0.3171, simple_loss=0.3801, pruned_loss=0.1271, over 28811.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3791, pruned_loss=0.1256, over 5672511.50 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3648, pruned_loss=0.1181, over 5706534.65 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.38, pruned_loss=0.126, over 5667196.03 frames. ], batch size: 186, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:09:52,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.139e+03 1.678e+03 2.118e+03 2.495e+03 4.270e+03, threshold=4.237e+03, percent-clipped=1.0
+2023-03-09 16:09:52,954 INFO [train.py:968] (0/2) Epoch 18, batch 43700, giga_loss[loss=0.3091, simple_loss=0.3697, pruned_loss=0.1242, over 28855.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3804, pruned_loss=0.1273, over 5665877.03 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3648, pruned_loss=0.1181, over 5704803.20 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3812, pruned_loss=0.1276, over 5662878.08 frames. ], batch size: 199, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:10:27,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=820452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:10:37,294 INFO [train.py:968] (0/2) Epoch 18, batch 43750, giga_loss[loss=0.3936, simple_loss=0.4181, pruned_loss=0.1845, over 23490.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3788, pruned_loss=0.1266, over 5673493.52 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3648, pruned_loss=0.1181, over 5707675.37 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3796, pruned_loss=0.127, over 5668254.42 frames. ], batch size: 705, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:10:41,795 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=820466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:11:29,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.806e+02 1.497e+03 1.969e+03 2.718e+03 8.578e+03, threshold=3.938e+03, percent-clipped=7.0
+2023-03-09 16:11:29,114 INFO [train.py:968] (0/2) Epoch 18, batch 43800, giga_loss[loss=0.3359, simple_loss=0.3905, pruned_loss=0.1407, over 27545.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3772, pruned_loss=0.1264, over 5664766.44 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3641, pruned_loss=0.1177, over 5710844.81 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3787, pruned_loss=0.1273, over 5657154.57 frames. ], batch size: 472, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:11:43,884 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=820528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:12:17,576 INFO [train.py:968] (0/2) Epoch 18, batch 43850, giga_loss[loss=0.2704, simple_loss=0.342, pruned_loss=0.09942, over 28987.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3758, pruned_loss=0.1262, over 5667681.96 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3643, pruned_loss=0.1177, over 5712643.28 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3769, pruned_loss=0.1269, over 5659886.98 frames. ], batch size: 128, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:12:39,004 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-09 16:12:46,122 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=820592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:12:50,513 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=820595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:12:55,405 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=820598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:13:12,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.911e+02 1.716e+03 2.026e+03 2.799e+03 7.530e+03, threshold=4.052e+03, percent-clipped=12.0
+2023-03-09 16:13:12,191 INFO [train.py:968] (0/2) Epoch 18, batch 43900, giga_loss[loss=0.3846, simple_loss=0.4125, pruned_loss=0.1783, over 26622.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3751, pruned_loss=0.1266, over 5638953.36 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3644, pruned_loss=0.1178, over 5697795.63 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3762, pruned_loss=0.1274, over 5644170.52 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:13:24,318 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=820627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:13:52,684 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 16:13:56,975 INFO [train.py:968] (0/2) Epoch 18, batch 43950, giga_loss[loss=0.2743, simple_loss=0.3465, pruned_loss=0.101, over 28971.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3726, pruned_loss=0.1247, over 5649647.13 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3635, pruned_loss=0.1172, over 5705635.25 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3747, pruned_loss=0.1262, over 5644436.12 frames. ], batch size: 213, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:14:07,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3887, 1.7632, 1.3663, 1.3234], device='cuda:0'), covar=tensor([0.2524, 0.2489, 0.2921, 0.2239], device='cuda:0'), in_proj_covar=tensor([0.1443, 0.1049, 0.1283, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 16:14:38,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5527, 4.4624, 1.7873, 1.7036], device='cuda:0'), covar=tensor([0.0977, 0.0332, 0.0841, 0.1293], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0548, 0.0374, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 16:14:45,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.129e+03 1.654e+03 2.300e+03 2.916e+03 8.510e+03, threshold=4.599e+03, percent-clipped=9.0
+2023-03-09 16:14:45,178 INFO [train.py:968] (0/2) Epoch 18, batch 44000, libri_loss[loss=0.2435, simple_loss=0.3146, pruned_loss=0.08618, over 29570.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.371, pruned_loss=0.124, over 5659212.16 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3632, pruned_loss=0.117, over 5705739.86 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.373, pruned_loss=0.1255, over 5654191.13 frames. ], batch size: 74, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:14:47,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-09 16:15:12,680 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-09 16:15:28,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6399, 1.7923, 1.3526, 1.3508], device='cuda:0'), covar=tensor([0.0978, 0.0629, 0.1060, 0.1186], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0447, 0.0512, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 16:15:30,816 INFO [train.py:968] (0/2) Epoch 18, batch 44050, giga_loss[loss=0.2604, simple_loss=0.3317, pruned_loss=0.09459, over 28546.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3701, pruned_loss=0.1234, over 5660389.93 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3634, pruned_loss=0.1171, over 5698939.64 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3717, pruned_loss=0.1246, over 5661176.00 frames. ], batch size: 85, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:15:34,298 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-09 16:16:16,365 INFO [train.py:968] (0/2) Epoch 18, batch 44100, giga_loss[loss=0.3521, simple_loss=0.4062, pruned_loss=0.149, over 27866.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3707, pruned_loss=0.1235, over 5662988.52 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3633, pruned_loss=0.1169, over 5705294.56 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3724, pruned_loss=0.1249, over 5656700.18 frames. ], batch size: 412, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:16:18,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.824e+02 1.679e+03 2.081e+03 2.909e+03 6.379e+03, threshold=4.162e+03, percent-clipped=5.0
+2023-03-09 16:16:24,900 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-09 16:16:48,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=820841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:17:10,646 INFO [train.py:968] (0/2) Epoch 18, batch 44150, libri_loss[loss=0.3283, simple_loss=0.3865, pruned_loss=0.135, over 26138.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3727, pruned_loss=0.1242, over 5654043.88 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3629, pruned_loss=0.1165, over 5704704.84 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3746, pruned_loss=0.1258, over 5648424.90 frames. ], batch size: 136, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:17:50,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=820903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:17:50,843 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2890, 1.5417, 1.2764, 1.0131], device='cuda:0'), covar=tensor([0.2200, 0.2220, 0.2390, 0.2012], device='cuda:0'), in_proj_covar=tensor([0.1445, 0.1050, 0.1284, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 16:17:57,781 INFO [train.py:968] (0/2) Epoch 18, batch 44200, giga_loss[loss=0.3179, simple_loss=0.3826, pruned_loss=0.1266, over 28865.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3743, pruned_loss=0.1254, over 5649420.65 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3633, pruned_loss=0.1168, over 5697877.48 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3756, pruned_loss=0.1265, over 5650468.53 frames. ], batch size: 174, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:17:58,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.640e+03 2.228e+03 3.302e+03 6.595e+03, threshold=4.456e+03, percent-clipped=15.0
+2023-03-09 16:18:48,884 INFO [train.py:968] (0/2) Epoch 18, batch 44250, giga_loss[loss=0.2779, simple_loss=0.3679, pruned_loss=0.09388, over 29119.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3753, pruned_loss=0.125, over 5652183.32 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3633, pruned_loss=0.1168, over 5689799.88 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3764, pruned_loss=0.126, over 5660792.64 frames. ], batch size: 128, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:18:54,643 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=820967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:19:11,002 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=820984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:19:12,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=820987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:19:36,565 INFO [train.py:968] (0/2) Epoch 18, batch 44300, giga_loss[loss=0.3626, simple_loss=0.413, pruned_loss=0.1561, over 27538.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.377, pruned_loss=0.1239, over 5655767.70 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3632, pruned_loss=0.1168, over 5690749.88 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.378, pruned_loss=0.1247, over 5661530.54 frames. ], batch size: 472, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:19:37,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.030e+02 1.562e+03 2.017e+03 3.053e+03 9.522e+03, threshold=4.034e+03, percent-clipped=6.0
+2023-03-09 16:19:40,871 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=821016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:20:05,417 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=821046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:20:08,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=821049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:20:23,798 INFO [train.py:968] (0/2) Epoch 18, batch 44350, giga_loss[loss=0.3551, simple_loss=0.4022, pruned_loss=0.154, over 27675.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3798, pruned_loss=0.1247, over 5660111.80 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3635, pruned_loss=0.117, over 5693003.76 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3807, pruned_loss=0.1253, over 5662000.65 frames. ], batch size: 472, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:20:40,843 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=821078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:21:10,801 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=821110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:21:14,408 INFO [train.py:968] (0/2) Epoch 18, batch 44400, giga_loss[loss=0.2932, simple_loss=0.3622, pruned_loss=0.1121, over 28711.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3813, pruned_loss=0.1263, over 5658139.83 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3636, pruned_loss=0.1169, over 5698214.29 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3825, pruned_loss=0.1271, over 5654147.91 frames. ], batch size: 85, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:21:14,994 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.654e+03 2.228e+03 3.243e+03 6.586e+03, threshold=4.457e+03, percent-clipped=8.0
+2023-03-09 16:21:15,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=821113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:21:22,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0013, 3.8260, 3.6618, 1.9926], device='cuda:0'), covar=tensor([0.0671, 0.0802, 0.0776, 0.1941], device='cuda:0'), in_proj_covar=tensor([0.1193, 0.1114, 0.0950, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 16:21:42,761 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=821142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:22:02,287 INFO [train.py:968] (0/2) Epoch 18, batch 44450, giga_loss[loss=0.3053, simple_loss=0.3702, pruned_loss=0.1203, over 28569.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3836, pruned_loss=0.1294, over 5653999.10 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3637, pruned_loss=0.117, over 5700339.79 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3846, pruned_loss=0.1301, over 5648567.69 frames. ], batch size: 71, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:22:25,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=821185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:22:51,210 INFO [train.py:968] (0/2) Epoch 18, batch 44500, giga_loss[loss=0.3228, simple_loss=0.3849, pruned_loss=0.1304, over 28257.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.382, pruned_loss=0.1286, over 5675696.26 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3634, pruned_loss=0.1169, over 5705617.02 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3835, pruned_loss=0.1295, over 5665974.08 frames. ], batch size: 368, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:22:52,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.159e+03 1.719e+03 2.269e+03 3.179e+03 8.467e+03, threshold=4.538e+03, percent-clipped=9.0
+2023-03-09 16:23:34,688 INFO [train.py:968] (0/2) Epoch 18, batch 44550, giga_loss[loss=0.3135, simple_loss=0.3877, pruned_loss=0.1197, over 29023.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3809, pruned_loss=0.1278, over 5675894.13 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3637, pruned_loss=0.117, over 5709515.79 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3823, pruned_loss=0.1287, over 5663853.56 frames. ], batch size: 155, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:24:16,067 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=821310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:24:17,124 INFO [train.py:968] (0/2) Epoch 18, batch 44600, giga_loss[loss=0.336, simple_loss=0.4005, pruned_loss=0.1358, over 28937.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3792, pruned_loss=0.1253, over 5684888.16 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3634, pruned_loss=0.1168, over 5714651.87 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.381, pruned_loss=0.1264, over 5669531.20 frames. ], batch size: 227, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:24:18,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.590e+02 1.655e+03 2.180e+03 2.961e+03 1.076e+04, threshold=4.361e+03, percent-clipped=7.0
+2023-03-09 16:25:06,088 INFO [train.py:968] (0/2) Epoch 18, batch 44650, giga_loss[loss=0.3583, simple_loss=0.4067, pruned_loss=0.155, over 27677.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3796, pruned_loss=0.1241, over 5677012.57 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3636, pruned_loss=0.1169, over 5708841.46 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3812, pruned_loss=0.125, over 5670324.77 frames. ], batch size: 472, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:25:53,476 INFO [train.py:968] (0/2) Epoch 18, batch 44700, giga_loss[loss=0.3336, simple_loss=0.3923, pruned_loss=0.1374, over 28297.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3809, pruned_loss=0.1252, over 5664022.56 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3637, pruned_loss=0.117, over 5707276.68 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3824, pruned_loss=0.126, over 5659550.67 frames. ], batch size: 368, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:25:55,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.547e+02 1.569e+03 2.072e+03 3.181e+03 8.241e+03, threshold=4.144e+03, percent-clipped=7.0
+2023-03-09 16:26:03,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4151, 1.5489, 1.6677, 1.2644], device='cuda:0'), covar=tensor([0.1431, 0.2342, 0.1213, 0.1542], device='cuda:0'), in_proj_covar=tensor([0.0882, 0.0705, 0.0929, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 16:26:40,811 INFO [train.py:968] (0/2) Epoch 18, batch 44750, giga_loss[loss=0.2679, simple_loss=0.3477, pruned_loss=0.09407, over 28858.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3813, pruned_loss=0.1263, over 5648552.14 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3639, pruned_loss=0.117, over 5693314.92 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3829, pruned_loss=0.1271, over 5656206.22 frames. ], batch size: 174, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:27:24,147 INFO [train.py:968] (0/2) Epoch 18, batch 44800, giga_loss[loss=0.3901, simple_loss=0.424, pruned_loss=0.178, over 28648.00 frames. ], tot_loss[loss=0.316, simple_loss=0.38, pruned_loss=0.1259, over 5645306.71 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3638, pruned_loss=0.117, over 5683688.83 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3816, pruned_loss=0.1268, over 5659160.53 frames. ], batch size: 307, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:27:26,305 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.808e+03 2.553e+03 3.576e+03 1.124e+04, threshold=5.105e+03, percent-clipped=18.0
+2023-03-09 16:27:47,331 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-09 16:28:14,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=821560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:28:15,490 INFO [train.py:968] (0/2) Epoch 18, batch 44850, giga_loss[loss=0.3206, simple_loss=0.3816, pruned_loss=0.1298, over 28678.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3784, pruned_loss=0.126, over 5649855.40 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3642, pruned_loss=0.1173, over 5686196.07 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3796, pruned_loss=0.1267, over 5658019.68 frames. ], batch size: 307, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:28:28,707 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 16:28:31,057 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 16:29:06,166 INFO [train.py:968] (0/2) Epoch 18, batch 44900, giga_loss[loss=0.3321, simple_loss=0.386, pruned_loss=0.1391, over 28350.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.377, pruned_loss=0.1259, over 5652520.88 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3642, pruned_loss=0.1174, over 5689532.33 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3781, pruned_loss=0.1264, over 5655503.52 frames. ], batch size: 369, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:29:08,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.159e+03 1.827e+03 2.321e+03 3.361e+03 6.052e+03, threshold=4.641e+03, percent-clipped=4.0
+2023-03-09 16:29:54,317 INFO [train.py:968] (0/2) Epoch 18, batch 44950, giga_loss[loss=0.2817, simple_loss=0.3523, pruned_loss=0.1056, over 28916.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3755, pruned_loss=0.1255, over 5652649.55 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3641, pruned_loss=0.1173, over 5690819.07 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3766, pruned_loss=0.1262, over 5653286.30 frames. ], batch size: 174, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:30:16,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=821685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:30:23,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3712, 1.6577, 1.3686, 1.0772], device='cuda:0'), covar=tensor([0.2374, 0.2370, 0.2668, 0.2156], device='cuda:0'), in_proj_covar=tensor([0.1446, 0.1051, 0.1284, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 16:30:38,633 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=821703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:30:39,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6856, 1.8169, 1.2647, 1.4613], device='cuda:0'), covar=tensor([0.0947, 0.0658, 0.1019, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0449, 0.0515, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 16:30:41,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=821706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:30:45,117 INFO [train.py:968] (0/2) Epoch 18, batch 45000, giga_loss[loss=0.2924, simple_loss=0.3611, pruned_loss=0.1119, over 28969.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3751, pruned_loss=0.1264, over 5642699.36 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3642, pruned_loss=0.1174, over 5692738.73 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3761, pruned_loss=0.1269, over 5641295.20 frames. ], batch size: 136, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:30:45,122 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 16:30:51,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0601, 1.5609, 1.6428, 1.3215], device='cuda:0'), covar=tensor([0.2255, 0.1566, 0.2147, 0.1946], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0748, 0.0707, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 16:30:55,170 INFO [train.py:1012] (0/2) Epoch 18, validation: loss=0.2106, simple_loss=0.3201, pruned_loss=0.05053, over 944034.00 frames. 
+2023-03-09 16:30:55,171 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 16:30:56,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.978e+02 1.628e+03 2.222e+03 2.872e+03 6.881e+03, threshold=4.443e+03, percent-clipped=9.0
+2023-03-09 16:31:08,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 16:31:15,168 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=821735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:31:39,244 INFO [train.py:968] (0/2) Epoch 18, batch 45050, giga_loss[loss=0.3398, simple_loss=0.39, pruned_loss=0.1448, over 26782.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.372, pruned_loss=0.1231, over 5643964.44 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3642, pruned_loss=0.1174, over 5692066.27 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3729, pruned_loss=0.1236, over 5642673.22 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:32:06,664 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=821788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:32:29,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3173, 1.1513, 4.3066, 3.4461], device='cuda:0'), covar=tensor([0.1733, 0.2983, 0.0401, 0.1105], device='cuda:0'), in_proj_covar=tensor([0.0738, 0.0638, 0.0943, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 16:32:29,340 INFO [train.py:968] (0/2) Epoch 18, batch 45100, giga_loss[loss=0.2738, simple_loss=0.3528, pruned_loss=0.09742, over 28716.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3673, pruned_loss=0.1179, over 5650795.69 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3641, pruned_loss=0.1173, over 5693163.94 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3681, pruned_loss=0.1184, over 5648533.02 frames. ], batch size: 262, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:32:32,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.607e+02 1.311e+03 1.628e+03 2.182e+03 9.246e+03, threshold=3.257e+03, percent-clipped=8.0
+2023-03-09 16:32:42,519 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=821827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:32:43,124 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=821828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:32:45,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=821831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:33:16,684 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 16:33:17,749 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=821860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:33:20,197 INFO [train.py:968] (0/2) Epoch 18, batch 45150, giga_loss[loss=0.308, simple_loss=0.3746, pruned_loss=0.1207, over 28578.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3669, pruned_loss=0.1181, over 5650327.41 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.364, pruned_loss=0.1173, over 5697693.49 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3677, pruned_loss=0.1185, over 5643724.29 frames. ], batch size: 307, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:33:30,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9348, 1.2420, 1.4252, 0.9488], device='cuda:0'), covar=tensor([0.1554, 0.1106, 0.1785, 0.1464], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0750, 0.0709, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 16:33:34,583 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=821878.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 16:33:36,080 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 16:33:52,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4326, 1.6709, 1.7137, 1.2538], device='cuda:0'), covar=tensor([0.1616, 0.2459, 0.1309, 0.1583], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0705, 0.0930, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 16:33:55,364 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2704, 3.4947, 2.5608, 1.5134], device='cuda:0'), covar=tensor([0.5918, 0.2115, 0.2892, 0.5166], device='cuda:0'), in_proj_covar=tensor([0.1698, 0.1609, 0.1567, 0.1385], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 16:34:05,032 INFO [train.py:968] (0/2) Epoch 18, batch 45200, giga_loss[loss=0.3293, simple_loss=0.3807, pruned_loss=0.139, over 28521.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3664, pruned_loss=0.1184, over 5667835.23 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.364, pruned_loss=0.1173, over 5700240.29 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3672, pruned_loss=0.1188, over 5658311.12 frames. ], batch size: 336, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:34:07,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.056e+02 1.668e+03 2.177e+03 2.695e+03 5.772e+03, threshold=4.353e+03, percent-clipped=18.0
+2023-03-09 16:34:12,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2341, 1.1395, 4.3719, 3.3927], device='cuda:0'), covar=tensor([0.1773, 0.2903, 0.0402, 0.1151], device='cuda:0'), in_proj_covar=tensor([0.0740, 0.0640, 0.0946, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-09 16:34:45,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7415, 1.7300, 1.4395, 1.3401], device='cuda:0'), covar=tensor([0.0901, 0.0651, 0.0977, 0.1194], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0449, 0.0514, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 16:34:56,907 INFO [train.py:968] (0/2) Epoch 18, batch 45250, giga_loss[loss=0.2541, simple_loss=0.3249, pruned_loss=0.09166, over 28768.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3635, pruned_loss=0.117, over 5675572.38 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.364, pruned_loss=0.1172, over 5703338.25 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3641, pruned_loss=0.1174, over 5664907.53 frames. ], batch size: 119, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:35:30,409 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-822000.pt
+2023-03-09 16:35:42,862 INFO [train.py:968] (0/2) Epoch 18, batch 45300, giga_loss[loss=0.2827, simple_loss=0.3606, pruned_loss=0.1023, over 29067.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3645, pruned_loss=0.1169, over 5688742.60 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.364, pruned_loss=0.1171, over 5706576.96 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3649, pruned_loss=0.1172, over 5677247.76 frames. ], batch size: 128, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:35:45,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.593e+02 1.696e+03 2.337e+03 3.339e+03 7.797e+03, threshold=4.674e+03, percent-clipped=7.0
+2023-03-09 16:36:17,315 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=822048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:36:28,385 INFO [train.py:968] (0/2) Epoch 18, batch 45350, giga_loss[loss=0.2942, simple_loss=0.3712, pruned_loss=0.1086, over 28638.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3676, pruned_loss=0.1184, over 5684229.49 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3648, pruned_loss=0.1176, over 5711544.75 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3673, pruned_loss=0.1184, over 5669689.51 frames. ], batch size: 60, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:37:04,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4128, 1.3077, 3.7856, 3.3375], device='cuda:0'), covar=tensor([0.1458, 0.2752, 0.0388, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0638, 0.0941, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 16:37:12,975 INFO [train.py:968] (0/2) Epoch 18, batch 45400, giga_loss[loss=0.3186, simple_loss=0.3663, pruned_loss=0.1355, over 28954.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3696, pruned_loss=0.12, over 5672910.61 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.365, pruned_loss=0.1177, over 5707556.38 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3694, pruned_loss=0.1199, over 5663617.10 frames. ], batch size: 106, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:37:15,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.497e+03 1.927e+03 2.521e+03 5.696e+03, threshold=3.854e+03, percent-clipped=2.0
+2023-03-09 16:37:36,782 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2435, 2.5045, 1.3306, 1.3717], device='cuda:0'), covar=tensor([0.0968, 0.0361, 0.0867, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0546, 0.0374, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 16:37:58,620 INFO [train.py:968] (0/2) Epoch 18, batch 45450, libri_loss[loss=0.3186, simple_loss=0.3844, pruned_loss=0.1264, over 29531.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3702, pruned_loss=0.1208, over 5677988.83 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.365, pruned_loss=0.1175, over 5710121.20 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3701, pruned_loss=0.1209, over 5667274.20 frames. ], batch size: 80, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:37:59,876 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:38:10,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7054, 4.5236, 4.3176, 1.9763], device='cuda:0'), covar=tensor([0.0736, 0.0910, 0.1110, 0.2106], device='cuda:0'), in_proj_covar=tensor([0.1195, 0.1117, 0.0955, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0011], device='cuda:0')
+2023-03-09 16:38:30,472 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-09 16:38:34,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:38:42,553 INFO [train.py:968] (0/2) Epoch 18, batch 45500, giga_loss[loss=0.2938, simple_loss=0.3605, pruned_loss=0.1135, over 28997.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3713, pruned_loss=0.1224, over 5645945.33 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3651, pruned_loss=0.1177, over 5687574.75 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3713, pruned_loss=0.1225, over 5655242.16 frames. ], batch size: 213, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:38:45,340 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.543e+03 1.925e+03 2.750e+03 7.366e+03, threshold=3.849e+03, percent-clipped=9.0
+2023-03-09 16:39:25,549 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822253.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 16:39:31,602 INFO [train.py:968] (0/2) Epoch 18, batch 45550, giga_loss[loss=0.2855, simple_loss=0.3633, pruned_loss=0.1039, over 28963.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3728, pruned_loss=0.1236, over 5633673.43 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3645, pruned_loss=0.1174, over 5692096.75 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3735, pruned_loss=0.124, over 5636056.54 frames. ], batch size: 106, lr: 1.75e-03, grad_scale: 2.0
+2023-03-09 16:39:51,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6811, 1.7284, 1.9661, 1.4893], device='cuda:0'), covar=tensor([0.1793, 0.2355, 0.1405, 0.1698], device='cuda:0'), in_proj_covar=tensor([0.0881, 0.0703, 0.0927, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 16:40:13,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=822306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:40:14,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=822309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:40:17,497 INFO [train.py:968] (0/2) Epoch 18, batch 45600, giga_loss[loss=0.3387, simple_loss=0.3944, pruned_loss=0.1415, over 28901.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3736, pruned_loss=0.1237, over 5642336.38 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3647, pruned_loss=0.1175, over 5683593.40 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3741, pruned_loss=0.1241, over 5650266.69 frames. ], batch size: 164, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:40:22,958 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.771e+03 2.365e+03 3.206e+03 7.922e+03, threshold=4.730e+03, percent-clipped=17.0
+2023-03-09 16:40:42,907 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=822338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:40:51,960 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=822345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:40:54,340 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=822348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:41:01,693 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6020, 1.5729, 1.8167, 1.3858], device='cuda:0'), covar=tensor([0.1744, 0.2511, 0.1398, 0.1682], device='cuda:0'), in_proj_covar=tensor([0.0881, 0.0704, 0.0927, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 16:41:09,548 INFO [train.py:968] (0/2) Epoch 18, batch 45650, giga_loss[loss=0.311, simple_loss=0.3811, pruned_loss=0.1204, over 29071.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3756, pruned_loss=0.1254, over 5628398.25 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3649, pruned_loss=0.1176, over 5671950.08 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3759, pruned_loss=0.1256, over 5644423.68 frames. ], batch size: 128, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:41:24,607 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=822377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:41:37,018 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-09 16:41:43,434 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=822396.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 16:41:45,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=822399.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 16:41:58,351 INFO [train.py:968] (0/2) Epoch 18, batch 45700, giga_loss[loss=0.2992, simple_loss=0.3726, pruned_loss=0.1129, over 29034.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3766, pruned_loss=0.1257, over 5634470.40 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3653, pruned_loss=0.1178, over 5667311.39 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3768, pruned_loss=0.126, over 5650020.84 frames. ], batch size: 136, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:42:05,493 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.088e+03 1.777e+03 2.253e+03 2.947e+03 8.818e+03, threshold=4.506e+03, percent-clipped=4.0
+2023-03-09 16:42:10,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:42:15,990 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=822428.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 16:42:16,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 16:42:45,563 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=822459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:42:47,809 INFO [train.py:968] (0/2) Epoch 18, batch 45750, giga_loss[loss=0.3275, simple_loss=0.3997, pruned_loss=0.1276, over 28593.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3759, pruned_loss=0.1243, over 5579888.49 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3657, pruned_loss=0.1184, over 5609032.52 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.376, pruned_loss=0.1241, over 5646029.59 frames. ], batch size: 336, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:43:20,156 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=822496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:43:37,001 INFO [train.py:968] (0/2) Epoch 18, batch 45800, giga_loss[loss=0.2959, simple_loss=0.3631, pruned_loss=0.1143, over 28945.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3755, pruned_loss=0.1239, over 5562260.10 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3663, pruned_loss=0.1189, over 5568023.45 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3753, pruned_loss=0.1234, over 5649637.48 frames. ], batch size: 164, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:43:43,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.827e+03 2.452e+03 3.043e+03 1.310e+04, threshold=4.904e+03, percent-clipped=7.0
+2023-03-09 16:44:28,290 INFO [train.py:968] (0/2) Epoch 18, batch 45850, libri_loss[loss=0.3975, simple_loss=0.4344, pruned_loss=0.1803, over 19648.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3748, pruned_loss=0.1244, over 5564339.51 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3668, pruned_loss=0.1194, over 5542964.13 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3743, pruned_loss=0.1236, over 5656761.51 frames. ], batch size: 186, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:44:31,748 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=822566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:44:33,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=822569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:44:35,670 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=822571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:45:00,360 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-09 16:45:02,948 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-18.pt
+2023-03-09 16:45:47,779 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=822598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:46:07,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.647e+02 1.479e+03 1.947e+03 2.516e+03 7.048e+03, threshold=3.893e+03, percent-clipped=5.0
+2023-03-09 16:46:28,327 INFO [train.py:968] (0/2) Epoch 19, batch 50, giga_loss[loss=0.2776, simple_loss=0.3624, pruned_loss=0.09636, over 28740.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3754, pruned_loss=0.1096, over 1261092.73 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3393, pruned_loss=0.08679, over 58655.78 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3771, pruned_loss=0.1106, over 1214674.18 frames. ], batch size: 284, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:46:46,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8641, 1.9768, 1.3650, 1.5630], device='cuda:0'), covar=tensor([0.0992, 0.0703, 0.1081, 0.1269], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0448, 0.0514, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 16:46:53,637 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-09 16:47:16,108 INFO [train.py:968] (0/2) Epoch 19, batch 100, giga_loss[loss=0.2943, simple_loss=0.3544, pruned_loss=0.1171, over 26672.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3636, pruned_loss=0.1032, over 2246323.33 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3516, pruned_loss=0.09523, over 202978.69 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3647, pruned_loss=0.1039, over 2117134.88 frames. ], batch size: 555, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:47:43,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.620e+02 1.116e+03 1.308e+03 1.660e+03 4.531e+03, threshold=2.616e+03, percent-clipped=1.0
+2023-03-09 16:48:03,778 INFO [train.py:968] (0/2) Epoch 19, batch 150, giga_loss[loss=0.2258, simple_loss=0.2969, pruned_loss=0.07729, over 28560.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3476, pruned_loss=0.09602, over 3009187.12 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3482, pruned_loss=0.09341, over 259473.91 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3479, pruned_loss=0.09634, over 2879104.29 frames. ], batch size: 85, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:48:44,564 INFO [train.py:968] (0/2) Epoch 19, batch 200, giga_loss[loss=0.2068, simple_loss=0.2923, pruned_loss=0.06065, over 28905.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.335, pruned_loss=0.09012, over 3615685.23 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3412, pruned_loss=0.08952, over 397885.75 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3351, pruned_loss=0.0905, over 3455316.98 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:49:10,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.497e+02 1.056e+03 1.379e+03 1.854e+03 5.545e+03, threshold=2.759e+03, percent-clipped=13.0
+2023-03-09 16:49:25,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:49:31,082 INFO [train.py:968] (0/2) Epoch 19, batch 250, giga_loss[loss=0.2167, simple_loss=0.3015, pruned_loss=0.06596, over 28830.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3247, pruned_loss=0.08545, over 4080463.23 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.344, pruned_loss=0.09263, over 533539.28 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3237, pruned_loss=0.08511, over 3907320.81 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:49:57,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:50:14,424 INFO [train.py:968] (0/2) Epoch 19, batch 300, giga_loss[loss=0.1798, simple_loss=0.2535, pruned_loss=0.05301, over 28591.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3153, pruned_loss=0.08128, over 4445296.35 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3388, pruned_loss=0.08928, over 666026.15 frames. ], giga_tot_loss[loss=0.2382, simple_loss=0.3143, pruned_loss=0.08107, over 4271265.98 frames. ], batch size: 78, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:50:21,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9466, 5.0183, 2.1620, 2.3395], device='cuda:0'), covar=tensor([0.0933, 0.0372, 0.0836, 0.1204], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0543, 0.0372, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 16:50:41,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.941e+02 9.938e+02 1.252e+03 1.693e+03 3.342e+03, threshold=2.505e+03, percent-clipped=4.0
+2023-03-09 16:50:58,790 INFO [train.py:968] (0/2) Epoch 19, batch 350, giga_loss[loss=0.2353, simple_loss=0.3035, pruned_loss=0.08358, over 28820.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3113, pruned_loss=0.08022, over 4720441.23 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3429, pruned_loss=0.0921, over 837158.21 frames. ], giga_tot_loss[loss=0.2335, simple_loss=0.3086, pruned_loss=0.07919, over 4546605.49 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:51:04,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:51:14,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3411, 1.6103, 1.6713, 1.4430], device='cuda:0'), covar=tensor([0.2102, 0.1983, 0.2339, 0.2036], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0746, 0.0703, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 16:51:30,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3784, 3.1395, 1.5446, 1.6051], device='cuda:0'), covar=tensor([0.1003, 0.0327, 0.0892, 0.1271], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0543, 0.0372, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 16:51:30,922 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=822977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:51:32,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=822980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:51:38,668 INFO [train.py:968] (0/2) Epoch 19, batch 400, giga_loss[loss=0.225, simple_loss=0.2955, pruned_loss=0.07726, over 27650.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3073, pruned_loss=0.07806, over 4948621.18 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3399, pruned_loss=0.08952, over 1060611.82 frames. ], giga_tot_loss[loss=0.229, simple_loss=0.304, pruned_loss=0.07703, over 4763918.23 frames. ], batch size: 472, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 16:51:53,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5521, 1.7953, 1.4780, 1.7173], device='cuda:0'), covar=tensor([0.0768, 0.0311, 0.0334, 0.0874], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0103], device='cuda:0')
+2023-03-09 16:51:56,458 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=823009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:51:59,591 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=823014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:52:01,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.007e+02 1.055e+03 1.367e+03 1.792e+03 4.258e+03, threshold=2.734e+03, percent-clipped=11.0
+2023-03-09 16:52:02,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=823017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:52:17,848 INFO [train.py:968] (0/2) Epoch 19, batch 450, libri_loss[loss=0.2585, simple_loss=0.3467, pruned_loss=0.08515, over 29244.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3048, pruned_loss=0.07659, over 5124202.98 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.339, pruned_loss=0.08831, over 1226687.56 frames. ], giga_tot_loss[loss=0.2261, simple_loss=0.301, pruned_loss=0.07555, over 4949052.69 frames. ], batch size: 94, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 16:52:21,061 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 16:52:27,821 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=823046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:53:02,634 INFO [train.py:968] (0/2) Epoch 19, batch 500, giga_loss[loss=0.2505, simple_loss=0.3211, pruned_loss=0.08991, over 27964.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3028, pruned_loss=0.07591, over 5257937.53 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3388, pruned_loss=0.08845, over 1296588.84 frames. ], giga_tot_loss[loss=0.2246, simple_loss=0.2994, pruned_loss=0.07487, over 5109988.79 frames. ], batch size: 412, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:53:06,468 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=823089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:53:08,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=823092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:53:28,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.885e+02 1.174e+03 1.512e+03 2.146e+03 4.720e+03, threshold=3.024e+03, percent-clipped=11.0
+2023-03-09 16:53:31,699 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=823121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:53:46,309 INFO [train.py:968] (0/2) Epoch 19, batch 550, giga_loss[loss=0.2582, simple_loss=0.3238, pruned_loss=0.09623, over 28965.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3012, pruned_loss=0.07527, over 5355392.63 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3413, pruned_loss=0.08988, over 1385615.37 frames. ], giga_tot_loss[loss=0.2224, simple_loss=0.2972, pruned_loss=0.07385, over 5229442.12 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:54:30,558 INFO [train.py:968] (0/2) Epoch 19, batch 600, giga_loss[loss=0.1963, simple_loss=0.2802, pruned_loss=0.05622, over 28905.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.2986, pruned_loss=0.0739, over 5432043.98 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3408, pruned_loss=0.08968, over 1517646.10 frames. ], giga_tot_loss[loss=0.2196, simple_loss=0.2944, pruned_loss=0.07239, over 5319166.54 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:54:48,931 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5901, 1.6451, 1.8285, 1.4315], device='cuda:0'), covar=tensor([0.1795, 0.2381, 0.1442, 0.1669], device='cuda:0'), in_proj_covar=tensor([0.0895, 0.0707, 0.0940, 0.0837], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-09 16:54:59,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.992e+02 9.670e+02 1.247e+03 1.656e+03 7.818e+03, threshold=2.493e+03, percent-clipped=9.0
+2023-03-09 16:55:00,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2303, 1.1153, 3.8529, 3.2706], device='cuda:0'), covar=tensor([0.1749, 0.3014, 0.0466, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0632, 0.0935, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 16:55:11,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=823234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:55:16,120 INFO [train.py:968] (0/2) Epoch 19, batch 650, libri_loss[loss=0.2345, simple_loss=0.3199, pruned_loss=0.07451, over 29504.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2975, pruned_loss=0.07327, over 5484951.51 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3399, pruned_loss=0.08864, over 1687165.63 frames. ], giga_tot_loss[loss=0.218, simple_loss=0.2925, pruned_loss=0.0717, over 5379864.51 frames. ], batch size: 84, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:55:55,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3773, 1.4390, 1.3726, 1.5318], device='cuda:0'), covar=tensor([0.0761, 0.0366, 0.0340, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 16:55:57,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7818, 1.0943, 2.8584, 2.7730], device='cuda:0'), covar=tensor([0.1734, 0.2714, 0.0606, 0.1116], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0632, 0.0935, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 16:56:02,113 INFO [train.py:968] (0/2) Epoch 19, batch 700, libri_loss[loss=0.2696, simple_loss=0.3531, pruned_loss=0.09308, over 29562.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2964, pruned_loss=0.07299, over 5536324.43 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3397, pruned_loss=0.08834, over 1852785.59 frames. ], giga_tot_loss[loss=0.2166, simple_loss=0.2908, pruned_loss=0.0712, over 5437172.17 frames. ], batch size: 89, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:56:25,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.579e+02 1.038e+03 1.438e+03 1.916e+03 4.530e+03, threshold=2.877e+03, percent-clipped=11.0
+2023-03-09 16:56:44,638 INFO [train.py:968] (0/2) Epoch 19, batch 750, giga_loss[loss=0.2134, simple_loss=0.2859, pruned_loss=0.07046, over 28855.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2942, pruned_loss=0.07165, over 5586378.67 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3394, pruned_loss=0.08808, over 1991840.78 frames. ], giga_tot_loss[loss=0.214, simple_loss=0.2884, pruned_loss=0.06977, over 5496362.88 frames. ], batch size: 199, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:56:52,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4852, 1.6667, 1.5964, 1.3554], device='cuda:0'), covar=tensor([0.3058, 0.2426, 0.1846, 0.2600], device='cuda:0'), in_proj_covar=tensor([0.1905, 0.1823, 0.1751, 0.1894], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 16:57:29,533 INFO [train.py:968] (0/2) Epoch 19, batch 800, giga_loss[loss=0.1973, simple_loss=0.2703, pruned_loss=0.06217, over 28612.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2906, pruned_loss=0.07056, over 5601311.07 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3396, pruned_loss=0.08821, over 2001135.58 frames. ], giga_tot_loss[loss=0.2119, simple_loss=0.2858, pruned_loss=0.06899, over 5538902.59 frames. ], batch size: 85, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 16:57:32,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9960, 2.9284, 2.0040, 1.0839], device='cuda:0'), covar=tensor([0.7175, 0.3343, 0.3796, 0.6420], device='cuda:0'), in_proj_covar=tensor([0.1692, 0.1603, 0.1567, 0.1381], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 16:57:58,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.529e+02 1.157e+03 1.501e+03 2.096e+03 5.337e+03, threshold=3.002e+03, percent-clipped=9.0
+2023-03-09 16:58:16,748 INFO [train.py:968] (0/2) Epoch 19, batch 850, giga_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.0902, over 29041.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2986, pruned_loss=0.07544, over 5606363.93 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3391, pruned_loss=0.08823, over 2125325.50 frames. ], giga_tot_loss[loss=0.2205, simple_loss=0.2936, pruned_loss=0.07373, over 5556266.46 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 16:58:45,348 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=823467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:59:02,484 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-09 16:59:04,608 INFO [train.py:968] (0/2) Epoch 19, batch 900, giga_loss[loss=0.2718, simple_loss=0.3521, pruned_loss=0.09579, over 28863.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3106, pruned_loss=0.08152, over 5626599.28 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3376, pruned_loss=0.08749, over 2235417.14 frames. ], giga_tot_loss[loss=0.2334, simple_loss=0.3063, pruned_loss=0.08024, over 5580179.14 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:59:09,602 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5969, 1.7028, 1.6562, 1.4360], device='cuda:0'), covar=tensor([0.2393, 0.2263, 0.1806, 0.2345], device='cuda:0'), in_proj_covar=tensor([0.1907, 0.1821, 0.1752, 0.1898], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 16:59:17,684 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=823504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:59:32,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.030e+02 1.376e+03 1.649e+03 2.196e+03 6.523e+03, threshold=3.298e+03, percent-clipped=10.0
+2023-03-09 16:59:46,916 INFO [train.py:968] (0/2) Epoch 19, batch 950, giga_loss[loss=0.2813, simple_loss=0.3669, pruned_loss=0.09783, over 29019.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3221, pruned_loss=0.08684, over 5650791.81 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3361, pruned_loss=0.08655, over 2324897.52 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3187, pruned_loss=0.08611, over 5609066.00 frames. ], batch size: 164, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:00:00,735 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7070, 2.6219, 2.0507, 2.1885], device='cuda:0'), covar=tensor([0.0790, 0.0678, 0.0965, 0.1068], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0447, 0.0513, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 17:00:13,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3255, 1.6186, 1.5395, 1.4612], device='cuda:0'), covar=tensor([0.2038, 0.1996, 0.2450, 0.1978], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0743, 0.0700, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 17:00:27,477 INFO [train.py:968] (0/2) Epoch 19, batch 1000, giga_loss[loss=0.2564, simple_loss=0.3439, pruned_loss=0.08441, over 28992.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3305, pruned_loss=0.09061, over 5663461.05 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3367, pruned_loss=0.08703, over 2394468.10 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3275, pruned_loss=0.08992, over 5627274.39 frames. ], batch size: 164, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:00:45,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=823609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:00:52,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.347e+02 1.233e+03 1.455e+03 2.040e+03 4.863e+03, threshold=2.910e+03, percent-clipped=3.0
+2023-03-09 17:01:06,176 INFO [train.py:968] (0/2) Epoch 19, batch 1050, giga_loss[loss=0.2589, simple_loss=0.3527, pruned_loss=0.08256, over 28624.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3345, pruned_loss=0.09108, over 5677144.59 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.338, pruned_loss=0.08783, over 2529264.19 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3314, pruned_loss=0.09032, over 5642219.77 frames. ], batch size: 336, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:01:50,674 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-09 17:01:51,441 INFO [train.py:968] (0/2) Epoch 19, batch 1100, giga_loss[loss=0.2438, simple_loss=0.3268, pruned_loss=0.08036, over 28685.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3371, pruned_loss=0.09144, over 5673832.87 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3394, pruned_loss=0.08876, over 2611138.75 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.334, pruned_loss=0.09052, over 5642259.54 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:02:14,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.824e+02 1.085e+03 1.439e+03 1.985e+03 5.315e+03, threshold=2.879e+03, percent-clipped=9.0
+2023-03-09 17:02:33,688 INFO [train.py:968] (0/2) Epoch 19, batch 1150, giga_loss[loss=0.2973, simple_loss=0.3677, pruned_loss=0.1135, over 28896.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3384, pruned_loss=0.09208, over 5693967.35 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3378, pruned_loss=0.08784, over 2770923.60 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3364, pruned_loss=0.09185, over 5659442.30 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:02:47,035 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=823752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:02:48,843 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=823755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:03:17,310 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=823784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:03:19,541 INFO [train.py:968] (0/2) Epoch 19, batch 1200, giga_loss[loss=0.2798, simple_loss=0.3573, pruned_loss=0.1012, over 28880.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3407, pruned_loss=0.09409, over 5685636.11 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3375, pruned_loss=0.08773, over 2846424.37 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3394, pruned_loss=0.0941, over 5655190.68 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:03:48,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.124e+02 1.131e+03 1.384e+03 1.795e+03 4.581e+03, threshold=2.768e+03, percent-clipped=7.0
+2023-03-09 17:04:04,769 INFO [train.py:968] (0/2) Epoch 19, batch 1250, giga_loss[loss=0.2605, simple_loss=0.3409, pruned_loss=0.09002, over 28993.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3442, pruned_loss=0.09632, over 5693475.63 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3382, pruned_loss=0.08794, over 2907311.59 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.343, pruned_loss=0.09639, over 5665251.69 frames. ], batch size: 136, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:04:10,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=823842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:04:20,188 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=823853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:04:42,083 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=823879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:04:51,859 INFO [train.py:968] (0/2) Epoch 19, batch 1300, giga_loss[loss=0.2787, simple_loss=0.3628, pruned_loss=0.09726, over 28684.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3479, pruned_loss=0.09837, over 5682561.07 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3383, pruned_loss=0.08797, over 2950133.29 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.347, pruned_loss=0.09856, over 5659097.70 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:04:56,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4377, 1.6956, 1.3743, 1.5817], device='cuda:0'), covar=tensor([0.2649, 0.2622, 0.2918, 0.2242], device='cuda:0'), in_proj_covar=tensor([0.1461, 0.1061, 0.1295, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 17:05:15,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.155e+02 1.264e+03 1.607e+03 2.237e+03 6.202e+03, threshold=3.215e+03, percent-clipped=15.0
+2023-03-09 17:05:23,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9669, 5.6722, 5.3696, 3.2208], device='cuda:0'), covar=tensor([0.0399, 0.0590, 0.0752, 0.1600], device='cuda:0'), in_proj_covar=tensor([0.1157, 0.1080, 0.0922, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 17:05:29,816 INFO [train.py:968] (0/2) Epoch 19, batch 1350, giga_loss[loss=0.2796, simple_loss=0.3586, pruned_loss=0.1003, over 28713.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3495, pruned_loss=0.09824, over 5688264.53 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3392, pruned_loss=0.0889, over 3103646.22 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3489, pruned_loss=0.09846, over 5675348.21 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:06:03,595 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=823976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:11,976 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=823985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:13,891 INFO [train.py:968] (0/2) Epoch 19, batch 1400, giga_loss[loss=0.2586, simple_loss=0.3433, pruned_loss=0.08696, over 28536.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3522, pruned_loss=0.09935, over 5682243.77 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3401, pruned_loss=0.08957, over 3147769.09 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3514, pruned_loss=0.09936, over 5679550.49 frames. ], batch size: 336, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:06:14,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=823988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:22,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2951, 1.7584, 1.5875, 1.4465], device='cuda:0'), covar=tensor([0.0804, 0.0319, 0.0316, 0.0874], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 17:06:23,857 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-09 17:06:26,309 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-824000.pt
+2023-03-09 17:06:29,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=824003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:34,999 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=824011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:35,850 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-09 17:06:38,960 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:40,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.215e+02 1.156e+03 1.361e+03 1.763e+03 5.108e+03, threshold=2.723e+03, percent-clipped=1.0
+2023-03-09 17:06:42,373 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:45,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:57,834 INFO [train.py:968] (0/2) Epoch 19, batch 1450, giga_loss[loss=0.2434, simple_loss=0.3345, pruned_loss=0.07614, over 28583.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3517, pruned_loss=0.09814, over 5684186.29 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3398, pruned_loss=0.08927, over 3228105.72 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3516, pruned_loss=0.09856, over 5677303.72 frames. ], batch size: 60, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:07:10,069 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:07:28,525 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-09 17:07:38,247 INFO [train.py:968] (0/2) Epoch 19, batch 1500, giga_loss[loss=0.2573, simple_loss=0.3389, pruned_loss=0.0879, over 28001.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3509, pruned_loss=0.09638, over 5694352.80 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3401, pruned_loss=0.08917, over 3310430.23 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.351, pruned_loss=0.09701, over 5690763.88 frames. ], batch size: 412, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:07:44,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5020, 3.2463, 1.4947, 1.5798], device='cuda:0'), covar=tensor([0.0974, 0.0248, 0.0890, 0.1334], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0537, 0.0371, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 17:07:47,401 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.26 vs. limit=5.0
+2023-03-09 17:07:49,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2302, 0.8273, 0.9496, 1.3640], device='cuda:0'), covar=tensor([0.0837, 0.0388, 0.0368, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 17:08:03,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.773e+02 1.117e+03 1.444e+03 1.975e+03 5.839e+03, threshold=2.888e+03, percent-clipped=12.0
+2023-03-09 17:08:17,001 INFO [train.py:968] (0/2) Epoch 19, batch 1550, giga_loss[loss=0.2274, simple_loss=0.3189, pruned_loss=0.06795, over 28517.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3476, pruned_loss=0.09341, over 5701918.06 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3394, pruned_loss=0.08868, over 3412549.20 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3484, pruned_loss=0.0944, over 5691316.17 frames. ], batch size: 60, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:09:00,410 INFO [train.py:968] (0/2) Epoch 19, batch 1600, giga_loss[loss=0.2742, simple_loss=0.3474, pruned_loss=0.1005, over 28843.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3468, pruned_loss=0.09303, over 5707280.85 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3392, pruned_loss=0.08844, over 3461683.47 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3477, pruned_loss=0.09401, over 5695699.36 frames. ], batch size: 199, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:09:10,668 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=824200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:09:29,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.303e+02 1.127e+03 1.378e+03 1.857e+03 4.297e+03, threshold=2.755e+03, percent-clipped=5.0
+2023-03-09 17:09:35,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:09:43,165 INFO [train.py:968] (0/2) Epoch 19, batch 1650, giga_loss[loss=0.3641, simple_loss=0.3871, pruned_loss=0.1706, over 26550.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3498, pruned_loss=0.09754, over 5705479.40 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3401, pruned_loss=0.089, over 3535196.46 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3503, pruned_loss=0.09823, over 5698990.93 frames. ], batch size: 555, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:10:07,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3787, 4.3818, 1.5267, 1.7806], device='cuda:0'), covar=tensor([0.1271, 0.0364, 0.1011, 0.1554], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0538, 0.0370, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 17:10:35,668 INFO [train.py:968] (0/2) Epoch 19, batch 1700, giga_loss[loss=0.2674, simple_loss=0.3437, pruned_loss=0.09559, over 28920.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3535, pruned_loss=0.1025, over 5698422.89 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3401, pruned_loss=0.089, over 3535196.46 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3539, pruned_loss=0.1031, over 5693372.82 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:11:04,742 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.208e+02 1.242e+03 1.567e+03 1.997e+03 3.586e+03, threshold=3.135e+03, percent-clipped=8.0
+2023-03-09 17:11:20,297 INFO [train.py:968] (0/2) Epoch 19, batch 1750, giga_loss[loss=0.2555, simple_loss=0.3353, pruned_loss=0.0879, over 28789.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3518, pruned_loss=0.1028, over 5691877.21 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3398, pruned_loss=0.08878, over 3570725.23 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3525, pruned_loss=0.1036, over 5684786.65 frames. ], batch size: 199, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:11:31,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:11:49,591 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:11:52,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:11:54,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:12:01,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:12:02,183 INFO [train.py:968] (0/2) Epoch 19, batch 1800, giga_loss[loss=0.272, simple_loss=0.3453, pruned_loss=0.09933, over 28612.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.349, pruned_loss=0.1011, over 5709022.35 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3398, pruned_loss=0.08869, over 3661061.05 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3499, pruned_loss=0.1022, over 5697358.71 frames. ], batch size: 307, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:12:14,666 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:12:20,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3593, 1.5014, 3.4708, 3.2754], device='cuda:0'), covar=tensor([0.1353, 0.2469, 0.0403, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0632, 0.0933, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 17:12:28,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.932e+02 1.263e+03 1.424e+03 1.935e+03 4.559e+03, threshold=2.848e+03, percent-clipped=4.0
+2023-03-09 17:12:43,946 INFO [train.py:968] (0/2) Epoch 19, batch 1850, giga_loss[loss=0.2534, simple_loss=0.3383, pruned_loss=0.08428, over 28988.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3479, pruned_loss=0.1002, over 5713019.24 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3397, pruned_loss=0.08864, over 3716580.61 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3489, pruned_loss=0.1015, over 5699369.12 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:13:11,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3658, 1.9636, 1.6120, 0.5434], device='cuda:0'), covar=tensor([0.4921, 0.2741, 0.3527, 0.6087], device='cuda:0'), in_proj_covar=tensor([0.1688, 0.1597, 0.1564, 0.1382], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 17:13:18,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3746, 4.2332, 4.0488, 1.8809], device='cuda:0'), covar=tensor([0.0555, 0.0678, 0.0654, 0.2026], device='cuda:0'), in_proj_covar=tensor([0.1162, 0.1081, 0.0925, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 17:13:27,413 INFO [train.py:968] (0/2) Epoch 19, batch 1900, giga_loss[loss=0.245, simple_loss=0.3202, pruned_loss=0.08488, over 29032.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3476, pruned_loss=0.09918, over 5719400.99 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3403, pruned_loss=0.08878, over 3745653.07 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3482, pruned_loss=0.1002, over 5709268.61 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:13:33,539 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:13:36,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:13:38,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4974, 1.7477, 1.4405, 1.5384], device='cuda:0'), covar=tensor([0.2503, 0.2470, 0.2647, 0.2390], device='cuda:0'), in_proj_covar=tensor([0.1454, 0.1055, 0.1287, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 17:13:59,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.095e+02 1.292e+03 1.684e+03 2.564e+03 6.005e+03, threshold=3.369e+03, percent-clipped=19.0
+2023-03-09 17:14:00,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:00,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4731, 1.5824, 1.7138, 1.3411], device='cuda:0'), covar=tensor([0.1517, 0.2108, 0.1236, 0.1539], device='cuda:0'), in_proj_covar=tensor([0.0888, 0.0700, 0.0935, 0.0832], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0013], device='cuda:0')
+2023-03-09 17:14:03,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:06,258 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:09,365 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:13,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:18,621 INFO [train.py:968] (0/2) Epoch 19, batch 1950, giga_loss[loss=0.2343, simple_loss=0.3186, pruned_loss=0.07497, over 28962.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.345, pruned_loss=0.09736, over 5697590.64 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3405, pruned_loss=0.0888, over 3787675.70 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3454, pruned_loss=0.09834, over 5687062.01 frames. ], batch size: 106, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:14:31,611 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:39,091 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:50,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:15:03,306 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-09 17:15:05,225 INFO [train.py:968] (0/2) Epoch 19, batch 2000, giga_loss[loss=0.2385, simple_loss=0.3174, pruned_loss=0.07976, over 28854.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3402, pruned_loss=0.09457, over 5697473.94 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3412, pruned_loss=0.08903, over 3829265.08 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3401, pruned_loss=0.09533, over 5685846.68 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:15:08,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4766, 1.6858, 1.6575, 1.5551], device='cuda:0'), covar=tensor([0.2025, 0.2238, 0.2397, 0.2019], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0744, 0.0701, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 17:15:11,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=824593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:15:37,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.404e+02 9.677e+02 1.156e+03 1.746e+03 7.133e+03, threshold=2.313e+03, percent-clipped=3.0
+2023-03-09 17:15:53,185 INFO [train.py:968] (0/2) Epoch 19, batch 2050, giga_loss[loss=0.2424, simple_loss=0.3142, pruned_loss=0.08534, over 28522.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3349, pruned_loss=0.09217, over 5688014.41 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3411, pruned_loss=0.08899, over 3860430.16 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3349, pruned_loss=0.09288, over 5675677.66 frames. ], batch size: 336, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:16:17,190 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=824665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:16:21,748 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.88 vs. limit=2.0
+2023-03-09 17:16:39,109 INFO [train.py:968] (0/2) Epoch 19, batch 2100, giga_loss[loss=0.2427, simple_loss=0.323, pruned_loss=0.08115, over 28796.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3309, pruned_loss=0.09022, over 5688264.83 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3413, pruned_loss=0.08928, over 3926733.99 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3306, pruned_loss=0.0907, over 5675512.46 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:17:05,971 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:17:07,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:17:08,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.672e+02 1.025e+03 1.443e+03 2.140e+03 8.589e+03, threshold=2.886e+03, percent-clipped=19.0
+2023-03-09 17:17:19,628 INFO [train.py:968] (0/2) Epoch 19, batch 2150, giga_loss[loss=0.2572, simple_loss=0.3456, pruned_loss=0.08441, over 28990.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3312, pruned_loss=0.08958, over 5700961.78 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3411, pruned_loss=0.0891, over 4012927.55 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3306, pruned_loss=0.09013, over 5682808.23 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:17:29,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:17:35,262 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 17:17:59,237 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-09 17:18:00,047 INFO [train.py:968] (0/2) Epoch 19, batch 2200, giga_loss[loss=0.2337, simple_loss=0.317, pruned_loss=0.07519, over 28795.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3319, pruned_loss=0.08948, over 5707566.84 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3415, pruned_loss=0.0892, over 4067938.30 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3309, pruned_loss=0.08988, over 5688364.48 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:18:28,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.999e+02 1.101e+03 1.330e+03 1.728e+03 4.224e+03, threshold=2.660e+03, percent-clipped=8.0
+2023-03-09 17:18:44,321 INFO [train.py:968] (0/2) Epoch 19, batch 2250, giga_loss[loss=0.2336, simple_loss=0.3143, pruned_loss=0.07646, over 28957.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3318, pruned_loss=0.08958, over 5704583.03 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3418, pruned_loss=0.08912, over 4113210.04 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3306, pruned_loss=0.08997, over 5684884.07 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:19:23,277 INFO [train.py:968] (0/2) Epoch 19, batch 2300, giga_loss[loss=0.223, simple_loss=0.3031, pruned_loss=0.07144, over 28848.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3289, pruned_loss=0.08825, over 5715098.52 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3417, pruned_loss=0.08895, over 4157530.11 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3278, pruned_loss=0.08866, over 5695092.14 frames. ], batch size: 199, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:19:51,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3274, 1.3729, 1.2566, 1.4822], device='cuda:0'), covar=tensor([0.0797, 0.0373, 0.0356, 0.0869], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 17:19:54,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.891e+02 1.034e+03 1.359e+03 1.964e+03 6.004e+03, threshold=2.719e+03, percent-clipped=8.0
+2023-03-09 17:20:07,441 INFO [train.py:968] (0/2) Epoch 19, batch 2350, giga_loss[loss=0.2457, simple_loss=0.3169, pruned_loss=0.08729, over 28932.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3265, pruned_loss=0.08742, over 5714717.11 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3418, pruned_loss=0.08907, over 4177835.82 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3253, pruned_loss=0.08765, over 5701265.58 frames. ], batch size: 106, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:20:34,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:20:48,677 INFO [train.py:968] (0/2) Epoch 19, batch 2400, libri_loss[loss=0.237, simple_loss=0.3189, pruned_loss=0.07753, over 29645.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3242, pruned_loss=0.08609, over 5725439.67 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3413, pruned_loss=0.08856, over 4244347.76 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.323, pruned_loss=0.08655, over 5708835.19 frames. ], batch size: 73, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:20:48,919 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0884, 5.8888, 5.6488, 3.3602], device='cuda:0'), covar=tensor([0.0499, 0.0636, 0.0833, 0.1325], device='cuda:0'), in_proj_covar=tensor([0.1156, 0.1073, 0.0916, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 17:21:17,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.771e+02 9.732e+02 1.215e+03 1.525e+03 3.128e+03, threshold=2.429e+03, percent-clipped=2.0
+2023-03-09 17:21:30,607 INFO [train.py:968] (0/2) Epoch 19, batch 2450, libri_loss[loss=0.2571, simple_loss=0.3584, pruned_loss=0.07787, over 29667.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3213, pruned_loss=0.08446, over 5732337.35 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3417, pruned_loss=0.08844, over 4285361.39 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3196, pruned_loss=0.08483, over 5715098.14 frames. ], batch size: 88, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:21:32,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=825040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:22:08,257 INFO [train.py:968] (0/2) Epoch 19, batch 2500, giga_loss[loss=0.2306, simple_loss=0.2993, pruned_loss=0.08099, over 28742.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3201, pruned_loss=0.08381, over 5738954.84 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3427, pruned_loss=0.08871, over 4325156.40 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3176, pruned_loss=0.08384, over 5721746.93 frames. ], batch size: 99, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:22:28,585 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=825111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:22:30,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=825114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:22:37,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.329e+02 1.034e+03 1.273e+03 1.682e+03 5.838e+03, threshold=2.546e+03, percent-clipped=10.0
+2023-03-09 17:22:50,247 INFO [train.py:968] (0/2) Epoch 19, batch 2550, libri_loss[loss=0.2795, simple_loss=0.3766, pruned_loss=0.09123, over 29233.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3186, pruned_loss=0.08322, over 5732011.68 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3434, pruned_loss=0.08897, over 4348112.01 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3157, pruned_loss=0.08298, over 5716493.42 frames. ], batch size: 94, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:22:54,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=825143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:23:20,479 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=825175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:23:26,821 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=825183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:23:29,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=825186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:23:30,697 INFO [train.py:968] (0/2) Epoch 19, batch 2600, giga_loss[loss=0.2507, simple_loss=0.3184, pruned_loss=0.09147, over 28675.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3185, pruned_loss=0.08345, over 5730716.25 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3443, pruned_loss=0.08932, over 4385290.94 frames. ], giga_tot_loss[loss=0.2403, simple_loss=0.3149, pruned_loss=0.08285, over 5715518.28 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:23:51,706 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=825215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:23:57,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.368e+02 1.035e+03 1.331e+03 2.109e+03 1.178e+04, threshold=2.662e+03, percent-clipped=16.0
+2023-03-09 17:24:10,350 INFO [train.py:968] (0/2) Epoch 19, batch 2650, giga_loss[loss=0.2394, simple_loss=0.3139, pruned_loss=0.08243, over 28740.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3176, pruned_loss=0.08285, over 5732961.20 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3452, pruned_loss=0.0897, over 4420804.69 frames. ], giga_tot_loss[loss=0.2388, simple_loss=0.3136, pruned_loss=0.082, over 5718914.19 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:24:51,051 INFO [train.py:968] (0/2) Epoch 19, batch 2700, giga_loss[loss=0.2148, simple_loss=0.291, pruned_loss=0.06936, over 28888.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3172, pruned_loss=0.08297, over 5725538.74 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3457, pruned_loss=0.08984, over 4435239.31 frames. ], giga_tot_loss[loss=0.2388, simple_loss=0.3134, pruned_loss=0.08213, over 5713304.30 frames. ], batch size: 112, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:25:23,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.087e+02 1.008e+03 1.307e+03 1.736e+03 4.744e+03, threshold=2.613e+03, percent-clipped=10.0
+2023-03-09 17:25:34,909 INFO [train.py:968] (0/2) Epoch 19, batch 2750, giga_loss[loss=0.3167, simple_loss=0.3859, pruned_loss=0.1238, over 28905.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3214, pruned_loss=0.08543, over 5725893.03 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3455, pruned_loss=0.08954, over 4470064.98 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.318, pruned_loss=0.08485, over 5713456.99 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:26:19,253 INFO [train.py:968] (0/2) Epoch 19, batch 2800, giga_loss[loss=0.3373, simple_loss=0.3977, pruned_loss=0.1385, over 28620.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3271, pruned_loss=0.08899, over 5709513.65 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3455, pruned_loss=0.08941, over 4523533.63 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3235, pruned_loss=0.08857, over 5700445.87 frames. ], batch size: 336, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:26:52,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.721e+02 1.213e+03 1.546e+03 1.940e+03 9.026e+03, threshold=3.092e+03, percent-clipped=13.0
+2023-03-09 17:27:07,848 INFO [train.py:968] (0/2) Epoch 19, batch 2850, giga_loss[loss=0.3139, simple_loss=0.3805, pruned_loss=0.1237, over 28797.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3355, pruned_loss=0.09499, over 5701161.87 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3454, pruned_loss=0.08949, over 4543168.12 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3325, pruned_loss=0.09464, over 5692280.59 frames. ], batch size: 243, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:27:24,688 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-09 17:27:52,862 INFO [train.py:968] (0/2) Epoch 19, batch 2900, giga_loss[loss=0.2606, simple_loss=0.3442, pruned_loss=0.08856, over 28944.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3404, pruned_loss=0.09704, over 5692388.11 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3451, pruned_loss=0.08929, over 4573767.95 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3381, pruned_loss=0.09705, over 5683404.06 frames. ], batch size: 164, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:28:30,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.476e+02 1.186e+03 1.421e+03 1.824e+03 8.430e+03, threshold=2.841e+03, percent-clipped=5.0
+2023-03-09 17:28:42,646 INFO [train.py:968] (0/2) Epoch 19, batch 2950, giga_loss[loss=0.324, simple_loss=0.3909, pruned_loss=0.1286, over 28729.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3458, pruned_loss=0.09957, over 5674241.86 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3451, pruned_loss=0.08919, over 4603174.71 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3439, pruned_loss=0.0999, over 5664965.83 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:28:53,852 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=825550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:29:28,824 INFO [train.py:968] (0/2) Epoch 19, batch 3000, giga_loss[loss=0.3068, simple_loss=0.3819, pruned_loss=0.1159, over 28641.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3494, pruned_loss=0.1008, over 5685483.17 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3444, pruned_loss=0.08902, over 4632111.23 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3485, pruned_loss=0.1015, over 5681291.54 frames. ], batch size: 307, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:29:28,830 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 17:29:37,563 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2112, simple_loss=0.3186, pruned_loss=0.05195, over 944034.00 frames. 
+2023-03-09 17:29:37,564 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 17:29:38,461 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=825589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:30:01,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4248, 3.3520, 1.5536, 1.4826], device='cuda:0'), covar=tensor([0.0921, 0.0380, 0.0880, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0540, 0.0370, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 17:30:05,361 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 17:30:09,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.031e+02 1.210e+03 1.659e+03 2.474e+03 1.025e+04, threshold=3.318e+03, percent-clipped=18.0
+2023-03-09 17:30:21,779 INFO [train.py:968] (0/2) Epoch 19, batch 3050, giga_loss[loss=0.225, simple_loss=0.3163, pruned_loss=0.06683, over 29102.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3533, pruned_loss=0.1036, over 5666870.56 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3448, pruned_loss=0.08937, over 4640506.56 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3523, pruned_loss=0.104, over 5669515.67 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:31:02,210 INFO [train.py:968] (0/2) Epoch 19, batch 3100, giga_loss[loss=0.2349, simple_loss=0.3182, pruned_loss=0.07576, over 28941.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3487, pruned_loss=0.09986, over 5683166.67 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3447, pruned_loss=0.08945, over 4682928.66 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3483, pruned_loss=0.1006, over 5679344.36 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:31:08,788 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=825693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:31:10,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=825696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:31:28,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.53 vs. limit=5.0
+2023-03-09 17:31:33,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.020e+02 1.134e+03 1.436e+03 1.949e+03 6.810e+03, threshold=2.872e+03, percent-clipped=2.0
+2023-03-09 17:31:33,929 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=825725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:31:47,859 INFO [train.py:968] (0/2) Epoch 19, batch 3150, giga_loss[loss=0.2838, simple_loss=0.3571, pruned_loss=0.1052, over 28885.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3461, pruned_loss=0.09791, over 5679300.49 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3446, pruned_loss=0.08955, over 4706845.08 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3459, pruned_loss=0.09857, over 5672599.01 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:32:04,946 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-09 17:32:31,703 INFO [train.py:968] (0/2) Epoch 19, batch 3200, giga_loss[loss=0.3136, simple_loss=0.371, pruned_loss=0.1281, over 23679.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3459, pruned_loss=0.09734, over 5670848.55 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.344, pruned_loss=0.08938, over 4726978.77 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3462, pruned_loss=0.09818, over 5668912.06 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:32:52,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7290, 1.8938, 1.3541, 1.3996], device='cuda:0'), covar=tensor([0.0924, 0.0586, 0.1025, 0.1126], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0447, 0.0515, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 17:32:55,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3451, 1.4700, 1.4458, 1.2369], device='cuda:0'), covar=tensor([0.2843, 0.2642, 0.1853, 0.2435], device='cuda:0'), in_proj_covar=tensor([0.1905, 0.1826, 0.1771, 0.1912], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 17:33:01,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.710e+02 1.203e+03 1.465e+03 2.053e+03 4.531e+03, threshold=2.931e+03, percent-clipped=6.0
+2023-03-09 17:33:16,276 INFO [train.py:968] (0/2) Epoch 19, batch 3250, giga_loss[loss=0.2556, simple_loss=0.3404, pruned_loss=0.08538, over 28838.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3483, pruned_loss=0.09875, over 5675231.57 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.344, pruned_loss=0.08953, over 4754035.36 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3486, pruned_loss=0.09952, over 5670250.66 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:33:44,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4385, 1.8653, 1.4476, 1.3582], device='cuda:0'), covar=tensor([0.2549, 0.2538, 0.2916, 0.2372], device='cuda:0'), in_proj_covar=tensor([0.1457, 0.1058, 0.1291, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 17:33:55,319 INFO [train.py:968] (0/2) Epoch 19, batch 3300, giga_loss[loss=0.3214, simple_loss=0.3875, pruned_loss=0.1277, over 28578.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3496, pruned_loss=0.09933, over 5691416.86 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.08967, over 4793534.75 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.35, pruned_loss=0.1002, over 5680196.97 frames. ], batch size: 307, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:34:16,082 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-09 17:34:30,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.042e+02 1.394e+03 1.703e+03 2.454e+03 6.148e+03, threshold=3.406e+03, percent-clipped=14.0
+2023-03-09 17:34:42,060 INFO [train.py:968] (0/2) Epoch 19, batch 3350, giga_loss[loss=0.2984, simple_loss=0.3739, pruned_loss=0.1115, over 28688.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3538, pruned_loss=0.1029, over 5688007.10 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3445, pruned_loss=0.08996, over 4815130.63 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.354, pruned_loss=0.1036, over 5675541.51 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:34:57,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9221, 2.1943, 1.7619, 2.2724], device='cuda:0'), covar=tensor([0.2633, 0.2662, 0.2977, 0.2317], device='cuda:0'), in_proj_covar=tensor([0.1461, 0.1062, 0.1293, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 17:35:02,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=825964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:35:04,456 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 17:35:20,038 INFO [train.py:968] (0/2) Epoch 19, batch 3400, giga_loss[loss=0.2796, simple_loss=0.3539, pruned_loss=0.1026, over 28895.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3551, pruned_loss=0.104, over 5704091.58 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3449, pruned_loss=0.09025, over 4852783.01 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3552, pruned_loss=0.1047, over 5687354.41 frames. ], batch size: 112, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:35:31,099 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-826000.pt
+2023-03-09 17:35:44,517 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:35:52,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.624e+02 1.307e+03 1.681e+03 2.553e+03 6.901e+03, threshold=3.361e+03, percent-clipped=13.0
+2023-03-09 17:35:55,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:36:04,972 INFO [train.py:968] (0/2) Epoch 19, batch 3450, giga_loss[loss=0.3491, simple_loss=0.3977, pruned_loss=0.1502, over 28044.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3555, pruned_loss=0.1049, over 5690056.19 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.345, pruned_loss=0.09021, over 4884229.16 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.356, pruned_loss=0.106, over 5678040.45 frames. ], batch size: 412, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:36:35,309 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2434, 1.8834, 1.3784, 0.3624], device='cuda:0'), covar=tensor([0.4121, 0.2600, 0.4217, 0.5881], device='cuda:0'), in_proj_covar=tensor([0.1689, 0.1594, 0.1563, 0.1382], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 17:36:47,784 INFO [train.py:968] (0/2) Epoch 19, batch 3500, giga_loss[loss=0.2612, simple_loss=0.3402, pruned_loss=0.09112, over 29062.00 frames. ], tot_loss[loss=0.283, simple_loss=0.356, pruned_loss=0.105, over 5679282.83 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3454, pruned_loss=0.09041, over 4895276.08 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3562, pruned_loss=0.106, over 5674986.08 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:37:02,870 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:37:05,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=826110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:37:17,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.084e+02 1.151e+03 1.436e+03 1.876e+03 4.324e+03, threshold=2.872e+03, percent-clipped=3.0
+2023-03-09 17:37:29,039 INFO [train.py:968] (0/2) Epoch 19, batch 3550, giga_loss[loss=0.2629, simple_loss=0.3488, pruned_loss=0.08847, over 29043.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3554, pruned_loss=0.1036, over 5690745.79 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3449, pruned_loss=0.09017, over 4910164.69 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.356, pruned_loss=0.1048, over 5684645.95 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:37:29,845 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:37:52,347 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:37:54,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0611, 1.2320, 3.3283, 2.9669], device='cuda:0'), covar=tensor([0.1700, 0.2723, 0.0488, 0.0972], device='cuda:0'), in_proj_covar=tensor([0.0730, 0.0627, 0.0924, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 17:38:13,585 INFO [train.py:968] (0/2) Epoch 19, batch 3600, giga_loss[loss=0.2703, simple_loss=0.3521, pruned_loss=0.09428, over 28580.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3547, pruned_loss=0.1021, over 5692401.67 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3449, pruned_loss=0.09003, over 4931749.31 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3555, pruned_loss=0.1035, over 5685379.64 frames. ], batch size: 60, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:38:45,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.336e+02 1.113e+03 1.465e+03 1.865e+03 3.929e+03, threshold=2.930e+03, percent-clipped=11.0
+2023-03-09 17:38:54,275 INFO [train.py:968] (0/2) Epoch 19, batch 3650, giga_loss[loss=0.2447, simple_loss=0.3239, pruned_loss=0.08275, over 28617.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3529, pruned_loss=0.1003, over 5695649.32 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3453, pruned_loss=0.09024, over 4946051.12 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3534, pruned_loss=0.1014, over 5690484.05 frames. ], batch size: 85, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:39:32,010 INFO [train.py:968] (0/2) Epoch 19, batch 3700, giga_loss[loss=0.2737, simple_loss=0.3446, pruned_loss=0.1014, over 28580.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3515, pruned_loss=0.09994, over 5694658.07 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3455, pruned_loss=0.09043, over 4992266.34 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.352, pruned_loss=0.1012, over 5681481.44 frames. ], batch size: 307, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:40:04,475 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.072e+02 1.063e+03 1.412e+03 1.971e+03 5.477e+03, threshold=2.825e+03, percent-clipped=7.0
+2023-03-09 17:40:12,648 INFO [train.py:968] (0/2) Epoch 19, batch 3750, giga_loss[loss=0.3037, simple_loss=0.3735, pruned_loss=0.1169, over 28964.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3486, pruned_loss=0.09811, over 5707789.33 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3455, pruned_loss=0.09038, over 5001124.77 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.349, pruned_loss=0.09917, over 5695898.09 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:40:15,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-09 17:40:23,063 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:40:36,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826368.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 17:40:51,706 INFO [train.py:968] (0/2) Epoch 19, batch 3800, giga_loss[loss=0.26, simple_loss=0.337, pruned_loss=0.0915, over 28756.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3459, pruned_loss=0.09683, over 5709665.89 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3445, pruned_loss=0.08981, over 5028144.75 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3472, pruned_loss=0.09838, over 5697805.24 frames. ], batch size: 284, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:40:53,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=826390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:41:08,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=826404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:41:22,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6176, 4.4247, 4.2040, 2.2001], device='cuda:0'), covar=tensor([0.0568, 0.0748, 0.0811, 0.1924], device='cuda:0'), in_proj_covar=tensor([0.1163, 0.1080, 0.0922, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 17:41:24,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.663e+02 1.101e+03 1.279e+03 1.618e+03 3.868e+03, threshold=2.557e+03, percent-clipped=3.0
+2023-03-09 17:41:37,497 INFO [train.py:968] (0/2) Epoch 19, batch 3850, giga_loss[loss=0.3127, simple_loss=0.3828, pruned_loss=0.1213, over 28792.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3482, pruned_loss=0.09874, over 5703937.24 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3444, pruned_loss=0.08984, over 5035252.04 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3492, pruned_loss=0.1, over 5694284.01 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:41:39,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2498, 1.5476, 1.5423, 1.1506], device='cuda:0'), covar=tensor([0.1578, 0.2478, 0.1318, 0.1583], device='cuda:0'), in_proj_covar=tensor([0.0887, 0.0701, 0.0934, 0.0829], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 17:41:41,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7887, 1.8913, 1.2984, 1.3966], device='cuda:0'), covar=tensor([0.0936, 0.0627, 0.1085, 0.1178], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0442, 0.0512, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 17:41:55,871 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-09 17:42:14,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4764, 2.1251, 1.6872, 0.6140], device='cuda:0'), covar=tensor([0.5263, 0.2463, 0.3675, 0.5847], device='cuda:0'), in_proj_covar=tensor([0.1683, 0.1577, 0.1556, 0.1377], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 17:42:16,039 INFO [train.py:968] (0/2) Epoch 19, batch 3900, giga_loss[loss=0.2727, simple_loss=0.3525, pruned_loss=0.0964, over 28842.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3482, pruned_loss=0.09818, over 5706939.08 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3443, pruned_loss=0.08979, over 5052761.37 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3492, pruned_loss=0.09938, over 5695586.59 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:42:48,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.632e+02 1.052e+03 1.307e+03 1.744e+03 5.066e+03, threshold=2.613e+03, percent-clipped=11.0
+2023-03-09 17:42:55,337 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:42:56,289 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 17:42:57,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=826536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:42:58,422 INFO [train.py:968] (0/2) Epoch 19, batch 3950, giga_loss[loss=0.2576, simple_loss=0.3392, pruned_loss=0.08795, over 28601.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3479, pruned_loss=0.0975, over 5715381.02 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3439, pruned_loss=0.08952, over 5082984.96 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3492, pruned_loss=0.09899, over 5702244.41 frames. ], batch size: 336, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:42:58,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=826538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:43:05,756 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:43:07,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=826550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:43:20,451 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:43:34,055 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:43:39,900 INFO [train.py:968] (0/2) Epoch 19, batch 4000, giga_loss[loss=0.3132, simple_loss=0.3765, pruned_loss=0.125, over 28738.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3477, pruned_loss=0.09739, over 5708008.84 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3444, pruned_loss=0.08991, over 5103501.87 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3485, pruned_loss=0.09853, over 5698503.05 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:43:47,161 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5712, 1.7420, 1.8306, 1.4026], device='cuda:0'), covar=tensor([0.1740, 0.2353, 0.1393, 0.1698], device='cuda:0'), in_proj_covar=tensor([0.0887, 0.0703, 0.0934, 0.0830], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 17:43:57,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1836, 1.2767, 3.6944, 3.1148], device='cuda:0'), covar=tensor([0.1658, 0.2774, 0.0422, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0728, 0.0626, 0.0921, 0.0868], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 17:44:10,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.321e+02 1.009e+03 1.230e+03 1.678e+03 7.061e+03, threshold=2.459e+03, percent-clipped=7.0
+2023-03-09 17:44:19,523 INFO [train.py:968] (0/2) Epoch 19, batch 4050, libri_loss[loss=0.2562, simple_loss=0.3428, pruned_loss=0.08478, over 29768.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3467, pruned_loss=0.09693, over 5703601.64 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3447, pruned_loss=0.09014, over 5109756.52 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3472, pruned_loss=0.09779, over 5700253.38 frames. ], batch size: 87, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:44:40,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:44:54,300 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:44:56,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=826684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:44:59,827 INFO [train.py:968] (0/2) Epoch 19, batch 4100, giga_loss[loss=0.2368, simple_loss=0.3178, pruned_loss=0.07788, over 28946.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3443, pruned_loss=0.09584, over 5690733.25 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3453, pruned_loss=0.09062, over 5110466.87 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3441, pruned_loss=0.09627, over 5703773.35 frames. ], batch size: 164, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:45:18,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:45:23,275 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:45:28,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=826725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:45:29,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.406e+02 1.139e+03 1.358e+03 1.929e+03 5.423e+03, threshold=2.717e+03, percent-clipped=13.0
+2023-03-09 17:45:37,501 INFO [train.py:968] (0/2) Epoch 19, batch 4150, giga_loss[loss=0.2428, simple_loss=0.3234, pruned_loss=0.08111, over 28989.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3411, pruned_loss=0.0945, over 5702422.23 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3451, pruned_loss=0.09059, over 5126313.68 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3411, pruned_loss=0.09495, over 5709281.44 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:45:42,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=826743.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:45:45,818 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:46:16,449 INFO [train.py:968] (0/2) Epoch 19, batch 4200, giga_loss[loss=0.2464, simple_loss=0.3277, pruned_loss=0.08257, over 29087.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3401, pruned_loss=0.09406, over 5705892.14 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3457, pruned_loss=0.09082, over 5153160.50 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3395, pruned_loss=0.09436, over 5705358.91 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:46:48,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.517e+02 1.118e+03 1.396e+03 1.758e+03 3.773e+03, threshold=2.792e+03, percent-clipped=5.0
+2023-03-09 17:46:57,871 INFO [train.py:968] (0/2) Epoch 19, batch 4250, giga_loss[loss=0.2492, simple_loss=0.3286, pruned_loss=0.08488, over 28856.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3396, pruned_loss=0.0941, over 5704359.39 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3458, pruned_loss=0.09095, over 5157244.46 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3389, pruned_loss=0.09429, over 5710297.85 frames. ], batch size: 199, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:47:25,669 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:47:27,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=826871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:47:33,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9560, 3.7751, 3.6523, 1.7303], device='cuda:0'), covar=tensor([0.0715, 0.0864, 0.0895, 0.1937], device='cuda:0'), in_proj_covar=tensor([0.1159, 0.1080, 0.0920, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 17:47:42,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:47:43,647 INFO [train.py:968] (0/2) Epoch 19, batch 4300, giga_loss[loss=0.2831, simple_loss=0.3559, pruned_loss=0.1052, over 28715.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.339, pruned_loss=0.09443, over 5705364.49 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.346, pruned_loss=0.09102, over 5171518.70 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3381, pruned_loss=0.09457, over 5706890.85 frames. ], batch size: 284, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:47:44,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=826889.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 17:47:49,530 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-09 17:47:55,699 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:48:08,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826918.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:48:15,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.694e+02 1.158e+03 1.498e+03 1.890e+03 6.483e+03, threshold=2.996e+03, percent-clipped=7.0
+2023-03-09 17:48:23,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4139, 2.0522, 1.6809, 1.6147], device='cuda:0'), covar=tensor([0.0719, 0.0269, 0.0305, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 17:48:24,449 INFO [train.py:968] (0/2) Epoch 19, batch 4350, giga_loss[loss=0.2446, simple_loss=0.3201, pruned_loss=0.08454, over 28634.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3369, pruned_loss=0.09391, over 5707616.25 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3461, pruned_loss=0.09101, over 5180711.48 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3361, pruned_loss=0.09409, over 5707720.58 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:49:02,353 INFO [train.py:968] (0/2) Epoch 19, batch 4400, giga_loss[loss=0.2776, simple_loss=0.347, pruned_loss=0.1041, over 28359.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3348, pruned_loss=0.09286, over 5713321.38 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3464, pruned_loss=0.09137, over 5207381.21 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3334, pruned_loss=0.09279, over 5707813.23 frames. ], batch size: 368, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:49:09,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2792, 2.2409, 2.1448, 2.0189], device='cuda:0'), covar=tensor([0.1732, 0.2360, 0.2110, 0.2114], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0742, 0.0701, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 17:49:10,278 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:49:33,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.328e+02 1.039e+03 1.435e+03 2.012e+03 6.579e+03, threshold=2.870e+03, percent-clipped=10.0
+2023-03-09 17:49:42,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:49:42,382 INFO [train.py:968] (0/2) Epoch 19, batch 4450, giga_loss[loss=0.2655, simple_loss=0.3455, pruned_loss=0.09276, over 28711.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3335, pruned_loss=0.09185, over 5711442.86 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3469, pruned_loss=0.09168, over 5223971.15 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3318, pruned_loss=0.09157, over 5703440.37 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:50:25,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5360, 3.3831, 1.5889, 1.5499], device='cuda:0'), covar=tensor([0.0894, 0.0274, 0.0945, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0540, 0.0371, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 17:50:27,028 INFO [train.py:968] (0/2) Epoch 19, batch 4500, giga_loss[loss=0.241, simple_loss=0.3153, pruned_loss=0.08333, over 28809.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3356, pruned_loss=0.09294, over 5707439.66 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.347, pruned_loss=0.09173, over 5230647.81 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.334, pruned_loss=0.09268, over 5699548.89 frames. ], batch size: 112, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:50:31,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:50:43,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:50:57,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:51:00,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.324e+02 9.793e+02 1.157e+03 1.462e+03 3.104e+03, threshold=2.314e+03, percent-clipped=2.0
+2023-03-09 17:51:11,171 INFO [train.py:968] (0/2) Epoch 19, batch 4550, giga_loss[loss=0.238, simple_loss=0.3188, pruned_loss=0.07862, over 28975.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3373, pruned_loss=0.09309, over 5718728.48 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3474, pruned_loss=0.09193, over 5246807.13 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3355, pruned_loss=0.09276, over 5708720.55 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:51:22,764 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827154.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:51:31,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3754, 1.7269, 1.6265, 1.4971], device='cuda:0'), covar=tensor([0.1905, 0.1918, 0.2123, 0.2104], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0741, 0.0701, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 17:51:33,222 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-09 17:51:45,720 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:51:49,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:51:53,873 INFO [train.py:968] (0/2) Epoch 19, batch 4600, giga_loss[loss=0.3292, simple_loss=0.3794, pruned_loss=0.1395, over 26569.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3381, pruned_loss=0.09274, over 5718922.71 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3469, pruned_loss=0.09169, over 5254868.69 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3369, pruned_loss=0.09268, over 5710305.93 frames. ], batch size: 555, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:52:16,876 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:52:29,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.304e+02 1.008e+03 1.249e+03 1.616e+03 4.814e+03, threshold=2.499e+03, percent-clipped=9.0
+2023-03-09 17:52:36,449 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:52:37,775 INFO [train.py:968] (0/2) Epoch 19, batch 4650, giga_loss[loss=0.2773, simple_loss=0.3514, pruned_loss=0.1016, over 28788.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3402, pruned_loss=0.09346, over 5702029.75 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3473, pruned_loss=0.09204, over 5261415.02 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3387, pruned_loss=0.09315, over 5701490.45 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:52:38,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:53:04,387 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:53:07,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:53:07,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:53:12,097 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-09 17:53:13,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2423, 2.1988, 1.7172, 1.7708], device='cuda:0'), covar=tensor([0.0877, 0.0787, 0.1045, 0.1244], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0440, 0.0512, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 17:53:25,203 INFO [train.py:968] (0/2) Epoch 19, batch 4700, libri_loss[loss=0.3167, simple_loss=0.3852, pruned_loss=0.1241, over 20744.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3403, pruned_loss=0.09328, over 5688115.89 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3476, pruned_loss=0.09229, over 5262266.49 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3387, pruned_loss=0.09284, over 5693782.14 frames. ], batch size: 187, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:53:33,819 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:53:51,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3745, 3.6467, 1.5147, 1.5563], device='cuda:0'), covar=tensor([0.0950, 0.0332, 0.0927, 0.1285], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0540, 0.0372, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 17:53:57,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.805e+02 1.173e+03 1.477e+03 2.373e+03 5.970e+03, threshold=2.954e+03, percent-clipped=22.0
+2023-03-09 17:54:01,068 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827332.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:54:04,575 INFO [train.py:968] (0/2) Epoch 19, batch 4750, giga_loss[loss=0.2798, simple_loss=0.3395, pruned_loss=0.1101, over 28685.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3412, pruned_loss=0.09394, over 5694921.23 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.348, pruned_loss=0.09257, over 5274266.41 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3395, pruned_loss=0.09337, over 5696112.83 frames. ], batch size: 99, lr: 1.70e-03, grad_scale: 2.0
+2023-03-09 17:54:33,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:54:47,795 INFO [train.py:968] (0/2) Epoch 19, batch 4800, giga_loss[loss=0.2357, simple_loss=0.3093, pruned_loss=0.08104, over 28822.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3409, pruned_loss=0.09381, over 5707327.07 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3479, pruned_loss=0.09263, over 5283676.44 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3395, pruned_loss=0.09332, over 5705644.72 frames. ], batch size: 112, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:54:48,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7107, 1.8734, 1.3258, 1.4680], device='cuda:0'), covar=tensor([0.0889, 0.0575, 0.1007, 0.1093], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0440, 0.0511, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 17:54:55,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6887, 1.7138, 1.4197, 1.2961], device='cuda:0'), covar=tensor([0.0850, 0.0599, 0.0943, 0.1136], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0440, 0.0511, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 17:54:58,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6265, 1.7650, 1.4319, 1.8032], device='cuda:0'), covar=tensor([0.2645, 0.2847, 0.3230, 0.2520], device='cuda:0'), in_proj_covar=tensor([0.1458, 0.1059, 0.1291, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 17:55:07,192 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:55:25,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.689e+02 1.300e+03 1.812e+03 2.411e+03 6.186e+03, threshold=3.625e+03, percent-clipped=15.0
+2023-03-09 17:55:34,607 INFO [train.py:968] (0/2) Epoch 19, batch 4850, giga_loss[loss=0.2811, simple_loss=0.3582, pruned_loss=0.102, over 28747.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3424, pruned_loss=0.09535, over 5704360.43 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3478, pruned_loss=0.09252, over 5286613.06 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3414, pruned_loss=0.09506, over 5702329.69 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:56:06,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:56:16,131 INFO [train.py:968] (0/2) Epoch 19, batch 4900, giga_loss[loss=0.2645, simple_loss=0.3421, pruned_loss=0.0934, over 28772.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3452, pruned_loss=0.09674, over 5699105.75 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3484, pruned_loss=0.09285, over 5291645.87 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3438, pruned_loss=0.09632, over 5703901.96 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:56:17,518 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:56:21,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3809, 1.2917, 1.1080, 1.4384], device='cuda:0'), covar=tensor([0.0710, 0.0349, 0.0358, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 17:56:40,318 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:56:42,163 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3321, 1.5532, 1.5013, 1.3980], device='cuda:0'), covar=tensor([0.1873, 0.1932, 0.2308, 0.2010], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0744, 0.0704, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 17:56:42,167 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:56:52,049 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.035e+02 1.311e+03 1.615e+03 2.041e+03 1.085e+04, threshold=3.229e+03, percent-clipped=6.0
+2023-03-09 17:56:52,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827529.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 17:56:59,871 INFO [train.py:968] (0/2) Epoch 19, batch 4950, giga_loss[loss=0.2751, simple_loss=0.3571, pruned_loss=0.09658, over 28929.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3482, pruned_loss=0.09839, over 5702377.49 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3482, pruned_loss=0.09272, over 5293220.64 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3473, pruned_loss=0.09819, over 5706538.97 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:57:02,107 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:57:07,751 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:57:38,319 INFO [train.py:968] (0/2) Epoch 19, batch 5000, giga_loss[loss=0.2705, simple_loss=0.3509, pruned_loss=0.09504, over 28908.00 frames. ], tot_loss[loss=0.273, simple_loss=0.349, pruned_loss=0.09847, over 5711285.97 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3487, pruned_loss=0.09296, over 5316920.85 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3478, pruned_loss=0.09835, over 5707722.41 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:57:41,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9210, 1.3200, 1.1353, 0.1916], device='cuda:0'), covar=tensor([0.4137, 0.2950, 0.4314, 0.6417], device='cuda:0'), in_proj_covar=tensor([0.1695, 0.1597, 0.1568, 0.1385], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 17:57:57,313 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2678, 1.7644, 1.3018, 1.3259], device='cuda:0'), covar=tensor([0.2523, 0.2408, 0.2995, 0.2283], device='cuda:0'), in_proj_covar=tensor([0.1446, 0.1051, 0.1281, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 17:58:08,732 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:58:11,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:58:14,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.331e+02 1.349e+03 1.695e+03 2.559e+03 8.214e+03, threshold=3.390e+03, percent-clipped=13.0
+2023-03-09 17:58:20,534 INFO [train.py:968] (0/2) Epoch 19, batch 5050, giga_loss[loss=0.29, simple_loss=0.3605, pruned_loss=0.1098, over 28870.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3505, pruned_loss=0.09964, over 5705461.57 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3488, pruned_loss=0.09292, over 5322422.60 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3495, pruned_loss=0.09964, over 5701188.21 frames. ], batch size: 112, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:58:35,142 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:58:48,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8308, 1.0802, 2.8510, 2.8734], device='cuda:0'), covar=tensor([0.1699, 0.2691, 0.0569, 0.1051], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0630, 0.0928, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 17:58:49,120 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827672.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 17:58:50,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827675.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:59:01,594 INFO [train.py:968] (0/2) Epoch 19, batch 5100, giga_loss[loss=0.2383, simple_loss=0.332, pruned_loss=0.07231, over 28984.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3501, pruned_loss=0.09916, over 5708817.85 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3493, pruned_loss=0.0931, over 5331941.86 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3489, pruned_loss=0.09912, over 5703757.54 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:59:15,474 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 17:59:18,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827707.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:59:36,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.908e+02 1.146e+03 1.417e+03 1.926e+03 6.902e+03, threshold=2.834e+03, percent-clipped=4.0
+2023-03-09 17:59:43,045 INFO [train.py:968] (0/2) Epoch 19, batch 5150, giga_loss[loss=0.2634, simple_loss=0.3427, pruned_loss=0.09199, over 29006.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3498, pruned_loss=0.09937, over 5709482.81 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.35, pruned_loss=0.09356, over 5339525.70 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3483, pruned_loss=0.09907, over 5705122.38 frames. ], batch size: 164, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:00:24,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:00:25,772 INFO [train.py:968] (0/2) Epoch 19, batch 5200, giga_loss[loss=0.2428, simple_loss=0.3141, pruned_loss=0.08575, over 28871.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3465, pruned_loss=0.09778, over 5704654.50 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3504, pruned_loss=0.09377, over 5343512.43 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3449, pruned_loss=0.09743, over 5702829.93 frames. ], batch size: 112, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:00:59,160 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2072, 1.6863, 1.5387, 1.3940], device='cuda:0'), covar=tensor([0.0805, 0.0300, 0.0305, 0.1091], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 18:01:01,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.972e+02 1.135e+03 1.393e+03 1.964e+03 5.638e+03, threshold=2.786e+03, percent-clipped=12.0
+2023-03-09 18:01:07,714 INFO [train.py:968] (0/2) Epoch 19, batch 5250, giga_loss[loss=0.2622, simple_loss=0.339, pruned_loss=0.09264, over 28878.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3441, pruned_loss=0.09645, over 5712869.77 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3506, pruned_loss=0.09395, over 5362647.43 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3425, pruned_loss=0.0961, over 5705511.80 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:01:09,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8818, 1.1613, 1.0877, 0.8461], device='cuda:0'), covar=tensor([0.2365, 0.2534, 0.1463, 0.2248], device='cuda:0'), in_proj_covar=tensor([0.1905, 0.1839, 0.1779, 0.1911], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 18:01:17,021 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827850.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:01:19,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827853.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:01:28,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:01:42,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827882.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:01:47,303 INFO [train.py:968] (0/2) Epoch 19, batch 5300, giga_loss[loss=0.2734, simple_loss=0.3596, pruned_loss=0.09357, over 29002.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.346, pruned_loss=0.09679, over 5713736.09 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3508, pruned_loss=0.09414, over 5380594.79 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3443, pruned_loss=0.09643, over 5702591.35 frames. ], batch size: 164, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:02:12,392 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:02:23,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.271e+02 1.209e+03 1.468e+03 1.929e+03 7.006e+03, threshold=2.937e+03, percent-clipped=11.0
+2023-03-09 18:02:23,929 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:02:25,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:02:29,617 INFO [train.py:968] (0/2) Epoch 19, batch 5350, giga_loss[loss=0.2546, simple_loss=0.3305, pruned_loss=0.08939, over 28928.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.348, pruned_loss=0.09665, over 5716829.82 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3511, pruned_loss=0.09439, over 5393214.89 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3463, pruned_loss=0.09623, over 5706618.56 frames. ], batch size: 106, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:02:42,194 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:02:49,371 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:03:12,510 INFO [train.py:968] (0/2) Epoch 19, batch 5400, giga_loss[loss=0.2411, simple_loss=0.3074, pruned_loss=0.08746, over 28541.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.347, pruned_loss=0.09603, over 5716501.17 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3512, pruned_loss=0.09449, over 5400144.28 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3455, pruned_loss=0.09567, over 5709290.65 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:03:20,410 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-828000.pt
+2023-03-09 18:03:27,315 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:03:30,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=828011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:03:45,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.741e+02 1.118e+03 1.400e+03 1.878e+03 3.524e+03, threshold=2.800e+03, percent-clipped=4.0
+2023-03-09 18:03:52,990 INFO [train.py:968] (0/2) Epoch 19, batch 5450, giga_loss[loss=0.2819, simple_loss=0.3496, pruned_loss=0.1071, over 28700.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3444, pruned_loss=0.09584, over 5722695.35 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3509, pruned_loss=0.09429, over 5405067.82 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3435, pruned_loss=0.09572, over 5715487.68 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:03:56,258 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=828040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:04:11,182 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:04:13,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=828062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:04:34,433 INFO [train.py:968] (0/2) Epoch 19, batch 5500, libri_loss[loss=0.2596, simple_loss=0.3494, pruned_loss=0.08496, over 29758.00 frames. ], tot_loss[loss=0.269, simple_loss=0.344, pruned_loss=0.097, over 5730572.03 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3512, pruned_loss=0.09449, over 5422770.24 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3428, pruned_loss=0.09686, over 5720059.17 frames. ], batch size: 87, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:04:37,582 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=828091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:04:38,238 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:05:06,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.915e+02 1.232e+03 1.519e+03 1.854e+03 4.260e+03, threshold=3.038e+03, percent-clipped=8.0
+2023-03-09 18:05:14,195 INFO [train.py:968] (0/2) Epoch 19, batch 5550, giga_loss[loss=0.2454, simple_loss=0.312, pruned_loss=0.08937, over 28939.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3412, pruned_loss=0.09668, over 5728234.08 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3504, pruned_loss=0.0943, over 5433930.92 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3405, pruned_loss=0.09684, over 5721596.57 frames. ], batch size: 106, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:05:42,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6475, 4.4432, 4.2140, 2.0384], device='cuda:0'), covar=tensor([0.0479, 0.0687, 0.0662, 0.2195], device='cuda:0'), in_proj_covar=tensor([0.1164, 0.1081, 0.0925, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 18:05:54,643 INFO [train.py:968] (0/2) Epoch 19, batch 5600, giga_loss[loss=0.2994, simple_loss=0.3677, pruned_loss=0.1155, over 28518.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3405, pruned_loss=0.09651, over 5731409.27 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3501, pruned_loss=0.09414, over 5448148.91 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3399, pruned_loss=0.09686, over 5722351.14 frames. ], batch size: 336, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:05:59,667 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828192.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:06:32,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.617e+02 1.166e+03 1.404e+03 1.780e+03 3.713e+03, threshold=2.807e+03, percent-clipped=5.0
+2023-03-09 18:06:39,179 INFO [train.py:968] (0/2) Epoch 19, batch 5650, libri_loss[loss=0.256, simple_loss=0.335, pruned_loss=0.0885, over 29579.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3383, pruned_loss=0.09522, over 5726176.58 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3494, pruned_loss=0.09364, over 5466696.80 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3381, pruned_loss=0.09604, over 5712747.67 frames. ], batch size: 74, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:07:19,104 INFO [train.py:968] (0/2) Epoch 19, batch 5700, libri_loss[loss=0.274, simple_loss=0.3616, pruned_loss=0.09321, over 29263.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3355, pruned_loss=0.0941, over 5725442.57 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3495, pruned_loss=0.09361, over 5482124.31 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3348, pruned_loss=0.09484, over 5708647.86 frames. ], batch size: 94, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:07:33,965 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-09 18:07:52,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=828328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:07:54,595 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.477e+02 1.151e+03 1.444e+03 2.057e+03 4.079e+03, threshold=2.888e+03, percent-clipped=8.0
+2023-03-09 18:07:59,201 INFO [train.py:968] (0/2) Epoch 19, batch 5750, giga_loss[loss=0.2355, simple_loss=0.3113, pruned_loss=0.0798, over 28890.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3306, pruned_loss=0.09147, over 5725376.23 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3493, pruned_loss=0.09347, over 5488239.90 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.33, pruned_loss=0.09214, over 5709948.44 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:08:40,164 INFO [train.py:968] (0/2) Epoch 19, batch 5800, giga_loss[loss=0.2312, simple_loss=0.2974, pruned_loss=0.08252, over 28593.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3295, pruned_loss=0.09096, over 5728630.86 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3489, pruned_loss=0.0934, over 5502601.80 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3285, pruned_loss=0.09146, over 5711552.91 frames. ], batch size: 85, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:08:47,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6197, 1.9474, 1.5716, 1.4657], device='cuda:0'), covar=tensor([0.2550, 0.2486, 0.3004, 0.2324], device='cuda:0'), in_proj_covar=tensor([0.1450, 0.1053, 0.1284, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 18:09:14,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.415e+02 1.178e+03 1.502e+03 1.800e+03 4.959e+03, threshold=3.004e+03, percent-clipped=6.0
+2023-03-09 18:09:19,820 INFO [train.py:968] (0/2) Epoch 19, batch 5850, giga_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09488, over 28954.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3324, pruned_loss=0.09267, over 5721933.71 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3488, pruned_loss=0.09331, over 5506508.48 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3316, pruned_loss=0.09314, over 5706952.88 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:09:39,313 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:09:42,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=828467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:09:45,708 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:09:48,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=828474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:10:01,889 INFO [train.py:968] (0/2) Epoch 19, batch 5900, giga_loss[loss=0.2838, simple_loss=0.3516, pruned_loss=0.108, over 28923.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3357, pruned_loss=0.09415, over 5716707.04 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3483, pruned_loss=0.09307, over 5515382.93 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3351, pruned_loss=0.09471, over 5701745.12 frames. ], batch size: 106, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:10:15,174 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=828503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:10:17,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2510, 1.2166, 3.7360, 3.1568], device='cuda:0'), covar=tensor([0.1616, 0.2806, 0.0425, 0.1063], device='cuda:0'), in_proj_covar=tensor([0.0736, 0.0634, 0.0936, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 18:10:36,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.511e+02 1.137e+03 1.437e+03 1.899e+03 3.376e+03, threshold=2.874e+03, percent-clipped=4.0
+2023-03-09 18:10:41,735 INFO [train.py:968] (0/2) Epoch 19, batch 5950, libri_loss[loss=0.3178, simple_loss=0.3812, pruned_loss=0.1272, over 29790.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3388, pruned_loss=0.09489, over 5719653.57 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3487, pruned_loss=0.0935, over 5520567.02 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3376, pruned_loss=0.095, over 5710667.85 frames. ], batch size: 87, lr: 1.70e-03, grad_scale: 2.0
+2023-03-09 18:11:08,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=828567.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:11:15,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3033, 1.1875, 1.1145, 1.4310], device='cuda:0'), covar=tensor([0.0755, 0.0362, 0.0363, 0.0874], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 18:11:19,128 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:11:27,689 INFO [train.py:968] (0/2) Epoch 19, batch 6000, giga_loss[loss=0.2617, simple_loss=0.3387, pruned_loss=0.09232, over 29143.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3424, pruned_loss=0.09623, over 5710488.56 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3487, pruned_loss=0.09355, over 5521182.26 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3413, pruned_loss=0.0963, over 5704706.58 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:11:27,693 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 18:11:35,446 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4907, 1.7014, 1.3237, 1.3574], device='cuda:0'), covar=tensor([0.0908, 0.0492, 0.0984, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0443, 0.0511, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 18:11:36,637 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2128, simple_loss=0.3196, pruned_loss=0.05298, over 944034.00 frames. 
+2023-03-09 18:11:36,637 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 18:11:39,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5979, 1.7170, 1.4174, 1.7652], device='cuda:0'), covar=tensor([0.2535, 0.2668, 0.2977, 0.2316], device='cuda:0'), in_proj_covar=tensor([0.1456, 0.1055, 0.1286, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 18:11:47,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1355, 1.4605, 1.4309, 1.0350], device='cuda:0'), covar=tensor([0.1571, 0.2308, 0.1341, 0.1563], device='cuda:0'), in_proj_covar=tensor([0.0886, 0.0699, 0.0931, 0.0830], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 18:11:54,002 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:11:57,323 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=828613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:12:12,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.608e+02 1.186e+03 1.450e+03 2.022e+03 9.072e+03, threshold=2.899e+03, percent-clipped=7.0
+2023-03-09 18:12:18,114 INFO [train.py:968] (0/2) Epoch 19, batch 6050, giga_loss[loss=0.2568, simple_loss=0.3375, pruned_loss=0.08808, over 29050.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3454, pruned_loss=0.09817, over 5711435.95 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3487, pruned_loss=0.09355, over 5529236.52 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3445, pruned_loss=0.0983, over 5703763.21 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:12:22,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=828642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:13:04,847 INFO [train.py:968] (0/2) Epoch 19, batch 6100, giga_loss[loss=0.3389, simple_loss=0.3966, pruned_loss=0.1406, over 28728.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3509, pruned_loss=0.1031, over 5703136.96 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3489, pruned_loss=0.09368, over 5534565.06 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.35, pruned_loss=0.1032, over 5695180.04 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:13:09,771 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 18:13:27,929 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828710.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:13:30,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=828713.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:13:46,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.919e+02 1.578e+03 2.227e+03 3.431e+03 8.845e+03, threshold=4.455e+03, percent-clipped=32.0
+2023-03-09 18:13:51,351 INFO [train.py:968] (0/2) Epoch 19, batch 6150, giga_loss[loss=0.3688, simple_loss=0.4139, pruned_loss=0.1619, over 27582.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3559, pruned_loss=0.1071, over 5702639.07 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.09339, over 5544475.91 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3558, pruned_loss=0.1079, over 5691499.76 frames. ], batch size: 472, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:13:55,978 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=828742.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:14:40,293 INFO [train.py:968] (0/2) Epoch 19, batch 6200, giga_loss[loss=0.4105, simple_loss=0.4419, pruned_loss=0.1895, over 26605.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3635, pruned_loss=0.112, over 5701015.58 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3482, pruned_loss=0.09322, over 5550663.38 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3637, pruned_loss=0.1131, over 5690203.94 frames. ], batch size: 555, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:14:56,027 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:15:25,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.761e+03 2.426e+03 3.510e+03 9.169e+03, threshold=4.852e+03, percent-clipped=11.0
+2023-03-09 18:15:31,007 INFO [train.py:968] (0/2) Epoch 19, batch 6250, giga_loss[loss=0.3313, simple_loss=0.4028, pruned_loss=0.1299, over 28945.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3696, pruned_loss=0.1173, over 5703231.99 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3484, pruned_loss=0.0933, over 5558862.56 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3701, pruned_loss=0.1186, over 5690548.37 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:15:31,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=828838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:15:36,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-09 18:15:47,533 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-09 18:15:57,477 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6683, 1.6835, 1.9005, 1.4421], device='cuda:0'), covar=tensor([0.1647, 0.2341, 0.1311, 0.1616], device='cuda:0'), in_proj_covar=tensor([0.0878, 0.0694, 0.0925, 0.0825], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 18:16:02,976 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:16:11,388 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-09 18:16:14,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4747, 1.5497, 1.1867, 1.1103], device='cuda:0'), covar=tensor([0.0730, 0.0404, 0.0905, 0.1018], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0446, 0.0513, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 18:16:17,296 INFO [train.py:968] (0/2) Epoch 19, batch 6300, giga_loss[loss=0.3307, simple_loss=0.4071, pruned_loss=0.1272, over 28943.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3765, pruned_loss=0.123, over 5700923.46 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3489, pruned_loss=0.09355, over 5564197.53 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.377, pruned_loss=0.1244, over 5689017.57 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:16:19,974 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:16:23,068 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:16:54,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3179, 2.6075, 1.2918, 1.4222], device='cuda:0'), covar=tensor([0.0956, 0.0411, 0.0884, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0544, 0.0373, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 18:16:58,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.827e+03 2.277e+03 3.048e+03 6.595e+03, threshold=4.554e+03, percent-clipped=11.0
+2023-03-09 18:17:05,670 INFO [train.py:968] (0/2) Epoch 19, batch 6350, giga_loss[loss=0.3581, simple_loss=0.4148, pruned_loss=0.1507, over 29014.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.381, pruned_loss=0.1268, over 5696105.64 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3486, pruned_loss=0.0934, over 5572050.33 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3824, pruned_loss=0.1288, over 5682257.75 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:17:17,155 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-09 18:17:21,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=828953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:17:50,289 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:17:53,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=828984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:17:56,840 INFO [train.py:968] (0/2) Epoch 19, batch 6400, giga_loss[loss=0.387, simple_loss=0.4211, pruned_loss=0.1765, over 27487.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3821, pruned_loss=0.1286, over 5681673.88 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3483, pruned_loss=0.09314, over 5581974.50 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3844, pruned_loss=0.1314, over 5664794.15 frames. ], batch size: 472, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:18:04,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-09 18:18:23,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:18:48,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.915e+02 1.771e+03 2.504e+03 3.585e+03 9.305e+03, threshold=5.008e+03, percent-clipped=18.0
+2023-03-09 18:18:51,577 INFO [train.py:968] (0/2) Epoch 19, batch 6450, giga_loss[loss=0.3643, simple_loss=0.4214, pruned_loss=0.1537, over 28916.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.386, pruned_loss=0.1327, over 5667408.73 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3483, pruned_loss=0.09307, over 5576786.28 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3884, pruned_loss=0.1357, over 5661233.33 frames. ], batch size: 199, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:19:36,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8174, 3.6315, 3.4940, 1.8247], device='cuda:0'), covar=tensor([0.0773, 0.0894, 0.0829, 0.2078], device='cuda:0'), in_proj_covar=tensor([0.1178, 0.1090, 0.0935, 0.0710], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 18:19:40,742 INFO [train.py:968] (0/2) Epoch 19, batch 6500, giga_loss[loss=0.3406, simple_loss=0.395, pruned_loss=0.1431, over 28555.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3873, pruned_loss=0.1347, over 5665807.15 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3488, pruned_loss=0.09329, over 5588388.41 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3903, pruned_loss=0.1384, over 5653535.49 frames. ], batch size: 60, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:19:52,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:19:56,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:19:58,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5133, 1.8044, 1.4801, 1.5050], device='cuda:0'), covar=tensor([0.2032, 0.1948, 0.2068, 0.1794], device='cuda:0'), in_proj_covar=tensor([0.1447, 0.1054, 0.1284, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 18:20:19,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1832, 1.4081, 0.9095, 1.0459], device='cuda:0'), covar=tensor([0.1245, 0.0699, 0.1635, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0448, 0.0516, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 18:20:26,344 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:20:31,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.895e+02 1.638e+03 2.225e+03 3.254e+03 8.859e+03, threshold=4.451e+03, percent-clipped=7.0
+2023-03-09 18:20:35,664 INFO [train.py:968] (0/2) Epoch 19, batch 6550, giga_loss[loss=0.3068, simple_loss=0.3847, pruned_loss=0.1144, over 28500.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3907, pruned_loss=0.1376, over 5646304.26 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3489, pruned_loss=0.09344, over 5583702.63 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3934, pruned_loss=0.141, over 5642145.93 frames. ], batch size: 65, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:21:14,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:21:24,165 INFO [train.py:968] (0/2) Epoch 19, batch 6600, libri_loss[loss=0.2792, simple_loss=0.3577, pruned_loss=0.1003, over 29534.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3883, pruned_loss=0.1367, over 5651054.60 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3481, pruned_loss=0.09296, over 5595482.11 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3927, pruned_loss=0.1415, over 5639078.23 frames. ], batch size: 83, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:22:11,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.811e+03 2.446e+03 3.244e+03 1.044e+04, threshold=4.891e+03, percent-clipped=7.0
+2023-03-09 18:22:15,765 INFO [train.py:968] (0/2) Epoch 19, batch 6650, giga_loss[loss=0.3049, simple_loss=0.3656, pruned_loss=0.1221, over 28712.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3885, pruned_loss=0.138, over 5652852.57 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3482, pruned_loss=0.09293, over 5599912.50 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3925, pruned_loss=0.1424, over 5640273.96 frames. ], batch size: 78, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:22:19,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2801, 1.2801, 3.9145, 3.2693], device='cuda:0'), covar=tensor([0.1596, 0.2553, 0.0433, 0.0923], device='cuda:0'), in_proj_covar=tensor([0.0739, 0.0634, 0.0938, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 18:22:27,862 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:22:40,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3481, 3.0014, 1.4761, 1.4347], device='cuda:0'), covar=tensor([0.0895, 0.0362, 0.0832, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0546, 0.0372, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 18:22:45,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:22:48,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:22:54,317 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:23:07,547 INFO [train.py:968] (0/2) Epoch 19, batch 6700, giga_loss[loss=0.3184, simple_loss=0.3852, pruned_loss=0.1258, over 28460.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3884, pruned_loss=0.1372, over 5641885.15 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3483, pruned_loss=0.0929, over 5604478.14 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3924, pruned_loss=0.1419, over 5628284.26 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:23:40,545 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:23:43,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:23:50,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.813e+03 2.543e+03 3.401e+03 1.270e+04, threshold=5.087e+03, percent-clipped=13.0
+2023-03-09 18:23:55,189 INFO [train.py:968] (0/2) Epoch 19, batch 6750, giga_loss[loss=0.3354, simple_loss=0.3954, pruned_loss=0.1377, over 28285.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3872, pruned_loss=0.135, over 5654183.60 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3479, pruned_loss=0.09261, over 5613404.39 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.392, pruned_loss=0.1402, over 5636484.36 frames. ], batch size: 368, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:24:08,958 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:24:13,294 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-09 18:24:45,987 INFO [train.py:968] (0/2) Epoch 19, batch 6800, giga_loss[loss=0.3848, simple_loss=0.428, pruned_loss=0.1708, over 28263.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3889, pruned_loss=0.1362, over 5633510.29 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3481, pruned_loss=0.09289, over 5602971.06 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3935, pruned_loss=0.1411, over 5629546.15 frames. ], batch size: 368, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:24:49,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:24:51,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:07,818 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:10,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:12,238 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:12,320 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:15,754 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:23,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:32,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 1.757e+03 2.378e+03 3.231e+03 7.906e+03, threshold=4.755e+03, percent-clipped=7.0
+2023-03-09 18:25:39,432 INFO [train.py:968] (0/2) Epoch 19, batch 6850, giga_loss[loss=0.2845, simple_loss=0.3608, pruned_loss=0.1041, over 28643.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3867, pruned_loss=0.1343, over 5623181.68 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3483, pruned_loss=0.09292, over 5602740.20 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3903, pruned_loss=0.1384, over 5620371.71 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:25:41,693 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:47,301 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:26:14,327 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-09 18:26:20,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-09 18:26:32,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2159, 0.8046, 0.8444, 1.3905], device='cuda:0'), covar=tensor([0.0725, 0.0428, 0.0381, 0.0808], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 18:26:34,077 INFO [train.py:968] (0/2) Epoch 19, batch 6900, giga_loss[loss=0.354, simple_loss=0.3889, pruned_loss=0.1595, over 24108.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3851, pruned_loss=0.1317, over 5632878.91 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3483, pruned_loss=0.09288, over 5605289.18 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3884, pruned_loss=0.1354, over 5628684.25 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:26:35,457 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829489.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:27:19,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.551e+03 2.049e+03 2.759e+03 4.959e+03, threshold=4.098e+03, percent-clipped=1.0
+2023-03-09 18:27:23,923 INFO [train.py:968] (0/2) Epoch 19, batch 6950, giga_loss[loss=0.2705, simple_loss=0.3468, pruned_loss=0.09713, over 29018.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3819, pruned_loss=0.1286, over 5638665.06 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3484, pruned_loss=0.093, over 5604993.60 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.385, pruned_loss=0.132, over 5635796.32 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:27:27,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2800, 1.4764, 1.5002, 1.2375], device='cuda:0'), covar=tensor([0.2292, 0.1986, 0.1387, 0.1884], device='cuda:0'), in_proj_covar=tensor([0.1924, 0.1857, 0.1790, 0.1922], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 18:27:56,926 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0684, 3.9062, 3.7108, 2.0563], device='cuda:0'), covar=tensor([0.0667, 0.0788, 0.0746, 0.1932], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.1099, 0.0940, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 18:28:12,578 INFO [train.py:968] (0/2) Epoch 19, batch 7000, giga_loss[loss=0.3256, simple_loss=0.3851, pruned_loss=0.133, over 28709.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3788, pruned_loss=0.1258, over 5640240.48 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3486, pruned_loss=0.09325, over 5600881.67 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3818, pruned_loss=0.1292, over 5642310.72 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:28:14,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8264, 3.6531, 3.4681, 1.6483], device='cuda:0'), covar=tensor([0.0751, 0.0834, 0.0810, 0.2184], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.1100, 0.0941, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 18:29:01,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.415e+02 1.747e+03 2.413e+03 3.466e+03 1.238e+04, threshold=4.825e+03, percent-clipped=21.0
+2023-03-09 18:29:04,973 INFO [train.py:968] (0/2) Epoch 19, batch 7050, giga_loss[loss=0.2908, simple_loss=0.3647, pruned_loss=0.1085, over 28853.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3767, pruned_loss=0.125, over 5638915.69 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3487, pruned_loss=0.09343, over 5602469.26 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3792, pruned_loss=0.1276, over 5639313.34 frames. ], batch size: 284, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:29:13,343 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:29:17,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:29:33,537 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:29:54,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1810, 1.2438, 3.4416, 3.0447], device='cuda:0'), covar=tensor([0.1630, 0.2740, 0.0498, 0.1365], device='cuda:0'), in_proj_covar=tensor([0.0745, 0.0641, 0.0946, 0.0888], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-09 18:29:56,684 INFO [train.py:968] (0/2) Epoch 19, batch 7100, giga_loss[loss=0.2928, simple_loss=0.3629, pruned_loss=0.1114, over 28435.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.376, pruned_loss=0.1246, over 5631651.92 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3489, pruned_loss=0.09361, over 5597305.86 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3781, pruned_loss=0.1268, over 5637735.81 frames. ], batch size: 78, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:30:10,179 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:30:50,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.931e+02 1.450e+03 1.844e+03 2.452e+03 7.750e+03, threshold=3.688e+03, percent-clipped=2.0
+2023-03-09 18:30:53,119 INFO [train.py:968] (0/2) Epoch 19, batch 7150, libri_loss[loss=0.2784, simple_loss=0.3648, pruned_loss=0.09595, over 29550.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3754, pruned_loss=0.1235, over 5635629.95 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3489, pruned_loss=0.09349, over 5599937.75 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3779, pruned_loss=0.1264, over 5639518.32 frames. ], batch size: 78, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:31:44,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:31:45,079 INFO [train.py:968] (0/2) Epoch 19, batch 7200, giga_loss[loss=0.3665, simple_loss=0.4318, pruned_loss=0.1506, over 28085.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3747, pruned_loss=0.1217, over 5647630.72 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3493, pruned_loss=0.09375, over 5605988.77 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3769, pruned_loss=0.1244, over 5645932.92 frames. ], batch size: 412, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:31:50,803 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:31:55,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:32:26,252 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:32:34,971 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:32:39,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.950e+02 1.535e+03 2.187e+03 2.987e+03 8.965e+03, threshold=4.375e+03, percent-clipped=14.0
+2023-03-09 18:32:42,628 INFO [train.py:968] (0/2) Epoch 19, batch 7250, giga_loss[loss=0.3469, simple_loss=0.4139, pruned_loss=0.1399, over 28861.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3744, pruned_loss=0.1191, over 5651275.35 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3492, pruned_loss=0.09399, over 5607265.62 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.377, pruned_loss=0.1219, over 5649839.03 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:33:05,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829864.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:33:30,249 INFO [train.py:968] (0/2) Epoch 19, batch 7300, giga_loss[loss=0.3008, simple_loss=0.3495, pruned_loss=0.1261, over 23835.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3753, pruned_loss=0.1188, over 5667815.07 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3494, pruned_loss=0.09413, over 5615297.91 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3781, pruned_loss=0.1217, over 5660932.72 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:33:38,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1807, 4.0240, 3.8436, 1.8398], device='cuda:0'), covar=tensor([0.0625, 0.0724, 0.0731, 0.2070], device='cuda:0'), in_proj_covar=tensor([0.1189, 0.1103, 0.0942, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 18:33:39,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2475, 1.5563, 1.5289, 1.3241], device='cuda:0'), covar=tensor([0.1762, 0.1578, 0.2132, 0.1775], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0748, 0.0705, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 18:33:47,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4250, 1.4796, 3.3022, 3.2086], device='cuda:0'), covar=tensor([0.1250, 0.2312, 0.0458, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0743, 0.0641, 0.0944, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-09 18:34:16,781 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:34:18,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:34:19,918 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.531e+03 1.955e+03 2.553e+03 6.807e+03, threshold=3.910e+03, percent-clipped=3.0
+2023-03-09 18:34:23,652 INFO [train.py:968] (0/2) Epoch 19, batch 7350, giga_loss[loss=0.3317, simple_loss=0.3922, pruned_loss=0.1356, over 28598.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3757, pruned_loss=0.1203, over 5661110.68 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.09386, over 5619854.84 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3787, pruned_loss=0.1232, over 5652079.76 frames. ], batch size: 307, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:34:36,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1730, 2.5988, 1.2305, 1.3539], device='cuda:0'), covar=tensor([0.1002, 0.0366, 0.0897, 0.1365], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0547, 0.0374, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 18:34:41,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:34:48,214 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:35:12,111 INFO [train.py:968] (0/2) Epoch 19, batch 7400, libri_loss[loss=0.297, simple_loss=0.3703, pruned_loss=0.1118, over 29745.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3736, pruned_loss=0.1192, over 5674131.91 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3486, pruned_loss=0.09367, over 5629413.21 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3772, pruned_loss=0.1226, over 5659448.36 frames. ], batch size: 87, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:35:22,448 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-830000.pt
+2023-03-09 18:35:31,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830007.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:35:34,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830010.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:35:38,911 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=830015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:35:43,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:35:56,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.908e+02 1.788e+03 2.480e+03 3.371e+03 1.099e+04, threshold=4.960e+03, percent-clipped=17.0
+2023-03-09 18:35:58,309 INFO [train.py:968] (0/2) Epoch 19, batch 7450, giga_loss[loss=0.2708, simple_loss=0.3431, pruned_loss=0.09922, over 28691.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3718, pruned_loss=0.1191, over 5675854.86 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3482, pruned_loss=0.09345, over 5638090.00 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3758, pruned_loss=0.1228, over 5657413.24 frames. ], batch size: 243, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:35:59,207 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830039.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:36:02,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:36:31,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:36:45,243 INFO [train.py:968] (0/2) Epoch 19, batch 7500, giga_loss[loss=0.3287, simple_loss=0.3932, pruned_loss=0.1321, over 28904.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3709, pruned_loss=0.1193, over 5681218.48 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3484, pruned_loss=0.09351, over 5639163.55 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.374, pruned_loss=0.1223, over 5665996.92 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:37:27,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=830128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:37:34,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.276e+02 1.453e+03 1.929e+03 2.730e+03 6.123e+03, threshold=3.859e+03, percent-clipped=4.0
+2023-03-09 18:37:36,798 INFO [train.py:968] (0/2) Epoch 19, batch 7550, giga_loss[loss=0.2805, simple_loss=0.3602, pruned_loss=0.1004, over 28835.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3696, pruned_loss=0.1176, over 5671405.20 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3482, pruned_loss=0.09349, over 5644184.77 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3728, pruned_loss=0.1207, over 5655623.45 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:37:43,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4027, 1.5983, 1.6580, 1.2597], device='cuda:0'), covar=tensor([0.1717, 0.2618, 0.1497, 0.1799], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0698, 0.0925, 0.0825], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 18:37:45,743 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=830147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:01,219 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:03,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:20,986 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:23,061 INFO [train.py:968] (0/2) Epoch 19, batch 7600, giga_loss[loss=0.305, simple_loss=0.3743, pruned_loss=0.1178, over 28649.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3695, pruned_loss=0.1166, over 5669634.74 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.348, pruned_loss=0.0934, over 5640770.67 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3728, pruned_loss=0.1198, over 5660012.60 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:38:23,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:31,931 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:43,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:50,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:50,870 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:52,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:39:03,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2740, 1.4700, 1.5527, 1.3373], device='cuda:0'), covar=tensor([0.1816, 0.1730, 0.2208, 0.1840], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0749, 0.0704, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 18:39:06,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.691e+03 2.061e+03 2.779e+03 7.115e+03, threshold=4.122e+03, percent-clipped=12.0
+2023-03-09 18:39:09,312 INFO [train.py:968] (0/2) Epoch 19, batch 7650, giga_loss[loss=0.2746, simple_loss=0.3534, pruned_loss=0.09789, over 29035.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3693, pruned_loss=0.1165, over 5670216.34 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3481, pruned_loss=0.09354, over 5636026.60 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3724, pruned_loss=0.1195, over 5667209.33 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:39:17,868 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:39:42,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7684, 1.9237, 1.7238, 1.7443], device='cuda:0'), covar=tensor([0.1751, 0.2099, 0.2377, 0.2079], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0748, 0.0705, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 18:39:46,039 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 18:39:57,159 INFO [train.py:968] (0/2) Epoch 19, batch 7700, giga_loss[loss=0.2432, simple_loss=0.3314, pruned_loss=0.07752, over 28870.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3668, pruned_loss=0.1147, over 5678141.55 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3482, pruned_loss=0.09368, over 5640930.11 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3696, pruned_loss=0.1174, over 5672300.30 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:40:37,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:40:46,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.557e+03 2.227e+03 3.138e+03 1.165e+04, threshold=4.455e+03, percent-clipped=17.0
+2023-03-09 18:40:47,228 INFO [train.py:968] (0/2) Epoch 19, batch 7750, giga_loss[loss=0.2762, simple_loss=0.3487, pruned_loss=0.1018, over 28312.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.366, pruned_loss=0.1153, over 5667398.56 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3485, pruned_loss=0.09389, over 5648190.51 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3685, pruned_loss=0.1179, over 5656492.05 frames. ], batch size: 65, lr: 1.70e-03, grad_scale: 2.0
+2023-03-09 18:40:58,780 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:41:00,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:41:27,783 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:41:33,707 INFO [train.py:968] (0/2) Epoch 19, batch 7800, libri_loss[loss=0.2512, simple_loss=0.3386, pruned_loss=0.08187, over 29542.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3642, pruned_loss=0.1147, over 5670968.84 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3481, pruned_loss=0.09366, over 5651888.98 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3671, pruned_loss=0.1176, over 5659401.66 frames. ], batch size: 79, lr: 1.70e-03, grad_scale: 2.0
+2023-03-09 18:41:35,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:42:18,935 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3228, 1.9604, 1.3867, 0.6378], device='cuda:0'), covar=tensor([0.5237, 0.2425, 0.3466, 0.5869], device='cuda:0'), in_proj_covar=tensor([0.1708, 0.1613, 0.1578, 0.1392], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 18:42:24,352 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.721e+02 1.622e+03 2.056e+03 2.699e+03 7.206e+03, threshold=4.111e+03, percent-clipped=4.0
+2023-03-09 18:42:25,113 INFO [train.py:968] (0/2) Epoch 19, batch 7850, giga_loss[loss=0.2553, simple_loss=0.3202, pruned_loss=0.09518, over 28906.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3645, pruned_loss=0.1159, over 5664347.14 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3482, pruned_loss=0.09378, over 5653064.85 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3669, pruned_loss=0.1185, over 5654318.04 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 18:42:55,012 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:42:59,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:10,308 INFO [train.py:968] (0/2) Epoch 19, batch 7900, giga_loss[loss=0.28, simple_loss=0.3521, pruned_loss=0.104, over 29019.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3629, pruned_loss=0.1153, over 5661739.40 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3479, pruned_loss=0.09366, over 5654889.87 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3656, pruned_loss=0.1181, over 5652256.12 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 18:43:27,142 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:27,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:42,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:55,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:58,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:58,550 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.410e+02 1.581e+03 2.268e+03 3.293e+03 6.257e+03, threshold=4.537e+03, percent-clipped=10.0
+2023-03-09 18:43:59,243 INFO [train.py:968] (0/2) Epoch 19, batch 7950, giga_loss[loss=0.3375, simple_loss=0.4066, pruned_loss=0.1342, over 28822.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3628, pruned_loss=0.1155, over 5662047.07 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3479, pruned_loss=0.09364, over 5658571.01 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3652, pruned_loss=0.1181, over 5651438.11 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 18:44:24,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3828, 1.5574, 1.5224, 1.2784], device='cuda:0'), covar=tensor([0.2543, 0.2106, 0.1488, 0.2112], device='cuda:0'), in_proj_covar=tensor([0.1924, 0.1848, 0.1789, 0.1923], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 18:44:26,685 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:44:49,048 INFO [train.py:968] (0/2) Epoch 19, batch 8000, giga_loss[loss=0.2764, simple_loss=0.3484, pruned_loss=0.1022, over 28623.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3649, pruned_loss=0.1166, over 5664617.07 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.348, pruned_loss=0.09372, over 5663518.92 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3671, pruned_loss=0.1192, over 5651764.17 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:44:54,172 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-09 18:45:40,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.312e+02 1.684e+03 2.231e+03 2.928e+03 1.412e+04, threshold=4.463e+03, percent-clipped=9.0
+2023-03-09 18:45:41,000 INFO [train.py:968] (0/2) Epoch 19, batch 8050, giga_loss[loss=0.2906, simple_loss=0.3587, pruned_loss=0.1112, over 28437.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3656, pruned_loss=0.1163, over 5669362.20 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3478, pruned_loss=0.09365, over 5664556.33 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3675, pruned_loss=0.1184, over 5658484.74 frames. ], batch size: 85, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:45:49,213 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:45:50,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:46:05,162 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:46:07,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:46:14,851 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:46:24,816 INFO [train.py:968] (0/2) Epoch 19, batch 8100, giga_loss[loss=0.282, simple_loss=0.3571, pruned_loss=0.1034, over 28740.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3645, pruned_loss=0.1142, over 5688176.28 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.347, pruned_loss=0.09316, over 5674354.74 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3674, pruned_loss=0.1173, over 5671045.24 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:46:35,534 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:47:14,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.489e+03 1.926e+03 2.619e+03 5.892e+03, threshold=3.852e+03, percent-clipped=8.0
+2023-03-09 18:47:15,068 INFO [train.py:968] (0/2) Epoch 19, batch 8150, giga_loss[loss=0.3627, simple_loss=0.4041, pruned_loss=0.1606, over 28295.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3659, pruned_loss=0.1155, over 5686122.92 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3469, pruned_loss=0.09305, over 5677989.82 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3686, pruned_loss=0.1183, over 5669193.34 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:47:54,602 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0447, 2.9551, 1.9915, 1.2491], device='cuda:0'), covar=tensor([0.6615, 0.3358, 0.3575, 0.5810], device='cuda:0'), in_proj_covar=tensor([0.1712, 0.1618, 0.1584, 0.1396], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 18:48:06,137 INFO [train.py:968] (0/2) Epoch 19, batch 8200, giga_loss[loss=0.3725, simple_loss=0.4133, pruned_loss=0.1659, over 28941.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3686, pruned_loss=0.1182, over 5668479.33 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3467, pruned_loss=0.09287, over 5666973.69 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3712, pruned_loss=0.1209, over 5665417.27 frames. ], batch size: 285, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:48:31,778 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-09 18:48:40,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8145, 2.7327, 1.8005, 1.0154], device='cuda:0'), covar=tensor([0.7181, 0.3455, 0.3455, 0.6443], device='cuda:0'), in_proj_covar=tensor([0.1712, 0.1618, 0.1584, 0.1396], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 18:48:40,862 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 18:48:53,309 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5254, 2.7921, 1.5740, 1.6396], device='cuda:0'), covar=tensor([0.0747, 0.0338, 0.0709, 0.1043], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0549, 0.0374, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 18:48:58,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.754e+03 2.291e+03 3.170e+03 1.230e+04, threshold=4.582e+03, percent-clipped=16.0
+2023-03-09 18:48:58,939 INFO [train.py:968] (0/2) Epoch 19, batch 8250, giga_loss[loss=0.2763, simple_loss=0.3504, pruned_loss=0.1011, over 28934.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3716, pruned_loss=0.1223, over 5657220.51 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3471, pruned_loss=0.09306, over 5673737.34 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3739, pruned_loss=0.125, over 5648808.57 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:49:36,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2758, 1.3049, 3.9369, 3.3561], device='cuda:0'), covar=tensor([0.1676, 0.2715, 0.0467, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0744, 0.0639, 0.0943, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-09 18:49:44,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1954, 1.1645, 3.5079, 3.0586], device='cuda:0'), covar=tensor([0.1557, 0.2598, 0.0509, 0.1274], device='cuda:0'), in_proj_covar=tensor([0.0744, 0.0639, 0.0942, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-09 18:49:51,326 INFO [train.py:968] (0/2) Epoch 19, batch 8300, giga_loss[loss=0.3323, simple_loss=0.3855, pruned_loss=0.1395, over 28646.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3719, pruned_loss=0.1234, over 5667593.43 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3473, pruned_loss=0.0932, over 5678265.76 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3741, pruned_loss=0.126, over 5656495.26 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:50:41,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.157e+03 1.929e+03 2.494e+03 3.579e+03 1.067e+04, threshold=4.989e+03, percent-clipped=11.0
+2023-03-09 18:50:41,949 INFO [train.py:968] (0/2) Epoch 19, batch 8350, giga_loss[loss=0.2787, simple_loss=0.3519, pruned_loss=0.1028, over 28850.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3742, pruned_loss=0.1258, over 5663483.65 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3478, pruned_loss=0.09339, over 5681859.42 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.376, pruned_loss=0.1283, over 5650999.05 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:51:25,327 INFO [train.py:968] (0/2) Epoch 19, batch 8400, giga_loss[loss=0.2533, simple_loss=0.3355, pruned_loss=0.08555, over 28810.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3713, pruned_loss=0.1231, over 5672796.12 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3475, pruned_loss=0.09317, over 5686839.85 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3739, pruned_loss=0.1265, over 5657675.75 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 18:51:51,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5768, 1.4180, 4.8636, 3.6445], device='cuda:0'), covar=tensor([0.1698, 0.2826, 0.0387, 0.0794], device='cuda:0'), in_proj_covar=tensor([0.0742, 0.0638, 0.0939, 0.0884], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 18:52:07,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.605e+03 2.277e+03 3.016e+03 1.248e+04, threshold=4.555e+03, percent-clipped=11.0
+2023-03-09 18:52:07,806 INFO [train.py:968] (0/2) Epoch 19, batch 8450, giga_loss[loss=0.2806, simple_loss=0.3536, pruned_loss=0.1038, over 28682.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3708, pruned_loss=0.1219, over 5685062.09 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3477, pruned_loss=0.09334, over 5691346.09 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3732, pruned_loss=0.1251, over 5668872.86 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:52:53,145 INFO [train.py:968] (0/2) Epoch 19, batch 8500, giga_loss[loss=0.3049, simple_loss=0.3751, pruned_loss=0.1174, over 29077.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3689, pruned_loss=0.1192, over 5689016.34 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3471, pruned_loss=0.09309, over 5694372.13 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3719, pruned_loss=0.1224, over 5673298.68 frames. ], batch size: 128, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:53:36,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.122e+03 1.598e+03 2.009e+03 3.234e+03 1.063e+04, threshold=4.018e+03, percent-clipped=10.0
+2023-03-09 18:53:36,978 INFO [train.py:968] (0/2) Epoch 19, batch 8550, giga_loss[loss=0.2637, simple_loss=0.3361, pruned_loss=0.09565, over 28806.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3668, pruned_loss=0.1177, over 5687168.65 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3473, pruned_loss=0.09319, over 5697263.77 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3693, pruned_loss=0.1207, over 5671883.45 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:54:07,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5379, 1.6918, 1.6333, 1.4737], device='cuda:0'), covar=tensor([0.1910, 0.2190, 0.2506, 0.2260], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0742, 0.0700, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 18:54:23,993 INFO [train.py:968] (0/2) Epoch 19, batch 8600, giga_loss[loss=0.2734, simple_loss=0.3447, pruned_loss=0.1011, over 29072.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3658, pruned_loss=0.1175, over 5675973.23 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3483, pruned_loss=0.09374, over 5692425.61 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3675, pruned_loss=0.1201, over 5667511.82 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:55:13,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.655e+03 2.210e+03 3.412e+03 8.614e+03, threshold=4.421e+03, percent-clipped=15.0
+2023-03-09 18:55:13,824 INFO [train.py:968] (0/2) Epoch 19, batch 8650, giga_loss[loss=0.3089, simple_loss=0.3692, pruned_loss=0.1243, over 29060.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.365, pruned_loss=0.1173, over 5674662.79 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3484, pruned_loss=0.09369, over 5696835.54 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3666, pruned_loss=0.1199, over 5663548.04 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:55:27,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5205, 1.8192, 1.8289, 1.5180], device='cuda:0'), covar=tensor([0.2484, 0.2043, 0.2017, 0.2276], device='cuda:0'), in_proj_covar=tensor([0.1903, 0.1836, 0.1768, 0.1901], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 18:56:05,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5770, 4.0268, 1.8048, 1.6577], device='cuda:0'), covar=tensor([0.0945, 0.0383, 0.0841, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0549, 0.0373, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 18:56:06,312 INFO [train.py:968] (0/2) Epoch 19, batch 8700, giga_loss[loss=0.3132, simple_loss=0.3746, pruned_loss=0.126, over 28790.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3679, pruned_loss=0.1191, over 5670937.07 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3485, pruned_loss=0.09379, over 5690538.75 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3694, pruned_loss=0.1214, over 5667814.85 frames. ], batch size: 284, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 18:56:39,865 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-09 18:56:57,406 INFO [train.py:968] (0/2) Epoch 19, batch 8750, giga_loss[loss=0.2987, simple_loss=0.3772, pruned_loss=0.1102, over 28927.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.371, pruned_loss=0.1186, over 5671611.76 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3485, pruned_loss=0.09373, over 5694902.23 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3725, pruned_loss=0.1209, over 5665006.29 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 18:56:58,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.919e+02 1.465e+03 2.069e+03 2.657e+03 1.473e+04, threshold=4.138e+03, percent-clipped=9.0
+2023-03-09 18:57:32,506 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3427, 1.7480, 1.8018, 1.4309], device='cuda:0'), covar=tensor([0.1926, 0.1653, 0.1966, 0.1870], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0738, 0.0697, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 18:57:41,272 INFO [train.py:968] (0/2) Epoch 19, batch 8800, giga_loss[loss=0.2694, simple_loss=0.3506, pruned_loss=0.09409, over 28999.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3712, pruned_loss=0.1172, over 5679085.32 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3476, pruned_loss=0.0933, over 5701124.70 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3742, pruned_loss=0.1205, over 5667179.68 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:58:27,667 INFO [train.py:968] (0/2) Epoch 19, batch 8850, giga_loss[loss=0.3132, simple_loss=0.3786, pruned_loss=0.1239, over 28994.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3739, pruned_loss=0.1192, over 5682501.51 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3477, pruned_loss=0.0933, over 5703435.25 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.377, pruned_loss=0.1227, over 5670139.92 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:58:28,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.499e+02 1.460e+03 1.933e+03 2.683e+03 5.543e+03, threshold=3.866e+03, percent-clipped=6.0
+2023-03-09 18:58:39,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2047, 1.4617, 1.4493, 1.3285], device='cuda:0'), covar=tensor([0.1736, 0.1656, 0.2298, 0.1735], device='cuda:0'), in_proj_covar=tensor([0.0462, 0.0735, 0.0694, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 18:58:54,956 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=831472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:59:13,941 INFO [train.py:968] (0/2) Epoch 19, batch 8900, giga_loss[loss=0.2858, simple_loss=0.3585, pruned_loss=0.1066, over 28598.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3743, pruned_loss=0.1201, over 5689713.29 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3475, pruned_loss=0.09323, over 5706230.27 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3773, pruned_loss=0.1232, over 5677310.28 frames. ], batch size: 78, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:59:57,137 INFO [train.py:968] (0/2) Epoch 19, batch 8950, giga_loss[loss=0.2801, simple_loss=0.3478, pruned_loss=0.1062, over 28854.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3749, pruned_loss=0.1215, over 5696040.58 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3475, pruned_loss=0.09319, over 5713821.47 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3786, pruned_loss=0.1253, over 5678388.00 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:59:57,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.846e+02 1.528e+03 2.223e+03 3.180e+03 9.992e+03, threshold=4.446e+03, percent-clipped=12.0
+2023-03-09 19:00:27,672 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2704, 1.3786, 3.7600, 3.3137], device='cuda:0'), covar=tensor([0.1648, 0.2757, 0.0461, 0.1104], device='cuda:0'), in_proj_covar=tensor([0.0742, 0.0638, 0.0942, 0.0888], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-09 19:00:45,988 INFO [train.py:968] (0/2) Epoch 19, batch 9000, giga_loss[loss=0.302, simple_loss=0.371, pruned_loss=0.1165, over 28346.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3744, pruned_loss=0.1219, over 5697323.19 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3473, pruned_loss=0.09307, over 5715553.29 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.378, pruned_loss=0.1256, over 5681444.59 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:00:45,994 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 19:00:54,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1048, 1.4793, 1.5118, 1.3256], device='cuda:0'), covar=tensor([0.1772, 0.1598, 0.2122, 0.1734], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0746, 0.0703, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 19:00:54,701 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2091, simple_loss=0.3164, pruned_loss=0.05091, over 944034.00 frames. 
+2023-03-09 19:00:54,702 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 19:01:38,823 INFO [train.py:968] (0/2) Epoch 19, batch 9050, giga_loss[loss=0.2866, simple_loss=0.3533, pruned_loss=0.1099, over 28900.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3719, pruned_loss=0.1208, over 5698090.10 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3472, pruned_loss=0.09295, over 5723262.91 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3759, pruned_loss=0.125, over 5677272.88 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:01:41,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.809e+03 2.335e+03 3.251e+03 1.253e+04, threshold=4.671e+03, percent-clipped=15.0
+2023-03-09 19:02:27,801 INFO [train.py:968] (0/2) Epoch 19, batch 9100, giga_loss[loss=0.3199, simple_loss=0.3767, pruned_loss=0.1316, over 28307.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3702, pruned_loss=0.1206, over 5681934.91 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3475, pruned_loss=0.09315, over 5714804.76 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3733, pruned_loss=0.124, over 5672147.57 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:02:41,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2153, 1.2002, 1.1773, 1.5183], device='cuda:0'), covar=tensor([0.0786, 0.0368, 0.0335, 0.0840], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0217], device='cuda:0'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 19:03:20,971 INFO [train.py:968] (0/2) Epoch 19, batch 9150, giga_loss[loss=0.2868, simple_loss=0.3562, pruned_loss=0.1087, over 28884.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3698, pruned_loss=0.1205, over 5686276.56 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3475, pruned_loss=0.09316, over 5716874.00 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3726, pruned_loss=0.1236, over 5676292.18 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:03:23,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.732e+03 2.423e+03 3.466e+03 1.429e+04, threshold=4.846e+03, percent-clipped=15.0
+2023-03-09 19:03:34,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4759, 1.6687, 1.6016, 1.5775], device='cuda:0'), covar=tensor([0.1745, 0.1796, 0.2241, 0.1758], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0748, 0.0705, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 19:03:49,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7747, 1.7712, 1.5636, 1.7917], device='cuda:0'), covar=tensor([0.2592, 0.2807, 0.3111, 0.2563], device='cuda:0'), in_proj_covar=tensor([0.1458, 0.1056, 0.1293, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 19:04:10,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4909, 1.6611, 1.6657, 1.5257], device='cuda:0'), covar=tensor([0.1701, 0.1916, 0.2074, 0.1968], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0748, 0.0705, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 19:04:11,235 INFO [train.py:968] (0/2) Epoch 19, batch 9200, giga_loss[loss=0.3076, simple_loss=0.3739, pruned_loss=0.1207, over 28679.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3707, pruned_loss=0.1219, over 5680559.94 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3475, pruned_loss=0.09327, over 5719489.37 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3732, pruned_loss=0.1246, over 5669802.96 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:05:01,083 INFO [train.py:968] (0/2) Epoch 19, batch 9250, giga_loss[loss=0.2612, simple_loss=0.3297, pruned_loss=0.09635, over 28102.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3699, pruned_loss=0.1221, over 5681037.90 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3476, pruned_loss=0.09334, over 5719505.31 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3722, pruned_loss=0.1247, over 5671853.34 frames. ], batch size: 77, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:05:02,660 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.519e+03 2.050e+03 2.805e+03 6.667e+03, threshold=4.100e+03, percent-clipped=3.0
+2023-03-09 19:05:08,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=831847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:05:48,098 INFO [train.py:968] (0/2) Epoch 19, batch 9300, giga_loss[loss=0.3691, simple_loss=0.4094, pruned_loss=0.1644, over 27937.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3693, pruned_loss=0.1217, over 5687568.05 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3479, pruned_loss=0.09347, over 5725601.45 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3716, pruned_loss=0.1245, over 5673751.40 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:06:37,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6793, 1.6694, 1.3325, 1.2260], device='cuda:0'), covar=tensor([0.0710, 0.0451, 0.0818, 0.1314], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0447, 0.0512, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 19:06:37,949 INFO [train.py:968] (0/2) Epoch 19, batch 9350, giga_loss[loss=0.3389, simple_loss=0.3748, pruned_loss=0.1515, over 23544.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3718, pruned_loss=0.1228, over 5672089.02 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3484, pruned_loss=0.09388, over 5719269.98 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3736, pruned_loss=0.1254, over 5665341.04 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:06:39,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.523e+03 1.981e+03 3.062e+03 1.005e+04, threshold=3.963e+03, percent-clipped=9.0
+2023-03-09 19:07:25,086 INFO [train.py:968] (0/2) Epoch 19, batch 9400, giga_loss[loss=0.3696, simple_loss=0.4119, pruned_loss=0.1637, over 28658.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3737, pruned_loss=0.1237, over 5675512.85 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3483, pruned_loss=0.09384, over 5719123.91 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3756, pruned_loss=0.1263, over 5669855.21 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:07:27,799 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=831990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:07:30,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=831993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:07:38,672 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3700, 1.8796, 1.3504, 0.6527], device='cuda:0'), covar=tensor([0.5284, 0.2908, 0.2958, 0.5946], device='cuda:0'), in_proj_covar=tensor([0.1719, 0.1622, 0.1586, 0.1404], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 19:07:39,565 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-832000.pt
+2023-03-09 19:07:58,758 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=832022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:08:14,163 INFO [train.py:968] (0/2) Epoch 19, batch 9450, giga_loss[loss=0.3268, simple_loss=0.3992, pruned_loss=0.1272, over 28887.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3723, pruned_loss=0.123, over 5672838.98 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3483, pruned_loss=0.09384, over 5722357.33 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3745, pruned_loss=0.1257, over 5664450.01 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:08:17,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.521e+02 1.663e+03 2.028e+03 3.324e+03 9.366e+03, threshold=4.055e+03, percent-clipped=13.0
+2023-03-09 19:09:02,954 INFO [train.py:968] (0/2) Epoch 19, batch 9500, giga_loss[loss=0.2796, simple_loss=0.3611, pruned_loss=0.09911, over 28616.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3722, pruned_loss=0.1205, over 5665326.64 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3478, pruned_loss=0.0937, over 5707937.31 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3752, pruned_loss=0.1238, over 5670640.55 frames. ], batch size: 85, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:09:21,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4000, 1.8064, 1.6479, 1.5728], device='cuda:0'), covar=tensor([0.1921, 0.1990, 0.2218, 0.2135], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0748, 0.0704, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 19:09:22,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5790, 1.9618, 1.8454, 1.7198], device='cuda:0'), covar=tensor([0.1855, 0.1871, 0.2126, 0.1949], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0748, 0.0704, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 19:09:49,661 INFO [train.py:968] (0/2) Epoch 19, batch 9550, libri_loss[loss=0.2782, simple_loss=0.3607, pruned_loss=0.0978, over 29759.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3739, pruned_loss=0.1198, over 5671839.62 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3476, pruned_loss=0.09349, over 5713261.99 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3772, pruned_loss=0.1232, over 5670337.52 frames. ], batch size: 87, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:09:51,199 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.740e+02 1.447e+03 1.882e+03 2.632e+03 7.091e+03, threshold=3.764e+03, percent-clipped=7.0
+2023-03-09 19:10:30,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0001, 5.2153, 2.2739, 2.0629], device='cuda:0'), covar=tensor([0.0898, 0.0374, 0.0824, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0550, 0.0374, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 19:10:38,595 INFO [train.py:968] (0/2) Epoch 19, batch 9600, giga_loss[loss=0.2689, simple_loss=0.3495, pruned_loss=0.09419, over 28533.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3763, pruned_loss=0.1205, over 5673986.91 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3472, pruned_loss=0.09329, over 5717373.15 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3796, pruned_loss=0.1239, over 5668369.04 frames. ], batch size: 85, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:11:28,996 INFO [train.py:968] (0/2) Epoch 19, batch 9650, giga_loss[loss=0.3504, simple_loss=0.4049, pruned_loss=0.1479, over 28232.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3793, pruned_loss=0.1234, over 5670822.64 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3473, pruned_loss=0.0933, over 5711175.73 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3824, pruned_loss=0.1266, over 5671951.45 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:11:29,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6443, 1.8372, 1.5376, 1.7311], device='cuda:0'), covar=tensor([0.2367, 0.2339, 0.2538, 0.2242], device='cuda:0'), in_proj_covar=tensor([0.1456, 0.1054, 0.1290, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 19:11:34,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.526e+03 1.837e+03 2.481e+03 5.878e+03, threshold=3.674e+03, percent-clipped=9.0
+2023-03-09 19:12:17,167 INFO [train.py:968] (0/2) Epoch 19, batch 9700, giga_loss[loss=0.3892, simple_loss=0.4193, pruned_loss=0.1796, over 26546.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3807, pruned_loss=0.1256, over 5672383.48 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3471, pruned_loss=0.09317, over 5714158.54 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3841, pruned_loss=0.129, over 5669730.05 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:13:10,360 INFO [train.py:968] (0/2) Epoch 19, batch 9750, giga_loss[loss=0.2906, simple_loss=0.3664, pruned_loss=0.1074, over 28957.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3803, pruned_loss=0.1266, over 5660897.32 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3472, pruned_loss=0.09319, over 5715322.75 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3831, pruned_loss=0.1293, over 5657631.40 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:13:12,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.171e+03 1.634e+03 2.148e+03 3.015e+03 8.168e+03, threshold=4.295e+03, percent-clipped=12.0
+2023-03-09 19:13:58,630 INFO [train.py:968] (0/2) Epoch 19, batch 9800, giga_loss[loss=0.3081, simple_loss=0.3807, pruned_loss=0.1178, over 28546.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3787, pruned_loss=0.1252, over 5650734.61 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3468, pruned_loss=0.093, over 5708883.67 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3819, pruned_loss=0.1282, over 5652533.83 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:14:26,068 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 19:14:44,525 INFO [train.py:968] (0/2) Epoch 19, batch 9850, giga_loss[loss=0.3361, simple_loss=0.4039, pruned_loss=0.1341, over 29006.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.378, pruned_loss=0.1227, over 5657954.75 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3465, pruned_loss=0.09284, over 5711878.58 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3812, pruned_loss=0.1257, over 5655962.18 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:14:47,104 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.560e+03 1.884e+03 2.463e+03 7.067e+03, threshold=3.767e+03, percent-clipped=5.0
+2023-03-09 19:15:17,734 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=832470.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:15:31,470 INFO [train.py:968] (0/2) Epoch 19, batch 9900, giga_loss[loss=0.2844, simple_loss=0.3641, pruned_loss=0.1023, over 28856.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3772, pruned_loss=0.1209, over 5669632.84 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3465, pruned_loss=0.09293, over 5716732.88 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3805, pruned_loss=0.1238, over 5662355.74 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:16:21,378 INFO [train.py:968] (0/2) Epoch 19, batch 9950, giga_loss[loss=0.4135, simple_loss=0.437, pruned_loss=0.195, over 26701.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3784, pruned_loss=0.1219, over 5660267.81 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3469, pruned_loss=0.09296, over 5711968.19 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3816, pruned_loss=0.1252, over 5657990.53 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:16:23,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.654e+03 2.089e+03 2.830e+03 4.598e+03, threshold=4.178e+03, percent-clipped=10.0
+2023-03-09 19:17:10,828 INFO [train.py:968] (0/2) Epoch 19, batch 10000, giga_loss[loss=0.2945, simple_loss=0.3673, pruned_loss=0.1108, over 28987.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3772, pruned_loss=0.1212, over 5665746.27 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3473, pruned_loss=0.09317, over 5717094.29 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3804, pruned_loss=0.1245, over 5657813.88 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:17:13,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2751, 1.5922, 1.3008, 0.9637], device='cuda:0'), covar=tensor([0.2396, 0.2441, 0.2742, 0.2242], device='cuda:0'), in_proj_covar=tensor([0.1457, 0.1057, 0.1291, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 19:17:59,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5012, 1.8901, 1.4892, 1.4914], device='cuda:0'), covar=tensor([0.2566, 0.2427, 0.2903, 0.2241], device='cuda:0'), in_proj_covar=tensor([0.1459, 0.1058, 0.1292, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 19:18:00,567 INFO [train.py:968] (0/2) Epoch 19, batch 10050, giga_loss[loss=0.2869, simple_loss=0.3509, pruned_loss=0.1114, over 28956.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3749, pruned_loss=0.1206, over 5674383.51 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3469, pruned_loss=0.09286, over 5720913.58 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3786, pruned_loss=0.1242, over 5663231.06 frames. ], batch size: 106, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:18:04,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.752e+03 2.048e+03 2.667e+03 5.818e+03, threshold=4.095e+03, percent-clipped=5.0
+2023-03-09 19:18:47,428 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 19:18:48,721 INFO [train.py:968] (0/2) Epoch 19, batch 10100, giga_loss[loss=0.3248, simple_loss=0.3797, pruned_loss=0.135, over 28749.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3749, pruned_loss=0.1216, over 5669277.56 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3475, pruned_loss=0.09309, over 5722075.97 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3782, pruned_loss=0.1251, over 5657680.66 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:19:36,394 INFO [train.py:968] (0/2) Epoch 19, batch 10150, giga_loss[loss=0.2542, simple_loss=0.3273, pruned_loss=0.09053, over 28991.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3724, pruned_loss=0.12, over 5670066.81 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3479, pruned_loss=0.09321, over 5718837.62 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3758, pruned_loss=0.1241, over 5661218.70 frames. ], batch size: 106, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:19:40,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.552e+03 2.087e+03 2.922e+03 5.543e+03, threshold=4.174e+03, percent-clipped=7.0
+2023-03-09 19:20:29,539 INFO [train.py:968] (0/2) Epoch 19, batch 10200, giga_loss[loss=0.3069, simple_loss=0.3675, pruned_loss=0.1231, over 28986.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3709, pruned_loss=0.1203, over 5673919.02 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3472, pruned_loss=0.09282, over 5721356.38 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3746, pruned_loss=0.1242, over 5664225.93 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:21:06,456 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 19:21:17,843 INFO [train.py:968] (0/2) Epoch 19, batch 10250, giga_loss[loss=0.2876, simple_loss=0.3615, pruned_loss=0.1068, over 28644.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.371, pruned_loss=0.1208, over 5675785.25 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3474, pruned_loss=0.09284, over 5725130.09 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3743, pruned_loss=0.1245, over 5663446.83 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:21:22,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.264e+02 1.645e+03 2.150e+03 3.716e+03 1.075e+04, threshold=4.299e+03, percent-clipped=18.0
+2023-03-09 19:21:26,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=832845.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:22:05,819 INFO [train.py:968] (0/2) Epoch 19, batch 10300, giga_loss[loss=0.2654, simple_loss=0.349, pruned_loss=0.09092, over 28673.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 5676068.45 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3472, pruned_loss=0.09272, over 5730800.00 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3717, pruned_loss=0.122, over 5659206.33 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:22:28,200 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=832913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:22:51,995 INFO [train.py:968] (0/2) Epoch 19, batch 10350, giga_loss[loss=0.2638, simple_loss=0.3416, pruned_loss=0.09296, over 29036.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3653, pruned_loss=0.1152, over 5676311.29 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3471, pruned_loss=0.09271, over 5735014.29 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3686, pruned_loss=0.1188, over 5657615.20 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:23:00,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.551e+02 1.290e+03 1.689e+03 2.298e+03 8.933e+03, threshold=3.379e+03, percent-clipped=4.0
+2023-03-09 19:23:05,537 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=832948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:23:42,900 INFO [train.py:968] (0/2) Epoch 19, batch 10400, giga_loss[loss=0.3541, simple_loss=0.4042, pruned_loss=0.1521, over 28884.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3646, pruned_loss=0.1147, over 5671432.75 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3469, pruned_loss=0.09262, over 5734309.04 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3678, pruned_loss=0.1182, over 5655615.38 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:23:43,420 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=832988.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:23:45,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=832991.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:24:13,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=833020.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:24:32,371 INFO [train.py:968] (0/2) Epoch 19, batch 10450, giga_loss[loss=0.2541, simple_loss=0.3261, pruned_loss=0.09101, over 29009.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3628, pruned_loss=0.114, over 5677162.73 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3472, pruned_loss=0.09268, over 5735156.87 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3655, pruned_loss=0.1173, over 5662241.66 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:24:41,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.302e+02 1.542e+03 1.997e+03 2.463e+03 6.681e+03, threshold=3.994e+03, percent-clipped=15.0
+2023-03-09 19:25:19,768 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=833082.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:25:25,033 INFO [train.py:968] (0/2) Epoch 19, batch 10500, giga_loss[loss=0.311, simple_loss=0.372, pruned_loss=0.125, over 28863.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3606, pruned_loss=0.1136, over 5675083.50 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3472, pruned_loss=0.09259, over 5738575.20 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3631, pruned_loss=0.1167, over 5658928.04 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:26:12,430 INFO [train.py:968] (0/2) Epoch 19, batch 10550, giga_loss[loss=0.2903, simple_loss=0.3567, pruned_loss=0.1119, over 28782.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3625, pruned_loss=0.1141, over 5676908.85 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3471, pruned_loss=0.09244, over 5740517.68 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3648, pruned_loss=0.1172, over 5660800.03 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:26:16,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.746e+03 2.399e+03 3.366e+03 8.289e+03, threshold=4.797e+03, percent-clipped=18.0
+2023-03-09 19:26:48,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3948, 1.4039, 3.3765, 3.2144], device='cuda:0'), covar=tensor([0.1270, 0.2415, 0.0410, 0.0957], device='cuda:0'), in_proj_covar=tensor([0.0739, 0.0636, 0.0939, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 19:26:57,482 INFO [train.py:968] (0/2) Epoch 19, batch 10600, giga_loss[loss=0.2728, simple_loss=0.3569, pruned_loss=0.09429, over 28909.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3638, pruned_loss=0.1139, over 5679055.25 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3473, pruned_loss=0.0925, over 5741490.18 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.366, pruned_loss=0.117, over 5663428.50 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:27:49,020 INFO [train.py:968] (0/2) Epoch 19, batch 10650, libri_loss[loss=0.2362, simple_loss=0.3215, pruned_loss=0.07547, over 29553.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3651, pruned_loss=0.1157, over 5647551.64 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3474, pruned_loss=0.09256, over 5740502.99 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3671, pruned_loss=0.1184, over 5634730.08 frames. ], batch size: 76, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:27:53,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.084e+02 1.392e+03 1.729e+03 2.206e+03 5.385e+03, threshold=3.457e+03, percent-clipped=1.0
+2023-03-09 19:28:36,406 INFO [train.py:968] (0/2) Epoch 19, batch 10700, giga_loss[loss=0.3248, simple_loss=0.3832, pruned_loss=0.1332, over 28895.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3658, pruned_loss=0.1164, over 5650020.74 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3478, pruned_loss=0.09267, over 5745528.10 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5631800.07 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:28:36,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=833288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:29:10,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=833323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:29:21,815 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.39 vs. limit=5.0
+2023-03-09 19:29:27,346 INFO [train.py:968] (0/2) Epoch 19, batch 10750, giga_loss[loss=0.3179, simple_loss=0.3828, pruned_loss=0.1264, over 28634.00 frames. ], tot_loss[loss=0.301, simple_loss=0.367, pruned_loss=0.1175, over 5642235.83 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.348, pruned_loss=0.0927, over 5737935.37 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3687, pruned_loss=0.1203, over 5633141.54 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:29:31,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.853e+02 1.495e+03 1.866e+03 2.729e+03 5.599e+03, threshold=3.732e+03, percent-clipped=15.0
+2023-03-09 19:30:05,351 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-09 19:30:19,040 INFO [train.py:968] (0/2) Epoch 19, batch 10800, giga_loss[loss=0.3169, simple_loss=0.3849, pruned_loss=0.1245, over 29025.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.369, pruned_loss=0.1184, over 5644880.16 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3483, pruned_loss=0.09292, over 5730787.08 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3704, pruned_loss=0.1209, over 5641454.71 frames. ], batch size: 128, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:30:41,784 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-09 19:30:44,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2589, 0.8312, 0.8705, 1.4370], device='cuda:0'), covar=tensor([0.0741, 0.0382, 0.0344, 0.0802], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 19:31:01,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=833431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:03,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=833434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:07,156 INFO [train.py:968] (0/2) Epoch 19, batch 10850, libri_loss[loss=0.2561, simple_loss=0.3446, pruned_loss=0.08383, over 29227.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3686, pruned_loss=0.1178, over 5651392.61 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3479, pruned_loss=0.09252, over 5733853.28 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3707, pruned_loss=0.121, over 5643078.91 frames. ], batch size: 94, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:31:12,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.068e+03 1.392e+03 1.863e+03 2.588e+03 8.817e+03, threshold=3.727e+03, percent-clipped=8.0
+2023-03-09 19:31:24,481 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=833456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:25,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=833457.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:31:31,215 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=833463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:31,229 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=833463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:33,764 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=833466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:37,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=833469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:54,707 INFO [train.py:968] (0/2) Epoch 19, batch 10900, giga_loss[loss=0.3711, simple_loss=0.4172, pruned_loss=0.1626, over 28301.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3713, pruned_loss=0.1204, over 5655418.43 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3479, pruned_loss=0.09265, over 5737570.74 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3735, pruned_loss=0.1233, over 5643834.34 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:32:05,283 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=833498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:32:18,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7723, 4.9359, 1.9467, 2.2598], device='cuda:0'), covar=tensor([0.0960, 0.0309, 0.0820, 0.1137], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0547, 0.0374, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 19:32:44,224 INFO [train.py:968] (0/2) Epoch 19, batch 10950, libri_loss[loss=0.2536, simple_loss=0.3297, pruned_loss=0.08877, over 29409.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3715, pruned_loss=0.1206, over 5654569.64 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3479, pruned_loss=0.09263, over 5739776.70 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3738, pruned_loss=0.1238, over 5641028.50 frames. ], batch size: 67, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:32:46,621 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7178, 2.7665, 1.6087, 0.7403], device='cuda:0'), covar=tensor([0.7368, 0.3318, 0.3836, 0.6058], device='cuda:0'), in_proj_covar=tensor([0.1714, 0.1619, 0.1580, 0.1401], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 19:32:50,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.694e+03 2.440e+03 3.692e+03 1.295e+04, threshold=4.880e+03, percent-clipped=24.0
+2023-03-09 19:33:35,250 INFO [train.py:968] (0/2) Epoch 19, batch 11000, giga_loss[loss=0.2652, simple_loss=0.3409, pruned_loss=0.09477, over 28824.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3707, pruned_loss=0.1186, over 5662411.41 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3475, pruned_loss=0.09249, over 5744868.03 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3738, pruned_loss=0.1221, over 5644785.52 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:33:48,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3428, 3.1602, 3.0077, 1.2602], device='cuda:0'), covar=tensor([0.0957, 0.1118, 0.1064, 0.2407], device='cuda:0'), in_proj_covar=tensor([0.1194, 0.1111, 0.0951, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 19:33:48,934 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=833600.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:33:51,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=833603.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:34:24,851 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=833632.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:34:29,919 INFO [train.py:968] (0/2) Epoch 19, batch 11050, giga_loss[loss=0.2983, simple_loss=0.3599, pruned_loss=0.1184, over 28842.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.1199, over 5656133.57 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3469, pruned_loss=0.09224, over 5747298.05 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3742, pruned_loss=0.1232, over 5639222.74 frames. ], batch size: 119, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:34:35,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.997e+02 1.661e+03 1.939e+03 2.895e+03 6.085e+03, threshold=3.878e+03, percent-clipped=3.0
+2023-03-09 19:35:20,743 INFO [train.py:968] (0/2) Epoch 19, batch 11100, giga_loss[loss=0.3047, simple_loss=0.3692, pruned_loss=0.1201, over 28004.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3685, pruned_loss=0.1184, over 5670400.98 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3466, pruned_loss=0.09209, over 5751501.87 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3719, pruned_loss=0.1219, over 5650868.13 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:35:24,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4616, 4.3084, 4.0871, 2.1229], device='cuda:0'), covar=tensor([0.0602, 0.0748, 0.0799, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.1194, 0.1110, 0.0949, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 19:36:14,770 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=833733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:36:19,286 INFO [train.py:968] (0/2) Epoch 19, batch 11150, giga_loss[loss=0.2868, simple_loss=0.3556, pruned_loss=0.109, over 29065.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3681, pruned_loss=0.119, over 5661553.33 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3466, pruned_loss=0.092, over 5751828.08 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3714, pruned_loss=0.1224, over 5643997.90 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:36:25,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.649e+02 1.625e+03 1.998e+03 2.595e+03 6.307e+03, threshold=3.997e+03, percent-clipped=10.0
+2023-03-09 19:37:07,998 INFO [train.py:968] (0/2) Epoch 19, batch 11200, libri_loss[loss=0.2113, simple_loss=0.2981, pruned_loss=0.06223, over 29333.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3663, pruned_loss=0.1178, over 5675459.69 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3465, pruned_loss=0.09193, over 5753575.86 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3693, pruned_loss=0.121, over 5658900.70 frames. ], batch size: 71, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:37:41,841 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=833824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:37:48,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=833831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:37:53,223 INFO [train.py:968] (0/2) Epoch 19, batch 11250, giga_loss[loss=0.3035, simple_loss=0.367, pruned_loss=0.12, over 27969.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3661, pruned_loss=0.1179, over 5675320.80 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3466, pruned_loss=0.09184, over 5754798.12 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.369, pruned_loss=0.1212, over 5658697.27 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:37:53,451 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=833838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:38:00,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.475e+02 1.571e+03 2.184e+03 3.017e+03 6.226e+03, threshold=4.367e+03, percent-clipped=10.0
+2023-03-09 19:38:44,567 INFO [train.py:968] (0/2) Epoch 19, batch 11300, giga_loss[loss=0.3089, simple_loss=0.3702, pruned_loss=0.1238, over 28684.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3667, pruned_loss=0.1186, over 5672394.65 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3467, pruned_loss=0.09191, over 5752517.05 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3691, pruned_loss=0.1216, over 5660440.84 frames. ], batch size: 78, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:39:35,651 INFO [train.py:968] (0/2) Epoch 19, batch 11350, giga_loss[loss=0.2745, simple_loss=0.3388, pruned_loss=0.1051, over 28709.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3676, pruned_loss=0.1197, over 5666178.40 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3465, pruned_loss=0.09187, over 5745953.80 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3701, pruned_loss=0.1227, over 5659710.99 frames. ], batch size: 92, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:39:40,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.186e+03 1.670e+03 2.100e+03 2.906e+03 6.054e+03, threshold=4.199e+03, percent-clipped=9.0
+2023-03-09 19:40:00,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2416, 1.5996, 1.2290, 0.9874], device='cuda:0'), covar=tensor([0.2593, 0.2612, 0.3018, 0.2268], device='cuda:0'), in_proj_covar=tensor([0.1460, 0.1058, 0.1294, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 19:40:11,319 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=833974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:14,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=833977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:16,919 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=833981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:19,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=833984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:22,116 INFO [train.py:968] (0/2) Epoch 19, batch 11400, giga_loss[loss=0.3067, simple_loss=0.3657, pruned_loss=0.1238, over 28677.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3701, pruned_loss=0.1216, over 5659855.32 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.347, pruned_loss=0.09218, over 5736064.12 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3723, pruned_loss=0.1244, over 5661658.30 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:40:25,257 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-09 19:40:36,217 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-834000.pt
+2023-03-09 19:40:42,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:47,684 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:48,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2746, 1.8256, 1.4267, 0.4538], device='cuda:0'), covar=tensor([0.4195, 0.2686, 0.3848, 0.5615], device='cuda:0'), in_proj_covar=tensor([0.1713, 0.1616, 0.1578, 0.1396], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 19:41:00,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9784, 3.7680, 3.6281, 1.6711], device='cuda:0'), covar=tensor([0.0712, 0.0900, 0.0874, 0.2158], device='cuda:0'), in_proj_covar=tensor([0.1195, 0.1112, 0.0952, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 19:41:11,405 INFO [train.py:968] (0/2) Epoch 19, batch 11450, giga_loss[loss=0.3008, simple_loss=0.364, pruned_loss=0.1188, over 28916.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.371, pruned_loss=0.1221, over 5664164.78 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3469, pruned_loss=0.09226, over 5739766.14 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3734, pruned_loss=0.1251, over 5660411.44 frames. ], batch size: 112, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:41:17,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.652e+03 2.358e+03 3.379e+03 9.804e+03, threshold=4.716e+03, percent-clipped=13.0
+2023-03-09 19:42:05,428 INFO [train.py:968] (0/2) Epoch 19, batch 11500, giga_loss[loss=0.3525, simple_loss=0.3877, pruned_loss=0.1586, over 23612.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.371, pruned_loss=0.1233, over 5654431.73 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3467, pruned_loss=0.09212, over 5743063.29 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3735, pruned_loss=0.1263, over 5647049.52 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:42:25,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:42:45,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5439, 1.6319, 1.2306, 1.2185], device='cuda:0'), covar=tensor([0.0818, 0.0515, 0.0922, 0.1233], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0446, 0.0509, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 19:42:52,264 INFO [train.py:968] (0/2) Epoch 19, batch 11550, giga_loss[loss=0.3298, simple_loss=0.3927, pruned_loss=0.1334, over 29075.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3706, pruned_loss=0.1228, over 5663428.42 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3465, pruned_loss=0.09197, over 5745810.95 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3733, pruned_loss=0.126, over 5653145.63 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:42:58,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.685e+03 2.193e+03 3.278e+03 7.982e+03, threshold=4.386e+03, percent-clipped=9.0
+2023-03-09 19:43:13,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4745, 1.5758, 1.2396, 1.1414], device='cuda:0'), covar=tensor([0.0972, 0.0603, 0.1114, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0446, 0.0509, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 19:43:40,925 INFO [train.py:968] (0/2) Epoch 19, batch 11600, giga_loss[loss=0.3607, simple_loss=0.4042, pruned_loss=0.1586, over 27629.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1224, over 5675831.36 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3466, pruned_loss=0.09201, over 5749309.03 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3733, pruned_loss=0.1255, over 5663120.35 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:43:53,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:43:59,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9412, 1.2492, 1.3162, 1.0586], device='cuda:0'), covar=tensor([0.1961, 0.1468, 0.2419, 0.1885], device='cuda:0'), in_proj_covar=tensor([0.0473, 0.0751, 0.0707, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 19:44:26,207 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:44:31,120 INFO [train.py:968] (0/2) Epoch 19, batch 11650, giga_loss[loss=0.3085, simple_loss=0.3767, pruned_loss=0.1202, over 28775.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3714, pruned_loss=0.1226, over 5672193.21 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.347, pruned_loss=0.0923, over 5750101.01 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3734, pruned_loss=0.1252, over 5660466.37 frames. ], batch size: 243, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:44:40,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.551e+02 1.587e+03 2.115e+03 3.139e+03 8.959e+03, threshold=4.230e+03, percent-clipped=14.0
+2023-03-09 19:44:45,082 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=834251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:44:49,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=834254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:44:59,651 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 19:45:20,406 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:45:25,852 INFO [train.py:968] (0/2) Epoch 19, batch 11700, giga_loss[loss=0.3259, simple_loss=0.3834, pruned_loss=0.1342, over 28704.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5681432.91 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3471, pruned_loss=0.09238, over 5750662.83 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.374, pruned_loss=0.1253, over 5670697.30 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:46:16,516 INFO [train.py:968] (0/2) Epoch 19, batch 11750, giga_loss[loss=0.3451, simple_loss=0.3955, pruned_loss=0.1474, over 28716.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3744, pruned_loss=0.1247, over 5674599.02 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3476, pruned_loss=0.0926, over 5750206.00 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.376, pruned_loss=0.1271, over 5664986.70 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:46:20,383 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=834342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:46:23,201 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=834345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:46:23,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.068e+03 1.558e+03 1.991e+03 2.411e+03 6.625e+03, threshold=3.982e+03, percent-clipped=4.0
+2023-03-09 19:46:32,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:46:51,191 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:47:04,431 INFO [train.py:968] (0/2) Epoch 19, batch 11800, giga_loss[loss=0.2828, simple_loss=0.3594, pruned_loss=0.103, over 28790.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.374, pruned_loss=0.1241, over 5685678.35 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3473, pruned_loss=0.0924, over 5752713.69 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3759, pruned_loss=0.1266, over 5674933.89 frames. ], batch size: 112, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:47:52,021 INFO [train.py:968] (0/2) Epoch 19, batch 11850, giga_loss[loss=0.298, simple_loss=0.3743, pruned_loss=0.1108, over 28741.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3738, pruned_loss=0.1229, over 5683832.64 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3473, pruned_loss=0.09234, over 5754558.41 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3762, pruned_loss=0.126, over 5671290.11 frames. ], batch size: 243, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:47:59,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.853e+02 1.491e+03 1.881e+03 2.981e+03 8.893e+03, threshold=3.763e+03, percent-clipped=10.0
+2023-03-09 19:48:12,015 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 19:48:17,157 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:48:39,793 INFO [train.py:968] (0/2) Epoch 19, batch 11900, libri_loss[loss=0.2503, simple_loss=0.3229, pruned_loss=0.08881, over 29485.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3725, pruned_loss=0.1211, over 5680809.75 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.347, pruned_loss=0.09223, over 5758974.57 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3757, pruned_loss=0.1248, over 5663935.32 frames. ], batch size: 70, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:49:11,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:49:27,374 INFO [train.py:968] (0/2) Epoch 19, batch 11950, giga_loss[loss=0.3954, simple_loss=0.4326, pruned_loss=0.1791, over 26618.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3714, pruned_loss=0.1202, over 5681566.37 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3469, pruned_loss=0.092, over 5759144.35 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3746, pruned_loss=0.1241, over 5666320.13 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:49:33,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.499e+03 1.919e+03 2.395e+03 6.254e+03, threshold=3.838e+03, percent-clipped=5.0
+2023-03-09 19:49:38,695 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834549.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:49:39,806 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:50:13,597 INFO [train.py:968] (0/2) Epoch 19, batch 12000, giga_loss[loss=0.3368, simple_loss=0.3728, pruned_loss=0.1504, over 23533.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3709, pruned_loss=0.1202, over 5680187.44 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3473, pruned_loss=0.09229, over 5756780.30 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3734, pruned_loss=0.1234, over 5669491.08 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:50:13,602 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 19:50:23,113 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2101, simple_loss=0.3175, pruned_loss=0.05135, over 944034.00 frames. 
+2023-03-09 19:50:23,114 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 19:50:39,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0053, 1.3376, 1.2718, 1.1578], device='cuda:0'), covar=tensor([0.1296, 0.0908, 0.1702, 0.1137], device='cuda:0'), in_proj_covar=tensor([0.0472, 0.0750, 0.0707, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 19:50:41,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:50:45,069 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:50:45,293 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 19:51:09,354 INFO [train.py:968] (0/2) Epoch 19, batch 12050, giga_loss[loss=0.2952, simple_loss=0.3693, pruned_loss=0.1105, over 28675.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3727, pruned_loss=0.1221, over 5672991.15 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.347, pruned_loss=0.09225, over 5759682.65 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3756, pruned_loss=0.1254, over 5659725.22 frames. ], batch size: 71, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:51:18,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.785e+03 2.076e+03 3.169e+03 5.678e+03, threshold=4.152e+03, percent-clipped=12.0
+2023-03-09 19:52:00,252 INFO [train.py:968] (0/2) Epoch 19, batch 12100, libri_loss[loss=0.2457, simple_loss=0.3244, pruned_loss=0.08346, over 29369.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3737, pruned_loss=0.1228, over 5674827.65 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3471, pruned_loss=0.09224, over 5761859.00 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3764, pruned_loss=0.1258, over 5661292.36 frames. ], batch size: 67, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:52:11,369 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:52:41,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:52:48,228 INFO [train.py:968] (0/2) Epoch 19, batch 12150, giga_loss[loss=0.321, simple_loss=0.3786, pruned_loss=0.1317, over 28182.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3734, pruned_loss=0.1231, over 5662543.11 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3473, pruned_loss=0.09233, over 5752952.81 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3761, pruned_loss=0.1264, over 5657751.99 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:52:55,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.757e+02 1.556e+03 2.112e+03 2.701e+03 5.422e+03, threshold=4.224e+03, percent-clipped=6.0
+2023-03-09 19:53:00,622 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=834750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:53:04,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=834753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:53:09,733 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:53:31,432 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:53:39,388 INFO [train.py:968] (0/2) Epoch 19, batch 12200, giga_loss[loss=0.3035, simple_loss=0.3712, pruned_loss=0.1179, over 28769.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3743, pruned_loss=0.1243, over 5665353.95 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3474, pruned_loss=0.09235, over 5753854.24 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3767, pruned_loss=0.1272, over 5659879.30 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:54:27,964 INFO [train.py:968] (0/2) Epoch 19, batch 12250, giga_loss[loss=0.3188, simple_loss=0.3864, pruned_loss=0.1255, over 28704.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3751, pruned_loss=0.1247, over 5663241.15 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3479, pruned_loss=0.09256, over 5756815.04 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3772, pruned_loss=0.1277, over 5654021.45 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:54:29,923 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:54:36,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.262e+02 1.728e+03 2.144e+03 2.833e+03 6.219e+03, threshold=4.288e+03, percent-clipped=8.0
+2023-03-09 19:54:59,386 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=834872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:55:01,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=834875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:55:03,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2805, 1.1514, 3.7287, 3.2602], device='cuda:0'), covar=tensor([0.1663, 0.2892, 0.0468, 0.1199], device='cuda:0'), in_proj_covar=tensor([0.0743, 0.0638, 0.0943, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-09 19:55:10,102 INFO [train.py:968] (0/2) Epoch 19, batch 12300, giga_loss[loss=0.2878, simple_loss=0.3572, pruned_loss=0.1092, over 28900.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3742, pruned_loss=0.1235, over 5670614.52 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3478, pruned_loss=0.09249, over 5759749.53 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3777, pruned_loss=0.1279, over 5655297.00 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:55:18,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:55:24,488 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:55:46,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834924.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:55:49,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:55:52,151 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:56:00,043 INFO [train.py:968] (0/2) Epoch 19, batch 12350, libri_loss[loss=0.2993, simple_loss=0.3696, pruned_loss=0.1145, over 20080.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3745, pruned_loss=0.1245, over 5637884.70 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.348, pruned_loss=0.09256, over 5753059.99 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3775, pruned_loss=0.1285, over 5630677.70 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:56:09,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.465e+03 1.909e+03 2.552e+03 1.015e+04, threshold=3.818e+03, percent-clipped=7.0
+2023-03-09 19:56:46,609 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=834982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:56:49,675 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:56:49,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=834985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:56:51,448 INFO [train.py:968] (0/2) Epoch 19, batch 12400, giga_loss[loss=0.2967, simple_loss=0.3694, pruned_loss=0.112, over 28962.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3737, pruned_loss=0.1237, over 5646628.07 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3479, pruned_loss=0.09247, over 5755775.04 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3768, pruned_loss=0.1276, over 5636363.16 frames. ], batch size: 227, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:56:56,148 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:57:15,199 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:57:34,306 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:57:34,679 INFO [train.py:968] (0/2) Epoch 19, batch 12450, giga_loss[loss=0.315, simple_loss=0.3786, pruned_loss=0.1257, over 27982.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3738, pruned_loss=0.123, over 5659258.00 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3476, pruned_loss=0.09231, over 5759609.46 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3772, pruned_loss=0.127, over 5644845.18 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:57:38,868 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:57:41,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:57:44,004 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.350e+02 1.558e+03 2.011e+03 2.880e+03 5.393e+03, threshold=4.021e+03, percent-clipped=7.0
+2023-03-09 19:57:58,784 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:58:01,197 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835067.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:58:04,554 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:58:05,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835070.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:58:07,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:58:07,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:58:14,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:58:25,561 INFO [train.py:968] (0/2) Epoch 19, batch 12500, giga_loss[loss=0.2674, simple_loss=0.3383, pruned_loss=0.09827, over 28633.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3739, pruned_loss=0.1233, over 5652202.64 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3481, pruned_loss=0.09255, over 5751936.89 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3767, pruned_loss=0.127, over 5644798.60 frames. ], batch size: 71, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:58:36,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835099.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:58:37,367 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:59:02,962 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:59:05,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:59:05,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:59:11,380 INFO [train.py:968] (0/2) Epoch 19, batch 12550, giga_loss[loss=0.2491, simple_loss=0.3264, pruned_loss=0.08586, over 28348.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3726, pruned_loss=0.1231, over 5650067.39 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.348, pruned_loss=0.09252, over 5745810.86 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3755, pruned_loss=0.1267, over 5647036.61 frames. ], batch size: 65, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:59:21,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.213e+03 1.680e+03 2.112e+03 2.916e+03 5.648e+03, threshold=4.225e+03, percent-clipped=6.0
+2023-03-09 19:59:34,210 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:59:58,027 INFO [train.py:968] (0/2) Epoch 19, batch 12600, giga_loss[loss=0.2405, simple_loss=0.314, pruned_loss=0.08353, over 28533.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3688, pruned_loss=0.1205, over 5664424.45 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3482, pruned_loss=0.09278, over 5742595.40 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3717, pruned_loss=0.1242, over 5662383.86 frames. ], batch size: 85, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:00:31,036 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:00:34,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:00:50,776 INFO [train.py:968] (0/2) Epoch 19, batch 12650, giga_loss[loss=0.3047, simple_loss=0.3704, pruned_loss=0.1195, over 28898.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3647, pruned_loss=0.1189, over 5649173.05 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3478, pruned_loss=0.09254, over 5744076.21 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3675, pruned_loss=0.1223, over 5645738.97 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:01:02,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.163e+03 1.659e+03 2.233e+03 3.542e+03 1.235e+04, threshold=4.466e+03, percent-clipped=14.0
+2023-03-09 20:01:07,437 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:01:27,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:01:29,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:01:40,675 INFO [train.py:968] (0/2) Epoch 19, batch 12700, giga_loss[loss=0.2875, simple_loss=0.3477, pruned_loss=0.1137, over 28831.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3644, pruned_loss=0.1192, over 5654319.37 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3481, pruned_loss=0.09252, over 5748628.15 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3669, pruned_loss=0.1227, over 5644642.00 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:01:56,347 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:01:59,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:02:01,862 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 20:02:26,613 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-09 20:02:29,976 INFO [train.py:968] (0/2) Epoch 19, batch 12750, libri_loss[loss=0.293, simple_loss=0.3743, pruned_loss=0.1058, over 25740.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3634, pruned_loss=0.1187, over 5649614.10 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.348, pruned_loss=0.09235, over 5748238.28 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3658, pruned_loss=0.1221, over 5640850.97 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:02:40,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.600e+03 2.023e+03 3.036e+03 8.262e+03, threshold=4.046e+03, percent-clipped=11.0
+2023-03-09 20:02:46,738 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:03:01,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:03:18,574 INFO [train.py:968] (0/2) Epoch 19, batch 12800, giga_loss[loss=0.275, simple_loss=0.3544, pruned_loss=0.09778, over 27658.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3628, pruned_loss=0.1174, over 5653812.05 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3476, pruned_loss=0.09237, over 5752114.78 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3655, pruned_loss=0.121, over 5640105.13 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:03:41,664 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 20:03:43,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:04:10,894 INFO [train.py:968] (0/2) Epoch 19, batch 12850, giga_loss[loss=0.2959, simple_loss=0.3703, pruned_loss=0.1108, over 28246.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3623, pruned_loss=0.115, over 5647401.49 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3479, pruned_loss=0.09269, over 5743961.80 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3645, pruned_loss=0.118, over 5642336.32 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:04:12,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:04:20,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:04:21,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.766e+03 2.417e+03 3.660e+03 1.319e+04, threshold=4.833e+03, percent-clipped=22.0
+2023-03-09 20:04:22,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:04:32,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:04:54,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:05:03,597 INFO [train.py:968] (0/2) Epoch 19, batch 12900, giga_loss[loss=0.2634, simple_loss=0.3417, pruned_loss=0.09253, over 28736.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3598, pruned_loss=0.1118, over 5650415.04 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3476, pruned_loss=0.09259, over 5746196.15 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3619, pruned_loss=0.1145, over 5643312.22 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:05:28,467 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:05:33,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:05:46,808 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-09 20:05:52,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9779, 3.7711, 3.6158, 1.8094], device='cuda:0'), covar=tensor([0.0851, 0.0985, 0.1076, 0.2240], device='cuda:0'), in_proj_covar=tensor([0.1186, 0.1107, 0.0943, 0.0712], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 20:05:54,768 INFO [train.py:968] (0/2) Epoch 19, batch 12950, giga_loss[loss=0.2681, simple_loss=0.339, pruned_loss=0.09862, over 27627.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3557, pruned_loss=0.1079, over 5655277.43 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3464, pruned_loss=0.0922, over 5752179.43 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3589, pruned_loss=0.1111, over 5640593.54 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:06:00,053 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:06:05,233 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.768e+02 1.339e+03 1.690e+03 2.368e+03 4.341e+03, threshold=3.381e+03, percent-clipped=0.0
+2023-03-09 20:06:11,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:06:16,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:06:41,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:06:44,051 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:06:45,443 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:06:45,822 INFO [train.py:968] (0/2) Epoch 19, batch 13000, giga_loss[loss=0.2575, simple_loss=0.3361, pruned_loss=0.08944, over 27961.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3521, pruned_loss=0.1047, over 5636445.25 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3462, pruned_loss=0.09216, over 5738943.03 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3551, pruned_loss=0.1078, over 5633322.07 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:07:14,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:07:15,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:07:36,887 INFO [train.py:968] (0/2) Epoch 19, batch 13050, giga_loss[loss=0.2411, simple_loss=0.3343, pruned_loss=0.07397, over 28988.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3507, pruned_loss=0.1011, over 5652681.42 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3458, pruned_loss=0.09194, over 5740784.18 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3535, pruned_loss=0.1037, over 5647760.56 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:07:48,732 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.840e+02 1.375e+03 1.789e+03 2.447e+03 5.814e+03, threshold=3.577e+03, percent-clipped=8.0
+2023-03-09 20:07:57,285 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-09 20:08:30,834 INFO [train.py:968] (0/2) Epoch 19, batch 13100, giga_loss[loss=0.2962, simple_loss=0.3659, pruned_loss=0.1133, over 28742.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3519, pruned_loss=0.1019, over 5652954.77 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3455, pruned_loss=0.09195, over 5744700.59 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3546, pruned_loss=0.1042, over 5643407.84 frames. ], batch size: 284, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:09:14,261 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:09:23,956 INFO [train.py:968] (0/2) Epoch 19, batch 13150, giga_loss[loss=0.2729, simple_loss=0.35, pruned_loss=0.09795, over 28544.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3512, pruned_loss=0.1013, over 5642321.94 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3453, pruned_loss=0.09195, over 5733909.67 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3536, pruned_loss=0.1033, over 5643532.92 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:09:35,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.507e+02 1.404e+03 1.971e+03 3.299e+03 9.968e+03, threshold=3.942e+03, percent-clipped=22.0
+2023-03-09 20:10:07,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4859, 3.7907, 1.6613, 1.7458], device='cuda:0'), covar=tensor([0.0990, 0.0282, 0.0930, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0550, 0.0375, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-09 20:10:15,597 INFO [train.py:968] (0/2) Epoch 19, batch 13200, giga_loss[loss=0.2913, simple_loss=0.3699, pruned_loss=0.1064, over 28884.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3476, pruned_loss=0.09896, over 5628696.54 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3447, pruned_loss=0.09173, over 5727065.49 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3501, pruned_loss=0.1009, over 5633380.27 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:10:26,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3868, 1.6233, 1.6297, 1.2638], device='cuda:0'), covar=tensor([0.1458, 0.2208, 0.1255, 0.1604], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0695, 0.0922, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 20:11:02,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:11:07,653 INFO [train.py:968] (0/2) Epoch 19, batch 13250, giga_loss[loss=0.2693, simple_loss=0.3454, pruned_loss=0.09657, over 28315.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.347, pruned_loss=0.09887, over 5636544.60 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3446, pruned_loss=0.09172, over 5729905.42 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3492, pruned_loss=0.1005, over 5636254.66 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:11:18,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.978e+02 1.412e+03 1.860e+03 2.637e+03 1.219e+04, threshold=3.721e+03, percent-clipped=10.0
+2023-03-09 20:11:41,760 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:11:45,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:11:57,687 INFO [train.py:968] (0/2) Epoch 19, batch 13300, giga_loss[loss=0.2943, simple_loss=0.3532, pruned_loss=0.1177, over 26710.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3469, pruned_loss=0.09854, over 5641746.49 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3441, pruned_loss=0.09152, over 5734073.53 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3491, pruned_loss=0.1002, over 5635804.46 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:12:11,955 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:12:16,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:12:47,139 INFO [train.py:968] (0/2) Epoch 19, batch 13350, giga_loss[loss=0.232, simple_loss=0.312, pruned_loss=0.07601, over 27548.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3454, pruned_loss=0.09698, over 5656678.56 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3435, pruned_loss=0.09127, over 5737974.37 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3478, pruned_loss=0.09874, over 5645745.03 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:13:00,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.034e+02 1.419e+03 1.991e+03 2.990e+03 9.303e+03, threshold=3.981e+03, percent-clipped=16.0
+2023-03-09 20:13:26,137 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:13:28,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:13:38,993 INFO [train.py:968] (0/2) Epoch 19, batch 13400, giga_loss[loss=0.2351, simple_loss=0.3232, pruned_loss=0.07353, over 28671.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3422, pruned_loss=0.09452, over 5641328.73 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3431, pruned_loss=0.09112, over 5727406.31 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3445, pruned_loss=0.09617, over 5640398.78 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:13:40,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:13:51,166 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-836000.pt
+2023-03-09 20:13:58,710 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=836007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:14:32,697 INFO [train.py:968] (0/2) Epoch 19, batch 13450, giga_loss[loss=0.2427, simple_loss=0.3172, pruned_loss=0.08408, over 28815.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3389, pruned_loss=0.09274, over 5643837.28 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3432, pruned_loss=0.09133, over 5728835.21 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3406, pruned_loss=0.09392, over 5639402.62 frames. ], batch size: 112, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:14:45,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.200e+02 1.304e+03 1.638e+03 2.462e+03 5.879e+03, threshold=3.276e+03, percent-clipped=1.0
+2023-03-09 20:15:26,083 INFO [train.py:968] (0/2) Epoch 19, batch 13500, giga_loss[loss=0.2943, simple_loss=0.3622, pruned_loss=0.1132, over 28785.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3371, pruned_loss=0.09193, over 5659273.60 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3427, pruned_loss=0.0911, over 5733531.80 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3387, pruned_loss=0.09313, over 5649028.11 frames. ], batch size: 243, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:16:13,863 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=836133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:16:17,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=836136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:16:18,526 INFO [train.py:968] (0/2) Epoch 19, batch 13550, giga_loss[loss=0.2545, simple_loss=0.3377, pruned_loss=0.08566, over 28596.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3362, pruned_loss=0.0923, over 5649071.01 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3424, pruned_loss=0.09117, over 5736660.18 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3376, pruned_loss=0.09322, over 5636703.30 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:16:32,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.298e+02 1.387e+03 1.751e+03 2.615e+03 1.279e+04, threshold=3.501e+03, percent-clipped=20.0
+2023-03-09 20:16:48,584 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=836165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:17:18,208 INFO [train.py:968] (0/2) Epoch 19, batch 13600, giga_loss[loss=0.3101, simple_loss=0.3741, pruned_loss=0.1231, over 27647.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3376, pruned_loss=0.09298, over 5648858.94 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3422, pruned_loss=0.09111, over 5737416.75 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3388, pruned_loss=0.09381, over 5636264.13 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:17:27,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5762, 1.8266, 1.7518, 1.4697], device='cuda:0'), covar=tensor([0.2451, 0.1805, 0.1648, 0.2108], device='cuda:0'), in_proj_covar=tensor([0.1894, 0.1810, 0.1741, 0.1877], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 20:18:11,165 INFO [train.py:968] (0/2) Epoch 19, batch 13650, libri_loss[loss=0.2749, simple_loss=0.3516, pruned_loss=0.09913, over 28150.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3393, pruned_loss=0.09257, over 5655138.47 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3415, pruned_loss=0.09076, over 5742472.03 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3408, pruned_loss=0.09361, over 5637987.71 frames. ], batch size: 116, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:18:28,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.951e+02 1.397e+03 1.874e+03 2.361e+03 5.079e+03, threshold=3.747e+03, percent-clipped=6.0
+2023-03-09 20:18:54,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=836275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:19:10,172 INFO [train.py:968] (0/2) Epoch 19, batch 13700, libri_loss[loss=0.2249, simple_loss=0.3089, pruned_loss=0.07043, over 29535.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3394, pruned_loss=0.09207, over 5659254.21 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3411, pruned_loss=0.09067, over 5735291.17 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3408, pruned_loss=0.09305, over 5648869.65 frames. ], batch size: 80, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:19:28,123 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=836303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:20:16,955 INFO [train.py:968] (0/2) Epoch 19, batch 13750, giga_loss[loss=0.2585, simple_loss=0.3334, pruned_loss=0.09186, over 27660.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3388, pruned_loss=0.09181, over 5652937.88 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3409, pruned_loss=0.09058, over 5726645.31 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3402, pruned_loss=0.09268, over 5651901.62 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:20:32,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.390e+02 1.356e+03 1.727e+03 2.456e+03 5.022e+03, threshold=3.454e+03, percent-clipped=5.0
+2023-03-09 20:21:19,094 INFO [train.py:968] (0/2) Epoch 19, batch 13800, giga_loss[loss=0.226, simple_loss=0.3062, pruned_loss=0.07292, over 24453.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3361, pruned_loss=0.08981, over 5652667.92 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3405, pruned_loss=0.09054, over 5720833.80 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3376, pruned_loss=0.09056, over 5655210.94 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:21:35,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5027, 3.3099, 1.5267, 1.6320], device='cuda:0'), covar=tensor([0.0941, 0.0326, 0.0941, 0.1292], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0543, 0.0372, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 20:21:51,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=836418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:21:54,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=836421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:22:15,780 INFO [train.py:968] (0/2) Epoch 19, batch 13850, giga_loss[loss=0.2374, simple_loss=0.321, pruned_loss=0.07695, over 28992.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3351, pruned_loss=0.08788, over 5656692.44 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.34, pruned_loss=0.0903, over 5717293.79 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3366, pruned_loss=0.08866, over 5659878.96 frames. ], batch size: 285, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:22:28,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=836450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:22:29,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.936e+02 1.280e+03 1.725e+03 2.343e+03 7.339e+03, threshold=3.449e+03, percent-clipped=11.0
+2023-03-09 20:22:34,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3137, 1.5706, 1.4996, 1.3898], device='cuda:0'), covar=tensor([0.2570, 0.1662, 0.1665, 0.1923], device='cuda:0'), in_proj_covar=tensor([0.1889, 0.1803, 0.1729, 0.1874], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 20:23:17,161 INFO [train.py:968] (0/2) Epoch 19, batch 13900, giga_loss[loss=0.29, simple_loss=0.3435, pruned_loss=0.1183, over 28993.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3326, pruned_loss=0.08765, over 5657470.33 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3388, pruned_loss=0.08981, over 5722843.17 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3347, pruned_loss=0.08865, over 5653164.29 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:24:13,825 INFO [train.py:968] (0/2) Epoch 19, batch 13950, giga_loss[loss=0.2681, simple_loss=0.3399, pruned_loss=0.09814, over 27761.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3314, pruned_loss=0.08766, over 5653628.52 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3389, pruned_loss=0.09007, over 5709363.17 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3329, pruned_loss=0.08819, over 5659193.73 frames. ], batch size: 474, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:24:30,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.495e+02 1.293e+03 1.735e+03 2.349e+03 7.948e+03, threshold=3.470e+03, percent-clipped=8.0
+2023-03-09 20:25:14,248 INFO [train.py:968] (0/2) Epoch 19, batch 14000, libri_loss[loss=0.2616, simple_loss=0.3428, pruned_loss=0.09019, over 27949.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3301, pruned_loss=0.08707, over 5658883.67 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3387, pruned_loss=0.09004, over 5711522.16 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3313, pruned_loss=0.08744, over 5660139.28 frames. ], batch size: 116, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:25:32,017 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=836605.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 20:26:03,720 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-09 20:26:07,604 INFO [train.py:968] (0/2) Epoch 19, batch 14050, giga_loss[loss=0.2784, simple_loss=0.3543, pruned_loss=0.1012, over 28175.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3311, pruned_loss=0.08702, over 5659163.64 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.338, pruned_loss=0.08969, over 5720124.11 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3323, pruned_loss=0.08754, over 5649608.98 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:26:20,448 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6723, 2.0968, 1.7071, 1.1121], device='cuda:0'), covar=tensor([0.4272, 0.2932, 0.3149, 0.4794], device='cuda:0'), in_proj_covar=tensor([0.1703, 0.1607, 0.1574, 0.1400], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 20:26:21,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.245e+02 1.416e+03 1.689e+03 2.313e+03 6.617e+03, threshold=3.378e+03, percent-clipped=8.0
+2023-03-09 20:26:30,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5098, 1.7004, 1.2047, 1.3362], device='cuda:0'), covar=tensor([0.1018, 0.0657, 0.1103, 0.1159], device='cuda:0'), in_proj_covar=tensor([0.0381, 0.0440, 0.0507, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 20:26:56,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2056, 1.1412, 3.6429, 3.0897], device='cuda:0'), covar=tensor([0.1650, 0.2950, 0.0443, 0.1452], device='cuda:0'), in_proj_covar=tensor([0.0738, 0.0637, 0.0933, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 20:26:57,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=836678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:27:11,230 INFO [train.py:968] (0/2) Epoch 19, batch 14100, giga_loss[loss=0.2217, simple_loss=0.3106, pruned_loss=0.06639, over 28957.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3335, pruned_loss=0.08757, over 5653698.34 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3378, pruned_loss=0.08958, over 5714588.70 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3345, pruned_loss=0.08803, over 5649624.72 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:27:34,483 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=836704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:28:24,400 INFO [train.py:968] (0/2) Epoch 19, batch 14150, giga_loss[loss=0.3329, simple_loss=0.3799, pruned_loss=0.1429, over 26870.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3295, pruned_loss=0.08488, over 5663489.94 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3377, pruned_loss=0.08956, over 5715561.76 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3303, pruned_loss=0.08525, over 5659279.10 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:28:43,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.640e+02 1.473e+03 1.873e+03 2.595e+03 4.676e+03, threshold=3.745e+03, percent-clipped=9.0
+2023-03-09 20:29:28,081 INFO [train.py:968] (0/2) Epoch 19, batch 14200, giga_loss[loss=0.2352, simple_loss=0.3172, pruned_loss=0.07661, over 29046.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3323, pruned_loss=0.08704, over 5669306.79 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3381, pruned_loss=0.08992, over 5711507.55 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3325, pruned_loss=0.08692, over 5667433.63 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:30:13,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=836821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:30:18,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=836824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:30:23,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6424, 3.0800, 2.6204, 2.0529], device='cuda:0'), covar=tensor([0.2432, 0.1524, 0.1820, 0.2329], device='cuda:0'), in_proj_covar=tensor([0.1886, 0.1805, 0.1736, 0.1880], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 20:30:32,820 INFO [train.py:968] (0/2) Epoch 19, batch 14250, giga_loss[loss=0.2873, simple_loss=0.3798, pruned_loss=0.09736, over 28552.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3355, pruned_loss=0.08751, over 5675729.08 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3376, pruned_loss=0.08971, over 5716002.72 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.336, pruned_loss=0.08753, over 5668874.04 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:30:51,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.616e+02 1.367e+03 1.827e+03 2.260e+03 4.202e+03, threshold=3.653e+03, percent-clipped=5.0
+2023-03-09 20:30:54,177 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=836853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:31:03,466 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 20:31:26,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4020, 1.6077, 1.5005, 1.2652], device='cuda:0'), covar=tensor([0.2647, 0.2185, 0.1807, 0.2309], device='cuda:0'), in_proj_covar=tensor([0.1882, 0.1798, 0.1732, 0.1876], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 20:31:32,824 INFO [train.py:968] (0/2) Epoch 19, batch 14300, giga_loss[loss=0.2689, simple_loss=0.3542, pruned_loss=0.09175, over 28300.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3386, pruned_loss=0.08697, over 5679259.55 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3376, pruned_loss=0.08973, over 5722116.21 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3389, pruned_loss=0.08689, over 5666929.32 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:32:33,884 INFO [train.py:968] (0/2) Epoch 19, batch 14350, giga_loss[loss=0.2551, simple_loss=0.3403, pruned_loss=0.08489, over 28372.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3393, pruned_loss=0.08675, over 5675735.56 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3368, pruned_loss=0.08948, over 5725551.91 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3403, pruned_loss=0.0868, over 5661319.26 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:32:50,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.402e+02 1.472e+03 1.908e+03 3.207e+03 8.895e+03, threshold=3.817e+03, percent-clipped=19.0
+2023-03-09 20:33:09,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4140, 2.0048, 1.4602, 0.6331], device='cuda:0'), covar=tensor([0.4427, 0.2719, 0.4014, 0.5766], device='cuda:0'), in_proj_covar=tensor([0.1702, 0.1603, 0.1572, 0.1397], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 20:33:23,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=836980.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 20:33:30,926 INFO [train.py:968] (0/2) Epoch 19, batch 14400, giga_loss[loss=0.2758, simple_loss=0.3488, pruned_loss=0.1014, over 29017.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3387, pruned_loss=0.08558, over 5676295.67 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3367, pruned_loss=0.08946, over 5721457.43 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3396, pruned_loss=0.08556, over 5667063.57 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:34:35,446 INFO [train.py:968] (0/2) Epoch 19, batch 14450, giga_loss[loss=0.3069, simple_loss=0.3714, pruned_loss=0.1212, over 28766.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.34, pruned_loss=0.08722, over 5673941.72 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3367, pruned_loss=0.08951, over 5722118.46 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3408, pruned_loss=0.08709, over 5665154.91 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:34:54,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.355e+02 1.350e+03 1.797e+03 2.228e+03 5.891e+03, threshold=3.594e+03, percent-clipped=1.0
+2023-03-09 20:35:17,919 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:35:26,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=837079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:35:35,178 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 20:35:37,134 INFO [train.py:968] (0/2) Epoch 19, batch 14500, giga_loss[loss=0.2986, simple_loss=0.3556, pruned_loss=0.1209, over 26881.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3385, pruned_loss=0.08793, over 5684293.94 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3359, pruned_loss=0.0891, over 5726964.88 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3401, pruned_loss=0.08813, over 5670832.15 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:36:27,917 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837123.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 20:36:31,512 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=837126.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 20:36:47,108 INFO [train.py:968] (0/2) Epoch 19, batch 14550, giga_loss[loss=0.2856, simple_loss=0.3519, pruned_loss=0.1097, over 28705.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3399, pruned_loss=0.08947, over 5677314.67 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.336, pruned_loss=0.08914, over 5713357.22 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3411, pruned_loss=0.08958, over 5677916.17 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:36:52,757 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:37:08,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.470e+02 1.429e+03 1.840e+03 2.772e+03 4.947e+03, threshold=3.680e+03, percent-clipped=12.0
+2023-03-09 20:37:14,910 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837155.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 20:37:48,111 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:38:10,754 INFO [train.py:968] (0/2) Epoch 19, batch 14600, giga_loss[loss=0.2503, simple_loss=0.3234, pruned_loss=0.08856, over 27642.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3356, pruned_loss=0.08792, over 5670532.29 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3354, pruned_loss=0.08883, over 5713132.25 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3372, pruned_loss=0.08829, over 5670243.46 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:38:59,784 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:39:02,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=837225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:39:17,404 INFO [train.py:968] (0/2) Epoch 19, batch 14650, giga_loss[loss=0.2209, simple_loss=0.314, pruned_loss=0.06387, over 28979.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3329, pruned_loss=0.08567, over 5673193.77 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3355, pruned_loss=0.0889, over 5713702.98 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.334, pruned_loss=0.08585, over 5671760.58 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:39:34,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.796e+02 1.209e+03 1.539e+03 1.980e+03 4.484e+03, threshold=3.078e+03, percent-clipped=2.0
+2023-03-09 20:39:39,273 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:40:29,698 INFO [train.py:968] (0/2) Epoch 19, batch 14700, giga_loss[loss=0.2718, simple_loss=0.3402, pruned_loss=0.1016, over 27622.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3305, pruned_loss=0.08488, over 5672843.73 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3354, pruned_loss=0.08891, over 5714932.31 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3314, pruned_loss=0.08497, over 5670207.82 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:41:05,177 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:41:32,906 INFO [train.py:968] (0/2) Epoch 19, batch 14750, giga_loss[loss=0.2808, simple_loss=0.3633, pruned_loss=0.0991, over 28673.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3343, pruned_loss=0.087, over 5671949.25 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3352, pruned_loss=0.08889, over 5717716.22 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3352, pruned_loss=0.08706, over 5666857.50 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:41:51,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.415e+02 1.509e+03 1.997e+03 3.066e+03 6.617e+03, threshold=3.994e+03, percent-clipped=23.0
+2023-03-09 20:42:14,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-09 20:42:33,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4774, 1.7662, 1.4674, 1.5323], device='cuda:0'), covar=tensor([0.2794, 0.2456, 0.2795, 0.2331], device='cuda:0'), in_proj_covar=tensor([0.1457, 0.1051, 0.1293, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 20:42:35,170 INFO [train.py:968] (0/2) Epoch 19, batch 14800, giga_loss[loss=0.2661, simple_loss=0.3394, pruned_loss=0.09647, over 28739.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3358, pruned_loss=0.08863, over 5678595.94 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3347, pruned_loss=0.08877, over 5718543.10 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3369, pruned_loss=0.08876, over 5672920.87 frames. ], batch size: 243, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:43:34,369 INFO [train.py:968] (0/2) Epoch 19, batch 14850, giga_loss[loss=0.315, simple_loss=0.3816, pruned_loss=0.1242, over 28883.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3357, pruned_loss=0.08973, over 5675230.06 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3347, pruned_loss=0.08891, over 5715053.55 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3366, pruned_loss=0.0897, over 5672037.01 frames. ], batch size: 284, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:43:36,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 20:43:47,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=837447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:43:54,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.604e+02 1.346e+03 1.667e+03 2.296e+03 6.123e+03, threshold=3.334e+03, percent-clipped=5.0
+2023-03-09 20:44:16,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6534, 4.3739, 1.6978, 1.8058], device='cuda:0'), covar=tensor([0.0940, 0.0294, 0.0928, 0.1282], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0543, 0.0374, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-09 20:44:35,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5169, 1.4185, 1.6575, 1.2054], device='cuda:0'), covar=tensor([0.1963, 0.3231, 0.1568, 0.1746], device='cuda:0'), in_proj_covar=tensor([0.0879, 0.0691, 0.0923, 0.0825], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 20:44:35,448 INFO [train.py:968] (0/2) Epoch 19, batch 14900, giga_loss[loss=0.2579, simple_loss=0.3385, pruned_loss=0.08866, over 28905.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3353, pruned_loss=0.09001, over 5684455.65 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3342, pruned_loss=0.08883, over 5719606.48 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3365, pruned_loss=0.09008, over 5676853.31 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:45:14,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=837517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:45:40,080 INFO [train.py:968] (0/2) Epoch 19, batch 14950, giga_loss[loss=0.2806, simple_loss=0.3541, pruned_loss=0.1036, over 27596.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3365, pruned_loss=0.09019, over 5684587.15 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.334, pruned_loss=0.08877, over 5726070.34 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3377, pruned_loss=0.09033, over 5671205.94 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:45:49,437 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=837547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:45:55,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.663e+02 1.499e+03 2.026e+03 3.083e+03 1.004e+04, threshold=4.052e+03, percent-clipped=19.0
+2023-03-09 20:46:45,331 INFO [train.py:968] (0/2) Epoch 19, batch 15000, giga_loss[loss=0.2457, simple_loss=0.3103, pruned_loss=0.09058, over 24353.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3391, pruned_loss=0.09085, over 5677732.63 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3341, pruned_loss=0.08909, over 5719455.01 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3401, pruned_loss=0.09072, over 5671318.78 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:46:45,335 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 20:46:52,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2801, 1.8473, 1.4462, 0.4588], device='cuda:0'), covar=tensor([0.4890, 0.3387, 0.4794, 0.6575], device='cuda:0'), in_proj_covar=tensor([0.1704, 0.1603, 0.1577, 0.1404], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 20:46:54,832 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.198, simple_loss=0.2981, pruned_loss=0.04895, over 944034.00 frames. 
+2023-03-09 20:46:54,832 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 20:46:57,673 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:47:00,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=837593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:47:45,294 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:47:54,190 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:47:56,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9355, 1.1199, 1.1011, 0.8963], device='cuda:0'), covar=tensor([0.1970, 0.2033, 0.1226, 0.1730], device='cuda:0'), in_proj_covar=tensor([0.1883, 0.1794, 0.1727, 0.1868], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 20:48:11,166 INFO [train.py:968] (0/2) Epoch 19, batch 15050, giga_loss[loss=0.2762, simple_loss=0.3474, pruned_loss=0.1025, over 28920.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3381, pruned_loss=0.09013, over 5671764.87 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3339, pruned_loss=0.08906, over 5723807.77 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3392, pruned_loss=0.0901, over 5662163.39 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:48:32,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.227e+02 1.488e+03 2.064e+03 3.086e+03 6.228e+03, threshold=4.128e+03, percent-clipped=11.0
+2023-03-09 20:48:46,094 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:48:51,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=837663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:49:24,237 INFO [train.py:968] (0/2) Epoch 19, batch 15100, giga_loss[loss=0.2393, simple_loss=0.3195, pruned_loss=0.07957, over 28644.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3346, pruned_loss=0.08906, over 5660839.05 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3337, pruned_loss=0.0889, over 5716707.33 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3357, pruned_loss=0.08918, over 5659021.13 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:49:29,474 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:49:31,241 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:49:31,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=837692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:49:34,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=837693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:49:50,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5161, 1.8224, 1.6759, 1.4826], device='cuda:0'), covar=tensor([0.2753, 0.1937, 0.1828, 0.2158], device='cuda:0'), in_proj_covar=tensor([0.1883, 0.1792, 0.1726, 0.1868], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 20:50:07,476 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:50:16,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:50:35,224 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2827, 1.8124, 1.2716, 0.5376], device='cuda:0'), covar=tensor([0.4671, 0.2552, 0.4315, 0.5787], device='cuda:0'), in_proj_covar=tensor([0.1715, 0.1611, 0.1586, 0.1412], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 20:50:35,722 INFO [train.py:968] (0/2) Epoch 19, batch 15150, giga_loss[loss=0.2711, simple_loss=0.3433, pruned_loss=0.09947, over 28073.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3286, pruned_loss=0.08634, over 5659111.35 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3339, pruned_loss=0.089, over 5715455.79 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3293, pruned_loss=0.08635, over 5658483.49 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:50:50,968 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:50:54,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.472e+02 1.526e+03 1.875e+03 2.583e+03 6.563e+03, threshold=3.750e+03, percent-clipped=5.0
+2023-03-09 20:51:33,496 INFO [train.py:968] (0/2) Epoch 19, batch 15200, giga_loss[loss=0.2572, simple_loss=0.329, pruned_loss=0.09271, over 28320.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3285, pruned_loss=0.08651, over 5657497.82 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3335, pruned_loss=0.08884, over 5713079.34 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3292, pruned_loss=0.0866, over 5657997.21 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:51:48,497 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-09 20:52:28,574 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:52:30,529 INFO [train.py:968] (0/2) Epoch 19, batch 15250, giga_loss[loss=0.2477, simple_loss=0.3237, pruned_loss=0.08584, over 28955.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3303, pruned_loss=0.08807, over 5662035.14 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3334, pruned_loss=0.08891, over 5718235.91 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3309, pruned_loss=0.08806, over 5656309.97 frames. ], batch size: 106, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:52:31,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=837838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:52:51,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.304e+02 1.460e+03 2.017e+03 2.695e+03 5.766e+03, threshold=4.034e+03, percent-clipped=6.0
+2023-03-09 20:53:08,000 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:53:29,819 INFO [train.py:968] (0/2) Epoch 19, batch 15300, giga_loss[loss=0.2297, simple_loss=0.3159, pruned_loss=0.07181, over 28963.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3285, pruned_loss=0.08661, over 5667133.14 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3335, pruned_loss=0.08895, over 5722002.96 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3288, pruned_loss=0.08652, over 5657504.61 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:54:31,479 INFO [train.py:968] (0/2) Epoch 19, batch 15350, giga_loss[loss=0.2186, simple_loss=0.3066, pruned_loss=0.06532, over 28817.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3274, pruned_loss=0.08519, over 5657721.65 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3334, pruned_loss=0.08884, over 5716679.08 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3276, pruned_loss=0.08514, over 5652919.67 frames. ], batch size: 119, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:54:48,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.425e+02 1.247e+03 1.717e+03 2.518e+03 7.746e+03, threshold=3.434e+03, percent-clipped=6.0
+2023-03-09 20:55:01,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6212, 1.7441, 1.9117, 1.4547], device='cuda:0'), covar=tensor([0.1908, 0.2557, 0.1520, 0.1889], device='cuda:0'), in_proj_covar=tensor([0.0878, 0.0688, 0.0922, 0.0824], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 20:55:39,208 INFO [train.py:968] (0/2) Epoch 19, batch 15400, giga_loss[loss=0.2394, simple_loss=0.3171, pruned_loss=0.08088, over 28819.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3261, pruned_loss=0.08514, over 5662314.03 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3331, pruned_loss=0.08884, over 5722101.28 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3263, pruned_loss=0.085, over 5652550.57 frames. ], batch size: 119, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:55:54,476 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-838000.pt
+2023-03-09 20:56:01,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:56:22,928 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-09 20:56:44,256 INFO [train.py:968] (0/2) Epoch 19, batch 15450, giga_loss[loss=0.2288, simple_loss=0.3149, pruned_loss=0.07129, over 28971.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3262, pruned_loss=0.08462, over 5662279.82 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3334, pruned_loss=0.08889, over 5722995.21 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3259, pruned_loss=0.08438, over 5652244.58 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:56:52,754 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:57:08,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.225e+02 1.351e+03 1.766e+03 2.394e+03 5.120e+03, threshold=3.532e+03, percent-clipped=3.0
+2023-03-09 20:57:22,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:57:53,683 INFO [train.py:968] (0/2) Epoch 19, batch 15500, giga_loss[loss=0.249, simple_loss=0.3267, pruned_loss=0.08563, over 28896.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.326, pruned_loss=0.08432, over 5658253.74 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3333, pruned_loss=0.08881, over 5724412.48 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3258, pruned_loss=0.08413, over 5648107.56 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:57:56,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:58:37,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:58:55,616 INFO [train.py:968] (0/2) Epoch 19, batch 15550, giga_loss[loss=0.2334, simple_loss=0.3139, pruned_loss=0.07649, over 28916.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3266, pruned_loss=0.08548, over 5662666.80 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3319, pruned_loss=0.08807, over 5728867.93 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3275, pruned_loss=0.08588, over 5648150.36 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:59:08,613 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=838147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:59:12,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=838150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:59:15,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.222e+02 1.308e+03 1.818e+03 2.548e+03 1.165e+04, threshold=3.636e+03, percent-clipped=9.0
+2023-03-09 20:59:40,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3998, 1.8974, 1.4046, 0.7174], device='cuda:0'), covar=tensor([0.5583, 0.2849, 0.3684, 0.5729], device='cuda:0'), in_proj_covar=tensor([0.1713, 0.1616, 0.1590, 0.1414], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 20:59:49,026 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=838179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:59:59,898 INFO [train.py:968] (0/2) Epoch 19, batch 15600, giga_loss[loss=0.3553, simple_loss=0.4195, pruned_loss=0.1456, over 28953.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3257, pruned_loss=0.0848, over 5665013.80 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3317, pruned_loss=0.08794, over 5727908.36 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3265, pruned_loss=0.08519, over 5653156.14 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:00:50,525 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=838233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:00:54,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=838236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:00:55,678 INFO [train.py:968] (0/2) Epoch 19, batch 15650, libri_loss[loss=0.2524, simple_loss=0.3321, pruned_loss=0.08638, over 29527.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3273, pruned_loss=0.08365, over 5677811.99 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3317, pruned_loss=0.08797, over 5733150.17 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3278, pruned_loss=0.08384, over 5661861.05 frames. ], batch size: 81, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:01:15,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.650e+02 1.240e+03 1.510e+03 1.961e+03 4.865e+03, threshold=3.021e+03, percent-clipped=1.0
+2023-03-09 21:01:20,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8174, 1.9680, 1.4209, 1.6720], device='cuda:0'), covar=tensor([0.0855, 0.0538, 0.0976, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0441, 0.0511, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:01:25,713 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=838265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:01:27,801 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=838267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:01:30,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=838270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:01:55,352 INFO [train.py:968] (0/2) Epoch 19, batch 15700, giga_loss[loss=0.2789, simple_loss=0.3634, pruned_loss=0.09717, over 28691.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3298, pruned_loss=0.08466, over 5674350.54 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3314, pruned_loss=0.08779, over 5739180.02 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3304, pruned_loss=0.08486, over 5654239.42 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:02:10,380 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=838299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:02:57,062 INFO [train.py:968] (0/2) Epoch 19, batch 15750, giga_loss[loss=0.2552, simple_loss=0.3335, pruned_loss=0.08846, over 27564.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3319, pruned_loss=0.08557, over 5675070.73 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3314, pruned_loss=0.0878, over 5741412.45 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3323, pruned_loss=0.08569, over 5656118.06 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:03:15,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.083e+02 1.542e+03 1.914e+03 2.645e+03 9.852e+03, threshold=3.829e+03, percent-clipped=18.0
+2023-03-09 21:03:46,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7285, 1.9815, 1.3659, 1.5906], device='cuda:0'), covar=tensor([0.0956, 0.0577, 0.0950, 0.1156], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0442, 0.0512, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:03:55,501 INFO [train.py:968] (0/2) Epoch 19, batch 15800, giga_loss[loss=0.2441, simple_loss=0.3338, pruned_loss=0.07715, over 28514.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.331, pruned_loss=0.08477, over 5689924.65 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3313, pruned_loss=0.08767, over 5746135.82 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3315, pruned_loss=0.08491, over 5668872.96 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 21:04:29,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4142, 1.5969, 1.1942, 1.1913], device='cuda:0'), covar=tensor([0.0859, 0.0399, 0.0973, 0.1042], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0442, 0.0512, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:04:32,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:04:56,008 INFO [train.py:968] (0/2) Epoch 19, batch 15850, giga_loss[loss=0.2515, simple_loss=0.3373, pruned_loss=0.08289, over 28682.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3285, pruned_loss=0.08304, over 5694731.76 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3309, pruned_loss=0.08747, over 5746177.54 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3293, pruned_loss=0.08327, over 5677051.29 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 21:05:01,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:05:12,659 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-09 21:05:19,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.667e+02 1.525e+03 2.037e+03 3.485e+03 1.448e+04, threshold=4.073e+03, percent-clipped=17.0
+2023-03-09 21:05:40,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2308, 0.8813, 0.9963, 1.3976], device='cuda:0'), covar=tensor([0.0778, 0.0367, 0.0354, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0116, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0067, 0.0061, 0.0104], device='cuda:0')
+2023-03-09 21:06:01,696 INFO [train.py:968] (0/2) Epoch 19, batch 15900, giga_loss[loss=0.2595, simple_loss=0.343, pruned_loss=0.08804, over 28925.00 frames. ], tot_loss[loss=0.247, simple_loss=0.328, pruned_loss=0.08298, over 5692941.45 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3307, pruned_loss=0.08736, over 5747751.79 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3287, pruned_loss=0.08321, over 5676780.16 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 21:06:06,242 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-09 21:06:59,772 INFO [train.py:968] (0/2) Epoch 19, batch 15950, giga_loss[loss=0.2865, simple_loss=0.3551, pruned_loss=0.109, over 28948.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3272, pruned_loss=0.08344, over 5682267.38 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3312, pruned_loss=0.08758, over 5739843.69 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3272, pruned_loss=0.08327, over 5675062.35 frames. ], batch size: 227, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 21:07:05,631 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:07:21,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.304e+02 1.406e+03 1.917e+03 2.547e+03 1.404e+04, threshold=3.834e+03, percent-clipped=5.0
+2023-03-09 21:07:27,232 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=838561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:07:29,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=838564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:07:54,711 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=838584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:07:56,931 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-09 21:07:59,144 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=838587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:07:59,454 INFO [train.py:968] (0/2) Epoch 19, batch 16000, giga_loss[loss=0.2615, simple_loss=0.3458, pruned_loss=0.08855, over 29081.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3281, pruned_loss=0.08403, over 5678276.29 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3308, pruned_loss=0.08735, over 5743005.87 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3283, pruned_loss=0.08399, over 5667939.67 frames. ], batch size: 285, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:08:03,265 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 21:08:07,234 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=838593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:08:21,776 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:08:34,148 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=838616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:08:42,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 21:08:56,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3650, 1.8425, 1.6423, 1.5780], device='cuda:0'), covar=tensor([0.1964, 0.1762, 0.2153, 0.1898], device='cuda:0'), in_proj_covar=tensor([0.0446, 0.0721, 0.0686, 0.0659], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 21:09:03,080 INFO [train.py:968] (0/2) Epoch 19, batch 16050, giga_loss[loss=0.2282, simple_loss=0.3158, pruned_loss=0.07025, over 28998.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3304, pruned_loss=0.08536, over 5686449.26 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3306, pruned_loss=0.08732, over 5747189.80 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3308, pruned_loss=0.08532, over 5673045.95 frames. ], batch size: 120, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:09:26,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.683e+02 1.333e+03 1.656e+03 2.490e+03 9.079e+03, threshold=3.312e+03, percent-clipped=8.0
+2023-03-09 21:09:56,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2219, 1.3366, 1.2871, 0.9365], device='cuda:0'), covar=tensor([0.0967, 0.0467, 0.0993, 0.0987], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0443, 0.0512, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:10:10,366 INFO [train.py:968] (0/2) Epoch 19, batch 16100, giga_loss[loss=0.2534, simple_loss=0.3404, pruned_loss=0.08324, over 28899.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3311, pruned_loss=0.08622, over 5680308.44 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3304, pruned_loss=0.08733, over 5748865.23 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3316, pruned_loss=0.08612, over 5666886.34 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:10:20,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5160, 1.8002, 1.4032, 1.6805], device='cuda:0'), covar=tensor([0.2578, 0.2489, 0.2885, 0.2173], device='cuda:0'), in_proj_covar=tensor([0.1450, 0.1049, 0.1292, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 21:10:26,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2727, 1.5660, 1.4276, 1.1751], device='cuda:0'), covar=tensor([0.2495, 0.2153, 0.1499, 0.2032], device='cuda:0'), in_proj_covar=tensor([0.1890, 0.1798, 0.1731, 0.1876], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 21:10:40,659 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 21:11:04,596 INFO [train.py:968] (0/2) Epoch 19, batch 16150, giga_loss[loss=0.2812, simple_loss=0.3406, pruned_loss=0.1109, over 28423.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3339, pruned_loss=0.08786, over 5675423.63 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3298, pruned_loss=0.08712, over 5736558.40 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3349, pruned_loss=0.08798, over 5673096.55 frames. ], batch size: 71, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:11:27,922 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.495e+02 1.339e+03 1.805e+03 2.370e+03 6.120e+03, threshold=3.611e+03, percent-clipped=8.0
+2023-03-09 21:11:30,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3314, 1.3951, 1.3235, 1.4729], device='cuda:0'), covar=tensor([0.0783, 0.0349, 0.0346, 0.0903], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0067, 0.0061, 0.0104], device='cuda:0')
+2023-03-09 21:11:54,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2773, 1.3836, 1.3674, 1.3020], device='cuda:0'), covar=tensor([0.2291, 0.1879, 0.1800, 0.2061], device='cuda:0'), in_proj_covar=tensor([0.1883, 0.1790, 0.1725, 0.1869], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 21:12:02,402 INFO [train.py:968] (0/2) Epoch 19, batch 16200, giga_loss[loss=0.262, simple_loss=0.3458, pruned_loss=0.08912, over 29021.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3358, pruned_loss=0.08778, over 5684433.88 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3297, pruned_loss=0.08703, over 5741051.87 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3368, pruned_loss=0.08799, over 5677147.30 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:12:05,416 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:12:08,134 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 21:12:21,505 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:13:11,345 INFO [train.py:968] (0/2) Epoch 19, batch 16250, giga_loss[loss=0.2507, simple_loss=0.3346, pruned_loss=0.08346, over 28666.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3364, pruned_loss=0.08805, over 5683488.51 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3297, pruned_loss=0.08699, over 5741841.61 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3373, pruned_loss=0.08826, over 5676794.63 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:13:20,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3200, 1.5972, 1.5636, 1.5207], device='cuda:0'), covar=tensor([0.1598, 0.1491, 0.1569, 0.1418], device='cuda:0'), in_proj_covar=tensor([0.0444, 0.0722, 0.0685, 0.0660], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 21:13:37,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.946e+02 1.396e+03 1.937e+03 2.744e+03 7.690e+03, threshold=3.874e+03, percent-clipped=15.0
+2023-03-09 21:13:53,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4670, 1.5359, 1.1698, 1.1417], device='cuda:0'), covar=tensor([0.0883, 0.0518, 0.1032, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0442, 0.0511, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:14:18,990 INFO [train.py:968] (0/2) Epoch 19, batch 16300, giga_loss[loss=0.2298, simple_loss=0.3165, pruned_loss=0.07157, over 28998.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3343, pruned_loss=0.08692, over 5690789.63 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3293, pruned_loss=0.08682, over 5744399.33 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3354, pruned_loss=0.08726, over 5681956.50 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:14:32,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5888, 1.8238, 1.4694, 1.7425], device='cuda:0'), covar=tensor([0.2659, 0.2564, 0.2922, 0.2316], device='cuda:0'), in_proj_covar=tensor([0.1455, 0.1051, 0.1295, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 21:14:54,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:15:27,774 INFO [train.py:968] (0/2) Epoch 19, batch 16350, giga_loss[loss=0.3009, simple_loss=0.3716, pruned_loss=0.1151, over 28757.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3343, pruned_loss=0.08734, over 5686099.79 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3295, pruned_loss=0.08684, over 5745657.22 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3351, pruned_loss=0.08759, over 5677683.43 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 21:15:50,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.186e+02 1.418e+03 1.950e+03 2.877e+03 1.886e+04, threshold=3.900e+03, percent-clipped=15.0
+2023-03-09 21:16:16,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:16:26,305 INFO [train.py:968] (0/2) Epoch 19, batch 16400, giga_loss[loss=0.2451, simple_loss=0.3251, pruned_loss=0.0826, over 29000.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3326, pruned_loss=0.0868, over 5668732.52 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.329, pruned_loss=0.08665, over 5737508.20 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3339, pruned_loss=0.0872, over 5666156.49 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:16:29,932 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:17:29,507 INFO [train.py:968] (0/2) Epoch 19, batch 16450, giga_loss[loss=0.2642, simple_loss=0.3426, pruned_loss=0.09295, over 28514.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3311, pruned_loss=0.08746, over 5670097.55 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3289, pruned_loss=0.08657, over 5739677.59 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3323, pruned_loss=0.08786, over 5665282.76 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:17:52,059 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.939e+02 1.343e+03 1.809e+03 2.308e+03 5.047e+03, threshold=3.617e+03, percent-clipped=2.0
+2023-03-09 21:17:54,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:17:58,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=839062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:18:30,210 INFO [train.py:968] (0/2) Epoch 19, batch 16500, giga_loss[loss=0.2482, simple_loss=0.3332, pruned_loss=0.08159, over 28487.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3308, pruned_loss=0.08705, over 5670588.80 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.329, pruned_loss=0.08669, over 5734403.52 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3317, pruned_loss=0.08728, over 5670301.24 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:18:34,032 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:19:05,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3768, 1.2706, 3.8778, 3.2896], device='cuda:0'), covar=tensor([0.1536, 0.2753, 0.0396, 0.1532], device='cuda:0'), in_proj_covar=tensor([0.0732, 0.0631, 0.0924, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:19:15,484 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:19:19,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=839125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:19:36,543 INFO [train.py:968] (0/2) Epoch 19, batch 16550, giga_loss[loss=0.2212, simple_loss=0.3075, pruned_loss=0.0674, over 28916.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.33, pruned_loss=0.08563, over 5664747.60 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.329, pruned_loss=0.08668, over 5735408.51 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3307, pruned_loss=0.08583, over 5663433.68 frames. ], batch size: 284, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:19:42,923 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-09 21:19:42,952 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-09 21:19:54,588 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:19:58,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.651e+02 1.370e+03 1.821e+03 2.899e+03 1.175e+04, threshold=3.642e+03, percent-clipped=12.0
+2023-03-09 21:20:04,667 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839161.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:20:08,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=839164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:20:24,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=839177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:20:34,251 INFO [train.py:968] (0/2) Epoch 19, batch 16600, giga_loss[loss=0.2816, simple_loss=0.3594, pruned_loss=0.1019, over 26733.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3309, pruned_loss=0.08453, over 5672175.65 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3291, pruned_loss=0.08668, over 5733891.77 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3314, pruned_loss=0.08465, over 5671199.27 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:21:35,517 INFO [train.py:968] (0/2) Epoch 19, batch 16650, giga_loss[loss=0.2669, simple_loss=0.3489, pruned_loss=0.09241, over 27685.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3323, pruned_loss=0.08391, over 5668209.84 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3289, pruned_loss=0.08661, over 5733810.85 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3329, pruned_loss=0.08405, over 5666950.76 frames. ], batch size: 474, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:21:57,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.078e+02 1.434e+03 1.925e+03 2.548e+03 6.591e+03, threshold=3.851e+03, percent-clipped=6.0
+2023-03-09 21:22:21,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3102, 1.7666, 1.3675, 1.6386], device='cuda:0'), covar=tensor([0.0818, 0.0292, 0.0341, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0116, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:0')
+2023-03-09 21:22:37,010 INFO [train.py:968] (0/2) Epoch 19, batch 16700, giga_loss[loss=0.2723, simple_loss=0.3577, pruned_loss=0.09346, over 28891.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3329, pruned_loss=0.08373, over 5683517.66 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3289, pruned_loss=0.08663, over 5735488.96 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3335, pruned_loss=0.08381, over 5680544.39 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:23:03,294 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:23:06,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=839310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:23:20,129 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:23:25,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=839323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:23:45,732 INFO [train.py:968] (0/2) Epoch 19, batch 16750, giga_loss[loss=0.2592, simple_loss=0.3442, pruned_loss=0.08709, over 28786.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3331, pruned_loss=0.08435, over 5673106.51 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3289, pruned_loss=0.0866, over 5735586.98 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3335, pruned_loss=0.08437, over 5668986.36 frames. ], batch size: 243, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:23:47,892 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:24:06,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:24:10,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.821e+02 1.475e+03 1.948e+03 2.916e+03 6.729e+03, threshold=3.896e+03, percent-clipped=13.0
+2023-03-09 21:24:25,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3444, 3.2753, 1.5143, 1.5028], device='cuda:0'), covar=tensor([0.0952, 0.0377, 0.0894, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0537, 0.0372, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 21:24:25,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=839366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:24:55,562 INFO [train.py:968] (0/2) Epoch 19, batch 16800, giga_loss[loss=0.2344, simple_loss=0.3291, pruned_loss=0.06985, over 28993.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3332, pruned_loss=0.08428, over 5676576.70 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3289, pruned_loss=0.08671, over 5739752.87 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3336, pruned_loss=0.08414, over 5668025.02 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:25:14,640 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-09 21:26:05,625 INFO [train.py:968] (0/2) Epoch 19, batch 16850, giga_loss[loss=0.24, simple_loss=0.3294, pruned_loss=0.07527, over 29174.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3339, pruned_loss=0.08431, over 5679648.50 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3294, pruned_loss=0.08707, over 5742665.16 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.334, pruned_loss=0.08379, over 5668009.11 frames. ], batch size: 200, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:26:32,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.689e+02 1.378e+03 1.918e+03 2.737e+03 9.646e+03, threshold=3.836e+03, percent-clipped=8.0
+2023-03-09 21:27:13,616 INFO [train.py:968] (0/2) Epoch 19, batch 16900, giga_loss[loss=0.2789, simple_loss=0.3633, pruned_loss=0.09726, over 28946.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3343, pruned_loss=0.0844, over 5687612.35 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3289, pruned_loss=0.08684, over 5747773.13 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3349, pruned_loss=0.0841, over 5672172.67 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:27:27,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1943, 1.4755, 1.4576, 1.0579], device='cuda:0'), covar=tensor([0.1542, 0.2514, 0.1350, 0.1640], device='cuda:0'), in_proj_covar=tensor([0.0873, 0.0684, 0.0919, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 21:27:41,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3099, 1.8019, 1.3488, 0.5695], device='cuda:0'), covar=tensor([0.4053, 0.2087, 0.2920, 0.5504], device='cuda:0'), in_proj_covar=tensor([0.1694, 0.1596, 0.1573, 0.1395], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 21:27:44,302 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:27:47,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=839512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:28:21,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=839536.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:28:23,621 INFO [train.py:968] (0/2) Epoch 19, batch 16950, giga_loss[loss=0.2499, simple_loss=0.3386, pruned_loss=0.08064, over 28075.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3381, pruned_loss=0.08647, over 5674469.70 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3288, pruned_loss=0.08682, over 5735327.61 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3389, pruned_loss=0.08624, over 5671374.64 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:28:27,521 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:28:43,463 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3446, 1.8009, 1.4269, 1.5085], device='cuda:0'), covar=tensor([0.0780, 0.0286, 0.0330, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0116, 0.0116, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0067, 0.0061, 0.0104], device='cuda:0')
+2023-03-09 21:28:50,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.791e+02 1.538e+03 1.994e+03 3.081e+03 8.343e+03, threshold=3.989e+03, percent-clipped=14.0
+2023-03-09 21:29:04,319 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8493, 1.9557, 1.6233, 2.2026], device='cuda:0'), covar=tensor([0.2491, 0.2696, 0.3129, 0.2321], device='cuda:0'), in_proj_covar=tensor([0.1450, 0.1046, 0.1291, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 21:29:34,280 INFO [train.py:968] (0/2) Epoch 19, batch 17000, giga_loss[loss=0.2179, simple_loss=0.3082, pruned_loss=0.06376, over 28855.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3375, pruned_loss=0.08622, over 5679461.34 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.329, pruned_loss=0.08689, over 5737125.43 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.338, pruned_loss=0.08597, over 5674734.39 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:29:36,010 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839589.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:30:27,752 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5152, 1.7775, 1.6817, 1.4958], device='cuda:0'), covar=tensor([0.1677, 0.2116, 0.2146, 0.2221], device='cuda:0'), in_proj_covar=tensor([0.0438, 0.0715, 0.0681, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 21:30:27,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4652, 2.1913, 1.4744, 0.7174], device='cuda:0'), covar=tensor([0.5720, 0.2530, 0.4320, 0.6074], device='cuda:0'), in_proj_covar=tensor([0.1685, 0.1588, 0.1569, 0.1391], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 21:30:40,960 INFO [train.py:968] (0/2) Epoch 19, batch 17050, libri_loss[loss=0.2197, simple_loss=0.2998, pruned_loss=0.06984, over 29562.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3352, pruned_loss=0.08543, over 5692406.75 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.329, pruned_loss=0.0869, over 5740587.69 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.336, pruned_loss=0.08519, over 5683498.25 frames. ], batch size: 77, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:31:05,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.559e+02 1.369e+03 1.699e+03 2.193e+03 5.975e+03, threshold=3.398e+03, percent-clipped=2.0
+2023-03-09 21:31:37,811 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839679.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:31:45,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=839682.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 21:31:51,782 INFO [train.py:968] (0/2) Epoch 19, batch 17100, giga_loss[loss=0.2317, simple_loss=0.323, pruned_loss=0.0702, over 27764.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.333, pruned_loss=0.08381, over 5696264.35 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3286, pruned_loss=0.08658, over 5745014.48 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3341, pruned_loss=0.08384, over 5684334.86 frames. ], batch size: 474, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:32:03,807 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-09 21:32:23,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839711.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:32:36,320 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-09 21:32:59,955 INFO [train.py:968] (0/2) Epoch 19, batch 17150, giga_loss[loss=0.2357, simple_loss=0.3189, pruned_loss=0.07626, over 28921.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3306, pruned_loss=0.08183, over 5705973.86 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3283, pruned_loss=0.0864, over 5749450.26 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3318, pruned_loss=0.08193, over 5691752.18 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:33:21,573 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:33:24,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.816e+02 1.307e+03 1.694e+03 2.310e+03 5.280e+03, threshold=3.387e+03, percent-clipped=7.0
+2023-03-09 21:33:55,829 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:34:03,436 INFO [train.py:968] (0/2) Epoch 19, batch 17200, giga_loss[loss=0.2297, simple_loss=0.3185, pruned_loss=0.07039, over 28983.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3333, pruned_loss=0.08424, over 5694483.44 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3284, pruned_loss=0.08642, over 5751814.56 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3342, pruned_loss=0.08427, over 5680586.02 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:35:04,654 INFO [train.py:968] (0/2) Epoch 19, batch 17250, giga_loss[loss=0.2298, simple_loss=0.3099, pruned_loss=0.07485, over 24517.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3367, pruned_loss=0.08636, over 5691766.40 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3282, pruned_loss=0.08635, over 5753870.54 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3377, pruned_loss=0.08644, over 5677693.70 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:35:29,049 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.214e+02 1.433e+03 1.789e+03 2.376e+03 5.933e+03, threshold=3.577e+03, percent-clipped=8.0
+2023-03-09 21:36:03,416 INFO [train.py:968] (0/2) Epoch 19, batch 17300, giga_loss[loss=0.2886, simple_loss=0.3497, pruned_loss=0.1137, over 27549.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3366, pruned_loss=0.08673, over 5688278.85 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3282, pruned_loss=0.0864, over 5753802.69 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3375, pruned_loss=0.08674, over 5676615.20 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:36:05,060 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3387, 1.1645, 4.4799, 3.5278], device='cuda:0'), covar=tensor([0.1741, 0.2886, 0.0433, 0.0984], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0633, 0.0923, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:36:39,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:36:41,277 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:37:02,766 INFO [train.py:968] (0/2) Epoch 19, batch 17350, giga_loss[loss=0.2723, simple_loss=0.3516, pruned_loss=0.09655, over 28262.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3348, pruned_loss=0.08715, over 5674806.50 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3283, pruned_loss=0.08643, over 5744620.07 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3356, pruned_loss=0.08714, over 5671308.67 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:37:25,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.944e+02 1.518e+03 1.960e+03 2.625e+03 5.801e+03, threshold=3.921e+03, percent-clipped=8.0
+2023-03-09 21:37:33,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=839964.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 21:37:54,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7660, 4.8049, 1.8564, 1.8023], device='cuda:0'), covar=tensor([0.0912, 0.0254, 0.0899, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0539, 0.0374, 0.0418], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 21:37:57,032 INFO [train.py:968] (0/2) Epoch 19, batch 17400, giga_loss[loss=0.2555, simple_loss=0.3423, pruned_loss=0.08433, over 28834.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3347, pruned_loss=0.08751, over 5684419.54 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3284, pruned_loss=0.08636, over 5746241.17 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3354, pruned_loss=0.0876, over 5678580.76 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:38:12,352 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-840000.pt
+2023-03-09 21:38:47,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3320, 2.8204, 1.5355, 1.4354], device='cuda:0'), covar=tensor([0.0916, 0.0372, 0.0889, 0.1302], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0538, 0.0373, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 21:38:56,956 INFO [train.py:968] (0/2) Epoch 19, batch 17450, giga_loss[loss=0.2771, simple_loss=0.3652, pruned_loss=0.09451, over 28947.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3412, pruned_loss=0.0918, over 5675866.04 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3285, pruned_loss=0.08648, over 5746778.50 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3419, pruned_loss=0.09181, over 5669389.05 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:39:14,908 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.673e+02 1.517e+03 2.007e+03 2.829e+03 6.616e+03, threshold=4.013e+03, percent-clipped=10.0
+2023-03-09 21:39:41,973 INFO [train.py:968] (0/2) Epoch 19, batch 17500, libri_loss[loss=0.2837, simple_loss=0.3609, pruned_loss=0.1032, over 27726.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3491, pruned_loss=0.09591, over 5690730.83 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3287, pruned_loss=0.0866, over 5748887.97 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3498, pruned_loss=0.09597, over 5682454.37 frames. ], batch size: 115, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:39:58,801 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840107.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:40:01,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840110.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:40:01,897 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 21:40:16,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:40:23,438 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=840135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:40:26,579 INFO [train.py:968] (0/2) Epoch 19, batch 17550, giga_loss[loss=0.2553, simple_loss=0.3195, pruned_loss=0.09558, over 23728.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.352, pruned_loss=0.09767, over 5690603.74 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3289, pruned_loss=0.08672, over 5752145.92 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3528, pruned_loss=0.09783, over 5680067.28 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:40:27,545 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840139.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:40:34,171 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=840147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:40:40,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:40:41,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.750e+02 1.230e+03 1.639e+03 2.242e+03 4.509e+03, threshold=3.277e+03, percent-clipped=4.0
+2023-03-09 21:41:09,419 INFO [train.py:968] (0/2) Epoch 19, batch 17600, giga_loss[loss=0.2608, simple_loss=0.3278, pruned_loss=0.09696, over 28849.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3469, pruned_loss=0.09617, over 5696850.89 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3284, pruned_loss=0.08652, over 5758039.55 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3487, pruned_loss=0.09684, over 5681184.15 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:41:54,126 INFO [train.py:968] (0/2) Epoch 19, batch 17650, giga_loss[loss=0.2173, simple_loss=0.3014, pruned_loss=0.0666, over 28926.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3385, pruned_loss=0.09217, over 5698685.67 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3283, pruned_loss=0.08632, over 5761262.50 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3402, pruned_loss=0.093, over 5682106.16 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:42:15,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.377e+02 1.109e+03 1.466e+03 1.921e+03 6.534e+03, threshold=2.932e+03, percent-clipped=8.0
+2023-03-09 21:42:23,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7614, 1.9852, 1.3253, 1.6440], device='cuda:0'), covar=tensor([0.0984, 0.0746, 0.1116, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0439, 0.0510, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:42:27,582 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:29,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:37,695 INFO [train.py:968] (0/2) Epoch 19, batch 17700, giga_loss[loss=0.238, simple_loss=0.3038, pruned_loss=0.08615, over 28680.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3329, pruned_loss=0.09, over 5695473.99 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.329, pruned_loss=0.08643, over 5760934.06 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3338, pruned_loss=0.0907, over 5680883.22 frames. ], batch size: 92, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:42:43,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:44,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:46,342 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:49,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:52,159 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:43:13,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6053, 1.6317, 1.8519, 1.4350], device='cuda:0'), covar=tensor([0.1748, 0.2506, 0.1447, 0.1700], device='cuda:0'), in_proj_covar=tensor([0.0885, 0.0691, 0.0929, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:0')
+2023-03-09 21:43:15,357 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:43:23,287 INFO [train.py:968] (0/2) Epoch 19, batch 17750, giga_loss[loss=0.2297, simple_loss=0.3119, pruned_loss=0.07375, over 28986.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3264, pruned_loss=0.08758, over 5695533.47 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3289, pruned_loss=0.08635, over 5766864.38 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3273, pruned_loss=0.0883, over 5676142.83 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:43:24,769 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=840340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:43:42,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.388e+02 1.062e+03 1.318e+03 2.028e+03 1.026e+04, threshold=2.636e+03, percent-clipped=12.0
+2023-03-09 21:44:07,327 INFO [train.py:968] (0/2) Epoch 19, batch 17800, giga_loss[loss=0.2251, simple_loss=0.3063, pruned_loss=0.07197, over 28874.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3195, pruned_loss=0.08479, over 5698453.16 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3288, pruned_loss=0.08636, over 5768413.55 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3202, pruned_loss=0.08534, over 5681199.63 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:44:22,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1906, 1.2295, 3.9049, 3.1143], device='cuda:0'), covar=tensor([0.1745, 0.2799, 0.0490, 0.1015], device='cuda:0'), in_proj_covar=tensor([0.0738, 0.0638, 0.0933, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:44:43,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0408, 3.8661, 3.6160, 1.7853], device='cuda:0'), covar=tensor([0.0644, 0.0800, 0.0793, 0.2242], device='cuda:0'), in_proj_covar=tensor([0.1168, 0.1078, 0.0918, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 21:44:49,391 INFO [train.py:968] (0/2) Epoch 19, batch 17850, giga_loss[loss=0.2535, simple_loss=0.3124, pruned_loss=0.09728, over 26650.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3151, pruned_loss=0.08236, over 5702432.85 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3288, pruned_loss=0.0862, over 5770275.79 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3154, pruned_loss=0.08286, over 5685405.53 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:44:49,690 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:44:50,308 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:44:51,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:44:52,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:45:04,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.131e+02 1.118e+03 1.376e+03 1.985e+03 4.266e+03, threshold=2.752e+03, percent-clipped=11.0
+2023-03-09 21:45:12,190 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:45:12,860 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:45:25,507 INFO [train.py:968] (0/2) Epoch 19, batch 17900, libri_loss[loss=0.2881, simple_loss=0.3653, pruned_loss=0.1055, over 29732.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3139, pruned_loss=0.08193, over 5711886.88 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3288, pruned_loss=0.08612, over 5772754.79 frames. ], giga_tot_loss[loss=0.239, simple_loss=0.3135, pruned_loss=0.08222, over 5692865.74 frames. ], batch size: 87, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:45:44,046 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:45:58,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:46:02,960 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7221, 1.8535, 1.9417, 1.5184], device='cuda:0'), covar=tensor([0.1863, 0.2413, 0.1492, 0.1697], device='cuda:0'), in_proj_covar=tensor([0.0888, 0.0691, 0.0931, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:0')
+2023-03-09 21:46:10,490 INFO [train.py:968] (0/2) Epoch 19, batch 17950, giga_loss[loss=0.2368, simple_loss=0.3084, pruned_loss=0.08264, over 28767.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3106, pruned_loss=0.08046, over 5697023.53 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.329, pruned_loss=0.0862, over 5764410.16 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.3096, pruned_loss=0.08049, over 5686503.50 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:46:29,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.674e+02 1.011e+03 1.375e+03 2.112e+03 7.570e+03, threshold=2.749e+03, percent-clipped=8.0
+2023-03-09 21:46:50,355 INFO [train.py:968] (0/2) Epoch 19, batch 18000, giga_loss[loss=0.2208, simple_loss=0.2972, pruned_loss=0.07224, over 28898.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3081, pruned_loss=0.07911, over 5712656.45 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3293, pruned_loss=0.08629, over 5769257.81 frames. ], giga_tot_loss[loss=0.2319, simple_loss=0.3062, pruned_loss=0.07882, over 5697703.53 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:46:50,359 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 21:46:58,871 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2037, simple_loss=0.3094, pruned_loss=0.04904, over 944034.00 frames. 
+2023-03-09 21:46:58,872 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 21:47:01,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4100, 3.2482, 3.0137, 1.9065], device='cuda:0'), covar=tensor([0.0766, 0.0912, 0.0840, 0.1815], device='cuda:0'), in_proj_covar=tensor([0.1171, 0.1084, 0.0922, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 21:47:19,411 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-09 21:47:44,711 INFO [train.py:968] (0/2) Epoch 19, batch 18050, giga_loss[loss=0.2691, simple_loss=0.3199, pruned_loss=0.1092, over 26451.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3062, pruned_loss=0.07863, over 5688875.13 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3295, pruned_loss=0.08626, over 5762139.78 frames. ], giga_tot_loss[loss=0.2304, simple_loss=0.3043, pruned_loss=0.0783, over 5681620.13 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:47:56,535 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:47:58,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:47:59,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3267, 1.2877, 4.0168, 3.2808], device='cuda:0'), covar=tensor([0.1730, 0.2870, 0.0435, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0635, 0.0931, 0.0868], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:48:01,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.704e+02 9.497e+02 1.281e+03 1.932e+03 4.461e+03, threshold=2.561e+03, percent-clipped=10.0
+2023-03-09 21:48:05,745 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:48:08,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:48:24,417 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:48:26,397 INFO [train.py:968] (0/2) Epoch 19, batch 18100, giga_loss[loss=0.2312, simple_loss=0.3085, pruned_loss=0.07699, over 28885.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3029, pruned_loss=0.07705, over 5694704.73 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3291, pruned_loss=0.08604, over 5764352.42 frames. ], giga_tot_loss[loss=0.2275, simple_loss=0.3013, pruned_loss=0.07686, over 5686205.96 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:48:36,413 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:48:52,083 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:48:58,442 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9700, 2.9969, 2.1165, 1.0728], device='cuda:0'), covar=tensor([0.7718, 0.3173, 0.3647, 0.6727], device='cuda:0'), in_proj_covar=tensor([0.1702, 0.1615, 0.1580, 0.1401], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 21:49:10,023 INFO [train.py:968] (0/2) Epoch 19, batch 18150, giga_loss[loss=0.2218, simple_loss=0.291, pruned_loss=0.07628, over 28485.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3008, pruned_loss=0.07622, over 5692823.82 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.33, pruned_loss=0.08629, over 5765180.76 frames. ], giga_tot_loss[loss=0.2245, simple_loss=0.2979, pruned_loss=0.07557, over 5683453.93 frames. ], batch size: 60, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:49:19,070 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=840748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:49:25,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4200, 2.7038, 1.5900, 1.5605], device='cuda:0'), covar=tensor([0.0862, 0.0325, 0.0769, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0538, 0.0374, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 21:49:30,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.189e+02 1.096e+03 1.463e+03 2.466e+03 6.328e+03, threshold=2.926e+03, percent-clipped=21.0
+2023-03-09 21:49:57,283 INFO [train.py:968] (0/2) Epoch 19, batch 18200, giga_loss[loss=0.197, simple_loss=0.2635, pruned_loss=0.06522, over 28466.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2988, pruned_loss=0.07522, over 5681560.71 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3306, pruned_loss=0.0865, over 5760613.95 frames. ], giga_tot_loss[loss=0.2218, simple_loss=0.2951, pruned_loss=0.07423, over 5675996.64 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:50:11,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3520, 2.5192, 2.3128, 2.1989], device='cuda:0'), covar=tensor([0.1597, 0.1889, 0.1826, 0.1811], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0732, 0.0696, 0.0667], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 21:50:38,309 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6306, 4.4630, 4.1961, 2.0378], device='cuda:0'), covar=tensor([0.0514, 0.0736, 0.0746, 0.2018], device='cuda:0'), in_proj_covar=tensor([0.1168, 0.1085, 0.0922, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 21:50:41,603 INFO [train.py:968] (0/2) Epoch 19, batch 18250, giga_loss[loss=0.2144, simple_loss=0.2838, pruned_loss=0.07247, over 28530.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.297, pruned_loss=0.07464, over 5689520.06 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3311, pruned_loss=0.08679, over 5762436.69 frames. ], giga_tot_loss[loss=0.2198, simple_loss=0.2929, pruned_loss=0.07335, over 5681833.04 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:51:01,960 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:51:03,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.355e+02 1.059e+03 1.308e+03 1.757e+03 4.059e+03, threshold=2.616e+03, percent-clipped=2.0
+2023-03-09 21:51:06,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:51:34,022 INFO [train.py:968] (0/2) Epoch 19, batch 18300, giga_loss[loss=0.2752, simple_loss=0.3513, pruned_loss=0.09958, over 28639.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3063, pruned_loss=0.08014, over 5673900.41 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3313, pruned_loss=0.08691, over 5760297.58 frames. ], giga_tot_loss[loss=0.2302, simple_loss=0.3025, pruned_loss=0.07889, over 5668530.29 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:51:35,681 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:52:20,246 INFO [train.py:968] (0/2) Epoch 19, batch 18350, giga_loss[loss=0.3193, simple_loss=0.3795, pruned_loss=0.1295, over 28895.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3205, pruned_loss=0.08769, over 5682123.84 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3317, pruned_loss=0.08711, over 5763315.40 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3168, pruned_loss=0.08647, over 5673810.39 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:52:24,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8127, 1.3766, 1.3006, 1.1087], device='cuda:0'), covar=tensor([0.1996, 0.1110, 0.2188, 0.1591], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0733, 0.0696, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 21:52:38,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.310e+02 1.403e+03 1.850e+03 2.847e+03 8.897e+03, threshold=3.699e+03, percent-clipped=30.0
+2023-03-09 21:52:59,659 INFO [train.py:968] (0/2) Epoch 19, batch 18400, giga_loss[loss=0.284, simple_loss=0.3649, pruned_loss=0.1015, over 28583.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3312, pruned_loss=0.09285, over 5689890.92 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3326, pruned_loss=0.08771, over 5758779.12 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3272, pruned_loss=0.0914, over 5684436.36 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:53:05,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4306, 1.9805, 1.5822, 1.5342], device='cuda:0'), covar=tensor([0.0801, 0.0292, 0.0319, 0.0907], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0061, 0.0104], device='cuda:0')
+2023-03-09 21:53:40,920 INFO [train.py:968] (0/2) Epoch 19, batch 18450, giga_loss[loss=0.3025, simple_loss=0.3589, pruned_loss=0.123, over 23517.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3369, pruned_loss=0.09443, over 5686364.54 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3331, pruned_loss=0.08764, over 5756919.01 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3333, pruned_loss=0.09364, over 5679965.59 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:54:02,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.367e+03 1.709e+03 2.294e+03 4.381e+03, threshold=3.419e+03, percent-clipped=4.0
+2023-03-09 21:54:15,431 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4116, 1.3275, 1.3130, 1.4925], device='cuda:0'), covar=tensor([0.0794, 0.0347, 0.0326, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0067, 0.0061, 0.0104], device='cuda:0')
+2023-03-09 21:54:22,046 INFO [train.py:968] (0/2) Epoch 19, batch 18500, giga_loss[loss=0.3117, simple_loss=0.3846, pruned_loss=0.1194, over 27587.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3402, pruned_loss=0.09473, over 5683868.62 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3331, pruned_loss=0.08759, over 5753324.94 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3375, pruned_loss=0.09436, over 5679920.31 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:54:22,236 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:54:53,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:55:07,069 INFO [train.py:968] (0/2) Epoch 19, batch 18550, giga_loss[loss=0.2605, simple_loss=0.3488, pruned_loss=0.08611, over 28727.00 frames. ], tot_loss[loss=0.267, simple_loss=0.343, pruned_loss=0.09551, over 5684309.30 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3336, pruned_loss=0.08785, over 5756763.78 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3405, pruned_loss=0.09515, over 5676209.12 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:55:28,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.553e+02 1.241e+03 1.654e+03 2.455e+03 8.702e+03, threshold=3.309e+03, percent-clipped=14.0
+2023-03-09 21:55:48,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4232, 1.6864, 1.3543, 1.3666], device='cuda:0'), covar=tensor([0.2525, 0.2513, 0.2840, 0.2222], device='cuda:0'), in_proj_covar=tensor([0.1449, 0.1053, 0.1290, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 21:55:49,785 INFO [train.py:968] (0/2) Epoch 19, batch 18600, giga_loss[loss=0.2745, simple_loss=0.346, pruned_loss=0.1015, over 28496.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.345, pruned_loss=0.09698, over 5675986.83 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3338, pruned_loss=0.08785, over 5758417.80 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3431, pruned_loss=0.09699, over 5665127.02 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:56:34,241 INFO [train.py:968] (0/2) Epoch 19, batch 18650, giga_loss[loss=0.3047, simple_loss=0.3648, pruned_loss=0.1223, over 28691.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.347, pruned_loss=0.09914, over 5686744.86 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3336, pruned_loss=0.08779, over 5762320.85 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3461, pruned_loss=0.09946, over 5672690.64 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:56:55,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.165e+02 1.205e+03 1.463e+03 2.014e+03 7.088e+03, threshold=2.926e+03, percent-clipped=7.0
+2023-03-09 21:56:58,509 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=841266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:57:00,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:57:17,063 INFO [train.py:968] (0/2) Epoch 19, batch 18700, giga_loss[loss=0.2419, simple_loss=0.327, pruned_loss=0.0784, over 28751.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3498, pruned_loss=0.1006, over 5692668.20 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3342, pruned_loss=0.08789, over 5767396.45 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3492, pruned_loss=0.1013, over 5673068.40 frames. ], batch size: 60, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:57:26,323 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=841298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:58:00,635 INFO [train.py:968] (0/2) Epoch 19, batch 18750, giga_loss[loss=0.261, simple_loss=0.348, pruned_loss=0.08695, over 28701.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3522, pruned_loss=0.1012, over 5694851.34 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3348, pruned_loss=0.08799, over 5767659.07 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3518, pruned_loss=0.1022, over 5675862.17 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:58:19,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.516e+02 1.243e+03 1.399e+03 1.892e+03 4.572e+03, threshold=2.798e+03, percent-clipped=2.0
+2023-03-09 21:58:38,680 INFO [train.py:968] (0/2) Epoch 19, batch 18800, giga_loss[loss=0.2848, simple_loss=0.3678, pruned_loss=0.1009, over 28713.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3532, pruned_loss=0.1007, over 5696990.45 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3349, pruned_loss=0.08791, over 5771528.37 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3535, pruned_loss=0.1021, over 5675314.87 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:58:54,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:59:01,222 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:59:20,006 INFO [train.py:968] (0/2) Epoch 19, batch 18850, giga_loss[loss=0.2735, simple_loss=0.3561, pruned_loss=0.09549, over 29040.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3543, pruned_loss=0.1007, over 5702284.35 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3344, pruned_loss=0.08765, over 5774831.34 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3555, pruned_loss=0.1024, over 5679686.00 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:59:27,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7706, 1.8529, 1.3856, 1.4650], device='cuda:0'), covar=tensor([0.0986, 0.0713, 0.1043, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0442, 0.0516, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 21:59:37,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.428e+02 1.307e+03 1.649e+03 2.145e+03 4.267e+03, threshold=3.299e+03, percent-clipped=11.0
+2023-03-09 21:59:37,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:59:59,345 INFO [train.py:968] (0/2) Epoch 19, batch 18900, giga_loss[loss=0.2985, simple_loss=0.3774, pruned_loss=0.1098, over 28686.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3549, pruned_loss=0.1002, over 5697255.56 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3351, pruned_loss=0.08796, over 5769964.34 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.356, pruned_loss=0.1017, over 5680146.26 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:00:38,132 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-09 22:00:38,362 INFO [train.py:968] (0/2) Epoch 19, batch 18950, giga_loss[loss=0.2849, simple_loss=0.3647, pruned_loss=0.1025, over 28909.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3549, pruned_loss=0.09899, over 5713153.22 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3352, pruned_loss=0.088, over 5773850.46 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3563, pruned_loss=0.1005, over 5693969.99 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:00:57,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.998e+02 1.245e+03 1.563e+03 2.004e+03 5.160e+03, threshold=3.125e+03, percent-clipped=4.0
+2023-03-09 22:01:07,246 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:01:17,231 INFO [train.py:968] (0/2) Epoch 19, batch 19000, giga_loss[loss=0.2921, simple_loss=0.3685, pruned_loss=0.1078, over 28527.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3534, pruned_loss=0.09708, over 5710121.67 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3359, pruned_loss=0.08834, over 5767518.51 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3546, pruned_loss=0.09844, over 5697267.07 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:01:25,992 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:01:30,499 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=841606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:01:34,296 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:01:56,848 INFO [train.py:968] (0/2) Epoch 19, batch 19050, giga_loss[loss=0.2536, simple_loss=0.3377, pruned_loss=0.08476, over 28951.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3529, pruned_loss=0.0969, over 5712035.51 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3358, pruned_loss=0.08821, over 5768900.51 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3542, pruned_loss=0.09823, over 5699722.27 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:01:57,681 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=841638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:02:20,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.120e+02 1.159e+03 1.530e+03 2.014e+03 8.785e+03, threshold=3.059e+03, percent-clipped=11.0
+2023-03-09 22:02:45,508 INFO [train.py:968] (0/2) Epoch 19, batch 19100, giga_loss[loss=0.2927, simple_loss=0.3579, pruned_loss=0.1137, over 28897.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3553, pruned_loss=0.1008, over 5715922.83 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3359, pruned_loss=0.08821, over 5770274.64 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3565, pruned_loss=0.102, over 5704219.15 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:02:57,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5581, 1.6020, 1.7951, 1.3864], device='cuda:0'), covar=tensor([0.1440, 0.2124, 0.1159, 0.1445], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0690, 0.0927, 0.0829], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 22:03:31,545 INFO [train.py:968] (0/2) Epoch 19, batch 19150, giga_loss[loss=0.3375, simple_loss=0.4077, pruned_loss=0.1336, over 29081.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3583, pruned_loss=0.1056, over 5717130.66 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3359, pruned_loss=0.0882, over 5770380.53 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3596, pruned_loss=0.1068, over 5706876.61 frames. ], batch size: 128, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:03:51,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.273e+02 1.379e+03 1.760e+03 2.332e+03 4.032e+03, threshold=3.520e+03, percent-clipped=5.0
+2023-03-09 22:04:05,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:04:11,225 INFO [train.py:968] (0/2) Epoch 19, batch 19200, giga_loss[loss=0.2582, simple_loss=0.3343, pruned_loss=0.09102, over 28532.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3572, pruned_loss=0.106, over 5709370.08 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3367, pruned_loss=0.08865, over 5770309.46 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3581, pruned_loss=0.107, over 5699717.38 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:04:13,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:04:55,035 INFO [train.py:968] (0/2) Epoch 19, batch 19250, giga_loss[loss=0.2747, simple_loss=0.3456, pruned_loss=0.1019, over 28440.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3538, pruned_loss=0.1046, over 5706351.44 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3365, pruned_loss=0.08852, over 5772994.71 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.355, pruned_loss=0.1058, over 5695090.36 frames. ], batch size: 65, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:05:05,866 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841851.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 22:05:15,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.498e+02 1.373e+03 1.772e+03 2.174e+03 5.782e+03, threshold=3.543e+03, percent-clipped=5.0
+2023-03-09 22:05:38,194 INFO [train.py:968] (0/2) Epoch 19, batch 19300, giga_loss[loss=0.2539, simple_loss=0.3345, pruned_loss=0.08661, over 28691.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3523, pruned_loss=0.1034, over 5716440.41 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3359, pruned_loss=0.08797, over 5776134.93 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3544, pruned_loss=0.1054, over 5702421.76 frames. ], batch size: 92, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:05:51,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5765, 1.7341, 1.8259, 1.3894], device='cuda:0'), covar=tensor([0.1795, 0.2441, 0.1455, 0.1645], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0689, 0.0925, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 22:06:01,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6742, 1.8161, 1.4991, 1.7313], device='cuda:0'), covar=tensor([0.2495, 0.2594, 0.2909, 0.2301], device='cuda:0'), in_proj_covar=tensor([0.1446, 0.1052, 0.1284, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 22:06:07,277 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=841924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:09,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:14,290 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=841932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:16,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1733, 1.2625, 1.1329, 0.9255], device='cuda:0'), covar=tensor([0.0952, 0.0483, 0.0995, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0443, 0.0515, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 22:06:16,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:19,252 INFO [train.py:968] (0/2) Epoch 19, batch 19350, giga_loss[loss=0.2749, simple_loss=0.3539, pruned_loss=0.09796, over 28581.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3525, pruned_loss=0.1027, over 5699935.10 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3372, pruned_loss=0.08867, over 5753248.13 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3536, pruned_loss=0.1043, over 5706252.05 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:06:28,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:33,235 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=841956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:40,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.161e+02 1.191e+03 1.412e+03 1.831e+03 4.567e+03, threshold=2.825e+03, percent-clipped=2.0
+2023-03-09 22:06:41,968 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=841964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:48,903 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:53,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:07:05,206 INFO [train.py:968] (0/2) Epoch 19, batch 19400, giga_loss[loss=0.2814, simple_loss=0.3547, pruned_loss=0.104, over 28609.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3498, pruned_loss=0.1009, over 5691413.20 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3368, pruned_loss=0.08845, over 5756492.28 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3513, pruned_loss=0.1026, over 5692416.19 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:07:15,466 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-842000.pt
+2023-03-09 22:07:50,284 INFO [train.py:968] (0/2) Epoch 19, batch 19450, giga_loss[loss=0.2352, simple_loss=0.3129, pruned_loss=0.07876, over 28772.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3438, pruned_loss=0.0974, over 5691719.47 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3371, pruned_loss=0.08855, over 5757417.17 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3449, pruned_loss=0.09873, over 5691153.89 frames. ], batch size: 284, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:08:18,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.657e+02 1.047e+03 1.308e+03 1.631e+03 6.843e+03, threshold=2.615e+03, percent-clipped=4.0
+2023-03-09 22:08:41,167 INFO [train.py:968] (0/2) Epoch 19, batch 19500, giga_loss[loss=0.2645, simple_loss=0.3355, pruned_loss=0.09681, over 28653.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3383, pruned_loss=0.09493, over 5678859.71 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3371, pruned_loss=0.08855, over 5758190.09 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3391, pruned_loss=0.09601, over 5677358.31 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:08:46,771 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:08:50,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:08:55,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3236, 1.2456, 1.2294, 1.4844], device='cuda:0'), covar=tensor([0.0817, 0.0360, 0.0344, 0.0893], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0104], device='cuda:0')
+2023-03-09 22:09:15,318 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:09:17,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:09:19,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:09:32,324 INFO [train.py:968] (0/2) Epoch 19, batch 19550, giga_loss[loss=0.2371, simple_loss=0.3152, pruned_loss=0.07953, over 28802.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3338, pruned_loss=0.09306, over 5658317.92 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3375, pruned_loss=0.08867, over 5758547.45 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3341, pruned_loss=0.09388, over 5655896.63 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:09:46,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:09:57,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.970e+02 9.499e+02 1.239e+03 1.569e+03 5.325e+03, threshold=2.478e+03, percent-clipped=3.0
+2023-03-09 22:10:15,623 INFO [train.py:968] (0/2) Epoch 19, batch 19600, giga_loss[loss=0.2519, simple_loss=0.3316, pruned_loss=0.08606, over 28983.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3336, pruned_loss=0.092, over 5663207.41 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3378, pruned_loss=0.08859, over 5761625.52 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3335, pruned_loss=0.09291, over 5654087.83 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:10:23,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5621, 1.8474, 1.4862, 1.5743], device='cuda:0'), covar=tensor([0.2541, 0.2568, 0.2943, 0.2323], device='cuda:0'), in_proj_covar=tensor([0.1449, 0.1053, 0.1286, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 22:10:35,847 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=842212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:10:47,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=842226.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 22:11:00,707 INFO [train.py:968] (0/2) Epoch 19, batch 19650, giga_loss[loss=0.2615, simple_loss=0.3358, pruned_loss=0.09358, over 28904.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3348, pruned_loss=0.09289, over 5667135.73 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3378, pruned_loss=0.08856, over 5763916.97 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3346, pruned_loss=0.09368, over 5656770.58 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:11:23,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.103e+02 1.043e+03 1.294e+03 2.029e+03 5.130e+03, threshold=2.589e+03, percent-clipped=13.0
+2023-03-09 22:11:40,565 INFO [train.py:968] (0/2) Epoch 19, batch 19700, libri_loss[loss=0.2899, simple_loss=0.3608, pruned_loss=0.1095, over 29567.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.335, pruned_loss=0.09308, over 5675497.58 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3385, pruned_loss=0.08882, over 5762279.02 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.334, pruned_loss=0.09362, over 5665247.80 frames. ], batch size: 77, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:12:20,465 INFO [train.py:968] (0/2) Epoch 19, batch 19750, giga_loss[loss=0.2261, simple_loss=0.2995, pruned_loss=0.0764, over 28880.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3331, pruned_loss=0.09156, over 5688703.49 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3393, pruned_loss=0.08902, over 5766990.74 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3314, pruned_loss=0.09192, over 5673313.31 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:12:26,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=842345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:12:40,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.457e+02 1.010e+03 1.285e+03 1.888e+03 5.916e+03, threshold=2.569e+03, percent-clipped=12.0
+2023-03-09 22:12:44,218 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842369.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 22:12:46,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842372.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 22:12:51,541 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=842378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:12:59,804 INFO [train.py:968] (0/2) Epoch 19, batch 19800, giga_loss[loss=0.2338, simple_loss=0.3136, pruned_loss=0.07698, over 28846.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3322, pruned_loss=0.09122, over 5684032.99 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3399, pruned_loss=0.08912, over 5757137.75 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3301, pruned_loss=0.09152, over 5677020.07 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:13:11,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842401.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 22:13:41,480 INFO [train.py:968] (0/2) Epoch 19, batch 19850, giga_loss[loss=0.2219, simple_loss=0.2996, pruned_loss=0.0721, over 28982.00 frames. ], tot_loss[loss=0.254, simple_loss=0.329, pruned_loss=0.08955, over 5698325.68 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3398, pruned_loss=0.08901, over 5758545.73 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3273, pruned_loss=0.0899, over 5690738.34 frames. ], batch size: 106, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:14:06,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.863e+02 1.069e+03 1.435e+03 2.415e+03 6.785e+03, threshold=2.871e+03, percent-clipped=23.0
+2023-03-09 22:14:22,213 INFO [train.py:968] (0/2) Epoch 19, batch 19900, giga_loss[loss=0.2359, simple_loss=0.3128, pruned_loss=0.07948, over 28866.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3263, pruned_loss=0.08814, over 5697978.89 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3404, pruned_loss=0.08905, over 5756699.21 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3241, pruned_loss=0.08837, over 5691746.40 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:14:22,533 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:14:24,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:14:34,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3121, 1.3271, 1.1246, 1.4704], device='cuda:0'), covar=tensor([0.0788, 0.0364, 0.0362, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0104], device='cuda:0')
+2023-03-09 22:14:48,640 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:14:57,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2845, 2.3777, 1.7390, 1.9022], device='cuda:0'), covar=tensor([0.0915, 0.0660, 0.0944, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0442, 0.0514, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 22:15:02,675 INFO [train.py:968] (0/2) Epoch 19, batch 19950, giga_loss[loss=0.222, simple_loss=0.2999, pruned_loss=0.07202, over 28811.00 frames. ], tot_loss[loss=0.248, simple_loss=0.323, pruned_loss=0.08646, over 5712487.14 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3403, pruned_loss=0.08885, over 5760289.33 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3209, pruned_loss=0.08677, over 5702610.38 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:15:26,291 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.584e+02 1.088e+03 1.365e+03 1.785e+03 4.765e+03, threshold=2.729e+03, percent-clipped=10.0
+2023-03-09 22:15:36,778 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-09 22:15:43,392 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=842587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:15:43,861 INFO [train.py:968] (0/2) Epoch 19, batch 20000, giga_loss[loss=0.2269, simple_loss=0.3027, pruned_loss=0.07552, over 29017.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3203, pruned_loss=0.08501, over 5720438.77 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3401, pruned_loss=0.08864, over 5762631.61 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3185, pruned_loss=0.08538, over 5710057.51 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:16:22,500 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-09 22:16:24,686 INFO [train.py:968] (0/2) Epoch 19, batch 20050, giga_loss[loss=0.2267, simple_loss=0.3008, pruned_loss=0.07632, over 28657.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3195, pruned_loss=0.08492, over 5709837.22 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3406, pruned_loss=0.08891, over 5757277.15 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3171, pruned_loss=0.08489, over 5705026.60 frames. ], batch size: 92, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:16:46,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.905e+02 1.130e+03 1.469e+03 2.199e+03 6.872e+03, threshold=2.939e+03, percent-clipped=16.0
+2023-03-09 22:16:47,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=842666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:17:04,950 INFO [train.py:968] (0/2) Epoch 19, batch 20100, giga_loss[loss=0.2442, simple_loss=0.3178, pruned_loss=0.08528, over 28951.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.318, pruned_loss=0.08423, over 5714591.81 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3407, pruned_loss=0.08893, over 5758906.08 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.3158, pruned_loss=0.08415, over 5709110.73 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:17:19,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=842708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:17:36,592 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:17:40,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:17:43,354 INFO [train.py:968] (0/2) Epoch 19, batch 20150, giga_loss[loss=0.2317, simple_loss=0.3087, pruned_loss=0.07738, over 28768.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3178, pruned_loss=0.08402, over 5720481.26 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3406, pruned_loss=0.08876, over 5760914.43 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.3157, pruned_loss=0.08401, over 5713358.68 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:17:54,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=842753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:18:02,489 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:18:05,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.317e+02 1.013e+03 1.355e+03 1.746e+03 3.798e+03, threshold=2.711e+03, percent-clipped=4.0
+2023-03-09 22:18:24,494 INFO [train.py:968] (0/2) Epoch 19, batch 20200, libri_loss[loss=0.2746, simple_loss=0.3642, pruned_loss=0.09251, over 29517.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3251, pruned_loss=0.08824, over 5717637.00 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3419, pruned_loss=0.08929, over 5764626.26 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3211, pruned_loss=0.08759, over 5706565.08 frames. ], batch size: 89, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:18:42,155 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3920, 3.3322, 1.4876, 1.5347], device='cuda:0'), covar=tensor([0.0982, 0.0314, 0.0903, 0.1305], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0534, 0.0371, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 22:19:08,817 INFO [train.py:968] (0/2) Epoch 19, batch 20250, giga_loss[loss=0.2737, simple_loss=0.3513, pruned_loss=0.09807, over 28887.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3308, pruned_loss=0.09201, over 5704678.45 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3418, pruned_loss=0.08932, over 5759759.09 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3272, pruned_loss=0.09147, over 5698864.95 frames. ], batch size: 227, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:19:25,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2626, 1.4842, 1.3999, 1.1985], device='cuda:0'), covar=tensor([0.2072, 0.1984, 0.1411, 0.1870], device='cuda:0'), in_proj_covar=tensor([0.1907, 0.1799, 0.1753, 0.1904], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 22:19:36,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.127e+02 1.349e+03 1.607e+03 2.226e+03 8.606e+03, threshold=3.215e+03, percent-clipped=15.0
+2023-03-09 22:19:59,570 INFO [train.py:968] (0/2) Epoch 19, batch 20300, giga_loss[loss=0.3803, simple_loss=0.4226, pruned_loss=0.169, over 26530.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3376, pruned_loss=0.09658, over 5702457.86 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3417, pruned_loss=0.08932, over 5762285.44 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3347, pruned_loss=0.09626, over 5694143.35 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:20:07,888 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:20:11,079 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:20:35,370 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:20:40,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9950, 1.2740, 3.3186, 2.7397], device='cuda:0'), covar=tensor([0.1750, 0.2689, 0.0484, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0731, 0.0630, 0.0923, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 22:20:45,588 INFO [train.py:968] (0/2) Epoch 19, batch 20350, giga_loss[loss=0.3217, simple_loss=0.3845, pruned_loss=0.1294, over 28576.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3445, pruned_loss=0.1004, over 5695063.83 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3426, pruned_loss=0.08982, over 5760998.28 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3414, pruned_loss=0.09992, over 5688616.64 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:21:05,525 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-09 22:21:14,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.354e+02 1.182e+03 1.494e+03 2.053e+03 8.945e+03, threshold=2.988e+03, percent-clipped=10.0
+2023-03-09 22:21:33,439 INFO [train.py:968] (0/2) Epoch 19, batch 20400, giga_loss[loss=0.3548, simple_loss=0.3952, pruned_loss=0.1572, over 23651.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3497, pruned_loss=0.1029, over 5692042.08 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3425, pruned_loss=0.08987, over 5763240.36 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3474, pruned_loss=0.1027, over 5683142.30 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:22:08,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4872, 1.8561, 1.4469, 1.7118], device='cuda:0'), covar=tensor([0.2668, 0.2727, 0.3008, 0.2375], device='cuda:0'), in_proj_covar=tensor([0.1454, 0.1057, 0.1291, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 22:22:17,288 INFO [train.py:968] (0/2) Epoch 19, batch 20450, giga_loss[loss=0.3449, simple_loss=0.4069, pruned_loss=0.1415, over 27621.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3534, pruned_loss=0.1041, over 5687600.03 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3433, pruned_loss=0.09047, over 5756085.11 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3512, pruned_loss=0.1038, over 5684887.52 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:22:22,813 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=843041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:22:47,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.978e+02 1.192e+03 1.521e+03 2.319e+03 4.786e+03, threshold=3.041e+03, percent-clipped=10.0
+2023-03-09 22:23:01,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=843083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:23:05,686 INFO [train.py:968] (0/2) Epoch 19, batch 20500, giga_loss[loss=0.2653, simple_loss=0.3478, pruned_loss=0.09134, over 28619.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3575, pruned_loss=0.1063, over 5690975.38 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3433, pruned_loss=0.09056, over 5756849.68 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3557, pruned_loss=0.1061, over 5687854.12 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:23:52,333 INFO [train.py:968] (0/2) Epoch 19, batch 20550, giga_loss[loss=0.2342, simple_loss=0.3201, pruned_loss=0.07418, over 28724.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3523, pruned_loss=0.1027, over 5691531.49 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3435, pruned_loss=0.09082, over 5759747.25 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.351, pruned_loss=0.1026, over 5685298.57 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:24:17,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.923e+02 1.160e+03 1.437e+03 1.847e+03 4.442e+03, threshold=2.874e+03, percent-clipped=6.0
+2023-03-09 22:24:30,126 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=843184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:24:33,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=843187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:24:33,987 INFO [train.py:968] (0/2) Epoch 19, batch 20600, giga_loss[loss=0.2932, simple_loss=0.3656, pruned_loss=0.1104, over 28631.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3496, pruned_loss=0.1001, over 5700127.81 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3434, pruned_loss=0.09083, over 5762779.71 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3488, pruned_loss=0.1002, over 5691299.89 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:24:57,805 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=843216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:25:08,064 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=843226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:25:11,587 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=843229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:25:18,617 INFO [train.py:968] (0/2) Epoch 19, batch 20650, giga_loss[loss=0.2515, simple_loss=0.3408, pruned_loss=0.08107, over 28974.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3491, pruned_loss=0.09965, over 5687683.02 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3436, pruned_loss=0.09124, over 5757848.27 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3484, pruned_loss=0.09956, over 5683917.54 frames. ], batch size: 164, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:25:36,627 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=843258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:25:43,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.373e+02 1.304e+03 1.738e+03 2.191e+03 5.394e+03, threshold=3.476e+03, percent-clipped=12.0
+2023-03-09 22:26:00,909 INFO [train.py:968] (0/2) Epoch 19, batch 20700, giga_loss[loss=0.2891, simple_loss=0.3585, pruned_loss=0.1099, over 28830.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3501, pruned_loss=0.09978, over 5695168.85 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3438, pruned_loss=0.09151, over 5760983.77 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3495, pruned_loss=0.09969, over 5687630.84 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:26:04,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6263, 1.7906, 1.6204, 1.6029], device='cuda:0'), covar=tensor([0.1954, 0.2205, 0.2394, 0.2193], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0741, 0.0705, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 22:26:45,707 INFO [train.py:968] (0/2) Epoch 19, batch 20750, giga_loss[loss=0.2748, simple_loss=0.3508, pruned_loss=0.09941, over 28892.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.353, pruned_loss=0.1016, over 5692913.24 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3442, pruned_loss=0.09183, over 5758864.40 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3523, pruned_loss=0.1014, over 5687369.24 frames. ], batch size: 227, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:26:51,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6645, 1.8453, 1.4753, 2.2003], device='cuda:0'), covar=tensor([0.2450, 0.2622, 0.2891, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.1457, 0.1060, 0.1294, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 22:27:10,191 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=843366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:27:10,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.812e+02 1.257e+03 1.575e+03 2.156e+03 5.137e+03, threshold=3.150e+03, percent-clipped=4.0
+2023-03-09 22:27:26,912 INFO [train.py:968] (0/2) Epoch 19, batch 20800, giga_loss[loss=0.3274, simple_loss=0.3954, pruned_loss=0.1297, over 28970.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3546, pruned_loss=0.1032, over 5699895.93 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3449, pruned_loss=0.09227, over 5763303.13 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3537, pruned_loss=0.1029, over 5689664.43 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:27:49,713 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9595, 1.1318, 1.0509, 0.8489], device='cuda:0'), covar=tensor([0.2038, 0.2442, 0.1526, 0.2081], device='cuda:0'), in_proj_covar=tensor([0.1921, 0.1819, 0.1765, 0.1921], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 22:28:14,657 INFO [train.py:968] (0/2) Epoch 19, batch 20850, libri_loss[loss=0.2882, simple_loss=0.3652, pruned_loss=0.1056, over 29490.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3546, pruned_loss=0.1031, over 5713561.65 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3455, pruned_loss=0.09267, over 5766804.06 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3536, pruned_loss=0.1028, over 5700411.68 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:28:40,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.251e+02 1.307e+03 1.686e+03 2.244e+03 5.102e+03, threshold=3.371e+03, percent-clipped=11.0
+2023-03-09 22:28:59,151 INFO [train.py:968] (0/2) Epoch 19, batch 20900, giga_loss[loss=0.3, simple_loss=0.3729, pruned_loss=0.1136, over 29023.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3569, pruned_loss=0.1053, over 5710947.16 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3458, pruned_loss=0.09288, over 5768657.79 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3561, pruned_loss=0.1051, over 5697906.75 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:29:39,247 INFO [train.py:968] (0/2) Epoch 19, batch 20950, giga_loss[loss=0.2909, simple_loss=0.3682, pruned_loss=0.1068, over 28906.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3571, pruned_loss=0.1054, over 5716470.23 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3462, pruned_loss=0.09326, over 5770413.01 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3563, pruned_loss=0.1051, over 5703905.05 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:30:02,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.137e+02 1.139e+03 1.361e+03 1.769e+03 3.767e+03, threshold=2.723e+03, percent-clipped=2.0
+2023-03-09 22:30:21,109 INFO [train.py:968] (0/2) Epoch 19, batch 21000, giga_loss[loss=0.2908, simple_loss=0.3681, pruned_loss=0.1068, over 28969.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3557, pruned_loss=0.1036, over 5714253.84 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3462, pruned_loss=0.09336, over 5772480.13 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3551, pruned_loss=0.1034, over 5701770.64 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:30:21,114 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 22:30:30,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5582, 1.7312, 1.3460, 1.3774], device='cuda:0'), covar=tensor([0.0906, 0.0440, 0.0931, 0.0867], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0438, 0.0509, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 22:30:31,411 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2105, simple_loss=0.319, pruned_loss=0.05096, over 944034.00 frames. 
+2023-03-09 22:30:31,411 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 22:30:34,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=843593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:30:42,646 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-09 22:31:11,571 INFO [train.py:968] (0/2) Epoch 19, batch 21050, giga_loss[loss=0.2821, simple_loss=0.3589, pruned_loss=0.1027, over 27948.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3563, pruned_loss=0.1029, over 5715290.85 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3465, pruned_loss=0.09369, over 5769616.19 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3558, pruned_loss=0.1027, over 5707081.07 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:31:37,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.904e+02 1.093e+03 1.309e+03 1.624e+03 2.842e+03, threshold=2.618e+03, percent-clipped=1.0
+2023-03-09 22:31:44,967 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=843677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:31:52,275 INFO [train.py:968] (0/2) Epoch 19, batch 21100, giga_loss[loss=0.2376, simple_loss=0.3222, pruned_loss=0.0765, over 28897.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3557, pruned_loss=0.1024, over 5719255.81 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3464, pruned_loss=0.09391, over 5768642.51 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3555, pruned_loss=0.1022, over 5712755.18 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:32:08,817 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=843708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:32:15,977 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 22:32:32,727 INFO [train.py:968] (0/2) Epoch 19, batch 21150, giga_loss[loss=0.2329, simple_loss=0.3146, pruned_loss=0.07561, over 28629.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3528, pruned_loss=0.1012, over 5712062.35 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3462, pruned_loss=0.09399, over 5768925.36 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3528, pruned_loss=0.101, over 5706159.88 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:32:34,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=843741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:32:45,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3294, 1.1860, 1.0628, 1.4368], device='cuda:0'), covar=tensor([0.0733, 0.0347, 0.0353, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0115, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 22:32:56,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.541e+02 1.071e+03 1.309e+03 1.774e+03 4.001e+03, threshold=2.618e+03, percent-clipped=7.0
+2023-03-09 22:33:13,380 INFO [train.py:968] (0/2) Epoch 19, batch 21200, giga_loss[loss=0.2895, simple_loss=0.3571, pruned_loss=0.1109, over 27944.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.351, pruned_loss=0.1006, over 5713062.08 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3464, pruned_loss=0.09425, over 5771635.61 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3511, pruned_loss=0.1004, over 5704339.04 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:33:32,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0785, 3.8884, 3.6576, 2.0024], device='cuda:0'), covar=tensor([0.0627, 0.0769, 0.0815, 0.2105], device='cuda:0'), in_proj_covar=tensor([0.1168, 0.1085, 0.0925, 0.0711], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 22:33:38,835 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=843821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:33:47,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4343, 4.4395, 1.6228, 1.7034], device='cuda:0'), covar=tensor([0.1063, 0.0222, 0.0905, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0535, 0.0371, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 22:33:52,577 INFO [train.py:968] (0/2) Epoch 19, batch 21250, giga_loss[loss=0.2877, simple_loss=0.3645, pruned_loss=0.1055, over 28605.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3501, pruned_loss=0.1001, over 5721437.27 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.347, pruned_loss=0.09472, over 5773209.28 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3497, pruned_loss=0.09961, over 5711513.32 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:34:16,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.299e+02 1.108e+03 1.505e+03 2.030e+03 4.795e+03, threshold=3.010e+03, percent-clipped=9.0
+2023-03-09 22:34:29,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=843884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:34:31,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=843887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:34:32,036 INFO [train.py:968] (0/2) Epoch 19, batch 21300, giga_loss[loss=0.2373, simple_loss=0.3155, pruned_loss=0.07956, over 28425.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3514, pruned_loss=0.1015, over 5721952.75 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.347, pruned_loss=0.09514, over 5775489.49 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3512, pruned_loss=0.1009, over 5710732.82 frames. ], batch size: 65, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:34:43,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2626, 1.7235, 1.4626, 1.4430], device='cuda:0'), covar=tensor([0.0743, 0.0357, 0.0319, 0.0807], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0115, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 22:34:59,284 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=843916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:35:16,412 INFO [train.py:968] (0/2) Epoch 19, batch 21350, giga_loss[loss=0.2735, simple_loss=0.3581, pruned_loss=0.09444, over 28952.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.351, pruned_loss=0.1012, over 5720477.11 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3467, pruned_loss=0.09517, over 5779256.31 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3512, pruned_loss=0.1009, over 5706202.45 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:35:42,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=843968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:35:42,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.565e+02 1.128e+03 1.431e+03 1.731e+03 4.124e+03, threshold=2.863e+03, percent-clipped=2.0
+2023-03-09 22:35:56,861 INFO [train.py:968] (0/2) Epoch 19, batch 21400, giga_loss[loss=0.2531, simple_loss=0.3397, pruned_loss=0.08321, over 28345.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3509, pruned_loss=0.1004, over 5721561.44 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3473, pruned_loss=0.09568, over 5777085.83 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3506, pruned_loss=0.09977, over 5711029.42 frames. ], batch size: 368, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:36:05,480 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-844000.pt
+2023-03-09 22:36:39,568 INFO [train.py:968] (0/2) Epoch 19, batch 21450, giga_loss[loss=0.2654, simple_loss=0.3535, pruned_loss=0.0886, over 28965.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3503, pruned_loss=0.09995, over 5711294.81 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3476, pruned_loss=0.09602, over 5778875.78 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3499, pruned_loss=0.0992, over 5700547.16 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:36:49,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:37:02,701 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.844e+02 1.155e+03 1.348e+03 1.882e+03 1.220e+04, threshold=2.697e+03, percent-clipped=6.0
+2023-03-09 22:37:14,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:37:17,428 INFO [train.py:968] (0/2) Epoch 19, batch 21500, giga_loss[loss=0.2579, simple_loss=0.3315, pruned_loss=0.09217, over 29016.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3491, pruned_loss=0.1, over 5709559.47 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3475, pruned_loss=0.0961, over 5782738.05 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3489, pruned_loss=0.09948, over 5695728.66 frames. ], batch size: 128, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:37:36,424 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:37:39,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:37:55,405 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=844137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:37:55,858 INFO [train.py:968] (0/2) Epoch 19, batch 21550, giga_loss[loss=0.2337, simple_loss=0.3095, pruned_loss=0.07895, over 28863.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3473, pruned_loss=0.09945, over 5704902.09 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3482, pruned_loss=0.09673, over 5775873.18 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3464, pruned_loss=0.09851, over 5698311.62 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:37:59,488 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=844143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:38:22,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.015e+02 1.098e+03 1.466e+03 2.090e+03 6.767e+03, threshold=2.931e+03, percent-clipped=12.0
+2023-03-09 22:38:38,315 INFO [train.py:968] (0/2) Epoch 19, batch 21600, giga_loss[loss=0.2472, simple_loss=0.3259, pruned_loss=0.0843, over 28786.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3435, pruned_loss=0.09735, over 5702225.34 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3488, pruned_loss=0.0974, over 5777429.13 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3421, pruned_loss=0.09599, over 5694265.07 frames. ], batch size: 92, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:38:43,115 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:38:43,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:38:45,478 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:38:52,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4311, 4.2562, 4.0154, 2.1052], device='cuda:0'), covar=tensor([0.0554, 0.0695, 0.0744, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.1168, 0.1085, 0.0925, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 22:38:59,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3750, 3.2053, 3.0010, 1.8370], device='cuda:0'), covar=tensor([0.0781, 0.0899, 0.0823, 0.2074], device='cuda:0'), in_proj_covar=tensor([0.1169, 0.1085, 0.0926, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 22:39:08,305 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:39:08,893 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=844227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:39:10,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:39:16,278 INFO [train.py:968] (0/2) Epoch 19, batch 21650, giga_loss[loss=0.2497, simple_loss=0.3334, pruned_loss=0.08303, over 28446.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3434, pruned_loss=0.09766, over 5696202.48 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3495, pruned_loss=0.09802, over 5770218.92 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3416, pruned_loss=0.09601, over 5693836.03 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:39:33,608 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=844258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:39:44,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.165e+02 1.185e+03 1.496e+03 1.969e+03 8.991e+03, threshold=2.991e+03, percent-clipped=12.0
+2023-03-09 22:39:57,572 INFO [train.py:968] (0/2) Epoch 19, batch 21700, giga_loss[loss=0.2565, simple_loss=0.3321, pruned_loss=0.0905, over 28957.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3428, pruned_loss=0.09769, over 5697277.19 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3499, pruned_loss=0.09838, over 5773834.55 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3407, pruned_loss=0.09605, over 5690223.44 frames. ], batch size: 106, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:40:29,465 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=844325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:40:39,277 INFO [train.py:968] (0/2) Epoch 19, batch 21750, giga_loss[loss=0.2379, simple_loss=0.3118, pruned_loss=0.08195, over 28380.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3414, pruned_loss=0.09758, over 5704751.58 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3503, pruned_loss=0.09882, over 5776989.28 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3392, pruned_loss=0.09582, over 5693544.67 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:40:40,333 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:40:42,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:41:04,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.855e+02 1.118e+03 1.357e+03 1.578e+03 5.140e+03, threshold=2.713e+03, percent-clipped=5.0
+2023-03-09 22:41:04,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=844370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:41:05,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=844371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:41:18,115 INFO [train.py:968] (0/2) Epoch 19, batch 21800, giga_loss[loss=0.2774, simple_loss=0.3493, pruned_loss=0.1028, over 27542.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3376, pruned_loss=0.09545, over 5710573.82 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3501, pruned_loss=0.09885, over 5780815.59 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3357, pruned_loss=0.09391, over 5696386.61 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:41:40,334 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 22:41:45,580 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8248, 1.9061, 1.3762, 1.5262], device='cuda:0'), covar=tensor([0.0974, 0.0816, 0.1106, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0444, 0.0514, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 22:41:56,637 INFO [train.py:968] (0/2) Epoch 19, batch 21850, libri_loss[loss=0.2606, simple_loss=0.3365, pruned_loss=0.0923, over 29569.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3363, pruned_loss=0.09493, over 5720172.34 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3502, pruned_loss=0.09911, over 5782791.93 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.334, pruned_loss=0.09327, over 5703970.40 frames. ], batch size: 76, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:42:22,742 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.218e+02 1.159e+03 1.401e+03 1.853e+03 6.941e+03, threshold=2.803e+03, percent-clipped=8.0
+2023-03-09 22:42:38,525 INFO [train.py:968] (0/2) Epoch 19, batch 21900, giga_loss[loss=0.2577, simple_loss=0.3414, pruned_loss=0.08704, over 28730.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3347, pruned_loss=0.09428, over 5718616.07 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3504, pruned_loss=0.09928, over 5782562.97 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3327, pruned_loss=0.09282, over 5705882.32 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:43:00,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:43:21,613 INFO [train.py:968] (0/2) Epoch 19, batch 21950, giga_loss[loss=0.2554, simple_loss=0.3344, pruned_loss=0.0882, over 28927.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3372, pruned_loss=0.09537, over 5708768.06 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3511, pruned_loss=0.09996, over 5772251.78 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3346, pruned_loss=0.09344, over 5706596.24 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:43:21,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=844538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:43:37,596 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9401, 1.6771, 5.4173, 3.8282], device='cuda:0'), covar=tensor([0.1526, 0.2611, 0.0338, 0.0743], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0630, 0.0928, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 22:43:52,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.121e+02 1.058e+03 1.302e+03 1.656e+03 9.128e+03, threshold=2.604e+03, percent-clipped=7.0
+2023-03-09 22:44:04,863 INFO [train.py:968] (0/2) Epoch 19, batch 22000, giga_loss[loss=0.2738, simple_loss=0.356, pruned_loss=0.09581, over 28952.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3406, pruned_loss=0.09666, over 5704590.59 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3517, pruned_loss=0.1005, over 5775445.29 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3377, pruned_loss=0.09451, over 5698569.33 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:44:43,101 INFO [train.py:968] (0/2) Epoch 19, batch 22050, giga_loss[loss=0.2683, simple_loss=0.3516, pruned_loss=0.09248, over 28694.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3442, pruned_loss=0.09816, over 5701915.14 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.352, pruned_loss=0.1011, over 5778028.20 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3409, pruned_loss=0.09562, over 5690324.46 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:44:56,646 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:44:58,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:45:09,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.639e+02 1.122e+03 1.344e+03 2.080e+03 6.411e+03, threshold=2.689e+03, percent-clipped=14.0
+2023-03-09 22:45:23,522 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=844687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:45:24,202 INFO [train.py:968] (0/2) Epoch 19, batch 22100, giga_loss[loss=0.3741, simple_loss=0.414, pruned_loss=0.1671, over 26700.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3459, pruned_loss=0.09854, over 5711186.80 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3525, pruned_loss=0.1017, over 5781622.70 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3427, pruned_loss=0.09588, over 5696846.00 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:45:35,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:46:07,935 INFO [train.py:968] (0/2) Epoch 19, batch 22150, giga_loss[loss=0.265, simple_loss=0.3389, pruned_loss=0.0955, over 28869.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3448, pruned_loss=0.09716, over 5714286.91 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3522, pruned_loss=0.1016, over 5782704.12 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3425, pruned_loss=0.09514, over 5701414.75 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:46:14,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:46:38,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.946e+02 1.052e+03 1.228e+03 1.653e+03 3.826e+03, threshold=2.455e+03, percent-clipped=6.0
+2023-03-09 22:46:54,608 INFO [train.py:968] (0/2) Epoch 19, batch 22200, giga_loss[loss=0.2647, simple_loss=0.3468, pruned_loss=0.09136, over 28694.00 frames. ], tot_loss[loss=0.27, simple_loss=0.345, pruned_loss=0.09753, over 5701820.77 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3524, pruned_loss=0.1019, over 5782922.31 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3429, pruned_loss=0.09567, over 5690877.61 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:47:23,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3324, 1.5898, 1.3080, 1.1406], device='cuda:0'), covar=tensor([0.2579, 0.2620, 0.3018, 0.2320], device='cuda:0'), in_proj_covar=tensor([0.1461, 0.1056, 0.1295, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 22:47:34,936 INFO [train.py:968] (0/2) Epoch 19, batch 22250, libri_loss[loss=0.347, simple_loss=0.4127, pruned_loss=0.1406, over 29194.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.346, pruned_loss=0.09846, over 5699556.01 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3538, pruned_loss=0.1029, over 5775440.13 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3428, pruned_loss=0.09592, over 5695832.88 frames. ], batch size: 97, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:47:40,513 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:47:42,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:03,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.473e+02 1.302e+03 1.536e+03 2.107e+03 4.517e+03, threshold=3.071e+03, percent-clipped=16.0
+2023-03-09 22:48:06,238 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=844875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:16,683 INFO [train.py:968] (0/2) Epoch 19, batch 22300, giga_loss[loss=0.263, simple_loss=0.3517, pruned_loss=0.08717, over 28893.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3468, pruned_loss=0.0993, over 5700777.23 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3539, pruned_loss=0.1031, over 5777467.36 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.344, pruned_loss=0.09702, over 5695029.18 frames. ], batch size: 164, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:48:17,031 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:19,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:38,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:42,346 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=844918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:43,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=844920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:49:00,363 INFO [train.py:968] (0/2) Epoch 19, batch 22350, giga_loss[loss=0.2824, simple_loss=0.361, pruned_loss=0.1019, over 28923.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3502, pruned_loss=0.1011, over 5712557.51 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3546, pruned_loss=0.1037, over 5781995.55 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3471, pruned_loss=0.09856, over 5701652.40 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:49:14,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3352, 1.4597, 1.3782, 1.3404], device='cuda:0'), covar=tensor([0.2573, 0.2123, 0.2157, 0.2265], device='cuda:0'), in_proj_covar=tensor([0.1924, 0.1831, 0.1781, 0.1916], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 22:49:25,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.315e+02 1.293e+03 1.543e+03 2.036e+03 1.064e+04, threshold=3.086e+03, percent-clipped=7.0
+2023-03-09 22:49:39,932 INFO [train.py:968] (0/2) Epoch 19, batch 22400, giga_loss[loss=0.2831, simple_loss=0.3609, pruned_loss=0.1027, over 28928.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3527, pruned_loss=0.102, over 5715503.20 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3548, pruned_loss=0.1038, over 5785355.66 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3501, pruned_loss=0.0999, over 5701896.81 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:49:41,037 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-09 22:50:18,394 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=845035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:50:20,140 INFO [train.py:968] (0/2) Epoch 19, batch 22450, giga_loss[loss=0.263, simple_loss=0.3453, pruned_loss=0.09037, over 28262.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3541, pruned_loss=0.1029, over 5721447.57 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3551, pruned_loss=0.1042, over 5786684.09 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3517, pruned_loss=0.1009, over 5707959.28 frames. ], batch size: 368, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:50:33,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=845056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:50:35,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=845059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:50:43,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.883e+02 1.267e+03 1.573e+03 2.447e+03 4.972e+03, threshold=3.146e+03, percent-clipped=14.0
+2023-03-09 22:50:58,469 INFO [train.py:968] (0/2) Epoch 19, batch 22500, giga_loss[loss=0.2607, simple_loss=0.3328, pruned_loss=0.09428, over 28472.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.354, pruned_loss=0.1026, over 5731044.72 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3553, pruned_loss=0.1045, over 5789801.15 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3518, pruned_loss=0.1006, over 5715840.32 frames. ], batch size: 60, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:50:58,703 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=845088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:51:27,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9189, 3.1240, 2.1284, 1.0006], device='cuda:0'), covar=tensor([0.7965, 0.2673, 0.3933, 0.7058], device='cuda:0'), in_proj_covar=tensor([0.1702, 0.1592, 0.1579, 0.1391], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 22:51:42,589 INFO [train.py:968] (0/2) Epoch 19, batch 22550, giga_loss[loss=0.2917, simple_loss=0.3677, pruned_loss=0.1078, over 28224.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.354, pruned_loss=0.1026, over 5720992.84 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3549, pruned_loss=0.1043, over 5788232.32 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3526, pruned_loss=0.1011, over 5709366.05 frames. ], batch size: 368, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:52:10,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.575e+02 1.308e+03 1.660e+03 2.322e+03 5.595e+03, threshold=3.320e+03, percent-clipped=7.0
+2023-03-09 22:52:24,609 INFO [train.py:968] (0/2) Epoch 19, batch 22600, giga_loss[loss=0.3127, simple_loss=0.3712, pruned_loss=0.1272, over 24011.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.352, pruned_loss=0.1018, over 5714019.19 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3548, pruned_loss=0.1044, over 5780786.31 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.351, pruned_loss=0.1005, over 5710504.53 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:52:34,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3373, 1.5608, 1.5057, 1.3799], device='cuda:0'), covar=tensor([0.1680, 0.1730, 0.2231, 0.1888], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0741, 0.0706, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 22:52:57,576 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 22:52:59,914 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-09 22:53:06,477 INFO [train.py:968] (0/2) Epoch 19, batch 22650, giga_loss[loss=0.2911, simple_loss=0.3713, pruned_loss=0.1055, over 27923.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3497, pruned_loss=0.1007, over 5711077.28 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3549, pruned_loss=0.1046, over 5774218.53 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3487, pruned_loss=0.09943, over 5713010.79 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:53:33,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.960e+02 1.117e+03 1.385e+03 1.843e+03 7.896e+03, threshold=2.769e+03, percent-clipped=4.0
+2023-03-09 22:53:40,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2332, 1.4869, 1.5492, 1.3087], device='cuda:0'), covar=tensor([0.2022, 0.1770, 0.2536, 0.2079], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0741, 0.0706, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 22:53:48,536 INFO [train.py:968] (0/2) Epoch 19, batch 22700, giga_loss[loss=0.2379, simple_loss=0.3176, pruned_loss=0.07907, over 28711.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3466, pruned_loss=0.09911, over 5711257.15 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3551, pruned_loss=0.1047, over 5774884.43 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3456, pruned_loss=0.098, over 5711813.13 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:53:53,737 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=845293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:54:29,116 INFO [train.py:968] (0/2) Epoch 19, batch 22750, giga_loss[loss=0.3297, simple_loss=0.379, pruned_loss=0.1402, over 26633.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3453, pruned_loss=0.09815, over 5705831.57 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3553, pruned_loss=0.1049, over 5766493.68 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3441, pruned_loss=0.09694, over 5712276.80 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:54:59,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.486e+02 1.184e+03 1.459e+03 1.929e+03 6.966e+03, threshold=2.918e+03, percent-clipped=10.0
+2023-03-09 22:55:11,943 INFO [train.py:968] (0/2) Epoch 19, batch 22800, giga_loss[loss=0.2588, simple_loss=0.3394, pruned_loss=0.08912, over 28959.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3474, pruned_loss=0.09809, over 5710503.94 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3563, pruned_loss=0.1057, over 5770246.67 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3452, pruned_loss=0.09609, over 5709801.43 frames. ], batch size: 106, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:55:33,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=845410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:55:54,804 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=845436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:55:56,009 INFO [train.py:968] (0/2) Epoch 19, batch 22850, giga_loss[loss=0.2524, simple_loss=0.3281, pruned_loss=0.0883, over 29036.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3477, pruned_loss=0.09773, over 5715219.76 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3563, pruned_loss=0.1058, over 5770769.44 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3458, pruned_loss=0.09601, over 5713922.25 frames. ], batch size: 128, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:55:57,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=845439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:56:19,957 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=845468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:56:22,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.049e+02 1.186e+03 1.479e+03 1.936e+03 7.229e+03, threshold=2.958e+03, percent-clipped=3.0
+2023-03-09 22:56:37,383 INFO [train.py:968] (0/2) Epoch 19, batch 22900, giga_loss[loss=0.26, simple_loss=0.3353, pruned_loss=0.09235, over 29021.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3463, pruned_loss=0.09768, over 5720916.41 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3564, pruned_loss=0.106, over 5769095.14 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3446, pruned_loss=0.09602, over 5720537.08 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:56:47,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4083, 1.8952, 1.3994, 0.6940], device='cuda:0'), covar=tensor([0.5721, 0.2729, 0.3113, 0.6653], device='cuda:0'), in_proj_covar=tensor([0.1691, 0.1586, 0.1569, 0.1383], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 22:57:00,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=845514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:57:19,925 INFO [train.py:968] (0/2) Epoch 19, batch 22950, giga_loss[loss=0.2553, simple_loss=0.331, pruned_loss=0.08979, over 28902.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3452, pruned_loss=0.09846, over 5720589.45 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3565, pruned_loss=0.1062, over 5770089.83 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3434, pruned_loss=0.0967, over 5718099.53 frames. ], batch size: 227, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:57:31,941 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=845553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:57:35,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=845556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:57:48,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.958e+02 1.214e+03 1.637e+03 2.278e+03 4.873e+03, threshold=3.274e+03, percent-clipped=11.0
+2023-03-09 22:57:58,480 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=845585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:58:00,411 INFO [train.py:968] (0/2) Epoch 19, batch 23000, giga_loss[loss=0.248, simple_loss=0.3222, pruned_loss=0.08695, over 28831.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3434, pruned_loss=0.09895, over 5721445.33 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3565, pruned_loss=0.1063, over 5772763.92 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3418, pruned_loss=0.09733, over 5716235.62 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:58:06,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4695, 3.4280, 1.5752, 1.5867], device='cuda:0'), covar=tensor([0.0929, 0.0445, 0.0867, 0.1293], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0539, 0.0372, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 22:58:13,537 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=845603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:58:42,801 INFO [train.py:968] (0/2) Epoch 19, batch 23050, giga_loss[loss=0.2694, simple_loss=0.3275, pruned_loss=0.1056, over 28469.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.342, pruned_loss=0.09906, over 5711682.16 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3569, pruned_loss=0.1067, over 5760765.67 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3403, pruned_loss=0.09744, over 5717861.46 frames. ], batch size: 65, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:59:10,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.899e+02 1.105e+03 1.290e+03 1.577e+03 3.540e+03, threshold=2.580e+03, percent-clipped=2.0
+2023-03-09 22:59:21,945 INFO [train.py:968] (0/2) Epoch 19, batch 23100, giga_loss[loss=0.2202, simple_loss=0.297, pruned_loss=0.07174, over 29116.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3407, pruned_loss=0.09862, over 5701226.10 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3577, pruned_loss=0.1073, over 5754165.61 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3383, pruned_loss=0.09655, over 5711784.60 frames. ], batch size: 128, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:00:02,510 INFO [train.py:968] (0/2) Epoch 19, batch 23150, libri_loss[loss=0.2611, simple_loss=0.3337, pruned_loss=0.09425, over 29583.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3372, pruned_loss=0.09694, over 5714579.69 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3575, pruned_loss=0.1074, over 5758573.09 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3347, pruned_loss=0.0948, over 5717047.11 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:00:17,742 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=845758.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:00:28,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.438e+02 1.271e+03 1.721e+03 2.360e+03 1.947e+04, threshold=3.442e+03, percent-clipped=21.0
+2023-03-09 23:00:41,160 INFO [train.py:968] (0/2) Epoch 19, batch 23200, giga_loss[loss=0.2403, simple_loss=0.3193, pruned_loss=0.08069, over 28561.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3339, pruned_loss=0.09542, over 5715639.85 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3577, pruned_loss=0.1078, over 5760772.67 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3308, pruned_loss=0.0929, over 5713964.98 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:01:21,758 INFO [train.py:968] (0/2) Epoch 19, batch 23250, giga_loss[loss=0.2691, simple_loss=0.3448, pruned_loss=0.09666, over 28941.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3308, pruned_loss=0.0936, over 5722088.63 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.358, pruned_loss=0.108, over 5761554.66 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3278, pruned_loss=0.09126, over 5719699.02 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:01:32,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=845851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:01:53,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.399e+02 1.152e+03 1.539e+03 2.300e+03 6.074e+03, threshold=3.078e+03, percent-clipped=5.0
+2023-03-09 23:02:04,710 INFO [train.py:968] (0/2) Epoch 19, batch 23300, giga_loss[loss=0.2997, simple_loss=0.3673, pruned_loss=0.116, over 28727.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3343, pruned_loss=0.09513, over 5708285.51 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3583, pruned_loss=0.1084, over 5753414.88 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3314, pruned_loss=0.09275, over 5712127.37 frames. ], batch size: 284, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:02:05,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=845889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:02:45,852 INFO [train.py:968] (0/2) Epoch 19, batch 23350, giga_loss[loss=0.2822, simple_loss=0.3566, pruned_loss=0.1039, over 28667.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3381, pruned_loss=0.09699, over 5698411.80 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3592, pruned_loss=0.1092, over 5745518.97 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.334, pruned_loss=0.09383, over 5707576.97 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:03:16,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.244e+02 1.162e+03 1.405e+03 1.855e+03 7.895e+03, threshold=2.811e+03, percent-clipped=4.0
+2023-03-09 23:03:19,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=845978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:03:27,753 INFO [train.py:968] (0/2) Epoch 19, batch 23400, giga_loss[loss=0.2681, simple_loss=0.3442, pruned_loss=0.09596, over 29032.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3421, pruned_loss=0.0983, over 5704975.72 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3597, pruned_loss=0.1096, over 5745406.67 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.338, pruned_loss=0.09525, over 5711556.31 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:03:38,277 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-846000.pt
+2023-03-09 23:04:04,960 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=846032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:04:08,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=846035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:04:10,320 INFO [train.py:968] (0/2) Epoch 19, batch 23450, giga_loss[loss=0.2404, simple_loss=0.3258, pruned_loss=0.07745, over 28949.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3443, pruned_loss=0.099, over 5701133.10 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3594, pruned_loss=0.1094, over 5738465.17 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3411, pruned_loss=0.09652, over 5711701.71 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:04:34,881 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=846064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:04:43,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.663e+02 1.210e+03 1.481e+03 2.108e+03 4.759e+03, threshold=2.962e+03, percent-clipped=10.0
+2023-03-09 23:04:55,503 INFO [train.py:968] (0/2) Epoch 19, batch 23500, giga_loss[loss=0.2647, simple_loss=0.3445, pruned_loss=0.09243, over 28688.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3462, pruned_loss=0.09974, over 5708065.02 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3596, pruned_loss=0.1096, over 5738258.79 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3433, pruned_loss=0.09759, over 5716174.61 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:04:58,055 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4619, 1.2098, 4.7198, 3.3585], device='cuda:0'), covar=tensor([0.1714, 0.2964, 0.0390, 0.1001], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0633, 0.0930, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 23:05:10,066 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1743, 1.1252, 3.9996, 3.2503], device='cuda:0'), covar=tensor([0.1732, 0.2839, 0.0425, 0.0950], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0632, 0.0930, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 23:05:29,199 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=846121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:05:31,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=846124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:05:38,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=846133.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 23:05:44,891 INFO [train.py:968] (0/2) Epoch 19, batch 23550, giga_loss[loss=0.3258, simple_loss=0.3919, pruned_loss=0.1299, over 28614.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3513, pruned_loss=0.104, over 5701132.06 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3596, pruned_loss=0.1096, over 5739977.70 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3489, pruned_loss=0.1022, over 5705571.08 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:05:59,244 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=846153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:06:19,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.111e+02 1.466e+03 1.943e+03 2.862e+03 1.307e+04, threshold=3.885e+03, percent-clipped=23.0
+2023-03-09 23:06:34,594 INFO [train.py:968] (0/2) Epoch 19, batch 23600, libri_loss[loss=0.2722, simple_loss=0.3355, pruned_loss=0.1045, over 29340.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.358, pruned_loss=0.1095, over 5684857.91 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.36, pruned_loss=0.1099, over 5731971.29 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3556, pruned_loss=0.1077, over 5693885.13 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:07:10,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=846226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:07:23,857 INFO [train.py:968] (0/2) Epoch 19, batch 23650, giga_loss[loss=0.3173, simple_loss=0.3857, pruned_loss=0.1244, over 28836.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3649, pruned_loss=0.1144, over 5670672.65 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3601, pruned_loss=0.1101, over 5725874.60 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3628, pruned_loss=0.1128, over 5682554.97 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:07:34,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9142, 1.9553, 1.4789, 1.5892], device='cuda:0'), covar=tensor([0.0741, 0.0507, 0.0913, 0.0930], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0444, 0.0511, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 23:07:59,610 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.845e+03 2.139e+03 3.143e+03 6.548e+03, threshold=4.279e+03, percent-clipped=13.0
+2023-03-09 23:08:01,974 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=846276.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:08:05,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=846279.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:08:13,323 INFO [train.py:968] (0/2) Epoch 19, batch 23700, giga_loss[loss=0.3048, simple_loss=0.3741, pruned_loss=0.1177, over 28713.00 frames. ], tot_loss[loss=0.305, simple_loss=0.371, pruned_loss=0.1195, over 5675585.80 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3608, pruned_loss=0.1107, over 5728838.38 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.369, pruned_loss=0.1178, over 5681063.34 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:08:35,989 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=846308.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 23:09:03,455 INFO [train.py:968] (0/2) Epoch 19, batch 23750, giga_loss[loss=0.3937, simple_loss=0.4474, pruned_loss=0.17, over 28251.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3776, pruned_loss=0.1251, over 5673651.09 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3608, pruned_loss=0.1109, over 5724291.80 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3763, pruned_loss=0.1239, over 5680691.49 frames. ], batch size: 368, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:09:32,189 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=846369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:09:36,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=846372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:09:40,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.173e+03 1.846e+03 2.331e+03 3.385e+03 7.475e+03, threshold=4.662e+03, percent-clipped=11.0
+2023-03-09 23:09:55,265 INFO [train.py:968] (0/2) Epoch 19, batch 23800, giga_loss[loss=0.4535, simple_loss=0.4524, pruned_loss=0.2273, over 23176.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3831, pruned_loss=0.1301, over 5670170.69 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3615, pruned_loss=0.1114, over 5729323.83 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.382, pruned_loss=0.1291, over 5670001.31 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:10:06,520 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=846401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:10:44,669 INFO [train.py:968] (0/2) Epoch 19, batch 23850, giga_loss[loss=0.3168, simple_loss=0.3793, pruned_loss=0.1272, over 28703.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3842, pruned_loss=0.1319, over 5647777.08 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3618, pruned_loss=0.1116, over 5713440.22 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3833, pruned_loss=0.1311, over 5661119.03 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:10:47,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8067, 3.6339, 3.4685, 1.8079], device='cuda:0'), covar=tensor([0.0774, 0.0887, 0.0831, 0.2146], device='cuda:0'), in_proj_covar=tensor([0.1180, 0.1101, 0.0935, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 23:11:18,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6560, 1.8706, 1.5510, 1.9205], device='cuda:0'), covar=tensor([0.2042, 0.2033, 0.2096, 0.1873], device='cuda:0'), in_proj_covar=tensor([0.1451, 0.1054, 0.1287, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 23:11:21,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.871e+03 2.575e+03 3.763e+03 1.655e+04, threshold=5.150e+03, percent-clipped=15.0
+2023-03-09 23:11:34,721 INFO [train.py:968] (0/2) Epoch 19, batch 23900, libri_loss[loss=0.3362, simple_loss=0.3851, pruned_loss=0.1436, over 29487.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3879, pruned_loss=0.136, over 5645991.48 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3621, pruned_loss=0.112, over 5711368.55 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3876, pruned_loss=0.1357, over 5656632.13 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:11:35,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4047, 3.1447, 1.5305, 1.5488], device='cuda:0'), covar=tensor([0.0943, 0.0361, 0.0859, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0546, 0.0376, 0.0418], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-09 23:12:23,931 INFO [train.py:968] (0/2) Epoch 19, batch 23950, giga_loss[loss=0.3791, simple_loss=0.4159, pruned_loss=0.1712, over 28224.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.391, pruned_loss=0.1397, over 5643029.64 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3622, pruned_loss=0.1122, over 5714537.55 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3912, pruned_loss=0.1399, over 5647205.52 frames. ], batch size: 368, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:12:32,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5846, 1.6278, 1.8205, 1.4444], device='cuda:0'), covar=tensor([0.1151, 0.1765, 0.0998, 0.1368], device='cuda:0'), in_proj_covar=tensor([0.0878, 0.0693, 0.0921, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 23:12:51,131 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-09 23:12:54,939 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-09 23:12:58,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9959, 1.3117, 1.1093, 0.1373], device='cuda:0'), covar=tensor([0.3225, 0.2598, 0.3320, 0.5428], device='cuda:0'), in_proj_covar=tensor([0.1706, 0.1606, 0.1580, 0.1396], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-09 23:13:10,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.927e+03 2.533e+03 3.207e+03 1.047e+04, threshold=5.066e+03, percent-clipped=5.0
+2023-03-09 23:13:16,727 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3274, 1.2215, 1.2298, 1.4464], device='cuda:0'), covar=tensor([0.0764, 0.0367, 0.0325, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0103], device='cuda:0')
+2023-03-09 23:13:23,134 INFO [train.py:968] (0/2) Epoch 19, batch 24000, giga_loss[loss=0.333, simple_loss=0.3874, pruned_loss=0.1393, over 28928.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.394, pruned_loss=0.1416, over 5628978.55 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3623, pruned_loss=0.1124, over 5698758.95 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3947, pruned_loss=0.1422, over 5643819.36 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:13:23,138 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-09 23:13:31,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3521, 3.1991, 1.4559, 1.5466], device='cuda:0'), covar=tensor([0.1104, 0.0428, 0.0995, 0.1504], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0546, 0.0375, 0.0418], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:0')
+2023-03-09 23:13:32,000 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2059, simple_loss=0.313, pruned_loss=0.04942, over 944034.00 frames. 
+2023-03-09 23:13:32,001 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-09 23:14:25,683 INFO [train.py:968] (0/2) Epoch 19, batch 24050, giga_loss[loss=0.2864, simple_loss=0.349, pruned_loss=0.112, over 28774.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3934, pruned_loss=0.142, over 5636156.27 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3625, pruned_loss=0.1126, over 5703603.15 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3948, pruned_loss=0.1431, over 5642076.04 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:15:00,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.190e+03 1.875e+03 2.317e+03 3.191e+03 8.076e+03, threshold=4.634e+03, percent-clipped=6.0
+2023-03-09 23:15:12,440 INFO [train.py:968] (0/2) Epoch 19, batch 24100, giga_loss[loss=0.2763, simple_loss=0.3427, pruned_loss=0.105, over 28801.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3906, pruned_loss=0.1402, over 5645134.81 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3623, pruned_loss=0.1125, over 5710797.85 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.393, pruned_loss=0.1423, over 5640731.48 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:16:03,938 INFO [train.py:968] (0/2) Epoch 19, batch 24150, giga_loss[loss=0.3114, simple_loss=0.3759, pruned_loss=0.1235, over 28847.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3902, pruned_loss=0.1399, over 5636068.72 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3625, pruned_loss=0.1128, over 5704220.85 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3925, pruned_loss=0.1418, over 5638077.91 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:16:22,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2542, 1.2525, 1.3203, 1.4934], device='cuda:0'), covar=tensor([0.0795, 0.0365, 0.0322, 0.0850], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0104], device='cuda:0')
+2023-03-09 23:16:27,752 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 23:16:37,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.196e+03 1.664e+03 2.096e+03 2.989e+03 6.147e+03, threshold=4.191e+03, percent-clipped=7.0
+2023-03-09 23:16:50,059 INFO [train.py:968] (0/2) Epoch 19, batch 24200, libri_loss[loss=0.2753, simple_loss=0.3449, pruned_loss=0.1029, over 29668.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3894, pruned_loss=0.1385, over 5621887.62 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3621, pruned_loss=0.1128, over 5690137.57 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3925, pruned_loss=0.1409, over 5634095.46 frames. ], batch size: 69, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:17:46,509 INFO [train.py:968] (0/2) Epoch 19, batch 24250, giga_loss[loss=0.3487, simple_loss=0.4039, pruned_loss=0.1467, over 28504.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3917, pruned_loss=0.1401, over 5617788.43 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3621, pruned_loss=0.1128, over 5688660.82 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3944, pruned_loss=0.1422, over 5628208.44 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:18:27,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.675e+02 1.638e+03 2.188e+03 2.783e+03 7.047e+03, threshold=4.375e+03, percent-clipped=9.0
+2023-03-09 23:18:30,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9317, 1.9485, 2.1606, 1.6824], device='cuda:0'), covar=tensor([0.1756, 0.2334, 0.1384, 0.1659], device='cuda:0'), in_proj_covar=tensor([0.0878, 0.0694, 0.0922, 0.0823], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 23:18:38,274 INFO [train.py:968] (0/2) Epoch 19, batch 24300, giga_loss[loss=0.2821, simple_loss=0.3574, pruned_loss=0.1034, over 29014.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3899, pruned_loss=0.1381, over 5623705.69 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.362, pruned_loss=0.1128, over 5692044.30 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3927, pruned_loss=0.1403, over 5627535.04 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:19:18,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3933, 1.7417, 1.3338, 1.2766], device='cuda:0'), covar=tensor([0.2839, 0.2719, 0.3236, 0.2511], device='cuda:0'), in_proj_covar=tensor([0.1455, 0.1055, 0.1288, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 23:19:31,410 INFO [train.py:968] (0/2) Epoch 19, batch 24350, giga_loss[loss=0.3554, simple_loss=0.4028, pruned_loss=0.1539, over 27557.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3869, pruned_loss=0.1344, over 5625383.26 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3615, pruned_loss=0.1125, over 5694432.83 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3897, pruned_loss=0.1366, over 5625532.44 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:20:09,660 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.914e+02 1.584e+03 2.222e+03 2.962e+03 7.296e+03, threshold=4.443e+03, percent-clipped=12.0
+2023-03-09 23:20:15,089 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-09 23:20:21,928 INFO [train.py:968] (0/2) Epoch 19, batch 24400, giga_loss[loss=0.3797, simple_loss=0.4105, pruned_loss=0.1744, over 26639.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3823, pruned_loss=0.1298, over 5639484.15 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.361, pruned_loss=0.1123, over 5694636.00 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3856, pruned_loss=0.1322, over 5638246.79 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:21:06,679 INFO [train.py:968] (0/2) Epoch 19, batch 24450, giga_loss[loss=0.3767, simple_loss=0.421, pruned_loss=0.1662, over 27940.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3797, pruned_loss=0.1271, over 5661227.73 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3609, pruned_loss=0.1122, over 5701003.92 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3832, pruned_loss=0.1297, over 5653004.20 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:21:45,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.799e+03 2.230e+03 3.104e+03 6.130e+03, threshold=4.460e+03, percent-clipped=5.0
+2023-03-09 23:21:49,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4277, 1.7922, 1.3843, 1.4618], device='cuda:0'), covar=tensor([0.0739, 0.0319, 0.0323, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0104], device='cuda:0')
+2023-03-09 23:21:54,632 INFO [train.py:968] (0/2) Epoch 19, batch 24500, giga_loss[loss=0.2791, simple_loss=0.3496, pruned_loss=0.1043, over 28773.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3764, pruned_loss=0.1248, over 5662616.51 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3606, pruned_loss=0.1121, over 5706232.72 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3798, pruned_loss=0.1273, over 5650644.93 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:22:44,858 INFO [train.py:968] (0/2) Epoch 19, batch 24550, giga_loss[loss=0.2746, simple_loss=0.3485, pruned_loss=0.1004, over 28247.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3753, pruned_loss=0.1238, over 5662318.57 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3608, pruned_loss=0.1122, over 5696165.09 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3781, pruned_loss=0.1258, over 5661457.55 frames. ], batch size: 65, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:23:28,097 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.615e+03 2.188e+03 2.532e+03 5.880e+03, threshold=4.376e+03, percent-clipped=3.0
+2023-03-09 23:23:38,243 INFO [train.py:968] (0/2) Epoch 19, batch 24600, giga_loss[loss=0.2879, simple_loss=0.3667, pruned_loss=0.1045, over 28974.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3761, pruned_loss=0.1244, over 5662747.83 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3611, pruned_loss=0.1126, over 5698549.25 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3785, pruned_loss=0.126, over 5659085.85 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:24:26,645 INFO [train.py:968] (0/2) Epoch 19, batch 24650, giga_loss[loss=0.2774, simple_loss=0.3497, pruned_loss=0.1025, over 28702.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3728, pruned_loss=0.1214, over 5667804.52 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3609, pruned_loss=0.1126, over 5702209.04 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3753, pruned_loss=0.1231, over 5660669.19 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:25:04,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.674e+02 1.674e+03 2.057e+03 2.852e+03 9.182e+03, threshold=4.114e+03, percent-clipped=3.0
+2023-03-09 23:25:16,509 INFO [train.py:968] (0/2) Epoch 19, batch 24700, giga_loss[loss=0.3601, simple_loss=0.4197, pruned_loss=0.1502, over 28005.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3734, pruned_loss=0.1195, over 5684105.53 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3608, pruned_loss=0.1127, over 5707488.21 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.376, pruned_loss=0.1211, over 5672726.08 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:26:12,398 INFO [train.py:968] (0/2) Epoch 19, batch 24750, libri_loss[loss=0.295, simple_loss=0.3415, pruned_loss=0.1243, over 29674.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3743, pruned_loss=0.1193, over 5662528.30 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3609, pruned_loss=0.1128, over 5709878.34 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3765, pruned_loss=0.1205, over 5650786.35 frames. ], batch size: 73, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:26:47,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.086e+03 1.720e+03 2.140e+03 2.929e+03 5.030e+03, threshold=4.280e+03, percent-clipped=4.0
+2023-03-09 23:26:58,593 INFO [train.py:968] (0/2) Epoch 19, batch 24800, libri_loss[loss=0.2327, simple_loss=0.3057, pruned_loss=0.0798, over 29618.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3751, pruned_loss=0.1204, over 5670441.32 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3605, pruned_loss=0.1126, over 5712273.87 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3779, pruned_loss=0.1219, over 5656986.42 frames. ], batch size: 74, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:27:18,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1952, 1.2570, 3.1891, 2.9027], device='cuda:0'), covar=tensor([0.1432, 0.2492, 0.0510, 0.2007], device='cuda:0'), in_proj_covar=tensor([0.0743, 0.0637, 0.0942, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 23:27:26,416 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=847417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:27:28,908 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=847420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:27:44,898 INFO [train.py:968] (0/2) Epoch 19, batch 24850, giga_loss[loss=0.2853, simple_loss=0.3585, pruned_loss=0.1061, over 28675.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3765, pruned_loss=0.1222, over 5670351.20 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3607, pruned_loss=0.1128, over 5717648.65 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.379, pruned_loss=0.1235, over 5653482.86 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:28:24,856 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=847475.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:28:28,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.674e+03 2.361e+03 3.060e+03 5.534e+03, threshold=4.721e+03, percent-clipped=8.0
+2023-03-09 23:28:38,389 INFO [train.py:968] (0/2) Epoch 19, batch 24900, giga_loss[loss=0.2865, simple_loss=0.3545, pruned_loss=0.1093, over 28805.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.374, pruned_loss=0.1218, over 5663288.92 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3603, pruned_loss=0.1126, over 5720269.41 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3766, pruned_loss=0.1232, over 5646966.54 frames. ], batch size: 284, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:28:38,582 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=847488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:29:23,101 INFO [train.py:968] (0/2) Epoch 19, batch 24950, giga_loss[loss=0.3294, simple_loss=0.3925, pruned_loss=0.1332, over 28775.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3726, pruned_loss=0.1214, over 5666895.30 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3605, pruned_loss=0.1126, over 5713666.20 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3746, pruned_loss=0.1226, over 5658940.70 frames. ], batch size: 284, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:30:01,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.502e+02 1.701e+03 2.131e+03 2.793e+03 6.566e+03, threshold=4.262e+03, percent-clipped=7.0
+2023-03-09 23:30:09,935 INFO [train.py:968] (0/2) Epoch 19, batch 25000, giga_loss[loss=0.3139, simple_loss=0.3788, pruned_loss=0.1245, over 28681.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3722, pruned_loss=0.1211, over 5671105.42 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3602, pruned_loss=0.1125, over 5716434.63 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3742, pruned_loss=0.1222, over 5661840.97 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:30:55,502 INFO [train.py:968] (0/2) Epoch 19, batch 25050, giga_loss[loss=0.3492, simple_loss=0.3999, pruned_loss=0.1492, over 27603.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.371, pruned_loss=0.1192, over 5678548.06 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.36, pruned_loss=0.1124, over 5717202.72 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3732, pruned_loss=0.1204, over 5669342.82 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:31:21,122 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-09 23:31:38,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 1.670e+03 2.246e+03 3.061e+03 6.482e+03, threshold=4.493e+03, percent-clipped=8.0
+2023-03-09 23:31:45,982 INFO [train.py:968] (0/2) Epoch 19, batch 25100, giga_loss[loss=0.2944, simple_loss=0.375, pruned_loss=0.1069, over 29022.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3709, pruned_loss=0.119, over 5671007.44 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3596, pruned_loss=0.1122, over 5717972.33 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3731, pruned_loss=0.1202, over 5662806.27 frames. ], batch size: 164, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:31:54,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8324, 1.0740, 2.8551, 2.6883], device='cuda:0'), covar=tensor([0.1661, 0.2589, 0.0627, 0.1291], device='cuda:0'), in_proj_covar=tensor([0.0746, 0.0637, 0.0945, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-09 23:32:01,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9998, 1.0821, 3.3387, 3.0218], device='cuda:0'), covar=tensor([0.1791, 0.2811, 0.0534, 0.1053], device='cuda:0'), in_proj_covar=tensor([0.0746, 0.0637, 0.0945, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-09 23:32:35,497 INFO [train.py:968] (0/2) Epoch 19, batch 25150, giga_loss[loss=0.3369, simple_loss=0.3867, pruned_loss=0.1435, over 28556.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.371, pruned_loss=0.1192, over 5654024.82 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3599, pruned_loss=0.1124, over 5698641.09 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3727, pruned_loss=0.1201, over 5664744.34 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:33:15,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.746e+02 1.523e+03 2.146e+03 2.792e+03 9.621e+03, threshold=4.291e+03, percent-clipped=6.0
+2023-03-09 23:33:23,996 INFO [train.py:968] (0/2) Epoch 19, batch 25200, giga_loss[loss=0.3113, simple_loss=0.3706, pruned_loss=0.126, over 28838.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3703, pruned_loss=0.1188, over 5665390.47 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3603, pruned_loss=0.1126, over 5693737.69 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3715, pruned_loss=0.1195, over 5677649.14 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:33:28,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=847792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:33:31,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=847795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:34:10,114 INFO [train.py:968] (0/2) Epoch 19, batch 25250, giga_loss[loss=0.308, simple_loss=0.365, pruned_loss=0.1255, over 28481.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3688, pruned_loss=0.1186, over 5667500.59 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3603, pruned_loss=0.1128, over 5697128.72 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.37, pruned_loss=0.1192, over 5673479.87 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:34:21,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=847850.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 23:34:24,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 23:34:31,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=847863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:34:46,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.758e+03 2.302e+03 3.107e+03 6.359e+03, threshold=4.605e+03, percent-clipped=6.0
+2023-03-09 23:34:51,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3765, 3.2101, 3.0769, 1.4145], device='cuda:0'), covar=tensor([0.0864, 0.0957, 0.0903, 0.2166], device='cuda:0'), in_proj_covar=tensor([0.1193, 0.1112, 0.0944, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 23:34:53,677 INFO [train.py:968] (0/2) Epoch 19, batch 25300, giga_loss[loss=0.3803, simple_loss=0.4142, pruned_loss=0.1732, over 27980.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3677, pruned_loss=0.1182, over 5677103.57 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3607, pruned_loss=0.113, over 5691505.48 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3686, pruned_loss=0.1187, over 5686008.51 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:35:03,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3958, 1.6955, 1.5905, 1.5402], device='cuda:0'), covar=tensor([0.1866, 0.1806, 0.2188, 0.1858], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0744, 0.0706, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 23:35:08,524 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-09 23:35:40,818 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=847935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:35:43,027 INFO [train.py:968] (0/2) Epoch 19, batch 25350, giga_loss[loss=0.3292, simple_loss=0.3659, pruned_loss=0.1462, over 23585.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3656, pruned_loss=0.1172, over 5682651.93 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3603, pruned_loss=0.1127, over 5698249.49 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3669, pruned_loss=0.118, over 5683427.13 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:35:43,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:35:43,861 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=847938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:35:47,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:36:09,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=847967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:36:11,889 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=847970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:36:20,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.155e+02 1.802e+03 2.379e+03 2.873e+03 7.101e+03, threshold=4.758e+03, percent-clipped=5.0
+2023-03-09 23:36:26,813 INFO [train.py:968] (0/2) Epoch 19, batch 25400, giga_loss[loss=0.2972, simple_loss=0.3629, pruned_loss=0.1157, over 28793.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3647, pruned_loss=0.1173, over 5677573.60 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.36, pruned_loss=0.1126, over 5693710.89 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3661, pruned_loss=0.1181, over 5681793.85 frames. ], batch size: 243, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:36:33,036 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=847993.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 23:36:38,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847996.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:36:41,969 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-848000.pt
+2023-03-09 23:36:48,239 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=848006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:36:50,686 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=848009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:37:07,920 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=848025.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 23:37:19,828 INFO [train.py:968] (0/2) Epoch 19, batch 25450, giga_loss[loss=0.2721, simple_loss=0.3509, pruned_loss=0.09666, over 28936.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3657, pruned_loss=0.1185, over 5674871.27 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.36, pruned_loss=0.1128, over 5694626.33 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3669, pruned_loss=0.119, over 5677322.74 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:37:20,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=848038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:38:00,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.913e+02 1.656e+03 2.063e+03 2.894e+03 7.347e+03, threshold=4.127e+03, percent-clipped=9.0
+2023-03-09 23:38:09,329 INFO [train.py:968] (0/2) Epoch 19, batch 25500, giga_loss[loss=0.3362, simple_loss=0.3889, pruned_loss=0.1418, over 27531.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3664, pruned_loss=0.118, over 5680601.84 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.36, pruned_loss=0.1128, over 5697865.76 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3674, pruned_loss=0.1185, over 5679556.45 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:38:22,965 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 23:38:51,227 INFO [train.py:968] (0/2) Epoch 19, batch 25550, giga_loss[loss=0.2886, simple_loss=0.3646, pruned_loss=0.1063, over 28952.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3674, pruned_loss=0.1181, over 5683682.51 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3603, pruned_loss=0.113, over 5702468.71 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3681, pruned_loss=0.1185, over 5678349.01 frames. ], batch size: 227, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:39:30,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.386e+02 1.591e+03 2.228e+03 3.484e+03 9.057e+03, threshold=4.456e+03, percent-clipped=11.0
+2023-03-09 23:39:40,233 INFO [train.py:968] (0/2) Epoch 19, batch 25600, giga_loss[loss=0.273, simple_loss=0.3523, pruned_loss=0.09681, over 28733.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3675, pruned_loss=0.1174, over 5678692.81 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3604, pruned_loss=0.113, over 5697707.90 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3682, pruned_loss=0.1178, over 5678407.48 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:39:58,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9964, 3.7935, 3.6297, 1.8528], device='cuda:0'), covar=tensor([0.0799, 0.0991, 0.1118, 0.2081], device='cuda:0'), in_proj_covar=tensor([0.1195, 0.1113, 0.0946, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 23:40:22,985 INFO [train.py:968] (0/2) Epoch 19, batch 25650, giga_loss[loss=0.281, simple_loss=0.3511, pruned_loss=0.1055, over 28963.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3674, pruned_loss=0.1177, over 5678230.51 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.113, over 5696910.66 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3684, pruned_loss=0.1183, over 5677316.67 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:40:45,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2243, 4.0554, 3.9071, 2.0018], device='cuda:0'), covar=tensor([0.0600, 0.0752, 0.0762, 0.1994], device='cuda:0'), in_proj_covar=tensor([0.1196, 0.1115, 0.0948, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 23:41:02,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.606e+03 2.178e+03 3.007e+03 8.347e+03, threshold=4.356e+03, percent-clipped=3.0
+2023-03-09 23:41:09,592 INFO [train.py:968] (0/2) Epoch 19, batch 25700, giga_loss[loss=0.3106, simple_loss=0.3697, pruned_loss=0.1258, over 28902.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3696, pruned_loss=0.1198, over 5684826.69 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.113, over 5702143.04 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3708, pruned_loss=0.1205, over 5678767.84 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:41:57,640 INFO [train.py:968] (0/2) Epoch 19, batch 25750, giga_loss[loss=0.2757, simple_loss=0.3573, pruned_loss=0.09703, over 28845.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.37, pruned_loss=0.1215, over 5677045.18 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5696567.50 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.372, pruned_loss=0.1228, over 5676008.86 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:42:21,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8228, 4.6671, 4.4360, 2.1438], device='cuda:0'), covar=tensor([0.0573, 0.0758, 0.0804, 0.2021], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.1118, 0.0951, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 23:42:37,718 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 23:42:40,375 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.853e+03 2.235e+03 3.234e+03 1.093e+04, threshold=4.470e+03, percent-clipped=15.0
+2023-03-09 23:42:46,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 23:42:49,055 INFO [train.py:968] (0/2) Epoch 19, batch 25800, giga_loss[loss=0.2995, simple_loss=0.3687, pruned_loss=0.1151, over 29050.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3712, pruned_loss=0.1235, over 5665783.47 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1128, over 5688023.45 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3727, pruned_loss=0.1245, over 5672338.69 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:43:33,769 INFO [train.py:968] (0/2) Epoch 19, batch 25850, giga_loss[loss=0.3206, simple_loss=0.3793, pruned_loss=0.1309, over 28593.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3709, pruned_loss=0.1232, over 5676158.50 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3599, pruned_loss=0.1128, over 5690345.02 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3721, pruned_loss=0.1242, over 5679007.57 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:44:12,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.734e+03 2.266e+03 3.177e+03 9.089e+03, threshold=4.532e+03, percent-clipped=7.0
+2023-03-09 23:44:19,732 INFO [train.py:968] (0/2) Epoch 19, batch 25900, giga_loss[loss=0.3436, simple_loss=0.3955, pruned_loss=0.1458, over 28612.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3696, pruned_loss=0.123, over 5662356.46 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3601, pruned_loss=0.1132, over 5686681.07 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3708, pruned_loss=0.1237, over 5666578.26 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:44:25,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2628, 1.4203, 1.4894, 1.3271], device='cuda:0'), covar=tensor([0.1388, 0.1375, 0.1876, 0.1453], device='cuda:0'), in_proj_covar=tensor([0.0465, 0.0741, 0.0704, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-09 23:44:53,979 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:45:00,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6841, 4.5015, 4.2599, 2.1295], device='cuda:0'), covar=tensor([0.0478, 0.0633, 0.0685, 0.2100], device='cuda:0'), in_proj_covar=tensor([0.1196, 0.1115, 0.0948, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 23:45:05,497 INFO [train.py:968] (0/2) Epoch 19, batch 25950, giga_loss[loss=0.2707, simple_loss=0.3517, pruned_loss=0.09487, over 28834.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.37, pruned_loss=0.1218, over 5670665.90 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.36, pruned_loss=0.1131, over 5690141.98 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3711, pruned_loss=0.1225, over 5670647.45 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:45:43,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.638e+03 2.080e+03 3.029e+03 9.980e+03, threshold=4.160e+03, percent-clipped=8.0
+2023-03-09 23:45:51,862 INFO [train.py:968] (0/2) Epoch 19, batch 26000, giga_loss[loss=0.2627, simple_loss=0.3472, pruned_loss=0.08913, over 28670.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1205, over 5658727.69 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3605, pruned_loss=0.1136, over 5685336.62 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3694, pruned_loss=0.1208, over 5663164.65 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:46:06,693 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:46:40,972 INFO [train.py:968] (0/2) Epoch 19, batch 26050, giga_loss[loss=0.2993, simple_loss=0.3636, pruned_loss=0.1176, over 28902.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.367, pruned_loss=0.1197, over 5657443.88 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3604, pruned_loss=0.1136, over 5686367.94 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3677, pruned_loss=0.12, over 5659937.09 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:46:41,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8231, 4.6417, 4.3821, 1.9339], device='cuda:0'), covar=tensor([0.0680, 0.0856, 0.0988, 0.2138], device='cuda:0'), in_proj_covar=tensor([0.1195, 0.1112, 0.0946, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 23:47:05,089 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-09 23:47:17,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.972e+03 2.564e+03 3.343e+03 1.074e+04, threshold=5.128e+03, percent-clipped=14.0
+2023-03-09 23:47:24,649 INFO [train.py:968] (0/2) Epoch 19, batch 26100, giga_loss[loss=0.3077, simple_loss=0.3656, pruned_loss=0.1249, over 28877.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3664, pruned_loss=0.1201, over 5668442.70 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3603, pruned_loss=0.1135, over 5693391.48 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3673, pruned_loss=0.1207, over 5663601.54 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:47:45,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0490, 5.8837, 5.5406, 3.0070], device='cuda:0'), covar=tensor([0.0513, 0.0653, 0.0829, 0.1568], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.1115, 0.0949, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-09 23:47:53,707 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:48:03,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3014, 1.1420, 3.7223, 3.2080], device='cuda:0'), covar=tensor([0.1562, 0.2900, 0.0477, 0.2060], device='cuda:0'), in_proj_covar=tensor([0.0749, 0.0642, 0.0946, 0.0885], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-09 23:48:13,907 INFO [train.py:968] (0/2) Epoch 19, batch 26150, giga_loss[loss=0.3327, simple_loss=0.3816, pruned_loss=0.1419, over 28719.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3673, pruned_loss=0.1213, over 5663669.80 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3599, pruned_loss=0.1131, over 5699432.23 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3687, pruned_loss=0.1225, over 5652456.59 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:48:44,602 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 23:48:51,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.752e+02 1.652e+03 2.349e+03 3.197e+03 1.069e+04, threshold=4.698e+03, percent-clipped=8.0
+2023-03-09 23:48:57,395 INFO [train.py:968] (0/2) Epoch 19, batch 26200, giga_loss[loss=0.3596, simple_loss=0.4123, pruned_loss=0.1535, over 28526.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3698, pruned_loss=0.1225, over 5659862.95 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3599, pruned_loss=0.1133, over 5694505.89 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3711, pruned_loss=0.1235, over 5654426.35 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:49:10,739 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:49:48,103 INFO [train.py:968] (0/2) Epoch 19, batch 26250, giga_loss[loss=0.3409, simple_loss=0.4055, pruned_loss=0.1381, over 28044.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3733, pruned_loss=0.1218, over 5665257.89 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3595, pruned_loss=0.1131, over 5697587.13 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3747, pruned_loss=0.1228, over 5657959.24 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:50:27,009 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:50:29,840 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.762e+03 2.142e+03 3.604e+03 1.094e+04, threshold=4.284e+03, percent-clipped=16.0
+2023-03-09 23:50:36,187 INFO [train.py:968] (0/2) Epoch 19, batch 26300, giga_loss[loss=0.2708, simple_loss=0.3473, pruned_loss=0.09709, over 28777.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3738, pruned_loss=0.1215, over 5663842.26 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3594, pruned_loss=0.113, over 5701560.13 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3755, pruned_loss=0.1226, over 5653621.74 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:50:46,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=848901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:50:53,489 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 23:51:13,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3481, 1.5383, 1.0872, 1.0591], device='cuda:0'), covar=tensor([0.0984, 0.0556, 0.1162, 0.1160], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0446, 0.0511, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 23:51:21,836 INFO [train.py:968] (0/2) Epoch 19, batch 26350, giga_loss[loss=0.3105, simple_loss=0.3779, pruned_loss=0.1216, over 28672.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3755, pruned_loss=0.1235, over 5662445.54 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3594, pruned_loss=0.1134, over 5709173.39 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3776, pruned_loss=0.1245, over 5645253.00 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:52:00,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:52:01,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=848979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:52:03,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.292e+03 1.874e+03 2.375e+03 3.103e+03 1.328e+04, threshold=4.751e+03, percent-clipped=8.0
+2023-03-09 23:52:08,510 INFO [train.py:968] (0/2) Epoch 19, batch 26400, giga_loss[loss=0.28, simple_loss=0.3555, pruned_loss=0.1022, over 28921.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3767, pruned_loss=0.1247, over 5671134.66 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3595, pruned_loss=0.1134, over 5710277.52 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3783, pruned_loss=0.1255, over 5656539.64 frames. ], batch size: 106, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:53:00,744 INFO [train.py:968] (0/2) Epoch 19, batch 26450, giga_loss[loss=0.3476, simple_loss=0.4074, pruned_loss=0.1439, over 28725.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3769, pruned_loss=0.1256, over 5662498.95 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3596, pruned_loss=0.1135, over 5711713.86 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3784, pruned_loss=0.1264, over 5647947.66 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:53:06,625 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:53:10,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:53:39,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:53:43,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.583e+03 2.290e+03 3.208e+03 6.277e+03, threshold=4.580e+03, percent-clipped=5.0
+2023-03-09 23:53:47,893 INFO [train.py:968] (0/2) Epoch 19, batch 26500, giga_loss[loss=0.3022, simple_loss=0.3622, pruned_loss=0.1212, over 28904.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.376, pruned_loss=0.1256, over 5663169.84 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3598, pruned_loss=0.1135, over 5716778.45 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3776, pruned_loss=0.1267, over 5644781.33 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:53:50,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=849091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:53:56,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5754, 1.8010, 1.8293, 1.3651], device='cuda:0'), covar=tensor([0.1678, 0.2705, 0.1455, 0.1803], device='cuda:0'), in_proj_covar=tensor([0.0883, 0.0700, 0.0927, 0.0826], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 23:54:17,259 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:54:19,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:54:32,200 INFO [train.py:968] (0/2) Epoch 19, batch 26550, giga_loss[loss=0.2981, simple_loss=0.3595, pruned_loss=0.1184, over 28832.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3733, pruned_loss=0.1244, over 5668731.83 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3598, pruned_loss=0.1137, over 5722425.42 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3751, pruned_loss=0.1255, over 5647010.39 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:54:45,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:55:12,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=849176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:55:16,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.011e+02 1.879e+03 2.649e+03 3.627e+03 1.203e+04, threshold=5.298e+03, percent-clipped=13.0
+2023-03-09 23:55:21,690 INFO [train.py:968] (0/2) Epoch 19, batch 26600, giga_loss[loss=0.3178, simple_loss=0.376, pruned_loss=0.1297, over 28908.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3733, pruned_loss=0.1254, over 5654532.87 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.36, pruned_loss=0.1139, over 5714320.82 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3749, pruned_loss=0.1263, over 5644011.68 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:56:04,843 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:56:07,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:56:07,594 INFO [train.py:968] (0/2) Epoch 19, batch 26650, libri_loss[loss=0.3101, simple_loss=0.373, pruned_loss=0.1236, over 29675.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3743, pruned_loss=0.1261, over 5656225.20 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3601, pruned_loss=0.1141, over 5720716.12 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3759, pruned_loss=0.1271, over 5639782.69 frames. ], batch size: 91, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:56:22,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=849254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:56:22,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8356, 1.8767, 1.3784, 1.4505], device='cuda:0'), covar=tensor([0.0844, 0.0615, 0.0999, 0.1163], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0448, 0.0514, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-09 23:56:32,782 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:56:44,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.812e+03 2.428e+03 3.205e+03 6.635e+03, threshold=4.856e+03, percent-clipped=6.0
+2023-03-09 23:56:50,329 INFO [train.py:968] (0/2) Epoch 19, batch 26700, giga_loss[loss=0.3012, simple_loss=0.365, pruned_loss=0.1187, over 28966.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3737, pruned_loss=0.1258, over 5666689.57 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3607, pruned_loss=0.1144, over 5723155.00 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3748, pruned_loss=0.1266, over 5650050.19 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:57:16,703 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:57:19,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:57:28,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4710, 1.8437, 1.8040, 1.3779], device='cuda:0'), covar=tensor([0.3125, 0.2458, 0.2528, 0.2965], device='cuda:0'), in_proj_covar=tensor([0.1919, 0.1847, 0.1791, 0.1929], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-09 23:57:31,782 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=849336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:57:34,543 INFO [train.py:968] (0/2) Epoch 19, batch 26750, giga_loss[loss=0.3682, simple_loss=0.4137, pruned_loss=0.1613, over 27672.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.371, pruned_loss=0.124, over 5671616.61 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.114, over 5716729.25 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3727, pruned_loss=0.1253, over 5661642.60 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:57:47,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:57:50,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=849353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:58:17,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.947e+02 1.781e+03 2.480e+03 3.072e+03 6.062e+03, threshold=4.959e+03, percent-clipped=6.0
+2023-03-09 23:58:24,694 INFO [train.py:968] (0/2) Epoch 19, batch 26800, giga_loss[loss=0.2814, simple_loss=0.3556, pruned_loss=0.1036, over 28969.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3704, pruned_loss=0.1236, over 5660897.85 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.1141, over 5706788.70 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3719, pruned_loss=0.1247, over 5661586.15 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:58:25,841 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6462, 1.6395, 1.9151, 1.4337], device='cuda:0'), covar=tensor([0.1785, 0.2374, 0.1391, 0.1696], device='cuda:0'), in_proj_covar=tensor([0.0885, 0.0701, 0.0928, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-09 23:58:33,037 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:58:37,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:59:02,699 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:59:10,923 INFO [train.py:968] (0/2) Epoch 19, batch 26850, giga_loss[loss=0.3206, simple_loss=0.3861, pruned_loss=0.1275, over 28532.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3724, pruned_loss=0.124, over 5655220.27 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3605, pruned_loss=0.1144, over 5700053.57 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3736, pruned_loss=0.1247, over 5661461.27 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:59:30,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5577, 3.6325, 1.6337, 1.6303], device='cuda:0'), covar=tensor([0.0972, 0.0385, 0.0901, 0.1400], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0550, 0.0377, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-09 23:59:31,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8153, 2.0546, 1.6017, 2.0047], device='cuda:0'), covar=tensor([0.2385, 0.2524, 0.2858, 0.2460], device='cuda:0'), in_proj_covar=tensor([0.1461, 0.1063, 0.1298, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-09 23:59:57,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.611e+02 1.783e+03 2.318e+03 3.609e+03 1.231e+04, threshold=4.635e+03, percent-clipped=11.0
+2023-03-10 00:00:00,756 INFO [train.py:968] (0/2) Epoch 19, batch 26900, giga_loss[loss=0.3031, simple_loss=0.3657, pruned_loss=0.1202, over 28902.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3739, pruned_loss=0.1251, over 5657558.45 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3603, pruned_loss=0.1143, over 5705052.25 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3754, pruned_loss=0.1261, over 5656905.78 frames. ], batch size: 227, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:00:09,924 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:00:12,908 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:00:13,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3478, 1.4938, 4.0810, 3.2495], device='cuda:0'), covar=tensor([0.1676, 0.2550, 0.0463, 0.0908], device='cuda:0'), in_proj_covar=tensor([0.0750, 0.0642, 0.0949, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 00:00:41,928 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:00:48,509 INFO [train.py:968] (0/2) Epoch 19, batch 26950, giga_loss[loss=0.3088, simple_loss=0.3765, pruned_loss=0.1205, over 28838.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3747, pruned_loss=0.1262, over 5663231.20 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3603, pruned_loss=0.1142, over 5708846.03 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3762, pruned_loss=0.1274, over 5658102.74 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:01:28,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.435e+02 1.605e+03 1.952e+03 2.759e+03 7.106e+03, threshold=3.904e+03, percent-clipped=5.0
+2023-03-10 00:01:34,802 INFO [train.py:968] (0/2) Epoch 19, batch 27000, giga_loss[loss=0.2906, simple_loss=0.3579, pruned_loss=0.1117, over 28702.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3753, pruned_loss=0.1231, over 5677004.18 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.36, pruned_loss=0.1139, over 5712054.99 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.377, pruned_loss=0.1245, over 5669146.37 frames. ], batch size: 284, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:01:34,807 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 00:01:43,567 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2067, simple_loss=0.3133, pruned_loss=0.05006, over 944034.00 frames. 
+2023-03-10 00:01:43,568 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 00:01:58,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8675, 1.2152, 1.3753, 0.9726], device='cuda:0'), covar=tensor([0.1970, 0.1318, 0.2135, 0.1887], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0741, 0.0705, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 00:02:28,962 INFO [train.py:968] (0/2) Epoch 19, batch 27050, giga_loss[loss=0.3164, simple_loss=0.3902, pruned_loss=0.1213, over 28917.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3761, pruned_loss=0.1222, over 5664689.37 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3597, pruned_loss=0.1137, over 5706488.37 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3783, pruned_loss=0.1238, over 5661641.87 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:03:08,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.954e+02 1.498e+03 1.821e+03 2.423e+03 5.613e+03, threshold=3.641e+03, percent-clipped=4.0
+2023-03-10 00:03:11,997 INFO [train.py:968] (0/2) Epoch 19, batch 27100, giga_loss[loss=0.3402, simple_loss=0.3956, pruned_loss=0.1424, over 28471.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.378, pruned_loss=0.1226, over 5670766.91 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3597, pruned_loss=0.1138, over 5707252.03 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3801, pruned_loss=0.1239, over 5666688.02 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:03:35,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=849711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:03:49,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2019, 1.4093, 1.2797, 1.1046], device='cuda:0'), covar=tensor([0.2605, 0.2093, 0.1813, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.1924, 0.1847, 0.1793, 0.1929], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 00:04:02,789 INFO [train.py:968] (0/2) Epoch 19, batch 27150, giga_loss[loss=0.3561, simple_loss=0.4122, pruned_loss=0.15, over 28933.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.382, pruned_loss=0.1269, over 5670627.23 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3597, pruned_loss=0.1138, over 5710770.53 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3841, pruned_loss=0.1282, over 5663443.93 frames. ], batch size: 227, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:04:46,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5069, 1.6078, 1.6990, 1.2943], device='cuda:0'), covar=tensor([0.1765, 0.2555, 0.1443, 0.1694], device='cuda:0'), in_proj_covar=tensor([0.0883, 0.0700, 0.0927, 0.0826], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-10 00:04:50,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.629e+03 2.129e+03 2.750e+03 6.936e+03, threshold=4.259e+03, percent-clipped=9.0
+2023-03-10 00:04:58,056 INFO [train.py:968] (0/2) Epoch 19, batch 27200, libri_loss[loss=0.3004, simple_loss=0.3727, pruned_loss=0.114, over 29540.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3824, pruned_loss=0.128, over 5680869.26 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3596, pruned_loss=0.1137, over 5712780.67 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3844, pruned_loss=0.1292, over 5673079.35 frames. ], batch size: 80, lr: 1.68e-03, grad_scale: 8.0
+2023-03-10 00:05:50,024 INFO [train.py:968] (0/2) Epoch 19, batch 27250, giga_loss[loss=0.3052, simple_loss=0.3739, pruned_loss=0.1183, over 28630.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3826, pruned_loss=0.1292, over 5674968.34 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3599, pruned_loss=0.1139, over 5715805.61 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3843, pruned_loss=0.1302, over 5665616.17 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:06:05,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1351, 3.4327, 2.1689, 0.9171], device='cuda:0'), covar=tensor([0.7307, 0.2624, 0.3799, 0.7168], device='cuda:0'), in_proj_covar=tensor([0.1716, 0.1624, 0.1586, 0.1397], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 00:06:05,249 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:06:08,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:06:33,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.271e+03 1.989e+03 2.354e+03 3.551e+03 8.979e+03, threshold=4.708e+03, percent-clipped=15.0
+2023-03-10 00:06:34,676 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=849884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:06:36,389 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:06:37,966 INFO [train.py:968] (0/2) Epoch 19, batch 27300, giga_loss[loss=0.2669, simple_loss=0.3501, pruned_loss=0.0918, over 28971.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3802, pruned_loss=0.1264, over 5676838.66 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3602, pruned_loss=0.1141, over 5709266.67 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3815, pruned_loss=0.1273, over 5673970.12 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:07:22,564 INFO [train.py:968] (0/2) Epoch 19, batch 27350, giga_loss[loss=0.2936, simple_loss=0.3661, pruned_loss=0.1106, over 28438.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3793, pruned_loss=0.1241, over 5663959.15 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3605, pruned_loss=0.1144, over 5704128.99 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3805, pruned_loss=0.1248, over 5666160.39 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:07:40,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1018, 1.1635, 3.2616, 2.8552], device='cuda:0'), covar=tensor([0.1644, 0.2711, 0.0550, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.0750, 0.0641, 0.0949, 0.0885], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 00:08:05,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.541e+02 1.588e+03 1.976e+03 2.578e+03 7.800e+03, threshold=3.951e+03, percent-clipped=3.0
+2023-03-10 00:08:09,268 INFO [train.py:968] (0/2) Epoch 19, batch 27400, giga_loss[loss=0.2817, simple_loss=0.3589, pruned_loss=0.1023, over 28685.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3796, pruned_loss=0.1239, over 5663521.20 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3608, pruned_loss=0.1144, over 5708097.67 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3811, pruned_loss=0.1249, over 5659847.77 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:08:21,210 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-850000.pt
+2023-03-10 00:09:01,481 INFO [train.py:968] (0/2) Epoch 19, batch 27450, giga_loss[loss=0.2884, simple_loss=0.3607, pruned_loss=0.1081, over 28252.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3823, pruned_loss=0.1263, over 5659118.93 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3609, pruned_loss=0.1145, over 5711178.48 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3839, pruned_loss=0.1273, over 5652652.87 frames. ], batch size: 65, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:09:28,051 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 00:09:44,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.139e+03 1.695e+03 2.176e+03 2.986e+03 9.277e+03, threshold=4.352e+03, percent-clipped=13.0
+2023-03-10 00:09:48,744 INFO [train.py:968] (0/2) Epoch 19, batch 27500, giga_loss[loss=0.3469, simple_loss=0.4056, pruned_loss=0.1441, over 28585.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3809, pruned_loss=0.1259, over 5663784.81 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3607, pruned_loss=0.1144, over 5713621.42 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3826, pruned_loss=0.1269, over 5656004.02 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:10:23,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3442, 1.4710, 1.3409, 1.4664], device='cuda:0'), covar=tensor([0.0760, 0.0347, 0.0321, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0104], device='cuda:0')
+2023-03-10 00:10:40,188 INFO [train.py:968] (0/2) Epoch 19, batch 27550, giga_loss[loss=0.2499, simple_loss=0.3302, pruned_loss=0.0848, over 28502.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.379, pruned_loss=0.1258, over 5663077.97 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3609, pruned_loss=0.1146, over 5712980.01 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3805, pruned_loss=0.1266, over 5656595.36 frames. ], batch size: 60, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:10:48,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:11:24,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.889e+03 2.451e+03 3.725e+03 9.043e+03, threshold=4.901e+03, percent-clipped=18.0
+2023-03-10 00:11:27,866 INFO [train.py:968] (0/2) Epoch 19, batch 27600, libri_loss[loss=0.2559, simple_loss=0.3267, pruned_loss=0.09255, over 29561.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3781, pruned_loss=0.1259, over 5666211.53 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3615, pruned_loss=0.1152, over 5707600.69 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3794, pruned_loss=0.1264, over 5663548.66 frames. ], batch size: 77, lr: 1.68e-03, grad_scale: 8.0
+2023-03-10 00:11:39,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4403, 1.7642, 1.5533, 1.5795], device='cuda:0'), covar=tensor([0.0632, 0.0279, 0.0274, 0.0655], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0104], device='cuda:0')
+2023-03-10 00:12:19,659 INFO [train.py:968] (0/2) Epoch 19, batch 27650, giga_loss[loss=0.3181, simple_loss=0.3737, pruned_loss=0.1312, over 27516.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3743, pruned_loss=0.1235, over 5656378.97 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3614, pruned_loss=0.1151, over 5700935.73 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3759, pruned_loss=0.1242, over 5658890.98 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 8.0
+2023-03-10 00:12:39,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=850259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:13:03,621 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.574e+03 2.042e+03 2.797e+03 8.076e+03, threshold=4.085e+03, percent-clipped=6.0
+2023-03-10 00:13:03,863 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7753, 1.3865, 4.9901, 3.6811], device='cuda:0'), covar=tensor([0.1576, 0.2745, 0.0403, 0.0777], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0639, 0.0947, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 00:13:07,186 INFO [train.py:968] (0/2) Epoch 19, batch 27700, giga_loss[loss=0.3045, simple_loss=0.3747, pruned_loss=0.1172, over 28853.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3733, pruned_loss=0.1234, over 5659839.72 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3619, pruned_loss=0.1154, over 5705425.69 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3743, pruned_loss=0.1238, over 5656828.57 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:13:53,578 INFO [train.py:968] (0/2) Epoch 19, batch 27750, giga_loss[loss=0.298, simple_loss=0.3593, pruned_loss=0.1183, over 28592.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3723, pruned_loss=0.123, over 5657909.40 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3618, pruned_loss=0.1153, over 5704724.77 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3733, pruned_loss=0.1236, over 5655395.50 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:14:23,168 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3963, 1.9992, 1.6891, 1.3829], device='cuda:0'), covar=tensor([0.0830, 0.0283, 0.0282, 0.1098], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0104], device='cuda:0')
+2023-03-10 00:14:36,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.894e+02 1.558e+03 2.030e+03 2.672e+03 5.094e+03, threshold=4.059e+03, percent-clipped=5.0
+2023-03-10 00:14:39,905 INFO [train.py:968] (0/2) Epoch 19, batch 27800, giga_loss[loss=0.2723, simple_loss=0.3613, pruned_loss=0.09162, over 29047.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.37, pruned_loss=0.1198, over 5665491.50 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3618, pruned_loss=0.1152, over 5706829.22 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.371, pruned_loss=0.1204, over 5660930.25 frames. ], batch size: 155, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:14:50,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=850402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:14:53,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=850405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:15:03,264 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850415.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 00:15:11,977 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 00:15:19,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=850434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:15:22,624 INFO [train.py:968] (0/2) Epoch 19, batch 27850, libri_loss[loss=0.3536, simple_loss=0.4069, pruned_loss=0.1502, over 26367.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3668, pruned_loss=0.117, over 5657497.31 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3612, pruned_loss=0.1148, over 5702534.26 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3685, pruned_loss=0.1181, over 5656518.83 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:15:42,338 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-10 00:16:06,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:16:11,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.016e+02 1.420e+03 1.826e+03 2.581e+03 1.472e+04, threshold=3.651e+03, percent-clipped=10.0
+2023-03-10 00:16:13,155 INFO [train.py:968] (0/2) Epoch 19, batch 27900, giga_loss[loss=0.2908, simple_loss=0.3638, pruned_loss=0.1089, over 29052.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3647, pruned_loss=0.1154, over 5660661.01 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3614, pruned_loss=0.1151, over 5707638.79 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3659, pruned_loss=0.116, over 5654191.09 frames. ], batch size: 128, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 00:16:46,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=850520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:17:05,922 INFO [train.py:968] (0/2) Epoch 19, batch 27950, giga_loss[loss=0.3832, simple_loss=0.4073, pruned_loss=0.1796, over 26664.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3642, pruned_loss=0.1164, over 5656358.71 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3616, pruned_loss=0.1152, over 5711713.32 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3651, pruned_loss=0.1168, over 5646988.11 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 00:17:18,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9312, 1.9544, 1.8655, 1.6946], device='cuda:0'), covar=tensor([0.1872, 0.2661, 0.2243, 0.2333], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0746, 0.0706, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 00:17:35,297 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:17:58,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.491e+02 2.066e+03 2.702e+03 4.348e+03 1.217e+04, threshold=5.404e+03, percent-clipped=32.0
+2023-03-10 00:18:00,505 INFO [train.py:968] (0/2) Epoch 19, batch 28000, giga_loss[loss=0.2509, simple_loss=0.3296, pruned_loss=0.08607, over 29030.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3615, pruned_loss=0.1153, over 5660793.93 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3608, pruned_loss=0.1148, over 5714632.47 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3628, pruned_loss=0.1159, over 5650080.11 frames. ], batch size: 155, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:18:46,512 INFO [train.py:968] (0/2) Epoch 19, batch 28050, giga_loss[loss=0.2839, simple_loss=0.3682, pruned_loss=0.09979, over 29001.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3656, pruned_loss=0.1176, over 5660628.15 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.362, pruned_loss=0.1156, over 5708500.12 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3657, pruned_loss=0.1174, over 5655877.14 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:19:09,559 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-10 00:19:12,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=850663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:19:15,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=850666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:19:32,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.268e+02 1.686e+03 2.194e+03 2.782e+03 7.455e+03, threshold=4.388e+03, percent-clipped=1.0
+2023-03-10 00:19:34,526 INFO [train.py:968] (0/2) Epoch 19, batch 28100, giga_loss[loss=0.3643, simple_loss=0.4103, pruned_loss=0.1592, over 27525.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3666, pruned_loss=0.1178, over 5647710.43 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3621, pruned_loss=0.1158, over 5707651.08 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3667, pruned_loss=0.1175, over 5643648.54 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:19:39,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6751, 1.7854, 1.4350, 2.0047], device='cuda:0'), covar=tensor([0.2589, 0.2777, 0.3009, 0.2392], device='cuda:0'), in_proj_covar=tensor([0.1466, 0.1070, 0.1303, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 00:19:41,344 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=850695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:19:55,882 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-10 00:20:04,249 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:20:20,892 INFO [train.py:968] (0/2) Epoch 19, batch 28150, giga_loss[loss=0.238, simple_loss=0.3225, pruned_loss=0.07679, over 28539.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.366, pruned_loss=0.1173, over 5660096.33 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3617, pruned_loss=0.1155, over 5712551.66 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3665, pruned_loss=0.1174, over 5651233.47 frames. ], batch size: 60, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:20:32,765 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:21:00,213 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7597, 1.7739, 1.3280, 1.4537], device='cuda:0'), covar=tensor([0.0715, 0.0445, 0.0911, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0441, 0.0509, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 00:21:04,146 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:21:07,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.658e+02 1.486e+03 1.873e+03 2.425e+03 5.105e+03, threshold=3.747e+03, percent-clipped=4.0
+2023-03-10 00:21:08,285 INFO [train.py:968] (0/2) Epoch 19, batch 28200, giga_loss[loss=0.2974, simple_loss=0.3617, pruned_loss=0.1165, over 28791.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3669, pruned_loss=0.1186, over 5649008.51 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3624, pruned_loss=0.1157, over 5713401.33 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3668, pruned_loss=0.1185, over 5639531.55 frames. ], batch size: 119, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:21:10,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=850790.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 00:21:15,821 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-10 00:21:53,569 INFO [train.py:968] (0/2) Epoch 19, batch 28250, giga_loss[loss=0.3659, simple_loss=0.4051, pruned_loss=0.1633, over 26643.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3666, pruned_loss=0.1188, over 5646947.79 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3623, pruned_loss=0.1157, over 5717372.07 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3667, pruned_loss=0.1188, over 5634732.33 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:22:07,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=850855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:22:37,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 1.808e+03 2.428e+03 3.315e+03 9.820e+03, threshold=4.857e+03, percent-clipped=19.0
+2023-03-10 00:22:38,519 INFO [train.py:968] (0/2) Epoch 19, batch 28300, giga_loss[loss=0.3029, simple_loss=0.3708, pruned_loss=0.1176, over 28889.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3679, pruned_loss=0.1191, over 5654199.98 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.362, pruned_loss=0.1156, over 5717760.23 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3683, pruned_loss=0.1193, over 5642588.87 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:22:43,702 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6710, 1.7868, 1.7169, 1.5915], device='cuda:0'), covar=tensor([0.1659, 0.2384, 0.2062, 0.2140], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0750, 0.0709, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 00:23:22,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=850933.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 00:23:25,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=850936.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 00:23:28,785 INFO [train.py:968] (0/2) Epoch 19, batch 28350, giga_loss[loss=0.435, simple_loss=0.4491, pruned_loss=0.2104, over 26568.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3706, pruned_loss=0.1215, over 5645234.95 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.363, pruned_loss=0.1166, over 5712785.86 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3701, pruned_loss=0.1209, over 5639519.38 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 00:23:30,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=850939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:23:35,754 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-10 00:23:58,848 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=850965.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 00:23:59,392 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:24:18,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.817e+03 2.292e+03 3.424e+03 7.999e+03, threshold=4.584e+03, percent-clipped=9.0
+2023-03-10 00:24:18,892 INFO [train.py:968] (0/2) Epoch 19, batch 28400, giga_loss[loss=0.284, simple_loss=0.3561, pruned_loss=0.106, over 28861.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5634059.83 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3633, pruned_loss=0.1167, over 5693364.08 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1223, over 5646947.09 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:24:29,639 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=850998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:24:34,324 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:25:04,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:25:14,209 INFO [train.py:968] (0/2) Epoch 19, batch 28450, giga_loss[loss=0.2754, simple_loss=0.3617, pruned_loss=0.09458, over 27946.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3723, pruned_loss=0.1227, over 5637759.90 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3628, pruned_loss=0.1164, over 5696698.10 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3725, pruned_loss=0.1226, over 5643978.12 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:26:00,177 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:26:03,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:26:04,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.656e+03 2.010e+03 2.799e+03 7.287e+03, threshold=4.020e+03, percent-clipped=7.0
+2023-03-10 00:26:05,030 INFO [train.py:968] (0/2) Epoch 19, batch 28500, giga_loss[loss=0.2695, simple_loss=0.3416, pruned_loss=0.09872, over 28712.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3711, pruned_loss=0.1209, over 5644907.90 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3625, pruned_loss=0.1164, over 5701393.54 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3719, pruned_loss=0.121, over 5643497.76 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:26:10,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:26:33,120 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:26:43,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:26:45,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6328, 2.4807, 1.5066, 0.8677], device='cuda:0'), covar=tensor([0.7028, 0.3081, 0.3411, 0.5311], device='cuda:0'), in_proj_covar=tensor([0.1715, 0.1618, 0.1580, 0.1396], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 00:26:54,690 INFO [train.py:968] (0/2) Epoch 19, batch 28550, giga_loss[loss=0.2854, simple_loss=0.3593, pruned_loss=0.1057, over 28580.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3722, pruned_loss=0.1222, over 5643185.98 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3624, pruned_loss=0.1163, over 5704110.72 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.373, pruned_loss=0.1225, over 5638792.57 frames. ], batch size: 242, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:27:11,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7788, 2.1746, 2.0143, 1.5879], device='cuda:0'), covar=tensor([0.3025, 0.2118, 0.2284, 0.2454], device='cuda:0'), in_proj_covar=tensor([0.1935, 0.1864, 0.1798, 0.1934], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 00:27:16,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:27:20,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5890, 1.6036, 1.2678, 1.2258], device='cuda:0'), covar=tensor([0.0773, 0.0518, 0.0876, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0445, 0.0511, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 00:27:48,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.144e+03 1.908e+03 2.659e+03 3.752e+03 1.293e+04, threshold=5.318e+03, percent-clipped=22.0
+2023-03-10 00:27:48,931 INFO [train.py:968] (0/2) Epoch 19, batch 28600, giga_loss[loss=0.2919, simple_loss=0.3587, pruned_loss=0.1125, over 28323.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3722, pruned_loss=0.123, over 5631252.32 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3626, pruned_loss=0.1164, over 5707674.23 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3729, pruned_loss=0.1232, over 5623322.81 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:28:10,896 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3548, 1.4925, 1.3929, 1.2748], device='cuda:0'), covar=tensor([0.1976, 0.1936, 0.1529, 0.1847], device='cuda:0'), in_proj_covar=tensor([0.1931, 0.1862, 0.1796, 0.1933], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 00:28:28,053 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.36 vs. limit=5.0
+2023-03-10 00:28:41,146 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=851228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:28:55,319 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:28:55,628 INFO [train.py:968] (0/2) Epoch 19, batch 28650, giga_loss[loss=0.3405, simple_loss=0.3756, pruned_loss=0.1527, over 23642.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3706, pruned_loss=0.1228, over 5617295.68 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3626, pruned_loss=0.1165, over 5699157.43 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.123, over 5617365.05 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:28:58,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:22,900 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:24,186 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:25,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:40,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.570e+03 1.965e+03 2.571e+03 6.176e+03, threshold=3.929e+03, percent-clipped=1.0
+2023-03-10 00:29:41,436 INFO [train.py:968] (0/2) Epoch 19, batch 28700, giga_loss[loss=0.281, simple_loss=0.3442, pruned_loss=0.1089, over 28746.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3693, pruned_loss=0.1221, over 5628488.42 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3628, pruned_loss=0.1165, over 5694256.71 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3698, pruned_loss=0.1224, over 5631405.38 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:29:49,891 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:50,700 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:52,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:30:22,127 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:30:29,504 INFO [train.py:968] (0/2) Epoch 19, batch 28750, giga_loss[loss=0.3096, simple_loss=0.3667, pruned_loss=0.1262, over 27888.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3679, pruned_loss=0.1211, over 5644283.63 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3622, pruned_loss=0.1161, over 5698561.63 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3689, pruned_loss=0.1218, over 5641535.12 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:30:33,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:31:04,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5606, 1.8161, 1.4615, 1.7572], device='cuda:0'), covar=tensor([0.2558, 0.2717, 0.2998, 0.2339], device='cuda:0'), in_proj_covar=tensor([0.1466, 0.1067, 0.1302, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 00:31:15,742 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.697e+03 2.119e+03 3.132e+03 6.145e+03, threshold=4.238e+03, percent-clipped=16.0
+2023-03-10 00:31:16,357 INFO [train.py:968] (0/2) Epoch 19, batch 28800, giga_loss[loss=0.3194, simple_loss=0.38, pruned_loss=0.1294, over 28561.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.368, pruned_loss=0.1213, over 5648799.20 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3623, pruned_loss=0.116, over 5701964.59 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3689, pruned_loss=0.1221, over 5642710.05 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:31:38,255 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=851408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:32:09,451 INFO [train.py:968] (0/2) Epoch 19, batch 28850, giga_loss[loss=0.276, simple_loss=0.3417, pruned_loss=0.1051, over 28543.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3691, pruned_loss=0.122, over 5651848.68 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3624, pruned_loss=0.116, over 5704233.86 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3698, pruned_loss=0.1227, over 5644298.09 frames. ], batch size: 60, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:32:52,065 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:32:55,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:32:55,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.274e+02 1.690e+03 2.216e+03 3.198e+03 6.420e+03, threshold=4.433e+03, percent-clipped=8.0
+2023-03-10 00:32:55,807 INFO [train.py:968] (0/2) Epoch 19, batch 28900, giga_loss[loss=0.308, simple_loss=0.3736, pruned_loss=0.1212, over 28497.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3714, pruned_loss=0.1237, over 5659917.61 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3624, pruned_loss=0.116, over 5698809.04 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.372, pruned_loss=0.1243, over 5657480.37 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:32:59,442 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-10 00:33:17,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1578, 2.5919, 1.2487, 1.3582], device='cuda:0'), covar=tensor([0.1043, 0.0400, 0.0886, 0.1412], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0552, 0.0377, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 00:33:26,923 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:33:49,225 INFO [train.py:968] (0/2) Epoch 19, batch 28950, giga_loss[loss=0.3344, simple_loss=0.3835, pruned_loss=0.1426, over 28964.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3721, pruned_loss=0.1244, over 5661119.95 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3626, pruned_loss=0.1161, over 5699075.91 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3726, pruned_loss=0.1249, over 5658551.28 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:34:22,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6075, 2.3459, 2.4610, 2.0438], device='cuda:0'), covar=tensor([0.1657, 0.2607, 0.2036, 0.2316], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0752, 0.0712, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 00:34:35,233 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.145e+03 1.685e+03 2.200e+03 3.285e+03 6.549e+03, threshold=4.401e+03, percent-clipped=11.0
+2023-03-10 00:34:35,245 INFO [train.py:968] (0/2) Epoch 19, batch 29000, giga_loss[loss=0.2545, simple_loss=0.3273, pruned_loss=0.09083, over 28853.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3732, pruned_loss=0.1256, over 5644948.76 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3631, pruned_loss=0.1166, over 5677108.60 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3733, pruned_loss=0.1257, over 5662220.19 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:34:48,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:34:49,317 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-10 00:35:18,329 INFO [train.py:968] (0/2) Epoch 19, batch 29050, libri_loss[loss=0.3075, simple_loss=0.3832, pruned_loss=0.1159, over 29265.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3738, pruned_loss=0.1255, over 5662232.62 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3632, pruned_loss=0.1165, over 5686210.75 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3743, pruned_loss=0.1262, over 5666863.96 frames. ], batch size: 94, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:36:01,995 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9317, 2.1030, 2.1589, 1.7243], device='cuda:0'), covar=tensor([0.1896, 0.2414, 0.1472, 0.1736], device='cuda:0'), in_proj_covar=tensor([0.0885, 0.0705, 0.0931, 0.0829], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-10 00:36:07,285 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.719e+02 1.726e+03 2.266e+03 2.998e+03 6.920e+03, threshold=4.531e+03, percent-clipped=8.0
+2023-03-10 00:36:07,297 INFO [train.py:968] (0/2) Epoch 19, batch 29100, giga_loss[loss=0.3179, simple_loss=0.3804, pruned_loss=0.1277, over 28668.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3741, pruned_loss=0.1256, over 5661365.40 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3629, pruned_loss=0.1164, over 5692270.59 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3751, pruned_loss=0.1265, over 5658789.43 frames. ], batch size: 242, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:36:21,580 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3904, 1.7035, 1.3397, 1.4884], device='cuda:0'), covar=tensor([0.2782, 0.2772, 0.3102, 0.2392], device='cuda:0'), in_proj_covar=tensor([0.1468, 0.1069, 0.1301, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 00:36:38,094 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 00:36:54,301 INFO [train.py:968] (0/2) Epoch 19, batch 29150, giga_loss[loss=0.3241, simple_loss=0.3853, pruned_loss=0.1315, over 28577.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3739, pruned_loss=0.1249, over 5663866.32 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3626, pruned_loss=0.1161, over 5685357.57 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3751, pruned_loss=0.126, over 5667507.90 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:36:54,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9850, 2.3164, 2.2402, 2.1896], device='cuda:0'), covar=tensor([0.0632, 0.0243, 0.0261, 0.0737], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0104], device='cuda:0')
+2023-03-10 00:37:01,162 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:37:01,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5963, 1.9082, 1.4840, 1.9832], device='cuda:0'), covar=tensor([0.2561, 0.2713, 0.2979, 0.2458], device='cuda:0'), in_proj_covar=tensor([0.1467, 0.1068, 0.1301, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 00:37:03,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:37:31,028 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:37:36,764 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:37:41,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.046e+02 1.756e+03 2.297e+03 3.151e+03 5.672e+03, threshold=4.593e+03, percent-clipped=7.0
+2023-03-10 00:37:41,713 INFO [train.py:968] (0/2) Epoch 19, batch 29200, giga_loss[loss=0.2787, simple_loss=0.3463, pruned_loss=0.1056, over 28813.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.375, pruned_loss=0.1263, over 5658917.52 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3627, pruned_loss=0.1162, over 5687619.79 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.376, pruned_loss=0.1272, over 5659553.59 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:37:46,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2607, 3.0691, 1.3578, 1.4710], device='cuda:0'), covar=tensor([0.1051, 0.0389, 0.0944, 0.1384], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0552, 0.0377, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 00:37:59,427 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 00:38:03,985 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.33 vs. limit=5.0
+2023-03-10 00:38:25,761 INFO [train.py:968] (0/2) Epoch 19, batch 29250, giga_loss[loss=0.3152, simple_loss=0.3804, pruned_loss=0.125, over 28669.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3762, pruned_loss=0.127, over 5662134.09 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3631, pruned_loss=0.1165, over 5686346.46 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3769, pruned_loss=0.1277, over 5663659.11 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:39:12,213 INFO [train.py:968] (0/2) Epoch 19, batch 29300, giga_loss[loss=0.2884, simple_loss=0.3639, pruned_loss=0.1065, over 28727.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3759, pruned_loss=0.1264, over 5663237.11 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.363, pruned_loss=0.1164, over 5689699.98 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3767, pruned_loss=0.1272, over 5660980.50 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:39:13,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.700e+03 2.292e+03 3.262e+03 1.011e+04, threshold=4.585e+03, percent-clipped=10.0
+2023-03-10 00:39:41,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1695, 1.7619, 1.3509, 0.3882], device='cuda:0'), covar=tensor([0.4121, 0.2821, 0.3662, 0.5540], device='cuda:0'), in_proj_covar=tensor([0.1727, 0.1633, 0.1596, 0.1404], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 00:39:53,461 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:39:56,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:40:05,810 INFO [train.py:968] (0/2) Epoch 19, batch 29350, giga_loss[loss=0.3497, simple_loss=0.4041, pruned_loss=0.1477, over 28286.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3759, pruned_loss=0.1258, over 5640608.57 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3632, pruned_loss=0.1164, over 5684247.11 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3767, pruned_loss=0.1266, over 5642617.93 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:40:11,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2426, 1.4316, 1.4897, 1.2889], device='cuda:0'), covar=tensor([0.1574, 0.1462, 0.1955, 0.1568], device='cuda:0'), in_proj_covar=tensor([0.0473, 0.0755, 0.0715, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 00:40:14,350 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-10 00:40:25,231 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:40:51,870 INFO [train.py:968] (0/2) Epoch 19, batch 29400, giga_loss[loss=0.302, simple_loss=0.3672, pruned_loss=0.1184, over 28555.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3749, pruned_loss=0.1247, over 5644708.13 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3632, pruned_loss=0.1164, over 5685771.02 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3758, pruned_loss=0.1254, over 5644000.64 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:40:53,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.795e+02 1.619e+03 2.060e+03 2.904e+03 9.813e+03, threshold=4.119e+03, percent-clipped=9.0
+2023-03-10 00:40:57,937 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-10 00:41:01,764 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-852000.pt
+2023-03-10 00:41:36,095 INFO [train.py:968] (0/2) Epoch 19, batch 29450, libri_loss[loss=0.2599, simple_loss=0.3287, pruned_loss=0.09561, over 29549.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3728, pruned_loss=0.1228, over 5656947.86 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3629, pruned_loss=0.1163, over 5689361.40 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3743, pruned_loss=0.1239, over 5651576.82 frames. ], batch size: 78, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:41:44,382 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-10 00:42:17,435 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6592, 1.9777, 1.5231, 2.0344], device='cuda:0'), covar=tensor([0.2550, 0.2626, 0.3015, 0.2454], device='cuda:0'), in_proj_covar=tensor([0.1461, 0.1064, 0.1297, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 00:42:21,663 INFO [train.py:968] (0/2) Epoch 19, batch 29500, giga_loss[loss=0.3417, simple_loss=0.3973, pruned_loss=0.143, over 28581.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3723, pruned_loss=0.1226, over 5660446.95 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.363, pruned_loss=0.1162, over 5692453.89 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3735, pruned_loss=0.1236, over 5653249.90 frames. ], batch size: 242, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:42:22,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.808e+02 1.595e+03 2.034e+03 2.550e+03 5.359e+03, threshold=4.067e+03, percent-clipped=1.0
+2023-03-10 00:42:33,452 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-10 00:43:10,336 INFO [train.py:968] (0/2) Epoch 19, batch 29550, giga_loss[loss=0.2817, simple_loss=0.361, pruned_loss=0.1012, over 28843.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3728, pruned_loss=0.1229, over 5658955.31 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.363, pruned_loss=0.1162, over 5696513.42 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3739, pruned_loss=0.1238, over 5648885.79 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:43:25,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3891, 3.6224, 1.5376, 1.5524], device='cuda:0'), covar=tensor([0.1022, 0.0407, 0.0926, 0.1400], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0553, 0.0377, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 00:44:01,635 INFO [train.py:968] (0/2) Epoch 19, batch 29600, giga_loss[loss=0.3922, simple_loss=0.4281, pruned_loss=0.1781, over 27679.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3742, pruned_loss=0.1244, over 5651714.81 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3632, pruned_loss=0.1164, over 5688680.75 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3751, pruned_loss=0.1251, over 5650727.77 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:44:02,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.067e+02 1.560e+03 1.855e+03 2.646e+03 7.388e+03, threshold=3.710e+03, percent-clipped=7.0
+2023-03-10 00:44:07,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7650, 3.5924, 3.4155, 1.7119], device='cuda:0'), covar=tensor([0.0741, 0.0841, 0.0805, 0.2276], device='cuda:0'), in_proj_covar=tensor([0.1219, 0.1131, 0.0963, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 00:44:48,940 INFO [train.py:968] (0/2) Epoch 19, batch 29650, giga_loss[loss=0.3063, simple_loss=0.3671, pruned_loss=0.1227, over 28835.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3726, pruned_loss=0.124, over 5656986.98 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3632, pruned_loss=0.1163, over 5692366.24 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3736, pruned_loss=0.1249, over 5652090.47 frames. ], batch size: 243, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:45:38,610 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:45:38,980 INFO [train.py:968] (0/2) Epoch 19, batch 29700, giga_loss[loss=0.3185, simple_loss=0.3809, pruned_loss=0.1281, over 28269.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3731, pruned_loss=0.1242, over 5666791.57 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.363, pruned_loss=0.1161, over 5694319.08 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3741, pruned_loss=0.1251, over 5661020.91 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:45:39,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2567, 4.0713, 3.8656, 1.7558], device='cuda:0'), covar=tensor([0.0570, 0.0737, 0.0727, 0.2217], device='cuda:0'), in_proj_covar=tensor([0.1215, 0.1129, 0.0963, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 00:45:40,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.660e+02 1.694e+03 2.362e+03 3.678e+03 9.321e+03, threshold=4.725e+03, percent-clipped=23.0
+2023-03-10 00:46:22,624 INFO [train.py:968] (0/2) Epoch 19, batch 29750, giga_loss[loss=0.2822, simple_loss=0.3559, pruned_loss=0.1042, over 28855.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3763, pruned_loss=0.1268, over 5674141.69 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3634, pruned_loss=0.1164, over 5701853.23 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3772, pruned_loss=0.1276, over 5661645.64 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:46:44,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5471, 1.6482, 1.2214, 1.2457], device='cuda:0'), covar=tensor([0.0960, 0.0620, 0.1060, 0.1157], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0450, 0.0515, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 00:47:09,756 INFO [train.py:968] (0/2) Epoch 19, batch 29800, giga_loss[loss=0.3126, simple_loss=0.3762, pruned_loss=0.1245, over 29161.00 frames. ], tot_loss[loss=0.314, simple_loss=0.375, pruned_loss=0.1265, over 5650192.05 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3626, pruned_loss=0.116, over 5695863.67 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.377, pruned_loss=0.1279, over 5643594.22 frames. ], batch size: 128, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:47:11,208 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.599e+03 1.902e+03 2.707e+03 7.073e+03, threshold=3.803e+03, percent-clipped=9.0
+2023-03-10 00:47:57,286 INFO [train.py:968] (0/2) Epoch 19, batch 29850, giga_loss[loss=0.3286, simple_loss=0.3808, pruned_loss=0.1382, over 27617.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3751, pruned_loss=0.1265, over 5642900.10 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3628, pruned_loss=0.1163, over 5689912.97 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3767, pruned_loss=0.1276, over 5642840.34 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:47:59,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.47 vs. limit=5.0
+2023-03-10 00:48:50,115 INFO [train.py:968] (0/2) Epoch 19, batch 29900, giga_loss[loss=0.3623, simple_loss=0.3965, pruned_loss=0.164, over 23550.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3756, pruned_loss=0.126, over 5650570.37 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3625, pruned_loss=0.1161, over 5692808.92 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3773, pruned_loss=0.1273, over 5646942.76 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:48:51,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.479e+02 1.564e+03 2.088e+03 2.872e+03 6.972e+03, threshold=4.175e+03, percent-clipped=8.0
+2023-03-10 00:49:04,268 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 00:49:35,368 INFO [train.py:968] (0/2) Epoch 19, batch 29950, libri_loss[loss=0.2681, simple_loss=0.328, pruned_loss=0.1041, over 29346.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3751, pruned_loss=0.1252, over 5653597.32 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3625, pruned_loss=0.1161, over 5691863.33 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.377, pruned_loss=0.1266, over 5650265.04 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:50:17,333 INFO [train.py:968] (0/2) Epoch 19, batch 30000, giga_loss[loss=0.3624, simple_loss=0.4124, pruned_loss=0.1562, over 28007.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3746, pruned_loss=0.1252, over 5651746.92 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3627, pruned_loss=0.1162, over 5688088.37 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3767, pruned_loss=0.1267, over 5651362.74 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:50:17,338 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 00:50:26,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4586, 1.6546, 1.2679, 1.3065], device='cuda:0'), covar=tensor([0.0882, 0.0494, 0.1005, 0.0964], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0450, 0.0515, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 00:50:28,074 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2051, simple_loss=0.3127, pruned_loss=0.04877, over 944034.00 frames. 
+2023-03-10 00:50:28,075 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 00:50:29,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.751e+03 2.286e+03 3.307e+03 1.084e+04, threshold=4.572e+03, percent-clipped=13.0
+2023-03-10 00:50:47,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3720, 1.5956, 1.0447, 1.1694], device='cuda:0'), covar=tensor([0.1143, 0.0750, 0.1512, 0.1441], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0450, 0.0515, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 00:51:10,862 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-10 00:51:14,149 INFO [train.py:968] (0/2) Epoch 19, batch 30050, giga_loss[loss=0.3386, simple_loss=0.385, pruned_loss=0.1462, over 27513.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3736, pruned_loss=0.1244, over 5664889.86 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3628, pruned_loss=0.1162, over 5694118.12 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3754, pruned_loss=0.1258, over 5658591.31 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:51:40,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=852662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:51:46,474 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:51:49,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.12 vs. limit=5.0
+2023-03-10 00:52:01,348 INFO [train.py:968] (0/2) Epoch 19, batch 30100, giga_loss[loss=0.2698, simple_loss=0.341, pruned_loss=0.09928, over 28926.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3691, pruned_loss=0.1216, over 5666921.25 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3622, pruned_loss=0.1158, over 5693051.44 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3713, pruned_loss=0.1232, over 5662256.22 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:52:04,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.236e+03 1.583e+03 2.008e+03 2.795e+03 5.234e+03, threshold=4.017e+03, percent-clipped=2.0
+2023-03-10 00:52:34,849 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-10 00:52:50,600 INFO [train.py:968] (0/2) Epoch 19, batch 30150, giga_loss[loss=0.2857, simple_loss=0.3503, pruned_loss=0.1106, over 28507.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3659, pruned_loss=0.1206, over 5657789.16 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3622, pruned_loss=0.1157, over 5695274.46 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3678, pruned_loss=0.1221, over 5651403.25 frames. ], batch size: 60, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:53:26,214 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:53:38,062 INFO [train.py:968] (0/2) Epoch 19, batch 30200, giga_loss[loss=0.2832, simple_loss=0.3426, pruned_loss=0.1119, over 28565.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3644, pruned_loss=0.1203, over 5664016.86 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3624, pruned_loss=0.1158, over 5697581.58 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3657, pruned_loss=0.1214, over 5656306.68 frames. ], batch size: 85, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:53:40,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.329e+02 1.777e+03 2.282e+03 3.289e+03 6.341e+03, threshold=4.564e+03, percent-clipped=10.0
+2023-03-10 00:53:56,173 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=852805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:53:58,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=852808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:54:01,431 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-10 00:54:12,199 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:54:29,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=852837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:54:30,401 INFO [train.py:968] (0/2) Epoch 19, batch 30250, giga_loss[loss=0.2792, simple_loss=0.3525, pruned_loss=0.103, over 28898.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3638, pruned_loss=0.1202, over 5648103.59 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3622, pruned_loss=0.1157, over 5699771.97 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3649, pruned_loss=0.1211, over 5640013.44 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:54:35,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4336, 1.8166, 1.4038, 1.5423], device='cuda:0'), covar=tensor([0.2651, 0.2706, 0.3074, 0.2385], device='cuda:0'), in_proj_covar=tensor([0.1464, 0.1064, 0.1301, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 00:54:55,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:55:18,434 INFO [train.py:968] (0/2) Epoch 19, batch 30300, giga_loss[loss=0.2457, simple_loss=0.3287, pruned_loss=0.08137, over 28935.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.364, pruned_loss=0.1182, over 5648395.21 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3627, pruned_loss=0.1161, over 5695226.93 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3645, pruned_loss=0.1187, over 5645262.98 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:55:23,927 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.252e+03 1.650e+03 2.198e+03 3.708e+03 9.726e+03, threshold=4.395e+03, percent-clipped=15.0
+2023-03-10 00:55:38,722 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:56:11,353 INFO [train.py:968] (0/2) Epoch 19, batch 30350, giga_loss[loss=0.278, simple_loss=0.3556, pruned_loss=0.1002, over 28884.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3616, pruned_loss=0.1151, over 5630398.73 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3624, pruned_loss=0.1161, over 5686648.73 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3624, pruned_loss=0.1156, over 5633679.15 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:56:31,746 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-10 00:56:44,515 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 00:56:58,932 INFO [train.py:968] (0/2) Epoch 19, batch 30400, giga_loss[loss=0.2703, simple_loss=0.3433, pruned_loss=0.09863, over 28014.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3589, pruned_loss=0.1117, over 5646637.72 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3616, pruned_loss=0.1157, over 5690930.78 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3601, pruned_loss=0.1123, over 5643945.39 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:57:01,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.771e+02 1.453e+03 1.948e+03 2.564e+03 5.295e+03, threshold=3.897e+03, percent-clipped=2.0
+2023-03-10 00:57:12,216 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:57:37,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2452, 2.7348, 1.3956, 1.4015], device='cuda:0'), covar=tensor([0.0953, 0.0362, 0.0975, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0551, 0.0376, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 00:57:42,514 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:57:48,052 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:57:52,455 INFO [train.py:968] (0/2) Epoch 19, batch 30450, giga_loss[loss=0.27, simple_loss=0.3461, pruned_loss=0.0969, over 28475.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3567, pruned_loss=0.109, over 5649150.09 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.362, pruned_loss=0.1161, over 5692671.82 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3574, pruned_loss=0.1091, over 5644985.88 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:58:00,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:58:41,712 INFO [train.py:968] (0/2) Epoch 19, batch 30500, giga_loss[loss=0.3112, simple_loss=0.3874, pruned_loss=0.1175, over 28568.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3531, pruned_loss=0.1057, over 5655280.19 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3615, pruned_loss=0.116, over 5697461.56 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3539, pruned_loss=0.1056, over 5646536.82 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:58:44,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.122e+02 1.285e+03 1.788e+03 2.352e+03 6.532e+03, threshold=3.576e+03, percent-clipped=5.0
+2023-03-10 00:58:54,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:59:31,119 INFO [train.py:968] (0/2) Epoch 19, batch 30550, giga_loss[loss=0.2891, simple_loss=0.3746, pruned_loss=0.1018, over 28924.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3532, pruned_loss=0.1031, over 5674662.07 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3615, pruned_loss=0.1162, over 5699365.19 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3535, pruned_loss=0.1025, over 5665050.60 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:59:38,677 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-10 00:59:45,293 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:00:24,023 INFO [train.py:968] (0/2) Epoch 19, batch 30600, giga_loss[loss=0.2595, simple_loss=0.3195, pruned_loss=0.0998, over 24177.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3526, pruned_loss=0.103, over 5654990.34 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3612, pruned_loss=0.1162, over 5684895.71 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.353, pruned_loss=0.1023, over 5658045.06 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:00:27,254 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:00:29,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.247e+02 1.514e+03 1.920e+03 2.944e+03 5.925e+03, threshold=3.840e+03, percent-clipped=9.0
+2023-03-10 01:00:29,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:00:33,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:00:49,259 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:00:58,142 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:01:16,453 INFO [train.py:968] (0/2) Epoch 19, batch 30650, giga_loss[loss=0.2532, simple_loss=0.3423, pruned_loss=0.08206, over 29142.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3506, pruned_loss=0.1011, over 5664047.64 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3613, pruned_loss=0.1164, over 5688448.44 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3507, pruned_loss=0.1002, over 5662907.94 frames. ], batch size: 113, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:01:17,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:01:56,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0959, 1.1831, 3.2264, 2.8993], device='cuda:0'), covar=tensor([0.1645, 0.2836, 0.0508, 0.1519], device='cuda:0'), in_proj_covar=tensor([0.0749, 0.0642, 0.0945, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 01:01:57,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:02:05,243 INFO [train.py:968] (0/2) Epoch 19, batch 30700, giga_loss[loss=0.2504, simple_loss=0.3321, pruned_loss=0.08433, over 28545.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3475, pruned_loss=0.09912, over 5667782.19 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3604, pruned_loss=0.1159, over 5694227.88 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3481, pruned_loss=0.09841, over 5661315.11 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:02:07,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.960e+02 1.433e+03 1.817e+03 2.402e+03 1.055e+04, threshold=3.633e+03, percent-clipped=7.0
+2023-03-10 01:02:09,326 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:02:13,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:02:19,812 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9828, 1.1984, 1.1332, 0.9370], device='cuda:0'), covar=tensor([0.2086, 0.2038, 0.1179, 0.1785], device='cuda:0'), in_proj_covar=tensor([0.1904, 0.1828, 0.1762, 0.1903], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 01:02:42,824 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:02:52,873 INFO [train.py:968] (0/2) Epoch 19, batch 30750, giga_loss[loss=0.2468, simple_loss=0.3291, pruned_loss=0.08224, over 28932.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3455, pruned_loss=0.09836, over 5657857.16 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3593, pruned_loss=0.1155, over 5692659.04 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3464, pruned_loss=0.09756, over 5653471.27 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:02:53,900 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:02:55,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:20,979 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:25,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:33,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:37,374 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:39,143 INFO [train.py:968] (0/2) Epoch 19, batch 30800, giga_loss[loss=0.2352, simple_loss=0.3207, pruned_loss=0.07486, over 28900.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3461, pruned_loss=0.09862, over 5670410.12 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3586, pruned_loss=0.1151, over 5699089.95 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.347, pruned_loss=0.09771, over 5660168.95 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:03:44,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.348e+02 1.652e+03 2.098e+03 2.883e+03 8.084e+03, threshold=4.197e+03, percent-clipped=14.0
+2023-03-10 01:03:55,568 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:58,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:04:04,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:04:13,090 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:04:15,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:04:26,532 INFO [train.py:968] (0/2) Epoch 19, batch 30850, libri_loss[loss=0.2433, simple_loss=0.3057, pruned_loss=0.0904, over 28121.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3448, pruned_loss=0.09776, over 5656729.76 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3582, pruned_loss=0.115, over 5689348.63 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3454, pruned_loss=0.09648, over 5655602.78 frames. ], batch size: 62, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:04:45,924 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:05:04,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5238, 1.7814, 1.7561, 1.3230], device='cuda:0'), covar=tensor([0.1869, 0.2735, 0.1581, 0.1948], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0697, 0.0927, 0.0828], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-10 01:05:06,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:05:17,325 INFO [train.py:968] (0/2) Epoch 19, batch 30900, giga_loss[loss=0.2452, simple_loss=0.3173, pruned_loss=0.0865, over 27714.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3409, pruned_loss=0.09456, over 5660895.94 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3575, pruned_loss=0.1146, over 5693490.40 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3418, pruned_loss=0.09353, over 5655591.34 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:05:22,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.394e+02 1.421e+03 1.809e+03 2.541e+03 7.972e+03, threshold=3.618e+03, percent-clipped=2.0
+2023-03-10 01:05:25,647 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:05:44,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:05:47,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:01,866 INFO [train.py:968] (0/2) Epoch 19, batch 30950, giga_loss[loss=0.2339, simple_loss=0.3147, pruned_loss=0.07652, over 28679.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3384, pruned_loss=0.09326, over 5668022.13 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3573, pruned_loss=0.1147, over 5691017.17 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3385, pruned_loss=0.09153, over 5665337.85 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:06:12,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:15,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:16,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:16,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:20,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:40,417 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-10 01:06:43,866 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:46,792 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:52,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:52,576 INFO [train.py:968] (0/2) Epoch 19, batch 31000, giga_loss[loss=0.2519, simple_loss=0.3315, pruned_loss=0.0862, over 28570.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3362, pruned_loss=0.09257, over 5660029.81 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3572, pruned_loss=0.1148, over 5684773.31 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3361, pruned_loss=0.09075, over 5663412.01 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:06:56,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.025e+02 1.391e+03 1.742e+03 2.205e+03 5.839e+03, threshold=3.484e+03, percent-clipped=5.0
+2023-03-10 01:07:22,945 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:07:26,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:07:41,840 INFO [train.py:968] (0/2) Epoch 19, batch 31050, giga_loss[loss=0.2505, simple_loss=0.3262, pruned_loss=0.08737, over 28805.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3365, pruned_loss=0.09327, over 5656787.87 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3572, pruned_loss=0.1149, over 5687481.18 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3358, pruned_loss=0.09109, over 5656026.36 frames. ], batch size: 92, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:07:56,538 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:08:08,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2718, 3.0921, 1.4816, 1.4146], device='cuda:0'), covar=tensor([0.0994, 0.0336, 0.0977, 0.1389], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0547, 0.0376, 0.0418], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 01:08:36,462 INFO [train.py:968] (0/2) Epoch 19, batch 31100, libri_loss[loss=0.3206, simple_loss=0.3829, pruned_loss=0.1292, over 28994.00 frames. ], tot_loss[loss=0.262, simple_loss=0.337, pruned_loss=0.09355, over 5640525.36 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3572, pruned_loss=0.115, over 5681832.65 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.336, pruned_loss=0.09122, over 5643617.40 frames. ], batch size: 107, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:08:41,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.008e+02 1.610e+03 2.125e+03 3.173e+03 8.997e+03, threshold=4.250e+03, percent-clipped=19.0
+2023-03-10 01:09:23,318 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:09:27,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:09:32,188 INFO [train.py:968] (0/2) Epoch 19, batch 31150, giga_loss[loss=0.2406, simple_loss=0.3308, pruned_loss=0.0752, over 28050.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3397, pruned_loss=0.09439, over 5641526.16 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3563, pruned_loss=0.1148, over 5687375.17 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3392, pruned_loss=0.0922, over 5638099.15 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:10:00,356 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:10:32,502 INFO [train.py:968] (0/2) Epoch 19, batch 31200, giga_loss[loss=0.2472, simple_loss=0.329, pruned_loss=0.08268, over 28662.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3394, pruned_loss=0.0935, over 5635168.54 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3561, pruned_loss=0.1147, over 5689007.11 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.339, pruned_loss=0.09157, over 5630492.05 frames. ], batch size: 60, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:10:38,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.816e+02 1.454e+03 2.019e+03 2.672e+03 7.007e+03, threshold=4.038e+03, percent-clipped=8.0
+2023-03-10 01:11:12,476 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:11:44,572 INFO [train.py:968] (0/2) Epoch 19, batch 31250, giga_loss[loss=0.2192, simple_loss=0.2883, pruned_loss=0.07505, over 24415.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3395, pruned_loss=0.09358, over 5637748.61 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3557, pruned_loss=0.1145, over 5691365.88 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3393, pruned_loss=0.09198, over 5631660.72 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:12:01,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9577, 1.1517, 1.1255, 0.9790], device='cuda:0'), covar=tensor([0.2144, 0.2312, 0.1345, 0.1829], device='cuda:0'), in_proj_covar=tensor([0.1885, 0.1812, 0.1745, 0.1879], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 01:12:26,713 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:12:28,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:12:48,124 INFO [train.py:968] (0/2) Epoch 19, batch 31300, giga_loss[loss=0.2614, simple_loss=0.3429, pruned_loss=0.0899, over 28876.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3373, pruned_loss=0.09134, over 5646680.94 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3556, pruned_loss=0.1145, over 5693206.46 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3371, pruned_loss=0.08994, over 5639775.30 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:12:56,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.113e+02 1.315e+03 1.622e+03 2.194e+03 5.799e+03, threshold=3.244e+03, percent-clipped=3.0
+2023-03-10 01:13:29,806 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853919.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 01:13:50,892 INFO [train.py:968] (0/2) Epoch 19, batch 31350, giga_loss[loss=0.2363, simple_loss=0.3171, pruned_loss=0.07776, over 28944.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3355, pruned_loss=0.0891, over 5645190.19 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3557, pruned_loss=0.1144, over 5697224.50 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3347, pruned_loss=0.08736, over 5634925.88 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:14:22,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3586, 1.8445, 1.7590, 1.5823], device='cuda:0'), covar=tensor([0.1855, 0.1644, 0.1866, 0.1855], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0736, 0.0699, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 01:14:53,283 INFO [train.py:968] (0/2) Epoch 19, batch 31400, giga_loss[loss=0.269, simple_loss=0.3308, pruned_loss=0.1036, over 26945.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3329, pruned_loss=0.08846, over 5655418.80 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3553, pruned_loss=0.1143, over 5701423.26 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3323, pruned_loss=0.08677, over 5642697.31 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:14:58,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.966e+02 1.364e+03 1.906e+03 3.073e+03 8.130e+03, threshold=3.813e+03, percent-clipped=22.0
+2023-03-10 01:15:08,983 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-854000.pt
+2023-03-10 01:15:26,930 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:15:29,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=854017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:15:51,516 INFO [train.py:968] (0/2) Epoch 19, batch 31450, giga_loss[loss=0.2861, simple_loss=0.3565, pruned_loss=0.1079, over 28979.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3332, pruned_loss=0.08947, over 5658610.60 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3552, pruned_loss=0.1143, over 5695421.06 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3322, pruned_loss=0.08736, over 5652719.80 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:15:52,080 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-10 01:16:01,464 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:16:07,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8868, 1.4524, 1.3325, 1.1991], device='cuda:0'), covar=tensor([0.2177, 0.1599, 0.2164, 0.1843], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0737, 0.0700, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 01:16:48,684 INFO [train.py:968] (0/2) Epoch 19, batch 31500, giga_loss[loss=0.3058, simple_loss=0.3651, pruned_loss=0.1232, over 26821.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3329, pruned_loss=0.0894, over 5670739.06 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3548, pruned_loss=0.1144, over 5700732.40 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3319, pruned_loss=0.08704, over 5660617.60 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:16:56,211 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-10 01:16:56,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.104e+02 1.585e+03 1.961e+03 2.585e+03 4.676e+03, threshold=3.922e+03, percent-clipped=5.0
+2023-03-10 01:17:45,802 INFO [train.py:968] (0/2) Epoch 19, batch 31550, giga_loss[loss=0.2167, simple_loss=0.312, pruned_loss=0.06069, over 29041.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3329, pruned_loss=0.08869, over 5675371.35 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3539, pruned_loss=0.114, over 5706937.46 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3324, pruned_loss=0.08647, over 5660943.91 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:17:55,975 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4793, 1.7409, 1.1136, 1.3138], device='cuda:0'), covar=tensor([0.1015, 0.0560, 0.1151, 0.1146], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0443, 0.0510, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 01:18:39,924 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-10 01:18:51,267 INFO [train.py:968] (0/2) Epoch 19, batch 31600, giga_loss[loss=0.2608, simple_loss=0.3401, pruned_loss=0.09076, over 28847.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3343, pruned_loss=0.08934, over 5668023.19 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3536, pruned_loss=0.1139, over 5712813.66 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3335, pruned_loss=0.08696, over 5650013.84 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:18:55,611 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=854190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:18:58,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.921e+02 1.505e+03 1.872e+03 2.484e+03 8.024e+03, threshold=3.745e+03, percent-clipped=9.0
+2023-03-10 01:19:17,129 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4151, 1.4099, 4.1304, 3.1684], device='cuda:0'), covar=tensor([0.1582, 0.2549, 0.0392, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0744, 0.0639, 0.0936, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 01:19:26,694 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-10 01:19:53,781 INFO [train.py:968] (0/2) Epoch 19, batch 31650, libri_loss[loss=0.2995, simple_loss=0.3648, pruned_loss=0.1171, over 27581.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3316, pruned_loss=0.08772, over 5676649.03 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3529, pruned_loss=0.1135, over 5707232.97 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.331, pruned_loss=0.08537, over 5665423.10 frames. ], batch size: 116, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:20:03,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=854245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:20:41,642 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 01:21:03,795 INFO [train.py:968] (0/2) Epoch 19, batch 31700, libri_loss[loss=0.2287, simple_loss=0.2971, pruned_loss=0.08014, over 29575.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3336, pruned_loss=0.08956, over 5664162.56 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3525, pruned_loss=0.1135, over 5700929.79 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3332, pruned_loss=0.08737, over 5659719.75 frames. ], batch size: 75, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:21:09,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=854294.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 01:21:10,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.706e+02 1.426e+03 1.924e+03 2.697e+03 1.048e+04, threshold=3.847e+03, percent-clipped=12.0
+2023-03-10 01:21:25,552 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-10 01:21:55,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:21:58,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=854336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:22:00,887 INFO [train.py:968] (0/2) Epoch 19, batch 31750, giga_loss[loss=0.2477, simple_loss=0.3461, pruned_loss=0.07468, over 28987.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3379, pruned_loss=0.09053, over 5669410.79 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3526, pruned_loss=0.1134, over 5698516.78 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3368, pruned_loss=0.08795, over 5666281.99 frames. ], batch size: 155, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:22:37,018 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:23:02,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2434, 1.2213, 4.0262, 3.3292], device='cuda:0'), covar=tensor([0.1759, 0.2817, 0.0379, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0741, 0.0637, 0.0933, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 01:23:04,332 INFO [train.py:968] (0/2) Epoch 19, batch 31800, giga_loss[loss=0.2367, simple_loss=0.339, pruned_loss=0.06719, over 28900.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.34, pruned_loss=0.08928, over 5668512.22 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3526, pruned_loss=0.1132, over 5704213.80 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3388, pruned_loss=0.08669, over 5660043.22 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:23:05,048 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:23:09,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=854391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:23:13,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.858e+02 1.397e+03 1.945e+03 2.961e+03 1.235e+04, threshold=3.889e+03, percent-clipped=10.0
+2023-03-10 01:23:19,023 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=854400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:23:40,516 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:23:55,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=854434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:24:00,334 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854437.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 01:24:00,823 INFO [train.py:968] (0/2) Epoch 19, batch 31850, giga_loss[loss=0.2762, simple_loss=0.3631, pruned_loss=0.09462, over 28972.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3412, pruned_loss=0.08971, over 5667764.57 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3521, pruned_loss=0.1131, over 5708625.37 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3402, pruned_loss=0.08666, over 5655376.90 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:24:03,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=854440.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 01:24:32,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3384, 1.3255, 3.7226, 3.0546], device='cuda:0'), covar=tensor([0.1606, 0.2773, 0.0461, 0.1100], device='cuda:0'), in_proj_covar=tensor([0.0740, 0.0636, 0.0932, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 01:24:35,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854469.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 01:24:58,390 INFO [train.py:968] (0/2) Epoch 19, batch 31900, giga_loss[loss=0.2074, simple_loss=0.3046, pruned_loss=0.05507, over 28856.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3412, pruned_loss=0.08883, over 5679530.15 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.352, pruned_loss=0.1131, over 5713907.65 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3401, pruned_loss=0.08562, over 5663633.39 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:25:06,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.814e+02 1.419e+03 1.741e+03 2.482e+03 5.506e+03, threshold=3.482e+03, percent-clipped=7.0
+2023-03-10 01:25:59,184 INFO [train.py:968] (0/2) Epoch 19, batch 31950, giga_loss[loss=0.2688, simple_loss=0.3527, pruned_loss=0.09244, over 28041.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3416, pruned_loss=0.08987, over 5689849.60 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3515, pruned_loss=0.1128, over 5719380.05 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.341, pruned_loss=0.08694, over 5671207.39 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:26:35,178 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 01:27:05,093 INFO [train.py:968] (0/2) Epoch 19, batch 32000, giga_loss[loss=0.2429, simple_loss=0.311, pruned_loss=0.08735, over 24380.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3392, pruned_loss=0.08987, over 5680965.21 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3508, pruned_loss=0.1124, over 5713457.15 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3388, pruned_loss=0.08702, over 5669017.44 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:27:13,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.672e+02 1.331e+03 1.765e+03 2.352e+03 4.473e+03, threshold=3.531e+03, percent-clipped=9.0
+2023-03-10 01:27:29,650 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4539, 1.3693, 3.9403, 3.3257], device='cuda:0'), covar=tensor([0.1549, 0.2715, 0.0441, 0.0816], device='cuda:0'), in_proj_covar=tensor([0.0739, 0.0635, 0.0933, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 01:28:23,973 INFO [train.py:968] (0/2) Epoch 19, batch 32050, giga_loss[loss=0.3183, simple_loss=0.3588, pruned_loss=0.1389, over 27144.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3408, pruned_loss=0.0918, over 5675612.46 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3508, pruned_loss=0.1124, over 5711930.52 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3403, pruned_loss=0.08902, over 5666645.06 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:29:38,069 INFO [train.py:968] (0/2) Epoch 19, batch 32100, giga_loss[loss=0.2459, simple_loss=0.3304, pruned_loss=0.0807, over 28699.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3363, pruned_loss=0.08897, over 5681255.80 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3504, pruned_loss=0.1122, over 5713280.83 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3361, pruned_loss=0.08654, over 5672284.59 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:29:46,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.072e+02 1.373e+03 1.816e+03 2.761e+03 5.653e+03, threshold=3.632e+03, percent-clipped=10.0
+2023-03-10 01:30:40,837 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-10 01:30:44,682 INFO [train.py:968] (0/2) Epoch 19, batch 32150, giga_loss[loss=0.248, simple_loss=0.3245, pruned_loss=0.08573, over 29113.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3339, pruned_loss=0.08729, over 5681904.49 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3503, pruned_loss=0.1121, over 5715867.31 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3336, pruned_loss=0.08512, over 5672020.43 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:31:04,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-10 01:31:34,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=854775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:31:49,696 INFO [train.py:968] (0/2) Epoch 19, batch 32200, giga_loss[loss=0.2475, simple_loss=0.3234, pruned_loss=0.0858, over 28471.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3325, pruned_loss=0.08729, over 5687553.61 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3502, pruned_loss=0.1121, over 5721168.81 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3318, pruned_loss=0.08486, over 5673860.78 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:31:59,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.845e+02 1.395e+03 1.872e+03 2.248e+03 4.139e+03, threshold=3.745e+03, percent-clipped=5.0
+2023-03-10 01:32:00,605 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-10 01:32:13,426 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=854809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:32:36,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3128, 1.8851, 1.3063, 0.5671], device='cuda:0'), covar=tensor([0.4117, 0.2308, 0.3565, 0.5238], device='cuda:0'), in_proj_covar=tensor([0.1713, 0.1619, 0.1582, 0.1405], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 01:32:45,700 INFO [train.py:968] (0/2) Epoch 19, batch 32250, libri_loss[loss=0.3474, simple_loss=0.3933, pruned_loss=0.1508, over 19328.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3356, pruned_loss=0.08945, over 5690545.07 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3497, pruned_loss=0.1121, over 5721249.85 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3346, pruned_loss=0.0861, over 5678479.94 frames. ], batch size: 187, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:33:44,053 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.43 vs. limit=5.0
+2023-03-10 01:33:44,136 INFO [train.py:968] (0/2) Epoch 19, batch 32300, giga_loss[loss=0.2291, simple_loss=0.3098, pruned_loss=0.0742, over 28846.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3378, pruned_loss=0.09082, over 5694414.49 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3492, pruned_loss=0.1117, over 5722575.96 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3371, pruned_loss=0.08792, over 5682921.58 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:33:51,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.124e+02 1.536e+03 1.997e+03 3.199e+03 1.016e+04, threshold=3.994e+03, percent-clipped=16.0
+2023-03-10 01:34:27,707 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:34:27,836 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-10 01:34:33,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=854921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:34:50,251 INFO [train.py:968] (0/2) Epoch 19, batch 32350, giga_loss[loss=0.252, simple_loss=0.3354, pruned_loss=0.08434, over 28352.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3359, pruned_loss=0.0906, over 5692291.43 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3491, pruned_loss=0.1116, over 5721486.21 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3354, pruned_loss=0.08815, over 5683445.36 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:35:06,046 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:35:08,336 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:35:12,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=854955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:35:46,435 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:35:50,648 INFO [train.py:968] (0/2) Epoch 19, batch 32400, giga_loss[loss=0.2319, simple_loss=0.3188, pruned_loss=0.07249, over 28456.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3368, pruned_loss=0.09211, over 5679784.97 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.349, pruned_loss=0.1117, over 5715743.90 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3359, pruned_loss=0.0894, over 5676810.49 frames. ], batch size: 369, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:36:00,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.020e+02 1.390e+03 1.913e+03 2.748e+03 1.303e+04, threshold=3.825e+03, percent-clipped=6.0
+2023-03-10 01:36:53,566 INFO [train.py:968] (0/2) Epoch 19, batch 32450, giga_loss[loss=0.2473, simple_loss=0.3464, pruned_loss=0.07414, over 28938.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3365, pruned_loss=0.09151, over 5682414.84 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3487, pruned_loss=0.1114, over 5721469.19 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3359, pruned_loss=0.08904, over 5674033.21 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:38:08,810 INFO [train.py:968] (0/2) Epoch 19, batch 32500, giga_loss[loss=0.2209, simple_loss=0.3169, pruned_loss=0.06241, over 28869.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3384, pruned_loss=0.09101, over 5679826.21 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3485, pruned_loss=0.1113, over 5722438.85 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3381, pruned_loss=0.08909, over 5672188.27 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:38:26,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.873e+02 1.429e+03 1.934e+03 2.735e+03 5.738e+03, threshold=3.868e+03, percent-clipped=9.0
+2023-03-10 01:39:29,288 INFO [train.py:968] (0/2) Epoch 19, batch 32550, giga_loss[loss=0.2347, simple_loss=0.3221, pruned_loss=0.07369, over 28865.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3381, pruned_loss=0.09082, over 5672163.50 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3489, pruned_loss=0.1117, over 5726405.27 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3372, pruned_loss=0.08854, over 5661584.14 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:40:42,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-10 01:40:42,840 INFO [train.py:968] (0/2) Epoch 19, batch 32600, giga_loss[loss=0.3079, simple_loss=0.3597, pruned_loss=0.1281, over 26870.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3342, pruned_loss=0.08979, over 5682653.08 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3489, pruned_loss=0.1118, over 5728306.76 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3334, pruned_loss=0.08771, over 5672077.58 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:40:55,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.276e+02 1.428e+03 1.924e+03 2.760e+03 6.787e+03, threshold=3.849e+03, percent-clipped=6.0
+2023-03-10 01:41:50,411 INFO [train.py:968] (0/2) Epoch 19, batch 32650, giga_loss[loss=0.2691, simple_loss=0.338, pruned_loss=0.1001, over 28823.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3288, pruned_loss=0.08748, over 5681663.10 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3491, pruned_loss=0.112, over 5730069.35 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3278, pruned_loss=0.08545, over 5671272.86 frames. ], batch size: 263, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:42:26,844 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=855264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:42:52,033 INFO [train.py:968] (0/2) Epoch 19, batch 32700, giga_loss[loss=0.2678, simple_loss=0.3303, pruned_loss=0.1026, over 26799.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3292, pruned_loss=0.08797, over 5671046.97 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3491, pruned_loss=0.1119, over 5722868.71 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3276, pruned_loss=0.08559, over 5667212.96 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:43:04,285 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.988e+02 1.561e+03 2.192e+03 3.685e+03 2.568e+04, threshold=4.383e+03, percent-clipped=21.0
+2023-03-10 01:43:53,249 INFO [train.py:968] (0/2) Epoch 19, batch 32750, giga_loss[loss=0.2806, simple_loss=0.3543, pruned_loss=0.1035, over 28660.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3311, pruned_loss=0.08919, over 5674223.33 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3486, pruned_loss=0.1116, over 5723644.72 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3302, pruned_loss=0.08726, over 5669823.09 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:44:53,919 INFO [train.py:968] (0/2) Epoch 19, batch 32800, giga_loss[loss=0.2244, simple_loss=0.3146, pruned_loss=0.06705, over 29041.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.329, pruned_loss=0.08763, over 5673736.26 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3486, pruned_loss=0.1117, over 5725296.86 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3278, pruned_loss=0.08541, over 5667539.38 frames. ], batch size: 155, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:45:06,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.078e+02 1.354e+03 1.855e+03 2.345e+03 5.895e+03, threshold=3.709e+03, percent-clipped=2.0
+2023-03-10 01:45:25,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4571, 1.3324, 4.3990, 3.5199], device='cuda:0'), covar=tensor([0.1665, 0.2853, 0.0419, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0635, 0.0930, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 01:45:56,291 INFO [train.py:968] (0/2) Epoch 19, batch 32850, giga_loss[loss=0.2857, simple_loss=0.3556, pruned_loss=0.1079, over 28332.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3281, pruned_loss=0.08662, over 5662692.83 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3484, pruned_loss=0.1117, over 5718345.49 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3268, pruned_loss=0.08429, over 5662921.03 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:46:56,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6487, 1.9448, 1.2861, 1.5637], device='cuda:0'), covar=tensor([0.0968, 0.0595, 0.1004, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0438, 0.0508, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 01:47:04,815 INFO [train.py:968] (0/2) Epoch 19, batch 32900, giga_loss[loss=0.2516, simple_loss=0.3288, pruned_loss=0.08716, over 29009.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3265, pruned_loss=0.08605, over 5664489.43 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3482, pruned_loss=0.1116, over 5721907.65 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3253, pruned_loss=0.08389, over 5660655.44 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:47:22,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.725e+02 1.364e+03 1.756e+03 2.409e+03 1.232e+04, threshold=3.513e+03, percent-clipped=11.0
+2023-03-10 01:47:29,089 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-10 01:47:53,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9666, 2.2642, 2.0077, 1.9431], device='cuda:0'), covar=tensor([0.1772, 0.2325, 0.2079, 0.2182], device='cuda:0'), in_proj_covar=tensor([0.0450, 0.0728, 0.0692, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 01:48:15,143 INFO [train.py:968] (0/2) Epoch 19, batch 32950, giga_loss[loss=0.2442, simple_loss=0.3351, pruned_loss=0.0766, over 29027.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3264, pruned_loss=0.085, over 5676757.18 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3481, pruned_loss=0.1115, over 5722226.71 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3251, pruned_loss=0.08283, over 5672147.60 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:49:18,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2748, 1.3349, 3.4808, 3.0967], device='cuda:0'), covar=tensor([0.1541, 0.2682, 0.0432, 0.1580], device='cuda:0'), in_proj_covar=tensor([0.0738, 0.0635, 0.0930, 0.0868], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 01:49:24,790 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=855587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:49:25,231 INFO [train.py:968] (0/2) Epoch 19, batch 33000, giga_loss[loss=0.2629, simple_loss=0.3357, pruned_loss=0.09499, over 28037.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3272, pruned_loss=0.08552, over 5672073.65 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3482, pruned_loss=0.1115, over 5715276.39 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3259, pruned_loss=0.08345, over 5673130.84 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:49:25,236 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 01:49:33,753 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.1964, simple_loss=0.2975, pruned_loss=0.0476, over 944034.00 frames. 
+2023-03-10 01:49:33,753 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 01:49:39,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9168, 2.3289, 2.1771, 1.7518], device='cuda:0'), covar=tensor([0.2598, 0.1882, 0.1954, 0.2173], device='cuda:0'), in_proj_covar=tensor([0.1896, 0.1817, 0.1741, 0.1883], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 01:49:45,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.829e+02 1.265e+03 1.598e+03 2.054e+03 3.837e+03, threshold=3.195e+03, percent-clipped=3.0
+2023-03-10 01:50:39,561 INFO [train.py:968] (0/2) Epoch 19, batch 33050, giga_loss[loss=0.2451, simple_loss=0.3352, pruned_loss=0.07755, over 28490.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3277, pruned_loss=0.0866, over 5675134.64 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.348, pruned_loss=0.1115, over 5715272.00 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3267, pruned_loss=0.08494, over 5675796.90 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:50:40,793 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=855639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:51:42,099 INFO [train.py:968] (0/2) Epoch 19, batch 33100, libri_loss[loss=0.2896, simple_loss=0.3559, pruned_loss=0.1117, over 28624.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3268, pruned_loss=0.08521, over 5671164.10 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3482, pruned_loss=0.1117, over 5716523.45 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3255, pruned_loss=0.08321, over 5669684.18 frames. ], batch size: 106, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:51:57,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.862e+02 1.516e+03 2.106e+03 3.214e+03 1.161e+04, threshold=4.212e+03, percent-clipped=26.0
+2023-03-10 01:52:40,293 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=855735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:52:42,919 INFO [train.py:968] (0/2) Epoch 19, batch 33150, libri_loss[loss=0.2909, simple_loss=0.3514, pruned_loss=0.1152, over 28626.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3287, pruned_loss=0.08496, over 5665663.20 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3483, pruned_loss=0.1118, over 5718703.89 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3272, pruned_loss=0.08288, over 5661663.81 frames. ], batch size: 106, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:53:24,449 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=855774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:53:25,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3788, 1.9693, 1.4138, 0.5768], device='cuda:0'), covar=tensor([0.4683, 0.2712, 0.4392, 0.5797], device='cuda:0'), in_proj_covar=tensor([0.1704, 0.1609, 0.1572, 0.1395], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 01:53:33,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=855782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:53:36,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=855785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:53:40,711 INFO [train.py:968] (0/2) Epoch 19, batch 33200, giga_loss[loss=0.2819, simple_loss=0.3565, pruned_loss=0.1036, over 28747.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3312, pruned_loss=0.08615, over 5670846.00 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3476, pruned_loss=0.1114, over 5723282.39 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3302, pruned_loss=0.08412, over 5662269.54 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:53:56,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.460e+02 1.499e+03 1.786e+03 2.467e+03 5.705e+03, threshold=3.572e+03, percent-clipped=7.0
+2023-03-10 01:54:07,850 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 01:54:14,857 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=855814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:54:45,211 INFO [train.py:968] (0/2) Epoch 19, batch 33250, giga_loss[loss=0.2543, simple_loss=0.3266, pruned_loss=0.091, over 26779.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3324, pruned_loss=0.08673, over 5673439.74 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3476, pruned_loss=0.1115, over 5726742.70 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3312, pruned_loss=0.08452, over 5662351.28 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:55:00,937 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=855850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:55:01,916 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 01:55:17,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7974, 1.6791, 4.7675, 3.3925], device='cuda:0'), covar=tensor([0.1476, 0.2551, 0.0407, 0.0971], device='cuda:0'), in_proj_covar=tensor([0.0735, 0.0633, 0.0926, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 01:55:48,210 INFO [train.py:968] (0/2) Epoch 19, batch 33300, giga_loss[loss=0.2705, simple_loss=0.3426, pruned_loss=0.09923, over 27612.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3328, pruned_loss=0.08734, over 5667923.45 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3472, pruned_loss=0.1114, over 5720729.38 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3319, pruned_loss=0.08533, over 5663212.28 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:55:59,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.662e+02 1.341e+03 1.707e+03 2.513e+03 6.528e+03, threshold=3.413e+03, percent-clipped=10.0
+2023-03-10 01:56:45,951 INFO [train.py:968] (0/2) Epoch 19, batch 33350, giga_loss[loss=0.2213, simple_loss=0.3127, pruned_loss=0.06492, over 28859.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3289, pruned_loss=0.08477, over 5666933.53 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3471, pruned_loss=0.1115, over 5711892.15 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3279, pruned_loss=0.08245, over 5668694.90 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:56:52,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-10 01:57:17,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=855962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:57:43,973 INFO [train.py:968] (0/2) Epoch 19, batch 33400, giga_loss[loss=0.292, simple_loss=0.3623, pruned_loss=0.1108, over 28780.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3285, pruned_loss=0.08512, over 5666003.58 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3467, pruned_loss=0.1113, over 5708367.25 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3275, pruned_loss=0.08258, over 5669242.26 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:57:59,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.726e+02 1.413e+03 1.865e+03 2.745e+03 1.204e+04, threshold=3.729e+03, percent-clipped=17.0
+2023-03-10 01:57:59,194 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-856000.pt
+2023-03-10 01:58:15,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0069, 1.1610, 5.3872, 3.6058], device='cuda:0'), covar=tensor([0.1453, 0.2825, 0.0409, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.0738, 0.0637, 0.0930, 0.0868], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 01:58:20,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2240, 1.5254, 1.4255, 1.1506], device='cuda:0'), covar=tensor([0.2710, 0.2142, 0.1393, 0.2045], device='cuda:0'), in_proj_covar=tensor([0.1893, 0.1809, 0.1730, 0.1880], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 01:58:43,413 INFO [train.py:968] (0/2) Epoch 19, batch 33450, giga_loss[loss=0.2403, simple_loss=0.3069, pruned_loss=0.08679, over 24694.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3272, pruned_loss=0.08509, over 5661927.38 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3463, pruned_loss=0.111, over 5704355.74 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3262, pruned_loss=0.08252, over 5667579.25 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:59:41,031 INFO [train.py:968] (0/2) Epoch 19, batch 33500, giga_loss[loss=0.2598, simple_loss=0.339, pruned_loss=0.09033, over 27834.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3297, pruned_loss=0.08653, over 5667907.06 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3462, pruned_loss=0.1111, over 5705992.13 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3285, pruned_loss=0.08366, over 5669573.89 frames. ], batch size: 476, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:59:57,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.315e+02 1.261e+03 1.803e+03 2.697e+03 5.935e+03, threshold=3.605e+03, percent-clipped=10.0
+2023-03-10 02:00:07,223 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:00:10,442 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:00:14,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:00:48,042 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=856137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:00:48,493 INFO [train.py:968] (0/2) Epoch 19, batch 33550, giga_loss[loss=0.2695, simple_loss=0.3453, pruned_loss=0.09685, over 28917.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3308, pruned_loss=0.0867, over 5668010.19 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3463, pruned_loss=0.111, over 5706977.17 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3297, pruned_loss=0.08435, over 5668265.73 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:01:07,835 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:01:12,041 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 02:01:53,532 INFO [train.py:968] (0/2) Epoch 19, batch 33600, giga_loss[loss=0.2788, simple_loss=0.3598, pruned_loss=0.09892, over 28468.00 frames. ], tot_loss[loss=0.254, simple_loss=0.332, pruned_loss=0.08802, over 5664211.11 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3459, pruned_loss=0.1108, over 5709272.86 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3311, pruned_loss=0.08568, over 5661336.44 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:02:10,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.579e+02 1.356e+03 1.657e+03 2.356e+03 5.027e+03, threshold=3.314e+03, percent-clipped=5.0
+2023-03-10 02:02:41,849 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:02:43,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:02:58,507 INFO [train.py:968] (0/2) Epoch 19, batch 33650, giga_loss[loss=0.2418, simple_loss=0.3326, pruned_loss=0.07555, over 28896.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3359, pruned_loss=0.08996, over 5656885.12 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3458, pruned_loss=0.1109, over 5707446.72 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.335, pruned_loss=0.08764, over 5655056.37 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:03:13,873 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:03:17,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:03:55,771 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=856285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:03:58,728 INFO [train.py:968] (0/2) Epoch 19, batch 33700, giga_loss[loss=0.3031, simple_loss=0.3612, pruned_loss=0.1225, over 26966.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3383, pruned_loss=0.0907, over 5661446.73 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.346, pruned_loss=0.1111, over 5709746.83 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3374, pruned_loss=0.08845, over 5657344.05 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:04:05,366 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:04:12,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:04:18,071 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.723e+02 1.333e+03 1.725e+03 2.253e+03 4.753e+03, threshold=3.450e+03, percent-clipped=9.0
+2023-03-10 02:04:23,060 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3967, 1.9020, 1.3877, 0.7724], device='cuda:0'), covar=tensor([0.5347, 0.2937, 0.3755, 0.5472], device='cuda:0'), in_proj_covar=tensor([0.1702, 0.1611, 0.1570, 0.1395], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 02:04:49,319 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=856324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:04:54,175 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-10 02:05:10,657 INFO [train.py:968] (0/2) Epoch 19, batch 33750, giga_loss[loss=0.2267, simple_loss=0.3125, pruned_loss=0.0705, over 28901.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3376, pruned_loss=0.0904, over 5674369.82 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3455, pruned_loss=0.111, over 5715954.48 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.337, pruned_loss=0.08809, over 5664254.60 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:05:47,875 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:05:53,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:06:14,150 INFO [train.py:968] (0/2) Epoch 19, batch 33800, giga_loss[loss=0.2578, simple_loss=0.339, pruned_loss=0.08827, over 28905.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3352, pruned_loss=0.08962, over 5670977.59 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.346, pruned_loss=0.1114, over 5707575.84 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3342, pruned_loss=0.08674, over 5668209.95 frames. ], batch size: 284, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:06:30,717 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=856400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:06:32,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.177e+02 1.619e+03 1.923e+03 2.869e+03 8.275e+03, threshold=3.847e+03, percent-clipped=11.0
+2023-03-10 02:07:16,636 INFO [train.py:968] (0/2) Epoch 19, batch 33850, giga_loss[loss=0.2333, simple_loss=0.3176, pruned_loss=0.0745, over 28669.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3346, pruned_loss=0.08939, over 5675507.22 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3458, pruned_loss=0.1114, over 5710444.12 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3336, pruned_loss=0.08667, over 5669931.10 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:07:39,428 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:07:56,866 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:08:26,179 INFO [train.py:968] (0/2) Epoch 19, batch 33900, libri_loss[loss=0.2542, simple_loss=0.3129, pruned_loss=0.09774, over 29480.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3327, pruned_loss=0.08883, over 5679369.92 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3452, pruned_loss=0.111, over 5712353.82 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3323, pruned_loss=0.08653, over 5672522.85 frames. ], batch size: 70, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:08:42,858 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.506e+02 1.376e+03 2.094e+03 2.952e+03 6.719e+03, threshold=4.188e+03, percent-clipped=12.0
+2023-03-10 02:09:32,515 INFO [train.py:968] (0/2) Epoch 19, batch 33950, giga_loss[loss=0.2028, simple_loss=0.2866, pruned_loss=0.05946, over 28896.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3303, pruned_loss=0.08794, over 5682936.20 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3452, pruned_loss=0.111, over 5713935.24 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3298, pruned_loss=0.08598, over 5675745.74 frames. ], batch size: 112, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:09:48,150 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:10:33,852 INFO [train.py:968] (0/2) Epoch 19, batch 34000, giga_loss[loss=0.2266, simple_loss=0.316, pruned_loss=0.06863, over 29049.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3304, pruned_loss=0.08717, over 5684014.26 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3457, pruned_loss=0.1114, over 5710164.66 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3292, pruned_loss=0.0847, over 5680983.72 frames. ], batch size: 128, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:10:46,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:10:48,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.681e+02 1.337e+03 1.685e+03 2.478e+03 5.216e+03, threshold=3.371e+03, percent-clipped=5.0
+2023-03-10 02:11:34,198 INFO [train.py:968] (0/2) Epoch 19, batch 34050, giga_loss[loss=0.2411, simple_loss=0.3191, pruned_loss=0.08155, over 28968.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3283, pruned_loss=0.08556, over 5675162.60 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3456, pruned_loss=0.1112, over 5716104.22 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.327, pruned_loss=0.08306, over 5666461.82 frames. ], batch size: 106, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:11:41,416 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:12:29,622 INFO [train.py:968] (0/2) Epoch 19, batch 34100, giga_loss[loss=0.2964, simple_loss=0.3563, pruned_loss=0.1183, over 24522.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3296, pruned_loss=0.08437, over 5656236.17 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.346, pruned_loss=0.1116, over 5691881.92 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3277, pruned_loss=0.0813, over 5669939.50 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:12:44,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.184e+02 1.438e+03 2.025e+03 2.692e+03 1.068e+04, threshold=4.050e+03, percent-clipped=13.0
+2023-03-10 02:13:15,843 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 02:13:29,293 INFO [train.py:968] (0/2) Epoch 19, batch 34150, giga_loss[loss=0.241, simple_loss=0.3311, pruned_loss=0.07546, over 28664.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.331, pruned_loss=0.08384, over 5659237.65 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3459, pruned_loss=0.1115, over 5685779.65 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3294, pruned_loss=0.08115, over 5675389.97 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:13:34,001 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:13:38,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:14:14,166 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=856773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:14:37,676 INFO [train.py:968] (0/2) Epoch 19, batch 34200, giga_loss[loss=0.2606, simple_loss=0.3389, pruned_loss=0.09112, over 28778.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3313, pruned_loss=0.08395, over 5659173.63 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3459, pruned_loss=0.1116, over 5688033.25 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3298, pruned_loss=0.08137, over 5669510.69 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:14:56,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.450e+02 1.350e+03 1.682e+03 2.224e+03 4.481e+03, threshold=3.363e+03, percent-clipped=3.0
+2023-03-10 02:15:35,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:15:50,816 INFO [train.py:968] (0/2) Epoch 19, batch 34250, giga_loss[loss=0.2213, simple_loss=0.2922, pruned_loss=0.07522, over 24551.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3324, pruned_loss=0.08508, over 5651248.40 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3462, pruned_loss=0.1118, over 5684867.69 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3307, pruned_loss=0.0823, over 5661948.00 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:15:54,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:16:57,902 INFO [train.py:968] (0/2) Epoch 19, batch 34300, giga_loss[loss=0.3084, simple_loss=0.3818, pruned_loss=0.1175, over 28783.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.333, pruned_loss=0.08557, over 5654059.53 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3461, pruned_loss=0.1117, over 5686701.24 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3316, pruned_loss=0.0831, over 5660555.83 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:17:01,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5680, 4.3913, 4.1407, 2.2052], device='cuda:0'), covar=tensor([0.0449, 0.0574, 0.0665, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.1161, 0.1076, 0.0917, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 02:17:13,352 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5672, 1.4486, 1.7216, 1.2545], device='cuda:0'), covar=tensor([0.2158, 0.3213, 0.1695, 0.1903], device='cuda:0'), in_proj_covar=tensor([0.0881, 0.0688, 0.0926, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-10 02:17:16,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.612e+03 1.982e+03 2.391e+03 4.715e+03, threshold=3.963e+03, percent-clipped=9.0
+2023-03-10 02:17:47,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:18:12,741 INFO [train.py:968] (0/2) Epoch 19, batch 34350, giga_loss[loss=0.2323, simple_loss=0.327, pruned_loss=0.06883, over 28888.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3335, pruned_loss=0.08478, over 5662539.64 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3461, pruned_loss=0.1117, over 5692290.46 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.332, pruned_loss=0.0822, over 5662161.91 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:18:17,511 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-10 02:18:24,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5107, 1.5777, 1.6089, 1.4536], device='cuda:0'), covar=tensor([0.2391, 0.2210, 0.1792, 0.2070], device='cuda:0'), in_proj_covar=tensor([0.1886, 0.1804, 0.1726, 0.1872], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 02:18:40,923 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 02:18:54,700 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:18:58,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:11,472 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:15,004 INFO [train.py:968] (0/2) Epoch 19, batch 34400, libri_loss[loss=0.2201, simple_loss=0.2871, pruned_loss=0.07657, over 29375.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3358, pruned_loss=0.08687, over 5667509.66 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3455, pruned_loss=0.1113, over 5691972.21 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3347, pruned_loss=0.08403, over 5665743.76 frames. ], batch size: 67, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:19:15,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:27,571 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:29,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.583e+02 1.460e+03 2.058e+03 2.846e+03 8.602e+03, threshold=4.116e+03, percent-clipped=11.0
+2023-03-10 02:19:33,087 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:42,316 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857011.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 02:19:49,717 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:57,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:20:15,155 INFO [train.py:968] (0/2) Epoch 19, batch 34450, giga_loss[loss=0.248, simple_loss=0.3409, pruned_loss=0.07757, over 28460.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3387, pruned_loss=0.08833, over 5675625.17 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3453, pruned_loss=0.1113, over 5695458.02 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3377, pruned_loss=0.08527, over 5670418.29 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:20:56,531 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:21:01,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:21:23,520 INFO [train.py:968] (0/2) Epoch 19, batch 34500, giga_loss[loss=0.2149, simple_loss=0.3045, pruned_loss=0.06266, over 29084.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3376, pruned_loss=0.0881, over 5662739.13 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3455, pruned_loss=0.1115, over 5680372.45 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3365, pruned_loss=0.08501, over 5671018.88 frames. ], batch size: 128, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:21:28,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5435, 2.2664, 1.5957, 0.8009], device='cuda:0'), covar=tensor([0.6001, 0.2998, 0.4198, 0.6471], device='cuda:0'), in_proj_covar=tensor([0.1695, 0.1601, 0.1568, 0.1391], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 02:21:42,602 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:21:48,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.255e+02 1.526e+03 1.907e+03 2.671e+03 7.402e+03, threshold=3.815e+03, percent-clipped=7.0
+2023-03-10 02:22:36,845 INFO [train.py:968] (0/2) Epoch 19, batch 34550, giga_loss[loss=0.2286, simple_loss=0.3157, pruned_loss=0.0707, over 28962.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3356, pruned_loss=0.08749, over 5670519.69 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3455, pruned_loss=0.1115, over 5679621.91 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3345, pruned_loss=0.08451, over 5677249.57 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:23:14,731 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:23:17,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:23:50,323 INFO [train.py:968] (0/2) Epoch 19, batch 34600, giga_loss[loss=0.2582, simple_loss=0.3194, pruned_loss=0.09846, over 25050.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3333, pruned_loss=0.086, over 5665220.58 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3452, pruned_loss=0.1114, over 5677257.23 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3324, pruned_loss=0.08297, over 5672120.36 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:24:01,693 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:24:09,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.028e+02 1.310e+03 1.535e+03 2.226e+03 6.677e+03, threshold=3.070e+03, percent-clipped=7.0
+2023-03-10 02:24:26,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2632, 0.8637, 0.8671, 1.3919], device='cuda:0'), covar=tensor([0.0796, 0.0407, 0.0390, 0.0896], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0116, 0.0118, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0068, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 02:24:52,361 INFO [train.py:968] (0/2) Epoch 19, batch 34650, giga_loss[loss=0.2458, simple_loss=0.3284, pruned_loss=0.08165, over 28685.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.332, pruned_loss=0.0848, over 5672293.54 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.345, pruned_loss=0.1111, over 5681992.71 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3311, pruned_loss=0.08183, over 5673327.23 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:25:47,608 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 02:25:50,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4259, 1.5686, 1.6328, 1.2291], device='cuda:0'), covar=tensor([0.1696, 0.2687, 0.1464, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.0882, 0.0688, 0.0927, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-10 02:25:55,625 INFO [train.py:968] (0/2) Epoch 19, batch 34700, giga_loss[loss=0.2384, simple_loss=0.3298, pruned_loss=0.07353, over 28491.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.333, pruned_loss=0.08562, over 5660778.16 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3446, pruned_loss=0.1111, over 5675078.56 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3324, pruned_loss=0.08287, over 5668117.40 frames. ], batch size: 370, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:26:02,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:26:12,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.155e+02 1.273e+03 1.609e+03 2.185e+03 6.812e+03, threshold=3.218e+03, percent-clipped=10.0
+2023-03-10 02:26:35,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9575, 2.5118, 2.3477, 1.8045], device='cuda:0'), covar=tensor([0.2966, 0.1805, 0.2015, 0.2329], device='cuda:0'), in_proj_covar=tensor([0.1885, 0.1797, 0.1721, 0.1868], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 02:26:53,707 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:26:56,760 INFO [train.py:968] (0/2) Epoch 19, batch 34750, libri_loss[loss=0.2885, simple_loss=0.3566, pruned_loss=0.1101, over 29282.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3361, pruned_loss=0.08707, over 5667362.35 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3447, pruned_loss=0.1109, over 5680542.47 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3353, pruned_loss=0.08442, over 5667907.29 frames. ], batch size: 94, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:27:36,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:27:51,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857386.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 02:27:53,714 INFO [train.py:968] (0/2) Epoch 19, batch 34800, giga_loss[loss=0.2624, simple_loss=0.3254, pruned_loss=0.09965, over 26750.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3351, pruned_loss=0.08689, over 5671228.63 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3443, pruned_loss=0.1107, over 5676181.06 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3346, pruned_loss=0.08449, over 5675704.02 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:28:11,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.728e+02 1.499e+03 2.060e+03 3.109e+03 1.448e+04, threshold=4.121e+03, percent-clipped=24.0
+2023-03-10 02:28:44,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4800, 1.5726, 1.2209, 1.5858], device='cuda:0'), covar=tensor([0.0707, 0.0306, 0.0343, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0116, 0.0118, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0067, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 02:28:51,599 INFO [train.py:968] (0/2) Epoch 19, batch 34850, giga_loss[loss=0.2292, simple_loss=0.2939, pruned_loss=0.0822, over 24310.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3332, pruned_loss=0.08712, over 5663282.15 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3441, pruned_loss=0.1104, over 5679641.24 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3329, pruned_loss=0.08512, over 5663410.60 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:29:51,464 INFO [train.py:968] (0/2) Epoch 19, batch 34900, giga_loss[loss=0.2842, simple_loss=0.3555, pruned_loss=0.1065, over 27534.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3339, pruned_loss=0.088, over 5668045.82 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3437, pruned_loss=0.1101, over 5680525.28 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3339, pruned_loss=0.08646, over 5667102.73 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:30:08,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.586e+02 1.578e+03 1.951e+03 2.716e+03 8.042e+03, threshold=3.903e+03, percent-clipped=6.0
+2023-03-10 02:30:21,253 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:30:23,551 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:30:31,783 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857529.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 02:30:34,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857532.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 02:30:35,560 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 02:30:40,150 INFO [train.py:968] (0/2) Epoch 19, batch 34950, giga_loss[loss=0.2789, simple_loss=0.3682, pruned_loss=0.09473, over 28706.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3423, pruned_loss=0.09303, over 5669861.31 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3439, pruned_loss=0.1101, over 5683464.18 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.342, pruned_loss=0.0913, over 5666206.58 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:30:48,496 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:30:59,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857561.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:31:28,907 INFO [train.py:968] (0/2) Epoch 19, batch 35000, giga_loss[loss=0.274, simple_loss=0.3598, pruned_loss=0.09413, over 28635.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3509, pruned_loss=0.09779, over 5671392.67 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.344, pruned_loss=0.1102, over 5685814.95 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3506, pruned_loss=0.09629, over 5666505.83 frames. ], batch size: 60, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:31:44,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.732e+02 1.299e+03 1.621e+03 2.167e+03 6.217e+03, threshold=3.243e+03, percent-clipped=6.0
+2023-03-10 02:32:00,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:32:00,505 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5962, 1.7790, 1.5204, 1.7243], device='cuda:0'), covar=tensor([0.2470, 0.2411, 0.2655, 0.2232], device='cuda:0'), in_proj_covar=tensor([0.1457, 0.1054, 0.1298, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 02:32:13,054 INFO [train.py:968] (0/2) Epoch 19, batch 35050, giga_loss[loss=0.2345, simple_loss=0.318, pruned_loss=0.07551, over 29019.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3519, pruned_loss=0.09919, over 5676349.30 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3437, pruned_loss=0.1098, over 5690541.15 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3522, pruned_loss=0.09796, over 5667489.97 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:32:38,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:32:53,268 INFO [train.py:968] (0/2) Epoch 19, batch 35100, giga_loss[loss=0.2226, simple_loss=0.3034, pruned_loss=0.0709, over 28966.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3465, pruned_loss=0.09724, over 5689751.03 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3442, pruned_loss=0.1101, over 5696696.79 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3464, pruned_loss=0.09571, over 5676926.62 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:33:05,915 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.222e+02 1.172e+03 1.545e+03 2.250e+03 1.434e+04, threshold=3.091e+03, percent-clipped=8.0
+2023-03-10 02:33:11,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:33:34,204 INFO [train.py:968] (0/2) Epoch 19, batch 35150, giga_loss[loss=0.2207, simple_loss=0.2994, pruned_loss=0.07105, over 28613.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3415, pruned_loss=0.09575, over 5681874.82 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3453, pruned_loss=0.1109, over 5685296.50 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3405, pruned_loss=0.09337, over 5682133.18 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:34:18,681 INFO [train.py:968] (0/2) Epoch 19, batch 35200, giga_loss[loss=0.2154, simple_loss=0.2952, pruned_loss=0.06778, over 28801.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.334, pruned_loss=0.09234, over 5673279.31 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3451, pruned_loss=0.1106, over 5678251.30 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3331, pruned_loss=0.09037, over 5679959.76 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:34:30,300 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.634e+02 1.166e+03 1.546e+03 2.135e+03 6.250e+03, threshold=3.093e+03, percent-clipped=10.0
+2023-03-10 02:34:32,014 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5784, 1.8593, 1.7745, 1.6763], device='cuda:0'), covar=tensor([0.1878, 0.1803, 0.2121, 0.1968], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0727, 0.0693, 0.0664], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 02:34:37,055 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:34:39,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:34:56,611 INFO [train.py:968] (0/2) Epoch 19, batch 35250, giga_loss[loss=0.2085, simple_loss=0.2872, pruned_loss=0.06488, over 28733.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3285, pruned_loss=0.0906, over 5678118.51 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3455, pruned_loss=0.1107, over 5679895.81 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3269, pruned_loss=0.08807, over 5682414.29 frames. ], batch size: 60, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:35:01,347 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:35:10,243 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:35:12,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:35:33,793 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857880.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:35:39,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:35:41,286 INFO [train.py:968] (0/2) Epoch 19, batch 35300, libri_loss[loss=0.2599, simple_loss=0.3344, pruned_loss=0.09273, over 29458.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.321, pruned_loss=0.08703, over 5679048.41 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3452, pruned_loss=0.1104, over 5680940.48 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3196, pruned_loss=0.08496, over 5681372.15 frames. ], batch size: 85, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:35:52,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.901e+02 1.130e+03 1.374e+03 1.891e+03 8.172e+03, threshold=2.747e+03, percent-clipped=7.0
+2023-03-10 02:35:54,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:35:54,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5044, 2.0230, 1.5906, 0.7346], device='cuda:0'), covar=tensor([0.5414, 0.2837, 0.3771, 0.6263], device='cuda:0'), in_proj_covar=tensor([0.1701, 0.1612, 0.1573, 0.1390], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 02:36:20,987 INFO [train.py:968] (0/2) Epoch 19, batch 35350, giga_loss[loss=0.1899, simple_loss=0.2724, pruned_loss=0.05371, over 29029.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3158, pruned_loss=0.08449, over 5683081.30 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3451, pruned_loss=0.1103, over 5686409.22 frames. ], giga_tot_loss[loss=0.2395, simple_loss=0.3143, pruned_loss=0.08241, over 5680243.69 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:37:03,285 INFO [train.py:968] (0/2) Epoch 19, batch 35400, giga_loss[loss=0.2069, simple_loss=0.2872, pruned_loss=0.06336, over 28830.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3127, pruned_loss=0.08267, over 5692852.71 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3451, pruned_loss=0.1102, over 5689710.42 frames. ], giga_tot_loss[loss=0.2363, simple_loss=0.3111, pruned_loss=0.08078, over 5687755.48 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:37:09,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:37:11,970 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-858000.pt
+2023-03-10 02:37:14,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.019e+02 1.104e+03 1.357e+03 1.604e+03 3.772e+03, threshold=2.713e+03, percent-clipped=6.0
+2023-03-10 02:37:43,100 INFO [train.py:968] (0/2) Epoch 19, batch 35450, giga_loss[loss=0.2284, simple_loss=0.2984, pruned_loss=0.07926, over 29147.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3097, pruned_loss=0.08122, over 5700930.57 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3453, pruned_loss=0.1102, over 5690952.98 frames. ], giga_tot_loss[loss=0.2335, simple_loss=0.308, pruned_loss=0.07947, over 5695865.34 frames. ], batch size: 128, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:38:28,093 INFO [train.py:968] (0/2) Epoch 19, batch 35500, giga_loss[loss=0.2009, simple_loss=0.2713, pruned_loss=0.06525, over 28796.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3079, pruned_loss=0.08065, over 5711743.70 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3457, pruned_loss=0.1103, over 5698588.37 frames. ], giga_tot_loss[loss=0.2305, simple_loss=0.3048, pruned_loss=0.07811, over 5700986.03 frames. ], batch size: 92, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:38:40,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.946e+02 1.041e+03 1.391e+03 2.178e+03 5.980e+03, threshold=2.783e+03, percent-clipped=13.0
+2023-03-10 02:39:08,042 INFO [train.py:968] (0/2) Epoch 19, batch 35550, giga_loss[loss=0.221, simple_loss=0.2796, pruned_loss=0.08122, over 23787.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3071, pruned_loss=0.08071, over 5702010.35 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3471, pruned_loss=0.1112, over 5698763.94 frames. ], giga_tot_loss[loss=0.2278, simple_loss=0.302, pruned_loss=0.07679, over 5693330.56 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:39:09,998 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=858140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:39:12,842 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:39:39,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=858172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:39:53,123 INFO [train.py:968] (0/2) Epoch 19, batch 35600, giga_loss[loss=0.2492, simple_loss=0.3101, pruned_loss=0.09413, over 26661.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3029, pruned_loss=0.07828, over 5702390.90 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3472, pruned_loss=0.1111, over 5701656.40 frames. ], giga_tot_loss[loss=0.2239, simple_loss=0.2981, pruned_loss=0.07483, over 5693078.53 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:39:56,955 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=858192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:40:07,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.268e+02 1.021e+03 1.274e+03 1.703e+03 4.094e+03, threshold=2.548e+03, percent-clipped=7.0
+2023-03-10 02:40:27,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4667, 2.4857, 1.9603, 2.1346], device='cuda:0'), covar=tensor([0.0872, 0.0717, 0.0998, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0441, 0.0509, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 02:40:38,938 INFO [train.py:968] (0/2) Epoch 19, batch 35650, giga_loss[loss=0.2106, simple_loss=0.2844, pruned_loss=0.06844, over 28749.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3002, pruned_loss=0.07685, over 5698858.51 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3481, pruned_loss=0.1115, over 5704210.41 frames. ], giga_tot_loss[loss=0.2207, simple_loss=0.2949, pruned_loss=0.07321, over 5688988.62 frames. ], batch size: 92, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:40:52,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=858255.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:41:17,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=858282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:41:21,863 INFO [train.py:968] (0/2) Epoch 19, batch 35700, giga_loss[loss=0.1814, simple_loss=0.2613, pruned_loss=0.05074, over 28896.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.299, pruned_loss=0.07673, over 5695959.63 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3492, pruned_loss=0.112, over 5695274.71 frames. ], giga_tot_loss[loss=0.2185, simple_loss=0.2924, pruned_loss=0.07232, over 5695391.71 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:41:39,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.283e+02 1.052e+03 1.461e+03 1.962e+03 6.498e+03, threshold=2.922e+03, percent-clipped=19.0
+2023-03-10 02:42:04,225 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=858332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:42:05,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3845, 2.0275, 1.6226, 0.6281], device='cuda:0'), covar=tensor([0.5180, 0.3013, 0.4541, 0.6137], device='cuda:0'), in_proj_covar=tensor([0.1710, 0.1614, 0.1576, 0.1394], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 02:42:09,318 INFO [train.py:968] (0/2) Epoch 19, batch 35750, giga_loss[loss=0.2585, simple_loss=0.3272, pruned_loss=0.09489, over 28836.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.299, pruned_loss=0.07735, over 5687170.31 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3492, pruned_loss=0.112, over 5695493.34 frames. ], giga_tot_loss[loss=0.2206, simple_loss=0.2936, pruned_loss=0.07377, over 5686524.64 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:42:53,565 INFO [train.py:968] (0/2) Epoch 19, batch 35800, giga_loss[loss=0.2196, simple_loss=0.2981, pruned_loss=0.07056, over 28814.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3101, pruned_loss=0.0832, over 5693217.71 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.349, pruned_loss=0.1119, over 5702999.99 frames. ], giga_tot_loss[loss=0.2317, simple_loss=0.3044, pruned_loss=0.07944, over 5685882.55 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:43:05,300 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=858398.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:43:07,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858401.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:43:11,078 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.462e+02 1.248e+03 1.587e+03 2.006e+03 9.066e+03, threshold=3.174e+03, percent-clipped=11.0
+2023-03-10 02:43:30,925 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=858425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:43:34,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:43:35,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=858430.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:43:43,235 INFO [train.py:968] (0/2) Epoch 19, batch 35850, giga_loss[loss=0.3159, simple_loss=0.3827, pruned_loss=0.1246, over 28553.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3237, pruned_loss=0.09029, over 5691897.39 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3491, pruned_loss=0.1118, over 5703432.92 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3189, pruned_loss=0.08713, over 5685621.98 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:44:00,752 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=858457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:44:28,192 INFO [train.py:968] (0/2) Epoch 19, batch 35900, libri_loss[loss=0.302, simple_loss=0.3475, pruned_loss=0.1282, over 29659.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3352, pruned_loss=0.09636, over 5679476.66 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3496, pruned_loss=0.1122, over 5689426.02 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3304, pruned_loss=0.09304, over 5687360.71 frames. ], batch size: 73, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:44:42,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.898e+02 1.310e+03 1.938e+03 2.559e+03 7.075e+03, threshold=3.876e+03, percent-clipped=16.0
+2023-03-10 02:45:10,791 INFO [train.py:968] (0/2) Epoch 19, batch 35950, giga_loss[loss=0.2526, simple_loss=0.3327, pruned_loss=0.08627, over 28813.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3392, pruned_loss=0.097, over 5686631.71 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3494, pruned_loss=0.112, over 5690742.93 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3354, pruned_loss=0.09432, over 5691648.41 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:45:29,791 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-10 02:45:35,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=858567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:45:55,605 INFO [train.py:968] (0/2) Epoch 19, batch 36000, giga_loss[loss=0.2839, simple_loss=0.3652, pruned_loss=0.1013, over 27898.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3418, pruned_loss=0.09704, over 5687860.62 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3496, pruned_loss=0.112, over 5694275.05 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3384, pruned_loss=0.09447, over 5688311.80 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:45:55,610 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 02:46:04,211 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2033, simple_loss=0.3095, pruned_loss=0.04854, over 944034.00 frames. 
+2023-03-10 02:46:04,212 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 02:46:05,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1961, 2.9415, 1.3524, 1.3763], device='cuda:0'), covar=tensor([0.1197, 0.0423, 0.1057, 0.1547], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0541, 0.0375, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-10 02:46:17,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 02:46:19,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.956e+02 1.255e+03 1.684e+03 2.191e+03 5.613e+03, threshold=3.368e+03, percent-clipped=4.0
+2023-03-10 02:46:50,117 INFO [train.py:968] (0/2) Epoch 19, batch 36050, libri_loss[loss=0.3253, simple_loss=0.3887, pruned_loss=0.131, over 25867.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3444, pruned_loss=0.09804, over 5686037.73 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3499, pruned_loss=0.1121, over 5694852.89 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3411, pruned_loss=0.09547, over 5685841.37 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:46:53,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3285, 1.3265, 3.7525, 3.0509], device='cuda:0'), covar=tensor([0.1668, 0.2762, 0.0458, 0.1050], device='cuda:0'), in_proj_covar=tensor([0.0732, 0.0631, 0.0928, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 02:47:02,839 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-10 02:47:31,292 INFO [train.py:968] (0/2) Epoch 19, batch 36100, giga_loss[loss=0.3171, simple_loss=0.3727, pruned_loss=0.1307, over 28870.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3465, pruned_loss=0.0996, over 5690023.22 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3503, pruned_loss=0.112, over 5699226.84 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3433, pruned_loss=0.0972, over 5685564.16 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:47:47,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.636e+02 1.233e+03 1.443e+03 1.894e+03 6.473e+03, threshold=2.885e+03, percent-clipped=4.0
+2023-03-10 02:47:48,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=858707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:47:50,493 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=858710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:47:52,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:48:14,911 INFO [train.py:968] (0/2) Epoch 19, batch 36150, giga_loss[loss=0.2516, simple_loss=0.3292, pruned_loss=0.08697, over 28453.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3486, pruned_loss=0.1012, over 5686970.53 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.35, pruned_loss=0.1118, over 5704227.72 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3464, pruned_loss=0.09919, over 5678475.07 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:48:18,163 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=858742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:48:20,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=858745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:48:41,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5622, 1.6380, 1.5621, 1.3370], device='cuda:0'), covar=tensor([0.2641, 0.2547, 0.1927, 0.2485], device='cuda:0'), in_proj_covar=tensor([0.1909, 0.1820, 0.1749, 0.1898], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 02:48:48,921 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=858778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:48:56,580 INFO [train.py:968] (0/2) Epoch 19, batch 36200, giga_loss[loss=0.2819, simple_loss=0.3725, pruned_loss=0.09564, over 28455.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3515, pruned_loss=0.1024, over 5683032.83 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3503, pruned_loss=0.112, over 5696475.40 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3495, pruned_loss=0.1005, over 5683451.06 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:49:07,452 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6657, 1.8324, 1.2425, 1.3839], device='cuda:0'), covar=tensor([0.1027, 0.0645, 0.1209, 0.1149], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0444, 0.0513, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 02:49:11,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.669e+02 1.239e+03 1.571e+03 2.009e+03 7.009e+03, threshold=3.143e+03, percent-clipped=6.0
+2023-03-10 02:49:20,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=858816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:49:20,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2801, 2.8287, 1.4437, 1.3655], device='cuda:0'), covar=tensor([0.0998, 0.0336, 0.0901, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0542, 0.0376, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 02:49:36,542 INFO [train.py:968] (0/2) Epoch 19, batch 36250, libri_loss[loss=0.259, simple_loss=0.3251, pruned_loss=0.09647, over 29377.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.356, pruned_loss=0.1045, over 5682200.39 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3508, pruned_loss=0.1122, over 5689853.02 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3541, pruned_loss=0.1025, over 5688153.62 frames. ], batch size: 67, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:49:46,301 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=858850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:49:48,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:50:13,102 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=858882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:50:16,995 INFO [train.py:968] (0/2) Epoch 19, batch 36300, giga_loss[loss=0.2798, simple_loss=0.3507, pruned_loss=0.1044, over 28584.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3572, pruned_loss=0.105, over 5677489.25 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3517, pruned_loss=0.1127, over 5691075.24 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.355, pruned_loss=0.1027, over 5681056.68 frames. ], batch size: 85, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:50:32,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.451e+02 1.315e+03 1.628e+03 2.177e+03 7.672e+03, threshold=3.255e+03, percent-clipped=14.0
+2023-03-10 02:50:46,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4146, 2.3707, 2.1987, 2.0060], device='cuda:0'), covar=tensor([0.1713, 0.2256, 0.2032, 0.2125], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0736, 0.0700, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 02:50:58,114 INFO [train.py:968] (0/2) Epoch 19, batch 36350, giga_loss[loss=0.2595, simple_loss=0.3405, pruned_loss=0.08923, over 28812.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3578, pruned_loss=0.1041, over 5685966.60 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3516, pruned_loss=0.1126, over 5694893.40 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3562, pruned_loss=0.1022, over 5685164.92 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:51:38,668 INFO [train.py:968] (0/2) Epoch 19, batch 36400, giga_loss[loss=0.2747, simple_loss=0.3567, pruned_loss=0.09637, over 28903.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3586, pruned_loss=0.1039, over 5700303.56 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3519, pruned_loss=0.1128, over 5702300.95 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3573, pruned_loss=0.1019, over 5692912.06 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:51:52,062 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.380e+02 1.139e+03 1.470e+03 2.261e+03 5.370e+03, threshold=2.941e+03, percent-clipped=5.0
+2023-03-10 02:52:15,912 INFO [train.py:968] (0/2) Epoch 19, batch 36450, giga_loss[loss=0.2573, simple_loss=0.3445, pruned_loss=0.08507, over 28629.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3576, pruned_loss=0.1026, over 5703911.05 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3528, pruned_loss=0.1135, over 5704146.70 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3559, pruned_loss=0.09991, over 5696248.94 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:52:33,512 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=859060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:52:59,554 INFO [train.py:968] (0/2) Epoch 19, batch 36500, giga_loss[loss=0.303, simple_loss=0.3788, pruned_loss=0.1136, over 28734.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3565, pruned_loss=0.1013, over 5712273.95 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3531, pruned_loss=0.1136, over 5707061.56 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.355, pruned_loss=0.09898, over 5703756.79 frames. ], batch size: 78, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:53:08,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3263, 2.2196, 2.0486, 1.8464], device='cuda:0'), covar=tensor([0.1833, 0.2391, 0.2343, 0.2342], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0740, 0.0703, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 02:53:14,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.040e+02 1.094e+03 1.311e+03 1.859e+03 3.588e+03, threshold=2.622e+03, percent-clipped=6.0
+2023-03-10 02:53:14,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3158, 4.1300, 3.9663, 1.7918], device='cuda:0'), covar=tensor([0.0681, 0.0844, 0.0911, 0.1996], device='cuda:0'), in_proj_covar=tensor([0.1162, 0.1079, 0.0918, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 02:53:23,006 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=859116.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:53:26,469 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:53:44,259 INFO [train.py:968] (0/2) Epoch 19, batch 36550, giga_loss[loss=0.3244, simple_loss=0.3827, pruned_loss=0.133, over 28679.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3588, pruned_loss=0.1046, over 5698816.99 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3535, pruned_loss=0.1139, over 5697016.28 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3573, pruned_loss=0.1023, over 5701290.34 frames. ], batch size: 284, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:53:47,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3791, 1.1329, 4.4175, 3.4281], device='cuda:0'), covar=tensor([0.1774, 0.3052, 0.0436, 0.1031], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0632, 0.0929, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 02:53:48,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9814, 3.7905, 3.6526, 1.4693], device='cuda:0'), covar=tensor([0.0796, 0.0970, 0.1035, 0.2243], device='cuda:0'), in_proj_covar=tensor([0.1164, 0.1081, 0.0920, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 02:53:57,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:54:27,957 INFO [train.py:968] (0/2) Epoch 19, batch 36600, giga_loss[loss=0.2961, simple_loss=0.362, pruned_loss=0.1152, over 28572.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3616, pruned_loss=0.1091, over 5693010.15 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.354, pruned_loss=0.1142, over 5695964.68 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3601, pruned_loss=0.1068, over 5696062.83 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:54:30,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:54:44,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.836e+02 1.333e+03 1.684e+03 2.449e+03 4.708e+03, threshold=3.367e+03, percent-clipped=16.0
+2023-03-10 02:54:47,110 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 02:55:13,325 INFO [train.py:968] (0/2) Epoch 19, batch 36650, giga_loss[loss=0.2824, simple_loss=0.3508, pruned_loss=0.107, over 29012.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3608, pruned_loss=0.1096, over 5695951.93 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3545, pruned_loss=0.1145, over 5698216.70 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3592, pruned_loss=0.1076, over 5696356.20 frames. ], batch size: 128, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:55:28,659 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-10 02:55:32,746 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3627, 2.0772, 1.5995, 0.5903], device='cuda:0'), covar=tensor([0.5570, 0.2942, 0.4048, 0.6230], device='cuda:0'), in_proj_covar=tensor([0.1704, 0.1606, 0.1579, 0.1390], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 02:55:37,201 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:55:40,248 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:55:41,132 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-10 02:56:01,209 INFO [train.py:968] (0/2) Epoch 19, batch 36700, giga_loss[loss=0.3082, simple_loss=0.3748, pruned_loss=0.1208, over 28906.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3577, pruned_loss=0.1084, over 5700605.67 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3544, pruned_loss=0.1144, over 5700113.21 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3567, pruned_loss=0.1069, over 5699176.23 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:56:07,243 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:07,924 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:09,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:16,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.345e+02 1.274e+03 1.633e+03 2.552e+03 5.578e+03, threshold=3.267e+03, percent-clipped=15.0
+2023-03-10 02:56:33,340 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:38,987 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:42,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:42,642 INFO [train.py:968] (0/2) Epoch 19, batch 36750, giga_loss[loss=0.3064, simple_loss=0.3812, pruned_loss=0.1158, over 29005.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3561, pruned_loss=0.1077, over 5701744.79 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3542, pruned_loss=0.1142, over 5704753.04 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3555, pruned_loss=0.1065, over 5696493.85 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:57:06,133 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:57:16,952 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 02:57:26,154 INFO [train.py:968] (0/2) Epoch 19, batch 36800, giga_loss[loss=0.2751, simple_loss=0.3511, pruned_loss=0.0995, over 27579.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3542, pruned_loss=0.1056, over 5706971.83 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3542, pruned_loss=0.114, over 5709579.69 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3537, pruned_loss=0.1046, over 5698563.97 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:57:33,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4307, 1.5134, 1.5111, 1.3378], device='cuda:0'), covar=tensor([0.2789, 0.2463, 0.2144, 0.2489], device='cuda:0'), in_proj_covar=tensor([0.1903, 0.1817, 0.1742, 0.1898], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 02:57:44,431 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7613, 1.8436, 1.3301, 1.3705], device='cuda:0'), covar=tensor([0.0933, 0.0655, 0.1075, 0.1207], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0444, 0.0514, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 02:57:45,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.007e+02 1.246e+03 1.544e+03 2.422e+03 9.786e+03, threshold=3.089e+03, percent-clipped=9.0
+2023-03-10 02:58:01,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=859425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:58:09,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:58:11,342 INFO [train.py:968] (0/2) Epoch 19, batch 36850, libri_loss[loss=0.2682, simple_loss=0.3312, pruned_loss=0.1026, over 29589.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3515, pruned_loss=0.1033, over 5696930.54 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3543, pruned_loss=0.1139, over 5710357.32 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.351, pruned_loss=0.1023, over 5689175.09 frames. ], batch size: 74, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:58:57,503 INFO [train.py:968] (0/2) Epoch 19, batch 36900, giga_loss[loss=0.2608, simple_loss=0.3298, pruned_loss=0.09588, over 28578.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3474, pruned_loss=0.1009, over 5677864.89 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3548, pruned_loss=0.1142, over 5702866.38 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3465, pruned_loss=0.09958, over 5678424.44 frames. ], batch size: 92, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:59:00,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859491.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 02:59:11,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0124, 3.8687, 3.6374, 1.8632], device='cuda:0'), covar=tensor([0.0610, 0.0709, 0.0679, 0.2124], device='cuda:0'), in_proj_covar=tensor([0.1166, 0.1086, 0.0924, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 02:59:20,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.920e+02 1.095e+03 1.362e+03 1.995e+03 7.157e+03, threshold=2.724e+03, percent-clipped=6.0
+2023-03-10 02:59:50,322 INFO [train.py:968] (0/2) Epoch 19, batch 36950, giga_loss[loss=0.2278, simple_loss=0.297, pruned_loss=0.07928, over 27521.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3406, pruned_loss=0.09751, over 5660338.94 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3552, pruned_loss=0.1145, over 5694755.17 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3395, pruned_loss=0.09612, over 5667919.20 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:00:18,585 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-10 03:00:38,040 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:00:40,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:00:45,036 INFO [train.py:968] (0/2) Epoch 19, batch 37000, libri_loss[loss=0.2956, simple_loss=0.3591, pruned_loss=0.116, over 29538.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3356, pruned_loss=0.09522, over 5654554.86 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3551, pruned_loss=0.1144, over 5699991.43 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3343, pruned_loss=0.09376, over 5654674.50 frames. ], batch size: 81, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:01:01,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4512, 1.6865, 1.0554, 1.2573], device='cuda:0'), covar=tensor([0.1290, 0.0890, 0.1688, 0.1526], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0444, 0.0514, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 03:01:05,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.607e+02 9.419e+02 1.179e+03 1.686e+03 3.845e+03, threshold=2.358e+03, percent-clipped=6.0
+2023-03-10 03:01:06,951 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:01:27,042 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859634.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 03:01:30,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859637.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 03:01:30,817 INFO [train.py:968] (0/2) Epoch 19, batch 37050, giga_loss[loss=0.2409, simple_loss=0.3205, pruned_loss=0.08059, over 28732.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3351, pruned_loss=0.09461, over 5661981.89 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3549, pruned_loss=0.1141, over 5701505.24 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3337, pruned_loss=0.09303, over 5658996.83 frames. ], batch size: 66, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:01:54,589 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859666.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 03:02:12,053 INFO [train.py:968] (0/2) Epoch 19, batch 37100, giga_loss[loss=0.2142, simple_loss=0.2949, pruned_loss=0.06679, over 28420.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3355, pruned_loss=0.094, over 5670223.18 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3554, pruned_loss=0.1143, over 5702220.89 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3334, pruned_loss=0.09216, over 5666673.46 frames. ], batch size: 65, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:02:28,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.048e+02 1.061e+03 1.445e+03 1.907e+03 5.890e+03, threshold=2.890e+03, percent-clipped=12.0
+2023-03-10 03:02:55,995 INFO [train.py:968] (0/2) Epoch 19, batch 37150, giga_loss[loss=0.248, simple_loss=0.3177, pruned_loss=0.08912, over 28787.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3353, pruned_loss=0.09391, over 5684201.90 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3559, pruned_loss=0.1144, over 5705362.00 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3329, pruned_loss=0.09203, over 5678318.52 frames. ], batch size: 119, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:03:08,645 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=859752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:03:36,851 INFO [train.py:968] (0/2) Epoch 19, batch 37200, giga_loss[loss=0.2482, simple_loss=0.323, pruned_loss=0.08669, over 28659.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3333, pruned_loss=0.09288, over 5697369.50 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3567, pruned_loss=0.1147, over 5706351.08 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3301, pruned_loss=0.09055, over 5691559.71 frames. ], batch size: 92, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:03:47,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:03:53,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.496e+02 1.166e+03 1.521e+03 2.269e+03 1.096e+04, threshold=3.042e+03, percent-clipped=11.0
+2023-03-10 03:04:16,950 INFO [train.py:968] (0/2) Epoch 19, batch 37250, libri_loss[loss=0.2696, simple_loss=0.3407, pruned_loss=0.09927, over 29711.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3308, pruned_loss=0.09168, over 5706115.55 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3571, pruned_loss=0.1148, over 5709235.70 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3275, pruned_loss=0.08945, over 5698817.04 frames. ], batch size: 69, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:04:59,010 INFO [train.py:968] (0/2) Epoch 19, batch 37300, giga_loss[loss=0.2267, simple_loss=0.3035, pruned_loss=0.07492, over 28691.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3288, pruned_loss=0.09098, over 5703881.05 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.358, pruned_loss=0.1153, over 5706796.09 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3251, pruned_loss=0.08851, over 5700487.52 frames. ], batch size: 119, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:05:15,313 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.917e+02 1.016e+03 1.235e+03 1.636e+03 8.132e+03, threshold=2.471e+03, percent-clipped=7.0
+2023-03-10 03:05:39,668 INFO [train.py:968] (0/2) Epoch 19, batch 37350, giga_loss[loss=0.2332, simple_loss=0.3139, pruned_loss=0.0763, over 28613.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3267, pruned_loss=0.08989, over 5708739.32 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3585, pruned_loss=0.1154, over 5707387.59 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3227, pruned_loss=0.08731, over 5705645.49 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:05:43,289 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:05:45,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:06:07,468 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:06:20,506 INFO [train.py:968] (0/2) Epoch 19, batch 37400, giga_loss[loss=0.2333, simple_loss=0.3107, pruned_loss=0.07795, over 28509.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3236, pruned_loss=0.08812, over 5714792.85 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3581, pruned_loss=0.115, over 5709482.06 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3201, pruned_loss=0.08594, over 5710607.71 frames. ], batch size: 78, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:06:30,091 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-860000.pt
+2023-03-10 03:06:36,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.700e+02 1.151e+03 1.505e+03 2.226e+03 1.283e+04, threshold=3.011e+03, percent-clipped=22.0
+2023-03-10 03:07:00,826 INFO [train.py:968] (0/2) Epoch 19, batch 37450, giga_loss[loss=0.2589, simple_loss=0.326, pruned_loss=0.09587, over 28545.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3243, pruned_loss=0.08904, over 5721493.81 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3591, pruned_loss=0.1155, over 5716692.98 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3193, pruned_loss=0.08578, over 5711584.77 frames. ], batch size: 60, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:07:04,126 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-10 03:07:13,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1430, 3.9860, 3.7715, 1.8749], device='cuda:0'), covar=tensor([0.0598, 0.0734, 0.0660, 0.2152], device='cuda:0'), in_proj_covar=tensor([0.1166, 0.1084, 0.0923, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 03:07:25,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7365, 1.8504, 1.9440, 1.5239], device='cuda:0'), covar=tensor([0.1825, 0.2339, 0.1413, 0.1609], device='cuda:0'), in_proj_covar=tensor([0.0889, 0.0697, 0.0934, 0.0832], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 03:07:40,684 INFO [train.py:968] (0/2) Epoch 19, batch 37500, giga_loss[loss=0.3336, simple_loss=0.3788, pruned_loss=0.1442, over 26576.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3228, pruned_loss=0.08824, over 5722023.56 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3596, pruned_loss=0.1157, over 5718645.74 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3177, pruned_loss=0.08498, over 5712553.22 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:07:58,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.540e+02 1.157e+03 1.350e+03 2.123e+03 6.591e+03, threshold=2.700e+03, percent-clipped=14.0
+2023-03-10 03:07:59,972 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=860110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:08:13,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=860127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:08:22,089 INFO [train.py:968] (0/2) Epoch 19, batch 37550, giga_loss[loss=0.2875, simple_loss=0.3532, pruned_loss=0.1109, over 27538.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3206, pruned_loss=0.08669, over 5719930.61 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3595, pruned_loss=0.1156, over 5721445.59 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.3161, pruned_loss=0.08389, over 5710094.30 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:09:04,930 INFO [train.py:968] (0/2) Epoch 19, batch 37600, libri_loss[loss=0.3261, simple_loss=0.3965, pruned_loss=0.1278, over 28561.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3222, pruned_loss=0.08778, over 5713438.96 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3605, pruned_loss=0.1159, over 5715436.40 frames. ], giga_tot_loss[loss=0.242, simple_loss=0.316, pruned_loss=0.08399, over 5710762.74 frames. ], batch size: 106, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:09:12,033 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=860198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:09:20,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.834e+02 1.033e+03 1.255e+03 1.814e+03 7.555e+03, threshold=2.511e+03, percent-clipped=7.0
+2023-03-10 03:09:41,090 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-10 03:09:47,284 INFO [train.py:968] (0/2) Epoch 19, batch 37650, giga_loss[loss=0.2328, simple_loss=0.311, pruned_loss=0.0773, over 28916.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3262, pruned_loss=0.08999, over 5714752.59 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3607, pruned_loss=0.1158, over 5714893.15 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3204, pruned_loss=0.08651, over 5713346.76 frames. ], batch size: 66, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:10:16,198 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=860270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:10:18,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=860273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:10:35,215 INFO [train.py:968] (0/2) Epoch 19, batch 37700, giga_loss[loss=0.3602, simple_loss=0.4156, pruned_loss=0.1524, over 27659.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3317, pruned_loss=0.09348, over 5711946.98 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3607, pruned_loss=0.1157, over 5718039.04 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3266, pruned_loss=0.09045, over 5708063.66 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:10:49,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=860302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:10:55,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.004e+02 1.156e+03 1.345e+03 1.890e+03 6.206e+03, threshold=2.689e+03, percent-clipped=11.0
+2023-03-10 03:11:25,251 INFO [train.py:968] (0/2) Epoch 19, batch 37750, giga_loss[loss=0.2852, simple_loss=0.3617, pruned_loss=0.1043, over 28906.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3412, pruned_loss=0.09988, over 5701577.27 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.361, pruned_loss=0.1157, over 5719144.95 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3367, pruned_loss=0.09727, over 5697593.78 frames. ], batch size: 106, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:12:16,618 INFO [train.py:968] (0/2) Epoch 19, batch 37800, giga_loss[loss=0.28, simple_loss=0.3522, pruned_loss=0.1039, over 28645.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3466, pruned_loss=0.1026, over 5691565.80 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3613, pruned_loss=0.1159, over 5721349.60 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3425, pruned_loss=0.1002, over 5685898.43 frames. ], batch size: 78, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:12:23,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9174, 1.2231, 4.9464, 3.5501], device='cuda:0'), covar=tensor([0.1508, 0.2981, 0.0374, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0733, 0.0630, 0.0922, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 03:12:34,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.794e+02 1.244e+03 1.513e+03 2.660e+03 5.459e+03, threshold=3.025e+03, percent-clipped=24.0
+2023-03-10 03:12:59,661 INFO [train.py:968] (0/2) Epoch 19, batch 37850, giga_loss[loss=0.3374, simple_loss=0.4054, pruned_loss=0.1347, over 28959.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3513, pruned_loss=0.1042, over 5686225.11 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3615, pruned_loss=0.1159, over 5710758.91 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3474, pruned_loss=0.1019, over 5690173.36 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:13:03,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6573, 1.7152, 1.8756, 1.4488], device='cuda:0'), covar=tensor([0.1771, 0.2491, 0.1443, 0.1727], device='cuda:0'), in_proj_covar=tensor([0.0887, 0.0697, 0.0933, 0.0829], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 03:13:41,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=860485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:13:44,758 INFO [train.py:968] (0/2) Epoch 19, batch 37900, giga_loss[loss=0.3406, simple_loss=0.4011, pruned_loss=0.14, over 28172.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3566, pruned_loss=0.1076, over 5676675.48 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3619, pruned_loss=0.1163, over 5709590.09 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3527, pruned_loss=0.1049, over 5679113.01 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:13:52,984 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=860498.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 03:13:55,725 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=860502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:14:03,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.139e+02 1.228e+03 1.624e+03 2.000e+03 4.237e+03, threshold=3.248e+03, percent-clipped=7.0
+2023-03-10 03:14:25,035 INFO [train.py:968] (0/2) Epoch 19, batch 37950, giga_loss[loss=0.2669, simple_loss=0.3442, pruned_loss=0.09479, over 28571.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3585, pruned_loss=0.1082, over 5689881.46 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3616, pruned_loss=0.116, over 5715661.37 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3556, pruned_loss=0.106, over 5685530.14 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 03:14:53,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=860573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:15:06,590 INFO [train.py:968] (0/2) Epoch 19, batch 38000, giga_loss[loss=0.321, simple_loss=0.3828, pruned_loss=0.1296, over 28608.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3556, pruned_loss=0.1059, over 5684073.24 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3619, pruned_loss=0.1163, over 5708795.74 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3529, pruned_loss=0.1038, over 5686751.60 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:15:24,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.820e+02 1.274e+03 1.616e+03 2.480e+03 8.965e+03, threshold=3.232e+03, percent-clipped=16.0
+2023-03-10 03:15:38,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=860628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:15:41,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=860631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:15:46,236 INFO [train.py:968] (0/2) Epoch 19, batch 38050, giga_loss[loss=0.2695, simple_loss=0.3467, pruned_loss=0.09615, over 28843.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3539, pruned_loss=0.1042, over 5693452.83 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3627, pruned_loss=0.1167, over 5712956.46 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3508, pruned_loss=0.1016, over 5691119.03 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:16:06,368 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=860660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:16:28,191 INFO [train.py:968] (0/2) Epoch 19, batch 38100, giga_loss[loss=0.2909, simple_loss=0.3572, pruned_loss=0.1123, over 28139.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3531, pruned_loss=0.1033, over 5686244.95 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3626, pruned_loss=0.1168, over 5701821.56 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3503, pruned_loss=0.1008, over 5693751.20 frames. ], batch size: 77, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:16:37,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=860698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:16:49,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.465e+02 1.165e+03 1.506e+03 1.993e+03 5.431e+03, threshold=3.012e+03, percent-clipped=6.0
+2023-03-10 03:16:54,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=860716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:16:56,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=860719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:17:13,203 INFO [train.py:968] (0/2) Epoch 19, batch 38150, giga_loss[loss=0.2662, simple_loss=0.3438, pruned_loss=0.09427, over 28858.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3535, pruned_loss=0.1029, over 5689050.98 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3627, pruned_loss=0.1168, over 5702491.45 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3511, pruned_loss=0.1008, over 5694271.73 frames. ], batch size: 119, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:17:23,310 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=860748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:17:47,995 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-10 03:17:55,951 INFO [train.py:968] (0/2) Epoch 19, batch 38200, giga_loss[loss=0.3058, simple_loss=0.3814, pruned_loss=0.1151, over 28695.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3556, pruned_loss=0.1044, over 5696218.16 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3631, pruned_loss=0.1171, over 5706707.79 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3531, pruned_loss=0.102, over 5696141.20 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:18:18,344 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.973e+02 1.380e+03 1.882e+03 2.626e+03 1.732e+04, threshold=3.764e+03, percent-clipped=17.0
+2023-03-10 03:18:43,035 INFO [train.py:968] (0/2) Epoch 19, batch 38250, giga_loss[loss=0.294, simple_loss=0.3602, pruned_loss=0.1139, over 27611.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3581, pruned_loss=0.1064, over 5697206.06 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3636, pruned_loss=0.1172, over 5709829.20 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3556, pruned_loss=0.1041, over 5694189.40 frames. ], batch size: 474, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:18:46,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5087, 3.6008, 1.7102, 1.6091], device='cuda:0'), covar=tensor([0.1020, 0.0296, 0.0841, 0.1368], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0538, 0.0372, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-10 03:19:04,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8042, 1.1047, 2.8769, 2.6947], device='cuda:0'), covar=tensor([0.1689, 0.2540, 0.0575, 0.0930], device='cuda:0'), in_proj_covar=tensor([0.0736, 0.0630, 0.0927, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 03:19:16,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=860873.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 03:19:19,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=860877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:19:29,841 INFO [train.py:968] (0/2) Epoch 19, batch 38300, giga_loss[loss=0.3205, simple_loss=0.379, pruned_loss=0.131, over 28769.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3587, pruned_loss=0.1075, over 5695363.58 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3635, pruned_loss=0.1172, over 5710956.05 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3568, pruned_loss=0.1057, over 5691979.17 frames. ], batch size: 99, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:19:51,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.926e+02 1.275e+03 1.582e+03 2.099e+03 6.251e+03, threshold=3.164e+03, percent-clipped=2.0
+2023-03-10 03:20:11,648 INFO [train.py:968] (0/2) Epoch 19, batch 38350, giga_loss[loss=0.2452, simple_loss=0.3279, pruned_loss=0.08127, over 28482.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3594, pruned_loss=0.1081, over 5705010.66 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3638, pruned_loss=0.117, over 5718192.50 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3574, pruned_loss=0.1065, over 5695216.75 frames. ], batch size: 60, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:20:53,217 INFO [train.py:968] (0/2) Epoch 19, batch 38400, giga_loss[loss=0.319, simple_loss=0.3842, pruned_loss=0.1269, over 27922.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3604, pruned_loss=0.109, over 5692612.24 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.364, pruned_loss=0.117, over 5712069.67 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3585, pruned_loss=0.1075, over 5690272.62 frames. ], batch size: 412, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:21:13,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.092e+02 1.223e+03 1.717e+03 2.762e+03 8.971e+03, threshold=3.434e+03, percent-clipped=21.0
+2023-03-10 03:21:17,049 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 03:21:20,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861019.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 03:21:21,550 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:21:23,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:21:33,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4802, 1.6229, 1.5734, 1.4610], device='cuda:0'), covar=tensor([0.1903, 0.2271, 0.2438, 0.2217], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0746, 0.0711, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 03:21:33,421 INFO [train.py:968] (0/2) Epoch 19, batch 38450, giga_loss[loss=0.2815, simple_loss=0.3611, pruned_loss=0.1009, over 28991.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3597, pruned_loss=0.1076, over 5694845.45 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3641, pruned_loss=0.1169, over 5706676.94 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3581, pruned_loss=0.1062, over 5696728.55 frames. ], batch size: 186, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:21:42,083 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=861048.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 03:21:43,634 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 03:21:44,630 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=861052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:22:02,924 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=861073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:22:02,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8237, 2.6188, 1.5661, 1.0383], device='cuda:0'), covar=tensor([0.7458, 0.3092, 0.4273, 0.6257], device='cuda:0'), in_proj_covar=tensor([0.1689, 0.1590, 0.1562, 0.1380], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 03:22:16,651 INFO [train.py:968] (0/2) Epoch 19, batch 38500, giga_loss[loss=0.2833, simple_loss=0.3609, pruned_loss=0.1028, over 28914.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3602, pruned_loss=0.107, over 5701398.88 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3648, pruned_loss=0.1174, over 5710098.88 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3581, pruned_loss=0.1053, over 5699379.95 frames. ], batch size: 213, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:22:20,693 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:22:22,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5499, 1.6562, 1.6234, 1.3750], device='cuda:0'), covar=tensor([0.2506, 0.2496, 0.1927, 0.2428], device='cuda:0'), in_proj_covar=tensor([0.1915, 0.1833, 0.1761, 0.1914], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 03:22:26,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1301, 1.2078, 1.1235, 0.8436], device='cuda:0'), covar=tensor([0.1071, 0.0602, 0.1120, 0.1155], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0441, 0.0511, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 03:22:35,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.411e+02 1.126e+03 1.362e+03 2.171e+03 9.129e+03, threshold=2.724e+03, percent-clipped=5.0
+2023-03-10 03:22:56,214 INFO [train.py:968] (0/2) Epoch 19, batch 38550, giga_loss[loss=0.2933, simple_loss=0.3606, pruned_loss=0.113, over 28848.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3594, pruned_loss=0.1062, over 5703790.24 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3648, pruned_loss=0.1173, over 5710478.07 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3577, pruned_loss=0.1047, over 5701764.08 frames. ], batch size: 106, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:23:42,663 INFO [train.py:968] (0/2) Epoch 19, batch 38600, giga_loss[loss=0.2678, simple_loss=0.324, pruned_loss=0.1058, over 23559.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3568, pruned_loss=0.1048, over 5690334.15 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.365, pruned_loss=0.1173, over 5702528.29 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3552, pruned_loss=0.1034, over 5695842.39 frames. ], batch size: 705, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:24:00,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.776e+02 1.044e+03 1.301e+03 1.939e+03 4.312e+03, threshold=2.603e+03, percent-clipped=10.0
+2023-03-10 03:24:02,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:24:05,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:24:22,403 INFO [train.py:968] (0/2) Epoch 19, batch 38650, giga_loss[loss=0.3159, simple_loss=0.3808, pruned_loss=0.1255, over 28251.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3541, pruned_loss=0.1032, over 5700453.84 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3649, pruned_loss=0.1172, over 5708721.46 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3524, pruned_loss=0.1017, over 5699030.37 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:24:31,119 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=861248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:24:48,054 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 03:25:02,293 INFO [train.py:968] (0/2) Epoch 19, batch 38700, giga_loss[loss=0.2686, simple_loss=0.3483, pruned_loss=0.09445, over 28666.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3523, pruned_loss=0.1024, over 5706185.68 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3648, pruned_loss=0.1172, over 5712535.63 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3508, pruned_loss=0.1009, over 5701597.52 frames. ], batch size: 99, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:25:24,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.777e+02 1.153e+03 1.404e+03 1.921e+03 1.221e+04, threshold=2.809e+03, percent-clipped=11.0
+2023-03-10 03:25:44,855 INFO [train.py:968] (0/2) Epoch 19, batch 38750, giga_loss[loss=0.3012, simple_loss=0.3736, pruned_loss=0.1143, over 28624.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3528, pruned_loss=0.1029, over 5709908.16 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3646, pruned_loss=0.117, over 5717388.79 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3516, pruned_loss=0.1016, over 5701818.48 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:25:52,812 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-10 03:26:26,504 INFO [train.py:968] (0/2) Epoch 19, batch 38800, giga_loss[loss=0.2673, simple_loss=0.3526, pruned_loss=0.09096, over 28914.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3529, pruned_loss=0.1026, over 5712317.74 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3645, pruned_loss=0.1169, over 5721470.27 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3517, pruned_loss=0.1014, over 5702269.18 frames. ], batch size: 213, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:26:31,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2719, 3.0743, 2.9242, 1.2156], device='cuda:0'), covar=tensor([0.0900, 0.1107, 0.0976, 0.2423], device='cuda:0'), in_proj_covar=tensor([0.1166, 0.1088, 0.0922, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 03:26:44,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.169e+02 1.051e+03 1.283e+03 1.826e+03 1.230e+04, threshold=2.565e+03, percent-clipped=8.0
+2023-03-10 03:26:46,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2645, 2.6207, 2.3328, 1.8387], device='cuda:0'), covar=tensor([0.2820, 0.1992, 0.2082, 0.2966], device='cuda:0'), in_proj_covar=tensor([0.1919, 0.1838, 0.1768, 0.1920], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 03:26:50,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:27:06,234 INFO [train.py:968] (0/2) Epoch 19, batch 38850, giga_loss[loss=0.3235, simple_loss=0.3847, pruned_loss=0.1311, over 28399.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3519, pruned_loss=0.1011, over 5713714.68 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3645, pruned_loss=0.1169, over 5721470.27 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.351, pruned_loss=0.1002, over 5705893.76 frames. ], batch size: 65, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:27:22,995 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:27:27,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=861468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:27:44,352 INFO [train.py:968] (0/2) Epoch 19, batch 38900, giga_loss[loss=0.3062, simple_loss=0.3796, pruned_loss=0.1164, over 28299.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3511, pruned_loss=0.1002, over 5711248.54 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.365, pruned_loss=0.1171, over 5715669.82 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3496, pruned_loss=0.09885, over 5709636.85 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:28:06,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.845e+02 1.045e+03 1.242e+03 1.591e+03 7.929e+03, threshold=2.483e+03, percent-clipped=10.0
+2023-03-10 03:28:23,575 INFO [train.py:968] (0/2) Epoch 19, batch 38950, giga_loss[loss=0.2745, simple_loss=0.3494, pruned_loss=0.0998, over 28892.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3507, pruned_loss=0.1008, over 5710282.57 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3651, pruned_loss=0.1173, over 5719759.65 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3491, pruned_loss=0.09913, over 5704890.74 frames. ], batch size: 227, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:28:27,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6014, 2.5168, 2.5688, 2.2492], device='cuda:0'), covar=tensor([0.1856, 0.2526, 0.2052, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.0463, 0.0739, 0.0703, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 03:29:05,163 INFO [train.py:968] (0/2) Epoch 19, batch 39000, giga_loss[loss=0.2371, simple_loss=0.3197, pruned_loss=0.07727, over 28345.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3481, pruned_loss=0.0997, over 5704733.71 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3652, pruned_loss=0.1174, over 5721674.49 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3464, pruned_loss=0.09804, over 5698462.63 frames. ], batch size: 71, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:29:05,167 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 03:29:14,768 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2074, simple_loss=0.3154, pruned_loss=0.04973, over 944034.00 frames. 
+2023-03-10 03:29:14,768 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 03:29:34,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:29:35,717 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.186e+02 1.195e+03 1.425e+03 1.856e+03 5.410e+03, threshold=2.849e+03, percent-clipped=13.0
+2023-03-10 03:29:36,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:29:37,342 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:29:46,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-10 03:29:52,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:29:54,596 INFO [train.py:968] (0/2) Epoch 19, batch 39050, giga_loss[loss=0.2422, simple_loss=0.3161, pruned_loss=0.08414, over 28505.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3453, pruned_loss=0.09871, over 5698046.10 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3658, pruned_loss=0.1178, over 5713176.99 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3432, pruned_loss=0.09682, over 5700825.51 frames. ], batch size: 78, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:30:00,194 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=861643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:30:18,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6070, 1.8344, 1.1909, 1.5579], device='cuda:0'), covar=tensor([0.0958, 0.0704, 0.1089, 0.1244], device='cuda:0'), in_proj_covar=tensor([0.0385, 0.0441, 0.0514, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 03:30:29,695 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:30:32,610 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:30:36,406 INFO [train.py:968] (0/2) Epoch 19, batch 39100, giga_loss[loss=0.2688, simple_loss=0.3417, pruned_loss=0.09796, over 29025.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3434, pruned_loss=0.09772, over 5702050.16 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.366, pruned_loss=0.1179, over 5715190.89 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3414, pruned_loss=0.09597, over 5702344.83 frames. ], batch size: 128, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:31:00,263 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.512e+02 1.155e+03 1.456e+03 1.884e+03 4.271e+03, threshold=2.913e+03, percent-clipped=9.0
+2023-03-10 03:31:19,672 INFO [train.py:968] (0/2) Epoch 19, batch 39150, libri_loss[loss=0.2769, simple_loss=0.3337, pruned_loss=0.1101, over 28586.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.342, pruned_loss=0.09719, over 5711656.38 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.366, pruned_loss=0.1179, over 5718099.98 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.34, pruned_loss=0.09536, over 5709097.50 frames. ], batch size: 63, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:31:48,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3052, 1.9677, 1.4508, 0.5957], device='cuda:0'), covar=tensor([0.5183, 0.2740, 0.4450, 0.5882], device='cuda:0'), in_proj_covar=tensor([0.1699, 0.1589, 0.1573, 0.1382], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 03:31:58,833 INFO [train.py:968] (0/2) Epoch 19, batch 39200, giga_loss[loss=0.288, simple_loss=0.3444, pruned_loss=0.1158, over 28858.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3406, pruned_loss=0.0972, over 5710194.37 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3658, pruned_loss=0.1177, over 5714282.67 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3385, pruned_loss=0.09533, over 5711065.70 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:32:04,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=861795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:32:19,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.376e+02 1.112e+03 1.565e+03 2.451e+03 9.839e+03, threshold=3.129e+03, percent-clipped=10.0
+2023-03-10 03:32:37,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=861837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:32:38,166 INFO [train.py:968] (0/2) Epoch 19, batch 39250, giga_loss[loss=0.2161, simple_loss=0.3032, pruned_loss=0.06451, over 29097.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3387, pruned_loss=0.09673, over 5686788.60 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3658, pruned_loss=0.1179, over 5692623.43 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3362, pruned_loss=0.09441, over 5707269.01 frames. ], batch size: 155, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:32:57,787 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:33:17,364 INFO [train.py:968] (0/2) Epoch 19, batch 39300, giga_loss[loss=0.232, simple_loss=0.3039, pruned_loss=0.08002, over 28974.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3344, pruned_loss=0.09418, over 5696360.73 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3654, pruned_loss=0.1175, over 5698055.18 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.332, pruned_loss=0.09211, over 5707519.19 frames. ], batch size: 106, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:33:18,498 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-10 03:33:24,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2431, 1.8947, 1.3801, 0.5195], device='cuda:0'), covar=tensor([0.3311, 0.2526, 0.3545, 0.4721], device='cuda:0'), in_proj_covar=tensor([0.1705, 0.1599, 0.1581, 0.1391], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 03:33:37,983 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:33:40,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.512e+02 1.089e+03 1.538e+03 2.131e+03 6.345e+03, threshold=3.076e+03, percent-clipped=9.0
+2023-03-10 03:33:49,200 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-10 03:33:59,843 INFO [train.py:968] (0/2) Epoch 19, batch 39350, libri_loss[loss=0.311, simple_loss=0.3834, pruned_loss=0.1193, over 28684.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3342, pruned_loss=0.09443, over 5698044.35 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3655, pruned_loss=0.1175, over 5701363.87 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.331, pruned_loss=0.09182, over 5704155.31 frames. ], batch size: 106, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:34:00,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:02,399 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:15,372 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:17,317 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:26,627 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=861970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:35,537 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:38,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:42,918 INFO [train.py:968] (0/2) Epoch 19, batch 39400, giga_loss[loss=0.292, simple_loss=0.3678, pruned_loss=0.1081, over 28781.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3357, pruned_loss=0.09494, over 5701039.00 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3655, pruned_loss=0.1174, over 5701836.80 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3326, pruned_loss=0.09263, over 5705456.72 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:34:46,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=861990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:54,236 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-862000.pt
+2023-03-10 03:35:05,191 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:35:06,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:35:08,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.406e+02 1.145e+03 1.469e+03 1.935e+03 9.138e+03, threshold=2.938e+03, percent-clipped=10.0
+2023-03-10 03:35:26,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4269, 1.9608, 1.4557, 0.7264], device='cuda:0'), covar=tensor([0.6559, 0.3064, 0.3825, 0.6760], device='cuda:0'), in_proj_covar=tensor([0.1708, 0.1605, 0.1582, 0.1395], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 03:35:30,913 INFO [train.py:968] (0/2) Epoch 19, batch 39450, giga_loss[loss=0.2583, simple_loss=0.335, pruned_loss=0.09084, over 29067.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3378, pruned_loss=0.09502, over 5705417.38 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3656, pruned_loss=0.1175, over 5702331.78 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.335, pruned_loss=0.09286, over 5708376.38 frames. ], batch size: 128, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:35:45,481 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:35:49,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:35:53,822 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:36:13,120 INFO [train.py:968] (0/2) Epoch 19, batch 39500, giga_loss[loss=0.3219, simple_loss=0.3916, pruned_loss=0.1261, over 28545.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3418, pruned_loss=0.09708, over 5697354.28 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3661, pruned_loss=0.118, over 5698079.51 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3384, pruned_loss=0.09435, over 5703881.28 frames. ], batch size: 336, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:36:17,150 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:36:35,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.122e+02 9.940e+02 1.318e+03 1.720e+03 4.233e+03, threshold=2.636e+03, percent-clipped=6.0
+2023-03-10 03:36:54,120 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:36:56,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:36:57,890 INFO [train.py:968] (0/2) Epoch 19, batch 39550, giga_loss[loss=0.2656, simple_loss=0.3454, pruned_loss=0.09285, over 28399.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3435, pruned_loss=0.09708, over 5695046.63 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3666, pruned_loss=0.1182, over 5701416.37 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3397, pruned_loss=0.09409, over 5697183.59 frames. ], batch size: 369, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:37:09,028 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:12,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:19,929 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:22,434 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:27,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3081, 1.6212, 1.3139, 1.4566], device='cuda:0'), covar=tensor([0.0719, 0.0306, 0.0329, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 03:37:35,966 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:37,766 INFO [train.py:968] (0/2) Epoch 19, batch 39600, giga_loss[loss=0.2402, simple_loss=0.328, pruned_loss=0.0762, over 28271.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3433, pruned_loss=0.09678, over 5696840.76 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3669, pruned_loss=0.1186, over 5710478.98 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.339, pruned_loss=0.09315, over 5690111.89 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:37:48,099 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:50,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:50,071 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:52,014 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:55,256 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7472, 1.8078, 1.2271, 1.3815], device='cuda:0'), covar=tensor([0.0909, 0.0733, 0.1156, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0440, 0.0511, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 03:37:58,457 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.897e+02 1.159e+03 1.366e+03 2.180e+03 7.323e+03, threshold=2.732e+03, percent-clipped=16.0
+2023-03-10 03:38:11,957 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:38:13,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:38:16,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:38:17,094 INFO [train.py:968] (0/2) Epoch 19, batch 39650, giga_loss[loss=0.2331, simple_loss=0.3188, pruned_loss=0.07369, over 29052.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3419, pruned_loss=0.09535, over 5704521.17 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.366, pruned_loss=0.1179, over 5711070.39 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3386, pruned_loss=0.09244, over 5698189.26 frames. ], batch size: 155, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:38:22,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2679, 0.7798, 0.7831, 1.4929], device='cuda:0'), covar=tensor([0.0714, 0.0364, 0.0362, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 03:38:32,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2790, 1.1568, 3.6212, 3.0765], device='cuda:0'), covar=tensor([0.1545, 0.2718, 0.0468, 0.0914], device='cuda:0'), in_proj_covar=tensor([0.0734, 0.0628, 0.0925, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 03:38:43,393 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:38:59,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:38:59,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1299, 1.1792, 1.0928, 0.7646], device='cuda:0'), covar=tensor([0.0960, 0.0557, 0.1062, 0.1193], device='cuda:0'), in_proj_covar=tensor([0.0383, 0.0439, 0.0509, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 03:39:02,901 INFO [train.py:968] (0/2) Epoch 19, batch 39700, giga_loss[loss=0.2774, simple_loss=0.3543, pruned_loss=0.1003, over 27673.00 frames. ], tot_loss[loss=0.268, simple_loss=0.343, pruned_loss=0.09655, over 5698484.50 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.366, pruned_loss=0.1178, over 5713699.95 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3401, pruned_loss=0.09403, over 5691144.87 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:39:25,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.601e+02 1.187e+03 1.506e+03 2.031e+03 5.977e+03, threshold=3.012e+03, percent-clipped=11.0
+2023-03-10 03:39:40,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:39:42,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:39:43,717 INFO [train.py:968] (0/2) Epoch 19, batch 39750, giga_loss[loss=0.3238, simple_loss=0.3833, pruned_loss=0.1321, over 28713.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3445, pruned_loss=0.09798, over 5700120.25 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3656, pruned_loss=0.1178, over 5719915.98 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3414, pruned_loss=0.09495, over 5687520.96 frames. ], batch size: 99, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:39:51,321 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 03:40:05,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2498, 1.9836, 1.5751, 0.4685], device='cuda:0'), covar=tensor([0.5429, 0.2665, 0.3954, 0.6222], device='cuda:0'), in_proj_covar=tensor([0.1706, 0.1597, 0.1579, 0.1390], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 03:40:20,080 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:40:20,169 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:40:24,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:40:28,401 INFO [train.py:968] (0/2) Epoch 19, batch 39800, giga_loss[loss=0.2651, simple_loss=0.3468, pruned_loss=0.09172, over 28900.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3467, pruned_loss=0.09865, over 5707893.94 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3657, pruned_loss=0.1178, over 5721074.40 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3441, pruned_loss=0.09624, over 5696911.86 frames. ], batch size: 186, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:40:48,905 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:40:52,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.879e+02 1.183e+03 1.531e+03 1.918e+03 6.513e+03, threshold=3.061e+03, percent-clipped=4.0
+2023-03-10 03:41:03,953 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:06,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:11,881 INFO [train.py:968] (0/2) Epoch 19, batch 39850, giga_loss[loss=0.297, simple_loss=0.3666, pruned_loss=0.1137, over 28901.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3492, pruned_loss=0.09974, over 5713449.57 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3659, pruned_loss=0.1178, over 5723620.33 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3468, pruned_loss=0.09756, over 5702450.37 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:41:12,083 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:23,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6843, 1.8638, 1.5163, 1.7735], device='cuda:0'), covar=tensor([0.2477, 0.2641, 0.3022, 0.2474], device='cuda:0'), in_proj_covar=tensor([0.1467, 0.1064, 0.1301, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 03:41:30,249 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:35,592 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:44,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:46,145 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:46,159 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:49,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:54,473 INFO [train.py:968] (0/2) Epoch 19, batch 39900, giga_loss[loss=0.2909, simple_loss=0.3658, pruned_loss=0.108, over 28970.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3505, pruned_loss=0.09983, over 5718587.84 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3659, pruned_loss=0.1177, over 5725566.40 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3483, pruned_loss=0.09779, over 5707985.97 frames. ], batch size: 213, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:42:11,206 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:42:13,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:42:15,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.635e+02 1.223e+03 1.558e+03 2.192e+03 5.838e+03, threshold=3.115e+03, percent-clipped=9.0
+2023-03-10 03:42:34,868 INFO [train.py:968] (0/2) Epoch 19, batch 39950, giga_loss[loss=0.2912, simple_loss=0.3706, pruned_loss=0.1058, over 28902.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3518, pruned_loss=0.1009, over 5714902.43 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3659, pruned_loss=0.1177, over 5720127.44 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3496, pruned_loss=0.09895, over 5711435.64 frames. ], batch size: 213, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:42:40,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:03,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1154, 1.4441, 0.9316, 1.1022], device='cuda:0'), covar=tensor([0.1160, 0.0677, 0.1612, 0.1284], device='cuda:0'), in_proj_covar=tensor([0.0384, 0.0443, 0.0512, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 03:43:11,329 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:13,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:16,612 INFO [train.py:968] (0/2) Epoch 19, batch 40000, giga_loss[loss=0.3318, simple_loss=0.4067, pruned_loss=0.1285, over 28699.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.352, pruned_loss=0.1013, over 5709504.38 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3658, pruned_loss=0.1175, over 5714301.68 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.35, pruned_loss=0.09953, over 5712215.82 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:43:33,928 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:36,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:37,205 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:38,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.528e+02 1.338e+03 1.735e+03 2.373e+03 7.302e+03, threshold=3.470e+03, percent-clipped=11.0
+2023-03-10 03:43:54,916 INFO [train.py:968] (0/2) Epoch 19, batch 40050, libri_loss[loss=0.2683, simple_loss=0.3427, pruned_loss=0.09689, over 29590.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3515, pruned_loss=0.1011, over 5709905.18 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3665, pruned_loss=0.1178, over 5715747.47 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3488, pruned_loss=0.09884, over 5710648.14 frames. ], batch size: 74, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:43:57,140 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:57,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:58,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6853, 1.7918, 1.7637, 1.5568], device='cuda:0'), covar=tensor([0.2825, 0.2295, 0.1878, 0.2420], device='cuda:0'), in_proj_covar=tensor([0.1931, 0.1854, 0.1782, 0.1918], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 03:44:36,582 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:44:36,906 INFO [train.py:968] (0/2) Epoch 19, batch 40100, giga_loss[loss=0.2464, simple_loss=0.3232, pruned_loss=0.08475, over 29035.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3485, pruned_loss=0.09977, over 5707388.06 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3661, pruned_loss=0.1176, over 5716358.70 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3465, pruned_loss=0.09792, over 5707440.35 frames. ], batch size: 136, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:44:38,574 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:45:00,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.054e+02 1.168e+03 1.417e+03 2.052e+03 6.069e+03, threshold=2.834e+03, percent-clipped=8.0
+2023-03-10 03:45:03,201 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:45:04,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5968, 1.9463, 1.6917, 1.7109], device='cuda:0'), covar=tensor([0.1816, 0.2136, 0.2190, 0.2231], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0738, 0.0706, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 03:45:19,920 INFO [train.py:968] (0/2) Epoch 19, batch 40150, giga_loss[loss=0.2938, simple_loss=0.3654, pruned_loss=0.1111, over 28265.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3451, pruned_loss=0.09803, over 5718720.23 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3658, pruned_loss=0.1175, over 5720870.03 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3432, pruned_loss=0.0961, over 5714771.81 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:45:33,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:45:56,986 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-10 03:45:57,428 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:45:59,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:46:00,453 INFO [train.py:968] (0/2) Epoch 19, batch 40200, giga_loss[loss=0.2479, simple_loss=0.3404, pruned_loss=0.0777, over 28628.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3444, pruned_loss=0.09756, over 5706027.08 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.366, pruned_loss=0.1178, over 5713626.71 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3421, pruned_loss=0.09524, over 5709276.83 frames. ], batch size: 307, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:46:22,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:46:22,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.104e+02 1.202e+03 1.707e+03 2.719e+03 1.245e+04, threshold=3.414e+03, percent-clipped=22.0
+2023-03-10 03:46:40,428 INFO [train.py:968] (0/2) Epoch 19, batch 40250, giga_loss[loss=0.2825, simple_loss=0.3496, pruned_loss=0.1077, over 23966.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3458, pruned_loss=0.09736, over 5693173.56 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.366, pruned_loss=0.1179, over 5701512.98 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3433, pruned_loss=0.09474, over 5706676.40 frames. ], batch size: 705, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:46:49,516 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.00 vs. limit=5.0
+2023-03-10 03:47:26,064 INFO [train.py:968] (0/2) Epoch 19, batch 40300, giga_loss[loss=0.2572, simple_loss=0.3374, pruned_loss=0.08846, over 28785.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3463, pruned_loss=0.09724, over 5694895.77 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3658, pruned_loss=0.1177, over 5703697.17 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3443, pruned_loss=0.09517, over 5703668.33 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:47:32,537 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:47:34,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:47:48,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.016e+02 1.191e+03 1.566e+03 2.144e+03 7.110e+03, threshold=3.132e+03, percent-clipped=9.0
+2023-03-10 03:47:57,407 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:48:05,528 INFO [train.py:968] (0/2) Epoch 19, batch 40350, giga_loss[loss=0.2714, simple_loss=0.3366, pruned_loss=0.1031, over 28804.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3452, pruned_loss=0.09749, over 5705308.44 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.366, pruned_loss=0.1179, over 5705168.86 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3431, pruned_loss=0.09537, over 5710930.48 frames. ], batch size: 119, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:48:06,769 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:48:29,928 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 03:48:34,435 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4749, 1.7644, 1.3974, 1.6888], device='cuda:0'), covar=tensor([0.0723, 0.0271, 0.0330, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 03:48:40,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6771, 4.9416, 1.9469, 1.9483], device='cuda:0'), covar=tensor([0.0948, 0.0354, 0.0835, 0.1225], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0540, 0.0372, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-10 03:48:45,422 INFO [train.py:968] (0/2) Epoch 19, batch 40400, giga_loss[loss=0.2755, simple_loss=0.3361, pruned_loss=0.1074, over 28563.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3439, pruned_loss=0.09812, over 5702212.73 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3661, pruned_loss=0.118, over 5696965.42 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3417, pruned_loss=0.09591, over 5714314.09 frames. ], batch size: 85, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:49:12,321 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.760e+02 1.154e+03 1.344e+03 1.733e+03 4.898e+03, threshold=2.688e+03, percent-clipped=1.0
+2023-03-10 03:49:31,420 INFO [train.py:968] (0/2) Epoch 19, batch 40450, giga_loss[loss=0.288, simple_loss=0.3453, pruned_loss=0.1154, over 24007.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3421, pruned_loss=0.09873, over 5687626.63 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3663, pruned_loss=0.1181, over 5686702.09 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.34, pruned_loss=0.09661, over 5707414.45 frames. ], batch size: 705, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:49:32,972 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-10 03:50:12,268 INFO [train.py:968] (0/2) Epoch 19, batch 40500, giga_loss[loss=0.2607, simple_loss=0.3401, pruned_loss=0.09065, over 28963.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3406, pruned_loss=0.09862, over 5687012.22 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3662, pruned_loss=0.1181, over 5685141.79 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3387, pruned_loss=0.09678, over 5704100.88 frames. ], batch size: 227, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:50:36,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.395e+02 1.037e+03 1.395e+03 1.937e+03 4.723e+03, threshold=2.790e+03, percent-clipped=6.0
+2023-03-10 03:50:52,524 INFO [train.py:968] (0/2) Epoch 19, batch 40550, giga_loss[loss=0.264, simple_loss=0.3291, pruned_loss=0.09947, over 29084.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3388, pruned_loss=0.09802, over 5690803.96 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3657, pruned_loss=0.1178, over 5689463.53 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.337, pruned_loss=0.09624, over 5700574.16 frames. ], batch size: 128, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:51:30,272 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 03:51:33,106 INFO [train.py:968] (0/2) Epoch 19, batch 40600, giga_loss[loss=0.2008, simple_loss=0.2823, pruned_loss=0.0597, over 28888.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3354, pruned_loss=0.09614, over 5694106.47 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3658, pruned_loss=0.1178, over 5683805.99 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3333, pruned_loss=0.09438, over 5707201.39 frames. ], batch size: 145, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:51:55,060 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7769, 1.8294, 1.4673, 1.4089], device='cuda:0'), covar=tensor([0.0867, 0.0634, 0.1014, 0.1208], device='cuda:0'), in_proj_covar=tensor([0.0382, 0.0440, 0.0507, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 03:51:56,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.997e+02 1.194e+03 1.404e+03 1.864e+03 7.633e+03, threshold=2.808e+03, percent-clipped=10.0
+2023-03-10 03:52:16,030 INFO [train.py:968] (0/2) Epoch 19, batch 40650, giga_loss[loss=0.2407, simple_loss=0.3183, pruned_loss=0.08152, over 28857.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.333, pruned_loss=0.09529, over 5690469.06 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3662, pruned_loss=0.118, over 5680625.89 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3301, pruned_loss=0.093, over 5704584.25 frames. ], batch size: 186, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:52:46,412 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=863277.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 03:52:48,956 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-10 03:52:55,626 INFO [train.py:968] (0/2) Epoch 19, batch 40700, giga_loss[loss=0.2526, simple_loss=0.3289, pruned_loss=0.08817, over 28894.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3329, pruned_loss=0.09496, over 5698379.40 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3663, pruned_loss=0.118, over 5679432.43 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3294, pruned_loss=0.09235, over 5711566.89 frames. ], batch size: 227, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:53:03,521 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 03:53:17,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=863313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:53:21,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.001e+02 1.203e+03 1.399e+03 1.776e+03 7.509e+03, threshold=2.797e+03, percent-clipped=11.0
+2023-03-10 03:53:38,516 INFO [train.py:968] (0/2) Epoch 19, batch 40750, giga_loss[loss=0.3256, simple_loss=0.377, pruned_loss=0.1371, over 23873.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.336, pruned_loss=0.09641, over 5698876.48 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3669, pruned_loss=0.1185, over 5682368.08 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3323, pruned_loss=0.09359, over 5706956.81 frames. ], batch size: 705, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:54:18,316 INFO [train.py:968] (0/2) Epoch 19, batch 40800, libri_loss[loss=0.2894, simple_loss=0.3718, pruned_loss=0.1035, over 29538.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.34, pruned_loss=0.0981, over 5707610.40 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3669, pruned_loss=0.1184, over 5690057.96 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3357, pruned_loss=0.09504, over 5707827.85 frames. ], batch size: 84, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:54:23,089 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-10 03:54:26,303 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=863400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:54:40,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.495e+02 1.219e+03 1.708e+03 2.712e+03 7.808e+03, threshold=3.416e+03, percent-clipped=22.0
+2023-03-10 03:54:53,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8959, 1.2417, 1.2475, 1.1253], device='cuda:0'), covar=tensor([0.1662, 0.1174, 0.2006, 0.1402], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0745, 0.0710, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 03:54:57,204 INFO [train.py:968] (0/2) Epoch 19, batch 40850, giga_loss[loss=0.2874, simple_loss=0.3659, pruned_loss=0.1045, over 28960.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3437, pruned_loss=0.1, over 5709843.90 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3669, pruned_loss=0.1185, over 5693546.77 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3394, pruned_loss=0.09679, over 5707448.24 frames. ], batch size: 128, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:54:58,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2354, 2.8967, 1.3689, 1.3729], device='cuda:0'), covar=tensor([0.0966, 0.0350, 0.0972, 0.1329], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0543, 0.0374, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-10 03:55:04,995 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4090, 2.0583, 1.4669, 0.6870], device='cuda:0'), covar=tensor([0.4780, 0.2271, 0.3200, 0.4702], device='cuda:0'), in_proj_covar=tensor([0.1709, 0.1606, 0.1583, 0.1392], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 03:55:07,551 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=863450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:55:14,153 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=863456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:55:16,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=863459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:55:42,542 INFO [train.py:968] (0/2) Epoch 19, batch 40900, giga_loss[loss=0.2833, simple_loss=0.3583, pruned_loss=0.1041, over 28898.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3473, pruned_loss=0.1015, over 5694116.62 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3668, pruned_loss=0.1184, over 5686071.87 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3436, pruned_loss=0.09877, over 5699575.46 frames. ], batch size: 145, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:55:42,790 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=863488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:56:07,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.914e+02 1.222e+03 1.511e+03 2.121e+03 6.852e+03, threshold=3.023e+03, percent-clipped=5.0
+2023-03-10 03:56:24,580 INFO [train.py:968] (0/2) Epoch 19, batch 40950, giga_loss[loss=0.2579, simple_loss=0.3335, pruned_loss=0.09118, over 28898.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3487, pruned_loss=0.1022, over 5696783.08 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3661, pruned_loss=0.1181, over 5682547.40 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3457, pruned_loss=0.09967, over 5704565.17 frames. ], batch size: 186, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:57:16,413 INFO [train.py:968] (0/2) Epoch 19, batch 41000, giga_loss[loss=0.2847, simple_loss=0.3598, pruned_loss=0.1048, over 29037.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3544, pruned_loss=0.1072, over 5692631.01 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3663, pruned_loss=0.1182, over 5684796.30 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3518, pruned_loss=0.1049, over 5697063.95 frames. ], batch size: 155, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:57:45,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.979e+02 1.668e+03 2.122e+03 2.863e+03 1.200e+04, threshold=4.243e+03, percent-clipped=23.0
+2023-03-10 03:58:03,271 INFO [train.py:968] (0/2) Epoch 19, batch 41050, giga_loss[loss=0.3142, simple_loss=0.3788, pruned_loss=0.1248, over 28779.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3601, pruned_loss=0.1115, over 5700264.64 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3672, pruned_loss=0.1188, over 5691518.35 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3568, pruned_loss=0.1089, over 5698054.64 frames. ], batch size: 284, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:58:18,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=863652.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 03:58:52,892 INFO [train.py:968] (0/2) Epoch 19, batch 41100, giga_loss[loss=0.3394, simple_loss=0.3993, pruned_loss=0.1398, over 28575.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3679, pruned_loss=0.117, over 5696692.65 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3667, pruned_loss=0.1185, over 5694897.55 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3656, pruned_loss=0.1152, over 5692234.22 frames. ], batch size: 336, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:59:20,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.749e+03 2.216e+03 2.956e+03 5.223e+03, threshold=4.432e+03, percent-clipped=7.0
+2023-03-10 03:59:38,318 INFO [train.py:968] (0/2) Epoch 19, batch 41150, giga_loss[loss=0.3534, simple_loss=0.4082, pruned_loss=0.1493, over 28672.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3713, pruned_loss=0.1199, over 5698824.50 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3661, pruned_loss=0.1181, over 5697216.69 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3701, pruned_loss=0.1187, over 5693184.94 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:59:52,388 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-10 04:00:11,830 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=863775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:00:23,349 INFO [train.py:968] (0/2) Epoch 19, batch 41200, giga_loss[loss=0.4145, simple_loss=0.453, pruned_loss=0.188, over 28605.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3787, pruned_loss=0.1263, over 5691304.46 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3663, pruned_loss=0.1183, over 5691271.76 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3778, pruned_loss=0.1253, over 5691928.65 frames. ], batch size: 307, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:00:32,243 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=863795.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 04:00:34,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=863798.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 04:00:52,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.915e+03 2.583e+03 4.326e+03 9.745e+03, threshold=5.166e+03, percent-clipped=22.0
+2023-03-10 04:01:01,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=863825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:01:03,168 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=863827.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 04:01:03,915 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=863828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:01:15,516 INFO [train.py:968] (0/2) Epoch 19, batch 41250, giga_loss[loss=0.3307, simple_loss=0.3873, pruned_loss=0.137, over 28883.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.384, pruned_loss=0.1312, over 5670314.25 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3666, pruned_loss=0.1185, over 5694434.30 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3832, pruned_loss=0.1304, over 5668000.72 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:01:42,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4571, 1.5946, 1.4975, 1.4546], device='cuda:0'), covar=tensor([0.1391, 0.1673, 0.1867, 0.1670], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0748, 0.0712, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 04:02:10,283 INFO [train.py:968] (0/2) Epoch 19, batch 41300, giga_loss[loss=0.342, simple_loss=0.4025, pruned_loss=0.1407, over 28649.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3864, pruned_loss=0.1343, over 5662570.27 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3661, pruned_loss=0.118, over 5697196.59 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3866, pruned_loss=0.1344, over 5657729.48 frames. ], batch size: 307, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:02:46,384 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=863918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:02:46,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.140e+03 1.715e+03 2.044e+03 2.528e+03 6.427e+03, threshold=4.089e+03, percent-clipped=3.0
+2023-03-10 04:02:48,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=863921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:03:12,745 INFO [train.py:968] (0/2) Epoch 19, batch 41350, libri_loss[loss=0.3354, simple_loss=0.3966, pruned_loss=0.1371, over 25685.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3894, pruned_loss=0.1378, over 5655596.84 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3663, pruned_loss=0.118, over 5696246.85 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3897, pruned_loss=0.1381, over 5652618.85 frames. ], batch size: 136, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:03:24,063 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=863950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:03:32,556 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-10 04:03:43,460 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=863968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:03:46,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=863971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:04:02,867 INFO [train.py:968] (0/2) Epoch 19, batch 41400, giga_loss[loss=0.4278, simple_loss=0.4529, pruned_loss=0.2014, over 27569.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3919, pruned_loss=0.1411, over 5626498.40 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3663, pruned_loss=0.1181, over 5680188.75 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3926, pruned_loss=0.1417, over 5637230.73 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:04:15,859 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-864000.pt
+2023-03-10 04:04:16,335 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:04:26,703 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:04:36,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.213e+03 1.995e+03 2.402e+03 3.314e+03 1.170e+04, threshold=4.805e+03, percent-clipped=15.0
+2023-03-10 04:04:55,962 INFO [train.py:968] (0/2) Epoch 19, batch 41450, libri_loss[loss=0.277, simple_loss=0.3546, pruned_loss=0.09966, over 29259.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3938, pruned_loss=0.1425, over 5618942.40 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3663, pruned_loss=0.1181, over 5676703.98 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3951, pruned_loss=0.1435, over 5629470.08 frames. ], batch size: 97, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:05:00,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3651, 1.8057, 1.5054, 1.5574], device='cuda:0'), covar=tensor([0.0779, 0.0296, 0.0311, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 04:05:36,289 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:05:49,008 INFO [train.py:968] (0/2) Epoch 19, batch 41500, giga_loss[loss=0.2734, simple_loss=0.3392, pruned_loss=0.1038, over 28573.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3933, pruned_loss=0.1431, over 5606645.30 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3659, pruned_loss=0.118, over 5671836.82 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3955, pruned_loss=0.1449, over 5617143.63 frames. ], batch size: 60, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:06:12,767 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:06:13,008 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 04:06:19,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.774e+03 2.377e+03 3.237e+03 9.345e+03, threshold=4.753e+03, percent-clipped=12.0
+2023-03-10 04:06:20,427 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:06:29,205 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-10 04:06:34,418 INFO [train.py:968] (0/2) Epoch 19, batch 41550, giga_loss[loss=0.3069, simple_loss=0.3731, pruned_loss=0.1203, over 28724.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3917, pruned_loss=0.1423, over 5627781.36 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3656, pruned_loss=0.1178, over 5676892.41 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3946, pruned_loss=0.1446, over 5629799.56 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:07:22,477 INFO [train.py:968] (0/2) Epoch 19, batch 41600, libri_loss[loss=0.3319, simple_loss=0.4005, pruned_loss=0.1316, over 29235.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3899, pruned_loss=0.1399, over 5640187.58 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3658, pruned_loss=0.118, over 5672548.83 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3929, pruned_loss=0.1424, over 5643584.56 frames. ], batch size: 94, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:07:31,016 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:07:39,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:07:58,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.762e+02 1.758e+03 2.141e+03 2.781e+03 5.680e+03, threshold=4.282e+03, percent-clipped=4.0
+2023-03-10 04:08:18,822 INFO [train.py:968] (0/2) Epoch 19, batch 41650, giga_loss[loss=0.2997, simple_loss=0.3677, pruned_loss=0.1158, over 28960.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3883, pruned_loss=0.1375, over 5647415.22 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3654, pruned_loss=0.1178, over 5676216.30 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3914, pruned_loss=0.14, over 5646458.64 frames. ], batch size: 145, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:08:22,607 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864241.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 04:09:10,793 INFO [train.py:968] (0/2) Epoch 19, batch 41700, giga_loss[loss=0.422, simple_loss=0.4407, pruned_loss=0.2016, over 26420.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3906, pruned_loss=0.1384, over 5652018.86 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3654, pruned_loss=0.1178, over 5671809.67 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3935, pruned_loss=0.1408, over 5655267.23 frames. ], batch size: 555, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:09:35,626 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:09:42,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.863e+03 2.210e+03 2.946e+03 1.086e+04, threshold=4.419e+03, percent-clipped=4.0
+2023-03-10 04:10:01,088 INFO [train.py:968] (0/2) Epoch 19, batch 41750, giga_loss[loss=0.2859, simple_loss=0.3665, pruned_loss=0.1026, over 28824.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3895, pruned_loss=0.1371, over 5644668.52 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3656, pruned_loss=0.1179, over 5676152.83 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3923, pruned_loss=0.1394, over 5642714.54 frames. ], batch size: 119, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:10:03,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8214, 3.1824, 1.8662, 2.0145], device='cuda:0'), covar=tensor([0.0711, 0.0365, 0.0676, 0.0974], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0544, 0.0374, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:0')
+2023-03-10 04:10:09,020 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:10:12,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:10:40,487 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:10:47,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:10:49,520 INFO [train.py:968] (0/2) Epoch 19, batch 41800, giga_loss[loss=0.3145, simple_loss=0.3821, pruned_loss=0.1234, over 28704.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3873, pruned_loss=0.1345, over 5633645.99 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3653, pruned_loss=0.1178, over 5667290.35 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3906, pruned_loss=0.1371, over 5638424.84 frames. ], batch size: 92, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:11:14,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7668, 2.1777, 1.6968, 2.1550], device='cuda:0'), covar=tensor([0.2538, 0.2517, 0.2899, 0.2262], device='cuda:0'), in_proj_covar=tensor([0.1466, 0.1064, 0.1301, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 04:11:23,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.680e+03 2.177e+03 2.958e+03 6.146e+03, threshold=4.354e+03, percent-clipped=8.0
+2023-03-10 04:11:40,720 INFO [train.py:968] (0/2) Epoch 19, batch 41850, giga_loss[loss=0.3391, simple_loss=0.4069, pruned_loss=0.1356, over 28761.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3835, pruned_loss=0.1304, over 5650087.58 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3651, pruned_loss=0.1178, over 5669137.49 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3865, pruned_loss=0.1326, over 5651813.25 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:11:57,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:12:23,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0786, 1.5588, 5.4893, 3.7149], device='cuda:0'), covar=tensor([0.1493, 0.2635, 0.0348, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0741, 0.0636, 0.0937, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 04:12:32,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:12:35,019 INFO [train.py:968] (0/2) Epoch 19, batch 41900, giga_loss[loss=0.2765, simple_loss=0.3517, pruned_loss=0.1007, over 28906.00 frames. ], tot_loss[loss=0.316, simple_loss=0.379, pruned_loss=0.1265, over 5651017.26 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1177, over 5672774.52 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3818, pruned_loss=0.1284, over 5648822.88 frames. ], batch size: 186, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:12:42,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:13:03,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.716e+03 2.107e+03 3.009e+03 1.291e+04, threshold=4.213e+03, percent-clipped=8.0
+2023-03-10 04:13:14,355 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:13:16,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:13:21,553 INFO [train.py:968] (0/2) Epoch 19, batch 41950, giga_loss[loss=0.3042, simple_loss=0.3693, pruned_loss=0.1195, over 28898.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3759, pruned_loss=0.1248, over 5631037.45 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3652, pruned_loss=0.1178, over 5659069.45 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3782, pruned_loss=0.1265, over 5640175.97 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:13:47,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:13:56,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:14:13,798 INFO [train.py:968] (0/2) Epoch 19, batch 42000, libri_loss[loss=0.2566, simple_loss=0.3235, pruned_loss=0.09485, over 28607.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3754, pruned_loss=0.1245, over 5638005.46 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3651, pruned_loss=0.1178, over 5651977.78 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3774, pruned_loss=0.1259, over 5651407.09 frames. ], batch size: 63, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:14:13,802 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 04:14:23,599 INFO [train.py:1012] (0/2) Epoch 19, validation: loss=0.2044, simple_loss=0.3117, pruned_loss=0.04848, over 944034.00 frames. 
+2023-03-10 04:14:23,600 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 04:14:29,851 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:14:31,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:14:47,892 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864616.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 04:14:53,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.582e+03 1.981e+03 2.574e+03 6.591e+03, threshold=3.963e+03, percent-clipped=2.0
+2023-03-10 04:15:00,178 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:02,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:05,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:14,377 INFO [train.py:968] (0/2) Epoch 19, batch 42050, giga_loss[loss=0.3269, simple_loss=0.3997, pruned_loss=0.1271, over 28883.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3754, pruned_loss=0.124, over 5649837.37 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3647, pruned_loss=0.1175, over 5657176.85 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3776, pruned_loss=0.1255, over 5655805.86 frames. ], batch size: 145, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:15:15,251 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:17,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:38,618 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:49,951 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:16:05,607 INFO [train.py:968] (0/2) Epoch 19, batch 42100, giga_loss[loss=0.362, simple_loss=0.3966, pruned_loss=0.1637, over 26709.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3728, pruned_loss=0.121, over 5661684.22 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1176, over 5658145.52 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3746, pruned_loss=0.1222, over 5665514.41 frames. ], batch size: 555, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:16:08,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:16:34,977 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:16:36,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:16:41,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.590e+02 1.545e+03 1.925e+03 2.401e+03 7.095e+03, threshold=3.850e+03, percent-clipped=6.0
+2023-03-10 04:16:57,314 INFO [train.py:968] (0/2) Epoch 19, batch 42150, giga_loss[loss=0.3138, simple_loss=0.3832, pruned_loss=0.1222, over 28657.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3727, pruned_loss=0.1185, over 5676263.01 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3648, pruned_loss=0.1175, over 5665176.22 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3744, pruned_loss=0.1196, over 5673260.99 frames. ], batch size: 85, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:17:04,496 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:17:19,023 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864759.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 04:17:23,399 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864762.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 04:17:45,384 INFO [train.py:968] (0/2) Epoch 19, batch 42200, libri_loss[loss=0.2965, simple_loss=0.3721, pruned_loss=0.1105, over 29546.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3746, pruned_loss=0.1195, over 5678243.78 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3642, pruned_loss=0.1171, over 5672931.53 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3769, pruned_loss=0.1208, over 5668973.76 frames. ], batch size: 83, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:17:50,808 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864791.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 04:18:19,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.759e+03 2.471e+03 3.861e+03 7.987e+03, threshold=4.941e+03, percent-clipped=25.0
+2023-03-10 04:18:21,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6453, 1.7306, 1.2075, 1.3206], device='cuda:0'), covar=tensor([0.0889, 0.0601, 0.1056, 0.1142], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0448, 0.0514, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 04:18:29,508 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:18:31,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:18:33,509 INFO [train.py:968] (0/2) Epoch 19, batch 42250, giga_loss[loss=0.3181, simple_loss=0.3814, pruned_loss=0.1274, over 28556.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3753, pruned_loss=0.1211, over 5668147.90 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3645, pruned_loss=0.1176, over 5668135.53 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3771, pruned_loss=0.1219, over 5665835.81 frames. ], batch size: 307, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:18:58,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:19:19,698 INFO [train.py:968] (0/2) Epoch 19, batch 42300, libri_loss[loss=0.312, simple_loss=0.3765, pruned_loss=0.1238, over 25707.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3743, pruned_loss=0.1209, over 5669607.50 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.364, pruned_loss=0.1173, over 5669110.72 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3766, pruned_loss=0.1218, over 5667279.55 frames. ], batch size: 136, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:19:21,446 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:19:51,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.654e+02 1.759e+03 2.203e+03 3.104e+03 9.271e+03, threshold=4.406e+03, percent-clipped=2.0
+2023-03-10 04:20:08,342 INFO [train.py:968] (0/2) Epoch 19, batch 42350, giga_loss[loss=0.2964, simple_loss=0.3678, pruned_loss=0.1125, over 29003.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3736, pruned_loss=0.1217, over 5675010.02 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3636, pruned_loss=0.117, over 5672827.05 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.376, pruned_loss=0.1228, over 5669886.25 frames. ], batch size: 155, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:20:32,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8193, 3.6574, 3.4996, 1.7472], device='cuda:0'), covar=tensor([0.0711, 0.0817, 0.0768, 0.2333], device='cuda:0'), in_proj_covar=tensor([0.1199, 0.1116, 0.0945, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 04:20:56,345 INFO [train.py:968] (0/2) Epoch 19, batch 42400, giga_loss[loss=0.2971, simple_loss=0.3658, pruned_loss=0.1141, over 28647.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3726, pruned_loss=0.1222, over 5666891.44 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3637, pruned_loss=0.1171, over 5676472.73 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3745, pruned_loss=0.123, over 5659699.95 frames. ], batch size: 92, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:20:56,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:20:56,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0907, 3.9214, 3.7275, 1.7056], device='cuda:0'), covar=tensor([0.0637, 0.0738, 0.0707, 0.2183], device='cuda:0'), in_proj_covar=tensor([0.1196, 0.1113, 0.0943, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 04:21:28,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.305e+02 1.751e+03 2.387e+03 3.110e+03 5.275e+03, threshold=4.773e+03, percent-clipped=3.0
+2023-03-10 04:21:46,694 INFO [train.py:968] (0/2) Epoch 19, batch 42450, giga_loss[loss=0.2711, simple_loss=0.3444, pruned_loss=0.09886, over 28820.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3704, pruned_loss=0.1199, over 5674876.21 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3636, pruned_loss=0.1171, over 5681683.56 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3722, pruned_loss=0.1207, over 5664352.12 frames. ], batch size: 119, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:22:02,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5680, 1.7655, 1.3370, 1.3393], device='cuda:0'), covar=tensor([0.1000, 0.0598, 0.1065, 0.1120], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0449, 0.0515, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 04:22:19,325 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 04:22:31,212 INFO [train.py:968] (0/2) Epoch 19, batch 42500, libri_loss[loss=0.2856, simple_loss=0.3555, pruned_loss=0.1078, over 29489.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3698, pruned_loss=0.1179, over 5690066.56 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3631, pruned_loss=0.1166, over 5684231.86 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3719, pruned_loss=0.1191, over 5679356.41 frames. ], batch size: 85, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:22:38,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9219, 1.1552, 1.2687, 1.0934], device='cuda:0'), covar=tensor([0.1472, 0.1050, 0.1684, 0.1147], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0748, 0.0712, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 04:23:06,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.491e+02 1.490e+03 1.907e+03 2.593e+03 7.943e+03, threshold=3.815e+03, percent-clipped=2.0
+2023-03-10 04:23:20,968 INFO [train.py:968] (0/2) Epoch 19, batch 42550, libri_loss[loss=0.3539, simple_loss=0.4019, pruned_loss=0.1529, over 29768.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3707, pruned_loss=0.1179, over 5689517.43 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3636, pruned_loss=0.117, over 5687140.02 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.372, pruned_loss=0.1185, over 5678268.42 frames. ], batch size: 87, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:23:29,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9209, 5.7204, 5.4298, 3.0346], device='cuda:0'), covar=tensor([0.0442, 0.0618, 0.0711, 0.1550], device='cuda:0'), in_proj_covar=tensor([0.1197, 0.1115, 0.0944, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 04:24:02,432 INFO [train.py:968] (0/2) Epoch 19, batch 42600, giga_loss[loss=0.2833, simple_loss=0.3552, pruned_loss=0.1057, over 28564.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3685, pruned_loss=0.1169, over 5694220.91 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3635, pruned_loss=0.1169, over 5694819.93 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3701, pruned_loss=0.1175, over 5678534.11 frames. ], batch size: 307, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:24:29,069 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=865218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:24:33,829 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.177e+03 1.585e+03 1.934e+03 2.528e+03 5.457e+03, threshold=3.867e+03, percent-clipped=5.0
+2023-03-10 04:24:48,856 INFO [train.py:968] (0/2) Epoch 19, batch 42650, giga_loss[loss=0.2833, simple_loss=0.3559, pruned_loss=0.1054, over 28843.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3679, pruned_loss=0.1172, over 5692580.65 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3634, pruned_loss=0.1169, over 5697505.53 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3693, pruned_loss=0.1177, over 5677887.39 frames. ], batch size: 186, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:25:15,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=865265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:25:41,254 INFO [train.py:968] (0/2) Epoch 19, batch 42700, giga_loss[loss=0.2941, simple_loss=0.3612, pruned_loss=0.1135, over 28840.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3677, pruned_loss=0.1184, over 5682366.01 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3632, pruned_loss=0.1168, over 5699515.58 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.369, pruned_loss=0.1189, over 5668928.91 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:25:45,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0951, 2.9784, 1.2440, 1.3924], device='cuda:0'), covar=tensor([0.1244, 0.0475, 0.1045, 0.1545], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0551, 0.0379, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 04:25:46,993 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9411, 3.7782, 3.5969, 1.9182], device='cuda:0'), covar=tensor([0.0763, 0.0871, 0.0898, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.1199, 0.1116, 0.0946, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 04:26:14,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.561e+03 2.213e+03 3.021e+03 8.444e+03, threshold=4.425e+03, percent-clipped=12.0
+2023-03-10 04:26:15,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4531, 1.2408, 4.3538, 3.4466], device='cuda:0'), covar=tensor([0.1584, 0.2800, 0.0395, 0.1141], device='cuda:0'), in_proj_covar=tensor([0.0746, 0.0640, 0.0945, 0.0885], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 04:26:20,852 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-10 04:26:30,343 INFO [train.py:968] (0/2) Epoch 19, batch 42750, giga_loss[loss=0.2657, simple_loss=0.3411, pruned_loss=0.09513, over 28517.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3681, pruned_loss=0.1194, over 5676384.73 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3633, pruned_loss=0.1167, over 5701536.42 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3691, pruned_loss=0.1199, over 5663673.82 frames. ], batch size: 78, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:26:34,946 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5459, 1.8388, 1.7360, 1.3918], device='cuda:0'), covar=tensor([0.2558, 0.2096, 0.2281, 0.2540], device='cuda:0'), in_proj_covar=tensor([0.1934, 0.1865, 0.1784, 0.1921], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 04:26:56,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=865363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:27:17,580 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5400, 1.8510, 1.4814, 1.8087], device='cuda:0'), covar=tensor([0.2567, 0.2619, 0.2899, 0.2512], device='cuda:0'), in_proj_covar=tensor([0.1466, 0.1062, 0.1303, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 04:27:20,632 INFO [train.py:968] (0/2) Epoch 19, batch 42800, libri_loss[loss=0.26, simple_loss=0.3399, pruned_loss=0.09002, over 29541.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.366, pruned_loss=0.1182, over 5689389.24 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3634, pruned_loss=0.1167, over 5706667.54 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3669, pruned_loss=0.1187, over 5673911.84 frames. ], batch size: 80, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:27:40,728 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=865408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:27:44,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=865411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:27:59,380 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.750e+02 1.781e+03 2.354e+03 3.192e+03 7.988e+03, threshold=4.709e+03, percent-clipped=7.0
+2023-03-10 04:28:12,083 INFO [train.py:968] (0/2) Epoch 19, batch 42850, libri_loss[loss=0.3392, simple_loss=0.3961, pruned_loss=0.1411, over 29656.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3666, pruned_loss=0.1191, over 5687509.43 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3636, pruned_loss=0.1169, over 5706435.51 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3673, pruned_loss=0.1194, over 5674513.20 frames. ], batch size: 88, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:28:14,595 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=865440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:28:40,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9222, 3.7424, 3.5808, 1.7112], device='cuda:0'), covar=tensor([0.0783, 0.0902, 0.0881, 0.2188], device='cuda:0'), in_proj_covar=tensor([0.1202, 0.1118, 0.0949, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 04:28:55,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3704, 1.2314, 4.1859, 3.3815], device='cuda:0'), covar=tensor([0.1701, 0.2848, 0.0477, 0.0999], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0642, 0.0947, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 04:28:59,425 INFO [train.py:968] (0/2) Epoch 19, batch 42900, libri_loss[loss=0.2568, simple_loss=0.3305, pruned_loss=0.09151, over 29577.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3668, pruned_loss=0.119, over 5683346.49 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3634, pruned_loss=0.1167, over 5700119.00 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3676, pruned_loss=0.1194, over 5677862.92 frames. ], batch size: 75, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:29:05,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9179, 1.2006, 1.3114, 0.9855], device='cuda:0'), covar=tensor([0.1772, 0.1414, 0.2242, 0.1788], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0749, 0.0713, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 04:29:15,868 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=865506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:29:18,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=865509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:29:30,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.171e+02 1.788e+03 2.406e+03 3.464e+03 9.328e+03, threshold=4.811e+03, percent-clipped=9.0
+2023-03-10 04:29:46,413 INFO [train.py:968] (0/2) Epoch 19, batch 42950, giga_loss[loss=0.3529, simple_loss=0.4021, pruned_loss=0.1518, over 28192.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3672, pruned_loss=0.1184, over 5679501.96 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3627, pruned_loss=0.1163, over 5703509.12 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3685, pruned_loss=0.1192, over 5671881.98 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:29:46,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=865538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:30:05,828 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 04:30:23,208 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.26 vs. limit=5.0
+2023-03-10 04:30:31,224 INFO [train.py:968] (0/2) Epoch 19, batch 43000, giga_loss[loss=0.34, simple_loss=0.4015, pruned_loss=0.1392, over 28701.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3665, pruned_loss=0.1172, over 5674105.31 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3628, pruned_loss=0.1165, over 5698011.75 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3675, pruned_loss=0.1176, over 5673015.58 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:30:37,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=865593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:31:03,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.595e+03 2.133e+03 2.861e+03 8.015e+03, threshold=4.267e+03, percent-clipped=8.0
+2023-03-10 04:31:17,582 INFO [train.py:968] (0/2) Epoch 19, batch 43050, libri_loss[loss=0.3174, simple_loss=0.3812, pruned_loss=0.1269, over 29519.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.367, pruned_loss=0.1172, over 5669288.38 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3629, pruned_loss=0.1166, over 5695992.61 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3678, pruned_loss=0.1174, over 5669259.07 frames. ], batch size: 82, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:32:11,556 INFO [train.py:968] (0/2) Epoch 19, batch 43100, giga_loss[loss=0.3318, simple_loss=0.3907, pruned_loss=0.1365, over 28612.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.369, pruned_loss=0.1196, over 5661753.88 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3627, pruned_loss=0.1165, over 5697833.33 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3699, pruned_loss=0.12, over 5659908.88 frames. ], batch size: 307, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:32:14,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2644, 1.2155, 3.6909, 3.1148], device='cuda:0'), covar=tensor([0.1595, 0.2632, 0.0487, 0.0914], device='cuda:0'), in_proj_covar=tensor([0.0747, 0.0642, 0.0947, 0.0888], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 04:32:43,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.841e+03 2.208e+03 2.854e+03 8.689e+03, threshold=4.417e+03, percent-clipped=1.0
+2023-03-10 04:32:58,884 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=865736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:33:00,327 INFO [train.py:968] (0/2) Epoch 19, batch 43150, giga_loss[loss=0.3094, simple_loss=0.362, pruned_loss=0.1284, over 28722.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3717, pruned_loss=0.1229, over 5669721.44 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3623, pruned_loss=0.1161, over 5700787.89 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3729, pruned_loss=0.1235, over 5665082.67 frames. ], batch size: 99, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:33:02,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=865739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:33:36,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=865768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:33:58,214 INFO [train.py:968] (0/2) Epoch 19, batch 43200, giga_loss[loss=0.2864, simple_loss=0.3475, pruned_loss=0.1126, over 28785.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3736, pruned_loss=0.1257, over 5655024.25 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3625, pruned_loss=0.1163, over 5701802.95 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3745, pruned_loss=0.1261, over 5650353.47 frames. ], batch size: 119, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:34:30,265 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-10 04:34:34,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.886e+03 2.391e+03 3.271e+03 1.034e+04, threshold=4.783e+03, percent-clipped=9.0
+2023-03-10 04:34:49,138 INFO [train.py:968] (0/2) Epoch 19, batch 43250, giga_loss[loss=0.288, simple_loss=0.3576, pruned_loss=0.1092, over 28880.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3752, pruned_loss=0.1272, over 5661323.90 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3625, pruned_loss=0.1164, over 5702661.13 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.376, pruned_loss=0.1277, over 5655922.10 frames. ], batch size: 186, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:35:21,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0170, 3.1115, 1.9302, 1.2443], device='cuda:0'), covar=tensor([0.7719, 0.2955, 0.4025, 0.6399], device='cuda:0'), in_proj_covar=tensor([0.1718, 0.1624, 0.1587, 0.1398], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 04:35:31,627 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 04:35:33,104 INFO [train.py:968] (0/2) Epoch 19, batch 43300, giga_loss[loss=0.3175, simple_loss=0.3688, pruned_loss=0.1331, over 28772.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.373, pruned_loss=0.1257, over 5665616.05 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3627, pruned_loss=0.1164, over 5695913.19 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3739, pruned_loss=0.1262, over 5667103.17 frames. ], batch size: 92, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:36:07,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.634e+03 2.176e+03 2.856e+03 6.948e+03, threshold=4.352e+03, percent-clipped=7.0
+2023-03-10 04:36:20,862 INFO [train.py:968] (0/2) Epoch 19, batch 43350, giga_loss[loss=0.298, simple_loss=0.3855, pruned_loss=0.1052, over 28950.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.37, pruned_loss=0.1223, over 5668182.13 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3624, pruned_loss=0.1161, over 5697293.78 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.371, pruned_loss=0.1231, over 5667606.23 frames. ], batch size: 155, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:37:06,440 INFO [train.py:968] (0/2) Epoch 19, batch 43400, giga_loss[loss=0.2846, simple_loss=0.3546, pruned_loss=0.1073, over 28585.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3694, pruned_loss=0.1203, over 5679371.41 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3625, pruned_loss=0.1162, over 5701423.76 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3703, pruned_loss=0.121, over 5674750.14 frames. ], batch size: 336, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:37:06,907 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 04:37:10,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0044, 1.4025, 1.1803, 0.1934], device='cuda:0'), covar=tensor([0.3483, 0.3026, 0.3857, 0.5700], device='cuda:0'), in_proj_covar=tensor([0.1712, 0.1622, 0.1583, 0.1395], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 04:37:16,451 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-866000.pt
+2023-03-10 04:37:38,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.542e+03 2.038e+03 3.306e+03 8.124e+03, threshold=4.076e+03, percent-clipped=11.0
+2023-03-10 04:37:50,702 INFO [train.py:968] (0/2) Epoch 19, batch 43450, giga_loss[loss=0.2976, simple_loss=0.3522, pruned_loss=0.1215, over 28642.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3672, pruned_loss=0.1192, over 5677409.09 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3624, pruned_loss=0.116, over 5708343.90 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3682, pruned_loss=0.12, over 5666994.92 frames. ], batch size: 92, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:38:16,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=866068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:38:36,098 INFO [train.py:968] (0/2) Epoch 19, batch 43500, giga_loss[loss=0.3301, simple_loss=0.3839, pruned_loss=0.1382, over 27937.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3668, pruned_loss=0.1199, over 5672166.21 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3628, pruned_loss=0.1163, over 5706884.01 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3673, pruned_loss=0.1204, over 5664399.83 frames. ], batch size: 412, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:39:12,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.974e+02 1.595e+03 2.027e+03 2.650e+03 6.226e+03, threshold=4.055e+03, percent-clipped=6.0
+2023-03-10 04:39:13,447 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4059, 3.2501, 3.0946, 1.9320], device='cuda:0'), covar=tensor([0.0820, 0.0922, 0.0827, 0.1801], device='cuda:0'), in_proj_covar=tensor([0.1202, 0.1115, 0.0949, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 04:39:25,376 INFO [train.py:968] (0/2) Epoch 19, batch 43550, libri_loss[loss=0.2337, simple_loss=0.3053, pruned_loss=0.08107, over 29477.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3661, pruned_loss=0.1202, over 5655433.72 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3627, pruned_loss=0.1163, over 5699827.88 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3667, pruned_loss=0.1206, over 5655615.91 frames. ], batch size: 70, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:39:58,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2336, 3.0607, 1.4216, 1.3748], device='cuda:0'), covar=tensor([0.1083, 0.0420, 0.0944, 0.1470], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0552, 0.0379, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 04:40:00,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4945, 1.8573, 1.4002, 1.5674], device='cuda:0'), covar=tensor([0.2579, 0.2643, 0.3042, 0.2455], device='cuda:0'), in_proj_covar=tensor([0.1465, 0.1062, 0.1303, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 04:40:10,848 INFO [train.py:968] (0/2) Epoch 19, batch 43600, giga_loss[loss=0.3801, simple_loss=0.4217, pruned_loss=0.1692, over 27935.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3694, pruned_loss=0.122, over 5668098.08 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.363, pruned_loss=0.1165, over 5703976.69 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3697, pruned_loss=0.1223, over 5663937.30 frames. ], batch size: 412, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:40:40,616 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.39 vs. limit=5.0
+2023-03-10 04:40:42,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4825, 1.7203, 1.3463, 1.6402], device='cuda:0'), covar=tensor([0.2779, 0.2790, 0.3271, 0.2280], device='cuda:0'), in_proj_covar=tensor([0.1466, 0.1063, 0.1303, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 04:40:46,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.719e+03 2.140e+03 2.768e+03 9.048e+03, threshold=4.281e+03, percent-clipped=12.0
+2023-03-10 04:40:59,193 INFO [train.py:968] (0/2) Epoch 19, batch 43650, libri_loss[loss=0.2684, simple_loss=0.3395, pruned_loss=0.0987, over 29576.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3727, pruned_loss=0.1218, over 5672484.01 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3628, pruned_loss=0.1162, over 5706915.62 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3733, pruned_loss=0.1224, over 5665691.21 frames. ], batch size: 75, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:41:00,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9930, 1.3213, 1.0708, 0.2323], device='cuda:0'), covar=tensor([0.3543, 0.3042, 0.3848, 0.5883], device='cuda:0'), in_proj_covar=tensor([0.1708, 0.1620, 0.1581, 0.1393], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 04:41:47,924 INFO [train.py:968] (0/2) Epoch 19, batch 43700, giga_loss[loss=0.3604, simple_loss=0.4113, pruned_loss=0.1548, over 27639.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3732, pruned_loss=0.1203, over 5670972.74 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3618, pruned_loss=0.1155, over 5714867.49 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3749, pruned_loss=0.1216, over 5657091.94 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:42:25,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.691e+03 2.281e+03 3.255e+03 1.405e+04, threshold=4.561e+03, percent-clipped=11.0
+2023-03-10 04:42:38,374 INFO [train.py:968] (0/2) Epoch 19, batch 43750, giga_loss[loss=0.2817, simple_loss=0.3581, pruned_loss=0.1026, over 28953.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3752, pruned_loss=0.1219, over 5668671.41 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.362, pruned_loss=0.1156, over 5708159.62 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3765, pruned_loss=0.1229, over 5662888.42 frames. ], batch size: 106, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:43:25,062 INFO [train.py:968] (0/2) Epoch 19, batch 43800, giga_loss[loss=0.3537, simple_loss=0.4106, pruned_loss=0.1484, over 28752.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3769, pruned_loss=0.1235, over 5675182.34 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3616, pruned_loss=0.1153, over 5713959.90 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3787, pruned_loss=0.1248, over 5664322.78 frames. ], batch size: 284, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:43:56,596 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5691, 1.6638, 1.7815, 1.3875], device='cuda:0'), covar=tensor([0.1729, 0.2424, 0.1416, 0.1655], device='cuda:0'), in_proj_covar=tensor([0.0886, 0.0699, 0.0930, 0.0829], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:0')
+2023-03-10 04:44:01,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6752, 4.4868, 4.3019, 2.3277], device='cuda:0'), covar=tensor([0.0542, 0.0679, 0.0714, 0.1920], device='cuda:0'), in_proj_covar=tensor([0.1200, 0.1117, 0.0948, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 04:44:01,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.813e+03 2.284e+03 3.234e+03 1.829e+04, threshold=4.567e+03, percent-clipped=6.0
+2023-03-10 04:44:12,033 INFO [train.py:968] (0/2) Epoch 19, batch 43850, libri_loss[loss=0.2557, simple_loss=0.3339, pruned_loss=0.08877, over 29541.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3763, pruned_loss=0.1239, over 5682097.19 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3615, pruned_loss=0.1152, over 5715374.79 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3782, pruned_loss=0.1253, over 5671413.01 frames. ], batch size: 78, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:44:16,478 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=866443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:44:59,442 INFO [train.py:968] (0/2) Epoch 19, batch 43900, giga_loss[loss=0.2718, simple_loss=0.3426, pruned_loss=0.1005, over 28669.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3752, pruned_loss=0.124, over 5675038.42 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3616, pruned_loss=0.1152, over 5716741.22 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3771, pruned_loss=0.1253, over 5664246.82 frames. ], batch size: 242, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:45:15,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-10 04:45:17,246 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=866506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:45:35,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.573e+03 2.161e+03 2.867e+03 5.667e+03, threshold=4.322e+03, percent-clipped=8.0
+2023-03-10 04:45:46,011 INFO [train.py:968] (0/2) Epoch 19, batch 43950, giga_loss[loss=0.4027, simple_loss=0.429, pruned_loss=0.1882, over 26673.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3738, pruned_loss=0.1242, over 5674182.02 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3613, pruned_loss=0.1152, over 5717869.40 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3758, pruned_loss=0.1254, over 5663191.44 frames. ], batch size: 555, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:46:17,342 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-10 04:46:35,909 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=866586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:46:37,213 INFO [train.py:968] (0/2) Epoch 19, batch 44000, giga_loss[loss=0.2842, simple_loss=0.3511, pruned_loss=0.1086, over 28863.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3734, pruned_loss=0.1249, over 5673642.78 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.361, pruned_loss=0.1151, over 5720232.74 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3753, pruned_loss=0.1261, over 5662528.79 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:46:38,138 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=866589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:47:09,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=866618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:47:16,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.914e+02 1.902e+03 2.705e+03 4.047e+03 2.666e+04, threshold=5.410e+03, percent-clipped=23.0
+2023-03-10 04:47:26,881 INFO [train.py:968] (0/2) Epoch 19, batch 44050, giga_loss[loss=0.3594, simple_loss=0.4021, pruned_loss=0.1584, over 26637.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3732, pruned_loss=0.125, over 5652262.39 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3615, pruned_loss=0.1152, over 5714425.73 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3746, pruned_loss=0.126, over 5648224.84 frames. ], batch size: 555, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:48:13,884 INFO [train.py:968] (0/2) Epoch 19, batch 44100, giga_loss[loss=0.334, simple_loss=0.3723, pruned_loss=0.1478, over 23887.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3719, pruned_loss=0.1244, over 5646458.74 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3616, pruned_loss=0.1152, over 5710617.16 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3733, pruned_loss=0.1255, over 5644312.58 frames. ], batch size: 705, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:48:21,875 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2362, 0.7723, 0.8913, 1.4450], device='cuda:0'), covar=tensor([0.0753, 0.0369, 0.0340, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 04:48:48,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.689e+03 2.229e+03 2.819e+03 6.099e+03, threshold=4.459e+03, percent-clipped=2.0
+2023-03-10 04:49:00,185 INFO [train.py:968] (0/2) Epoch 19, batch 44150, giga_loss[loss=0.2683, simple_loss=0.3383, pruned_loss=0.09916, over 29014.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3711, pruned_loss=0.1244, over 5661250.15 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3616, pruned_loss=0.1153, over 5712466.52 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3722, pruned_loss=0.1254, over 5657363.18 frames. ], batch size: 128, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:49:12,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 04:49:34,180 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3744, 2.0396, 1.6026, 0.5701], device='cuda:0'), covar=tensor([0.4845, 0.2779, 0.3984, 0.5912], device='cuda:0'), in_proj_covar=tensor([0.1714, 0.1630, 0.1582, 0.1399], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 04:49:44,029 INFO [train.py:968] (0/2) Epoch 19, batch 44200, giga_loss[loss=0.2829, simple_loss=0.3522, pruned_loss=0.1068, over 28830.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3685, pruned_loss=0.1224, over 5652416.27 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.361, pruned_loss=0.1148, over 5708064.98 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3703, pruned_loss=0.1238, over 5651861.07 frames. ], batch size: 119, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:50:19,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.450e+02 1.622e+03 1.913e+03 2.662e+03 5.063e+03, threshold=3.826e+03, percent-clipped=2.0
+2023-03-10 04:50:30,879 INFO [train.py:968] (0/2) Epoch 19, batch 44250, giga_loss[loss=0.2941, simple_loss=0.361, pruned_loss=0.1136, over 28766.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3687, pruned_loss=0.1214, over 5659603.98 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3606, pruned_loss=0.1144, over 5713025.16 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3708, pruned_loss=0.1232, over 5653323.47 frames. ], batch size: 92, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:50:58,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6757, 2.5067, 2.6043, 2.1505], device='cuda:0'), covar=tensor([0.1810, 0.2406, 0.1776, 0.1981], device='cuda:0'), in_proj_covar=tensor([0.0472, 0.0752, 0.0717, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 04:51:16,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=866881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:51:21,271 INFO [train.py:968] (0/2) Epoch 19, batch 44300, giga_loss[loss=0.3061, simple_loss=0.3689, pruned_loss=0.1216, over 28836.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.372, pruned_loss=0.1233, over 5657141.64 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3605, pruned_loss=0.1142, over 5712489.12 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.374, pruned_loss=0.125, over 5651752.27 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:52:03,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.313e+02 1.626e+03 1.948e+03 2.662e+03 4.420e+03, threshold=3.897e+03, percent-clipped=2.0
+2023-03-10 04:52:12,815 INFO [train.py:968] (0/2) Epoch 19, batch 44350, giga_loss[loss=0.2992, simple_loss=0.3629, pruned_loss=0.1177, over 28800.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1237, over 5665026.01 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3606, pruned_loss=0.1142, over 5715425.05 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3739, pruned_loss=0.1252, over 5657348.54 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:53:02,534 INFO [train.py:968] (0/2) Epoch 19, batch 44400, giga_loss[loss=0.3709, simple_loss=0.4142, pruned_loss=0.1638, over 27661.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3746, pruned_loss=0.123, over 5661677.49 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3608, pruned_loss=0.1144, over 5707438.50 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3758, pruned_loss=0.1242, over 5663183.80 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:53:33,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9235, 1.1269, 1.0357, 0.8986], device='cuda:0'), covar=tensor([0.2446, 0.2443, 0.1570, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.1934, 0.1868, 0.1796, 0.1925], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 04:53:33,670 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=867024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:53:35,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.252e+02 1.517e+03 1.997e+03 2.859e+03 6.935e+03, threshold=3.993e+03, percent-clipped=12.0
+2023-03-10 04:53:36,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=867027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:53:43,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1873, 1.6996, 0.9289, 1.2542], device='cuda:0'), covar=tensor([0.1348, 0.0708, 0.1738, 0.1351], device='cuda:0'), in_proj_covar=tensor([0.0387, 0.0446, 0.0513, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 04:53:45,434 INFO [train.py:968] (0/2) Epoch 19, batch 44450, giga_loss[loss=0.2853, simple_loss=0.3707, pruned_loss=0.09995, over 28941.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3763, pruned_loss=0.122, over 5636467.56 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3613, pruned_loss=0.1148, over 5680382.55 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.377, pruned_loss=0.1226, over 5662450.69 frames. ], batch size: 145, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:53:52,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2246, 2.6050, 1.1846, 1.4717], device='cuda:0'), covar=tensor([0.1020, 0.0473, 0.0993, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0552, 0.0379, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 04:54:03,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=867056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:54:33,355 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-10 04:54:36,062 INFO [train.py:968] (0/2) Epoch 19, batch 44500, libri_loss[loss=0.3401, simple_loss=0.3912, pruned_loss=0.1445, over 19289.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3806, pruned_loss=0.125, over 5594595.62 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3621, pruned_loss=0.1155, over 5635597.95 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3807, pruned_loss=0.125, over 5655680.11 frames. ], batch size: 187, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:55:13,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.796e+02 1.720e+03 2.545e+03 3.504e+03 1.085e+04, threshold=5.091e+03, percent-clipped=11.0
+2023-03-10 04:55:24,294 INFO [train.py:968] (0/2) Epoch 19, batch 44550, giga_loss[loss=0.3903, simple_loss=0.4282, pruned_loss=0.1762, over 28334.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3833, pruned_loss=0.1281, over 5577949.68 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3624, pruned_loss=0.1158, over 5601102.91 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3833, pruned_loss=0.1279, over 5656291.90 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:55:45,863 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-10 04:55:48,239 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-19.pt
+2023-03-10 04:56:39,965 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4576, 1.5575, 1.6570, 1.2261], device='cuda:0'), covar=tensor([0.1922, 0.2654, 0.1554, 0.1892], device='cuda:0'), in_proj_covar=tensor([0.0883, 0.0698, 0.0928, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-10 04:57:00,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4687, 3.4802, 1.5516, 1.5368], device='cuda:0'), covar=tensor([0.1047, 0.0271, 0.0962, 0.1405], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0550, 0.0378, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 04:57:04,580 INFO [train.py:968] (0/2) Epoch 20, batch 50, giga_loss[loss=0.276, simple_loss=0.3723, pruned_loss=0.08987, over 28851.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.368, pruned_loss=0.1052, over 1267265.45 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3305, pruned_loss=0.08363, over 203390.68 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3743, pruned_loss=0.1089, over 1102653.71 frames. ], batch size: 174, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 04:57:22,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.469e+02 1.329e+03 1.833e+03 2.827e+03 2.115e+04, threshold=3.666e+03, percent-clipped=10.0
+2023-03-10 04:57:39,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0658, 3.8618, 3.6778, 1.7426], device='cuda:0'), covar=tensor([0.0644, 0.0802, 0.0770, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.1201, 0.1114, 0.0946, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 04:57:53,951 INFO [train.py:968] (0/2) Epoch 20, batch 100, giga_loss[loss=0.2492, simple_loss=0.323, pruned_loss=0.08766, over 28422.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3628, pruned_loss=0.1034, over 2248784.47 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3422, pruned_loss=0.08955, over 342976.53 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3656, pruned_loss=0.1053, over 2026738.42 frames. ], batch size: 71, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 04:58:03,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5494, 1.8319, 1.4544, 1.6950], device='cuda:0'), covar=tensor([0.2764, 0.2799, 0.3202, 0.2404], device='cuda:0'), in_proj_covar=tensor([0.1468, 0.1066, 0.1305, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 04:58:24,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5626, 1.7789, 1.7905, 1.3593], device='cuda:0'), covar=tensor([0.1831, 0.2584, 0.1574, 0.1737], device='cuda:0'), in_proj_covar=tensor([0.0887, 0.0700, 0.0933, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 04:58:40,038 INFO [train.py:968] (0/2) Epoch 20, batch 150, giga_loss[loss=0.3398, simple_loss=0.3755, pruned_loss=0.152, over 26635.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3502, pruned_loss=0.0989, over 3012978.69 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3438, pruned_loss=0.09145, over 426163.03 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3511, pruned_loss=0.09971, over 2793708.84 frames. ], batch size: 555, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 04:58:45,882 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=867318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:58:54,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.395e+02 1.075e+03 1.350e+03 1.794e+03 3.521e+03, threshold=2.700e+03, percent-clipped=0.0
+2023-03-10 04:59:05,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4940, 1.7938, 1.3908, 1.5873], device='cuda:0'), covar=tensor([0.2659, 0.2609, 0.3034, 0.2449], device='cuda:0'), in_proj_covar=tensor([0.1469, 0.1067, 0.1306, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 04:59:19,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 04:59:21,781 INFO [train.py:968] (0/2) Epoch 20, batch 200, libri_loss[loss=0.2586, simple_loss=0.3452, pruned_loss=0.08597, over 29526.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3367, pruned_loss=0.09203, over 3616817.47 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3457, pruned_loss=0.09202, over 559601.76 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3362, pruned_loss=0.09225, over 3386055.42 frames. ], batch size: 84, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 05:00:02,422 INFO [train.py:968] (0/2) Epoch 20, batch 250, giga_loss[loss=0.2197, simple_loss=0.295, pruned_loss=0.07214, over 28588.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3263, pruned_loss=0.08658, over 4084592.63 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3426, pruned_loss=0.08979, over 743337.56 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3251, pruned_loss=0.08667, over 3834660.44 frames. ], batch size: 307, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 05:00:15,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.008e+02 1.027e+03 1.332e+03 1.863e+03 3.105e+03, threshold=2.664e+03, percent-clipped=7.0
+2023-03-10 05:00:27,511 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6735, 1.9353, 1.8524, 1.7307], device='cuda:0'), covar=tensor([0.2158, 0.2267, 0.2340, 0.2240], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0743, 0.0709, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 05:00:34,621 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3943, 1.2586, 4.2790, 3.4041], device='cuda:0'), covar=tensor([0.1756, 0.2963, 0.0403, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0742, 0.0638, 0.0944, 0.0884], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 05:00:40,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6758, 1.8361, 1.5236, 1.6884], device='cuda:0'), covar=tensor([0.2883, 0.2801, 0.3233, 0.2670], device='cuda:0'), in_proj_covar=tensor([0.1473, 0.1068, 0.1310, 0.0990], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 05:00:43,236 INFO [train.py:968] (0/2) Epoch 20, batch 300, giga_loss[loss=0.2253, simple_loss=0.2969, pruned_loss=0.07681, over 27934.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3174, pruned_loss=0.08219, over 4448234.80 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3368, pruned_loss=0.08604, over 945771.18 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3162, pruned_loss=0.08252, over 4187750.03 frames. ], batch size: 412, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 05:00:47,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6436, 1.8403, 1.5397, 1.6038], device='cuda:0'), covar=tensor([0.2590, 0.2648, 0.2941, 0.2644], device='cuda:0'), in_proj_covar=tensor([0.1473, 0.1068, 0.1309, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 05:01:27,617 INFO [train.py:968] (0/2) Epoch 20, batch 350, giga_loss[loss=0.2081, simple_loss=0.2889, pruned_loss=0.06364, over 29140.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3117, pruned_loss=0.07998, over 4728741.86 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3389, pruned_loss=0.0875, over 1067839.36 frames. ], giga_tot_loss[loss=0.2344, simple_loss=0.3094, pruned_loss=0.07968, over 4492195.66 frames. ], batch size: 128, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 05:01:33,965 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=867520.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:01:40,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.213e+02 9.579e+02 1.343e+03 1.699e+03 6.861e+03, threshold=2.686e+03, percent-clipped=5.0
+2023-03-10 05:02:09,008 INFO [train.py:968] (0/2) Epoch 20, batch 400, libri_loss[loss=0.2529, simple_loss=0.3404, pruned_loss=0.08265, over 29538.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3065, pruned_loss=0.07767, over 4950326.45 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3393, pruned_loss=0.08747, over 1116119.95 frames. ], giga_tot_loss[loss=0.2293, simple_loss=0.3041, pruned_loss=0.07726, over 4754906.11 frames. ], batch size: 82, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:02:44,750 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=867604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:02:50,081 INFO [train.py:968] (0/2) Epoch 20, batch 450, giga_loss[loss=0.2024, simple_loss=0.2876, pruned_loss=0.05862, over 28931.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3041, pruned_loss=0.07637, over 5114954.59 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3395, pruned_loss=0.08816, over 1247889.93 frames. ], giga_tot_loss[loss=0.2261, simple_loss=0.3011, pruned_loss=0.07552, over 4944871.04 frames. ], batch size: 174, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:02:54,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-10 05:03:04,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0381, 3.8506, 3.6198, 1.9603], device='cuda:0'), covar=tensor([0.0668, 0.0831, 0.0832, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1190, 0.1104, 0.0937, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 05:03:06,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.901e+02 1.050e+03 1.321e+03 1.712e+03 4.608e+03, threshold=2.642e+03, percent-clipped=2.0
+2023-03-10 05:03:32,224 INFO [train.py:968] (0/2) Epoch 20, batch 500, giga_loss[loss=0.205, simple_loss=0.2745, pruned_loss=0.06774, over 28462.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3029, pruned_loss=0.07588, over 5252218.46 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3415, pruned_loss=0.08964, over 1406556.14 frames. ], giga_tot_loss[loss=0.2238, simple_loss=0.2988, pruned_loss=0.0744, over 5094301.31 frames. ], batch size: 85, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:03:59,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=867693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:04:16,268 INFO [train.py:968] (0/2) Epoch 20, batch 550, giga_loss[loss=0.2758, simple_loss=0.3324, pruned_loss=0.1096, over 26730.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3017, pruned_loss=0.07545, over 5349703.28 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.342, pruned_loss=0.09032, over 1536092.19 frames. ], giga_tot_loss[loss=0.2223, simple_loss=0.2971, pruned_loss=0.07369, over 5210404.18 frames. ], batch size: 555, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:04:29,935 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9441, 1.9827, 1.9500, 1.6854], device='cuda:0'), covar=tensor([0.2088, 0.2705, 0.2428, 0.2524], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0742, 0.0710, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 05:04:30,556 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.674e+02 1.014e+03 1.406e+03 1.954e+03 4.741e+03, threshold=2.812e+03, percent-clipped=12.0
+2023-03-10 05:04:49,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4922, 1.6459, 1.5543, 1.4577], device='cuda:0'), covar=tensor([0.3076, 0.2384, 0.1971, 0.2386], device='cuda:0'), in_proj_covar=tensor([0.1924, 0.1846, 0.1779, 0.1911], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 05:04:58,433 INFO [train.py:968] (0/2) Epoch 20, batch 600, giga_loss[loss=0.1839, simple_loss=0.2624, pruned_loss=0.05266, over 29021.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3, pruned_loss=0.07468, over 5426824.35 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3418, pruned_loss=0.09031, over 1676815.46 frames. ], giga_tot_loss[loss=0.2203, simple_loss=0.2951, pruned_loss=0.07278, over 5307859.47 frames. ], batch size: 136, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:05:24,773 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-10 05:05:45,927 INFO [train.py:968] (0/2) Epoch 20, batch 650, giga_loss[loss=0.2187, simple_loss=0.2982, pruned_loss=0.06964, over 28205.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.2986, pruned_loss=0.07396, over 5478159.91 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3428, pruned_loss=0.09054, over 1799839.67 frames. ], giga_tot_loss[loss=0.2185, simple_loss=0.2931, pruned_loss=0.07191, over 5372601.77 frames. ], batch size: 368, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:05:48,387 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 05:06:01,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.930e+02 1.076e+03 1.272e+03 1.686e+03 4.629e+03, threshold=2.545e+03, percent-clipped=7.0
+2023-03-10 05:06:09,806 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=867836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:06:12,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=867839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:06:28,856 INFO [train.py:968] (0/2) Epoch 20, batch 700, giga_loss[loss=0.2628, simple_loss=0.3214, pruned_loss=0.1021, over 28534.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.2971, pruned_loss=0.0737, over 5523365.69 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3427, pruned_loss=0.09124, over 1891670.91 frames. ], giga_tot_loss[loss=0.2174, simple_loss=0.2918, pruned_loss=0.07148, over 5439413.93 frames. ], batch size: 336, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:06:38,347 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=867868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:06:45,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4154, 1.5684, 1.3354, 1.5602], device='cuda:0'), covar=tensor([0.0789, 0.0333, 0.0351, 0.0925], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 05:07:02,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=867895.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:07:15,486 INFO [train.py:968] (0/2) Epoch 20, batch 750, giga_loss[loss=0.1997, simple_loss=0.2751, pruned_loss=0.06217, over 28855.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2939, pruned_loss=0.07199, over 5566438.97 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3417, pruned_loss=0.09047, over 1968107.12 frames. ], giga_tot_loss[loss=0.2147, simple_loss=0.2891, pruned_loss=0.07011, over 5496362.88 frames. ], batch size: 199, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:07:32,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.690e+02 9.576e+02 1.212e+03 1.705e+03 5.658e+03, threshold=2.424e+03, percent-clipped=10.0
+2023-03-10 05:07:42,721 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=867942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:07:44,243 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=867943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:08:00,294 INFO [train.py:968] (0/2) Epoch 20, batch 800, giga_loss[loss=0.1744, simple_loss=0.2559, pruned_loss=0.04647, over 28674.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2907, pruned_loss=0.07092, over 5595646.45 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3427, pruned_loss=0.09111, over 2007079.80 frames. ], giga_tot_loss[loss=0.2121, simple_loss=0.2862, pruned_loss=0.06907, over 5537980.49 frames. ], batch size: 66, lr: 1.62e-03, grad_scale: 8.0
+2023-03-10 05:08:15,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=867979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:08:18,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4110, 1.4927, 3.8912, 3.2280], device='cuda:0'), covar=tensor([0.1590, 0.2625, 0.0457, 0.0964], device='cuda:0'), in_proj_covar=tensor([0.0741, 0.0636, 0.0940, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 05:08:31,447 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2189, 1.4874, 1.2107, 0.9737], device='cuda:0'), covar=tensor([0.2520, 0.2563, 0.2887, 0.2244], device='cuda:0'), in_proj_covar=tensor([0.1478, 0.1070, 0.1313, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 05:08:37,031 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-868000.pt
+2023-03-10 05:08:45,839 INFO [train.py:968] (0/2) Epoch 20, batch 850, giga_loss[loss=0.2816, simple_loss=0.3591, pruned_loss=0.102, over 29041.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2985, pruned_loss=0.07555, over 5606011.13 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3424, pruned_loss=0.09093, over 2117289.35 frames. ], giga_tot_loss[loss=0.2205, simple_loss=0.2937, pruned_loss=0.07367, over 5556266.46 frames. ], batch size: 128, lr: 1.62e-03, grad_scale: 8.0
+2023-03-10 05:09:03,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.031e+02 1.304e+03 1.733e+03 2.375e+03 4.397e+03, threshold=3.466e+03, percent-clipped=24.0
+2023-03-10 05:09:11,449 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868038.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:09:14,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868041.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:09:24,023 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 05:09:35,605 INFO [train.py:968] (0/2) Epoch 20, batch 900, giga_loss[loss=0.2916, simple_loss=0.368, pruned_loss=0.1076, over 28316.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3121, pruned_loss=0.08255, over 5626202.54 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3424, pruned_loss=0.09081, over 2192672.46 frames. ], giga_tot_loss[loss=0.2348, simple_loss=0.3077, pruned_loss=0.08092, over 5581381.40 frames. ], batch size: 77, lr: 1.62e-03, grad_scale: 8.0
+2023-03-10 05:09:41,676 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868070.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:10:15,713 INFO [train.py:968] (0/2) Epoch 20, batch 950, giga_loss[loss=0.3, simple_loss=0.3643, pruned_loss=0.1178, over 28532.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3229, pruned_loss=0.08766, over 5653086.60 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3414, pruned_loss=0.09077, over 2337685.74 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3187, pruned_loss=0.08615, over 5608087.43 frames. ], batch size: 85, lr: 1.62e-03, grad_scale: 8.0
+2023-03-10 05:10:26,976 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:10:28,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:10:30,686 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.429e+03 1.775e+03 2.311e+03 5.217e+03, threshold=3.550e+03, percent-clipped=8.0
+2023-03-10 05:10:54,611 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:10:59,056 INFO [train.py:968] (0/2) Epoch 20, batch 1000, giga_loss[loss=0.2824, simple_loss=0.3596, pruned_loss=0.1027, over 28913.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3311, pruned_loss=0.09119, over 5661713.19 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3409, pruned_loss=0.09026, over 2403122.48 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3278, pruned_loss=0.09018, over 5626414.46 frames. ], batch size: 199, lr: 1.62e-03, grad_scale: 8.0
+2023-03-10 05:11:17,111 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:11:38,970 INFO [train.py:968] (0/2) Epoch 20, batch 1050, giga_loss[loss=0.2448, simple_loss=0.3344, pruned_loss=0.07762, over 28799.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3343, pruned_loss=0.09135, over 5678346.78 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3401, pruned_loss=0.08993, over 2555986.61 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3317, pruned_loss=0.09075, over 5641366.62 frames. ], batch size: 285, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:11:50,943 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 05:11:54,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.675e+02 1.246e+03 1.653e+03 2.121e+03 4.608e+03, threshold=3.305e+03, percent-clipped=4.0
+2023-03-10 05:12:25,774 INFO [train.py:968] (0/2) Epoch 20, batch 1100, giga_loss[loss=0.2964, simple_loss=0.3718, pruned_loss=0.1104, over 28540.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3367, pruned_loss=0.0916, over 5671414.88 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3402, pruned_loss=0.09009, over 2633850.20 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3345, pruned_loss=0.09109, over 5641401.46 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:12:43,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9107, 1.1581, 3.3502, 2.9080], device='cuda:0'), covar=tensor([0.1853, 0.2831, 0.0513, 0.1039], device='cuda:0'), in_proj_covar=tensor([0.0738, 0.0635, 0.0936, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 05:13:10,731 INFO [train.py:968] (0/2) Epoch 20, batch 1150, giga_loss[loss=0.2792, simple_loss=0.3606, pruned_loss=0.09895, over 29083.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3379, pruned_loss=0.09187, over 5686102.53 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3396, pruned_loss=0.08964, over 2695330.07 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3363, pruned_loss=0.09168, over 5661435.32 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:13:12,819 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:13:15,568 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:13:16,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:13:23,970 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.144e+02 1.188e+03 1.344e+03 1.693e+03 6.343e+03, threshold=2.687e+03, percent-clipped=1.0
+2023-03-10 05:13:53,105 INFO [train.py:968] (0/2) Epoch 20, batch 1200, giga_loss[loss=0.2943, simple_loss=0.367, pruned_loss=0.1108, over 28675.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3407, pruned_loss=0.09412, over 5680523.44 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3403, pruned_loss=0.08999, over 2788495.36 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3392, pruned_loss=0.09391, over 5656691.22 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:14:20,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6436, 1.9667, 1.7318, 1.8665], device='cuda:0'), covar=tensor([0.0775, 0.0281, 0.0305, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 05:14:34,512 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-10 05:14:40,082 INFO [train.py:968] (0/2) Epoch 20, batch 1250, giga_loss[loss=0.2949, simple_loss=0.3708, pruned_loss=0.1095, over 28026.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3443, pruned_loss=0.09641, over 5681748.78 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3402, pruned_loss=0.08988, over 2818671.28 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3431, pruned_loss=0.09635, over 5661896.00 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:14:50,642 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:14:56,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.175e+02 1.184e+03 1.427e+03 2.028e+03 7.737e+03, threshold=2.855e+03, percent-clipped=12.0
+2023-03-10 05:15:08,904 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:23,984 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:24,408 INFO [train.py:968] (0/2) Epoch 20, batch 1300, libri_loss[loss=0.2154, simple_loss=0.2984, pruned_loss=0.06617, over 29398.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3469, pruned_loss=0.09752, over 5678531.54 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3399, pruned_loss=0.0895, over 2920045.66 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3463, pruned_loss=0.0979, over 5659842.22 frames. ], batch size: 67, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:15:24,642 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:26,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:27,404 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:39,717 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:49,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5572, 2.2292, 1.6372, 0.7687], device='cuda:0'), covar=tensor([0.5969, 0.3037, 0.4490, 0.6281], device='cuda:0'), in_proj_covar=tensor([0.1702, 0.1610, 0.1573, 0.1388], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 05:15:50,593 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:51,249 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:16:04,978 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4229, 1.6745, 1.3638, 1.3206], device='cuda:0'), covar=tensor([0.2639, 0.2664, 0.2946, 0.2357], device='cuda:0'), in_proj_covar=tensor([0.1476, 0.1072, 0.1311, 0.0990], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 05:16:05,360 INFO [train.py:968] (0/2) Epoch 20, batch 1350, giga_loss[loss=0.2813, simple_loss=0.3582, pruned_loss=0.1022, over 28671.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3497, pruned_loss=0.09787, over 5693412.68 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3412, pruned_loss=0.08997, over 3002622.97 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3489, pruned_loss=0.09816, over 5677769.04 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:16:08,696 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:16:20,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.444e+02 1.199e+03 1.473e+03 1.924e+03 3.103e+03, threshold=2.946e+03, percent-clipped=4.0
+2023-03-10 05:16:42,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:16:49,902 INFO [train.py:968] (0/2) Epoch 20, batch 1400, giga_loss[loss=0.2776, simple_loss=0.3556, pruned_loss=0.09982, over 28244.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3515, pruned_loss=0.09855, over 5692294.86 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3413, pruned_loss=0.08985, over 3074152.55 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3511, pruned_loss=0.09904, over 5675458.51 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:17:01,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4117, 3.8073, 1.4784, 1.6908], device='cuda:0'), covar=tensor([0.0968, 0.0241, 0.0945, 0.1308], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0542, 0.0376, 0.0418], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 05:17:26,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9436, 1.1536, 1.0648, 0.8824], device='cuda:0'), covar=tensor([0.1999, 0.2452, 0.1526, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.1914, 0.1845, 0.1767, 0.1903], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 05:17:32,755 INFO [train.py:968] (0/2) Epoch 20, batch 1450, giga_loss[loss=0.293, simple_loss=0.3751, pruned_loss=0.1054, over 28882.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3528, pruned_loss=0.09869, over 5694865.47 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3412, pruned_loss=0.08964, over 3116173.39 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3527, pruned_loss=0.09928, over 5679170.87 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:17:48,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1126, 1.2474, 1.0511, 0.8804], device='cuda:0'), covar=tensor([0.1067, 0.0531, 0.1149, 0.1133], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0445, 0.0515, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 05:17:48,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.031e+02 1.211e+03 1.468e+03 1.803e+03 4.020e+03, threshold=2.936e+03, percent-clipped=3.0
+2023-03-10 05:17:53,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8582, 1.9985, 2.1065, 1.6355], device='cuda:0'), covar=tensor([0.1996, 0.2436, 0.1562, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0898, 0.0702, 0.0943, 0.0839], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 05:18:12,384 INFO [train.py:968] (0/2) Epoch 20, batch 1500, giga_loss[loss=0.2396, simple_loss=0.3327, pruned_loss=0.07329, over 28776.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3508, pruned_loss=0.09583, over 5706759.82 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3419, pruned_loss=0.0897, over 3206492.35 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3506, pruned_loss=0.09651, over 5693189.76 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:18:24,124 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868674.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:18:27,328 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:18:36,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:18:43,542 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:18:45,686 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:18:52,233 INFO [train.py:968] (0/2) Epoch 20, batch 1550, giga_loss[loss=0.2373, simple_loss=0.3275, pruned_loss=0.07355, over 28719.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3487, pruned_loss=0.094, over 5701747.64 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3419, pruned_loss=0.08971, over 3290480.37 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3488, pruned_loss=0.09468, over 5692524.69 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:19:09,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.205e+02 1.126e+03 1.312e+03 1.804e+03 4.737e+03, threshold=2.624e+03, percent-clipped=7.0
+2023-03-10 05:19:09,342 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:19:28,782 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-10 05:19:35,539 INFO [train.py:968] (0/2) Epoch 20, batch 1600, giga_loss[loss=0.274, simple_loss=0.3501, pruned_loss=0.09897, over 28904.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3481, pruned_loss=0.0942, over 5710488.77 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3418, pruned_loss=0.08961, over 3368008.01 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3484, pruned_loss=0.09492, over 5697635.49 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:20:06,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:20:20,583 INFO [train.py:968] (0/2) Epoch 20, batch 1650, giga_loss[loss=0.2876, simple_loss=0.3633, pruned_loss=0.1059, over 29020.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3504, pruned_loss=0.09823, over 5716469.43 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.342, pruned_loss=0.08988, over 3443252.01 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3507, pruned_loss=0.09885, over 5701260.16 frames. ], batch size: 164, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:20:29,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:20:39,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.987e+02 1.313e+03 1.566e+03 2.236e+03 6.196e+03, threshold=3.133e+03, percent-clipped=14.0
+2023-03-10 05:20:41,133 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:20:43,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:20:59,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:21:05,225 INFO [train.py:968] (0/2) Epoch 20, batch 1700, giga_loss[loss=0.3198, simple_loss=0.3779, pruned_loss=0.1308, over 28743.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3525, pruned_loss=0.1017, over 5701411.71 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3429, pruned_loss=0.09043, over 3481190.49 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3524, pruned_loss=0.1021, over 5694892.23 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:21:09,471 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:21:31,754 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:21:49,680 INFO [train.py:968] (0/2) Epoch 20, batch 1750, giga_loss[loss=0.271, simple_loss=0.343, pruned_loss=0.09951, over 28950.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3523, pruned_loss=0.1032, over 5695925.08 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3429, pruned_loss=0.09024, over 3540963.41 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3525, pruned_loss=0.1039, over 5686259.54 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:21:59,477 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 05:22:09,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.850e+02 1.254e+03 1.507e+03 2.131e+03 9.976e+03, threshold=3.014e+03, percent-clipped=15.0
+2023-03-10 05:22:16,769 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:22:19,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:22:33,481 INFO [train.py:968] (0/2) Epoch 20, batch 1800, giga_loss[loss=0.2451, simple_loss=0.3201, pruned_loss=0.08499, over 28740.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3493, pruned_loss=0.1016, over 5708521.59 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3429, pruned_loss=0.0905, over 3587449.62 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3496, pruned_loss=0.1023, over 5697662.95 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:22:36,116 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:22:38,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:22:43,631 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:03,833 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:04,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:06,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:16,894 INFO [train.py:968] (0/2) Epoch 20, batch 1850, giga_loss[loss=0.2668, simple_loss=0.345, pruned_loss=0.0943, over 28867.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3486, pruned_loss=0.1008, over 5712415.68 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.343, pruned_loss=0.09035, over 3631327.72 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3489, pruned_loss=0.1016, over 5702274.65 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:23:30,204 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:33,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.199e+02 1.241e+03 1.554e+03 2.012e+03 5.923e+03, threshold=3.109e+03, percent-clipped=10.0
+2023-03-10 05:23:35,180 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=869034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:37,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=869037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:45,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 05:23:47,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=869049.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:23:50,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=869054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:56,389 INFO [train.py:968] (0/2) Epoch 20, batch 1900, giga_loss[loss=0.2636, simple_loss=0.3435, pruned_loss=0.0918, over 28685.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3478, pruned_loss=0.09937, over 5715515.72 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3429, pruned_loss=0.09004, over 3703499.76 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3483, pruned_loss=0.1005, over 5707052.86 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:24:01,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:24:47,514 INFO [train.py:968] (0/2) Epoch 20, batch 1950, libri_loss[loss=0.2153, simple_loss=0.3036, pruned_loss=0.06343, over 29681.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3455, pruned_loss=0.09784, over 5698441.51 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3426, pruned_loss=0.08976, over 3758038.77 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3461, pruned_loss=0.09911, over 5687324.43 frames. ], batch size: 73, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:25:04,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.685e+02 1.086e+03 1.261e+03 1.736e+03 5.611e+03, threshold=2.523e+03, percent-clipped=4.0
+2023-03-10 05:25:33,113 INFO [train.py:968] (0/2) Epoch 20, batch 2000, libri_loss[loss=0.2939, simple_loss=0.3725, pruned_loss=0.1077, over 29676.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3404, pruned_loss=0.09499, over 5697702.09 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.343, pruned_loss=0.09003, over 3808109.19 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3407, pruned_loss=0.09598, over 5686715.78 frames. ], batch size: 88, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:26:03,289 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=869192.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:26:06,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=869195.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:26:07,721 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=869197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:26:11,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=869200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:26:21,347 INFO [train.py:968] (0/2) Epoch 20, batch 2050, giga_loss[loss=0.2431, simple_loss=0.3171, pruned_loss=0.0846, over 28947.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3343, pruned_loss=0.09202, over 5688479.86 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3428, pruned_loss=0.09, over 3849620.77 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3346, pruned_loss=0.09289, over 5676246.27 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:26:34,096 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-10 05:26:35,386 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869224.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:26:38,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:26:42,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.160e+02 9.932e+02 1.270e+03 1.970e+03 1.036e+04, threshold=2.540e+03, percent-clipped=12.0
+2023-03-10 05:27:13,003 INFO [train.py:968] (0/2) Epoch 20, batch 2100, giga_loss[loss=0.2732, simple_loss=0.3471, pruned_loss=0.09962, over 28697.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3304, pruned_loss=0.08955, over 5687819.51 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3428, pruned_loss=0.08981, over 3886684.80 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3303, pruned_loss=0.09036, over 5677326.04 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:27:49,717 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=869304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:27:55,414 INFO [train.py:968] (0/2) Epoch 20, batch 2150, giga_loss[loss=0.2687, simple_loss=0.3426, pruned_loss=0.09741, over 29043.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3318, pruned_loss=0.09011, over 5677547.44 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3428, pruned_loss=0.09005, over 3917141.56 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3316, pruned_loss=0.09062, over 5680589.63 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:28:11,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.897e+02 1.006e+03 1.245e+03 1.810e+03 5.554e+03, threshold=2.490e+03, percent-clipped=12.0
+2023-03-10 05:28:19,137 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 05:28:35,546 INFO [train.py:968] (0/2) Epoch 20, batch 2200, giga_loss[loss=0.2317, simple_loss=0.313, pruned_loss=0.07524, over 29036.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3324, pruned_loss=0.09009, over 5694590.55 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3427, pruned_loss=0.08989, over 3985438.74 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3319, pruned_loss=0.09062, over 5690496.86 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:29:17,559 INFO [train.py:968] (0/2) Epoch 20, batch 2250, giga_loss[loss=0.2515, simple_loss=0.3265, pruned_loss=0.08826, over 28840.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3319, pruned_loss=0.08958, over 5696116.15 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.08986, over 4050416.37 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3308, pruned_loss=0.09002, over 5687373.19 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:29:17,833 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=869411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:29:35,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.837e+02 1.101e+03 1.280e+03 1.522e+03 3.298e+03, threshold=2.559e+03, percent-clipped=6.0
+2023-03-10 05:29:59,262 INFO [train.py:968] (0/2) Epoch 20, batch 2300, giga_loss[loss=0.2819, simple_loss=0.3506, pruned_loss=0.1066, over 28048.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3295, pruned_loss=0.08887, over 5705903.36 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3434, pruned_loss=0.08984, over 4068775.03 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3285, pruned_loss=0.08923, over 5697356.41 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:30:42,425 INFO [train.py:968] (0/2) Epoch 20, batch 2350, giga_loss[loss=0.2484, simple_loss=0.3185, pruned_loss=0.08918, over 28907.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3283, pruned_loss=0.08863, over 5701742.31 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3439, pruned_loss=0.09018, over 4100860.18 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3268, pruned_loss=0.08864, over 5703531.29 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:30:52,470 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=869524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:30:59,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.098e+02 9.884e+02 1.185e+03 1.682e+03 5.679e+03, threshold=2.370e+03, percent-clipped=7.0
+2023-03-10 05:31:25,428 INFO [train.py:968] (0/2) Epoch 20, batch 2400, giga_loss[loss=0.2274, simple_loss=0.301, pruned_loss=0.07693, over 28428.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3251, pruned_loss=0.08712, over 5709938.58 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.344, pruned_loss=0.09006, over 4126854.93 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3236, pruned_loss=0.08718, over 5709578.56 frames. ], batch size: 78, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:31:32,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6956, 4.5244, 4.3026, 2.1556], device='cuda:0'), covar=tensor([0.0508, 0.0623, 0.0598, 0.1986], device='cuda:0'), in_proj_covar=tensor([0.1171, 0.1092, 0.0924, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 05:31:56,114 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-10 05:32:05,921 INFO [train.py:968] (0/2) Epoch 20, batch 2450, giga_loss[loss=0.2706, simple_loss=0.3363, pruned_loss=0.1025, over 28689.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3242, pruned_loss=0.08732, over 5717052.25 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3447, pruned_loss=0.09043, over 4151533.96 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3223, pruned_loss=0.08707, over 5715866.77 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:32:22,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.981e+02 1.041e+03 1.298e+03 1.672e+03 3.324e+03, threshold=2.596e+03, percent-clipped=10.0
+2023-03-10 05:32:25,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6594, 1.6931, 1.3799, 1.2949], device='cuda:0'), covar=tensor([0.0936, 0.0644, 0.1017, 0.1197], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0445, 0.0514, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 05:32:43,045 INFO [train.py:968] (0/2) Epoch 20, batch 2500, giga_loss[loss=0.2473, simple_loss=0.3208, pruned_loss=0.0869, over 28591.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.322, pruned_loss=0.08606, over 5727773.39 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3452, pruned_loss=0.09072, over 4203715.57 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3195, pruned_loss=0.08554, over 5721688.27 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:32:58,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=869679.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:33:25,783 INFO [train.py:968] (0/2) Epoch 20, batch 2550, giga_loss[loss=0.2347, simple_loss=0.2998, pruned_loss=0.08484, over 28684.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.32, pruned_loss=0.08533, over 5719660.37 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3457, pruned_loss=0.09093, over 4229249.29 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3173, pruned_loss=0.08468, over 5712103.77 frames. ], batch size: 85, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:33:33,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5197, 2.1917, 1.6958, 0.7766], device='cuda:0'), covar=tensor([0.5442, 0.2708, 0.3885, 0.6000], device='cuda:0'), in_proj_covar=tensor([0.1690, 0.1594, 0.1564, 0.1383], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 05:33:43,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.981e+02 1.080e+03 1.442e+03 2.064e+03 5.389e+03, threshold=2.885e+03, percent-clipped=13.0
+2023-03-10 05:33:50,675 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=869743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:34:05,985 INFO [train.py:968] (0/2) Epoch 20, batch 2600, giga_loss[loss=0.222, simple_loss=0.3044, pruned_loss=0.06984, over 28884.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3185, pruned_loss=0.08436, over 5726388.45 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3461, pruned_loss=0.09104, over 4253390.29 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3158, pruned_loss=0.0837, over 5718880.05 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:34:28,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=869786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:34:47,380 INFO [train.py:968] (0/2) Epoch 20, batch 2650, giga_loss[loss=0.2214, simple_loss=0.2971, pruned_loss=0.07282, over 28736.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3168, pruned_loss=0.08291, over 5726350.66 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3467, pruned_loss=0.09096, over 4293478.77 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3136, pruned_loss=0.08227, over 5717016.93 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:34:57,187 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=869822.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:34:59,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=869825.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:35:06,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.715e+02 9.655e+02 1.166e+03 1.459e+03 5.336e+03, threshold=2.333e+03, percent-clipped=4.0
+2023-03-10 05:35:24,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869854.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:35:30,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0228, 3.1861, 2.2103, 1.1078], device='cuda:0'), covar=tensor([0.7285, 0.2339, 0.3858, 0.6860], device='cuda:0'), in_proj_covar=tensor([0.1689, 0.1593, 0.1565, 0.1382], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 05:35:31,694 INFO [train.py:968] (0/2) Epoch 20, batch 2700, libri_loss[loss=0.2818, simple_loss=0.369, pruned_loss=0.09732, over 29500.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3193, pruned_loss=0.08456, over 5717338.71 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3477, pruned_loss=0.09136, over 4338169.65 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.3152, pruned_loss=0.08357, over 5707667.74 frames. ], batch size: 85, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:35:41,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4372, 2.1397, 1.6566, 0.6719], device='cuda:0'), covar=tensor([0.5316, 0.2702, 0.4198, 0.6060], device='cuda:0'), in_proj_covar=tensor([0.1692, 0.1596, 0.1567, 0.1384], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 05:35:46,422 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 05:36:06,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=869899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:36:18,291 INFO [train.py:968] (0/2) Epoch 20, batch 2750, giga_loss[loss=0.2518, simple_loss=0.3281, pruned_loss=0.08769, over 28889.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3231, pruned_loss=0.0867, over 5716647.53 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3478, pruned_loss=0.09125, over 4353397.88 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3196, pruned_loss=0.08595, over 5707777.71 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:36:34,100 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=869929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:36:36,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=869932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:36:38,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.167e+02 1.093e+03 1.358e+03 1.740e+03 7.340e+03, threshold=2.716e+03, percent-clipped=11.0
+2023-03-10 05:36:40,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9071, 2.1091, 1.7375, 2.0281], device='cuda:0'), covar=tensor([0.2464, 0.2559, 0.2868, 0.2508], device='cuda:0'), in_proj_covar=tensor([0.1477, 0.1070, 0.1310, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 05:37:04,802 INFO [train.py:968] (0/2) Epoch 20, batch 2800, giga_loss[loss=0.3805, simple_loss=0.4289, pruned_loss=0.1661, over 27637.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3291, pruned_loss=0.09067, over 5704741.49 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3475, pruned_loss=0.09087, over 4395510.10 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3259, pruned_loss=0.09023, over 5696057.09 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:37:05,088 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:37:38,367 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-870000.pt
+2023-03-10 05:37:48,384 INFO [train.py:968] (0/2) Epoch 20, batch 2850, giga_loss[loss=0.3042, simple_loss=0.3811, pruned_loss=0.1136, over 28973.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3365, pruned_loss=0.09511, over 5694193.66 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3475, pruned_loss=0.09079, over 4444939.45 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3334, pruned_loss=0.09489, over 5689169.01 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:38:10,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.476e+02 1.277e+03 1.595e+03 2.267e+03 6.215e+03, threshold=3.191e+03, percent-clipped=14.0
+2023-03-10 05:38:20,471 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=870042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:38:22,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=870045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:38:37,562 INFO [train.py:968] (0/2) Epoch 20, batch 2900, giga_loss[loss=0.287, simple_loss=0.3462, pruned_loss=0.1139, over 23745.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3414, pruned_loss=0.09706, over 5686668.22 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3472, pruned_loss=0.09063, over 4466982.52 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.339, pruned_loss=0.0971, over 5679331.57 frames. ], batch size: 710, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:38:53,390 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=870074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:39:24,984 INFO [train.py:968] (0/2) Epoch 20, batch 2950, giga_loss[loss=0.3104, simple_loss=0.3897, pruned_loss=0.1156, over 28609.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3462, pruned_loss=0.09907, over 5678320.10 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3466, pruned_loss=0.09027, over 4506133.81 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3447, pruned_loss=0.09958, over 5669272.30 frames. ], batch size: 336, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:39:30,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=870118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:39:45,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.554e+02 1.088e+03 1.302e+03 1.700e+03 3.082e+03, threshold=2.605e+03, percent-clipped=0.0
+2023-03-10 05:39:46,618 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5995, 1.9773, 1.6030, 1.6178], device='cuda:0'), covar=tensor([0.0786, 0.0278, 0.0320, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 05:40:12,112 INFO [train.py:968] (0/2) Epoch 20, batch 3000, giga_loss[loss=0.3252, simple_loss=0.3906, pruned_loss=0.1299, over 29101.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3516, pruned_loss=0.1018, over 5691448.60 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3461, pruned_loss=0.09004, over 4546065.71 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3508, pruned_loss=0.1026, over 5679790.27 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:40:12,117 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 05:40:20,896 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2119, simple_loss=0.3184, pruned_loss=0.05275, over 944034.00 frames. 
+2023-03-10 05:40:20,896 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 05:40:24,436 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=870166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:40:56,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5543, 1.8045, 1.4655, 1.4039], device='cuda:0'), covar=tensor([0.2724, 0.2712, 0.2999, 0.2362], device='cuda:0'), in_proj_covar=tensor([0.1471, 0.1067, 0.1305, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 05:41:05,821 INFO [train.py:968] (0/2) Epoch 20, batch 3050, libri_loss[loss=0.2759, simple_loss=0.3665, pruned_loss=0.09267, over 29674.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3521, pruned_loss=0.1022, over 5683045.58 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3459, pruned_loss=0.08999, over 4586572.39 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3518, pruned_loss=0.1033, over 5667369.40 frames. ], batch size: 91, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:41:24,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.597e+02 1.221e+03 1.582e+03 2.082e+03 6.018e+03, threshold=3.164e+03, percent-clipped=14.0
+2023-03-10 05:41:41,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2998, 1.4312, 1.4183, 1.3222], device='cuda:0'), covar=tensor([0.2290, 0.2075, 0.1909, 0.2080], device='cuda:0'), in_proj_covar=tensor([0.1911, 0.1840, 0.1778, 0.1913], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 05:41:41,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-10 05:41:45,984 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 05:41:47,082 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=870260.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:41:47,490 INFO [train.py:968] (0/2) Epoch 20, batch 3100, giga_loss[loss=0.2665, simple_loss=0.336, pruned_loss=0.09849, over 28581.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3474, pruned_loss=0.09867, over 5690592.61 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3456, pruned_loss=0.08991, over 4612897.28 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3474, pruned_loss=0.09976, over 5674263.48 frames. ], batch size: 85, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:41:47,776 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=870261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:41:50,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=870264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:42:06,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1315, 1.2736, 3.8031, 3.2326], device='cuda:0'), covar=tensor([0.1865, 0.2914, 0.0440, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0631, 0.0925, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 05:42:15,707 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=870293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:42:34,343 INFO [train.py:968] (0/2) Epoch 20, batch 3150, giga_loss[loss=0.2679, simple_loss=0.3445, pruned_loss=0.09562, over 28727.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3461, pruned_loss=0.09727, over 5686030.49 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3453, pruned_loss=0.08975, over 4636125.72 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3463, pruned_loss=0.09839, over 5671521.59 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:42:44,024 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-10 05:42:52,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.992e+02 1.099e+03 1.454e+03 2.038e+03 6.442e+03, threshold=2.908e+03, percent-clipped=8.0
+2023-03-10 05:43:01,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9142, 1.0993, 1.0195, 0.7970], device='cuda:0'), covar=tensor([0.2405, 0.2797, 0.1667, 0.2367], device='cuda:0'), in_proj_covar=tensor([0.1908, 0.1837, 0.1779, 0.1911], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 05:43:19,724 INFO [train.py:968] (0/2) Epoch 20, batch 3200, giga_loss[loss=0.2617, simple_loss=0.3422, pruned_loss=0.0906, over 28996.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3453, pruned_loss=0.09677, over 5683431.15 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3449, pruned_loss=0.08957, over 4648900.92 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3458, pruned_loss=0.09783, over 5670109.64 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:43:41,232 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-10 05:44:04,884 INFO [train.py:968] (0/2) Epoch 20, batch 3250, giga_loss[loss=0.2874, simple_loss=0.3691, pruned_loss=0.1029, over 28848.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3474, pruned_loss=0.09789, over 5684264.30 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3448, pruned_loss=0.08953, over 4673736.20 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3479, pruned_loss=0.09892, over 5669626.31 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:44:26,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.116e+02 1.198e+03 1.533e+03 2.006e+03 5.224e+03, threshold=3.067e+03, percent-clipped=9.0
+2023-03-10 05:44:36,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8852, 3.7221, 3.4759, 1.8757], device='cuda:0'), covar=tensor([0.0641, 0.0740, 0.0694, 0.2307], device='cuda:0'), in_proj_covar=tensor([0.1169, 0.1087, 0.0920, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 05:44:48,839 INFO [train.py:968] (0/2) Epoch 20, batch 3300, giga_loss[loss=0.2626, simple_loss=0.3432, pruned_loss=0.09106, over 28341.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3487, pruned_loss=0.09877, over 5698632.53 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3442, pruned_loss=0.08927, over 4697799.63 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3496, pruned_loss=0.09995, over 5683444.27 frames. ], batch size: 65, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:45:32,682 INFO [train.py:968] (0/2) Epoch 20, batch 3350, giga_loss[loss=0.2482, simple_loss=0.3238, pruned_loss=0.08631, over 28297.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3498, pruned_loss=0.09976, over 5688920.20 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3439, pruned_loss=0.08905, over 4723729.41 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3509, pruned_loss=0.1012, over 5680153.18 frames. ], batch size: 77, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:45:50,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.580e+02 1.300e+03 1.674e+03 2.279e+03 7.458e+03, threshold=3.348e+03, percent-clipped=12.0
+2023-03-10 05:45:55,990 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=870541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:46:14,259 INFO [train.py:968] (0/2) Epoch 20, batch 3400, giga_loss[loss=0.2903, simple_loss=0.3551, pruned_loss=0.1127, over 29031.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.351, pruned_loss=0.1008, over 5695167.29 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3442, pruned_loss=0.0891, over 4774454.34 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3519, pruned_loss=0.1024, over 5680801.08 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:47:00,198 INFO [train.py:968] (0/2) Epoch 20, batch 3450, giga_loss[loss=0.3285, simple_loss=0.384, pruned_loss=0.1365, over 27578.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3524, pruned_loss=0.1025, over 5689092.66 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3442, pruned_loss=0.08903, over 4791399.09 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3533, pruned_loss=0.104, over 5674762.34 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:47:00,474 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5608, 2.6557, 2.7304, 2.4990], device='cuda:0'), covar=tensor([0.1969, 0.2415, 0.1950, 0.2045], device='cuda:0'), in_proj_covar=tensor([0.0472, 0.0741, 0.0709, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 05:47:20,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=870635.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:47:20,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.266e+02 1.210e+03 1.352e+03 1.618e+03 4.305e+03, threshold=2.704e+03, percent-clipped=1.0
+2023-03-10 05:47:39,432 INFO [train.py:968] (0/2) Epoch 20, batch 3500, giga_loss[loss=0.2717, simple_loss=0.3506, pruned_loss=0.09639, over 28864.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3522, pruned_loss=0.1013, over 5691704.37 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3451, pruned_loss=0.08961, over 4820953.73 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3524, pruned_loss=0.1025, over 5681039.05 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:47:58,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=870684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:48:00,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=870687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:48:21,520 INFO [train.py:968] (0/2) Epoch 20, batch 3550, giga_loss[loss=0.3054, simple_loss=0.3834, pruned_loss=0.1137, over 28898.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.352, pruned_loss=0.1, over 5696731.62 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3453, pruned_loss=0.08974, over 4837259.74 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3521, pruned_loss=0.101, over 5685450.01 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:48:26,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=870716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:48:42,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.155e+02 1.101e+03 1.359e+03 1.673e+03 3.449e+03, threshold=2.718e+03, percent-clipped=5.0
+2023-03-10 05:49:02,323 INFO [train.py:968] (0/2) Epoch 20, batch 3600, giga_loss[loss=0.2756, simple_loss=0.3527, pruned_loss=0.0992, over 28944.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3525, pruned_loss=0.09959, over 5699780.78 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3457, pruned_loss=0.08988, over 4866568.84 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3524, pruned_loss=0.1005, over 5687306.72 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:49:17,426 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=870778.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:49:21,201 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=870781.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:49:21,999 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 05:49:42,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=870810.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:49:43,224 INFO [train.py:968] (0/2) Epoch 20, batch 3650, giga_loss[loss=0.2735, simple_loss=0.3463, pruned_loss=0.1004, over 28584.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3509, pruned_loss=0.09878, over 5705528.28 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3455, pruned_loss=0.08984, over 4885880.74 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3512, pruned_loss=0.09971, over 5693028.66 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:49:59,255 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=870830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:50:03,527 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.503e+02 1.182e+03 1.464e+03 1.918e+03 5.152e+03, threshold=2.928e+03, percent-clipped=7.0
+2023-03-10 05:50:25,550 INFO [train.py:968] (0/2) Epoch 20, batch 3700, giga_loss[loss=0.2156, simple_loss=0.3039, pruned_loss=0.06364, over 28410.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3478, pruned_loss=0.09767, over 5698852.87 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3454, pruned_loss=0.08991, over 4913296.80 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3482, pruned_loss=0.09861, over 5685942.14 frames. ], batch size: 65, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:50:27,655 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=870864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:50:44,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2380, 2.8574, 2.4751, 1.8352], device='cuda:0'), covar=tensor([0.2751, 0.1672, 0.2017, 0.2546], device='cuda:0'), in_proj_covar=tensor([0.1903, 0.1840, 0.1772, 0.1906], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 05:51:02,114 INFO [train.py:968] (0/2) Epoch 20, batch 3750, libri_loss[loss=0.2332, simple_loss=0.3225, pruned_loss=0.072, over 29523.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.344, pruned_loss=0.09518, over 5711401.16 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3446, pruned_loss=0.08932, over 4947552.30 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3451, pruned_loss=0.09661, over 5694768.68 frames. ], batch size: 80, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:51:05,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4418, 2.4809, 1.9067, 1.9648], device='cuda:0'), covar=tensor([0.0897, 0.0687, 0.0959, 0.1116], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0444, 0.0515, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 05:51:22,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.765e+02 1.055e+03 1.236e+03 1.475e+03 5.422e+03, threshold=2.472e+03, percent-clipped=1.0
+2023-03-10 05:51:47,412 INFO [train.py:968] (0/2) Epoch 20, batch 3800, giga_loss[loss=0.2877, simple_loss=0.3624, pruned_loss=0.1065, over 28724.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3445, pruned_loss=0.09622, over 5705289.24 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3442, pruned_loss=0.08914, over 4956989.71 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3456, pruned_loss=0.09755, over 5690542.89 frames. ], batch size: 60, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:51:55,309 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8439, 1.9380, 2.0549, 1.6536], device='cuda:0'), covar=tensor([0.1860, 0.2505, 0.1439, 0.1712], device='cuda:0'), in_proj_covar=tensor([0.0894, 0.0703, 0.0941, 0.0837], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 05:52:29,789 INFO [train.py:968] (0/2) Epoch 20, batch 3850, giga_loss[loss=0.2766, simple_loss=0.3418, pruned_loss=0.1058, over 28766.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3446, pruned_loss=0.09628, over 5708537.76 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3434, pruned_loss=0.08875, over 4985757.48 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3461, pruned_loss=0.0979, over 5693828.87 frames. ], batch size: 78, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:52:47,042 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-10 05:52:48,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.849e+02 1.126e+03 1.451e+03 1.854e+03 3.611e+03, threshold=2.901e+03, percent-clipped=12.0
+2023-03-10 05:53:11,084 INFO [train.py:968] (0/2) Epoch 20, batch 3900, giga_loss[loss=0.2487, simple_loss=0.3352, pruned_loss=0.08112, over 28960.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3452, pruned_loss=0.09593, over 5710526.86 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3437, pruned_loss=0.0889, over 5000150.71 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3462, pruned_loss=0.09723, over 5698916.40 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:53:21,475 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 05:53:55,429 INFO [train.py:968] (0/2) Epoch 20, batch 3950, giga_loss[loss=0.2896, simple_loss=0.3627, pruned_loss=0.1083, over 28547.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.345, pruned_loss=0.09522, over 5715599.61 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3433, pruned_loss=0.08869, over 5013878.66 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3461, pruned_loss=0.0965, over 5703770.65 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:54:10,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4803, 1.6044, 1.7009, 1.3000], device='cuda:0'), covar=tensor([0.1825, 0.2582, 0.1490, 0.1802], device='cuda:0'), in_proj_covar=tensor([0.0893, 0.0702, 0.0940, 0.0836], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 05:54:11,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1629, 1.1507, 3.6612, 3.1153], device='cuda:0'), covar=tensor([0.1726, 0.2850, 0.0443, 0.0926], device='cuda:0'), in_proj_covar=tensor([0.0735, 0.0628, 0.0926, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 05:54:16,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.896e+02 1.058e+03 1.449e+03 1.776e+03 5.832e+03, threshold=2.897e+03, percent-clipped=7.0
+2023-03-10 05:54:17,004 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=871136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:54:39,436 INFO [train.py:968] (0/2) Epoch 20, batch 4000, giga_loss[loss=0.3142, simple_loss=0.3761, pruned_loss=0.1261, over 27642.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.345, pruned_loss=0.09578, over 5707229.18 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3434, pruned_loss=0.08868, over 5018030.26 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3458, pruned_loss=0.09682, over 5697383.08 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:55:01,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4617, 1.5835, 1.2850, 1.1288], device='cuda:0'), covar=tensor([0.1016, 0.0579, 0.1074, 0.1162], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0444, 0.0515, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 05:55:05,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4311, 1.7471, 1.5413, 1.6350], device='cuda:0'), covar=tensor([0.0786, 0.0297, 0.0313, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 05:55:14,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=871205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:55:19,261 INFO [train.py:968] (0/2) Epoch 20, batch 4050, giga_loss[loss=0.2633, simple_loss=0.3429, pruned_loss=0.09181, over 28358.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3432, pruned_loss=0.09517, over 5715631.72 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3433, pruned_loss=0.08861, over 5031259.66 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3439, pruned_loss=0.09612, over 5705429.11 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:55:37,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.047e+02 1.058e+03 1.306e+03 1.700e+03 6.292e+03, threshold=2.611e+03, percent-clipped=10.0
+2023-03-10 05:55:41,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=871239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:55:59,157 INFO [train.py:968] (0/2) Epoch 20, batch 4100, giga_loss[loss=0.2466, simple_loss=0.3263, pruned_loss=0.08344, over 29109.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3408, pruned_loss=0.09416, over 5716718.11 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3433, pruned_loss=0.08863, over 5044260.69 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3414, pruned_loss=0.09498, over 5706325.19 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:56:39,654 INFO [train.py:968] (0/2) Epoch 20, batch 4150, giga_loss[loss=0.2372, simple_loss=0.3166, pruned_loss=0.07884, over 28801.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3375, pruned_loss=0.09239, over 5718503.97 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3429, pruned_loss=0.0884, over 5056929.21 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3382, pruned_loss=0.09328, over 5707965.72 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:56:59,239 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.927e+02 1.144e+03 1.416e+03 1.805e+03 6.498e+03, threshold=2.832e+03, percent-clipped=7.0
+2023-03-10 05:57:06,788 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-10 05:57:08,234 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=871348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:57:11,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=871351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:57:17,789 INFO [train.py:968] (0/2) Epoch 20, batch 4200, giga_loss[loss=0.2534, simple_loss=0.3316, pruned_loss=0.08762, over 28197.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3377, pruned_loss=0.09236, over 5723478.66 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3428, pruned_loss=0.08817, over 5097745.73 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3382, pruned_loss=0.09346, over 5706921.42 frames. ], batch size: 77, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:57:33,440 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=871380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:57:37,378 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=871382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:57:39,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=871385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:57:53,653 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-10 05:57:57,335 INFO [train.py:968] (0/2) Epoch 20, batch 4250, giga_loss[loss=0.2786, simple_loss=0.35, pruned_loss=0.1036, over 28826.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3373, pruned_loss=0.09286, over 5726770.97 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3431, pruned_loss=0.08859, over 5125610.39 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3372, pruned_loss=0.09355, over 5707614.84 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:58:01,989 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=871414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:58:20,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.618e+02 1.112e+03 1.431e+03 2.070e+03 5.881e+03, threshold=2.863e+03, percent-clipped=11.0
+2023-03-10 05:58:39,876 INFO [train.py:968] (0/2) Epoch 20, batch 4300, giga_loss[loss=0.2694, simple_loss=0.3424, pruned_loss=0.09821, over 28916.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3368, pruned_loss=0.09284, over 5724587.14 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3437, pruned_loss=0.08897, over 5159846.01 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3359, pruned_loss=0.09325, over 5701694.32 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:58:46,417 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3271, 1.5121, 3.7443, 3.1404], device='cuda:0'), covar=tensor([0.1547, 0.2475, 0.0407, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0630, 0.0928, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 05:58:56,769 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=871484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:59:19,098 INFO [train.py:968] (0/2) Epoch 20, batch 4350, giga_loss[loss=0.2311, simple_loss=0.3132, pruned_loss=0.07453, over 28858.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3345, pruned_loss=0.09194, over 5717595.02 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3438, pruned_loss=0.08911, over 5167315.65 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3335, pruned_loss=0.09222, over 5704923.79 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:59:19,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=871511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:59:38,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.064e+02 1.124e+03 1.422e+03 2.205e+03 5.422e+03, threshold=2.844e+03, percent-clipped=10.0
+2023-03-10 05:59:57,909 INFO [train.py:968] (0/2) Epoch 20, batch 4400, giga_loss[loss=0.2488, simple_loss=0.3187, pruned_loss=0.08948, over 28739.00 frames. ], tot_loss[loss=0.258, simple_loss=0.333, pruned_loss=0.09156, over 5720167.96 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.344, pruned_loss=0.08925, over 5189280.19 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3317, pruned_loss=0.09173, over 5704854.13 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:00:38,257 INFO [train.py:968] (0/2) Epoch 20, batch 4450, giga_loss[loss=0.3428, simple_loss=0.406, pruned_loss=0.1398, over 27765.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.333, pruned_loss=0.09136, over 5719796.38 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3435, pruned_loss=0.089, over 5199924.50 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3323, pruned_loss=0.09173, over 5705231.28 frames. ], batch size: 474, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:01:01,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.573e+02 1.136e+03 1.397e+03 2.033e+03 4.347e+03, threshold=2.794e+03, percent-clipped=8.0
+2023-03-10 06:01:16,188 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=871654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:01:20,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=871657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:01:24,473 INFO [train.py:968] (0/2) Epoch 20, batch 4500, giga_loss[loss=0.2391, simple_loss=0.3187, pruned_loss=0.07979, over 28747.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3352, pruned_loss=0.09234, over 5718134.74 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3436, pruned_loss=0.08906, over 5221191.22 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3342, pruned_loss=0.09269, over 5703078.49 frames. ], batch size: 60, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:01:44,153 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=871686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:02:03,364 INFO [train.py:968] (0/2) Epoch 20, batch 4550, giga_loss[loss=0.2639, simple_loss=0.3492, pruned_loss=0.08928, over 28805.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3373, pruned_loss=0.09278, over 5720918.21 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3434, pruned_loss=0.08918, over 5245524.33 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3362, pruned_loss=0.09312, over 5710930.44 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:02:10,862 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-10 06:02:25,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.239e+02 1.162e+03 1.382e+03 1.706e+03 4.914e+03, threshold=2.764e+03, percent-clipped=7.0
+2023-03-10 06:02:44,351 INFO [train.py:968] (0/2) Epoch 20, batch 4600, giga_loss[loss=0.2558, simple_loss=0.3376, pruned_loss=0.08698, over 28660.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3391, pruned_loss=0.09308, over 5723207.88 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3434, pruned_loss=0.08926, over 5264034.96 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3381, pruned_loss=0.09337, over 5711278.77 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:03:28,969 INFO [train.py:968] (0/2) Epoch 20, batch 4650, giga_loss[loss=0.2299, simple_loss=0.3114, pruned_loss=0.07419, over 28792.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3393, pruned_loss=0.09255, over 5708705.67 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3434, pruned_loss=0.0894, over 5269564.45 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3383, pruned_loss=0.09274, over 5703980.66 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:03:53,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.008e+02 1.137e+03 1.517e+03 2.213e+03 7.200e+03, threshold=3.034e+03, percent-clipped=15.0
+2023-03-10 06:04:09,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=871859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:04:09,964 INFO [train.py:968] (0/2) Epoch 20, batch 4700, giga_loss[loss=0.2504, simple_loss=0.3297, pruned_loss=0.08557, over 28973.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3388, pruned_loss=0.09189, over 5706793.02 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3435, pruned_loss=0.08957, over 5299349.20 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3378, pruned_loss=0.09201, over 5694804.71 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:04:22,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=871874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:04:34,847 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=871890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:04:51,551 INFO [train.py:968] (0/2) Epoch 20, batch 4750, giga_loss[loss=0.2956, simple_loss=0.3498, pruned_loss=0.1207, over 28814.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3392, pruned_loss=0.09206, over 5710369.00 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08942, over 5313975.11 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3386, pruned_loss=0.09237, over 5696446.27 frames. ], batch size: 66, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:05:12,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.044e+02 1.311e+03 1.681e+03 2.208e+03 4.873e+03, threshold=3.361e+03, percent-clipped=8.0
+2023-03-10 06:05:32,532 INFO [train.py:968] (0/2) Epoch 20, batch 4800, giga_loss[loss=0.2705, simple_loss=0.3481, pruned_loss=0.09641, over 28907.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3408, pruned_loss=0.09303, over 5721506.80 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08923, over 5331205.69 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3404, pruned_loss=0.09354, over 5705408.51 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:06:04,057 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-872000.pt
+2023-03-10 06:06:06,183 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:06:10,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:06:15,021 INFO [train.py:968] (0/2) Epoch 20, batch 4850, giga_loss[loss=0.2895, simple_loss=0.3575, pruned_loss=0.1107, over 28979.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3418, pruned_loss=0.09401, over 5717051.35 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08928, over 5342678.72 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3413, pruned_loss=0.09445, over 5700971.47 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:06:21,053 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-10 06:06:36,268 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:06:40,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.079e+02 1.254e+03 1.455e+03 2.046e+03 6.138e+03, threshold=2.911e+03, percent-clipped=4.0
+2023-03-10 06:07:01,850 INFO [train.py:968] (0/2) Epoch 20, batch 4900, giga_loss[loss=0.2646, simple_loss=0.3397, pruned_loss=0.09477, over 28772.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3446, pruned_loss=0.09586, over 5717252.40 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.0892, over 5345455.32 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3444, pruned_loss=0.0963, over 5703901.96 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:07:22,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4097, 1.5203, 1.4109, 1.5425], device='cuda:0'), covar=tensor([0.0623, 0.0308, 0.0296, 0.0700], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 06:07:44,602 INFO [train.py:968] (0/2) Epoch 20, batch 4950, giga_loss[loss=0.3059, simple_loss=0.3741, pruned_loss=0.1188, over 27607.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3477, pruned_loss=0.09778, over 5719657.74 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08911, over 5355565.70 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.348, pruned_loss=0.09836, over 5706431.50 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:08:08,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.624e+02 1.326e+03 1.569e+03 2.005e+03 3.979e+03, threshold=3.139e+03, percent-clipped=6.0
+2023-03-10 06:08:10,700 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:08:25,577 INFO [train.py:968] (0/2) Epoch 20, batch 5000, giga_loss[loss=0.2971, simple_loss=0.3679, pruned_loss=0.1131, over 28908.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3498, pruned_loss=0.09878, over 5722539.07 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08942, over 5369199.25 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3496, pruned_loss=0.0992, over 5707722.41 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:08:59,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4952, 1.8834, 1.8157, 1.7340], device='cuda:0'), covar=tensor([0.1805, 0.1605, 0.1890, 0.1623], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0740, 0.0708, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 06:09:06,437 INFO [train.py:968] (0/2) Epoch 20, batch 5050, giga_loss[loss=0.2813, simple_loss=0.361, pruned_loss=0.1008, over 28342.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3516, pruned_loss=0.1001, over 5715959.46 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3436, pruned_loss=0.08968, over 5383938.80 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3512, pruned_loss=0.1005, over 5700080.41 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:09:27,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.431e+02 1.257e+03 1.563e+03 2.114e+03 5.669e+03, threshold=3.126e+03, percent-clipped=7.0
+2023-03-10 06:09:36,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:09:46,250 INFO [train.py:968] (0/2) Epoch 20, batch 5100, giga_loss[loss=0.2604, simple_loss=0.3272, pruned_loss=0.09682, over 28497.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3508, pruned_loss=0.09961, over 5718632.44 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3438, pruned_loss=0.08978, over 5396755.72 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3505, pruned_loss=0.1001, over 5702066.78 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:09:50,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:10:27,555 INFO [train.py:968] (0/2) Epoch 20, batch 5150, giga_loss[loss=0.2328, simple_loss=0.3207, pruned_loss=0.07247, over 28651.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3499, pruned_loss=0.09914, over 5722482.05 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3439, pruned_loss=0.08983, over 5411663.50 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3498, pruned_loss=0.09974, over 5704152.16 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:10:51,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.468e+02 1.144e+03 1.389e+03 1.679e+03 3.764e+03, threshold=2.779e+03, percent-clipped=4.0
+2023-03-10 06:11:11,306 INFO [train.py:968] (0/2) Epoch 20, batch 5200, giga_loss[loss=0.2702, simple_loss=0.3424, pruned_loss=0.09903, over 28920.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3468, pruned_loss=0.09819, over 5715470.89 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3439, pruned_loss=0.08981, over 5413821.18 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3467, pruned_loss=0.09871, over 5700595.12 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:11:37,791 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:11:41,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:11:50,898 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:11:52,558 INFO [train.py:968] (0/2) Epoch 20, batch 5250, giga_loss[loss=0.2388, simple_loss=0.3106, pruned_loss=0.0835, over 28934.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3425, pruned_loss=0.0957, over 5721526.75 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3433, pruned_loss=0.08951, over 5428186.95 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3429, pruned_loss=0.09658, over 5704547.71 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:11:53,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:11:59,264 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:12:03,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4272, 3.5571, 1.5569, 1.4860], device='cuda:0'), covar=tensor([0.0961, 0.0299, 0.0959, 0.1348], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0545, 0.0377, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 06:12:03,935 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:12:04,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9310, 2.0662, 1.7218, 2.1027], device='cuda:0'), covar=tensor([0.2410, 0.2630, 0.2942, 0.2520], device='cuda:0'), in_proj_covar=tensor([0.1467, 0.1063, 0.1301, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 06:12:05,462 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-10 06:12:12,016 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:12:14,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.196e+02 1.078e+03 1.384e+03 1.876e+03 4.276e+03, threshold=2.768e+03, percent-clipped=5.0
+2023-03-10 06:12:15,207 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:12:34,109 INFO [train.py:968] (0/2) Epoch 20, batch 5300, giga_loss[loss=0.3522, simple_loss=0.4133, pruned_loss=0.1455, over 27706.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3431, pruned_loss=0.09517, over 5712345.96 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3438, pruned_loss=0.08989, over 5429539.92 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3431, pruned_loss=0.09567, over 5703861.16 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:13:13,669 INFO [train.py:968] (0/2) Epoch 20, batch 5350, giga_loss[loss=0.2784, simple_loss=0.3515, pruned_loss=0.1027, over 28765.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.345, pruned_loss=0.09503, over 5718865.02 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3439, pruned_loss=0.09013, over 5450953.46 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3448, pruned_loss=0.09548, over 5704908.96 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:13:17,168 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:13:35,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.715e+02 1.195e+03 1.350e+03 1.916e+03 4.528e+03, threshold=2.699e+03, percent-clipped=10.0
+2023-03-10 06:13:54,705 INFO [train.py:968] (0/2) Epoch 20, batch 5400, giga_loss[loss=0.2478, simple_loss=0.327, pruned_loss=0.08431, over 29022.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3448, pruned_loss=0.09468, over 5726192.35 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3436, pruned_loss=0.09, over 5461167.82 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.345, pruned_loss=0.09528, over 5713084.05 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:13:58,618 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-10 06:14:22,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:14:29,279 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:14:34,689 INFO [train.py:968] (0/2) Epoch 20, batch 5450, giga_loss[loss=0.2927, simple_loss=0.3589, pruned_loss=0.1132, over 28968.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3428, pruned_loss=0.09492, over 5728854.23 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3432, pruned_loss=0.08967, over 5471961.02 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3434, pruned_loss=0.09582, over 5714387.04 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:14:57,113 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.699e+02 1.257e+03 1.501e+03 1.912e+03 4.735e+03, threshold=3.002e+03, percent-clipped=7.0
+2023-03-10 06:15:15,343 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:15:16,545 INFO [train.py:968] (0/2) Epoch 20, batch 5500, giga_loss[loss=0.2928, simple_loss=0.3639, pruned_loss=0.1108, over 28954.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3416, pruned_loss=0.09532, over 5734623.45 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3431, pruned_loss=0.08968, over 5482845.88 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3421, pruned_loss=0.09617, over 5719022.29 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:15:17,451 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:15:41,711 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:15:58,138 INFO [train.py:968] (0/2) Epoch 20, batch 5550, giga_loss[loss=0.3652, simple_loss=0.4038, pruned_loss=0.1633, over 26765.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3403, pruned_loss=0.09596, over 5737028.97 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3436, pruned_loss=0.08988, over 5492615.18 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3403, pruned_loss=0.09658, over 5721424.01 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:16:18,982 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-10 06:16:22,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.708e+02 1.128e+03 1.382e+03 1.846e+03 4.707e+03, threshold=2.764e+03, percent-clipped=6.0
+2023-03-10 06:16:42,810 INFO [train.py:968] (0/2) Epoch 20, batch 5600, libri_loss[loss=0.2645, simple_loss=0.347, pruned_loss=0.09101, over 29539.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3391, pruned_loss=0.09547, over 5735305.16 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3436, pruned_loss=0.08987, over 5496603.42 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3389, pruned_loss=0.09603, over 5721579.60 frames. ], batch size: 77, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:17:10,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:17:26,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:17:27,035 INFO [train.py:968] (0/2) Epoch 20, batch 5650, giga_loss[loss=0.2265, simple_loss=0.2935, pruned_loss=0.07976, over 28717.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3365, pruned_loss=0.09473, over 5722421.32 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3434, pruned_loss=0.08976, over 5500830.07 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3365, pruned_loss=0.0953, over 5709964.20 frames. ], batch size: 60, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:17:51,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.187e+02 1.149e+03 1.422e+03 1.862e+03 3.966e+03, threshold=2.844e+03, percent-clipped=5.0
+2023-03-10 06:18:07,747 INFO [train.py:968] (0/2) Epoch 20, batch 5700, giga_loss[loss=0.2214, simple_loss=0.2995, pruned_loss=0.07162, over 29051.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.332, pruned_loss=0.09214, over 5725679.88 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3435, pruned_loss=0.08989, over 5509297.18 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3318, pruned_loss=0.09254, over 5712075.26 frames. ], batch size: 164, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:18:22,619 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:18:49,081 INFO [train.py:968] (0/2) Epoch 20, batch 5750, libri_loss[loss=0.2613, simple_loss=0.3426, pruned_loss=0.09003, over 29508.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3304, pruned_loss=0.09145, over 5722330.22 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3441, pruned_loss=0.09015, over 5522409.81 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3292, pruned_loss=0.09157, over 5705822.26 frames. ], batch size: 80, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:19:07,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:09,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:11,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.006e+02 1.178e+03 1.456e+03 2.016e+03 3.944e+03, threshold=2.912e+03, percent-clipped=7.0
+2023-03-10 06:19:22,981 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:25,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:28,912 INFO [train.py:968] (0/2) Epoch 20, batch 5800, giga_loss[loss=0.2405, simple_loss=0.3207, pruned_loss=0.08011, over 28700.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3295, pruned_loss=0.09098, over 5723369.93 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3443, pruned_loss=0.09038, over 5530138.71 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3281, pruned_loss=0.09088, over 5707092.96 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:19:33,740 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:34,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:35,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4265, 1.8108, 1.6741, 1.2243], device='cuda:0'), covar=tensor([0.3662, 0.2470, 0.2927, 0.3249], device='cuda:0'), in_proj_covar=tensor([0.1920, 0.1856, 0.1780, 0.1912], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 06:19:41,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:42,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:46,950 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:20:09,017 INFO [train.py:968] (0/2) Epoch 20, batch 5850, giga_loss[loss=0.2498, simple_loss=0.3316, pruned_loss=0.08397, over 28852.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3329, pruned_loss=0.09249, over 5715282.36 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3447, pruned_loss=0.09067, over 5533910.42 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.331, pruned_loss=0.09219, over 5705181.80 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:20:17,353 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:20:33,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.867e+02 1.319e+03 1.645e+03 2.317e+03 5.414e+03, threshold=3.291e+03, percent-clipped=11.0
+2023-03-10 06:20:51,683 INFO [train.py:968] (0/2) Epoch 20, batch 5900, giga_loss[loss=0.2722, simple_loss=0.3535, pruned_loss=0.09543, over 28654.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3365, pruned_loss=0.09372, over 5715207.90 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3447, pruned_loss=0.09055, over 5541002.36 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3348, pruned_loss=0.09363, over 5704098.79 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:21:02,557 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 06:21:31,333 INFO [train.py:968] (0/2) Epoch 20, batch 5950, giga_loss[loss=0.2834, simple_loss=0.3591, pruned_loss=0.1038, over 28893.00 frames. ], tot_loss[loss=0.265, simple_loss=0.34, pruned_loss=0.095, over 5716285.53 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3445, pruned_loss=0.0905, over 5545950.91 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3386, pruned_loss=0.09511, over 5709102.69 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:21:31,721 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:21:33,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:21:40,546 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:21:42,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:21:56,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.330e+02 1.293e+03 1.627e+03 2.090e+03 1.304e+04, threshold=3.253e+03, percent-clipped=6.0
+2023-03-10 06:21:58,359 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:22:08,528 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:22:14,853 INFO [train.py:968] (0/2) Epoch 20, batch 6000, giga_loss[loss=0.3107, simple_loss=0.3761, pruned_loss=0.1226, over 29009.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3427, pruned_loss=0.09614, over 5717869.60 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3442, pruned_loss=0.09056, over 5556640.89 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3417, pruned_loss=0.09632, over 5706966.96 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:22:14,858 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 06:22:23,594 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2106, simple_loss=0.3172, pruned_loss=0.05204, over 944034.00 frames. 
+2023-03-10 06:22:23,594 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 06:23:07,640 INFO [train.py:968] (0/2) Epoch 20, batch 6050, giga_loss[loss=0.2795, simple_loss=0.3555, pruned_loss=0.1017, over 28909.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3448, pruned_loss=0.09725, over 5715553.55 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3444, pruned_loss=0.09073, over 5564107.60 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3438, pruned_loss=0.0974, over 5704545.82 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:23:18,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3842, 1.7478, 1.5324, 1.6030], device='cuda:0'), covar=tensor([0.0628, 0.0277, 0.0270, 0.0653], device='cuda:0'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0218], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0067, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 06:23:32,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.056e+02 1.332e+03 1.609e+03 2.281e+03 5.164e+03, threshold=3.219e+03, percent-clipped=7.0
+2023-03-10 06:23:42,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:23:52,782 INFO [train.py:968] (0/2) Epoch 20, batch 6100, giga_loss[loss=0.27, simple_loss=0.343, pruned_loss=0.09851, over 28859.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.35, pruned_loss=0.1015, over 5707582.77 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3452, pruned_loss=0.09111, over 5571155.53 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3486, pruned_loss=0.1015, over 5696156.26 frames. ], batch size: 112, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:24:08,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5878, 1.7224, 1.4590, 1.8838], device='cuda:0'), covar=tensor([0.2535, 0.2718, 0.2952, 0.2414], device='cuda:0'), in_proj_covar=tensor([0.1470, 0.1067, 0.1305, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 06:24:44,400 INFO [train.py:968] (0/2) Epoch 20, batch 6150, giga_loss[loss=0.2949, simple_loss=0.3686, pruned_loss=0.1105, over 28956.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3566, pruned_loss=0.1071, over 5703271.80 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3452, pruned_loss=0.09102, over 5577694.02 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3557, pruned_loss=0.1074, over 5690879.06 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:24:49,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-10 06:25:04,885 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-10 06:25:10,423 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:25:15,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.708e+03 2.104e+03 2.673e+03 7.097e+03, threshold=4.208e+03, percent-clipped=16.0
+2023-03-10 06:25:24,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:25:26,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873355.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:25:32,175 INFO [train.py:968] (0/2) Epoch 20, batch 6200, giga_loss[loss=0.451, simple_loss=0.4664, pruned_loss=0.2178, over 26502.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3634, pruned_loss=0.1117, over 5703116.76 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3453, pruned_loss=0.09123, over 5586908.53 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.363, pruned_loss=0.1124, over 5688055.21 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:25:36,931 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873366.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:26:06,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:26:09,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:26:09,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:26:22,879 INFO [train.py:968] (0/2) Epoch 20, batch 6250, giga_loss[loss=0.3387, simple_loss=0.4002, pruned_loss=0.1386, over 28980.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3689, pruned_loss=0.1164, over 5706874.75 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3452, pruned_loss=0.09111, over 5592201.31 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3691, pruned_loss=0.1174, over 5692239.76 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:26:35,253 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:26:47,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.735e+03 2.269e+03 2.984e+03 6.866e+03, threshold=4.538e+03, percent-clipped=10.0
+2023-03-10 06:27:09,071 INFO [train.py:968] (0/2) Epoch 20, batch 6300, giga_loss[loss=0.2809, simple_loss=0.3617, pruned_loss=0.1001, over 28355.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3749, pruned_loss=0.1218, over 5692522.06 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3453, pruned_loss=0.09116, over 5589939.37 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3755, pruned_loss=0.1231, over 5685207.75 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:27:16,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3745, 1.6196, 1.3426, 1.4671], device='cuda:0'), covar=tensor([0.0793, 0.0323, 0.0334, 0.0894], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:0')
+2023-03-10 06:27:43,101 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:27:45,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:27:59,790 INFO [train.py:968] (0/2) Epoch 20, batch 6350, giga_loss[loss=0.3167, simple_loss=0.3923, pruned_loss=0.1205, over 28871.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3791, pruned_loss=0.1255, over 5672149.16 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3458, pruned_loss=0.09171, over 5579327.62 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.38, pruned_loss=0.1269, over 5678383.92 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 06:28:17,750 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:28:32,736 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:28:34,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.111e+03 1.795e+03 2.421e+03 3.378e+03 1.196e+04, threshold=4.843e+03, percent-clipped=9.0
+2023-03-10 06:28:35,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:28:43,676 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8808, 2.9902, 1.9520, 1.0667], device='cuda:0'), covar=tensor([0.6965, 0.2830, 0.3378, 0.6079], device='cuda:0'), in_proj_covar=tensor([0.1706, 0.1609, 0.1574, 0.1392], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 06:28:49,374 INFO [train.py:968] (0/2) Epoch 20, batch 6400, giga_loss[loss=0.2782, simple_loss=0.3517, pruned_loss=0.1024, over 28742.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3808, pruned_loss=0.1279, over 5669128.95 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3459, pruned_loss=0.09199, over 5591010.80 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3826, pruned_loss=0.13, over 5665804.23 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:29:01,619 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:29:42,459 INFO [train.py:968] (0/2) Epoch 20, batch 6450, giga_loss[loss=0.3543, simple_loss=0.4072, pruned_loss=0.1507, over 28916.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3838, pruned_loss=0.1314, over 5658695.98 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3459, pruned_loss=0.09204, over 5588098.33 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3863, pruned_loss=0.1341, over 5661233.33 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:30:07,921 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873635.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:30:16,413 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.904e+02 1.762e+03 2.204e+03 2.783e+03 7.890e+03, threshold=4.409e+03, percent-clipped=5.0
+2023-03-10 06:30:32,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8996, 2.6003, 2.6258, 2.4449], device='cuda:0'), covar=tensor([0.1526, 0.2274, 0.1813, 0.1961], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0740, 0.0707, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 06:30:35,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5520, 1.7097, 1.7551, 1.3865], device='cuda:0'), covar=tensor([0.1526, 0.2179, 0.1276, 0.1571], device='cuda:0'), in_proj_covar=tensor([0.0884, 0.0694, 0.0930, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:0')
+2023-03-10 06:30:37,384 INFO [train.py:968] (0/2) Epoch 20, batch 6500, giga_loss[loss=0.34, simple_loss=0.3967, pruned_loss=0.1416, over 28920.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3865, pruned_loss=0.1351, over 5645411.26 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3459, pruned_loss=0.09216, over 5585116.91 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3892, pruned_loss=0.1379, over 5650624.32 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:31:24,192 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:31:31,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:31:31,967 INFO [train.py:968] (0/2) Epoch 20, batch 6550, giga_loss[loss=0.2966, simple_loss=0.3623, pruned_loss=0.1155, over 28692.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3888, pruned_loss=0.1367, over 5635657.56 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3456, pruned_loss=0.09199, over 5588421.21 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3915, pruned_loss=0.1394, over 5637322.20 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:31:53,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873730.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:32:02,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873741.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:32:02,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.243e+03 2.160e+03 2.645e+03 3.875e+03 8.579e+03, threshold=5.290e+03, percent-clipped=18.0
+2023-03-10 06:32:21,765 INFO [train.py:968] (0/2) Epoch 20, batch 6600, giga_loss[loss=0.2873, simple_loss=0.3602, pruned_loss=0.1072, over 29062.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3867, pruned_loss=0.1356, over 5631089.93 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3459, pruned_loss=0.09221, over 5582089.75 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3895, pruned_loss=0.1386, over 5639278.19 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 06:32:49,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6127, 1.7725, 1.4715, 1.8320], device='cuda:0'), covar=tensor([0.2512, 0.2550, 0.2850, 0.2099], device='cuda:0'), in_proj_covar=tensor([0.1471, 0.1069, 0.1308, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 06:33:01,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6471, 4.4678, 4.2125, 2.1347], device='cuda:0'), covar=tensor([0.0659, 0.0854, 0.0993, 0.2145], device='cuda:0'), in_proj_covar=tensor([0.1205, 0.1120, 0.0950, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-10 06:33:14,666 INFO [train.py:968] (0/2) Epoch 20, batch 6650, giga_loss[loss=0.3909, simple_loss=0.4117, pruned_loss=0.185, over 23504.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3853, pruned_loss=0.1354, over 5625864.63 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3457, pruned_loss=0.09212, over 5580947.23 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3884, pruned_loss=0.1386, over 5634473.56 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 06:33:23,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4447, 1.7276, 1.4109, 1.4406], device='cuda:0'), covar=tensor([0.2436, 0.2328, 0.2587, 0.1998], device='cuda:0'), in_proj_covar=tensor([0.1468, 0.1066, 0.1304, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 06:33:41,482 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.47 vs. limit=5.0
+2023-03-10 06:33:44,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.174e+03 1.739e+03 2.250e+03 3.043e+03 1.269e+04, threshold=4.501e+03, percent-clipped=8.0
+2023-03-10 06:33:54,251 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:33:57,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:34:01,206 INFO [train.py:968] (0/2) Epoch 20, batch 6700, libri_loss[loss=0.2702, simple_loss=0.3551, pruned_loss=0.09268, over 29270.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3851, pruned_loss=0.1344, over 5629823.52 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3457, pruned_loss=0.09201, over 5589741.59 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3893, pruned_loss=0.1389, over 5630170.50 frames. ], batch size: 94, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 06:34:12,966 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873873.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:34:18,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873876.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:34:26,258 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873884.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:34:27,061 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:34:28,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873887.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:34:45,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873905.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:34:50,675 INFO [train.py:968] (0/2) Epoch 20, batch 6750, giga_loss[loss=0.3467, simple_loss=0.4069, pruned_loss=0.1433, over 28994.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3862, pruned_loss=0.1342, over 5642953.09 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3458, pruned_loss=0.09207, over 5595095.07 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3902, pruned_loss=0.1385, over 5639168.50 frames. ], batch size: 164, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 06:34:56,484 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873916.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:35:11,875 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 06:35:23,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.666e+03 2.044e+03 2.793e+03 5.274e+03, threshold=4.088e+03, percent-clipped=3.0
+2023-03-10 06:35:42,248 INFO [train.py:968] (0/2) Epoch 20, batch 6800, giga_loss[loss=0.3169, simple_loss=0.373, pruned_loss=0.1303, over 28722.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3865, pruned_loss=0.1342, over 5632677.09 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3453, pruned_loss=0.09178, over 5602467.78 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3911, pruned_loss=0.139, over 5623691.40 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:36:18,826 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4799, 0.9994, 4.9033, 3.6859], device='cuda:0'), covar=tensor([0.1838, 0.3172, 0.0415, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0745, 0.0636, 0.0942, 0.0891], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 06:36:20,237 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-874000.pt
+2023-03-10 06:36:32,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=874010.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:36:32,634 INFO [train.py:968] (0/2) Epoch 20, batch 6850, giga_loss[loss=0.3635, simple_loss=0.4118, pruned_loss=0.1576, over 28740.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3843, pruned_loss=0.1321, over 5633603.30 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3453, pruned_loss=0.09181, over 5606685.68 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3886, pruned_loss=0.1364, over 5623217.73 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:37:08,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.666e+03 2.194e+03 3.291e+03 7.440e+03, threshold=4.389e+03, percent-clipped=11.0
+2023-03-10 06:37:25,020 INFO [train.py:968] (0/2) Epoch 20, batch 6900, libri_loss[loss=0.2961, simple_loss=0.376, pruned_loss=0.108, over 29527.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3824, pruned_loss=0.1292, over 5645717.18 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3453, pruned_loss=0.09185, over 5616910.93 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3872, pruned_loss=0.1339, over 5628684.25 frames. ], batch size: 83, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:37:35,443 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-10 06:37:42,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=874079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:38:15,012 INFO [train.py:968] (0/2) Epoch 20, batch 6950, giga_loss[loss=0.2862, simple_loss=0.3383, pruned_loss=0.1171, over 24005.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3794, pruned_loss=0.1266, over 5652919.05 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.345, pruned_loss=0.09177, over 5622705.36 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3841, pruned_loss=0.1311, over 5634953.09 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:38:47,063 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.651e+03 2.245e+03 3.004e+03 7.790e+03, threshold=4.491e+03, percent-clipped=9.0
+2023-03-10 06:38:51,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1209, 2.8760, 1.2097, 1.2953], device='cuda:0'), covar=tensor([0.1215, 0.0514, 0.1042, 0.1620], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0550, 0.0378, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 06:38:58,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=874153.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:39:03,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=874156.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:39:07,145 INFO [train.py:968] (0/2) Epoch 20, batch 7000, giga_loss[loss=0.2936, simple_loss=0.3748, pruned_loss=0.1063, over 28897.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.377, pruned_loss=0.1243, over 5659039.77 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3449, pruned_loss=0.0917, over 5624723.91 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3812, pruned_loss=0.1284, over 5643500.08 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:39:30,980 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=874185.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:39:34,709 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-10 06:39:58,926 INFO [train.py:968] (0/2) Epoch 20, batch 7050, libri_loss[loss=0.2268, simple_loss=0.3141, pruned_loss=0.06973, over 29549.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.375, pruned_loss=0.1236, over 5654967.36 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3445, pruned_loss=0.09152, over 5628787.66 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3791, pruned_loss=0.1275, over 5639313.34 frames. ], batch size: 78, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:40:11,484 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=874222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:40:14,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=874225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:40:32,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.593e+03 2.005e+03 2.842e+03 8.259e+03, threshold=4.009e+03, percent-clipped=2.0
+2023-03-10 06:40:44,289 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=874254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:40:51,145 INFO [train.py:968] (0/2) Epoch 20, batch 7100, giga_loss[loss=0.3504, simple_loss=0.4098, pruned_loss=0.1455, over 28751.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.374, pruned_loss=0.1229, over 5652768.45 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3447, pruned_loss=0.0916, over 5631236.62 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3774, pruned_loss=0.1262, over 5638298.13 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:41:03,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=874272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:41:44,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-10 06:41:48,951 INFO [train.py:968] (0/2) Epoch 20, batch 7150, giga_loss[loss=0.2786, simple_loss=0.3502, pruned_loss=0.1035, over 28748.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.373, pruned_loss=0.1218, over 5657770.80 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3446, pruned_loss=0.09157, over 5636530.92 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3762, pruned_loss=0.125, over 5641889.00 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:41:52,883 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=874315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:42:23,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.470e+03 1.943e+03 2.762e+03 6.369e+03, threshold=3.885e+03, percent-clipped=10.0
+2023-03-10 06:42:40,645 INFO [train.py:968] (0/2) Epoch 20, batch 7200, giga_loss[loss=0.3798, simple_loss=0.4158, pruned_loss=0.1719, over 23640.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.372, pruned_loss=0.1196, over 5660966.06 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3452, pruned_loss=0.09207, over 5643488.34 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3747, pruned_loss=0.1224, over 5642321.91 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:43:37,401 INFO [train.py:968] (0/2) Epoch 20, batch 7250, giga_loss[loss=0.2908, simple_loss=0.3704, pruned_loss=0.1056, over 29064.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3734, pruned_loss=0.1183, over 5672429.01 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3449, pruned_loss=0.09182, over 5650283.00 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3765, pruned_loss=0.1214, over 5651884.72 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:43:58,599 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-10 06:44:10,562 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.603e+02 1.685e+03 2.185e+03 3.139e+03 1.038e+04, threshold=4.370e+03, percent-clipped=13.0
+2023-03-10 06:44:20,641 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-10 06:44:24,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3602, 3.1781, 1.5000, 1.4235], device='cuda:0'), covar=tensor([0.0944, 0.0327, 0.0865, 0.1312], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0550, 0.0379, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 06:44:28,788 INFO [train.py:968] (0/2) Epoch 20, batch 7300, giga_loss[loss=0.3009, simple_loss=0.3666, pruned_loss=0.1176, over 28913.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3746, pruned_loss=0.1188, over 5682252.09 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3447, pruned_loss=0.09178, over 5654030.90 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3776, pruned_loss=0.1215, over 5663216.81 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:45:18,265 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5134, 1.5772, 1.2620, 1.2072], device='cuda:0'), covar=tensor([0.0938, 0.0592, 0.1079, 0.1091], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0446, 0.0515, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 06:45:25,569 INFO [train.py:968] (0/2) Epoch 20, batch 7350, giga_loss[loss=0.2985, simple_loss=0.3679, pruned_loss=0.1146, over 29030.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3749, pruned_loss=0.1201, over 5663329.24 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3445, pruned_loss=0.09167, over 5655305.74 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3775, pruned_loss=0.1225, over 5647394.82 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:45:54,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.968e+02 1.789e+03 2.177e+03 3.062e+03 7.271e+03, threshold=4.354e+03, percent-clipped=8.0
+2023-03-10 06:46:10,783 INFO [train.py:968] (0/2) Epoch 20, batch 7400, giga_loss[loss=0.3205, simple_loss=0.3792, pruned_loss=0.1309, over 28031.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3723, pruned_loss=0.1187, over 5673907.12 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3446, pruned_loss=0.09161, over 5659339.33 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3752, pruned_loss=0.1215, over 5657957.47 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:46:58,124 INFO [train.py:968] (0/2) Epoch 20, batch 7450, giga_loss[loss=0.3387, simple_loss=0.393, pruned_loss=0.1422, over 28469.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3713, pruned_loss=0.1193, over 5672516.79 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3449, pruned_loss=0.09174, over 5658615.39 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3736, pruned_loss=0.1217, over 5660779.09 frames. ], batch size: 336, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:47:04,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3797, 1.4462, 3.3539, 3.2871], device='cuda:0'), covar=tensor([0.1250, 0.2293, 0.0444, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0744, 0.0636, 0.0941, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 06:47:20,785 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 06:47:27,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.716e+02 1.592e+03 2.068e+03 2.754e+03 5.126e+03, threshold=4.137e+03, percent-clipped=5.0
+2023-03-10 06:47:30,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=874647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:47:34,254 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=874650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:47:45,339 INFO [train.py:968] (0/2) Epoch 20, batch 7500, giga_loss[loss=0.3273, simple_loss=0.3635, pruned_loss=0.1456, over 23713.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3705, pruned_loss=0.1191, over 5670558.79 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3446, pruned_loss=0.09149, over 5663640.54 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3732, pruned_loss=0.1219, over 5656945.98 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:48:13,240 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=874687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:48:15,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=874690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:48:34,655 INFO [train.py:968] (0/2) Epoch 20, batch 7550, giga_loss[loss=0.3155, simple_loss=0.3858, pruned_loss=0.1226, over 28919.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3697, pruned_loss=0.1173, over 5673391.70 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3443, pruned_loss=0.0913, over 5669950.53 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3728, pruned_loss=0.1204, over 5656691.76 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:48:38,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=874715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:48:54,379 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.87 vs. limit=2.0
+2023-03-10 06:49:03,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.570e+03 2.009e+03 2.838e+03 6.576e+03, threshold=4.019e+03, percent-clipped=10.0
+2023-03-10 06:49:19,800 INFO [train.py:968] (0/2) Epoch 20, batch 7600, giga_loss[loss=0.291, simple_loss=0.3573, pruned_loss=0.1124, over 28718.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3705, pruned_loss=0.1179, over 5677813.31 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3443, pruned_loss=0.09161, over 5676596.66 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.374, pruned_loss=0.121, over 5658515.87 frames. ], batch size: 60, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:49:45,181 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=874790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:49:48,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=874793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:50:03,030 INFO [train.py:968] (0/2) Epoch 20, batch 7650, giga_loss[loss=0.2952, simple_loss=0.3686, pruned_loss=0.1109, over 28287.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3692, pruned_loss=0.1165, over 5691492.26 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3441, pruned_loss=0.0916, over 5683457.02 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3728, pruned_loss=0.1198, over 5670016.45 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:50:10,939 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=874822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:50:24,411 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=874833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:50:26,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7811, 1.1997, 2.8711, 2.7847], device='cuda:0'), covar=tensor([0.1673, 0.2384, 0.0580, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.0743, 0.0635, 0.0941, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 06:50:27,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=874836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:50:36,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.969e+02 1.691e+03 2.393e+03 3.639e+03 9.253e+03, threshold=4.787e+03, percent-clipped=21.0
+2023-03-10 06:50:36,525 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-10 06:50:52,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7184, 2.0138, 1.6569, 1.8843], device='cuda:0'), covar=tensor([0.2427, 0.2529, 0.2782, 0.2464], device='cuda:0'), in_proj_covar=tensor([0.1476, 0.1071, 0.1311, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 06:50:53,266 INFO [train.py:968] (0/2) Epoch 20, batch 7700, giga_loss[loss=0.2723, simple_loss=0.3449, pruned_loss=0.09983, over 28649.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3677, pruned_loss=0.1162, over 5694378.38 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3438, pruned_loss=0.09159, over 5689916.22 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3715, pruned_loss=0.1195, over 5671484.88 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:50:57,735 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=874865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:51:07,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-10 06:51:10,420 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7034, 1.0110, 2.8471, 2.7406], device='cuda:0'), covar=tensor([0.1744, 0.2585, 0.0569, 0.1009], device='cuda:0'), in_proj_covar=tensor([0.0745, 0.0637, 0.0943, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 06:51:13,582 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 06:51:43,906 INFO [train.py:968] (0/2) Epoch 20, batch 7750, giga_loss[loss=0.2993, simple_loss=0.3448, pruned_loss=0.1269, over 23655.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3664, pruned_loss=0.1162, over 5677229.55 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3443, pruned_loss=0.0918, over 5695197.33 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3695, pruned_loss=0.1192, over 5654376.65 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:52:01,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-10 06:52:16,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.803e+03 2.332e+03 3.133e+03 6.065e+03, threshold=4.664e+03, percent-clipped=3.0
+2023-03-10 06:52:17,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5643, 1.7003, 1.8055, 1.3962], device='cuda:0'), covar=tensor([0.1695, 0.2441, 0.1410, 0.1687], device='cuda:0'), in_proj_covar=tensor([0.0890, 0.0703, 0.0935, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 06:52:32,524 INFO [train.py:968] (0/2) Epoch 20, batch 7800, giga_loss[loss=0.3423, simple_loss=0.3917, pruned_loss=0.1464, over 26613.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3663, pruned_loss=0.117, over 5671838.60 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3439, pruned_loss=0.09149, over 5696633.48 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3697, pruned_loss=0.1202, over 5652117.70 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:53:20,062 INFO [train.py:968] (0/2) Epoch 20, batch 7850, giga_loss[loss=0.2746, simple_loss=0.3492, pruned_loss=0.1001, over 29152.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3668, pruned_loss=0.1178, over 5667645.81 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3447, pruned_loss=0.09187, over 5694823.27 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3696, pruned_loss=0.1209, over 5652702.10 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:53:32,016 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=875023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:53:33,446 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:53:37,994 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 06:53:45,896 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=875036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:53:54,347 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.188e+02 1.887e+03 2.349e+03 3.578e+03 5.703e+03, threshold=4.697e+03, percent-clipped=4.0
+2023-03-10 06:54:10,361 INFO [train.py:968] (0/2) Epoch 20, batch 7900, giga_loss[loss=0.2943, simple_loss=0.3638, pruned_loss=0.1124, over 28932.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3653, pruned_loss=0.1172, over 5664790.27 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3448, pruned_loss=0.09181, over 5697966.65 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3677, pruned_loss=0.1201, over 5649962.80 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:54:13,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:54:20,266 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8984, 3.7259, 3.5302, 1.7542], device='cuda:0'), covar=tensor([0.0737, 0.0826, 0.0798, 0.2073], device='cuda:0'), in_proj_covar=tensor([0.1217, 0.1131, 0.0959, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 06:54:36,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:54:56,197 INFO [train.py:968] (0/2) Epoch 20, batch 7950, giga_loss[loss=0.2899, simple_loss=0.3642, pruned_loss=0.1078, over 28887.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3644, pruned_loss=0.1165, over 5661189.08 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3448, pruned_loss=0.09175, over 5697114.44 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3668, pruned_loss=0.1195, over 5649087.11 frames. ], batch size: 112, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:55:30,918 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.569e+02 1.660e+03 2.112e+03 3.131e+03 6.513e+03, threshold=4.224e+03, percent-clipped=4.0
+2023-03-10 06:55:36,172 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-10 06:55:47,002 INFO [train.py:968] (0/2) Epoch 20, batch 8000, giga_loss[loss=0.2941, simple_loss=0.3687, pruned_loss=0.1098, over 28900.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3652, pruned_loss=0.1162, over 5664763.95 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3449, pruned_loss=0.09166, over 5700956.87 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3674, pruned_loss=0.1191, over 5651255.20 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:55:52,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:55:55,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:24,518 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:28,139 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:30,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:33,976 INFO [train.py:968] (0/2) Epoch 20, batch 8050, giga_loss[loss=0.3314, simple_loss=0.3825, pruned_loss=0.1402, over 27551.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.366, pruned_loss=0.1159, over 5671614.88 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3449, pruned_loss=0.09162, over 5703877.66 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3682, pruned_loss=0.1188, over 5657529.03 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:56:44,662 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=875224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:52,507 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:54,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:55,120 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:57:00,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.624e+03 2.093e+03 2.946e+03 1.164e+04, threshold=4.185e+03, percent-clipped=10.0
+2023-03-10 06:57:16,081 INFO [train.py:968] (0/2) Epoch 20, batch 8100, giga_loss[loss=0.2751, simple_loss=0.3488, pruned_loss=0.1007, over 28869.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3652, pruned_loss=0.1141, over 5676155.92 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3448, pruned_loss=0.09164, over 5693703.75 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3682, pruned_loss=0.1176, over 5672101.96 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:57:22,063 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:57:42,695 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0233, 2.1278, 2.2461, 1.7844], device='cuda:0'), covar=tensor([0.1871, 0.2233, 0.1483, 0.1672], device='cuda:0'), in_proj_covar=tensor([0.0889, 0.0700, 0.0932, 0.0829], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 06:58:03,543 INFO [train.py:968] (0/2) Epoch 20, batch 8150, giga_loss[loss=0.3444, simple_loss=0.3948, pruned_loss=0.147, over 28295.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3665, pruned_loss=0.1153, over 5666172.55 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.345, pruned_loss=0.09173, over 5686355.02 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3692, pruned_loss=0.1186, over 5669193.34 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:58:37,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.999e+02 1.648e+03 1.982e+03 2.816e+03 8.156e+03, threshold=3.965e+03, percent-clipped=9.0
+2023-03-10 06:58:55,046 INFO [train.py:968] (0/2) Epoch 20, batch 8200, giga_loss[loss=0.2963, simple_loss=0.3668, pruned_loss=0.1129, over 28948.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3698, pruned_loss=0.1185, over 5664441.47 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3456, pruned_loss=0.09211, over 5689157.87 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.372, pruned_loss=0.1215, over 5663900.17 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:58:56,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4646, 1.8086, 1.4631, 1.3648], device='cuda:0'), covar=tensor([0.2465, 0.2283, 0.2665, 0.2053], device='cuda:0'), in_proj_covar=tensor([0.1475, 0.1074, 0.1308, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 06:59:01,695 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-10 06:59:03,942 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-10 06:59:12,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4119, 1.7317, 1.3986, 1.4250], device='cuda:0'), covar=tensor([0.2416, 0.2451, 0.2747, 0.2096], device='cuda:0'), in_proj_covar=tensor([0.1475, 0.1073, 0.1308, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 06:59:38,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:59:50,686 INFO [train.py:968] (0/2) Epoch 20, batch 8250, giga_loss[loss=0.2952, simple_loss=0.3617, pruned_loss=0.1144, over 28934.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3726, pruned_loss=0.1226, over 5649472.83 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3459, pruned_loss=0.09222, over 5688972.73 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3746, pruned_loss=0.1253, over 5648808.57 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:59:50,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:00:22,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.269e+03 1.878e+03 2.334e+03 3.465e+03 7.832e+03, threshold=4.668e+03, percent-clipped=18.0
+2023-03-10 07:00:35,985 INFO [train.py:968] (0/2) Epoch 20, batch 8300, giga_loss[loss=0.3288, simple_loss=0.3852, pruned_loss=0.1362, over 28333.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3726, pruned_loss=0.1233, over 5655732.18 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3455, pruned_loss=0.09191, over 5687631.16 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3756, pruned_loss=0.1271, over 5654322.87 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:00:48,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=875475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:01:27,852 INFO [train.py:968] (0/2) Epoch 20, batch 8350, giga_loss[loss=0.3318, simple_loss=0.3824, pruned_loss=0.1406, over 27983.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.374, pruned_loss=0.1252, over 5654485.26 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3456, pruned_loss=0.09196, over 5691809.77 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3767, pruned_loss=0.1288, over 5649116.28 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:01:57,841 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:01:59,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:00,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.864e+03 2.440e+03 3.359e+03 1.094e+04, threshold=4.880e+03, percent-clipped=12.0
+2023-03-10 07:02:06,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:09,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:13,135 INFO [train.py:968] (0/2) Epoch 20, batch 8400, giga_loss[loss=0.3259, simple_loss=0.3848, pruned_loss=0.1335, over 28793.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3716, pruned_loss=0.1233, over 5665763.48 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3456, pruned_loss=0.09198, over 5695759.69 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3743, pruned_loss=0.1267, over 5657151.51 frames. ], batch size: 285, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:02:25,178 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:36,313 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:48,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:59,729 INFO [train.py:968] (0/2) Epoch 20, batch 8450, giga_loss[loss=0.2571, simple_loss=0.349, pruned_loss=0.08262, over 28979.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.121, over 5675504.85 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3454, pruned_loss=0.09187, over 5694156.22 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3725, pruned_loss=0.124, over 5669811.60 frames. ], batch size: 164, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:03:20,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4817, 1.3964, 4.1019, 3.2862], device='cuda:0'), covar=tensor([0.1626, 0.2707, 0.0487, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0639, 0.0945, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 07:03:31,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.583e+03 2.146e+03 3.005e+03 1.287e+04, threshold=4.292e+03, percent-clipped=7.0
+2023-03-10 07:03:44,656 INFO [train.py:968] (0/2) Epoch 20, batch 8500, giga_loss[loss=0.2723, simple_loss=0.3377, pruned_loss=0.1035, over 27970.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3676, pruned_loss=0.1176, over 5683129.33 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3455, pruned_loss=0.09176, over 5699833.73 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3704, pruned_loss=0.121, over 5672584.60 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:04:27,300 INFO [train.py:968] (0/2) Epoch 20, batch 8550, giga_loss[loss=0.2796, simple_loss=0.349, pruned_loss=0.1051, over 28933.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.366, pruned_loss=0.1167, over 5684789.18 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3456, pruned_loss=0.09192, over 5705073.59 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3687, pruned_loss=0.1199, over 5671207.66 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:04:52,470 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 07:04:58,378 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:05:00,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:05:01,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.616e+03 2.143e+03 3.157e+03 7.978e+03, threshold=4.285e+03, percent-clipped=11.0
+2023-03-10 07:05:14,800 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 07:05:15,582 INFO [train.py:968] (0/2) Epoch 20, batch 8600, giga_loss[loss=0.2791, simple_loss=0.3538, pruned_loss=0.1022, over 28900.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.365, pruned_loss=0.1171, over 5674237.05 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3454, pruned_loss=0.0918, over 5699530.62 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3676, pruned_loss=0.1202, over 5668074.26 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:05:19,553 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-10 07:05:28,346 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:06:02,690 INFO [train.py:968] (0/2) Epoch 20, batch 8650, giga_loss[loss=0.3065, simple_loss=0.3689, pruned_loss=0.122, over 28397.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3641, pruned_loss=0.1165, over 5672307.73 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3457, pruned_loss=0.09184, over 5703979.91 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3666, pruned_loss=0.1199, over 5662142.62 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:06:39,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.739e+02 1.558e+03 1.868e+03 2.618e+03 1.122e+04, threshold=3.737e+03, percent-clipped=8.0
+2023-03-10 07:06:45,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:06:55,795 INFO [train.py:968] (0/2) Epoch 20, batch 8700, giga_loss[loss=0.4756, simple_loss=0.4838, pruned_loss=0.2336, over 26578.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3681, pruned_loss=0.1192, over 5674228.15 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3458, pruned_loss=0.09188, over 5704120.56 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5665816.83 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:07:44,796 INFO [train.py:968] (0/2) Epoch 20, batch 8750, giga_loss[loss=0.3662, simple_loss=0.4134, pruned_loss=0.1595, over 26671.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3717, pruned_loss=0.1194, over 5666071.91 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.346, pruned_loss=0.09196, over 5698720.75 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3739, pruned_loss=0.1224, over 5663671.84 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:08:15,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.242e+02 1.564e+03 2.051e+03 2.952e+03 9.139e+03, threshold=4.102e+03, percent-clipped=12.0
+2023-03-10 07:08:27,911 INFO [train.py:968] (0/2) Epoch 20, batch 8800, giga_loss[loss=0.3539, simple_loss=0.4154, pruned_loss=0.1462, over 28942.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3721, pruned_loss=0.1178, over 5668460.21 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3452, pruned_loss=0.09158, over 5697184.07 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3755, pruned_loss=0.1215, over 5666948.77 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:09:00,032 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:09:02,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:09:05,098 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-876000.pt
+2023-03-10 07:09:16,527 INFO [train.py:968] (0/2) Epoch 20, batch 8850, giga_loss[loss=0.3245, simple_loss=0.3895, pruned_loss=0.1298, over 28922.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3739, pruned_loss=0.1193, over 5674207.59 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3452, pruned_loss=0.09156, over 5700957.43 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3773, pruned_loss=0.1229, over 5668899.88 frames. ], batch size: 112, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:09:27,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=876025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:09:39,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5579, 4.5270, 1.6820, 1.6908], device='cuda:0'), covar=tensor([0.1001, 0.0385, 0.0877, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0553, 0.0380, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 07:09:41,158 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 07:09:47,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.564e+02 1.627e+03 2.320e+03 3.305e+03 6.857e+03, threshold=4.640e+03, percent-clipped=14.0
+2023-03-10 07:10:04,975 INFO [train.py:968] (0/2) Epoch 20, batch 8900, giga_loss[loss=0.2642, simple_loss=0.3408, pruned_loss=0.09383, over 28823.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3743, pruned_loss=0.1199, over 5681447.06 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.345, pruned_loss=0.09146, over 5700930.63 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3772, pruned_loss=0.1231, over 5677097.77 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:10:43,613 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 07:10:49,836 INFO [train.py:968] (0/2) Epoch 20, batch 8950, libri_loss[loss=0.225, simple_loss=0.3078, pruned_loss=0.07112, over 29560.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3742, pruned_loss=0.1209, over 5685770.41 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3447, pruned_loss=0.09133, over 5706201.98 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3777, pruned_loss=0.1243, over 5676874.93 frames. ], batch size: 77, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:11:22,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.616e+03 2.124e+03 2.947e+03 6.876e+03, threshold=4.249e+03, percent-clipped=5.0
+2023-03-10 07:11:32,948 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-10 07:11:35,765 INFO [train.py:968] (0/2) Epoch 20, batch 9000, libri_loss[loss=0.2796, simple_loss=0.3619, pruned_loss=0.09868, over 27923.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3715, pruned_loss=0.1194, over 5693133.15 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3443, pruned_loss=0.091, over 5711098.69 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3759, pruned_loss=0.1237, over 5680781.90 frames. ], batch size: 116, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:11:35,769 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 07:11:45,121 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2064, simple_loss=0.3134, pruned_loss=0.04969, over 944034.00 frames. 
+2023-03-10 07:11:45,122 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 07:12:00,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9828, 3.2105, 2.0357, 1.2566], device='cuda:0'), covar=tensor([0.7747, 0.3406, 0.4414, 0.6149], device='cuda:0'), in_proj_covar=tensor([0.1727, 0.1635, 0.1584, 0.1408], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 07:12:24,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2114, 1.3645, 1.2702, 1.1142], device='cuda:0'), covar=tensor([0.2634, 0.2491, 0.1662, 0.2262], device='cuda:0'), in_proj_covar=tensor([0.1940, 0.1872, 0.1807, 0.1926], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 07:12:34,116 INFO [train.py:968] (0/2) Epoch 20, batch 9050, giga_loss[loss=0.3406, simple_loss=0.3923, pruned_loss=0.1445, over 28574.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3697, pruned_loss=0.119, over 5690012.94 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3442, pruned_loss=0.09093, over 5713686.21 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3736, pruned_loss=0.1229, over 5677601.45 frames. ], batch size: 336, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:12:36,367 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 07:13:09,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.620e+03 2.024e+03 2.852e+03 9.312e+03, threshold=4.048e+03, percent-clipped=7.0
+2023-03-10 07:13:22,382 INFO [train.py:968] (0/2) Epoch 20, batch 9100, giga_loss[loss=0.3498, simple_loss=0.403, pruned_loss=0.1483, over 28307.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3703, pruned_loss=0.1206, over 5686430.51 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3443, pruned_loss=0.09101, over 5718149.05 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.374, pruned_loss=0.1243, over 5672147.57 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:14:12,689 INFO [train.py:968] (0/2) Epoch 20, batch 9150, giga_loss[loss=0.3375, simple_loss=0.3921, pruned_loss=0.1415, over 28592.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3697, pruned_loss=0.1204, over 5690950.72 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3442, pruned_loss=0.09091, over 5721750.56 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3731, pruned_loss=0.124, over 5675787.11 frames. ], batch size: 336, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:14:13,114 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 07:14:53,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.736e+03 2.132e+03 2.866e+03 7.148e+03, threshold=4.264e+03, percent-clipped=6.0
+2023-03-10 07:15:01,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2579, 1.6524, 1.4612, 1.4415], device='cuda:0'), covar=tensor([0.0722, 0.0387, 0.0313, 0.0768], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 07:15:05,286 INFO [train.py:968] (0/2) Epoch 20, batch 9200, giga_loss[loss=0.2765, simple_loss=0.3427, pruned_loss=0.1051, over 28930.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3694, pruned_loss=0.1209, over 5684200.42 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3441, pruned_loss=0.0909, over 5724212.19 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3726, pruned_loss=0.1242, over 5669471.32 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:15:51,538 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:15:53,222 INFO [train.py:968] (0/2) Epoch 20, batch 9250, giga_loss[loss=0.2908, simple_loss=0.3609, pruned_loss=0.1103, over 29020.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3688, pruned_loss=0.121, over 5687995.67 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.344, pruned_loss=0.09086, over 5728775.47 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3721, pruned_loss=0.1245, over 5670895.14 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:16:32,159 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.399e+02 1.964e+03 2.639e+03 3.705e+03 8.488e+03, threshold=5.277e+03, percent-clipped=17.0
+2023-03-10 07:16:36,283 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:16:39,724 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:16:44,174 INFO [train.py:968] (0/2) Epoch 20, batch 9300, giga_loss[loss=0.3167, simple_loss=0.3834, pruned_loss=0.125, over 28694.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3676, pruned_loss=0.1202, over 5690702.51 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3439, pruned_loss=0.09085, over 5731471.82 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3706, pruned_loss=0.1234, over 5674076.64 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:17:35,097 INFO [train.py:968] (0/2) Epoch 20, batch 9350, giga_loss[loss=0.3737, simple_loss=0.4229, pruned_loss=0.1622, over 28621.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3699, pruned_loss=0.1215, over 5674979.85 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3437, pruned_loss=0.09074, over 5724696.27 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.373, pruned_loss=0.1246, over 5666264.47 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 07:17:48,904 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.48 vs. limit=5.0
+2023-03-10 07:17:51,618 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4505, 2.1473, 1.6435, 0.6563], device='cuda:0'), covar=tensor([0.5725, 0.2919, 0.4270, 0.6246], device='cuda:0'), in_proj_covar=tensor([0.1722, 0.1633, 0.1583, 0.1406], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 07:18:01,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3404, 3.3231, 1.4805, 1.5254], device='cuda:0'), covar=tensor([0.1038, 0.0446, 0.0897, 0.1417], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0555, 0.0381, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 07:18:11,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.731e+02 1.576e+03 2.150e+03 3.722e+03 1.335e+04, threshold=4.299e+03, percent-clipped=8.0
+2023-03-10 07:18:21,401 INFO [train.py:968] (0/2) Epoch 20, batch 9400, giga_loss[loss=0.2886, simple_loss=0.3646, pruned_loss=0.1063, over 28992.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1224, over 5681662.63 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3437, pruned_loss=0.09075, over 5727852.53 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3746, pruned_loss=0.1254, over 5671155.35 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 07:19:01,280 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-10 07:19:08,288 INFO [train.py:968] (0/2) Epoch 20, batch 9450, giga_loss[loss=0.297, simple_loss=0.3775, pruned_loss=0.1082, over 29041.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3704, pruned_loss=0.1216, over 5677461.98 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3437, pruned_loss=0.0906, over 5730745.70 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3735, pruned_loss=0.1251, over 5665168.76 frames. ], batch size: 164, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 07:19:19,277 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:19:31,752 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:19:47,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.618e+02 1.611e+03 1.970e+03 2.828e+03 7.731e+03, threshold=3.940e+03, percent-clipped=6.0
+2023-03-10 07:19:58,608 INFO [train.py:968] (0/2) Epoch 20, batch 9500, giga_loss[loss=0.272, simple_loss=0.3503, pruned_loss=0.09682, over 28858.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3725, pruned_loss=0.1203, over 5682482.90 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3442, pruned_loss=0.09088, over 5730663.14 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3749, pruned_loss=0.1232, over 5672217.46 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 07:20:25,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 07:20:42,792 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-10 07:20:43,049 INFO [train.py:968] (0/2) Epoch 20, batch 9550, giga_loss[loss=0.2718, simple_loss=0.3588, pruned_loss=0.09237, over 28977.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3731, pruned_loss=0.1186, over 5683768.69 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3439, pruned_loss=0.09073, over 5731682.20 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3759, pruned_loss=0.1217, over 5673450.44 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 07:21:20,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.263e+02 1.522e+03 1.831e+03 2.292e+03 7.105e+03, threshold=3.663e+03, percent-clipped=4.0
+2023-03-10 07:21:32,531 INFO [train.py:968] (0/2) Epoch 20, batch 9600, giga_loss[loss=0.3216, simple_loss=0.3922, pruned_loss=0.1255, over 28742.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3759, pruned_loss=0.1204, over 5678401.15 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3438, pruned_loss=0.09065, over 5734359.69 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3787, pruned_loss=0.1233, over 5667047.52 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:21:35,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3156, 1.2557, 3.8414, 3.1977], device='cuda:0'), covar=tensor([0.1745, 0.2869, 0.0507, 0.1094], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0638, 0.0944, 0.0891], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 07:21:54,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=876783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:22:20,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5249, 2.7020, 1.5953, 1.6554], device='cuda:0'), covar=tensor([0.0748, 0.0340, 0.0688, 0.1017], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0553, 0.0381, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 07:22:21,070 INFO [train.py:968] (0/2) Epoch 20, batch 9650, giga_loss[loss=0.35, simple_loss=0.4082, pruned_loss=0.1459, over 28624.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3778, pruned_loss=0.1222, over 5682464.77 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3438, pruned_loss=0.09063, over 5739448.53 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3812, pruned_loss=0.1257, over 5666942.72 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:22:34,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=876827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:22:35,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=876829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:22:55,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.688e+03 2.092e+03 2.937e+03 1.144e+04, threshold=4.184e+03, percent-clipped=12.0
+2023-03-10 07:23:04,663 INFO [train.py:968] (0/2) Epoch 20, batch 9700, giga_loss[loss=0.4812, simple_loss=0.4805, pruned_loss=0.2409, over 26546.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3773, pruned_loss=0.1228, over 5690438.14 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.343, pruned_loss=0.09017, over 5746204.92 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3821, pruned_loss=0.1273, over 5669730.05 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:23:33,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:23:51,966 INFO [train.py:968] (0/2) Epoch 20, batch 9750, giga_loss[loss=0.3687, simple_loss=0.3907, pruned_loss=0.1734, over 23832.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3771, pruned_loss=0.1239, over 5672105.11 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3428, pruned_loss=0.09018, over 5741710.81 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3825, pruned_loss=0.1288, over 5657286.81 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:23:59,167 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-10 07:24:05,079 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=876926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:06,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=876929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:26,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.701e+03 2.333e+03 2.979e+03 5.740e+03, threshold=4.666e+03, percent-clipped=7.0
+2023-03-10 07:24:34,704 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=876958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:35,527 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 07:24:36,804 INFO [train.py:968] (0/2) Epoch 20, batch 9800, giga_loss[loss=0.3004, simple_loss=0.3735, pruned_loss=0.1137, over 29030.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3763, pruned_loss=0.1232, over 5665645.07 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3431, pruned_loss=0.09033, over 5745392.94 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3811, pruned_loss=0.1277, over 5649109.57 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:24:45,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=876970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:47,345 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=876972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:48,823 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=876973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:50,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=876975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:05,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=876996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:10,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:13,211 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:16,363 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2827, 5.0597, 4.8061, 2.4628], device='cuda:0'), covar=tensor([0.0488, 0.0676, 0.0716, 0.1806], device='cuda:0'), in_proj_covar=tensor([0.1219, 0.1133, 0.0963, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 07:25:17,740 INFO [train.py:968] (0/2) Epoch 20, batch 9850, giga_loss[loss=0.3056, simple_loss=0.3782, pruned_loss=0.1166, over 28703.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3744, pruned_loss=0.1201, over 5667673.44 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.343, pruned_loss=0.09019, over 5741019.93 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3795, pruned_loss=0.125, over 5655435.83 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:25:18,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:24,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:53,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.851e+02 1.498e+03 2.114e+03 2.848e+03 6.405e+03, threshold=4.228e+03, percent-clipped=4.0
+2023-03-10 07:26:02,258 INFO [train.py:968] (0/2) Epoch 20, batch 9900, giga_loss[loss=0.2974, simple_loss=0.3736, pruned_loss=0.1106, over 29041.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3746, pruned_loss=0.1189, over 5675592.69 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3426, pruned_loss=0.08995, over 5744023.22 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3796, pruned_loss=0.1235, over 5661909.46 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:26:15,633 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:26:51,631 INFO [train.py:968] (0/2) Epoch 20, batch 9950, giga_loss[loss=0.2911, simple_loss=0.3635, pruned_loss=0.1094, over 28867.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3752, pruned_loss=0.1193, over 5677943.56 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3424, pruned_loss=0.08992, over 5747892.93 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3803, pruned_loss=0.1238, over 5661729.52 frames. ], batch size: 112, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:26:56,637 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:27:17,557 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 07:27:22,488 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:27:26,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:27:33,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.533e+03 2.203e+03 2.940e+03 8.007e+03, threshold=4.406e+03, percent-clipped=6.0
+2023-03-10 07:27:40,457 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:27:42,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:27:46,232 INFO [train.py:968] (0/2) Epoch 20, batch 10000, giga_loss[loss=0.3028, simple_loss=0.3715, pruned_loss=0.117, over 28777.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3756, pruned_loss=0.1204, over 5668276.12 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3423, pruned_loss=0.08979, over 5749574.11 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3801, pruned_loss=0.1244, over 5653287.00 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:27:58,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:28:13,419 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:28:30,614 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 07:28:34,936 INFO [train.py:968] (0/2) Epoch 20, batch 10050, libri_loss[loss=0.2268, simple_loss=0.3061, pruned_loss=0.07379, over 29661.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3745, pruned_loss=0.1204, over 5672938.05 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3425, pruned_loss=0.08994, over 5740898.14 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3788, pruned_loss=0.1243, over 5667223.05 frames. ], batch size: 69, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:29:16,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.126e+03 1.584e+03 1.985e+03 2.886e+03 7.350e+03, threshold=3.970e+03, percent-clipped=3.0
+2023-03-10 07:29:25,590 INFO [train.py:968] (0/2) Epoch 20, batch 10100, libri_loss[loss=0.2734, simple_loss=0.3652, pruned_loss=0.09079, over 29531.00 frames. ], tot_loss[loss=0.308, simple_loss=0.374, pruned_loss=0.121, over 5665135.21 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3429, pruned_loss=0.09013, over 5744869.33 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3778, pruned_loss=0.1247, over 5655001.50 frames. ], batch size: 89, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:29:28,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:29:57,716 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9324, 1.0868, 1.0470, 0.8811], device='cuda:0'), covar=tensor([0.2017, 0.2276, 0.1419, 0.1939], device='cuda:0'), in_proj_covar=tensor([0.1941, 0.1870, 0.1799, 0.1929], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 07:30:16,426 INFO [train.py:968] (0/2) Epoch 20, batch 10150, libri_loss[loss=0.2208, simple_loss=0.3035, pruned_loss=0.06903, over 29667.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3713, pruned_loss=0.1196, over 5669610.32 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.343, pruned_loss=0.09015, over 5744087.38 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3752, pruned_loss=0.1234, over 5660026.48 frames. ], batch size: 69, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:30:57,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.720e+03 2.310e+03 3.688e+03 1.435e+04, threshold=4.620e+03, percent-clipped=21.0
+2023-03-10 07:31:00,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5634, 1.7589, 1.4095, 1.7881], device='cuda:0'), covar=tensor([0.2511, 0.2633, 0.2974, 0.2313], device='cuda:0'), in_proj_covar=tensor([0.1473, 0.1071, 0.1306, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 07:31:07,376 INFO [train.py:968] (0/2) Epoch 20, batch 10200, giga_loss[loss=0.4046, simple_loss=0.4324, pruned_loss=0.1884, over 26628.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3701, pruned_loss=0.1196, over 5670958.94 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3433, pruned_loss=0.09021, over 5744541.68 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3735, pruned_loss=0.1232, over 5661402.71 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:31:30,213 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:31:37,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:31:51,413 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:31:53,715 INFO [train.py:968] (0/2) Epoch 20, batch 10250, giga_loss[loss=0.2791, simple_loss=0.3479, pruned_loss=0.1052, over 28899.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3706, pruned_loss=0.1205, over 5674447.78 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3437, pruned_loss=0.09045, over 5747860.20 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3735, pruned_loss=0.1239, over 5662279.31 frames. ], batch size: 285, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:31:53,917 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:31:58,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2405, 1.3445, 1.2268, 1.0539], device='cuda:0'), covar=tensor([0.0946, 0.0468, 0.1045, 0.0943], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0448, 0.0513, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 07:32:20,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:32:30,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:32:30,708 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.324e+02 1.880e+03 2.338e+03 3.680e+03 1.361e+04, threshold=4.676e+03, percent-clipped=16.0
+2023-03-10 07:32:40,518 INFO [train.py:968] (0/2) Epoch 20, batch 10300, giga_loss[loss=0.3037, simple_loss=0.3807, pruned_loss=0.1134, over 28904.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3684, pruned_loss=0.1183, over 5673198.31 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.344, pruned_loss=0.09053, over 5752290.55 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3714, pruned_loss=0.1219, over 5656966.10 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:33:03,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:33:25,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1016, 4.9440, 4.6759, 2.2795], device='cuda:0'), covar=tensor([0.0409, 0.0524, 0.0596, 0.1876], device='cuda:0'), in_proj_covar=tensor([0.1216, 0.1131, 0.0961, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 07:33:25,561 INFO [train.py:968] (0/2) Epoch 20, batch 10350, giga_loss[loss=0.2546, simple_loss=0.3387, pruned_loss=0.08523, over 28930.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3652, pruned_loss=0.1149, over 5678636.25 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.344, pruned_loss=0.09063, over 5757543.26 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3683, pruned_loss=0.1185, over 5658524.47 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:33:49,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-10 07:33:52,124 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:33:54,131 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:34:07,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.911e+02 1.433e+03 1.856e+03 2.280e+03 8.493e+03, threshold=3.713e+03, percent-clipped=1.0
+2023-03-10 07:34:17,319 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:34:17,684 INFO [train.py:968] (0/2) Epoch 20, batch 10400, giga_loss[loss=0.3743, simple_loss=0.4088, pruned_loss=0.1698, over 26656.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3638, pruned_loss=0.1139, over 5673917.29 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.344, pruned_loss=0.09075, over 5760095.54 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3668, pruned_loss=0.1173, over 5653541.14 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:34:22,977 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:34:25,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3519, 3.2158, 1.5155, 1.4852], device='cuda:0'), covar=tensor([0.0983, 0.0352, 0.0885, 0.1334], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0552, 0.0381, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 07:34:37,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5871, 1.6627, 1.7908, 1.3729], device='cuda:0'), covar=tensor([0.1785, 0.2471, 0.1434, 0.1696], device='cuda:0'), in_proj_covar=tensor([0.0893, 0.0705, 0.0938, 0.0834], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 07:34:50,646 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:34:52,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:35:03,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3152, 1.5478, 1.4412, 1.5435], device='cuda:0'), covar=tensor([0.0717, 0.0378, 0.0317, 0.0769], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 07:35:06,699 INFO [train.py:968] (0/2) Epoch 20, batch 10450, giga_loss[loss=0.2701, simple_loss=0.3333, pruned_loss=0.1035, over 28703.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3617, pruned_loss=0.113, over 5672184.85 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3439, pruned_loss=0.0906, over 5754448.32 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3646, pruned_loss=0.1163, over 5659307.82 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:35:10,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5460, 1.7549, 1.4499, 1.7615], device='cuda:0'), covar=tensor([0.2434, 0.2520, 0.2742, 0.2172], device='cuda:0'), in_proj_covar=tensor([0.1476, 0.1073, 0.1310, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 07:35:23,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:35:28,709 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:35:30,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:35:46,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 2.055e+03 3.043e+03 4.488e+03 1.392e+04, threshold=6.086e+03, percent-clipped=35.0
+2023-03-10 07:35:55,858 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:35:56,561 INFO [train.py:968] (0/2) Epoch 20, batch 10500, giga_loss[loss=0.324, simple_loss=0.3625, pruned_loss=0.1428, over 23359.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1126, over 5661177.15 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3444, pruned_loss=0.09091, over 5745097.15 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3621, pruned_loss=0.1155, over 5657224.42 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:35:58,805 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:36:11,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6137, 1.8006, 1.7770, 1.5756], device='cuda:0'), covar=tensor([0.2483, 0.2110, 0.1602, 0.1997], device='cuda:0'), in_proj_covar=tensor([0.1943, 0.1875, 0.1798, 0.1930], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 07:36:36,121 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-10 07:36:40,931 INFO [train.py:968] (0/2) Epoch 20, batch 10550, libri_loss[loss=0.2788, simple_loss=0.3537, pruned_loss=0.102, over 29468.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.362, pruned_loss=0.1137, over 5675465.27 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3441, pruned_loss=0.09069, over 5748686.44 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3646, pruned_loss=0.117, over 5666188.25 frames. ], batch size: 70, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:37:18,690 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:37:19,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.607e+03 2.052e+03 2.497e+03 5.862e+03, threshold=4.104e+03, percent-clipped=0.0
+2023-03-10 07:37:27,767 INFO [train.py:968] (0/2) Epoch 20, batch 10600, libri_loss[loss=0.2591, simple_loss=0.3489, pruned_loss=0.08468, over 29750.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3645, pruned_loss=0.1146, over 5679495.93 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3442, pruned_loss=0.09065, over 5753651.71 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3671, pruned_loss=0.118, over 5665237.60 frames. ], batch size: 87, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:37:28,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:37:57,814 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-10 07:38:13,971 INFO [train.py:968] (0/2) Epoch 20, batch 10650, giga_loss[loss=0.2754, simple_loss=0.3636, pruned_loss=0.09362, over 28828.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3654, pruned_loss=0.1151, over 5666001.31 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3448, pruned_loss=0.09084, over 5757446.11 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3677, pruned_loss=0.1185, over 5648662.29 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:38:36,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:38:52,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.859e+02 1.497e+03 1.945e+03 2.391e+03 8.384e+03, threshold=3.889e+03, percent-clipped=8.0
+2023-03-10 07:39:02,933 INFO [train.py:968] (0/2) Epoch 20, batch 10700, giga_loss[loss=0.2756, simple_loss=0.347, pruned_loss=0.1021, over 28391.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3657, pruned_loss=0.116, over 5647803.42 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3447, pruned_loss=0.09078, over 5760945.59 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3681, pruned_loss=0.1194, over 5628726.99 frames. ], batch size: 60, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:39:13,135 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:39:43,668 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:39:47,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:39:49,620 INFO [train.py:968] (0/2) Epoch 20, batch 10750, libri_loss[loss=0.2621, simple_loss=0.3527, pruned_loss=0.08573, over 27788.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3669, pruned_loss=0.117, over 5653651.33 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3448, pruned_loss=0.09069, over 5757214.88 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3692, pruned_loss=0.1204, over 5639436.11 frames. ], batch size: 115, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:40:13,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:40:13,740 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:40:26,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7259, 1.8883, 1.8249, 1.7441], device='cuda:0'), covar=tensor([0.1984, 0.1706, 0.1422, 0.1482], device='cuda:0'), in_proj_covar=tensor([0.1945, 0.1875, 0.1794, 0.1930], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 07:40:27,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.350e+02 1.601e+03 2.064e+03 2.767e+03 7.146e+03, threshold=4.128e+03, percent-clipped=10.0
+2023-03-10 07:40:34,128 INFO [train.py:968] (0/2) Epoch 20, batch 10800, giga_loss[loss=0.3014, simple_loss=0.3741, pruned_loss=0.1144, over 28793.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3679, pruned_loss=0.1173, over 5646875.57 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3455, pruned_loss=0.09101, over 5754237.96 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3704, pruned_loss=0.1212, over 5632110.01 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:40:40,869 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:41:08,899 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-878000.pt
+2023-03-10 07:41:12,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5737, 1.3856, 4.9748, 3.7225], device='cuda:0'), covar=tensor([0.1682, 0.2683, 0.0398, 0.0782], device='cuda:0'), in_proj_covar=tensor([0.0751, 0.0642, 0.0949, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 07:41:14,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1487, 1.3230, 1.1795, 1.0324], device='cuda:0'), covar=tensor([0.0953, 0.0487, 0.1053, 0.0980], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0448, 0.0515, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 07:41:20,444 INFO [train.py:968] (0/2) Epoch 20, batch 10850, giga_loss[loss=0.3306, simple_loss=0.3899, pruned_loss=0.1356, over 28563.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3688, pruned_loss=0.1175, over 5648089.02 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3456, pruned_loss=0.09105, over 5750970.42 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3717, pruned_loss=0.1218, over 5634452.26 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:41:43,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:41:56,495 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=878050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:41:58,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.265e+03 1.576e+03 2.234e+03 2.727e+03 7.097e+03, threshold=4.468e+03, percent-clipped=8.0
+2023-03-10 07:42:05,094 INFO [train.py:968] (0/2) Epoch 20, batch 10900, libri_loss[loss=0.2031, simple_loss=0.2835, pruned_loss=0.06135, over 29479.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3697, pruned_loss=0.1189, over 5658933.96 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3449, pruned_loss=0.09086, over 5755910.94 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3734, pruned_loss=0.1233, over 5640813.02 frames. ], batch size: 70, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:42:23,667 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:42:26,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:42:56,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:42:57,069 INFO [train.py:968] (0/2) Epoch 20, batch 10950, giga_loss[loss=0.2857, simple_loss=0.363, pruned_loss=0.1042, over 28958.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3711, pruned_loss=0.1206, over 5662787.16 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3446, pruned_loss=0.09069, over 5757151.49 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3747, pruned_loss=0.1246, over 5646062.51 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:43:03,854 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.33 vs. limit=5.0
+2023-03-10 07:43:13,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:43:21,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6439, 1.8299, 1.7138, 1.5078], device='cuda:0'), covar=tensor([0.2690, 0.2263, 0.2079, 0.2414], device='cuda:0'), in_proj_covar=tensor([0.1938, 0.1869, 0.1786, 0.1921], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 07:43:40,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.625e+03 2.131e+03 2.720e+03 9.272e+03, threshold=4.261e+03, percent-clipped=4.0
+2023-03-10 07:43:48,371 INFO [train.py:968] (0/2) Epoch 20, batch 11000, giga_loss[loss=0.2778, simple_loss=0.3614, pruned_loss=0.09717, over 28868.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3731, pruned_loss=0.1208, over 5652075.78 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3453, pruned_loss=0.09117, over 5746630.69 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3758, pruned_loss=0.1241, over 5645712.89 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:44:08,285 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:44:10,354 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:44:42,153 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:44:42,610 INFO [train.py:968] (0/2) Epoch 20, batch 11050, giga_loss[loss=0.2583, simple_loss=0.3288, pruned_loss=0.09394, over 28713.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3724, pruned_loss=0.1203, over 5646105.33 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3453, pruned_loss=0.09112, over 5745140.53 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3749, pruned_loss=0.1233, over 5641027.37 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:44:43,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:45:05,651 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 07:45:18,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:45:24,736 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.907e+02 1.570e+03 1.934e+03 2.633e+03 6.497e+03, threshold=3.868e+03, percent-clipped=3.0
+2023-03-10 07:45:30,548 INFO [train.py:968] (0/2) Epoch 20, batch 11100, giga_loss[loss=0.32, simple_loss=0.3746, pruned_loss=0.1327, over 28816.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3696, pruned_loss=0.1186, over 5656235.83 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3449, pruned_loss=0.09088, over 5741012.90 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3729, pruned_loss=0.1224, over 5652353.25 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:45:41,035 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:45:45,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:46:18,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1325, 1.3345, 1.1421, 0.9339], device='cuda:0'), covar=tensor([0.0984, 0.0507, 0.1063, 0.1072], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0448, 0.0514, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 07:46:19,204 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:46:29,674 INFO [train.py:968] (0/2) Epoch 20, batch 11150, giga_loss[loss=0.2907, simple_loss=0.3619, pruned_loss=0.1098, over 29052.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3699, pruned_loss=0.1197, over 5645332.24 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3452, pruned_loss=0.091, over 5733802.94 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3729, pruned_loss=0.1232, over 5647390.41 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:46:37,183 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-10 07:46:58,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:09,062 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.850e+02 1.646e+03 2.415e+03 3.530e+03 1.171e+04, threshold=4.829e+03, percent-clipped=18.0
+2023-03-10 07:47:12,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:15,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:17,337 INFO [train.py:968] (0/2) Epoch 20, batch 11200, giga_loss[loss=0.377, simple_loss=0.4128, pruned_loss=0.1706, over 27671.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3689, pruned_loss=0.119, over 5656371.18 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3454, pruned_loss=0.09109, over 5737958.10 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3721, pruned_loss=0.1228, over 5651300.61 frames. ], batch size: 474, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:47:19,225 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-10 07:47:37,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=878383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:41,511 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:43,656 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:45,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:48:02,554 INFO [train.py:968] (0/2) Epoch 20, batch 11250, giga_loss[loss=0.3074, simple_loss=0.3695, pruned_loss=0.1226, over 28840.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3687, pruned_loss=0.1196, over 5667232.43 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3456, pruned_loss=0.09124, over 5741482.15 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3717, pruned_loss=0.1234, over 5657407.30 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:48:12,549 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:48:15,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:48:40,265 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.538e+03 1.962e+03 2.803e+03 6.015e+03, threshold=3.923e+03, percent-clipped=1.0
+2023-03-10 07:48:43,007 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=878456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:48:47,859 INFO [train.py:968] (0/2) Epoch 20, batch 11300, giga_loss[loss=0.3772, simple_loss=0.3984, pruned_loss=0.178, over 23594.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3672, pruned_loss=0.1189, over 5668405.52 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3451, pruned_loss=0.09109, over 5746475.37 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3709, pruned_loss=0.1231, over 5653407.42 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:49:11,459 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:49:13,395 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:49:28,797 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.11 vs. limit=5.0
+2023-03-10 07:49:38,578 INFO [train.py:968] (0/2) Epoch 20, batch 11350, giga_loss[loss=0.3403, simple_loss=0.3842, pruned_loss=0.1482, over 26628.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3676, pruned_loss=0.1196, over 5664350.42 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3446, pruned_loss=0.09074, over 5749752.06 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3713, pruned_loss=0.1237, over 5648246.51 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:49:45,472 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:50:18,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.707e+03 2.173e+03 3.310e+03 1.127e+04, threshold=4.346e+03, percent-clipped=12.0
+2023-03-10 07:50:25,508 INFO [train.py:968] (0/2) Epoch 20, batch 11400, giga_loss[loss=0.4538, simple_loss=0.4563, pruned_loss=0.2257, over 26692.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3693, pruned_loss=0.1208, over 5662521.27 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3446, pruned_loss=0.09072, over 5739092.60 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.373, pruned_loss=0.1249, over 5657825.37 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:50:34,302 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:50:37,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:50:39,554 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7950, 2.1077, 2.0943, 1.7427], device='cuda:0'), covar=tensor([0.2991, 0.2485, 0.2343, 0.2496], device='cuda:0'), in_proj_covar=tensor([0.1939, 0.1873, 0.1789, 0.1925], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 07:50:45,390 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=878579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:51:03,101 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:51:14,407 INFO [train.py:968] (0/2) Epoch 20, batch 11450, giga_loss[loss=0.3235, simple_loss=0.385, pruned_loss=0.131, over 28740.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3703, pruned_loss=0.1213, over 5666336.58 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3446, pruned_loss=0.09067, over 5743043.89 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3738, pruned_loss=0.1254, over 5657251.97 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:51:50,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7344, 1.9090, 1.6439, 1.9562], device='cuda:0'), covar=tensor([0.2000, 0.2033, 0.2016, 0.1816], device='cuda:0'), in_proj_covar=tensor([0.1479, 0.1073, 0.1309, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 07:51:53,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.825e+02 1.630e+03 2.078e+03 3.165e+03 1.183e+04, threshold=4.156e+03, percent-clipped=11.0
+2023-03-10 07:52:02,934 INFO [train.py:968] (0/2) Epoch 20, batch 11500, giga_loss[loss=0.2825, simple_loss=0.3461, pruned_loss=0.1094, over 28998.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3716, pruned_loss=0.1225, over 5652069.82 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3453, pruned_loss=0.09097, over 5733121.85 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3745, pruned_loss=0.1263, over 5651507.46 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:52:51,701 INFO [train.py:968] (0/2) Epoch 20, batch 11550, giga_loss[loss=0.3061, simple_loss=0.3665, pruned_loss=0.1229, over 28547.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3714, pruned_loss=0.1231, over 5653248.18 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3455, pruned_loss=0.09115, over 5735768.49 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3742, pruned_loss=0.1268, over 5648165.21 frames. ], batch size: 336, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:52:59,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4683, 1.6273, 1.6202, 1.4867], device='cuda:0'), covar=tensor([0.1670, 0.2021, 0.2011, 0.1824], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0749, 0.0713, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 07:53:32,853 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.552e+03 2.012e+03 2.937e+03 6.482e+03, threshold=4.024e+03, percent-clipped=11.0
+2023-03-10 07:53:36,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:53:39,326 INFO [train.py:968] (0/2) Epoch 20, batch 11600, giga_loss[loss=0.2928, simple_loss=0.3645, pruned_loss=0.1105, over 28861.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1234, over 5660590.23 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3452, pruned_loss=0.09102, over 5738092.95 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3746, pruned_loss=0.1271, over 5653088.53 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:54:19,292 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-10 07:54:28,456 INFO [train.py:968] (0/2) Epoch 20, batch 11650, giga_loss[loss=0.2826, simple_loss=0.3574, pruned_loss=0.1039, over 28782.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3709, pruned_loss=0.1222, over 5666504.71 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3448, pruned_loss=0.09085, over 5741368.13 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3743, pruned_loss=0.1262, over 5655616.30 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:54:47,378 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:54:48,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7893, 1.8635, 1.4527, 1.4029], device='cuda:0'), covar=tensor([0.0946, 0.0649, 0.1023, 0.1178], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0449, 0.0515, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 07:55:03,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4348, 1.7663, 1.4949, 1.5026], device='cuda:0'), covar=tensor([0.0777, 0.0294, 0.0309, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 07:55:08,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.424e+02 1.622e+03 2.195e+03 2.840e+03 8.285e+03, threshold=4.390e+03, percent-clipped=12.0
+2023-03-10 07:55:15,042 INFO [train.py:968] (0/2) Epoch 20, batch 11700, giga_loss[loss=0.2779, simple_loss=0.3534, pruned_loss=0.1012, over 28943.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3704, pruned_loss=0.1212, over 5671178.47 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3446, pruned_loss=0.09061, over 5745930.16 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3741, pruned_loss=0.1255, over 5656316.74 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:55:16,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6396, 1.7777, 1.2730, 1.4405], device='cuda:0'), covar=tensor([0.0972, 0.0728, 0.1098, 0.1263], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0449, 0.0515, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 07:55:56,074 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:55:58,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:56:02,684 INFO [train.py:968] (0/2) Epoch 20, batch 11750, giga_loss[loss=0.3411, simple_loss=0.3948, pruned_loss=0.1437, over 28897.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3717, pruned_loss=0.1216, over 5689396.54 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3446, pruned_loss=0.09065, over 5749986.70 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3752, pruned_loss=0.1257, over 5672506.00 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:56:29,067 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:56:48,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:56:48,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.749e+03 2.084e+03 2.809e+03 4.920e+03, threshold=4.168e+03, percent-clipped=2.0
+2023-03-10 07:56:55,062 INFO [train.py:968] (0/2) Epoch 20, batch 11800, giga_loss[loss=0.3109, simple_loss=0.3737, pruned_loss=0.124, over 28879.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3733, pruned_loss=0.1233, over 5686380.99 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3446, pruned_loss=0.09071, over 5753719.79 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3767, pruned_loss=0.1272, over 5667934.02 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:57:08,080 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:57:10,407 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:57:27,253 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 07:57:38,026 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=879006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:57:39,263 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=879008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:57:41,222 INFO [train.py:968] (0/2) Epoch 20, batch 11850, libri_loss[loss=0.2526, simple_loss=0.3415, pruned_loss=0.08184, over 29526.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3729, pruned_loss=0.1226, over 5696099.63 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3449, pruned_loss=0.09078, over 5757294.48 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3762, pruned_loss=0.1265, over 5676354.39 frames. ], batch size: 84, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:58:18,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.702e+02 1.669e+03 2.117e+03 3.014e+03 6.059e+03, threshold=4.233e+03, percent-clipped=9.0
+2023-03-10 07:58:26,308 INFO [train.py:968] (0/2) Epoch 20, batch 11900, giga_loss[loss=0.351, simple_loss=0.4067, pruned_loss=0.1476, over 28724.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.374, pruned_loss=0.1225, over 5681620.44 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3456, pruned_loss=0.09132, over 5752267.63 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.377, pruned_loss=0.1263, over 5667211.88 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:59:04,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=879097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:59:07,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=879100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:59:17,839 INFO [train.py:968] (0/2) Epoch 20, batch 11950, giga_loss[loss=0.3161, simple_loss=0.381, pruned_loss=0.1256, over 28685.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3738, pruned_loss=0.122, over 5674374.55 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3456, pruned_loss=0.09132, over 5752976.29 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3762, pruned_loss=0.1251, over 5662223.25 frames. ], batch size: 262, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 07:59:34,597 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=879129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:00:00,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.844e+03 2.369e+03 3.379e+03 1.042e+04, threshold=4.738e+03, percent-clipped=12.0
+2023-03-10 08:00:05,054 INFO [train.py:968] (0/2) Epoch 20, batch 12000, libri_loss[loss=0.2851, simple_loss=0.3655, pruned_loss=0.1024, over 29525.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.12, over 5686891.22 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3454, pruned_loss=0.09122, over 5757392.55 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3739, pruned_loss=0.1234, over 5671173.17 frames. ], batch size: 84, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:00:05,058 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 08:00:14,019 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2083, simple_loss=0.3156, pruned_loss=0.05049, over 944034.00 frames. 
+2023-03-10 08:00:14,019 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 08:00:59,290 INFO [train.py:968] (0/2) Epoch 20, batch 12050, giga_loss[loss=0.2798, simple_loss=0.3487, pruned_loss=0.1054, over 28774.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3707, pruned_loss=0.1201, over 5679384.35 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3454, pruned_loss=0.09123, over 5757691.20 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3735, pruned_loss=0.1233, over 5665210.54 frames. ], batch size: 99, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:01:01,351 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=879213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:01:41,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.832e+02 1.502e+03 2.146e+03 3.062e+03 8.510e+03, threshold=4.292e+03, percent-clipped=9.0
+2023-03-10 08:01:47,085 INFO [train.py:968] (0/2) Epoch 20, batch 12100, giga_loss[loss=0.2939, simple_loss=0.364, pruned_loss=0.1119, over 28893.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3708, pruned_loss=0.12, over 5677901.28 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3449, pruned_loss=0.09103, over 5759380.77 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3744, pruned_loss=0.1238, over 5662267.21 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:02:30,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2111, 2.5213, 1.2442, 1.3693], device='cuda:0'), covar=tensor([0.1024, 0.0435, 0.0938, 0.1400], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0553, 0.0380, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 08:02:34,019 INFO [train.py:968] (0/2) Epoch 20, batch 12150, giga_loss[loss=0.2758, simple_loss=0.3457, pruned_loss=0.1029, over 28811.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3716, pruned_loss=0.1209, over 5678282.24 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3446, pruned_loss=0.09078, over 5760362.58 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3757, pruned_loss=0.1252, over 5661796.90 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:03:18,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.930e+02 1.550e+03 2.032e+03 3.141e+03 6.488e+03, threshold=4.064e+03, percent-clipped=7.0
+2023-03-10 08:03:24,783 INFO [train.py:968] (0/2) Epoch 20, batch 12200, giga_loss[loss=0.3264, simple_loss=0.3812, pruned_loss=0.1358, over 27635.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3725, pruned_loss=0.123, over 5675566.95 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3441, pruned_loss=0.09047, over 5763143.95 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3768, pruned_loss=0.1273, over 5658609.62 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:03:34,791 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=879370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:03:46,922 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=879383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:04:13,565 INFO [train.py:968] (0/2) Epoch 20, batch 12250, giga_loss[loss=0.3172, simple_loss=0.3839, pruned_loss=0.1253, over 28769.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3734, pruned_loss=0.1235, over 5675725.17 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3447, pruned_loss=0.09082, over 5762916.01 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.377, pruned_loss=0.1276, over 5659879.30 frames. ], batch size: 99, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:04:54,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.649e+03 2.102e+03 2.895e+03 9.664e+03, threshold=4.203e+03, percent-clipped=9.0
+2023-03-10 08:04:59,129 INFO [train.py:968] (0/2) Epoch 20, batch 12300, giga_loss[loss=0.3002, simple_loss=0.3656, pruned_loss=0.1174, over 28849.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3741, pruned_loss=0.1242, over 5668622.42 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.345, pruned_loss=0.09088, over 5762127.26 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3777, pruned_loss=0.1284, over 5653585.38 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:05:41,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1304, 1.3211, 1.2965, 1.0906], device='cuda:0'), covar=tensor([0.2531, 0.2313, 0.1557, 0.2091], device='cuda:0'), in_proj_covar=tensor([0.1939, 0.1868, 0.1779, 0.1926], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 08:05:49,862 INFO [train.py:968] (0/2) Epoch 20, batch 12350, giga_loss[loss=0.3217, simple_loss=0.3739, pruned_loss=0.1348, over 28699.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3748, pruned_loss=0.1247, over 5663269.90 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3451, pruned_loss=0.0909, over 5763581.29 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3779, pruned_loss=0.1284, over 5649400.50 frames. ], batch size: 92, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:06:03,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=879526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:06:05,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=879529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:06:34,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.548e+03 2.029e+03 2.858e+03 7.842e+03, threshold=4.058e+03, percent-clipped=9.0
+2023-03-10 08:06:37,664 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=879558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:06:39,316 INFO [train.py:968] (0/2) Epoch 20, batch 12400, giga_loss[loss=0.3339, simple_loss=0.3865, pruned_loss=0.1406, over 28700.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3748, pruned_loss=0.1252, over 5648398.39 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3451, pruned_loss=0.09087, over 5767102.79 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3779, pruned_loss=0.129, over 5631754.57 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:07:07,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=879588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:07:31,687 INFO [train.py:968] (0/2) Epoch 20, batch 12450, giga_loss[loss=0.343, simple_loss=0.3819, pruned_loss=0.1521, over 23708.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3752, pruned_loss=0.1253, over 5648076.58 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.345, pruned_loss=0.09083, over 5767643.28 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3778, pruned_loss=0.1284, over 5634127.09 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:08:10,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.583e+02 1.563e+03 1.975e+03 2.654e+03 9.733e+03, threshold=3.949e+03, percent-clipped=8.0
+2023-03-10 08:08:14,362 INFO [train.py:968] (0/2) Epoch 20, batch 12500, giga_loss[loss=0.2764, simple_loss=0.3493, pruned_loss=0.1018, over 28870.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3743, pruned_loss=0.1238, over 5658152.81 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.345, pruned_loss=0.09071, over 5771455.10 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3773, pruned_loss=0.1274, over 5640343.95 frames. ], batch size: 112, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:08:31,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5681, 1.7938, 1.6380, 1.5694], device='cuda:0'), covar=tensor([0.1812, 0.2141, 0.2403, 0.2193], device='cuda:0'), in_proj_covar=tensor([0.0476, 0.0751, 0.0716, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 08:09:04,228 INFO [train.py:968] (0/2) Epoch 20, batch 12550, libri_loss[loss=0.2984, simple_loss=0.3775, pruned_loss=0.1097, over 29519.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3716, pruned_loss=0.1218, over 5662319.24 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3447, pruned_loss=0.09061, over 5771026.06 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3754, pruned_loss=0.1262, over 5643633.09 frames. ], batch size: 81, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:09:17,826 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-10 08:09:21,383 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=879731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:09:24,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4032, 3.7418, 1.5128, 1.7419], device='cuda:0'), covar=tensor([0.1030, 0.0339, 0.0903, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0553, 0.0380, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 08:09:24,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=879734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:09:34,824 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=879745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:09:47,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.300e+02 1.532e+03 2.162e+03 2.886e+03 6.529e+03, threshold=4.325e+03, percent-clipped=8.0
+2023-03-10 08:09:51,495 INFO [train.py:968] (0/2) Epoch 20, batch 12600, libri_loss[loss=0.2517, simple_loss=0.3444, pruned_loss=0.07952, over 29541.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3691, pruned_loss=0.1201, over 5670294.63 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3447, pruned_loss=0.09051, over 5773370.89 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3726, pruned_loss=0.1242, over 5651305.60 frames. ], batch size: 84, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:09:53,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=879763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:10:37,123 INFO [train.py:968] (0/2) Epoch 20, batch 12650, libri_loss[loss=0.2816, simple_loss=0.3623, pruned_loss=0.1005, over 29368.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3675, pruned_loss=0.1195, over 5687030.75 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3445, pruned_loss=0.09034, over 5776399.05 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.371, pruned_loss=0.1236, over 5666931.59 frames. ], batch size: 92, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:10:41,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8523, 2.0167, 1.3514, 1.6166], device='cuda:0'), covar=tensor([0.0981, 0.0707, 0.1129, 0.1233], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0448, 0.0513, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 08:11:18,244 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=879854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:11:22,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.801e+03 2.385e+03 3.462e+03 6.548e+03, threshold=4.770e+03, percent-clipped=15.0
+2023-03-10 08:11:25,668 INFO [train.py:968] (0/2) Epoch 20, batch 12700, giga_loss[loss=0.2905, simple_loss=0.3541, pruned_loss=0.1135, over 28792.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3648, pruned_loss=0.1188, over 5659898.73 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3447, pruned_loss=0.09041, over 5777140.44 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.368, pruned_loss=0.1227, over 5641271.01 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:11:43,940 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=879878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:11:51,877 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=879888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:11:54,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=879891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:12:15,314 INFO [train.py:968] (0/2) Epoch 20, batch 12750, giga_loss[loss=0.2633, simple_loss=0.3349, pruned_loss=0.09583, over 28997.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3651, pruned_loss=0.1198, over 5664820.06 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.09041, over 5779005.66 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3678, pruned_loss=0.1232, over 5647013.09 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:12:16,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9477, 1.1930, 1.3670, 1.0340], device='cuda:0'), covar=tensor([0.1991, 0.1540, 0.2352, 0.1836], device='cuda:0'), in_proj_covar=tensor([0.0476, 0.0750, 0.0716, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 08:12:25,624 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=879920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:12:46,546 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=879940.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 08:13:04,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.712e+03 2.222e+03 2.883e+03 7.345e+03, threshold=4.443e+03, percent-clipped=6.0
+2023-03-10 08:13:07,306 INFO [train.py:968] (0/2) Epoch 20, batch 12800, giga_loss[loss=0.3409, simple_loss=0.3857, pruned_loss=0.1481, over 26518.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3641, pruned_loss=0.1194, over 5651524.76 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3447, pruned_loss=0.09038, over 5772321.67 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3665, pruned_loss=0.1226, over 5641991.12 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:13:48,468 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-880000.pt
+2023-03-10 08:13:56,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2961, 1.5033, 1.4190, 1.3138], device='cuda:0'), covar=tensor([0.2092, 0.1861, 0.1349, 0.1688], device='cuda:0'), in_proj_covar=tensor([0.1936, 0.1870, 0.1786, 0.1924], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 08:13:58,199 INFO [train.py:968] (0/2) Epoch 20, batch 12850, libri_loss[loss=0.2561, simple_loss=0.3291, pruned_loss=0.09152, over 29376.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3627, pruned_loss=0.1167, over 5656962.56 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3443, pruned_loss=0.09024, over 5774485.24 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3656, pruned_loss=0.1201, over 5644110.37 frames. ], batch size: 67, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:14:13,406 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3856, 1.5466, 1.4244, 1.5716], device='cuda:0'), covar=tensor([0.0764, 0.0335, 0.0325, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0069, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 08:14:44,313 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.658e+02 1.497e+03 1.965e+03 2.601e+03 7.463e+03, threshold=3.929e+03, percent-clipped=6.0
+2023-03-10 08:14:46,473 INFO [train.py:968] (0/2) Epoch 20, batch 12900, giga_loss[loss=0.2474, simple_loss=0.3355, pruned_loss=0.07964, over 28662.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.361, pruned_loss=0.1134, over 5658218.29 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3443, pruned_loss=0.09026, over 5775555.85 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3639, pruned_loss=0.1169, over 5643239.06 frames. ], batch size: 262, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:15:41,941 INFO [train.py:968] (0/2) Epoch 20, batch 12950, giga_loss[loss=0.2555, simple_loss=0.3351, pruned_loss=0.0879, over 28836.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3576, pruned_loss=0.1095, over 5656110.89 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3441, pruned_loss=0.09015, over 5776995.33 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3602, pruned_loss=0.1125, over 5641791.37 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:16:15,654 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3769, 1.7139, 1.5796, 1.1720], device='cuda:0'), covar=tensor([0.1882, 0.2727, 0.1623, 0.1912], device='cuda:0'), in_proj_covar=tensor([0.0889, 0.0699, 0.0934, 0.0830], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 08:16:29,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.628e+02 1.315e+03 1.727e+03 2.565e+03 1.437e+04, threshold=3.454e+03, percent-clipped=11.0
+2023-03-10 08:16:31,759 INFO [train.py:968] (0/2) Epoch 20, batch 13000, giga_loss[loss=0.2573, simple_loss=0.337, pruned_loss=0.08875, over 27998.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3547, pruned_loss=0.1066, over 5651258.45 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3435, pruned_loss=0.09003, over 5768675.42 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3576, pruned_loss=0.1097, over 5643439.34 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:17:17,045 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 08:17:26,948 INFO [train.py:968] (0/2) Epoch 20, batch 13050, giga_loss[loss=0.2265, simple_loss=0.2954, pruned_loss=0.07883, over 24217.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3518, pruned_loss=0.1035, over 5642968.72 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3434, pruned_loss=0.09004, over 5767203.71 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3543, pruned_loss=0.1062, over 5635979.21 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:17:46,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=880229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:18:11,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=880253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:18:17,269 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.060e+02 1.345e+03 1.696e+03 2.209e+03 4.452e+03, threshold=3.393e+03, percent-clipped=3.0
+2023-03-10 08:18:20,490 INFO [train.py:968] (0/2) Epoch 20, batch 13100, giga_loss[loss=0.2874, simple_loss=0.3651, pruned_loss=0.1049, over 28950.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3517, pruned_loss=0.1007, over 5652047.23 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3435, pruned_loss=0.09015, over 5758862.81 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3536, pruned_loss=0.1028, over 5652783.11 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:19:13,428 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.91 vs. limit=5.0
+2023-03-10 08:19:14,299 INFO [train.py:968] (0/2) Epoch 20, batch 13150, libri_loss[loss=0.2186, simple_loss=0.2938, pruned_loss=0.07172, over 29484.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3527, pruned_loss=0.102, over 5628834.32 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3433, pruned_loss=0.09029, over 5742010.35 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3547, pruned_loss=0.1038, over 5641673.77 frames. ], batch size: 70, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:19:18,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=880315.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 08:19:39,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3087, 1.6879, 1.3553, 1.0886], device='cuda:0'), covar=tensor([0.2674, 0.2585, 0.3046, 0.2306], device='cuda:0'), in_proj_covar=tensor([0.1486, 0.1077, 0.1320, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 08:19:46,193 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 08:19:56,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3395, 3.3866, 1.4104, 1.6213], device='cuda:0'), covar=tensor([0.1003, 0.0312, 0.0966, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0551, 0.0380, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 08:20:02,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.183e+02 1.334e+03 1.860e+03 3.157e+03 1.025e+04, threshold=3.720e+03, percent-clipped=22.0
+2023-03-10 08:20:05,598 INFO [train.py:968] (0/2) Epoch 20, batch 13200, giga_loss[loss=0.272, simple_loss=0.3519, pruned_loss=0.09602, over 28664.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3512, pruned_loss=0.1008, over 5639867.68 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3434, pruned_loss=0.09054, over 5740817.85 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3529, pruned_loss=0.1022, over 5649468.57 frames. ], batch size: 242, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:20:17,844 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=880372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:20:19,165 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2584, 1.6660, 1.6689, 1.4522], device='cuda:0'), covar=tensor([0.1854, 0.1549, 0.1854, 0.1745], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0738, 0.0705, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 08:20:19,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=880375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:20:42,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=880396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:20:44,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=880399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:20:48,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=880404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:20:52,926 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5852, 2.3940, 1.7400, 0.7379], device='cuda:0'), covar=tensor([0.6653, 0.3490, 0.4488, 0.6749], device='cuda:0'), in_proj_covar=tensor([0.1720, 0.1627, 0.1587, 0.1404], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 08:20:55,292 INFO [train.py:968] (0/2) Epoch 20, batch 13250, giga_loss[loss=0.2839, simple_loss=0.3552, pruned_loss=0.1063, over 28563.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3476, pruned_loss=0.09912, over 5631682.05 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3432, pruned_loss=0.09067, over 5745212.59 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3493, pruned_loss=0.1004, over 5632565.75 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:20:57,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2985, 0.9385, 1.0070, 1.4456], device='cuda:0'), covar=tensor([0.0710, 0.0439, 0.0350, 0.0793], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0069, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 08:21:14,563 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=880428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:21:44,142 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=880458.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 08:21:44,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.234e+02 1.452e+03 1.942e+03 2.786e+03 1.275e+04, threshold=3.885e+03, percent-clipped=16.0
+2023-03-10 08:21:45,820 INFO [train.py:968] (0/2) Epoch 20, batch 13300, giga_loss[loss=0.242, simple_loss=0.3258, pruned_loss=0.07914, over 29038.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3476, pruned_loss=0.09916, over 5623618.98 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3431, pruned_loss=0.09085, over 5734710.94 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.349, pruned_loss=0.1001, over 5632615.69 frames. ], batch size: 128, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:21:46,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=880461.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 08:22:04,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7254, 1.8962, 1.5422, 1.9749], device='cuda:0'), covar=tensor([0.2760, 0.2672, 0.3065, 0.2395], device='cuda:0'), in_proj_covar=tensor([0.1484, 0.1075, 0.1318, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 08:22:14,408 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=880490.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 08:22:34,567 INFO [train.py:968] (0/2) Epoch 20, batch 13350, giga_loss[loss=0.2474, simple_loss=0.3323, pruned_loss=0.0812, over 29059.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3471, pruned_loss=0.09881, over 5633267.30 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3427, pruned_loss=0.09085, over 5737434.21 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3489, pruned_loss=0.09984, over 5634637.06 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:23:18,527 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 08:23:23,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4724, 1.8962, 1.6600, 1.6504], device='cuda:0'), covar=tensor([0.0702, 0.0282, 0.0294, 0.0713], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0069, 0.0062, 0.0105], device='cuda:0')
+2023-03-10 08:23:24,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.104e+02 1.426e+03 1.835e+03 2.400e+03 7.811e+03, threshold=3.669e+03, percent-clipped=9.0
+2023-03-10 08:23:27,295 INFO [train.py:968] (0/2) Epoch 20, batch 13400, giga_loss[loss=0.2369, simple_loss=0.3228, pruned_loss=0.07548, over 28039.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3439, pruned_loss=0.09595, over 5638701.53 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3425, pruned_loss=0.09082, over 5738991.34 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3454, pruned_loss=0.09693, over 5636808.60 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:23:32,292 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=880564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:24:17,122 INFO [train.py:968] (0/2) Epoch 20, batch 13450, giga_loss[loss=0.2481, simple_loss=0.3211, pruned_loss=0.08751, over 27647.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3404, pruned_loss=0.09332, over 5641706.20 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3417, pruned_loss=0.09046, over 5738831.62 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3424, pruned_loss=0.09456, over 5637436.43 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:25:08,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.327e+02 1.170e+03 1.577e+03 2.062e+03 5.092e+03, threshold=3.155e+03, percent-clipped=5.0
+2023-03-10 08:25:10,634 INFO [train.py:968] (0/2) Epoch 20, batch 13500, giga_loss[loss=0.2347, simple_loss=0.3127, pruned_loss=0.07831, over 27899.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.337, pruned_loss=0.09144, over 5643150.96 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3418, pruned_loss=0.09051, over 5732210.30 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3384, pruned_loss=0.09242, over 5643322.38 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:25:15,912 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.96 vs. limit=2.0
+2023-03-10 08:25:50,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3016, 1.4934, 1.3764, 1.2832], device='cuda:0'), covar=tensor([0.2223, 0.1701, 0.1339, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.1910, 0.1838, 0.1754, 0.1892], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 08:25:59,440 INFO [train.py:968] (0/2) Epoch 20, batch 13550, giga_loss[loss=0.2404, simple_loss=0.3232, pruned_loss=0.07884, over 28833.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.336, pruned_loss=0.09139, over 5651012.21 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.341, pruned_loss=0.0902, over 5732132.05 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3377, pruned_loss=0.09249, over 5647500.12 frames. ], batch size: 285, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:26:02,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2619, 1.2992, 3.3998, 3.0776], device='cuda:0'), covar=tensor([0.1539, 0.2731, 0.0473, 0.1500], device='cuda:0'), in_proj_covar=tensor([0.0747, 0.0638, 0.0941, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 08:26:53,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.510e+02 1.397e+03 2.021e+03 3.047e+03 1.079e+04, threshold=4.043e+03, percent-clipped=24.0
+2023-03-10 08:26:55,423 INFO [train.py:968] (0/2) Epoch 20, batch 13600, giga_loss[loss=0.2967, simple_loss=0.35, pruned_loss=0.1217, over 26684.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3347, pruned_loss=0.09114, over 5640403.87 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3406, pruned_loss=0.08998, over 5733861.13 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3363, pruned_loss=0.09224, over 5634348.66 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:27:54,554 INFO [train.py:968] (0/2) Epoch 20, batch 13650, giga_loss[loss=0.2756, simple_loss=0.3589, pruned_loss=0.09614, over 28909.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3373, pruned_loss=0.0919, over 5650966.20 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3403, pruned_loss=0.08981, over 5736661.33 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3388, pruned_loss=0.09293, over 5642760.30 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:28:55,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.553e+02 1.416e+03 1.868e+03 2.440e+03 6.106e+03, threshold=3.736e+03, percent-clipped=8.0
+2023-03-10 08:28:56,507 INFO [train.py:968] (0/2) Epoch 20, batch 13700, giga_loss[loss=0.2542, simple_loss=0.3352, pruned_loss=0.08658, over 29132.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3407, pruned_loss=0.09279, over 5650631.63 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3407, pruned_loss=0.09019, over 5735401.75 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3414, pruned_loss=0.09329, over 5644090.94 frames. ], batch size: 113, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:29:06,292 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4168, 1.7426, 1.5250, 1.6543], device='cuda:0'), covar=tensor([0.0743, 0.0311, 0.0317, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 08:29:58,773 INFO [train.py:968] (0/2) Epoch 20, batch 13750, giga_loss[loss=0.3451, simple_loss=0.4035, pruned_loss=0.1433, over 28426.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3405, pruned_loss=0.09282, over 5659939.92 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3401, pruned_loss=0.08993, over 5734348.06 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3415, pruned_loss=0.09347, over 5654706.87 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:30:37,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=880939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:30:58,742 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.267e+02 1.423e+03 2.022e+03 2.873e+03 6.460e+03, threshold=4.043e+03, percent-clipped=20.0
+2023-03-10 08:30:59,572 INFO [train.py:968] (0/2) Epoch 20, batch 13800, giga_loss[loss=0.237, simple_loss=0.3125, pruned_loss=0.08071, over 28726.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3376, pruned_loss=0.09097, over 5663465.64 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3398, pruned_loss=0.08978, over 5735104.14 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3387, pruned_loss=0.09168, over 5656420.75 frames. ], batch size: 60, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:31:58,101 INFO [train.py:968] (0/2) Epoch 20, batch 13850, giga_loss[loss=0.2238, simple_loss=0.3217, pruned_loss=0.06294, over 29030.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.337, pruned_loss=0.08985, over 5667741.14 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3389, pruned_loss=0.08937, over 5741840.50 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3386, pruned_loss=0.0908, over 5653195.96 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:31:59,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-10 08:32:11,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5684, 1.5293, 1.2086, 1.2340], device='cuda:0'), covar=tensor([0.0831, 0.0468, 0.0877, 0.1078], device='cuda:0'), in_proj_covar=tensor([0.0386, 0.0442, 0.0509, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 08:32:55,402 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.410e+02 1.244e+03 1.605e+03 2.362e+03 6.881e+03, threshold=3.210e+03, percent-clipped=10.0
+2023-03-10 08:32:56,310 INFO [train.py:968] (0/2) Epoch 20, batch 13900, giga_loss[loss=0.2529, simple_loss=0.3319, pruned_loss=0.08696, over 28494.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3351, pruned_loss=0.08775, over 5668923.11 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3383, pruned_loss=0.08919, over 5737231.31 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3369, pruned_loss=0.08866, over 5658036.16 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:33:06,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7633, 2.4710, 1.5861, 1.0325], device='cuda:0'), covar=tensor([0.6980, 0.3546, 0.3734, 0.6206], device='cuda:0'), in_proj_covar=tensor([0.1715, 0.1622, 0.1580, 0.1402], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 08:33:23,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=881082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:33:28,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=881085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:33:58,214 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:33:59,508 INFO [train.py:968] (0/2) Epoch 20, batch 13950, giga_loss[loss=0.2904, simple_loss=0.3585, pruned_loss=0.1111, over 28944.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3334, pruned_loss=0.0882, over 5662609.69 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3384, pruned_loss=0.08939, over 5730400.29 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3346, pruned_loss=0.08873, over 5658559.85 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:34:04,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=881114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:34:26,266 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:34:59,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.762e+02 1.302e+03 1.689e+03 2.425e+03 5.385e+03, threshold=3.378e+03, percent-clipped=6.0
+2023-03-10 08:35:02,549 INFO [train.py:968] (0/2) Epoch 20, batch 14000, giga_loss[loss=0.2602, simple_loss=0.3288, pruned_loss=0.09582, over 27675.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3321, pruned_loss=0.0879, over 5654512.81 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3381, pruned_loss=0.08938, over 5724149.07 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3332, pruned_loss=0.0883, over 5656447.64 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:35:18,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3683, 1.6117, 1.3929, 1.6043], device='cuda:0'), covar=tensor([0.0717, 0.0404, 0.0351, 0.0784], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 08:36:01,069 INFO [train.py:968] (0/2) Epoch 20, batch 14050, giga_loss[loss=0.2532, simple_loss=0.3346, pruned_loss=0.0859, over 28977.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3325, pruned_loss=0.08861, over 5657283.79 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.338, pruned_loss=0.08955, over 5726762.60 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3334, pruned_loss=0.08874, over 5654922.38 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:37:04,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.065e+02 1.435e+03 1.807e+03 2.769e+03 4.601e+03, threshold=3.613e+03, percent-clipped=13.0
+2023-03-10 08:37:05,720 INFO [train.py:968] (0/2) Epoch 20, batch 14100, giga_loss[loss=0.2334, simple_loss=0.3265, pruned_loss=0.07011, over 28483.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3345, pruned_loss=0.08877, over 5640288.45 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3381, pruned_loss=0.08972, over 5716888.98 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.335, pruned_loss=0.08871, over 5645363.39 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:38:13,298 INFO [train.py:968] (0/2) Epoch 20, batch 14150, giga_loss[loss=0.2344, simple_loss=0.319, pruned_loss=0.07485, over 28639.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3331, pruned_loss=0.0872, over 5649329.10 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3378, pruned_loss=0.08955, over 5715495.86 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3337, pruned_loss=0.08729, over 5653147.03 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:38:55,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7009, 5.0530, 1.9121, 2.0215], device='cuda:0'), covar=tensor([0.0933, 0.0218, 0.0874, 0.1228], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0547, 0.0380, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 08:39:11,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2314, 1.1543, 3.6057, 3.0523], device='cuda:0'), covar=tensor([0.1620, 0.2842, 0.0420, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0744, 0.0637, 0.0937, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 08:39:18,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.421e+02 1.414e+03 1.852e+03 2.501e+03 5.177e+03, threshold=3.705e+03, percent-clipped=7.0
+2023-03-10 08:39:18,379 INFO [train.py:968] (0/2) Epoch 20, batch 14200, giga_loss[loss=0.2519, simple_loss=0.3242, pruned_loss=0.08983, over 29191.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3326, pruned_loss=0.08762, over 5664432.63 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3378, pruned_loss=0.08958, over 5720571.99 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.333, pruned_loss=0.08759, over 5661299.87 frames. ], batch size: 107, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:40:20,958 INFO [train.py:968] (0/2) Epoch 20, batch 14250, giga_loss[loss=0.2544, simple_loss=0.34, pruned_loss=0.08436, over 28887.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3356, pruned_loss=0.08897, over 5663797.94 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3376, pruned_loss=0.08959, over 5707031.70 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.336, pruned_loss=0.08891, over 5671720.75 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:40:25,541 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-10 08:41:22,619 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.508e+02 1.461e+03 1.783e+03 2.496e+03 8.049e+03, threshold=3.567e+03, percent-clipped=7.0
+2023-03-10 08:41:22,631 INFO [train.py:968] (0/2) Epoch 20, batch 14300, giga_loss[loss=0.2577, simple_loss=0.3589, pruned_loss=0.07827, over 28859.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3388, pruned_loss=0.08895, over 5653498.94 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3372, pruned_loss=0.08963, over 5696992.68 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3394, pruned_loss=0.08885, over 5666866.14 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:41:32,773 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:41:52,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=881484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:42:21,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=881506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:42:27,035 INFO [train.py:968] (0/2) Epoch 20, batch 14350, giga_loss[loss=0.2405, simple_loss=0.3337, pruned_loss=0.07366, over 28898.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3405, pruned_loss=0.08776, over 5648529.63 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3371, pruned_loss=0.0896, over 5688542.38 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3411, pruned_loss=0.0877, over 5665769.55 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:42:35,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-10 08:42:42,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4017, 1.6647, 1.3893, 1.2772], device='cuda:0'), covar=tensor([0.2553, 0.2477, 0.2754, 0.2351], device='cuda:0'), in_proj_covar=tensor([0.1484, 0.1075, 0.1319, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 08:43:15,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3677, 1.6005, 1.4770, 1.2230], device='cuda:0'), covar=tensor([0.2447, 0.2163, 0.1647, 0.2217], device='cuda:0'), in_proj_covar=tensor([0.1897, 0.1824, 0.1742, 0.1881], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 08:43:30,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.153e+02 1.361e+03 1.728e+03 2.372e+03 4.563e+03, threshold=3.455e+03, percent-clipped=5.0
+2023-03-10 08:43:30,105 INFO [train.py:968] (0/2) Epoch 20, batch 14400, giga_loss[loss=0.2539, simple_loss=0.338, pruned_loss=0.08485, over 28071.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3411, pruned_loss=0.08704, over 5652497.90 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3373, pruned_loss=0.08968, over 5692132.73 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3415, pruned_loss=0.0869, over 5662649.75 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:43:35,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4737, 4.3905, 1.7767, 1.6608], device='cuda:0'), covar=tensor([0.1016, 0.0253, 0.0900, 0.1380], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0547, 0.0380, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 08:43:42,632 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-10 08:43:56,343 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:44:34,227 INFO [train.py:968] (0/2) Epoch 20, batch 14450, giga_loss[loss=0.2377, simple_loss=0.3261, pruned_loss=0.07469, over 27665.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.341, pruned_loss=0.08744, over 5659188.85 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3371, pruned_loss=0.0897, over 5697028.97 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3415, pruned_loss=0.08725, over 5661946.19 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:44:55,355 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=881627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:44:59,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=881630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:45:22,066 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=881649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:45:24,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=881652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:45:31,920 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=881659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:45:34,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.265e+02 1.273e+03 1.648e+03 2.283e+03 4.356e+03, threshold=3.296e+03, percent-clipped=3.0
+2023-03-10 08:45:34,202 INFO [train.py:968] (0/2) Epoch 20, batch 14500, giga_loss[loss=0.267, simple_loss=0.3448, pruned_loss=0.09454, over 28450.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3396, pruned_loss=0.08769, over 5670022.78 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3367, pruned_loss=0.08943, over 5701229.25 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3405, pruned_loss=0.08772, over 5667341.97 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:46:03,767 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=881681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:46:11,287 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:46:47,580 INFO [train.py:968] (0/2) Epoch 20, batch 14550, giga_loss[loss=0.259, simple_loss=0.3409, pruned_loss=0.08853, over 29033.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3402, pruned_loss=0.08876, over 5683494.07 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3366, pruned_loss=0.0894, over 5703438.26 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3411, pruned_loss=0.08881, over 5679192.24 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:47:30,650 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4031, 1.8618, 1.5339, 1.5123], device='cuda:0'), covar=tensor([0.2395, 0.2427, 0.2414, 0.2430], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0730, 0.0696, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 08:48:12,391 INFO [train.py:968] (0/2) Epoch 20, batch 14600, giga_loss[loss=0.2231, simple_loss=0.3115, pruned_loss=0.06738, over 28493.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3389, pruned_loss=0.08912, over 5682021.95 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3365, pruned_loss=0.08948, over 5709786.33 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3398, pruned_loss=0.08907, over 5671813.54 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:48:14,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9861, 1.3228, 1.0714, 0.1743], device='cuda:0'), covar=tensor([0.3946, 0.3236, 0.5331, 0.6706], device='cuda:0'), in_proj_covar=tensor([0.1724, 0.1629, 0.1591, 0.1410], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 08:48:14,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.362e+02 1.414e+03 1.751e+03 2.332e+03 5.378e+03, threshold=3.502e+03, percent-clipped=15.0
+2023-03-10 08:48:46,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 08:48:56,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5276, 1.7175, 1.8091, 1.3309], device='cuda:0'), covar=tensor([0.2076, 0.2886, 0.1731, 0.2025], device='cuda:0'), in_proj_covar=tensor([0.0889, 0.0692, 0.0934, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 08:49:26,966 INFO [train.py:968] (0/2) Epoch 20, batch 14650, giga_loss[loss=0.2505, simple_loss=0.3285, pruned_loss=0.08622, over 27659.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3337, pruned_loss=0.08576, over 5684066.62 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3356, pruned_loss=0.08902, over 5713701.01 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3352, pruned_loss=0.0861, over 5671850.84 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:50:05,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=881843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:50:28,316 INFO [train.py:968] (0/2) Epoch 20, batch 14700, giga_loss[loss=0.231, simple_loss=0.3169, pruned_loss=0.07256, over 28763.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3319, pruned_loss=0.08541, over 5690648.41 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3351, pruned_loss=0.08906, over 5719935.54 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3334, pruned_loss=0.08548, over 5674059.41 frames. ], batch size: 263, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:50:31,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.332e+02 1.355e+03 1.639e+03 2.222e+03 7.773e+03, threshold=3.279e+03, percent-clipped=5.0
+2023-03-10 08:50:47,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.46 vs. limit=5.0
+2023-03-10 08:51:02,295 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.42 vs. limit=5.0
+2023-03-10 08:51:31,805 INFO [train.py:968] (0/2) Epoch 20, batch 14750, libri_loss[loss=0.2764, simple_loss=0.3526, pruned_loss=0.1001, over 29535.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3326, pruned_loss=0.08609, over 5687140.20 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3355, pruned_loss=0.08935, over 5725631.85 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3333, pruned_loss=0.08577, over 5667580.57 frames. ], batch size: 83, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:51:51,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:51:54,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6745, 1.8804, 1.5843, 1.7402], device='cuda:0'), covar=tensor([0.2411, 0.2298, 0.2446, 0.2320], device='cuda:0'), in_proj_covar=tensor([0.1483, 0.1073, 0.1316, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 08:52:30,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=881956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:52:37,934 INFO [train.py:968] (0/2) Epoch 20, batch 14800, giga_loss[loss=0.2814, simple_loss=0.3394, pruned_loss=0.1117, over 26879.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3365, pruned_loss=0.08823, over 5679038.81 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3352, pruned_loss=0.08927, over 5718482.35 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3374, pruned_loss=0.08802, over 5669397.37 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:52:39,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.960e+02 1.516e+03 1.942e+03 2.726e+03 1.105e+04, threshold=3.885e+03, percent-clipped=17.0
+2023-03-10 08:53:12,496 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=881986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:53:14,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=881989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:53:26,255 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-882000.pt
+2023-03-10 08:53:38,608 INFO [train.py:968] (0/2) Epoch 20, batch 14850, giga_loss[loss=0.2389, simple_loss=0.3212, pruned_loss=0.07833, over 28667.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3343, pruned_loss=0.08807, over 5685335.21 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3345, pruned_loss=0.08907, over 5724350.93 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3356, pruned_loss=0.08805, over 5671325.89 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:53:48,081 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:54:42,417 INFO [train.py:968] (0/2) Epoch 20, batch 14900, giga_loss[loss=0.2633, simple_loss=0.3371, pruned_loss=0.09478, over 28963.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3351, pruned_loss=0.08928, over 5684758.09 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3347, pruned_loss=0.08933, over 5720870.83 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3359, pruned_loss=0.089, over 5675910.05 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:54:45,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=882062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:54:45,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.375e+03 1.802e+03 2.660e+03 9.763e+03, threshold=3.604e+03, percent-clipped=3.0
+2023-03-10 08:55:26,827 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=882099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:55:29,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=882102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:55:40,028 INFO [train.py:968] (0/2) Epoch 20, batch 14950, giga_loss[loss=0.2778, simple_loss=0.3639, pruned_loss=0.09586, over 28872.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3359, pruned_loss=0.08969, over 5690716.44 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3343, pruned_loss=0.08896, over 5726935.30 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.337, pruned_loss=0.08983, over 5676779.81 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:55:52,107 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 08:56:08,857 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:56:44,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 08:56:47,387 INFO [train.py:968] (0/2) Epoch 20, batch 15000, giga_loss[loss=0.2408, simple_loss=0.3363, pruned_loss=0.0726, over 29073.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3384, pruned_loss=0.09029, over 5689416.25 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3339, pruned_loss=0.08888, over 5729069.09 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3398, pruned_loss=0.09051, over 5675448.28 frames. ], batch size: 128, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:56:47,392 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 08:56:51,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1644, 1.5678, 1.5936, 1.4118], device='cuda:0'), covar=tensor([0.1781, 0.1376, 0.2044, 0.1472], device='cuda:0'), in_proj_covar=tensor([0.0456, 0.0730, 0.0698, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 08:56:56,184 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.1969, simple_loss=0.2981, pruned_loss=0.04791, over 944034.00 frames. 
+2023-03-10 08:56:56,185 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 08:56:57,115 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:56:59,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.187e+02 1.409e+03 2.195e+03 2.973e+03 8.729e+03, threshold=4.390e+03, percent-clipped=18.0
+2023-03-10 08:57:40,888 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:58:01,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6287, 2.2030, 1.5778, 0.9226], device='cuda:0'), covar=tensor([0.5273, 0.2978, 0.4497, 0.5665], device='cuda:0'), in_proj_covar=tensor([0.1725, 0.1630, 0.1592, 0.1407], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 08:58:06,753 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=882205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:58:11,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=882208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:58:13,927 INFO [train.py:968] (0/2) Epoch 20, batch 15050, giga_loss[loss=0.2567, simple_loss=0.3276, pruned_loss=0.09294, over 26911.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3384, pruned_loss=0.08968, over 5684772.97 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3337, pruned_loss=0.08866, over 5732603.39 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3398, pruned_loss=0.09007, over 5669647.66 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:59:02,687 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:59:38,648 INFO [train.py:968] (0/2) Epoch 20, batch 15100, giga_loss[loss=0.2828, simple_loss=0.3457, pruned_loss=0.1099, over 28132.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3359, pruned_loss=0.08943, over 5662067.91 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3335, pruned_loss=0.08858, over 5724718.80 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3371, pruned_loss=0.08983, over 5657363.60 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:59:40,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.211e+02 1.534e+03 1.865e+03 2.551e+03 5.814e+03, threshold=3.730e+03, percent-clipped=3.0
+2023-03-10 09:00:29,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=882302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:00:34,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4684, 1.6322, 1.3086, 1.6796], device='cuda:0'), covar=tensor([0.0750, 0.0311, 0.0344, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0118, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 09:00:44,100 INFO [train.py:968] (0/2) Epoch 20, batch 15150, giga_loss[loss=0.2457, simple_loss=0.3161, pruned_loss=0.08761, over 28723.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3305, pruned_loss=0.08731, over 5668306.03 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3331, pruned_loss=0.08832, over 5727325.27 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3319, pruned_loss=0.08786, over 5660204.38 frames. ], batch size: 99, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:01:06,747 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-10 09:01:42,910 INFO [train.py:968] (0/2) Epoch 20, batch 15200, giga_loss[loss=0.2728, simple_loss=0.3539, pruned_loss=0.09589, over 28324.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3281, pruned_loss=0.08623, over 5662384.01 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3332, pruned_loss=0.08842, over 5719859.69 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.329, pruned_loss=0.08654, over 5660981.52 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:01:48,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.586e+02 1.452e+03 1.971e+03 3.045e+03 1.202e+04, threshold=3.942e+03, percent-clipped=18.0
+2023-03-10 09:02:41,595 INFO [train.py:968] (0/2) Epoch 20, batch 15250, giga_loss[loss=0.2771, simple_loss=0.3555, pruned_loss=0.09934, over 28427.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.33, pruned_loss=0.08768, over 5662814.89 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3332, pruned_loss=0.08836, over 5725867.06 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3306, pruned_loss=0.08794, over 5654502.44 frames. ], batch size: 369, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:03:19,186 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=882445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:03:22,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=882448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:03:37,371 INFO [train.py:968] (0/2) Epoch 20, batch 15300, giga_loss[loss=0.2119, simple_loss=0.2951, pruned_loss=0.06438, over 27559.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3294, pruned_loss=0.08715, over 5669719.01 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3328, pruned_loss=0.08819, over 5727454.69 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3303, pruned_loss=0.0875, over 5660169.11 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:03:43,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.766e+02 1.381e+03 1.748e+03 2.162e+03 6.222e+03, threshold=3.496e+03, percent-clipped=2.0
+2023-03-10 09:04:01,299 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:04:45,885 INFO [train.py:968] (0/2) Epoch 20, batch 15350, giga_loss[loss=0.2461, simple_loss=0.3278, pruned_loss=0.08218, over 28947.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3272, pruned_loss=0.08539, over 5657348.22 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3326, pruned_loss=0.08812, over 5729304.43 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.328, pruned_loss=0.0857, over 5647623.37 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:05:18,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=882536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:05:18,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4709, 1.7205, 1.5966, 1.4275], device='cuda:0'), covar=tensor([0.2568, 0.1969, 0.1571, 0.1978], device='cuda:0'), in_proj_covar=tensor([0.1890, 0.1808, 0.1729, 0.1868], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 09:05:24,971 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:05:50,735 INFO [train.py:968] (0/2) Epoch 20, batch 15400, giga_loss[loss=0.2939, simple_loss=0.354, pruned_loss=0.1169, over 27688.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3258, pruned_loss=0.08438, over 5671458.61 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3322, pruned_loss=0.08796, over 5732916.61 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3267, pruned_loss=0.08469, over 5659035.45 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:05:54,852 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=882563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:05:55,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.795e+02 1.442e+03 1.830e+03 2.372e+03 6.704e+03, threshold=3.660e+03, percent-clipped=9.0
+2023-03-10 09:06:21,627 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6104, 1.8725, 1.7293, 1.5088], device='cuda:0'), covar=tensor([0.2665, 0.2033, 0.1703, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1884, 0.1801, 0.1723, 0.1861], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 09:06:59,776 INFO [train.py:968] (0/2) Epoch 20, batch 15450, giga_loss[loss=0.2053, simple_loss=0.2949, pruned_loss=0.05781, over 29086.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3257, pruned_loss=0.08456, over 5666552.24 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3321, pruned_loss=0.08802, over 5736875.31 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3262, pruned_loss=0.08465, over 5651264.11 frames. ], batch size: 128, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:07:37,734 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882638.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 09:08:04,553 INFO [train.py:968] (0/2) Epoch 20, batch 15500, giga_loss[loss=0.234, simple_loss=0.3135, pruned_loss=0.07727, over 28563.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3261, pruned_loss=0.08439, over 5658114.16 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.332, pruned_loss=0.08802, over 5730662.45 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3264, pruned_loss=0.08438, over 5650148.78 frames. ], batch size: 92, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:08:09,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.606e+02 1.237e+03 1.556e+03 2.143e+03 4.641e+03, threshold=3.111e+03, percent-clipped=4.0
+2023-03-10 09:08:32,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:08:32,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=882679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:08:34,814 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=882682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:08:37,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1602, 4.9829, 4.6990, 2.2181], device='cuda:0'), covar=tensor([0.0444, 0.0555, 0.0701, 0.1994], device='cuda:0'), in_proj_covar=tensor([0.1181, 0.1093, 0.0927, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 09:08:48,344 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-10 09:08:49,866 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:09:09,825 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=882706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:09:12,512 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=882709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:09:13,877 INFO [train.py:968] (0/2) Epoch 20, batch 15550, giga_loss[loss=0.2624, simple_loss=0.3427, pruned_loss=0.0911, over 28448.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3261, pruned_loss=0.08449, over 5663794.64 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3318, pruned_loss=0.08786, over 5733558.10 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3265, pruned_loss=0.08455, over 5653511.68 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:09:14,303 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:09:48,570 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:10:06,397 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 09:10:15,846 INFO [train.py:968] (0/2) Epoch 20, batch 15600, giga_loss[loss=0.2366, simple_loss=0.3197, pruned_loss=0.07674, over 28375.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3256, pruned_loss=0.08465, over 5664608.58 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3318, pruned_loss=0.08789, over 5738838.76 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3257, pruned_loss=0.08459, over 5649538.17 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:10:21,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.237e+02 1.364e+03 1.794e+03 2.580e+03 8.257e+03, threshold=3.588e+03, percent-clipped=20.0
+2023-03-10 09:11:03,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3391, 2.0512, 1.5372, 0.5596], device='cuda:0'), covar=tensor([0.5266, 0.3015, 0.4279, 0.5938], device='cuda:0'), in_proj_covar=tensor([0.1713, 0.1622, 0.1581, 0.1398], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 09:11:16,589 INFO [train.py:968] (0/2) Epoch 20, batch 15650, giga_loss[loss=0.2547, simple_loss=0.3425, pruned_loss=0.08344, over 28773.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3261, pruned_loss=0.08311, over 5659451.92 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3316, pruned_loss=0.0878, over 5724218.58 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3262, pruned_loss=0.08308, over 5659792.96 frames. ], batch size: 243, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:11:50,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2295, 3.4818, 2.3812, 1.0687], device='cuda:0'), covar=tensor([0.7486, 0.2975, 0.3687, 0.6778], device='cuda:0'), in_proj_covar=tensor([0.1705, 0.1615, 0.1574, 0.1392], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 09:12:15,410 INFO [train.py:968] (0/2) Epoch 20, batch 15700, giga_loss[loss=0.2607, simple_loss=0.3457, pruned_loss=0.08783, over 29006.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3296, pruned_loss=0.08421, over 5662965.36 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3315, pruned_loss=0.08768, over 5725493.02 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3298, pruned_loss=0.08418, over 5660396.55 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:12:20,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.409e+02 1.266e+03 1.652e+03 2.348e+03 4.426e+03, threshold=3.304e+03, percent-clipped=3.0
+2023-03-10 09:13:21,763 INFO [train.py:968] (0/2) Epoch 20, batch 15750, giga_loss[loss=0.2744, simple_loss=0.3416, pruned_loss=0.1036, over 26781.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3321, pruned_loss=0.08572, over 5656910.15 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3316, pruned_loss=0.08772, over 5726880.57 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3321, pruned_loss=0.08564, over 5652923.44 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:13:28,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=882917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:14:16,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4941, 2.1832, 1.5835, 0.7506], device='cuda:0'), covar=tensor([0.5083, 0.2704, 0.3866, 0.5321], device='cuda:0'), in_proj_covar=tensor([0.1710, 0.1618, 0.1578, 0.1393], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 09:14:20,526 INFO [train.py:968] (0/2) Epoch 20, batch 15800, giga_loss[loss=0.2349, simple_loss=0.319, pruned_loss=0.0754, over 28727.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3316, pruned_loss=0.085, over 5667423.61 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3316, pruned_loss=0.08768, over 5725598.60 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3317, pruned_loss=0.08493, over 5664220.77 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:14:23,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.028e+02 1.381e+03 2.057e+03 2.834e+03 1.118e+04, threshold=4.114e+03, percent-clipped=15.0
+2023-03-10 09:15:21,912 INFO [train.py:968] (0/2) Epoch 20, batch 15850, giga_loss[loss=0.213, simple_loss=0.3027, pruned_loss=0.0617, over 28332.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3315, pruned_loss=0.08508, over 5677783.91 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3317, pruned_loss=0.08779, over 5728748.35 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3315, pruned_loss=0.08489, over 5671709.71 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:15:24,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883013.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 09:15:45,801 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 09:16:18,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:16:21,924 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 09:16:26,765 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:16:28,000 INFO [train.py:968] (0/2) Epoch 20, batch 15900, giga_loss[loss=0.2496, simple_loss=0.332, pruned_loss=0.08362, over 28977.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3281, pruned_loss=0.08276, over 5683890.72 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3315, pruned_loss=0.08774, over 5730872.38 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3282, pruned_loss=0.08257, over 5676329.68 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:16:31,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:16:32,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.276e+02 1.268e+03 1.697e+03 2.534e+03 5.195e+03, threshold=3.394e+03, percent-clipped=4.0
+2023-03-10 09:16:36,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:16:39,638 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7490, 2.0364, 1.7058, 1.8220], device='cuda:0'), covar=tensor([0.2597, 0.2605, 0.3000, 0.2369], device='cuda:0'), in_proj_covar=tensor([0.1472, 0.1068, 0.1308, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 09:17:06,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:17:25,248 INFO [train.py:968] (0/2) Epoch 20, batch 15950, libri_loss[loss=0.2627, simple_loss=0.3502, pruned_loss=0.08763, over 29738.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3269, pruned_loss=0.08324, over 5682027.88 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3314, pruned_loss=0.08776, over 5734627.19 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3269, pruned_loss=0.0829, over 5670883.98 frames. ], batch size: 87, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:17:44,124 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:18:24,553 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883156.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 09:18:27,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883159.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 09:18:29,882 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:18:30,329 INFO [train.py:968] (0/2) Epoch 20, batch 16000, giga_loss[loss=0.2861, simple_loss=0.358, pruned_loss=0.107, over 29050.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3276, pruned_loss=0.0839, over 5678211.94 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3312, pruned_loss=0.08767, over 5734764.72 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3277, pruned_loss=0.08368, over 5668931.51 frames. ], batch size: 200, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:18:35,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.298e+02 1.412e+03 1.942e+03 2.755e+03 9.115e+03, threshold=3.884e+03, percent-clipped=9.0
+2023-03-10 09:19:05,449 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883188.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 09:19:14,848 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:19:19,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:19:32,877 INFO [train.py:968] (0/2) Epoch 20, batch 16050, giga_loss[loss=0.2571, simple_loss=0.3283, pruned_loss=0.09292, over 27631.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3297, pruned_loss=0.08493, over 5680648.82 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3309, pruned_loss=0.08749, over 5738776.38 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3301, pruned_loss=0.08484, over 5668466.27 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:19:33,360 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:19:37,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:19:38,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-10 09:19:42,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4941, 1.7460, 1.4717, 1.3183], device='cuda:0'), covar=tensor([0.2726, 0.2733, 0.3131, 0.2385], device='cuda:0'), in_proj_covar=tensor([0.1474, 0.1067, 0.1309, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 09:19:58,019 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:20:00,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5545, 1.5491, 1.7383, 1.2307], device='cuda:0'), covar=tensor([0.2228, 0.3381, 0.1730, 0.1958], device='cuda:0'), in_proj_covar=tensor([0.0887, 0.0689, 0.0932, 0.0832], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:0')
+2023-03-10 09:20:14,790 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:20:17,604 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:20:36,860 INFO [train.py:968] (0/2) Epoch 20, batch 16100, giga_loss[loss=0.2446, simple_loss=0.3242, pruned_loss=0.08255, over 28957.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3304, pruned_loss=0.08587, over 5682620.69 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3308, pruned_loss=0.08749, over 5744412.86 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3307, pruned_loss=0.08569, over 5665747.89 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:20:40,443 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0319, 4.8394, 4.6156, 1.9791], device='cuda:0'), covar=tensor([0.0458, 0.0629, 0.0774, 0.2224], device='cuda:0'), in_proj_covar=tensor([0.1181, 0.1092, 0.0930, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 09:20:41,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.912e+02 1.411e+03 1.830e+03 2.767e+03 6.942e+03, threshold=3.660e+03, percent-clipped=11.0
+2023-03-10 09:21:09,105 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 09:21:21,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5198, 1.7657, 1.4376, 1.7534], device='cuda:0'), covar=tensor([0.2542, 0.2416, 0.2734, 0.2154], device='cuda:0'), in_proj_covar=tensor([0.1472, 0.1067, 0.1308, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 09:21:39,360 INFO [train.py:968] (0/2) Epoch 20, batch 16150, giga_loss[loss=0.2678, simple_loss=0.3549, pruned_loss=0.09035, over 28942.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3337, pruned_loss=0.08761, over 5685893.27 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3307, pruned_loss=0.08748, over 5746246.16 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.334, pruned_loss=0.08747, over 5670521.88 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:21:53,185 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3671, 3.5381, 1.5494, 1.5048], device='cuda:0'), covar=tensor([0.1002, 0.0322, 0.0947, 0.1389], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0544, 0.0379, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 09:22:03,612 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 09:22:06,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3250, 4.1320, 3.9350, 1.6966], device='cuda:0'), covar=tensor([0.0695, 0.0841, 0.0970, 0.2269], device='cuda:0'), in_proj_covar=tensor([0.1180, 0.1089, 0.0928, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 09:22:25,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4194, 1.7791, 1.3424, 1.6478], device='cuda:0'), covar=tensor([0.2810, 0.2509, 0.3137, 0.2110], device='cuda:0'), in_proj_covar=tensor([0.1473, 0.1068, 0.1309, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 09:22:26,081 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 09:22:35,739 INFO [train.py:968] (0/2) Epoch 20, batch 16200, giga_loss[loss=0.2488, simple_loss=0.3327, pruned_loss=0.08246, over 28929.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3364, pruned_loss=0.08832, over 5693407.28 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3306, pruned_loss=0.08737, over 5750302.28 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3369, pruned_loss=0.08832, over 5676185.08 frames. ], batch size: 106, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:22:42,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.551e+02 1.455e+03 1.896e+03 2.591e+03 6.465e+03, threshold=3.791e+03, percent-clipped=4.0
+2023-03-10 09:23:34,334 INFO [train.py:968] (0/2) Epoch 20, batch 16250, libri_loss[loss=0.294, simple_loss=0.3621, pruned_loss=0.113, over 29281.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3377, pruned_loss=0.089, over 5688900.79 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3307, pruned_loss=0.08749, over 5747457.53 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3382, pruned_loss=0.08895, over 5674163.58 frames. ], batch size: 94, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:23:53,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4924, 2.2791, 1.6275, 0.6514], device='cuda:0'), covar=tensor([0.6285, 0.2912, 0.4392, 0.6381], device='cuda:0'), in_proj_covar=tensor([0.1707, 0.1615, 0.1577, 0.1393], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 09:24:42,582 INFO [train.py:968] (0/2) Epoch 20, batch 16300, giga_loss[loss=0.2233, simple_loss=0.3085, pruned_loss=0.06906, over 28641.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3353, pruned_loss=0.08774, over 5694916.18 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3301, pruned_loss=0.08732, over 5751249.33 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3365, pruned_loss=0.08788, over 5678161.74 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:24:47,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6057, 3.8020, 1.7270, 1.7031], device='cuda:0'), covar=tensor([0.0946, 0.0340, 0.0941, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0546, 0.0380, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 09:24:48,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.653e+02 1.618e+03 2.231e+03 3.000e+03 8.841e+03, threshold=4.463e+03, percent-clipped=9.0
+2023-03-10 09:25:37,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:25:50,005 INFO [train.py:968] (0/2) Epoch 20, batch 16350, giga_loss[loss=0.2503, simple_loss=0.3391, pruned_loss=0.08079, over 29036.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3338, pruned_loss=0.08747, over 5702342.38 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.33, pruned_loss=0.08728, over 5753722.95 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3349, pruned_loss=0.0876, over 5686069.69 frames. ], batch size: 285, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:26:14,403 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:26:27,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:26:51,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3629, 1.3764, 3.6585, 3.0573], device='cuda:0'), covar=tensor([0.1531, 0.2660, 0.0489, 0.1197], device='cuda:0'), in_proj_covar=tensor([0.0746, 0.0636, 0.0939, 0.0876], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 09:27:02,307 INFO [train.py:968] (0/2) Epoch 20, batch 16400, libri_loss[loss=0.2192, simple_loss=0.2897, pruned_loss=0.07429, over 29365.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3325, pruned_loss=0.08673, over 5679399.45 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.33, pruned_loss=0.08737, over 5753491.88 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3334, pruned_loss=0.08676, over 5666105.64 frames. ], batch size: 71, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:27:06,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.813e+02 1.445e+03 1.811e+03 2.628e+03 5.457e+03, threshold=3.622e+03, percent-clipped=5.0
+2023-03-10 09:28:05,004 INFO [train.py:968] (0/2) Epoch 20, batch 16450, giga_loss[loss=0.2183, simple_loss=0.3075, pruned_loss=0.06451, over 28930.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.331, pruned_loss=0.08668, over 5683967.20 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3297, pruned_loss=0.08716, over 5756589.89 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.332, pruned_loss=0.08689, over 5669435.94 frames. ], batch size: 174, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:28:20,172 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:28:38,969 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:28:45,119 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-10 09:28:50,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:28:53,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:28:56,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5012, 2.4289, 2.0985, 1.7093], device='cuda:0'), covar=tensor([0.0813, 0.0221, 0.0259, 0.1050], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0117, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 09:29:10,559 INFO [train.py:968] (0/2) Epoch 20, batch 16500, giga_loss[loss=0.2666, simple_loss=0.3462, pruned_loss=0.09352, over 27613.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3298, pruned_loss=0.08632, over 5682416.27 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3295, pruned_loss=0.08704, over 5758810.79 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3308, pruned_loss=0.08658, over 5667533.38 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:29:17,299 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.440e+02 1.437e+03 1.920e+03 2.673e+03 7.377e+03, threshold=3.840e+03, percent-clipped=7.0
+2023-03-10 09:29:27,401 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:29:31,588 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:29:34,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2500, 2.8075, 2.4247, 2.5150], device='cuda:0'), covar=tensor([0.1804, 0.1668, 0.1788, 0.1657], device='cuda:0'), in_proj_covar=tensor([0.0448, 0.0724, 0.0690, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 09:29:35,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:30:13,955 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:30:14,335 INFO [train.py:968] (0/2) Epoch 20, batch 16550, giga_loss[loss=0.2323, simple_loss=0.3241, pruned_loss=0.07031, over 28951.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3294, pruned_loss=0.085, over 5680204.55 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3293, pruned_loss=0.08693, over 5761132.49 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3304, pruned_loss=0.08529, over 5665337.67 frames. ], batch size: 164, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:30:17,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4760, 4.3298, 4.0767, 1.9052], device='cuda:0'), covar=tensor([0.0589, 0.0719, 0.0840, 0.2151], device='cuda:0'), in_proj_covar=tensor([0.1179, 0.1092, 0.0929, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 09:31:20,133 INFO [train.py:968] (0/2) Epoch 20, batch 16600, giga_loss[loss=0.276, simple_loss=0.3686, pruned_loss=0.09172, over 28178.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3304, pruned_loss=0.08442, over 5681966.31 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3291, pruned_loss=0.08678, over 5762702.98 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3314, pruned_loss=0.08476, over 5668243.86 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:31:22,883 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:31:25,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:31:26,291 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.137e+02 1.402e+03 1.796e+03 2.501e+03 1.670e+04, threshold=3.592e+03, percent-clipped=14.0
+2023-03-10 09:32:00,759 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:32:06,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6730, 3.5223, 3.3059, 1.8040], device='cuda:0'), covar=tensor([0.0569, 0.0730, 0.0745, 0.1939], device='cuda:0'), in_proj_covar=tensor([0.1177, 0.1090, 0.0927, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 09:32:06,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-10 09:32:19,426 INFO [train.py:968] (0/2) Epoch 20, batch 16650, giga_loss[loss=0.234, simple_loss=0.3221, pruned_loss=0.07296, over 28919.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3325, pruned_loss=0.08435, over 5674883.81 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3287, pruned_loss=0.08663, over 5762759.21 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3337, pruned_loss=0.08472, over 5662498.58 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:32:26,330 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-10 09:32:57,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7221, 4.5556, 4.3069, 1.8964], device='cuda:0'), covar=tensor([0.0506, 0.0631, 0.0724, 0.2055], device='cuda:0'), in_proj_covar=tensor([0.1174, 0.1088, 0.0924, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 09:33:13,025 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 09:33:15,978 INFO [train.py:968] (0/2) Epoch 20, batch 16700, giga_loss[loss=0.246, simple_loss=0.3328, pruned_loss=0.07963, over 28545.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3332, pruned_loss=0.08403, over 5680801.40 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3287, pruned_loss=0.08663, over 5755892.66 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3342, pruned_loss=0.08427, over 5675933.45 frames. ], batch size: 78, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:33:25,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.089e+02 1.329e+03 1.652e+03 2.255e+03 5.421e+03, threshold=3.304e+03, percent-clipped=3.0
+2023-03-10 09:34:05,355 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-10 09:34:13,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:34:27,059 INFO [train.py:968] (0/2) Epoch 20, batch 16750, giga_loss[loss=0.2992, simple_loss=0.3664, pruned_loss=0.116, over 26890.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3338, pruned_loss=0.08478, over 5680813.95 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3285, pruned_loss=0.08655, over 5757521.03 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3348, pruned_loss=0.08502, over 5674859.09 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:35:00,894 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 09:35:23,208 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 09:35:26,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:35:32,700 INFO [train.py:968] (0/2) Epoch 20, batch 16800, libri_loss[loss=0.3059, simple_loss=0.3681, pruned_loss=0.1219, over 20518.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3343, pruned_loss=0.08509, over 5672142.29 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3287, pruned_loss=0.08664, over 5748873.98 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3351, pruned_loss=0.08515, over 5673335.74 frames. ], batch size: 188, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:35:41,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.770e+02 1.419e+03 1.812e+03 2.651e+03 7.287e+03, threshold=3.624e+03, percent-clipped=14.0
+2023-03-10 09:36:25,871 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-884000.pt
+2023-03-10 09:36:36,363 INFO [train.py:968] (0/2) Epoch 20, batch 16850, giga_loss[loss=0.2672, simple_loss=0.3636, pruned_loss=0.08543, over 28736.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.334, pruned_loss=0.08479, over 5674270.62 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3284, pruned_loss=0.08653, over 5752921.02 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3351, pruned_loss=0.08481, over 5667617.01 frames. ], batch size: 262, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:36:39,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=884012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:37:26,550 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=884043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:37:29,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=884046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:37:49,829 INFO [train.py:968] (0/2) Epoch 20, batch 16900, giga_loss[loss=0.3238, simple_loss=0.3917, pruned_loss=0.128, over 28936.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3345, pruned_loss=0.08454, over 5679217.84 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3281, pruned_loss=0.08641, over 5752778.62 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3358, pruned_loss=0.08463, over 5672367.33 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:38:00,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.450e+02 1.305e+03 1.635e+03 2.257e+03 8.335e+03, threshold=3.270e+03, percent-clipped=4.0
+2023-03-10 09:38:13,071 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=884075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:38:21,350 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.50 vs. limit=5.0
+2023-03-10 09:38:41,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6579, 1.9447, 1.5635, 2.0532], device='cuda:0'), covar=tensor([0.2704, 0.2678, 0.3116, 0.2374], device='cuda:0'), in_proj_covar=tensor([0.1477, 0.1070, 0.1313, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 09:39:05,822 INFO [train.py:968] (0/2) Epoch 20, batch 16950, giga_loss[loss=0.2581, simple_loss=0.3424, pruned_loss=0.08694, over 27621.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3387, pruned_loss=0.08652, over 5679866.65 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3278, pruned_loss=0.08621, over 5754356.69 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3401, pruned_loss=0.08678, over 5672368.72 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:39:44,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=884138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:40:07,040 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=884155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:40:10,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=884158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:40:15,581 INFO [train.py:968] (0/2) Epoch 20, batch 17000, giga_loss[loss=0.2468, simple_loss=0.3269, pruned_loss=0.08331, over 28589.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3384, pruned_loss=0.08615, over 5685394.62 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3277, pruned_loss=0.08618, over 5757453.17 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3398, pruned_loss=0.08638, over 5675043.46 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:40:25,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.331e+02 1.449e+03 1.874e+03 2.659e+03 8.902e+03, threshold=3.748e+03, percent-clipped=14.0
+2023-03-10 09:40:51,246 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=884187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:41:26,779 INFO [train.py:968] (0/2) Epoch 20, batch 17050, giga_loss[loss=0.2226, simple_loss=0.3072, pruned_loss=0.06904, over 28953.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3369, pruned_loss=0.08613, over 5693995.75 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3279, pruned_loss=0.08614, over 5760009.09 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.338, pruned_loss=0.08634, over 5682590.63 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:41:49,602 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8164, 4.8457, 1.9485, 1.8265], device='cuda:0'), covar=tensor([0.0909, 0.0277, 0.0875, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0546, 0.0381, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 09:42:39,159 INFO [train.py:968] (0/2) Epoch 20, batch 17100, giga_loss[loss=0.219, simple_loss=0.3178, pruned_loss=0.06012, over 28420.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3342, pruned_loss=0.08435, over 5697355.06 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.328, pruned_loss=0.0861, over 5763009.73 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3351, pruned_loss=0.08453, over 5684520.34 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:42:50,337 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.285e+02 1.315e+03 1.569e+03 1.986e+03 6.222e+03, threshold=3.139e+03, percent-clipped=2.0
+2023-03-10 09:43:51,515 INFO [train.py:968] (0/2) Epoch 20, batch 17150, giga_loss[loss=0.2106, simple_loss=0.3014, pruned_loss=0.05987, over 28798.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3321, pruned_loss=0.08227, over 5708637.03 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3283, pruned_loss=0.08623, over 5766330.05 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3326, pruned_loss=0.08221, over 5693846.52 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:44:18,421 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=884330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:44:54,564 INFO [train.py:968] (0/2) Epoch 20, batch 17200, giga_loss[loss=0.2514, simple_loss=0.3381, pruned_loss=0.08232, over 28754.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3327, pruned_loss=0.08339, over 5697135.27 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3283, pruned_loss=0.08635, over 5767109.91 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3333, pruned_loss=0.08312, over 5682855.19 frames. ], batch size: 262, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:45:03,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.357e+02 1.284e+03 1.578e+03 2.199e+03 1.134e+04, threshold=3.157e+03, percent-clipped=11.0
+2023-03-10 09:45:30,563 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 09:45:55,446 INFO [train.py:968] (0/2) Epoch 20, batch 17250, giga_loss[loss=0.3188, simple_loss=0.3918, pruned_loss=0.1229, over 28895.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3344, pruned_loss=0.084, over 5696962.82 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3283, pruned_loss=0.08627, over 5769148.85 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.335, pruned_loss=0.08383, over 5682918.05 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:46:21,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3459, 1.6847, 1.3202, 1.3129], device='cuda:0'), covar=tensor([0.2691, 0.2635, 0.3087, 0.2305], device='cuda:0'), in_proj_covar=tensor([0.1476, 0.1070, 0.1313, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 09:46:53,101 INFO [train.py:968] (0/2) Epoch 20, batch 17300, giga_loss[loss=0.3006, simple_loss=0.3685, pruned_loss=0.1163, over 28685.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3355, pruned_loss=0.0854, over 5683716.93 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3283, pruned_loss=0.08636, over 5762198.60 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3361, pruned_loss=0.08519, over 5676646.49 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:47:04,619 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.004e+02 1.437e+03 2.009e+03 2.897e+03 8.661e+03, threshold=4.018e+03, percent-clipped=22.0
+2023-03-10 09:47:07,498 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=884473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:47:13,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=884476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:47:44,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=884505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:47:50,118 INFO [train.py:968] (0/2) Epoch 20, batch 17350, giga_loss[loss=0.3177, simple_loss=0.3726, pruned_loss=0.1314, over 27626.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3328, pruned_loss=0.08497, over 5677431.25 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3285, pruned_loss=0.0864, over 5757518.92 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3332, pruned_loss=0.08474, over 5674566.84 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:47:54,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=884513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:48:26,353 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=884541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:48:38,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3782, 1.9171, 1.4364, 1.6522], device='cuda:0'), covar=tensor([0.0774, 0.0305, 0.0340, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 09:48:48,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0723, 1.4636, 1.3061, 1.2395], device='cuda:0'), covar=tensor([0.1784, 0.1482, 0.1934, 0.1594], device='cuda:0'), in_proj_covar=tensor([0.0447, 0.0724, 0.0692, 0.0663], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 09:48:50,651 INFO [train.py:968] (0/2) Epoch 20, batch 17400, giga_loss[loss=0.234, simple_loss=0.3193, pruned_loss=0.07433, over 29011.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3316, pruned_loss=0.0849, over 5680960.88 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3281, pruned_loss=0.08628, over 5759119.39 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3323, pruned_loss=0.08479, over 5675655.87 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:48:58,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.699e+02 1.375e+03 1.763e+03 2.917e+03 7.960e+03, threshold=3.526e+03, percent-clipped=9.0
+2023-03-10 09:49:06,615 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=884575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:49:44,588 INFO [train.py:968] (0/2) Epoch 20, batch 17450, giga_loss[loss=0.3055, simple_loss=0.3586, pruned_loss=0.1262, over 24010.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3361, pruned_loss=0.08788, over 5686077.62 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3279, pruned_loss=0.08617, over 5762370.27 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3371, pruned_loss=0.08791, over 5676645.89 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:50:33,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=884656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:50:36,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=884659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:50:38,489 INFO [train.py:968] (0/2) Epoch 20, batch 17500, giga_loss[loss=0.2959, simple_loss=0.3761, pruned_loss=0.1079, over 28841.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.345, pruned_loss=0.0929, over 5689438.14 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3279, pruned_loss=0.08613, over 5763678.15 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.346, pruned_loss=0.09301, over 5679956.35 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:50:45,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.139e+02 1.398e+03 1.813e+03 2.497e+03 9.310e+03, threshold=3.627e+03, percent-clipped=8.0
+2023-03-10 09:51:01,107 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=884688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:51:20,181 INFO [train.py:968] (0/2) Epoch 20, batch 17550, giga_loss[loss=0.2744, simple_loss=0.3528, pruned_loss=0.09796, over 28234.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3507, pruned_loss=0.09594, over 5700890.08 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3281, pruned_loss=0.08611, over 5767692.37 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3518, pruned_loss=0.09625, over 5688037.89 frames. ], batch size: 77, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:52:07,680 INFO [train.py:968] (0/2) Epoch 20, batch 17600, giga_loss[loss=0.2379, simple_loss=0.3219, pruned_loss=0.07692, over 29027.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3473, pruned_loss=0.09524, over 5689209.67 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3285, pruned_loss=0.08639, over 5759513.66 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3481, pruned_loss=0.09533, over 5686328.75 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:52:18,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.322e+02 1.159e+03 1.494e+03 1.979e+03 7.982e+03, threshold=2.988e+03, percent-clipped=5.0
+2023-03-10 09:52:52,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1698, 1.3546, 3.3019, 2.9795], device='cuda:0'), covar=tensor([0.1554, 0.2558, 0.0496, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0743, 0.0635, 0.0937, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 09:52:55,324 INFO [train.py:968] (0/2) Epoch 20, batch 17650, giga_loss[loss=0.2624, simple_loss=0.3225, pruned_loss=0.1011, over 29035.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3383, pruned_loss=0.09138, over 5683830.35 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3285, pruned_loss=0.08639, over 5759513.66 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3389, pruned_loss=0.09145, over 5681588.10 frames. ], batch size: 106, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:53:41,250 INFO [train.py:968] (0/2) Epoch 20, batch 17700, giga_loss[loss=0.2104, simple_loss=0.2845, pruned_loss=0.06816, over 28680.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3313, pruned_loss=0.08852, over 5686117.73 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3286, pruned_loss=0.08641, over 5761332.44 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3318, pruned_loss=0.08863, over 5680883.22 frames. ], batch size: 92, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:53:48,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.010e+02 1.071e+03 1.389e+03 2.050e+03 4.181e+03, threshold=2.777e+03, percent-clipped=8.0
+2023-03-10 09:54:26,701 INFO [train.py:968] (0/2) Epoch 20, batch 17750, libri_loss[loss=0.2166, simple_loss=0.2998, pruned_loss=0.06665, over 29603.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3244, pruned_loss=0.08559, over 5686650.79 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3287, pruned_loss=0.08633, over 5765014.93 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3246, pruned_loss=0.08577, over 5676910.27 frames. ], batch size: 74, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:54:30,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=884916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:54:58,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=884950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:55:07,009 INFO [train.py:968] (0/2) Epoch 20, batch 17800, giga_loss[loss=0.1822, simple_loss=0.2617, pruned_loss=0.05135, over 28676.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3195, pruned_loss=0.08374, over 5687120.79 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3289, pruned_loss=0.08637, over 5760049.71 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3192, pruned_loss=0.08379, over 5680426.30 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:55:14,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.008e+02 1.256e+03 1.513e+03 2.277e+03 1.042e+04, threshold=3.027e+03, percent-clipped=14.0
+2023-03-10 09:55:18,807 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=884976.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 09:55:41,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9578, 1.3014, 1.0801, 0.1813], device='cuda:0'), covar=tensor([0.4679, 0.4036, 0.5405, 0.6992], device='cuda:0'), in_proj_covar=tensor([0.1717, 0.1632, 0.1586, 0.1399], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 09:55:49,422 INFO [train.py:968] (0/2) Epoch 20, batch 17850, libri_loss[loss=0.3391, simple_loss=0.402, pruned_loss=0.1382, over 29653.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3149, pruned_loss=0.08152, over 5695356.83 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3294, pruned_loss=0.08662, over 5762874.04 frames. ], giga_tot_loss[loss=0.2382, simple_loss=0.3139, pruned_loss=0.08121, over 5686008.64 frames. ], batch size: 88, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:56:27,275 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=885059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:56:28,098 INFO [train.py:968] (0/2) Epoch 20, batch 17900, libri_loss[loss=0.3313, simple_loss=0.391, pruned_loss=0.1358, over 29646.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3124, pruned_loss=0.08054, over 5702218.29 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3297, pruned_loss=0.08671, over 5764138.51 frames. ], giga_tot_loss[loss=0.2355, simple_loss=0.3109, pruned_loss=0.08004, over 5692034.41 frames. ], batch size: 91, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:56:30,245 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=885062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:56:39,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.871e+02 1.086e+03 1.583e+03 2.507e+03 6.719e+03, threshold=3.166e+03, percent-clipped=12.0
+2023-03-10 09:56:57,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=885091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:57:01,344 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=885093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:57:03,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=885096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:57:14,391 INFO [train.py:968] (0/2) Epoch 20, batch 17950, giga_loss[loss=0.2268, simple_loss=0.295, pruned_loss=0.07926, over 28600.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3093, pruned_loss=0.07936, over 5699088.36 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3299, pruned_loss=0.08677, over 5766691.42 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.3075, pruned_loss=0.07876, over 5687338.39 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:57:25,556 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=885125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:57:54,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5229, 2.8850, 1.6929, 1.6412], device='cuda:0'), covar=tensor([0.0801, 0.0315, 0.0710, 0.1095], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0544, 0.0379, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 09:57:58,433 INFO [train.py:968] (0/2) Epoch 20, batch 18000, giga_loss[loss=0.2107, simple_loss=0.2949, pruned_loss=0.06325, over 29073.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3066, pruned_loss=0.07821, over 5694849.36 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3302, pruned_loss=0.08693, over 5757804.20 frames. ], giga_tot_loss[loss=0.2296, simple_loss=0.3044, pruned_loss=0.07742, over 5693158.73 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:57:58,437 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 09:58:07,477 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2054, simple_loss=0.3118, pruned_loss=0.04946, over 944034.00 frames. 
+2023-03-10 09:58:07,478 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 09:58:13,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.918e+02 9.587e+02 1.223e+03 1.542e+03 2.916e+03, threshold=2.446e+03, percent-clipped=0.0
+2023-03-10 09:58:52,553 INFO [train.py:968] (0/2) Epoch 20, batch 18050, giga_loss[loss=0.2149, simple_loss=0.2846, pruned_loss=0.07262, over 28732.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3037, pruned_loss=0.07696, over 5678590.44 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3303, pruned_loss=0.08686, over 5748496.10 frames. ], giga_tot_loss[loss=0.2269, simple_loss=0.3014, pruned_loss=0.07621, over 5683644.64 frames. ], batch size: 99, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:59:33,361 INFO [train.py:968] (0/2) Epoch 20, batch 18100, giga_loss[loss=0.2365, simple_loss=0.3103, pruned_loss=0.08132, over 28553.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3023, pruned_loss=0.07606, over 5677070.36 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3309, pruned_loss=0.0871, over 5739797.51 frames. ], giga_tot_loss[loss=0.2243, simple_loss=0.299, pruned_loss=0.07485, over 5685749.71 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:59:43,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.145e+02 1.136e+03 1.431e+03 1.824e+03 6.512e+03, threshold=2.861e+03, percent-clipped=10.0
+2023-03-10 10:00:21,160 INFO [train.py:968] (0/2) Epoch 20, batch 18150, giga_loss[loss=0.2071, simple_loss=0.2827, pruned_loss=0.06572, over 28865.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.2987, pruned_loss=0.07469, over 5677112.83 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3313, pruned_loss=0.08727, over 5737982.03 frames. ], giga_tot_loss[loss=0.2212, simple_loss=0.2954, pruned_loss=0.07348, over 5684975.00 frames. ], batch size: 112, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:00:46,830 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:00:52,969 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-10 10:01:01,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=885351.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:01:08,799 INFO [train.py:968] (0/2) Epoch 20, batch 18200, giga_loss[loss=0.2055, simple_loss=0.2763, pruned_loss=0.06733, over 28983.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2967, pruned_loss=0.07406, over 5673704.51 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3314, pruned_loss=0.0873, over 5739461.39 frames. ], giga_tot_loss[loss=0.2195, simple_loss=0.2933, pruned_loss=0.07282, over 5677289.44 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:01:16,927 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.153e+02 1.098e+03 1.396e+03 2.015e+03 5.202e+03, threshold=2.791e+03, percent-clipped=8.0
+2023-03-10 10:01:17,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4285, 1.2502, 4.4457, 3.3126], device='cuda:0'), covar=tensor([0.1713, 0.2910, 0.0410, 0.1125], device='cuda:0'), in_proj_covar=tensor([0.0743, 0.0634, 0.0938, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 10:01:20,931 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6460, 1.8001, 1.8673, 1.4495], device='cuda:0'), covar=tensor([0.1751, 0.2453, 0.1415, 0.1735], device='cuda:0'), in_proj_covar=tensor([0.0897, 0.0695, 0.0944, 0.0841], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 10:01:48,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885407.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:01:51,697 INFO [train.py:968] (0/2) Epoch 20, batch 18250, giga_loss[loss=0.2503, simple_loss=0.3203, pruned_loss=0.09021, over 28801.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2953, pruned_loss=0.07385, over 5679688.33 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3318, pruned_loss=0.08751, over 5743001.35 frames. ], giga_tot_loss[loss=0.2183, simple_loss=0.2917, pruned_loss=0.07243, over 5678321.04 frames. ], batch size: 199, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:02:38,849 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 10:02:45,938 INFO [train.py:968] (0/2) Epoch 20, batch 18300, giga_loss[loss=0.2706, simple_loss=0.3458, pruned_loss=0.09769, over 28703.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3059, pruned_loss=0.07952, over 5672665.35 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3319, pruned_loss=0.08755, over 5744383.09 frames. ], giga_tot_loss[loss=0.2296, simple_loss=0.3026, pruned_loss=0.07827, over 5669929.22 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:02:54,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.415e+02 1.123e+03 1.388e+03 2.047e+03 8.150e+03, threshold=2.776e+03, percent-clipped=9.0
+2023-03-10 10:03:18,971 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=885494.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:03:21,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=885497.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:03:33,784 INFO [train.py:968] (0/2) Epoch 20, batch 18350, giga_loss[loss=0.3141, simple_loss=0.3775, pruned_loss=0.1254, over 27869.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.32, pruned_loss=0.0869, over 5678692.25 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3321, pruned_loss=0.08764, over 5743496.45 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3168, pruned_loss=0.08572, over 5675604.23 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:03:46,677 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=885526.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:03:49,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7217, 1.8459, 1.2320, 1.4225], device='cuda:0'), covar=tensor([0.0972, 0.0696, 0.1181, 0.1172], device='cuda:0'), in_proj_covar=tensor([0.0388, 0.0440, 0.0512, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 10:04:15,676 INFO [train.py:968] (0/2) Epoch 20, batch 18400, giga_loss[loss=0.2856, simple_loss=0.3604, pruned_loss=0.1054, over 28115.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.331, pruned_loss=0.09238, over 5687346.18 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3323, pruned_loss=0.08758, over 5746861.94 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3282, pruned_loss=0.09156, over 5680766.98 frames. ], batch size: 77, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:04:24,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.191e+02 1.570e+03 2.019e+03 2.867e+03 9.532e+03, threshold=4.039e+03, percent-clipped=25.0
+2023-03-10 10:04:58,394 INFO [train.py:968] (0/2) Epoch 20, batch 18450, giga_loss[loss=0.3719, simple_loss=0.4279, pruned_loss=0.158, over 27966.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3381, pruned_loss=0.09498, over 5693971.74 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.333, pruned_loss=0.08785, over 5752242.63 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3353, pruned_loss=0.09428, over 5681950.55 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:05:23,798 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:05:39,504 INFO [train.py:968] (0/2) Epoch 20, batch 18500, giga_loss[loss=0.2691, simple_loss=0.353, pruned_loss=0.09261, over 28743.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.34, pruned_loss=0.09443, over 5686226.30 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3336, pruned_loss=0.08812, over 5744180.68 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3374, pruned_loss=0.09376, over 5683077.76 frames. ], batch size: 262, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:05:48,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.673e+02 1.143e+03 1.410e+03 1.876e+03 4.673e+03, threshold=2.821e+03, percent-clipped=2.0
+2023-03-10 10:05:53,471 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:05:54,588 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:06:07,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:06:16,540 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-10 10:06:27,441 INFO [train.py:968] (0/2) Epoch 20, batch 18550, libri_loss[loss=0.2062, simple_loss=0.2895, pruned_loss=0.06147, over 29365.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3411, pruned_loss=0.09404, over 5679244.40 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3337, pruned_loss=0.08802, over 5749245.69 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3391, pruned_loss=0.09381, over 5669757.61 frames. ], batch size: 67, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:06:29,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=885712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:06:54,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3647, 1.5114, 1.3299, 1.3856], device='cuda:0'), covar=tensor([0.0772, 0.0338, 0.0331, 0.0869], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0118, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 10:07:13,541 INFO [train.py:968] (0/2) Epoch 20, batch 18600, giga_loss[loss=0.278, simple_loss=0.3342, pruned_loss=0.1109, over 23582.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3432, pruned_loss=0.09595, over 5676863.53 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3335, pruned_loss=0.08783, over 5750679.64 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.342, pruned_loss=0.09608, over 5666629.93 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:07:21,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.901e+02 1.129e+03 1.340e+03 1.757e+03 4.576e+03, threshold=2.680e+03, percent-clipped=5.0
+2023-03-10 10:07:30,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=885782.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:07:59,033 INFO [train.py:968] (0/2) Epoch 20, batch 18650, giga_loss[loss=0.3213, simple_loss=0.3867, pruned_loss=0.128, over 28915.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3465, pruned_loss=0.0985, over 5677268.90 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3335, pruned_loss=0.08765, over 5752364.26 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3457, pruned_loss=0.09894, over 5666193.45 frames. ], batch size: 174, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:08:37,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=885855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:08:40,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=885858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:08:42,560 INFO [train.py:968] (0/2) Epoch 20, batch 18700, giga_loss[loss=0.2644, simple_loss=0.3389, pruned_loss=0.09491, over 28593.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3498, pruned_loss=0.1003, over 5673582.89 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3335, pruned_loss=0.08766, over 5744436.67 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3494, pruned_loss=0.1008, over 5670365.57 frames. ], batch size: 78, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:08:52,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.588e+02 1.229e+03 1.499e+03 1.911e+03 6.298e+03, threshold=2.998e+03, percent-clipped=8.0
+2023-03-10 10:09:05,365 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-10 10:09:05,898 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=885887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:09:09,735 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6559, 2.2739, 1.5873, 0.8760], device='cuda:0'), covar=tensor([0.6084, 0.2998, 0.4629, 0.5801], device='cuda:0'), in_proj_covar=tensor([0.1718, 0.1626, 0.1586, 0.1398], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 10:09:25,779 INFO [train.py:968] (0/2) Epoch 20, batch 18750, giga_loss[loss=0.2843, simple_loss=0.3693, pruned_loss=0.09961, over 28725.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3526, pruned_loss=0.101, over 5679110.25 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3337, pruned_loss=0.08773, over 5744159.07 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3523, pruned_loss=0.1015, over 5675663.72 frames. ], batch size: 99, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:09:37,220 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=885925.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:09:41,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=885928.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:10:05,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-10 10:10:07,088 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=885957.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:10:11,711 INFO [train.py:968] (0/2) Epoch 20, batch 18800, giga_loss[loss=0.2786, simple_loss=0.3583, pruned_loss=0.0994, over 28818.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3552, pruned_loss=0.1017, over 5671626.80 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.334, pruned_loss=0.08777, over 5736630.73 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.355, pruned_loss=0.1022, over 5675404.22 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:10:20,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.475e+02 1.160e+03 1.466e+03 1.804e+03 3.272e+03, threshold=2.933e+03, percent-clipped=5.0
+2023-03-10 10:10:29,450 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:10:41,860 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-886000.pt
+2023-03-10 10:10:53,577 INFO [train.py:968] (0/2) Epoch 20, batch 18850, giga_loss[loss=0.2745, simple_loss=0.364, pruned_loss=0.09253, over 28986.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3563, pruned_loss=0.1013, over 5681272.91 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3346, pruned_loss=0.08794, over 5739969.90 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.356, pruned_loss=0.1018, over 5680235.10 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:10:56,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:11:24,571 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:11:25,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:11:37,157 INFO [train.py:968] (0/2) Epoch 20, batch 18900, giga_loss[loss=0.258, simple_loss=0.3467, pruned_loss=0.08465, over 28975.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3559, pruned_loss=0.09967, over 5695898.46 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3347, pruned_loss=0.08791, over 5741655.03 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3558, pruned_loss=0.1002, over 5693085.06 frames. ], batch size: 128, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:11:38,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:11:45,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.265e+02 1.192e+03 1.433e+03 1.923e+03 3.848e+03, threshold=2.866e+03, percent-clipped=3.0
+2023-03-10 10:12:17,100 INFO [train.py:968] (0/2) Epoch 20, batch 18950, giga_loss[loss=0.2704, simple_loss=0.3514, pruned_loss=0.09466, over 28527.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3535, pruned_loss=0.09712, over 5701916.17 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3349, pruned_loss=0.08771, over 5742862.52 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3539, pruned_loss=0.09808, over 5697267.07 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:12:54,786 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:12:56,492 INFO [train.py:968] (0/2) Epoch 20, batch 19000, giga_loss[loss=0.3565, simple_loss=0.4219, pruned_loss=0.1455, over 28679.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.353, pruned_loss=0.09701, over 5700618.92 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3347, pruned_loss=0.08749, over 5738732.90 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.354, pruned_loss=0.09828, over 5699456.28 frames. ], batch size: 242, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:12:56,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:07,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.035e+02 1.168e+03 1.344e+03 1.713e+03 6.596e+03, threshold=2.687e+03, percent-clipped=7.0
+2023-03-10 10:13:20,177 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:23,938 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:24,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:25,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:26,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:36,563 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:40,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:41,752 INFO [train.py:968] (0/2) Epoch 20, batch 19050, giga_loss[loss=0.295, simple_loss=0.3588, pruned_loss=0.1156, over 28620.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.356, pruned_loss=0.1013, over 5708669.98 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3351, pruned_loss=0.08767, over 5744073.97 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.357, pruned_loss=0.1024, over 5702153.44 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:13:50,012 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:53,502 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:54,221 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:14:05,013 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:14:25,932 INFO [train.py:968] (0/2) Epoch 20, batch 19100, giga_loss[loss=0.3158, simple_loss=0.3694, pruned_loss=0.131, over 28872.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3578, pruned_loss=0.1049, over 5702653.45 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3357, pruned_loss=0.08782, over 5736485.47 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3587, pruned_loss=0.1061, over 5703816.78 frames. ], batch size: 99, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:14:36,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.482e+03 2.075e+03 3.172e+03 2.287e+04, threshold=4.150e+03, percent-clipped=31.0
+2023-03-10 10:15:04,959 INFO [train.py:968] (0/2) Epoch 20, batch 19150, libri_loss[loss=0.2282, simple_loss=0.3153, pruned_loss=0.07054, over 29539.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3568, pruned_loss=0.1051, over 5699887.42 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3358, pruned_loss=0.08781, over 5730020.68 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.358, pruned_loss=0.1066, over 5705911.46 frames. ], batch size: 78, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:15:24,934 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:15:45,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2540, 1.3092, 1.1668, 1.2129], device='cuda:0'), covar=tensor([0.2043, 0.1967, 0.1666, 0.1885], device='cuda:0'), in_proj_covar=tensor([0.1922, 0.1828, 0.1763, 0.1907], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 10:15:47,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:15:50,294 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:15:50,717 INFO [train.py:968] (0/2) Epoch 20, batch 19200, giga_loss[loss=0.3178, simple_loss=0.3666, pruned_loss=0.1345, over 26523.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3539, pruned_loss=0.1042, over 5693477.77 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3361, pruned_loss=0.08799, over 5732875.62 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3549, pruned_loss=0.1055, over 5695192.76 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:16:01,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.878e+02 1.185e+03 1.501e+03 1.990e+03 5.939e+03, threshold=3.003e+03, percent-clipped=2.0
+2023-03-10 10:16:31,333 INFO [train.py:968] (0/2) Epoch 20, batch 19250, libri_loss[loss=0.2774, simple_loss=0.3637, pruned_loss=0.0956, over 26175.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3521, pruned_loss=0.1029, over 5700641.20 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3369, pruned_loss=0.08822, over 5733331.83 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3528, pruned_loss=0.1043, over 5700065.79 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:16:40,770 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:16:48,657 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886429.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:16:50,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5073, 4.3315, 4.1110, 2.0763], device='cuda:0'), covar=tensor([0.0493, 0.0605, 0.0624, 0.2149], device='cuda:0'), in_proj_covar=tensor([0.1179, 0.1092, 0.0928, 0.0710], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 10:17:00,761 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:17:02,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9639, 1.3140, 1.0986, 0.1724], device='cuda:0'), covar=tensor([0.3947, 0.3060, 0.4612, 0.6432], device='cuda:0'), in_proj_covar=tensor([0.1706, 0.1615, 0.1578, 0.1388], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 10:17:13,747 INFO [train.py:968] (0/2) Epoch 20, batch 19300, giga_loss[loss=0.2655, simple_loss=0.3462, pruned_loss=0.09242, over 28952.00 frames. ], tot_loss[loss=0.277, simple_loss=0.351, pruned_loss=0.1015, over 5700494.37 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.337, pruned_loss=0.08819, over 5726846.84 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3517, pruned_loss=0.1029, over 5706015.35 frames. ], batch size: 164, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:17:24,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.744e+02 1.273e+03 1.588e+03 2.209e+03 7.805e+03, threshold=3.176e+03, percent-clipped=10.0
+2023-03-10 10:17:51,372 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:17:54,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:18:02,588 INFO [train.py:968] (0/2) Epoch 20, batch 19350, giga_loss[loss=0.2615, simple_loss=0.3336, pruned_loss=0.09468, over 28416.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3483, pruned_loss=0.09971, over 5694416.32 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3369, pruned_loss=0.08814, over 5727135.03 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3492, pruned_loss=0.101, over 5698239.71 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:18:12,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4270, 4.1873, 3.9832, 1.9334], device='cuda:0'), covar=tensor([0.0602, 0.0793, 0.0759, 0.2184], device='cuda:0'), in_proj_covar=tensor([0.1178, 0.1093, 0.0930, 0.0710], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 10:18:17,328 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:18:20,138 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:18:46,552 INFO [train.py:968] (0/2) Epoch 20, batch 19400, giga_loss[loss=0.218, simple_loss=0.3026, pruned_loss=0.06666, over 29006.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3428, pruned_loss=0.09668, over 5691090.77 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3364, pruned_loss=0.08788, over 5730179.76 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.344, pruned_loss=0.0981, over 5690766.43 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:18:58,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.402e+02 1.048e+03 1.247e+03 1.648e+03 4.745e+03, threshold=2.494e+03, percent-clipped=2.0
+2023-03-10 10:19:21,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:19:34,973 INFO [train.py:968] (0/2) Epoch 20, batch 19450, libri_loss[loss=0.2859, simple_loss=0.3642, pruned_loss=0.1038, over 29656.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3383, pruned_loss=0.09447, over 5681952.78 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3369, pruned_loss=0.08805, over 5734941.53 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3389, pruned_loss=0.09567, over 5676058.04 frames. ], batch size: 88, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:20:04,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2489, 1.4170, 1.2562, 1.4517], device='cuda:0'), covar=tensor([0.0778, 0.0355, 0.0348, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0116, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 10:20:21,192 INFO [train.py:968] (0/2) Epoch 20, batch 19500, giga_loss[loss=0.2148, simple_loss=0.3019, pruned_loss=0.06378, over 28865.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3329, pruned_loss=0.09179, over 5672250.53 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3373, pruned_loss=0.08816, over 5738058.09 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.333, pruned_loss=0.09278, over 5663275.12 frames. ], batch size: 174, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:20:34,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.603e+02 1.010e+03 1.250e+03 1.817e+03 6.095e+03, threshold=2.500e+03, percent-clipped=11.0
+2023-03-10 10:20:49,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1896, 1.7859, 1.4748, 0.4362], device='cuda:0'), covar=tensor([0.4563, 0.2836, 0.4101, 0.5693], device='cuda:0'), in_proj_covar=tensor([0.1701, 0.1610, 0.1570, 0.1383], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 10:21:09,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:21:09,393 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-10 10:21:12,150 INFO [train.py:968] (0/2) Epoch 20, batch 19550, giga_loss[loss=0.2429, simple_loss=0.3228, pruned_loss=0.0815, over 28872.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3311, pruned_loss=0.09069, over 5660437.46 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3371, pruned_loss=0.08801, over 5739965.77 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3313, pruned_loss=0.09161, over 5651261.66 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:21:32,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:21:34,405 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:21:36,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:21:50,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3082, 1.5540, 1.5994, 1.1628], device='cuda:0'), covar=tensor([0.1606, 0.2399, 0.1277, 0.1515], device='cuda:0'), in_proj_covar=tensor([0.0889, 0.0691, 0.0935, 0.0834], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 10:21:53,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0625, 1.1806, 1.1251, 1.0247], device='cuda:0'), covar=tensor([0.2803, 0.2541, 0.1781, 0.2291], device='cuda:0'), in_proj_covar=tensor([0.1927, 0.1834, 0.1771, 0.1917], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 10:21:54,234 INFO [train.py:968] (0/2) Epoch 20, batch 19600, giga_loss[loss=0.3004, simple_loss=0.3731, pruned_loss=0.1139, over 28083.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3331, pruned_loss=0.09164, over 5670923.42 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3376, pruned_loss=0.08811, over 5743900.25 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3326, pruned_loss=0.09238, over 5657956.56 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:22:04,013 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:22:07,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.785e+02 1.135e+03 1.335e+03 1.704e+03 6.303e+03, threshold=2.670e+03, percent-clipped=8.0
+2023-03-10 10:22:24,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:22:30,155 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 10:22:31,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886804.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:22:36,608 INFO [train.py:968] (0/2) Epoch 20, batch 19650, giga_loss[loss=0.2363, simple_loss=0.317, pruned_loss=0.07775, over 28963.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3321, pruned_loss=0.09102, over 5673059.34 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3376, pruned_loss=0.08808, over 5738537.68 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3316, pruned_loss=0.09172, over 5665314.24 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:22:41,861 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-10 10:22:43,424 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:13,034 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:15,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:19,522 INFO [train.py:968] (0/2) Epoch 20, batch 19700, giga_loss[loss=0.2175, simple_loss=0.3026, pruned_loss=0.06619, over 28889.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3316, pruned_loss=0.0906, over 5672678.93 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3384, pruned_loss=0.08841, over 5732838.03 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3303, pruned_loss=0.09091, over 5671077.08 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:23:29,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.453e+02 1.045e+03 1.327e+03 1.827e+03 4.647e+03, threshold=2.654e+03, percent-clipped=11.0
+2023-03-10 10:23:34,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:36,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:38,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:54,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:01,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4422, 3.5589, 1.5665, 1.6249], device='cuda:0'), covar=tensor([0.1045, 0.0350, 0.0893, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0542, 0.0378, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 10:24:02,462 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:02,921 INFO [train.py:968] (0/2) Epoch 20, batch 19750, libri_loss[loss=0.2444, simple_loss=0.3294, pruned_loss=0.07966, over 29657.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.329, pruned_loss=0.08943, over 5672432.26 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3385, pruned_loss=0.08842, over 5726008.81 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3278, pruned_loss=0.08968, over 5676556.86 frames. ], batch size: 73, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:24:16,290 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 10:24:24,653 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:26,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:29,687 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886947.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:24:33,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886950.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:24:40,556 INFO [train.py:968] (0/2) Epoch 20, batch 19800, giga_loss[loss=0.2587, simple_loss=0.3334, pruned_loss=0.09196, over 28744.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3282, pruned_loss=0.08897, over 5693038.26 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3391, pruned_loss=0.08856, over 5732707.16 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3263, pruned_loss=0.08905, over 5689055.09 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:24:40,790 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:43,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:51,312 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:52,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.429e+02 1.134e+03 1.415e+03 1.936e+03 6.462e+03, threshold=2.830e+03, percent-clipped=13.0
+2023-03-10 10:24:58,012 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886979.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:25:09,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:25:22,880 INFO [train.py:968] (0/2) Epoch 20, batch 19850, giga_loss[loss=0.2178, simple_loss=0.2986, pruned_loss=0.06847, over 28884.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3259, pruned_loss=0.08766, over 5698972.33 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3396, pruned_loss=0.08863, over 5737686.59 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3236, pruned_loss=0.08765, over 5690002.66 frames. ], batch size: 174, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:25:41,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2946, 2.9665, 1.3630, 1.5080], device='cuda:0'), covar=tensor([0.1060, 0.0368, 0.0934, 0.1372], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0541, 0.0377, 0.0418], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 10:25:52,487 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=887047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:25:54,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=887050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:25:58,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4824, 1.6479, 1.2971, 1.2330], device='cuda:0'), covar=tensor([0.0989, 0.0549, 0.1039, 0.1068], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0439, 0.0512, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 10:26:05,449 INFO [train.py:968] (0/2) Epoch 20, batch 19900, libri_loss[loss=0.2097, simple_loss=0.2965, pruned_loss=0.06148, over 29380.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3225, pruned_loss=0.08652, over 5710458.53 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3394, pruned_loss=0.08849, over 5738378.15 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3208, pruned_loss=0.08664, over 5702610.38 frames. ], batch size: 67, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:26:15,908 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.621e+02 1.076e+03 1.324e+03 1.738e+03 4.904e+03, threshold=2.648e+03, percent-clipped=8.0
+2023-03-10 10:26:18,646 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=887079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:26:44,927 INFO [train.py:968] (0/2) Epoch 20, batch 19950, giga_loss[loss=0.2302, simple_loss=0.3128, pruned_loss=0.07383, over 28994.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3209, pruned_loss=0.08555, over 5720060.05 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3392, pruned_loss=0.08829, over 5742636.33 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3192, pruned_loss=0.08575, over 5709215.95 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:26:48,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:27:23,584 INFO [train.py:968] (0/2) Epoch 20, batch 20000, giga_loss[loss=0.2466, simple_loss=0.3204, pruned_loss=0.08636, over 28972.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3206, pruned_loss=0.08495, over 5721573.57 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3403, pruned_loss=0.08852, over 5750436.06 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3174, pruned_loss=0.08478, over 5704858.99 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:27:27,655 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3425, 1.3267, 4.0482, 3.1658], device='cuda:0'), covar=tensor([0.1724, 0.2804, 0.0428, 0.1006], device='cuda:0'), in_proj_covar=tensor([0.0737, 0.0631, 0.0928, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 10:27:35,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.227e+02 1.069e+03 1.344e+03 1.816e+03 6.351e+03, threshold=2.688e+03, percent-clipped=6.0
+2023-03-10 10:27:46,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4583, 2.6530, 2.4587, 1.9579], device='cuda:0'), covar=tensor([0.2726, 0.1989, 0.1942, 0.2716], device='cuda:0'), in_proj_covar=tensor([0.1916, 0.1820, 0.1758, 0.1907], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 10:27:52,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:28:03,472 INFO [train.py:968] (0/2) Epoch 20, batch 20050, giga_loss[loss=0.2614, simple_loss=0.3348, pruned_loss=0.09397, over 27653.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3194, pruned_loss=0.08454, over 5721074.80 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3405, pruned_loss=0.08854, over 5752513.07 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3164, pruned_loss=0.08434, over 5705665.80 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:28:40,616 INFO [train.py:968] (0/2) Epoch 20, batch 20100, giga_loss[loss=0.2158, simple_loss=0.2935, pruned_loss=0.06906, over 28955.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3192, pruned_loss=0.08377, over 5727789.21 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3412, pruned_loss=0.08866, over 5753189.83 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.3156, pruned_loss=0.08337, over 5714071.47 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:28:52,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.381e+02 9.959e+02 1.223e+03 1.532e+03 4.218e+03, threshold=2.447e+03, percent-clipped=3.0
+2023-03-10 10:29:26,214 INFO [train.py:968] (0/2) Epoch 20, batch 20150, giga_loss[loss=0.2295, simple_loss=0.3091, pruned_loss=0.07495, over 28902.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3235, pruned_loss=0.08692, over 5721528.81 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3414, pruned_loss=0.08867, over 5752243.81 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3203, pruned_loss=0.08658, over 5711304.60 frames. ], batch size: 66, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:29:51,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2156, 4.0423, 3.7960, 1.8828], device='cuda:0'), covar=tensor([0.0590, 0.0698, 0.0654, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.1175, 0.1089, 0.0925, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 10:29:56,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3386, 1.4885, 1.3419, 1.2030], device='cuda:0'), covar=tensor([0.2434, 0.2407, 0.1650, 0.2216], device='cuda:0'), in_proj_covar=tensor([0.1922, 0.1828, 0.1767, 0.1913], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 10:30:18,183 INFO [train.py:968] (0/2) Epoch 20, batch 20200, giga_loss[loss=0.247, simple_loss=0.3226, pruned_loss=0.08572, over 28632.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3288, pruned_loss=0.09078, over 5706178.31 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3412, pruned_loss=0.08852, over 5752863.59 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3264, pruned_loss=0.09062, over 5697575.75 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:30:31,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.884e+02 1.218e+03 1.557e+03 2.003e+03 4.420e+03, threshold=3.114e+03, percent-clipped=9.0
+2023-03-10 10:31:09,342 INFO [train.py:968] (0/2) Epoch 20, batch 20250, libri_loss[loss=0.2702, simple_loss=0.3582, pruned_loss=0.09109, over 29552.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3375, pruned_loss=0.09638, over 5704755.75 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3412, pruned_loss=0.08839, over 5754679.68 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3352, pruned_loss=0.09652, over 5694576.72 frames. ], batch size: 77, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:31:55,140 INFO [train.py:968] (0/2) Epoch 20, batch 20300, libri_loss[loss=0.2457, simple_loss=0.3217, pruned_loss=0.08487, over 29381.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3427, pruned_loss=0.09899, over 5700757.69 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3411, pruned_loss=0.08838, over 5756528.67 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3409, pruned_loss=0.09929, over 5690149.08 frames. ], batch size: 67, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:32:05,670 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-10 10:32:08,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.815e+02 1.289e+03 1.765e+03 2.269e+03 4.996e+03, threshold=3.529e+03, percent-clipped=13.0
+2023-03-10 10:32:24,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=887491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:32:43,877 INFO [train.py:968] (0/2) Epoch 20, batch 20350, giga_loss[loss=0.2783, simple_loss=0.3516, pruned_loss=0.1025, over 28548.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3486, pruned_loss=0.1018, over 5696096.37 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3415, pruned_loss=0.0885, over 5760379.96 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.347, pruned_loss=0.1022, over 5682975.39 frames. ], batch size: 60, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:33:07,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5327, 2.1664, 1.6205, 0.7608], device='cuda:0'), covar=tensor([0.6088, 0.3103, 0.3856, 0.5965], device='cuda:0'), in_proj_covar=tensor([0.1703, 0.1606, 0.1576, 0.1384], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 10:33:30,912 INFO [train.py:968] (0/2) Epoch 20, batch 20400, giga_loss[loss=0.3123, simple_loss=0.3919, pruned_loss=0.1164, over 29118.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3542, pruned_loss=0.1045, over 5678758.64 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3419, pruned_loss=0.08875, over 5744278.43 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3527, pruned_loss=0.1048, over 5681850.44 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:33:44,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=887573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:33:45,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.731e+02 1.213e+03 1.560e+03 2.003e+03 4.664e+03, threshold=3.119e+03, percent-clipped=3.0
+2023-03-10 10:34:16,382 INFO [train.py:968] (0/2) Epoch 20, batch 20450, giga_loss[loss=0.2653, simple_loss=0.3433, pruned_loss=0.09368, over 27894.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3577, pruned_loss=0.1062, over 5687985.32 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3416, pruned_loss=0.08853, over 5746335.60 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3571, pruned_loss=0.1069, over 5687742.37 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:34:25,997 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887623.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:34:39,083 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=887634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:34:42,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=887637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:35:03,567 INFO [train.py:968] (0/2) Epoch 20, batch 20500, giga_loss[loss=0.2127, simple_loss=0.3027, pruned_loss=0.06129, over 28221.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3514, pruned_loss=0.1016, over 5687471.76 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3417, pruned_loss=0.08864, over 5746207.26 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3509, pruned_loss=0.1022, over 5686954.71 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:35:09,001 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=887666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:35:16,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.832e+02 1.263e+03 1.569e+03 2.218e+03 3.523e+03, threshold=3.137e+03, percent-clipped=7.0
+2023-03-10 10:35:46,667 INFO [train.py:968] (0/2) Epoch 20, batch 20550, giga_loss[loss=0.2909, simple_loss=0.3639, pruned_loss=0.1089, over 28322.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3491, pruned_loss=0.09924, over 5698862.54 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3417, pruned_loss=0.08867, over 5750800.95 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.349, pruned_loss=0.1, over 5692838.65 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:35:49,666 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-10 10:35:50,130 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=887716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:35:52,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=887719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:36:10,031 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:36:18,269 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=887748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:36:27,937 INFO [train.py:968] (0/2) Epoch 20, batch 20600, giga_loss[loss=0.2515, simple_loss=0.3332, pruned_loss=0.08491, over 29041.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.348, pruned_loss=0.09819, over 5696128.60 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3418, pruned_loss=0.08891, over 5755571.53 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.348, pruned_loss=0.09886, over 5685521.65 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:36:43,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.608e+02 1.127e+03 1.412e+03 2.243e+03 7.115e+03, threshold=2.825e+03, percent-clipped=7.0
+2023-03-10 10:36:44,104 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887777.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:37:12,391 INFO [train.py:968] (0/2) Epoch 20, batch 20650, giga_loss[loss=0.2601, simple_loss=0.338, pruned_loss=0.09105, over 28525.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3496, pruned_loss=0.09898, over 5700541.63 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.342, pruned_loss=0.08907, over 5756960.98 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3495, pruned_loss=0.09954, over 5689803.28 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:37:44,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:37:52,609 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-10 10:37:54,771 INFO [train.py:968] (0/2) Epoch 20, batch 20700, giga_loss[loss=0.2976, simple_loss=0.3667, pruned_loss=0.1143, over 29032.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3518, pruned_loss=0.1004, over 5695862.86 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3426, pruned_loss=0.08946, over 5751997.73 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3516, pruned_loss=0.101, over 5687964.39 frames. ], batch size: 106, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:38:05,109 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:38:09,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.786e+02 1.340e+03 1.608e+03 2.136e+03 4.639e+03, threshold=3.217e+03, percent-clipped=11.0
+2023-03-10 10:38:26,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9350, 2.9629, 1.9099, 1.1316], device='cuda:0'), covar=tensor([0.7950, 0.3009, 0.3937, 0.6327], device='cuda:0'), in_proj_covar=tensor([0.1703, 0.1603, 0.1576, 0.1383], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 10:38:35,542 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 10:38:37,845 INFO [train.py:968] (0/2) Epoch 20, batch 20750, giga_loss[loss=0.2803, simple_loss=0.3505, pruned_loss=0.1051, over 28657.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3528, pruned_loss=0.1014, over 5691478.28 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3427, pruned_loss=0.08957, over 5745369.10 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3527, pruned_loss=0.102, over 5690422.10 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:38:52,129 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:39:18,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7214, 1.7555, 1.8345, 1.6314], device='cuda:0'), covar=tensor([0.2505, 0.2482, 0.1940, 0.2215], device='cuda:0'), in_proj_covar=tensor([0.1929, 0.1838, 0.1780, 0.1917], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 10:39:24,369 INFO [train.py:968] (0/2) Epoch 20, batch 20800, giga_loss[loss=0.253, simple_loss=0.3345, pruned_loss=0.08577, over 28831.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3537, pruned_loss=0.1025, over 5702433.58 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3432, pruned_loss=0.09024, over 5744610.61 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3535, pruned_loss=0.1027, over 5700953.20 frames. ], batch size: 112, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:39:39,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.776e+02 1.329e+03 1.574e+03 2.361e+03 5.779e+03, threshold=3.148e+03, percent-clipped=10.0
+2023-03-10 10:40:00,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=887998.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:40:01,228 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-888000.pt
+2023-03-10 10:40:10,073 INFO [train.py:968] (0/2) Epoch 20, batch 20850, giga_loss[loss=0.2928, simple_loss=0.3648, pruned_loss=0.1104, over 28920.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3539, pruned_loss=0.1029, over 5703190.15 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3431, pruned_loss=0.09018, over 5747207.03 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.354, pruned_loss=0.1034, over 5698896.50 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:40:24,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5502, 1.6406, 1.3813, 1.7379], device='cuda:0'), covar=tensor([0.2787, 0.2883, 0.3120, 0.2607], device='cuda:0'), in_proj_covar=tensor([0.1475, 0.1072, 0.1308, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 10:40:34,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4964, 1.5410, 1.5532, 1.4447], device='cuda:0'), covar=tensor([0.2441, 0.2315, 0.1768, 0.1995], device='cuda:0'), in_proj_covar=tensor([0.1927, 0.1835, 0.1778, 0.1915], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 10:40:49,351 INFO [train.py:968] (0/2) Epoch 20, batch 20900, giga_loss[loss=0.2872, simple_loss=0.361, pruned_loss=0.1067, over 28916.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3549, pruned_loss=0.1035, over 5710306.18 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3435, pruned_loss=0.09044, over 5749034.87 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3548, pruned_loss=0.1039, over 5704301.53 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:41:03,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.288e+02 1.228e+03 1.547e+03 2.026e+03 4.033e+03, threshold=3.094e+03, percent-clipped=8.0
+2023-03-10 10:41:16,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9320, 3.7481, 3.5273, 1.6768], device='cuda:0'), covar=tensor([0.0667, 0.0825, 0.0804, 0.2198], device='cuda:0'), in_proj_covar=tensor([0.1175, 0.1091, 0.0926, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 10:41:30,211 INFO [train.py:968] (0/2) Epoch 20, batch 20950, giga_loss[loss=0.2842, simple_loss=0.3578, pruned_loss=0.1053, over 28614.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3549, pruned_loss=0.1026, over 5694582.36 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3443, pruned_loss=0.09088, over 5734873.27 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3544, pruned_loss=0.1028, over 5701289.63 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:41:33,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:41:55,973 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888141.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:41:57,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888144.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:42:05,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888152.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:42:12,250 INFO [train.py:968] (0/2) Epoch 20, batch 21000, giga_loss[loss=0.2908, simple_loss=0.3651, pruned_loss=0.1082, over 28623.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3553, pruned_loss=0.1016, over 5700574.60 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3445, pruned_loss=0.0911, over 5732492.87 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3549, pruned_loss=0.1018, over 5707164.05 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:42:12,255 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 10:42:15,926 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3201, 1.6705, 1.6083, 1.1465], device='cuda:0'), covar=tensor([0.2102, 0.3082, 0.1870, 0.2076], device='cuda:0'), in_proj_covar=tensor([0.0891, 0.0696, 0.0936, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 10:42:20,611 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2084, simple_loss=0.3162, pruned_loss=0.05028, over 944034.00 frames. 
+2023-03-10 10:42:20,612 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 10:42:30,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888173.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:42:33,595 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.187e+02 1.166e+03 1.569e+03 1.910e+03 5.211e+03, threshold=3.139e+03, percent-clipped=8.0
+2023-03-10 10:43:01,231 INFO [train.py:968] (0/2) Epoch 20, batch 21050, giga_loss[loss=0.2422, simple_loss=0.3266, pruned_loss=0.07888, over 29000.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3556, pruned_loss=0.1018, over 5710259.49 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3446, pruned_loss=0.09118, over 5736022.44 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3555, pruned_loss=0.1021, over 5711828.22 frames. ], batch size: 164, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:43:10,568 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:43:31,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:43:39,546 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:43:41,786 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:43:42,116 INFO [train.py:968] (0/2) Epoch 20, batch 21100, giga_loss[loss=0.276, simple_loss=0.3469, pruned_loss=0.1025, over 28895.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3525, pruned_loss=0.1006, over 5709923.29 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3446, pruned_loss=0.09137, over 5739315.46 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3526, pruned_loss=0.1009, over 5707469.95 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:43:53,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.334e+02 1.210e+03 1.486e+03 1.959e+03 8.897e+03, threshold=2.971e+03, percent-clipped=7.0
+2023-03-10 10:44:06,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:44:10,363 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888295.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:44:12,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888298.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:44:13,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:44:14,821 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 10:44:20,941 INFO [train.py:968] (0/2) Epoch 20, batch 21150, giga_loss[loss=0.2537, simple_loss=0.334, pruned_loss=0.08667, over 28484.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3505, pruned_loss=0.09963, over 5715930.07 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3449, pruned_loss=0.09158, over 5744775.60 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3506, pruned_loss=0.09993, over 5708254.17 frames. ], batch size: 78, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:44:35,791 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888327.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:45:00,407 INFO [train.py:968] (0/2) Epoch 20, batch 21200, giga_loss[loss=0.2672, simple_loss=0.3411, pruned_loss=0.0967, over 28543.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3494, pruned_loss=0.09919, over 5722477.82 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3453, pruned_loss=0.09192, over 5748866.65 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3492, pruned_loss=0.09937, over 5711792.37 frames. ], batch size: 60, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:45:04,461 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:45:08,634 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:45:14,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.775e+02 1.123e+03 1.470e+03 1.884e+03 4.342e+03, threshold=2.940e+03, percent-clipped=5.0
+2023-03-10 10:45:26,391 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:45:28,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:45:31,216 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:45:43,730 INFO [train.py:968] (0/2) Epoch 20, batch 21250, giga_loss[loss=0.2498, simple_loss=0.3353, pruned_loss=0.08218, over 29048.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.349, pruned_loss=0.09941, over 5723893.92 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3447, pruned_loss=0.09174, over 5753557.27 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3495, pruned_loss=0.09993, over 5710223.34 frames. ], batch size: 164, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:45:51,708 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:46:05,792 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-10 10:46:10,576 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:46:12,688 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:46:22,519 INFO [train.py:968] (0/2) Epoch 20, batch 21300, giga_loss[loss=0.2411, simple_loss=0.3294, pruned_loss=0.07637, over 29050.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3495, pruned_loss=0.09965, over 5708523.31 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3452, pruned_loss=0.09224, over 5742980.00 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3495, pruned_loss=0.09982, over 5705299.52 frames. ], batch size: 128, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:46:34,024 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:46:37,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.902e+02 1.173e+03 1.358e+03 1.819e+03 5.095e+03, threshold=2.717e+03, percent-clipped=4.0
+2023-03-10 10:47:06,829 INFO [train.py:968] (0/2) Epoch 20, batch 21350, giga_loss[loss=0.278, simple_loss=0.3607, pruned_loss=0.09761, over 28846.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3488, pruned_loss=0.09837, over 5713963.94 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3452, pruned_loss=0.09221, over 5743796.10 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3489, pruned_loss=0.09855, over 5710663.05 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:47:08,045 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=888512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:47:39,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1715, 1.7294, 1.3539, 0.4433], device='cuda:0'), covar=tensor([0.4387, 0.2520, 0.3976, 0.5436], device='cuda:0'), in_proj_covar=tensor([0.1696, 0.1585, 0.1566, 0.1378], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 10:47:46,618 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 10:47:47,527 INFO [train.py:968] (0/2) Epoch 20, batch 21400, giga_loss[loss=0.2323, simple_loss=0.3159, pruned_loss=0.07438, over 29038.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3484, pruned_loss=0.09827, over 5697451.66 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3463, pruned_loss=0.0934, over 5731248.45 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3476, pruned_loss=0.09754, over 5705077.89 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 1.0
+2023-03-10 10:47:58,109 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 10:48:01,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.491e+02 1.070e+03 1.415e+03 2.018e+03 7.833e+03, threshold=2.830e+03, percent-clipped=13.0
+2023-03-10 10:48:24,507 INFO [train.py:968] (0/2) Epoch 20, batch 21450, libri_loss[loss=0.3181, simple_loss=0.3877, pruned_loss=0.1243, over 29744.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3483, pruned_loss=0.0987, over 5691541.89 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3471, pruned_loss=0.09407, over 5726334.67 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3469, pruned_loss=0.09763, over 5700898.03 frames. ], batch size: 87, lr: 1.60e-03, grad_scale: 1.0
+2023-03-10 10:49:04,568 INFO [train.py:968] (0/2) Epoch 20, batch 21500, giga_loss[loss=0.3282, simple_loss=0.3764, pruned_loss=0.14, over 26845.00 frames. ], tot_loss[loss=0.272, simple_loss=0.347, pruned_loss=0.0985, over 5684853.19 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3482, pruned_loss=0.095, over 5719512.79 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.345, pruned_loss=0.09694, over 5696742.99 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 1.0
+2023-03-10 10:49:20,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.360e+02 1.164e+03 1.437e+03 1.950e+03 7.015e+03, threshold=2.874e+03, percent-clipped=12.0
+2023-03-10 10:49:45,905 INFO [train.py:968] (0/2) Epoch 20, batch 21550, giga_loss[loss=0.2764, simple_loss=0.3523, pruned_loss=0.1003, over 28611.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3439, pruned_loss=0.09711, over 5690576.84 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3486, pruned_loss=0.09541, over 5723248.67 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3419, pruned_loss=0.09553, over 5696083.16 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 1.0
+2023-03-10 10:50:25,975 INFO [train.py:968] (0/2) Epoch 20, batch 21600, giga_loss[loss=0.2621, simple_loss=0.3386, pruned_loss=0.09278, over 28819.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3421, pruned_loss=0.0964, over 5690818.17 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3486, pruned_loss=0.09563, over 5728903.77 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3403, pruned_loss=0.09497, over 5689211.56 frames. ], batch size: 92, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:50:40,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.104e+02 1.222e+03 1.555e+03 1.939e+03 6.440e+03, threshold=3.110e+03, percent-clipped=7.0
+2023-03-10 10:50:48,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=888790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:51:05,442 INFO [train.py:968] (0/2) Epoch 20, batch 21650, giga_loss[loss=0.2474, simple_loss=0.324, pruned_loss=0.0854, over 28926.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3417, pruned_loss=0.09671, over 5688018.71 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3489, pruned_loss=0.0961, over 5722898.69 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3397, pruned_loss=0.09515, over 5691800.24 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:51:18,703 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=888829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:51:19,669 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-10 10:51:46,458 INFO [train.py:968] (0/2) Epoch 20, batch 21700, giga_loss[loss=0.2531, simple_loss=0.324, pruned_loss=0.0911, over 28523.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3394, pruned_loss=0.09571, over 5688981.11 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3489, pruned_loss=0.09618, over 5723567.05 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3378, pruned_loss=0.09442, over 5691127.27 frames. ], batch size: 65, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:52:02,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.968e+02 1.104e+03 1.395e+03 2.022e+03 1.151e+04, threshold=2.791e+03, percent-clipped=12.0
+2023-03-10 10:52:09,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:52:27,003 INFO [train.py:968] (0/2) Epoch 20, batch 21750, giga_loss[loss=0.2413, simple_loss=0.312, pruned_loss=0.08528, over 28680.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3366, pruned_loss=0.09439, over 5698208.19 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3492, pruned_loss=0.09654, over 5726431.52 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3348, pruned_loss=0.09301, over 5696727.12 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:52:46,281 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-10 10:53:08,510 INFO [train.py:968] (0/2) Epoch 20, batch 21800, giga_loss[loss=0.2265, simple_loss=0.3079, pruned_loss=0.07256, over 28771.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3336, pruned_loss=0.09273, over 5701620.86 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3493, pruned_loss=0.09674, over 5721850.55 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3318, pruned_loss=0.09141, over 5703468.75 frames. ], batch size: 112, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:53:24,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.555e+02 1.088e+03 1.292e+03 1.793e+03 1.691e+04, threshold=2.584e+03, percent-clipped=5.0
+2023-03-10 10:53:48,742 INFO [train.py:968] (0/2) Epoch 20, batch 21850, giga_loss[loss=0.312, simple_loss=0.3667, pruned_loss=0.1287, over 24083.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3328, pruned_loss=0.09255, over 5708381.35 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3495, pruned_loss=0.09696, over 5726333.81 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3308, pruned_loss=0.0912, over 5705617.88 frames. ], batch size: 710, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:54:04,077 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:54:07,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:54:26,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4430, 4.2397, 1.5900, 1.6510], device='cuda:0'), covar=tensor([0.0977, 0.0300, 0.0986, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0544, 0.0377, 0.0418], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 10:54:30,257 INFO [train.py:968] (0/2) Epoch 20, batch 21900, giga_loss[loss=0.2557, simple_loss=0.3348, pruned_loss=0.08831, over 29044.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3339, pruned_loss=0.09276, over 5702244.37 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3497, pruned_loss=0.09731, over 5718185.74 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3319, pruned_loss=0.09128, over 5706088.11 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:54:31,294 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:54:47,701 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.455e+02 1.110e+03 1.390e+03 1.857e+03 6.027e+03, threshold=2.779e+03, percent-clipped=5.0
+2023-03-10 10:55:14,028 INFO [train.py:968] (0/2) Epoch 20, batch 21950, giga_loss[loss=0.2679, simple_loss=0.3511, pruned_loss=0.0923, over 28644.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3386, pruned_loss=0.09524, over 5697375.40 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3506, pruned_loss=0.0981, over 5720266.32 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3358, pruned_loss=0.09328, over 5698156.28 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:55:33,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=889131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:55:44,466 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 10:55:44,988 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=889143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:55:59,428 INFO [train.py:968] (0/2) Epoch 20, batch 22000, giga_loss[loss=0.2631, simple_loss=0.346, pruned_loss=0.09016, over 28942.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3415, pruned_loss=0.09588, over 5684925.89 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3508, pruned_loss=0.09844, over 5713744.80 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3388, pruned_loss=0.09392, over 5691258.98 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:56:01,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4105, 1.5823, 1.6495, 1.3792], device='cuda:0'), covar=tensor([0.3719, 0.2570, 0.2241, 0.2754], device='cuda:0'), in_proj_covar=tensor([0.1945, 0.1856, 0.1797, 0.1928], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 10:56:02,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=889165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:56:13,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.030e+02 1.094e+03 1.349e+03 1.945e+03 5.500e+03, threshold=2.697e+03, percent-clipped=9.0
+2023-03-10 10:56:36,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=889204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:56:43,467 INFO [train.py:968] (0/2) Epoch 20, batch 22050, giga_loss[loss=0.2329, simple_loss=0.3148, pruned_loss=0.07548, over 29021.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3423, pruned_loss=0.09569, over 5697467.29 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3509, pruned_loss=0.09869, over 5717959.41 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.34, pruned_loss=0.09389, over 5698539.11 frames. ], batch size: 128, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:57:27,683 INFO [train.py:968] (0/2) Epoch 20, batch 22100, giga_loss[loss=0.2971, simple_loss=0.374, pruned_loss=0.1101, over 27982.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3421, pruned_loss=0.09476, over 5703435.44 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3513, pruned_loss=0.09909, over 5720524.22 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3397, pruned_loss=0.0929, over 5701646.21 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:57:46,604 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.704e+02 1.126e+03 1.387e+03 2.048e+03 4.514e+03, threshold=2.775e+03, percent-clipped=10.0
+2023-03-10 10:58:11,250 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:58:13,724 INFO [train.py:968] (0/2) Epoch 20, batch 22150, giga_loss[loss=0.2626, simple_loss=0.3454, pruned_loss=0.08987, over 28694.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3431, pruned_loss=0.09597, over 5693678.71 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3519, pruned_loss=0.09977, over 5721375.93 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3403, pruned_loss=0.09375, over 5690877.61 frames. ], batch size: 262, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:58:14,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:58:39,440 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:58:45,317 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:58:47,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:58:54,529 INFO [train.py:968] (0/2) Epoch 20, batch 22200, giga_loss[loss=0.2649, simple_loss=0.3362, pruned_loss=0.09681, over 28743.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3431, pruned_loss=0.09642, over 5701390.89 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3519, pruned_loss=0.09985, over 5724157.03 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3408, pruned_loss=0.09452, over 5696096.71 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:59:05,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5040, 1.6846, 1.7091, 1.4700], device='cuda:0'), covar=tensor([0.1837, 0.2377, 0.2270, 0.2397], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0742, 0.0708, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 10:59:09,775 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:59:10,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.638e+02 1.232e+03 1.510e+03 2.180e+03 8.757e+03, threshold=3.020e+03, percent-clipped=11.0
+2023-03-10 10:59:35,546 INFO [train.py:968] (0/2) Epoch 20, batch 22250, giga_loss[loss=0.2785, simple_loss=0.3529, pruned_loss=0.1021, over 28834.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3439, pruned_loss=0.09718, over 5702464.39 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3525, pruned_loss=0.1003, over 5727611.63 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3414, pruned_loss=0.09515, over 5694609.23 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:59:48,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9133, 1.3112, 1.1113, 0.1723], device='cuda:0'), covar=tensor([0.4039, 0.2890, 0.4465, 0.6344], device='cuda:0'), in_proj_covar=tensor([0.1714, 0.1605, 0.1583, 0.1395], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 10:59:58,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=889440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:00:11,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5357, 1.8095, 1.4933, 1.5035], device='cuda:0'), covar=tensor([0.2335, 0.2380, 0.2612, 0.2378], device='cuda:0'), in_proj_covar=tensor([0.1477, 0.1071, 0.1308, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 11:00:15,906 INFO [train.py:968] (0/2) Epoch 20, batch 22300, giga_loss[loss=0.2797, simple_loss=0.3591, pruned_loss=0.1001, over 29006.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3473, pruned_loss=0.09906, over 5697965.72 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3522, pruned_loss=0.1005, over 5716529.50 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3452, pruned_loss=0.09718, over 5700509.73 frames. ], batch size: 164, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:00:30,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.290e+02 1.313e+03 1.600e+03 2.205e+03 6.533e+03, threshold=3.200e+03, percent-clipped=10.0
+2023-03-10 11:00:51,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=889506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:00:55,219 INFO [train.py:968] (0/2) Epoch 20, batch 22350, giga_loss[loss=0.3298, simple_loss=0.3958, pruned_loss=0.1319, over 28868.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3506, pruned_loss=0.1006, over 5700317.92 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3528, pruned_loss=0.101, over 5718239.49 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3483, pruned_loss=0.09866, over 5700371.83 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:01:00,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=889518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:01:35,499 INFO [train.py:968] (0/2) Epoch 20, batch 22400, libri_loss[loss=0.2662, simple_loss=0.3461, pruned_loss=0.09316, over 29518.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1011, over 5710446.12 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3524, pruned_loss=0.1009, over 5721335.13 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3499, pruned_loss=0.09958, over 5707358.79 frames. ], batch size: 80, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 11:01:49,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.543e+02 1.266e+03 1.599e+03 2.060e+03 4.653e+03, threshold=3.197e+03, percent-clipped=5.0
+2023-03-10 11:02:19,514 INFO [train.py:968] (0/2) Epoch 20, batch 22450, giga_loss[loss=0.2607, simple_loss=0.3434, pruned_loss=0.08898, over 29045.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3525, pruned_loss=0.1016, over 5715161.94 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3527, pruned_loss=0.1014, over 5719333.09 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3509, pruned_loss=0.1, over 5714294.20 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:02:48,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1548, 1.6673, 1.3043, 0.3806], device='cuda:0'), covar=tensor([0.4443, 0.2558, 0.3645, 0.6053], device='cuda:0'), in_proj_covar=tensor([0.1717, 0.1614, 0.1588, 0.1398], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 11:02:53,958 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:02:55,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4056, 1.8107, 1.4375, 1.3159], device='cuda:0'), covar=tensor([0.2468, 0.2499, 0.2949, 0.2324], device='cuda:0'), in_proj_covar=tensor([0.1473, 0.1068, 0.1303, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 11:02:56,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:03:05,102 INFO [train.py:968] (0/2) Epoch 20, batch 22500, giga_loss[loss=0.2769, simple_loss=0.3591, pruned_loss=0.09737, over 28983.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3529, pruned_loss=0.1021, over 5708830.43 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.353, pruned_loss=0.1015, over 5720198.85 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3514, pruned_loss=0.1006, over 5707397.39 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:03:05,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:03:07,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:03:22,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.387e+02 1.324e+03 1.638e+03 2.270e+03 5.859e+03, threshold=3.276e+03, percent-clipped=6.0
+2023-03-10 11:03:23,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:03:32,833 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:03:48,813 INFO [train.py:968] (0/2) Epoch 20, batch 22550, giga_loss[loss=0.2779, simple_loss=0.347, pruned_loss=0.1044, over 28635.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3515, pruned_loss=0.1012, over 5714304.76 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3531, pruned_loss=0.1016, over 5719282.85 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3502, pruned_loss=0.1, over 5713892.45 frames. ], batch size: 60, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:04:32,964 INFO [train.py:968] (0/2) Epoch 20, batch 22600, libri_loss[loss=0.276, simple_loss=0.34, pruned_loss=0.1059, over 29546.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3505, pruned_loss=0.1011, over 5719077.77 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3533, pruned_loss=0.1018, over 5723072.34 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3494, pruned_loss=0.09993, over 5715046.84 frames. ], batch size: 76, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:04:50,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.452e+02 1.126e+03 1.504e+03 2.023e+03 5.535e+03, threshold=3.008e+03, percent-clipped=8.0
+2023-03-10 11:05:15,651 INFO [train.py:968] (0/2) Epoch 20, batch 22650, giga_loss[loss=0.2493, simple_loss=0.3293, pruned_loss=0.08469, over 29001.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3478, pruned_loss=0.09992, over 5717921.91 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3539, pruned_loss=0.1023, over 5725154.33 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3463, pruned_loss=0.09858, over 5712903.78 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:05:18,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=889815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:05:54,063 INFO [train.py:968] (0/2) Epoch 20, batch 22700, giga_loss[loss=0.3593, simple_loss=0.3966, pruned_loss=0.161, over 23957.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3456, pruned_loss=0.09835, over 5722613.56 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3537, pruned_loss=0.1024, over 5731698.68 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3442, pruned_loss=0.09704, over 5712375.18 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:05:56,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3539, 2.3636, 1.8555, 2.1978], device='cuda:0'), covar=tensor([0.0779, 0.0577, 0.0888, 0.0933], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0442, 0.0515, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 11:06:10,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.566e+02 1.197e+03 1.638e+03 2.372e+03 1.209e+04, threshold=3.277e+03, percent-clipped=15.0
+2023-03-10 11:06:35,427 INFO [train.py:968] (0/2) Epoch 20, batch 22750, giga_loss[loss=0.2992, simple_loss=0.3766, pruned_loss=0.111, over 29110.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3478, pruned_loss=0.09869, over 5720339.70 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3543, pruned_loss=0.103, over 5736479.81 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3459, pruned_loss=0.09694, over 5707613.25 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:06:38,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5657, 1.7330, 1.4142, 1.8406], device='cuda:0'), covar=tensor([0.2715, 0.2797, 0.3241, 0.2364], device='cuda:0'), in_proj_covar=tensor([0.1475, 0.1071, 0.1306, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 11:07:11,627 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8732, 2.1688, 2.0037, 1.7034], device='cuda:0'), covar=tensor([0.2193, 0.2492, 0.2292, 0.2687], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0745, 0.0711, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 11:07:15,937 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:07:18,061 INFO [train.py:968] (0/2) Epoch 20, batch 22800, giga_loss[loss=0.2657, simple_loss=0.3468, pruned_loss=0.09226, over 28686.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3488, pruned_loss=0.09814, over 5723005.27 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3543, pruned_loss=0.103, over 5737971.61 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3473, pruned_loss=0.09676, over 5711553.07 frames. ], batch size: 262, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 11:07:19,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:07:30,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4394, 4.0057, 1.6507, 1.4508], device='cuda:0'), covar=tensor([0.0943, 0.0324, 0.0913, 0.1350], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0546, 0.0379, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 11:07:36,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.258e+02 1.124e+03 1.421e+03 1.719e+03 3.526e+03, threshold=2.841e+03, percent-clipped=2.0
+2023-03-10 11:07:42,692 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:07:49,871 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-890000.pt
+2023-03-10 11:07:59,535 INFO [train.py:968] (0/2) Epoch 20, batch 22850, giga_loss[loss=0.307, simple_loss=0.3763, pruned_loss=0.1188, over 28156.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3474, pruned_loss=0.09803, over 5728712.93 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3547, pruned_loss=0.1034, over 5740009.33 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3457, pruned_loss=0.09646, over 5717646.17 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:08:06,398 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:08:20,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5923, 1.8652, 1.4668, 1.6671], device='cuda:0'), covar=tensor([0.2635, 0.2659, 0.3081, 0.2391], device='cuda:0'), in_proj_covar=tensor([0.1475, 0.1071, 0.1306, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 11:08:40,721 INFO [train.py:968] (0/2) Epoch 20, batch 22900, giga_loss[loss=0.2864, simple_loss=0.3541, pruned_loss=0.1094, over 27699.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3456, pruned_loss=0.09799, over 5730639.83 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3548, pruned_loss=0.1037, over 5745191.75 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3439, pruned_loss=0.09634, over 5716750.25 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:08:57,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.678e+02 1.163e+03 1.404e+03 1.935e+03 4.750e+03, threshold=2.808e+03, percent-clipped=7.0
+2023-03-10 11:09:22,187 INFO [train.py:968] (0/2) Epoch 20, batch 22950, giga_loss[loss=0.2563, simple_loss=0.3291, pruned_loss=0.09171, over 28622.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3434, pruned_loss=0.09809, over 5730761.65 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3545, pruned_loss=0.1037, over 5749429.08 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3421, pruned_loss=0.09664, over 5715599.93 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:09:37,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3456, 1.7546, 1.5133, 1.5317], device='cuda:0'), covar=tensor([0.0753, 0.0316, 0.0322, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0185, 0.0116, 0.0117, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 11:10:05,398 INFO [train.py:968] (0/2) Epoch 20, batch 23000, giga_loss[loss=0.2336, simple_loss=0.3133, pruned_loss=0.07692, over 28999.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3425, pruned_loss=0.09866, over 5730292.17 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3546, pruned_loss=0.1038, over 5748358.88 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3412, pruned_loss=0.0973, over 5718888.91 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:10:22,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.273e+02 1.180e+03 1.445e+03 1.753e+03 4.804e+03, threshold=2.890e+03, percent-clipped=5.0
+2023-03-10 11:10:46,243 INFO [train.py:968] (0/2) Epoch 20, batch 23050, giga_loss[loss=0.2356, simple_loss=0.3211, pruned_loss=0.07505, over 28824.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3413, pruned_loss=0.09897, over 5726231.27 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3548, pruned_loss=0.1043, over 5751320.53 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3396, pruned_loss=0.0973, over 5713911.51 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:11:22,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7247, 1.7911, 1.9411, 1.4922], device='cuda:0'), covar=tensor([0.1900, 0.2426, 0.1530, 0.1711], device='cuda:0'), in_proj_covar=tensor([0.0890, 0.0696, 0.0934, 0.0834], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 11:11:26,588 INFO [train.py:968] (0/2) Epoch 20, batch 23100, giga_loss[loss=0.2172, simple_loss=0.2946, pruned_loss=0.06993, over 28790.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3379, pruned_loss=0.09701, over 5714937.46 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3556, pruned_loss=0.105, over 5735926.25 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3357, pruned_loss=0.09494, over 5716490.21 frames. ], batch size: 119, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:11:44,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.908e+02 1.169e+03 1.535e+03 2.235e+03 1.681e+04, threshold=3.070e+03, percent-clipped=13.0
+2023-03-10 11:12:08,460 INFO [train.py:968] (0/2) Epoch 20, batch 23150, giga_loss[loss=0.2222, simple_loss=0.2931, pruned_loss=0.0757, over 28942.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3338, pruned_loss=0.09482, over 5715048.91 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3556, pruned_loss=0.1051, over 5738531.18 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3317, pruned_loss=0.09296, over 5713761.05 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:12:48,257 INFO [train.py:968] (0/2) Epoch 20, batch 23200, libri_loss[loss=0.3108, simple_loss=0.3664, pruned_loss=0.1276, over 29547.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3302, pruned_loss=0.09293, over 5716645.27 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3563, pruned_loss=0.1057, over 5732885.10 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3273, pruned_loss=0.09063, over 5719164.34 frames. ], batch size: 76, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:13:04,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.833e+02 1.281e+03 1.621e+03 2.428e+03 9.523e+03, threshold=3.241e+03, percent-clipped=14.0
+2023-03-10 11:13:16,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=890395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:13:31,332 INFO [train.py:968] (0/2) Epoch 20, batch 23250, giga_loss[loss=0.3024, simple_loss=0.366, pruned_loss=0.1194, over 28702.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3333, pruned_loss=0.09424, over 5700589.69 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3569, pruned_loss=0.106, over 5725426.32 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.33, pruned_loss=0.09179, over 5709327.34 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:13:51,620 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-10 11:14:15,145 INFO [train.py:968] (0/2) Epoch 20, batch 23300, giga_loss[loss=0.3047, simple_loss=0.3809, pruned_loss=0.1143, over 28750.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3367, pruned_loss=0.09584, over 5700667.62 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3569, pruned_loss=0.1062, over 5725514.62 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3336, pruned_loss=0.09358, over 5707287.20 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:14:32,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.015e+02 1.202e+03 1.576e+03 2.160e+03 1.219e+04, threshold=3.152e+03, percent-clipped=11.0
+2023-03-10 11:14:42,874 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:14:42,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3765, 1.6939, 1.6618, 1.5850], device='cuda:0'), covar=tensor([0.1842, 0.1828, 0.2199, 0.1866], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0745, 0.0712, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 11:14:56,930 INFO [train.py:968] (0/2) Epoch 20, batch 23350, giga_loss[loss=0.2677, simple_loss=0.3511, pruned_loss=0.09216, over 28725.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3393, pruned_loss=0.09649, over 5708296.22 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.357, pruned_loss=0.1065, over 5729220.14 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3361, pruned_loss=0.09412, over 5709772.29 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:15:08,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4582, 2.3223, 2.0896, 2.1095], device='cuda:0'), covar=tensor([0.1706, 0.2460, 0.2378, 0.2304], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0745, 0.0712, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 11:15:20,357 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=890538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:15:22,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=890541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:15:38,989 INFO [train.py:968] (0/2) Epoch 20, batch 23400, libri_loss[loss=0.3219, simple_loss=0.3798, pruned_loss=0.132, over 25852.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3425, pruned_loss=0.09787, over 5711800.06 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3572, pruned_loss=0.1069, over 5731066.56 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3392, pruned_loss=0.09521, over 5710738.32 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:15:46,140 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=890570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:15:57,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.572e+02 1.338e+03 1.721e+03 2.364e+03 6.984e+03, threshold=3.442e+03, percent-clipped=12.0
+2023-03-10 11:16:21,993 INFO [train.py:968] (0/2) Epoch 20, batch 23450, giga_loss[loss=0.2474, simple_loss=0.3324, pruned_loss=0.08117, over 28957.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3439, pruned_loss=0.09802, over 5724297.58 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3574, pruned_loss=0.1072, over 5737254.20 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3405, pruned_loss=0.09524, over 5717170.91 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:17:04,304 INFO [train.py:968] (0/2) Epoch 20, batch 23500, giga_loss[loss=0.2344, simple_loss=0.3122, pruned_loss=0.07833, over 28568.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3455, pruned_loss=0.09961, over 5720465.85 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3574, pruned_loss=0.1075, over 5743778.92 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3423, pruned_loss=0.09673, over 5708470.58 frames. ], batch size: 65, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:17:09,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:17:25,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.497e+02 1.237e+03 1.623e+03 2.232e+03 4.159e+03, threshold=3.247e+03, percent-clipped=4.0
+2023-03-10 11:17:52,912 INFO [train.py:968] (0/2) Epoch 20, batch 23550, giga_loss[loss=0.2699, simple_loss=0.3399, pruned_loss=0.09992, over 28911.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3521, pruned_loss=0.1048, over 5715737.42 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3573, pruned_loss=0.1075, over 5746138.57 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3495, pruned_loss=0.1024, over 5703461.81 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:18:19,006 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:18:21,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5897, 1.7709, 1.4435, 1.8762], device='cuda:0'), covar=tensor([0.2317, 0.2407, 0.2622, 0.2177], device='cuda:0'), in_proj_covar=tensor([0.1475, 0.1070, 0.1306, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 11:18:42,511 INFO [train.py:968] (0/2) Epoch 20, batch 23600, giga_loss[loss=0.3203, simple_loss=0.3919, pruned_loss=0.1244, over 28995.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3595, pruned_loss=0.1107, over 5705254.62 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3575, pruned_loss=0.108, over 5750839.91 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3572, pruned_loss=0.1083, over 5689620.93 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:18:52,669 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:19:06,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.782e+03 2.104e+03 2.962e+03 1.045e+04, threshold=4.208e+03, percent-clipped=20.0
+2023-03-10 11:19:21,316 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:19:29,463 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:19:35,700 INFO [train.py:968] (0/2) Epoch 20, batch 23650, giga_loss[loss=0.3276, simple_loss=0.3939, pruned_loss=0.1307, over 28997.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3661, pruned_loss=0.1155, over 5683116.06 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3576, pruned_loss=0.1081, over 5742414.05 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3643, pruned_loss=0.1136, over 5678321.75 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:20:13,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8072, 2.5682, 1.6573, 0.9124], device='cuda:0'), covar=tensor([0.6540, 0.3044, 0.3725, 0.5989], device='cuda:0'), in_proj_covar=tensor([0.1713, 0.1613, 0.1583, 0.1397], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 11:20:24,628 INFO [train.py:968] (0/2) Epoch 20, batch 23700, giga_loss[loss=0.292, simple_loss=0.3632, pruned_loss=0.1104, over 29006.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3724, pruned_loss=0.121, over 5671814.36 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3575, pruned_loss=0.1081, over 5733319.44 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3712, pruned_loss=0.1197, over 5674692.17 frames. ], batch size: 128, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:20:35,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=890869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:20:46,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.813e+02 1.833e+03 2.315e+03 3.001e+03 6.368e+03, threshold=4.630e+03, percent-clipped=3.0
+2023-03-10 11:21:13,765 INFO [train.py:968] (0/2) Epoch 20, batch 23750, giga_loss[loss=0.3629, simple_loss=0.4103, pruned_loss=0.1577, over 28946.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3774, pruned_loss=0.125, over 5678431.46 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3575, pruned_loss=0.1082, over 5735137.37 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3769, pruned_loss=0.1242, over 5677834.53 frames. ], batch size: 227, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:22:03,686 INFO [train.py:968] (0/2) Epoch 20, batch 23800, giga_loss[loss=0.2936, simple_loss=0.3579, pruned_loss=0.1147, over 28463.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3804, pruned_loss=0.1281, over 5668196.64 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.358, pruned_loss=0.1086, over 5729158.74 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.38, pruned_loss=0.1276, over 5671775.29 frames. ], batch size: 71, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:22:27,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.236e+03 1.763e+03 2.225e+03 3.302e+03 8.259e+03, threshold=4.451e+03, percent-clipped=12.0
+2023-03-10 11:22:57,068 INFO [train.py:968] (0/2) Epoch 20, batch 23850, giga_loss[loss=0.3483, simple_loss=0.3965, pruned_loss=0.1501, over 28653.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3838, pruned_loss=0.1324, over 5658432.11 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3577, pruned_loss=0.1085, over 5731076.51 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3839, pruned_loss=0.1323, over 5659025.70 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:23:00,055 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:23:00,180 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 11:23:02,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:23:06,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1198, 1.1493, 3.4035, 2.9714], device='cuda:0'), covar=tensor([0.1665, 0.2632, 0.0486, 0.0939], device='cuda:0'), in_proj_covar=tensor([0.0743, 0.0636, 0.0941, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 11:23:30,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:23:32,244 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=891044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:23:49,299 INFO [train.py:968] (0/2) Epoch 20, batch 23900, giga_loss[loss=0.3728, simple_loss=0.4185, pruned_loss=0.1636, over 28633.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3869, pruned_loss=0.1361, over 5643555.42 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3573, pruned_loss=0.1083, over 5724881.10 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3883, pruned_loss=0.1369, over 5646494.74 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:24:12,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.771e+03 2.402e+03 3.529e+03 8.736e+03, threshold=4.804e+03, percent-clipped=15.0
+2023-03-10 11:24:48,169 INFO [train.py:968] (0/2) Epoch 20, batch 23950, giga_loss[loss=0.3345, simple_loss=0.3953, pruned_loss=0.1368, over 28782.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3908, pruned_loss=0.1392, over 5641904.79 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3572, pruned_loss=0.1083, over 5727687.11 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3923, pruned_loss=0.1402, over 5640883.96 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:24:49,344 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:24:59,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4859, 2.1369, 1.6244, 0.8263], device='cuda:0'), covar=tensor([0.5110, 0.2811, 0.3718, 0.5366], device='cuda:0'), in_proj_covar=tensor([0.1727, 0.1625, 0.1589, 0.1405], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 11:25:00,842 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 11:25:19,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3178, 1.2933, 3.7955, 3.2753], device='cuda:0'), covar=tensor([0.1594, 0.2664, 0.0470, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0747, 0.0640, 0.0947, 0.0891], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 11:25:27,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:25:41,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2855, 2.2087, 1.3296, 1.4305], device='cuda:0'), covar=tensor([0.0812, 0.0379, 0.0720, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0551, 0.0381, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 11:25:42,845 INFO [train.py:968] (0/2) Epoch 20, batch 24000, giga_loss[loss=0.3765, simple_loss=0.4174, pruned_loss=0.1678, over 28570.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3903, pruned_loss=0.139, over 5651205.41 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3568, pruned_loss=0.1082, over 5733769.46 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3931, pruned_loss=0.141, over 5642300.22 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:25:42,850 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 11:25:51,388 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2047, simple_loss=0.3123, pruned_loss=0.04853, over 944034.00 frames. 
+2023-03-10 11:25:51,389 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 11:26:01,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:26:08,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:26:14,041 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:26:14,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.965e+03 2.751e+03 4.165e+03 1.143e+04, threshold=5.502e+03, percent-clipped=16.0
+2023-03-10 11:26:17,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:26:19,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4965, 1.6044, 1.5029, 1.6202], device='cuda:0'), covar=tensor([0.0621, 0.0295, 0.0270, 0.0630], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0068, 0.0062, 0.0105], device='cuda:0')
+2023-03-10 11:26:34,863 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7033, 1.6540, 1.9363, 1.4974], device='cuda:0'), covar=tensor([0.1551, 0.2195, 0.1218, 0.1562], device='cuda:0'), in_proj_covar=tensor([0.0882, 0.0694, 0.0927, 0.0827], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:0')
+2023-03-10 11:26:40,361 INFO [train.py:968] (0/2) Epoch 20, batch 24050, giga_loss[loss=0.2928, simple_loss=0.3628, pruned_loss=0.1114, over 29106.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3901, pruned_loss=0.1397, over 5643012.78 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3571, pruned_loss=0.1084, over 5733290.12 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3927, pruned_loss=0.1417, over 5634686.03 frames. ], batch size: 128, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:26:45,150 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=891216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:27:09,713 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2028, 1.3339, 1.2517, 1.1814], device='cuda:0'), covar=tensor([0.1591, 0.1734, 0.1247, 0.1573], device='cuda:0'), in_proj_covar=tensor([0.1959, 0.1886, 0.1811, 0.1947], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 11:27:28,000 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:27:30,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:27:32,849 INFO [train.py:968] (0/2) Epoch 20, batch 24100, giga_loss[loss=0.3717, simple_loss=0.4118, pruned_loss=0.1658, over 28719.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3901, pruned_loss=0.1398, over 5648727.31 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3571, pruned_loss=0.1084, over 5733290.12 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3921, pruned_loss=0.1413, over 5642246.50 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:27:51,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4143, 1.6794, 1.3681, 1.2439], device='cuda:0'), covar=tensor([0.2527, 0.2493, 0.2938, 0.2254], device='cuda:0'), in_proj_covar=tensor([0.1478, 0.1071, 0.1311, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 11:27:55,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.549e+02 1.653e+03 2.127e+03 3.258e+03 9.924e+03, threshold=4.255e+03, percent-clipped=7.0
+2023-03-10 11:27:56,602 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=891287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:27:58,813 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:00,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:23,339 INFO [train.py:968] (0/2) Epoch 20, batch 24150, giga_loss[loss=0.3087, simple_loss=0.3816, pruned_loss=0.1179, over 28898.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.39, pruned_loss=0.1388, over 5641193.22 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3577, pruned_loss=0.1089, over 5729070.92 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3921, pruned_loss=0.1405, over 5637167.41 frames. ], batch size: 243, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:28:28,012 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:31,354 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:34,154 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=891321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:34,262 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:37,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:40,429 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-10 11:29:01,934 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=891347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:29:08,496 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=891353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:29:16,130 INFO [train.py:968] (0/2) Epoch 20, batch 24200, giga_loss[loss=0.3387, simple_loss=0.3981, pruned_loss=0.1397, over 27894.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3909, pruned_loss=0.1391, over 5634629.17 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3577, pruned_loss=0.1088, over 5727830.83 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3931, pruned_loss=0.1409, over 5631311.15 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:29:44,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.617e+03 1.961e+03 2.569e+03 7.252e+03, threshold=3.923e+03, percent-clipped=6.0
+2023-03-10 11:30:07,234 INFO [train.py:968] (0/2) Epoch 20, batch 24250, giga_loss[loss=0.3115, simple_loss=0.3831, pruned_loss=0.1199, over 28877.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3905, pruned_loss=0.1386, over 5629766.30 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3576, pruned_loss=0.1088, over 5731518.32 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3932, pruned_loss=0.1409, over 5621298.28 frames. ], batch size: 112, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:30:07,555 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9581, 2.9365, 1.9290, 0.9767], device='cuda:0'), covar=tensor([0.7802, 0.3319, 0.3938, 0.7065], device='cuda:0'), in_proj_covar=tensor([0.1719, 0.1617, 0.1585, 0.1398], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 11:30:58,500 INFO [train.py:968] (0/2) Epoch 20, batch 24300, giga_loss[loss=0.2762, simple_loss=0.3551, pruned_loss=0.09861, over 28653.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.386, pruned_loss=0.1336, over 5638758.60 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3578, pruned_loss=0.109, over 5735697.70 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3886, pruned_loss=0.1359, over 5625701.73 frames. ], batch size: 242, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:31:25,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.623e+03 2.047e+03 2.804e+03 8.130e+03, threshold=4.094e+03, percent-clipped=10.0
+2023-03-10 11:31:46,067 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=891505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:31:51,486 INFO [train.py:968] (0/2) Epoch 20, batch 24350, giga_loss[loss=0.302, simple_loss=0.3706, pruned_loss=0.1167, over 28899.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3844, pruned_loss=0.1314, over 5642420.85 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3579, pruned_loss=0.1093, over 5732428.64 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3866, pruned_loss=0.1331, over 5634380.14 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:31:56,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7103, 1.9557, 1.5893, 1.7569], device='cuda:0'), covar=tensor([0.2687, 0.2659, 0.3144, 0.2381], device='cuda:0'), in_proj_covar=tensor([0.1479, 0.1072, 0.1311, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 11:32:04,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5655, 1.9078, 1.7048, 1.6519], device='cuda:0'), covar=tensor([0.1984, 0.2371, 0.2218, 0.2273], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0745, 0.0709, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 11:32:38,191 INFO [train.py:968] (0/2) Epoch 20, batch 24400, giga_loss[loss=0.3416, simple_loss=0.3988, pruned_loss=0.1422, over 28487.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3817, pruned_loss=0.1289, over 5664710.28 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3585, pruned_loss=0.1098, over 5737414.40 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3836, pruned_loss=0.1303, over 5651579.23 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:33:03,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.807e+02 1.956e+03 2.445e+03 3.361e+03 1.237e+04, threshold=4.890e+03, percent-clipped=10.0
+2023-03-10 11:33:29,400 INFO [train.py:968] (0/2) Epoch 20, batch 24450, giga_loss[loss=0.2806, simple_loss=0.3538, pruned_loss=0.1037, over 28825.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.377, pruned_loss=0.1254, over 5664022.55 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3578, pruned_loss=0.1096, over 5742177.36 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3797, pruned_loss=0.1272, over 5647342.57 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:33:58,649 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=891642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:34:17,693 INFO [train.py:968] (0/2) Epoch 20, batch 24500, giga_loss[loss=0.3035, simple_loss=0.376, pruned_loss=0.1155, over 29038.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.375, pruned_loss=0.1241, over 5677995.00 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3574, pruned_loss=0.1094, over 5745525.83 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3779, pruned_loss=0.126, over 5660589.35 frames. ], batch size: 128, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:34:44,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.838e+02 1.676e+03 2.100e+03 2.788e+03 5.496e+03, threshold=4.199e+03, percent-clipped=3.0
+2023-03-10 11:35:15,433 INFO [train.py:968] (0/2) Epoch 20, batch 24550, giga_loss[loss=0.3074, simple_loss=0.3759, pruned_loss=0.1194, over 29026.00 frames. ], tot_loss[loss=0.311, simple_loss=0.375, pruned_loss=0.1235, over 5675404.99 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3574, pruned_loss=0.1096, over 5747475.71 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3775, pruned_loss=0.1251, over 5659017.74 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:36:04,451 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=891759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:36:05,882 INFO [train.py:968] (0/2) Epoch 20, batch 24600, giga_loss[loss=0.2928, simple_loss=0.3744, pruned_loss=0.1056, over 28916.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.373, pruned_loss=0.1216, over 5669118.24 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3572, pruned_loss=0.1095, over 5740538.94 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3755, pruned_loss=0.1232, over 5660268.54 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:36:08,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7374, 1.8702, 1.5823, 1.8876], device='cuda:0'), covar=tensor([0.2917, 0.2892, 0.3312, 0.2562], device='cuda:0'), in_proj_covar=tensor([0.1477, 0.1071, 0.1310, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 11:36:23,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1488, 1.4575, 1.4565, 1.2906], device='cuda:0'), covar=tensor([0.1835, 0.1618, 0.2094, 0.1775], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0751, 0.0714, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 11:36:36,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.713e+02 1.506e+03 1.907e+03 2.600e+03 8.257e+03, threshold=3.815e+03, percent-clipped=9.0
+2023-03-10 11:36:59,638 INFO [train.py:968] (0/2) Epoch 20, batch 24650, giga_loss[loss=0.269, simple_loss=0.3482, pruned_loss=0.09488, over 28523.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3739, pruned_loss=0.1193, over 5685200.29 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3574, pruned_loss=0.1097, over 5743584.48 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3761, pruned_loss=0.1206, over 5674247.48 frames. ], batch size: 60, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:37:15,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6036, 1.5921, 1.8296, 1.4093], device='cuda:0'), covar=tensor([0.1759, 0.2439, 0.1402, 0.1729], device='cuda:0'), in_proj_covar=tensor([0.0886, 0.0696, 0.0930, 0.0830], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:0')
+2023-03-10 11:37:54,244 INFO [train.py:968] (0/2) Epoch 20, batch 24700, libri_loss[loss=0.3017, simple_loss=0.3717, pruned_loss=0.1159, over 29657.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3745, pruned_loss=0.1192, over 5663556.16 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3574, pruned_loss=0.1097, over 5744746.77 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3765, pruned_loss=0.1204, over 5652320.45 frames. ], batch size: 91, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:38:09,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5230, 3.5466, 1.6300, 1.6320], device='cuda:0'), covar=tensor([0.0954, 0.0371, 0.0870, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0554, 0.0382, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 11:38:14,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:38:20,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.677e+02 1.653e+03 2.001e+03 2.540e+03 5.511e+03, threshold=4.002e+03, percent-clipped=5.0
+2023-03-10 11:38:47,856 INFO [train.py:968] (0/2) Epoch 20, batch 24750, giga_loss[loss=0.2872, simple_loss=0.3583, pruned_loss=0.108, over 28840.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3744, pruned_loss=0.1196, over 5663917.75 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3572, pruned_loss=0.1096, over 5745023.88 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3767, pruned_loss=0.1209, over 5652919.46 frames. ], batch size: 112, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:39:06,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3763, 3.0721, 1.5566, 1.4483], device='cuda:0'), covar=tensor([0.0967, 0.0353, 0.0881, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0555, 0.0382, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 11:39:23,867 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 11:39:32,865 INFO [train.py:968] (0/2) Epoch 20, batch 24800, giga_loss[loss=0.3368, simple_loss=0.3928, pruned_loss=0.1405, over 28896.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3741, pruned_loss=0.1201, over 5666811.28 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3574, pruned_loss=0.11, over 5745404.69 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.376, pruned_loss=0.121, over 5655744.85 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:39:59,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.129e+03 1.985e+03 2.589e+03 3.747e+03 8.105e+03, threshold=5.178e+03, percent-clipped=17.0
+2023-03-10 11:40:13,654 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-892000.pt
+2023-03-10 11:40:16,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7622, 1.8240, 1.6837, 1.5585], device='cuda:0'), covar=tensor([0.1754, 0.2328, 0.2514, 0.2310], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0747, 0.0710, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 11:40:23,436 INFO [train.py:968] (0/2) Epoch 20, batch 24850, giga_loss[loss=0.2811, simple_loss=0.3525, pruned_loss=0.1048, over 28841.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3732, pruned_loss=0.121, over 5659223.09 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3579, pruned_loss=0.1103, over 5743821.77 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3751, pruned_loss=0.1218, over 5648207.69 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:40:30,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=892017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:40:37,764 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=892023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:40:40,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=892026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:41:04,697 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=892055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:41:10,484 INFO [train.py:968] (0/2) Epoch 20, batch 24900, giga_loss[loss=0.3164, simple_loss=0.3777, pruned_loss=0.1276, over 28775.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3714, pruned_loss=0.1201, over 5671060.51 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3582, pruned_loss=0.1105, over 5744701.28 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.373, pruned_loss=0.1209, over 5658940.70 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:41:34,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.648e+03 2.073e+03 2.864e+03 8.680e+03, threshold=4.147e+03, percent-clipped=5.0
+2023-03-10 11:41:49,254 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=892102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:41:57,222 INFO [train.py:968] (0/2) Epoch 20, batch 24950, giga_loss[loss=0.2966, simple_loss=0.366, pruned_loss=0.1135, over 28608.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3709, pruned_loss=0.1198, over 5675121.71 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3582, pruned_loss=0.1106, over 5747768.87 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3724, pruned_loss=0.1205, over 5661467.30 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:42:17,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=892134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:42:42,790 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=892160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:42:43,143 INFO [train.py:968] (0/2) Epoch 20, batch 25000, giga_loss[loss=0.2723, simple_loss=0.3504, pruned_loss=0.09706, over 28883.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3696, pruned_loss=0.1177, over 5688294.49 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3578, pruned_loss=0.1105, over 5753649.65 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3715, pruned_loss=0.1187, over 5669676.66 frames. ], batch size: 227, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:42:44,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=892163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:43:09,632 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-10 11:43:12,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.600e+02 1.438e+03 1.861e+03 2.331e+03 4.993e+03, threshold=3.722e+03, percent-clipped=5.0
+2023-03-10 11:43:15,939 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=892192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:43:34,338 INFO [train.py:968] (0/2) Epoch 20, batch 25050, giga_loss[loss=0.3093, simple_loss=0.3849, pruned_loss=0.1168, over 28845.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3699, pruned_loss=0.1179, over 5678003.62 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3589, pruned_loss=0.1113, over 5755026.18 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.371, pruned_loss=0.1183, over 5658938.77 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:43:43,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3814, 1.6332, 1.3143, 1.2746], device='cuda:0'), covar=tensor([0.2728, 0.2747, 0.3162, 0.2449], device='cuda:0'), in_proj_covar=tensor([0.1476, 0.1071, 0.1310, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 11:44:20,576 INFO [train.py:968] (0/2) Epoch 20, batch 25100, giga_loss[loss=0.3994, simple_loss=0.4273, pruned_loss=0.1858, over 26624.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3692, pruned_loss=0.1172, over 5682059.63 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3589, pruned_loss=0.1114, over 5753306.97 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3705, pruned_loss=0.1176, over 5665193.24 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:44:32,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2637, 1.5050, 1.0350, 1.0977], device='cuda:0'), covar=tensor([0.1069, 0.0612, 0.1289, 0.1572], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0446, 0.0514, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 11:44:36,595 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=892277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:44:39,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=892280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:44:44,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.726e+03 2.260e+03 3.389e+03 1.119e+04, threshold=4.520e+03, percent-clipped=20.0
+2023-03-10 11:45:06,280 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=892309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:45:07,358 INFO [train.py:968] (0/2) Epoch 20, batch 25150, giga_loss[loss=0.3268, simple_loss=0.3754, pruned_loss=0.1391, over 28546.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3679, pruned_loss=0.1171, over 5696784.83 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3584, pruned_loss=0.1113, over 5756422.88 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3698, pruned_loss=0.1179, over 5677808.88 frames. ], batch size: 78, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:45:25,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3041, 1.6502, 1.3385, 1.0255], device='cuda:0'), covar=tensor([0.2472, 0.2561, 0.2857, 0.2286], device='cuda:0'), in_proj_covar=tensor([0.1478, 0.1074, 0.1312, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 11:45:58,806 INFO [train.py:968] (0/2) Epoch 20, batch 25200, giga_loss[loss=0.3082, simple_loss=0.3685, pruned_loss=0.1239, over 28455.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3689, pruned_loss=0.1189, over 5691523.04 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3592, pruned_loss=0.1119, over 5759224.29 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3699, pruned_loss=0.1191, over 5672960.37 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:45:59,268 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.44 vs. limit=5.0
+2023-03-10 11:46:23,143 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.497e+02 1.763e+03 2.284e+03 3.714e+03 8.264e+03, threshold=4.569e+03, percent-clipped=16.0
+2023-03-10 11:46:34,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7421, 1.8929, 1.3680, 1.4201], device='cuda:0'), covar=tensor([0.0999, 0.0669, 0.1080, 0.1171], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0446, 0.0514, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 11:46:46,077 INFO [train.py:968] (0/2) Epoch 20, batch 25250, giga_loss[loss=0.3544, simple_loss=0.4006, pruned_loss=0.1541, over 28637.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3692, pruned_loss=0.1197, over 5699265.43 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3599, pruned_loss=0.1123, over 5757778.25 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3696, pruned_loss=0.1197, over 5684248.84 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:47:21,089 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-10 11:47:23,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6867, 2.0922, 1.8878, 1.4628], device='cuda:0'), covar=tensor([0.2720, 0.2137, 0.2374, 0.2687], device='cuda:0'), in_proj_covar=tensor([0.1950, 0.1880, 0.1809, 0.1941], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 11:47:23,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=892449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:47:36,564 INFO [train.py:968] (0/2) Epoch 20, batch 25300, giga_loss[loss=0.268, simple_loss=0.331, pruned_loss=0.1025, over 28999.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3657, pruned_loss=0.1174, over 5700909.38 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3594, pruned_loss=0.1121, over 5759942.19 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3666, pruned_loss=0.1177, over 5685846.13 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:47:40,683 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-10 11:47:53,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=892477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:48:02,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.192e+03 1.782e+03 2.250e+03 2.988e+03 9.938e+03, threshold=4.499e+03, percent-clipped=10.0
+2023-03-10 11:48:03,609 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 11:48:16,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8864, 1.1219, 2.8202, 2.6960], device='cuda:0'), covar=tensor([0.1609, 0.2543, 0.0616, 0.1645], device='cuda:0'), in_proj_covar=tensor([0.0747, 0.0640, 0.0949, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 11:48:23,232 INFO [train.py:968] (0/2) Epoch 20, batch 25350, giga_loss[loss=0.3206, simple_loss=0.3779, pruned_loss=0.1317, over 28769.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.365, pruned_loss=0.1174, over 5701665.98 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3597, pruned_loss=0.1122, over 5764159.18 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3656, pruned_loss=0.1176, over 5683862.42 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:48:36,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7616, 1.7941, 1.9824, 1.5214], device='cuda:0'), covar=tensor([0.2000, 0.2513, 0.1577, 0.1818], device='cuda:0'), in_proj_covar=tensor([0.0887, 0.0698, 0.0932, 0.0831], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 11:49:12,063 INFO [train.py:968] (0/2) Epoch 20, batch 25400, libri_loss[loss=0.2871, simple_loss=0.359, pruned_loss=0.1076, over 29548.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3647, pruned_loss=0.1177, over 5696126.34 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3594, pruned_loss=0.1121, over 5766820.58 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3655, pruned_loss=0.1182, over 5677524.45 frames. ], batch size: 83, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:49:35,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.799e+03 2.324e+03 2.890e+03 1.054e+04, threshold=4.648e+03, percent-clipped=6.0
+2023-03-10 11:49:58,686 INFO [train.py:968] (0/2) Epoch 20, batch 25450, giga_loss[loss=0.2783, simple_loss=0.3583, pruned_loss=0.09914, over 28436.00 frames. ], tot_loss[loss=0.301, simple_loss=0.366, pruned_loss=0.118, over 5691516.59 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3595, pruned_loss=0.1123, over 5759700.73 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3668, pruned_loss=0.1184, over 5679996.36 frames. ], batch size: 65, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:50:04,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4318, 4.5680, 1.5232, 1.7285], device='cuda:0'), covar=tensor([0.1036, 0.0379, 0.0939, 0.1330], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0552, 0.0380, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 11:50:06,854 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=892620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:50:09,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=892623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:50:35,901 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=892652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:50:45,874 INFO [train.py:968] (0/2) Epoch 20, batch 25500, giga_loss[loss=0.2504, simple_loss=0.3304, pruned_loss=0.08522, over 28509.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3667, pruned_loss=0.1177, over 5690157.56 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3594, pruned_loss=0.1124, over 5760284.89 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3675, pruned_loss=0.118, over 5680018.85 frames. ], batch size: 65, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:51:10,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4774, 1.5485, 1.6837, 1.5370], device='cuda:0'), covar=tensor([0.3144, 0.2739, 0.2169, 0.2413], device='cuda:0'), in_proj_covar=tensor([0.1948, 0.1880, 0.1808, 0.1940], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 11:51:12,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.595e+03 2.024e+03 2.939e+03 8.207e+03, threshold=4.048e+03, percent-clipped=4.0
+2023-03-10 11:51:29,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4124, 1.4634, 1.2388, 1.4611], device='cuda:0'), covar=tensor([0.0783, 0.0345, 0.0356, 0.0889], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0069, 0.0062, 0.0105], device='cuda:0')
+2023-03-10 11:51:32,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-10 11:51:32,754 INFO [train.py:968] (0/2) Epoch 20, batch 25550, libri_loss[loss=0.3067, simple_loss=0.3713, pruned_loss=0.121, over 29256.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.366, pruned_loss=0.1165, over 5695149.60 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1125, over 5764211.29 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3669, pruned_loss=0.1168, over 5681551.18 frames. ], batch size: 94, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:51:35,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.33 vs. limit=5.0
+2023-03-10 11:51:41,759 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=892718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:51:54,224 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=892733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:52:22,295 INFO [train.py:968] (0/2) Epoch 20, batch 25600, giga_loss[loss=0.3828, simple_loss=0.4284, pruned_loss=0.1686, over 27582.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3671, pruned_loss=0.1175, over 5686101.62 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1124, over 5761753.99 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.368, pruned_loss=0.1178, over 5676246.04 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:52:28,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2528, 1.5267, 1.5106, 1.2873], device='cuda:0'), covar=tensor([0.1784, 0.1600, 0.2252, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0748, 0.0710, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 11:52:47,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.664e+02 1.739e+03 2.218e+03 3.353e+03 1.139e+04, threshold=4.436e+03, percent-clipped=17.0
+2023-03-10 11:53:12,114 INFO [train.py:968] (0/2) Epoch 20, batch 25650, giga_loss[loss=0.2942, simple_loss=0.3624, pruned_loss=0.113, over 29015.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3699, pruned_loss=0.1199, over 5690961.79 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3591, pruned_loss=0.1123, over 5765448.29 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.371, pruned_loss=0.1205, over 5678172.86 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:53:19,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1591, 1.1480, 3.6601, 3.2108], device='cuda:0'), covar=tensor([0.1679, 0.2814, 0.0489, 0.1148], device='cuda:0'), in_proj_covar=tensor([0.0750, 0.0642, 0.0952, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 11:53:21,311 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8741, 5.7032, 5.4512, 3.1299], device='cuda:0'), covar=tensor([0.0462, 0.0584, 0.0714, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.1218, 0.1131, 0.0960, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 11:53:22,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=892824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:54:00,550 INFO [train.py:968] (0/2) Epoch 20, batch 25700, giga_loss[loss=0.417, simple_loss=0.437, pruned_loss=0.1985, over 27973.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3726, pruned_loss=0.1238, over 5672709.27 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5749598.41 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3735, pruned_loss=0.1242, over 5675221.04 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:54:30,244 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-10 11:54:33,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.190e+03 1.961e+03 2.683e+03 4.137e+03 7.783e+03, threshold=5.366e+03, percent-clipped=20.0
+2023-03-10 11:54:57,687 INFO [train.py:968] (0/2) Epoch 20, batch 25750, giga_loss[loss=0.2736, simple_loss=0.3476, pruned_loss=0.0998, over 29050.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3746, pruned_loss=0.1265, over 5672676.03 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5751302.52 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3751, pruned_loss=0.1268, over 5672338.69 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:55:00,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-10 11:55:05,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4153, 3.1348, 1.4274, 1.5139], device='cuda:0'), covar=tensor([0.0924, 0.0355, 0.0844, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0553, 0.0381, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 11:55:42,873 INFO [train.py:968] (0/2) Epoch 20, batch 25800, giga_loss[loss=0.3301, simple_loss=0.3856, pruned_loss=0.1373, over 28951.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3748, pruned_loss=0.127, over 5673070.15 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1127, over 5741996.39 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3757, pruned_loss=0.1276, over 5678808.61 frames. ], batch size: 213, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:55:49,759 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=892967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:55:52,405 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=892970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:56:13,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 2.026e+03 2.584e+03 3.601e+03 6.530e+03, threshold=5.167e+03, percent-clipped=6.0
+2023-03-10 11:56:20,721 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=892999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:56:35,352 INFO [train.py:968] (0/2) Epoch 20, batch 25850, giga_loss[loss=0.3677, simple_loss=0.4105, pruned_loss=0.1625, over 27649.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3736, pruned_loss=0.1266, over 5666019.61 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3599, pruned_loss=0.1129, over 5744579.97 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3743, pruned_loss=0.1271, over 5667024.51 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:56:40,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3555, 1.7494, 1.5486, 1.4408], device='cuda:0'), covar=tensor([0.0758, 0.0306, 0.0300, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0069, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 11:56:50,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8936, 1.2004, 1.3603, 1.0313], device='cuda:0'), covar=tensor([0.1958, 0.1417, 0.2345, 0.1766], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0748, 0.0711, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 11:57:14,939 INFO [train.py:968] (0/2) Epoch 20, batch 25900, giga_loss[loss=0.3159, simple_loss=0.3803, pruned_loss=0.1258, over 28330.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3729, pruned_loss=0.1243, over 5680522.63 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1132, over 5749489.53 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3738, pruned_loss=0.125, over 5674090.37 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:57:24,701 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=893071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:57:42,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.517e+03 2.032e+03 2.617e+03 6.843e+03, threshold=4.064e+03, percent-clipped=1.0
+2023-03-10 11:57:46,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=893093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:58:02,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=893108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:58:06,113 INFO [train.py:968] (0/2) Epoch 20, batch 25950, giga_loss[loss=0.2926, simple_loss=0.3639, pruned_loss=0.1106, over 28652.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1212, over 5669857.96 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3596, pruned_loss=0.1129, over 5750307.40 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3711, pruned_loss=0.1221, over 5662808.69 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:58:50,926 INFO [train.py:968] (0/2) Epoch 20, batch 26000, giga_loss[loss=0.2974, simple_loss=0.3609, pruned_loss=0.117, over 28689.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3676, pruned_loss=0.1197, over 5673019.53 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5753635.87 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.1209, over 5661888.34 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:59:17,530 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.918e+02 1.739e+03 2.234e+03 2.978e+03 6.197e+03, threshold=4.469e+03, percent-clipped=8.0
+2023-03-10 11:59:39,591 INFO [train.py:968] (0/2) Epoch 20, batch 26050, giga_loss[loss=0.298, simple_loss=0.3609, pruned_loss=0.1176, over 28886.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3661, pruned_loss=0.1193, over 5674814.84 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3597, pruned_loss=0.1131, over 5755756.86 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3672, pruned_loss=0.1201, over 5663129.38 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:59:47,995 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 12:00:09,091 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=893236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:12,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=893239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:24,780 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=893251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:28,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=893254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:34,870 INFO [train.py:968] (0/2) Epoch 20, batch 26100, giga_loss[loss=0.3181, simple_loss=0.3829, pruned_loss=0.1266, over 28688.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3685, pruned_loss=0.1221, over 5661310.52 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3597, pruned_loss=0.1131, over 5756895.74 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3694, pruned_loss=0.1228, over 5650470.87 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:00:37,941 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3817, 3.5757, 1.5326, 1.5793], device='cuda:0'), covar=tensor([0.1013, 0.0366, 0.0894, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0556, 0.0382, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 12:00:39,866 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=893268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:40,570 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=893269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:52,945 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=893283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:01:00,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.135e+03 1.700e+03 2.221e+03 3.217e+03 8.968e+03, threshold=4.443e+03, percent-clipped=4.0
+2023-03-10 12:01:18,266 INFO [train.py:968] (0/2) Epoch 20, batch 26150, libri_loss[loss=0.267, simple_loss=0.3434, pruned_loss=0.09534, over 29537.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3705, pruned_loss=0.1225, over 5673207.44 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3596, pruned_loss=0.113, over 5762699.90 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3717, pruned_loss=0.1235, over 5655252.82 frames. ], batch size: 81, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:02:06,878 INFO [train.py:968] (0/2) Epoch 20, batch 26200, giga_loss[loss=0.2967, simple_loss=0.3794, pruned_loss=0.107, over 28918.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3745, pruned_loss=0.1225, over 5676630.61 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3599, pruned_loss=0.1133, over 5764622.50 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3756, pruned_loss=0.1232, over 5658206.27 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:02:33,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.754e+02 1.655e+03 1.993e+03 3.011e+03 7.204e+03, threshold=3.985e+03, percent-clipped=7.0
+2023-03-10 12:02:58,005 INFO [train.py:968] (0/2) Epoch 20, batch 26250, giga_loss[loss=0.316, simple_loss=0.3769, pruned_loss=0.1276, over 28745.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3766, pruned_loss=0.1233, over 5672529.23 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3603, pruned_loss=0.1137, over 5767932.79 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3775, pruned_loss=0.1239, over 5653110.29 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:03:33,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=893446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:03:48,644 INFO [train.py:968] (0/2) Epoch 20, batch 26300, giga_loss[loss=0.3071, simple_loss=0.3751, pruned_loss=0.1195, over 28508.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3775, pruned_loss=0.1243, over 5666976.16 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3603, pruned_loss=0.1136, over 5770713.08 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3785, pruned_loss=0.1249, over 5647415.51 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:03:55,224 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-10 12:04:14,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.192e+02 1.746e+03 2.208e+03 2.971e+03 6.173e+03, threshold=4.417e+03, percent-clipped=13.0
+2023-03-10 12:04:31,769 INFO [train.py:968] (0/2) Epoch 20, batch 26350, giga_loss[loss=0.3175, simple_loss=0.376, pruned_loss=0.1295, over 28852.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3778, pruned_loss=0.1251, over 5665035.83 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3605, pruned_loss=0.114, over 5760110.40 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3788, pruned_loss=0.1254, over 5655898.13 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:04:50,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0902, 1.2408, 3.2562, 2.9458], device='cuda:0'), covar=tensor([0.1653, 0.2676, 0.0536, 0.1477], device='cuda:0'), in_proj_covar=tensor([0.0751, 0.0642, 0.0952, 0.0895], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 12:05:22,596 INFO [train.py:968] (0/2) Epoch 20, batch 26400, giga_loss[loss=0.2455, simple_loss=0.323, pruned_loss=0.08399, over 28354.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3775, pruned_loss=0.1257, over 5659913.43 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3604, pruned_loss=0.1139, over 5763216.52 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3788, pruned_loss=0.1264, over 5647459.96 frames. ], batch size: 65, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 12:05:36,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2300, 1.4635, 1.5450, 1.2632], device='cuda:0'), covar=tensor([0.1905, 0.1692, 0.2279, 0.1952], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0752, 0.0714, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 12:05:48,713 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=893589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:05:49,719 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.808e+03 2.275e+03 3.218e+03 8.379e+03, threshold=4.551e+03, percent-clipped=10.0
+2023-03-10 12:05:50,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=893592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:05:58,571 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-10 12:06:09,729 INFO [train.py:968] (0/2) Epoch 20, batch 26450, libri_loss[loss=0.3205, simple_loss=0.3795, pruned_loss=0.1308, over 26044.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3747, pruned_loss=0.1244, over 5642526.76 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3599, pruned_loss=0.1137, over 5747281.84 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3771, pruned_loss=0.1259, over 5641478.92 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:06:16,371 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=893621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:06:29,598 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-10 12:06:35,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4489, 1.7215, 1.5985, 1.6366], device='cuda:0'), covar=tensor([0.0642, 0.0280, 0.0265, 0.0668], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0069, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 12:06:36,903 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=893644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:06:52,952 INFO [train.py:968] (0/2) Epoch 20, batch 26500, libri_loss[loss=0.3204, simple_loss=0.3873, pruned_loss=0.1267, over 29357.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3711, pruned_loss=0.1221, over 5657767.13 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3597, pruned_loss=0.1135, over 5753625.81 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3737, pruned_loss=0.1237, over 5647805.43 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:07:25,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.851e+02 1.889e+03 2.490e+03 3.321e+03 9.194e+03, threshold=4.980e+03, percent-clipped=11.0
+2023-03-10 12:07:44,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0057, 5.3324, 2.0628, 2.2618], device='cuda:0'), covar=tensor([0.0872, 0.0314, 0.0830, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0554, 0.0381, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 12:07:50,663 INFO [train.py:968] (0/2) Epoch 20, batch 26550, giga_loss[loss=0.2622, simple_loss=0.3299, pruned_loss=0.0972, over 28717.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3709, pruned_loss=0.1232, over 5650673.26 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.36, pruned_loss=0.1136, over 5753459.68 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3728, pruned_loss=0.1245, over 5642616.37 frames. ], batch size: 71, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:08:37,454 INFO [train.py:968] (0/2) Epoch 20, batch 26600, giga_loss[loss=0.2917, simple_loss=0.3593, pruned_loss=0.1121, over 28647.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3721, pruned_loss=0.1242, over 5643428.89 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3602, pruned_loss=0.1139, over 5745226.08 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3736, pruned_loss=0.1252, over 5640721.34 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:09:02,536 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=893787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:09:04,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=893790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:09:04,908 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.808e+03 2.186e+03 3.216e+03 1.521e+04, threshold=4.373e+03, percent-clipped=5.0
+2023-03-10 12:09:22,661 INFO [train.py:968] (0/2) Epoch 20, batch 26650, giga_loss[loss=0.4051, simple_loss=0.4298, pruned_loss=0.1901, over 26691.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3705, pruned_loss=0.1233, over 5653754.91 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.36, pruned_loss=0.1137, over 5745130.38 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3724, pruned_loss=0.1246, over 5648782.23 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:09:28,029 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 12:09:29,368 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=893819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:09:33,233 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 12:10:04,722 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 12:10:11,467 INFO [train.py:968] (0/2) Epoch 20, batch 26700, giga_loss[loss=0.271, simple_loss=0.3494, pruned_loss=0.09627, over 28922.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3685, pruned_loss=0.1221, over 5669573.76 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.36, pruned_loss=0.1137, over 5747203.95 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3701, pruned_loss=0.1233, over 5662256.39 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:10:43,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.852e+03 2.322e+03 3.330e+03 1.215e+04, threshold=4.643e+03, percent-clipped=15.0
+2023-03-10 12:11:03,996 INFO [train.py:968] (0/2) Epoch 20, batch 26750, giga_loss[loss=0.2719, simple_loss=0.3385, pruned_loss=0.1027, over 28351.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3686, pruned_loss=0.122, over 5668593.25 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.36, pruned_loss=0.1136, over 5747697.19 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.37, pruned_loss=0.1231, over 5661293.97 frames. ], batch size: 71, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:11:50,061 INFO [train.py:968] (0/2) Epoch 20, batch 26800, giga_loss[loss=0.287, simple_loss=0.3574, pruned_loss=0.1083, over 28652.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1224, over 5672588.33 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.36, pruned_loss=0.1138, over 5751304.05 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3719, pruned_loss=0.1234, over 5661805.97 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 12:12:18,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.677e+03 2.126e+03 2.684e+03 1.001e+04, threshold=4.253e+03, percent-clipped=1.0
+2023-03-10 12:12:27,118 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-894000.pt
+2023-03-10 12:12:41,255 INFO [train.py:968] (0/2) Epoch 20, batch 26850, giga_loss[loss=0.3605, simple_loss=0.3902, pruned_loss=0.1654, over 23804.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.373, pruned_loss=0.1244, over 5660542.33 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3602, pruned_loss=0.1138, over 5749244.73 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3742, pruned_loss=0.1253, over 5652425.26 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:12:53,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3799, 1.7196, 1.3500, 1.2850], device='cuda:0'), covar=tensor([0.2408, 0.2317, 0.2605, 0.2077], device='cuda:0'), in_proj_covar=tensor([0.1485, 0.1077, 0.1319, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 12:13:04,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6577, 1.8527, 1.5076, 1.8323], device='cuda:0'), covar=tensor([0.2457, 0.2576, 0.2756, 0.2341], device='cuda:0'), in_proj_covar=tensor([0.1484, 0.1077, 0.1319, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 12:13:26,031 INFO [train.py:968] (0/2) Epoch 20, batch 26900, giga_loss[loss=0.2949, simple_loss=0.3801, pruned_loss=0.1049, over 29004.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3737, pruned_loss=0.1253, over 5654210.78 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3606, pruned_loss=0.1141, over 5736764.22 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3747, pruned_loss=0.1262, over 5656661.79 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:13:54,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.788e+03 2.234e+03 3.069e+03 8.481e+03, threshold=4.468e+03, percent-clipped=12.0
+2023-03-10 12:14:10,365 INFO [train.py:968] (0/2) Epoch 20, batch 26950, giga_loss[loss=0.2618, simple_loss=0.3507, pruned_loss=0.08643, over 28907.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3739, pruned_loss=0.1223, over 5666521.73 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3601, pruned_loss=0.1138, over 5732998.08 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3756, pruned_loss=0.1236, over 5668390.20 frames. ], batch size: 99, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:14:42,066 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4044, 2.2201, 1.4063, 1.5472], device='cuda:0'), covar=tensor([0.0757, 0.0425, 0.0680, 0.1044], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0556, 0.0381, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 12:14:51,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4549, 1.7031, 1.6690, 1.2473], device='cuda:0'), covar=tensor([0.1954, 0.2561, 0.1624, 0.1900], device='cuda:0'), in_proj_covar=tensor([0.0890, 0.0700, 0.0934, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 12:14:58,700 INFO [train.py:968] (0/2) Epoch 20, batch 27000, giga_loss[loss=0.3454, simple_loss=0.3892, pruned_loss=0.1509, over 23671.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3749, pruned_loss=0.1217, over 5656052.62 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3602, pruned_loss=0.1139, over 5727552.74 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3766, pruned_loss=0.1229, over 5660095.58 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:14:58,704 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 12:15:07,804 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2052, simple_loss=0.3114, pruned_loss=0.04954, over 944034.00 frames. 
+2023-03-10 12:15:07,805 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 12:15:19,029 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-10 12:15:35,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.000e+02 1.419e+03 1.952e+03 2.494e+03 8.550e+03, threshold=3.903e+03, percent-clipped=4.0
+2023-03-10 12:15:52,949 INFO [train.py:968] (0/2) Epoch 20, batch 27050, libri_loss[loss=0.2866, simple_loss=0.3543, pruned_loss=0.1094, over 26407.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3773, pruned_loss=0.122, over 5663861.75 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3601, pruned_loss=0.1138, over 5727637.09 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.379, pruned_loss=0.1232, over 5666189.07 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:16:41,061 INFO [train.py:968] (0/2) Epoch 20, batch 27100, libri_loss[loss=0.2786, simple_loss=0.3513, pruned_loss=0.103, over 29491.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3799, pruned_loss=0.125, over 5658440.50 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3602, pruned_loss=0.1139, over 5721738.62 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3816, pruned_loss=0.1261, over 5663692.93 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:16:51,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4693, 1.6815, 1.6419, 1.5301], device='cuda:0'), covar=tensor([0.1716, 0.1787, 0.1988, 0.1791], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0748, 0.0710, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 12:17:14,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.168e+03 1.684e+03 2.220e+03 3.334e+03 1.885e+04, threshold=4.440e+03, percent-clipped=12.0
+2023-03-10 12:17:16,363 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=894294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:17:30,486 INFO [train.py:968] (0/2) Epoch 20, batch 27150, giga_loss[loss=0.3368, simple_loss=0.3924, pruned_loss=0.1405, over 28217.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3805, pruned_loss=0.1262, over 5671488.10 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3602, pruned_loss=0.1139, over 5724770.10 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3823, pruned_loss=0.1272, over 5672032.43 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:18:21,567 INFO [train.py:968] (0/2) Epoch 20, batch 27200, giga_loss[loss=0.3304, simple_loss=0.3921, pruned_loss=0.1343, over 29014.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3801, pruned_loss=0.1271, over 5664118.83 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3596, pruned_loss=0.1137, over 5719421.17 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3825, pruned_loss=0.1284, over 5668673.86 frames. ], batch size: 128, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:18:50,273 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-10 12:18:57,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.637e+03 2.101e+03 2.785e+03 9.285e+03, threshold=4.202e+03, percent-clipped=6.0
+2023-03-10 12:19:12,284 INFO [train.py:968] (0/2) Epoch 20, batch 27250, giga_loss[loss=0.2665, simple_loss=0.348, pruned_loss=0.09252, over 28779.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3776, pruned_loss=0.1248, over 5665718.80 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3597, pruned_loss=0.1138, over 5716799.93 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3798, pruned_loss=0.1261, over 5670584.77 frames. ], batch size: 119, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:19:58,723 INFO [train.py:968] (0/2) Epoch 20, batch 27300, giga_loss[loss=0.2771, simple_loss=0.3647, pruned_loss=0.09468, over 28934.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3779, pruned_loss=0.1236, over 5651478.10 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3599, pruned_loss=0.1139, over 5707343.05 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3799, pruned_loss=0.1247, over 5663002.82 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:20:27,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.321e+02 1.495e+03 2.033e+03 2.638e+03 5.839e+03, threshold=4.065e+03, percent-clipped=6.0
+2023-03-10 12:20:41,398 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=894505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:20:48,104 INFO [train.py:968] (0/2) Epoch 20, batch 27350, giga_loss[loss=0.3344, simple_loss=0.398, pruned_loss=0.1354, over 27937.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3776, pruned_loss=0.1227, over 5654746.82 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3595, pruned_loss=0.1136, over 5713482.34 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.38, pruned_loss=0.124, over 5657212.01 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:20:56,428 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=894521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:21:37,674 INFO [train.py:968] (0/2) Epoch 20, batch 27400, giga_loss[loss=0.3278, simple_loss=0.3878, pruned_loss=0.1339, over 27890.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3791, pruned_loss=0.1239, over 5662080.44 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3592, pruned_loss=0.1133, over 5719090.79 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3817, pruned_loss=0.1254, over 5657780.40 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:22:12,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.068e+03 1.713e+03 2.573e+03 3.776e+03 9.761e+03, threshold=5.145e+03, percent-clipped=22.0
+2023-03-10 12:22:29,083 INFO [train.py:968] (0/2) Epoch 20, batch 27450, libri_loss[loss=0.2896, simple_loss=0.3618, pruned_loss=0.1087, over 27841.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3802, pruned_loss=0.1254, over 5657398.09 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.36, pruned_loss=0.1139, over 5720042.10 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3821, pruned_loss=0.1264, over 5651862.72 frames. ], batch size: 116, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:22:29,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9478, 1.3049, 1.0583, 0.3613], device='cuda:0'), covar=tensor([0.1955, 0.1759, 0.2282, 0.3215], device='cuda:0'), in_proj_covar=tensor([0.1727, 0.1639, 0.1592, 0.1404], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 12:23:13,338 INFO [train.py:968] (0/2) Epoch 20, batch 27500, giga_loss[loss=0.3198, simple_loss=0.3762, pruned_loss=0.1317, over 29012.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3787, pruned_loss=0.1254, over 5662485.76 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.36, pruned_loss=0.1139, over 5723702.58 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3812, pruned_loss=0.1268, over 5651847.97 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:23:21,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=894669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:23:23,229 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=894672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:23:39,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5391, 1.6341, 1.7538, 1.3386], device='cuda:0'), covar=tensor([0.1750, 0.2560, 0.1396, 0.1655], device='cuda:0'), in_proj_covar=tensor([0.0891, 0.0702, 0.0935, 0.0834], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 12:23:45,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3290, 1.8746, 1.6235, 1.5525], device='cuda:0'), covar=tensor([0.0751, 0.0330, 0.0302, 0.0789], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0118, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0105], device='cuda:0')
+2023-03-10 12:23:45,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.850e+03 2.396e+03 3.740e+03 1.560e+04, threshold=4.791e+03, percent-clipped=13.0
+2023-03-10 12:24:00,621 INFO [train.py:968] (0/2) Epoch 20, batch 27550, giga_loss[loss=0.259, simple_loss=0.3398, pruned_loss=0.08913, over 28981.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3777, pruned_loss=0.1252, over 5672978.18 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3602, pruned_loss=0.1141, over 5717661.52 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3801, pruned_loss=0.1266, over 5668586.71 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 1.0
+2023-03-10 12:24:09,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1753, 1.2618, 1.1087, 0.9035], device='cuda:0'), covar=tensor([0.0913, 0.0470, 0.0997, 0.1070], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0448, 0.0516, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 12:24:57,536 INFO [train.py:968] (0/2) Epoch 20, batch 27600, giga_loss[loss=0.2595, simple_loss=0.3396, pruned_loss=0.08972, over 28982.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3752, pruned_loss=0.1244, over 5658819.61 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3598, pruned_loss=0.1138, over 5711162.53 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3778, pruned_loss=0.1259, over 5660044.32 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:25:27,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5145, 1.6217, 1.5781, 1.4121], device='cuda:0'), covar=tensor([0.2438, 0.2320, 0.2106, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.1964, 0.1885, 0.1821, 0.1959], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 12:25:31,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.784e+03 2.350e+03 3.404e+03 8.697e+03, threshold=4.700e+03, percent-clipped=12.0
+2023-03-10 12:25:44,895 INFO [train.py:968] (0/2) Epoch 20, batch 27650, giga_loss[loss=0.2958, simple_loss=0.3604, pruned_loss=0.1156, over 28215.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3734, pruned_loss=0.1234, over 5669044.68 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1136, over 5718224.98 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3762, pruned_loss=0.1252, over 5662245.76 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:25:46,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=894812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:25:48,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=894815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:26:14,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=894844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:26:31,007 INFO [train.py:968] (0/2) Epoch 20, batch 27700, giga_loss[loss=0.3338, simple_loss=0.3986, pruned_loss=0.1345, over 29007.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3739, pruned_loss=0.1247, over 5669480.69 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3601, pruned_loss=0.1139, over 5719700.79 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3761, pruned_loss=0.1261, over 5661571.78 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:26:50,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=894880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:27:02,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.695e+03 2.192e+03 3.338e+03 9.206e+03, threshold=4.384e+03, percent-clipped=7.0
+2023-03-10 12:27:03,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=894896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:27:14,910 INFO [train.py:968] (0/2) Epoch 20, batch 27750, libri_loss[loss=0.349, simple_loss=0.3986, pruned_loss=0.1497, over 19614.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3716, pruned_loss=0.1224, over 5647445.06 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3607, pruned_loss=0.1143, over 5696602.77 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3733, pruned_loss=0.1236, over 5660267.08 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:27:53,119 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-10 12:28:04,650 INFO [train.py:968] (0/2) Epoch 20, batch 27800, libri_loss[loss=0.2673, simple_loss=0.3412, pruned_loss=0.09664, over 29533.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3687, pruned_loss=0.1188, over 5653432.60 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3609, pruned_loss=0.1144, over 5698526.36 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3701, pruned_loss=0.1197, over 5661187.08 frames. ], batch size: 80, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:28:21,356 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-10 12:28:36,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.979e+02 1.440e+03 1.790e+03 2.278e+03 7.644e+03, threshold=3.579e+03, percent-clipped=1.0
+2023-03-10 12:28:38,928 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=894996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:28:54,608 INFO [train.py:968] (0/2) Epoch 20, batch 27850, giga_loss[loss=0.2859, simple_loss=0.3562, pruned_loss=0.1078, over 28573.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3661, pruned_loss=0.1166, over 5653061.56 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3608, pruned_loss=0.1143, over 5699157.75 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3674, pruned_loss=0.1175, over 5657970.40 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:29:08,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:11,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:23,524 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:26,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:32,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:39,223 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:44,586 INFO [train.py:968] (0/2) Epoch 20, batch 27900, giga_loss[loss=0.2862, simple_loss=0.3505, pruned_loss=0.1109, over 28689.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3654, pruned_loss=0.1165, over 5659137.78 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3604, pruned_loss=0.114, over 5705000.72 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.367, pruned_loss=0.1177, over 5656395.74 frames. ], batch size: 242, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:29:57,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:30:21,823 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:30:23,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.682e+03 2.064e+03 2.685e+03 9.706e+03, threshold=4.128e+03, percent-clipped=8.0
+2023-03-10 12:30:38,830 INFO [train.py:968] (0/2) Epoch 20, batch 27950, giga_loss[loss=0.2803, simple_loss=0.3601, pruned_loss=0.1003, over 29080.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3632, pruned_loss=0.1162, over 5642537.61 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3607, pruned_loss=0.1141, over 5697409.44 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3643, pruned_loss=0.1171, over 5646188.96 frames. ], batch size: 128, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:31:00,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5156, 2.9095, 1.6523, 1.5674], device='cuda:0'), covar=tensor([0.0793, 0.0329, 0.0703, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0554, 0.0381, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 12:31:32,424 INFO [train.py:968] (0/2) Epoch 20, batch 28000, libri_loss[loss=0.274, simple_loss=0.3374, pruned_loss=0.1053, over 28611.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3637, pruned_loss=0.1165, over 5647967.57 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3607, pruned_loss=0.114, over 5699583.97 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3646, pruned_loss=0.1173, over 5648223.41 frames. ], batch size: 63, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:31:59,926 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:32:04,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:32:07,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.753e+02 1.759e+03 2.335e+03 3.152e+03 8.742e+03, threshold=4.671e+03, percent-clipped=12.0
+2023-03-10 12:32:17,191 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:32:22,852 INFO [train.py:968] (0/2) Epoch 20, batch 28050, giga_loss[loss=0.3389, simple_loss=0.3813, pruned_loss=0.1483, over 23401.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3672, pruned_loss=0.1183, over 5651957.35 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3607, pruned_loss=0.1141, over 5702818.64 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3679, pruned_loss=0.1188, over 5648481.53 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:32:34,432 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:33:12,540 INFO [train.py:968] (0/2) Epoch 20, batch 28100, giga_loss[loss=0.4713, simple_loss=0.4725, pruned_loss=0.235, over 26484.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3682, pruned_loss=0.1188, over 5639918.50 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3612, pruned_loss=0.1143, over 5693920.27 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3685, pruned_loss=0.1192, over 5644587.69 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:33:36,727 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 12:33:36,742 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 12:33:45,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.584e+03 2.201e+03 3.072e+03 1.240e+04, threshold=4.402e+03, percent-clipped=9.0
+2023-03-10 12:34:01,530 INFO [train.py:968] (0/2) Epoch 20, batch 28150, giga_loss[loss=0.3262, simple_loss=0.3755, pruned_loss=0.1384, over 28947.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3677, pruned_loss=0.1192, over 5643618.44 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1143, over 5697343.30 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3681, pruned_loss=0.1195, over 5643571.43 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:34:01,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8092, 1.8502, 1.3805, 1.4036], device='cuda:0'), covar=tensor([0.0975, 0.0738, 0.1075, 0.1289], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0447, 0.0516, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 12:34:47,610 INFO [train.py:968] (0/2) Epoch 20, batch 28200, giga_loss[loss=0.2643, simple_loss=0.3411, pruned_loss=0.0938, over 28666.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3675, pruned_loss=0.1192, over 5647333.49 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3613, pruned_loss=0.1144, over 5697679.84 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3677, pruned_loss=0.1195, over 5645790.49 frames. ], batch size: 78, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:34:58,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:35:21,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.876e+02 1.640e+03 2.084e+03 3.017e+03 8.820e+03, threshold=4.168e+03, percent-clipped=10.0
+2023-03-10 12:35:36,805 INFO [train.py:968] (0/2) Epoch 20, batch 28250, giga_loss[loss=0.2763, simple_loss=0.3574, pruned_loss=0.0976, over 28818.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3693, pruned_loss=0.1203, over 5652753.22 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3614, pruned_loss=0.1143, over 5702228.00 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5646194.01 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:36:20,998 INFO [train.py:968] (0/2) Epoch 20, batch 28300, giga_loss[loss=0.2951, simple_loss=0.3676, pruned_loss=0.1112, over 28736.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3692, pruned_loss=0.1199, over 5659767.56 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3607, pruned_loss=0.1139, over 5707943.68 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3703, pruned_loss=0.1208, over 5648062.30 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:36:28,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:36:34,986 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-10 12:36:55,984 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:36:57,078 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.581e+03 2.033e+03 2.892e+03 9.658e+03, threshold=4.066e+03, percent-clipped=11.0
+2023-03-10 12:37:13,133 INFO [train.py:968] (0/2) Epoch 20, batch 28350, giga_loss[loss=0.2811, simple_loss=0.3506, pruned_loss=0.1058, over 28896.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3719, pruned_loss=0.1225, over 5643266.13 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1144, over 5697881.57 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3723, pruned_loss=0.1229, over 5641064.08 frames. ], batch size: 112, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:37:16,465 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:37:19,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:37:28,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7425, 1.8960, 1.9436, 1.4877], device='cuda:0'), covar=tensor([0.1831, 0.2482, 0.1448, 0.1742], device='cuda:0'), in_proj_covar=tensor([0.0893, 0.0705, 0.0938, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 12:37:28,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9099, 1.1945, 1.3210, 0.9855], device='cuda:0'), covar=tensor([0.1938, 0.1471, 0.2381, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0750, 0.0712, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 12:37:50,439 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:37:50,453 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:37:52,735 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 12:37:59,738 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:38:05,548 INFO [train.py:968] (0/2) Epoch 20, batch 28400, giga_loss[loss=0.2684, simple_loss=0.3413, pruned_loss=0.09774, over 28581.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3735, pruned_loss=0.124, over 5647896.12 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3617, pruned_loss=0.1144, over 5700549.05 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3737, pruned_loss=0.1244, over 5642973.91 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 12:38:18,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0584, 1.0710, 5.4523, 3.7763], device='cuda:0'), covar=tensor([0.1522, 0.3008, 0.0457, 0.1013], device='cuda:0'), in_proj_covar=tensor([0.0749, 0.0642, 0.0948, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 12:38:24,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:38:44,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.217e+03 1.811e+03 2.306e+03 3.249e+03 5.840e+03, threshold=4.611e+03, percent-clipped=7.0
+2023-03-10 12:38:59,701 INFO [train.py:968] (0/2) Epoch 20, batch 28450, giga_loss[loss=0.291, simple_loss=0.373, pruned_loss=0.1045, over 28541.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3723, pruned_loss=0.1214, over 5654465.29 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3615, pruned_loss=0.1144, over 5707594.79 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3731, pruned_loss=0.122, over 5642525.79 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:39:00,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:39:02,167 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:39:31,979 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:39:51,798 INFO [train.py:968] (0/2) Epoch 20, batch 28500, giga_loss[loss=0.3145, simple_loss=0.3743, pruned_loss=0.1273, over 28845.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3724, pruned_loss=0.122, over 5628947.54 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3614, pruned_loss=0.1146, over 5692297.75 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3733, pruned_loss=0.1225, over 5632699.81 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:40:28,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.896e+03 2.511e+03 3.660e+03 1.032e+04, threshold=5.021e+03, percent-clipped=10.0
+2023-03-10 12:40:44,165 INFO [train.py:968] (0/2) Epoch 20, batch 28550, giga_loss[loss=0.3254, simple_loss=0.3878, pruned_loss=0.1315, over 27989.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3742, pruned_loss=0.124, over 5627087.32 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1147, over 5692380.64 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3748, pruned_loss=0.1244, over 5629276.66 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:40:59,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:41:02,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:41:37,096 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:41:44,501 INFO [train.py:968] (0/2) Epoch 20, batch 28600, giga_loss[loss=0.2477, simple_loss=0.3274, pruned_loss=0.08396, over 29044.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3732, pruned_loss=0.1245, over 5632463.37 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3614, pruned_loss=0.1147, over 5698941.63 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3742, pruned_loss=0.125, over 5626420.86 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:41:45,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7063, 2.4642, 1.5616, 0.8470], device='cuda:0'), covar=tensor([0.8235, 0.3449, 0.4004, 0.7421], device='cuda:0'), in_proj_covar=tensor([0.1735, 0.1637, 0.1589, 0.1407], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 12:42:24,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.718e+03 2.291e+03 3.296e+03 1.796e+04, threshold=4.582e+03, percent-clipped=12.0
+2023-03-10 12:42:35,753 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-10 12:42:40,481 INFO [train.py:968] (0/2) Epoch 20, batch 28650, giga_loss[loss=0.3149, simple_loss=0.3769, pruned_loss=0.1265, over 29085.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.373, pruned_loss=0.125, over 5633667.16 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3621, pruned_loss=0.1152, over 5703232.40 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3735, pruned_loss=0.1251, over 5623842.35 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:43:20,203 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895853.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 12:43:26,830 INFO [train.py:968] (0/2) Epoch 20, batch 28700, giga_loss[loss=0.3455, simple_loss=0.4014, pruned_loss=0.1448, over 28732.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3719, pruned_loss=0.1245, over 5638189.75 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3622, pruned_loss=0.1154, over 5688298.00 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3724, pruned_loss=0.1247, over 5642425.54 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:43:36,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:44:01,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.554e+03 1.952e+03 2.725e+03 8.595e+03, threshold=3.905e+03, percent-clipped=7.0
+2023-03-10 12:44:17,844 INFO [train.py:968] (0/2) Epoch 20, batch 28750, giga_loss[loss=0.2791, simple_loss=0.3561, pruned_loss=0.1011, over 28923.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3711, pruned_loss=0.1242, over 5639807.49 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3622, pruned_loss=0.1154, over 5690369.67 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3717, pruned_loss=0.1245, over 5640148.25 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:44:29,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:44:36,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:44:40,521 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-10 12:44:59,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:45:08,469 INFO [train.py:968] (0/2) Epoch 20, batch 28800, giga_loss[loss=0.2798, simple_loss=0.3464, pruned_loss=0.1067, over 28488.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.372, pruned_loss=0.1247, over 5650961.67 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3624, pruned_loss=0.1156, over 5692917.75 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3724, pruned_loss=0.1249, over 5648068.19 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:45:31,913 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2603, 1.2328, 3.9043, 3.2389], device='cuda:0'), covar=tensor([0.1618, 0.2737, 0.0447, 0.0998], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0639, 0.0945, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 12:45:44,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.616e+03 2.061e+03 2.925e+03 6.530e+03, threshold=4.122e+03, percent-clipped=13.0
+2023-03-10 12:45:47,736 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-896000.pt
+2023-03-10 12:45:57,315 INFO [train.py:968] (0/2) Epoch 20, batch 28850, libri_loss[loss=0.2876, simple_loss=0.3581, pruned_loss=0.1086, over 29524.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3718, pruned_loss=0.1247, over 5660629.95 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3622, pruned_loss=0.1156, over 5698762.06 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3726, pruned_loss=0.1251, over 5651659.15 frames. ], batch size: 81, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:45:59,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:46:01,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:46:30,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:46:45,451 INFO [train.py:968] (0/2) Epoch 20, batch 28900, giga_loss[loss=0.3223, simple_loss=0.3866, pruned_loss=0.129, over 28988.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3741, pruned_loss=0.1263, over 5661752.59 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3623, pruned_loss=0.1156, over 5697327.67 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3751, pruned_loss=0.127, over 5653816.48 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:46:48,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:46:53,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:46:57,325 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:47:00,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:47:18,810 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:47:19,241 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.322e+03 2.072e+03 2.589e+03 3.942e+03 9.768e+03, threshold=5.177e+03, percent-clipped=22.0
+2023-03-10 12:47:26,244 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:47:32,030 INFO [train.py:968] (0/2) Epoch 20, batch 28950, giga_loss[loss=0.3009, simple_loss=0.366, pruned_loss=0.1179, over 28906.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.375, pruned_loss=0.1272, over 5655103.69 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3628, pruned_loss=0.1159, over 5684053.28 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3758, pruned_loss=0.1278, over 5660116.08 frames. ], batch size: 213, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:47:35,696 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=896114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:47:56,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1905, 1.4797, 1.4561, 1.0907], device='cuda:0'), covar=tensor([0.1523, 0.2362, 0.1252, 0.1493], device='cuda:0'), in_proj_covar=tensor([0.0890, 0.0703, 0.0935, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 12:48:15,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7607, 4.5940, 4.3735, 2.1384], device='cuda:0'), covar=tensor([0.0472, 0.0620, 0.0642, 0.2047], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.1141, 0.0970, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 12:48:22,221 INFO [train.py:968] (0/2) Epoch 20, batch 29000, giga_loss[loss=0.2868, simple_loss=0.3588, pruned_loss=0.1074, over 28816.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3741, pruned_loss=0.1265, over 5659307.46 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3626, pruned_loss=0.1158, over 5686077.16 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3749, pruned_loss=0.1272, over 5661373.40 frames. ], batch size: 242, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:48:56,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.086e+03 1.759e+03 2.084e+03 2.806e+03 8.270e+03, threshold=4.168e+03, percent-clipped=4.0
+2023-03-10 12:49:10,028 INFO [train.py:968] (0/2) Epoch 20, batch 29050, giga_loss[loss=0.327, simple_loss=0.3631, pruned_loss=0.1454, over 23786.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3744, pruned_loss=0.126, over 5660810.17 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3623, pruned_loss=0.1156, over 5679473.27 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3757, pruned_loss=0.1269, over 5667994.82 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:49:12,322 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2734, 1.1713, 1.1242, 1.4543], device='cuda:0'), covar=tensor([0.0772, 0.0369, 0.0356, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0069, 0.0062, 0.0105], device='cuda:0')
+2023-03-10 12:49:27,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=896228.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 12:49:40,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2200, 1.4168, 1.3605, 1.1523], device='cuda:0'), covar=tensor([0.2387, 0.2500, 0.1657, 0.2060], device='cuda:0'), in_proj_covar=tensor([0.1954, 0.1884, 0.1810, 0.1951], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 12:49:59,178 INFO [train.py:968] (0/2) Epoch 20, batch 29100, giga_loss[loss=0.3606, simple_loss=0.4025, pruned_loss=0.1594, over 28271.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3766, pruned_loss=0.1278, over 5649636.43 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3622, pruned_loss=0.1156, over 5671588.44 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.378, pruned_loss=0.129, over 5661883.80 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:50:31,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.193e+02 1.822e+03 2.456e+03 3.300e+03 7.857e+03, threshold=4.911e+03, percent-clipped=14.0
+2023-03-10 12:50:36,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1913, 1.5471, 1.4779, 1.0737], device='cuda:0'), covar=tensor([0.1555, 0.2646, 0.1403, 0.1664], device='cuda:0'), in_proj_covar=tensor([0.0890, 0.0702, 0.0934, 0.0834], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 12:50:43,613 INFO [train.py:968] (0/2) Epoch 20, batch 29150, giga_loss[loss=0.2827, simple_loss=0.3552, pruned_loss=0.1051, over 28790.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3771, pruned_loss=0.128, over 5664214.19 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3623, pruned_loss=0.1156, over 5677528.77 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3785, pruned_loss=0.1291, over 5668381.30 frames. ], batch size: 99, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:51:01,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=896328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:51:19,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2672, 1.4676, 1.4655, 1.3273], device='cuda:0'), covar=tensor([0.1411, 0.1300, 0.1954, 0.1402], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0755, 0.0717, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 12:51:19,896 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-10 12:51:32,358 INFO [train.py:968] (0/2) Epoch 20, batch 29200, giga_loss[loss=0.2722, simple_loss=0.3453, pruned_loss=0.09956, over 28651.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3788, pruned_loss=0.1292, over 5662365.94 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3622, pruned_loss=0.1155, over 5681011.47 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3801, pruned_loss=0.1304, over 5662395.08 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 12:51:40,478 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896371.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 12:51:42,957 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896374.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 12:52:00,706 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=896393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:52:04,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.982e+02 1.577e+03 2.002e+03 2.654e+03 7.239e+03, threshold=4.003e+03, percent-clipped=2.0
+2023-03-10 12:52:09,861 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896403.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 12:52:17,305 INFO [train.py:968] (0/2) Epoch 20, batch 29250, giga_loss[loss=0.3262, simple_loss=0.3879, pruned_loss=0.1322, over 28302.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3777, pruned_loss=0.1281, over 5665160.87 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3621, pruned_loss=0.1153, over 5684287.41 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3792, pruned_loss=0.1296, over 5661927.89 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 12:52:47,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8920, 3.7086, 3.5063, 1.7010], device='cuda:0'), covar=tensor([0.0728, 0.0869, 0.0904, 0.2203], device='cuda:0'), in_proj_covar=tensor([0.1231, 0.1142, 0.0970, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 12:52:56,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4956, 2.9406, 1.5927, 1.6789], device='cuda:0'), covar=tensor([0.0775, 0.0357, 0.0725, 0.1025], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0555, 0.0381, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 12:53:08,230 INFO [train.py:968] (0/2) Epoch 20, batch 29300, giga_loss[loss=0.303, simple_loss=0.3705, pruned_loss=0.1177, over 28615.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3772, pruned_loss=0.1271, over 5640012.58 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3619, pruned_loss=0.1152, over 5675296.17 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3789, pruned_loss=0.1286, over 5644529.55 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:53:17,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:53:21,388 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:53:36,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=896489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:53:44,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.553e+03 1.906e+03 2.353e+03 5.320e+03, threshold=3.812e+03, percent-clipped=2.0
+2023-03-10 12:53:50,259 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:53:58,751 INFO [train.py:968] (0/2) Epoch 20, batch 29350, libri_loss[loss=0.3057, simple_loss=0.3742, pruned_loss=0.1185, over 29529.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3768, pruned_loss=0.1261, over 5634429.75 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3617, pruned_loss=0.1152, over 5666946.85 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3787, pruned_loss=0.1275, over 5644191.36 frames. ], batch size: 83, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:54:37,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2525, 0.8159, 0.8911, 1.3220], device='cuda:0'), covar=tensor([0.0748, 0.0396, 0.0353, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0069, 0.0062, 0.0105], device='cuda:0')
+2023-03-10 12:54:47,926 INFO [train.py:968] (0/2) Epoch 20, batch 29400, giga_loss[loss=0.2922, simple_loss=0.3616, pruned_loss=0.1114, over 28971.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3756, pruned_loss=0.1248, over 5639876.91 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3616, pruned_loss=0.1151, over 5668470.63 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3774, pruned_loss=0.1262, over 5645924.09 frames. ], batch size: 227, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:55:22,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.719e+03 2.067e+03 2.831e+03 8.324e+03, threshold=4.133e+03, percent-clipped=12.0
+2023-03-10 12:55:36,466 INFO [train.py:968] (0/2) Epoch 20, batch 29450, giga_loss[loss=0.2829, simple_loss=0.3527, pruned_loss=0.1065, over 28805.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3741, pruned_loss=0.124, over 5649492.20 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3618, pruned_loss=0.1152, over 5671399.71 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3756, pruned_loss=0.1251, over 5651199.39 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:55:46,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3614, 1.7968, 1.6735, 1.5561], device='cuda:0'), covar=tensor([0.1734, 0.1355, 0.1847, 0.1566], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0754, 0.0715, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 12:55:48,160 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-10 12:55:56,022 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:55:57,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:56:09,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3036, 3.2756, 1.4487, 1.4459], device='cuda:0'), covar=tensor([0.1031, 0.0377, 0.0929, 0.1392], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0555, 0.0381, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 12:56:24,326 INFO [train.py:968] (0/2) Epoch 20, batch 29500, giga_loss[loss=0.2848, simple_loss=0.3611, pruned_loss=0.1043, over 28546.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3745, pruned_loss=0.1245, over 5650452.07 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3618, pruned_loss=0.1154, over 5673496.30 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.376, pruned_loss=0.1255, over 5649126.82 frames. ], batch size: 60, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:56:24,492 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=896661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:56:27,981 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:57:01,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.899e+02 1.593e+03 2.150e+03 2.868e+03 6.137e+03, threshold=4.300e+03, percent-clipped=8.0
+2023-03-10 12:57:13,695 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3212, 2.0668, 1.5185, 0.5961], device='cuda:0'), covar=tensor([0.4334, 0.2443, 0.3757, 0.4737], device='cuda:0'), in_proj_covar=tensor([0.1737, 0.1646, 0.1593, 0.1412], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 12:57:14,038 INFO [train.py:968] (0/2) Epoch 20, batch 29550, giga_loss[loss=0.2813, simple_loss=0.3563, pruned_loss=0.1031, over 28893.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3739, pruned_loss=0.124, over 5652600.87 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3612, pruned_loss=0.1149, over 5675055.54 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.376, pruned_loss=0.1254, over 5649793.38 frames. ], batch size: 112, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:58:03,061 INFO [train.py:968] (0/2) Epoch 20, batch 29600, giga_loss[loss=0.3276, simple_loss=0.3857, pruned_loss=0.1348, over 28894.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.374, pruned_loss=0.1251, over 5656404.44 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3613, pruned_loss=0.1151, over 5679647.03 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3758, pruned_loss=0.1263, over 5649883.35 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:58:11,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=896768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:58:22,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9610, 1.3359, 1.0715, 0.3792], device='cuda:0'), covar=tensor([0.3282, 0.2603, 0.3933, 0.4604], device='cuda:0'), in_proj_covar=tensor([0.1735, 0.1644, 0.1592, 0.1411], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 12:58:38,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.830e+02 1.649e+03 2.203e+03 3.274e+03 1.008e+04, threshold=4.407e+03, percent-clipped=12.0
+2023-03-10 12:58:50,842 INFO [train.py:968] (0/2) Epoch 20, batch 29650, giga_loss[loss=0.407, simple_loss=0.4309, pruned_loss=0.1916, over 26611.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3749, pruned_loss=0.1261, over 5660905.01 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3614, pruned_loss=0.1152, over 5673764.45 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3764, pruned_loss=0.1271, over 5660842.15 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:58:51,157 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=896811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:59:42,683 INFO [train.py:968] (0/2) Epoch 20, batch 29700, giga_loss[loss=0.3154, simple_loss=0.3792, pruned_loss=0.1258, over 28604.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3783, pruned_loss=0.129, over 5658037.20 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3621, pruned_loss=0.1158, over 5675377.50 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.379, pruned_loss=0.1294, over 5656144.14 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:00:24,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.689e+02 1.621e+03 2.064e+03 2.782e+03 5.394e+03, threshold=4.128e+03, percent-clipped=7.0
+2023-03-10 13:00:35,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9796, 2.1197, 1.4457, 1.8096], device='cuda:0'), covar=tensor([0.0992, 0.0707, 0.1105, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0449, 0.0519, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 13:00:39,068 INFO [train.py:968] (0/2) Epoch 20, batch 29750, giga_loss[loss=0.3483, simple_loss=0.3836, pruned_loss=0.1565, over 23544.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3777, pruned_loss=0.1287, over 5644539.91 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3624, pruned_loss=0.1159, over 5676691.61 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3782, pruned_loss=0.129, over 5641848.96 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:00:39,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:00:42,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:01:09,061 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:01:26,266 INFO [train.py:968] (0/2) Epoch 20, batch 29800, giga_loss[loss=0.2787, simple_loss=0.3491, pruned_loss=0.1041, over 28925.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3769, pruned_loss=0.1276, over 5653088.84 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3623, pruned_loss=0.1159, over 5679036.84 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3775, pruned_loss=0.1279, over 5648680.53 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:01:34,794 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-10 13:01:48,139 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=896979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:01:59,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7727, 1.8034, 1.2413, 1.3899], device='cuda:0'), covar=tensor([0.0952, 0.0661, 0.1121, 0.1177], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0446, 0.0516, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 13:02:03,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.599e+03 2.186e+03 3.100e+03 1.001e+04, threshold=4.373e+03, percent-clipped=16.0
+2023-03-10 13:02:14,910 INFO [train.py:968] (0/2) Epoch 20, batch 29850, giga_loss[loss=0.2538, simple_loss=0.3277, pruned_loss=0.08996, over 28845.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3764, pruned_loss=0.1264, over 5661978.88 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3626, pruned_loss=0.1159, over 5687606.24 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3771, pruned_loss=0.1271, over 5649645.19 frames. ], batch size: 99, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:02:42,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:02:58,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3074, 2.0007, 1.6089, 1.6662], device='cuda:0'), covar=tensor([0.0791, 0.0292, 0.0304, 0.0820], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0118, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 13:03:05,962 INFO [train.py:968] (0/2) Epoch 20, batch 29900, giga_loss[loss=0.342, simple_loss=0.4097, pruned_loss=0.1371, over 28679.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3757, pruned_loss=0.1252, over 5663981.75 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3624, pruned_loss=0.1158, over 5691125.89 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3767, pruned_loss=0.126, over 5650676.61 frames. ], batch size: 242, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:03:10,266 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 13:03:41,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.614e+03 2.009e+03 2.647e+03 9.455e+03, threshold=4.017e+03, percent-clipped=3.0
+2023-03-10 13:03:41,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2163, 2.5442, 1.2348, 1.3916], device='cuda:0'), covar=tensor([0.1006, 0.0392, 0.0903, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0555, 0.0381, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 13:03:53,687 INFO [train.py:968] (0/2) Epoch 20, batch 29950, giga_loss[loss=0.3867, simple_loss=0.4263, pruned_loss=0.1736, over 28007.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3758, pruned_loss=0.1257, over 5658378.23 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3632, pruned_loss=0.1163, over 5686500.10 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3763, pruned_loss=0.1261, over 5651362.74 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:04:42,939 INFO [train.py:968] (0/2) Epoch 20, batch 30000, giga_loss[loss=0.3003, simple_loss=0.3735, pruned_loss=0.1135, over 28346.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3758, pruned_loss=0.1259, over 5665014.68 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3633, pruned_loss=0.1163, over 5688356.10 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3762, pruned_loss=0.1263, over 5657637.70 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:04:42,945 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 13:04:51,922 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2085, simple_loss=0.317, pruned_loss=0.05001, over 944034.00 frames. 
+2023-03-10 13:04:51,923 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 13:04:52,919 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=897162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:05:03,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1490, 1.2998, 1.3777, 1.1415], device='cuda:0'), covar=tensor([0.2486, 0.2482, 0.1482, 0.2145], device='cuda:0'), in_proj_covar=tensor([0.1964, 0.1890, 0.1812, 0.1957], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 13:05:07,283 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:05:11,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:05:15,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:05:24,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9456, 3.7807, 3.5627, 1.7603], device='cuda:0'), covar=tensor([0.0715, 0.0869, 0.0829, 0.2295], device='cuda:0'), in_proj_covar=tensor([0.1230, 0.1143, 0.0971, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 13:05:27,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.478e+02 1.685e+03 2.177e+03 2.958e+03 6.645e+03, threshold=4.355e+03, percent-clipped=10.0
+2023-03-10 13:05:39,100 INFO [train.py:968] (0/2) Epoch 20, batch 30050, giga_loss[loss=0.2664, simple_loss=0.3426, pruned_loss=0.09514, over 28938.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.371, pruned_loss=0.1229, over 5670144.62 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3632, pruned_loss=0.1163, over 5688822.71 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3715, pruned_loss=0.1234, over 5663739.02 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:05:39,337 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:06:28,687 INFO [train.py:968] (0/2) Epoch 20, batch 30100, giga_loss[loss=0.3297, simple_loss=0.3747, pruned_loss=0.1423, over 28907.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3673, pruned_loss=0.1217, over 5648009.74 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.363, pruned_loss=0.1161, over 5682080.37 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3681, pruned_loss=0.1224, over 5648137.16 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:07:03,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.176e+03 1.713e+03 2.145e+03 3.138e+03 6.182e+03, threshold=4.290e+03, percent-clipped=10.0
+2023-03-10 13:07:11,848 INFO [train.py:968] (0/2) Epoch 20, batch 30150, giga_loss[loss=0.281, simple_loss=0.3402, pruned_loss=0.1109, over 28832.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3668, pruned_loss=0.1219, over 5659897.79 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3629, pruned_loss=0.1162, over 5685458.47 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3675, pruned_loss=0.1226, over 5656021.79 frames. ], batch size: 99, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:07:23,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3435, 1.9923, 1.4319, 0.5744], device='cuda:0'), covar=tensor([0.4864, 0.2661, 0.4156, 0.6056], device='cuda:0'), in_proj_covar=tensor([0.1735, 0.1647, 0.1594, 0.1412], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 13:07:31,317 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:07:33,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:07:54,046 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:07:58,759 INFO [train.py:968] (0/2) Epoch 20, batch 30200, giga_loss[loss=0.2567, simple_loss=0.3357, pruned_loss=0.08886, over 28751.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3653, pruned_loss=0.1213, over 5640222.30 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3629, pruned_loss=0.116, over 5681852.95 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.366, pruned_loss=0.1222, over 5639345.29 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:07:59,149 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:08:31,376 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=897393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:08:38,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.596e+03 2.022e+03 2.697e+03 5.514e+03, threshold=4.043e+03, percent-clipped=5.0
+2023-03-10 13:08:48,554 INFO [train.py:968] (0/2) Epoch 20, batch 30250, giga_loss[loss=0.2661, simple_loss=0.3483, pruned_loss=0.09193, over 28477.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3647, pruned_loss=0.1189, over 5644578.35 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3633, pruned_loss=0.1164, over 5676456.80 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.365, pruned_loss=0.1194, over 5647641.89 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:09:22,377 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=897441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:09:43,836 INFO [train.py:968] (0/2) Epoch 20, batch 30300, giga_loss[loss=0.2334, simple_loss=0.3224, pruned_loss=0.0722, over 28291.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.362, pruned_loss=0.1154, over 5638000.75 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3628, pruned_loss=0.1161, over 5682095.57 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3627, pruned_loss=0.116, over 5634272.51 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:10:22,159 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:10:24,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:10:25,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.791e+02 1.450e+03 2.032e+03 2.847e+03 8.772e+03, threshold=4.064e+03, percent-clipped=14.0
+2023-03-10 13:10:32,582 INFO [train.py:968] (0/2) Epoch 20, batch 30350, giga_loss[loss=0.2881, simple_loss=0.359, pruned_loss=0.1086, over 28861.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.359, pruned_loss=0.1121, over 5655066.47 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.362, pruned_loss=0.116, over 5687006.15 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3602, pruned_loss=0.1126, over 5645919.33 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:10:51,790 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:10:58,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:11:26,025 INFO [train.py:968] (0/2) Epoch 20, batch 30400, giga_loss[loss=0.254, simple_loss=0.3319, pruned_loss=0.08809, over 28897.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3559, pruned_loss=0.1088, over 5647412.84 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3614, pruned_loss=0.1157, over 5681940.75 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3573, pruned_loss=0.1093, over 5643431.64 frames. ], batch size: 213, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:12:04,007 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.694e+02 1.482e+03 2.129e+03 3.115e+03 8.207e+03, threshold=4.257e+03, percent-clipped=13.0
+2023-03-10 13:12:12,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6467, 1.9509, 1.3727, 1.8016], device='cuda:0'), covar=tensor([0.0697, 0.0260, 0.0328, 0.0807], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0118, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 13:12:13,143 INFO [train.py:968] (0/2) Epoch 20, batch 30450, libri_loss[loss=0.2893, simple_loss=0.3576, pruned_loss=0.1105, over 29531.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3518, pruned_loss=0.1052, over 5654807.45 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3605, pruned_loss=0.1155, over 5686234.93 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3534, pruned_loss=0.1054, over 5646026.40 frames. ], batch size: 84, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:12:37,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=897636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:13:02,789 INFO [train.py:968] (0/2) Epoch 20, batch 30500, giga_loss[loss=0.2494, simple_loss=0.3403, pruned_loss=0.07928, over 28905.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.352, pruned_loss=0.1028, over 5671540.57 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.361, pruned_loss=0.1159, over 5689677.81 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3527, pruned_loss=0.1024, over 5660692.71 frames. ], batch size: 112, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:13:27,774 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:13:29,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:13:37,206 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7197, 1.9246, 1.3658, 1.4747], device='cuda:0'), covar=tensor([0.0956, 0.0549, 0.0940, 0.1123], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0441, 0.0510, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 13:13:38,731 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 13:13:49,647 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.594e+02 1.382e+03 1.819e+03 2.765e+03 8.228e+03, threshold=3.638e+03, percent-clipped=6.0
+2023-03-10 13:13:55,790 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-10 13:14:00,726 INFO [train.py:968] (0/2) Epoch 20, batch 30550, giga_loss[loss=0.2633, simple_loss=0.3461, pruned_loss=0.09026, over 28729.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3531, pruned_loss=0.1031, over 5671848.69 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3611, pruned_loss=0.1161, over 5690770.42 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3534, pruned_loss=0.1025, over 5662286.37 frames. ], batch size: 243, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:14:01,772 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:14:35,995 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2921, 1.3370, 1.3134, 1.5005], device='cuda:0'), covar=tensor([0.0777, 0.0332, 0.0341, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 13:14:51,427 INFO [train.py:968] (0/2) Epoch 20, batch 30600, giga_loss[loss=0.2744, simple_loss=0.3484, pruned_loss=0.1002, over 27977.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3506, pruned_loss=0.1013, over 5662488.51 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3606, pruned_loss=0.1161, over 5686151.72 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3511, pruned_loss=0.1005, over 5658716.30 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:14:57,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:15:16,393 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=897786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:15:31,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.275e+02 1.319e+03 1.858e+03 2.789e+03 7.301e+03, threshold=3.716e+03, percent-clipped=12.0
+2023-03-10 13:15:39,926 INFO [train.py:968] (0/2) Epoch 20, batch 30650, libri_loss[loss=0.3515, simple_loss=0.3929, pruned_loss=0.1551, over 29497.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.348, pruned_loss=0.09948, over 5662955.22 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3601, pruned_loss=0.116, over 5682060.59 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3487, pruned_loss=0.09857, over 5662182.25 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:15:46,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:16:18,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2145, 2.7902, 1.3386, 1.3660], device='cuda:0'), covar=tensor([0.1010, 0.0473, 0.1014, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0552, 0.0380, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 13:16:27,897 INFO [train.py:968] (0/2) Epoch 20, batch 30700, giga_loss[loss=0.2356, simple_loss=0.3215, pruned_loss=0.07487, over 28967.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.347, pruned_loss=0.09942, over 5642406.61 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.36, pruned_loss=0.1161, over 5667831.96 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3472, pruned_loss=0.09809, over 5653943.58 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:17:08,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.888e+02 1.444e+03 1.857e+03 2.493e+03 4.972e+03, threshold=3.713e+03, percent-clipped=7.0
+2023-03-10 13:17:16,337 INFO [train.py:968] (0/2) Epoch 20, batch 30750, giga_loss[loss=0.2442, simple_loss=0.328, pruned_loss=0.08019, over 28870.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3469, pruned_loss=0.09921, over 5652195.09 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3594, pruned_loss=0.1159, over 5670192.69 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3474, pruned_loss=0.09799, over 5658528.03 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:17:16,708 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:17:19,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:17:51,457 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:18:08,295 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:18:09,225 INFO [train.py:968] (0/2) Epoch 20, batch 30800, giga_loss[loss=0.2464, simple_loss=0.3326, pruned_loss=0.08008, over 29009.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3446, pruned_loss=0.09697, over 5651974.61 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3592, pruned_loss=0.1158, over 5669939.91 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3449, pruned_loss=0.09587, over 5656890.95 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:18:11,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:18:42,228 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:18:48,849 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-898000.pt
+2023-03-10 13:18:50,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.982e+02 1.331e+03 2.008e+03 3.006e+03 8.134e+03, threshold=4.015e+03, percent-clipped=12.0
+2023-03-10 13:18:58,778 INFO [train.py:968] (0/2) Epoch 20, batch 30850, giga_loss[loss=0.2382, simple_loss=0.324, pruned_loss=0.07619, over 28242.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3411, pruned_loss=0.09464, over 5659102.31 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3583, pruned_loss=0.1155, over 5676589.57 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3416, pruned_loss=0.09325, over 5656417.21 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:18:59,035 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=898011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:19:36,233 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=898048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:19:48,611 INFO [train.py:968] (0/2) Epoch 20, batch 30900, giga_loss[loss=0.2535, simple_loss=0.3305, pruned_loss=0.08818, over 28547.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3377, pruned_loss=0.0926, over 5672494.81 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3574, pruned_loss=0.115, over 5681411.79 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3385, pruned_loss=0.0913, over 5665558.16 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:20:24,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4079, 1.6891, 1.6803, 1.3992], device='cuda:0'), covar=tensor([0.2798, 0.1998, 0.1457, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.1926, 0.1844, 0.1768, 0.1909], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 13:20:32,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.408e+02 1.296e+03 1.689e+03 2.258e+03 4.720e+03, threshold=3.379e+03, percent-clipped=2.0
+2023-03-10 13:20:39,809 INFO [train.py:968] (0/2) Epoch 20, batch 30950, giga_loss[loss=0.2265, simple_loss=0.3074, pruned_loss=0.07284, over 28716.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3342, pruned_loss=0.09081, over 5673516.73 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3565, pruned_loss=0.1146, over 5685851.97 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3352, pruned_loss=0.08976, over 5663810.95 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:21:25,660 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=898154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:21:28,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=898157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:21:31,268 INFO [train.py:968] (0/2) Epoch 20, batch 31000, giga_loss[loss=0.2558, simple_loss=0.3172, pruned_loss=0.09722, over 24145.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3347, pruned_loss=0.09193, over 5650313.92 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3563, pruned_loss=0.1145, over 5674735.24 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3352, pruned_loss=0.09059, over 5652455.26 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:21:31,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=898161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:22:00,898 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=898186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:22:09,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8027, 1.9833, 1.4007, 1.4980], device='cuda:0'), covar=tensor([0.0944, 0.0520, 0.0959, 0.1086], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0443, 0.0513, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 13:22:18,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.677e+02 1.524e+03 2.093e+03 3.178e+03 9.950e+03, threshold=4.186e+03, percent-clipped=19.0
+2023-03-10 13:22:26,156 INFO [train.py:968] (0/2) Epoch 20, batch 31050, giga_loss[loss=0.2789, simple_loss=0.3548, pruned_loss=0.1015, over 28811.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3364, pruned_loss=0.09321, over 5647237.33 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3559, pruned_loss=0.1145, over 5678733.99 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3366, pruned_loss=0.09163, over 5644844.04 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:22:43,075 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 13:23:24,875 INFO [train.py:968] (0/2) Epoch 20, batch 31100, giga_loss[loss=0.2781, simple_loss=0.3592, pruned_loss=0.09851, over 28802.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3397, pruned_loss=0.09428, over 5647186.11 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.356, pruned_loss=0.1146, over 5684034.88 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3394, pruned_loss=0.09243, over 5639684.52 frames. ], batch size: 243, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:24:10,937 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.371e+02 1.529e+03 2.216e+03 3.431e+03 7.215e+03, threshold=4.431e+03, percent-clipped=12.0
+2023-03-10 13:24:13,686 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=898304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:24:19,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=898307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:24:22,533 INFO [train.py:968] (0/2) Epoch 20, batch 31150, libri_loss[loss=0.2817, simple_loss=0.3501, pruned_loss=0.1067, over 29475.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3401, pruned_loss=0.09466, over 5642955.45 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3555, pruned_loss=0.1144, over 5690327.48 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3398, pruned_loss=0.09276, over 5629979.95 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:24:58,829 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=898336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:25:25,447 INFO [train.py:968] (0/2) Epoch 20, batch 31200, libri_loss[loss=0.2869, simple_loss=0.3386, pruned_loss=0.1176, over 29597.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3398, pruned_loss=0.09458, over 5643329.95 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3553, pruned_loss=0.1145, over 5688068.69 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3392, pruned_loss=0.09225, over 5633031.14 frames. ], batch size: 75, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:25:39,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4345, 1.3330, 3.8402, 3.3514], device='cuda:0'), covar=tensor([0.1543, 0.2837, 0.0503, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0743, 0.0638, 0.0940, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 13:25:51,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5167, 1.8239, 1.4267, 1.7130], device='cuda:0'), covar=tensor([0.2730, 0.2597, 0.3068, 0.2238], device='cuda:0'), in_proj_covar=tensor([0.1486, 0.1072, 0.1321, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 13:26:15,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.720e+02 1.312e+03 1.636e+03 2.318e+03 4.986e+03, threshold=3.272e+03, percent-clipped=1.0
+2023-03-10 13:26:19,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3341, 1.6262, 1.5429, 1.2726], device='cuda:0'), covar=tensor([0.2920, 0.2281, 0.1695, 0.2311], device='cuda:0'), in_proj_covar=tensor([0.1923, 0.1843, 0.1771, 0.1909], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 13:26:26,775 INFO [train.py:968] (0/2) Epoch 20, batch 31250, giga_loss[loss=0.2784, simple_loss=0.3603, pruned_loss=0.09828, over 28083.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3377, pruned_loss=0.09265, over 5655877.60 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3548, pruned_loss=0.1143, over 5693700.05 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3373, pruned_loss=0.09055, over 5641630.53 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:26:44,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4181, 2.7368, 1.5241, 1.5321], device='cuda:0'), covar=tensor([0.0819, 0.0303, 0.0806, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0552, 0.0381, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 13:26:44,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=898423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:27:15,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4678, 3.0322, 1.4783, 1.5751], device='cuda:0'), covar=tensor([0.0949, 0.0353, 0.0965, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0552, 0.0381, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 13:27:23,007 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.29 vs. limit=5.0
+2023-03-10 13:27:31,514 INFO [train.py:968] (0/2) Epoch 20, batch 31300, giga_loss[loss=0.2244, simple_loss=0.3232, pruned_loss=0.06281, over 28874.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3356, pruned_loss=0.09, over 5636012.21 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3545, pruned_loss=0.1141, over 5686911.60 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3352, pruned_loss=0.08791, over 5628857.90 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:28:18,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.046e+02 1.383e+03 1.806e+03 2.866e+03 8.868e+03, threshold=3.612e+03, percent-clipped=20.0
+2023-03-10 13:28:25,029 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 13:28:27,837 INFO [train.py:968] (0/2) Epoch 20, batch 31350, giga_loss[loss=0.2285, simple_loss=0.3064, pruned_loss=0.07531, over 28981.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.334, pruned_loss=0.08959, over 5636167.49 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3538, pruned_loss=0.1139, over 5677578.46 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3335, pruned_loss=0.08708, over 5637394.82 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:29:25,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5489, 1.8099, 1.7495, 1.3055], device='cuda:0'), covar=tensor([0.1442, 0.2513, 0.1321, 0.1666], device='cuda:0'), in_proj_covar=tensor([0.0892, 0.0694, 0.0935, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 13:29:29,956 INFO [train.py:968] (0/2) Epoch 20, batch 31400, giga_loss[loss=0.2549, simple_loss=0.3342, pruned_loss=0.0878, over 28893.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3329, pruned_loss=0.08975, over 5645529.25 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3534, pruned_loss=0.1139, over 5670308.75 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3322, pruned_loss=0.08705, over 5651846.51 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:29:37,483 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=898566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:29:41,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=898569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:29:52,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3622, 1.5702, 1.5991, 1.2256], device='cuda:0'), covar=tensor([0.1723, 0.2484, 0.1446, 0.1780], device='cuda:0'), in_proj_covar=tensor([0.0891, 0.0693, 0.0934, 0.0837], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 13:30:14,624 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=898598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:30:21,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.823e+02 1.398e+03 2.044e+03 3.214e+03 8.959e+03, threshold=4.089e+03, percent-clipped=19.0
+2023-03-10 13:30:28,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4355, 1.6085, 1.6795, 1.2596], device='cuda:0'), covar=tensor([0.1844, 0.2808, 0.1552, 0.1870], device='cuda:0'), in_proj_covar=tensor([0.0891, 0.0693, 0.0935, 0.0837], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 13:30:32,669 INFO [train.py:968] (0/2) Epoch 20, batch 31450, giga_loss[loss=0.259, simple_loss=0.3416, pruned_loss=0.0882, over 28509.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3316, pruned_loss=0.08872, over 5654686.21 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3535, pruned_loss=0.114, over 5671876.34 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3308, pruned_loss=0.08634, over 5658056.60 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:30:33,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0480, 3.8693, 3.6760, 1.7718], device='cuda:0'), covar=tensor([0.0668, 0.0782, 0.0868, 0.2216], device='cuda:0'), in_proj_covar=tensor([0.1192, 0.1108, 0.0938, 0.0711], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 13:31:25,344 INFO [train.py:968] (0/2) Epoch 20, batch 31500, giga_loss[loss=0.3126, simple_loss=0.3908, pruned_loss=0.1172, over 28885.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.334, pruned_loss=0.08985, over 5653636.68 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3532, pruned_loss=0.1137, over 5671198.74 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3328, pruned_loss=0.08713, over 5657028.93 frames. ], batch size: 227, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:31:51,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3543, 2.6162, 1.2651, 1.5317], device='cuda:0'), covar=tensor([0.0960, 0.0366, 0.0944, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0551, 0.0381, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 13:32:20,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.453e+02 1.480e+03 1.996e+03 2.992e+03 8.465e+03, threshold=3.992e+03, percent-clipped=5.0
+2023-03-10 13:32:32,644 INFO [train.py:968] (0/2) Epoch 20, batch 31550, giga_loss[loss=0.2756, simple_loss=0.3563, pruned_loss=0.09743, over 28017.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3362, pruned_loss=0.09041, over 5645412.03 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3533, pruned_loss=0.1138, over 5670228.25 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3349, pruned_loss=0.08788, over 5648528.50 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:33:31,926 INFO [train.py:968] (0/2) Epoch 20, batch 31600, giga_loss[loss=0.227, simple_loss=0.3009, pruned_loss=0.07655, over 27832.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3327, pruned_loss=0.08818, over 5647188.57 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3527, pruned_loss=0.1134, over 5657592.60 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3317, pruned_loss=0.08564, over 5661461.17 frames. ], batch size: 476, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:33:41,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3050, 1.6735, 1.6856, 1.3940], device='cuda:0'), covar=tensor([0.2026, 0.1971, 0.2099, 0.2063], device='cuda:0'), in_proj_covar=tensor([0.0462, 0.0737, 0.0699, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 13:34:05,159 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 13:34:35,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.023e+02 1.537e+03 1.999e+03 2.874e+03 6.248e+03, threshold=3.999e+03, percent-clipped=7.0
+2023-03-10 13:34:45,591 INFO [train.py:968] (0/2) Epoch 20, batch 31650, giga_loss[loss=0.256, simple_loss=0.3362, pruned_loss=0.08795, over 28876.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3338, pruned_loss=0.08908, over 5660559.26 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3522, pruned_loss=0.1132, over 5662851.15 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3329, pruned_loss=0.0867, over 5666931.57 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:34:52,160 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4753, 1.6897, 1.7543, 1.3153], device='cuda:0'), covar=tensor([0.1863, 0.2693, 0.1603, 0.1915], device='cuda:0'), in_proj_covar=tensor([0.0892, 0.0693, 0.0935, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 13:35:52,282 INFO [train.py:968] (0/2) Epoch 20, batch 31700, giga_loss[loss=0.2644, simple_loss=0.3456, pruned_loss=0.09156, over 28135.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3369, pruned_loss=0.08988, over 5663830.26 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3524, pruned_loss=0.1133, over 5667040.25 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3358, pruned_loss=0.08753, over 5664851.82 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:36:30,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=898891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:36:35,286 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=898894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:36:46,460 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.363e+02 1.397e+03 1.965e+03 2.795e+03 8.026e+03, threshold=3.931e+03, percent-clipped=11.0
+2023-03-10 13:36:53,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-10 13:36:56,570 INFO [train.py:968] (0/2) Epoch 20, batch 31750, giga_loss[loss=0.2451, simple_loss=0.3411, pruned_loss=0.07455, over 28903.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3397, pruned_loss=0.08917, over 5653797.37 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3525, pruned_loss=0.1135, over 5663300.58 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3385, pruned_loss=0.08656, over 5657414.28 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:38:01,059 INFO [train.py:968] (0/2) Epoch 20, batch 31800, giga_loss[loss=0.2482, simple_loss=0.3358, pruned_loss=0.08027, over 27520.00 frames. ], tot_loss[loss=0.258, simple_loss=0.34, pruned_loss=0.08798, over 5652737.51 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3522, pruned_loss=0.1135, over 5664097.57 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.339, pruned_loss=0.08549, over 5654678.29 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:38:55,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.920e+02 1.327e+03 1.700e+03 2.355e+03 4.440e+03, threshold=3.401e+03, percent-clipped=2.0
+2023-03-10 13:39:01,663 INFO [train.py:968] (0/2) Epoch 20, batch 31850, giga_loss[loss=0.2491, simple_loss=0.3348, pruned_loss=0.08172, over 29066.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3393, pruned_loss=0.08653, over 5656070.36 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3523, pruned_loss=0.1137, over 5656358.93 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3382, pruned_loss=0.08385, over 5664381.22 frames. ], batch size: 128, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:39:24,710 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=899032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:39:58,691 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=899057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:40:02,828 INFO [train.py:968] (0/2) Epoch 20, batch 31900, giga_loss[loss=0.3065, simple_loss=0.3854, pruned_loss=0.1138, over 28041.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3411, pruned_loss=0.0888, over 5671572.27 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3525, pruned_loss=0.114, over 5664449.62 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3396, pruned_loss=0.08548, over 5671207.39 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:40:09,351 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-10 13:41:04,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.920e+02 1.277e+03 1.804e+03 2.444e+03 5.668e+03, threshold=3.609e+03, percent-clipped=7.0
+2023-03-10 13:41:13,581 INFO [train.py:968] (0/2) Epoch 20, batch 31950, giga_loss[loss=0.2253, simple_loss=0.3126, pruned_loss=0.06903, over 29104.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3396, pruned_loss=0.08943, over 5672784.45 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3525, pruned_loss=0.114, over 5667214.47 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3382, pruned_loss=0.08621, over 5670028.15 frames. ], batch size: 214, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:41:35,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3288, 1.5778, 1.3336, 1.5380], device='cuda:0'), covar=tensor([0.0754, 0.0381, 0.0352, 0.0840], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 13:42:37,676 INFO [train.py:968] (0/2) Epoch 20, batch 32000, giga_loss[loss=0.2233, simple_loss=0.3138, pruned_loss=0.06637, over 28973.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3404, pruned_loss=0.09057, over 5673669.21 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3525, pruned_loss=0.1141, over 5669239.21 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3391, pruned_loss=0.08774, over 5669846.39 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:42:46,579 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=899168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:43:35,580 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0581, 1.2293, 3.3310, 2.9660], device='cuda:0'), covar=tensor([0.1652, 0.2702, 0.0501, 0.0986], device='cuda:0'), in_proj_covar=tensor([0.0740, 0.0635, 0.0936, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 13:43:44,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.512e+02 1.259e+03 1.525e+03 2.041e+03 4.858e+03, threshold=3.049e+03, percent-clipped=3.0
+2023-03-10 13:43:51,370 INFO [train.py:968] (0/2) Epoch 20, batch 32050, giga_loss[loss=0.228, simple_loss=0.3089, pruned_loss=0.07357, over 28947.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3359, pruned_loss=0.08796, over 5679449.21 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3525, pruned_loss=0.114, over 5672911.85 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3346, pruned_loss=0.08542, over 5673287.49 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:44:59,403 INFO [train.py:968] (0/2) Epoch 20, batch 32100, giga_loss[loss=0.2339, simple_loss=0.3203, pruned_loss=0.07374, over 28480.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.333, pruned_loss=0.08613, over 5677848.01 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3525, pruned_loss=0.1139, over 5677279.66 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3317, pruned_loss=0.08366, over 5668961.42 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:45:07,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:45:11,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:45:57,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.154e+02 1.312e+03 1.730e+03 2.294e+03 4.259e+03, threshold=3.460e+03, percent-clipped=8.0
+2023-03-10 13:46:02,685 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-10 13:46:05,599 INFO [train.py:968] (0/2) Epoch 20, batch 32150, giga_loss[loss=0.2525, simple_loss=0.3392, pruned_loss=0.08285, over 28865.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3311, pruned_loss=0.08556, over 5688433.03 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3519, pruned_loss=0.1136, over 5681710.52 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3301, pruned_loss=0.08324, over 5677367.62 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:47:12,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5979, 1.8949, 1.5152, 1.8127], device='cuda:0'), covar=tensor([0.2708, 0.2594, 0.2951, 0.2502], device='cuda:0'), in_proj_covar=tensor([0.1483, 0.1072, 0.1316, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 13:47:13,867 INFO [train.py:968] (0/2) Epoch 20, batch 32200, giga_loss[loss=0.2657, simple_loss=0.3451, pruned_loss=0.09314, over 28744.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3357, pruned_loss=0.08789, over 5685498.52 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.352, pruned_loss=0.1136, over 5683901.92 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3347, pruned_loss=0.08582, over 5675014.25 frames. ], batch size: 99, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:47:16,655 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4757, 3.7532, 1.5852, 1.6317], device='cuda:0'), covar=tensor([0.0930, 0.0289, 0.0912, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0550, 0.0381, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 13:48:04,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3521, 2.8352, 1.4924, 1.4695], device='cuda:0'), covar=tensor([0.0911, 0.0400, 0.0895, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0551, 0.0382, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 13:48:07,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.771e+02 1.587e+03 2.139e+03 2.928e+03 1.163e+04, threshold=4.277e+03, percent-clipped=15.0
+2023-03-10 13:48:08,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:13,498 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:14,822 INFO [train.py:968] (0/2) Epoch 20, batch 32250, giga_loss[loss=0.2496, simple_loss=0.326, pruned_loss=0.08656, over 28650.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3358, pruned_loss=0.08902, over 5697001.05 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.352, pruned_loss=0.1136, over 5687482.81 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3346, pruned_loss=0.08689, over 5685686.75 frames. ], batch size: 242, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:48:17,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:17,539 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:22,264 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:47,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:50,449 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=899435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:58,906 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:49:01,549 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:49:19,462 INFO [train.py:968] (0/2) Epoch 20, batch 32300, giga_loss[loss=0.263, simple_loss=0.3363, pruned_loss=0.09485, over 28967.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3355, pruned_loss=0.09019, over 5672167.06 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3521, pruned_loss=0.1139, over 5669960.39 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3342, pruned_loss=0.08772, over 5680212.40 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:50:12,635 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 13:50:19,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.803e+02 1.349e+03 1.833e+03 3.111e+03 1.585e+04, threshold=3.665e+03, percent-clipped=9.0
+2023-03-10 13:50:26,162 INFO [train.py:968] (0/2) Epoch 20, batch 32350, giga_loss[loss=0.2334, simple_loss=0.3244, pruned_loss=0.07123, over 28951.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3343, pruned_loss=0.08966, over 5673814.12 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3521, pruned_loss=0.114, over 5672253.10 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3329, pruned_loss=0.08722, over 5678316.35 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:51:03,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:51:13,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:51:18,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:51:30,533 INFO [train.py:968] (0/2) Epoch 20, batch 32400, giga_loss[loss=0.233, simple_loss=0.3264, pruned_loss=0.0698, over 28883.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3359, pruned_loss=0.09026, over 5673610.17 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3515, pruned_loss=0.1137, over 5675502.17 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3345, pruned_loss=0.08746, over 5674762.61 frames. ], batch size: 112, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:51:51,385 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:51:53,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:52:00,379 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:52:37,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.990e+02 1.531e+03 1.928e+03 2.888e+03 1.217e+04, threshold=3.855e+03, percent-clipped=16.0
+2023-03-10 13:52:39,678 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:52:43,654 INFO [train.py:968] (0/2) Epoch 20, batch 32450, giga_loss[loss=0.2324, simple_loss=0.3246, pruned_loss=0.07006, over 28452.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3366, pruned_loss=0.08975, over 5670049.05 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.351, pruned_loss=0.1134, over 5677718.85 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3357, pruned_loss=0.08729, over 5668904.14 frames. ], batch size: 65, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:53:11,784 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-10 13:54:00,465 INFO [train.py:968] (0/2) Epoch 20, batch 32500, giga_loss[loss=0.2001, simple_loss=0.2895, pruned_loss=0.05535, over 29049.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3361, pruned_loss=0.0901, over 5662737.33 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.351, pruned_loss=0.1135, over 5674979.51 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3348, pruned_loss=0.08725, over 5664039.38 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:54:15,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8522, 1.9611, 1.3555, 1.7280], device='cuda:0'), covar=tensor([0.0927, 0.0664, 0.1023, 0.1123], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0441, 0.0513, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 13:54:33,463 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:54:36,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:55:00,717 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.501e+03 2.031e+03 2.943e+03 6.625e+03, threshold=4.063e+03, percent-clipped=11.0
+2023-03-10 13:55:07,006 INFO [train.py:968] (0/2) Epoch 20, batch 32550, giga_loss[loss=0.2593, simple_loss=0.3228, pruned_loss=0.09786, over 27003.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3314, pruned_loss=0.08865, over 5672239.91 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3508, pruned_loss=0.1135, over 5679703.53 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3303, pruned_loss=0.08592, over 5669009.60 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:55:14,608 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:56:10,862 INFO [train.py:968] (0/2) Epoch 20, batch 32600, giga_loss[loss=0.2391, simple_loss=0.3198, pruned_loss=0.07922, over 28823.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.327, pruned_loss=0.08681, over 5681491.53 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3506, pruned_loss=0.1134, over 5688031.04 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3253, pruned_loss=0.08366, over 5671272.86 frames. ], batch size: 263, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:57:06,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.542e+03 2.096e+03 3.289e+03 9.744e+03, threshold=4.192e+03, percent-clipped=10.0
+2023-03-10 13:57:11,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:57:12,352 INFO [train.py:968] (0/2) Epoch 20, batch 32650, giga_loss[loss=0.2269, simple_loss=0.2932, pruned_loss=0.08028, over 24039.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3275, pruned_loss=0.08733, over 5675681.58 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3506, pruned_loss=0.1133, over 5692872.03 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3255, pruned_loss=0.08411, over 5662915.89 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:58:11,587 INFO [train.py:968] (0/2) Epoch 20, batch 32700, giga_loss[loss=0.2419, simple_loss=0.3237, pruned_loss=0.08, over 28884.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3292, pruned_loss=0.08831, over 5677517.20 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.35, pruned_loss=0.1131, over 5690544.31 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3276, pruned_loss=0.08525, over 5668813.71 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:59:05,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.568e+02 1.589e+03 2.183e+03 3.304e+03 6.944e+03, threshold=4.367e+03, percent-clipped=10.0
+2023-03-10 13:59:11,141 INFO [train.py:968] (0/2) Epoch 20, batch 32750, giga_loss[loss=0.2117, simple_loss=0.2905, pruned_loss=0.0665, over 27584.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3276, pruned_loss=0.08695, over 5667426.54 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.35, pruned_loss=0.1133, over 5684154.76 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3258, pruned_loss=0.08367, over 5666220.64 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:59:21,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4451, 1.7106, 1.6305, 1.4042], device='cuda:0'), covar=tensor([0.2865, 0.2239, 0.1763, 0.2240], device='cuda:0'), in_proj_covar=tensor([0.1908, 0.1813, 0.1744, 0.1887], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 14:00:08,361 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:00:12,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:00:17,850 INFO [train.py:968] (0/2) Epoch 20, batch 32800, giga_loss[loss=0.2269, simple_loss=0.313, pruned_loss=0.07039, over 28845.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3264, pruned_loss=0.08537, over 5666104.38 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3498, pruned_loss=0.1132, over 5686139.70 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3248, pruned_loss=0.08264, over 5663451.42 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:00:40,743 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1971, 1.5711, 0.9671, 1.0930], device='cuda:0'), covar=tensor([0.1205, 0.0583, 0.1540, 0.1231], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0441, 0.0513, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 14:00:54,365 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:01:13,765 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-900000.pt
+2023-03-10 14:01:23,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.776e+02 1.357e+03 1.781e+03 2.494e+03 5.586e+03, threshold=3.561e+03, percent-clipped=4.0
+2023-03-10 14:01:32,112 INFO [train.py:968] (0/2) Epoch 20, batch 32850, giga_loss[loss=0.2007, simple_loss=0.2866, pruned_loss=0.05742, over 28802.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3259, pruned_loss=0.08553, over 5666148.92 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3498, pruned_loss=0.1132, over 5688314.73 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3244, pruned_loss=0.08314, over 5661935.37 frames. ], batch size: 119, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:02:12,928 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:02:41,906 INFO [train.py:968] (0/2) Epoch 20, batch 32900, giga_loss[loss=0.2216, simple_loss=0.3141, pruned_loss=0.06459, over 29032.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3266, pruned_loss=0.08501, over 5680290.84 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3497, pruned_loss=0.1131, over 5692098.66 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3252, pruned_loss=0.08264, over 5673132.08 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:03:47,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.963e+02 1.319e+03 1.777e+03 2.661e+03 6.934e+03, threshold=3.555e+03, percent-clipped=13.0
+2023-03-10 14:03:51,604 INFO [train.py:968] (0/2) Epoch 20, batch 32950, giga_loss[loss=0.2514, simple_loss=0.3271, pruned_loss=0.08783, over 28950.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.327, pruned_loss=0.0853, over 5682249.79 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3496, pruned_loss=0.113, over 5694250.29 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3258, pruned_loss=0.08331, over 5674703.30 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:04:25,466 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-10 14:04:49,241 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:04:50,787 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:04:56,171 INFO [train.py:968] (0/2) Epoch 20, batch 33000, giga_loss[loss=0.2584, simple_loss=0.3363, pruned_loss=0.09027, over 28490.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3281, pruned_loss=0.08686, over 5686059.65 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3495, pruned_loss=0.1129, over 5698550.44 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3267, pruned_loss=0.08473, over 5675796.90 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:04:56,175 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 14:05:05,151 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.195, simple_loss=0.2962, pruned_loss=0.0469, over 944034.00 frames. 
+2023-03-10 14:05:05,152 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 14:05:58,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.405e+02 1.441e+03 1.787e+03 2.166e+03 4.756e+03, threshold=3.574e+03, percent-clipped=6.0
+2023-03-10 14:06:05,578 INFO [train.py:968] (0/2) Epoch 20, batch 33050, giga_loss[loss=0.2176, simple_loss=0.3061, pruned_loss=0.06457, over 28057.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3281, pruned_loss=0.08621, over 5681642.78 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3498, pruned_loss=0.113, over 5703165.18 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3262, pruned_loss=0.08383, over 5669115.76 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:07:03,753 INFO [train.py:968] (0/2) Epoch 20, batch 33100, giga_loss[loss=0.2496, simple_loss=0.3331, pruned_loss=0.08304, over 28902.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3291, pruned_loss=0.08555, over 5677900.19 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.349, pruned_loss=0.1124, over 5708954.33 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3277, pruned_loss=0.08334, over 5661812.88 frames. ], batch size: 227, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:07:22,744 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4060, 3.2280, 1.5506, 1.5138], device='cuda:0'), covar=tensor([0.0963, 0.0290, 0.0904, 0.1336], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0549, 0.0381, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 14:07:34,729 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:07:59,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.914e+02 1.347e+03 1.855e+03 2.375e+03 7.973e+03, threshold=3.710e+03, percent-clipped=10.0
+2023-03-10 14:08:05,083 INFO [train.py:968] (0/2) Epoch 20, batch 33150, giga_loss[loss=0.229, simple_loss=0.3177, pruned_loss=0.07016, over 28487.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3318, pruned_loss=0.0865, over 5675181.18 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3483, pruned_loss=0.112, over 5708404.44 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.331, pruned_loss=0.0847, over 5662537.38 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:09:11,460 INFO [train.py:968] (0/2) Epoch 20, batch 33200, giga_loss[loss=0.2711, simple_loss=0.3474, pruned_loss=0.09739, over 28741.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3333, pruned_loss=0.08722, over 5669812.11 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3483, pruned_loss=0.112, over 5702562.04 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3323, pruned_loss=0.08537, over 5663701.24 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:10:08,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.592e+02 1.443e+03 2.002e+03 2.920e+03 8.528e+03, threshold=4.003e+03, percent-clipped=13.0
+2023-03-10 14:10:12,233 INFO [train.py:968] (0/2) Epoch 20, batch 33250, giga_loss[loss=0.2684, simple_loss=0.3513, pruned_loss=0.09278, over 28731.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.334, pruned_loss=0.0885, over 5670550.34 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.348, pruned_loss=0.1117, over 5701475.05 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.333, pruned_loss=0.08626, over 5665420.26 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:10:16,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=900414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:10:37,615 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1853, 1.1763, 3.5865, 3.0806], device='cuda:0'), covar=tensor([0.1665, 0.2894, 0.0486, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0742, 0.0638, 0.0937, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 14:11:11,702 INFO [train.py:968] (0/2) Epoch 20, batch 33300, giga_loss[loss=0.2654, simple_loss=0.3525, pruned_loss=0.08918, over 28441.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3295, pruned_loss=0.08537, over 5678603.47 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3475, pruned_loss=0.1114, over 5705281.43 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3288, pruned_loss=0.08338, over 5670608.65 frames. ], batch size: 370, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:11:20,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0421, 3.0305, 1.9397, 1.0542], device='cuda:0'), covar=tensor([0.8371, 0.3271, 0.4589, 0.7776], device='cuda:0'), in_proj_covar=tensor([0.1720, 0.1625, 0.1581, 0.1406], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 14:11:43,756 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:12:12,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.069e+02 1.378e+03 1.816e+03 2.692e+03 7.950e+03, threshold=3.632e+03, percent-clipped=6.0
+2023-03-10 14:12:14,130 INFO [train.py:968] (0/2) Epoch 20, batch 33350, libri_loss[loss=0.2949, simple_loss=0.3557, pruned_loss=0.117, over 29395.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3284, pruned_loss=0.08496, over 5677815.29 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3473, pruned_loss=0.1115, over 5706579.63 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3275, pruned_loss=0.0827, over 5669834.63 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:12:30,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=900529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:12:32,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=900530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:13:08,604 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=900557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:13:10,606 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4795, 1.8532, 1.4218, 1.6491], device='cuda:0'), covar=tensor([0.2819, 0.2626, 0.3204, 0.2299], device='cuda:0'), in_proj_covar=tensor([0.1479, 0.1066, 0.1313, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 14:13:11,700 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=900560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:13:11,981 INFO [train.py:968] (0/2) Epoch 20, batch 33400, giga_loss[loss=0.2532, simple_loss=0.3373, pruned_loss=0.08459, over 28752.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3267, pruned_loss=0.08459, over 5681036.13 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3467, pruned_loss=0.1111, over 5710241.28 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3261, pruned_loss=0.08246, over 5670884.27 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:13:45,624 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=900589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:14:12,125 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.560e+02 1.298e+03 1.754e+03 2.418e+03 5.620e+03, threshold=3.508e+03, percent-clipped=6.0
+2023-03-10 14:14:16,671 INFO [train.py:968] (0/2) Epoch 20, batch 33450, giga_loss[loss=0.2549, simple_loss=0.3421, pruned_loss=0.0839, over 28537.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3288, pruned_loss=0.08552, over 5668972.61 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3467, pruned_loss=0.1113, over 5699477.67 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3281, pruned_loss=0.08337, over 5669763.02 frames. ], batch size: 370, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:14:32,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1230, 1.5117, 1.4146, 1.0511], device='cuda:0'), covar=tensor([0.1510, 0.2441, 0.1323, 0.1689], device='cuda:0'), in_proj_covar=tensor([0.0885, 0.0687, 0.0929, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:0')
+2023-03-10 14:14:52,953 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:20,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=900660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:20,467 INFO [train.py:968] (0/2) Epoch 20, batch 33500, giga_loss[loss=0.2545, simple_loss=0.3409, pruned_loss=0.08401, over 28480.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3304, pruned_loss=0.08642, over 5675160.36 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3466, pruned_loss=0.1112, over 5703750.24 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3296, pruned_loss=0.08438, over 5671540.87 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:15:33,042 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:37,030 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=900672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:38,125 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=900673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:40,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=900675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:42,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=900676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:55,644 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:16:20,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=900704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:16:21,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=900705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:16:25,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.165e+02 1.375e+03 2.012e+03 2.513e+03 8.806e+03, threshold=4.024e+03, percent-clipped=15.0
+2023-03-10 14:16:30,360 INFO [train.py:968] (0/2) Epoch 20, batch 33550, giga_loss[loss=0.2803, simple_loss=0.3627, pruned_loss=0.09894, over 28468.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3322, pruned_loss=0.08841, over 5659514.56 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3467, pruned_loss=0.1112, over 5697263.20 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3312, pruned_loss=0.08636, over 5661336.44 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:17:10,202 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900738.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:17:22,296 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:17:25,468 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:17:34,665 INFO [train.py:968] (0/2) Epoch 20, batch 33600, giga_loss[loss=0.2644, simple_loss=0.3484, pruned_loss=0.09025, over 28998.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3363, pruned_loss=0.09051, over 5654121.83 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3462, pruned_loss=0.1109, over 5697994.73 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3356, pruned_loss=0.08857, over 5653640.74 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:18:19,148 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=900803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:18:23,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=900806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:18:25,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.316e+02 1.395e+03 1.861e+03 2.483e+03 5.136e+03, threshold=3.721e+03, percent-clipped=3.0
+2023-03-10 14:18:28,025 INFO [train.py:968] (0/2) Epoch 20, batch 33650, giga_loss[loss=0.2503, simple_loss=0.3352, pruned_loss=0.08265, over 28134.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.338, pruned_loss=0.09069, over 5656700.34 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.346, pruned_loss=0.1106, over 5694177.06 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3374, pruned_loss=0.08863, over 5658032.21 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:19:05,871 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=900835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:19:44,137 INFO [train.py:968] (0/2) Epoch 20, batch 33700, giga_loss[loss=0.2471, simple_loss=0.326, pruned_loss=0.08412, over 29047.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3382, pruned_loss=0.09034, over 5662577.14 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3459, pruned_loss=0.1106, over 5694896.17 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3377, pruned_loss=0.08867, over 5662762.78 frames. ], batch size: 187, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:19:44,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=900861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:19:58,017 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:20:45,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.117e+02 1.257e+03 1.700e+03 2.363e+03 6.112e+03, threshold=3.399e+03, percent-clipped=6.0
+2023-03-10 14:20:48,767 INFO [train.py:968] (0/2) Epoch 20, batch 33750, giga_loss[loss=0.2159, simple_loss=0.304, pruned_loss=0.06389, over 28770.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3357, pruned_loss=0.0892, over 5670226.28 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3462, pruned_loss=0.1106, over 5697123.56 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3349, pruned_loss=0.08725, over 5667643.17 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:21:52,454 INFO [train.py:968] (0/2) Epoch 20, batch 33800, giga_loss[loss=0.2387, simple_loss=0.3228, pruned_loss=0.07727, over 28534.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3349, pruned_loss=0.08907, over 5673389.04 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3456, pruned_loss=0.1103, over 5698462.77 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3344, pruned_loss=0.0872, over 5669505.68 frames. ], batch size: 370, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:22:10,442 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4830, 1.8022, 1.4348, 1.6496], device='cuda:0'), covar=tensor([0.2858, 0.2637, 0.3003, 0.2325], device='cuda:0'), in_proj_covar=tensor([0.1480, 0.1069, 0.1312, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 14:22:13,157 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:22:55,704 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:22:58,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:22:59,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.197e+02 1.525e+03 1.869e+03 2.624e+03 7.024e+03, threshold=3.739e+03, percent-clipped=14.0
+2023-03-10 14:23:04,194 INFO [train.py:968] (0/2) Epoch 20, batch 33850, giga_loss[loss=0.2649, simple_loss=0.3442, pruned_loss=0.09273, over 28879.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.334, pruned_loss=0.08909, over 5674654.03 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3456, pruned_loss=0.1104, over 5697079.48 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3335, pruned_loss=0.08741, over 5672522.85 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:23:07,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:23:40,289 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:23:47,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4115, 1.7500, 1.3691, 1.3626], device='cuda:0'), covar=tensor([0.2572, 0.2408, 0.2863, 0.2209], device='cuda:0'), in_proj_covar=tensor([0.1476, 0.1066, 0.1309, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 14:23:49,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:23:59,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5316, 5.3330, 5.1300, 3.0766], device='cuda:0'), covar=tensor([0.0486, 0.0741, 0.0810, 0.1411], device='cuda:0'), in_proj_covar=tensor([0.1192, 0.1097, 0.0933, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 14:24:03,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3488, 1.5230, 1.2585, 1.5682], device='cuda:0'), covar=tensor([0.0708, 0.0373, 0.0352, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0117, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-10 14:24:10,271 INFO [train.py:968] (0/2) Epoch 20, batch 33900, giga_loss[loss=0.2022, simple_loss=0.2891, pruned_loss=0.05767, over 29023.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3318, pruned_loss=0.08846, over 5665615.81 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3459, pruned_loss=0.1105, over 5684920.88 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.331, pruned_loss=0.08672, over 5675225.86 frames. ], batch size: 128, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 14:24:11,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:24:56,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1441, 1.4223, 1.4036, 1.0312], device='cuda:0'), covar=tensor([0.1659, 0.2465, 0.1379, 0.1660], device='cuda:0'), in_proj_covar=tensor([0.0887, 0.0689, 0.0933, 0.0834], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 14:25:08,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.076e+02 1.500e+03 1.895e+03 2.708e+03 9.129e+03, threshold=3.789e+03, percent-clipped=15.0
+2023-03-10 14:25:10,513 INFO [train.py:968] (0/2) Epoch 20, batch 33950, giga_loss[loss=0.25, simple_loss=0.338, pruned_loss=0.08093, over 28515.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3327, pruned_loss=0.08852, over 5678792.38 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3465, pruned_loss=0.1112, over 5691385.29 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3311, pruned_loss=0.08592, over 5680168.25 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 14:25:14,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901113.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 14:25:26,344 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:25:29,753 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:04,548 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:08,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:09,083 INFO [train.py:968] (0/2) Epoch 20, batch 34000, libri_loss[loss=0.2698, simple_loss=0.3349, pruned_loss=0.1023, over 26120.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3307, pruned_loss=0.08719, over 5669101.54 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3463, pruned_loss=0.1111, over 5691043.39 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.329, pruned_loss=0.0842, over 5669928.20 frames. ], batch size: 137, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:26:39,768 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:41,030 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:42,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:55,457 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=901201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:27:01,101 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:27:05,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:27:06,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.335e+02 1.417e+03 1.638e+03 2.001e+03 4.478e+03, threshold=3.276e+03, percent-clipped=1.0
+2023-03-10 14:27:07,400 INFO [train.py:968] (0/2) Epoch 20, batch 34050, giga_loss[loss=0.2274, simple_loss=0.3277, pruned_loss=0.06354, over 28840.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.331, pruned_loss=0.08535, over 5675492.44 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3457, pruned_loss=0.1107, over 5693850.17 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.33, pruned_loss=0.08285, over 5673317.02 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:27:15,093 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:27:36,207 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:27:48,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:27:59,183 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901256.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:28:03,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901259.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:28:04,779 INFO [train.py:968] (0/2) Epoch 20, batch 34100, giga_loss[loss=0.2532, simple_loss=0.3379, pruned_loss=0.08429, over 28937.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3335, pruned_loss=0.0863, over 5660898.71 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3457, pruned_loss=0.1109, over 5676871.03 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3322, pruned_loss=0.0833, over 5675131.89 frames. ], batch size: 227, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:28:09,521 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:12,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:12,087 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:16,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:34,123 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901288.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 14:28:47,932 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:52,762 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:29:04,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.592e+02 1.457e+03 1.810e+03 2.757e+03 7.910e+03, threshold=3.620e+03, percent-clipped=15.0
+2023-03-10 14:29:04,715 INFO [train.py:968] (0/2) Epoch 20, batch 34150, giga_loss[loss=0.228, simple_loss=0.3204, pruned_loss=0.06776, over 28688.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3335, pruned_loss=0.08582, over 5664738.42 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3457, pruned_loss=0.1109, over 5679562.15 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3323, pruned_loss=0.08302, over 5673701.36 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:30:02,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:30:15,605 INFO [train.py:968] (0/2) Epoch 20, batch 34200, giga_loss[loss=0.2831, simple_loss=0.3657, pruned_loss=0.1002, over 28468.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3327, pruned_loss=0.08561, over 5669283.24 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3448, pruned_loss=0.1104, over 5683399.07 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3323, pruned_loss=0.08326, over 5672612.70 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:30:44,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2280, 1.2031, 3.8117, 3.1062], device='cuda:0'), covar=tensor([0.1609, 0.2798, 0.0450, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.0743, 0.0637, 0.0937, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 14:30:57,542 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:31:00,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:31:12,435 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5058, 1.6580, 1.2080, 1.2086], device='cuda:0'), covar=tensor([0.0937, 0.0499, 0.0987, 0.1192], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0443, 0.0517, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 14:31:25,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.711e+02 1.463e+03 1.666e+03 2.509e+03 5.932e+03, threshold=3.332e+03, percent-clipped=9.0
+2023-03-10 14:31:25,314 INFO [train.py:968] (0/2) Epoch 20, batch 34250, giga_loss[loss=0.2538, simple_loss=0.3407, pruned_loss=0.08347, over 28685.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3329, pruned_loss=0.08601, over 5669269.04 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3449, pruned_loss=0.1105, over 5688712.54 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3321, pruned_loss=0.08327, over 5666459.15 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:31:37,964 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:32:32,985 INFO [train.py:968] (0/2) Epoch 20, batch 34300, giga_loss[loss=0.2622, simple_loss=0.343, pruned_loss=0.0907, over 28677.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3338, pruned_loss=0.08564, over 5677229.10 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.345, pruned_loss=0.1105, over 5693885.94 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3327, pruned_loss=0.08272, over 5669746.59 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:33:26,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:33:29,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:33:48,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.891e+02 1.425e+03 1.820e+03 2.665e+03 7.481e+03, threshold=3.641e+03, percent-clipped=14.0
+2023-03-10 14:33:48,972 INFO [train.py:968] (0/2) Epoch 20, batch 34350, giga_loss[loss=0.326, simple_loss=0.3827, pruned_loss=0.1346, over 26888.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3352, pruned_loss=0.08624, over 5670653.07 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3449, pruned_loss=0.1104, over 5693145.63 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3343, pruned_loss=0.08364, over 5665344.12 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:34:08,081 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:34:36,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2776, 1.6316, 1.2377, 1.3755], device='cuda:0'), covar=tensor([0.2749, 0.2801, 0.3382, 0.2293], device='cuda:0'), in_proj_covar=tensor([0.1477, 0.1069, 0.1313, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 14:34:51,372 INFO [train.py:968] (0/2) Epoch 20, batch 34400, giga_loss[loss=0.2413, simple_loss=0.3393, pruned_loss=0.07171, over 29035.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3385, pruned_loss=0.08766, over 5679773.66 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3446, pruned_loss=0.1102, over 5699148.87 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3379, pruned_loss=0.08515, over 5669645.74 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:35:18,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:35:39,255 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5491, 1.8505, 1.4644, 1.6999], device='cuda:0'), covar=tensor([0.2775, 0.2802, 0.3380, 0.2481], device='cuda:0'), in_proj_covar=tensor([0.1476, 0.1067, 0.1311, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 14:35:48,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9525, 2.0831, 1.4473, 1.6672], device='cuda:0'), covar=tensor([0.0804, 0.0514, 0.0957, 0.0981], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0442, 0.0516, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 14:35:50,027 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-10 14:35:50,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7717, 1.9269, 1.3207, 1.5729], device='cuda:0'), covar=tensor([0.0883, 0.0628, 0.1006, 0.1192], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0442, 0.0516, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 14:36:02,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.874e+02 1.572e+03 2.011e+03 2.912e+03 1.203e+04, threshold=4.022e+03, percent-clipped=14.0
+2023-03-10 14:36:02,503 INFO [train.py:968] (0/2) Epoch 20, batch 34450, giga_loss[loss=0.2666, simple_loss=0.3459, pruned_loss=0.09363, over 28180.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3384, pruned_loss=0.08777, over 5680522.26 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3445, pruned_loss=0.11, over 5703074.70 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3378, pruned_loss=0.08551, over 5668425.02 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:36:21,223 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5050, 1.8400, 1.2210, 1.4938], device='cuda:0'), covar=tensor([0.0845, 0.0483, 0.1017, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0441, 0.0515, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 14:36:35,568 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=901636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:37:08,850 INFO [train.py:968] (0/2) Epoch 20, batch 34500, libri_loss[loss=0.3202, simple_loss=0.3824, pruned_loss=0.129, over 29099.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.338, pruned_loss=0.08871, over 5681623.88 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3455, pruned_loss=0.1106, over 5700296.00 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3364, pruned_loss=0.08553, over 5672991.47 frames. ], batch size: 101, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:37:57,469 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=901692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:38:27,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.304e+02 1.374e+03 1.974e+03 2.789e+03 9.076e+03, threshold=3.948e+03, percent-clipped=14.0
+2023-03-10 14:38:27,719 INFO [train.py:968] (0/2) Epoch 20, batch 34550, giga_loss[loss=0.241, simple_loss=0.3355, pruned_loss=0.07329, over 28177.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3349, pruned_loss=0.08608, over 5683287.51 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3455, pruned_loss=0.1107, over 5702384.69 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3336, pruned_loss=0.08336, over 5674673.83 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:38:29,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6863, 1.8965, 1.1938, 1.5057], device='cuda:0'), covar=tensor([0.1041, 0.0687, 0.1113, 0.1281], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0440, 0.0512, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 14:38:41,671 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:38:45,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:39:16,753 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:39:28,144 INFO [train.py:968] (0/2) Epoch 20, batch 34600, giga_loss[loss=0.2641, simple_loss=0.3533, pruned_loss=0.08743, over 28076.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3339, pruned_loss=0.08565, over 5695094.61 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3456, pruned_loss=0.1108, over 5712308.08 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3321, pruned_loss=0.08199, over 5678312.16 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:40:10,761 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-10 14:40:35,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.655e+02 1.371e+03 1.921e+03 2.536e+03 4.984e+03, threshold=3.841e+03, percent-clipped=6.0
+2023-03-10 14:40:35,576 INFO [train.py:968] (0/2) Epoch 20, batch 34650, giga_loss[loss=0.253, simple_loss=0.3344, pruned_loss=0.08583, over 28664.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3351, pruned_loss=0.08676, over 5681706.84 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3455, pruned_loss=0.1108, over 5711384.18 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3335, pruned_loss=0.0834, over 5668763.66 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:40:54,159 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-10 14:41:37,376 INFO [train.py:968] (0/2) Epoch 20, batch 34700, giga_loss[loss=0.2524, simple_loss=0.336, pruned_loss=0.08436, over 27594.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3382, pruned_loss=0.08839, over 5683428.78 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3456, pruned_loss=0.1108, over 5714029.03 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3366, pruned_loss=0.0852, over 5669868.86 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:42:38,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.676e+02 1.478e+03 1.976e+03 2.687e+03 5.669e+03, threshold=3.952e+03, percent-clipped=9.0
+2023-03-10 14:42:38,056 INFO [train.py:968] (0/2) Epoch 20, batch 34750, giga_loss[loss=0.2494, simple_loss=0.3286, pruned_loss=0.08515, over 28687.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3376, pruned_loss=0.08857, over 5684227.78 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3454, pruned_loss=0.1107, over 5711895.71 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3365, pruned_loss=0.08597, over 5675324.79 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:43:02,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4415, 2.0652, 1.5043, 0.7222], device='cuda:0'), covar=tensor([0.6447, 0.3246, 0.4329, 0.6322], device='cuda:0'), in_proj_covar=tensor([0.1721, 0.1627, 0.1586, 0.1409], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 14:43:33,539 INFO [train.py:968] (0/2) Epoch 20, batch 34800, giga_loss[loss=0.2657, simple_loss=0.3314, pruned_loss=0.09996, over 26735.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3346, pruned_loss=0.08851, over 5677142.07 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.345, pruned_loss=0.1105, over 5716402.21 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3337, pruned_loss=0.08584, over 5664733.15 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 14:43:55,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2478, 1.4010, 1.1340, 1.0326], device='cuda:0'), covar=tensor([0.0957, 0.0438, 0.1037, 0.0943], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0441, 0.0514, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 14:44:18,949 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-902000.pt
+2023-03-10 14:44:32,693 INFO [train.py:968] (0/2) Epoch 20, batch 34850, giga_loss[loss=0.249, simple_loss=0.3312, pruned_loss=0.08338, over 28843.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3334, pruned_loss=0.08802, over 5677609.80 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.345, pruned_loss=0.1105, over 5718615.67 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3325, pruned_loss=0.08554, over 5665136.00 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:44:33,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:44:33,300 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.509e+03 2.206e+03 3.336e+03 8.776e+03, threshold=4.413e+03, percent-clipped=17.0
+2023-03-10 14:45:26,963 INFO [train.py:968] (0/2) Epoch 20, batch 34900, giga_loss[loss=0.3128, simple_loss=0.3921, pruned_loss=0.1168, over 29007.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.34, pruned_loss=0.09177, over 5680631.83 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3453, pruned_loss=0.1106, over 5722076.70 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3389, pruned_loss=0.08928, over 5667027.49 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:45:33,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:46:18,096 INFO [train.py:968] (0/2) Epoch 20, batch 34950, giga_loss[loss=0.3784, simple_loss=0.434, pruned_loss=0.1614, over 26743.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3489, pruned_loss=0.09638, over 5677106.71 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3453, pruned_loss=0.1106, over 5722076.70 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3481, pruned_loss=0.09445, over 5666518.29 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:46:18,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.932e+02 1.430e+03 1.761e+03 2.405e+03 7.505e+03, threshold=3.521e+03, percent-clipped=6.0
+2023-03-10 14:47:00,942 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=902154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:47:03,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=902157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:47:05,572 INFO [train.py:968] (0/2) Epoch 20, batch 35000, libri_loss[loss=0.266, simple_loss=0.3374, pruned_loss=0.09736, over 29516.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3525, pruned_loss=0.09895, over 5666746.47 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3456, pruned_loss=0.1108, over 5714800.33 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3516, pruned_loss=0.09712, over 5665030.61 frames. ], batch size: 80, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:47:05,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7846, 2.0256, 1.7572, 1.8816], device='cuda:0'), covar=tensor([0.0730, 0.0281, 0.0312, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 14:47:28,624 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=902186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:47:45,776 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 14:47:47,512 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=902210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:47:47,880 INFO [train.py:968] (0/2) Epoch 20, batch 35050, giga_loss[loss=0.2439, simple_loss=0.3138, pruned_loss=0.08694, over 29005.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3479, pruned_loss=0.09757, over 5668587.82 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3462, pruned_loss=0.1112, over 5704778.66 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3468, pruned_loss=0.09544, over 5675290.39 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:47:48,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.985e+02 1.193e+03 1.489e+03 1.888e+03 3.998e+03, threshold=2.978e+03, percent-clipped=1.0
+2023-03-10 14:47:50,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=902213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:48:15,451 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=902242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:48:33,502 INFO [train.py:968] (0/2) Epoch 20, batch 35100, giga_loss[loss=0.2648, simple_loss=0.3366, pruned_loss=0.09651, over 28268.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3404, pruned_loss=0.09425, over 5677843.78 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3462, pruned_loss=0.1112, over 5705801.77 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3395, pruned_loss=0.09258, over 5682062.56 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:48:59,594 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:49:11,424 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6487, 1.6525, 1.8231, 1.4705], device='cuda:0'), covar=tensor([0.1750, 0.2312, 0.1396, 0.1658], device='cuda:0'), in_proj_covar=tensor([0.0889, 0.0688, 0.0933, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 14:49:16,586 INFO [train.py:968] (0/2) Epoch 20, batch 35150, giga_loss[loss=0.2569, simple_loss=0.3234, pruned_loss=0.09525, over 28886.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3327, pruned_loss=0.09109, over 5680599.41 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.346, pruned_loss=0.111, over 5708922.15 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.332, pruned_loss=0.08966, over 5680844.00 frames. ], batch size: 112, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:49:18,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.074e+02 1.147e+03 1.547e+03 2.135e+03 7.428e+03, threshold=3.095e+03, percent-clipped=18.0
+2023-03-10 14:49:20,918 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.38 vs. limit=5.0
+2023-03-10 14:50:00,363 INFO [train.py:968] (0/2) Epoch 20, batch 35200, giga_loss[loss=0.2153, simple_loss=0.2867, pruned_loss=0.07193, over 27571.00 frames. ], tot_loss[loss=0.249, simple_loss=0.324, pruned_loss=0.08701, over 5686177.93 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.346, pruned_loss=0.1109, over 5710801.49 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3233, pruned_loss=0.08582, over 5684449.17 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:50:23,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902386.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:50:37,856 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:50:43,604 INFO [train.py:968] (0/2) Epoch 20, batch 35250, giga_loss[loss=0.209, simple_loss=0.2923, pruned_loss=0.06285, over 29062.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3197, pruned_loss=0.08555, over 5688019.56 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3463, pruned_loss=0.1108, over 5712581.37 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3181, pruned_loss=0.08393, over 5684572.98 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:50:45,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.619e+02 9.738e+02 1.248e+03 1.845e+03 8.680e+03, threshold=2.496e+03, percent-clipped=10.0
+2023-03-10 14:50:50,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4773, 2.9570, 2.4990, 2.0801], device='cuda:0'), covar=tensor([0.2397, 0.1599, 0.1880, 0.2113], device='cuda:0'), in_proj_covar=tensor([0.1938, 0.1830, 0.1766, 0.1914], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 14:50:51,693 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-10 14:51:26,265 INFO [train.py:968] (0/2) Epoch 20, batch 35300, giga_loss[loss=0.2185, simple_loss=0.2987, pruned_loss=0.06914, over 28943.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3147, pruned_loss=0.08296, over 5686971.98 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3463, pruned_loss=0.1107, over 5711118.47 frames. ], giga_tot_loss[loss=0.2382, simple_loss=0.3133, pruned_loss=0.08155, over 5685197.83 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:52:00,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1790, 1.2755, 1.1984, 0.8435], device='cuda:0'), covar=tensor([0.1045, 0.0560, 0.1121, 0.1168], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0441, 0.0514, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 14:52:11,590 INFO [train.py:968] (0/2) Epoch 20, batch 35350, libri_loss[loss=0.3017, simple_loss=0.3666, pruned_loss=0.1184, over 28864.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3126, pruned_loss=0.08221, over 5694686.38 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3471, pruned_loss=0.1112, over 5713024.83 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.3102, pruned_loss=0.08025, over 5691187.25 frames. ], batch size: 107, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:52:13,278 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:52:13,736 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.902e+02 9.981e+02 1.257e+03 1.613e+03 5.389e+03, threshold=2.513e+03, percent-clipped=6.0
+2023-03-10 14:52:53,833 INFO [train.py:968] (0/2) Epoch 20, batch 35400, libri_loss[loss=0.3035, simple_loss=0.3778, pruned_loss=0.1146, over 29658.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3104, pruned_loss=0.08156, over 5705517.13 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3475, pruned_loss=0.1112, over 5717959.30 frames. ], giga_tot_loss[loss=0.2329, simple_loss=0.3072, pruned_loss=0.07925, over 5697843.09 frames. ], batch size: 88, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:52:58,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4400, 1.5491, 1.3303, 1.6253], device='cuda:0'), covar=tensor([0.0794, 0.0351, 0.0352, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0117, 0.0118, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 14:53:16,904 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:53:40,145 INFO [train.py:968] (0/2) Epoch 20, batch 35450, giga_loss[loss=0.2139, simple_loss=0.2892, pruned_loss=0.06932, over 28675.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3064, pruned_loss=0.07946, over 5706575.06 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3473, pruned_loss=0.111, over 5719302.17 frames. ], giga_tot_loss[loss=0.229, simple_loss=0.3034, pruned_loss=0.07726, over 5699017.78 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:53:43,208 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.394e+02 1.101e+03 1.379e+03 1.993e+03 7.030e+03, threshold=2.757e+03, percent-clipped=14.0
+2023-03-10 14:54:26,604 INFO [train.py:968] (0/2) Epoch 20, batch 35500, giga_loss[loss=0.2199, simple_loss=0.2871, pruned_loss=0.07635, over 28749.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.302, pruned_loss=0.07722, over 5704723.50 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3474, pruned_loss=0.111, over 5720226.66 frames. ], giga_tot_loss[loss=0.225, simple_loss=0.2994, pruned_loss=0.07533, over 5697966.31 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:54:29,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:54:54,189 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:54:55,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6859, 1.9952, 1.8224, 1.7543], device='cuda:0'), covar=tensor([0.2055, 0.2069, 0.2421, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0738, 0.0702, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 14:55:06,885 INFO [train.py:968] (0/2) Epoch 20, batch 35550, giga_loss[loss=0.1932, simple_loss=0.2718, pruned_loss=0.05727, over 29071.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.2994, pruned_loss=0.07614, over 5703216.66 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3472, pruned_loss=0.1109, over 5721189.30 frames. ], giga_tot_loss[loss=0.2221, simple_loss=0.2964, pruned_loss=0.07392, over 5696525.28 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:55:09,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.982e+02 1.002e+03 1.317e+03 1.908e+03 1.513e+04, threshold=2.635e+03, percent-clipped=7.0
+2023-03-10 14:55:44,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1360, 3.1160, 2.0371, 1.2097], device='cuda:0'), covar=tensor([0.8039, 0.3120, 0.4453, 0.7075], device='cuda:0'), in_proj_covar=tensor([0.1727, 0.1637, 0.1589, 0.1410], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 14:55:51,657 INFO [train.py:968] (0/2) Epoch 20, batch 35600, giga_loss[loss=0.1964, simple_loss=0.2772, pruned_loss=0.05785, over 28976.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.2971, pruned_loss=0.07528, over 5697884.31 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3476, pruned_loss=0.111, over 5725779.69 frames. ], giga_tot_loss[loss=0.2193, simple_loss=0.2933, pruned_loss=0.07264, over 5688162.53 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:55:51,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902761.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:55:57,989 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:56:05,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:56:32,006 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=902808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:56:33,879 INFO [train.py:968] (0/2) Epoch 20, batch 35650, giga_loss[loss=0.2409, simple_loss=0.3086, pruned_loss=0.08664, over 28902.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2947, pruned_loss=0.07426, over 5697116.36 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3475, pruned_loss=0.1109, over 5719864.29 frames. ], giga_tot_loss[loss=0.2163, simple_loss=0.2901, pruned_loss=0.07118, over 5693666.09 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:56:34,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=902811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:56:37,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.115e+02 1.083e+03 1.400e+03 2.147e+03 1.691e+04, threshold=2.800e+03, percent-clipped=17.0
+2023-03-10 14:56:59,001 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=902840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:57:07,462 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-10 14:57:17,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3505, 3.0284, 1.5490, 1.4936], device='cuda:0'), covar=tensor([0.0977, 0.0352, 0.0903, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0546, 0.0378, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 14:57:21,892 INFO [train.py:968] (0/2) Epoch 20, batch 35700, giga_loss[loss=0.2828, simple_loss=0.3577, pruned_loss=0.1039, over 29121.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.2994, pruned_loss=0.07718, over 5694089.02 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3476, pruned_loss=0.1108, over 5724167.60 frames. ], giga_tot_loss[loss=0.2214, simple_loss=0.2946, pruned_loss=0.07407, over 5687361.46 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:57:27,680 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4689, 4.1796, 1.6825, 1.7516], device='cuda:0'), covar=tensor([0.1044, 0.0286, 0.0948, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0546, 0.0379, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 14:57:47,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:58:01,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=902904.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:58:05,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=902907.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 14:58:09,700 INFO [train.py:968] (0/2) Epoch 20, batch 35750, giga_loss[loss=0.2793, simple_loss=0.3353, pruned_loss=0.1117, over 23910.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3116, pruned_loss=0.08321, over 5690835.30 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3478, pruned_loss=0.1108, over 5725304.92 frames. ], giga_tot_loss[loss=0.2338, simple_loss=0.307, pruned_loss=0.0803, over 5684052.53 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:58:12,562 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.235e+02 1.215e+03 1.746e+03 2.631e+03 8.166e+03, threshold=3.492e+03, percent-clipped=22.0
+2023-03-10 14:58:18,436 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=902920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:58:22,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=902923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:58:34,700 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=902936.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:58:50,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=902952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:58:56,401 INFO [train.py:968] (0/2) Epoch 20, batch 35800, giga_loss[loss=0.307, simple_loss=0.3787, pruned_loss=0.1176, over 28988.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3254, pruned_loss=0.09053, over 5695695.16 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3478, pruned_loss=0.1108, over 5726431.19 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3214, pruned_loss=0.08792, over 5688919.22 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:58:56,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:59:41,560 INFO [train.py:968] (0/2) Epoch 20, batch 35850, giga_loss[loss=0.2519, simple_loss=0.3241, pruned_loss=0.0899, over 28604.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3335, pruned_loss=0.09398, over 5694818.40 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3476, pruned_loss=0.1105, over 5728997.52 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3303, pruned_loss=0.09192, over 5686732.90 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:59:44,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.519e+02 1.306e+03 1.659e+03 2.055e+03 5.958e+03, threshold=3.318e+03, percent-clipped=4.0
+2023-03-10 14:59:59,496 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=903031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:00:01,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=903034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:00:16,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4766, 3.4351, 1.6024, 1.6396], device='cuda:0'), covar=tensor([0.0988, 0.0309, 0.0915, 0.1325], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0544, 0.0377, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 15:00:25,932 INFO [train.py:968] (0/2) Epoch 20, batch 35900, libri_loss[loss=0.282, simple_loss=0.3565, pruned_loss=0.1038, over 26097.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3369, pruned_loss=0.09417, over 5688056.00 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3473, pruned_loss=0.1102, over 5729748.61 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3343, pruned_loss=0.09245, over 5680403.46 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:00:27,481 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=903063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:00:29,845 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-10 15:00:31,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=903069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:00:50,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7019, 2.0990, 1.7208, 1.7116], device='cuda:0'), covar=tensor([0.0766, 0.0273, 0.0309, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 15:00:59,898 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 15:01:01,669 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=903104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:01:04,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=903107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:01:07,333 INFO [train.py:968] (0/2) Epoch 20, batch 35950, giga_loss[loss=0.2973, simple_loss=0.3671, pruned_loss=0.1137, over 26742.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3407, pruned_loss=0.09559, over 5676710.41 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3482, pruned_loss=0.1107, over 5709142.33 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3374, pruned_loss=0.09309, over 5686837.88 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:01:11,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.913e+02 1.231e+03 1.461e+03 1.940e+03 5.432e+03, threshold=2.923e+03, percent-clipped=6.0
+2023-03-10 15:01:35,614 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=903136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:01:42,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=903144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:01:55,299 INFO [train.py:968] (0/2) Epoch 20, batch 36000, giga_loss[loss=0.3572, simple_loss=0.3994, pruned_loss=0.1575, over 26494.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3433, pruned_loss=0.09687, over 5682790.79 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3483, pruned_loss=0.1106, over 5711811.52 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3405, pruned_loss=0.09474, over 5687938.10 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:01:55,304 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 15:01:59,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3491, 3.1084, 1.4240, 1.5312], device='cuda:0'), covar=tensor([0.1069, 0.0344, 0.1003, 0.1494], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0545, 0.0378, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 15:02:04,853 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2027, simple_loss=0.3093, pruned_loss=0.04808, over 944034.00 frames. 
+2023-03-10 15:02:04,854 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 15:02:07,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8613, 1.9608, 1.9525, 1.7234], device='cuda:0'), covar=tensor([0.2546, 0.2455, 0.2390, 0.2470], device='cuda:0'), in_proj_covar=tensor([0.1930, 0.1832, 0.1764, 0.1920], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 15:02:49,292 INFO [train.py:968] (0/2) Epoch 20, batch 36050, giga_loss[loss=0.3083, simple_loss=0.3665, pruned_loss=0.125, over 29071.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3463, pruned_loss=0.09928, over 5681774.57 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3487, pruned_loss=0.1106, over 5716349.93 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3436, pruned_loss=0.09725, over 5680843.95 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:02:50,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=903212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:02:51,282 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.048e+02 1.151e+03 1.436e+03 2.215e+03 5.362e+03, threshold=2.872e+03, percent-clipped=13.0
+2023-03-10 15:02:52,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=903215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:03:17,145 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=903244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:03:31,546 INFO [train.py:968] (0/2) Epoch 20, batch 36100, giga_loss[loss=0.2931, simple_loss=0.3629, pruned_loss=0.1117, over 28956.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3511, pruned_loss=0.1026, over 5669642.09 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3494, pruned_loss=0.1111, over 5701691.44 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3484, pruned_loss=0.1004, over 5680069.31 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:03:53,963 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=903287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:03:56,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=903290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:03:56,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4210, 1.8242, 1.4180, 1.3884], device='cuda:0'), covar=tensor([0.2744, 0.2606, 0.2947, 0.2254], device='cuda:0'), in_proj_covar=tensor([0.1482, 0.1076, 0.1313, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 15:04:14,577 INFO [train.py:968] (0/2) Epoch 20, batch 36150, giga_loss[loss=0.3084, simple_loss=0.3781, pruned_loss=0.1193, over 28764.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3544, pruned_loss=0.104, over 5671848.39 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3494, pruned_loss=0.1111, over 5694751.31 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3522, pruned_loss=0.1021, over 5686075.06 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:04:17,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.176e+02 1.201e+03 1.497e+03 2.253e+03 1.192e+04, threshold=2.995e+03, percent-clipped=10.0
+2023-03-10 15:04:22,047 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=903319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:04:25,946 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6270, 1.7374, 1.6187, 1.3915], device='cuda:0'), covar=tensor([0.2646, 0.2670, 0.2464, 0.2703], device='cuda:0'), in_proj_covar=tensor([0.1929, 0.1830, 0.1768, 0.1921], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 15:04:56,282 INFO [train.py:968] (0/2) Epoch 20, batch 36200, giga_loss[loss=0.2899, simple_loss=0.3683, pruned_loss=0.1057, over 28306.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3556, pruned_loss=0.1038, over 5662283.39 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3495, pruned_loss=0.111, over 5684204.31 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3539, pruned_loss=0.1024, over 5682349.41 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:05:40,626 INFO [train.py:968] (0/2) Epoch 20, batch 36250, giga_loss[loss=0.2674, simple_loss=0.3554, pruned_loss=0.0897, over 28783.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3573, pruned_loss=0.104, over 5657332.38 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.35, pruned_loss=0.1112, over 5676038.58 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3556, pruned_loss=0.1025, over 5680084.08 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:05:43,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.641e+02 1.175e+03 1.417e+03 1.902e+03 5.290e+03, threshold=2.834e+03, percent-clipped=6.0
+2023-03-10 15:06:21,672 INFO [train.py:968] (0/2) Epoch 20, batch 36300, giga_loss[loss=0.2391, simple_loss=0.3335, pruned_loss=0.07233, over 28605.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.357, pruned_loss=0.102, over 5671825.27 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3504, pruned_loss=0.1113, over 5677029.55 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3554, pruned_loss=0.1007, over 5688739.84 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:06:43,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8382, 4.6394, 4.4329, 2.0972], device='cuda:0'), covar=tensor([0.0417, 0.0593, 0.0609, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.1170, 0.1083, 0.0920, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 15:07:03,532 INFO [train.py:968] (0/2) Epoch 20, batch 36350, giga_loss[loss=0.2453, simple_loss=0.3343, pruned_loss=0.07813, over 28973.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.356, pruned_loss=0.1006, over 5680932.60 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.351, pruned_loss=0.1115, over 5679934.96 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3543, pruned_loss=0.09916, over 5691599.63 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:07:06,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.919e+02 1.124e+03 1.311e+03 1.712e+03 3.794e+03, threshold=2.621e+03, percent-clipped=3.0
+2023-03-10 15:07:42,996 INFO [train.py:968] (0/2) Epoch 20, batch 36400, giga_loss[loss=0.2318, simple_loss=0.3223, pruned_loss=0.07063, over 28525.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.354, pruned_loss=0.09864, over 5697692.85 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.351, pruned_loss=0.1113, over 5686581.57 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3527, pruned_loss=0.09729, over 5700590.25 frames. ], batch size: 71, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:08:00,724 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-10 15:08:26,568 INFO [train.py:968] (0/2) Epoch 20, batch 36450, giga_loss[loss=0.4435, simple_loss=0.4482, pruned_loss=0.2195, over 26548.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3557, pruned_loss=0.1009, over 5697168.86 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3517, pruned_loss=0.1117, over 5689763.45 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3542, pruned_loss=0.09927, over 5696832.25 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:08:29,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.850e+02 1.196e+03 1.498e+03 2.127e+03 7.204e+03, threshold=2.996e+03, percent-clipped=12.0
+2023-03-10 15:08:44,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8972, 1.2107, 2.9020, 2.8277], device='cuda:0'), covar=tensor([0.1652, 0.2506, 0.0636, 0.1045], device='cuda:0'), in_proj_covar=tensor([0.0745, 0.0641, 0.0940, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 15:08:54,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5920, 1.7974, 1.4735, 1.6238], device='cuda:0'), covar=tensor([0.2605, 0.2560, 0.2852, 0.2232], device='cuda:0'), in_proj_covar=tensor([0.1483, 0.1077, 0.1313, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 15:09:13,065 INFO [train.py:968] (0/2) Epoch 20, batch 36500, giga_loss[loss=0.3681, simple_loss=0.4128, pruned_loss=0.1617, over 27947.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3581, pruned_loss=0.1052, over 5685472.14 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3517, pruned_loss=0.1117, over 5680336.27 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.357, pruned_loss=0.1038, over 5693278.90 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:09:18,505 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 15:09:52,630 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-10 15:09:54,103 INFO [train.py:968] (0/2) Epoch 20, batch 36550, giga_loss[loss=0.3254, simple_loss=0.3912, pruned_loss=0.1298, over 28864.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3585, pruned_loss=0.1072, over 5678249.33 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.352, pruned_loss=0.1118, over 5666155.27 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3574, pruned_loss=0.1058, over 5696584.68 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:09:59,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.360e+02 1.293e+03 1.618e+03 2.064e+03 4.558e+03, threshold=3.235e+03, percent-clipped=4.0
+2023-03-10 15:10:16,878 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-10 15:10:42,032 INFO [train.py:968] (0/2) Epoch 20, batch 36600, giga_loss[loss=0.2842, simple_loss=0.3514, pruned_loss=0.1085, over 28672.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3563, pruned_loss=0.1064, over 5681058.85 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3522, pruned_loss=0.1119, over 5660583.95 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3554, pruned_loss=0.1052, over 5701457.83 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:11:24,039 INFO [train.py:968] (0/2) Epoch 20, batch 36650, giga_loss[loss=0.2646, simple_loss=0.3436, pruned_loss=0.09277, over 28894.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3539, pruned_loss=0.1058, over 5677292.75 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.352, pruned_loss=0.1117, over 5656728.41 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3535, pruned_loss=0.1049, over 5696921.73 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:11:27,604 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.391e+02 1.378e+03 1.767e+03 2.370e+03 1.031e+04, threshold=3.533e+03, percent-clipped=11.0
+2023-03-10 15:11:52,325 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=903843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:12:07,268 INFO [train.py:968] (0/2) Epoch 20, batch 36700, giga_loss[loss=0.2801, simple_loss=0.3522, pruned_loss=0.104, over 28943.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3537, pruned_loss=0.1052, over 5688401.06 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3518, pruned_loss=0.1114, over 5663944.12 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3536, pruned_loss=0.1046, over 5698358.13 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:12:51,291 INFO [train.py:968] (0/2) Epoch 20, batch 36750, giga_loss[loss=0.2459, simple_loss=0.3309, pruned_loss=0.08049, over 28862.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3513, pruned_loss=0.103, over 5679749.24 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3518, pruned_loss=0.1113, over 5660936.12 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3512, pruned_loss=0.1024, over 5690608.18 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:12:58,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.306e+02 1.271e+03 1.603e+03 2.194e+03 5.249e+03, threshold=3.206e+03, percent-clipped=4.0
+2023-03-10 15:13:37,223 INFO [train.py:968] (0/2) Epoch 20, batch 36800, giga_loss[loss=0.2489, simple_loss=0.3233, pruned_loss=0.08726, over 28544.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3484, pruned_loss=0.1009, over 5681111.24 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3525, pruned_loss=0.1117, over 5668203.51 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3477, pruned_loss=0.09989, over 5683795.45 frames. ], batch size: 336, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:14:12,526 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-904000.pt
+2023-03-10 15:14:24,569 INFO [train.py:968] (0/2) Epoch 20, batch 36850, giga_loss[loss=0.2182, simple_loss=0.3122, pruned_loss=0.06204, over 28905.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3423, pruned_loss=0.09784, over 5669740.97 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3528, pruned_loss=0.1119, over 5670883.62 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3414, pruned_loss=0.09651, over 5669549.76 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:14:28,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.206e+02 1.030e+03 1.338e+03 1.904e+03 6.455e+03, threshold=2.677e+03, percent-clipped=7.0
+2023-03-10 15:14:42,358 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-10 15:15:19,013 INFO [train.py:968] (0/2) Epoch 20, batch 36900, giga_loss[loss=0.2795, simple_loss=0.331, pruned_loss=0.114, over 26545.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3366, pruned_loss=0.09518, over 5660187.46 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3529, pruned_loss=0.1119, over 5674261.82 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3356, pruned_loss=0.09391, over 5657071.61 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:15:21,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5790, 1.5801, 1.8120, 1.4157], device='cuda:0'), covar=tensor([0.1648, 0.2322, 0.1333, 0.1576], device='cuda:0'), in_proj_covar=tensor([0.0891, 0.0694, 0.0936, 0.0836], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 15:15:47,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5647, 1.7805, 1.7169, 1.4954], device='cuda:0'), covar=tensor([0.2496, 0.2412, 0.2498, 0.2733], device='cuda:0'), in_proj_covar=tensor([0.1934, 0.1833, 0.1774, 0.1921], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 15:16:09,746 INFO [train.py:968] (0/2) Epoch 20, batch 36950, giga_loss[loss=0.2649, simple_loss=0.3384, pruned_loss=0.09567, over 28916.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3341, pruned_loss=0.09404, over 5651434.86 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3532, pruned_loss=0.1119, over 5669700.58 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3325, pruned_loss=0.09248, over 5652559.56 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:16:13,995 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.884e+02 1.110e+03 1.348e+03 2.190e+03 9.916e+03, threshold=2.696e+03, percent-clipped=17.0
+2023-03-10 15:16:57,028 INFO [train.py:968] (0/2) Epoch 20, batch 37000, giga_loss[loss=0.2556, simple_loss=0.3367, pruned_loss=0.08728, over 28619.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3344, pruned_loss=0.09321, over 5665949.34 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3533, pruned_loss=0.1119, over 5673300.58 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3327, pruned_loss=0.09171, over 5663397.80 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:17:38,389 INFO [train.py:968] (0/2) Epoch 20, batch 37050, giga_loss[loss=0.2509, simple_loss=0.3192, pruned_loss=0.09136, over 28723.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3345, pruned_loss=0.09311, over 5669814.97 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.354, pruned_loss=0.1124, over 5671680.75 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3319, pruned_loss=0.09093, over 5668910.24 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:17:43,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.091e+02 1.028e+03 1.348e+03 1.637e+03 4.030e+03, threshold=2.695e+03, percent-clipped=8.0
+2023-03-10 15:17:44,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=904218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:18:05,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1756, 0.9104, 1.0315, 1.3720], device='cuda:0'), covar=tensor([0.0801, 0.0384, 0.0358, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 15:18:11,893 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-10 15:18:13,167 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-10 15:18:16,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2651, 1.3989, 1.2873, 1.2469], device='cuda:0'), covar=tensor([0.2312, 0.2194, 0.1670, 0.2145], device='cuda:0'), in_proj_covar=tensor([0.1936, 0.1834, 0.1779, 0.1925], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 15:18:19,645 INFO [train.py:968] (0/2) Epoch 20, batch 37100, giga_loss[loss=0.2652, simple_loss=0.3311, pruned_loss=0.0997, over 28990.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3329, pruned_loss=0.09225, over 5687789.39 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3539, pruned_loss=0.1121, over 5675786.84 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3306, pruned_loss=0.09042, over 5683482.08 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:18:22,916 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 15:18:29,943 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=904271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:18:32,752 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5978, 4.5969, 1.8765, 1.8903], device='cuda:0'), covar=tensor([0.1016, 0.0258, 0.0863, 0.1247], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0547, 0.0378, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 15:18:43,393 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-10 15:19:01,317 INFO [train.py:968] (0/2) Epoch 20, batch 37150, giga_loss[loss=0.213, simple_loss=0.2894, pruned_loss=0.06827, over 28593.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3308, pruned_loss=0.09129, over 5706025.40 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3541, pruned_loss=0.112, over 5680411.06 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3283, pruned_loss=0.0895, over 5698866.03 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:19:06,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.416e+02 1.107e+03 1.417e+03 1.806e+03 5.408e+03, threshold=2.833e+03, percent-clipped=9.0
+2023-03-10 15:19:14,375 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=904326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:19:17,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5035, 3.3370, 3.1518, 1.8687], device='cuda:0'), covar=tensor([0.0678, 0.0793, 0.0694, 0.1918], device='cuda:0'), in_proj_covar=tensor([0.1180, 0.1092, 0.0929, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 15:19:30,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4752, 1.6655, 1.6771, 1.2824], device='cuda:0'), covar=tensor([0.1608, 0.2641, 0.1467, 0.1705], device='cuda:0'), in_proj_covar=tensor([0.0891, 0.0693, 0.0935, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 15:19:44,115 INFO [train.py:968] (0/2) Epoch 20, batch 37200, giga_loss[loss=0.2189, simple_loss=0.2991, pruned_loss=0.06929, over 28950.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3286, pruned_loss=0.0904, over 5709242.79 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3546, pruned_loss=0.1121, over 5686419.06 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3255, pruned_loss=0.08833, over 5698897.12 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:19:44,349 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=904361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:19:46,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=904364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:20:03,313 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2502, 1.5839, 1.2579, 0.9191], device='cuda:0'), covar=tensor([0.2457, 0.2540, 0.2808, 0.2465], device='cuda:0'), in_proj_covar=tensor([0.1483, 0.1075, 0.1312, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 15:20:07,959 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=904393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:20:22,624 INFO [train.py:968] (0/2) Epoch 20, batch 37250, giga_loss[loss=0.2399, simple_loss=0.3165, pruned_loss=0.08161, over 28923.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3265, pruned_loss=0.0892, over 5706424.57 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3551, pruned_loss=0.1121, over 5684872.14 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3229, pruned_loss=0.08697, over 5700068.30 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:20:26,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.220e+02 1.051e+03 1.286e+03 2.011e+03 9.805e+03, threshold=2.571e+03, percent-clipped=11.0
+2023-03-10 15:21:02,470 INFO [train.py:968] (0/2) Epoch 20, batch 37300, giga_loss[loss=0.2248, simple_loss=0.2994, pruned_loss=0.07506, over 28760.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3259, pruned_loss=0.08902, over 5711357.93 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3559, pruned_loss=0.1122, over 5682886.94 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3214, pruned_loss=0.08637, over 5709256.62 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:21:42,513 INFO [train.py:968] (0/2) Epoch 20, batch 37350, giga_loss[loss=0.2117, simple_loss=0.2937, pruned_loss=0.06479, over 28513.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.324, pruned_loss=0.08809, over 5717087.59 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3561, pruned_loss=0.1122, over 5687198.71 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3196, pruned_loss=0.08553, over 5712251.19 frames. ], batch size: 78, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:21:46,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.113e+02 1.011e+03 1.293e+03 1.924e+03 6.008e+03, threshold=2.586e+03, percent-clipped=8.0
+2023-03-10 15:22:05,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4448, 4.2671, 4.0324, 1.9611], device='cuda:0'), covar=tensor([0.0561, 0.0701, 0.0675, 0.2073], device='cuda:0'), in_proj_covar=tensor([0.1187, 0.1100, 0.0934, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 15:22:21,670 INFO [train.py:968] (0/2) Epoch 20, batch 37400, libri_loss[loss=0.3406, simple_loss=0.4163, pruned_loss=0.1325, over 29613.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3217, pruned_loss=0.08701, over 5721082.27 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3564, pruned_loss=0.1121, over 5693066.10 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3171, pruned_loss=0.08448, over 5712477.99 frames. ], batch size: 91, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:22:35,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0006, 1.2308, 1.1742, 0.9782], device='cuda:0'), covar=tensor([0.1939, 0.2016, 0.1271, 0.1665], device='cuda:0'), in_proj_covar=tensor([0.1943, 0.1839, 0.1785, 0.1931], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 15:23:02,261 INFO [train.py:968] (0/2) Epoch 20, batch 37450, giga_loss[loss=0.2243, simple_loss=0.3021, pruned_loss=0.07328, over 29077.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3202, pruned_loss=0.08604, over 5719494.44 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3566, pruned_loss=0.1121, over 5694641.74 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.316, pruned_loss=0.08371, over 5711705.32 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:23:07,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.989e+02 9.313e+02 1.127e+03 1.418e+03 5.723e+03, threshold=2.254e+03, percent-clipped=5.0
+2023-03-10 15:23:11,275 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 15:23:30,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=904646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:23:43,608 INFO [train.py:968] (0/2) Epoch 20, batch 37500, giga_loss[loss=0.3173, simple_loss=0.3682, pruned_loss=0.1332, over 26595.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3205, pruned_loss=0.08644, over 5715616.96 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.357, pruned_loss=0.1122, over 5699183.41 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3156, pruned_loss=0.08372, over 5705863.91 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:24:18,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=904701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:24:21,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6788, 1.7204, 1.8862, 1.4551], device='cuda:0'), covar=tensor([0.1798, 0.2475, 0.1454, 0.1681], device='cuda:0'), in_proj_covar=tensor([0.0897, 0.0697, 0.0942, 0.0841], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 15:24:25,551 INFO [train.py:968] (0/2) Epoch 20, batch 37550, giga_loss[loss=0.2457, simple_loss=0.3162, pruned_loss=0.08763, over 28670.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3217, pruned_loss=0.08687, over 5725174.05 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3576, pruned_loss=0.1124, over 5702123.84 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3168, pruned_loss=0.08422, over 5715042.19 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:24:29,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3722, 1.6203, 1.3344, 1.0456], device='cuda:0'), covar=tensor([0.2650, 0.2705, 0.3082, 0.2367], device='cuda:0'), in_proj_covar=tensor([0.1485, 0.1076, 0.1314, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 15:24:31,679 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.694e+02 1.098e+03 1.500e+03 2.047e+03 7.033e+03, threshold=3.000e+03, percent-clipped=21.0
+2023-03-10 15:25:08,419 INFO [train.py:968] (0/2) Epoch 20, batch 37600, giga_loss[loss=0.2565, simple_loss=0.3338, pruned_loss=0.08961, over 28794.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3274, pruned_loss=0.0904, over 5706716.57 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3582, pruned_loss=0.1126, over 5691033.94 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3221, pruned_loss=0.08747, over 5709385.07 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:25:32,419 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=904789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:25:34,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=904792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:25:50,975 INFO [train.py:968] (0/2) Epoch 20, batch 37650, libri_loss[loss=0.2581, simple_loss=0.3263, pruned_loss=0.09494, over 28103.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.335, pruned_loss=0.09521, over 5710111.80 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3593, pruned_loss=0.1131, over 5694704.89 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3285, pruned_loss=0.09152, over 5709629.82 frames. ], batch size: 62, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:25:59,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.063e+02 1.331e+03 1.743e+03 2.648e+03 1.046e+04, threshold=3.486e+03, percent-clipped=18.0
+2023-03-10 15:26:01,771 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=904821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:26:26,739 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=904844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:26:28,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=904847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:26:41,365 INFO [train.py:968] (0/2) Epoch 20, batch 37700, libri_loss[loss=0.2835, simple_loss=0.3372, pruned_loss=0.1148, over 29364.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3438, pruned_loss=0.1014, over 5698604.98 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3594, pruned_loss=0.1132, over 5697847.94 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3384, pruned_loss=0.09814, over 5695279.98 frames. ], batch size: 67, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:26:57,146 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=904876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:27:29,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=904908.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 15:27:32,156 INFO [train.py:968] (0/2) Epoch 20, batch 37750, giga_loss[loss=0.3024, simple_loss=0.3593, pruned_loss=0.1227, over 23178.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3481, pruned_loss=0.1033, over 5687204.80 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3593, pruned_loss=0.1132, over 5700889.91 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3438, pruned_loss=0.1006, over 5681715.41 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:27:38,957 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.030e+02 1.257e+03 1.582e+03 2.078e+03 4.048e+03, threshold=3.164e+03, percent-clipped=2.0
+2023-03-10 15:27:44,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5336, 3.5496, 1.6819, 1.5874], device='cuda:0'), covar=tensor([0.1009, 0.0291, 0.0899, 0.1386], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0549, 0.0379, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 15:27:49,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0533, 2.3708, 2.1772, 2.1960], device='cuda:0'), covar=tensor([0.1455, 0.1229, 0.1414, 0.1258], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0746, 0.0710, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 15:28:13,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1269, 2.9481, 2.7835, 1.5482], device='cuda:0'), covar=tensor([0.1004, 0.1089, 0.0987, 0.2565], device='cuda:0'), in_proj_covar=tensor([0.1196, 0.1104, 0.0939, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 15:28:18,278 INFO [train.py:968] (0/2) Epoch 20, batch 37800, giga_loss[loss=0.284, simple_loss=0.365, pruned_loss=0.1015, over 28925.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3533, pruned_loss=0.1055, over 5679333.25 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3596, pruned_loss=0.1134, over 5693643.26 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3495, pruned_loss=0.1031, over 5681133.38 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:28:46,958 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 15:28:47,039 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 15:29:01,746 INFO [train.py:968] (0/2) Epoch 20, batch 37850, giga_loss[loss=0.2634, simple_loss=0.3488, pruned_loss=0.08898, over 28329.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3583, pruned_loss=0.1082, over 5668826.63 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.36, pruned_loss=0.1137, over 5680380.97 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3547, pruned_loss=0.1058, over 5681534.66 frames. ], batch size: 65, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:29:09,049 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.586e+02 1.284e+03 1.609e+03 2.055e+03 7.072e+03, threshold=3.218e+03, percent-clipped=10.0
+2023-03-10 15:29:28,842 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:29:45,879 INFO [train.py:968] (0/2) Epoch 20, batch 37900, giga_loss[loss=0.2834, simple_loss=0.3319, pruned_loss=0.1175, over 23901.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3563, pruned_loss=0.1065, over 5672683.81 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.36, pruned_loss=0.1138, over 5682700.35 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3535, pruned_loss=0.1045, over 5680586.07 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:29:51,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9120, 1.1487, 1.0111, 0.8089], device='cuda:0'), covar=tensor([0.2415, 0.2547, 0.1668, 0.2375], device='cuda:0'), in_proj_covar=tensor([0.1950, 0.1847, 0.1792, 0.1941], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 15:30:25,595 INFO [train.py:968] (0/2) Epoch 20, batch 37950, giga_loss[loss=0.264, simple_loss=0.3476, pruned_loss=0.09024, over 28917.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3528, pruned_loss=0.1037, over 5682447.60 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.36, pruned_loss=0.114, over 5678743.80 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3503, pruned_loss=0.1017, over 5691466.51 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:30:35,493 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.625e+02 1.268e+03 1.612e+03 2.027e+03 7.304e+03, threshold=3.224e+03, percent-clipped=9.0
+2023-03-10 15:31:09,010 INFO [train.py:968] (0/2) Epoch 20, batch 38000, giga_loss[loss=0.2633, simple_loss=0.3402, pruned_loss=0.09314, over 28427.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3509, pruned_loss=0.1018, over 5688746.40 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3598, pruned_loss=0.1138, over 5683020.90 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.349, pruned_loss=0.1001, over 5692247.32 frames. ], batch size: 71, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:31:19,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1888, 1.2326, 1.1350, 0.8685], device='cuda:0'), covar=tensor([0.1069, 0.0535, 0.1098, 0.1083], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0444, 0.0518, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 15:31:51,009 INFO [train.py:968] (0/2) Epoch 20, batch 38050, giga_loss[loss=0.3014, simple_loss=0.377, pruned_loss=0.1129, over 27829.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3512, pruned_loss=0.1013, over 5697651.33 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.36, pruned_loss=0.1138, over 5686554.97 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3494, pruned_loss=0.09978, over 5697593.09 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:31:59,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.832e+02 1.250e+03 1.485e+03 2.009e+03 9.873e+03, threshold=2.970e+03, percent-clipped=5.0
+2023-03-10 15:32:33,487 INFO [train.py:968] (0/2) Epoch 20, batch 38100, giga_loss[loss=0.302, simple_loss=0.37, pruned_loss=0.117, over 29150.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.355, pruned_loss=0.1043, over 5698889.82 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3606, pruned_loss=0.1144, over 5690637.37 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3527, pruned_loss=0.102, over 5695811.85 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:32:52,164 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905283.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 15:33:16,077 INFO [train.py:968] (0/2) Epoch 20, batch 38150, giga_loss[loss=0.2629, simple_loss=0.3384, pruned_loss=0.09374, over 28687.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3561, pruned_loss=0.105, over 5691249.03 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3607, pruned_loss=0.1146, over 5682291.62 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3541, pruned_loss=0.1029, over 5696568.20 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:33:24,356 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:33:24,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.993e+02 1.350e+03 1.651e+03 2.374e+03 7.297e+03, threshold=3.301e+03, percent-clipped=14.0
+2023-03-10 15:33:26,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1900, 2.5965, 1.2067, 1.4028], device='cuda:0'), covar=tensor([0.1104, 0.0403, 0.0916, 0.1467], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0550, 0.0379, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 15:33:51,938 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:34:01,668 INFO [train.py:968] (0/2) Epoch 20, batch 38200, giga_loss[loss=0.2965, simple_loss=0.3722, pruned_loss=0.1104, over 29012.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3589, pruned_loss=0.1074, over 5689212.97 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3611, pruned_loss=0.1148, over 5686484.52 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3568, pruned_loss=0.1053, over 5689832.32 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:34:42,604 INFO [train.py:968] (0/2) Epoch 20, batch 38250, giga_loss[loss=0.2802, simple_loss=0.35, pruned_loss=0.1052, over 28880.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3592, pruned_loss=0.1081, over 5693132.01 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3613, pruned_loss=0.1148, over 5683119.98 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3572, pruned_loss=0.1061, over 5696243.59 frames. ], batch size: 112, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:34:47,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:34:51,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.903e+02 1.322e+03 1.786e+03 2.629e+03 6.335e+03, threshold=3.572e+03, percent-clipped=18.0
+2023-03-10 15:34:56,518 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=905426.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 15:34:59,523 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=905429.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 15:35:24,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=905458.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 15:35:26,277 INFO [train.py:968] (0/2) Epoch 20, batch 38300, giga_loss[loss=0.2646, simple_loss=0.3394, pruned_loss=0.09485, over 28551.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3598, pruned_loss=0.109, over 5677247.30 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3614, pruned_loss=0.1148, over 5673809.14 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3581, pruned_loss=0.1073, over 5687970.39 frames. ], batch size: 78, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:35:49,506 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6521, 4.4683, 4.3125, 1.9929], device='cuda:0'), covar=tensor([0.0687, 0.0869, 0.0950, 0.2005], device='cuda:0'), in_proj_covar=tensor([0.1187, 0.1102, 0.0936, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 15:36:06,537 INFO [train.py:968] (0/2) Epoch 20, batch 38350, giga_loss[loss=0.2738, simple_loss=0.3517, pruned_loss=0.09793, over 28663.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.36, pruned_loss=0.1085, over 5687010.71 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3618, pruned_loss=0.1148, over 5676229.77 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3583, pruned_loss=0.107, over 5693727.66 frames. ], batch size: 78, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:36:11,414 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:36:14,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.363e+02 1.210e+03 1.453e+03 2.129e+03 5.762e+03, threshold=2.905e+03, percent-clipped=4.0
+2023-03-10 15:36:46,278 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=905558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:36:47,975 INFO [train.py:968] (0/2) Epoch 20, batch 38400, giga_loss[loss=0.3479, simple_loss=0.4163, pruned_loss=0.1397, over 28560.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3591, pruned_loss=0.1067, over 5694919.47 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3618, pruned_loss=0.1148, over 5680948.56 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3577, pruned_loss=0.1054, over 5696232.55 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:36:48,259 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=905561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:36:55,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9029, 2.1226, 2.0301, 1.7488], device='cuda:0'), covar=tensor([0.3024, 0.2402, 0.2334, 0.2678], device='cuda:0'), in_proj_covar=tensor([0.1962, 0.1862, 0.1810, 0.1951], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 15:37:09,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8342, 2.0202, 2.0767, 1.6226], device='cuda:0'), covar=tensor([0.1938, 0.2463, 0.1521, 0.1788], device='cuda:0'), in_proj_covar=tensor([0.0896, 0.0699, 0.0941, 0.0839], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 15:37:10,190 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:37:13,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=905590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:37:19,892 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:37:29,508 INFO [train.py:968] (0/2) Epoch 20, batch 38450, giga_loss[loss=0.2782, simple_loss=0.3529, pruned_loss=0.1017, over 28362.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3587, pruned_loss=0.1056, over 5703480.18 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1152, over 5681144.52 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3573, pruned_loss=0.1041, over 5704543.21 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:37:37,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.107e+02 1.181e+03 1.432e+03 2.277e+03 5.255e+03, threshold=2.864e+03, percent-clipped=12.0
+2023-03-10 15:37:43,344 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:38:09,840 INFO [train.py:968] (0/2) Epoch 20, batch 38500, libri_loss[loss=0.2767, simple_loss=0.3502, pruned_loss=0.1016, over 29566.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3568, pruned_loss=0.1047, over 5703723.52 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3618, pruned_loss=0.1149, over 5685420.60 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3558, pruned_loss=0.1035, over 5701397.34 frames. ], batch size: 83, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:38:35,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:38:37,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-10 15:38:48,611 INFO [train.py:968] (0/2) Epoch 20, batch 38550, giga_loss[loss=0.2469, simple_loss=0.3248, pruned_loss=0.08449, over 28604.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3541, pruned_loss=0.1034, over 5705985.41 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3616, pruned_loss=0.1148, over 5691056.76 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3532, pruned_loss=0.1021, over 5699606.42 frames. ], batch size: 66, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:38:55,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.175e+02 1.248e+03 1.615e+03 2.347e+03 1.043e+04, threshold=3.231e+03, percent-clipped=16.0
+2023-03-10 15:38:59,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:39:27,059 INFO [train.py:968] (0/2) Epoch 20, batch 38600, giga_loss[loss=0.2692, simple_loss=0.3481, pruned_loss=0.09515, over 28942.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3516, pruned_loss=0.1018, over 5708740.13 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5693738.54 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3507, pruned_loss=0.1004, over 5701486.10 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:39:39,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7863, 2.4305, 1.5936, 1.0176], device='cuda:0'), covar=tensor([0.7041, 0.3312, 0.3430, 0.6156], device='cuda:0'), in_proj_covar=tensor([0.1706, 0.1608, 0.1568, 0.1391], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 15:39:39,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2697, 1.1261, 4.0163, 3.3385], device='cuda:0'), covar=tensor([0.1682, 0.3009, 0.0419, 0.1391], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0639, 0.0944, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 15:40:09,638 INFO [train.py:968] (0/2) Epoch 20, batch 38650, giga_loss[loss=0.3271, simple_loss=0.3929, pruned_loss=0.1307, over 28881.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3508, pruned_loss=0.1014, over 5707559.65 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3613, pruned_loss=0.1145, over 5696046.53 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3502, pruned_loss=0.1004, over 5699978.35 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:40:16,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.982e+02 1.066e+03 1.270e+03 1.727e+03 8.836e+03, threshold=2.540e+03, percent-clipped=4.0
+2023-03-10 15:40:28,094 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=905837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:40:30,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=905840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:40:43,420 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 15:40:47,368 INFO [train.py:968] (0/2) Epoch 20, batch 38700, giga_loss[loss=0.2453, simple_loss=0.333, pruned_loss=0.07883, over 28915.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3505, pruned_loss=0.1012, over 5710899.34 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3608, pruned_loss=0.114, over 5701305.04 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3501, pruned_loss=0.1003, over 5700708.39 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:40:53,505 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=905868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:40:54,377 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=905869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:40:57,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=905871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:41:13,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:41:13,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0152, 1.3046, 1.0696, 0.2498], device='cuda:0'), covar=tensor([0.3602, 0.2717, 0.4187, 0.6227], device='cuda:0'), in_proj_covar=tensor([0.1708, 0.1609, 0.1570, 0.1391], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 15:41:16,901 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-10 15:41:18,016 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=905900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:41:27,675 INFO [train.py:968] (0/2) Epoch 20, batch 38750, giga_loss[loss=0.2561, simple_loss=0.3398, pruned_loss=0.08616, over 28855.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3512, pruned_loss=0.1013, over 5714131.13 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3608, pruned_loss=0.1141, over 5704100.96 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3507, pruned_loss=0.1003, over 5703679.77 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:41:35,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.016e+02 1.160e+03 1.468e+03 1.758e+03 1.379e+04, threshold=2.937e+03, percent-clipped=11.0
+2023-03-10 15:42:05,703 INFO [train.py:968] (0/2) Epoch 20, batch 38800, giga_loss[loss=0.2796, simple_loss=0.3586, pruned_loss=0.1003, over 28885.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3515, pruned_loss=0.101, over 5710182.25 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3614, pruned_loss=0.1145, over 5697643.59 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3502, pruned_loss=0.0994, over 5707750.04 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:42:05,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:42:14,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:42:36,191 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-906000.pt
+2023-03-10 15:42:37,037 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-10 15:42:40,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:42:43,943 INFO [train.py:968] (0/2) Epoch 20, batch 38850, libri_loss[loss=0.2712, simple_loss=0.3417, pruned_loss=0.1003, over 29539.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3501, pruned_loss=0.09982, over 5717607.26 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3612, pruned_loss=0.1144, over 5703903.28 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3489, pruned_loss=0.09819, over 5710251.85 frames. ], batch size: 80, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:42:47,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:42:52,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.145e+02 9.583e+02 1.284e+03 1.773e+03 3.206e+03, threshold=2.568e+03, percent-clipped=1.0
+2023-03-10 15:42:55,771 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:43:07,563 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:43:09,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:43:14,277 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 15:43:28,641 INFO [train.py:968] (0/2) Epoch 20, batch 38900, giga_loss[loss=0.2355, simple_loss=0.3182, pruned_loss=0.07644, over 28783.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.349, pruned_loss=0.09967, over 5706817.62 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1143, over 5704802.03 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.348, pruned_loss=0.09831, over 5700257.80 frames. ], batch size: 66, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:43:30,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2800, 0.8301, 0.9844, 1.4499], device='cuda:0'), covar=tensor([0.0773, 0.0380, 0.0340, 0.0840], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 15:43:32,938 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:02,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:05,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:09,199 INFO [train.py:968] (0/2) Epoch 20, batch 38950, libri_loss[loss=0.2879, simple_loss=0.3447, pruned_loss=0.1156, over 29369.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3465, pruned_loss=0.09882, over 5700546.34 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3612, pruned_loss=0.1145, over 5698760.58 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3454, pruned_loss=0.09725, over 5701239.31 frames. ], batch size: 67, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:44:13,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:15,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:17,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.524e+02 1.172e+03 1.548e+03 2.276e+03 6.737e+03, threshold=3.095e+03, percent-clipped=19.0
+2023-03-10 15:44:29,307 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:37,449 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:37,534 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4076, 1.6887, 1.7203, 1.4187], device='cuda:0'), covar=tensor([0.1602, 0.1546, 0.1863, 0.1736], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0747, 0.0710, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 15:44:37,552 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:40,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:47,676 INFO [train.py:968] (0/2) Epoch 20, batch 39000, giga_loss[loss=0.2076, simple_loss=0.2972, pruned_loss=0.05897, over 28598.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3423, pruned_loss=0.09645, over 5707300.41 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.361, pruned_loss=0.1144, over 5704181.57 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3411, pruned_loss=0.09486, over 5703032.90 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:44:47,682 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 15:44:55,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3049, 3.1689, 1.3857, 1.4652], device='cuda:0'), covar=tensor([0.1121, 0.0328, 0.1033, 0.1533], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0544, 0.0376, 0.0418], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 15:44:56,766 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2055, simple_loss=0.3137, pruned_loss=0.04862, over 944034.00 frames. 
+2023-03-10 15:44:56,767 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 15:45:03,675 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4181, 1.5147, 1.4565, 1.3130], device='cuda:0'), covar=tensor([0.2794, 0.2647, 0.1821, 0.2414], device='cuda:0'), in_proj_covar=tensor([0.1957, 0.1866, 0.1807, 0.1953], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 15:45:12,094 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:45:12,679 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:45:35,429 INFO [train.py:968] (0/2) Epoch 20, batch 39050, giga_loss[loss=0.2868, simple_loss=0.3558, pruned_loss=0.109, over 28740.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3419, pruned_loss=0.09673, over 5699520.82 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3612, pruned_loss=0.1147, over 5697577.97 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.34, pruned_loss=0.09456, over 5702573.10 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:45:44,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.364e+02 1.198e+03 1.487e+03 1.963e+03 8.891e+03, threshold=2.974e+03, percent-clipped=7.0
+2023-03-10 15:46:16,044 INFO [train.py:968] (0/2) Epoch 20, batch 39100, giga_loss[loss=0.2186, simple_loss=0.3019, pruned_loss=0.06764, over 29006.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3403, pruned_loss=0.09593, over 5701441.58 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3611, pruned_loss=0.1146, over 5692879.62 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3385, pruned_loss=0.09386, over 5709097.50 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:46:32,596 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 15:46:42,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1299, 1.1410, 3.2990, 2.9878], device='cuda:0'), covar=tensor([0.1593, 0.2711, 0.0471, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0747, 0.0637, 0.0941, 0.0888], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 15:46:54,315 INFO [train.py:968] (0/2) Epoch 20, batch 39150, giga_loss[loss=0.2623, simple_loss=0.3387, pruned_loss=0.09295, over 28716.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3386, pruned_loss=0.09551, over 5707414.59 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3615, pruned_loss=0.1149, over 5696133.67 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3362, pruned_loss=0.09304, over 5710761.50 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:47:02,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.711e+02 1.122e+03 1.432e+03 1.752e+03 1.048e+04, threshold=2.863e+03, percent-clipped=7.0
+2023-03-10 15:47:23,180 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:47:33,658 INFO [train.py:968] (0/2) Epoch 20, batch 39200, giga_loss[loss=0.2756, simple_loss=0.3399, pruned_loss=0.1056, over 28870.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3371, pruned_loss=0.09535, over 5707501.66 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3617, pruned_loss=0.115, over 5698483.73 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3348, pruned_loss=0.09304, over 5708339.78 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:47:58,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:48:03,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:48:12,206 INFO [train.py:968] (0/2) Epoch 20, batch 39250, giga_loss[loss=0.2445, simple_loss=0.3208, pruned_loss=0.08412, over 28730.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3353, pruned_loss=0.09458, over 5710092.27 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3618, pruned_loss=0.115, over 5701838.61 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3324, pruned_loss=0.09213, over 5707830.40 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:48:23,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.521e+02 1.120e+03 1.422e+03 1.897e+03 4.603e+03, threshold=2.845e+03, percent-clipped=10.0
+2023-03-10 15:48:58,021 INFO [train.py:968] (0/2) Epoch 20, batch 39300, giga_loss[loss=0.275, simple_loss=0.3399, pruned_loss=0.105, over 28862.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3341, pruned_loss=0.09424, over 5709105.60 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3615, pruned_loss=0.1148, over 5704606.35 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3318, pruned_loss=0.09216, over 5704878.68 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:49:37,633 INFO [train.py:968] (0/2) Epoch 20, batch 39350, giga_loss[loss=0.2911, simple_loss=0.3663, pruned_loss=0.108, over 28815.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3354, pruned_loss=0.09452, over 5714511.78 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3615, pruned_loss=0.1148, over 5711916.00 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3327, pruned_loss=0.0922, over 5704756.46 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:49:46,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.869e+02 1.064e+03 1.310e+03 1.769e+03 4.040e+03, threshold=2.619e+03, percent-clipped=4.0
+2023-03-10 15:49:58,845 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:49:58,900 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:00,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:04,679 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:07,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:18,754 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:22,341 INFO [train.py:968] (0/2) Epoch 20, batch 39400, giga_loss[loss=0.2727, simple_loss=0.3632, pruned_loss=0.09106, over 28763.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3381, pruned_loss=0.09476, over 5717603.96 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3612, pruned_loss=0.1146, over 5713377.56 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3357, pruned_loss=0.0927, over 5708783.68 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:50:28,016 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:36,993 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:51:06,219 INFO [train.py:968] (0/2) Epoch 20, batch 39450, libri_loss[loss=0.3236, simple_loss=0.3853, pruned_loss=0.131, over 29524.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3422, pruned_loss=0.09704, over 5699952.52 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3619, pruned_loss=0.1151, over 5706671.51 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.339, pruned_loss=0.09429, over 5699138.87 frames. ], batch size: 81, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:51:14,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.267e+02 1.147e+03 1.411e+03 2.099e+03 5.897e+03, threshold=2.823e+03, percent-clipped=11.0
+2023-03-10 15:51:30,743 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:51:50,280 INFO [train.py:968] (0/2) Epoch 20, batch 39500, giga_loss[loss=0.2314, simple_loss=0.3132, pruned_loss=0.07477, over 28479.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3426, pruned_loss=0.09617, over 5701070.07 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3615, pruned_loss=0.1148, over 5708862.33 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3402, pruned_loss=0.09408, over 5698397.81 frames. ], batch size: 71, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:52:01,548 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-10 15:52:11,669 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-10 15:52:22,418 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:52:25,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:52:32,847 INFO [train.py:968] (0/2) Epoch 20, batch 39550, giga_loss[loss=0.2431, simple_loss=0.3262, pruned_loss=0.07998, over 28902.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3427, pruned_loss=0.09584, over 5686024.66 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3616, pruned_loss=0.115, over 5703780.93 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3401, pruned_loss=0.09342, over 5687269.61 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:52:41,259 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:52:42,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.170e+02 1.074e+03 1.355e+03 1.989e+03 6.684e+03, threshold=2.710e+03, percent-clipped=16.0
+2023-03-10 15:52:47,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:52:54,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4411, 1.5288, 1.6301, 1.2287], device='cuda:0'), covar=tensor([0.1729, 0.2677, 0.1476, 0.1787], device='cuda:0'), in_proj_covar=tensor([0.0895, 0.0696, 0.0939, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 15:53:13,267 INFO [train.py:968] (0/2) Epoch 20, batch 39600, giga_loss[loss=0.3447, simple_loss=0.4043, pruned_loss=0.1425, over 28579.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3425, pruned_loss=0.09543, over 5697935.63 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.362, pruned_loss=0.1153, over 5705795.85 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.34, pruned_loss=0.09313, over 5697091.62 frames. ], batch size: 336, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:53:28,849 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 15:53:34,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7828, 2.7576, 1.7654, 0.8536], device='cuda:0'), covar=tensor([0.8284, 0.3251, 0.4148, 0.7595], device='cuda:0'), in_proj_covar=tensor([0.1712, 0.1610, 0.1575, 0.1394], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 15:53:46,583 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-10 15:53:53,871 INFO [train.py:968] (0/2) Epoch 20, batch 39650, giga_loss[loss=0.2579, simple_loss=0.3359, pruned_loss=0.08991, over 28872.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3434, pruned_loss=0.09651, over 5698832.38 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3624, pruned_loss=0.1156, over 5709723.40 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3405, pruned_loss=0.09391, over 5694260.58 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:54:04,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.451e+02 1.248e+03 1.512e+03 1.898e+03 5.075e+03, threshold=3.024e+03, percent-clipped=11.0
+2023-03-10 15:54:35,329 INFO [train.py:968] (0/2) Epoch 20, batch 39700, libri_loss[loss=0.2582, simple_loss=0.3221, pruned_loss=0.09718, over 27750.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3462, pruned_loss=0.09844, over 5692288.83 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3631, pruned_loss=0.1161, over 5709512.31 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3429, pruned_loss=0.09545, over 5688590.08 frames. ], batch size: 61, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:54:38,209 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:54:40,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:55:01,417 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:55:11,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:55:12,840 INFO [train.py:968] (0/2) Epoch 20, batch 39750, libri_loss[loss=0.3407, simple_loss=0.4013, pruned_loss=0.1401, over 27777.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3505, pruned_loss=0.1012, over 5690735.33 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.364, pruned_loss=0.1165, over 5697148.75 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3463, pruned_loss=0.09774, over 5697780.35 frames. ], batch size: 116, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:55:25,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2607, 1.4215, 1.2302, 1.4598], device='cuda:0'), covar=tensor([0.0731, 0.0385, 0.0364, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 15:55:25,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.259e+02 1.347e+03 1.694e+03 2.406e+03 6.836e+03, threshold=3.388e+03, percent-clipped=12.0
+2023-03-10 15:55:47,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7320, 1.1632, 4.9804, 3.5512], device='cuda:0'), covar=tensor([0.1608, 0.3011, 0.0380, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0747, 0.0639, 0.0943, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 15:55:54,231 INFO [train.py:968] (0/2) Epoch 20, batch 39800, giga_loss[loss=0.2615, simple_loss=0.3522, pruned_loss=0.08543, over 28962.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3522, pruned_loss=0.1014, over 5700930.08 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3643, pruned_loss=0.1166, over 5699220.12 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3485, pruned_loss=0.0984, over 5704453.13 frames. ], batch size: 164, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:56:33,403 INFO [train.py:968] (0/2) Epoch 20, batch 39850, giga_loss[loss=0.2783, simple_loss=0.3483, pruned_loss=0.1042, over 28583.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3537, pruned_loss=0.1025, over 5700509.84 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3646, pruned_loss=0.1169, over 5695792.26 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3502, pruned_loss=0.09936, over 5706974.03 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:56:35,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=907013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:56:46,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.399e+02 1.322e+03 1.595e+03 2.277e+03 1.214e+04, threshold=3.190e+03, percent-clipped=12.0
+2023-03-10 15:57:03,488 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=907046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:57:09,880 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:57:10,686 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 15:57:11,081 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5465, 1.8431, 1.4932, 1.6032], device='cuda:0'), covar=tensor([0.2486, 0.2568, 0.2925, 0.2449], device='cuda:0'), in_proj_covar=tensor([0.1475, 0.1071, 0.1303, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 15:57:11,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=907055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:57:16,177 INFO [train.py:968] (0/2) Epoch 20, batch 39900, giga_loss[loss=0.2805, simple_loss=0.3505, pruned_loss=0.1052, over 28925.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3527, pruned_loss=0.1017, over 5705329.75 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3645, pruned_loss=0.1167, over 5699104.22 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3499, pruned_loss=0.0992, over 5707612.80 frames. ], batch size: 112, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:57:34,093 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:57:34,365 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 15:57:54,273 INFO [train.py:968] (0/2) Epoch 20, batch 39950, giga_loss[loss=0.276, simple_loss=0.3677, pruned_loss=0.09217, over 29034.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3524, pruned_loss=0.1015, over 5707100.04 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3651, pruned_loss=0.1171, over 5693800.64 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3492, pruned_loss=0.0987, over 5713935.10 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:58:01,812 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=907118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:58:07,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.242e+02 1.152e+03 1.492e+03 2.129e+03 5.293e+03, threshold=2.983e+03, percent-clipped=13.0
+2023-03-10 15:58:31,568 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:58:33,692 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=907159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:58:35,194 INFO [train.py:968] (0/2) Epoch 20, batch 40000, giga_loss[loss=0.2751, simple_loss=0.3522, pruned_loss=0.09897, over 28976.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.351, pruned_loss=0.1011, over 5698074.87 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3654, pruned_loss=0.1173, over 5684426.77 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.348, pruned_loss=0.09845, over 5711491.75 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:59:00,005 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:59:20,094 INFO [train.py:968] (0/2) Epoch 20, batch 40050, giga_loss[loss=0.2455, simple_loss=0.328, pruned_loss=0.08151, over 28946.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3475, pruned_loss=0.09949, over 5700138.60 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3655, pruned_loss=0.1174, over 5686469.10 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.345, pruned_loss=0.09723, over 5708983.38 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:59:31,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.319e+02 1.171e+03 1.487e+03 2.107e+03 1.045e+04, threshold=2.973e+03, percent-clipped=8.0
+2023-03-10 16:00:01,886 INFO [train.py:968] (0/2) Epoch 20, batch 40100, giga_loss[loss=0.2545, simple_loss=0.3324, pruned_loss=0.08829, over 28999.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3439, pruned_loss=0.09733, over 5706388.46 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3656, pruned_loss=0.1173, over 5689199.53 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3414, pruned_loss=0.0953, over 5711168.74 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:00:42,040 INFO [train.py:968] (0/2) Epoch 20, batch 40150, giga_loss[loss=0.2871, simple_loss=0.363, pruned_loss=0.1057, over 28750.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3444, pruned_loss=0.09686, over 5694625.74 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3663, pruned_loss=0.1178, over 5675435.29 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3413, pruned_loss=0.09435, over 5710637.42 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:00:49,763 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=907320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:00:53,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.862e+02 1.164e+03 1.449e+03 1.999e+03 9.720e+03, threshold=2.897e+03, percent-clipped=12.0
+2023-03-10 16:01:25,534 INFO [train.py:968] (0/2) Epoch 20, batch 40200, giga_loss[loss=0.2318, simple_loss=0.3173, pruned_loss=0.07313, over 29019.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3454, pruned_loss=0.09664, over 5685509.51 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3661, pruned_loss=0.1178, over 5671727.24 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3426, pruned_loss=0.09413, over 5702474.36 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:01:40,850 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 16:01:46,137 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-10 16:02:07,956 INFO [train.py:968] (0/2) Epoch 20, batch 40250, giga_loss[loss=0.2467, simple_loss=0.3154, pruned_loss=0.08907, over 28525.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3462, pruned_loss=0.09728, over 5692606.18 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3659, pruned_loss=0.1176, over 5673693.07 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3439, pruned_loss=0.09517, over 5704579.06 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:02:15,678 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=907421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:02:18,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.098e+02 1.130e+03 1.520e+03 2.051e+03 4.428e+03, threshold=3.040e+03, percent-clipped=6.0
+2023-03-10 16:02:24,123 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=907432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:02:46,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6706, 1.0482, 4.6298, 3.5733], device='cuda:0'), covar=tensor([0.1571, 0.3060, 0.0417, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0745, 0.0637, 0.0941, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 16:02:48,983 INFO [train.py:968] (0/2) Epoch 20, batch 40300, giga_loss[loss=0.2187, simple_loss=0.2976, pruned_loss=0.06987, over 28548.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3433, pruned_loss=0.09623, over 5701820.50 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3657, pruned_loss=0.1174, over 5677141.89 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3414, pruned_loss=0.09446, over 5708605.01 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:03:13,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=907493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:03:27,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.1409, 5.9291, 5.6293, 3.4031], device='cuda:0'), covar=tensor([0.0494, 0.0677, 0.0747, 0.1476], device='cuda:0'), in_proj_covar=tensor([0.1199, 0.1106, 0.0942, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 16:03:28,040 INFO [train.py:968] (0/2) Epoch 20, batch 40350, giga_loss[loss=0.2069, simple_loss=0.2819, pruned_loss=0.06595, over 28420.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3428, pruned_loss=0.09729, over 5701991.72 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3662, pruned_loss=0.1179, over 5668601.24 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3401, pruned_loss=0.09491, over 5716136.61 frames. ], batch size: 65, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:03:40,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.770e+02 1.180e+03 1.536e+03 2.231e+03 8.312e+03, threshold=3.072e+03, percent-clipped=8.0
+2023-03-10 16:03:49,185 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9070, 3.7265, 3.5197, 1.9923], device='cuda:0'), covar=tensor([0.0639, 0.0824, 0.0719, 0.1980], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.1105, 0.0940, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 16:03:56,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5601, 1.7836, 1.4430, 1.7279], device='cuda:0'), covar=tensor([0.2516, 0.2641, 0.2892, 0.2398], device='cuda:0'), in_proj_covar=tensor([0.1476, 0.1072, 0.1304, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 16:04:10,966 INFO [train.py:968] (0/2) Epoch 20, batch 40400, giga_loss[loss=0.282, simple_loss=0.3487, pruned_loss=0.1077, over 28997.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3414, pruned_loss=0.09813, over 5690789.15 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3664, pruned_loss=0.1181, over 5661127.32 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3388, pruned_loss=0.09573, over 5709076.55 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:04:14,389 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:04:16,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=907567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:04:17,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3628, 1.7064, 1.6762, 1.5698], device='cuda:0'), covar=tensor([0.1949, 0.1741, 0.2251, 0.2075], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0745, 0.0710, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 16:04:42,447 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:04:54,105 INFO [train.py:968] (0/2) Epoch 20, batch 40450, giga_loss[loss=0.3168, simple_loss=0.3667, pruned_loss=0.1334, over 23934.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3406, pruned_loss=0.09833, over 5691475.47 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3663, pruned_loss=0.118, over 5664560.04 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3383, pruned_loss=0.09629, over 5703310.18 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:05:06,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.556e+02 1.320e+03 1.570e+03 2.176e+03 5.707e+03, threshold=3.139e+03, percent-clipped=7.0
+2023-03-10 16:05:15,671 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:05:17,692 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=907639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:05:34,998 INFO [train.py:968] (0/2) Epoch 20, batch 40500, giga_loss[loss=0.2287, simple_loss=0.3017, pruned_loss=0.07783, over 28643.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3393, pruned_loss=0.09785, over 5697652.04 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3667, pruned_loss=0.118, over 5668582.69 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3368, pruned_loss=0.09585, over 5704025.55 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:05:40,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:06:04,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=907695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:06:05,107 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-10 16:06:16,801 INFO [train.py:968] (0/2) Epoch 20, batch 40550, giga_loss[loss=0.2065, simple_loss=0.2805, pruned_loss=0.06626, over 28536.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3354, pruned_loss=0.09605, over 5700985.21 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3667, pruned_loss=0.118, over 5672201.36 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.333, pruned_loss=0.09422, over 5703212.69 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:06:28,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.949e+02 1.106e+03 1.342e+03 1.833e+03 6.264e+03, threshold=2.685e+03, percent-clipped=2.0
+2023-03-10 16:06:57,790 INFO [train.py:968] (0/2) Epoch 20, batch 40600, giga_loss[loss=0.2253, simple_loss=0.3012, pruned_loss=0.07473, over 28889.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3316, pruned_loss=0.09425, over 5704630.92 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3665, pruned_loss=0.1179, over 5670757.02 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3288, pruned_loss=0.0921, over 5708229.13 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:07:32,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 16:07:35,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=907807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:07:38,418 INFO [train.py:968] (0/2) Epoch 20, batch 40650, giga_loss[loss=0.2258, simple_loss=0.2964, pruned_loss=0.07764, over 28416.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3319, pruned_loss=0.09376, over 5713306.94 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3665, pruned_loss=0.1179, over 5674491.28 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.329, pruned_loss=0.09164, over 5713445.84 frames. ], batch size: 78, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:07:41,406 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1586, 1.6174, 1.4143, 1.3537], device='cuda:0'), covar=tensor([0.2303, 0.2026, 0.2403, 0.2257], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0748, 0.0713, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 16:07:51,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.898e+02 1.258e+03 1.566e+03 2.123e+03 4.099e+03, threshold=3.131e+03, percent-clipped=11.0
+2023-03-10 16:08:02,167 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:08:05,573 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=907841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:08:21,899 INFO [train.py:968] (0/2) Epoch 20, batch 40700, giga_loss[loss=0.2442, simple_loss=0.327, pruned_loss=0.08069, over 29011.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3363, pruned_loss=0.0959, over 5711622.74 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3661, pruned_loss=0.1178, over 5678805.81 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3337, pruned_loss=0.09393, over 5708385.26 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:08:29,223 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:09:00,493 INFO [train.py:968] (0/2) Epoch 20, batch 40750, giga_loss[loss=0.273, simple_loss=0.354, pruned_loss=0.09602, over 28265.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3387, pruned_loss=0.0968, over 5708635.53 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3656, pruned_loss=0.1174, over 5675026.47 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3363, pruned_loss=0.09484, over 5710895.04 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:09:14,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.287e+02 1.227e+03 1.638e+03 2.236e+03 6.834e+03, threshold=3.276e+03, percent-clipped=11.0
+2023-03-10 16:09:33,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:09:36,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=907953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:09:42,776 INFO [train.py:968] (0/2) Epoch 20, batch 40800, giga_loss[loss=0.2823, simple_loss=0.359, pruned_loss=0.1028, over 28250.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3421, pruned_loss=0.09786, over 5705637.91 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3658, pruned_loss=0.1176, over 5678889.17 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3394, pruned_loss=0.09571, over 5704816.49 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:10:03,656 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:10:18,089 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-908000.pt
+2023-03-10 16:10:25,659 INFO [train.py:968] (0/2) Epoch 20, batch 40850, giga_loss[loss=0.2564, simple_loss=0.3394, pruned_loss=0.08669, over 28914.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3468, pruned_loss=0.1008, over 5707449.73 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3663, pruned_loss=0.1179, over 5681577.42 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3438, pruned_loss=0.0984, over 5704566.97 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:10:39,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.605e+02 1.232e+03 1.481e+03 2.106e+03 4.512e+03, threshold=2.963e+03, percent-clipped=10.0
+2023-03-10 16:11:13,310 INFO [train.py:968] (0/2) Epoch 20, batch 40900, giga_loss[loss=0.3386, simple_loss=0.3935, pruned_loss=0.1419, over 28599.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3504, pruned_loss=0.1032, over 5708161.38 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3663, pruned_loss=0.118, over 5683614.26 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3479, pruned_loss=0.1012, over 5704273.58 frames. ], batch size: 336, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:11:53,796 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-10 16:12:03,774 INFO [train.py:968] (0/2) Epoch 20, batch 40950, giga_loss[loss=0.4131, simple_loss=0.4349, pruned_loss=0.1956, over 26596.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3562, pruned_loss=0.1083, over 5704010.90 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3662, pruned_loss=0.1179, over 5686021.82 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3538, pruned_loss=0.1064, over 5699333.34 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:12:19,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.076e+02 1.488e+03 2.190e+03 3.195e+03 8.114e+03, threshold=4.381e+03, percent-clipped=26.0
+2023-03-10 16:12:22,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5177, 1.7992, 1.4788, 1.6309], device='cuda:0'), covar=tensor([0.2452, 0.2379, 0.2584, 0.2083], device='cuda:0'), in_proj_covar=tensor([0.1484, 0.1077, 0.1308, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 16:12:50,793 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-10 16:12:51,551 INFO [train.py:968] (0/2) Epoch 20, batch 41000, giga_loss[loss=0.2985, simple_loss=0.3768, pruned_loss=0.1101, over 28914.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3632, pruned_loss=0.1137, over 5701208.43 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.366, pruned_loss=0.1177, over 5688000.27 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3615, pruned_loss=0.1122, over 5695901.15 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:13:39,612 INFO [train.py:968] (0/2) Epoch 20, batch 41050, giga_loss[loss=0.3345, simple_loss=0.4011, pruned_loss=0.134, over 28514.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3701, pruned_loss=0.1189, over 5693933.43 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3655, pruned_loss=0.1174, over 5689604.33 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3692, pruned_loss=0.1179, over 5688593.04 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:13:53,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.854e+02 1.656e+03 2.153e+03 2.869e+03 1.012e+04, threshold=4.306e+03, percent-clipped=8.0
+2023-03-10 16:14:25,501 INFO [train.py:968] (0/2) Epoch 20, batch 41100, giga_loss[loss=0.3105, simple_loss=0.3826, pruned_loss=0.1192, over 28874.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3756, pruned_loss=0.1232, over 5698320.85 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3656, pruned_loss=0.1174, over 5692719.40 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3749, pruned_loss=0.1226, over 5691548.74 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:14:40,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3852, 1.1550, 4.1066, 3.3709], device='cuda:0'), covar=tensor([0.1635, 0.2864, 0.0430, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0746, 0.0637, 0.0944, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 16:15:14,833 INFO [train.py:968] (0/2) Epoch 20, batch 41150, giga_loss[loss=0.4609, simple_loss=0.4734, pruned_loss=0.2242, over 27446.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3808, pruned_loss=0.1277, over 5689219.26 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3657, pruned_loss=0.1174, over 5687786.81 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3803, pruned_loss=0.1273, over 5688191.69 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:15:32,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2807, 4.0614, 3.8681, 1.8346], device='cuda:0'), covar=tensor([0.0755, 0.0931, 0.1010, 0.2163], device='cuda:0'), in_proj_covar=tensor([0.1215, 0.1121, 0.0953, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-10 16:15:33,806 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.794e+03 2.243e+03 3.341e+03 2.243e+04, threshold=4.486e+03, percent-clipped=9.0
+2023-03-10 16:16:07,931 INFO [train.py:968] (0/2) Epoch 20, batch 41200, giga_loss[loss=0.3879, simple_loss=0.4315, pruned_loss=0.1721, over 28626.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3831, pruned_loss=0.1307, over 5658913.68 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3654, pruned_loss=0.1172, over 5683838.10 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3836, pruned_loss=0.131, over 5660997.36 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:16:44,709 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-10 16:17:00,483 INFO [train.py:968] (0/2) Epoch 20, batch 41250, giga_loss[loss=0.2885, simple_loss=0.3604, pruned_loss=0.1083, over 28847.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3836, pruned_loss=0.1321, over 5654200.86 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3648, pruned_loss=0.1169, over 5675545.57 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3851, pruned_loss=0.1331, over 5662563.01 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:17:15,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.157e+03 1.624e+03 2.212e+03 3.304e+03 7.519e+03, threshold=4.424e+03, percent-clipped=15.0
+2023-03-10 16:17:22,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5257, 1.8195, 1.5271, 1.3923], device='cuda:0'), covar=tensor([0.1819, 0.1753, 0.1781, 0.1661], device='cuda:0'), in_proj_covar=tensor([0.1480, 0.1077, 0.1308, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 16:17:52,549 INFO [train.py:968] (0/2) Epoch 20, batch 41300, giga_loss[loss=0.3894, simple_loss=0.418, pruned_loss=0.1804, over 23451.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3854, pruned_loss=0.1354, over 5643813.64 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3634, pruned_loss=0.1161, over 5683695.47 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3886, pruned_loss=0.1374, over 5642574.86 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:17:52,897 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 16:18:26,295 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-10 16:18:44,113 INFO [train.py:968] (0/2) Epoch 20, batch 41350, giga_loss[loss=0.2884, simple_loss=0.3528, pruned_loss=0.112, over 28882.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3902, pruned_loss=0.1393, over 5643565.30 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3639, pruned_loss=0.1165, over 5688133.46 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3927, pruned_loss=0.141, over 5637693.92 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:18:48,326 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=908515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:19:03,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.192e+03 1.837e+03 2.132e+03 2.911e+03 6.886e+03, threshold=4.264e+03, percent-clipped=3.0
+2023-03-10 16:19:29,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1961, 1.2721, 3.7236, 3.1244], device='cuda:0'), covar=tensor([0.1889, 0.2734, 0.0837, 0.1292], device='cuda:0'), in_proj_covar=tensor([0.0752, 0.0642, 0.0952, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 16:19:41,490 INFO [train.py:968] (0/2) Epoch 20, batch 41400, giga_loss[loss=0.4862, simple_loss=0.4766, pruned_loss=0.248, over 23671.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3933, pruned_loss=0.1426, over 5626262.86 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3638, pruned_loss=0.1163, over 5687905.48 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3958, pruned_loss=0.1443, over 5621127.16 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:20:35,075 INFO [train.py:968] (0/2) Epoch 20, batch 41450, giga_loss[loss=0.3668, simple_loss=0.4118, pruned_loss=0.1609, over 28284.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3941, pruned_loss=0.1442, over 5621581.46 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3642, pruned_loss=0.1168, over 5681801.40 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3962, pruned_loss=0.1457, over 5622392.36 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:20:50,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.145e+03 1.977e+03 2.601e+03 3.438e+03 1.334e+04, threshold=5.201e+03, percent-clipped=17.0
+2023-03-10 16:21:23,761 INFO [train.py:968] (0/2) Epoch 20, batch 41500, giga_loss[loss=0.2514, simple_loss=0.3373, pruned_loss=0.08271, over 28626.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3901, pruned_loss=0.1406, over 5645726.89 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3636, pruned_loss=0.1163, over 5688533.13 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3931, pruned_loss=0.143, over 5639010.61 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:22:14,832 INFO [train.py:968] (0/2) Epoch 20, batch 41550, giga_loss[loss=0.2864, simple_loss=0.3621, pruned_loss=0.1054, over 28784.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3899, pruned_loss=0.1393, over 5653233.73 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3635, pruned_loss=0.1164, over 5694292.13 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.393, pruned_loss=0.1416, over 5641928.97 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:22:31,208 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.674e+03 2.108e+03 3.026e+03 1.348e+04, threshold=4.216e+03, percent-clipped=5.0
+2023-03-10 16:23:03,485 INFO [train.py:968] (0/2) Epoch 20, batch 41600, giga_loss[loss=0.3125, simple_loss=0.3839, pruned_loss=0.1206, over 29000.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3904, pruned_loss=0.1386, over 5648926.08 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3645, pruned_loss=0.1173, over 5676870.80 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3927, pruned_loss=0.1402, over 5653483.45 frames. ], batch size: 164, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:23:56,054 INFO [train.py:968] (0/2) Epoch 20, batch 41650, giga_loss[loss=0.4931, simple_loss=0.498, pruned_loss=0.2441, over 24025.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3912, pruned_loss=0.1389, over 5635930.33 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3643, pruned_loss=0.117, over 5680497.34 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3939, pruned_loss=0.1411, over 5635481.87 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:24:12,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.971e+02 1.776e+03 2.265e+03 2.873e+03 6.055e+03, threshold=4.530e+03, percent-clipped=4.0
+2023-03-10 16:24:25,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1882, 2.5401, 1.2310, 1.3479], device='cuda:0'), covar=tensor([0.1003, 0.0365, 0.0893, 0.1386], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0551, 0.0380, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 16:24:31,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3436, 1.7546, 1.3718, 1.4331], device='cuda:0'), covar=tensor([0.2596, 0.2717, 0.2940, 0.2293], device='cuda:0'), in_proj_covar=tensor([0.1480, 0.1079, 0.1309, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 16:24:41,871 INFO [train.py:968] (0/2) Epoch 20, batch 41700, giga_loss[loss=0.3125, simple_loss=0.3793, pruned_loss=0.1228, over 28551.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3872, pruned_loss=0.1353, over 5635849.52 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3641, pruned_loss=0.117, over 5671480.12 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3909, pruned_loss=0.1381, over 5641851.37 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:25:09,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=908890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:25:27,809 INFO [train.py:968] (0/2) Epoch 20, batch 41750, giga_loss[loss=0.3166, simple_loss=0.388, pruned_loss=0.1226, over 29062.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3847, pruned_loss=0.132, over 5640651.49 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3641, pruned_loss=0.117, over 5673267.98 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3883, pruned_loss=0.1347, over 5643077.43 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:25:48,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.206e+03 1.736e+03 2.163e+03 2.811e+03 6.131e+03, threshold=4.327e+03, percent-clipped=3.0
+2023-03-10 16:25:49,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=908930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:26:03,474 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-10 16:26:13,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3667, 1.3468, 3.4410, 3.0768], device='cuda:0'), covar=tensor([0.1450, 0.2598, 0.0486, 0.1533], device='cuda:0'), in_proj_covar=tensor([0.0751, 0.0643, 0.0953, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 16:26:22,041 INFO [train.py:968] (0/2) Epoch 20, batch 41800, giga_loss[loss=0.3009, simple_loss=0.3718, pruned_loss=0.115, over 28636.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3807, pruned_loss=0.128, over 5653532.47 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3639, pruned_loss=0.1169, over 5675634.42 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3839, pruned_loss=0.1305, over 5653074.20 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:27:00,656 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 16:27:10,746 INFO [train.py:968] (0/2) Epoch 20, batch 41850, giga_loss[loss=0.2877, simple_loss=0.3599, pruned_loss=0.1077, over 28315.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3766, pruned_loss=0.1248, over 5659297.44 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3638, pruned_loss=0.117, over 5680709.38 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3797, pruned_loss=0.127, over 5653523.90 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:27:28,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.598e+02 1.539e+03 2.184e+03 2.957e+03 5.903e+03, threshold=4.369e+03, percent-clipped=3.0
+2023-03-10 16:27:33,567 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=909033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:27:37,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=909036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:28:01,671 INFO [train.py:968] (0/2) Epoch 20, batch 41900, giga_loss[loss=0.2804, simple_loss=0.3509, pruned_loss=0.1049, over 28588.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3745, pruned_loss=0.1239, over 5647389.11 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3639, pruned_loss=0.1169, over 5683543.67 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3771, pruned_loss=0.1257, over 5639594.74 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:28:04,842 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=909065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:28:48,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6876, 1.2309, 4.7927, 3.7149], device='cuda:0'), covar=tensor([0.1671, 0.2997, 0.0414, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0750, 0.0641, 0.0952, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 16:28:49,249 INFO [train.py:968] (0/2) Epoch 20, batch 41950, giga_loss[loss=0.3024, simple_loss=0.3741, pruned_loss=0.1154, over 28981.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3757, pruned_loss=0.1243, over 5667698.22 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.364, pruned_loss=0.117, over 5686744.73 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3779, pruned_loss=0.1258, over 5658374.77 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:29:05,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.168e+03 1.641e+03 2.164e+03 2.710e+03 7.018e+03, threshold=4.328e+03, percent-clipped=3.0
+2023-03-10 16:29:39,991 INFO [train.py:968] (0/2) Epoch 20, batch 42000, giga_loss[loss=0.2739, simple_loss=0.3487, pruned_loss=0.0996, over 28575.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3747, pruned_loss=0.1231, over 5674336.73 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.364, pruned_loss=0.1169, over 5693169.15 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3768, pruned_loss=0.1246, over 5660801.03 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:29:39,996 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 16:29:48,874 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2038, simple_loss=0.3111, pruned_loss=0.04823, over 944034.00 frames. 
+2023-03-10 16:29:48,875 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 16:29:59,944 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2971, 5.0834, 4.8208, 2.6132], device='cuda:0'), covar=tensor([0.0463, 0.0665, 0.0701, 0.1813], device='cuda:0'), in_proj_covar=tensor([0.1226, 0.1136, 0.0965, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 16:30:23,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5329, 1.7381, 1.4236, 1.6349], device='cuda:0'), covar=tensor([0.2724, 0.2777, 0.3151, 0.2377], device='cuda:0'), in_proj_covar=tensor([0.1482, 0.1078, 0.1311, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 16:30:46,605 INFO [train.py:968] (0/2) Epoch 20, batch 42050, giga_loss[loss=0.3502, simple_loss=0.391, pruned_loss=0.1546, over 23915.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3734, pruned_loss=0.1207, over 5677349.19 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.364, pruned_loss=0.1169, over 5694963.75 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3751, pruned_loss=0.1219, over 5665087.76 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:31:05,760 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.505e+03 1.931e+03 2.961e+03 5.486e+03, threshold=3.862e+03, percent-clipped=6.0
+2023-03-10 16:31:41,473 INFO [train.py:968] (0/2) Epoch 20, batch 42100, giga_loss[loss=0.3188, simple_loss=0.3962, pruned_loss=0.1207, over 28617.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3749, pruned_loss=0.1198, over 5676993.28 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3641, pruned_loss=0.117, over 5693383.59 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3763, pruned_loss=0.1207, over 5668017.24 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:32:11,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2051, 1.2620, 1.1874, 0.8839], device='cuda:0'), covar=tensor([0.1004, 0.0549, 0.1028, 0.1073], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0449, 0.0520, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 16:32:21,332 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=909305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:32:22,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3941, 1.6947, 1.6604, 1.4899], device='cuda:0'), covar=tensor([0.2056, 0.1992, 0.2436, 0.2202], device='cuda:0'), in_proj_covar=tensor([0.0472, 0.0749, 0.0713, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 16:32:26,922 INFO [train.py:968] (0/2) Epoch 20, batch 42150, giga_loss[loss=0.3086, simple_loss=0.3722, pruned_loss=0.1225, over 27582.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3741, pruned_loss=0.1193, over 5677740.40 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3637, pruned_loss=0.1168, over 5694262.86 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.376, pruned_loss=0.1204, over 5669235.75 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:32:47,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.661e+03 2.042e+03 2.642e+03 5.270e+03, threshold=4.084e+03, percent-clipped=8.0
+2023-03-10 16:33:16,651 INFO [train.py:968] (0/2) Epoch 20, batch 42200, giga_loss[loss=0.2602, simple_loss=0.3358, pruned_loss=0.0923, over 28830.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3755, pruned_loss=0.121, over 5677629.56 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3637, pruned_loss=0.1168, over 5694873.86 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3771, pruned_loss=0.1219, over 5670043.39 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:34:05,896 INFO [train.py:968] (0/2) Epoch 20, batch 42250, giga_loss[loss=0.34, simple_loss=0.4024, pruned_loss=0.1388, over 28943.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3738, pruned_loss=0.1205, over 5682281.58 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3638, pruned_loss=0.1168, over 5697014.83 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3751, pruned_loss=0.1213, over 5674206.16 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:34:07,410 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=909412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:34:09,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4822, 1.6880, 1.3840, 1.4867], device='cuda:0'), covar=tensor([0.2813, 0.2769, 0.3137, 0.2360], device='cuda:0'), in_proj_covar=tensor([0.1484, 0.1081, 0.1315, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 16:34:16,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 16:34:25,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.156e+03 1.888e+03 2.542e+03 3.530e+03 8.276e+03, threshold=5.085e+03, percent-clipped=20.0
+2023-03-10 16:34:41,490 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 16:34:43,959 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=909448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:34:48,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=909451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:34:57,754 INFO [train.py:968] (0/2) Epoch 20, batch 42300, giga_loss[loss=0.318, simple_loss=0.3594, pruned_loss=0.1383, over 23463.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3732, pruned_loss=0.1217, over 5666943.29 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3642, pruned_loss=0.1171, over 5694668.51 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.374, pruned_loss=0.1221, over 5662551.99 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:35:14,757 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=909480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:35:45,397 INFO [train.py:968] (0/2) Epoch 20, batch 42350, giga_loss[loss=0.2833, simple_loss=0.3569, pruned_loss=0.1049, over 28978.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3716, pruned_loss=0.1209, over 5648284.13 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3647, pruned_loss=0.1175, over 5672622.30 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.372, pruned_loss=0.121, over 5663885.91 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:36:06,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.775e+02 1.837e+03 2.495e+03 3.879e+03 1.390e+04, threshold=4.990e+03, percent-clipped=14.0
+2023-03-10 16:36:35,407 INFO [train.py:968] (0/2) Epoch 20, batch 42400, giga_loss[loss=0.2732, simple_loss=0.3543, pruned_loss=0.09604, over 29027.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.371, pruned_loss=0.119, over 5664574.81 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3642, pruned_loss=0.1173, over 5676409.65 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3719, pruned_loss=0.1194, over 5673168.89 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:36:35,848 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-10 16:37:21,905 INFO [train.py:968] (0/2) Epoch 20, batch 42450, giga_loss[loss=0.3244, simple_loss=0.3922, pruned_loss=0.1284, over 28698.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3715, pruned_loss=0.1187, over 5670097.58 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3642, pruned_loss=0.1174, over 5680853.57 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3723, pruned_loss=0.1189, over 5673043.95 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:37:43,530 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.426e+02 1.479e+03 1.957e+03 2.610e+03 7.581e+03, threshold=3.914e+03, percent-clipped=3.0
+2023-03-10 16:38:09,615 INFO [train.py:968] (0/2) Epoch 20, batch 42500, giga_loss[loss=0.2954, simple_loss=0.3603, pruned_loss=0.1153, over 28879.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3701, pruned_loss=0.1178, over 5670838.56 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3642, pruned_loss=0.1174, over 5668313.44 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3709, pruned_loss=0.1179, over 5684521.70 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:38:53,891 INFO [train.py:968] (0/2) Epoch 20, batch 42550, giga_loss[loss=0.2845, simple_loss=0.3558, pruned_loss=0.1066, over 28815.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3686, pruned_loss=0.1174, over 5666058.82 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3639, pruned_loss=0.1173, over 5665141.89 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3697, pruned_loss=0.1176, over 5679939.98 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:39:13,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1064, 3.9391, 3.7537, 1.8850], device='cuda:0'), covar=tensor([0.0670, 0.0750, 0.0763, 0.2032], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.1134, 0.0961, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 16:39:15,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.586e+03 2.022e+03 2.688e+03 1.102e+04, threshold=4.043e+03, percent-clipped=13.0
+2023-03-10 16:39:46,437 INFO [train.py:968] (0/2) Epoch 20, batch 42600, giga_loss[loss=0.3222, simple_loss=0.3612, pruned_loss=0.1416, over 23473.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3686, pruned_loss=0.118, over 5666163.35 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3639, pruned_loss=0.1172, over 5668877.03 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3696, pruned_loss=0.1183, over 5673846.74 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:40:12,548 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=909787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:40:14,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5836, 4.4231, 4.2374, 1.9851], device='cuda:0'), covar=tensor([0.0572, 0.0705, 0.0721, 0.2053], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.1138, 0.0963, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 16:40:14,486 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 16:40:33,359 INFO [train.py:968] (0/2) Epoch 20, batch 42650, giga_loss[loss=0.3273, simple_loss=0.3836, pruned_loss=0.1355, over 27467.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3677, pruned_loss=0.1183, over 5665388.15 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3636, pruned_loss=0.1168, over 5669836.35 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3689, pruned_loss=0.1189, over 5670841.16 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:40:54,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.677e+03 2.124e+03 2.676e+03 5.319e+03, threshold=4.247e+03, percent-clipped=4.0
+2023-03-10 16:40:54,776 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=909832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:41:18,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4343, 2.0426, 1.5042, 0.6763], device='cuda:0'), covar=tensor([0.4534, 0.2538, 0.3878, 0.5655], device='cuda:0'), in_proj_covar=tensor([0.1735, 0.1640, 0.1593, 0.1408], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 16:41:19,972 INFO [train.py:968] (0/2) Epoch 20, batch 42700, giga_loss[loss=0.2606, simple_loss=0.3307, pruned_loss=0.09524, over 28718.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3664, pruned_loss=0.1182, over 5667062.04 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3631, pruned_loss=0.1164, over 5675337.28 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3679, pruned_loss=0.1191, over 5666436.78 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:42:00,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6530, 1.9235, 1.7783, 1.6660], device='cuda:0'), covar=tensor([0.1949, 0.2231, 0.2400, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0748, 0.0712, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 16:42:13,321 INFO [train.py:968] (0/2) Epoch 20, batch 42750, giga_loss[loss=0.2908, simple_loss=0.3571, pruned_loss=0.1122, over 29068.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3656, pruned_loss=0.1179, over 5672144.65 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3634, pruned_loss=0.1166, over 5674959.11 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3665, pruned_loss=0.1184, over 5671856.32 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:42:33,309 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=909930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:42:34,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.846e+03 2.383e+03 3.452e+03 1.028e+04, threshold=4.765e+03, percent-clipped=13.0
+2023-03-10 16:42:36,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=909933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:43:04,699 INFO [train.py:968] (0/2) Epoch 20, batch 42800, giga_loss[loss=0.2685, simple_loss=0.3383, pruned_loss=0.09933, over 28781.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3658, pruned_loss=0.1181, over 5676260.80 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3638, pruned_loss=0.1169, over 5674209.73 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3663, pruned_loss=0.1183, over 5676745.47 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:43:05,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=909962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:43:19,716 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0464, 1.9663, 1.4822, 1.7166], device='cuda:0'), covar=tensor([0.0921, 0.0784, 0.1058, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0449, 0.0520, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 16:43:42,455 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-910000.pt
+2023-03-10 16:43:53,490 INFO [train.py:968] (0/2) Epoch 20, batch 42850, giga_loss[loss=0.3473, simple_loss=0.3884, pruned_loss=0.1531, over 26708.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3664, pruned_loss=0.1175, over 5679381.29 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3635, pruned_loss=0.1166, over 5676619.53 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.367, pruned_loss=0.1179, over 5677776.70 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:44:16,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.660e+03 2.258e+03 3.118e+03 1.558e+04, threshold=4.515e+03, percent-clipped=9.0
+2023-03-10 16:44:40,433 INFO [train.py:968] (0/2) Epoch 20, batch 42900, libri_loss[loss=0.2798, simple_loss=0.357, pruned_loss=0.1013, over 29251.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3669, pruned_loss=0.1171, over 5685769.02 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.363, pruned_loss=0.1161, over 5682090.87 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.368, pruned_loss=0.1179, over 5679472.25 frames. ], batch size: 97, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:44:46,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2818, 2.2717, 2.1883, 2.0278], device='cuda:0'), covar=tensor([0.1664, 0.2186, 0.1934, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0748, 0.0712, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 16:44:58,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3221, 1.4836, 1.4395, 1.2513], device='cuda:0'), covar=tensor([0.2298, 0.2582, 0.1651, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.1967, 0.1885, 0.1828, 0.1963], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 16:45:25,293 INFO [train.py:968] (0/2) Epoch 20, batch 42950, libri_loss[loss=0.3015, simple_loss=0.3613, pruned_loss=0.1209, over 20029.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3668, pruned_loss=0.1166, over 5679907.99 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3623, pruned_loss=0.1155, over 5682084.55 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3685, pruned_loss=0.118, over 5675319.70 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:45:48,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.657e+03 2.173e+03 2.700e+03 8.844e+03, threshold=4.347e+03, percent-clipped=4.0
+2023-03-10 16:46:21,360 INFO [train.py:968] (0/2) Epoch 20, batch 43000, giga_loss[loss=0.3453, simple_loss=0.384, pruned_loss=0.1533, over 23792.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3669, pruned_loss=0.1172, over 5665206.76 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3621, pruned_loss=0.1153, over 5682500.06 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3686, pruned_loss=0.1185, over 5660857.74 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:46:34,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2902, 4.1039, 3.8919, 1.9096], device='cuda:0'), covar=tensor([0.0651, 0.0778, 0.0864, 0.1938], device='cuda:0'), in_proj_covar=tensor([0.1226, 0.1136, 0.0962, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 16:46:38,636 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=910179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:47:01,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3551, 3.1725, 3.0505, 1.3664], device='cuda:0'), covar=tensor([0.0894, 0.0986, 0.0933, 0.2204], device='cuda:0'), in_proj_covar=tensor([0.1226, 0.1136, 0.0961, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 16:47:02,172 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-10 16:47:04,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=910207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:47:09,560 INFO [train.py:968] (0/2) Epoch 20, batch 43050, giga_loss[loss=0.3621, simple_loss=0.4028, pruned_loss=0.1607, over 27962.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3687, pruned_loss=0.1194, over 5659622.81 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3614, pruned_loss=0.1148, over 5676749.80 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3708, pruned_loss=0.1209, over 5661609.79 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:47:12,615 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=910213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:47:33,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.668e+02 1.700e+03 2.321e+03 3.303e+03 9.802e+03, threshold=4.643e+03, percent-clipped=12.0
+2023-03-10 16:48:07,004 INFO [train.py:968] (0/2) Epoch 20, batch 43100, giga_loss[loss=0.2594, simple_loss=0.334, pruned_loss=0.09244, over 29043.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3711, pruned_loss=0.1227, over 5653686.71 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3613, pruned_loss=0.1147, over 5678425.46 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3729, pruned_loss=0.1241, over 5653575.85 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:48:56,384 INFO [train.py:968] (0/2) Epoch 20, batch 43150, giga_loss[loss=0.2905, simple_loss=0.3668, pruned_loss=0.1071, over 28909.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3725, pruned_loss=0.1248, over 5659446.31 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3618, pruned_loss=0.115, over 5684536.48 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3739, pruned_loss=0.1259, over 5653051.26 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:49:21,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.815e+03 2.459e+03 3.326e+03 6.583e+03, threshold=4.918e+03, percent-clipped=11.0
+2023-03-10 16:49:30,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6143, 1.8872, 1.3254, 1.4734], device='cuda:0'), covar=tensor([0.1007, 0.0635, 0.1069, 0.1167], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0449, 0.0520, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 16:49:40,256 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=910350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:49:42,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=910353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:49:50,671 INFO [train.py:968] (0/2) Epoch 20, batch 43200, giga_loss[loss=0.2555, simple_loss=0.3209, pruned_loss=0.09507, over 28627.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3722, pruned_loss=0.125, over 5664209.04 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3615, pruned_loss=0.1148, over 5685595.80 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3736, pruned_loss=0.1262, over 5658105.56 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:50:09,082 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=910382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:50:34,701 INFO [train.py:968] (0/2) Epoch 20, batch 43250, giga_loss[loss=0.3755, simple_loss=0.4102, pruned_loss=0.1704, over 26540.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3705, pruned_loss=0.1237, over 5663762.11 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3611, pruned_loss=0.1145, over 5679971.82 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3723, pruned_loss=0.1252, over 5662522.42 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:50:45,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2398, 1.5689, 1.5247, 1.0945], device='cuda:0'), covar=tensor([0.1471, 0.2585, 0.1341, 0.1594], device='cuda:0'), in_proj_covar=tensor([0.0890, 0.0700, 0.0936, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 16:50:54,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.686e+03 2.315e+03 3.087e+03 1.161e+04, threshold=4.630e+03, percent-clipped=10.0
+2023-03-10 16:51:22,966 INFO [train.py:968] (0/2) Epoch 20, batch 43300, giga_loss[loss=0.2844, simple_loss=0.3581, pruned_loss=0.1053, over 28641.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.369, pruned_loss=0.1209, over 5672311.61 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3612, pruned_loss=0.1145, over 5681110.97 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3704, pruned_loss=0.1221, over 5670355.17 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:52:13,975 INFO [train.py:968] (0/2) Epoch 20, batch 43350, giga_loss[loss=0.2651, simple_loss=0.3415, pruned_loss=0.0944, over 28688.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3667, pruned_loss=0.1189, over 5661332.04 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3613, pruned_loss=0.1147, over 5675497.92 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3678, pruned_loss=0.1197, over 5665284.14 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:52:33,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.710e+03 2.044e+03 2.884e+03 8.359e+03, threshold=4.088e+03, percent-clipped=7.0
+2023-03-10 16:52:39,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3461, 1.6149, 1.4021, 1.5678], device='cuda:0'), covar=tensor([0.0794, 0.0328, 0.0330, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0105], device='cuda:0')
+2023-03-10 16:52:53,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=910554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:53:00,302 INFO [train.py:968] (0/2) Epoch 20, batch 43400, giga_loss[loss=0.3034, simple_loss=0.3634, pruned_loss=0.1217, over 28268.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3653, pruned_loss=0.1183, over 5649760.70 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3615, pruned_loss=0.1147, over 5665515.65 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.366, pruned_loss=0.1189, over 5661237.82 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:53:24,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=910588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:53:29,630 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=910592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:53:47,494 INFO [train.py:968] (0/2) Epoch 20, batch 43450, giga_loss[loss=0.2934, simple_loss=0.3592, pruned_loss=0.1138, over 28929.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3648, pruned_loss=0.1188, over 5656650.01 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3617, pruned_loss=0.1148, over 5666349.81 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3653, pruned_loss=0.1193, over 5664774.07 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:53:58,830 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.85 vs. limit=2.0
+2023-03-10 16:54:09,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.758e+03 2.230e+03 2.861e+03 8.646e+03, threshold=4.461e+03, percent-clipped=5.0
+2023-03-10 16:54:32,437 INFO [train.py:968] (0/2) Epoch 20, batch 43500, giga_loss[loss=0.3037, simple_loss=0.3697, pruned_loss=0.1188, over 28418.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3651, pruned_loss=0.1189, over 5662557.75 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3615, pruned_loss=0.1146, over 5672280.50 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3657, pruned_loss=0.1197, over 5663300.38 frames. ], batch size: 78, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:54:44,309 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8331, 2.0934, 2.0994, 1.7091], device='cuda:0'), covar=tensor([0.2554, 0.2188, 0.2080, 0.2387], device='cuda:0'), in_proj_covar=tensor([0.1967, 0.1887, 0.1828, 0.1967], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 16:54:51,954 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=910682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:55:08,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=910697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:55:12,408 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=910700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:55:16,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5839, 2.2223, 1.5608, 0.8317], device='cuda:0'), covar=tensor([0.5658, 0.3116, 0.4097, 0.6524], device='cuda:0'), in_proj_covar=tensor([0.1729, 0.1637, 0.1585, 0.1403], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 16:55:19,668 INFO [train.py:968] (0/2) Epoch 20, batch 43550, libri_loss[loss=0.2703, simple_loss=0.3357, pruned_loss=0.1025, over 29582.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3675, pruned_loss=0.1198, over 5671596.48 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3613, pruned_loss=0.1144, over 5682760.28 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3685, pruned_loss=0.1209, over 5661698.01 frames. ], batch size: 75, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:55:37,694 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=910729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:55:38,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-10 16:55:39,139 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=910731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:55:40,873 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.683e+03 2.181e+03 2.955e+03 5.083e+03, threshold=4.362e+03, percent-clipped=5.0
+2023-03-10 16:55:41,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=910734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:56:08,664 INFO [train.py:968] (0/2) Epoch 20, batch 43600, giga_loss[loss=0.3027, simple_loss=0.3774, pruned_loss=0.1139, over 28856.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3703, pruned_loss=0.1189, over 5669066.27 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3613, pruned_loss=0.1144, over 5686297.18 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3713, pruned_loss=0.1198, over 5657557.42 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:56:10,589 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=910763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:56:38,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4664, 1.7779, 1.4317, 1.3565], device='cuda:0'), covar=tensor([0.2535, 0.2484, 0.2839, 0.2344], device='cuda:0'), in_proj_covar=tensor([0.1491, 0.1081, 0.1316, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 16:57:01,379 INFO [train.py:968] (0/2) Epoch 20, batch 43650, giga_loss[loss=0.2977, simple_loss=0.3654, pruned_loss=0.115, over 28556.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3725, pruned_loss=0.1195, over 5666156.06 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3611, pruned_loss=0.1144, over 5680943.91 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3737, pruned_loss=0.1205, over 5660685.36 frames. ], batch size: 78, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:57:23,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.680e+03 2.129e+03 2.838e+03 1.111e+04, threshold=4.258e+03, percent-clipped=8.0
+2023-03-10 16:57:47,481 INFO [train.py:968] (0/2) Epoch 20, batch 43700, giga_loss[loss=0.3082, simple_loss=0.3755, pruned_loss=0.1205, over 28728.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3745, pruned_loss=0.1214, over 5667542.12 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3609, pruned_loss=0.1144, over 5676709.36 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3762, pruned_loss=0.1223, over 5666689.80 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:58:37,874 INFO [train.py:968] (0/2) Epoch 20, batch 43750, giga_loss[loss=0.2702, simple_loss=0.3521, pruned_loss=0.09415, over 28879.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3769, pruned_loss=0.1238, over 5670899.15 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3613, pruned_loss=0.1147, over 5680691.45 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3783, pruned_loss=0.1245, over 5666378.64 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:58:53,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5183, 1.5506, 1.7146, 1.3307], device='cuda:0'), covar=tensor([0.1607, 0.2385, 0.1299, 0.1582], device='cuda:0'), in_proj_covar=tensor([0.0891, 0.0698, 0.0937, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 16:58:58,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.163e+03 1.754e+03 2.393e+03 3.265e+03 7.320e+03, threshold=4.787e+03, percent-clipped=10.0
+2023-03-10 16:59:23,293 INFO [train.py:968] (0/2) Epoch 20, batch 43800, giga_loss[loss=0.4123, simple_loss=0.4373, pruned_loss=0.1936, over 26582.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3762, pruned_loss=0.1241, over 5669507.75 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3617, pruned_loss=0.1149, over 5675566.92 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3772, pruned_loss=0.1246, over 5670368.92 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:59:27,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=910967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:59:33,499 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 16:59:46,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2875, 1.3598, 3.3681, 3.0718], device='cuda:0'), covar=tensor([0.1486, 0.2604, 0.0497, 0.1921], device='cuda:0'), in_proj_covar=tensor([0.0755, 0.0645, 0.0958, 0.0901], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 17:00:13,393 INFO [train.py:968] (0/2) Epoch 20, batch 43850, giga_loss[loss=0.2653, simple_loss=0.345, pruned_loss=0.0928, over 28952.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3742, pruned_loss=0.1234, over 5664385.55 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3613, pruned_loss=0.1146, over 5680019.66 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3757, pruned_loss=0.1243, over 5660838.34 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:00:35,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.937e+02 1.691e+03 2.090e+03 2.969e+03 6.834e+03, threshold=4.179e+03, percent-clipped=4.0
+2023-03-10 17:00:57,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:01:01,051 INFO [train.py:968] (0/2) Epoch 20, batch 43900, giga_loss[loss=0.3987, simple_loss=0.424, pruned_loss=0.1867, over 27663.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3734, pruned_loss=0.124, over 5657903.17 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3615, pruned_loss=0.1147, over 5675789.67 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3747, pruned_loss=0.1249, over 5658909.56 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:01:22,590 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-10 17:01:41,829 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 17:01:50,245 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=911110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:01:50,585 INFO [train.py:968] (0/2) Epoch 20, batch 43950, giga_loss[loss=0.2859, simple_loss=0.3531, pruned_loss=0.1094, over 28677.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3725, pruned_loss=0.1238, over 5653113.19 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3619, pruned_loss=0.115, over 5671534.58 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3735, pruned_loss=0.1245, over 5657349.27 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:01:56,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=911113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:02:13,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=911130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:02:16,266 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.997e+03 2.588e+03 4.748e+03 7.735e+03, threshold=5.176e+03, percent-clipped=29.0
+2023-03-10 17:02:24,611 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=911142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:02:41,586 INFO [train.py:968] (0/2) Epoch 20, batch 44000, giga_loss[loss=0.2763, simple_loss=0.3466, pruned_loss=0.103, over 28769.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3726, pruned_loss=0.1241, over 5643921.05 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1152, over 5669640.30 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3734, pruned_loss=0.1247, over 5648487.51 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 17:03:19,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=911200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:03:22,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=911203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:03:29,619 INFO [train.py:968] (0/2) Epoch 20, batch 44050, giga_loss[loss=0.332, simple_loss=0.3909, pruned_loss=0.1366, over 28643.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3697, pruned_loss=0.1227, over 5651567.65 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3617, pruned_loss=0.1149, over 5675642.00 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.371, pruned_loss=0.1237, over 5648789.62 frames. ], batch size: 336, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 17:03:47,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=911232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:03:50,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.698e+03 2.164e+03 3.056e+03 1.132e+04, threshold=4.328e+03, percent-clipped=4.0
+2023-03-10 17:04:16,659 INFO [train.py:968] (0/2) Epoch 20, batch 44100, giga_loss[loss=0.302, simple_loss=0.3664, pruned_loss=0.1187, over 28631.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3681, pruned_loss=0.1216, over 5665636.44 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3617, pruned_loss=0.115, over 5679239.39 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3692, pruned_loss=0.1224, over 5659925.56 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:05:06,081 INFO [train.py:968] (0/2) Epoch 20, batch 44150, giga_loss[loss=0.2927, simple_loss=0.3651, pruned_loss=0.1101, over 28735.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3675, pruned_loss=0.1209, over 5661330.98 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3618, pruned_loss=0.115, over 5680362.19 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3684, pruned_loss=0.1216, over 5655781.34 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:05:30,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.618e+03 2.002e+03 2.982e+03 1.907e+04, threshold=4.004e+03, percent-clipped=7.0
+2023-03-10 17:06:01,524 INFO [train.py:968] (0/2) Epoch 20, batch 44200, giga_loss[loss=0.319, simple_loss=0.3849, pruned_loss=0.1265, over 28904.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.371, pruned_loss=0.1231, over 5645949.50 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.362, pruned_loss=0.1152, over 5674388.33 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3716, pruned_loss=0.1235, over 5646427.04 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:06:49,709 INFO [train.py:968] (0/2) Epoch 20, batch 44250, giga_loss[loss=0.3171, simple_loss=0.3723, pruned_loss=0.1309, over 28706.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3731, pruned_loss=0.1248, over 5649831.33 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3623, pruned_loss=0.1153, over 5675693.01 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3734, pruned_loss=0.1251, over 5648746.75 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:06:51,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=911413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:07:16,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.134e+03 1.650e+03 2.181e+03 3.154e+03 7.467e+03, threshold=4.362e+03, percent-clipped=11.0
+2023-03-10 17:07:40,555 INFO [train.py:968] (0/2) Epoch 20, batch 44300, giga_loss[loss=0.3089, simple_loss=0.3871, pruned_loss=0.1153, over 28928.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3727, pruned_loss=0.1234, over 5661483.39 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3624, pruned_loss=0.1154, over 5678055.30 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.373, pruned_loss=0.1237, over 5658308.53 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:07:45,132 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-10 17:07:52,984 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=911473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:08:06,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6170, 1.7764, 1.4015, 1.5982], device='cuda:0'), covar=tensor([0.2987, 0.2929, 0.3580, 0.2401], device='cuda:0'), in_proj_covar=tensor([0.1487, 0.1078, 0.1314, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 17:08:22,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:08:27,248 INFO [train.py:968] (0/2) Epoch 20, batch 44350, libri_loss[loss=0.3191, simple_loss=0.3793, pruned_loss=0.1294, over 25653.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3727, pruned_loss=0.1205, over 5664002.90 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3625, pruned_loss=0.1155, over 5679071.25 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3731, pruned_loss=0.1208, over 5660325.70 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:08:47,177 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.348e+02 1.506e+03 2.081e+03 2.657e+03 8.029e+03, threshold=4.162e+03, percent-clipped=8.0
+2023-03-10 17:09:06,525 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=911555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:09:13,012 INFO [train.py:968] (0/2) Epoch 20, batch 44400, giga_loss[loss=0.3305, simple_loss=0.4051, pruned_loss=0.128, over 28764.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3749, pruned_loss=0.1203, over 5672348.58 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3621, pruned_loss=0.1153, over 5683197.59 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3758, pruned_loss=0.1208, over 5665629.28 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:09:43,703 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 17:10:02,776 INFO [train.py:968] (0/2) Epoch 20, batch 44450, giga_loss[loss=0.3936, simple_loss=0.4318, pruned_loss=0.1777, over 28597.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3786, pruned_loss=0.1238, over 5650909.68 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3626, pruned_loss=0.1155, over 5676043.24 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3792, pruned_loss=0.1242, over 5652191.34 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:10:10,013 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=911617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:10:27,927 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.740e+02 1.765e+03 2.255e+03 3.249e+03 9.842e+03, threshold=4.511e+03, percent-clipped=11.0
+2023-03-10 17:10:38,302 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=911648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:10:42,643 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=911651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:10:51,713 INFO [train.py:968] (0/2) Epoch 20, batch 44500, libri_loss[loss=0.2585, simple_loss=0.3329, pruned_loss=0.09202, over 29572.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3803, pruned_loss=0.1264, over 5653025.60 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3622, pruned_loss=0.1151, over 5682329.12 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3819, pruned_loss=0.1275, over 5647589.61 frames. ], batch size: 76, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:11:10,525 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=911680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:11:11,567 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-10 17:11:28,993 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-10 17:11:43,594 INFO [train.py:968] (0/2) Epoch 20, batch 44550, giga_loss[loss=0.2965, simple_loss=0.3699, pruned_loss=0.1115, over 28847.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3805, pruned_loss=0.1271, over 5667000.98 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.362, pruned_loss=0.1149, over 5683441.65 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3823, pruned_loss=0.1283, over 5661446.01 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:12:08,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 1.650e+03 2.315e+03 3.027e+03 5.705e+03, threshold=4.630e+03, percent-clipped=1.0
+2023-03-10 17:12:30,074 INFO [train.py:968] (0/2) Epoch 20, batch 44600, giga_loss[loss=0.3002, simple_loss=0.3702, pruned_loss=0.1151, over 27896.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3785, pruned_loss=0.1259, over 5669051.31 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1146, over 5687518.50 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3805, pruned_loss=0.1273, over 5660685.49 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:12:51,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:13:09,945 INFO [train.py:968] (0/2) Epoch 20, batch 44650, libri_loss[loss=0.3385, simple_loss=0.3943, pruned_loss=0.1413, over 29516.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3772, pruned_loss=0.1244, over 5661380.22 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3623, pruned_loss=0.1153, over 5675285.80 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3791, pruned_loss=0.1255, over 5663277.76 frames. ], batch size: 84, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:13:16,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4814, 1.8192, 1.7427, 1.5789], device='cuda:0'), covar=tensor([0.1801, 0.1803, 0.2165, 0.2048], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0747, 0.0710, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 17:13:17,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3355, 1.6131, 1.5695, 1.2269], device='cuda:0'), covar=tensor([0.3114, 0.2715, 0.2049, 0.2587], device='cuda:0'), in_proj_covar=tensor([0.1963, 0.1874, 0.1824, 0.1960], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 17:13:34,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.681e+03 2.152e+03 3.029e+03 9.256e+03, threshold=4.305e+03, percent-clipped=5.0
+2023-03-10 17:13:42,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:13:56,235 INFO [train.py:968] (0/2) Epoch 20, batch 44700, giga_loss[loss=0.2932, simple_loss=0.3728, pruned_loss=0.1068, over 29057.00 frames. ], tot_loss[loss=0.311, simple_loss=0.377, pruned_loss=0.1225, over 5674640.59 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3616, pruned_loss=0.115, over 5681825.66 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3795, pruned_loss=0.1239, over 5670180.48 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:14:29,667 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-10 17:14:44,073 INFO [train.py:968] (0/2) Epoch 20, batch 44750, giga_loss[loss=0.3642, simple_loss=0.4088, pruned_loss=0.1598, over 27637.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3775, pruned_loss=0.122, over 5683081.40 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3614, pruned_loss=0.1148, over 5683009.88 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.38, pruned_loss=0.1234, over 5678469.34 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:15:06,156 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4776, 3.9220, 1.6175, 1.7375], device='cuda:0'), covar=tensor([0.0934, 0.0397, 0.0875, 0.1234], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0558, 0.0383, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 17:15:06,167 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:15:06,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=911931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:15:11,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=911934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:15:13,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.556e+03 1.940e+03 2.893e+03 9.350e+03, threshold=3.880e+03, percent-clipped=11.0
+2023-03-10 17:15:37,891 INFO [train.py:968] (0/2) Epoch 20, batch 44800, giga_loss[loss=0.285, simple_loss=0.3642, pruned_loss=0.1029, over 28956.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3785, pruned_loss=0.1237, over 5670144.36 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3612, pruned_loss=0.1147, over 5688318.44 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.381, pruned_loss=0.1251, over 5661527.37 frames. ], batch size: 164, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:15:42,610 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=911963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:16:08,360 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=911991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:16:09,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:16:10,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=911994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:16:15,952 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-912000.pt
+2023-03-10 17:16:25,917 INFO [train.py:968] (0/2) Epoch 20, batch 44850, giga_loss[loss=0.2635, simple_loss=0.336, pruned_loss=0.09548, over 28795.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3765, pruned_loss=0.123, over 5657346.32 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.361, pruned_loss=0.1146, over 5682172.70 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3788, pruned_loss=0.1243, over 5655345.29 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:16:38,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=912023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:16:42,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5885, 1.8204, 1.4904, 1.8505], device='cuda:0'), covar=tensor([0.2532, 0.2633, 0.2842, 0.2454], device='cuda:0'), in_proj_covar=tensor([0.1486, 0.1077, 0.1315, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 17:16:49,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.515e+03 2.076e+03 2.809e+03 9.296e+03, threshold=4.153e+03, percent-clipped=12.0
+2023-03-10 17:17:12,886 INFO [train.py:968] (0/2) Epoch 20, batch 44900, giga_loss[loss=0.3209, simple_loss=0.3788, pruned_loss=0.1315, over 27916.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3749, pruned_loss=0.1229, over 5669275.09 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3612, pruned_loss=0.1145, over 5687139.10 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3771, pruned_loss=0.1242, over 5662877.94 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:17:27,665 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=912073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:17:29,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=912076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:17:57,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=912105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:18:02,295 INFO [train.py:968] (0/2) Epoch 20, batch 44950, giga_loss[loss=0.26, simple_loss=0.3326, pruned_loss=0.09368, over 28520.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3741, pruned_loss=0.1234, over 5664540.72 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3614, pruned_loss=0.1147, over 5691382.38 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3759, pruned_loss=0.1244, over 5655261.52 frames. ], batch size: 78, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:18:03,907 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-10 17:18:26,380 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=912135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:18:27,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.805e+03 2.275e+03 3.293e+03 8.026e+03, threshold=4.550e+03, percent-clipped=16.0
+2023-03-10 17:18:29,200 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=912138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:18:53,908 INFO [train.py:968] (0/2) Epoch 20, batch 45000, giga_loss[loss=0.3263, simple_loss=0.3842, pruned_loss=0.1342, over 28933.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3728, pruned_loss=0.1228, over 5652946.36 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.362, pruned_loss=0.1151, over 5681612.82 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3739, pruned_loss=0.1234, over 5654157.35 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:18:53,914 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 17:19:02,810 INFO [train.py:1012] (0/2) Epoch 20, validation: loss=0.2075, simple_loss=0.3165, pruned_loss=0.04919, over 944034.00 frames. 
+2023-03-10 17:19:02,811 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 17:19:07,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=912167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:19:34,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5137, 4.3566, 4.1442, 2.0432], device='cuda:0'), covar=tensor([0.0605, 0.0730, 0.0758, 0.1881], device='cuda:0'), in_proj_covar=tensor([0.1232, 0.1147, 0.0970, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 17:19:45,524 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-10 17:19:48,474 INFO [train.py:968] (0/2) Epoch 20, batch 45050, libri_loss[loss=0.3326, simple_loss=0.3884, pruned_loss=0.1383, over 19209.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3744, pruned_loss=0.1253, over 5609426.69 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3627, pruned_loss=0.1157, over 5638392.14 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3748, pruned_loss=0.1255, over 5650687.31 frames. ], batch size: 187, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:20:11,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.292e+03 2.135e+03 2.881e+03 3.856e+03 8.256e+03, threshold=5.763e+03, percent-clipped=19.0
+2023-03-10 17:20:34,331 INFO [train.py:968] (0/2) Epoch 20, batch 45100, giga_loss[loss=0.3136, simple_loss=0.3734, pruned_loss=0.1269, over 29001.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3737, pruned_loss=0.1251, over 5573625.96 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3634, pruned_loss=0.1163, over 5594867.22 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3737, pruned_loss=0.1249, over 5642957.70 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:21:11,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0676, 1.2410, 3.3613, 3.0497], device='cuda:0'), covar=tensor([0.1803, 0.2790, 0.0512, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0756, 0.0648, 0.0960, 0.0904], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 17:21:19,585 INFO [train.py:968] (0/2) Epoch 20, batch 45150, giga_loss[loss=0.2884, simple_loss=0.3696, pruned_loss=0.1036, over 28663.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3705, pruned_loss=0.121, over 5536684.33 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3645, pruned_loss=0.1173, over 5529985.13 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3698, pruned_loss=0.1202, over 5650315.92 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:21:43,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.455e+02 1.350e+03 1.907e+03 2.807e+03 1.538e+04, threshold=3.813e+03, percent-clipped=1.0
+2023-03-10 17:21:59,594 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-10 17:22:02,846 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-20.pt
+2023-03-10 17:23:23,373 INFO [train.py:968] (0/2) Epoch 21, batch 50, giga_loss[loss=0.2738, simple_loss=0.3633, pruned_loss=0.09214, over 28851.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.365, pruned_loss=0.1048, over 1265977.00 frames. ], libri_tot_loss[loss=0.2428, simple_loss=0.3198, pruned_loss=0.08288, over 202269.78 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3723, pruned_loss=0.1083, over 1102653.71 frames. ], batch size: 174, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:24:00,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.229e+02 1.292e+03 1.678e+03 2.218e+03 1.177e+04, threshold=3.357e+03, percent-clipped=8.0
+2023-03-10 17:24:14,475 INFO [train.py:968] (0/2) Epoch 21, batch 100, giga_loss[loss=0.231, simple_loss=0.3082, pruned_loss=0.07694, over 28422.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3594, pruned_loss=0.1017, over 2247425.03 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3321, pruned_loss=0.0885, over 341536.75 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3629, pruned_loss=0.1034, over 2026738.42 frames. ], batch size: 71, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:24:34,845 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 17:24:45,874 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-10 17:24:50,082 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=912489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:25:02,758 INFO [train.py:968] (0/2) Epoch 21, batch 150, giga_loss[loss=0.2731, simple_loss=0.3305, pruned_loss=0.1078, over 26635.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3467, pruned_loss=0.09598, over 3011170.88 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.337, pruned_loss=0.08893, over 424136.46 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3478, pruned_loss=0.09683, over 2793708.84 frames. ], batch size: 555, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:25:21,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4388, 1.7660, 1.4171, 1.5571], device='cuda:0'), covar=tensor([0.0745, 0.0353, 0.0337, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 17:25:33,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.494e+02 1.174e+03 1.485e+03 2.122e+03 4.749e+03, threshold=2.969e+03, percent-clipped=14.0
+2023-03-10 17:25:44,341 INFO [train.py:968] (0/2) Epoch 21, batch 200, giga_loss[loss=0.2022, simple_loss=0.2872, pruned_loss=0.05867, over 28935.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3348, pruned_loss=0.08995, over 3616392.01 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3388, pruned_loss=0.08788, over 636977.95 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3347, pruned_loss=0.09058, over 3349715.38 frames. ], batch size: 164, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:26:27,248 INFO [train.py:968] (0/2) Epoch 21, batch 250, giga_loss[loss=0.2194, simple_loss=0.2967, pruned_loss=0.07103, over 28588.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.325, pruned_loss=0.08525, over 4080879.62 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3428, pruned_loss=0.0895, over 739348.04 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3233, pruned_loss=0.0851, over 3834660.44 frames. ], batch size: 307, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:27:00,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.437e+02 1.051e+03 1.394e+03 1.990e+03 4.190e+03, threshold=2.788e+03, percent-clipped=5.0
+2023-03-10 17:27:12,212 INFO [train.py:968] (0/2) Epoch 21, batch 300, giga_loss[loss=0.1886, simple_loss=0.2728, pruned_loss=0.05216, over 29085.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3165, pruned_loss=0.08151, over 4443092.49 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3437, pruned_loss=0.08964, over 842063.43 frames. ], giga_tot_loss[loss=0.2381, simple_loss=0.3141, pruned_loss=0.08102, over 4218121.06 frames. ], batch size: 155, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:27:45,063 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=912684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:27:59,556 INFO [train.py:968] (0/2) Epoch 21, batch 350, giga_loss[loss=0.1974, simple_loss=0.2715, pruned_loss=0.06164, over 28510.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.31, pruned_loss=0.0789, over 4725458.49 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3418, pruned_loss=0.08878, over 990909.21 frames. ], giga_tot_loss[loss=0.2318, simple_loss=0.3071, pruned_loss=0.0782, over 4510549.28 frames. ], batch size: 85, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:28:29,187 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.524e+02 1.035e+03 1.274e+03 1.737e+03 4.175e+03, threshold=2.547e+03, percent-clipped=7.0
+2023-03-10 17:28:38,619 INFO [train.py:968] (0/2) Epoch 21, batch 400, giga_loss[loss=0.2343, simple_loss=0.3043, pruned_loss=0.0822, over 28968.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3068, pruned_loss=0.07769, over 4948021.47 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3423, pruned_loss=0.08933, over 1133175.31 frames. ], giga_tot_loss[loss=0.2283, simple_loss=0.3032, pruned_loss=0.07665, over 4749876.49 frames. ], batch size: 106, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:28:42,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3975, 4.2098, 3.9959, 2.1311], device='cuda:0'), covar=tensor([0.0469, 0.0676, 0.0721, 0.2192], device='cuda:0'), in_proj_covar=tensor([0.1211, 0.1125, 0.0953, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-10 17:29:20,858 INFO [train.py:968] (0/2) Epoch 21, batch 450, giga_loss[loss=0.2474, simple_loss=0.3105, pruned_loss=0.09216, over 28480.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3054, pruned_loss=0.07695, over 5122278.64 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3426, pruned_loss=0.08982, over 1296045.63 frames. ], giga_tot_loss[loss=0.2261, simple_loss=0.3011, pruned_loss=0.07552, over 4936712.82 frames. ], batch size: 78, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:29:55,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.314e+02 1.052e+03 1.373e+03 1.951e+03 6.552e+03, threshold=2.746e+03, percent-clipped=13.0
+2023-03-10 17:30:05,756 INFO [train.py:968] (0/2) Epoch 21, batch 500, giga_loss[loss=0.1793, simple_loss=0.2576, pruned_loss=0.05051, over 28462.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3029, pruned_loss=0.07571, over 5253524.19 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3441, pruned_loss=0.09061, over 1405093.15 frames. ], giga_tot_loss[loss=0.2231, simple_loss=0.2982, pruned_loss=0.07401, over 5094301.31 frames. ], batch size: 85, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:30:19,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=912864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:30:53,169 INFO [train.py:968] (0/2) Epoch 21, batch 550, giga_loss[loss=0.2185, simple_loss=0.3017, pruned_loss=0.06769, over 28671.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3014, pruned_loss=0.0753, over 5351722.06 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3443, pruned_loss=0.09043, over 1493179.22 frames. ], giga_tot_loss[loss=0.2221, simple_loss=0.2968, pruned_loss=0.07371, over 5215630.04 frames. ], batch size: 242, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:30:58,516 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=912908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:31:25,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.474e+02 1.049e+03 1.205e+03 1.495e+03 3.253e+03, threshold=2.409e+03, percent-clipped=1.0
+2023-03-10 17:31:36,570 INFO [train.py:968] (0/2) Epoch 21, batch 600, giga_loss[loss=0.2242, simple_loss=0.2813, pruned_loss=0.08359, over 23826.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2996, pruned_loss=0.07452, over 5431101.76 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3407, pruned_loss=0.08861, over 1643075.07 frames. ], giga_tot_loss[loss=0.2209, simple_loss=0.2953, pruned_loss=0.07322, over 5307320.94 frames. ], batch size: 705, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:31:56,963 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3980, 1.8082, 1.2845, 0.7976], device='cuda:0'), covar=tensor([0.6425, 0.2984, 0.3408, 0.6674], device='cuda:0'), in_proj_covar=tensor([0.1720, 0.1622, 0.1574, 0.1395], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 17:32:23,786 INFO [train.py:968] (0/2) Epoch 21, batch 650, giga_loss[loss=0.2004, simple_loss=0.2776, pruned_loss=0.06157, over 28858.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2987, pruned_loss=0.0744, over 5474562.72 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3417, pruned_loss=0.08934, over 1755344.96 frames. ], giga_tot_loss[loss=0.2197, simple_loss=0.2939, pruned_loss=0.07277, over 5376257.41 frames. ], batch size: 112, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:32:32,638 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:32:34,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:32:56,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.520e+02 9.654e+02 1.179e+03 1.726e+03 5.414e+03, threshold=2.359e+03, percent-clipped=11.0
+2023-03-10 17:32:57,820 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=913039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:33:06,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2528, 1.3277, 3.5290, 3.1145], device='cuda:0'), covar=tensor([0.1622, 0.2724, 0.0505, 0.1159], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0639, 0.0949, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 17:33:09,486 INFO [train.py:968] (0/2) Epoch 21, batch 700, giga_loss[loss=0.2412, simple_loss=0.3032, pruned_loss=0.08956, over 26581.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2965, pruned_loss=0.07333, over 5525334.99 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3418, pruned_loss=0.08922, over 1874958.85 frames. ], giga_tot_loss[loss=0.2173, simple_loss=0.2914, pruned_loss=0.07163, over 5438797.86 frames. ], batch size: 555, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:33:19,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=913059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:33:57,766 INFO [train.py:968] (0/2) Epoch 21, batch 750, giga_loss[loss=0.1937, simple_loss=0.2656, pruned_loss=0.06085, over 28744.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2934, pruned_loss=0.07168, over 5563644.61 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3425, pruned_loss=0.08961, over 1924745.58 frames. ], giga_tot_loss[loss=0.2143, simple_loss=0.2886, pruned_loss=0.07, over 5499108.79 frames. ], batch size: 119, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:34:30,475 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.646e+02 9.493e+02 1.269e+03 1.630e+03 3.635e+03, threshold=2.538e+03, percent-clipped=6.0
+2023-03-10 17:34:42,698 INFO [train.py:968] (0/2) Epoch 21, batch 800, giga_loss[loss=0.193, simple_loss=0.2638, pruned_loss=0.06109, over 28612.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2899, pruned_loss=0.07031, over 5579178.95 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3428, pruned_loss=0.09008, over 1965753.15 frames. ], giga_tot_loss[loss=0.2114, simple_loss=0.2854, pruned_loss=0.06865, over 5538902.59 frames. ], batch size: 85, lr: 1.54e-03, grad_scale: 8.0
+2023-03-10 17:35:31,611 INFO [train.py:968] (0/2) Epoch 21, batch 850, giga_loss[loss=0.2587, simple_loss=0.3276, pruned_loss=0.09487, over 28977.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2977, pruned_loss=0.07483, over 5595400.22 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3416, pruned_loss=0.08946, over 2079108.85 frames. ], giga_tot_loss[loss=0.2198, simple_loss=0.2931, pruned_loss=0.07322, over 5557462.13 frames. ], batch size: 213, lr: 1.54e-03, grad_scale: 8.0
+2023-03-10 17:35:32,556 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:35:34,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:36:00,457 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=913230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:36:03,764 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=913234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:36:07,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.465e+02 1.236e+03 1.509e+03 1.865e+03 3.885e+03, threshold=3.017e+03, percent-clipped=5.0
+2023-03-10 17:36:23,808 INFO [train.py:968] (0/2) Epoch 21, batch 900, giga_loss[loss=0.2717, simple_loss=0.3513, pruned_loss=0.09607, over 28827.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3108, pruned_loss=0.08132, over 5617104.23 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3418, pruned_loss=0.08947, over 2136354.97 frames. ], giga_tot_loss[loss=0.2333, simple_loss=0.3067, pruned_loss=0.07992, over 5583068.73 frames. ], batch size: 174, lr: 1.54e-03, grad_scale: 8.0
+2023-03-10 17:36:53,488 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=913283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:37:08,420 INFO [train.py:968] (0/2) Epoch 21, batch 950, giga_loss[loss=0.2465, simple_loss=0.3276, pruned_loss=0.08272, over 28607.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3222, pruned_loss=0.08676, over 5641800.41 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3422, pruned_loss=0.08969, over 2227571.36 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3183, pruned_loss=0.08547, over 5610320.80 frames. ], batch size: 60, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:37:23,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=913317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:37:40,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.861e+02 1.288e+03 1.554e+03 2.020e+03 8.380e+03, threshold=3.109e+03, percent-clipped=8.0
+2023-03-10 17:37:53,785 INFO [train.py:968] (0/2) Epoch 21, batch 1000, giga_loss[loss=0.3511, simple_loss=0.4117, pruned_loss=0.1453, over 28771.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3307, pruned_loss=0.09048, over 5659521.70 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3417, pruned_loss=0.08952, over 2300726.31 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3275, pruned_loss=0.0895, over 5630018.72 frames. ], batch size: 199, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:38:33,112 INFO [train.py:968] (0/2) Epoch 21, batch 1050, giga_loss[loss=0.2745, simple_loss=0.3551, pruned_loss=0.09694, over 28876.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3339, pruned_loss=0.09033, over 5675572.64 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3413, pruned_loss=0.08905, over 2389819.48 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3312, pruned_loss=0.08975, over 5646716.59 frames. ], batch size: 112, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:39:00,668 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:39:05,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:39:14,902 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.430e+02 1.105e+03 1.412e+03 1.962e+03 5.642e+03, threshold=2.825e+03, percent-clipped=6.0
+2023-03-10 17:39:24,282 INFO [train.py:968] (0/2) Epoch 21, batch 1100, giga_loss[loss=0.3202, simple_loss=0.379, pruned_loss=0.1307, over 26607.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.336, pruned_loss=0.09058, over 5668553.61 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3412, pruned_loss=0.08886, over 2440636.15 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3339, pruned_loss=0.09023, over 5644083.83 frames. ], batch size: 555, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:39:29,332 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=913458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:39:41,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=913472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:40:05,593 INFO [train.py:968] (0/2) Epoch 21, batch 1150, giga_loss[loss=0.2569, simple_loss=0.3386, pruned_loss=0.08763, over 28868.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3379, pruned_loss=0.09162, over 5673231.58 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3412, pruned_loss=0.08907, over 2556582.57 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.336, pruned_loss=0.09132, over 5663357.04 frames. ], batch size: 174, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:40:39,535 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.096e+02 1.167e+03 1.431e+03 1.994e+03 4.195e+03, threshold=2.863e+03, percent-clipped=11.0
+2023-03-10 17:40:48,779 INFO [train.py:968] (0/2) Epoch 21, batch 1200, giga_loss[loss=0.2766, simple_loss=0.3523, pruned_loss=0.1005, over 28331.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3404, pruned_loss=0.0937, over 5665796.45 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3406, pruned_loss=0.08874, over 2678475.17 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3391, pruned_loss=0.09373, over 5657569.45 frames. ], batch size: 368, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:41:35,248 INFO [train.py:968] (0/2) Epoch 21, batch 1250, giga_loss[loss=0.2773, simple_loss=0.3489, pruned_loss=0.1029, over 28513.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3433, pruned_loss=0.09584, over 5667544.19 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3402, pruned_loss=0.08857, over 2726159.31 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3424, pruned_loss=0.09603, over 5658258.91 frames. ], batch size: 85, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:41:35,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5279, 4.4281, 1.7492, 1.7651], device='cuda:0'), covar=tensor([0.1052, 0.0201, 0.0898, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0549, 0.0381, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 17:41:38,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=913605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:41:42,284 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-10 17:42:12,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.350e+02 1.182e+03 1.469e+03 1.862e+03 3.661e+03, threshold=2.938e+03, percent-clipped=6.0
+2023-03-10 17:42:19,419 INFO [train.py:968] (0/2) Epoch 21, batch 1300, giga_loss[loss=0.2578, simple_loss=0.3447, pruned_loss=0.08547, over 28716.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.346, pruned_loss=0.09671, over 5677011.63 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3405, pruned_loss=0.08867, over 2835925.81 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3453, pruned_loss=0.09705, over 5662150.46 frames. ], batch size: 66, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:42:40,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3401, 1.8816, 1.6258, 1.5621], device='cuda:0'), covar=tensor([0.0837, 0.0300, 0.0327, 0.0907], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 17:42:55,598 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=913692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:43:01,323 INFO [train.py:968] (0/2) Epoch 21, batch 1350, giga_loss[loss=0.2925, simple_loss=0.38, pruned_loss=0.1025, over 28695.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3475, pruned_loss=0.09614, over 5693965.05 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3408, pruned_loss=0.08875, over 2923158.11 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.347, pruned_loss=0.09658, over 5679683.88 frames. ], batch size: 60, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:43:36,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.211e+02 1.214e+03 1.520e+03 2.034e+03 5.280e+03, threshold=3.040e+03, percent-clipped=8.0
+2023-03-10 17:43:37,473 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 17:43:44,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:43:46,611 INFO [train.py:968] (0/2) Epoch 21, batch 1400, giga_loss[loss=0.2541, simple_loss=0.3443, pruned_loss=0.08197, over 29013.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3495, pruned_loss=0.09682, over 5681700.62 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3407, pruned_loss=0.08862, over 2984600.33 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3493, pruned_loss=0.0974, over 5676660.30 frames. ], batch size: 164, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:43:46,972 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:43:58,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7310, 1.1009, 2.8764, 2.5836], device='cuda:0'), covar=tensor([0.1809, 0.2681, 0.0580, 0.1017], device='cuda:0'), in_proj_covar=tensor([0.0747, 0.0639, 0.0943, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 17:44:11,946 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=913780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:44:28,777 INFO [train.py:968] (0/2) Epoch 21, batch 1450, giga_loss[loss=0.2332, simple_loss=0.3237, pruned_loss=0.07135, over 28417.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3495, pruned_loss=0.09589, over 5678134.52 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3401, pruned_loss=0.08846, over 3018015.55 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3498, pruned_loss=0.09654, over 5680735.66 frames. ], batch size: 71, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:44:57,708 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:44:59,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:45:00,970 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.774e+02 1.198e+03 1.443e+03 1.852e+03 8.741e+03, threshold=2.886e+03, percent-clipped=6.0
+2023-03-10 17:45:05,500 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-10 17:45:07,164 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=913847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:45:09,747 INFO [train.py:968] (0/2) Epoch 21, batch 1500, giga_loss[loss=0.2453, simple_loss=0.3373, pruned_loss=0.07663, over 28517.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3484, pruned_loss=0.09402, over 5692915.58 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3412, pruned_loss=0.08909, over 3103404.88 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3483, pruned_loss=0.0944, over 5689237.55 frames. ], batch size: 60, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:45:23,386 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=913867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:45:48,193 INFO [train.py:968] (0/2) Epoch 21, batch 1550, giga_loss[loss=0.2504, simple_loss=0.3298, pruned_loss=0.08548, over 28806.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.346, pruned_loss=0.09214, over 5703587.65 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3404, pruned_loss=0.0888, over 3212490.33 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3465, pruned_loss=0.09272, over 5694167.90 frames. ], batch size: 99, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:45:52,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4102, 1.6368, 1.3166, 1.3020], device='cuda:0'), covar=tensor([0.2910, 0.2897, 0.3243, 0.2545], device='cuda:0'), in_proj_covar=tensor([0.1494, 0.1079, 0.1318, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 17:46:03,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3914, 1.0784, 4.0397, 3.3231], device='cuda:0'), covar=tensor([0.1549, 0.2927, 0.0428, 0.0938], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0641, 0.0944, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 17:46:11,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7520, 1.8015, 1.5366, 1.9666], device='cuda:0'), covar=tensor([0.2712, 0.2779, 0.3068, 0.2323], device='cuda:0'), in_proj_covar=tensor([0.1492, 0.1079, 0.1317, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 17:46:24,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.847e+02 1.109e+03 1.420e+03 1.907e+03 5.902e+03, threshold=2.840e+03, percent-clipped=6.0
+2023-03-10 17:46:38,958 INFO [train.py:968] (0/2) Epoch 21, batch 1600, giga_loss[loss=0.237, simple_loss=0.3163, pruned_loss=0.07885, over 28799.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3469, pruned_loss=0.09351, over 5709340.98 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3408, pruned_loss=0.08881, over 3278252.60 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3472, pruned_loss=0.09405, over 5698175.92 frames. ], batch size: 60, lr: 1.54e-03, grad_scale: 8.0
+2023-03-10 17:47:09,949 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:47:13,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:47:22,058 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-914000.pt
+2023-03-10 17:47:22,931 INFO [train.py:968] (0/2) Epoch 21, batch 1650, giga_loss[loss=0.3088, simple_loss=0.3847, pruned_loss=0.1165, over 28909.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3484, pruned_loss=0.09692, over 5715160.08 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3397, pruned_loss=0.08845, over 3366215.64 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3494, pruned_loss=0.09771, over 5702508.11 frames. ], batch size: 174, lr: 1.54e-03, grad_scale: 8.0
+2023-03-10 17:47:43,734 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=914022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:48:00,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.948e+02 1.357e+03 1.652e+03 2.183e+03 4.517e+03, threshold=3.303e+03, percent-clipped=9.0
+2023-03-10 17:48:09,377 INFO [train.py:968] (0/2) Epoch 21, batch 1700, giga_loss[loss=0.261, simple_loss=0.3339, pruned_loss=0.09403, over 28377.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3507, pruned_loss=0.1005, over 5708871.37 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3395, pruned_loss=0.0882, over 3428910.12 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3519, pruned_loss=0.1016, over 5694571.60 frames. ], batch size: 65, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:48:54,371 INFO [train.py:968] (0/2) Epoch 21, batch 1750, giga_loss[loss=0.269, simple_loss=0.342, pruned_loss=0.09803, over 28516.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3507, pruned_loss=0.1015, over 5693661.57 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.34, pruned_loss=0.08839, over 3491702.31 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3517, pruned_loss=0.1026, over 5686543.94 frames. ], batch size: 336, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:49:31,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.390e+02 1.302e+03 1.682e+03 2.225e+03 6.485e+03, threshold=3.365e+03, percent-clipped=9.0
+2023-03-10 17:49:39,802 INFO [train.py:968] (0/2) Epoch 21, batch 1800, giga_loss[loss=0.2413, simple_loss=0.3219, pruned_loss=0.08033, over 29069.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3484, pruned_loss=0.1008, over 5708839.08 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3396, pruned_loss=0.08813, over 3551723.11 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3497, pruned_loss=0.1021, over 5698243.63 frames. ], batch size: 155, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:50:03,719 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9770, 1.5185, 5.1733, 3.8836], device='cuda:0'), covar=tensor([0.1571, 0.2808, 0.0368, 0.0747], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0640, 0.0946, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 17:50:06,577 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-10 17:50:21,282 INFO [train.py:968] (0/2) Epoch 21, batch 1850, giga_loss[loss=0.2536, simple_loss=0.3344, pruned_loss=0.08639, over 28867.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3483, pruned_loss=0.1002, over 5713382.01 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.34, pruned_loss=0.08817, over 3607936.20 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3493, pruned_loss=0.1016, over 5702274.65 frames. ], batch size: 145, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:50:56,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.362e+02 1.183e+03 1.448e+03 2.128e+03 5.877e+03, threshold=2.895e+03, percent-clipped=3.0
+2023-03-10 17:51:05,335 INFO [train.py:968] (0/2) Epoch 21, batch 1900, giga_loss[loss=0.2511, simple_loss=0.3358, pruned_loss=0.08319, over 28685.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3466, pruned_loss=0.09837, over 5722127.48 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3393, pruned_loss=0.08765, over 3687369.37 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3481, pruned_loss=0.1001, over 5707052.86 frames. ], batch size: 242, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:51:53,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4864, 1.5475, 1.6891, 1.4293], device='cuda:0'), covar=tensor([0.3349, 0.2715, 0.2188, 0.2676], device='cuda:0'), in_proj_covar=tensor([0.1951, 0.1865, 0.1817, 0.1953], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 17:51:57,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2734, 3.1040, 2.9552, 1.4667], device='cuda:0'), covar=tensor([0.0910, 0.1001, 0.0873, 0.2308], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.1112, 0.0943, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 17:51:59,100 INFO [train.py:968] (0/2) Epoch 21, batch 1950, giga_loss[loss=0.2421, simple_loss=0.3218, pruned_loss=0.08114, over 28723.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.344, pruned_loss=0.09684, over 5702285.75 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3396, pruned_loss=0.08785, over 3720501.53 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3451, pruned_loss=0.09825, over 5687879.83 frames. ], batch size: 119, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:52:31,511 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3364, 1.3127, 1.3185, 1.4664], device='cuda:0'), covar=tensor([0.0807, 0.0340, 0.0333, 0.0924], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 17:52:39,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.106e+02 1.138e+03 1.504e+03 1.941e+03 5.620e+03, threshold=3.008e+03, percent-clipped=6.0
+2023-03-10 17:52:47,817 INFO [train.py:968] (0/2) Epoch 21, batch 2000, giga_loss[loss=0.2426, simple_loss=0.3253, pruned_loss=0.07994, over 28908.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3384, pruned_loss=0.09357, over 5697546.84 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3394, pruned_loss=0.08773, over 3763476.39 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3394, pruned_loss=0.09488, over 5682615.54 frames. ], batch size: 174, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:52:53,050 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.11 vs. limit=5.0
+2023-03-10 17:53:35,008 INFO [train.py:968] (0/2) Epoch 21, batch 2050, giga_loss[loss=0.237, simple_loss=0.3101, pruned_loss=0.08193, over 28531.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3333, pruned_loss=0.09098, over 5691336.94 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3399, pruned_loss=0.08802, over 3804630.44 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3336, pruned_loss=0.0919, over 5677159.81 frames. ], batch size: 71, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:54:15,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.332e+02 1.011e+03 1.241e+03 1.675e+03 8.278e+03, threshold=2.482e+03, percent-clipped=5.0
+2023-03-10 17:54:25,841 INFO [train.py:968] (0/2) Epoch 21, batch 2100, giga_loss[loss=0.254, simple_loss=0.3358, pruned_loss=0.08616, over 28874.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3299, pruned_loss=0.08935, over 5693954.03 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3395, pruned_loss=0.08784, over 3865713.21 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3302, pruned_loss=0.09024, over 5677813.41 frames. ], batch size: 145, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:55:09,456 INFO [train.py:968] (0/2) Epoch 21, batch 2150, giga_loss[loss=0.2665, simple_loss=0.345, pruned_loss=0.09407, over 28639.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3308, pruned_loss=0.08943, over 5697082.14 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3394, pruned_loss=0.08769, over 3914835.20 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3309, pruned_loss=0.09026, over 5680825.69 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:55:44,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.689e+02 1.049e+03 1.218e+03 1.595e+03 5.264e+03, threshold=2.436e+03, percent-clipped=8.0
+2023-03-10 17:55:52,246 INFO [train.py:968] (0/2) Epoch 21, batch 2200, giga_loss[loss=0.2434, simple_loss=0.3204, pruned_loss=0.08326, over 28859.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3301, pruned_loss=0.08858, over 5693449.64 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3401, pruned_loss=0.08811, over 3935153.47 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3296, pruned_loss=0.08897, over 5686705.46 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:55:55,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6125, 5.0809, 1.6562, 2.0733], device='cuda:0'), covar=tensor([0.0955, 0.0215, 0.0889, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0548, 0.0380, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 17:56:00,067 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=914561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:56:33,264 INFO [train.py:968] (0/2) Epoch 21, batch 2250, giga_loss[loss=0.2589, simple_loss=0.3322, pruned_loss=0.09285, over 29002.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3293, pruned_loss=0.08831, over 5701083.79 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3403, pruned_loss=0.08808, over 3993302.80 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3285, pruned_loss=0.08866, over 5690183.89 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:56:44,957 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=914614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:56:49,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5830, 2.2066, 1.8370, 1.7581], device='cuda:0'), covar=tensor([0.0774, 0.0255, 0.0293, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 17:57:06,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3902, 1.4472, 3.9405, 3.2533], device='cuda:0'), covar=tensor([0.1637, 0.2695, 0.0415, 0.0982], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0637, 0.0942, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 17:57:07,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.846e+02 1.027e+03 1.270e+03 1.562e+03 3.546e+03, threshold=2.539e+03, percent-clipped=4.0
+2023-03-10 17:57:16,306 INFO [train.py:968] (0/2) Epoch 21, batch 2300, libri_loss[loss=0.246, simple_loss=0.3336, pruned_loss=0.07925, over 29600.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3274, pruned_loss=0.08767, over 5703800.67 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3406, pruned_loss=0.08804, over 4028068.76 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3263, pruned_loss=0.08797, over 5694153.12 frames. ], batch size: 74, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:57:55,967 INFO [train.py:968] (0/2) Epoch 21, batch 2350, libri_loss[loss=0.2211, simple_loss=0.3044, pruned_loss=0.06885, over 29667.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3252, pruned_loss=0.08664, over 5713789.72 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.0881, over 4070943.71 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3239, pruned_loss=0.08682, over 5704937.98 frames. ], batch size: 73, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:58:27,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4430, 1.6738, 1.7077, 1.4478], device='cuda:0'), covar=tensor([0.2173, 0.1948, 0.2537, 0.2271], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0747, 0.0709, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 17:58:31,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.342e+02 1.075e+03 1.379e+03 2.058e+03 5.948e+03, threshold=2.758e+03, percent-clipped=18.0
+2023-03-10 17:58:37,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0234, 3.8486, 3.7193, 1.7147], device='cuda:0'), covar=tensor([0.0758, 0.0834, 0.0918, 0.2046], device='cuda:0'), in_proj_covar=tensor([0.1196, 0.1107, 0.0939, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 17:58:38,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1026, 1.5731, 1.7477, 1.3505], device='cuda:0'), covar=tensor([0.1926, 0.1395, 0.1914, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0747, 0.0709, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 17:58:39,109 INFO [train.py:968] (0/2) Epoch 21, batch 2400, libri_loss[loss=0.2134, simple_loss=0.3022, pruned_loss=0.06231, over 29409.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3234, pruned_loss=0.086, over 5720524.78 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3409, pruned_loss=0.08829, over 4115231.38 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3218, pruned_loss=0.08595, over 5709926.68 frames. ], batch size: 67, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 17:59:15,334 INFO [train.py:968] (0/2) Epoch 21, batch 2450, giga_loss[loss=0.2999, simple_loss=0.3563, pruned_loss=0.1218, over 26707.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3227, pruned_loss=0.08575, over 5729400.97 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3415, pruned_loss=0.08835, over 4184466.32 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3203, pruned_loss=0.08558, over 5715467.25 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:59:23,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1951, 1.3054, 1.2073, 0.9187], device='cuda:0'), covar=tensor([0.1012, 0.0540, 0.1079, 0.1123], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0447, 0.0519, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 17:59:24,984 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-10 17:59:45,613 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.586e+02 1.056e+03 1.366e+03 1.955e+03 4.607e+03, threshold=2.731e+03, percent-clipped=8.0
+2023-03-10 17:59:51,710 INFO [train.py:968] (0/2) Epoch 21, batch 2500, giga_loss[loss=0.2406, simple_loss=0.3193, pruned_loss=0.08101, over 28591.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3203, pruned_loss=0.08459, over 5735044.39 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3413, pruned_loss=0.08801, over 4210121.80 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3183, pruned_loss=0.08464, over 5721688.27 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:00:32,149 INFO [train.py:968] (0/2) Epoch 21, batch 2550, giga_loss[loss=0.2374, simple_loss=0.3126, pruned_loss=0.08107, over 28873.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3182, pruned_loss=0.08342, over 5725512.40 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3413, pruned_loss=0.08786, over 4250902.16 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3161, pruned_loss=0.08346, over 5712120.27 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:00:35,406 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3185, 2.0081, 1.5108, 0.5567], device='cuda:0'), covar=tensor([0.5096, 0.2368, 0.4317, 0.6090], device='cuda:0'), in_proj_covar=tensor([0.1731, 0.1619, 0.1580, 0.1404], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 18:00:40,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9115, 1.3490, 1.2449, 1.1198], device='cuda:0'), covar=tensor([0.2442, 0.1827, 0.2535, 0.2196], device='cuda:0'), in_proj_covar=tensor([0.0473, 0.0743, 0.0706, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 18:01:00,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=914936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:01:06,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.934e+02 1.027e+03 1.261e+03 1.662e+03 5.855e+03, threshold=2.522e+03, percent-clipped=5.0
+2023-03-10 18:01:11,573 INFO [train.py:968] (0/2) Epoch 21, batch 2600, giga_loss[loss=0.2234, simple_loss=0.3019, pruned_loss=0.07242, over 28754.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3172, pruned_loss=0.08266, over 5728180.43 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.342, pruned_loss=0.08819, over 4294780.99 frames. ], giga_tot_loss[loss=0.2395, simple_loss=0.3143, pruned_loss=0.08234, over 5717468.76 frames. ], batch size: 243, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:01:42,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=914989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:01:51,378 INFO [train.py:968] (0/2) Epoch 21, batch 2650, giga_loss[loss=0.2493, simple_loss=0.3174, pruned_loss=0.09058, over 24292.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.317, pruned_loss=0.08263, over 5727332.31 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3426, pruned_loss=0.08838, over 4333454.81 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3138, pruned_loss=0.08214, over 5715989.41 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:02:25,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.135e+02 1.026e+03 1.249e+03 1.787e+03 9.941e+03, threshold=2.498e+03, percent-clipped=13.0
+2023-03-10 18:02:31,129 INFO [train.py:968] (0/2) Epoch 21, batch 2700, giga_loss[loss=0.2496, simple_loss=0.3295, pruned_loss=0.08483, over 28674.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3186, pruned_loss=0.08378, over 5717916.66 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08885, over 4363018.21 frames. ], giga_tot_loss[loss=0.2404, simple_loss=0.315, pruned_loss=0.08293, over 5712452.53 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:02:59,758 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=915079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:03:01,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=915082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:03:18,432 INFO [train.py:968] (0/2) Epoch 21, batch 2750, giga_loss[loss=0.3601, simple_loss=0.4072, pruned_loss=0.1565, over 27661.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3224, pruned_loss=0.08638, over 5715233.19 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3431, pruned_loss=0.08859, over 4386036.72 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3194, pruned_loss=0.08586, over 5708214.24 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:03:26,624 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=915111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:03:47,776 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=915132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:03:49,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=915135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:03:57,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.966e+02 1.261e+03 1.651e+03 2.054e+03 1.084e+04, threshold=3.303e+03, percent-clipped=13.0
+2023-03-10 18:04:04,933 INFO [train.py:968] (0/2) Epoch 21, batch 2800, giga_loss[loss=0.2677, simple_loss=0.3421, pruned_loss=0.09662, over 28815.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3287, pruned_loss=0.09074, over 5704300.92 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3428, pruned_loss=0.08839, over 4401290.19 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3264, pruned_loss=0.09048, over 5696904.62 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:04:10,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5532, 4.2159, 1.6959, 1.6199], device='cuda:0'), covar=tensor([0.0952, 0.0255, 0.0830, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0548, 0.0380, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 18:04:17,538 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=915164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:04:52,601 INFO [train.py:968] (0/2) Epoch 21, batch 2850, giga_loss[loss=0.2863, simple_loss=0.343, pruned_loss=0.1148, over 23675.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3377, pruned_loss=0.09651, over 5694432.60 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3429, pruned_loss=0.08832, over 4423966.19 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3357, pruned_loss=0.09648, over 5685431.87 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:05:30,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.867e+02 1.270e+03 1.510e+03 1.869e+03 4.161e+03, threshold=3.019e+03, percent-clipped=1.0
+2023-03-10 18:05:38,294 INFO [train.py:968] (0/2) Epoch 21, batch 2900, giga_loss[loss=0.2849, simple_loss=0.3653, pruned_loss=0.1022, over 28878.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3426, pruned_loss=0.09832, over 5691663.97 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3428, pruned_loss=0.08818, over 4467153.54 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.341, pruned_loss=0.09861, over 5679812.91 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:06:22,631 INFO [train.py:968] (0/2) Epoch 21, batch 2950, giga_loss[loss=0.2854, simple_loss=0.3636, pruned_loss=0.1035, over 28899.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3481, pruned_loss=0.1006, over 5673295.76 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.343, pruned_loss=0.0883, over 4489331.43 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3468, pruned_loss=0.101, over 5670204.67 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:07:02,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.987e+02 1.212e+03 1.463e+03 1.843e+03 5.083e+03, threshold=2.927e+03, percent-clipped=4.0
+2023-03-10 18:07:05,204 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6405, 1.9146, 1.4973, 1.9396], device='cuda:0'), covar=tensor([0.2589, 0.2619, 0.2875, 0.2335], device='cuda:0'), in_proj_covar=tensor([0.1489, 0.1078, 0.1314, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 18:07:08,319 INFO [train.py:968] (0/2) Epoch 21, batch 3000, giga_loss[loss=0.3506, simple_loss=0.4101, pruned_loss=0.1456, over 28225.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3527, pruned_loss=0.1027, over 5679661.16 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.343, pruned_loss=0.08836, over 4528638.56 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3518, pruned_loss=0.1034, over 5679616.32 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:07:08,324 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 18:07:17,023 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2109, simple_loss=0.3176, pruned_loss=0.05215, over 944034.00 frames. 
+2023-03-10 18:07:17,024 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 18:07:58,742 INFO [train.py:968] (0/2) Epoch 21, batch 3050, giga_loss[loss=0.2127, simple_loss=0.2998, pruned_loss=0.06282, over 28528.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3514, pruned_loss=0.1017, over 5672118.41 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3429, pruned_loss=0.08844, over 4555274.11 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3509, pruned_loss=0.1025, over 5668579.08 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:08:17,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1876, 1.0785, 3.9647, 3.1539], device='cuda:0'), covar=tensor([0.1877, 0.3167, 0.0446, 0.0926], device='cuda:0'), in_proj_covar=tensor([0.0745, 0.0634, 0.0937, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 18:08:34,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.091e+02 1.259e+03 1.675e+03 2.291e+03 5.748e+03, threshold=3.351e+03, percent-clipped=12.0
+2023-03-10 18:08:36,397 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=915446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:08:40,046 INFO [train.py:968] (0/2) Epoch 21, batch 3100, giga_loss[loss=0.2195, simple_loss=0.3053, pruned_loss=0.06687, over 28906.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3468, pruned_loss=0.09813, over 5680547.14 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.343, pruned_loss=0.08862, over 4580683.40 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3464, pruned_loss=0.09881, over 5675126.04 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:08:47,942 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=915460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:09:04,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9674, 1.3362, 1.3147, 1.1407], device='cuda:0'), covar=tensor([0.1896, 0.1365, 0.1864, 0.1618], device='cuda:0'), in_proj_covar=tensor([0.0476, 0.0749, 0.0711, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 18:09:24,590 INFO [train.py:968] (0/2) Epoch 21, batch 3150, libri_loss[loss=0.2663, simple_loss=0.3365, pruned_loss=0.09802, over 29554.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.346, pruned_loss=0.09728, over 5680761.33 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3431, pruned_loss=0.08881, over 4618452.26 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3458, pruned_loss=0.09794, over 5671835.98 frames. ], batch size: 75, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:10:02,016 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.072e+02 1.262e+03 1.625e+03 2.141e+03 5.211e+03, threshold=3.251e+03, percent-clipped=4.0
+2023-03-10 18:10:08,166 INFO [train.py:968] (0/2) Epoch 21, batch 3200, giga_loss[loss=0.3102, simple_loss=0.3831, pruned_loss=0.1187, over 28996.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3469, pruned_loss=0.09791, over 5679732.51 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3429, pruned_loss=0.08867, over 4637016.14 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.347, pruned_loss=0.09867, over 5670109.64 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:10:10,087 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=915552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:10:48,101 INFO [train.py:968] (0/2) Epoch 21, batch 3250, giga_loss[loss=0.3024, simple_loss=0.3573, pruned_loss=0.1237, over 23768.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3489, pruned_loss=0.09928, over 5665663.43 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3422, pruned_loss=0.08853, over 4655282.98 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3495, pruned_loss=0.1003, over 5668797.91 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:11:10,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8539, 3.6623, 3.4385, 1.6839], device='cuda:0'), covar=tensor([0.0707, 0.0861, 0.0767, 0.2191], device='cuda:0'), in_proj_covar=tensor([0.1196, 0.1108, 0.0939, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 18:11:27,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.850e+02 1.327e+03 1.703e+03 2.302e+03 3.776e+03, threshold=3.406e+03, percent-clipped=5.0
+2023-03-10 18:11:30,967 INFO [train.py:968] (0/2) Epoch 21, batch 3300, libri_loss[loss=0.2824, simple_loss=0.3654, pruned_loss=0.0997, over 26285.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3521, pruned_loss=0.1016, over 5676316.44 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3423, pruned_loss=0.08853, over 4660792.58 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3527, pruned_loss=0.1025, over 5683444.27 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:11:48,326 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.15 vs. limit=5.0
+2023-03-10 18:12:01,446 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9420, 1.2994, 1.0554, 0.2318], device='cuda:0'), covar=tensor([0.4221, 0.2917, 0.4343, 0.6293], device='cuda:0'), in_proj_covar=tensor([0.1729, 0.1622, 0.1583, 0.1404], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 18:12:07,684 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=915692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:12:13,566 INFO [train.py:968] (0/2) Epoch 21, batch 3350, giga_loss[loss=0.2879, simple_loss=0.3527, pruned_loss=0.1115, over 28297.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3532, pruned_loss=0.1026, over 5672697.60 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3423, pruned_loss=0.08854, over 4688556.33 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3539, pruned_loss=0.1037, over 5680153.18 frames. ], batch size: 77, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:12:41,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4970, 1.6277, 1.3587, 1.2670], device='cuda:0'), covar=tensor([0.0912, 0.0475, 0.0952, 0.0901], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0445, 0.0516, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:12:48,868 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=915743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:12:50,905 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.132e+02 1.342e+03 1.696e+03 2.286e+03 1.171e+04, threshold=3.392e+03, percent-clipped=10.0
+2023-03-10 18:12:54,989 INFO [train.py:968] (0/2) Epoch 21, batch 3400, giga_loss[loss=0.2808, simple_loss=0.3519, pruned_loss=0.1048, over 28702.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.353, pruned_loss=0.1029, over 5670582.66 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3424, pruned_loss=0.08871, over 4713149.86 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3539, pruned_loss=0.104, over 5680953.14 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:13:37,716 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1600, 1.2743, 1.1487, 0.8496], device='cuda:0'), covar=tensor([0.1034, 0.0549, 0.1101, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0444, 0.0515, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:13:41,088 INFO [train.py:968] (0/2) Epoch 21, batch 3450, giga_loss[loss=0.2453, simple_loss=0.332, pruned_loss=0.07934, over 28874.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3535, pruned_loss=0.1038, over 5665055.93 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3422, pruned_loss=0.08857, over 4724881.42 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3544, pruned_loss=0.1049, over 5671235.08 frames. ], batch size: 66, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:13:59,065 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=915821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:14:09,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=915835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:14:20,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.738e+02 1.218e+03 1.416e+03 1.795e+03 3.853e+03, threshold=2.831e+03, percent-clipped=2.0
+2023-03-10 18:14:23,823 INFO [train.py:968] (0/2) Epoch 21, batch 3500, giga_loss[loss=0.2502, simple_loss=0.3298, pruned_loss=0.08533, over 28508.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3531, pruned_loss=0.1028, over 5676342.56 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.342, pruned_loss=0.08857, over 4736840.49 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.354, pruned_loss=0.1038, over 5678858.23 frames. ], batch size: 85, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:14:48,642 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-10 18:15:03,271 INFO [train.py:968] (0/2) Epoch 21, batch 3550, giga_loss[loss=0.266, simple_loss=0.3491, pruned_loss=0.09149, over 29035.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3527, pruned_loss=0.1014, over 5685011.11 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.08842, over 4759639.93 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3538, pruned_loss=0.1026, over 5683355.50 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:15:26,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=915927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:15:28,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7842, 1.9844, 1.3564, 1.5144], device='cuda:0'), covar=tensor([0.0958, 0.0607, 0.1080, 0.1081], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0445, 0.0516, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:15:40,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.966e+02 1.096e+03 1.335e+03 1.732e+03 3.476e+03, threshold=2.669e+03, percent-clipped=4.0
+2023-03-10 18:15:44,125 INFO [train.py:968] (0/2) Epoch 21, batch 3600, giga_loss[loss=0.2575, simple_loss=0.3401, pruned_loss=0.08745, over 29003.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3517, pruned_loss=0.09984, over 5692038.44 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3412, pruned_loss=0.08806, over 4787941.09 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3533, pruned_loss=0.1013, over 5685921.94 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:15:56,071 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=915964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:15:58,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=915967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:16:06,651 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=915978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:16:08,129 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5999, 1.8001, 1.7761, 1.6472], device='cuda:0'), covar=tensor([0.1849, 0.2129, 0.2176, 0.2081], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0745, 0.0708, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 18:16:09,091 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=915981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:16:11,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3984, 1.5238, 1.2933, 1.5487], device='cuda:0'), covar=tensor([0.0779, 0.0360, 0.0349, 0.0869], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0116, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0061, 0.0106], device='cuda:0')
+2023-03-10 18:16:20,433 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=915996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:16:23,936 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-916000.pt
+2023-03-10 18:16:24,970 INFO [train.py:968] (0/2) Epoch 21, batch 3650, libri_loss[loss=0.279, simple_loss=0.3627, pruned_loss=0.0976, over 29368.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3502, pruned_loss=0.09886, over 5694778.33 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3414, pruned_loss=0.08815, over 4795812.40 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3514, pruned_loss=0.1001, over 5691156.03 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:16:32,120 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:17:04,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.160e+02 1.142e+03 1.498e+03 2.275e+03 8.409e+03, threshold=2.996e+03, percent-clipped=20.0
+2023-03-10 18:17:07,485 INFO [train.py:968] (0/2) Epoch 21, batch 3700, giga_loss[loss=0.2469, simple_loss=0.321, pruned_loss=0.08641, over 28096.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.349, pruned_loss=0.09878, over 5695673.05 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3417, pruned_loss=0.08832, over 4816272.64 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3498, pruned_loss=0.09981, over 5690490.10 frames. ], batch size: 77, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:17:18,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:17:19,981 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:17:22,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:17:42,777 INFO [train.py:968] (0/2) Epoch 21, batch 3750, giga_loss[loss=0.2908, simple_loss=0.3649, pruned_loss=0.1084, over 29044.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3469, pruned_loss=0.09738, over 5702194.38 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3421, pruned_loss=0.08856, over 4866619.50 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3476, pruned_loss=0.09854, over 5695321.00 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:17:43,657 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:17:54,818 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:18:22,818 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.651e+02 1.187e+03 1.372e+03 1.836e+03 8.833e+03, threshold=2.743e+03, percent-clipped=9.0
+2023-03-10 18:18:26,216 INFO [train.py:968] (0/2) Epoch 21, batch 3800, giga_loss[loss=0.3018, simple_loss=0.3774, pruned_loss=0.1131, over 28644.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3483, pruned_loss=0.09883, over 5701702.72 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3416, pruned_loss=0.08829, over 4892409.88 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3493, pruned_loss=0.1001, over 5692013.17 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:18:30,728 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=916156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:06,539 INFO [train.py:968] (0/2) Epoch 21, batch 3850, giga_loss[loss=0.2807, simple_loss=0.361, pruned_loss=0.1002, over 28840.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3489, pruned_loss=0.09901, over 5704752.22 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3421, pruned_loss=0.08858, over 4904245.18 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3494, pruned_loss=0.09995, over 5697198.90 frames. ], batch size: 66, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:19:12,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:15,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:30,844 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=916232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:38,218 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:39,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1737, 2.3231, 2.2968, 2.1351], device='cuda:0'), covar=tensor([0.1460, 0.1607, 0.1588, 0.1568], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0743, 0.0706, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 18:19:40,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.348e+02 1.091e+03 1.444e+03 2.046e+03 7.271e+03, threshold=2.887e+03, percent-clipped=9.0
+2023-03-10 18:19:45,128 INFO [train.py:968] (0/2) Epoch 21, batch 3900, giga_loss[loss=0.2511, simple_loss=0.3294, pruned_loss=0.0864, over 28915.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.348, pruned_loss=0.09763, over 5700065.00 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08856, over 4919623.74 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3486, pruned_loss=0.09858, over 5698777.31 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:19:49,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5101, 1.7873, 1.4619, 1.3517], device='cuda:0'), covar=tensor([0.2697, 0.2623, 0.3016, 0.2281], device='cuda:0'), in_proj_covar=tensor([0.1486, 0.1077, 0.1312, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 18:19:53,245 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:55,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:20:14,370 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=916283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:20:21,150 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:20:27,623 INFO [train.py:968] (0/2) Epoch 21, batch 3950, giga_loss[loss=0.2543, simple_loss=0.3325, pruned_loss=0.08803, over 28647.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3465, pruned_loss=0.09647, over 5707867.25 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08856, over 4929479.57 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3471, pruned_loss=0.09728, over 5705047.06 frames. ], batch size: 85, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:21:03,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.460e+02 1.013e+03 1.155e+03 1.494e+03 4.022e+03, threshold=2.309e+03, percent-clipped=4.0
+2023-03-10 18:21:08,588 INFO [train.py:968] (0/2) Epoch 21, batch 4000, giga_loss[loss=0.2785, simple_loss=0.3509, pruned_loss=0.1031, over 28912.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3458, pruned_loss=0.09645, over 5705209.87 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3409, pruned_loss=0.088, over 4965758.11 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3472, pruned_loss=0.09779, over 5697790.64 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:21:21,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=916367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:21:47,511 INFO [train.py:968] (0/2) Epoch 21, batch 4050, libri_loss[loss=0.2729, simple_loss=0.3592, pruned_loss=0.09326, over 29529.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3437, pruned_loss=0.09543, over 5709664.60 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3406, pruned_loss=0.08788, over 4992936.43 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3451, pruned_loss=0.09691, over 5705492.57 frames. ], batch size: 89, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:22:02,760 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-10 18:22:16,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2040, 1.5702, 1.0365, 1.1216], device='cuda:0'), covar=tensor([0.1216, 0.0629, 0.1450, 0.1350], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0446, 0.0517, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:22:20,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.710e+02 1.346e+03 1.736e+03 2.403e+03 5.455e+03, threshold=3.472e+03, percent-clipped=28.0
+2023-03-10 18:22:23,967 INFO [train.py:968] (0/2) Epoch 21, batch 4100, giga_loss[loss=0.2501, simple_loss=0.3255, pruned_loss=0.08732, over 28956.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3417, pruned_loss=0.09459, over 5713281.15 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3406, pruned_loss=0.08783, over 5018156.59 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3428, pruned_loss=0.096, over 5706412.49 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:22:57,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-10 18:23:05,639 INFO [train.py:968] (0/2) Epoch 21, batch 4150, giga_loss[loss=0.2389, simple_loss=0.3196, pruned_loss=0.07913, over 29076.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3387, pruned_loss=0.09315, over 5714261.55 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.08797, over 5022727.81 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3395, pruned_loss=0.09418, over 5708004.86 frames. ], batch size: 128, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:23:06,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1241, 2.1954, 1.8968, 2.2404], device='cuda:0'), covar=tensor([0.2251, 0.2512, 0.2728, 0.2237], device='cuda:0'), in_proj_covar=tensor([0.1487, 0.1078, 0.1315, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 18:23:31,506 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:23:32,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6699, 1.8438, 1.3081, 1.4182], device='cuda:0'), covar=tensor([0.0926, 0.0636, 0.1016, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0445, 0.0516, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:23:42,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9484, 2.2934, 1.8264, 2.1190], device='cuda:0'), covar=tensor([0.2487, 0.2485, 0.2862, 0.2545], device='cuda:0'), in_proj_covar=tensor([0.1487, 0.1076, 0.1314, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 18:23:43,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.743e+02 1.125e+03 1.368e+03 1.901e+03 3.971e+03, threshold=2.737e+03, percent-clipped=1.0
+2023-03-10 18:23:46,545 INFO [train.py:968] (0/2) Epoch 21, batch 4200, giga_loss[loss=0.2101, simple_loss=0.2972, pruned_loss=0.06148, over 28958.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3378, pruned_loss=0.0929, over 5714518.75 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.08805, over 5027108.17 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3384, pruned_loss=0.09367, over 5708891.41 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:24:03,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3236, 1.6830, 1.1134, 1.1758], device='cuda:0'), covar=tensor([0.1222, 0.0772, 0.1526, 0.1419], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0445, 0.0516, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:24:28,767 INFO [train.py:968] (0/2) Epoch 21, batch 4250, giga_loss[loss=0.2782, simple_loss=0.3468, pruned_loss=0.1048, over 28905.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3371, pruned_loss=0.09319, over 5716140.51 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.08802, over 5035922.12 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3376, pruned_loss=0.09386, over 5710125.76 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:24:34,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:24:40,480 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 18:25:04,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.399e+02 1.195e+03 1.382e+03 1.927e+03 5.955e+03, threshold=2.763e+03, percent-clipped=9.0
+2023-03-10 18:25:07,871 INFO [train.py:968] (0/2) Epoch 21, batch 4300, libri_loss[loss=0.2458, simple_loss=0.3332, pruned_loss=0.0792, over 29564.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3361, pruned_loss=0.09309, over 5707728.63 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3411, pruned_loss=0.0882, over 5056054.96 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3361, pruned_loss=0.09367, over 5705449.54 frames. ], batch size: 76, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:25:12,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3392, 3.5599, 1.6415, 1.4115], device='cuda:0'), covar=tensor([0.0966, 0.0405, 0.0868, 0.1357], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0545, 0.0379, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 18:25:13,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:25:27,241 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:25:29,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:25:46,053 INFO [train.py:968] (0/2) Epoch 21, batch 4350, libri_loss[loss=0.26, simple_loss=0.3411, pruned_loss=0.08947, over 29538.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3341, pruned_loss=0.09228, over 5710721.59 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3418, pruned_loss=0.08865, over 5076420.32 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3334, pruned_loss=0.09248, over 5704959.68 frames. ], batch size: 80, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:25:50,479 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:26:18,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:26:22,067 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.168e+02 1.106e+03 1.308e+03 1.618e+03 5.214e+03, threshold=2.617e+03, percent-clipped=7.0
+2023-03-10 18:26:24,471 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:26:24,809 INFO [train.py:968] (0/2) Epoch 21, batch 4400, giga_loss[loss=0.2283, simple_loss=0.3067, pruned_loss=0.0749, over 28816.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3321, pruned_loss=0.09119, over 5709520.02 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3419, pruned_loss=0.0887, over 5088553.17 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3313, pruned_loss=0.09135, over 5702525.69 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:26:26,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:26:39,340 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=916769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:26:48,709 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:27:04,772 INFO [train.py:968] (0/2) Epoch 21, batch 4450, giga_loss[loss=0.2449, simple_loss=0.3308, pruned_loss=0.07951, over 28696.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3338, pruned_loss=0.09174, over 5714639.19 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3421, pruned_loss=0.08875, over 5116822.71 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3328, pruned_loss=0.09191, over 5702870.47 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:27:05,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:27:08,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:27:15,903 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5471, 1.7304, 1.4575, 1.4782], device='cuda:0'), covar=tensor([0.2830, 0.2860, 0.3255, 0.2545], device='cuda:0'), in_proj_covar=tensor([0.1492, 0.1079, 0.1317, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 18:27:24,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9844, 1.0877, 3.4563, 2.9885], device='cuda:0'), covar=tensor([0.1731, 0.2800, 0.0462, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0746, 0.0636, 0.0939, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 18:27:29,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9608, 2.1987, 1.7167, 1.7378], device='cuda:0'), covar=tensor([0.0965, 0.0690, 0.0933, 0.1234], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0445, 0.0517, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:27:32,212 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:27:46,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.021e+02 1.049e+03 1.272e+03 1.715e+03 5.931e+03, threshold=2.545e+03, percent-clipped=9.0
+2023-03-10 18:27:48,458 INFO [train.py:968] (0/2) Epoch 21, batch 4500, giga_loss[loss=0.2583, simple_loss=0.3459, pruned_loss=0.08537, over 28661.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3349, pruned_loss=0.09211, over 5720145.19 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3418, pruned_loss=0.08879, over 5140258.63 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3341, pruned_loss=0.09232, over 5705519.47 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:28:17,572 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:28:20,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:28:30,562 INFO [train.py:968] (0/2) Epoch 21, batch 4550, libri_loss[loss=0.2713, simple_loss=0.3534, pruned_loss=0.09461, over 29493.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3378, pruned_loss=0.09282, over 5724013.87 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3418, pruned_loss=0.08885, over 5151439.36 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3371, pruned_loss=0.093, over 5710125.63 frames. ], batch size: 85, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:28:43,850 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:29:10,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.746e+02 1.108e+03 1.295e+03 1.702e+03 4.251e+03, threshold=2.590e+03, percent-clipped=9.0
+2023-03-10 18:29:12,616 INFO [train.py:968] (0/2) Epoch 21, batch 4600, giga_loss[loss=0.3043, simple_loss=0.3784, pruned_loss=0.1151, over 29009.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3402, pruned_loss=0.09363, over 5717014.18 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.342, pruned_loss=0.08888, over 5171158.12 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3394, pruned_loss=0.09388, over 5703033.32 frames. ], batch size: 128, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:29:36,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2396, 1.3205, 1.2463, 1.2555], device='cuda:0'), covar=tensor([0.2199, 0.1946, 0.1818, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.1947, 0.1860, 0.1809, 0.1944], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 18:29:57,400 INFO [train.py:968] (0/2) Epoch 21, batch 4650, giga_loss[loss=0.2305, simple_loss=0.318, pruned_loss=0.07156, over 28800.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3398, pruned_loss=0.09307, over 5694225.71 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3426, pruned_loss=0.08931, over 5174355.21 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3386, pruned_loss=0.093, over 5692395.14 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:30:13,896 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3853, 3.3527, 1.4904, 1.5041], device='cuda:0'), covar=tensor([0.0946, 0.0326, 0.0942, 0.1303], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0547, 0.0380, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 18:30:35,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.856e+02 1.116e+03 1.358e+03 1.816e+03 7.405e+03, threshold=2.716e+03, percent-clipped=5.0
+2023-03-10 18:30:36,755 INFO [train.py:968] (0/2) Epoch 21, batch 4700, giga_loss[loss=0.2766, simple_loss=0.355, pruned_loss=0.09909, over 28985.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3398, pruned_loss=0.09245, over 5701795.56 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3427, pruned_loss=0.08924, over 5192468.44 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3387, pruned_loss=0.09253, over 5695580.83 frames. ], batch size: 145, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:31:10,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4641, 1.7244, 1.6719, 1.5301], device='cuda:0'), covar=tensor([0.1823, 0.2045, 0.2286, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0745, 0.0708, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 18:31:17,021 INFO [train.py:968] (0/2) Epoch 21, batch 4750, giga_loss[loss=0.2229, simple_loss=0.3065, pruned_loss=0.06963, over 28931.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3405, pruned_loss=0.09303, over 5707704.07 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08933, over 5202963.47 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3395, pruned_loss=0.09308, over 5702198.26 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:31:44,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7032, 1.8321, 1.4383, 1.4537], device='cuda:0'), covar=tensor([0.1052, 0.0834, 0.1063, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0445, 0.0515, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:31:53,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:31:56,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.703e+02 1.260e+03 1.563e+03 1.987e+03 4.944e+03, threshold=3.126e+03, percent-clipped=10.0
+2023-03-10 18:31:58,141 INFO [train.py:968] (0/2) Epoch 21, batch 4800, giga_loss[loss=0.2507, simple_loss=0.3283, pruned_loss=0.08655, over 28758.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3404, pruned_loss=0.09338, over 5712156.66 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3424, pruned_loss=0.0891, over 5213313.98 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3399, pruned_loss=0.09367, over 5705382.15 frames. ], batch size: 99, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:32:17,200 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:32:30,469 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:32:37,359 INFO [train.py:968] (0/2) Epoch 21, batch 4850, giga_loss[loss=0.2774, simple_loss=0.3539, pruned_loss=0.1004, over 29070.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.343, pruned_loss=0.09473, over 5698237.02 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3425, pruned_loss=0.08919, over 5221550.73 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3424, pruned_loss=0.09506, over 5705848.37 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:32:59,359 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:33:18,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.714e+02 1.213e+03 1.639e+03 2.392e+03 6.753e+03, threshold=3.278e+03, percent-clipped=16.0
+2023-03-10 18:33:20,155 INFO [train.py:968] (0/2) Epoch 21, batch 4900, giga_loss[loss=0.2526, simple_loss=0.3414, pruned_loss=0.08189, over 28921.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.345, pruned_loss=0.09583, over 5702751.53 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3426, pruned_loss=0.08924, over 5231138.93 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3445, pruned_loss=0.09614, over 5706610.15 frames. ], batch size: 145, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:33:49,945 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:33:52,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:33:59,193 INFO [train.py:968] (0/2) Epoch 21, batch 4950, giga_loss[loss=0.2726, simple_loss=0.3421, pruned_loss=0.1015, over 28884.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3479, pruned_loss=0.09752, over 5706060.80 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3432, pruned_loss=0.08955, over 5252711.73 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3472, pruned_loss=0.09777, over 5704332.11 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:34:04,028 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:34:14,605 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=917319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:34:30,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8677, 1.2689, 1.3782, 1.0138], device='cuda:0'), covar=tensor([0.1896, 0.1313, 0.2112, 0.1695], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0743, 0.0706, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 18:34:37,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.795e+02 1.296e+03 1.652e+03 2.246e+03 4.746e+03, threshold=3.305e+03, percent-clipped=5.0
+2023-03-10 18:34:38,773 INFO [train.py:968] (0/2) Epoch 21, batch 5000, giga_loss[loss=0.252, simple_loss=0.332, pruned_loss=0.08602, over 28710.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3495, pruned_loss=0.09851, over 5714461.17 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09002, over 5271760.14 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3486, pruned_loss=0.09857, over 5708098.53 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:35:19,062 INFO [train.py:968] (0/2) Epoch 21, batch 5050, giga_loss[loss=0.322, simple_loss=0.3909, pruned_loss=0.1265, over 28354.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3503, pruned_loss=0.09906, over 5707114.54 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3438, pruned_loss=0.08996, over 5280963.01 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3496, pruned_loss=0.09927, over 5699849.44 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:35:30,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5275, 3.6524, 1.5802, 1.6650], device='cuda:0'), covar=tensor([0.0925, 0.0256, 0.0917, 0.1308], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0548, 0.0380, 0.0420], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 18:35:46,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5871, 1.5693, 1.7879, 1.3719], device='cuda:0'), covar=tensor([0.1874, 0.2398, 0.1508, 0.1727], device='cuda:0'), in_proj_covar=tensor([0.0897, 0.0699, 0.0941, 0.0839], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 18:35:58,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.729e+02 1.283e+03 1.552e+03 1.867e+03 3.909e+03, threshold=3.105e+03, percent-clipped=3.0
+2023-03-10 18:36:00,079 INFO [train.py:968] (0/2) Epoch 21, batch 5100, giga_loss[loss=0.2426, simple_loss=0.3191, pruned_loss=0.08306, over 29021.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3493, pruned_loss=0.09834, over 5712649.47 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3437, pruned_loss=0.08987, over 5284115.20 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3489, pruned_loss=0.0986, over 5706127.69 frames. ], batch size: 128, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:36:23,084 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:36:25,669 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:36:44,545 INFO [train.py:968] (0/2) Epoch 21, batch 5150, giga_loss[loss=0.3118, simple_loss=0.3648, pruned_loss=0.1294, over 23896.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3478, pruned_loss=0.09816, over 5704313.13 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3442, pruned_loss=0.09025, over 5292088.56 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3471, pruned_loss=0.09817, over 5697635.79 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:36:50,857 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-10 18:37:08,641 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:37:22,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:37:24,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.128e+02 1.156e+03 1.654e+03 2.239e+03 5.571e+03, threshold=3.308e+03, percent-clipped=8.0
+2023-03-10 18:37:24,807 INFO [train.py:968] (0/2) Epoch 21, batch 5200, giga_loss[loss=0.2572, simple_loss=0.3314, pruned_loss=0.09144, over 28938.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3439, pruned_loss=0.09633, over 5711794.13 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3442, pruned_loss=0.09024, over 5298255.32 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3434, pruned_loss=0.09641, over 5704750.94 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:37:32,212 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:37:33,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:37:35,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:38:04,957 INFO [train.py:968] (0/2) Epoch 21, batch 5250, giga_loss[loss=0.2252, simple_loss=0.304, pruned_loss=0.07316, over 28419.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3423, pruned_loss=0.09449, over 5712109.36 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3446, pruned_loss=0.0904, over 5310844.52 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3414, pruned_loss=0.09454, over 5707929.84 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:38:05,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:38:06,730 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-10 18:38:08,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1390, 1.4761, 1.1566, 0.5433], device='cuda:0'), covar=tensor([0.2613, 0.1587, 0.2187, 0.4627], device='cuda:0'), in_proj_covar=tensor([0.1722, 0.1611, 0.1577, 0.1395], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 18:38:44,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.413e+02 1.254e+03 1.505e+03 1.883e+03 3.467e+03, threshold=3.010e+03, percent-clipped=0.0
+2023-03-10 18:38:44,793 INFO [train.py:968] (0/2) Epoch 21, batch 5300, giga_loss[loss=0.2655, simple_loss=0.3553, pruned_loss=0.08785, over 28724.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3435, pruned_loss=0.0938, over 5711157.42 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.345, pruned_loss=0.09055, over 5320758.01 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3424, pruned_loss=0.0938, over 5707743.21 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:38:53,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4808, 2.9217, 1.5830, 1.6185], device='cuda:0'), covar=tensor([0.0784, 0.0263, 0.0761, 0.1085], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0550, 0.0382, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 18:39:03,094 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-10 18:39:03,680 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5087, 1.5889, 1.7019, 1.3080], device='cuda:0'), covar=tensor([0.1951, 0.2435, 0.1581, 0.1795], device='cuda:0'), in_proj_covar=tensor([0.0898, 0.0699, 0.0942, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 18:39:07,754 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-10 18:39:09,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:18,899 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:22,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:22,752 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8987, 2.0206, 1.8769, 1.7168], device='cuda:0'), covar=tensor([0.1848, 0.2380, 0.2186, 0.2559], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0744, 0.0707, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 18:39:23,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:26,656 INFO [train.py:968] (0/2) Epoch 21, batch 5350, libri_loss[loss=0.2662, simple_loss=0.3566, pruned_loss=0.08795, over 29240.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3445, pruned_loss=0.09361, over 5719655.97 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.345, pruned_loss=0.09045, over 5337650.34 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3436, pruned_loss=0.09379, over 5712035.05 frames. ], batch size: 97, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:39:32,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:35,486 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:44,443 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=917723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:56,659 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=917740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:59,418 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:40:01,424 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:40:04,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.107e+02 1.222e+03 1.473e+03 1.913e+03 6.157e+03, threshold=2.947e+03, percent-clipped=4.0
+2023-03-10 18:40:05,700 INFO [train.py:968] (0/2) Epoch 21, batch 5400, libri_loss[loss=0.2105, simple_loss=0.2944, pruned_loss=0.0633, over 28543.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3433, pruned_loss=0.09417, over 5718635.32 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3453, pruned_loss=0.09066, over 5354989.27 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3424, pruned_loss=0.09429, over 5710388.30 frames. ], batch size: 63, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:40:19,553 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-10 18:40:24,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=917776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:40:45,430 INFO [train.py:968] (0/2) Epoch 21, batch 5450, giga_loss[loss=0.2555, simple_loss=0.3369, pruned_loss=0.08709, over 28571.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3418, pruned_loss=0.09445, over 5723563.82 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3452, pruned_loss=0.09071, over 5364068.66 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.341, pruned_loss=0.09456, over 5715438.00 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:40:52,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1647, 1.2757, 1.1487, 0.8821], device='cuda:0'), covar=tensor([0.0967, 0.0539, 0.1091, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0445, 0.0513, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:41:02,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1974, 1.2543, 1.1203, 0.8801], device='cuda:0'), covar=tensor([0.0945, 0.0531, 0.1090, 0.1110], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0445, 0.0514, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:41:05,165 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:41:07,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:41:13,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3644, 1.1715, 1.0478, 1.5174], device='cuda:0'), covar=tensor([0.0753, 0.0361, 0.0368, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0116, 0.0219], device='cuda:0'), out_proj_covar=tensor([0.0096, 0.0069, 0.0061, 0.0105], device='cuda:0')
+2023-03-10 18:41:24,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.476e+02 1.263e+03 1.488e+03 1.901e+03 7.083e+03, threshold=2.977e+03, percent-clipped=12.0
+2023-03-10 18:41:25,614 INFO [train.py:968] (0/2) Epoch 21, batch 5500, giga_loss[loss=0.2656, simple_loss=0.3293, pruned_loss=0.101, over 28539.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.34, pruned_loss=0.09485, over 5731525.34 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3446, pruned_loss=0.09048, over 5381808.08 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3398, pruned_loss=0.09527, over 5720246.11 frames. ], batch size: 60, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:41:27,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:41:29,767 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=917856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:41:30,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:41:42,741 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-10 18:42:03,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0493, 1.3850, 1.1247, 0.2022], device='cuda:0'), covar=tensor([0.3452, 0.3153, 0.4360, 0.6126], device='cuda:0'), in_proj_covar=tensor([0.1719, 0.1609, 0.1574, 0.1396], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 18:42:03,322 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-10 18:42:06,232 INFO [train.py:968] (0/2) Epoch 21, batch 5550, giga_loss[loss=0.2248, simple_loss=0.3079, pruned_loss=0.07081, over 28919.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3379, pruned_loss=0.09488, over 5738174.60 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3438, pruned_loss=0.0901, over 5402925.39 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3382, pruned_loss=0.09573, over 5722264.36 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:42:09,480 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:42:36,935 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:42:40,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:42:40,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:42:48,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.464e+02 1.224e+03 1.556e+03 1.970e+03 4.210e+03, threshold=3.112e+03, percent-clipped=4.0
+2023-03-10 18:42:49,742 INFO [train.py:968] (0/2) Epoch 21, batch 5600, giga_loss[loss=0.2861, simple_loss=0.346, pruned_loss=0.1131, over 24034.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3381, pruned_loss=0.09523, over 5724895.99 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3438, pruned_loss=0.09009, over 5410578.74 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3382, pruned_loss=0.09598, over 5709884.82 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:43:26,522 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:43:30,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:43:30,497 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-918000.pt
+2023-03-10 18:43:31,117 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:43:31,419 INFO [train.py:968] (0/2) Epoch 21, batch 5650, giga_loss[loss=0.2362, simple_loss=0.3082, pruned_loss=0.0821, over 28910.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3335, pruned_loss=0.09269, over 5716657.66 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3437, pruned_loss=0.09011, over 5410501.56 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3335, pruned_loss=0.09333, over 5710089.33 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:43:34,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:43:53,001 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:43:56,082 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:10,056 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:10,321 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.310e+02 1.213e+03 1.477e+03 2.011e+03 4.962e+03, threshold=2.954e+03, percent-clipped=9.0
+2023-03-10 18:44:10,961 INFO [train.py:968] (0/2) Epoch 21, batch 5700, giga_loss[loss=0.2373, simple_loss=0.3113, pruned_loss=0.08166, over 29000.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3305, pruned_loss=0.09136, over 5715463.62 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09003, over 5421006.70 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3302, pruned_loss=0.09198, over 5707623.40 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:44:13,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:29,582 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=918072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:34,935 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:36,249 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:36,309 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:37,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:38,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:51,728 INFO [train.py:968] (0/2) Epoch 21, batch 5750, giga_loss[loss=0.2308, simple_loss=0.3027, pruned_loss=0.07942, over 28745.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3282, pruned_loss=0.09031, over 5718721.28 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09022, over 5425806.43 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3276, pruned_loss=0.09063, over 5710833.54 frames. ], batch size: 99, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:45:01,873 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:45:03,347 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:45:09,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=918122.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 18:45:16,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=918132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:45:32,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.586e+02 1.254e+03 1.665e+03 2.185e+03 6.547e+03, threshold=3.330e+03, percent-clipped=10.0
+2023-03-10 18:45:32,464 INFO [train.py:968] (0/2) Epoch 21, batch 5800, giga_loss[loss=0.2316, simple_loss=0.3157, pruned_loss=0.07372, over 29021.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3299, pruned_loss=0.09108, over 5714267.25 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.0902, over 5428220.40 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3294, pruned_loss=0.09136, over 5707225.29 frames. ], batch size: 128, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:46:05,972 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-10 18:46:13,902 INFO [train.py:968] (0/2) Epoch 21, batch 5850, giga_loss[loss=0.2613, simple_loss=0.329, pruned_loss=0.09676, over 23882.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3334, pruned_loss=0.09227, over 5711774.80 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.09024, over 5436894.27 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3327, pruned_loss=0.09248, over 5703584.24 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:46:27,492 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:46:30,434 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:46:39,497 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-10 18:46:54,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:46:54,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3580, 1.5141, 1.5030, 1.2595], device='cuda:0'), covar=tensor([0.2687, 0.2501, 0.1743, 0.2384], device='cuda:0'), in_proj_covar=tensor([0.1955, 0.1876, 0.1814, 0.1953], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 18:46:58,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.475e+02 1.166e+03 1.383e+03 1.760e+03 4.333e+03, threshold=2.766e+03, percent-clipped=3.0
+2023-03-10 18:46:58,026 INFO [train.py:968] (0/2) Epoch 21, batch 5900, giga_loss[loss=0.2543, simple_loss=0.3409, pruned_loss=0.0838, over 28619.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3368, pruned_loss=0.09329, over 5719377.37 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.344, pruned_loss=0.09013, over 5443552.25 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3362, pruned_loss=0.09358, over 5710667.85 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:47:16,483 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-10 18:47:40,956 INFO [train.py:968] (0/2) Epoch 21, batch 5950, giga_loss[loss=0.2791, simple_loss=0.357, pruned_loss=0.1006, over 28641.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3401, pruned_loss=0.09471, over 5714770.87 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3438, pruned_loss=0.09013, over 5453014.12 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3397, pruned_loss=0.09502, over 5704042.21 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:47:50,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=918314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:48:21,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.918e+02 1.156e+03 1.600e+03 2.302e+03 9.598e+03, threshold=3.201e+03, percent-clipped=19.0
+2023-03-10 18:48:22,001 INFO [train.py:968] (0/2) Epoch 21, batch 6000, giga_loss[loss=0.2695, simple_loss=0.3457, pruned_loss=0.09663, over 29112.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.342, pruned_loss=0.09549, over 5714920.08 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.344, pruned_loss=0.09011, over 5461912.15 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3414, pruned_loss=0.09585, over 5703229.36 frames. ], batch size: 128, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:48:22,005 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 18:48:30,631 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2114, simple_loss=0.3187, pruned_loss=0.05201, over 944034.00 frames. 
+2023-03-10 18:48:30,631 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 18:49:14,681 INFO [train.py:968] (0/2) Epoch 21, batch 6050, giga_loss[loss=0.4141, simple_loss=0.436, pruned_loss=0.1961, over 26608.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3467, pruned_loss=0.09965, over 5700788.06 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3438, pruned_loss=0.08999, over 5466069.39 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3464, pruned_loss=0.1002, over 5694484.76 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:49:32,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5567, 1.6832, 1.2738, 1.2305], device='cuda:0'), covar=tensor([0.0956, 0.0603, 0.1065, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0446, 0.0514, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 18:50:03,728 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.532e+03 1.966e+03 2.523e+03 6.163e+03, threshold=3.932e+03, percent-clipped=16.0
+2023-03-10 18:50:03,740 INFO [train.py:968] (0/2) Epoch 21, batch 6100, giga_loss[loss=0.3765, simple_loss=0.4158, pruned_loss=0.1686, over 26740.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.354, pruned_loss=0.1056, over 5696416.38 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3436, pruned_loss=0.08989, over 5470350.86 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3541, pruned_loss=0.1062, over 5689782.26 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:50:08,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:50:14,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:50:38,924 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:50:46,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=918497.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 18:50:48,652 INFO [train.py:968] (0/2) Epoch 21, batch 6150, giga_loss[loss=0.4116, simple_loss=0.4426, pruned_loss=0.1903, over 26605.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3603, pruned_loss=0.1094, over 5701186.59 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3438, pruned_loss=0.09008, over 5483435.11 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3605, pruned_loss=0.1102, over 5690203.94 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:50:54,374 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=918507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:51:05,829 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=918519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:51:37,272 INFO [train.py:968] (0/2) Epoch 21, batch 6200, giga_loss[loss=0.3929, simple_loss=0.4309, pruned_loss=0.1775, over 28831.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3659, pruned_loss=0.1143, over 5702066.13 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3439, pruned_loss=0.09022, over 5490537.73 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3663, pruned_loss=0.1152, over 5690926.63 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:51:37,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.144e+03 1.812e+03 2.511e+03 2.994e+03 4.886e+03, threshold=5.023e+03, percent-clipped=6.0
+2023-03-10 18:52:22,608 INFO [train.py:968] (0/2) Epoch 21, batch 6250, giga_loss[loss=0.3388, simple_loss=0.3965, pruned_loss=0.1406, over 28973.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3738, pruned_loss=0.1212, over 5699275.10 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09031, over 5496969.72 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3745, pruned_loss=0.1224, over 5689545.48 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:52:46,123 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=918627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:52:59,829 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918640.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 18:53:03,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918643.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 18:53:09,418 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:53:09,764 INFO [train.py:968] (0/2) Epoch 21, batch 6300, libri_loss[loss=0.2915, simple_loss=0.3722, pruned_loss=0.1054, over 25470.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3778, pruned_loss=0.1246, over 5688567.07 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3443, pruned_loss=0.09048, over 5501391.42 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.379, pruned_loss=0.1263, over 5682257.75 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:53:11,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.683e+03 2.294e+03 3.343e+03 1.127e+04, threshold=4.588e+03, percent-clipped=7.0
+2023-03-10 18:53:12,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:53:33,642 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918672.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 18:53:43,137 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:53:50,274 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3339, 1.2037, 4.0028, 3.2735], device='cuda:0'), covar=tensor([0.1603, 0.2773, 0.0442, 0.1515], device='cuda:0'), in_proj_covar=tensor([0.0751, 0.0637, 0.0945, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 18:54:03,534 INFO [train.py:968] (0/2) Epoch 21, batch 6350, giga_loss[loss=0.3726, simple_loss=0.4014, pruned_loss=0.1719, over 23275.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3791, pruned_loss=0.1266, over 5667513.63 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.0902, over 5505883.12 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3808, pruned_loss=0.1287, over 5660750.21 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:54:33,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-10 18:54:52,780 INFO [train.py:968] (0/2) Epoch 21, batch 6400, giga_loss[loss=0.319, simple_loss=0.3828, pruned_loss=0.1276, over 28916.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3821, pruned_loss=0.1296, over 5668614.13 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3439, pruned_loss=0.09023, over 5513536.43 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3847, pruned_loss=0.1324, over 5661233.33 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:54:54,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.661e+03 2.214e+03 3.112e+03 7.505e+03, threshold=4.428e+03, percent-clipped=9.0
+2023-03-10 18:55:44,206 INFO [train.py:968] (0/2) Epoch 21, batch 6450, giga_loss[loss=0.4276, simple_loss=0.4333, pruned_loss=0.2109, over 23622.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3845, pruned_loss=0.1328, over 5648238.66 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3441, pruned_loss=0.09019, over 5510396.33 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3876, pruned_loss=0.1362, over 5649547.83 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:56:37,187 INFO [train.py:968] (0/2) Epoch 21, batch 6500, libri_loss[loss=0.3098, simple_loss=0.3864, pruned_loss=0.1166, over 29260.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3886, pruned_loss=0.1361, over 5636862.03 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3444, pruned_loss=0.09035, over 5514485.91 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3913, pruned_loss=0.1393, over 5635830.46 frames. ], batch size: 94, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:56:39,523 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.371e+02 2.261e+03 3.077e+03 4.734e+03 1.459e+04, threshold=6.155e+03, percent-clipped=29.0
+2023-03-10 18:57:10,760 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-10 18:57:17,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=918894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:57:24,645 INFO [train.py:968] (0/2) Epoch 21, batch 6550, giga_loss[loss=0.302, simple_loss=0.3637, pruned_loss=0.1202, over 28831.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3882, pruned_loss=0.137, over 5642536.45 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3443, pruned_loss=0.09035, over 5521717.59 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3912, pruned_loss=0.1403, over 5637617.32 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:58:13,421 INFO [train.py:968] (0/2) Epoch 21, batch 6600, giga_loss[loss=0.3286, simple_loss=0.3775, pruned_loss=0.1398, over 28678.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.386, pruned_loss=0.1357, over 5645859.72 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3447, pruned_loss=0.09066, over 5529131.24 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.389, pruned_loss=0.1391, over 5637970.71 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:58:17,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.814e+02 1.913e+03 2.362e+03 3.143e+03 6.562e+03, threshold=4.723e+03, percent-clipped=1.0
+2023-03-10 18:59:03,129 INFO [train.py:968] (0/2) Epoch 21, batch 6650, giga_loss[loss=0.3146, simple_loss=0.3843, pruned_loss=0.1225, over 28708.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3849, pruned_loss=0.1341, over 5643196.96 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3444, pruned_loss=0.09054, over 5539485.52 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3888, pruned_loss=0.1383, over 5630170.50 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:59:04,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=919002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:59:38,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=919037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:59:40,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=919040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:59:47,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7802, 2.0237, 1.5712, 2.0879], device='cuda:0'), covar=tensor([0.2464, 0.2595, 0.2913, 0.2322], device='cuda:0'), in_proj_covar=tensor([0.1487, 0.1077, 0.1316, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 18:59:50,709 INFO [train.py:968] (0/2) Epoch 21, batch 6700, giga_loss[loss=0.2915, simple_loss=0.372, pruned_loss=0.1055, over 28994.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3846, pruned_loss=0.1328, over 5646051.65 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3443, pruned_loss=0.09054, over 5536972.19 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3885, pruned_loss=0.1369, over 5639168.50 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:59:53,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.845e+02 1.631e+03 2.065e+03 2.712e+03 9.158e+03, threshold=4.129e+03, percent-clipped=4.0
+2023-03-10 19:00:07,195 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=919069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:00:36,608 INFO [train.py:968] (0/2) Epoch 21, batch 6750, giga_loss[loss=0.3425, simple_loss=0.4037, pruned_loss=0.1407, over 29067.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3867, pruned_loss=0.1344, over 5639236.84 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3443, pruned_loss=0.09066, over 5550321.32 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3916, pruned_loss=0.1393, over 5626093.66 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:01:17,929 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=919145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:01:19,643 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=919148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:01:22,597 INFO [train.py:968] (0/2) Epoch 21, batch 6800, giga_loss[loss=0.2588, simple_loss=0.3428, pruned_loss=0.08745, over 28331.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.385, pruned_loss=0.1328, over 5639685.47 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3447, pruned_loss=0.09084, over 5561619.57 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3901, pruned_loss=0.1383, over 5621976.79 frames. ], batch size: 60, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:01:24,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.773e+03 2.212e+03 3.107e+03 1.004e+04, threshold=4.425e+03, percent-clipped=8.0
+2023-03-10 19:01:51,198 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=919177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:02:09,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3875, 5.2183, 4.9590, 2.6781], device='cuda:0'), covar=tensor([0.0481, 0.0618, 0.0721, 0.1589], device='cuda:0'), in_proj_covar=tensor([0.1214, 0.1125, 0.0954, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-10 19:02:12,781 INFO [train.py:968] (0/2) Epoch 21, batch 6850, giga_loss[loss=0.3032, simple_loss=0.374, pruned_loss=0.1162, over 28709.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3825, pruned_loss=0.1295, over 5651191.26 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3442, pruned_loss=0.09072, over 5571872.49 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3883, pruned_loss=0.1354, over 5630095.90 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:03:02,086 INFO [train.py:968] (0/2) Epoch 21, batch 6900, libri_loss[loss=0.2346, simple_loss=0.3141, pruned_loss=0.07753, over 29351.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3794, pruned_loss=0.1267, over 5658234.93 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3438, pruned_loss=0.09063, over 5578189.15 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3851, pruned_loss=0.1321, over 5637156.25 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:03:04,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.239e+02 1.632e+03 2.013e+03 2.752e+03 1.043e+04, threshold=4.026e+03, percent-clipped=6.0
+2023-03-10 19:03:43,088 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 19:03:51,759 INFO [train.py:968] (0/2) Epoch 21, batch 6950, giga_loss[loss=0.3155, simple_loss=0.3612, pruned_loss=0.1349, over 23779.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3759, pruned_loss=0.124, over 5659011.32 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3434, pruned_loss=0.0905, over 5580587.12 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3811, pruned_loss=0.1287, over 5640957.50 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:04:18,791 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-10 19:04:39,525 INFO [train.py:968] (0/2) Epoch 21, batch 7000, giga_loss[loss=0.2714, simple_loss=0.3491, pruned_loss=0.09687, over 28863.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3739, pruned_loss=0.1226, over 5661128.14 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3435, pruned_loss=0.09071, over 5588500.97 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3787, pruned_loss=0.1269, over 5641324.32 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:04:42,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.588e+02 1.754e+03 2.229e+03 3.013e+03 5.321e+03, threshold=4.459e+03, percent-clipped=7.0
+2023-03-10 19:04:53,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3312, 2.9918, 1.3839, 1.4746], device='cuda:0'), covar=tensor([0.1003, 0.0419, 0.0899, 0.1350], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0555, 0.0384, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 19:05:24,034 INFO [train.py:968] (0/2) Epoch 21, batch 7050, giga_loss[loss=0.3323, simple_loss=0.3843, pruned_loss=0.1401, over 28547.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3724, pruned_loss=0.1216, over 5660942.04 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3435, pruned_loss=0.09064, over 5599448.59 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3773, pruned_loss=0.1262, over 5637103.45 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:06:00,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4208, 2.0276, 1.4224, 0.7914], device='cuda:0'), covar=tensor([0.4649, 0.2370, 0.3148, 0.5525], device='cuda:0'), in_proj_covar=tensor([0.1730, 0.1623, 0.1579, 0.1401], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 19:06:02,034 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=919434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:06:18,198 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-10 19:06:19,001 INFO [train.py:968] (0/2) Epoch 21, batch 7100, giga_loss[loss=0.3589, simple_loss=0.4047, pruned_loss=0.1566, over 27504.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3727, pruned_loss=0.1216, over 5657170.40 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3437, pruned_loss=0.09076, over 5599721.06 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3768, pruned_loss=0.1257, over 5638810.67 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:06:19,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3616, 1.8604, 1.4694, 1.5312], device='cuda:0'), covar=tensor([0.0796, 0.0329, 0.0338, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 19:06:20,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.143e+02 1.599e+03 1.968e+03 2.657e+03 7.862e+03, threshold=3.937e+03, percent-clipped=5.0
+2023-03-10 19:06:34,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7086, 1.6915, 1.9761, 1.4909], device='cuda:0'), covar=tensor([0.1720, 0.2353, 0.1343, 0.1674], device='cuda:0'), in_proj_covar=tensor([0.0893, 0.0698, 0.0936, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 19:06:45,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8061, 2.0329, 1.3887, 1.6067], device='cuda:0'), covar=tensor([0.0909, 0.0517, 0.0991, 0.1009], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0448, 0.0516, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 19:07:13,017 INFO [train.py:968] (0/2) Epoch 21, batch 7150, giga_loss[loss=0.2748, simple_loss=0.3489, pruned_loss=0.1004, over 28683.00 frames. ], tot_loss[loss=0.304, simple_loss=0.37, pruned_loss=0.119, over 5661708.81 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3435, pruned_loss=0.09055, over 5606525.80 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3741, pruned_loss=0.1231, over 5642196.61 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:08:09,367 INFO [train.py:968] (0/2) Epoch 21, batch 7200, giga_loss[loss=0.3326, simple_loss=0.4077, pruned_loss=0.1288, over 28634.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3709, pruned_loss=0.1174, over 5656275.94 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3437, pruned_loss=0.09092, over 5603110.08 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3745, pruned_loss=0.1209, over 5645088.35 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:08:13,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.441e+02 1.490e+03 2.058e+03 2.929e+03 2.456e+04, threshold=4.116e+03, percent-clipped=11.0
+2023-03-10 19:08:17,294 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 19:08:58,829 INFO [train.py:968] (0/2) Epoch 21, batch 7250, giga_loss[loss=0.3181, simple_loss=0.3869, pruned_loss=0.1246, over 28495.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3738, pruned_loss=0.1181, over 5674661.89 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3437, pruned_loss=0.09094, over 5605561.00 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3769, pruned_loss=0.1211, over 5664218.80 frames. ], batch size: 78, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:09:44,168 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-10 19:09:48,141 INFO [train.py:968] (0/2) Epoch 21, batch 7300, giga_loss[loss=0.291, simple_loss=0.3684, pruned_loss=0.1069, over 28899.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3741, pruned_loss=0.1194, over 5669001.47 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3429, pruned_loss=0.09064, over 5614788.89 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3784, pruned_loss=0.123, over 5653993.94 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:09:51,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.129e+03 1.661e+03 2.129e+03 3.107e+03 8.003e+03, threshold=4.258e+03, percent-clipped=8.0
+2023-03-10 19:10:34,861 INFO [train.py:968] (0/2) Epoch 21, batch 7350, giga_loss[loss=0.3033, simple_loss=0.3734, pruned_loss=0.1167, over 29089.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3736, pruned_loss=0.1197, over 5669642.70 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3431, pruned_loss=0.09087, over 5615667.74 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3779, pruned_loss=0.1233, over 5657888.92 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:10:35,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7001, 1.7510, 1.9023, 1.4744], device='cuda:0'), covar=tensor([0.1827, 0.2511, 0.1433, 0.1724], device='cuda:0'), in_proj_covar=tensor([0.0893, 0.0698, 0.0936, 0.0835], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 19:10:38,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=919706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:11:19,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4487, 1.7367, 1.3359, 1.5797], device='cuda:0'), covar=tensor([0.2606, 0.2722, 0.3042, 0.2255], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1080, 0.1321, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 19:11:22,072 INFO [train.py:968] (0/2) Epoch 21, batch 7400, giga_loss[loss=0.3383, simple_loss=0.3903, pruned_loss=0.1432, over 29096.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3718, pruned_loss=0.1193, over 5672148.73 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3432, pruned_loss=0.09083, over 5616929.97 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3757, pruned_loss=0.1227, over 5662426.29 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:11:27,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.562e+03 1.957e+03 2.575e+03 5.368e+03, threshold=3.914e+03, percent-clipped=3.0
+2023-03-10 19:12:06,801 INFO [train.py:968] (0/2) Epoch 21, batch 7450, giga_loss[loss=0.2514, simple_loss=0.3361, pruned_loss=0.08337, over 28970.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3706, pruned_loss=0.1196, over 5673689.97 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3429, pruned_loss=0.09056, over 5622633.74 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3745, pruned_loss=0.1231, over 5661807.60 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:12:13,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=919809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:12:57,645 INFO [train.py:968] (0/2) Epoch 21, batch 7500, giga_loss[loss=0.2956, simple_loss=0.3725, pruned_loss=0.1094, over 28992.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3702, pruned_loss=0.1185, over 5672040.04 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3429, pruned_loss=0.0906, over 5626556.11 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3736, pruned_loss=0.1216, over 5659928.74 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:13:02,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.750e+02 1.666e+03 2.151e+03 3.159e+03 7.709e+03, threshold=4.302e+03, percent-clipped=11.0
+2023-03-10 19:13:43,947 INFO [train.py:968] (0/2) Epoch 21, batch 7550, giga_loss[loss=0.3085, simple_loss=0.3958, pruned_loss=0.1107, over 28983.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3691, pruned_loss=0.1163, over 5663146.83 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.343, pruned_loss=0.09062, over 5623945.70 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3726, pruned_loss=0.1196, over 5656938.30 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:14:28,852 INFO [train.py:968] (0/2) Epoch 21, batch 7600, giga_loss[loss=0.2945, simple_loss=0.3602, pruned_loss=0.1144, over 29115.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3694, pruned_loss=0.1161, over 5673916.98 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.343, pruned_loss=0.09062, over 5628115.21 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3727, pruned_loss=0.1193, over 5665824.94 frames. ], batch size: 113, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:14:29,279 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 19:14:29,773 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=919952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:14:32,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=919955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:14:32,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.626e+03 2.269e+03 3.237e+03 1.217e+04, threshold=4.537e+03, percent-clipped=13.0
+2023-03-10 19:14:53,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6410, 1.7401, 1.3613, 1.3019], device='cuda:0'), covar=tensor([0.0970, 0.0659, 0.1041, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0448, 0.0515, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 19:14:59,064 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=919984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:15:12,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=919997.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:15:15,196 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-920000.pt
+2023-03-10 19:15:16,474 INFO [train.py:968] (0/2) Epoch 21, batch 7650, giga_loss[loss=0.2738, simple_loss=0.3497, pruned_loss=0.09898, over 28656.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3679, pruned_loss=0.1154, over 5681522.29 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3429, pruned_loss=0.09055, over 5632323.21 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3709, pruned_loss=0.1182, over 5672137.40 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:15:28,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6710, 1.7752, 1.2717, 1.3209], device='cuda:0'), covar=tensor([0.0948, 0.0605, 0.1081, 0.1172], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0448, 0.0515, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 19:16:07,371 INFO [train.py:968] (0/2) Epoch 21, batch 7700, giga_loss[loss=0.2662, simple_loss=0.3422, pruned_loss=0.09515, over 28312.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3673, pruned_loss=0.1163, over 5666337.09 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3431, pruned_loss=0.09053, over 5635774.42 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3699, pruned_loss=0.1189, over 5656492.05 frames. ], batch size: 65, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:16:11,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.587e+03 1.808e+03 2.285e+03 7.273e+03, threshold=3.615e+03, percent-clipped=5.0
+2023-03-10 19:16:35,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=920081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:16:46,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9885, 1.3122, 1.0855, 0.2488], device='cuda:0'), covar=tensor([0.3808, 0.3120, 0.4754, 0.6470], device='cuda:0'), in_proj_covar=tensor([0.1738, 0.1637, 0.1590, 0.1409], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 19:16:53,595 INFO [train.py:968] (0/2) Epoch 21, batch 7750, giga_loss[loss=0.2945, simple_loss=0.3565, pruned_loss=0.1163, over 28719.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3668, pruned_loss=0.1166, over 5672950.19 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3434, pruned_loss=0.0906, over 5642513.07 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3691, pruned_loss=0.1192, over 5659823.65 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:17:38,487 INFO [train.py:968] (0/2) Epoch 21, batch 7800, giga_loss[loss=0.2931, simple_loss=0.362, pruned_loss=0.1121, over 28764.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.366, pruned_loss=0.1166, over 5672990.68 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3429, pruned_loss=0.09019, over 5652159.76 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3692, pruned_loss=0.1201, over 5654572.53 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:17:44,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.618e+03 2.091e+03 2.640e+03 7.960e+03, threshold=4.183e+03, percent-clipped=16.0
+2023-03-10 19:18:21,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6965, 1.7640, 1.9479, 1.4854], device='cuda:0'), covar=tensor([0.1862, 0.2562, 0.1513, 0.1833], device='cuda:0'), in_proj_covar=tensor([0.0893, 0.0699, 0.0937, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 19:18:29,939 INFO [train.py:968] (0/2) Epoch 21, batch 7850, giga_loss[loss=0.3187, simple_loss=0.3607, pruned_loss=0.1384, over 23667.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3651, pruned_loss=0.1168, over 5661854.22 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3433, pruned_loss=0.09047, over 5652068.04 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3676, pruned_loss=0.1197, over 5647661.84 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:18:44,675 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=920216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:18:51,588 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=920224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:18:53,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=920227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:19:18,206 INFO [train.py:968] (0/2) Epoch 21, batch 7900, giga_loss[loss=0.3464, simple_loss=0.3916, pruned_loss=0.1506, over 27719.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3644, pruned_loss=0.1169, over 5664537.14 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3431, pruned_loss=0.09046, over 5654935.01 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3667, pruned_loss=0.1195, over 5650899.92 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:19:22,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.693e+03 2.259e+03 3.112e+03 6.859e+03, threshold=4.517e+03, percent-clipped=9.0
+2023-03-10 19:19:23,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=920256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:19:31,506 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-10 19:20:00,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 19:20:04,743 INFO [train.py:968] (0/2) Epoch 21, batch 7950, giga_loss[loss=0.3267, simple_loss=0.3877, pruned_loss=0.1328, over 28836.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3655, pruned_loss=0.117, over 5668638.81 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3432, pruned_loss=0.09055, over 5661039.27 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3678, pruned_loss=0.1196, over 5652341.35 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:20:08,143 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 19:20:50,263 INFO [train.py:968] (0/2) Epoch 21, batch 8000, giga_loss[loss=0.3292, simple_loss=0.3821, pruned_loss=0.1382, over 27613.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3663, pruned_loss=0.1164, over 5668981.96 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3438, pruned_loss=0.0909, over 5659709.61 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3683, pruned_loss=0.119, over 5656819.36 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:20:55,669 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.131e+03 1.729e+03 2.252e+03 3.139e+03 7.996e+03, threshold=4.504e+03, percent-clipped=8.0
+2023-03-10 19:21:06,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=920372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:21:26,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5824, 2.0291, 1.5422, 1.8459], device='cuda:0'), covar=tensor([0.2838, 0.2708, 0.3200, 0.2369], device='cuda:0'), in_proj_covar=tensor([0.1490, 0.1079, 0.1316, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 19:21:32,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1846, 1.3794, 1.3003, 1.1633], device='cuda:0'), covar=tensor([0.2382, 0.2231, 0.1519, 0.1941], device='cuda:0'), in_proj_covar=tensor([0.1950, 0.1877, 0.1804, 0.1944], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 19:21:33,957 INFO [train.py:968] (0/2) Epoch 21, batch 8050, giga_loss[loss=0.2677, simple_loss=0.3429, pruned_loss=0.09632, over 28717.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3654, pruned_loss=0.1146, over 5677748.19 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3438, pruned_loss=0.09088, over 5656401.17 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3675, pruned_loss=0.1173, over 5670658.53 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:22:08,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0845, 1.2683, 1.0258, 0.8334], device='cuda:0'), covar=tensor([0.0884, 0.0420, 0.1004, 0.1239], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0449, 0.0516, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 19:22:18,898 INFO [train.py:968] (0/2) Epoch 21, batch 8100, giga_loss[loss=0.3014, simple_loss=0.3681, pruned_loss=0.1174, over 28852.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3643, pruned_loss=0.1138, over 5681830.88 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3434, pruned_loss=0.09056, over 5663347.44 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.367, pruned_loss=0.1169, over 5670239.67 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:22:24,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.310e+02 1.475e+03 1.909e+03 2.834e+03 9.519e+03, threshold=3.819e+03, percent-clipped=5.0
+2023-03-10 19:23:07,379 INFO [train.py:968] (0/2) Epoch 21, batch 8150, giga_loss[loss=0.3338, simple_loss=0.3887, pruned_loss=0.1395, over 28808.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3667, pruned_loss=0.1163, over 5673580.79 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3433, pruned_loss=0.09041, over 5666415.62 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3696, pruned_loss=0.1197, over 5661915.18 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:23:21,421 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=920515.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:23:23,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=920518.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:23:57,070 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=920547.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:24:01,296 INFO [train.py:968] (0/2) Epoch 21, batch 8200, giga_loss[loss=0.2632, simple_loss=0.3335, pruned_loss=0.09644, over 28978.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3703, pruned_loss=0.1211, over 5659944.75 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3433, pruned_loss=0.09041, over 5668843.49 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3728, pruned_loss=0.1241, over 5648730.45 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:24:08,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.849e+03 2.374e+03 3.165e+03 6.794e+03, threshold=4.749e+03, percent-clipped=15.0
+2023-03-10 19:24:34,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7256, 1.8392, 1.3379, 1.4566], device='cuda:0'), covar=tensor([0.0897, 0.0628, 0.1026, 0.1094], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0448, 0.0515, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 19:24:39,924 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=920591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:24:49,000 INFO [train.py:968] (0/2) Epoch 21, batch 8250, giga_loss[loss=0.3022, simple_loss=0.3686, pruned_loss=0.1179, over 29027.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1227, over 5669001.32 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3433, pruned_loss=0.09034, over 5675079.39 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.374, pruned_loss=0.126, over 5654261.18 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:25:33,635 INFO [train.py:968] (0/2) Epoch 21, batch 8300, giga_loss[loss=0.2932, simple_loss=0.3612, pruned_loss=0.1126, over 29082.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.372, pruned_loss=0.1234, over 5673289.14 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3439, pruned_loss=0.09067, over 5682136.85 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3748, pruned_loss=0.127, over 5654964.14 frames. ], batch size: 128, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:25:42,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.924e+03 2.306e+03 3.240e+03 1.545e+04, threshold=4.612e+03, percent-clipped=12.0
+2023-03-10 19:25:59,712 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-10 19:26:21,609 INFO [train.py:968] (0/2) Epoch 21, batch 8350, giga_loss[loss=0.3238, simple_loss=0.3867, pruned_loss=0.1305, over 28950.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3702, pruned_loss=0.1219, over 5675980.12 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3438, pruned_loss=0.09064, over 5687603.32 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3731, pruned_loss=0.1256, over 5656229.78 frames. ], batch size: 145, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:26:32,765 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=920715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:26:49,773 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=920734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:26:52,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=920737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:27:02,928 INFO [train.py:968] (0/2) Epoch 21, batch 8400, giga_loss[loss=0.2835, simple_loss=0.3593, pruned_loss=0.1038, over 28912.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3686, pruned_loss=0.12, over 5688261.28 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3442, pruned_loss=0.09079, over 5696041.22 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3718, pruned_loss=0.1242, over 5663925.92 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:27:03,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4693, 1.8290, 1.7670, 1.5793], device='cuda:0'), covar=tensor([0.2048, 0.2072, 0.2447, 0.2433], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0752, 0.0716, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 19:27:07,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.641e+03 2.171e+03 3.351e+03 7.109e+03, threshold=4.342e+03, percent-clipped=8.0
+2023-03-10 19:27:15,409 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=920766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:27:42,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3437, 1.6868, 1.4076, 1.6086], device='cuda:0'), covar=tensor([0.0783, 0.0349, 0.0334, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 19:27:46,890 INFO [train.py:968] (0/2) Epoch 21, batch 8450, libri_loss[loss=0.2731, simple_loss=0.3481, pruned_loss=0.09905, over 21084.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3671, pruned_loss=0.1167, over 5689789.97 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3439, pruned_loss=0.09054, over 5695362.94 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3708, pruned_loss=0.1213, over 5671263.41 frames. ], batch size: 188, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:27:54,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=920809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:28:31,990 INFO [train.py:968] (0/2) Epoch 21, batch 8500, giga_loss[loss=0.3261, simple_loss=0.3644, pruned_loss=0.1439, over 24098.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3657, pruned_loss=0.1157, over 5687010.56 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3443, pruned_loss=0.09071, over 5700990.49 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3688, pruned_loss=0.1198, over 5667060.55 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:28:36,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.537e+03 1.885e+03 2.461e+03 5.376e+03, threshold=3.771e+03, percent-clipped=1.0
+2023-03-10 19:29:01,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-10 19:29:12,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4228, 1.4268, 3.8610, 3.2734], device='cuda:0'), covar=tensor([0.1568, 0.2630, 0.0465, 0.1006], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0646, 0.0958, 0.0906], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 19:29:14,231 INFO [train.py:968] (0/2) Epoch 21, batch 8550, giga_loss[loss=0.2802, simple_loss=0.3569, pruned_loss=0.1017, over 28310.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3639, pruned_loss=0.1149, over 5673324.69 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3444, pruned_loss=0.09086, over 5691916.21 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3671, pruned_loss=0.1189, over 5664596.45 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:29:35,139 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 19:30:01,406 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-10 19:30:03,581 INFO [train.py:968] (0/2) Epoch 21, batch 8600, giga_loss[loss=0.3897, simple_loss=0.4048, pruned_loss=0.1874, over 23575.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3633, pruned_loss=0.1157, over 5676425.72 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3441, pruned_loss=0.09064, over 5695087.28 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3665, pruned_loss=0.1196, over 5666265.14 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:30:09,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.401e+02 1.518e+03 1.884e+03 2.644e+03 7.703e+03, threshold=3.768e+03, percent-clipped=10.0
+2023-03-10 19:30:32,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-10 19:30:49,965 INFO [train.py:968] (0/2) Epoch 21, batch 8650, giga_loss[loss=0.2846, simple_loss=0.3537, pruned_loss=0.1077, over 28507.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3655, pruned_loss=0.1171, over 5674985.98 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3448, pruned_loss=0.09107, over 5692574.59 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.368, pruned_loss=0.1207, over 5668745.95 frames. ], batch size: 60, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:31:22,179 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-10 19:31:22,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 19:31:34,474 INFO [train.py:968] (0/2) Epoch 21, batch 8700, giga_loss[loss=0.3388, simple_loss=0.4265, pruned_loss=0.1255, over 29045.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3696, pruned_loss=0.1186, over 5679934.22 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.345, pruned_loss=0.09122, over 5699106.23 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3724, pruned_loss=0.1225, over 5668502.41 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:31:37,647 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:31:43,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.516e+02 1.700e+03 2.238e+03 3.270e+03 5.839e+03, threshold=4.475e+03, percent-clipped=17.0
+2023-03-10 19:32:10,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:32:19,540 INFO [train.py:968] (0/2) Epoch 21, batch 8750, giga_loss[loss=0.3244, simple_loss=0.3926, pruned_loss=0.1281, over 28594.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3717, pruned_loss=0.1174, over 5680433.28 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3448, pruned_loss=0.09105, over 5702337.91 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3749, pruned_loss=0.1216, over 5667648.30 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:32:35,180 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:33:00,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2503, 1.6043, 1.2508, 0.9738], device='cuda:0'), covar=tensor([0.2585, 0.2595, 0.2967, 0.2230], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1081, 0.1318, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 19:33:04,192 INFO [train.py:968] (0/2) Epoch 21, batch 8800, giga_loss[loss=0.2847, simple_loss=0.3646, pruned_loss=0.1024, over 28920.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.372, pruned_loss=0.1171, over 5690940.86 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3443, pruned_loss=0.09071, over 5709192.50 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3761, pruned_loss=0.1216, over 5673679.42 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:33:08,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2599, 1.6619, 1.2433, 0.7579], device='cuda:0'), covar=tensor([0.4538, 0.2745, 0.2709, 0.5424], device='cuda:0'), in_proj_covar=tensor([0.1733, 0.1636, 0.1580, 0.1402], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 19:33:13,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.970e+02 1.547e+03 2.065e+03 2.842e+03 1.463e+04, threshold=4.130e+03, percent-clipped=11.0
+2023-03-10 19:33:15,006 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:33:27,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6070, 1.8062, 1.6423, 1.4257], device='cuda:0'), covar=tensor([0.2721, 0.2357, 0.2309, 0.2630], device='cuda:0'), in_proj_covar=tensor([0.1948, 0.1881, 0.1807, 0.1950], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 19:33:35,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:33:49,866 INFO [train.py:968] (0/2) Epoch 21, batch 8850, giga_loss[loss=0.2615, simple_loss=0.3455, pruned_loss=0.08871, over 28878.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3734, pruned_loss=0.1188, over 5693212.21 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3442, pruned_loss=0.09076, over 5712001.45 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3771, pruned_loss=0.1227, over 5676673.56 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:33:54,058 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2980, 1.2876, 1.2113, 1.4375], device='cuda:0'), covar=tensor([0.0729, 0.0368, 0.0326, 0.0787], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 19:33:55,962 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:34:15,886 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:34:20,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:34:22,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0373, 1.1259, 3.3470, 3.0576], device='cuda:0'), covar=tensor([0.1715, 0.2755, 0.0533, 0.1053], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0647, 0.0959, 0.0908], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 19:34:22,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:34:36,242 INFO [train.py:968] (0/2) Epoch 21, batch 8900, giga_loss[loss=0.2604, simple_loss=0.3375, pruned_loss=0.09168, over 28636.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3733, pruned_loss=0.1194, over 5687342.46 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3444, pruned_loss=0.09074, over 5711965.32 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3767, pruned_loss=0.1232, over 5673728.58 frames. ], batch size: 85, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:34:44,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.646e+03 2.187e+03 2.877e+03 1.308e+04, threshold=4.374e+03, percent-clipped=8.0
+2023-03-10 19:34:50,238 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:35:22,298 INFO [train.py:968] (0/2) Epoch 21, batch 8950, giga_loss[loss=0.3023, simple_loss=0.3629, pruned_loss=0.1209, over 28970.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3738, pruned_loss=0.121, over 5689134.55 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3445, pruned_loss=0.09079, over 5711237.93 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3768, pruned_loss=0.1244, over 5678742.12 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:35:49,851 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:35:53,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:35:58,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:36:05,972 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-10 19:36:11,433 INFO [train.py:968] (0/2) Epoch 21, batch 9000, giga_loss[loss=0.2737, simple_loss=0.3464, pruned_loss=0.1005, over 28780.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3718, pruned_loss=0.1203, over 5676146.93 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3449, pruned_loss=0.09101, over 5700998.91 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3741, pruned_loss=0.1231, over 5677673.05 frames. ], batch size: 99, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:36:11,437 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 19:36:19,998 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2074, simple_loss=0.3156, pruned_loss=0.04961, over 944034.00 frames. 
+2023-03-10 19:36:19,999 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 19:36:26,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.687e+03 2.046e+03 2.521e+03 5.610e+03, threshold=4.093e+03, percent-clipped=6.0
+2023-03-10 19:36:26,494 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:36:34,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4716, 1.6200, 1.3526, 1.5115], device='cuda:0'), covar=tensor([0.0747, 0.0327, 0.0342, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 19:37:08,670 INFO [train.py:968] (0/2) Epoch 21, batch 9050, giga_loss[loss=0.3196, simple_loss=0.3787, pruned_loss=0.1302, over 27867.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3704, pruned_loss=0.1208, over 5667620.28 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.345, pruned_loss=0.09113, over 5703045.85 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3724, pruned_loss=0.1231, over 5666860.61 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:37:18,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3140, 3.5967, 1.5188, 1.5536], device='cuda:0'), covar=tensor([0.1059, 0.0383, 0.0862, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0555, 0.0384, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 19:37:19,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3650, 4.2059, 4.0055, 1.9237], device='cuda:0'), covar=tensor([0.0542, 0.0696, 0.0745, 0.2035], device='cuda:0'), in_proj_covar=tensor([0.1217, 0.1131, 0.0960, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 19:37:19,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5604, 1.7895, 1.4541, 1.6705], device='cuda:0'), covar=tensor([0.2583, 0.2670, 0.2855, 0.2453], device='cuda:0'), in_proj_covar=tensor([0.1495, 0.1081, 0.1321, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 19:37:28,160 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:37:37,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:37:43,255 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6796, 1.9595, 1.6273, 1.7862], device='cuda:0'), covar=tensor([0.2601, 0.2638, 0.2952, 0.2388], device='cuda:0'), in_proj_covar=tensor([0.1494, 0.1080, 0.1320, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 19:37:58,045 INFO [train.py:968] (0/2) Epoch 21, batch 9100, giga_loss[loss=0.3022, simple_loss=0.3659, pruned_loss=0.1193, over 29091.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3689, pruned_loss=0.1201, over 5665238.93 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3451, pruned_loss=0.09118, over 5695298.62 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3706, pruned_loss=0.1221, over 5672024.51 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:38:07,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.208e+03 1.761e+03 2.188e+03 3.069e+03 1.140e+04, threshold=4.376e+03, percent-clipped=13.0
+2023-03-10 19:38:36,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:38:48,806 INFO [train.py:968] (0/2) Epoch 21, batch 9150, giga_loss[loss=0.2902, simple_loss=0.3576, pruned_loss=0.1114, over 28749.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3705, pruned_loss=0.1215, over 5671936.55 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3454, pruned_loss=0.09135, over 5702642.66 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3723, pruned_loss=0.1239, over 5670110.76 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:38:52,885 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:39:20,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:39:36,193 INFO [train.py:968] (0/2) Epoch 21, batch 9200, giga_loss[loss=0.3269, simple_loss=0.3835, pruned_loss=0.1352, over 27665.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1213, over 5673123.87 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3453, pruned_loss=0.09133, over 5706891.69 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3712, pruned_loss=0.1238, over 5667439.34 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:39:45,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.929e+02 1.625e+03 1.977e+03 2.807e+03 1.155e+04, threshold=3.954e+03, percent-clipped=9.0
+2023-03-10 19:39:55,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:39:57,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:08,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:21,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5617, 4.4559, 1.6063, 1.8717], device='cuda:0'), covar=tensor([0.0957, 0.0244, 0.0874, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0556, 0.0384, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 19:40:25,526 INFO [train.py:968] (0/2) Epoch 21, batch 9250, giga_loss[loss=0.335, simple_loss=0.3949, pruned_loss=0.1376, over 28313.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3681, pruned_loss=0.1206, over 5679570.94 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3455, pruned_loss=0.09138, over 5708800.65 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3697, pruned_loss=0.1228, over 5673140.71 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:40:27,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:27,689 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:52,475 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:54,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:57,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921641.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:41:06,870 INFO [train.py:968] (0/2) Epoch 21, batch 9300, giga_loss[loss=0.3207, simple_loss=0.3877, pruned_loss=0.1269, over 28810.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3683, pruned_loss=0.1199, over 5685577.42 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3453, pruned_loss=0.09129, over 5714421.42 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3705, pruned_loss=0.1227, over 5674553.34 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:41:15,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8019, 1.0027, 2.8664, 2.8040], device='cuda:0'), covar=tensor([0.1801, 0.2754, 0.0689, 0.1127], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0646, 0.0960, 0.0908], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 19:41:19,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.637e+03 2.067e+03 2.798e+03 7.736e+03, threshold=4.134e+03, percent-clipped=11.0
+2023-03-10 19:41:21,985 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3887, 1.9058, 1.4556, 1.5938], device='cuda:0'), covar=tensor([0.0771, 0.0308, 0.0327, 0.0862], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 19:41:26,059 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:41:38,685 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:41:41,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:00,031 INFO [train.py:968] (0/2) Epoch 21, batch 9350, giga_loss[loss=0.2938, simple_loss=0.3665, pruned_loss=0.1105, over 28665.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3692, pruned_loss=0.1199, over 5681414.64 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3456, pruned_loss=0.09143, over 5717687.44 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3712, pruned_loss=0.1225, over 5669009.54 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:42:07,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:08,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:25,713 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:27,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:32,765 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:40,913 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:43,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:44,529 INFO [train.py:968] (0/2) Epoch 21, batch 9400, giga_loss[loss=0.3042, simple_loss=0.374, pruned_loss=0.1172, over 28984.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3683, pruned_loss=0.1194, over 5672053.75 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3458, pruned_loss=0.09161, over 5711928.56 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3702, pruned_loss=0.122, over 5665787.47 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:42:51,914 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:53,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.734e+02 1.623e+03 2.129e+03 3.079e+03 1.058e+04, threshold=4.257e+03, percent-clipped=12.0
+2023-03-10 19:43:12,365 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:43:26,426 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:43:32,527 INFO [train.py:968] (0/2) Epoch 21, batch 9450, giga_loss[loss=0.2935, simple_loss=0.3793, pruned_loss=0.1039, over 29024.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3693, pruned_loss=0.1186, over 5676290.67 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3457, pruned_loss=0.09159, over 5714857.77 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3712, pruned_loss=0.121, over 5668139.25 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:44:17,188 INFO [train.py:968] (0/2) Epoch 21, batch 9500, giga_loss[loss=0.4214, simple_loss=0.4455, pruned_loss=0.1987, over 26546.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3706, pruned_loss=0.1172, over 5682860.65 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3457, pruned_loss=0.09166, over 5718617.33 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3725, pruned_loss=0.1195, over 5672499.57 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:44:18,034 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:44:20,453 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:44:20,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:44:25,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.821e+02 1.405e+03 1.805e+03 2.368e+03 7.826e+03, threshold=3.610e+03, percent-clipped=4.0
+2023-03-10 19:44:37,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4563, 1.8874, 1.7796, 1.4691], device='cuda:0'), covar=tensor([0.1972, 0.1962, 0.2156, 0.2539], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0748, 0.0712, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 19:44:43,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:44:46,204 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:45:04,205 INFO [train.py:968] (0/2) Epoch 21, batch 9550, giga_loss[loss=0.3166, simple_loss=0.3854, pruned_loss=0.124, over 28743.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3735, pruned_loss=0.1178, over 5679064.61 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.346, pruned_loss=0.09175, over 5716845.46 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3752, pruned_loss=0.1199, over 5672014.43 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:45:42,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:45:44,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:45:53,531 INFO [train.py:968] (0/2) Epoch 21, batch 9600, libri_loss[loss=0.219, simple_loss=0.2994, pruned_loss=0.06931, over 29361.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3753, pruned_loss=0.1198, over 5679996.64 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3452, pruned_loss=0.09138, over 5720428.02 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3782, pruned_loss=0.1226, over 5669881.85 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:46:01,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.765e+02 1.702e+03 2.386e+03 3.275e+03 9.258e+03, threshold=4.771e+03, percent-clipped=19.0
+2023-03-10 19:46:10,471 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:46:35,105 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-922000.pt
+2023-03-10 19:46:36,827 INFO [train.py:968] (0/2) Epoch 21, batch 9650, giga_loss[loss=0.3898, simple_loss=0.4245, pruned_loss=0.1775, over 27638.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3771, pruned_loss=0.1222, over 5671047.63 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3458, pruned_loss=0.09183, over 5711838.91 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3797, pruned_loss=0.1249, over 5669549.79 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:46:51,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=922016.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:47:00,164 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=922023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:47:03,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=922026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:47:22,484 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 19:47:26,266 INFO [train.py:968] (0/2) Epoch 21, batch 9700, giga_loss[loss=0.45, simple_loss=0.4561, pruned_loss=0.222, over 26635.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3774, pruned_loss=0.1238, over 5670549.84 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3458, pruned_loss=0.09189, over 5716129.54 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3802, pruned_loss=0.1266, over 5664746.24 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:47:30,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=922055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:47:36,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.912e+02 1.782e+03 2.398e+03 3.733e+03 1.721e+04, threshold=4.797e+03, percent-clipped=11.0
+2023-03-10 19:48:14,176 INFO [train.py:968] (0/2) Epoch 21, batch 9750, giga_loss[loss=0.316, simple_loss=0.3794, pruned_loss=0.1263, over 28655.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3752, pruned_loss=0.1225, over 5655462.80 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3458, pruned_loss=0.09192, over 5718945.05 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3779, pruned_loss=0.1251, over 5647578.09 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:48:21,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=922111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:48:27,224 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2343, 3.0344, 1.4183, 1.4271], device='cuda:0'), covar=tensor([0.1051, 0.0385, 0.0969, 0.1438], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0556, 0.0384, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-10 19:48:37,611 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-10 19:48:55,872 INFO [train.py:968] (0/2) Epoch 21, batch 9800, giga_loss[loss=0.2971, simple_loss=0.3764, pruned_loss=0.1089, over 28780.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3734, pruned_loss=0.1196, over 5658194.85 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3455, pruned_loss=0.0918, over 5714484.69 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3763, pruned_loss=0.1224, over 5655007.86 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:49:03,730 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=922159.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:49:05,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.712e+02 1.692e+03 2.053e+03 2.899e+03 1.127e+04, threshold=4.107e+03, percent-clipped=6.0
+2023-03-10 19:49:05,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=922162.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:49:27,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3370, 2.9640, 1.4044, 1.5043], device='cuda:0'), covar=tensor([0.0968, 0.0394, 0.0928, 0.1284], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0556, 0.0384, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 19:49:30,068 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=922191.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:49:38,220 INFO [train.py:968] (0/2) Epoch 21, batch 9850, giga_loss[loss=0.3331, simple_loss=0.396, pruned_loss=0.1351, over 28764.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3736, pruned_loss=0.1182, over 5658686.69 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3457, pruned_loss=0.09182, over 5708524.76 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3765, pruned_loss=0.1212, over 5660436.75 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:50:04,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=922230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:50:28,107 INFO [train.py:968] (0/2) Epoch 21, batch 9900, giga_loss[loss=0.318, simple_loss=0.3846, pruned_loss=0.1257, over 28867.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3743, pruned_loss=0.1188, over 5663467.77 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3455, pruned_loss=0.0919, over 5711399.09 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3772, pruned_loss=0.1214, over 5661729.52 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:50:30,963 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=922254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:50:34,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=922257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:50:36,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.709e+03 2.333e+03 3.552e+03 1.257e+04, threshold=4.665e+03, percent-clipped=16.0
+2023-03-10 19:50:59,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=922286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:51:14,871 INFO [train.py:968] (0/2) Epoch 21, batch 9950, giga_loss[loss=0.3073, simple_loss=0.3744, pruned_loss=0.1201, over 28903.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3734, pruned_loss=0.1188, over 5660675.55 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3453, pruned_loss=0.09176, over 5718359.05 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3768, pruned_loss=0.1219, over 5651457.56 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:51:15,400 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-10 19:51:15,664 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=922302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:51:59,058 INFO [train.py:968] (0/2) Epoch 21, batch 10000, giga_loss[loss=0.2776, simple_loss=0.3547, pruned_loss=0.1003, over 28970.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3713, pruned_loss=0.1175, over 5678554.29 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3455, pruned_loss=0.09183, over 5723368.12 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3747, pruned_loss=0.1207, over 5665258.81 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:52:09,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.585e+03 2.138e+03 2.637e+03 7.768e+03, threshold=4.277e+03, percent-clipped=3.0
+2023-03-10 19:52:22,273 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=922373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:52:24,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=922376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:52:47,821 INFO [train.py:968] (0/2) Epoch 21, batch 10050, giga_loss[loss=0.393, simple_loss=0.4075, pruned_loss=0.1893, over 23878.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3707, pruned_loss=0.1184, over 5668926.17 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3451, pruned_loss=0.09162, over 5728763.25 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3745, pruned_loss=0.1219, over 5652249.58 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:52:51,413 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=922405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:53:06,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1963, 1.4749, 1.4833, 1.0711], device='cuda:0'), covar=tensor([0.1702, 0.2610, 0.1384, 0.1626], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0706, 0.0943, 0.0841], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 19:53:35,331 INFO [train.py:968] (0/2) Epoch 21, batch 10100, libri_loss[loss=0.2602, simple_loss=0.3499, pruned_loss=0.08524, over 29224.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3688, pruned_loss=0.1174, over 5674496.57 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3453, pruned_loss=0.09157, over 5730878.34 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3724, pruned_loss=0.1211, over 5657329.49 frames. ], batch size: 94, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:53:45,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.149e+03 1.609e+03 1.999e+03 2.775e+03 7.012e+03, threshold=3.998e+03, percent-clipped=8.0
+2023-03-10 19:54:20,520 INFO [train.py:968] (0/2) Epoch 21, batch 10150, giga_loss[loss=0.3597, simple_loss=0.3883, pruned_loss=0.1656, over 23433.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3673, pruned_loss=0.1172, over 5681865.55 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3451, pruned_loss=0.09167, over 5738570.05 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3716, pruned_loss=0.1214, over 5658347.66 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:54:32,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3865, 1.2861, 3.7778, 3.3312], device='cuda:0'), covar=tensor([0.1482, 0.2650, 0.0440, 0.1953], device='cuda:0'), in_proj_covar=tensor([0.0763, 0.0649, 0.0963, 0.0913], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 19:55:08,668 INFO [train.py:968] (0/2) Epoch 21, batch 10200, giga_loss[loss=0.2881, simple_loss=0.3576, pruned_loss=0.1093, over 28727.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3688, pruned_loss=0.1195, over 5679172.50 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3449, pruned_loss=0.09166, over 5740509.97 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3725, pruned_loss=0.1231, over 5658372.64 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:55:16,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6078, 1.8017, 1.4977, 1.7634], device='cuda:0'), covar=tensor([0.2479, 0.2581, 0.2842, 0.2292], device='cuda:0'), in_proj_covar=tensor([0.1496, 0.1082, 0.1319, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 19:55:17,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 1.652e+03 2.015e+03 2.738e+03 6.749e+03, threshold=4.029e+03, percent-clipped=6.0
+2023-03-10 19:55:58,833 INFO [train.py:968] (0/2) Epoch 21, batch 10250, giga_loss[loss=0.3432, simple_loss=0.3918, pruned_loss=0.1474, over 26590.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3673, pruned_loss=0.1181, over 5672859.21 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3449, pruned_loss=0.09166, over 5740509.97 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3702, pruned_loss=0.1209, over 5656670.41 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:56:04,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4595, 1.6083, 1.5695, 1.3577], device='cuda:0'), covar=tensor([0.2579, 0.2238, 0.2041, 0.2434], device='cuda:0'), in_proj_covar=tensor([0.1955, 0.1890, 0.1823, 0.1961], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 19:56:08,423 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=922612.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:56:12,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1662, 1.2020, 3.5833, 3.0894], device='cuda:0'), covar=tensor([0.1766, 0.2999, 0.0475, 0.1077], device='cuda:0'), in_proj_covar=tensor([0.0762, 0.0647, 0.0961, 0.0911], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 19:56:28,950 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 19:56:43,977 INFO [train.py:968] (0/2) Epoch 21, batch 10300, giga_loss[loss=0.2813, simple_loss=0.3594, pruned_loss=0.1016, over 28630.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3634, pruned_loss=0.1139, over 5676306.26 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3446, pruned_loss=0.09144, over 5745029.69 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3666, pruned_loss=0.1171, over 5657602.90 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:56:54,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.615e+02 1.394e+03 1.808e+03 2.536e+03 7.205e+03, threshold=3.616e+03, percent-clipped=9.0
+2023-03-10 19:57:10,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=922677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:57:17,798 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-10 19:57:24,124 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 19:57:33,017 INFO [train.py:968] (0/2) Epoch 21, batch 10350, giga_loss[loss=0.275, simple_loss=0.3516, pruned_loss=0.09923, over 28616.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3627, pruned_loss=0.1132, over 5673611.65 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3446, pruned_loss=0.0913, over 5747679.73 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3656, pruned_loss=0.1162, over 5655160.95 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:57:42,361 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 19:58:20,857 INFO [train.py:968] (0/2) Epoch 21, batch 10400, giga_loss[loss=0.2479, simple_loss=0.3235, pruned_loss=0.08618, over 28818.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3625, pruned_loss=0.1137, over 5679051.85 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3452, pruned_loss=0.09167, over 5752149.50 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3647, pruned_loss=0.1163, over 5658899.31 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:58:30,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3681, 1.5970, 1.4767, 1.5104], device='cuda:0'), covar=tensor([0.0673, 0.0405, 0.0311, 0.0719], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0070, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 19:58:33,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.564e+03 2.249e+03 3.140e+03 7.427e+03, threshold=4.497e+03, percent-clipped=18.0
+2023-03-10 19:59:05,409 INFO [train.py:968] (0/2) Epoch 21, batch 10450, giga_loss[loss=0.2626, simple_loss=0.3372, pruned_loss=0.09398, over 28863.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3596, pruned_loss=0.1125, over 5686341.33 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3455, pruned_loss=0.09183, over 5758407.76 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3618, pruned_loss=0.1154, over 5661359.76 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:59:24,644 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=922820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:59:26,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=922823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:59:53,100 INFO [train.py:968] (0/2) Epoch 21, batch 10500, giga_loss[loss=0.3239, simple_loss=0.395, pruned_loss=0.1264, over 28883.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3609, pruned_loss=0.1133, over 5691075.70 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.345, pruned_loss=0.09159, over 5762244.77 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3634, pruned_loss=0.1163, over 5666286.33 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:59:54,089 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=922852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:00:04,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.862e+03 2.346e+03 3.449e+03 8.270e+03, threshold=4.691e+03, percent-clipped=13.0
+2023-03-10 20:00:38,333 INFO [train.py:968] (0/2) Epoch 21, batch 10550, giga_loss[loss=0.3017, simple_loss=0.3722, pruned_loss=0.1156, over 28774.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.364, pruned_loss=0.1148, over 5688986.38 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3454, pruned_loss=0.09185, over 5764988.96 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.366, pruned_loss=0.1173, over 5665237.60 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:01:24,533 INFO [train.py:968] (0/2) Epoch 21, batch 10600, giga_loss[loss=0.2807, simple_loss=0.3515, pruned_loss=0.1049, over 28921.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3648, pruned_loss=0.1155, over 5662746.54 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3453, pruned_loss=0.09179, over 5767329.72 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3668, pruned_loss=0.118, over 5639943.06 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:01:36,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.691e+02 1.651e+03 2.191e+03 3.204e+03 6.890e+03, threshold=4.383e+03, percent-clipped=8.0
+2023-03-10 20:02:02,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=922987.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:02:07,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5767, 1.5559, 1.7856, 1.3740], device='cuda:0'), covar=tensor([0.1624, 0.2398, 0.1306, 0.1592], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0707, 0.0944, 0.0843], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 20:02:15,149 INFO [train.py:968] (0/2) Epoch 21, batch 10650, giga_loss[loss=0.3291, simple_loss=0.384, pruned_loss=0.1371, over 27929.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3663, pruned_loss=0.1176, over 5649348.45 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3454, pruned_loss=0.09179, over 5768446.38 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1198, over 5629429.60 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:02:57,692 INFO [train.py:968] (0/2) Epoch 21, batch 10700, giga_loss[loss=0.3725, simple_loss=0.3997, pruned_loss=0.1727, over 23429.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3682, pruned_loss=0.1189, over 5659547.39 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3457, pruned_loss=0.09176, over 5771784.04 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3698, pruned_loss=0.1214, over 5638157.53 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:03:12,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.837e+03 2.413e+03 3.507e+03 1.185e+04, threshold=4.826e+03, percent-clipped=12.0
+2023-03-10 20:03:26,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=923075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:03:37,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5890, 1.3111, 4.8687, 3.8002], device='cuda:0'), covar=tensor([0.1692, 0.2847, 0.0383, 0.0752], device='cuda:0'), in_proj_covar=tensor([0.0757, 0.0645, 0.0955, 0.0906], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 20:03:51,389 INFO [train.py:968] (0/2) Epoch 21, batch 10750, giga_loss[loss=0.3947, simple_loss=0.4263, pruned_loss=0.1816, over 26615.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3711, pruned_loss=0.1208, over 5657411.77 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.346, pruned_loss=0.09189, over 5771913.04 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3724, pruned_loss=0.123, over 5638694.50 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:04:17,783 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=923130.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:04:19,885 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=923133.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:04:35,073 INFO [train.py:968] (0/2) Epoch 21, batch 10800, giga_loss[loss=0.2761, simple_loss=0.3413, pruned_loss=0.1054, over 28946.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3707, pruned_loss=0.1202, over 5661981.99 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3459, pruned_loss=0.09177, over 5773731.43 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3726, pruned_loss=0.1228, over 5642116.50 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:04:44,379 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=923162.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:04:44,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.928e+02 1.650e+03 2.028e+03 2.668e+03 4.918e+03, threshold=4.056e+03, percent-clipped=1.0
+2023-03-10 20:05:21,327 INFO [train.py:968] (0/2) Epoch 21, batch 10850, libri_loss[loss=0.2529, simple_loss=0.3433, pruned_loss=0.08128, over 29190.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3713, pruned_loss=0.1206, over 5664497.97 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3458, pruned_loss=0.0916, over 5776484.07 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3734, pruned_loss=0.1236, over 5643604.61 frames. ], batch size: 97, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:06:07,911 INFO [train.py:968] (0/2) Epoch 21, batch 10900, giga_loss[loss=0.3384, simple_loss=0.3877, pruned_loss=0.1446, over 28559.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3741, pruned_loss=0.1232, over 5658158.30 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3456, pruned_loss=0.0915, over 5768024.13 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3767, pruned_loss=0.1265, over 5645807.96 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:06:21,413 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.692e+03 2.165e+03 2.899e+03 1.028e+04, threshold=4.330e+03, percent-clipped=11.0
+2023-03-10 20:06:59,850 INFO [train.py:968] (0/2) Epoch 21, batch 10950, giga_loss[loss=0.3097, simple_loss=0.3607, pruned_loss=0.1293, over 23831.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3736, pruned_loss=0.1212, over 5656894.94 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3458, pruned_loss=0.09161, over 5770618.59 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3761, pruned_loss=0.1243, over 5643129.38 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:07:48,347 INFO [train.py:968] (0/2) Epoch 21, batch 11000, giga_loss[loss=0.2754, simple_loss=0.3507, pruned_loss=0.1001, over 28815.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3726, pruned_loss=0.1208, over 5652296.16 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3458, pruned_loss=0.09158, over 5771382.61 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3754, pruned_loss=0.1242, over 5637034.21 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:07:59,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.580e+03 2.209e+03 3.192e+03 8.865e+03, threshold=4.418e+03, percent-clipped=11.0
+2023-03-10 20:08:30,998 INFO [train.py:968] (0/2) Epoch 21, batch 11050, giga_loss[loss=0.2827, simple_loss=0.3489, pruned_loss=0.1083, over 28844.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3693, pruned_loss=0.1188, over 5674128.07 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3453, pruned_loss=0.09143, over 5777173.61 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3732, pruned_loss=0.1231, over 5651415.65 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:09:27,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=923450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:09:27,997 INFO [train.py:968] (0/2) Epoch 21, batch 11100, giga_loss[loss=0.3003, simple_loss=0.3662, pruned_loss=0.1172, over 28665.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3702, pruned_loss=0.1204, over 5654125.77 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3454, pruned_loss=0.09156, over 5767745.10 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3736, pruned_loss=0.1241, over 5641985.42 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:09:35,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5822, 1.6996, 1.2222, 1.2370], device='cuda:0'), covar=tensor([0.1036, 0.0658, 0.1167, 0.1290], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0449, 0.0513, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 20:09:43,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.696e+03 2.472e+03 3.545e+03 1.127e+04, threshold=4.944e+03, percent-clipped=17.0
+2023-03-10 20:09:55,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4111, 3.2656, 1.5491, 1.5279], device='cuda:0'), covar=tensor([0.0977, 0.0342, 0.0879, 0.1330], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0558, 0.0385, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-10 20:10:14,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7982, 1.0404, 5.1381, 3.5630], device='cuda:0'), covar=tensor([0.1635, 0.3103, 0.0474, 0.1114], device='cuda:0'), in_proj_covar=tensor([0.0757, 0.0644, 0.0955, 0.0905], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 20:10:15,327 INFO [train.py:968] (0/2) Epoch 21, batch 11150, libri_loss[loss=0.248, simple_loss=0.3211, pruned_loss=0.08748, over 29327.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.369, pruned_loss=0.1196, over 5663900.14 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3452, pruned_loss=0.09147, over 5761858.11 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3724, pruned_loss=0.1233, over 5656630.45 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:10:31,756 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3334, 1.6262, 1.5036, 1.4458], device='cuda:0'), covar=tensor([0.0741, 0.0326, 0.0309, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 20:10:40,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-10 20:11:01,515 INFO [train.py:968] (0/2) Epoch 21, batch 11200, giga_loss[loss=0.3063, simple_loss=0.3724, pruned_loss=0.1201, over 28618.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3696, pruned_loss=0.121, over 5665282.60 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3454, pruned_loss=0.09156, over 5763026.22 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3723, pruned_loss=0.1241, over 5657719.70 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:11:16,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.149e+03 1.589e+03 1.983e+03 2.844e+03 6.551e+03, threshold=3.966e+03, percent-clipped=1.0
+2023-03-10 20:11:19,273 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=923569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:11:19,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5603, 1.7272, 1.4170, 1.6951], device='cuda:0'), covar=tensor([0.2584, 0.2761, 0.2963, 0.2443], device='cuda:0'), in_proj_covar=tensor([0.1498, 0.1084, 0.1320, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 20:11:41,525 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=923593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:11:44,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=923596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:11:50,266 INFO [train.py:968] (0/2) Epoch 21, batch 11250, giga_loss[loss=0.3042, simple_loss=0.3642, pruned_loss=0.1222, over 27988.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3698, pruned_loss=0.1213, over 5657399.62 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3457, pruned_loss=0.09162, over 5756167.93 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3722, pruned_loss=0.1243, over 5655018.98 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:12:15,282 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=923625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:12:37,677 INFO [train.py:968] (0/2) Epoch 21, batch 11300, giga_loss[loss=0.3233, simple_loss=0.381, pruned_loss=0.1328, over 28503.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.37, pruned_loss=0.1217, over 5662489.24 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3455, pruned_loss=0.09145, over 5759432.74 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3726, pruned_loss=0.1248, over 5655675.55 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:12:50,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.856e+02 1.680e+03 2.393e+03 3.274e+03 8.715e+03, threshold=4.787e+03, percent-clipped=12.0
+2023-03-10 20:13:25,841 INFO [train.py:968] (0/2) Epoch 21, batch 11350, giga_loss[loss=0.3925, simple_loss=0.4378, pruned_loss=0.1736, over 28688.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3711, pruned_loss=0.1225, over 5667757.81 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3457, pruned_loss=0.09157, over 5762238.33 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3735, pruned_loss=0.1256, over 5657827.00 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:13:26,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5237, 2.1975, 1.6256, 0.7798], device='cuda:0'), covar=tensor([0.5840, 0.2837, 0.4005, 0.6103], device='cuda:0'), in_proj_covar=tensor([0.1748, 0.1638, 0.1595, 0.1420], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 20:13:48,961 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-10 20:14:05,099 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-10 20:14:13,616 INFO [train.py:968] (0/2) Epoch 21, batch 11400, giga_loss[loss=0.3269, simple_loss=0.3898, pruned_loss=0.132, over 28622.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3713, pruned_loss=0.1223, over 5671734.34 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3458, pruned_loss=0.0916, over 5764567.11 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3735, pruned_loss=0.1251, over 5660806.65 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:14:25,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8521, 5.0575, 2.1622, 2.2032], device='cuda:0'), covar=tensor([0.0940, 0.0281, 0.0771, 0.1173], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0557, 0.0385, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-10 20:14:26,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.644e+03 2.447e+03 3.219e+03 8.351e+03, threshold=4.893e+03, percent-clipped=8.0
+2023-03-10 20:15:00,206 INFO [train.py:968] (0/2) Epoch 21, batch 11450, giga_loss[loss=0.3456, simple_loss=0.39, pruned_loss=0.1506, over 27563.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3722, pruned_loss=0.1236, over 5668716.33 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3461, pruned_loss=0.09178, over 5767943.90 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3744, pruned_loss=0.1266, over 5654358.43 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:15:50,786 INFO [train.py:968] (0/2) Epoch 21, batch 11500, libri_loss[loss=0.2211, simple_loss=0.3037, pruned_loss=0.06922, over 29459.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3722, pruned_loss=0.1245, over 5663692.21 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3459, pruned_loss=0.09166, over 5768563.18 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3743, pruned_loss=0.1272, over 5651314.85 frames. ], batch size: 70, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:16:03,295 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.122e+03 1.560e+03 1.982e+03 2.704e+03 5.773e+03, threshold=3.964e+03, percent-clipped=1.0
+2023-03-10 20:16:11,287 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5130, 2.2310, 1.5618, 0.7436], device='cuda:0'), covar=tensor([0.6119, 0.2943, 0.4003, 0.6506], device='cuda:0'), in_proj_covar=tensor([0.1744, 0.1634, 0.1586, 0.1414], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 20:16:35,780 INFO [train.py:968] (0/2) Epoch 21, batch 11550, giga_loss[loss=0.2934, simple_loss=0.3573, pruned_loss=0.1148, over 28680.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.372, pruned_loss=0.1232, over 5673803.75 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.346, pruned_loss=0.09161, over 5768519.61 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3744, pruned_loss=0.1264, over 5660442.47 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:17:15,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=923944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:17:21,843 INFO [train.py:968] (0/2) Epoch 21, batch 11600, giga_loss[loss=0.3217, simple_loss=0.3865, pruned_loss=0.1284, over 28865.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3726, pruned_loss=0.1233, over 5669724.32 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3463, pruned_loss=0.09173, over 5767849.85 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3748, pruned_loss=0.1265, over 5657450.46 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:17:34,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.602e+03 2.183e+03 3.254e+03 8.813e+03, threshold=4.366e+03, percent-clipped=15.0
+2023-03-10 20:18:12,101 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-924000.pt
+2023-03-10 20:18:13,378 INFO [train.py:968] (0/2) Epoch 21, batch 11650, giga_loss[loss=0.2925, simple_loss=0.3706, pruned_loss=0.1072, over 28909.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3722, pruned_loss=0.1225, over 5680729.02 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3461, pruned_loss=0.09163, over 5770567.60 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3747, pruned_loss=0.1257, over 5666947.73 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:18:13,622 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:18:42,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2119, 5.0511, 4.8008, 2.5930], device='cuda:0'), covar=tensor([0.0484, 0.0625, 0.0673, 0.1670], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.1138, 0.0967, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 20:18:50,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4215, 3.3432, 1.5119, 1.6055], device='cuda:0'), covar=tensor([0.0963, 0.0377, 0.0897, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0559, 0.0385, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-10 20:18:57,730 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 20:19:03,980 INFO [train.py:968] (0/2) Epoch 21, batch 11700, giga_loss[loss=0.2567, simple_loss=0.3362, pruned_loss=0.08861, over 28444.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3759, pruned_loss=0.1259, over 5674365.23 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3464, pruned_loss=0.09178, over 5769958.31 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3778, pruned_loss=0.1285, over 5663242.24 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:19:17,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.720e+02 1.515e+03 2.124e+03 2.648e+03 7.985e+03, threshold=4.248e+03, percent-clipped=8.0
+2023-03-10 20:19:24,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6853, 1.8414, 1.5838, 1.7168], device='cuda:0'), covar=tensor([0.2036, 0.1952, 0.1978, 0.1740], device='cuda:0'), in_proj_covar=tensor([0.1503, 0.1086, 0.1323, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 20:19:34,991 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:19:40,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=924090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:19:40,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5612, 1.8277, 1.4635, 1.7755], device='cuda:0'), covar=tensor([0.2666, 0.2716, 0.3199, 0.2252], device='cuda:0'), in_proj_covar=tensor([0.1501, 0.1085, 0.1322, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 20:19:45,400 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:19:48,878 INFO [train.py:968] (0/2) Epoch 21, batch 11750, giga_loss[loss=0.3344, simple_loss=0.3913, pruned_loss=0.1388, over 28711.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3744, pruned_loss=0.1244, over 5688592.80 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3461, pruned_loss=0.09161, over 5772994.05 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3768, pruned_loss=0.1274, over 5675006.18 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:20:07,380 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:20:08,912 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:20:35,457 INFO [train.py:968] (0/2) Epoch 21, batch 11800, giga_loss[loss=0.3118, simple_loss=0.3804, pruned_loss=0.1216, over 28648.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3756, pruned_loss=0.1243, over 5689478.52 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3461, pruned_loss=0.09152, over 5774962.64 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3781, pruned_loss=0.1274, over 5675340.64 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:20:49,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.763e+02 1.537e+03 1.846e+03 2.441e+03 8.224e+03, threshold=3.691e+03, percent-clipped=4.0
+2023-03-10 20:21:19,193 INFO [train.py:968] (0/2) Epoch 21, batch 11850, giga_loss[loss=0.3199, simple_loss=0.3566, pruned_loss=0.1416, over 23424.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3736, pruned_loss=0.1222, over 5682099.26 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3454, pruned_loss=0.09133, over 5774838.71 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3772, pruned_loss=0.1259, over 5667757.93 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:22:06,139 INFO [train.py:968] (0/2) Epoch 21, batch 11900, giga_loss[loss=0.3336, simple_loss=0.3903, pruned_loss=0.1385, over 28594.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3721, pruned_loss=0.1211, over 5682519.03 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3455, pruned_loss=0.09141, over 5777121.24 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3754, pruned_loss=0.1245, over 5667195.30 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:22:07,023 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:22:19,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.735e+02 1.591e+03 2.102e+03 2.711e+03 7.119e+03, threshold=4.203e+03, percent-clipped=12.0
+2023-03-10 20:22:44,661 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 20:22:50,142 INFO [train.py:968] (0/2) Epoch 21, batch 11950, libri_loss[loss=0.2609, simple_loss=0.3519, pruned_loss=0.0849, over 29683.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3712, pruned_loss=0.1206, over 5694534.10 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3453, pruned_loss=0.09129, over 5780539.17 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3746, pruned_loss=0.1242, over 5677294.13 frames. ], batch size: 88, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:23:20,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6834, 3.0075, 2.7165, 2.7447], device='cuda:0'), covar=tensor([0.1665, 0.1612, 0.1559, 0.1468], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0752, 0.0715, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 20:23:25,859 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.91 vs. limit=2.0
+2023-03-10 20:23:38,081 INFO [train.py:968] (0/2) Epoch 21, batch 12000, giga_loss[loss=0.3617, simple_loss=0.4034, pruned_loss=0.16, over 26600.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3724, pruned_loss=0.1217, over 5656348.66 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3456, pruned_loss=0.09145, over 5760584.54 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3759, pruned_loss=0.1256, over 5657574.11 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:23:38,086 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 20:23:46,876 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2074, simple_loss=0.3145, pruned_loss=0.05016, over 944034.00 frames. 
+2023-03-10 20:23:46,877 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 20:23:59,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.636e+03 2.073e+03 2.532e+03 7.830e+03, threshold=4.145e+03, percent-clipped=7.0
+2023-03-10 20:24:09,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:24:22,722 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924391.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:24:32,027 INFO [train.py:968] (0/2) Epoch 21, batch 12050, giga_loss[loss=0.4172, simple_loss=0.4349, pruned_loss=0.1997, over 26497.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3742, pruned_loss=0.1225, over 5667543.91 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.346, pruned_loss=0.09178, over 5763207.73 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3771, pruned_loss=0.1259, over 5664441.38 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:25:21,522 INFO [train.py:968] (0/2) Epoch 21, batch 12100, giga_loss[loss=0.302, simple_loss=0.3685, pruned_loss=0.1178, over 28911.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3753, pruned_loss=0.1248, over 5667753.30 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3459, pruned_loss=0.09174, over 5765389.65 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.378, pruned_loss=0.1279, over 5662130.33 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:25:35,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.668e+03 2.008e+03 2.724e+03 6.618e+03, threshold=4.016e+03, percent-clipped=6.0
+2023-03-10 20:25:39,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:25:58,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3783, 1.7099, 1.3533, 1.4972], device='cuda:0'), covar=tensor([0.0700, 0.0372, 0.0330, 0.0778], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 20:26:03,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:26:06,187 INFO [train.py:968] (0/2) Epoch 21, batch 12150, giga_loss[loss=0.2982, simple_loss=0.3692, pruned_loss=0.1136, over 28893.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3744, pruned_loss=0.1241, over 5662307.58 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3462, pruned_loss=0.09177, over 5761675.88 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3774, pruned_loss=0.1278, over 5658076.95 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:26:25,442 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:26:30,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=924522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:26:54,034 INFO [train.py:968] (0/2) Epoch 21, batch 12200, giga_loss[loss=0.3424, simple_loss=0.3984, pruned_loss=0.1432, over 28870.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3747, pruned_loss=0.1246, over 5667750.53 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3457, pruned_loss=0.09144, over 5763886.46 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3783, pruned_loss=0.1286, over 5660221.76 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:26:54,535 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:27:11,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.626e+03 1.972e+03 2.482e+03 4.838e+03, threshold=3.945e+03, percent-clipped=4.0
+2023-03-10 20:27:17,960 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:27:41,867 INFO [train.py:968] (0/2) Epoch 21, batch 12250, giga_loss[loss=0.3011, simple_loss=0.371, pruned_loss=0.1156, over 28733.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3759, pruned_loss=0.1254, over 5644327.21 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3457, pruned_loss=0.09146, over 5746116.12 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3791, pruned_loss=0.1291, over 5652800.20 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:27:57,269 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:27:59,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=924618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:28:02,476 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924621.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:28:07,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:28:18,983 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:28:22,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=924642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:28:25,040 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:28:28,895 INFO [train.py:968] (0/2) Epoch 21, batch 12300, giga_loss[loss=0.3156, simple_loss=0.379, pruned_loss=0.1261, over 28863.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3739, pruned_loss=0.124, over 5642793.35 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3459, pruned_loss=0.09158, over 5750589.71 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3771, pruned_loss=0.1277, over 5642754.75 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:28:42,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.598e+03 2.123e+03 2.709e+03 8.962e+03, threshold=4.246e+03, percent-clipped=9.0
+2023-03-10 20:28:46,838 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:29:13,984 INFO [train.py:968] (0/2) Epoch 21, batch 12350, giga_loss[loss=0.3185, simple_loss=0.386, pruned_loss=0.1255, over 28953.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3738, pruned_loss=0.1236, over 5649134.32 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3458, pruned_loss=0.09161, over 5756240.75 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3776, pruned_loss=0.1278, over 5640090.21 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:29:19,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9232, 1.3670, 1.4497, 1.1312], device='cuda:0'), covar=tensor([0.1906, 0.1091, 0.2103, 0.1559], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0754, 0.0719, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 20:29:51,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3077, 1.4692, 1.5186, 1.3455], device='cuda:0'), covar=tensor([0.1586, 0.1738, 0.1959, 0.1720], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0754, 0.0719, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 20:29:55,504 INFO [train.py:968] (0/2) Epoch 21, batch 12400, giga_loss[loss=0.2893, simple_loss=0.3634, pruned_loss=0.1076, over 28731.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3741, pruned_loss=0.1234, over 5654333.71 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3461, pruned_loss=0.09185, over 5761033.28 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3778, pruned_loss=0.1276, over 5639422.43 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:30:10,145 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924766.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:30:10,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.535e+03 1.924e+03 2.697e+03 1.170e+04, threshold=3.849e+03, percent-clipped=6.0
+2023-03-10 20:30:13,673 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:30:16,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=924773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:30:45,190 INFO [train.py:968] (0/2) Epoch 21, batch 12450, giga_loss[loss=0.255, simple_loss=0.3334, pruned_loss=0.08829, over 28461.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3732, pruned_loss=0.1227, over 5656004.19 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3458, pruned_loss=0.0916, over 5762220.06 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3768, pruned_loss=0.1267, over 5641802.03 frames. ], batch size: 65, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:30:46,105 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:30:54,152 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-10 20:31:14,614 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4087, 1.6542, 1.6664, 1.2292], device='cuda:0'), covar=tensor([0.1770, 0.2507, 0.1432, 0.1704], device='cuda:0'), in_proj_covar=tensor([0.0898, 0.0704, 0.0943, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 20:31:29,988 INFO [train.py:968] (0/2) Epoch 21, batch 12500, giga_loss[loss=0.2944, simple_loss=0.3595, pruned_loss=0.1147, over 28695.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.371, pruned_loss=0.1214, over 5663396.00 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3459, pruned_loss=0.09166, over 5764526.26 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3744, pruned_loss=0.1252, over 5647779.20 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:31:38,112 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924858.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:31:47,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.641e+02 1.729e+03 2.166e+03 2.927e+03 9.729e+03, threshold=4.331e+03, percent-clipped=15.0
+2023-03-10 20:32:16,213 INFO [train.py:968] (0/2) Epoch 21, batch 12550, giga_loss[loss=0.259, simple_loss=0.3289, pruned_loss=0.09459, over 28774.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3681, pruned_loss=0.1195, over 5678520.68 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3456, pruned_loss=0.09136, over 5768844.48 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3717, pruned_loss=0.1237, over 5659304.53 frames. ], batch size: 66, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:32:24,776 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924909.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:32:26,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4227, 1.8055, 1.7754, 1.5216], device='cuda:0'), covar=tensor([0.1892, 0.1591, 0.2064, 0.1821], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0755, 0.0719, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 20:32:26,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=924912.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:32:47,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2797, 1.6885, 1.3182, 1.0125], device='cuda:0'), covar=tensor([0.2678, 0.2785, 0.3135, 0.2486], device='cuda:0'), in_proj_covar=tensor([0.1499, 0.1085, 0.1320, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 20:32:50,140 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924941.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:32:50,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3725, 1.3401, 4.1510, 3.4076], device='cuda:0'), covar=tensor([0.1606, 0.2579, 0.0443, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0759, 0.0646, 0.0957, 0.0906], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 20:32:57,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:32:58,936 INFO [train.py:968] (0/2) Epoch 21, batch 12600, giga_loss[loss=0.2609, simple_loss=0.3376, pruned_loss=0.0921, over 28658.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3656, pruned_loss=0.1184, over 5673939.17 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3455, pruned_loss=0.09132, over 5773142.93 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3695, pruned_loss=0.123, over 5650856.52 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:33:16,748 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-10 20:33:16,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.788e+03 2.339e+03 3.735e+03 1.279e+04, threshold=4.678e+03, percent-clipped=18.0
+2023-03-10 20:33:45,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924996.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:33:48,180 INFO [train.py:968] (0/2) Epoch 21, batch 12650, giga_loss[loss=0.2803, simple_loss=0.3443, pruned_loss=0.1082, over 28853.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3644, pruned_loss=0.119, over 5667433.06 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3456, pruned_loss=0.09137, over 5774818.07 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3676, pruned_loss=0.1229, over 5646760.95 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:33:58,240 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:34:27,506 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.86 vs. limit=2.0
+2023-03-10 20:34:31,549 INFO [train.py:968] (0/2) Epoch 21, batch 12700, giga_loss[loss=0.292, simple_loss=0.3473, pruned_loss=0.1183, over 28820.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.363, pruned_loss=0.118, over 5666260.50 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3455, pruned_loss=0.09122, over 5779110.78 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3667, pruned_loss=0.1228, over 5640209.58 frames. ], batch size: 99, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:34:51,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.800e+03 2.392e+03 2.943e+03 6.984e+03, threshold=4.785e+03, percent-clipped=7.0
+2023-03-10 20:35:14,290 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:35:15,008 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:35:18,095 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:35:22,652 INFO [train.py:968] (0/2) Epoch 21, batch 12750, giga_loss[loss=0.2945, simple_loss=0.3622, pruned_loss=0.1134, over 28910.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3629, pruned_loss=0.1179, over 5665752.61 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3456, pruned_loss=0.09125, over 5780209.81 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3659, pruned_loss=0.1219, over 5643122.22 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:35:46,652 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:36:00,542 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925139.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:36:02,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925142.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:36:11,635 INFO [train.py:968] (0/2) Epoch 21, batch 12800, giga_loss[loss=0.2833, simple_loss=0.3598, pruned_loss=0.1034, over 28860.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3614, pruned_loss=0.115, over 5653035.01 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3453, pruned_loss=0.09122, over 5772000.55 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3642, pruned_loss=0.1185, over 5641397.30 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:36:27,811 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:36:28,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.700e+03 2.051e+03 2.772e+03 1.046e+04, threshold=4.102e+03, percent-clipped=5.0
+2023-03-10 20:36:32,852 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925171.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:36:59,675 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:37:03,331 INFO [train.py:968] (0/2) Epoch 21, batch 12850, giga_loss[loss=0.2543, simple_loss=0.3389, pruned_loss=0.08489, over 28911.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3588, pruned_loss=0.1113, over 5650874.97 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3449, pruned_loss=0.09105, over 5773709.48 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1147, over 5637900.92 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:37:32,463 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925233.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:37:40,389 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:37:50,454 INFO [train.py:968] (0/2) Epoch 21, batch 12900, libri_loss[loss=0.2085, simple_loss=0.2849, pruned_loss=0.06608, over 29656.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3549, pruned_loss=0.1071, over 5652174.43 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.344, pruned_loss=0.09068, over 5775503.36 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3586, pruned_loss=0.111, over 5634941.17 frames. ], batch size: 69, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:38:06,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.498e+03 1.802e+03 2.402e+03 5.396e+03, threshold=3.605e+03, percent-clipped=2.0
+2023-03-10 20:38:06,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4816, 4.2984, 4.0897, 2.0000], device='cuda:0'), covar=tensor([0.0676, 0.0877, 0.1013, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.1218, 0.1129, 0.0958, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-10 20:38:37,478 INFO [train.py:968] (0/2) Epoch 21, batch 12950, giga_loss[loss=0.255, simple_loss=0.3363, pruned_loss=0.08683, over 28389.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3522, pruned_loss=0.1047, over 5660981.75 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3437, pruned_loss=0.0907, over 5781188.03 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3561, pruned_loss=0.1086, over 5635592.20 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:39:27,205 INFO [train.py:968] (0/2) Epoch 21, batch 13000, giga_loss[loss=0.2378, simple_loss=0.3367, pruned_loss=0.06948, over 29044.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3508, pruned_loss=0.1018, over 5659761.80 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3438, pruned_loss=0.09072, over 5781821.09 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3538, pruned_loss=0.1049, over 5638716.99 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:39:40,739 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.47 vs. limit=5.0
+2023-03-10 20:39:44,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.217e+02 1.379e+03 1.757e+03 2.394e+03 5.428e+03, threshold=3.515e+03, percent-clipped=8.0
+2023-03-10 20:39:53,338 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925376.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:39:55,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925379.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:40:04,116 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:40:18,126 INFO [train.py:968] (0/2) Epoch 21, batch 13050, giga_loss[loss=0.2877, simple_loss=0.3593, pruned_loss=0.108, over 27832.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3506, pruned_loss=0.1004, over 5666685.01 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3435, pruned_loss=0.09064, over 5781269.43 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3534, pruned_loss=0.1031, over 5647922.47 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:40:22,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0798, 1.5685, 1.5578, 1.3459], device='cuda:0'), covar=tensor([0.1870, 0.1340, 0.1869, 0.1597], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0746, 0.0711, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 20:40:27,446 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925408.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:40:29,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:41:07,599 INFO [train.py:968] (0/2) Epoch 21, batch 13100, giga_loss[loss=0.2546, simple_loss=0.334, pruned_loss=0.08759, over 28586.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3509, pruned_loss=0.1004, over 5657692.10 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3433, pruned_loss=0.09068, over 5779854.65 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3534, pruned_loss=0.1028, over 5641803.41 frames. ], batch size: 85, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:41:24,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:41:27,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.508e+02 1.428e+03 1.834e+03 2.718e+03 7.963e+03, threshold=3.668e+03, percent-clipped=14.0
+2023-03-10 20:41:57,370 INFO [train.py:968] (0/2) Epoch 21, batch 13150, giga_loss[loss=0.2688, simple_loss=0.3484, pruned_loss=0.09459, over 28703.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3489, pruned_loss=0.09906, over 5656578.98 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3437, pruned_loss=0.09111, over 5781920.68 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3506, pruned_loss=0.1007, over 5639739.28 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:42:29,340 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:42:32,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:42:38,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:42:46,035 INFO [train.py:968] (0/2) Epoch 21, batch 13200, giga_loss[loss=0.2422, simple_loss=0.3281, pruned_loss=0.07811, over 28972.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3457, pruned_loss=0.09729, over 5655074.99 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3434, pruned_loss=0.09105, over 5784683.38 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3474, pruned_loss=0.0988, over 5636176.95 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:42:59,969 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:43:03,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.201e+02 1.337e+03 1.733e+03 2.306e+03 9.066e+03, threshold=3.466e+03, percent-clipped=6.0
+2023-03-10 20:43:05,813 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:43:37,042 INFO [train.py:968] (0/2) Epoch 21, batch 13250, giga_loss[loss=0.2777, simple_loss=0.3531, pruned_loss=0.1011, over 28527.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3463, pruned_loss=0.09762, over 5648831.09 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3432, pruned_loss=0.09093, over 5786246.37 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.348, pruned_loss=0.09904, over 5630675.88 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:43:46,214 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:43:47,957 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:43:51,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:43:53,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1469, 1.5019, 1.4117, 1.0553], device='cuda:0'), covar=tensor([0.1605, 0.2470, 0.1374, 0.1683], device='cuda:0'), in_proj_covar=tensor([0.0896, 0.0698, 0.0940, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 20:44:16,311 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:44:25,578 INFO [train.py:968] (0/2) Epoch 21, batch 13300, giga_loss[loss=0.2731, simple_loss=0.3472, pruned_loss=0.09951, over 27738.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3452, pruned_loss=0.09692, over 5656069.67 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3427, pruned_loss=0.09079, over 5788191.69 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.347, pruned_loss=0.0983, over 5637068.28 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:44:44,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.089e+02 1.469e+03 2.043e+03 2.646e+03 7.140e+03, threshold=4.085e+03, percent-clipped=10.0
+2023-03-10 20:45:02,726 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:45:04,498 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:45:07,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:45:17,477 INFO [train.py:968] (0/2) Epoch 21, batch 13350, giga_loss[loss=0.2374, simple_loss=0.3208, pruned_loss=0.07701, over 28955.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3423, pruned_loss=0.0944, over 5656617.17 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3429, pruned_loss=0.09099, over 5789755.59 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3435, pruned_loss=0.09538, over 5637665.17 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:45:20,002 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7574, 1.9677, 1.4235, 1.5537], device='cuda:0'), covar=tensor([0.0945, 0.0539, 0.0943, 0.1091], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0445, 0.0510, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 20:45:20,787 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-10 20:45:29,857 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:45:32,684 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:45:32,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:03,708 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:08,032 INFO [train.py:968] (0/2) Epoch 21, batch 13400, giga_loss[loss=0.2261, simple_loss=0.3114, pruned_loss=0.07042, over 28771.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3381, pruned_loss=0.09152, over 5657274.44 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3424, pruned_loss=0.09084, over 5792107.52 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3395, pruned_loss=0.09248, over 5637178.53 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:46:15,989 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:17,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:24,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.182e+02 1.270e+03 1.573e+03 1.918e+03 3.602e+03, threshold=3.146e+03, percent-clipped=0.0
+2023-03-10 20:46:41,002 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:44,287 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:57,564 INFO [train.py:968] (0/2) Epoch 21, batch 13450, giga_loss[loss=0.2223, simple_loss=0.3047, pruned_loss=0.07, over 28596.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3349, pruned_loss=0.08989, over 5666859.57 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3419, pruned_loss=0.0906, over 5795252.53 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3363, pruned_loss=0.09087, over 5642540.24 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:47:11,009 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-10 20:47:20,013 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:47:51,800 INFO [train.py:968] (0/2) Epoch 21, batch 13500, giga_loss[loss=0.2378, simple_loss=0.3024, pruned_loss=0.08658, over 23902.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3345, pruned_loss=0.0903, over 5662497.98 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.342, pruned_loss=0.09066, over 5795484.27 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3354, pruned_loss=0.09102, over 5642629.94 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:48:09,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.867e+02 1.475e+03 1.814e+03 2.583e+03 8.003e+03, threshold=3.628e+03, percent-clipped=17.0
+2023-03-10 20:48:49,699 INFO [train.py:968] (0/2) Epoch 21, batch 13550, giga_loss[loss=0.2573, simple_loss=0.3445, pruned_loss=0.08507, over 28278.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3344, pruned_loss=0.0902, over 5656493.44 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3418, pruned_loss=0.09069, over 5796553.13 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3352, pruned_loss=0.09074, over 5638268.63 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:49:20,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:49:24,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:49:40,953 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4032, 1.6185, 1.2187, 1.2036], device='cuda:0'), covar=tensor([0.0998, 0.0496, 0.1024, 0.1056], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0445, 0.0510, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 20:49:46,614 INFO [train.py:968] (0/2) Epoch 21, batch 13600, libri_loss[loss=0.2986, simple_loss=0.3634, pruned_loss=0.1169, over 29650.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3378, pruned_loss=0.09123, over 5649382.31 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.342, pruned_loss=0.09092, over 5794270.42 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3381, pruned_loss=0.09143, over 5635583.00 frames. ], batch size: 88, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:49:57,058 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:50:06,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.875e+02 1.335e+03 1.747e+03 2.262e+03 5.545e+03, threshold=3.495e+03, percent-clipped=7.0
+2023-03-10 20:50:25,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5507, 1.7151, 1.7364, 1.5393], device='cuda:0'), covar=tensor([0.2533, 0.2018, 0.1657, 0.1959], device='cuda:0'), in_proj_covar=tensor([0.1923, 0.1847, 0.1769, 0.1918], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 20:50:39,254 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-926000.pt
+2023-03-10 20:50:41,422 INFO [train.py:968] (0/2) Epoch 21, batch 13650, giga_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08846, over 28710.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3381, pruned_loss=0.09076, over 5668571.56 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3414, pruned_loss=0.09079, over 5796376.19 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3388, pruned_loss=0.09103, over 5650781.35 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:51:05,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.29 vs. limit=5.0
+2023-03-10 20:51:08,837 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-10 20:51:40,453 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-10 20:51:41,340 INFO [train.py:968] (0/2) Epoch 21, batch 13700, giga_loss[loss=0.2682, simple_loss=0.3415, pruned_loss=0.09745, over 27765.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3383, pruned_loss=0.09112, over 5664202.67 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.341, pruned_loss=0.0906, over 5789139.75 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3392, pruned_loss=0.09152, over 5654593.66 frames. ], batch size: 474, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:51:53,245 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=926061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:52:01,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.464e+03 1.769e+03 2.319e+03 4.971e+03, threshold=3.538e+03, percent-clipped=5.0
+2023-03-10 20:52:23,515 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=926091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:52:34,823 INFO [train.py:968] (0/2) Epoch 21, batch 13750, giga_loss[loss=0.2235, simple_loss=0.3161, pruned_loss=0.06543, over 28427.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3348, pruned_loss=0.08903, over 5674995.33 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3396, pruned_loss=0.09009, over 5793319.46 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3367, pruned_loss=0.08983, over 5658354.90 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:52:46,524 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=926110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:53:33,519 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-10 20:53:35,185 INFO [train.py:968] (0/2) Epoch 21, batch 13800, giga_loss[loss=0.2343, simple_loss=0.3203, pruned_loss=0.07417, over 28458.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3345, pruned_loss=0.08718, over 5673298.50 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3397, pruned_loss=0.0902, over 5794094.87 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3357, pruned_loss=0.08766, over 5658934.17 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:53:57,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.723e+02 1.317e+03 1.692e+03 2.575e+03 5.912e+03, threshold=3.385e+03, percent-clipped=8.0
+2023-03-10 20:54:29,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=926195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:54:37,096 INFO [train.py:968] (0/2) Epoch 21, batch 13850, giga_loss[loss=0.3091, simple_loss=0.3529, pruned_loss=0.1327, over 26684.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3329, pruned_loss=0.08711, over 5667333.87 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3392, pruned_loss=0.09005, over 5797108.35 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3342, pruned_loss=0.08754, over 5650108.94 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:54:41,135 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=926204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:54:46,296 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=926207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:55:19,873 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=926236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:55:35,636 INFO [train.py:968] (0/2) Epoch 21, batch 13900, giga_loss[loss=0.283, simple_loss=0.3541, pruned_loss=0.106, over 28023.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3313, pruned_loss=0.08714, over 5660667.87 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.339, pruned_loss=0.09009, over 5779930.56 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3324, pruned_loss=0.0874, over 5660142.39 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:56:01,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.782e+02 1.469e+03 1.799e+03 2.464e+03 1.174e+04, threshold=3.599e+03, percent-clipped=15.0
+2023-03-10 20:56:38,047 INFO [train.py:968] (0/2) Epoch 21, batch 13950, giga_loss[loss=0.2235, simple_loss=0.2933, pruned_loss=0.07689, over 24458.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3296, pruned_loss=0.0865, over 5656705.57 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.339, pruned_loss=0.09009, over 5779930.56 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3305, pruned_loss=0.0867, over 5656296.59 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:57:23,134 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=926338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:57:25,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=926341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:57:36,547 INFO [train.py:968] (0/2) Epoch 21, batch 14000, giga_loss[loss=0.2543, simple_loss=0.3327, pruned_loss=0.08796, over 27461.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3315, pruned_loss=0.087, over 5651606.68 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3385, pruned_loss=0.08989, over 5778878.23 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3325, pruned_loss=0.08731, over 5649864.55 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:58:00,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=926370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:58:02,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.417e+02 1.504e+03 1.964e+03 2.781e+03 1.102e+04, threshold=3.927e+03, percent-clipped=10.0
+2023-03-10 20:58:37,047 INFO [train.py:968] (0/2) Epoch 21, batch 14050, giga_loss[loss=0.2337, simple_loss=0.321, pruned_loss=0.07315, over 28427.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3321, pruned_loss=0.08656, over 5658551.56 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3384, pruned_loss=0.09006, over 5781864.02 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3328, pruned_loss=0.08659, over 5651873.42 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:58:45,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1933, 2.5786, 1.2034, 1.3977], device='cuda:0'), covar=tensor([0.1001, 0.0322, 0.0948, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0549, 0.0382, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 20:59:44,461 INFO [train.py:968] (0/2) Epoch 21, batch 14100, giga_loss[loss=0.3103, simple_loss=0.3783, pruned_loss=0.1212, over 27744.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3293, pruned_loss=0.08485, over 5668023.84 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3383, pruned_loss=0.09004, over 5783133.21 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.33, pruned_loss=0.08486, over 5660762.75 frames. ], batch size: 474, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:59:44,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=926451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:00:05,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=926466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:00:06,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3909, 3.8283, 1.6385, 1.6245], device='cuda:0'), covar=tensor([0.1025, 0.0269, 0.0955, 0.1355], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0549, 0.0382, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 21:00:09,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.964e+02 1.408e+03 1.722e+03 2.343e+03 6.527e+03, threshold=3.444e+03, percent-clipped=2.0
+2023-03-10 21:00:10,169 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-10 21:00:24,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=926485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:00:40,343 INFO [train.py:968] (0/2) Epoch 21, batch 14150, giga_loss[loss=0.2779, simple_loss=0.3526, pruned_loss=0.1016, over 28936.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3318, pruned_loss=0.08665, over 5671130.10 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3383, pruned_loss=0.09005, over 5774468.41 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.332, pruned_loss=0.08647, over 5668621.34 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 21:01:40,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8921, 2.0470, 1.4298, 1.7002], device='cuda:0'), covar=tensor([0.0903, 0.0572, 0.1013, 0.1014], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0442, 0.0508, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 21:01:45,167 INFO [train.py:968] (0/2) Epoch 21, batch 14200, giga_loss[loss=0.2927, simple_loss=0.3653, pruned_loss=0.1101, over 26840.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3347, pruned_loss=0.08692, over 5660694.73 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3375, pruned_loss=0.08968, over 5765647.63 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3355, pruned_loss=0.08701, over 5664227.20 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:01:53,905 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-10 21:02:05,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-10 21:02:11,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.479e+02 1.497e+03 1.864e+03 2.492e+03 6.802e+03, threshold=3.727e+03, percent-clipped=11.0
+2023-03-10 21:02:49,803 INFO [train.py:968] (0/2) Epoch 21, batch 14250, giga_loss[loss=0.274, simple_loss=0.3595, pruned_loss=0.09421, over 28635.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3385, pruned_loss=0.08658, over 5665468.23 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3374, pruned_loss=0.08965, over 5767584.79 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3391, pruned_loss=0.08664, over 5665406.61 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:02:57,034 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=926609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:03:00,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=926612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:03:22,775 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=926628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:03:25,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=926631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:03:35,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=926641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:03:47,759 INFO [train.py:968] (0/2) Epoch 21, batch 14300, giga_loss[loss=0.2489, simple_loss=0.3389, pruned_loss=0.07949, over 28489.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3401, pruned_loss=0.08637, over 5668419.71 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3375, pruned_loss=0.08965, over 5770332.41 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3406, pruned_loss=0.08635, over 5664015.64 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:03:58,446 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=926660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:04:13,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.717e+02 1.359e+03 1.688e+03 2.289e+03 4.921e+03, threshold=3.376e+03, percent-clipped=4.0
+2023-03-10 21:04:49,631 INFO [train.py:968] (0/2) Epoch 21, batch 14350, giga_loss[loss=0.2361, simple_loss=0.3079, pruned_loss=0.08218, over 24713.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3393, pruned_loss=0.08569, over 5666509.38 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3375, pruned_loss=0.08966, over 5771138.75 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3396, pruned_loss=0.08565, over 5661878.27 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:05:04,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1370, 1.2886, 1.1784, 0.9113], device='cuda:0'), covar=tensor([0.1062, 0.0500, 0.1050, 0.1026], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0442, 0.0509, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 21:05:51,684 INFO [train.py:968] (0/2) Epoch 21, batch 14400, giga_loss[loss=0.2461, simple_loss=0.3268, pruned_loss=0.08263, over 28879.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.34, pruned_loss=0.08717, over 5671323.93 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3373, pruned_loss=0.08956, over 5773068.59 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3405, pruned_loss=0.08715, over 5663435.90 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:06:16,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.860e+02 1.311e+03 1.756e+03 2.110e+03 6.645e+03, threshold=3.513e+03, percent-clipped=5.0
+2023-03-10 21:06:31,121 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=926782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:06:55,765 INFO [train.py:968] (0/2) Epoch 21, batch 14450, giga_loss[loss=0.2865, simple_loss=0.3623, pruned_loss=0.1054, over 28885.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3395, pruned_loss=0.08805, over 5687043.66 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3371, pruned_loss=0.08952, over 5776132.00 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3401, pruned_loss=0.08804, over 5675747.81 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:07:24,634 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=926821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:07:33,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=926826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:08:07,582 INFO [train.py:968] (0/2) Epoch 21, batch 14500, giga_loss[loss=0.2595, simple_loss=0.3482, pruned_loss=0.08536, over 28701.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3397, pruned_loss=0.0891, over 5688829.60 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3376, pruned_loss=0.08993, over 5777445.74 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3398, pruned_loss=0.08869, over 5677170.02 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:08:47,383 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.849e+02 1.398e+03 1.892e+03 2.612e+03 6.495e+03, threshold=3.783e+03, percent-clipped=5.0
+2023-03-10 21:09:30,150 INFO [train.py:968] (0/2) Epoch 21, batch 14550, giga_loss[loss=0.2285, simple_loss=0.3156, pruned_loss=0.07072, over 28179.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3354, pruned_loss=0.08651, over 5681810.49 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3377, pruned_loss=0.08995, over 5774351.52 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3354, pruned_loss=0.08616, over 5674909.68 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:10:29,616 INFO [train.py:968] (0/2) Epoch 21, batch 14600, giga_loss[loss=0.2519, simple_loss=0.3116, pruned_loss=0.09606, over 24134.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3334, pruned_loss=0.08561, over 5682689.19 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3376, pruned_loss=0.09006, over 5774850.87 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3333, pruned_loss=0.08506, over 5672703.99 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:10:54,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=926969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:10:57,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=926972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:10:57,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.162e+02 1.475e+03 1.954e+03 2.757e+03 6.416e+03, threshold=3.908e+03, percent-clipped=7.0
+2023-03-10 21:11:09,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-10 21:11:32,991 INFO [train.py:968] (0/2) Epoch 21, batch 14650, giga_loss[loss=0.2658, simple_loss=0.3488, pruned_loss=0.09141, over 28115.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3329, pruned_loss=0.08612, over 5670372.71 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3376, pruned_loss=0.09016, over 5766434.79 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3327, pruned_loss=0.08546, over 5668627.35 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:11:33,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:11:51,012 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=927017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:12:32,759 INFO [train.py:968] (0/2) Epoch 21, batch 14700, giga_loss[loss=0.2611, simple_loss=0.3423, pruned_loss=0.08993, over 28798.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3367, pruned_loss=0.08798, over 5667777.01 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3373, pruned_loss=0.08996, over 5760531.46 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3368, pruned_loss=0.08755, over 5669518.38 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:12:45,353 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 21:12:53,368 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 21:12:58,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.622e+03 2.056e+03 2.686e+03 6.271e+03, threshold=4.112e+03, percent-clipped=13.0
+2023-03-10 21:13:35,604 INFO [train.py:968] (0/2) Epoch 21, batch 14750, giga_loss[loss=0.2458, simple_loss=0.3266, pruned_loss=0.08247, over 28049.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.335, pruned_loss=0.08783, over 5673273.94 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3372, pruned_loss=0.08994, over 5761289.80 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3351, pruned_loss=0.08751, over 5673621.24 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:14:40,325 INFO [train.py:968] (0/2) Epoch 21, batch 14800, giga_loss[loss=0.2221, simple_loss=0.3094, pruned_loss=0.06743, over 28736.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3346, pruned_loss=0.08878, over 5682337.36 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3364, pruned_loss=0.08957, over 5764160.40 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3354, pruned_loss=0.08881, over 5678572.45 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:14:48,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:15:05,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.598e+02 1.536e+03 2.119e+03 3.011e+03 8.601e+03, threshold=4.239e+03, percent-clipped=8.0
+2023-03-10 21:15:37,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:15:42,617 INFO [train.py:968] (0/2) Epoch 21, batch 14850, giga_loss[loss=0.2773, simple_loss=0.36, pruned_loss=0.09726, over 28593.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3355, pruned_loss=0.08948, over 5681792.93 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3363, pruned_loss=0.08956, over 5765581.16 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3361, pruned_loss=0.08951, over 5676988.91 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:16:42,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4723, 4.6404, 1.7711, 1.7157], device='cuda:0'), covar=tensor([0.0999, 0.0333, 0.0906, 0.1325], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0550, 0.0383, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 21:16:45,879 INFO [train.py:968] (0/2) Epoch 21, batch 14900, giga_loss[loss=0.255, simple_loss=0.3406, pruned_loss=0.08467, over 28968.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3377, pruned_loss=0.08971, over 5673430.97 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3366, pruned_loss=0.08991, over 5757250.23 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3379, pruned_loss=0.0894, over 5676039.04 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:17:01,338 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=927261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:17:22,025 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.189e+02 1.455e+03 1.795e+03 2.158e+03 4.490e+03, threshold=3.591e+03, percent-clipped=1.0
+2023-03-10 21:18:00,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:18:01,102 INFO [train.py:968] (0/2) Epoch 21, batch 14950, libri_loss[loss=0.2232, simple_loss=0.2907, pruned_loss=0.07786, over 28545.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3378, pruned_loss=0.08956, over 5662137.05 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3368, pruned_loss=0.09023, over 5749490.86 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3378, pruned_loss=0.08902, over 5669067.43 frames. ], batch size: 63, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:18:04,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:18:21,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8745, 1.0757, 0.8600, 0.2577], device='cuda:0'), covar=tensor([0.3982, 0.2702, 0.3870, 0.6355], device='cuda:0'), in_proj_covar=tensor([0.1735, 0.1627, 0.1591, 0.1419], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 21:18:26,114 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=927320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:18:43,828 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:18:49,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:18:57,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:19:06,812 INFO [train.py:968] (0/2) Epoch 21, batch 15000, libri_loss[loss=0.1918, simple_loss=0.2763, pruned_loss=0.05364, over 28073.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3355, pruned_loss=0.08907, over 5652011.82 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3363, pruned_loss=0.09004, over 5741553.05 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.336, pruned_loss=0.08879, over 5660845.98 frames. ], batch size: 62, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:19:06,816 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 21:19:14,676 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3039, 1.5089, 1.3967, 1.2598], device='cuda:0'), covar=tensor([0.2498, 0.2164, 0.1425, 0.1771], device='cuda:0'), in_proj_covar=tensor([0.1908, 0.1828, 0.1749, 0.1900], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 21:19:15,256 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.1967, simple_loss=0.2975, pruned_loss=0.04796, over 944034.00 frames. 
+2023-03-10 21:19:15,257 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 21:19:41,975 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:19:44,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.339e+02 1.504e+03 1.916e+03 2.624e+03 7.052e+03, threshold=3.831e+03, percent-clipped=10.0
+2023-03-10 21:20:11,823 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:20:26,646 INFO [train.py:968] (0/2) Epoch 21, batch 15050, giga_loss[loss=0.2104, simple_loss=0.2919, pruned_loss=0.06446, over 29108.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3301, pruned_loss=0.08732, over 5653665.46 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3363, pruned_loss=0.08999, over 5742122.29 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3306, pruned_loss=0.08714, over 5659780.28 frames. ], batch size: 285, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:20:34,850 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=927409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:21:25,419 INFO [train.py:968] (0/2) Epoch 21, batch 15100, giga_loss[loss=0.2797, simple_loss=0.3578, pruned_loss=0.1008, over 28821.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3269, pruned_loss=0.08561, over 5660962.03 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3357, pruned_loss=0.08959, over 5746600.73 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3276, pruned_loss=0.08571, over 5659713.14 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 21:21:53,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.571e+02 1.474e+03 1.968e+03 2.827e+03 1.249e+04, threshold=3.936e+03, percent-clipped=14.0
+2023-03-10 21:22:22,551 INFO [train.py:968] (0/2) Epoch 21, batch 15150, giga_loss[loss=0.3108, simple_loss=0.3781, pruned_loss=0.1217, over 28879.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3284, pruned_loss=0.08679, over 5659336.35 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3356, pruned_loss=0.08952, over 5748496.07 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3289, pruned_loss=0.08687, over 5654347.18 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 21:22:58,256 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:23:00,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1299, 4.9787, 4.6771, 2.1974], device='cuda:0'), covar=tensor([0.0469, 0.0640, 0.0804, 0.2086], device='cuda:0'), in_proj_covar=tensor([0.1190, 0.1099, 0.0934, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 21:23:01,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:23:14,665 INFO [train.py:968] (0/2) Epoch 21, batch 15200, giga_loss[loss=0.229, simple_loss=0.312, pruned_loss=0.07298, over 28675.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3284, pruned_loss=0.08646, over 5669526.83 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3351, pruned_loss=0.08924, over 5751072.59 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3291, pruned_loss=0.08673, over 5660693.15 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:23:38,928 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:23:47,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.484e+02 1.391e+03 1.736e+03 2.380e+03 6.132e+03, threshold=3.473e+03, percent-clipped=4.0
+2023-03-10 21:24:17,283 INFO [train.py:968] (0/2) Epoch 21, batch 15250, giga_loss[loss=0.2332, simple_loss=0.3181, pruned_loss=0.07413, over 27556.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3262, pruned_loss=0.08481, over 5652065.66 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3349, pruned_loss=0.08928, over 5744868.58 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3267, pruned_loss=0.08487, over 5648569.04 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:24:59,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:25:06,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5471, 1.8011, 1.4277, 1.8127], device='cuda:0'), covar=tensor([0.2689, 0.2659, 0.2993, 0.2244], device='cuda:0'), in_proj_covar=tensor([0.1496, 0.1077, 0.1323, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 21:25:21,776 INFO [train.py:968] (0/2) Epoch 21, batch 15300, giga_loss[loss=0.2606, simple_loss=0.3374, pruned_loss=0.09184, over 28145.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3257, pruned_loss=0.08433, over 5664570.87 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3348, pruned_loss=0.08934, over 5748996.22 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3261, pruned_loss=0.08424, over 5656235.38 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:25:53,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.430e+02 1.327e+03 1.654e+03 2.466e+03 6.353e+03, threshold=3.307e+03, percent-clipped=6.0
+2023-03-10 21:26:24,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:26:31,176 INFO [train.py:968] (0/2) Epoch 21, batch 15350, giga_loss[loss=0.2919, simple_loss=0.3569, pruned_loss=0.1135, over 26802.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3252, pruned_loss=0.08412, over 5651349.87 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3349, pruned_loss=0.08941, over 5741087.90 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3252, pruned_loss=0.08393, over 5650543.75 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:26:47,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2961, 1.5319, 1.5366, 1.1555], device='cuda:0'), covar=tensor([0.1665, 0.2581, 0.1472, 0.1780], device='cuda:0'), in_proj_covar=tensor([0.0894, 0.0692, 0.0940, 0.0839], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 21:27:34,396 INFO [train.py:968] (0/2) Epoch 21, batch 15400, giga_loss[loss=0.2747, simple_loss=0.3522, pruned_loss=0.09862, over 27998.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3251, pruned_loss=0.08336, over 5653615.53 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3348, pruned_loss=0.08925, over 5743716.44 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.325, pruned_loss=0.08325, over 5648952.72 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:27:43,980 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=927758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:28:06,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.699e+02 1.266e+03 1.598e+03 2.190e+03 7.026e+03, threshold=3.196e+03, percent-clipped=7.0
+2023-03-10 21:28:13,436 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:28:15,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:28:17,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:28:40,649 INFO [train.py:968] (0/2) Epoch 21, batch 15450, giga_loss[loss=0.279, simple_loss=0.3446, pruned_loss=0.1067, over 28918.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3256, pruned_loss=0.08411, over 5652767.45 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3345, pruned_loss=0.08918, over 5737286.96 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3256, pruned_loss=0.08398, over 5651971.44 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:28:43,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7923, 5.6181, 5.3075, 2.6747], device='cuda:0'), covar=tensor([0.0415, 0.0551, 0.0746, 0.1720], device='cuda:0'), in_proj_covar=tensor([0.1188, 0.1096, 0.0931, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 21:28:51,738 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:29:20,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0797, 3.1689, 2.1110, 1.0909], device='cuda:0'), covar=tensor([0.7528, 0.3000, 0.3880, 0.6588], device='cuda:0'), in_proj_covar=tensor([0.1727, 0.1625, 0.1586, 0.1414], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 21:29:25,006 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:29:28,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:29:41,693 INFO [train.py:968] (0/2) Epoch 21, batch 15500, giga_loss[loss=0.2547, simple_loss=0.3403, pruned_loss=0.08455, over 29020.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3244, pruned_loss=0.08379, over 5654411.15 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3337, pruned_loss=0.08879, over 5740116.42 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.325, pruned_loss=0.0839, over 5649410.22 frames. ], batch size: 285, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:30:05,121 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:30:09,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.877e+02 1.271e+03 1.660e+03 2.181e+03 5.945e+03, threshold=3.320e+03, percent-clipped=8.0
+2023-03-10 21:30:40,050 INFO [train.py:968] (0/2) Epoch 21, batch 15550, giga_loss[loss=0.225, simple_loss=0.3198, pruned_loss=0.06509, over 28648.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.326, pruned_loss=0.0832, over 5654753.40 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3339, pruned_loss=0.08899, over 5729816.72 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3261, pruned_loss=0.08302, over 5659316.55 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:31:12,404 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:31:16,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:31:41,636 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=927949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:31:43,090 INFO [train.py:968] (0/2) Epoch 21, batch 15600, giga_loss[loss=0.2384, simple_loss=0.3262, pruned_loss=0.07533, over 29001.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3297, pruned_loss=0.08441, over 5655889.47 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3339, pruned_loss=0.08902, over 5730692.64 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3298, pruned_loss=0.08421, over 5658462.11 frames. ], batch size: 128, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:31:55,162 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:32:12,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.409e+02 1.422e+03 1.898e+03 2.580e+03 5.192e+03, threshold=3.796e+03, percent-clipped=9.0
+2023-03-10 21:32:22,023 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 21:32:33,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2132, 1.5064, 1.4498, 1.0561], device='cuda:0'), covar=tensor([0.1789, 0.2850, 0.1613, 0.1958], device='cuda:0'), in_proj_covar=tensor([0.0897, 0.0693, 0.0941, 0.0841], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 21:32:41,656 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-928000.pt
+2023-03-10 21:32:43,751 INFO [train.py:968] (0/2) Epoch 21, batch 15650, giga_loss[loss=0.3105, simple_loss=0.3708, pruned_loss=0.1251, over 26781.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3315, pruned_loss=0.08539, over 5655316.45 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3338, pruned_loss=0.089, over 5733422.01 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3315, pruned_loss=0.08513, over 5652923.44 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:33:34,035 INFO [train.py:968] (0/2) Epoch 21, batch 15700, giga_loss[loss=0.236, simple_loss=0.3233, pruned_loss=0.0744, over 28970.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3313, pruned_loss=0.08508, over 5670414.03 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.333, pruned_loss=0.08862, over 5738195.91 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3319, pruned_loss=0.08502, over 5660770.76 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:33:35,308 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6219, 1.7156, 1.8845, 1.4303], device='cuda:0'), covar=tensor([0.1916, 0.2568, 0.1521, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.0896, 0.0692, 0.0941, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 21:34:04,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.154e+02 1.531e+03 1.990e+03 2.965e+03 1.207e+04, threshold=3.980e+03, percent-clipped=14.0
+2023-03-10 21:34:31,688 INFO [train.py:968] (0/2) Epoch 21, batch 15750, giga_loss[loss=0.2398, simple_loss=0.3195, pruned_loss=0.08009, over 27563.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3305, pruned_loss=0.08468, over 5682484.10 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3325, pruned_loss=0.08839, over 5741742.66 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3315, pruned_loss=0.08477, over 5670591.22 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:35:13,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=928133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:35:20,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2756, 1.3940, 1.3380, 1.3084], device='cuda:0'), covar=tensor([0.2093, 0.1857, 0.1427, 0.1654], device='cuda:0'), in_proj_covar=tensor([0.1913, 0.1834, 0.1759, 0.1902], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 21:35:33,216 INFO [train.py:968] (0/2) Epoch 21, batch 15800, libri_loss[loss=0.2554, simple_loss=0.3332, pruned_loss=0.08882, over 29521.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3277, pruned_loss=0.0826, over 5693550.59 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3322, pruned_loss=0.08827, over 5745587.39 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3286, pruned_loss=0.08264, over 5678643.21 frames. ], batch size: 82, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:36:03,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.716e+02 1.258e+03 1.608e+03 2.254e+03 6.590e+03, threshold=3.216e+03, percent-clipped=5.0
+2023-03-10 21:36:31,252 INFO [train.py:968] (0/2) Epoch 21, batch 15850, giga_loss[loss=0.2479, simple_loss=0.324, pruned_loss=0.08591, over 28730.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.326, pruned_loss=0.08225, over 5684400.58 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.332, pruned_loss=0.08806, over 5744661.32 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3268, pruned_loss=0.08235, over 5671991.71 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:36:48,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5040, 1.9245, 1.8839, 1.7000], device='cuda:0'), covar=tensor([0.1937, 0.1903, 0.2004, 0.1929], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0730, 0.0696, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 21:36:59,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0880, 3.3660, 1.2627, 1.5452], device='cuda:0'), covar=tensor([0.1290, 0.0437, 0.1092, 0.1494], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0550, 0.0383, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 21:37:35,505 INFO [train.py:968] (0/2) Epoch 21, batch 15900, giga_loss[loss=0.2543, simple_loss=0.3339, pruned_loss=0.08739, over 28862.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3252, pruned_loss=0.08231, over 5681221.41 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3318, pruned_loss=0.08803, over 5746045.83 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.326, pruned_loss=0.08236, over 5669842.94 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:37:44,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3805, 3.4518, 1.5180, 1.5416], device='cuda:0'), covar=tensor([0.1016, 0.0262, 0.0933, 0.1389], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0549, 0.0383, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 21:38:07,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.309e+02 1.418e+03 1.944e+03 2.979e+03 8.365e+03, threshold=3.889e+03, percent-clipped=23.0
+2023-03-10 21:38:08,150 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=928276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:38:12,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=928279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:38:36,289 INFO [train.py:968] (0/2) Epoch 21, batch 15950, giga_loss[loss=0.3012, simple_loss=0.3686, pruned_loss=0.1169, over 28698.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3281, pruned_loss=0.08362, over 5679920.09 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3312, pruned_loss=0.08774, over 5747827.54 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3291, pruned_loss=0.08385, over 5667950.00 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:38:41,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=928308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:38:56,720 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=928319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:39:05,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=928324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:39:07,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4568, 1.7784, 1.5255, 1.6290], device='cuda:0'), covar=tensor([0.0719, 0.0358, 0.0325, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-10 21:39:14,331 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=928331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:39:28,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2445, 1.5702, 1.2204, 0.9498], device='cuda:0'), covar=tensor([0.2677, 0.2605, 0.3059, 0.2346], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1075, 0.1321, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 21:39:39,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1177, 1.5788, 1.5037, 1.4311], device='cuda:0'), covar=tensor([0.2096, 0.1394, 0.1666, 0.1512], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0727, 0.0692, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 21:39:41,868 INFO [train.py:968] (0/2) Epoch 21, batch 16000, giga_loss[loss=0.2936, simple_loss=0.3633, pruned_loss=0.1119, over 28936.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3294, pruned_loss=0.08502, over 5678654.24 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.331, pruned_loss=0.08766, over 5751093.39 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3304, pruned_loss=0.0852, over 5665257.97 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:39:44,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4369, 1.7869, 1.4338, 1.3165], device='cuda:0'), covar=tensor([0.2496, 0.2293, 0.2627, 0.2160], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1075, 0.1321, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 21:40:10,891 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-10 21:40:11,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.840e+02 1.431e+03 1.916e+03 2.786e+03 8.572e+03, threshold=3.831e+03, percent-clipped=11.0
+2023-03-10 21:40:16,576 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.45 vs. limit=5.0
+2023-03-10 21:40:37,889 INFO [train.py:968] (0/2) Epoch 21, batch 16050, giga_loss[loss=0.2788, simple_loss=0.3545, pruned_loss=0.1015, over 28854.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3322, pruned_loss=0.08664, over 5674555.78 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3309, pruned_loss=0.08758, over 5743788.72 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3331, pruned_loss=0.08681, over 5668514.41 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:40:53,648 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 21:41:30,867 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-10 21:41:39,939 INFO [train.py:968] (0/2) Epoch 21, batch 16100, giga_loss[loss=0.2494, simple_loss=0.3388, pruned_loss=0.08004, over 29027.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3352, pruned_loss=0.08698, over 5681248.92 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3311, pruned_loss=0.08776, over 5745278.50 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3357, pruned_loss=0.08695, over 5674549.18 frames. ], batch size: 285, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:42:01,141 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=928467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:42:03,688 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=928470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:42:09,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.859e+02 1.514e+03 1.851e+03 2.590e+03 5.603e+03, threshold=3.703e+03, percent-clipped=6.0
+2023-03-10 21:42:39,469 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=928499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:42:42,330 INFO [train.py:968] (0/2) Epoch 21, batch 16150, giga_loss[loss=0.2059, simple_loss=0.2785, pruned_loss=0.06664, over 24300.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3355, pruned_loss=0.087, over 5680545.52 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3309, pruned_loss=0.08752, over 5748255.06 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3363, pruned_loss=0.08719, over 5671299.05 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:42:46,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2414, 3.0676, 1.3642, 1.3957], device='cuda:0'), covar=tensor([0.1000, 0.0399, 0.0985, 0.1382], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0548, 0.0382, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 21:43:46,121 INFO [train.py:968] (0/2) Epoch 21, batch 16200, giga_loss[loss=0.2443, simple_loss=0.3258, pruned_loss=0.08141, over 28959.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3347, pruned_loss=0.08661, over 5687007.88 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3306, pruned_loss=0.08734, over 5746462.13 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3357, pruned_loss=0.0869, over 5678729.93 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:43:57,351 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5615, 1.8367, 1.7045, 1.4283], device='cuda:0'), covar=tensor([0.2150, 0.1606, 0.1498, 0.1843], device='cuda:0'), in_proj_covar=tensor([0.1923, 0.1838, 0.1763, 0.1910], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 21:44:17,071 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.429e+03 1.885e+03 2.667e+03 6.015e+03, threshold=3.771e+03, percent-clipped=10.0
+2023-03-10 21:44:44,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6461, 5.4693, 5.2483, 2.4929], device='cuda:0'), covar=tensor([0.0387, 0.0552, 0.0652, 0.1786], device='cuda:0'), in_proj_covar=tensor([0.1189, 0.1096, 0.0931, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 21:44:47,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3942, 1.5109, 1.5779, 1.2278], device='cuda:0'), covar=tensor([0.1710, 0.2638, 0.1420, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0894, 0.0690, 0.0938, 0.0837], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 21:44:48,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5070, 1.9451, 1.7847, 1.6600], device='cuda:0'), covar=tensor([0.2031, 0.2332, 0.2141, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0728, 0.0694, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 21:44:49,237 INFO [train.py:968] (0/2) Epoch 21, batch 16250, giga_loss[loss=0.2614, simple_loss=0.3504, pruned_loss=0.08624, over 28480.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3346, pruned_loss=0.0875, over 5687362.88 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3307, pruned_loss=0.0875, over 5740027.07 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3354, pruned_loss=0.08758, over 5685461.00 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:45:18,247 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-10 21:45:28,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2866, 4.1089, 3.9079, 1.8332], device='cuda:0'), covar=tensor([0.0747, 0.0914, 0.1005, 0.2203], device='cuda:0'), in_proj_covar=tensor([0.1187, 0.1094, 0.0930, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 21:45:51,945 INFO [train.py:968] (0/2) Epoch 21, batch 16300, libri_loss[loss=0.2053, simple_loss=0.2799, pruned_loss=0.06528, over 28494.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3336, pruned_loss=0.08728, over 5669056.08 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3307, pruned_loss=0.08758, over 5741252.19 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3344, pruned_loss=0.08728, over 5664435.74 frames. ], batch size: 63, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:46:25,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.256e+02 1.555e+03 2.091e+03 3.260e+03 6.529e+03, threshold=4.182e+03, percent-clipped=18.0
+2023-03-10 21:46:48,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=928694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:46:49,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3625, 1.7433, 1.8399, 1.4783], device='cuda:0'), covar=tensor([0.2932, 0.2045, 0.1977, 0.2353], device='cuda:0'), in_proj_covar=tensor([0.1918, 0.1833, 0.1756, 0.1904], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 21:46:54,749 INFO [train.py:968] (0/2) Epoch 21, batch 16350, giga_loss[loss=0.2275, simple_loss=0.307, pruned_loss=0.07404, over 28936.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3325, pruned_loss=0.08758, over 5676036.38 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3306, pruned_loss=0.08746, over 5743765.71 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3333, pruned_loss=0.0877, over 5669122.95 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:47:03,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=928706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:47:59,457 INFO [train.py:968] (0/2) Epoch 21, batch 16400, giga_loss[loss=0.253, simple_loss=0.3326, pruned_loss=0.08668, over 28911.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.331, pruned_loss=0.08735, over 5675900.76 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3301, pruned_loss=0.0873, over 5746068.76 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.332, pruned_loss=0.08758, over 5667386.76 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:48:02,432 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=928755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:48:04,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3252, 1.4322, 1.2976, 1.4555], device='cuda:0'), covar=tensor([0.0756, 0.0331, 0.0341, 0.0878], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-10 21:48:27,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.284e+02 1.286e+03 1.796e+03 2.642e+03 1.028e+04, threshold=3.592e+03, percent-clipped=10.0
+2023-03-10 21:48:57,997 INFO [train.py:968] (0/2) Epoch 21, batch 16450, giga_loss[loss=0.2393, simple_loss=0.3233, pruned_loss=0.07766, over 28612.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3313, pruned_loss=0.08664, over 5678615.71 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.33, pruned_loss=0.08721, over 5748708.48 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3322, pruned_loss=0.08687, over 5668756.04 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:49:01,300 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-10 21:49:41,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=928837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:49:45,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=928840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:49:46,533 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-10 21:49:55,578 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=928849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:49:55,716 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-10 21:49:56,534 INFO [train.py:968] (0/2) Epoch 21, batch 16500, giga_loss[loss=0.2552, simple_loss=0.3356, pruned_loss=0.08736, over 28063.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.331, pruned_loss=0.08515, over 5680617.41 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3301, pruned_loss=0.08731, over 5751618.87 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3316, pruned_loss=0.08521, over 5668463.19 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:49:57,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=928852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:50:06,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 21:50:17,479 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=928869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:50:25,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.500e+02 1.235e+03 1.657e+03 2.139e+03 4.219e+03, threshold=3.314e+03, percent-clipped=3.0
+2023-03-10 21:50:31,093 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=928881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:50:33,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8865, 1.1402, 2.8609, 2.6724], device='cuda:0'), covar=tensor([0.1583, 0.2628, 0.0520, 0.1577], device='cuda:0'), in_proj_covar=tensor([0.0749, 0.0643, 0.0940, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 21:50:39,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4930, 1.7602, 1.2716, 1.3135], device='cuda:0'), covar=tensor([0.1067, 0.0576, 0.1067, 0.1149], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0442, 0.0511, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 21:50:53,271 INFO [train.py:968] (0/2) Epoch 21, batch 16550, giga_loss[loss=0.241, simple_loss=0.335, pruned_loss=0.07351, over 28942.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3326, pruned_loss=0.08451, over 5676057.87 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3299, pruned_loss=0.08707, over 5753653.24 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3333, pruned_loss=0.08472, over 5663057.97 frames. ], batch size: 120, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:51:46,073 INFO [train.py:968] (0/2) Epoch 21, batch 16600, giga_loss[loss=0.2583, simple_loss=0.3413, pruned_loss=0.08763, over 28843.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3333, pruned_loss=0.08429, over 5680672.78 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3302, pruned_loss=0.0873, over 5748012.11 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3338, pruned_loss=0.08419, over 5673433.98 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:52:20,483 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.668e+02 1.481e+03 2.064e+03 2.746e+03 5.019e+03, threshold=4.129e+03, percent-clipped=14.0
+2023-03-10 21:52:47,852 INFO [train.py:968] (0/2) Epoch 21, batch 16650, giga_loss[loss=0.2604, simple_loss=0.3419, pruned_loss=0.08946, over 28381.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3329, pruned_loss=0.0842, over 5685642.99 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3299, pruned_loss=0.08724, over 5749950.93 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3336, pruned_loss=0.08407, over 5676205.15 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:53:09,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:53:34,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4161, 1.8500, 1.7159, 1.6043], device='cuda:0'), covar=tensor([0.1997, 0.2003, 0.2104, 0.2019], device='cuda:0'), in_proj_covar=tensor([0.0453, 0.0728, 0.0695, 0.0669], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 21:53:52,186 INFO [train.py:968] (0/2) Epoch 21, batch 16700, giga_loss[loss=0.27, simple_loss=0.3571, pruned_loss=0.09149, over 28736.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3328, pruned_loss=0.08421, over 5684449.84 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3295, pruned_loss=0.08714, over 5754662.20 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3338, pruned_loss=0.08414, over 5670798.81 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:54:31,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.873e+02 1.424e+03 1.849e+03 2.896e+03 7.588e+03, threshold=3.698e+03, percent-clipped=12.0
+2023-03-10 21:55:02,821 INFO [train.py:968] (0/2) Epoch 21, batch 16750, giga_loss[loss=0.2685, simple_loss=0.3526, pruned_loss=0.09218, over 28631.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3336, pruned_loss=0.08475, over 5677970.81 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3294, pruned_loss=0.08717, over 5756145.22 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3346, pruned_loss=0.0846, over 5664118.20 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:55:43,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=929130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:56:10,863 INFO [train.py:968] (0/2) Epoch 21, batch 16800, giga_loss[loss=0.2952, simple_loss=0.351, pruned_loss=0.1196, over 26853.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3341, pruned_loss=0.08445, over 5686232.87 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.329, pruned_loss=0.08691, over 5759028.74 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3354, pruned_loss=0.08448, over 5670243.43 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:56:55,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.901e+02 1.472e+03 1.917e+03 3.009e+03 8.493e+03, threshold=3.834e+03, percent-clipped=13.0
+2023-03-10 21:57:23,864 INFO [train.py:968] (0/2) Epoch 21, batch 16850, libri_loss[loss=0.2467, simple_loss=0.329, pruned_loss=0.0822, over 29534.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3372, pruned_loss=0.08599, over 5686473.66 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3289, pruned_loss=0.08673, over 5762126.55 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3386, pruned_loss=0.08616, over 5669389.08 frames. ], batch size: 89, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:57:53,652 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:58:07,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5279, 1.7706, 1.4639, 1.6111], device='cuda:0'), covar=tensor([0.2752, 0.2859, 0.3352, 0.2433], device='cuda:0'), in_proj_covar=tensor([0.1489, 0.1073, 0.1317, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 21:58:23,535 INFO [train.py:968] (0/2) Epoch 21, batch 16900, giga_loss[loss=0.2421, simple_loss=0.3341, pruned_loss=0.07503, over 28471.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3381, pruned_loss=0.0857, over 5687199.72 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.329, pruned_loss=0.08678, over 5752554.44 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3393, pruned_loss=0.0858, over 5680122.32 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:58:56,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=929273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:59:00,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=929276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:59:05,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.897e+02 1.372e+03 1.787e+03 2.630e+03 1.097e+04, threshold=3.573e+03, percent-clipped=8.0
+2023-03-10 21:59:07,746 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6797, 1.7709, 1.4589, 1.7575], device='cuda:0'), covar=tensor([0.2587, 0.2725, 0.3169, 0.2649], device='cuda:0'), in_proj_covar=tensor([0.1490, 0.1074, 0.1319, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 21:59:36,919 INFO [train.py:968] (0/2) Epoch 21, batch 16950, giga_loss[loss=0.2791, simple_loss=0.3452, pruned_loss=0.1065, over 27591.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3363, pruned_loss=0.08536, over 5687397.40 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3288, pruned_loss=0.08668, over 5754120.99 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3374, pruned_loss=0.08551, over 5679846.83 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 21:59:42,371 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=929305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:00:47,388 INFO [train.py:968] (0/2) Epoch 21, batch 17000, libri_loss[loss=0.2804, simple_loss=0.3588, pruned_loss=0.101, over 29513.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.334, pruned_loss=0.08436, over 5692473.09 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3288, pruned_loss=0.08666, over 5756177.74 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.335, pruned_loss=0.08447, over 5683642.05 frames. ], batch size: 89, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:01:13,801 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-10 22:01:29,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.793e+02 1.368e+03 1.815e+03 2.920e+03 6.594e+03, threshold=3.630e+03, percent-clipped=11.0
+2023-03-10 22:01:47,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=929391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:01:59,098 INFO [train.py:968] (0/2) Epoch 21, batch 17050, giga_loss[loss=0.2266, simple_loss=0.3099, pruned_loss=0.07171, over 28974.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3318, pruned_loss=0.08215, over 5700382.76 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3286, pruned_loss=0.0865, over 5757834.61 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3329, pruned_loss=0.08234, over 5691554.74 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:02:31,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1897, 1.2418, 3.4759, 3.0601], device='cuda:0'), covar=tensor([0.1611, 0.2816, 0.0444, 0.1700], device='cuda:0'), in_proj_covar=tensor([0.0750, 0.0644, 0.0941, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 22:03:00,174 INFO [train.py:968] (0/2) Epoch 21, batch 17100, giga_loss[loss=0.2574, simple_loss=0.3281, pruned_loss=0.09333, over 24254.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3312, pruned_loss=0.08197, over 5687602.35 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3283, pruned_loss=0.08625, over 5756646.20 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3323, pruned_loss=0.08228, over 5681112.72 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:03:30,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.849e+02 1.274e+03 1.606e+03 2.048e+03 4.232e+03, threshold=3.211e+03, percent-clipped=3.0
+2023-03-10 22:03:54,141 INFO [train.py:968] (0/2) Epoch 21, batch 17150, giga_loss[loss=0.283, simple_loss=0.3643, pruned_loss=0.1009, over 28939.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3336, pruned_loss=0.08362, over 5695454.86 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3282, pruned_loss=0.08626, over 5761828.76 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3347, pruned_loss=0.08374, over 5683608.35 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:04:33,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=929534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:04:36,415 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=929537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:04:46,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5450, 1.3396, 4.8152, 3.7740], device='cuda:0'), covar=tensor([0.2001, 0.3169, 0.0699, 0.0934], device='cuda:0'), in_proj_covar=tensor([0.0750, 0.0645, 0.0941, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 22:04:49,173 INFO [train.py:968] (0/2) Epoch 21, batch 17200, giga_loss[loss=0.2287, simple_loss=0.3147, pruned_loss=0.07134, over 28621.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3359, pruned_loss=0.0854, over 5691003.82 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3283, pruned_loss=0.08622, over 5764118.63 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3368, pruned_loss=0.08549, over 5677233.43 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:05:05,837 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=929566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:05:21,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.517e+03 2.224e+03 3.273e+03 7.228e+03, threshold=4.449e+03, percent-clipped=26.0
+2023-03-10 22:05:39,147 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=929597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:05:44,581 INFO [train.py:968] (0/2) Epoch 21, batch 17250, giga_loss[loss=0.2378, simple_loss=0.3233, pruned_loss=0.07612, over 28878.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3337, pruned_loss=0.08545, over 5688317.61 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.328, pruned_loss=0.08601, over 5764679.12 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3349, pruned_loss=0.08569, over 5675530.13 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:06:15,135 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:06:40,566 INFO [train.py:968] (0/2) Epoch 21, batch 17300, giga_loss[loss=0.2694, simple_loss=0.3427, pruned_loss=0.09799, over 28560.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3343, pruned_loss=0.08672, over 5684661.11 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.328, pruned_loss=0.08594, over 5763415.10 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3353, pruned_loss=0.08699, over 5674723.36 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:07:15,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.005e+02 1.347e+03 1.790e+03 2.394e+03 1.151e+04, threshold=3.580e+03, percent-clipped=3.0
+2023-03-10 22:07:36,937 INFO [train.py:968] (0/2) Epoch 21, batch 17350, giga_loss[loss=0.2572, simple_loss=0.3458, pruned_loss=0.08433, over 28683.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3367, pruned_loss=0.08825, over 5693244.90 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.328, pruned_loss=0.08585, over 5766350.64 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3376, pruned_loss=0.08856, over 5681500.02 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:08:18,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=929740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:08:20,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=929743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:08:25,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6844, 1.8520, 1.6288, 1.7816], device='cuda:0'), covar=tensor([0.0770, 0.0286, 0.0318, 0.0876], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0117, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-10 22:08:27,393 INFO [train.py:968] (0/2) Epoch 21, batch 17400, giga_loss[loss=0.2799, simple_loss=0.3664, pruned_loss=0.09672, over 28993.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3448, pruned_loss=0.09313, over 5694962.85 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3278, pruned_loss=0.08564, over 5770311.14 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3461, pruned_loss=0.09371, over 5679851.28 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:08:45,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=929772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:08:51,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.182e+02 1.473e+03 1.798e+03 2.248e+03 7.738e+03, threshold=3.596e+03, percent-clipped=5.0
+2023-03-10 22:09:11,630 INFO [train.py:968] (0/2) Epoch 21, batch 17450, giga_loss[loss=0.3029, simple_loss=0.3784, pruned_loss=0.1137, over 28881.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3508, pruned_loss=0.09645, over 5694934.33 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3271, pruned_loss=0.08525, over 5765586.76 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.353, pruned_loss=0.09755, over 5684845.71 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:09:16,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3919, 1.4797, 4.0185, 3.3053], device='cuda:0'), covar=tensor([0.1678, 0.2652, 0.0446, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0752, 0.0647, 0.0945, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 22:09:47,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7034, 1.8897, 1.5154, 1.8822], device='cuda:0'), covar=tensor([0.2690, 0.2724, 0.3110, 0.2577], device='cuda:0'), in_proj_covar=tensor([0.1490, 0.1074, 0.1317, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 22:09:51,403 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:09:54,896 INFO [train.py:968] (0/2) Epoch 21, batch 17500, giga_loss[loss=0.2983, simple_loss=0.347, pruned_loss=0.1248, over 23696.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.348, pruned_loss=0.0958, over 5694046.42 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3267, pruned_loss=0.08507, over 5767594.96 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3504, pruned_loss=0.09698, over 5683413.32 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:10:09,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2114, 1.7084, 1.2365, 0.5836], device='cuda:0'), covar=tensor([0.4946, 0.2474, 0.2948, 0.6111], device='cuda:0'), in_proj_covar=tensor([0.1731, 0.1630, 0.1586, 0.1411], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 22:10:22,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.411e+02 1.238e+03 1.490e+03 2.032e+03 1.036e+04, threshold=2.980e+03, percent-clipped=4.0
+2023-03-10 22:10:28,128 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:10:33,011 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:10:39,885 INFO [train.py:968] (0/2) Epoch 21, batch 17550, giga_loss[loss=0.228, simple_loss=0.309, pruned_loss=0.07347, over 28556.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3407, pruned_loss=0.09293, over 5673953.91 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3269, pruned_loss=0.0852, over 5749091.48 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3428, pruned_loss=0.09392, over 5680962.00 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:11:06,637 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-10 22:11:22,413 INFO [train.py:968] (0/2) Epoch 21, batch 17600, giga_loss[loss=0.2171, simple_loss=0.2926, pruned_loss=0.07081, over 28257.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.334, pruned_loss=0.08978, over 5673061.71 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.327, pruned_loss=0.08493, over 5750386.34 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3359, pruned_loss=0.09103, over 5675062.43 frames. ], batch size: 65, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:11:45,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.105e+02 1.070e+03 1.357e+03 1.688e+03 5.995e+03, threshold=2.714e+03, percent-clipped=4.0
+2023-03-10 22:12:02,480 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-930000.pt
+2023-03-10 22:12:03,484 INFO [train.py:968] (0/2) Epoch 21, batch 17650, libri_loss[loss=0.2545, simple_loss=0.3403, pruned_loss=0.08432, over 29162.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3272, pruned_loss=0.08681, over 5673105.16 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3277, pruned_loss=0.08523, over 5744443.18 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3282, pruned_loss=0.08761, over 5677882.42 frames. ], batch size: 101, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:12:03,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=930001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:12:08,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5720, 3.2143, 1.6032, 1.5708], device='cuda:0'), covar=tensor([0.0928, 0.0332, 0.0885, 0.1291], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0547, 0.0382, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 22:12:43,794 INFO [train.py:968] (0/2) Epoch 21, batch 17700, giga_loss[loss=0.2295, simple_loss=0.3013, pruned_loss=0.07883, over 28805.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.321, pruned_loss=0.08404, over 5668662.42 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3275, pruned_loss=0.08508, over 5729636.08 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3219, pruned_loss=0.08486, over 5682670.77 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:13:09,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.551e+02 1.072e+03 1.462e+03 1.762e+03 3.476e+03, threshold=2.923e+03, percent-clipped=6.0
+2023-03-10 22:13:10,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6908, 1.8947, 1.2815, 1.3365], device='cuda:0'), covar=tensor([0.1303, 0.0890, 0.1677, 0.1588], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0441, 0.0511, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 22:13:25,615 INFO [train.py:968] (0/2) Epoch 21, batch 17750, libri_loss[loss=0.2819, simple_loss=0.366, pruned_loss=0.09893, over 27621.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3167, pruned_loss=0.08218, over 5680431.24 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3282, pruned_loss=0.0854, over 5733016.35 frames. ], giga_tot_loss[loss=0.2407, simple_loss=0.3165, pruned_loss=0.08248, over 5687177.97 frames. ], batch size: 115, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:14:01,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=930144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:14:03,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=930147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:14:07,032 INFO [train.py:968] (0/2) Epoch 21, batch 17800, giga_loss[loss=0.2719, simple_loss=0.3271, pruned_loss=0.1083, over 27519.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3144, pruned_loss=0.08149, over 5676921.64 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3285, pruned_loss=0.08553, over 5726130.17 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3137, pruned_loss=0.08154, over 5688181.98 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:14:28,969 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=930176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:14:33,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.742e+02 1.086e+03 1.433e+03 2.058e+03 5.440e+03, threshold=2.866e+03, percent-clipped=12.0
+2023-03-10 22:14:46,656 INFO [train.py:968] (0/2) Epoch 21, batch 17850, giga_loss[loss=0.2293, simple_loss=0.3032, pruned_loss=0.07766, over 28669.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3116, pruned_loss=0.08012, over 5687756.69 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.329, pruned_loss=0.08568, over 5730266.01 frames. ], giga_tot_loss[loss=0.2349, simple_loss=0.3101, pruned_loss=0.07983, over 5691232.84 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:15:01,498 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 22:15:03,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=930223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:15:13,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7893, 5.5560, 5.2286, 2.9843], device='cuda:0'), covar=tensor([0.0373, 0.0604, 0.0676, 0.1620], device='cuda:0'), in_proj_covar=tensor([0.1187, 0.1097, 0.0933, 0.0710], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 22:15:25,316 INFO [train.py:968] (0/2) Epoch 21, batch 17900, giga_loss[loss=0.2121, simple_loss=0.2912, pruned_loss=0.06652, over 28785.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3098, pruned_loss=0.07936, over 5693958.89 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3292, pruned_loss=0.08559, over 5734383.60 frames. ], giga_tot_loss[loss=0.2328, simple_loss=0.3076, pruned_loss=0.079, over 5691825.37 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:15:28,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1772, 2.3077, 2.0311, 2.1396], device='cuda:0'), covar=tensor([0.1943, 0.2470, 0.2407, 0.2170], device='cuda:0'), in_proj_covar=tensor([0.0457, 0.0733, 0.0701, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 22:15:34,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=930261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:15:38,871 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=930267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:15:50,535 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.170e+02 1.050e+03 1.268e+03 1.588e+03 4.625e+03, threshold=2.537e+03, percent-clipped=3.0
+2023-03-10 22:16:04,990 INFO [train.py:968] (0/2) Epoch 21, batch 17950, giga_loss[loss=0.2227, simple_loss=0.301, pruned_loss=0.07221, over 28782.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3068, pruned_loss=0.07791, over 5700251.65 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3296, pruned_loss=0.08555, over 5739018.45 frames. ], giga_tot_loss[loss=0.2295, simple_loss=0.3042, pruned_loss=0.07746, over 5693284.32 frames. ], batch size: 243, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:16:33,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3353, 1.7184, 1.0030, 1.2332], device='cuda:0'), covar=tensor([0.1320, 0.0729, 0.1687, 0.1422], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0442, 0.0512, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 22:16:47,689 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-10 22:16:49,182 INFO [train.py:968] (0/2) Epoch 21, batch 18000, giga_loss[loss=0.1882, simple_loss=0.2638, pruned_loss=0.05635, over 28960.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3034, pruned_loss=0.07645, over 5695191.32 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3295, pruned_loss=0.08548, over 5740547.43 frames. ], giga_tot_loss[loss=0.2267, simple_loss=0.3011, pruned_loss=0.0761, over 5687980.56 frames. ], batch size: 106, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:16:49,185 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 22:16:57,370 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2048, simple_loss=0.311, pruned_loss=0.04933, over 944034.00 frames. 
+2023-03-10 22:16:57,371 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 22:17:11,428 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=930366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:13,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=930369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:22,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.810e+02 1.012e+03 1.306e+03 1.855e+03 7.500e+03, threshold=2.612e+03, percent-clipped=9.0
+2023-03-10 22:17:36,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=930398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:38,364 INFO [train.py:968] (0/2) Epoch 21, batch 18050, giga_loss[loss=0.2322, simple_loss=0.3057, pruned_loss=0.0794, over 29094.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3009, pruned_loss=0.07564, over 5688191.63 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3294, pruned_loss=0.08553, over 5735739.16 frames. ], giga_tot_loss[loss=0.2243, simple_loss=0.2985, pruned_loss=0.07506, over 5684775.72 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:17:42,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=930404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:44,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=930407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:46,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=930410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:46,428 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=930410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:48,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5457, 1.6333, 1.3945, 1.6727], device='cuda:0'), covar=tensor([0.0756, 0.0330, 0.0331, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-10 22:17:48,324 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=930413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:18:08,660 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=930436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:18:15,350 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=930442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:18:21,905 INFO [train.py:968] (0/2) Epoch 21, batch 18100, giga_loss[loss=0.2024, simple_loss=0.2824, pruned_loss=0.06113, over 28699.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2974, pruned_loss=0.07395, over 5689853.64 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3295, pruned_loss=0.08542, over 5738468.19 frames. ], giga_tot_loss[loss=0.2209, simple_loss=0.295, pruned_loss=0.07343, over 5683794.83 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:18:53,045 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.364e+02 1.055e+03 1.263e+03 1.658e+03 3.306e+03, threshold=2.526e+03, percent-clipped=3.0
+2023-03-10 22:19:03,818 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 22:19:07,563 INFO [train.py:968] (0/2) Epoch 21, batch 18150, giga_loss[loss=0.2081, simple_loss=0.2729, pruned_loss=0.07163, over 28812.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2939, pruned_loss=0.07223, over 5682348.34 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3298, pruned_loss=0.08545, over 5735472.59 frames. ], giga_tot_loss[loss=0.2171, simple_loss=0.291, pruned_loss=0.07155, over 5679229.59 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:19:52,566 INFO [train.py:968] (0/2) Epoch 21, batch 18200, giga_loss[loss=0.2691, simple_loss=0.3517, pruned_loss=0.09327, over 28870.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.2991, pruned_loss=0.07595, over 5674343.55 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.33, pruned_loss=0.08567, over 5737278.07 frames. ], giga_tot_loss[loss=0.2228, simple_loss=0.2958, pruned_loss=0.07492, over 5668860.58 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:20:22,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.179e+02 1.133e+03 1.503e+03 1.823e+03 4.018e+03, threshold=3.006e+03, percent-clipped=12.0
+2023-03-10 22:20:29,527 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-10 22:20:40,533 INFO [train.py:968] (0/2) Epoch 21, batch 18250, giga_loss[loss=0.3254, simple_loss=0.393, pruned_loss=0.1288, over 27535.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3119, pruned_loss=0.08232, over 5677713.16 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3303, pruned_loss=0.08574, over 5739447.24 frames. ], giga_tot_loss[loss=0.2358, simple_loss=0.3088, pruned_loss=0.08137, over 5670442.93 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:21:20,092 INFO [train.py:968] (0/2) Epoch 21, batch 18300, giga_loss[loss=0.2942, simple_loss=0.3739, pruned_loss=0.1072, over 28529.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3246, pruned_loss=0.08878, over 5692007.98 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3304, pruned_loss=0.08584, over 5743330.15 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3217, pruned_loss=0.08796, over 5681394.32 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:21:22,706 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=930654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:21:43,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.602e+02 1.427e+03 1.694e+03 2.282e+03 6.133e+03, threshold=3.388e+03, percent-clipped=9.0
+2023-03-10 22:21:58,016 INFO [train.py:968] (0/2) Epoch 21, batch 18350, giga_loss[loss=0.3733, simple_loss=0.4178, pruned_loss=0.1644, over 26493.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3327, pruned_loss=0.09242, over 5692906.54 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3303, pruned_loss=0.08577, over 5744998.37 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3305, pruned_loss=0.09199, over 5681697.32 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:22:38,735 INFO [train.py:968] (0/2) Epoch 21, batch 18400, giga_loss[loss=0.2612, simple_loss=0.3443, pruned_loss=0.08908, over 28420.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3389, pruned_loss=0.09473, over 5691040.39 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3309, pruned_loss=0.08606, over 5749218.80 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3367, pruned_loss=0.09431, over 5677392.90 frames. ], batch size: 71, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:22:53,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5964, 1.7994, 1.5018, 1.7431], device='cuda:0'), covar=tensor([0.2661, 0.2652, 0.2773, 0.2562], device='cuda:0'), in_proj_covar=tensor([0.1483, 0.1073, 0.1311, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 22:22:56,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5711, 1.7249, 1.2585, 1.2293], device='cuda:0'), covar=tensor([0.1025, 0.0658, 0.1092, 0.1285], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0442, 0.0513, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 22:23:02,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.255e+02 1.284e+03 1.821e+03 2.476e+03 8.626e+03, threshold=3.641e+03, percent-clipped=11.0
+2023-03-10 22:23:05,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=930785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:23:19,197 INFO [train.py:968] (0/2) Epoch 21, batch 18450, giga_loss[loss=0.2557, simple_loss=0.3349, pruned_loss=0.08825, over 28908.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.341, pruned_loss=0.0948, over 5690902.41 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3314, pruned_loss=0.08644, over 5752607.09 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3391, pruned_loss=0.09432, over 5675496.82 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:24:02,998 INFO [train.py:968] (0/2) Epoch 21, batch 18500, giga_loss[loss=0.2375, simple_loss=0.3165, pruned_loss=0.0793, over 28368.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3426, pruned_loss=0.09555, over 5680468.90 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3313, pruned_loss=0.08639, over 5754664.74 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3412, pruned_loss=0.09533, over 5665696.60 frames. ], batch size: 65, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:24:17,101 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-10 22:24:29,205 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.582e+02 1.169e+03 1.456e+03 1.928e+03 4.250e+03, threshold=2.912e+03, percent-clipped=1.0
+2023-03-10 22:24:44,745 INFO [train.py:968] (0/2) Epoch 21, batch 18550, giga_loss[loss=0.3006, simple_loss=0.3706, pruned_loss=0.1153, over 28937.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3452, pruned_loss=0.0976, over 5686151.48 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3317, pruned_loss=0.08654, over 5756472.38 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.344, pruned_loss=0.09748, over 5671564.34 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:25:02,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2774, 1.3385, 1.2046, 1.1966], device='cuda:0'), covar=tensor([0.1885, 0.2111, 0.1679, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.1953, 0.1864, 0.1790, 0.1946], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 22:25:02,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5392, 1.7877, 1.4525, 1.7094], device='cuda:0'), covar=tensor([0.2692, 0.2705, 0.2882, 0.2523], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1080, 0.1319, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 22:25:05,597 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.35 vs. limit=5.0
+2023-03-10 22:25:07,354 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=930928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:25:09,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=930931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:25:24,906 INFO [train.py:968] (0/2) Epoch 21, batch 18600, libri_loss[loss=0.3288, simple_loss=0.3926, pruned_loss=0.1325, over 20170.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3492, pruned_loss=0.1002, over 5677414.51 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3327, pruned_loss=0.08693, over 5747141.09 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3479, pruned_loss=0.1002, over 5672085.23 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:25:32,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=930960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:25:51,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.072e+02 1.365e+03 1.758e+03 2.291e+03 8.119e+03, threshold=3.515e+03, percent-clipped=14.0
+2023-03-10 22:26:01,360 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 22:26:06,520 INFO [train.py:968] (0/2) Epoch 21, batch 18650, giga_loss[loss=0.2773, simple_loss=0.3564, pruned_loss=0.09906, over 28953.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3517, pruned_loss=0.101, over 5672661.35 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3329, pruned_loss=0.0871, over 5737283.39 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3506, pruned_loss=0.1009, over 5677540.78 frames. ], batch size: 106, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:26:26,927 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:26:29,643 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:26:33,000 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:26:38,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-10 22:26:47,468 INFO [train.py:968] (0/2) Epoch 21, batch 18700, giga_loss[loss=0.2726, simple_loss=0.3634, pruned_loss=0.09085, over 29074.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3531, pruned_loss=0.1007, over 5670526.07 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3327, pruned_loss=0.08681, over 5739217.67 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3527, pruned_loss=0.1012, over 5671586.24 frames. ], batch size: 128, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:27:05,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8856, 3.6857, 3.4784, 1.5944], device='cuda:0'), covar=tensor([0.0658, 0.0843, 0.0810, 0.2325], device='cuda:0'), in_proj_covar=tensor([0.1179, 0.1093, 0.0928, 0.0711], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 22:27:13,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.946e+02 1.212e+03 1.465e+03 2.243e+03 5.489e+03, threshold=2.930e+03, percent-clipped=4.0
+2023-03-10 22:27:27,412 INFO [train.py:968] (0/2) Epoch 21, batch 18750, giga_loss[loss=0.2415, simple_loss=0.33, pruned_loss=0.07649, over 28472.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3553, pruned_loss=0.1012, over 5683642.33 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.333, pruned_loss=0.08681, over 5742133.15 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3552, pruned_loss=0.1018, over 5680763.22 frames. ], batch size: 65, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:27:54,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1986, 1.1669, 1.2474, 1.4211], device='cuda:0'), covar=tensor([0.0840, 0.0382, 0.0346, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0116, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 22:27:55,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3493, 3.2251, 1.4844, 1.4962], device='cuda:0'), covar=tensor([0.1065, 0.0290, 0.0963, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0543, 0.0380, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 22:28:03,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3015, 1.5839, 1.2848, 1.0047], device='cuda:0'), covar=tensor([0.2551, 0.2514, 0.2891, 0.2353], device='cuda:0'), in_proj_covar=tensor([0.1491, 0.1078, 0.1316, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 22:28:09,234 INFO [train.py:968] (0/2) Epoch 21, batch 18800, giga_loss[loss=0.2893, simple_loss=0.3606, pruned_loss=0.1091, over 28340.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3557, pruned_loss=0.1005, over 5689363.64 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3329, pruned_loss=0.08679, over 5744644.82 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.356, pruned_loss=0.1012, over 5683923.46 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:28:25,959 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=931172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:28:29,164 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=931175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:28:35,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.063e+02 1.256e+03 1.618e+03 2.015e+03 6.444e+03, threshold=3.237e+03, percent-clipped=11.0
+2023-03-10 22:28:40,155 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 22:28:49,034 INFO [train.py:968] (0/2) Epoch 21, batch 18850, giga_loss[loss=0.2514, simple_loss=0.3321, pruned_loss=0.08535, over 28799.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3542, pruned_loss=0.09815, over 5704880.68 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3328, pruned_loss=0.08667, over 5746974.51 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3548, pruned_loss=0.099, over 5697822.23 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:28:52,451 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=931204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:29:28,342 INFO [train.py:968] (0/2) Epoch 21, batch 18900, giga_loss[loss=0.2522, simple_loss=0.3419, pruned_loss=0.08124, over 28875.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3526, pruned_loss=0.09709, over 5702947.87 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3333, pruned_loss=0.08689, over 5750296.31 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3532, pruned_loss=0.09786, over 5693114.22 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:29:33,693 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:29:37,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3138, 3.0448, 1.4662, 1.4155], device='cuda:0'), covar=tensor([0.1055, 0.0272, 0.0924, 0.1476], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0544, 0.0380, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 22:29:42,357 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 22:29:44,851 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931275.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:29:47,727 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4301, 1.3141, 1.5139, 1.1497], device='cuda:0'), covar=tensor([0.1670, 0.2760, 0.1343, 0.1516], device='cuda:0'), in_proj_covar=tensor([0.0903, 0.0697, 0.0949, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 22:29:51,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.406e+02 1.189e+03 1.480e+03 1.997e+03 5.905e+03, threshold=2.960e+03, percent-clipped=6.0
+2023-03-10 22:30:05,557 INFO [train.py:968] (0/2) Epoch 21, batch 18950, giga_loss[loss=0.2814, simple_loss=0.3476, pruned_loss=0.1076, over 28514.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3553, pruned_loss=0.09968, over 5705435.10 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3337, pruned_loss=0.08713, over 5741891.63 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3561, pruned_loss=0.1004, over 5703138.40 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:30:47,557 INFO [train.py:968] (0/2) Epoch 21, batch 19000, giga_loss[loss=0.291, simple_loss=0.3584, pruned_loss=0.1118, over 28576.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3577, pruned_loss=0.1037, over 5713961.90 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3345, pruned_loss=0.08759, over 5747495.14 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3583, pruned_loss=0.1044, over 5706051.04 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:31:10,168 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2364, 1.1704, 3.9626, 3.2138], device='cuda:0'), covar=tensor([0.1766, 0.2899, 0.0456, 0.0945], device='cuda:0'), in_proj_covar=tensor([0.0750, 0.0645, 0.0949, 0.0895], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 22:31:15,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.623e+02 1.429e+03 1.832e+03 2.434e+03 5.567e+03, threshold=3.663e+03, percent-clipped=12.0
+2023-03-10 22:31:29,551 INFO [train.py:968] (0/2) Epoch 21, batch 19050, libri_loss[loss=0.2579, simple_loss=0.3441, pruned_loss=0.08587, over 29668.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3582, pruned_loss=0.1059, over 5710954.79 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3347, pruned_loss=0.08756, over 5750325.95 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3591, pruned_loss=0.1069, over 5701358.13 frames. ], batch size: 88, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:31:29,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:31:36,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:31:36,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9566, 2.0537, 1.7966, 1.8493], device='cuda:0'), covar=tensor([0.1977, 0.2507, 0.2492, 0.2194], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0741, 0.0708, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 22:31:51,305 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-10 22:32:07,420 INFO [train.py:968] (0/2) Epoch 21, batch 19100, giga_loss[loss=0.2785, simple_loss=0.3561, pruned_loss=0.1004, over 28987.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3563, pruned_loss=0.1057, over 5709682.45 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3347, pruned_loss=0.0875, over 5753121.98 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3576, pruned_loss=0.107, over 5698213.71 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:32:33,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.699e+02 1.304e+03 1.600e+03 2.047e+03 1.142e+04, threshold=3.200e+03, percent-clipped=3.0
+2023-03-10 22:32:46,617 INFO [train.py:968] (0/2) Epoch 21, batch 19150, giga_loss[loss=0.3131, simple_loss=0.3824, pruned_loss=0.1219, over 28917.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.354, pruned_loss=0.1048, over 5706836.56 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3356, pruned_loss=0.08807, over 5755799.53 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.355, pruned_loss=0.106, over 5693316.46 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:32:46,856 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:20,924 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=931544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:22,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=931547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:24,633 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-10 22:33:27,212 INFO [train.py:968] (0/2) Epoch 21, batch 19200, libri_loss[loss=0.295, simple_loss=0.3716, pruned_loss=0.1092, over 29484.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3534, pruned_loss=0.1039, over 5712353.80 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3357, pruned_loss=0.0882, over 5751210.13 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3549, pruned_loss=0.1056, over 5703391.05 frames. ], batch size: 85, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:33:28,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=931552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:30,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=931555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:47,751 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=931576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:53,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=931584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:54,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.215e+02 1.430e+03 1.852e+03 2.541e+03 7.165e+03, threshold=3.704e+03, percent-clipped=9.0
+2023-03-10 22:34:04,986 INFO [train.py:968] (0/2) Epoch 21, batch 19250, giga_loss[loss=0.2982, simple_loss=0.361, pruned_loss=0.1177, over 26542.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3525, pruned_loss=0.1027, over 5718683.30 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3358, pruned_loss=0.08814, over 5755813.70 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3542, pruned_loss=0.1045, over 5706160.20 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:34:34,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:34:49,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931650.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:34:49,939 INFO [train.py:968] (0/2) Epoch 21, batch 19300, giga_loss[loss=0.2742, simple_loss=0.3501, pruned_loss=0.09915, over 28541.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3493, pruned_loss=0.1007, over 5701298.60 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3355, pruned_loss=0.08796, over 5757472.29 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.351, pruned_loss=0.1025, over 5689383.16 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:34:52,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9760, 1.2739, 1.0287, 0.2472], device='cuda:0'), covar=tensor([0.4200, 0.3215, 0.4840, 0.6737], device='cuda:0'), in_proj_covar=tensor([0.1729, 0.1629, 0.1586, 0.1408], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 22:35:13,972 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 22:35:19,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.265e+02 1.230e+03 1.490e+03 1.975e+03 4.501e+03, threshold=2.980e+03, percent-clipped=2.0
+2023-03-10 22:35:24,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6311, 1.6809, 4.1865, 3.6293], device='cuda:0'), covar=tensor([0.1392, 0.2405, 0.0417, 0.0814], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0642, 0.0947, 0.0895], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 22:35:33,210 INFO [train.py:968] (0/2) Epoch 21, batch 19350, giga_loss[loss=0.3083, simple_loss=0.3525, pruned_loss=0.132, over 26577.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3442, pruned_loss=0.09765, over 5700317.07 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3357, pruned_loss=0.08796, over 5757288.24 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3457, pruned_loss=0.0993, over 5689936.22 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:36:14,446 INFO [train.py:968] (0/2) Epoch 21, batch 19400, giga_loss[loss=0.229, simple_loss=0.3035, pruned_loss=0.07731, over 28779.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3395, pruned_loss=0.09523, over 5693428.79 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3356, pruned_loss=0.0876, over 5761998.40 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3411, pruned_loss=0.09726, over 5678441.71 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:36:40,498 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=931777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:36:43,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=931780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:36:49,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.309e+02 1.045e+03 1.294e+03 2.040e+03 7.548e+03, threshold=2.588e+03, percent-clipped=10.0
+2023-03-10 22:36:54,963 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:36:56,844 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=931793.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:36:59,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=931796.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:37:02,773 INFO [train.py:968] (0/2) Epoch 21, batch 19450, giga_loss[loss=0.2295, simple_loss=0.2897, pruned_loss=0.08463, over 23537.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3341, pruned_loss=0.0926, over 5670105.29 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3357, pruned_loss=0.08754, over 5764430.94 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3352, pruned_loss=0.09435, over 5654639.01 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:37:10,681 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=931809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:37:25,144 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=931825.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:37:33,186 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:37:47,419 INFO [train.py:968] (0/2) Epoch 21, batch 19500, giga_loss[loss=0.252, simple_loss=0.3289, pruned_loss=0.08754, over 29078.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3333, pruned_loss=0.09178, over 5666631.72 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3356, pruned_loss=0.08753, over 5762594.20 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3343, pruned_loss=0.09326, over 5654359.64 frames. ], batch size: 113, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:38:09,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:38:11,873 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:38:15,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.237e+02 1.129e+03 1.445e+03 1.939e+03 7.715e+03, threshold=2.889e+03, percent-clipped=12.0
+2023-03-10 22:38:31,458 INFO [train.py:968] (0/2) Epoch 21, batch 19550, giga_loss[loss=0.2733, simple_loss=0.3378, pruned_loss=0.1044, over 23900.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.334, pruned_loss=0.09221, over 5666042.05 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3357, pruned_loss=0.08756, over 5764719.69 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3346, pruned_loss=0.09343, over 5653049.55 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:39:11,010 INFO [train.py:968] (0/2) Epoch 21, batch 19600, giga_loss[loss=0.3056, simple_loss=0.3666, pruned_loss=0.1224, over 27660.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3329, pruned_loss=0.09141, over 5673029.96 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3362, pruned_loss=0.08773, over 5757903.44 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3329, pruned_loss=0.09239, over 5665166.38 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:39:36,637 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-10 22:39:37,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.196e+02 1.064e+03 1.339e+03 1.723e+03 5.166e+03, threshold=2.678e+03, percent-clipped=4.0
+2023-03-10 22:39:44,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0015, 2.1756, 2.2096, 1.7706], device='cuda:0'), covar=tensor([0.1944, 0.2200, 0.1432, 0.1653], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0694, 0.0944, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 22:39:49,194 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-932000.pt
+2023-03-10 22:39:50,228 INFO [train.py:968] (0/2) Epoch 21, batch 19650, giga_loss[loss=0.2485, simple_loss=0.3177, pruned_loss=0.08964, over 28851.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3306, pruned_loss=0.09001, over 5685429.92 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3361, pruned_loss=0.08745, over 5760423.97 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3306, pruned_loss=0.0911, over 5675144.37 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:40:05,941 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:40:07,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:40:28,397 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:40:30,399 INFO [train.py:968] (0/2) Epoch 21, batch 19700, giga_loss[loss=0.223, simple_loss=0.2957, pruned_loss=0.07513, over 28602.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3287, pruned_loss=0.08897, over 5690995.89 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3364, pruned_loss=0.0875, over 5761596.30 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3284, pruned_loss=0.0898, over 5680943.92 frames. ], batch size: 85, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:40:30,693 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:40:54,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.268e+02 1.098e+03 1.354e+03 2.027e+03 5.763e+03, threshold=2.708e+03, percent-clipped=11.0
+2023-03-10 22:41:07,667 INFO [train.py:968] (0/2) Epoch 21, batch 19750, giga_loss[loss=0.2239, simple_loss=0.3075, pruned_loss=0.07017, over 28658.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3285, pruned_loss=0.0889, over 5682091.98 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3378, pruned_loss=0.08816, over 5744384.69 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3266, pruned_loss=0.089, over 5687190.69 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:41:47,080 INFO [train.py:968] (0/2) Epoch 21, batch 19800, giga_loss[loss=0.2223, simple_loss=0.3035, pruned_loss=0.07056, over 28908.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3256, pruned_loss=0.08776, over 5693570.83 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3382, pruned_loss=0.0883, over 5748228.80 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3235, pruned_loss=0.08774, over 5693024.46 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:41:47,422 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6641, 1.7281, 1.7394, 1.5155], device='cuda:0'), covar=tensor([0.3376, 0.2838, 0.2005, 0.2995], device='cuda:0'), in_proj_covar=tensor([0.1949, 0.1854, 0.1793, 0.1948], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 22:42:00,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:42:03,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5526, 1.8906, 1.5445, 1.6071], device='cuda:0'), covar=tensor([0.0787, 0.0300, 0.0323, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0116, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 22:42:15,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.455e+02 1.080e+03 1.287e+03 1.696e+03 6.607e+03, threshold=2.575e+03, percent-clipped=9.0
+2023-03-10 22:42:24,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6919, 1.9990, 1.4671, 1.4908], device='cuda:0'), covar=tensor([0.1028, 0.0587, 0.1014, 0.1160], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0445, 0.0515, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 22:42:25,952 INFO [train.py:968] (0/2) Epoch 21, batch 19850, giga_loss[loss=0.2145, simple_loss=0.2916, pruned_loss=0.06869, over 28812.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3226, pruned_loss=0.08637, over 5705172.95 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3386, pruned_loss=0.08828, over 5750852.02 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3202, pruned_loss=0.08634, over 5701354.49 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:42:33,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:43:06,160 INFO [train.py:968] (0/2) Epoch 21, batch 19900, giga_loss[loss=0.234, simple_loss=0.3058, pruned_loss=0.08108, over 28702.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3202, pruned_loss=0.085, over 5715633.14 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3389, pruned_loss=0.08836, over 5752130.56 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.318, pruned_loss=0.08489, over 5711172.24 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:43:08,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:43:34,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.025e+02 1.035e+03 1.222e+03 1.492e+03 4.414e+03, threshold=2.444e+03, percent-clipped=3.0
+2023-03-10 22:43:38,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:43:44,646 INFO [train.py:968] (0/2) Epoch 21, batch 19950, giga_loss[loss=0.2239, simple_loss=0.304, pruned_loss=0.07194, over 28996.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3186, pruned_loss=0.08451, over 5713373.48 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3393, pruned_loss=0.08847, over 5754279.40 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3162, pruned_loss=0.08427, over 5707414.34 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:43:50,435 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:43:53,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:44:16,414 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:44:23,810 INFO [train.py:968] (0/2) Epoch 21, batch 20000, giga_loss[loss=0.2144, simple_loss=0.294, pruned_loss=0.06736, over 28960.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3167, pruned_loss=0.08309, over 5717081.53 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3396, pruned_loss=0.08857, over 5755432.57 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.3143, pruned_loss=0.08276, over 5710930.75 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:44:24,632 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:44:26,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:44:47,024 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:44:49,869 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.650e+02 9.661e+02 1.175e+03 1.464e+03 4.163e+03, threshold=2.350e+03, percent-clipped=5.0
+2023-03-10 22:44:50,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4345, 2.4565, 2.4922, 2.0668], device='cuda:0'), covar=tensor([0.2817, 0.2245, 0.1781, 0.2494], device='cuda:0'), in_proj_covar=tensor([0.1941, 0.1849, 0.1792, 0.1938], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 22:44:58,809 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:45:01,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:45:02,058 INFO [train.py:968] (0/2) Epoch 21, batch 20050, giga_loss[loss=0.2344, simple_loss=0.3161, pruned_loss=0.0763, over 28851.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3176, pruned_loss=0.08328, over 5714955.86 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3404, pruned_loss=0.08889, over 5750363.85 frames. ], giga_tot_loss[loss=0.2401, simple_loss=0.3148, pruned_loss=0.08269, over 5713737.37 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:45:15,087 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:45:19,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:45:26,479 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:45:27,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6290, 1.7301, 1.8979, 1.4402], device='cuda:0'), covar=tensor([0.1731, 0.2310, 0.1359, 0.1650], device='cuda:0'), in_proj_covar=tensor([0.0903, 0.0698, 0.0948, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 22:45:44,917 INFO [train.py:968] (0/2) Epoch 21, batch 20100, giga_loss[loss=0.2829, simple_loss=0.3542, pruned_loss=0.1058, over 28835.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3243, pruned_loss=0.08792, over 5712538.31 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3407, pruned_loss=0.08897, over 5751145.03 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3218, pruned_loss=0.08738, over 5710774.18 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:46:06,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:46:20,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.234e+02 1.163e+03 1.370e+03 1.754e+03 4.755e+03, threshold=2.740e+03, percent-clipped=10.0
+2023-03-10 22:46:34,678 INFO [train.py:968] (0/2) Epoch 21, batch 20150, giga_loss[loss=0.2626, simple_loss=0.3352, pruned_loss=0.09496, over 28685.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3313, pruned_loss=0.09271, over 5697895.29 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3405, pruned_loss=0.08891, over 5752526.01 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3293, pruned_loss=0.09234, over 5694851.34 frames. ], batch size: 71, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:47:02,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2459, 1.6149, 1.2393, 0.7492], device='cuda:0'), covar=tensor([0.3800, 0.2014, 0.2477, 0.4647], device='cuda:0'), in_proj_covar=tensor([0.1737, 0.1635, 0.1596, 0.1414], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 22:47:08,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6172, 1.8885, 1.5314, 1.9178], device='cuda:0'), covar=tensor([0.2594, 0.2701, 0.2983, 0.2189], device='cuda:0'), in_proj_covar=tensor([0.1492, 0.1080, 0.1315, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 22:47:15,975 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3924, 1.9259, 1.5814, 1.5780], device='cuda:0'), covar=tensor([0.0787, 0.0289, 0.0310, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-10 22:47:17,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:47:23,985 INFO [train.py:968] (0/2) Epoch 21, batch 20200, giga_loss[loss=0.2725, simple_loss=0.3524, pruned_loss=0.09632, over 29000.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3395, pruned_loss=0.09798, over 5696719.52 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3411, pruned_loss=0.08927, over 5753108.00 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3374, pruned_loss=0.09744, over 5693416.68 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:47:38,317 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:47:40,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:47:48,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3360, 1.7448, 1.7394, 1.4798], device='cuda:0'), covar=tensor([0.1898, 0.1668, 0.2072, 0.1935], device='cuda:0'), in_proj_covar=tensor([0.0473, 0.0746, 0.0712, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 22:47:57,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.939e+02 1.359e+03 1.590e+03 2.170e+03 7.509e+03, threshold=3.179e+03, percent-clipped=14.0
+2023-03-10 22:48:05,625 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:48:07,655 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:48:10,308 INFO [train.py:968] (0/2) Epoch 21, batch 20250, giga_loss[loss=0.3324, simple_loss=0.4038, pruned_loss=0.1305, over 28907.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3458, pruned_loss=0.1009, over 5689197.46 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.341, pruned_loss=0.08921, over 5755567.66 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3443, pruned_loss=0.1007, over 5683446.48 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:48:51,401 INFO [train.py:968] (0/2) Epoch 21, batch 20300, giga_loss[loss=0.3348, simple_loss=0.4047, pruned_loss=0.1325, over 27908.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3515, pruned_loss=0.103, over 5693873.47 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3416, pruned_loss=0.0893, over 5760115.94 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3501, pruned_loss=0.1032, over 5682684.53 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:49:07,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:49:15,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932677.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:49:22,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.433e+02 1.206e+03 1.535e+03 1.945e+03 3.779e+03, threshold=3.070e+03, percent-clipped=6.0
+2023-03-10 22:49:33,818 INFO [train.py:968] (0/2) Epoch 21, batch 20350, libri_loss[loss=0.2391, simple_loss=0.3155, pruned_loss=0.08137, over 29638.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.356, pruned_loss=0.1054, over 5703843.91 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3413, pruned_loss=0.08932, over 5765663.85 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3556, pruned_loss=0.1062, over 5687042.44 frames. ], batch size: 73, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:49:42,336 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932713.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:50:11,005 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:50:15,414 INFO [train.py:968] (0/2) Epoch 21, batch 20400, giga_loss[loss=0.2229, simple_loss=0.3113, pruned_loss=0.06722, over 28969.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3546, pruned_loss=0.1041, over 5694754.10 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3413, pruned_loss=0.08929, over 5765202.16 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3545, pruned_loss=0.105, over 5681307.50 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:50:19,115 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2541, 1.2602, 1.1851, 1.3726], device='cuda:0'), covar=tensor([0.0590, 0.0301, 0.0269, 0.0626], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0116, 0.0116, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 22:50:31,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1490, 3.9684, 3.7287, 2.0429], device='cuda:0'), covar=tensor([0.0561, 0.0670, 0.0668, 0.1937], device='cuda:0'), in_proj_covar=tensor([0.1190, 0.1103, 0.0934, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 22:50:46,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.318e+02 1.348e+03 1.737e+03 2.440e+03 5.503e+03, threshold=3.474e+03, percent-clipped=16.0
+2023-03-10 22:50:49,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:50:56,704 INFO [train.py:968] (0/2) Epoch 21, batch 20450, giga_loss[loss=0.2498, simple_loss=0.3304, pruned_loss=0.08458, over 28766.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3508, pruned_loss=0.101, over 5687943.10 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3417, pruned_loss=0.0897, over 5746691.27 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3506, pruned_loss=0.1015, over 5691533.78 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:50:59,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1571, 1.3281, 3.4783, 3.1457], device='cuda:0'), covar=tensor([0.1680, 0.2834, 0.0458, 0.1648], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0642, 0.0947, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 22:51:04,892 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:51:07,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:51:30,834 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:51:32,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:51:37,147 INFO [train.py:968] (0/2) Epoch 21, batch 20500, giga_loss[loss=0.2921, simple_loss=0.3523, pruned_loss=0.116, over 23302.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3496, pruned_loss=0.09996, over 5686330.82 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3417, pruned_loss=0.08969, over 5748447.51 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3495, pruned_loss=0.1005, over 5687096.57 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:51:41,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2340, 1.1802, 3.9016, 3.4094], device='cuda:0'), covar=tensor([0.2170, 0.3290, 0.0770, 0.0921], device='cuda:0'), in_proj_covar=tensor([0.0750, 0.0642, 0.0947, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 22:52:08,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.540e+02 1.206e+03 1.488e+03 1.984e+03 5.900e+03, threshold=2.977e+03, percent-clipped=5.0
+2023-03-10 22:52:17,591 INFO [train.py:968] (0/2) Epoch 21, batch 20550, giga_loss[loss=0.2648, simple_loss=0.3419, pruned_loss=0.09381, over 28413.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3491, pruned_loss=0.09896, over 5688820.62 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3417, pruned_loss=0.08982, over 5742261.69 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3492, pruned_loss=0.09956, over 5693532.94 frames. ], batch size: 65, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:52:32,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:52:44,481 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:52:45,410 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:52:46,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:52:58,612 INFO [train.py:968] (0/2) Epoch 21, batch 20600, giga_loss[loss=0.2699, simple_loss=0.3467, pruned_loss=0.09654, over 28634.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3507, pruned_loss=0.1002, over 5690705.27 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3416, pruned_loss=0.08984, over 5746939.11 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3512, pruned_loss=0.1009, over 5688904.30 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:53:13,053 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:53:15,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:53:32,590 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:53:33,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.644e+02 1.315e+03 1.808e+03 2.490e+03 4.982e+03, threshold=3.615e+03, percent-clipped=10.0
+2023-03-10 22:53:34,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:53:43,196 INFO [train.py:968] (0/2) Epoch 21, batch 20650, giga_loss[loss=0.3083, simple_loss=0.3735, pruned_loss=0.1216, over 28601.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3536, pruned_loss=0.1024, over 5688336.91 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3414, pruned_loss=0.08972, over 5749684.75 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3543, pruned_loss=0.1033, over 5683549.77 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:53:58,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:54:23,896 INFO [train.py:968] (0/2) Epoch 21, batch 20700, giga_loss[loss=0.3218, simple_loss=0.3788, pruned_loss=0.1324, over 26712.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3529, pruned_loss=0.1017, over 5703794.70 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3413, pruned_loss=0.08957, over 5753256.73 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.354, pruned_loss=0.1029, over 5694984.10 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:54:24,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933052.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:54:33,492 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:54:35,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:54:36,844 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:54:54,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933088.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:54:54,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.814e+02 1.385e+03 1.913e+03 2.707e+03 6.816e+03, threshold=3.826e+03, percent-clipped=6.0
+2023-03-10 22:55:00,001 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:05,389 INFO [train.py:968] (0/2) Epoch 21, batch 20750, giga_loss[loss=0.3302, simple_loss=0.3797, pruned_loss=0.1403, over 23967.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3548, pruned_loss=0.1032, over 5700859.46 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3413, pruned_loss=0.08956, over 5744552.74 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3558, pruned_loss=0.1044, over 5700586.85 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:55:17,800 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:19,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:21,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:35,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4544, 1.5151, 1.6751, 1.2985], device='cuda:0'), covar=tensor([0.1482, 0.2257, 0.1226, 0.1545], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0697, 0.0943, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 22:55:40,957 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:44,094 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:45,186 INFO [train.py:968] (0/2) Epoch 21, batch 20800, giga_loss[loss=0.3082, simple_loss=0.371, pruned_loss=0.1226, over 28779.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3564, pruned_loss=0.1043, over 5705966.80 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3424, pruned_loss=0.09001, over 5747904.85 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3567, pruned_loss=0.1054, over 5701158.41 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:56:14,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.605e+02 1.292e+03 1.632e+03 2.257e+03 8.848e+03, threshold=3.264e+03, percent-clipped=7.0
+2023-03-10 22:56:18,936 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933195.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:56:20,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933198.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:56:22,734 INFO [train.py:968] (0/2) Epoch 21, batch 20850, giga_loss[loss=0.2779, simple_loss=0.3576, pruned_loss=0.09911, over 28899.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3555, pruned_loss=0.1032, over 5708559.78 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3426, pruned_loss=0.09003, over 5748229.78 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3558, pruned_loss=0.1042, over 5703916.45 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:56:44,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933227.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:56:46,932 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933231.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:56:49,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933234.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:57:02,894 INFO [train.py:968] (0/2) Epoch 21, batch 20900, libri_loss[loss=0.3421, simple_loss=0.4075, pruned_loss=0.1383, over 20229.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3557, pruned_loss=0.1023, over 5693412.32 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3431, pruned_loss=0.09032, over 5733247.56 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3556, pruned_loss=0.103, over 5702463.94 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:57:12,318 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:57:12,939 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933263.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:57:14,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:57:15,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4056, 1.2093, 4.2224, 3.3903], device='cuda:0'), covar=tensor([0.1518, 0.2747, 0.0424, 0.0999], device='cuda:0'), in_proj_covar=tensor([0.0748, 0.0639, 0.0944, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 22:57:30,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.702e+02 1.118e+03 1.436e+03 1.877e+03 5.927e+03, threshold=2.872e+03, percent-clipped=9.0
+2023-03-10 22:57:34,509 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:57:39,794 INFO [train.py:968] (0/2) Epoch 21, batch 20950, giga_loss[loss=0.2629, simple_loss=0.3425, pruned_loss=0.09164, over 28246.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3558, pruned_loss=0.1015, over 5706251.33 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3426, pruned_loss=0.08996, over 5737772.27 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3567, pruned_loss=0.1029, over 5708377.91 frames. ], batch size: 77, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:57:49,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:58:19,541 INFO [train.py:968] (0/2) Epoch 21, batch 21000, giga_loss[loss=0.2857, simple_loss=0.3672, pruned_loss=0.1021, over 28716.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3544, pruned_loss=0.1011, over 5710668.90 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3424, pruned_loss=0.08972, over 5741045.10 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3556, pruned_loss=0.1026, over 5708882.11 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:58:19,545 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 22:58:28,139 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2087, simple_loss=0.3155, pruned_loss=0.05099, over 944034.00 frames. 
+2023-03-10 22:58:28,140 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 22:58:41,392 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:58:44,869 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:58:47,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9390, 1.2538, 1.3117, 1.0953], device='cuda:0'), covar=tensor([0.2006, 0.1431, 0.2383, 0.1722], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0743, 0.0711, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 22:58:52,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9951, 1.1597, 1.1476, 0.9080], device='cuda:0'), covar=tensor([0.2131, 0.2545, 0.1514, 0.2134], device='cuda:0'), in_proj_covar=tensor([0.1961, 0.1870, 0.1816, 0.1959], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 22:58:55,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.417e+02 1.106e+03 1.447e+03 1.848e+03 5.803e+03, threshold=2.893e+03, percent-clipped=6.0
+2023-03-10 22:59:05,097 INFO [train.py:968] (0/2) Epoch 21, batch 21050, giga_loss[loss=0.258, simple_loss=0.3332, pruned_loss=0.09143, over 28637.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3518, pruned_loss=0.1001, over 5711923.34 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3425, pruned_loss=0.08976, over 5742328.17 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3527, pruned_loss=0.1014, over 5708963.31 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:59:35,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:59:41,683 INFO [train.py:968] (0/2) Epoch 21, batch 21100, giga_loss[loss=0.2898, simple_loss=0.3586, pruned_loss=0.1106, over 28855.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3494, pruned_loss=0.09872, over 5710349.91 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3426, pruned_loss=0.08998, over 5741997.40 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3503, pruned_loss=0.09984, over 5707204.69 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:59:44,034 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:59:46,172 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:00:09,193 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:00:11,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.576e+02 1.076e+03 1.290e+03 1.742e+03 5.325e+03, threshold=2.581e+03, percent-clipped=5.0
+2023-03-10 23:00:22,407 INFO [train.py:968] (0/2) Epoch 21, batch 21150, giga_loss[loss=0.2724, simple_loss=0.3532, pruned_loss=0.09584, over 28898.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3492, pruned_loss=0.09915, over 5713677.18 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3425, pruned_loss=0.08988, over 5744431.16 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3502, pruned_loss=0.1003, over 5708594.52 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:00:39,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:00:43,281 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:00:45,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2972, 3.1196, 2.9440, 1.4676], device='cuda:0'), covar=tensor([0.0904, 0.1054, 0.0959, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.1181, 0.1096, 0.0930, 0.0712], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 23:01:02,803 INFO [train.py:968] (0/2) Epoch 21, batch 21200, giga_loss[loss=0.2907, simple_loss=0.3661, pruned_loss=0.1077, over 29010.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3498, pruned_loss=0.1001, over 5693211.98 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3425, pruned_loss=0.08999, over 5730462.91 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3508, pruned_loss=0.1011, over 5701949.89 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:01:27,775 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:01:30,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:01:31,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.222e+02 1.145e+03 1.443e+03 1.965e+03 8.086e+03, threshold=2.887e+03, percent-clipped=11.0
+2023-03-10 23:01:40,832 INFO [train.py:968] (0/2) Epoch 21, batch 21250, giga_loss[loss=0.2413, simple_loss=0.3273, pruned_loss=0.07768, over 28968.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3487, pruned_loss=0.09854, over 5705319.20 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3427, pruned_loss=0.09019, over 5731925.15 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3493, pruned_loss=0.09927, over 5710654.33 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:01:55,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:02:20,311 INFO [train.py:968] (0/2) Epoch 21, batch 21300, giga_loss[loss=0.2609, simple_loss=0.3441, pruned_loss=0.08884, over 28602.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3473, pruned_loss=0.09658, over 5707941.24 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3429, pruned_loss=0.09029, over 5735249.07 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3476, pruned_loss=0.09721, over 5708649.56 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:02:37,013 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:02:38,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:02:40,233 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4085, 1.6441, 1.6648, 1.1963], device='cuda:0'), covar=tensor([0.1680, 0.2882, 0.1537, 0.1924], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0701, 0.0946, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 23:02:52,243 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.632e+02 1.103e+03 1.430e+03 1.729e+03 3.987e+03, threshold=2.860e+03, percent-clipped=2.0
+2023-03-10 23:02:59,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:03:00,899 INFO [train.py:968] (0/2) Epoch 21, batch 21350, giga_loss[loss=0.2859, simple_loss=0.36, pruned_loss=0.1059, over 28980.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3477, pruned_loss=0.0976, over 5701124.95 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3432, pruned_loss=0.09042, over 5737648.64 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3477, pruned_loss=0.09808, over 5699048.68 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:03:23,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3725, 5.1806, 4.9218, 2.4856], device='cuda:0'), covar=tensor([0.0412, 0.0573, 0.0666, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.1182, 0.1099, 0.0931, 0.0712], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 23:03:32,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4018, 2.6954, 2.0664, 2.0022], device='cuda:0'), covar=tensor([0.0918, 0.0639, 0.0915, 0.1093], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0442, 0.0513, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 23:03:33,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:03:36,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:03:37,615 INFO [train.py:968] (0/2) Epoch 21, batch 21400, giga_loss[loss=0.2671, simple_loss=0.3456, pruned_loss=0.09432, over 28637.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.346, pruned_loss=0.09737, over 5695565.46 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3435, pruned_loss=0.09079, over 5732424.72 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3459, pruned_loss=0.09759, over 5697487.99 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:04:06,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.026e+02 1.144e+03 1.400e+03 2.124e+03 6.329e+03, threshold=2.800e+03, percent-clipped=13.0
+2023-03-10 23:04:15,078 INFO [train.py:968] (0/2) Epoch 21, batch 21450, giga_loss[loss=0.2465, simple_loss=0.3242, pruned_loss=0.08442, over 29071.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3425, pruned_loss=0.09528, over 5698486.76 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3432, pruned_loss=0.09053, over 5730225.90 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3427, pruned_loss=0.09583, over 5701217.44 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:04:53,983 INFO [train.py:968] (0/2) Epoch 21, batch 21500, giga_loss[loss=0.294, simple_loss=0.3568, pruned_loss=0.1156, over 23580.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3405, pruned_loss=0.09464, over 5688129.57 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3433, pruned_loss=0.09069, over 5733630.80 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3405, pruned_loss=0.095, over 5686363.80 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:04:55,775 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-10 23:05:01,542 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:12,382 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-10 23:05:21,111 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:22,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.460e+02 1.218e+03 1.659e+03 2.225e+03 6.115e+03, threshold=3.317e+03, percent-clipped=13.0
+2023-03-10 23:05:22,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:24,302 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:26,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:29,900 INFO [train.py:968] (0/2) Epoch 21, batch 21550, giga_loss[loss=0.2519, simple_loss=0.3292, pruned_loss=0.08736, over 28932.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3407, pruned_loss=0.09487, over 5696184.36 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.343, pruned_loss=0.0904, over 5737801.98 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3409, pruned_loss=0.09553, over 5689803.36 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:05:32,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:34,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6725, 1.8932, 1.3890, 1.3315], device='cuda:0'), covar=tensor([0.1015, 0.0642, 0.1074, 0.1200], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0442, 0.0513, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 23:05:43,479 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:46,680 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:06:06,762 INFO [train.py:968] (0/2) Epoch 21, batch 21600, giga_loss[loss=0.2365, simple_loss=0.3151, pruned_loss=0.07897, over 28966.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.341, pruned_loss=0.09537, over 5705441.71 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.344, pruned_loss=0.09111, over 5746547.46 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.34, pruned_loss=0.09554, over 5690372.05 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:06:15,050 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-10 23:06:38,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.364e+02 1.209e+03 1.527e+03 2.222e+03 6.250e+03, threshold=3.054e+03, percent-clipped=2.0
+2023-03-10 23:06:46,586 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-934000.pt
+2023-03-10 23:06:47,535 INFO [train.py:968] (0/2) Epoch 21, batch 21650, giga_loss[loss=0.2451, simple_loss=0.3186, pruned_loss=0.0858, over 28915.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3379, pruned_loss=0.09402, over 5712904.03 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3443, pruned_loss=0.09133, over 5747895.46 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3368, pruned_loss=0.09402, over 5699145.79 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:07:15,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2044, 1.5115, 1.4755, 1.0481], device='cuda:0'), covar=tensor([0.1689, 0.2722, 0.1483, 0.1748], device='cuda:0'), in_proj_covar=tensor([0.0895, 0.0697, 0.0942, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 23:07:23,118 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=934047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:07:26,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=934050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:07:26,467 INFO [train.py:968] (0/2) Epoch 21, batch 21700, giga_loss[loss=0.2896, simple_loss=0.3575, pruned_loss=0.1109, over 28286.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3349, pruned_loss=0.09256, over 5712150.52 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3445, pruned_loss=0.09148, over 5749205.07 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3338, pruned_loss=0.09244, over 5699824.54 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:07:50,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=934079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:07:55,070 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 23:07:58,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.181e+02 1.058e+03 1.311e+03 1.834e+03 5.485e+03, threshold=2.623e+03, percent-clipped=4.0
+2023-03-10 23:08:06,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9057, 2.8425, 1.9902, 0.9896], device='cuda:0'), covar=tensor([0.8401, 0.3637, 0.3948, 0.7322], device='cuda:0'), in_proj_covar=tensor([0.1727, 0.1623, 0.1588, 0.1408], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 23:08:07,547 INFO [train.py:968] (0/2) Epoch 21, batch 21750, giga_loss[loss=0.2622, simple_loss=0.3339, pruned_loss=0.09522, over 28954.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3326, pruned_loss=0.09152, over 5719021.71 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3444, pruned_loss=0.09153, over 5751402.59 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3316, pruned_loss=0.09141, over 5706690.15 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:08:46,400 INFO [train.py:968] (0/2) Epoch 21, batch 21800, giga_loss[loss=0.2293, simple_loss=0.3062, pruned_loss=0.07617, over 28313.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3332, pruned_loss=0.09219, over 5721004.93 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3447, pruned_loss=0.09198, over 5755279.16 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3319, pruned_loss=0.09173, over 5706924.04 frames. ], batch size: 65, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:09:00,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1004, 1.9750, 4.5143, 3.9510], device='cuda:0'), covar=tensor([0.1145, 0.2266, 0.0374, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0751, 0.0641, 0.0948, 0.0901], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 23:09:19,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.942e+02 1.089e+03 1.444e+03 1.875e+03 5.360e+03, threshold=2.888e+03, percent-clipped=15.0
+2023-03-10 23:09:27,862 INFO [train.py:968] (0/2) Epoch 21, batch 21850, giga_loss[loss=0.2449, simple_loss=0.3129, pruned_loss=0.08851, over 28917.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3357, pruned_loss=0.09283, over 5700051.97 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3448, pruned_loss=0.0921, over 5739321.39 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3343, pruned_loss=0.09236, over 5702210.58 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:09:50,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9066, 1.1708, 1.3782, 1.0121], device='cuda:0'), covar=tensor([0.1886, 0.1452, 0.2278, 0.1797], device='cuda:0'), in_proj_covar=tensor([0.0470, 0.0739, 0.0708, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-10 23:09:51,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3856, 1.6176, 1.5563, 1.3032], device='cuda:0'), covar=tensor([0.3201, 0.2587, 0.2046, 0.2663], device='cuda:0'), in_proj_covar=tensor([0.1965, 0.1876, 0.1820, 0.1958], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-10 23:10:01,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=934237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:10:03,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5911, 1.7991, 1.2488, 1.3878], device='cuda:0'), covar=tensor([0.0955, 0.0649, 0.1047, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0443, 0.0513, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 23:10:11,928 INFO [train.py:968] (0/2) Epoch 21, batch 21900, giga_loss[loss=0.2395, simple_loss=0.3309, pruned_loss=0.07408, over 28779.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3396, pruned_loss=0.09466, over 5695413.46 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3449, pruned_loss=0.09219, over 5742168.50 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3382, pruned_loss=0.09423, over 5693649.76 frames. ], batch size: 243, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:10:46,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.590e+02 1.102e+03 1.300e+03 1.635e+03 4.431e+03, threshold=2.601e+03, percent-clipped=3.0
+2023-03-10 23:10:55,186 INFO [train.py:968] (0/2) Epoch 21, batch 21950, giga_loss[loss=0.2774, simple_loss=0.3579, pruned_loss=0.09849, over 28844.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3416, pruned_loss=0.09494, over 5692513.82 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.345, pruned_loss=0.09231, over 5734193.30 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3405, pruned_loss=0.09454, over 5697048.75 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:11:15,585 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 23:11:36,957 INFO [train.py:968] (0/2) Epoch 21, batch 22000, giga_loss[loss=0.2315, simple_loss=0.3158, pruned_loss=0.07357, over 28887.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.341, pruned_loss=0.09376, over 5700234.68 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3456, pruned_loss=0.09272, over 5737451.52 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3396, pruned_loss=0.09311, over 5700178.91 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:12:03,645 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=934380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:12:05,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=934383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:12:12,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.257e+02 1.147e+03 1.433e+03 2.243e+03 8.252e+03, threshold=2.867e+03, percent-clipped=19.0
+2023-03-10 23:12:19,835 INFO [train.py:968] (0/2) Epoch 21, batch 22050, giga_loss[loss=0.2682, simple_loss=0.3411, pruned_loss=0.09761, over 28976.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3399, pruned_loss=0.09318, over 5691984.29 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3455, pruned_loss=0.09287, over 5733195.21 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3386, pruned_loss=0.09251, over 5695062.13 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:12:28,311 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=934412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:12:29,752 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3290, 1.9338, 1.4911, 1.6130], device='cuda:0'), covar=tensor([0.0740, 0.0272, 0.0313, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0116, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 23:13:00,205 INFO [train.py:968] (0/2) Epoch 21, batch 22100, giga_loss[loss=0.315, simple_loss=0.3854, pruned_loss=0.1224, over 27645.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3419, pruned_loss=0.0945, over 5695466.91 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.346, pruned_loss=0.0932, over 5738039.69 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3403, pruned_loss=0.09369, over 5692667.72 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:13:30,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.160e+02 1.349e+03 1.744e+03 2.318e+03 5.291e+03, threshold=3.487e+03, percent-clipped=10.0
+2023-03-10 23:13:38,136 INFO [train.py:968] (0/2) Epoch 21, batch 22150, giga_loss[loss=0.2375, simple_loss=0.325, pruned_loss=0.07503, over 28996.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3423, pruned_loss=0.09527, over 5695844.58 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3465, pruned_loss=0.09385, over 5733679.90 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3404, pruned_loss=0.0941, over 5696188.47 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:14:18,633 INFO [train.py:968] (0/2) Epoch 21, batch 22200, giga_loss[loss=0.2734, simple_loss=0.3552, pruned_loss=0.0958, over 29013.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3447, pruned_loss=0.09655, over 5698827.47 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3468, pruned_loss=0.09406, over 5733518.41 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3429, pruned_loss=0.09544, over 5698961.48 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:14:24,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4809, 3.1936, 1.5425, 1.5137], device='cuda:0'), covar=tensor([0.0923, 0.0343, 0.0921, 0.1292], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0547, 0.0381, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 23:14:52,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.412e+02 1.320e+03 1.636e+03 2.294e+03 6.874e+03, threshold=3.273e+03, percent-clipped=7.0
+2023-03-10 23:14:59,059 INFO [train.py:968] (0/2) Epoch 21, batch 22250, giga_loss[loss=0.2766, simple_loss=0.3593, pruned_loss=0.09693, over 29000.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3497, pruned_loss=0.09947, over 5707724.76 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3477, pruned_loss=0.09471, over 5734834.12 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3475, pruned_loss=0.09807, over 5706040.09 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:15:16,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8749, 1.8570, 1.9848, 1.5662], device='cuda:0'), covar=tensor([0.1932, 0.2419, 0.1559, 0.1746], device='cuda:0'), in_proj_covar=tensor([0.0895, 0.0694, 0.0941, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 23:15:20,782 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3440, 1.6980, 1.3611, 0.9630], device='cuda:0'), covar=tensor([0.2484, 0.2613, 0.2912, 0.2399], device='cuda:0'), in_proj_covar=tensor([0.1491, 0.1080, 0.1314, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-10 23:15:38,470 INFO [train.py:968] (0/2) Epoch 21, batch 22300, giga_loss[loss=0.2535, simple_loss=0.3404, pruned_loss=0.0833, over 28823.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3525, pruned_loss=0.1011, over 5711513.55 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3485, pruned_loss=0.09533, over 5737080.43 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.0996, over 5707280.63 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:16:11,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.490e+02 1.452e+03 1.898e+03 2.582e+03 6.041e+03, threshold=3.797e+03, percent-clipped=11.0
+2023-03-10 23:16:16,019 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:16:18,569 INFO [train.py:968] (0/2) Epoch 21, batch 22350, giga_loss[loss=0.2468, simple_loss=0.3274, pruned_loss=0.08311, over 28742.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3534, pruned_loss=0.1016, over 5718312.53 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3487, pruned_loss=0.09553, over 5739455.15 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3513, pruned_loss=0.1003, over 5712605.10 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:16:37,316 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5020, 3.8827, 1.7625, 1.5833], device='cuda:0'), covar=tensor([0.0920, 0.0302, 0.0842, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0547, 0.0382, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 23:17:00,612 INFO [train.py:968] (0/2) Epoch 21, batch 22400, libri_loss[loss=0.2661, simple_loss=0.3485, pruned_loss=0.09184, over 29539.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3531, pruned_loss=0.1013, over 5719755.70 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3492, pruned_loss=0.09583, over 5744757.48 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3513, pruned_loss=0.1002, over 5709557.19 frames. ], batch size: 82, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:17:09,203 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 23:17:24,922 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:17:36,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.779e+02 1.225e+03 1.657e+03 2.190e+03 7.202e+03, threshold=3.315e+03, percent-clipped=6.0
+2023-03-10 23:17:38,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3258, 1.5682, 1.3570, 1.5276], device='cuda:0'), covar=tensor([0.0768, 0.0312, 0.0345, 0.0876], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 23:17:38,812 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:17:44,489 INFO [train.py:968] (0/2) Epoch 21, batch 22450, giga_loss[loss=0.2592, simple_loss=0.3426, pruned_loss=0.08788, over 29001.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3523, pruned_loss=0.1009, over 5718539.18 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3499, pruned_loss=0.09651, over 5743192.39 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3501, pruned_loss=0.0995, over 5711585.01 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:18:12,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8319, 4.6825, 4.4825, 2.0445], device='cuda:0'), covar=tensor([0.0643, 0.0779, 0.0919, 0.1945], device='cuda:0'), in_proj_covar=tensor([0.1197, 0.1106, 0.0942, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-10 23:18:13,649 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 23:18:23,162 INFO [train.py:968] (0/2) Epoch 21, batch 22500, giga_loss[loss=0.276, simple_loss=0.3529, pruned_loss=0.09951, over 28562.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3494, pruned_loss=0.0992, over 5724418.87 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.35, pruned_loss=0.09676, over 5747373.90 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3477, pruned_loss=0.09797, over 5714359.65 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:18:57,702 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1981, 1.5223, 0.9552, 1.0797], device='cuda:0'), covar=tensor([0.1245, 0.0786, 0.1700, 0.1430], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0446, 0.0515, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 23:18:58,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.478e+02 1.251e+03 1.493e+03 2.144e+03 6.289e+03, threshold=2.985e+03, percent-clipped=8.0
+2023-03-10 23:19:03,758 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:19:04,208 INFO [train.py:968] (0/2) Epoch 21, batch 22550, giga_loss[loss=0.2984, simple_loss=0.3654, pruned_loss=0.1157, over 28598.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3473, pruned_loss=0.09828, over 5721301.80 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3508, pruned_loss=0.09737, over 5741178.41 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3452, pruned_loss=0.09682, over 5717481.74 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:19:20,355 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:19:42,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6099, 4.9722, 1.8102, 1.9248], device='cuda:0'), covar=tensor([0.0966, 0.0278, 0.0904, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0549, 0.0382, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 23:19:46,177 INFO [train.py:968] (0/2) Epoch 21, batch 22600, giga_loss[loss=0.3274, simple_loss=0.3961, pruned_loss=0.1293, over 28377.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3444, pruned_loss=0.09686, over 5717659.50 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.351, pruned_loss=0.09754, over 5741996.52 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3426, pruned_loss=0.09558, over 5713923.66 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:19:52,623 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:19:55,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0628, 1.2657, 1.1132, 0.8933], device='cuda:0'), covar=tensor([0.1008, 0.0541, 0.1147, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0446, 0.0515, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 23:20:19,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.502e+02 1.187e+03 1.534e+03 1.922e+03 4.188e+03, threshold=3.068e+03, percent-clipped=7.0
+2023-03-10 23:20:25,266 INFO [train.py:968] (0/2) Epoch 21, batch 22650, giga_loss[loss=0.2805, simple_loss=0.3653, pruned_loss=0.09787, over 28313.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3453, pruned_loss=0.09638, over 5717689.24 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3516, pruned_loss=0.0981, over 5741773.21 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3431, pruned_loss=0.09478, over 5713868.18 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:21:08,585 INFO [train.py:968] (0/2) Epoch 21, batch 22700, giga_loss[loss=0.2811, simple_loss=0.3578, pruned_loss=0.1022, over 28973.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3468, pruned_loss=0.09642, over 5713582.92 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3517, pruned_loss=0.09825, over 5742623.35 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3449, pruned_loss=0.09502, over 5709682.79 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:21:24,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:21:40,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.588e+02 1.097e+03 1.458e+03 1.937e+03 5.390e+03, threshold=2.915e+03, percent-clipped=7.0
+2023-03-10 23:21:46,591 INFO [train.py:968] (0/2) Epoch 21, batch 22750, giga_loss[loss=0.263, simple_loss=0.3356, pruned_loss=0.09514, over 28662.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3458, pruned_loss=0.09622, over 5726312.37 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3518, pruned_loss=0.09841, over 5745662.46 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3442, pruned_loss=0.09495, over 5720151.20 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:22:06,705 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 23:22:27,579 INFO [train.py:968] (0/2) Epoch 21, batch 22800, giga_loss[loss=0.2738, simple_loss=0.3311, pruned_loss=0.1082, over 28719.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3445, pruned_loss=0.09655, over 5717197.98 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3517, pruned_loss=0.09853, over 5737916.34 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3431, pruned_loss=0.09538, over 5718448.94 frames. ], batch size: 92, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:22:27,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:22:44,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:23:00,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.607e+02 1.267e+03 1.592e+03 2.148e+03 4.931e+03, threshold=3.183e+03, percent-clipped=8.0
+2023-03-10 23:23:07,779 INFO [train.py:968] (0/2) Epoch 21, batch 22850, giga_loss[loss=0.2821, simple_loss=0.3527, pruned_loss=0.1058, over 28966.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3431, pruned_loss=0.09739, over 5717042.75 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3518, pruned_loss=0.09875, over 5740807.40 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3417, pruned_loss=0.09622, over 5714946.96 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:23:13,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0887, 3.0422, 1.1966, 1.3933], device='cuda:0'), covar=tensor([0.1118, 0.0512, 0.1084, 0.1448], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0548, 0.0382, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 23:23:20,020 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:23:22,079 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:23:46,447 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=935248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:23:48,037 INFO [train.py:968] (0/2) Epoch 21, batch 22900, giga_loss[loss=0.2565, simple_loss=0.3342, pruned_loss=0.08934, over 29029.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.342, pruned_loss=0.09811, over 5722243.49 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3519, pruned_loss=0.09907, over 5745129.35 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3405, pruned_loss=0.09684, over 5716007.06 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:24:07,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:22,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.593e+02 1.198e+03 1.488e+03 1.987e+03 4.702e+03, threshold=2.976e+03, percent-clipped=5.0
+2023-03-10 23:24:22,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:23,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:24,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:27,904 INFO [train.py:968] (0/2) Epoch 21, batch 22950, giga_loss[loss=0.2356, simple_loss=0.3059, pruned_loss=0.0826, over 28501.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3406, pruned_loss=0.09768, over 5724682.04 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3522, pruned_loss=0.09936, over 5747828.69 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.339, pruned_loss=0.09639, over 5717058.52 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:24:36,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:40,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:47,283 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=935326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:54,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:25:02,492 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=935345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:25:06,348 INFO [train.py:968] (0/2) Epoch 21, batch 23000, giga_loss[loss=0.2176, simple_loss=0.2974, pruned_loss=0.06891, over 28892.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3377, pruned_loss=0.0964, over 5708515.45 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3526, pruned_loss=0.09984, over 5732493.86 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3358, pruned_loss=0.09489, over 5715194.93 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:25:21,155 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1642, 1.7192, 1.3637, 1.4563], device='cuda:0'), covar=tensor([0.2487, 0.2004, 0.2674, 0.2271], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0746, 0.0714, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 23:25:38,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.557e+02 1.249e+03 1.663e+03 2.311e+03 6.011e+03, threshold=3.326e+03, percent-clipped=14.0
+2023-03-10 23:25:44,938 INFO [train.py:968] (0/2) Epoch 21, batch 23050, giga_loss[loss=0.2624, simple_loss=0.334, pruned_loss=0.09535, over 27606.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3341, pruned_loss=0.0948, over 5711827.03 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3529, pruned_loss=0.1001, over 5735104.59 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3321, pruned_loss=0.09331, over 5714450.60 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:25:57,495 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:00,531 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:14,180 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:16,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:21,723 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=935450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:22,118 INFO [train.py:968] (0/2) Epoch 21, batch 23100, giga_loss[loss=0.2194, simple_loss=0.2955, pruned_loss=0.07164, over 28980.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3298, pruned_loss=0.09233, over 5721469.31 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3529, pruned_loss=0.1003, over 5740607.36 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3276, pruned_loss=0.09078, over 5718218.75 frames. ], batch size: 128, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:26:38,237 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=935471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:44,676 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:46,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:57,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.047e+02 1.351e+03 1.759e+03 2.353e+03 5.756e+03, threshold=3.517e+03, percent-clipped=10.0
+2023-03-10 23:27:02,442 INFO [train.py:968] (0/2) Epoch 21, batch 23150, giga_loss[loss=0.2626, simple_loss=0.3387, pruned_loss=0.09325, over 28752.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3314, pruned_loss=0.09299, over 5715270.30 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3531, pruned_loss=0.1008, over 5741504.63 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.329, pruned_loss=0.09116, over 5711475.59 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:27:10,867 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=935510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:27:12,622 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-10 23:27:33,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3845, 3.0009, 1.4693, 1.5064], device='cuda:0'), covar=tensor([0.0946, 0.0366, 0.0929, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0549, 0.0382, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 23:27:43,462 INFO [train.py:968] (0/2) Epoch 21, batch 23200, libri_loss[loss=0.2532, simple_loss=0.3201, pruned_loss=0.09315, over 28229.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3346, pruned_loss=0.09446, over 5716919.20 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3528, pruned_loss=0.1008, over 5744097.93 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3325, pruned_loss=0.09282, over 5711093.49 frames. ], batch size: 62, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:28:04,495 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=935575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:28:10,004 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1711, 1.2452, 1.1062, 0.8983], device='cuda:0'), covar=tensor([0.0937, 0.0557, 0.1109, 0.1103], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0443, 0.0511, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 23:28:19,970 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.488e+02 1.204e+03 1.393e+03 1.839e+03 4.636e+03, threshold=2.785e+03, percent-clipped=2.0
+2023-03-10 23:28:25,979 INFO [train.py:968] (0/2) Epoch 21, batch 23250, giga_loss[loss=0.3011, simple_loss=0.3847, pruned_loss=0.1087, over 28725.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3385, pruned_loss=0.09577, over 5714306.51 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3528, pruned_loss=0.1008, over 5744097.93 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3369, pruned_loss=0.09449, over 5709772.29 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:29:05,165 INFO [train.py:968] (0/2) Epoch 21, batch 23300, giga_loss[loss=0.2654, simple_loss=0.3402, pruned_loss=0.09532, over 28885.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3412, pruned_loss=0.09636, over 5706344.81 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.353, pruned_loss=0.1011, over 5733411.83 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3392, pruned_loss=0.09493, over 5711069.63 frames. ], batch size: 66, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:29:43,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.425e+02 1.247e+03 1.600e+03 2.124e+03 4.308e+03, threshold=3.200e+03, percent-clipped=9.0
+2023-03-10 23:29:47,770 INFO [train.py:968] (0/2) Epoch 21, batch 23350, giga_loss[loss=0.3122, simple_loss=0.3756, pruned_loss=0.1244, over 28984.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3429, pruned_loss=0.09692, over 5716518.23 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3532, pruned_loss=0.1014, over 5736070.21 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3412, pruned_loss=0.09546, over 5717496.77 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:30:32,040 INFO [train.py:968] (0/2) Epoch 21, batch 23400, giga_loss[loss=0.2272, simple_loss=0.3168, pruned_loss=0.06882, over 28831.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3468, pruned_loss=0.1001, over 5712044.65 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3532, pruned_loss=0.1016, over 5738325.54 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3451, pruned_loss=0.09869, over 5710477.33 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:30:56,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2807, 3.3308, 1.4158, 1.5068], device='cuda:0'), covar=tensor([0.1010, 0.0457, 0.0945, 0.1328], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0550, 0.0382, 0.0422], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-10 23:31:13,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.201e+02 1.625e+03 1.962e+03 2.492e+03 7.887e+03, threshold=3.924e+03, percent-clipped=14.0
+2023-03-10 23:31:19,378 INFO [train.py:968] (0/2) Epoch 21, batch 23450, giga_loss[loss=0.3535, simple_loss=0.4085, pruned_loss=0.1492, over 27676.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3529, pruned_loss=0.1054, over 5705316.51 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.353, pruned_loss=0.1018, over 5739859.51 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3517, pruned_loss=0.1041, over 5701816.79 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:32:01,856 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.76 vs. limit=5.0
+2023-03-10 23:32:10,047 INFO [train.py:968] (0/2) Epoch 21, batch 23500, giga_loss[loss=0.2914, simple_loss=0.3625, pruned_loss=0.1102, over 28877.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3604, pruned_loss=0.1108, over 5692883.53 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3533, pruned_loss=0.102, over 5742707.76 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3592, pruned_loss=0.1098, over 5686762.31 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:32:22,742 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=935863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:32:27,809 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 23:32:50,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.795e+03 2.359e+03 3.522e+03 8.706e+03, threshold=4.719e+03, percent-clipped=18.0
+2023-03-10 23:32:55,502 INFO [train.py:968] (0/2) Epoch 21, batch 23550, giga_loss[loss=0.3651, simple_loss=0.4204, pruned_loss=0.1549, over 28531.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3668, pruned_loss=0.1157, over 5682052.58 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3537, pruned_loss=0.1023, over 5737437.97 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3658, pruned_loss=0.1148, over 5680477.31 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:33:32,669 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=935938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:33:43,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:33:46,577 INFO [train.py:968] (0/2) Epoch 21, batch 23600, giga_loss[loss=0.3102, simple_loss=0.3802, pruned_loss=0.1201, over 28912.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3723, pruned_loss=0.1208, over 5682416.72 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3533, pruned_loss=0.1023, over 5740793.05 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3722, pruned_loss=0.1204, over 5677204.92 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:33:51,534 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6514, 1.7140, 1.8775, 1.4232], device='cuda:0'), covar=tensor([0.1555, 0.2482, 0.1293, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.0888, 0.0691, 0.0935, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 23:33:54,749 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=935960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:34:12,372 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-10 23:34:28,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.149e+03 1.780e+03 2.277e+03 3.089e+03 6.127e+03, threshold=4.554e+03, percent-clipped=7.0
+2023-03-10 23:34:33,251 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-936000.pt
+2023-03-10 23:34:34,188 INFO [train.py:968] (0/2) Epoch 21, batch 23650, giga_loss[loss=0.2956, simple_loss=0.3594, pruned_loss=0.1158, over 28781.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3768, pruned_loss=0.1241, over 5669292.72 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3536, pruned_loss=0.1026, over 5724718.54 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3769, pruned_loss=0.124, over 5678980.05 frames. ], batch size: 92, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:34:57,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3679, 1.6264, 1.6402, 1.5002], device='cuda:0'), covar=tensor([0.1260, 0.1042, 0.1324, 0.1125], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0748, 0.0714, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 23:35:19,540 INFO [train.py:968] (0/2) Epoch 21, batch 23700, giga_loss[loss=0.293, simple_loss=0.3588, pruned_loss=0.1136, over 28285.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3795, pruned_loss=0.1275, over 5656522.06 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3539, pruned_loss=0.1029, over 5717755.86 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3798, pruned_loss=0.1275, over 5669409.32 frames. ], batch size: 77, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:35:55,612 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4217, 1.5255, 1.6442, 1.2841], device='cuda:0'), covar=tensor([0.1118, 0.1753, 0.0969, 0.1289], device='cuda:0'), in_proj_covar=tensor([0.0888, 0.0691, 0.0935, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 23:35:58,345 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:36:01,650 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.154e+03 1.720e+03 2.131e+03 3.016e+03 1.320e+04, threshold=4.262e+03, percent-clipped=5.0
+2023-03-10 23:36:02,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:36:07,188 INFO [train.py:968] (0/2) Epoch 21, batch 23750, giga_loss[loss=0.3104, simple_loss=0.3727, pruned_loss=0.1241, over 28708.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3831, pruned_loss=0.132, over 5649424.57 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3534, pruned_loss=0.1029, over 5723550.03 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3845, pruned_loss=0.1328, over 5652943.84 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:36:12,908 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:36:31,260 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=936125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:36:33,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3740, 1.4973, 1.5786, 1.2256], device='cuda:0'), covar=tensor([0.1144, 0.1946, 0.1044, 0.1418], device='cuda:0'), in_proj_covar=tensor([0.0889, 0.0692, 0.0935, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 23:36:54,405 INFO [train.py:968] (0/2) Epoch 21, batch 23800, giga_loss[loss=0.4594, simple_loss=0.4666, pruned_loss=0.2261, over 23458.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.386, pruned_loss=0.1353, over 5638123.60 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3535, pruned_loss=0.1032, over 5719148.14 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3881, pruned_loss=0.1366, over 5642410.43 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:36:54,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8570, 1.8490, 2.0598, 1.5963], device='cuda:0'), covar=tensor([0.1454, 0.2032, 0.1148, 0.1443], device='cuda:0'), in_proj_covar=tensor([0.0889, 0.0692, 0.0935, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 23:37:40,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.303e+03 1.908e+03 2.320e+03 3.590e+03 1.162e+04, threshold=4.640e+03, percent-clipped=15.0
+2023-03-10 23:37:46,114 INFO [train.py:968] (0/2) Epoch 21, batch 23850, giga_loss[loss=0.3295, simple_loss=0.3936, pruned_loss=0.1327, over 28877.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3881, pruned_loss=0.1368, over 5638880.13 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3534, pruned_loss=0.1033, over 5718332.39 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3909, pruned_loss=0.1389, over 5640303.48 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:38:29,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:38:37,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3969, 1.4435, 1.4250, 1.5812], device='cuda:0'), covar=tensor([0.0602, 0.0301, 0.0271, 0.0616], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0116, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 23:38:39,689 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 23:38:43,253 INFO [train.py:968] (0/2) Epoch 21, batch 23900, giga_loss[loss=0.3305, simple_loss=0.3876, pruned_loss=0.1367, over 28605.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3899, pruned_loss=0.1386, over 5640803.71 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3538, pruned_loss=0.1037, over 5720254.41 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3921, pruned_loss=0.1402, over 5639395.99 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:38:47,579 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-10 23:39:25,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.844e+03 2.607e+03 3.581e+03 1.491e+04, threshold=5.214e+03, percent-clipped=15.0
+2023-03-10 23:39:28,900 INFO [train.py:968] (0/2) Epoch 21, batch 23950, giga_loss[loss=0.3199, simple_loss=0.3823, pruned_loss=0.1288, over 29106.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3892, pruned_loss=0.1389, over 5644917.25 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3545, pruned_loss=0.1045, over 5727001.96 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3916, pruned_loss=0.1408, over 5634686.03 frames. ], batch size: 128, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:39:40,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:39:47,011 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-10 23:39:47,062 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-10 23:40:01,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:40:09,785 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:40:14,504 INFO [train.py:968] (0/2) Epoch 21, batch 24000, giga_loss[loss=0.3152, simple_loss=0.3775, pruned_loss=0.1265, over 29023.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3878, pruned_loss=0.138, over 5652277.45 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3543, pruned_loss=0.1045, over 5730844.84 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3909, pruned_loss=0.1404, over 5638433.59 frames. ], batch size: 128, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:40:14,508 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-10 23:40:23,182 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2051, simple_loss=0.3129, pruned_loss=0.0486, over 944034.00 frames. 
+2023-03-10 23:40:23,183 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-10 23:40:24,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-10 23:40:49,527 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:40:52,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:41:04,344 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.751e+03 2.144e+03 3.325e+03 9.996e+03, threshold=4.288e+03, percent-clipped=10.0
+2023-03-10 23:41:08,649 INFO [train.py:968] (0/2) Epoch 21, batch 24050, libri_loss[loss=0.3115, simple_loss=0.3839, pruned_loss=0.1196, over 28650.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.387, pruned_loss=0.1361, over 5652141.60 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.1049, over 5732338.22 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3903, pruned_loss=0.1387, over 5637250.23 frames. ], batch size: 106, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:41:22,135 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=936413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:41:57,422 INFO [train.py:968] (0/2) Epoch 21, batch 24100, giga_loss[loss=0.4357, simple_loss=0.4502, pruned_loss=0.2106, over 26668.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3887, pruned_loss=0.1375, over 5642792.93 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3551, pruned_loss=0.1053, over 5732858.57 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3919, pruned_loss=0.1401, over 5627556.67 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:42:00,901 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:04,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:24,931 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:25,629 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:27,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:29,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:35,545 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=936488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:41,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.293e+03 1.871e+03 2.221e+03 3.138e+03 9.072e+03, threshold=4.442e+03, percent-clipped=11.0
+2023-03-10 23:42:48,308 INFO [train.py:968] (0/2) Epoch 21, batch 24150, giga_loss[loss=0.4792, simple_loss=0.4882, pruned_loss=0.2351, over 24376.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3892, pruned_loss=0.1377, over 5638068.43 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.355, pruned_loss=0.1055, over 5737548.28 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.393, pruned_loss=0.1407, over 5618434.78 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:42:56,584 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=936510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:43:39,207 INFO [train.py:968] (0/2) Epoch 21, batch 24200, giga_loss[loss=0.3464, simple_loss=0.4011, pruned_loss=0.1459, over 28611.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3842, pruned_loss=0.1323, over 5644302.62 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3552, pruned_loss=0.1057, over 5741097.22 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3876, pruned_loss=0.1351, over 5624179.42 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:44:04,654 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936582.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 23:44:18,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.648e+03 1.998e+03 3.064e+03 6.357e+03, threshold=3.995e+03, percent-clipped=4.0
+2023-03-10 23:44:22,762 INFO [train.py:968] (0/2) Epoch 21, batch 24250, giga_loss[loss=0.2618, simple_loss=0.3453, pruned_loss=0.08917, over 28394.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3816, pruned_loss=0.1291, over 5653457.00 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.355, pruned_loss=0.1058, over 5746506.10 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3853, pruned_loss=0.132, over 5629488.29 frames. ], batch size: 60, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:44:34,367 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1694, 3.2930, 2.1610, 1.4068], device='cuda:0'), covar=tensor([0.6715, 0.3166, 0.3712, 0.5644], device='cuda:0'), in_proj_covar=tensor([0.1746, 0.1644, 0.1599, 0.1418], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 23:44:46,571 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:44:48,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4938, 1.6310, 1.2171, 1.2348], device='cuda:0'), covar=tensor([0.0881, 0.0571, 0.0964, 0.1264], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0448, 0.0516, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 23:44:48,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:45:09,664 INFO [train.py:968] (0/2) Epoch 21, batch 24300, giga_loss[loss=0.3395, simple_loss=0.3994, pruned_loss=0.1399, over 28960.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3791, pruned_loss=0.1267, over 5670376.81 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3553, pruned_loss=0.1061, over 5749567.77 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3823, pruned_loss=0.1292, over 5647112.21 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:45:14,694 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=936657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:45:53,132 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-10 23:45:54,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.999e+02 1.799e+03 2.399e+03 3.556e+03 1.560e+04, threshold=4.799e+03, percent-clipped=18.0
+2023-03-10 23:45:58,964 INFO [train.py:968] (0/2) Epoch 21, batch 24350, giga_loss[loss=0.2789, simple_loss=0.354, pruned_loss=0.1019, over 28968.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3762, pruned_loss=0.1242, over 5659511.63 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3553, pruned_loss=0.1062, over 5741194.76 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.379, pruned_loss=0.1262, over 5648128.82 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:46:14,870 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:46:44,447 INFO [train.py:968] (0/2) Epoch 21, batch 24400, giga_loss[loss=0.3004, simple_loss=0.3677, pruned_loss=0.1165, over 28879.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3741, pruned_loss=0.1228, over 5670449.61 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3553, pruned_loss=0.1063, over 5745152.87 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3766, pruned_loss=0.1247, over 5656186.13 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:46:48,827 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:47:31,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.916e+02 1.525e+03 1.850e+03 2.455e+03 4.854e+03, threshold=3.701e+03, percent-clipped=1.0
+2023-03-10 23:47:34,643 INFO [train.py:968] (0/2) Epoch 21, batch 24450, giga_loss[loss=0.3625, simple_loss=0.4159, pruned_loss=0.1545, over 28618.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3746, pruned_loss=0.1231, over 5658686.14 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3556, pruned_loss=0.1067, over 5736080.54 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3768, pruned_loss=0.1245, over 5654042.32 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:47:39,538 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:47:53,720 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:48:14,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7214, 1.9536, 2.0199, 1.4904], device='cuda:0'), covar=tensor([0.1827, 0.2847, 0.1569, 0.1983], device='cuda:0'), in_proj_covar=tensor([0.0891, 0.0696, 0.0937, 0.0833], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-10 23:48:24,273 INFO [train.py:968] (0/2) Epoch 21, batch 24500, giga_loss[loss=0.2845, simple_loss=0.359, pruned_loss=0.105, over 28892.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.374, pruned_loss=0.1221, over 5661796.83 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3561, pruned_loss=0.1071, over 5730368.93 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3758, pruned_loss=0.1234, over 5660603.60 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:48:25,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:48:36,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:48:39,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:48:51,120 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-10 23:49:05,543 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-10 23:49:09,847 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=936894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:49:14,805 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.600e+03 2.053e+03 2.653e+03 1.207e+04, threshold=4.107e+03, percent-clipped=8.0
+2023-03-10 23:49:17,032 INFO [train.py:968] (0/2) Epoch 21, batch 24550, libri_loss[loss=0.36, simple_loss=0.4031, pruned_loss=0.1584, over 19301.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3743, pruned_loss=0.1203, over 5664275.59 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3567, pruned_loss=0.1077, over 5722896.89 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3754, pruned_loss=0.1209, over 5670065.96 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:49:29,727 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9779, 1.5692, 5.5039, 3.7947], device='cuda:0'), covar=tensor([0.1688, 0.2787, 0.0423, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0757, 0.0650, 0.0964, 0.0914], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 23:49:39,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4075, 1.5293, 1.4770, 1.5371], device='cuda:0'), covar=tensor([0.0622, 0.0296, 0.0277, 0.0667], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-10 23:50:09,501 INFO [train.py:968] (0/2) Epoch 21, batch 24600, giga_loss[loss=0.3675, simple_loss=0.4002, pruned_loss=0.1674, over 23567.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3763, pruned_loss=0.1203, over 5649950.00 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3568, pruned_loss=0.1077, over 5723970.90 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3774, pruned_loss=0.121, over 5652699.94 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:50:18,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936957.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 23:50:54,503 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:50:56,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.126e+03 1.856e+03 2.166e+03 3.562e+03 6.995e+03, threshold=4.332e+03, percent-clipped=12.0
+2023-03-10 23:50:57,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:50:59,548 INFO [train.py:968] (0/2) Epoch 21, batch 24650, giga_loss[loss=0.3026, simple_loss=0.3721, pruned_loss=0.1166, over 28506.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3765, pruned_loss=0.1208, over 5658696.36 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3569, pruned_loss=0.1079, over 5727568.02 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3775, pruned_loss=0.1214, over 5656423.66 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:51:22,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:51:44,620 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=937049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:51:45,678 INFO [train.py:968] (0/2) Epoch 21, batch 24700, giga_loss[loss=0.2881, simple_loss=0.364, pruned_loss=0.1061, over 28473.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3758, pruned_loss=0.1211, over 5661321.48 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3567, pruned_loss=0.108, over 5734180.54 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3778, pruned_loss=0.1221, over 5650705.90 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:52:10,397 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=937075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:52:31,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.857e+03 2.240e+03 2.808e+03 6.422e+03, threshold=4.481e+03, percent-clipped=8.0
+2023-03-10 23:52:34,048 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937100.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 23:52:34,382 INFO [train.py:968] (0/2) Epoch 21, batch 24750, giga_loss[loss=0.3092, simple_loss=0.3746, pruned_loss=0.1219, over 28689.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3733, pruned_loss=0.1204, over 5660286.59 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.357, pruned_loss=0.1083, over 5737038.53 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3749, pruned_loss=0.1211, over 5648017.54 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:52:37,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937103.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 23:52:37,346 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-10 23:53:03,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:53:04,300 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937132.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 23:53:18,545 INFO [train.py:968] (0/2) Epoch 21, batch 24800, giga_loss[loss=0.2789, simple_loss=0.353, pruned_loss=0.1023, over 28182.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3719, pruned_loss=0.12, over 5665837.30 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3571, pruned_loss=0.1083, over 5731109.38 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3734, pruned_loss=0.1209, over 5660418.02 frames. ], batch size: 77, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:53:25,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9922, 1.3098, 1.0624, 0.2147], device='cuda:0'), covar=tensor([0.3726, 0.2992, 0.4345, 0.6498], device='cuda:0'), in_proj_covar=tensor([0.1745, 0.1642, 0.1599, 0.1413], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 23:53:42,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:53:54,365 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:54:00,413 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.664e+03 2.302e+03 3.619e+03 8.980e+03, threshold=4.604e+03, percent-clipped=16.0
+2023-03-10 23:54:02,480 INFO [train.py:968] (0/2) Epoch 21, batch 24850, giga_loss[loss=0.3293, simple_loss=0.39, pruned_loss=0.1344, over 28757.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3713, pruned_loss=0.1202, over 5654491.91 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3571, pruned_loss=0.1084, over 5720619.06 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3727, pruned_loss=0.1209, over 5658452.61 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:54:46,021 INFO [train.py:968] (0/2) Epoch 21, batch 24900, giga_loss[loss=0.3046, simple_loss=0.377, pruned_loss=0.1161, over 28878.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3712, pruned_loss=0.1186, over 5668060.92 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3577, pruned_loss=0.1087, over 5721251.03 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3722, pruned_loss=0.1193, over 5668981.73 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:55:08,268 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:55:10,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:55:29,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.522e+02 1.492e+03 1.887e+03 2.568e+03 6.600e+03, threshold=3.774e+03, percent-clipped=2.0
+2023-03-10 23:55:30,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8073, 1.9330, 1.7716, 1.6260], device='cuda:0'), covar=tensor([0.1973, 0.2635, 0.2450, 0.2481], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0749, 0.0716, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-10 23:55:31,165 INFO [train.py:968] (0/2) Epoch 21, batch 24950, giga_loss[loss=0.2733, simple_loss=0.3512, pruned_loss=0.09767, over 28651.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3712, pruned_loss=0.1192, over 5662928.41 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.357, pruned_loss=0.1085, over 5726934.21 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3731, pruned_loss=0.1203, over 5656648.53 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:55:33,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3965, 2.9106, 1.3857, 1.5503], device='cuda:0'), covar=tensor([0.0973, 0.0413, 0.0910, 0.1306], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0558, 0.0386, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-10 23:55:34,106 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:55:34,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4794, 3.2961, 1.6116, 1.5502], device='cuda:0'), covar=tensor([0.0964, 0.0395, 0.0881, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0558, 0.0386, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-10 23:55:46,951 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:55:49,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:56:01,233 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:56:03,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3506, 1.1985, 3.8766, 3.2763], device='cuda:0'), covar=tensor([0.1623, 0.2838, 0.0505, 0.0918], device='cuda:0'), in_proj_covar=tensor([0.0757, 0.0648, 0.0963, 0.0912], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-10 23:56:03,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:56:14,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6924, 2.4196, 1.5043, 0.8498], device='cuda:0'), covar=tensor([0.8432, 0.3697, 0.4404, 0.7535], device='cuda:0'), in_proj_covar=tensor([0.1739, 0.1636, 0.1593, 0.1408], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-10 23:56:17,023 INFO [train.py:968] (0/2) Epoch 21, batch 25000, giga_loss[loss=0.278, simple_loss=0.355, pruned_loss=0.1005, over 28763.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3704, pruned_loss=0.1182, over 5667617.84 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.357, pruned_loss=0.1088, over 5721436.80 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3721, pruned_loss=0.1189, over 5666903.76 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:56:20,103 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:56:29,068 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 23:56:33,788 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:56:40,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4790, 2.7668, 2.2030, 2.4085], device='cuda:0'), covar=tensor([0.0796, 0.0443, 0.0757, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0447, 0.0515, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 23:56:55,674 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=937391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:57:02,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.277e+03 1.769e+03 2.374e+03 3.225e+03 1.093e+04, threshold=4.748e+03, percent-clipped=15.0
+2023-03-10 23:57:04,422 INFO [train.py:968] (0/2) Epoch 21, batch 25050, giga_loss[loss=0.374, simple_loss=0.4113, pruned_loss=0.1683, over 26651.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3692, pruned_loss=0.1181, over 5672211.69 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3574, pruned_loss=0.1092, over 5714555.20 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3705, pruned_loss=0.1186, over 5676762.32 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:57:24,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:57:48,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:57:49,170 INFO [train.py:968] (0/2) Epoch 21, batch 25100, giga_loss[loss=0.3311, simple_loss=0.3907, pruned_loss=0.1358, over 28568.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3692, pruned_loss=0.1187, over 5675519.84 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3576, pruned_loss=0.1093, over 5719882.40 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3705, pruned_loss=0.1192, over 5673248.96 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:57:59,680 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7304, 1.8583, 1.3991, 1.5251], device='cuda:0'), covar=tensor([0.0816, 0.0458, 0.0946, 0.0908], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0447, 0.0515, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-10 23:58:31,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.193e+03 1.688e+03 2.223e+03 3.259e+03 7.538e+03, threshold=4.446e+03, percent-clipped=2.0
+2023-03-10 23:58:33,139 INFO [train.py:968] (0/2) Epoch 21, batch 25150, libri_loss[loss=0.3431, simple_loss=0.4032, pruned_loss=0.1414, over 25790.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3683, pruned_loss=0.1184, over 5688833.34 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.358, pruned_loss=0.1096, over 5721099.61 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3692, pruned_loss=0.1189, over 5685230.88 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:59:16,788 INFO [train.py:968] (0/2) Epoch 21, batch 25200, giga_loss[loss=0.259, simple_loss=0.3356, pruned_loss=0.09124, over 28899.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3663, pruned_loss=0.1178, over 5683304.23 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3582, pruned_loss=0.11, over 5716583.05 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3671, pruned_loss=0.1179, over 5683920.06 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:59:34,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:59:36,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:59:57,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:59:59,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:00:01,695 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.614e+02 1.951e+03 2.503e+03 3.427e+03 1.081e+04, threshold=5.005e+03, percent-clipped=12.0
+2023-03-11 00:00:02,012 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:00:04,540 INFO [train.py:968] (0/2) Epoch 21, batch 25250, giga_loss[loss=0.2583, simple_loss=0.3329, pruned_loss=0.09187, over 28500.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3645, pruned_loss=0.117, over 5685086.71 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3581, pruned_loss=0.1099, over 5718487.34 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3654, pruned_loss=0.1174, over 5683510.98 frames. ], batch size: 60, lr: 1.52e-03, grad_scale: 8.0
+2023-03-11 00:00:31,471 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:00:46,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6405, 1.7248, 1.7412, 1.4824], device='cuda:0'), covar=tensor([0.2026, 0.2389, 0.2500, 0.2544], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0752, 0.0717, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 00:00:53,654 INFO [train.py:968] (0/2) Epoch 21, batch 25300, giga_loss[loss=0.3071, simple_loss=0.3748, pruned_loss=0.1197, over 28887.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3642, pruned_loss=0.1172, over 5682469.25 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3579, pruned_loss=0.1097, over 5721155.75 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3651, pruned_loss=0.1176, over 5678256.03 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:01:17,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0526, 2.1963, 1.5782, 1.8435], device='cuda:0'), covar=tensor([0.0970, 0.0737, 0.1060, 0.1128], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0446, 0.0514, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 00:01:35,220 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-11 00:01:38,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.344e+02 1.761e+03 2.297e+03 3.729e+03 1.207e+04, threshold=4.594e+03, percent-clipped=11.0
+2023-03-11 00:01:38,906 INFO [train.py:968] (0/2) Epoch 21, batch 25350, giga_loss[loss=0.2819, simple_loss=0.3561, pruned_loss=0.1039, over 28944.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3652, pruned_loss=0.117, over 5681632.94 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3581, pruned_loss=0.1099, over 5715719.33 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.366, pruned_loss=0.1175, over 5681501.26 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:01:45,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2206, 1.4403, 1.3238, 1.1416], device='cuda:0'), covar=tensor([0.2872, 0.2608, 0.1829, 0.2357], device='cuda:0'), in_proj_covar=tensor([0.1975, 0.1893, 0.1836, 0.1968], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 00:01:55,524 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=937721.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 00:02:12,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0079, 1.2939, 1.0404, 0.2477], device='cuda:0'), covar=tensor([0.3076, 0.2298, 0.3363, 0.5427], device='cuda:0'), in_proj_covar=tensor([0.1740, 0.1637, 0.1589, 0.1407], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 00:02:16,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2432, 1.8983, 1.5689, 1.3440], device='cuda:0'), covar=tensor([0.0736, 0.0370, 0.0293, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:0')
+2023-03-11 00:02:23,782 INFO [train.py:968] (0/2) Epoch 21, batch 25400, giga_loss[loss=0.264, simple_loss=0.3395, pruned_loss=0.09423, over 28509.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3655, pruned_loss=0.1166, over 5685077.78 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3577, pruned_loss=0.1097, over 5720502.43 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3666, pruned_loss=0.1173, over 5680018.85 frames. ], batch size: 65, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:02:36,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:03:05,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.664e+02 1.642e+03 2.040e+03 2.789e+03 7.413e+03, threshold=4.081e+03, percent-clipped=7.0
+2023-03-11 00:03:06,591 INFO [train.py:968] (0/2) Epoch 21, batch 25450, giga_loss[loss=0.2827, simple_loss=0.3632, pruned_loss=0.1011, over 28598.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3653, pruned_loss=0.1157, over 5691292.36 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3577, pruned_loss=0.1099, over 5725687.73 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3665, pruned_loss=0.1164, over 5681306.71 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:03:40,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3750, 1.5528, 1.6276, 1.4780], device='cuda:0'), covar=tensor([0.1421, 0.1282, 0.1521, 0.1310], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0752, 0.0716, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 00:03:53,055 INFO [train.py:968] (0/2) Epoch 21, batch 25500, giga_loss[loss=0.2674, simple_loss=0.3356, pruned_loss=0.09965, over 28351.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3654, pruned_loss=0.1161, over 5687527.35 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3569, pruned_loss=0.1093, over 5727729.01 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3672, pruned_loss=0.1172, over 5677049.29 frames. ], batch size: 71, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:04:42,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 2.027e+03 2.582e+03 3.501e+03 9.833e+03, threshold=5.165e+03, percent-clipped=18.0
+2023-03-11 00:04:43,011 INFO [train.py:968] (0/2) Epoch 21, batch 25550, giga_loss[loss=0.3121, simple_loss=0.3759, pruned_loss=0.1241, over 28592.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3691, pruned_loss=0.1192, over 5686075.00 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3569, pruned_loss=0.1093, over 5726624.52 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3707, pruned_loss=0.1202, over 5678373.99 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:04:50,557 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:04:52,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:05:23,272 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:05:30,698 INFO [train.py:968] (0/2) Epoch 21, batch 25600, giga_loss[loss=0.2664, simple_loss=0.3503, pruned_loss=0.09126, over 28845.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3707, pruned_loss=0.122, over 5685483.22 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3567, pruned_loss=0.1092, over 5729897.71 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3724, pruned_loss=0.1231, over 5676008.86 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:06:19,830 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-938000.pt
+2023-03-11 00:06:20,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.976e+03 2.639e+03 3.879e+03 1.012e+04, threshold=5.278e+03, percent-clipped=8.0
+2023-03-11 00:06:20,806 INFO [train.py:968] (0/2) Epoch 21, batch 25650, libri_loss[loss=0.2986, simple_loss=0.3643, pruned_loss=0.1165, over 29567.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.372, pruned_loss=0.1241, over 5684260.55 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3569, pruned_loss=0.1092, over 5733888.03 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3736, pruned_loss=0.1254, over 5671312.49 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:07:07,002 INFO [train.py:968] (0/2) Epoch 21, batch 25700, giga_loss[loss=0.352, simple_loss=0.3944, pruned_loss=0.1548, over 28681.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3717, pruned_loss=0.1238, over 5691640.92 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3568, pruned_loss=0.1091, over 5736450.49 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3734, pruned_loss=0.1252, over 5678248.86 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:07:23,839 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=938071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:07:46,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=938096.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 00:07:51,875 INFO [train.py:968] (0/2) Epoch 21, batch 25750, giga_loss[loss=0.2895, simple_loss=0.3527, pruned_loss=0.1131, over 28451.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3708, pruned_loss=0.1237, over 5682996.12 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3572, pruned_loss=0.1096, over 5741621.07 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3723, pruned_loss=0.125, over 5665748.86 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:07:52,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.243e+03 1.804e+03 2.419e+03 3.523e+03 2.296e+04, threshold=4.839e+03, percent-clipped=10.0
+2023-03-11 00:08:33,517 INFO [train.py:968] (0/2) Epoch 21, batch 25800, giga_loss[loss=0.4291, simple_loss=0.4465, pruned_loss=0.2058, over 26444.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3708, pruned_loss=0.1224, over 5683564.63 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3571, pruned_loss=0.1097, over 5735400.51 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3724, pruned_loss=0.1237, over 5673328.84 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:09:19,657 INFO [train.py:968] (0/2) Epoch 21, batch 25850, giga_loss[loss=0.2617, simple_loss=0.3435, pruned_loss=0.08998, over 28756.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3684, pruned_loss=0.1199, over 5679038.13 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3568, pruned_loss=0.1096, over 5738896.71 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3702, pruned_loss=0.1211, over 5666855.97 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:09:21,908 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.767e+02 1.623e+03 1.991e+03 2.679e+03 7.321e+03, threshold=3.983e+03, percent-clipped=5.0
+2023-03-11 00:09:37,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4506, 1.9497, 1.4010, 0.6603], device='cuda:0'), covar=tensor([0.5397, 0.3211, 0.3653, 0.6314], device='cuda:0'), in_proj_covar=tensor([0.1740, 0.1645, 0.1594, 0.1411], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 00:09:51,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5274, 1.8582, 1.4770, 1.4696], device='cuda:0'), covar=tensor([0.2609, 0.2708, 0.3043, 0.2430], device='cuda:0'), in_proj_covar=tensor([0.1493, 0.1084, 0.1318, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 00:09:58,324 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=938239.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 00:10:03,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=938242.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 00:10:10,279 INFO [train.py:968] (0/2) Epoch 21, batch 25900, giga_loss[loss=0.2834, simple_loss=0.3541, pruned_loss=0.1063, over 28689.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3659, pruned_loss=0.1183, over 5675483.36 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3567, pruned_loss=0.1096, over 5741669.46 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3677, pruned_loss=0.1196, over 5661888.34 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:10:27,937 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=938271.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 00:10:56,607 INFO [train.py:968] (0/2) Epoch 21, batch 25950, giga_loss[loss=0.2531, simple_loss=0.3245, pruned_loss=0.09083, over 28226.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3648, pruned_loss=0.1183, over 5675716.46 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3568, pruned_loss=0.1097, over 5743321.74 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3661, pruned_loss=0.1192, over 5663039.74 frames. ], batch size: 77, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:10:58,375 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.660e+03 2.033e+03 2.676e+03 8.691e+03, threshold=4.067e+03, percent-clipped=5.0
+2023-03-11 00:11:48,100 INFO [train.py:968] (0/2) Epoch 21, batch 26000, giga_loss[loss=0.2807, simple_loss=0.35, pruned_loss=0.1057, over 28880.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3675, pruned_loss=0.1211, over 5659398.68 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3577, pruned_loss=0.1104, over 5740908.60 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.368, pruned_loss=0.1214, over 5649685.47 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:12:32,699 INFO [train.py:968] (0/2) Epoch 21, batch 26050, giga_loss[loss=0.3245, simple_loss=0.3984, pruned_loss=0.1253, over 28845.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3703, pruned_loss=0.1225, over 5670733.24 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3577, pruned_loss=0.1106, over 5746747.93 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.371, pruned_loss=0.1229, over 5655252.82 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:12:34,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.839e+03 2.628e+03 3.359e+03 9.134e+03, threshold=5.257e+03, percent-clipped=20.0
+2023-03-11 00:13:12,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=938446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:13:15,791 INFO [train.py:968] (0/2) Epoch 21, batch 26100, giga_loss[loss=0.2881, simple_loss=0.3816, pruned_loss=0.09736, over 28900.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3731, pruned_loss=0.1217, over 5669057.87 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.358, pruned_loss=0.1111, over 5740387.06 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3741, pruned_loss=0.122, over 5658512.12 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:14:05,682 INFO [train.py:968] (0/2) Epoch 21, batch 26150, libri_loss[loss=0.2963, simple_loss=0.3658, pruned_loss=0.1134, over 29157.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3752, pruned_loss=0.1221, over 5668419.74 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3579, pruned_loss=0.1111, over 5742377.83 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3763, pruned_loss=0.1226, over 5657227.57 frames. ], batch size: 101, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:14:07,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.850e+03 2.513e+03 3.211e+03 1.115e+04, threshold=5.027e+03, percent-clipped=8.0
+2023-03-11 00:14:15,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 00:14:19,867 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=938519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:14:29,347 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=938527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:14:37,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3746, 2.1399, 1.5675, 0.5841], device='cuda:0'), covar=tensor([0.5285, 0.2721, 0.3607, 0.6085], device='cuda:0'), in_proj_covar=tensor([0.1742, 0.1640, 0.1592, 0.1410], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 00:14:53,172 INFO [train.py:968] (0/2) Epoch 21, batch 26200, giga_loss[loss=0.3242, simple_loss=0.3927, pruned_loss=0.1279, over 28867.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3765, pruned_loss=0.1237, over 5659859.29 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3577, pruned_loss=0.1112, over 5746305.58 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.378, pruned_loss=0.1243, over 5645893.74 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:15:05,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5031, 1.1257, 4.4695, 3.5789], device='cuda:0'), covar=tensor([0.1650, 0.2879, 0.0415, 0.0951], device='cuda:0'), in_proj_covar=tensor([0.0760, 0.0648, 0.0965, 0.0912], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 00:15:18,929 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-11 00:15:28,859 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=938589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:15:30,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=938592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:15:36,991 INFO [train.py:968] (0/2) Epoch 21, batch 26250, giga_loss[loss=0.3755, simple_loss=0.4189, pruned_loss=0.1661, over 28921.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3776, pruned_loss=0.1248, over 5669769.08 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3578, pruned_loss=0.1113, over 5747801.27 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3788, pruned_loss=0.1253, over 5656539.64 frames. ], batch size: 106, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:15:38,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.208e+03 1.723e+03 2.243e+03 3.200e+03 7.737e+03, threshold=4.485e+03, percent-clipped=3.0
+2023-03-11 00:15:58,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=938621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:16:07,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6488, 1.6890, 1.8606, 1.4285], device='cuda:0'), covar=tensor([0.1780, 0.2490, 0.1418, 0.1750], device='cuda:0'), in_proj_covar=tensor([0.0897, 0.0702, 0.0943, 0.0839], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 00:16:25,592 INFO [train.py:968] (0/2) Epoch 21, batch 26300, giga_loss[loss=0.2774, simple_loss=0.3489, pruned_loss=0.1029, over 28354.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.378, pruned_loss=0.1262, over 5655430.35 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3585, pruned_loss=0.1119, over 5741243.25 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3789, pruned_loss=0.1264, over 5647459.96 frames. ], batch size: 65, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:16:30,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1108, 4.9369, 4.7012, 2.4168], device='cuda:0'), covar=tensor([0.0477, 0.0625, 0.0707, 0.1823], device='cuda:0'), in_proj_covar=tensor([0.1226, 0.1138, 0.0963, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 00:16:56,307 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.42 vs. limit=5.0
+2023-03-11 00:17:09,542 INFO [train.py:968] (0/2) Epoch 21, batch 26350, giga_loss[loss=0.2761, simple_loss=0.3441, pruned_loss=0.104, over 28583.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3766, pruned_loss=0.1256, over 5657044.85 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5745782.43 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3779, pruned_loss=0.1263, over 5642276.02 frames. ], batch size: 85, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:17:10,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.875e+03 2.298e+03 3.246e+03 7.959e+03, threshold=4.595e+03, percent-clipped=6.0
+2023-03-11 00:17:15,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4127, 1.5387, 3.2794, 3.1394], device='cuda:0'), covar=tensor([0.1294, 0.2411, 0.0474, 0.0992], device='cuda:0'), in_proj_covar=tensor([0.0762, 0.0651, 0.0969, 0.0916], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 00:17:56,336 INFO [train.py:968] (0/2) Epoch 21, batch 26400, giga_loss[loss=0.3434, simple_loss=0.3887, pruned_loss=0.1491, over 28832.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3738, pruned_loss=0.1244, over 5655147.68 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3593, pruned_loss=0.1126, over 5739628.45 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3747, pruned_loss=0.1247, over 5647010.39 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:18:22,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2936, 1.6054, 0.9496, 1.1751], device='cuda:0'), covar=tensor([0.1297, 0.0804, 0.1661, 0.1401], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0447, 0.0515, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 00:18:44,022 INFO [train.py:968] (0/2) Epoch 21, batch 26450, giga_loss[loss=0.322, simple_loss=0.3867, pruned_loss=0.1286, over 28597.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.373, pruned_loss=0.1241, over 5653573.39 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3601, pruned_loss=0.113, over 5740803.73 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3736, pruned_loss=0.1244, over 5642277.30 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:18:46,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.876e+03 2.613e+03 3.849e+03 8.611e+03, threshold=5.225e+03, percent-clipped=15.0
+2023-03-11 00:19:28,962 INFO [train.py:968] (0/2) Epoch 21, batch 26500, libri_loss[loss=0.3441, simple_loss=0.3931, pruned_loss=0.1476, over 19763.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3733, pruned_loss=0.1246, over 5645495.89 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1133, over 5734709.12 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.1249, over 5640275.72 frames. ], batch size: 187, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:19:44,611 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 00:20:05,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=938894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:20:11,300 INFO [train.py:968] (0/2) Epoch 21, batch 26550, libri_loss[loss=0.2811, simple_loss=0.3401, pruned_loss=0.1111, over 29353.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.372, pruned_loss=0.1238, over 5660009.06 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1132, over 5738548.71 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.373, pruned_loss=0.1244, over 5650342.94 frames. ], batch size: 71, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:20:12,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=938902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:20:13,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.723e+03 2.129e+03 3.064e+03 9.752e+03, threshold=4.259e+03, percent-clipped=8.0
+2023-03-11 00:20:57,254 INFO [train.py:968] (0/2) Epoch 21, batch 26600, giga_loss[loss=0.3516, simple_loss=0.4033, pruned_loss=0.1499, over 27672.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3693, pruned_loss=0.1223, over 5673771.38 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3601, pruned_loss=0.1132, over 5742012.62 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3703, pruned_loss=0.123, over 5661642.60 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:20:59,759 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-11 00:21:07,947 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-11 00:21:41,281 INFO [train.py:968] (0/2) Epoch 21, batch 26650, giga_loss[loss=0.2898, simple_loss=0.3582, pruned_loss=0.1107, over 28826.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3694, pruned_loss=0.1225, over 5674030.26 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3602, pruned_loss=0.1131, over 5744359.40 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3704, pruned_loss=0.1234, over 5659983.17 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:21:43,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.781e+03 2.225e+03 3.269e+03 8.037e+03, threshold=4.451e+03, percent-clipped=9.0
+2023-03-11 00:22:13,664 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=939037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:19,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=939040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:23,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=939045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:26,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=939048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:28,011 INFO [train.py:968] (0/2) Epoch 21, batch 26700, giga_loss[loss=0.3094, simple_loss=0.3795, pruned_loss=0.1197, over 28843.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3707, pruned_loss=0.1226, over 5672713.53 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3601, pruned_loss=0.1131, over 5744437.29 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3719, pruned_loss=0.1236, over 5659932.57 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:22:32,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3538, 2.7735, 1.4345, 1.4249], device='cuda:0'), covar=tensor([0.0928, 0.0382, 0.0867, 0.1314], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0558, 0.0385, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 00:22:45,669 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=939069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:52,312 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=939077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:58,918 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-11 00:23:17,177 INFO [train.py:968] (0/2) Epoch 21, batch 26750, giga_loss[loss=0.3148, simple_loss=0.3734, pruned_loss=0.1281, over 28896.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3724, pruned_loss=0.1236, over 5659549.90 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3605, pruned_loss=0.1134, over 5737048.08 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3731, pruned_loss=0.1242, over 5655784.13 frames. ], batch size: 112, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:23:18,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.659e+03 2.067e+03 2.876e+03 6.506e+03, threshold=4.133e+03, percent-clipped=6.0
+2023-03-11 00:24:03,450 INFO [train.py:968] (0/2) Epoch 21, batch 26800, giga_loss[loss=0.3582, simple_loss=0.4124, pruned_loss=0.152, over 29018.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3732, pruned_loss=0.1249, over 5663441.30 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1132, over 5738761.30 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3743, pruned_loss=0.1258, over 5657396.48 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 00:24:16,649 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=939166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:24:47,776 INFO [train.py:968] (0/2) Epoch 21, batch 26850, libri_loss[loss=0.2821, simple_loss=0.3556, pruned_loss=0.1043, over 29530.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3736, pruned_loss=0.122, over 5669817.95 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3599, pruned_loss=0.113, over 5731229.42 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3751, pruned_loss=0.1231, over 5669875.86 frames. ], batch size: 84, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 00:24:49,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.748e+02 1.558e+03 1.957e+03 2.456e+03 8.858e+03, threshold=3.915e+03, percent-clipped=3.0
+2023-03-11 00:25:34,179 INFO [train.py:968] (0/2) Epoch 21, batch 26900, giga_loss[loss=0.273, simple_loss=0.3582, pruned_loss=0.0939, over 29064.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3752, pruned_loss=0.1215, over 5662388.39 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3599, pruned_loss=0.113, over 5729468.62 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3764, pruned_loss=0.1225, over 5663503.63 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:26:10,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1769, 1.2464, 3.3172, 3.0515], device='cuda:0'), covar=tensor([0.1583, 0.2648, 0.0557, 0.1118], device='cuda:0'), in_proj_covar=tensor([0.0763, 0.0651, 0.0972, 0.0916], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 00:26:16,272 INFO [train.py:968] (0/2) Epoch 21, batch 26950, giga_loss[loss=0.2912, simple_loss=0.3676, pruned_loss=0.1074, over 28910.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3771, pruned_loss=0.1219, over 5671189.45 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.1131, over 5733661.90 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3782, pruned_loss=0.1228, over 5667227.94 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:26:18,592 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 00:26:20,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.647e+03 2.010e+03 2.836e+03 5.730e+03, threshold=4.019e+03, percent-clipped=7.0
+2023-03-11 00:27:07,944 INFO [train.py:968] (0/2) Epoch 21, batch 27000, giga_loss[loss=0.3367, simple_loss=0.3955, pruned_loss=0.1389, over 28226.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3804, pruned_loss=0.1254, over 5670678.27 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3603, pruned_loss=0.1132, over 5735575.36 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3814, pruned_loss=0.1262, over 5665459.61 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:27:07,955 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 00:27:16,795 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2056, simple_loss=0.3131, pruned_loss=0.04898, over 944034.00 frames. 
+2023-03-11 00:27:16,795 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 00:28:02,630 INFO [train.py:968] (0/2) Epoch 21, batch 27050, giga_loss[loss=0.3131, simple_loss=0.3793, pruned_loss=0.1234, over 28838.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3804, pruned_loss=0.126, over 5684055.13 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3607, pruned_loss=0.1135, over 5741442.55 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3815, pruned_loss=0.1268, over 5672446.59 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:28:06,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.356e+02 1.818e+03 2.667e+03 3.568e+03 1.097e+04, threshold=5.334e+03, percent-clipped=15.0
+2023-03-11 00:28:51,698 INFO [train.py:968] (0/2) Epoch 21, batch 27100, libri_loss[loss=0.314, simple_loss=0.3823, pruned_loss=0.1228, over 26433.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3808, pruned_loss=0.1274, over 5666449.82 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3613, pruned_loss=0.1139, over 5731490.45 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3815, pruned_loss=0.1279, over 5665100.23 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 1.0
+2023-03-11 00:29:13,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=939475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:29:36,447 INFO [train.py:968] (0/2) Epoch 21, batch 27150, giga_loss[loss=0.3128, simple_loss=0.3736, pruned_loss=0.126, over 27549.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3786, pruned_loss=0.1254, over 5651372.36 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1143, over 5706032.18 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3793, pruned_loss=0.1257, over 5671378.35 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 1.0
+2023-03-11 00:29:40,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.181e+03 1.923e+03 2.717e+03 4.276e+03 1.984e+04, threshold=5.434e+03, percent-clipped=18.0
+2023-03-11 00:30:12,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5987, 3.4362, 3.2906, 2.0220], device='cuda:0'), covar=tensor([0.0716, 0.0881, 0.0834, 0.1736], device='cuda:0'), in_proj_covar=tensor([0.1237, 0.1150, 0.0970, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 00:30:13,463 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=939541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:30:21,538 INFO [train.py:968] (0/2) Epoch 21, batch 27200, giga_loss[loss=0.3077, simple_loss=0.3818, pruned_loss=0.1168, over 28943.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3789, pruned_loss=0.124, over 5652641.31 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1144, over 5708049.90 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3795, pruned_loss=0.1243, over 5666072.02 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:30:30,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0941, 1.3452, 2.7358, 2.7271], device='cuda:0'), covar=tensor([0.1262, 0.2177, 0.0586, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0762, 0.0652, 0.0971, 0.0917], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 00:30:32,429 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.30 vs. limit=5.0
+2023-03-11 00:31:06,469 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-11 00:31:08,041 INFO [train.py:968] (0/2) Epoch 21, batch 27250, giga_loss[loss=0.2779, simple_loss=0.3607, pruned_loss=0.09755, over 28990.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3789, pruned_loss=0.1236, over 5654009.57 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.362, pruned_loss=0.1146, over 5710704.81 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3796, pruned_loss=0.1239, over 5661150.22 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:31:14,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.777e+02 1.417e+03 1.728e+03 2.534e+03 4.500e+03, threshold=3.457e+03, percent-clipped=0.0
+2023-03-11 00:31:43,511 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=939633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:31:59,392 INFO [train.py:968] (0/2) Epoch 21, batch 27300, giga_loss[loss=0.3322, simple_loss=0.3993, pruned_loss=0.1326, over 29016.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.38, pruned_loss=0.1249, over 5647768.30 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3619, pruned_loss=0.1146, over 5713698.32 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3809, pruned_loss=0.1253, over 5649667.19 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:32:32,198 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=939684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:32:35,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=939687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:32:46,411 INFO [train.py:968] (0/2) Epoch 21, batch 27350, libri_loss[loss=0.3102, simple_loss=0.3802, pruned_loss=0.1201, over 29241.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3781, pruned_loss=0.1238, over 5650642.63 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.362, pruned_loss=0.1147, over 5708001.72 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3791, pruned_loss=0.1242, over 5656902.45 frames. ], batch size: 94, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:32:49,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.109e+03 1.780e+03 2.116e+03 2.706e+03 8.498e+03, threshold=4.232e+03, percent-clipped=13.0
+2023-03-11 00:32:58,848 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=939716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:33:34,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 00:33:35,389 INFO [train.py:968] (0/2) Epoch 21, batch 27400, giga_loss[loss=0.28, simple_loss=0.3556, pruned_loss=0.1022, over 28723.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3757, pruned_loss=0.123, over 5656252.70 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3616, pruned_loss=0.1144, over 5710827.39 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.377, pruned_loss=0.1237, over 5658075.71 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:34:25,640 INFO [train.py:968] (0/2) Epoch 21, batch 27450, giga_loss[loss=0.3645, simple_loss=0.3965, pruned_loss=0.1663, over 28737.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3737, pruned_loss=0.1222, over 5665670.40 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3615, pruned_loss=0.1143, over 5711959.55 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.375, pruned_loss=0.1229, over 5665553.94 frames. ], batch size: 92, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:34:33,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.725e+03 2.343e+03 3.132e+03 7.951e+03, threshold=4.686e+03, percent-clipped=12.0
+2023-03-11 00:35:16,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=939850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:35:17,028 INFO [train.py:968] (0/2) Epoch 21, batch 27500, giga_loss[loss=0.3285, simple_loss=0.3897, pruned_loss=0.1336, over 27953.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.371, pruned_loss=0.1207, over 5663062.83 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3614, pruned_loss=0.1143, over 5712196.33 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3724, pruned_loss=0.1214, over 5661812.30 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:35:57,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8849, 1.8422, 2.0894, 1.6605], device='cuda:0'), covar=tensor([0.1795, 0.2485, 0.1417, 0.1698], device='cuda:0'), in_proj_covar=tensor([0.0895, 0.0703, 0.0943, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 00:36:02,001 INFO [train.py:968] (0/2) Epoch 21, batch 27550, giga_loss[loss=0.3844, simple_loss=0.4138, pruned_loss=0.1775, over 26684.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3705, pruned_loss=0.1211, over 5662880.67 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3616, pruned_loss=0.1144, over 5713007.47 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3716, pruned_loss=0.1216, over 5660374.61 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:36:06,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.835e+03 2.269e+03 3.075e+03 6.579e+03, threshold=4.537e+03, percent-clipped=10.0
+2023-03-11 00:36:42,073 INFO [train.py:968] (0/2) Epoch 21, batch 27600, giga_loss[loss=0.2525, simple_loss=0.3343, pruned_loss=0.08541, over 28915.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.371, pruned_loss=0.1222, over 5664342.13 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3621, pruned_loss=0.1148, over 5717693.00 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3718, pruned_loss=0.1226, over 5656361.45 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:36:51,967 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=939962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:37:21,148 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=939993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:37:24,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=939996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:37:28,221 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-940000.pt
+2023-03-11 00:37:29,344 INFO [train.py:968] (0/2) Epoch 21, batch 27650, giga_loss[loss=0.2495, simple_loss=0.3282, pruned_loss=0.0854, over 28824.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3678, pruned_loss=0.1183, over 5665574.76 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.362, pruned_loss=0.1148, over 5718629.53 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3685, pruned_loss=0.1186, over 5658340.59 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:37:33,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.475e+02 1.616e+03 2.010e+03 2.764e+03 9.999e+03, threshold=4.020e+03, percent-clipped=5.0
+2023-03-11 00:37:34,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4545, 1.8515, 1.0849, 1.5079], device='cuda:0'), covar=tensor([0.1152, 0.0675, 0.1321, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0447, 0.0516, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 00:37:34,770 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:37:51,062 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:38:15,515 INFO [train.py:968] (0/2) Epoch 21, batch 27700, giga_loss[loss=0.2451, simple_loss=0.3113, pruned_loss=0.08942, over 23938.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3658, pruned_loss=0.1164, over 5657049.91 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3623, pruned_loss=0.1149, over 5710798.34 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3661, pruned_loss=0.1166, over 5656825.02 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:39:06,992 INFO [train.py:968] (0/2) Epoch 21, batch 27750, giga_loss[loss=0.2826, simple_loss=0.3503, pruned_loss=0.1074, over 29010.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3656, pruned_loss=0.1167, over 5658652.46 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3621, pruned_loss=0.1147, over 5714721.62 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3661, pruned_loss=0.117, over 5653427.37 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:39:12,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.253e+02 1.511e+03 2.044e+03 2.993e+03 6.969e+03, threshold=4.088e+03, percent-clipped=12.0
+2023-03-11 00:39:30,565 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-11 00:39:50,954 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:39:59,206 INFO [train.py:968] (0/2) Epoch 21, batch 27800, giga_loss[loss=0.2988, simple_loss=0.3513, pruned_loss=0.1232, over 23396.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3629, pruned_loss=0.1163, over 5654362.74 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3622, pruned_loss=0.1148, over 5718835.24 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3633, pruned_loss=0.1165, over 5645366.75 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:39:59,464 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:40:04,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:40:34,534 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:40:52,420 INFO [train.py:968] (0/2) Epoch 21, batch 27850, giga_loss[loss=0.2744, simple_loss=0.3488, pruned_loss=0.1, over 28397.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.363, pruned_loss=0.1169, over 5654207.49 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3624, pruned_loss=0.115, over 5716970.65 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3631, pruned_loss=0.1169, over 5647930.17 frames. ], batch size: 71, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:40:57,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.215e+02 1.740e+03 2.260e+03 3.215e+03 1.205e+04, threshold=4.520e+03, percent-clipped=13.0
+2023-03-11 00:41:01,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4757, 1.6534, 1.3155, 1.5105], device='cuda:0'), covar=tensor([0.2732, 0.2723, 0.3111, 0.2348], device='cuda:0'), in_proj_covar=tensor([0.1496, 0.1085, 0.1319, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 00:41:38,452 INFO [train.py:968] (0/2) Epoch 21, batch 27900, giga_loss[loss=0.3386, simple_loss=0.3941, pruned_loss=0.1416, over 28674.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3644, pruned_loss=0.1167, over 5663415.82 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3624, pruned_loss=0.1149, over 5719553.03 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3646, pruned_loss=0.1169, over 5654738.63 frames. ], batch size: 284, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:42:23,142 INFO [train.py:968] (0/2) Epoch 21, batch 27950, giga_loss[loss=0.2916, simple_loss=0.3671, pruned_loss=0.108, over 28884.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3664, pruned_loss=0.1177, over 5665607.66 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3621, pruned_loss=0.1148, over 5726749.76 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3669, pruned_loss=0.1181, over 5649655.37 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:42:24,666 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:42:28,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.435e+02 1.613e+03 2.013e+03 2.660e+03 7.486e+03, threshold=4.026e+03, percent-clipped=4.0
+2023-03-11 00:42:46,743 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:42:57,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:43:09,539 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 00:43:09,714 INFO [train.py:968] (0/2) Epoch 21, batch 28000, giga_loss[loss=0.3049, simple_loss=0.3608, pruned_loss=0.1245, over 28967.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3662, pruned_loss=0.1179, over 5654602.51 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3623, pruned_loss=0.115, over 5720629.53 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3665, pruned_loss=0.1181, over 5645993.78 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 00:43:53,489 INFO [train.py:968] (0/2) Epoch 21, batch 28050, giga_loss[loss=0.2971, simple_loss=0.367, pruned_loss=0.1136, over 28777.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3653, pruned_loss=0.1177, over 5656736.05 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3618, pruned_loss=0.1145, over 5726159.71 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3661, pruned_loss=0.1183, over 5643049.23 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:43:58,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.701e+02 1.608e+03 2.042e+03 2.755e+03 6.683e+03, threshold=4.084e+03, percent-clipped=9.0
+2023-03-11 00:44:20,734 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:44:35,015 INFO [train.py:968] (0/2) Epoch 21, batch 28100, giga_loss[loss=0.3401, simple_loss=0.3999, pruned_loss=0.1402, over 29008.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3671, pruned_loss=0.119, over 5636850.44 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.362, pruned_loss=0.1146, over 5711274.97 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3678, pruned_loss=0.1196, over 5638208.72 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:45:01,738 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:45:05,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:45:14,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940494.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 00:45:21,546 INFO [train.py:968] (0/2) Epoch 21, batch 28150, giga_loss[loss=0.2938, simple_loss=0.3628, pruned_loss=0.1124, over 28711.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3698, pruned_loss=0.1203, over 5645667.78 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3623, pruned_loss=0.1148, over 5709898.25 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3702, pruned_loss=0.1207, over 5646538.61 frames. ], batch size: 242, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:45:24,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3474, 3.0938, 1.5025, 1.4287], device='cuda:0'), covar=tensor([0.0948, 0.0351, 0.0852, 0.1311], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0557, 0.0385, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 00:45:27,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.252e+03 1.690e+03 2.132e+03 2.758e+03 7.950e+03, threshold=4.264e+03, percent-clipped=12.0
+2023-03-11 00:45:31,698 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:45:39,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:45:52,700 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 00:46:11,664 INFO [train.py:968] (0/2) Epoch 21, batch 28200, giga_loss[loss=0.4075, simple_loss=0.4412, pruned_loss=0.1869, over 28592.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1223, over 5645035.94 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3623, pruned_loss=0.1147, over 5711590.27 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3721, pruned_loss=0.1227, over 5643721.19 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:46:44,316 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:46:55,960 INFO [train.py:968] (0/2) Epoch 21, batch 28250, giga_loss[loss=0.3367, simple_loss=0.3882, pruned_loss=0.1426, over 27552.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3722, pruned_loss=0.1231, over 5649841.61 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3619, pruned_loss=0.1146, over 5717563.44 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3731, pruned_loss=0.1238, over 5641819.10 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:47:03,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.122e+03 1.698e+03 2.351e+03 3.095e+03 1.202e+04, threshold=4.701e+03, percent-clipped=9.0
+2023-03-11 00:47:36,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4643, 1.6226, 1.6646, 1.4257], device='cuda:0'), covar=tensor([0.1885, 0.2024, 0.2344, 0.2113], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0752, 0.0719, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 00:47:40,934 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-11 00:47:47,927 INFO [train.py:968] (0/2) Epoch 21, batch 28300, giga_loss[loss=0.289, simple_loss=0.3616, pruned_loss=0.1082, over 28890.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3734, pruned_loss=0.1226, over 5651280.21 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.362, pruned_loss=0.1146, over 5718383.24 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3742, pruned_loss=0.1232, over 5643418.97 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:48:03,016 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:05,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:08,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3492, 1.4475, 3.9267, 3.2675], device='cuda:0'), covar=tensor([0.1687, 0.2651, 0.0493, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0759, 0.0649, 0.0961, 0.0909], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 00:48:12,918 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-11 00:48:14,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:33,275 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:33,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:39,214 INFO [train.py:968] (0/2) Epoch 21, batch 28350, giga_loss[loss=0.3027, simple_loss=0.3719, pruned_loss=0.1168, over 28625.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3728, pruned_loss=0.1217, over 5655944.69 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3617, pruned_loss=0.1145, over 5721339.27 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3739, pruned_loss=0.1224, over 5646271.46 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:48:40,451 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:45,241 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.248e+03 1.880e+03 2.306e+03 3.041e+03 9.214e+03, threshold=4.613e+03, percent-clipped=4.0
+2023-03-11 00:49:04,335 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-11 00:49:30,957 INFO [train.py:968] (0/2) Epoch 21, batch 28400, giga_loss[loss=0.3064, simple_loss=0.3689, pruned_loss=0.122, over 27592.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3733, pruned_loss=0.1231, over 5639595.24 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3621, pruned_loss=0.1147, over 5723786.61 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.374, pruned_loss=0.1236, over 5628975.90 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:50:22,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5102, 1.7123, 1.1479, 1.3416], device='cuda:0'), covar=tensor([0.1076, 0.0691, 0.1193, 0.1234], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0449, 0.0517, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 00:50:22,779 INFO [train.py:968] (0/2) Epoch 21, batch 28450, giga_loss[loss=0.3, simple_loss=0.3627, pruned_loss=0.1186, over 28648.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3725, pruned_loss=0.1231, over 5642274.56 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.362, pruned_loss=0.1147, over 5726694.01 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3734, pruned_loss=0.1237, over 5628430.44 frames. ], batch size: 242, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:50:30,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.759e+03 2.361e+03 3.493e+03 8.024e+03, threshold=4.722e+03, percent-clipped=12.0
+2023-03-11 00:50:33,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:50:46,912 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:50:49,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:51:15,005 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:51:16,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:51:18,509 INFO [train.py:968] (0/2) Epoch 21, batch 28500, giga_loss[loss=0.2977, simple_loss=0.3622, pruned_loss=0.1166, over 28948.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3718, pruned_loss=0.1238, over 5628750.06 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3624, pruned_loss=0.1151, over 5725181.47 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3723, pruned_loss=0.1241, over 5617773.00 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:51:19,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:51:34,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940869.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 00:51:43,434 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:52:02,948 INFO [train.py:968] (0/2) Epoch 21, batch 28550, giga_loss[loss=0.2892, simple_loss=0.3556, pruned_loss=0.1114, over 28912.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3705, pruned_loss=0.1232, over 5631482.68 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1152, over 5708911.02 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.371, pruned_loss=0.1235, over 5634116.59 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:52:12,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.078e+03 1.654e+03 2.185e+03 3.221e+03 8.662e+03, threshold=4.370e+03, percent-clipped=10.0
+2023-03-11 00:52:13,706 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8834, 1.8959, 1.3693, 1.6012], device='cuda:0'), covar=tensor([0.1093, 0.0848, 0.1241, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0450, 0.0518, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 00:52:50,415 INFO [train.py:968] (0/2) Epoch 21, batch 28600, giga_loss[loss=0.3139, simple_loss=0.3749, pruned_loss=0.1264, over 28924.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3696, pruned_loss=0.1226, over 5635856.72 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3624, pruned_loss=0.115, over 5703878.05 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3703, pruned_loss=0.1232, over 5640860.76 frames. ], batch size: 112, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:52:52,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:52:54,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:53:02,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:53:19,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:53:26,765 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:53:31,296 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-11 00:53:31,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3010, 1.5887, 0.9687, 1.1325], device='cuda:0'), covar=tensor([0.1156, 0.0678, 0.1457, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0449, 0.0517, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 00:53:32,140 INFO [train.py:968] (0/2) Epoch 21, batch 28650, giga_loss[loss=0.3433, simple_loss=0.4027, pruned_loss=0.1419, over 28846.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3694, pruned_loss=0.1224, over 5642941.15 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1151, over 5703123.80 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3701, pruned_loss=0.1231, over 5644510.21 frames. ], batch size: 285, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:53:38,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.162e+02 1.742e+03 2.295e+03 3.232e+03 9.913e+03, threshold=4.589e+03, percent-clipped=9.0
+2023-03-11 00:53:42,830 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941012.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 00:53:45,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941015.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 00:54:13,621 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941044.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 00:54:17,915 INFO [train.py:968] (0/2) Epoch 21, batch 28700, giga_loss[loss=0.2774, simple_loss=0.3527, pruned_loss=0.1011, over 28934.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3695, pruned_loss=0.1225, over 5653868.10 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3618, pruned_loss=0.1145, over 5710518.97 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.371, pruned_loss=0.124, over 5646450.39 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:54:19,896 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:54:36,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:54:53,439 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:55:08,524 INFO [train.py:968] (0/2) Epoch 21, batch 28750, giga_loss[loss=0.3393, simple_loss=0.3973, pruned_loss=0.1406, over 28828.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3724, pruned_loss=0.1245, over 5654133.67 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3618, pruned_loss=0.1145, over 5710518.97 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3736, pruned_loss=0.1256, over 5648360.37 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:55:15,707 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:55:15,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.699e+03 2.215e+03 3.477e+03 8.525e+03, threshold=4.430e+03, percent-clipped=9.0
+2023-03-11 00:55:17,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:55:41,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5074, 3.5252, 1.5760, 1.7534], device='cuda:0'), covar=tensor([0.0948, 0.0340, 0.0861, 0.1204], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0558, 0.0385, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 00:55:47,119 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:55:55,601 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-11 00:55:56,911 INFO [train.py:968] (0/2) Epoch 21, batch 28800, giga_loss[loss=0.3367, simple_loss=0.3989, pruned_loss=0.1372, over 28621.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3725, pruned_loss=0.1249, over 5663896.30 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3615, pruned_loss=0.1144, over 5712451.11 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3739, pruned_loss=0.1261, over 5657048.65 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 00:56:42,199 INFO [train.py:968] (0/2) Epoch 21, batch 28850, giga_loss[loss=0.273, simple_loss=0.3434, pruned_loss=0.1013, over 28802.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3726, pruned_loss=0.1252, over 5668416.86 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3614, pruned_loss=0.1145, over 5714399.86 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3739, pruned_loss=0.1261, over 5660925.25 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:56:49,114 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.666e+03 2.116e+03 2.745e+03 7.305e+03, threshold=4.232e+03, percent-clipped=5.0
+2023-03-11 00:56:54,984 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:56:56,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:56:56,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:57:11,325 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:57:20,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9478, 2.0841, 1.8025, 2.1068], device='cuda:0'), covar=tensor([0.2596, 0.2806, 0.3051, 0.2516], device='cuda:0'), in_proj_covar=tensor([0.1503, 0.1091, 0.1327, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 00:57:24,136 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:57:29,975 INFO [train.py:968] (0/2) Epoch 21, batch 28900, giga_loss[loss=0.2532, simple_loss=0.3357, pruned_loss=0.0854, over 28342.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3727, pruned_loss=0.1243, over 5679557.03 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3616, pruned_loss=0.1145, over 5715492.70 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3738, pruned_loss=0.1253, over 5671949.52 frames. ], batch size: 65, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:57:35,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:58:14,801 INFO [train.py:968] (0/2) Epoch 21, batch 28950, giga_loss[loss=0.2826, simple_loss=0.3518, pruned_loss=0.1067, over 28663.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3738, pruned_loss=0.125, over 5678333.79 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3612, pruned_loss=0.1142, over 5722058.69 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3754, pruned_loss=0.1264, over 5664739.33 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:58:15,361 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.84 vs. limit=5.0
+2023-03-11 00:58:23,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.945e+02 1.752e+03 2.446e+03 3.391e+03 1.494e+04, threshold=4.891e+03, percent-clipped=17.0
+2023-03-11 00:58:43,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4113, 1.5791, 1.4450, 1.5565], device='cuda:0'), covar=tensor([0.0780, 0.0333, 0.0320, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 00:59:02,001 INFO [train.py:968] (0/2) Epoch 21, batch 29000, giga_loss[loss=0.3925, simple_loss=0.4203, pruned_loss=0.1823, over 26486.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3749, pruned_loss=0.1258, over 5670624.11 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1144, over 5712292.64 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3762, pruned_loss=0.1269, over 5667672.87 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:59:15,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:59:17,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5621, 4.3979, 4.2399, 2.1689], device='cuda:0'), covar=tensor([0.0559, 0.0691, 0.0691, 0.2124], device='cuda:0'), in_proj_covar=tensor([0.1241, 0.1151, 0.0974, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 00:59:44,834 INFO [train.py:968] (0/2) Epoch 21, batch 29050, giga_loss[loss=0.3745, simple_loss=0.4145, pruned_loss=0.1672, over 27524.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3766, pruned_loss=0.1273, over 5664779.75 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3618, pruned_loss=0.1146, over 5708395.38 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3777, pruned_loss=0.1282, over 5664582.59 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:59:50,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.162e+03 1.658e+03 2.057e+03 2.663e+03 4.230e+03, threshold=4.115e+03, percent-clipped=0.0
+2023-03-11 01:00:10,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:00:27,921 INFO [train.py:968] (0/2) Epoch 21, batch 29100, giga_loss[loss=0.2832, simple_loss=0.359, pruned_loss=0.1037, over 29117.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.1271, over 5669599.69 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3615, pruned_loss=0.1144, over 5714419.90 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3777, pruned_loss=0.1285, over 5662771.90 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:00:39,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:01:14,345 INFO [train.py:968] (0/2) Epoch 21, batch 29150, giga_loss[loss=0.2782, simple_loss=0.3583, pruned_loss=0.09908, over 28818.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3757, pruned_loss=0.1257, over 5659592.07 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3619, pruned_loss=0.1147, over 5713232.73 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3767, pruned_loss=0.1266, over 5654919.77 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:01:24,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.685e+03 2.149e+03 3.369e+03 6.510e+03, threshold=4.299e+03, percent-clipped=15.0
+2023-03-11 01:01:26,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:01:28,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:01:58,411 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:05,900 INFO [train.py:968] (0/2) Epoch 21, batch 29200, giga_loss[loss=0.3025, simple_loss=0.3712, pruned_loss=0.117, over 28519.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3752, pruned_loss=0.1249, over 5652588.26 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1144, over 5714761.17 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3768, pruned_loss=0.1261, over 5646673.68 frames. ], batch size: 65, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:02:13,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7822, 1.8380, 2.0283, 1.5711], device='cuda:0'), covar=tensor([0.1883, 0.2676, 0.1512, 0.1842], device='cuda:0'), in_proj_covar=tensor([0.0897, 0.0704, 0.0945, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 01:02:14,882 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:25,398 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:28,333 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:30,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:42,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:42,715 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-11 01:02:51,775 INFO [train.py:968] (0/2) Epoch 21, batch 29250, giga_loss[loss=0.279, simple_loss=0.3538, pruned_loss=0.1021, over 28909.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3742, pruned_loss=0.124, over 5655528.50 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.361, pruned_loss=0.1143, over 5719004.35 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3759, pruned_loss=0.1252, over 5645857.85 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:02:54,199 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:54,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:55,605 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:58,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:58,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.047e+02 1.543e+03 2.013e+03 2.536e+03 7.241e+03, threshold=4.026e+03, percent-clipped=6.0
+2023-03-11 01:03:19,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:03:24,603 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:03:34,208 INFO [train.py:968] (0/2) Epoch 21, batch 29300, giga_loss[loss=0.3262, simple_loss=0.3866, pruned_loss=0.1329, over 28574.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1226, over 5671017.97 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.361, pruned_loss=0.1142, over 5725036.91 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3743, pruned_loss=0.1239, over 5655847.77 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:04:02,885 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:04:14,692 INFO [train.py:968] (0/2) Epoch 21, batch 29350, giga_loss[loss=0.3138, simple_loss=0.3822, pruned_loss=0.1227, over 27875.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3718, pruned_loss=0.1223, over 5674511.97 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3612, pruned_loss=0.1145, over 5727831.00 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1235, over 5657562.16 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:04:25,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.995e+02 1.702e+03 2.326e+03 3.147e+03 9.923e+03, threshold=4.652e+03, percent-clipped=13.0
+2023-03-11 01:04:36,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2105, 1.4621, 1.4804, 1.2580], device='cuda:0'), covar=tensor([0.1893, 0.1737, 0.2344, 0.1990], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0754, 0.0718, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 01:04:45,405 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:04:47,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:00,755 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:04,094 INFO [train.py:968] (0/2) Epoch 21, batch 29400, giga_loss[loss=0.313, simple_loss=0.3818, pruned_loss=0.1221, over 28925.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3744, pruned_loss=0.1243, over 5660278.15 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1148, over 5718852.31 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3756, pruned_loss=0.1252, over 5652455.36 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:05:04,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:18,414 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:26,561 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:28,801 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:31,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:43,174 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 01:05:51,433 INFO [train.py:968] (0/2) Epoch 21, batch 29450, giga_loss[loss=0.2869, simple_loss=0.3588, pruned_loss=0.1075, over 28997.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3734, pruned_loss=0.124, over 5664125.78 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3616, pruned_loss=0.1148, over 5720918.89 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3745, pruned_loss=0.1248, over 5655585.84 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:05:59,550 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:06:03,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.388e+02 1.750e+03 2.460e+03 3.488e+03 1.460e+04, threshold=4.920e+03, percent-clipped=15.0
+2023-03-11 01:06:36,920 INFO [train.py:968] (0/2) Epoch 21, batch 29500, giga_loss[loss=0.3132, simple_loss=0.3786, pruned_loss=0.1239, over 28863.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3732, pruned_loss=0.1248, over 5669544.53 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3616, pruned_loss=0.1149, over 5725317.33 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3745, pruned_loss=0.1257, over 5656566.76 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:07:12,596 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:07:26,290 INFO [train.py:968] (0/2) Epoch 21, batch 29550, giga_loss[loss=0.2807, simple_loss=0.3542, pruned_loss=0.1036, over 29077.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3751, pruned_loss=0.1262, over 5668664.87 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3616, pruned_loss=0.1149, over 5720986.56 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3763, pruned_loss=0.127, over 5661473.89 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:07:34,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.584e+03 2.104e+03 3.073e+03 7.144e+03, threshold=4.209e+03, percent-clipped=9.0
+2023-03-11 01:07:36,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3657, 1.4960, 1.2655, 1.0513], device='cuda:0'), covar=tensor([0.1000, 0.0566, 0.0984, 0.1152], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0449, 0.0516, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 01:07:37,699 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 01:07:57,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:07:59,712 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-11 01:08:06,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:08:14,476 INFO [train.py:968] (0/2) Epoch 21, batch 29600, giga_loss[loss=0.3451, simple_loss=0.3784, pruned_loss=0.1559, over 23515.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3765, pruned_loss=0.1275, over 5652502.52 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3616, pruned_loss=0.115, over 5722108.79 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3777, pruned_loss=0.1282, over 5645124.15 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:08:18,346 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:09:01,106 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-942000.pt
+2023-03-11 01:09:02,091 INFO [train.py:968] (0/2) Epoch 21, batch 29650, libri_loss[loss=0.3134, simple_loss=0.3793, pruned_loss=0.1238, over 27792.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3771, pruned_loss=0.1278, over 5656930.76 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3619, pruned_loss=0.1152, over 5724428.81 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3781, pruned_loss=0.1285, over 5647556.46 frames. ], batch size: 116, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:09:11,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.184e+03 1.699e+03 2.259e+03 2.970e+03 6.789e+03, threshold=4.518e+03, percent-clipped=8.0
+2023-03-11 01:09:50,611 INFO [train.py:968] (0/2) Epoch 21, batch 29700, giga_loss[loss=0.2783, simple_loss=0.3515, pruned_loss=0.1026, over 28851.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3758, pruned_loss=0.1265, over 5644881.27 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3619, pruned_loss=0.1152, over 5716859.73 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3767, pruned_loss=0.1271, over 5643989.07 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:09:53,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6447, 1.6941, 1.8832, 1.4618], device='cuda:0'), covar=tensor([0.1921, 0.2509, 0.1513, 0.1790], device='cuda:0'), in_proj_covar=tensor([0.0896, 0.0704, 0.0944, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 01:09:59,446 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:05,599 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=942069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:13,367 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:18,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:24,361 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:26,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:38,280 INFO [train.py:968] (0/2) Epoch 21, batch 29750, giga_loss[loss=0.2837, simple_loss=0.3602, pruned_loss=0.1036, over 28994.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3752, pruned_loss=0.1248, over 5640214.21 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3622, pruned_loss=0.1153, over 5699790.00 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3759, pruned_loss=0.1254, over 5653718.61 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:10:44,111 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 01:10:45,480 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:45,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.588e+03 2.126e+03 3.011e+03 9.909e+03, threshold=4.252e+03, percent-clipped=11.0
+2023-03-11 01:10:53,654 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:11:24,958 INFO [train.py:968] (0/2) Epoch 21, batch 29800, giga_loss[loss=0.3121, simple_loss=0.3722, pruned_loss=0.126, over 29026.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3741, pruned_loss=0.1239, over 5640207.25 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3625, pruned_loss=0.1154, over 5700077.40 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3747, pruned_loss=0.1245, over 5649348.86 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:12:06,180 INFO [train.py:968] (0/2) Epoch 21, batch 29850, giga_loss[loss=0.3291, simple_loss=0.3808, pruned_loss=0.1387, over 27450.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3738, pruned_loss=0.1238, over 5661456.97 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1157, over 5706597.54 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3742, pruned_loss=0.1243, over 5660834.70 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:12:09,227 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:12:11,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:12:14,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.205e+03 1.674e+03 2.054e+03 3.424e+03 7.914e+03, threshold=4.108e+03, percent-clipped=14.0
+2023-03-11 01:12:33,978 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8540, 1.8375, 1.8241, 1.6280], device='cuda:0'), covar=tensor([0.1702, 0.2229, 0.2202, 0.2170], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0755, 0.0718, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 01:12:38,598 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:12:51,683 INFO [train.py:968] (0/2) Epoch 21, batch 29900, giga_loss[loss=0.2696, simple_loss=0.3429, pruned_loss=0.09813, over 28644.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3718, pruned_loss=0.1227, over 5664264.30 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3632, pruned_loss=0.1158, over 5707546.23 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3721, pruned_loss=0.1232, over 5662138.53 frames. ], batch size: 242, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:13:02,764 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:13:40,452 INFO [train.py:968] (0/2) Epoch 21, batch 29950, giga_loss[loss=0.2653, simple_loss=0.3324, pruned_loss=0.09908, over 28821.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3681, pruned_loss=0.1213, over 5647731.51 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3636, pruned_loss=0.1162, over 5699791.30 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3681, pruned_loss=0.1215, over 5651822.12 frames. ], batch size: 66, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:13:50,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.155e+03 1.713e+03 2.339e+03 3.303e+03 7.678e+03, threshold=4.678e+03, percent-clipped=14.0
+2023-03-11 01:14:07,829 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:14:24,953 INFO [train.py:968] (0/2) Epoch 21, batch 30000, giga_loss[loss=0.3006, simple_loss=0.3697, pruned_loss=0.1157, over 28874.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3664, pruned_loss=0.1212, over 5651160.89 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3637, pruned_loss=0.1162, over 5700622.72 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3664, pruned_loss=0.1214, over 5652510.33 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:14:24,957 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 01:14:33,836 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2068, simple_loss=0.3154, pruned_loss=0.04915, over 944034.00 frames. 
+2023-03-11 01:14:33,837 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 01:14:50,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3714, 1.6572, 1.5026, 1.4909], device='cuda:0'), covar=tensor([0.0753, 0.0320, 0.0313, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 01:15:13,805 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=942391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:15:13,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7291, 1.8571, 1.7965, 1.6892], device='cuda:0'), covar=tensor([0.1890, 0.2158, 0.2320, 0.2149], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0755, 0.0720, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 01:15:22,417 INFO [train.py:968] (0/2) Epoch 21, batch 30050, giga_loss[loss=0.3335, simple_loss=0.3904, pruned_loss=0.1383, over 27918.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3645, pruned_loss=0.1207, over 5649526.97 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3636, pruned_loss=0.1161, over 5702566.08 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3646, pruned_loss=0.121, over 5648416.59 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:15:28,961 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 01:15:29,533 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:15:32,563 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:15:34,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.789e+02 1.705e+03 2.111e+03 2.580e+03 5.761e+03, threshold=4.221e+03, percent-clipped=4.0
+2023-03-11 01:15:50,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5128, 3.3984, 1.5987, 1.6344], device='cuda:0'), covar=tensor([0.0953, 0.0386, 0.0909, 0.1290], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0562, 0.0387, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 01:15:58,804 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:16:02,034 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-11 01:16:03,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:16:11,274 INFO [train.py:968] (0/2) Epoch 21, batch 30100, giga_loss[loss=0.2772, simple_loss=0.3585, pruned_loss=0.09793, over 28631.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3643, pruned_loss=0.1196, over 5637767.11 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3641, pruned_loss=0.1164, over 5696742.74 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3639, pruned_loss=0.1196, over 5639629.63 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:16:30,993 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:16:35,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:16:48,629 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-11 01:17:00,577 INFO [train.py:968] (0/2) Epoch 21, batch 30150, giga_loss[loss=0.2848, simple_loss=0.3616, pruned_loss=0.104, over 28294.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3613, pruned_loss=0.1154, over 5635887.84 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3633, pruned_loss=0.116, over 5693238.07 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3617, pruned_loss=0.1159, over 5640119.56 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:17:06,090 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:17:15,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.760e+03 2.254e+03 3.053e+03 7.105e+03, threshold=4.508e+03, percent-clipped=10.0
+2023-03-11 01:17:48,072 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=942545.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 01:17:53,263 INFO [train.py:968] (0/2) Epoch 21, batch 30200, giga_loss[loss=0.2937, simple_loss=0.3587, pruned_loss=0.1144, over 27915.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1127, over 5639483.79 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.363, pruned_loss=0.1159, over 5695824.00 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1132, over 5639526.36 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:18:27,548 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:18:29,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:18:43,564 INFO [train.py:968] (0/2) Epoch 21, batch 30250, giga_loss[loss=0.2675, simple_loss=0.3433, pruned_loss=0.09586, over 27546.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3571, pruned_loss=0.1094, over 5639081.75 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3631, pruned_loss=0.1159, over 5693508.52 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3574, pruned_loss=0.1097, over 5640294.23 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:18:55,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.929e+02 1.472e+03 1.784e+03 2.474e+03 7.161e+03, threshold=3.569e+03, percent-clipped=3.0
+2023-03-11 01:19:01,959 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:19:19,163 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=942637.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 01:19:32,743 INFO [train.py:968] (0/2) Epoch 21, batch 30300, giga_loss[loss=0.2455, simple_loss=0.3348, pruned_loss=0.07815, over 29106.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3533, pruned_loss=0.1057, over 5644453.29 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3628, pruned_loss=0.1158, over 5694703.98 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3537, pruned_loss=0.106, over 5644012.35 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:20:08,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2662, 1.2296, 3.6205, 3.1952], device='cuda:0'), covar=tensor([0.1632, 0.2864, 0.0500, 0.0998], device='cuda:0'), in_proj_covar=tensor([0.0763, 0.0651, 0.0966, 0.0909], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 01:20:21,831 INFO [train.py:968] (0/2) Epoch 21, batch 30350, giga_loss[loss=0.3138, simple_loss=0.3882, pruned_loss=0.1197, over 28542.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3522, pruned_loss=0.1026, over 5656696.25 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3621, pruned_loss=0.1154, over 5689909.31 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.353, pruned_loss=0.1028, over 5660319.99 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:20:31,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.516e+02 1.371e+03 1.770e+03 2.471e+03 6.873e+03, threshold=3.539e+03, percent-clipped=11.0
+2023-03-11 01:21:11,140 INFO [train.py:968] (0/2) Epoch 21, batch 30400, giga_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08923, over 29029.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3527, pruned_loss=0.1023, over 5654731.29 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3619, pruned_loss=0.1154, over 5684974.48 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3533, pruned_loss=0.1023, over 5661302.21 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:21:30,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:22:05,806 INFO [train.py:968] (0/2) Epoch 21, batch 30450, giga_loss[loss=0.2725, simple_loss=0.3249, pruned_loss=0.1101, over 24026.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3517, pruned_loss=0.1013, over 5653091.87 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3619, pruned_loss=0.1154, over 5684382.61 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3521, pruned_loss=0.1012, over 5658578.36 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:22:19,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.835e+02 1.283e+03 1.673e+03 2.109e+03 5.079e+03, threshold=3.346e+03, percent-clipped=8.0
+2023-03-11 01:22:41,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5424, 1.8112, 1.4590, 1.5776], device='cuda:0'), covar=tensor([0.2620, 0.2458, 0.2785, 0.2289], device='cuda:0'), in_proj_covar=tensor([0.1499, 0.1083, 0.1325, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 01:22:54,633 INFO [train.py:968] (0/2) Epoch 21, batch 30500, giga_loss[loss=0.2354, simple_loss=0.3218, pruned_loss=0.0745, over 29031.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.349, pruned_loss=0.09965, over 5661882.14 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3621, pruned_loss=0.1159, over 5689058.96 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3488, pruned_loss=0.09879, over 5661220.64 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:22:59,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8857, 2.0998, 1.6774, 2.1647], device='cuda:0'), covar=tensor([0.2699, 0.2661, 0.3109, 0.2358], device='cuda:0'), in_proj_covar=tensor([0.1500, 0.1083, 0.1326, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 01:23:42,297 INFO [train.py:968] (0/2) Epoch 21, batch 30550, giga_loss[loss=0.2537, simple_loss=0.3381, pruned_loss=0.08468, over 28894.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3476, pruned_loss=0.09944, over 5660244.05 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3614, pruned_loss=0.1156, over 5696559.92 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3477, pruned_loss=0.09842, over 5652012.22 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:23:49,887 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:23:53,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:23:53,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.630e+02 1.549e+03 2.005e+03 3.018e+03 7.741e+03, threshold=4.009e+03, percent-clipped=19.0
+2023-03-11 01:24:01,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942920.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:24:17,727 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:24:27,744 INFO [train.py:968] (0/2) Epoch 21, batch 30600, giga_loss[loss=0.3109, simple_loss=0.3623, pruned_loss=0.1297, over 26739.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3475, pruned_loss=0.09919, over 5665445.36 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3608, pruned_loss=0.1155, over 5698171.16 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3478, pruned_loss=0.09813, over 5656620.36 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:24:44,134 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-11 01:25:16,959 INFO [train.py:968] (0/2) Epoch 21, batch 30650, giga_loss[loss=0.2346, simple_loss=0.3185, pruned_loss=0.0753, over 28016.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3454, pruned_loss=0.09735, over 5664774.00 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3606, pruned_loss=0.1154, over 5700921.75 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3456, pruned_loss=0.09629, over 5654922.39 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:25:27,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=943012.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:25:28,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.898e+02 1.368e+03 1.774e+03 2.211e+03 5.344e+03, threshold=3.547e+03, percent-clipped=3.0
+2023-03-11 01:26:06,923 INFO [train.py:968] (0/2) Epoch 21, batch 30700, giga_loss[loss=0.244, simple_loss=0.3162, pruned_loss=0.08594, over 27714.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3424, pruned_loss=0.09489, over 5665230.37 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3604, pruned_loss=0.1153, over 5702361.25 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3425, pruned_loss=0.09387, over 5655591.34 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:26:20,803 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=943063.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:26:22,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=943066.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 01:26:49,527 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=943095.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 01:26:53,121 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 01:26:56,208 INFO [train.py:968] (0/2) Epoch 21, batch 30750, giga_loss[loss=0.2691, simple_loss=0.3274, pruned_loss=0.1053, over 26746.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3392, pruned_loss=0.0931, over 5676197.39 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3597, pruned_loss=0.1151, over 5705427.28 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3394, pruned_loss=0.092, over 5664921.72 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:27:08,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.383e+02 1.243e+03 1.839e+03 3.116e+03 8.310e+03, threshold=3.677e+03, percent-clipped=19.0
+2023-03-11 01:27:18,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4801, 3.2346, 1.5407, 1.4961], device='cuda:0'), covar=tensor([0.0929, 0.0344, 0.0964, 0.1319], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0555, 0.0384, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 01:27:43,214 INFO [train.py:968] (0/2) Epoch 21, batch 30800, giga_loss[loss=0.2892, simple_loss=0.3705, pruned_loss=0.104, over 28908.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3367, pruned_loss=0.09234, over 5675153.59 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3592, pruned_loss=0.1149, over 5706410.74 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3369, pruned_loss=0.09113, over 5664462.53 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:27:46,368 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=943155.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:27:49,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=943158.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 01:28:01,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2870, 1.3284, 3.5828, 3.0466], device='cuda:0'), covar=tensor([0.1696, 0.2836, 0.0504, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0649, 0.0963, 0.0904], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 01:28:18,866 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=943187.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:28:30,384 INFO [train.py:968] (0/2) Epoch 21, batch 30850, giga_loss[loss=0.2471, simple_loss=0.3335, pruned_loss=0.0803, over 28847.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3376, pruned_loss=0.09382, over 5654544.13 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3593, pruned_loss=0.1153, over 5691736.92 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3369, pruned_loss=0.09178, over 5656593.23 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:28:42,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.750e+02 1.378e+03 1.941e+03 2.766e+03 7.880e+03, threshold=3.882e+03, percent-clipped=9.0
+2023-03-11 01:29:22,906 INFO [train.py:968] (0/2) Epoch 21, batch 30900, giga_loss[loss=0.2593, simple_loss=0.334, pruned_loss=0.09231, over 28811.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3378, pruned_loss=0.09392, over 5639686.01 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3592, pruned_loss=0.1153, over 5686165.29 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.337, pruned_loss=0.09189, over 5644844.04 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:30:18,671 INFO [train.py:968] (0/2) Epoch 21, batch 30950, giga_loss[loss=0.2721, simple_loss=0.3534, pruned_loss=0.09534, over 28693.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3398, pruned_loss=0.09381, over 5629730.92 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3591, pruned_loss=0.1154, over 5677538.74 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.339, pruned_loss=0.09182, over 5640179.09 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:30:31,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.078e+03 1.579e+03 2.386e+03 3.361e+03 8.861e+03, threshold=4.772e+03, percent-clipped=18.0
+2023-03-11 01:31:13,999 INFO [train.py:968] (0/2) Epoch 21, batch 31000, giga_loss[loss=0.2574, simple_loss=0.3386, pruned_loss=0.08803, over 28811.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3399, pruned_loss=0.09382, over 5627223.96 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3587, pruned_loss=0.1153, over 5681921.10 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3388, pruned_loss=0.09155, over 5629979.95 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:32:01,952 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-11 01:32:21,452 INFO [train.py:968] (0/2) Epoch 21, batch 31050, giga_loss[loss=0.2545, simple_loss=0.3334, pruned_loss=0.08782, over 28640.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3404, pruned_loss=0.09415, over 5628119.80 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3583, pruned_loss=0.1151, over 5676282.15 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3396, pruned_loss=0.09212, over 5634014.31 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:32:35,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.907e+02 1.415e+03 1.788e+03 2.447e+03 5.429e+03, threshold=3.575e+03, percent-clipped=2.0
+2023-03-11 01:33:14,260 INFO [train.py:968] (0/2) Epoch 21, batch 31100, giga_loss[loss=0.27, simple_loss=0.3472, pruned_loss=0.09636, over 28971.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3381, pruned_loss=0.0929, over 5633941.52 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.358, pruned_loss=0.1155, over 5671544.02 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.337, pruned_loss=0.08999, over 5641756.31 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:34:08,410 INFO [train.py:968] (0/2) Epoch 21, batch 31150, giga_loss[loss=0.2581, simple_loss=0.3338, pruned_loss=0.09125, over 28907.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3363, pruned_loss=0.09101, over 5638291.05 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3572, pruned_loss=0.1151, over 5682894.98 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3349, pruned_loss=0.0876, over 5631959.60 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:34:24,604 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.610e+02 1.430e+03 1.934e+03 2.874e+03 2.219e+04, threshold=3.867e+03, percent-clipped=14.0
+2023-03-11 01:35:08,674 INFO [train.py:968] (0/2) Epoch 21, batch 31200, giga_loss[loss=0.2611, simple_loss=0.3247, pruned_loss=0.09876, over 26882.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3337, pruned_loss=0.08931, over 5635626.20 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.357, pruned_loss=0.1152, over 5675587.20 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3325, pruned_loss=0.08627, over 5636596.81 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:36:00,891 INFO [train.py:968] (0/2) Epoch 21, batch 31250, giga_loss[loss=0.24, simple_loss=0.3227, pruned_loss=0.07861, over 29040.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3335, pruned_loss=0.09055, over 5647777.30 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3571, pruned_loss=0.1156, over 5671838.47 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3313, pruned_loss=0.08651, over 5651255.97 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:36:17,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.515e+02 1.474e+03 2.202e+03 3.215e+03 8.307e+03, threshold=4.404e+03, percent-clipped=13.0
+2023-03-11 01:36:29,291 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 01:36:39,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3112, 1.5958, 1.5272, 1.1475], device='cuda:0'), covar=tensor([0.1814, 0.2731, 0.1544, 0.1878], device='cuda:0'), in_proj_covar=tensor([0.0893, 0.0694, 0.0939, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 01:36:59,463 INFO [train.py:968] (0/2) Epoch 21, batch 31300, giga_loss[loss=0.22, simple_loss=0.2873, pruned_loss=0.07639, over 24711.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3334, pruned_loss=0.09062, over 5648006.19 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3571, pruned_loss=0.1157, over 5664897.40 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3314, pruned_loss=0.08709, over 5656946.20 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:37:57,845 INFO [train.py:968] (0/2) Epoch 21, batch 31350, giga_loss[loss=0.2652, simple_loss=0.3523, pruned_loss=0.08907, over 28885.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3336, pruned_loss=0.08952, over 5653154.02 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3572, pruned_loss=0.1159, over 5668332.14 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3316, pruned_loss=0.08619, over 5657028.93 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:38:12,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.635e+02 1.473e+03 1.948e+03 2.718e+03 5.871e+03, threshold=3.896e+03, percent-clipped=4.0
+2023-03-11 01:38:54,509 INFO [train.py:968] (0/2) Epoch 21, batch 31400, libri_loss[loss=0.2648, simple_loss=0.3164, pruned_loss=0.1066, over 28143.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3348, pruned_loss=0.08988, over 5652027.70 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3561, pruned_loss=0.1154, over 5672899.20 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3333, pruned_loss=0.08667, over 5650338.46 frames. ], batch size: 62, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:39:53,663 INFO [train.py:968] (0/2) Epoch 21, batch 31450, giga_loss[loss=0.274, simple_loss=0.344, pruned_loss=0.102, over 28873.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.333, pruned_loss=0.08873, over 5661135.86 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.356, pruned_loss=0.1154, over 5670482.62 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3313, pruned_loss=0.08547, over 5660929.37 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:40:15,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.501e+02 1.355e+03 1.683e+03 2.251e+03 3.875e+03, threshold=3.367e+03, percent-clipped=0.0
+2023-03-11 01:40:27,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5998, 1.8413, 1.5052, 1.6543], device='cuda:0'), covar=tensor([0.2582, 0.2452, 0.2728, 0.2363], device='cuda:0'), in_proj_covar=tensor([0.1499, 0.1081, 0.1324, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 01:41:00,831 INFO [train.py:968] (0/2) Epoch 21, batch 31500, giga_loss[loss=0.2593, simple_loss=0.3367, pruned_loss=0.09099, over 28370.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3336, pruned_loss=0.08938, over 5660599.70 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3558, pruned_loss=0.1153, over 5664973.06 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3321, pruned_loss=0.08639, over 5666054.78 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:41:38,006 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-11 01:42:00,298 INFO [train.py:968] (0/2) Epoch 21, batch 31550, giga_loss[loss=0.2623, simple_loss=0.3554, pruned_loss=0.08462, over 28623.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3347, pruned_loss=0.0888, over 5658076.04 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3552, pruned_loss=0.115, over 5662130.83 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3334, pruned_loss=0.08601, over 5664138.06 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:42:18,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.887e+02 1.456e+03 2.125e+03 3.268e+03 1.185e+04, threshold=4.249e+03, percent-clipped=21.0
+2023-03-11 01:42:59,913 INFO [train.py:968] (0/2) Epoch 21, batch 31600, giga_loss[loss=0.2109, simple_loss=0.2857, pruned_loss=0.06803, over 24602.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3382, pruned_loss=0.08865, over 5663646.84 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3547, pruned_loss=0.1147, over 5670200.18 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.337, pruned_loss=0.08566, over 5661132.86 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:43:11,088 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=943960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:43:33,869 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-11 01:43:54,125 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-944000.pt
+2023-03-11 01:43:55,752 INFO [train.py:968] (0/2) Epoch 21, batch 31650, giga_loss[loss=0.2514, simple_loss=0.3455, pruned_loss=0.07863, over 29062.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3402, pruned_loss=0.08885, over 5656171.23 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3546, pruned_loss=0.1146, over 5669640.60 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3387, pruned_loss=0.08558, over 5654701.23 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:44:13,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.055e+02 1.429e+03 1.902e+03 2.767e+03 7.521e+03, threshold=3.804e+03, percent-clipped=11.0
+2023-03-11 01:44:54,369 INFO [train.py:968] (0/2) Epoch 21, batch 31700, giga_loss[loss=0.2721, simple_loss=0.3542, pruned_loss=0.09501, over 29103.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.08801, over 5662662.72 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3546, pruned_loss=0.1146, over 5672008.90 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3394, pruned_loss=0.085, over 5659222.43 frames. ], batch size: 113, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:44:57,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4956, 1.6175, 1.7557, 1.3365], device='cuda:0'), covar=tensor([0.2099, 0.2919, 0.1668, 0.2011], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0698, 0.0944, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 01:45:53,193 INFO [train.py:968] (0/2) Epoch 21, batch 31750, giga_loss[loss=0.2783, simple_loss=0.3562, pruned_loss=0.1002, over 28720.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3413, pruned_loss=0.08876, over 5676936.05 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3541, pruned_loss=0.1143, over 5678110.67 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3403, pruned_loss=0.08596, over 5668698.54 frames. ], batch size: 243, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:46:08,958 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.265e+02 1.343e+03 1.793e+03 2.753e+03 8.153e+03, threshold=3.586e+03, percent-clipped=11.0
+2023-03-11 01:46:41,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4835, 1.5961, 1.7352, 1.3294], device='cuda:0'), covar=tensor([0.1914, 0.2745, 0.1576, 0.1862], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0698, 0.0946, 0.0843], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 01:46:54,152 INFO [train.py:968] (0/2) Epoch 21, batch 31800, giga_loss[loss=0.2297, simple_loss=0.2959, pruned_loss=0.08176, over 24823.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3391, pruned_loss=0.08885, over 5685319.03 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3532, pruned_loss=0.1137, over 5685590.88 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3387, pruned_loss=0.08629, over 5672267.76 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:47:13,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2305, 4.0624, 3.8376, 1.9542], device='cuda:0'), covar=tensor([0.0589, 0.0740, 0.0784, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.1214, 0.1124, 0.0949, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 01:48:06,985 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-11 01:48:11,071 INFO [train.py:968] (0/2) Epoch 21, batch 31850, giga_loss[loss=0.2811, simple_loss=0.354, pruned_loss=0.1041, over 27524.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3404, pruned_loss=0.09026, over 5685482.41 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.353, pruned_loss=0.1136, over 5687429.11 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3401, pruned_loss=0.08815, over 5673612.08 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:48:37,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.749e+02 1.348e+03 1.757e+03 2.517e+03 9.774e+03, threshold=3.515e+03, percent-clipped=7.0
+2023-03-11 01:49:28,226 INFO [train.py:968] (0/2) Epoch 21, batch 31900, giga_loss[loss=0.2319, simple_loss=0.3159, pruned_loss=0.07397, over 28978.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3356, pruned_loss=0.08762, over 5682241.79 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3527, pruned_loss=0.1134, over 5689868.78 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3355, pruned_loss=0.08591, over 5670747.92 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:50:17,886 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=944291.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 01:50:29,998 INFO [train.py:968] (0/2) Epoch 21, batch 31950, giga_loss[loss=0.2483, simple_loss=0.3282, pruned_loss=0.08423, over 29019.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3332, pruned_loss=0.08633, over 5671529.41 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3525, pruned_loss=0.1134, over 5680100.96 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3329, pruned_loss=0.08446, over 5670485.65 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:50:50,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.068e+02 1.350e+03 1.624e+03 2.806e+03 6.250e+03, threshold=3.249e+03, percent-clipped=15.0
+2023-03-11 01:51:13,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=944335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:51:28,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4579, 1.8288, 1.3995, 1.6073], device='cuda:0'), covar=tensor([0.2726, 0.2552, 0.3111, 0.2255], device='cuda:0'), in_proj_covar=tensor([0.1499, 0.1077, 0.1322, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 01:51:31,559 INFO [train.py:968] (0/2) Epoch 21, batch 32000, giga_loss[loss=0.2506, simple_loss=0.3373, pruned_loss=0.08198, over 28649.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3306, pruned_loss=0.08561, over 5678266.78 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3517, pruned_loss=0.113, over 5685081.48 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3305, pruned_loss=0.08366, over 5672796.99 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:51:54,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4469, 1.7398, 1.6769, 1.4523], device='cuda:0'), covar=tensor([0.1909, 0.2088, 0.2073, 0.2178], device='cuda:0'), in_proj_covar=tensor([0.0460, 0.0731, 0.0700, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 01:52:10,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3808, 1.2528, 4.4572, 3.5067], device='cuda:0'), covar=tensor([0.1744, 0.2864, 0.0424, 0.0985], device='cuda:0'), in_proj_covar=tensor([0.0753, 0.0644, 0.0954, 0.0895], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 01:52:32,244 INFO [train.py:968] (0/2) Epoch 21, batch 32050, giga_loss[loss=0.2532, simple_loss=0.3407, pruned_loss=0.08279, over 28936.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3337, pruned_loss=0.08751, over 5689861.09 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3516, pruned_loss=0.1132, over 5692549.28 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.333, pruned_loss=0.08486, over 5678479.94 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:52:51,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.854e+02 1.439e+03 1.908e+03 2.720e+03 6.180e+03, threshold=3.817e+03, percent-clipped=16.0
+2023-03-11 01:53:28,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5500, 1.7247, 1.2165, 1.3583], device='cuda:0'), covar=tensor([0.0887, 0.0499, 0.1022, 0.1026], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0446, 0.0516, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 01:53:29,939 INFO [train.py:968] (0/2) Epoch 21, batch 32100, giga_loss[loss=0.2484, simple_loss=0.3187, pruned_loss=0.08901, over 28931.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3362, pruned_loss=0.08932, over 5695997.53 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3512, pruned_loss=0.113, over 5696306.09 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3357, pruned_loss=0.08688, over 5683437.97 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:53:52,025 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4358, 3.6383, 1.6920, 1.5260], device='cuda:0'), covar=tensor([0.0993, 0.0376, 0.0926, 0.1336], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0553, 0.0386, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 01:53:54,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5824, 1.6879, 1.8340, 1.4064], device='cuda:0'), covar=tensor([0.1841, 0.2627, 0.1485, 0.1820], device='cuda:0'), in_proj_covar=tensor([0.0895, 0.0696, 0.0942, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 01:54:05,889 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=944478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:54:10,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=944481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:54:34,111 INFO [train.py:968] (0/2) Epoch 21, batch 32150, libri_loss[loss=0.2411, simple_loss=0.3062, pruned_loss=0.08798, over 29562.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3335, pruned_loss=0.0888, over 5697126.19 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3503, pruned_loss=0.1125, over 5700509.72 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3336, pruned_loss=0.0868, over 5683445.36 frames. ], batch size: 75, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:54:44,970 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=944510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:54:51,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.464e+02 1.443e+03 1.809e+03 2.217e+03 6.082e+03, threshold=3.617e+03, percent-clipped=8.0
+2023-03-11 01:55:31,771 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=944548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:55:34,477 INFO [train.py:968] (0/2) Epoch 21, batch 32200, giga_loss[loss=0.2461, simple_loss=0.324, pruned_loss=0.08413, over 28872.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.334, pruned_loss=0.0898, over 5689789.27 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3501, pruned_loss=0.1124, over 5701235.89 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3339, pruned_loss=0.08796, over 5678106.28 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:55:37,650 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=944553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:56:01,677 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-11 01:56:39,096 INFO [train.py:968] (0/2) Epoch 21, batch 32250, giga_loss[loss=0.2491, simple_loss=0.3371, pruned_loss=0.0805, over 28707.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3352, pruned_loss=0.0901, over 5685922.31 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3499, pruned_loss=0.1122, over 5701802.90 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3351, pruned_loss=0.08841, over 5675619.78 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:56:58,750 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.668e+02 1.610e+03 2.053e+03 2.654e+03 7.054e+03, threshold=4.105e+03, percent-clipped=11.0
+2023-03-11 01:57:36,455 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.95 vs. limit=5.0
+2023-03-11 01:57:47,111 INFO [train.py:968] (0/2) Epoch 21, batch 32300, libri_loss[loss=0.2889, simple_loss=0.3428, pruned_loss=0.1175, over 29540.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3366, pruned_loss=0.0896, over 5680834.49 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3495, pruned_loss=0.112, over 5702484.86 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3365, pruned_loss=0.08786, over 5671373.33 frames. ], batch size: 76, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:58:05,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=944666.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:58:50,854 INFO [train.py:968] (0/2) Epoch 21, batch 32350, libri_loss[loss=0.3595, simple_loss=0.3949, pruned_loss=0.1621, over 29537.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3371, pruned_loss=0.09067, over 5675738.35 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3494, pruned_loss=0.1121, over 5703196.91 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3364, pruned_loss=0.08788, over 5664865.47 frames. ], batch size: 84, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:59:14,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.962e+02 1.518e+03 2.017e+03 2.674e+03 6.480e+03, threshold=4.035e+03, percent-clipped=5.0
+2023-03-11 01:59:57,577 INFO [train.py:968] (0/2) Epoch 21, batch 32400, giga_loss[loss=0.2323, simple_loss=0.3147, pruned_loss=0.07492, over 28642.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.333, pruned_loss=0.08896, over 5678598.12 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3493, pruned_loss=0.112, over 5702401.94 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3323, pruned_loss=0.08651, over 5670693.10 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:01:03,350 INFO [train.py:968] (0/2) Epoch 21, batch 32450, giga_loss[loss=0.2366, simple_loss=0.315, pruned_loss=0.07911, over 27554.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3274, pruned_loss=0.0864, over 5679637.34 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3492, pruned_loss=0.112, over 5703819.90 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3266, pruned_loss=0.08416, over 5671801.42 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:01:14,791 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=944809.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 02:01:17,934 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=944812.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 02:01:26,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.657e+02 1.549e+03 2.174e+03 3.240e+03 9.754e+03, threshold=4.348e+03, percent-clipped=13.0
+2023-03-11 02:01:57,242 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=944841.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 02:02:04,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5315, 1.6644, 1.7648, 1.3792], device='cuda:0'), covar=tensor([0.1741, 0.2637, 0.1460, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.0895, 0.0693, 0.0942, 0.0839], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 02:02:07,945 INFO [train.py:968] (0/2) Epoch 21, batch 32500, giga_loss[loss=0.2632, simple_loss=0.3456, pruned_loss=0.0904, over 27556.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3277, pruned_loss=0.08672, over 5674162.22 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3494, pruned_loss=0.1121, over 5704105.80 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3269, pruned_loss=0.08472, over 5667653.19 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:03:05,728 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=944899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:03:06,954 INFO [train.py:968] (0/2) Epoch 21, batch 32550, libri_loss[loss=0.2187, simple_loss=0.2928, pruned_loss=0.07229, over 29553.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3292, pruned_loss=0.08768, over 5672642.18 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3491, pruned_loss=0.1119, over 5702773.65 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3286, pruned_loss=0.08599, over 5667973.94 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:03:27,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.084e+02 1.484e+03 1.800e+03 2.742e+03 7.492e+03, threshold=3.601e+03, percent-clipped=6.0
+2023-03-11 02:03:33,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=944923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:03:35,293 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-11 02:03:39,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=944928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:04:04,092 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 02:04:05,654 INFO [train.py:968] (0/2) Epoch 21, batch 32600, giga_loss[loss=0.1973, simple_loss=0.2796, pruned_loss=0.05753, over 27584.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3276, pruned_loss=0.08606, over 5674587.27 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3485, pruned_loss=0.1115, over 5706985.83 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3272, pruned_loss=0.08457, over 5666220.64 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:04:26,942 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=944967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:04:35,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3503, 1.5112, 1.4564, 1.3034], device='cuda:0'), covar=tensor([0.2426, 0.2188, 0.1697, 0.2045], device='cuda:0'), in_proj_covar=tensor([0.1919, 0.1845, 0.1770, 0.1908], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 02:05:04,263 INFO [train.py:968] (0/2) Epoch 21, batch 32650, giga_loss[loss=0.2887, simple_loss=0.3624, pruned_loss=0.1075, over 28640.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3272, pruned_loss=0.08557, over 5664532.41 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3484, pruned_loss=0.1116, over 5702222.06 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3264, pruned_loss=0.0836, over 5661497.67 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:05:22,610 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.485e+02 1.426e+03 1.767e+03 2.430e+03 4.549e+03, threshold=3.534e+03, percent-clipped=6.0
+2023-03-11 02:05:23,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2246, 0.8740, 0.8794, 1.4044], device='cuda:0'), covar=tensor([0.0745, 0.0390, 0.0369, 0.0837], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0108], device='cuda:0')
+2023-03-11 02:06:03,038 INFO [train.py:968] (0/2) Epoch 21, batch 32700, giga_loss[loss=0.248, simple_loss=0.3239, pruned_loss=0.08607, over 27582.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3261, pruned_loss=0.08581, over 5667195.59 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3475, pruned_loss=0.1111, over 5709560.51 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3254, pruned_loss=0.08364, over 5657012.05 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:06:23,325 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:06:25,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:06:28,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:06:32,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:07:05,068 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:07:07,106 INFO [train.py:968] (0/2) Epoch 21, batch 32750, giga_loss[loss=0.2357, simple_loss=0.3134, pruned_loss=0.07897, over 28923.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3264, pruned_loss=0.0853, over 5681645.82 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3473, pruned_loss=0.1109, over 5712433.34 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3254, pruned_loss=0.08304, over 5669910.66 frames. ], batch size: 93, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:07:09,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:07:27,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.612e+02 1.355e+03 1.950e+03 2.478e+03 7.033e+03, threshold=3.900e+03, percent-clipped=12.0
+2023-03-11 02:08:08,996 INFO [train.py:968] (0/2) Epoch 21, batch 32800, giga_loss[loss=0.2136, simple_loss=0.3008, pruned_loss=0.06321, over 29036.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3268, pruned_loss=0.08538, over 5675260.54 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3478, pruned_loss=0.1112, over 5702714.69 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3253, pruned_loss=0.08289, over 5674153.93 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:09:03,077 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-11 02:09:03,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1180, 1.6348, 1.4745, 1.3674], device='cuda:0'), covar=tensor([0.1985, 0.1722, 0.1960, 0.1881], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0730, 0.0700, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 02:09:05,591 INFO [train.py:968] (0/2) Epoch 21, batch 32850, libri_loss[loss=0.3014, simple_loss=0.3633, pruned_loss=0.1198, over 29528.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3285, pruned_loss=0.087, over 5682179.88 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3478, pruned_loss=0.1113, over 5704993.24 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3267, pruned_loss=0.08426, over 5678506.80 frames. ], batch size: 84, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:09:27,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.263e+02 1.366e+03 1.807e+03 2.469e+03 5.874e+03, threshold=3.614e+03, percent-clipped=6.0
+2023-03-11 02:10:07,126 INFO [train.py:968] (0/2) Epoch 21, batch 32900, giga_loss[loss=0.2534, simple_loss=0.3392, pruned_loss=0.08383, over 29051.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.327, pruned_loss=0.08586, over 5666794.31 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3475, pruned_loss=0.1111, over 5699384.37 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3255, pruned_loss=0.08347, over 5669081.99 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:10:10,238 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-11 02:10:32,986 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=945274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:11:00,245 INFO [train.py:968] (0/2) Epoch 21, batch 32950, giga_loss[loss=0.24, simple_loss=0.3305, pruned_loss=0.07472, over 28879.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3282, pruned_loss=0.08593, over 5650893.21 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3471, pruned_loss=0.111, over 5689651.59 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3266, pruned_loss=0.08303, over 5660578.33 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:11:18,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.480e+02 1.331e+03 1.937e+03 2.508e+03 6.406e+03, threshold=3.873e+03, percent-clipped=10.0
+2023-03-11 02:11:33,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3009, 2.9913, 1.4659, 1.4345], device='cuda:0'), covar=tensor([0.0935, 0.0309, 0.0910, 0.1330], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0549, 0.0384, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 02:11:46,686 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=945341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:11:47,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=945342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:11:57,923 INFO [train.py:968] (0/2) Epoch 21, batch 33000, giga_loss[loss=0.2678, simple_loss=0.3565, pruned_loss=0.08955, over 28628.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3322, pruned_loss=0.08741, over 5654298.94 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3469, pruned_loss=0.1109, over 5694073.16 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3307, pruned_loss=0.08478, over 5657657.51 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:11:57,927 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 02:12:06,526 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.1961, simple_loss=0.2972, pruned_loss=0.04752, over 944034.00 frames. 
+2023-03-11 02:12:06,527 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 02:13:03,144 INFO [train.py:968] (0/2) Epoch 21, batch 33050, giga_loss[loss=0.244, simple_loss=0.331, pruned_loss=0.07854, over 28900.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.334, pruned_loss=0.08794, over 5663792.50 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3468, pruned_loss=0.1107, over 5698142.84 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3327, pruned_loss=0.0856, over 5662064.07 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:13:25,644 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:13:26,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.988e+03 2.636e+03 3.821e+03 7.016e+03, threshold=5.272e+03, percent-clipped=22.0
+2023-03-11 02:13:27,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:13:33,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3949, 1.5968, 1.6167, 1.2515], device='cuda:0'), covar=tensor([0.1644, 0.2420, 0.1394, 0.1837], device='cuda:0'), in_proj_covar=tensor([0.0893, 0.0691, 0.0941, 0.0839], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 02:13:49,383 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=945436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:14:04,953 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:14:06,735 INFO [train.py:968] (0/2) Epoch 21, batch 33100, giga_loss[loss=0.2663, simple_loss=0.3427, pruned_loss=0.095, over 28754.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.334, pruned_loss=0.08827, over 5660985.20 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3464, pruned_loss=0.1105, over 5697963.16 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.333, pruned_loss=0.08607, over 5659110.67 frames. ], batch size: 243, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:14:44,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:14:48,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:15:01,260 INFO [train.py:968] (0/2) Epoch 21, batch 33150, giga_loss[loss=0.2041, simple_loss=0.2893, pruned_loss=0.05945, over 28820.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3313, pruned_loss=0.08663, over 5679013.35 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3465, pruned_loss=0.1107, over 5705066.04 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.33, pruned_loss=0.08394, over 5670194.14 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:15:21,634 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:15:24,380 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.437e+02 1.450e+03 2.095e+03 2.899e+03 1.098e+04, threshold=4.189e+03, percent-clipped=5.0
+2023-03-11 02:16:00,650 INFO [train.py:968] (0/2) Epoch 21, batch 33200, giga_loss[loss=0.2359, simple_loss=0.3238, pruned_loss=0.07402, over 28465.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3292, pruned_loss=0.08523, over 5684575.53 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3464, pruned_loss=0.1106, over 5709064.39 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3279, pruned_loss=0.08251, over 5673133.32 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:16:18,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3783, 1.5967, 1.4322, 1.5751], device='cuda:0'), covar=tensor([0.0785, 0.0318, 0.0332, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0062, 0.0108], device='cuda:0')
+2023-03-11 02:16:28,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3946, 1.2622, 1.2706, 1.6772], device='cuda:0'), covar=tensor([0.0760, 0.0337, 0.0338, 0.0817], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0062, 0.0108], device='cuda:0')
+2023-03-11 02:17:00,580 INFO [train.py:968] (0/2) Epoch 21, batch 33250, libri_loss[loss=0.2416, simple_loss=0.2993, pruned_loss=0.0919, over 29524.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3276, pruned_loss=0.08512, over 5687694.15 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.346, pruned_loss=0.1105, over 5712202.47 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3266, pruned_loss=0.08261, over 5675162.66 frames. ], batch size: 70, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:17:21,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.428e+02 1.403e+03 1.794e+03 2.348e+03 8.312e+03, threshold=3.588e+03, percent-clipped=5.0
+2023-03-11 02:17:33,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8119, 2.4979, 1.4941, 1.0092], device='cuda:0'), covar=tensor([0.7183, 0.3579, 0.4184, 0.6437], device='cuda:0'), in_proj_covar=tensor([0.1728, 0.1631, 0.1583, 0.1410], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 02:17:40,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5937, 1.7938, 1.7648, 1.5865], device='cuda:0'), covar=tensor([0.2485, 0.2040, 0.1770, 0.2052], device='cuda:0'), in_proj_covar=tensor([0.1921, 0.1847, 0.1762, 0.1905], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 02:17:59,398 INFO [train.py:968] (0/2) Epoch 21, batch 33300, giga_loss[loss=0.2259, simple_loss=0.3108, pruned_loss=0.07051, over 28826.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3276, pruned_loss=0.08501, over 5685425.91 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3453, pruned_loss=0.1102, over 5716222.80 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3271, pruned_loss=0.08274, over 5671179.91 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:19:01,344 INFO [train.py:968] (0/2) Epoch 21, batch 33350, giga_loss[loss=0.2332, simple_loss=0.3115, pruned_loss=0.07745, over 28996.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3305, pruned_loss=0.08673, over 5685826.57 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3458, pruned_loss=0.1107, over 5719534.41 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3293, pruned_loss=0.08391, over 5670882.27 frames. ], batch size: 93, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:19:19,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=945716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:19:24,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.136e+02 1.382e+03 1.947e+03 2.677e+03 6.722e+03, threshold=3.893e+03, percent-clipped=13.0
+2023-03-11 02:20:03,628 INFO [train.py:968] (0/2) Epoch 21, batch 33400, giga_loss[loss=0.2154, simple_loss=0.2974, pruned_loss=0.06667, over 29055.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3311, pruned_loss=0.08722, over 5683106.85 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3458, pruned_loss=0.1106, over 5722130.86 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3299, pruned_loss=0.08471, over 5668214.22 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:20:41,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2034, 1.3109, 3.9663, 3.2237], device='cuda:0'), covar=tensor([0.1737, 0.2798, 0.0428, 0.1016], device='cuda:0'), in_proj_covar=tensor([0.0752, 0.0643, 0.0949, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 02:21:01,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0164, 3.8543, 3.6281, 2.1261], device='cuda:0'), covar=tensor([0.0645, 0.0792, 0.0853, 0.2086], device='cuda:0'), in_proj_covar=tensor([0.1193, 0.1104, 0.0934, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 02:21:09,754 INFO [train.py:968] (0/2) Epoch 21, batch 33450, giga_loss[loss=0.3082, simple_loss=0.3665, pruned_loss=0.1249, over 26730.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3346, pruned_loss=0.08946, over 5655425.32 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3457, pruned_loss=0.1107, over 5713679.91 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3335, pruned_loss=0.08711, over 5650937.83 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:21:20,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=945811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:21:29,402 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.927e+02 1.399e+03 1.759e+03 2.367e+03 6.971e+03, threshold=3.519e+03, percent-clipped=4.0
+2023-03-11 02:22:01,075 INFO [train.py:968] (0/2) Epoch 21, batch 33500, giga_loss[loss=0.2709, simple_loss=0.356, pruned_loss=0.09291, over 28519.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3366, pruned_loss=0.08997, over 5666572.94 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3452, pruned_loss=0.1104, over 5716835.64 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3359, pruned_loss=0.08768, over 5658715.13 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:22:10,439 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:22:15,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:22:54,087 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:23:05,862 INFO [train.py:968] (0/2) Epoch 21, batch 33550, giga_loss[loss=0.2061, simple_loss=0.2942, pruned_loss=0.05897, over 28216.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3372, pruned_loss=0.09, over 5651512.24 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3455, pruned_loss=0.1108, over 5702562.21 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3362, pruned_loss=0.08737, over 5658058.52 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:23:32,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.311e+02 1.402e+03 2.070e+03 2.894e+03 6.331e+03, threshold=4.140e+03, percent-clipped=15.0
+2023-03-11 02:23:52,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5039, 1.5957, 1.5577, 1.4820], device='cuda:0'), covar=tensor([0.1876, 0.1862, 0.1475, 0.1613], device='cuda:0'), in_proj_covar=tensor([0.1933, 0.1857, 0.1771, 0.1917], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 02:24:10,759 INFO [train.py:968] (0/2) Epoch 21, batch 33600, giga_loss[loss=0.2342, simple_loss=0.3201, pruned_loss=0.07417, over 28646.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3348, pruned_loss=0.08846, over 5645452.02 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3455, pruned_loss=0.1108, over 5687059.88 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3338, pruned_loss=0.08593, over 5663325.92 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:24:17,461 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:24:22,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:24:40,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7529, 1.9359, 1.9514, 1.5489], device='cuda:0'), covar=tensor([0.1267, 0.1878, 0.1085, 0.1386], device='cuda:0'), in_proj_covar=tensor([0.0893, 0.0692, 0.0941, 0.0839], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 02:24:54,273 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:25:12,335 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-946000.pt
+2023-03-11 02:25:13,317 INFO [train.py:968] (0/2) Epoch 21, batch 33650, giga_loss[loss=0.2476, simple_loss=0.3271, pruned_loss=0.08403, over 28422.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3341, pruned_loss=0.08848, over 5648000.70 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3459, pruned_loss=0.1112, over 5674481.71 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3327, pruned_loss=0.08554, over 5672964.67 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:25:20,273 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-11 02:25:39,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.111e+02 1.416e+03 1.778e+03 2.171e+03 4.997e+03, threshold=3.555e+03, percent-clipped=3.0
+2023-03-11 02:26:18,245 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-11 02:26:19,331 INFO [train.py:968] (0/2) Epoch 21, batch 33700, giga_loss[loss=0.2613, simple_loss=0.3361, pruned_loss=0.09328, over 28146.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3339, pruned_loss=0.08875, over 5652947.75 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3455, pruned_loss=0.111, over 5678033.32 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3329, pruned_loss=0.08631, over 5669241.35 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:27:00,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=946080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:27:28,025 INFO [train.py:968] (0/2) Epoch 21, batch 33750, giga_loss[loss=0.3551, simple_loss=0.4058, pruned_loss=0.1522, over 28440.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3311, pruned_loss=0.08794, over 5658664.87 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3453, pruned_loss=0.1108, over 5679950.21 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3305, pruned_loss=0.08601, over 5669551.09 frames. ], batch size: 369, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:27:52,511 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.343e+02 1.379e+03 1.797e+03 2.445e+03 4.421e+03, threshold=3.595e+03, percent-clipped=5.0
+2023-03-11 02:28:28,892 INFO [train.py:968] (0/2) Epoch 21, batch 33800, giga_loss[loss=0.2687, simple_loss=0.3534, pruned_loss=0.092, over 28676.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3294, pruned_loss=0.08654, over 5671178.10 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3448, pruned_loss=0.1105, over 5683250.49 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3291, pruned_loss=0.08488, over 5676378.41 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:29:30,119 INFO [train.py:968] (0/2) Epoch 21, batch 33850, giga_loss[loss=0.2211, simple_loss=0.3084, pruned_loss=0.06689, over 29034.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3282, pruned_loss=0.08472, over 5665961.06 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.345, pruned_loss=0.1107, over 5684851.39 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3277, pruned_loss=0.08298, over 5668666.22 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:29:52,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.433e+02 1.422e+03 2.079e+03 3.118e+03 9.970e+03, threshold=4.157e+03, percent-clipped=15.0
+2023-03-11 02:29:55,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7374, 1.9519, 1.8447, 1.6630], device='cuda:0'), covar=tensor([0.2848, 0.2283, 0.1908, 0.2146], device='cuda:0'), in_proj_covar=tensor([0.1928, 0.1853, 0.1767, 0.1910], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 02:30:23,209 INFO [train.py:968] (0/2) Epoch 21, batch 33900, giga_loss[loss=0.2413, simple_loss=0.3367, pruned_loss=0.07299, over 28778.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3294, pruned_loss=0.08441, over 5676809.19 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3453, pruned_loss=0.111, over 5690812.11 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3281, pruned_loss=0.08185, over 5673148.33 frames. ], batch size: 243, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:30:42,684 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-11 02:30:43,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3945, 1.7989, 1.5237, 1.5887], device='cuda:0'), covar=tensor([0.0793, 0.0291, 0.0339, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 02:31:16,411 INFO [train.py:968] (0/2) Epoch 21, batch 33950, giga_loss[loss=0.2173, simple_loss=0.3085, pruned_loss=0.06301, over 29123.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3301, pruned_loss=0.08338, over 5684611.12 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3445, pruned_loss=0.1105, over 5697540.54 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3293, pruned_loss=0.08089, over 5675244.83 frames. ], batch size: 113, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 02:31:28,548 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-11 02:31:38,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1976, 1.3517, 1.1906, 1.1897], device='cuda:0'), covar=tensor([0.2179, 0.1857, 0.1537, 0.1674], device='cuda:0'), in_proj_covar=tensor([0.1924, 0.1847, 0.1762, 0.1907], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 02:31:41,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.186e+02 1.332e+03 2.169e+03 2.947e+03 1.135e+04, threshold=4.337e+03, percent-clipped=9.0
+2023-03-11 02:32:03,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8246, 1.2019, 1.3519, 0.9909], device='cuda:0'), covar=tensor([0.2082, 0.1418, 0.2281, 0.1854], device='cuda:0'), in_proj_covar=tensor([0.0455, 0.0727, 0.0697, 0.0665], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 02:32:10,133 INFO [train.py:968] (0/2) Epoch 21, batch 34000, giga_loss[loss=0.2555, simple_loss=0.3249, pruned_loss=0.09309, over 26820.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3313, pruned_loss=0.08383, over 5679336.95 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3441, pruned_loss=0.1102, over 5689763.93 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3306, pruned_loss=0.0813, over 5678562.00 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:32:38,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6377, 1.8083, 1.2270, 1.4842], device='cuda:0'), covar=tensor([0.0913, 0.0557, 0.0936, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0441, 0.0512, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 02:32:38,922 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-11 02:33:16,604 INFO [train.py:968] (0/2) Epoch 21, batch 34050, giga_loss[loss=0.2539, simple_loss=0.3335, pruned_loss=0.08712, over 28452.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3315, pruned_loss=0.08445, over 5672877.02 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3439, pruned_loss=0.11, over 5691034.41 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3308, pruned_loss=0.08197, over 5670720.93 frames. ], batch size: 369, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:33:47,569 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.002e+02 1.420e+03 1.905e+03 2.723e+03 6.193e+03, threshold=3.809e+03, percent-clipped=4.0
+2023-03-11 02:33:51,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6049, 1.8962, 1.4783, 2.0672], device='cuda:0'), covar=tensor([0.2741, 0.2714, 0.3047, 0.2343], device='cuda:0'), in_proj_covar=tensor([0.1499, 0.1075, 0.1324, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 02:34:21,516 INFO [train.py:968] (0/2) Epoch 21, batch 34100, giga_loss[loss=0.2706, simple_loss=0.3514, pruned_loss=0.09491, over 28905.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3311, pruned_loss=0.08409, over 5669172.84 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3439, pruned_loss=0.11, over 5690945.19 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3304, pruned_loss=0.08182, over 5667086.59 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:34:22,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4960, 1.6415, 1.1780, 1.2168], device='cuda:0'), covar=tensor([0.0973, 0.0522, 0.1027, 0.1178], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0442, 0.0514, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 02:34:28,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=946455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:35:20,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2618, 3.0915, 1.2917, 1.4812], device='cuda:0'), covar=tensor([0.1045, 0.0270, 0.1006, 0.1380], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0549, 0.0385, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 02:35:27,900 INFO [train.py:968] (0/2) Epoch 21, batch 34150, giga_loss[loss=0.2209, simple_loss=0.3122, pruned_loss=0.06479, over 28210.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3314, pruned_loss=0.08368, over 5658691.17 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3435, pruned_loss=0.1099, over 5679756.45 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3308, pruned_loss=0.08133, over 5665953.50 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:36:02,550 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.038e+02 1.386e+03 1.955e+03 3.274e+03 1.286e+04, threshold=3.910e+03, percent-clipped=17.0
+2023-03-11 02:36:14,683 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 02:36:33,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3411, 1.5581, 1.4867, 1.2577], device='cuda:0'), covar=tensor([0.2643, 0.2265, 0.1720, 0.2265], device='cuda:0'), in_proj_covar=tensor([0.1922, 0.1842, 0.1753, 0.1902], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 02:36:41,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5753, 2.2814, 1.5551, 0.6878], device='cuda:0'), covar=tensor([0.6153, 0.2936, 0.4423, 0.6638], device='cuda:0'), in_proj_covar=tensor([0.1731, 0.1634, 0.1589, 0.1414], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 02:36:42,742 INFO [train.py:968] (0/2) Epoch 21, batch 34200, giga_loss[loss=0.2379, simple_loss=0.3207, pruned_loss=0.0775, over 28153.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3327, pruned_loss=0.08413, over 5660575.89 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3436, pruned_loss=0.1099, over 5680502.67 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3322, pruned_loss=0.08211, over 5665428.52 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:36:59,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5122, 1.5909, 1.7895, 1.3459], device='cuda:0'), covar=tensor([0.1852, 0.2734, 0.1491, 0.1899], device='cuda:0'), in_proj_covar=tensor([0.0891, 0.0689, 0.0938, 0.0837], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 02:37:09,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8813, 2.1181, 1.4121, 1.6459], device='cuda:0'), covar=tensor([0.0842, 0.0468, 0.0966, 0.0928], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0441, 0.0513, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 02:37:37,949 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=946598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:37:44,332 INFO [train.py:968] (0/2) Epoch 21, batch 34250, giga_loss[loss=0.268, simple_loss=0.3427, pruned_loss=0.09668, over 24600.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.337, pruned_loss=0.08691, over 5658928.92 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3437, pruned_loss=0.1102, over 5678838.57 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.336, pruned_loss=0.08406, over 5664545.56 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:37:44,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=946601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:38:09,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.109e+02 1.419e+03 1.759e+03 2.555e+03 8.951e+03, threshold=3.518e+03, percent-clipped=9.0
+2023-03-11 02:38:22,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=946630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:38:35,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-11 02:38:45,070 INFO [train.py:968] (0/2) Epoch 21, batch 34300, giga_loss[loss=0.2675, simple_loss=0.3398, pruned_loss=0.09756, over 27612.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3366, pruned_loss=0.08641, over 5654923.59 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.344, pruned_loss=0.1103, over 5663748.87 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3354, pruned_loss=0.08342, over 5672148.05 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:39:51,045 INFO [train.py:968] (0/2) Epoch 21, batch 34350, libri_loss[loss=0.2268, simple_loss=0.3027, pruned_loss=0.07549, over 29576.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3358, pruned_loss=0.0871, over 5661257.73 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3439, pruned_loss=0.1102, over 5669540.64 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3347, pruned_loss=0.08401, over 5669352.02 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:40:12,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6382, 1.2749, 5.0418, 3.5625], device='cuda:0'), covar=tensor([0.1600, 0.2925, 0.0388, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0753, 0.0644, 0.0950, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 02:40:17,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.237e+02 1.401e+03 1.795e+03 2.490e+03 7.909e+03, threshold=3.590e+03, percent-clipped=8.0
+2023-03-11 02:40:52,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5949, 1.8166, 1.2472, 1.4017], device='cuda:0'), covar=tensor([0.0949, 0.0567, 0.1002, 0.1118], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0440, 0.0512, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 02:40:59,291 INFO [train.py:968] (0/2) Epoch 21, batch 34400, giga_loss[loss=0.2876, simple_loss=0.3615, pruned_loss=0.1068, over 27913.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3332, pruned_loss=0.08537, over 5670675.63 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3438, pruned_loss=0.1102, over 5671549.04 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3322, pruned_loss=0.08238, over 5675582.44 frames. ], batch size: 476, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:41:03,357 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=946753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:41:29,793 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-11 02:42:03,650 INFO [train.py:968] (0/2) Epoch 21, batch 34450, giga_loss[loss=0.223, simple_loss=0.3047, pruned_loss=0.07066, over 28746.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3318, pruned_loss=0.08392, over 5677813.32 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3439, pruned_loss=0.1102, over 5674730.05 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3307, pruned_loss=0.08097, over 5678909.32 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:42:33,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.874e+02 1.255e+03 1.523e+03 1.925e+03 6.868e+03, threshold=3.045e+03, percent-clipped=8.0
+2023-03-11 02:42:40,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4058, 1.6015, 1.0585, 1.2331], device='cuda:0'), covar=tensor([0.1028, 0.0604, 0.1164, 0.1344], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0441, 0.0512, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 02:43:01,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6130, 1.6991, 1.7457, 1.5735], device='cuda:0'), covar=tensor([0.2644, 0.2302, 0.1750, 0.2149], device='cuda:0'), in_proj_covar=tensor([0.1931, 0.1844, 0.1756, 0.1906], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 02:43:07,763 INFO [train.py:968] (0/2) Epoch 21, batch 34500, giga_loss[loss=0.2768, simple_loss=0.3483, pruned_loss=0.1027, over 26815.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3322, pruned_loss=0.08451, over 5663349.65 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3438, pruned_loss=0.1102, over 5669293.12 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3312, pruned_loss=0.08173, over 5668777.33 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:43:57,000 INFO [train.py:968] (0/2) Epoch 21, batch 34550, libri_loss[loss=0.3654, simple_loss=0.4017, pruned_loss=0.1646, over 18707.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3348, pruned_loss=0.08709, over 5662564.76 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3435, pruned_loss=0.1101, over 5670590.82 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3336, pruned_loss=0.0833, over 5666163.15 frames. ], batch size: 188, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:44:19,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.780e+02 1.567e+03 1.890e+03 2.346e+03 6.913e+03, threshold=3.780e+03, percent-clipped=10.0
+2023-03-11 02:44:50,318 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 02:44:50,583 INFO [train.py:968] (0/2) Epoch 21, batch 34600, libri_loss[loss=0.2672, simple_loss=0.3277, pruned_loss=0.1034, over 29579.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3362, pruned_loss=0.08776, over 5677229.34 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3439, pruned_loss=0.1104, over 5677462.90 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3347, pruned_loss=0.08368, over 5673794.83 frames. ], batch size: 75, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:45:08,292 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=946965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:45:13,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5628, 1.7576, 1.7655, 1.3277], device='cuda:0'), covar=tensor([0.1697, 0.2641, 0.1448, 0.1733], device='cuda:0'), in_proj_covar=tensor([0.0892, 0.0690, 0.0939, 0.0838], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 02:45:47,571 INFO [train.py:968] (0/2) Epoch 21, batch 34650, libri_loss[loss=0.3399, simple_loss=0.3899, pruned_loss=0.1449, over 29154.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3339, pruned_loss=0.08775, over 5671847.88 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3441, pruned_loss=0.1104, over 5682752.87 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3323, pruned_loss=0.08391, over 5664134.17 frames. ], batch size: 101, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:45:50,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5391, 1.7206, 1.2042, 1.2879], device='cuda:0'), covar=tensor([0.0901, 0.0491, 0.0960, 0.1154], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0442, 0.0513, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 02:46:10,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.031e+02 1.495e+03 1.825e+03 2.270e+03 9.873e+03, threshold=3.650e+03, percent-clipped=6.0
+2023-03-11 02:46:40,993 INFO [train.py:968] (0/2) Epoch 21, batch 34700, giga_loss[loss=0.2363, simple_loss=0.3148, pruned_loss=0.07888, over 28909.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3331, pruned_loss=0.08802, over 5667180.55 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3439, pruned_loss=0.1103, over 5679726.26 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3317, pruned_loss=0.08442, over 5663570.04 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:47:31,541 INFO [train.py:968] (0/2) Epoch 21, batch 34750, giga_loss[loss=0.284, simple_loss=0.368, pruned_loss=0.1, over 28911.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3339, pruned_loss=0.08894, over 5671241.88 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3436, pruned_loss=0.1101, over 5689481.84 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3325, pruned_loss=0.08528, over 5658921.58 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:47:52,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.889e+02 1.525e+03 1.932e+03 2.797e+03 1.040e+04, threshold=3.864e+03, percent-clipped=7.0
+2023-03-11 02:47:55,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=947128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:48:13,475 INFO [train.py:968] (0/2) Epoch 21, batch 34800, giga_loss[loss=0.2975, simple_loss=0.3779, pruned_loss=0.1085, over 28919.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3424, pruned_loss=0.09381, over 5674152.53 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3434, pruned_loss=0.11, over 5685397.84 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3413, pruned_loss=0.09044, over 5667716.75 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:48:57,415 INFO [train.py:968] (0/2) Epoch 21, batch 34850, giga_loss[loss=0.2757, simple_loss=0.3611, pruned_loss=0.0951, over 29085.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3495, pruned_loss=0.09782, over 5676874.65 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3435, pruned_loss=0.1099, over 5689226.98 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3487, pruned_loss=0.0947, over 5667858.59 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:49:16,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4534, 2.2643, 1.7473, 0.6133], device='cuda:0'), covar=tensor([0.5145, 0.3137, 0.3993, 0.5917], device='cuda:0'), in_proj_covar=tensor([0.1742, 0.1646, 0.1596, 0.1422], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 02:49:17,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.369e+03 1.673e+03 2.508e+03 7.757e+03, threshold=3.347e+03, percent-clipped=7.0
+2023-03-11 02:49:38,063 INFO [train.py:968] (0/2) Epoch 21, batch 34900, libri_loss[loss=0.2976, simple_loss=0.3623, pruned_loss=0.1165, over 29258.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3511, pruned_loss=0.0997, over 5683162.27 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3435, pruned_loss=0.1099, over 5697448.58 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3507, pruned_loss=0.09675, over 5667489.97 frames. ], batch size: 97, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:49:53,843 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=947271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:49:56,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=947274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:50:12,015 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=947292.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 02:50:18,099 INFO [train.py:968] (0/2) Epoch 21, batch 34950, giga_loss[loss=0.2597, simple_loss=0.331, pruned_loss=0.09416, over 28684.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.345, pruned_loss=0.09712, over 5693840.44 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3438, pruned_loss=0.11, over 5701586.23 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3446, pruned_loss=0.0944, over 5677626.34 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:50:19,608 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=947303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:50:36,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.297e+02 1.177e+03 1.506e+03 1.864e+03 9.365e+03, threshold=3.012e+03, percent-clipped=3.0
+2023-03-11 02:50:49,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=947340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:50:59,382 INFO [train.py:968] (0/2) Epoch 21, batch 35000, giga_loss[loss=0.2622, simple_loss=0.3251, pruned_loss=0.09972, over 26631.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3392, pruned_loss=0.09476, over 5685189.94 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3435, pruned_loss=0.1098, over 5692812.92 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.339, pruned_loss=0.09245, over 5680283.25 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:51:30,361 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=947389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:51:39,564 INFO [train.py:968] (0/2) Epoch 21, batch 35050, giga_loss[loss=0.2133, simple_loss=0.2903, pruned_loss=0.06812, over 28750.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3322, pruned_loss=0.09174, over 5681429.73 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3429, pruned_loss=0.1093, over 5690665.89 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3323, pruned_loss=0.08959, over 5679556.54 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:51:57,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.186e+02 1.086e+03 1.428e+03 2.156e+03 9.688e+03, threshold=2.856e+03, percent-clipped=8.0
+2023-03-11 02:52:11,461 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 02:52:18,650 INFO [train.py:968] (0/2) Epoch 21, batch 35100, giga_loss[loss=0.2347, simple_loss=0.3038, pruned_loss=0.08277, over 27901.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3253, pruned_loss=0.08863, over 5689697.38 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3428, pruned_loss=0.1091, over 5696782.72 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3251, pruned_loss=0.08659, over 5682521.69 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:52:43,340 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=947483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:52:46,957 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=947486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:52:58,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3154, 1.2992, 3.7087, 3.2528], device='cuda:0'), covar=tensor([0.1563, 0.2855, 0.0422, 0.1926], device='cuda:0'), in_proj_covar=tensor([0.0755, 0.0644, 0.0956, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 02:53:00,240 INFO [train.py:968] (0/2) Epoch 21, batch 35150, libri_loss[loss=0.3395, simple_loss=0.3859, pruned_loss=0.1466, over 20082.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3199, pruned_loss=0.08664, over 5680035.09 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3428, pruned_loss=0.1091, over 5689852.35 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3191, pruned_loss=0.0844, over 5681372.15 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:53:12,715 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=947515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:53:19,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.502e+02 1.176e+03 1.794e+03 2.770e+03 8.390e+03, threshold=3.588e+03, percent-clipped=21.0
+2023-03-11 02:53:34,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=947542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:53:41,146 INFO [train.py:968] (0/2) Epoch 21, batch 35200, libri_loss[loss=0.324, simple_loss=0.3846, pruned_loss=0.1317, over 25709.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3165, pruned_loss=0.08512, over 5678400.76 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3432, pruned_loss=0.1091, over 5689682.42 frames. ], giga_tot_loss[loss=0.24, simple_loss=0.3148, pruned_loss=0.08264, over 5679612.75 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:54:21,822 INFO [train.py:968] (0/2) Epoch 21, batch 35250, giga_loss[loss=0.2114, simple_loss=0.2845, pruned_loss=0.06921, over 28997.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3135, pruned_loss=0.08345, over 5692754.13 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3429, pruned_loss=0.1087, over 5698366.71 frames. ], giga_tot_loss[loss=0.2365, simple_loss=0.3113, pruned_loss=0.08083, over 5685336.46 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:54:38,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.884e+02 1.011e+03 1.285e+03 1.689e+03 3.651e+03, threshold=2.570e+03, percent-clipped=1.0
+2023-03-11 02:55:00,719 INFO [train.py:968] (0/2) Epoch 21, batch 35300, giga_loss[loss=0.2552, simple_loss=0.3262, pruned_loss=0.09209, over 28633.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3102, pruned_loss=0.08153, over 5703684.07 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3428, pruned_loss=0.1084, over 5703437.97 frames. ], giga_tot_loss[loss=0.2328, simple_loss=0.3076, pruned_loss=0.07897, over 5693103.45 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:55:13,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=947667.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 02:55:37,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5818, 1.6649, 1.4179, 1.6423], device='cuda:0'), covar=tensor([0.0740, 0.0322, 0.0325, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 02:55:45,084 INFO [train.py:968] (0/2) Epoch 21, batch 35350, libri_loss[loss=0.4074, simple_loss=0.4455, pruned_loss=0.1846, over 25750.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.309, pruned_loss=0.08142, over 5708289.80 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3438, pruned_loss=0.1091, over 5704651.70 frames. ], giga_tot_loss[loss=0.2302, simple_loss=0.3049, pruned_loss=0.07781, over 5699056.05 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:56:03,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.108e+02 1.131e+03 1.489e+03 2.057e+03 4.910e+03, threshold=2.979e+03, percent-clipped=14.0
+2023-03-11 02:56:23,725 INFO [train.py:968] (0/2) Epoch 21, batch 35400, giga_loss[loss=0.1932, simple_loss=0.2725, pruned_loss=0.05698, over 29022.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3068, pruned_loss=0.08043, over 5692804.74 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3443, pruned_loss=0.1093, over 5691017.76 frames. ], giga_tot_loss[loss=0.2271, simple_loss=0.3017, pruned_loss=0.07631, over 5698703.34 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:56:35,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=947764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:57:06,937 INFO [train.py:968] (0/2) Epoch 21, batch 35450, giga_loss[loss=0.1976, simple_loss=0.2759, pruned_loss=0.0596, over 28915.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3032, pruned_loss=0.0785, over 5679123.71 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3447, pruned_loss=0.1096, over 5682307.94 frames. ], giga_tot_loss[loss=0.224, simple_loss=0.2984, pruned_loss=0.07478, over 5692215.28 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:57:14,103 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=947810.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 02:57:16,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=947813.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 02:57:23,530 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5998, 2.1899, 1.7023, 0.8730], device='cuda:0'), covar=tensor([0.5941, 0.3123, 0.4046, 0.6755], device='cuda:0'), in_proj_covar=tensor([0.1738, 0.1646, 0.1589, 0.1413], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 02:57:24,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.908e+02 1.068e+03 1.320e+03 2.138e+03 1.691e+04, threshold=2.641e+03, percent-clipped=15.0
+2023-03-11 02:57:39,928 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=947842.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 02:57:46,925 INFO [train.py:968] (0/2) Epoch 21, batch 35500, giga_loss[loss=0.2065, simple_loss=0.2812, pruned_loss=0.06591, over 28792.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3011, pruned_loss=0.07756, over 5669189.64 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3455, pruned_loss=0.1099, over 5674011.16 frames. ], giga_tot_loss[loss=0.2212, simple_loss=0.2955, pruned_loss=0.07342, over 5687595.92 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:58:30,364 INFO [train.py:968] (0/2) Epoch 21, batch 35550, giga_loss[loss=0.2039, simple_loss=0.2688, pruned_loss=0.06956, over 28597.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.2981, pruned_loss=0.07608, over 5682444.14 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3456, pruned_loss=0.1099, over 5678570.76 frames. ], giga_tot_loss[loss=0.2186, simple_loss=0.2927, pruned_loss=0.07223, over 5692839.82 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:58:37,241 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=947907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:58:39,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=947910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:58:40,468 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=947912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:58:43,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=947917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:58:53,380 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.096e+02 1.015e+03 1.365e+03 1.930e+03 4.485e+03, threshold=2.731e+03, percent-clipped=10.0
+2023-03-11 02:59:04,325 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=947939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:59:13,533 INFO [train.py:968] (0/2) Epoch 21, batch 35600, giga_loss[loss=0.2338, simple_loss=0.294, pruned_loss=0.08676, over 23724.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.2971, pruned_loss=0.07613, over 5683049.38 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3462, pruned_loss=0.11, over 5679521.47 frames. ], giga_tot_loss[loss=0.2177, simple_loss=0.2912, pruned_loss=0.07215, over 5690699.10 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:59:26,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5045, 1.7596, 1.7139, 1.2762], device='cuda:0'), covar=tensor([0.1711, 0.2701, 0.1512, 0.1800], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0700, 0.0953, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 02:59:55,704 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-948000.pt
+2023-03-11 02:59:56,585 INFO [train.py:968] (0/2) Epoch 21, batch 35650, giga_loss[loss=0.2627, simple_loss=0.3433, pruned_loss=0.09105, over 28742.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3066, pruned_loss=0.08099, over 5684670.68 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3465, pruned_loss=0.1099, over 5686934.48 frames. ], giga_tot_loss[loss=0.2269, simple_loss=0.3001, pruned_loss=0.07684, over 5684289.28 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:00:17,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.360e+02 1.324e+03 1.797e+03 2.381e+03 9.717e+03, threshold=3.593e+03, percent-clipped=17.0
+2023-03-11 03:00:41,166 INFO [train.py:968] (0/2) Epoch 21, batch 35700, giga_loss[loss=0.3275, simple_loss=0.395, pruned_loss=0.13, over 29018.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3194, pruned_loss=0.08751, over 5684564.60 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3467, pruned_loss=0.11, over 5681807.05 frames. ], giga_tot_loss[loss=0.2402, simple_loss=0.3133, pruned_loss=0.08356, over 5688664.92 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:00:49,573 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=948060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:00:52,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=948063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:01:16,102 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=948092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:01:22,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3977, 2.6498, 2.3614, 1.9018], device='cuda:0'), covar=tensor([0.2767, 0.2021, 0.2398, 0.2859], device='cuda:0'), in_proj_covar=tensor([0.1959, 0.1867, 0.1791, 0.1937], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 03:01:22,429 INFO [train.py:968] (0/2) Epoch 21, batch 35750, giga_loss[loss=0.2773, simple_loss=0.3497, pruned_loss=0.1024, over 28345.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3295, pruned_loss=0.09238, over 5684683.13 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3465, pruned_loss=0.1098, over 5679810.29 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3238, pruned_loss=0.08859, over 5690815.14 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:01:42,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.809e+02 1.296e+03 1.743e+03 2.179e+03 4.709e+03, threshold=3.487e+03, percent-clipped=5.0
+2023-03-11 03:02:04,576 INFO [train.py:968] (0/2) Epoch 21, batch 35800, giga_loss[loss=0.2481, simple_loss=0.3317, pruned_loss=0.08225, over 28874.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3366, pruned_loss=0.09525, over 5674007.47 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3472, pruned_loss=0.1102, over 5670753.77 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3313, pruned_loss=0.09164, over 5686493.65 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:02:44,065 INFO [train.py:968] (0/2) Epoch 21, batch 35850, giga_loss[loss=0.3094, simple_loss=0.3734, pruned_loss=0.1228, over 29036.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3404, pruned_loss=0.09629, over 5660032.82 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.348, pruned_loss=0.1106, over 5658453.71 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3348, pruned_loss=0.09226, over 5682135.88 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:03:08,212 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.803e+02 1.140e+03 1.522e+03 2.021e+03 6.990e+03, threshold=3.045e+03, percent-clipped=4.0
+2023-03-11 03:03:17,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4082, 3.3322, 1.5701, 1.4805], device='cuda:0'), covar=tensor([0.0946, 0.0356, 0.0893, 0.1348], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0547, 0.0382, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-11 03:03:32,251 INFO [train.py:968] (0/2) Epoch 21, batch 35900, giga_loss[loss=0.3308, simple_loss=0.3988, pruned_loss=0.1314, over 28288.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3419, pruned_loss=0.09555, over 5672085.76 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3481, pruned_loss=0.1105, over 5660571.49 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3375, pruned_loss=0.09236, over 5687707.82 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:04:05,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=948287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:04:12,255 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3505, 1.6202, 1.3026, 1.0388], device='cuda:0'), covar=tensor([0.2896, 0.2735, 0.3200, 0.2409], device='cuda:0'), in_proj_covar=tensor([0.1495, 0.1079, 0.1321, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 03:04:18,055 INFO [train.py:968] (0/2) Epoch 21, batch 35950, giga_loss[loss=0.2643, simple_loss=0.3515, pruned_loss=0.08855, over 28818.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3439, pruned_loss=0.09692, over 5680291.99 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3478, pruned_loss=0.1104, over 5664289.00 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3404, pruned_loss=0.09415, over 5689752.92 frames. ], batch size: 284, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:04:18,992 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=948302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:04:42,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.063e+02 1.192e+03 1.634e+03 2.283e+03 8.773e+03, threshold=3.268e+03, percent-clipped=10.0
+2023-03-11 03:04:58,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5950, 2.1900, 1.5924, 0.7807], device='cuda:0'), covar=tensor([0.6150, 0.2950, 0.4024, 0.6173], device='cuda:0'), in_proj_covar=tensor([0.1742, 0.1647, 0.1599, 0.1417], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 03:05:00,310 INFO [train.py:968] (0/2) Epoch 21, batch 36000, giga_loss[loss=0.3044, simple_loss=0.3583, pruned_loss=0.1252, over 23526.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3474, pruned_loss=0.09986, over 5675914.97 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3484, pruned_loss=0.1106, over 5670003.94 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3441, pruned_loss=0.09716, over 5678434.28 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:05:00,315 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 03:05:09,655 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2043, simple_loss=0.3112, pruned_loss=0.04869, over 944034.00 frames. 
+2023-03-11 03:05:09,656 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 03:05:51,983 INFO [train.py:968] (0/2) Epoch 21, batch 36050, giga_loss[loss=0.2624, simple_loss=0.3447, pruned_loss=0.09007, over 28954.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3508, pruned_loss=0.102, over 5682188.20 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3481, pruned_loss=0.1104, over 5670980.59 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3484, pruned_loss=0.09992, over 5683319.61 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:06:12,501 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.788e+02 1.344e+03 1.682e+03 2.145e+03 6.412e+03, threshold=3.363e+03, percent-clipped=5.0
+2023-03-11 03:06:15,294 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=948430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:06:17,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=948433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:06:29,538 INFO [train.py:968] (0/2) Epoch 21, batch 36100, giga_loss[loss=0.2814, simple_loss=0.3576, pruned_loss=0.1026, over 28156.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.354, pruned_loss=0.1032, over 5691192.79 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3481, pruned_loss=0.1103, over 5678348.83 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3522, pruned_loss=0.1015, over 5685858.59 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:06:29,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3101, 1.2295, 4.0384, 3.2716], device='cuda:0'), covar=tensor([0.1755, 0.3005, 0.0450, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0746, 0.0638, 0.0949, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 03:06:36,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9684, 1.0971, 3.3230, 2.8503], device='cuda:0'), covar=tensor([0.1848, 0.2971, 0.0505, 0.1068], device='cuda:0'), in_proj_covar=tensor([0.0746, 0.0638, 0.0949, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 03:06:38,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=948462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:07:09,113 INFO [train.py:968] (0/2) Epoch 21, batch 36150, giga_loss[loss=0.353, simple_loss=0.4111, pruned_loss=0.1475, over 26602.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3564, pruned_loss=0.1044, over 5683433.89 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.349, pruned_loss=0.1107, over 5675992.95 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3544, pruned_loss=0.1023, over 5681705.54 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:07:32,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.468e+02 1.189e+03 1.401e+03 1.790e+03 4.038e+03, threshold=2.802e+03, percent-clipped=4.0
+2023-03-11 03:07:49,770 INFO [train.py:968] (0/2) Epoch 21, batch 36200, giga_loss[loss=0.2965, simple_loss=0.373, pruned_loss=0.11, over 28654.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3573, pruned_loss=0.1037, over 5677129.01 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3494, pruned_loss=0.1111, over 5668601.93 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3554, pruned_loss=0.1017, over 5682172.55 frames. ], batch size: 242, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:08:28,976 INFO [train.py:968] (0/2) Epoch 21, batch 36250, giga_loss[loss=0.2757, simple_loss=0.3582, pruned_loss=0.09662, over 28672.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3566, pruned_loss=0.1022, over 5690730.54 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3495, pruned_loss=0.1109, over 5675361.37 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3552, pruned_loss=0.1005, over 5689091.23 frames. ], batch size: 71, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:08:46,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.497e+02 1.177e+03 1.496e+03 1.827e+03 4.829e+03, threshold=2.992e+03, percent-clipped=4.0
+2023-03-11 03:09:05,756 INFO [train.py:968] (0/2) Epoch 21, batch 36300, giga_loss[loss=0.2552, simple_loss=0.3422, pruned_loss=0.08412, over 28694.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3548, pruned_loss=0.1003, over 5697522.21 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3499, pruned_loss=0.111, over 5682286.50 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3535, pruned_loss=0.09837, over 5691082.04 frames. ], batch size: 92, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:09:28,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=948677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:09:46,151 INFO [train.py:968] (0/2) Epoch 21, batch 36350, giga_loss[loss=0.2475, simple_loss=0.3338, pruned_loss=0.08062, over 28933.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3529, pruned_loss=0.09825, over 5703492.90 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3502, pruned_loss=0.111, over 5680054.00 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3517, pruned_loss=0.09644, over 5701020.30 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:10:06,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.408e+02 1.095e+03 1.496e+03 2.183e+03 7.734e+03, threshold=2.992e+03, percent-clipped=8.0
+2023-03-11 03:10:24,054 INFO [train.py:968] (0/2) Epoch 21, batch 36400, libri_loss[loss=0.3631, simple_loss=0.4092, pruned_loss=0.1585, over 29516.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.353, pruned_loss=0.09917, over 5714366.44 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3501, pruned_loss=0.111, over 5686002.37 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3523, pruned_loss=0.09719, over 5707749.46 frames. ], batch size: 89, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:11:02,009 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=948790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:11:12,209 INFO [train.py:968] (0/2) Epoch 21, batch 36450, giga_loss[loss=0.3521, simple_loss=0.3786, pruned_loss=0.1628, over 23534.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3564, pruned_loss=0.1042, over 5699362.30 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3501, pruned_loss=0.1109, over 5687224.36 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3558, pruned_loss=0.1027, over 5693261.22 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:11:27,750 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=948820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:11:29,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=948823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:11:33,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.344e+02 1.439e+03 1.675e+03 2.140e+03 8.484e+03, threshold=3.349e+03, percent-clipped=9.0
+2023-03-11 03:11:50,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8015, 1.8876, 1.3615, 1.5688], device='cuda:0'), covar=tensor([0.0791, 0.0532, 0.0937, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0442, 0.0515, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 03:11:50,711 INFO [train.py:968] (0/2) Epoch 21, batch 36500, libri_loss[loss=0.3246, simple_loss=0.3861, pruned_loss=0.1315, over 29753.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3575, pruned_loss=0.1065, over 5703578.32 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3509, pruned_loss=0.1113, over 5691734.39 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3565, pruned_loss=0.1047, over 5694798.03 frames. ], batch size: 87, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:11:51,643 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=948852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:12:35,171 INFO [train.py:968] (0/2) Epoch 21, batch 36550, giga_loss[loss=0.2915, simple_loss=0.3556, pruned_loss=0.1137, over 28973.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3562, pruned_loss=0.1064, over 5710127.79 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3507, pruned_loss=0.1111, over 5693973.43 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3557, pruned_loss=0.1051, over 5701420.76 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:12:43,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8404, 1.7786, 2.0099, 1.6317], device='cuda:0'), covar=tensor([0.1897, 0.2496, 0.1423, 0.1717], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0695, 0.0947, 0.0844], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 03:12:58,671 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7058, 1.9646, 1.4442, 1.5131], device='cuda:0'), covar=tensor([0.0970, 0.0606, 0.1041, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0444, 0.0516, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 03:13:00,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.039e+02 1.279e+03 1.650e+03 2.280e+03 1.028e+04, threshold=3.299e+03, percent-clipped=12.0
+2023-03-11 03:13:20,171 INFO [train.py:968] (0/2) Epoch 21, batch 36600, libri_loss[loss=0.3232, simple_loss=0.3897, pruned_loss=0.1283, over 29528.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3544, pruned_loss=0.1059, over 5705214.21 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.351, pruned_loss=0.1113, over 5696166.51 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3538, pruned_loss=0.1046, over 5696510.28 frames. ], batch size: 84, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:13:59,762 INFO [train.py:968] (0/2) Epoch 21, batch 36650, giga_loss[loss=0.2716, simple_loss=0.3463, pruned_loss=0.09845, over 28943.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3538, pruned_loss=0.105, over 5699935.71 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3514, pruned_loss=0.1115, over 5690184.27 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.353, pruned_loss=0.1036, over 5698358.13 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:14:05,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4230, 3.3105, 1.4950, 1.5824], device='cuda:0'), covar=tensor([0.0987, 0.0300, 0.0903, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0549, 0.0383, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-11 03:14:13,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949015.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 03:14:23,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.762e+02 1.167e+03 1.404e+03 2.127e+03 6.812e+03, threshold=2.808e+03, percent-clipped=7.0
+2023-03-11 03:14:43,371 INFO [train.py:968] (0/2) Epoch 21, batch 36700, giga_loss[loss=0.2513, simple_loss=0.3338, pruned_loss=0.08435, over 28713.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3539, pruned_loss=0.1047, over 5693770.67 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3524, pruned_loss=0.112, over 5693158.79 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3525, pruned_loss=0.103, over 5690197.17 frames. ], batch size: 243, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:14:43,594 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949051.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 03:15:28,889 INFO [train.py:968] (0/2) Epoch 21, batch 36750, libri_loss[loss=0.2981, simple_loss=0.369, pruned_loss=0.1136, over 27738.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3505, pruned_loss=0.1023, over 5680211.33 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3526, pruned_loss=0.1121, over 5687285.44 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3491, pruned_loss=0.1005, over 5683669.80 frames. ], batch size: 115, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:15:51,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4464, 1.6695, 1.4323, 1.5618], device='cuda:0'), covar=tensor([0.0810, 0.0327, 0.0332, 0.0919], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 03:15:54,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.778e+02 1.207e+03 1.577e+03 2.109e+03 1.312e+04, threshold=3.153e+03, percent-clipped=12.0
+2023-03-11 03:16:17,162 INFO [train.py:968] (0/2) Epoch 21, batch 36800, giga_loss[loss=0.2205, simple_loss=0.3013, pruned_loss=0.06985, over 28862.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3443, pruned_loss=0.09905, over 5667460.10 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3529, pruned_loss=0.1122, over 5686915.06 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3429, pruned_loss=0.09743, over 5670064.01 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:16:29,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=949165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:17:08,454 INFO [train.py:968] (0/2) Epoch 21, batch 36850, giga_loss[loss=0.2314, simple_loss=0.3067, pruned_loss=0.07802, over 28910.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3378, pruned_loss=0.09592, over 5658668.61 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3527, pruned_loss=0.1121, over 5690043.59 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3366, pruned_loss=0.09457, over 5657696.26 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:17:40,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.845e+02 9.925e+02 1.236e+03 1.729e+03 3.798e+03, threshold=2.473e+03, percent-clipped=1.0
+2023-03-11 03:17:40,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7232, 1.8481, 1.2836, 1.3876], device='cuda:0'), covar=tensor([0.1001, 0.0627, 0.1105, 0.1181], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0442, 0.0515, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 03:17:54,872 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 03:17:57,637 INFO [train.py:968] (0/2) Epoch 21, batch 36900, libri_loss[loss=0.3566, simple_loss=0.3985, pruned_loss=0.1574, over 20075.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3359, pruned_loss=0.09524, over 5641455.62 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3532, pruned_loss=0.1125, over 5675561.95 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3341, pruned_loss=0.09338, over 5652694.03 frames. ], batch size: 187, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:18:16,408 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 03:18:39,131 INFO [train.py:968] (0/2) Epoch 21, batch 36950, giga_loss[loss=0.2661, simple_loss=0.3381, pruned_loss=0.09705, over 28778.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3368, pruned_loss=0.09502, over 5658821.32 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3537, pruned_loss=0.1126, over 5679946.41 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3346, pruned_loss=0.09296, over 5663094.27 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:18:44,134 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=949308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:18:47,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=949311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:19:02,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.332e+02 1.154e+03 1.430e+03 2.138e+03 8.202e+03, threshold=2.860e+03, percent-clipped=17.0
+2023-03-11 03:19:11,123 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=949340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:19:18,638 INFO [train.py:968] (0/2) Epoch 21, batch 37000, giga_loss[loss=0.2705, simple_loss=0.3502, pruned_loss=0.09541, over 29010.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3368, pruned_loss=0.09475, over 5672525.36 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.354, pruned_loss=0.1124, over 5687225.51 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.334, pruned_loss=0.09263, over 5668529.89 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:19:51,066 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=949390.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:19:59,300 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:19:59,771 INFO [train.py:968] (0/2) Epoch 21, batch 37050, libri_loss[loss=0.3193, simple_loss=0.3826, pruned_loss=0.128, over 29558.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3353, pruned_loss=0.09374, over 5686714.58 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3545, pruned_loss=0.1126, over 5687476.59 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3323, pruned_loss=0.09154, over 5682906.61 frames. ], batch size: 76, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:20:15,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1679, 1.1451, 3.9533, 3.1847], device='cuda:0'), covar=tensor([0.1644, 0.2698, 0.0446, 0.1064], device='cuda:0'), in_proj_covar=tensor([0.0751, 0.0640, 0.0952, 0.0897], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 03:20:19,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=949426.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 03:20:22,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.051e+02 1.188e+03 1.418e+03 2.008e+03 1.335e+04, threshold=2.836e+03, percent-clipped=15.0
+2023-03-11 03:20:23,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6239, 1.5528, 5.1530, 3.6356], device='cuda:0'), covar=tensor([0.1741, 0.2741, 0.0357, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0752, 0.0640, 0.0952, 0.0897], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 03:20:38,929 INFO [train.py:968] (0/2) Epoch 21, batch 37100, giga_loss[loss=0.2618, simple_loss=0.3363, pruned_loss=0.09369, over 28955.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3328, pruned_loss=0.09232, over 5705035.76 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3547, pruned_loss=0.1126, over 5691491.68 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3297, pruned_loss=0.09014, over 5698573.74 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:21:18,402 INFO [train.py:968] (0/2) Epoch 21, batch 37150, giga_loss[loss=0.2531, simple_loss=0.3285, pruned_loss=0.08887, over 28981.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3313, pruned_loss=0.09191, over 5709244.47 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3555, pruned_loss=0.1126, over 5698159.25 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3273, pruned_loss=0.08939, over 5698340.61 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:21:39,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.276e+02 1.080e+03 1.401e+03 1.822e+03 6.080e+03, threshold=2.802e+03, percent-clipped=7.0
+2023-03-11 03:21:42,804 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=949533.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:21:44,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=949536.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:21:56,757 INFO [train.py:968] (0/2) Epoch 21, batch 37200, giga_loss[loss=0.2392, simple_loss=0.3112, pruned_loss=0.08366, over 24340.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3287, pruned_loss=0.09067, over 5708767.40 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3553, pruned_loss=0.1123, over 5698929.33 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3251, pruned_loss=0.08852, over 5699638.71 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:21:57,226 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 03:22:07,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=949565.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:22:09,926 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=949569.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 03:22:12,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=949572.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:22:35,025 INFO [train.py:968] (0/2) Epoch 21, batch 37250, giga_loss[loss=0.2226, simple_loss=0.3009, pruned_loss=0.0721, over 28760.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.327, pruned_loss=0.09005, over 5719743.80 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3557, pruned_loss=0.1125, over 5703021.74 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3232, pruned_loss=0.08771, over 5709256.62 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:22:35,306 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=949601.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:22:50,357 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:22:57,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.595e+02 1.043e+03 1.200e+03 1.547e+03 5.652e+03, threshold=2.401e+03, percent-clipped=4.0
+2023-03-11 03:23:15,835 INFO [train.py:968] (0/2) Epoch 21, batch 37300, giga_loss[loss=0.2272, simple_loss=0.3033, pruned_loss=0.07549, over 28513.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3241, pruned_loss=0.08849, over 5723884.30 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3557, pruned_loss=0.1123, over 5707064.08 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3205, pruned_loss=0.08635, over 5712251.19 frames. ], batch size: 78, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:23:55,048 INFO [train.py:968] (0/2) Epoch 21, batch 37350, giga_loss[loss=0.2106, simple_loss=0.2918, pruned_loss=0.06469, over 28848.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3232, pruned_loss=0.08794, over 5716749.87 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3566, pruned_loss=0.1126, over 5702311.49 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3186, pruned_loss=0.08531, over 5712586.92 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:24:18,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.574e+02 1.026e+03 1.337e+03 1.952e+03 5.880e+03, threshold=2.673e+03, percent-clipped=15.0
+2023-03-11 03:24:36,433 INFO [train.py:968] (0/2) Epoch 21, batch 37400, giga_loss[loss=0.227, simple_loss=0.2987, pruned_loss=0.07766, over 28387.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3213, pruned_loss=0.08664, over 5716650.79 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3567, pruned_loss=0.1124, over 5705433.69 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3169, pruned_loss=0.08427, over 5710822.35 frames. ], batch size: 78, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:24:55,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=949775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:25:14,859 INFO [train.py:968] (0/2) Epoch 21, batch 37450, giga_loss[loss=0.2202, simple_loss=0.2966, pruned_loss=0.0719, over 28723.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3214, pruned_loss=0.08689, over 5707470.16 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3574, pruned_loss=0.1127, over 5700772.23 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3165, pruned_loss=0.08413, over 5707807.95 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:25:24,256 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:25:33,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:25:38,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.965e+02 1.110e+03 1.467e+03 1.891e+03 1.317e+04, threshold=2.933e+03, percent-clipped=11.0
+2023-03-11 03:25:48,153 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:25:56,310 INFO [train.py:968] (0/2) Epoch 21, batch 37500, giga_loss[loss=0.2616, simple_loss=0.3269, pruned_loss=0.09817, over 28737.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.323, pruned_loss=0.08794, over 5717450.39 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3575, pruned_loss=0.1126, over 5704287.59 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3184, pruned_loss=0.08533, over 5714771.79 frames. ], batch size: 92, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:26:27,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5046, 1.5975, 1.3186, 1.1069], device='cuda:0'), covar=tensor([0.1028, 0.0564, 0.1038, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0442, 0.0514, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 03:26:43,047 INFO [train.py:968] (0/2) Epoch 21, batch 37550, giga_loss[loss=0.2912, simple_loss=0.3557, pruned_loss=0.1133, over 28837.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3281, pruned_loss=0.09123, over 5714019.58 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3582, pruned_loss=0.1129, over 5706978.79 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3234, pruned_loss=0.08862, over 5709675.14 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:26:56,930 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=949918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:27:01,149 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=949921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:27:10,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.528e+02 1.183e+03 1.549e+03 1.950e+03 4.283e+03, threshold=3.099e+03, percent-clipped=7.0
+2023-03-11 03:27:30,470 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=949950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:27:30,867 INFO [train.py:968] (0/2) Epoch 21, batch 37600, giga_loss[loss=0.2876, simple_loss=0.3579, pruned_loss=0.1086, over 28841.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3357, pruned_loss=0.09634, over 5699226.55 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3586, pruned_loss=0.1132, over 5699905.17 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3313, pruned_loss=0.09376, over 5702964.42 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:28:09,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=949995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:28:14,485 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-950000.pt
+2023-03-11 03:28:15,386 INFO [train.py:968] (0/2) Epoch 21, batch 37650, giga_loss[loss=0.314, simple_loss=0.3887, pruned_loss=0.1197, over 28942.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3437, pruned_loss=0.1016, over 5695909.67 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3588, pruned_loss=0.1134, over 5703799.38 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3394, pruned_loss=0.09895, over 5695654.22 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:28:21,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4096, 1.5786, 1.1642, 1.1270], device='cuda:0'), covar=tensor([0.0985, 0.0532, 0.1102, 0.1229], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0445, 0.0517, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 03:28:44,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6807, 1.8617, 1.5660, 1.5941], device='cuda:0'), covar=tensor([0.2797, 0.2792, 0.3138, 0.2486], device='cuda:0'), in_proj_covar=tensor([0.1497, 0.1082, 0.1319, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 03:28:49,352 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.373e+03 1.745e+03 2.491e+03 5.489e+03, threshold=3.489e+03, percent-clipped=15.0
+2023-03-11 03:29:05,941 INFO [train.py:968] (0/2) Epoch 21, batch 37700, giga_loss[loss=0.2998, simple_loss=0.3802, pruned_loss=0.1098, over 28882.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3475, pruned_loss=0.1026, over 5684766.57 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3592, pruned_loss=0.1135, over 5704337.41 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3437, pruned_loss=0.1003, over 5683885.72 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:29:54,549 INFO [train.py:968] (0/2) Epoch 21, batch 37750, libri_loss[loss=0.2762, simple_loss=0.3457, pruned_loss=0.1034, over 29561.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3528, pruned_loss=0.1049, over 5684511.03 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3591, pruned_loss=0.1135, over 5706374.03 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3497, pruned_loss=0.103, over 5681681.67 frames. ], batch size: 79, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:30:21,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.302e+02 1.207e+03 1.611e+03 2.282e+03 1.309e+04, threshold=3.222e+03, percent-clipped=11.0
+2023-03-11 03:30:25,786 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:30:26,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5605, 1.8143, 1.7986, 1.3469], device='cuda:0'), covar=tensor([0.1771, 0.2559, 0.1510, 0.1787], device='cuda:0'), in_proj_covar=tensor([0.0901, 0.0700, 0.0949, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 03:30:27,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:30:34,886 INFO [train.py:968] (0/2) Epoch 21, batch 37800, giga_loss[loss=0.2924, simple_loss=0.3637, pruned_loss=0.1106, over 27929.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3585, pruned_loss=0.1081, over 5691659.12 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3594, pruned_loss=0.1135, over 5712985.03 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3557, pruned_loss=0.1064, over 5682493.07 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:30:50,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4150, 1.5468, 1.2510, 1.5978], device='cuda:0'), covar=tensor([0.0795, 0.0351, 0.0338, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0116, 0.0116, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 03:30:51,560 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=950170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:31:04,278 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=950186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:31:13,754 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=950196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:31:16,912 INFO [train.py:968] (0/2) Epoch 21, batch 37850, giga_loss[loss=0.2641, simple_loss=0.3431, pruned_loss=0.09255, over 28828.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3545, pruned_loss=0.1047, over 5689654.79 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3595, pruned_loss=0.1136, over 5714053.11 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3522, pruned_loss=0.1032, over 5681470.63 frames. ], batch size: 112, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:31:26,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=950214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:31:30,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4917, 1.9518, 1.7792, 1.5999], device='cuda:0'), covar=tensor([0.2144, 0.2266, 0.2204, 0.2211], device='cuda:0'), in_proj_covar=tensor([0.0472, 0.0745, 0.0713, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 03:31:40,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.314e+02 1.238e+03 1.489e+03 2.274e+03 6.277e+03, threshold=2.978e+03, percent-clipped=7.0
+2023-03-11 03:31:46,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6291, 1.8773, 1.5012, 1.5818], device='cuda:0'), covar=tensor([0.2756, 0.2775, 0.3093, 0.2453], device='cuda:0'), in_proj_covar=tensor([0.1496, 0.1080, 0.1318, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 03:31:55,802 INFO [train.py:968] (0/2) Epoch 21, batch 37900, giga_loss[loss=0.2561, simple_loss=0.3373, pruned_loss=0.08747, over 28915.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3506, pruned_loss=0.1014, over 5704688.26 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3592, pruned_loss=0.1137, over 5719272.29 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3489, pruned_loss=0.09994, over 5693131.48 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:32:07,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4102, 1.7163, 1.3992, 1.4793], device='cuda:0'), covar=tensor([0.0833, 0.0317, 0.0332, 0.0899], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0116, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 03:32:15,655 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=950272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:32:15,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2652, 1.4850, 1.5495, 1.3849], device='cuda:0'), covar=tensor([0.1558, 0.1199, 0.1755, 0.1355], device='cuda:0'), in_proj_covar=tensor([0.0472, 0.0745, 0.0713, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 03:32:25,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4191, 3.5830, 1.6085, 1.4759], device='cuda:0'), covar=tensor([0.1077, 0.0285, 0.0932, 0.1480], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0547, 0.0382, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-11 03:32:39,560 INFO [train.py:968] (0/2) Epoch 21, batch 37950, giga_loss[loss=0.2943, simple_loss=0.3652, pruned_loss=0.1117, over 28548.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3501, pruned_loss=0.1005, over 5689255.69 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3594, pruned_loss=0.1138, over 5705086.86 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3484, pruned_loss=0.09891, over 5692140.27 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:32:47,016 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-11 03:33:04,487 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:06,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.378e+02 1.270e+03 1.898e+03 2.713e+03 9.508e+03, threshold=3.795e+03, percent-clipped=21.0
+2023-03-11 03:33:06,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:11,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:14,885 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:22,717 INFO [train.py:968] (0/2) Epoch 21, batch 38000, giga_loss[loss=0.294, simple_loss=0.3656, pruned_loss=0.1112, over 28384.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3507, pruned_loss=0.1008, over 5685881.48 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3596, pruned_loss=0.1141, over 5699254.35 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.349, pruned_loss=0.0991, over 5692914.99 frames. ], batch size: 65, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:33:27,628 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:31,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:32,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=950361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:40,829 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=950371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:43,182 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=950374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:47,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7207, 1.1897, 2.7250, 2.7590], device='cuda:0'), covar=tensor([0.2157, 0.2890, 0.1089, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.0749, 0.0638, 0.0946, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 03:33:56,228 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=950389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:34:05,232 INFO [train.py:968] (0/2) Epoch 21, batch 38050, libri_loss[loss=0.3609, simple_loss=0.4195, pruned_loss=0.1511, over 29755.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3536, pruned_loss=0.1026, over 5688245.46 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3593, pruned_loss=0.114, over 5693994.64 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3522, pruned_loss=0.101, over 5697462.17 frames. ], batch size: 87, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:34:07,851 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 03:34:13,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3831, 1.8371, 1.2275, 0.9077], device='cuda:0'), covar=tensor([0.6614, 0.3433, 0.3213, 0.5797], device='cuda:0'), in_proj_covar=tensor([0.1735, 0.1637, 0.1598, 0.1414], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 03:34:31,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.172e+02 1.287e+03 1.607e+03 2.185e+03 6.398e+03, threshold=3.214e+03, percent-clipped=7.0
+2023-03-11 03:34:49,120 INFO [train.py:968] (0/2) Epoch 21, batch 38100, giga_loss[loss=0.3293, simple_loss=0.3923, pruned_loss=0.1332, over 27953.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3558, pruned_loss=0.1043, over 5689794.17 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3596, pruned_loss=0.114, over 5695878.49 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3544, pruned_loss=0.1028, over 5695412.71 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:35:31,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=950499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:35:32,867 INFO [train.py:968] (0/2) Epoch 21, batch 38150, giga_loss[loss=0.2566, simple_loss=0.3429, pruned_loss=0.08514, over 29067.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3561, pruned_loss=0.1049, over 5691991.86 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3595, pruned_loss=0.114, over 5699172.48 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.355, pruned_loss=0.1035, over 5693382.24 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:35:59,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.941e+02 1.346e+03 1.718e+03 2.506e+03 6.503e+03, threshold=3.436e+03, percent-clipped=11.0
+2023-03-11 03:36:13,307 INFO [train.py:968] (0/2) Epoch 21, batch 38200, libri_loss[loss=0.3402, simple_loss=0.3962, pruned_loss=0.1421, over 29514.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3564, pruned_loss=0.1056, over 5700186.20 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.36, pruned_loss=0.1141, over 5708003.73 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.355, pruned_loss=0.104, over 5692676.72 frames. ], batch size: 89, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:36:52,658 INFO [train.py:968] (0/2) Epoch 21, batch 38250, giga_loss[loss=0.2809, simple_loss=0.355, pruned_loss=0.1034, over 28875.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3564, pruned_loss=0.1057, over 5698145.15 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.36, pruned_loss=0.114, over 5711158.33 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3551, pruned_loss=0.1043, over 5689164.10 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:37:19,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.873e+02 1.262e+03 1.601e+03 2.356e+03 7.409e+03, threshold=3.202e+03, percent-clipped=8.0
+2023-03-11 03:37:25,355 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=950639.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:37:30,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=950647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:37:32,935 INFO [train.py:968] (0/2) Epoch 21, batch 38300, giga_loss[loss=0.291, simple_loss=0.3679, pruned_loss=0.107, over 28966.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3561, pruned_loss=0.1051, over 5705463.45 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3602, pruned_loss=0.1143, over 5713951.37 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3547, pruned_loss=0.1035, over 5695436.96 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:37:49,862 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-11 03:38:12,719 INFO [train.py:968] (0/2) Epoch 21, batch 38350, giga_loss[loss=0.2589, simple_loss=0.3516, pruned_loss=0.08313, over 28991.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3563, pruned_loss=0.1043, over 5690325.85 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3605, pruned_loss=0.1144, over 5698004.71 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3548, pruned_loss=0.1027, over 5696151.02 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:38:35,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.239e+02 1.227e+03 1.451e+03 1.951e+03 8.062e+03, threshold=2.903e+03, percent-clipped=9.0
+2023-03-11 03:38:47,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=950749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:38:49,103 INFO [train.py:968] (0/2) Epoch 21, batch 38400, giga_loss[loss=0.2475, simple_loss=0.3271, pruned_loss=0.08394, over 28678.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.356, pruned_loss=0.1035, over 5705742.74 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3602, pruned_loss=0.1142, over 5704056.22 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3549, pruned_loss=0.102, over 5704891.81 frames. ], batch size: 92, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:38:57,346 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 03:39:22,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:39:24,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:39:29,841 INFO [train.py:968] (0/2) Epoch 21, batch 38450, giga_loss[loss=0.2546, simple_loss=0.3278, pruned_loss=0.09073, over 29008.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3532, pruned_loss=0.1022, over 5698587.29 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3601, pruned_loss=0.1141, over 5698856.77 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3524, pruned_loss=0.1009, over 5703146.98 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:39:35,831 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 03:39:46,241 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=950822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:39:54,550 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.397e+02 1.238e+03 1.587e+03 2.260e+03 5.231e+03, threshold=3.174e+03, percent-clipped=12.0
+2023-03-11 03:40:07,746 INFO [train.py:968] (0/2) Epoch 21, batch 38500, giga_loss[loss=0.2687, simple_loss=0.3465, pruned_loss=0.09544, over 29001.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3513, pruned_loss=0.1015, over 5697160.15 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3597, pruned_loss=0.114, over 5697150.82 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3507, pruned_loss=0.1001, over 5701598.15 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:40:27,892 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=950874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:40:41,438 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:40:43,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:40:46,236 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-11 03:40:48,428 INFO [train.py:968] (0/2) Epoch 21, batch 38550, giga_loss[loss=0.2564, simple_loss=0.3352, pruned_loss=0.08887, over 28667.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3494, pruned_loss=0.1003, over 5700684.06 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3596, pruned_loss=0.1139, over 5700173.55 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3489, pruned_loss=0.09909, over 5701640.04 frames. ], batch size: 66, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:41:06,808 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=950924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:41:14,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.541e+02 1.031e+03 1.293e+03 1.651e+03 5.462e+03, threshold=2.586e+03, percent-clipped=5.0
+2023-03-11 03:41:28,598 INFO [train.py:968] (0/2) Epoch 21, batch 38600, giga_loss[loss=0.2434, simple_loss=0.3207, pruned_loss=0.08303, over 28659.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3494, pruned_loss=0.1006, over 5700317.19 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3599, pruned_loss=0.1141, over 5697651.46 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3487, pruned_loss=0.09936, over 5703251.06 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:41:43,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8160, 2.8041, 1.8914, 0.9034], device='cuda:0'), covar=tensor([0.7884, 0.3059, 0.3814, 0.6998], device='cuda:0'), in_proj_covar=tensor([0.1731, 0.1627, 0.1589, 0.1411], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 03:42:09,682 INFO [train.py:968] (0/2) Epoch 21, batch 38650, giga_loss[loss=0.273, simple_loss=0.3474, pruned_loss=0.09927, over 28569.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3503, pruned_loss=0.1011, over 5703198.54 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3603, pruned_loss=0.1143, over 5702959.01 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3492, pruned_loss=0.09968, over 5700872.62 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:42:21,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=951014.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:42:23,398 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=951017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:42:25,133 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=951020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:42:35,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.760e+02 1.076e+03 1.323e+03 1.740e+03 5.275e+03, threshold=2.645e+03, percent-clipped=7.0
+2023-03-11 03:42:46,978 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=951049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:42:48,717 INFO [train.py:968] (0/2) Epoch 21, batch 38700, giga_loss[loss=0.307, simple_loss=0.363, pruned_loss=0.1255, over 28627.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.35, pruned_loss=0.1003, over 5710088.99 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.36, pruned_loss=0.1141, over 5705079.09 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3492, pruned_loss=0.09918, over 5706432.00 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:43:26,091 INFO [train.py:968] (0/2) Epoch 21, batch 38750, giga_loss[loss=0.2655, simple_loss=0.3393, pruned_loss=0.09589, over 28885.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3493, pruned_loss=0.09949, over 5709932.07 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.36, pruned_loss=0.114, over 5701980.72 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3484, pruned_loss=0.09814, over 5710711.29 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:43:26,281 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=951101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:43:51,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.926e+02 1.051e+03 1.337e+03 1.917e+03 1.286e+04, threshold=2.673e+03, percent-clipped=15.0
+2023-03-11 03:44:04,492 INFO [train.py:968] (0/2) Epoch 21, batch 38800, giga_loss[loss=0.2413, simple_loss=0.3224, pruned_loss=0.0801, over 28495.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3505, pruned_loss=0.1007, over 5700253.03 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3611, pruned_loss=0.115, over 5697000.91 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3483, pruned_loss=0.09823, over 5706101.36 frames. ], batch size: 65, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:44:10,806 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=951157.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 03:44:13,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=951160.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:44:36,178 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=951189.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:44:45,160 INFO [train.py:968] (0/2) Epoch 21, batch 38850, giga_loss[loss=0.2935, simple_loss=0.3635, pruned_loss=0.1118, over 28562.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3498, pruned_loss=0.101, over 5681203.10 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3615, pruned_loss=0.1152, over 5680210.46 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3473, pruned_loss=0.0982, over 5701539.56 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:45:08,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3676, 3.0292, 1.4489, 1.6296], device='cuda:0'), covar=tensor([0.1032, 0.0311, 0.0939, 0.1346], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0545, 0.0381, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-11 03:45:10,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.033e+02 1.115e+03 1.434e+03 2.038e+03 6.666e+03, threshold=2.868e+03, percent-clipped=12.0
+2023-03-11 03:45:20,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3880, 2.4832, 1.8150, 2.0108], device='cuda:0'), covar=tensor([0.0971, 0.0717, 0.1050, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0443, 0.0516, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 03:45:25,616 INFO [train.py:968] (0/2) Epoch 21, batch 38900, giga_loss[loss=0.2202, simple_loss=0.3048, pruned_loss=0.06782, over 28855.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3453, pruned_loss=0.09846, over 5688253.38 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3613, pruned_loss=0.1151, over 5682477.64 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3433, pruned_loss=0.09624, over 5702172.78 frames. ], batch size: 112, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:45:28,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3976, 1.7033, 1.3426, 1.3508], device='cuda:0'), covar=tensor([0.2665, 0.2625, 0.2951, 0.2333], device='cuda:0'), in_proj_covar=tensor([0.1500, 0.1085, 0.1321, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 03:46:03,787 INFO [train.py:968] (0/2) Epoch 21, batch 38950, giga_loss[loss=0.2378, simple_loss=0.308, pruned_loss=0.08382, over 28716.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.343, pruned_loss=0.09726, over 5676705.88 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3621, pruned_loss=0.1156, over 5664627.11 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3404, pruned_loss=0.09469, over 5704433.81 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:46:24,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6657, 1.8125, 1.7647, 1.6140], device='cuda:0'), covar=tensor([0.1878, 0.2026, 0.2256, 0.2119], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0749, 0.0714, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 03:46:29,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.085e+02 1.169e+03 1.537e+03 2.340e+03 4.836e+03, threshold=3.074e+03, percent-clipped=15.0
+2023-03-11 03:46:42,486 INFO [train.py:968] (0/2) Epoch 21, batch 39000, giga_loss[loss=0.2435, simple_loss=0.3184, pruned_loss=0.08428, over 28792.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3422, pruned_loss=0.09747, over 5683373.93 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3612, pruned_loss=0.1151, over 5671232.10 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3401, pruned_loss=0.09503, over 5700349.17 frames. ], batch size: 92, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 03:46:42,491 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 03:46:52,169 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.205, simple_loss=0.3122, pruned_loss=0.04886, over 944034.00 frames. 
+2023-03-11 03:46:52,170 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 03:47:04,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2696, 4.0983, 3.8982, 1.6979], device='cuda:0'), covar=tensor([0.0562, 0.0702, 0.0717, 0.2323], device='cuda:0'), in_proj_covar=tensor([0.1198, 0.1117, 0.0940, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 03:47:30,437 INFO [train.py:968] (0/2) Epoch 21, batch 39050, giga_loss[loss=0.2693, simple_loss=0.3464, pruned_loss=0.09607, over 28918.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3404, pruned_loss=0.09658, over 5687956.86 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3613, pruned_loss=0.1151, over 5666254.72 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3382, pruned_loss=0.09428, over 5705678.73 frames. ], batch size: 186, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 03:47:54,760 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-11 03:47:57,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.148e+02 1.177e+03 1.566e+03 2.051e+03 4.990e+03, threshold=3.133e+03, percent-clipped=7.0
+2023-03-11 03:48:10,917 INFO [train.py:968] (0/2) Epoch 21, batch 39100, giga_loss[loss=0.2438, simple_loss=0.3162, pruned_loss=0.0857, over 28671.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3372, pruned_loss=0.09478, over 5690966.26 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3614, pruned_loss=0.1152, over 5668664.21 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3352, pruned_loss=0.09272, over 5702973.29 frames. ], batch size: 92, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 03:48:30,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=951476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:48:49,270 INFO [train.py:968] (0/2) Epoch 21, batch 39150, libri_loss[loss=0.3407, simple_loss=0.3971, pruned_loss=0.1421, over 29504.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3354, pruned_loss=0.09442, over 5694989.63 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3615, pruned_loss=0.1152, over 5671221.36 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3334, pruned_loss=0.09247, over 5702492.62 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 03:48:56,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3618, 1.5344, 1.4564, 1.2916], device='cuda:0'), covar=tensor([0.2973, 0.2622, 0.1733, 0.2361], device='cuda:0'), in_proj_covar=tensor([0.1962, 0.1880, 0.1811, 0.1949], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 03:49:17,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.513e+02 1.091e+03 1.336e+03 1.744e+03 3.291e+03, threshold=2.673e+03, percent-clipped=2.0
+2023-03-11 03:49:33,895 INFO [train.py:968] (0/2) Epoch 21, batch 39200, giga_loss[loss=0.2551, simple_loss=0.3226, pruned_loss=0.09375, over 28981.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3326, pruned_loss=0.09291, over 5699118.97 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3616, pruned_loss=0.1153, over 5672634.25 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3308, pruned_loss=0.0912, over 5703876.02 frames. ], batch size: 106, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:49:48,770 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=951570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:50:13,958 INFO [train.py:968] (0/2) Epoch 21, batch 39250, giga_loss[loss=0.2331, simple_loss=0.3096, pruned_loss=0.07832, over 28466.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3332, pruned_loss=0.09313, over 5708276.53 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3617, pruned_loss=0.1153, over 5680238.84 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3307, pruned_loss=0.09111, over 5706272.35 frames. ], batch size: 71, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:50:28,782 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=951619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:50:31,737 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=951622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:50:42,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.000e+02 1.129e+03 1.375e+03 1.705e+03 5.129e+03, threshold=2.750e+03, percent-clipped=7.0
+2023-03-11 03:50:51,604 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-11 03:50:53,009 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.88 vs. limit=2.0
+2023-03-11 03:50:55,104 INFO [train.py:968] (0/2) Epoch 21, batch 39300, giga_loss[loss=0.2662, simple_loss=0.3435, pruned_loss=0.09447, over 28841.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3371, pruned_loss=0.095, over 5712117.58 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3619, pruned_loss=0.1153, over 5683761.64 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3342, pruned_loss=0.0928, over 5707986.23 frames. ], batch size: 186, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:50:55,403 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=951651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:51:40,531 INFO [train.py:968] (0/2) Epoch 21, batch 39350, giga_loss[loss=0.245, simple_loss=0.3238, pruned_loss=0.08308, over 28839.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.339, pruned_loss=0.09496, over 5705833.10 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3618, pruned_loss=0.1152, over 5683579.47 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3365, pruned_loss=0.09306, over 5703039.82 frames. ], batch size: 112, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:52:08,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.042e+02 1.102e+03 1.378e+03 2.046e+03 4.120e+03, threshold=2.756e+03, percent-clipped=8.0
+2023-03-11 03:52:21,818 INFO [train.py:968] (0/2) Epoch 21, batch 39400, giga_loss[loss=0.2655, simple_loss=0.34, pruned_loss=0.09553, over 28697.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3415, pruned_loss=0.09546, over 5708354.13 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3614, pruned_loss=0.1148, over 5688043.86 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3392, pruned_loss=0.09371, over 5702674.67 frames. ], batch size: 92, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:52:23,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3751, 2.0283, 1.5113, 0.6753], device='cuda:0'), covar=tensor([0.5154, 0.2808, 0.4397, 0.6276], device='cuda:0'), in_proj_covar=tensor([0.1731, 0.1632, 0.1593, 0.1409], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 03:52:43,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4529, 1.6381, 1.2737, 1.6032], device='cuda:0'), covar=tensor([0.0759, 0.0305, 0.0359, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0116, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 03:53:05,258 INFO [train.py:968] (0/2) Epoch 21, batch 39450, giga_loss[loss=0.2506, simple_loss=0.3383, pruned_loss=0.08148, over 28638.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3411, pruned_loss=0.09441, over 5681371.87 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3613, pruned_loss=0.1148, over 5672770.96 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.339, pruned_loss=0.09272, over 5691113.39 frames. ], batch size: 307, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:53:32,189 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=951829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:53:36,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.202e+02 1.103e+03 1.332e+03 1.709e+03 6.584e+03, threshold=2.664e+03, percent-clipped=7.0
+2023-03-11 03:53:38,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3560, 1.6425, 1.3418, 1.5067], device='cuda:0'), covar=tensor([0.0740, 0.0329, 0.0344, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 03:53:46,851 INFO [train.py:968] (0/2) Epoch 21, batch 39500, libri_loss[loss=0.3075, simple_loss=0.3766, pruned_loss=0.1192, over 29537.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3405, pruned_loss=0.09369, over 5684935.59 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3612, pruned_loss=0.1147, over 5672957.14 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3385, pruned_loss=0.09194, over 5692852.42 frames. ], batch size: 83, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:54:26,465 INFO [train.py:968] (0/2) Epoch 21, batch 39550, libri_loss[loss=0.2816, simple_loss=0.3515, pruned_loss=0.1058, over 29516.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3426, pruned_loss=0.0956, over 5687960.51 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3609, pruned_loss=0.1144, over 5677760.38 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3405, pruned_loss=0.09364, over 5690994.30 frames. ], batch size: 81, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:54:53,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.063e+02 1.239e+03 1.746e+03 2.271e+03 7.194e+03, threshold=3.492e+03, percent-clipped=18.0
+2023-03-11 03:55:01,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=951945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:55:07,043 INFO [train.py:968] (0/2) Epoch 21, batch 39600, giga_loss[loss=0.2599, simple_loss=0.3347, pruned_loss=0.0926, over 28784.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.343, pruned_loss=0.09636, over 5694249.42 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3607, pruned_loss=0.1142, over 5682119.59 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3412, pruned_loss=0.09462, over 5693076.13 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 03:55:47,644 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-952000.pt
+2023-03-11 03:55:48,643 INFO [train.py:968] (0/2) Epoch 21, batch 39650, libri_loss[loss=0.2978, simple_loss=0.3715, pruned_loss=0.112, over 29235.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3458, pruned_loss=0.09797, over 5697623.38 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3611, pruned_loss=0.1144, over 5687764.61 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3434, pruned_loss=0.09584, over 5691767.62 frames. ], batch size: 97, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 03:56:17,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.966e+02 1.286e+03 1.604e+03 2.440e+03 5.894e+03, threshold=3.208e+03, percent-clipped=6.0
+2023-03-11 03:56:21,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3971, 1.6113, 1.2778, 1.5267], device='cuda:0'), covar=tensor([0.0748, 0.0301, 0.0350, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 03:56:29,995 INFO [train.py:968] (0/2) Epoch 21, batch 39700, giga_loss[loss=0.3161, simple_loss=0.3868, pruned_loss=0.1227, over 27973.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3498, pruned_loss=0.1007, over 5692823.65 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3618, pruned_loss=0.115, over 5679490.20 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3468, pruned_loss=0.09799, over 5696575.61 frames. ], batch size: 412, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:56:59,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=952088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:57:01,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=952091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:57:09,714 INFO [train.py:968] (0/2) Epoch 21, batch 39750, giga_loss[loss=0.2742, simple_loss=0.3522, pruned_loss=0.09812, over 28885.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3516, pruned_loss=0.1011, over 5707773.37 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3613, pruned_loss=0.1146, over 5684199.82 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3494, pruned_loss=0.09905, over 5707171.77 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:57:24,872 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=952120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:57:37,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.692e+02 1.341e+03 1.831e+03 2.811e+03 1.074e+04, threshold=3.661e+03, percent-clipped=17.0
+2023-03-11 03:57:43,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4907, 1.8041, 1.4744, 1.3858], device='cuda:0'), covar=tensor([0.2314, 0.2383, 0.2625, 0.2171], device='cuda:0'), in_proj_covar=tensor([0.1499, 0.1084, 0.1323, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 03:57:50,162 INFO [train.py:968] (0/2) Epoch 21, batch 39800, giga_loss[loss=0.2898, simple_loss=0.3625, pruned_loss=0.1086, over 28651.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3519, pruned_loss=0.1011, over 5710794.25 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.361, pruned_loss=0.1144, over 5687364.23 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3502, pruned_loss=0.09937, over 5707837.85 frames. ], batch size: 71, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:58:31,941 INFO [train.py:968] (0/2) Epoch 21, batch 39850, libri_loss[loss=0.3352, simple_loss=0.3917, pruned_loss=0.1394, over 19136.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3515, pruned_loss=0.101, over 5705551.36 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.361, pruned_loss=0.1144, over 5683143.83 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3499, pruned_loss=0.0993, over 5708559.44 frames. ], batch size: 187, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:58:34,278 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:58:53,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4643, 1.7899, 1.5881, 1.7317], device='cuda:0'), covar=tensor([0.0734, 0.0274, 0.0312, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 03:58:54,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2921, 0.7685, 0.8333, 1.4027], device='cuda:0'), covar=tensor([0.0732, 0.0368, 0.0370, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 03:58:59,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.390e+02 1.304e+03 1.560e+03 1.834e+03 5.561e+03, threshold=3.119e+03, percent-clipped=5.0
+2023-03-11 03:59:12,696 INFO [train.py:968] (0/2) Epoch 21, batch 39900, giga_loss[loss=0.2589, simple_loss=0.3274, pruned_loss=0.09517, over 28862.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3519, pruned_loss=0.1013, over 5711542.34 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5685813.40 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3501, pruned_loss=0.09946, over 5711988.86 frames. ], batch size: 86, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:59:49,260 INFO [train.py:968] (0/2) Epoch 21, batch 39950, giga_loss[loss=0.2192, simple_loss=0.298, pruned_loss=0.07025, over 28495.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3505, pruned_loss=0.1007, over 5710472.13 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1153, over 5683394.70 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3479, pruned_loss=0.09828, over 5713234.64 frames. ], batch size: 71, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:00:14,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7458, 1.8193, 1.9957, 1.5348], device='cuda:0'), covar=tensor([0.1999, 0.2485, 0.1614, 0.1799], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0699, 0.0946, 0.0843], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 04:00:16,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.102e+02 1.303e+03 1.672e+03 2.327e+03 6.904e+03, threshold=3.345e+03, percent-clipped=15.0
+2023-03-11 04:00:26,428 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=952347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:00:29,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=952350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:00:29,623 INFO [train.py:968] (0/2) Epoch 21, batch 40000, giga_loss[loss=0.2686, simple_loss=0.3507, pruned_loss=0.0933, over 28752.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3471, pruned_loss=0.09914, over 5712883.60 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3624, pruned_loss=0.1152, over 5690021.34 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3446, pruned_loss=0.09679, over 5710123.61 frames. ], batch size: 284, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:00:30,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9871, 2.0901, 2.0292, 1.8603], device='cuda:0'), covar=tensor([0.2088, 0.2681, 0.2254, 0.2586], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0754, 0.0719, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 04:00:42,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4162, 2.1377, 1.6985, 0.6878], device='cuda:0'), covar=tensor([0.6739, 0.3051, 0.4070, 0.7009], device='cuda:0'), in_proj_covar=tensor([0.1737, 0.1638, 0.1599, 0.1417], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 04:00:53,429 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=952379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:00:56,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1738, 1.7098, 1.2771, 0.4541], device='cuda:0'), covar=tensor([0.3232, 0.1836, 0.2766, 0.4465], device='cuda:0'), in_proj_covar=tensor([0.1736, 0.1637, 0.1599, 0.1416], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 04:01:08,205 INFO [train.py:968] (0/2) Epoch 21, batch 40050, giga_loss[loss=0.2466, simple_loss=0.328, pruned_loss=0.08262, over 28999.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3437, pruned_loss=0.09743, over 5715718.38 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.362, pruned_loss=0.115, over 5693569.39 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3415, pruned_loss=0.09508, over 5711168.74 frames. ], batch size: 145, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:01:27,567 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 04:01:34,708 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.610e+02 1.177e+03 1.434e+03 2.029e+03 4.626e+03, threshold=2.868e+03, percent-clipped=6.0
+2023-03-11 04:01:48,181 INFO [train.py:968] (0/2) Epoch 21, batch 40100, giga_loss[loss=0.2994, simple_loss=0.3787, pruned_loss=0.11, over 27954.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3441, pruned_loss=0.09697, over 5714020.97 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.362, pruned_loss=0.115, over 5695120.84 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3419, pruned_loss=0.09466, over 5710038.23 frames. ], batch size: 412, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:01:54,044 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=952460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:02:27,306 INFO [train.py:968] (0/2) Epoch 21, batch 40150, giga_loss[loss=0.2529, simple_loss=0.333, pruned_loss=0.08637, over 28824.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3446, pruned_loss=0.09622, over 5709476.81 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3615, pruned_loss=0.1148, over 5699403.65 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3429, pruned_loss=0.09429, over 5703045.31 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:02:30,154 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=952503.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 04:02:56,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.481e+02 1.268e+03 1.652e+03 2.277e+03 5.619e+03, threshold=3.305e+03, percent-clipped=16.0
+2023-03-11 04:03:06,629 INFO [train.py:968] (0/2) Epoch 21, batch 40200, giga_loss[loss=0.2401, simple_loss=0.3093, pruned_loss=0.08542, over 28525.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3453, pruned_loss=0.09676, over 5714182.08 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3618, pruned_loss=0.1149, over 5704519.67 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3433, pruned_loss=0.09472, over 5704579.06 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:03:10,327 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 04:03:29,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7461, 1.8469, 1.9688, 1.5274], device='cuda:0'), covar=tensor([0.2007, 0.2440, 0.1597, 0.1786], device='cuda:0'), in_proj_covar=tensor([0.0901, 0.0699, 0.0947, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 04:03:45,273 INFO [train.py:968] (0/2) Epoch 21, batch 40250, giga_loss[loss=0.292, simple_loss=0.3596, pruned_loss=0.1123, over 29040.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3441, pruned_loss=0.09742, over 5720039.85 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3615, pruned_loss=0.1148, over 5709486.21 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3424, pruned_loss=0.09543, over 5708258.30 frames. ], batch size: 164, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:03:46,958 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-11 04:03:48,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4089, 1.5933, 1.1805, 1.1773], device='cuda:0'), covar=tensor([0.1071, 0.0694, 0.1198, 0.1273], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0444, 0.0515, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 04:03:49,745 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=952608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:03:53,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5911, 1.7827, 1.3988, 1.7501], device='cuda:0'), covar=tensor([0.2611, 0.2761, 0.3168, 0.2433], device='cuda:0'), in_proj_covar=tensor([0.1501, 0.1087, 0.1323, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 04:04:12,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.528e+02 1.163e+03 1.465e+03 2.108e+03 7.817e+03, threshold=2.929e+03, percent-clipped=5.0
+2023-03-11 04:04:24,139 INFO [train.py:968] (0/2) Epoch 21, batch 40300, giga_loss[loss=0.2246, simple_loss=0.3, pruned_loss=0.0746, over 28795.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3425, pruned_loss=0.09753, over 5726462.54 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3614, pruned_loss=0.1147, over 5710662.74 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3409, pruned_loss=0.0958, over 5716350.93 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:04:36,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3042, 1.5276, 1.5411, 1.3208], device='cuda:0'), covar=tensor([0.1615, 0.1662, 0.2013, 0.1948], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0751, 0.0716, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 04:04:39,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2445, 1.1868, 3.3769, 3.0566], device='cuda:0'), covar=tensor([0.1543, 0.2789, 0.0456, 0.2031], device='cuda:0'), in_proj_covar=tensor([0.0751, 0.0641, 0.0953, 0.0901], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 04:05:05,668 INFO [train.py:968] (0/2) Epoch 21, batch 40350, giga_loss[loss=0.2718, simple_loss=0.3412, pruned_loss=0.1012, over 28924.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3397, pruned_loss=0.09724, over 5712657.03 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3614, pruned_loss=0.1147, over 5704716.78 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3382, pruned_loss=0.09556, over 5709455.17 frames. ], batch size: 227, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:05:33,457 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.546e+02 1.298e+03 1.630e+03 2.176e+03 5.059e+03, threshold=3.260e+03, percent-clipped=12.0
+2023-03-11 04:05:43,800 INFO [train.py:968] (0/2) Epoch 21, batch 40400, giga_loss[loss=0.2804, simple_loss=0.3569, pruned_loss=0.102, over 28703.00 frames. ], tot_loss[loss=0.268, simple_loss=0.34, pruned_loss=0.098, over 5714651.42 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3616, pruned_loss=0.1146, over 5709449.25 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3381, pruned_loss=0.09634, over 5707915.76 frames. ], batch size: 242, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:06:26,523 INFO [train.py:968] (0/2) Epoch 21, batch 40450, giga_loss[loss=0.2141, simple_loss=0.2846, pruned_loss=0.07174, over 28640.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3376, pruned_loss=0.09675, over 5707517.76 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3618, pruned_loss=0.1147, over 5707498.78 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3357, pruned_loss=0.09525, over 5704145.42 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:06:54,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:06:56,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.712e+02 1.217e+03 1.569e+03 1.954e+03 4.874e+03, threshold=3.138e+03, percent-clipped=3.0
+2023-03-11 04:07:06,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-11 04:07:07,287 INFO [train.py:968] (0/2) Epoch 21, batch 40500, giga_loss[loss=0.2142, simple_loss=0.2871, pruned_loss=0.07064, over 28474.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3338, pruned_loss=0.09513, over 5703473.10 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.115, over 5704586.91 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3317, pruned_loss=0.09338, over 5703170.62 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:07:20,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-11 04:07:26,340 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952878.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 04:07:36,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3898, 2.0585, 1.5308, 0.6284], device='cuda:0'), covar=tensor([0.6273, 0.2936, 0.4363, 0.6936], device='cuda:0'), in_proj_covar=tensor([0.1732, 0.1627, 0.1595, 0.1412], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 04:07:43,822 INFO [train.py:968] (0/2) Epoch 21, batch 40550, giga_loss[loss=0.2503, simple_loss=0.3186, pruned_loss=0.09102, over 28889.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3313, pruned_loss=0.0941, over 5704252.00 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3625, pruned_loss=0.1153, over 5699666.90 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3282, pruned_loss=0.0917, over 5708229.13 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:08:12,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.511e+02 1.453e+03 1.893e+03 2.895e+03 9.972e+03, threshold=3.787e+03, percent-clipped=15.0
+2023-03-11 04:08:23,323 INFO [train.py:968] (0/2) Epoch 21, batch 40600, giga_loss[loss=0.239, simple_loss=0.3214, pruned_loss=0.07825, over 28976.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3315, pruned_loss=0.09358, over 5713927.16 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3627, pruned_loss=0.1154, over 5703815.98 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3281, pruned_loss=0.09114, over 5713597.83 frames. ], batch size: 128, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:08:24,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4232, 1.6359, 1.4926, 1.2822], device='cuda:0'), covar=tensor([0.3082, 0.2661, 0.2228, 0.2645], device='cuda:0'), in_proj_covar=tensor([0.1974, 0.1909, 0.1829, 0.1966], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 04:08:32,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5074, 4.7437, 1.7540, 1.9417], device='cuda:0'), covar=tensor([0.0989, 0.0334, 0.0922, 0.1196], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0548, 0.0383, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-11 04:08:40,302 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 04:08:45,165 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=952978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:08:47,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=952981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:08:49,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:08:59,011 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=952994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:09:03,035 INFO [train.py:968] (0/2) Epoch 21, batch 40650, giga_loss[loss=0.2625, simple_loss=0.3467, pruned_loss=0.08917, over 28722.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3357, pruned_loss=0.09555, over 5710665.94 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3625, pruned_loss=0.1153, over 5707960.27 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3321, pruned_loss=0.09287, over 5706849.04 frames. ], batch size: 284, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:09:11,652 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=953010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:09:19,314 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=953021.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:09:21,147 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=953024.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:09:32,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.422e+02 1.209e+03 1.525e+03 2.304e+03 1.150e+04, threshold=3.050e+03, percent-clipped=8.0
+2023-03-11 04:09:42,727 INFO [train.py:968] (0/2) Epoch 21, batch 40700, giga_loss[loss=0.2708, simple_loss=0.3457, pruned_loss=0.09792, over 28915.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3378, pruned_loss=0.09612, over 5716295.92 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1147, over 5710431.97 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3352, pruned_loss=0.09414, over 5711156.75 frames. ], batch size: 145, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:09:44,499 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=953053.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:10:21,020 INFO [train.py:968] (0/2) Epoch 21, batch 40750, giga_loss[loss=0.2862, simple_loss=0.3655, pruned_loss=0.1034, over 29005.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3416, pruned_loss=0.09765, over 5709031.77 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.362, pruned_loss=0.1149, over 5710562.27 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3388, pruned_loss=0.09559, over 5705091.95 frames. ], batch size: 164, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:10:42,577 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=953126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:10:45,424 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=953129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:10:51,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.439e+02 1.262e+03 1.442e+03 2.125e+03 4.385e+03, threshold=2.884e+03, percent-clipped=9.0
+2023-03-11 04:10:57,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3199, 3.3436, 1.4672, 1.4905], device='cuda:0'), covar=tensor([0.1013, 0.0348, 0.0952, 0.1355], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0548, 0.0382, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-11 04:11:01,480 INFO [train.py:968] (0/2) Epoch 21, batch 40800, giga_loss[loss=0.2857, simple_loss=0.3592, pruned_loss=0.1061, over 29144.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3454, pruned_loss=0.09935, over 5710266.84 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3618, pruned_loss=0.1148, over 5713505.33 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3429, pruned_loss=0.09743, over 5704236.02 frames. ], batch size: 128, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:11:06,449 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=953158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:11:35,896 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2163, 2.2671, 1.7421, 1.8871], device='cuda:0'), covar=tensor([0.0962, 0.0730, 0.1030, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0449, 0.0519, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 04:11:40,072 INFO [train.py:968] (0/2) Epoch 21, batch 40850, libri_loss[loss=0.3171, simple_loss=0.3826, pruned_loss=0.1258, over 25874.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3471, pruned_loss=0.1004, over 5701402.57 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.362, pruned_loss=0.1151, over 5704519.79 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3443, pruned_loss=0.09802, over 5704843.22 frames. ], batch size: 136, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:12:19,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.997e+02 1.443e+03 1.746e+03 2.518e+03 6.468e+03, threshold=3.493e+03, percent-clipped=18.0
+2023-03-11 04:12:31,353 INFO [train.py:968] (0/2) Epoch 21, batch 40900, giga_loss[loss=0.2908, simple_loss=0.3746, pruned_loss=0.1035, over 28581.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3535, pruned_loss=0.1061, over 5700571.11 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3622, pruned_loss=0.1152, over 5708328.36 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3509, pruned_loss=0.1039, over 5699775.96 frames. ], batch size: 307, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:12:42,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2291, 1.2679, 3.5568, 3.2438], device='cuda:0'), covar=tensor([0.1571, 0.2724, 0.0483, 0.1693], device='cuda:0'), in_proj_covar=tensor([0.0752, 0.0642, 0.0957, 0.0905], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 04:13:05,234 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 04:13:13,617 INFO [train.py:968] (0/2) Epoch 21, batch 40950, giga_loss[loss=0.3529, simple_loss=0.4177, pruned_loss=0.144, over 28853.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.359, pruned_loss=0.1103, over 5687271.11 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1152, over 5695769.42 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3567, pruned_loss=0.1083, over 5696940.44 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:13:26,481 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-11 04:13:47,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.911e+02 1.795e+03 2.516e+03 3.521e+03 9.485e+03, threshold=5.032e+03, percent-clipped=27.0
+2023-03-11 04:13:59,474 INFO [train.py:968] (0/2) Epoch 21, batch 41000, giga_loss[loss=0.3215, simple_loss=0.388, pruned_loss=0.1274, over 28312.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3657, pruned_loss=0.1151, over 5689174.74 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3622, pruned_loss=0.1153, over 5699706.23 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3639, pruned_loss=0.1134, over 5693059.68 frames. ], batch size: 368, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:14:17,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=953369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:14:45,201 INFO [train.py:968] (0/2) Epoch 21, batch 41050, giga_loss[loss=0.3208, simple_loss=0.3893, pruned_loss=0.1262, over 28815.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3727, pruned_loss=0.1212, over 5681879.43 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3624, pruned_loss=0.1154, over 5692176.50 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3711, pruned_loss=0.1197, over 5691430.16 frames. ], batch size: 186, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:15:14,703 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=953434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:15:17,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.718e+03 2.177e+03 2.880e+03 8.975e+03, threshold=4.355e+03, percent-clipped=3.0
+2023-03-11 04:15:27,846 INFO [train.py:968] (0/2) Epoch 21, batch 41100, giga_loss[loss=0.2943, simple_loss=0.3632, pruned_loss=0.1127, over 28753.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3786, pruned_loss=0.1263, over 5677972.34 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3626, pruned_loss=0.1155, over 5689207.03 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3776, pruned_loss=0.1253, over 5687862.98 frames. ], batch size: 262, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:15:44,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-11 04:16:14,945 INFO [train.py:968] (0/2) Epoch 21, batch 41150, giga_loss[loss=0.352, simple_loss=0.4036, pruned_loss=0.1502, over 28690.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3819, pruned_loss=0.1296, over 5654843.66 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1153, over 5684625.74 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.382, pruned_loss=0.1296, over 5666104.52 frames. ], batch size: 242, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:16:16,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=953503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:16:25,403 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=953512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:16:29,678 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=953515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:16:52,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.896e+03 2.496e+03 3.766e+03 1.021e+04, threshold=4.992e+03, percent-clipped=12.0
+2023-03-11 04:16:56,558 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=953544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:16:56,814 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-11 04:17:04,750 INFO [train.py:968] (0/2) Epoch 21, batch 41200, giga_loss[loss=0.2811, simple_loss=0.3473, pruned_loss=0.1074, over 28812.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3836, pruned_loss=0.132, over 5657405.49 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3625, pruned_loss=0.1155, over 5690642.48 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3842, pruned_loss=0.1322, over 5660216.29 frames. ], batch size: 99, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:17:58,453 INFO [train.py:968] (0/2) Epoch 21, batch 41250, giga_loss[loss=0.386, simple_loss=0.4121, pruned_loss=0.18, over 23652.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3868, pruned_loss=0.136, over 5639391.20 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3625, pruned_loss=0.1155, over 5683094.78 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3876, pruned_loss=0.1364, over 5648423.36 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:18:34,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.826e+03 2.252e+03 3.773e+03 1.146e+04, threshold=4.505e+03, percent-clipped=13.0
+2023-03-11 04:18:43,404 INFO [train.py:968] (0/2) Epoch 21, batch 41300, libri_loss[loss=0.3189, simple_loss=0.3751, pruned_loss=0.1313, over 29500.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3884, pruned_loss=0.1382, over 5625044.63 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3618, pruned_loss=0.1152, over 5679224.28 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3907, pruned_loss=0.1397, over 5633183.87 frames. ], batch size: 82, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:19:33,262 INFO [train.py:968] (0/2) Epoch 21, batch 41350, giga_loss[loss=0.2941, simple_loss=0.363, pruned_loss=0.1126, over 28814.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3915, pruned_loss=0.1404, over 5615001.58 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3625, pruned_loss=0.1157, over 5670446.74 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3932, pruned_loss=0.1414, over 5628036.17 frames. ], batch size: 99, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:20:17,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.284e+03 2.039e+03 2.425e+03 3.400e+03 1.288e+04, threshold=4.850e+03, percent-clipped=7.0
+2023-03-11 04:20:25,897 INFO [train.py:968] (0/2) Epoch 21, batch 41400, giga_loss[loss=0.3249, simple_loss=0.3849, pruned_loss=0.1325, over 28921.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3912, pruned_loss=0.1412, over 5614192.91 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3619, pruned_loss=0.1153, over 5676118.65 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3939, pruned_loss=0.1431, over 5618116.89 frames. ], batch size: 174, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:21:09,126 INFO [train.py:968] (0/2) Epoch 21, batch 41450, libri_loss[loss=0.2789, simple_loss=0.3551, pruned_loss=0.1014, over 29527.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3894, pruned_loss=0.1401, over 5625550.49 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3618, pruned_loss=0.1152, over 5671563.41 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3928, pruned_loss=0.1428, over 5631736.93 frames. ], batch size: 84, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:21:17,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=953809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:21:47,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.792e+03 2.486e+03 3.462e+03 9.526e+03, threshold=4.971e+03, percent-clipped=13.0
+2023-03-11 04:21:57,351 INFO [train.py:968] (0/2) Epoch 21, batch 41500, giga_loss[loss=0.2657, simple_loss=0.3547, pruned_loss=0.08837, over 28952.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.389, pruned_loss=0.1384, over 5643352.05 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3617, pruned_loss=0.1152, over 5673472.15 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3921, pruned_loss=0.1408, over 5646103.89 frames. ], batch size: 145, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:22:28,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=953878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:22:53,246 INFO [train.py:968] (0/2) Epoch 21, batch 41550, giga_loss[loss=0.3254, simple_loss=0.3953, pruned_loss=0.1277, over 28569.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3885, pruned_loss=0.1367, over 5652350.34 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3613, pruned_loss=0.1149, over 5677224.75 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3917, pruned_loss=0.1392, over 5650731.53 frames. ], batch size: 307, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:23:14,743 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-11 04:23:33,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.960e+02 1.670e+03 2.196e+03 3.080e+03 7.991e+03, threshold=4.391e+03, percent-clipped=8.0
+2023-03-11 04:23:43,401 INFO [train.py:968] (0/2) Epoch 21, batch 41600, giga_loss[loss=0.3035, simple_loss=0.3804, pruned_loss=0.1133, over 28871.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3895, pruned_loss=0.1371, over 5651264.19 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3608, pruned_loss=0.1146, over 5679324.94 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3931, pruned_loss=0.1398, over 5647486.22 frames. ], batch size: 174, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:23:45,740 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=953952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:23:49,014 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=953955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:24:17,163 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=953984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:24:31,853 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-954000.pt
+2023-03-11 04:24:32,838 INFO [train.py:968] (0/2) Epoch 21, batch 41650, giga_loss[loss=0.3303, simple_loss=0.385, pruned_loss=0.1378, over 27512.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3872, pruned_loss=0.1352, over 5656339.97 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3608, pruned_loss=0.1146, over 5687602.74 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3911, pruned_loss=0.1382, over 5644479.26 frames. ], batch size: 472, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:24:53,166 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=954021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:24:56,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=954024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:25:09,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.771e+03 2.466e+03 3.330e+03 8.997e+03, threshold=4.932e+03, percent-clipped=12.0
+2023-03-11 04:25:20,340 INFO [train.py:968] (0/2) Epoch 21, batch 41700, giga_loss[loss=0.3111, simple_loss=0.3807, pruned_loss=0.1207, over 28902.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.386, pruned_loss=0.1333, over 5644846.75 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3607, pruned_loss=0.1143, over 5684585.27 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3901, pruned_loss=0.1366, over 5637409.02 frames. ], batch size: 285, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:25:22,216 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=954053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:25:42,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4629, 1.9470, 1.8678, 1.4128], device='cuda:0'), covar=tensor([0.3446, 0.2324, 0.2584, 0.2994], device='cuda:0'), in_proj_covar=tensor([0.1984, 0.1913, 0.1839, 0.1971], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 04:26:02,907 INFO [train.py:968] (0/2) Epoch 21, batch 41750, giga_loss[loss=0.2917, simple_loss=0.3607, pruned_loss=0.1113, over 28270.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3817, pruned_loss=0.1288, over 5657111.86 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3604, pruned_loss=0.1142, over 5682868.53 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3864, pruned_loss=0.1324, over 5651308.79 frames. ], batch size: 368, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:26:34,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3818, 1.6259, 1.5444, 1.5093], device='cuda:0'), covar=tensor([0.1743, 0.1935, 0.2046, 0.1854], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0754, 0.0716, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 04:26:39,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.778e+03 2.407e+03 3.380e+03 1.982e+04, threshold=4.815e+03, percent-clipped=5.0
+2023-03-11 04:26:43,945 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:26:49,604 INFO [train.py:968] (0/2) Epoch 21, batch 41800, giga_loss[loss=0.2734, simple_loss=0.3542, pruned_loss=0.09632, over 28741.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3781, pruned_loss=0.1257, over 5654226.45 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3605, pruned_loss=0.1143, over 5685137.37 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3821, pruned_loss=0.1288, over 5647097.46 frames. ], batch size: 242, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:27:40,382 INFO [train.py:968] (0/2) Epoch 21, batch 41850, giga_loss[loss=0.3328, simple_loss=0.3927, pruned_loss=0.1364, over 28845.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3752, pruned_loss=0.1241, over 5648954.21 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3604, pruned_loss=0.1142, over 5687505.79 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3787, pruned_loss=0.1267, over 5640820.10 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:27:42,812 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-11 04:27:44,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6136, 1.7138, 1.8361, 1.3800], device='cuda:0'), covar=tensor([0.1821, 0.2524, 0.1487, 0.1739], device='cuda:0'), in_proj_covar=tensor([0.0896, 0.0698, 0.0941, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 04:28:20,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.870e+02 1.563e+03 1.983e+03 2.858e+03 5.796e+03, threshold=3.966e+03, percent-clipped=4.0
+2023-03-11 04:28:28,805 INFO [train.py:968] (0/2) Epoch 21, batch 41900, giga_loss[loss=0.2969, simple_loss=0.3691, pruned_loss=0.1123, over 29013.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3752, pruned_loss=0.1243, over 5649688.80 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3605, pruned_loss=0.1145, over 5677500.27 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3781, pruned_loss=0.1263, over 5651407.09 frames. ], batch size: 128, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:28:32,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2828, 1.0185, 4.0225, 3.3850], device='cuda:0'), covar=tensor([0.1749, 0.3024, 0.0438, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0759, 0.0648, 0.0965, 0.0913], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 04:28:45,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3133, 2.7756, 1.3850, 1.4717], device='cuda:0'), covar=tensor([0.0948, 0.0379, 0.0864, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0555, 0.0385, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 04:29:17,975 INFO [train.py:968] (0/2) Epoch 21, batch 41950, giga_loss[loss=0.2933, simple_loss=0.3659, pruned_loss=0.1103, over 28850.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3755, pruned_loss=0.1241, over 5657850.27 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3604, pruned_loss=0.1145, over 5679653.38 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.378, pruned_loss=0.1258, over 5656962.85 frames. ], batch size: 112, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:29:58,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.610e+03 2.130e+03 2.780e+03 6.542e+03, threshold=4.261e+03, percent-clipped=12.0
+2023-03-11 04:30:06,115 INFO [train.py:968] (0/2) Epoch 21, batch 42000, giga_loss[loss=0.2568, simple_loss=0.3324, pruned_loss=0.0906, over 28933.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3737, pruned_loss=0.1218, over 5672062.16 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3611, pruned_loss=0.115, over 5685036.35 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3754, pruned_loss=0.1229, over 5666119.84 frames. ], batch size: 112, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:30:06,119 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 04:30:15,117 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2039, simple_loss=0.3106, pruned_loss=0.0486, over 944034.00 frames. 
+2023-03-11 04:30:15,118 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 04:30:54,401 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-11 04:31:01,574 INFO [train.py:968] (0/2) Epoch 21, batch 42050, giga_loss[loss=0.2795, simple_loss=0.3612, pruned_loss=0.0989, over 28657.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3734, pruned_loss=0.1191, over 5676201.90 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3606, pruned_loss=0.1148, over 5682252.04 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3757, pruned_loss=0.1203, over 5673260.99 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:31:21,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0613, 5.2693, 1.9781, 2.5747], device='cuda:0'), covar=tensor([0.0880, 0.0431, 0.0868, 0.1005], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0554, 0.0385, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 04:31:42,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.692e+03 1.963e+03 2.634e+03 6.437e+03, threshold=3.926e+03, percent-clipped=6.0
+2023-03-11 04:31:52,637 INFO [train.py:968] (0/2) Epoch 21, batch 42100, giga_loss[loss=0.2924, simple_loss=0.366, pruned_loss=0.1094, over 28900.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3751, pruned_loss=0.1196, over 5653671.88 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3609, pruned_loss=0.115, over 5666005.63 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.377, pruned_loss=0.1205, over 5665490.59 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:32:29,237 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8625, 1.2082, 2.8603, 2.8345], device='cuda:0'), covar=tensor([0.1640, 0.2439, 0.0587, 0.0978], device='cuda:0'), in_proj_covar=tensor([0.0756, 0.0645, 0.0961, 0.0910], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 04:32:40,801 INFO [train.py:968] (0/2) Epoch 21, batch 42150, libri_loss[loss=0.2371, simple_loss=0.3126, pruned_loss=0.08077, over 29553.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3755, pruned_loss=0.1206, over 5659674.48 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.361, pruned_loss=0.115, over 5668436.50 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3771, pruned_loss=0.1215, over 5666490.82 frames. ], batch size: 75, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:32:56,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=954521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:33:16,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.700e+03 2.206e+03 2.937e+03 7.970e+03, threshold=4.413e+03, percent-clipped=13.0
+2023-03-11 04:33:24,245 INFO [train.py:968] (0/2) Epoch 21, batch 42200, giga_loss[loss=0.2683, simple_loss=0.3442, pruned_loss=0.09624, over 28645.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3755, pruned_loss=0.121, over 5665477.70 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3618, pruned_loss=0.1154, over 5669430.09 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3765, pruned_loss=0.1215, over 5669980.86 frames. ], batch size: 242, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:33:30,301 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 04:34:06,727 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:34:07,179 INFO [train.py:968] (0/2) Epoch 21, batch 42250, libri_loss[loss=0.3143, simple_loss=0.3805, pruned_loss=0.1241, over 29155.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3738, pruned_loss=0.1212, over 5662539.17 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3619, pruned_loss=0.1155, over 5663907.54 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3749, pruned_loss=0.1217, over 5670244.30 frames. ], batch size: 101, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:34:11,770 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:34:37,823 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:34:45,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.737e+03 2.224e+03 3.340e+03 7.607e+03, threshold=4.448e+03, percent-clipped=11.0
+2023-03-11 04:34:53,259 INFO [train.py:968] (0/2) Epoch 21, batch 42300, libri_loss[loss=0.2658, simple_loss=0.3412, pruned_loss=0.09521, over 29530.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3714, pruned_loss=0.1203, over 5660602.43 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.361, pruned_loss=0.1148, over 5671214.18 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3735, pruned_loss=0.1215, over 5659699.95 frames. ], batch size: 79, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:35:04,423 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=954664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:35:06,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=954667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:35:08,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:35:20,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.5688, 3.4048, 3.2338, 1.7793], device='cuda:0'), covar=tensor([0.0806, 0.0899, 0.0807, 0.2316], device='cuda:0'), in_proj_covar=tensor([0.1234, 0.1148, 0.0968, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 04:35:37,850 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=954696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:35:43,144 INFO [train.py:968] (0/2) Epoch 21, batch 42350, giga_loss[loss=0.289, simple_loss=0.3708, pruned_loss=0.1036, over 28613.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3701, pruned_loss=0.1189, over 5668542.55 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3607, pruned_loss=0.1146, over 5672662.42 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3721, pruned_loss=0.12, over 5666549.67 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:36:20,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.584e+02 1.503e+03 2.022e+03 2.885e+03 6.028e+03, threshold=4.044e+03, percent-clipped=7.0
+2023-03-11 04:36:30,254 INFO [train.py:968] (0/2) Epoch 21, batch 42400, giga_loss[loss=0.3452, simple_loss=0.4015, pruned_loss=0.1444, over 28305.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3692, pruned_loss=0.1165, over 5687057.14 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3605, pruned_loss=0.1143, over 5678313.66 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3711, pruned_loss=0.1178, over 5680660.18 frames. ], batch size: 368, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:37:03,406 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1680, 1.1764, 3.5193, 3.1152], device='cuda:0'), covar=tensor([0.1709, 0.2816, 0.0536, 0.1190], device='cuda:0'), in_proj_covar=tensor([0.0760, 0.0648, 0.0966, 0.0914], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 04:37:16,259 INFO [train.py:968] (0/2) Epoch 21, batch 42450, giga_loss[loss=0.258, simple_loss=0.3339, pruned_loss=0.09099, over 28916.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3692, pruned_loss=0.1162, over 5686928.13 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.1141, over 5680390.54 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3712, pruned_loss=0.1174, over 5680275.09 frames. ], batch size: 106, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:37:16,461 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:37:55,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.667e+03 2.071e+03 2.814e+03 7.309e+03, threshold=4.142e+03, percent-clipped=11.0
+2023-03-11 04:37:57,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3552, 2.5866, 1.8138, 2.1128], device='cuda:0'), covar=tensor([0.0990, 0.0697, 0.1085, 0.1143], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0449, 0.0518, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 04:38:01,950 INFO [train.py:968] (0/2) Epoch 21, batch 42500, libri_loss[loss=0.2786, simple_loss=0.3585, pruned_loss=0.09934, over 29244.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.368, pruned_loss=0.1161, over 5681071.55 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3602, pruned_loss=0.114, over 5681637.99 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3699, pruned_loss=0.1172, over 5674636.51 frames. ], batch size: 94, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:38:49,249 INFO [train.py:968] (0/2) Epoch 21, batch 42550, giga_loss[loss=0.3154, simple_loss=0.3794, pruned_loss=0.1257, over 28589.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3677, pruned_loss=0.1167, over 5672215.45 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3605, pruned_loss=0.1143, over 5675475.56 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.369, pruned_loss=0.1173, over 5672990.32 frames. ], batch size: 307, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:38:51,364 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-11 04:39:08,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:39:22,796 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1533, 1.2530, 1.1817, 1.1325], device='cuda:0'), covar=tensor([0.1839, 0.1879, 0.1466, 0.1778], device='cuda:0'), in_proj_covar=tensor([0.1973, 0.1904, 0.1828, 0.1970], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 04:39:30,915 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.747e+03 2.145e+03 3.109e+03 6.654e+03, threshold=4.289e+03, percent-clipped=8.0
+2023-03-11 04:39:37,653 INFO [train.py:968] (0/2) Epoch 21, batch 42600, giga_loss[loss=0.2835, simple_loss=0.3459, pruned_loss=0.1106, over 28886.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3673, pruned_loss=0.1173, over 5676601.25 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3607, pruned_loss=0.1144, over 5680299.64 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3682, pruned_loss=0.1178, over 5672952.96 frames. ], batch size: 106, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:40:03,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=954975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:40:07,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=954980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:40:16,974 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954991.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:40:23,856 INFO [train.py:968] (0/2) Epoch 21, batch 42650, giga_loss[loss=0.3048, simple_loss=0.3727, pruned_loss=0.1185, over 28630.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3665, pruned_loss=0.1179, over 5673433.70 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3603, pruned_loss=0.1142, over 5686845.94 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3679, pruned_loss=0.1187, over 5664072.81 frames. ], batch size: 307, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:40:25,827 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6508, 1.7820, 1.3061, 1.3284], device='cuda:0'), covar=tensor([0.1015, 0.0672, 0.1092, 0.1203], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0449, 0.0518, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 04:40:30,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=955007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:40:43,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3689, 1.5267, 1.3148, 1.5077], device='cuda:0'), covar=tensor([0.0756, 0.0359, 0.0336, 0.0832], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 04:41:03,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.669e+03 2.096e+03 2.714e+03 1.073e+04, threshold=4.192e+03, percent-clipped=7.0
+2023-03-11 04:41:04,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=955045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:41:09,844 INFO [train.py:968] (0/2) Epoch 21, batch 42700, giga_loss[loss=0.3224, simple_loss=0.3854, pruned_loss=0.1297, over 27950.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3654, pruned_loss=0.1178, over 5681725.67 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3606, pruned_loss=0.1143, over 5689502.32 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3664, pruned_loss=0.1183, over 5671753.46 frames. ], batch size: 412, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:41:34,251 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-11 04:41:59,816 INFO [train.py:968] (0/2) Epoch 21, batch 42750, giga_loss[loss=0.2573, simple_loss=0.3337, pruned_loss=0.09049, over 28899.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3644, pruned_loss=0.1174, over 5687789.55 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3602, pruned_loss=0.1141, over 5694245.62 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3656, pruned_loss=0.1182, over 5675039.64 frames. ], batch size: 174, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:42:18,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:20,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:21,810 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:25,959 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:41,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.878e+03 2.255e+03 3.337e+03 1.512e+04, threshold=4.510e+03, percent-clipped=13.0
+2023-03-11 04:42:43,442 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3023, 3.2848, 1.4773, 1.4434], device='cuda:0'), covar=tensor([0.1013, 0.0373, 0.0892, 0.1340], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0554, 0.0384, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 04:42:47,762 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:47,807 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:48,193 INFO [train.py:968] (0/2) Epoch 21, batch 42800, giga_loss[loss=0.3042, simple_loss=0.3681, pruned_loss=0.1201, over 27942.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3647, pruned_loss=0.117, over 5682294.54 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3603, pruned_loss=0.1141, over 5687213.88 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3656, pruned_loss=0.1176, over 5678685.30 frames. ], batch size: 412, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:42:49,695 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:50,968 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:52,029 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-11 04:43:12,046 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=955176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:43:17,138 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:43:21,339 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:43:27,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:43:34,489 INFO [train.py:968] (0/2) Epoch 21, batch 42850, giga_loss[loss=0.3987, simple_loss=0.4203, pruned_loss=0.1885, over 26520.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3656, pruned_loss=0.1168, over 5678018.42 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3605, pruned_loss=0.1142, over 5690636.70 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3663, pruned_loss=0.1172, over 5671916.07 frames. ], batch size: 555, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:43:51,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:44:10,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.743e+02 1.495e+03 2.027e+03 2.770e+03 6.733e+03, threshold=4.054e+03, percent-clipped=4.0
+2023-03-11 04:44:17,289 INFO [train.py:968] (0/2) Epoch 21, batch 42900, libri_loss[loss=0.2903, simple_loss=0.3594, pruned_loss=0.1106, over 29522.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3646, pruned_loss=0.1153, over 5670636.38 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3606, pruned_loss=0.1143, over 5686052.15 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3651, pruned_loss=0.1156, over 5669508.17 frames. ], batch size: 83, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:44:21,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6349, 1.8964, 1.2412, 1.5433], device='cuda:0'), covar=tensor([0.1044, 0.0670, 0.1092, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0449, 0.0518, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 04:44:37,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6336, 1.7597, 1.6596, 1.5091], device='cuda:0'), covar=tensor([0.2636, 0.2603, 0.2399, 0.2464], device='cuda:0'), in_proj_covar=tensor([0.1971, 0.1900, 0.1827, 0.1967], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 04:44:41,776 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-11 04:45:00,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=955298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:45:05,217 INFO [train.py:968] (0/2) Epoch 21, batch 42950, giga_loss[loss=0.3441, simple_loss=0.3809, pruned_loss=0.1537, over 23659.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3664, pruned_loss=0.1166, over 5667566.19 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3611, pruned_loss=0.1146, over 5686152.29 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3665, pruned_loss=0.1165, over 5666285.65 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:45:21,513 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:45:26,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:45:44,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.835e+03 2.422e+03 3.487e+03 9.957e+03, threshold=4.845e+03, percent-clipped=17.0
+2023-03-11 04:45:49,753 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5852, 2.3319, 1.8157, 0.8247], device='cuda:0'), covar=tensor([0.5826, 0.2986, 0.3476, 0.6222], device='cuda:0'), in_proj_covar=tensor([0.1751, 0.1656, 0.1608, 0.1425], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 04:45:52,188 INFO [train.py:968] (0/2) Epoch 21, batch 43000, giga_loss[loss=0.3342, simple_loss=0.3907, pruned_loss=0.1389, over 28908.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3672, pruned_loss=0.118, over 5639799.48 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3607, pruned_loss=0.1146, over 5661892.86 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3676, pruned_loss=0.1181, over 5660722.76 frames. ], batch size: 174, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:45:52,411 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:46:05,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=955366.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:46:37,290 INFO [train.py:968] (0/2) Epoch 21, batch 43050, giga_loss[loss=0.2863, simple_loss=0.3491, pruned_loss=0.1118, over 28722.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.371, pruned_loss=0.1222, over 5645303.49 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3607, pruned_loss=0.1146, over 5657097.60 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1224, over 5665082.67 frames. ], batch size: 99, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:47:05,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5691, 1.3643, 4.8711, 3.5258], device='cuda:0'), covar=tensor([0.1724, 0.2758, 0.0399, 0.0912], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0650, 0.0968, 0.0916], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 04:47:22,797 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:47:23,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2099, 2.5780, 2.3003, 1.7966], device='cuda:0'), covar=tensor([0.3174, 0.2190, 0.2254, 0.2948], device='cuda:0'), in_proj_covar=tensor([0.1972, 0.1898, 0.1827, 0.1964], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 04:47:24,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.843e+03 2.408e+03 3.638e+03 7.860e+03, threshold=4.817e+03, percent-clipped=5.0
+2023-03-11 04:47:24,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:47:30,831 INFO [train.py:968] (0/2) Epoch 21, batch 43100, giga_loss[loss=0.3073, simple_loss=0.374, pruned_loss=0.1203, over 28884.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3731, pruned_loss=0.1254, over 5637482.90 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3607, pruned_loss=0.1146, over 5661230.86 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3737, pruned_loss=0.1257, over 5649109.10 frames. ], batch size: 164, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:47:50,926 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:47:53,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2072, 1.5033, 1.1275, 0.5993], device='cuda:0'), covar=tensor([0.2583, 0.1792, 0.2388, 0.4676], device='cuda:0'), in_proj_covar=tensor([0.1749, 0.1656, 0.1608, 0.1425], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 04:48:18,406 INFO [train.py:968] (0/2) Epoch 21, batch 43150, giga_loss[loss=0.2842, simple_loss=0.3533, pruned_loss=0.1076, over 28676.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3742, pruned_loss=0.127, over 5649510.32 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3604, pruned_loss=0.1144, over 5665215.65 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3753, pruned_loss=0.1276, over 5654916.13 frames. ], batch size: 262, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:48:26,441 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955509.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 04:48:29,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955512.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 04:48:51,328 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955541.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:48:54,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.139e+03 1.836e+03 2.343e+03 3.235e+03 8.052e+03, threshold=4.685e+03, percent-clipped=7.0
+2023-03-11 04:49:00,421 INFO [train.py:968] (0/2) Epoch 21, batch 43200, giga_loss[loss=0.3213, simple_loss=0.3911, pruned_loss=0.1257, over 28859.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3712, pruned_loss=0.1243, over 5668742.31 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.36, pruned_loss=0.114, over 5672705.49 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3728, pruned_loss=0.1255, over 5666146.83 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:49:42,778 INFO [train.py:968] (0/2) Epoch 21, batch 43250, giga_loss[loss=0.267, simple_loss=0.3431, pruned_loss=0.09548, over 29007.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3699, pruned_loss=0.1229, over 5675470.08 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3603, pruned_loss=0.1142, over 5679821.99 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3714, pruned_loss=0.1241, over 5666715.45 frames. ], batch size: 128, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:50:21,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.206e+03 1.685e+03 2.513e+03 3.180e+03 8.321e+03, threshold=5.026e+03, percent-clipped=8.0
+2023-03-11 04:50:25,413 INFO [train.py:968] (0/2) Epoch 21, batch 43300, giga_loss[loss=0.2919, simple_loss=0.3686, pruned_loss=0.1076, over 29045.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3698, pruned_loss=0.1209, over 5679070.35 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3602, pruned_loss=0.1142, over 5676698.60 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3713, pruned_loss=0.1221, over 5674444.56 frames. ], batch size: 155, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:50:30,584 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-11 04:51:07,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2393, 1.5764, 1.4991, 1.3629], device='cuda:0'), covar=tensor([0.1881, 0.1582, 0.2294, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0758, 0.0720, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 04:51:12,338 INFO [train.py:968] (0/2) Epoch 21, batch 43350, giga_loss[loss=0.2944, simple_loss=0.3411, pruned_loss=0.1239, over 23539.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.367, pruned_loss=0.1188, over 5671284.38 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3602, pruned_loss=0.1141, over 5678351.34 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3684, pruned_loss=0.1199, over 5666315.96 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:51:13,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5095, 1.7781, 1.6886, 1.5068], device='cuda:0'), covar=tensor([0.2514, 0.2295, 0.2383, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.1973, 0.1900, 0.1828, 0.1965], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 04:51:36,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8772, 1.1885, 2.9117, 2.8097], device='cuda:0'), covar=tensor([0.1720, 0.2516, 0.0607, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0652, 0.0971, 0.0919], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 04:51:49,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.821e+03 2.350e+03 3.284e+03 6.733e+03, threshold=4.699e+03, percent-clipped=4.0
+2023-03-11 04:51:55,625 INFO [train.py:968] (0/2) Epoch 21, batch 43400, giga_loss[loss=0.2772, simple_loss=0.3395, pruned_loss=0.1075, over 28928.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3659, pruned_loss=0.1187, over 5657835.75 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3608, pruned_loss=0.1145, over 5667276.52 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3667, pruned_loss=0.1194, over 5664407.97 frames. ], batch size: 106, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:52:40,961 INFO [train.py:968] (0/2) Epoch 21, batch 43450, giga_loss[loss=0.322, simple_loss=0.3605, pruned_loss=0.1417, over 23490.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3647, pruned_loss=0.1185, over 5649498.70 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3606, pruned_loss=0.1144, over 5663717.70 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3656, pruned_loss=0.1193, over 5657838.93 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:53:18,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.725e+03 2.185e+03 3.173e+03 7.488e+03, threshold=4.369e+03, percent-clipped=6.0
+2023-03-11 04:53:23,408 INFO [train.py:968] (0/2) Epoch 21, batch 43500, libri_loss[loss=0.3076, simple_loss=0.3763, pruned_loss=0.1194, over 29657.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3672, pruned_loss=0.1205, over 5661528.16 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3604, pruned_loss=0.1141, over 5672531.19 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3683, pruned_loss=0.1215, over 5659820.87 frames. ], batch size: 91, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:54:10,461 INFO [train.py:968] (0/2) Epoch 21, batch 43550, giga_loss[loss=0.2871, simple_loss=0.3688, pruned_loss=0.1027, over 28517.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3715, pruned_loss=0.1223, over 5660444.28 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3607, pruned_loss=0.1145, over 5668638.67 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5662598.37 frames. ], batch size: 78, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:54:55,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.874e+02 1.576e+03 2.021e+03 2.815e+03 9.032e+03, threshold=4.041e+03, percent-clipped=10.0
+2023-03-11 04:55:02,801 INFO [train.py:968] (0/2) Epoch 21, batch 43600, libri_loss[loss=0.4067, simple_loss=0.4317, pruned_loss=0.1909, over 19195.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.374, pruned_loss=0.1209, over 5646739.10 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.361, pruned_loss=0.1147, over 5661710.84 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3745, pruned_loss=0.1213, over 5655567.85 frames. ], batch size: 187, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:55:44,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5724, 1.6997, 1.7114, 1.5159], device='cuda:0'), covar=tensor([0.2833, 0.2567, 0.2065, 0.2493], device='cuda:0'), in_proj_covar=tensor([0.1974, 0.1900, 0.1834, 0.1969], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 04:55:53,031 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-956000.pt
+2023-03-11 04:55:54,931 INFO [train.py:968] (0/2) Epoch 21, batch 43650, giga_loss[loss=0.3142, simple_loss=0.3784, pruned_loss=0.125, over 28880.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3757, pruned_loss=0.122, over 5660547.40 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3607, pruned_loss=0.1146, over 5662794.29 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3765, pruned_loss=0.1224, over 5666361.29 frames. ], batch size: 112, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:56:29,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.698e+03 2.214e+03 3.417e+03 9.836e+03, threshold=4.427e+03, percent-clipped=16.0
+2023-03-11 04:56:35,004 INFO [train.py:968] (0/2) Epoch 21, batch 43700, giga_loss[loss=0.3288, simple_loss=0.3882, pruned_loss=0.1347, over 28597.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3762, pruned_loss=0.1229, over 5671179.05 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3605, pruned_loss=0.1147, over 5670594.72 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3777, pruned_loss=0.1235, over 5668865.53 frames. ], batch size: 336, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:56:43,161 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:57:23,503 INFO [train.py:968] (0/2) Epoch 21, batch 43750, libri_loss[loss=0.312, simple_loss=0.3698, pruned_loss=0.1272, over 19429.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3762, pruned_loss=0.1237, over 5663525.32 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3602, pruned_loss=0.1146, over 5663902.77 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3778, pruned_loss=0.1244, over 5668578.13 frames. ], batch size: 187, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:57:50,779 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:57:52,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4320, 1.6846, 1.3284, 1.5232], device='cuda:0'), covar=tensor([0.0727, 0.0315, 0.0324, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 04:57:54,758 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2918, 2.2250, 1.6319, 1.4383], device='cuda:0'), covar=tensor([0.0843, 0.0266, 0.0294, 0.1060], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 04:58:04,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.285e+02 1.628e+03 2.069e+03 2.697e+03 5.842e+03, threshold=4.138e+03, percent-clipped=6.0
+2023-03-11 04:58:08,105 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-11 04:58:09,998 INFO [train.py:968] (0/2) Epoch 21, batch 43800, giga_loss[loss=0.3267, simple_loss=0.3858, pruned_loss=0.1338, over 28273.00 frames. ], tot_loss[loss=0.311, simple_loss=0.375, pruned_loss=0.1235, over 5661460.73 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3603, pruned_loss=0.1145, over 5666217.28 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3763, pruned_loss=0.1242, over 5663331.91 frames. ], batch size: 368, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:59:00,059 INFO [train.py:968] (0/2) Epoch 21, batch 43850, giga_loss[loss=0.3228, simple_loss=0.3798, pruned_loss=0.1329, over 29046.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3731, pruned_loss=0.1231, over 5664443.10 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3602, pruned_loss=0.1145, over 5667413.19 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3742, pruned_loss=0.1237, over 5664842.65 frames. ], batch size: 128, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:59:21,048 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:59:42,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.794e+03 2.282e+03 3.106e+03 8.710e+03, threshold=4.565e+03, percent-clipped=9.0
+2023-03-11 04:59:42,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6834, 2.3861, 1.6034, 0.8473], device='cuda:0'), covar=tensor([0.7053, 0.3381, 0.3219, 0.6558], device='cuda:0'), in_proj_covar=tensor([0.1743, 0.1655, 0.1596, 0.1423], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 04:59:47,981 INFO [train.py:968] (0/2) Epoch 21, batch 43900, giga_loss[loss=0.3038, simple_loss=0.3666, pruned_loss=0.1205, over 28966.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3734, pruned_loss=0.1243, over 5655120.33 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3611, pruned_loss=0.115, over 5657753.71 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3739, pruned_loss=0.1245, over 5664007.24 frames. ], batch size: 213, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:00:13,107 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:00:23,879 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:00:27,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4363, 1.5619, 1.4164, 1.3017], device='cuda:0'), covar=tensor([0.2258, 0.2164, 0.2001, 0.2304], device='cuda:0'), in_proj_covar=tensor([0.1975, 0.1904, 0.1839, 0.1975], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 05:00:34,744 INFO [train.py:968] (0/2) Epoch 21, batch 43950, giga_loss[loss=0.2548, simple_loss=0.3301, pruned_loss=0.08979, over 28375.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3728, pruned_loss=0.1243, over 5650360.85 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.361, pruned_loss=0.1149, over 5665537.54 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3736, pruned_loss=0.1249, over 5650293.57 frames. ], batch size: 60, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:01:22,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.942e+03 2.552e+03 3.943e+03 1.322e+04, threshold=5.103e+03, percent-clipped=19.0
+2023-03-11 05:01:23,816 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:01:23,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5858, 1.8214, 1.5405, 1.8003], device='cuda:0'), covar=tensor([0.2084, 0.2039, 0.2084, 0.1845], device='cuda:0'), in_proj_covar=tensor([0.1505, 0.1091, 0.1326, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 05:01:25,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5919, 1.7194, 1.5841, 1.4556], device='cuda:0'), covar=tensor([0.2802, 0.2383, 0.2195, 0.2595], device='cuda:0'), in_proj_covar=tensor([0.1978, 0.1907, 0.1840, 0.1975], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 05:01:26,327 INFO [train.py:968] (0/2) Epoch 21, batch 44000, giga_loss[loss=0.2828, simple_loss=0.3498, pruned_loss=0.1079, over 28734.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3721, pruned_loss=0.1243, over 5646326.82 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3611, pruned_loss=0.1149, over 5666453.85 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3727, pruned_loss=0.1248, over 5645486.13 frames. ], batch size: 262, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:01:51,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5013, 1.6324, 1.6932, 1.3140], device='cuda:0'), covar=tensor([0.1624, 0.2326, 0.1321, 0.1598], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0703, 0.0945, 0.0843], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 05:02:12,524 INFO [train.py:968] (0/2) Epoch 21, batch 44050, giga_loss[loss=0.276, simple_loss=0.3427, pruned_loss=0.1047, over 29014.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3711, pruned_loss=0.1238, over 5662260.27 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3614, pruned_loss=0.1151, over 5671006.01 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3715, pruned_loss=0.1242, over 5657363.18 frames. ], batch size: 128, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:02:33,363 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:02:40,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:02:53,027 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.697e+03 2.012e+03 2.669e+03 7.974e+03, threshold=4.023e+03, percent-clipped=2.0
+2023-03-11 05:02:55,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1276, 2.4681, 2.3043, 1.8604], device='cuda:0'), covar=tensor([0.2870, 0.2291, 0.2459, 0.2770], device='cuda:0'), in_proj_covar=tensor([0.1977, 0.1906, 0.1842, 0.1974], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 05:02:57,138 INFO [train.py:968] (0/2) Epoch 21, batch 44100, giga_loss[loss=0.2803, simple_loss=0.352, pruned_loss=0.1043, over 28936.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3698, pruned_loss=0.1229, over 5651837.38 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3617, pruned_loss=0.1152, over 5665729.39 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3701, pruned_loss=0.1233, over 5652891.68 frames. ], batch size: 112, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:03:44,190 INFO [train.py:968] (0/2) Epoch 21, batch 44150, giga_loss[loss=0.3568, simple_loss=0.3898, pruned_loss=0.1619, over 23528.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3705, pruned_loss=0.1224, over 5649418.22 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3619, pruned_loss=0.1151, over 5667262.57 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.371, pruned_loss=0.1232, over 5648584.86 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:03:51,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:04:28,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 1.722e+03 2.178e+03 2.794e+03 9.015e+03, threshold=4.357e+03, percent-clipped=11.0
+2023-03-11 05:04:31,892 INFO [train.py:968] (0/2) Epoch 21, batch 44200, giga_loss[loss=0.2949, simple_loss=0.3692, pruned_loss=0.1103, over 28758.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3735, pruned_loss=0.124, over 5646161.83 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3622, pruned_loss=0.1154, over 5658760.25 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3737, pruned_loss=0.1244, over 5652251.51 frames. ], batch size: 262, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:04:32,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3346, 1.6416, 1.5389, 1.1320], device='cuda:0'), covar=tensor([0.1678, 0.2691, 0.1500, 0.1773], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0704, 0.0945, 0.0844], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 05:04:55,957 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:04:57,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:05:16,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:05:19,024 INFO [train.py:968] (0/2) Epoch 21, batch 44250, libri_loss[loss=0.3032, simple_loss=0.3551, pruned_loss=0.1256, over 29520.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3727, pruned_loss=0.124, over 5648487.94 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.362, pruned_loss=0.1153, over 5655352.61 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3733, pruned_loss=0.1247, over 5656227.85 frames. ], batch size: 80, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:05:24,637 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:05:43,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2972, 1.2885, 4.0402, 3.3671], device='cuda:0'), covar=tensor([0.1816, 0.3008, 0.0493, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0651, 0.0973, 0.0919], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 05:06:01,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.050e+02 1.745e+03 2.347e+03 2.958e+03 8.388e+03, threshold=4.693e+03, percent-clipped=6.0
+2023-03-11 05:06:04,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:04,999 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:05,854 INFO [train.py:968] (0/2) Epoch 21, batch 44300, giga_loss[loss=0.2703, simple_loss=0.3507, pruned_loss=0.09489, over 28681.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3737, pruned_loss=0.1222, over 5651181.71 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3617, pruned_loss=0.1152, over 5647755.78 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3746, pruned_loss=0.123, over 5663842.01 frames. ], batch size: 242, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:06:07,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:16,133 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:26,250 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-11 05:06:32,828 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:34,263 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:48,873 INFO [train.py:968] (0/2) Epoch 21, batch 44350, giga_loss[loss=0.2941, simple_loss=0.3804, pruned_loss=0.1039, over 28941.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3752, pruned_loss=0.1208, over 5654338.75 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3619, pruned_loss=0.1152, over 5650140.79 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3759, pruned_loss=0.1214, over 5662450.69 frames. ], batch size: 145, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:06:52,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7178, 1.9033, 1.6304, 1.7184], device='cuda:0'), covar=tensor([0.2763, 0.2868, 0.3259, 0.2460], device='cuda:0'), in_proj_covar=tensor([0.1504, 0.1091, 0.1327, 0.0991], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 05:07:08,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:07:27,846 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:07:30,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:07:35,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.533e+02 1.463e+03 1.965e+03 2.661e+03 7.972e+03, threshold=3.930e+03, percent-clipped=4.0
+2023-03-11 05:07:38,226 INFO [train.py:968] (0/2) Epoch 21, batch 44400, giga_loss[loss=0.3782, simple_loss=0.4246, pruned_loss=0.1659, over 28566.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.377, pruned_loss=0.1218, over 5652048.08 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3614, pruned_loss=0.1148, over 5655126.11 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3784, pruned_loss=0.1229, over 5654437.31 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 05:07:55,901 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:15,738 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:18,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:20,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:23,155 INFO [train.py:968] (0/2) Epoch 21, batch 44450, giga_loss[loss=0.3142, simple_loss=0.382, pruned_loss=0.1232, over 28376.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3797, pruned_loss=0.125, over 5661430.98 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3612, pruned_loss=0.1148, over 5665069.95 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3817, pruned_loss=0.1261, over 5654488.91 frames. ], batch size: 77, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:08:26,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:28,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:43,228 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:49,199 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:56,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:09:06,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.884e+03 2.427e+03 3.067e+03 7.300e+03, threshold=4.853e+03, percent-clipped=13.0
+2023-03-11 05:09:08,623 INFO [train.py:968] (0/2) Epoch 21, batch 44500, giga_loss[loss=0.3804, simple_loss=0.4193, pruned_loss=0.1708, over 27529.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3806, pruned_loss=0.127, over 5661511.60 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3609, pruned_loss=0.1147, over 5670940.78 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3829, pruned_loss=0.1283, over 5650600.97 frames. ], batch size: 472, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:09:17,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0445, 1.9598, 1.9050, 1.6394], device='cuda:0'), covar=tensor([0.1745, 0.2510, 0.2147, 0.2461], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0750, 0.0710, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 05:09:20,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4231, 3.5695, 1.5447, 1.5944], device='cuda:0'), covar=tensor([0.0986, 0.0403, 0.0897, 0.1307], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0557, 0.0385, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 05:09:24,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:09:27,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:09:53,685 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:09:55,782 INFO [train.py:968] (0/2) Epoch 21, batch 44550, giga_loss[loss=0.2727, simple_loss=0.3439, pruned_loss=0.1007, over 28838.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3789, pruned_loss=0.1262, over 5677133.28 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3606, pruned_loss=0.1145, over 5675249.19 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3813, pruned_loss=0.1276, over 5664568.45 frames. ], batch size: 99, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:09:57,566 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 05:10:15,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3693, 1.4585, 1.4541, 1.2993], device='cuda:0'), covar=tensor([0.2157, 0.2008, 0.1782, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.1965, 0.1894, 0.1829, 0.1962], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 05:10:28,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:10:30,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:10:35,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.794e+03 2.186e+03 3.280e+03 1.591e+04, threshold=4.372e+03, percent-clipped=8.0
+2023-03-11 05:10:37,114 INFO [train.py:968] (0/2) Epoch 21, batch 44600, giga_loss[loss=0.2648, simple_loss=0.3489, pruned_loss=0.09036, over 28870.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3784, pruned_loss=0.126, over 5666219.37 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3607, pruned_loss=0.1147, over 5672348.03 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3809, pruned_loss=0.1273, over 5658065.65 frames. ], batch size: 174, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:10:48,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5660, 1.8401, 1.7095, 1.6078], device='cuda:0'), covar=tensor([0.1930, 0.1976, 0.2380, 0.2157], device='cuda:0'), in_proj_covar=tensor([0.0476, 0.0754, 0.0714, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 05:10:55,285 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:11:18,452 INFO [train.py:968] (0/2) Epoch 21, batch 44650, libri_loss[loss=0.3058, simple_loss=0.3663, pruned_loss=0.1227, over 19256.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3774, pruned_loss=0.1235, over 5655778.79 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3609, pruned_loss=0.115, over 5648352.84 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3796, pruned_loss=0.1246, over 5672368.09 frames. ], batch size: 187, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:12:04,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.647e+03 2.169e+03 3.007e+03 8.770e+03, threshold=4.337e+03, percent-clipped=4.0
+2023-03-11 05:12:05,878 INFO [train.py:968] (0/2) Epoch 21, batch 44700, giga_loss[loss=0.2803, simple_loss=0.3599, pruned_loss=0.1003, over 28903.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3777, pruned_loss=0.1225, over 5655373.62 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3612, pruned_loss=0.1153, over 5644504.81 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3795, pruned_loss=0.1232, over 5672255.90 frames. ], batch size: 145, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:12:06,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6128, 1.8496, 1.6685, 1.5797], device='cuda:0'), covar=tensor([0.1918, 0.2342, 0.2434, 0.2411], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0754, 0.0713, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 05:12:11,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:12:42,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0804, 1.1802, 3.3727, 3.0839], device='cuda:0'), covar=tensor([0.1746, 0.2799, 0.0567, 0.0901], device='cuda:0'), in_proj_covar=tensor([0.0767, 0.0652, 0.0976, 0.0921], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 05:12:52,569 INFO [train.py:968] (0/2) Epoch 21, batch 44750, giga_loss[loss=0.3758, simple_loss=0.409, pruned_loss=0.1714, over 26595.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3782, pruned_loss=0.1231, over 5652479.07 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3612, pruned_loss=0.1151, over 5651519.21 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3802, pruned_loss=0.1242, over 5659546.29 frames. ], batch size: 555, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:13:35,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.206e+03 1.847e+03 2.305e+03 3.001e+03 6.090e+03, threshold=4.610e+03, percent-clipped=7.0
+2023-03-11 05:13:38,207 INFO [train.py:968] (0/2) Epoch 21, batch 44800, giga_loss[loss=0.2786, simple_loss=0.3526, pruned_loss=0.1023, over 29087.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3781, pruned_loss=0.1238, over 5653311.66 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3614, pruned_loss=0.1154, over 5658696.69 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.38, pruned_loss=0.1246, over 5652529.83 frames. ], batch size: 155, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 05:13:53,707 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-11 05:14:23,570 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:14:23,966 INFO [train.py:968] (0/2) Epoch 21, batch 44850, giga_loss[loss=0.2849, simple_loss=0.3551, pruned_loss=0.1074, over 29074.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3765, pruned_loss=0.1231, over 5656160.17 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3618, pruned_loss=0.1156, over 5650021.00 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3778, pruned_loss=0.1237, over 5662522.89 frames. ], batch size: 155, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:14:24,215 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:14:26,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:14:27,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:14:51,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2182, 1.4750, 1.3660, 1.0706], device='cuda:0'), covar=tensor([0.2208, 0.2321, 0.1621, 0.2243], device='cuda:0'), in_proj_covar=tensor([0.1969, 0.1898, 0.1840, 0.1969], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 05:14:54,756 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:15:11,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.727e+03 2.164e+03 2.817e+03 7.247e+03, threshold=4.328e+03, percent-clipped=6.0
+2023-03-11 05:15:13,528 INFO [train.py:968] (0/2) Epoch 21, batch 44900, giga_loss[loss=0.2761, simple_loss=0.3417, pruned_loss=0.1053, over 28723.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3757, pruned_loss=0.1241, over 5645980.93 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3619, pruned_loss=0.1157, over 5643673.15 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3769, pruned_loss=0.1245, over 5657476.42 frames. ], batch size: 99, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:15:24,574 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:15:34,376 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:16:00,005 INFO [train.py:968] (0/2) Epoch 21, batch 44950, libri_loss[loss=0.2899, simple_loss=0.3647, pruned_loss=0.1076, over 29504.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3739, pruned_loss=0.1229, over 5653932.84 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3619, pruned_loss=0.1154, over 5651159.19 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3751, pruned_loss=0.1238, over 5656524.20 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:16:26,570 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:16:43,322 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3182, 1.6185, 1.3065, 0.8955], device='cuda:0'), covar=tensor([0.2486, 0.2542, 0.2843, 0.2326], device='cuda:0'), in_proj_covar=tensor([0.1506, 0.1092, 0.1328, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 05:16:45,077 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:16:46,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.181e+03 1.611e+03 2.220e+03 3.582e+03 7.332e+03, threshold=4.440e+03, percent-clipped=16.0
+2023-03-11 05:16:47,572 INFO [train.py:968] (0/2) Epoch 21, batch 45000, giga_loss[loss=0.3044, simple_loss=0.3706, pruned_loss=0.1191, over 29003.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3729, pruned_loss=0.1232, over 5654823.80 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3616, pruned_loss=0.1151, over 5654718.21 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3744, pruned_loss=0.1243, over 5653836.06 frames. ], batch size: 136, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:16:47,577 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 05:16:56,694 INFO [train.py:1012] (0/2) Epoch 21, validation: loss=0.2082, simple_loss=0.3173, pruned_loss=0.04959, over 944034.00 frames. 
+2023-03-11 05:16:56,695 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 05:16:57,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:17:13,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4041, 3.3573, 1.6463, 1.4398], device='cuda:0'), covar=tensor([0.0899, 0.0353, 0.0783, 0.1284], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0557, 0.0385, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 05:17:20,356 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.36 vs. limit=2.0
+2023-03-11 05:17:24,180 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:17:28,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6362, 1.1831, 4.2714, 3.5575], device='cuda:0'), covar=tensor([0.1850, 0.3131, 0.0946, 0.1119], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0650, 0.0971, 0.0917], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 05:17:39,788 INFO [train.py:968] (0/2) Epoch 21, batch 45050, giga_loss[loss=0.321, simple_loss=0.381, pruned_loss=0.1306, over 28716.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3733, pruned_loss=0.1247, over 5647613.83 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3618, pruned_loss=0.1154, over 5659014.48 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3745, pruned_loss=0.1255, over 5642481.84 frames. ], batch size: 284, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:17:55,352 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:18:21,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.963e+02 1.716e+03 2.338e+03 3.209e+03 9.277e+03, threshold=4.675e+03, percent-clipped=10.0
+2023-03-11 05:18:23,230 INFO [train.py:968] (0/2) Epoch 21, batch 45100, giga_loss[loss=0.3147, simple_loss=0.3716, pruned_loss=0.1289, over 26782.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3709, pruned_loss=0.1219, over 5649164.02 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3619, pruned_loss=0.1153, over 5661894.89 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3721, pruned_loss=0.1229, over 5642673.22 frames. ], batch size: 555, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:19:06,499 INFO [train.py:968] (0/2) Epoch 21, batch 45150, libri_loss[loss=0.3108, simple_loss=0.3771, pruned_loss=0.1222, over 25618.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3675, pruned_loss=0.1179, over 5659597.46 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3622, pruned_loss=0.1154, over 5666699.13 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3685, pruned_loss=0.1186, over 5649770.10 frames. ], batch size: 136, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:19:43,324 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-11 05:19:50,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2550, 3.2528, 1.4998, 1.3822], device='cuda:0'), covar=tensor([0.1043, 0.0407, 0.0902, 0.1448], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0556, 0.0385, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 05:19:54,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.779e+02 1.413e+03 1.722e+03 2.355e+03 8.456e+03, threshold=3.444e+03, percent-clipped=3.0
+2023-03-11 05:19:55,469 INFO [train.py:968] (0/2) Epoch 21, batch 45200, giga_loss[loss=0.2746, simple_loss=0.3469, pruned_loss=0.1012, over 28662.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3661, pruned_loss=0.1168, over 5643064.73 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3621, pruned_loss=0.1154, over 5659352.78 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.367, pruned_loss=0.1174, over 5641824.84 frames. ], batch size: 92, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:20:10,418 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-11 05:20:20,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:20:28,716 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4163, 1.6430, 1.6487, 1.2358], device='cuda:0'), covar=tensor([0.1813, 0.2598, 0.1471, 0.1740], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0707, 0.0948, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 05:20:41,722 INFO [train.py:968] (0/2) Epoch 21, batch 45250, giga_loss[loss=0.2834, simple_loss=0.3576, pruned_loss=0.1046, over 28840.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3659, pruned_loss=0.1173, over 5635829.55 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3623, pruned_loss=0.1155, over 5633962.20 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3665, pruned_loss=0.1177, over 5657703.11 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:21:03,617 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-11 05:21:18,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:21:26,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:21:28,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.915e+02 1.778e+03 2.552e+03 3.679e+03 1.500e+04, threshold=5.104e+03, percent-clipped=30.0
+2023-03-11 05:21:28,962 INFO [train.py:968] (0/2) Epoch 21, batch 45300, giga_loss[loss=0.3457, simple_loss=0.3698, pruned_loss=0.1608, over 23598.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3644, pruned_loss=0.1175, over 5599875.01 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3631, pruned_loss=0.1163, over 5581706.47 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3643, pruned_loss=0.1172, over 5663822.77 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:21:37,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4022, 1.6941, 1.3723, 1.2422], device='cuda:0'), covar=tensor([0.2561, 0.2560, 0.2944, 0.2321], device='cuda:0'), in_proj_covar=tensor([0.1510, 0.1095, 0.1332, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 05:22:11,749 INFO [train.py:968] (0/2) Epoch 21, batch 45350, giga_loss[loss=0.3203, simple_loss=0.3885, pruned_loss=0.1261, over 28685.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3667, pruned_loss=0.1189, over 5578912.07 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3641, pruned_loss=0.1172, over 5531143.20 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3657, pruned_loss=0.1178, over 5676531.92 frames. ], batch size: 262, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:22:14,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:22:27,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:22:28,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-11 05:22:30,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:22:38,688 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-11 05:22:41,450 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-21.pt
+2023-03-11 05:23:32,131 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:23:32,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.236e+02 1.546e+03 2.109e+03 3.131e+03 2.910e+04, threshold=4.217e+03, percent-clipped=13.0
+2023-03-11 05:24:01,357 INFO [train.py:968] (0/2) Epoch 22, batch 50, libri_loss[loss=0.2647, simple_loss=0.3508, pruned_loss=0.08929, over 29266.00 frames. ], tot_loss[loss=0.298, simple_loss=0.376, pruned_loss=0.11, over 1263957.07 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3359, pruned_loss=0.08229, over 145648.63 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3809, pruned_loss=0.1133, over 1147001.26 frames. ], batch size: 94, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:24:02,278 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:05,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:09,414 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:12,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:12,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:32,883 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:40,128 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:47,491 INFO [train.py:968] (0/2) Epoch 22, batch 100, giga_loss[loss=0.3216, simple_loss=0.3839, pruned_loss=0.1297, over 28710.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3653, pruned_loss=0.1046, over 2249103.26 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3421, pruned_loss=0.0882, over 259796.93 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3678, pruned_loss=0.1063, over 2082669.07 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:24:52,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:25:02,344 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:25:04,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:25:04,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.291e+02 1.243e+03 1.479e+03 2.114e+03 5.765e+03, threshold=2.959e+03, percent-clipped=1.0
+2023-03-11 05:25:30,664 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:25:31,801 INFO [train.py:968] (0/2) Epoch 22, batch 150, giga_loss[loss=0.2562, simple_loss=0.3321, pruned_loss=0.09013, over 27957.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3503, pruned_loss=0.0977, over 3005629.36 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3429, pruned_loss=0.09005, over 390017.82 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3516, pruned_loss=0.0987, over 2807697.29 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:26:14,247 INFO [train.py:968] (0/2) Epoch 22, batch 200, libri_loss[loss=0.254, simple_loss=0.3282, pruned_loss=0.08995, over 29314.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3364, pruned_loss=0.09131, over 3611887.46 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3413, pruned_loss=0.08946, over 497985.76 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3367, pruned_loss=0.09182, over 3409875.68 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:26:19,758 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:26:23,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:26:30,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.351e+02 1.093e+03 1.326e+03 1.803e+03 3.787e+03, threshold=2.652e+03, percent-clipped=4.0
+2023-03-11 05:26:40,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4620, 2.9158, 1.5242, 1.6247], device='cuda:0'), covar=tensor([0.0949, 0.0319, 0.0887, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0552, 0.0384, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 05:26:45,892 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:26:57,324 INFO [train.py:968] (0/2) Epoch 22, batch 250, giga_loss[loss=0.225, simple_loss=0.3046, pruned_loss=0.07266, over 28257.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.325, pruned_loss=0.086, over 4072644.88 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3447, pruned_loss=0.09291, over 575285.23 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.324, pruned_loss=0.08562, over 3888705.24 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:27:12,357 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-958000.pt
+2023-03-11 05:27:36,977 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=958029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:27:38,018 INFO [train.py:968] (0/2) Epoch 22, batch 300, giga_loss[loss=0.2016, simple_loss=0.2834, pruned_loss=0.0599, over 29085.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3175, pruned_loss=0.08228, over 4433206.40 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3435, pruned_loss=0.09281, over 824734.86 frames. ], giga_tot_loss[loss=0.2388, simple_loss=0.3151, pruned_loss=0.08128, over 4218121.06 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:27:49,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4495, 1.3142, 3.9998, 3.3340], device='cuda:0'), covar=tensor([0.1559, 0.2858, 0.0439, 0.1454], device='cuda:0'), in_proj_covar=tensor([0.0762, 0.0650, 0.0971, 0.0915], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 05:27:55,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.489e+02 1.080e+03 1.481e+03 1.984e+03 4.370e+03, threshold=2.962e+03, percent-clipped=9.0
+2023-03-11 05:28:15,741 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=958072.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 05:28:23,790 INFO [train.py:968] (0/2) Epoch 22, batch 350, giga_loss[loss=0.1949, simple_loss=0.2763, pruned_loss=0.05676, over 28381.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3104, pruned_loss=0.07938, over 4713512.84 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3424, pruned_loss=0.09256, over 945808.63 frames. ], giga_tot_loss[loss=0.2322, simple_loss=0.3078, pruned_loss=0.07826, over 4516377.53 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:28:59,109 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4168, 4.2459, 4.0352, 2.1369], device='cuda:0'), covar=tensor([0.0523, 0.0742, 0.0777, 0.1963], device='cuda:0'), in_proj_covar=tensor([0.1222, 0.1138, 0.0961, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 05:29:04,354 INFO [train.py:968] (0/2) Epoch 22, batch 400, libri_loss[loss=0.2832, simple_loss=0.3653, pruned_loss=0.1006, over 27714.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3067, pruned_loss=0.07783, over 4939610.59 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.341, pruned_loss=0.09131, over 1066645.79 frames. ], giga_tot_loss[loss=0.2287, simple_loss=0.3039, pruned_loss=0.07678, over 4760068.58 frames. ], batch size: 116, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:29:21,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.383e+02 1.061e+03 1.281e+03 1.808e+03 6.273e+03, threshold=2.562e+03, percent-clipped=5.0
+2023-03-11 05:29:45,626 INFO [train.py:968] (0/2) Epoch 22, batch 450, libri_loss[loss=0.2897, simple_loss=0.3681, pruned_loss=0.1057, over 29485.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3039, pruned_loss=0.07655, over 5110472.08 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3411, pruned_loss=0.09125, over 1138859.44 frames. ], giga_tot_loss[loss=0.226, simple_loss=0.3011, pruned_loss=0.07548, over 4956810.65 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:30:13,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=958212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:30:31,238 INFO [train.py:968] (0/2) Epoch 22, batch 500, libri_loss[loss=0.2331, simple_loss=0.3141, pruned_loss=0.07607, over 28644.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.301, pruned_loss=0.07511, over 5243293.76 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3409, pruned_loss=0.09072, over 1185654.02 frames. ], giga_tot_loss[loss=0.2234, simple_loss=0.2984, pruned_loss=0.07422, over 5116812.86 frames. ], batch size: 63, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:30:47,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.920e+02 1.160e+03 1.449e+03 1.834e+03 4.330e+03, threshold=2.898e+03, percent-clipped=8.0
+2023-03-11 05:30:51,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3983, 1.9070, 1.4432, 0.7779], device='cuda:0'), covar=tensor([0.6698, 0.3439, 0.4003, 0.6926], device='cuda:0'), in_proj_covar=tensor([0.1743, 0.1648, 0.1591, 0.1418], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 05:31:15,144 INFO [train.py:968] (0/2) Epoch 22, batch 550, libri_loss[loss=0.269, simple_loss=0.3522, pruned_loss=0.09287, over 29663.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2996, pruned_loss=0.07449, over 5351299.09 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3421, pruned_loss=0.09058, over 1279705.16 frames. ], giga_tot_loss[loss=0.2218, simple_loss=0.2965, pruned_loss=0.07349, over 5239698.66 frames. ], batch size: 88, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:31:58,464 INFO [train.py:968] (0/2) Epoch 22, batch 600, libri_loss[loss=0.2194, simple_loss=0.3059, pruned_loss=0.06648, over 29569.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.2991, pruned_loss=0.07455, over 5419009.24 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3422, pruned_loss=0.09052, over 1427977.55 frames. ], giga_tot_loss[loss=0.2208, simple_loss=0.2951, pruned_loss=0.07324, over 5322488.41 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:32:15,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.834e+02 1.021e+03 1.272e+03 1.667e+03 3.853e+03, threshold=2.543e+03, percent-clipped=5.0
+2023-03-11 05:32:19,884 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=958355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:32:23,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6467, 1.9388, 1.8441, 1.6766], device='cuda:0'), covar=tensor([0.2148, 0.2234, 0.2531, 0.2415], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0748, 0.0709, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 05:32:24,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=958358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:32:42,755 INFO [train.py:968] (0/2) Epoch 22, batch 650, giga_loss[loss=0.2029, simple_loss=0.2857, pruned_loss=0.06006, over 29056.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2976, pruned_loss=0.07388, over 5477641.25 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3413, pruned_loss=0.0898, over 1559468.50 frames. ], giga_tot_loss[loss=0.2192, simple_loss=0.2934, pruned_loss=0.07253, over 5389004.26 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:32:49,828 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=958387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:33:02,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=958404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:33:28,654 INFO [train.py:968] (0/2) Epoch 22, batch 700, giga_loss[loss=0.1815, simple_loss=0.2586, pruned_loss=0.05219, over 28306.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2938, pruned_loss=0.07204, over 5525193.40 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3409, pruned_loss=0.08961, over 1622825.20 frames. ], giga_tot_loss[loss=0.2158, simple_loss=0.2899, pruned_loss=0.07079, over 5451194.42 frames. ], batch size: 65, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:33:32,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6954, 1.1180, 4.7840, 3.5933], device='cuda:0'), covar=tensor([0.1634, 0.3111, 0.0402, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0759, 0.0648, 0.0969, 0.0912], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 05:33:39,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4501, 1.7895, 1.7210, 1.5395], device='cuda:0'), covar=tensor([0.1704, 0.1568, 0.1959, 0.1667], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0747, 0.0709, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 05:33:41,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=958447.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 05:33:44,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.371e+02 1.002e+03 1.267e+03 1.710e+03 5.863e+03, threshold=2.533e+03, percent-clipped=8.0
+2023-03-11 05:34:12,964 INFO [train.py:968] (0/2) Epoch 22, batch 750, libri_loss[loss=0.2196, simple_loss=0.3113, pruned_loss=0.06391, over 29568.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2919, pruned_loss=0.07096, over 5575269.66 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3408, pruned_loss=0.08968, over 1770278.01 frames. ], giga_tot_loss[loss=0.2129, simple_loss=0.2872, pruned_loss=0.06932, over 5504649.17 frames. ], batch size: 75, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:34:55,172 INFO [train.py:968] (0/2) Epoch 22, batch 800, giga_loss[loss=0.2476, simple_loss=0.3089, pruned_loss=0.09313, over 28744.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2907, pruned_loss=0.07033, over 5602078.96 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3397, pruned_loss=0.08894, over 1942305.13 frames. ], giga_tot_loss[loss=0.211, simple_loss=0.2852, pruned_loss=0.06841, over 5541129.65 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:35:11,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=958547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:35:14,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=958550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:35:15,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.798e+02 1.098e+03 1.295e+03 1.775e+03 4.271e+03, threshold=2.589e+03, percent-clipped=8.0
+2023-03-11 05:35:42,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=958579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:35:43,973 INFO [train.py:968] (0/2) Epoch 22, batch 850, giga_loss[loss=0.2837, simple_loss=0.3585, pruned_loss=0.1044, over 28827.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2987, pruned_loss=0.07522, over 5611477.07 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3404, pruned_loss=0.08938, over 1960432.61 frames. ], giga_tot_loss[loss=0.2205, simple_loss=0.294, pruned_loss=0.07356, over 5563941.43 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:35:46,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3346, 1.5090, 1.6036, 1.4472], device='cuda:0'), covar=tensor([0.1531, 0.1365, 0.1549, 0.1393], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0746, 0.0709, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 05:35:55,172 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=958590.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 05:35:57,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=958593.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 05:36:25,004 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=958622.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 05:36:31,054 INFO [train.py:968] (0/2) Epoch 22, batch 900, giga_loss[loss=0.2817, simple_loss=0.3533, pruned_loss=0.1051, over 28545.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3116, pruned_loss=0.08164, over 5630165.23 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3396, pruned_loss=0.08893, over 2112000.38 frames. ], giga_tot_loss[loss=0.2335, simple_loss=0.3069, pruned_loss=0.08008, over 5584593.54 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:36:49,639 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.738e+02 1.331e+03 1.696e+03 2.292e+03 5.333e+03, threshold=3.392e+03, percent-clipped=16.0
+2023-03-11 05:36:57,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5242, 1.7558, 1.4329, 1.6937], device='cuda:0'), covar=tensor([0.2530, 0.2625, 0.2943, 0.2120], device='cuda:0'), in_proj_covar=tensor([0.1512, 0.1091, 0.1334, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 05:37:02,905 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-11 05:37:12,031 INFO [train.py:968] (0/2) Epoch 22, batch 950, giga_loss[loss=0.2781, simple_loss=0.3561, pruned_loss=0.1, over 28789.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3236, pruned_loss=0.08748, over 5651703.31 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3395, pruned_loss=0.08888, over 2220940.48 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3193, pruned_loss=0.08619, over 5611058.20 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:37:12,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0062, 2.2091, 1.5535, 1.7639], device='cuda:0'), covar=tensor([0.0997, 0.0706, 0.1108, 0.1188], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0449, 0.0520, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 05:37:53,936 INFO [train.py:968] (0/2) Epoch 22, batch 1000, giga_loss[loss=0.2545, simple_loss=0.3345, pruned_loss=0.0873, over 28829.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3323, pruned_loss=0.09144, over 5660504.43 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3404, pruned_loss=0.08955, over 2253551.64 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3286, pruned_loss=0.09023, over 5629764.07 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:38:02,049 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.78 vs. limit=5.0
+2023-03-11 05:38:09,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.573e+02 1.267e+03 1.754e+03 2.375e+03 6.121e+03, threshold=3.507e+03, percent-clipped=5.0
+2023-03-11 05:38:28,736 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 05:38:33,842 INFO [train.py:968] (0/2) Epoch 22, batch 1050, giga_loss[loss=0.2358, simple_loss=0.3127, pruned_loss=0.07941, over 23772.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3352, pruned_loss=0.09109, over 5673203.53 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3411, pruned_loss=0.08964, over 2361170.45 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3319, pruned_loss=0.09014, over 5642956.90 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:38:45,797 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=958793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:39:05,090 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-11 05:39:22,538 INFO [train.py:968] (0/2) Epoch 22, batch 1100, giga_loss[loss=0.2671, simple_loss=0.3497, pruned_loss=0.09225, over 29039.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3378, pruned_loss=0.09175, over 5672138.47 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3409, pruned_loss=0.08944, over 2396582.11 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3352, pruned_loss=0.0911, over 5646370.10 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:39:39,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.051e+02 1.168e+03 1.433e+03 1.879e+03 7.510e+03, threshold=2.866e+03, percent-clipped=3.0
+2023-03-11 05:40:03,639 INFO [train.py:968] (0/2) Epoch 22, batch 1150, giga_loss[loss=0.3067, simple_loss=0.3821, pruned_loss=0.1157, over 28686.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3404, pruned_loss=0.09349, over 5689807.10 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.341, pruned_loss=0.08943, over 2465452.95 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3382, pruned_loss=0.09302, over 5666183.62 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:40:14,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9571, 1.1475, 1.2537, 1.0960], device='cuda:0'), covar=tensor([0.1590, 0.1154, 0.1827, 0.1185], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0752, 0.0715, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 05:40:26,024 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 05:40:30,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2984, 1.1224, 3.7975, 3.1831], device='cuda:0'), covar=tensor([0.1723, 0.2973, 0.0458, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0752, 0.0642, 0.0959, 0.0905], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 05:40:32,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9960, 1.0614, 3.3421, 2.8991], device='cuda:0'), covar=tensor([0.1844, 0.2968, 0.0518, 0.1119], device='cuda:0'), in_proj_covar=tensor([0.0752, 0.0642, 0.0959, 0.0905], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 05:40:34,674 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5514, 1.7961, 1.4475, 1.8219], device='cuda:0'), covar=tensor([0.2642, 0.2626, 0.2908, 0.2213], device='cuda:0'), in_proj_covar=tensor([0.1514, 0.1096, 0.1336, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 05:40:47,407 INFO [train.py:968] (0/2) Epoch 22, batch 1200, giga_loss[loss=0.3114, simple_loss=0.3755, pruned_loss=0.1236, over 28617.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3426, pruned_loss=0.09526, over 5672662.20 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.341, pruned_loss=0.08939, over 2574020.03 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3408, pruned_loss=0.09505, over 5655791.79 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:41:07,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.409e+02 1.198e+03 1.420e+03 1.706e+03 4.089e+03, threshold=2.841e+03, percent-clipped=2.0
+2023-03-11 05:41:29,240 INFO [train.py:968] (0/2) Epoch 22, batch 1250, giga_loss[loss=0.2609, simple_loss=0.3452, pruned_loss=0.08827, over 29065.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3454, pruned_loss=0.09687, over 5677783.30 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3417, pruned_loss=0.08987, over 2655442.20 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3437, pruned_loss=0.09667, over 5660228.96 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:41:40,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-11 05:42:12,641 INFO [train.py:968] (0/2) Epoch 22, batch 1300, libri_loss[loss=0.2683, simple_loss=0.3509, pruned_loss=0.09291, over 29260.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.348, pruned_loss=0.0977, over 5676172.76 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3419, pruned_loss=0.08994, over 2756206.68 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3468, pruned_loss=0.09776, over 5664631.57 frames. ], batch size: 97, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:42:27,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.403e+02 1.306e+03 1.583e+03 2.021e+03 5.542e+03, threshold=3.165e+03, percent-clipped=9.0
+2023-03-11 05:42:45,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5717, 4.3671, 4.1586, 1.9871], device='cuda:0'), covar=tensor([0.0542, 0.0736, 0.0751, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.1204, 0.1121, 0.0949, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 05:42:50,690 INFO [train.py:968] (0/2) Epoch 22, batch 1350, giga_loss[loss=0.2381, simple_loss=0.3315, pruned_loss=0.07236, over 28717.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3498, pruned_loss=0.09737, over 5695038.96 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.342, pruned_loss=0.08982, over 2803231.50 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3489, pruned_loss=0.09757, over 5683007.44 frames. ], batch size: 60, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:43:28,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8299, 2.0372, 1.6482, 1.9658], device='cuda:0'), covar=tensor([0.0757, 0.0280, 0.0325, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 05:43:36,296 INFO [train.py:968] (0/2) Epoch 22, batch 1400, giga_loss[loss=0.3003, simple_loss=0.3696, pruned_loss=0.1155, over 28856.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.351, pruned_loss=0.09762, over 5691250.15 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3422, pruned_loss=0.08983, over 2833867.42 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3503, pruned_loss=0.09786, over 5680063.83 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:43:49,008 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-11 05:43:51,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.099e+02 1.215e+03 1.412e+03 1.933e+03 4.045e+03, threshold=2.825e+03, percent-clipped=2.0
+2023-03-11 05:44:05,301 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=959168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:44:16,462 INFO [train.py:968] (0/2) Epoch 22, batch 1450, giga_loss[loss=0.2392, simple_loss=0.3314, pruned_loss=0.07354, over 28816.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3503, pruned_loss=0.09616, over 5696702.89 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3411, pruned_loss=0.08915, over 2950483.90 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3506, pruned_loss=0.09694, over 5683740.29 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:44:21,909 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-11 05:44:41,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3820, 2.0545, 1.5369, 0.6337], device='cuda:0'), covar=tensor([0.6698, 0.3438, 0.4721, 0.6995], device='cuda:0'), in_proj_covar=tensor([0.1742, 0.1637, 0.1592, 0.1415], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 05:44:55,493 INFO [train.py:968] (0/2) Epoch 22, batch 1500, giga_loss[loss=0.2514, simple_loss=0.3398, pruned_loss=0.08151, over 28873.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3493, pruned_loss=0.09457, over 5695742.35 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3414, pruned_loss=0.08929, over 2969886.85 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3494, pruned_loss=0.09517, over 5691858.01 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:45:11,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.561e+02 1.249e+03 1.529e+03 2.045e+03 4.734e+03, threshold=3.058e+03, percent-clipped=7.0
+2023-03-11 05:45:23,965 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=959267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:45:36,685 INFO [train.py:968] (0/2) Epoch 22, batch 1550, giga_loss[loss=0.2442, simple_loss=0.3259, pruned_loss=0.08128, over 28824.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3477, pruned_loss=0.09324, over 5699189.93 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3416, pruned_loss=0.08941, over 3040769.38 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3479, pruned_loss=0.09377, over 5692967.24 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:45:45,976 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 05:45:53,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2039, 1.4092, 1.2765, 1.0754], device='cuda:0'), covar=tensor([0.2929, 0.2682, 0.1924, 0.2509], device='cuda:0'), in_proj_covar=tensor([0.1956, 0.1876, 0.1817, 0.1952], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 05:46:02,766 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=959311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:46:05,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=959314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:46:17,362 INFO [train.py:968] (0/2) Epoch 22, batch 1600, libri_loss[loss=0.2756, simple_loss=0.3583, pruned_loss=0.0964, over 28650.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3482, pruned_loss=0.0948, over 5711105.53 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3421, pruned_loss=0.08944, over 3149803.26 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3484, pruned_loss=0.0954, over 5701848.87 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:46:25,311 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=959343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:46:33,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.779e+02 1.203e+03 1.464e+03 1.734e+03 3.005e+03, threshold=2.929e+03, percent-clipped=0.0
+2023-03-11 05:47:00,167 INFO [train.py:968] (0/2) Epoch 22, batch 1650, giga_loss[loss=0.3512, simple_loss=0.3986, pruned_loss=0.1519, over 28722.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3508, pruned_loss=0.09941, over 5711072.56 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3415, pruned_loss=0.08908, over 3218075.87 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3515, pruned_loss=0.1002, over 5699407.74 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:47:41,563 INFO [train.py:968] (0/2) Epoch 22, batch 1700, giga_loss[loss=0.2871, simple_loss=0.3567, pruned_loss=0.1088, over 28939.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3521, pruned_loss=0.1016, over 5707949.57 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3419, pruned_loss=0.08921, over 3310956.93 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3527, pruned_loss=0.1026, over 5692072.28 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:48:00,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.006e+02 1.290e+03 1.600e+03 2.075e+03 4.597e+03, threshold=3.200e+03, percent-clipped=7.0
+2023-03-11 05:48:26,236 INFO [train.py:968] (0/2) Epoch 22, batch 1750, giga_loss[loss=0.252, simple_loss=0.3288, pruned_loss=0.0876, over 28738.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3497, pruned_loss=0.101, over 5707749.48 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3411, pruned_loss=0.08873, over 3349574.00 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3507, pruned_loss=0.1022, over 5692420.23 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:49:04,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8102, 1.9774, 1.3112, 1.5555], device='cuda:0'), covar=tensor([0.0978, 0.0644, 0.1128, 0.1171], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0450, 0.0522, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 05:49:07,373 INFO [train.py:968] (0/2) Epoch 22, batch 1800, giga_loss[loss=0.2448, simple_loss=0.3232, pruned_loss=0.08322, over 29070.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3473, pruned_loss=0.09972, over 5721007.25 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3397, pruned_loss=0.08777, over 3450411.42 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3491, pruned_loss=0.1016, over 5701744.75 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:49:27,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.880e+02 1.334e+03 1.678e+03 2.325e+03 6.352e+03, threshold=3.355e+03, percent-clipped=9.0
+2023-03-11 05:49:29,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=959556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:49:37,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2087, 0.7848, 0.8523, 1.3730], device='cuda:0'), covar=tensor([0.0829, 0.0398, 0.0377, 0.0903], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 05:49:49,582 INFO [train.py:968] (0/2) Epoch 22, batch 1850, giga_loss[loss=0.2507, simple_loss=0.3297, pruned_loss=0.08589, over 28986.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3471, pruned_loss=0.09899, over 5720002.49 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3398, pruned_loss=0.08774, over 3474989.12 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3485, pruned_loss=0.1006, over 5703092.40 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:50:35,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5470, 2.3136, 1.6142, 0.6990], device='cuda:0'), covar=tensor([0.7601, 0.3201, 0.4879, 0.7554], device='cuda:0'), in_proj_covar=tensor([0.1747, 0.1643, 0.1596, 0.1418], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 05:50:36,159 INFO [train.py:968] (0/2) Epoch 22, batch 1900, giga_loss[loss=0.2583, simple_loss=0.3366, pruned_loss=0.08993, over 28743.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3452, pruned_loss=0.09724, over 5711990.46 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3406, pruned_loss=0.08818, over 3519801.04 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.346, pruned_loss=0.09846, over 5698000.69 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:50:40,337 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-11 05:50:46,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=959642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:50:58,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.793e+02 1.112e+03 1.403e+03 1.995e+03 4.931e+03, threshold=2.807e+03, percent-clipped=7.0
+2023-03-11 05:51:21,144 INFO [train.py:968] (0/2) Epoch 22, batch 1950, giga_loss[loss=0.209, simple_loss=0.2962, pruned_loss=0.06089, over 29002.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3421, pruned_loss=0.0951, over 5708630.86 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3408, pruned_loss=0.08813, over 3590543.59 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3427, pruned_loss=0.09633, over 5691992.74 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:52:07,048 INFO [train.py:968] (0/2) Epoch 22, batch 2000, giga_loss[loss=0.2191, simple_loss=0.3009, pruned_loss=0.06863, over 28969.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3361, pruned_loss=0.09194, over 5697307.49 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3405, pruned_loss=0.08804, over 3657058.71 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3368, pruned_loss=0.09314, over 5680784.05 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:52:27,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.992e+02 1.031e+03 1.397e+03 1.998e+03 5.378e+03, threshold=2.794e+03, percent-clipped=12.0
+2023-03-11 05:52:51,935 INFO [train.py:968] (0/2) Epoch 22, batch 2050, giga_loss[loss=0.2505, simple_loss=0.3182, pruned_loss=0.09143, over 28987.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3308, pruned_loss=0.08966, over 5690910.25 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3397, pruned_loss=0.08759, over 3701582.01 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3317, pruned_loss=0.09093, over 5673978.21 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:52:55,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4016, 2.9426, 1.6219, 1.4685], device='cuda:0'), covar=tensor([0.0931, 0.0298, 0.0853, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0547, 0.0382, 0.0424], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-11 05:52:56,168 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=959785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:52:58,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=959788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:53:26,942 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=959817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:53:39,733 INFO [train.py:968] (0/2) Epoch 22, batch 2100, giga_loss[loss=0.2321, simple_loss=0.3129, pruned_loss=0.07564, over 28649.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3291, pruned_loss=0.08857, over 5693977.74 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3401, pruned_loss=0.08779, over 3734566.24 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3294, pruned_loss=0.08947, over 5677697.48 frames. ], batch size: 60, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:53:56,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.362e+02 1.200e+03 1.463e+03 2.043e+03 4.567e+03, threshold=2.927e+03, percent-clipped=8.0
+2023-03-11 05:54:18,693 INFO [train.py:968] (0/2) Epoch 22, batch 2150, giga_loss[loss=0.2285, simple_loss=0.3091, pruned_loss=0.07391, over 28422.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3299, pruned_loss=0.08855, over 5691705.28 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3405, pruned_loss=0.08816, over 3767613.26 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3297, pruned_loss=0.08906, over 5685025.10 frames. ], batch size: 60, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:55:00,857 INFO [train.py:968] (0/2) Epoch 22, batch 2200, giga_loss[loss=0.265, simple_loss=0.3292, pruned_loss=0.1004, over 23845.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.33, pruned_loss=0.08857, over 5694439.51 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3413, pruned_loss=0.08854, over 3818817.51 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3291, pruned_loss=0.08875, over 5686114.04 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:55:01,095 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=959931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:55:20,978 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.029e+02 1.108e+03 1.365e+03 1.855e+03 3.937e+03, threshold=2.729e+03, percent-clipped=4.0
+2023-03-11 05:55:41,048 INFO [train.py:968] (0/2) Epoch 22, batch 2250, giga_loss[loss=0.2253, simple_loss=0.2962, pruned_loss=0.07722, over 28347.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3284, pruned_loss=0.08804, over 5705351.72 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3412, pruned_loss=0.08832, over 3870259.75 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3275, pruned_loss=0.08832, over 5694254.70 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:55:56,747 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-960000.pt
+2023-03-11 05:56:06,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5496, 1.6007, 1.7877, 1.3496], device='cuda:0'), covar=tensor([0.1798, 0.2575, 0.1438, 0.1790], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0703, 0.0952, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 05:56:20,514 INFO [train.py:968] (0/2) Epoch 22, batch 2300, giga_loss[loss=0.2165, simple_loss=0.3004, pruned_loss=0.06629, over 29016.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3279, pruned_loss=0.08783, over 5709340.86 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08849, over 3958490.53 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3261, pruned_loss=0.08789, over 5693510.34 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:56:39,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.144e+02 1.067e+03 1.245e+03 1.642e+03 3.989e+03, threshold=2.491e+03, percent-clipped=3.0
+2023-03-11 05:56:54,770 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=960072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:56:54,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2989, 0.7874, 0.8983, 1.3932], device='cuda:0'), covar=tensor([0.0795, 0.0398, 0.0377, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 05:56:56,178 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=960074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:56:58,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=960077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:56:59,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7534, 4.5573, 4.3322, 1.9329], device='cuda:0'), covar=tensor([0.0518, 0.0711, 0.0690, 0.2195], device='cuda:0'), in_proj_covar=tensor([0.1205, 0.1120, 0.0949, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 05:57:01,163 INFO [train.py:968] (0/2) Epoch 22, batch 2350, giga_loss[loss=0.2477, simple_loss=0.3241, pruned_loss=0.08559, over 28308.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3254, pruned_loss=0.08639, over 5722100.27 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3424, pruned_loss=0.08855, over 3997070.32 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3235, pruned_loss=0.08639, over 5706330.71 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:57:10,578 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-11 05:57:22,773 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=960106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:57:36,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2708, 1.2236, 3.7643, 3.1463], device='cuda:0'), covar=tensor([0.1697, 0.2778, 0.0452, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0752, 0.0642, 0.0954, 0.0901], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 05:57:36,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4329, 1.5114, 1.1672, 1.1230], device='cuda:0'), covar=tensor([0.0999, 0.0546, 0.1104, 0.1157], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0447, 0.0518, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 05:57:43,872 INFO [train.py:968] (0/2) Epoch 22, batch 2400, giga_loss[loss=0.2364, simple_loss=0.3102, pruned_loss=0.08132, over 29015.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3238, pruned_loss=0.0856, over 5729632.92 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3423, pruned_loss=0.08827, over 4053628.53 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3217, pruned_loss=0.08572, over 5711985.45 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:58:01,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.374e+02 9.738e+02 1.177e+03 1.456e+03 5.188e+03, threshold=2.354e+03, percent-clipped=8.0
+2023-03-11 05:58:02,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=960155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:58:02,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7181, 2.3191, 2.0189, 1.8205], device='cuda:0'), covar=tensor([0.0741, 0.0257, 0.0282, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 05:58:21,953 INFO [train.py:968] (0/2) Epoch 22, batch 2450, giga_loss[loss=0.2338, simple_loss=0.3122, pruned_loss=0.07772, over 28995.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3228, pruned_loss=0.08537, over 5725833.37 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3435, pruned_loss=0.0889, over 4098061.30 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3199, pruned_loss=0.08498, over 5717237.27 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:58:47,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8777, 2.0827, 2.0385, 1.5442], device='cuda:0'), covar=tensor([0.3623, 0.2731, 0.2348, 0.3385], device='cuda:0'), in_proj_covar=tensor([0.1949, 0.1873, 0.1812, 0.1948], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 05:59:00,121 INFO [train.py:968] (0/2) Epoch 22, batch 2500, giga_loss[loss=0.2205, simple_loss=0.2882, pruned_loss=0.07641, over 28528.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3198, pruned_loss=0.084, over 5724985.32 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3434, pruned_loss=0.08876, over 4121653.69 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3173, pruned_loss=0.08372, over 5718827.41 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:59:19,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.927e+02 1.103e+03 1.345e+03 1.830e+03 7.326e+03, threshold=2.690e+03, percent-clipped=15.0
+2023-03-11 05:59:40,466 INFO [train.py:968] (0/2) Epoch 22, batch 2550, giga_loss[loss=0.2185, simple_loss=0.2948, pruned_loss=0.07105, over 28884.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3176, pruned_loss=0.08326, over 5720544.30 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3437, pruned_loss=0.08903, over 4138196.44 frames. ], giga_tot_loss[loss=0.2404, simple_loss=0.3152, pruned_loss=0.08282, over 5715161.76 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:59:44,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4372, 1.7079, 1.6640, 1.2425], device='cuda:0'), covar=tensor([0.1689, 0.2836, 0.1564, 0.1903], device='cuda:0'), in_proj_covar=tensor([0.0908, 0.0706, 0.0954, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 05:59:57,318 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4978, 1.6292, 1.5056, 1.3449], device='cuda:0'), covar=tensor([0.3110, 0.2599, 0.2078, 0.2916], device='cuda:0'), in_proj_covar=tensor([0.1953, 0.1876, 0.1815, 0.1951], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 06:00:18,412 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=960329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:00:19,491 INFO [train.py:968] (0/2) Epoch 22, batch 2600, giga_loss[loss=0.244, simple_loss=0.3229, pruned_loss=0.08255, over 28991.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3181, pruned_loss=0.08326, over 5723203.61 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3451, pruned_loss=0.08946, over 4189085.00 frames. ], giga_tot_loss[loss=0.2398, simple_loss=0.3145, pruned_loss=0.08249, over 5715392.05 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:00:38,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.725e+02 1.054e+03 1.440e+03 1.934e+03 4.790e+03, threshold=2.879e+03, percent-clipped=12.0
+2023-03-11 06:00:44,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3058, 1.5858, 1.4730, 1.2191], device='cuda:0'), covar=tensor([0.2259, 0.2087, 0.1459, 0.1938], device='cuda:0'), in_proj_covar=tensor([0.1953, 0.1875, 0.1813, 0.1948], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 06:00:57,688 INFO [train.py:968] (0/2) Epoch 22, batch 2650, giga_loss[loss=0.2139, simple_loss=0.2951, pruned_loss=0.06631, over 28909.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3173, pruned_loss=0.08253, over 5727107.40 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3458, pruned_loss=0.0896, over 4236370.69 frames. ], giga_tot_loss[loss=0.2382, simple_loss=0.3132, pruned_loss=0.08163, over 5719014.31 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:01:02,423 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=960388.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:01:03,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-11 06:01:18,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5465, 1.8340, 1.4824, 1.6375], device='cuda:0'), covar=tensor([0.2637, 0.2769, 0.3156, 0.2436], device='cuda:0'), in_proj_covar=tensor([0.1510, 0.1094, 0.1334, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 06:01:40,459 INFO [train.py:968] (0/2) Epoch 22, batch 2700, giga_loss[loss=0.2374, simple_loss=0.3143, pruned_loss=0.08024, over 29033.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3195, pruned_loss=0.08432, over 5717306.66 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3454, pruned_loss=0.08922, over 4275850.18 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3159, pruned_loss=0.08371, over 5708416.47 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:01:54,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=960447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:02:00,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.561e+02 1.178e+03 1.534e+03 2.071e+03 5.534e+03, threshold=3.068e+03, percent-clipped=8.0
+2023-03-11 06:02:22,431 INFO [train.py:968] (0/2) Epoch 22, batch 2750, giga_loss[loss=0.2638, simple_loss=0.3425, pruned_loss=0.09258, over 28845.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3239, pruned_loss=0.08688, over 5716688.10 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3455, pruned_loss=0.08913, over 4304585.51 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3206, pruned_loss=0.0864, over 5709439.34 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:02:24,856 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 06:02:29,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2812, 2.2505, 1.3851, 1.4278], device='cuda:0'), covar=tensor([0.0843, 0.0386, 0.0729, 0.1191], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0548, 0.0384, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 06:03:00,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8794, 3.7193, 3.5124, 1.5529], device='cuda:0'), covar=tensor([0.0671, 0.0778, 0.0731, 0.2216], device='cuda:0'), in_proj_covar=tensor([0.1204, 0.1119, 0.0948, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 06:03:07,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=960530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:03:08,262 INFO [train.py:968] (0/2) Epoch 22, batch 2800, giga_loss[loss=0.265, simple_loss=0.3403, pruned_loss=0.09485, over 29074.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3328, pruned_loss=0.09305, over 5708355.29 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3463, pruned_loss=0.0896, over 4344552.80 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3292, pruned_loss=0.09235, over 5697946.92 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:03:31,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.188e+02 1.260e+03 1.690e+03 2.395e+03 5.677e+03, threshold=3.381e+03, percent-clipped=11.0
+2023-03-11 06:03:52,986 INFO [train.py:968] (0/2) Epoch 22, batch 2850, giga_loss[loss=0.2923, simple_loss=0.3662, pruned_loss=0.1092, over 27817.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3395, pruned_loss=0.09714, over 5698816.66 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3458, pruned_loss=0.08937, over 4374875.62 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3368, pruned_loss=0.09688, over 5687528.33 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:04:02,349 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=960590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:04:05,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=960593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:04:32,775 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=960622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:04:43,782 INFO [train.py:968] (0/2) Epoch 22, batch 2900, giga_loss[loss=0.2896, simple_loss=0.3716, pruned_loss=0.1038, over 28962.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3453, pruned_loss=0.1001, over 5672572.17 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3462, pruned_loss=0.08958, over 4408387.49 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3428, pruned_loss=0.09996, over 5662894.89 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:04:59,322 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-11 06:05:05,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.153e+02 1.271e+03 1.579e+03 2.292e+03 8.235e+03, threshold=3.159e+03, percent-clipped=11.0
+2023-03-11 06:05:17,649 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=960673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:05:19,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=960676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:05:22,337 INFO [train.py:968] (0/2) Epoch 22, batch 2950, giga_loss[loss=0.3619, simple_loss=0.417, pruned_loss=0.1534, over 29042.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.349, pruned_loss=0.1008, over 5682929.48 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3459, pruned_loss=0.08952, over 4460349.78 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3472, pruned_loss=0.1011, over 5672938.82 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:05:44,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=960704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:05:45,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=960705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:06:07,628 INFO [train.py:968] (0/2) Epoch 22, batch 3000, giga_loss[loss=0.3648, simple_loss=0.398, pruned_loss=0.1658, over 23470.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3541, pruned_loss=0.1036, over 5687646.61 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3459, pruned_loss=0.08961, over 4506834.06 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3528, pruned_loss=0.1042, over 5675882.23 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:06:07,632 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 06:06:17,126 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2104, simple_loss=0.3164, pruned_loss=0.05217, over 944034.00 frames. 
+2023-03-11 06:06:17,127 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 06:06:37,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.964e+02 1.234e+03 1.660e+03 2.133e+03 4.756e+03, threshold=3.320e+03, percent-clipped=6.0
+2023-03-11 06:06:43,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=960763.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 06:06:57,499 INFO [train.py:968] (0/2) Epoch 22, batch 3050, giga_loss[loss=0.2385, simple_loss=0.3301, pruned_loss=0.0734, over 28896.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3506, pruned_loss=0.1007, over 5684953.32 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3454, pruned_loss=0.08932, over 4534177.16 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3501, pruned_loss=0.1017, over 5672086.87 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:07:32,597 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 06:07:35,636 INFO [train.py:968] (0/2) Epoch 22, batch 3100, giga_loss[loss=0.2756, simple_loss=0.3503, pruned_loss=0.1004, over 28532.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.347, pruned_loss=0.09783, over 5688286.50 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3449, pruned_loss=0.08906, over 4604269.01 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3471, pruned_loss=0.09936, over 5674593.00 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:07:48,965 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=960847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:07:51,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=960850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:07:54,121 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=960854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:07:56,758 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.941e+02 1.224e+03 1.575e+03 2.356e+03 8.672e+03, threshold=3.150e+03, percent-clipped=9.0
+2023-03-11 06:07:59,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-11 06:08:16,511 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=960879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:08:17,596 INFO [train.py:968] (0/2) Epoch 22, batch 3150, giga_loss[loss=0.2882, simple_loss=0.3564, pruned_loss=0.11, over 28727.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3451, pruned_loss=0.09611, over 5688113.76 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3442, pruned_loss=0.08879, over 4642649.95 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3457, pruned_loss=0.09772, over 5671521.59 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:08:36,214 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=960906.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:08:38,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=960909.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:08:59,344 INFO [train.py:968] (0/2) Epoch 22, batch 3200, libri_loss[loss=0.2224, simple_loss=0.3043, pruned_loss=0.07028, over 29354.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3443, pruned_loss=0.09562, over 5688590.66 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3432, pruned_loss=0.08834, over 4684992.95 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3456, pruned_loss=0.0975, over 5669441.17 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:09:04,758 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=960938.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 06:09:18,511 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.030e+02 1.261e+03 1.526e+03 2.139e+03 1.139e+04, threshold=3.051e+03, percent-clipped=6.0
+2023-03-11 06:09:36,951 INFO [train.py:968] (0/2) Epoch 22, batch 3250, giga_loss[loss=0.2961, simple_loss=0.3648, pruned_loss=0.1137, over 28892.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3455, pruned_loss=0.09592, over 5689177.74 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3423, pruned_loss=0.08799, over 4739966.82 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3473, pruned_loss=0.09809, over 5672294.88 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:10:17,801 INFO [train.py:968] (0/2) Epoch 22, batch 3300, giga_loss[loss=0.2722, simple_loss=0.3447, pruned_loss=0.09986, over 28991.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3484, pruned_loss=0.09798, over 5695836.46 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3424, pruned_loss=0.08821, over 4767801.60 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3499, pruned_loss=0.09976, over 5678439.59 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:10:34,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5049, 2.6173, 2.3414, 1.9625], device='cuda:0'), covar=tensor([0.2308, 0.1941, 0.2350, 0.2587], device='cuda:0'), in_proj_covar=tensor([0.1962, 0.1891, 0.1827, 0.1963], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 06:10:37,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.540e+02 1.290e+03 1.579e+03 2.175e+03 5.317e+03, threshold=3.158e+03, percent-clipped=6.0
+2023-03-11 06:11:01,465 INFO [train.py:968] (0/2) Epoch 22, batch 3350, giga_loss[loss=0.2773, simple_loss=0.3555, pruned_loss=0.09958, over 29031.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3499, pruned_loss=0.09949, over 5695498.01 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3422, pruned_loss=0.08802, over 4795650.86 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3514, pruned_loss=0.1013, over 5677617.35 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:11:14,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2675, 1.9832, 1.4884, 0.5787], device='cuda:0'), covar=tensor([0.5359, 0.2763, 0.4231, 0.5849], device='cuda:0'), in_proj_covar=tensor([0.1730, 0.1627, 0.1584, 0.1411], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 06:11:42,593 INFO [train.py:968] (0/2) Epoch 22, batch 3400, libri_loss[loss=0.275, simple_loss=0.3591, pruned_loss=0.09547, over 29235.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3508, pruned_loss=0.1005, over 5701108.37 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3425, pruned_loss=0.08819, over 4827776.12 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3519, pruned_loss=0.1021, over 5681851.84 frames. ], batch size: 97, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:11:55,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2680, 3.1167, 2.9292, 1.3162], device='cuda:0'), covar=tensor([0.0923, 0.0977, 0.0856, 0.2418], device='cuda:0'), in_proj_covar=tensor([0.1206, 0.1127, 0.0951, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 06:12:05,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.536e+02 1.395e+03 1.820e+03 2.354e+03 7.207e+03, threshold=3.641e+03, percent-clipped=7.0
+2023-03-11 06:12:25,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-11 06:12:25,766 INFO [train.py:968] (0/2) Epoch 22, batch 3450, giga_loss[loss=0.3201, simple_loss=0.3804, pruned_loss=0.1299, over 27539.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3516, pruned_loss=0.1018, over 5690565.00 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3422, pruned_loss=0.088, over 4850501.43 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3529, pruned_loss=0.1035, over 5673996.22 frames. ], batch size: 472, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:13:02,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=961229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:13:04,182 INFO [train.py:968] (0/2) Epoch 22, batch 3500, giga_loss[loss=0.247, simple_loss=0.3338, pruned_loss=0.08013, over 28675.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3511, pruned_loss=0.1008, over 5692622.70 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3424, pruned_loss=0.08798, over 4873243.65 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3523, pruned_loss=0.1025, over 5677758.54 frames. ], batch size: 60, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:13:12,422 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6416, 1.7510, 1.8527, 1.4479], device='cuda:0'), covar=tensor([0.1822, 0.2511, 0.1500, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.0905, 0.0704, 0.0952, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 06:13:26,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.133e+02 1.190e+03 1.589e+03 2.155e+03 4.970e+03, threshold=3.177e+03, percent-clipped=3.0
+2023-03-11 06:13:44,688 INFO [train.py:968] (0/2) Epoch 22, batch 3550, giga_loss[loss=0.283, simple_loss=0.3628, pruned_loss=0.1016, over 28199.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3506, pruned_loss=0.0995, over 5702043.97 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3419, pruned_loss=0.08783, over 4898590.13 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3521, pruned_loss=0.1013, over 5685503.53 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:13:45,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-11 06:14:27,492 INFO [train.py:968] (0/2) Epoch 22, batch 3600, giga_loss[loss=0.2656, simple_loss=0.3573, pruned_loss=0.08693, over 29041.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3505, pruned_loss=0.09848, over 5703482.75 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3421, pruned_loss=0.08787, over 4928077.59 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3517, pruned_loss=0.1002, over 5685005.68 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:14:45,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.597e+02 1.168e+03 1.604e+03 2.146e+03 6.910e+03, threshold=3.208e+03, percent-clipped=6.0
+2023-03-11 06:14:57,771 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=961372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:14:58,408 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-11 06:14:59,610 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=961375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:15:03,371 INFO [train.py:968] (0/2) Epoch 22, batch 3650, giga_loss[loss=0.2739, simple_loss=0.3486, pruned_loss=0.09956, over 27938.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3486, pruned_loss=0.09721, over 5710459.99 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3418, pruned_loss=0.08756, over 4960966.32 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3502, pruned_loss=0.09912, over 5690120.45 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:15:03,837 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-11 06:15:21,874 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=961404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:15:25,843 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3789, 1.6554, 1.6387, 1.5830], device='cuda:0'), covar=tensor([0.2068, 0.1968, 0.2536, 0.1979], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0751, 0.0716, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 06:15:43,826 INFO [train.py:968] (0/2) Epoch 22, batch 3700, giga_loss[loss=0.2854, simple_loss=0.3549, pruned_loss=0.108, over 28368.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3474, pruned_loss=0.09729, over 5700751.58 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3416, pruned_loss=0.08756, over 4985329.38 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3488, pruned_loss=0.09907, over 5682828.24 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:15:52,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5553, 1.7098, 1.4642, 1.5448], device='cuda:0'), covar=tensor([0.2461, 0.2493, 0.2613, 0.2298], device='cuda:0'), in_proj_covar=tensor([0.1510, 0.1093, 0.1331, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 06:16:04,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.901e+02 1.105e+03 1.340e+03 1.814e+03 4.428e+03, threshold=2.680e+03, percent-clipped=3.0
+2023-03-11 06:16:10,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4402, 1.6938, 1.6508, 1.2434], device='cuda:0'), covar=tensor([0.1949, 0.3020, 0.1678, 0.2115], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0704, 0.0951, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 06:16:22,744 INFO [train.py:968] (0/2) Epoch 22, batch 3750, giga_loss[loss=0.252, simple_loss=0.3294, pruned_loss=0.08725, over 28977.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3448, pruned_loss=0.09552, over 5701192.93 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3419, pruned_loss=0.0877, over 4999989.35 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3459, pruned_loss=0.09708, over 5696395.60 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:16:59,312 INFO [train.py:968] (0/2) Epoch 22, batch 3800, libri_loss[loss=0.2622, simple_loss=0.3472, pruned_loss=0.08864, over 29573.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3436, pruned_loss=0.09526, over 5698716.83 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3419, pruned_loss=0.08781, over 5020327.45 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3446, pruned_loss=0.09668, over 5697869.22 frames. ], batch size: 77, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:17:18,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5361, 1.3954, 4.3425, 3.5314], device='cuda:0'), covar=tensor([0.1609, 0.2761, 0.0417, 0.1009], device='cuda:0'), in_proj_covar=tensor([0.0755, 0.0641, 0.0957, 0.0906], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 06:17:24,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.415e+02 1.121e+03 1.370e+03 1.837e+03 4.083e+03, threshold=2.741e+03, percent-clipped=6.0
+2023-03-11 06:17:42,142 INFO [train.py:968] (0/2) Epoch 22, batch 3850, libri_loss[loss=0.253, simple_loss=0.339, pruned_loss=0.08355, over 29522.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3455, pruned_loss=0.0968, over 5699508.82 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3416, pruned_loss=0.0876, over 5041781.55 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3465, pruned_loss=0.0983, over 5694445.05 frames. ], batch size: 82, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:18:22,626 INFO [train.py:968] (0/2) Epoch 22, batch 3900, libri_loss[loss=0.3014, simple_loss=0.3678, pruned_loss=0.1176, over 29554.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.345, pruned_loss=0.09618, over 5703013.81 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3411, pruned_loss=0.08745, over 5059207.92 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3462, pruned_loss=0.09766, over 5694899.19 frames. ], batch size: 76, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:18:41,326 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.566e+02 1.068e+03 1.364e+03 1.959e+03 1.079e+04, threshold=2.728e+03, percent-clipped=9.0
+2023-03-11 06:19:00,521 INFO [train.py:968] (0/2) Epoch 22, batch 3950, giga_loss[loss=0.2753, simple_loss=0.3444, pruned_loss=0.1031, over 28691.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3453, pruned_loss=0.09572, over 5707255.04 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3415, pruned_loss=0.08775, over 5085239.26 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3462, pruned_loss=0.09701, over 5702228.02 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:19:40,581 INFO [train.py:968] (0/2) Epoch 22, batch 4000, giga_loss[loss=0.2641, simple_loss=0.3362, pruned_loss=0.09594, over 28762.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3448, pruned_loss=0.09528, over 5697188.84 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3416, pruned_loss=0.08779, over 5098153.24 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3455, pruned_loss=0.09648, over 5697834.51 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:19:56,660 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-11 06:20:01,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.498e+02 1.065e+03 1.244e+03 1.696e+03 5.325e+03, threshold=2.487e+03, percent-clipped=2.0
+2023-03-11 06:20:20,652 INFO [train.py:968] (0/2) Epoch 22, batch 4050, giga_loss[loss=0.2377, simple_loss=0.3253, pruned_loss=0.075, over 28926.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3445, pruned_loss=0.09553, over 5701351.48 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3414, pruned_loss=0.08766, over 5109339.80 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3452, pruned_loss=0.09668, over 5699883.80 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:20:34,994 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=961800.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:21:00,365 INFO [train.py:968] (0/2) Epoch 22, batch 4100, libri_loss[loss=0.282, simple_loss=0.3633, pruned_loss=0.1004, over 27830.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3418, pruned_loss=0.09411, over 5707489.86 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3414, pruned_loss=0.08766, over 5127299.63 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3425, pruned_loss=0.09517, over 5703773.35 frames. ], batch size: 116, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:21:08,963 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2459, 2.2158, 2.1681, 1.9606], device='cuda:0'), covar=tensor([0.1996, 0.2652, 0.2330, 0.2419], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0747, 0.0711, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 06:21:22,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.190e+02 1.098e+03 1.478e+03 1.906e+03 5.862e+03, threshold=2.956e+03, percent-clipped=14.0
+2023-03-11 06:21:37,363 INFO [train.py:968] (0/2) Epoch 22, batch 4150, libri_loss[loss=0.247, simple_loss=0.3381, pruned_loss=0.07793, over 29688.00 frames. ], tot_loss[loss=0.262, simple_loss=0.339, pruned_loss=0.09251, over 5715290.82 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3418, pruned_loss=0.08787, over 5142183.28 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3392, pruned_loss=0.0933, over 5709281.44 frames. ], batch size: 91, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:21:38,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5290, 3.3466, 1.5553, 1.6029], device='cuda:0'), covar=tensor([0.0986, 0.0268, 0.0911, 0.1395], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0545, 0.0382, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:0')
+2023-03-11 06:21:49,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=961895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:22:16,590 INFO [train.py:968] (0/2) Epoch 22, batch 4200, giga_loss[loss=0.2882, simple_loss=0.3637, pruned_loss=0.1063, over 28692.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3384, pruned_loss=0.09291, over 5705316.97 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3417, pruned_loss=0.08801, over 5147474.22 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3386, pruned_loss=0.0935, over 5705745.93 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:22:37,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.892e+02 1.226e+03 1.485e+03 2.018e+03 4.277e+03, threshold=2.970e+03, percent-clipped=7.0
+2023-03-11 06:22:46,599 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=961971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:22:54,557 INFO [train.py:968] (0/2) Epoch 22, batch 4250, giga_loss[loss=0.2511, simple_loss=0.3267, pruned_loss=0.08771, over 29039.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3381, pruned_loss=0.09278, over 5711205.85 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.342, pruned_loss=0.08831, over 5164678.76 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3378, pruned_loss=0.09313, over 5710785.36 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:23:03,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=961991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:23:06,021 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1305, 2.4253, 2.1918, 1.9265], device='cuda:0'), covar=tensor([0.3214, 0.2214, 0.2443, 0.2837], device='cuda:0'), in_proj_covar=tensor([0.1972, 0.1900, 0.1834, 0.1975], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 06:23:11,573 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-962000.pt
+2023-03-11 06:23:37,643 INFO [train.py:968] (0/2) Epoch 22, batch 4300, giga_loss[loss=0.268, simple_loss=0.3387, pruned_loss=0.0986, over 28876.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3383, pruned_loss=0.09363, over 5706671.10 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3425, pruned_loss=0.08851, over 5174463.89 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3376, pruned_loss=0.09383, over 5707232.40 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:23:46,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2499, 4.0627, 3.8796, 1.8399], device='cuda:0'), covar=tensor([0.0591, 0.0791, 0.0769, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1208, 0.1122, 0.0950, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 06:23:58,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7013, 1.9179, 1.5583, 1.9303], device='cuda:0'), covar=tensor([0.2671, 0.2811, 0.3102, 0.2578], device='cuda:0'), in_proj_covar=tensor([0.1510, 0.1092, 0.1330, 0.0990], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 06:24:00,908 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.309e+02 1.186e+03 1.460e+03 2.095e+03 7.029e+03, threshold=2.920e+03, percent-clipped=9.0
+2023-03-11 06:24:17,996 INFO [train.py:968] (0/2) Epoch 22, batch 4350, giga_loss[loss=0.2692, simple_loss=0.3445, pruned_loss=0.09697, over 28751.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3358, pruned_loss=0.09267, over 5710006.63 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3429, pruned_loss=0.08865, over 5185210.52 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3349, pruned_loss=0.09279, over 5707932.97 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:24:56,844 INFO [train.py:968] (0/2) Epoch 22, batch 4400, giga_loss[loss=0.2961, simple_loss=0.3593, pruned_loss=0.1165, over 28853.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3339, pruned_loss=0.09236, over 5706906.77 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3429, pruned_loss=0.08869, over 5192319.55 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3331, pruned_loss=0.09245, over 5703461.99 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:25:07,440 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=962143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:25:10,204 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-03-11 06:25:21,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.337e+02 1.110e+03 1.390e+03 1.780e+03 4.837e+03, threshold=2.780e+03, percent-clipped=9.0
+2023-03-11 06:25:31,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7239, 5.5552, 5.2829, 2.7492], device='cuda:0'), covar=tensor([0.0484, 0.0645, 0.0800, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.1213, 0.1126, 0.0956, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 06:25:31,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=962175.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:25:35,604 INFO [train.py:968] (0/2) Epoch 22, batch 4450, giga_loss[loss=0.3023, simple_loss=0.3612, pruned_loss=0.1217, over 28813.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.333, pruned_loss=0.09167, over 5710285.35 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3431, pruned_loss=0.08882, over 5209626.19 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3319, pruned_loss=0.0917, over 5703325.11 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:26:06,834 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 06:26:08,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2980, 1.5125, 1.4273, 1.4783], device='cuda:0'), covar=tensor([0.0625, 0.0290, 0.0301, 0.0718], device='cuda:0'), in_proj_covar=tensor([0.0186, 0.0117, 0.0116, 0.0220], device='cuda:0'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 06:26:16,590 INFO [train.py:968] (0/2) Epoch 22, batch 4500, giga_loss[loss=0.2687, simple_loss=0.3435, pruned_loss=0.09694, over 28724.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3354, pruned_loss=0.09293, over 5710249.17 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3431, pruned_loss=0.08893, over 5226486.50 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3344, pruned_loss=0.09296, over 5700631.31 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:26:39,236 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-11 06:26:42,469 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.257e+02 1.170e+03 1.397e+03 1.704e+03 5.043e+03, threshold=2.793e+03, percent-clipped=4.0
+2023-03-11 06:26:51,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=962270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:26:59,359 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 06:27:00,982 INFO [train.py:968] (0/2) Epoch 22, batch 4550, giga_loss[loss=0.2376, simple_loss=0.3283, pruned_loss=0.07345, over 28934.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3371, pruned_loss=0.09289, over 5718438.62 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.343, pruned_loss=0.08888, over 5229891.07 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3363, pruned_loss=0.09297, over 5710053.43 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:27:29,209 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=962318.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:27:30,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=962321.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 06:27:38,047 INFO [train.py:968] (0/2) Epoch 22, batch 4600, giga_loss[loss=0.2986, simple_loss=0.3696, pruned_loss=0.1138, over 28580.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3393, pruned_loss=0.09354, over 5724188.91 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3433, pruned_loss=0.08925, over 5256353.87 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3382, pruned_loss=0.09344, over 5710752.28 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:27:51,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=962346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:27:55,302 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=962350.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:28:04,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.079e+02 1.117e+03 1.301e+03 1.805e+03 1.046e+04, threshold=2.601e+03, percent-clipped=13.0
+2023-03-11 06:28:12,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=962366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:28:23,781 INFO [train.py:968] (0/2) Epoch 22, batch 4650, giga_loss[loss=0.2632, simple_loss=0.3443, pruned_loss=0.091, over 28962.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3403, pruned_loss=0.0935, over 5716548.38 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3434, pruned_loss=0.08927, over 5265758.49 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3394, pruned_loss=0.09346, over 5703757.12 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:28:51,471 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=962413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:28:53,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=962416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:29:06,830 INFO [train.py:968] (0/2) Epoch 22, batch 4700, giga_loss[loss=0.2611, simple_loss=0.3391, pruned_loss=0.0915, over 28915.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3407, pruned_loss=0.09372, over 5696144.59 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3445, pruned_loss=0.09023, over 5271809.29 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.339, pruned_loss=0.093, over 5690591.05 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:29:14,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8125, 4.6441, 4.4291, 2.0447], device='cuda:0'), covar=tensor([0.0636, 0.0796, 0.0831, 0.2065], device='cuda:0'), in_proj_covar=tensor([0.1217, 0.1127, 0.0956, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 06:29:19,515 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=962445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:29:29,639 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.681e+02 1.106e+03 1.351e+03 1.755e+03 4.750e+03, threshold=2.702e+03, percent-clipped=7.0
+2023-03-11 06:29:48,273 INFO [train.py:968] (0/2) Epoch 22, batch 4750, giga_loss[loss=0.2877, simple_loss=0.3671, pruned_loss=0.1041, over 28755.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3407, pruned_loss=0.09343, over 5696830.28 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3447, pruned_loss=0.09045, over 5277743.98 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.339, pruned_loss=0.09276, over 5697211.33 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:29:54,232 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=962489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:29:56,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=962492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:06,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4875, 2.3122, 1.7079, 0.7295], device='cuda:0'), covar=tensor([0.6933, 0.2744, 0.4556, 0.6926], device='cuda:0'), in_proj_covar=tensor([0.1738, 0.1624, 0.1589, 0.1415], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 06:30:10,587 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=962509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:11,413 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-11 06:30:13,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=962512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:17,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=962518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:19,961 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=962521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:27,322 INFO [train.py:968] (0/2) Epoch 22, batch 4800, giga_loss[loss=0.245, simple_loss=0.3268, pruned_loss=0.08161, over 28937.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3406, pruned_loss=0.0932, over 5710213.44 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.0904, over 5290687.77 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3392, pruned_loss=0.09274, over 5706477.14 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:30:29,467 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.12 vs. limit=2.0
+2023-03-11 06:30:36,059 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=962541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:49,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.418e+02 1.220e+03 1.554e+03 2.064e+03 5.007e+03, threshold=3.109e+03, percent-clipped=11.0
+2023-03-11 06:31:04,172 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-11 06:31:05,548 INFO [train.py:968] (0/2) Epoch 22, batch 4850, giga_loss[loss=0.2598, simple_loss=0.3376, pruned_loss=0.09098, over 28979.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3424, pruned_loss=0.09468, over 5710288.28 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3449, pruned_loss=0.09039, over 5313940.58 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3411, pruned_loss=0.09447, over 5700971.47 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:31:48,331 INFO [train.py:968] (0/2) Epoch 22, batch 4900, giga_loss[loss=0.2992, simple_loss=0.3706, pruned_loss=0.1139, over 28862.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3456, pruned_loss=0.09665, over 5709365.41 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3451, pruned_loss=0.09049, over 5316253.71 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3443, pruned_loss=0.09645, over 5703648.20 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:32:12,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.137e+02 1.379e+03 1.665e+03 2.340e+03 5.606e+03, threshold=3.331e+03, percent-clipped=9.0
+2023-03-11 06:32:12,451 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=962661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:32:15,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=962664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:32:28,718 INFO [train.py:968] (0/2) Epoch 22, batch 4950, giga_loss[loss=0.3229, simple_loss=0.386, pruned_loss=0.1299, over 27607.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3483, pruned_loss=0.09801, over 5713418.70 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3453, pruned_loss=0.09053, over 5325014.30 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3472, pruned_loss=0.0979, over 5706431.50 frames. ], batch size: 472, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:32:39,698 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=962693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:33:12,753 INFO [train.py:968] (0/2) Epoch 22, batch 5000, giga_loss[loss=0.2981, simple_loss=0.3714, pruned_loss=0.1124, over 28947.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3505, pruned_loss=0.0991, over 5712997.84 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3453, pruned_loss=0.09053, over 5327957.22 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3496, pruned_loss=0.09906, over 5706722.16 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:33:37,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.554e+02 1.300e+03 1.670e+03 2.305e+03 5.465e+03, threshold=3.339e+03, percent-clipped=6.0
+2023-03-11 06:33:39,902 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=962763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:33:52,341 INFO [train.py:968] (0/2) Epoch 22, batch 5050, giga_loss[loss=0.2469, simple_loss=0.3217, pruned_loss=0.08607, over 28870.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3517, pruned_loss=0.09985, over 5702519.80 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.346, pruned_loss=0.09086, over 5335619.29 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3506, pruned_loss=0.09977, over 5701188.21 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:34:32,943 INFO [train.py:968] (0/2) Epoch 22, batch 5100, giga_loss[loss=0.2694, simple_loss=0.3576, pruned_loss=0.09058, over 28984.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3506, pruned_loss=0.09916, over 5708125.80 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3459, pruned_loss=0.09086, over 5346561.62 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3499, pruned_loss=0.09927, over 5703757.54 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:34:54,995 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=962859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:34:56,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.908e+02 1.164e+03 1.470e+03 1.815e+03 3.978e+03, threshold=2.941e+03, percent-clipped=4.0
+2023-03-11 06:34:59,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1023, 1.2505, 1.1089, 0.8752], device='cuda:0'), covar=tensor([0.0988, 0.0559, 0.1124, 0.1156], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0445, 0.0517, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 06:35:11,164 INFO [train.py:968] (0/2) Epoch 22, batch 5150, libri_loss[loss=0.3168, simple_loss=0.3938, pruned_loss=0.1199, over 28647.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3499, pruned_loss=0.09882, over 5703185.47 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3462, pruned_loss=0.09088, over 5355702.91 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3492, pruned_loss=0.09914, over 5704818.89 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:35:13,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-11 06:35:14,979 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-11 06:35:38,411 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-11 06:35:50,674 INFO [train.py:968] (0/2) Epoch 22, batch 5200, giga_loss[loss=0.2739, simple_loss=0.3481, pruned_loss=0.09983, over 28594.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3472, pruned_loss=0.09791, over 5699655.34 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3465, pruned_loss=0.09109, over 5365105.68 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3465, pruned_loss=0.09816, over 5700176.00 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:36:08,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5718, 3.7033, 1.6008, 1.6976], device='cuda:0'), covar=tensor([0.0931, 0.0297, 0.0917, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0550, 0.0384, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 06:36:15,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.382e+02 1.196e+03 1.429e+03 1.992e+03 5.207e+03, threshold=2.858e+03, percent-clipped=6.0
+2023-03-11 06:36:18,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4471, 1.6920, 1.6726, 1.5219], device='cuda:0'), covar=tensor([0.2062, 0.2291, 0.2379, 0.2257], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0748, 0.0713, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 06:36:29,611 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=962980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:36:30,038 INFO [train.py:968] (0/2) Epoch 22, batch 5250, giga_loss[loss=0.2584, simple_loss=0.3264, pruned_loss=0.09517, over 28934.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3438, pruned_loss=0.09604, over 5708827.92 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3464, pruned_loss=0.09099, over 5378670.36 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3433, pruned_loss=0.09647, over 5704547.71 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:37:10,541 INFO [train.py:968] (0/2) Epoch 22, batch 5300, giga_loss[loss=0.2957, simple_loss=0.3653, pruned_loss=0.1131, over 26758.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3431, pruned_loss=0.09463, over 5700002.00 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3464, pruned_loss=0.09103, over 5379171.73 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3426, pruned_loss=0.095, over 5702099.85 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:37:36,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.025e+02 1.090e+03 1.352e+03 1.842e+03 4.321e+03, threshold=2.703e+03, percent-clipped=6.0
+2023-03-11 06:37:53,733 INFO [train.py:968] (0/2) Epoch 22, batch 5350, giga_loss[loss=0.2749, simple_loss=0.3603, pruned_loss=0.0948, over 27936.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3442, pruned_loss=0.09358, over 5706581.67 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3465, pruned_loss=0.09102, over 5384473.27 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3437, pruned_loss=0.09391, over 5706333.66 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:38:23,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7286, 1.8661, 1.5746, 1.7091], device='cuda:0'), covar=tensor([0.2414, 0.2607, 0.2906, 0.2340], device='cuda:0'), in_proj_covar=tensor([0.1503, 0.1085, 0.1322, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 06:38:35,768 INFO [train.py:968] (0/2) Epoch 22, batch 5400, giga_loss[loss=0.2744, simple_loss=0.3376, pruned_loss=0.1056, over 28551.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3438, pruned_loss=0.09387, over 5712827.90 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3467, pruned_loss=0.09111, over 5389252.64 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3432, pruned_loss=0.09409, over 5711203.62 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:38:40,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=963138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:38:48,741 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-11 06:38:57,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:38:59,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.534e+02 1.147e+03 1.433e+03 2.031e+03 4.598e+03, threshold=2.867e+03, percent-clipped=9.0
+2023-03-11 06:39:02,021 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:39:02,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6996, 1.9497, 1.5540, 1.7507], device='cuda:0'), covar=tensor([0.2523, 0.2577, 0.2990, 0.2338], device='cuda:0'), in_proj_covar=tensor([0.1504, 0.1087, 0.1324, 0.0990], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 06:39:15,163 INFO [train.py:968] (0/2) Epoch 22, batch 5450, giga_loss[loss=0.2575, simple_loss=0.3216, pruned_loss=0.09672, over 28779.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3421, pruned_loss=0.09413, over 5719637.81 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3469, pruned_loss=0.09124, over 5399066.55 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3414, pruned_loss=0.09424, over 5715428.13 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:39:54,939 INFO [train.py:968] (0/2) Epoch 22, batch 5500, giga_loss[loss=0.2695, simple_loss=0.3416, pruned_loss=0.09872, over 28975.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3407, pruned_loss=0.09452, over 5710632.21 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3474, pruned_loss=0.09158, over 5393741.17 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3396, pruned_loss=0.09444, over 5721197.23 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:39:57,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=963234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:40:19,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.140e+02 1.169e+03 1.389e+03 1.731e+03 4.071e+03, threshold=2.778e+03, percent-clipped=4.0
+2023-03-11 06:40:34,602 INFO [train.py:968] (0/2) Epoch 22, batch 5550, giga_loss[loss=0.2331, simple_loss=0.3117, pruned_loss=0.07731, over 28850.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3384, pruned_loss=0.09414, over 5720257.71 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3474, pruned_loss=0.09156, over 5410719.80 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3372, pruned_loss=0.09417, over 5723501.31 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:40:34,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:40:37,770 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=963284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:41:00,933 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=963313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:41:17,402 INFO [train.py:968] (0/2) Epoch 22, batch 5600, giga_loss[loss=0.2568, simple_loss=0.3249, pruned_loss=0.09441, over 28793.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3384, pruned_loss=0.0943, over 5723714.06 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3477, pruned_loss=0.09173, over 5430125.96 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3369, pruned_loss=0.09428, over 5720085.66 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:41:19,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3878, 1.2561, 4.1273, 3.3500], device='cuda:0'), covar=tensor([0.1633, 0.2785, 0.0430, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0755, 0.0642, 0.0956, 0.0904], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 06:41:38,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=963355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:41:43,185 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.900e+02 1.201e+03 1.466e+03 1.944e+03 7.476e+03, threshold=2.932e+03, percent-clipped=11.0
+2023-03-11 06:41:53,673 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:41:55,147 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:41:57,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=963380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:41:58,292 INFO [train.py:968] (0/2) Epoch 22, batch 5650, giga_loss[loss=0.2028, simple_loss=0.2787, pruned_loss=0.0635, over 28717.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.336, pruned_loss=0.09343, over 5716283.78 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3476, pruned_loss=0.09169, over 5442294.57 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3346, pruned_loss=0.09351, over 5709964.20 frames. ], batch size: 60, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:42:21,037 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=963409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:42:38,836 INFO [train.py:968] (0/2) Epoch 22, batch 5700, giga_loss[loss=0.2196, simple_loss=0.306, pruned_loss=0.06661, over 28927.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3324, pruned_loss=0.09172, over 5716507.38 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3475, pruned_loss=0.09161, over 5444989.92 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3312, pruned_loss=0.09186, over 5711549.00 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:43:04,918 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.471e+02 1.245e+03 1.593e+03 2.266e+03 6.482e+03, threshold=3.185e+03, percent-clipped=17.0
+2023-03-11 06:43:18,696 INFO [train.py:968] (0/2) Epoch 22, batch 5750, giga_loss[loss=0.2231, simple_loss=0.304, pruned_loss=0.07111, over 29041.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3294, pruned_loss=0.09039, over 5699276.13 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3477, pruned_loss=0.09169, over 5438910.93 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.328, pruned_loss=0.09045, over 5706975.61 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:43:31,155 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:43:33,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=963501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:43:55,099 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=963530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:43:55,539 INFO [train.py:968] (0/2) Epoch 22, batch 5800, giga_loss[loss=0.2478, simple_loss=0.3227, pruned_loss=0.08645, over 28700.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3287, pruned_loss=0.08981, over 5707170.02 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3472, pruned_loss=0.09143, over 5454845.55 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3273, pruned_loss=0.09002, over 5707092.96 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:43:58,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=963534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:44:02,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=963540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:44:20,106 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.906e+02 1.185e+03 1.593e+03 2.086e+03 4.885e+03, threshold=3.186e+03, percent-clipped=2.0
+2023-03-11 06:44:20,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9259, 3.7485, 3.5460, 1.7900], device='cuda:0'), covar=tensor([0.0695, 0.0814, 0.0768, 0.2073], device='cuda:0'), in_proj_covar=tensor([0.1220, 0.1135, 0.0962, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 06:44:26,545 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-11 06:44:34,382 INFO [train.py:968] (0/2) Epoch 22, batch 5850, giga_loss[loss=0.2619, simple_loss=0.3334, pruned_loss=0.09525, over 28935.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3313, pruned_loss=0.09086, over 5710646.78 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3471, pruned_loss=0.09148, over 5467999.38 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3298, pruned_loss=0.09094, over 5705590.89 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:44:56,074 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-11 06:45:13,516 INFO [train.py:968] (0/2) Epoch 22, batch 5900, giga_loss[loss=0.2463, simple_loss=0.32, pruned_loss=0.08634, over 28495.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3351, pruned_loss=0.09247, over 5710949.67 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3472, pruned_loss=0.09157, over 5474752.04 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3336, pruned_loss=0.09246, over 5704165.34 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:45:40,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.919e+02 1.242e+03 1.520e+03 2.018e+03 7.331e+03, threshold=3.040e+03, percent-clipped=6.0
+2023-03-11 06:45:51,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:45:53,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=963680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:45:54,173 INFO [train.py:968] (0/2) Epoch 22, batch 5950, giga_loss[loss=0.246, simple_loss=0.3371, pruned_loss=0.07748, over 28872.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3387, pruned_loss=0.09372, over 5714914.65 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3474, pruned_loss=0.09167, over 5483225.04 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.337, pruned_loss=0.09368, over 5707986.28 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:45:56,712 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:45:59,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=963686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:46:09,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2774, 2.5977, 1.2380, 1.4726], device='cuda:0'), covar=tensor([0.0968, 0.0328, 0.0905, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0552, 0.0384, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 06:46:17,197 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=963709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:46:23,066 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=963715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:46:36,745 INFO [train.py:968] (0/2) Epoch 22, batch 6000, giga_loss[loss=0.2582, simple_loss=0.334, pruned_loss=0.09126, over 28261.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3409, pruned_loss=0.0946, over 5715602.51 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3474, pruned_loss=0.09169, over 5494151.64 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3394, pruned_loss=0.0946, over 5705642.61 frames. ], batch size: 77, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:46:36,749 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 06:46:45,308 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2108, simple_loss=0.3175, pruned_loss=0.05206, over 944034.00 frames. 
+2023-03-11 06:46:45,309 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 06:47:01,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=963750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:47:12,701 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.162e+02 1.307e+03 1.567e+03 2.199e+03 4.189e+03, threshold=3.135e+03, percent-clipped=10.0
+2023-03-11 06:47:24,725 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:47:25,197 INFO [train.py:968] (0/2) Epoch 22, batch 6050, giga_loss[loss=0.2369, simple_loss=0.3135, pruned_loss=0.08018, over 28468.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.343, pruned_loss=0.09585, over 5720141.17 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3469, pruned_loss=0.09149, over 5512143.44 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3421, pruned_loss=0.09619, over 5704195.85 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:48:09,942 INFO [train.py:968] (0/2) Epoch 22, batch 6100, giga_loss[loss=0.3026, simple_loss=0.3742, pruned_loss=0.1154, over 28676.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.348, pruned_loss=0.1003, over 5712567.87 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3465, pruned_loss=0.09137, over 5520952.63 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3475, pruned_loss=0.1008, over 5696400.06 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:48:43,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.795e+02 1.693e+03 2.054e+03 3.272e+03 6.230e+03, threshold=4.108e+03, percent-clipped=26.0
+2023-03-11 06:48:56,325 INFO [train.py:968] (0/2) Epoch 22, batch 6150, giga_loss[loss=0.3354, simple_loss=0.4, pruned_loss=0.1353, over 28822.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3549, pruned_loss=0.1061, over 5706202.69 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3463, pruned_loss=0.09129, over 5527507.66 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3549, pruned_loss=0.1069, over 5691435.32 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:49:08,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:49:12,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=963896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:49:17,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8030, 2.0888, 2.0562, 1.5894], device='cuda:0'), covar=tensor([0.1984, 0.2624, 0.1588, 0.1992], device='cuda:0'), in_proj_covar=tensor([0.0901, 0.0702, 0.0950, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 06:49:19,998 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:49:38,192 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=963925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:49:45,498 INFO [train.py:968] (0/2) Epoch 22, batch 6200, giga_loss[loss=0.2773, simple_loss=0.3519, pruned_loss=0.1014, over 28863.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3628, pruned_loss=0.1119, over 5694484.05 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3458, pruned_loss=0.09099, over 5525126.40 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3634, pruned_loss=0.113, over 5687450.34 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:49:54,207 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4491, 1.6637, 1.7238, 1.2742], device='cuda:0'), covar=tensor([0.1622, 0.2553, 0.1359, 0.1681], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0701, 0.0948, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 06:50:15,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.134e+03 1.904e+03 2.365e+03 3.507e+03 8.813e+03, threshold=4.731e+03, percent-clipped=16.0
+2023-03-11 06:50:29,349 INFO [train.py:968] (0/2) Epoch 22, batch 6250, giga_loss[loss=0.2719, simple_loss=0.3557, pruned_loss=0.09401, over 28980.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3666, pruned_loss=0.1151, over 5703793.70 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3454, pruned_loss=0.09085, over 5538173.89 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3681, pruned_loss=0.1169, over 5692239.76 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:50:32,798 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:50:34,810 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6259, 1.3049, 4.3456, 3.5600], device='cuda:0'), covar=tensor([0.1520, 0.2757, 0.0422, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0758, 0.0644, 0.0962, 0.0911], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 06:50:46,809 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-964000.pt
+2023-03-11 06:51:15,731 INFO [train.py:968] (0/2) Epoch 22, batch 6300, giga_loss[loss=0.3582, simple_loss=0.4128, pruned_loss=0.1518, over 28945.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3743, pruned_loss=0.1217, over 5695521.03 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3456, pruned_loss=0.09093, over 5543419.79 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3757, pruned_loss=0.1235, over 5683838.16 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:51:45,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.608e+03 2.085e+03 2.987e+03 5.428e+03, threshold=4.171e+03, percent-clipped=4.0
+2023-03-11 06:52:01,509 INFO [train.py:968] (0/2) Epoch 22, batch 6350, libri_loss[loss=0.2606, simple_loss=0.3484, pruned_loss=0.08638, over 25974.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3768, pruned_loss=0.1235, over 5691212.11 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3455, pruned_loss=0.09074, over 5554693.12 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3793, pruned_loss=0.1264, over 5677902.43 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:52:12,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1045, 2.4131, 1.1051, 1.4034], device='cuda:0'), covar=tensor([0.1110, 0.0510, 0.0949, 0.1376], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0554, 0.0385, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 06:52:22,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3386, 1.8654, 1.5400, 1.5183], device='cuda:0'), covar=tensor([0.0762, 0.0324, 0.0305, 0.0837], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 06:52:30,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9359, 2.0680, 1.6881, 2.1061], device='cuda:0'), covar=tensor([0.2321, 0.2555, 0.2837, 0.2354], device='cuda:0'), in_proj_covar=tensor([0.1507, 0.1092, 0.1329, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 06:52:51,621 INFO [train.py:968] (0/2) Epoch 22, batch 6400, giga_loss[loss=0.3035, simple_loss=0.3764, pruned_loss=0.1153, over 28805.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.379, pruned_loss=0.1261, over 5674582.68 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3455, pruned_loss=0.09078, over 5559332.89 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3814, pruned_loss=0.1289, over 5661654.68 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:52:56,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.99 vs. limit=2.0
+2023-03-11 06:53:17,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=964155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:53:28,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.666e+03 2.141e+03 2.967e+03 7.785e+03, threshold=4.283e+03, percent-clipped=11.0
+2023-03-11 06:53:44,290 INFO [train.py:968] (0/2) Epoch 22, batch 6450, libri_loss[loss=0.3161, simple_loss=0.3894, pruned_loss=0.1214, over 26029.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3823, pruned_loss=0.1296, over 5673704.55 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3455, pruned_loss=0.09072, over 5562684.12 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.385, pruned_loss=0.1327, over 5663314.32 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:54:16,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3149, 1.6779, 1.2171, 0.7709], device='cuda:0'), covar=tensor([0.3708, 0.2205, 0.2221, 0.4672], device='cuda:0'), in_proj_covar=tensor([0.1748, 0.1638, 0.1596, 0.1418], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 06:54:37,722 INFO [train.py:968] (0/2) Epoch 22, batch 6500, giga_loss[loss=0.3465, simple_loss=0.4064, pruned_loss=0.1433, over 28826.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3849, pruned_loss=0.133, over 5660045.24 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3457, pruned_loss=0.09089, over 5568026.06 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3878, pruned_loss=0.1363, over 5649114.38 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:55:13,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.216e+03 1.922e+03 2.360e+03 3.509e+03 6.812e+03, threshold=4.720e+03, percent-clipped=11.0
+2023-03-11 06:55:19,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1315, 4.9572, 4.7470, 2.1762], device='cuda:0'), covar=tensor([0.0479, 0.0625, 0.0643, 0.1886], device='cuda:0'), in_proj_covar=tensor([0.1226, 0.1138, 0.0966, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 06:55:27,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=964278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:55:29,654 INFO [train.py:968] (0/2) Epoch 22, batch 6550, giga_loss[loss=0.3369, simple_loss=0.3909, pruned_loss=0.1414, over 28674.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3868, pruned_loss=0.1346, over 5650986.14 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3453, pruned_loss=0.09066, over 5574467.47 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3902, pruned_loss=0.1383, over 5637809.59 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:55:48,068 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=964298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:55:51,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=964301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:56:16,782 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=964330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:56:17,384 INFO [train.py:968] (0/2) Epoch 22, batch 6600, giga_loss[loss=0.2887, simple_loss=0.3551, pruned_loss=0.1112, over 28979.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3864, pruned_loss=0.1353, over 5656333.98 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3449, pruned_loss=0.09044, over 5585140.25 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3909, pruned_loss=0.14, over 5638408.23 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:56:45,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3274, 1.6466, 1.3560, 1.0263], device='cuda:0'), covar=tensor([0.2432, 0.2407, 0.2761, 0.2176], device='cuda:0'), in_proj_covar=tensor([0.1504, 0.1089, 0.1328, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 06:56:45,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=964358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:56:52,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 1.794e+03 2.190e+03 3.413e+03 8.344e+03, threshold=4.380e+03, percent-clipped=8.0
+2023-03-11 06:56:55,466 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1924, 1.1965, 3.9135, 3.3588], device='cuda:0'), covar=tensor([0.1727, 0.2770, 0.0465, 0.0945], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0651, 0.0971, 0.0918], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 06:57:05,956 INFO [train.py:968] (0/2) Epoch 22, batch 6650, giga_loss[loss=0.2706, simple_loss=0.3426, pruned_loss=0.09933, over 28678.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3852, pruned_loss=0.1352, over 5652237.93 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3454, pruned_loss=0.0911, over 5587315.03 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3895, pruned_loss=0.1396, over 5637970.71 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:57:10,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5794, 1.6418, 1.2159, 1.2093], device='cuda:0'), covar=tensor([0.0807, 0.0475, 0.0944, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0451, 0.0520, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 06:57:44,761 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=964421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:57:48,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3414, 1.5486, 1.4689, 1.3889], device='cuda:0'), covar=tensor([0.1389, 0.1382, 0.1888, 0.1436], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0756, 0.0719, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 06:57:48,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=964424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:57:54,617 INFO [train.py:968] (0/2) Epoch 22, batch 6700, giga_loss[loss=0.3121, simple_loss=0.3826, pruned_loss=0.1208, over 28714.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3847, pruned_loss=0.1342, over 5647241.61 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3453, pruned_loss=0.09103, over 5596875.77 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3898, pruned_loss=0.1395, over 5628856.83 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:57:56,705 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5121, 1.5507, 1.7413, 1.3650], device='cuda:0'), covar=tensor([0.1227, 0.2008, 0.1018, 0.1406], device='cuda:0'), in_proj_covar=tensor([0.0895, 0.0700, 0.0942, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 06:57:59,638 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-11 06:58:14,271 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=964453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:58:23,209 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=964463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:58:26,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.713e+03 2.148e+03 2.981e+03 7.699e+03, threshold=4.296e+03, percent-clipped=6.0
+2023-03-11 06:58:39,989 INFO [train.py:968] (0/2) Epoch 22, batch 6750, giga_loss[loss=0.3944, simple_loss=0.4324, pruned_loss=0.1782, over 28285.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3838, pruned_loss=0.1325, over 5658408.14 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3446, pruned_loss=0.09063, over 5606762.82 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3898, pruned_loss=0.1385, over 5636484.36 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:58:51,412 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 06:58:59,650 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=964501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:59:00,195 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7290, 1.0011, 2.8935, 2.8349], device='cuda:0'), covar=tensor([0.2134, 0.2887, 0.1013, 0.1329], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0649, 0.0970, 0.0918], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 06:59:02,407 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=964504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:59:29,772 INFO [train.py:968] (0/2) Epoch 22, batch 6800, giga_loss[loss=0.3073, simple_loss=0.3813, pruned_loss=0.1167, over 29067.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3863, pruned_loss=0.1343, over 5647965.22 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3449, pruned_loss=0.0909, over 5612745.66 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3916, pruned_loss=0.1397, over 5626093.66 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:59:31,240 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=964533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:59:44,054 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5231, 1.7598, 1.6438, 1.5510], device='cuda:0'), covar=tensor([0.1857, 0.1930, 0.2248, 0.2057], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0756, 0.0718, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 07:00:01,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.808e+03 2.335e+03 3.293e+03 1.016e+04, threshold=4.669e+03, percent-clipped=13.0
+2023-03-11 07:00:12,960 INFO [train.py:968] (0/2) Epoch 22, batch 6850, giga_loss[loss=0.3132, simple_loss=0.3869, pruned_loss=0.1198, over 28589.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3835, pruned_loss=0.1317, over 5638819.24 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3453, pruned_loss=0.09118, over 5612277.97 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3887, pruned_loss=0.1371, over 5622455.91 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:01:02,743 INFO [train.py:968] (0/2) Epoch 22, batch 6900, giga_loss[loss=0.3443, simple_loss=0.3882, pruned_loss=0.1502, over 26750.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.382, pruned_loss=0.1293, over 5647439.25 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3454, pruned_loss=0.09125, over 5618791.14 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.387, pruned_loss=0.1346, over 5629450.14 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:01:30,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3174, 1.4951, 1.5252, 1.2717], device='cuda:0'), covar=tensor([0.2702, 0.2483, 0.1743, 0.2360], device='cuda:0'), in_proj_covar=tensor([0.1966, 0.1905, 0.1835, 0.1962], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 07:01:36,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.516e+03 1.949e+03 2.569e+03 6.981e+03, threshold=3.898e+03, percent-clipped=3.0
+2023-03-11 07:01:48,291 INFO [train.py:968] (0/2) Epoch 22, batch 6950, giga_loss[loss=0.2911, simple_loss=0.3664, pruned_loss=0.1079, over 28790.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3787, pruned_loss=0.1259, over 5657493.34 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.345, pruned_loss=0.0911, over 5627243.55 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3842, pruned_loss=0.1314, over 5636432.42 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:02:10,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6333, 1.8802, 1.5043, 1.8455], device='cuda:0'), covar=tensor([0.2495, 0.2606, 0.2804, 0.2412], device='cuda:0'), in_proj_covar=tensor([0.1504, 0.1087, 0.1326, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 07:02:24,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2151, 1.5307, 1.4957, 1.3896], device='cuda:0'), covar=tensor([0.1999, 0.1737, 0.2460, 0.1936], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0754, 0.0716, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 07:02:36,813 INFO [train.py:968] (0/2) Epoch 22, batch 7000, giga_loss[loss=0.2896, simple_loss=0.358, pruned_loss=0.1106, over 28961.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3737, pruned_loss=0.1217, over 5663492.28 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3448, pruned_loss=0.09091, over 5629622.90 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3787, pruned_loss=0.1267, over 5645017.61 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:03:06,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.736e+03 2.275e+03 3.068e+03 1.045e+04, threshold=4.550e+03, percent-clipped=16.0
+2023-03-11 07:03:11,349 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=964770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:03:22,729 INFO [train.py:968] (0/2) Epoch 22, batch 7050, giga_loss[loss=0.3031, simple_loss=0.3658, pruned_loss=0.1203, over 27605.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3704, pruned_loss=0.1196, over 5649050.48 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3444, pruned_loss=0.09079, over 5625346.41 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3757, pruned_loss=0.1246, over 5638418.03 frames. ], batch size: 472, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:04:08,720 INFO [train.py:968] (0/2) Epoch 22, batch 7100, giga_loss[loss=0.3992, simple_loss=0.4093, pruned_loss=0.1945, over 23702.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3704, pruned_loss=0.1195, over 5649874.93 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.344, pruned_loss=0.09046, over 5629712.74 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3755, pruned_loss=0.1244, over 5638262.46 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:04:15,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=964838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:04:45,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.536e+03 1.988e+03 2.505e+03 4.847e+03, threshold=3.977e+03, percent-clipped=2.0
+2023-03-11 07:05:03,442 INFO [train.py:968] (0/2) Epoch 22, batch 7150, giga_loss[loss=0.2803, simple_loss=0.357, pruned_loss=0.1018, over 28727.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3705, pruned_loss=0.1195, over 5644100.33 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3437, pruned_loss=0.09033, over 5623196.48 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3754, pruned_loss=0.1242, over 5641377.85 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:05:27,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=964907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:05:51,020 INFO [train.py:968] (0/2) Epoch 22, batch 7200, giga_loss[loss=0.2755, simple_loss=0.3493, pruned_loss=0.1008, over 28823.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3695, pruned_loss=0.1185, over 5652343.75 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3435, pruned_loss=0.09022, over 5631047.71 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3748, pruned_loss=0.1236, over 5643869.04 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:06:20,314 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-11 07:06:27,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.715e+02 1.577e+03 2.034e+03 2.711e+03 8.770e+03, threshold=4.068e+03, percent-clipped=9.0
+2023-03-11 07:06:44,766 INFO [train.py:968] (0/2) Epoch 22, batch 7250, giga_loss[loss=0.2833, simple_loss=0.3409, pruned_loss=0.1129, over 23793.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.37, pruned_loss=0.1165, over 5657746.71 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3434, pruned_loss=0.09014, over 5636765.57 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3751, pruned_loss=0.1214, over 5646441.78 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:06:45,031 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=964981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:06:48,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=964984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:06:58,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5158, 1.7045, 1.6534, 1.3838], device='cuda:0'), covar=tensor([0.3019, 0.2557, 0.2010, 0.2595], device='cuda:0'), in_proj_covar=tensor([0.1966, 0.1903, 0.1832, 0.1960], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 07:07:18,665 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=965013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:07:35,500 INFO [train.py:968] (0/2) Epoch 22, batch 7300, giga_loss[loss=0.3167, simple_loss=0.3825, pruned_loss=0.1254, over 28780.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3727, pruned_loss=0.1167, over 5661216.21 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3434, pruned_loss=0.09016, over 5628456.74 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3767, pruned_loss=0.1206, over 5660292.02 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:07:43,794 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-11 07:08:02,281 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.92 vs. limit=5.0
+2023-03-11 07:08:12,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.724e+03 2.211e+03 2.851e+03 6.307e+03, threshold=4.422e+03, percent-clipped=11.0
+2023-03-11 07:08:27,949 INFO [train.py:968] (0/2) Epoch 22, batch 7350, giga_loss[loss=0.2901, simple_loss=0.3651, pruned_loss=0.1075, over 29046.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3726, pruned_loss=0.1174, over 5658664.03 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3431, pruned_loss=0.09012, over 5635226.68 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3767, pruned_loss=0.1211, over 5652934.73 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:09:04,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1868, 1.4786, 1.4751, 1.3306], device='cuda:0'), covar=tensor([0.2021, 0.1750, 0.2355, 0.1901], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0756, 0.0718, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 07:09:17,135 INFO [train.py:968] (0/2) Epoch 22, batch 7400, giga_loss[loss=0.2743, simple_loss=0.3525, pruned_loss=0.09805, over 29089.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3738, pruned_loss=0.1192, over 5662049.42 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3432, pruned_loss=0.09009, over 5634939.54 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.377, pruned_loss=0.1223, over 5657888.92 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:09:29,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=965145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:09:33,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4229, 1.6114, 1.6649, 1.1850], device='cuda:0'), covar=tensor([0.1932, 0.3009, 0.1699, 0.1969], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0706, 0.0948, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 07:09:54,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.160e+03 1.845e+03 2.518e+03 3.323e+03 9.841e+03, threshold=5.037e+03, percent-clipped=12.0
+2023-03-11 07:10:06,023 INFO [train.py:968] (0/2) Epoch 22, batch 7450, giga_loss[loss=0.31, simple_loss=0.3554, pruned_loss=0.1323, over 23753.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3702, pruned_loss=0.1174, over 5669992.23 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3426, pruned_loss=0.08959, over 5643853.82 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3744, pruned_loss=0.1214, over 5659667.80 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:10:54,473 INFO [train.py:968] (0/2) Epoch 22, batch 7500, giga_loss[loss=0.2774, simple_loss=0.3524, pruned_loss=0.1012, over 28529.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3688, pruned_loss=0.1174, over 5656168.71 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.343, pruned_loss=0.08984, over 5629579.91 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.372, pruned_loss=0.1206, over 5660194.11 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:11:19,618 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3473, 4.1569, 3.9584, 1.9696], device='cuda:0'), covar=tensor([0.0636, 0.0774, 0.0770, 0.2056], device='cuda:0'), in_proj_covar=tensor([0.1235, 0.1146, 0.0968, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 07:11:31,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.698e+03 1.982e+03 2.642e+03 6.990e+03, threshold=3.963e+03, percent-clipped=2.0
+2023-03-11 07:11:41,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2964, 1.3962, 1.3544, 1.2632], device='cuda:0'), covar=tensor([0.2203, 0.2223, 0.1928, 0.2095], device='cuda:0'), in_proj_covar=tensor([0.1973, 0.1913, 0.1845, 0.1971], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 07:11:43,537 INFO [train.py:968] (0/2) Epoch 22, batch 7550, giga_loss[loss=0.3082, simple_loss=0.3803, pruned_loss=0.1181, over 28923.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3684, pruned_loss=0.1165, over 5658456.10 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.0899, over 5634248.75 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3713, pruned_loss=0.1195, over 5657720.00 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:11:45,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=965282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:11:51,363 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=965288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:11:55,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=965291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:12:10,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=965307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:12:22,352 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=965320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:12:28,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5017, 2.1440, 1.5393, 0.7663], device='cuda:0'), covar=tensor([0.5633, 0.2830, 0.3822, 0.6111], device='cuda:0'), in_proj_covar=tensor([0.1761, 0.1652, 0.1603, 0.1425], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 07:12:31,721 INFO [train.py:968] (0/2) Epoch 22, batch 7600, giga_loss[loss=0.2678, simple_loss=0.3487, pruned_loss=0.09349, over 29003.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3681, pruned_loss=0.1154, over 5661266.69 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08997, over 5639367.63 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3708, pruned_loss=0.1182, over 5656236.48 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:13:00,717 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-11 07:13:01,913 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0908, 2.0823, 1.6677, 1.7308], device='cuda:0'), covar=tensor([0.0971, 0.0766, 0.1004, 0.1195], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0450, 0.0520, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 07:13:03,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.740e+03 2.073e+03 2.851e+03 1.093e+04, threshold=4.146e+03, percent-clipped=9.0
+2023-03-11 07:13:16,133 INFO [train.py:968] (0/2) Epoch 22, batch 7650, giga_loss[loss=0.2606, simple_loss=0.3472, pruned_loss=0.087, over 29002.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3683, pruned_loss=0.1153, over 5674569.36 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3431, pruned_loss=0.08985, over 5647020.54 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3715, pruned_loss=0.1184, over 5664225.11 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:13:57,394 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=965425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:13:59,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=965428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:14:02,446 INFO [train.py:968] (0/2) Epoch 22, batch 7700, giga_loss[loss=0.3429, simple_loss=0.3774, pruned_loss=0.1542, over 23712.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3668, pruned_loss=0.1144, over 5681275.12 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08995, over 5649341.82 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3693, pruned_loss=0.117, over 5671465.93 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:14:29,265 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=965457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:14:36,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.637e+03 2.227e+03 2.991e+03 6.499e+03, threshold=4.454e+03, percent-clipped=7.0
+2023-03-11 07:14:51,759 INFO [train.py:968] (0/2) Epoch 22, batch 7750, libri_loss[loss=0.2438, simple_loss=0.3264, pruned_loss=0.08059, over 29558.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3662, pruned_loss=0.1152, over 5668460.76 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3431, pruned_loss=0.08987, over 5656020.80 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.369, pruned_loss=0.1181, over 5655122.87 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:14:58,039 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=965488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:15:38,407 INFO [train.py:968] (0/2) Epoch 22, batch 7800, libri_loss[loss=0.2609, simple_loss=0.3402, pruned_loss=0.09078, over 29603.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3644, pruned_loss=0.1145, over 5669458.61 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.08991, over 5652060.76 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.367, pruned_loss=0.1173, over 5663444.66 frames. ], batch size: 75, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:15:45,898 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=965539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:16:10,783 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6867, 1.9451, 1.3481, 1.6214], device='cuda:0'), covar=tensor([0.0930, 0.0539, 0.1019, 0.0996], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0450, 0.0520, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 07:16:11,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.708e+03 2.285e+03 3.160e+03 1.229e+04, threshold=4.570e+03, percent-clipped=8.0
+2023-03-11 07:16:21,218 INFO [train.py:968] (0/2) Epoch 22, batch 7850, giga_loss[loss=0.3264, simple_loss=0.3837, pruned_loss=0.1345, over 27957.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3635, pruned_loss=0.1145, over 5669276.10 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3431, pruned_loss=0.08994, over 5663913.87 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3668, pruned_loss=0.118, over 5654015.11 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:16:27,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2223, 2.8424, 1.3763, 1.4366], device='cuda:0'), covar=tensor([0.0993, 0.0368, 0.0862, 0.1280], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0559, 0.0387, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 07:16:43,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6538, 1.5763, 1.9282, 1.4838], device='cuda:0'), covar=tensor([0.1534, 0.2054, 0.1236, 0.1515], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0704, 0.0947, 0.0844], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 07:17:09,382 INFO [train.py:968] (0/2) Epoch 22, batch 7900, giga_loss[loss=0.3059, simple_loss=0.3772, pruned_loss=0.1173, over 28930.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3632, pruned_loss=0.1152, over 5663580.57 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08998, over 5668676.84 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3661, pruned_loss=0.1184, over 5647308.61 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:17:38,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7257, 2.6081, 1.5732, 0.9137], device='cuda:0'), covar=tensor([0.8451, 0.3812, 0.4557, 0.7405], device='cuda:0'), in_proj_covar=tensor([0.1764, 0.1655, 0.1602, 0.1423], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 07:17:42,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.223e+03 1.819e+03 2.557e+03 3.602e+03 9.748e+03, threshold=5.113e+03, percent-clipped=14.0
+2023-03-11 07:17:52,259 INFO [train.py:968] (0/2) Epoch 22, batch 7950, giga_loss[loss=0.3095, simple_loss=0.3699, pruned_loss=0.1245, over 27648.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3628, pruned_loss=0.1151, over 5662565.47 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3434, pruned_loss=0.09001, over 5665666.85 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3656, pruned_loss=0.1184, over 5651149.69 frames. ], batch size: 472, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:17:54,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=965682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:18:26,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3093, 3.4315, 1.4075, 1.5123], device='cuda:0'), covar=tensor([0.1088, 0.0411, 0.0924, 0.1462], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0559, 0.0387, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 07:18:35,630 INFO [train.py:968] (0/2) Epoch 22, batch 8000, giga_loss[loss=0.2496, simple_loss=0.3339, pruned_loss=0.08268, over 28505.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3614, pruned_loss=0.1138, over 5668544.89 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3422, pruned_loss=0.08949, over 5672552.89 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3654, pruned_loss=0.1179, over 5653005.89 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:18:46,610 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 07:19:01,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5551, 2.2807, 1.6515, 0.8034], device='cuda:0'), covar=tensor([0.6222, 0.3112, 0.4047, 0.6624], device='cuda:0'), in_proj_covar=tensor([0.1761, 0.1653, 0.1599, 0.1420], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 07:19:11,063 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.078e+03 1.666e+03 2.032e+03 3.011e+03 6.669e+03, threshold=4.063e+03, percent-clipped=5.0
+2023-03-11 07:19:20,927 INFO [train.py:968] (0/2) Epoch 22, batch 8050, giga_loss[loss=0.3268, simple_loss=0.3883, pruned_loss=0.1327, over 28556.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3619, pruned_loss=0.1132, over 5674522.70 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3421, pruned_loss=0.08945, over 5679504.18 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3658, pruned_loss=0.1172, over 5655565.19 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:19:24,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2355, 1.4754, 1.4912, 1.2894], device='cuda:0'), covar=tensor([0.1723, 0.1517, 0.2134, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0755, 0.0718, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 07:20:01,809 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=965825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:20:04,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=965828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:20:05,753 INFO [train.py:968] (0/2) Epoch 22, batch 8100, giga_loss[loss=0.3661, simple_loss=0.4193, pruned_loss=0.1564, over 28801.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3637, pruned_loss=0.1135, over 5682563.03 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3422, pruned_loss=0.08951, over 5677830.27 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3669, pruned_loss=0.1168, over 5668787.17 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:20:32,410 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=965857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:20:36,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=965863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:20:41,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.538e+02 1.684e+03 2.246e+03 3.604e+03 8.483e+03, threshold=4.491e+03, percent-clipped=16.0
+2023-03-11 07:20:53,685 INFO [train.py:968] (0/2) Epoch 22, batch 8150, giga_loss[loss=0.3014, simple_loss=0.3675, pruned_loss=0.1177, over 28768.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3644, pruned_loss=0.1135, over 5688024.20 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3425, pruned_loss=0.08965, over 5682887.13 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3672, pruned_loss=0.1166, over 5672606.46 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:21:25,343 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=965914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:21:28,604 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-11 07:21:41,020 INFO [train.py:968] (0/2) Epoch 22, batch 8200, giga_loss[loss=0.3644, simple_loss=0.425, pruned_loss=0.1519, over 28886.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3665, pruned_loss=0.1156, over 5679707.82 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3426, pruned_loss=0.08974, over 5677778.22 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.369, pruned_loss=0.1185, over 5671318.20 frames. ], batch size: 285, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:22:18,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.821e+03 2.335e+03 3.209e+03 7.144e+03, threshold=4.669e+03, percent-clipped=9.0
+2023-03-11 07:22:31,725 INFO [train.py:968] (0/2) Epoch 22, batch 8250, giga_loss[loss=0.3461, simple_loss=0.3987, pruned_loss=0.1468, over 28974.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3688, pruned_loss=0.1187, over 5658065.91 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3426, pruned_loss=0.08975, over 5670337.52 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3713, pruned_loss=0.1215, over 5658036.65 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:22:50,389 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-966000.pt
+2023-03-11 07:22:55,753 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=966006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:22:57,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=966009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:23:20,639 INFO [train.py:968] (0/2) Epoch 22, batch 8300, giga_loss[loss=0.3284, simple_loss=0.3842, pruned_loss=0.1363, over 29102.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3705, pruned_loss=0.1217, over 5650311.46 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3427, pruned_loss=0.08976, over 5665406.54 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3733, pruned_loss=0.1249, over 5654120.61 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:23:27,082 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=966038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:23:46,051 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=966057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:23:48,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=966060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:23:56,069 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4408, 1.6666, 1.6955, 1.2687], device='cuda:0'), covar=tensor([0.1756, 0.2530, 0.1457, 0.1645], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0702, 0.0946, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 07:23:57,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.913e+02 1.804e+03 2.648e+03 3.718e+03 9.024e+03, threshold=5.295e+03, percent-clipped=17.0
+2023-03-11 07:24:09,224 INFO [train.py:968] (0/2) Epoch 22, batch 8350, giga_loss[loss=0.4098, simple_loss=0.4449, pruned_loss=0.1873, over 24287.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3728, pruned_loss=0.1243, over 5652568.98 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.343, pruned_loss=0.0899, over 5669945.56 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3756, pruned_loss=0.1277, over 5651320.19 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:24:16,162 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=966089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:24:51,988 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=966125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:24:56,340 INFO [train.py:968] (0/2) Epoch 22, batch 8400, giga_loss[loss=0.3205, simple_loss=0.3746, pruned_loss=0.1332, over 28730.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3733, pruned_loss=0.1246, over 5653351.21 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3433, pruned_loss=0.09006, over 5671159.07 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3757, pruned_loss=0.1278, over 5651252.64 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:25:29,160 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=966167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:25:31,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.607e+03 1.994e+03 2.870e+03 1.319e+04, threshold=3.988e+03, percent-clipped=3.0
+2023-03-11 07:25:38,975 INFO [train.py:968] (0/2) Epoch 22, batch 8450, giga_loss[loss=0.3441, simple_loss=0.4049, pruned_loss=0.1416, over 27852.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3707, pruned_loss=0.1219, over 5668683.28 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.08989, over 5674503.50 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3732, pruned_loss=0.1251, over 5663698.23 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:26:17,164 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 07:26:21,492 INFO [train.py:968] (0/2) Epoch 22, batch 8500, giga_loss[loss=0.2552, simple_loss=0.3404, pruned_loss=0.08502, over 29033.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3688, pruned_loss=0.1187, over 5685617.54 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3427, pruned_loss=0.08968, over 5681339.66 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.372, pruned_loss=0.1222, over 5675619.51 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:26:53,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.700e+03 2.188e+03 3.017e+03 6.154e+03, threshold=4.376e+03, percent-clipped=12.0
+2023-03-11 07:27:02,278 INFO [train.py:968] (0/2) Epoch 22, batch 8550, giga_loss[loss=0.3084, simple_loss=0.3703, pruned_loss=0.1232, over 28939.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3672, pruned_loss=0.1171, over 5676306.60 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3429, pruned_loss=0.08987, over 5679974.88 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3705, pruned_loss=0.1209, over 5669888.12 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:27:48,105 INFO [train.py:968] (0/2) Epoch 22, batch 8600, giga_loss[loss=0.3105, simple_loss=0.3706, pruned_loss=0.1252, over 28893.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3664, pruned_loss=0.1173, over 5670467.14 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.08993, over 5681442.50 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3694, pruned_loss=0.121, over 5663924.12 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:28:22,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.701e+03 2.019e+03 2.888e+03 1.260e+04, threshold=4.037e+03, percent-clipped=9.0
+2023-03-11 07:28:31,631 INFO [train.py:968] (0/2) Epoch 22, batch 8650, giga_loss[loss=0.3286, simple_loss=0.3852, pruned_loss=0.136, over 28683.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3657, pruned_loss=0.1175, over 5680277.52 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3433, pruned_loss=0.09004, over 5686383.23 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3687, pruned_loss=0.1211, over 5670060.17 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:29:00,672 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-11 07:29:18,658 INFO [train.py:968] (0/2) Epoch 22, batch 8700, giga_loss[loss=0.3149, simple_loss=0.3885, pruned_loss=0.1207, over 28797.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3664, pruned_loss=0.1179, over 5682974.87 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3429, pruned_loss=0.08981, over 5694439.61 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1222, over 5666678.19 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:29:53,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.794e+03 2.169e+03 2.944e+03 7.772e+03, threshold=4.338e+03, percent-clipped=6.0
+2023-03-11 07:29:59,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0647, 5.8585, 5.5521, 3.1701], device='cuda:0'), covar=tensor([0.0537, 0.0738, 0.0950, 0.1561], device='cuda:0'), in_proj_covar=tensor([0.1245, 0.1153, 0.0976, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 07:30:01,691 INFO [train.py:968] (0/2) Epoch 22, batch 8750, giga_loss[loss=0.2914, simple_loss=0.379, pruned_loss=0.1019, over 28958.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3675, pruned_loss=0.1179, over 5687849.33 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3423, pruned_loss=0.08949, over 5703786.65 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3723, pruned_loss=0.123, over 5665532.29 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:30:05,646 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3012, 3.1511, 1.4388, 1.5037], device='cuda:0'), covar=tensor([0.1020, 0.0510, 0.0951, 0.1364], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0557, 0.0387, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 07:30:19,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=966500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:30:20,282 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=966502.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 07:30:42,671 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3565, 1.7426, 1.5948, 1.5238], device='cuda:0'), covar=tensor([0.2026, 0.1942, 0.2263, 0.2142], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0751, 0.0714, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 07:30:47,012 INFO [train.py:968] (0/2) Epoch 22, batch 8800, giga_loss[loss=0.3405, simple_loss=0.4062, pruned_loss=0.1374, over 27901.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3713, pruned_loss=0.1179, over 5689442.33 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3422, pruned_loss=0.08955, over 5707393.85 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3758, pruned_loss=0.1225, over 5668013.66 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:30:50,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 07:30:59,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=966542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:31:26,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.491e+02 1.621e+03 2.341e+03 3.351e+03 9.761e+03, threshold=4.682e+03, percent-clipped=11.0
+2023-03-11 07:31:36,627 INFO [train.py:968] (0/2) Epoch 22, batch 8850, giga_loss[loss=0.2987, simple_loss=0.3724, pruned_loss=0.1125, over 28920.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3738, pruned_loss=0.1188, over 5692775.78 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3425, pruned_loss=0.08961, over 5709547.17 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3775, pruned_loss=0.1229, over 5673679.42 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:32:17,736 INFO [train.py:968] (0/2) Epoch 22, batch 8900, giga_loss[loss=0.3392, simple_loss=0.393, pruned_loss=0.1427, over 28944.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3741, pruned_loss=0.1194, over 5697367.04 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3416, pruned_loss=0.08912, over 5715691.75 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3788, pruned_loss=0.1241, over 5675945.95 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:32:21,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0935, 3.3140, 1.2300, 1.5540], device='cuda:0'), covar=tensor([0.1270, 0.0436, 0.1001, 0.1476], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0557, 0.0387, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 07:32:29,725 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=966643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:32:33,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=966646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:32:57,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.380e+02 1.765e+03 2.062e+03 2.936e+03 8.719e+03, threshold=4.125e+03, percent-clipped=4.0
+2023-03-11 07:33:00,680 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=966675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:33:03,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4310, 1.6456, 1.5638, 1.2882], device='cuda:0'), covar=tensor([0.2910, 0.2636, 0.1861, 0.2546], device='cuda:0'), in_proj_covar=tensor([0.1964, 0.1912, 0.1835, 0.1965], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 07:33:05,618 INFO [train.py:968] (0/2) Epoch 22, batch 8950, giga_loss[loss=0.2713, simple_loss=0.3429, pruned_loss=0.09985, over 28546.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3745, pruned_loss=0.1206, over 5696963.63 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3418, pruned_loss=0.08924, over 5717584.39 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3786, pruned_loss=0.1246, over 5677948.63 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:33:09,902 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=966685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:33:12,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=966688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:33:39,454 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=966717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:33:54,507 INFO [train.py:968] (0/2) Epoch 22, batch 9000, giga_loss[loss=0.3271, simple_loss=0.3882, pruned_loss=0.1329, over 28714.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3734, pruned_loss=0.1205, over 5696221.83 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3417, pruned_loss=0.08909, over 5722146.11 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3776, pruned_loss=0.1245, over 5676549.97 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:33:54,511 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 07:34:00,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3607, 1.2982, 1.1558, 1.5145], device='cuda:0'), covar=tensor([0.0863, 0.0385, 0.0385, 0.0962], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:0')
+2023-03-11 07:34:03,091 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2063, simple_loss=0.314, pruned_loss=0.04931, over 944034.00 frames. 
+2023-03-11 07:34:03,091 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 07:34:48,579 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.740e+02 1.630e+03 2.077e+03 2.812e+03 1.036e+04, threshold=4.155e+03, percent-clipped=8.0
+2023-03-11 07:34:56,972 INFO [train.py:968] (0/2) Epoch 22, batch 9050, giga_loss[loss=0.2456, simple_loss=0.3279, pruned_loss=0.08166, over 28902.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3718, pruned_loss=0.1204, over 5692590.30 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3417, pruned_loss=0.08909, over 5722146.11 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.375, pruned_loss=0.1235, over 5677279.44 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:35:16,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7112, 1.7885, 1.6896, 1.6398], device='cuda:0'), covar=tensor([0.1870, 0.2423, 0.2497, 0.2196], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0756, 0.0718, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 07:35:36,959 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-11 07:35:46,112 INFO [train.py:968] (0/2) Epoch 22, batch 9100, giga_loss[loss=0.2947, simple_loss=0.3723, pruned_loss=0.1086, over 28963.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3707, pruned_loss=0.1207, over 5687704.64 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3421, pruned_loss=0.08935, over 5725262.98 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3734, pruned_loss=0.1236, over 5671829.87 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:36:24,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.629e+03 2.094e+03 2.878e+03 7.782e+03, threshold=4.188e+03, percent-clipped=10.0
+2023-03-11 07:36:29,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=966877.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 07:36:33,325 INFO [train.py:968] (0/2) Epoch 22, batch 9150, giga_loss[loss=0.3094, simple_loss=0.3685, pruned_loss=0.1251, over 28649.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3706, pruned_loss=0.121, over 5685771.85 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3426, pruned_loss=0.0895, over 5729271.32 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.373, pruned_loss=0.1239, over 5668693.05 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:36:51,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3233, 3.0986, 1.5005, 1.4127], device='cuda:0'), covar=tensor([0.0975, 0.0337, 0.0873, 0.1367], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0558, 0.0387, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 07:37:26,191 INFO [train.py:968] (0/2) Epoch 22, batch 9200, giga_loss[loss=0.3438, simple_loss=0.3982, pruned_loss=0.1447, over 28250.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3714, pruned_loss=0.122, over 5685797.32 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3425, pruned_loss=0.0894, over 5732099.69 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3737, pruned_loss=0.1248, over 5669247.55 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:37:42,596 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2196, 0.8150, 0.9688, 1.4274], device='cuda:0'), covar=tensor([0.0775, 0.0378, 0.0345, 0.0813], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 07:38:03,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.872e+03 2.369e+03 3.188e+03 8.171e+03, threshold=4.738e+03, percent-clipped=11.0
+2023-03-11 07:38:13,593 INFO [train.py:968] (0/2) Epoch 22, batch 9250, giga_loss[loss=0.2801, simple_loss=0.3497, pruned_loss=0.1053, over 28649.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.369, pruned_loss=0.1207, over 5684073.48 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3425, pruned_loss=0.08942, over 5734804.87 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3712, pruned_loss=0.1233, over 5668114.92 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:38:13,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8877, 3.7204, 3.5296, 1.9245], device='cuda:0'), covar=tensor([0.0688, 0.0806, 0.0777, 0.2267], device='cuda:0'), in_proj_covar=tensor([0.1247, 0.1153, 0.0976, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 07:38:52,975 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=967020.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 07:38:56,172 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=967023.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 07:39:03,070 INFO [train.py:968] (0/2) Epoch 22, batch 9300, giga_loss[loss=0.272, simple_loss=0.3393, pruned_loss=0.1024, over 28912.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3683, pruned_loss=0.1205, over 5688387.42 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3425, pruned_loss=0.0894, over 5737382.46 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3704, pruned_loss=0.1231, over 5672679.07 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:39:22,114 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=967052.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 07:39:36,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.547e+03 1.921e+03 2.342e+03 6.253e+03, threshold=3.843e+03, percent-clipped=2.0
+2023-03-11 07:39:43,927 INFO [train.py:968] (0/2) Epoch 22, batch 9350, giga_loss[loss=0.3964, simple_loss=0.4272, pruned_loss=0.1828, over 26656.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3687, pruned_loss=0.1199, over 5688115.88 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3425, pruned_loss=0.08937, over 5732494.31 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3712, pruned_loss=0.1229, over 5678917.50 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:40:11,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.2005, 6.0307, 5.7737, 3.0839], device='cuda:0'), covar=tensor([0.0436, 0.0598, 0.0621, 0.1539], device='cuda:0'), in_proj_covar=tensor([0.1246, 0.1153, 0.0977, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 07:40:35,293 INFO [train.py:968] (0/2) Epoch 22, batch 9400, giga_loss[loss=0.2961, simple_loss=0.3692, pruned_loss=0.1115, over 28819.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3699, pruned_loss=0.1205, over 5678816.21 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3422, pruned_loss=0.0892, over 5734951.50 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3726, pruned_loss=0.1236, over 5668485.41 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:40:43,865 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:41:15,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.645e+03 2.020e+03 2.867e+03 7.361e+03, threshold=4.039e+03, percent-clipped=11.0
+2023-03-11 07:41:22,424 INFO [train.py:968] (0/2) Epoch 22, batch 9450, giga_loss[loss=0.3148, simple_loss=0.3743, pruned_loss=0.1277, over 28692.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3693, pruned_loss=0.1204, over 5677450.92 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3422, pruned_loss=0.08914, over 5738178.98 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3721, pruned_loss=0.1235, over 5665129.12 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:41:48,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3602, 3.1581, 1.5412, 1.4712], device='cuda:0'), covar=tensor([0.0979, 0.0369, 0.0860, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0559, 0.0388, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 07:42:06,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5479, 1.7682, 1.7620, 1.3260], device='cuda:0'), covar=tensor([0.1938, 0.2812, 0.1662, 0.1993], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0704, 0.0947, 0.0843], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 07:42:06,975 INFO [train.py:968] (0/2) Epoch 22, batch 9500, libri_loss[loss=0.273, simple_loss=0.3575, pruned_loss=0.09425, over 28695.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3702, pruned_loss=0.1194, over 5685741.61 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.342, pruned_loss=0.08905, over 5741675.86 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3732, pruned_loss=0.1227, over 5671383.97 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:42:39,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:42:47,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7121, 1.9254, 1.5786, 1.7614], device='cuda:0'), covar=tensor([0.2859, 0.2874, 0.3433, 0.2600], device='cuda:0'), in_proj_covar=tensor([0.1502, 0.1085, 0.1327, 0.0991], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 07:42:47,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.121e+02 1.438e+03 1.881e+03 2.844e+03 1.054e+04, threshold=3.762e+03, percent-clipped=8.0
+2023-03-11 07:42:53,487 INFO [train.py:968] (0/2) Epoch 22, batch 9550, giga_loss[loss=0.3136, simple_loss=0.3902, pruned_loss=0.1184, over 28648.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3706, pruned_loss=0.1174, over 5689070.23 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3419, pruned_loss=0.08903, over 5745248.39 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3737, pruned_loss=0.1208, over 5672890.17 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:43:36,119 INFO [train.py:968] (0/2) Epoch 22, batch 9600, giga_loss[loss=0.2776, simple_loss=0.359, pruned_loss=0.09815, over 28986.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.372, pruned_loss=0.1167, over 5678206.90 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3422, pruned_loss=0.08906, over 5737100.63 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.375, pruned_loss=0.1202, over 5671258.32 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:44:13,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3692, 1.4499, 1.2860, 1.4768], device='cuda:0'), covar=tensor([0.0786, 0.0350, 0.0351, 0.0891], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0107], device='cuda:0')
+2023-03-11 07:44:15,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.400e+02 1.496e+03 1.824e+03 2.423e+03 4.231e+03, threshold=3.648e+03, percent-clipped=3.0
+2023-03-11 07:44:23,788 INFO [train.py:968] (0/2) Epoch 22, batch 9650, giga_loss[loss=0.3828, simple_loss=0.4242, pruned_loss=0.1707, over 28633.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3755, pruned_loss=0.1198, over 5685094.40 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.342, pruned_loss=0.08903, over 5741297.65 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3788, pruned_loss=0.1232, over 5674645.70 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:44:37,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2702, 1.3623, 1.2614, 1.4880], device='cuda:0'), covar=tensor([0.0751, 0.0385, 0.0337, 0.0815], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 07:45:08,834 INFO [train.py:968] (0/2) Epoch 22, batch 9700, giga_loss[loss=0.4355, simple_loss=0.4463, pruned_loss=0.2123, over 23359.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3771, pruned_loss=0.1227, over 5681277.05 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3415, pruned_loss=0.08878, over 5744162.37 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3808, pruned_loss=0.1261, over 5669626.70 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:45:10,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3530, 3.6680, 1.5200, 1.5174], device='cuda:0'), covar=tensor([0.0953, 0.0398, 0.0880, 0.1264], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0559, 0.0388, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 07:45:48,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.230e+03 1.707e+03 2.312e+03 2.897e+03 6.650e+03, threshold=4.624e+03, percent-clipped=11.0
+2023-03-11 07:45:54,066 INFO [train.py:968] (0/2) Epoch 22, batch 9750, giga_loss[loss=0.3289, simple_loss=0.3891, pruned_loss=0.1344, over 28715.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3771, pruned_loss=0.1232, over 5677024.20 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3421, pruned_loss=0.08898, over 5743359.82 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3806, pruned_loss=0.1268, over 5666298.37 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:46:22,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=967515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:46:35,898 INFO [train.py:968] (0/2) Epoch 22, batch 9800, giga_loss[loss=0.2985, simple_loss=0.3683, pruned_loss=0.1143, over 28851.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3752, pruned_loss=0.122, over 5653609.95 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.342, pruned_loss=0.08901, over 5731396.91 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3797, pruned_loss=0.1265, over 5652476.02 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:46:40,435 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-11 07:46:47,123 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:46:54,501 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-11 07:47:06,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3048, 1.5066, 1.2385, 1.4947], device='cuda:0'), covar=tensor([0.0720, 0.0399, 0.0346, 0.0795], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 07:47:14,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.595e+03 2.144e+03 3.127e+03 1.061e+04, threshold=4.288e+03, percent-clipped=4.0
+2023-03-11 07:47:20,746 INFO [train.py:968] (0/2) Epoch 22, batch 9850, libri_loss[loss=0.2527, simple_loss=0.3357, pruned_loss=0.08487, over 29521.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3738, pruned_loss=0.1202, over 5655912.37 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08915, over 5734323.47 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.378, pruned_loss=0.1244, over 5650596.85 frames. ], batch size: 80, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:47:49,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5062, 2.2882, 1.6459, 0.7490], device='cuda:0'), covar=tensor([0.7051, 0.3695, 0.4801, 0.7462], device='cuda:0'), in_proj_covar=tensor([0.1761, 0.1657, 0.1598, 0.1422], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 07:47:52,474 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:47:58,523 INFO [train.py:968] (0/2) Epoch 22, batch 9900, giga_loss[loss=0.3179, simple_loss=0.3818, pruned_loss=0.127, over 28600.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3735, pruned_loss=0.1183, over 5673058.12 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3421, pruned_loss=0.08925, over 5741250.88 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3782, pruned_loss=0.1229, over 5659595.19 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:48:02,233 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-11 07:48:06,573 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=967638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:48:18,068 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 07:48:24,086 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=967658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:48:26,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=967661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:48:37,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.414e+03 1.831e+03 2.614e+03 6.780e+03, threshold=3.661e+03, percent-clipped=8.0
+2023-03-11 07:48:44,106 INFO [train.py:968] (0/2) Epoch 22, batch 9950, giga_loss[loss=0.3395, simple_loss=0.396, pruned_loss=0.1415, over 28029.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3747, pruned_loss=0.1184, over 5681826.79 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3424, pruned_loss=0.08944, over 5742900.33 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3788, pruned_loss=0.1224, over 5668339.18 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:48:56,972 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=967690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:49:36,140 INFO [train.py:968] (0/2) Epoch 22, batch 10000, giga_loss[loss=0.3161, simple_loss=0.372, pruned_loss=0.1301, over 27967.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3753, pruned_loss=0.1199, over 5666862.14 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3421, pruned_loss=0.08922, over 5744469.90 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3793, pruned_loss=0.1238, over 5653611.64 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:50:08,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9325, 2.2195, 2.2128, 1.6711], device='cuda:0'), covar=tensor([0.3139, 0.2489, 0.2461, 0.3073], device='cuda:0'), in_proj_covar=tensor([0.1969, 0.1915, 0.1839, 0.1971], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 07:50:18,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.730e+02 1.687e+03 2.480e+03 3.584e+03 6.728e+03, threshold=4.961e+03, percent-clipped=21.0
+2023-03-11 07:50:20,141 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-11 07:50:25,974 INFO [train.py:968] (0/2) Epoch 22, batch 10050, giga_loss[loss=0.292, simple_loss=0.3648, pruned_loss=0.1096, over 28938.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3752, pruned_loss=0.1208, over 5674812.32 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3419, pruned_loss=0.0891, over 5745403.55 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3787, pruned_loss=0.1241, over 5663328.48 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:50:26,317 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=967781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:50:28,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=967784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:51:01,621 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=967813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:51:18,054 INFO [train.py:968] (0/2) Epoch 22, batch 10100, giga_loss[loss=0.2781, simple_loss=0.3533, pruned_loss=0.1014, over 28910.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3744, pruned_loss=0.1217, over 5667297.67 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3415, pruned_loss=0.08888, over 5747121.07 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3777, pruned_loss=0.1248, over 5656264.84 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:52:01,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.058e+02 1.685e+03 2.317e+03 3.334e+03 7.185e+03, threshold=4.635e+03, percent-clipped=5.0
+2023-03-11 07:52:10,482 INFO [train.py:968] (0/2) Epoch 22, batch 10150, giga_loss[loss=0.2497, simple_loss=0.322, pruned_loss=0.08872, over 28757.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3721, pruned_loss=0.1209, over 5668612.82 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3416, pruned_loss=0.0889, over 5748628.72 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.375, pruned_loss=0.1235, over 5657799.84 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:52:48,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2075, 0.8099, 0.9491, 1.3926], device='cuda:0'), covar=tensor([0.0785, 0.0379, 0.0345, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0107], device='cuda:0')
+2023-03-11 07:52:48,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=967918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:52:58,654 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:53:00,428 INFO [train.py:968] (0/2) Epoch 22, batch 10200, giga_loss[loss=0.2703, simple_loss=0.3378, pruned_loss=0.1014, over 28478.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.37, pruned_loss=0.12, over 5674153.35 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3417, pruned_loss=0.08904, over 5751074.17 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3725, pruned_loss=0.1224, over 5662279.48 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:53:14,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967945.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 07:53:42,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.825e+03 2.219e+03 3.271e+03 2.484e+04, threshold=4.437e+03, percent-clipped=11.0
+2023-03-11 07:53:48,633 INFO [train.py:968] (0/2) Epoch 22, batch 10250, giga_loss[loss=0.3371, simple_loss=0.3893, pruned_loss=0.1424, over 27857.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3711, pruned_loss=0.1213, over 5671576.22 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08915, over 5751036.03 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3737, pruned_loss=0.1242, over 5659318.34 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:53:50,237 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1582, 1.3274, 1.3020, 1.0593], device='cuda:0'), covar=tensor([0.2459, 0.2372, 0.1705, 0.2285], device='cuda:0'), in_proj_covar=tensor([0.1974, 0.1918, 0.1846, 0.1980], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 07:53:59,946 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:54:05,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=967999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:54:05,866 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-968000.pt
+2023-03-11 07:54:35,660 INFO [train.py:968] (0/2) Epoch 22, batch 10300, giga_loss[loss=0.3033, simple_loss=0.3779, pruned_loss=0.1143, over 28261.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3693, pruned_loss=0.1197, over 5669210.07 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3426, pruned_loss=0.08947, over 5753123.78 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3714, pruned_loss=0.1221, over 5656648.06 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:54:40,217 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:55:03,224 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:55:05,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:55:17,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.468e+02 1.521e+03 2.044e+03 2.780e+03 9.467e+03, threshold=4.088e+03, percent-clipped=7.0
+2023-03-11 07:55:22,933 INFO [train.py:968] (0/2) Epoch 22, batch 10350, giga_loss[loss=0.2822, simple_loss=0.3543, pruned_loss=0.1051, over 28896.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3654, pruned_loss=0.1154, over 5671660.89 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3426, pruned_loss=0.08947, over 5754003.16 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3671, pruned_loss=0.1174, over 5660670.38 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:55:37,067 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:56:10,242 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:56:10,643 INFO [train.py:968] (0/2) Epoch 22, batch 10400, giga_loss[loss=0.2659, simple_loss=0.3448, pruned_loss=0.09347, over 28883.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3639, pruned_loss=0.1143, over 5658701.24 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3428, pruned_loss=0.08974, over 5747161.23 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3657, pruned_loss=0.1164, over 5653210.17 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:56:20,388 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:56:24,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:56:50,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:56:50,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.104e+03 1.657e+03 2.115e+03 3.086e+03 7.081e+03, threshold=4.231e+03, percent-clipped=13.0
+2023-03-11 07:56:57,356 INFO [train.py:968] (0/2) Epoch 22, batch 10450, giga_loss[loss=0.2759, simple_loss=0.3462, pruned_loss=0.1028, over 29016.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3613, pruned_loss=0.1128, over 5667368.41 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3428, pruned_loss=0.0896, over 5747426.81 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3633, pruned_loss=0.1152, over 5660790.90 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:57:48,812 INFO [train.py:968] (0/2) Epoch 22, batch 10500, giga_loss[loss=0.2873, simple_loss=0.3631, pruned_loss=0.1057, over 28736.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3604, pruned_loss=0.1134, over 5665410.29 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3431, pruned_loss=0.08985, over 5749147.86 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3618, pruned_loss=0.1153, over 5658033.99 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:58:15,407 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 07:58:26,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.834e+03 2.199e+03 3.016e+03 1.317e+04, threshold=4.399e+03, percent-clipped=7.0
+2023-03-11 07:58:33,384 INFO [train.py:968] (0/2) Epoch 22, batch 10550, giga_loss[loss=0.285, simple_loss=0.3621, pruned_loss=0.104, over 28737.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3643, pruned_loss=0.1155, over 5664996.60 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.09017, over 5746679.32 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3651, pruned_loss=0.1173, over 5659227.55 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:58:43,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2766, 1.1193, 3.8834, 3.2339], device='cuda:0'), covar=tensor([0.1687, 0.2880, 0.0458, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0649, 0.0970, 0.0916], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 07:58:51,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:58:52,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5910, 1.7518, 1.4723, 1.8219], device='cuda:0'), covar=tensor([0.2539, 0.2614, 0.2764, 0.2432], device='cuda:0'), in_proj_covar=tensor([0.1509, 0.1093, 0.1329, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 07:59:06,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968320.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 07:59:16,523 INFO [train.py:968] (0/2) Epoch 22, batch 10600, giga_loss[loss=0.2869, simple_loss=0.3531, pruned_loss=0.1104, over 29006.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.366, pruned_loss=0.1159, over 5671031.71 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3439, pruned_loss=0.09001, over 5749636.65 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3672, pruned_loss=0.118, over 5662279.07 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:59:30,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:59:30,768 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:59:56,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:00:03,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.902e+02 1.540e+03 2.056e+03 2.841e+03 4.993e+03, threshold=4.111e+03, percent-clipped=2.0
+2023-03-11 08:00:10,351 INFO [train.py:968] (0/2) Epoch 22, batch 10650, giga_loss[loss=0.3389, simple_loss=0.3962, pruned_loss=0.1408, over 27934.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3676, pruned_loss=0.1178, over 5642960.96 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3444, pruned_loss=0.09028, over 5751114.28 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3683, pruned_loss=0.1195, over 5634018.17 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:00:40,630 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:00:40,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:00:57,210 INFO [train.py:968] (0/2) Epoch 22, batch 10700, giga_loss[loss=0.3, simple_loss=0.3652, pruned_loss=0.1174, over 28591.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3681, pruned_loss=0.1186, over 5641962.81 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3447, pruned_loss=0.09045, over 5753021.74 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3687, pruned_loss=0.1202, over 5631481.96 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:01:11,429 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:01:13,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:01:27,064 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968463.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 08:01:30,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968466.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 08:01:39,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.609e+02 1.610e+03 1.962e+03 2.930e+03 9.018e+03, threshold=3.923e+03, percent-clipped=12.0
+2023-03-11 08:01:42,755 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:01:44,542 INFO [train.py:968] (0/2) Epoch 22, batch 10750, giga_loss[loss=0.2826, simple_loss=0.3562, pruned_loss=0.1045, over 28982.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3695, pruned_loss=0.12, over 5642193.35 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.345, pruned_loss=0.09061, over 5748691.78 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3702, pruned_loss=0.1217, over 5635738.79 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:01:58,497 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968495.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 08:02:06,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:15,573 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:17,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:30,071 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:31,128 INFO [train.py:968] (0/2) Epoch 22, batch 10800, libri_loss[loss=0.3, simple_loss=0.3836, pruned_loss=0.1082, over 29645.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.371, pruned_loss=0.1202, over 5651763.68 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3456, pruned_loss=0.09085, over 5749964.15 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3719, pruned_loss=0.1225, over 5641454.71 frames. ], batch size: 91, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:02:34,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4724, 1.6372, 1.6478, 1.5528], device='cuda:0'), covar=tensor([0.1640, 0.1703, 0.1736, 0.1685], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0755, 0.0719, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 08:02:43,191 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:54,135 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:56,165 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:03:09,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2686, 1.5114, 1.4945, 1.1715], device='cuda:0'), covar=tensor([0.2691, 0.2594, 0.1567, 0.2309], device='cuda:0'), in_proj_covar=tensor([0.1967, 0.1912, 0.1838, 0.1976], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 08:03:13,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.532e+02 1.885e+03 2.548e+03 3.772e+03 6.684e+03, threshold=5.096e+03, percent-clipped=22.0
+2023-03-11 08:03:18,219 INFO [train.py:968] (0/2) Epoch 22, batch 10850, giga_loss[loss=0.3162, simple_loss=0.3753, pruned_loss=0.1285, over 28564.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3711, pruned_loss=0.1203, over 5653344.75 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3455, pruned_loss=0.09082, over 5749256.51 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3724, pruned_loss=0.1227, over 5644245.98 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:03:23,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:04:06,575 INFO [train.py:968] (0/2) Epoch 22, batch 10900, libri_loss[loss=0.2715, simple_loss=0.3572, pruned_loss=0.09288, over 29272.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3718, pruned_loss=0.121, over 5653988.48 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3456, pruned_loss=0.09077, over 5749435.91 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3733, pruned_loss=0.1236, over 5644133.99 frames. ], batch size: 94, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:04:23,583 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:04:25,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:04:50,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.150e+03 1.711e+03 2.120e+03 2.753e+03 6.785e+03, threshold=4.240e+03, percent-clipped=5.0
+2023-03-11 08:04:53,738 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:04:54,099 INFO [train.py:968] (0/2) Epoch 22, batch 10950, giga_loss[loss=0.2852, simple_loss=0.3698, pruned_loss=0.1003, over 28634.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3732, pruned_loss=0.1221, over 5654919.14 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3456, pruned_loss=0.09086, over 5751554.32 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3749, pruned_loss=0.1247, over 5643379.52 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:05:00,699 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3479, 1.3207, 3.5874, 3.1876], device='cuda:0'), covar=tensor([0.1566, 0.2738, 0.0499, 0.1342], device='cuda:0'), in_proj_covar=tensor([0.0766, 0.0651, 0.0968, 0.0915], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 08:05:33,030 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:05:34,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:05:47,571 INFO [train.py:968] (0/2) Epoch 22, batch 11000, giga_loss[loss=0.2997, simple_loss=0.3672, pruned_loss=0.1161, over 28693.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3736, pruned_loss=0.1215, over 5652399.14 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3456, pruned_loss=0.09096, over 5753491.84 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3753, pruned_loss=0.1238, over 5640283.34 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:06:30,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.749e+03 2.314e+03 4.095e+03 9.376e+03, threshold=4.627e+03, percent-clipped=22.0
+2023-03-11 08:06:35,721 INFO [train.py:968] (0/2) Epoch 22, batch 11050, giga_loss[loss=0.297, simple_loss=0.3664, pruned_loss=0.1138, over 28745.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3722, pruned_loss=0.1211, over 5656820.81 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3454, pruned_loss=0.09092, over 5756340.17 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3744, pruned_loss=0.1237, over 5642442.61 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:06:41,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:06:43,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4753, 1.1471, 4.5170, 3.2973], device='cuda:0'), covar=tensor([0.1689, 0.2925, 0.0436, 0.1141], device='cuda:0'), in_proj_covar=tensor([0.0766, 0.0652, 0.0969, 0.0916], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 08:06:45,243 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:07:24,253 INFO [train.py:968] (0/2) Epoch 22, batch 11100, libri_loss[loss=0.2833, simple_loss=0.3644, pruned_loss=0.1011, over 27883.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3698, pruned_loss=0.1195, over 5668573.37 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3455, pruned_loss=0.09098, over 5757337.06 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3722, pruned_loss=0.1225, over 5653143.57 frames. ], batch size: 116, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:07:55,823 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:07:56,465 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:07:58,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:07:59,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:08:15,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.203e+03 1.778e+03 2.458e+03 3.396e+03 9.068e+03, threshold=4.916e+03, percent-clipped=11.0
+2023-03-11 08:08:17,777 INFO [train.py:968] (0/2) Epoch 22, batch 11150, giga_loss[loss=0.3234, simple_loss=0.3852, pruned_loss=0.1308, over 28734.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3701, pruned_loss=0.1203, over 5664709.10 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3457, pruned_loss=0.09105, over 5762302.94 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3728, pruned_loss=0.1237, over 5644511.91 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:08:27,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:08:28,632 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:08:29,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4270, 1.5185, 1.5225, 1.3688], device='cuda:0'), covar=tensor([0.2556, 0.2293, 0.1807, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.1970, 0.1913, 0.1835, 0.1976], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 08:08:39,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:08:42,509 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:08:46,549 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-11 08:09:02,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:09:04,747 INFO [train.py:968] (0/2) Epoch 22, batch 11200, giga_loss[loss=0.2626, simple_loss=0.3356, pruned_loss=0.0948, over 28292.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3688, pruned_loss=0.1196, over 5677553.69 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3457, pruned_loss=0.09102, over 5764073.93 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3713, pruned_loss=0.1228, over 5658898.19 frames. ], batch size: 65, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:09:05,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:09:31,923 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:09:42,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2134, 1.2407, 3.7348, 3.0886], device='cuda:0'), covar=tensor([0.1746, 0.2823, 0.0477, 0.1186], device='cuda:0'), in_proj_covar=tensor([0.0767, 0.0652, 0.0969, 0.0917], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 08:09:44,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.689e+03 2.275e+03 3.268e+03 9.753e+03, threshold=4.550e+03, percent-clipped=8.0
+2023-03-11 08:09:49,240 INFO [train.py:968] (0/2) Epoch 22, batch 11250, giga_loss[loss=0.3034, simple_loss=0.3743, pruned_loss=0.1163, over 28490.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3692, pruned_loss=0.1206, over 5669890.83 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3457, pruned_loss=0.09105, over 5757367.90 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3718, pruned_loss=0.1237, over 5659594.77 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:09:51,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8674, 1.0588, 0.8658, 0.2190], device='cuda:0'), covar=tensor([0.3194, 0.2470, 0.2590, 0.5511], device='cuda:0'), in_proj_covar=tensor([0.1760, 0.1653, 0.1594, 0.1427], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 08:09:55,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5430, 1.6361, 1.6740, 1.4055], device='cuda:0'), covar=tensor([0.2743, 0.2608, 0.2077, 0.2535], device='cuda:0'), in_proj_covar=tensor([0.1967, 0.1910, 0.1833, 0.1975], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 08:10:29,690 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-11 08:10:35,485 INFO [train.py:968] (0/2) Epoch 22, batch 11300, libri_loss[loss=0.243, simple_loss=0.3284, pruned_loss=0.07885, over 29543.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3692, pruned_loss=0.1206, over 5666603.52 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.346, pruned_loss=0.09118, over 5751713.27 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3716, pruned_loss=0.1238, over 5660057.13 frames. ], batch size: 77, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:10:51,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=969047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:10:54,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=969050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:11:16,108 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 08:11:18,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.492e+02 1.731e+03 2.086e+03 2.842e+03 6.696e+03, threshold=4.172e+03, percent-clipped=7.0
+2023-03-11 08:11:20,065 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=969079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:11:22,692 INFO [train.py:968] (0/2) Epoch 22, batch 11350, giga_loss[loss=0.3274, simple_loss=0.3748, pruned_loss=0.1399, over 28709.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3697, pruned_loss=0.1212, over 5669185.98 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3459, pruned_loss=0.09118, over 5754865.38 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3721, pruned_loss=0.1243, over 5659710.99 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:11:30,887 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 08:11:40,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 08:11:45,993 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-11 08:11:56,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5229, 2.2005, 1.6777, 0.7262], device='cuda:0'), covar=tensor([0.5738, 0.3030, 0.4078, 0.6449], device='cuda:0'), in_proj_covar=tensor([0.1760, 0.1655, 0.1596, 0.1427], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 08:12:02,452 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8078, 2.7349, 1.7759, 0.9767], device='cuda:0'), covar=tensor([0.7832, 0.3477, 0.3973, 0.6934], device='cuda:0'), in_proj_covar=tensor([0.1760, 0.1655, 0.1597, 0.1427], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 08:12:08,293 INFO [train.py:968] (0/2) Epoch 22, batch 11400, giga_loss[loss=0.3269, simple_loss=0.3844, pruned_loss=0.1347, over 28677.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3716, pruned_loss=0.1227, over 5674756.84 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3459, pruned_loss=0.09127, over 5758199.62 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3743, pruned_loss=0.126, over 5661658.30 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:12:40,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:12:51,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.779e+03 2.555e+03 3.459e+03 7.560e+03, threshold=5.110e+03, percent-clipped=15.0
+2023-03-11 08:12:54,154 INFO [train.py:968] (0/2) Epoch 22, batch 11450, giga_loss[loss=0.3138, simple_loss=0.3753, pruned_loss=0.1262, over 28763.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1223, over 5679814.10 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3462, pruned_loss=0.09152, over 5762254.98 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3746, pruned_loss=0.1259, over 5662211.14 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:13:44,420 INFO [train.py:968] (0/2) Epoch 22, batch 11500, libri_loss[loss=0.2332, simple_loss=0.3214, pruned_loss=0.07249, over 29565.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3721, pruned_loss=0.1237, over 5663653.20 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3462, pruned_loss=0.09149, over 5753614.03 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3747, pruned_loss=0.127, over 5656462.31 frames. ], batch size: 77, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:14:18,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4659, 1.7680, 1.4082, 1.5266], device='cuda:0'), covar=tensor([0.2464, 0.2517, 0.2874, 0.2333], device='cuda:0'), in_proj_covar=tensor([0.1507, 0.1091, 0.1329, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 08:14:28,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.953e+03 2.767e+03 4.693e+03 1.439e+04, threshold=5.533e+03, percent-clipped=17.0
+2023-03-11 08:14:32,098 INFO [train.py:968] (0/2) Epoch 22, batch 11550, giga_loss[loss=0.3034, simple_loss=0.3694, pruned_loss=0.1187, over 28897.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.1241, over 5661594.62 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3461, pruned_loss=0.09144, over 5754923.31 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3747, pruned_loss=0.127, over 5653776.90 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:14:35,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:15:02,136 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=969308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:15:04,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=969311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:15:20,435 INFO [train.py:968] (0/2) Epoch 22, batch 11600, giga_loss[loss=0.3121, simple_loss=0.3733, pruned_loss=0.1254, over 29103.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3715, pruned_loss=0.1227, over 5674389.50 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.346, pruned_loss=0.09137, over 5757447.07 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3744, pruned_loss=0.1263, over 5663092.87 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:15:20,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 08:15:30,931 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=969340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:15:39,090 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:16:07,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.788e+03 2.487e+03 3.457e+03 8.643e+03, threshold=4.974e+03, percent-clipped=8.0
+2023-03-11 08:16:08,579 INFO [train.py:968] (0/2) Epoch 22, batch 11650, giga_loss[loss=0.2654, simple_loss=0.3395, pruned_loss=0.09562, over 28722.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3709, pruned_loss=0.1218, over 5671013.30 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3457, pruned_loss=0.09115, over 5758973.87 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3737, pruned_loss=0.1251, over 5659991.32 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:16:53,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:16:54,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4541, 1.6965, 1.5655, 1.6338], device='cuda:0'), covar=tensor([0.0802, 0.0322, 0.0310, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0108], device='cuda:0')
+2023-03-11 08:16:54,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8162, 1.9591, 1.3947, 1.4939], device='cuda:0'), covar=tensor([0.0995, 0.0684, 0.1119, 0.1185], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0446, 0.0516, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 08:16:58,034 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=969427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:17:00,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=969430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:17:00,682 INFO [train.py:968] (0/2) Epoch 22, batch 11700, giga_loss[loss=0.3195, simple_loss=0.3802, pruned_loss=0.1294, over 28704.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3724, pruned_loss=0.1229, over 5680555.38 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3457, pruned_loss=0.0911, over 5759421.79 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3748, pruned_loss=0.1258, over 5670697.30 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:17:16,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2836, 1.9445, 1.4745, 0.4974], device='cuda:0'), covar=tensor([0.5062, 0.2900, 0.3703, 0.6028], device='cuda:0'), in_proj_covar=tensor([0.1762, 0.1659, 0.1598, 0.1428], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 08:17:29,691 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=969459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:17:48,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.636e+03 2.070e+03 2.609e+03 6.273e+03, threshold=4.140e+03, percent-clipped=2.0
+2023-03-11 08:17:49,757 INFO [train.py:968] (0/2) Epoch 22, batch 11750, giga_loss[loss=0.3559, simple_loss=0.4044, pruned_loss=0.1536, over 28684.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3757, pruned_loss=0.1262, over 5668241.49 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3454, pruned_loss=0.09095, over 5752709.01 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3783, pruned_loss=0.1292, over 5665345.77 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:18:02,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:18:23,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4542, 1.7282, 1.6765, 1.2599], device='cuda:0'), covar=tensor([0.1972, 0.2856, 0.1701, 0.2007], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0707, 0.0949, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 08:18:34,864 INFO [train.py:968] (0/2) Epoch 22, batch 11800, libri_loss[loss=0.2459, simple_loss=0.3304, pruned_loss=0.08068, over 29331.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3741, pruned_loss=0.1245, over 5682245.92 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3451, pruned_loss=0.09084, over 5756719.74 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3773, pruned_loss=0.1279, over 5674199.47 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:19:16,604 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.158e+03 1.592e+03 1.956e+03 2.843e+03 5.747e+03, threshold=3.911e+03, percent-clipped=6.0
+2023-03-11 08:19:20,691 INFO [train.py:968] (0/2) Epoch 22, batch 11850, libri_loss[loss=0.3146, simple_loss=0.3876, pruned_loss=0.1208, over 25807.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3737, pruned_loss=0.1231, over 5680800.90 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3455, pruned_loss=0.09112, over 5758917.75 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3769, pruned_loss=0.1267, over 5670374.50 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:19:41,825 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:20:10,152 INFO [train.py:968] (0/2) Epoch 22, batch 11900, giga_loss[loss=0.3395, simple_loss=0.3993, pruned_loss=0.1399, over 28343.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3735, pruned_loss=0.1225, over 5665447.18 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3454, pruned_loss=0.09106, over 5755931.06 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3763, pruned_loss=0.1256, over 5658969.76 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:20:17,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3953, 3.8198, 1.6447, 1.6809], device='cuda:0'), covar=tensor([0.1071, 0.0375, 0.0896, 0.1378], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0559, 0.0388, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 08:20:41,531 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-11 08:20:45,411 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5893, 1.8808, 1.4945, 1.8299], device='cuda:0'), covar=tensor([0.2561, 0.2603, 0.2819, 0.2453], device='cuda:0'), in_proj_covar=tensor([0.1512, 0.1092, 0.1332, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 08:20:48,667 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 08:20:53,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.342e+02 1.539e+03 1.983e+03 2.536e+03 5.288e+03, threshold=3.967e+03, percent-clipped=6.0
+2023-03-11 08:20:55,144 INFO [train.py:968] (0/2) Epoch 22, batch 11950, giga_loss[loss=0.2947, simple_loss=0.3645, pruned_loss=0.1124, over 28610.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3708, pruned_loss=0.1201, over 5666513.24 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3455, pruned_loss=0.09118, over 5746054.08 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3734, pruned_loss=0.1231, over 5668220.76 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:21:34,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:21:39,702 INFO [train.py:968] (0/2) Epoch 22, batch 12000, giga_loss[loss=0.3198, simple_loss=0.3804, pruned_loss=0.1296, over 28599.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.37, pruned_loss=0.1197, over 5667280.48 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3456, pruned_loss=0.09116, over 5748632.08 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1226, over 5665141.13 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:21:39,706 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 08:21:48,238 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2088, simple_loss=0.3166, pruned_loss=0.05047, over 944034.00 frames. 
+2023-03-11 08:21:48,239 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 08:22:10,883 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:22:32,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.989e+02 1.638e+03 2.367e+03 3.247e+03 1.480e+04, threshold=4.734e+03, percent-clipped=14.0
+2023-03-11 08:22:34,488 INFO [train.py:968] (0/2) Epoch 22, batch 12050, giga_loss[loss=0.3067, simple_loss=0.3738, pruned_loss=0.1198, over 28893.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3714, pruned_loss=0.1207, over 5666005.27 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3453, pruned_loss=0.09102, over 5750261.69 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3738, pruned_loss=0.1234, over 5662267.21 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:22:34,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2134, 1.3924, 1.2778, 1.1040], device='cuda:0'), covar=tensor([0.1994, 0.2058, 0.1599, 0.2007], device='cuda:0'), in_proj_covar=tensor([0.1967, 0.1911, 0.1834, 0.1972], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 08:22:51,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:22:56,884 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:23:18,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:23:21,418 INFO [train.py:968] (0/2) Epoch 22, batch 12100, giga_loss[loss=0.3463, simple_loss=0.396, pruned_loss=0.1483, over 27958.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3723, pruned_loss=0.1214, over 5658754.45 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3458, pruned_loss=0.09119, over 5742225.08 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3744, pruned_loss=0.1242, over 5661046.94 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:24:02,119 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:24:02,220 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=969869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:24:02,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:24:04,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=969872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:24:10,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.316e+02 1.587e+03 2.023e+03 2.813e+03 7.396e+03, threshold=4.046e+03, percent-clipped=6.0
+2023-03-11 08:24:11,743 INFO [train.py:968] (0/2) Epoch 22, batch 12150, giga_loss[loss=0.316, simple_loss=0.3784, pruned_loss=0.1268, over 29027.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3719, pruned_loss=0.1218, over 5661958.14 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3464, pruned_loss=0.09155, over 5744435.36 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3733, pruned_loss=0.124, over 5660754.09 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:24:17,898 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2843, 1.7349, 1.4257, 1.5142], device='cuda:0'), covar=tensor([0.0732, 0.0353, 0.0320, 0.0812], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 08:24:32,835 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=969901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:24:58,884 INFO [train.py:968] (0/2) Epoch 22, batch 12200, giga_loss[loss=0.3151, simple_loss=0.3716, pruned_loss=0.1293, over 28810.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3714, pruned_loss=0.1217, over 5666015.67 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3463, pruned_loss=0.09152, over 5746367.67 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3733, pruned_loss=0.1242, over 5661454.28 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:25:05,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2434, 1.5116, 1.3272, 1.0962], device='cuda:0'), covar=tensor([0.2118, 0.2033, 0.1568, 0.2038], device='cuda:0'), in_proj_covar=tensor([0.1965, 0.1910, 0.1832, 0.1968], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 08:25:08,639 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=969942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:25:11,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=969945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:25:41,815 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=969974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:25:45,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.478e+02 1.610e+03 2.062e+03 2.771e+03 5.450e+03, threshold=4.124e+03, percent-clipped=7.0
+2023-03-11 08:25:46,749 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:25:47,134 INFO [train.py:968] (0/2) Epoch 22, batch 12250, giga_loss[loss=0.2981, simple_loss=0.3662, pruned_loss=0.115, over 28919.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3739, pruned_loss=0.124, over 5655896.32 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.346, pruned_loss=0.09128, over 5746615.59 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.376, pruned_loss=0.1266, over 5651164.00 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:25:47,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:26:04,921 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-970000.pt
+2023-03-11 08:26:17,560 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:26:21,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:26:34,715 INFO [train.py:968] (0/2) Epoch 22, batch 12300, giga_loss[loss=0.3473, simple_loss=0.3787, pruned_loss=0.158, over 23599.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.374, pruned_loss=0.1242, over 5651560.05 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.346, pruned_loss=0.09124, over 5747397.46 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3759, pruned_loss=0.1265, over 5646515.57 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:26:47,789 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:26:51,117 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-11 08:27:21,194 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.570e+03 1.997e+03 2.940e+03 7.968e+03, threshold=3.994e+03, percent-clipped=10.0
+2023-03-11 08:27:22,527 INFO [train.py:968] (0/2) Epoch 22, batch 12350, giga_loss[loss=0.2777, simple_loss=0.3561, pruned_loss=0.09964, over 28486.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3741, pruned_loss=0.1243, over 5641822.11 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3462, pruned_loss=0.09132, over 5747798.58 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3761, pruned_loss=0.127, over 5634654.16 frames. ], batch size: 65, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:28:02,978 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:28:06,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:28:09,244 INFO [train.py:968] (0/2) Epoch 22, batch 12400, giga_loss[loss=0.275, simple_loss=0.3422, pruned_loss=0.1039, over 28205.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3737, pruned_loss=0.1235, over 5645392.10 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3462, pruned_loss=0.0912, over 5750307.73 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3759, pruned_loss=0.1264, over 5635718.84 frames. ], batch size: 77, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:28:10,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:28:30,626 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:28:37,316 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.95 vs. limit=2.0
+2023-03-11 08:28:52,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.572e+03 2.161e+03 2.952e+03 7.781e+03, threshold=4.322e+03, percent-clipped=7.0
+2023-03-11 08:28:53,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:28:54,014 INFO [train.py:968] (0/2) Epoch 22, batch 12450, giga_loss[loss=0.3452, simple_loss=0.4009, pruned_loss=0.1447, over 28971.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3742, pruned_loss=0.1239, over 5652568.34 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3463, pruned_loss=0.0913, over 5752141.99 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3762, pruned_loss=0.1266, over 5641945.91 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:29:15,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:29:32,707 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=970218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:29:43,323 INFO [train.py:968] (0/2) Epoch 22, batch 12500, giga_loss[loss=0.315, simple_loss=0.3775, pruned_loss=0.1263, over 28657.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3719, pruned_loss=0.1222, over 5659709.28 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3462, pruned_loss=0.09136, over 5751212.95 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3739, pruned_loss=0.1246, over 5650671.70 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:29:57,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:29:57,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3456, 1.5597, 1.4547, 1.2744], device='cuda:0'), covar=tensor([0.2769, 0.2422, 0.2042, 0.2458], device='cuda:0'), in_proj_covar=tensor([0.1974, 0.1914, 0.1836, 0.1975], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 08:29:59,951 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=970248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:30:24,205 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:30:27,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:30:30,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.748e+03 2.418e+03 3.156e+03 1.056e+04, threshold=4.836e+03, percent-clipped=6.0
+2023-03-11 08:30:30,844 INFO [train.py:968] (0/2) Epoch 22, batch 12550, giga_loss[loss=0.2926, simple_loss=0.3548, pruned_loss=0.1152, over 28933.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.369, pruned_loss=0.1202, over 5670248.58 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3464, pruned_loss=0.09162, over 5756298.47 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3713, pruned_loss=0.1229, over 5655298.22 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:30:40,836 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=970293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:30:52,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:07,078 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:10,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:14,151 INFO [train.py:968] (0/2) Epoch 22, batch 12600, giga_loss[loss=0.3213, simple_loss=0.3786, pruned_loss=0.132, over 28670.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3656, pruned_loss=0.1183, over 5676967.33 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3463, pruned_loss=0.09154, over 5758545.60 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3679, pruned_loss=0.1211, over 5661551.52 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:31:20,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=970336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:27,272 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:30,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:36,888 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:37,424 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:32:00,297 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:32:03,114 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.226e+03 1.769e+03 2.360e+03 3.230e+03 9.048e+03, threshold=4.721e+03, percent-clipped=8.0
+2023-03-11 08:32:03,885 INFO [train.py:968] (0/2) Epoch 22, batch 12650, giga_loss[loss=0.2761, simple_loss=0.3467, pruned_loss=0.1027, over 28744.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3635, pruned_loss=0.1183, over 5657586.83 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.346, pruned_loss=0.09142, over 5759989.97 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3659, pruned_loss=0.1211, over 5642384.37 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:32:10,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:32:12,164 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:32:39,284 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:32:49,152 INFO [train.py:968] (0/2) Epoch 22, batch 12700, giga_loss[loss=0.2894, simple_loss=0.3561, pruned_loss=0.1113, over 28344.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3623, pruned_loss=0.1179, over 5649820.29 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3458, pruned_loss=0.09122, over 5753820.23 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3651, pruned_loss=0.1213, over 5639974.02 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:33:02,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6222, 1.7795, 1.4247, 1.6227], device='cuda:0'), covar=tensor([0.2563, 0.2734, 0.3056, 0.2454], device='cuda:0'), in_proj_covar=tensor([0.1517, 0.1095, 0.1336, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 08:33:12,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7669, 2.7538, 2.5490, 2.4224], device='cuda:0'), covar=tensor([0.1784, 0.2246, 0.2090, 0.2141], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0758, 0.0722, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 08:33:34,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.670e+03 2.153e+03 2.952e+03 6.408e+03, threshold=4.307e+03, percent-clipped=4.0
+2023-03-11 08:33:35,234 INFO [train.py:968] (0/2) Epoch 22, batch 12750, giga_loss[loss=0.3306, simple_loss=0.3759, pruned_loss=0.1426, over 28792.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3618, pruned_loss=0.1176, over 5655699.22 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3457, pruned_loss=0.09113, over 5756447.70 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3643, pruned_loss=0.1209, over 5643824.80 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:33:53,261 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:33:55,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:34:02,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2317, 1.6615, 1.5428, 1.5063], device='cuda:0'), covar=tensor([0.1775, 0.1453, 0.1921, 0.1568], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0756, 0.0720, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 08:34:22,839 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:34:23,184 INFO [train.py:968] (0/2) Epoch 22, batch 12800, giga_loss[loss=0.2987, simple_loss=0.3676, pruned_loss=0.1149, over 28928.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3604, pruned_loss=0.1143, over 5661585.21 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3456, pruned_loss=0.09115, over 5761150.03 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.363, pruned_loss=0.1177, over 5645135.35 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:35:10,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.379e+02 1.512e+03 2.160e+03 2.957e+03 9.205e+03, threshold=4.319e+03, percent-clipped=13.0
+2023-03-11 08:35:13,751 INFO [train.py:968] (0/2) Epoch 22, batch 12850, giga_loss[loss=0.2755, simple_loss=0.3286, pruned_loss=0.1113, over 24345.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3585, pruned_loss=0.1113, over 5652801.15 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3453, pruned_loss=0.091, over 5762297.33 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3613, pruned_loss=0.1147, over 5635704.60 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:35:25,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:35:51,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:35:57,201 INFO [train.py:968] (0/2) Epoch 22, batch 12900, giga_loss[loss=0.2515, simple_loss=0.3336, pruned_loss=0.08469, over 28836.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3553, pruned_loss=0.1078, over 5652795.34 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3446, pruned_loss=0.09088, over 5754016.41 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3587, pruned_loss=0.1114, over 5641791.37 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:36:12,705 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4567, 1.6815, 1.6584, 1.4042], device='cuda:0'), covar=tensor([0.2420, 0.1988, 0.1457, 0.1921], device='cuda:0'), in_proj_covar=tensor([0.1958, 0.1897, 0.1817, 0.1959], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 08:36:16,090 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-11 08:36:26,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5075, 1.5953, 1.6893, 1.3095], device='cuda:0'), covar=tensor([0.1807, 0.2730, 0.1547, 0.1894], device='cuda:0'), in_proj_covar=tensor([0.0898, 0.0701, 0.0946, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 08:36:38,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:36:52,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.215e+02 1.447e+03 1.991e+03 2.639e+03 5.469e+03, threshold=3.982e+03, percent-clipped=3.0
+2023-03-11 08:36:52,169 INFO [train.py:968] (0/2) Epoch 22, batch 12950, giga_loss[loss=0.2895, simple_loss=0.3618, pruned_loss=0.1086, over 28840.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3528, pruned_loss=0.1051, over 5652169.72 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3445, pruned_loss=0.09086, over 5754993.33 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3555, pruned_loss=0.108, over 5642268.97 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:37:22,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:37:30,328 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=970722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:37:37,870 INFO [train.py:968] (0/2) Epoch 22, batch 13000, giga_loss[loss=0.2575, simple_loss=0.3469, pruned_loss=0.08406, over 28818.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3493, pruned_loss=0.1017, over 5650402.85 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3434, pruned_loss=0.0905, over 5756933.64 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3529, pruned_loss=0.1049, over 5636617.31 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:37:45,013 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:37:49,132 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:38:14,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:38:15,731 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:38:16,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:38:28,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.888e+02 1.417e+03 1.755e+03 2.127e+03 7.792e+03, threshold=3.510e+03, percent-clipped=4.0
+2023-03-11 08:38:28,250 INFO [train.py:968] (0/2) Epoch 22, batch 13050, giga_loss[loss=0.2952, simple_loss=0.3735, pruned_loss=0.1084, over 28950.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3502, pruned_loss=0.09978, over 5664101.11 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3433, pruned_loss=0.09047, over 5756036.97 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3531, pruned_loss=0.1024, over 5652783.11 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:38:47,480 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:39:00,828 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:39:03,277 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:39:05,406 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4180, 3.1929, 1.5071, 1.5252], device='cuda:0'), covar=tensor([0.0945, 0.0333, 0.0959, 0.1326], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0557, 0.0387, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 08:39:20,260 INFO [train.py:968] (0/2) Epoch 22, batch 13100, giga_loss[loss=0.3017, simple_loss=0.3711, pruned_loss=0.1162, over 28610.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3517, pruned_loss=0.1013, over 5655729.95 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3432, pruned_loss=0.09046, over 5759118.10 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3544, pruned_loss=0.1036, over 5642075.40 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:39:34,037 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:39:43,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:39:46,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:40:09,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.881e+02 1.331e+03 1.798e+03 3.077e+03 5.918e+03, threshold=3.596e+03, percent-clipped=20.0
+2023-03-11 08:40:09,485 INFO [train.py:968] (0/2) Epoch 22, batch 13150, giga_loss[loss=0.243, simple_loss=0.3058, pruned_loss=0.0901, over 24250.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3494, pruned_loss=0.09948, over 5659181.51 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3427, pruned_loss=0.09026, over 5760879.12 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.352, pruned_loss=0.1016, over 5645471.23 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:40:15,991 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:40:35,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2674, 1.2355, 3.7501, 3.2360], device='cuda:0'), covar=tensor([0.1633, 0.2694, 0.0470, 0.0944], device='cuda:0'), in_proj_covar=tensor([0.0760, 0.0647, 0.0959, 0.0906], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 08:40:52,568 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.85 vs. limit=2.0
+2023-03-11 08:40:58,885 INFO [train.py:968] (0/2) Epoch 22, batch 13200, giga_loss[loss=0.2583, simple_loss=0.343, pruned_loss=0.08685, over 28971.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3463, pruned_loss=0.09783, over 5643949.03 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3428, pruned_loss=0.09038, over 5759393.56 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3484, pruned_loss=0.0996, over 5632405.43 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 08:41:46,905 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.624e+02 1.552e+03 1.989e+03 2.593e+03 4.103e+03, threshold=3.978e+03, percent-clipped=5.0
+2023-03-11 08:41:46,917 INFO [train.py:968] (0/2) Epoch 22, batch 13250, giga_loss[loss=0.271, simple_loss=0.3493, pruned_loss=0.09639, over 27906.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3464, pruned_loss=0.09794, over 5646783.55 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3425, pruned_loss=0.09039, over 5761462.65 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3484, pruned_loss=0.09952, over 5633118.71 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 08:42:34,838 INFO [train.py:968] (0/2) Epoch 22, batch 13300, giga_loss[loss=0.3026, simple_loss=0.3689, pruned_loss=0.1181, over 28251.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3453, pruned_loss=0.09731, over 5649392.46 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3417, pruned_loss=0.09004, over 5764389.96 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3478, pruned_loss=0.09904, over 5633725.73 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:42:59,388 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-11 08:43:27,851 INFO [train.py:968] (0/2) Epoch 22, batch 13350, giga_loss[loss=0.28, simple_loss=0.3566, pruned_loss=0.1017, over 28943.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3424, pruned_loss=0.09458, over 5651051.76 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3415, pruned_loss=0.08998, over 5765309.51 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3445, pruned_loss=0.09606, over 5636200.71 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:43:28,443 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.508e+02 1.357e+03 1.897e+03 2.592e+03 4.982e+03, threshold=3.794e+03, percent-clipped=5.0
+2023-03-11 08:43:33,396 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=971087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:43:42,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=971097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:43:49,152 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-11 08:44:15,628 INFO [train.py:968] (0/2) Epoch 22, batch 13400, giga_loss[loss=0.2549, simple_loss=0.3325, pruned_loss=0.08864, over 28705.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3398, pruned_loss=0.09286, over 5654676.90 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3414, pruned_loss=0.09013, over 5765931.03 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3416, pruned_loss=0.094, over 5639757.91 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:45:06,998 INFO [train.py:968] (0/2) Epoch 22, batch 13450, giga_loss[loss=0.252, simple_loss=0.3276, pruned_loss=0.08819, over 28840.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3353, pruned_loss=0.09036, over 5655133.23 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3409, pruned_loss=0.08983, over 5766118.18 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3371, pruned_loss=0.09158, over 5640146.84 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:45:07,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.638e+02 1.368e+03 1.810e+03 2.150e+03 4.618e+03, threshold=3.619e+03, percent-clipped=2.0
+2023-03-11 08:46:00,135 INFO [train.py:968] (0/2) Epoch 22, batch 13500, giga_loss[loss=0.2802, simple_loss=0.3393, pruned_loss=0.1106, over 26721.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3354, pruned_loss=0.09113, over 5650028.97 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3409, pruned_loss=0.09003, over 5759470.11 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3366, pruned_loss=0.09191, over 5642437.33 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:46:10,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=971240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:46:13,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=971243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:46:40,802 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-11 08:46:44,914 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=971272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:46:55,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5408, 1.7590, 1.7501, 1.5222], device='cuda:0'), covar=tensor([0.2499, 0.1961, 0.1872, 0.2090], device='cuda:0'), in_proj_covar=tensor([0.1929, 0.1861, 0.1784, 0.1921], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 08:46:58,384 INFO [train.py:968] (0/2) Epoch 22, batch 13550, giga_loss[loss=0.2249, simple_loss=0.312, pruned_loss=0.06892, over 28880.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3354, pruned_loss=0.09124, over 5642969.94 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3409, pruned_loss=0.09004, over 5760199.76 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3364, pruned_loss=0.09185, over 5635943.95 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:46:59,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.572e+02 1.428e+03 1.984e+03 2.849e+03 1.273e+04, threshold=3.968e+03, percent-clipped=11.0
+2023-03-11 08:47:51,160 INFO [train.py:968] (0/2) Epoch 22, batch 13600, giga_loss[loss=0.2968, simple_loss=0.3688, pruned_loss=0.1124, over 28488.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3375, pruned_loss=0.09143, over 5641980.24 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3405, pruned_loss=0.08995, over 5753953.54 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3386, pruned_loss=0.09206, over 5637943.91 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 08:48:47,716 INFO [train.py:968] (0/2) Epoch 22, batch 13650, giga_loss[loss=0.2463, simple_loss=0.3329, pruned_loss=0.07989, over 28470.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3388, pruned_loss=0.09111, over 5653598.38 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3405, pruned_loss=0.08994, over 5754858.50 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3397, pruned_loss=0.09167, over 5647191.29 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:48:49,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.666e+02 1.543e+03 1.850e+03 2.369e+03 9.821e+03, threshold=3.699e+03, percent-clipped=5.0
+2023-03-11 08:49:27,571 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=971415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:49:44,983 INFO [train.py:968] (0/2) Epoch 22, batch 13700, giga_loss[loss=0.261, simple_loss=0.3364, pruned_loss=0.09283, over 28052.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3396, pruned_loss=0.09176, over 5652988.32 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3407, pruned_loss=0.09034, over 5741094.75 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.34, pruned_loss=0.09186, over 5657693.97 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:50:19,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=971462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:50:41,195 INFO [train.py:968] (0/2) Epoch 22, batch 13750, libri_loss[loss=0.2709, simple_loss=0.3444, pruned_loss=0.09871, over 29519.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3362, pruned_loss=0.08976, over 5650880.00 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.34, pruned_loss=0.09006, over 5737240.62 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.337, pruned_loss=0.0901, over 5655274.01 frames. ], batch size: 80, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:50:42,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.660e+02 1.309e+03 1.862e+03 2.738e+03 7.014e+03, threshold=3.724e+03, percent-clipped=8.0
+2023-03-11 08:51:22,612 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1439, 1.4707, 1.1352, 0.5638], device='cuda:0'), covar=tensor([0.3459, 0.2188, 0.3536, 0.5156], device='cuda:0'), in_proj_covar=tensor([0.1751, 0.1645, 0.1588, 0.1417], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 08:51:39,024 INFO [train.py:968] (0/2) Epoch 22, batch 13800, giga_loss[loss=0.2319, simple_loss=0.3149, pruned_loss=0.07446, over 24379.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3355, pruned_loss=0.08816, over 5655940.92 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3392, pruned_loss=0.08982, over 5741278.77 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3368, pruned_loss=0.08862, over 5653670.15 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:52:24,138 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-11 08:52:32,657 INFO [train.py:968] (0/2) Epoch 22, batch 13850, giga_loss[loss=0.3024, simple_loss=0.3655, pruned_loss=0.1197, over 28640.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3344, pruned_loss=0.08764, over 5666106.27 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3391, pruned_loss=0.09016, over 5748816.71 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3354, pruned_loss=0.08761, over 5653964.49 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:52:36,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.687e+02 1.390e+03 2.001e+03 2.915e+03 7.226e+03, threshold=4.002e+03, percent-clipped=10.0
+2023-03-11 08:53:04,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=971605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:53:07,793 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=971608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:53:09,275 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=971610.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 08:53:33,670 INFO [train.py:968] (0/2) Epoch 22, batch 13900, giga_loss[loss=0.2456, simple_loss=0.3314, pruned_loss=0.07997, over 28751.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3326, pruned_loss=0.08776, over 5671063.44 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3389, pruned_loss=0.0902, over 5750330.72 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3335, pruned_loss=0.08767, over 5658717.72 frames. ], batch size: 243, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:53:42,243 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=971637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:54:14,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3252, 2.8554, 1.4256, 1.4532], device='cuda:0'), covar=tensor([0.0952, 0.0384, 0.0960, 0.1328], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0556, 0.0389, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 08:54:33,429 INFO [train.py:968] (0/2) Epoch 22, batch 13950, giga_loss[loss=0.2071, simple_loss=0.295, pruned_loss=0.05957, over 28806.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3305, pruned_loss=0.08704, over 5673468.62 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3388, pruned_loss=0.09027, over 5752024.42 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3312, pruned_loss=0.08688, over 5661407.02 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:54:36,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.263e+03 1.600e+03 2.396e+03 4.417e+03, threshold=3.199e+03, percent-clipped=4.0
+2023-03-11 08:55:30,031 INFO [train.py:968] (0/2) Epoch 22, batch 14000, libri_loss[loss=0.224, simple_loss=0.3032, pruned_loss=0.07246, over 29592.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.331, pruned_loss=0.08726, over 5667126.12 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3382, pruned_loss=0.08994, over 5754964.32 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.332, pruned_loss=0.0874, over 5653136.32 frames. ], batch size: 74, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 08:56:28,034 INFO [train.py:968] (0/2) Epoch 22, batch 14050, giga_loss[loss=0.2761, simple_loss=0.3553, pruned_loss=0.09848, over 28667.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3336, pruned_loss=0.08822, over 5663345.06 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3378, pruned_loss=0.08976, over 5756053.26 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3346, pruned_loss=0.08843, over 5648300.21 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:56:31,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.715e+02 1.505e+03 1.900e+03 2.842e+03 6.303e+03, threshold=3.800e+03, percent-clipped=19.0
+2023-03-11 08:56:32,153 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=971784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:56:40,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=971790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:56:51,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4108, 1.6775, 1.6865, 1.2046], device='cuda:0'), covar=tensor([0.1900, 0.2779, 0.1600, 0.1938], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0700, 0.0949, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 08:57:30,441 INFO [train.py:968] (0/2) Epoch 22, batch 14100, giga_loss[loss=0.2343, simple_loss=0.3153, pruned_loss=0.07667, over 28343.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3314, pruned_loss=0.08618, over 5675780.86 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3375, pruned_loss=0.08965, over 5763038.10 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3323, pruned_loss=0.08636, over 5653590.72 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:58:21,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3303, 1.6997, 1.3143, 1.2966], device='cuda:0'), covar=tensor([0.2536, 0.2347, 0.2784, 0.2155], device='cuda:0'), in_proj_covar=tensor([0.1512, 0.1091, 0.1336, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 08:58:34,762 INFO [train.py:968] (0/2) Epoch 22, batch 14150, libri_loss[loss=0.2034, simple_loss=0.2858, pruned_loss=0.0605, over 29363.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3309, pruned_loss=0.0862, over 5684671.79 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3369, pruned_loss=0.08935, over 5765787.65 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3321, pruned_loss=0.08654, over 5663042.29 frames. ], batch size: 67, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:58:38,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.130e+02 1.492e+03 1.928e+03 2.619e+03 4.940e+03, threshold=3.855e+03, percent-clipped=7.0
+2023-03-11 08:59:38,712 INFO [train.py:968] (0/2) Epoch 22, batch 14200, giga_loss[loss=0.2355, simple_loss=0.3189, pruned_loss=0.07609, over 28718.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3341, pruned_loss=0.0877, over 5688389.60 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3372, pruned_loss=0.08949, over 5765105.10 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3347, pruned_loss=0.08781, over 5670458.35 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:59:43,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=971933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:59:47,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=971936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:00:01,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5942, 1.7948, 1.5089, 1.6630], device='cuda:0'), covar=tensor([0.2781, 0.2789, 0.3096, 0.2547], device='cuda:0'), in_proj_covar=tensor([0.1515, 0.1093, 0.1338, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 09:00:24,212 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=971965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:00:44,080 INFO [train.py:968] (0/2) Epoch 22, batch 14250, giga_loss[loss=0.2451, simple_loss=0.3369, pruned_loss=0.07669, over 28089.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3378, pruned_loss=0.08709, over 5677920.68 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3369, pruned_loss=0.08937, over 5765327.75 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3386, pruned_loss=0.08727, over 5662836.67 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:00:47,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.168e+02 1.559e+03 2.026e+03 2.756e+03 5.418e+03, threshold=4.052e+03, percent-clipped=9.0
+2023-03-11 09:00:49,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=971985.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 09:01:07,599 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-972000.pt
+2023-03-11 09:01:45,128 INFO [train.py:968] (0/2) Epoch 22, batch 14300, giga_loss[loss=0.263, simple_loss=0.3319, pruned_loss=0.09704, over 26898.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3397, pruned_loss=0.08655, over 5678894.32 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3371, pruned_loss=0.08941, over 5767984.83 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3402, pruned_loss=0.0866, over 5663094.26 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:02:42,229 INFO [train.py:968] (0/2) Epoch 22, batch 14350, giga_loss[loss=0.3144, simple_loss=0.3703, pruned_loss=0.1292, over 26851.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3393, pruned_loss=0.08589, over 5679474.09 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3366, pruned_loss=0.08931, over 5770372.76 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3401, pruned_loss=0.08593, over 5662434.85 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:02:45,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.007e+02 1.429e+03 1.717e+03 2.342e+03 4.464e+03, threshold=3.434e+03, percent-clipped=3.0
+2023-03-11 09:03:43,699 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=972128.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 09:03:45,897 INFO [train.py:968] (0/2) Epoch 22, batch 14400, giga_loss[loss=0.2523, simple_loss=0.3311, pruned_loss=0.08673, over 28997.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3399, pruned_loss=0.08669, over 5679493.17 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3366, pruned_loss=0.08927, over 5772676.53 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3406, pruned_loss=0.0867, over 5663035.33 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:03:46,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=972131.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 09:04:02,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4379, 1.5485, 1.6531, 1.3255], device='cuda:0'), covar=tensor([0.1588, 0.2421, 0.1372, 0.1669], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0699, 0.0949, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 09:04:18,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=972159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:04:21,427 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=972160.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 09:04:47,112 INFO [train.py:968] (0/2) Epoch 22, batch 14450, giga_loss[loss=0.2405, simple_loss=0.3228, pruned_loss=0.07914, over 28763.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3381, pruned_loss=0.08659, over 5679870.43 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3364, pruned_loss=0.08921, over 5765637.80 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3388, pruned_loss=0.08663, over 5671773.33 frames. ], batch size: 243, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:04:50,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.113e+02 1.347e+03 1.733e+03 2.370e+03 6.241e+03, threshold=3.465e+03, percent-clipped=6.0
+2023-03-11 09:06:01,684 INFO [train.py:968] (0/2) Epoch 22, batch 14500, giga_loss[loss=0.243, simple_loss=0.329, pruned_loss=0.07847, over 28963.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3394, pruned_loss=0.08828, over 5679160.64 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3368, pruned_loss=0.08955, over 5758163.46 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3397, pruned_loss=0.08799, over 5677599.16 frames. ], batch size: 106, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:06:21,313 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=972245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:07:22,022 INFO [train.py:968] (0/2) Epoch 22, batch 14550, giga_loss[loss=0.2335, simple_loss=0.3109, pruned_loss=0.07808, over 27642.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3363, pruned_loss=0.08741, over 5671749.38 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3367, pruned_loss=0.08956, over 5757107.98 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3367, pruned_loss=0.08713, over 5670243.46 frames. ], batch size: 472, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:07:25,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.479e+02 1.293e+03 1.733e+03 2.480e+03 6.508e+03, threshold=3.467e+03, percent-clipped=9.0
+2023-03-11 09:07:42,593 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-11 09:07:49,347 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=972302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:07:54,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=972305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:08:25,705 INFO [train.py:968] (0/2) Epoch 22, batch 14600, giga_loss[loss=0.2573, simple_loss=0.3341, pruned_loss=0.09024, over 28125.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3337, pruned_loss=0.08568, over 5678050.44 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.336, pruned_loss=0.08918, over 5761137.85 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3345, pruned_loss=0.08572, over 5671137.27 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:08:29,666 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=972334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:09:31,757 INFO [train.py:968] (0/2) Epoch 22, batch 14650, giga_loss[loss=0.2356, simple_loss=0.3011, pruned_loss=0.08507, over 24263.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3302, pruned_loss=0.08404, over 5678906.51 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.336, pruned_loss=0.08923, over 5763250.78 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3308, pruned_loss=0.08396, over 5670605.55 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:09:36,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.281e+02 1.374e+03 1.737e+03 2.606e+03 8.300e+03, threshold=3.474e+03, percent-clipped=9.0
+2023-03-11 09:10:31,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5884, 1.5764, 1.2460, 1.2301], device='cuda:0'), covar=tensor([0.0835, 0.0476, 0.0863, 0.1114], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0442, 0.0516, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 09:10:31,383 INFO [train.py:968] (0/2) Epoch 22, batch 14700, giga_loss[loss=0.2218, simple_loss=0.295, pruned_loss=0.07433, over 24540.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3348, pruned_loss=0.08728, over 5667318.50 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3353, pruned_loss=0.08899, over 5756703.05 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3359, pruned_loss=0.08733, over 5664426.93 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:11:13,115 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0107, 4.8506, 4.6402, 2.3882], device='cuda:0'), covar=tensor([0.0442, 0.0606, 0.0723, 0.1759], device='cuda:0'), in_proj_covar=tensor([0.1207, 0.1116, 0.0948, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 09:11:22,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5514, 1.7214, 1.2060, 1.3883], device='cuda:0'), covar=tensor([0.0814, 0.0430, 0.0909, 0.1169], device='cuda:0'), in_proj_covar=tensor([0.0391, 0.0442, 0.0515, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 09:11:27,130 INFO [train.py:968] (0/2) Epoch 22, batch 14750, giga_loss[loss=0.2042, simple_loss=0.2958, pruned_loss=0.05629, over 29012.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3351, pruned_loss=0.08769, over 5682890.19 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3342, pruned_loss=0.08856, over 5760855.73 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.337, pruned_loss=0.08811, over 5673984.51 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:11:31,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.673e+02 1.676e+03 2.085e+03 3.270e+03 7.780e+03, threshold=4.171e+03, percent-clipped=16.0
+2023-03-11 09:11:35,215 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=972487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:12:20,812 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4442, 1.5238, 1.4272, 1.4364], device='cuda:0'), covar=tensor([0.0744, 0.0318, 0.0322, 0.0919], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0117, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 09:12:28,317 INFO [train.py:968] (0/2) Epoch 22, batch 14800, giga_loss[loss=0.2635, simple_loss=0.3207, pruned_loss=0.1031, over 24533.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3346, pruned_loss=0.08873, over 5678239.55 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.334, pruned_loss=0.0884, over 5761237.42 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3364, pruned_loss=0.0892, over 5669217.00 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:12:47,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6304, 1.3562, 4.2373, 3.3862], device='cuda:0'), covar=tensor([0.1515, 0.2752, 0.0493, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.0759, 0.0649, 0.0957, 0.0900], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 09:12:54,270 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-11 09:13:20,397 INFO [train.py:968] (0/2) Epoch 22, batch 14850, giga_loss[loss=0.2553, simple_loss=0.333, pruned_loss=0.08876, over 28834.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3343, pruned_loss=0.08912, over 5682528.31 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.333, pruned_loss=0.08804, over 5758681.33 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3366, pruned_loss=0.08987, over 5673826.59 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:13:26,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.302e+02 1.423e+03 1.960e+03 2.633e+03 7.150e+03, threshold=3.920e+03, percent-clipped=8.0
+2023-03-11 09:14:10,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=972620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:14:25,426 INFO [train.py:968] (0/2) Epoch 22, batch 14900, giga_loss[loss=0.2709, simple_loss=0.3488, pruned_loss=0.09649, over 28075.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3355, pruned_loss=0.0895, over 5683847.99 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3332, pruned_loss=0.08812, over 5759324.21 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3372, pruned_loss=0.09003, over 5675850.92 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:15:38,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-11 09:15:39,395 INFO [train.py:968] (0/2) Epoch 22, batch 14950, giga_loss[loss=0.2519, simple_loss=0.3289, pruned_loss=0.08743, over 27628.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3367, pruned_loss=0.08883, over 5677837.21 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3329, pruned_loss=0.08796, over 5760083.59 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3383, pruned_loss=0.08938, over 5670590.18 frames. ], batch size: 474, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:15:47,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.484e+02 1.619e+03 2.075e+03 2.773e+03 1.128e+04, threshold=4.150e+03, percent-clipped=11.0
+2023-03-11 09:17:00,993 INFO [train.py:968] (0/2) Epoch 22, batch 15000, giga_loss[loss=0.2476, simple_loss=0.3133, pruned_loss=0.091, over 24359.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3358, pruned_loss=0.08795, over 5666456.60 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3331, pruned_loss=0.088, over 5761906.96 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.337, pruned_loss=0.08837, over 5658198.30 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:17:00,997 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 09:17:09,797 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.1979, simple_loss=0.2984, pruned_loss=0.04871, over 944034.00 frames. 
+2023-03-11 09:17:09,798 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 09:17:53,814 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=972763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:17:56,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=972766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:18:00,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9907, 2.2959, 1.6268, 1.8328], device='cuda:0'), covar=tensor([0.0938, 0.0553, 0.0887, 0.1053], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0439, 0.0513, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 09:18:10,832 INFO [train.py:968] (0/2) Epoch 22, batch 15050, giga_loss[loss=0.2249, simple_loss=0.3072, pruned_loss=0.07134, over 28644.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3336, pruned_loss=0.08831, over 5673127.30 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3331, pruned_loss=0.08808, over 5766622.87 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3347, pruned_loss=0.0886, over 5659021.13 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:18:16,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.375e+02 1.488e+03 1.896e+03 2.646e+03 5.298e+03, threshold=3.792e+03, percent-clipped=1.0
+2023-03-11 09:18:28,059 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=972795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:19:12,362 INFO [train.py:968] (0/2) Epoch 22, batch 15100, giga_loss[loss=0.2533, simple_loss=0.3281, pruned_loss=0.08923, over 28436.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3275, pruned_loss=0.08567, over 5671625.23 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3327, pruned_loss=0.08793, over 5765870.55 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3286, pruned_loss=0.086, over 5657797.98 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:19:12,956 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-11 09:19:49,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=972862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:20:11,111 INFO [train.py:968] (0/2) Epoch 22, batch 15150, giga_loss[loss=0.2634, simple_loss=0.3428, pruned_loss=0.09196, over 28451.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3272, pruned_loss=0.08554, over 5665352.78 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3325, pruned_loss=0.08792, over 5757990.56 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3281, pruned_loss=0.08572, over 5660080.06 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:20:16,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.582e+03 2.034e+03 2.988e+03 7.698e+03, threshold=4.069e+03, percent-clipped=14.0
+2023-03-11 09:20:38,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6383, 1.6966, 1.8925, 1.4692], device='cuda:0'), covar=tensor([0.1525, 0.2245, 0.1280, 0.1645], device='cuda:0'), in_proj_covar=tensor([0.0897, 0.0697, 0.0947, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 09:20:45,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3397, 1.3440, 3.9920, 3.2708], device='cuda:0'), covar=tensor([0.1596, 0.2630, 0.0453, 0.0767], device='cuda:0'), in_proj_covar=tensor([0.0756, 0.0646, 0.0953, 0.0895], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 09:21:06,118 INFO [train.py:968] (0/2) Epoch 22, batch 15200, libri_loss[loss=0.2477, simple_loss=0.3305, pruned_loss=0.08252, over 29150.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3287, pruned_loss=0.08686, over 5664655.55 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3322, pruned_loss=0.08777, over 5761418.32 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3296, pruned_loss=0.0871, over 5655095.23 frames. ], batch size: 101, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:21:08,848 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6617, 1.9062, 1.5260, 1.9033], device='cuda:0'), covar=tensor([0.2597, 0.2639, 0.2991, 0.2400], device='cuda:0'), in_proj_covar=tensor([0.1511, 0.1089, 0.1333, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 09:21:27,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5072, 1.9314, 1.9520, 1.6542], device='cuda:0'), covar=tensor([0.2024, 0.1964, 0.2147, 0.2046], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0735, 0.0702, 0.0673], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 09:21:58,240 INFO [train.py:968] (0/2) Epoch 22, batch 15250, giga_loss[loss=0.2452, simple_loss=0.3282, pruned_loss=0.08108, over 28938.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3267, pruned_loss=0.08543, over 5664281.45 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3318, pruned_loss=0.08771, over 5755945.98 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3276, pruned_loss=0.08562, over 5658395.51 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:22:05,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.687e+02 1.510e+03 1.794e+03 2.615e+03 8.686e+03, threshold=3.587e+03, percent-clipped=5.0
+2023-03-11 09:22:23,617 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=973002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:22:28,183 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=973005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:22:31,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=973008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:22:55,854 INFO [train.py:968] (0/2) Epoch 22, batch 15300, giga_loss[loss=0.2373, simple_loss=0.3199, pruned_loss=0.07734, over 28737.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3254, pruned_loss=0.08406, over 5661834.77 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3314, pruned_loss=0.08749, over 5759545.40 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3263, pruned_loss=0.08432, over 5651254.50 frames. ], batch size: 243, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:23:02,605 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=973037.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 09:23:02,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=973037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:23:06,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-11 09:23:20,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=973055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:23:58,085 INFO [train.py:968] (0/2) Epoch 22, batch 15350, giga_loss[loss=0.2639, simple_loss=0.3387, pruned_loss=0.09457, over 28131.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3249, pruned_loss=0.08397, over 5669520.57 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3311, pruned_loss=0.08735, over 5762728.04 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3257, pruned_loss=0.08419, over 5655495.26 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:24:06,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.444e+02 1.485e+03 2.039e+03 2.548e+03 6.058e+03, threshold=4.078e+03, percent-clipped=9.0
+2023-03-11 09:25:04,016 INFO [train.py:968] (0/2) Epoch 22, batch 15400, giga_loss[loss=0.2092, simple_loss=0.2949, pruned_loss=0.06173, over 28861.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3236, pruned_loss=0.083, over 5665348.62 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3308, pruned_loss=0.08715, over 5765420.32 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3244, pruned_loss=0.08329, over 5650151.51 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:25:39,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6336, 1.9092, 1.2568, 1.4478], device='cuda:0'), covar=tensor([0.0983, 0.0570, 0.1111, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0438, 0.0512, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 09:26:03,346 INFO [train.py:968] (0/2) Epoch 22, batch 15450, giga_loss[loss=0.2924, simple_loss=0.3609, pruned_loss=0.112, over 28083.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3245, pruned_loss=0.08324, over 5665164.45 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3305, pruned_loss=0.08701, over 5767874.96 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3252, pruned_loss=0.08348, over 5647516.41 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:26:07,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5628, 1.7161, 1.8298, 1.3647], device='cuda:0'), covar=tensor([0.1882, 0.2662, 0.1540, 0.1821], device='cuda:0'), in_proj_covar=tensor([0.0897, 0.0695, 0.0946, 0.0844], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 09:26:08,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.418e+02 1.254e+03 1.646e+03 2.238e+03 5.052e+03, threshold=3.293e+03, percent-clipped=6.0
+2023-03-11 09:26:37,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7938, 1.9227, 1.8766, 1.7414], device='cuda:0'), covar=tensor([0.2196, 0.2011, 0.1683, 0.1861], device='cuda:0'), in_proj_covar=tensor([0.1927, 0.1854, 0.1764, 0.1914], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 09:27:06,065 INFO [train.py:968] (0/2) Epoch 22, batch 15500, giga_loss[loss=0.315, simple_loss=0.3699, pruned_loss=0.1301, over 28672.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3252, pruned_loss=0.0844, over 5671815.37 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3301, pruned_loss=0.08686, over 5769704.37 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3261, pruned_loss=0.08466, over 5653996.09 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:27:20,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5496, 1.6213, 1.7928, 1.4078], device='cuda:0'), covar=tensor([0.1648, 0.2446, 0.1373, 0.1704], device='cuda:0'), in_proj_covar=tensor([0.0896, 0.0694, 0.0946, 0.0844], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 09:28:10,377 INFO [train.py:968] (0/2) Epoch 22, batch 15550, giga_loss[loss=0.2222, simple_loss=0.3097, pruned_loss=0.06739, over 28955.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3238, pruned_loss=0.08361, over 5666642.36 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3301, pruned_loss=0.08702, over 5771602.14 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3243, pruned_loss=0.0836, over 5649318.69 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:28:18,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.102e+02 1.342e+03 1.597e+03 2.347e+03 7.941e+03, threshold=3.194e+03, percent-clipped=7.0
+2023-03-11 09:29:10,667 INFO [train.py:968] (0/2) Epoch 22, batch 15600, giga_loss[loss=0.264, simple_loss=0.3622, pruned_loss=0.08289, over 28741.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3258, pruned_loss=0.08272, over 5676313.33 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3299, pruned_loss=0.08701, over 5773035.52 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3263, pruned_loss=0.08268, over 5660495.40 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:30:08,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=973377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:30:13,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3501, 1.2374, 1.1326, 1.4724], device='cuda:0'), covar=tensor([0.0799, 0.0359, 0.0364, 0.0890], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 09:30:14,569 INFO [train.py:968] (0/2) Epoch 22, batch 15650, giga_loss[loss=0.2633, simple_loss=0.3524, pruned_loss=0.08707, over 28691.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3294, pruned_loss=0.08429, over 5659376.40 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.33, pruned_loss=0.08703, over 5764138.86 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3298, pruned_loss=0.08422, over 5654239.42 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:30:21,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.251e+02 1.476e+03 2.085e+03 3.239e+03 1.039e+04, threshold=4.171e+03, percent-clipped=24.0
+2023-03-11 09:30:46,163 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=973412.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 09:31:08,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=973430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:31:08,805 INFO [train.py:968] (0/2) Epoch 22, batch 15700, giga_loss[loss=0.2854, simple_loss=0.355, pruned_loss=0.1079, over 28151.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3319, pruned_loss=0.08562, over 5652632.18 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.33, pruned_loss=0.08704, over 5752257.11 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3323, pruned_loss=0.08549, over 5655464.75 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:31:29,613 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 09:31:55,282 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4636, 1.6455, 1.5679, 1.4065], device='cuda:0'), covar=tensor([0.2278, 0.1974, 0.1682, 0.2091], device='cuda:0'), in_proj_covar=tensor([0.1931, 0.1857, 0.1765, 0.1920], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 09:32:06,221 INFO [train.py:968] (0/2) Epoch 22, batch 15750, giga_loss[loss=0.2628, simple_loss=0.3364, pruned_loss=0.09462, over 28987.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3318, pruned_loss=0.08544, over 5660374.85 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3302, pruned_loss=0.0872, over 5744761.70 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3319, pruned_loss=0.08516, over 5668165.30 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:32:17,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.711e+02 1.391e+03 1.777e+03 2.267e+03 4.542e+03, threshold=3.553e+03, percent-clipped=2.0
+2023-03-11 09:32:53,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=973520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:32:56,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=973523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:33:07,168 INFO [train.py:968] (0/2) Epoch 22, batch 15800, giga_loss[loss=0.2078, simple_loss=0.2969, pruned_loss=0.05936, over 28932.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3292, pruned_loss=0.08325, over 5673946.63 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3301, pruned_loss=0.0871, over 5747382.94 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3293, pruned_loss=0.0831, over 5676753.05 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:33:31,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5325, 2.1639, 1.6123, 0.6950], device='cuda:0'), covar=tensor([0.6135, 0.3015, 0.4078, 0.6517], device='cuda:0'), in_proj_covar=tensor([0.1748, 0.1640, 0.1586, 0.1419], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 09:33:32,093 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=973552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:33:35,934 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=973555.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 09:33:40,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=973558.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 09:33:57,048 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=973573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:34:00,298 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=973576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:34:04,719 INFO [train.py:968] (0/2) Epoch 22, batch 15850, giga_loss[loss=0.2531, simple_loss=0.3426, pruned_loss=0.08184, over 28744.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3286, pruned_loss=0.08313, over 5674066.79 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3307, pruned_loss=0.08774, over 5745020.03 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.328, pruned_loss=0.0822, over 5675820.35 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:34:12,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=973587.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 09:34:12,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.858e+02 1.288e+03 1.595e+03 2.484e+03 6.766e+03, threshold=3.189e+03, percent-clipped=10.0
+2023-03-11 09:34:19,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=973594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:34:35,391 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=973605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:34:50,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5049, 1.8110, 1.4152, 1.6859], device='cuda:0'), covar=tensor([0.2609, 0.2573, 0.3014, 0.2341], device='cuda:0'), in_proj_covar=tensor([0.1505, 0.1085, 0.1331, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 09:35:01,526 INFO [train.py:968] (0/2) Epoch 22, batch 15900, giga_loss[loss=0.1889, simple_loss=0.2765, pruned_loss=0.05071, over 28933.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3261, pruned_loss=0.08284, over 5673623.31 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3303, pruned_loss=0.08752, over 5745421.39 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3261, pruned_loss=0.08223, over 5673120.87 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:35:55,537 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 09:36:02,325 INFO [train.py:968] (0/2) Epoch 22, batch 15950, giga_loss[loss=0.301, simple_loss=0.3766, pruned_loss=0.1127, over 28388.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3278, pruned_loss=0.08383, over 5670595.04 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3302, pruned_loss=0.08738, over 5747784.42 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3278, pruned_loss=0.08341, over 5667194.64 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:36:03,830 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-11 09:36:11,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.829e+02 1.336e+03 1.786e+03 2.367e+03 9.183e+03, threshold=3.572e+03, percent-clipped=11.0
+2023-03-11 09:37:01,674 INFO [train.py:968] (0/2) Epoch 22, batch 16000, giga_loss[loss=0.2378, simple_loss=0.3223, pruned_loss=0.07664, over 28933.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3306, pruned_loss=0.08527, over 5672906.09 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3306, pruned_loss=0.08762, over 5743308.82 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3302, pruned_loss=0.08465, over 5671863.32 frames. ], batch size: 106, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:37:58,787 INFO [train.py:968] (0/2) Epoch 22, batch 16050, giga_loss[loss=0.2614, simple_loss=0.341, pruned_loss=0.09093, over 28704.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3311, pruned_loss=0.08622, over 5659364.56 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3307, pruned_loss=0.08786, over 5729039.97 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3307, pruned_loss=0.08539, over 5667324.10 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:38:06,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.873e+02 1.327e+03 1.719e+03 2.512e+03 7.167e+03, threshold=3.438e+03, percent-clipped=7.0
+2023-03-11 09:38:22,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3279, 3.4917, 1.3553, 1.6432], device='cuda:0'), covar=tensor([0.1019, 0.0448, 0.0967, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0551, 0.0386, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 09:38:55,470 INFO [train.py:968] (0/2) Epoch 22, batch 16100, giga_loss[loss=0.2832, simple_loss=0.3723, pruned_loss=0.09711, over 28688.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3342, pruned_loss=0.08804, over 5669692.86 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3299, pruned_loss=0.08755, over 5733349.89 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3346, pruned_loss=0.08762, over 5670857.27 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:39:48,563 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=973879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:39:50,556 INFO [train.py:968] (0/2) Epoch 22, batch 16150, giga_loss[loss=0.3016, simple_loss=0.3581, pruned_loss=0.1226, over 26920.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3361, pruned_loss=0.08806, over 5678038.30 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3297, pruned_loss=0.08745, over 5736509.69 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3367, pruned_loss=0.08783, over 5674724.15 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:39:59,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.805e+02 1.471e+03 1.847e+03 2.708e+03 7.287e+03, threshold=3.693e+03, percent-clipped=12.0
+2023-03-11 09:40:55,416 INFO [train.py:968] (0/2) Epoch 22, batch 16200, giga_loss[loss=0.2511, simple_loss=0.3371, pruned_loss=0.08252, over 28977.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3368, pruned_loss=0.08807, over 5679513.29 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3296, pruned_loss=0.08742, over 5737479.14 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3374, pruned_loss=0.08792, over 5675732.89 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:41:50,323 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=973969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:42:07,631 INFO [train.py:968] (0/2) Epoch 22, batch 16250, giga_loss[loss=0.2887, simple_loss=0.3519, pruned_loss=0.1128, over 29062.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3351, pruned_loss=0.08717, over 5682250.17 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3297, pruned_loss=0.08744, over 5735343.92 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3356, pruned_loss=0.08704, over 5680769.48 frames. ], batch size: 200, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:42:17,888 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 09:42:18,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.460e+03 1.938e+03 2.582e+03 6.996e+03, threshold=3.875e+03, percent-clipped=9.0
+2023-03-11 09:42:29,207 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-974000.pt
+2023-03-11 09:42:45,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6283, 1.8930, 1.3250, 1.4485], device='cuda:0'), covar=tensor([0.1007, 0.0576, 0.1047, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0437, 0.0512, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 09:43:08,155 INFO [train.py:968] (0/2) Epoch 22, batch 16300, giga_loss[loss=0.2508, simple_loss=0.3283, pruned_loss=0.08667, over 28120.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3336, pruned_loss=0.08684, over 5682849.63 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3291, pruned_loss=0.08717, over 5737975.13 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3347, pruned_loss=0.08699, over 5678075.67 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:44:13,379 INFO [train.py:968] (0/2) Epoch 22, batch 16350, giga_loss[loss=0.2541, simple_loss=0.3319, pruned_loss=0.08816, over 29000.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3319, pruned_loss=0.0861, over 5673963.03 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3292, pruned_loss=0.08727, over 5741153.15 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3327, pruned_loss=0.08609, over 5666156.49 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:44:21,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.871e+02 1.366e+03 1.787e+03 2.994e+03 1.181e+04, threshold=3.575e+03, percent-clipped=13.0
+2023-03-11 09:44:46,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=974109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:44:50,067 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=974112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:44:53,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=974115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:45:14,201 INFO [train.py:968] (0/2) Epoch 22, batch 16400, giga_loss[loss=0.2674, simple_loss=0.3443, pruned_loss=0.09519, over 28514.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3299, pruned_loss=0.0863, over 5673450.93 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.329, pruned_loss=0.08711, over 5742536.79 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3307, pruned_loss=0.08642, over 5665282.76 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:45:21,232 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-11 09:45:28,197 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=974144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:45:45,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5224, 1.9064, 1.6864, 1.5542], device='cuda:0'), covar=tensor([0.1978, 0.2282, 0.2029, 0.2258], device='cuda:0'), in_proj_covar=tensor([0.0458, 0.0729, 0.0695, 0.0668], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 09:46:01,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5651, 1.8866, 1.2702, 1.3575], device='cuda:0'), covar=tensor([0.1048, 0.0578, 0.1086, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0437, 0.0512, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 09:46:10,061 INFO [train.py:968] (0/2) Epoch 22, batch 16450, giga_loss[loss=0.2383, simple_loss=0.329, pruned_loss=0.07384, over 28887.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3301, pruned_loss=0.08639, over 5674728.25 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3291, pruned_loss=0.08732, over 5737922.50 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3307, pruned_loss=0.08627, over 5669545.80 frames. ], batch size: 284, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:46:20,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.562e+02 1.297e+03 1.635e+03 2.271e+03 6.104e+03, threshold=3.270e+03, percent-clipped=4.0
+2023-03-11 09:46:53,689 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2887, 1.9325, 1.4300, 0.4940], device='cuda:0'), covar=tensor([0.5303, 0.2885, 0.4604, 0.6373], device='cuda:0'), in_proj_covar=tensor([0.1754, 0.1650, 0.1595, 0.1426], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 09:47:05,485 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7906, 2.3643, 1.4086, 1.0920], device='cuda:0'), covar=tensor([0.7393, 0.4154, 0.4705, 0.6445], device='cuda:0'), in_proj_covar=tensor([0.1755, 0.1650, 0.1595, 0.1427], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 09:47:15,196 INFO [train.py:968] (0/2) Epoch 22, batch 16500, giga_loss[loss=0.2291, simple_loss=0.3109, pruned_loss=0.07364, over 28511.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3295, pruned_loss=0.08496, over 5666631.69 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3293, pruned_loss=0.08739, over 5736703.89 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3299, pruned_loss=0.08479, over 5663368.68 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:47:42,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=974254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:48:10,331 INFO [train.py:968] (0/2) Epoch 22, batch 16550, giga_loss[loss=0.2492, simple_loss=0.3447, pruned_loss=0.07683, over 28760.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3305, pruned_loss=0.0838, over 5679728.86 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3292, pruned_loss=0.0873, over 5740013.01 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3308, pruned_loss=0.08369, over 5672830.43 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:48:21,619 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.667e+02 1.418e+03 1.979e+03 2.475e+03 5.166e+03, threshold=3.958e+03, percent-clipped=12.0
+2023-03-11 09:48:28,484 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-11 09:49:08,831 INFO [train.py:968] (0/2) Epoch 22, batch 16600, giga_loss[loss=0.2338, simple_loss=0.3244, pruned_loss=0.07165, over 28868.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3324, pruned_loss=0.08367, over 5664461.94 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3293, pruned_loss=0.08739, over 5730070.68 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3326, pruned_loss=0.08346, over 5667431.94 frames. ], batch size: 284, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:50:04,284 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-11 09:50:08,419 INFO [train.py:968] (0/2) Epoch 22, batch 16650, giga_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08885, over 28484.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3334, pruned_loss=0.08378, over 5679809.00 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3293, pruned_loss=0.08742, over 5732377.41 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3336, pruned_loss=0.08354, over 5679221.67 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:50:19,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.847e+02 1.430e+03 1.862e+03 2.698e+03 7.176e+03, threshold=3.724e+03, percent-clipped=11.0
+2023-03-11 09:50:28,772 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=974397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:50:32,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=974400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:50:38,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5141, 1.1545, 4.2892, 3.3060], device='cuda:0'), covar=tensor([0.1575, 0.2798, 0.0419, 0.0985], device='cuda:0'), in_proj_covar=tensor([0.0756, 0.0647, 0.0948, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 09:51:01,166 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 09:51:11,085 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=974429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:51:12,401 INFO [train.py:968] (0/2) Epoch 22, batch 16700, giga_loss[loss=0.2653, simple_loss=0.3266, pruned_loss=0.102, over 24328.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3335, pruned_loss=0.08455, over 5664798.00 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3292, pruned_loss=0.08747, over 5726876.74 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3339, pruned_loss=0.08423, over 5667771.09 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:51:44,934 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.55 vs. limit=5.0
+2023-03-11 09:52:16,097 INFO [train.py:968] (0/2) Epoch 22, batch 16750, giga_loss[loss=0.2447, simple_loss=0.3363, pruned_loss=0.07658, over 28673.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3325, pruned_loss=0.08374, over 5672636.46 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3291, pruned_loss=0.08741, over 5732249.39 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.333, pruned_loss=0.08344, over 5668497.23 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:52:21,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=974484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:52:27,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.770e+02 1.375e+03 1.855e+03 2.542e+03 6.360e+03, threshold=3.711e+03, percent-clipped=10.0
+2023-03-11 09:52:34,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4765, 1.5760, 1.6945, 1.3201], device='cuda:0'), covar=tensor([0.1519, 0.2379, 0.1299, 0.1663], device='cuda:0'), in_proj_covar=tensor([0.0896, 0.0694, 0.0946, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 09:53:26,101 INFO [train.py:968] (0/2) Epoch 22, batch 16800, giga_loss[loss=0.2189, simple_loss=0.3158, pruned_loss=0.06095, over 29174.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3321, pruned_loss=0.08258, over 5664563.78 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3292, pruned_loss=0.08742, over 5724047.05 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3324, pruned_loss=0.08228, over 5668009.11 frames. ], batch size: 200, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:53:34,011 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=974536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:54:39,069 INFO [train.py:968] (0/2) Epoch 22, batch 16850, giga_loss[loss=0.2826, simple_loss=0.3761, pruned_loss=0.09454, over 28432.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3332, pruned_loss=0.08291, over 5672967.93 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3289, pruned_loss=0.08721, over 5727617.42 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3337, pruned_loss=0.08276, over 5671471.82 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:54:52,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.060e+02 1.442e+03 1.830e+03 2.390e+03 4.577e+03, threshold=3.661e+03, percent-clipped=2.0
+2023-03-11 09:55:23,383 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3475, 1.6742, 1.6908, 1.3676], device='cuda:0'), covar=tensor([0.3009, 0.1950, 0.2130, 0.2370], device='cuda:0'), in_proj_covar=tensor([0.1927, 0.1850, 0.1756, 0.1915], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 09:55:40,824 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=974627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:55:45,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=974630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:55:46,142 INFO [train.py:968] (0/2) Epoch 22, batch 16900, giga_loss[loss=0.2635, simple_loss=0.3539, pruned_loss=0.08658, over 28987.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3358, pruned_loss=0.08415, over 5680944.70 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.328, pruned_loss=0.08667, over 5734208.17 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3373, pruned_loss=0.08443, over 5672075.54 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:56:08,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4943, 1.8038, 1.3610, 1.3923], device='cuda:0'), covar=tensor([0.1019, 0.0596, 0.1068, 0.1138], device='cuda:0'), in_proj_covar=tensor([0.0389, 0.0436, 0.0511, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 09:56:26,454 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=974659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:56:53,914 INFO [train.py:968] (0/2) Epoch 22, batch 16950, giga_loss[loss=0.2443, simple_loss=0.3305, pruned_loss=0.07904, over 28361.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3347, pruned_loss=0.08335, over 5684160.30 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3279, pruned_loss=0.08657, over 5735920.16 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3361, pruned_loss=0.08363, over 5674721.72 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:57:06,998 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.315e+02 1.446e+03 1.758e+03 2.491e+03 7.149e+03, threshold=3.517e+03, percent-clipped=5.0
+2023-03-11 09:58:06,672 INFO [train.py:968] (0/2) Epoch 22, batch 17000, libri_loss[loss=0.206, simple_loss=0.2889, pruned_loss=0.06155, over 29549.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3327, pruned_loss=0.08299, over 5687664.01 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3281, pruned_loss=0.08671, over 5730852.53 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3339, pruned_loss=0.08304, over 5683407.21 frames. ], batch size: 77, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:59:14,256 INFO [train.py:968] (0/2) Epoch 22, batch 17050, giga_loss[loss=0.201, simple_loss=0.291, pruned_loss=0.05552, over 28687.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3307, pruned_loss=0.08136, over 5689835.29 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.328, pruned_loss=0.08663, over 5730368.90 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3317, pruned_loss=0.08136, over 5685673.92 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:59:30,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.588e+02 1.364e+03 1.806e+03 2.540e+03 8.983e+03, threshold=3.613e+03, percent-clipped=10.0
+2023-03-11 09:59:32,554 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=974794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:00:22,141 INFO [train.py:968] (0/2) Epoch 22, batch 17100, giga_loss[loss=0.2584, simple_loss=0.3443, pruned_loss=0.08627, over 28805.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3301, pruned_loss=0.08095, over 5698973.24 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3281, pruned_loss=0.08679, over 5734874.14 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3309, pruned_loss=0.08068, over 5690933.78 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:00:52,256 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=974855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:01:10,477 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=974870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:01:23,826 INFO [train.py:968] (0/2) Epoch 22, batch 17150, giga_loss[loss=0.2432, simple_loss=0.333, pruned_loss=0.07666, over 28441.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3311, pruned_loss=0.08198, over 5680276.98 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3282, pruned_loss=0.08678, over 5725901.58 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3317, pruned_loss=0.0817, over 5680848.92 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:01:35,265 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.305e+02 1.391e+03 1.879e+03 2.573e+03 4.887e+03, threshold=3.758e+03, percent-clipped=9.0
+2023-03-11 10:01:54,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=974911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:02:15,150 INFO [train.py:968] (0/2) Epoch 22, batch 17200, giga_loss[loss=0.2764, simple_loss=0.3524, pruned_loss=0.1002, over 28898.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3336, pruned_loss=0.08392, over 5689809.09 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3277, pruned_loss=0.08663, over 5734134.91 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3347, pruned_loss=0.08364, over 5680803.67 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:02:40,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3939, 1.7204, 1.3692, 1.3360], device='cuda:0'), covar=tensor([0.2639, 0.2491, 0.2887, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.1503, 0.1086, 0.1331, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 10:03:11,435 INFO [train.py:968] (0/2) Epoch 22, batch 17250, giga_loss[loss=0.2367, simple_loss=0.3175, pruned_loss=0.07796, over 28331.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3342, pruned_loss=0.08456, over 5686700.37 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.328, pruned_loss=0.08679, over 5736918.83 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3348, pruned_loss=0.08415, over 5676415.26 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:03:23,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.651e+02 1.547e+03 2.097e+03 3.248e+03 1.382e+04, threshold=4.194e+03, percent-clipped=18.0
+2023-03-11 10:04:03,347 INFO [train.py:968] (0/2) Epoch 22, batch 17300, giga_loss[loss=0.2448, simple_loss=0.3354, pruned_loss=0.07709, over 28403.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3315, pruned_loss=0.08441, over 5680806.26 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3277, pruned_loss=0.08671, over 5732787.65 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3325, pruned_loss=0.08406, over 5675167.92 frames. ], batch size: 65, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:04:33,551 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=975054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:04:36,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=975057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:04:38,309 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=975059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:05:02,636 INFO [train.py:968] (0/2) Epoch 22, batch 17350, giga_loss[loss=0.2303, simple_loss=0.3092, pruned_loss=0.07565, over 28712.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3309, pruned_loss=0.08468, over 5685168.19 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3274, pruned_loss=0.08656, over 5736109.83 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.332, pruned_loss=0.08449, over 5676691.08 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:05:10,135 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=975086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:05:15,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.461e+02 1.558e+03 1.911e+03 2.397e+03 5.341e+03, threshold=3.823e+03, percent-clipped=5.0
+2023-03-11 10:05:55,764 INFO [train.py:968] (0/2) Epoch 22, batch 17400, giga_loss[loss=0.3784, simple_loss=0.4234, pruned_loss=0.1667, over 27594.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3354, pruned_loss=0.0875, over 5689200.55 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3272, pruned_loss=0.0864, over 5741905.26 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3367, pruned_loss=0.08746, over 5675399.73 frames. ], batch size: 472, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:06:12,787 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-11 10:06:31,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4376, 1.7243, 1.6776, 1.2580], device='cuda:0'), covar=tensor([0.1898, 0.2573, 0.1611, 0.1858], device='cuda:0'), in_proj_covar=tensor([0.0896, 0.0692, 0.0946, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 10:06:34,597 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=975169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:06:45,684 INFO [train.py:968] (0/2) Epoch 22, batch 17450, giga_loss[loss=0.2913, simple_loss=0.3719, pruned_loss=0.1054, over 28607.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3448, pruned_loss=0.09292, over 5694594.02 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3275, pruned_loss=0.08659, over 5744690.26 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3458, pruned_loss=0.09283, over 5680163.57 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:06:50,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3812, 1.3503, 1.2511, 1.4048], device='cuda:0'), covar=tensor([0.0803, 0.0364, 0.0357, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0117, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 10:06:54,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.639e+02 1.375e+03 1.703e+03 2.213e+03 4.382e+03, threshold=3.406e+03, percent-clipped=3.0
+2023-03-11 10:07:26,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=975230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:07:27,492 INFO [train.py:968] (0/2) Epoch 22, batch 17500, giga_loss[loss=0.2624, simple_loss=0.3375, pruned_loss=0.0937, over 28904.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3491, pruned_loss=0.09532, over 5695734.83 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3273, pruned_loss=0.08659, over 5738357.68 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3506, pruned_loss=0.09543, over 5688501.70 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:07:43,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=975245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:07:47,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-11 10:07:53,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2079, 2.5034, 1.2158, 1.4135], device='cuda:0'), covar=tensor([0.1067, 0.0356, 0.0960, 0.1449], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0550, 0.0385, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 10:08:13,308 INFO [train.py:968] (0/2) Epoch 22, batch 17550, giga_loss[loss=0.2536, simple_loss=0.3328, pruned_loss=0.08718, over 29027.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3453, pruned_loss=0.0943, over 5696103.53 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3271, pruned_loss=0.08639, over 5741939.47 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3471, pruned_loss=0.09473, over 5686328.75 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:08:24,569 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.807e+02 1.249e+03 1.577e+03 2.207e+03 5.804e+03, threshold=3.155e+03, percent-clipped=6.0
+2023-03-11 10:08:40,315 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=975312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:08:42,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=975315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:08:54,083 INFO [train.py:968] (0/2) Epoch 22, batch 17600, giga_loss[loss=0.2498, simple_loss=0.3275, pruned_loss=0.08605, over 28308.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3391, pruned_loss=0.09188, over 5697025.62 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3276, pruned_loss=0.08656, over 5746796.70 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3406, pruned_loss=0.0923, over 5683204.47 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:09:07,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=975344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:09:36,253 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=975373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:09:38,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=975376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:09:41,679 INFO [train.py:968] (0/2) Epoch 22, batch 17650, giga_loss[loss=0.1901, simple_loss=0.2657, pruned_loss=0.05726, over 28477.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3314, pruned_loss=0.08848, over 5693218.49 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3274, pruned_loss=0.08642, over 5748540.13 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3328, pruned_loss=0.08896, over 5680151.17 frames. ], batch size: 60, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:09:48,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=975388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:09:50,121 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=975391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:09:50,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.349e+02 1.052e+03 1.283e+03 1.637e+03 3.428e+03, threshold=2.566e+03, percent-clipped=2.0
+2023-03-11 10:10:03,835 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=975405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:10:17,583 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=975420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:10:25,550 INFO [train.py:968] (0/2) Epoch 22, batch 17700, libri_loss[loss=0.2666, simple_loss=0.3583, pruned_loss=0.08744, over 29126.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3239, pruned_loss=0.08513, over 5689155.65 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3278, pruned_loss=0.08642, over 5748419.50 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3246, pruned_loss=0.08549, over 5677432.66 frames. ], batch size: 101, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:10:30,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=975434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:10:43,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-11 10:11:05,452 INFO [train.py:968] (0/2) Epoch 22, batch 17750, giga_loss[loss=0.2103, simple_loss=0.2807, pruned_loss=0.06991, over 28756.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3188, pruned_loss=0.08253, over 5690981.29 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3285, pruned_loss=0.08645, over 5743545.35 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.3184, pruned_loss=0.08267, over 5683742.41 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:11:15,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.316e+02 1.069e+03 1.353e+03 1.656e+03 5.502e+03, threshold=2.705e+03, percent-clipped=8.0
+2023-03-11 10:11:44,129 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5294, 1.6564, 1.2743, 1.2631], device='cuda:0'), covar=tensor([0.0986, 0.0609, 0.1050, 0.1305], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0437, 0.0512, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 10:11:45,290 INFO [train.py:968] (0/2) Epoch 22, batch 17800, giga_loss[loss=0.2089, simple_loss=0.2881, pruned_loss=0.06488, over 29007.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3136, pruned_loss=0.08022, over 5694992.12 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3285, pruned_loss=0.08643, over 5746150.17 frames. ], giga_tot_loss[loss=0.2368, simple_loss=0.313, pruned_loss=0.08026, over 5686169.31 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:12:24,845 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=975577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:12:26,972 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=975580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:12:28,220 INFO [train.py:968] (0/2) Epoch 22, batch 17850, giga_loss[loss=0.207, simple_loss=0.2862, pruned_loss=0.06394, over 28923.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.3108, pruned_loss=0.079, over 5701551.85 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3283, pruned_loss=0.08621, over 5747844.31 frames. ], giga_tot_loss[loss=0.2343, simple_loss=0.3103, pruned_loss=0.07914, over 5692420.84 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:12:41,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.877e+02 1.043e+03 1.391e+03 1.966e+03 7.849e+03, threshold=2.781e+03, percent-clipped=13.0
+2023-03-11 10:12:55,733 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=975609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:13:02,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9258, 3.7356, 3.5098, 1.4855], device='cuda:0'), covar=tensor([0.0669, 0.0847, 0.0804, 0.2354], device='cuda:0'), in_proj_covar=tensor([0.1217, 0.1118, 0.0946, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 10:13:04,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3243, 1.5371, 1.4594, 1.2538], device='cuda:0'), covar=tensor([0.2903, 0.2791, 0.1841, 0.2495], device='cuda:0'), in_proj_covar=tensor([0.1940, 0.1861, 0.1767, 0.1932], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 10:13:13,793 INFO [train.py:968] (0/2) Epoch 22, batch 17900, giga_loss[loss=0.2254, simple_loss=0.2986, pruned_loss=0.07609, over 28665.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3074, pruned_loss=0.07755, over 5688294.52 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3286, pruned_loss=0.08632, over 5739788.86 frames. ], giga_tot_loss[loss=0.2307, simple_loss=0.3065, pruned_loss=0.07748, over 5687174.27 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:13:52,624 INFO [train.py:968] (0/2) Epoch 22, batch 17950, giga_loss[loss=0.2111, simple_loss=0.2839, pruned_loss=0.06914, over 28801.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3062, pruned_loss=0.07674, over 5701780.55 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3294, pruned_loss=0.08642, over 5741166.49 frames. ], giga_tot_loss[loss=0.2282, simple_loss=0.3039, pruned_loss=0.07625, over 5698016.01 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:14:04,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.974e+02 1.114e+03 1.389e+03 1.893e+03 1.093e+04, threshold=2.778e+03, percent-clipped=12.0
+2023-03-11 10:14:36,964 INFO [train.py:968] (0/2) Epoch 22, batch 18000, giga_loss[loss=0.218, simple_loss=0.2947, pruned_loss=0.07068, over 28823.00 frames. ], tot_loss[loss=0.228, simple_loss=0.304, pruned_loss=0.07599, over 5691334.14 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3297, pruned_loss=0.08642, over 5745207.00 frames. ], giga_tot_loss[loss=0.2259, simple_loss=0.3012, pruned_loss=0.07531, over 5683587.06 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:14:36,968 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 10:14:45,417 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2035, simple_loss=0.3091, pruned_loss=0.04889, over 944034.00 frames. 
+2023-03-11 10:14:45,417 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 10:15:28,540 INFO [train.py:968] (0/2) Epoch 22, batch 18050, giga_loss[loss=0.2365, simple_loss=0.3142, pruned_loss=0.07939, over 28553.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3016, pruned_loss=0.07537, over 5685110.07 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.33, pruned_loss=0.08644, over 5737636.25 frames. ], giga_tot_loss[loss=0.2241, simple_loss=0.2989, pruned_loss=0.07467, over 5685749.71 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:15:40,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.523e+02 1.062e+03 1.264e+03 1.786e+03 7.812e+03, threshold=2.528e+03, percent-clipped=7.0
+2023-03-11 10:15:52,971 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5101, 4.2217, 1.6273, 1.7797], device='cuda:0'), covar=tensor([0.0992, 0.0296, 0.0920, 0.1249], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0550, 0.0385, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 10:16:11,023 INFO [train.py:968] (0/2) Epoch 22, batch 18100, libri_loss[loss=0.2726, simple_loss=0.3669, pruned_loss=0.0891, over 29296.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.2989, pruned_loss=0.07439, over 5678854.83 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3305, pruned_loss=0.08652, over 5730919.45 frames. ], giga_tot_loss[loss=0.2213, simple_loss=0.2956, pruned_loss=0.07351, over 5683453.93 frames. ], batch size: 97, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:16:58,219 INFO [train.py:968] (0/2) Epoch 22, batch 18150, giga_loss[loss=0.2098, simple_loss=0.2836, pruned_loss=0.06802, over 27889.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2961, pruned_loss=0.07332, over 5676379.66 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3304, pruned_loss=0.08649, over 5734201.05 frames. ], giga_tot_loss[loss=0.2188, simple_loss=0.2928, pruned_loss=0.0724, over 5676005.17 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:17:09,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1958, 3.0330, 2.8520, 1.3954], device='cuda:0'), covar=tensor([0.1001, 0.1106, 0.0946, 0.2492], device='cuda:0'), in_proj_covar=tensor([0.1216, 0.1122, 0.0949, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 10:17:10,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.126e+02 1.036e+03 1.363e+03 1.922e+03 4.880e+03, threshold=2.727e+03, percent-clipped=9.0
+2023-03-11 10:17:43,093 INFO [train.py:968] (0/2) Epoch 22, batch 18200, giga_loss[loss=0.2086, simple_loss=0.284, pruned_loss=0.06663, over 28792.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2937, pruned_loss=0.07254, over 5680745.17 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3299, pruned_loss=0.08621, over 5736559.46 frames. ], giga_tot_loss[loss=0.2174, simple_loss=0.2911, pruned_loss=0.07187, over 5677546.32 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:18:10,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8643, 2.9891, 2.1341, 2.6470], device='cuda:0'), covar=tensor([0.0782, 0.0567, 0.0978, 0.0964], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0440, 0.0516, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 10:18:34,709 INFO [train.py:968] (0/2) Epoch 22, batch 18250, giga_loss[loss=0.2805, simple_loss=0.3554, pruned_loss=0.1028, over 28625.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3035, pruned_loss=0.07786, over 5663217.53 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3301, pruned_loss=0.08638, over 5725907.78 frames. ], giga_tot_loss[loss=0.2275, simple_loss=0.3009, pruned_loss=0.07704, over 5670033.12 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:18:45,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.523e+02 1.168e+03 1.477e+03 2.167e+03 2.347e+04, threshold=2.953e+03, percent-clipped=16.0
+2023-03-11 10:18:50,400 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-976000.pt
+2023-03-11 10:19:19,558 INFO [train.py:968] (0/2) Epoch 22, batch 18300, giga_loss[loss=0.3344, simple_loss=0.4059, pruned_loss=0.1315, over 28979.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3169, pruned_loss=0.08463, over 5671907.21 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3299, pruned_loss=0.08619, over 5726826.67 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.3144, pruned_loss=0.08401, over 5675036.90 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:19:58,243 INFO [train.py:968] (0/2) Epoch 22, batch 18350, giga_loss[loss=0.2995, simple_loss=0.3739, pruned_loss=0.1126, over 28964.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3285, pruned_loss=0.0904, over 5677179.14 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3303, pruned_loss=0.08628, over 5718116.45 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.326, pruned_loss=0.08988, over 5686589.44 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:20:09,657 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.441e+02 1.379e+03 1.745e+03 2.313e+03 4.958e+03, threshold=3.490e+03, percent-clipped=11.0
+2023-03-11 10:20:26,807 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:20:40,273 INFO [train.py:968] (0/2) Epoch 22, batch 18400, giga_loss[loss=0.259, simple_loss=0.3404, pruned_loss=0.08879, over 28924.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3338, pruned_loss=0.09192, over 5677356.07 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3301, pruned_loss=0.08605, over 5721125.65 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.332, pruned_loss=0.09184, over 5681047.38 frames. ], batch size: 106, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:21:22,506 INFO [train.py:968] (0/2) Epoch 22, batch 18450, giga_loss[loss=0.2547, simple_loss=0.3405, pruned_loss=0.08443, over 28672.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3366, pruned_loss=0.09192, over 5680764.94 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3303, pruned_loss=0.08606, over 5722394.88 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3351, pruned_loss=0.0919, over 5682214.57 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:21:33,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.590e+02 1.241e+03 1.577e+03 1.951e+03 6.984e+03, threshold=3.154e+03, percent-clipped=1.0
+2023-03-11 10:21:47,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4716, 1.4212, 1.1781, 1.5844], device='cuda:0'), covar=tensor([0.0770, 0.0351, 0.0372, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 10:22:06,931 INFO [train.py:968] (0/2) Epoch 22, batch 18500, giga_loss[loss=0.2635, simple_loss=0.3257, pruned_loss=0.1007, over 23698.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3384, pruned_loss=0.09192, over 5673473.01 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3307, pruned_loss=0.08615, over 5726615.75 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.337, pruned_loss=0.09198, over 5669662.34 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:22:48,013 INFO [train.py:968] (0/2) Epoch 22, batch 18550, libri_loss[loss=0.25, simple_loss=0.3411, pruned_loss=0.07945, over 29524.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.341, pruned_loss=0.09378, over 5677422.30 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3309, pruned_loss=0.08594, over 5728880.49 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.34, pruned_loss=0.09434, over 5669871.95 frames. ], batch size: 82, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:23:01,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.547e+02 1.270e+03 1.528e+03 1.903e+03 6.300e+03, threshold=3.056e+03, percent-clipped=10.0
+2023-03-11 10:23:15,917 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5546, 3.3904, 1.6147, 1.7705], device='cuda:0'), covar=tensor([0.1035, 0.0269, 0.0893, 0.1304], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0549, 0.0384, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 10:23:30,815 INFO [train.py:968] (0/2) Epoch 22, batch 18600, giga_loss[loss=0.2743, simple_loss=0.3599, pruned_loss=0.09437, over 28402.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3434, pruned_loss=0.09566, over 5673028.11 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3312, pruned_loss=0.08597, over 5724478.77 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3427, pruned_loss=0.0963, over 5670213.43 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:23:42,032 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:24:13,140 INFO [train.py:968] (0/2) Epoch 22, batch 18650, giga_loss[loss=0.3117, simple_loss=0.3872, pruned_loss=0.1181, over 28751.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3467, pruned_loss=0.09777, over 5665822.84 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3316, pruned_loss=0.0862, over 5719438.34 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3461, pruned_loss=0.09831, over 5667652.58 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:24:25,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.298e+02 1.275e+03 1.544e+03 2.067e+03 8.054e+03, threshold=3.089e+03, percent-clipped=6.0
+2023-03-11 10:24:33,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8577, 2.0398, 1.6818, 2.0265], device='cuda:0'), covar=tensor([0.2465, 0.2571, 0.2849, 0.2363], device='cuda:0'), in_proj_covar=tensor([0.1508, 0.1091, 0.1333, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 10:24:54,454 INFO [train.py:968] (0/2) Epoch 22, batch 18700, giga_loss[loss=0.3089, simple_loss=0.3782, pruned_loss=0.1198, over 29072.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3498, pruned_loss=0.09886, over 5673079.63 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.332, pruned_loss=0.08631, over 5720625.19 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3494, pruned_loss=0.09952, over 5672294.75 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:24:58,524 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976435.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 10:25:00,511 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8928, 1.1765, 2.8209, 2.6567], device='cuda:0'), covar=tensor([0.1652, 0.2675, 0.0591, 0.1481], device='cuda:0'), in_proj_covar=tensor([0.0755, 0.0647, 0.0955, 0.0899], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 10:25:14,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3612, 1.5966, 1.2027, 1.1173], device='cuda:0'), covar=tensor([0.1015, 0.0530, 0.1079, 0.1138], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0440, 0.0517, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 10:25:26,882 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-03-11 10:25:32,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5371, 1.5608, 1.5777, 1.4198], device='cuda:0'), covar=tensor([0.2601, 0.2809, 0.2132, 0.2622], device='cuda:0'), in_proj_covar=tensor([0.1948, 0.1872, 0.1787, 0.1944], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 10:25:37,326 INFO [train.py:968] (0/2) Epoch 22, batch 18750, giga_loss[loss=0.2787, simple_loss=0.3605, pruned_loss=0.09839, over 28531.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3526, pruned_loss=0.09976, over 5674802.34 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3321, pruned_loss=0.08633, over 5722209.18 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3524, pruned_loss=0.1004, over 5672215.58 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:25:45,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:25:48,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.786e+02 1.250e+03 1.725e+03 2.208e+03 4.398e+03, threshold=3.450e+03, percent-clipped=4.0
+2023-03-11 10:25:53,711 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5893, 1.8080, 1.4617, 1.6089], device='cuda:0'), covar=tensor([0.2787, 0.2691, 0.3107, 0.2250], device='cuda:0'), in_proj_covar=tensor([0.1511, 0.1094, 0.1336, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 10:26:03,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4823, 2.6236, 2.0352, 2.0786], device='cuda:0'), covar=tensor([0.0930, 0.0676, 0.1001, 0.1145], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0442, 0.0519, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 10:26:15,732 INFO [train.py:968] (0/2) Epoch 22, batch 18800, giga_loss[loss=0.3158, simple_loss=0.3948, pruned_loss=0.1184, over 28787.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3534, pruned_loss=0.09919, over 5689720.87 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3322, pruned_loss=0.08631, over 5726108.78 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3536, pruned_loss=0.1, over 5683399.83 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:26:29,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:26:32,808 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-11 10:26:51,744 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:26:53,670 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:26:54,373 INFO [train.py:968] (0/2) Epoch 22, batch 18850, giga_loss[loss=0.2655, simple_loss=0.3525, pruned_loss=0.08924, over 28869.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3538, pruned_loss=0.09825, over 5695558.37 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3323, pruned_loss=0.0861, over 5729183.51 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3548, pruned_loss=0.09969, over 5686125.06 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:26:55,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5495, 5.3123, 5.0215, 2.5338], device='cuda:0'), covar=tensor([0.0391, 0.0570, 0.0623, 0.1870], device='cuda:0'), in_proj_covar=tensor([0.1205, 0.1118, 0.0944, 0.0713], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 10:27:02,227 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-11 10:27:04,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.512e+02 1.233e+03 1.552e+03 2.139e+03 6.475e+03, threshold=3.103e+03, percent-clipped=7.0
+2023-03-11 10:27:05,843 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-11 10:27:06,643 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 10:27:13,148 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 10:27:34,521 INFO [train.py:968] (0/2) Epoch 22, batch 18900, giga_loss[loss=0.2496, simple_loss=0.3453, pruned_loss=0.07698, over 28438.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.353, pruned_loss=0.09682, over 5704401.48 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3329, pruned_loss=0.08641, over 5732188.65 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3538, pruned_loss=0.09789, over 5693868.01 frames. ], batch size: 60, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:27:36,523 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=976634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:27:39,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=976637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:27:44,166 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:28:00,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=976666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:28:08,179 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5625, 1.8130, 1.5319, 1.5353], device='cuda:0'), covar=tensor([0.2153, 0.2231, 0.2318, 0.2210], device='cuda:0'), in_proj_covar=tensor([0.1508, 0.1092, 0.1333, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 10:28:14,982 INFO [train.py:968] (0/2) Epoch 22, batch 18950, giga_loss[loss=0.2904, simple_loss=0.364, pruned_loss=0.1084, over 28716.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3516, pruned_loss=0.09576, over 5707756.10 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3332, pruned_loss=0.08642, over 5736090.49 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3523, pruned_loss=0.09684, over 5695327.90 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:28:18,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9382, 3.7135, 3.5499, 1.6641], device='cuda:0'), covar=tensor([0.0707, 0.0915, 0.0818, 0.2336], device='cuda:0'), in_proj_covar=tensor([0.1204, 0.1117, 0.0942, 0.0712], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 10:28:25,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.309e+02 1.153e+03 1.376e+03 1.765e+03 5.108e+03, threshold=2.751e+03, percent-clipped=2.0
+2023-03-11 10:28:46,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:28:58,188 INFO [train.py:968] (0/2) Epoch 22, batch 19000, giga_loss[loss=0.2955, simple_loss=0.3578, pruned_loss=0.1166, over 28899.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3535, pruned_loss=0.09839, over 5713262.59 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.333, pruned_loss=0.0863, over 5737975.60 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3545, pruned_loss=0.09944, over 5701650.85 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:29:43,317 INFO [train.py:968] (0/2) Epoch 22, batch 19050, giga_loss[loss=0.2691, simple_loss=0.3434, pruned_loss=0.09739, over 29057.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3565, pruned_loss=0.1033, over 5715191.47 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3334, pruned_loss=0.08633, over 5741058.22 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3574, pruned_loss=0.1044, over 5702723.91 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:29:56,604 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.747e+02 1.381e+03 1.582e+03 2.293e+03 5.267e+03, threshold=3.165e+03, percent-clipped=11.0
+2023-03-11 10:29:59,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1882, 2.3247, 1.3207, 1.2917], device='cuda:0'), covar=tensor([0.0962, 0.0403, 0.0814, 0.1370], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0548, 0.0384, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 10:30:06,761 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 10:30:07,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976810.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 10:30:21,879 INFO [train.py:968] (0/2) Epoch 22, batch 19100, giga_loss[loss=0.2616, simple_loss=0.3387, pruned_loss=0.09228, over 28783.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3566, pruned_loss=0.1047, over 5715553.51 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3336, pruned_loss=0.08646, over 5741640.72 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3573, pruned_loss=0.1057, over 5704843.55 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:30:47,345 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=976861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:30:49,266 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=976864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:31:02,510 INFO [train.py:968] (0/2) Epoch 22, batch 19150, giga_loss[loss=0.2634, simple_loss=0.3381, pruned_loss=0.09436, over 28192.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3544, pruned_loss=0.1041, over 5713168.36 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3337, pruned_loss=0.08643, over 5745616.14 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3554, pruned_loss=0.1054, over 5700474.35 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:31:13,255 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=976893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:31:17,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.394e+03 1.681e+03 2.199e+03 4.974e+03, threshold=3.363e+03, percent-clipped=10.0
+2023-03-11 10:31:39,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:31:46,662 INFO [train.py:968] (0/2) Epoch 22, batch 19200, libri_loss[loss=0.2187, simple_loss=0.2985, pruned_loss=0.06948, over 29657.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3516, pruned_loss=0.1026, over 5713411.83 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3335, pruned_loss=0.08621, over 5749697.78 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3531, pruned_loss=0.1042, over 5698659.87 frames. ], batch size: 69, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:31:53,055 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-11 10:32:06,506 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:32:07,184 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=976953.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 10:32:08,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:32:09,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=976956.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 10:32:28,831 INFO [train.py:968] (0/2) Epoch 22, batch 19250, giga_loss[loss=0.2844, simple_loss=0.3569, pruned_loss=0.1059, over 28620.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3503, pruned_loss=0.1007, over 5721095.49 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3338, pruned_loss=0.08622, over 5753117.11 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3516, pruned_loss=0.1023, over 5705570.89 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:32:31,632 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=976985.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 10:32:38,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.785e+02 1.259e+03 1.543e+03 2.022e+03 5.729e+03, threshold=3.087e+03, percent-clipped=3.0
+2023-03-11 10:32:56,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3079, 3.1374, 1.4455, 1.4662], device='cuda:0'), covar=tensor([0.1027, 0.0288, 0.0897, 0.1368], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0550, 0.0386, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 10:32:57,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=977018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:33:09,688 INFO [train.py:968] (0/2) Epoch 22, batch 19300, giga_loss[loss=0.2662, simple_loss=0.3465, pruned_loss=0.09296, over 28644.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3487, pruned_loss=0.0992, over 5712464.34 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3343, pruned_loss=0.08652, over 5750446.76 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3499, pruned_loss=0.1007, over 5700690.95 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:33:43,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=977067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:33:46,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=977070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:33:57,059 INFO [train.py:968] (0/2) Epoch 22, batch 19350, giga_loss[loss=0.2669, simple_loss=0.3322, pruned_loss=0.1008, over 28586.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.344, pruned_loss=0.09653, over 5696434.33 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3347, pruned_loss=0.08666, over 5750220.00 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3449, pruned_loss=0.09789, over 5686546.53 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:34:08,750 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=977095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:10,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.452e+02 1.215e+03 1.456e+03 2.210e+03 5.048e+03, threshold=2.913e+03, percent-clipped=4.0
+2023-03-11 10:34:11,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=977098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:11,494 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=977098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:12,260 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=977099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:12,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0860, 3.9321, 3.6982, 1.8935], device='cuda:0'), covar=tensor([0.0573, 0.0685, 0.0684, 0.2217], device='cuda:0'), in_proj_covar=tensor([0.1209, 0.1122, 0.0947, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 10:34:13,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=977101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:24,887 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 10:34:39,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=977127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:42,146 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=977130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:42,810 INFO [train.py:968] (0/2) Epoch 22, batch 19400, giga_loss[loss=0.213, simple_loss=0.2873, pruned_loss=0.06939, over 28778.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3392, pruned_loss=0.09443, over 5688087.58 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3354, pruned_loss=0.08716, over 5751133.82 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3393, pruned_loss=0.09519, over 5678444.60 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:35:10,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=977161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:35:13,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=977164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:35:15,357 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9722, 5.7667, 5.4881, 2.9922], device='cuda:0'), covar=tensor([0.0348, 0.0468, 0.0596, 0.1597], device='cuda:0'), in_proj_covar=tensor([0.1209, 0.1123, 0.0947, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 10:35:28,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-11 10:35:28,779 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 10:35:29,945 INFO [train.py:968] (0/2) Epoch 22, batch 19450, giga_loss[loss=0.225, simple_loss=0.2875, pruned_loss=0.08131, over 23439.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3345, pruned_loss=0.0923, over 5676172.52 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3358, pruned_loss=0.08729, over 5746944.99 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3343, pruned_loss=0.09289, over 5671293.61 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:35:41,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=977193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:35:44,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.605e+02 1.063e+03 1.295e+03 1.960e+03 4.597e+03, threshold=2.590e+03, percent-clipped=9.0
+2023-03-11 10:36:15,976 INFO [train.py:968] (0/2) Epoch 22, batch 19500, giga_loss[loss=0.2523, simple_loss=0.3317, pruned_loss=0.08649, over 28912.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3318, pruned_loss=0.09093, over 5654057.52 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3359, pruned_loss=0.08722, over 5750700.88 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3313, pruned_loss=0.09158, over 5644607.26 frames. ], batch size: 106, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:36:55,670 INFO [train.py:968] (0/2) Epoch 22, batch 19550, libri_loss[loss=0.2423, simple_loss=0.3303, pruned_loss=0.0772, over 29554.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3327, pruned_loss=0.09093, over 5669295.93 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3358, pruned_loss=0.08707, over 5754149.13 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3324, pruned_loss=0.09178, over 5654392.51 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:37:08,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.438e+02 1.158e+03 1.480e+03 2.046e+03 8.326e+03, threshold=2.960e+03, percent-clipped=12.0
+2023-03-11 10:37:40,307 INFO [train.py:968] (0/2) Epoch 22, batch 19600, giga_loss[loss=0.2404, simple_loss=0.3274, pruned_loss=0.07674, over 28317.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3327, pruned_loss=0.09108, over 5678253.39 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3358, pruned_loss=0.08712, over 5757919.89 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3324, pruned_loss=0.09182, over 5661375.48 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:37:48,915 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=977342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:38:01,891 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-11 10:38:21,028 INFO [train.py:968] (0/2) Epoch 22, batch 19650, giga_loss[loss=0.2642, simple_loss=0.3335, pruned_loss=0.09746, over 28975.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3317, pruned_loss=0.09064, over 5682990.10 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.336, pruned_loss=0.08715, over 5757753.36 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3312, pruned_loss=0.09124, over 5668803.80 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:38:31,419 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 10:38:33,663 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.308e+02 1.116e+03 1.391e+03 1.821e+03 7.814e+03, threshold=2.782e+03, percent-clipped=4.0
+2023-03-11 10:38:59,213 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2000, 0.8093, 0.9209, 1.3927], device='cuda:0'), covar=tensor([0.0784, 0.0387, 0.0361, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 10:38:59,631 INFO [train.py:968] (0/2) Epoch 22, batch 19700, libri_loss[loss=0.2659, simple_loss=0.3529, pruned_loss=0.08951, over 29555.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3297, pruned_loss=0.08948, over 5693219.59 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3365, pruned_loss=0.08729, over 5762015.82 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3287, pruned_loss=0.08992, over 5676043.33 frames. ], batch size: 75, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:39:14,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0839, 1.2363, 1.1094, 0.8430], device='cuda:0'), covar=tensor([0.1077, 0.0552, 0.1118, 0.1197], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0443, 0.0517, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 10:39:37,530 INFO [train.py:968] (0/2) Epoch 22, batch 19750, giga_loss[loss=0.2188, simple_loss=0.2935, pruned_loss=0.07203, over 28511.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3281, pruned_loss=0.08864, over 5702371.98 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3372, pruned_loss=0.0876, over 5764589.92 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3265, pruned_loss=0.08873, over 5685382.97 frames. ], batch size: 65, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:39:51,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.865e+02 1.122e+03 1.345e+03 1.814e+03 8.239e+03, threshold=2.691e+03, percent-clipped=9.0
+2023-03-11 10:40:16,192 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-11 10:40:20,636 INFO [train.py:968] (0/2) Epoch 22, batch 19800, giga_loss[loss=0.293, simple_loss=0.352, pruned_loss=0.117, over 26680.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3262, pruned_loss=0.088, over 5703228.75 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3374, pruned_loss=0.08761, over 5767683.45 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3246, pruned_loss=0.08809, over 5685618.41 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:40:57,445 INFO [train.py:968] (0/2) Epoch 22, batch 19850, giga_loss[loss=0.2371, simple_loss=0.3094, pruned_loss=0.08236, over 28876.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3233, pruned_loss=0.08666, over 5706307.45 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3382, pruned_loss=0.08803, over 5759832.18 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.321, pruned_loss=0.08635, over 5698054.15 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:41:01,680 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5621, 1.6552, 1.6372, 1.4133], device='cuda:0'), covar=tensor([0.3251, 0.2786, 0.2085, 0.3076], device='cuda:0'), in_proj_covar=tensor([0.1953, 0.1871, 0.1796, 0.1959], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 10:41:12,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.143e+02 1.085e+03 1.361e+03 2.077e+03 1.989e+04, threshold=2.722e+03, percent-clipped=15.0
+2023-03-11 10:41:33,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 10:41:38,919 INFO [train.py:968] (0/2) Epoch 22, batch 19900, giga_loss[loss=0.2505, simple_loss=0.3201, pruned_loss=0.09045, over 28832.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3217, pruned_loss=0.08612, over 5714374.50 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3384, pruned_loss=0.08806, over 5760311.02 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3196, pruned_loss=0.08584, over 5707326.02 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:42:08,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=977667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:42:19,764 INFO [train.py:968] (0/2) Epoch 22, batch 19950, giga_loss[loss=0.2405, simple_loss=0.3119, pruned_loss=0.08451, over 28349.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3208, pruned_loss=0.08592, over 5713607.68 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3395, pruned_loss=0.08852, over 5762659.99 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.318, pruned_loss=0.08525, over 5705290.48 frames. ], batch size: 65, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:42:27,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2683, 1.4355, 1.3131, 1.0956], device='cuda:0'), covar=tensor([0.2962, 0.2611, 0.1706, 0.2686], device='cuda:0'), in_proj_covar=tensor([0.1949, 0.1865, 0.1792, 0.1957], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 10:42:33,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.224e+02 1.034e+03 1.308e+03 1.674e+03 3.477e+03, threshold=2.615e+03, percent-clipped=3.0
+2023-03-11 10:42:34,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2674, 1.2176, 1.1935, 1.4500], device='cuda:0'), covar=tensor([0.0800, 0.0410, 0.0367, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 10:42:48,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=977717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:42:57,538 INFO [train.py:968] (0/2) Epoch 22, batch 20000, giga_loss[loss=0.216, simple_loss=0.2962, pruned_loss=0.06789, over 29033.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3187, pruned_loss=0.08463, over 5720014.40 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3397, pruned_loss=0.08853, over 5764824.98 frames. ], giga_tot_loss[loss=0.242, simple_loss=0.316, pruned_loss=0.08405, over 5711007.57 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:43:36,433 INFO [train.py:968] (0/2) Epoch 22, batch 20050, giga_loss[loss=0.2372, simple_loss=0.3058, pruned_loss=0.0843, over 28331.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.319, pruned_loss=0.08441, over 5721745.20 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3403, pruned_loss=0.08871, over 5764838.03 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3158, pruned_loss=0.08367, over 5713684.89 frames. ], batch size: 65, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:43:38,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4842, 4.3210, 4.0830, 1.9478], device='cuda:0'), covar=tensor([0.0542, 0.0697, 0.0651, 0.2068], device='cuda:0'), in_proj_covar=tensor([0.1206, 0.1115, 0.0941, 0.0712], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 10:43:44,918 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2735, 1.8916, 1.4499, 0.5837], device='cuda:0'), covar=tensor([0.5633, 0.3023, 0.4433, 0.6434], device='cuda:0'), in_proj_covar=tensor([0.1749, 0.1641, 0.1595, 0.1420], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 10:43:51,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.197e+02 9.767e+02 1.221e+03 1.524e+03 4.300e+03, threshold=2.442e+03, percent-clipped=3.0
+2023-03-11 10:44:22,149 INFO [train.py:968] (0/2) Epoch 22, batch 20100, giga_loss[loss=0.3144, simple_loss=0.3683, pruned_loss=0.1302, over 23817.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3243, pruned_loss=0.08847, over 5712838.52 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3403, pruned_loss=0.08871, over 5764838.03 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3218, pruned_loss=0.0879, over 5706565.08 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:44:34,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7400, 1.9650, 1.7285, 1.7065], device='cuda:0'), covar=tensor([0.2136, 0.2427, 0.2435, 0.2268], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0750, 0.0713, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 10:44:47,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4240, 1.1942, 4.3730, 3.3677], device='cuda:0'), covar=tensor([0.1673, 0.2975, 0.0376, 0.1140], device='cuda:0'), in_proj_covar=tensor([0.0753, 0.0641, 0.0951, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 10:44:47,428 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=977860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:44:50,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=977863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:45:09,446 INFO [train.py:968] (0/2) Epoch 22, batch 20150, giga_loss[loss=0.2581, simple_loss=0.3385, pruned_loss=0.08888, over 28919.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.331, pruned_loss=0.09261, over 5698006.44 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3409, pruned_loss=0.08895, over 5758553.00 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3282, pruned_loss=0.09197, over 5697402.79 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:45:17,801 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=977892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:45:22,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.086e+02 1.265e+03 1.600e+03 2.189e+03 5.449e+03, threshold=3.201e+03, percent-clipped=22.0
+2023-03-11 10:45:54,652 INFO [train.py:968] (0/2) Epoch 22, batch 20200, giga_loss[loss=0.3389, simple_loss=0.3892, pruned_loss=0.1443, over 26530.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3385, pruned_loss=0.09709, over 5691982.85 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3413, pruned_loss=0.08899, over 5754285.65 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3356, pruned_loss=0.09669, over 5694143.35 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:46:08,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 10:46:18,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 10:46:39,445 INFO [train.py:968] (0/2) Epoch 22, batch 20250, giga_loss[loss=0.3382, simple_loss=0.4065, pruned_loss=0.135, over 28576.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3442, pruned_loss=0.1002, over 5680840.14 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3415, pruned_loss=0.08914, over 5746975.15 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3418, pruned_loss=0.09996, over 5688616.64 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:46:55,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.924e+02 1.441e+03 1.828e+03 2.511e+03 6.934e+03, threshold=3.656e+03, percent-clipped=10.0
+2023-03-11 10:46:56,321 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-978000.pt
+2023-03-11 10:47:08,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3283, 1.7579, 1.5308, 1.4661], device='cuda:0'), covar=tensor([0.2243, 0.1931, 0.2397, 0.2154], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0750, 0.0712, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 10:47:25,009 INFO [train.py:968] (0/2) Epoch 22, batch 20300, giga_loss[loss=0.3228, simple_loss=0.3734, pruned_loss=0.1361, over 23651.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3494, pruned_loss=0.1024, over 5678579.11 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3415, pruned_loss=0.08911, over 5747495.52 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3476, pruned_loss=0.1025, over 5683142.30 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:47:28,099 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9933, 1.3153, 1.0833, 0.2876], device='cuda:0'), covar=tensor([0.4074, 0.3426, 0.4432, 0.6260], device='cuda:0'), in_proj_covar=tensor([0.1759, 0.1651, 0.1602, 0.1429], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 10:47:34,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=978042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:48:01,287 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6363, 1.7917, 1.5252, 1.5644], device='cuda:0'), covar=tensor([0.2828, 0.2879, 0.3191, 0.2391], device='cuda:0'), in_proj_covar=tensor([0.1503, 0.1090, 0.1327, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 10:48:11,523 INFO [train.py:968] (0/2) Epoch 22, batch 20350, giga_loss[loss=0.2898, simple_loss=0.368, pruned_loss=0.1058, over 28922.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3539, pruned_loss=0.1042, over 5676092.71 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3417, pruned_loss=0.08914, over 5745587.85 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3523, pruned_loss=0.1044, over 5680652.39 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:48:26,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.479e+02 1.189e+03 1.533e+03 2.236e+03 5.525e+03, threshold=3.067e+03, percent-clipped=5.0
+2023-03-11 10:48:53,158 INFO [train.py:968] (0/2) Epoch 22, batch 20400, giga_loss[loss=0.2871, simple_loss=0.3527, pruned_loss=0.1107, over 27522.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3571, pruned_loss=0.1056, over 5684194.16 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.342, pruned_loss=0.08923, over 5745369.36 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3559, pruned_loss=0.1061, over 5686936.85 frames. ], batch size: 472, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:49:00,474 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5341, 3.5556, 1.6387, 1.6119], device='cuda:0'), covar=tensor([0.1023, 0.0281, 0.0946, 0.1512], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0550, 0.0386, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 10:49:40,099 INFO [train.py:968] (0/2) Epoch 22, batch 20450, giga_loss[loss=0.2402, simple_loss=0.32, pruned_loss=0.08023, over 28686.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3526, pruned_loss=0.1023, over 5677866.96 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3423, pruned_loss=0.08936, over 5738232.61 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3516, pruned_loss=0.1027, over 5685312.82 frames. ], batch size: 284, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:49:43,228 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=978185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:49:46,453 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=978188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:49:52,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0872, 1.4169, 1.4606, 1.2473], device='cuda:0'), covar=tensor([0.2110, 0.1747, 0.2363, 0.1892], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0750, 0.0713, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 10:49:55,181 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.701e+02 1.401e+03 1.815e+03 2.392e+03 7.016e+03, threshold=3.630e+03, percent-clipped=14.0
+2023-03-11 10:50:02,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5380, 1.6773, 1.6035, 1.4229], device='cuda:0'), covar=tensor([0.2969, 0.2549, 0.2127, 0.2622], device='cuda:0'), in_proj_covar=tensor([0.1949, 0.1872, 0.1795, 0.1958], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 10:50:05,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4313, 1.5744, 1.6875, 1.2468], device='cuda:0'), covar=tensor([0.1807, 0.2552, 0.1499, 0.1790], device='cuda:0'), in_proj_covar=tensor([0.0901, 0.0699, 0.0951, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 10:50:08,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=978217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:50:18,388 INFO [train.py:968] (0/2) Epoch 22, batch 20500, giga_loss[loss=0.3494, simple_loss=0.4096, pruned_loss=0.1446, over 28631.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3502, pruned_loss=0.09978, over 5687889.80 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3426, pruned_loss=0.08956, over 5739491.44 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3493, pruned_loss=0.1002, over 5691299.89 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:50:30,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4921, 1.7430, 1.3272, 1.6511], device='cuda:0'), covar=tensor([0.0755, 0.0301, 0.0335, 0.0842], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 10:50:58,369 INFO [train.py:968] (0/2) Epoch 22, batch 20550, giga_loss[loss=0.2712, simple_loss=0.3537, pruned_loss=0.09439, over 28974.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3488, pruned_loss=0.09863, over 5679337.44 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3427, pruned_loss=0.08969, over 5735049.65 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3482, pruned_loss=0.09913, over 5683917.54 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:51:14,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.333e+02 1.260e+03 1.614e+03 2.143e+03 4.033e+03, threshold=3.228e+03, percent-clipped=2.0
+2023-03-11 10:51:32,012 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=978319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:51:41,290 INFO [train.py:968] (0/2) Epoch 22, batch 20600, giga_loss[loss=0.264, simple_loss=0.3395, pruned_loss=0.09427, over 28721.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.349, pruned_loss=0.09829, over 5685015.91 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3429, pruned_loss=0.08972, over 5733747.92 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3485, pruned_loss=0.09877, over 5688903.54 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:52:04,538 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-11 10:52:25,160 INFO [train.py:968] (0/2) Epoch 22, batch 20650, giga_loss[loss=0.3111, simple_loss=0.3783, pruned_loss=0.122, over 28786.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3531, pruned_loss=0.1014, over 5687893.21 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.343, pruned_loss=0.08978, over 5736358.58 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3526, pruned_loss=0.1018, over 5688101.98 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:52:40,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.205e+03 1.541e+03 2.509e+03 6.347e+03, threshold=3.081e+03, percent-clipped=14.0
+2023-03-11 10:52:53,031 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 10:53:06,756 INFO [train.py:968] (0/2) Epoch 22, batch 20700, giga_loss[loss=0.2497, simple_loss=0.3329, pruned_loss=0.0832, over 28999.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3535, pruned_loss=0.102, over 5695681.15 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3423, pruned_loss=0.08939, over 5740587.66 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.354, pruned_loss=0.103, over 5690968.99 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:53:35,074 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=978460.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 10:53:51,209 INFO [train.py:968] (0/2) Epoch 22, batch 20750, giga_loss[loss=0.304, simple_loss=0.369, pruned_loss=0.1196, over 28912.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3549, pruned_loss=0.1031, over 5707877.36 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3424, pruned_loss=0.08932, over 5743017.91 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3555, pruned_loss=0.1042, over 5701327.97 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:53:59,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-11 10:54:06,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.557e+02 1.281e+03 1.532e+03 1.913e+03 5.892e+03, threshold=3.064e+03, percent-clipped=6.0
+2023-03-11 10:54:15,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9681, 1.9548, 2.2140, 1.7524], device='cuda:0'), covar=tensor([0.1735, 0.2217, 0.1308, 0.1585], device='cuda:0'), in_proj_covar=tensor([0.0901, 0.0699, 0.0950, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 10:54:35,179 INFO [train.py:968] (0/2) Epoch 22, batch 20800, giga_loss[loss=0.295, simple_loss=0.367, pruned_loss=0.1115, over 28487.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3562, pruned_loss=0.1044, over 5697743.98 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3431, pruned_loss=0.08969, over 5735981.94 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3565, pruned_loss=0.1054, over 5698468.85 frames. ], batch size: 65, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:55:11,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3923, 1.6034, 1.2027, 1.1890], device='cuda:0'), covar=tensor([0.0984, 0.0529, 0.1073, 0.1088], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0443, 0.0519, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 10:55:12,967 INFO [train.py:968] (0/2) Epoch 22, batch 20850, giga_loss[loss=0.3023, simple_loss=0.3809, pruned_loss=0.1118, over 28703.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.357, pruned_loss=0.1047, over 5706770.87 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3434, pruned_loss=0.08986, over 5738567.94 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3571, pruned_loss=0.1056, over 5704483.02 frames. ], batch size: 284, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:55:17,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9958, 2.8474, 2.6597, 1.5012], device='cuda:0'), covar=tensor([0.0978, 0.1020, 0.0942, 0.2092], device='cuda:0'), in_proj_covar=tensor([0.1214, 0.1126, 0.0950, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 10:55:29,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.318e+02 1.240e+03 1.614e+03 2.087e+03 6.023e+03, threshold=3.228e+03, percent-clipped=7.0
+2023-03-11 10:55:30,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2720, 1.5103, 1.2138, 1.2812], device='cuda:0'), covar=tensor([0.2946, 0.2712, 0.3286, 0.2190], device='cuda:0'), in_proj_covar=tensor([0.1508, 0.1094, 0.1332, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 10:55:51,825 INFO [train.py:968] (0/2) Epoch 22, batch 20900, giga_loss[loss=0.2756, simple_loss=0.3534, pruned_loss=0.09894, over 28876.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3556, pruned_loss=0.1028, over 5707499.57 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.343, pruned_loss=0.08948, over 5742068.33 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3565, pruned_loss=0.1044, over 5701181.54 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:56:06,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4790, 1.2743, 1.2478, 1.5573], device='cuda:0'), covar=tensor([0.0726, 0.0353, 0.0336, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 10:56:10,265 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.35 vs. limit=5.0
+2023-03-11 10:56:27,620 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=978675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:56:31,889 INFO [train.py:968] (0/2) Epoch 22, batch 20950, giga_loss[loss=0.2952, simple_loss=0.3678, pruned_loss=0.1113, over 28623.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3561, pruned_loss=0.1022, over 5715588.81 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.343, pruned_loss=0.08959, over 5745302.62 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3571, pruned_loss=0.1036, over 5707164.05 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:56:42,849 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-11 10:56:43,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=978694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:56:46,173 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=978698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:56:48,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.964e+02 1.260e+03 1.609e+03 2.150e+03 5.514e+03, threshold=3.218e+03, percent-clipped=8.0
+2023-03-11 10:57:12,018 INFO [train.py:968] (0/2) Epoch 22, batch 21000, giga_loss[loss=0.2432, simple_loss=0.3269, pruned_loss=0.07978, over 29000.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3552, pruned_loss=0.1015, over 5721513.93 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3434, pruned_loss=0.08988, over 5748091.30 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3558, pruned_loss=0.1026, over 5711828.22 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:57:12,022 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 10:57:19,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1698, 1.7877, 1.4050, 0.4346], device='cuda:0'), covar=tensor([0.4125, 0.2596, 0.3627, 0.5412], device='cuda:0'), in_proj_covar=tensor([0.1742, 0.1630, 0.1584, 0.1417], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 10:57:20,794 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2077, simple_loss=0.3151, pruned_loss=0.0502, over 944034.00 frames. 
+2023-03-11 10:57:20,795 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 10:57:38,319 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-11 10:57:48,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5227, 1.6814, 1.4541, 1.7305], device='cuda:0'), covar=tensor([0.0799, 0.0313, 0.0328, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0062, 0.0108], device='cuda:0')
+2023-03-11 10:58:00,150 INFO [train.py:968] (0/2) Epoch 22, batch 21050, giga_loss[loss=0.2612, simple_loss=0.3334, pruned_loss=0.09456, over 28646.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3524, pruned_loss=0.1002, over 5710356.54 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09, over 5741960.28 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3528, pruned_loss=0.1011, over 5707110.51 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:58:06,705 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=978791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:58:13,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.208e+02 1.179e+03 1.654e+03 2.273e+03 6.210e+03, threshold=3.307e+03, percent-clipped=5.0
+2023-03-11 10:58:35,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 10:58:36,226 INFO [train.py:968] (0/2) Epoch 22, batch 21100, giga_loss[loss=0.2398, simple_loss=0.3193, pruned_loss=0.08015, over 28872.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3493, pruned_loss=0.09837, over 5707513.60 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3433, pruned_loss=0.08976, over 5737476.19 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3503, pruned_loss=0.09967, over 5708311.73 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:58:40,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=978835.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 10:58:41,546 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=978837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:58:43,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=978840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:59:06,397 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=978869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:59:14,982 INFO [train.py:968] (0/2) Epoch 22, batch 21150, giga_loss[loss=0.2942, simple_loss=0.3668, pruned_loss=0.1108, over 28933.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.349, pruned_loss=0.09823, over 5714203.55 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09022, over 5739794.31 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3491, pruned_loss=0.09901, over 5712166.41 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:59:18,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8195, 3.6499, 3.4336, 1.9247], device='cuda:0'), covar=tensor([0.0674, 0.0829, 0.0817, 0.2291], device='cuda:0'), in_proj_covar=tensor([0.1211, 0.1122, 0.0948, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 10:59:31,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.779e+02 1.118e+03 1.287e+03 1.785e+03 5.624e+03, threshold=2.573e+03, percent-clipped=3.0
+2023-03-11 10:59:56,437 INFO [train.py:968] (0/2) Epoch 22, batch 21200, giga_loss[loss=0.3094, simple_loss=0.3821, pruned_loss=0.1184, over 28887.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3489, pruned_loss=0.09858, over 5714151.66 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3436, pruned_loss=0.08992, over 5741910.39 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3497, pruned_loss=0.09984, over 5709677.35 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:00:35,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=978978.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:00:36,846 INFO [train.py:968] (0/2) Epoch 22, batch 21250, giga_loss[loss=0.2611, simple_loss=0.3393, pruned_loss=0.09145, over 29104.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3487, pruned_loss=0.09855, over 5713035.33 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3432, pruned_loss=0.08981, over 5746242.73 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3499, pruned_loss=0.09986, over 5704773.38 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:00:37,185 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=978981.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:00:45,477 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-11 11:00:52,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.599e+02 1.206e+03 1.537e+03 2.274e+03 7.915e+03, threshold=3.073e+03, percent-clipped=19.0
+2023-03-11 11:01:00,457 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979010.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:01:04,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 11:01:16,190 INFO [train.py:968] (0/2) Epoch 22, batch 21300, giga_loss[loss=0.2718, simple_loss=0.3428, pruned_loss=0.1004, over 28671.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3493, pruned_loss=0.09851, over 5720113.43 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3436, pruned_loss=0.09037, over 5749767.16 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.35, pruned_loss=0.09929, over 5709543.34 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:01:30,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:01:51,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:01:53,352 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:01:57,337 INFO [train.py:968] (0/2) Epoch 22, batch 21350, giga_loss[loss=0.248, simple_loss=0.326, pruned_loss=0.08502, over 28828.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3485, pruned_loss=0.09787, over 5716022.62 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3438, pruned_loss=0.09076, over 5752978.10 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.349, pruned_loss=0.09837, over 5703761.56 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:02:14,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.796e+02 1.098e+03 1.377e+03 2.065e+03 7.124e+03, threshold=2.755e+03, percent-clipped=8.0
+2023-03-11 11:02:37,714 INFO [train.py:968] (0/2) Epoch 22, batch 21400, libri_loss[loss=0.2346, simple_loss=0.3124, pruned_loss=0.07841, over 29644.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3472, pruned_loss=0.09767, over 5713923.66 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3438, pruned_loss=0.0908, over 5755246.97 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3477, pruned_loss=0.09814, over 5701537.54 frames. ], batch size: 69, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:02:57,080 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:05,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:11,443 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3741, 1.6152, 1.1291, 1.1929], device='cuda:0'), covar=tensor([0.1052, 0.0589, 0.1149, 0.1300], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0443, 0.0517, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 11:03:12,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:16,827 INFO [train.py:968] (0/2) Epoch 22, batch 21450, giga_loss[loss=0.2717, simple_loss=0.3341, pruned_loss=0.1046, over 28720.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3447, pruned_loss=0.09669, over 5710158.18 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3437, pruned_loss=0.09085, over 5756687.61 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3452, pruned_loss=0.09707, over 5698631.20 frames. ], batch size: 60, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:03:27,441 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:29,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=979196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:34,091 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 11:03:34,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.006e+02 1.099e+03 1.409e+03 1.841e+03 4.796e+03, threshold=2.819e+03, percent-clipped=6.0
+2023-03-11 11:03:46,285 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:48,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=979219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:53,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:58,663 INFO [train.py:968] (0/2) Epoch 22, batch 21500, giga_loss[loss=0.2536, simple_loss=0.3323, pruned_loss=0.08746, over 28939.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3416, pruned_loss=0.09553, over 5703827.47 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3439, pruned_loss=0.09096, over 5758114.07 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3418, pruned_loss=0.09578, over 5693130.67 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:04:09,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1435, 1.1341, 3.7050, 3.0736], device='cuda:0'), covar=tensor([0.1765, 0.2860, 0.0452, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0754, 0.0641, 0.0949, 0.0897], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 11:04:11,487 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:04:37,293 INFO [train.py:968] (0/2) Epoch 22, batch 21550, giga_loss[loss=0.2621, simple_loss=0.347, pruned_loss=0.08862, over 28745.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3411, pruned_loss=0.09531, over 5700968.69 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.344, pruned_loss=0.09102, over 5755619.07 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3411, pruned_loss=0.09553, over 5693412.39 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:04:52,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.155e+02 1.195e+03 1.539e+03 2.035e+03 4.011e+03, threshold=3.078e+03, percent-clipped=12.0
+2023-03-11 11:05:01,274 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:05:04,201 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=979312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:05:17,862 INFO [train.py:968] (0/2) Epoch 22, batch 21600, giga_loss[loss=0.2409, simple_loss=0.323, pruned_loss=0.07945, over 28919.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3401, pruned_loss=0.0954, over 5691329.11 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3443, pruned_loss=0.09124, over 5748104.59 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3399, pruned_loss=0.09545, over 5690691.32 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:05:24,604 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:06:02,811 INFO [train.py:968] (0/2) Epoch 22, batch 21650, giga_loss[loss=0.222, simple_loss=0.2995, pruned_loss=0.07225, over 29046.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3382, pruned_loss=0.09474, over 5698963.12 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3443, pruned_loss=0.09131, over 5750914.68 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.338, pruned_loss=0.09479, over 5695271.38 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:06:17,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.636e+02 1.256e+03 1.647e+03 2.297e+03 4.212e+03, threshold=3.295e+03, percent-clipped=6.0
+2023-03-11 11:06:39,925 INFO [train.py:968] (0/2) Epoch 22, batch 21700, giga_loss[loss=0.3065, simple_loss=0.3699, pruned_loss=0.1215, over 28510.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3355, pruned_loss=0.09324, over 5703449.05 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.344, pruned_loss=0.09127, over 5752657.85 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3353, pruned_loss=0.09342, over 5696790.70 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:06:57,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:07:21,054 INFO [train.py:968] (0/2) Epoch 22, batch 21750, giga_loss[loss=0.2542, simple_loss=0.3307, pruned_loss=0.08883, over 28922.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3332, pruned_loss=0.0919, over 5705396.25 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3448, pruned_loss=0.0919, over 5745162.54 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3321, pruned_loss=0.09144, over 5706426.89 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:07:21,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979481.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:07:36,097 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.102e+02 1.113e+03 1.397e+03 1.916e+03 5.212e+03, threshold=2.793e+03, percent-clipped=3.0
+2023-03-11 11:07:57,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:07:58,085 INFO [train.py:968] (0/2) Epoch 22, batch 21800, giga_loss[loss=0.2673, simple_loss=0.3334, pruned_loss=0.1006, over 28491.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3336, pruned_loss=0.09257, over 5700691.31 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3452, pruned_loss=0.09217, over 5737890.20 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3321, pruned_loss=0.092, over 5706051.74 frames. ], batch size: 60, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:07:59,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3716, 1.5727, 1.6486, 1.2243], device='cuda:0'), covar=tensor([0.1805, 0.2566, 0.1538, 0.1824], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0702, 0.0953, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 11:08:14,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:08:40,687 INFO [train.py:968] (0/2) Epoch 22, batch 21850, libri_loss[loss=0.2953, simple_loss=0.3675, pruned_loss=0.1115, over 29535.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3355, pruned_loss=0.09307, over 5706105.39 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3456, pruned_loss=0.09249, over 5739818.51 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3337, pruned_loss=0.09233, over 5707971.10 frames. ], batch size: 89, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:08:52,359 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:08:54,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=979597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:08:58,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.654e+02 1.206e+03 1.517e+03 2.062e+03 5.880e+03, threshold=3.034e+03, percent-clipped=11.0
+2023-03-11 11:09:19,439 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:09:25,100 INFO [train.py:968] (0/2) Epoch 22, batch 21900, giga_loss[loss=0.3452, simple_loss=0.3877, pruned_loss=0.1514, over 23760.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3393, pruned_loss=0.09505, over 5695675.55 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3457, pruned_loss=0.09262, over 5741544.49 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3377, pruned_loss=0.09436, over 5695071.48 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:09:59,562 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:10:02,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=979675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:10:07,330 INFO [train.py:968] (0/2) Epoch 22, batch 21950, giga_loss[loss=0.2352, simple_loss=0.3166, pruned_loss=0.07688, over 28800.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3416, pruned_loss=0.09544, over 5699242.39 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3458, pruned_loss=0.09289, over 5745350.72 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3402, pruned_loss=0.09472, over 5694354.89 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:10:13,612 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5267, 1.7136, 1.7571, 1.3296], device='cuda:0'), covar=tensor([0.1988, 0.2578, 0.1660, 0.1960], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0699, 0.0949, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 11:10:16,276 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:10:18,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=979695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:10:23,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.167e+02 1.082e+03 1.264e+03 1.591e+03 4.039e+03, threshold=2.527e+03, percent-clipped=3.0
+2023-03-11 11:10:25,936 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:10:43,005 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4118, 1.3425, 4.3532, 3.6102], device='cuda:0'), covar=tensor([0.2044, 0.3161, 0.0679, 0.1163], device='cuda:0'), in_proj_covar=tensor([0.0758, 0.0646, 0.0957, 0.0903], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 11:10:43,008 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:10:49,195 INFO [train.py:968] (0/2) Epoch 22, batch 22000, giga_loss[loss=0.2626, simple_loss=0.347, pruned_loss=0.08908, over 28696.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.342, pruned_loss=0.09476, over 5705054.54 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3456, pruned_loss=0.0929, over 5747495.49 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3409, pruned_loss=0.09421, over 5698593.60 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:11:24,672 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1841, 1.4032, 1.2817, 1.1041], device='cuda:0'), covar=tensor([0.3017, 0.2577, 0.1874, 0.2453], device='cuda:0'), in_proj_covar=tensor([0.1955, 0.1881, 0.1811, 0.1963], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 11:11:32,034 INFO [train.py:968] (0/2) Epoch 22, batch 22050, giga_loss[loss=0.2643, simple_loss=0.3314, pruned_loss=0.09856, over 23909.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.341, pruned_loss=0.09388, over 5694648.88 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3453, pruned_loss=0.09301, over 5739655.13 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3404, pruned_loss=0.09338, over 5695142.49 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:11:50,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.398e+02 1.156e+03 1.593e+03 2.383e+03 8.888e+03, threshold=3.186e+03, percent-clipped=19.0
+2023-03-11 11:12:12,817 INFO [train.py:968] (0/2) Epoch 22, batch 22100, giga_loss[loss=0.2742, simple_loss=0.3513, pruned_loss=0.09853, over 27710.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3418, pruned_loss=0.09454, over 5687118.72 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3455, pruned_loss=0.09338, over 5733669.99 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.341, pruned_loss=0.0938, over 5692677.00 frames. ], batch size: 472, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:12:32,959 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979856.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:12:51,860 INFO [train.py:968] (0/2) Epoch 22, batch 22150, giga_loss[loss=0.2908, simple_loss=0.3671, pruned_loss=0.1073, over 28695.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3436, pruned_loss=0.09611, over 5692961.90 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3464, pruned_loss=0.09406, over 5733121.90 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3421, pruned_loss=0.09492, over 5697185.26 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:13:12,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.019e+02 1.397e+03 1.706e+03 2.547e+03 8.059e+03, threshold=3.411e+03, percent-clipped=19.0
+2023-03-11 11:13:34,262 INFO [train.py:968] (0/2) Epoch 22, batch 22200, giga_loss[loss=0.2852, simple_loss=0.3594, pruned_loss=0.1055, over 28820.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.345, pruned_loss=0.09721, over 5684517.36 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3464, pruned_loss=0.0942, over 5726998.39 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3437, pruned_loss=0.09616, over 5693225.46 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:13:57,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1927, 0.8078, 0.8818, 1.3476], device='cuda:0'), covar=tensor([0.0732, 0.0420, 0.0382, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 11:14:03,636 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:14:13,664 INFO [train.py:968] (0/2) Epoch 22, batch 22250, giga_loss[loss=0.3029, simple_loss=0.3742, pruned_loss=0.1158, over 28800.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3477, pruned_loss=0.09831, over 5699586.05 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.347, pruned_loss=0.0946, over 5730399.33 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3462, pruned_loss=0.09722, over 5702612.63 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:14:17,583 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:14:20,366 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4264, 1.5720, 1.5372, 1.3207], device='cuda:0'), covar=tensor([0.2943, 0.2485, 0.2276, 0.2770], device='cuda:0'), in_proj_covar=tensor([0.1966, 0.1896, 0.1825, 0.1977], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 11:14:28,833 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979999.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:14:29,313 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-980000.pt
+2023-03-11 11:14:32,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=980002.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:14:32,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.717e+02 1.211e+03 1.491e+03 1.868e+03 4.648e+03, threshold=2.982e+03, percent-clipped=3.0
+2023-03-11 11:14:52,149 INFO [train.py:968] (0/2) Epoch 22, batch 22300, giga_loss[loss=0.3317, simple_loss=0.4027, pruned_loss=0.1304, over 29039.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.351, pruned_loss=0.1002, over 5706640.17 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3482, pruned_loss=0.09567, over 5733966.37 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3487, pruned_loss=0.09847, over 5704298.98 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:14:52,362 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=980031.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:15:30,109 INFO [train.py:968] (0/2) Epoch 22, batch 22350, giga_loss[loss=0.2576, simple_loss=0.3446, pruned_loss=0.08531, over 29106.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3522, pruned_loss=0.1007, over 5717423.61 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3486, pruned_loss=0.09604, over 5739470.64 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3501, pruned_loss=0.09918, over 5709670.01 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:15:40,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9777, 3.7931, 3.6086, 1.9055], device='cuda:0'), covar=tensor([0.0771, 0.0892, 0.0846, 0.1947], device='cuda:0'), in_proj_covar=tensor([0.1207, 0.1118, 0.0949, 0.0712], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 11:15:46,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.237e+02 1.354e+03 1.758e+03 2.576e+03 1.210e+04, threshold=3.515e+03, percent-clipped=15.0
+2023-03-11 11:16:09,206 INFO [train.py:968] (0/2) Epoch 22, batch 22400, giga_loss[loss=0.2563, simple_loss=0.3396, pruned_loss=0.08648, over 28817.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3522, pruned_loss=0.1005, over 5723295.12 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3488, pruned_loss=0.09612, over 5740149.29 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3505, pruned_loss=0.09926, over 5716511.65 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:16:31,016 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-11 11:16:44,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1542, 1.3908, 1.3456, 1.1045], device='cuda:0'), covar=tensor([0.3012, 0.2620, 0.1696, 0.2577], device='cuda:0'), in_proj_covar=tensor([0.1967, 0.1900, 0.1829, 0.1976], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 11:16:54,078 INFO [train.py:968] (0/2) Epoch 22, batch 22450, giga_loss[loss=0.3032, simple_loss=0.3714, pruned_loss=0.1175, over 28810.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3525, pruned_loss=0.1008, over 5714329.60 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3487, pruned_loss=0.09608, over 5739132.54 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3512, pruned_loss=0.09991, over 5709843.32 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:17:13,511 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.706e+02 1.282e+03 1.591e+03 2.054e+03 6.983e+03, threshold=3.182e+03, percent-clipped=4.0
+2023-03-11 11:17:35,787 INFO [train.py:968] (0/2) Epoch 22, batch 22500, giga_loss[loss=0.258, simple_loss=0.3347, pruned_loss=0.0907, over 28774.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3506, pruned_loss=0.1001, over 5718487.62 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3493, pruned_loss=0.09677, over 5743919.52 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3492, pruned_loss=0.09886, over 5709783.94 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:17:48,823 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=980250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:18:14,619 INFO [train.py:968] (0/2) Epoch 22, batch 22550, libri_loss[loss=0.2267, simple_loss=0.3042, pruned_loss=0.07466, over 29649.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3486, pruned_loss=0.09912, over 5717219.39 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3496, pruned_loss=0.09707, over 5737515.83 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3472, pruned_loss=0.09795, over 5714473.36 frames. ], batch size: 69, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:18:32,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.812e+02 1.205e+03 1.546e+03 1.881e+03 5.233e+03, threshold=3.091e+03, percent-clipped=7.0
+2023-03-11 11:18:55,654 INFO [train.py:968] (0/2) Epoch 22, batch 22600, giga_loss[loss=0.2576, simple_loss=0.3375, pruned_loss=0.08884, over 28582.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3457, pruned_loss=0.0981, over 5715991.44 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3504, pruned_loss=0.09788, over 5738915.26 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3438, pruned_loss=0.09649, over 5712032.01 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:19:03,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=980342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:19:16,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=980360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:19:23,378 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 11:19:33,163 INFO [train.py:968] (0/2) Epoch 22, batch 22650, giga_loss[loss=0.2366, simple_loss=0.3174, pruned_loss=0.07787, over 28801.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3439, pruned_loss=0.09647, over 5720037.61 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3511, pruned_loss=0.09837, over 5743232.60 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3416, pruned_loss=0.09472, over 5712516.42 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:19:55,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.914e+02 1.130e+03 1.459e+03 1.924e+03 7.031e+03, threshold=2.917e+03, percent-clipped=9.0
+2023-03-11 11:20:21,246 INFO [train.py:968] (0/2) Epoch 22, batch 22700, giga_loss[loss=0.2732, simple_loss=0.3589, pruned_loss=0.09371, over 28575.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3455, pruned_loss=0.09571, over 5710920.80 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3512, pruned_loss=0.09847, over 5740421.44 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3435, pruned_loss=0.09424, over 5707431.22 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:20:31,661 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-11 11:20:50,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4079, 4.4048, 1.6280, 1.6017], device='cuda:0'), covar=tensor([0.1034, 0.0433, 0.0946, 0.1384], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0550, 0.0386, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 11:21:00,758 INFO [train.py:968] (0/2) Epoch 22, batch 22750, libri_loss[loss=0.3156, simple_loss=0.3766, pruned_loss=0.1273, over 27690.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3459, pruned_loss=0.09639, over 5718883.85 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3519, pruned_loss=0.09925, over 5741218.94 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3437, pruned_loss=0.09445, over 5715060.83 frames. ], batch size: 116, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:21:04,601 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=980485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:21:06,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=980488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:21:07,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8502, 1.8711, 1.4360, 1.5350], device='cuda:0'), covar=tensor([0.1007, 0.0824, 0.1082, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0392, 0.0443, 0.0516, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 11:21:17,273 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=980503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:21:17,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.767e+02 1.171e+03 1.438e+03 1.901e+03 5.681e+03, threshold=2.875e+03, percent-clipped=8.0
+2023-03-11 11:21:19,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=980506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:21:21,724 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=980509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:21:30,621 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=980517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:21:39,698 INFO [train.py:968] (0/2) Epoch 22, batch 22800, giga_loss[loss=0.2668, simple_loss=0.3408, pruned_loss=0.09639, over 28786.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3452, pruned_loss=0.09651, over 5715421.46 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3522, pruned_loss=0.09964, over 5732190.44 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.343, pruned_loss=0.09458, over 5720486.19 frames. ], batch size: 284, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:21:40,333 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-11 11:21:43,927 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=980535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:22:21,616 INFO [train.py:968] (0/2) Epoch 22, batch 22850, giga_loss[loss=0.2639, simple_loss=0.3207, pruned_loss=0.1035, over 28411.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3442, pruned_loss=0.09763, over 5718589.05 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3524, pruned_loss=0.09976, over 5736862.28 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3421, pruned_loss=0.09589, over 5717959.18 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:22:37,026 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 11:22:40,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.470e+02 1.210e+03 1.628e+03 2.281e+03 6.367e+03, threshold=3.256e+03, percent-clipped=16.0
+2023-03-11 11:22:41,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-11 11:22:56,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=980625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:23:01,363 INFO [train.py:968] (0/2) Epoch 22, batch 22900, giga_loss[loss=0.3351, simple_loss=0.3819, pruned_loss=0.1441, over 26784.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3431, pruned_loss=0.09804, over 5708556.33 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.353, pruned_loss=0.1002, over 5729472.76 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3405, pruned_loss=0.09612, over 5714438.44 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:23:40,688 INFO [train.py:968] (0/2) Epoch 22, batch 22950, giga_loss[loss=0.2341, simple_loss=0.317, pruned_loss=0.07554, over 28962.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3416, pruned_loss=0.09806, over 5715954.01 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3531, pruned_loss=0.1005, over 5732302.98 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3392, pruned_loss=0.09624, over 5717673.74 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:23:54,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5443, 1.1896, 4.9158, 3.4336], device='cuda:0'), covar=tensor([0.1750, 0.3010, 0.0369, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.0755, 0.0643, 0.0956, 0.0902], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 11:23:58,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.735e+02 1.308e+03 1.565e+03 2.035e+03 5.141e+03, threshold=3.130e+03, percent-clipped=5.0
+2023-03-11 11:24:03,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5402, 3.4515, 1.6341, 1.6883], device='cuda:0'), covar=tensor([0.0908, 0.0399, 0.0882, 0.1234], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0549, 0.0386, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 11:24:20,892 INFO [train.py:968] (0/2) Epoch 22, batch 23000, libri_loss[loss=0.2467, simple_loss=0.3143, pruned_loss=0.08956, over 29662.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3397, pruned_loss=0.09753, over 5714015.10 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.353, pruned_loss=0.1005, over 5735739.73 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3376, pruned_loss=0.096, over 5711784.60 frames. ], batch size: 69, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:24:33,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1687, 4.0062, 3.7597, 1.9023], device='cuda:0'), covar=tensor([0.0651, 0.0797, 0.0804, 0.2082], device='cuda:0'), in_proj_covar=tensor([0.1220, 0.1126, 0.0954, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 11:24:48,575 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=980768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:24:50,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=980771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:24:57,851 INFO [train.py:968] (0/2) Epoch 22, batch 23050, giga_loss[loss=0.2516, simple_loss=0.3168, pruned_loss=0.09324, over 28713.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.336, pruned_loss=0.09572, over 5723891.03 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3533, pruned_loss=0.101, over 5740385.12 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3336, pruned_loss=0.09399, over 5717511.95 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:25:03,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7427, 4.5907, 4.3083, 2.2792], device='cuda:0'), covar=tensor([0.0459, 0.0562, 0.0620, 0.1952], device='cuda:0'), in_proj_covar=tensor([0.1223, 0.1130, 0.0957, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 11:25:12,409 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=980800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:25:14,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.481e+02 1.243e+03 1.630e+03 2.450e+03 7.724e+03, threshold=3.261e+03, percent-clipped=14.0
+2023-03-11 11:25:15,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0864, 2.3374, 2.2687, 1.8612], device='cuda:0'), covar=tensor([0.3267, 0.2301, 0.2226, 0.2994], device='cuda:0'), in_proj_covar=tensor([0.1966, 0.1896, 0.1819, 0.1972], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 11:25:29,281 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.47 vs. limit=5.0
+2023-03-11 11:25:39,963 INFO [train.py:968] (0/2) Epoch 22, batch 23100, giga_loss[loss=0.2345, simple_loss=0.3116, pruned_loss=0.07867, over 28886.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3318, pruned_loss=0.09333, over 5719340.65 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3534, pruned_loss=0.1011, over 5739911.83 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3293, pruned_loss=0.09164, over 5714196.01 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:26:08,362 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-11 11:26:18,983 INFO [train.py:968] (0/2) Epoch 22, batch 23150, giga_loss[loss=0.235, simple_loss=0.3135, pruned_loss=0.0782, over 29141.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3295, pruned_loss=0.09215, over 5727121.59 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3533, pruned_loss=0.1011, over 5742383.93 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3274, pruned_loss=0.09069, over 5720650.79 frames. ], batch size: 113, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:26:22,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=980884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:26:39,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.639e+02 1.244e+03 1.572e+03 2.283e+03 3.851e+03, threshold=3.144e+03, percent-clipped=4.0
+2023-03-11 11:26:40,114 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=980905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:26:53,747 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=980919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:27:01,463 INFO [train.py:968] (0/2) Epoch 22, batch 23200, giga_loss[loss=0.236, simple_loss=0.3102, pruned_loss=0.08095, over 28999.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3328, pruned_loss=0.09339, over 5709368.69 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3538, pruned_loss=0.1015, over 5733422.62 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3301, pruned_loss=0.0917, over 5712565.73 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:27:40,596 INFO [train.py:968] (0/2) Epoch 22, batch 23250, giga_loss[loss=0.2989, simple_loss=0.3694, pruned_loss=0.1142, over 27687.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3367, pruned_loss=0.0955, over 5699075.96 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3536, pruned_loss=0.1017, over 5726788.11 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3342, pruned_loss=0.09377, over 5707223.80 frames. ], batch size: 472, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:27:53,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5979, 1.8026, 1.4880, 1.6898], device='cuda:0'), covar=tensor([0.2552, 0.2665, 0.2986, 0.2419], device='cuda:0'), in_proj_covar=tensor([0.1503, 0.1085, 0.1325, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 11:28:01,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.797e+02 1.254e+03 1.605e+03 2.109e+03 6.654e+03, threshold=3.210e+03, percent-clipped=5.0
+2023-03-11 11:28:20,114 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:28:22,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:28:22,614 INFO [train.py:968] (0/2) Epoch 22, batch 23300, giga_loss[loss=0.2544, simple_loss=0.3347, pruned_loss=0.08703, over 28971.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3397, pruned_loss=0.09624, over 5701112.69 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3537, pruned_loss=0.1017, over 5727787.17 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3377, pruned_loss=0.09484, over 5706479.85 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:28:46,168 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:29:04,391 INFO [train.py:968] (0/2) Epoch 22, batch 23350, giga_loss[loss=0.2711, simple_loss=0.3484, pruned_loss=0.09694, over 28949.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.343, pruned_loss=0.0978, over 5711236.20 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3542, pruned_loss=0.1024, over 5731495.50 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3406, pruned_loss=0.09593, over 5711627.31 frames. ], batch size: 106, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:29:07,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:29:23,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.038e+02 1.367e+03 1.760e+03 2.335e+03 4.600e+03, threshold=3.519e+03, percent-clipped=7.0
+2023-03-11 11:29:45,466 INFO [train.py:968] (0/2) Epoch 22, batch 23400, giga_loss[loss=0.2823, simple_loss=0.3468, pruned_loss=0.1089, over 28919.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3455, pruned_loss=0.09915, over 5713460.70 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3547, pruned_loss=0.103, over 5727982.31 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3426, pruned_loss=0.09691, over 5716066.95 frames. ], batch size: 66, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:29:58,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4042, 1.8149, 1.4537, 1.3275], device='cuda:0'), covar=tensor([0.2519, 0.2537, 0.2947, 0.2347], device='cuda:0'), in_proj_covar=tensor([0.1507, 0.1088, 0.1329, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 11:30:17,269 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981168.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:30:28,498 INFO [train.py:968] (0/2) Epoch 22, batch 23450, giga_loss[loss=0.288, simple_loss=0.3666, pruned_loss=0.1048, over 28950.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3494, pruned_loss=0.1025, over 5711555.35 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3546, pruned_loss=0.1031, over 5732064.62 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3471, pruned_loss=0.1006, over 5709814.39 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:30:53,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.964e+02 1.574e+03 1.895e+03 2.320e+03 6.469e+03, threshold=3.789e+03, percent-clipped=7.0
+2023-03-11 11:31:18,788 INFO [train.py:968] (0/2) Epoch 22, batch 23500, giga_loss[loss=0.3311, simple_loss=0.3732, pruned_loss=0.1445, over 23773.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3566, pruned_loss=0.1081, over 5698480.48 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3551, pruned_loss=0.1035, over 5735399.00 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3543, pruned_loss=0.1063, over 5693613.19 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:31:30,563 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-11 11:31:31,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:31:33,059 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:32:07,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:32:08,066 INFO [train.py:968] (0/2) Epoch 22, batch 23550, giga_loss[loss=0.2867, simple_loss=0.3567, pruned_loss=0.1084, over 28761.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3633, pruned_loss=0.113, over 5679402.79 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.355, pruned_loss=0.1035, over 5726948.55 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3616, pruned_loss=0.1116, over 5682903.20 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:32:15,531 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:32:21,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:32:36,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.749e+03 2.321e+03 3.061e+03 9.886e+03, threshold=4.642e+03, percent-clipped=11.0
+2023-03-11 11:32:43,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4431, 3.1051, 1.5065, 1.5661], device='cuda:0'), covar=tensor([0.0912, 0.0340, 0.0872, 0.1230], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0549, 0.0385, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 11:32:58,367 INFO [train.py:968] (0/2) Epoch 22, batch 23600, giga_loss[loss=0.3524, simple_loss=0.4028, pruned_loss=0.1509, over 28639.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.37, pruned_loss=0.1182, over 5683570.18 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3557, pruned_loss=0.104, over 5730415.13 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3683, pruned_loss=0.1169, over 5682414.33 frames. ], batch size: 60, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:33:08,787 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:33:50,407 INFO [train.py:968] (0/2) Epoch 22, batch 23650, giga_loss[loss=0.3831, simple_loss=0.4233, pruned_loss=0.1714, over 28578.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3766, pruned_loss=0.1238, over 5684413.87 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3561, pruned_loss=0.1045, over 5733345.08 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3753, pruned_loss=0.1228, over 5679391.19 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:33:56,579 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7937, 1.8928, 1.7959, 1.6599], device='cuda:0'), covar=tensor([0.2456, 0.2308, 0.2099, 0.2264], device='cuda:0'), in_proj_covar=tensor([0.1971, 0.1906, 0.1823, 0.1970], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 11:34:13,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.300e+03 1.830e+03 2.475e+03 3.610e+03 1.179e+04, threshold=4.950e+03, percent-clipped=16.0
+2023-03-11 11:34:34,232 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:34:37,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:34:41,025 INFO [train.py:968] (0/2) Epoch 22, batch 23700, giga_loss[loss=0.3115, simple_loss=0.3769, pruned_loss=0.123, over 28862.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3805, pruned_loss=0.1274, over 5672923.64 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3563, pruned_loss=0.1047, over 5728770.92 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3799, pruned_loss=0.127, over 5670814.24 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:34:46,268 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:34:48,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:35:02,637 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:35:09,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:35:18,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:35:26,981 INFO [train.py:968] (0/2) Epoch 22, batch 23750, giga_loss[loss=0.3073, simple_loss=0.3742, pruned_loss=0.1202, over 28843.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3819, pruned_loss=0.1296, over 5649666.17 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3565, pruned_loss=0.1049, over 5711337.86 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3816, pruned_loss=0.1294, over 5663123.23 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:35:51,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.171e+03 1.703e+03 2.107e+03 3.202e+03 1.013e+04, threshold=4.214e+03, percent-clipped=8.0
+2023-03-11 11:36:18,951 INFO [train.py:968] (0/2) Epoch 22, batch 23800, giga_loss[loss=0.3737, simple_loss=0.4168, pruned_loss=0.1652, over 28955.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3866, pruned_loss=0.1349, over 5645155.72 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3566, pruned_loss=0.1051, over 5715482.38 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3867, pruned_loss=0.1351, over 5651421.38 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:36:20,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3459, 3.1569, 3.0227, 1.4610], device='cuda:0'), covar=tensor([0.1041, 0.1205, 0.1133, 0.2198], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.1138, 0.0963, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 11:36:31,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981543.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:37:07,313 INFO [train.py:968] (0/2) Epoch 22, batch 23850, giga_loss[loss=0.2885, simple_loss=0.3634, pruned_loss=0.1068, over 28739.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3901, pruned_loss=0.1392, over 5645606.60 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.357, pruned_loss=0.1057, over 5719787.40 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3907, pruned_loss=0.1397, over 5644785.12 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:37:33,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7687, 2.5810, 1.6008, 1.0764], device='cuda:0'), covar=tensor([0.7777, 0.3856, 0.4060, 0.6283], device='cuda:0'), in_proj_covar=tensor([0.1755, 0.1647, 0.1602, 0.1424], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 11:37:33,898 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:37:37,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.213e+03 2.004e+03 2.394e+03 3.729e+03 1.101e+04, threshold=4.787e+03, percent-clipped=18.0
+2023-03-11 11:37:39,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:37:50,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:37:52,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:38:04,169 INFO [train.py:968] (0/2) Epoch 22, batch 23900, giga_loss[loss=0.4557, simple_loss=0.462, pruned_loss=0.2247, over 26593.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3933, pruned_loss=0.141, over 5647479.73 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3576, pruned_loss=0.1062, over 5719931.75 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3941, pruned_loss=0.1417, over 5644760.70 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:38:11,181 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:38:37,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:38:48,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1695, 1.2696, 1.0683, 0.8610], device='cuda:0'), covar=tensor([0.0812, 0.0397, 0.0884, 0.1068], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0449, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 11:38:56,311 INFO [train.py:968] (0/2) Epoch 22, batch 23950, libri_loss[loss=0.3186, simple_loss=0.3822, pruned_loss=0.1275, over 29217.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3928, pruned_loss=0.1413, over 5636657.41 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3578, pruned_loss=0.1065, over 5713643.64 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.394, pruned_loss=0.1423, over 5637931.14 frames. ], batch size: 97, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:39:03,089 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981686.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:39:07,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981689.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:39:23,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.151e+03 1.885e+03 2.493e+03 3.415e+03 9.722e+03, threshold=4.986e+03, percent-clipped=6.0
+2023-03-11 11:39:32,181 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:39:35,054 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981718.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:39:46,108 INFO [train.py:968] (0/2) Epoch 22, batch 24000, giga_loss[loss=0.2976, simple_loss=0.3635, pruned_loss=0.1159, over 28629.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3907, pruned_loss=0.1403, over 5631299.65 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3576, pruned_loss=0.1065, over 5706667.72 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3922, pruned_loss=0.1416, over 5636961.77 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:39:46,113 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 11:39:55,526 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2061, simple_loss=0.3148, pruned_loss=0.04865, over 944034.00 frames. 
+2023-03-11 11:39:55,527 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 11:39:58,646 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6824, 1.7997, 1.6466, 1.4735], device='cuda:0'), covar=tensor([0.2743, 0.2520, 0.2283, 0.2654], device='cuda:0'), in_proj_covar=tensor([0.1980, 0.1921, 0.1833, 0.1982], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 11:40:18,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6590, 1.7469, 1.6325, 1.5360], device='cuda:0'), covar=tensor([0.2205, 0.2331, 0.2024, 0.2161], device='cuda:0'), in_proj_covar=tensor([0.1980, 0.1920, 0.1832, 0.1981], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 11:40:21,220 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:24,362 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:25,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:27,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:33,394 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.99 vs. limit=2.0
+2023-03-11 11:40:40,641 INFO [train.py:968] (0/2) Epoch 22, batch 24050, giga_loss[loss=0.3192, simple_loss=0.3819, pruned_loss=0.1283, over 28714.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3897, pruned_loss=0.1385, over 5631304.80 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3575, pruned_loss=0.1065, over 5699255.83 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3915, pruned_loss=0.14, over 5640892.75 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:40:44,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5223, 4.0519, 1.5916, 1.7318], device='cuda:0'), covar=tensor([0.1018, 0.0356, 0.0894, 0.1307], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0553, 0.0388, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 11:40:51,110 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:53,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:53,863 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2183, 1.1773, 1.1487, 1.4324], device='cuda:0'), covar=tensor([0.0762, 0.0419, 0.0355, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 11:41:00,941 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:41:04,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.133e+03 1.715e+03 2.139e+03 2.810e+03 8.018e+03, threshold=4.277e+03, percent-clipped=2.0
+2023-03-11 11:41:04,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:41:12,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7263, 1.9420, 1.3550, 1.5083], device='cuda:0'), covar=tensor([0.1036, 0.0655, 0.1072, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0448, 0.0520, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 11:41:20,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.37 vs. limit=5.0
+2023-03-11 11:41:30,090 INFO [train.py:968] (0/2) Epoch 22, batch 24100, giga_loss[loss=0.3096, simple_loss=0.3758, pruned_loss=0.1217, over 28564.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3883, pruned_loss=0.1366, over 5631339.59 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3574, pruned_loss=0.1065, over 5703411.78 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3909, pruned_loss=0.1389, over 5633385.85 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:41:32,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4818, 1.7850, 1.4244, 1.4653], device='cuda:0'), covar=tensor([0.2528, 0.2524, 0.2801, 0.2273], device='cuda:0'), in_proj_covar=tensor([0.1502, 0.1086, 0.1327, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 11:41:32,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4790, 1.5852, 1.7091, 1.2978], device='cuda:0'), covar=tensor([0.1611, 0.2381, 0.1339, 0.1630], device='cuda:0'), in_proj_covar=tensor([0.0892, 0.0696, 0.0941, 0.0840], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 11:41:34,743 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:41:34,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5463, 1.7164, 1.6864, 1.4787], device='cuda:0'), covar=tensor([0.2055, 0.2242, 0.2503, 0.2267], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0758, 0.0722, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 11:41:57,561 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:41:59,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:41:59,905 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-11 11:42:19,761 INFO [train.py:968] (0/2) Epoch 22, batch 24150, giga_loss[loss=0.3266, simple_loss=0.3916, pruned_loss=0.1308, over 28946.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3904, pruned_loss=0.138, over 5624564.99 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3577, pruned_loss=0.1067, over 5703871.79 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3927, pruned_loss=0.1401, over 5624654.57 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:42:31,993 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:42:48,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.238e+03 1.947e+03 2.427e+03 3.441e+03 7.856e+03, threshold=4.854e+03, percent-clipped=12.0
+2023-03-11 11:43:15,129 INFO [train.py:968] (0/2) Epoch 22, batch 24200, giga_loss[loss=0.3275, simple_loss=0.3993, pruned_loss=0.1278, over 28774.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3874, pruned_loss=0.1351, over 5622379.43 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3575, pruned_loss=0.1066, over 5704937.44 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3896, pruned_loss=0.1371, over 5621203.14 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:44:05,307 INFO [train.py:968] (0/2) Epoch 22, batch 24250, giga_loss[loss=0.2697, simple_loss=0.3532, pruned_loss=0.09314, over 28773.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3845, pruned_loss=0.1312, over 5619523.09 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3577, pruned_loss=0.1068, over 5695859.53 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3863, pruned_loss=0.1327, over 5626330.53 frames. ], batch size: 243, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:44:23,148 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-982000.pt
+2023-03-11 11:44:29,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.813e+03 2.263e+03 3.239e+03 7.874e+03, threshold=4.526e+03, percent-clipped=8.0
+2023-03-11 11:44:41,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:44:51,760 INFO [train.py:968] (0/2) Epoch 22, batch 24300, giga_loss[loss=0.3285, simple_loss=0.3937, pruned_loss=0.1317, over 28731.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3815, pruned_loss=0.1284, over 5644212.47 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3579, pruned_loss=0.1073, over 5701692.12 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3838, pruned_loss=0.1302, over 5641666.60 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:45:37,176 INFO [train.py:968] (0/2) Epoch 22, batch 24350, giga_loss[loss=0.2852, simple_loss=0.3598, pruned_loss=0.1053, over 28910.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3791, pruned_loss=0.126, over 5652209.60 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3582, pruned_loss=0.1077, over 5692835.35 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3811, pruned_loss=0.1274, over 5657217.26 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:46:01,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.864e+03 2.358e+03 3.142e+03 1.579e+04, threshold=4.715e+03, percent-clipped=16.0
+2023-03-11 11:46:06,072 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:46:09,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0859, 1.0344, 3.3959, 3.0702], device='cuda:0'), covar=tensor([0.1946, 0.3018, 0.0822, 0.1206], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0650, 0.0966, 0.0912], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 11:46:20,855 INFO [train.py:968] (0/2) Epoch 22, batch 24400, giga_loss[loss=0.2848, simple_loss=0.3536, pruned_loss=0.1081, over 28612.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3764, pruned_loss=0.1249, over 5646266.18 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3584, pruned_loss=0.1081, over 5692882.47 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3787, pruned_loss=0.1265, over 5648097.67 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:46:24,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3943, 1.6135, 1.4604, 1.4392], device='cuda:0'), covar=tensor([0.1646, 0.1850, 0.2122, 0.1846], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0752, 0.0716, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 11:46:59,258 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:47:09,876 INFO [train.py:968] (0/2) Epoch 22, batch 24450, giga_loss[loss=0.3152, simple_loss=0.3789, pruned_loss=0.1258, over 28717.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3756, pruned_loss=0.1239, over 5666384.73 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3581, pruned_loss=0.1079, over 5694075.06 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3777, pruned_loss=0.1253, over 5666630.44 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:47:13,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4540, 1.7322, 1.5497, 1.5982], device='cuda:0'), covar=tensor([0.0612, 0.0284, 0.0270, 0.0658], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 11:47:41,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.191e+03 1.649e+03 2.253e+03 2.879e+03 6.129e+03, threshold=4.506e+03, percent-clipped=5.0
+2023-03-11 11:48:04,222 INFO [train.py:968] (0/2) Epoch 22, batch 24500, libri_loss[loss=0.2623, simple_loss=0.3216, pruned_loss=0.1014, over 28546.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3749, pruned_loss=0.1234, over 5667546.26 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3582, pruned_loss=0.1082, over 5697330.65 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3769, pruned_loss=0.1247, over 5664247.58 frames. ], batch size: 63, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:48:18,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3464, 1.4950, 1.6128, 1.2331], device='cuda:0'), covar=tensor([0.1309, 0.2038, 0.1116, 0.1524], device='cuda:0'), in_proj_covar=tensor([0.0892, 0.0697, 0.0941, 0.0839], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 11:48:26,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3639, 1.2369, 4.0253, 3.2849], device='cuda:0'), covar=tensor([0.1725, 0.2956, 0.0507, 0.0955], device='cuda:0'), in_proj_covar=tensor([0.0762, 0.0651, 0.0968, 0.0913], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 11:48:56,573 INFO [train.py:968] (0/2) Epoch 22, batch 24550, giga_loss[loss=0.2834, simple_loss=0.3627, pruned_loss=0.102, over 28677.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3733, pruned_loss=0.1211, over 5667680.37 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3586, pruned_loss=0.1085, over 5698665.51 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3747, pruned_loss=0.122, over 5663475.79 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:49:26,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.738e+02 1.493e+03 2.010e+03 3.101e+03 8.620e+03, threshold=4.021e+03, percent-clipped=9.0
+2023-03-11 11:49:47,054 INFO [train.py:968] (0/2) Epoch 22, batch 24600, giga_loss[loss=0.4519, simple_loss=0.4611, pruned_loss=0.2213, over 26820.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3733, pruned_loss=0.1188, over 5670128.12 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3584, pruned_loss=0.1085, over 5701422.08 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3751, pruned_loss=0.1198, over 5663644.98 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:50:35,797 INFO [train.py:968] (0/2) Epoch 22, batch 24650, giga_loss[loss=0.3375, simple_loss=0.3912, pruned_loss=0.1419, over 27988.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3737, pruned_loss=0.1191, over 5670528.30 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3582, pruned_loss=0.1086, over 5708585.11 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3759, pruned_loss=0.1202, over 5657730.99 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:50:49,402 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=982395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:51:00,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.820e+03 2.279e+03 3.177e+03 6.763e+03, threshold=4.558e+03, percent-clipped=8.0
+2023-03-11 11:51:23,272 INFO [train.py:968] (0/2) Epoch 22, batch 24700, giga_loss[loss=0.3108, simple_loss=0.3654, pruned_loss=0.1281, over 23496.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3736, pruned_loss=0.1201, over 5670647.81 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3576, pruned_loss=0.1087, over 5715532.99 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3766, pruned_loss=0.1214, over 5652335.45 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:51:47,413 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 11:52:10,765 INFO [train.py:968] (0/2) Epoch 22, batch 24750, giga_loss[loss=0.3266, simple_loss=0.3872, pruned_loss=0.133, over 28863.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3727, pruned_loss=0.1202, over 5662266.05 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.357, pruned_loss=0.1085, over 5719192.62 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3762, pruned_loss=0.1219, over 5642853.60 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:52:13,973 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-11 11:52:17,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=982488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:52:23,862 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-11 11:52:34,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.210e+03 1.602e+03 2.148e+03 2.987e+03 6.334e+03, threshold=4.295e+03, percent-clipped=6.0
+2023-03-11 11:52:54,536 INFO [train.py:968] (0/2) Epoch 22, batch 24800, giga_loss[loss=0.3186, simple_loss=0.3812, pruned_loss=0.128, over 28638.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3714, pruned_loss=0.1202, over 5661746.92 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3577, pruned_loss=0.1091, over 5714771.42 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3741, pruned_loss=0.1213, over 5648609.03 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:53:01,709 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=982538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:53:05,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=982541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:53:08,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=982544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:53:30,069 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=982570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:53:34,749 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.65 vs. limit=5.0
+2023-03-11 11:53:39,119 INFO [train.py:968] (0/2) Epoch 22, batch 24850, giga_loss[loss=0.2732, simple_loss=0.353, pruned_loss=0.09666, over 29068.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3703, pruned_loss=0.1197, over 5671556.05 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3576, pruned_loss=0.1091, over 5716745.37 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3726, pruned_loss=0.1206, over 5659022.65 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:54:05,909 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.577e+02 1.752e+03 2.266e+03 3.106e+03 8.648e+03, threshold=4.531e+03, percent-clipped=10.0
+2023-03-11 11:54:23,631 INFO [train.py:968] (0/2) Epoch 22, batch 24900, giga_loss[loss=0.2939, simple_loss=0.3681, pruned_loss=0.1098, over 28644.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3705, pruned_loss=0.119, over 5671344.07 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3576, pruned_loss=0.1092, over 5709537.84 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3725, pruned_loss=0.1199, over 5666830.52 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:54:23,970 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=982631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:54:26,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=982634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:54:56,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=982663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:55:12,063 INFO [train.py:968] (0/2) Epoch 22, batch 24950, giga_loss[loss=0.2961, simple_loss=0.3634, pruned_loss=0.1144, over 29095.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3701, pruned_loss=0.118, over 5669860.96 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3575, pruned_loss=0.1092, over 5712251.73 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3719, pruned_loss=0.1188, over 5663067.57 frames. ], batch size: 113, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:55:12,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3271, 3.6020, 1.4734, 1.5559], device='cuda:0'), covar=tensor([0.1054, 0.0424, 0.0964, 0.1393], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0557, 0.0389, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 11:55:19,244 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=982687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:55:20,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3734, 2.0227, 1.5250, 0.5994], device='cuda:0'), covar=tensor([0.5337, 0.2955, 0.4239, 0.6342], device='cuda:0'), in_proj_covar=tensor([0.1763, 0.1649, 0.1604, 0.1427], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 11:55:21,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=982690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:55:39,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.988e+02 1.567e+03 1.953e+03 2.827e+03 5.667e+03, threshold=3.907e+03, percent-clipped=3.0
+2023-03-11 11:55:47,461 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=982719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:55:55,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1465, 1.6131, 5.5796, 3.9463], device='cuda:0'), covar=tensor([0.1594, 0.2698, 0.0428, 0.0760], device='cuda:0'), in_proj_covar=tensor([0.0766, 0.0654, 0.0974, 0.0919], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 11:55:58,326 INFO [train.py:968] (0/2) Epoch 22, batch 25000, giga_loss[loss=0.2979, simple_loss=0.3614, pruned_loss=0.1172, over 28632.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3706, pruned_loss=0.1181, over 5662552.53 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3579, pruned_loss=0.1096, over 5705553.84 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.372, pruned_loss=0.1186, over 5662939.93 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:56:03,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9890, 3.0133, 1.1574, 1.3487], device='cuda:0'), covar=tensor([0.1309, 0.0484, 0.1034, 0.1601], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0557, 0.0390, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 11:56:10,676 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:56:46,008 INFO [train.py:968] (0/2) Epoch 22, batch 25050, giga_loss[loss=0.2777, simple_loss=0.3503, pruned_loss=0.1026, over 28664.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3701, pruned_loss=0.118, over 5671587.05 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3582, pruned_loss=0.1096, over 5705660.09 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3712, pruned_loss=0.1185, over 5671291.63 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:57:12,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.773e+02 1.701e+03 2.120e+03 2.869e+03 5.778e+03, threshold=4.240e+03, percent-clipped=7.0
+2023-03-11 11:57:31,419 INFO [train.py:968] (0/2) Epoch 22, batch 25100, giga_loss[loss=0.3095, simple_loss=0.3748, pruned_loss=0.1221, over 28913.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3691, pruned_loss=0.1181, over 5674016.79 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3581, pruned_loss=0.11, over 5702503.21 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3703, pruned_loss=0.1184, over 5676397.77 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:57:52,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:58:22,881 INFO [train.py:968] (0/2) Epoch 22, batch 25150, giga_loss[loss=0.309, simple_loss=0.3756, pruned_loss=0.1213, over 28896.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3686, pruned_loss=0.1186, over 5675144.98 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3583, pruned_loss=0.1101, over 5699691.69 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3695, pruned_loss=0.1188, over 5679170.96 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:58:27,715 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 11:58:38,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3786, 1.8843, 1.5618, 1.5390], device='cuda:0'), covar=tensor([0.0698, 0.0377, 0.0307, 0.0737], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 11:58:47,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.273e+03 1.747e+03 2.262e+03 2.968e+03 8.493e+03, threshold=4.525e+03, percent-clipped=10.0
+2023-03-11 11:59:05,644 INFO [train.py:968] (0/2) Epoch 22, batch 25200, giga_loss[loss=0.306, simple_loss=0.3702, pruned_loss=0.1209, over 28566.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3665, pruned_loss=0.1173, over 5690215.53 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3582, pruned_loss=0.1099, over 5707138.10 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3676, pruned_loss=0.118, over 5685815.77 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:59:06,447 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2596, 1.2530, 3.7860, 3.1549], device='cuda:0'), covar=tensor([0.1658, 0.2784, 0.0456, 0.1201], device='cuda:0'), in_proj_covar=tensor([0.0768, 0.0655, 0.0976, 0.0920], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 11:59:29,646 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982954.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:59:53,223 INFO [train.py:968] (0/2) Epoch 22, batch 25250, giga_loss[loss=0.2678, simple_loss=0.3439, pruned_loss=0.09591, over 29002.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3645, pruned_loss=0.1166, over 5678487.95 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3583, pruned_loss=0.1099, over 5697830.84 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3654, pruned_loss=0.1172, over 5683237.37 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:59:56,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5641, 1.8255, 1.7205, 1.6674], device='cuda:0'), covar=tensor([0.2097, 0.2179, 0.2469, 0.2122], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0753, 0.0716, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 12:00:18,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.649e+03 2.295e+03 3.046e+03 5.573e+03, threshold=4.591e+03, percent-clipped=6.0
+2023-03-11 12:00:40,547 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6904, 1.9899, 1.8915, 1.6257], device='cuda:0'), covar=tensor([0.2466, 0.2003, 0.1707, 0.1967], device='cuda:0'), in_proj_covar=tensor([0.1985, 0.1923, 0.1836, 0.1987], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 12:00:40,922 INFO [train.py:968] (0/2) Epoch 22, batch 25300, libri_loss[loss=0.2854, simple_loss=0.3408, pruned_loss=0.115, over 29382.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3655, pruned_loss=0.1181, over 5673837.44 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3591, pruned_loss=0.1106, over 5695623.19 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3658, pruned_loss=0.1182, over 5679083.72 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:01:30,976 INFO [train.py:968] (0/2) Epoch 22, batch 25350, giga_loss[loss=0.3388, simple_loss=0.3983, pruned_loss=0.1396, over 28492.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3658, pruned_loss=0.1182, over 5675946.24 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3591, pruned_loss=0.1107, over 5697934.34 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3661, pruned_loss=0.1182, over 5677877.52 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:01:57,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.653e+02 1.745e+03 2.092e+03 2.800e+03 5.320e+03, threshold=4.184e+03, percent-clipped=2.0
+2023-03-11 12:02:06,610 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=983119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:02:15,677 INFO [train.py:968] (0/2) Epoch 22, batch 25400, giga_loss[loss=0.3039, simple_loss=0.3829, pruned_loss=0.1124, over 28667.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3662, pruned_loss=0.1172, over 5684858.42 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3589, pruned_loss=0.1106, over 5701355.89 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3666, pruned_loss=0.1175, over 5683169.18 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:02:20,574 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-11 12:02:49,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2301, 1.3167, 1.4699, 1.0668], device='cuda:0'), covar=tensor([0.1630, 0.2856, 0.1316, 0.1482], device='cuda:0'), in_proj_covar=tensor([0.0897, 0.0703, 0.0946, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 12:03:04,214 INFO [train.py:968] (0/2) Epoch 22, batch 25450, libri_loss[loss=0.2672, simple_loss=0.3444, pruned_loss=0.09494, over 28568.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3664, pruned_loss=0.1165, over 5690040.84 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.359, pruned_loss=0.1108, over 5703597.37 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3667, pruned_loss=0.1166, over 5686470.10 frames. ], batch size: 106, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:03:14,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 12:03:20,537 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 12:03:34,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.841e+02 1.668e+03 2.146e+03 3.179e+03 7.553e+03, threshold=4.291e+03, percent-clipped=9.0
+2023-03-11 12:03:49,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=983226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:03:52,392 INFO [train.py:968] (0/2) Epoch 22, batch 25500, giga_loss[loss=0.3463, simple_loss=0.3929, pruned_loss=0.1499, over 28622.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3674, pruned_loss=0.1175, over 5683301.83 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3594, pruned_loss=0.1111, over 5704190.38 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3675, pruned_loss=0.1174, over 5679865.55 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:04:22,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=983262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:04:25,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=983265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:04:40,437 INFO [train.py:968] (0/2) Epoch 22, batch 25550, libri_loss[loss=0.2768, simple_loss=0.3443, pruned_loss=0.1046, over 29562.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3705, pruned_loss=0.1203, over 5677962.47 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3593, pruned_loss=0.1111, over 5698211.34 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3707, pruned_loss=0.1203, over 5680359.20 frames. ], batch size: 75, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:04:52,914 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=983294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:05:08,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.304e+03 2.013e+03 2.458e+03 3.258e+03 6.385e+03, threshold=4.917e+03, percent-clipped=12.0
+2023-03-11 12:05:23,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=983325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:05:27,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=983329.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 12:05:29,843 INFO [train.py:968] (0/2) Epoch 22, batch 25600, giga_loss[loss=0.4071, simple_loss=0.4256, pruned_loss=0.1943, over 26686.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.371, pruned_loss=0.1219, over 5676950.42 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3591, pruned_loss=0.111, over 5702149.47 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3716, pruned_loss=0.1222, over 5674949.26 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:06:07,060 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=983369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:06:09,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=983372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:06:16,386 INFO [train.py:968] (0/2) Epoch 22, batch 25650, giga_loss[loss=0.3127, simple_loss=0.3716, pruned_loss=0.1269, over 28882.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3723, pruned_loss=0.1242, over 5678925.01 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3589, pruned_loss=0.1109, over 5707496.14 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3736, pruned_loss=0.125, over 5671129.76 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:06:38,048 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=983401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:06:46,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.184e+03 1.951e+03 2.580e+03 3.453e+03 9.373e+03, threshold=5.160e+03, percent-clipped=5.0
+2023-03-11 12:06:56,777 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=983422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:07:04,024 INFO [train.py:968] (0/2) Epoch 22, batch 25700, giga_loss[loss=0.2455, simple_loss=0.3318, pruned_loss=0.07962, over 28995.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.373, pruned_loss=0.125, over 5686121.03 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.359, pruned_loss=0.1111, over 5707631.32 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3742, pruned_loss=0.1257, over 5679495.11 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:07:40,552 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=983472.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 12:07:43,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=983475.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 12:07:47,614 INFO [train.py:968] (0/2) Epoch 22, batch 25750, giga_loss[loss=0.3295, simple_loss=0.3891, pruned_loss=0.1349, over 28748.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3716, pruned_loss=0.1242, over 5676518.53 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3591, pruned_loss=0.1111, over 5708898.47 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3728, pruned_loss=0.1251, over 5669345.13 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:07:51,951 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 12:08:12,614 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=983504.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 12:08:17,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.770e+03 2.237e+03 2.878e+03 8.358e+03, threshold=4.475e+03, percent-clipped=5.0
+2023-03-11 12:08:33,867 INFO [train.py:968] (0/2) Epoch 22, batch 25800, giga_loss[loss=0.2927, simple_loss=0.369, pruned_loss=0.1082, over 28612.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3716, pruned_loss=0.1227, over 5683701.54 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3594, pruned_loss=0.1113, over 5711882.73 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3724, pruned_loss=0.1234, over 5674896.03 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:09:00,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9982, 1.3177, 1.1180, 0.2126], device='cuda:0'), covar=tensor([0.4476, 0.3727, 0.4829, 0.7593], device='cuda:0'), in_proj_covar=tensor([0.1764, 0.1658, 0.1607, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 12:09:21,648 INFO [train.py:968] (0/2) Epoch 22, batch 25850, giga_loss[loss=0.2862, simple_loss=0.3578, pruned_loss=0.1073, over 28756.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3701, pruned_loss=0.1208, over 5674661.74 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3596, pruned_loss=0.1115, over 5712840.32 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5666855.97 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:09:46,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8972, 5.7278, 5.4078, 2.9399], device='cuda:0'), covar=tensor([0.0481, 0.0597, 0.0670, 0.1525], device='cuda:0'), in_proj_covar=tensor([0.1244, 0.1150, 0.0973, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 12:09:49,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.741e+03 2.248e+03 3.237e+03 8.593e+03, threshold=4.496e+03, percent-clipped=14.0
+2023-03-11 12:10:09,482 INFO [train.py:968] (0/2) Epoch 22, batch 25900, giga_loss[loss=0.3318, simple_loss=0.3875, pruned_loss=0.138, over 28689.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1201, over 5672053.11 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3603, pruned_loss=0.1121, over 5716428.32 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1201, over 5661888.34 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:10:25,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4649, 3.7742, 1.5534, 1.6972], device='cuda:0'), covar=tensor([0.0974, 0.0375, 0.0914, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0560, 0.0390, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 12:10:53,594 INFO [train.py:968] (0/2) Epoch 22, batch 25950, giga_loss[loss=0.3283, simple_loss=0.3615, pruned_loss=0.1475, over 23553.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3666, pruned_loss=0.1194, over 5673594.41 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3601, pruned_loss=0.1119, over 5718867.08 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.367, pruned_loss=0.1198, over 5661950.37 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:11:11,392 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=983700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:11:23,915 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.756e+03 2.484e+03 3.386e+03 1.682e+04, threshold=4.968e+03, percent-clipped=13.0
+2023-03-11 12:11:37,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4916, 1.7245, 1.4308, 1.6371], device='cuda:0'), covar=tensor([0.0737, 0.0352, 0.0333, 0.0788], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 12:11:40,956 INFO [train.py:968] (0/2) Epoch 22, batch 26000, giga_loss[loss=0.2897, simple_loss=0.3636, pruned_loss=0.1079, over 28940.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3676, pruned_loss=0.1211, over 5652944.32 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3602, pruned_loss=0.1123, over 5707259.36 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.368, pruned_loss=0.1214, over 5651837.10 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:12:10,070 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 12:12:13,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5634, 1.8553, 1.4365, 1.8019], device='cuda:0'), covar=tensor([0.2556, 0.2587, 0.2986, 0.2313], device='cuda:0'), in_proj_covar=tensor([0.1505, 0.1088, 0.1327, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 12:12:15,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5898, 1.7806, 1.7755, 1.3212], device='cuda:0'), covar=tensor([0.1755, 0.2706, 0.1492, 0.1847], device='cuda:0'), in_proj_covar=tensor([0.0895, 0.0701, 0.0944, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 12:12:24,188 INFO [train.py:968] (0/2) Epoch 22, batch 26050, giga_loss[loss=0.3248, simple_loss=0.3884, pruned_loss=0.1306, over 28860.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3688, pruned_loss=0.1217, over 5648618.68 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.36, pruned_loss=0.1123, over 5696938.74 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3698, pruned_loss=0.1224, over 5654549.52 frames. ], batch size: 66, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:12:38,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=983797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:12:51,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.552e+02 1.763e+03 2.235e+03 3.395e+03 1.002e+04, threshold=4.470e+03, percent-clipped=5.0
+2023-03-11 12:13:06,592 INFO [train.py:968] (0/2) Epoch 22, batch 26100, libri_loss[loss=0.2953, simple_loss=0.3684, pruned_loss=0.1111, over 29162.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3724, pruned_loss=0.1222, over 5657923.13 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3597, pruned_loss=0.1122, over 5702289.90 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.374, pruned_loss=0.1233, over 5655759.61 frames. ], batch size: 101, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:13:15,108 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-11 12:13:16,906 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=983843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:13:19,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=983846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:13:44,147 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=983875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:13:46,254 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=983878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:13:48,320 INFO [train.py:968] (0/2) Epoch 22, batch 26150, giga_loss[loss=0.3047, simple_loss=0.3803, pruned_loss=0.1145, over 28628.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3727, pruned_loss=0.1204, over 5665935.57 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3592, pruned_loss=0.1121, over 5703983.72 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3751, pruned_loss=0.1218, over 5660557.57 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:14:18,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.323e+02 1.544e+03 2.191e+03 3.434e+03 1.120e+04, threshold=4.383e+03, percent-clipped=12.0
+2023-03-11 12:14:35,399 INFO [train.py:968] (0/2) Epoch 22, batch 26200, giga_loss[loss=0.3092, simple_loss=0.3746, pruned_loss=0.1219, over 28880.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3743, pruned_loss=0.1218, over 5657869.62 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3596, pruned_loss=0.1125, over 5700203.73 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3761, pruned_loss=0.1228, over 5656302.97 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:14:46,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=983940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:14:51,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=983943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:14:52,836 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=983946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:15:16,010 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=983972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:15:22,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3352, 1.0826, 4.1316, 3.4379], device='cuda:0'), covar=tensor([0.1717, 0.2885, 0.0473, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0652, 0.0970, 0.0916], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 12:15:24,144 INFO [train.py:968] (0/2) Epoch 22, batch 26250, giga_loss[loss=0.3628, simple_loss=0.4117, pruned_loss=0.1569, over 28261.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3767, pruned_loss=0.124, over 5650375.01 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.36, pruned_loss=0.1127, over 5699355.49 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3779, pruned_loss=0.1246, over 5649527.48 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:15:39,073 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-984000.pt
+2023-03-11 12:15:54,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.229e+03 1.762e+03 2.122e+03 3.271e+03 6.998e+03, threshold=4.243e+03, percent-clipped=7.0
+2023-03-11 12:16:14,010 INFO [train.py:968] (0/2) Epoch 22, batch 26300, giga_loss[loss=0.2655, simple_loss=0.3455, pruned_loss=0.09278, over 28894.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3758, pruned_loss=0.1242, over 5643070.06 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3603, pruned_loss=0.113, over 5698780.72 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3768, pruned_loss=0.1246, over 5642309.33 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:16:17,978 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2962, 1.5493, 1.5392, 1.4064], device='cuda:0'), covar=tensor([0.1937, 0.1704, 0.2554, 0.1831], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0748, 0.0715, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 12:17:03,425 INFO [train.py:968] (0/2) Epoch 22, batch 26350, giga_loss[loss=0.2885, simple_loss=0.3641, pruned_loss=0.1064, over 29059.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3756, pruned_loss=0.1247, over 5642943.67 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3604, pruned_loss=0.113, over 5700903.78 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3765, pruned_loss=0.1253, over 5639631.00 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:17:06,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9118, 1.1012, 1.0450, 0.8481], device='cuda:0'), covar=tensor([0.2022, 0.2354, 0.1443, 0.1987], device='cuda:0'), in_proj_covar=tensor([0.1979, 0.1922, 0.1834, 0.1982], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 12:17:32,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.799e+02 1.740e+03 2.407e+03 3.663e+03 1.132e+04, threshold=4.814e+03, percent-clipped=14.0
+2023-03-11 12:17:50,021 INFO [train.py:968] (0/2) Epoch 22, batch 26400, giga_loss[loss=0.2247, simple_loss=0.3035, pruned_loss=0.07295, over 28444.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3728, pruned_loss=0.1235, over 5651494.15 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3601, pruned_loss=0.1128, over 5701747.26 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3738, pruned_loss=0.1241, over 5648010.40 frames. ], batch size: 60, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:18:23,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2510, 2.4012, 1.3119, 1.3677], device='cuda:0'), covar=tensor([0.0968, 0.0388, 0.0842, 0.1345], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0558, 0.0388, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 12:18:43,465 INFO [train.py:968] (0/2) Epoch 22, batch 26450, giga_loss[loss=0.3169, simple_loss=0.3813, pruned_loss=0.1263, over 28908.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3727, pruned_loss=0.1245, over 5646723.13 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3601, pruned_loss=0.1128, over 5701747.26 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3735, pruned_loss=0.125, over 5644011.68 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:18:43,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3999, 1.3969, 3.9867, 3.2723], device='cuda:0'), covar=tensor([0.1608, 0.2626, 0.0461, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.0766, 0.0653, 0.0970, 0.0915], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 12:18:58,891 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 12:19:00,784 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:19:15,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.631e+02 1.769e+03 2.184e+03 3.064e+03 7.652e+03, threshold=4.369e+03, percent-clipped=6.0
+2023-03-11 12:19:32,506 INFO [train.py:968] (0/2) Epoch 22, batch 26500, giga_loss[loss=0.323, simple_loss=0.3836, pruned_loss=0.1312, over 28880.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3744, pruned_loss=0.1261, over 5645481.43 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1132, over 5701512.37 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3749, pruned_loss=0.1263, over 5642547.22 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:19:46,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8314, 1.0934, 2.8443, 2.8199], device='cuda:0'), covar=tensor([0.1697, 0.2661, 0.0659, 0.1007], device='cuda:0'), in_proj_covar=tensor([0.0768, 0.0655, 0.0973, 0.0918], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 12:19:52,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:20:18,103 INFO [train.py:968] (0/2) Epoch 22, batch 26550, giga_loss[loss=0.3315, simple_loss=0.3892, pruned_loss=0.1369, over 27614.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3737, pruned_loss=0.1259, over 5650512.55 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3606, pruned_loss=0.1133, over 5702142.19 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3741, pruned_loss=0.1261, over 5647263.16 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:20:46,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.532e+02 1.902e+03 2.504e+03 3.451e+03 8.746e+03, threshold=5.008e+03, percent-clipped=13.0
+2023-03-11 12:20:54,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:21:01,672 INFO [train.py:968] (0/2) Epoch 22, batch 26600, giga_loss[loss=0.2599, simple_loss=0.3366, pruned_loss=0.09159, over 28953.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3711, pruned_loss=0.1239, over 5671778.13 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3609, pruned_loss=0.1135, over 5706757.59 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3715, pruned_loss=0.1242, over 5663924.11 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:21:11,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6974, 1.8442, 1.2994, 1.4869], device='cuda:0'), covar=tensor([0.0776, 0.0448, 0.0975, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0448, 0.0518, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 12:21:16,303 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-11 12:21:48,767 INFO [train.py:968] (0/2) Epoch 22, batch 26650, libri_loss[loss=0.3471, simple_loss=0.4064, pruned_loss=0.1439, over 29234.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3712, pruned_loss=0.1239, over 5674772.28 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3614, pruned_loss=0.1138, over 5710970.07 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3714, pruned_loss=0.1242, over 5663240.75 frames. ], batch size: 97, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:22:01,094 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=984396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:22:04,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=984399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:22:17,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.668e+03 2.200e+03 2.717e+03 7.477e+03, threshold=4.400e+03, percent-clipped=2.0
+2023-03-11 12:22:22,908 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:22:32,829 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=984428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:22:34,836 INFO [train.py:968] (0/2) Epoch 22, batch 26700, giga_loss[loss=0.3075, simple_loss=0.3837, pruned_loss=0.1157, over 28949.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3724, pruned_loss=0.1236, over 5671822.10 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3614, pruned_loss=0.1139, over 5710459.60 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3728, pruned_loss=0.124, over 5662508.83 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:23:07,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=984464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:23:13,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=984467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:23:26,155 INFO [train.py:968] (0/2) Epoch 22, batch 26750, libri_loss[loss=0.3298, simple_loss=0.3935, pruned_loss=0.1331, over 29358.00 frames. ], tot_loss[loss=0.311, simple_loss=0.373, pruned_loss=0.1245, over 5648276.05 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3616, pruned_loss=0.1141, over 5700674.90 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3734, pruned_loss=0.1248, over 5648956.07 frames. ], batch size: 92, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:23:38,967 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=984496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:23:53,431 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:23:56,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.621e+03 2.068e+03 2.699e+03 4.684e+03, threshold=4.135e+03, percent-clipped=2.0
+2023-03-11 12:24:10,044 INFO [train.py:968] (0/2) Epoch 22, batch 26800, libri_loss[loss=0.3479, simple_loss=0.403, pruned_loss=0.1464, over 29481.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3744, pruned_loss=0.1256, over 5655746.09 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.362, pruned_loss=0.1143, over 5701091.66 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3746, pruned_loss=0.1259, over 5655011.17 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 12:24:27,968 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:24:34,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2880, 1.5968, 1.2370, 1.0102], device='cuda:0'), covar=tensor([0.2891, 0.2830, 0.3365, 0.2494], device='cuda:0'), in_proj_covar=tensor([0.1505, 0.1090, 0.1330, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 12:24:47,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:24:55,021 INFO [train.py:968] (0/2) Epoch 22, batch 26850, giga_loss[loss=0.2828, simple_loss=0.3634, pruned_loss=0.1012, over 28855.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3746, pruned_loss=0.1226, over 5668741.89 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.362, pruned_loss=0.1143, over 5702123.20 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3749, pruned_loss=0.1229, over 5667089.49 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:25:18,282 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:25:27,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.389e+02 1.569e+03 2.113e+03 3.232e+03 8.020e+03, threshold=4.226e+03, percent-clipped=11.0
+2023-03-11 12:25:27,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:25:42,932 INFO [train.py:968] (0/2) Epoch 22, batch 26900, giga_loss[loss=0.2628, simple_loss=0.3547, pruned_loss=0.08549, over 29075.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.376, pruned_loss=0.1213, over 5672751.28 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.362, pruned_loss=0.1143, over 5704974.48 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3763, pruned_loss=0.1216, over 5668497.43 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:26:24,513 INFO [train.py:968] (0/2) Epoch 22, batch 26950, giga_loss[loss=0.2828, simple_loss=0.3558, pruned_loss=0.1049, over 28899.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3788, pruned_loss=0.1233, over 5669428.41 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.362, pruned_loss=0.1143, over 5700383.49 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3795, pruned_loss=0.1238, over 5669706.28 frames. ], batch size: 243, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:26:58,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.725e+02 1.571e+03 2.086e+03 2.807e+03 6.787e+03, threshold=4.171e+03, percent-clipped=7.0
+2023-03-11 12:26:59,143 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=984715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:27:01,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=984718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:27:12,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1396, 2.2298, 1.6766, 1.8596], device='cuda:0'), covar=tensor([0.0942, 0.0715, 0.1011, 0.1178], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0450, 0.0520, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 12:27:15,998 INFO [train.py:968] (0/2) Epoch 22, batch 27000, giga_loss[loss=0.3477, simple_loss=0.4017, pruned_loss=0.1469, over 28871.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3819, pruned_loss=0.1267, over 5670219.34 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.362, pruned_loss=0.1142, over 5701102.57 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3825, pruned_loss=0.1272, over 5669689.21 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:27:16,002 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 12:27:24,930 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2059, simple_loss=0.3133, pruned_loss=0.0492, over 944034.00 frames. 
+2023-03-11 12:27:24,931 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 12:27:34,256 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3969, 2.0483, 1.4063, 0.5784], device='cuda:0'), covar=tensor([0.5265, 0.2938, 0.4555, 0.6395], device='cuda:0'), in_proj_covar=tensor([0.1750, 0.1645, 0.1594, 0.1423], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 12:27:39,639 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=984747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:28:11,978 INFO [train.py:968] (0/2) Epoch 22, batch 27050, libri_loss[loss=0.2732, simple_loss=0.342, pruned_loss=0.1022, over 29580.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3821, pruned_loss=0.1274, over 5675841.73 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3629, pruned_loss=0.1148, over 5698672.09 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3825, pruned_loss=0.1277, over 5676508.63 frames. ], batch size: 76, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:28:13,815 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.38 vs. limit=5.0
+2023-03-11 12:28:23,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:28:41,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.158e+03 1.768e+03 2.291e+03 3.263e+03 8.454e+03, threshold=4.582e+03, percent-clipped=9.0
+2023-03-11 12:28:56,251 INFO [train.py:968] (0/2) Epoch 22, batch 27100, giga_loss[loss=0.32, simple_loss=0.3866, pruned_loss=0.1267, over 28676.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3822, pruned_loss=0.1287, over 5673389.53 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3627, pruned_loss=0.1149, over 5705454.59 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3835, pruned_loss=0.1294, over 5666594.05 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:29:11,246 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:29:27,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5527, 2.2766, 1.5940, 0.8354], device='cuda:0'), covar=tensor([0.5031, 0.2574, 0.3347, 0.4835], device='cuda:0'), in_proj_covar=tensor([0.1757, 0.1649, 0.1600, 0.1428], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 12:29:44,815 INFO [train.py:968] (0/2) Epoch 22, batch 27150, libri_loss[loss=0.2965, simple_loss=0.3589, pruned_loss=0.117, over 20120.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3791, pruned_loss=0.1252, over 5675435.36 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3625, pruned_loss=0.1147, over 5699410.74 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3808, pruned_loss=0.1262, over 5675491.40 frames. ], batch size: 188, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:29:48,002 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:30:12,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.773e+02 1.752e+03 2.198e+03 3.094e+03 5.562e+03, threshold=4.395e+03, percent-clipped=3.0
+2023-03-11 12:30:21,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:30:25,692 INFO [train.py:968] (0/2) Epoch 22, batch 27200, giga_loss[loss=0.274, simple_loss=0.3653, pruned_loss=0.09133, over 28952.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.38, pruned_loss=0.1246, over 5670886.33 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3627, pruned_loss=0.115, over 5703232.62 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3816, pruned_loss=0.1254, over 5666789.69 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 12:30:33,506 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=984935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:30:37,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=984938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:30:40,426 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:31:04,165 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=984967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:31:14,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:31:17,458 INFO [train.py:968] (0/2) Epoch 22, batch 27250, libri_loss[loss=0.2919, simple_loss=0.3479, pruned_loss=0.118, over 29575.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3811, pruned_loss=0.1251, over 5670129.54 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3629, pruned_loss=0.1152, over 5706991.33 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3826, pruned_loss=0.1257, over 5662496.32 frames. ], batch size: 74, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:31:22,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:31:50,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.878e+02 1.683e+03 2.595e+03 3.902e+03 8.579e+03, threshold=5.190e+03, percent-clipped=14.0
+2023-03-11 12:32:05,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:32:08,484 INFO [train.py:968] (0/2) Epoch 22, batch 27300, giga_loss[loss=0.3145, simple_loss=0.3846, pruned_loss=0.1222, over 29032.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3829, pruned_loss=0.127, over 5658037.54 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.363, pruned_loss=0.1153, over 5705657.69 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3842, pruned_loss=0.1276, over 5652337.17 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:32:09,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:32:22,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4237, 1.5927, 1.5596, 1.2904], device='cuda:0'), covar=tensor([0.3260, 0.2768, 0.2287, 0.2862], device='cuda:0'), in_proj_covar=tensor([0.1980, 0.1931, 0.1841, 0.1986], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 12:32:34,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:32:40,329 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:32:43,683 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:32:47,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7427, 2.3868, 1.7128, 1.0888], device='cuda:0'), covar=tensor([0.5134, 0.2907, 0.3693, 0.5184], device='cuda:0'), in_proj_covar=tensor([0.1750, 0.1644, 0.1594, 0.1422], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 12:32:49,521 INFO [train.py:968] (0/2) Epoch 22, batch 27350, giga_loss[loss=0.287, simple_loss=0.3611, pruned_loss=0.1065, over 28694.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3811, pruned_loss=0.1262, over 5664719.21 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3635, pruned_loss=0.1159, over 5706414.84 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3824, pruned_loss=0.1266, over 5657888.53 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:33:09,918 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:33:24,033 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.230e+03 1.872e+03 2.647e+03 4.241e+03 7.408e+03, threshold=5.293e+03, percent-clipped=15.0
+2023-03-11 12:33:30,961 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:33:34,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:33:38,545 INFO [train.py:968] (0/2) Epoch 22, batch 27400, giga_loss[loss=0.3204, simple_loss=0.3898, pruned_loss=0.1255, over 29107.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3792, pruned_loss=0.1256, over 5668008.71 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3635, pruned_loss=0.1158, over 5709697.82 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3805, pruned_loss=0.1261, over 5658892.33 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:33:39,766 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:33:43,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:34:02,080 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:34:12,661 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:34:28,346 INFO [train.py:968] (0/2) Epoch 22, batch 27450, giga_loss[loss=0.2706, simple_loss=0.3371, pruned_loss=0.102, over 28737.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3769, pruned_loss=0.1246, over 5677241.40 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3635, pruned_loss=0.1158, over 5713913.06 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3783, pruned_loss=0.1251, over 5665553.94 frames. ], batch size: 92, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:35:03,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.176e+03 1.711e+03 2.290e+03 3.126e+03 8.090e+03, threshold=4.580e+03, percent-clipped=5.0
+2023-03-11 12:35:08,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=985219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:35:20,191 INFO [train.py:968] (0/2) Epoch 22, batch 27500, giga_loss[loss=0.2826, simple_loss=0.3557, pruned_loss=0.1048, over 29072.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.374, pruned_loss=0.1229, over 5674705.40 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3634, pruned_loss=0.1158, over 5716805.19 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3753, pruned_loss=0.1235, over 5662715.47 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:35:26,353 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 12:36:03,810 INFO [train.py:968] (0/2) Epoch 22, batch 27550, giga_loss[loss=0.4053, simple_loss=0.433, pruned_loss=0.1888, over 26684.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3737, pruned_loss=0.1238, over 5673390.04 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3637, pruned_loss=0.116, over 5719816.98 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3747, pruned_loss=0.1242, over 5660374.61 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:36:33,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.668e+03 2.187e+03 3.160e+03 5.899e+03, threshold=4.373e+03, percent-clipped=7.0
+2023-03-11 12:36:34,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=985317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:36:45,590 INFO [train.py:968] (0/2) Epoch 22, batch 27600, libri_loss[loss=0.2924, simple_loss=0.3653, pruned_loss=0.1097, over 29557.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3723, pruned_loss=0.1229, over 5676713.54 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.364, pruned_loss=0.116, over 5728922.33 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3733, pruned_loss=0.1236, over 5655725.08 frames. ], batch size: 78, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:36:45,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4227, 1.5877, 1.2671, 1.1661], device='cuda:0'), covar=tensor([0.1071, 0.0589, 0.1115, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0449, 0.0520, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 12:37:10,833 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=985360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:37:12,205 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:37:14,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:37:26,733 INFO [train.py:968] (0/2) Epoch 22, batch 27650, giga_loss[loss=0.2775, simple_loss=0.3532, pruned_loss=0.1009, over 29047.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3697, pruned_loss=0.1196, over 5680992.23 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.364, pruned_loss=0.116, over 5730672.63 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3707, pruned_loss=0.1203, over 5660930.25 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:37:39,932 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:37:59,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.502e+02 1.532e+03 1.981e+03 2.883e+03 5.822e+03, threshold=3.961e+03, percent-clipped=6.0
+2023-03-11 12:38:12,445 INFO [train.py:968] (0/2) Epoch 22, batch 27700, giga_loss[loss=0.3714, simple_loss=0.4079, pruned_loss=0.1675, over 26576.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3676, pruned_loss=0.1177, over 5677128.56 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3645, pruned_loss=0.1162, over 5734222.38 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3681, pruned_loss=0.1182, over 5655431.53 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:38:37,328 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:38:38,914 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-11 12:38:40,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:39:00,096 INFO [train.py:968] (0/2) Epoch 22, batch 27750, giga_loss[loss=0.2828, simple_loss=0.3533, pruned_loss=0.1062, over 28918.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3666, pruned_loss=0.1169, over 5669063.69 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3647, pruned_loss=0.1166, over 5729112.98 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3669, pruned_loss=0.117, over 5654838.14 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:39:11,658 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:39:13,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3051, 2.9462, 1.4296, 1.3886], device='cuda:0'), covar=tensor([0.1004, 0.0353, 0.0872, 0.1369], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0559, 0.0389, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 12:39:34,291 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.760e+02 1.520e+03 2.119e+03 2.989e+03 5.054e+03, threshold=4.237e+03, percent-clipped=9.0
+2023-03-11 12:39:39,716 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=985523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:39:41,240 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=985524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:39:46,098 INFO [train.py:968] (0/2) Epoch 22, batch 27800, giga_loss[loss=0.2682, simple_loss=0.3392, pruned_loss=0.09856, over 28757.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.365, pruned_loss=0.1167, over 5657490.01 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.365, pruned_loss=0.1167, over 5724500.86 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3649, pruned_loss=0.1167, over 5647824.82 frames. ], batch size: 243, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:40:37,330 INFO [train.py:968] (0/2) Epoch 22, batch 27850, giga_loss[loss=0.3323, simple_loss=0.372, pruned_loss=0.1463, over 23525.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.363, pruned_loss=0.116, over 5663518.55 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3651, pruned_loss=0.1168, over 5730636.37 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3629, pruned_loss=0.1159, over 5648239.96 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 1.0
+2023-03-11 12:41:11,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.242e+03 1.919e+03 2.601e+03 3.553e+03 1.336e+04, threshold=5.203e+03, percent-clipped=17.0
+2023-03-11 12:41:20,109 INFO [train.py:968] (0/2) Epoch 22, batch 27900, giga_loss[loss=0.3505, simple_loss=0.4048, pruned_loss=0.1481, over 28950.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3657, pruned_loss=0.1177, over 5668677.22 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3657, pruned_loss=0.1173, over 5731875.13 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.365, pruned_loss=0.1172, over 5653084.59 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 1.0
+2023-03-11 12:42:04,232 INFO [train.py:968] (0/2) Epoch 22, batch 27950, giga_loss[loss=0.3414, simple_loss=0.4032, pruned_loss=0.1398, over 28804.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3667, pruned_loss=0.1183, over 5643842.47 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3648, pruned_loss=0.1168, over 5719752.54 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.367, pruned_loss=0.1182, over 5638621.10 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 1.0
+2023-03-11 12:42:09,305 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-11 12:42:39,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.611e+03 2.095e+03 2.667e+03 7.774e+03, threshold=4.189e+03, percent-clipped=1.0
+2023-03-11 12:42:51,269 INFO [train.py:968] (0/2) Epoch 22, batch 28000, giga_loss[loss=0.2872, simple_loss=0.3665, pruned_loss=0.104, over 29003.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3677, pruned_loss=0.1186, over 5651314.67 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3648, pruned_loss=0.1169, over 5717733.25 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3679, pruned_loss=0.1185, over 5648427.59 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:42:56,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=985735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:43:31,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=985775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:43:37,332 INFO [train.py:968] (0/2) Epoch 22, batch 28050, giga_loss[loss=0.3022, simple_loss=0.3684, pruned_loss=0.1179, over 28809.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3684, pruned_loss=0.1194, over 5651281.82 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3655, pruned_loss=0.1171, over 5721272.88 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3681, pruned_loss=0.1192, over 5644275.41 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:43:42,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4788, 4.3509, 4.1402, 1.8650], device='cuda:0'), covar=tensor([0.0600, 0.0690, 0.0726, 0.2110], device='cuda:0'), in_proj_covar=tensor([0.1249, 0.1160, 0.0980, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 12:43:49,903 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-11 12:44:08,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.613e+03 2.052e+03 2.857e+03 5.641e+03, threshold=4.105e+03, percent-clipped=7.0
+2023-03-11 12:44:20,430 INFO [train.py:968] (0/2) Epoch 22, batch 28100, giga_loss[loss=0.3211, simple_loss=0.3885, pruned_loss=0.1269, over 29011.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3686, pruned_loss=0.12, over 5642686.75 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3658, pruned_loss=0.1175, over 5714454.18 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.368, pruned_loss=0.1195, over 5641919.02 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:44:26,281 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=985837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:44:38,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5185, 4.5517, 1.6909, 1.8037], device='cuda:0'), covar=tensor([0.1030, 0.0361, 0.0906, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0558, 0.0388, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 12:45:04,268 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:45:04,316 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7328, 1.8567, 1.9498, 1.5053], device='cuda:0'), covar=tensor([0.1992, 0.2438, 0.1595, 0.1841], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0705, 0.0951, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 12:45:05,964 INFO [train.py:968] (0/2) Epoch 22, batch 28150, giga_loss[loss=0.3334, simple_loss=0.3898, pruned_loss=0.1386, over 27698.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3688, pruned_loss=0.1196, over 5652799.61 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3653, pruned_loss=0.117, over 5717977.97 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3689, pruned_loss=0.1198, over 5646992.38 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:45:06,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:45:20,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=985898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:45:21,426 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=985899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:45:30,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:45:37,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.176e+03 1.736e+03 2.420e+03 3.185e+03 1.169e+04, threshold=4.840e+03, percent-clipped=15.0
+2023-03-11 12:45:52,025 INFO [train.py:968] (0/2) Epoch 22, batch 28200, giga_loss[loss=0.3631, simple_loss=0.4061, pruned_loss=0.1601, over 27568.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3704, pruned_loss=0.121, over 5650476.63 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3652, pruned_loss=0.1169, over 5720699.73 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3708, pruned_loss=0.1213, over 5641273.93 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:45:58,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4704, 1.6471, 1.5973, 1.3617], device='cuda:0'), covar=tensor([0.2684, 0.2627, 0.2062, 0.2464], device='cuda:0'), in_proj_covar=tensor([0.1979, 0.1929, 0.1840, 0.1982], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 12:46:41,397 INFO [train.py:968] (0/2) Epoch 22, batch 28250, giga_loss[loss=0.2612, simple_loss=0.3414, pruned_loss=0.09047, over 29016.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.372, pruned_loss=0.1223, over 5651523.57 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.365, pruned_loss=0.1168, over 5721983.12 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3726, pruned_loss=0.1227, over 5642523.50 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:46:57,438 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-986000.pt
+2023-03-11 12:47:15,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.760e+03 2.308e+03 3.362e+03 1.268e+04, threshold=4.617e+03, percent-clipped=12.0
+2023-03-11 12:47:27,386 INFO [train.py:968] (0/2) Epoch 22, batch 28300, giga_loss[loss=0.3025, simple_loss=0.3698, pruned_loss=0.1176, over 28930.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3724, pruned_loss=0.123, over 5646935.81 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3647, pruned_loss=0.1166, over 5715261.13 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3734, pruned_loss=0.1238, over 5642605.94 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:47:39,359 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:47:39,910 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:47:41,531 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:47:42,095 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:47:44,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3380, 1.6099, 1.3164, 0.9507], device='cuda:0'), covar=tensor([0.2568, 0.2567, 0.2890, 0.2252], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1098, 0.1339, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 12:47:55,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4337, 2.6300, 1.5762, 1.5856], device='cuda:0'), covar=tensor([0.0813, 0.0330, 0.0717, 0.1133], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0559, 0.0388, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 12:48:10,258 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:48:10,914 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:48:18,283 INFO [train.py:968] (0/2) Epoch 22, batch 28350, giga_loss[loss=0.2959, simple_loss=0.3703, pruned_loss=0.1108, over 28828.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3723, pruned_loss=0.1213, over 5643287.74 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3648, pruned_loss=0.1166, over 5709762.42 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3733, pruned_loss=0.122, over 5643141.16 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:48:37,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3194, 3.3464, 1.4876, 1.5315], device='cuda:0'), covar=tensor([0.0983, 0.0362, 0.0884, 0.1286], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0558, 0.0388, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 12:48:51,970 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:48:55,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.906e+03 2.671e+03 3.923e+03 1.406e+04, threshold=5.343e+03, percent-clipped=17.0
+2023-03-11 12:49:08,582 INFO [train.py:968] (0/2) Epoch 22, batch 28400, giga_loss[loss=0.2872, simple_loss=0.3582, pruned_loss=0.1081, over 29060.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.373, pruned_loss=0.1222, over 5635503.11 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3652, pruned_loss=0.1169, over 5712110.07 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3735, pruned_loss=0.1226, over 5632621.71 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:49:28,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=986150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:50:00,027 INFO [train.py:968] (0/2) Epoch 22, batch 28450, giga_loss[loss=0.4089, simple_loss=0.4405, pruned_loss=0.1887, over 28713.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3731, pruned_loss=0.123, over 5636213.36 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3646, pruned_loss=0.1166, over 5715131.12 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3741, pruned_loss=0.1237, over 5630290.65 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:50:36,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=986212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:50:41,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.224e+03 1.985e+03 2.472e+03 3.635e+03 8.771e+03, threshold=4.944e+03, percent-clipped=11.0
+2023-03-11 12:50:58,452 INFO [train.py:968] (0/2) Epoch 22, batch 28500, giga_loss[loss=0.3027, simple_loss=0.3681, pruned_loss=0.1186, over 28571.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3713, pruned_loss=0.123, over 5636366.19 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3639, pruned_loss=0.1163, over 5719657.22 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3729, pruned_loss=0.1239, over 5625743.55 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:51:11,224 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2759, 3.1680, 1.5137, 1.4176], device='cuda:0'), covar=tensor([0.0987, 0.0340, 0.0865, 0.1361], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0558, 0.0388, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 12:51:23,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:51:48,838 INFO [train.py:968] (0/2) Epoch 22, batch 28550, giga_loss[loss=0.2836, simple_loss=0.3447, pruned_loss=0.1112, over 28479.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3696, pruned_loss=0.1223, over 5639637.93 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3637, pruned_loss=0.1161, over 5724435.37 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3712, pruned_loss=0.1234, over 5624936.24 frames. ], batch size: 78, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:51:59,787 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:52:03,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:52:23,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.588e+03 2.179e+03 2.605e+03 5.989e+03, threshold=4.359e+03, percent-clipped=1.0
+2023-03-11 12:52:29,065 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:52:34,259 INFO [train.py:968] (0/2) Epoch 22, batch 28600, libri_loss[loss=0.3314, simple_loss=0.3883, pruned_loss=0.1372, over 29751.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3685, pruned_loss=0.1213, over 5658926.22 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3637, pruned_loss=0.1161, over 5727332.70 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3699, pruned_loss=0.1224, over 5643051.50 frames. ], batch size: 87, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:52:58,095 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:53:01,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:53:04,216 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-11 12:53:22,543 INFO [train.py:968] (0/2) Epoch 22, batch 28650, giga_loss[loss=0.3023, simple_loss=0.368, pruned_loss=0.1183, over 28846.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3684, pruned_loss=0.1217, over 5650722.51 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3637, pruned_loss=0.116, over 5727112.26 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3697, pruned_loss=0.1227, over 5636741.98 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:53:27,273 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:53:31,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1986, 0.8038, 0.8680, 1.3578], device='cuda:0'), covar=tensor([0.0727, 0.0434, 0.0366, 0.0783], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 12:53:43,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4918, 1.8013, 1.5114, 1.5644], device='cuda:0'), covar=tensor([0.1972, 0.1990, 0.2190, 0.1713], device='cuda:0'), in_proj_covar=tensor([0.1517, 0.1096, 0.1338, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 12:53:54,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.204e+03 1.700e+03 2.225e+03 2.951e+03 1.190e+04, threshold=4.451e+03, percent-clipped=11.0
+2023-03-11 12:54:06,954 INFO [train.py:968] (0/2) Epoch 22, batch 28700, giga_loss[loss=0.3171, simple_loss=0.3799, pruned_loss=0.1271, over 28804.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.369, pruned_loss=0.1217, over 5646260.00 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.364, pruned_loss=0.1161, over 5709290.57 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3698, pruned_loss=0.1226, over 5647819.28 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:54:21,699 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:54:55,180 INFO [train.py:968] (0/2) Epoch 22, batch 28750, giga_loss[loss=0.3189, simple_loss=0.3857, pruned_loss=0.1261, over 28916.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3696, pruned_loss=0.1221, over 5652653.45 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3643, pruned_loss=0.1163, over 5709373.07 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3701, pruned_loss=0.1227, over 5652952.37 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:55:02,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=986490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:55:04,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3659, 2.8795, 1.5065, 1.4827], device='cuda:0'), covar=tensor([0.0900, 0.0371, 0.0852, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0559, 0.0388, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 12:55:06,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 12:55:08,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4778, 1.5584, 1.7148, 1.2794], device='cuda:0'), covar=tensor([0.1636, 0.2481, 0.1373, 0.1665], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0706, 0.0949, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 12:55:30,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.139e+03 1.821e+03 2.534e+03 3.852e+03 1.014e+04, threshold=5.069e+03, percent-clipped=18.0
+2023-03-11 12:55:42,217 INFO [train.py:968] (0/2) Epoch 22, batch 28800, giga_loss[loss=0.3174, simple_loss=0.3779, pruned_loss=0.1284, over 28740.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3708, pruned_loss=0.1232, over 5652778.55 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3636, pruned_loss=0.1158, over 5706255.06 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3721, pruned_loss=0.1243, over 5653896.24 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 12:56:30,968 INFO [train.py:968] (0/2) Epoch 22, batch 28850, giga_loss[loss=0.3648, simple_loss=0.4128, pruned_loss=0.1584, over 28575.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3709, pruned_loss=0.1233, over 5654005.51 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3638, pruned_loss=0.1159, over 5697144.97 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3717, pruned_loss=0.1241, over 5663007.00 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:57:06,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 1.794e+03 2.090e+03 2.784e+03 5.477e+03, threshold=4.180e+03, percent-clipped=1.0
+2023-03-11 12:57:14,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=986627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:57:17,140 INFO [train.py:968] (0/2) Epoch 22, batch 28900, giga_loss[loss=0.2903, simple_loss=0.3565, pruned_loss=0.1121, over 28641.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3701, pruned_loss=0.1226, over 5659994.61 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1157, over 5700393.48 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.371, pruned_loss=0.1236, over 5663791.15 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:57:19,936 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:57:22,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:57:50,100 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:58:06,810 INFO [train.py:968] (0/2) Epoch 22, batch 28950, giga_loss[loss=0.392, simple_loss=0.4087, pruned_loss=0.1876, over 23511.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.1229, over 5648381.88 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3638, pruned_loss=0.1158, over 5685700.93 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3719, pruned_loss=0.1237, over 5662183.62 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:58:43,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.654e+03 2.220e+03 3.286e+03 1.150e+04, threshold=4.441e+03, percent-clipped=12.0
+2023-03-11 12:58:54,369 INFO [train.py:968] (0/2) Epoch 22, batch 29000, giga_loss[loss=0.3618, simple_loss=0.4047, pruned_loss=0.1594, over 27897.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3715, pruned_loss=0.1231, over 5658825.39 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3641, pruned_loss=0.1161, over 5689069.79 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.372, pruned_loss=0.1236, over 5666202.73 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:59:29,423 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:59:33,030 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:59:40,545 INFO [train.py:968] (0/2) Epoch 22, batch 29050, giga_loss[loss=0.3032, simple_loss=0.3744, pruned_loss=0.116, over 28750.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3734, pruned_loss=0.1248, over 5665054.81 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3638, pruned_loss=0.116, over 5692059.93 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3741, pruned_loss=0.1254, over 5667808.23 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:59:55,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:00:13,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.477e+02 1.711e+03 2.267e+03 3.092e+03 1.024e+04, threshold=4.534e+03, percent-clipped=7.0
+2023-03-11 13:00:13,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=986821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:00:14,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6843, 1.8339, 1.8215, 1.5655], device='cuda:0'), covar=tensor([0.1936, 0.2267, 0.2349, 0.2348], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0759, 0.0724, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 13:00:22,160 INFO [train.py:968] (0/2) Epoch 22, batch 29100, giga_loss[loss=0.2939, simple_loss=0.3624, pruned_loss=0.1127, over 28836.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3755, pruned_loss=0.1267, over 5661843.42 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.364, pruned_loss=0.1162, over 5695126.42 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3762, pruned_loss=0.1274, over 5660910.12 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:00:59,612 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-11 13:01:04,147 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:01:09,338 INFO [train.py:968] (0/2) Epoch 22, batch 29150, giga_loss[loss=0.3125, simple_loss=0.3781, pruned_loss=0.1234, over 28747.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3748, pruned_loss=0.1263, over 5659690.84 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3639, pruned_loss=0.1161, over 5697230.02 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3756, pruned_loss=0.1269, over 5656759.40 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:01:09,667 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986881.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 13:01:17,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6059, 1.7737, 1.5995, 1.6136], device='cuda:0'), covar=tensor([0.0763, 0.0310, 0.0317, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 13:01:37,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-11 13:01:41,561 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:01:43,236 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-11 13:01:47,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.952e+02 1.712e+03 2.393e+03 3.485e+03 1.401e+04, threshold=4.786e+03, percent-clipped=11.0
+2023-03-11 13:01:58,496 INFO [train.py:968] (0/2) Epoch 22, batch 29200, giga_loss[loss=0.2927, simple_loss=0.3696, pruned_loss=0.1079, over 28830.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3753, pruned_loss=0.1255, over 5645334.96 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3636, pruned_loss=0.1158, over 5692957.73 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3764, pruned_loss=0.1265, over 5646435.55 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:02:33,370 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:02:36,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:02:50,328 INFO [train.py:968] (0/2) Epoch 22, batch 29250, giga_loss[loss=0.2858, simple_loss=0.349, pruned_loss=0.1113, over 28619.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3764, pruned_loss=0.1261, over 5640726.70 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3635, pruned_loss=0.1158, over 5693230.95 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3775, pruned_loss=0.127, over 5640971.67 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:02:51,291 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2538, 0.7284, 0.8363, 1.3649], device='cuda:0'), covar=tensor([0.0793, 0.0395, 0.0370, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 13:02:52,832 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:03:03,819 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:03:10,000 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=987002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:03:23,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.388e+02 1.475e+03 1.835e+03 2.798e+03 7.843e+03, threshold=3.670e+03, percent-clipped=3.0
+2023-03-11 13:03:28,567 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=987026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:03:34,242 INFO [train.py:968] (0/2) Epoch 22, batch 29300, giga_loss[loss=0.3162, simple_loss=0.3757, pruned_loss=0.1283, over 28193.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3745, pruned_loss=0.1244, over 5653664.68 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3635, pruned_loss=0.1159, over 5699192.21 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3758, pruned_loss=0.1254, over 5646992.23 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:03:50,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8891, 1.1112, 2.8615, 2.8142], device='cuda:0'), covar=tensor([0.1649, 0.2525, 0.0608, 0.0882], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0654, 0.0977, 0.0920], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 13:04:20,009 INFO [train.py:968] (0/2) Epoch 22, batch 29350, giga_loss[loss=0.2896, simple_loss=0.3618, pruned_loss=0.1086, over 28743.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3729, pruned_loss=0.1236, over 5655890.71 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3635, pruned_loss=0.1159, over 5700345.90 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.374, pruned_loss=0.1244, over 5649047.63 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:04:25,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3421, 1.2187, 3.7757, 3.2484], device='cuda:0'), covar=tensor([0.1698, 0.2954, 0.0472, 0.1151], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0655, 0.0978, 0.0921], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 13:04:26,463 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1567, 1.2301, 1.1363, 0.8465], device='cuda:0'), covar=tensor([0.0974, 0.0520, 0.1023, 0.1086], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0448, 0.0519, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 13:04:53,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.683e+03 2.303e+03 2.936e+03 8.162e+03, threshold=4.607e+03, percent-clipped=14.0
+2023-03-11 13:05:02,173 INFO [train.py:968] (0/2) Epoch 22, batch 29400, giga_loss[loss=0.367, simple_loss=0.409, pruned_loss=0.1625, over 27532.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3725, pruned_loss=0.1229, over 5655566.11 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3634, pruned_loss=0.1157, over 5702576.89 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3738, pruned_loss=0.1241, over 5646888.18 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:05:54,933 INFO [train.py:968] (0/2) Epoch 22, batch 29450, giga_loss[loss=0.2684, simple_loss=0.3424, pruned_loss=0.09715, over 28957.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3728, pruned_loss=0.123, over 5665396.07 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.363, pruned_loss=0.1154, over 5706179.63 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3745, pruned_loss=0.1244, over 5653824.15 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:06:22,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7687, 2.5146, 1.5873, 0.9400], device='cuda:0'), covar=tensor([0.6961, 0.3405, 0.3858, 0.6504], device='cuda:0'), in_proj_covar=tensor([0.1771, 0.1666, 0.1608, 0.1441], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 13:06:31,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.734e+03 2.231e+03 3.092e+03 9.950e+03, threshold=4.462e+03, percent-clipped=7.0
+2023-03-11 13:06:42,891 INFO [train.py:968] (0/2) Epoch 22, batch 29500, libri_loss[loss=0.3701, simple_loss=0.4184, pruned_loss=0.1609, over 19529.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3724, pruned_loss=0.1237, over 5652013.24 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3628, pruned_loss=0.1154, over 5699689.04 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3741, pruned_loss=0.125, over 5648499.73 frames. ], batch size: 187, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:06:58,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:07:03,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987256.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 13:07:24,972 INFO [train.py:968] (0/2) Epoch 22, batch 29550, giga_loss[loss=0.3046, simple_loss=0.3767, pruned_loss=0.1162, over 28491.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1235, over 5674115.78 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3628, pruned_loss=0.1154, over 5708000.34 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.374, pruned_loss=0.1248, over 5662410.77 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:07:31,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:07:55,048 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-11 13:08:00,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.740e+03 2.458e+03 3.138e+03 8.401e+03, threshold=4.915e+03, percent-clipped=9.0
+2023-03-11 13:08:11,615 INFO [train.py:968] (0/2) Epoch 22, batch 29600, giga_loss[loss=0.2861, simple_loss=0.3589, pruned_loss=0.1067, over 28753.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3752, pruned_loss=0.1261, over 5667017.89 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3633, pruned_loss=0.1157, over 5710202.00 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3763, pruned_loss=0.127, over 5655229.16 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 13:08:37,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:08:38,428 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7990, 1.1569, 2.8712, 2.6883], device='cuda:0'), covar=tensor([0.1786, 0.2605, 0.0621, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0657, 0.0981, 0.0925], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 13:08:53,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:08:57,867 INFO [train.py:968] (0/2) Epoch 22, batch 29650, giga_loss[loss=0.3633, simple_loss=0.4199, pruned_loss=0.1534, over 28578.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3757, pruned_loss=0.1265, over 5657310.33 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3633, pruned_loss=0.1157, over 5711602.97 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3768, pruned_loss=0.1274, over 5645699.15 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:09:10,663 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:09:13,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:09:16,138 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987399.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:09:18,332 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:09:19,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987402.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:09:36,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.599e+03 1.911e+03 2.650e+03 6.125e+03, threshold=3.821e+03, percent-clipped=3.0
+2023-03-11 13:09:40,387 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:09:45,226 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:09:45,624 INFO [train.py:968] (0/2) Epoch 22, batch 29700, giga_loss[loss=0.2811, simple_loss=0.3526, pruned_loss=0.1048, over 28591.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3751, pruned_loss=0.126, over 5657905.40 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3626, pruned_loss=0.1153, over 5713439.05 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3769, pruned_loss=0.1274, over 5646076.91 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:09:45,978 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987431.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:09:47,323 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:10:16,620 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:10:30,376 INFO [train.py:968] (0/2) Epoch 22, batch 29750, giga_loss[loss=0.2833, simple_loss=0.359, pruned_loss=0.1038, over 28120.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3749, pruned_loss=0.1251, over 5661176.06 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.363, pruned_loss=0.1155, over 5715134.75 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3764, pruned_loss=0.1263, over 5648494.89 frames. ], batch size: 77, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:10:37,109 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1659, 1.4049, 1.4307, 1.0560], device='cuda:0'), covar=tensor([0.1664, 0.2539, 0.1395, 0.1673], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0704, 0.0948, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 13:10:48,819 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:10:50,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2894, 1.1911, 3.5334, 3.0480], device='cuda:0'), covar=tensor([0.1657, 0.2932, 0.0503, 0.1446], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0655, 0.0977, 0.0922], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 13:10:52,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:07,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5165, 1.7862, 1.4539, 1.3606], device='cuda:0'), covar=tensor([0.2498, 0.2546, 0.2896, 0.2213], device='cuda:0'), in_proj_covar=tensor([0.1520, 0.1100, 0.1340, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 13:11:09,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:09,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.809e+02 1.642e+03 2.242e+03 3.002e+03 8.763e+03, threshold=4.483e+03, percent-clipped=14.0
+2023-03-11 13:11:11,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:17,078 INFO [train.py:968] (0/2) Epoch 22, batch 29800, giga_loss[loss=0.2843, simple_loss=0.3585, pruned_loss=0.1051, over 28740.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3751, pruned_loss=0.1245, over 5655744.05 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3631, pruned_loss=0.1155, over 5709299.77 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3764, pruned_loss=0.1257, over 5648746.11 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:11:19,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:31,973 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:35,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:39,084 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:12:02,684 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:12:07,171 INFO [train.py:968] (0/2) Epoch 22, batch 29850, giga_loss[loss=0.3172, simple_loss=0.3896, pruned_loss=0.1224, over 28765.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3744, pruned_loss=0.124, over 5657606.35 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.363, pruned_loss=0.1155, over 5708617.27 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3756, pruned_loss=0.125, over 5652476.07 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:12:42,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.964e+02 1.828e+03 2.417e+03 2.860e+03 1.181e+04, threshold=4.833e+03, percent-clipped=5.0
+2023-03-11 13:12:51,041 INFO [train.py:968] (0/2) Epoch 22, batch 29900, libri_loss[loss=0.2518, simple_loss=0.3209, pruned_loss=0.09136, over 29481.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.372, pruned_loss=0.1226, over 5671963.05 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3626, pruned_loss=0.1153, over 5716696.17 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3737, pruned_loss=0.1239, over 5658591.31 frames. ], batch size: 70, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:13:35,087 INFO [train.py:968] (0/2) Epoch 22, batch 29950, giga_loss[loss=0.2749, simple_loss=0.335, pruned_loss=0.1074, over 28573.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3684, pruned_loss=0.1205, over 5675222.10 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3627, pruned_loss=0.1152, over 5718930.31 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3698, pruned_loss=0.1217, over 5661638.41 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:14:13,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.644e+03 2.148e+03 2.844e+03 6.383e+03, threshold=4.296e+03, percent-clipped=5.0
+2023-03-11 13:14:20,330 INFO [train.py:968] (0/2) Epoch 22, batch 30000, giga_loss[loss=0.2333, simple_loss=0.3017, pruned_loss=0.08241, over 28690.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3645, pruned_loss=0.1189, over 5668201.99 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3623, pruned_loss=0.1151, over 5723858.97 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3662, pruned_loss=0.1201, over 5651138.40 frames. ], batch size: 92, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:14:20,333 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 13:14:26,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9543, 1.1369, 3.1173, 2.8420], device='cuda:0'), covar=tensor([0.1905, 0.3009, 0.0573, 0.1010], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0659, 0.0981, 0.0925], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 13:14:29,028 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2048, simple_loss=0.3132, pruned_loss=0.0482, over 944034.00 frames. 
+2023-03-11 13:14:29,029 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 13:14:29,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5961, 1.6460, 1.8098, 1.3869], device='cuda:0'), covar=tensor([0.1749, 0.2648, 0.1459, 0.1797], device='cuda:0'), in_proj_covar=tensor([0.0901, 0.0706, 0.0949, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 13:14:55,443 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-11 13:14:58,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2547, 1.8758, 1.5165, 1.4626], device='cuda:0'), covar=tensor([0.0805, 0.0301, 0.0305, 0.1039], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 13:15:13,777 INFO [train.py:968] (0/2) Epoch 22, batch 30050, giga_loss[loss=0.2714, simple_loss=0.3429, pruned_loss=0.09997, over 28400.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3632, pruned_loss=0.1187, over 5675719.06 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3619, pruned_loss=0.1148, over 5726563.69 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3648, pruned_loss=0.12, over 5659236.74 frames. ], batch size: 65, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:15:43,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5288, 1.6391, 1.7378, 1.3369], device='cuda:0'), covar=tensor([0.1746, 0.2521, 0.1436, 0.1680], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0705, 0.0948, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 13:15:52,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.263e+03 1.827e+03 2.373e+03 3.287e+03 1.067e+04, threshold=4.746e+03, percent-clipped=12.0
+2023-03-11 13:15:59,685 INFO [train.py:968] (0/2) Epoch 22, batch 30100, giga_loss[loss=0.2718, simple_loss=0.3426, pruned_loss=0.1005, over 28892.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3621, pruned_loss=0.1187, over 5657528.69 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3623, pruned_loss=0.115, over 5731438.51 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3631, pruned_loss=0.1197, over 5638343.95 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:16:26,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6174, 1.7400, 1.2820, 1.3510], device='cuda:0'), covar=tensor([0.0908, 0.0538, 0.0922, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0448, 0.0517, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 13:16:45,422 INFO [train.py:968] (0/2) Epoch 22, batch 30150, giga_loss[loss=0.2756, simple_loss=0.3638, pruned_loss=0.0937, over 28596.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3615, pruned_loss=0.1167, over 5656971.36 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3619, pruned_loss=0.1149, over 5724119.24 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3627, pruned_loss=0.1176, over 5646871.47 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:17:25,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.485e+02 1.514e+03 2.011e+03 2.617e+03 5.029e+03, threshold=4.023e+03, percent-clipped=3.0
+2023-03-11 13:17:34,122 INFO [train.py:968] (0/2) Epoch 22, batch 30200, giga_loss[loss=0.2614, simple_loss=0.3476, pruned_loss=0.0876, over 28921.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3593, pruned_loss=0.1135, over 5640056.05 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3619, pruned_loss=0.115, over 5719481.95 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3602, pruned_loss=0.1141, over 5633705.13 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:18:22,822 INFO [train.py:968] (0/2) Epoch 22, batch 30250, giga_loss[loss=0.2592, simple_loss=0.3424, pruned_loss=0.08795, over 28322.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3584, pruned_loss=0.1115, over 5652975.38 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.362, pruned_loss=0.1153, over 5721359.04 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.359, pruned_loss=0.1116, over 5643369.38 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:18:39,959 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-988000.pt
+2023-03-11 13:19:00,821 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-11 13:19:02,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.694e+03 2.196e+03 2.972e+03 8.217e+03, threshold=4.391e+03, percent-clipped=7.0
+2023-03-11 13:19:13,379 INFO [train.py:968] (0/2) Epoch 22, batch 30300, giga_loss[loss=0.2542, simple_loss=0.3359, pruned_loss=0.08629, over 28875.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3556, pruned_loss=0.1086, over 5653997.42 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3612, pruned_loss=0.1152, over 5724894.35 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3566, pruned_loss=0.1087, over 5641793.44 frames. ], batch size: 186, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:20:01,097 INFO [train.py:968] (0/2) Epoch 22, batch 30350, giga_loss[loss=0.2875, simple_loss=0.3649, pruned_loss=0.105, over 28821.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3529, pruned_loss=0.1058, over 5660233.87 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3609, pruned_loss=0.1151, over 5728442.52 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3538, pruned_loss=0.1058, over 5646026.40 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:20:25,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1140, 3.9361, 3.7390, 1.8326], device='cuda:0'), covar=tensor([0.0663, 0.0817, 0.0887, 0.2189], device='cuda:0'), in_proj_covar=tensor([0.1245, 0.1153, 0.0975, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 13:20:39,239 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.690e+02 1.396e+03 1.734e+03 2.281e+03 4.036e+03, threshold=3.468e+03, percent-clipped=0.0
+2023-03-11 13:20:49,430 INFO [train.py:968] (0/2) Epoch 22, batch 30400, giga_loss[loss=0.2711, simple_loss=0.3563, pruned_loss=0.09299, over 28905.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3519, pruned_loss=0.1029, over 5677651.16 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3604, pruned_loss=0.115, over 5733108.71 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3529, pruned_loss=0.1027, over 5660692.71 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 13:21:03,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7826, 3.6228, 3.4139, 2.0308], device='cuda:0'), covar=tensor([0.0700, 0.0883, 0.0923, 0.2402], device='cuda:0'), in_proj_covar=tensor([0.1246, 0.1153, 0.0976, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 13:21:41,541 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=988179.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:21:43,238 INFO [train.py:968] (0/2) Epoch 22, batch 30450, giga_loss[loss=0.3056, simple_loss=0.3753, pruned_loss=0.118, over 28703.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3517, pruned_loss=0.1021, over 5677410.40 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3603, pruned_loss=0.1151, over 5735316.18 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3525, pruned_loss=0.1017, over 5661431.86 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:22:29,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.851e+02 1.493e+03 1.978e+03 3.066e+03 1.342e+04, threshold=3.957e+03, percent-clipped=20.0
+2023-03-11 13:22:35,511 INFO [train.py:968] (0/2) Epoch 22, batch 30500, giga_loss[loss=0.2772, simple_loss=0.3486, pruned_loss=0.1029, over 28243.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.35, pruned_loss=0.1005, over 5670285.89 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3605, pruned_loss=0.1153, over 5733335.69 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3503, pruned_loss=0.09991, over 5658954.68 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:22:55,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3535, 1.5465, 1.3953, 1.5336], device='cuda:0'), covar=tensor([0.0743, 0.0392, 0.0347, 0.0828], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 13:23:29,406 INFO [train.py:968] (0/2) Epoch 22, batch 30550, giga_loss[loss=0.3049, simple_loss=0.3607, pruned_loss=0.1246, over 26722.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3475, pruned_loss=0.09859, over 5666756.03 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3605, pruned_loss=0.1153, over 5730459.01 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3478, pruned_loss=0.09808, over 5660110.61 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:24:09,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.517e+02 1.499e+03 1.973e+03 2.732e+03 9.201e+03, threshold=3.945e+03, percent-clipped=8.0
+2023-03-11 13:24:15,826 INFO [train.py:968] (0/2) Epoch 22, batch 30600, libri_loss[loss=0.2456, simple_loss=0.3052, pruned_loss=0.09307, over 29481.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3453, pruned_loss=0.09776, over 5670871.61 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3592, pruned_loss=0.1146, over 5738373.71 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.346, pruned_loss=0.09704, over 5654310.86 frames. ], batch size: 70, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:24:37,526 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-11 13:24:43,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4667, 3.5824, 1.6272, 1.5785], device='cuda:0'), covar=tensor([0.0988, 0.0330, 0.0948, 0.1355], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0556, 0.0387, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 13:25:00,039 INFO [train.py:968] (0/2) Epoch 22, batch 30650, giga_loss[loss=0.277, simple_loss=0.3379, pruned_loss=0.1081, over 26739.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.346, pruned_loss=0.09825, over 5664315.43 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3591, pruned_loss=0.1148, over 5728422.52 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3461, pruned_loss=0.09686, over 5656620.36 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:25:27,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5073, 1.7671, 1.4645, 1.4391], device='cuda:0'), covar=tensor([0.2747, 0.2513, 0.2742, 0.2414], device='cuda:0'), in_proj_covar=tensor([0.1519, 0.1098, 0.1342, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 13:25:33,641 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=988419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:25:37,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.587e+02 1.369e+03 1.801e+03 2.492e+03 5.078e+03, threshold=3.601e+03, percent-clipped=2.0
+2023-03-11 13:25:46,132 INFO [train.py:968] (0/2) Epoch 22, batch 30700, giga_loss[loss=0.2617, simple_loss=0.3388, pruned_loss=0.09232, over 28299.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3442, pruned_loss=0.09701, over 5650779.65 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3582, pruned_loss=0.1145, over 5715440.34 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3447, pruned_loss=0.09561, over 5653186.85 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:26:37,216 INFO [train.py:968] (0/2) Epoch 22, batch 30750, giga_loss[loss=0.285, simple_loss=0.3441, pruned_loss=0.1129, over 27540.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3412, pruned_loss=0.09444, over 5656737.01 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3581, pruned_loss=0.1144, over 5718266.78 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3415, pruned_loss=0.09312, over 5655378.49 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:27:20,413 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.958e+02 1.394e+03 1.881e+03 2.612e+03 5.427e+03, threshold=3.762e+03, percent-clipped=6.0
+2023-03-11 13:27:25,652 INFO [train.py:968] (0/2) Epoch 22, batch 30800, libri_loss[loss=0.2758, simple_loss=0.3372, pruned_loss=0.1072, over 29741.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3392, pruned_loss=0.09328, over 5673618.54 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3577, pruned_loss=0.1143, over 5725260.80 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3392, pruned_loss=0.09161, over 5664102.09 frames. ], batch size: 87, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:27:46,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=988554.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:27:51,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0970, 5.4748, 2.2380, 2.5505], device='cuda:0'), covar=tensor([0.0939, 0.0274, 0.0859, 0.1113], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0557, 0.0388, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 13:28:08,113 INFO [train.py:968] (0/2) Epoch 22, batch 30850, giga_loss[loss=0.2435, simple_loss=0.3253, pruned_loss=0.08085, over 28802.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.337, pruned_loss=0.09287, over 5678512.76 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3571, pruned_loss=0.1142, over 5729666.80 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3363, pruned_loss=0.09028, over 5663544.27 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:28:47,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.234e+02 1.528e+03 1.983e+03 2.667e+03 9.847e+03, threshold=3.966e+03, percent-clipped=7.0
+2023-03-11 13:28:55,084 INFO [train.py:968] (0/2) Epoch 22, batch 30900, giga_loss[loss=0.281, simple_loss=0.3568, pruned_loss=0.1026, over 28171.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3365, pruned_loss=0.09322, over 5677225.35 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3567, pruned_loss=0.1142, over 5733976.26 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3358, pruned_loss=0.09071, over 5660360.99 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:29:33,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2906, 1.2463, 3.4886, 3.0620], device='cuda:0'), covar=tensor([0.1634, 0.2926, 0.0520, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0767, 0.0653, 0.0970, 0.0913], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 13:29:47,616 INFO [train.py:968] (0/2) Epoch 22, batch 30950, giga_loss[loss=0.3088, simple_loss=0.3731, pruned_loss=0.1222, over 28881.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3385, pruned_loss=0.09486, over 5654524.89 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3566, pruned_loss=0.1141, over 5725122.88 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3377, pruned_loss=0.09247, over 5646526.72 frames. ], batch size: 186, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:30:04,067 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=988697.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:30:07,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=988700.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:30:35,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.382e+02 1.491e+03 2.017e+03 2.815e+03 8.273e+03, threshold=4.034e+03, percent-clipped=7.0
+2023-03-11 13:30:44,095 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=988729.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:30:45,404 INFO [train.py:968] (0/2) Epoch 22, batch 31000, giga_loss[loss=0.2743, simple_loss=0.353, pruned_loss=0.09783, over 27934.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3407, pruned_loss=0.09527, over 5648463.27 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3559, pruned_loss=0.1137, over 5720555.00 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3402, pruned_loss=0.09309, over 5643491.74 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:31:40,526 INFO [train.py:968] (0/2) Epoch 22, batch 31050, giga_loss[loss=0.2699, simple_loss=0.3509, pruned_loss=0.09445, over 28773.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3407, pruned_loss=0.09415, over 5633768.35 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3556, pruned_loss=0.1137, over 5713164.21 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3402, pruned_loss=0.09222, over 5635835.82 frames. ], batch size: 243, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:31:59,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=988794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:32:40,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.390e+02 1.575e+03 1.900e+03 2.932e+03 1.165e+04, threshold=3.801e+03, percent-clipped=9.0
+2023-03-11 13:32:49,879 INFO [train.py:968] (0/2) Epoch 22, batch 31100, giga_loss[loss=0.2706, simple_loss=0.3494, pruned_loss=0.0959, over 28824.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.09477, over 5632104.57 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3553, pruned_loss=0.1136, over 5714198.30 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3409, pruned_loss=0.09315, over 5631970.78 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:32:55,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6018, 1.6494, 1.8688, 1.4210], device='cuda:0'), covar=tensor([0.1986, 0.2638, 0.1567, 0.1906], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0701, 0.0948, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 13:33:37,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3825, 1.4630, 1.3185, 1.4980], device='cuda:0'), covar=tensor([0.0746, 0.0345, 0.0347, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 13:33:47,518 INFO [train.py:968] (0/2) Epoch 22, batch 31150, giga_loss[loss=0.2502, simple_loss=0.3332, pruned_loss=0.08358, over 28653.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3389, pruned_loss=0.09321, over 5644183.54 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3551, pruned_loss=0.1136, over 5719243.23 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3385, pruned_loss=0.09129, over 5637106.87 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:34:46,818 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.627e+02 1.382e+03 1.777e+03 2.368e+03 7.372e+03, threshold=3.554e+03, percent-clipped=5.0
+2023-03-11 13:34:54,457 INFO [train.py:968] (0/2) Epoch 22, batch 31200, giga_loss[loss=0.2808, simple_loss=0.3571, pruned_loss=0.1023, over 27593.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3374, pruned_loss=0.09081, over 5638896.30 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3551, pruned_loss=0.1137, over 5720906.51 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3368, pruned_loss=0.08893, over 5630776.11 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:34:54,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5013, 1.9399, 1.7206, 1.6603], device='cuda:0'), covar=tensor([0.2036, 0.2377, 0.2091, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0742, 0.0710, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 13:34:59,807 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=988937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:35:02,261 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=988940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:35:38,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=988969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:35:54,896 INFO [train.py:968] (0/2) Epoch 22, batch 31250, giga_loss[loss=0.2705, simple_loss=0.338, pruned_loss=0.1015, over 28981.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3358, pruned_loss=0.09002, over 5640392.12 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3548, pruned_loss=0.1135, over 5714781.77 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3353, pruned_loss=0.08828, over 5637394.82 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:36:47,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.377e+02 1.434e+03 1.870e+03 2.407e+03 7.227e+03, threshold=3.740e+03, percent-clipped=8.0
+2023-03-11 13:36:56,425 INFO [train.py:968] (0/2) Epoch 22, batch 31300, giga_loss[loss=0.2734, simple_loss=0.3456, pruned_loss=0.1006, over 27698.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3344, pruned_loss=0.0901, over 5656436.70 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3544, pruned_loss=0.1133, over 5716132.13 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3339, pruned_loss=0.08834, over 5651285.28 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:37:09,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4163, 1.5294, 1.3368, 1.5635], device='cuda:0'), covar=tensor([0.0780, 0.0324, 0.0352, 0.0893], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 13:37:15,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 13:37:52,559 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-11 13:37:56,323 INFO [train.py:968] (0/2) Epoch 22, batch 31350, giga_loss[loss=0.2398, simple_loss=0.329, pruned_loss=0.07531, over 28509.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3339, pruned_loss=0.09011, over 5655686.46 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3541, pruned_loss=0.1133, over 5709234.76 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3336, pruned_loss=0.08847, over 5658056.60 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:38:45,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.404e+02 1.407e+03 1.897e+03 2.922e+03 9.016e+03, threshold=3.794e+03, percent-clipped=13.0
+2023-03-11 13:38:51,271 INFO [train.py:968] (0/2) Epoch 22, batch 31400, giga_loss[loss=0.2349, simple_loss=0.3175, pruned_loss=0.07613, over 28331.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3343, pruned_loss=0.08982, over 5662889.99 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3532, pruned_loss=0.1129, over 5715204.72 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3343, pruned_loss=0.08822, over 5657979.83 frames. ], batch size: 65, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:39:27,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 13:39:56,834 INFO [train.py:968] (0/2) Epoch 22, batch 31450, giga_loss[loss=0.232, simple_loss=0.3241, pruned_loss=0.06994, over 28103.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3361, pruned_loss=0.09017, over 5652228.42 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.353, pruned_loss=0.113, over 5715616.23 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3358, pruned_loss=0.0884, over 5647043.97 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:39:59,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6684, 2.0116, 1.3206, 1.5824], device='cuda:0'), covar=tensor([0.1036, 0.0550, 0.1037, 0.1078], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0444, 0.0518, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 13:40:50,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.734e+02 1.375e+03 1.833e+03 2.501e+03 7.266e+03, threshold=3.665e+03, percent-clipped=8.0
+2023-03-11 13:41:00,972 INFO [train.py:968] (0/2) Epoch 22, batch 31500, libri_loss[loss=0.3446, simple_loss=0.3792, pruned_loss=0.155, over 29564.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.333, pruned_loss=0.08812, over 5671527.93 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3532, pruned_loss=0.1133, over 5719105.66 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3321, pruned_loss=0.08586, over 5663007.16 frames. ], batch size: 76, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:42:08,038 INFO [train.py:968] (0/2) Epoch 22, batch 31550, giga_loss[loss=0.3029, simple_loss=0.3502, pruned_loss=0.1278, over 24439.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.335, pruned_loss=0.08973, over 5671409.14 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3527, pruned_loss=0.113, over 5721778.56 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3346, pruned_loss=0.08783, over 5661700.44 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:42:09,571 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=989282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:42:49,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6211, 4.6524, 1.7964, 1.8448], device='cuda:0'), covar=tensor([0.0962, 0.0194, 0.0918, 0.1289], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0555, 0.0389, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 13:43:06,007 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.109e+02 1.533e+03 2.002e+03 2.777e+03 6.411e+03, threshold=4.003e+03, percent-clipped=8.0
+2023-03-11 13:43:11,362 INFO [train.py:968] (0/2) Epoch 22, batch 31600, giga_loss[loss=0.2703, simple_loss=0.3598, pruned_loss=0.0904, over 28701.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3366, pruned_loss=0.08875, over 5676399.88 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.352, pruned_loss=0.1127, over 5724533.69 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3366, pruned_loss=0.08722, over 5665623.11 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 13:44:13,712 INFO [train.py:968] (0/2) Epoch 22, batch 31650, giga_loss[loss=0.2349, simple_loss=0.3349, pruned_loss=0.06745, over 28725.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3389, pruned_loss=0.08809, over 5673547.01 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.352, pruned_loss=0.1128, over 5729009.81 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3385, pruned_loss=0.08604, over 5659440.42 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:45:10,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.222e+02 1.407e+03 2.081e+03 3.051e+03 6.535e+03, threshold=4.162e+03, percent-clipped=12.0
+2023-03-11 13:45:14,876 INFO [train.py:968] (0/2) Epoch 22, batch 31700, giga_loss[loss=0.2555, simple_loss=0.3457, pruned_loss=0.0826, over 28815.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.341, pruned_loss=0.08808, over 5671805.69 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.352, pruned_loss=0.1129, over 5730282.10 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3405, pruned_loss=0.08593, over 5658156.03 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:45:27,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3598, 1.6331, 1.5875, 1.1515], device='cuda:0'), covar=tensor([0.1870, 0.2822, 0.1611, 0.1972], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0702, 0.0952, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 13:46:15,741 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-11 13:46:15,932 INFO [train.py:968] (0/2) Epoch 22, batch 31750, giga_loss[loss=0.2894, simple_loss=0.3662, pruned_loss=0.1063, over 28501.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.34, pruned_loss=0.08656, over 5677011.61 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3521, pruned_loss=0.1128, over 5731796.10 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3394, pruned_loss=0.08442, over 5663902.28 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:46:41,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3617, 1.2037, 3.8517, 3.3975], device='cuda:0'), covar=tensor([0.1581, 0.2897, 0.0438, 0.0954], device='cuda:0'), in_proj_covar=tensor([0.0764, 0.0655, 0.0967, 0.0910], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 13:46:42,296 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-11 13:46:57,356 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-11 13:47:06,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.668e+02 1.424e+03 1.834e+03 2.247e+03 7.698e+03, threshold=3.669e+03, percent-clipped=7.0
+2023-03-11 13:47:10,641 INFO [train.py:968] (0/2) Epoch 22, batch 31800, giga_loss[loss=0.2685, simple_loss=0.3504, pruned_loss=0.09327, over 28472.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3417, pruned_loss=0.08919, over 5687841.36 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3522, pruned_loss=0.1132, over 5734861.75 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3406, pruned_loss=0.08621, over 5672416.01 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:47:23,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2633, 1.1082, 3.6603, 3.2462], device='cuda:0'), covar=tensor([0.1626, 0.3003, 0.0469, 0.1244], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0656, 0.0968, 0.0911], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 13:47:39,424 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=989552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:48:01,812 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2635, 1.2642, 3.2065, 2.9442], device='cuda:0'), covar=tensor([0.1439, 0.2707, 0.0474, 0.1995], device='cuda:0'), in_proj_covar=tensor([0.0764, 0.0656, 0.0968, 0.0910], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 13:48:19,877 INFO [train.py:968] (0/2) Epoch 22, batch 31850, giga_loss[loss=0.2967, simple_loss=0.3655, pruned_loss=0.114, over 28420.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3398, pruned_loss=0.08946, over 5688855.14 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3522, pruned_loss=0.1132, over 5738256.68 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3388, pruned_loss=0.08669, over 5672978.59 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:49:04,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4632, 2.0213, 1.4234, 0.7087], device='cuda:0'), covar=tensor([0.5402, 0.2658, 0.4303, 0.6287], device='cuda:0'), in_proj_covar=tensor([0.1754, 0.1644, 0.1593, 0.1429], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 13:49:26,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2692, 5.0740, 4.8492, 2.4516], device='cuda:0'), covar=tensor([0.0433, 0.0606, 0.0735, 0.1825], device='cuda:0'), in_proj_covar=tensor([0.1232, 0.1137, 0.0960, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 13:49:35,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.934e+02 1.440e+03 1.847e+03 2.808e+03 8.265e+03, threshold=3.694e+03, percent-clipped=13.0
+2023-03-11 13:49:39,906 INFO [train.py:968] (0/2) Epoch 22, batch 31900, giga_loss[loss=0.2157, simple_loss=0.3038, pruned_loss=0.06373, over 29140.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3398, pruned_loss=0.08998, over 5678204.31 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3521, pruned_loss=0.1132, over 5729342.29 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3389, pruned_loss=0.08755, over 5673319.67 frames. ], batch size: 120, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:50:21,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=989657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:50:54,936 INFO [train.py:968] (0/2) Epoch 22, batch 31950, libri_loss[loss=0.2665, simple_loss=0.3232, pruned_loss=0.1049, over 29401.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3339, pruned_loss=0.08667, over 5678461.88 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3516, pruned_loss=0.1131, over 5731773.85 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3335, pruned_loss=0.08454, over 5671564.81 frames. ], batch size: 67, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:51:38,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5273, 1.8244, 1.4770, 1.6289], device='cuda:0'), covar=tensor([0.2688, 0.2527, 0.2885, 0.2340], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1095, 0.1341, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 13:51:51,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.298e+02 1.333e+03 1.683e+03 2.361e+03 6.013e+03, threshold=3.365e+03, percent-clipped=4.0
+2023-03-11 13:51:57,558 INFO [train.py:968] (0/2) Epoch 22, batch 32000, giga_loss[loss=0.2056, simple_loss=0.2922, pruned_loss=0.05948, over 28943.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3309, pruned_loss=0.08463, over 5678022.05 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3513, pruned_loss=0.1129, over 5733415.14 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3306, pruned_loss=0.08276, over 5670456.20 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:52:17,618 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3507, 1.2261, 1.2153, 1.5189], device='cuda:0'), covar=tensor([0.0780, 0.0353, 0.0350, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 13:52:23,896 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=989751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:53:01,206 INFO [train.py:968] (0/2) Epoch 22, batch 32050, giga_loss[loss=0.2964, simple_loss=0.3715, pruned_loss=0.1107, over 28724.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3302, pruned_loss=0.08498, over 5683432.33 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3516, pruned_loss=0.113, over 5733431.98 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3291, pruned_loss=0.08264, over 5675871.01 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:53:01,849 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-11 13:53:02,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=989782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:53:07,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=989786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:53:22,076 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=989800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:53:24,489 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=989803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:53:52,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.558e+03 2.028e+03 2.594e+03 9.603e+03, threshold=4.056e+03, percent-clipped=13.0
+2023-03-11 13:53:57,678 INFO [train.py:968] (0/2) Epoch 22, batch 32100, giga_loss[loss=0.257, simple_loss=0.3391, pruned_loss=0.08741, over 28141.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3346, pruned_loss=0.08724, over 5678904.29 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3515, pruned_loss=0.113, over 5728201.75 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3333, pruned_loss=0.08466, over 5675781.62 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:54:00,055 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=989832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:54:54,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2602, 1.5176, 1.4355, 1.2467], device='cuda:0'), covar=tensor([0.2447, 0.2053, 0.1399, 0.1990], device='cuda:0'), in_proj_covar=tensor([0.1936, 0.1873, 0.1791, 0.1936], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 13:54:58,742 INFO [train.py:968] (0/2) Epoch 22, batch 32150, giga_loss[loss=0.2329, simple_loss=0.3136, pruned_loss=0.0761, over 28964.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3338, pruned_loss=0.08813, over 5691810.54 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3511, pruned_loss=0.1128, over 5731863.53 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3329, pruned_loss=0.08578, over 5685328.85 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:55:22,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4281, 1.6062, 1.3555, 1.6047], device='cuda:0'), covar=tensor([0.0752, 0.0347, 0.0346, 0.0894], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 13:55:57,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.462e+03 1.998e+03 2.841e+03 1.046e+04, threshold=3.997e+03, percent-clipped=7.0
+2023-03-11 13:55:57,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=989927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:56:00,845 INFO [train.py:968] (0/2) Epoch 22, batch 32200, giga_loss[loss=0.276, simple_loss=0.3568, pruned_loss=0.09759, over 28701.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3334, pruned_loss=0.0887, over 5690577.36 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3507, pruned_loss=0.1126, over 5735423.42 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3327, pruned_loss=0.0865, over 5681441.06 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:56:06,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3609, 1.2550, 3.6033, 3.1699], device='cuda:0'), covar=tensor([0.1555, 0.2749, 0.0534, 0.1005], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0656, 0.0970, 0.0910], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 13:56:18,095 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-11 13:57:00,328 INFO [train.py:968] (0/2) Epoch 22, batch 32250, giga_loss[loss=0.2315, simple_loss=0.3207, pruned_loss=0.0712, over 28951.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3337, pruned_loss=0.08956, over 5682736.32 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3504, pruned_loss=0.1125, over 5729899.27 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3329, pruned_loss=0.08707, over 5678316.35 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:57:24,915 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-990000.pt
+2023-03-11 13:57:37,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7315, 2.0613, 1.9272, 1.7385], device='cuda:0'), covar=tensor([0.1715, 0.1678, 0.1931, 0.1781], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0737, 0.0706, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 13:57:39,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5897, 4.4046, 4.2126, 2.0427], device='cuda:0'), covar=tensor([0.0622, 0.0821, 0.0863, 0.2005], device='cuda:0'), in_proj_covar=tensor([0.1229, 0.1137, 0.0959, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 13:57:53,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8781, 1.2530, 1.2635, 1.0916], device='cuda:0'), covar=tensor([0.1892, 0.1328, 0.2137, 0.1569], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0738, 0.0706, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 13:58:02,319 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.581e+03 2.146e+03 2.804e+03 7.939e+03, threshold=4.293e+03, percent-clipped=4.0
+2023-03-11 13:58:06,502 INFO [train.py:968] (0/2) Epoch 22, batch 32300, libri_loss[loss=0.3215, simple_loss=0.3788, pruned_loss=0.1321, over 29266.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3361, pruned_loss=0.09044, over 5670315.11 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3506, pruned_loss=0.1128, over 5719911.01 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3347, pruned_loss=0.08745, over 5674310.71 frames. ], batch size: 94, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:58:58,116 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:59:05,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:59:15,648 INFO [train.py:968] (0/2) Epoch 22, batch 32350, libri_loss[loss=0.2667, simple_loss=0.3396, pruned_loss=0.09691, over 29767.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3377, pruned_loss=0.09075, over 5664681.14 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3504, pruned_loss=0.1127, over 5716833.84 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3362, pruned_loss=0.08756, over 5668796.12 frames. ], batch size: 87, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:59:47,259 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:59:51,690 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:00:13,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-11 14:00:18,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:00:19,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.618e+02 1.374e+03 1.967e+03 2.840e+03 7.376e+03, threshold=3.934e+03, percent-clipped=7.0
+2023-03-11 14:00:27,307 INFO [train.py:968] (0/2) Epoch 22, batch 32400, giga_loss[loss=0.2693, simple_loss=0.3401, pruned_loss=0.09932, over 28463.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3358, pruned_loss=0.08982, over 5663950.89 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3496, pruned_loss=0.1123, over 5719511.82 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3349, pruned_loss=0.08692, over 5663306.91 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:00:48,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3299, 1.8710, 1.5155, 1.6083], device='cuda:0'), covar=tensor([0.1940, 0.1593, 0.1933, 0.1709], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0736, 0.0704, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 14:00:59,962 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:01:04,549 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:01:31,390 INFO [train.py:968] (0/2) Epoch 22, batch 32450, libri_loss[loss=0.2987, simple_loss=0.3602, pruned_loss=0.1186, over 25767.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3313, pruned_loss=0.08815, over 5671528.28 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3493, pruned_loss=0.1121, over 5720126.93 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3304, pruned_loss=0.08549, over 5669768.23 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:02:20,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.495e+02 1.460e+03 1.784e+03 2.712e+03 6.255e+03, threshold=3.568e+03, percent-clipped=6.0
+2023-03-11 14:02:23,044 INFO [train.py:968] (0/2) Epoch 22, batch 32500, giga_loss[loss=0.2221, simple_loss=0.3041, pruned_loss=0.06999, over 28934.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.327, pruned_loss=0.08673, over 5672325.43 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3485, pruned_loss=0.1116, over 5715658.99 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3258, pruned_loss=0.08349, over 5671913.60 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:02:45,948 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:02:48,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6193, 1.8217, 1.2700, 1.4117], device='cuda:0'), covar=tensor([0.0940, 0.0522, 0.1008, 0.1120], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0442, 0.0517, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 14:03:13,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8411, 1.1203, 2.9014, 2.7315], device='cuda:0'), covar=tensor([0.1648, 0.2496, 0.0585, 0.0878], device='cuda:0'), in_proj_covar=tensor([0.0764, 0.0656, 0.0968, 0.0908], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 14:03:15,534 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:18,923 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:28,752 INFO [train.py:968] (0/2) Epoch 22, batch 32550, giga_loss[loss=0.2154, simple_loss=0.3098, pruned_loss=0.0605, over 28876.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.327, pruned_loss=0.08693, over 5663284.16 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3481, pruned_loss=0.1113, over 5709112.77 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3258, pruned_loss=0.0839, over 5668439.38 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:03:38,348 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:50,300 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:50,918 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:52,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:53,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:58,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:04:01,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4437, 1.7533, 1.4167, 1.5485], device='cuda:0'), covar=tensor([0.2814, 0.2631, 0.3029, 0.2267], device='cuda:0'), in_proj_covar=tensor([0.1517, 0.1094, 0.1341, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 14:04:23,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.144e+02 1.540e+03 2.045e+03 2.785e+03 7.506e+03, threshold=4.089e+03, percent-clipped=9.0
+2023-03-11 14:04:28,217 INFO [train.py:968] (0/2) Epoch 22, batch 32600, giga_loss[loss=0.2499, simple_loss=0.3323, pruned_loss=0.08372, over 28890.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3288, pruned_loss=0.08788, over 5665846.95 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.348, pruned_loss=0.1112, over 5711573.96 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3277, pruned_loss=0.08529, over 5667259.23 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:04:29,649 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:04:36,880 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:05:05,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:05:33,585 INFO [train.py:968] (0/2) Epoch 22, batch 32650, giga_loss[loss=0.2043, simple_loss=0.2997, pruned_loss=0.05445, over 29041.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3268, pruned_loss=0.08575, over 5665792.62 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3478, pruned_loss=0.1111, over 5710662.09 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.326, pruned_loss=0.08369, over 5667539.38 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:05:35,331 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 14:06:36,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.938e+02 1.456e+03 1.908e+03 2.842e+03 6.480e+03, threshold=3.817e+03, percent-clipped=10.0
+2023-03-11 14:06:40,046 INFO [train.py:968] (0/2) Epoch 22, batch 32700, giga_loss[loss=0.2594, simple_loss=0.3296, pruned_loss=0.09458, over 27647.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3256, pruned_loss=0.08437, over 5660951.74 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3476, pruned_loss=0.111, over 5709474.55 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3248, pruned_loss=0.08248, over 5662781.16 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:06:50,627 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2735, 4.1151, 3.8733, 2.0382], device='cuda:0'), covar=tensor([0.0557, 0.0716, 0.0847, 0.2029], device='cuda:0'), in_proj_covar=tensor([0.1226, 0.1129, 0.0952, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 14:07:39,620 INFO [train.py:968] (0/2) Epoch 22, batch 32750, giga_loss[loss=0.2166, simple_loss=0.3013, pruned_loss=0.06591, over 28943.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3247, pruned_loss=0.08459, over 5665554.10 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3476, pruned_loss=0.111, over 5715635.53 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3233, pruned_loss=0.0822, over 5659946.17 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:07:39,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:08:45,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.413e+02 1.441e+03 1.858e+03 2.433e+03 5.991e+03, threshold=3.716e+03, percent-clipped=7.0
+2023-03-11 14:08:50,900 INFO [train.py:968] (0/2) Epoch 22, batch 32800, giga_loss[loss=0.2473, simple_loss=0.3287, pruned_loss=0.08296, over 29027.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3249, pruned_loss=0.08355, over 5679672.02 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3476, pruned_loss=0.1109, over 5718495.51 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3235, pruned_loss=0.08137, over 5672147.60 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:09:17,130 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:09:52,880 INFO [train.py:968] (0/2) Epoch 22, batch 32850, giga_loss[loss=0.2494, simple_loss=0.3373, pruned_loss=0.08076, over 28420.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3264, pruned_loss=0.08485, over 5668270.26 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3482, pruned_loss=0.1114, over 5704060.20 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3243, pruned_loss=0.082, over 5673022.39 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:10:41,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:10:47,545 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:10:50,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:10:51,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.511e+02 1.308e+03 1.637e+03 2.176e+03 7.493e+03, threshold=3.273e+03, percent-clipped=5.0
+2023-03-11 14:10:52,807 INFO [train.py:968] (0/2) Epoch 22, batch 32900, giga_loss[loss=0.2399, simple_loss=0.3228, pruned_loss=0.07852, over 29006.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3264, pruned_loss=0.08539, over 5676699.83 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3477, pruned_loss=0.1112, over 5707271.75 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3249, pruned_loss=0.08303, over 5677151.59 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:11:25,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2163, 1.5367, 1.2267, 0.6801], device='cuda:0'), covar=tensor([0.3741, 0.2264, 0.2998, 0.5436], device='cuda:0'), in_proj_covar=tensor([0.1758, 0.1656, 0.1601, 0.1439], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 14:11:26,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:11:34,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:11:54,070 INFO [train.py:968] (0/2) Epoch 22, batch 32950, giga_loss[loss=0.2579, simple_loss=0.3414, pruned_loss=0.08717, over 28678.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3259, pruned_loss=0.08489, over 5674593.68 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3478, pruned_loss=0.1112, over 5711203.54 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3243, pruned_loss=0.08248, over 5670918.03 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:12:52,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.624e+02 1.494e+03 1.952e+03 2.747e+03 8.243e+03, threshold=3.905e+03, percent-clipped=12.0
+2023-03-11 14:12:56,125 INFO [train.py:968] (0/2) Epoch 22, batch 33000, giga_loss[loss=0.2654, simple_loss=0.3471, pruned_loss=0.09188, over 28902.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3279, pruned_loss=0.08459, over 5666906.32 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3478, pruned_loss=0.1113, over 5713271.00 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3264, pruned_loss=0.08243, over 5661812.88 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:12:56,129 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 14:13:01,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3203, 3.0943, 1.4081, 1.5164], device='cuda:0'), covar=tensor([0.1132, 0.0382, 0.1059, 0.1529], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0551, 0.0387, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 14:13:04,890 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.1952, simple_loss=0.2965, pruned_loss=0.04696, over 944034.00 frames. 
+2023-03-11 14:13:04,890 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 14:13:08,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4050, 1.7012, 1.3097, 1.6945], device='cuda:0'), covar=tensor([0.2842, 0.2815, 0.3279, 0.2509], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1096, 0.1341, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 14:13:08,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:13:43,509 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:13:44,242 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-11 14:13:45,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:14:01,201 INFO [train.py:968] (0/2) Epoch 22, batch 33050, giga_loss[loss=0.2271, simple_loss=0.3251, pruned_loss=0.06457, over 28887.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3306, pruned_loss=0.08556, over 5663354.96 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3471, pruned_loss=0.1109, over 5708140.20 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3297, pruned_loss=0.0836, over 5662362.19 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:14:23,134 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:14:37,403 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:14:41,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:15:05,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.387e+02 1.468e+03 1.956e+03 2.616e+03 6.434e+03, threshold=3.911e+03, percent-clipped=7.0
+2023-03-11 14:15:07,183 INFO [train.py:968] (0/2) Epoch 22, batch 33100, giga_loss[loss=0.2776, simple_loss=0.3516, pruned_loss=0.1018, over 28509.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3315, pruned_loss=0.08551, over 5666847.29 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3471, pruned_loss=0.1109, over 5709968.59 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3306, pruned_loss=0.08377, over 5663891.74 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:15:19,497 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:03,837 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:07,168 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:07,494 INFO [train.py:968] (0/2) Epoch 22, batch 33150, giga_loss[loss=0.2444, simple_loss=0.3359, pruned_loss=0.07639, over 28939.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3319, pruned_loss=0.08623, over 5667582.30 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.347, pruned_loss=0.1109, over 5711509.77 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3309, pruned_loss=0.0842, over 5663163.69 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:16:35,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:36,336 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:41,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-11 14:16:55,861 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-11 14:16:56,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:59,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.592e+02 1.537e+03 1.971e+03 3.297e+03 1.700e+04, threshold=3.942e+03, percent-clipped=18.0
+2023-03-11 14:17:00,598 INFO [train.py:968] (0/2) Epoch 22, batch 33200, giga_loss[loss=0.2068, simple_loss=0.297, pruned_loss=0.05826, over 28599.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3295, pruned_loss=0.08529, over 5658620.67 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3474, pruned_loss=0.1115, over 5694254.12 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3275, pruned_loss=0.08183, over 5669817.34 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:17:22,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3025, 1.3833, 3.4461, 3.1514], device='cuda:0'), covar=tensor([0.1553, 0.2761, 0.0473, 0.1593], device='cuda:0'), in_proj_covar=tensor([0.0759, 0.0649, 0.0959, 0.0901], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 14:17:58,601 INFO [train.py:968] (0/2) Epoch 22, batch 33250, libri_loss[loss=0.2275, simple_loss=0.2925, pruned_loss=0.08126, over 29474.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3286, pruned_loss=0.08484, over 5665083.16 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3471, pruned_loss=0.1113, over 5699324.58 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3266, pruned_loss=0.08133, over 5668430.90 frames. ], batch size: 70, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:18:54,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8558, 1.0831, 1.0740, 0.8636], device='cuda:0'), covar=tensor([0.2172, 0.2101, 0.1304, 0.1970], device='cuda:0'), in_proj_covar=tensor([0.1931, 0.1866, 0.1784, 0.1929], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 14:18:56,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.498e+02 1.343e+03 1.710e+03 2.407e+03 5.547e+03, threshold=3.419e+03, percent-clipped=4.0
+2023-03-11 14:18:57,460 INFO [train.py:968] (0/2) Epoch 22, batch 33300, giga_loss[loss=0.2304, simple_loss=0.3098, pruned_loss=0.07545, over 26984.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3272, pruned_loss=0.08479, over 5670428.24 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3471, pruned_loss=0.1113, over 5703033.39 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3251, pruned_loss=0.08122, over 5668889.52 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:18:58,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2222, 1.2432, 1.3675, 1.0112], device='cuda:0'), covar=tensor([0.1847, 0.3430, 0.1503, 0.1732], device='cuda:0'), in_proj_covar=tensor([0.0898, 0.0695, 0.0945, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 14:19:41,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3773, 1.5710, 1.5889, 1.2453], device='cuda:0'), covar=tensor([0.1708, 0.2624, 0.1536, 0.1901], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0696, 0.0946, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 14:19:43,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=991069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:19:46,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=991072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:19:54,448 INFO [train.py:968] (0/2) Epoch 22, batch 33350, giga_loss[loss=0.2434, simple_loss=0.3283, pruned_loss=0.07929, over 29078.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3277, pruned_loss=0.08485, over 5665639.80 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3461, pruned_loss=0.1107, over 5697310.20 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3261, pruned_loss=0.08157, over 5669034.68 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:20:23,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=991101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:20:56,067 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.563e+02 1.379e+03 1.854e+03 2.690e+03 4.971e+03, threshold=3.708e+03, percent-clipped=11.0
+2023-03-11 14:20:57,513 INFO [train.py:968] (0/2) Epoch 22, batch 33400, giga_loss[loss=0.2393, simple_loss=0.3262, pruned_loss=0.0762, over 28664.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3293, pruned_loss=0.08563, over 5670822.04 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3458, pruned_loss=0.1105, over 5701555.96 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3279, pruned_loss=0.08245, over 5668848.11 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:22:01,585 INFO [train.py:968] (0/2) Epoch 22, batch 33450, giga_loss[loss=0.2319, simple_loss=0.3177, pruned_loss=0.07303, over 28803.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3294, pruned_loss=0.08595, over 5668657.07 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3454, pruned_loss=0.1101, over 5702989.61 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3283, pruned_loss=0.08322, over 5665048.97 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:22:07,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-11 14:23:06,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.366e+02 1.293e+03 1.732e+03 2.197e+03 5.566e+03, threshold=3.464e+03, percent-clipped=6.0
+2023-03-11 14:23:07,069 INFO [train.py:968] (0/2) Epoch 22, batch 33500, giga_loss[loss=0.3162, simple_loss=0.3983, pruned_loss=0.1171, over 28997.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3328, pruned_loss=0.08798, over 5657338.04 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3447, pruned_loss=0.1097, over 5705456.99 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3324, pruned_loss=0.08591, over 5651709.06 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:23:28,818 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 14:24:06,316 INFO [train.py:968] (0/2) Epoch 22, batch 33550, giga_loss[loss=0.2718, simple_loss=0.3446, pruned_loss=0.09952, over 26746.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3357, pruned_loss=0.0886, over 5663631.28 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3447, pruned_loss=0.1097, over 5707445.51 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3353, pruned_loss=0.08672, over 5657022.80 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:24:08,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=991283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:25:12,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2188, 2.5174, 1.3022, 1.3871], device='cuda:0'), covar=tensor([0.1001, 0.0462, 0.0974, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0553, 0.0387, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 14:25:13,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.580e+02 1.599e+03 2.140e+03 3.053e+03 9.623e+03, threshold=4.280e+03, percent-clipped=18.0
+2023-03-11 14:25:15,075 INFO [train.py:968] (0/2) Epoch 22, batch 33600, giga_loss[loss=0.2134, simple_loss=0.3076, pruned_loss=0.05962, over 28883.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3367, pruned_loss=0.08913, over 5667962.23 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3449, pruned_loss=0.1098, over 5707990.26 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.336, pruned_loss=0.08719, over 5661410.87 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:25:39,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5983, 4.4431, 4.2021, 1.9897], device='cuda:0'), covar=tensor([0.0576, 0.0717, 0.0827, 0.1953], device='cuda:0'), in_proj_covar=tensor([0.1220, 0.1122, 0.0947, 0.0710], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 14:26:27,131 INFO [train.py:968] (0/2) Epoch 22, batch 33650, giga_loss[loss=0.2843, simple_loss=0.3532, pruned_loss=0.1077, over 28975.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3341, pruned_loss=0.08803, over 5666942.22 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3452, pruned_loss=0.11, over 5702544.83 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3332, pruned_loss=0.08597, over 5666735.47 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:27:24,049 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=991426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:27:27,686 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=991429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:27:29,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.385e+02 1.426e+03 1.784e+03 2.340e+03 8.859e+03, threshold=3.568e+03, percent-clipped=3.0
+2023-03-11 14:27:29,045 INFO [train.py:968] (0/2) Epoch 22, batch 33700, giga_loss[loss=0.2604, simple_loss=0.3232, pruned_loss=0.09879, over 24396.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3337, pruned_loss=0.08786, over 5665391.39 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3449, pruned_loss=0.1099, over 5695656.16 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3329, pruned_loss=0.08586, over 5670322.77 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:28:10,898 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=991458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:28:41,134 INFO [train.py:968] (0/2) Epoch 22, batch 33750, giga_loss[loss=0.2338, simple_loss=0.3143, pruned_loss=0.07663, over 28544.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3317, pruned_loss=0.08696, over 5668721.48 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3448, pruned_loss=0.1098, over 5696752.88 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3311, pruned_loss=0.08537, over 5671383.79 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:29:44,973 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.383e+02 1.442e+03 1.821e+03 2.603e+03 7.759e+03, threshold=3.643e+03, percent-clipped=12.0
+2023-03-11 14:29:44,986 INFO [train.py:968] (0/2) Epoch 22, batch 33800, giga_loss[loss=0.2359, simple_loss=0.3203, pruned_loss=0.07575, over 28976.00 frames. ], tot_loss[loss=0.252, simple_loss=0.33, pruned_loss=0.08702, over 5674800.42 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3447, pruned_loss=0.1098, over 5699670.51 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3295, pruned_loss=0.08544, over 5673910.29 frames. ], batch size: 285, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:30:21,108 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=991560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:30:23,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8861, 1.1501, 2.8414, 2.6529], device='cuda:0'), covar=tensor([0.1667, 0.2697, 0.0605, 0.1888], device='cuda:0'), in_proj_covar=tensor([0.0760, 0.0648, 0.0957, 0.0899], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 14:30:43,180 INFO [train.py:968] (0/2) Epoch 22, batch 33850, giga_loss[loss=0.2221, simple_loss=0.3134, pruned_loss=0.06538, over 28952.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3296, pruned_loss=0.08634, over 5676312.84 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3444, pruned_loss=0.1095, over 5693418.66 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3291, pruned_loss=0.08462, over 5679672.89 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:31:49,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.877e+02 1.516e+03 1.847e+03 2.579e+03 6.808e+03, threshold=3.695e+03, percent-clipped=7.0
+2023-03-11 14:31:49,125 INFO [train.py:968] (0/2) Epoch 22, batch 33900, giga_loss[loss=0.2138, simple_loss=0.2915, pruned_loss=0.06802, over 24476.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3273, pruned_loss=0.08409, over 5665257.03 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3445, pruned_loss=0.1096, over 5695543.16 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3266, pruned_loss=0.08246, over 5665822.93 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:31:54,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4442, 4.3209, 4.0805, 1.8607], device='cuda:0'), covar=tensor([0.0538, 0.0644, 0.0698, 0.1941], device='cuda:0'), in_proj_covar=tensor([0.1223, 0.1124, 0.0948, 0.0711], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 14:32:41,294 INFO [train.py:968] (0/2) Epoch 22, batch 33950, giga_loss[loss=0.2256, simple_loss=0.3154, pruned_loss=0.06789, over 28836.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3294, pruned_loss=0.08374, over 5661740.11 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3449, pruned_loss=0.1101, over 5682280.75 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3279, pruned_loss=0.08118, over 5673786.43 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:33:04,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3884, 1.7544, 1.3501, 1.4400], device='cuda:0'), covar=tensor([0.2371, 0.2396, 0.2806, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1516, 0.1091, 0.1339, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 14:33:38,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.268e+02 1.379e+03 1.760e+03 2.510e+03 7.463e+03, threshold=3.520e+03, percent-clipped=8.0
+2023-03-11 14:33:38,978 INFO [train.py:968] (0/2) Epoch 22, batch 34000, giga_loss[loss=0.2525, simple_loss=0.3276, pruned_loss=0.08874, over 26815.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3312, pruned_loss=0.0837, over 5669725.47 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3447, pruned_loss=0.11, over 5687604.53 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3299, pruned_loss=0.08114, over 5674355.58 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:34:39,019 INFO [train.py:968] (0/2) Epoch 22, batch 34050, giga_loss[loss=0.2142, simple_loss=0.3061, pruned_loss=0.06115, over 28918.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3321, pruned_loss=0.08419, over 5664959.29 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3444, pruned_loss=0.1098, over 5681661.42 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3311, pruned_loss=0.0819, over 5674119.79 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:35:48,309 INFO [train.py:968] (0/2) Epoch 22, batch 34100, giga_loss[loss=0.2983, simple_loss=0.3616, pruned_loss=0.1175, over 26868.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3328, pruned_loss=0.08544, over 5650568.35 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3443, pruned_loss=0.1099, over 5667667.04 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3319, pruned_loss=0.08302, over 5669324.93 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:35:52,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.779e+02 1.559e+03 2.034e+03 2.581e+03 6.961e+03, threshold=4.068e+03, percent-clipped=8.0
+2023-03-11 14:36:47,149 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 14:36:55,808 INFO [train.py:968] (0/2) Epoch 22, batch 34150, giga_loss[loss=0.3126, simple_loss=0.3721, pruned_loss=0.1266, over 27771.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.334, pruned_loss=0.0865, over 5655197.27 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3442, pruned_loss=0.11, over 5673685.57 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3329, pruned_loss=0.08375, over 5664282.53 frames. ], batch size: 474, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:36:57,634 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-11 14:37:27,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2559, 1.7210, 1.5376, 1.3760], device='cuda:0'), covar=tensor([0.0859, 0.0299, 0.0308, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0110], device='cuda:0')
+2023-03-11 14:37:41,207 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3649, 1.2603, 1.1278, 1.5652], device='cuda:0'), covar=tensor([0.0802, 0.0343, 0.0379, 0.0898], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0110], device='cuda:0')
+2023-03-11 14:38:09,863 INFO [train.py:968] (0/2) Epoch 22, batch 34200, giga_loss[loss=0.2161, simple_loss=0.3117, pruned_loss=0.06023, over 28947.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3341, pruned_loss=0.08553, over 5647473.44 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3442, pruned_loss=0.11, over 5663418.86 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3332, pruned_loss=0.08307, over 5663996.38 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:38:11,614 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.460e+03 1.992e+03 2.978e+03 6.689e+03, threshold=3.984e+03, percent-clipped=9.0
+2023-03-11 14:38:17,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=991935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:39:03,933 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 14:39:15,364 INFO [train.py:968] (0/2) Epoch 22, batch 34250, giga_loss[loss=0.2663, simple_loss=0.3436, pruned_loss=0.09447, over 28764.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3354, pruned_loss=0.08635, over 5660602.30 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3436, pruned_loss=0.1096, over 5670524.08 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3349, pruned_loss=0.08403, over 5666822.90 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:39:40,410 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-992000.pt
+2023-03-11 14:40:20,037 INFO [train.py:968] (0/2) Epoch 22, batch 34300, giga_loss[loss=0.2039, simple_loss=0.3006, pruned_loss=0.05359, over 28943.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3391, pruned_loss=0.08822, over 5671761.93 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3448, pruned_loss=0.1107, over 5677395.19 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3375, pruned_loss=0.08482, over 5670309.83 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:40:21,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.140e+03 1.558e+03 2.121e+03 3.046e+03 5.382e+03, threshold=4.243e+03, percent-clipped=10.0
+2023-03-11 14:40:42,175 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=992048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:41:16,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4598, 1.8566, 1.4526, 1.3854], device='cuda:0'), covar=tensor([0.2479, 0.2545, 0.2904, 0.2289], device='cuda:0'), in_proj_covar=tensor([0.1515, 0.1092, 0.1340, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 14:41:20,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=992078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:41:23,238 INFO [train.py:968] (0/2) Epoch 22, batch 34350, giga_loss[loss=0.2257, simple_loss=0.3133, pruned_loss=0.06903, over 28895.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3388, pruned_loss=0.08858, over 5674971.21 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3448, pruned_loss=0.1105, over 5682624.02 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3373, pruned_loss=0.08517, over 5668977.89 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:41:23,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=992081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:41:51,959 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-11 14:42:00,779 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=992110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:42:21,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2958, 0.8294, 0.9782, 1.3530], device='cuda:0'), covar=tensor([0.0716, 0.0353, 0.0335, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0110], device='cuda:0')
+2023-03-11 14:42:26,596 INFO [train.py:968] (0/2) Epoch 22, batch 34400, libri_loss[loss=0.3078, simple_loss=0.3615, pruned_loss=0.127, over 25707.00 frames. ], tot_loss[loss=0.258, simple_loss=0.338, pruned_loss=0.08894, over 5679882.22 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3454, pruned_loss=0.1108, over 5684871.54 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.336, pruned_loss=0.0851, over 5673069.51 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:42:28,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.416e+02 1.423e+03 1.838e+03 2.804e+03 8.178e+03, threshold=3.676e+03, percent-clipped=5.0
+2023-03-11 14:42:47,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2262, 1.2891, 1.1896, 1.5140], device='cuda:0'), covar=tensor([0.0826, 0.0351, 0.0365, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0110], device='cuda:0')
+2023-03-11 14:43:34,413 INFO [train.py:968] (0/2) Epoch 22, batch 34450, giga_loss[loss=0.2151, simple_loss=0.3148, pruned_loss=0.05773, over 28868.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3353, pruned_loss=0.08719, over 5674482.87 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3454, pruned_loss=0.1108, over 5681105.92 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3334, pruned_loss=0.08335, over 5673357.19 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:44:39,334 INFO [train.py:968] (0/2) Epoch 22, batch 34500, giga_loss[loss=0.2369, simple_loss=0.3224, pruned_loss=0.07567, over 29037.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3331, pruned_loss=0.085, over 5679842.79 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3451, pruned_loss=0.1106, over 5683758.72 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3317, pruned_loss=0.08182, over 5676451.34 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:44:41,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.871e+02 1.309e+03 1.690e+03 2.427e+03 6.968e+03, threshold=3.379e+03, percent-clipped=11.0
+2023-03-11 14:44:55,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5094, 1.5983, 1.7159, 1.3091], device='cuda:0'), covar=tensor([0.1889, 0.2718, 0.1632, 0.1853], device='cuda:0'), in_proj_covar=tensor([0.0897, 0.0695, 0.0945, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 14:45:42,512 INFO [train.py:968] (0/2) Epoch 22, batch 34550, giga_loss[loss=0.256, simple_loss=0.3387, pruned_loss=0.08661, over 28993.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.335, pruned_loss=0.08627, over 5674831.94 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.345, pruned_loss=0.1106, over 5687454.10 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3338, pruned_loss=0.08346, over 5668734.71 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:45:55,251 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 14:46:43,749 INFO [train.py:968] (0/2) Epoch 22, batch 34600, giga_loss[loss=0.2104, simple_loss=0.3003, pruned_loss=0.06022, over 28857.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3384, pruned_loss=0.0883, over 5674740.00 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3453, pruned_loss=0.1109, over 5690388.04 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3371, pruned_loss=0.08539, over 5666956.18 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:46:45,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.454e+03 2.259e+03 3.276e+03 1.312e+04, threshold=4.518e+03, percent-clipped=20.0
+2023-03-11 14:47:35,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2321, 2.1140, 1.7366, 1.4428], device='cuda:0'), covar=tensor([0.0833, 0.0280, 0.0292, 0.1094], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0110], device='cuda:0')
+2023-03-11 14:47:38,991 INFO [train.py:968] (0/2) Epoch 22, batch 34650, giga_loss[loss=0.2543, simple_loss=0.3403, pruned_loss=0.08414, over 28869.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3378, pruned_loss=0.08855, over 5668385.77 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3459, pruned_loss=0.1114, over 5674646.57 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.336, pruned_loss=0.08523, over 5676297.13 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 14:48:07,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5128, 5.3772, 5.1071, 2.3849], device='cuda:0'), covar=tensor([0.0432, 0.0528, 0.0626, 0.1844], device='cuda:0'), in_proj_covar=tensor([0.1219, 0.1119, 0.0944, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0011], device='cuda:0')
+2023-03-11 14:48:24,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=992423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:48:32,121 INFO [train.py:968] (0/2) Epoch 22, batch 34700, giga_loss[loss=0.2829, simple_loss=0.3646, pruned_loss=0.1006, over 28934.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3351, pruned_loss=0.08829, over 5667783.92 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3455, pruned_loss=0.1109, over 5683491.72 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3337, pruned_loss=0.08516, over 5665343.48 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 14:48:34,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.213e+02 1.548e+03 2.027e+03 2.901e+03 1.345e+04, threshold=4.053e+03, percent-clipped=8.0
+2023-03-11 14:49:29,986 INFO [train.py:968] (0/2) Epoch 22, batch 34750, libri_loss[loss=0.2585, simple_loss=0.3316, pruned_loss=0.09268, over 29521.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3344, pruned_loss=0.08857, over 5671868.56 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3453, pruned_loss=0.1107, over 5688483.24 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3331, pruned_loss=0.08571, over 5665136.00 frames. ], batch size: 81, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 14:50:23,367 INFO [train.py:968] (0/2) Epoch 22, batch 34800, giga_loss[loss=0.3178, simple_loss=0.3938, pruned_loss=0.1209, over 29007.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3401, pruned_loss=0.09182, over 5667367.71 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3452, pruned_loss=0.1107, over 5681954.96 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3391, pruned_loss=0.08922, over 5667027.49 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:50:26,535 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.933e+02 1.457e+03 1.903e+03 2.460e+03 7.700e+03, threshold=3.806e+03, percent-clipped=2.0
+2023-03-11 14:50:52,359 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=992566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:50:56,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=992569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:51:07,485 INFO [train.py:968] (0/2) Epoch 22, batch 34850, giga_loss[loss=0.2777, simple_loss=0.364, pruned_loss=0.0957, over 28494.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3486, pruned_loss=0.09672, over 5664854.15 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.345, pruned_loss=0.1106, over 5680790.30 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3478, pruned_loss=0.09434, over 5665619.55 frames. ], batch size: 60, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:51:14,461 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-11 14:51:23,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=992598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:51:43,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4337, 2.0309, 1.5522, 0.6106], device='cuda:0'), covar=tensor([0.5275, 0.3255, 0.4725, 0.6467], device='cuda:0'), in_proj_covar=tensor([0.1760, 0.1664, 0.1605, 0.1444], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 14:51:51,963 INFO [train.py:968] (0/2) Epoch 22, batch 34900, giga_loss[loss=0.287, simple_loss=0.3541, pruned_loss=0.1099, over 27627.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3531, pruned_loss=0.09989, over 5659016.05 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3452, pruned_loss=0.1107, over 5675303.62 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3525, pruned_loss=0.09769, over 5664493.36 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:51:54,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.755e+02 1.496e+03 1.882e+03 2.747e+03 1.512e+04, threshold=3.765e+03, percent-clipped=12.0
+2023-03-11 14:52:30,620 INFO [train.py:968] (0/2) Epoch 22, batch 34950, giga_loss[loss=0.2599, simple_loss=0.3352, pruned_loss=0.09229, over 28887.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.349, pruned_loss=0.0984, over 5678993.15 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3452, pruned_loss=0.1105, over 5683522.10 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3488, pruned_loss=0.09632, over 5675596.83 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:52:44,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=992697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:53:12,379 INFO [train.py:968] (0/2) Epoch 22, batch 35000, giga_loss[loss=0.2642, simple_loss=0.3412, pruned_loss=0.09359, over 28725.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3425, pruned_loss=0.09582, over 5688641.95 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3449, pruned_loss=0.1104, over 5687638.45 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3425, pruned_loss=0.09399, over 5682203.84 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:53:14,818 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.141e+02 1.214e+03 1.436e+03 2.159e+03 6.867e+03, threshold=2.872e+03, percent-clipped=2.0
+2023-03-11 14:53:54,795 INFO [train.py:968] (0/2) Epoch 22, batch 35050, giga_loss[loss=0.2996, simple_loss=0.3574, pruned_loss=0.1209, over 28575.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3359, pruned_loss=0.09337, over 5677326.23 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3449, pruned_loss=0.1104, over 5682269.53 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3358, pruned_loss=0.09157, over 5676887.18 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:54:36,122 INFO [train.py:968] (0/2) Epoch 22, batch 35100, giga_loss[loss=0.2005, simple_loss=0.2788, pruned_loss=0.06114, over 28822.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3295, pruned_loss=0.0907, over 5675558.89 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3455, pruned_loss=0.1105, over 5677664.02 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3286, pruned_loss=0.08867, over 5679100.24 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:54:37,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.644e+02 1.051e+03 1.261e+03 1.776e+03 3.696e+03, threshold=2.522e+03, percent-clipped=6.0
+2023-03-11 14:55:18,685 INFO [train.py:968] (0/2) Epoch 22, batch 35150, giga_loss[loss=0.2075, simple_loss=0.2711, pruned_loss=0.07194, over 23817.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3229, pruned_loss=0.08805, over 5679983.28 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3459, pruned_loss=0.1109, over 5680198.49 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3217, pruned_loss=0.086, over 5680551.44 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:55:46,291 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-11 14:55:57,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-11 14:56:01,216 INFO [train.py:968] (0/2) Epoch 22, batch 35200, giga_loss[loss=0.2132, simple_loss=0.2876, pruned_loss=0.06944, over 27689.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3187, pruned_loss=0.08615, over 5684536.54 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.346, pruned_loss=0.1106, over 5689884.75 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.3167, pruned_loss=0.08378, over 5676548.04 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:56:03,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.564e+02 1.174e+03 1.607e+03 2.029e+03 8.082e+03, threshold=3.214e+03, percent-clipped=17.0
+2023-03-11 14:56:41,163 INFO [train.py:968] (0/2) Epoch 22, batch 35250, giga_loss[loss=0.2187, simple_loss=0.2929, pruned_loss=0.07227, over 28922.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3169, pruned_loss=0.08525, over 5681701.62 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3463, pruned_loss=0.1105, over 5682606.94 frames. ], giga_tot_loss[loss=0.2395, simple_loss=0.3139, pruned_loss=0.08253, over 5681973.39 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:57:22,773 INFO [train.py:968] (0/2) Epoch 22, batch 35300, giga_loss[loss=0.2334, simple_loss=0.3095, pruned_loss=0.0786, over 29054.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3137, pruned_loss=0.08346, over 5693220.44 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.347, pruned_loss=0.1108, over 5686128.59 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.31, pruned_loss=0.08033, over 5690529.33 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:57:25,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.974e+02 1.061e+03 1.377e+03 2.042e+03 4.891e+03, threshold=2.754e+03, percent-clipped=6.0
+2023-03-11 14:57:59,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=993072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:58:05,029 INFO [train.py:968] (0/2) Epoch 22, batch 35350, giga_loss[loss=0.2311, simple_loss=0.3011, pruned_loss=0.08048, over 28779.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3104, pruned_loss=0.08187, over 5706011.13 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.347, pruned_loss=0.1108, over 5691192.01 frames. ], giga_tot_loss[loss=0.2321, simple_loss=0.3066, pruned_loss=0.07876, over 5699552.31 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:58:32,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9709, 2.1028, 1.8531, 2.0705], device='cuda:0'), covar=tensor([0.2550, 0.2817, 0.3003, 0.2601], device='cuda:0'), in_proj_covar=tensor([0.1509, 0.1092, 0.1334, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 14:58:47,766 INFO [train.py:968] (0/2) Epoch 22, batch 35400, giga_loss[loss=0.2111, simple_loss=0.2851, pruned_loss=0.0686, over 29030.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3067, pruned_loss=0.07987, over 5706854.15 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3465, pruned_loss=0.1102, over 5696286.34 frames. ], giga_tot_loss[loss=0.2288, simple_loss=0.3033, pruned_loss=0.07719, over 5697659.36 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:58:50,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.159e+02 1.223e+03 1.659e+03 2.336e+03 6.648e+03, threshold=3.319e+03, percent-clipped=15.0
+2023-03-11 14:59:31,476 INFO [train.py:968] (0/2) Epoch 22, batch 35450, libri_loss[loss=0.2828, simple_loss=0.3526, pruned_loss=0.1065, over 29521.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3042, pruned_loss=0.07886, over 5697572.48 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3469, pruned_loss=0.1104, over 5689629.95 frames. ], giga_tot_loss[loss=0.2259, simple_loss=0.3, pruned_loss=0.0759, over 5696779.22 frames. ], batch size: 84, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:59:38,707 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 15:00:00,239 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=993215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:00:02,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=993218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:00:02,274 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4662, 1.6444, 1.7210, 1.2958], device='cuda:0'), covar=tensor([0.1853, 0.2607, 0.1530, 0.1749], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0700, 0.0953, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 15:00:16,303 INFO [train.py:968] (0/2) Epoch 22, batch 35500, giga_loss[loss=0.1981, simple_loss=0.2826, pruned_loss=0.0568, over 28887.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.2999, pruned_loss=0.07655, over 5689146.95 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3469, pruned_loss=0.1103, over 5690839.80 frames. ], giga_tot_loss[loss=0.2224, simple_loss=0.2965, pruned_loss=0.07417, over 5687477.51 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:00:18,732 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.00 vs. limit=2.0
+2023-03-11 15:00:18,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.446e+02 1.001e+03 1.374e+03 1.858e+03 7.184e+03, threshold=2.748e+03, percent-clipped=5.0
+2023-03-11 15:00:27,050 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.46 vs. limit=5.0
+2023-03-11 15:00:28,356 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=993247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:00:34,265 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=993254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:00:58,651 INFO [train.py:968] (0/2) Epoch 22, batch 35550, giga_loss[loss=0.1904, simple_loss=0.2741, pruned_loss=0.05331, over 28987.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2981, pruned_loss=0.07574, over 5697836.22 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3471, pruned_loss=0.1102, over 5696716.90 frames. ], giga_tot_loss[loss=0.2198, simple_loss=0.2938, pruned_loss=0.07294, over 5691321.19 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:01:42,190 INFO [train.py:968] (0/2) Epoch 22, batch 35600, giga_loss[loss=0.3232, simple_loss=0.3777, pruned_loss=0.1343, over 28686.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2969, pruned_loss=0.07554, over 5691826.30 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3477, pruned_loss=0.1104, over 5687711.11 frames. ], giga_tot_loss[loss=0.2183, simple_loss=0.2919, pruned_loss=0.07239, over 5694687.33 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:01:45,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.432e+02 1.050e+03 1.314e+03 2.059e+03 6.428e+03, threshold=2.628e+03, percent-clipped=13.0
+2023-03-11 15:02:29,131 INFO [train.py:968] (0/2) Epoch 22, batch 35650, giga_loss[loss=0.2893, simple_loss=0.3646, pruned_loss=0.107, over 28614.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3063, pruned_loss=0.08091, over 5682345.56 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3478, pruned_loss=0.1104, over 5688552.37 frames. ], giga_tot_loss[loss=0.2286, simple_loss=0.3015, pruned_loss=0.07783, over 5683967.11 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:03:05,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=993419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:03:17,230 INFO [train.py:968] (0/2) Epoch 22, batch 35700, giga_loss[loss=0.2854, simple_loss=0.3576, pruned_loss=0.1066, over 28780.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3192, pruned_loss=0.08742, over 5684409.81 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.348, pruned_loss=0.1104, over 5687068.51 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3149, pruned_loss=0.08475, over 5687128.68 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:03:20,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.010e+02 1.290e+03 1.710e+03 2.343e+03 9.364e+03, threshold=3.419e+03, percent-clipped=17.0
+2023-03-11 15:03:45,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6491, 1.8598, 1.6025, 1.7032], device='cuda:0'), covar=tensor([0.1836, 0.2202, 0.2106, 0.1981], device='cuda:0'), in_proj_covar=tensor([0.0468, 0.0742, 0.0707, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 15:04:00,695 INFO [train.py:968] (0/2) Epoch 22, batch 35750, giga_loss[loss=0.351, simple_loss=0.4052, pruned_loss=0.1484, over 28759.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3306, pruned_loss=0.09314, over 5690873.57 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3477, pruned_loss=0.1103, over 5690713.52 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.327, pruned_loss=0.0908, over 5689710.00 frames. ], batch size: 92, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:04:43,879 INFO [train.py:968] (0/2) Epoch 22, batch 35800, giga_loss[loss=0.3015, simple_loss=0.3678, pruned_loss=0.1176, over 28232.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3371, pruned_loss=0.09544, over 5685986.75 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3481, pruned_loss=0.1103, over 5687382.24 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3335, pruned_loss=0.09312, over 5688164.87 frames. ], batch size: 77, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:04:47,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.347e+03 1.617e+03 2.020e+03 3.443e+03, threshold=3.234e+03, percent-clipped=1.0
+2023-03-11 15:05:12,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=993564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:05:27,924 INFO [train.py:968] (0/2) Epoch 22, batch 35850, giga_loss[loss=0.2971, simple_loss=0.3737, pruned_loss=0.1102, over 28607.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3402, pruned_loss=0.09572, over 5687118.15 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3486, pruned_loss=0.1105, over 5690671.30 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3368, pruned_loss=0.09343, over 5685910.22 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:06:14,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=993629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:06:16,942 INFO [train.py:968] (0/2) Epoch 22, batch 35900, giga_loss[loss=0.2982, simple_loss=0.3518, pruned_loss=0.1223, over 23761.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3416, pruned_loss=0.0957, over 5687432.08 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3485, pruned_loss=0.1104, over 5694932.38 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3388, pruned_loss=0.0936, over 5682616.22 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:06:21,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.274e+02 1.186e+03 1.596e+03 2.516e+03 8.115e+03, threshold=3.192e+03, percent-clipped=13.0
+2023-03-11 15:07:02,032 INFO [train.py:968] (0/2) Epoch 22, batch 35950, giga_loss[loss=0.2647, simple_loss=0.3353, pruned_loss=0.09705, over 28958.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3449, pruned_loss=0.09772, over 5689941.44 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3488, pruned_loss=0.1106, over 5692703.06 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3424, pruned_loss=0.09581, over 5688407.10 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:07:47,758 INFO [train.py:968] (0/2) Epoch 22, batch 36000, giga_loss[loss=0.3006, simple_loss=0.379, pruned_loss=0.1111, over 28672.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.349, pruned_loss=0.101, over 5680708.97 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3491, pruned_loss=0.1107, over 5695749.35 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3467, pruned_loss=0.09918, over 5676883.97 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:07:47,762 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 15:07:56,283 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2052, simple_loss=0.313, pruned_loss=0.04868, over 944034.00 frames. 
+2023-03-11 15:07:56,284 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 15:07:59,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.124e+02 1.347e+03 1.622e+03 2.231e+03 6.163e+03, threshold=3.245e+03, percent-clipped=8.0
+2023-03-11 15:08:09,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3946, 1.3208, 1.2054, 1.5778], device='cuda:0'), covar=tensor([0.0831, 0.0373, 0.0360, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 15:08:14,711 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-11 15:08:29,768 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=993772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:08:32,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=993775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:08:36,916 INFO [train.py:968] (0/2) Epoch 22, batch 36050, giga_loss[loss=0.2792, simple_loss=0.3589, pruned_loss=0.0998, over 28751.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3516, pruned_loss=0.1021, over 5692563.62 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3496, pruned_loss=0.1109, over 5699415.95 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3494, pruned_loss=0.1003, over 5686299.71 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:08:48,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=993794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:08:57,469 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=993804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:09:18,315 INFO [train.py:968] (0/2) Epoch 22, batch 36100, giga_loss[loss=0.2997, simple_loss=0.3737, pruned_loss=0.1129, over 28180.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3544, pruned_loss=0.1028, over 5695675.05 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3496, pruned_loss=0.1109, over 5701422.87 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3526, pruned_loss=0.1013, over 5688965.48 frames. ], batch size: 77, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:09:22,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.251e+02 1.162e+03 1.508e+03 1.943e+03 3.926e+03, threshold=3.017e+03, percent-clipped=3.0
+2023-03-11 15:10:02,212 INFO [train.py:968] (0/2) Epoch 22, batch 36150, giga_loss[loss=0.2759, simple_loss=0.3581, pruned_loss=0.0968, over 28904.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3561, pruned_loss=0.1034, over 5672435.57 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3502, pruned_loss=0.1113, over 5685448.86 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3542, pruned_loss=0.1018, over 5681555.40 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:10:41,969 INFO [train.py:968] (0/2) Epoch 22, batch 36200, giga_loss[loss=0.2585, simple_loss=0.3458, pruned_loss=0.08561, over 28812.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3575, pruned_loss=0.1031, over 5681908.96 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.351, pruned_loss=0.1117, over 5689570.94 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3554, pruned_loss=0.1012, over 5685164.92 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:10:45,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.748e+02 1.196e+03 1.441e+03 1.763e+03 4.048e+03, threshold=2.882e+03, percent-clipped=6.0
+2023-03-11 15:10:46,537 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=993937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:10:49,861 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=993939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:10:50,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=993940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:10:55,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4143, 1.6592, 1.6742, 1.2529], device='cuda:0'), covar=tensor([0.1864, 0.2509, 0.1523, 0.1774], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0696, 0.0948, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 15:11:12,593 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=993969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:11:21,903 INFO [train.py:968] (0/2) Epoch 22, batch 36250, libri_loss[loss=0.2862, simple_loss=0.3587, pruned_loss=0.1069, over 29203.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3567, pruned_loss=0.1017, over 5696837.30 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3515, pruned_loss=0.1119, over 5696486.00 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3548, pruned_loss=0.09961, over 5692912.06 frames. ], batch size: 101, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:11:22,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3897, 2.0682, 1.5089, 0.5479], device='cuda:0'), covar=tensor([0.6760, 0.3396, 0.5104, 0.7372], device='cuda:0'), in_proj_covar=tensor([0.1745, 0.1653, 0.1593, 0.1430], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 15:11:36,845 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-994000.pt
+2023-03-11 15:12:04,630 INFO [train.py:968] (0/2) Epoch 22, batch 36300, giga_loss[loss=0.2696, simple_loss=0.351, pruned_loss=0.09414, over 28788.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3545, pruned_loss=0.09912, over 5694002.45 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3519, pruned_loss=0.1122, over 5689104.36 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3526, pruned_loss=0.09696, over 5698227.06 frames. ], batch size: 243, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:12:09,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.787e+02 1.157e+03 1.500e+03 1.950e+03 6.257e+03, threshold=2.999e+03, percent-clipped=12.0
+2023-03-11 15:12:46,244 INFO [train.py:968] (0/2) Epoch 22, batch 36350, giga_loss[loss=0.3187, simple_loss=0.389, pruned_loss=0.1242, over 28580.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3543, pruned_loss=0.09898, over 5703053.96 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3521, pruned_loss=0.1122, over 5692395.56 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3528, pruned_loss=0.09713, over 5703577.75 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:12:46,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7047, 1.8729, 1.3557, 1.4728], device='cuda:0'), covar=tensor([0.0982, 0.0623, 0.1065, 0.1162], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0442, 0.0519, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 15:12:47,021 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=994082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:12:47,184 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-11 15:12:48,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=994085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:13:12,915 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=994114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:13:19,376 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-11 15:13:29,674 INFO [train.py:968] (0/2) Epoch 22, batch 36400, giga_loss[loss=0.2815, simple_loss=0.3467, pruned_loss=0.1082, over 28796.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3554, pruned_loss=0.1016, over 5705624.66 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3523, pruned_loss=0.1122, over 5696994.00 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.354, pruned_loss=0.09978, over 5702081.78 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:13:35,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.459e+02 1.375e+03 1.617e+03 2.028e+03 5.205e+03, threshold=3.233e+03, percent-clipped=6.0
+2023-03-11 15:14:13,552 INFO [train.py:968] (0/2) Epoch 22, batch 36450, giga_loss[loss=0.3098, simple_loss=0.3713, pruned_loss=0.1241, over 28900.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3586, pruned_loss=0.1062, over 5702478.29 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3524, pruned_loss=0.1122, over 5701169.12 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3576, pruned_loss=0.1046, over 5696119.55 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:14:57,334 INFO [train.py:968] (0/2) Epoch 22, batch 36500, giga_loss[loss=0.2441, simple_loss=0.322, pruned_loss=0.0831, over 28748.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3596, pruned_loss=0.1086, over 5694161.71 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3531, pruned_loss=0.1126, over 5694204.23 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3583, pruned_loss=0.1068, over 5695823.32 frames. ], batch size: 66, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:15:05,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.353e+03 1.671e+03 2.095e+03 6.941e+03, threshold=3.342e+03, percent-clipped=6.0
+2023-03-11 15:15:40,143 INFO [train.py:968] (0/2) Epoch 22, batch 36550, libri_loss[loss=0.3225, simple_loss=0.3865, pruned_loss=0.1292, over 25853.00 frames. ], tot_loss[loss=0.286, simple_loss=0.357, pruned_loss=0.1075, over 5688804.12 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3533, pruned_loss=0.1125, over 5685753.07 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3559, pruned_loss=0.106, over 5698105.67 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:15:55,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2708, 4.0895, 3.8976, 1.7508], device='cuda:0'), covar=tensor([0.0671, 0.0800, 0.0723, 0.2314], device='cuda:0'), in_proj_covar=tensor([0.1221, 0.1131, 0.0951, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 15:16:20,194 INFO [train.py:968] (0/2) Epoch 22, batch 36600, giga_loss[loss=0.307, simple_loss=0.3719, pruned_loss=0.1211, over 28737.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.356, pruned_loss=0.1073, over 5665359.35 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3535, pruned_loss=0.1128, over 5653617.02 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3549, pruned_loss=0.1058, over 5700411.06 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:16:27,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.978e+02 1.298e+03 1.629e+03 2.365e+03 1.526e+04, threshold=3.258e+03, percent-clipped=12.0
+2023-03-11 15:16:34,685 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 15:17:02,525 INFO [train.py:968] (0/2) Epoch 22, batch 36650, giga_loss[loss=0.2505, simple_loss=0.3396, pruned_loss=0.08068, over 28946.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3546, pruned_loss=0.1056, over 5665580.86 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.354, pruned_loss=0.1129, over 5648797.82 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3534, pruned_loss=0.1041, over 5699007.45 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:17:02,739 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=994381.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:17:06,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2195, 1.3440, 3.5284, 3.0330], device='cuda:0'), covar=tensor([0.1541, 0.2612, 0.0410, 0.1682], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0650, 0.0965, 0.0906], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 15:17:45,311 INFO [train.py:968] (0/2) Epoch 22, batch 36700, giga_loss[loss=0.2873, simple_loss=0.3567, pruned_loss=0.109, over 27878.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3521, pruned_loss=0.1034, over 5666809.33 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3539, pruned_loss=0.1127, over 5657033.66 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3511, pruned_loss=0.1021, over 5687012.80 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:17:51,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.353e+02 1.231e+03 1.512e+03 2.019e+03 8.285e+03, threshold=3.024e+03, percent-clipped=9.0
+2023-03-11 15:18:31,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3929, 1.4934, 1.4691, 1.6098], device='cuda:0'), covar=tensor([0.0684, 0.0301, 0.0288, 0.0674], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 15:18:33,807 INFO [train.py:968] (0/2) Epoch 22, batch 36750, giga_loss[loss=0.2356, simple_loss=0.3144, pruned_loss=0.07837, over 28880.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3475, pruned_loss=0.1007, over 5663044.39 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3543, pruned_loss=0.1127, over 5661632.64 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3463, pruned_loss=0.09928, over 5675630.62 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:19:09,764 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-11 15:19:18,243 INFO [train.py:968] (0/2) Epoch 22, batch 36800, giga_loss[loss=0.2536, simple_loss=0.3026, pruned_loss=0.1023, over 23159.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3413, pruned_loss=0.09752, over 5662635.05 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3541, pruned_loss=0.1126, over 5667867.46 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3402, pruned_loss=0.09615, over 5667114.20 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:19:27,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.977e+02 1.058e+03 1.324e+03 1.863e+03 5.215e+03, threshold=2.647e+03, percent-clipped=11.0
+2023-03-11 15:20:10,750 INFO [train.py:968] (0/2) Epoch 22, batch 36850, giga_loss[loss=0.2534, simple_loss=0.3299, pruned_loss=0.08843, over 28660.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3372, pruned_loss=0.09575, over 5656975.85 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3543, pruned_loss=0.1127, over 5669597.29 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3358, pruned_loss=0.09422, over 5658831.81 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:20:57,766 INFO [train.py:968] (0/2) Epoch 22, batch 36900, giga_loss[loss=0.2535, simple_loss=0.3413, pruned_loss=0.08283, over 28804.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3363, pruned_loss=0.09524, over 5646514.97 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3548, pruned_loss=0.1128, over 5663584.96 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3344, pruned_loss=0.0935, over 5653503.98 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:21:06,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.972e+02 1.021e+03 1.401e+03 2.284e+03 7.812e+03, threshold=2.802e+03, percent-clipped=17.0
+2023-03-11 15:21:23,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=994659.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:21:43,885 INFO [train.py:968] (0/2) Epoch 22, batch 36950, giga_loss[loss=0.2516, simple_loss=0.3267, pruned_loss=0.08824, over 29052.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3362, pruned_loss=0.09428, over 5660019.33 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.355, pruned_loss=0.1128, over 5666006.00 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3344, pruned_loss=0.09274, over 5663259.85 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:22:01,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6890, 4.5250, 4.3125, 2.1704], device='cuda:0'), covar=tensor([0.0544, 0.0652, 0.0634, 0.1973], device='cuda:0'), in_proj_covar=tensor([0.1211, 0.1122, 0.0943, 0.0716], device='cuda:0'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 15:22:26,842 INFO [train.py:968] (0/2) Epoch 22, batch 37000, libri_loss[loss=0.2655, simple_loss=0.3375, pruned_loss=0.09678, over 28570.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3354, pruned_loss=0.09352, over 5675316.62 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3551, pruned_loss=0.1128, over 5668242.01 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3336, pruned_loss=0.0921, over 5675842.08 frames. ], batch size: 63, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:22:29,346 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 15:22:33,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.432e+02 1.103e+03 1.330e+03 2.061e+03 6.314e+03, threshold=2.660e+03, percent-clipped=16.0
+2023-03-11 15:22:47,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=994756.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:22:48,414 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=994757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:23:08,005 INFO [train.py:968] (0/2) Epoch 22, batch 37050, libri_loss[loss=0.3397, simple_loss=0.4042, pruned_loss=0.1376, over 27965.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.334, pruned_loss=0.09293, over 5690648.25 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3561, pruned_loss=0.1131, over 5670239.60 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3312, pruned_loss=0.09101, over 5689646.34 frames. ], batch size: 116, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:23:19,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2456, 1.5566, 1.5519, 1.1000], device='cuda:0'), covar=tensor([0.1317, 0.2294, 0.1185, 0.1418], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0697, 0.0947, 0.0849], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 15:23:46,489 INFO [train.py:968] (0/2) Epoch 22, batch 37100, giga_loss[loss=0.2185, simple_loss=0.2958, pruned_loss=0.07056, over 28979.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3319, pruned_loss=0.09176, over 5706184.86 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3565, pruned_loss=0.1132, over 5673591.79 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3287, pruned_loss=0.08966, over 5703305.95 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:23:54,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.745e+02 1.080e+03 1.343e+03 1.956e+03 1.062e+04, threshold=2.685e+03, percent-clipped=10.0
+2023-03-11 15:24:09,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2502, 2.5708, 1.2700, 1.3748], device='cuda:0'), covar=tensor([0.1031, 0.0343, 0.0908, 0.1416], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0549, 0.0386, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 15:24:29,049 INFO [train.py:968] (0/2) Epoch 22, batch 37150, giga_loss[loss=0.2507, simple_loss=0.3305, pruned_loss=0.08541, over 28891.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3301, pruned_loss=0.09122, over 5702982.16 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3572, pruned_loss=0.1135, over 5676618.67 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3265, pruned_loss=0.08899, over 5698255.34 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:24:43,643 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=994899.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:24:45,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=994902.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:25:12,427 INFO [train.py:968] (0/2) Epoch 22, batch 37200, giga_loss[loss=0.2572, simple_loss=0.3274, pruned_loss=0.09353, over 28758.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.327, pruned_loss=0.08976, over 5702496.82 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3574, pruned_loss=0.1136, over 5670498.84 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3237, pruned_loss=0.08765, over 5703988.19 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:25:12,685 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=994931.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:25:22,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.797e+02 1.117e+03 1.491e+03 1.899e+03 1.564e+04, threshold=2.982e+03, percent-clipped=18.0
+2023-03-11 15:25:56,008 INFO [train.py:968] (0/2) Epoch 22, batch 37250, libri_loss[loss=0.4179, simple_loss=0.4679, pruned_loss=0.1839, over 28590.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3259, pruned_loss=0.08933, over 5703930.98 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3588, pruned_loss=0.1144, over 5661395.70 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3214, pruned_loss=0.08649, over 5714363.78 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:26:16,977 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-11 15:26:30,133 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-11 15:26:34,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4146, 1.0936, 1.0887, 1.6724], device='cuda:0'), covar=tensor([0.0770, 0.0383, 0.0365, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 15:26:38,165 INFO [train.py:968] (0/2) Epoch 22, batch 37300, giga_loss[loss=0.2354, simple_loss=0.3184, pruned_loss=0.07626, over 28387.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3251, pruned_loss=0.08895, over 5708312.09 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3595, pruned_loss=0.1146, over 5665913.08 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3198, pruned_loss=0.08568, over 5713896.70 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:26:40,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=995034.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:26:45,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.225e+02 1.237e+03 1.849e+03 2.681e+03 9.824e+03, threshold=3.698e+03, percent-clipped=22.0
+2023-03-11 15:27:07,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2681, 3.1097, 2.9092, 1.3852], device='cuda:0'), covar=tensor([0.0896, 0.0993, 0.0825, 0.2485], device='cuda:0'), in_proj_covar=tensor([0.1214, 0.1123, 0.0945, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 15:27:16,313 INFO [train.py:968] (0/2) Epoch 22, batch 37350, libri_loss[loss=0.2905, simple_loss=0.3579, pruned_loss=0.1116, over 29560.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3229, pruned_loss=0.0874, over 5714032.41 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3599, pruned_loss=0.1145, over 5673253.22 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3168, pruned_loss=0.0838, over 5713251.82 frames. ], batch size: 76, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:27:49,576 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-11 15:27:58,086 INFO [train.py:968] (0/2) Epoch 22, batch 37400, libri_loss[loss=0.3289, simple_loss=0.3929, pruned_loss=0.1325, over 29549.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3228, pruned_loss=0.08746, over 5703090.34 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3606, pruned_loss=0.1149, over 5666659.36 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3161, pruned_loss=0.08329, over 5710111.23 frames. ], batch size: 82, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:27:58,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=995132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:28:04,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.039e+02 1.061e+03 1.318e+03 1.726e+03 7.796e+03, threshold=2.637e+03, percent-clipped=4.0
+2023-03-11 15:28:27,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0026, 3.8523, 3.6807, 1.6928], device='cuda:0'), covar=tensor([0.0772, 0.0908, 0.0974, 0.1960], device='cuda:0'), in_proj_covar=tensor([0.1212, 0.1122, 0.0945, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 15:28:36,098 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=995177.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:28:40,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=995180.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:28:40,917 INFO [train.py:968] (0/2) Epoch 22, batch 37450, giga_loss[loss=0.2221, simple_loss=0.2953, pruned_loss=0.07445, over 28751.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3226, pruned_loss=0.08754, over 5697291.55 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3608, pruned_loss=0.115, over 5660317.84 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.316, pruned_loss=0.08342, over 5709048.53 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:28:45,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6845, 1.9537, 1.6047, 1.6562], device='cuda:0'), covar=tensor([0.2610, 0.2639, 0.3089, 0.2430], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1098, 0.1338, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 15:29:02,001 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=995209.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:29:21,826 INFO [train.py:968] (0/2) Epoch 22, batch 37500, giga_loss[loss=0.2517, simple_loss=0.3239, pruned_loss=0.08979, over 28809.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3258, pruned_loss=0.08942, over 5712569.19 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3613, pruned_loss=0.115, over 5668466.31 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3186, pruned_loss=0.08501, over 5716684.42 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:29:28,369 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.619e+02 1.187e+03 1.439e+03 2.038e+03 6.749e+03, threshold=2.878e+03, percent-clipped=14.0
+2023-03-11 15:30:04,408 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=995275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:30:06,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=995278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:30:08,133 INFO [train.py:968] (0/2) Epoch 22, batch 37550, giga_loss[loss=0.2692, simple_loss=0.3421, pruned_loss=0.09814, over 28916.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3299, pruned_loss=0.09218, over 5710510.93 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3613, pruned_loss=0.1148, over 5670432.51 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.324, pruned_loss=0.08861, over 5712332.41 frames. ], batch size: 186, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:30:13,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7528, 1.9436, 1.4411, 1.4659], device='cuda:0'), covar=tensor([0.1053, 0.0649, 0.1046, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0444, 0.0521, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 15:30:36,900 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=995307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:30:47,673 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-11 15:31:04,460 INFO [train.py:968] (0/2) Epoch 22, batch 37600, giga_loss[loss=0.4581, simple_loss=0.4758, pruned_loss=0.2202, over 26575.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3383, pruned_loss=0.0978, over 5697981.85 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3611, pruned_loss=0.1147, over 5671461.35 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3337, pruned_loss=0.09501, over 5698683.44 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:31:11,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.674e+02 1.261e+03 1.634e+03 2.034e+03 7.327e+03, threshold=3.269e+03, percent-clipped=12.0
+2023-03-11 15:31:37,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4435, 2.0109, 1.4057, 0.8137], device='cuda:0'), covar=tensor([0.6404, 0.2892, 0.3819, 0.6307], device='cuda:0'), in_proj_covar=tensor([0.1744, 0.1651, 0.1592, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 15:31:50,812 INFO [train.py:968] (0/2) Epoch 22, batch 37650, giga_loss[loss=0.3202, simple_loss=0.3829, pruned_loss=0.1287, over 27614.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3441, pruned_loss=0.1013, over 5693995.58 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3611, pruned_loss=0.1148, over 5676060.92 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3396, pruned_loss=0.09845, over 5691562.66 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:31:57,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3544, 1.5612, 1.4223, 1.2852], device='cuda:0'), covar=tensor([0.2830, 0.2363, 0.2436, 0.2636], device='cuda:0'), in_proj_covar=tensor([0.1961, 0.1893, 0.1830, 0.1977], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 15:32:01,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8724, 2.8244, 1.7306, 0.9943], device='cuda:0'), covar=tensor([0.8335, 0.4073, 0.4431, 0.7603], device='cuda:0'), in_proj_covar=tensor([0.1745, 0.1652, 0.1592, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 15:32:43,051 INFO [train.py:968] (0/2) Epoch 22, batch 37700, giga_loss[loss=0.3319, simple_loss=0.3942, pruned_loss=0.1348, over 27634.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3472, pruned_loss=0.1018, over 5686420.26 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.361, pruned_loss=0.1147, over 5676999.77 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3434, pruned_loss=0.09937, over 5684406.53 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:32:51,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.452e+02 1.202e+03 1.528e+03 2.054e+03 6.847e+03, threshold=3.057e+03, percent-clipped=6.0
+2023-03-11 15:33:27,201 INFO [train.py:968] (0/2) Epoch 22, batch 37750, giga_loss[loss=0.2572, simple_loss=0.3372, pruned_loss=0.08862, over 28702.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3534, pruned_loss=0.1051, over 5688709.56 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3615, pruned_loss=0.1152, over 5683111.50 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3492, pruned_loss=0.1021, over 5681975.26 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:34:12,491 INFO [train.py:968] (0/2) Epoch 22, batch 37800, giga_loss[loss=0.289, simple_loss=0.3585, pruned_loss=0.1098, over 28707.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3579, pruned_loss=0.1078, over 5694120.18 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3613, pruned_loss=0.1151, over 5687906.99 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3547, pruned_loss=0.1053, over 5684469.48 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:34:20,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.029e+02 1.369e+03 1.640e+03 2.131e+03 4.666e+03, threshold=3.279e+03, percent-clipped=8.0
+2023-03-11 15:34:57,287 INFO [train.py:968] (0/2) Epoch 22, batch 37850, giga_loss[loss=0.2579, simple_loss=0.3524, pruned_loss=0.08163, over 28861.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.354, pruned_loss=0.1047, over 5684433.09 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3614, pruned_loss=0.1152, over 5681055.81 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3514, pruned_loss=0.1026, over 5682475.15 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:35:07,727 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 15:35:08,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-11 15:35:37,879 INFO [train.py:968] (0/2) Epoch 22, batch 37900, giga_loss[loss=0.2909, simple_loss=0.3586, pruned_loss=0.1116, over 28915.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3516, pruned_loss=0.1023, over 5700459.15 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3614, pruned_loss=0.1152, over 5687909.63 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3492, pruned_loss=0.1002, over 5693131.48 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:35:50,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.232e+02 1.201e+03 1.478e+03 1.823e+03 5.269e+03, threshold=2.956e+03, percent-clipped=8.0
+2023-03-11 15:36:26,260 INFO [train.py:968] (0/2) Epoch 22, batch 37950, giga_loss[loss=0.2651, simple_loss=0.3448, pruned_loss=0.09269, over 28994.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3508, pruned_loss=0.1014, over 5703918.89 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3612, pruned_loss=0.1151, over 5694999.74 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3487, pruned_loss=0.09947, over 5692052.53 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:36:38,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=995696.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:37:07,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4295, 1.6286, 1.5008, 1.5105], device='cuda:0'), covar=tensor([0.1906, 0.2171, 0.2329, 0.1985], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0750, 0.0716, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 15:37:08,925 INFO [train.py:968] (0/2) Epoch 22, batch 38000, giga_loss[loss=0.2632, simple_loss=0.3359, pruned_loss=0.09522, over 28838.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3502, pruned_loss=0.1007, over 5695358.08 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3612, pruned_loss=0.1152, over 5688406.79 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3484, pruned_loss=0.09879, over 5692995.97 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:37:11,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=995734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:37:17,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2161, 0.8096, 0.8690, 1.3442], device='cuda:0'), covar=tensor([0.0778, 0.0437, 0.0367, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 15:37:18,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.633e+02 1.257e+03 1.484e+03 2.094e+03 5.522e+03, threshold=2.968e+03, percent-clipped=7.0
+2023-03-11 15:37:42,093 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=995768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:37:43,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2717, 1.7539, 1.3964, 0.4941], device='cuda:0'), covar=tensor([0.4370, 0.2763, 0.3837, 0.5900], device='cuda:0'), in_proj_covar=tensor([0.1748, 0.1652, 0.1599, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 15:37:54,145 INFO [train.py:968] (0/2) Epoch 22, batch 38050, giga_loss[loss=0.2818, simple_loss=0.3598, pruned_loss=0.1019, over 29083.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3537, pruned_loss=0.1028, over 5702602.53 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3612, pruned_loss=0.115, over 5692671.67 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.352, pruned_loss=0.1011, over 5697125.80 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:38:41,590 INFO [train.py:968] (0/2) Epoch 22, batch 38100, giga_loss[loss=0.2824, simple_loss=0.3542, pruned_loss=0.1053, over 28693.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3558, pruned_loss=0.1044, over 5702720.37 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3614, pruned_loss=0.1151, over 5696005.89 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3541, pruned_loss=0.1028, over 5695628.65 frames. ], batch size: 92, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:38:51,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.471e+02 1.250e+03 1.518e+03 2.048e+03 6.104e+03, threshold=3.035e+03, percent-clipped=8.0
+2023-03-11 15:39:15,116 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 15:39:23,126 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-11 15:39:25,369 INFO [train.py:968] (0/2) Epoch 22, batch 38150, giga_loss[loss=0.2887, simple_loss=0.3626, pruned_loss=0.1073, over 28911.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3566, pruned_loss=0.1056, over 5695475.23 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3606, pruned_loss=0.1145, over 5692167.02 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3559, pruned_loss=0.1046, over 5692655.41 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:40:14,796 INFO [train.py:968] (0/2) Epoch 22, batch 38200, giga_loss[loss=0.2538, simple_loss=0.3377, pruned_loss=0.08493, over 28829.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3564, pruned_loss=0.1057, over 5695200.43 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3608, pruned_loss=0.1144, over 5691042.89 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3557, pruned_loss=0.1048, over 5694102.68 frames. ], batch size: 186, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:40:24,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.590e+02 1.385e+03 1.906e+03 2.514e+03 1.478e+04, threshold=3.812e+03, percent-clipped=20.0
+2023-03-11 15:41:01,020 INFO [train.py:968] (0/2) Epoch 22, batch 38250, giga_loss[loss=0.2824, simple_loss=0.3662, pruned_loss=0.09926, over 29091.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3561, pruned_loss=0.1053, over 5692087.19 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3607, pruned_loss=0.1143, over 5692854.86 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3555, pruned_loss=0.1047, over 5689762.60 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:41:16,388 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-996000.pt
+2023-03-11 15:41:42,773 INFO [train.py:968] (0/2) Epoch 22, batch 38300, giga_loss[loss=0.266, simple_loss=0.3443, pruned_loss=0.09379, over 28562.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3573, pruned_loss=0.1055, over 5693048.81 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3611, pruned_loss=0.1145, over 5688126.67 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3565, pruned_loss=0.1046, over 5695061.73 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:41:53,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.439e+02 1.154e+03 1.425e+03 1.693e+03 4.463e+03, threshold=2.850e+03, percent-clipped=2.0
+2023-03-11 15:42:19,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=996071.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:42:28,726 INFO [train.py:968] (0/2) Epoch 22, batch 38350, giga_loss[loss=0.2723, simple_loss=0.3617, pruned_loss=0.09151, over 28885.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3575, pruned_loss=0.1045, over 5699329.03 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3613, pruned_loss=0.1148, over 5690554.79 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3566, pruned_loss=0.1035, over 5698920.35 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:42:52,932 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=996109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:42:56,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8084, 1.9554, 1.6708, 1.9500], device='cuda:0'), covar=tensor([0.2567, 0.2651, 0.2945, 0.2461], device='cuda:0'), in_proj_covar=tensor([0.1514, 0.1097, 0.1335, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 15:43:11,094 INFO [train.py:968] (0/2) Epoch 22, batch 38400, giga_loss[loss=0.2546, simple_loss=0.3409, pruned_loss=0.08417, over 28525.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3566, pruned_loss=0.1037, over 5703529.87 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3612, pruned_loss=0.1147, over 5692694.20 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3559, pruned_loss=0.1028, over 5701423.19 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:43:17,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5462, 1.6998, 1.2390, 1.3735], device='cuda:0'), covar=tensor([0.0831, 0.0442, 0.1021, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0443, 0.0522, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 15:43:21,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5051, 1.8377, 1.4830, 1.4152], device='cuda:0'), covar=tensor([0.2499, 0.2454, 0.2826, 0.2214], device='cuda:0'), in_proj_covar=tensor([0.1512, 0.1096, 0.1333, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 15:43:22,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.878e+02 1.156e+03 1.466e+03 2.123e+03 4.710e+03, threshold=2.933e+03, percent-clipped=9.0
+2023-03-11 15:43:23,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=996143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:43:31,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0016, 2.1610, 1.5032, 1.7490], device='cuda:0'), covar=tensor([0.0998, 0.0695, 0.1082, 0.1165], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0443, 0.0522, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 15:43:53,533 INFO [train.py:968] (0/2) Epoch 22, batch 38450, giga_loss[loss=0.2769, simple_loss=0.3475, pruned_loss=0.1032, over 29021.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3552, pruned_loss=0.1035, over 5695032.26 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3617, pruned_loss=0.115, over 5686115.85 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.354, pruned_loss=0.1022, over 5699884.40 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:44:17,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5361, 1.6680, 1.6585, 1.3721], device='cuda:0'), covar=tensor([0.2713, 0.2631, 0.1920, 0.2655], device='cuda:0'), in_proj_covar=tensor([0.1953, 0.1888, 0.1818, 0.1974], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 15:44:21,227 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=996214.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:44:24,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=996217.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:44:36,610 INFO [train.py:968] (0/2) Epoch 22, batch 38500, giga_loss[loss=0.2399, simple_loss=0.3288, pruned_loss=0.07554, over 28967.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3525, pruned_loss=0.1018, over 5698176.08 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3616, pruned_loss=0.1149, over 5689356.80 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3516, pruned_loss=0.1008, over 5699275.75 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:44:46,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.248e+02 1.202e+03 1.470e+03 2.097e+03 6.231e+03, threshold=2.941e+03, percent-clipped=11.0
+2023-03-11 15:44:48,709 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=996246.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:44:49,325 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=996247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:44:53,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=996252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:44:55,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=996255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:45:15,960 INFO [train.py:968] (0/2) Epoch 22, batch 38550, giga_loss[loss=0.3323, simple_loss=0.3857, pruned_loss=0.1395, over 27635.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3507, pruned_loss=0.1007, over 5707684.12 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1144, over 5696190.79 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.35, pruned_loss=0.09986, over 5702979.45 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:45:19,657 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=996284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:45:20,978 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=996286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:45:22,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=996289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:45:51,092 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=996318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:45:57,394 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.37 vs. limit=5.0
+2023-03-11 15:46:00,874 INFO [train.py:968] (0/2) Epoch 22, batch 38600, giga_loss[loss=0.278, simple_loss=0.3555, pruned_loss=0.1002, over 28819.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3511, pruned_loss=0.1013, over 5696809.18 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1143, over 5687916.66 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3505, pruned_loss=0.1005, over 5701102.54 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:46:12,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.340e+02 1.095e+03 1.386e+03 1.890e+03 3.192e+03, threshold=2.772e+03, percent-clipped=3.0
+2023-03-11 15:46:43,291 INFO [train.py:968] (0/2) Epoch 22, batch 38650, giga_loss[loss=0.2638, simple_loss=0.3476, pruned_loss=0.09003, over 28717.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.351, pruned_loss=0.1012, over 5704946.25 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3616, pruned_loss=0.1145, over 5693435.29 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3497, pruned_loss=0.09989, over 5703431.32 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:46:49,434 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-11 15:47:22,884 INFO [train.py:968] (0/2) Epoch 22, batch 38700, giga_loss[loss=0.2689, simple_loss=0.3471, pruned_loss=0.09537, over 28523.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3507, pruned_loss=0.1004, over 5698558.46 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3615, pruned_loss=0.1145, over 5685459.51 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3495, pruned_loss=0.09916, over 5704002.40 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:47:33,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.965e+02 1.125e+03 1.475e+03 2.139e+03 1.088e+04, threshold=2.950e+03, percent-clipped=12.0
+2023-03-11 15:47:36,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5120, 1.9269, 1.5497, 1.6685], device='cuda:0'), covar=tensor([0.0804, 0.0282, 0.0333, 0.0912], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0117, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 15:47:36,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6296, 1.7509, 1.6711, 1.4423], device='cuda:0'), covar=tensor([0.2964, 0.2572, 0.2037, 0.2710], device='cuda:0'), in_proj_covar=tensor([0.1966, 0.1900, 0.1832, 0.1987], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 15:47:44,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 15:48:01,925 INFO [train.py:968] (0/2) Epoch 22, batch 38750, giga_loss[loss=0.3116, simple_loss=0.3729, pruned_loss=0.1252, over 27509.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3494, pruned_loss=0.09911, over 5706706.40 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1144, over 5686336.85 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3484, pruned_loss=0.09782, over 5710528.48 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:48:13,031 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=996495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:48:42,277 INFO [train.py:968] (0/2) Epoch 22, batch 38800, giga_loss[loss=0.224, simple_loss=0.3143, pruned_loss=0.0668, over 28546.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3488, pruned_loss=0.09949, over 5688824.95 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3608, pruned_loss=0.1143, over 5675657.64 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3479, pruned_loss=0.09775, over 5702600.24 frames. ], batch size: 60, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:48:54,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.562e+02 1.114e+03 1.404e+03 2.106e+03 6.431e+03, threshold=2.807e+03, percent-clipped=13.0
+2023-03-11 15:49:18,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2522, 2.5401, 1.3021, 1.3835], device='cuda:0'), covar=tensor([0.1050, 0.0320, 0.0947, 0.1441], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0548, 0.0384, 0.0423], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 15:49:23,933 INFO [train.py:968] (0/2) Epoch 22, batch 38850, giga_loss[loss=0.2544, simple_loss=0.3283, pruned_loss=0.0902, over 29025.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.348, pruned_loss=0.09976, over 5683352.14 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3615, pruned_loss=0.115, over 5665640.91 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3464, pruned_loss=0.09751, over 5703342.20 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:49:25,896 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-11 15:49:57,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=996622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:50:03,646 INFO [train.py:968] (0/2) Epoch 22, batch 38900, giga_loss[loss=0.2179, simple_loss=0.303, pruned_loss=0.06645, over 28769.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3448, pruned_loss=0.09804, over 5679700.33 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3619, pruned_loss=0.1152, over 5656278.99 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3429, pruned_loss=0.09575, over 5704009.45 frames. ], batch size: 60, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:50:07,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8475, 0.9094, 0.8213, 0.8114], device='cuda:0'), covar=tensor([0.1784, 0.2346, 0.1540, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.1962, 0.1903, 0.1828, 0.1983], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 15:50:14,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.565e+02 1.201e+03 1.589e+03 2.096e+03 5.042e+03, threshold=3.178e+03, percent-clipped=11.0
+2023-03-11 15:50:32,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5049, 1.7539, 1.4007, 1.7676], device='cuda:0'), covar=tensor([0.2742, 0.2767, 0.2989, 0.2577], device='cuda:0'), in_proj_covar=tensor([0.1515, 0.1098, 0.1339, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 15:50:42,064 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-11 15:50:43,284 INFO [train.py:968] (0/2) Epoch 22, batch 38950, giga_loss[loss=0.2357, simple_loss=0.3154, pruned_loss=0.07796, over 28703.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3421, pruned_loss=0.09668, over 5693812.03 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3611, pruned_loss=0.1148, over 5666094.36 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3405, pruned_loss=0.09457, over 5705575.61 frames. ], batch size: 78, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:50:45,134 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-11 15:51:08,604 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-11 15:51:29,015 INFO [train.py:968] (0/2) Epoch 22, batch 39000, giga_loss[loss=0.2573, simple_loss=0.3351, pruned_loss=0.08977, over 28698.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3411, pruned_loss=0.09637, over 5697100.17 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3612, pruned_loss=0.1147, over 5669845.16 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3396, pruned_loss=0.09455, over 5703420.83 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:51:29,021 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 15:51:33,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2516, 1.8326, 1.4312, 0.4547], device='cuda:0'), covar=tensor([0.4885, 0.3538, 0.4684, 0.6644], device='cuda:0'), in_proj_covar=tensor([0.1740, 0.1636, 0.1583, 0.1421], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 15:51:37,978 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2064, simple_loss=0.3137, pruned_loss=0.04954, over 944034.00 frames. 
+2023-03-11 15:51:37,979 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 15:51:49,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.182e+02 1.178e+03 1.418e+03 1.982e+03 4.718e+03, threshold=2.836e+03, percent-clipped=6.0
+2023-03-11 15:52:03,265 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3897, 1.5306, 1.3037, 1.5461], device='cuda:0'), covar=tensor([0.0762, 0.0333, 0.0350, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 15:52:05,941 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=996765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:52:07,906 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=996768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:52:18,177 INFO [train.py:968] (0/2) Epoch 22, batch 39050, libri_loss[loss=0.4214, simple_loss=0.4468, pruned_loss=0.1979, over 19467.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3398, pruned_loss=0.09601, over 5697743.94 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3615, pruned_loss=0.1149, over 5667291.11 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3376, pruned_loss=0.09371, over 5706885.13 frames. ], batch size: 188, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:52:32,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=996797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:53:02,073 INFO [train.py:968] (0/2) Epoch 22, batch 39100, giga_loss[loss=0.2298, simple_loss=0.3097, pruned_loss=0.07495, over 28966.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3383, pruned_loss=0.09606, over 5701512.87 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3621, pruned_loss=0.1156, over 5672907.48 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3353, pruned_loss=0.09302, over 5704464.62 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:53:09,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4989, 4.3079, 4.1003, 2.0849], device='cuda:0'), covar=tensor([0.0554, 0.0729, 0.0761, 0.1959], device='cuda:0'), in_proj_covar=tensor([0.1216, 0.1130, 0.0951, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:0')
+2023-03-11 15:53:12,295 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.741e+02 1.237e+03 1.629e+03 2.837e+03 1.655e+04, threshold=3.259e+03, percent-clipped=25.0
+2023-03-11 15:53:35,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=996870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:53:44,867 INFO [train.py:968] (0/2) Epoch 22, batch 39150, giga_loss[loss=0.2318, simple_loss=0.3068, pruned_loss=0.07838, over 28855.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3345, pruned_loss=0.09405, over 5702052.65 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3621, pruned_loss=0.1156, over 5672907.48 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3322, pruned_loss=0.09168, over 5704350.04 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:53:51,034 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-11 15:54:13,897 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9717, 2.7158, 1.7523, 0.9863], device='cuda:0'), covar=tensor([0.6027, 0.3107, 0.3527, 0.6029], device='cuda:0'), in_proj_covar=tensor([0.1739, 0.1639, 0.1585, 0.1420], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 15:54:20,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 15:54:24,641 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=996927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:54:26,988 INFO [train.py:968] (0/2) Epoch 22, batch 39200, libri_loss[loss=0.3189, simple_loss=0.3829, pruned_loss=0.1275, over 29527.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3351, pruned_loss=0.09467, over 5705624.41 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.363, pruned_loss=0.1161, over 5677756.72 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3311, pruned_loss=0.09133, over 5704282.05 frames. ], batch size: 81, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:54:33,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6094, 1.8212, 1.4822, 1.6797], device='cuda:0'), covar=tensor([0.2713, 0.2876, 0.3290, 0.2565], device='cuda:0'), in_proj_covar=tensor([0.1516, 0.1098, 0.1337, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 15:54:36,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.605e+02 1.092e+03 1.385e+03 2.078e+03 5.604e+03, threshold=2.770e+03, percent-clipped=7.0
+2023-03-11 15:55:09,562 INFO [train.py:968] (0/2) Epoch 22, batch 39250, giga_loss[loss=0.3125, simple_loss=0.3829, pruned_loss=0.121, over 28611.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3359, pruned_loss=0.09565, over 5711760.74 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3628, pruned_loss=0.116, over 5682827.69 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3323, pruned_loss=0.09255, over 5707104.22 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:55:25,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6384, 1.7969, 1.5110, 2.0852], device='cuda:0'), covar=tensor([0.2643, 0.2836, 0.3100, 0.2382], device='cuda:0'), in_proj_covar=tensor([0.1513, 0.1095, 0.1335, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 15:55:40,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=997013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:55:42,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=997016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:55:55,842 INFO [train.py:968] (0/2) Epoch 22, batch 39300, giga_loss[loss=0.2274, simple_loss=0.3061, pruned_loss=0.07434, over 28644.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3382, pruned_loss=0.09623, over 5718382.82 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3626, pruned_loss=0.1159, over 5690619.49 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3346, pruned_loss=0.09311, over 5708308.27 frames. ], batch size: 78, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:56:06,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.766e+02 1.146e+03 1.477e+03 2.006e+03 5.936e+03, threshold=2.953e+03, percent-clipped=15.0
+2023-03-11 15:56:07,082 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=997045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:56:37,611 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=997076.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:56:41,968 INFO [train.py:968] (0/2) Epoch 22, batch 39350, giga_loss[loss=0.2522, simple_loss=0.3364, pruned_loss=0.08401, over 28999.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3411, pruned_loss=0.0969, over 5714547.63 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3626, pruned_loss=0.116, over 5694465.98 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3377, pruned_loss=0.09398, over 5703523.62 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:56:43,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3316, 1.5301, 1.4191, 1.2194], device='cuda:0'), covar=tensor([0.3217, 0.2796, 0.2051, 0.2812], device='cuda:0'), in_proj_covar=tensor([0.1970, 0.1902, 0.1830, 0.1982], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 15:57:25,739 INFO [train.py:968] (0/2) Epoch 22, batch 39400, giga_loss[loss=0.2623, simple_loss=0.3469, pruned_loss=0.08881, over 27942.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3443, pruned_loss=0.09779, over 5711234.17 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3629, pruned_loss=0.116, over 5695305.54 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3408, pruned_loss=0.09494, over 5702103.30 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:57:39,170 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.244e+02 1.171e+03 1.658e+03 2.262e+03 8.283e+03, threshold=3.317e+03, percent-clipped=13.0
+2023-03-11 15:57:42,165 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3395, 1.5527, 1.5396, 1.2881], device='cuda:0'), covar=tensor([0.3408, 0.2725, 0.2147, 0.2852], device='cuda:0'), in_proj_covar=tensor([0.1968, 0.1901, 0.1831, 0.1981], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 15:57:58,860 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=997168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:58:08,184 INFO [train.py:968] (0/2) Epoch 22, batch 39450, giga_loss[loss=0.2518, simple_loss=0.3188, pruned_loss=0.09242, over 23765.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3446, pruned_loss=0.09757, over 5698191.70 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3623, pruned_loss=0.1155, over 5696312.05 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3414, pruned_loss=0.09478, over 5690072.57 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:58:53,106 INFO [train.py:968] (0/2) Epoch 22, batch 39500, giga_loss[loss=0.2144, simple_loss=0.3001, pruned_loss=0.06435, over 28885.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3426, pruned_loss=0.09546, over 5699746.78 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3623, pruned_loss=0.1155, over 5696312.05 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3401, pruned_loss=0.09328, over 5693427.56 frames. ], batch size: 186, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:59:04,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.934e+02 1.183e+03 1.376e+03 1.829e+03 6.357e+03, threshold=2.752e+03, percent-clipped=5.0
+2023-03-11 15:59:37,035 INFO [train.py:968] (0/2) Epoch 22, batch 39550, libri_loss[loss=0.2977, simple_loss=0.3755, pruned_loss=0.1099, over 29281.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3432, pruned_loss=0.09614, over 5699992.65 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3622, pruned_loss=0.1153, over 5699683.82 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3407, pruned_loss=0.09401, over 5691907.28 frames. ], batch size: 94, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:59:55,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=997302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:00:18,848 INFO [train.py:968] (0/2) Epoch 22, batch 39600, giga_loss[loss=0.3065, simple_loss=0.3778, pruned_loss=0.1176, over 28601.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3438, pruned_loss=0.09706, over 5697777.68 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3624, pruned_loss=0.1154, over 5702843.95 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3413, pruned_loss=0.09496, over 5688401.54 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 16:00:31,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.465e+02 1.249e+03 1.658e+03 2.247e+03 6.837e+03, threshold=3.315e+03, percent-clipped=12.0
+2023-03-11 16:01:02,878 INFO [train.py:968] (0/2) Epoch 22, batch 39650, giga_loss[loss=0.2943, simple_loss=0.3646, pruned_loss=0.112, over 29009.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3467, pruned_loss=0.09879, over 5703513.01 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3625, pruned_loss=0.1154, over 5705420.85 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3443, pruned_loss=0.09675, over 5693732.42 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:01:16,761 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-11 16:01:45,050 INFO [train.py:968] (0/2) Epoch 22, batch 39700, giga_loss[loss=0.2608, simple_loss=0.3366, pruned_loss=0.09253, over 28338.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3483, pruned_loss=0.09986, over 5711456.74 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3616, pruned_loss=0.1149, over 5712095.45 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3466, pruned_loss=0.09814, over 5697508.04 frames. ], batch size: 65, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:01:55,210 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=997445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:01:55,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.103e+02 1.377e+03 1.712e+03 2.492e+03 8.599e+03, threshold=3.425e+03, percent-clipped=16.0
+2023-03-11 16:01:57,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=997448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:02:00,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=997451.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:02:21,429 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=997477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:02:24,372 INFO [train.py:968] (0/2) Epoch 22, batch 39750, giga_loss[loss=0.2507, simple_loss=0.3279, pruned_loss=0.08676, over 28797.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1011, over 5713951.70 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3622, pruned_loss=0.1151, over 5706186.14 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3494, pruned_loss=0.09921, over 5708154.33 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:03:06,535 INFO [train.py:968] (0/2) Epoch 22, batch 39800, giga_loss[loss=0.2594, simple_loss=0.3328, pruned_loss=0.09302, over 28767.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3536, pruned_loss=0.1024, over 5707989.19 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1156, over 5701792.57 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3508, pruned_loss=0.1001, over 5708065.66 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:03:16,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=997543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:03:18,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.436e+02 1.292e+03 1.579e+03 1.904e+03 5.592e+03, threshold=3.158e+03, percent-clipped=5.0
+2023-03-11 16:03:32,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5741, 3.4785, 1.5658, 1.6155], device='cuda:0'), covar=tensor([0.0898, 0.0355, 0.0902, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0550, 0.0385, 0.0425], device='cuda:0'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 16:03:49,172 INFO [train.py:968] (0/2) Epoch 22, batch 39850, giga_loss[loss=0.2508, simple_loss=0.3226, pruned_loss=0.0895, over 28572.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3531, pruned_loss=0.1018, over 5712829.97 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3633, pruned_loss=0.1157, over 5706773.56 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3505, pruned_loss=0.09962, over 5708728.14 frames. ], batch size: 78, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:03:54,477 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 16:03:59,692 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=997594.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 16:04:01,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=997597.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:04:26,211 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=997626.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 16:04:29,314 INFO [train.py:968] (0/2) Epoch 22, batch 39900, giga_loss[loss=0.2444, simple_loss=0.3302, pruned_loss=0.07931, over 28864.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3524, pruned_loss=0.1013, over 5699080.59 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.364, pruned_loss=0.1162, over 5690115.51 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3492, pruned_loss=0.09868, over 5711359.08 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:04:29,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3424, 1.7727, 1.4930, 1.5709], device='cuda:0'), covar=tensor([0.0764, 0.0288, 0.0329, 0.0882], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 16:04:37,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4174, 1.4964, 1.2378, 1.4907], device='cuda:0'), covar=tensor([0.0769, 0.0325, 0.0361, 0.0913], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 16:04:42,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.700e+02 1.223e+03 1.570e+03 2.044e+03 4.921e+03, threshold=3.140e+03, percent-clipped=6.0
+2023-03-11 16:04:48,333 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-11 16:05:09,884 INFO [train.py:968] (0/2) Epoch 22, batch 39950, giga_loss[loss=0.2239, simple_loss=0.3002, pruned_loss=0.0738, over 28495.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3502, pruned_loss=0.1002, over 5704383.87 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3642, pruned_loss=0.1162, over 5691588.64 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3472, pruned_loss=0.09767, over 5713234.64 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:05:13,521 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=997686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:05:15,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=997689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:05:41,558 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=997718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:05:51,420 INFO [train.py:968] (0/2) Epoch 22, batch 40000, giga_loss[loss=0.242, simple_loss=0.3212, pruned_loss=0.08141, over 29041.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3473, pruned_loss=0.09926, over 5709105.46 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3642, pruned_loss=0.1164, over 5698334.36 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3445, pruned_loss=0.09658, over 5710351.69 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 16:05:59,802 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-11 16:06:04,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.230e+02 1.093e+03 1.319e+03 1.965e+03 8.046e+03, threshold=2.638e+03, percent-clipped=4.0
+2023-03-11 16:06:07,744 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=997751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:06:32,013 INFO [train.py:968] (0/2) Epoch 22, batch 40050, giga_loss[loss=0.225, simple_loss=0.3127, pruned_loss=0.06867, over 28993.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3444, pruned_loss=0.09725, over 5712031.46 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3646, pruned_loss=0.1163, over 5699814.71 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.341, pruned_loss=0.09445, over 5711930.43 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:06:37,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7515, 1.8295, 1.3618, 1.4129], device='cuda:0'), covar=tensor([0.0929, 0.0644, 0.1078, 0.1269], device='cuda:0'), in_proj_covar=tensor([0.0390, 0.0442, 0.0516, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 16:07:13,983 INFO [train.py:968] (0/2) Epoch 22, batch 40100, giga_loss[loss=0.2743, simple_loss=0.3534, pruned_loss=0.09758, over 28781.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3446, pruned_loss=0.09666, over 5710508.02 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3647, pruned_loss=0.1164, over 5699963.14 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3413, pruned_loss=0.09392, over 5710451.93 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:07:27,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.970e+02 1.188e+03 1.442e+03 1.872e+03 5.223e+03, threshold=2.883e+03, percent-clipped=11.0
+2023-03-11 16:07:56,987 INFO [train.py:968] (0/2) Epoch 22, batch 40150, giga_loss[loss=0.2659, simple_loss=0.335, pruned_loss=0.09841, over 28935.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3456, pruned_loss=0.09639, over 5698098.29 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3647, pruned_loss=0.1164, over 5693283.23 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3428, pruned_loss=0.09396, over 5703949.07 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:08:38,288 INFO [train.py:968] (0/2) Epoch 22, batch 40200, giga_loss[loss=0.2679, simple_loss=0.3419, pruned_loss=0.09693, over 29072.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3461, pruned_loss=0.09713, over 5704398.27 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3644, pruned_loss=0.1162, over 5697214.28 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3439, pruned_loss=0.09507, over 5705552.91 frames. ], batch size: 155, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:08:52,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.321e+02 1.238e+03 1.523e+03 2.009e+03 5.625e+03, threshold=3.047e+03, percent-clipped=5.0
+2023-03-11 16:09:20,708 INFO [train.py:968] (0/2) Epoch 22, batch 40250, giga_loss[loss=0.265, simple_loss=0.3324, pruned_loss=0.09882, over 29067.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.345, pruned_loss=0.09786, over 5711421.58 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3647, pruned_loss=0.1166, over 5700804.95 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3425, pruned_loss=0.0954, over 5709277.22 frames. ], batch size: 128, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:09:37,261 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-998000.pt
+2023-03-11 16:10:02,959 INFO [train.py:968] (0/2) Epoch 22, batch 40300, giga_loss[loss=0.2703, simple_loss=0.333, pruned_loss=0.1038, over 28960.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3427, pruned_loss=0.09782, over 5720328.11 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3652, pruned_loss=0.1169, over 5703642.88 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3399, pruned_loss=0.0953, over 5716210.46 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:10:15,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4089, 1.6235, 1.4078, 1.5812], device='cuda:0'), covar=tensor([0.0755, 0.0317, 0.0331, 0.0870], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 16:10:19,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.740e+02 1.162e+03 1.434e+03 2.080e+03 8.174e+03, threshold=2.868e+03, percent-clipped=10.0
+2023-03-11 16:10:45,561 INFO [train.py:968] (0/2) Epoch 22, batch 40350, giga_loss[loss=0.2377, simple_loss=0.3099, pruned_loss=0.08272, over 28843.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3408, pruned_loss=0.09776, over 5713684.32 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3654, pruned_loss=0.1168, over 5704610.82 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3379, pruned_loss=0.09537, over 5709799.05 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:10:54,429 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2416, 1.1291, 1.0644, 1.4652], device='cuda:0'), covar=tensor([0.0738, 0.0398, 0.0372, 0.0845], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 16:11:21,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=998126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:11:26,353 INFO [train.py:968] (0/2) Epoch 22, batch 40400, libri_loss[loss=0.2602, simple_loss=0.3313, pruned_loss=0.09457, over 29611.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3389, pruned_loss=0.09706, over 5702082.07 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3651, pruned_loss=0.1168, over 5699330.69 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3362, pruned_loss=0.09477, over 5703358.42 frames. ], batch size: 74, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:11:40,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.440e+02 1.281e+03 1.719e+03 2.378e+03 6.851e+03, threshold=3.437e+03, percent-clipped=15.0
+2023-03-11 16:12:02,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4245, 1.5423, 1.4463, 1.5890], device='cuda:0'), covar=tensor([0.0753, 0.0322, 0.0328, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 16:12:07,686 INFO [train.py:968] (0/2) Epoch 22, batch 40450, giga_loss[loss=0.2257, simple_loss=0.2927, pruned_loss=0.07937, over 28640.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3358, pruned_loss=0.09535, over 5707895.43 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3646, pruned_loss=0.1165, over 5704480.28 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3335, pruned_loss=0.09335, over 5704145.42 frames. ], batch size: 85, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:12:29,676 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2630, 1.3682, 3.3217, 2.9274], device='cuda:0'), covar=tensor([0.1461, 0.2492, 0.0492, 0.1118], device='cuda:0'), in_proj_covar=tensor([0.0758, 0.0646, 0.0961, 0.0908], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 16:12:33,451 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-11 16:12:38,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9396, 1.1309, 1.0276, 0.8632], device='cuda:0'), covar=tensor([0.2460, 0.2748, 0.1683, 0.2427], device='cuda:0'), in_proj_covar=tensor([0.1980, 0.1914, 0.1845, 0.1990], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 16:12:50,400 INFO [train.py:968] (0/2) Epoch 22, batch 40500, giga_loss[loss=0.2546, simple_loss=0.3248, pruned_loss=0.09221, over 28888.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3328, pruned_loss=0.09425, over 5704825.39 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3647, pruned_loss=0.1166, over 5702773.43 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3306, pruned_loss=0.09237, over 5703542.77 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:13:02,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.641e+02 1.214e+03 1.528e+03 2.158e+03 5.234e+03, threshold=3.057e+03, percent-clipped=5.0
+2023-03-11 16:13:06,482 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3284, 1.7086, 1.4687, 1.4090], device='cuda:0'), covar=tensor([0.0760, 0.0330, 0.0344, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 16:13:20,089 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=998269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:13:23,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=998272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:13:25,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5280, 3.6604, 1.5458, 1.6537], device='cuda:0'), covar=tensor([0.0950, 0.0346, 0.0967, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0553, 0.0387, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 16:13:28,989 INFO [train.py:968] (0/2) Epoch 22, batch 40550, giga_loss[loss=0.2513, simple_loss=0.3206, pruned_loss=0.09095, over 28819.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3292, pruned_loss=0.09236, over 5712354.53 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3644, pruned_loss=0.1165, over 5705982.74 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3266, pruned_loss=0.09021, over 5708714.79 frames. ], batch size: 199, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:13:37,500 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 16:13:44,747 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=998301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:14:09,945 INFO [train.py:968] (0/2) Epoch 22, batch 40600, giga_loss[loss=0.2226, simple_loss=0.294, pruned_loss=0.07563, over 28416.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3298, pruned_loss=0.09229, over 5703345.49 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3649, pruned_loss=0.1168, over 5690715.07 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3264, pruned_loss=0.08965, over 5713445.84 frames. ], batch size: 78, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:14:23,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.026e+02 1.162e+03 1.568e+03 2.109e+03 4.879e+03, threshold=3.137e+03, percent-clipped=10.0
+2023-03-11 16:14:51,898 INFO [train.py:968] (0/2) Epoch 22, batch 40650, giga_loss[loss=0.279, simple_loss=0.3488, pruned_loss=0.1046, over 28697.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3342, pruned_loss=0.09425, over 5700597.33 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3646, pruned_loss=0.1167, over 5691608.60 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3311, pruned_loss=0.09175, over 5707913.83 frames. ], batch size: 92, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:15:33,905 INFO [train.py:968] (0/2) Epoch 22, batch 40700, giga_loss[loss=0.2499, simple_loss=0.3318, pruned_loss=0.08399, over 28886.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3378, pruned_loss=0.096, over 5701475.57 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3645, pruned_loss=0.1166, over 5689215.51 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3342, pruned_loss=0.09317, over 5710795.73 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:15:40,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2816, 4.0757, 3.9051, 1.8257], device='cuda:0'), covar=tensor([0.0624, 0.0792, 0.0718, 0.2156], device='cuda:0'), in_proj_covar=tensor([0.1227, 0.1136, 0.0958, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 16:15:43,833 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=998444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:15:47,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.463e+02 1.313e+03 1.668e+03 2.295e+03 6.466e+03, threshold=3.336e+03, percent-clipped=12.0
+2023-03-11 16:16:14,461 INFO [train.py:968] (0/2) Epoch 22, batch 40750, giga_loss[loss=0.271, simple_loss=0.3597, pruned_loss=0.09119, over 29076.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3416, pruned_loss=0.09776, over 5700167.56 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3641, pruned_loss=0.1165, over 5692847.35 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3386, pruned_loss=0.09525, over 5704610.00 frames. ], batch size: 128, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:16:57,113 INFO [train.py:968] (0/2) Epoch 22, batch 40800, giga_loss[loss=0.2904, simple_loss=0.3533, pruned_loss=0.1138, over 28758.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3454, pruned_loss=0.09965, over 5696875.55 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.364, pruned_loss=0.1163, over 5689610.06 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3424, pruned_loss=0.09727, over 5703209.29 frames. ], batch size: 92, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:17:13,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.324e+02 1.175e+03 1.448e+03 1.901e+03 1.046e+04, threshold=2.896e+03, percent-clipped=6.0
+2023-03-11 16:17:41,556 INFO [train.py:968] (0/2) Epoch 22, batch 40850, libri_loss[loss=0.4015, simple_loss=0.4253, pruned_loss=0.1889, over 20085.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3478, pruned_loss=0.1009, over 5692978.13 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3638, pruned_loss=0.1162, over 5682648.72 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3453, pruned_loss=0.09879, over 5705417.28 frames. ], batch size: 187, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:18:31,399 INFO [train.py:968] (0/2) Epoch 22, batch 40900, giga_loss[loss=0.2777, simple_loss=0.3496, pruned_loss=0.1029, over 29005.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3527, pruned_loss=0.1055, over 5687114.67 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3636, pruned_loss=0.1159, over 5678372.17 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3503, pruned_loss=0.1035, over 5701103.42 frames. ], batch size: 136, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:18:47,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.268e+02 1.488e+03 1.914e+03 2.425e+03 4.577e+03, threshold=3.827e+03, percent-clipped=15.0
+2023-03-11 16:19:02,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3581, 1.4724, 1.1503, 1.0945], device='cuda:0'), covar=tensor([0.0928, 0.0530, 0.1037, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0448, 0.0520, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 16:19:22,274 INFO [train.py:968] (0/2) Epoch 22, batch 40950, giga_loss[loss=0.3263, simple_loss=0.3939, pruned_loss=0.1294, over 28614.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3587, pruned_loss=0.1098, over 5687274.58 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3637, pruned_loss=0.116, over 5679759.23 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3567, pruned_loss=0.1081, over 5697219.29 frames. ], batch size: 78, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:20:13,069 INFO [train.py:968] (0/2) Epoch 22, batch 41000, giga_loss[loss=0.3138, simple_loss=0.38, pruned_loss=0.1239, over 28932.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3663, pruned_loss=0.1154, over 5686508.57 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3636, pruned_loss=0.1159, over 5682090.87 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3648, pruned_loss=0.1142, over 5692265.92 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:20:28,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.109e+03 1.839e+03 2.373e+03 3.054e+03 5.992e+03, threshold=4.746e+03, percent-clipped=13.0
+2023-03-11 16:20:57,297 INFO [train.py:968] (0/2) Epoch 22, batch 41050, giga_loss[loss=0.3428, simple_loss=0.4054, pruned_loss=0.1401, over 28900.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3726, pruned_loss=0.1208, over 5681744.89 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3635, pruned_loss=0.1159, over 5674063.25 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3715, pruned_loss=0.1198, over 5692729.99 frames. ], batch size: 213, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:21:09,374 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-11 16:21:32,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=998819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:21:46,310 INFO [train.py:968] (0/2) Epoch 22, batch 41100, giga_loss[loss=0.3443, simple_loss=0.4039, pruned_loss=0.1423, over 28260.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3783, pruned_loss=0.1256, over 5683059.14 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1156, over 5677480.04 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3781, pruned_loss=0.1252, over 5688923.96 frames. ], batch size: 368, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:22:04,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.141e+03 1.788e+03 2.282e+03 3.180e+03 7.394e+03, threshold=4.564e+03, percent-clipped=7.0
+2023-03-11 16:22:09,744 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=998851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:22:39,223 INFO [train.py:968] (0/2) Epoch 22, batch 41150, giga_loss[loss=0.3084, simple_loss=0.3759, pruned_loss=0.1205, over 28913.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3819, pruned_loss=0.1295, over 5662126.26 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3633, pruned_loss=0.1158, over 5681563.28 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3818, pruned_loss=0.1293, over 5662854.31 frames. ], batch size: 119, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:23:34,062 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-11 16:23:34,272 INFO [train.py:968] (0/2) Epoch 22, batch 41200, giga_loss[loss=0.3175, simple_loss=0.3704, pruned_loss=0.1323, over 28752.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3838, pruned_loss=0.1322, over 5663267.77 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3633, pruned_loss=0.1158, over 5683569.05 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3843, pruned_loss=0.1324, over 5661960.95 frames. ], batch size: 92, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:23:53,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.131e+03 1.787e+03 2.299e+03 3.015e+03 8.047e+03, threshold=4.598e+03, percent-clipped=8.0
+2023-03-11 16:24:07,205 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=998962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:24:10,342 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=998965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:24:27,687 INFO [train.py:968] (0/2) Epoch 22, batch 41250, giga_loss[loss=0.3486, simple_loss=0.4022, pruned_loss=0.1475, over 28557.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3875, pruned_loss=0.1364, over 5646620.01 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3633, pruned_loss=0.1156, over 5687065.32 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3883, pruned_loss=0.1371, over 5642202.71 frames. ], batch size: 336, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:24:44,801 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=998994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:25:21,915 INFO [train.py:968] (0/2) Epoch 22, batch 41300, giga_loss[loss=0.3192, simple_loss=0.3808, pruned_loss=0.1288, over 28829.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3894, pruned_loss=0.1382, over 5638224.02 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3629, pruned_loss=0.1154, over 5681731.23 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3909, pruned_loss=0.1393, over 5638334.45 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:25:42,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.751e+03 2.399e+03 3.224e+03 7.848e+03, threshold=4.798e+03, percent-clipped=8.0
+2023-03-11 16:26:12,454 INFO [train.py:968] (0/2) Epoch 22, batch 41350, giga_loss[loss=0.4015, simple_loss=0.4338, pruned_loss=0.1846, over 27554.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3893, pruned_loss=0.1383, over 5631409.17 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3629, pruned_loss=0.1154, over 5687161.97 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3913, pruned_loss=0.1399, over 5625069.43 frames. ], batch size: 472, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:26:36,147 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 16:27:07,362 INFO [train.py:968] (0/2) Epoch 22, batch 41400, giga_loss[loss=0.3339, simple_loss=0.3954, pruned_loss=0.1362, over 28432.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3888, pruned_loss=0.1388, over 5633079.23 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.363, pruned_loss=0.1154, over 5690767.23 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3908, pruned_loss=0.1406, over 5623739.16 frames. ], batch size: 71, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 16:27:26,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.170e+03 1.886e+03 2.344e+03 3.144e+03 5.836e+03, threshold=4.688e+03, percent-clipped=7.0
+2023-03-11 16:27:58,661 INFO [train.py:968] (0/2) Epoch 22, batch 41450, giga_loss[loss=0.3081, simple_loss=0.3793, pruned_loss=0.1184, over 28901.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3869, pruned_loss=0.1374, over 5641013.25 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3629, pruned_loss=0.1153, over 5684373.57 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3893, pruned_loss=0.1395, over 5637824.58 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 16:28:43,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=999226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:28:50,982 INFO [train.py:968] (0/2) Epoch 22, batch 41500, giga_loss[loss=0.3031, simple_loss=0.3675, pruned_loss=0.1193, over 28941.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3862, pruned_loss=0.1357, over 5651963.03 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3626, pruned_loss=0.1152, over 5690526.57 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.389, pruned_loss=0.1381, over 5642970.50 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 16:29:11,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.200e+03 1.903e+03 2.186e+03 3.447e+03 8.628e+03, threshold=4.372e+03, percent-clipped=5.0
+2023-03-11 16:29:44,678 INFO [train.py:968] (0/2) Epoch 22, batch 41550, giga_loss[loss=0.2993, simple_loss=0.3727, pruned_loss=0.113, over 28938.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3878, pruned_loss=0.136, over 5654130.73 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3626, pruned_loss=0.1152, over 5682665.62 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3904, pruned_loss=0.1381, over 5652747.19 frames. ], batch size: 128, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 16:30:07,333 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:30:38,848 INFO [train.py:968] (0/2) Epoch 22, batch 41600, giga_loss[loss=0.4803, simple_loss=0.4943, pruned_loss=0.2331, over 24025.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3889, pruned_loss=0.1366, over 5640535.66 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3626, pruned_loss=0.1151, over 5686009.22 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3916, pruned_loss=0.1389, over 5635481.87 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:31:03,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.154e+03 1.720e+03 2.231e+03 2.889e+03 7.747e+03, threshold=4.462e+03, percent-clipped=3.0
+2023-03-11 16:31:24,233 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=999369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:31:26,194 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=999372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:31:27,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2792, 4.0885, 3.8705, 1.9366], device='cuda:0'), covar=tensor([0.0622, 0.0758, 0.0787, 0.1951], device='cuda:0'), in_proj_covar=tensor([0.1243, 0.1148, 0.0972, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 16:31:35,018 INFO [train.py:968] (0/2) Epoch 22, batch 41650, giga_loss[loss=0.2669, simple_loss=0.3542, pruned_loss=0.08978, over 28248.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3853, pruned_loss=0.1329, over 5642833.83 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3626, pruned_loss=0.1152, over 5687132.17 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3876, pruned_loss=0.1348, over 5637601.33 frames. ], batch size: 77, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:31:49,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5055, 1.5985, 1.7122, 1.3041], device='cuda:0'), covar=tensor([0.1852, 0.2728, 0.1574, 0.1832], device='cuda:0'), in_proj_covar=tensor([0.0895, 0.0700, 0.0942, 0.0842], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 16:31:54,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=999401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:32:21,707 INFO [train.py:968] (0/2) Epoch 22, batch 41700, giga_loss[loss=0.2869, simple_loss=0.364, pruned_loss=0.1049, over 28962.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3821, pruned_loss=0.1292, over 5657472.62 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3622, pruned_loss=0.1149, over 5692398.22 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3848, pruned_loss=0.1314, over 5647603.79 frames. ], batch size: 186, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:32:43,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.135e+03 1.645e+03 2.102e+03 3.345e+03 6.502e+03, threshold=4.203e+03, percent-clipped=7.0
+2023-03-11 16:33:16,037 INFO [train.py:968] (0/2) Epoch 22, batch 41750, giga_loss[loss=0.3017, simple_loss=0.3691, pruned_loss=0.1172, over 27960.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3782, pruned_loss=0.1255, over 5667201.13 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3619, pruned_loss=0.1146, over 5695300.32 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3808, pruned_loss=0.1276, over 5656426.94 frames. ], batch size: 412, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:33:46,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0250, 3.8540, 3.6787, 2.0317], device='cuda:0'), covar=tensor([0.0659, 0.0747, 0.0778, 0.2124], device='cuda:0'), in_proj_covar=tensor([0.1243, 0.1150, 0.0973, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 16:34:05,560 INFO [train.py:968] (0/2) Epoch 22, batch 41800, libri_loss[loss=0.305, simple_loss=0.3729, pruned_loss=0.1186, over 29658.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3736, pruned_loss=0.1221, over 5659725.85 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3615, pruned_loss=0.1143, over 5693457.04 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3765, pruned_loss=0.1244, over 5652231.73 frames. ], batch size: 91, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:34:23,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:34:27,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.304e+03 1.828e+03 2.433e+03 3.473e+03 9.514e+03, threshold=4.866e+03, percent-clipped=17.0
+2023-03-11 16:34:48,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3284, 1.4956, 1.4051, 1.2481], device='cuda:0'), covar=tensor([0.2513, 0.2297, 0.1879, 0.2310], device='cuda:0'), in_proj_covar=tensor([0.1976, 0.1914, 0.1844, 0.1980], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 16:34:56,299 INFO [train.py:968] (0/2) Epoch 22, batch 41850, giga_loss[loss=0.2831, simple_loss=0.3612, pruned_loss=0.1025, over 29045.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3724, pruned_loss=0.1219, over 5656075.05 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3611, pruned_loss=0.1141, over 5698893.69 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3754, pruned_loss=0.1242, over 5643810.46 frames. ], batch size: 164, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:35:09,474 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3087, 1.8695, 1.4488, 0.4813], device='cuda:0'), covar=tensor([0.4045, 0.3282, 0.3868, 0.5812], device='cuda:0'), in_proj_covar=tensor([0.1765, 0.1663, 0.1605, 0.1441], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 16:35:25,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2821, 1.7563, 1.3616, 0.5223], device='cuda:0'), covar=tensor([0.3911, 0.2460, 0.3491, 0.5882], device='cuda:0'), in_proj_covar=tensor([0.1764, 0.1661, 0.1604, 0.1440], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 16:35:43,074 INFO [train.py:968] (0/2) Epoch 22, batch 41900, giga_loss[loss=0.3234, simple_loss=0.3846, pruned_loss=0.1311, over 28307.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3736, pruned_loss=0.1226, over 5665023.96 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1143, over 5693445.52 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3763, pruned_loss=0.1244, over 5659595.27 frames. ], batch size: 368, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:36:03,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.662e+03 1.922e+03 2.879e+03 6.501e+03, threshold=3.844e+03, percent-clipped=2.0
+2023-03-11 16:36:22,591 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 16:36:30,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=999674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:36:38,187 INFO [train.py:968] (0/2) Epoch 22, batch 41950, giga_loss[loss=0.2738, simple_loss=0.3603, pruned_loss=0.09358, over 28936.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3729, pruned_loss=0.1217, over 5668046.70 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1144, over 5693842.66 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3751, pruned_loss=0.1232, over 5662650.17 frames. ], batch size: 136, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:37:11,620 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-11 16:37:36,263 INFO [train.py:968] (0/2) Epoch 22, batch 42000, giga_loss[loss=0.3901, simple_loss=0.4389, pruned_loss=0.1707, over 27585.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3734, pruned_loss=0.1202, over 5669909.91 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3614, pruned_loss=0.1144, over 5695797.67 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3752, pruned_loss=0.1214, over 5663545.85 frames. ], batch size: 472, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:37:36,266 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 16:37:44,529 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2037, simple_loss=0.3101, pruned_loss=0.04862, over 944034.00 frames. 
+2023-03-11 16:37:44,530 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 16:37:50,638 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1086, 1.0803, 4.0673, 3.3336], device='cuda:0'), covar=tensor([0.1929, 0.3065, 0.0494, 0.0989], device='cuda:0'), in_proj_covar=tensor([0.0766, 0.0656, 0.0973, 0.0923], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 16:38:04,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.715e+02 1.560e+03 2.092e+03 3.052e+03 1.054e+04, threshold=4.183e+03, percent-clipped=11.0
+2023-03-11 16:38:29,613 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:38:35,496 INFO [train.py:968] (0/2) Epoch 22, batch 42050, giga_loss[loss=0.2719, simple_loss=0.3582, pruned_loss=0.09277, over 28451.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3747, pruned_loss=0.1189, over 5676280.43 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3616, pruned_loss=0.1145, over 5698082.75 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3761, pruned_loss=0.1199, over 5668948.69 frames. ], batch size: 60, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:39:09,320 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=999817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:39:12,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=999820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:39:16,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6519, 1.8547, 1.5593, 1.7150], device='cuda:0'), covar=tensor([0.2051, 0.2139, 0.2125, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.1516, 0.1097, 0.1338, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 16:39:25,576 INFO [train.py:968] (0/2) Epoch 22, batch 42100, giga_loss[loss=0.2945, simple_loss=0.3659, pruned_loss=0.1116, over 28792.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3744, pruned_loss=0.1196, over 5677045.60 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3609, pruned_loss=0.1142, over 5703079.41 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3765, pruned_loss=0.1207, over 5666358.96 frames. ], batch size: 119, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:39:39,414 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 16:39:41,424 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:39:42,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=999849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:39:43,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.676e+03 2.336e+03 3.135e+03 1.332e+04, threshold=4.672e+03, percent-clipped=9.0
+2023-03-11 16:39:47,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1503, 3.0108, 2.8779, 1.4461], device='cuda:0'), covar=tensor([0.1127, 0.1114, 0.0949, 0.2564], device='cuda:0'), in_proj_covar=tensor([0.1247, 0.1155, 0.0975, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 16:39:53,621 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-11 16:40:12,445 INFO [train.py:968] (0/2) Epoch 22, batch 42150, giga_loss[loss=0.3301, simple_loss=0.384, pruned_loss=0.1381, over 27917.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3754, pruned_loss=0.1212, over 5663283.34 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3608, pruned_loss=0.1142, over 5689788.02 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3774, pruned_loss=0.1223, over 5666199.83 frames. ], batch size: 412, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:40:50,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=999922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:40:57,777 INFO [train.py:968] (0/2) Epoch 22, batch 42200, giga_loss[loss=0.2801, simple_loss=0.3523, pruned_loss=0.1039, over 28543.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3742, pruned_loss=0.1213, over 5676008.15 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3611, pruned_loss=0.1145, over 5693156.76 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3757, pruned_loss=0.1219, over 5674955.54 frames. ], batch size: 92, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:41:16,975 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.162e+03 1.819e+03 2.494e+03 3.418e+03 7.678e+03, threshold=4.988e+03, percent-clipped=11.0
+2023-03-11 16:41:27,163 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:41:46,893 INFO [train.py:968] (0/2) Epoch 22, batch 42250, libri_loss[loss=0.2968, simple_loss=0.3538, pruned_loss=0.1199, over 29357.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3725, pruned_loss=0.1216, over 5663793.82 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3611, pruned_loss=0.1145, over 5698354.34 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3741, pruned_loss=0.1223, over 5657745.23 frames. ], batch size: 67, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:42:04,513 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1000000.pt
+2023-03-11 16:42:08,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1945, 2.4884, 1.2476, 1.4019], device='cuda:0'), covar=tensor([0.1032, 0.0409, 0.0921, 0.1357], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0558, 0.0388, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 16:42:40,601 INFO [train.py:968] (0/2) Epoch 22, batch 42300, giga_loss[loss=0.2737, simple_loss=0.356, pruned_loss=0.09574, over 28567.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3716, pruned_loss=0.1208, over 5662977.20 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3612, pruned_loss=0.1146, over 5689156.57 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3728, pruned_loss=0.1213, over 5665343.26 frames. ], batch size: 85, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:42:50,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000041.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 16:43:00,921 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.201e+02 1.569e+03 2.169e+03 2.994e+03 6.457e+03, threshold=4.338e+03, percent-clipped=5.0
+2023-03-11 16:43:01,971 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4067, 1.5931, 1.5358, 1.3628], device='cuda:0'), covar=tensor([0.2987, 0.2704, 0.2115, 0.2471], device='cuda:0'), in_proj_covar=tensor([0.1978, 0.1918, 0.1849, 0.1987], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 16:43:13,689 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:43:16,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:43:29,698 INFO [train.py:968] (0/2) Epoch 22, batch 42350, libri_loss[loss=0.3764, simple_loss=0.4234, pruned_loss=0.1647, over 29186.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3694, pruned_loss=0.1175, over 5680818.14 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3611, pruned_loss=0.1146, over 5695359.89 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3706, pruned_loss=0.118, over 5676731.66 frames. ], batch size: 97, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:43:34,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2646, 2.5634, 1.2271, 1.4411], device='cuda:0'), covar=tensor([0.1005, 0.0455, 0.0924, 0.1370], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0559, 0.0388, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 16:43:47,656 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:43:50,812 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3621, 1.6264, 1.1397, 1.1427], device='cuda:0'), covar=tensor([0.1081, 0.0563, 0.1228, 0.1141], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0448, 0.0519, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 16:44:02,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6702, 1.7158, 1.9064, 1.4592], device='cuda:0'), covar=tensor([0.1538, 0.2403, 0.1262, 0.1552], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0703, 0.0945, 0.0844], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 16:44:17,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5483, 1.7720, 1.3034, 1.3421], device='cuda:0'), covar=tensor([0.1018, 0.0606, 0.1087, 0.1125], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0448, 0.0519, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 16:44:22,277 INFO [train.py:968] (0/2) Epoch 22, batch 42400, giga_loss[loss=0.3059, simple_loss=0.3705, pruned_loss=0.1207, over 28300.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3707, pruned_loss=0.118, over 5680986.69 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.361, pruned_loss=0.1145, over 5697507.92 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3719, pruned_loss=0.1185, over 5675500.59 frames. ], batch size: 368, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:44:37,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:44:40,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.735e+02 1.596e+03 1.938e+03 2.654e+03 5.947e+03, threshold=3.877e+03, percent-clipped=3.0
+2023-03-11 16:44:53,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5242, 1.8329, 1.6336, 1.6514], device='cuda:0'), covar=tensor([0.1937, 0.1930, 0.2478, 0.2074], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0750, 0.0718, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 16:45:10,325 INFO [train.py:968] (0/2) Epoch 22, batch 42450, giga_loss[loss=0.2732, simple_loss=0.3543, pruned_loss=0.09607, over 28861.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3687, pruned_loss=0.1166, over 5687666.63 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3605, pruned_loss=0.114, over 5703962.74 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3703, pruned_loss=0.1175, over 5677371.15 frames. ], batch size: 112, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:45:14,009 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000184.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 16:45:15,514 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:45:16,133 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000187.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:45:29,048 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.09 vs. limit=2.0
+2023-03-11 16:45:40,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000216.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:45:46,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:45:55,172 INFO [train.py:968] (0/2) Epoch 22, batch 42500, giga_loss[loss=0.2476, simple_loss=0.3307, pruned_loss=0.08226, over 28947.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.368, pruned_loss=0.1168, over 5679164.97 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3607, pruned_loss=0.114, over 5698185.39 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3693, pruned_loss=0.1176, over 5675426.25 frames. ], batch size: 145, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:45:56,685 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:46:17,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.246e+02 1.571e+03 2.121e+03 3.056e+03 6.704e+03, threshold=4.242e+03, percent-clipped=11.0
+2023-03-11 16:46:41,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2916, 1.5555, 1.5394, 1.3489], device='cuda:0'), covar=tensor([0.1985, 0.1977, 0.2408, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0751, 0.0717, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 16:46:46,703 INFO [train.py:968] (0/2) Epoch 22, batch 42550, giga_loss[loss=0.4234, simple_loss=0.4391, pruned_loss=0.2039, over 26643.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3681, pruned_loss=0.118, over 5670795.85 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3607, pruned_loss=0.114, over 5699067.46 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3692, pruned_loss=0.1187, over 5666894.53 frames. ], batch size: 555, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:46:57,701 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:46:59,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:47:07,378 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-11 16:47:27,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:47:35,008 INFO [train.py:968] (0/2) Epoch 22, batch 42600, giga_loss[loss=0.2992, simple_loss=0.3685, pruned_loss=0.115, over 28721.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3679, pruned_loss=0.119, over 5668015.95 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3609, pruned_loss=0.1142, over 5700131.82 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3687, pruned_loss=0.1195, over 5663244.56 frames. ], batch size: 262, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:47:40,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:47:54,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 2.155e+03 2.972e+03 3.703e+03 8.270e+03, threshold=5.944e+03, percent-clipped=19.0
+2023-03-11 16:48:06,132 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:48:08,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:48:22,746 INFO [train.py:968] (0/2) Epoch 22, batch 42650, giga_loss[loss=0.3105, simple_loss=0.3722, pruned_loss=0.1244, over 28821.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3676, pruned_loss=0.1192, over 5672657.06 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5695765.50 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.368, pruned_loss=0.1193, over 5671312.86 frames. ], batch size: 284, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:48:29,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 16:48:41,798 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:49:15,201 INFO [train.py:968] (0/2) Epoch 22, batch 42700, giga_loss[loss=0.3564, simple_loss=0.4073, pruned_loss=0.1528, over 28609.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3666, pruned_loss=0.1188, over 5680384.48 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3611, pruned_loss=0.1143, over 5700974.00 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3675, pruned_loss=0.1193, over 5673941.28 frames. ], batch size: 336, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:49:16,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4073, 1.5993, 1.4982, 1.4210], device='cuda:0'), covar=tensor([0.1741, 0.2145, 0.2239, 0.2123], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0752, 0.0719, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 16:49:34,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.738e+03 2.750e+03 4.178e+03 1.059e+04, threshold=5.501e+03, percent-clipped=7.0
+2023-03-11 16:49:59,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:50:00,332 INFO [train.py:968] (0/2) Epoch 22, batch 42750, giga_loss[loss=0.2946, simple_loss=0.368, pruned_loss=0.1106, over 28917.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3668, pruned_loss=0.1191, over 5682816.27 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3604, pruned_loss=0.1141, over 5700173.51 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3684, pruned_loss=0.12, over 5677538.10 frames. ], batch size: 145, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 16:50:02,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:50:27,295 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:50:47,000 INFO [train.py:968] (0/2) Epoch 22, batch 42800, giga_loss[loss=0.2539, simple_loss=0.3359, pruned_loss=0.08592, over 29005.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3668, pruned_loss=0.1181, over 5673408.29 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3605, pruned_loss=0.1141, over 5687758.39 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3682, pruned_loss=0.1188, over 5678677.52 frames. ], batch size: 136, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:51:09,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.743e+03 2.284e+03 3.380e+03 7.059e+03, threshold=4.569e+03, percent-clipped=6.0
+2023-03-11 16:51:15,786 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:51:17,343 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000562.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 16:51:33,324 INFO [train.py:968] (0/2) Epoch 22, batch 42850, giga_loss[loss=0.2585, simple_loss=0.3369, pruned_loss=0.09008, over 28301.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3668, pruned_loss=0.1172, over 5678040.48 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3605, pruned_loss=0.1141, over 5689957.66 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3681, pruned_loss=0.118, over 5679865.07 frames. ], batch size: 65, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:51:59,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:52:20,444 INFO [train.py:968] (0/2) Epoch 22, batch 42900, giga_loss[loss=0.3407, simple_loss=0.402, pruned_loss=0.1397, over 29012.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3669, pruned_loss=0.1168, over 5675921.14 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.36, pruned_loss=0.1138, over 5694649.12 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3686, pruned_loss=0.1177, over 5672765.37 frames. ], batch size: 155, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:52:39,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1980, 1.5482, 1.2191, 0.6996], device='cuda:0'), covar=tensor([0.3029, 0.2068, 0.2201, 0.4775], device='cuda:0'), in_proj_covar=tensor([0.1768, 0.1667, 0.1604, 0.1437], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 16:52:44,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.671e+03 2.023e+03 3.108e+03 6.250e+03, threshold=4.045e+03, percent-clipped=5.0
+2023-03-11 16:52:50,514 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:53:13,166 INFO [train.py:968] (0/2) Epoch 22, batch 42950, giga_loss[loss=0.3385, simple_loss=0.3992, pruned_loss=0.139, over 28926.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3689, pruned_loss=0.1191, over 5664802.76 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3599, pruned_loss=0.1136, over 5693573.24 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3705, pruned_loss=0.1201, over 5662990.77 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:53:34,642 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:53:36,732 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:54:01,153 INFO [train.py:968] (0/2) Epoch 22, batch 43000, giga_loss[loss=0.3404, simple_loss=0.3995, pruned_loss=0.1406, over 28953.00 frames. ], tot_loss[loss=0.308, simple_loss=0.372, pruned_loss=0.122, over 5658723.90 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.36, pruned_loss=0.1137, over 5688558.59 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3735, pruned_loss=0.1229, over 5661181.02 frames. ], batch size: 186, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:54:05,981 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:54:22,986 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:54:25,802 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.207e+03 1.910e+03 2.354e+03 3.274e+03 8.343e+03, threshold=4.708e+03, percent-clipped=17.0
+2023-03-11 16:54:26,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:54:36,551 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.66 vs. limit=5.0
+2023-03-11 16:54:53,974 INFO [train.py:968] (0/2) Epoch 22, batch 43050, giga_loss[loss=0.2841, simple_loss=0.3536, pruned_loss=0.1074, over 29043.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3726, pruned_loss=0.1239, over 5655185.14 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3596, pruned_loss=0.1135, over 5691438.92 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3746, pruned_loss=0.1251, over 5653575.85 frames. ], batch size: 128, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:54:57,063 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:55:08,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6528, 1.6898, 1.8400, 1.4272], device='cuda:0'), covar=tensor([0.1688, 0.2369, 0.1430, 0.1668], device='cuda:0'), in_proj_covar=tensor([0.0898, 0.0703, 0.0945, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 16:55:44,661 INFO [train.py:968] (0/2) Epoch 22, batch 43100, giga_loss[loss=0.3095, simple_loss=0.3628, pruned_loss=0.1281, over 28972.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.373, pruned_loss=0.1251, over 5660418.94 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3596, pruned_loss=0.1135, over 5696128.70 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3749, pruned_loss=0.1264, over 5653758.01 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:56:07,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.255e+03 1.851e+03 2.406e+03 3.452e+03 7.655e+03, threshold=4.811e+03, percent-clipped=8.0
+2023-03-11 16:56:34,930 INFO [train.py:968] (0/2) Epoch 22, batch 43150, giga_loss[loss=0.2595, simple_loss=0.3332, pruned_loss=0.09289, over 28778.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3744, pruned_loss=0.1264, over 5662313.18 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3598, pruned_loss=0.1136, over 5696044.27 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.376, pruned_loss=0.1275, over 5656690.51 frames. ], batch size: 99, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:56:44,004 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 16:57:20,047 INFO [train.py:968] (0/2) Epoch 22, batch 43200, giga_loss[loss=0.295, simple_loss=0.363, pruned_loss=0.1135, over 28750.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3716, pruned_loss=0.1242, over 5666456.34 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3598, pruned_loss=0.1135, over 5693836.92 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3734, pruned_loss=0.1256, over 5663271.63 frames. ], batch size: 242, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:57:26,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000937.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:57:37,265 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4294, 1.8467, 1.3130, 0.9819], device='cuda:0'), covar=tensor([0.6713, 0.4758, 0.2987, 0.6122], device='cuda:0'), in_proj_covar=tensor([0.1772, 0.1675, 0.1610, 0.1443], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 16:57:37,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8394, 2.1628, 1.6723, 2.0217], device='cuda:0'), covar=tensor([0.2603, 0.2671, 0.3134, 0.2603], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1100, 0.1340, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 16:57:41,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.254e+03 1.810e+03 2.269e+03 3.209e+03 7.686e+03, threshold=4.537e+03, percent-clipped=7.0
+2023-03-11 16:57:49,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5725, 1.6763, 1.7432, 1.3308], device='cuda:0'), covar=tensor([0.2042, 0.2758, 0.1749, 0.1990], device='cuda:0'), in_proj_covar=tensor([0.0899, 0.0704, 0.0946, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 16:57:59,115 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9942, 1.1588, 1.1503, 1.0048], device='cuda:0'), covar=tensor([0.2306, 0.2772, 0.1507, 0.1991], device='cuda:0'), in_proj_covar=tensor([0.1980, 0.1916, 0.1844, 0.1987], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 16:58:07,142 INFO [train.py:968] (0/2) Epoch 22, batch 43250, giga_loss[loss=0.3064, simple_loss=0.3847, pruned_loss=0.1141, over 28978.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3689, pruned_loss=0.1206, over 5674872.41 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3594, pruned_loss=0.1133, over 5696935.57 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3708, pruned_loss=0.1222, over 5668743.60 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:58:13,217 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 16:58:34,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5260, 1.3439, 1.6386, 1.2829], device='cuda:0'), covar=tensor([0.1522, 0.2490, 0.1275, 0.1476], device='cuda:0'), in_proj_covar=tensor([0.0901, 0.0704, 0.0946, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 16:58:54,622 INFO [train.py:968] (0/2) Epoch 22, batch 43300, giga_loss[loss=0.2824, simple_loss=0.3605, pruned_loss=0.1022, over 28860.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3673, pruned_loss=0.1184, over 5679993.19 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3592, pruned_loss=0.1131, over 5699606.79 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3692, pruned_loss=0.1199, over 5672487.16 frames. ], batch size: 112, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:58:57,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1001033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:59:19,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.608e+03 2.198e+03 2.944e+03 6.542e+03, threshold=4.396e+03, percent-clipped=5.0
+2023-03-11 16:59:43,245 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1001080.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:59:43,635 INFO [train.py:968] (0/2) Epoch 22, batch 43350, giga_loss[loss=0.2953, simple_loss=0.3719, pruned_loss=0.1093, over 28660.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3666, pruned_loss=0.1189, over 5661243.36 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3596, pruned_loss=0.1134, over 5688769.60 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3678, pruned_loss=0.1199, over 5663904.72 frames. ], batch size: 262, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:59:46,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1001083.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 17:00:12,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1001112.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 17:00:32,574 INFO [train.py:968] (0/2) Epoch 22, batch 43400, giga_loss[loss=0.3278, simple_loss=0.3791, pruned_loss=0.1383, over 27509.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3655, pruned_loss=0.1192, over 5661689.07 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3595, pruned_loss=0.1133, over 5689248.42 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3666, pruned_loss=0.12, over 5663161.32 frames. ], batch size: 472, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:00:57,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.685e+03 2.002e+03 3.035e+03 8.067e+03, threshold=4.005e+03, percent-clipped=9.0
+2023-03-11 17:01:18,024 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1001176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:01:21,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1001179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:01:22,556 INFO [train.py:968] (0/2) Epoch 22, batch 43450, giga_loss[loss=0.2988, simple_loss=0.3722, pruned_loss=0.1127, over 28698.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3668, pruned_loss=0.1204, over 5665389.20 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1136, over 5692532.00 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3676, pruned_loss=0.1209, over 5663198.38 frames. ], batch size: 307, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:01:35,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2228, 1.5196, 1.5116, 1.4141], device='cuda:0'), covar=tensor([0.1744, 0.1554, 0.2156, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0751, 0.0718, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 17:01:48,447 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1001208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:02:10,278 INFO [train.py:968] (0/2) Epoch 22, batch 43500, giga_loss[loss=0.2965, simple_loss=0.3716, pruned_loss=0.1107, over 28506.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1222, over 5670808.16 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3592, pruned_loss=0.1132, over 5699213.81 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3716, pruned_loss=0.1233, over 5662225.37 frames. ], batch size: 85, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:02:22,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9536, 3.7823, 3.5311, 1.8812], device='cuda:0'), covar=tensor([0.0687, 0.0947, 0.1021, 0.1917], device='cuda:0'), in_proj_covar=tensor([0.1248, 0.1160, 0.0980, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 17:02:30,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.565e+03 2.035e+03 2.504e+03 6.227e+03, threshold=4.070e+03, percent-clipped=3.0
+2023-03-11 17:02:46,978 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1001270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:02:57,338 INFO [train.py:968] (0/2) Epoch 22, batch 43550, giga_loss[loss=0.2796, simple_loss=0.3637, pruned_loss=0.09772, over 28850.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3721, pruned_loss=0.1207, over 5670420.11 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3585, pruned_loss=0.1127, over 5704632.90 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3741, pruned_loss=0.1221, over 5658385.51 frames. ], batch size: 119, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:03:13,578 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5444, 4.4271, 1.6485, 1.6707], device='cuda:0'), covar=tensor([0.0996, 0.0416, 0.0975, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0558, 0.0388, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 17:03:14,168 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1001295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:03:46,991 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1001330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:03:47,499 INFO [train.py:968] (0/2) Epoch 22, batch 43600, giga_loss[loss=0.3479, simple_loss=0.3947, pruned_loss=0.1506, over 26670.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3731, pruned_loss=0.1205, over 5671714.43 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.358, pruned_loss=0.1124, over 5705925.62 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3756, pruned_loss=0.1222, over 5659616.67 frames. ], batch size: 555, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 17:04:00,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4419, 1.7207, 3.9400, 3.2811], device='cuda:0'), covar=tensor([0.1609, 0.2395, 0.0496, 0.0973], device='cuda:0'), in_proj_covar=tensor([0.0766, 0.0652, 0.0971, 0.0921], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 17:04:15,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.577e+03 1.973e+03 2.455e+03 6.542e+03, threshold=3.947e+03, percent-clipped=5.0
+2023-03-11 17:04:40,565 INFO [train.py:968] (0/2) Epoch 22, batch 43650, giga_loss[loss=0.2993, simple_loss=0.3675, pruned_loss=0.1155, over 28526.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3757, pruned_loss=0.1225, over 5676890.75 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.358, pruned_loss=0.1124, over 5705539.15 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3779, pruned_loss=0.124, over 5667386.05 frames. ], batch size: 78, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:05:27,596 INFO [train.py:968] (0/2) Epoch 22, batch 43700, giga_loss[loss=0.3947, simple_loss=0.4288, pruned_loss=0.1803, over 27509.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3766, pruned_loss=0.124, over 5672863.39 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3575, pruned_loss=0.1121, over 5708444.78 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3791, pruned_loss=0.1256, over 5662096.23 frames. ], batch size: 472, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:05:50,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.575e+02 1.692e+03 2.167e+03 2.953e+03 7.973e+03, threshold=4.334e+03, percent-clipped=12.0
+2023-03-11 17:06:13,458 INFO [train.py:968] (0/2) Epoch 22, batch 43750, giga_loss[loss=0.3487, simple_loss=0.398, pruned_loss=0.1497, over 27560.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3752, pruned_loss=0.1236, over 5673921.14 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3574, pruned_loss=0.112, over 5700068.97 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3776, pruned_loss=0.1252, over 5672907.61 frames. ], batch size: 472, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:06:41,272 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1001510.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 17:07:02,056 INFO [train.py:968] (0/2) Epoch 22, batch 43800, giga_loss[loss=0.3045, simple_loss=0.3752, pruned_loss=0.1169, over 28693.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3738, pruned_loss=0.1234, over 5661987.17 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3574, pruned_loss=0.112, over 5704294.23 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3761, pruned_loss=0.1249, over 5656836.00 frames. ], batch size: 307, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:07:27,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.695e+03 2.437e+03 3.012e+03 6.961e+03, threshold=4.875e+03, percent-clipped=10.0
+2023-03-11 17:07:51,814 INFO [train.py:968] (0/2) Epoch 22, batch 43850, giga_loss[loss=0.3682, simple_loss=0.3978, pruned_loss=0.1693, over 23447.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3722, pruned_loss=0.1232, over 5666033.74 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3575, pruned_loss=0.1118, over 5710287.72 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3744, pruned_loss=0.1249, over 5655491.58 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:08:25,452 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4858, 1.8373, 1.5638, 1.5563], device='cuda:0'), covar=tensor([0.0629, 0.0270, 0.0275, 0.0671], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0119, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 17:08:34,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6696, 1.6980, 1.8592, 1.4213], device='cuda:0'), covar=tensor([0.1775, 0.2609, 0.1445, 0.1763], device='cuda:0'), in_proj_covar=tensor([0.0901, 0.0704, 0.0947, 0.0845], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 17:08:35,630 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-11 17:08:41,030 INFO [train.py:968] (0/2) Epoch 22, batch 43900, giga_loss[loss=0.2873, simple_loss=0.3636, pruned_loss=0.1055, over 29019.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3709, pruned_loss=0.1225, over 5663345.11 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3571, pruned_loss=0.1114, over 5715995.09 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3734, pruned_loss=0.1246, over 5648355.86 frames. ], batch size: 164, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:08:54,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1001645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:09:04,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.203e+03 1.772e+03 2.563e+03 3.651e+03 7.034e+03, threshold=5.127e+03, percent-clipped=9.0
+2023-03-11 17:09:08,228 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5775, 2.2700, 1.5842, 0.8147], device='cuda:0'), covar=tensor([0.4869, 0.3110, 0.4342, 0.5817], device='cuda:0'), in_proj_covar=tensor([0.1763, 0.1670, 0.1606, 0.1438], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 17:09:18,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1001670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:09:28,668 INFO [train.py:968] (0/2) Epoch 22, batch 43950, giga_loss[loss=0.2698, simple_loss=0.3321, pruned_loss=0.1037, over 28735.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3718, pruned_loss=0.1232, over 5661458.37 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3573, pruned_loss=0.1115, over 5710819.54 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3741, pruned_loss=0.1252, over 5651786.56 frames. ], batch size: 99, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:09:54,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1001705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:10:12,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9274, 1.1861, 1.3244, 1.0072], device='cuda:0'), covar=tensor([0.2014, 0.1584, 0.2460, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0753, 0.0717, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 17:10:17,561 INFO [train.py:968] (0/2) Epoch 22, batch 44000, giga_loss[loss=0.2902, simple_loss=0.3571, pruned_loss=0.1117, over 28816.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.37, pruned_loss=0.1228, over 5665059.53 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3574, pruned_loss=0.1117, over 5714786.78 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3719, pruned_loss=0.1244, over 5652677.68 frames. ], batch size: 186, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 17:10:39,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.798e+03 2.352e+03 3.199e+03 7.127e+03, threshold=4.704e+03, percent-clipped=3.0
+2023-03-11 17:11:04,188 INFO [train.py:968] (0/2) Epoch 22, batch 44050, giga_loss[loss=0.3138, simple_loss=0.3763, pruned_loss=0.1256, over 28618.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3692, pruned_loss=0.1222, over 5669050.06 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3577, pruned_loss=0.1118, over 5714908.39 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3707, pruned_loss=0.1235, over 5658623.13 frames. ], batch size: 336, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 17:11:11,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1001788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:11:14,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1001791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:11:38,630 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1001813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:11:41,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1001816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:11:45,226 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1001820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:11:56,764 INFO [train.py:968] (0/2) Epoch 22, batch 44100, giga_loss[loss=0.3439, simple_loss=0.4006, pruned_loss=0.1436, over 27829.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3691, pruned_loss=0.1216, over 5664361.10 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3577, pruned_loss=0.1118, over 5714908.39 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3703, pruned_loss=0.1227, over 5656245.68 frames. ], batch size: 412, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:12:10,176 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1001845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:12:17,240 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1001848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:12:19,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1001851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:12:26,404 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.882e+02 1.561e+03 2.198e+03 3.246e+03 6.203e+03, threshold=4.396e+03, percent-clipped=6.0
+2023-03-11 17:12:50,092 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1001880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:12:50,465 INFO [train.py:968] (0/2) Epoch 22, batch 44150, giga_loss[loss=0.3179, simple_loss=0.3816, pruned_loss=0.1272, over 28696.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.372, pruned_loss=0.1233, over 5659861.40 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3581, pruned_loss=0.112, over 5718377.08 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.373, pruned_loss=0.1242, over 5648878.77 frames. ], batch size: 262, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:12:54,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1001885.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 17:13:36,911 INFO [train.py:968] (0/2) Epoch 22, batch 44200, giga_loss[loss=0.303, simple_loss=0.3756, pruned_loss=0.1152, over 28865.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3731, pruned_loss=0.1239, over 5663469.76 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3583, pruned_loss=0.112, over 5721994.05 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3739, pruned_loss=0.1248, over 5650468.53 frames. ], batch size: 174, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:14:04,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.809e+03 2.593e+03 3.668e+03 7.161e+03, threshold=5.186e+03, percent-clipped=15.0
+2023-03-11 17:14:23,791 INFO [train.py:968] (0/2) Epoch 22, batch 44250, giga_loss[loss=0.3954, simple_loss=0.4141, pruned_loss=0.1883, over 23597.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3739, pruned_loss=0.1243, over 5665449.09 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3589, pruned_loss=0.1126, over 5717918.18 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3747, pruned_loss=0.1251, over 5656199.40 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 17:14:27,634 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1001985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:14:42,953 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1002000.pt
+2023-03-11 17:15:10,114 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1002028.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 17:15:11,924 INFO [train.py:968] (0/2) Epoch 22, batch 44300, giga_loss[loss=0.3017, simple_loss=0.3695, pruned_loss=0.117, over 27959.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3739, pruned_loss=0.1212, over 5674069.70 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3586, pruned_loss=0.1123, over 5720189.21 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.375, pruned_loss=0.1222, over 5664213.59 frames. ], batch size: 412, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 17:15:14,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1002031.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 17:15:37,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.665e+02 1.552e+03 1.829e+03 2.436e+03 6.555e+03, threshold=3.658e+03, percent-clipped=4.0
+2023-03-11 17:15:38,680 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1002060.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 17:16:00,680 INFO [train.py:968] (0/2) Epoch 22, batch 44350, libri_loss[loss=0.228, simple_loss=0.2951, pruned_loss=0.08043, over 28420.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3752, pruned_loss=0.1204, over 5676176.35 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3582, pruned_loss=0.1122, over 5722760.13 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3768, pruned_loss=0.1215, over 5665191.24 frames. ], batch size: 63, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 17:16:02,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5971, 1.8471, 1.3320, 1.5183], device='cuda:0'), covar=tensor([0.0838, 0.0482, 0.0931, 0.0899], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0449, 0.0519, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 17:16:07,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9696, 1.3317, 1.0827, 0.1928], device='cuda:0'), covar=tensor([0.4280, 0.3417, 0.4725, 0.7362], device='cuda:0'), in_proj_covar=tensor([0.1759, 0.1661, 0.1600, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 17:16:31,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5981, 3.8857, 1.6615, 1.7964], device='cuda:0'), covar=tensor([0.0943, 0.0493, 0.0898, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0563, 0.0390, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 17:16:49,143 INFO [train.py:968] (0/2) Epoch 22, batch 44400, giga_loss[loss=0.3073, simple_loss=0.3786, pruned_loss=0.1179, over 28877.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3781, pruned_loss=0.1233, over 5667697.31 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3581, pruned_loss=0.112, over 5729339.86 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3801, pruned_loss=0.1246, over 5651227.74 frames. ], batch size: 186, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:16:54,397 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1002137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:17:17,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.753e+03 2.344e+03 3.086e+03 1.171e+04, threshold=4.688e+03, percent-clipped=19.0
+2023-03-11 17:17:25,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5075, 1.7705, 1.4354, 1.4962], device='cuda:0'), covar=tensor([0.2514, 0.2583, 0.2935, 0.2301], device='cuda:0'), in_proj_covar=tensor([0.1515, 0.1096, 0.1340, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 17:17:42,133 INFO [train.py:968] (0/2) Epoch 22, batch 44450, giga_loss[loss=0.2729, simple_loss=0.3525, pruned_loss=0.09665, over 29075.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3796, pruned_loss=0.1255, over 5665957.09 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3578, pruned_loss=0.1119, over 5732737.39 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3818, pruned_loss=0.127, over 5648889.38 frames. ], batch size: 155, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:18:13,005 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-11 17:18:31,622 INFO [train.py:968] (0/2) Epoch 22, batch 44500, giga_loss[loss=0.3045, simple_loss=0.3696, pruned_loss=0.1197, over 28313.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3783, pruned_loss=0.1252, over 5670803.68 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3572, pruned_loss=0.1116, over 5725559.83 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.381, pruned_loss=0.1269, over 5661451.78 frames. ], batch size: 368, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:18:59,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.721e+03 2.214e+03 2.686e+03 7.420e+03, threshold=4.428e+03, percent-clipped=1.0
+2023-03-11 17:19:20,474 INFO [train.py:968] (0/2) Epoch 22, batch 44550, giga_loss[loss=0.3071, simple_loss=0.3745, pruned_loss=0.1199, over 27896.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3782, pruned_loss=0.1256, over 5673573.75 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3572, pruned_loss=0.1118, over 5730161.92 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3808, pruned_loss=0.1271, over 5660685.49 frames. ], batch size: 412, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:20:07,828 INFO [train.py:968] (0/2) Epoch 22, batch 44600, giga_loss[loss=0.2601, simple_loss=0.3472, pruned_loss=0.08648, over 29006.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3759, pruned_loss=0.1229, over 5680512.76 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.357, pruned_loss=0.1116, over 5733411.71 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3786, pruned_loss=0.1245, over 5666315.47 frames. ], batch size: 155, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:20:33,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.966e+02 1.745e+03 2.201e+03 2.985e+03 5.938e+03, threshold=4.402e+03, percent-clipped=8.0
+2023-03-11 17:20:36,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1002360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:20:53,017 INFO [train.py:968] (0/2) Epoch 22, batch 44650, giga_loss[loss=0.2928, simple_loss=0.3785, pruned_loss=0.1036, over 28498.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3756, pruned_loss=0.1208, over 5686642.13 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3572, pruned_loss=0.1118, over 5734513.50 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3782, pruned_loss=0.1223, over 5672821.21 frames. ], batch size: 78, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:20:54,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5832, 1.5686, 1.7591, 1.3440], device='cuda:0'), covar=tensor([0.1741, 0.2585, 0.1468, 0.1754], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0705, 0.0948, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 17:21:05,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4176, 1.6694, 1.3605, 1.3117], device='cuda:0'), covar=tensor([0.2748, 0.2711, 0.3099, 0.2328], device='cuda:0'), in_proj_covar=tensor([0.1517, 0.1096, 0.1339, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 17:21:41,704 INFO [train.py:968] (0/2) Epoch 22, batch 44700, giga_loss[loss=0.4007, simple_loss=0.426, pruned_loss=0.1877, over 26681.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.377, pruned_loss=0.122, over 5671938.91 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3576, pruned_loss=0.1121, over 5726631.03 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.379, pruned_loss=0.123, over 5667944.93 frames. ], batch size: 555, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:22:07,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.857e+03 2.345e+03 2.859e+03 8.073e+03, threshold=4.690e+03, percent-clipped=5.0
+2023-03-11 17:22:33,339 INFO [train.py:968] (0/2) Epoch 22, batch 44750, giga_loss[loss=0.3255, simple_loss=0.368, pruned_loss=0.1415, over 23204.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3789, pruned_loss=0.1244, over 5658932.95 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3576, pruned_loss=0.1122, over 5729121.71 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3807, pruned_loss=0.1252, over 5652779.07 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:22:49,417 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-11 17:22:53,222 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1002503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:22:56,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1002506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:23:01,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1002512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:23:22,297 INFO [train.py:968] (0/2) Epoch 22, batch 44800, giga_loss[loss=0.293, simple_loss=0.3646, pruned_loss=0.1107, over 28639.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3774, pruned_loss=0.1236, over 5668320.52 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3576, pruned_loss=0.1122, over 5732663.05 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3793, pruned_loss=0.1245, over 5659401.80 frames. ], batch size: 262, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 17:23:27,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1002535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:23:49,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.230e+02 1.567e+03 2.074e+03 2.908e+03 7.805e+03, threshold=4.148e+03, percent-clipped=7.0
+2023-03-11 17:24:07,674 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3876, 1.6298, 1.6808, 1.4379], device='cuda:0'), covar=tensor([0.2057, 0.2115, 0.2306, 0.2199], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0752, 0.0715, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 17:24:16,558 INFO [train.py:968] (0/2) Epoch 22, batch 44850, giga_loss[loss=0.2962, simple_loss=0.3639, pruned_loss=0.1143, over 29048.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.376, pruned_loss=0.1241, over 5667427.48 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.358, pruned_loss=0.1126, over 5735507.22 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3773, pruned_loss=0.1246, over 5657059.74 frames. ], batch size: 136, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:25:04,566 INFO [train.py:968] (0/2) Epoch 22, batch 44900, libri_loss[loss=0.34, simple_loss=0.4011, pruned_loss=0.1395, over 25772.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3748, pruned_loss=0.1238, over 5667570.88 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3582, pruned_loss=0.1126, over 5735168.99 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3759, pruned_loss=0.1243, over 5658602.46 frames. ], batch size: 136, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:25:28,874 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1002655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:25:31,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1002658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:25:32,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.827e+02 1.787e+03 2.513e+03 3.551e+03 8.989e+03, threshold=5.025e+03, percent-clipped=14.0
+2023-03-11 17:25:51,693 INFO [train.py:968] (0/2) Epoch 22, batch 44950, giga_loss[loss=0.3023, simple_loss=0.3653, pruned_loss=0.1196, over 28657.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3724, pruned_loss=0.1225, over 5670358.00 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.358, pruned_loss=0.1124, over 5741154.39 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3741, pruned_loss=0.1235, over 5655358.77 frames. ], batch size: 262, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:25:57,926 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1002687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:26:12,244 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1002702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:26:13,007 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-11 17:26:40,670 INFO [train.py:968] (0/2) Epoch 22, batch 45000, libri_loss[loss=0.3219, simple_loss=0.3869, pruned_loss=0.1284, over 28628.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.373, pruned_loss=0.1242, over 5646601.15 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3584, pruned_loss=0.1128, over 5732556.23 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3741, pruned_loss=0.1248, over 5641645.45 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:26:40,675 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 17:26:50,445 INFO [train.py:1012] (0/2) Epoch 22, validation: loss=0.2077, simple_loss=0.317, pruned_loss=0.04922, over 944034.00 frames. 
+2023-03-11 17:26:50,446 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 17:26:52,853 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-11 17:27:17,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.216e+03 1.831e+03 2.225e+03 2.941e+03 8.681e+03, threshold=4.450e+03, percent-clipped=5.0
+2023-03-11 17:27:36,233 INFO [train.py:968] (0/2) Epoch 22, batch 45050, giga_loss[loss=0.2898, simple_loss=0.3639, pruned_loss=0.1078, over 28325.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3709, pruned_loss=0.1223, over 5646629.23 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3585, pruned_loss=0.1128, over 5735897.23 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3719, pruned_loss=0.123, over 5638112.53 frames. ], batch size: 368, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:27:45,614 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-11 17:28:23,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3106, 3.1462, 1.4876, 1.4785], device='cuda:0'), covar=tensor([0.1023, 0.0419, 0.0927, 0.1379], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0562, 0.0390, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 17:28:25,797 INFO [train.py:968] (0/2) Epoch 22, batch 45100, giga_loss[loss=0.271, simple_loss=0.3491, pruned_loss=0.09645, over 28925.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3683, pruned_loss=0.1185, over 5656518.74 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3586, pruned_loss=0.1128, over 5737479.08 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3692, pruned_loss=0.1191, over 5646983.11 frames. ], batch size: 145, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:28:48,269 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.130e+02 1.488e+03 1.852e+03 2.303e+03 6.572e+03, threshold=3.705e+03, percent-clipped=5.0
+2023-03-11 17:29:09,360 INFO [train.py:968] (0/2) Epoch 22, batch 45150, giga_loss[loss=0.3253, simple_loss=0.3625, pruned_loss=0.1441, over 24001.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3656, pruned_loss=0.1166, over 5655443.78 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3589, pruned_loss=0.113, over 5742145.53 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3664, pruned_loss=0.1171, over 5640226.97 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 17:29:17,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1002890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:30:01,949 INFO [train.py:968] (0/2) Epoch 22, batch 45200, giga_loss[loss=0.2785, simple_loss=0.3446, pruned_loss=0.1062, over 28696.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3648, pruned_loss=0.1158, over 5668208.15 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3588, pruned_loss=0.1129, over 5742326.19 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3656, pruned_loss=0.1164, over 5655043.24 frames. ], batch size: 242, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:30:32,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.757e+03 2.127e+03 2.745e+03 6.793e+03, threshold=4.253e+03, percent-clipped=11.0
+2023-03-11 17:30:54,670 INFO [train.py:968] (0/2) Epoch 22, batch 45250, giga_loss[loss=0.253, simple_loss=0.3199, pruned_loss=0.09308, over 28346.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3622, pruned_loss=0.1149, over 5679989.85 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3586, pruned_loss=0.1126, over 5743133.66 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3632, pruned_loss=0.1156, over 5666929.12 frames. ], batch size: 77, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:31:38,940 INFO [train.py:968] (0/2) Epoch 22, batch 45300, giga_loss[loss=0.2746, simple_loss=0.3583, pruned_loss=0.09545, over 28829.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3628, pruned_loss=0.1151, over 5680723.25 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3587, pruned_loss=0.1127, over 5737642.69 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3635, pruned_loss=0.1158, over 5673330.43 frames. ], batch size: 186, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:32:08,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.151e+03 1.779e+03 2.411e+03 3.361e+03 7.096e+03, threshold=4.823e+03, percent-clipped=12.0
+2023-03-11 17:32:18,209 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:32:22,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1003077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:32:26,035 INFO [train.py:968] (0/2) Epoch 22, batch 45350, giga_loss[loss=0.2744, simple_loss=0.3482, pruned_loss=0.1003, over 28788.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.365, pruned_loss=0.116, over 5681304.64 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3588, pruned_loss=0.1126, over 5737282.73 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3656, pruned_loss=0.1167, over 5674377.57 frames. ], batch size: 119, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:33:04,518 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:33:16,464 INFO [train.py:968] (0/2) Epoch 22, batch 45400, libri_loss[loss=0.3344, simple_loss=0.3941, pruned_loss=0.1373, over 26179.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3666, pruned_loss=0.1167, over 5666683.75 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3592, pruned_loss=0.1128, over 5729362.02 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3669, pruned_loss=0.1171, over 5666379.98 frames. ], batch size: 136, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:33:44,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.795e+02 1.809e+03 2.191e+03 2.695e+03 5.747e+03, threshold=4.382e+03, percent-clipped=5.0
+2023-03-11 17:34:06,393 INFO [train.py:968] (0/2) Epoch 22, batch 45450, giga_loss[loss=0.3173, simple_loss=0.3809, pruned_loss=0.1269, over 28857.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3683, pruned_loss=0.1186, over 5671306.94 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3595, pruned_loss=0.1131, over 5731188.87 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3684, pruned_loss=0.1187, over 5668840.14 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:34:12,537 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-11 17:34:44,603 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1003220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:34:46,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1003223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:34:55,376 INFO [train.py:968] (0/2) Epoch 22, batch 45500, giga_loss[loss=0.3086, simple_loss=0.3714, pruned_loss=0.1229, over 28257.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3682, pruned_loss=0.1192, over 5659592.72 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3594, pruned_loss=0.113, over 5734220.97 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3685, pruned_loss=0.1195, over 5653586.23 frames. ], batch size: 368, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:35:13,341 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1003252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:35:20,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.452e+03 1.898e+03 2.625e+03 6.156e+03, threshold=3.796e+03, percent-clipped=6.0
+2023-03-11 17:35:25,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1003265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:35:42,444 INFO [train.py:968] (0/2) Epoch 22, batch 45550, giga_loss[loss=0.3377, simple_loss=0.4007, pruned_loss=0.1373, over 28359.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3702, pruned_loss=0.1208, over 5651039.45 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3596, pruned_loss=0.1133, over 5737499.60 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3704, pruned_loss=0.1209, over 5641823.66 frames. ], batch size: 60, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:36:25,792 INFO [train.py:968] (0/2) Epoch 22, batch 45600, giga_loss[loss=0.2847, simple_loss=0.3517, pruned_loss=0.1089, over 28792.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3712, pruned_loss=0.1212, over 5609681.72 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3599, pruned_loss=0.1135, over 5684571.26 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3715, pruned_loss=0.1213, over 5646768.37 frames. ], batch size: 99, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 17:36:52,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.747e+03 2.198e+03 3.070e+03 5.921e+03, threshold=4.396e+03, percent-clipped=12.0
+2023-03-11 17:37:10,688 INFO [train.py:968] (0/2) Epoch 22, batch 45650, giga_loss[loss=0.3306, simple_loss=0.3936, pruned_loss=0.1338, over 28980.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3727, pruned_loss=0.1226, over 5557670.38 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3607, pruned_loss=0.1142, over 5614661.58 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3726, pruned_loss=0.1223, over 5647961.69 frames. ], batch size: 186, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:37:22,017 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-11 17:37:23,301 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-22.pt
+2023-03-11 17:38:10,597 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1003408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:38:13,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1003411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:38:40,279 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1003440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:38:40,651 INFO [train.py:968] (0/2) Epoch 23, batch 50, giga_loss[loss=0.274, simple_loss=0.3589, pruned_loss=0.09457, over 28695.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.374, pruned_loss=0.1089, over 1264513.61 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3396, pruned_loss=0.08733, over 172678.08 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3785, pruned_loss=0.1118, over 1125835.44 frames. ], batch size: 242, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:38:49,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1003448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:38:59,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.308e+02 1.454e+03 1.840e+03 2.324e+03 7.307e+03, threshold=3.680e+03, percent-clipped=4.0
+2023-03-11 17:39:28,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4224, 1.5529, 1.2858, 1.1009], device='cuda:0'), covar=tensor([0.1083, 0.0611, 0.1117, 0.1179], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0451, 0.0522, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 17:39:29,499 INFO [train.py:968] (0/2) Epoch 23, batch 100, giga_loss[loss=0.3036, simple_loss=0.3738, pruned_loss=0.1167, over 27552.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3628, pruned_loss=0.1032, over 2245198.76 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3339, pruned_loss=0.08692, over 392311.58 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3679, pruned_loss=0.106, over 1989437.40 frames. ], batch size: 472, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:39:33,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1003494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:39:34,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1797, 3.9917, 3.7640, 1.9773], device='cuda:0'), covar=tensor([0.0617, 0.0793, 0.0800, 0.1977], device='cuda:0'), in_proj_covar=tensor([0.1251, 0.1160, 0.0981, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 17:40:14,348 INFO [train.py:968] (0/2) Epoch 23, batch 150, giga_loss[loss=0.2391, simple_loss=0.3265, pruned_loss=0.07582, over 28623.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3491, pruned_loss=0.09694, over 3009520.31 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3353, pruned_loss=0.08804, over 525127.22 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3516, pruned_loss=0.09841, over 2735646.86 frames. ], batch size: 78, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:40:34,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.177e+02 1.161e+03 1.441e+03 1.740e+03 5.974e+03, threshold=2.883e+03, percent-clipped=3.0
+2023-03-11 17:40:49,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2343, 4.0429, 3.8005, 1.7863], device='cuda:0'), covar=tensor([0.0563, 0.0758, 0.0749, 0.2200], device='cuda:0'), in_proj_covar=tensor([0.1250, 0.1158, 0.0980, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 17:40:59,909 INFO [train.py:968] (0/2) Epoch 23, batch 200, libri_loss[loss=0.2655, simple_loss=0.3467, pruned_loss=0.09211, over 29268.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3347, pruned_loss=0.09001, over 3611552.29 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3359, pruned_loss=0.08826, over 551769.58 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3353, pruned_loss=0.09048, over 3386055.42 frames. ], batch size: 94, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:41:00,150 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1003591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:02,480 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1003594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:25,259 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1003623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:25,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4255, 1.7298, 1.7442, 1.5269], device='cuda:0'), covar=tensor([0.2294, 0.2205, 0.2410, 0.2338], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0754, 0.0718, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 17:41:28,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:37,340 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1003637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:41,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1003640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:43,149 INFO [train.py:968] (0/2) Epoch 23, batch 250, giga_loss[loss=0.188, simple_loss=0.2673, pruned_loss=0.05433, over 28707.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3232, pruned_loss=0.08443, over 4079353.86 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3316, pruned_loss=0.08653, over 684501.83 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3234, pruned_loss=0.08471, over 3853583.65 frames. ], batch size: 92, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:41:58,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.163e+02 1.103e+03 1.323e+03 2.037e+03 4.202e+03, threshold=2.646e+03, percent-clipped=6.0
+2023-03-11 17:42:06,957 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1003669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:42:27,424 INFO [train.py:968] (0/2) Epoch 23, batch 300, giga_loss[loss=0.1962, simple_loss=0.2799, pruned_loss=0.05626, over 28541.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3159, pruned_loss=0.08142, over 4442042.31 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3342, pruned_loss=0.08721, over 788180.16 frames. ], giga_tot_loss[loss=0.2388, simple_loss=0.315, pruned_loss=0.08128, over 4233580.60 frames. ], batch size: 336, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:42:31,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:42:34,154 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:43:13,653 INFO [train.py:968] (0/2) Epoch 23, batch 350, giga_loss[loss=0.2132, simple_loss=0.2944, pruned_loss=0.06597, over 28894.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.309, pruned_loss=0.07842, over 4723609.64 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3342, pruned_loss=0.0866, over 890104.59 frames. ], giga_tot_loss[loss=0.2319, simple_loss=0.3075, pruned_loss=0.07813, over 4534398.92 frames. ], batch size: 145, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:43:31,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.151e+02 1.064e+03 1.476e+03 2.450e+03 9.142e+03, threshold=2.952e+03, percent-clipped=20.0
+2023-03-11 17:43:40,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3625, 1.0796, 3.8502, 3.4546], device='cuda:0'), covar=tensor([0.1921, 0.3292, 0.0810, 0.0988], device='cuda:0'), in_proj_covar=tensor([0.0765, 0.0653, 0.0973, 0.0920], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 17:43:57,394 INFO [train.py:968] (0/2) Epoch 23, batch 400, giga_loss[loss=0.2336, simple_loss=0.2901, pruned_loss=0.08852, over 24135.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3058, pruned_loss=0.07727, over 4941040.13 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3361, pruned_loss=0.08736, over 1010835.85 frames. ], giga_tot_loss[loss=0.2284, simple_loss=0.3035, pruned_loss=0.07661, over 4769289.64 frames. ], batch size: 705, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:44:40,956 INFO [train.py:968] (0/2) Epoch 23, batch 450, giga_loss[loss=0.2149, simple_loss=0.2992, pruned_loss=0.06531, over 28898.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3036, pruned_loss=0.07621, over 5114845.22 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3377, pruned_loss=0.08827, over 1082581.54 frames. ], giga_tot_loss[loss=0.2258, simple_loss=0.3009, pruned_loss=0.07531, over 4968580.06 frames. ], batch size: 174, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:44:59,902 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.172e+02 9.814e+02 1.177e+03 1.545e+03 3.973e+03, threshold=2.354e+03, percent-clipped=1.0
+2023-03-11 17:45:10,460 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:45:16,506 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:45:28,684 INFO [train.py:968] (0/2) Epoch 23, batch 500, giga_loss[loss=0.2041, simple_loss=0.2792, pruned_loss=0.06455, over 28967.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3001, pruned_loss=0.07457, over 5247616.04 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3385, pruned_loss=0.08851, over 1106387.64 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2977, pruned_loss=0.07374, over 5130280.17 frames. ], batch size: 227, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:45:32,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6561, 1.7546, 1.8461, 1.4210], device='cuda:0'), covar=tensor([0.1970, 0.2622, 0.1582, 0.1802], device='cuda:0'), in_proj_covar=tensor([0.0914, 0.0711, 0.0962, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 17:46:11,890 INFO [train.py:968] (0/2) Epoch 23, batch 550, giga_loss[loss=0.213, simple_loss=0.2851, pruned_loss=0.07041, over 28825.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2992, pruned_loss=0.07415, over 5345803.16 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3392, pruned_loss=0.08912, over 1238595.95 frames. ], giga_tot_loss[loss=0.221, simple_loss=0.296, pruned_loss=0.07297, over 5243987.32 frames. ], batch size: 285, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:46:22,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:46:34,351 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.117e+02 1.142e+03 1.388e+03 2.188e+03 6.191e+03, threshold=2.775e+03, percent-clipped=21.0
+2023-03-11 17:46:38,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9513, 2.3621, 2.1942, 1.7531], device='cuda:0'), covar=tensor([0.3500, 0.2435, 0.2357, 0.3023], device='cuda:0'), in_proj_covar=tensor([0.1987, 0.1914, 0.1855, 0.1995], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 17:47:03,102 INFO [train.py:968] (0/2) Epoch 23, batch 600, giga_loss[loss=0.2058, simple_loss=0.2766, pruned_loss=0.06743, over 27697.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.297, pruned_loss=0.07347, over 5415488.63 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3383, pruned_loss=0.08873, over 1285305.54 frames. ], giga_tot_loss[loss=0.2196, simple_loss=0.2942, pruned_loss=0.07247, over 5330698.82 frames. ], batch size: 472, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:47:11,076 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1004000.pt
+2023-03-11 17:47:12,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:47:49,897 INFO [train.py:968] (0/2) Epoch 23, batch 650, giga_loss[loss=0.1975, simple_loss=0.2749, pruned_loss=0.06002, over 28445.00 frames. ], tot_loss[loss=0.221, simple_loss=0.296, pruned_loss=0.07297, over 5469155.83 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3398, pruned_loss=0.08923, over 1393292.06 frames. ], giga_tot_loss[loss=0.2179, simple_loss=0.2924, pruned_loss=0.07165, over 5397015.53 frames. ], batch size: 78, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:48:08,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.735e+02 1.006e+03 1.221e+03 1.620e+03 4.724e+03, threshold=2.442e+03, percent-clipped=6.0
+2023-03-11 17:48:18,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:48:19,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:48:34,784 INFO [train.py:968] (0/2) Epoch 23, batch 700, giga_loss[loss=0.2092, simple_loss=0.2809, pruned_loss=0.06882, over 28820.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2932, pruned_loss=0.07155, over 5525815.42 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3403, pruned_loss=0.08935, over 1459690.59 frames. ], giga_tot_loss[loss=0.215, simple_loss=0.2896, pruned_loss=0.07023, over 5464268.67 frames. ], batch size: 112, lr: 1.40e-03, grad_scale: 2.0
+2023-03-11 17:49:21,269 INFO [train.py:968] (0/2) Epoch 23, batch 750, giga_loss[loss=0.21, simple_loss=0.2768, pruned_loss=0.07166, over 28875.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2912, pruned_loss=0.07044, over 5573583.32 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3395, pruned_loss=0.08879, over 1611931.89 frames. ], giga_tot_loss[loss=0.2124, simple_loss=0.2869, pruned_loss=0.06893, over 5513707.66 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 2.0
+2023-03-11 17:49:25,083 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:49:27,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:49:41,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.785e+02 9.774e+02 1.159e+03 1.644e+03 3.853e+03, threshold=2.318e+03, percent-clipped=5.0
+2023-03-11 17:49:53,694 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:49:56,442 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1898, 1.2885, 5.3760, 3.9285], device='cuda:0'), covar=tensor([0.1506, 0.3130, 0.0396, 0.0820], device='cuda:0'), in_proj_covar=tensor([0.0763, 0.0652, 0.0971, 0.0917], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 17:50:05,801 INFO [train.py:968] (0/2) Epoch 23, batch 800, giga_loss[loss=0.2206, simple_loss=0.2957, pruned_loss=0.07275, over 28855.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2909, pruned_loss=0.0711, over 5597852.59 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3395, pruned_loss=0.08891, over 1695012.89 frames. ], giga_tot_loss[loss=0.2129, simple_loss=0.2867, pruned_loss=0.06955, over 5546219.87 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:50:27,576 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:29,888 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:30,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:30,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7303, 2.0926, 1.9980, 1.5793], device='cuda:0'), covar=tensor([0.3277, 0.2317, 0.2360, 0.3008], device='cuda:0'), in_proj_covar=tensor([0.1995, 0.1919, 0.1861, 0.2001], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 17:50:32,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:54,534 INFO [train.py:968] (0/2) Epoch 23, batch 850, giga_loss[loss=0.2539, simple_loss=0.3386, pruned_loss=0.08456, over 28730.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.301, pruned_loss=0.07619, over 5611887.90 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3405, pruned_loss=0.08918, over 1818267.38 frames. ], giga_tot_loss[loss=0.2225, simple_loss=0.2962, pruned_loss=0.07447, over 5563917.37 frames. ], batch size: 242, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:50:59,922 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:59,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:51:01,257 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:51:07,035 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:51:16,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.524e+02 1.205e+03 1.555e+03 2.170e+03 5.877e+03, threshold=3.110e+03, percent-clipped=21.0
+2023-03-11 17:51:41,428 INFO [train.py:968] (0/2) Epoch 23, batch 900, giga_loss[loss=0.3065, simple_loss=0.3739, pruned_loss=0.1196, over 28668.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3144, pruned_loss=0.08279, over 5632151.36 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3404, pruned_loss=0.08897, over 1898455.72 frames. ], giga_tot_loss[loss=0.2364, simple_loss=0.3101, pruned_loss=0.08135, over 5589930.16 frames. ], batch size: 99, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:51:50,675 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3100, 1.4431, 1.3622, 1.5028], device='cuda:0'), covar=tensor([0.0773, 0.0409, 0.0341, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 17:51:54,957 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.11 vs. limit=5.0
+2023-03-11 17:52:17,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:52:28,485 INFO [train.py:968] (0/2) Epoch 23, batch 950, giga_loss[loss=0.2599, simple_loss=0.3404, pruned_loss=0.08975, over 29018.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3253, pruned_loss=0.08801, over 5650770.41 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3396, pruned_loss=0.08862, over 1938412.09 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3221, pruned_loss=0.08701, over 5615097.83 frames. ], batch size: 128, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:52:46,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.690e+02 1.439e+03 1.904e+03 2.595e+03 8.260e+03, threshold=3.809e+03, percent-clipped=16.0
+2023-03-11 17:52:56,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2785, 1.2756, 3.1389, 2.8807], device='cuda:0'), covar=tensor([0.1426, 0.2686, 0.0504, 0.2220], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0650, 0.0969, 0.0914], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 17:53:09,565 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:12,024 INFO [train.py:968] (0/2) Epoch 23, batch 1000, giga_loss[loss=0.2651, simple_loss=0.354, pruned_loss=0.0881, over 28328.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3311, pruned_loss=0.0898, over 5663417.64 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3381, pruned_loss=0.08776, over 2014353.03 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3288, pruned_loss=0.0893, over 5633253.69 frames. ], batch size: 368, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:53:12,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:16,716 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:18,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:35,320 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:42,705 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:53,529 INFO [train.py:968] (0/2) Epoch 23, batch 1050, giga_loss[loss=0.2503, simple_loss=0.336, pruned_loss=0.0823, over 28362.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3352, pruned_loss=0.09084, over 5658658.33 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3391, pruned_loss=0.08851, over 2120142.63 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.333, pruned_loss=0.09026, over 5636504.71 frames. ], batch size: 368, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:54:14,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.506e+02 1.131e+03 1.406e+03 1.957e+03 4.261e+03, threshold=2.812e+03, percent-clipped=1.0
+2023-03-11 17:54:20,069 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:54:22,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:54:38,426 INFO [train.py:968] (0/2) Epoch 23, batch 1100, giga_loss[loss=0.2359, simple_loss=0.325, pruned_loss=0.07338, over 28930.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3362, pruned_loss=0.09041, over 5673517.88 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3391, pruned_loss=0.08834, over 2247237.05 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3343, pruned_loss=0.09009, over 5651154.01 frames. ], batch size: 174, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:54:48,731 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:55:23,665 INFO [train.py:968] (0/2) Epoch 23, batch 1150, giga_loss[loss=0.257, simple_loss=0.3342, pruned_loss=0.08985, over 28791.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3378, pruned_loss=0.09148, over 5677312.10 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.339, pruned_loss=0.08837, over 2327545.82 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3363, pruned_loss=0.09129, over 5663549.11 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:55:44,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.198e+02 1.237e+03 1.530e+03 2.047e+03 5.805e+03, threshold=3.059e+03, percent-clipped=12.0
+2023-03-11 17:56:02,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 17:56:09,360 INFO [train.py:968] (0/2) Epoch 23, batch 1200, giga_loss[loss=0.2891, simple_loss=0.3661, pruned_loss=0.106, over 28563.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3419, pruned_loss=0.09463, over 5672756.16 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3401, pruned_loss=0.08909, over 2380645.34 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3403, pruned_loss=0.09427, over 5658902.15 frames. ], batch size: 307, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:56:50,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2144, 1.3491, 3.4534, 3.1166], device='cuda:0'), covar=tensor([0.1621, 0.2746, 0.0489, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0756, 0.0645, 0.0959, 0.0909], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 17:56:52,671 INFO [train.py:968] (0/2) Epoch 23, batch 1250, giga_loss[loss=0.3081, simple_loss=0.3684, pruned_loss=0.1239, over 28978.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3449, pruned_loss=0.09657, over 5675530.20 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3402, pruned_loss=0.08889, over 2463132.26 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3437, pruned_loss=0.09652, over 5663530.82 frames. ], batch size: 106, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:56:53,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7167, 2.0084, 1.8347, 1.8469], device='cuda:0'), covar=tensor([0.0778, 0.0290, 0.0314, 0.0863], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 17:57:13,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.828e+02 1.318e+03 1.599e+03 2.255e+03 5.869e+03, threshold=3.197e+03, percent-clipped=7.0
+2023-03-11 17:57:30,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6939, 2.0885, 1.8364, 1.8597], device='cuda:0'), covar=tensor([0.2164, 0.2016, 0.2421, 0.2031], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0751, 0.0716, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 17:57:34,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1004687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:57:38,604 INFO [train.py:968] (0/2) Epoch 23, batch 1300, libri_loss[loss=0.2636, simple_loss=0.3491, pruned_loss=0.0891, over 29371.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3475, pruned_loss=0.09729, over 5685179.02 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3397, pruned_loss=0.08863, over 2564788.38 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3469, pruned_loss=0.0976, over 5669697.95 frames. ], batch size: 92, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:57:39,353 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1004692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:58:02,316 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4303, 1.5947, 1.4509, 1.2919], device='cuda:0'), covar=tensor([0.2563, 0.2843, 0.2171, 0.2607], device='cuda:0'), in_proj_covar=tensor([0.1973, 0.1906, 0.1845, 0.1986], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 17:58:21,454 INFO [train.py:968] (0/2) Epoch 23, batch 1350, giga_loss[loss=0.2622, simple_loss=0.3512, pruned_loss=0.08658, over 28670.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3491, pruned_loss=0.09703, over 5691958.19 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3405, pruned_loss=0.08904, over 2628434.90 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3485, pruned_loss=0.09726, over 5678227.24 frames. ], batch size: 284, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:58:40,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.891e+02 1.264e+03 1.520e+03 1.869e+03 3.315e+03, threshold=3.041e+03, percent-clipped=3.0
+2023-03-11 17:59:02,630 INFO [train.py:968] (0/2) Epoch 23, batch 1400, giga_loss[loss=0.2803, simple_loss=0.3625, pruned_loss=0.09906, over 28700.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3499, pruned_loss=0.09662, over 5687807.23 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3409, pruned_loss=0.08936, over 2651393.93 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3493, pruned_loss=0.09673, over 5683564.35 frames. ], batch size: 284, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:59:46,898 INFO [train.py:968] (0/2) Epoch 23, batch 1450, giga_loss[loss=0.268, simple_loss=0.3471, pruned_loss=0.09447, over 28587.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3489, pruned_loss=0.09481, over 5687613.93 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3415, pruned_loss=0.08963, over 2706132.74 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3483, pruned_loss=0.09489, over 5690100.14 frames. ], batch size: 336, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:59:49,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1004844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:00:05,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.126e+02 1.190e+03 1.407e+03 1.726e+03 5.165e+03, threshold=2.813e+03, percent-clipped=8.0
+2023-03-11 18:00:27,637 INFO [train.py:968] (0/2) Epoch 23, batch 1500, giga_loss[loss=0.2164, simple_loss=0.3095, pruned_loss=0.06162, over 28538.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3462, pruned_loss=0.09213, over 5683418.71 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3413, pruned_loss=0.08961, over 2743928.96 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3459, pruned_loss=0.09227, over 5690874.83 frames. ], batch size: 71, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:01:11,621 INFO [train.py:968] (0/2) Epoch 23, batch 1550, libri_loss[loss=0.2644, simple_loss=0.35, pruned_loss=0.08937, over 29525.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3459, pruned_loss=0.09217, over 5694062.68 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3417, pruned_loss=0.08947, over 2837603.09 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3457, pruned_loss=0.0924, over 5694204.56 frames. ], batch size: 89, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:01:32,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.815e+02 1.124e+03 1.434e+03 1.784e+03 3.732e+03, threshold=2.867e+03, percent-clipped=7.0
+2023-03-11 18:01:56,082 INFO [train.py:968] (0/2) Epoch 23, batch 1600, giga_loss[loss=0.4032, simple_loss=0.4164, pruned_loss=0.195, over 26550.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3489, pruned_loss=0.09686, over 5701676.11 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3408, pruned_loss=0.08921, over 2883432.24 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3492, pruned_loss=0.09722, over 5698990.93 frames. ], batch size: 555, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:02:35,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2706, 2.2735, 2.1518, 1.9438], device='cuda:0'), covar=tensor([0.2638, 0.2471, 0.2687, 0.2517], device='cuda:0'), in_proj_covar=tensor([0.1968, 0.1902, 0.1842, 0.1981], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 18:02:44,114 INFO [train.py:968] (0/2) Epoch 23, batch 1650, giga_loss[loss=0.2512, simple_loss=0.3364, pruned_loss=0.08295, over 28505.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1011, over 5696269.51 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3413, pruned_loss=0.08965, over 2941763.46 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3517, pruned_loss=0.1014, over 5691976.42 frames. ], batch size: 71, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:03:04,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:03:05,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.813e+02 1.414e+03 1.750e+03 2.264e+03 6.095e+03, threshold=3.500e+03, percent-clipped=11.0
+2023-03-11 18:03:07,872 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:03:15,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4391, 1.4825, 1.3773, 1.6074], device='cuda:0'), covar=tensor([0.0634, 0.0308, 0.0283, 0.0658], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 18:03:16,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3564, 1.2473, 3.8032, 3.3114], device='cuda:0'), covar=tensor([0.1629, 0.2854, 0.0443, 0.1036], device='cuda:0'), in_proj_covar=tensor([0.0760, 0.0649, 0.0966, 0.0914], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 18:03:27,478 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4018, 1.6099, 1.2911, 1.1896], device='cuda:0'), covar=tensor([0.1035, 0.0568, 0.1050, 0.1108], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0448, 0.0521, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 18:03:27,830 INFO [train.py:968] (0/2) Epoch 23, batch 1700, giga_loss[loss=0.2819, simple_loss=0.3535, pruned_loss=0.1052, over 28959.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3513, pruned_loss=0.1022, over 5695391.56 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3416, pruned_loss=0.08972, over 2999986.92 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3515, pruned_loss=0.1027, over 5688603.82 frames. ], batch size: 106, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:04:01,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6043, 1.7448, 1.8090, 1.3864], device='cuda:0'), covar=tensor([0.1779, 0.2457, 0.1415, 0.1654], device='cuda:0'), in_proj_covar=tensor([0.0909, 0.0704, 0.0955, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 18:04:13,614 INFO [train.py:968] (0/2) Epoch 23, batch 1750, giga_loss[loss=0.2467, simple_loss=0.3247, pruned_loss=0.08431, over 28804.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3487, pruned_loss=0.1008, over 5707831.99 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3411, pruned_loss=0.08941, over 3086008.64 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3493, pruned_loss=0.1016, over 5696890.97 frames. ], batch size: 99, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:04:28,729 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:04:33,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.464e+02 1.275e+03 1.607e+03 2.314e+03 6.332e+03, threshold=3.215e+03, percent-clipped=5.0
+2023-03-11 18:04:56,477 INFO [train.py:968] (0/2) Epoch 23, batch 1800, libri_loss[loss=0.2578, simple_loss=0.3415, pruned_loss=0.08706, over 29518.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3481, pruned_loss=0.1005, over 5711006.11 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.341, pruned_loss=0.08919, over 3193503.28 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3489, pruned_loss=0.1017, over 5698013.59 frames. ], batch size: 80, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:05:08,087 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:11,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:12,495 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:14,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:19,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:22,675 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2839, 3.4530, 1.5132, 1.4502], device='cuda:0'), covar=tensor([0.1072, 0.0281, 0.0884, 0.1451], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0556, 0.0388, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 18:05:38,917 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:41,216 INFO [train.py:968] (0/2) Epoch 23, batch 1850, giga_loss[loss=0.252, simple_loss=0.3272, pruned_loss=0.08844, over 28676.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3466, pruned_loss=0.09835, over 5718940.25 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3416, pruned_loss=0.08949, over 3220174.40 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.347, pruned_loss=0.09922, over 5707418.32 frames. ], batch size: 92, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:05:42,109 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:47,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9467, 2.1663, 2.0159, 1.7288], device='cuda:0'), covar=tensor([0.3422, 0.2580, 0.2894, 0.3081], device='cuda:0'), in_proj_covar=tensor([0.1973, 0.1907, 0.1844, 0.1982], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 18:06:06,311 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.781e+02 1.162e+03 1.408e+03 2.017e+03 5.118e+03, threshold=2.815e+03, percent-clipped=5.0
+2023-03-11 18:06:19,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9165, 1.2704, 1.0441, 0.1820], device='cuda:0'), covar=tensor([0.4465, 0.3267, 0.4743, 0.7221], device='cuda:0'), in_proj_covar=tensor([0.1754, 0.1656, 0.1597, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 18:06:33,977 INFO [train.py:968] (0/2) Epoch 23, batch 1900, giga_loss[loss=0.2634, simple_loss=0.3405, pruned_loss=0.09315, over 28299.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3436, pruned_loss=0.09641, over 5694181.89 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3413, pruned_loss=0.08936, over 3231341.52 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.344, pruned_loss=0.09721, over 5686373.54 frames. ], batch size: 368, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:06:53,700 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:06:54,270 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:07:09,890 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-11 18:07:23,050 INFO [train.py:968] (0/2) Epoch 23, batch 1950, giga_loss[loss=0.2417, simple_loss=0.3129, pruned_loss=0.08529, over 27905.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3385, pruned_loss=0.09342, over 5686051.69 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3413, pruned_loss=0.08928, over 3273791.31 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3389, pruned_loss=0.09419, over 5684880.18 frames. ], batch size: 412, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:07:37,249 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-11 18:07:44,017 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:07:45,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.877e+02 1.016e+03 1.346e+03 1.987e+03 5.471e+03, threshold=2.691e+03, percent-clipped=9.0
+2023-03-11 18:07:46,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:07:55,455 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3854, 1.6132, 1.3823, 1.5704], device='cuda:0'), covar=tensor([0.0763, 0.0327, 0.0337, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 18:08:09,871 INFO [train.py:968] (0/2) Epoch 23, batch 2000, giga_loss[loss=0.246, simple_loss=0.3243, pruned_loss=0.08387, over 28945.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3325, pruned_loss=0.08989, over 5687405.33 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3408, pruned_loss=0.08885, over 3349364.15 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3329, pruned_loss=0.09079, over 5682717.74 frames. ], batch size: 227, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:08:13,502 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:08:56,896 INFO [train.py:968] (0/2) Epoch 23, batch 2050, giga_loss[loss=0.2176, simple_loss=0.2982, pruned_loss=0.06853, over 28744.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3285, pruned_loss=0.08817, over 5686289.54 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3408, pruned_loss=0.08899, over 3462710.76 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3284, pruned_loss=0.08886, over 5673280.31 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:09:18,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.588e+02 1.037e+03 1.316e+03 1.833e+03 4.804e+03, threshold=2.633e+03, percent-clipped=9.0
+2023-03-11 18:09:38,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3671, 2.8931, 1.5565, 1.4002], device='cuda:0'), covar=tensor([0.1010, 0.0321, 0.0879, 0.1392], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0553, 0.0387, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 18:09:40,861 INFO [train.py:968] (0/2) Epoch 23, batch 2100, giga_loss[loss=0.2319, simple_loss=0.317, pruned_loss=0.07336, over 28884.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3294, pruned_loss=0.08815, over 5688306.68 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3414, pruned_loss=0.08927, over 3512985.48 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3288, pruned_loss=0.08853, over 5681734.17 frames. ], batch size: 66, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:10:08,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9706, 1.3252, 1.0880, 0.1996], device='cuda:0'), covar=tensor([0.4077, 0.3148, 0.5222, 0.6980], device='cuda:0'), in_proj_covar=tensor([0.1740, 0.1642, 0.1584, 0.1422], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 18:10:16,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:10:17,079 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:10:22,193 INFO [train.py:968] (0/2) Epoch 23, batch 2150, giga_loss[loss=0.2439, simple_loss=0.3176, pruned_loss=0.08506, over 28165.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3311, pruned_loss=0.08883, over 5697609.00 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3419, pruned_loss=0.08933, over 3571829.88 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.33, pruned_loss=0.08908, over 5687728.17 frames. ], batch size: 77, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:10:28,149 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:10:41,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.373e+02 1.173e+03 1.527e+03 1.829e+03 6.467e+03, threshold=3.054e+03, percent-clipped=6.0
+2023-03-11 18:11:02,250 INFO [train.py:968] (0/2) Epoch 23, batch 2200, giga_loss[loss=0.3539, simple_loss=0.3997, pruned_loss=0.1541, over 26734.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.331, pruned_loss=0.08858, over 5700177.38 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3425, pruned_loss=0.08936, over 3669746.97 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3293, pruned_loss=0.08874, over 5688441.33 frames. ], batch size: 555, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:11:16,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5427, 2.1306, 1.6094, 0.7790], device='cuda:0'), covar=tensor([0.5747, 0.3101, 0.4432, 0.6581], device='cuda:0'), in_proj_covar=tensor([0.1741, 0.1644, 0.1587, 0.1425], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 18:11:33,243 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 18:11:40,626 INFO [train.py:968] (0/2) Epoch 23, batch 2250, giga_loss[loss=0.2195, simple_loss=0.2981, pruned_loss=0.07048, over 28677.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3289, pruned_loss=0.08792, over 5712831.45 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3426, pruned_loss=0.08964, over 3755669.11 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.327, pruned_loss=0.08782, over 5698010.09 frames. ], batch size: 242, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:11:42,179 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 18:11:42,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3810, 3.5932, 2.4167, 1.4015], device='cuda:0'), covar=tensor([0.6710, 0.3019, 0.3934, 0.6353], device='cuda:0'), in_proj_covar=tensor([0.1741, 0.1645, 0.1587, 0.1425], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 18:11:58,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4676, 1.5526, 1.4405, 1.3278], device='cuda:0'), covar=tensor([0.2722, 0.2465, 0.2140, 0.2858], device='cuda:0'), in_proj_covar=tensor([0.1969, 0.1897, 0.1833, 0.1975], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 18:12:03,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.513e+02 1.136e+03 1.400e+03 1.903e+03 6.644e+03, threshold=2.799e+03, percent-clipped=7.0
+2023-03-11 18:12:11,996 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-11 18:12:12,566 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:12:15,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:12:24,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:12:24,359 INFO [train.py:968] (0/2) Epoch 23, batch 2300, giga_loss[loss=0.2204, simple_loss=0.3014, pruned_loss=0.06973, over 28797.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3259, pruned_loss=0.08636, over 5712470.69 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3423, pruned_loss=0.08927, over 3796159.15 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3243, pruned_loss=0.08645, over 5699120.39 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:12:24,568 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:12:30,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8705, 4.7035, 4.4698, 2.0227], device='cuda:0'), covar=tensor([0.0592, 0.0708, 0.0873, 0.1908], device='cuda:0'), in_proj_covar=tensor([0.1219, 0.1131, 0.0958, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 18:12:38,383 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:13:05,488 INFO [train.py:968] (0/2) Epoch 23, batch 2350, giga_loss[loss=0.2402, simple_loss=0.3068, pruned_loss=0.08683, over 28709.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3239, pruned_loss=0.08548, over 5720302.48 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3413, pruned_loss=0.08848, over 3856779.54 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3227, pruned_loss=0.08592, over 5706428.26 frames. ], batch size: 66, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:13:25,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.656e+02 1.050e+03 1.335e+03 1.766e+03 5.937e+03, threshold=2.671e+03, percent-clipped=5.0
+2023-03-11 18:13:48,834 INFO [train.py:968] (0/2) Epoch 23, batch 2400, giga_loss[loss=0.2331, simple_loss=0.3108, pruned_loss=0.07774, over 28889.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3217, pruned_loss=0.08436, over 5720756.19 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3417, pruned_loss=0.08848, over 3907348.20 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3201, pruned_loss=0.08461, over 5714179.05 frames. ], batch size: 145, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:14:14,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:23,725 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:24,465 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:25,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:27,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:29,609 INFO [train.py:968] (0/2) Epoch 23, batch 2450, giga_loss[loss=0.2002, simple_loss=0.2822, pruned_loss=0.05914, over 29042.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3196, pruned_loss=0.08354, over 5727475.49 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3416, pruned_loss=0.08829, over 3945284.79 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3179, pruned_loss=0.08379, over 5720503.05 frames. ], batch size: 128, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:14:35,545 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-11 18:14:35,961 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:47,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.096e+02 1.061e+03 1.287e+03 1.604e+03 4.342e+03, threshold=2.573e+03, percent-clipped=5.0
+2023-03-11 18:14:47,514 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:48,104 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:54,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1820, 2.3193, 2.4213, 1.9211], device='cuda:0'), covar=tensor([0.1830, 0.2314, 0.1396, 0.1693], device='cuda:0'), in_proj_covar=tensor([0.0908, 0.0704, 0.0954, 0.0851], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 18:14:58,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8129, 2.1888, 2.0526, 1.5448], device='cuda:0'), covar=tensor([0.3804, 0.2542, 0.2306, 0.3683], device='cuda:0'), in_proj_covar=tensor([0.1968, 0.1894, 0.1829, 0.1972], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 18:15:08,414 INFO [train.py:968] (0/2) Epoch 23, batch 2500, giga_loss[loss=0.2239, simple_loss=0.304, pruned_loss=0.07194, over 28742.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3178, pruned_loss=0.08266, over 5718581.49 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3414, pruned_loss=0.08815, over 4003026.93 frames. ], giga_tot_loss[loss=0.2407, simple_loss=0.3159, pruned_loss=0.0828, over 5715891.97 frames. ], batch size: 262, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:15:17,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0338, 2.1476, 1.5325, 1.8210], device='cuda:0'), covar=tensor([0.1023, 0.0732, 0.1074, 0.1188], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0450, 0.0525, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 18:15:22,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:15:35,727 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:15:49,346 INFO [train.py:968] (0/2) Epoch 23, batch 2550, giga_loss[loss=0.2215, simple_loss=0.3014, pruned_loss=0.07083, over 28889.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3164, pruned_loss=0.08193, over 5719376.30 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3408, pruned_loss=0.08779, over 4058979.04 frames. ], giga_tot_loss[loss=0.2394, simple_loss=0.3146, pruned_loss=0.08212, over 5712667.54 frames. ], batch size: 145, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:16:07,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.542e+02 1.053e+03 1.335e+03 1.884e+03 4.273e+03, threshold=2.670e+03, percent-clipped=6.0
+2023-03-11 18:16:27,983 INFO [train.py:968] (0/2) Epoch 23, batch 2600, giga_loss[loss=0.2673, simple_loss=0.3456, pruned_loss=0.09447, over 29060.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3156, pruned_loss=0.08118, over 5727279.59 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3417, pruned_loss=0.08807, over 4104917.30 frames. ], giga_tot_loss[loss=0.2375, simple_loss=0.313, pruned_loss=0.08099, over 5717519.26 frames. ], batch size: 164, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:16:28,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4777, 4.4486, 1.6453, 1.7487], device='cuda:0'), covar=tensor([0.1067, 0.0311, 0.0945, 0.1345], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0553, 0.0387, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 18:16:28,326 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 18:16:34,975 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1006000.pt
+2023-03-11 18:17:02,180 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1006030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:11,275 INFO [train.py:968] (0/2) Epoch 23, batch 2650, libri_loss[loss=0.2577, simple_loss=0.357, pruned_loss=0.07915, over 29548.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3149, pruned_loss=0.08081, over 5723058.13 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.342, pruned_loss=0.08798, over 4128317.47 frames. ], giga_tot_loss[loss=0.2367, simple_loss=0.3121, pruned_loss=0.08062, over 5716057.68 frames. ], batch size: 89, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:17:21,115 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:23,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:33,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.138e+02 1.091e+03 1.278e+03 1.530e+03 4.370e+03, threshold=2.555e+03, percent-clipped=8.0
+2023-03-11 18:17:33,914 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:35,925 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:50,193 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:57,975 INFO [train.py:968] (0/2) Epoch 23, batch 2700, giga_loss[loss=0.2816, simple_loss=0.3621, pruned_loss=0.1006, over 28594.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3178, pruned_loss=0.08268, over 5720211.17 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3424, pruned_loss=0.0881, over 4163651.14 frames. ], giga_tot_loss[loss=0.2398, simple_loss=0.3149, pruned_loss=0.08233, over 5711310.41 frames. ], batch size: 307, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:18:04,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:18:18,410 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1006112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:18:38,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6245, 1.8444, 1.7398, 1.6742], device='cuda:0'), covar=tensor([0.2159, 0.2222, 0.2493, 0.2180], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0755, 0.0722, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 18:18:42,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3397, 3.4016, 1.4470, 1.5200], device='cuda:0'), covar=tensor([0.1069, 0.0363, 0.0914, 0.1395], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0553, 0.0387, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 18:18:45,943 INFO [train.py:968] (0/2) Epoch 23, batch 2750, libri_loss[loss=0.2674, simple_loss=0.3543, pruned_loss=0.09022, over 29661.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3243, pruned_loss=0.08715, over 5709479.41 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3425, pruned_loss=0.08812, over 4172493.88 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3219, pruned_loss=0.08687, over 5701397.82 frames. ], batch size: 91, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:19:07,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.949e+02 1.251e+03 1.644e+03 1.995e+03 6.525e+03, threshold=3.287e+03, percent-clipped=7.0
+2023-03-11 18:19:22,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6904, 1.7243, 1.8813, 1.4742], device='cuda:0'), covar=tensor([0.1679, 0.2392, 0.1366, 0.1648], device='cuda:0'), in_proj_covar=tensor([0.0912, 0.0708, 0.0959, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 18:19:30,449 INFO [train.py:968] (0/2) Epoch 23, batch 2800, libri_loss[loss=0.2497, simple_loss=0.3445, pruned_loss=0.07739, over 28617.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3324, pruned_loss=0.09243, over 5706204.73 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3425, pruned_loss=0.08788, over 4230504.10 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.33, pruned_loss=0.09245, over 5695054.38 frames. ], batch size: 106, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:19:30,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7781, 1.9258, 1.8344, 1.8042], device='cuda:0'), covar=tensor([0.2079, 0.2197, 0.2509, 0.2046], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0755, 0.0722, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 18:19:33,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:19:40,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1006199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:19:42,058 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3504, 1.8739, 1.3239, 0.6917], device='cuda:0'), covar=tensor([0.5392, 0.2686, 0.3098, 0.5891], device='cuda:0'), in_proj_covar=tensor([0.1749, 0.1653, 0.1594, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 18:19:58,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5271, 1.8213, 1.4701, 1.5747], device='cuda:0'), covar=tensor([0.2592, 0.2676, 0.3067, 0.2410], device='cuda:0'), in_proj_covar=tensor([0.1526, 0.1103, 0.1353, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 18:20:02,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:20:15,309 INFO [train.py:968] (0/2) Epoch 23, batch 2850, giga_loss[loss=0.2759, simple_loss=0.3542, pruned_loss=0.09883, over 28851.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.338, pruned_loss=0.09515, over 5697340.47 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3427, pruned_loss=0.08805, over 4271518.43 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3358, pruned_loss=0.09521, over 5683913.71 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:20:38,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.325e+02 1.221e+03 1.557e+03 2.121e+03 7.447e+03, threshold=3.114e+03, percent-clipped=10.0
+2023-03-11 18:21:03,679 INFO [train.py:968] (0/2) Epoch 23, batch 2900, giga_loss[loss=0.2788, simple_loss=0.3526, pruned_loss=0.1025, over 28573.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3435, pruned_loss=0.09816, over 5667826.38 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3427, pruned_loss=0.08806, over 4302264.97 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3417, pruned_loss=0.0984, over 5660803.17 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:21:48,015 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:21:51,711 INFO [train.py:968] (0/2) Epoch 23, batch 2950, giga_loss[loss=0.3431, simple_loss=0.4046, pruned_loss=0.1408, over 28205.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3486, pruned_loss=0.09996, over 5687034.70 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3431, pruned_loss=0.08832, over 4317970.14 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3469, pruned_loss=0.1001, over 5679744.89 frames. ], batch size: 368, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:21:51,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:22:15,443 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.446e+02 1.241e+03 1.539e+03 2.131e+03 3.642e+03, threshold=3.078e+03, percent-clipped=4.0
+2023-03-11 18:22:18,895 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:22:21,897 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:22:21,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:22:27,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3620, 1.8039, 1.7569, 1.3560], device='cuda:0'), covar=tensor([0.2851, 0.2150, 0.2313, 0.2658], device='cuda:0'), in_proj_covar=tensor([0.1964, 0.1895, 0.1831, 0.1973], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 18:22:37,924 INFO [train.py:968] (0/2) Epoch 23, batch 3000, libri_loss[loss=0.2531, simple_loss=0.336, pruned_loss=0.08508, over 29531.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3536, pruned_loss=0.1034, over 5683427.08 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3435, pruned_loss=0.08879, over 4378264.08 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3523, pruned_loss=0.1037, over 5671168.92 frames. ], batch size: 80, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:22:37,928 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 18:22:45,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2561, 1.8078, 1.3692, 0.4575], device='cuda:0'), covar=tensor([0.4523, 0.3499, 0.4824, 0.6327], device='cuda:0'), in_proj_covar=tensor([0.1751, 0.1652, 0.1599, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 18:22:46,528 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2116, simple_loss=0.3172, pruned_loss=0.05299, over 944034.00 frames. 
+2023-03-11 18:22:46,529 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 18:22:52,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 18:22:54,932 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:22:59,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:23:02,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6723, 1.8837, 1.6206, 1.6193], device='cuda:0'), covar=tensor([0.2284, 0.2046, 0.2114, 0.1964], device='cuda:0'), in_proj_covar=tensor([0.1518, 0.1097, 0.1344, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 18:23:21,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2829, 1.6490, 5.3940, 4.2162], device='cuda:0'), covar=tensor([0.1657, 0.2858, 0.0576, 0.0626], device='cuda:0'), in_proj_covar=tensor([0.0754, 0.0641, 0.0952, 0.0905], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 18:23:26,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3126, 1.6268, 1.4858, 1.2225], device='cuda:0'), covar=tensor([0.2951, 0.2453, 0.1767, 0.2561], device='cuda:0'), in_proj_covar=tensor([0.1962, 0.1894, 0.1828, 0.1972], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 18:23:30,868 INFO [train.py:968] (0/2) Epoch 23, batch 3050, giga_loss[loss=0.2394, simple_loss=0.3251, pruned_loss=0.07689, over 28694.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3492, pruned_loss=0.1001, over 5690003.28 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3432, pruned_loss=0.08867, over 4393451.30 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3485, pruned_loss=0.1005, over 5678478.24 frames. ], batch size: 262, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:23:43,211 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1006452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:23:57,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.867e+02 1.311e+03 1.722e+03 2.433e+03 4.500e+03, threshold=3.444e+03, percent-clipped=12.0
+2023-03-11 18:24:15,171 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:24:17,510 INFO [train.py:968] (0/2) Epoch 23, batch 3100, giga_loss[loss=0.2356, simple_loss=0.3237, pruned_loss=0.07372, over 28980.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.347, pruned_loss=0.09823, over 5687623.66 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3437, pruned_loss=0.08895, over 4408216.19 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3462, pruned_loss=0.09846, over 5676826.40 frames. ], batch size: 164, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:25:00,179 INFO [train.py:968] (0/2) Epoch 23, batch 3150, giga_loss[loss=0.2478, simple_loss=0.3289, pruned_loss=0.08332, over 28801.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3462, pruned_loss=0.09724, over 5683591.84 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08885, over 4454324.71 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3458, pruned_loss=0.09779, over 5672589.33 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:25:08,818 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:25:11,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:25:24,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.731e+02 1.280e+03 1.624e+03 2.424e+03 4.316e+03, threshold=3.249e+03, percent-clipped=4.0
+2023-03-11 18:25:30,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:25:34,643 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:25:46,236 INFO [train.py:968] (0/2) Epoch 23, batch 3200, giga_loss[loss=0.3168, simple_loss=0.3754, pruned_loss=0.1291, over 26688.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3472, pruned_loss=0.09754, over 5680143.13 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3432, pruned_loss=0.0888, over 4475545.35 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.347, pruned_loss=0.09813, over 5668813.16 frames. ], batch size: 555, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:26:18,310 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:26:20,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:26:21,238 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-11 18:26:26,559 INFO [train.py:968] (0/2) Epoch 23, batch 3250, giga_loss[loss=0.2977, simple_loss=0.3604, pruned_loss=0.1175, over 28878.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3486, pruned_loss=0.09833, over 5694856.88 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3426, pruned_loss=0.08835, over 4524314.93 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3492, pruned_loss=0.09947, over 5679104.56 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:26:27,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9145, 3.7455, 3.5133, 1.8225], device='cuda:0'), covar=tensor([0.0712, 0.0811, 0.0791, 0.2112], device='cuda:0'), in_proj_covar=tensor([0.1220, 0.1128, 0.0957, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 18:26:46,395 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:26:50,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.010e+02 1.312e+03 1.739e+03 2.561e+03 5.562e+03, threshold=3.478e+03, percent-clipped=16.0
+2023-03-11 18:27:01,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5509, 1.6713, 1.5726, 1.5079], device='cuda:0'), covar=tensor([0.1977, 0.1973, 0.1757, 0.1740], device='cuda:0'), in_proj_covar=tensor([0.1968, 0.1900, 0.1835, 0.1976], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 18:27:13,577 INFO [train.py:968] (0/2) Epoch 23, batch 3300, giga_loss[loss=0.3587, simple_loss=0.401, pruned_loss=0.1581, over 26613.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3516, pruned_loss=0.1011, over 5680709.10 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3427, pruned_loss=0.08847, over 4534811.70 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.352, pruned_loss=0.102, over 5674986.59 frames. ], batch size: 555, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:27:14,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-11 18:27:37,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:27:39,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:27:57,016 INFO [train.py:968] (0/2) Epoch 23, batch 3350, giga_loss[loss=0.257, simple_loss=0.3395, pruned_loss=0.0872, over 28837.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3521, pruned_loss=0.1018, over 5689635.05 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3427, pruned_loss=0.08847, over 4548285.29 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3526, pruned_loss=0.1027, over 5683296.01 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:28:04,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:28:21,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.600e+02 1.454e+03 1.946e+03 2.881e+03 5.875e+03, threshold=3.891e+03, percent-clipped=13.0
+2023-03-11 18:28:23,447 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.90 vs. limit=2.0
+2023-03-11 18:28:39,696 INFO [train.py:968] (0/2) Epoch 23, batch 3400, giga_loss[loss=0.2621, simple_loss=0.3423, pruned_loss=0.09091, over 29001.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3526, pruned_loss=0.1026, over 5683011.42 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3425, pruned_loss=0.08818, over 4584515.84 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3534, pruned_loss=0.1039, over 5679943.09 frames. ], batch size: 136, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:29:09,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:29:21,744 INFO [train.py:968] (0/2) Epoch 23, batch 3450, giga_loss[loss=0.2731, simple_loss=0.3497, pruned_loss=0.09826, over 28579.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.354, pruned_loss=0.1033, over 5681588.58 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3435, pruned_loss=0.08876, over 4620962.84 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3542, pruned_loss=0.1043, over 5675727.52 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:29:46,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.474e+02 1.188e+03 1.488e+03 2.044e+03 7.626e+03, threshold=2.977e+03, percent-clipped=1.0
+2023-03-11 18:29:57,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9373, 1.1715, 1.0853, 0.8344], device='cuda:0'), covar=tensor([0.2250, 0.2682, 0.1842, 0.2497], device='cuda:0'), in_proj_covar=tensor([0.1980, 0.1909, 0.1845, 0.1987], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 18:30:04,523 INFO [train.py:968] (0/2) Epoch 23, batch 3500, giga_loss[loss=0.2467, simple_loss=0.3371, pruned_loss=0.07817, over 28934.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.353, pruned_loss=0.1014, over 5690719.55 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3433, pruned_loss=0.08857, over 4643224.86 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3536, pruned_loss=0.1026, over 5685156.72 frames. ], batch size: 213, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:30:07,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1006895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:30:11,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4272, 3.1509, 1.4734, 1.5531], device='cuda:0'), covar=tensor([0.0951, 0.0265, 0.0907, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0552, 0.0387, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 18:30:37,930 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5925, 1.9489, 1.3159, 1.4710], device='cuda:0'), covar=tensor([0.0999, 0.0528, 0.1050, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0446, 0.0522, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 18:30:48,359 INFO [train.py:968] (0/2) Epoch 23, batch 3550, giga_loss[loss=0.2664, simple_loss=0.3459, pruned_loss=0.09349, over 28319.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3525, pruned_loss=0.1003, over 5693991.41 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3431, pruned_loss=0.08868, over 4668405.49 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3533, pruned_loss=0.1014, over 5685653.93 frames. ], batch size: 368, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:31:13,719 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.783e+02 1.125e+03 1.319e+03 1.816e+03 5.659e+03, threshold=2.637e+03, percent-clipped=6.0
+2023-03-11 18:31:15,596 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:31:18,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:31:26,637 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 18:31:29,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4717, 1.6266, 1.2031, 1.2433], device='cuda:0'), covar=tensor([0.0989, 0.0575, 0.1042, 0.1175], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0447, 0.0522, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 18:31:34,217 INFO [train.py:968] (0/2) Epoch 23, batch 3600, giga_loss[loss=0.3134, simple_loss=0.3912, pruned_loss=0.1178, over 28983.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3518, pruned_loss=0.09991, over 5698323.38 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3431, pruned_loss=0.08868, over 4668405.49 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3524, pruned_loss=0.1008, over 5691834.21 frames. ], batch size: 145, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:31:42,991 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1007002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:32:15,663 INFO [train.py:968] (0/2) Epoch 23, batch 3650, giga_loss[loss=0.2648, simple_loss=0.3378, pruned_loss=0.09591, over 28789.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.35, pruned_loss=0.09926, over 5692185.59 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3434, pruned_loss=0.08876, over 4696939.96 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3505, pruned_loss=0.1002, over 5684214.32 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:32:38,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.294e+02 1.150e+03 1.470e+03 1.729e+03 4.337e+03, threshold=2.940e+03, percent-clipped=8.0
+2023-03-11 18:32:55,234 INFO [train.py:968] (0/2) Epoch 23, batch 3700, giga_loss[loss=0.2507, simple_loss=0.3302, pruned_loss=0.08554, over 29063.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3466, pruned_loss=0.09707, over 5702645.26 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08889, over 4728826.82 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3471, pruned_loss=0.09802, over 5698044.00 frames. ], batch size: 128, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:33:28,680 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4830, 1.7019, 1.6827, 1.4951], device='cuda:0'), covar=tensor([0.2047, 0.2351, 0.2392, 0.2438], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0749, 0.0719, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 18:33:36,001 INFO [train.py:968] (0/2) Epoch 23, batch 3750, giga_loss[loss=0.2811, simple_loss=0.356, pruned_loss=0.1031, over 28876.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3453, pruned_loss=0.0965, over 5700425.55 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.343, pruned_loss=0.08859, over 4754449.17 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3461, pruned_loss=0.09763, over 5695193.88 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:33:48,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1007151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:34:00,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.960e+02 1.148e+03 1.428e+03 1.960e+03 5.743e+03, threshold=2.857e+03, percent-clipped=7.0
+2023-03-11 18:34:05,451 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1007174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:34:20,733 INFO [train.py:968] (0/2) Epoch 23, batch 3800, giga_loss[loss=0.2736, simple_loss=0.3479, pruned_loss=0.09964, over 28577.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3468, pruned_loss=0.09789, over 5698309.25 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3429, pruned_loss=0.0885, over 4766054.29 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3475, pruned_loss=0.09895, over 5692098.94 frames. ], batch size: 78, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:34:34,463 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9605, 4.9950, 2.1711, 2.2663], device='cuda:0'), covar=tensor([0.0944, 0.0208, 0.0832, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0548, 0.0385, 0.0426], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 18:35:01,611 INFO [train.py:968] (0/2) Epoch 23, batch 3850, libri_loss[loss=0.2557, simple_loss=0.3395, pruned_loss=0.08596, over 29571.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3469, pruned_loss=0.09716, over 5705496.29 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3433, pruned_loss=0.08881, over 4788555.98 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3472, pruned_loss=0.09793, over 5696814.55 frames. ], batch size: 77, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:35:05,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3153, 3.0863, 1.3849, 1.4635], device='cuda:0'), covar=tensor([0.1068, 0.0330, 0.0931, 0.1427], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0549, 0.0386, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:0')
+2023-03-11 18:35:23,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.299e+02 1.078e+03 1.301e+03 1.661e+03 4.723e+03, threshold=2.603e+03, percent-clipped=6.0
+2023-03-11 18:35:24,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1007270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:35:44,381 INFO [train.py:968] (0/2) Epoch 23, batch 3900, giga_loss[loss=0.2703, simple_loss=0.3449, pruned_loss=0.09792, over 28968.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3455, pruned_loss=0.09555, over 5715567.03 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08887, over 4825648.91 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3458, pruned_loss=0.09636, over 5703899.89 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:36:01,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4483, 2.0808, 1.6234, 0.7130], device='cuda:0'), covar=tensor([0.5789, 0.2929, 0.4197, 0.6473], device='cuda:0'), in_proj_covar=tensor([0.1735, 0.1628, 0.1584, 0.1418], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 18:36:25,906 INFO [train.py:968] (0/2) Epoch 23, batch 3950, giga_loss[loss=0.2395, simple_loss=0.3223, pruned_loss=0.07835, over 28482.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.344, pruned_loss=0.09483, over 5708455.03 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3424, pruned_loss=0.08841, over 4846753.77 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.345, pruned_loss=0.09592, over 5696154.30 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:36:49,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.279e+02 1.193e+03 1.439e+03 1.785e+03 7.617e+03, threshold=2.877e+03, percent-clipped=11.0
+2023-03-11 18:37:06,403 INFO [train.py:968] (0/2) Epoch 23, batch 4000, giga_loss[loss=0.2502, simple_loss=0.3307, pruned_loss=0.08487, over 29113.00 frames. ], tot_loss[loss=0.267, simple_loss=0.344, pruned_loss=0.09501, over 5717797.85 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3426, pruned_loss=0.08833, over 4886945.15 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3448, pruned_loss=0.09624, over 5702070.31 frames. ], batch size: 128, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:37:12,144 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4452, 1.7062, 1.3556, 1.6855], device='cuda:0'), covar=tensor([0.2617, 0.2637, 0.2993, 0.2355], device='cuda:0'), in_proj_covar=tensor([0.1519, 0.1100, 0.1345, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 18:37:23,984 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1007413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:37:26,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1007416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:37:45,172 INFO [train.py:968] (0/2) Epoch 23, batch 4050, giga_loss[loss=0.2461, simple_loss=0.3278, pruned_loss=0.08216, over 28620.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.341, pruned_loss=0.09332, over 5714289.61 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3412, pruned_loss=0.08756, over 4926201.50 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3427, pruned_loss=0.09519, over 5702926.40 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:37:47,936 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1007445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:38:07,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.095e+02 1.179e+03 1.492e+03 1.864e+03 6.436e+03, threshold=2.984e+03, percent-clipped=9.0
+2023-03-11 18:38:23,007 INFO [train.py:968] (0/2) Epoch 23, batch 4100, giga_loss[loss=0.2835, simple_loss=0.3646, pruned_loss=0.1012, over 28341.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3377, pruned_loss=0.09153, over 5722666.52 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3403, pruned_loss=0.08708, over 4950646.94 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3397, pruned_loss=0.0935, over 5709281.95 frames. ], batch size: 368, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:38:27,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0659, 2.1846, 1.9063, 2.4364], device='cuda:0'), covar=tensor([0.2430, 0.2626, 0.2820, 0.2264], device='cuda:0'), in_proj_covar=tensor([0.1522, 0.1101, 0.1347, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 18:38:52,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1007526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:39:05,498 INFO [train.py:968] (0/2) Epoch 23, batch 4150, giga_loss[loss=0.2402, simple_loss=0.3221, pruned_loss=0.07917, over 29088.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3371, pruned_loss=0.09156, over 5713807.09 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3402, pruned_loss=0.08703, over 4961510.03 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3387, pruned_loss=0.09321, over 5703872.35 frames. ], batch size: 155, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:39:12,886 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1007549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:39:12,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1007549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:39:20,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4654, 1.9057, 1.4791, 1.6076], device='cuda:0'), covar=tensor([0.0721, 0.0283, 0.0331, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 18:39:29,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.588e+02 1.316e+03 1.657e+03 2.536e+03 6.187e+03, threshold=3.314e+03, percent-clipped=10.0
+2023-03-11 18:39:30,625 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-11 18:39:45,752 INFO [train.py:968] (0/2) Epoch 23, batch 4200, giga_loss[loss=0.2511, simple_loss=0.3273, pruned_loss=0.08745, over 28776.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3362, pruned_loss=0.09157, over 5716814.55 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3401, pruned_loss=0.0873, over 4992502.79 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3374, pruned_loss=0.09282, over 5704252.14 frames. ], batch size: 284, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:40:21,789 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1007630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:40:24,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9669, 1.2799, 1.0651, 0.2320], device='cuda:0'), covar=tensor([0.4106, 0.3337, 0.4799, 0.6773], device='cuda:0'), in_proj_covar=tensor([0.1745, 0.1639, 0.1595, 0.1428], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 18:40:29,509 INFO [train.py:968] (0/2) Epoch 23, batch 4250, giga_loss[loss=0.2423, simple_loss=0.3173, pruned_loss=0.0837, over 28646.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3349, pruned_loss=0.09137, over 5714210.87 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3407, pruned_loss=0.08763, over 4996133.62 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3354, pruned_loss=0.09212, over 5707703.74 frames. ], batch size: 262, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:40:53,725 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1007669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:40:54,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.001e+02 1.141e+03 1.463e+03 1.910e+03 5.542e+03, threshold=2.927e+03, percent-clipped=7.0
+2023-03-11 18:40:55,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1007672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:40:59,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9612, 2.1145, 1.4861, 1.6941], device='cuda:0'), covar=tensor([0.0963, 0.0758, 0.1097, 0.1245], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0447, 0.0524, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 18:41:12,767 INFO [train.py:968] (0/2) Epoch 23, batch 4300, giga_loss[loss=0.2196, simple_loss=0.2975, pruned_loss=0.07081, over 28169.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3324, pruned_loss=0.09048, over 5701604.26 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3403, pruned_loss=0.08751, over 4999449.30 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.333, pruned_loss=0.09121, over 5702728.28 frames. ], batch size: 77, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:41:13,611 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1007692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:41:15,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1007695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:41:20,335 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1007701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:41:37,732 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1007724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:41:51,321 INFO [train.py:968] (0/2) Epoch 23, batch 4350, giga_loss[loss=0.305, simple_loss=0.3622, pruned_loss=0.1239, over 26805.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3305, pruned_loss=0.08987, over 5705561.37 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3406, pruned_loss=0.08759, over 5025617.64 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3306, pruned_loss=0.09045, over 5701749.68 frames. ], batch size: 555, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:42:15,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.674e+02 1.137e+03 1.389e+03 2.002e+03 6.360e+03, threshold=2.777e+03, percent-clipped=10.0
+2023-03-11 18:42:31,417 INFO [train.py:968] (0/2) Epoch 23, batch 4400, giga_loss[loss=0.2616, simple_loss=0.3374, pruned_loss=0.09294, over 28931.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3306, pruned_loss=0.08991, over 5709851.67 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3409, pruned_loss=0.08788, over 5042379.15 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3301, pruned_loss=0.0902, over 5703977.66 frames. ], batch size: 186, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:43:15,604 INFO [train.py:968] (0/2) Epoch 23, batch 4450, giga_loss[loss=0.2534, simple_loss=0.3413, pruned_loss=0.08274, over 28980.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3326, pruned_loss=0.09085, over 5706136.55 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3412, pruned_loss=0.08799, over 5051876.29 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.332, pruned_loss=0.09102, over 5701752.40 frames. ], batch size: 164, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:43:44,591 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.44 vs. limit=5.0
+2023-03-11 18:43:45,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.961e+02 1.043e+03 1.264e+03 1.574e+03 3.899e+03, threshold=2.528e+03, percent-clipped=4.0
+2023-03-11 18:44:03,561 INFO [train.py:968] (0/2) Epoch 23, batch 4500, giga_loss[loss=0.2971, simple_loss=0.3764, pruned_loss=0.1089, over 28717.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3345, pruned_loss=0.09089, over 5715418.70 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3413, pruned_loss=0.08811, over 5056144.91 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3338, pruned_loss=0.09093, over 5711092.78 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:44:34,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1007924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:44:50,118 INFO [train.py:968] (0/2) Epoch 23, batch 4550, giga_loss[loss=0.2424, simple_loss=0.3166, pruned_loss=0.08414, over 28713.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3369, pruned_loss=0.09162, over 5709691.07 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3415, pruned_loss=0.08819, over 5060055.18 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3362, pruned_loss=0.09161, over 5705768.39 frames. ], batch size: 99, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:44:54,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7823, 2.5771, 1.6400, 1.0730], device='cuda:0'), covar=tensor([0.9305, 0.3866, 0.4706, 0.7525], device='cuda:0'), in_proj_covar=tensor([0.1745, 0.1636, 0.1594, 0.1426], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 18:45:16,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.916e+02 1.129e+03 1.373e+03 1.800e+03 5.663e+03, threshold=2.747e+03, percent-clipped=9.0
+2023-03-11 18:45:35,151 INFO [train.py:968] (0/2) Epoch 23, batch 4600, giga_loss[loss=0.2397, simple_loss=0.3064, pruned_loss=0.08653, over 23905.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3375, pruned_loss=0.0914, over 5702239.82 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3412, pruned_loss=0.08805, over 5087478.07 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3371, pruned_loss=0.09164, over 5694362.76 frames. ], batch size: 705, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:45:42,660 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1008000.pt
+2023-03-11 18:45:46,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1008005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:46:03,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1343, 1.5982, 1.3323, 1.3326], device='cuda:0'), covar=tensor([0.2317, 0.1912, 0.2392, 0.2377], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0742, 0.0710, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 18:46:17,756 INFO [train.py:968] (0/2) Epoch 23, batch 4650, giga_loss[loss=0.2515, simple_loss=0.3323, pruned_loss=0.08538, over 28946.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3368, pruned_loss=0.09058, over 5700876.78 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3411, pruned_loss=0.08796, over 5098831.68 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3365, pruned_loss=0.09086, over 5692630.03 frames. ], batch size: 136, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:46:39,906 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1008067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:46:42,610 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1008070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:46:43,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.072e+02 1.178e+03 1.436e+03 2.010e+03 6.506e+03, threshold=2.872e+03, percent-clipped=12.0
+2023-03-11 18:46:59,984 INFO [train.py:968] (0/2) Epoch 23, batch 4700, giga_loss[loss=0.2367, simple_loss=0.3165, pruned_loss=0.07845, over 29052.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3383, pruned_loss=0.09122, over 5706727.44 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3412, pruned_loss=0.08798, over 5113201.80 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3379, pruned_loss=0.09146, over 5698176.98 frames. ], batch size: 106, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:47:07,639 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1008099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:47:10,993 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-11 18:47:38,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6497, 1.7594, 1.5076, 1.6262], device='cuda:0'), covar=tensor([0.2508, 0.2571, 0.2824, 0.2458], device='cuda:0'), in_proj_covar=tensor([0.1527, 0.1103, 0.1347, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 18:47:45,481 INFO [train.py:968] (0/2) Epoch 23, batch 4750, giga_loss[loss=0.2878, simple_loss=0.3631, pruned_loss=0.1063, over 27985.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3389, pruned_loss=0.09181, over 5712032.13 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3411, pruned_loss=0.08789, over 5121167.80 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3386, pruned_loss=0.09212, over 5703449.80 frames. ], batch size: 412, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:47:50,686 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1008148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:47:53,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1008151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:48:09,010 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.951e+02 1.338e+03 1.657e+03 2.329e+03 7.355e+03, threshold=3.315e+03, percent-clipped=16.0
+2023-03-11 18:48:13,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2056, 1.5251, 5.5509, 3.9446], device='cuda:0'), covar=tensor([0.1392, 0.2644, 0.0374, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0760, 0.0648, 0.0961, 0.0914], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 18:48:15,129 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-11 18:48:18,706 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1008180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:48:27,048 INFO [train.py:968] (0/2) Epoch 23, batch 4800, giga_loss[loss=0.2907, simple_loss=0.3651, pruned_loss=0.1082, over 28827.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3409, pruned_loss=0.09336, over 5715710.83 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3414, pruned_loss=0.08804, over 5157504.65 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3405, pruned_loss=0.09374, over 5701700.16 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:49:07,849 INFO [train.py:968] (0/2) Epoch 23, batch 4850, giga_loss[loss=0.2446, simple_loss=0.3328, pruned_loss=0.07822, over 28967.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3435, pruned_loss=0.09459, over 5720066.50 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3418, pruned_loss=0.08821, over 5172011.86 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3429, pruned_loss=0.09485, over 5706011.98 frames. ], batch size: 174, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:49:32,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.397e+02 1.265e+03 1.510e+03 2.020e+03 5.206e+03, threshold=3.019e+03, percent-clipped=4.0
+2023-03-11 18:49:49,130 INFO [train.py:968] (0/2) Epoch 23, batch 4900, giga_loss[loss=0.2553, simple_loss=0.3317, pruned_loss=0.08947, over 28612.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3459, pruned_loss=0.09579, over 5713125.44 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3414, pruned_loss=0.08806, over 5182699.92 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3457, pruned_loss=0.09632, over 5706796.68 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:50:30,606 INFO [train.py:968] (0/2) Epoch 23, batch 4950, giga_loss[loss=0.2887, simple_loss=0.3661, pruned_loss=0.1057, over 28663.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3479, pruned_loss=0.09688, over 5713570.62 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.342, pruned_loss=0.08842, over 5197299.04 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3474, pruned_loss=0.09719, over 5706851.55 frames. ], batch size: 307, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:50:48,850 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1008365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:50:54,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.878e+02 1.371e+03 1.645e+03 2.176e+03 6.613e+03, threshold=3.290e+03, percent-clipped=9.0
+2023-03-11 18:51:11,015 INFO [train.py:968] (0/2) Epoch 23, batch 5000, giga_loss[loss=0.2427, simple_loss=0.3244, pruned_loss=0.08052, over 28754.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3496, pruned_loss=0.09841, over 5710039.09 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3425, pruned_loss=0.08863, over 5207619.74 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3489, pruned_loss=0.09861, over 5702532.45 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:51:28,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2306, 2.5804, 1.2776, 1.3295], device='cuda:0'), covar=tensor([0.0994, 0.0387, 0.0948, 0.1384], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0553, 0.0387, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 18:51:50,480 INFO [train.py:968] (0/2) Epoch 23, batch 5050, giga_loss[loss=0.2512, simple_loss=0.3359, pruned_loss=0.0833, over 28942.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3496, pruned_loss=0.09818, over 5709044.08 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3428, pruned_loss=0.0887, over 5221106.74 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3491, pruned_loss=0.09855, over 5704344.14 frames. ], batch size: 128, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:52:14,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.474e+02 1.351e+03 1.683e+03 2.384e+03 6.555e+03, threshold=3.367e+03, percent-clipped=13.0
+2023-03-11 18:52:30,437 INFO [train.py:968] (0/2) Epoch 23, batch 5100, libri_loss[loss=0.2211, simple_loss=0.3069, pruned_loss=0.06764, over 29567.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3474, pruned_loss=0.09705, over 5713159.95 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3423, pruned_loss=0.08845, over 5245798.62 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3476, pruned_loss=0.0979, over 5704412.65 frames. ], batch size: 75, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:52:48,929 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-11 18:53:07,420 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4096, 2.7575, 1.5095, 1.4811], device='cuda:0'), covar=tensor([0.0897, 0.0360, 0.0895, 0.1267], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0554, 0.0388, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 18:53:10,913 INFO [train.py:968] (0/2) Epoch 23, batch 5150, giga_loss[loss=0.2601, simple_loss=0.3343, pruned_loss=0.09299, over 28920.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3439, pruned_loss=0.09561, over 5705961.74 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3423, pruned_loss=0.08845, over 5257261.25 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3442, pruned_loss=0.09652, over 5700595.12 frames. ], batch size: 227, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:53:19,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3574, 1.6466, 1.3206, 1.0906], device='cuda:0'), covar=tensor([0.2603, 0.2718, 0.3147, 0.2430], device='cuda:0'), in_proj_covar=tensor([0.1519, 0.1097, 0.1342, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 18:53:37,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.490e+02 1.116e+03 1.389e+03 1.723e+03 3.277e+03, threshold=2.779e+03, percent-clipped=0.0
+2023-03-11 18:53:52,283 INFO [train.py:968] (0/2) Epoch 23, batch 5200, giga_loss[loss=0.275, simple_loss=0.3506, pruned_loss=0.09972, over 28787.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3412, pruned_loss=0.09427, over 5712501.74 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3425, pruned_loss=0.0886, over 5269932.19 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3413, pruned_loss=0.09497, over 5704769.44 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:54:28,266 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1008632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:54:36,307 INFO [train.py:968] (0/2) Epoch 23, batch 5250, giga_loss[loss=0.3205, simple_loss=0.4022, pruned_loss=0.1194, over 28803.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3427, pruned_loss=0.09431, over 5710856.40 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3431, pruned_loss=0.08903, over 5278171.36 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3423, pruned_loss=0.09459, over 5703362.02 frames. ], batch size: 285, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:55:04,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.137e+02 1.242e+03 1.672e+03 2.710e+03 5.256e+03, threshold=3.343e+03, percent-clipped=22.0
+2023-03-11 18:55:12,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5402, 1.7745, 1.4210, 1.5595], device='cuda:0'), covar=tensor([0.2777, 0.2848, 0.3301, 0.2566], device='cuda:0'), in_proj_covar=tensor([0.1520, 0.1099, 0.1344, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 18:55:14,020 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1008682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:55:21,031 INFO [train.py:968] (0/2) Epoch 23, batch 5300, giga_loss[loss=0.2622, simple_loss=0.3476, pruned_loss=0.08846, over 28018.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3444, pruned_loss=0.09383, over 5715292.47 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3432, pruned_loss=0.08909, over 5290796.01 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3439, pruned_loss=0.09412, over 5705819.99 frames. ], batch size: 412, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:56:03,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1008740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:56:03,861 INFO [train.py:968] (0/2) Epoch 23, batch 5350, giga_loss[loss=0.2427, simple_loss=0.3189, pruned_loss=0.08331, over 28869.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3446, pruned_loss=0.09444, over 5719841.50 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3438, pruned_loss=0.08938, over 5305736.28 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3438, pruned_loss=0.09458, over 5710630.61 frames. ], batch size: 112, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:56:24,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4012, 1.7382, 1.3561, 1.0201], device='cuda:0'), covar=tensor([0.2608, 0.2758, 0.3102, 0.2459], device='cuda:0'), in_proj_covar=tensor([0.1516, 0.1096, 0.1340, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 18:56:28,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.482e+02 1.190e+03 1.599e+03 2.113e+03 6.257e+03, threshold=3.197e+03, percent-clipped=3.0
+2023-03-11 18:56:35,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1008779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:56:45,939 INFO [train.py:968] (0/2) Epoch 23, batch 5400, giga_loss[loss=0.2535, simple_loss=0.3335, pruned_loss=0.08677, over 28987.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3424, pruned_loss=0.09429, over 5724522.63 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3441, pruned_loss=0.08952, over 5315349.48 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3416, pruned_loss=0.09438, over 5715966.46 frames. ], batch size: 213, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:57:29,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4953, 1.9882, 1.4855, 1.7500], device='cuda:0'), covar=tensor([0.0727, 0.0265, 0.0316, 0.0800], device='cuda:0'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:0')
+2023-03-11 18:57:29,451 INFO [train.py:968] (0/2) Epoch 23, batch 5450, giga_loss[loss=0.2263, simple_loss=0.306, pruned_loss=0.07327, over 29013.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3403, pruned_loss=0.09455, over 5730314.16 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.344, pruned_loss=0.08945, over 5323591.19 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3398, pruned_loss=0.09476, over 5721604.24 frames. ], batch size: 164, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:57:52,756 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-11 18:57:57,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.975e+02 1.218e+03 1.489e+03 1.910e+03 6.734e+03, threshold=2.977e+03, percent-clipped=5.0
+2023-03-11 18:58:06,313 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1008883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:58:09,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1008886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:58:12,056 INFO [train.py:968] (0/2) Epoch 23, batch 5500, giga_loss[loss=0.2903, simple_loss=0.3468, pruned_loss=0.1169, over 28797.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3387, pruned_loss=0.09489, over 5733867.74 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.0897, over 5338081.22 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3379, pruned_loss=0.09496, over 5722901.30 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:58:33,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1008915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:58:56,824 INFO [train.py:968] (0/2) Epoch 23, batch 5550, giga_loss[loss=0.3153, simple_loss=0.3751, pruned_loss=0.1278, over 23986.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3386, pruned_loss=0.09524, over 5713561.86 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3446, pruned_loss=0.08998, over 5336061.73 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3375, pruned_loss=0.09513, over 5714473.32 frames. ], batch size: 705, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:58:58,838 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-11 18:59:22,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.369e+02 1.242e+03 1.494e+03 2.044e+03 4.118e+03, threshold=2.987e+03, percent-clipped=6.0
+2023-03-11 18:59:40,252 INFO [train.py:968] (0/2) Epoch 23, batch 5600, giga_loss[loss=0.2343, simple_loss=0.3088, pruned_loss=0.07995, over 28887.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3346, pruned_loss=0.09326, over 5713628.58 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3442, pruned_loss=0.0898, over 5344718.38 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3339, pruned_loss=0.09338, over 5711736.68 frames. ], batch size: 145, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:59:54,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:00:17,966 INFO [train.py:968] (0/2) Epoch 23, batch 5650, giga_loss[loss=0.2222, simple_loss=0.2924, pruned_loss=0.07601, over 28613.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3328, pruned_loss=0.09239, over 5711379.75 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3446, pruned_loss=0.09009, over 5362094.60 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3315, pruned_loss=0.09241, over 5711604.25 frames. ], batch size: 78, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 19:00:30,489 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:00:31,968 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-11 19:00:33,526 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 19:00:41,506 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:00:42,048 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.000e+02 1.375e+03 1.734e+03 2.688e+03 5.233e+03, threshold=3.468e+03, percent-clipped=18.0
+2023-03-11 19:00:49,667 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 19:00:56,532 INFO [train.py:968] (0/2) Epoch 23, batch 5700, giga_loss[loss=0.2141, simple_loss=0.2989, pruned_loss=0.06465, over 29041.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3293, pruned_loss=0.0906, over 5711217.11 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3441, pruned_loss=0.08988, over 5378999.79 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3282, pruned_loss=0.09083, over 5706975.61 frames. ], batch size: 164, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 19:01:37,434 INFO [train.py:968] (0/2) Epoch 23, batch 5750, giga_loss[loss=0.2656, simple_loss=0.3404, pruned_loss=0.09538, over 28729.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3276, pruned_loss=0.08937, over 5715500.06 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3431, pruned_loss=0.08933, over 5394236.62 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3271, pruned_loss=0.09002, over 5707286.50 frames. ], batch size: 262, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 19:01:44,222 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:01:45,982 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009152.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:01:46,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:01:47,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:02:02,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.108e+02 1.305e+03 1.576e+03 2.214e+03 4.668e+03, threshold=3.153e+03, percent-clipped=6.0
+2023-03-11 19:02:10,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:02:12,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4699, 1.6064, 1.1681, 1.2115], device='cuda:0'), covar=tensor([0.0900, 0.0563, 0.1051, 0.1278], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0445, 0.0518, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 19:02:16,771 INFO [train.py:968] (0/2) Epoch 23, batch 5800, giga_loss[loss=0.2803, simple_loss=0.3541, pruned_loss=0.1032, over 28852.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3306, pruned_loss=0.09068, over 5714836.74 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.343, pruned_loss=0.08945, over 5410211.79 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3297, pruned_loss=0.09111, over 5705181.80 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 19:02:25,927 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:02:28,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:02:36,782 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3210, 3.3462, 1.4287, 1.4949], device='cuda:0'), covar=tensor([0.0993, 0.0355, 0.0944, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0554, 0.0388, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 19:02:51,142 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:02:58,563 INFO [train.py:968] (0/2) Epoch 23, batch 5850, libri_loss[loss=0.2294, simple_loss=0.314, pruned_loss=0.07243, over 29586.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3345, pruned_loss=0.0921, over 5717028.62 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3429, pruned_loss=0.08936, over 5426995.34 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3335, pruned_loss=0.09261, over 5703466.19 frames. ], batch size: 74, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:03:03,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4950, 1.7285, 1.7651, 1.3141], device='cuda:0'), covar=tensor([0.1882, 0.2452, 0.1537, 0.1701], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0701, 0.0950, 0.0849], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 19:03:25,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.575e+02 1.319e+03 1.663e+03 2.249e+03 4.961e+03, threshold=3.326e+03, percent-clipped=7.0
+2023-03-11 19:03:41,421 INFO [train.py:968] (0/2) Epoch 23, batch 5900, giga_loss[loss=0.2611, simple_loss=0.3354, pruned_loss=0.09337, over 28495.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3376, pruned_loss=0.09316, over 5720128.00 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3429, pruned_loss=0.08932, over 5431881.91 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3369, pruned_loss=0.09363, over 5707652.55 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:03:49,688 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:03:51,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:04:13,312 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-11 19:04:15,009 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:04:25,488 INFO [train.py:968] (0/2) Epoch 23, batch 5950, giga_loss[loss=0.2561, simple_loss=0.3417, pruned_loss=0.08526, over 29009.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.09475, over 5724042.90 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08943, over 5452076.54 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3404, pruned_loss=0.09523, over 5706966.96 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:04:29,814 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:04:54,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.020e+02 1.325e+03 1.935e+03 2.755e+03 8.783e+03, threshold=3.870e+03, percent-clipped=10.0
+2023-03-11 19:05:10,526 INFO [train.py:968] (0/2) Epoch 23, batch 6000, giga_loss[loss=0.2759, simple_loss=0.3554, pruned_loss=0.09821, over 28658.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3444, pruned_loss=0.09707, over 5716608.48 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08941, over 5453280.04 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3439, pruned_loss=0.09751, over 5704385.63 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:05:10,531 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 19:05:18,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4947, 1.6966, 1.3193, 1.3777], device='cuda:0'), covar=tensor([0.0962, 0.0434, 0.0948, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0446, 0.0520, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 19:05:19,129 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2101, simple_loss=0.3173, pruned_loss=0.05143, over 944034.00 frames. 
+2023-03-11 19:05:19,130 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 19:06:04,422 INFO [train.py:968] (0/2) Epoch 23, batch 6050, giga_loss[loss=0.3468, simple_loss=0.4061, pruned_loss=0.1438, over 28028.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.35, pruned_loss=0.1019, over 5710113.51 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08945, over 5464074.91 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3494, pruned_loss=0.1025, over 5696417.69 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:06:04,568 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:06:10,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:06:38,045 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.780e+02 1.594e+03 1.937e+03 2.699e+03 6.563e+03, threshold=3.875e+03, percent-clipped=7.0
+2023-03-11 19:06:54,318 INFO [train.py:968] (0/2) Epoch 23, batch 6100, giga_loss[loss=0.3089, simple_loss=0.385, pruned_loss=0.1164, over 28919.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3566, pruned_loss=0.1073, over 5707404.89 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08926, over 5475124.14 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3566, pruned_loss=0.1083, over 5692128.81 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:07:04,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3892, 1.5542, 1.4885, 1.3520], device='cuda:0'), covar=tensor([0.2347, 0.1961, 0.1927, 0.2083], device='cuda:0'), in_proj_covar=tensor([0.1974, 0.1923, 0.1852, 0.1977], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 19:07:30,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009527.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:07:42,580 INFO [train.py:968] (0/2) Epoch 23, batch 6150, giga_loss[loss=0.2998, simple_loss=0.3649, pruned_loss=0.1173, over 28863.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3651, pruned_loss=0.1133, over 5703112.58 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3433, pruned_loss=0.08954, over 5485726.09 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3652, pruned_loss=0.1144, over 5687450.34 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:08:20,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.723e+02 1.682e+03 2.213e+03 2.737e+03 6.416e+03, threshold=4.426e+03, percent-clipped=9.0
+2023-03-11 19:08:36,113 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:08:37,398 INFO [train.py:968] (0/2) Epoch 23, batch 6200, giga_loss[loss=0.3223, simple_loss=0.3832, pruned_loss=0.1307, over 28554.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.37, pruned_loss=0.1177, over 5704705.83 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3436, pruned_loss=0.08977, over 5492402.30 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3703, pruned_loss=0.1188, over 5690094.29 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:08:38,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:09:06,502 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:09:15,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9557, 3.0194, 1.9330, 1.0996], device='cuda:0'), covar=tensor([0.7316, 0.3013, 0.3727, 0.6640], device='cuda:0'), in_proj_covar=tensor([0.1747, 0.1649, 0.1601, 0.1431], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 19:09:17,359 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 19:09:24,541 INFO [train.py:968] (0/2) Epoch 23, batch 6250, libri_loss[loss=0.2454, simple_loss=0.3341, pruned_loss=0.07837, over 29567.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3753, pruned_loss=0.1221, over 5699297.04 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3434, pruned_loss=0.08962, over 5498792.58 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3761, pruned_loss=0.1236, over 5684696.85 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:09:53,636 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009670.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:09:55,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009673.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:09:56,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.104e+03 1.772e+03 2.454e+03 3.454e+03 8.557e+03, threshold=4.908e+03, percent-clipped=10.0
+2023-03-11 19:10:14,580 INFO [train.py:968] (0/2) Epoch 23, batch 6300, giga_loss[loss=0.287, simple_loss=0.3519, pruned_loss=0.1111, over 28812.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3767, pruned_loss=0.1234, over 5683800.68 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3435, pruned_loss=0.08958, over 5503906.39 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3786, pruned_loss=0.1259, over 5674728.13 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:10:26,733 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009702.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:10:28,028 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8810, 1.2350, 1.3116, 0.9729], device='cuda:0'), covar=tensor([0.2098, 0.1443, 0.2532, 0.1937], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0753, 0.0717, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 19:10:41,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6048, 1.5389, 1.7363, 1.2744], device='cuda:0'), covar=tensor([0.1942, 0.2985, 0.1529, 0.1726], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0701, 0.0948, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 19:10:48,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:10:58,683 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:11:05,797 INFO [train.py:968] (0/2) Epoch 23, batch 6350, giga_loss[loss=0.385, simple_loss=0.4138, pruned_loss=0.178, over 23510.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3801, pruned_loss=0.1271, over 5668465.28 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3436, pruned_loss=0.0896, over 5509296.77 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.382, pruned_loss=0.1296, over 5658787.06 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:11:09,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 19:11:16,840 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:11:44,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.788e+03 2.304e+03 3.173e+03 1.378e+04, threshold=4.607e+03, percent-clipped=7.0
+2023-03-11 19:12:01,385 INFO [train.py:968] (0/2) Epoch 23, batch 6400, giga_loss[loss=0.2761, simple_loss=0.3421, pruned_loss=0.105, over 28866.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3828, pruned_loss=0.1302, over 5670086.34 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3432, pruned_loss=0.08929, over 5514884.27 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3855, pruned_loss=0.1334, over 5660930.30 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:12:18,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2628, 1.1922, 3.4734, 3.0820], device='cuda:0'), covar=tensor([0.1544, 0.2723, 0.0486, 0.1480], device='cuda:0'), in_proj_covar=tensor([0.0763, 0.0649, 0.0964, 0.0915], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 19:12:27,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:12:30,367 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-11 19:12:42,474 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:12:58,708 INFO [train.py:968] (0/2) Epoch 23, batch 6450, giga_loss[loss=0.4049, simple_loss=0.4309, pruned_loss=0.1894, over 26389.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3865, pruned_loss=0.1345, over 5641748.06 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3435, pruned_loss=0.08942, over 5512486.75 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.389, pruned_loss=0.1375, over 5638750.32 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:13:16,393 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009856.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:13:26,277 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:13:28,975 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9412, 1.3215, 1.0964, 0.2751], device='cuda:0'), covar=tensor([0.4223, 0.3269, 0.4547, 0.6318], device='cuda:0'), in_proj_covar=tensor([0.1746, 0.1646, 0.1598, 0.1429], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 19:13:29,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:13:38,963 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 2.022e+03 2.536e+03 4.291e+03 1.203e+04, threshold=5.072e+03, percent-clipped=20.0
+2023-03-11 19:13:57,171 INFO [train.py:968] (0/2) Epoch 23, batch 6500, giga_loss[loss=0.3507, simple_loss=0.3911, pruned_loss=0.1552, over 28767.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3895, pruned_loss=0.1371, over 5641400.59 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3433, pruned_loss=0.08929, over 5515429.68 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3921, pruned_loss=0.1401, over 5637454.30 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:13:58,343 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-11 19:14:03,058 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:14:39,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2607, 1.5522, 1.0632, 1.1247], device='cuda:0'), covar=tensor([0.1100, 0.0568, 0.1211, 0.1107], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0448, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 19:14:39,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:14:46,536 INFO [train.py:968] (0/2) Epoch 23, batch 6550, giga_loss[loss=0.285, simple_loss=0.3529, pruned_loss=0.1086, over 28782.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3879, pruned_loss=0.1365, over 5644493.64 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3432, pruned_loss=0.08923, over 5523173.88 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3909, pruned_loss=0.1399, over 5636622.83 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:15:05,873 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:15:08,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:15:22,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.964e+03 2.713e+03 3.537e+03 7.994e+03, threshold=5.426e+03, percent-clipped=7.0
+2023-03-11 19:15:38,613 INFO [train.py:968] (0/2) Epoch 23, batch 6600, giga_loss[loss=0.2875, simple_loss=0.3653, pruned_loss=0.1048, over 28873.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3875, pruned_loss=0.1372, over 5624739.87 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3437, pruned_loss=0.08954, over 5521833.84 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3904, pruned_loss=0.1406, over 5622687.23 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:15:38,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:15:48,565 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1010000.pt
+2023-03-11 19:16:28,196 INFO [train.py:968] (0/2) Epoch 23, batch 6650, libri_loss[loss=0.2357, simple_loss=0.3197, pruned_loss=0.07582, over 29582.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3872, pruned_loss=0.1357, over 5640374.62 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3433, pruned_loss=0.08935, over 5530002.06 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3912, pruned_loss=0.1399, over 5634201.63 frames. ], batch size: 74, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:17:03,958 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.15 vs. limit=5.0
+2023-03-11 19:17:04,104 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.684e+02 1.684e+03 2.157e+03 2.978e+03 9.558e+03, threshold=4.314e+03, percent-clipped=3.0
+2023-03-11 19:17:12,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:17:20,550 INFO [train.py:968] (0/2) Epoch 23, batch 6700, giga_loss[loss=0.4416, simple_loss=0.4728, pruned_loss=0.2052, over 28285.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.388, pruned_loss=0.1357, over 5644909.42 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3429, pruned_loss=0.08904, over 5535216.39 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3923, pruned_loss=0.1402, over 5637040.97 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:17:21,339 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:17:34,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:17:55,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:17:57,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:18:13,743 INFO [train.py:968] (0/2) Epoch 23, batch 6750, giga_loss[loss=0.2752, simple_loss=0.3553, pruned_loss=0.09751, over 29023.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3886, pruned_loss=0.1363, over 5625524.25 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3432, pruned_loss=0.08937, over 5533035.89 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3925, pruned_loss=0.1404, over 5622610.66 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:18:47,243 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.186e+03 1.816e+03 2.263e+03 3.188e+03 6.293e+03, threshold=4.526e+03, percent-clipped=9.0
+2023-03-11 19:19:05,530 INFO [train.py:968] (0/2) Epoch 23, batch 6800, giga_loss[loss=0.295, simple_loss=0.3631, pruned_loss=0.1134, over 28635.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3848, pruned_loss=0.1326, over 5628763.57 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08923, over 5545732.95 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3895, pruned_loss=0.1374, over 5618165.20 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:19:16,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:19:47,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010231.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:19:54,856 INFO [train.py:968] (0/2) Epoch 23, batch 6850, giga_loss[loss=0.2583, simple_loss=0.3406, pruned_loss=0.08797, over 28934.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3822, pruned_loss=0.1288, over 5649308.28 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3431, pruned_loss=0.08931, over 5552780.00 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3865, pruned_loss=0.1332, over 5636424.78 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:20:05,897 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:20:08,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:20:21,022 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:20:23,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:20:27,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.572e+03 1.880e+03 2.233e+03 5.785e+03, threshold=3.761e+03, percent-clipped=2.0
+2023-03-11 19:20:38,325 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:20:46,925 INFO [train.py:968] (0/2) Epoch 23, batch 6900, giga_loss[loss=0.3014, simple_loss=0.3689, pruned_loss=0.1169, over 28658.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3777, pruned_loss=0.1251, over 5651116.55 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3431, pruned_loss=0.08933, over 5557994.41 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3816, pruned_loss=0.1291, over 5637721.00 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:20:49,109 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010293.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:20:57,378 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:21:04,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:21:37,795 INFO [train.py:968] (0/2) Epoch 23, batch 6950, libri_loss[loss=0.2171, simple_loss=0.311, pruned_loss=0.06165, over 29530.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.373, pruned_loss=0.1214, over 5651955.93 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3424, pruned_loss=0.08901, over 5568009.50 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3778, pruned_loss=0.1261, over 5634911.49 frames. ], batch size: 81, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:21:42,411 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:21:44,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:21:45,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9496, 3.4101, 2.3535, 1.1085], device='cuda:0'), covar=tensor([0.8931, 0.2758, 0.3808, 0.7742], device='cuda:0'), in_proj_covar=tensor([0.1756, 0.1650, 0.1603, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 19:22:07,463 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010374.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:22:08,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.754e+03 2.247e+03 2.915e+03 1.052e+04, threshold=4.494e+03, percent-clipped=12.0
+2023-03-11 19:22:10,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010377.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:22:11,591 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:22:23,141 INFO [train.py:968] (0/2) Epoch 23, batch 7000, giga_loss[loss=0.3226, simple_loss=0.3867, pruned_loss=0.1292, over 28479.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3716, pruned_loss=0.1207, over 5656504.62 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3423, pruned_loss=0.08896, over 5577664.40 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3764, pruned_loss=0.1254, over 5636434.93 frames. ], batch size: 65, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:22:27,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3564, 1.4686, 1.3629, 1.4589], device='cuda:0'), covar=tensor([0.0753, 0.0365, 0.0327, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 19:22:29,549 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-11 19:22:38,695 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010406.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:23:00,527 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:23:16,235 INFO [train.py:968] (0/2) Epoch 23, batch 7050, giga_loss[loss=0.4038, simple_loss=0.4192, pruned_loss=0.1942, over 23662.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3718, pruned_loss=0.1212, over 5645586.14 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08906, over 5571570.83 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3756, pruned_loss=0.125, over 5635580.96 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:23:27,237 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:23:30,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:23:35,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:23:43,531 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-11 19:23:45,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:23:54,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.817e+02 1.581e+03 2.068e+03 3.158e+03 1.044e+04, threshold=4.136e+03, percent-clipped=15.0
+2023-03-11 19:24:03,657 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:24:11,624 INFO [train.py:968] (0/2) Epoch 23, batch 7100, libri_loss[loss=0.2567, simple_loss=0.3496, pruned_loss=0.08189, over 29572.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.1199, over 5646979.64 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08933, over 5573129.53 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3746, pruned_loss=0.1237, over 5640647.57 frames. ], batch size: 78, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:24:23,354 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:25:06,861 INFO [train.py:968] (0/2) Epoch 23, batch 7150, giga_loss[loss=0.2709, simple_loss=0.363, pruned_loss=0.08943, over 28960.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3706, pruned_loss=0.1179, over 5652177.04 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08937, over 5576901.94 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3738, pruned_loss=0.1212, over 5644844.87 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:25:29,299 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:25:41,343 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2265, 2.2867, 2.3409, 1.9398], device='cuda:0'), covar=tensor([0.1985, 0.2599, 0.2081, 0.2645], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0755, 0.0719, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 19:25:52,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.324e+02 1.655e+03 2.181e+03 2.714e+03 6.075e+03, threshold=4.362e+03, percent-clipped=9.0
+2023-03-11 19:26:05,052 INFO [train.py:968] (0/2) Epoch 23, batch 7200, giga_loss[loss=0.2908, simple_loss=0.3628, pruned_loss=0.1094, over 28544.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3721, pruned_loss=0.1168, over 5654245.20 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08944, over 5571057.83 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3747, pruned_loss=0.1197, over 5654556.21 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:26:15,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:26:18,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:26:23,500 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:26:27,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:26:50,050 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:26:58,894 INFO [train.py:968] (0/2) Epoch 23, batch 7250, giga_loss[loss=0.2814, simple_loss=0.3567, pruned_loss=0.1031, over 28473.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3727, pruned_loss=0.1173, over 5660250.33 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3431, pruned_loss=0.08957, over 5579230.81 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3756, pruned_loss=0.12, over 5655154.14 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:27:00,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:27:03,775 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:27:06,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:27:25,568 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010668.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:27:34,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.792e+03 2.189e+03 3.518e+03 1.109e+04, threshold=4.378e+03, percent-clipped=17.0
+2023-03-11 19:27:35,321 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:27:49,110 INFO [train.py:968] (0/2) Epoch 23, batch 7300, giga_loss[loss=0.29, simple_loss=0.361, pruned_loss=0.1094, over 28924.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3742, pruned_loss=0.1195, over 5653312.71 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08947, over 5580269.79 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3769, pruned_loss=0.1222, over 5649047.39 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:28:09,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2727, 3.3560, 1.4032, 1.5958], device='cuda:0'), covar=tensor([0.1091, 0.0421, 0.0937, 0.1417], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0562, 0.0392, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 19:28:37,544 INFO [train.py:968] (0/2) Epoch 23, batch 7350, giga_loss[loss=0.2579, simple_loss=0.3291, pruned_loss=0.09337, over 28833.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3719, pruned_loss=0.1184, over 5669180.91 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08968, over 5590601.00 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.375, pruned_loss=0.1215, over 5659118.79 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:29:11,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.675e+03 2.211e+03 2.889e+03 6.994e+03, threshold=4.422e+03, percent-clipped=5.0
+2023-03-11 19:29:15,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5268, 1.5959, 1.7088, 1.3041], device='cuda:0'), covar=tensor([0.1758, 0.2567, 0.1460, 0.1646], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0704, 0.0949, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 19:29:22,564 INFO [train.py:968] (0/2) Epoch 23, batch 7400, giga_loss[loss=0.2653, simple_loss=0.3395, pruned_loss=0.09557, over 28699.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.369, pruned_loss=0.1176, over 5673633.33 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08938, over 5599471.09 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3729, pruned_loss=0.1213, over 5659962.97 frames. ], batch size: 92, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:29:34,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:29:39,456 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010811.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:29:41,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010814.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:29:57,294 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-11 19:30:10,017 INFO [train.py:968] (0/2) Epoch 23, batch 7450, giga_loss[loss=0.2921, simple_loss=0.3577, pruned_loss=0.1133, over 28010.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3677, pruned_loss=0.1166, over 5672906.07 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3429, pruned_loss=0.08932, over 5604123.16 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3714, pruned_loss=0.1205, over 5660315.26 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:30:11,905 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010843.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:30:48,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.762e+03 2.144e+03 3.143e+03 8.222e+03, threshold=4.288e+03, percent-clipped=9.0
+2023-03-11 19:31:03,147 INFO [train.py:968] (0/2) Epoch 23, batch 7500, giga_loss[loss=0.2905, simple_loss=0.3656, pruned_loss=0.1077, over 28918.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.367, pruned_loss=0.1151, over 5668123.53 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08926, over 5607132.42 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3703, pruned_loss=0.1185, over 5656053.02 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:31:41,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:31:51,302 INFO [train.py:968] (0/2) Epoch 23, batch 7550, giga_loss[loss=0.2912, simple_loss=0.3774, pruned_loss=0.1025, over 28976.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3678, pruned_loss=0.1152, over 5674053.85 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08939, over 5613333.62 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3707, pruned_loss=0.1182, over 5659910.44 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:31:59,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:32:01,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:32:27,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.534e+02 1.646e+03 2.039e+03 2.727e+03 5.768e+03, threshold=4.078e+03, percent-clipped=6.0
+2023-03-11 19:32:29,461 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:32:37,901 INFO [train.py:968] (0/2) Epoch 23, batch 7600, giga_loss[loss=0.2849, simple_loss=0.3555, pruned_loss=0.1072, over 28927.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3671, pruned_loss=0.1145, over 5685345.22 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08937, over 5616265.89 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3695, pruned_loss=0.1171, over 5672167.34 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:33:20,453 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1011029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:33:31,090 INFO [train.py:968] (0/2) Epoch 23, batch 7650, giga_loss[loss=0.2586, simple_loss=0.3328, pruned_loss=0.09216, over 28339.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3659, pruned_loss=0.1144, over 5682997.79 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3433, pruned_loss=0.08949, over 5621836.64 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3681, pruned_loss=0.1169, over 5669041.85 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:34:02,910 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1011075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:34:05,115 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 19:34:06,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6180, 1.9900, 1.5162, 1.8050], device='cuda:0'), covar=tensor([0.2949, 0.2831, 0.3357, 0.2508], device='cuda:0'), in_proj_covar=tensor([0.1520, 0.1096, 0.1347, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 19:34:06,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.734e+03 2.209e+03 3.145e+03 6.550e+03, threshold=4.418e+03, percent-clipped=12.0
+2023-03-11 19:34:07,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1011078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:34:19,212 INFO [train.py:968] (0/2) Epoch 23, batch 7700, giga_loss[loss=0.2471, simple_loss=0.3281, pruned_loss=0.08308, over 29009.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3641, pruned_loss=0.1135, over 5675295.01 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3437, pruned_loss=0.08953, over 5632439.20 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3665, pruned_loss=0.1166, over 5656780.36 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:34:33,836 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-11 19:34:34,903 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1011107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:35:06,807 INFO [train.py:968] (0/2) Epoch 23, batch 7750, giga_loss[loss=0.2842, simple_loss=0.3548, pruned_loss=0.1068, over 29102.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3639, pruned_loss=0.1144, over 5671275.94 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3439, pruned_loss=0.08965, over 5636614.00 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3661, pruned_loss=0.1174, over 5653888.66 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:35:37,050 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1011167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:35:37,877 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 19:35:45,033 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.177e+03 1.718e+03 2.266e+03 3.121e+03 9.935e+03, threshold=4.533e+03, percent-clipped=8.0
+2023-03-11 19:35:55,902 INFO [train.py:968] (0/2) Epoch 23, batch 7800, libri_loss[loss=0.2407, simple_loss=0.3271, pruned_loss=0.07717, over 29549.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3629, pruned_loss=0.1139, over 5667444.81 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3443, pruned_loss=0.0898, over 5636837.50 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3651, pruned_loss=0.1171, over 5654539.83 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:36:44,740 INFO [train.py:968] (0/2) Epoch 23, batch 7850, giga_loss[loss=0.2828, simple_loss=0.3489, pruned_loss=0.1084, over 28721.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3616, pruned_loss=0.1136, over 5658310.30 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3443, pruned_loss=0.08973, over 5639198.69 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3636, pruned_loss=0.1165, over 5646158.63 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:36:45,126 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1011241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:37:19,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.679e+03 2.308e+03 3.508e+03 1.079e+04, threshold=4.617e+03, percent-clipped=10.0
+2023-03-11 19:37:30,840 INFO [train.py:968] (0/2) Epoch 23, batch 7900, giga_loss[loss=0.3084, simple_loss=0.3747, pruned_loss=0.121, over 28860.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3613, pruned_loss=0.1133, over 5666049.53 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3441, pruned_loss=0.0895, over 5645960.79 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3634, pruned_loss=0.1165, over 5650365.65 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:38:21,365 INFO [train.py:968] (0/2) Epoch 23, batch 7950, giga_loss[loss=0.2618, simple_loss=0.3434, pruned_loss=0.09007, over 28921.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3616, pruned_loss=0.1129, over 5669559.32 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3437, pruned_loss=0.08926, over 5650901.14 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3641, pruned_loss=0.1163, over 5653102.55 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:38:56,894 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.626e+03 2.083e+03 2.985e+03 1.085e+04, threshold=4.166e+03, percent-clipped=12.0
+2023-03-11 19:39:09,539 INFO [train.py:968] (0/2) Epoch 23, batch 8000, libri_loss[loss=0.2739, simple_loss=0.3579, pruned_loss=0.09497, over 28654.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3628, pruned_loss=0.1129, over 5668356.97 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3441, pruned_loss=0.08953, over 5647027.18 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3648, pruned_loss=0.1158, over 5659300.45 frames. ], batch size: 106, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:39:21,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1011404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:39:58,791 INFO [train.py:968] (0/2) Epoch 23, batch 8050, giga_loss[loss=0.2789, simple_loss=0.356, pruned_loss=0.1009, over 29016.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3624, pruned_loss=0.1119, over 5677422.55 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.344, pruned_loss=0.0896, over 5646628.89 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3644, pruned_loss=0.1145, over 5670865.33 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:40:37,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.558e+03 1.879e+03 2.327e+03 1.222e+04, threshold=3.757e+03, percent-clipped=2.0
+2023-03-11 19:40:47,764 INFO [train.py:968] (0/2) Epoch 23, batch 8100, giga_loss[loss=0.3439, simple_loss=0.3943, pruned_loss=0.1468, over 28585.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3635, pruned_loss=0.1132, over 5682058.06 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.344, pruned_loss=0.08971, over 5653064.80 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3656, pruned_loss=0.1158, over 5671696.24 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:41:39,289 INFO [train.py:968] (0/2) Epoch 23, batch 8150, giga_loss[loss=0.3714, simple_loss=0.4014, pruned_loss=0.1706, over 23621.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3679, pruned_loss=0.1174, over 5661095.19 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.08992, over 5647600.30 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3696, pruned_loss=0.1196, over 5658099.70 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:41:40,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1011542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:41:44,788 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1011547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:41:51,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1011550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:42:21,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.824e+03 2.503e+03 3.229e+03 7.377e+03, threshold=5.005e+03, percent-clipped=18.0
+2023-03-11 19:42:21,949 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1011579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:42:35,483 INFO [train.py:968] (0/2) Epoch 23, batch 8200, giga_loss[loss=0.2741, simple_loss=0.3456, pruned_loss=0.1013, over 28921.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3701, pruned_loss=0.1213, over 5654412.05 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3444, pruned_loss=0.09003, over 5648900.30 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3714, pruned_loss=0.123, over 5651003.55 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:42:41,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2812, 2.8959, 1.3337, 1.4833], device='cuda:0'), covar=tensor([0.0982, 0.0414, 0.0897, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0561, 0.0391, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 19:42:52,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2344, 5.0794, 4.8174, 2.4383], device='cuda:0'), covar=tensor([0.0549, 0.0661, 0.0765, 0.1762], device='cuda:0'), in_proj_covar=tensor([0.1260, 0.1166, 0.0992, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 19:43:02,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1011616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:43:02,625 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 19:43:08,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2327, 1.5351, 1.5071, 1.1049], device='cuda:0'), covar=tensor([0.1598, 0.2397, 0.1332, 0.1548], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0704, 0.0948, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 19:43:27,024 INFO [train.py:968] (0/2) Epoch 23, batch 8250, giga_loss[loss=0.2584, simple_loss=0.3272, pruned_loss=0.09477, over 28647.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.371, pruned_loss=0.123, over 5667584.19 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.08993, over 5653970.08 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3727, pruned_loss=0.1251, over 5660416.63 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:43:43,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7279, 2.6042, 1.6914, 0.8486], device='cuda:0'), covar=tensor([0.8786, 0.3870, 0.3975, 0.7546], device='cuda:0'), in_proj_covar=tensor([0.1764, 0.1664, 0.1605, 0.1440], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 19:44:07,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.199e+03 1.810e+03 2.416e+03 3.329e+03 7.463e+03, threshold=4.833e+03, percent-clipped=7.0
+2023-03-11 19:44:13,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1011685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:44:17,308 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1011688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:44:20,152 INFO [train.py:968] (0/2) Epoch 23, batch 8300, giga_loss[loss=0.3716, simple_loss=0.4051, pruned_loss=0.169, over 24128.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3716, pruned_loss=0.1238, over 5651124.07 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3445, pruned_loss=0.09003, over 5646871.86 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.373, pruned_loss=0.1256, over 5652249.62 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:44:44,830 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1011717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:45:07,784 INFO [train.py:968] (0/2) Epoch 23, batch 8350, giga_loss[loss=0.2535, simple_loss=0.3365, pruned_loss=0.08524, over 28891.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3698, pruned_loss=0.1222, over 5664595.36 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3446, pruned_loss=0.09009, over 5651806.45 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3712, pruned_loss=0.1242, over 5661175.52 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:45:10,061 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1011744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:45:23,180 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1011759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:45:25,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1011762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:45:41,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.560e+03 2.109e+03 3.050e+03 1.264e+04, threshold=4.219e+03, percent-clipped=8.0
+2023-03-11 19:45:50,787 INFO [train.py:968] (0/2) Epoch 23, batch 8400, giga_loss[loss=0.2938, simple_loss=0.3689, pruned_loss=0.1094, over 28667.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3683, pruned_loss=0.1193, over 5679810.39 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3449, pruned_loss=0.09043, over 5658668.41 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3699, pruned_loss=0.1214, over 5671448.71 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:45:50,961 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1011791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:46:03,756 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1011804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:46:20,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4733, 3.5776, 1.5715, 1.7088], device='cuda:0'), covar=tensor([0.1026, 0.0329, 0.0892, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0562, 0.0391, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 19:46:34,503 INFO [train.py:968] (0/2) Epoch 23, batch 8450, giga_loss[loss=0.3344, simple_loss=0.3934, pruned_loss=0.1377, over 28676.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3661, pruned_loss=0.1168, over 5681790.48 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.345, pruned_loss=0.09048, over 5668259.51 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3683, pruned_loss=0.1197, over 5666931.49 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:47:09,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.687e+03 2.126e+03 3.017e+03 5.185e+03, threshold=4.252e+03, percent-clipped=9.0
+2023-03-11 19:47:20,188 INFO [train.py:968] (0/2) Epoch 23, batch 8500, giga_loss[loss=0.2544, simple_loss=0.3315, pruned_loss=0.08866, over 28886.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.364, pruned_loss=0.1158, over 5681538.51 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.345, pruned_loss=0.09048, over 5673255.68 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3662, pruned_loss=0.1186, over 5665714.00 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:47:25,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3561, 3.3013, 1.5389, 1.4907], device='cuda:0'), covar=tensor([0.1042, 0.0323, 0.0866, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0561, 0.0390, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 19:47:49,890 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2522, 2.1406, 2.1570, 1.8490], device='cuda:0'), covar=tensor([0.1821, 0.2710, 0.2287, 0.2650], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0748, 0.0714, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 19:47:54,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6109, 1.8740, 1.5112, 1.6864], device='cuda:0'), covar=tensor([0.2586, 0.2701, 0.2990, 0.2787], device='cuda:0'), in_proj_covar=tensor([0.1523, 0.1097, 0.1345, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 19:48:08,890 INFO [train.py:968] (0/2) Epoch 23, batch 8550, giga_loss[loss=0.2611, simple_loss=0.335, pruned_loss=0.09354, over 29054.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3618, pruned_loss=0.1145, over 5685032.03 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3451, pruned_loss=0.09059, over 5676654.41 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3636, pruned_loss=0.117, over 5669565.67 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:48:47,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.705e+03 2.333e+03 3.386e+03 8.776e+03, threshold=4.667e+03, percent-clipped=13.0
+2023-03-11 19:48:59,414 INFO [train.py:968] (0/2) Epoch 23, batch 8600, giga_loss[loss=0.3035, simple_loss=0.3633, pruned_loss=0.1218, over 28703.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3641, pruned_loss=0.1166, over 5676611.97 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.345, pruned_loss=0.09054, over 5676787.83 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3665, pruned_loss=0.1196, over 5663893.93 frames. ], batch size: 92, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:49:09,653 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1012000.pt
+2023-03-11 19:49:53,164 INFO [train.py:968] (0/2) Epoch 23, batch 8650, giga_loss[loss=0.338, simple_loss=0.4045, pruned_loss=0.1358, over 28879.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3683, pruned_loss=0.1187, over 5677998.59 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3447, pruned_loss=0.09034, over 5678934.18 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3706, pruned_loss=0.1215, over 5666083.63 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:50:27,845 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.32 vs. limit=2.0
+2023-03-11 19:50:32,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.966e+02 1.597e+03 1.932e+03 3.035e+03 5.385e+03, threshold=3.864e+03, percent-clipped=6.0
+2023-03-11 19:50:41,985 INFO [train.py:968] (0/2) Epoch 23, batch 8700, giga_loss[loss=0.2721, simple_loss=0.3606, pruned_loss=0.09186, over 28594.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3718, pruned_loss=0.1182, over 5679276.54 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3447, pruned_loss=0.09047, over 5681401.42 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3738, pruned_loss=0.1206, over 5667648.30 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:51:07,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:51:27,069 INFO [train.py:968] (0/2) Epoch 23, batch 8750, giga_loss[loss=0.311, simple_loss=0.3837, pruned_loss=0.1192, over 28934.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3746, pruned_loss=0.1195, over 5687963.17 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3446, pruned_loss=0.09041, over 5686301.10 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3769, pruned_loss=0.122, over 5674560.04 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:51:44,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7708, 5.6122, 5.3287, 2.8070], device='cuda:0'), covar=tensor([0.0516, 0.0699, 0.0751, 0.1641], device='cuda:0'), in_proj_covar=tensor([0.1252, 0.1157, 0.0986, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 19:52:01,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:52:02,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.597e+03 2.231e+03 3.143e+03 1.517e+04, threshold=4.462e+03, percent-clipped=16.0
+2023-03-11 19:52:13,239 INFO [train.py:968] (0/2) Epoch 23, batch 8800, giga_loss[loss=0.386, simple_loss=0.438, pruned_loss=0.167, over 28878.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3767, pruned_loss=0.1216, over 5691180.55 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3454, pruned_loss=0.09094, over 5690527.45 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3786, pruned_loss=0.1239, over 5676673.56 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:52:19,888 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1012199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:52:32,747 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1012211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:52:33,456 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1012212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:52:37,551 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-11 19:52:59,052 INFO [train.py:968] (0/2) Epoch 23, batch 8850, giga_loss[loss=0.2941, simple_loss=0.3592, pruned_loss=0.1146, over 28546.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3764, pruned_loss=0.1218, over 5694715.64 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3455, pruned_loss=0.0909, over 5696367.27 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3788, pruned_loss=0.1247, over 5677948.63 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:53:21,685 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:53:23,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1012265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:53:27,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-11 19:53:32,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7207, 3.5460, 3.3634, 1.5892], device='cuda:0'), covar=tensor([0.0759, 0.0864, 0.0818, 0.2374], device='cuda:0'), in_proj_covar=tensor([0.1256, 0.1161, 0.0989, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 19:53:39,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.200e+03 1.629e+03 2.119e+03 2.843e+03 5.506e+03, threshold=4.238e+03, percent-clipped=4.0
+2023-03-11 19:53:48,462 INFO [train.py:968] (0/2) Epoch 23, batch 8900, libri_loss[loss=0.241, simple_loss=0.3254, pruned_loss=0.07831, over 29592.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3779, pruned_loss=0.1241, over 5694492.53 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3457, pruned_loss=0.09096, over 5701171.20 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3804, pruned_loss=0.1271, over 5676549.97 frames. ], batch size: 75, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:53:51,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1012294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:54:18,654 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6674, 3.5227, 3.3464, 1.6658], device='cuda:0'), covar=tensor([0.0783, 0.0837, 0.0803, 0.2289], device='cuda:0'), in_proj_covar=tensor([0.1256, 0.1161, 0.0989, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 19:54:18,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3945, 1.6983, 1.3811, 1.2705], device='cuda:0'), covar=tensor([0.2399, 0.2351, 0.2582, 0.2190], device='cuda:0'), in_proj_covar=tensor([0.1524, 0.1100, 0.1347, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 19:54:21,833 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:54:26,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1012325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:54:41,589 INFO [train.py:968] (0/2) Epoch 23, batch 8950, giga_loss[loss=0.2563, simple_loss=0.3213, pruned_loss=0.09569, over 28644.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.376, pruned_loss=0.1235, over 5690344.85 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3459, pruned_loss=0.09107, over 5701129.37 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3784, pruned_loss=0.1264, over 5675866.55 frames. ], batch size: 92, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:54:48,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-11 19:54:52,105 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1012354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:55:19,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.148e+03 1.835e+03 2.275e+03 3.462e+03 9.683e+03, threshold=4.550e+03, percent-clipped=16.0
+2023-03-11 19:55:30,040 INFO [train.py:968] (0/2) Epoch 23, batch 9000, giga_loss[loss=0.2968, simple_loss=0.3568, pruned_loss=0.1184, over 28601.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3724, pruned_loss=0.1216, over 5687551.46 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3459, pruned_loss=0.09101, over 5706986.88 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3751, pruned_loss=0.1248, over 5670540.83 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:55:30,044 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 19:55:39,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3700, 1.2631, 1.1822, 1.5510], device='cuda:0'), covar=tensor([0.0831, 0.0366, 0.0364, 0.0930], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 19:55:39,578 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2075, simple_loss=0.3151, pruned_loss=0.04992, over 944034.00 frames. 
+2023-03-11 19:55:39,579 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 19:56:26,469 INFO [train.py:968] (0/2) Epoch 23, batch 9050, giga_loss[loss=0.3189, simple_loss=0.3802, pruned_loss=0.1288, over 27988.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3712, pruned_loss=0.1215, over 5686884.45 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3459, pruned_loss=0.09121, over 5711714.18 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3739, pruned_loss=0.1246, over 5668795.44 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:57:05,023 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1012480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:57:06,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.728e+03 2.373e+03 3.117e+03 6.032e+03, threshold=4.746e+03, percent-clipped=5.0
+2023-03-11 19:57:15,234 INFO [train.py:968] (0/2) Epoch 23, batch 9100, giga_loss[loss=0.3407, simple_loss=0.394, pruned_loss=0.1437, over 27904.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3701, pruned_loss=0.1207, over 5683349.26 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3454, pruned_loss=0.091, over 5711856.46 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3737, pruned_loss=0.1245, over 5667487.98 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:58:03,527 INFO [train.py:968] (0/2) Epoch 23, batch 9150, giga_loss[loss=0.2887, simple_loss=0.3491, pruned_loss=0.1142, over 28810.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3704, pruned_loss=0.1218, over 5684787.56 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3454, pruned_loss=0.0911, over 5715292.36 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3737, pruned_loss=0.1253, over 5668726.80 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:58:35,571 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:58:43,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.967e+02 1.632e+03 2.047e+03 2.672e+03 5.159e+03, threshold=4.094e+03, percent-clipped=3.0
+2023-03-11 19:58:47,740 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:58:50,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:58:52,791 INFO [train.py:968] (0/2) Epoch 23, batch 9200, giga_loss[loss=0.3042, simple_loss=0.37, pruned_loss=0.1192, over 28246.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3693, pruned_loss=0.1212, over 5685989.05 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3453, pruned_loss=0.09091, over 5718313.14 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3724, pruned_loss=0.1246, over 5670238.55 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:59:08,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9733, 3.8170, 3.6210, 1.7676], device='cuda:0'), covar=tensor([0.0700, 0.0822, 0.0824, 0.2010], device='cuda:0'), in_proj_covar=tensor([0.1256, 0.1160, 0.0987, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 19:59:10,718 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-11 19:59:40,089 INFO [train.py:968] (0/2) Epoch 23, batch 9250, giga_loss[loss=0.2973, simple_loss=0.3718, pruned_loss=0.1114, over 28950.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3688, pruned_loss=0.1202, over 5687059.64 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3453, pruned_loss=0.09091, over 5710800.93 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3713, pruned_loss=0.123, over 5680355.14 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:00:20,097 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.774e+03 2.353e+03 3.082e+03 1.288e+04, threshold=4.706e+03, percent-clipped=10.0
+2023-03-11 20:00:27,959 INFO [train.py:968] (0/2) Epoch 23, batch 9300, giga_loss[loss=0.3211, simple_loss=0.3886, pruned_loss=0.1268, over 28590.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3698, pruned_loss=0.1202, over 5678427.45 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3454, pruned_loss=0.09117, over 5710684.39 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3731, pruned_loss=0.1238, over 5670570.64 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:00:52,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:00:54,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1012720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:00:55,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2508, 4.1127, 3.9033, 1.9170], device='cuda:0'), covar=tensor([0.0620, 0.0725, 0.0779, 0.2123], device='cuda:0'), in_proj_covar=tensor([0.1259, 0.1165, 0.0990, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 20:01:04,347 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:05,641 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:07,972 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1012732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:08,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1012733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:15,800 INFO [train.py:968] (0/2) Epoch 23, batch 9350, giga_loss[loss=0.2925, simple_loss=0.3536, pruned_loss=0.1157, over 28752.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3707, pruned_loss=0.121, over 5672681.64 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3451, pruned_loss=0.09102, over 5713401.80 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3743, pruned_loss=0.1248, over 5662856.03 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:01:24,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1012749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:37,025 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1012761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:37,631 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1012762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:58,351 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.693e+03 2.034e+03 2.762e+03 5.198e+03, threshold=4.067e+03, percent-clipped=1.0
+2023-03-11 20:02:07,013 INFO [train.py:968] (0/2) Epoch 23, batch 9400, giga_loss[loss=0.3099, simple_loss=0.396, pruned_loss=0.1119, over 28689.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3711, pruned_loss=0.1212, over 5675509.00 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3451, pruned_loss=0.09102, over 5714393.79 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.374, pruned_loss=0.1243, over 5666873.64 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:02:57,132 INFO [train.py:968] (0/2) Epoch 23, batch 9450, giga_loss[loss=0.3494, simple_loss=0.3836, pruned_loss=0.1576, over 23422.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3727, pruned_loss=0.1195, over 5678704.69 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3455, pruned_loss=0.09121, over 5716390.59 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.375, pruned_loss=0.1222, over 5669648.39 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:03:10,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:03:11,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5958, 2.0881, 1.2989, 0.9716], device='cuda:0'), covar=tensor([0.7010, 0.4040, 0.3076, 0.6575], device='cuda:0'), in_proj_covar=tensor([0.1748, 0.1653, 0.1593, 0.1430], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 20:03:35,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.470e+03 2.058e+03 3.082e+03 1.057e+04, threshold=4.115e+03, percent-clipped=8.0
+2023-03-11 20:03:40,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3727, 1.6202, 1.5819, 1.1690], device='cuda:0'), covar=tensor([0.1987, 0.2737, 0.1689, 0.1934], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0709, 0.0954, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 20:03:41,754 INFO [train.py:968] (0/2) Epoch 23, batch 9500, giga_loss[loss=0.2944, simple_loss=0.3744, pruned_loss=0.1072, over 28977.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3737, pruned_loss=0.1181, over 5684475.48 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3452, pruned_loss=0.09102, over 5718891.62 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3763, pruned_loss=0.1209, over 5674423.65 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:04:27,516 INFO [train.py:968] (0/2) Epoch 23, batch 9550, giga_loss[loss=0.3525, simple_loss=0.4067, pruned_loss=0.1491, over 28985.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3764, pruned_loss=0.1206, over 5675683.86 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.345, pruned_loss=0.09105, over 5713861.01 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3799, pruned_loss=0.124, over 5670469.84 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:04:34,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6001, 1.8639, 1.7399, 1.6287], device='cuda:0'), covar=tensor([0.1908, 0.2046, 0.2277, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0750, 0.0716, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 20:04:50,747 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 20:05:09,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.625e+03 2.028e+03 3.462e+03 6.675e+03, threshold=4.056e+03, percent-clipped=17.0
+2023-03-11 20:05:18,108 INFO [train.py:968] (0/2) Epoch 23, batch 9600, giga_loss[loss=0.2765, simple_loss=0.3448, pruned_loss=0.104, over 28528.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3772, pruned_loss=0.1224, over 5669917.63 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3446, pruned_loss=0.0909, over 5705550.03 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.381, pruned_loss=0.1259, over 5672268.79 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:05:25,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:05:28,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1013001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:05:58,741 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1013030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:06:08,307 INFO [train.py:968] (0/2) Epoch 23, batch 9650, giga_loss[loss=0.2981, simple_loss=0.3622, pruned_loss=0.1171, over 28551.00 frames. ], tot_loss[loss=0.313, simple_loss=0.378, pruned_loss=0.124, over 5655844.48 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3447, pruned_loss=0.09091, over 5692493.66 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3816, pruned_loss=0.1274, over 5669213.92 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:06:46,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.203e+03 1.794e+03 2.327e+03 3.195e+03 6.047e+03, threshold=4.654e+03, percent-clipped=10.0
+2023-03-11 20:06:53,322 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 20:06:55,041 INFO [train.py:968] (0/2) Epoch 23, batch 9700, giga_loss[loss=0.2942, simple_loss=0.3644, pruned_loss=0.112, over 28732.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3762, pruned_loss=0.1236, over 5648207.95 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.344, pruned_loss=0.09057, over 5695504.84 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3802, pruned_loss=0.127, over 5655563.38 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:07:14,612 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1561, 1.2678, 3.3133, 2.9882], device='cuda:0'), covar=tensor([0.1612, 0.2700, 0.0529, 0.1177], device='cuda:0'), in_proj_covar=tensor([0.0768, 0.0654, 0.0971, 0.0923], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 20:07:22,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3899, 1.5530, 1.5148, 1.3498], device='cuda:0'), covar=tensor([0.2526, 0.2268, 0.2142, 0.2290], device='cuda:0'), in_proj_covar=tensor([0.1983, 0.1931, 0.1867, 0.1990], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 20:07:31,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3695, 2.0520, 1.4844, 0.6271], device='cuda:0'), covar=tensor([0.5547, 0.2799, 0.3645, 0.5597], device='cuda:0'), in_proj_covar=tensor([0.1755, 0.1659, 0.1597, 0.1436], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 20:07:33,666 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013132.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:07:41,075 INFO [train.py:968] (0/2) Epoch 23, batch 9750, libri_loss[loss=0.2576, simple_loss=0.3508, pruned_loss=0.08221, over 29764.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3744, pruned_loss=0.121, over 5656376.33 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3436, pruned_loss=0.09034, over 5698055.63 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3784, pruned_loss=0.1245, over 5658929.76 frames. ], batch size: 87, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:07:48,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7838, 2.6328, 1.6633, 0.9908], device='cuda:0'), covar=tensor([0.9192, 0.4047, 0.4332, 0.7649], device='cuda:0'), in_proj_covar=tensor([0.1759, 0.1662, 0.1600, 0.1440], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 20:08:10,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3094, 4.1124, 3.9196, 1.9788], device='cuda:0'), covar=tensor([0.0675, 0.0816, 0.0860, 0.2102], device='cuda:0'), in_proj_covar=tensor([0.1262, 0.1167, 0.0991, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 20:08:13,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2473, 1.8016, 1.3213, 0.5011], device='cuda:0'), covar=tensor([0.4717, 0.3147, 0.4524, 0.6463], device='cuda:0'), in_proj_covar=tensor([0.1758, 0.1661, 0.1599, 0.1440], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 20:08:22,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.542e+03 1.973e+03 2.664e+03 8.209e+03, threshold=3.947e+03, percent-clipped=7.0
+2023-03-11 20:08:28,922 INFO [train.py:968] (0/2) Epoch 23, batch 9800, giga_loss[loss=0.2875, simple_loss=0.3721, pruned_loss=0.1015, over 29053.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3738, pruned_loss=0.1188, over 5660796.52 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3438, pruned_loss=0.09038, over 5701279.11 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3773, pruned_loss=0.122, over 5659232.08 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:09:15,993 INFO [train.py:968] (0/2) Epoch 23, batch 9850, giga_loss[loss=0.3263, simple_loss=0.3882, pruned_loss=0.1322, over 28650.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.374, pruned_loss=0.1182, over 5651954.67 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3443, pruned_loss=0.09077, over 5686985.84 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3769, pruned_loss=0.121, over 5663813.08 frames. ], batch size: 78, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:09:39,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6245, 1.9861, 1.4571, 1.6570], device='cuda:0'), covar=tensor([0.0741, 0.0286, 0.0333, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 20:09:58,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.650e+02 1.619e+03 1.960e+03 2.722e+03 4.540e+03, threshold=3.921e+03, percent-clipped=9.0
+2023-03-11 20:10:08,079 INFO [train.py:968] (0/2) Epoch 23, batch 9900, giga_loss[loss=0.2741, simple_loss=0.3595, pruned_loss=0.09439, over 28877.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3743, pruned_loss=0.1189, over 5648871.16 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3442, pruned_loss=0.09054, over 5690812.23 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3772, pruned_loss=0.1218, over 5654220.58 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:10:33,497 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-11 20:10:57,916 INFO [train.py:968] (0/2) Epoch 23, batch 9950, giga_loss[loss=0.2679, simple_loss=0.3399, pruned_loss=0.09792, over 29053.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3742, pruned_loss=0.1197, over 5660546.50 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3441, pruned_loss=0.0905, over 5692392.88 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3771, pruned_loss=0.1225, over 5662704.00 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:11:03,261 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-11 20:11:30,329 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-11 20:11:41,576 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.748e+03 2.391e+03 3.418e+03 7.899e+03, threshold=4.781e+03, percent-clipped=18.0
+2023-03-11 20:11:50,076 INFO [train.py:968] (0/2) Epoch 23, batch 10000, giga_loss[loss=0.3053, simple_loss=0.3729, pruned_loss=0.1189, over 28839.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3728, pruned_loss=0.1201, over 5648722.37 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3441, pruned_loss=0.0905, over 5686761.15 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3757, pruned_loss=0.1228, over 5654381.17 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:11:56,716 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-11 20:12:35,418 INFO [train.py:968] (0/2) Epoch 23, batch 10050, libri_loss[loss=0.2371, simple_loss=0.322, pruned_loss=0.07607, over 29578.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3711, pruned_loss=0.1194, over 5663721.99 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3442, pruned_loss=0.09056, over 5694261.92 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3742, pruned_loss=0.1225, over 5660562.92 frames. ], batch size: 74, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:13:21,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.213e+02 1.540e+03 2.242e+03 3.089e+03 4.861e+03, threshold=4.484e+03, percent-clipped=1.0
+2023-03-11 20:13:29,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6573, 1.6254, 1.9112, 1.4752], device='cuda:0'), covar=tensor([0.1428, 0.2005, 0.1135, 0.1489], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0710, 0.0953, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 20:13:29,768 INFO [train.py:968] (0/2) Epoch 23, batch 10100, giga_loss[loss=0.3544, simple_loss=0.3795, pruned_loss=0.1646, over 23542.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3684, pruned_loss=0.1179, over 5673141.66 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3439, pruned_loss=0.09034, over 5699849.02 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3718, pruned_loss=0.1212, over 5664934.49 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:13:43,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1013507.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:13:46,730 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-11 20:14:13,640 INFO [train.py:968] (0/2) Epoch 23, batch 10150, giga_loss[loss=0.2823, simple_loss=0.3545, pruned_loss=0.105, over 28927.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3674, pruned_loss=0.1178, over 5672286.67 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08996, over 5705944.64 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3715, pruned_loss=0.1219, over 5659014.56 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:14:34,980 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:14:47,191 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:14:48,461 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:14:53,994 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.769e+03 2.116e+03 3.260e+03 1.305e+04, threshold=4.231e+03, percent-clipped=12.0
+2023-03-11 20:15:02,502 INFO [train.py:968] (0/2) Epoch 23, batch 10200, libri_loss[loss=0.2345, simple_loss=0.3188, pruned_loss=0.0751, over 29541.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3666, pruned_loss=0.1173, over 5672036.05 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3434, pruned_loss=0.08998, over 5707322.99 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3703, pruned_loss=0.121, over 5659599.07 frames. ], batch size: 79, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:15:50,433 INFO [train.py:968] (0/2) Epoch 23, batch 10250, giga_loss[loss=0.2426, simple_loss=0.3234, pruned_loss=0.0809, over 28865.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3638, pruned_loss=0.1141, over 5669304.75 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.08988, over 5706945.38 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3669, pruned_loss=0.1172, over 5659731.85 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:15:57,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:16:00,569 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1013650.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:16:04,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1013653.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:16:06,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8772, 2.8799, 1.8193, 1.0132], device='cuda:0'), covar=tensor([0.8650, 0.3689, 0.4576, 0.7915], device='cuda:0'), in_proj_covar=tensor([0.1759, 0.1662, 0.1596, 0.1441], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 20:16:35,106 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1013682.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:16:37,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.313e+02 1.520e+03 1.926e+03 3.018e+03 6.447e+03, threshold=3.852e+03, percent-clipped=8.0
+2023-03-11 20:16:42,992 INFO [train.py:968] (0/2) Epoch 23, batch 10300, libri_loss[loss=0.2094, simple_loss=0.2914, pruned_loss=0.06364, over 29660.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3626, pruned_loss=0.1129, over 5665759.62 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3432, pruned_loss=0.08976, over 5711172.56 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3657, pruned_loss=0.1161, over 5652962.97 frames. ], batch size: 69, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:17:28,837 INFO [train.py:968] (0/2) Epoch 23, batch 10350, giga_loss[loss=0.2616, simple_loss=0.3368, pruned_loss=0.0932, over 29057.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3616, pruned_loss=0.1127, over 5671233.31 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08994, over 5713750.72 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3646, pruned_loss=0.1158, over 5657137.78 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:17:49,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:18:12,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.633e+03 2.053e+03 2.919e+03 7.961e+03, threshold=4.106e+03, percent-clipped=13.0
+2023-03-11 20:18:20,695 INFO [train.py:968] (0/2) Epoch 23, batch 10400, giga_loss[loss=0.2642, simple_loss=0.3341, pruned_loss=0.09713, over 28948.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3583, pruned_loss=0.1114, over 5676656.83 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3433, pruned_loss=0.08989, over 5719589.82 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3613, pruned_loss=0.1147, over 5658771.70 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:18:47,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4233, 1.4804, 1.3480, 1.5687], device='cuda:0'), covar=tensor([0.0757, 0.0347, 0.0323, 0.0824], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 20:19:09,317 INFO [train.py:968] (0/2) Epoch 23, batch 10450, giga_loss[loss=0.3092, simple_loss=0.3789, pruned_loss=0.1198, over 28681.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3602, pruned_loss=0.1128, over 5684357.55 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3431, pruned_loss=0.0897, over 5723387.16 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3629, pruned_loss=0.116, over 5666119.36 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:19:45,213 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2751, 1.2317, 3.4540, 3.1239], device='cuda:0'), covar=tensor([0.1571, 0.2747, 0.0473, 0.1291], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0657, 0.0975, 0.0927], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 20:19:49,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.149e+03 1.647e+03 2.330e+03 3.039e+03 1.309e+04, threshold=4.660e+03, percent-clipped=13.0
+2023-03-11 20:19:56,210 INFO [train.py:968] (0/2) Epoch 23, batch 10500, giga_loss[loss=0.2464, simple_loss=0.327, pruned_loss=0.08291, over 28487.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3629, pruned_loss=0.1138, over 5681196.62 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.343, pruned_loss=0.08955, over 5727388.92 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3656, pruned_loss=0.117, over 5662163.21 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:20:03,973 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4596, 1.7017, 1.3702, 1.3601], device='cuda:0'), covar=tensor([0.2514, 0.2490, 0.2769, 0.2320], device='cuda:0'), in_proj_covar=tensor([0.1528, 0.1102, 0.1353, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 20:20:10,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5497, 1.8105, 1.4775, 1.7211], device='cuda:0'), covar=tensor([0.2656, 0.2637, 0.2968, 0.2273], device='cuda:0'), in_proj_covar=tensor([0.1529, 0.1103, 0.1353, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 20:20:42,699 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6355, 1.6161, 1.6495, 1.4939], device='cuda:0'), covar=tensor([0.2407, 0.2889, 0.2246, 0.2524], device='cuda:0'), in_proj_covar=tensor([0.1997, 0.1945, 0.1876, 0.2005], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 20:20:44,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1013939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:20:45,896 INFO [train.py:968] (0/2) Epoch 23, batch 10550, libri_loss[loss=0.2365, simple_loss=0.32, pruned_loss=0.07645, over 28657.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3631, pruned_loss=0.1135, over 5667639.43 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.343, pruned_loss=0.08946, over 5731695.84 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.366, pruned_loss=0.1171, over 5646123.12 frames. ], batch size: 63, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:20:53,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1013950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:20:56,373 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:20:56,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1013952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:21:27,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.613e+03 2.010e+03 2.852e+03 9.406e+03, threshold=4.020e+03, percent-clipped=5.0
+2023-03-11 20:21:32,494 INFO [train.py:968] (0/2) Epoch 23, batch 10600, giga_loss[loss=0.3572, simple_loss=0.3963, pruned_loss=0.1591, over 26474.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3633, pruned_loss=0.1136, over 5649361.52 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3433, pruned_loss=0.08938, over 5731861.39 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3661, pruned_loss=0.1174, over 5629534.99 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:21:42,859 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1014000.pt
+2023-03-11 20:22:03,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:22:05,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-11 20:22:22,088 INFO [train.py:968] (0/2) Epoch 23, batch 10650, giga_loss[loss=0.333, simple_loss=0.394, pruned_loss=0.136, over 28889.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3648, pruned_loss=0.1155, over 5654775.82 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3434, pruned_loss=0.08938, over 5733730.38 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3671, pruned_loss=0.1187, over 5636640.83 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:22:50,166 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 20:22:52,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014071.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:23:04,251 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:07,551 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:07,869 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.682e+03 2.335e+03 3.354e+03 2.038e+04, threshold=4.670e+03, percent-clipped=17.0
+2023-03-11 20:23:14,298 INFO [train.py:968] (0/2) Epoch 23, batch 10700, giga_loss[loss=0.2575, simple_loss=0.3372, pruned_loss=0.08892, over 28422.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3655, pruned_loss=0.1161, over 5647504.71 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3433, pruned_loss=0.08936, over 5732245.85 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3679, pruned_loss=0.1191, over 5633174.88 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:23:16,037 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:17,379 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:18,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:19,392 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:34,224 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:44,777 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:46,223 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:52,045 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-11 20:23:54,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:24:00,265 INFO [train.py:968] (0/2) Epoch 23, batch 10750, giga_loss[loss=0.2689, simple_loss=0.3507, pruned_loss=0.09351, over 28808.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3675, pruned_loss=0.1171, over 5648608.80 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3439, pruned_loss=0.0897, over 5729070.38 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3695, pruned_loss=0.1201, over 5637083.46 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:24:21,551 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:24:25,121 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:24:41,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5346, 1.6652, 1.6055, 1.4232], device='cuda:0'), covar=tensor([0.2949, 0.2746, 0.2122, 0.2484], device='cuda:0'), in_proj_covar=tensor([0.1998, 0.1951, 0.1882, 0.2008], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 20:24:42,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.650e+03 2.095e+03 2.681e+03 8.672e+03, threshold=4.190e+03, percent-clipped=6.0
+2023-03-11 20:24:47,952 INFO [train.py:968] (0/2) Epoch 23, batch 10800, giga_loss[loss=0.3891, simple_loss=0.42, pruned_loss=0.1791, over 26599.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3695, pruned_loss=0.1191, over 5650564.94 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3438, pruned_loss=0.08966, over 5730733.73 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3715, pruned_loss=0.1219, over 5639091.17 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:24:51,888 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:25:38,913 INFO [train.py:968] (0/2) Epoch 23, batch 10850, giga_loss[loss=0.3301, simple_loss=0.3895, pruned_loss=0.1354, over 28052.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3706, pruned_loss=0.1207, over 5652102.67 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08927, over 5730467.85 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3735, pruned_loss=0.1239, over 5641237.90 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:26:17,419 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:26:19,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:26:24,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.690e+03 2.186e+03 2.967e+03 5.460e+03, threshold=4.371e+03, percent-clipped=7.0
+2023-03-11 20:26:30,023 INFO [train.py:968] (0/2) Epoch 23, batch 10900, giga_loss[loss=0.315, simple_loss=0.3786, pruned_loss=0.1257, over 28671.00 frames. ], tot_loss[loss=0.305, simple_loss=0.371, pruned_loss=0.1195, over 5660448.09 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3423, pruned_loss=0.08895, over 5733527.95 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3745, pruned_loss=0.1229, over 5647776.81 frames. ], batch size: 92, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:26:48,269 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:27:06,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:27:19,390 INFO [train.py:968] (0/2) Epoch 23, batch 10950, giga_loss[loss=0.3742, simple_loss=0.399, pruned_loss=0.1747, over 23415.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3705, pruned_loss=0.1192, over 5652702.31 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08892, over 5737049.25 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3738, pruned_loss=0.1227, over 5637807.25 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:27:47,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8549, 1.0240, 2.8678, 2.8287], device='cuda:0'), covar=tensor([0.1694, 0.2629, 0.0639, 0.1013], device='cuda:0'), in_proj_covar=tensor([0.0769, 0.0655, 0.0969, 0.0924], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 20:27:54,594 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:28:05,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.633e+02 1.696e+03 2.123e+03 2.921e+03 8.747e+03, threshold=4.245e+03, percent-clipped=11.0
+2023-03-11 20:28:08,920 INFO [train.py:968] (0/2) Epoch 23, batch 11000, giga_loss[loss=0.3825, simple_loss=0.4156, pruned_loss=0.1747, over 26578.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3689, pruned_loss=0.1185, over 5659830.33 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08939, over 5731672.86 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3719, pruned_loss=0.1216, over 5651396.25 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:29:05,720 INFO [train.py:968] (0/2) Epoch 23, batch 11050, giga_loss[loss=0.3416, simple_loss=0.3887, pruned_loss=0.1472, over 28869.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3695, pruned_loss=0.1196, over 5661283.36 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08926, over 5735127.44 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3724, pruned_loss=0.1227, over 5650530.76 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:29:13,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014446.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:29:37,946 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:29:40,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:29:50,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.621e+02 1.727e+03 2.220e+03 2.980e+03 9.762e+03, threshold=4.440e+03, percent-clipped=9.0
+2023-03-11 20:29:57,243 INFO [train.py:968] (0/2) Epoch 23, batch 11100, giga_loss[loss=0.2735, simple_loss=0.3478, pruned_loss=0.09962, over 28572.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3674, pruned_loss=0.1181, over 5669801.05 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08912, over 5740723.17 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3708, pruned_loss=0.1217, over 5653640.64 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:30:05,161 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3749, 1.5803, 1.6108, 1.2279], device='cuda:0'), covar=tensor([0.1519, 0.2278, 0.1280, 0.1536], device='cuda:0'), in_proj_covar=tensor([0.0907, 0.0711, 0.0955, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 20:30:08,591 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:30:28,266 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:30:44,164 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:30:44,653 INFO [train.py:968] (0/2) Epoch 23, batch 11150, libri_loss[loss=0.3209, simple_loss=0.3918, pruned_loss=0.125, over 29506.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3668, pruned_loss=0.1181, over 5677068.83 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08935, over 5742062.45 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3695, pruned_loss=0.1213, over 5661653.20 frames. ], batch size: 85, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:31:29,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1642, 2.1385, 2.0110, 1.9166], device='cuda:0'), covar=tensor([0.1952, 0.2762, 0.2345, 0.2375], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0759, 0.0724, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 20:31:31,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.797e+03 2.336e+03 3.616e+03 8.132e+03, threshold=4.672e+03, percent-clipped=11.0
+2023-03-11 20:31:34,022 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014589.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:31:35,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6115, 1.8882, 1.5575, 1.5949], device='cuda:0'), covar=tensor([0.2289, 0.2232, 0.2388, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.1528, 0.1103, 0.1351, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 20:31:35,594 INFO [train.py:968] (0/2) Epoch 23, batch 11200, giga_loss[loss=0.3781, simple_loss=0.4107, pruned_loss=0.1727, over 27653.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3676, pruned_loss=0.1192, over 5666684.42 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08926, over 5743319.75 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.37, pruned_loss=0.122, over 5652833.26 frames. ], batch size: 474, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:31:37,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014592.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:32:11,673 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014621.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:32:27,045 INFO [train.py:968] (0/2) Epoch 23, batch 11250, giga_loss[loss=0.3246, simple_loss=0.3852, pruned_loss=0.132, over 28643.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3694, pruned_loss=0.1207, over 5667080.80 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3432, pruned_loss=0.08934, over 5744623.51 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3713, pruned_loss=0.1232, over 5654211.79 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:32:54,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 20:33:16,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.694e+03 2.303e+03 3.044e+03 7.781e+03, threshold=4.605e+03, percent-clipped=10.0
+2023-03-11 20:33:19,330 INFO [train.py:968] (0/2) Epoch 23, batch 11300, giga_loss[loss=0.3473, simple_loss=0.3925, pruned_loss=0.1511, over 27582.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3703, pruned_loss=0.1217, over 5672098.56 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3433, pruned_loss=0.08936, over 5747760.47 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3722, pruned_loss=0.1242, over 5657538.46 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:33:35,266 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4912, 4.3346, 4.1031, 1.9013], device='cuda:0'), covar=tensor([0.0590, 0.0715, 0.0790, 0.2129], device='cuda:0'), in_proj_covar=tensor([0.1266, 0.1172, 0.0993, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 20:33:57,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:34:08,171 INFO [train.py:968] (0/2) Epoch 23, batch 11350, giga_loss[loss=0.3163, simple_loss=0.3827, pruned_loss=0.125, over 29047.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3707, pruned_loss=0.1216, over 5674046.92 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08916, over 5750261.56 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3735, pruned_loss=0.1249, over 5657755.38 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:34:16,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:34:32,664 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:34:54,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.554e+02 1.701e+03 2.309e+03 3.260e+03 7.495e+03, threshold=4.618e+03, percent-clipped=8.0
+2023-03-11 20:34:57,242 INFO [train.py:968] (0/2) Epoch 23, batch 11400, giga_loss[loss=0.3184, simple_loss=0.3802, pruned_loss=0.1283, over 28550.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3719, pruned_loss=0.1232, over 5667873.99 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.0892, over 5751508.40 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3744, pruned_loss=0.1263, over 5652566.12 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:35:47,506 INFO [train.py:968] (0/2) Epoch 23, batch 11450, giga_loss[loss=0.3592, simple_loss=0.3896, pruned_loss=0.1644, over 23731.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3715, pruned_loss=0.1234, over 5661980.97 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3434, pruned_loss=0.08931, over 5750622.89 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3743, pruned_loss=0.127, over 5647643.36 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:36:22,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9960, 3.8017, 3.6516, 1.7191], device='cuda:0'), covar=tensor([0.0785, 0.0949, 0.0844, 0.2141], device='cuda:0'), in_proj_covar=tensor([0.1264, 0.1169, 0.0989, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 20:36:34,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.195e+03 1.798e+03 2.293e+03 3.131e+03 9.930e+03, threshold=4.586e+03, percent-clipped=7.0
+2023-03-11 20:36:35,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-11 20:36:36,237 INFO [train.py:968] (0/2) Epoch 23, batch 11500, giga_loss[loss=0.3282, simple_loss=0.3888, pruned_loss=0.1338, over 28958.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5677554.70 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3437, pruned_loss=0.08968, over 5754922.72 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3731, pruned_loss=0.1257, over 5659949.89 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:36:38,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:36:38,733 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:36:42,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:36:44,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9465, 1.1424, 1.2353, 1.0234], device='cuda:0'), covar=tensor([0.1405, 0.1011, 0.1710, 0.1184], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0758, 0.0723, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 20:37:01,770 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:37:10,296 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:37:24,668 INFO [train.py:968] (0/2) Epoch 23, batch 11550, libri_loss[loss=0.2739, simple_loss=0.3624, pruned_loss=0.09272, over 29181.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3707, pruned_loss=0.1222, over 5671798.88 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3433, pruned_loss=0.08952, over 5757416.12 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3736, pruned_loss=0.1257, over 5653944.56 frames. ], batch size: 97, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:37:24,985 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3971, 3.3796, 1.5498, 1.5105], device='cuda:0'), covar=tensor([0.0994, 0.0402, 0.0884, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0561, 0.0391, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 20:38:05,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.574e+03 2.047e+03 2.525e+03 5.887e+03, threshold=4.094e+03, percent-clipped=3.0
+2023-03-11 20:38:07,019 INFO [train.py:968] (0/2) Epoch 23, batch 11600, giga_loss[loss=0.3174, simple_loss=0.3862, pruned_loss=0.1243, over 29106.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3711, pruned_loss=0.1219, over 5675611.26 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3432, pruned_loss=0.08942, over 5753685.51 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3745, pruned_loss=0.1259, over 5662105.17 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:38:59,535 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:01,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:02,076 INFO [train.py:968] (0/2) Epoch 23, batch 11650, giga_loss[loss=0.377, simple_loss=0.4131, pruned_loss=0.1705, over 27617.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3727, pruned_loss=0.1231, over 5681282.75 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3432, pruned_loss=0.08943, over 5756859.37 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3761, pruned_loss=0.127, over 5666221.79 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:39:22,039 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:23,840 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:23,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:32,356 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:33,593 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:50,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.252e+03 1.780e+03 2.462e+03 3.602e+03 7.043e+03, threshold=4.923e+03, percent-clipped=16.0
+2023-03-11 20:39:53,009 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:53,544 INFO [train.py:968] (0/2) Epoch 23, batch 11700, giga_loss[loss=0.2534, simple_loss=0.3374, pruned_loss=0.08472, over 28420.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3732, pruned_loss=0.1239, over 5684759.34 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.0893, over 5757420.07 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3762, pruned_loss=0.1272, over 5672069.26 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:40:09,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:40:10,481 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:40:11,590 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-11 20:40:40,730 INFO [train.py:968] (0/2) Epoch 23, batch 11750, giga_loss[loss=0.3922, simple_loss=0.4219, pruned_loss=0.1813, over 26487.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3736, pruned_loss=0.1231, over 5685487.13 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3432, pruned_loss=0.08951, over 5757808.77 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3762, pruned_loss=0.1262, over 5673869.44 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:40:40,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:41:26,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.536e+02 1.684e+03 2.176e+03 2.791e+03 1.227e+04, threshold=4.352e+03, percent-clipped=4.0
+2023-03-11 20:41:29,672 INFO [train.py:968] (0/2) Epoch 23, batch 11800, giga_loss[loss=0.3083, simple_loss=0.3756, pruned_loss=0.1205, over 27933.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3728, pruned_loss=0.1216, over 5683521.49 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3429, pruned_loss=0.08949, over 5759552.64 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3757, pruned_loss=0.1246, over 5671380.56 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:41:51,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3998, 1.6902, 1.5268, 1.3933], device='cuda:0'), covar=tensor([0.2218, 0.2155, 0.2383, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0756, 0.0720, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 20:42:16,528 INFO [train.py:968] (0/2) Epoch 23, batch 11850, giga_loss[loss=0.3249, simple_loss=0.3832, pruned_loss=0.1333, over 28592.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3722, pruned_loss=0.1214, over 5669900.75 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08938, over 5752995.34 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3755, pruned_loss=0.1248, over 5663840.16 frames. ], batch size: 85, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:42:28,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:42:31,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:42:47,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1714, 2.6863, 1.8964, 2.4055], device='cuda:0'), covar=tensor([0.0950, 0.0524, 0.0901, 0.0925], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0451, 0.0522, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 20:42:57,898 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:43:00,484 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:43:03,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:43:04,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.550e+03 2.034e+03 2.921e+03 9.436e+03, threshold=4.069e+03, percent-clipped=9.0
+2023-03-11 20:43:06,667 INFO [train.py:968] (0/2) Epoch 23, batch 11900, giga_loss[loss=0.2859, simple_loss=0.355, pruned_loss=0.1084, over 29053.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3701, pruned_loss=0.1201, over 5684072.90 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3425, pruned_loss=0.0893, over 5756160.00 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1234, over 5675174.89 frames. ], batch size: 106, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:43:29,660 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:43:53,598 INFO [train.py:968] (0/2) Epoch 23, batch 11950, giga_loss[loss=0.3211, simple_loss=0.383, pruned_loss=0.1296, over 28894.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3716, pruned_loss=0.1212, over 5663458.42 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3434, pruned_loss=0.08973, over 5750530.43 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3746, pruned_loss=0.1247, over 5658909.67 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:44:37,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.716e+03 2.135e+03 2.793e+03 5.289e+03, threshold=4.270e+03, percent-clipped=6.0
+2023-03-11 20:44:40,319 INFO [train.py:968] (0/2) Epoch 23, batch 12000, giga_loss[loss=0.2702, simple_loss=0.3455, pruned_loss=0.09749, over 28820.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3726, pruned_loss=0.1211, over 5674034.63 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.343, pruned_loss=0.08956, over 5753201.69 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.376, pruned_loss=0.1247, over 5665998.11 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:44:40,323 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 20:44:49,124 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2074, simple_loss=0.3148, pruned_loss=0.05002, over 944034.00 frames. 
+2023-03-11 20:44:49,125 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 20:45:29,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:45:34,719 INFO [train.py:968] (0/2) Epoch 23, batch 12050, libri_loss[loss=0.2233, simple_loss=0.3035, pruned_loss=0.07154, over 28526.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3726, pruned_loss=0.1224, over 5666405.66 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3427, pruned_loss=0.08942, over 5747232.94 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3768, pruned_loss=0.1265, over 5662358.83 frames. ], batch size: 63, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:45:37,318 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:45:41,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:46:00,730 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015467.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:46:08,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-11 20:46:15,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:46:21,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.624e+02 1.619e+03 1.977e+03 2.496e+03 5.170e+03, threshold=3.954e+03, percent-clipped=3.0
+2023-03-11 20:46:23,762 INFO [train.py:968] (0/2) Epoch 23, batch 12100, giga_loss[loss=0.2762, simple_loss=0.3497, pruned_loss=0.1014, over 28853.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3735, pruned_loss=0.1235, over 5663563.03 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.343, pruned_loss=0.08961, over 5750722.75 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3773, pruned_loss=0.1275, over 5655237.31 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:46:54,511 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-11 20:47:03,469 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2012, 1.4430, 1.3088, 1.0967], device='cuda:0'), covar=tensor([0.2403, 0.2291, 0.1673, 0.2262], device='cuda:0'), in_proj_covar=tensor([0.1986, 0.1943, 0.1860, 0.1998], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 20:47:13,382 INFO [train.py:968] (0/2) Epoch 23, batch 12150, giga_loss[loss=0.3336, simple_loss=0.4054, pruned_loss=0.1309, over 28903.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3742, pruned_loss=0.1241, over 5666584.52 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3429, pruned_loss=0.08957, over 5752217.89 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3778, pruned_loss=0.128, over 5657008.24 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:47:28,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:47:53,289 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:47:56,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:02,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.828e+03 2.088e+03 2.681e+03 5.311e+03, threshold=4.175e+03, percent-clipped=5.0
+2023-03-11 20:48:02,753 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:03,796 INFO [train.py:968] (0/2) Epoch 23, batch 12200, giga_loss[loss=0.3071, simple_loss=0.3826, pruned_loss=0.1159, over 28793.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3764, pruned_loss=0.1256, over 5665118.40 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3429, pruned_loss=0.08963, over 5753693.82 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3796, pruned_loss=0.1289, over 5655426.04 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:48:05,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:23,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:33,607 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:36,486 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:39,449 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:42,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:50,839 INFO [train.py:968] (0/2) Epoch 23, batch 12250, giga_loss[loss=0.4483, simple_loss=0.4582, pruned_loss=0.2192, over 26738.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3756, pruned_loss=0.1256, over 5657516.36 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3427, pruned_loss=0.08955, over 5756931.83 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3797, pruned_loss=0.1298, over 5643162.70 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:49:08,967 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:49:09,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3012, 1.7278, 1.2738, 0.7342], device='cuda:0'), covar=tensor([0.4494, 0.2443, 0.2771, 0.5446], device='cuda:0'), in_proj_covar=tensor([0.1763, 0.1671, 0.1602, 0.1443], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 20:49:32,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3699, 3.9089, 1.5215, 1.7396], device='cuda:0'), covar=tensor([0.1042, 0.0271, 0.0982, 0.1307], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0562, 0.0392, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 20:49:36,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.643e+03 2.186e+03 3.192e+03 7.458e+03, threshold=4.371e+03, percent-clipped=12.0
+2023-03-11 20:49:39,110 INFO [train.py:968] (0/2) Epoch 23, batch 12300, giga_loss[loss=0.2856, simple_loss=0.3575, pruned_loss=0.1068, over 28896.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3755, pruned_loss=0.1252, over 5649970.59 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3431, pruned_loss=0.08978, over 5752588.49 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3792, pruned_loss=0.1294, over 5638721.01 frames. ], batch size: 106, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:49:39,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8555, 1.1211, 2.8787, 2.8028], device='cuda:0'), covar=tensor([0.1676, 0.2611, 0.0606, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0660, 0.0978, 0.0931], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 20:49:43,631 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:50:26,621 INFO [train.py:968] (0/2) Epoch 23, batch 12350, giga_loss[loss=0.2744, simple_loss=0.3494, pruned_loss=0.09972, over 28418.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3768, pruned_loss=0.1261, over 5649179.72 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3432, pruned_loss=0.08981, over 5753069.87 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3801, pruned_loss=0.1297, over 5638885.86 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:51:14,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.734e+02 1.663e+03 2.053e+03 2.673e+03 6.116e+03, threshold=4.106e+03, percent-clipped=4.0
+2023-03-11 20:51:16,545 INFO [train.py:968] (0/2) Epoch 23, batch 12400, giga_loss[loss=0.2761, simple_loss=0.3517, pruned_loss=0.1003, over 28313.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3762, pruned_loss=0.1255, over 5653316.44 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.08987, over 5755413.53 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3793, pruned_loss=0.1289, over 5641548.77 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:51:42,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:52:01,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1841, 2.2115, 1.7531, 1.7727], device='cuda:0'), covar=tensor([0.0965, 0.0731, 0.0987, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0450, 0.0523, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 20:52:03,438 INFO [train.py:968] (0/2) Epoch 23, batch 12450, giga_loss[loss=0.2767, simple_loss=0.3433, pruned_loss=0.1051, over 28440.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3732, pruned_loss=0.1231, over 5664922.56 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3435, pruned_loss=0.09004, over 5756722.73 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3763, pruned_loss=0.1267, over 5651432.53 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:52:06,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015842.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:52:09,351 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3851, 1.3411, 4.3504, 3.4697], device='cuda:0'), covar=tensor([0.1680, 0.2836, 0.0434, 0.1123], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0660, 0.0979, 0.0933], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 20:52:55,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.697e+02 1.853e+03 2.419e+03 3.220e+03 8.496e+03, threshold=4.838e+03, percent-clipped=12.0
+2023-03-11 20:52:55,745 INFO [train.py:968] (0/2) Epoch 23, batch 12500, giga_loss[loss=0.2995, simple_loss=0.355, pruned_loss=0.122, over 28774.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3707, pruned_loss=0.1217, over 5672737.18 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3435, pruned_loss=0.09001, over 5758598.45 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3735, pruned_loss=0.125, over 5659304.53 frames. ], batch size: 66, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:53:05,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2912, 1.3718, 3.4542, 3.0776], device='cuda:0'), covar=tensor([0.1473, 0.2575, 0.0471, 0.1605], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0659, 0.0977, 0.0931], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 20:53:39,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:53:49,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-11 20:53:49,868 INFO [train.py:968] (0/2) Epoch 23, batch 12550, giga_loss[loss=0.2568, simple_loss=0.324, pruned_loss=0.09477, over 28799.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3661, pruned_loss=0.1195, over 5657202.31 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3436, pruned_loss=0.09009, over 5750559.61 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3685, pruned_loss=0.1222, over 5652518.24 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:54:10,075 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:54:13,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:54:32,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015985.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:54:34,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015988.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:54:37,078 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.126e+03 1.842e+03 2.775e+03 3.889e+03 1.509e+04, threshold=5.550e+03, percent-clipped=15.0
+2023-03-11 20:54:37,761 INFO [train.py:968] (0/2) Epoch 23, batch 12600, giga_loss[loss=0.2645, simple_loss=0.3405, pruned_loss=0.09425, over 28871.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3655, pruned_loss=0.12, over 5657665.32 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09031, over 5753757.32 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3674, pruned_loss=0.1225, over 5649563.84 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:54:39,749 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:54:46,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:54:47,260 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1016000.pt
+2023-03-11 20:55:04,684 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016017.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:55:28,080 INFO [train.py:968] (0/2) Epoch 23, batch 12650, giga_loss[loss=0.3189, simple_loss=0.3766, pruned_loss=0.1306, over 27966.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3663, pruned_loss=0.1211, over 5647401.06 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3445, pruned_loss=0.09054, over 5755063.32 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.368, pruned_loss=0.1237, over 5637335.19 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:55:32,977 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:55:57,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4380, 1.5179, 1.4885, 1.5829], device='cuda:0'), covar=tensor([0.0640, 0.0301, 0.0286, 0.0658], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0119, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 20:55:59,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:56:02,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:56:05,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:56:20,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.686e+02 1.696e+03 2.045e+03 2.796e+03 5.384e+03, threshold=4.091e+03, percent-clipped=0.0
+2023-03-11 20:56:22,730 INFO [train.py:968] (0/2) Epoch 23, batch 12700, giga_loss[loss=0.2587, simple_loss=0.3502, pruned_loss=0.08361, over 28687.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3651, pruned_loss=0.1195, over 5648484.78 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3445, pruned_loss=0.09054, over 5755063.32 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3664, pruned_loss=0.1215, over 5640650.38 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:56:38,370 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:57:00,028 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016128.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:57:08,020 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 20:57:13,646 INFO [train.py:968] (0/2) Epoch 23, batch 12750, giga_loss[loss=0.3139, simple_loss=0.3864, pruned_loss=0.1207, over 28246.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3638, pruned_loss=0.1161, over 5650535.84 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3443, pruned_loss=0.09053, over 5755653.22 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3652, pruned_loss=0.1181, over 5642336.32 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:57:14,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:57:16,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:57:27,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0744, 4.9225, 4.6474, 2.7708], device='cuda:0'), covar=tensor([0.0495, 0.0615, 0.0801, 0.1557], device='cuda:0'), in_proj_covar=tensor([0.1260, 0.1165, 0.0989, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 20:57:31,640 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:57:49,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:58:04,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.670e+03 2.304e+03 3.447e+03 9.434e+03, threshold=4.609e+03, percent-clipped=15.0
+2023-03-11 20:58:04,963 INFO [train.py:968] (0/2) Epoch 23, batch 12800, giga_loss[loss=0.2807, simple_loss=0.3543, pruned_loss=0.1036, over 28517.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3601, pruned_loss=0.1118, over 5649447.26 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3436, pruned_loss=0.09017, over 5754122.18 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3625, pruned_loss=0.1144, over 5640999.14 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:58:29,301 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:58:31,277 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:58:48,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2351, 3.2369, 1.3456, 1.5543], device='cuda:0'), covar=tensor([0.1238, 0.0425, 0.1120, 0.1529], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0560, 0.0391, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 20:58:59,899 INFO [train.py:968] (0/2) Epoch 23, batch 12850, giga_loss[loss=0.2819, simple_loss=0.354, pruned_loss=0.105, over 28537.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3571, pruned_loss=0.1086, over 5650618.33 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3437, pruned_loss=0.09035, over 5756081.27 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3591, pruned_loss=0.1108, over 5640927.57 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:59:03,767 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:59:47,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.716e+02 1.550e+03 2.164e+03 2.813e+03 9.731e+03, threshold=4.328e+03, percent-clipped=3.0
+2023-03-11 20:59:47,027 INFO [train.py:968] (0/2) Epoch 23, batch 12900, giga_loss[loss=0.2709, simple_loss=0.3547, pruned_loss=0.09354, over 28645.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3532, pruned_loss=0.1053, over 5640551.30 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.343, pruned_loss=0.0902, over 5750544.09 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3559, pruned_loss=0.1079, over 5633528.72 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:00:34,343 INFO [train.py:968] (0/2) Epoch 23, batch 12950, giga_loss[loss=0.2645, simple_loss=0.3535, pruned_loss=0.08779, over 28932.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3524, pruned_loss=0.1024, over 5656314.38 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.343, pruned_loss=0.09045, over 5753685.21 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3548, pruned_loss=0.1046, over 5644939.96 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:01:30,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.510e+02 1.268e+03 1.657e+03 2.254e+03 7.898e+03, threshold=3.313e+03, percent-clipped=2.0
+2023-03-11 21:01:30,974 INFO [train.py:968] (0/2) Epoch 23, batch 13000, giga_loss[loss=0.2838, simple_loss=0.3691, pruned_loss=0.09922, over 28889.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3519, pruned_loss=0.1016, over 5656019.63 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3424, pruned_loss=0.0902, over 5757124.80 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3546, pruned_loss=0.1038, over 5642229.74 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:01:51,850 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:02:02,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:02:20,535 INFO [train.py:968] (0/2) Epoch 23, batch 13050, giga_loss[loss=0.2285, simple_loss=0.3173, pruned_loss=0.0699, over 28907.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3508, pruned_loss=0.1007, over 5657085.29 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.342, pruned_loss=0.09004, over 5758258.93 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3535, pruned_loss=0.1028, over 5643269.46 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:02:52,767 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:02:54,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3840, 1.7384, 1.3168, 1.6238], device='cuda:0'), covar=tensor([0.0809, 0.0301, 0.0353, 0.0901], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 21:03:10,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.180e+02 1.301e+03 1.614e+03 2.083e+03 6.552e+03, threshold=3.228e+03, percent-clipped=8.0
+2023-03-11 21:03:10,559 INFO [train.py:968] (0/2) Epoch 23, batch 13100, giga_loss[loss=0.2252, simple_loss=0.3127, pruned_loss=0.06883, over 28351.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3477, pruned_loss=0.09844, over 5655562.46 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3418, pruned_loss=0.08998, over 5761801.50 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3502, pruned_loss=0.1004, over 5638905.43 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:03:23,138 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016503.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:03:51,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:04:00,045 INFO [train.py:968] (0/2) Epoch 23, batch 13150, giga_loss[loss=0.3337, simple_loss=0.3963, pruned_loss=0.1355, over 28721.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3462, pruned_loss=0.09764, over 5655184.82 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3414, pruned_loss=0.08972, over 5764783.12 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3487, pruned_loss=0.09959, over 5637344.20 frames. ], batch size: 284, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:04:07,616 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8847, 1.0003, 0.8986, 0.8605], device='cuda:0'), covar=tensor([0.1606, 0.1793, 0.1328, 0.1622], device='cuda:0'), in_proj_covar=tensor([0.1952, 0.1908, 0.1834, 0.1962], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 21:04:26,682 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:04:29,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:04:42,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7137, 1.9487, 1.9415, 1.4653], device='cuda:0'), covar=tensor([0.1881, 0.2692, 0.1564, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.0905, 0.0704, 0.0950, 0.0851], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 21:04:50,833 INFO [train.py:968] (0/2) Epoch 23, batch 13200, giga_loss[loss=0.2622, simple_loss=0.3395, pruned_loss=0.09245, over 28764.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.346, pruned_loss=0.09752, over 5654522.39 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3408, pruned_loss=0.08945, over 5767023.66 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3486, pruned_loss=0.09945, over 5636652.16 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:04:51,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.376e+02 1.486e+03 1.980e+03 2.794e+03 6.378e+03, threshold=3.960e+03, percent-clipped=16.0
+2023-03-11 21:04:56,711 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:04:58,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3971, 4.2466, 4.0068, 1.9713], device='cuda:0'), covar=tensor([0.0531, 0.0651, 0.0789, 0.2113], device='cuda:0'), in_proj_covar=tensor([0.1241, 0.1149, 0.0973, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 21:05:35,746 INFO [train.py:968] (0/2) Epoch 23, batch 13250, giga_loss[loss=0.2307, simple_loss=0.3208, pruned_loss=0.07033, over 28857.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3447, pruned_loss=0.0962, over 5663508.62 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3401, pruned_loss=0.0891, over 5767746.54 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3477, pruned_loss=0.0984, over 5643523.32 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:05:40,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8904, 2.3339, 2.3662, 1.8477], device='cuda:0'), covar=tensor([0.2842, 0.1910, 0.1793, 0.2237], device='cuda:0'), in_proj_covar=tensor([0.1951, 0.1907, 0.1830, 0.1958], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 21:05:40,509 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016646.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:05:43,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016649.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 21:06:13,012 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:06:15,689 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016678.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 21:06:15,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:06:28,658 INFO [train.py:968] (0/2) Epoch 23, batch 13300, giga_loss[loss=0.23, simple_loss=0.2969, pruned_loss=0.08153, over 24118.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3416, pruned_loss=0.09396, over 5657472.83 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3396, pruned_loss=0.08904, over 5770316.81 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3444, pruned_loss=0.09586, over 5637699.95 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:06:30,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.734e+02 1.380e+03 1.733e+03 2.403e+03 9.101e+03, threshold=3.466e+03, percent-clipped=3.0
+2023-03-11 21:06:42,855 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:07:20,181 INFO [train.py:968] (0/2) Epoch 23, batch 13350, giga_loss[loss=0.3023, simple_loss=0.3634, pruned_loss=0.1206, over 28866.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3371, pruned_loss=0.09134, over 5658734.92 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3387, pruned_loss=0.08866, over 5774216.68 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3402, pruned_loss=0.0933, over 5636273.50 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 1.0
+2023-03-11 21:08:15,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:08:19,952 INFO [train.py:968] (0/2) Epoch 23, batch 13400, giga_loss[loss=0.2188, simple_loss=0.3017, pruned_loss=0.06798, over 28880.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3363, pruned_loss=0.09122, over 5668386.21 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3385, pruned_loss=0.08854, over 5775509.96 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3388, pruned_loss=0.09293, over 5647470.00 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 1.0
+2023-03-11 21:08:23,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.210e+02 1.545e+03 1.937e+03 2.930e+03 7.577e+03, threshold=3.873e+03, percent-clipped=14.0
+2023-03-11 21:08:23,846 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:08:31,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8697, 2.1680, 1.7094, 2.0777], device='cuda:0'), covar=tensor([0.2842, 0.2643, 0.3147, 0.2414], device='cuda:0'), in_proj_covar=tensor([0.1531, 0.1102, 0.1355, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 21:09:13,757 INFO [train.py:968] (0/2) Epoch 23, batch 13450, giga_loss[loss=0.2679, simple_loss=0.3526, pruned_loss=0.09154, over 29016.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3356, pruned_loss=0.09155, over 5655670.20 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3383, pruned_loss=0.08843, over 5777387.09 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3378, pruned_loss=0.09307, over 5635839.94 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 1.0
+2023-03-11 21:09:18,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:10:12,274 INFO [train.py:968] (0/2) Epoch 23, batch 13500, giga_loss[loss=0.278, simple_loss=0.3622, pruned_loss=0.0969, over 28691.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3363, pruned_loss=0.09158, over 5659988.56 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3377, pruned_loss=0.08831, over 5780742.52 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3385, pruned_loss=0.093, over 5638198.12 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 1.0
+2023-03-11 21:10:16,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.658e+02 1.584e+03 2.270e+03 2.784e+03 5.771e+03, threshold=4.539e+03, percent-clipped=8.0
+2023-03-11 21:10:51,034 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:10:58,529 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:11:02,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:11:10,128 INFO [train.py:968] (0/2) Epoch 23, batch 13550, giga_loss[loss=0.2823, simple_loss=0.364, pruned_loss=0.1002, over 28376.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3394, pruned_loss=0.09218, over 5656098.62 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3373, pruned_loss=0.08807, over 5778500.75 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3416, pruned_loss=0.09363, over 5637987.71 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 1.0
+2023-03-11 21:11:39,628 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:12:07,964 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:12:09,740 INFO [train.py:968] (0/2) Epoch 23, batch 13600, giga_loss[loss=0.2667, simple_loss=0.348, pruned_loss=0.09266, over 28929.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3389, pruned_loss=0.09134, over 5661640.55 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3371, pruned_loss=0.088, over 5773167.98 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3409, pruned_loss=0.0927, over 5648177.54 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:12:11,354 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:12:13,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.594e+02 1.463e+03 1.768e+03 2.265e+03 7.426e+03, threshold=3.536e+03, percent-clipped=2.0
+2023-03-11 21:12:48,125 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1017021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:13:06,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1017035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:13:09,674 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-11 21:13:13,889 INFO [train.py:968] (0/2) Epoch 23, batch 13650, giga_loss[loss=0.2646, simple_loss=0.3361, pruned_loss=0.09656, over 27722.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3379, pruned_loss=0.09077, over 5666818.54 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.337, pruned_loss=0.08799, over 5775553.24 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3397, pruned_loss=0.09194, over 5651565.05 frames. ], batch size: 474, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:14:18,116 INFO [train.py:968] (0/2) Epoch 23, batch 13700, giga_loss[loss=0.3162, simple_loss=0.3817, pruned_loss=0.1254, over 26782.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3364, pruned_loss=0.08937, over 5674102.34 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.337, pruned_loss=0.08798, over 5778041.57 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3379, pruned_loss=0.09034, over 5657864.07 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:14:21,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.837e+02 1.302e+03 1.642e+03 2.285e+03 5.476e+03, threshold=3.284e+03, percent-clipped=9.0
+2023-03-11 21:15:22,120 INFO [train.py:968] (0/2) Epoch 23, batch 13750, giga_loss[loss=0.3307, simple_loss=0.3739, pruned_loss=0.1437, over 26898.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3347, pruned_loss=0.08706, over 5672271.56 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3369, pruned_loss=0.08797, over 5778701.37 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3359, pruned_loss=0.08783, over 5658453.06 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:15:36,400 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5028, 2.2981, 1.6384, 0.6752], device='cuda:0'), covar=tensor([0.5364, 0.3107, 0.4182, 0.6594], device='cuda:0'), in_proj_covar=tensor([0.1745, 0.1648, 0.1589, 0.1427], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 21:15:57,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1017169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:16:25,084 INFO [train.py:968] (0/2) Epoch 23, batch 13800, giga_loss[loss=0.3079, simple_loss=0.3542, pruned_loss=0.1308, over 26821.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3329, pruned_loss=0.08731, over 5667986.91 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.337, pruned_loss=0.08817, over 5779016.08 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3337, pruned_loss=0.08771, over 5653030.09 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:16:29,905 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.525e+02 1.467e+03 1.793e+03 2.485e+03 6.173e+03, threshold=3.587e+03, percent-clipped=14.0
+2023-03-11 21:17:01,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5890, 1.9033, 1.4941, 1.7863], device='cuda:0'), covar=tensor([0.2903, 0.2676, 0.3102, 0.2579], device='cuda:0'), in_proj_covar=tensor([0.1528, 0.1099, 0.1355, 0.0997], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 21:17:27,601 INFO [train.py:968] (0/2) Epoch 23, batch 13850, giga_loss[loss=0.2317, simple_loss=0.3016, pruned_loss=0.0809, over 24658.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3317, pruned_loss=0.08706, over 5663414.85 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.337, pruned_loss=0.08832, over 5770071.53 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3323, pruned_loss=0.08723, over 5657796.18 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:18:08,554 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-11 21:18:31,034 INFO [train.py:968] (0/2) Epoch 23, batch 13900, giga_loss[loss=0.2789, simple_loss=0.3609, pruned_loss=0.09845, over 28891.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3307, pruned_loss=0.08701, over 5663043.99 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3365, pruned_loss=0.08806, over 5772911.99 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3315, pruned_loss=0.08736, over 5653992.89 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:18:35,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.838e+02 1.304e+03 1.755e+03 2.731e+03 1.105e+04, threshold=3.510e+03, percent-clipped=12.0
+2023-03-11 21:18:39,466 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3650, 1.6442, 1.4897, 1.5291], device='cuda:0'), covar=tensor([0.0766, 0.0293, 0.0321, 0.0903], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 21:18:40,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1017299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:18:54,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1017312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:19:00,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1017315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:19:31,040 INFO [train.py:968] (0/2) Epoch 23, batch 13950, giga_loss[loss=0.272, simple_loss=0.3509, pruned_loss=0.09653, over 28858.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3339, pruned_loss=0.08854, over 5653745.56 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3367, pruned_loss=0.0883, over 5765275.37 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3342, pruned_loss=0.0886, over 5650414.19 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:19:35,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1017344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:19:37,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3174, 1.9645, 1.3971, 0.5746], device='cuda:0'), covar=tensor([0.4160, 0.2354, 0.3481, 0.5400], device='cuda:0'), in_proj_covar=tensor([0.1745, 0.1648, 0.1588, 0.1427], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 21:19:40,434 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5572, 1.6624, 1.8615, 1.4352], device='cuda:0'), covar=tensor([0.1823, 0.2307, 0.1456, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.0907, 0.0700, 0.0952, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 21:19:42,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2323, 1.4664, 1.4083, 1.1628], device='cuda:0'), covar=tensor([0.2085, 0.2154, 0.1452, 0.1974], device='cuda:0'), in_proj_covar=tensor([0.1940, 0.1889, 0.1810, 0.1945], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 21:20:31,376 INFO [train.py:968] (0/2) Epoch 23, batch 14000, giga_loss[loss=0.2349, simple_loss=0.3159, pruned_loss=0.07694, over 28928.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3339, pruned_loss=0.08768, over 5655752.57 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3364, pruned_loss=0.08801, over 5767331.64 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3344, pruned_loss=0.08801, over 5648668.20 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:20:36,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.528e+02 1.440e+03 2.142e+03 2.685e+03 7.775e+03, threshold=4.285e+03, percent-clipped=12.0
+2023-03-11 21:20:58,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1017410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:21:45,039 INFO [train.py:968] (0/2) Epoch 23, batch 14050, libri_loss[loss=0.2385, simple_loss=0.3148, pruned_loss=0.08104, over 29568.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3312, pruned_loss=0.08611, over 5670066.69 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3364, pruned_loss=0.08812, over 5770289.67 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3316, pruned_loss=0.08624, over 5659841.16 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:21:47,875 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1017442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:21:52,418 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1017445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:22:30,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1017474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:22:54,315 INFO [train.py:968] (0/2) Epoch 23, batch 14100, giga_loss[loss=0.2381, simple_loss=0.3261, pruned_loss=0.07504, over 28710.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3329, pruned_loss=0.08706, over 5682284.83 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3366, pruned_loss=0.08823, over 5772094.92 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3329, pruned_loss=0.08704, over 5671270.99 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:23:00,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.449e+02 1.362e+03 1.847e+03 2.793e+03 1.219e+04, threshold=3.695e+03, percent-clipped=11.0
+2023-03-11 21:24:02,536 INFO [train.py:968] (0/2) Epoch 23, batch 14150, giga_loss[loss=0.2764, simple_loss=0.3694, pruned_loss=0.09171, over 28792.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3368, pruned_loss=0.08746, over 5671340.91 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3363, pruned_loss=0.08817, over 5765876.93 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3371, pruned_loss=0.08749, over 5666114.06 frames. ], batch size: 243, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:24:17,000 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1017553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:24:19,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1017556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:24:54,859 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-11 21:24:57,134 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1017585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:25:02,205 INFO [train.py:968] (0/2) Epoch 23, batch 14200, giga_loss[loss=0.2692, simple_loss=0.3611, pruned_loss=0.0887, over 28641.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3387, pruned_loss=0.08692, over 5677112.92 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3359, pruned_loss=0.08818, over 5769328.67 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3393, pruned_loss=0.08686, over 5665720.57 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:25:05,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.455e+03 1.806e+03 2.515e+03 7.529e+03, threshold=3.612e+03, percent-clipped=10.0
+2023-03-11 21:25:51,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8981, 1.0694, 1.0948, 0.8875], device='cuda:0'), covar=tensor([0.2026, 0.2358, 0.1359, 0.1984], device='cuda:0'), in_proj_covar=tensor([0.1936, 0.1883, 0.1799, 0.1940], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 21:26:01,371 INFO [train.py:968] (0/2) Epoch 23, batch 14250, giga_loss[loss=0.2387, simple_loss=0.3353, pruned_loss=0.07109, over 29007.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3389, pruned_loss=0.08584, over 5679195.13 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3354, pruned_loss=0.08801, over 5773145.54 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3399, pruned_loss=0.08589, over 5663845.87 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:26:10,980 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1017648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:26:31,986 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-11 21:27:03,903 INFO [train.py:968] (0/2) Epoch 23, batch 14300, giga_loss[loss=0.2498, simple_loss=0.3352, pruned_loss=0.08221, over 28965.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3386, pruned_loss=0.08532, over 5680728.59 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3355, pruned_loss=0.08811, over 5775144.50 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3394, pruned_loss=0.08523, over 5665059.99 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:27:09,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.886e+02 1.428e+03 1.871e+03 2.480e+03 7.105e+03, threshold=3.742e+03, percent-clipped=11.0
+2023-03-11 21:27:15,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3476, 2.5293, 1.7204, 2.2361], device='cuda:0'), covar=tensor([0.0857, 0.0545, 0.0936, 0.0931], device='cuda:0'), in_proj_covar=tensor([0.0393, 0.0443, 0.0517, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 21:27:27,867 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1017708.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 21:27:42,921 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-11 21:28:06,386 INFO [train.py:968] (0/2) Epoch 23, batch 14350, giga_loss[loss=0.2535, simple_loss=0.3336, pruned_loss=0.08673, over 28485.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3389, pruned_loss=0.08642, over 5683523.03 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3354, pruned_loss=0.08815, over 5778694.65 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3398, pruned_loss=0.08625, over 5665391.24 frames. ], batch size: 65, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:29:07,215 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-11 21:29:12,192 INFO [train.py:968] (0/2) Epoch 23, batch 14400, giga_loss[loss=0.2702, simple_loss=0.3531, pruned_loss=0.09362, over 28900.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.338, pruned_loss=0.08686, over 5688009.44 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3354, pruned_loss=0.08821, over 5773085.23 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3388, pruned_loss=0.08665, over 5675220.66 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:29:18,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.243e+02 1.376e+03 2.016e+03 2.748e+03 6.377e+03, threshold=4.031e+03, percent-clipped=12.0
+2023-03-11 21:30:33,198 INFO [train.py:968] (0/2) Epoch 23, batch 14450, giga_loss[loss=0.2412, simple_loss=0.322, pruned_loss=0.08027, over 28851.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3391, pruned_loss=0.08845, over 5688991.56 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3354, pruned_loss=0.08823, over 5773777.81 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3397, pruned_loss=0.08827, over 5678087.28 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:31:34,099 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1017875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:31:54,807 INFO [train.py:968] (0/2) Epoch 23, batch 14500, giga_loss[loss=0.2184, simple_loss=0.3075, pruned_loss=0.06469, over 28420.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3355, pruned_loss=0.08685, over 5685359.63 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3356, pruned_loss=0.08844, over 5774411.87 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3359, pruned_loss=0.0865, over 5673705.24 frames. ], batch size: 78, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:32:00,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.382e+03 1.740e+03 2.345e+03 5.008e+03, threshold=3.480e+03, percent-clipped=2.0
+2023-03-11 21:33:01,557 INFO [train.py:968] (0/2) Epoch 23, batch 14550, giga_loss[loss=0.2564, simple_loss=0.3357, pruned_loss=0.08857, over 28692.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3338, pruned_loss=0.08571, over 5688785.51 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3354, pruned_loss=0.08834, over 5774515.10 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3341, pruned_loss=0.08543, over 5676434.98 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:33:16,932 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1017949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:34:03,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2335, 2.9041, 1.4084, 1.3930], device='cuda:0'), covar=tensor([0.0997, 0.0353, 0.0947, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0556, 0.0391, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 21:34:05,585 INFO [train.py:968] (0/2) Epoch 23, batch 14600, giga_loss[loss=0.287, simple_loss=0.3665, pruned_loss=0.1037, over 28708.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3312, pruned_loss=0.08498, over 5686188.16 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3348, pruned_loss=0.08805, over 5777939.92 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.332, pruned_loss=0.08493, over 5670238.76 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:34:10,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.475e+02 1.524e+03 2.046e+03 2.717e+03 1.025e+04, threshold=4.092e+03, percent-clipped=12.0
+2023-03-11 21:34:16,417 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1018000.pt
+2023-03-11 21:34:18,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4765, 1.6210, 1.6737, 1.2600], device='cuda:0'), covar=tensor([0.1668, 0.2531, 0.1403, 0.1700], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0699, 0.0953, 0.0853], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 21:34:22,570 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:34:35,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3340, 1.4055, 1.2265, 1.5632], device='cuda:0'), covar=tensor([0.0747, 0.0411, 0.0363, 0.0799], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-11 21:34:42,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:35:06,167 INFO [train.py:968] (0/2) Epoch 23, batch 14650, giga_loss[loss=0.2713, simple_loss=0.3452, pruned_loss=0.09869, over 28131.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3361, pruned_loss=0.08779, over 5674025.30 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3348, pruned_loss=0.08799, over 5770167.52 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3367, pruned_loss=0.08777, over 5666093.57 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:35:55,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018083.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 21:36:05,963 INFO [train.py:968] (0/2) Epoch 23, batch 14700, giga_loss[loss=0.2349, simple_loss=0.3148, pruned_loss=0.07755, over 28894.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3352, pruned_loss=0.08801, over 5684075.13 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.334, pruned_loss=0.08775, over 5775599.18 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3364, pruned_loss=0.08821, over 5669661.70 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:36:10,369 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.812e+02 1.469e+03 2.000e+03 2.795e+03 7.099e+03, threshold=4.001e+03, percent-clipped=8.0
+2023-03-11 21:36:56,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4204, 2.1518, 1.5806, 0.6463], device='cuda:0'), covar=tensor([0.4365, 0.2547, 0.4177, 0.5748], device='cuda:0'), in_proj_covar=tensor([0.1753, 0.1658, 0.1595, 0.1438], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 21:37:11,159 INFO [train.py:968] (0/2) Epoch 23, batch 14750, giga_loss[loss=0.2529, simple_loss=0.3281, pruned_loss=0.08887, over 28917.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3339, pruned_loss=0.08813, over 5687875.97 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3338, pruned_loss=0.08774, over 5776784.39 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.335, pruned_loss=0.08829, over 5673564.04 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:37:16,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3776, 3.6670, 1.5301, 1.5758], device='cuda:0'), covar=tensor([0.1032, 0.0336, 0.1001, 0.1386], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0555, 0.0392, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 21:37:43,656 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:37:47,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:38:13,592 INFO [train.py:968] (0/2) Epoch 23, batch 14800, giga_loss[loss=0.2293, simple_loss=0.3123, pruned_loss=0.07313, over 28853.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3341, pruned_loss=0.08868, over 5694398.83 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3339, pruned_loss=0.08785, over 5780518.12 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.335, pruned_loss=0.08873, over 5677356.93 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 21:38:17,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.300e+02 1.456e+03 1.814e+03 2.739e+03 6.619e+03, threshold=3.627e+03, percent-clipped=6.0
+2023-03-11 21:38:21,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:39:05,019 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018226.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:39:08,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018229.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:39:22,613 INFO [train.py:968] (0/2) Epoch 23, batch 14850, giga_loss[loss=0.2893, simple_loss=0.3673, pruned_loss=0.1057, over 28426.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3363, pruned_loss=0.08939, over 5687968.68 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3337, pruned_loss=0.08779, over 5781058.53 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3371, pruned_loss=0.0895, over 5673843.79 frames. ], batch size: 369, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:39:38,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:39:48,665 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018258.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:39:57,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4077, 3.2543, 3.0274, 1.8033], device='cuda:0'), covar=tensor([0.0749, 0.0871, 0.0883, 0.1872], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.1134, 0.0963, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 21:40:06,118 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:40:38,719 INFO [train.py:968] (0/2) Epoch 23, batch 14900, giga_loss[loss=0.2772, simple_loss=0.3589, pruned_loss=0.09772, over 28983.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3362, pruned_loss=0.08845, over 5681128.68 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3333, pruned_loss=0.0876, over 5781007.59 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3373, pruned_loss=0.08871, over 5668101.18 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:40:47,025 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.725e+02 1.614e+03 2.360e+03 3.274e+03 1.125e+04, threshold=4.720e+03, percent-clipped=21.0
+2023-03-11 21:41:29,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:41:53,406 INFO [train.py:968] (0/2) Epoch 23, batch 14950, giga_loss[loss=0.2509, simple_loss=0.329, pruned_loss=0.08636, over 29093.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3357, pruned_loss=0.0884, over 5679862.16 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3331, pruned_loss=0.0875, over 5785354.66 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3368, pruned_loss=0.08874, over 5662487.41 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:42:41,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:42:56,303 INFO [train.py:968] (0/2) Epoch 23, batch 15000, giga_loss[loss=0.2288, simple_loss=0.3076, pruned_loss=0.07498, over 29011.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3332, pruned_loss=0.08846, over 5678121.41 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3333, pruned_loss=0.08775, over 5785956.07 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3339, pruned_loss=0.08852, over 5659770.11 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:42:56,307 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 21:43:05,334 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.1963, simple_loss=0.2973, pruned_loss=0.04759, over 944034.00 frames. 
+2023-03-11 21:43:05,335 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 21:43:07,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:43:09,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.830e+02 1.378e+03 1.687e+03 2.453e+03 4.256e+03, threshold=3.373e+03, percent-clipped=0.0
+2023-03-11 21:43:09,954 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:43:49,080 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:44:11,535 INFO [train.py:968] (0/2) Epoch 23, batch 15050, giga_loss[loss=0.2034, simple_loss=0.2924, pruned_loss=0.05714, over 28816.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3273, pruned_loss=0.08565, over 5667484.84 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3333, pruned_loss=0.08769, over 5777023.90 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3279, pruned_loss=0.08574, over 5659041.43 frames. ], batch size: 243, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:44:44,731 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:44:48,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:45:10,968 INFO [train.py:968] (0/2) Epoch 23, batch 15100, giga_loss[loss=0.2696, simple_loss=0.3456, pruned_loss=0.09682, over 28608.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3273, pruned_loss=0.08548, over 5660277.33 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3334, pruned_loss=0.08776, over 5770767.04 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3275, pruned_loss=0.08543, over 5656092.17 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:45:16,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.591e+02 1.546e+03 2.042e+03 3.198e+03 6.729e+03, threshold=4.083e+03, percent-clipped=17.0
+2023-03-11 21:45:20,187 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:45:45,807 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:45:48,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:46:06,529 INFO [train.py:968] (0/2) Epoch 23, batch 15150, giga_loss[loss=0.2522, simple_loss=0.3376, pruned_loss=0.08336, over 28615.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3287, pruned_loss=0.08661, over 5665795.91 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3331, pruned_loss=0.08763, over 5774494.47 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3291, pruned_loss=0.08667, over 5656725.98 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:46:11,655 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 21:46:22,856 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:47:10,726 INFO [train.py:968] (0/2) Epoch 23, batch 15200, giga_loss[loss=0.2934, simple_loss=0.3562, pruned_loss=0.1153, over 27660.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3266, pruned_loss=0.0851, over 5663586.69 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.333, pruned_loss=0.08771, over 5777835.35 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3269, pruned_loss=0.08505, over 5651214.81 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 21:47:16,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3897, 1.2945, 4.1402, 3.4105], device='cuda:0'), covar=tensor([0.1639, 0.2973, 0.0477, 0.1269], device='cuda:0'), in_proj_covar=tensor([0.0761, 0.0654, 0.0958, 0.0909], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 21:47:17,099 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.373e+02 1.299e+03 1.953e+03 2.790e+03 1.059e+04, threshold=3.906e+03, percent-clipped=10.0
+2023-03-11 21:47:29,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4358, 2.1375, 1.6253, 0.5951], device='cuda:0'), covar=tensor([0.4582, 0.3073, 0.3973, 0.6052], device='cuda:0'), in_proj_covar=tensor([0.1759, 0.1661, 0.1596, 0.1443], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 21:47:35,148 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:48:04,511 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:48:12,006 INFO [train.py:968] (0/2) Epoch 23, batch 15250, giga_loss[loss=0.2403, simple_loss=0.3222, pruned_loss=0.07924, over 28597.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3257, pruned_loss=0.08362, over 5671790.66 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3326, pruned_loss=0.0875, over 5781001.91 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.326, pruned_loss=0.08369, over 5656564.38 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 21:48:17,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:48:17,867 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:48:21,874 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:48:50,399 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-11 21:48:51,591 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 21:49:10,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:49:25,506 INFO [train.py:968] (0/2) Epoch 23, batch 15300, giga_loss[loss=0.2448, simple_loss=0.3286, pruned_loss=0.08052, over 28899.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3249, pruned_loss=0.0838, over 5663178.24 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3326, pruned_loss=0.08748, over 5779849.11 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3251, pruned_loss=0.0838, over 5649925.91 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:49:27,022 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2009, 1.3206, 1.1277, 0.9999], device='cuda:0'), covar=tensor([0.1014, 0.0481, 0.1087, 0.1013], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0443, 0.0518, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 21:49:34,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.467e+02 1.439e+03 1.902e+03 2.531e+03 8.962e+03, threshold=3.804e+03, percent-clipped=8.0
+2023-03-11 21:50:31,233 INFO [train.py:968] (0/2) Epoch 23, batch 15350, giga_loss[loss=0.2648, simple_loss=0.3466, pruned_loss=0.09151, over 29020.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3254, pruned_loss=0.08356, over 5659885.59 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3324, pruned_loss=0.08736, over 5780272.36 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3255, pruned_loss=0.08358, over 5646919.14 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:51:07,322 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5005, 1.6688, 1.3925, 1.2434], device='cuda:0'), covar=tensor([0.0956, 0.0485, 0.0963, 0.1054], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0444, 0.0520, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 21:51:33,329 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:51:35,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:51:35,853 INFO [train.py:968] (0/2) Epoch 23, batch 15400, giga_loss[loss=0.2422, simple_loss=0.3182, pruned_loss=0.08311, over 28572.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3253, pruned_loss=0.08345, over 5661852.05 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.332, pruned_loss=0.08707, over 5782842.29 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3257, pruned_loss=0.08366, over 5647166.10 frames. ], batch size: 85, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:51:43,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.403e+02 1.327e+03 1.820e+03 2.586e+03 8.358e+03, threshold=3.639e+03, percent-clipped=6.0
+2023-03-11 21:52:14,888 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:52:44,306 INFO [train.py:968] (0/2) Epoch 23, batch 15450, giga_loss[loss=0.2149, simple_loss=0.297, pruned_loss=0.06642, over 28878.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3255, pruned_loss=0.08427, over 5663893.67 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3318, pruned_loss=0.08695, over 5785741.20 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3259, pruned_loss=0.08449, over 5647700.37 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:53:44,937 INFO [train.py:968] (0/2) Epoch 23, batch 15500, giga_loss[loss=0.2453, simple_loss=0.3326, pruned_loss=0.07898, over 28698.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3255, pruned_loss=0.08371, over 5670260.03 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3316, pruned_loss=0.08688, over 5785430.91 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3259, pruned_loss=0.08389, over 5654956.63 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:53:52,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.058e+02 1.370e+03 1.722e+03 2.252e+03 5.587e+03, threshold=3.445e+03, percent-clipped=4.0
+2023-03-11 21:54:50,810 INFO [train.py:968] (0/2) Epoch 23, batch 15550, giga_loss[loss=0.2366, simple_loss=0.3296, pruned_loss=0.07181, over 28606.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3284, pruned_loss=0.08382, over 5672427.58 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3315, pruned_loss=0.08692, over 5785104.01 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3287, pruned_loss=0.0839, over 5659780.21 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:55:03,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1592, 1.3021, 1.0856, 0.9562], device='cuda:0'), covar=tensor([0.0990, 0.0469, 0.1047, 0.1081], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0444, 0.0519, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 21:55:47,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:55:52,169 INFO [train.py:968] (0/2) Epoch 23, batch 15600, giga_loss[loss=0.2665, simple_loss=0.3588, pruned_loss=0.08706, over 28912.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3305, pruned_loss=0.08465, over 5666899.10 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3314, pruned_loss=0.08684, over 5785309.39 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3308, pruned_loss=0.08472, over 5652906.00 frames. ], batch size: 120, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 21:55:58,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.054e+02 1.412e+03 1.919e+03 2.707e+03 7.200e+03, threshold=3.839e+03, percent-clipped=9.0
+2023-03-11 21:56:13,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:56:28,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:56:34,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:56:53,545 INFO [train.py:968] (0/2) Epoch 23, batch 15650, giga_loss[loss=0.2314, simple_loss=0.3175, pruned_loss=0.07265, over 28864.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3314, pruned_loss=0.08446, over 5674224.54 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3314, pruned_loss=0.08675, over 5785994.09 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3317, pruned_loss=0.08457, over 5661178.18 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:57:12,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:57:14,146 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 21:57:58,308 INFO [train.py:968] (0/2) Epoch 23, batch 15700, giga_loss[loss=0.2237, simple_loss=0.3065, pruned_loss=0.07049, over 29060.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3298, pruned_loss=0.08366, over 5685897.79 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3306, pruned_loss=0.08645, over 5787129.33 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3307, pruned_loss=0.08394, over 5671523.09 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:58:04,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.043e+02 1.417e+03 1.830e+03 2.648e+03 9.567e+03, threshold=3.660e+03, percent-clipped=7.0
+2023-03-11 21:58:12,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1383, 3.9816, 3.7508, 1.8258], device='cuda:0'), covar=tensor([0.0649, 0.0780, 0.0850, 0.2231], device='cuda:0'), in_proj_covar=tensor([0.1222, 0.1128, 0.0956, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 21:58:45,919 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:58:49,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:58:57,917 INFO [train.py:968] (0/2) Epoch 23, batch 15750, giga_loss[loss=0.284, simple_loss=0.3609, pruned_loss=0.1036, over 28464.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3267, pruned_loss=0.0815, over 5691888.47 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3304, pruned_loss=0.08639, over 5787809.40 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3276, pruned_loss=0.08168, over 5677738.11 frames. ], batch size: 369, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:59:15,670 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:21,794 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:28,845 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:31,615 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1019163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:31,705 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:34,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:34,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:40,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:57,283 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:07,611 INFO [train.py:968] (0/2) Epoch 23, batch 15800, giga_loss[loss=0.1939, simple_loss=0.2801, pruned_loss=0.05381, over 29013.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3259, pruned_loss=0.08187, over 5682670.61 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3305, pruned_loss=0.08648, over 5788517.36 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3265, pruned_loss=0.08191, over 5670644.12 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:00:12,661 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:15,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.694e+02 1.361e+03 1.904e+03 2.736e+03 5.957e+03, threshold=3.809e+03, percent-clipped=10.0
+2023-03-11 22:00:16,468 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:16,512 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:19,813 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:28,175 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1019209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:54,241 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:01:06,656 INFO [train.py:968] (0/2) Epoch 23, batch 15850, giga_loss[loss=0.2774, simple_loss=0.3469, pruned_loss=0.104, over 26940.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3241, pruned_loss=0.08168, over 5681425.13 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3293, pruned_loss=0.08595, over 5787152.46 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3254, pruned_loss=0.08204, over 5669940.02 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:02:13,009 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1019290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:02:13,472 INFO [train.py:968] (0/2) Epoch 23, batch 15900, libri_loss[loss=0.2804, simple_loss=0.3474, pruned_loss=0.1067, over 29574.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3277, pruned_loss=0.08337, over 5677189.87 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3295, pruned_loss=0.08604, over 5787341.39 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3284, pruned_loss=0.08346, over 5665492.11 frames. ], batch size: 75, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:02:21,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.675e+02 1.365e+03 1.848e+03 2.852e+03 1.020e+04, threshold=3.696e+03, percent-clipped=6.0
+2023-03-11 22:02:31,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3314, 3.3615, 1.4348, 1.5264], device='cuda:0'), covar=tensor([0.1034, 0.0344, 0.0978, 0.1367], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0551, 0.0390, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 22:03:12,410 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-11 22:03:20,416 INFO [train.py:968] (0/2) Epoch 23, batch 15950, giga_loss[loss=0.2863, simple_loss=0.3483, pruned_loss=0.1122, over 26812.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3288, pruned_loss=0.08391, over 5684292.98 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.329, pruned_loss=0.08571, over 5788621.51 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3298, pruned_loss=0.08424, over 5671498.31 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:04:17,278 INFO [train.py:968] (0/2) Epoch 23, batch 16000, giga_loss[loss=0.2558, simple_loss=0.3395, pruned_loss=0.0861, over 28857.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3311, pruned_loss=0.08593, over 5683164.28 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3292, pruned_loss=0.08578, over 5791862.01 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3318, pruned_loss=0.08609, over 5666033.49 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:04:25,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.288e+02 1.389e+03 1.955e+03 2.668e+03 5.449e+03, threshold=3.911e+03, percent-clipped=10.0
+2023-03-11 22:04:41,747 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1019410.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:05:20,430 INFO [train.py:968] (0/2) Epoch 23, batch 16050, giga_loss[loss=0.267, simple_loss=0.3514, pruned_loss=0.0913, over 27982.00 frames. ], tot_loss[loss=0.255, simple_loss=0.335, pruned_loss=0.08748, over 5682230.60 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3289, pruned_loss=0.08568, over 5784933.13 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3359, pruned_loss=0.0877, over 5673827.65 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:05:30,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1231, 1.2311, 1.0783, 0.8651], device='cuda:0'), covar=tensor([0.1067, 0.0561, 0.1160, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0444, 0.0519, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 22:06:21,447 INFO [train.py:968] (0/2) Epoch 23, batch 16100, giga_loss[loss=0.2418, simple_loss=0.3228, pruned_loss=0.08039, over 27540.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3361, pruned_loss=0.0876, over 5682060.99 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3291, pruned_loss=0.08586, over 5787727.52 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3368, pruned_loss=0.08764, over 5670766.47 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:06:31,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.708e+02 1.496e+03 2.042e+03 2.872e+03 7.785e+03, threshold=4.085e+03, percent-clipped=9.0
+2023-03-11 22:07:29,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:07:33,963 INFO [train.py:968] (0/2) Epoch 23, batch 16150, giga_loss[loss=0.2566, simple_loss=0.3396, pruned_loss=0.08685, over 28430.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3355, pruned_loss=0.08715, over 5688726.81 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.329, pruned_loss=0.08582, over 5789203.13 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3363, pruned_loss=0.08726, over 5676242.72 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:07:52,981 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-11 22:08:26,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1019579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:08:30,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:08:37,085 INFO [train.py:968] (0/2) Epoch 23, batch 16200, giga_loss[loss=0.2358, simple_loss=0.3205, pruned_loss=0.07554, over 29052.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3336, pruned_loss=0.08654, over 5701648.42 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3292, pruned_loss=0.08612, over 5792134.48 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3341, pruned_loss=0.08632, over 5686096.18 frames. ], batch size: 285, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:08:47,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.500e+02 1.359e+03 1.794e+03 2.408e+03 4.844e+03, threshold=3.589e+03, percent-clipped=7.0
+2023-03-11 22:09:21,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5656, 1.5879, 1.7852, 1.3957], device='cuda:0'), covar=tensor([0.1874, 0.2543, 0.1514, 0.1879], device='cuda:0'), in_proj_covar=tensor([0.0901, 0.0694, 0.0948, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 22:09:45,802 INFO [train.py:968] (0/2) Epoch 23, batch 16250, giga_loss[loss=0.2649, simple_loss=0.3349, pruned_loss=0.09744, over 27599.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3337, pruned_loss=0.08689, over 5688286.11 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3294, pruned_loss=0.08622, over 5794086.05 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.334, pruned_loss=0.08664, over 5672756.27 frames. ], batch size: 473, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:10:18,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:10:40,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:10:45,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:10:55,177 INFO [train.py:968] (0/2) Epoch 23, batch 16300, giga_loss[loss=0.2509, simple_loss=0.3302, pruned_loss=0.0858, over 27589.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3321, pruned_loss=0.08679, over 5681818.06 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3291, pruned_loss=0.08613, over 5795359.35 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3326, pruned_loss=0.0867, over 5667358.11 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:11:04,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.888e+02 1.424e+03 1.991e+03 2.797e+03 7.720e+03, threshold=3.983e+03, percent-clipped=16.0
+2023-03-11 22:11:21,853 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-11 22:11:24,761 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:11:43,675 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:11:44,849 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5271, 1.3560, 4.6983, 3.6245], device='cuda:0'), covar=tensor([0.1706, 0.2895, 0.0397, 0.0899], device='cuda:0'), in_proj_covar=tensor([0.0760, 0.0652, 0.0955, 0.0906], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 22:11:46,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:12:00,617 INFO [train.py:968] (0/2) Epoch 23, batch 16350, libri_loss[loss=0.2453, simple_loss=0.331, pruned_loss=0.07985, over 27762.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3309, pruned_loss=0.08713, over 5674565.43 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3293, pruned_loss=0.08626, over 5794558.67 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3312, pruned_loss=0.08696, over 5662982.04 frames. ], batch size: 115, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:12:22,713 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:12:40,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4524, 1.2895, 4.0525, 3.4926], device='cuda:0'), covar=tensor([0.1585, 0.2952, 0.0431, 0.1439], device='cuda:0'), in_proj_covar=tensor([0.0759, 0.0650, 0.0953, 0.0905], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 22:12:55,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019785.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 22:13:02,642 INFO [train.py:968] (0/2) Epoch 23, batch 16400, giga_loss[loss=0.3155, simple_loss=0.3841, pruned_loss=0.1235, over 27533.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3314, pruned_loss=0.08664, over 5681461.89 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3294, pruned_loss=0.08652, over 5793882.39 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3315, pruned_loss=0.0863, over 5669826.10 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:13:12,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.056e+02 1.549e+03 2.012e+03 3.093e+03 1.274e+04, threshold=4.024e+03, percent-clipped=11.0
+2023-03-11 22:13:15,200 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-11 22:13:23,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:13:25,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:13:37,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7081, 4.5492, 4.3055, 2.3794], device='cuda:0'), covar=tensor([0.0661, 0.0805, 0.0990, 0.1793], device='cuda:0'), in_proj_covar=tensor([0.1219, 0.1125, 0.0953, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 22:14:01,267 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:14:02,581 INFO [train.py:968] (0/2) Epoch 23, batch 16450, giga_loss[loss=0.2473, simple_loss=0.3245, pruned_loss=0.08503, over 26812.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3294, pruned_loss=0.0847, over 5680520.02 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3294, pruned_loss=0.08653, over 5796858.90 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3296, pruned_loss=0.0844, over 5666043.38 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:14:15,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3268, 1.4371, 1.3000, 1.5334], device='cuda:0'), covar=tensor([0.0790, 0.0380, 0.0360, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-11 22:14:58,401 INFO [train.py:968] (0/2) Epoch 23, batch 16500, giga_loss[loss=0.2613, simple_loss=0.3562, pruned_loss=0.08323, over 28393.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.331, pruned_loss=0.08376, over 5676420.42 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3289, pruned_loss=0.08622, over 5790318.44 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3318, pruned_loss=0.08376, over 5667372.64 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:15:08,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.630e+02 1.307e+03 1.623e+03 2.030e+03 5.632e+03, threshold=3.246e+03, percent-clipped=2.0
+2023-03-11 22:15:42,113 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019928.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 22:15:45,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019931.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 22:15:55,267 INFO [train.py:968] (0/2) Epoch 23, batch 16550, libri_loss[loss=0.2709, simple_loss=0.3419, pruned_loss=0.09991, over 19155.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3322, pruned_loss=0.08358, over 5671996.33 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3286, pruned_loss=0.08608, over 5783753.88 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3331, pruned_loss=0.08362, over 5668318.46 frames. ], batch size: 187, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:16:09,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:16:15,085 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019960.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 22:16:56,350 INFO [train.py:968] (0/2) Epoch 23, batch 16600, giga_loss[loss=0.279, simple_loss=0.3523, pruned_loss=0.1028, over 27994.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3319, pruned_loss=0.08318, over 5674372.57 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3283, pruned_loss=0.08596, over 5777041.13 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.333, pruned_loss=0.08326, over 5674824.10 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:17:09,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.385e+02 1.452e+03 1.844e+03 2.429e+03 5.862e+03, threshold=3.689e+03, percent-clipped=13.0
+2023-03-11 22:17:09,730 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1020000.pt
+2023-03-11 22:17:47,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4453, 2.0640, 1.5396, 0.6927], device='cuda:0'), covar=tensor([0.5641, 0.3106, 0.4406, 0.6424], device='cuda:0'), in_proj_covar=tensor([0.1763, 0.1664, 0.1601, 0.1442], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 22:18:07,182 INFO [train.py:968] (0/2) Epoch 23, batch 16650, giga_loss[loss=0.2162, simple_loss=0.3134, pruned_loss=0.05949, over 28906.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3324, pruned_loss=0.0835, over 5662631.16 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3285, pruned_loss=0.08603, over 5768318.20 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3331, pruned_loss=0.08346, over 5668386.97 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 22:19:13,115 INFO [train.py:968] (0/2) Epoch 23, batch 16700, giga_loss[loss=0.2284, simple_loss=0.3227, pruned_loss=0.06708, over 28763.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3321, pruned_loss=0.08327, over 5661611.96 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3283, pruned_loss=0.08585, over 5762653.69 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.333, pruned_loss=0.08331, over 5667887.48 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 22:19:23,556 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1020097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:19:25,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1020100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:19:25,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.625e+02 1.547e+03 1.952e+03 2.863e+03 1.707e+04, threshold=3.904e+03, percent-clipped=19.0
+2023-03-11 22:20:05,518 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1020129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:20:26,500 INFO [train.py:968] (0/2) Epoch 23, batch 16750, giga_loss[loss=0.2474, simple_loss=0.3322, pruned_loss=0.08124, over 29043.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3327, pruned_loss=0.08302, over 5647368.52 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3281, pruned_loss=0.08591, over 5737395.07 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3336, pruned_loss=0.08294, over 5670326.00 frames. ], batch size: 165, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 22:21:37,966 INFO [train.py:968] (0/2) Epoch 23, batch 16800, giga_loss[loss=0.274, simple_loss=0.359, pruned_loss=0.09443, over 28629.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3346, pruned_loss=0.08422, over 5661562.24 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3277, pruned_loss=0.08569, over 5743125.59 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.336, pruned_loss=0.08429, over 5672031.02 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:21:53,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.904e+02 1.478e+03 1.975e+03 2.573e+03 7.519e+03, threshold=3.949e+03, percent-clipped=7.0
+2023-03-11 22:22:54,106 INFO [train.py:968] (0/2) Epoch 23, batch 16850, giga_loss[loss=0.2744, simple_loss=0.3594, pruned_loss=0.09473, over 28881.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3369, pruned_loss=0.08473, over 5662514.92 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3277, pruned_loss=0.08561, over 5736723.67 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3381, pruned_loss=0.08484, over 5675272.26 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:23:10,442 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1020254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:23:19,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2962, 3.1373, 1.4105, 1.4430], device='cuda:0'), covar=tensor([0.0990, 0.0387, 0.0967, 0.1328], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0551, 0.0390, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 22:23:50,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0170, 3.1744, 2.0250, 0.9815], device='cuda:0'), covar=tensor([0.8058, 0.3755, 0.4477, 0.7767], device='cuda:0'), in_proj_covar=tensor([0.1758, 0.1661, 0.1595, 0.1438], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 22:23:55,682 INFO [train.py:968] (0/2) Epoch 23, batch 16900, giga_loss[loss=0.2434, simple_loss=0.3259, pruned_loss=0.08047, over 28750.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3367, pruned_loss=0.08519, over 5668921.39 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3274, pruned_loss=0.08536, over 5737613.29 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3383, pruned_loss=0.08549, over 5676124.14 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:24:11,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.307e+02 1.346e+03 1.825e+03 2.536e+03 6.592e+03, threshold=3.650e+03, percent-clipped=7.0
+2023-03-11 22:24:44,179 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4163, 1.9180, 1.3363, 0.7078], device='cuda:0'), covar=tensor([0.5261, 0.2601, 0.3392, 0.6503], device='cuda:0'), in_proj_covar=tensor([0.1758, 0.1659, 0.1596, 0.1438], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 22:25:08,911 INFO [train.py:968] (0/2) Epoch 23, batch 16950, giga_loss[loss=0.2615, simple_loss=0.3398, pruned_loss=0.09161, over 28668.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3347, pruned_loss=0.08483, over 5682644.73 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3275, pruned_loss=0.08541, over 5740203.79 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3361, pruned_loss=0.085, over 5684200.42 frames. ], batch size: 370, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:26:08,456 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 22:26:15,314 INFO [train.py:968] (0/2) Epoch 23, batch 17000, giga_loss[loss=0.2232, simple_loss=0.3169, pruned_loss=0.06475, over 29052.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3313, pruned_loss=0.08237, over 5691631.69 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3272, pruned_loss=0.08532, over 5745628.81 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3329, pruned_loss=0.08252, over 5686297.55 frames. ], batch size: 175, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:26:31,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4480, 1.7912, 1.4406, 1.3800], device='cuda:0'), covar=tensor([0.3095, 0.3033, 0.3532, 0.2582], device='cuda:0'), in_proj_covar=tensor([0.1521, 0.1095, 0.1346, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 22:26:31,469 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.307e+02 1.332e+03 1.556e+03 2.074e+03 6.208e+03, threshold=3.113e+03, percent-clipped=3.0
+2023-03-11 22:27:25,530 INFO [train.py:968] (0/2) Epoch 23, batch 17050, giga_loss[loss=0.2391, simple_loss=0.3264, pruned_loss=0.07593, over 28785.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3296, pruned_loss=0.08092, over 5696892.16 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3272, pruned_loss=0.08522, over 5746253.28 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3309, pruned_loss=0.08104, over 5691264.11 frames. ], batch size: 243, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:28:07,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1020474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:28:27,515 INFO [train.py:968] (0/2) Epoch 23, batch 17100, giga_loss[loss=0.2623, simple_loss=0.347, pruned_loss=0.08881, over 28441.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3307, pruned_loss=0.08183, over 5689043.47 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.327, pruned_loss=0.08508, over 5749290.31 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3321, pruned_loss=0.08198, over 5680848.92 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:28:40,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.810e+02 1.497e+03 1.914e+03 2.736e+03 1.007e+04, threshold=3.828e+03, percent-clipped=18.0
+2023-03-11 22:29:15,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2831, 1.4870, 1.4954, 1.1611], device='cuda:0'), covar=tensor([0.1499, 0.2302, 0.1311, 0.1662], device='cuda:0'), in_proj_covar=tensor([0.0900, 0.0692, 0.0947, 0.0849], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 22:29:28,930 INFO [train.py:968] (0/2) Epoch 23, batch 17150, giga_loss[loss=0.2537, simple_loss=0.3392, pruned_loss=0.08413, over 28959.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.334, pruned_loss=0.08414, over 5688834.04 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3266, pruned_loss=0.08487, over 5752529.78 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3355, pruned_loss=0.08441, over 5678460.85 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:29:44,590 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1020555.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 22:29:55,032 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-11 22:30:24,714 INFO [train.py:968] (0/2) Epoch 23, batch 17200, giga_loss[loss=0.2856, simple_loss=0.3412, pruned_loss=0.115, over 27549.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3327, pruned_loss=0.08372, over 5689579.33 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.326, pruned_loss=0.08454, over 5756243.98 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3346, pruned_loss=0.0842, over 5676615.20 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:30:37,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.257e+02 1.383e+03 1.714e+03 2.558e+03 6.706e+03, threshold=3.428e+03, percent-clipped=7.0
+2023-03-11 22:31:03,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4109, 3.3750, 1.5428, 1.6138], device='cuda:0'), covar=tensor([0.1000, 0.0467, 0.0968, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0551, 0.0391, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 22:31:10,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8253, 5.1373, 1.8702, 2.2457], device='cuda:0'), covar=tensor([0.0896, 0.0248, 0.0882, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0551, 0.0391, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 22:31:12,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1020629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:31:27,889 INFO [train.py:968] (0/2) Epoch 23, batch 17250, giga_loss[loss=0.2263, simple_loss=0.3154, pruned_loss=0.06859, over 28984.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3301, pruned_loss=0.08326, over 5686199.26 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3258, pruned_loss=0.08435, over 5758952.64 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3319, pruned_loss=0.08381, over 5671936.12 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:32:32,004 INFO [train.py:968] (0/2) Epoch 23, batch 17300, giga_loss[loss=0.2594, simple_loss=0.3465, pruned_loss=0.08612, over 28881.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3314, pruned_loss=0.08478, over 5690217.24 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3257, pruned_loss=0.08423, over 5756989.94 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3329, pruned_loss=0.08531, over 5679570.43 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:32:45,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.977e+02 1.650e+03 2.293e+03 4.095e+03 1.214e+04, threshold=4.585e+03, percent-clipped=32.0
+2023-03-11 22:32:47,781 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-11 22:32:49,918 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-11 22:32:53,311 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-11 22:33:28,598 INFO [train.py:968] (0/2) Epoch 23, batch 17350, giga_loss[loss=0.2966, simple_loss=0.3781, pruned_loss=0.1075, over 28846.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.337, pruned_loss=0.08855, over 5681474.30 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.325, pruned_loss=0.08391, over 5758197.56 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3393, pruned_loss=0.08939, over 5669888.10 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:33:58,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1020772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:34:01,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1020775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:34:13,412 INFO [train.py:968] (0/2) Epoch 23, batch 17400, giga_loss[loss=0.2968, simple_loss=0.3766, pruned_loss=0.1085, over 28589.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3447, pruned_loss=0.09217, over 5696167.47 frames. ], libri_tot_loss[loss=0.2463, simple_loss=0.3249, pruned_loss=0.08389, over 5761342.25 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3469, pruned_loss=0.09299, over 5682900.47 frames. ], batch size: 85, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:34:20,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.749e+02 1.442e+03 1.852e+03 2.500e+03 4.217e+03, threshold=3.704e+03, percent-clipped=0.0
+2023-03-11 22:34:22,852 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1020804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:34:59,671 INFO [train.py:968] (0/2) Epoch 23, batch 17450, libri_loss[loss=0.2506, simple_loss=0.3347, pruned_loss=0.08328, over 29515.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3481, pruned_loss=0.09429, over 5700926.22 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3254, pruned_loss=0.08417, over 5764881.79 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3501, pruned_loss=0.09501, over 5684966.93 frames. ], batch size: 81, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:35:05,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1020849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:35:17,019 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-11 22:35:44,827 INFO [train.py:968] (0/2) Epoch 23, batch 17500, giga_loss[loss=0.2072, simple_loss=0.2931, pruned_loss=0.06061, over 29040.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3436, pruned_loss=0.09301, over 5697308.77 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3253, pruned_loss=0.08405, over 5766245.45 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3458, pruned_loss=0.09393, over 5681818.23 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:35:46,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4165, 1.8001, 1.7256, 1.5198], device='cuda:0'), covar=tensor([0.2045, 0.1789, 0.2281, 0.1952], device='cuda:0'), in_proj_covar=tensor([0.0464, 0.0733, 0.0703, 0.0671], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-11 22:35:54,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.238e+02 1.276e+03 1.601e+03 2.200e+03 5.282e+03, threshold=3.203e+03, percent-clipped=2.0
+2023-03-11 22:35:58,452 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6137, 2.2756, 1.7835, 0.7869], device='cuda:0'), covar=tensor([0.6696, 0.3272, 0.3979, 0.7243], device='cuda:0'), in_proj_covar=tensor([0.1756, 0.1661, 0.1597, 0.1435], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 22:36:03,100 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1020914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:36:10,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1020922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:36:17,194 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1020930.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:36:20,661 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4065, 1.3434, 3.3670, 3.2225], device='cuda:0'), covar=tensor([0.1348, 0.2673, 0.0444, 0.0990], device='cuda:0'), in_proj_covar=tensor([0.0762, 0.0654, 0.0961, 0.0913], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 22:36:25,873 INFO [train.py:968] (0/2) Epoch 23, batch 17550, giga_loss[loss=0.2659, simple_loss=0.3332, pruned_loss=0.09933, over 28696.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3367, pruned_loss=0.09017, over 5687192.98 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3254, pruned_loss=0.08417, over 5756639.83 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3388, pruned_loss=0.09103, over 5680957.89 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:37:11,501 INFO [train.py:968] (0/2) Epoch 23, batch 17600, libri_loss[loss=0.2543, simple_loss=0.3504, pruned_loss=0.07906, over 29226.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3302, pruned_loss=0.08757, over 5689249.63 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3257, pruned_loss=0.08421, over 5759017.19 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3318, pruned_loss=0.08831, over 5680883.22 frames. ], batch size: 97, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:37:12,504 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1020992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:37:14,921 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1020995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:37:20,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.509e+02 1.098e+03 1.412e+03 1.763e+03 3.682e+03, threshold=2.824e+03, percent-clipped=5.0
+2023-03-11 22:37:39,697 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:37:55,018 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 22:37:57,224 INFO [train.py:968] (0/2) Epoch 23, batch 17650, libri_loss[loss=0.2339, simple_loss=0.3068, pruned_loss=0.08043, over 27990.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3236, pruned_loss=0.08503, over 5688761.41 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3256, pruned_loss=0.08415, over 5762723.80 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.325, pruned_loss=0.08576, over 5676369.11 frames. ], batch size: 62, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:37:57,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7513, 1.3946, 4.5046, 3.5151], device='cuda:0'), covar=tensor([0.1554, 0.2921, 0.0396, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0763, 0.0654, 0.0962, 0.0914], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 22:38:24,249 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1021073.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:38:27,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1021076.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 22:38:40,211 INFO [train.py:968] (0/2) Epoch 23, batch 17700, giga_loss[loss=0.1916, simple_loss=0.2745, pruned_loss=0.05441, over 28874.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3177, pruned_loss=0.08275, over 5693747.38 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3256, pruned_loss=0.08413, over 5764869.79 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3186, pruned_loss=0.08332, over 5681199.63 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:38:49,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.971e+02 1.206e+03 1.517e+03 2.183e+03 5.297e+03, threshold=3.035e+03, percent-clipped=12.0
+2023-03-11 22:38:52,844 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021105.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:39:03,002 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-11 22:39:23,145 INFO [train.py:968] (0/2) Epoch 23, batch 17750, giga_loss[loss=0.2175, simple_loss=0.2951, pruned_loss=0.06998, over 28914.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3141, pruned_loss=0.08118, over 5699676.36 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3257, pruned_loss=0.08409, over 5765250.42 frames. ], giga_tot_loss[loss=0.2389, simple_loss=0.3146, pruned_loss=0.0816, over 5687191.49 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:39:27,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7787, 2.3744, 1.5024, 0.9833], device='cuda:0'), covar=tensor([0.8659, 0.4279, 0.4264, 0.7352], device='cuda:0'), in_proj_covar=tensor([0.1754, 0.1660, 0.1597, 0.1433], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 22:39:53,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3166, 3.0933, 1.3777, 1.5785], device='cuda:0'), covar=tensor([0.0979, 0.0298, 0.0939, 0.1311], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0550, 0.0390, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 22:40:03,423 INFO [train.py:968] (0/2) Epoch 23, batch 17800, giga_loss[loss=0.2243, simple_loss=0.2924, pruned_loss=0.07812, over 27653.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3114, pruned_loss=0.08008, over 5703889.24 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3259, pruned_loss=0.08417, over 5765702.84 frames. ], giga_tot_loss[loss=0.2359, simple_loss=0.3113, pruned_loss=0.08027, over 5691592.23 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:40:10,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2044, 4.0278, 3.7719, 1.9347], device='cuda:0'), covar=tensor([0.0518, 0.0723, 0.0725, 0.2426], device='cuda:0'), in_proj_covar=tensor([0.1225, 0.1130, 0.0960, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 22:40:14,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.642e+02 1.175e+03 1.541e+03 2.201e+03 6.143e+03, threshold=3.083e+03, percent-clipped=9.0
+2023-03-11 22:40:46,333 INFO [train.py:968] (0/2) Epoch 23, batch 17850, giga_loss[loss=0.2113, simple_loss=0.2806, pruned_loss=0.07104, over 28600.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3089, pruned_loss=0.079, over 5700144.91 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3262, pruned_loss=0.08421, over 5766933.86 frames. ], giga_tot_loss[loss=0.233, simple_loss=0.3081, pruned_loss=0.07898, over 5687338.39 frames. ], batch size: 85, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:40:56,083 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-11 22:41:14,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8247, 2.1378, 1.7610, 2.0118], device='cuda:0'), covar=tensor([0.2574, 0.2665, 0.3040, 0.2552], device='cuda:0'), in_proj_covar=tensor([0.1525, 0.1101, 0.1351, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 22:41:17,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1021278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:41:26,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1021289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:41:27,828 INFO [train.py:968] (0/2) Epoch 23, batch 17900, giga_loss[loss=0.1989, simple_loss=0.277, pruned_loss=0.06041, over 29073.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3062, pruned_loss=0.07762, over 5706709.57 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.327, pruned_loss=0.08457, over 5769475.20 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.3045, pruned_loss=0.07718, over 5693158.73 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:41:35,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1021297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:41:37,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1021300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:41:38,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.505e+02 1.029e+03 1.227e+03 1.650e+03 7.387e+03, threshold=2.454e+03, percent-clipped=3.0
+2023-03-11 22:41:42,906 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1620, 1.1487, 3.6611, 3.1074], device='cuda:0'), covar=tensor([0.1759, 0.3016, 0.0460, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.0763, 0.0652, 0.0961, 0.0914], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 22:42:12,916 INFO [train.py:968] (0/2) Epoch 23, batch 17950, giga_loss[loss=0.2325, simple_loss=0.3076, pruned_loss=0.07869, over 28924.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3032, pruned_loss=0.07645, over 5697838.83 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.327, pruned_loss=0.08454, over 5771723.43 frames. ], giga_tot_loss[loss=0.2268, simple_loss=0.3014, pruned_loss=0.07603, over 5684287.51 frames. ], batch size: 106, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:42:59,494 INFO [train.py:968] (0/2) Epoch 23, batch 18000, giga_loss[loss=0.2002, simple_loss=0.278, pruned_loss=0.06122, over 28540.00 frames. ], tot_loss[loss=0.226, simple_loss=0.301, pruned_loss=0.07556, over 5698566.82 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3272, pruned_loss=0.08454, over 5771539.26 frames. ], giga_tot_loss[loss=0.2247, simple_loss=0.2991, pruned_loss=0.07511, over 5687269.14 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:42:59,499 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 22:43:08,692 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2051, simple_loss=0.3118, pruned_loss=0.04918, over 944034.00 frames. 
+2023-03-11 22:43:08,693 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 22:43:17,231 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.306e+02 1.131e+03 1.528e+03 2.158e+03 6.119e+03, threshold=3.055e+03, percent-clipped=17.0
+2023-03-11 22:43:29,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7067, 1.9863, 1.3978, 1.5746], device='cuda:0'), covar=tensor([0.1033, 0.0598, 0.1157, 0.1106], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0441, 0.0516, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 22:43:31,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5268, 1.7599, 1.7857, 1.3450], device='cuda:0'), covar=tensor([0.1813, 0.2496, 0.1470, 0.1703], device='cuda:0'), in_proj_covar=tensor([0.0911, 0.0700, 0.0959, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 22:43:41,277 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1021432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:43:44,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1021435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:43:50,334 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1021440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:43:50,708 INFO [train.py:968] (0/2) Epoch 23, batch 18050, giga_loss[loss=0.2139, simple_loss=0.2859, pruned_loss=0.07095, over 28570.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.2984, pruned_loss=0.07404, over 5699206.77 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3275, pruned_loss=0.08442, over 5773764.75 frames. ], giga_tot_loss[loss=0.2211, simple_loss=0.2955, pruned_loss=0.07337, over 5685120.12 frames. ], batch size: 78, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:43:52,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1021443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:44:12,393 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:44:19,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:44:38,250 INFO [train.py:968] (0/2) Epoch 23, batch 18100, giga_loss[loss=0.2317, simple_loss=0.2815, pruned_loss=0.09091, over 23634.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2952, pruned_loss=0.07285, over 5686060.81 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3275, pruned_loss=0.08433, over 5775643.09 frames. ], giga_tot_loss[loss=0.2185, simple_loss=0.2925, pruned_loss=0.07224, over 5672315.66 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:44:47,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.869e+02 1.007e+03 1.328e+03 1.810e+03 3.233e+03, threshold=2.657e+03, percent-clipped=1.0
+2023-03-11 22:45:24,497 INFO [train.py:968] (0/2) Epoch 23, batch 18150, giga_loss[loss=0.2116, simple_loss=0.286, pruned_loss=0.06857, over 28472.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2941, pruned_loss=0.07288, over 5690260.61 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3275, pruned_loss=0.08436, over 5777836.48 frames. ], giga_tot_loss[loss=0.218, simple_loss=0.2915, pruned_loss=0.07223, over 5676515.26 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:45:27,531 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1021543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:46:15,913 INFO [train.py:968] (0/2) Epoch 23, batch 18200, giga_loss[loss=0.348, simple_loss=0.4051, pruned_loss=0.1454, over 27951.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3051, pruned_loss=0.07859, over 5685388.68 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3275, pruned_loss=0.08423, over 5780420.33 frames. ], giga_tot_loss[loss=0.2291, simple_loss=0.3023, pruned_loss=0.07794, over 5669574.09 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:46:26,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.427e+02 1.261e+03 1.826e+03 2.236e+03 6.140e+03, threshold=3.653e+03, percent-clipped=16.0
+2023-03-11 22:47:01,956 INFO [train.py:968] (0/2) Epoch 23, batch 18250, giga_loss[loss=0.2789, simple_loss=0.3548, pruned_loss=0.1015, over 28912.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3184, pruned_loss=0.08514, over 5692850.60 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3281, pruned_loss=0.08453, over 5782624.55 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.3152, pruned_loss=0.08431, over 5676115.81 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:47:05,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3317, 1.3635, 1.1686, 1.4487], device='cuda:0'), covar=tensor([0.0784, 0.0355, 0.0361, 0.0900], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-11 22:47:12,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1021653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:47:28,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1021675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:47:34,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7758, 4.5694, 4.3740, 1.8752], device='cuda:0'), covar=tensor([0.0654, 0.0898, 0.0994, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.1218, 0.1126, 0.0955, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 22:47:42,768 INFO [train.py:968] (0/2) Epoch 23, batch 18300, giga_loss[loss=0.2565, simple_loss=0.3418, pruned_loss=0.08556, over 28827.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3297, pruned_loss=0.09065, over 5700922.16 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3289, pruned_loss=0.08491, over 5785705.50 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3265, pruned_loss=0.08975, over 5682786.19 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:47:53,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.388e+03 1.646e+03 2.039e+03 3.934e+03, threshold=3.292e+03, percent-clipped=4.0
+2023-03-11 22:47:54,383 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1021703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:48:09,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6082, 1.9282, 1.2769, 1.5297], device='cuda:0'), covar=tensor([0.1142, 0.0709, 0.1135, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0442, 0.0519, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-11 22:48:26,316 INFO [train.py:968] (0/2) Epoch 23, batch 18350, giga_loss[loss=0.2867, simple_loss=0.3705, pruned_loss=0.1015, over 27966.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3359, pruned_loss=0.0927, over 5700515.83 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3291, pruned_loss=0.08495, over 5788141.24 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3332, pruned_loss=0.09211, over 5681950.55 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:49:10,208 INFO [train.py:968] (0/2) Epoch 23, batch 18400, giga_loss[loss=0.2369, simple_loss=0.3208, pruned_loss=0.07651, over 28399.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3376, pruned_loss=0.09222, over 5690408.10 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3292, pruned_loss=0.08507, over 5779080.25 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3354, pruned_loss=0.0917, over 5682925.10 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:49:15,449 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1021796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:49:20,469 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1021799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:49:24,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.714e+02 1.258e+03 1.538e+03 1.916e+03 4.707e+03, threshold=3.075e+03, percent-clipped=2.0
+2023-03-11 22:49:38,863 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1021818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:49:41,191 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1021821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:49:49,311 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:49:58,953 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1021839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:50:00,023 INFO [train.py:968] (0/2) Epoch 23, batch 18450, libri_loss[loss=0.2395, simple_loss=0.325, pruned_loss=0.07701, over 29555.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3401, pruned_loss=0.09314, over 5664705.08 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3298, pruned_loss=0.08533, over 5769519.78 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.338, pruned_loss=0.09265, over 5665167.19 frames. ], batch size: 78, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:50:08,874 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:50:29,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6408, 2.2276, 1.7543, 0.8860], device='cuda:0'), covar=tensor([0.5225, 0.2833, 0.3478, 0.5948], device='cuda:0'), in_proj_covar=tensor([0.1751, 0.1656, 0.1595, 0.1428], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 22:50:43,715 INFO [train.py:968] (0/2) Epoch 23, batch 18500, giga_loss[loss=0.2503, simple_loss=0.3347, pruned_loss=0.08295, over 28990.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3426, pruned_loss=0.09529, over 5672534.98 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3302, pruned_loss=0.08567, over 5772702.05 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3409, pruned_loss=0.0948, over 5667794.41 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:50:54,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.269e+02 1.308e+03 1.630e+03 2.267e+03 5.419e+03, threshold=3.261e+03, percent-clipped=15.0
+2023-03-11 22:51:08,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1021918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:51:29,380 INFO [train.py:968] (0/2) Epoch 23, batch 18550, libri_loss[loss=0.2878, simple_loss=0.3704, pruned_loss=0.1026, over 29526.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.346, pruned_loss=0.09788, over 5676700.32 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3304, pruned_loss=0.08563, over 5776373.09 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3448, pruned_loss=0.09783, over 5666947.09 frames. ], batch size: 89, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:52:11,147 INFO [train.py:968] (0/2) Epoch 23, batch 18600, libri_loss[loss=0.2439, simple_loss=0.3358, pruned_loss=0.07597, over 29540.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3477, pruned_loss=0.09824, over 5681146.16 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3303, pruned_loss=0.08542, over 5776213.30 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3474, pruned_loss=0.09877, over 5670365.57 frames. ], batch size: 82, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:52:18,719 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1022000.pt
+2023-03-11 22:52:19,922 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1022001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:52:21,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.712e+02 1.234e+03 1.456e+03 1.911e+03 4.241e+03, threshold=2.912e+03, percent-clipped=3.0
+2023-03-11 22:52:52,943 INFO [train.py:968] (0/2) Epoch 23, batch 18650, giga_loss[loss=0.3215, simple_loss=0.3947, pruned_loss=0.1241, over 28893.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3501, pruned_loss=0.0987, over 5692770.00 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3304, pruned_loss=0.08545, over 5782834.03 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3506, pruned_loss=0.09972, over 5673792.75 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:53:07,726 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1022061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:53:12,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1022064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:53:25,091 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1022078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:53:33,503 INFO [train.py:968] (0/2) Epoch 23, batch 18700, giga_loss[loss=0.2873, simple_loss=0.3631, pruned_loss=0.1057, over 28444.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3529, pruned_loss=0.09969, over 5691081.68 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3312, pruned_loss=0.08575, over 5783596.92 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3531, pruned_loss=0.1005, over 5673191.03 frames. ], batch size: 85, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:53:35,741 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1022093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:53:44,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.217e+02 1.398e+03 1.692e+03 2.392e+03 1.034e+04, threshold=3.384e+03, percent-clipped=15.0
+2023-03-11 22:54:16,106 INFO [train.py:968] (0/2) Epoch 23, batch 18750, giga_loss[loss=0.2775, simple_loss=0.3621, pruned_loss=0.09646, over 28451.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3536, pruned_loss=0.09924, over 5700928.61 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3315, pruned_loss=0.08596, over 5784902.62 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3539, pruned_loss=0.09996, over 5684056.50 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:54:43,674 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-11 22:54:56,549 INFO [train.py:968] (0/2) Epoch 23, batch 18800, giga_loss[loss=0.2658, simple_loss=0.3519, pruned_loss=0.08983, over 28762.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3532, pruned_loss=0.09775, over 5708375.72 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.332, pruned_loss=0.08601, over 5788260.31 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3537, pruned_loss=0.09869, over 5688939.78 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:55:05,305 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.642e+02 1.184e+03 1.451e+03 1.900e+03 5.102e+03, threshold=2.903e+03, percent-clipped=9.0
+2023-03-11 22:55:15,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1022214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:55:21,720 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1022221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:55:23,917 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1022224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:55:37,875 INFO [train.py:968] (0/2) Epoch 23, batch 18850, giga_loss[loss=0.2225, simple_loss=0.3173, pruned_loss=0.06386, over 28685.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3515, pruned_loss=0.09568, over 5714977.52 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.332, pruned_loss=0.08591, over 5790239.64 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3525, pruned_loss=0.09685, over 5695284.04 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:55:48,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1022253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:56:19,195 INFO [train.py:968] (0/2) Epoch 23, batch 18900, giga_loss[loss=0.2745, simple_loss=0.3605, pruned_loss=0.0942, over 28593.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3498, pruned_loss=0.09432, over 5713369.75 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3321, pruned_loss=0.08583, over 5789786.55 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3507, pruned_loss=0.09546, over 5697029.45 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:56:28,421 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.553e+02 1.150e+03 1.390e+03 1.798e+03 4.391e+03, threshold=2.780e+03, percent-clipped=7.0
+2023-03-11 22:57:03,364 INFO [train.py:968] (0/2) Epoch 23, batch 18950, giga_loss[loss=0.2543, simple_loss=0.3313, pruned_loss=0.0886, over 28941.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3514, pruned_loss=0.0969, over 5719831.11 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3319, pruned_loss=0.08578, over 5792624.30 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3527, pruned_loss=0.09807, over 5702632.54 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:57:19,176 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1022357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:57:21,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1022360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:57:35,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1022376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:57:48,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1022389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:57:49,927 INFO [train.py:968] (0/2) Epoch 23, batch 19000, giga_loss[loss=0.3207, simple_loss=0.3801, pruned_loss=0.1306, over 28262.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3549, pruned_loss=0.1023, over 5720054.23 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3319, pruned_loss=0.08573, over 5794867.69 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3564, pruned_loss=0.1035, over 5702844.80 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:57:51,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-11 22:57:59,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.754e+02 1.415e+03 1.760e+03 2.299e+03 4.439e+03, threshold=3.520e+03, percent-clipped=9.0
+2023-03-11 22:58:18,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8854, 1.9565, 2.1177, 1.6698], device='cuda:0'), covar=tensor([0.1837, 0.2324, 0.1422, 0.1701], device='cuda:0'), in_proj_covar=tensor([0.0907, 0.0697, 0.0954, 0.0853], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 22:58:27,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4576, 1.7779, 1.4264, 1.5421], device='cuda:0'), covar=tensor([0.2555, 0.2538, 0.2812, 0.2238], device='cuda:0'), in_proj_covar=tensor([0.1523, 0.1099, 0.1346, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 22:58:29,206 INFO [train.py:968] (0/2) Epoch 23, batch 19050, giga_loss[loss=0.2647, simple_loss=0.3428, pruned_loss=0.09328, over 28783.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3543, pruned_loss=0.1033, over 5722415.91 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.332, pruned_loss=0.08587, over 5797541.90 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.356, pruned_loss=0.1046, over 5704843.55 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:59:13,677 INFO [train.py:968] (0/2) Epoch 23, batch 19100, giga_loss[loss=0.2741, simple_loss=0.3493, pruned_loss=0.09939, over 28926.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3522, pruned_loss=0.1027, over 5707390.45 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3326, pruned_loss=0.08615, over 5788191.52 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3534, pruned_loss=0.1038, over 5700897.98 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:59:17,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2599, 1.5014, 1.5104, 1.3141], device='cuda:0'), covar=tensor([0.1988, 0.1830, 0.2439, 0.1935], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0746, 0.0714, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 22:59:25,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.100e+02 1.257e+03 1.548e+03 1.970e+03 5.984e+03, threshold=3.096e+03, percent-clipped=2.0
+2023-03-11 22:59:30,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-11 22:59:38,780 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1022519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:59:41,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1022522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:59:44,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7430, 1.8164, 1.5391, 1.7573], device='cuda:0'), covar=tensor([0.2853, 0.2933, 0.3234, 0.2647], device='cuda:0'), in_proj_covar=tensor([0.1522, 0.1100, 0.1346, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 22:59:59,682 INFO [train.py:968] (0/2) Epoch 23, batch 19150, giga_loss[loss=0.255, simple_loss=0.328, pruned_loss=0.09096, over 28470.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3497, pruned_loss=0.1013, over 5706055.21 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3323, pruned_loss=0.08603, over 5788378.37 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.351, pruned_loss=0.1024, over 5700035.47 frames. ], batch size: 85, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:00:10,368 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1022551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:00:45,022 INFO [train.py:968] (0/2) Epoch 23, batch 19200, giga_loss[loss=0.2863, simple_loss=0.3695, pruned_loss=0.1016, over 28873.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3493, pruned_loss=0.1003, over 5714324.20 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3323, pruned_loss=0.08594, over 5790381.52 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3507, pruned_loss=0.1015, over 5706358.28 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:00:56,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.967e+02 1.317e+03 1.681e+03 2.094e+03 3.656e+03, threshold=3.363e+03, percent-clipped=4.0
+2023-03-11 23:01:26,589 INFO [train.py:968] (0/2) Epoch 23, batch 19250, libri_loss[loss=0.2484, simple_loss=0.34, pruned_loss=0.07839, over 29539.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3485, pruned_loss=0.09922, over 5699984.57 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.333, pruned_loss=0.08617, over 5782806.35 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3494, pruned_loss=0.1004, over 5697790.50 frames. ], batch size: 84, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:01:26,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8211, 4.6739, 4.4390, 2.0497], device='cuda:0'), covar=tensor([0.0673, 0.0741, 0.0880, 0.1881], device='cuda:0'), in_proj_covar=tensor([0.1221, 0.1129, 0.0957, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 23:01:41,839 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1022657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:01:55,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2792, 3.2391, 1.3862, 1.5522], device='cuda:0'), covar=tensor([0.1268, 0.0402, 0.1052, 0.1554], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0553, 0.0390, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 23:02:02,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7037, 1.6157, 1.9794, 1.5562], device='cuda:0'), covar=tensor([0.1414, 0.1995, 0.1177, 0.1501], device='cuda:0'), in_proj_covar=tensor([0.0907, 0.0698, 0.0954, 0.0853], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 23:02:12,289 INFO [train.py:968] (0/2) Epoch 23, batch 19300, giga_loss[loss=0.2322, simple_loss=0.3164, pruned_loss=0.07402, over 28926.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3428, pruned_loss=0.09587, over 5695133.01 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.333, pruned_loss=0.08608, over 5785289.41 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3438, pruned_loss=0.0971, over 5689845.41 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:02:24,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.472e+02 1.110e+03 1.309e+03 1.731e+03 3.650e+03, threshold=2.618e+03, percent-clipped=1.0
+2023-03-11 23:02:54,036 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 23:02:57,675 INFO [train.py:968] (0/2) Epoch 23, batch 19350, giga_loss[loss=0.2264, simple_loss=0.3021, pruned_loss=0.0753, over 28853.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3384, pruned_loss=0.09386, over 5684986.60 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3336, pruned_loss=0.08652, over 5783587.54 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3388, pruned_loss=0.09468, over 5679909.26 frames. ], batch size: 66, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:03:19,789 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1022764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:03:41,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3408, 1.3217, 1.1912, 1.2681], device='cuda:0'), covar=tensor([0.1959, 0.2043, 0.1866, 0.1920], device='cuda:0'), in_proj_covar=tensor([0.1970, 0.1903, 0.1823, 0.1974], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 23:03:46,295 INFO [train.py:968] (0/2) Epoch 23, batch 19400, giga_loss[loss=0.2529, simple_loss=0.3252, pruned_loss=0.09034, over 28385.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3329, pruned_loss=0.09136, over 5670063.32 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3341, pruned_loss=0.0867, over 5786104.67 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3328, pruned_loss=0.09196, over 5661985.67 frames. ], batch size: 369, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:04:00,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.833e+02 1.035e+03 1.291e+03 1.704e+03 4.086e+03, threshold=2.582e+03, percent-clipped=7.0
+2023-03-11 23:04:16,452 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2949, 1.1685, 4.0704, 3.3647], device='cuda:0'), covar=tensor([0.2153, 0.3270, 0.0735, 0.1008], device='cuda:0'), in_proj_covar=tensor([0.0763, 0.0652, 0.0961, 0.0917], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 23:04:33,103 INFO [train.py:968] (0/2) Epoch 23, batch 19450, giga_loss[loss=0.2675, simple_loss=0.3431, pruned_loss=0.0959, over 28881.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3309, pruned_loss=0.09022, over 5658643.66 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3339, pruned_loss=0.08651, over 5786862.94 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.331, pruned_loss=0.091, over 5647234.71 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:04:40,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0742, 1.6590, 1.2711, 0.3631], device='cuda:0'), covar=tensor([0.5319, 0.2872, 0.4296, 0.6009], device='cuda:0'), in_proj_covar=tensor([0.1739, 0.1639, 0.1577, 0.1415], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 23:04:47,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2914, 3.1236, 2.9623, 1.3637], device='cuda:0'), covar=tensor([0.0935, 0.1048, 0.0795, 0.2385], device='cuda:0'), in_proj_covar=tensor([0.1217, 0.1127, 0.0955, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 23:05:04,234 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 23:05:16,217 INFO [train.py:968] (0/2) Epoch 23, batch 19500, giga_loss[loss=0.2453, simple_loss=0.3159, pruned_loss=0.08738, over 28700.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3314, pruned_loss=0.09012, over 5668750.93 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3337, pruned_loss=0.08628, over 5788586.60 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3315, pruned_loss=0.09098, over 5656691.15 frames. ], batch size: 92, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:05:27,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.161e+02 1.107e+03 1.333e+03 1.617e+03 4.572e+03, threshold=2.666e+03, percent-clipped=6.0
+2023-03-11 23:05:56,876 INFO [train.py:968] (0/2) Epoch 23, batch 19550, giga_loss[loss=0.3377, simple_loss=0.3859, pruned_loss=0.1447, over 26664.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3307, pruned_loss=0.08926, over 5681238.41 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3336, pruned_loss=0.08601, over 5793425.38 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3308, pruned_loss=0.09038, over 5662372.04 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:06:40,008 INFO [train.py:968] (0/2) Epoch 23, batch 19600, libri_loss[loss=0.2702, simple_loss=0.3625, pruned_loss=0.08897, over 29125.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.33, pruned_loss=0.0889, over 5679313.29 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3346, pruned_loss=0.08641, over 5785740.15 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3291, pruned_loss=0.08946, over 5669274.16 frames. ], batch size: 101, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:06:50,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.211e+02 1.064e+03 1.376e+03 2.114e+03 1.220e+04, threshold=2.751e+03, percent-clipped=11.0
+2023-03-11 23:07:11,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1023032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:07:16,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5512, 1.6643, 1.7820, 1.3482], device='cuda:0'), covar=tensor([0.1876, 0.2652, 0.1491, 0.1736], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0699, 0.0952, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 23:07:18,374 INFO [train.py:968] (0/2) Epoch 23, batch 19650, giga_loss[loss=0.2362, simple_loss=0.3138, pruned_loss=0.07925, over 28596.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3286, pruned_loss=0.08838, over 5688461.15 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3351, pruned_loss=0.08658, over 5787016.49 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3273, pruned_loss=0.08869, over 5677920.84 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:07:30,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2116, 1.5504, 1.4989, 1.0752], device='cuda:0'), covar=tensor([0.1729, 0.2773, 0.1541, 0.1809], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0698, 0.0952, 0.0851], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 23:07:34,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9814, 1.2903, 1.0710, 0.2448], device='cuda:0'), covar=tensor([0.4118, 0.3156, 0.5188, 0.6628], device='cuda:0'), in_proj_covar=tensor([0.1742, 0.1645, 0.1582, 0.1422], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 23:07:44,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3500, 1.5861, 1.6355, 1.4517], device='cuda:0'), covar=tensor([0.2210, 0.2073, 0.2421, 0.2210], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0753, 0.0720, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 23:07:45,631 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-11 23:07:59,611 INFO [train.py:968] (0/2) Epoch 23, batch 19700, giga_loss[loss=0.2232, simple_loss=0.3029, pruned_loss=0.07171, over 29021.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3271, pruned_loss=0.08761, over 5701977.49 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3358, pruned_loss=0.08681, over 5790785.48 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3251, pruned_loss=0.08771, over 5686195.33 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:08:08,984 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-11 23:08:13,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.694e+02 1.142e+03 1.444e+03 2.179e+03 1.516e+04, threshold=2.887e+03, percent-clipped=12.0
+2023-03-11 23:08:41,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1023139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:08:42,523 INFO [train.py:968] (0/2) Epoch 23, batch 19750, giga_loss[loss=0.1957, simple_loss=0.2768, pruned_loss=0.05733, over 28518.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3242, pruned_loss=0.08631, over 5703611.47 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3357, pruned_loss=0.08667, over 5793070.82 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3225, pruned_loss=0.08653, over 5687459.25 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:09:10,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1023175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:09:12,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1023178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:09:21,309 INFO [train.py:968] (0/2) Epoch 23, batch 19800, giga_loss[loss=0.235, simple_loss=0.3166, pruned_loss=0.0767, over 28613.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3211, pruned_loss=0.08495, over 5716511.15 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3357, pruned_loss=0.08655, over 5796005.45 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3194, pruned_loss=0.08517, over 5699012.48 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:09:35,087 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1023207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:09:35,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.421e+02 1.024e+03 1.205e+03 1.639e+03 5.517e+03, threshold=2.410e+03, percent-clipped=4.0
+2023-03-11 23:09:42,055 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 23:10:03,365 INFO [train.py:968] (0/2) Epoch 23, batch 19850, giga_loss[loss=0.2331, simple_loss=0.3067, pruned_loss=0.07974, over 28832.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3197, pruned_loss=0.08433, over 5724408.87 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3361, pruned_loss=0.08649, over 5797543.96 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3176, pruned_loss=0.08453, over 5707326.02 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:10:36,556 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1023282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:10:38,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1023285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:10:45,162 INFO [train.py:968] (0/2) Epoch 23, batch 19900, giga_loss[loss=0.2349, simple_loss=0.3031, pruned_loss=0.08332, over 24004.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3173, pruned_loss=0.08305, over 5721976.78 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3362, pruned_loss=0.08631, over 5799549.16 frames. ], giga_tot_loss[loss=0.2409, simple_loss=0.3153, pruned_loss=0.08331, over 5705468.83 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:10:59,003 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.548e+02 1.056e+03 1.270e+03 1.694e+03 4.590e+03, threshold=2.539e+03, percent-clipped=9.0
+2023-03-11 23:11:04,205 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1023314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:11:23,759 INFO [train.py:968] (0/2) Epoch 23, batch 19950, giga_loss[loss=0.2376, simple_loss=0.3146, pruned_loss=0.0803, over 28719.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3161, pruned_loss=0.08213, over 5726144.07 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3366, pruned_loss=0.08628, over 5800365.95 frames. ], giga_tot_loss[loss=0.239, simple_loss=0.3136, pruned_loss=0.08225, over 5710559.56 frames. ], batch size: 66, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:11:37,493 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 23:11:52,041 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1023376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:12:03,512 INFO [train.py:968] (0/2) Epoch 23, batch 20000, giga_loss[loss=0.2589, simple_loss=0.3367, pruned_loss=0.0905, over 28375.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3155, pruned_loss=0.08142, over 5729206.99 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.337, pruned_loss=0.08644, over 5802213.90 frames. ], giga_tot_loss[loss=0.2376, simple_loss=0.3127, pruned_loss=0.0813, over 5713911.06 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:12:16,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.486e+02 1.120e+03 1.398e+03 1.909e+03 4.881e+03, threshold=2.795e+03, percent-clipped=10.0
+2023-03-11 23:12:36,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5385, 2.2494, 1.7245, 0.6724], device='cuda:0'), covar=tensor([0.6329, 0.2966, 0.4420, 0.7154], device='cuda:0'), in_proj_covar=tensor([0.1748, 0.1646, 0.1589, 0.1427], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 23:12:44,597 INFO [train.py:968] (0/2) Epoch 23, batch 20050, giga_loss[loss=0.2543, simple_loss=0.3283, pruned_loss=0.09014, over 29024.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3196, pruned_loss=0.08376, over 5725437.62 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3382, pruned_loss=0.08685, over 5802002.10 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3158, pruned_loss=0.08317, over 5711179.14 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:13:34,524 INFO [train.py:968] (0/2) Epoch 23, batch 20100, giga_loss[loss=0.313, simple_loss=0.3652, pruned_loss=0.1303, over 23565.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3256, pruned_loss=0.08789, over 5709632.39 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3381, pruned_loss=0.08668, over 5801690.78 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3224, pruned_loss=0.08758, over 5697493.97 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:13:49,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.711e+02 1.316e+03 1.622e+03 2.357e+03 5.186e+03, threshold=3.243e+03, percent-clipped=11.0
+2023-03-11 23:13:59,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1023518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:14:25,679 INFO [train.py:968] (0/2) Epoch 23, batch 20150, giga_loss[loss=0.2763, simple_loss=0.3461, pruned_loss=0.1032, over 28832.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3328, pruned_loss=0.09223, over 5709606.24 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3385, pruned_loss=0.08674, over 5804140.40 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3296, pruned_loss=0.09198, over 5695331.56 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:14:25,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8082, 1.9615, 1.8585, 1.6782], device='cuda:0'), covar=tensor([0.2394, 0.2124, 0.1763, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.1975, 0.1913, 0.1831, 0.1983], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 23:15:11,372 INFO [train.py:968] (0/2) Epoch 23, batch 20200, giga_loss[loss=0.3023, simple_loss=0.3705, pruned_loss=0.1171, over 28236.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.339, pruned_loss=0.09609, over 5702376.30 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3385, pruned_loss=0.08676, over 5797594.20 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3364, pruned_loss=0.09608, over 5693492.51 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:15:20,395 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1023601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:15:25,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.085e+02 1.332e+03 1.617e+03 2.096e+03 3.962e+03, threshold=3.233e+03, percent-clipped=4.0
+2023-03-11 23:15:55,688 INFO [train.py:968] (0/2) Epoch 23, batch 20250, giga_loss[loss=0.252, simple_loss=0.3421, pruned_loss=0.081, over 28941.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3441, pruned_loss=0.09815, over 5695269.36 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3386, pruned_loss=0.08673, over 5797055.38 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.342, pruned_loss=0.09856, over 5684877.72 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:16:42,864 INFO [train.py:968] (0/2) Epoch 23, batch 20300, giga_loss[loss=0.2864, simple_loss=0.3632, pruned_loss=0.1048, over 28358.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3487, pruned_loss=0.0999, over 5692819.51 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3387, pruned_loss=0.08673, over 5795132.21 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3471, pruned_loss=0.1004, over 5684837.35 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:17:00,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.207e+02 1.277e+03 1.817e+03 2.754e+03 8.840e+03, threshold=3.634e+03, percent-clipped=17.0
+2023-03-11 23:17:01,225 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1023710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:17:29,186 INFO [train.py:968] (0/2) Epoch 23, batch 20350, giga_loss[loss=0.2881, simple_loss=0.3598, pruned_loss=0.1081, over 28815.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3549, pruned_loss=0.1035, over 5701210.48 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.339, pruned_loss=0.08692, over 5796440.67 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3536, pruned_loss=0.1039, over 5692011.52 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:17:39,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1023751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:18:14,714 INFO [train.py:968] (0/2) Epoch 23, batch 20400, giga_loss[loss=0.2573, simple_loss=0.3426, pruned_loss=0.08599, over 28563.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3513, pruned_loss=0.1009, over 5694322.24 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.339, pruned_loss=0.08701, over 5798773.68 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3506, pruned_loss=0.1016, over 5682824.78 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:18:32,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.781e+02 1.212e+03 1.474e+03 2.287e+03 5.584e+03, threshold=2.949e+03, percent-clipped=10.0
+2023-03-11 23:18:39,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2153, 3.9239, 1.3445, 1.4995], device='cuda:0'), covar=tensor([0.1333, 0.0358, 0.1086, 0.1731], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0551, 0.0389, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 23:18:57,004 INFO [train.py:968] (0/2) Epoch 23, batch 20450, giga_loss[loss=0.2905, simple_loss=0.3688, pruned_loss=0.1061, over 28716.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3477, pruned_loss=0.09772, over 5705754.19 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.339, pruned_loss=0.08701, over 5800372.24 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3473, pruned_loss=0.09852, over 5692897.35 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:19:21,012 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1023867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:19:39,008 INFO [train.py:968] (0/2) Epoch 23, batch 20500, giga_loss[loss=0.259, simple_loss=0.3327, pruned_loss=0.09263, over 28806.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3469, pruned_loss=0.09691, over 5695054.06 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3393, pruned_loss=0.08737, over 5792601.29 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3467, pruned_loss=0.0977, over 5687467.09 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:19:42,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1023893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:19:42,824 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1023894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:19:46,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1023897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:19:54,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.945e+02 1.296e+03 1.538e+03 2.250e+03 7.233e+03, threshold=3.076e+03, percent-clipped=11.0
+2023-03-11 23:20:11,051 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1023926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:20:25,258 INFO [train.py:968] (0/2) Epoch 23, batch 20550, giga_loss[loss=0.2728, simple_loss=0.3374, pruned_loss=0.1041, over 23532.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3473, pruned_loss=0.09684, over 5683333.72 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3397, pruned_loss=0.08768, over 5781770.79 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3469, pruned_loss=0.09733, over 5685443.64 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:20:55,453 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1023976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:06,528 INFO [train.py:968] (0/2) Epoch 23, batch 20600, libri_loss[loss=0.2809, simple_loss=0.3624, pruned_loss=0.0997, over 29460.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3492, pruned_loss=0.09812, over 5693593.29 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3397, pruned_loss=0.08764, over 5783627.65 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3492, pruned_loss=0.09886, over 5690433.57 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:21:14,236 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1024000.pt
+2023-03-11 23:21:20,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.641e+02 1.383e+03 1.751e+03 2.825e+03 9.298e+03, threshold=3.503e+03, percent-clipped=22.0
+2023-03-11 23:21:29,813 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:36,497 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:41,517 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:45,692 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:47,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:49,422 INFO [train.py:968] (0/2) Epoch 23, batch 20650, giga_loss[loss=0.2458, simple_loss=0.3297, pruned_loss=0.08093, over 28597.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3505, pruned_loss=0.09929, over 5689526.82 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3391, pruned_loss=0.08737, over 5784486.31 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3511, pruned_loss=0.1003, over 5684667.35 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:22:11,228 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:22:30,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:22:34,606 INFO [train.py:968] (0/2) Epoch 23, batch 20700, giga_loss[loss=0.2925, simple_loss=0.3651, pruned_loss=0.11, over 29043.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3518, pruned_loss=0.1005, over 5706458.72 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3392, pruned_loss=0.08739, over 5788674.22 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3527, pruned_loss=0.1017, over 5696089.95 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:22:53,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.866e+02 1.271e+03 1.581e+03 2.161e+03 4.549e+03, threshold=3.163e+03, percent-clipped=4.0
+2023-03-11 23:23:04,047 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:23:05,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:23:22,694 INFO [train.py:968] (0/2) Epoch 23, batch 20750, giga_loss[loss=0.2502, simple_loss=0.3291, pruned_loss=0.08565, over 28525.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3525, pruned_loss=0.1012, over 5712030.37 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3391, pruned_loss=0.08724, over 5790365.00 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3535, pruned_loss=0.1025, over 5701463.26 frames. ], batch size: 65, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:23:33,017 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:24:03,295 INFO [train.py:968] (0/2) Epoch 23, batch 20800, giga_loss[loss=0.287, simple_loss=0.3653, pruned_loss=0.1044, over 28955.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3525, pruned_loss=0.1012, over 5701610.81 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3392, pruned_loss=0.08714, over 5782579.18 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3537, pruned_loss=0.1029, over 5697852.15 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:24:17,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.419e+02 1.228e+03 1.543e+03 1.945e+03 3.555e+03, threshold=3.085e+03, percent-clipped=5.0
+2023-03-11 23:24:32,506 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:24:35,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:24:42,206 INFO [train.py:968] (0/2) Epoch 23, batch 20850, giga_loss[loss=0.279, simple_loss=0.3571, pruned_loss=0.1004, over 28904.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3522, pruned_loss=0.1004, over 5708800.33 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3388, pruned_loss=0.08694, over 5782115.47 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3536, pruned_loss=0.1021, over 5705098.00 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:24:43,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:25:00,384 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:25:07,260 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:25:24,502 INFO [train.py:968] (0/2) Epoch 23, batch 20900, giga_loss[loss=0.2788, simple_loss=0.3631, pruned_loss=0.09723, over 28639.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3527, pruned_loss=0.1001, over 5712539.76 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3392, pruned_loss=0.0873, over 5786657.30 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.354, pruned_loss=0.1015, over 5702842.04 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:25:40,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.012e+02 1.252e+03 1.486e+03 1.983e+03 4.963e+03, threshold=2.972e+03, percent-clipped=5.0
+2023-03-11 23:25:46,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7222, 1.9855, 2.0162, 1.5818], device='cuda:0'), covar=tensor([0.3338, 0.2690, 0.2553, 0.3133], device='cuda:0'), in_proj_covar=tensor([0.1978, 0.1913, 0.1838, 0.1986], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-11 23:26:06,587 INFO [train.py:968] (0/2) Epoch 23, batch 20950, giga_loss[loss=0.3436, simple_loss=0.39, pruned_loss=0.1487, over 26585.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3537, pruned_loss=0.1003, over 5700112.12 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3395, pruned_loss=0.08757, over 5767984.81 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3548, pruned_loss=0.1015, over 5707003.90 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:26:41,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:26:42,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:26:44,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:26:46,761 INFO [train.py:968] (0/2) Epoch 23, batch 21000, giga_loss[loss=0.2444, simple_loss=0.3275, pruned_loss=0.08072, over 28877.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3519, pruned_loss=0.0992, over 5707238.47 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3397, pruned_loss=0.08766, over 5770014.04 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3529, pruned_loss=0.1004, over 5709647.12 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:26:46,766 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-11 23:26:56,891 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2072, simple_loss=0.314, pruned_loss=0.05021, over 944034.00 frames. 
+2023-03-11 23:26:56,892 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-11 23:26:59,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:27:05,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:27:05,752 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:27:09,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:27:11,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.417e+02 1.197e+03 1.531e+03 2.081e+03 6.435e+03, threshold=3.063e+03, percent-clipped=10.0
+2023-03-11 23:27:16,531 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:27:35,814 INFO [train.py:968] (0/2) Epoch 23, batch 21050, giga_loss[loss=0.3206, simple_loss=0.3796, pruned_loss=0.1309, over 28637.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3498, pruned_loss=0.09859, over 5711444.35 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3394, pruned_loss=0.08754, over 5773330.92 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3512, pruned_loss=0.09993, over 5708963.31 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:28:09,363 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:28:15,364 INFO [train.py:968] (0/2) Epoch 23, batch 21100, giga_loss[loss=0.2363, simple_loss=0.3213, pruned_loss=0.07562, over 28420.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3483, pruned_loss=0.09764, over 5711799.17 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3393, pruned_loss=0.08742, over 5775549.17 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3496, pruned_loss=0.09894, over 5707187.92 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:28:30,482 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.620e+02 1.219e+03 1.542e+03 1.992e+03 3.968e+03, threshold=3.084e+03, percent-clipped=4.0
+2023-03-11 23:28:55,862 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:28:57,740 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:28:58,212 INFO [train.py:968] (0/2) Epoch 23, batch 21150, giga_loss[loss=0.2825, simple_loss=0.3549, pruned_loss=0.105, over 28588.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3479, pruned_loss=0.09778, over 5709082.79 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3395, pruned_loss=0.08742, over 5770094.98 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.349, pruned_loss=0.09905, over 5708639.55 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:28:59,676 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:01,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:04,808 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:05,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1453, 3.4612, 2.3797, 1.0655], device='cuda:0'), covar=tensor([0.8433, 0.2229, 0.3541, 0.7361], device='cuda:0'), in_proj_covar=tensor([0.1750, 0.1641, 0.1586, 0.1425], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 23:29:06,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:20,221 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:24,280 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:28,693 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:39,422 INFO [train.py:968] (0/2) Epoch 23, batch 21200, giga_loss[loss=0.262, simple_loss=0.3406, pruned_loss=0.09168, over 28863.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3478, pruned_loss=0.0979, over 5698087.46 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3397, pruned_loss=0.0877, over 5763292.60 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3489, pruned_loss=0.09909, over 5701659.42 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:29:52,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.229e+02 1.116e+03 1.319e+03 1.769e+03 3.324e+03, threshold=2.638e+03, percent-clipped=1.0
+2023-03-11 23:30:18,875 INFO [train.py:968] (0/2) Epoch 23, batch 21250, giga_loss[loss=0.3091, simple_loss=0.3698, pruned_loss=0.1242, over 26553.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3478, pruned_loss=0.09763, over 5698304.14 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3395, pruned_loss=0.08773, over 5753345.32 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3489, pruned_loss=0.09877, over 5709630.78 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:30:20,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:30:57,810 INFO [train.py:968] (0/2) Epoch 23, batch 21300, giga_loss[loss=0.2669, simple_loss=0.3481, pruned_loss=0.09283, over 28865.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3469, pruned_loss=0.09636, over 5702187.56 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3399, pruned_loss=0.08802, over 5756835.16 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3477, pruned_loss=0.09722, over 5706817.63 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:31:13,523 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.694e+02 1.090e+03 1.363e+03 1.899e+03 3.998e+03, threshold=2.727e+03, percent-clipped=11.0
+2023-03-11 23:31:38,805 INFO [train.py:968] (0/2) Epoch 23, batch 21350, giga_loss[loss=0.2955, simple_loss=0.3622, pruned_loss=0.1145, over 28814.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3467, pruned_loss=0.09697, over 5701065.50 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3401, pruned_loss=0.08822, over 5757391.48 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3473, pruned_loss=0.09773, over 5702672.31 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:31:52,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:32:09,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:32:18,463 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:32:20,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:32:21,466 INFO [train.py:968] (0/2) Epoch 23, batch 21400, giga_loss[loss=0.3036, simple_loss=0.3699, pruned_loss=0.1187, over 28271.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3453, pruned_loss=0.09676, over 5696030.20 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3401, pruned_loss=0.08822, over 5757391.48 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3457, pruned_loss=0.09736, over 5697280.81 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:32:34,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.602e+02 1.032e+03 1.317e+03 1.632e+03 5.591e+03, threshold=2.634e+03, percent-clipped=5.0
+2023-03-11 23:32:38,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5704, 4.4059, 4.2147, 1.9214], device='cuda:0'), covar=tensor([0.0574, 0.0780, 0.0745, 0.2135], device='cuda:0'), in_proj_covar=tensor([0.1225, 0.1135, 0.0960, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 23:32:40,827 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-11 23:32:41,974 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:33:00,761 INFO [train.py:968] (0/2) Epoch 23, batch 21450, giga_loss[loss=0.2303, simple_loss=0.3147, pruned_loss=0.07289, over 28883.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3428, pruned_loss=0.09575, over 5704035.65 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3401, pruned_loss=0.08831, over 5761011.81 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3433, pruned_loss=0.09635, over 5700339.39 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:33:13,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:33:18,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3450, 1.0911, 3.9532, 3.2584], device='cuda:0'), covar=tensor([0.1648, 0.2897, 0.0428, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0760, 0.0647, 0.0956, 0.0914], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 23:33:40,744 INFO [train.py:968] (0/2) Epoch 23, batch 21500, giga_loss[loss=0.2644, simple_loss=0.3333, pruned_loss=0.09774, over 28392.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3402, pruned_loss=0.09459, over 5693086.64 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3401, pruned_loss=0.08846, over 5753952.13 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3406, pruned_loss=0.09508, over 5695669.22 frames. ], batch size: 78, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:33:50,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:33:52,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:33:56,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.247e+02 1.221e+03 1.511e+03 1.895e+03 5.507e+03, threshold=3.022e+03, percent-clipped=12.0
+2023-03-11 23:34:04,049 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:34:05,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:34:15,435 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:34:20,780 INFO [train.py:968] (0/2) Epoch 23, batch 21550, giga_loss[loss=0.2316, simple_loss=0.316, pruned_loss=0.07361, over 28932.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3414, pruned_loss=0.09579, over 5691194.80 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3406, pruned_loss=0.08893, over 5756513.55 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3413, pruned_loss=0.09584, over 5689803.36 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:34:30,268 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:34:54,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6887, 1.8196, 1.5863, 1.5895], device='cuda:0'), covar=tensor([0.2444, 0.2402, 0.2448, 0.2358], device='cuda:0'), in_proj_covar=tensor([0.1526, 0.1103, 0.1349, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 23:35:05,591 INFO [train.py:968] (0/2) Epoch 23, batch 21600, giga_loss[loss=0.2506, simple_loss=0.3331, pruned_loss=0.08404, over 29082.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3383, pruned_loss=0.09417, over 5696227.69 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3407, pruned_loss=0.08906, over 5757544.44 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3381, pruned_loss=0.0942, over 5693119.30 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:35:14,591 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1025001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:35:16,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1025004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:35:23,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.722e+02 1.242e+03 1.517e+03 1.974e+03 5.219e+03, threshold=3.035e+03, percent-clipped=3.0
+2023-03-11 23:35:23,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:35:42,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1025033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:35:49,139 INFO [train.py:968] (0/2) Epoch 23, batch 21650, giga_loss[loss=0.254, simple_loss=0.3238, pruned_loss=0.09208, over 28807.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3365, pruned_loss=0.09371, over 5699194.79 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3411, pruned_loss=0.08931, over 5758627.22 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.336, pruned_loss=0.09355, over 5695054.51 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:36:07,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3892, 1.2024, 3.6501, 3.1235], device='cuda:0'), covar=tensor([0.1508, 0.2823, 0.0445, 0.0986], device='cuda:0'), in_proj_covar=tensor([0.0759, 0.0648, 0.0955, 0.0913], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-11 23:36:10,127 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:36:29,571 INFO [train.py:968] (0/2) Epoch 23, batch 21700, giga_loss[loss=0.2401, simple_loss=0.3175, pruned_loss=0.08131, over 29049.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.335, pruned_loss=0.09319, over 5707053.97 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3411, pruned_loss=0.08936, over 5761311.61 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3343, pruned_loss=0.09313, over 5699898.23 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:36:41,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5762, 1.9049, 1.4863, 1.8057], device='cuda:0'), covar=tensor([0.2708, 0.2803, 0.3240, 0.2577], device='cuda:0'), in_proj_covar=tensor([0.1526, 0.1103, 0.1347, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 23:36:44,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.696e+02 1.180e+03 1.447e+03 1.963e+03 4.752e+03, threshold=2.894e+03, percent-clipped=12.0
+2023-03-11 23:37:09,613 INFO [train.py:968] (0/2) Epoch 23, batch 21750, giga_loss[loss=0.2421, simple_loss=0.3162, pruned_loss=0.08398, over 28892.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3328, pruned_loss=0.09173, over 5717078.51 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3412, pruned_loss=0.08933, over 5763728.20 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.332, pruned_loss=0.09178, over 5707822.36 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:37:52,061 INFO [train.py:968] (0/2) Epoch 23, batch 21800, giga_loss[loss=0.3243, simple_loss=0.3793, pruned_loss=0.1347, over 23821.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3338, pruned_loss=0.09255, over 5698068.07 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3417, pruned_loss=0.08975, over 5747954.03 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3325, pruned_loss=0.09227, over 5702809.60 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:38:08,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.523e+02 1.150e+03 1.576e+03 2.141e+03 6.346e+03, threshold=3.153e+03, percent-clipped=12.0
+2023-03-11 23:38:13,897 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.37 vs. limit=5.0
+2023-03-11 23:38:34,730 INFO [train.py:968] (0/2) Epoch 23, batch 21850, giga_loss[loss=0.2463, simple_loss=0.3204, pruned_loss=0.08611, over 28657.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3366, pruned_loss=0.09364, over 5700291.08 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3421, pruned_loss=0.09008, over 5747579.32 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3351, pruned_loss=0.09318, over 5703815.53 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:38:57,439 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:39:02,328 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-11 23:39:11,223 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-11 23:39:21,420 INFO [train.py:968] (0/2) Epoch 23, batch 21900, giga_loss[loss=0.3277, simple_loss=0.3986, pruned_loss=0.1284, over 28549.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3405, pruned_loss=0.09527, over 5686902.75 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3418, pruned_loss=0.08999, over 5748311.21 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3396, pruned_loss=0.09505, over 5688434.03 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:39:39,429 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.129e+02 1.076e+03 1.331e+03 1.674e+03 4.100e+03, threshold=2.661e+03, percent-clipped=4.0
+2023-03-11 23:40:04,873 INFO [train.py:968] (0/2) Epoch 23, batch 21950, giga_loss[loss=0.2466, simple_loss=0.3333, pruned_loss=0.07992, over 28889.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3413, pruned_loss=0.09439, over 5696428.47 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.342, pruned_loss=0.09012, over 5746811.35 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3404, pruned_loss=0.09415, over 5698549.94 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:40:11,444 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025349.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 23:40:42,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1025386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:40:46,821 INFO [train.py:968] (0/2) Epoch 23, batch 22000, giga_loss[loss=0.3389, simple_loss=0.3878, pruned_loss=0.145, over 26695.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3412, pruned_loss=0.09362, over 5702650.39 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3427, pruned_loss=0.09064, over 5751215.84 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3399, pruned_loss=0.09309, over 5699542.04 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:41:08,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.020e+02 1.046e+03 1.295e+03 1.589e+03 1.104e+04, threshold=2.589e+03, percent-clipped=12.0
+2023-03-11 23:41:33,674 INFO [train.py:968] (0/2) Epoch 23, batch 22050, giga_loss[loss=0.2534, simple_loss=0.3332, pruned_loss=0.0868, over 28955.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3399, pruned_loss=0.09286, over 5702090.80 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3428, pruned_loss=0.09088, over 5753633.14 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3388, pruned_loss=0.09225, over 5696680.99 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:41:36,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1025444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:42:14,202 INFO [train.py:968] (0/2) Epoch 23, batch 22100, giga_loss[loss=0.2926, simple_loss=0.3615, pruned_loss=0.1118, over 28933.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3413, pruned_loss=0.0944, over 5696641.76 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3428, pruned_loss=0.09103, over 5747748.42 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3404, pruned_loss=0.09387, over 5695477.49 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:42:32,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.521e+02 1.321e+03 1.615e+03 2.000e+03 5.734e+03, threshold=3.229e+03, percent-clipped=13.0
+2023-03-11 23:42:46,845 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1025529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:42:48,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1025532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:42:54,590 INFO [train.py:968] (0/2) Epoch 23, batch 22150, giga_loss[loss=0.2466, simple_loss=0.3266, pruned_loss=0.08329, over 28846.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.342, pruned_loss=0.09493, over 5706352.62 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3429, pruned_loss=0.09132, over 5753556.19 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.341, pruned_loss=0.09439, over 5697791.45 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:43:12,296 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1025561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:43:32,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1025587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:43:35,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1025590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:43:35,351 INFO [train.py:968] (0/2) Epoch 23, batch 22200, giga_loss[loss=0.2958, simple_loss=0.3748, pruned_loss=0.1084, over 28712.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3454, pruned_loss=0.09732, over 5707454.97 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3433, pruned_loss=0.09195, over 5753030.06 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3443, pruned_loss=0.09645, over 5699483.33 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:43:50,447 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3044, 2.5808, 1.2828, 1.3879], device='cuda:0'), covar=tensor([0.0979, 0.0361, 0.0929, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0553, 0.0390, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 23:43:54,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.937e+02 1.380e+03 1.702e+03 2.236e+03 6.450e+03, threshold=3.404e+03, percent-clipped=7.0
+2023-03-11 23:44:00,860 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1025619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:44:15,677 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1025639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:44:17,473 INFO [train.py:968] (0/2) Epoch 23, batch 22250, giga_loss[loss=0.293, simple_loss=0.366, pruned_loss=0.11, over 27563.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3499, pruned_loss=0.0998, over 5714372.23 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.344, pruned_loss=0.09248, over 5754627.67 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3485, pruned_loss=0.09873, over 5706055.54 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:45:01,834 INFO [train.py:968] (0/2) Epoch 23, batch 22300, giga_loss[loss=0.3151, simple_loss=0.3753, pruned_loss=0.1274, over 26658.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3517, pruned_loss=0.1009, over 5705524.53 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3441, pruned_loss=0.0926, over 5745185.32 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3506, pruned_loss=0.1, over 5706617.08 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:45:18,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.863e+02 1.342e+03 1.662e+03 2.292e+03 6.753e+03, threshold=3.324e+03, percent-clipped=8.0
+2023-03-11 23:45:27,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1025724.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 23:45:41,208 INFO [train.py:968] (0/2) Epoch 23, batch 22350, libri_loss[loss=0.2823, simple_loss=0.3658, pruned_loss=0.0994, over 29670.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3533, pruned_loss=0.1016, over 5719807.65 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3455, pruned_loss=0.09368, over 5751141.34 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3514, pruned_loss=0.1002, over 5713818.29 frames. ], batch size: 88, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:45:49,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4211, 1.8694, 1.5530, 1.5140], device='cuda:0'), covar=tensor([0.0751, 0.0282, 0.0322, 0.0878], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 23:46:15,770 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1025782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:46:19,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1025785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:46:23,453 INFO [train.py:968] (0/2) Epoch 23, batch 22400, giga_loss[loss=0.2429, simple_loss=0.3293, pruned_loss=0.07823, over 28893.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3536, pruned_loss=0.1017, over 5719073.16 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3462, pruned_loss=0.09413, over 5754109.51 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3517, pruned_loss=0.1005, over 5711271.51 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:46:25,031 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6003, 1.8465, 1.5312, 1.7776], device='cuda:0'), covar=tensor([0.2185, 0.2257, 0.2253, 0.2217], device='cuda:0'), in_proj_covar=tensor([0.1521, 0.1100, 0.1343, 0.0992], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 23:46:43,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.130e+02 1.362e+03 1.710e+03 2.199e+03 6.107e+03, threshold=3.420e+03, percent-clipped=3.0
+2023-03-11 23:46:45,284 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1025814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:47:00,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4950, 1.5357, 1.7075, 1.3235], device='cuda:0'), covar=tensor([0.1617, 0.2359, 0.1365, 0.1608], device='cuda:0'), in_proj_covar=tensor([0.0910, 0.0704, 0.0957, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 23:47:08,141 INFO [train.py:968] (0/2) Epoch 23, batch 22450, giga_loss[loss=0.2657, simple_loss=0.3441, pruned_loss=0.09359, over 28732.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3534, pruned_loss=0.1021, over 5712493.92 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3472, pruned_loss=0.09497, over 5746065.61 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3512, pruned_loss=0.1005, over 5712445.40 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:47:13,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:47:26,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:47:29,500 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1025867.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 23:47:31,349 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:47:32,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1025870.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 23:47:48,594 INFO [train.py:968] (0/2) Epoch 23, batch 22500, giga_loss[loss=0.2618, simple_loss=0.3402, pruned_loss=0.09165, over 28682.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.352, pruned_loss=0.1016, over 5718169.33 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.349, pruned_loss=0.09661, over 5748781.86 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3489, pruned_loss=0.09917, over 5714469.85 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:47:56,133 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1025899.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 23:48:09,701 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.376e+02 1.271e+03 1.540e+03 2.302e+03 8.173e+03, threshold=3.080e+03, percent-clipped=8.0
+2023-03-11 23:48:22,792 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.11 vs. limit=5.0
+2023-03-11 23:48:30,917 INFO [train.py:968] (0/2) Epoch 23, batch 22550, giga_loss[loss=0.2571, simple_loss=0.3237, pruned_loss=0.09519, over 28537.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3478, pruned_loss=0.09947, over 5715251.11 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3485, pruned_loss=0.09652, over 5742760.23 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3457, pruned_loss=0.09768, over 5717411.02 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:49:11,813 INFO [train.py:968] (0/2) Epoch 23, batch 22600, giga_loss[loss=0.2597, simple_loss=0.3418, pruned_loss=0.08882, over 27920.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3447, pruned_loss=0.09772, over 5715759.39 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3487, pruned_loss=0.09674, over 5746380.47 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3429, pruned_loss=0.09616, over 5713791.20 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:49:12,254 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 23:49:17,951 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1026000.pt
+2023-03-11 23:49:31,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.253e+02 1.172e+03 1.360e+03 1.717e+03 3.671e+03, threshold=2.720e+03, percent-clipped=2.0
+2023-03-11 23:49:46,909 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.43 vs. limit=5.0
+2023-03-11 23:49:56,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1026040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:49:57,536 INFO [train.py:968] (0/2) Epoch 23, batch 22650, giga_loss[loss=0.224, simple_loss=0.3031, pruned_loss=0.07244, over 28383.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.345, pruned_loss=0.09677, over 5713809.97 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3484, pruned_loss=0.0966, over 5747316.57 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3439, pruned_loss=0.09568, over 5711246.99 frames. ], batch size: 78, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:50:40,440 INFO [train.py:968] (0/2) Epoch 23, batch 22700, giga_loss[loss=0.2903, simple_loss=0.3656, pruned_loss=0.1076, over 28810.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3461, pruned_loss=0.09641, over 5714543.06 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3486, pruned_loss=0.09678, over 5749850.70 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3449, pruned_loss=0.09539, over 5709944.38 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:50:58,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.979e+02 1.340e+03 1.649e+03 2.150e+03 5.992e+03, threshold=3.299e+03, percent-clipped=10.0
+2023-03-11 23:51:19,647 INFO [train.py:968] (0/2) Epoch 23, batch 22750, giga_loss[loss=0.2361, simple_loss=0.3063, pruned_loss=0.083, over 28542.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3464, pruned_loss=0.09711, over 5722417.99 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3491, pruned_loss=0.09732, over 5748048.46 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.345, pruned_loss=0.09581, over 5719445.65 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:52:02,263 INFO [train.py:968] (0/2) Epoch 23, batch 22800, libri_loss[loss=0.2705, simple_loss=0.3487, pruned_loss=0.09611, over 29467.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3459, pruned_loss=0.09774, over 5720548.78 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3495, pruned_loss=0.0977, over 5747921.46 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3443, pruned_loss=0.09635, over 5717596.04 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:52:20,840 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.662e+02 1.240e+03 1.526e+03 1.975e+03 4.265e+03, threshold=3.052e+03, percent-clipped=3.0
+2023-03-11 23:52:28,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1026224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:52:29,668 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-11 23:52:39,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1026237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:52:42,661 INFO [train.py:968] (0/2) Epoch 23, batch 22850, giga_loss[loss=0.2435, simple_loss=0.3173, pruned_loss=0.08485, over 28971.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3434, pruned_loss=0.09762, over 5721997.64 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3497, pruned_loss=0.09778, over 5750932.02 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.342, pruned_loss=0.09645, over 5716540.88 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:52:45,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1026243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:52:55,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5229, 4.4335, 1.5928, 1.7712], device='cuda:0'), covar=tensor([0.0967, 0.0395, 0.0942, 0.1261], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0553, 0.0389, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 23:53:23,866 INFO [train.py:968] (0/2) Epoch 23, batch 22900, giga_loss[loss=0.3081, simple_loss=0.3652, pruned_loss=0.1255, over 26762.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3422, pruned_loss=0.09837, over 5718800.74 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3496, pruned_loss=0.09813, over 5748533.21 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3409, pruned_loss=0.09716, over 5715504.26 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:53:45,815 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.193e+02 1.218e+03 1.565e+03 2.022e+03 6.222e+03, threshold=3.130e+03, percent-clipped=4.0
+2023-03-11 23:54:06,619 INFO [train.py:968] (0/2) Epoch 23, batch 22950, giga_loss[loss=0.3161, simple_loss=0.3688, pruned_loss=0.1317, over 24126.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3421, pruned_loss=0.09857, over 5721746.75 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3508, pruned_loss=0.09912, over 5751689.37 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3397, pruned_loss=0.09667, over 5715500.33 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:54:28,395 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1026367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:30,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1026370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:39,125 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1026380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:41,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1026383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:43,022 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1026386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:46,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1026389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:47,510 INFO [train.py:968] (0/2) Epoch 23, batch 23000, giga_loss[loss=0.2395, simple_loss=0.32, pruned_loss=0.07951, over 28620.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.339, pruned_loss=0.09702, over 5720485.58 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3509, pruned_loss=0.09919, over 5753776.30 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3369, pruned_loss=0.09543, over 5713137.63 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:54:50,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6328, 4.6739, 1.6893, 1.8186], device='cuda:0'), covar=tensor([0.0963, 0.0321, 0.0952, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0554, 0.0389, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-11 23:54:53,036 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:55:03,762 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:55:05,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.452e+02 1.237e+03 1.498e+03 2.016e+03 5.998e+03, threshold=2.995e+03, percent-clipped=8.0
+2023-03-11 23:55:06,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1026415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:55:08,212 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:55:22,352 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1026439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:55:24,576 INFO [train.py:968] (0/2) Epoch 23, batch 23050, giga_loss[loss=0.2275, simple_loss=0.3074, pruned_loss=0.0738, over 28644.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3365, pruned_loss=0.09585, over 5698256.04 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3518, pruned_loss=0.09999, over 5721833.77 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3335, pruned_loss=0.09378, over 5718119.44 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:55:47,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9020, 1.2834, 1.3373, 1.0594], device='cuda:0'), covar=tensor([0.1954, 0.1316, 0.2249, 0.1777], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0755, 0.0722, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-11 23:56:03,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3054, 4.1756, 3.9178, 1.8966], device='cuda:0'), covar=tensor([0.0609, 0.0703, 0.0656, 0.2093], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.1138, 0.0964, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 23:56:07,255 INFO [train.py:968] (0/2) Epoch 23, batch 23100, giga_loss[loss=0.233, simple_loss=0.3092, pruned_loss=0.07841, over 28900.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3327, pruned_loss=0.09405, over 5706828.71 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3524, pruned_loss=0.1006, over 5727214.98 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3294, pruned_loss=0.09169, over 5717334.93 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:56:27,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.898e+02 1.261e+03 1.639e+03 2.511e+03 5.852e+03, threshold=3.277e+03, percent-clipped=12.0
+2023-03-11 23:56:32,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4988, 1.5677, 1.4220, 1.6962], device='cuda:0'), covar=tensor([0.0751, 0.0327, 0.0333, 0.0870], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-11 23:56:48,160 INFO [train.py:968] (0/2) Epoch 23, batch 23150, giga_loss[loss=0.286, simple_loss=0.3535, pruned_loss=0.1093, over 27605.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3324, pruned_loss=0.09346, over 5706374.73 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3525, pruned_loss=0.1007, over 5729504.14 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3293, pruned_loss=0.09133, over 5712407.81 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:57:04,476 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1026558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:57:06,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1026561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:57:24,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9038, 1.3049, 1.1030, 0.2102], device='cuda:0'), covar=tensor([0.3449, 0.2526, 0.3621, 0.5037], device='cuda:0'), in_proj_covar=tensor([0.1763, 0.1657, 0.1602, 0.1436], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-11 23:57:30,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5899, 1.8027, 1.4343, 1.7683], device='cuda:0'), covar=tensor([0.2627, 0.2698, 0.3162, 0.2395], device='cuda:0'), in_proj_covar=tensor([0.1522, 0.1101, 0.1344, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 23:57:31,352 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:57:32,777 INFO [train.py:968] (0/2) Epoch 23, batch 23200, giga_loss[loss=0.3372, simple_loss=0.3963, pruned_loss=0.1391, over 27527.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3364, pruned_loss=0.09547, over 5709821.28 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3527, pruned_loss=0.1008, over 5732823.28 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3334, pruned_loss=0.0936, over 5711302.33 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:57:46,963 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3654, 1.5510, 1.6005, 1.2560], device='cuda:0'), covar=tensor([0.1324, 0.1876, 0.1146, 0.1428], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0700, 0.0952, 0.0851], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-11 23:57:54,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.074e+02 1.250e+03 1.477e+03 2.113e+03 5.579e+03, threshold=2.953e+03, percent-clipped=8.0
+2023-03-11 23:58:14,586 INFO [train.py:968] (0/2) Epoch 23, batch 23250, giga_loss[loss=0.2297, simple_loss=0.3112, pruned_loss=0.07412, over 28492.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3405, pruned_loss=0.09766, over 5712067.03 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.353, pruned_loss=0.1013, over 5734615.50 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3375, pruned_loss=0.0956, over 5711044.28 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:58:32,702 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4753, 1.7713, 1.3921, 1.8555], device='cuda:0'), covar=tensor([0.2538, 0.2669, 0.3003, 0.2520], device='cuda:0'), in_proj_covar=tensor([0.1524, 0.1102, 0.1345, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-11 23:58:55,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8586, 4.7055, 4.5010, 2.2966], device='cuda:0'), covar=tensor([0.0505, 0.0629, 0.0670, 0.1755], device='cuda:0'), in_proj_covar=tensor([0.1232, 0.1136, 0.0965, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 23:58:56,160 INFO [train.py:968] (0/2) Epoch 23, batch 23300, giga_loss[loss=0.2671, simple_loss=0.3367, pruned_loss=0.09872, over 28398.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3436, pruned_loss=0.09876, over 5706755.62 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3532, pruned_loss=0.1015, over 5731211.36 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3406, pruned_loss=0.09678, over 5708716.17 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:59:17,426 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.716e+02 1.207e+03 1.543e+03 2.223e+03 5.495e+03, threshold=3.086e+03, percent-clipped=16.0
+2023-03-11 23:59:29,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3408, 3.1701, 3.0549, 1.3601], device='cuda:0'), covar=tensor([0.0945, 0.1085, 0.0891, 0.2247], device='cuda:0'), in_proj_covar=tensor([0.1237, 0.1141, 0.0969, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-11 23:59:40,442 INFO [train.py:968] (0/2) Epoch 23, batch 23350, giga_loss[loss=0.2724, simple_loss=0.3521, pruned_loss=0.09636, over 28717.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.345, pruned_loss=0.09883, over 5716356.26 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3531, pruned_loss=0.1017, over 5733887.42 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3426, pruned_loss=0.09703, over 5715115.11 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:00:04,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4546, 1.3609, 3.8097, 3.2506], device='cuda:0'), covar=tensor([0.1478, 0.2730, 0.0431, 0.1037], device='cuda:0'), in_proj_covar=tensor([0.0762, 0.0651, 0.0960, 0.0917], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 00:00:25,453 INFO [train.py:968] (0/2) Epoch 23, batch 23400, giga_loss[loss=0.2815, simple_loss=0.3514, pruned_loss=0.1058, over 29080.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3464, pruned_loss=0.09958, over 5705297.45 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3534, pruned_loss=0.102, over 5718391.70 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.344, pruned_loss=0.09788, over 5717470.71 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:00:49,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1026814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:00:51,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.567e+02 1.270e+03 1.627e+03 2.082e+03 7.430e+03, threshold=3.254e+03, percent-clipped=8.0
+2023-03-12 00:01:01,261 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 00:01:13,940 INFO [train.py:968] (0/2) Epoch 23, batch 23450, libri_loss[loss=0.2678, simple_loss=0.3409, pruned_loss=0.09735, over 29552.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3517, pruned_loss=0.1045, over 5694841.86 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.353, pruned_loss=0.1019, over 5721711.07 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3502, pruned_loss=0.1033, over 5700739.37 frames. ], batch size: 79, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:01:23,789 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5525, 1.7647, 1.2831, 1.3001], device='cuda:0'), covar=tensor([0.1015, 0.0617, 0.1093, 0.1174], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0447, 0.0521, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 00:01:30,351 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1026858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:02:01,025 INFO [train.py:968] (0/2) Epoch 23, batch 23500, giga_loss[loss=0.3, simple_loss=0.3677, pruned_loss=0.1161, over 28971.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3566, pruned_loss=0.1083, over 5684885.26 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3526, pruned_loss=0.1019, over 5719107.20 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3558, pruned_loss=0.1075, over 5690699.89 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:02:21,782 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5842, 1.6422, 1.8297, 1.4038], device='cuda:0'), covar=tensor([0.1575, 0.2243, 0.1319, 0.1708], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0700, 0.0951, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 00:02:23,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.723e+03 2.240e+03 3.346e+03 1.230e+04, threshold=4.480e+03, percent-clipped=26.0
+2023-03-12 00:02:47,706 INFO [train.py:968] (0/2) Epoch 23, batch 23550, giga_loss[loss=0.3283, simple_loss=0.3951, pruned_loss=0.1307, over 28675.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3624, pruned_loss=0.1123, over 5679633.34 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3526, pruned_loss=0.102, over 5718124.62 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3619, pruned_loss=0.1117, over 5684471.36 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:03:04,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1026957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:03:06,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1026960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:03:16,990 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1026967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:03:37,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:03:42,162 INFO [train.py:968] (0/2) Epoch 23, batch 23600, giga_loss[loss=0.2933, simple_loss=0.3647, pruned_loss=0.1109, over 28443.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3693, pruned_loss=0.1178, over 5680046.05 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.353, pruned_loss=0.1023, over 5720678.70 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3687, pruned_loss=0.1172, over 5681230.68 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:04:08,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.758e+02 1.715e+03 2.288e+03 3.106e+03 6.078e+03, threshold=4.575e+03, percent-clipped=6.0
+2023-03-12 00:04:32,734 INFO [train.py:968] (0/2) Epoch 23, batch 23650, giga_loss[loss=0.3246, simple_loss=0.388, pruned_loss=0.1306, over 28844.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3761, pruned_loss=0.1238, over 5677230.06 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3529, pruned_loss=0.1025, over 5723285.37 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3759, pruned_loss=0.1235, over 5675309.31 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:04:33,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2619, 1.4319, 1.4202, 1.2495], device='cuda:0'), covar=tensor([0.2314, 0.2239, 0.1703, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.1990, 0.1942, 0.1862, 0.1994], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 00:04:57,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0910, 3.4418, 1.2456, 1.4932], device='cuda:0'), covar=tensor([0.1265, 0.0469, 0.1041, 0.1508], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0557, 0.0391, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 00:05:18,464 INFO [train.py:968] (0/2) Epoch 23, batch 23700, giga_loss[loss=0.3297, simple_loss=0.3894, pruned_loss=0.135, over 28702.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3782, pruned_loss=0.1257, over 5674289.29 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3528, pruned_loss=0.1025, over 5720435.42 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3792, pruned_loss=0.1264, over 5673593.41 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:05:43,509 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.87 vs. limit=5.0
+2023-03-12 00:05:46,095 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.837e+03 2.224e+03 3.069e+03 7.838e+03, threshold=4.448e+03, percent-clipped=1.0
+2023-03-12 00:06:07,537 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-12 00:06:12,024 INFO [train.py:968] (0/2) Epoch 23, batch 23750, giga_loss[loss=0.4402, simple_loss=0.4534, pruned_loss=0.2135, over 26475.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3824, pruned_loss=0.1304, over 5658227.62 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3529, pruned_loss=0.1026, over 5719425.77 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3835, pruned_loss=0.1312, over 5657778.77 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:06:25,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7589, 1.8560, 1.9125, 1.6945], device='cuda:0'), covar=tensor([0.1937, 0.2292, 0.2147, 0.2152], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0757, 0.0724, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 00:07:04,642 INFO [train.py:968] (0/2) Epoch 23, batch 23800, giga_loss[loss=0.2915, simple_loss=0.3624, pruned_loss=0.1103, over 28643.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3856, pruned_loss=0.1341, over 5655436.05 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.353, pruned_loss=0.1027, over 5723126.81 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.387, pruned_loss=0.1352, over 5650610.13 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:07:07,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 00:07:15,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3133, 1.6035, 1.2755, 0.9660], device='cuda:0'), covar=tensor([0.2072, 0.2059, 0.2320, 0.1975], device='cuda:0'), in_proj_covar=tensor([0.1525, 0.1103, 0.1346, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 00:07:31,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.943e+03 2.546e+03 3.972e+03 1.876e+04, threshold=5.091e+03, percent-clipped=19.0
+2023-03-12 00:07:39,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9316, 3.7786, 3.6250, 2.0862], device='cuda:0'), covar=tensor([0.0691, 0.0814, 0.0829, 0.2213], device='cuda:0'), in_proj_covar=tensor([0.1245, 0.1147, 0.0976, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 00:07:50,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1027233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:07:50,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4922, 1.7759, 1.4014, 1.6085], device='cuda:0'), covar=tensor([0.2617, 0.2683, 0.3020, 0.2311], device='cuda:0'), in_proj_covar=tensor([0.1525, 0.1104, 0.1347, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 00:07:56,645 INFO [train.py:968] (0/2) Epoch 23, batch 23850, giga_loss[loss=0.3589, simple_loss=0.4115, pruned_loss=0.1532, over 28723.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.39, pruned_loss=0.1385, over 5646375.09 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.353, pruned_loss=0.1029, over 5724408.82 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3918, pruned_loss=0.14, over 5639732.85 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:08:03,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8157, 2.6565, 1.5879, 1.0045], device='cuda:0'), covar=tensor([0.7573, 0.3655, 0.4327, 0.6571], device='cuda:0'), in_proj_covar=tensor([0.1770, 0.1675, 0.1610, 0.1442], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 00:08:59,215 INFO [train.py:968] (0/2) Epoch 23, batch 23900, giga_loss[loss=0.3497, simple_loss=0.4017, pruned_loss=0.1488, over 28811.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3917, pruned_loss=0.1396, over 5650101.59 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3532, pruned_loss=0.1032, over 5724170.71 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3937, pruned_loss=0.1412, over 5643515.07 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:09:21,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0418, 2.0202, 1.8463, 1.8753], device='cuda:0'), covar=tensor([0.2006, 0.2795, 0.2540, 0.2408], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0757, 0.0724, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 00:09:27,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.161e+03 2.016e+03 2.333e+03 3.365e+03 8.793e+03, threshold=4.666e+03, percent-clipped=6.0
+2023-03-12 00:09:48,333 INFO [train.py:968] (0/2) Epoch 23, batch 23950, giga_loss[loss=0.3138, simple_loss=0.3735, pruned_loss=0.1271, over 28976.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3902, pruned_loss=0.1391, over 5620549.86 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3537, pruned_loss=0.1036, over 5707982.10 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3924, pruned_loss=0.141, over 5627975.33 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:09:49,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1027342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:10:00,949 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:10:06,746 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2822, 2.6070, 1.2938, 1.4039], device='cuda:0'), covar=tensor([0.0969, 0.0363, 0.0855, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0560, 0.0392, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 00:10:22,779 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1027376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:10:24,919 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1027379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:10:35,882 INFO [train.py:968] (0/2) Epoch 23, batch 24000, giga_loss[loss=0.3528, simple_loss=0.4051, pruned_loss=0.1503, over 28673.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3896, pruned_loss=0.1393, over 5636592.81 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3536, pruned_loss=0.1037, over 5710624.42 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.392, pruned_loss=0.1413, over 5639052.38 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:10:35,886 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 00:10:44,786 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2055, simple_loss=0.314, pruned_loss=0.0485, over 944034.00 frames. 
+2023-03-12 00:10:44,787 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 00:11:00,569 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1027408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:11:04,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:11:12,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.913e+03 2.687e+03 3.275e+03 1.003e+04, threshold=5.375e+03, percent-clipped=11.0
+2023-03-12 00:11:31,060 INFO [train.py:968] (0/2) Epoch 23, batch 24050, giga_loss[loss=0.304, simple_loss=0.3683, pruned_loss=0.1198, over 29070.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3887, pruned_loss=0.1376, over 5648053.53 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3535, pruned_loss=0.1037, over 5716414.43 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3919, pruned_loss=0.1403, over 5642665.93 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:12:21,599 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1027485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:12:23,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1027488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:12:23,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1081, 2.2332, 1.8774, 2.2403], device='cuda:0'), covar=tensor([0.2254, 0.2510, 0.2690, 0.2260], device='cuda:0'), in_proj_covar=tensor([0.1524, 0.1104, 0.1347, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 00:12:25,900 INFO [train.py:968] (0/2) Epoch 23, batch 24100, giga_loss[loss=0.3624, simple_loss=0.4159, pruned_loss=0.1545, over 27581.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3881, pruned_loss=0.1365, over 5645005.90 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3536, pruned_loss=0.1038, over 5718723.14 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3911, pruned_loss=0.1391, over 5637274.19 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:12:52,859 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1027517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:12:53,271 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.311e+03 1.638e+03 2.010e+03 2.763e+03 5.198e+03, threshold=4.020e+03, percent-clipped=0.0
+2023-03-12 00:12:55,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3964, 1.4638, 1.3672, 1.5359], device='cuda:0'), covar=tensor([0.0744, 0.0370, 0.0331, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-12 00:13:10,202 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:13:18,094 INFO [train.py:968] (0/2) Epoch 23, batch 24150, giga_loss[loss=0.3904, simple_loss=0.4288, pruned_loss=0.176, over 27552.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3899, pruned_loss=0.1379, over 5633681.79 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3539, pruned_loss=0.104, over 5720567.69 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3931, pruned_loss=0.1407, over 5623494.91 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:14:05,661 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 00:14:15,358 INFO [train.py:968] (0/2) Epoch 23, batch 24200, giga_loss[loss=0.3086, simple_loss=0.3749, pruned_loss=0.1211, over 28899.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3856, pruned_loss=0.1337, over 5631778.27 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3538, pruned_loss=0.1041, over 5722216.16 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3884, pruned_loss=0.1361, over 5621520.69 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:14:43,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.153e+03 1.925e+03 2.719e+03 3.540e+03 9.036e+03, threshold=5.438e+03, percent-clipped=16.0
+2023-03-12 00:15:09,632 INFO [train.py:968] (0/2) Epoch 23, batch 24250, giga_loss[loss=0.2981, simple_loss=0.3646, pruned_loss=0.1158, over 28031.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3821, pruned_loss=0.1292, over 5639679.63 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3535, pruned_loss=0.104, over 5723241.07 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3847, pruned_loss=0.1313, over 5630219.14 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:15:57,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3776, 3.3191, 1.4917, 1.5077], device='cuda:0'), covar=tensor([0.1042, 0.0487, 0.0942, 0.1417], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0560, 0.0393, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 00:16:00,229 INFO [train.py:968] (0/2) Epoch 23, batch 24300, giga_loss[loss=0.2957, simple_loss=0.368, pruned_loss=0.1117, over 28973.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3786, pruned_loss=0.126, over 5661205.39 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3538, pruned_loss=0.1044, over 5727754.91 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3811, pruned_loss=0.1278, over 5647849.65 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:16:24,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.196e+03 1.818e+03 2.464e+03 3.039e+03 8.416e+03, threshold=4.927e+03, percent-clipped=3.0
+2023-03-12 00:16:31,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3608, 0.8027, 0.8595, 1.3983], device='cuda:0'), covar=tensor([0.0735, 0.0403, 0.0361, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-12 00:16:35,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1027729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:16:48,220 INFO [train.py:968] (0/2) Epoch 23, batch 24350, giga_loss[loss=0.2957, simple_loss=0.3637, pruned_loss=0.1138, over 28761.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3758, pruned_loss=0.1237, over 5654826.51 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.354, pruned_loss=0.1047, over 5722280.83 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.378, pruned_loss=0.1254, over 5647397.81 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:17:01,975 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-12 00:17:20,127 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:17:34,611 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1027787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:17:37,454 INFO [train.py:968] (0/2) Epoch 23, batch 24400, giga_loss[loss=0.3051, simple_loss=0.3781, pruned_loss=0.1161, over 28982.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3735, pruned_loss=0.1221, over 5658454.71 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.354, pruned_loss=0.1048, over 5717787.64 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3756, pruned_loss=0.1237, over 5655585.05 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:17:45,757 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-12 00:18:03,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.726e+03 2.199e+03 2.790e+03 4.446e+03, threshold=4.397e+03, percent-clipped=0.0
+2023-03-12 00:18:27,791 INFO [train.py:968] (0/2) Epoch 23, batch 24450, libri_loss[loss=0.2937, simple_loss=0.363, pruned_loss=0.1122, over 29745.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3733, pruned_loss=0.1221, over 5664832.06 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3544, pruned_loss=0.1052, over 5726129.29 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3755, pruned_loss=0.1237, over 5652299.64 frames. ], batch size: 87, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:18:44,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7638, 1.0411, 2.8466, 2.6555], device='cuda:0'), covar=tensor([0.1843, 0.2788, 0.0634, 0.1050], device='cuda:0'), in_proj_covar=tensor([0.0770, 0.0657, 0.0972, 0.0928], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 00:18:52,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:19:02,308 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1027872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:19:05,876 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1027875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:19:23,325 INFO [train.py:968] (0/2) Epoch 23, batch 24500, giga_loss[loss=0.2835, simple_loss=0.3557, pruned_loss=0.1057, over 28892.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3723, pruned_loss=0.1209, over 5668385.33 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3547, pruned_loss=0.1054, over 5723509.65 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3739, pruned_loss=0.1221, over 5660603.60 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:19:28,172 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6124, 1.6411, 1.8335, 1.3915], device='cuda:0'), covar=tensor([0.1755, 0.2598, 0.1468, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0902, 0.0701, 0.0949, 0.0847], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 00:19:36,057 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1027904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:19:41,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1027909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:19:52,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.598e+02 1.526e+03 2.006e+03 2.698e+03 5.670e+03, threshold=4.012e+03, percent-clipped=2.0
+2023-03-12 00:19:58,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:20:06,047 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1027930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:20:07,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1027933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:20:18,528 INFO [train.py:968] (0/2) Epoch 23, batch 24550, giga_loss[loss=0.3901, simple_loss=0.4397, pruned_loss=0.1703, over 28652.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3727, pruned_loss=0.1188, over 5677488.55 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3553, pruned_loss=0.1059, over 5724235.66 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3736, pruned_loss=0.1195, over 5670065.96 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:20:41,095 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1027962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:21:09,136 INFO [train.py:968] (0/2) Epoch 23, batch 24600, giga_loss[loss=0.271, simple_loss=0.3543, pruned_loss=0.09386, over 29004.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3738, pruned_loss=0.1183, over 5668241.35 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3558, pruned_loss=0.1063, over 5728592.43 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3745, pruned_loss=0.1187, over 5656895.52 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:21:17,501 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1028000.pt
+2023-03-12 00:21:40,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.256e+02 1.734e+03 2.156e+03 3.046e+03 7.324e+03, threshold=4.312e+03, percent-clipped=13.0
+2023-03-12 00:22:03,282 INFO [train.py:968] (0/2) Epoch 23, batch 24650, giga_loss[loss=0.2982, simple_loss=0.3638, pruned_loss=0.1163, over 28492.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3742, pruned_loss=0.119, over 5666911.74 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3558, pruned_loss=0.1063, over 5729906.07 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3751, pruned_loss=0.1194, over 5656198.65 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:22:15,506 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:22:17,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:22:45,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:22:50,242 INFO [train.py:968] (0/2) Epoch 23, batch 24700, giga_loss[loss=0.3035, simple_loss=0.3723, pruned_loss=0.1173, over 28869.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3752, pruned_loss=0.1205, over 5664010.75 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3565, pruned_loss=0.107, over 5731333.96 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3758, pruned_loss=0.1207, over 5651866.10 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:23:12,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2012, 0.8790, 0.9174, 1.3604], device='cuda:0'), covar=tensor([0.0691, 0.0486, 0.0345, 0.0732], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:0')
+2023-03-12 00:23:19,878 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.770e+03 2.218e+03 3.102e+03 9.082e+03, threshold=4.435e+03, percent-clipped=11.0
+2023-03-12 00:23:30,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4270, 1.2022, 4.3921, 3.4648], device='cuda:0'), covar=tensor([0.1691, 0.2940, 0.0411, 0.1074], device='cuda:0'), in_proj_covar=tensor([0.0768, 0.0655, 0.0970, 0.0926], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 00:23:37,916 INFO [train.py:968] (0/2) Epoch 23, batch 24750, giga_loss[loss=0.3443, simple_loss=0.399, pruned_loss=0.1448, over 28673.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3743, pruned_loss=0.1212, over 5662229.92 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.357, pruned_loss=0.1074, over 5733547.78 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3746, pruned_loss=0.1213, over 5649110.05 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:23:44,395 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1028147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:23:44,743 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-12 00:24:15,286 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1028181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:24:26,037 INFO [train.py:968] (0/2) Epoch 23, batch 24800, giga_loss[loss=0.3287, simple_loss=0.3676, pruned_loss=0.1449, over 23553.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3713, pruned_loss=0.1199, over 5660615.37 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3569, pruned_loss=0.1075, over 5725120.41 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3721, pruned_loss=0.1202, over 5655668.93 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:24:53,071 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.180e+03 1.870e+03 2.337e+03 3.141e+03 9.317e+03, threshold=4.674e+03, percent-clipped=10.0
+2023-03-12 00:25:09,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1028237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:25:12,967 INFO [train.py:968] (0/2) Epoch 23, batch 24850, giga_loss[loss=0.295, simple_loss=0.3736, pruned_loss=0.1082, over 28642.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3704, pruned_loss=0.1193, over 5656097.21 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3572, pruned_loss=0.1077, over 5716967.55 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3711, pruned_loss=0.1196, over 5658802.35 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:25:58,223 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:25:58,621 INFO [train.py:968] (0/2) Epoch 23, batch 24900, libri_loss[loss=0.2847, simple_loss=0.3581, pruned_loss=0.1056, over 29533.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3702, pruned_loss=0.1178, over 5672777.62 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3574, pruned_loss=0.1078, over 5721463.79 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3708, pruned_loss=0.1181, over 5669660.33 frames. ], batch size: 81, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:26:00,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:26:08,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1028300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:26:25,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.708e+02 1.620e+03 1.990e+03 2.834e+03 6.554e+03, threshold=3.979e+03, percent-clipped=4.0
+2023-03-12 00:26:29,332 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:26:33,654 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-12 00:26:46,475 INFO [train.py:968] (0/2) Epoch 23, batch 24950, giga_loss[loss=0.2825, simple_loss=0.3588, pruned_loss=0.1031, over 28700.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3696, pruned_loss=0.1176, over 5661733.75 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3575, pruned_loss=0.1081, over 5721151.60 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3704, pruned_loss=0.118, over 5657975.58 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:26:58,173 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1028354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:27:25,211 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:27:29,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:27:38,890 INFO [train.py:968] (0/2) Epoch 23, batch 25000, giga_loss[loss=0.336, simple_loss=0.3775, pruned_loss=0.1473, over 26505.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3682, pruned_loss=0.1163, over 5667997.82 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3573, pruned_loss=0.108, over 5721141.20 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.369, pruned_loss=0.1166, over 5664592.58 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:27:56,161 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:28:01,562 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.706e+03 2.164e+03 3.154e+03 9.795e+03, threshold=4.328e+03, percent-clipped=11.0
+2023-03-12 00:28:20,998 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 00:28:26,865 INFO [train.py:968] (0/2) Epoch 23, batch 25050, libri_loss[loss=0.2952, simple_loss=0.3751, pruned_loss=0.1077, over 29520.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3666, pruned_loss=0.1155, over 5684848.27 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3572, pruned_loss=0.1079, over 5724868.94 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3676, pruned_loss=0.1161, over 5677617.27 frames. ], batch size: 81, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:28:28,691 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:28:30,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:29:00,614 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:29:18,642 INFO [train.py:968] (0/2) Epoch 23, batch 25100, giga_loss[loss=0.3008, simple_loss=0.3702, pruned_loss=0.1157, over 29022.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3665, pruned_loss=0.1166, over 5685615.83 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.357, pruned_loss=0.108, over 5728128.81 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3678, pruned_loss=0.1173, over 5675876.57 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:29:25,230 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 00:29:46,806 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.647e+03 2.104e+03 2.932e+03 1.126e+04, threshold=4.207e+03, percent-clipped=10.0
+2023-03-12 00:30:05,102 INFO [train.py:968] (0/2) Epoch 23, batch 25150, giga_loss[loss=0.271, simple_loss=0.3457, pruned_loss=0.09818, over 28662.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3654, pruned_loss=0.1161, over 5690500.10 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3571, pruned_loss=0.108, over 5724311.20 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3667, pruned_loss=0.1169, over 5685392.81 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:30:18,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5987, 1.6617, 1.8217, 1.3465], device='cuda:0'), covar=tensor([0.1737, 0.2606, 0.1480, 0.1751], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0704, 0.0952, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 00:30:20,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1028556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:30:27,074 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-12 00:30:58,742 INFO [train.py:968] (0/2) Epoch 23, batch 25200, giga_loss[loss=0.2455, simple_loss=0.322, pruned_loss=0.08451, over 28990.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3638, pruned_loss=0.1155, over 5691683.79 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3573, pruned_loss=0.1081, over 5723510.64 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3647, pruned_loss=0.116, over 5688283.55 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:31:10,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1515, 1.3134, 1.1245, 0.8932], device='cuda:0'), covar=tensor([0.1020, 0.0510, 0.1134, 0.1130], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0448, 0.0522, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 00:31:12,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8844, 1.0650, 1.0410, 0.8436], device='cuda:0'), covar=tensor([0.2044, 0.2203, 0.1438, 0.2000], device='cuda:0'), in_proj_covar=tensor([0.1985, 0.1941, 0.1859, 0.2003], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 00:31:25,851 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.744e+03 2.295e+03 2.993e+03 5.587e+03, threshold=4.590e+03, percent-clipped=6.0
+2023-03-12 00:31:47,760 INFO [train.py:968] (0/2) Epoch 23, batch 25250, giga_loss[loss=0.2844, simple_loss=0.3518, pruned_loss=0.1086, over 28793.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3638, pruned_loss=0.1161, over 5688385.44 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3575, pruned_loss=0.1083, over 5726633.20 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3644, pruned_loss=0.1166, over 5681793.85 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:32:18,226 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 00:32:43,277 INFO [train.py:968] (0/2) Epoch 23, batch 25300, giga_loss[loss=0.3016, simple_loss=0.3693, pruned_loss=0.1169, over 28877.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3639, pruned_loss=0.1164, over 5681289.67 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3577, pruned_loss=0.1086, over 5725965.84 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3643, pruned_loss=0.1166, over 5676354.36 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:32:50,588 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:32:53,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:33:14,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.728e+02 1.771e+03 2.280e+03 3.151e+03 1.154e+04, threshold=4.559e+03, percent-clipped=9.0
+2023-03-12 00:33:23,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1028729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:33:25,233 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:33:34,458 INFO [train.py:968] (0/2) Epoch 23, batch 25350, giga_loss[loss=0.292, simple_loss=0.3691, pruned_loss=0.1074, over 29031.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3638, pruned_loss=0.1154, over 5676489.29 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3576, pruned_loss=0.1086, over 5716938.97 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3642, pruned_loss=0.1157, over 5679781.49 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:33:38,345 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1028745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:33:40,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4534, 3.3126, 1.5483, 1.5651], device='cuda:0'), covar=tensor([0.0997, 0.0416, 0.0919, 0.1344], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0560, 0.0392, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 00:34:17,515 INFO [train.py:968] (0/2) Epoch 23, batch 25400, giga_loss[loss=0.2974, simple_loss=0.3704, pruned_loss=0.1122, over 28952.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3644, pruned_loss=0.1152, over 5670630.64 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3574, pruned_loss=0.1084, over 5710688.26 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3651, pruned_loss=0.1157, over 5678349.01 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:34:31,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0111, 5.8399, 5.5356, 3.2269], device='cuda:0'), covar=tensor([0.0448, 0.0605, 0.0711, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.1263, 0.1168, 0.0989, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 00:34:37,208 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1028810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:34:45,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.236e+02 1.658e+03 1.990e+03 2.593e+03 9.118e+03, threshold=3.980e+03, percent-clipped=6.0
+2023-03-12 00:35:07,395 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-12 00:35:07,456 INFO [train.py:968] (0/2) Epoch 23, batch 25450, giga_loss[loss=0.2931, simple_loss=0.3645, pruned_loss=0.1109, over 28239.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3656, pruned_loss=0.1158, over 5661389.27 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3574, pruned_loss=0.1087, over 5697713.13 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3664, pruned_loss=0.1162, over 5677395.85 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:35:35,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:35:39,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:35:52,828 INFO [train.py:968] (0/2) Epoch 23, batch 25500, giga_loss[loss=0.2797, simple_loss=0.3516, pruned_loss=0.1039, over 28963.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3662, pruned_loss=0.1166, over 5666961.13 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3572, pruned_loss=0.1086, over 5699398.34 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3674, pruned_loss=0.1173, over 5677316.67 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:35:54,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9399, 1.2456, 1.2916, 1.0934], device='cuda:0'), covar=tensor([0.1719, 0.1283, 0.2171, 0.1535], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0757, 0.0721, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 00:36:06,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:36:24,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.266e+03 1.770e+03 2.320e+03 2.957e+03 8.180e+03, threshold=4.640e+03, percent-clipped=7.0
+2023-03-12 00:36:45,500 INFO [train.py:968] (0/2) Epoch 23, batch 25550, giga_loss[loss=0.3154, simple_loss=0.3759, pruned_loss=0.1275, over 28649.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3703, pruned_loss=0.1205, over 5664428.43 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3577, pruned_loss=0.109, over 5692529.08 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3708, pruned_loss=0.1207, over 5679201.51 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:36:48,719 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-12 00:37:12,049 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1028967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:37:17,160 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5692, 1.8966, 1.6901, 1.6851], device='cuda:0'), covar=tensor([0.2094, 0.2222, 0.2514, 0.2238], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0756, 0.0722, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 00:37:33,476 INFO [train.py:968] (0/2) Epoch 23, batch 25600, giga_loss[loss=0.2935, simple_loss=0.3556, pruned_loss=0.1157, over 28627.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3714, pruned_loss=0.1229, over 5671476.56 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3574, pruned_loss=0.1089, over 5698612.28 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3725, pruned_loss=0.1236, over 5677041.15 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:38:07,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.883e+03 2.329e+03 2.947e+03 1.059e+04, threshold=4.658e+03, percent-clipped=7.0
+2023-03-12 00:38:22,254 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6642, 1.7776, 1.3418, 1.3458], device='cuda:0'), covar=tensor([0.0951, 0.0642, 0.1062, 0.1142], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0448, 0.0520, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 00:38:27,841 INFO [train.py:968] (0/2) Epoch 23, batch 25650, giga_loss[loss=0.2865, simple_loss=0.3571, pruned_loss=0.108, over 28692.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3724, pruned_loss=0.1248, over 5665736.48 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3577, pruned_loss=0.1091, over 5692851.45 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3733, pruned_loss=0.1254, over 5674904.05 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:38:28,755 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:39:14,793 INFO [train.py:968] (0/2) Epoch 23, batch 25700, giga_loss[loss=0.28, simple_loss=0.3472, pruned_loss=0.1064, over 28995.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3716, pruned_loss=0.1241, over 5667509.40 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3575, pruned_loss=0.109, over 5685725.40 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3727, pruned_loss=0.1249, over 5679875.47 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:39:36,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4987, 1.5946, 1.7102, 1.3397], device='cuda:0'), covar=tensor([0.1433, 0.2079, 0.1204, 0.1494], device='cuda:0'), in_proj_covar=tensor([0.0903, 0.0702, 0.0948, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 00:39:41,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:39:43,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.144e+03 1.819e+03 2.451e+03 3.382e+03 7.522e+03, threshold=4.902e+03, percent-clipped=7.0
+2023-03-12 00:40:03,900 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:40:05,144 INFO [train.py:968] (0/2) Epoch 23, batch 25750, giga_loss[loss=0.2952, simple_loss=0.3645, pruned_loss=0.1129, over 28570.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3712, pruned_loss=0.1244, over 5652171.79 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3577, pruned_loss=0.1093, over 5679363.97 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.372, pruned_loss=0.125, over 5667929.24 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:40:07,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5083, 1.8131, 1.4623, 1.3740], device='cuda:0'), covar=tensor([0.2676, 0.2628, 0.3129, 0.2351], device='cuda:0'), in_proj_covar=tensor([0.1527, 0.1106, 0.1350, 0.0997], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 00:40:44,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:40:48,520 INFO [train.py:968] (0/2) Epoch 23, batch 25800, giga_loss[loss=0.2948, simple_loss=0.3591, pruned_loss=0.1153, over 28578.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3707, pruned_loss=0.1219, over 5657562.36 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3584, pruned_loss=0.1097, over 5676849.58 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3713, pruned_loss=0.1226, over 5672043.83 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:41:01,068 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:41:19,155 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.702e+03 2.278e+03 3.654e+03 9.583e+03, threshold=4.556e+03, percent-clipped=7.0
+2023-03-12 00:41:39,567 INFO [train.py:968] (0/2) Epoch 23, batch 25850, giga_loss[loss=0.3015, simple_loss=0.3397, pruned_loss=0.1316, over 23907.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.369, pruned_loss=0.1204, over 5643181.63 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3586, pruned_loss=0.1099, over 5676603.86 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3694, pruned_loss=0.1209, over 5654575.86 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:41:57,878 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:42:01,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:42:12,034 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:42:15,727 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 00:42:25,696 INFO [train.py:968] (0/2) Epoch 23, batch 25900, giga_loss[loss=0.2929, simple_loss=0.3577, pruned_loss=0.114, over 28753.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1191, over 5654301.15 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3586, pruned_loss=0.11, over 5680411.85 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3674, pruned_loss=0.1196, over 5659361.48 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:42:28,820 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:42:53,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.811e+03 2.206e+03 3.131e+03 1.015e+04, threshold=4.413e+03, percent-clipped=12.0
+2023-03-12 00:43:00,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:43:03,274 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:43:13,834 INFO [train.py:968] (0/2) Epoch 23, batch 25950, giga_loss[loss=0.3488, simple_loss=0.38, pruned_loss=0.1588, over 23967.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.366, pruned_loss=0.1195, over 5652580.14 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3586, pruned_loss=0.1099, over 5683234.74 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3666, pruned_loss=0.1201, over 5653779.10 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:43:15,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:43:31,661 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:44:02,582 INFO [train.py:968] (0/2) Epoch 23, batch 26000, giga_loss[loss=0.321, simple_loss=0.3884, pruned_loss=0.1268, over 28959.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3684, pruned_loss=0.1214, over 5650098.88 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3593, pruned_loss=0.1104, over 5678524.21 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3684, pruned_loss=0.1217, over 5653820.01 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:44:05,151 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-12 00:44:13,144 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5098, 2.2067, 1.5706, 0.6281], device='cuda:0'), covar=tensor([0.4205, 0.2854, 0.4090, 0.5780], device='cuda:0'), in_proj_covar=tensor([0.1761, 0.1671, 0.1604, 0.1440], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 00:44:14,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3561, 1.4308, 1.2282, 1.5326], device='cuda:0'), covar=tensor([0.0771, 0.0384, 0.0356, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-12 00:44:24,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:44:29,925 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.722e+03 2.242e+03 2.976e+03 7.611e+03, threshold=4.484e+03, percent-clipped=10.0
+2023-03-12 00:44:49,812 INFO [train.py:968] (0/2) Epoch 23, batch 26050, giga_loss[loss=0.3484, simple_loss=0.4038, pruned_loss=0.1465, over 27570.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3712, pruned_loss=0.1226, over 5659132.79 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3592, pruned_loss=0.1106, over 5684355.60 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.123, over 5656681.38 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:45:32,725 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:45:34,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:45:36,574 INFO [train.py:968] (0/2) Epoch 23, batch 26100, giga_loss[loss=0.2772, simple_loss=0.3627, pruned_loss=0.09579, over 29050.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3743, pruned_loss=0.122, over 5658522.07 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3591, pruned_loss=0.1107, over 5680640.75 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.375, pruned_loss=0.1224, over 5659186.33 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:45:57,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:46:01,421 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:46:08,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.548e+03 2.150e+03 2.730e+03 7.227e+03, threshold=4.300e+03, percent-clipped=3.0
+2023-03-12 00:46:17,347 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:46:25,192 INFO [train.py:968] (0/2) Epoch 23, batch 26150, giga_loss[loss=0.3458, simple_loss=0.3969, pruned_loss=0.1474, over 27602.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3747, pruned_loss=0.1218, over 5652667.80 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.359, pruned_loss=0.1108, over 5677342.66 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3758, pruned_loss=0.1224, over 5655078.12 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:46:41,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:46:44,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:47:01,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:47:14,513 INFO [train.py:968] (0/2) Epoch 23, batch 26200, giga_loss[loss=0.3959, simple_loss=0.4308, pruned_loss=0.1805, over 27628.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3767, pruned_loss=0.1238, over 5639647.50 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3593, pruned_loss=0.111, over 5668053.05 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3777, pruned_loss=0.1244, over 5649241.64 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:47:15,535 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:47:44,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.651e+03 2.054e+03 2.826e+03 4.995e+03, threshold=4.108e+03, percent-clipped=4.0
+2023-03-12 00:48:00,163 INFO [train.py:968] (0/2) Epoch 23, batch 26250, giga_loss[loss=0.3296, simple_loss=0.3722, pruned_loss=0.1435, over 23391.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3769, pruned_loss=0.1244, over 5645141.14 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3595, pruned_loss=0.1112, over 5670615.02 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3778, pruned_loss=0.1248, over 5650255.78 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:48:12,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:48:18,964 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:48:21,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:48:30,403 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:48:49,463 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:48:51,081 INFO [train.py:968] (0/2) Epoch 23, batch 26300, giga_loss[loss=0.2883, simple_loss=0.3594, pruned_loss=0.1086, over 28685.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3757, pruned_loss=0.1243, over 5641953.43 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3599, pruned_loss=0.1114, over 5672380.54 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3762, pruned_loss=0.1247, over 5644241.99 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:49:18,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-12 00:49:21,662 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:49:21,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.206e+03 1.774e+03 2.389e+03 3.319e+03 7.179e+03, threshold=4.779e+03, percent-clipped=16.0
+2023-03-12 00:49:23,063 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:49:23,692 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:49:26,839 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 00:49:37,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4939, 1.3792, 1.6154, 1.2279], device='cuda:0'), covar=tensor([0.1539, 0.2675, 0.1230, 0.1405], device='cuda:0'), in_proj_covar=tensor([0.0905, 0.0704, 0.0950, 0.0849], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 00:49:37,687 INFO [train.py:968] (0/2) Epoch 23, batch 26350, giga_loss[loss=0.4019, simple_loss=0.428, pruned_loss=0.1878, over 26709.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3746, pruned_loss=0.1243, over 5637255.32 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3606, pruned_loss=0.1121, over 5668547.31 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3748, pruned_loss=0.1242, over 5642591.79 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:49:52,088 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:50:27,124 INFO [train.py:968] (0/2) Epoch 23, batch 26400, giga_loss[loss=0.2927, simple_loss=0.3513, pruned_loss=0.117, over 28523.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3726, pruned_loss=0.1237, over 5649557.45 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3609, pruned_loss=0.1122, over 5670525.05 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3726, pruned_loss=0.1237, over 5651823.51 frames. ], batch size: 65, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:50:29,550 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:50:33,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:50:50,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:50:58,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.931e+03 2.423e+03 3.202e+03 1.208e+04, threshold=4.847e+03, percent-clipped=6.0
+2023-03-12 00:51:00,314 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:51:14,414 INFO [train.py:968] (0/2) Epoch 23, batch 26450, giga_loss[loss=0.2969, simple_loss=0.3623, pruned_loss=0.1158, over 28843.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3722, pruned_loss=0.1242, over 5636820.74 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3613, pruned_loss=0.1126, over 5666786.06 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3722, pruned_loss=0.1242, over 5640206.29 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:52:01,925 INFO [train.py:968] (0/2) Epoch 23, batch 26500, giga_loss[loss=0.2846, simple_loss=0.3643, pruned_loss=0.1024, over 28939.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3738, pruned_loss=0.1255, over 5634096.01 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3618, pruned_loss=0.1132, over 5659090.15 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3735, pruned_loss=0.1253, over 5643583.82 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:52:14,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:52:29,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.172e+03 1.777e+03 2.699e+03 3.768e+03 7.094e+03, threshold=5.399e+03, percent-clipped=12.0
+2023-03-12 00:52:45,587 INFO [train.py:968] (0/2) Epoch 23, batch 26550, giga_loss[loss=0.3438, simple_loss=0.3918, pruned_loss=0.1479, over 28601.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.371, pruned_loss=0.1239, over 5649213.11 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3615, pruned_loss=0.1131, over 5662899.13 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3713, pruned_loss=0.124, over 5652694.64 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:53:30,458 INFO [train.py:968] (0/2) Epoch 23, batch 26600, giga_loss[loss=0.3572, simple_loss=0.409, pruned_loss=0.1526, over 27538.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3712, pruned_loss=0.1244, over 5658643.85 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3621, pruned_loss=0.1137, over 5662622.03 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3712, pruned_loss=0.1243, over 5662130.54 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:53:37,314 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1030000.pt
+2023-03-12 00:53:50,967 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:54:01,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.759e+03 2.151e+03 2.860e+03 6.663e+03, threshold=4.302e+03, percent-clipped=4.0
+2023-03-12 00:54:16,464 INFO [train.py:968] (0/2) Epoch 23, batch 26650, giga_loss[loss=0.3061, simple_loss=0.3758, pruned_loss=0.1182, over 28224.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3703, pruned_loss=0.1231, over 5664917.02 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.362, pruned_loss=0.1136, over 5666306.42 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3707, pruned_loss=0.1235, over 5664513.87 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:54:20,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:54:25,241 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:54:28,095 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:54:53,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0244, 1.3400, 1.1372, 0.1973], device='cuda:0'), covar=tensor([0.4069, 0.3303, 0.4668, 0.6724], device='cuda:0'), in_proj_covar=tensor([0.1759, 0.1666, 0.1598, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 00:54:55,999 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:55:06,028 INFO [train.py:968] (0/2) Epoch 23, batch 26700, libri_loss[loss=0.2583, simple_loss=0.3279, pruned_loss=0.09431, over 29649.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3727, pruned_loss=0.1241, over 5661920.89 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3617, pruned_loss=0.1134, over 5669852.34 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3735, pruned_loss=0.1247, over 5658241.54 frames. ], batch size: 69, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:55:13,132 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:55:20,853 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:55:39,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.167e+03 1.676e+03 2.247e+03 2.895e+03 8.287e+03, threshold=4.494e+03, percent-clipped=4.0
+2023-03-12 00:55:55,988 INFO [train.py:968] (0/2) Epoch 23, batch 26750, giga_loss[loss=0.2738, simple_loss=0.3412, pruned_loss=0.1032, over 28959.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.371, pruned_loss=0.1231, over 5651357.95 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3614, pruned_loss=0.1133, over 5663011.92 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.372, pruned_loss=0.1237, over 5654535.57 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:56:40,215 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:56:43,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:56:43,738 INFO [train.py:968] (0/2) Epoch 23, batch 26800, giga_loss[loss=0.275, simple_loss=0.3659, pruned_loss=0.09206, over 28980.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.372, pruned_loss=0.1225, over 5656579.92 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.362, pruned_loss=0.114, over 5659314.34 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3724, pruned_loss=0.1226, over 5661516.01 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:56:43,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:57:10,442 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:57:15,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.154e+02 1.554e+03 1.895e+03 2.728e+03 7.126e+03, threshold=3.790e+03, percent-clipped=5.0
+2023-03-12 00:57:31,994 INFO [train.py:968] (0/2) Epoch 23, batch 26850, giga_loss[loss=0.3287, simple_loss=0.4042, pruned_loss=0.1266, over 28509.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3726, pruned_loss=0.1203, over 5666012.35 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5661014.15 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3734, pruned_loss=0.1206, over 5668298.49 frames. ], batch size: 78, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:57:33,859 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:57:38,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:58:01,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6529, 1.6005, 1.8038, 1.3690], device='cuda:0'), covar=tensor([0.1836, 0.2720, 0.1556, 0.1914], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0706, 0.0952, 0.0851], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 00:58:04,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:58:21,238 INFO [train.py:968] (0/2) Epoch 23, batch 26900, giga_loss[loss=0.4127, simple_loss=0.4388, pruned_loss=0.1933, over 27622.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3752, pruned_loss=0.1201, over 5673089.27 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.362, pruned_loss=0.114, over 5665009.66 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3757, pruned_loss=0.1203, over 5671402.28 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:58:35,043 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:58:52,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.587e+02 1.528e+03 1.956e+03 2.593e+03 6.358e+03, threshold=3.913e+03, percent-clipped=9.0
+2023-03-12 00:58:58,899 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:59:02,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:59:02,864 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 00:59:08,337 INFO [train.py:968] (0/2) Epoch 23, batch 26950, giga_loss[loss=0.2831, simple_loss=0.3571, pruned_loss=0.1046, over 28669.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3788, pruned_loss=0.1234, over 5666594.59 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3619, pruned_loss=0.114, over 5660212.07 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3795, pruned_loss=0.1237, over 5669638.24 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:59:19,828 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 00:59:29,515 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:59:33,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6766, 1.8298, 1.7196, 1.6403], device='cuda:0'), covar=tensor([0.1885, 0.2258, 0.2335, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0758, 0.0722, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 00:59:50,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:59:54,426 INFO [train.py:968] (0/2) Epoch 23, batch 27000, giga_loss[loss=0.3764, simple_loss=0.4038, pruned_loss=0.1746, over 23584.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.381, pruned_loss=0.1264, over 5647476.94 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.362, pruned_loss=0.1142, over 5640971.03 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3821, pruned_loss=0.1268, over 5666805.30 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:59:54,431 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 01:00:01,756 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3206, 1.2649, 1.1759, 1.4667], device='cuda:0'), covar=tensor([0.0810, 0.0366, 0.0359, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:0')
+2023-03-12 01:00:04,550 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2036, simple_loss=0.3103, pruned_loss=0.04842, over 944034.00 frames. 
+2023-03-12 01:00:04,551 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 01:00:38,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.252e+03 1.882e+03 2.372e+03 3.194e+03 9.071e+03, threshold=4.743e+03, percent-clipped=15.0
+2023-03-12 01:00:42,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:00:52,041 INFO [train.py:968] (0/2) Epoch 23, batch 27050, giga_loss[loss=0.3511, simple_loss=0.4006, pruned_loss=0.1508, over 28645.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3804, pruned_loss=0.1267, over 5668722.07 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3625, pruned_loss=0.1146, over 5651319.89 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3816, pruned_loss=0.1271, over 5675802.99 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:01:00,358 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:01:33,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:01:40,060 INFO [train.py:968] (0/2) Epoch 23, batch 27100, giga_loss[loss=0.3005, simple_loss=0.3693, pruned_loss=0.1159, over 28966.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3807, pruned_loss=0.1278, over 5658055.35 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3622, pruned_loss=0.1145, over 5650773.64 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3823, pruned_loss=0.1286, over 5664607.34 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:02:14,829 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.764e+03 2.283e+03 3.018e+03 9.187e+03, threshold=4.567e+03, percent-clipped=7.0
+2023-03-12 01:02:20,861 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:02:23,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:02:31,662 INFO [train.py:968] (0/2) Epoch 23, batch 27150, giga_loss[loss=0.2711, simple_loss=0.3622, pruned_loss=0.09002, over 28511.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3788, pruned_loss=0.1251, over 5668544.86 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3625, pruned_loss=0.1146, over 5654476.53 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3802, pruned_loss=0.1258, over 5670853.24 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:02:51,063 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:03:12,318 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4629, 1.6966, 1.1766, 1.3030], device='cuda:0'), covar=tensor([0.1031, 0.0547, 0.1158, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0448, 0.0521, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 01:03:18,937 INFO [train.py:968] (0/2) Epoch 23, batch 27200, libri_loss[loss=0.2528, simple_loss=0.3183, pruned_loss=0.09362, over 29632.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3774, pruned_loss=0.1229, over 5659897.00 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3623, pruned_loss=0.1147, over 5659366.55 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.379, pruned_loss=0.1236, over 5657469.35 frames. ], batch size: 69, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 01:03:54,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.537e+03 1.850e+03 2.547e+03 5.737e+03, threshold=3.700e+03, percent-clipped=4.0
+2023-03-12 01:03:54,837 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:03:58,813 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:04:11,867 INFO [train.py:968] (0/2) Epoch 23, batch 27250, giga_loss[loss=0.3618, simple_loss=0.3902, pruned_loss=0.1667, over 23525.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3788, pruned_loss=0.1234, over 5662620.62 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3623, pruned_loss=0.1147, over 5659366.55 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3801, pruned_loss=0.1239, over 5660731.16 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:04:28,411 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:04:37,848 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030667.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 01:04:52,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:05:01,710 INFO [train.py:968] (0/2) Epoch 23, batch 27300, libri_loss[loss=0.3069, simple_loss=0.3695, pruned_loss=0.1221, over 29567.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3788, pruned_loss=0.1241, over 5659370.06 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3622, pruned_loss=0.1146, over 5667704.63 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3806, pruned_loss=0.1249, over 5650065.20 frames. ], batch size: 76, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:05:31,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.761e+03 2.286e+03 2.995e+03 9.507e+03, threshold=4.573e+03, percent-clipped=15.0
+2023-03-12 01:05:45,934 INFO [train.py:968] (0/2) Epoch 23, batch 27350, giga_loss[loss=0.29, simple_loss=0.3627, pruned_loss=0.1087, over 28820.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3771, pruned_loss=0.123, over 5670967.36 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1145, over 5673836.14 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3788, pruned_loss=0.124, over 5657948.77 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:06:36,564 INFO [train.py:968] (0/2) Epoch 23, batch 27400, giga_loss[loss=0.3183, simple_loss=0.3758, pruned_loss=0.1304, over 28315.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3753, pruned_loss=0.1227, over 5671946.38 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1145, over 5672359.10 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3771, pruned_loss=0.1237, over 5662509.21 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:06:48,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:07:14,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:07:14,360 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:07:14,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.832e+03 2.344e+03 3.137e+03 1.141e+04, threshold=4.687e+03, percent-clipped=6.0
+2023-03-12 01:07:16,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:07:32,284 INFO [train.py:968] (0/2) Epoch 23, batch 27450, giga_loss[loss=0.2641, simple_loss=0.333, pruned_loss=0.09758, over 28668.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3745, pruned_loss=0.1227, over 5667695.67 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3624, pruned_loss=0.1148, over 5667390.45 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3758, pruned_loss=0.1233, over 5664844.96 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:07:49,343 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:08:22,394 INFO [train.py:968] (0/2) Epoch 23, batch 27500, giga_loss[loss=0.2713, simple_loss=0.3443, pruned_loss=0.09915, over 28701.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3721, pruned_loss=0.1217, over 5664825.41 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3625, pruned_loss=0.1149, over 5669364.62 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3732, pruned_loss=0.1222, over 5661006.06 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:08:41,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3359, 3.7852, 1.5488, 1.4673], device='cuda:0'), covar=tensor([0.0997, 0.0357, 0.0889, 0.1363], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0564, 0.0394, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 01:08:58,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.772e+03 2.400e+03 4.025e+03 1.299e+04, threshold=4.800e+03, percent-clipped=18.0
+2023-03-12 01:09:08,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4608, 1.7779, 1.4401, 1.4576], device='cuda:0'), covar=tensor([0.2397, 0.2335, 0.2598, 0.2244], device='cuda:0'), in_proj_covar=tensor([0.1525, 0.1105, 0.1350, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 01:09:10,808 INFO [train.py:968] (0/2) Epoch 23, batch 27550, giga_loss[loss=0.2863, simple_loss=0.3566, pruned_loss=0.108, over 28835.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3716, pruned_loss=0.1221, over 5671620.89 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5676208.64 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3726, pruned_loss=0.1227, over 5662056.84 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:09:14,596 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:09:17,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:09:35,347 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:09:38,340 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:09:42,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2076, 2.2911, 1.7976, 1.7790], device='cuda:0'), covar=tensor([0.0982, 0.0717, 0.1001, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0448, 0.0521, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 01:09:44,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:09:49,547 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 01:09:54,731 INFO [train.py:968] (0/2) Epoch 23, batch 27600, giga_loss[loss=0.2771, simple_loss=0.3595, pruned_loss=0.09741, over 28916.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3701, pruned_loss=0.1209, over 5664997.54 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5672993.37 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3712, pruned_loss=0.1215, over 5660665.41 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 01:10:04,292 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1031000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:10:16,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3096, 1.4937, 1.3690, 1.4690], device='cuda:0'), covar=tensor([0.0796, 0.0346, 0.0336, 0.0878], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:0')
+2023-03-12 01:10:28,663 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.568e+03 1.898e+03 2.641e+03 7.375e+03, threshold=3.796e+03, percent-clipped=5.0
+2023-03-12 01:10:41,199 INFO [train.py:968] (0/2) Epoch 23, batch 27650, giga_loss[loss=0.2703, simple_loss=0.3475, pruned_loss=0.09656, over 28969.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3668, pruned_loss=0.1172, over 5666596.96 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3624, pruned_loss=0.1148, over 5676194.26 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.368, pruned_loss=0.1179, over 5660106.17 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 01:10:42,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1031042.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 01:11:32,712 INFO [train.py:968] (0/2) Epoch 23, batch 27700, giga_loss[loss=0.2961, simple_loss=0.3659, pruned_loss=0.1132, over 28521.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3647, pruned_loss=0.1155, over 5666090.43 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5680865.20 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3655, pruned_loss=0.1159, over 5656542.49 frames. ], batch size: 78, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:12:11,311 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.575e+02 1.473e+03 2.020e+03 2.876e+03 5.717e+03, threshold=4.040e+03, percent-clipped=11.0
+2023-03-12 01:12:27,173 INFO [train.py:968] (0/2) Epoch 23, batch 27750, giga_loss[loss=0.3175, simple_loss=0.3793, pruned_loss=0.1278, over 28689.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3638, pruned_loss=0.1154, over 5666994.19 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3624, pruned_loss=0.1148, over 5684098.07 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3646, pruned_loss=0.1159, over 5656395.74 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:12:33,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2487, 4.1021, 3.9072, 1.9729], device='cuda:0'), covar=tensor([0.0637, 0.0756, 0.0770, 0.1920], device='cuda:0'), in_proj_covar=tensor([0.1265, 0.1171, 0.0992, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 01:13:13,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1031185.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 01:13:15,200 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1031188.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 01:13:16,890 INFO [train.py:968] (0/2) Epoch 23, batch 27800, giga_loss[loss=0.4061, simple_loss=0.4236, pruned_loss=0.1943, over 26764.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3616, pruned_loss=0.1151, over 5641951.48 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.363, pruned_loss=0.1152, over 5669037.61 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3618, pruned_loss=0.1151, over 5645999.64 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:13:18,952 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1031191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:13:44,139 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1031217.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 01:13:54,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.217e+03 1.995e+03 2.356e+03 3.173e+03 1.718e+04, threshold=4.712e+03, percent-clipped=14.0
+2023-03-12 01:14:05,067 INFO [train.py:968] (0/2) Epoch 23, batch 27850, giga_loss[loss=0.3077, simple_loss=0.3838, pruned_loss=0.1158, over 28897.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3622, pruned_loss=0.116, over 5638997.03 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3625, pruned_loss=0.1151, over 5665775.61 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3627, pruned_loss=0.1161, over 5644624.18 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:14:06,679 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-12 01:14:56,041 INFO [train.py:968] (0/2) Epoch 23, batch 27900, giga_loss[loss=0.3434, simple_loss=0.4003, pruned_loss=0.1433, over 28219.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3635, pruned_loss=0.1156, over 5642976.12 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3624, pruned_loss=0.1151, over 5660048.63 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.364, pruned_loss=0.1158, over 5652078.20 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:15:32,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.708e+03 2.230e+03 3.024e+03 7.007e+03, threshold=4.460e+03, percent-clipped=4.0
+2023-03-12 01:15:45,134 INFO [train.py:968] (0/2) Epoch 23, batch 27950, giga_loss[loss=0.3868, simple_loss=0.4168, pruned_loss=0.1784, over 26594.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3648, pruned_loss=0.116, over 5643391.59 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3627, pruned_loss=0.1151, over 5663351.59 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3649, pruned_loss=0.116, over 5647083.95 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:15:52,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0637, 1.4091, 1.1651, 0.2738], device='cuda:0'), covar=tensor([0.3465, 0.3307, 0.4065, 0.6045], device='cuda:0'), in_proj_covar=tensor([0.1769, 0.1668, 0.1600, 0.1439], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 01:16:10,895 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2622, 1.3121, 1.3233, 1.4574], device='cuda:0'), covar=tensor([0.0760, 0.0411, 0.0329, 0.0813], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:0')
+2023-03-12 01:16:31,294 INFO [train.py:968] (0/2) Epoch 23, batch 28000, giga_loss[loss=0.3253, simple_loss=0.3814, pruned_loss=0.1346, over 28698.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.366, pruned_loss=0.1174, over 5644404.92 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3634, pruned_loss=0.1157, over 5667435.05 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3656, pruned_loss=0.1171, over 5643354.46 frames. ], batch size: 78, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:16:50,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1150, 4.9542, 4.7242, 2.5001], device='cuda:0'), covar=tensor([0.0476, 0.0572, 0.0646, 0.1718], device='cuda:0'), in_proj_covar=tensor([0.1272, 0.1175, 0.1000, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 01:16:51,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2184, 1.7671, 1.3861, 0.3982], device='cuda:0'), covar=tensor([0.5070, 0.3064, 0.4524, 0.6379], device='cuda:0'), in_proj_covar=tensor([0.1777, 0.1673, 0.1605, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 01:16:52,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1031414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:17:04,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.650e+03 2.093e+03 2.760e+03 5.034e+03, threshold=4.185e+03, percent-clipped=3.0
+2023-03-12 01:17:14,401 INFO [train.py:968] (0/2) Epoch 23, batch 28050, giga_loss[loss=0.3197, simple_loss=0.3822, pruned_loss=0.1286, over 28655.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3665, pruned_loss=0.1186, over 5653085.01 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3628, pruned_loss=0.1152, over 5674712.18 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3669, pruned_loss=0.1188, over 5644587.43 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:18:01,533 INFO [train.py:968] (0/2) Epoch 23, batch 28100, libri_loss[loss=0.2998, simple_loss=0.3725, pruned_loss=0.1135, over 28549.00 frames. ], tot_loss[loss=0.302, simple_loss=0.367, pruned_loss=0.1185, over 5651767.77 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3628, pruned_loss=0.1151, over 5678583.96 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3674, pruned_loss=0.1189, over 5640536.64 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:18:37,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.246e+03 1.755e+03 2.299e+03 3.175e+03 7.548e+03, threshold=4.598e+03, percent-clipped=13.0
+2023-03-12 01:18:48,405 INFO [train.py:968] (0/2) Epoch 23, batch 28150, giga_loss[loss=0.266, simple_loss=0.3393, pruned_loss=0.09629, over 28939.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3699, pruned_loss=0.1205, over 5661745.73 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3629, pruned_loss=0.1154, over 5682018.67 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3702, pruned_loss=0.1207, over 5649151.95 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:19:11,932 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1031566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:19:37,733 INFO [train.py:968] (0/2) Epoch 23, batch 28200, libri_loss[loss=0.2648, simple_loss=0.3358, pruned_loss=0.09694, over 29556.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1218, over 5662583.33 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3622, pruned_loss=0.1148, over 5690082.83 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3721, pruned_loss=0.1226, over 5644339.58 frames. ], batch size: 76, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:20:10,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.197e+03 1.791e+03 2.257e+03 2.878e+03 6.807e+03, threshold=4.513e+03, percent-clipped=7.0
+2023-03-12 01:20:22,549 INFO [train.py:968] (0/2) Epoch 23, batch 28250, giga_loss[loss=0.2838, simple_loss=0.3464, pruned_loss=0.1107, over 28819.00 frames. ], tot_loss[loss=0.309, simple_loss=0.372, pruned_loss=0.123, over 5645944.63 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3621, pruned_loss=0.1147, over 5677388.07 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3734, pruned_loss=0.124, over 5642479.49 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:21:14,665 INFO [train.py:968] (0/2) Epoch 23, batch 28300, giga_loss[loss=0.2736, simple_loss=0.3552, pruned_loss=0.09597, over 28895.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.373, pruned_loss=0.1227, over 5651402.68 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3621, pruned_loss=0.1147, over 5679974.61 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3744, pruned_loss=0.1238, over 5645413.39 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:21:35,479 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1031709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:21:38,577 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1031712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:21:56,421 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.239e+03 1.830e+03 2.495e+03 3.382e+03 9.214e+03, threshold=4.990e+03, percent-clipped=15.0
+2023-03-12 01:22:01,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5490, 1.7345, 1.3882, 1.7460], device='cuda:0'), covar=tensor([0.2660, 0.2742, 0.3134, 0.2361], device='cuda:0'), in_proj_covar=tensor([0.1532, 0.1108, 0.1356, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 01:22:06,300 INFO [train.py:968] (0/2) Epoch 23, batch 28350, giga_loss[loss=0.277, simple_loss=0.3536, pruned_loss=0.1002, over 29042.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.373, pruned_loss=0.1219, over 5651600.62 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3622, pruned_loss=0.1147, over 5680795.48 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3741, pruned_loss=0.1228, over 5645487.02 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:22:06,552 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1031741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:22:55,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1031789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:22:56,690 INFO [train.py:968] (0/2) Epoch 23, batch 28400, giga_loss[loss=0.2598, simple_loss=0.3396, pruned_loss=0.09002, over 28527.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3735, pruned_loss=0.1233, over 5626294.91 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3624, pruned_loss=0.1149, over 5673259.50 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3745, pruned_loss=0.124, over 5627174.14 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:23:34,854 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.202e+03 1.721e+03 2.203e+03 3.507e+03 1.065e+04, threshold=4.406e+03, percent-clipped=9.0
+2023-03-12 01:23:49,245 INFO [train.py:968] (0/2) Epoch 23, batch 28450, giga_loss[loss=0.2848, simple_loss=0.3509, pruned_loss=0.1093, over 28782.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3726, pruned_loss=0.1231, over 5615517.49 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3623, pruned_loss=0.115, over 5658955.99 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3736, pruned_loss=0.1238, over 5628144.54 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:24:50,689 INFO [train.py:968] (0/2) Epoch 23, batch 28500, giga_loss[loss=0.3487, simple_loss=0.4067, pruned_loss=0.1453, over 28740.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3718, pruned_loss=0.1237, over 5601789.16 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3619, pruned_loss=0.1149, over 5654271.81 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3732, pruned_loss=0.1245, over 5614480.67 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:25:26,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.697e+03 1.934e+03 2.473e+03 5.171e+03, threshold=3.869e+03, percent-clipped=3.0
+2023-03-12 01:25:29,788 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1031932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:25:32,903 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1031935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:25:38,024 INFO [train.py:968] (0/2) Epoch 23, batch 28550, giga_loss[loss=0.3344, simple_loss=0.3918, pruned_loss=0.1385, over 28605.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3715, pruned_loss=0.124, over 5626824.02 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3619, pruned_loss=0.1149, over 5655118.46 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3727, pruned_loss=0.1248, over 5635050.00 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:25:38,427 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1031941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:25:39,751 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3223, 1.1868, 1.1907, 1.4261], device='cuda:0'), covar=tensor([0.0782, 0.0375, 0.0352, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 01:26:01,319 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1031964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:26:21,416 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1031981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:26:28,811 INFO [train.py:968] (0/2) Epoch 23, batch 28600, giga_loss[loss=0.2687, simple_loss=0.3386, pruned_loss=0.09938, over 28694.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3707, pruned_loss=0.1233, over 5643649.67 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3621, pruned_loss=0.1149, over 5661717.00 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3718, pruned_loss=0.1242, over 5643728.79 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:26:37,123 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1032000.pt
+2023-03-12 01:27:07,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.164e+02 1.766e+03 2.226e+03 3.029e+03 9.580e+03, threshold=4.452e+03, percent-clipped=10.0
+2023-03-12 01:27:18,863 INFO [train.py:968] (0/2) Epoch 23, batch 28650, giga_loss[loss=0.322, simple_loss=0.3801, pruned_loss=0.1319, over 28785.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3702, pruned_loss=0.1232, over 5645829.11 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3618, pruned_loss=0.1148, over 5664029.54 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3714, pruned_loss=0.124, over 5643516.92 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:27:52,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5717, 1.7145, 1.8686, 1.5468], device='cuda:0'), covar=tensor([0.2788, 0.2688, 0.2590, 0.2655], device='cuda:0'), in_proj_covar=tensor([0.1998, 0.1949, 0.1868, 0.2014], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 01:27:53,559 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1032075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:28:10,271 INFO [train.py:968] (0/2) Epoch 23, batch 28700, giga_loss[loss=0.2805, simple_loss=0.3473, pruned_loss=0.1069, over 28905.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3707, pruned_loss=0.1235, over 5652237.51 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3619, pruned_loss=0.1148, over 5666171.15 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3717, pruned_loss=0.1243, over 5647999.49 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:28:30,105 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1032109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:28:32,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5707, 3.3796, 1.6311, 1.6471], device='cuda:0'), covar=tensor([0.0942, 0.0370, 0.0865, 0.1277], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0562, 0.0393, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 01:28:38,162 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5586, 1.6187, 1.7849, 1.3520], device='cuda:0'), covar=tensor([0.1600, 0.2348, 0.1347, 0.1616], device='cuda:0'), in_proj_covar=tensor([0.0905, 0.0708, 0.0953, 0.0851], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 01:28:48,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.793e+03 2.398e+03 2.993e+03 6.297e+03, threshold=4.795e+03, percent-clipped=7.0
+2023-03-12 01:29:03,006 INFO [train.py:968] (0/2) Epoch 23, batch 28750, giga_loss[loss=0.3724, simple_loss=0.4251, pruned_loss=0.1599, over 28327.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.373, pruned_loss=0.125, over 5653411.37 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.362, pruned_loss=0.1148, over 5667752.87 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3739, pruned_loss=0.1258, over 5648445.57 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:29:52,470 INFO [train.py:968] (0/2) Epoch 23, batch 28800, giga_loss[loss=0.3175, simple_loss=0.3806, pruned_loss=0.1272, over 28621.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.372, pruned_loss=0.1245, over 5664325.12 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3613, pruned_loss=0.1144, over 5671404.76 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3736, pruned_loss=0.1256, over 5657048.65 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 01:30:12,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1048, 1.3257, 3.6544, 3.1340], device='cuda:0'), covar=tensor([0.1764, 0.2693, 0.0509, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0661, 0.0980, 0.0932], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 01:30:18,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9158, 1.1335, 2.8153, 2.7520], device='cuda:0'), covar=tensor([0.1614, 0.2548, 0.0669, 0.1126], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0661, 0.0980, 0.0932], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 01:30:27,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.216e+03 1.742e+03 2.128e+03 2.717e+03 6.025e+03, threshold=4.255e+03, percent-clipped=3.0
+2023-03-12 01:30:36,643 INFO [train.py:968] (0/2) Epoch 23, batch 28850, giga_loss[loss=0.2906, simple_loss=0.3503, pruned_loss=0.1155, over 28566.00 frames. ], tot_loss[loss=0.312, simple_loss=0.373, pruned_loss=0.1254, over 5660955.93 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3615, pruned_loss=0.1146, over 5665147.35 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3744, pruned_loss=0.1265, over 5659745.32 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:30:44,807 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 01:31:25,759 INFO [train.py:968] (0/2) Epoch 23, batch 28900, giga_loss[loss=0.2952, simple_loss=0.3761, pruned_loss=0.1072, over 29075.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3731, pruned_loss=0.1251, over 5676260.91 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1144, over 5669207.12 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3746, pruned_loss=0.1263, over 5671771.46 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:31:31,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4641, 1.7051, 1.7448, 1.2637], device='cuda:0'), covar=tensor([0.1709, 0.2503, 0.1425, 0.1677], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0706, 0.0951, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 01:31:32,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3389, 2.9498, 1.5096, 1.4482], device='cuda:0'), covar=tensor([0.0944, 0.0372, 0.0879, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0562, 0.0393, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 01:31:53,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1032316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:32:06,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.330e+02 1.705e+03 2.102e+03 3.201e+03 9.654e+03, threshold=4.204e+03, percent-clipped=15.0
+2023-03-12 01:32:06,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2111, 2.5331, 1.2677, 1.2890], device='cuda:0'), covar=tensor([0.1025, 0.0436, 0.0943, 0.1448], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0563, 0.0394, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 01:32:15,642 INFO [train.py:968] (0/2) Epoch 23, batch 28950, libri_loss[loss=0.271, simple_loss=0.3443, pruned_loss=0.09883, over 29587.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.374, pruned_loss=0.1255, over 5671087.53 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3613, pruned_loss=0.1145, over 5673140.60 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3754, pruned_loss=0.1266, over 5663756.10 frames. ], batch size: 77, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:32:30,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1032356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:33:02,841 INFO [train.py:968] (0/2) Epoch 23, batch 29000, giga_loss[loss=0.2929, simple_loss=0.3574, pruned_loss=0.1142, over 28637.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3749, pruned_loss=0.126, over 5676252.86 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1147, over 5674226.54 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3759, pruned_loss=0.1269, over 5669493.01 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:33:40,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.043e+02 1.726e+03 2.411e+03 3.604e+03 1.607e+04, threshold=4.822e+03, percent-clipped=18.0
+2023-03-12 01:33:48,551 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-12 01:33:48,779 INFO [train.py:968] (0/2) Epoch 23, batch 29050, giga_loss[loss=0.3202, simple_loss=0.3829, pruned_loss=0.1287, over 28540.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.376, pruned_loss=0.1271, over 5668982.99 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1148, over 5671880.76 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3771, pruned_loss=0.1281, over 5665141.22 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:33:57,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1032450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:06,868 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1032459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:10,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1032462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:13,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 01:34:30,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1032484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:36,250 INFO [train.py:968] (0/2) Epoch 23, batch 29100, giga_loss[loss=0.3423, simple_loss=0.392, pruned_loss=0.1463, over 28593.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3771, pruned_loss=0.1282, over 5658889.52 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3619, pruned_loss=0.1149, over 5665039.73 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3782, pruned_loss=0.1292, over 5661988.67 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:34:36,827 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1032491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:41,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1032499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:46,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1032502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:35:09,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.673e+03 2.184e+03 3.424e+03 7.259e+03, threshold=4.369e+03, percent-clipped=9.0
+2023-03-12 01:35:11,688 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1032531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:35:21,208 INFO [train.py:968] (0/2) Epoch 23, batch 29150, libri_loss[loss=0.3379, simple_loss=0.3906, pruned_loss=0.1426, over 29536.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3762, pruned_loss=0.1269, over 5650315.23 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3619, pruned_loss=0.1151, over 5659887.45 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3776, pruned_loss=0.1279, over 5657706.61 frames. ], batch size: 84, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:35:44,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-12 01:36:12,565 INFO [train.py:968] (0/2) Epoch 23, batch 29200, giga_loss[loss=0.3542, simple_loss=0.4066, pruned_loss=0.1509, over 27937.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3759, pruned_loss=0.126, over 5644266.19 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3615, pruned_loss=0.1148, over 5667203.71 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3778, pruned_loss=0.1275, over 5643504.78 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 01:36:14,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1032593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:15,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1032594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:17,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1032596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:44,796 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1032625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:46,375 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1032627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:48,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1032630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:48,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.610e+03 1.949e+03 2.347e+03 7.408e+03, threshold=3.898e+03, percent-clipped=7.0
+2023-03-12 01:36:57,311 INFO [train.py:968] (0/2) Epoch 23, batch 29250, giga_loss[loss=0.3104, simple_loss=0.3797, pruned_loss=0.1205, over 28303.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3756, pruned_loss=0.1255, over 5638213.10 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.362, pruned_loss=0.1152, over 5659479.49 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3771, pruned_loss=0.1266, over 5644083.63 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:37:12,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1032659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:37:15,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7669, 4.6009, 4.3578, 2.2611], device='cuda:0'), covar=tensor([0.0562, 0.0738, 0.0797, 0.1882], device='cuda:0'), in_proj_covar=tensor([0.1274, 0.1179, 0.0998, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 01:37:42,776 INFO [train.py:968] (0/2) Epoch 23, batch 29300, giga_loss[loss=0.2632, simple_loss=0.3399, pruned_loss=0.09327, over 29108.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.373, pruned_loss=0.1232, over 5652671.93 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3617, pruned_loss=0.1149, over 5665940.00 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3749, pruned_loss=0.1247, over 5650994.05 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:37:53,029 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 01:37:57,895 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2575, 1.8284, 1.4816, 0.4742], device='cuda:0'), covar=tensor([0.4741, 0.3185, 0.4155, 0.6445], device='cuda:0'), in_proj_covar=tensor([0.1777, 0.1679, 0.1611, 0.1449], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 01:38:15,734 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1032729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:38:18,530 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.149e+03 1.538e+03 1.883e+03 2.388e+03 4.471e+03, threshold=3.766e+03, percent-clipped=1.0
+2023-03-12 01:38:24,948 INFO [train.py:968] (0/2) Epoch 23, batch 29350, giga_loss[loss=0.2941, simple_loss=0.3664, pruned_loss=0.1109, over 28726.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3725, pruned_loss=0.1229, over 5657846.60 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3622, pruned_loss=0.1152, over 5669820.94 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3739, pruned_loss=0.124, over 5652261.88 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:38:42,814 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1032760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:39:04,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-12 01:39:13,374 INFO [train.py:968] (0/2) Epoch 23, batch 29400, giga_loss[loss=0.3234, simple_loss=0.3754, pruned_loss=0.1357, over 28938.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3729, pruned_loss=0.1229, over 5656976.90 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5675055.12 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.374, pruned_loss=0.124, over 5647628.35 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:39:58,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.666e+03 2.123e+03 2.733e+03 5.113e+03, threshold=4.247e+03, percent-clipped=8.0
+2023-03-12 01:40:08,950 INFO [train.py:968] (0/2) Epoch 23, batch 29450, giga_loss[loss=0.2896, simple_loss=0.3608, pruned_loss=0.1092, over 29061.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3741, pruned_loss=0.124, over 5659563.03 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5675055.12 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.375, pruned_loss=0.1248, over 5652286.94 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:40:57,373 INFO [train.py:968] (0/2) Epoch 23, batch 29500, giga_loss[loss=0.3148, simple_loss=0.3728, pruned_loss=0.1285, over 28835.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.373, pruned_loss=0.1242, over 5654472.66 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3622, pruned_loss=0.1152, over 5669959.66 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3744, pruned_loss=0.1252, over 5652090.47 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:40:59,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-12 01:41:05,421 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-03-12 01:41:37,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.643e+03 2.124e+03 3.102e+03 9.489e+03, threshold=4.248e+03, percent-clipped=13.0
+2023-03-12 01:41:46,037 INFO [train.py:968] (0/2) Epoch 23, batch 29550, giga_loss[loss=0.3237, simple_loss=0.3854, pruned_loss=0.131, over 28055.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3747, pruned_loss=0.1255, over 5664253.38 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3629, pruned_loss=0.1158, over 5671353.46 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3753, pruned_loss=0.126, over 5661057.20 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:42:14,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1032969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:42:32,971 INFO [train.py:968] (0/2) Epoch 23, batch 29600, giga_loss[loss=0.3389, simple_loss=0.3978, pruned_loss=0.14, over 28445.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3762, pruned_loss=0.1267, over 5663811.33 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3625, pruned_loss=0.1154, over 5674909.46 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3772, pruned_loss=0.1275, over 5657819.48 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:43:14,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.742e+03 2.216e+03 3.183e+03 1.049e+04, threshold=4.431e+03, percent-clipped=8.0
+2023-03-12 01:43:20,531 INFO [train.py:968] (0/2) Epoch 23, batch 29650, libri_loss[loss=0.25, simple_loss=0.3185, pruned_loss=0.09075, over 29639.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3764, pruned_loss=0.1272, over 5651140.47 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3625, pruned_loss=0.1155, over 5671974.97 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3777, pruned_loss=0.1282, over 5647076.33 frames. ], batch size: 73, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:43:51,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2063, 1.4357, 1.4627, 1.0819], device='cuda:0'), covar=tensor([0.1600, 0.2596, 0.1361, 0.1626], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0707, 0.0952, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 01:44:06,987 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-12 01:44:10,144 INFO [train.py:968] (0/2) Epoch 23, batch 29700, giga_loss[loss=0.4011, simple_loss=0.4281, pruned_loss=0.1871, over 26670.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3767, pruned_loss=0.1274, over 5646123.72 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3627, pruned_loss=0.1157, over 5675116.61 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3777, pruned_loss=0.1282, over 5639968.89 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:44:12,289 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 01:44:22,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1033104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:44:27,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4244, 4.2689, 4.0739, 2.1991], device='cuda:0'), covar=tensor([0.0516, 0.0625, 0.0672, 0.2058], device='cuda:0'), in_proj_covar=tensor([0.1272, 0.1178, 0.1000, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 01:44:29,585 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1033112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:44:32,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1033115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:44:33,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3542, 1.6641, 1.6328, 1.1830], device='cuda:0'), covar=tensor([0.1641, 0.2669, 0.1446, 0.1728], device='cuda:0'), in_proj_covar=tensor([0.0905, 0.0707, 0.0952, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 01:44:46,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.583e+03 1.978e+03 2.554e+03 7.416e+03, threshold=3.956e+03, percent-clipped=5.0
+2023-03-12 01:44:50,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1033135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:44:54,778 INFO [train.py:968] (0/2) Epoch 23, batch 29750, giga_loss[loss=0.2672, simple_loss=0.3465, pruned_loss=0.09399, over 29051.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.377, pruned_loss=0.1267, over 5656485.05 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3632, pruned_loss=0.1159, over 5677843.88 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3777, pruned_loss=0.1274, over 5648515.35 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:44:59,708 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1033144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:45:46,093 INFO [train.py:968] (0/2) Epoch 23, batch 29800, giga_loss[loss=0.3427, simple_loss=0.3767, pruned_loss=0.1543, over 23676.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3765, pruned_loss=0.1262, over 5654656.29 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3635, pruned_loss=0.1161, over 5679325.34 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3771, pruned_loss=0.1268, over 5646697.70 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:46:27,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.856e+03 2.627e+03 4.692e+03 8.312e+03, threshold=5.255e+03, percent-clipped=31.0
+2023-03-12 01:46:34,970 INFO [train.py:968] (0/2) Epoch 23, batch 29850, giga_loss[loss=0.2581, simple_loss=0.3319, pruned_loss=0.0921, over 28914.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3746, pruned_loss=0.1244, over 5667687.56 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3638, pruned_loss=0.1161, over 5680619.08 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3749, pruned_loss=0.1249, over 5660064.61 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:46:43,045 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1033247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:46:45,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1033250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:47:06,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8790, 5.6719, 5.4060, 2.9620], device='cuda:0'), covar=tensor([0.0560, 0.0765, 0.0928, 0.1618], device='cuda:0'), in_proj_covar=tensor([0.1272, 0.1180, 0.0999, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 01:47:12,643 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1033278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:47:13,529 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1033279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:47:15,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1033281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:47:27,588 INFO [train.py:968] (0/2) Epoch 23, batch 29900, giga_loss[loss=0.3146, simple_loss=0.374, pruned_loss=0.1276, over 28799.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3724, pruned_loss=0.1232, over 5672993.14 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3635, pruned_loss=0.116, over 5682441.13 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.373, pruned_loss=0.1238, over 5665299.74 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:47:44,683 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1033310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:48:07,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.809e+03 2.396e+03 3.552e+03 1.112e+04, threshold=4.792e+03, percent-clipped=8.0
+2023-03-12 01:48:15,558 INFO [train.py:968] (0/2) Epoch 23, batch 29950, giga_loss[loss=0.2831, simple_loss=0.3411, pruned_loss=0.1126, over 28859.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3669, pruned_loss=0.1202, over 5664908.26 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3627, pruned_loss=0.1154, over 5686496.73 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3682, pruned_loss=0.1213, over 5654928.86 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:48:58,456 INFO [train.py:968] (0/2) Epoch 23, batch 30000, giga_loss[loss=0.2808, simple_loss=0.3441, pruned_loss=0.1088, over 28436.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.364, pruned_loss=0.1194, over 5656044.78 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3626, pruned_loss=0.1154, over 5680122.27 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3653, pruned_loss=0.1205, over 5652485.58 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:48:58,460 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 01:49:07,292 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.205, simple_loss=0.3136, pruned_loss=0.0482, over 944034.00 frames. 
+2023-03-12 01:49:07,292 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 01:49:44,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6973, 1.8629, 1.3323, 1.4238], device='cuda:0'), covar=tensor([0.0969, 0.0621, 0.1067, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0450, 0.0522, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 01:49:52,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.757e+03 2.291e+03 3.125e+03 6.617e+03, threshold=4.582e+03, percent-clipped=4.0
+2023-03-12 01:49:58,995 INFO [train.py:968] (0/2) Epoch 23, batch 30050, giga_loss[loss=0.3122, simple_loss=0.362, pruned_loss=0.1312, over 28756.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3635, pruned_loss=0.1196, over 5650766.04 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3629, pruned_loss=0.1157, over 5682151.48 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3643, pruned_loss=0.1203, over 5645269.65 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:50:48,002 INFO [train.py:968] (0/2) Epoch 23, batch 30100, giga_loss[loss=0.3266, simple_loss=0.3918, pruned_loss=0.1307, over 28866.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3632, pruned_loss=0.119, over 5638678.03 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3629, pruned_loss=0.1157, over 5676258.31 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3638, pruned_loss=0.1196, over 5638941.66 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:51:24,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5460, 1.5603, 1.2575, 1.2113], device='cuda:0'), covar=tensor([0.0750, 0.0368, 0.0822, 0.1085], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0449, 0.0521, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 01:51:29,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.672e+03 2.116e+03 2.870e+03 5.666e+03, threshold=4.231e+03, percent-clipped=5.0
+2023-03-12 01:51:37,088 INFO [train.py:968] (0/2) Epoch 23, batch 30150, giga_loss[loss=0.2738, simple_loss=0.3324, pruned_loss=0.1076, over 24132.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3612, pruned_loss=0.1156, over 5634949.86 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.363, pruned_loss=0.1157, over 5674239.21 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3617, pruned_loss=0.1161, over 5636219.34 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:51:54,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3295, 1.5784, 1.4925, 1.2822], device='cuda:0'), covar=tensor([0.2600, 0.2098, 0.1635, 0.2194], device='cuda:0'), in_proj_covar=tensor([0.1986, 0.1938, 0.1855, 0.1999], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 01:52:00,851 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1033565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:52:23,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9389, 2.4578, 2.3580, 1.7954], device='cuda:0'), covar=tensor([0.3080, 0.2036, 0.1946, 0.2722], device='cuda:0'), in_proj_covar=tensor([0.1985, 0.1936, 0.1852, 0.1998], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 01:52:32,125 INFO [train.py:968] (0/2) Epoch 23, batch 30200, giga_loss[loss=0.2832, simple_loss=0.3545, pruned_loss=0.1059, over 28924.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3591, pruned_loss=0.1121, over 5643690.75 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3629, pruned_loss=0.1157, over 5679342.35 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3595, pruned_loss=0.1125, over 5639084.18 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:53:12,761 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.326e+02 1.465e+03 1.949e+03 2.646e+03 6.973e+03, threshold=3.899e+03, percent-clipped=7.0
+2023-03-12 01:53:21,413 INFO [train.py:968] (0/2) Epoch 23, batch 30250, giga_loss[loss=0.2869, simple_loss=0.3566, pruned_loss=0.1086, over 28979.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3567, pruned_loss=0.109, over 5657612.37 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3624, pruned_loss=0.1156, over 5683138.80 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3572, pruned_loss=0.1092, over 5649846.79 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:53:25,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3505, 1.7951, 1.3558, 0.7848], device='cuda:0'), covar=tensor([0.5917, 0.3017, 0.3094, 0.6427], device='cuda:0'), in_proj_covar=tensor([0.1768, 0.1666, 0.1608, 0.1443], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 01:54:11,803 INFO [train.py:968] (0/2) Epoch 23, batch 30300, giga_loss[loss=0.2628, simple_loss=0.3466, pruned_loss=0.08951, over 29011.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3534, pruned_loss=0.1059, over 5660372.26 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3617, pruned_loss=0.1153, over 5687743.45 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3544, pruned_loss=0.1063, over 5649473.20 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:54:15,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4859, 2.6192, 1.5447, 1.6382], device='cuda:0'), covar=tensor([0.0815, 0.0325, 0.0808, 0.1136], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0561, 0.0392, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 01:54:31,923 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1033709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:54:43,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7122, 4.5216, 4.3125, 1.9643], device='cuda:0'), covar=tensor([0.0696, 0.0849, 0.1065, 0.2160], device='cuda:0'), in_proj_covar=tensor([0.1261, 0.1171, 0.0991, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 01:54:52,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.770e+02 1.317e+03 1.754e+03 2.338e+03 5.593e+03, threshold=3.509e+03, percent-clipped=5.0
+2023-03-12 01:55:00,499 INFO [train.py:968] (0/2) Epoch 23, batch 30350, giga_loss[loss=0.2903, simple_loss=0.3746, pruned_loss=0.103, over 28700.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3513, pruned_loss=0.103, over 5656699.31 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3615, pruned_loss=0.1153, over 5680002.92 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3521, pruned_loss=0.103, over 5653626.09 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:55:36,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5151, 1.6886, 1.7651, 1.3253], device='cuda:0'), covar=tensor([0.2031, 0.2841, 0.1675, 0.2056], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0702, 0.0950, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 01:55:59,428 INFO [train.py:968] (0/2) Epoch 23, batch 30400, giga_loss[loss=0.2578, simple_loss=0.3412, pruned_loss=0.08719, over 28836.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3513, pruned_loss=0.1011, over 5667063.34 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3613, pruned_loss=0.1153, over 5681483.30 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3519, pruned_loss=0.101, over 5663163.78 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:56:24,579 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-12 01:56:43,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.827e+02 1.430e+03 1.919e+03 2.653e+03 7.156e+03, threshold=3.838e+03, percent-clipped=10.0
+2023-03-12 01:56:51,712 INFO [train.py:968] (0/2) Epoch 23, batch 30450, giga_loss[loss=0.2465, simple_loss=0.3321, pruned_loss=0.08042, over 28865.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3514, pruned_loss=0.1014, over 5669683.66 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.361, pruned_loss=0.1153, over 5689016.65 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3519, pruned_loss=0.1009, over 5659234.73 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:57:42,871 INFO [train.py:968] (0/2) Epoch 23, batch 30500, giga_loss[loss=0.256, simple_loss=0.3304, pruned_loss=0.09084, over 27612.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3488, pruned_loss=0.09963, over 5666022.34 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3605, pruned_loss=0.1151, over 5683698.30 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3493, pruned_loss=0.09888, over 5662606.40 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:58:25,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.149e+02 1.613e+03 2.044e+03 2.782e+03 7.870e+03, threshold=4.089e+03, percent-clipped=6.0
+2023-03-12 01:58:33,051 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1033940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:58:33,397 INFO [train.py:968] (0/2) Epoch 23, batch 30550, giga_loss[loss=0.246, simple_loss=0.3319, pruned_loss=0.08006, over 29006.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3457, pruned_loss=0.09747, over 5665536.28 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3602, pruned_loss=0.1149, over 5688124.11 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3461, pruned_loss=0.09675, over 5658294.64 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:59:25,319 INFO [train.py:968] (0/2) Epoch 23, batch 30600, giga_loss[loss=0.2835, simple_loss=0.3601, pruned_loss=0.1035, over 29003.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.346, pruned_loss=0.09786, over 5655102.56 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3601, pruned_loss=0.115, over 5680111.84 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3462, pruned_loss=0.09694, over 5656683.16 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:59:32,978 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1034000.pt
+2023-03-12 02:00:02,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.049e+02 1.477e+03 1.797e+03 2.590e+03 1.242e+04, threshold=3.594e+03, percent-clipped=7.0
+2023-03-12 02:00:04,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5123, 1.6238, 1.7207, 1.3355], device='cuda:0'), covar=tensor([0.1788, 0.2723, 0.1504, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.0903, 0.0700, 0.0948, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 02:00:08,394 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1034039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:00:10,328 INFO [train.py:968] (0/2) Epoch 23, batch 30650, giga_loss[loss=0.2327, simple_loss=0.3183, pruned_loss=0.07349, over 28899.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3458, pruned_loss=0.09818, over 5647043.06 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3597, pruned_loss=0.1153, over 5667810.96 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3459, pruned_loss=0.09662, over 5658784.48 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:00:17,343 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4355, 4.2868, 4.0943, 2.3460], device='cuda:0'), covar=tensor([0.0567, 0.0712, 0.0779, 0.1808], device='cuda:0'), in_proj_covar=tensor([0.1254, 0.1162, 0.0982, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 02:00:40,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9231, 1.2510, 5.1675, 3.6910], device='cuda:0'), covar=tensor([0.1566, 0.3037, 0.0464, 0.0959], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0659, 0.0977, 0.0927], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 02:00:50,778 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1034083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:00:51,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1034084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:00:52,764 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1034086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:00:58,746 INFO [train.py:968] (0/2) Epoch 23, batch 30700, giga_loss[loss=0.2539, simple_loss=0.34, pruned_loss=0.08389, over 28993.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3429, pruned_loss=0.09575, over 5644289.73 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.359, pruned_loss=0.115, over 5664044.48 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3431, pruned_loss=0.09422, over 5656988.76 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:01:22,549 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1034115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:01:43,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.437e+02 1.327e+03 1.732e+03 2.312e+03 6.369e+03, threshold=3.463e+03, percent-clipped=7.0
+2023-03-12 02:01:48,900 INFO [train.py:968] (0/2) Epoch 23, batch 30750, libri_loss[loss=0.3397, simple_loss=0.3828, pruned_loss=0.1483, over 18857.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3391, pruned_loss=0.09332, over 5634162.16 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3582, pruned_loss=0.1149, over 5648867.59 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3395, pruned_loss=0.09163, over 5659448.29 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 02:02:41,584 INFO [train.py:968] (0/2) Epoch 23, batch 30800, giga_loss[loss=0.2665, simple_loss=0.3336, pruned_loss=0.09971, over 27568.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3366, pruned_loss=0.09213, over 5637522.36 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3581, pruned_loss=0.115, over 5644316.95 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3366, pruned_loss=0.09035, over 5661192.04 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:02:53,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6561, 2.0264, 1.4144, 1.5265], device='cuda:0'), covar=tensor([0.0956, 0.0513, 0.0967, 0.1046], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0447, 0.0519, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 02:03:16,246 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1034227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:03:21,065 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1034230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:03:25,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.224e+02 1.624e+03 2.096e+03 3.558e+03 8.489e+03, threshold=4.192e+03, percent-clipped=24.0
+2023-03-12 02:03:32,338 INFO [train.py:968] (0/2) Epoch 23, batch 30850, giga_loss[loss=0.2164, simple_loss=0.2991, pruned_loss=0.06687, over 29019.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3354, pruned_loss=0.09193, over 5645744.51 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3576, pruned_loss=0.1146, over 5645953.10 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3355, pruned_loss=0.09038, over 5663246.98 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:03:52,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1034259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:03:55,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2099, 1.2737, 3.5227, 2.9953], device='cuda:0'), covar=tensor([0.1666, 0.2848, 0.0494, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0771, 0.0655, 0.0969, 0.0921], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 02:03:57,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6731, 1.9996, 1.6748, 1.6788], device='cuda:0'), covar=tensor([0.2319, 0.1965, 0.2113, 0.1920], device='cuda:0'), in_proj_covar=tensor([0.1538, 0.1109, 0.1362, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 02:04:00,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3117, 0.8494, 1.0147, 1.3885], device='cuda:0'), covar=tensor([0.0751, 0.0349, 0.0357, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 02:04:28,105 INFO [train.py:968] (0/2) Epoch 23, batch 30900, giga_loss[loss=0.2425, simple_loss=0.3272, pruned_loss=0.07893, over 28510.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3358, pruned_loss=0.09262, over 5625456.20 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3572, pruned_loss=0.1145, over 5638940.57 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3359, pruned_loss=0.09121, over 5645352.05 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:05:13,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.481e+02 1.526e+03 2.149e+03 2.791e+03 8.197e+03, threshold=4.297e+03, percent-clipped=8.0
+2023-03-12 02:05:20,970 INFO [train.py:968] (0/2) Epoch 23, batch 30950, giga_loss[loss=0.289, simple_loss=0.3517, pruned_loss=0.1132, over 26650.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3389, pruned_loss=0.0943, over 5629626.20 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3571, pruned_loss=0.1146, over 5644393.57 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3384, pruned_loss=0.09236, over 5640361.82 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:06:19,437 INFO [train.py:968] (0/2) Epoch 23, batch 31000, libri_loss[loss=0.2526, simple_loss=0.3154, pruned_loss=0.09491, over 29655.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3393, pruned_loss=0.09319, over 5626942.88 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3567, pruned_loss=0.1144, over 5649894.89 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3389, pruned_loss=0.09148, over 5630283.65 frames. ], batch size: 73, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:06:49,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1034414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:07:14,889 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 02:07:17,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.471e+03 1.840e+03 2.628e+03 8.174e+03, threshold=3.679e+03, percent-clipped=8.0
+2023-03-12 02:07:23,828 INFO [train.py:968] (0/2) Epoch 23, batch 31050, libri_loss[loss=0.241, simple_loss=0.3026, pruned_loss=0.08969, over 29672.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.339, pruned_loss=0.09308, over 5628757.51 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.356, pruned_loss=0.114, over 5654276.57 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3391, pruned_loss=0.09165, over 5626817.17 frames. ], batch size: 69, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:08:29,441 INFO [train.py:968] (0/2) Epoch 23, batch 31100, giga_loss[loss=0.2334, simple_loss=0.3221, pruned_loss=0.07233, over 28639.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3373, pruned_loss=0.09177, over 5623794.29 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3562, pruned_loss=0.1143, over 5638149.12 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3369, pruned_loss=0.09015, over 5637607.48 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:09:22,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.205e+02 1.324e+03 1.723e+03 2.470e+03 5.141e+03, threshold=3.446e+03, percent-clipped=7.0
+2023-03-12 02:09:29,072 INFO [train.py:968] (0/2) Epoch 23, batch 31150, giga_loss[loss=0.2774, simple_loss=0.3595, pruned_loss=0.09763, over 28385.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3367, pruned_loss=0.09069, over 5615268.00 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3559, pruned_loss=0.1143, over 5636143.10 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3361, pruned_loss=0.08864, over 5628122.72 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:09:52,651 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1034557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:09:54,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1034560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:10:32,198 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1034589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:10:34,023 INFO [train.py:968] (0/2) Epoch 23, batch 31200, giga_loss[loss=0.2673, simple_loss=0.3471, pruned_loss=0.09373, over 28228.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3343, pruned_loss=0.08833, over 5626292.39 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3563, pruned_loss=0.1147, over 5636196.58 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3331, pruned_loss=0.086, over 5636152.90 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:11:11,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6847, 1.8576, 1.4009, 1.4633], device='cuda:0'), covar=tensor([0.1068, 0.0718, 0.1006, 0.1307], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0447, 0.0520, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 02:11:33,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.466e+02 1.342e+03 1.788e+03 2.415e+03 6.888e+03, threshold=3.576e+03, percent-clipped=8.0
+2023-03-12 02:11:38,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5494, 1.7920, 1.2347, 1.3815], device='cuda:0'), covar=tensor([0.0996, 0.0556, 0.0997, 0.1142], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0446, 0.0520, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 02:11:38,822 INFO [train.py:968] (0/2) Epoch 23, batch 31250, giga_loss[loss=0.2799, simple_loss=0.3582, pruned_loss=0.1008, over 28530.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3328, pruned_loss=0.08887, over 5647836.47 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3565, pruned_loss=0.1151, over 5644044.50 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3311, pruned_loss=0.08593, over 5648543.18 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:12:25,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3299, 1.6354, 1.3007, 1.0497], device='cuda:0'), covar=tensor([0.2619, 0.2598, 0.3031, 0.2336], device='cuda:0'), in_proj_covar=tensor([0.1534, 0.1107, 0.1357, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 02:12:39,050 INFO [train.py:968] (0/2) Epoch 23, batch 31300, giga_loss[loss=0.2375, simple_loss=0.3213, pruned_loss=0.07682, over 28569.00 frames. ], tot_loss[loss=0.256, simple_loss=0.333, pruned_loss=0.08948, over 5661545.92 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3565, pruned_loss=0.1153, over 5648676.65 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3311, pruned_loss=0.08638, over 5658122.61 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:12:42,811 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1034694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:12:48,756 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0619, 1.4002, 1.3276, 1.0140], device='cuda:0'), covar=tensor([0.1417, 0.2125, 0.1208, 0.1628], device='cuda:0'), in_proj_covar=tensor([0.0905, 0.0700, 0.0950, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 02:13:35,757 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.198e+02 1.464e+03 2.087e+03 2.500e+03 5.316e+03, threshold=4.175e+03, percent-clipped=7.0
+2023-03-12 02:13:41,357 INFO [train.py:968] (0/2) Epoch 23, batch 31350, giga_loss[loss=0.2194, simple_loss=0.2888, pruned_loss=0.07501, over 24312.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3331, pruned_loss=0.08909, over 5652421.91 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3563, pruned_loss=0.1152, over 5639971.33 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3314, pruned_loss=0.08623, over 5657728.40 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:13:43,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5977, 1.8191, 1.3055, 1.3885], device='cuda:0'), covar=tensor([0.0971, 0.0557, 0.0927, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0447, 0.0521, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 02:14:40,624 INFO [train.py:968] (0/2) Epoch 23, batch 31400, libri_loss[loss=0.2622, simple_loss=0.3378, pruned_loss=0.09326, over 29639.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3355, pruned_loss=0.08989, over 5650146.54 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3562, pruned_loss=0.1152, over 5646648.08 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3334, pruned_loss=0.08665, over 5648334.33 frames. ], batch size: 88, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:14:52,352 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1580, 2.5579, 2.3017, 2.0893], device='cuda:0'), covar=tensor([0.2129, 0.2194, 0.1946, 0.2143], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0744, 0.0710, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 02:14:57,572 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1034805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:15:34,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.963e+02 1.467e+03 2.015e+03 2.930e+03 7.172e+03, threshold=4.029e+03, percent-clipped=11.0
+2023-03-12 02:15:39,722 INFO [train.py:968] (0/2) Epoch 23, batch 31450, giga_loss[loss=0.2243, simple_loss=0.3028, pruned_loss=0.0729, over 27554.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3336, pruned_loss=0.08867, over 5666235.01 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3557, pruned_loss=0.1151, over 5656507.42 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3316, pruned_loss=0.08526, over 5656258.03 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:16:49,967 INFO [train.py:968] (0/2) Epoch 23, batch 31500, giga_loss[loss=0.2463, simple_loss=0.3229, pruned_loss=0.0848, over 27538.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3324, pruned_loss=0.08777, over 5678827.11 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3557, pruned_loss=0.115, over 5659320.55 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3303, pruned_loss=0.08452, over 5668533.60 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:17:04,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3224, 2.0478, 1.4576, 0.6248], device='cuda:0'), covar=tensor([0.4795, 0.2752, 0.3915, 0.5114], device='cuda:0'), in_proj_covar=tensor([0.1779, 0.1670, 0.1608, 0.1448], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 02:17:40,555 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5042, 1.7776, 1.7151, 1.3007], device='cuda:0'), covar=tensor([0.1712, 0.2718, 0.1473, 0.1802], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0701, 0.0952, 0.0853], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 02:17:47,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.274e+02 1.360e+03 1.709e+03 2.269e+03 6.183e+03, threshold=3.418e+03, percent-clipped=7.0
+2023-03-12 02:17:57,270 INFO [train.py:968] (0/2) Epoch 23, batch 31550, giga_loss[loss=0.1952, simple_loss=0.276, pruned_loss=0.0572, over 24423.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3343, pruned_loss=0.08881, over 5674613.91 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3551, pruned_loss=0.1147, over 5666685.55 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3325, pruned_loss=0.08572, over 5660241.27 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:19:03,663 INFO [train.py:968] (0/2) Epoch 23, batch 31600, giga_loss[loss=0.2473, simple_loss=0.3393, pruned_loss=0.07761, over 27561.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3378, pruned_loss=0.08833, over 5678531.43 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3546, pruned_loss=0.1144, over 5671313.44 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3365, pruned_loss=0.08561, over 5662937.18 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:19:55,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.042e+02 1.456e+03 2.011e+03 2.793e+03 1.133e+04, threshold=4.022e+03, percent-clipped=15.0
+2023-03-12 02:20:00,706 INFO [train.py:968] (0/2) Epoch 23, batch 31650, giga_loss[loss=0.2405, simple_loss=0.3433, pruned_loss=0.06886, over 28922.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3391, pruned_loss=0.08822, over 5673939.25 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3543, pruned_loss=0.1142, over 5680886.35 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3377, pruned_loss=0.08508, over 5652867.25 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:20:17,031 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-12 02:20:37,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1035069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:20:51,795 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:21:04,804 INFO [train.py:968] (0/2) Epoch 23, batch 31700, giga_loss[loss=0.2774, simple_loss=0.3588, pruned_loss=0.09794, over 28487.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3389, pruned_loss=0.08638, over 5672355.20 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3539, pruned_loss=0.114, over 5680114.81 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.338, pruned_loss=0.08376, over 5656332.71 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:21:20,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2706, 1.6691, 1.6465, 1.4732], device='cuda:0'), covar=tensor([0.2181, 0.2192, 0.2233, 0.2077], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0744, 0.0708, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 02:22:07,067 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.974e+02 1.497e+03 1.882e+03 2.714e+03 8.134e+03, threshold=3.765e+03, percent-clipped=10.0
+2023-03-12 02:22:11,392 INFO [train.py:968] (0/2) Epoch 23, batch 31750, giga_loss[loss=0.2248, simple_loss=0.3198, pruned_loss=0.06486, over 28148.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3389, pruned_loss=0.08617, over 5683243.12 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3539, pruned_loss=0.114, over 5682627.92 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3379, pruned_loss=0.08381, over 5668209.21 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:22:34,949 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-12 02:22:54,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4655, 1.5550, 1.6915, 1.3097], device='cuda:0'), covar=tensor([0.1864, 0.2797, 0.1570, 0.1884], device='cuda:0'), in_proj_covar=tensor([0.0906, 0.0700, 0.0952, 0.0853], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 02:23:09,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1035180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:23:21,123 INFO [train.py:968] (0/2) Epoch 23, batch 31800, giga_loss[loss=0.2437, simple_loss=0.3233, pruned_loss=0.08204, over 28613.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3377, pruned_loss=0.08692, over 5692733.22 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3532, pruned_loss=0.1136, over 5688143.98 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3372, pruned_loss=0.08471, over 5675733.55 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:23:42,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4030, 3.2802, 1.4273, 1.5473], device='cuda:0'), covar=tensor([0.0974, 0.0325, 0.0927, 0.1326], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0558, 0.0393, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 02:23:47,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3790, 1.4325, 1.3501, 1.5443], device='cuda:0'), covar=tensor([0.0773, 0.0344, 0.0338, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 02:23:51,675 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1035212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:23:56,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1035215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:24:20,559 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-12 02:24:27,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.344e+02 1.515e+03 2.014e+03 2.830e+03 6.508e+03, threshold=4.028e+03, percent-clipped=12.0
+2023-03-12 02:24:36,341 INFO [train.py:968] (0/2) Epoch 23, batch 31850, giga_loss[loss=0.2593, simple_loss=0.341, pruned_loss=0.08877, over 29073.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3383, pruned_loss=0.08828, over 5688406.86 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3526, pruned_loss=0.1131, over 5690457.28 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.338, pruned_loss=0.08625, over 5672564.82 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:24:40,280 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1035244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:25:51,846 INFO [train.py:968] (0/2) Epoch 23, batch 31900, giga_loss[loss=0.2129, simple_loss=0.3005, pruned_loss=0.0627, over 28459.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3361, pruned_loss=0.08782, over 5684928.52 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3528, pruned_loss=0.1133, over 5685931.47 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3354, pruned_loss=0.08543, over 5676237.41 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:26:32,245 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-12 02:26:32,897 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1035323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:26:37,147 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1035326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:26:49,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.172e+02 1.517e+03 1.903e+03 2.890e+03 8.044e+03, threshold=3.806e+03, percent-clipped=10.0
+2023-03-12 02:26:54,835 INFO [train.py:968] (0/2) Epoch 23, batch 31950, giga_loss[loss=0.301, simple_loss=0.3637, pruned_loss=0.1191, over 26886.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3332, pruned_loss=0.08653, over 5685209.01 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.352, pruned_loss=0.1129, over 5694557.53 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3325, pruned_loss=0.08383, over 5670142.31 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:27:12,821 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:27:17,219 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1035355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:28:07,785 INFO [train.py:968] (0/2) Epoch 23, batch 32000, giga_loss[loss=0.1995, simple_loss=0.2873, pruned_loss=0.05584, over 28797.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3299, pruned_loss=0.08457, over 5681044.93 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.352, pruned_loss=0.113, over 5688164.11 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3292, pruned_loss=0.08208, over 5674174.09 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:29:13,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.980e+02 1.330e+03 1.701e+03 2.605e+03 4.934e+03, threshold=3.402e+03, percent-clipped=8.0
+2023-03-12 02:29:18,552 INFO [train.py:968] (0/2) Epoch 23, batch 32050, giga_loss[loss=0.3127, simple_loss=0.3883, pruned_loss=0.1185, over 28506.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3324, pruned_loss=0.08642, over 5677330.15 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3522, pruned_loss=0.1131, over 5679331.08 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3315, pruned_loss=0.0841, over 5680285.63 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:29:40,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1035456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:29:45,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8159, 2.0155, 1.4064, 1.5310], device='cuda:0'), covar=tensor([0.0994, 0.0579, 0.1019, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0446, 0.0520, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 02:30:23,874 INFO [train.py:968] (0/2) Epoch 23, batch 32100, giga_loss[loss=0.2877, simple_loss=0.3702, pruned_loss=0.1026, over 28630.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3363, pruned_loss=0.08858, over 5681843.11 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3523, pruned_loss=0.1131, over 5681357.50 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3352, pruned_loss=0.08638, over 5682186.98 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:31:29,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.532e+02 1.482e+03 1.944e+03 2.750e+03 1.051e+04, threshold=3.887e+03, percent-clipped=14.0
+2023-03-12 02:31:31,850 INFO [train.py:968] (0/2) Epoch 23, batch 32150, libri_loss[loss=0.2666, simple_loss=0.32, pruned_loss=0.1066, over 29661.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.335, pruned_loss=0.08922, over 5683214.42 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3521, pruned_loss=0.1132, over 5682107.98 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3339, pruned_loss=0.08699, over 5682670.30 frames. ], batch size: 73, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:31:39,862 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:32:37,654 INFO [train.py:968] (0/2) Epoch 23, batch 32200, giga_loss[loss=0.27, simple_loss=0.3427, pruned_loss=0.09865, over 28583.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3356, pruned_loss=0.09048, over 5679168.59 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3521, pruned_loss=0.1133, over 5684056.60 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3346, pruned_loss=0.08841, over 5677009.43 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 02:32:48,431 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1035599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:32:53,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1035602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:33:29,923 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1035631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:33:37,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2945, 1.6187, 1.5477, 1.4005], device='cuda:0'), covar=tensor([0.1928, 0.2096, 0.2068, 0.2091], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0740, 0.0706, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 02:33:41,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.487e+02 1.664e+03 2.097e+03 3.099e+03 7.009e+03, threshold=4.193e+03, percent-clipped=10.0
+2023-03-12 02:33:44,680 INFO [train.py:968] (0/2) Epoch 23, batch 32250, giga_loss[loss=0.2359, simple_loss=0.2983, pruned_loss=0.08677, over 24960.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3354, pruned_loss=0.0905, over 5679191.83 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3516, pruned_loss=0.1132, over 5688616.09 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3345, pruned_loss=0.08803, over 5673462.52 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 02:34:42,979 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:34:58,837 INFO [train.py:968] (0/2) Epoch 23, batch 32300, giga_loss[loss=0.2478, simple_loss=0.3341, pruned_loss=0.08072, over 28996.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3378, pruned_loss=0.09062, over 5667723.15 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3515, pruned_loss=0.1132, over 5680815.17 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3369, pruned_loss=0.0883, over 5670768.64 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 02:35:57,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1035728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:36:09,352 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.624e+03 2.155e+03 2.910e+03 1.034e+04, threshold=4.311e+03, percent-clipped=9.0
+2023-03-12 02:36:14,953 INFO [train.py:968] (0/2) Epoch 23, batch 32350, giga_loss[loss=0.2435, simple_loss=0.3157, pruned_loss=0.0856, over 24373.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3382, pruned_loss=0.09095, over 5653152.52 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3516, pruned_loss=0.1133, over 5676291.89 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.337, pruned_loss=0.08834, over 5659400.42 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 02:37:20,776 INFO [train.py:968] (0/2) Epoch 23, batch 32400, giga_loss[loss=0.2492, simple_loss=0.3256, pruned_loss=0.08638, over 29055.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3354, pruned_loss=0.09055, over 5666928.73 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3514, pruned_loss=0.1134, over 5678261.53 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3341, pruned_loss=0.08768, over 5670085.54 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:37:48,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3208, 1.1631, 3.3376, 3.1446], device='cuda:0'), covar=tensor([0.1820, 0.3229, 0.0943, 0.2242], device='cuda:0'), in_proj_covar=tensor([0.0767, 0.0655, 0.0966, 0.0913], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 02:38:02,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6258, 2.0654, 1.8467, 1.7622], device='cuda:0'), covar=tensor([0.1873, 0.1841, 0.2103, 0.1872], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0741, 0.0706, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 02:38:30,878 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.269e+02 1.386e+03 1.783e+03 2.420e+03 4.938e+03, threshold=3.566e+03, percent-clipped=2.0
+2023-03-12 02:38:33,785 INFO [train.py:968] (0/2) Epoch 23, batch 32450, giga_loss[loss=0.2127, simple_loss=0.2972, pruned_loss=0.0641, over 28911.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.33, pruned_loss=0.08832, over 5671835.49 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3515, pruned_loss=0.1135, over 5680187.22 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3288, pruned_loss=0.08577, over 5672610.47 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:39:05,447 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:39:20,363 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1035871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:39:22,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1035874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:39:37,293 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-12 02:39:42,675 INFO [train.py:968] (0/2) Epoch 23, batch 32500, giga_loss[loss=0.2627, simple_loss=0.3386, pruned_loss=0.09341, over 28961.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3285, pruned_loss=0.08761, over 5670170.65 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.351, pruned_loss=0.1133, over 5683753.40 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3276, pruned_loss=0.08539, over 5667297.39 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:39:56,232 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1035903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:40:02,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3006, 1.6296, 1.2848, 0.9894], device='cuda:0'), covar=tensor([0.2641, 0.2499, 0.2965, 0.2359], device='cuda:0'), in_proj_covar=tensor([0.1533, 0.1107, 0.1357, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 02:40:20,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1035922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:40:41,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.548e+02 1.634e+03 2.013e+03 2.868e+03 6.255e+03, threshold=4.025e+03, percent-clipped=15.0
+2023-03-12 02:40:42,675 INFO [train.py:968] (0/2) Epoch 23, batch 32550, libri_loss[loss=0.262, simple_loss=0.3355, pruned_loss=0.09427, over 29533.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3301, pruned_loss=0.08848, over 5664339.82 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.351, pruned_loss=0.1133, over 5678002.76 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.329, pruned_loss=0.08622, over 5666104.94 frames. ], batch size: 81, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:40:48,120 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:41:44,087 INFO [train.py:968] (0/2) Epoch 23, batch 32600, giga_loss[loss=0.2569, simple_loss=0.331, pruned_loss=0.09144, over 26830.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3289, pruned_loss=0.0874, over 5657423.11 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3508, pruned_loss=0.1131, over 5669617.57 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3278, pruned_loss=0.0852, over 5665391.98 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:41:55,198 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1036000.pt
+2023-03-12 02:42:00,944 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9613, 2.4777, 2.3023, 1.8756], device='cuda:0'), covar=tensor([0.2880, 0.1928, 0.2255, 0.2458], device='cuda:0'), in_proj_covar=tensor([0.1947, 0.1879, 0.1803, 0.1949], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 02:42:42,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.718e+02 1.522e+03 2.242e+03 3.169e+03 9.586e+03, threshold=4.484e+03, percent-clipped=10.0
+2023-03-12 02:42:45,443 INFO [train.py:968] (0/2) Epoch 23, batch 32650, giga_loss[loss=0.2645, simple_loss=0.3419, pruned_loss=0.09356, over 27577.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3274, pruned_loss=0.08573, over 5650873.43 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3507, pruned_loss=0.1132, over 5665701.18 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3261, pruned_loss=0.08342, over 5660288.81 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:42:58,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1036053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:43:11,146 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1036065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:43:13,077 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1036068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:43:41,047 INFO [train.py:968] (0/2) Epoch 23, batch 32700, giga_loss[loss=0.2006, simple_loss=0.2744, pruned_loss=0.06333, over 24096.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3257, pruned_loss=0.0856, over 5666201.45 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3496, pruned_loss=0.1125, over 5680573.15 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3242, pruned_loss=0.0826, over 5659931.72 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:43:50,237 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1036097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:44:49,199 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.207e+02 1.301e+03 1.653e+03 2.251e+03 4.828e+03, threshold=3.306e+03, percent-clipped=2.0
+2023-03-12 02:44:53,605 INFO [train.py:968] (0/2) Epoch 23, batch 32750, giga_loss[loss=0.2397, simple_loss=0.3263, pruned_loss=0.07654, over 29092.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.325, pruned_loss=0.08461, over 5671724.61 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3493, pruned_loss=0.1124, over 5680919.95 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3239, pruned_loss=0.08216, over 5666479.43 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:45:38,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3280, 3.1783, 3.0036, 1.3866], device='cuda:0'), covar=tensor([0.0949, 0.1028, 0.0926, 0.2370], device='cuda:0'), in_proj_covar=tensor([0.1234, 0.1140, 0.0963, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 02:46:10,646 INFO [train.py:968] (0/2) Epoch 23, batch 32800, giga_loss[loss=0.257, simple_loss=0.3365, pruned_loss=0.0887, over 28172.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3263, pruned_loss=0.08477, over 5677837.52 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3493, pruned_loss=0.1124, over 5680919.95 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3254, pruned_loss=0.08286, over 5673755.13 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:46:18,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9063, 1.2298, 1.3155, 1.0172], device='cuda:0'), covar=tensor([0.1649, 0.1218, 0.1893, 0.1431], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0736, 0.0702, 0.0674], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 02:46:18,834 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1036196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:46:21,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1036199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:46:37,991 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1036213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:46:56,367 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1036228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:47:09,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1036237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:47:11,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.486e+02 1.390e+03 1.716e+03 2.314e+03 4.285e+03, threshold=3.432e+03, percent-clipped=9.0
+2023-03-12 02:47:13,134 INFO [train.py:968] (0/2) Epoch 23, batch 32850, giga_loss[loss=0.2395, simple_loss=0.3176, pruned_loss=0.08075, over 28752.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3273, pruned_loss=0.08608, over 5686019.21 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3488, pruned_loss=0.1122, over 5685792.82 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3267, pruned_loss=0.08421, over 5678506.80 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:48:11,850 INFO [train.py:968] (0/2) Epoch 23, batch 32900, libri_loss[loss=0.3002, simple_loss=0.368, pruned_loss=0.1162, over 27666.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3266, pruned_loss=0.08586, over 5664411.20 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3488, pruned_loss=0.1123, over 5670868.83 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3253, pruned_loss=0.08342, over 5671657.53 frames. ], batch size: 115, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:48:45,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1036320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:49:08,380 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.415e+02 1.672e+03 1.974e+03 2.900e+03 5.935e+03, threshold=3.949e+03, percent-clipped=15.0
+2023-03-12 02:49:09,089 INFO [train.py:968] (0/2) Epoch 23, batch 32950, libri_loss[loss=0.233, simple_loss=0.3017, pruned_loss=0.08217, over 29360.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3269, pruned_loss=0.08478, over 5652038.12 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3481, pruned_loss=0.1119, over 5668176.06 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3258, pruned_loss=0.0823, over 5659469.67 frames. ], batch size: 67, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:49:45,695 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4100, 4.2524, 4.0271, 1.8292], device='cuda:0'), covar=tensor([0.0619, 0.0752, 0.0924, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.1231, 0.1138, 0.0961, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 02:49:55,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1036380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:49:58,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1036383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:50:00,311 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-12 02:50:07,964 INFO [train.py:968] (0/2) Epoch 23, batch 33000, giga_loss[loss=0.2309, simple_loss=0.3212, pruned_loss=0.07025, over 28797.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3309, pruned_loss=0.08628, over 5653326.67 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3483, pruned_loss=0.1121, over 5670103.18 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3297, pruned_loss=0.08383, over 5657316.09 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:50:07,968 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 02:50:16,958 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.1952, simple_loss=0.2965, pruned_loss=0.04699, over 944034.00 frames. 
+2023-03-12 02:50:16,959 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 02:50:30,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4455, 1.7229, 1.2361, 1.2490], device='cuda:0'), covar=tensor([0.1108, 0.0549, 0.1051, 0.1151], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0444, 0.0519, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 02:50:30,867 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 02:50:33,235 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-12 02:50:42,924 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1036412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:51:19,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.652e+02 1.521e+03 1.923e+03 2.810e+03 7.842e+03, threshold=3.845e+03, percent-clipped=13.0
+2023-03-12 02:51:19,871 INFO [train.py:968] (0/2) Epoch 23, batch 33050, giga_loss[loss=0.2446, simple_loss=0.3311, pruned_loss=0.07909, over 28630.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3325, pruned_loss=0.08673, over 5657843.19 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3485, pruned_loss=0.1122, over 5669316.21 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3311, pruned_loss=0.0842, over 5661471.42 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:51:43,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1254, 1.1927, 3.3706, 3.0402], device='cuda:0'), covar=tensor([0.1656, 0.2809, 0.0501, 0.1041], device='cuda:0'), in_proj_covar=tensor([0.0764, 0.0652, 0.0960, 0.0908], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 02:51:49,355 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1036463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:51:54,925 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1036466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:52:21,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5264, 1.7246, 1.1998, 1.2961], device='cuda:0'), covar=tensor([0.1078, 0.0593, 0.1039, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0444, 0.0520, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 02:52:22,785 INFO [train.py:968] (0/2) Epoch 23, batch 33100, giga_loss[loss=0.2178, simple_loss=0.3067, pruned_loss=0.06444, over 28949.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3329, pruned_loss=0.08721, over 5660210.25 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3483, pruned_loss=0.1121, over 5669430.21 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3314, pruned_loss=0.08444, over 5662078.38 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:52:28,369 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1036495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:53:15,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.914e+02 1.524e+03 2.111e+03 2.841e+03 9.384e+03, threshold=4.222e+03, percent-clipped=10.0
+2023-03-12 02:53:17,812 INFO [train.py:968] (0/2) Epoch 23, batch 33150, giga_loss[loss=0.2183, simple_loss=0.3034, pruned_loss=0.06664, over 28885.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3314, pruned_loss=0.08654, over 5663771.62 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3479, pruned_loss=0.1119, over 5665362.97 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3302, pruned_loss=0.08378, over 5668813.66 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:54:00,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4030, 1.7324, 1.4103, 1.5492], device='cuda:0'), covar=tensor([0.0743, 0.0394, 0.0361, 0.0823], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 02:54:18,025 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1036588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:54:22,410 INFO [train.py:968] (0/2) Epoch 23, batch 33200, giga_loss[loss=0.2402, simple_loss=0.3268, pruned_loss=0.07676, over 28699.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3286, pruned_loss=0.08487, over 5671835.13 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3477, pruned_loss=0.1118, over 5668957.28 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3275, pruned_loss=0.08225, over 5672543.20 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:55:17,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.279e+02 1.443e+03 1.899e+03 2.620e+03 6.620e+03, threshold=3.799e+03, percent-clipped=8.0
+2023-03-12 02:55:19,337 INFO [train.py:968] (0/2) Epoch 23, batch 33250, giga_loss[loss=0.264, simple_loss=0.3495, pruned_loss=0.08925, over 28909.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3263, pruned_loss=0.08446, over 5671315.36 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3467, pruned_loss=0.1113, over 5665623.83 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3257, pruned_loss=0.08205, over 5674307.25 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:56:14,718 INFO [train.py:968] (0/2) Epoch 23, batch 33300, giga_loss[loss=0.26, simple_loss=0.3434, pruned_loss=0.08826, over 28721.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3266, pruned_loss=0.08459, over 5672827.96 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3461, pruned_loss=0.1108, over 5670492.72 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3259, pruned_loss=0.08203, over 5670776.84 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:57:07,749 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1036731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:57:12,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1036734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:57:21,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.427e+02 1.488e+03 1.880e+03 2.414e+03 5.646e+03, threshold=3.760e+03, percent-clipped=5.0
+2023-03-12 02:57:21,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5372, 1.7528, 1.2315, 1.3647], device='cuda:0'), covar=tensor([0.0939, 0.0484, 0.0925, 0.1165], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0445, 0.0521, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 02:57:22,071 INFO [train.py:968] (0/2) Epoch 23, batch 33350, giga_loss[loss=0.2632, simple_loss=0.3318, pruned_loss=0.0973, over 26885.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3292, pruned_loss=0.08561, over 5669533.12 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3462, pruned_loss=0.1109, over 5670123.25 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3282, pruned_loss=0.08305, over 5668414.44 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:57:49,641 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1036763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:58:25,545 INFO [train.py:968] (0/2) Epoch 23, batch 33400, giga_loss[loss=0.2532, simple_loss=0.3295, pruned_loss=0.08848, over 28061.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3295, pruned_loss=0.08615, over 5679977.41 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3454, pruned_loss=0.1103, over 5678291.72 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3289, pruned_loss=0.08379, over 5671717.80 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:59:29,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.213e+02 1.380e+03 1.738e+03 2.473e+03 7.029e+03, threshold=3.477e+03, percent-clipped=6.0
+2023-03-12 02:59:30,925 INFO [train.py:968] (0/2) Epoch 23, batch 33450, giga_loss[loss=0.2497, simple_loss=0.3351, pruned_loss=0.08216, over 28979.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3312, pruned_loss=0.08728, over 5666616.39 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3449, pruned_loss=0.11, over 5680044.44 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3307, pruned_loss=0.08512, over 5658270.53 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 03:00:22,643 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1036889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:00:24,125 INFO [train.py:968] (0/2) Epoch 23, batch 33500, giga_loss[loss=0.2509, simple_loss=0.3378, pruned_loss=0.08201, over 28941.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3339, pruned_loss=0.08826, over 5669307.55 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3443, pruned_loss=0.1096, over 5687766.51 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3337, pruned_loss=0.08617, over 5654861.75 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:00:28,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-12 03:01:26,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-12 03:01:27,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.523e+02 1.456e+03 1.852e+03 2.638e+03 5.453e+03, threshold=3.704e+03, percent-clipped=9.0
+2023-03-12 03:01:27,710 INFO [train.py:968] (0/2) Epoch 23, batch 33550, giga_loss[loss=0.2713, simple_loss=0.3488, pruned_loss=0.09694, over 28989.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3369, pruned_loss=0.08978, over 5665627.72 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3446, pruned_loss=0.1098, over 5686201.03 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3363, pruned_loss=0.08727, over 5653823.25 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:01:35,669 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-12 03:02:11,476 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1036969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:02:20,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4173, 1.2483, 1.2162, 1.5075], device='cuda:0'), covar=tensor([0.0760, 0.0344, 0.0353, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 03:02:43,722 INFO [train.py:968] (0/2) Epoch 23, batch 33600, giga_loss[loss=0.2518, simple_loss=0.3282, pruned_loss=0.08771, over 28996.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3361, pruned_loss=0.08931, over 5679118.89 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3442, pruned_loss=0.1096, over 5688451.30 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3359, pruned_loss=0.08736, over 5667679.90 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:02:51,799 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-12 03:03:06,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4048, 1.6276, 1.3589, 1.5374], device='cuda:0'), covar=tensor([0.0780, 0.0300, 0.0335, 0.0911], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 03:04:03,502 INFO [train.py:968] (0/2) Epoch 23, batch 33650, giga_loss[loss=0.222, simple_loss=0.3158, pruned_loss=0.06403, over 28858.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3333, pruned_loss=0.08759, over 5680344.31 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3442, pruned_loss=0.1096, over 5688451.30 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3331, pruned_loss=0.08606, over 5671441.21 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:04:06,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.973e+02 1.809e+03 2.485e+03 3.720e+03 1.206e+04, threshold=4.970e+03, percent-clipped=26.0
+2023-03-12 03:04:23,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1037055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:04:36,577 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-12 03:05:10,584 INFO [train.py:968] (0/2) Epoch 23, batch 33700, giga_loss[loss=0.2528, simple_loss=0.3271, pruned_loss=0.08922, over 28919.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3325, pruned_loss=0.08733, over 5672191.82 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3443, pruned_loss=0.1097, over 5682739.03 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3321, pruned_loss=0.08553, over 5669874.64 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:05:45,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4384, 1.9601, 1.5231, 1.6643], device='cuda:0'), covar=tensor([0.0779, 0.0264, 0.0334, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 03:06:17,420 INFO [train.py:968] (0/2) Epoch 23, batch 33750, giga_loss[loss=0.2127, simple_loss=0.2867, pruned_loss=0.06937, over 28675.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3314, pruned_loss=0.08817, over 5671035.85 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3442, pruned_loss=0.1097, over 5685793.04 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3308, pruned_loss=0.08636, over 5666268.87 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:06:18,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.327e+02 1.450e+03 1.809e+03 2.369e+03 5.573e+03, threshold=3.619e+03, percent-clipped=2.0
+2023-03-12 03:06:25,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5570, 2.1164, 1.6530, 1.6121], device='cuda:0'), covar=tensor([0.0766, 0.0258, 0.0327, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 03:07:23,265 INFO [train.py:968] (0/2) Epoch 23, batch 33800, giga_loss[loss=0.2431, simple_loss=0.3322, pruned_loss=0.07695, over 27515.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.33, pruned_loss=0.08803, over 5673791.16 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3443, pruned_loss=0.1097, over 5681223.78 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3292, pruned_loss=0.08597, over 5673449.84 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:08:17,317 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1037235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:08:25,820 INFO [train.py:968] (0/2) Epoch 23, batch 33850, giga_loss[loss=0.2218, simple_loss=0.315, pruned_loss=0.0643, over 28965.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3293, pruned_loss=0.08601, over 5668994.53 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3445, pruned_loss=0.1099, over 5674766.86 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3282, pruned_loss=0.08381, over 5674350.10 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:08:28,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.497e+02 1.397e+03 1.737e+03 2.490e+03 8.200e+03, threshold=3.474e+03, percent-clipped=10.0
+2023-03-12 03:08:54,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:09:24,168 INFO [train.py:968] (0/2) Epoch 23, batch 33900, giga_loss[loss=0.2251, simple_loss=0.3223, pruned_loss=0.06392, over 29030.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3283, pruned_loss=0.08448, over 5662443.14 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3444, pruned_loss=0.1099, over 5668622.75 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3271, pruned_loss=0.08203, over 5671804.50 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:10:23,389 INFO [train.py:968] (0/2) Epoch 23, batch 33950, giga_loss[loss=0.2467, simple_loss=0.3394, pruned_loss=0.077, over 28943.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3304, pruned_loss=0.08366, over 5662399.98 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3446, pruned_loss=0.11, over 5661191.08 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.329, pruned_loss=0.08125, over 5676022.15 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:10:26,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.160e+02 1.617e+03 2.066e+03 2.958e+03 8.110e+03, threshold=4.131e+03, percent-clipped=21.0
+2023-03-12 03:10:29,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:11:22,837 INFO [train.py:968] (0/2) Epoch 23, batch 34000, giga_loss[loss=0.2622, simple_loss=0.347, pruned_loss=0.08868, over 28446.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.331, pruned_loss=0.08366, over 5669532.86 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3441, pruned_loss=0.1096, over 5664777.17 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.33, pruned_loss=0.08136, over 5676984.03 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:11:43,120 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1037404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:11:43,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7732, 3.6235, 3.4136, 1.7095], device='cuda:0'), covar=tensor([0.0745, 0.0845, 0.0846, 0.2204], device='cuda:0'), in_proj_covar=tensor([0.1230, 0.1135, 0.0958, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 03:11:47,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:11:50,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:12:09,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4994, 1.6100, 1.7133, 1.3394], device='cuda:0'), covar=tensor([0.1860, 0.2611, 0.1520, 0.1896], device='cuda:0'), in_proj_covar=tensor([0.0904, 0.0696, 0.0951, 0.0851], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 03:12:16,410 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1037427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:12:19,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:12:34,031 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:12:36,555 INFO [train.py:968] (0/2) Epoch 23, batch 34050, giga_loss[loss=0.243, simple_loss=0.3282, pruned_loss=0.07892, over 28623.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3317, pruned_loss=0.08422, over 5660485.89 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.344, pruned_loss=0.1096, over 5658189.24 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3308, pruned_loss=0.08207, over 5672404.54 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:12:39,194 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.352e+03 1.727e+03 2.584e+03 1.439e+04, threshold=3.455e+03, percent-clipped=6.0
+2023-03-12 03:13:10,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6637, 1.9787, 1.3028, 1.5979], device='cuda:0'), covar=tensor([0.1009, 0.0595, 0.1005, 0.1209], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0444, 0.0519, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 03:13:19,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6397, 3.4911, 3.2974, 1.6531], device='cuda:0'), covar=tensor([0.0773, 0.0872, 0.0863, 0.2429], device='cuda:0'), in_proj_covar=tensor([0.1233, 0.1139, 0.0961, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 03:13:35,414 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:13:39,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:13:39,886 INFO [train.py:968] (0/2) Epoch 23, batch 34100, giga_loss[loss=0.2814, simple_loss=0.3604, pruned_loss=0.1012, over 28658.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3322, pruned_loss=0.08465, over 5659807.72 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3441, pruned_loss=0.1096, over 5661705.41 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3312, pruned_loss=0.08238, over 5666219.00 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:14:01,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4321, 1.4528, 1.6025, 1.2315], device='cuda:0'), covar=tensor([0.1605, 0.2730, 0.1331, 0.1663], device='cuda:0'), in_proj_covar=tensor([0.0903, 0.0696, 0.0950, 0.0851], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 03:14:19,958 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:14:53,237 INFO [train.py:968] (0/2) Epoch 23, batch 34150, giga_loss[loss=0.2457, simple_loss=0.3412, pruned_loss=0.07507, over 28747.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3336, pruned_loss=0.08541, over 5658132.13 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3445, pruned_loss=0.1101, over 5659205.13 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3321, pruned_loss=0.08252, over 5665434.53 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:14:53,719 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 03:14:56,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.590e+02 1.530e+03 2.217e+03 3.197e+03 1.121e+04, threshold=4.434e+03, percent-clipped=19.0
+2023-03-12 03:15:40,741 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:15:44,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:16:05,197 INFO [train.py:968] (0/2) Epoch 23, batch 34200, giga_loss[loss=0.2768, simple_loss=0.3603, pruned_loss=0.09665, over 27717.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3338, pruned_loss=0.08518, over 5661259.38 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3445, pruned_loss=0.1102, over 5663678.90 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3324, pruned_loss=0.08237, over 5663276.49 frames. ], batch size: 474, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:16:28,627 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:16:35,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:17:15,882 INFO [train.py:968] (0/2) Epoch 23, batch 34250, giga_loss[loss=0.2349, simple_loss=0.3292, pruned_loss=0.07027, over 28777.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3376, pruned_loss=0.08688, over 5668090.56 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3443, pruned_loss=0.1101, over 5666195.66 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3365, pruned_loss=0.08443, over 5667391.51 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:17:22,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.249e+02 1.567e+03 2.190e+03 3.118e+03 7.125e+03, threshold=4.379e+03, percent-clipped=9.0
+2023-03-12 03:17:24,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7252, 3.5674, 3.3978, 1.8963], device='cuda:0'), covar=tensor([0.0706, 0.0857, 0.0823, 0.2443], device='cuda:0'), in_proj_covar=tensor([0.1229, 0.1134, 0.0958, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 03:18:27,996 INFO [train.py:968] (0/2) Epoch 23, batch 34300, giga_loss[loss=0.2569, simple_loss=0.3381, pruned_loss=0.08779, over 29099.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3377, pruned_loss=0.08642, over 5672605.68 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3443, pruned_loss=0.11, over 5664778.94 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3368, pruned_loss=0.08439, over 5673080.96 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:18:53,121 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-12 03:19:39,688 INFO [train.py:968] (0/2) Epoch 23, batch 34350, giga_loss[loss=0.2398, simple_loss=0.3224, pruned_loss=0.07865, over 27800.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3368, pruned_loss=0.087, over 5671367.92 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3445, pruned_loss=0.11, over 5668301.27 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3359, pruned_loss=0.08495, over 5668805.26 frames. ], batch size: 476, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:19:42,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.667e+02 1.304e+03 1.655e+03 2.011e+03 8.000e+03, threshold=3.310e+03, percent-clipped=3.0
+2023-03-12 03:19:56,540 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:20:00,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:20:25,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7081, 1.9569, 1.6267, 1.7950], device='cuda:0'), covar=tensor([0.2475, 0.2294, 0.2529, 0.2315], device='cuda:0'), in_proj_covar=tensor([0.1527, 0.1102, 0.1352, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 03:20:35,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:20:44,211 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:20:53,453 INFO [train.py:968] (0/2) Epoch 23, batch 34400, giga_loss[loss=0.2163, simple_loss=0.3075, pruned_loss=0.06257, over 28861.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3339, pruned_loss=0.08511, over 5677906.00 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.345, pruned_loss=0.1104, over 5671774.46 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3326, pruned_loss=0.08288, over 5672991.10 frames. ], batch size: 112, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:21:07,829 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:21:17,878 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1037809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:21:56,988 INFO [train.py:968] (0/2) Epoch 23, batch 34450, giga_loss[loss=0.241, simple_loss=0.329, pruned_loss=0.07652, over 28526.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3321, pruned_loss=0.08347, over 5685220.46 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3449, pruned_loss=0.1103, over 5675130.00 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.331, pruned_loss=0.08139, over 5678629.30 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:22:00,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.047e+02 1.317e+03 1.805e+03 2.327e+03 7.119e+03, threshold=3.611e+03, percent-clipped=14.0
+2023-03-12 03:22:02,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3975, 3.2489, 3.0664, 1.7631], device='cuda:0'), covar=tensor([0.0747, 0.0880, 0.0870, 0.2038], device='cuda:0'), in_proj_covar=tensor([0.1226, 0.1127, 0.0955, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 03:22:51,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5169, 1.4967, 1.7185, 1.3120], device='cuda:0'), covar=tensor([0.1796, 0.2733, 0.1552, 0.1960], device='cuda:0'), in_proj_covar=tensor([0.0901, 0.0693, 0.0949, 0.0850], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 03:23:01,819 INFO [train.py:968] (0/2) Epoch 23, batch 34500, giga_loss[loss=0.2612, simple_loss=0.3414, pruned_loss=0.09044, over 28726.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3332, pruned_loss=0.08438, over 5676366.35 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.345, pruned_loss=0.1103, over 5678221.08 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.332, pruned_loss=0.08224, over 5668441.87 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:23:39,173 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:23:42,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:24:03,241 INFO [train.py:968] (0/2) Epoch 23, batch 34550, giga_loss[loss=0.2456, simple_loss=0.332, pruned_loss=0.07961, over 29106.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3362, pruned_loss=0.08623, over 5682241.02 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3448, pruned_loss=0.1102, over 5684547.43 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3351, pruned_loss=0.08395, over 5669998.99 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:24:06,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.445e+02 1.500e+03 1.916e+03 2.475e+03 7.457e+03, threshold=3.831e+03, percent-clipped=8.0
+2023-03-12 03:24:09,150 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:24:12,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:24:20,769 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:24:47,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:25:04,265 INFO [train.py:968] (0/2) Epoch 23, batch 34600, giga_loss[loss=0.252, simple_loss=0.3311, pruned_loss=0.08641, over 28930.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3352, pruned_loss=0.08618, over 5690684.97 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3447, pruned_loss=0.1102, over 5691713.37 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3342, pruned_loss=0.08358, over 5674212.96 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:25:10,860 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.68 vs. limit=5.0
+2023-03-12 03:25:14,173 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1038000.pt
+2023-03-12 03:26:01,923 INFO [train.py:968] (0/2) Epoch 23, batch 34650, giga_loss[loss=0.2287, simple_loss=0.3105, pruned_loss=0.07343, over 28492.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3322, pruned_loss=0.08579, over 5671128.37 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3445, pruned_loss=0.1101, over 5683332.38 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3313, pruned_loss=0.08328, over 5664396.15 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:26:04,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.994e+02 1.522e+03 1.911e+03 2.591e+03 4.097e+03, threshold=3.823e+03, percent-clipped=4.0
+2023-03-12 03:26:13,451 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 03:26:55,546 INFO [train.py:968] (0/2) Epoch 23, batch 34700, giga_loss[loss=0.2872, simple_loss=0.3593, pruned_loss=0.1075, over 27493.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.332, pruned_loss=0.0867, over 5672409.04 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3438, pruned_loss=0.1095, over 5689865.91 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3314, pruned_loss=0.08422, over 5660603.49 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:27:03,358 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-12 03:27:20,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2439, 1.8650, 0.9413, 1.2476], device='cuda:0'), covar=tensor([0.1368, 0.0613, 0.1632, 0.1372], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0442, 0.0518, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 03:27:40,545 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-12 03:27:47,099 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:27:51,977 INFO [train.py:968] (0/2) Epoch 23, batch 34750, giga_loss[loss=0.3317, simple_loss=0.3966, pruned_loss=0.1334, over 27592.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3361, pruned_loss=0.0893, over 5671905.02 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3435, pruned_loss=0.1093, over 5687041.49 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3357, pruned_loss=0.0869, over 5664718.24 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:27:54,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.991e+02 1.501e+03 1.955e+03 2.685e+03 1.079e+04, threshold=3.909e+03, percent-clipped=9.0
+2023-03-12 03:28:31,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:28:39,068 INFO [train.py:968] (0/2) Epoch 23, batch 34800, giga_loss[loss=0.2845, simple_loss=0.3434, pruned_loss=0.1128, over 23908.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3442, pruned_loss=0.0937, over 5675713.56 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3436, pruned_loss=0.1093, over 5689813.97 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3437, pruned_loss=0.09148, over 5667372.50 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 03:28:46,503 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:29:27,551 INFO [train.py:968] (0/2) Epoch 23, batch 34850, giga_loss[loss=0.2822, simple_loss=0.3559, pruned_loss=0.1042, over 28538.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.35, pruned_loss=0.09689, over 5669601.15 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3436, pruned_loss=0.1092, over 5689332.61 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3496, pruned_loss=0.09504, over 5663209.82 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:29:30,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.723e+02 1.363e+03 1.704e+03 2.227e+03 7.694e+03, threshold=3.408e+03, percent-clipped=7.0
+2023-03-12 03:30:12,443 INFO [train.py:968] (0/2) Epoch 23, batch 34900, giga_loss[loss=0.2039, simple_loss=0.2918, pruned_loss=0.05802, over 28890.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3469, pruned_loss=0.09593, over 5674730.97 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3435, pruned_loss=0.1091, over 5683474.76 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3467, pruned_loss=0.09436, over 5674262.84 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:30:42,742 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:30:45,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:30:48,135 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:30:54,546 INFO [train.py:968] (0/2) Epoch 23, batch 34950, giga_loss[loss=0.2385, simple_loss=0.307, pruned_loss=0.08495, over 28513.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3403, pruned_loss=0.09347, over 5683011.20 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3429, pruned_loss=0.1085, over 5685248.37 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3408, pruned_loss=0.09232, over 5680740.54 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:30:57,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.696e+02 1.250e+03 1.795e+03 2.770e+03 7.631e+03, threshold=3.590e+03, percent-clipped=14.0
+2023-03-12 03:31:12,563 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1038359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:31:20,144 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5370, 1.8433, 1.5250, 1.5568], device='cuda:0'), covar=tensor([0.2534, 0.2454, 0.2705, 0.2385], device='cuda:0'), in_proj_covar=tensor([0.1529, 0.1105, 0.1356, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 03:31:27,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-12 03:31:37,355 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:31:38,977 INFO [train.py:968] (0/2) Epoch 23, batch 35000, giga_loss[loss=0.231, simple_loss=0.3069, pruned_loss=0.07757, over 27702.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3346, pruned_loss=0.09157, over 5685422.06 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3431, pruned_loss=0.1085, over 5689696.67 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3346, pruned_loss=0.09025, over 5679592.02 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:32:08,786 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-12 03:32:18,645 INFO [train.py:968] (0/2) Epoch 23, batch 35050, giga_loss[loss=0.2311, simple_loss=0.299, pruned_loss=0.08162, over 28602.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3286, pruned_loss=0.08945, over 5690086.85 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3435, pruned_loss=0.1087, over 5694556.58 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3277, pruned_loss=0.08754, over 5680705.36 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:32:23,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.144e+02 1.193e+03 1.525e+03 2.285e+03 5.494e+03, threshold=3.051e+03, percent-clipped=6.0
+2023-03-12 03:32:33,192 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:32:44,037 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 03:33:00,113 INFO [train.py:968] (0/2) Epoch 23, batch 35100, giga_loss[loss=0.201, simple_loss=0.281, pruned_loss=0.06049, over 28874.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3208, pruned_loss=0.08563, over 5696714.85 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3434, pruned_loss=0.1084, over 5699921.05 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3199, pruned_loss=0.08389, over 5684449.59 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:33:20,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:33:29,345 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-12 03:33:45,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:33:46,188 INFO [train.py:968] (0/2) Epoch 23, batch 35150, giga_loss[loss=0.2103, simple_loss=0.2867, pruned_loss=0.06691, over 28128.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3156, pruned_loss=0.08327, over 5697920.13 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3434, pruned_loss=0.1083, over 5702374.27 frames. ], giga_tot_loss[loss=0.2385, simple_loss=0.3142, pruned_loss=0.08143, over 5685845.95 frames. ], batch size: 77, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:33:49,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.966e+02 1.094e+03 1.485e+03 1.907e+03 5.624e+03, threshold=2.969e+03, percent-clipped=5.0
+2023-03-12 03:34:07,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5898, 1.7987, 1.3239, 1.2625], device='cuda:0'), covar=tensor([0.1079, 0.0659, 0.1152, 0.1306], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0444, 0.0520, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 03:34:10,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:34:27,595 INFO [train.py:968] (0/2) Epoch 23, batch 35200, giga_loss[loss=0.205, simple_loss=0.2786, pruned_loss=0.06572, over 28443.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3121, pruned_loss=0.08166, over 5695898.72 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3433, pruned_loss=0.108, over 5705954.41 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.3106, pruned_loss=0.07998, over 5682991.56 frames. ], batch size: 65, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:35:13,964 INFO [train.py:968] (0/2) Epoch 23, batch 35250, giga_loss[loss=0.2143, simple_loss=0.2973, pruned_loss=0.06563, over 29054.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.308, pruned_loss=0.07946, over 5701536.77 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3434, pruned_loss=0.108, over 5706941.64 frames. ], giga_tot_loss[loss=0.2312, simple_loss=0.3065, pruned_loss=0.07796, over 5690529.33 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:35:18,719 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.616e+02 1.153e+03 1.593e+03 2.264e+03 6.173e+03, threshold=3.186e+03, percent-clipped=11.0
+2023-03-12 03:35:25,861 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:35:28,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:35:36,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3450, 1.6241, 1.3383, 0.9982], device='cuda:0'), covar=tensor([0.2801, 0.2948, 0.3379, 0.2571], device='cuda:0'), in_proj_covar=tensor([0.1530, 0.1105, 0.1356, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 03:35:54,201 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1038687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:35:58,656 INFO [train.py:968] (0/2) Epoch 23, batch 35300, giga_loss[loss=0.2015, simple_loss=0.2833, pruned_loss=0.05991, over 28885.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3074, pruned_loss=0.07931, over 5713824.91 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.344, pruned_loss=0.1083, over 5713425.68 frames. ], giga_tot_loss[loss=0.2293, simple_loss=0.3045, pruned_loss=0.07703, over 5699320.26 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:36:11,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2855, 1.5343, 1.3279, 1.1451], device='cuda:0'), covar=tensor([0.2853, 0.2744, 0.1951, 0.2582], device='cuda:0'), in_proj_covar=tensor([0.1981, 0.1899, 0.1814, 0.1966], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 03:36:15,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:36:20,689 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:36:22,838 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:36:30,157 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-12 03:36:42,713 INFO [train.py:968] (0/2) Epoch 23, batch 35350, giga_loss[loss=0.1963, simple_loss=0.2754, pruned_loss=0.05863, over 28900.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3047, pruned_loss=0.0783, over 5710336.12 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3449, pruned_loss=0.1086, over 5716236.50 frames. ], giga_tot_loss[loss=0.2257, simple_loss=0.3007, pruned_loss=0.07535, over 5696096.15 frames. ], batch size: 112, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:36:46,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.874e+02 1.214e+03 1.433e+03 1.833e+03 7.017e+03, threshold=2.866e+03, percent-clipped=8.0
+2023-03-12 03:36:47,992 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1038747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:37:00,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:37:02,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:37:20,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 03:37:26,914 INFO [train.py:968] (0/2) Epoch 23, batch 35400, giga_loss[loss=0.2057, simple_loss=0.2884, pruned_loss=0.06147, over 29002.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3015, pruned_loss=0.07681, over 5707738.92 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3455, pruned_loss=0.109, over 5716376.93 frames. ], giga_tot_loss[loss=0.2222, simple_loss=0.2971, pruned_loss=0.07364, over 5696320.26 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:37:59,735 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:38:03,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3148, 3.4857, 2.3688, 1.3320], device='cuda:0'), covar=tensor([0.6130, 0.2629, 0.3429, 0.5755], device='cuda:0'), in_proj_covar=tensor([0.1770, 0.1674, 0.1607, 0.1439], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 03:38:07,920 INFO [train.py:968] (0/2) Epoch 23, batch 35450, giga_loss[loss=0.1942, simple_loss=0.266, pruned_loss=0.06123, over 28545.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3, pruned_loss=0.07643, over 5691217.33 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3455, pruned_loss=0.1091, over 5701580.17 frames. ], giga_tot_loss[loss=0.2201, simple_loss=0.2949, pruned_loss=0.07269, over 5694101.19 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:38:12,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.061e+02 1.135e+03 1.425e+03 1.968e+03 1.009e+04, threshold=2.851e+03, percent-clipped=12.0
+2023-03-12 03:38:18,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:38:20,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:38:46,194 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1038883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:38:51,534 INFO [train.py:968] (0/2) Epoch 23, batch 35500, giga_loss[loss=0.1867, simple_loss=0.2657, pruned_loss=0.05383, over 28996.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2972, pruned_loss=0.07541, over 5682753.82 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3455, pruned_loss=0.1092, over 5696412.04 frames. ], giga_tot_loss[loss=0.2176, simple_loss=0.292, pruned_loss=0.07155, over 5688665.83 frames. ], batch size: 136, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:39:09,533 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:39:11,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:39:15,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:39:32,124 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-12 03:39:40,650 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1038939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:39:41,656 INFO [train.py:968] (0/2) Epoch 23, batch 35550, giga_loss[loss=0.2515, simple_loss=0.3212, pruned_loss=0.09085, over 28934.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2943, pruned_loss=0.07415, over 5690352.02 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3455, pruned_loss=0.1092, over 5699703.25 frames. ], giga_tot_loss[loss=0.2154, simple_loss=0.2895, pruned_loss=0.07064, over 5691986.95 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:39:44,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.610e+02 1.050e+03 1.335e+03 2.002e+03 4.660e+03, threshold=2.670e+03, percent-clipped=11.0
+2023-03-12 03:40:05,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5879, 1.6880, 1.8080, 1.3804], device='cuda:0'), covar=tensor([0.1894, 0.2669, 0.1552, 0.1807], device='cuda:0'), in_proj_covar=tensor([0.0913, 0.0701, 0.0961, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 03:40:09,046 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:40:11,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:40:24,007 INFO [train.py:968] (0/2) Epoch 23, batch 35600, giga_loss[loss=0.2608, simple_loss=0.3434, pruned_loss=0.08906, over 28773.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3029, pruned_loss=0.07905, over 5688961.52 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3465, pruned_loss=0.1097, over 5704750.40 frames. ], giga_tot_loss[loss=0.2231, simple_loss=0.2967, pruned_loss=0.07477, over 5685516.52 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:40:27,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6330, 1.8550, 1.3736, 1.3566], device='cuda:0'), covar=tensor([0.1044, 0.0600, 0.1031, 0.1202], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0443, 0.0519, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 03:40:41,109 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1039006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:41:05,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4080, 1.5122, 1.6177, 1.2920], device='cuda:0'), covar=tensor([0.1340, 0.1948, 0.1119, 0.1431], device='cuda:0'), in_proj_covar=tensor([0.0911, 0.0701, 0.0960, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 03:41:12,878 INFO [train.py:968] (0/2) Epoch 23, batch 35650, giga_loss[loss=0.2848, simple_loss=0.3681, pruned_loss=0.1007, over 28629.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3137, pruned_loss=0.08442, over 5693390.41 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3461, pruned_loss=0.1096, over 5711016.00 frames. ], giga_tot_loss[loss=0.234, simple_loss=0.3077, pruned_loss=0.0801, over 5684369.93 frames. ], batch size: 60, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:41:15,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.281e+02 1.309e+03 1.644e+03 2.282e+03 6.810e+03, threshold=3.289e+03, percent-clipped=15.0
+2023-03-12 03:41:28,936 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1039058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:41:31,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1039061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:41:58,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1039090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:41:58,719 INFO [train.py:968] (0/2) Epoch 23, batch 35700, giga_loss[loss=0.2875, simple_loss=0.3708, pruned_loss=0.1021, over 28775.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3262, pruned_loss=0.09054, over 5687368.31 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3465, pruned_loss=0.1098, over 5700611.41 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3207, pruned_loss=0.08658, over 5689584.60 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:42:40,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1039136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:42:43,996 INFO [train.py:968] (0/2) Epoch 23, batch 35750, giga_loss[loss=0.2997, simple_loss=0.3803, pruned_loss=0.1095, over 28933.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3346, pruned_loss=0.09409, over 5684615.42 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3466, pruned_loss=0.1098, over 5701882.94 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3299, pruned_loss=0.09071, over 5685207.21 frames. ], batch size: 164, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:42:48,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.290e+02 1.293e+03 1.671e+03 2.274e+03 6.557e+03, threshold=3.341e+03, percent-clipped=10.0
+2023-03-12 03:43:27,713 INFO [train.py:968] (0/2) Epoch 23, batch 35800, giga_loss[loss=0.2995, simple_loss=0.3775, pruned_loss=0.1108, over 28865.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3377, pruned_loss=0.09421, over 5682562.69 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.347, pruned_loss=0.11, over 5703221.99 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3335, pruned_loss=0.09099, over 5681507.42 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:44:13,620 INFO [train.py:968] (0/2) Epoch 23, batch 35850, giga_loss[loss=0.2725, simple_loss=0.3553, pruned_loss=0.09489, over 28558.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3405, pruned_loss=0.09502, over 5695027.31 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3478, pruned_loss=0.1103, over 5711213.37 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3359, pruned_loss=0.09148, over 5686318.30 frames. ], batch size: 307, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:44:18,527 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.569e+02 1.297e+03 1.724e+03 2.455e+03 8.402e+03, threshold=3.448e+03, percent-clipped=5.0
+2023-03-12 03:44:43,692 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1039274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:44:48,165 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1039279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:44:50,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1039282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:44:54,939 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1039289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:44:56,105 INFO [train.py:968] (0/2) Epoch 23, batch 35900, giga_loss[loss=0.3664, simple_loss=0.4088, pruned_loss=0.162, over 26494.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3427, pruned_loss=0.09619, over 5697615.85 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3485, pruned_loss=0.1106, over 5713532.20 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3381, pruned_loss=0.09245, over 5687938.10 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:45:00,798 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5370, 2.3200, 1.6816, 0.5736], device='cuda:0'), covar=tensor([0.5865, 0.3334, 0.4313, 0.6708], device='cuda:0'), in_proj_covar=tensor([0.1776, 0.1681, 0.1616, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 03:45:14,364 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1039311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:45:20,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3430, 1.3777, 1.3218, 1.3135], device='cuda:0'), covar=tensor([0.2044, 0.2247, 0.2071, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.1984, 0.1910, 0.1820, 0.1977], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 03:45:29,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8376, 2.3850, 1.8706, 2.1964], device='cuda:0'), covar=tensor([0.2536, 0.2503, 0.2813, 0.2451], device='cuda:0'), in_proj_covar=tensor([0.1527, 0.1103, 0.1352, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 03:45:40,658 INFO [train.py:968] (0/2) Epoch 23, batch 35950, giga_loss[loss=0.3107, simple_loss=0.3687, pruned_loss=0.1263, over 28521.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3453, pruned_loss=0.09834, over 5691651.63 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3483, pruned_loss=0.1102, over 5717165.12 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3417, pruned_loss=0.09528, over 5680173.82 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:45:44,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.356e+02 1.300e+03 1.544e+03 1.949e+03 4.299e+03, threshold=3.089e+03, percent-clipped=4.0
+2023-03-12 03:45:52,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6560, 1.9707, 1.5607, 2.0291], device='cuda:0'), covar=tensor([0.2566, 0.2576, 0.2896, 0.2386], device='cuda:0'), in_proj_covar=tensor([0.1527, 0.1102, 0.1351, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 03:46:21,612 INFO [train.py:968] (0/2) Epoch 23, batch 36000, giga_loss[loss=0.343, simple_loss=0.3956, pruned_loss=0.1452, over 26564.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3487, pruned_loss=0.1003, over 5692342.97 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3484, pruned_loss=0.1102, over 5720571.87 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3457, pruned_loss=0.09767, over 5679510.86 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:46:21,616 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 03:46:30,745 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2031, simple_loss=0.3104, pruned_loss=0.04787, over 944034.00 frames. 
+2023-03-12 03:46:30,745 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 03:46:49,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4790, 1.7532, 1.3987, 1.7551], device='cuda:0'), covar=tensor([0.2790, 0.2879, 0.3183, 0.2438], device='cuda:0'), in_proj_covar=tensor([0.1529, 0.1104, 0.1352, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 03:47:10,514 INFO [train.py:968] (0/2) Epoch 23, batch 36050, giga_loss[loss=0.3771, simple_loss=0.4268, pruned_loss=0.1637, over 27917.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3514, pruned_loss=0.1018, over 5688641.58 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3486, pruned_loss=0.1102, over 5713360.31 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3488, pruned_loss=0.09929, over 5683328.01 frames. ], batch size: 412, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:47:15,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.471e+02 1.236e+03 1.653e+03 2.105e+03 7.015e+03, threshold=3.306e+03, percent-clipped=13.0
+2023-03-12 03:47:48,534 INFO [train.py:968] (0/2) Epoch 23, batch 36100, giga_loss[loss=0.3205, simple_loss=0.3622, pruned_loss=0.1394, over 23637.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.354, pruned_loss=0.1026, over 5683797.74 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3491, pruned_loss=0.1104, over 5707062.81 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3515, pruned_loss=0.1001, over 5685175.79 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:48:34,096 INFO [train.py:968] (0/2) Epoch 23, batch 36150, giga_loss[loss=0.2835, simple_loss=0.3636, pruned_loss=0.1017, over 28881.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3551, pruned_loss=0.1029, over 5677535.64 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3491, pruned_loss=0.1103, over 5709941.89 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3532, pruned_loss=0.1008, over 5675806.24 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:48:38,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.521e+02 1.259e+03 1.563e+03 1.910e+03 6.683e+03, threshold=3.126e+03, percent-clipped=5.0
+2023-03-12 03:48:48,461 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-12 03:49:14,872 INFO [train.py:968] (0/2) Epoch 23, batch 36200, giga_loss[loss=0.2477, simple_loss=0.3351, pruned_loss=0.08018, over 28735.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3548, pruned_loss=0.1009, over 5692703.46 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3488, pruned_loss=0.1101, over 5712050.53 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3537, pruned_loss=0.0995, over 5689280.77 frames. ], batch size: 66, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:49:16,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4854, 1.5994, 1.1763, 1.1807], device='cuda:0'), covar=tensor([0.0961, 0.0515, 0.0995, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0444, 0.0521, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 03:49:33,000 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2144, 2.5390, 2.2172, 1.7336], device='cuda:0'), covar=tensor([0.3052, 0.2406, 0.2884, 0.3168], device='cuda:0'), in_proj_covar=tensor([0.1983, 0.1913, 0.1818, 0.1978], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 03:49:38,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3374, 1.8273, 1.4679, 1.4849], device='cuda:0'), covar=tensor([0.0776, 0.0372, 0.0338, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 03:49:54,150 INFO [train.py:968] (0/2) Epoch 23, batch 36250, giga_loss[loss=0.2765, simple_loss=0.3581, pruned_loss=0.09749, over 28709.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3528, pruned_loss=0.09854, over 5706677.67 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3486, pruned_loss=0.1097, over 5718826.32 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3523, pruned_loss=0.09733, over 5697174.01 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:50:01,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.061e+02 1.076e+03 1.420e+03 2.012e+03 5.706e+03, threshold=2.840e+03, percent-clipped=5.0
+2023-03-12 03:50:02,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1039649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:50:13,577 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1039664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:50:37,584 INFO [train.py:968] (0/2) Epoch 23, batch 36300, giga_loss[loss=0.2691, simple_loss=0.3542, pruned_loss=0.09201, over 28895.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3507, pruned_loss=0.09649, over 5702330.45 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3487, pruned_loss=0.1097, over 5711447.11 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3503, pruned_loss=0.09499, over 5700832.73 frames. ], batch size: 112, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 03:50:37,927 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2603, 1.5341, 1.5219, 1.1281], device='cuda:0'), covar=tensor([0.1836, 0.2659, 0.1519, 0.1803], device='cuda:0'), in_proj_covar=tensor([0.0912, 0.0702, 0.0961, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 03:51:17,569 INFO [train.py:968] (0/2) Epoch 23, batch 36350, giga_loss[loss=0.2738, simple_loss=0.3515, pruned_loss=0.09804, over 28492.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3509, pruned_loss=0.09678, over 5704072.01 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3489, pruned_loss=0.1096, over 5708419.96 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3505, pruned_loss=0.09524, over 5704570.86 frames. ], batch size: 65, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 03:51:22,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5600, 1.6834, 1.5842, 1.3881], device='cuda:0'), covar=tensor([0.2840, 0.2834, 0.2431, 0.2801], device='cuda:0'), in_proj_covar=tensor([0.1985, 0.1917, 0.1822, 0.1981], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 03:51:25,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.102e+02 1.372e+03 1.677e+03 2.104e+03 6.660e+03, threshold=3.354e+03, percent-clipped=14.0
+2023-03-12 03:51:59,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4071, 3.1921, 1.5609, 1.5399], device='cuda:0'), covar=tensor([0.1014, 0.0318, 0.0850, 0.1348], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0552, 0.0392, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 03:52:05,098 INFO [train.py:968] (0/2) Epoch 23, batch 36400, giga_loss[loss=0.4074, simple_loss=0.4227, pruned_loss=0.196, over 26558.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3543, pruned_loss=0.1014, over 5694011.23 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3498, pruned_loss=0.1101, over 5699983.38 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3532, pruned_loss=0.09942, over 5701377.45 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:52:06,183 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1039792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:07,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1039795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:20,792 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1039807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:23,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1039810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:34,267 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1039824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:47,121 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1039839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:48,365 INFO [train.py:968] (0/2) Epoch 23, batch 36450, libri_loss[loss=0.2316, simple_loss=0.3081, pruned_loss=0.07756, over 29458.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3562, pruned_loss=0.1049, over 5692131.18 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3497, pruned_loss=0.11, over 5698898.13 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3556, pruned_loss=0.1033, over 5698490.90 frames. ], batch size: 70, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:52:55,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.925e+02 1.464e+03 1.807e+03 2.375e+03 6.931e+03, threshold=3.614e+03, percent-clipped=14.0
+2023-03-12 03:53:36,942 INFO [train.py:968] (0/2) Epoch 23, batch 36500, giga_loss[loss=0.2777, simple_loss=0.3506, pruned_loss=0.1024, over 28599.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3557, pruned_loss=0.1055, over 5678528.73 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3501, pruned_loss=0.1103, over 5682513.90 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3549, pruned_loss=0.104, over 5698831.54 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:53:37,263 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7436, 1.7112, 1.9154, 1.5145], device='cuda:0'), covar=tensor([0.1657, 0.2458, 0.1353, 0.1613], device='cuda:0'), in_proj_covar=tensor([0.0908, 0.0699, 0.0955, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 03:54:21,363 INFO [train.py:968] (0/2) Epoch 23, batch 36550, giga_loss[loss=0.2458, simple_loss=0.3067, pruned_loss=0.09244, over 23849.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3532, pruned_loss=0.1045, over 5679770.09 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3506, pruned_loss=0.1104, over 5683336.87 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3521, pruned_loss=0.103, over 5695328.55 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:54:22,780 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1039943.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 03:54:27,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.611e+02 1.284e+03 1.792e+03 2.507e+03 5.777e+03, threshold=3.585e+03, percent-clipped=11.0
+2023-03-12 03:55:06,150 INFO [train.py:968] (0/2) Epoch 23, batch 36600, libri_loss[loss=0.3271, simple_loss=0.3892, pruned_loss=0.1326, over 29368.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3525, pruned_loss=0.1041, over 5688563.54 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3506, pruned_loss=0.1105, over 5687279.11 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3517, pruned_loss=0.1028, over 5697288.10 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:55:14,267 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1040000.pt
+2023-03-12 03:55:42,214 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-12 03:55:51,295 INFO [train.py:968] (0/2) Epoch 23, batch 36650, giga_loss[loss=0.2463, simple_loss=0.3328, pruned_loss=0.07991, over 28956.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3512, pruned_loss=0.1023, over 5682398.63 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.351, pruned_loss=0.1107, over 5680374.78 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3502, pruned_loss=0.101, over 5694663.12 frames. ], batch size: 136, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:55:59,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.223e+02 1.247e+03 1.559e+03 2.063e+03 5.173e+03, threshold=3.118e+03, percent-clipped=5.0
+2023-03-12 03:56:01,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7460, 2.5861, 1.7397, 0.9114], device='cuda:0'), covar=tensor([0.9235, 0.3670, 0.4129, 0.7698], device='cuda:0'), in_proj_covar=tensor([0.1773, 0.1677, 0.1611, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 03:56:39,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2476, 3.0968, 2.9179, 1.3341], device='cuda:0'), covar=tensor([0.0972, 0.1073, 0.0894, 0.2524], device='cuda:0'), in_proj_covar=tensor([0.1237, 0.1142, 0.0969, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 03:56:41,295 INFO [train.py:968] (0/2) Epoch 23, batch 36700, giga_loss[loss=0.2564, simple_loss=0.3313, pruned_loss=0.09071, over 28712.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3494, pruned_loss=0.101, over 5673661.35 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3512, pruned_loss=0.1107, over 5682303.95 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3484, pruned_loss=0.0998, over 5681636.41 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:57:29,407 INFO [train.py:968] (0/2) Epoch 23, batch 36750, giga_loss[loss=0.2167, simple_loss=0.2954, pruned_loss=0.069, over 28923.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3427, pruned_loss=0.09743, over 5664606.86 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3514, pruned_loss=0.1107, over 5683391.43 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3418, pruned_loss=0.09639, over 5669923.70 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:57:35,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.753e+02 1.127e+03 1.364e+03 1.725e+03 3.518e+03, threshold=2.727e+03, percent-clipped=3.0
+2023-03-12 03:58:13,300 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 03:58:19,839 INFO [train.py:968] (0/2) Epoch 23, batch 36800, giga_loss[loss=0.2552, simple_loss=0.326, pruned_loss=0.09222, over 28291.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3367, pruned_loss=0.09478, over 5658407.83 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3513, pruned_loss=0.1108, over 5689110.81 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3358, pruned_loss=0.09359, over 5657222.58 frames. ], batch size: 368, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:59:11,771 INFO [train.py:968] (0/2) Epoch 23, batch 36850, libri_loss[loss=0.3086, simple_loss=0.379, pruned_loss=0.1191, over 29673.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3335, pruned_loss=0.09339, over 5657677.31 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3515, pruned_loss=0.1108, over 5692526.73 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3324, pruned_loss=0.09219, over 5653282.81 frames. ], batch size: 91, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:59:18,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.282e+02 1.019e+03 1.222e+03 1.579e+03 4.296e+03, threshold=2.444e+03, percent-clipped=6.0
+2023-03-12 03:59:54,838 INFO [train.py:968] (0/2) Epoch 23, batch 36900, giga_loss[loss=0.2454, simple_loss=0.3264, pruned_loss=0.08218, over 28396.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3346, pruned_loss=0.09314, over 5675618.24 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.352, pruned_loss=0.1109, over 5698536.42 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3328, pruned_loss=0.09164, over 5665874.85 frames. ], batch size: 65, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:00:09,422 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2761, 1.6149, 1.3122, 0.9477], device='cuda:0'), covar=tensor([0.2373, 0.2348, 0.2564, 0.2435], device='cuda:0'), in_proj_covar=tensor([0.1519, 0.1097, 0.1344, 0.0991], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 04:00:18,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1040318.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:00:34,474 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1040336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:00:39,160 INFO [train.py:968] (0/2) Epoch 23, batch 36950, giga_loss[loss=0.2458, simple_loss=0.3265, pruned_loss=0.08252, over 28900.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3344, pruned_loss=0.0929, over 5677837.81 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3521, pruned_loss=0.1109, over 5700492.46 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3327, pruned_loss=0.09146, over 5667859.18 frames. ], batch size: 106, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:00:46,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.413e+02 1.156e+03 1.466e+03 2.011e+03 4.955e+03, threshold=2.932e+03, percent-clipped=20.0
+2023-03-12 04:00:55,065 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 04:01:22,880 INFO [train.py:968] (0/2) Epoch 23, batch 37000, libri_loss[loss=0.2937, simple_loss=0.3701, pruned_loss=0.1087, over 29525.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3327, pruned_loss=0.09163, over 5696074.39 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3526, pruned_loss=0.111, over 5706670.07 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3303, pruned_loss=0.08982, over 5682119.95 frames. ], batch size: 89, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:02:04,076 INFO [train.py:968] (0/2) Epoch 23, batch 37050, giga_loss[loss=0.254, simple_loss=0.3291, pruned_loss=0.08942, over 28988.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3313, pruned_loss=0.09103, over 5701768.56 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3535, pruned_loss=0.1113, over 5701289.12 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3277, pruned_loss=0.08855, over 5695918.56 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:02:11,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.543e+02 1.145e+03 1.558e+03 2.171e+03 7.134e+03, threshold=3.115e+03, percent-clipped=15.0
+2023-03-12 04:02:21,175 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1040461.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:02:23,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1040464.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:02:44,734 INFO [train.py:968] (0/2) Epoch 23, batch 37100, giga_loss[loss=0.2261, simple_loss=0.2985, pruned_loss=0.0768, over 28543.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3299, pruned_loss=0.09055, over 5703354.47 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3538, pruned_loss=0.1113, over 5702475.45 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3262, pruned_loss=0.08811, over 5697891.96 frames. ], batch size: 78, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:02:46,594 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1040493.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:02:55,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3194, 3.1307, 3.0010, 1.3519], device='cuda:0'), covar=tensor([0.0975, 0.1119, 0.0875, 0.2344], device='cuda:0'), in_proj_covar=tensor([0.1233, 0.1138, 0.0962, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 04:03:01,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1040511.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:03:06,168 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1349, 1.2367, 1.1257, 0.8758], device='cuda:0'), covar=tensor([0.1062, 0.0556, 0.1105, 0.1102], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0444, 0.0521, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 04:03:23,832 INFO [train.py:968] (0/2) Epoch 23, batch 37150, giga_loss[loss=0.2427, simple_loss=0.3158, pruned_loss=0.08478, over 28868.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3277, pruned_loss=0.08966, over 5712554.91 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3539, pruned_loss=0.1111, over 5708233.65 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3239, pruned_loss=0.08731, over 5703011.09 frames. ], batch size: 112, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:03:31,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.071e+02 1.179e+03 1.482e+03 2.186e+03 4.493e+03, threshold=2.963e+03, percent-clipped=6.0
+2023-03-12 04:04:07,266 INFO [train.py:968] (0/2) Epoch 23, batch 37200, giga_loss[loss=0.2455, simple_loss=0.3199, pruned_loss=0.08549, over 28940.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3256, pruned_loss=0.08884, over 5715967.22 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3539, pruned_loss=0.111, over 5708623.42 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3221, pruned_loss=0.08665, over 5708185.35 frames. ], batch size: 136, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:04:25,820 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 04:04:44,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6045, 1.9048, 1.8804, 1.5911], device='cuda:0'), covar=tensor([0.1555, 0.1508, 0.1837, 0.1731], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0752, 0.0720, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 04:04:45,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7686, 1.7943, 1.7603, 1.5672], device='cuda:0'), covar=tensor([0.2770, 0.2622, 0.2040, 0.2790], device='cuda:0'), in_proj_covar=tensor([0.1990, 0.1912, 0.1828, 0.1992], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 04:04:48,771 INFO [train.py:968] (0/2) Epoch 23, batch 37250, giga_loss[loss=0.2387, simple_loss=0.3125, pruned_loss=0.08239, over 28913.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3234, pruned_loss=0.08766, over 5703637.19 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3545, pruned_loss=0.1113, over 5691094.12 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3196, pruned_loss=0.08526, over 5711891.40 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:04:56,717 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.652e+02 1.067e+03 1.317e+03 1.877e+03 6.706e+03, threshold=2.634e+03, percent-clipped=8.0
+2023-03-12 04:05:29,256 INFO [train.py:968] (0/2) Epoch 23, batch 37300, giga_loss[loss=0.2022, simple_loss=0.2865, pruned_loss=0.05895, over 28842.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3221, pruned_loss=0.08708, over 5708211.14 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3553, pruned_loss=0.1116, over 5694341.99 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3177, pruned_loss=0.08441, over 5712015.77 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:05:44,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1040711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:06:09,461 INFO [train.py:968] (0/2) Epoch 23, batch 37350, giga_loss[loss=0.2558, simple_loss=0.333, pruned_loss=0.08934, over 28705.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3204, pruned_loss=0.08591, over 5702283.26 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3558, pruned_loss=0.1117, over 5688090.33 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.3159, pruned_loss=0.08324, over 5711605.75 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:06:17,687 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.049e+02 1.126e+03 1.362e+03 1.761e+03 5.803e+03, threshold=2.725e+03, percent-clipped=11.0
+2023-03-12 04:06:49,352 INFO [train.py:968] (0/2) Epoch 23, batch 37400, giga_loss[loss=0.2294, simple_loss=0.3064, pruned_loss=0.07622, over 29036.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3206, pruned_loss=0.08624, over 5706041.11 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3562, pruned_loss=0.1118, over 5695833.98 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3154, pruned_loss=0.08316, over 5706881.30 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:07:30,028 INFO [train.py:968] (0/2) Epoch 23, batch 37450, giga_loss[loss=0.2835, simple_loss=0.3552, pruned_loss=0.1059, over 28599.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3207, pruned_loss=0.08594, over 5707016.95 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3565, pruned_loss=0.1117, over 5691343.28 frames. ], giga_tot_loss[loss=0.2403, simple_loss=0.3151, pruned_loss=0.08276, over 5712752.19 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:07:30,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3085, 3.1490, 2.9549, 1.4710], device='cuda:0'), covar=tensor([0.0954, 0.1072, 0.0903, 0.2415], device='cuda:0'), in_proj_covar=tensor([0.1236, 0.1140, 0.0966, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 04:07:38,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.089e+02 1.208e+03 1.886e+03 2.811e+03 1.124e+04, threshold=3.773e+03, percent-clipped=26.0
+2023-03-12 04:07:42,545 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1040854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:07:46,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1040857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:08:10,466 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1040886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:08:10,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1040886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:08:13,720 INFO [train.py:968] (0/2) Epoch 23, batch 37500, libri_loss[loss=0.2832, simple_loss=0.3566, pruned_loss=0.1049, over 29555.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3247, pruned_loss=0.0884, over 5710548.20 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.357, pruned_loss=0.1119, over 5693069.49 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3188, pruned_loss=0.08497, over 5714026.96 frames. ], batch size: 77, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:08:57,152 INFO [train.py:968] (0/2) Epoch 23, batch 37550, giga_loss[loss=0.2832, simple_loss=0.3574, pruned_loss=0.1045, over 28900.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3313, pruned_loss=0.09261, over 5707261.17 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3573, pruned_loss=0.1119, over 5695791.30 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3255, pruned_loss=0.0892, over 5708146.14 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:09:05,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-12 04:09:06,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.266e+02 1.241e+03 1.518e+03 2.210e+03 6.156e+03, threshold=3.037e+03, percent-clipped=7.0
+2023-03-12 04:09:09,676 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-12 04:09:36,499 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-12 04:09:41,043 INFO [train.py:968] (0/2) Epoch 23, batch 37600, giga_loss[loss=0.273, simple_loss=0.3482, pruned_loss=0.09893, over 28961.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3395, pruned_loss=0.09841, over 5682168.25 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3571, pruned_loss=0.1118, over 5679894.18 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3339, pruned_loss=0.09503, over 5697688.09 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:09:50,013 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:10:18,693 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041029.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:10:20,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041032.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:10:30,921 INFO [train.py:968] (0/2) Epoch 23, batch 37650, giga_loss[loss=0.2645, simple_loss=0.3512, pruned_loss=0.08886, over 29020.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3448, pruned_loss=0.1013, over 5677277.17 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3577, pruned_loss=0.1123, over 5685080.97 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3393, pruned_loss=0.0978, over 5684864.38 frames. ], batch size: 164, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:10:41,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.747e+02 1.240e+03 1.706e+03 2.357e+03 7.046e+03, threshold=3.412e+03, percent-clipped=9.0
+2023-03-12 04:10:51,228 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041061.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:11:20,808 INFO [train.py:968] (0/2) Epoch 23, batch 37700, giga_loss[loss=0.3301, simple_loss=0.3801, pruned_loss=0.14, over 23722.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3492, pruned_loss=0.1026, over 5677026.86 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3579, pruned_loss=0.1124, over 5682477.57 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3447, pruned_loss=0.09972, over 5685444.50 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:11:57,417 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:12:01,420 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041133.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:12:08,217 INFO [train.py:968] (0/2) Epoch 23, batch 37750, giga_loss[loss=0.2676, simple_loss=0.3489, pruned_loss=0.0932, over 28720.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.355, pruned_loss=0.1058, over 5669799.24 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3581, pruned_loss=0.1125, over 5678291.36 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3512, pruned_loss=0.1033, over 5679965.08 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:12:17,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.374e+02 1.179e+03 1.452e+03 2.000e+03 6.966e+03, threshold=2.904e+03, percent-clipped=8.0
+2023-03-12 04:12:52,000 INFO [train.py:968] (0/2) Epoch 23, batch 37800, giga_loss[loss=0.2414, simple_loss=0.3205, pruned_loss=0.08116, over 28364.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3557, pruned_loss=0.1058, over 5679480.96 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3576, pruned_loss=0.1123, over 5684214.52 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.353, pruned_loss=0.1037, over 5682106.65 frames. ], batch size: 65, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:13:29,429 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5541, 1.8801, 1.4601, 1.6684], device='cuda:0'), covar=tensor([0.2946, 0.2734, 0.3146, 0.2353], device='cuda:0'), in_proj_covar=tensor([0.1525, 0.1103, 0.1350, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 04:13:32,367 INFO [train.py:968] (0/2) Epoch 23, batch 37850, giga_loss[loss=0.2483, simple_loss=0.3314, pruned_loss=0.08265, over 28251.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3515, pruned_loss=0.1023, over 5689999.16 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3578, pruned_loss=0.1125, over 5688685.59 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3492, pruned_loss=0.1005, over 5688069.20 frames. ], batch size: 368, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:13:41,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.169e+02 1.247e+03 1.592e+03 2.049e+03 4.156e+03, threshold=3.184e+03, percent-clipped=8.0
+2023-03-12 04:13:51,571 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041264.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:13:56,438 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:14:13,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0099, 4.8304, 4.5959, 2.0983], device='cuda:0'), covar=tensor([0.0489, 0.0642, 0.0637, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1237, 0.1145, 0.0966, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 04:14:15,995 INFO [train.py:968] (0/2) Epoch 23, batch 37900, giga_loss[loss=0.2727, simple_loss=0.3465, pruned_loss=0.0995, over 28811.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3499, pruned_loss=0.1006, over 5695565.66 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3583, pruned_loss=0.1129, over 5694228.79 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3474, pruned_loss=0.09852, over 5689180.69 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:14:23,107 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-12 04:14:53,971 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:15:01,576 INFO [train.py:968] (0/2) Epoch 23, batch 37950, giga_loss[loss=0.2898, simple_loss=0.3644, pruned_loss=0.1076, over 28710.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3497, pruned_loss=0.1002, over 5690805.56 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3588, pruned_loss=0.1134, over 5684338.70 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3472, pruned_loss=0.09791, over 5695454.90 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:15:11,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.440e+02 1.232e+03 1.481e+03 2.108e+03 6.004e+03, threshold=2.963e+03, percent-clipped=9.0
+2023-03-12 04:15:20,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3967, 1.5975, 1.4890, 1.5064], device='cuda:0'), covar=tensor([0.0810, 0.0340, 0.0325, 0.0891], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 04:15:29,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:15:44,579 INFO [train.py:968] (0/2) Epoch 23, batch 38000, libri_loss[loss=0.3071, simple_loss=0.383, pruned_loss=0.1156, over 29529.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3517, pruned_loss=0.1011, over 5692935.20 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3591, pruned_loss=0.1135, over 5687088.17 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3491, pruned_loss=0.09887, over 5694216.66 frames. ], batch size: 82, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:16:16,122 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-12 04:16:26,403 INFO [train.py:968] (0/2) Epoch 23, batch 38050, giga_loss[loss=0.3232, simple_loss=0.3901, pruned_loss=0.1281, over 28631.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3534, pruned_loss=0.1024, over 5694276.22 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3593, pruned_loss=0.1137, over 5681239.42 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3511, pruned_loss=0.1002, over 5699994.85 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:16:36,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.987e+02 1.354e+03 1.875e+03 2.469e+03 7.375e+03, threshold=3.749e+03, percent-clipped=17.0
+2023-03-12 04:17:08,266 INFO [train.py:968] (0/2) Epoch 23, batch 38100, libri_loss[loss=0.2534, simple_loss=0.3293, pruned_loss=0.08874, over 29543.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3553, pruned_loss=0.104, over 5696793.54 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3591, pruned_loss=0.1132, over 5686768.09 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3535, pruned_loss=0.1023, over 5696943.99 frames. ], batch size: 79, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:17:12,723 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-12 04:17:20,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:17:23,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041508.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:17:33,881 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:17:35,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:17:45,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6412, 4.4828, 4.2378, 2.1184], device='cuda:0'), covar=tensor([0.0486, 0.0611, 0.0644, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.1235, 0.1144, 0.0967, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 04:17:52,031 INFO [train.py:968] (0/2) Epoch 23, batch 38150, giga_loss[loss=0.275, simple_loss=0.3508, pruned_loss=0.09958, over 28833.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3571, pruned_loss=0.1057, over 5697935.95 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3596, pruned_loss=0.1135, over 5692943.96 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.355, pruned_loss=0.1038, over 5692618.04 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:18:01,143 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:18:03,319 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.995e+02 1.451e+03 1.828e+03 2.430e+03 1.139e+04, threshold=3.656e+03, percent-clipped=7.0
+2023-03-12 04:18:35,238 INFO [train.py:968] (0/2) Epoch 23, batch 38200, giga_loss[loss=0.2973, simple_loss=0.3683, pruned_loss=0.1131, over 28734.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3582, pruned_loss=0.1067, over 5694285.43 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3599, pruned_loss=0.1137, over 5687067.87 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3563, pruned_loss=0.1048, over 5695865.04 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:19:05,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:19:18,463 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041639.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:19:19,528 INFO [train.py:968] (0/2) Epoch 23, batch 38250, libri_loss[loss=0.2705, simple_loss=0.3371, pruned_loss=0.1019, over 29358.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.358, pruned_loss=0.1067, over 5692058.58 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3599, pruned_loss=0.1136, over 5688084.83 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3565, pruned_loss=0.1051, over 5691955.20 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:19:21,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:19:25,040 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:19:28,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:19:28,652 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041651.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:19:29,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.836e+02 1.251e+03 1.557e+03 2.040e+03 5.238e+03, threshold=3.113e+03, percent-clipped=3.0
+2023-03-12 04:19:30,709 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041654.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:19:48,789 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2472, 2.4740, 1.8111, 1.9959], device='cuda:0'), covar=tensor([0.0962, 0.0665, 0.0993, 0.1125], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0445, 0.0522, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 04:19:50,984 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:19:53,534 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041683.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:20:01,748 INFO [train.py:968] (0/2) Epoch 23, batch 38300, giga_loss[loss=0.2726, simple_loss=0.3581, pruned_loss=0.09357, over 28959.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3567, pruned_loss=0.1048, over 5700871.11 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3593, pruned_loss=0.1132, over 5691609.10 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.356, pruned_loss=0.1038, over 5697932.86 frames. ], batch size: 136, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:20:16,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:20:45,568 INFO [train.py:968] (0/2) Epoch 23, batch 38350, giga_loss[loss=0.2681, simple_loss=0.3618, pruned_loss=0.08721, over 28923.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3567, pruned_loss=0.1037, over 5706101.70 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3591, pruned_loss=0.1132, over 5694811.44 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3561, pruned_loss=0.1028, over 5701052.72 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:20:55,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.777e+02 1.142e+03 1.396e+03 1.740e+03 7.337e+03, threshold=2.793e+03, percent-clipped=3.0
+2023-03-12 04:21:18,090 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041782.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:21:21,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041785.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:21:22,249 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:21:25,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:21:26,315 INFO [train.py:968] (0/2) Epoch 23, batch 38400, giga_loss[loss=0.2654, simple_loss=0.3376, pruned_loss=0.09654, over 28532.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3561, pruned_loss=0.1038, over 5693774.34 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3596, pruned_loss=0.1135, over 5687014.67 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3552, pruned_loss=0.1024, over 5696758.32 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:21:43,640 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041814.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:21:46,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:22:06,129 INFO [train.py:968] (0/2) Epoch 23, batch 38450, giga_loss[loss=0.2452, simple_loss=0.326, pruned_loss=0.08223, over 28567.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3544, pruned_loss=0.1033, over 5700188.78 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3598, pruned_loss=0.1137, over 5694147.29 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3532, pruned_loss=0.1017, over 5696469.50 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:22:08,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9790, 2.4351, 1.4831, 2.0242], device='cuda:0'), covar=tensor([0.0988, 0.0587, 0.1057, 0.0978], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0445, 0.0522, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 04:22:12,425 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:22:13,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1813, 2.2021, 2.0356, 2.0246], device='cuda:0'), covar=tensor([0.1959, 0.2440, 0.2286, 0.2143], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0753, 0.0721, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 04:22:15,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:22:15,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.561e+02 1.185e+03 1.518e+03 2.583e+03 6.000e+03, threshold=3.037e+03, percent-clipped=17.0
+2023-03-12 04:22:38,933 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:22:46,741 INFO [train.py:968] (0/2) Epoch 23, batch 38500, giga_loss[loss=0.2707, simple_loss=0.3479, pruned_loss=0.0968, over 28972.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.351, pruned_loss=0.1012, over 5707012.28 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3593, pruned_loss=0.1133, over 5699004.17 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3504, pruned_loss=0.1, over 5700039.05 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:23:26,594 INFO [train.py:968] (0/2) Epoch 23, batch 38550, giga_loss[loss=0.2624, simple_loss=0.3363, pruned_loss=0.09428, over 28426.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3501, pruned_loss=0.1007, over 5701970.83 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3598, pruned_loss=0.1136, over 5692759.01 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.349, pruned_loss=0.09928, over 5703218.42 frames. ], batch size: 65, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:23:39,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.682e+02 1.128e+03 1.388e+03 1.808e+03 5.883e+03, threshold=2.776e+03, percent-clipped=5.0
+2023-03-12 04:24:07,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1600, 1.7325, 1.3302, 0.4443], device='cuda:0'), covar=tensor([0.4763, 0.2800, 0.4288, 0.5879], device='cuda:0'), in_proj_covar=tensor([0.1759, 0.1659, 0.1600, 0.1438], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 04:24:08,806 INFO [train.py:968] (0/2) Epoch 23, batch 38600, giga_loss[loss=0.2703, simple_loss=0.3516, pruned_loss=0.09453, over 28692.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3512, pruned_loss=0.1017, over 5697611.88 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3601, pruned_loss=0.114, over 5687456.82 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3497, pruned_loss=0.09981, over 5703567.47 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:24:15,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:24:15,854 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1042000.pt
+2023-03-12 04:24:48,061 INFO [train.py:968] (0/2) Epoch 23, batch 38650, giga_loss[loss=0.2473, simple_loss=0.3367, pruned_loss=0.07898, over 29135.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3508, pruned_loss=0.1012, over 5683377.74 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.36, pruned_loss=0.114, over 5670226.68 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3496, pruned_loss=0.09955, over 5703993.74 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:24:58,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.292e+02 1.224e+03 1.601e+03 2.276e+03 6.654e+03, threshold=3.202e+03, percent-clipped=16.0
+2023-03-12 04:25:04,630 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 04:25:08,482 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0565, 3.9008, 3.6599, 1.8765], device='cuda:0'), covar=tensor([0.0625, 0.0758, 0.0733, 0.2164], device='cuda:0'), in_proj_covar=tensor([0.1236, 0.1141, 0.0967, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 04:25:14,309 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5485, 3.0825, 1.7055, 1.6212], device='cuda:0'), covar=tensor([0.0811, 0.0262, 0.0760, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0551, 0.0390, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 04:25:27,853 INFO [train.py:968] (0/2) Epoch 23, batch 38700, giga_loss[loss=0.2588, simple_loss=0.3371, pruned_loss=0.0902, over 28569.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3497, pruned_loss=0.09981, over 5693656.50 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.36, pruned_loss=0.1139, over 5675158.30 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3486, pruned_loss=0.09833, over 5705933.29 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:25:33,946 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4834, 1.8298, 1.3510, 1.2914], device='cuda:0'), covar=tensor([0.1132, 0.0557, 0.1045, 0.1168], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0444, 0.0521, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 04:25:51,337 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1042121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:26:08,943 INFO [train.py:968] (0/2) Epoch 23, batch 38750, giga_loss[loss=0.2578, simple_loss=0.3449, pruned_loss=0.08538, over 28933.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.349, pruned_loss=0.09886, over 5701159.06 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3603, pruned_loss=0.114, over 5676953.09 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3477, pruned_loss=0.09729, over 5710021.67 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:26:09,886 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1042142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:26:11,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1042145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:26:19,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.177e+02 1.031e+03 1.328e+03 1.801e+03 8.907e+03, threshold=2.656e+03, percent-clipped=7.0
+2023-03-12 04:26:38,294 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1042174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:26:50,819 INFO [train.py:968] (0/2) Epoch 23, batch 38800, giga_loss[loss=0.2144, simple_loss=0.2996, pruned_loss=0.06465, over 28595.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3486, pruned_loss=0.09917, over 5697854.25 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3604, pruned_loss=0.114, over 5677540.65 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3473, pruned_loss=0.09772, over 5704488.90 frames. ], batch size: 60, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:27:31,666 INFO [train.py:968] (0/2) Epoch 23, batch 38850, giga_loss[loss=0.2891, simple_loss=0.3572, pruned_loss=0.1105, over 28791.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3478, pruned_loss=0.0999, over 5691053.21 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3607, pruned_loss=0.1145, over 5675441.54 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3462, pruned_loss=0.09788, over 5698761.31 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:27:42,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.705e+02 1.272e+03 1.638e+03 2.177e+03 8.760e+03, threshold=3.276e+03, percent-clipped=17.0
+2023-03-12 04:27:42,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1042255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:28:10,118 INFO [train.py:968] (0/2) Epoch 23, batch 38900, giga_loss[loss=0.27, simple_loss=0.3372, pruned_loss=0.1013, over 28956.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3445, pruned_loss=0.09815, over 5689035.45 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3608, pruned_loss=0.1145, over 5669593.14 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3428, pruned_loss=0.09613, over 5700473.00 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:28:49,363 INFO [train.py:968] (0/2) Epoch 23, batch 38950, giga_loss[loss=0.2298, simple_loss=0.3085, pruned_loss=0.07551, over 28789.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3438, pruned_loss=0.09768, over 5692709.68 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3615, pruned_loss=0.1149, over 5667634.13 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3412, pruned_loss=0.09523, over 5704983.56 frames. ], batch size: 86, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:28:59,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.454e+02 1.236e+03 1.657e+03 2.500e+03 7.494e+03, threshold=3.313e+03, percent-clipped=13.0
+2023-03-12 04:29:31,990 INFO [train.py:968] (0/2) Epoch 23, batch 39000, giga_loss[loss=0.2301, simple_loss=0.3091, pruned_loss=0.07551, over 28943.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3425, pruned_loss=0.09728, over 5700759.03 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3606, pruned_loss=0.1142, over 5672893.92 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3408, pruned_loss=0.09541, over 5706615.79 frames. ], batch size: 136, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:29:31,996 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 04:29:40,812 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2045, simple_loss=0.3123, pruned_loss=0.0484, over 944034.00 frames. 
+2023-03-12 04:29:40,813 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 04:30:21,703 INFO [train.py:968] (0/2) Epoch 23, batch 39050, giga_loss[loss=0.2326, simple_loss=0.3091, pruned_loss=0.07809, over 28879.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3405, pruned_loss=0.09657, over 5708002.17 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3606, pruned_loss=0.1142, over 5675337.56 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3385, pruned_loss=0.09456, over 5711468.53 frames. ], batch size: 106, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:30:33,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.223e+02 1.185e+03 1.431e+03 1.889e+03 3.850e+03, threshold=2.862e+03, percent-clipped=1.0
+2023-03-12 04:30:52,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8807, 1.9582, 1.4333, 1.5864], device='cuda:0'), covar=tensor([0.0967, 0.0782, 0.1073, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0444, 0.0520, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 04:31:01,231 INFO [train.py:968] (0/2) Epoch 23, batch 39100, giga_loss[loss=0.2476, simple_loss=0.3243, pruned_loss=0.08545, over 28507.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3389, pruned_loss=0.09631, over 5708762.31 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3605, pruned_loss=0.1141, over 5683549.64 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3365, pruned_loss=0.094, over 5705039.67 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:31:06,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1042496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:31:40,265 INFO [train.py:968] (0/2) Epoch 23, batch 39150, giga_loss[loss=0.2161, simple_loss=0.2879, pruned_loss=0.07216, over 28793.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3358, pruned_loss=0.09475, over 5714588.67 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3601, pruned_loss=0.1137, over 5691744.27 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3334, pruned_loss=0.09253, over 5705130.36 frames. ], batch size: 99, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:31:49,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2139, 2.5723, 1.2264, 1.5018], device='cuda:0'), covar=tensor([0.0995, 0.0437, 0.0970, 0.1289], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0551, 0.0389, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 04:31:53,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.912e+02 1.183e+03 1.419e+03 1.849e+03 5.641e+03, threshold=2.838e+03, percent-clipped=8.0
+2023-03-12 04:32:17,555 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-12 04:32:24,337 INFO [train.py:968] (0/2) Epoch 23, batch 39200, giga_loss[loss=0.2378, simple_loss=0.3118, pruned_loss=0.08194, over 28946.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3344, pruned_loss=0.09405, over 5715371.50 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3604, pruned_loss=0.1139, over 5694580.08 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3317, pruned_loss=0.09181, over 5705585.71 frames. ], batch size: 106, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:32:58,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1042630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:33:06,673 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1042639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:33:08,178 INFO [train.py:968] (0/2) Epoch 23, batch 39250, libri_loss[loss=0.3182, simple_loss=0.3796, pruned_loss=0.1284, over 29526.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3352, pruned_loss=0.09448, over 5703021.77 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3606, pruned_loss=0.114, over 5680485.24 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3323, pruned_loss=0.09214, over 5708706.54 frames. ], batch size: 81, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:33:09,083 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1042642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:33:20,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.851e+02 1.108e+03 1.420e+03 1.879e+03 8.493e+03, threshold=2.840e+03, percent-clipped=11.0
+2023-03-12 04:33:35,046 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1042671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:33:44,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7641, 1.8623, 2.0028, 1.6346], device='cuda:0'), covar=tensor([0.1715, 0.1953, 0.1812, 0.1925], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0753, 0.0720, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 04:33:50,411 INFO [train.py:968] (0/2) Epoch 23, batch 39300, giga_loss[loss=0.2816, simple_loss=0.352, pruned_loss=0.1056, over 28473.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3381, pruned_loss=0.09543, over 5706166.71 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3608, pruned_loss=0.1139, over 5687059.09 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3346, pruned_loss=0.09285, over 5705845.82 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:34:30,254 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4182, 3.4175, 1.4926, 1.6073], device='cuda:0'), covar=tensor([0.0976, 0.0320, 0.0960, 0.1310], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0553, 0.0391, 0.0428], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 04:34:36,370 INFO [train.py:968] (0/2) Epoch 23, batch 39350, giga_loss[loss=0.2557, simple_loss=0.3456, pruned_loss=0.08293, over 28682.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3419, pruned_loss=0.09693, over 5699970.20 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3607, pruned_loss=0.1139, over 5682566.61 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3385, pruned_loss=0.09429, over 5705150.57 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:34:48,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.237e+02 1.198e+03 1.497e+03 2.082e+03 4.581e+03, threshold=2.994e+03, percent-clipped=13.0
+2023-03-12 04:35:04,301 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1042773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:35:06,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1042776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:35:08,191 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1042778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:35:22,012 INFO [train.py:968] (0/2) Epoch 23, batch 39400, giga_loss[loss=0.2562, simple_loss=0.3301, pruned_loss=0.09114, over 29004.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3455, pruned_loss=0.09861, over 5683873.90 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3608, pruned_loss=0.1141, over 5672026.00 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3424, pruned_loss=0.096, over 5698180.13 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:35:34,325 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1042805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:36:03,091 INFO [train.py:968] (0/2) Epoch 23, batch 39450, giga_loss[loss=0.2545, simple_loss=0.3405, pruned_loss=0.08422, over 28692.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3439, pruned_loss=0.09668, over 5677232.87 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1143, over 5666866.85 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3408, pruned_loss=0.09422, over 5693852.48 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:36:17,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.936e+02 1.152e+03 1.477e+03 2.376e+03 7.797e+03, threshold=2.953e+03, percent-clipped=10.0
+2023-03-12 04:36:31,496 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3552, 1.5389, 1.1584, 1.0885], device='cuda:0'), covar=tensor([0.1013, 0.0560, 0.1116, 0.1150], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0446, 0.0522, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 04:36:46,117 INFO [train.py:968] (0/2) Epoch 23, batch 39500, giga_loss[loss=0.2623, simple_loss=0.3419, pruned_loss=0.09138, over 28740.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3429, pruned_loss=0.09529, over 5687049.79 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3609, pruned_loss=0.114, over 5671304.19 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3403, pruned_loss=0.09327, over 5696609.01 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:36:48,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2664, 3.3836, 1.4201, 1.4184], device='cuda:0'), covar=tensor([0.1037, 0.0413, 0.0960, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0552, 0.0390, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 04:36:59,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5853, 1.8257, 1.8365, 1.6412], device='cuda:0'), covar=tensor([0.2079, 0.2308, 0.2277, 0.2385], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0754, 0.0721, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 04:37:01,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3293, 1.1398, 1.0872, 1.5114], device='cuda:0'), covar=tensor([0.0764, 0.0363, 0.0366, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:0')
+2023-03-12 04:37:29,849 INFO [train.py:968] (0/2) Epoch 23, batch 39550, giga_loss[loss=0.2643, simple_loss=0.3307, pruned_loss=0.09901, over 28726.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3441, pruned_loss=0.09633, over 5687653.63 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3613, pruned_loss=0.1141, over 5677388.86 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3412, pruned_loss=0.09417, over 5690122.31 frames. ], batch size: 99, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:37:39,991 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.984e+02 1.258e+03 1.492e+03 1.883e+03 3.836e+03, threshold=2.985e+03, percent-clipped=4.0
+2023-03-12 04:37:59,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5609, 2.8884, 1.5916, 1.6834], device='cuda:0'), covar=tensor([0.0761, 0.0290, 0.0749, 0.1059], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0551, 0.0390, 0.0427], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 04:38:07,899 INFO [train.py:968] (0/2) Epoch 23, batch 39600, giga_loss[loss=0.2829, simple_loss=0.3546, pruned_loss=0.1056, over 28572.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3445, pruned_loss=0.09694, over 5679313.97 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3613, pruned_loss=0.1141, over 5668923.37 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3414, pruned_loss=0.09436, over 5688996.67 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:38:53,887 INFO [train.py:968] (0/2) Epoch 23, batch 39650, giga_loss[loss=0.2738, simple_loss=0.3456, pruned_loss=0.101, over 28698.00 frames. ], tot_loss[loss=0.272, simple_loss=0.347, pruned_loss=0.09849, over 5680180.48 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3614, pruned_loss=0.1142, over 5659955.75 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3445, pruned_loss=0.09637, over 5695477.19 frames. ], batch size: 66, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:39:04,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.325e+02 1.274e+03 1.534e+03 1.942e+03 4.164e+03, threshold=3.068e+03, percent-clipped=9.0
+2023-03-12 04:39:34,917 INFO [train.py:968] (0/2) Epoch 23, batch 39700, giga_loss[loss=0.2687, simple_loss=0.358, pruned_loss=0.0897, over 28920.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3499, pruned_loss=0.1003, over 5690489.71 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.1141, over 5664235.22 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3475, pruned_loss=0.09819, over 5700069.90 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:40:17,588 INFO [train.py:968] (0/2) Epoch 23, batch 39750, giga_loss[loss=0.2467, simple_loss=0.3319, pruned_loss=0.08077, over 29014.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3508, pruned_loss=0.09995, over 5704387.08 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3612, pruned_loss=0.114, over 5665376.04 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.349, pruned_loss=0.09832, over 5711032.23 frames. ], batch size: 136, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:40:29,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1043153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:40:30,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.350e+02 1.282e+03 1.627e+03 2.148e+03 5.307e+03, threshold=3.254e+03, percent-clipped=8.0
+2023-03-12 04:40:59,684 INFO [train.py:968] (0/2) Epoch 23, batch 39800, giga_loss[loss=0.3457, simple_loss=0.4012, pruned_loss=0.1451, over 27648.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3521, pruned_loss=0.1009, over 5708242.15 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.1141, over 5670178.89 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3502, pruned_loss=0.09914, over 5710229.25 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:41:42,389 INFO [train.py:968] (0/2) Epoch 23, batch 39850, libri_loss[loss=0.3106, simple_loss=0.3733, pruned_loss=0.124, over 29539.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3512, pruned_loss=0.1003, over 5710753.28 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5674657.21 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3497, pruned_loss=0.09887, over 5708971.89 frames. ], batch size: 81, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:41:52,788 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.215e+02 1.468e+03 1.866e+03 2.563e+03 5.993e+03, threshold=3.732e+03, percent-clipped=14.0
+2023-03-12 04:42:21,560 INFO [train.py:968] (0/2) Epoch 23, batch 39900, giga_loss[loss=0.2916, simple_loss=0.3479, pruned_loss=0.1176, over 23983.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3512, pruned_loss=0.1006, over 5696763.54 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3618, pruned_loss=0.1141, over 5662111.65 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3492, pruned_loss=0.09877, over 5707674.18 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:42:24,889 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1043296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:42:26,813 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1043299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:42:52,010 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1043328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:43:02,519 INFO [train.py:968] (0/2) Epoch 23, batch 39950, giga_loss[loss=0.2577, simple_loss=0.3365, pruned_loss=0.08949, over 28688.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3491, pruned_loss=0.09951, over 5702117.64 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3617, pruned_loss=0.1141, over 5665494.21 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3473, pruned_loss=0.0979, over 5708203.71 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:43:16,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.326e+02 1.326e+03 1.656e+03 2.320e+03 1.115e+04, threshold=3.312e+03, percent-clipped=11.0
+2023-03-12 04:43:25,683 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1043368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:43:44,895 INFO [train.py:968] (0/2) Epoch 23, batch 40000, giga_loss[loss=0.2237, simple_loss=0.3001, pruned_loss=0.07366, over 28533.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3443, pruned_loss=0.09686, over 5701136.25 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.362, pruned_loss=0.1143, over 5658223.54 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3426, pruned_loss=0.09528, over 5713346.06 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:44:15,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8547, 1.2419, 1.3295, 1.0325], device='cuda:0'), covar=tensor([0.2201, 0.1447, 0.2455, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0755, 0.0723, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 04:44:19,242 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1043433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:44:26,950 INFO [train.py:968] (0/2) Epoch 23, batch 40050, giga_loss[loss=0.2326, simple_loss=0.305, pruned_loss=0.08014, over 28437.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3426, pruned_loss=0.09604, over 5699490.81 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3622, pruned_loss=0.1142, over 5660963.00 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3405, pruned_loss=0.09425, over 5708467.11 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:44:38,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.608e+02 1.103e+03 1.482e+03 1.942e+03 4.147e+03, threshold=2.964e+03, percent-clipped=4.0
+2023-03-12 04:45:05,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7602, 1.8930, 1.4874, 1.5206], device='cuda:0'), covar=tensor([0.1057, 0.0791, 0.1027, 0.1327], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0444, 0.0518, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 04:45:06,902 INFO [train.py:968] (0/2) Epoch 23, batch 40100, giga_loss[loss=0.2594, simple_loss=0.3493, pruned_loss=0.08471, over 28802.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3436, pruned_loss=0.09496, over 5703055.19 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3621, pruned_loss=0.1142, over 5661425.39 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3418, pruned_loss=0.09338, over 5710269.88 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:45:46,434 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4266, 4.2782, 4.0442, 1.8780], device='cuda:0'), covar=tensor([0.0691, 0.0813, 0.0757, 0.2166], device='cuda:0'), in_proj_covar=tensor([0.1238, 0.1141, 0.0966, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 04:45:50,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5939, 1.7433, 1.2225, 1.4056], device='cuda:0'), covar=tensor([0.0939, 0.0626, 0.1061, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0444, 0.0518, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 04:45:50,510 INFO [train.py:968] (0/2) Epoch 23, batch 40150, giga_loss[loss=0.2716, simple_loss=0.3502, pruned_loss=0.09653, over 28209.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3444, pruned_loss=0.09512, over 5697732.91 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5664046.79 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3429, pruned_loss=0.09367, over 5701865.41 frames. ], batch size: 368, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:46:03,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2449, 1.0733, 3.8077, 3.2141], device='cuda:0'), covar=tensor([0.1746, 0.3082, 0.0478, 0.1119], device='cuda:0'), in_proj_covar=tensor([0.0768, 0.0653, 0.0970, 0.0918], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 04:46:04,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.025e+02 1.153e+03 1.448e+03 1.848e+03 5.099e+03, threshold=2.896e+03, percent-clipped=3.0
+2023-03-12 04:46:32,392 INFO [train.py:968] (0/2) Epoch 23, batch 40200, giga_loss[loss=0.2957, simple_loss=0.3652, pruned_loss=0.1131, over 28873.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3431, pruned_loss=0.09475, over 5712796.54 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3618, pruned_loss=0.1138, over 5671818.95 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3413, pruned_loss=0.09309, over 5710461.23 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:47:13,244 INFO [train.py:968] (0/2) Epoch 23, batch 40250, giga_loss[loss=0.2744, simple_loss=0.3379, pruned_loss=0.1054, over 28892.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3412, pruned_loss=0.09477, over 5719767.97 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3614, pruned_loss=0.1135, over 5676626.34 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3396, pruned_loss=0.09325, over 5714394.59 frames. ], batch size: 106, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:47:28,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.903e+02 1.289e+03 1.658e+03 2.224e+03 5.396e+03, threshold=3.316e+03, percent-clipped=13.0
+2023-03-12 04:47:57,800 INFO [train.py:968] (0/2) Epoch 23, batch 40300, giga_loss[loss=0.2286, simple_loss=0.2951, pruned_loss=0.08104, over 28169.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3398, pruned_loss=0.09557, over 5700846.07 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3616, pruned_loss=0.1137, over 5658511.41 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.338, pruned_loss=0.09392, over 5713597.53 frames. ], batch size: 77, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:48:40,568 INFO [train.py:968] (0/2) Epoch 23, batch 40350, giga_loss[loss=0.2376, simple_loss=0.3058, pruned_loss=0.08474, over 28530.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3391, pruned_loss=0.09659, over 5695592.28 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.362, pruned_loss=0.114, over 5661663.05 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3371, pruned_loss=0.09483, over 5703480.93 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:48:42,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1043743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:48:55,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.459e+02 1.220e+03 1.496e+03 2.289e+03 7.242e+03, threshold=2.991e+03, percent-clipped=6.0
+2023-03-12 04:49:22,820 INFO [train.py:968] (0/2) Epoch 23, batch 40400, libri_loss[loss=0.3076, simple_loss=0.3698, pruned_loss=0.1227, over 29553.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3386, pruned_loss=0.09662, over 5694147.77 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.362, pruned_loss=0.1139, over 5657143.79 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3363, pruned_loss=0.09475, over 5704747.96 frames. ], batch size: 78, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:49:36,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1043808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:49:59,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0608, 2.2039, 1.6140, 1.8129], device='cuda:0'), covar=tensor([0.0975, 0.0747, 0.1073, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0396, 0.0446, 0.0520, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 04:50:03,269 INFO [train.py:968] (0/2) Epoch 23, batch 40450, giga_loss[loss=0.2196, simple_loss=0.2929, pruned_loss=0.07311, over 28432.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3356, pruned_loss=0.09515, over 5695993.93 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3618, pruned_loss=0.1138, over 5655834.93 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3333, pruned_loss=0.09334, over 5706574.59 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:50:08,757 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1043848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:50:18,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.259e+02 1.286e+03 1.656e+03 2.372e+03 5.403e+03, threshold=3.311e+03, percent-clipped=14.0
+2023-03-12 04:50:21,207 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7500, 5.1160, 1.8300, 2.1263], device='cuda:0'), covar=tensor([0.0888, 0.0362, 0.0852, 0.1102], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0555, 0.0391, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 04:50:40,429 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1043886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:50:42,368 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1043889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:50:43,437 INFO [train.py:968] (0/2) Epoch 23, batch 40500, giga_loss[loss=0.3161, simple_loss=0.3732, pruned_loss=0.1295, over 26682.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.332, pruned_loss=0.09366, over 5693758.52 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3618, pruned_loss=0.1139, over 5654549.11 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3294, pruned_loss=0.09156, over 5704211.06 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:50:50,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4558, 3.1305, 1.5527, 1.6447], device='cuda:0'), covar=tensor([0.0981, 0.0357, 0.0958, 0.1261], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0555, 0.0391, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 04:51:05,245 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1043918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:51:24,336 INFO [train.py:968] (0/2) Epoch 23, batch 40550, giga_loss[loss=0.2486, simple_loss=0.3191, pruned_loss=0.08905, over 28883.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3294, pruned_loss=0.09237, over 5704473.81 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3621, pruned_loss=0.1142, over 5661272.17 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3261, pruned_loss=0.08981, over 5707957.86 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:51:32,509 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1043951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:51:35,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1043954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:51:40,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.407e+02 1.182e+03 1.524e+03 1.861e+03 5.214e+03, threshold=3.048e+03, percent-clipped=4.0
+2023-03-12 04:51:58,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1043983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:52:05,276 INFO [train.py:968] (0/2) Epoch 23, batch 40600, giga_loss[loss=0.297, simple_loss=0.3699, pruned_loss=0.112, over 28655.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3307, pruned_loss=0.09272, over 5695903.16 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3626, pruned_loss=0.1147, over 5650502.47 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3264, pruned_loss=0.08936, over 5710509.71 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:52:13,158 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1044000.pt
+2023-03-12 04:52:13,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9196, 1.1387, 1.0360, 0.8499], device='cuda:0'), covar=tensor([0.2201, 0.2542, 0.1658, 0.2213], device='cuda:0'), in_proj_covar=tensor([0.1996, 0.1937, 0.1861, 0.2007], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 04:52:35,225 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-12 04:52:44,720 INFO [train.py:968] (0/2) Epoch 23, batch 40650, giga_loss[loss=0.2689, simple_loss=0.34, pruned_loss=0.09886, over 28274.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3351, pruned_loss=0.09472, over 5702217.46 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.362, pruned_loss=0.1142, over 5659845.08 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.331, pruned_loss=0.09172, over 5707070.24 frames. ], batch size: 77, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:53:02,319 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5486, 2.3090, 1.7618, 0.7492], device='cuda:0'), covar=tensor([0.6292, 0.2842, 0.3929, 0.6820], device='cuda:0'), in_proj_covar=tensor([0.1768, 0.1666, 0.1608, 0.1439], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 04:53:03,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.965e+02 1.338e+03 1.624e+03 2.183e+03 5.538e+03, threshold=3.247e+03, percent-clipped=10.0
+2023-03-12 04:53:30,226 INFO [train.py:968] (0/2) Epoch 23, batch 40700, giga_loss[loss=0.3139, simple_loss=0.3767, pruned_loss=0.1255, over 28971.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3374, pruned_loss=0.09555, over 5708428.91 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3616, pruned_loss=0.1139, over 5662520.11 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3343, pruned_loss=0.09319, over 5710247.02 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:53:39,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3691, 1.5602, 1.5597, 1.5021], device='cuda:0'), covar=tensor([0.0759, 0.0323, 0.0311, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 04:53:41,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1044106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:53:44,452 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1044110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:54:10,368 INFO [train.py:968] (0/2) Epoch 23, batch 40750, giga_loss[loss=0.2612, simple_loss=0.3413, pruned_loss=0.09058, over 28863.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3414, pruned_loss=0.09725, over 5700719.67 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3623, pruned_loss=0.1144, over 5664959.00 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3377, pruned_loss=0.09454, over 5701037.96 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:54:28,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.370e+02 1.328e+03 1.523e+03 2.207e+03 6.962e+03, threshold=3.046e+03, percent-clipped=10.0
+2023-03-12 04:54:32,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5601, 1.6099, 1.7374, 1.3015], device='cuda:0'), covar=tensor([0.1815, 0.2609, 0.1514, 0.1807], device='cuda:0'), in_proj_covar=tensor([0.0908, 0.0701, 0.0953, 0.0854], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 04:54:33,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3666, 1.5804, 1.1801, 1.1230], device='cuda:0'), covar=tensor([0.0943, 0.0552, 0.1077, 0.1152], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0446, 0.0520, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 04:54:53,028 INFO [train.py:968] (0/2) Epoch 23, batch 40800, giga_loss[loss=0.2575, simple_loss=0.3418, pruned_loss=0.08665, over 28657.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3436, pruned_loss=0.09821, over 5709009.07 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3619, pruned_loss=0.1141, over 5670027.35 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3404, pruned_loss=0.0957, over 5705739.40 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:55:20,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1044223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:55:23,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4440, 1.6443, 1.7113, 1.2401], device='cuda:0'), covar=tensor([0.1678, 0.2453, 0.1435, 0.1678], device='cuda:0'), in_proj_covar=tensor([0.0909, 0.0702, 0.0954, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 04:55:36,756 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1044240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:55:37,229 INFO [train.py:968] (0/2) Epoch 23, batch 40850, giga_loss[loss=0.2701, simple_loss=0.3385, pruned_loss=0.1008, over 28951.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3459, pruned_loss=0.0998, over 5710487.41 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3614, pruned_loss=0.1139, over 5673476.57 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3435, pruned_loss=0.09784, over 5705444.23 frames. ], batch size: 106, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:55:58,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4252, 3.1910, 1.2816, 1.6004], device='cuda:0'), covar=tensor([0.1099, 0.0508, 0.1039, 0.1444], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0555, 0.0392, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 04:56:02,412 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4158, 1.6935, 1.4505, 1.5455], device='cuda:0'), covar=tensor([0.0792, 0.0316, 0.0347, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 04:56:02,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.384e+02 1.386e+03 1.852e+03 2.588e+03 5.669e+03, threshold=3.705e+03, percent-clipped=13.0
+2023-03-12 04:56:23,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5680, 1.8216, 1.5042, 1.8804], device='cuda:0'), covar=tensor([0.2141, 0.2204, 0.2276, 0.2073], device='cuda:0'), in_proj_covar=tensor([0.1536, 0.1106, 0.1356, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 04:56:36,175 INFO [train.py:968] (0/2) Epoch 23, batch 40900, giga_loss[loss=0.2657, simple_loss=0.337, pruned_loss=0.09717, over 28216.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3532, pruned_loss=0.1059, over 5704839.97 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3614, pruned_loss=0.1139, over 5673476.57 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3513, pruned_loss=0.1044, over 5700914.80 frames. ], batch size: 77, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:57:24,965 INFO [train.py:968] (0/2) Epoch 23, batch 40950, giga_loss[loss=0.409, simple_loss=0.4429, pruned_loss=0.1876, over 26635.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3607, pruned_loss=0.1115, over 5694568.67 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3609, pruned_loss=0.1136, over 5675993.56 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3596, pruned_loss=0.1105, over 5689720.00 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:57:38,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.144e+03 1.697e+03 2.127e+03 2.658e+03 5.908e+03, threshold=4.253e+03, percent-clipped=10.0
+2023-03-12 04:57:46,818 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1044366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:57:49,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1044369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:58:06,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1443, 1.5383, 1.4374, 1.3913], device='cuda:0'), covar=tensor([0.1612, 0.1221, 0.1836, 0.1308], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0750, 0.0719, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 04:58:09,105 INFO [train.py:968] (0/2) Epoch 23, batch 41000, libri_loss[loss=0.2963, simple_loss=0.3623, pruned_loss=0.1151, over 29540.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3656, pruned_loss=0.1151, over 5676882.07 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3609, pruned_loss=0.1135, over 5658109.18 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3648, pruned_loss=0.1143, over 5689035.57 frames. ], batch size: 76, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:58:14,052 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1044398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:58:14,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5921, 1.8558, 1.7915, 1.5845], device='cuda:0'), covar=tensor([0.1615, 0.1701, 0.1947, 0.1812], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0751, 0.0719, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 04:58:50,666 INFO [train.py:968] (0/2) Epoch 23, batch 41050, giga_loss[loss=0.3941, simple_loss=0.4361, pruned_loss=0.1761, over 27954.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3716, pruned_loss=0.1202, over 5682091.51 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.361, pruned_loss=0.1136, over 5660733.33 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3711, pruned_loss=0.1195, over 5690290.91 frames. ], batch size: 412, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:59:09,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.252e+03 1.806e+03 2.295e+03 3.344e+03 1.218e+04, threshold=4.590e+03, percent-clipped=8.0
+2023-03-12 04:59:28,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2309, 2.6355, 1.2290, 1.4373], device='cuda:0'), covar=tensor([0.0982, 0.0477, 0.0929, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0554, 0.0391, 0.0429], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 04:59:30,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1044481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:59:36,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1044485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:59:41,836 INFO [train.py:968] (0/2) Epoch 23, batch 41100, giga_loss[loss=0.351, simple_loss=0.4071, pruned_loss=0.1475, over 28945.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3778, pruned_loss=0.1255, over 5673700.93 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3607, pruned_loss=0.1134, over 5663241.89 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3778, pruned_loss=0.1253, over 5678123.82 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:00:34,408 INFO [train.py:968] (0/2) Epoch 23, batch 41150, giga_loss[loss=0.3452, simple_loss=0.397, pruned_loss=0.1467, over 28207.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3807, pruned_loss=0.1287, over 5662094.48 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3609, pruned_loss=0.1135, over 5665333.04 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3807, pruned_loss=0.1286, over 5663696.20 frames. ], batch size: 368, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:00:55,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.227e+03 1.656e+03 2.199e+03 3.137e+03 8.018e+03, threshold=4.398e+03, percent-clipped=5.0
+2023-03-12 05:01:18,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1044579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:01:33,087 INFO [train.py:968] (0/2) Epoch 23, batch 41200, giga_loss[loss=0.3035, simple_loss=0.3663, pruned_loss=0.1204, over 28725.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3844, pruned_loss=0.1329, over 5645741.28 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3619, pruned_loss=0.1141, over 5650824.97 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.384, pruned_loss=0.1326, over 5661113.14 frames. ], batch size: 112, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:01:58,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1044615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:00,116 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.87 vs. limit=2.0
+2023-03-12 05:02:08,628 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1044624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:10,745 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1044627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:11,595 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1044628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:14,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1044631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:26,211 INFO [train.py:968] (0/2) Epoch 23, batch 41250, giga_loss[loss=0.3302, simple_loss=0.3925, pruned_loss=0.134, over 28886.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3874, pruned_loss=0.1366, over 5628355.13 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.362, pruned_loss=0.1141, over 5654385.36 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3873, pruned_loss=0.1366, over 5637189.89 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:02:40,487 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1044656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:49,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1044660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:50,299 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.666e+03 2.155e+03 3.098e+03 6.967e+03, threshold=4.309e+03, percent-clipped=5.0
+2023-03-12 05:03:20,521 INFO [train.py:968] (0/2) Epoch 23, batch 41300, giga_loss[loss=0.3106, simple_loss=0.3764, pruned_loss=0.1224, over 28994.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3898, pruned_loss=0.1383, over 5630476.22 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3619, pruned_loss=0.114, over 5659213.09 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3903, pruned_loss=0.1389, over 5632817.28 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:04:13,239 INFO [train.py:968] (0/2) Epoch 23, batch 41350, giga_loss[loss=0.2794, simple_loss=0.3508, pruned_loss=0.104, over 28953.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.391, pruned_loss=0.1401, over 5623145.60 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3619, pruned_loss=0.114, over 5661771.17 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.392, pruned_loss=0.1411, over 5621810.10 frames. ], batch size: 106, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:04:33,424 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1044758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:04:35,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1044761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:04:35,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.247e+03 1.930e+03 2.493e+03 3.865e+03 1.231e+04, threshold=4.986e+03, percent-clipped=17.0
+2023-03-12 05:05:03,133 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1044790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:05:03,448 INFO [train.py:968] (0/2) Epoch 23, batch 41400, giga_loss[loss=0.3865, simple_loss=0.4213, pruned_loss=0.1758, over 27446.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3894, pruned_loss=0.1398, over 5625031.74 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3617, pruned_loss=0.114, over 5657280.59 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3911, pruned_loss=0.1412, over 5626303.00 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:05:55,189 INFO [train.py:968] (0/2) Epoch 23, batch 41450, giga_loss[loss=0.2659, simple_loss=0.3458, pruned_loss=0.093, over 29033.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3873, pruned_loss=0.1376, over 5647671.68 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3614, pruned_loss=0.1137, over 5664788.67 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3898, pruned_loss=0.1398, over 5641414.98 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:05:58,650 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-12 05:06:06,763 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1044853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:06:17,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.860e+03 2.484e+03 3.281e+03 7.835e+03, threshold=4.968e+03, percent-clipped=4.0
+2023-03-12 05:06:53,456 INFO [train.py:968] (0/2) Epoch 23, batch 41500, giga_loss[loss=0.3186, simple_loss=0.3827, pruned_loss=0.1273, over 29045.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3872, pruned_loss=0.1361, over 5644549.50 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3614, pruned_loss=0.1138, over 5658246.52 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3894, pruned_loss=0.138, over 5645620.38 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:07:10,581 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 05:07:47,274 INFO [train.py:968] (0/2) Epoch 23, batch 41550, giga_loss[loss=0.3152, simple_loss=0.3812, pruned_loss=0.1246, over 28675.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3878, pruned_loss=0.1358, over 5659708.59 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3608, pruned_loss=0.1135, over 5663027.86 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3906, pruned_loss=0.138, over 5656348.29 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:08:02,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1044954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:08:09,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.224e+03 1.898e+03 2.394e+03 3.331e+03 7.317e+03, threshold=4.788e+03, percent-clipped=3.0
+2023-03-12 05:08:43,822 INFO [train.py:968] (0/2) Epoch 23, batch 41600, giga_loss[loss=0.3059, simple_loss=0.3717, pruned_loss=0.12, over 28596.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3889, pruned_loss=0.1368, over 5647260.85 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3609, pruned_loss=0.1137, over 5666344.50 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3914, pruned_loss=0.1388, over 5641552.81 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:09:28,983 INFO [train.py:968] (0/2) Epoch 23, batch 41650, libri_loss[loss=0.2808, simple_loss=0.3509, pruned_loss=0.1054, over 29670.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3848, pruned_loss=0.1325, over 5643412.16 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3606, pruned_loss=0.1135, over 5662424.62 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3884, pruned_loss=0.1352, over 5640641.57 frames. ], batch size: 88, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:09:29,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2885, 1.2601, 4.0765, 3.3741], device='cuda:0'), covar=tensor([0.1767, 0.2875, 0.0467, 0.0924], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0660, 0.0978, 0.0926], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 05:09:48,337 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.161e+03 1.586e+03 1.972e+03 2.806e+03 6.488e+03, threshold=3.944e+03, percent-clipped=2.0
+2023-03-12 05:09:50,461 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045063.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:09:51,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:10:16,366 INFO [train.py:968] (0/2) Epoch 23, batch 41700, libri_loss[loss=0.3118, simple_loss=0.3747, pruned_loss=0.1244, over 29553.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3818, pruned_loss=0.1293, over 5648276.73 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3603, pruned_loss=0.1134, over 5660232.16 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3857, pruned_loss=0.1321, over 5647715.88 frames. ], batch size: 83, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:10:23,056 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:10:25,354 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:10:59,423 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:11:10,074 INFO [train.py:968] (0/2) Epoch 23, batch 41750, giga_loss[loss=0.2703, simple_loss=0.3413, pruned_loss=0.09963, over 28717.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3779, pruned_loss=0.1259, over 5652670.18 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.36, pruned_loss=0.1133, over 5661377.56 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3814, pruned_loss=0.1284, over 5651239.90 frames. ], batch size: 60, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:11:33,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.756e+03 2.427e+03 3.496e+03 1.263e+04, threshold=4.854e+03, percent-clipped=20.0
+2023-03-12 05:11:49,808 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:11:50,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:11:57,967 INFO [train.py:968] (0/2) Epoch 23, batch 41800, giga_loss[loss=0.2998, simple_loss=0.3676, pruned_loss=0.116, over 28620.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3745, pruned_loss=0.1234, over 5653507.82 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3599, pruned_loss=0.1131, over 5660120.92 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3779, pruned_loss=0.1259, over 5652327.92 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:12:35,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:12:43,439 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:12:50,556 INFO [train.py:968] (0/2) Epoch 23, batch 41850, giga_loss[loss=0.2776, simple_loss=0.3608, pruned_loss=0.09721, over 28890.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3746, pruned_loss=0.1237, over 5647727.05 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.36, pruned_loss=0.1132, over 5659878.08 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3774, pruned_loss=0.1257, over 5646855.97 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:13:09,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.610e+03 2.083e+03 2.521e+03 5.233e+03, threshold=4.167e+03, percent-clipped=2.0
+2023-03-12 05:13:41,213 INFO [train.py:968] (0/2) Epoch 23, batch 41900, giga_loss[loss=0.3079, simple_loss=0.3738, pruned_loss=0.121, over 28478.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3756, pruned_loss=0.1239, over 5660051.88 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1132, over 5662478.52 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3778, pruned_loss=0.1256, over 5656923.05 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:14:24,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2432, 1.1367, 3.9527, 3.3452], device='cuda:0'), covar=tensor([0.1784, 0.3047, 0.0517, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0661, 0.0980, 0.0926], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 05:14:35,101 INFO [train.py:968] (0/2) Epoch 23, batch 41950, giga_loss[loss=0.2768, simple_loss=0.3473, pruned_loss=0.1032, over 28874.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3717, pruned_loss=0.1204, over 5663005.89 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1133, over 5659211.09 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3741, pruned_loss=0.1221, over 5662976.18 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:14:39,977 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:14:49,632 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-12 05:14:53,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.688e+02 1.840e+03 2.411e+03 3.621e+03 1.041e+04, threshold=4.822e+03, percent-clipped=15.0
+2023-03-12 05:15:05,359 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:15:10,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:15:27,895 INFO [train.py:968] (0/2) Epoch 23, batch 42000, giga_loss[loss=0.2922, simple_loss=0.3781, pruned_loss=0.1031, over 29024.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3731, pruned_loss=0.1191, over 5672285.74 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3607, pruned_loss=0.1138, over 5663192.44 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3747, pruned_loss=0.1201, over 5669008.17 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:15:27,898 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 05:15:36,895 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2023, simple_loss=0.3083, pruned_loss=0.0482, over 944034.00 frames. 
+2023-03-12 05:15:36,896 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 05:15:48,061 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:16:19,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045438.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 05:16:22,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:16:23,018 INFO [train.py:968] (0/2) Epoch 23, batch 42050, giga_loss[loss=0.3292, simple_loss=0.3881, pruned_loss=0.1351, over 27581.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3743, pruned_loss=0.1189, over 5660565.12 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3606, pruned_loss=0.1139, over 5652710.55 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3763, pruned_loss=0.1199, over 5667090.90 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:16:24,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2681, 1.5901, 1.2846, 0.9220], device='cuda:0'), covar=tensor([0.2404, 0.2482, 0.2669, 0.2335], device='cuda:0'), in_proj_covar=tensor([0.1535, 0.1108, 0.1356, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 05:16:43,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.711e+03 2.289e+03 2.867e+03 7.222e+03, threshold=4.578e+03, percent-clipped=4.0
+2023-03-12 05:16:58,611 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 05:17:12,483 INFO [train.py:968] (0/2) Epoch 23, batch 42100, giga_loss[loss=0.3078, simple_loss=0.3769, pruned_loss=0.1194, over 28443.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3747, pruned_loss=0.12, over 5648308.47 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3607, pruned_loss=0.1142, over 5638739.75 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3764, pruned_loss=0.1206, over 5667238.52 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:17:57,087 INFO [train.py:968] (0/2) Epoch 23, batch 42150, giga_loss[loss=0.3029, simple_loss=0.3622, pruned_loss=0.1218, over 28823.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3741, pruned_loss=0.1202, over 5659871.34 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3603, pruned_loss=0.114, over 5647691.65 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3765, pruned_loss=0.1212, over 5667538.39 frames. ], batch size: 99, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:18:10,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:18:11,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:18:19,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.144e+03 1.729e+03 2.191e+03 3.193e+03 1.147e+04, threshold=4.382e+03, percent-clipped=10.0
+2023-03-12 05:18:34,801 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045581.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:18:35,419 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:18:37,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045584.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 05:18:38,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:18:43,187 INFO [train.py:968] (0/2) Epoch 23, batch 42200, giga_loss[loss=0.277, simple_loss=0.3497, pruned_loss=0.1021, over 28543.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3724, pruned_loss=0.1196, over 5665804.95 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3604, pruned_loss=0.114, over 5644798.23 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3746, pruned_loss=0.1205, over 5674955.54 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:19:03,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:19:05,288 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045613.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:19:06,147 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:19:33,686 INFO [train.py:968] (0/2) Epoch 23, batch 42250, giga_loss[loss=0.2646, simple_loss=0.3349, pruned_loss=0.09717, over 28481.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3713, pruned_loss=0.1203, over 5654838.45 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3606, pruned_loss=0.1141, over 5649144.26 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.373, pruned_loss=0.121, over 5658571.59 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:19:55,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.218e+03 1.556e+03 2.039e+03 2.718e+03 8.073e+03, threshold=4.079e+03, percent-clipped=8.0
+2023-03-12 05:19:59,789 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:20:25,094 INFO [train.py:968] (0/2) Epoch 23, batch 42300, libri_loss[loss=0.2669, simple_loss=0.3423, pruned_loss=0.09579, over 29528.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3698, pruned_loss=0.119, over 5668165.52 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3606, pruned_loss=0.114, over 5656532.05 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3714, pruned_loss=0.1198, over 5664865.53 frames. ], batch size: 81, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:20:33,276 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:20:33,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:20:35,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:20:36,549 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:20:54,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:21:06,225 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:21:06,957 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:21:15,653 INFO [train.py:968] (0/2) Epoch 23, batch 42350, giga_loss[loss=0.2731, simple_loss=0.352, pruned_loss=0.09706, over 28848.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3687, pruned_loss=0.1164, over 5681372.06 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3606, pruned_loss=0.1139, over 5658441.56 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3701, pruned_loss=0.1172, over 5677196.01 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:21:17,828 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5743, 1.5671, 1.7858, 1.3780], device='cuda:0'), covar=tensor([0.1690, 0.2483, 0.1377, 0.1692], device='cuda:0'), in_proj_covar=tensor([0.0909, 0.0706, 0.0954, 0.0853], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 05:21:27,958 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:21:31,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:21:37,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.015e+02 1.507e+03 1.921e+03 2.624e+03 8.745e+03, threshold=3.842e+03, percent-clipped=8.0
+2023-03-12 05:21:57,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3928, 3.6754, 1.5227, 1.6772], device='cuda:0'), covar=tensor([0.1016, 0.0435, 0.0899, 0.1310], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0560, 0.0394, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 05:22:00,949 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:22:10,407 INFO [train.py:968] (0/2) Epoch 23, batch 42400, libri_loss[loss=0.2578, simple_loss=0.3172, pruned_loss=0.09916, over 28554.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3685, pruned_loss=0.1158, over 5673667.24 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3608, pruned_loss=0.1141, over 5652355.49 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3696, pruned_loss=0.1163, over 5675578.48 frames. ], batch size: 63, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:22:25,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:22:45,525 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:22:58,114 INFO [train.py:968] (0/2) Epoch 23, batch 42450, giga_loss[loss=0.2924, simple_loss=0.3648, pruned_loss=0.1099, over 28861.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3671, pruned_loss=0.1149, over 5672584.02 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.1141, over 5648578.49 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3681, pruned_loss=0.1153, over 5677371.15 frames. ], batch size: 112, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:23:17,077 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:23:17,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.811e+03 2.288e+03 3.067e+03 8.775e+03, threshold=4.576e+03, percent-clipped=11.0
+2023-03-12 05:23:19,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:23:45,412 INFO [train.py:968] (0/2) Epoch 23, batch 42500, giga_loss[loss=0.2869, simple_loss=0.361, pruned_loss=0.1064, over 28604.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3661, pruned_loss=0.1151, over 5676217.75 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3605, pruned_loss=0.114, over 5653633.08 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3672, pruned_loss=0.1156, over 5676070.89 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:23:49,366 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:24:02,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-12 05:24:30,405 INFO [train.py:968] (0/2) Epoch 23, batch 42550, giga_loss[loss=0.2825, simple_loss=0.3504, pruned_loss=0.1073, over 29089.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3658, pruned_loss=0.1157, over 5677077.96 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3607, pruned_loss=0.1138, over 5659419.91 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3668, pruned_loss=0.1162, over 5672299.40 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:24:52,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.892e+03 2.326e+03 3.108e+03 7.816e+03, threshold=4.651e+03, percent-clipped=12.0
+2023-03-12 05:25:17,989 INFO [train.py:968] (0/2) Epoch 23, batch 42600, libri_loss[loss=0.3202, simple_loss=0.3812, pruned_loss=0.1296, over 19607.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3652, pruned_loss=0.1164, over 5667094.19 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3606, pruned_loss=0.1139, over 5657976.47 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3661, pruned_loss=0.1168, over 5665912.19 frames. ], batch size: 187, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:25:26,812 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1046000.pt
+2023-03-12 05:25:58,557 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-12 05:26:07,576 INFO [train.py:968] (0/2) Epoch 23, batch 42650, giga_loss[loss=0.2498, simple_loss=0.3258, pruned_loss=0.08692, over 28962.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3642, pruned_loss=0.1164, over 5666040.84 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3608, pruned_loss=0.1139, over 5652265.10 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3649, pruned_loss=0.1168, over 5670223.21 frames. ], batch size: 164, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:26:10,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1046043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:26:32,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.932e+02 1.834e+03 2.308e+03 3.065e+03 1.050e+04, threshold=4.616e+03, percent-clipped=7.0
+2023-03-12 05:26:53,154 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5027, 4.3513, 4.1409, 1.8363], device='cuda:0'), covar=tensor([0.0625, 0.0743, 0.0767, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.1270, 0.1173, 0.0990, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 05:27:03,296 INFO [train.py:968] (0/2) Epoch 23, batch 42700, libri_loss[loss=0.3238, simple_loss=0.383, pruned_loss=0.1323, over 29759.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3637, pruned_loss=0.1162, over 5674137.29 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3607, pruned_loss=0.1138, over 5656294.96 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3644, pruned_loss=0.1166, over 5673941.28 frames. ], batch size: 87, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:27:31,213 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-12 05:27:53,378 INFO [train.py:968] (0/2) Epoch 23, batch 42750, giga_loss[loss=0.3226, simple_loss=0.3763, pruned_loss=0.1345, over 28716.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.116, over 5682348.69 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3603, pruned_loss=0.1134, over 5660295.87 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3645, pruned_loss=0.1168, over 5678810.39 frames. ], batch size: 99, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:28:14,483 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.150e+03 1.543e+03 2.150e+03 2.924e+03 7.532e+03, threshold=4.299e+03, percent-clipped=9.0
+2023-03-12 05:28:34,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1046183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:28:36,503 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1046186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:28:40,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1046189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:28:42,263 INFO [train.py:968] (0/2) Epoch 23, batch 42800, giga_loss[loss=0.2867, simple_loss=0.368, pruned_loss=0.1026, over 28986.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.364, pruned_loss=0.1155, over 5673919.24 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3603, pruned_loss=0.1135, over 5653964.16 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3648, pruned_loss=0.1161, over 5676498.63 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:28:54,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1046204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:29:07,232 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1046218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:29:21,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4656, 1.7806, 1.3999, 1.5519], device='cuda:0'), covar=tensor([0.0725, 0.0311, 0.0330, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 05:29:30,169 INFO [train.py:968] (0/2) Epoch 23, batch 42850, libri_loss[loss=0.2747, simple_loss=0.3484, pruned_loss=0.1004, over 29549.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3645, pruned_loss=0.1152, over 5677431.34 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3603, pruned_loss=0.1135, over 5659250.94 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3652, pruned_loss=0.1156, over 5674975.66 frames. ], batch size: 80, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:29:51,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.583e+03 2.100e+03 2.962e+03 5.851e+03, threshold=4.200e+03, percent-clipped=5.0
+2023-03-12 05:30:04,945 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1046279.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:30:08,692 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-12 05:30:16,819 INFO [train.py:968] (0/2) Epoch 23, batch 42900, giga_loss[loss=0.2982, simple_loss=0.3704, pruned_loss=0.113, over 28816.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.366, pruned_loss=0.1159, over 5678788.68 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1133, over 5663823.79 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3668, pruned_loss=0.1165, over 5672894.58 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:30:54,169 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1046326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:30:56,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1046329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:31:07,422 INFO [train.py:968] (0/2) Epoch 23, batch 42950, giga_loss[loss=0.3099, simple_loss=0.3731, pruned_loss=0.1234, over 28896.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3673, pruned_loss=0.1178, over 5654617.65 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1133, over 5650663.85 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3683, pruned_loss=0.1184, over 5662278.37 frames. ], batch size: 99, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:31:11,724 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1046347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:31:14,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1046350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:31:20,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1046358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:31:27,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.604e+03 2.123e+03 2.883e+03 6.513e+03, threshold=4.246e+03, percent-clipped=6.0
+2023-03-12 05:31:43,919 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1046379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:31:55,487 INFO [train.py:968] (0/2) Epoch 23, batch 43000, giga_loss[loss=0.3356, simple_loss=0.3902, pruned_loss=0.1405, over 28820.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3707, pruned_loss=0.1213, over 5651958.77 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3599, pruned_loss=0.1131, over 5646688.93 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3718, pruned_loss=0.1221, over 5661110.55 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:32:48,660 INFO [train.py:968] (0/2) Epoch 23, batch 43050, giga_loss[loss=0.4185, simple_loss=0.4423, pruned_loss=0.1974, over 27600.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.372, pruned_loss=0.1238, over 5644441.97 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3603, pruned_loss=0.1133, over 5647616.03 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3728, pruned_loss=0.1245, over 5651094.05 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:33:14,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.341e+03 2.097e+03 2.728e+03 3.952e+03 1.036e+04, threshold=5.456e+03, percent-clipped=23.0
+2023-03-12 05:33:36,589 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1046486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:33:41,071 INFO [train.py:968] (0/2) Epoch 23, batch 43100, giga_loss[loss=0.2613, simple_loss=0.3352, pruned_loss=0.09366, over 28407.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3727, pruned_loss=0.1254, over 5650437.09 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3605, pruned_loss=0.1134, over 5649989.62 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3735, pruned_loss=0.126, over 5653858.17 frames. ], batch size: 65, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:33:50,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4408, 1.6451, 1.5885, 1.4663], device='cuda:0'), covar=tensor([0.1880, 0.1868, 0.2388, 0.2072], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0755, 0.0723, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 05:34:10,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2960, 3.1348, 2.9722, 1.5984], device='cuda:0'), covar=tensor([0.1028, 0.1129, 0.0996, 0.2193], device='cuda:0'), in_proj_covar=tensor([0.1275, 0.1178, 0.0995, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 05:34:31,150 INFO [train.py:968] (0/2) Epoch 23, batch 43150, giga_loss[loss=0.2657, simple_loss=0.3425, pruned_loss=0.09443, over 28903.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3733, pruned_loss=0.1258, over 5661644.32 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3604, pruned_loss=0.1133, over 5654064.09 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3741, pruned_loss=0.1266, over 5660953.73 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:34:52,265 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.256e+03 1.818e+03 2.543e+03 3.744e+03 9.587e+03, threshold=5.087e+03, percent-clipped=5.0
+2023-03-12 05:35:20,233 INFO [train.py:968] (0/2) Epoch 23, batch 43200, libri_loss[loss=0.2851, simple_loss=0.3539, pruned_loss=0.1082, over 29536.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3715, pruned_loss=0.1245, over 5674444.69 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3607, pruned_loss=0.1136, over 5663650.00 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3723, pruned_loss=0.1254, over 5665930.06 frames. ], batch size: 80, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:36:01,436 INFO [train.py:968] (0/2) Epoch 23, batch 43250, giga_loss[loss=0.2953, simple_loss=0.3645, pruned_loss=0.113, over 28641.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3706, pruned_loss=0.1219, over 5679745.23 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.114, over 5666586.26 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1226, over 5670355.17 frames. ], batch size: 307, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:36:11,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1556, 1.3685, 1.1308, 1.0285], device='cuda:0'), covar=tensor([0.1059, 0.0494, 0.1111, 0.1045], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0446, 0.0518, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 05:36:16,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1046654.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:36:22,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4217, 1.7505, 1.4272, 1.5378], device='cuda:0'), covar=tensor([0.0777, 0.0321, 0.0333, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 05:36:27,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.749e+03 2.098e+03 2.814e+03 5.085e+03, threshold=4.196e+03, percent-clipped=0.0
+2023-03-12 05:36:44,755 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-12 05:36:50,689 INFO [train.py:968] (0/2) Epoch 23, batch 43300, giga_loss[loss=0.2759, simple_loss=0.3476, pruned_loss=0.1021, over 28688.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3687, pruned_loss=0.12, over 5675768.79 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3615, pruned_loss=0.1141, over 5670124.94 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3692, pruned_loss=0.1207, over 5665284.14 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:37:01,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2934, 1.5993, 1.2679, 0.9497], device='cuda:0'), covar=tensor([0.2621, 0.2581, 0.2988, 0.2339], device='cuda:0'), in_proj_covar=tensor([0.1542, 0.1113, 0.1359, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 05:37:05,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4983, 1.8257, 1.4255, 1.4900], device='cuda:0'), covar=tensor([0.2685, 0.2669, 0.3109, 0.2430], device='cuda:0'), in_proj_covar=tensor([0.1542, 0.1113, 0.1359, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 05:37:10,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6775, 1.9094, 1.5390, 2.0019], device='cuda:0'), covar=tensor([0.2550, 0.2708, 0.2986, 0.2426], device='cuda:0'), in_proj_covar=tensor([0.1542, 0.1113, 0.1360, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 05:37:35,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9202, 1.1575, 1.3202, 0.9761], device='cuda:0'), covar=tensor([0.2030, 0.1669, 0.2614, 0.2007], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0752, 0.0720, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 05:37:36,658 INFO [train.py:968] (0/2) Epoch 23, batch 43350, giga_loss[loss=0.2901, simple_loss=0.3551, pruned_loss=0.1126, over 28907.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.367, pruned_loss=0.1193, over 5673374.57 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5674640.94 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3678, pruned_loss=0.1202, over 5661276.20 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:37:59,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.866e+03 2.406e+03 3.363e+03 1.574e+04, threshold=4.813e+03, percent-clipped=15.0
+2023-03-12 05:38:13,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7241, 2.0359, 1.4956, 2.0917], device='cuda:0'), covar=tensor([0.2665, 0.2798, 0.3250, 0.2549], device='cuda:0'), in_proj_covar=tensor([0.1543, 0.1113, 0.1361, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 05:38:24,929 INFO [train.py:968] (0/2) Epoch 23, batch 43400, giga_loss[loss=0.2592, simple_loss=0.3367, pruned_loss=0.09089, over 28929.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3649, pruned_loss=0.1185, over 5677033.19 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5677685.11 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3655, pruned_loss=0.1192, over 5664774.07 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:38:30,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1046797.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:38:35,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1046800.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 05:39:04,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1046829.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 05:39:15,756 INFO [train.py:968] (0/2) Epoch 23, batch 43450, giga_loss[loss=0.2979, simple_loss=0.3721, pruned_loss=0.1118, over 28843.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3666, pruned_loss=0.1199, over 5668351.28 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5677685.11 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3671, pruned_loss=0.1206, over 5658809.85 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:39:35,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1046861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:39:40,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.537e+03 1.936e+03 2.903e+03 7.449e+03, threshold=3.872e+03, percent-clipped=6.0
+2023-03-12 05:39:55,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3327, 1.5670, 1.3661, 1.6565], device='cuda:0'), covar=tensor([0.0772, 0.0351, 0.0335, 0.0815], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:0')
+2023-03-12 05:40:06,593 INFO [train.py:968] (0/2) Epoch 23, batch 43500, giga_loss[loss=0.3152, simple_loss=0.3899, pruned_loss=0.1203, over 28851.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3697, pruned_loss=0.1206, over 5671159.23 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3608, pruned_loss=0.1135, over 5680198.87 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3705, pruned_loss=0.1215, over 5661241.71 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:40:20,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3451, 1.6697, 1.5883, 1.5023], device='cuda:0'), covar=tensor([0.1963, 0.2016, 0.2232, 0.2098], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0753, 0.0720, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 05:40:49,604 INFO [train.py:968] (0/2) Epoch 23, batch 43550, giga_loss[loss=0.2877, simple_loss=0.3709, pruned_loss=0.1023, over 28953.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3722, pruned_loss=0.1196, over 5664542.74 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3611, pruned_loss=0.1137, over 5676230.52 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3729, pruned_loss=0.1203, over 5659950.92 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:41:17,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.634e+03 2.115e+03 3.307e+03 1.143e+04, threshold=4.229e+03, percent-clipped=15.0
+2023-03-12 05:41:41,714 INFO [train.py:968] (0/2) Epoch 23, batch 43600, giga_loss[loss=0.3881, simple_loss=0.432, pruned_loss=0.1721, over 28329.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3739, pruned_loss=0.1209, over 5669327.66 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3605, pruned_loss=0.1134, over 5681878.27 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3754, pruned_loss=0.1219, over 5660384.11 frames. ], batch size: 369, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:41:55,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:41:59,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:42:24,198 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:42:27,175 INFO [train.py:968] (0/2) Epoch 23, batch 43650, giga_loss[loss=0.3099, simple_loss=0.3838, pruned_loss=0.118, over 29016.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3752, pruned_loss=0.1218, over 5672970.19 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3602, pruned_loss=0.1134, over 5677925.88 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3771, pruned_loss=0.1229, over 5668625.91 frames. ], batch size: 136, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:42:52,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.057e+02 1.846e+03 2.532e+03 3.805e+03 8.135e+03, threshold=5.064e+03, percent-clipped=16.0
+2023-03-12 05:43:04,438 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:43:15,333 INFO [train.py:968] (0/2) Epoch 23, batch 43700, libri_loss[loss=0.3367, simple_loss=0.4028, pruned_loss=0.1353, over 29273.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3768, pruned_loss=0.1237, over 5671443.99 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3608, pruned_loss=0.1137, over 5684158.40 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3784, pruned_loss=0.1247, over 5662096.23 frames. ], batch size: 94, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:43:58,233 INFO [train.py:968] (0/2) Epoch 23, batch 43750, giga_loss[loss=0.2738, simple_loss=0.3455, pruned_loss=0.101, over 28949.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3761, pruned_loss=0.124, over 5683818.72 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3611, pruned_loss=0.1138, over 5687690.40 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3775, pruned_loss=0.1248, over 5673202.20 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:44:27,351 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.179e+03 1.874e+03 2.412e+03 3.261e+03 7.918e+03, threshold=4.823e+03, percent-clipped=4.0
+2023-03-12 05:44:39,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:44:42,109 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3819, 1.5874, 1.3677, 1.0197], device='cuda:0'), covar=tensor([0.2312, 0.2395, 0.2559, 0.2155], device='cuda:0'), in_proj_covar=tensor([0.1537, 0.1110, 0.1356, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 05:44:46,972 INFO [train.py:968] (0/2) Epoch 23, batch 43800, giga_loss[loss=0.3348, simple_loss=0.389, pruned_loss=0.1403, over 28960.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3748, pruned_loss=0.1242, over 5675880.30 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3607, pruned_loss=0.1135, over 5693851.14 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3768, pruned_loss=0.1255, over 5660882.22 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:44:55,434 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:45:33,916 INFO [train.py:968] (0/2) Epoch 23, batch 43850, giga_loss[loss=0.2716, simple_loss=0.3436, pruned_loss=0.09984, over 28879.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3735, pruned_loss=0.1241, over 5665947.54 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3608, pruned_loss=0.1137, over 5686804.69 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3757, pruned_loss=0.1256, over 5659173.14 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:45:57,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.847e+03 2.328e+03 3.280e+03 5.458e+03, threshold=4.655e+03, percent-clipped=2.0
+2023-03-12 05:46:20,449 INFO [train.py:968] (0/2) Epoch 23, batch 43900, giga_loss[loss=0.2972, simple_loss=0.3609, pruned_loss=0.1167, over 28507.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1233, over 5668808.01 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3607, pruned_loss=0.1136, over 5687966.71 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3739, pruned_loss=0.1249, over 5661655.47 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:46:40,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9040, 5.6962, 5.3696, 3.1707], device='cuda:0'), covar=tensor([0.0521, 0.0689, 0.0840, 0.1428], device='cuda:0'), in_proj_covar=tensor([0.1269, 0.1175, 0.0993, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 05:47:08,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3408, 1.6225, 1.3252, 1.4558], device='cuda:0'), covar=tensor([0.0747, 0.0346, 0.0342, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 05:47:10,753 INFO [train.py:968] (0/2) Epoch 23, batch 43950, giga_loss[loss=0.3382, simple_loss=0.3966, pruned_loss=0.1399, over 28820.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.124, over 5658241.96 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3605, pruned_loss=0.1134, over 5693470.15 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3746, pruned_loss=0.1258, over 5646392.77 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:47:18,373 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:47:37,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.178e+03 1.866e+03 2.484e+03 3.603e+03 7.269e+03, threshold=4.967e+03, percent-clipped=8.0
+2023-03-12 05:47:54,740 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:48:01,449 INFO [train.py:968] (0/2) Epoch 23, batch 44000, giga_loss[loss=0.266, simple_loss=0.3417, pruned_loss=0.09514, over 28881.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3696, pruned_loss=0.1225, over 5661033.21 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3604, pruned_loss=0.1132, over 5695741.48 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3716, pruned_loss=0.1241, over 5649426.67 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:48:17,031 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:48:20,224 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3261, 2.9544, 1.3475, 1.5894], device='cuda:0'), covar=tensor([0.1000, 0.0348, 0.0908, 0.1276], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0561, 0.0393, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 05:48:21,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 05:48:51,088 INFO [train.py:968] (0/2) Epoch 23, batch 44050, giga_loss[loss=0.28, simple_loss=0.3515, pruned_loss=0.1043, over 28498.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3673, pruned_loss=0.1208, over 5666425.72 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3604, pruned_loss=0.1132, over 5695741.48 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3689, pruned_loss=0.1221, over 5657392.20 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:49:04,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:49:08,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5222, 2.5986, 1.5724, 1.6429], device='cuda:0'), covar=tensor([0.0774, 0.0338, 0.0702, 0.1046], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0561, 0.0393, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 05:49:13,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.881e+03 2.588e+03 3.839e+03 7.606e+03, threshold=5.176e+03, percent-clipped=9.0
+2023-03-12 05:49:37,314 INFO [train.py:968] (0/2) Epoch 23, batch 44100, libri_loss[loss=0.2888, simple_loss=0.366, pruned_loss=0.1059, over 29221.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.367, pruned_loss=0.1201, over 5667652.90 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3605, pruned_loss=0.1131, over 5700533.39 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3685, pruned_loss=0.1215, over 5654708.22 frames. ], batch size: 94, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:49:40,644 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:50:26,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:50:29,677 INFO [train.py:968] (0/2) Epoch 23, batch 44150, giga_loss[loss=0.3262, simple_loss=0.3821, pruned_loss=0.1351, over 28717.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3712, pruned_loss=0.1226, over 5654996.21 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3612, pruned_loss=0.1137, over 5701741.08 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3718, pruned_loss=0.1234, over 5643101.43 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:50:44,262 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-12 05:50:44,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:50:53,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.651e+03 2.108e+03 2.710e+03 6.422e+03, threshold=4.215e+03, percent-clipped=1.0
+2023-03-12 05:50:59,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:51:14,818 INFO [train.py:968] (0/2) Epoch 23, batch 44200, giga_loss[loss=0.3436, simple_loss=0.381, pruned_loss=0.1531, over 23700.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3716, pruned_loss=0.1228, over 5661502.52 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3608, pruned_loss=0.1133, over 5707795.78 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3728, pruned_loss=0.124, over 5645221.21 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:51:23,146 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:51:25,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:51:51,798 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7196, 1.5175, 1.8199, 1.3632], device='cuda:0'), covar=tensor([0.2361, 0.3343, 0.1722, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.0908, 0.0707, 0.0954, 0.0853], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 05:51:56,117 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:52:03,000 INFO [train.py:968] (0/2) Epoch 23, batch 44250, giga_loss[loss=0.2838, simple_loss=0.373, pruned_loss=0.09726, over 28852.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3712, pruned_loss=0.1224, over 5667112.00 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3609, pruned_loss=0.1136, over 5701679.96 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3724, pruned_loss=0.1234, over 5657947.34 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:52:03,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7046, 1.6768, 1.8927, 1.4632], device='cuda:0'), covar=tensor([0.1984, 0.2631, 0.1572, 0.1915], device='cuda:0'), in_proj_covar=tensor([0.0908, 0.0707, 0.0955, 0.0853], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 05:52:27,155 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.180e+03 1.726e+03 2.349e+03 3.234e+03 1.165e+04, threshold=4.698e+03, percent-clipped=10.0
+2023-03-12 05:52:49,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7471, 1.3260, 4.8740, 3.5757], device='cuda:0'), covar=tensor([0.1698, 0.3027, 0.0489, 0.0939], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0661, 0.0980, 0.0929], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 05:52:49,405 INFO [train.py:968] (0/2) Epoch 23, batch 44300, giga_loss[loss=0.2733, simple_loss=0.3538, pruned_loss=0.09635, over 29061.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3741, pruned_loss=0.1217, over 5675946.56 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3616, pruned_loss=0.1139, over 5707630.45 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3747, pruned_loss=0.1224, over 5662393.02 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:52:58,046 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:00,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:12,528 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:14,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:16,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:25,022 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:33,061 INFO [train.py:968] (0/2) Epoch 23, batch 44350, giga_loss[loss=0.3326, simple_loss=0.3944, pruned_loss=0.1354, over 28961.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3757, pruned_loss=0.1211, over 5677431.73 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5713472.76 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3769, pruned_loss=0.122, over 5660741.25 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:53:40,102 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:48,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:54:01,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.836e+02 1.506e+03 1.952e+03 2.628e+03 4.533e+03, threshold=3.904e+03, percent-clipped=0.0
+2023-03-12 05:54:14,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:54:21,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5938, 2.2036, 1.5834, 0.7571], device='cuda:0'), covar=tensor([0.6753, 0.3125, 0.4031, 0.7300], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1681, 0.1608, 0.1443], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 05:54:23,211 INFO [train.py:968] (0/2) Epoch 23, batch 44400, giga_loss[loss=0.3434, simple_loss=0.4016, pruned_loss=0.1426, over 28786.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.377, pruned_loss=0.122, over 5670746.64 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.361, pruned_loss=0.1137, over 5715160.56 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3786, pruned_loss=0.123, over 5655117.04 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:54:28,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4551, 4.4813, 1.6900, 1.6612], device='cuda:0'), covar=tensor([0.1065, 0.0471, 0.0995, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0562, 0.0394, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 05:55:11,258 INFO [train.py:968] (0/2) Epoch 23, batch 44450, giga_loss[loss=0.2783, simple_loss=0.3463, pruned_loss=0.1052, over 28856.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3802, pruned_loss=0.1254, over 5667461.72 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3609, pruned_loss=0.1136, over 5710311.07 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3819, pruned_loss=0.1265, over 5657473.90 frames. ], batch size: 106, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:55:24,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0128, 2.1025, 2.2774, 1.7700], device='cuda:0'), covar=tensor([0.1812, 0.2437, 0.1418, 0.1757], device='cuda:0'), in_proj_covar=tensor([0.0908, 0.0708, 0.0955, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 05:55:37,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:55:38,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.887e+03 2.368e+03 3.317e+03 7.309e+03, threshold=4.736e+03, percent-clipped=15.0
+2023-03-12 05:55:38,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:55:41,408 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:55:44,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1247, 1.7069, 1.3218, 0.4384], device='cuda:0'), covar=tensor([0.4774, 0.2812, 0.3646, 0.6145], device='cuda:0'), in_proj_covar=tensor([0.1784, 0.1686, 0.1612, 0.1447], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 05:55:58,777 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-12 05:56:04,194 INFO [train.py:968] (0/2) Epoch 23, batch 44500, giga_loss[loss=0.331, simple_loss=0.3921, pruned_loss=0.1349, over 28984.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3819, pruned_loss=0.1278, over 5668235.64 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1143, over 5711778.17 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3829, pruned_loss=0.1284, over 5658279.35 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:56:05,049 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047892.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:56:11,061 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:15,348 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:17,395 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:18,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047907.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 05:56:23,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:38,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:40,829 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:45,233 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:50,548 INFO [train.py:968] (0/2) Epoch 23, batch 44550, giga_loss[loss=0.3073, simple_loss=0.3836, pruned_loss=0.1155, over 29080.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3797, pruned_loss=0.1267, over 5672460.68 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3611, pruned_loss=0.114, over 5704763.56 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3817, pruned_loss=0.1278, over 5669195.15 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:57:08,222 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:57:08,349 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4933, 1.6846, 1.6871, 1.2682], device='cuda:0'), covar=tensor([0.1714, 0.2723, 0.1523, 0.1843], device='cuda:0'), in_proj_covar=tensor([0.0907, 0.0707, 0.0955, 0.0854], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 05:57:17,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.173e+03 1.889e+03 2.290e+03 3.325e+03 6.965e+03, threshold=4.581e+03, percent-clipped=5.0
+2023-03-12 05:57:38,386 INFO [train.py:968] (0/2) Epoch 23, batch 44600, giga_loss[loss=0.3411, simple_loss=0.4079, pruned_loss=0.1372, over 28540.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3775, pruned_loss=0.1249, over 5666346.17 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3612, pruned_loss=0.1141, over 5708305.58 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3794, pruned_loss=0.1259, over 5659661.77 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:57:45,856 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1048000.pt
+2023-03-12 05:58:00,071 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:00,672 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1942, 1.7457, 1.3591, 0.4917], device='cuda:0'), covar=tensor([0.5101, 0.3250, 0.4374, 0.6478], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1682, 0.1611, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 05:58:01,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:18,516 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:24,520 INFO [train.py:968] (0/2) Epoch 23, batch 44650, libri_loss[loss=0.3149, simple_loss=0.3801, pruned_loss=0.1248, over 29500.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3761, pruned_loss=0.1217, over 5671222.77 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3611, pruned_loss=0.1141, over 5701003.59 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3781, pruned_loss=0.1228, over 5671052.52 frames. ], batch size: 82, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:58:30,592 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:37,280 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:39,357 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:41,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:54,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.271e+02 1.485e+03 1.858e+03 2.633e+03 6.526e+03, threshold=3.715e+03, percent-clipped=4.0
+2023-03-12 05:59:12,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:59:14,713 INFO [train.py:968] (0/2) Epoch 23, batch 44700, giga_loss[loss=0.3211, simple_loss=0.3891, pruned_loss=0.1265, over 28966.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3755, pruned_loss=0.1199, over 5672892.71 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3611, pruned_loss=0.1141, over 5693367.18 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3772, pruned_loss=0.1208, over 5678841.86 frames. ], batch size: 112, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:59:22,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3559, 5.1597, 2.1726, 2.7131], device='cuda:0'), covar=tensor([0.0843, 0.0357, 0.0803, 0.1051], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0564, 0.0394, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 05:59:25,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:00:05,786 INFO [train.py:968] (0/2) Epoch 23, batch 44750, giga_loss[loss=0.3147, simple_loss=0.38, pruned_loss=0.1247, over 28982.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3781, pruned_loss=0.1231, over 5647690.92 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3608, pruned_loss=0.1138, over 5685848.57 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.38, pruned_loss=0.1242, over 5659467.03 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:00:31,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.840e+03 2.223e+03 3.207e+03 6.311e+03, threshold=4.446e+03, percent-clipped=17.0
+2023-03-12 06:00:50,625 INFO [train.py:968] (0/2) Epoch 23, batch 44800, giga_loss[loss=0.3273, simple_loss=0.3901, pruned_loss=0.1322, over 28702.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3765, pruned_loss=0.1222, over 5644543.35 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3611, pruned_loss=0.1142, over 5683816.70 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3783, pruned_loss=0.1231, over 5654187.63 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 06:01:46,902 INFO [train.py:968] (0/2) Epoch 23, batch 44850, giga_loss[loss=0.3104, simple_loss=0.3705, pruned_loss=0.1251, over 28624.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3748, pruned_loss=0.122, over 5645127.83 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1144, over 5673992.61 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3761, pruned_loss=0.1226, over 5662015.87 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 06:02:15,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048267.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:02:18,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.809e+03 2.160e+03 2.678e+03 8.239e+03, threshold=4.321e+03, percent-clipped=9.0
+2023-03-12 06:02:27,014 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048282.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:02:36,320 INFO [train.py:968] (0/2) Epoch 23, batch 44900, giga_loss[loss=0.3544, simple_loss=0.3978, pruned_loss=0.1554, over 26742.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3739, pruned_loss=0.1223, over 5632986.87 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3615, pruned_loss=0.1145, over 5660003.22 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3749, pruned_loss=0.1228, over 5658300.67 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:03:25,205 INFO [train.py:968] (0/2) Epoch 23, batch 44950, libri_loss[loss=0.3544, simple_loss=0.413, pruned_loss=0.1479, over 29388.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3714, pruned_loss=0.121, over 5647613.75 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.362, pruned_loss=0.1148, over 5665541.09 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3721, pruned_loss=0.1212, over 5662275.41 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:03:50,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.221e+03 1.876e+03 2.255e+03 2.949e+03 6.541e+03, threshold=4.510e+03, percent-clipped=7.0
+2023-03-12 06:03:58,652 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:04:08,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7663, 2.5965, 1.6470, 0.9922], device='cuda:0'), covar=tensor([0.7497, 0.3357, 0.4078, 0.7088], device='cuda:0'), in_proj_covar=tensor([0.1783, 0.1686, 0.1615, 0.1446], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 06:04:08,887 INFO [train.py:968] (0/2) Epoch 23, batch 45000, giga_loss[loss=0.3028, simple_loss=0.37, pruned_loss=0.1178, over 28718.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3715, pruned_loss=0.1221, over 5643754.17 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.362, pruned_loss=0.1148, over 5662155.26 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3723, pruned_loss=0.1225, over 5657308.47 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:04:08,891 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 06:04:19,228 INFO [train.py:1012] (0/2) Epoch 23, validation: loss=0.2058, simple_loss=0.3149, pruned_loss=0.04835, over 944034.00 frames. 
+2023-03-12 06:04:19,229 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 06:04:33,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:04:35,332 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048410.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:04:37,437 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048413.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:04:46,826 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048425.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:04:48,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:04:49,030 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048428.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:05:00,396 INFO [train.py:968] (0/2) Epoch 23, batch 45050, giga_loss[loss=0.2935, simple_loss=0.3614, pruned_loss=0.1128, over 28301.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1227, over 5623727.98 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3613, pruned_loss=0.1143, over 5653265.04 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3728, pruned_loss=0.1238, over 5640744.77 frames. ], batch size: 368, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:05:02,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048442.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:05:04,337 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:05:14,595 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:05:14,610 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048457.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:05:24,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.627e+02 1.610e+03 2.278e+03 3.238e+03 7.213e+03, threshold=4.555e+03, percent-clipped=6.0
+2023-03-12 06:05:28,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:05:44,386 INFO [train.py:968] (0/2) Epoch 23, batch 45100, giga_loss[loss=0.2577, simple_loss=0.3464, pruned_loss=0.08454, over 28653.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3677, pruned_loss=0.1187, over 5641351.53 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3611, pruned_loss=0.1141, over 5661237.14 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3696, pruned_loss=0.1201, over 5647004.44 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:06:29,102 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:06:30,990 INFO [train.py:968] (0/2) Epoch 23, batch 45150, giga_loss[loss=0.2524, simple_loss=0.3323, pruned_loss=0.08621, over 28942.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3632, pruned_loss=0.1144, over 5641794.81 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3608, pruned_loss=0.1139, over 5660452.89 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3651, pruned_loss=0.1157, over 5646871.21 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:06:40,339 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:06:43,049 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:06:59,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.169e+02 1.385e+03 1.731e+03 2.429e+03 6.523e+03, threshold=3.462e+03, percent-clipped=2.0
+2023-03-12 06:06:59,940 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:02,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:12,295 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:21,613 INFO [train.py:968] (0/2) Epoch 23, batch 45200, libri_loss[loss=0.3157, simple_loss=0.363, pruned_loss=0.1342, over 29552.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3631, pruned_loss=0.1148, over 5647108.24 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3605, pruned_loss=0.1139, over 5669057.43 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3649, pruned_loss=0.1158, over 5642701.39 frames. ], batch size: 76, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 06:07:32,972 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:45,398 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:45,965 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:48,908 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:08:08,528 INFO [train.py:968] (0/2) Epoch 23, batch 45250, giga_loss[loss=0.2592, simple_loss=0.3289, pruned_loss=0.09477, over 28997.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3625, pruned_loss=0.115, over 5669696.57 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3602, pruned_loss=0.1137, over 5675408.51 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3643, pruned_loss=0.1161, over 5660237.34 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:08:19,485 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:08:30,219 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3987, 4.2340, 4.0289, 1.8014], device='cuda:0'), covar=tensor([0.0636, 0.0779, 0.0765, 0.2168], device='cuda:0'), in_proj_covar=tensor([0.1278, 0.1183, 0.0997, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 06:08:37,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8361, 1.1031, 2.8754, 2.7886], device='cuda:0'), covar=tensor([0.1703, 0.2648, 0.0636, 0.1081], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0664, 0.0987, 0.0934], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 06:08:38,158 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-12 06:08:43,685 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.111e+03 1.944e+03 2.684e+03 3.455e+03 1.221e+04, threshold=5.368e+03, percent-clipped=25.0
+2023-03-12 06:08:46,485 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-12 06:09:01,074 INFO [train.py:968] (0/2) Epoch 23, batch 45300, giga_loss[loss=0.2656, simple_loss=0.3522, pruned_loss=0.0895, over 28960.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3609, pruned_loss=0.1146, over 5667319.90 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3604, pruned_loss=0.1138, over 5668942.95 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3622, pruned_loss=0.1154, over 5665875.78 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:09:46,484 INFO [train.py:968] (0/2) Epoch 23, batch 45350, giga_loss[loss=0.2879, simple_loss=0.3664, pruned_loss=0.1047, over 28736.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3625, pruned_loss=0.1153, over 5638102.41 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3608, pruned_loss=0.1141, over 5627719.11 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3632, pruned_loss=0.1156, over 5676272.94 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:09:57,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:10:13,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.321e+02 1.710e+03 2.215e+03 2.951e+03 8.808e+03, threshold=4.431e+03, percent-clipped=4.0
+2023-03-12 06:10:34,565 INFO [train.py:968] (0/2) Epoch 23, batch 45400, giga_loss[loss=0.4285, simple_loss=0.4429, pruned_loss=0.2071, over 26499.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3649, pruned_loss=0.1167, over 5622323.22 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3611, pruned_loss=0.1144, over 5609760.72 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3652, pruned_loss=0.1167, over 5667767.98 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:11:05,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:11:16,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:11:26,368 INFO [train.py:968] (0/2) Epoch 23, batch 45450, giga_loss[loss=0.2599, simple_loss=0.3381, pruned_loss=0.09082, over 28923.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3674, pruned_loss=0.1183, over 5590301.47 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.362, pruned_loss=0.1151, over 5567202.12 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3669, pruned_loss=0.1177, over 5665878.92 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:11:53,942 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6805, 1.7820, 1.9157, 1.4522], device='cuda:0'), covar=tensor([0.1803, 0.2646, 0.1461, 0.1709], device='cuda:0'), in_proj_covar=tensor([0.0910, 0.0711, 0.0958, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 06:11:54,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.109e+03 1.799e+03 2.363e+03 3.399e+03 1.192e+04, threshold=4.726e+03, percent-clipped=13.0
+2023-03-12 06:12:11,994 INFO [train.py:968] (0/2) Epoch 23, batch 45500, giga_loss[loss=0.3363, simple_loss=0.3887, pruned_loss=0.142, over 28005.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3693, pruned_loss=0.1203, over 5572540.71 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3627, pruned_loss=0.1157, over 5535621.92 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3683, pruned_loss=0.1194, over 5662437.00 frames. ], batch size: 412, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:12:17,569 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:12:19,892 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-12 06:12:21,365 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-23.pt
+2023-03-12 06:12:54,581 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:13:08,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:13:12,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0912, 1.3423, 1.0828, 0.9308], device='cuda:0'), covar=tensor([0.1194, 0.0505, 0.1224, 0.1113], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0451, 0.0523, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 06:13:24,236 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:13:39,794 INFO [train.py:968] (0/2) Epoch 24, batch 50, giga_loss[loss=0.2882, simple_loss=0.3686, pruned_loss=0.1039, over 28695.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3747, pruned_loss=0.1088, over 1256823.28 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3405, pruned_loss=0.08844, over 163992.08 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3792, pruned_loss=0.1116, over 1125835.44 frames. ], batch size: 242, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:13:55,391 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:13:57,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:02,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.067e+02 1.287e+03 1.557e+03 2.352e+03 6.019e+03, threshold=3.113e+03, percent-clipped=4.0
+2023-03-12 06:14:08,770 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:12,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:28,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:31,111 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:33,442 INFO [train.py:968] (0/2) Epoch 24, batch 100, giga_loss[loss=0.2384, simple_loss=0.3265, pruned_loss=0.0751, over 28857.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3634, pruned_loss=0.1028, over 2232386.30 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3413, pruned_loss=0.08911, over 268222.35 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3658, pruned_loss=0.1043, over 2064280.47 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:14:40,654 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:42,099 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:15:17,368 INFO [train.py:968] (0/2) Epoch 24, batch 150, giga_loss[loss=0.2376, simple_loss=0.3171, pruned_loss=0.07909, over 28651.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3494, pruned_loss=0.09659, over 3003966.08 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3372, pruned_loss=0.08522, over 489020.47 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3517, pruned_loss=0.09851, over 2750619.62 frames. ], batch size: 307, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:15:24,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:15:26,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:15:38,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.918e+02 1.149e+03 1.381e+03 1.844e+03 9.374e+03, threshold=2.763e+03, percent-clipped=3.0
+2023-03-12 06:15:52,239 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:15:59,503 INFO [train.py:968] (0/2) Epoch 24, batch 200, giga_loss[loss=0.235, simple_loss=0.313, pruned_loss=0.07851, over 29076.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3353, pruned_loss=0.08959, over 3608089.85 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3325, pruned_loss=0.08304, over 672204.37 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3369, pruned_loss=0.09117, over 3325428.87 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:16:07,122 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:16:32,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:16:34,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:16:39,284 INFO [train.py:968] (0/2) Epoch 24, batch 250, libri_loss[loss=0.2897, simple_loss=0.3792, pruned_loss=0.1001, over 29252.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3263, pruned_loss=0.08543, over 4079323.65 frames. ], libri_tot_loss[loss=0.2457, simple_loss=0.3294, pruned_loss=0.08102, over 927598.47 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3273, pruned_loss=0.08702, over 3755801.97 frames. ], batch size: 97, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:17:02,054 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:17:03,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.429e+02 1.147e+03 1.450e+03 1.890e+03 3.738e+03, threshold=2.899e+03, percent-clipped=7.0
+2023-03-12 06:17:24,747 INFO [train.py:968] (0/2) Epoch 24, batch 300, giga_loss[loss=0.2213, simple_loss=0.299, pruned_loss=0.07185, over 27983.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3175, pruned_loss=0.08158, over 4443367.26 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3295, pruned_loss=0.08163, over 1050695.20 frames. ], giga_tot_loss[loss=0.2411, simple_loss=0.3174, pruned_loss=0.0824, over 4148634.93 frames. ], batch size: 412, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:18:11,761 INFO [train.py:968] (0/2) Epoch 24, batch 350, giga_loss[loss=0.2312, simple_loss=0.2959, pruned_loss=0.08327, over 26594.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3102, pruned_loss=0.0782, over 4726168.62 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3305, pruned_loss=0.08182, over 1193202.30 frames. ], giga_tot_loss[loss=0.2329, simple_loss=0.3088, pruned_loss=0.07849, over 4455282.16 frames. ], batch size: 555, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:18:28,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1992, 3.2923, 2.3818, 1.3424], device='cuda:0'), covar=tensor([0.7868, 0.2779, 0.3513, 0.6443], device='cuda:0'), in_proj_covar=tensor([0.1784, 0.1682, 0.1614, 0.1443], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 06:18:31,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.872e+02 1.111e+03 1.344e+03 1.678e+03 4.302e+03, threshold=2.689e+03, percent-clipped=5.0
+2023-03-12 06:18:39,337 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 06:18:51,447 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7548, 1.3235, 4.8155, 3.6551], device='cuda:0'), covar=tensor([0.1578, 0.2891, 0.0368, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0660, 0.0978, 0.0926], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 06:18:53,770 INFO [train.py:968] (0/2) Epoch 24, batch 400, libri_loss[loss=0.2658, simple_loss=0.3488, pruned_loss=0.09136, over 29657.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3065, pruned_loss=0.07715, over 4949141.83 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3304, pruned_loss=0.08214, over 1310180.92 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.3047, pruned_loss=0.07711, over 4708662.48 frames. ], batch size: 91, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:19:09,287 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:19:36,203 INFO [train.py:968] (0/2) Epoch 24, batch 450, giga_loss[loss=0.1844, simple_loss=0.2649, pruned_loss=0.05193, over 28568.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3041, pruned_loss=0.0762, over 5123936.57 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3313, pruned_loss=0.08316, over 1423274.85 frames. ], giga_tot_loss[loss=0.2265, simple_loss=0.3017, pruned_loss=0.07571, over 4913036.29 frames. ], batch size: 60, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:20:00,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.866e+02 1.056e+03 1.364e+03 1.814e+03 5.445e+03, threshold=2.728e+03, percent-clipped=11.0
+2023-03-12 06:20:05,863 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2354, 1.8115, 1.4722, 0.4197], device='cuda:0'), covar=tensor([0.5324, 0.3236, 0.5012, 0.7018], device='cuda:0'), in_proj_covar=tensor([0.1777, 0.1679, 0.1611, 0.1441], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 06:20:06,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1049380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:20:18,511 INFO [train.py:968] (0/2) Epoch 24, batch 500, giga_loss[loss=0.2363, simple_loss=0.3009, pruned_loss=0.08584, over 27666.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3028, pruned_loss=0.07541, over 5258678.47 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3341, pruned_loss=0.08452, over 1572820.32 frames. ], giga_tot_loss[loss=0.2237, simple_loss=0.2989, pruned_loss=0.07428, over 5070831.64 frames. ], batch size: 472, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:20:40,447 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:21:02,492 INFO [train.py:968] (0/2) Epoch 24, batch 550, libri_loss[loss=0.2773, simple_loss=0.3597, pruned_loss=0.09751, over 26017.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3013, pruned_loss=0.07499, over 5345685.24 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3356, pruned_loss=0.0854, over 1666319.04 frames. ], giga_tot_loss[loss=0.2222, simple_loss=0.297, pruned_loss=0.07364, over 5194959.99 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:21:04,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0215, 2.0442, 2.0793, 1.8313], device='cuda:0'), covar=tensor([0.2258, 0.3078, 0.2461, 0.2692], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0753, 0.0718, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 06:21:22,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.926e+02 1.051e+03 1.294e+03 1.750e+03 4.063e+03, threshold=2.588e+03, percent-clipped=6.0
+2023-03-12 06:21:34,871 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1049481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:21:49,546 INFO [train.py:968] (0/2) Epoch 24, batch 600, giga_loss[loss=0.2405, simple_loss=0.3091, pruned_loss=0.08592, over 28715.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2991, pruned_loss=0.07419, over 5417856.46 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3366, pruned_loss=0.08598, over 1717358.67 frames. ], giga_tot_loss[loss=0.2203, simple_loss=0.2951, pruned_loss=0.07282, over 5303212.67 frames. ], batch size: 262, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:22:05,069 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5719, 1.4608, 1.2989, 1.6396], device='cuda:0'), covar=tensor([0.0772, 0.0364, 0.0355, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0063, 0.0110], device='cuda:0')
+2023-03-12 06:22:15,291 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:22:17,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:22:34,537 INFO [train.py:968] (0/2) Epoch 24, batch 650, giga_loss[loss=0.2012, simple_loss=0.2833, pruned_loss=0.05949, over 28867.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2972, pruned_loss=0.0729, over 5474162.24 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3367, pruned_loss=0.08561, over 1857990.72 frames. ], giga_tot_loss[loss=0.2177, simple_loss=0.2925, pruned_loss=0.07148, over 5370633.47 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:22:42,609 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:22:56,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.881e+02 1.068e+03 1.253e+03 1.821e+03 6.646e+03, threshold=2.507e+03, percent-clipped=10.0
+2023-03-12 06:22:57,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6361, 1.8985, 1.5702, 1.6641], device='cuda:0'), covar=tensor([0.2493, 0.2506, 0.2721, 0.2611], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1118, 0.1370, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 06:23:20,880 INFO [train.py:968] (0/2) Epoch 24, batch 700, giga_loss[loss=0.2093, simple_loss=0.2781, pruned_loss=0.07027, over 29008.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2944, pruned_loss=0.07184, over 5522831.72 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3361, pruned_loss=0.08512, over 1936754.59 frames. ], giga_tot_loss[loss=0.2157, simple_loss=0.2902, pruned_loss=0.07062, over 5434660.28 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:23:43,565 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:23:45,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:24:05,869 INFO [train.py:968] (0/2) Epoch 24, batch 750, giga_loss[loss=0.206, simple_loss=0.2761, pruned_loss=0.06794, over 28778.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2936, pruned_loss=0.07142, over 5569453.99 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3373, pruned_loss=0.08572, over 2072865.75 frames. ], giga_tot_loss[loss=0.214, simple_loss=0.2884, pruned_loss=0.0698, over 5487720.29 frames. ], batch size: 99, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:24:15,026 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:24:15,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6026, 1.6154, 1.3148, 1.2885], device='cuda:0'), covar=tensor([0.0890, 0.0593, 0.0956, 0.1238], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0448, 0.0520, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 06:24:29,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.043e+02 1.084e+03 1.414e+03 1.934e+03 5.873e+03, threshold=2.828e+03, percent-clipped=12.0
+2023-03-12 06:24:44,922 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1049691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:24:52,042 INFO [train.py:968] (0/2) Epoch 24, batch 800, giga_loss[loss=0.1897, simple_loss=0.2629, pruned_loss=0.05829, over 28692.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2911, pruned_loss=0.07024, over 5607562.13 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3379, pruned_loss=0.08642, over 2205177.45 frames. ], giga_tot_loss[loss=0.2108, simple_loss=0.2852, pruned_loss=0.06819, over 5531624.59 frames. ], batch size: 92, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:25:12,897 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:25:36,649 INFO [train.py:968] (0/2) Epoch 24, batch 850, libri_loss[loss=0.26, simple_loss=0.3443, pruned_loss=0.08783, over 29261.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2949, pruned_loss=0.07291, over 5606634.05 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3373, pruned_loss=0.08591, over 2320009.96 frames. ], giga_tot_loss[loss=0.2155, simple_loss=0.289, pruned_loss=0.07099, over 5548418.28 frames. ], batch size: 97, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:25:41,743 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3220, 1.4792, 1.3747, 1.2584], device='cuda:0'), covar=tensor([0.2286, 0.2166, 0.1772, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.1997, 0.1933, 0.1858, 0.2001], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 06:25:58,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.653e+02 1.146e+03 1.411e+03 1.780e+03 3.820e+03, threshold=2.823e+03, percent-clipped=7.0
+2023-03-12 06:26:11,307 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 06:26:18,855 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 06:26:24,026 INFO [train.py:968] (0/2) Epoch 24, batch 900, giga_loss[loss=0.3086, simple_loss=0.3753, pruned_loss=0.1209, over 27705.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3071, pruned_loss=0.07919, over 5631482.93 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3369, pruned_loss=0.08594, over 2408808.72 frames. ], giga_tot_loss[loss=0.2284, simple_loss=0.3018, pruned_loss=0.07748, over 5578452.16 frames. ], batch size: 472, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:26:24,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1049798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:26:58,281 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:27:03,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:27:11,385 INFO [train.py:968] (0/2) Epoch 24, batch 950, giga_loss[loss=0.2477, simple_loss=0.3301, pruned_loss=0.08267, over 29094.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3192, pruned_loss=0.08514, over 5648318.91 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3358, pruned_loss=0.08536, over 2477255.63 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.315, pruned_loss=0.08398, over 5602340.31 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:27:25,354 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:27:27,269 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:27:31,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.692e+02 1.438e+03 1.660e+03 2.509e+03 5.701e+03, threshold=3.320e+03, percent-clipped=16.0
+2023-03-12 06:27:53,617 INFO [train.py:968] (0/2) Epoch 24, batch 1000, giga_loss[loss=0.2551, simple_loss=0.3327, pruned_loss=0.08875, over 28252.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.328, pruned_loss=0.08881, over 5656563.34 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3357, pruned_loss=0.08531, over 2569384.76 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3245, pruned_loss=0.08799, over 5621161.64 frames. ], batch size: 77, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:28:16,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7161, 1.9151, 1.8151, 1.6090], device='cuda:0'), covar=tensor([0.1984, 0.2212, 0.2415, 0.2368], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0751, 0.0719, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 06:28:30,351 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6311, 1.9001, 1.7499, 1.7297], device='cuda:0'), covar=tensor([0.2226, 0.2344, 0.2486, 0.2277], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0751, 0.0719, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 06:28:31,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:28:33,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:28:36,050 INFO [train.py:968] (0/2) Epoch 24, batch 1050, giga_loss[loss=0.2504, simple_loss=0.3413, pruned_loss=0.07978, over 28866.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3329, pruned_loss=0.08999, over 5671108.69 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3358, pruned_loss=0.08531, over 2618897.97 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.33, pruned_loss=0.08942, over 5640017.09 frames. ], batch size: 285, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:28:42,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9874, 3.2245, 2.0987, 1.1323], device='cuda:0'), covar=tensor([0.8171, 0.3329, 0.4519, 0.7020], device='cuda:0'), in_proj_covar=tensor([0.1770, 0.1672, 0.1609, 0.1438], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 06:28:48,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8277, 2.1305, 2.1469, 1.5751], device='cuda:0'), covar=tensor([0.3278, 0.2604, 0.2681, 0.3413], device='cuda:0'), in_proj_covar=tensor([0.1991, 0.1931, 0.1855, 0.2002], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 06:28:53,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.023e+02 1.183e+03 1.564e+03 2.021e+03 4.535e+03, threshold=3.128e+03, percent-clipped=3.0
+2023-03-12 06:28:55,936 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:29:05,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5293, 1.8654, 1.6520, 1.6298], device='cuda:0'), covar=tensor([0.2222, 0.2550, 0.2476, 0.2419], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0752, 0.0721, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 06:29:20,821 INFO [train.py:968] (0/2) Epoch 24, batch 1100, giga_loss[loss=0.2711, simple_loss=0.344, pruned_loss=0.09911, over 28472.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3364, pruned_loss=0.09114, over 5674527.29 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3357, pruned_loss=0.08528, over 2746767.57 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3341, pruned_loss=0.09088, over 5642451.87 frames. ], batch size: 65, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:29:23,233 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1050000.pt
+2023-03-12 06:29:27,210 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 06:29:36,851 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 06:29:56,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1970, 2.3247, 2.2144, 1.8839], device='cuda:0'), covar=tensor([0.2542, 0.2276, 0.2347, 0.2574], device='cuda:0'), in_proj_covar=tensor([0.1988, 0.1931, 0.1851, 0.2000], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 06:29:58,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 06:30:04,926 INFO [train.py:968] (0/2) Epoch 24, batch 1150, libri_loss[loss=0.2311, simple_loss=0.326, pruned_loss=0.06817, over 29520.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3377, pruned_loss=0.09132, over 5675464.95 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.336, pruned_loss=0.08529, over 2811766.73 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3358, pruned_loss=0.09126, over 5656860.52 frames. ], batch size: 84, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:30:12,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6796, 1.9235, 1.5774, 1.8626], device='cuda:0'), covar=tensor([0.2643, 0.2685, 0.3104, 0.2381], device='cuda:0'), in_proj_covar=tensor([0.1540, 0.1111, 0.1363, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 06:30:24,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.304e+03 1.621e+03 2.281e+03 5.264e+03, threshold=3.243e+03, percent-clipped=13.0
+2023-03-12 06:30:48,368 INFO [train.py:968] (0/2) Epoch 24, batch 1200, giga_loss[loss=0.2645, simple_loss=0.344, pruned_loss=0.09249, over 29003.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3388, pruned_loss=0.09241, over 5676870.14 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3359, pruned_loss=0.0854, over 2929692.54 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3374, pruned_loss=0.09256, over 5657272.54 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:30:48,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1050098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:31:26,115 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-12 06:31:33,598 INFO [train.py:968] (0/2) Epoch 24, batch 1250, giga_loss[loss=0.2516, simple_loss=0.3312, pruned_loss=0.08599, over 28290.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3431, pruned_loss=0.09546, over 5683055.97 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3361, pruned_loss=0.08559, over 3017781.78 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.342, pruned_loss=0.09576, over 5661038.93 frames. ], batch size: 77, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:31:40,697 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.35 vs. limit=5.0
+2023-03-12 06:31:52,206 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.424e+02 1.333e+03 1.558e+03 1.877e+03 6.171e+03, threshold=3.117e+03, percent-clipped=3.0
+2023-03-12 06:32:12,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3033, 2.7687, 1.4843, 1.4202], device='cuda:0'), covar=tensor([0.0979, 0.0326, 0.0863, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0553, 0.0392, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 06:32:14,653 INFO [train.py:968] (0/2) Epoch 24, batch 1300, giga_loss[loss=0.2897, simple_loss=0.3616, pruned_loss=0.1089, over 28497.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3457, pruned_loss=0.09676, over 5676771.51 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3366, pruned_loss=0.08582, over 3078595.89 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3449, pruned_loss=0.09713, over 5664472.31 frames. ], batch size: 85, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:32:28,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8739, 1.3052, 1.1083, 0.1793], device='cuda:0'), covar=tensor([0.4513, 0.3212, 0.4112, 0.6400], device='cuda:0'), in_proj_covar=tensor([0.1776, 0.1675, 0.1613, 0.1439], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 06:32:36,732 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1050220.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:32:51,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1050239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:32:53,441 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1050241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:32:57,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1050244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:33:00,899 INFO [train.py:968] (0/2) Epoch 24, batch 1350, giga_loss[loss=0.2622, simple_loss=0.3412, pruned_loss=0.09162, over 28732.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3478, pruned_loss=0.0969, over 5682923.36 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3364, pruned_loss=0.08572, over 3116227.02 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3474, pruned_loss=0.09738, over 5674046.23 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:33:18,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.694e+02 1.320e+03 1.590e+03 2.033e+03 3.934e+03, threshold=3.180e+03, percent-clipped=5.0
+2023-03-12 06:33:22,935 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1050273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:33:45,667 INFO [train.py:968] (0/2) Epoch 24, batch 1400, giga_loss[loss=0.2676, simple_loss=0.352, pruned_loss=0.09161, over 28654.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3493, pruned_loss=0.09697, over 5685501.03 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.336, pruned_loss=0.0855, over 3154419.18 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3492, pruned_loss=0.0976, over 5678849.79 frames. ], batch size: 307, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:34:19,954 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-12 06:34:28,433 INFO [train.py:968] (0/2) Epoch 24, batch 1450, giga_loss[loss=0.2919, simple_loss=0.3642, pruned_loss=0.1098, over 27926.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3496, pruned_loss=0.09646, over 5686418.30 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3357, pruned_loss=0.08527, over 3221669.12 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.35, pruned_loss=0.09733, over 5677106.25 frames. ], batch size: 412, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:34:38,142 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-12 06:34:42,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3094, 5.1017, 4.8594, 2.3973], device='cuda:0'), covar=tensor([0.0398, 0.0566, 0.0611, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.1236, 0.1146, 0.0964, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 06:34:47,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.702e+02 1.185e+03 1.418e+03 1.943e+03 4.654e+03, threshold=2.837e+03, percent-clipped=2.0
+2023-03-12 06:34:56,367 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1050382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:34:59,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1050385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:35:02,500 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1050390.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:35:08,460 INFO [train.py:968] (0/2) Epoch 24, batch 1500, giga_loss[loss=0.2765, simple_loss=0.3514, pruned_loss=0.1008, over 28001.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3486, pruned_loss=0.09466, over 5702611.31 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3358, pruned_loss=0.08547, over 3299647.31 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3493, pruned_loss=0.09554, over 5690763.88 frames. ], batch size: 412, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:35:22,528 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1050414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:35:50,042 INFO [train.py:968] (0/2) Epoch 24, batch 1550, libri_loss[loss=0.2367, simple_loss=0.3169, pruned_loss=0.0783, over 29590.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3468, pruned_loss=0.09267, over 5706115.51 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3365, pruned_loss=0.08581, over 3375822.28 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3473, pruned_loss=0.09344, over 5691913.78 frames. ], batch size: 74, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:36:09,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.336e+02 1.155e+03 1.371e+03 1.858e+03 3.471e+03, threshold=2.741e+03, percent-clipped=4.0
+2023-03-12 06:36:33,144 INFO [train.py:968] (0/2) Epoch 24, batch 1600, giga_loss[loss=0.2658, simple_loss=0.3473, pruned_loss=0.09208, over 28874.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3459, pruned_loss=0.09206, over 5711023.42 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3367, pruned_loss=0.08588, over 3413628.45 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3463, pruned_loss=0.09269, over 5697070.63 frames. ], batch size: 145, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:36:50,262 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1050514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:37:00,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3707, 1.5601, 1.3189, 1.5909], device='cuda:0'), covar=tensor([0.0739, 0.0321, 0.0338, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 06:37:19,883 INFO [train.py:968] (0/2) Epoch 24, batch 1650, giga_loss[loss=0.3027, simple_loss=0.3676, pruned_loss=0.1189, over 28796.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3497, pruned_loss=0.09744, over 5711331.60 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3365, pruned_loss=0.08573, over 3458931.18 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3503, pruned_loss=0.09817, over 5700486.97 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:37:31,772 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 06:37:41,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.217e+02 1.286e+03 1.618e+03 1.997e+03 4.923e+03, threshold=3.236e+03, percent-clipped=10.0
+2023-03-12 06:38:03,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1050595.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:38:06,011 INFO [train.py:968] (0/2) Epoch 24, batch 1700, giga_loss[loss=0.2751, simple_loss=0.347, pruned_loss=0.1016, over 28920.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3523, pruned_loss=0.1012, over 5705710.36 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.337, pruned_loss=0.08591, over 3529637.54 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3529, pruned_loss=0.1021, over 5693372.82 frames. ], batch size: 186, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:38:13,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2159, 1.2731, 3.3797, 3.0386], device='cuda:0'), covar=tensor([0.1600, 0.2788, 0.0502, 0.1645], device='cuda:0'), in_proj_covar=tensor([0.0770, 0.0656, 0.0973, 0.0922], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 06:38:22,420 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3252, 3.1130, 1.4745, 1.5095], device='cuda:0'), covar=tensor([0.1046, 0.0327, 0.0902, 0.1374], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0557, 0.0393, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 06:38:38,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4297, 1.5007, 1.5234, 1.3694], device='cuda:0'), covar=tensor([0.2636, 0.2438, 0.2229, 0.2543], device='cuda:0'), in_proj_covar=tensor([0.1985, 0.1932, 0.1857, 0.2002], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 06:38:49,426 INFO [train.py:968] (0/2) Epoch 24, batch 1750, giga_loss[loss=0.2631, simple_loss=0.3376, pruned_loss=0.09427, over 28751.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3511, pruned_loss=0.1018, over 5699045.06 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.337, pruned_loss=0.08584, over 3588626.79 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3519, pruned_loss=0.103, over 5684256.03 frames. ], batch size: 284, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:39:11,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.941e+02 1.297e+03 1.590e+03 2.197e+03 6.383e+03, threshold=3.180e+03, percent-clipped=6.0
+2023-03-12 06:39:30,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0533, 2.4531, 1.6507, 2.0043], device='cuda:0'), covar=tensor([0.1019, 0.0641, 0.1050, 0.1117], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0448, 0.0521, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 06:39:33,352 INFO [train.py:968] (0/2) Epoch 24, batch 1800, giga_loss[loss=0.2374, simple_loss=0.3168, pruned_loss=0.07897, over 28612.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3485, pruned_loss=0.1005, over 5714522.89 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3364, pruned_loss=0.08568, over 3657404.30 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3499, pruned_loss=0.1019, over 5697358.71 frames. ], batch size: 307, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:40:09,640 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1050738.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:40:11,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1050741.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:40:15,794 INFO [train.py:968] (0/2) Epoch 24, batch 1850, giga_loss[loss=0.2628, simple_loss=0.3375, pruned_loss=0.09404, over 28847.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3477, pruned_loss=0.1001, over 5714475.77 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3361, pruned_loss=0.08553, over 3720103.00 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3493, pruned_loss=0.1018, over 5698875.50 frames. ], batch size: 199, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:40:30,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1050765.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:40:35,581 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1050770.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:40:36,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.306e+02 1.183e+03 1.526e+03 2.149e+03 3.436e+03, threshold=3.052e+03, percent-clipped=4.0
+2023-03-12 06:40:59,265 INFO [train.py:968] (0/2) Epoch 24, batch 1900, giga_loss[loss=0.2477, simple_loss=0.3315, pruned_loss=0.08192, over 28881.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3472, pruned_loss=0.09873, over 5722219.02 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3372, pruned_loss=0.0861, over 3781216.12 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3481, pruned_loss=0.1, over 5707762.23 frames. ], batch size: 227, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:41:47,870 INFO [train.py:968] (0/2) Epoch 24, batch 1950, giga_loss[loss=0.2689, simple_loss=0.3262, pruned_loss=0.1058, over 23711.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3458, pruned_loss=0.0978, over 5699916.06 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3376, pruned_loss=0.08619, over 3840508.70 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3465, pruned_loss=0.09913, over 5686281.70 frames. ], batch size: 705, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:41:52,482 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1050854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:41:54,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4024, 1.5304, 1.6116, 1.3562], device='cuda:0'), covar=tensor([0.2928, 0.2487, 0.2232, 0.2648], device='cuda:0'), in_proj_covar=tensor([0.1989, 0.1933, 0.1860, 0.2000], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 06:42:07,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.926e+02 1.252e+03 1.459e+03 2.018e+03 7.599e+03, threshold=2.917e+03, percent-clipped=9.0
+2023-03-12 06:42:17,419 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 06:42:23,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1050889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:42:31,024 INFO [train.py:968] (0/2) Epoch 24, batch 2000, giga_loss[loss=0.2348, simple_loss=0.3106, pruned_loss=0.07945, over 28955.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3399, pruned_loss=0.09399, over 5699766.03 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3371, pruned_loss=0.08583, over 3908463.53 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3409, pruned_loss=0.09556, over 5685486.98 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:42:39,300 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1050908.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:42:41,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1050911.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:42:44,655 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-12 06:42:53,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4867, 2.8564, 1.6941, 1.7244], device='cuda:0'), covar=tensor([0.0874, 0.0322, 0.0765, 0.1145], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0555, 0.0393, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 06:43:07,603 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1050940.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:43:14,892 INFO [train.py:968] (0/2) Epoch 24, batch 2050, giga_loss[loss=0.218, simple_loss=0.2993, pruned_loss=0.0683, over 28889.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3346, pruned_loss=0.09112, over 5693257.39 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3369, pruned_loss=0.08567, over 3989537.45 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3356, pruned_loss=0.09278, over 5679727.14 frames. ], batch size: 213, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:43:38,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.293e+02 1.089e+03 1.468e+03 2.213e+03 4.692e+03, threshold=2.935e+03, percent-clipped=10.0
+2023-03-12 06:44:02,549 INFO [train.py:968] (0/2) Epoch 24, batch 2100, libri_loss[loss=0.2499, simple_loss=0.3324, pruned_loss=0.08373, over 29540.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3296, pruned_loss=0.08865, over 5688280.51 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3371, pruned_loss=0.08588, over 4054623.27 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3301, pruned_loss=0.09, over 5671731.01 frames. ], batch size: 79, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:44:07,298 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-12 06:44:35,694 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1051032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:44:37,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1051035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:44:47,000 INFO [train.py:968] (0/2) Epoch 24, batch 2150, libri_loss[loss=0.283, simple_loss=0.3645, pruned_loss=0.1008, over 29255.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3287, pruned_loss=0.08753, over 5687992.97 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3373, pruned_loss=0.086, over 4096943.81 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3287, pruned_loss=0.08857, over 5680246.06 frames. ], batch size: 94, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:45:00,484 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1051064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:45:03,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4565, 3.5828, 1.6070, 1.6157], device='cuda:0'), covar=tensor([0.1084, 0.0315, 0.0940, 0.1384], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0555, 0.0392, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 06:45:08,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.612e+02 1.073e+03 1.257e+03 1.579e+03 3.835e+03, threshold=2.514e+03, percent-clipped=2.0
+2023-03-12 06:45:10,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6511, 4.5043, 4.2678, 2.1584], device='cuda:0'), covar=tensor([0.0528, 0.0653, 0.0604, 0.1977], device='cuda:0'), in_proj_covar=tensor([0.1234, 0.1144, 0.0960, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 06:45:27,876 INFO [train.py:968] (0/2) Epoch 24, batch 2200, giga_loss[loss=0.2499, simple_loss=0.3312, pruned_loss=0.08427, over 28914.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3296, pruned_loss=0.0878, over 5694712.04 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3374, pruned_loss=0.08588, over 4124069.89 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3294, pruned_loss=0.08873, over 5685569.55 frames. ], batch size: 145, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:45:58,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4026, 3.2330, 3.0397, 1.8948], device='cuda:0'), covar=tensor([0.0749, 0.0871, 0.0748, 0.1872], device='cuda:0'), in_proj_covar=tensor([0.1237, 0.1146, 0.0961, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 06:46:06,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1770, 0.7674, 0.9308, 1.4047], device='cuda:0'), covar=tensor([0.0803, 0.0401, 0.0377, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 06:46:09,347 INFO [train.py:968] (0/2) Epoch 24, batch 2250, giga_loss[loss=0.2682, simple_loss=0.3416, pruned_loss=0.0974, over 28682.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3289, pruned_loss=0.08775, over 5700028.49 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3372, pruned_loss=0.08582, over 4150678.60 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3288, pruned_loss=0.08853, over 5690158.13 frames. ], batch size: 242, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:46:22,940 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 06:46:31,067 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.479e+02 1.199e+03 1.454e+03 1.913e+03 5.042e+03, threshold=2.908e+03, percent-clipped=12.0
+2023-03-12 06:46:52,133 INFO [train.py:968] (0/2) Epoch 24, batch 2300, giga_loss[loss=0.2254, simple_loss=0.3034, pruned_loss=0.07368, over 28677.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3273, pruned_loss=0.08705, over 5702942.40 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3374, pruned_loss=0.08568, over 4200375.20 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3269, pruned_loss=0.08782, over 5698010.09 frames. ], batch size: 242, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:46:52,873 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1051199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:47:20,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1051229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:47:35,554 INFO [train.py:968] (0/2) Epoch 24, batch 2350, giga_loss[loss=0.2413, simple_loss=0.3259, pruned_loss=0.07833, over 28590.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3257, pruned_loss=0.08676, over 5705083.35 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.337, pruned_loss=0.08533, over 4224970.68 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3254, pruned_loss=0.08761, over 5699214.79 frames. ], batch size: 307, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:47:58,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.990e+02 1.110e+03 1.398e+03 2.102e+03 6.935e+03, threshold=2.796e+03, percent-clipped=8.0
+2023-03-12 06:48:16,526 INFO [train.py:968] (0/2) Epoch 24, batch 2400, giga_loss[loss=0.2345, simple_loss=0.3142, pruned_loss=0.07736, over 29016.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3262, pruned_loss=0.0872, over 5706733.62 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3388, pruned_loss=0.08624, over 4278796.28 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3243, pruned_loss=0.0873, over 5706250.51 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:48:55,409 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1051346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:48:56,420 INFO [train.py:968] (0/2) Epoch 24, batch 2450, giga_loss[loss=0.2517, simple_loss=0.3192, pruned_loss=0.09213, over 28843.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3242, pruned_loss=0.08595, over 5722232.99 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.339, pruned_loss=0.08606, over 4365609.51 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3218, pruned_loss=0.08617, over 5712611.07 frames. ], batch size: 186, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:49:16,341 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1051372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:49:18,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.317e+02 1.131e+03 1.495e+03 2.197e+03 7.857e+03, threshold=2.991e+03, percent-clipped=15.0
+2023-03-12 06:49:18,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1051375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:49:34,408 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-12 06:49:36,924 INFO [train.py:968] (0/2) Epoch 24, batch 2500, giga_loss[loss=0.2296, simple_loss=0.305, pruned_loss=0.07704, over 28882.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3215, pruned_loss=0.0845, over 5722786.08 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3389, pruned_loss=0.08597, over 4400973.66 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3192, pruned_loss=0.08473, over 5718881.74 frames. ], batch size: 99, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:49:39,505 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-12 06:49:42,044 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1051404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:49:42,744 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2100, 4.0482, 3.8451, 1.7825], device='cuda:0'), covar=tensor([0.0605, 0.0737, 0.0651, 0.2120], device='cuda:0'), in_proj_covar=tensor([0.1235, 0.1145, 0.0962, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 06:50:15,756 INFO [train.py:968] (0/2) Epoch 24, batch 2550, giga_loss[loss=0.2344, simple_loss=0.3105, pruned_loss=0.07918, over 28844.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3193, pruned_loss=0.08309, over 5710829.61 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3394, pruned_loss=0.08609, over 4432825.30 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3166, pruned_loss=0.08312, over 5719617.54 frames. ], batch size: 199, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:50:16,214 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 06:50:38,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.995e+02 1.049e+03 1.211e+03 1.434e+03 5.183e+03, threshold=2.422e+03, percent-clipped=6.0
+2023-03-12 06:50:57,814 INFO [train.py:968] (0/2) Epoch 24, batch 2600, giga_loss[loss=0.2373, simple_loss=0.3186, pruned_loss=0.07802, over 28374.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3171, pruned_loss=0.08203, over 5710940.52 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3399, pruned_loss=0.08621, over 4454847.57 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3143, pruned_loss=0.08194, over 5715286.50 frames. ], batch size: 368, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:51:37,869 INFO [train.py:968] (0/2) Epoch 24, batch 2650, libri_loss[loss=0.3172, simple_loss=0.4019, pruned_loss=0.1162, over 29282.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3166, pruned_loss=0.08135, over 5713184.80 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3409, pruned_loss=0.08653, over 4499598.88 frames. ], giga_tot_loss[loss=0.2372, simple_loss=0.3128, pruned_loss=0.08086, over 5715305.09 frames. ], batch size: 94, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:51:40,325 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-12 06:51:58,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1051574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:51:58,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.830e+02 1.024e+03 1.322e+03 1.659e+03 4.200e+03, threshold=2.644e+03, percent-clipped=7.0
+2023-03-12 06:51:59,779 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 06:52:16,396 INFO [train.py:968] (0/2) Epoch 24, batch 2700, libri_loss[loss=0.2618, simple_loss=0.3451, pruned_loss=0.08926, over 29572.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3157, pruned_loss=0.0808, over 5722431.43 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3403, pruned_loss=0.08611, over 4553259.85 frames. ], giga_tot_loss[loss=0.2365, simple_loss=0.3121, pruned_loss=0.08049, over 5718698.80 frames. ], batch size: 75, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:52:58,515 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-12 06:53:04,549 INFO [train.py:968] (0/2) Epoch 24, batch 2750, giga_loss[loss=0.2341, simple_loss=0.3105, pruned_loss=0.07882, over 28553.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.318, pruned_loss=0.08268, over 5711951.36 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3407, pruned_loss=0.08626, over 4565081.23 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3147, pruned_loss=0.0823, over 5708885.00 frames. ], batch size: 85, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:53:26,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.141e+02 1.282e+03 1.527e+03 2.090e+03 7.497e+03, threshold=3.054e+03, percent-clipped=13.0
+2023-03-12 06:53:47,702 INFO [train.py:968] (0/2) Epoch 24, batch 2800, giga_loss[loss=0.2986, simple_loss=0.3647, pruned_loss=0.1163, over 29071.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.324, pruned_loss=0.08682, over 5715047.51 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3411, pruned_loss=0.08653, over 4590364.67 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3207, pruned_loss=0.08633, over 5710345.03 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:53:54,203 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1051704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:54:06,543 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1051717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:54:08,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1051720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:54:09,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1051721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:54:34,111 INFO [train.py:968] (0/2) Epoch 24, batch 2850, giga_loss[loss=0.3397, simple_loss=0.4021, pruned_loss=0.1386, over 29045.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3319, pruned_loss=0.0922, over 5706059.96 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3412, pruned_loss=0.08658, over 4621440.08 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3289, pruned_loss=0.09185, over 5699055.67 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:54:34,966 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1051749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:54:59,337 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.682e+02 1.504e+03 1.904e+03 2.534e+03 5.409e+03, threshold=3.808e+03, percent-clipped=17.0
+2023-03-12 06:55:10,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5507, 1.7172, 1.2572, 1.2685], device='cuda:0'), covar=tensor([0.1015, 0.0581, 0.1053, 0.1104], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0447, 0.0521, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 06:55:19,818 INFO [train.py:968] (0/2) Epoch 24, batch 2900, giga_loss[loss=0.3156, simple_loss=0.3661, pruned_loss=0.1325, over 23584.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3395, pruned_loss=0.09677, over 5691931.46 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3423, pruned_loss=0.08719, over 4652081.27 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3362, pruned_loss=0.09624, over 5682674.69 frames. ], batch size: 710, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:56:10,833 INFO [train.py:968] (0/2) Epoch 24, batch 2950, giga_loss[loss=0.2976, simple_loss=0.3732, pruned_loss=0.1109, over 28343.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3455, pruned_loss=0.1001, over 5673357.33 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3424, pruned_loss=0.0873, over 4664626.91 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3428, pruned_loss=0.09968, over 5664193.12 frames. ], batch size: 368, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:56:20,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6978, 1.8415, 1.8891, 1.4637], device='cuda:0'), covar=tensor([0.1883, 0.2484, 0.1542, 0.1767], device='cuda:0'), in_proj_covar=tensor([0.0919, 0.0711, 0.0966, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 06:56:24,540 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1051864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:56:27,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1051867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:56:36,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.012e+02 1.204e+03 1.493e+03 1.923e+03 4.737e+03, threshold=2.986e+03, percent-clipped=1.0
+2023-03-12 06:56:48,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4000, 4.1884, 1.6883, 1.5193], device='cuda:0'), covar=tensor([0.1098, 0.0230, 0.0934, 0.1453], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0556, 0.0394, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 06:56:53,661 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1051896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:56:54,818 INFO [train.py:968] (0/2) Epoch 24, batch 3000, giga_loss[loss=0.3386, simple_loss=0.4068, pruned_loss=0.1352, over 28727.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3491, pruned_loss=0.101, over 5681657.15 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3418, pruned_loss=0.08698, over 4688260.64 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3476, pruned_loss=0.1012, over 5675425.99 frames. ], batch size: 262, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:56:54,822 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 06:57:00,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3913, 1.6403, 1.2232, 1.2958], device='cuda:0'), covar=tensor([0.0993, 0.0443, 0.0997, 0.0949], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0448, 0.0522, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 06:57:04,256 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2095, simple_loss=0.3175, pruned_loss=0.05072, over 944034.00 frames. 
+2023-03-12 06:57:04,257 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 06:57:47,799 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4791, 1.6655, 1.7153, 1.2979], device='cuda:0'), covar=tensor([0.1874, 0.2783, 0.1612, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.0918, 0.0711, 0.0965, 0.0861], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 06:57:52,027 INFO [train.py:968] (0/2) Epoch 24, batch 3050, giga_loss[loss=0.2749, simple_loss=0.3521, pruned_loss=0.09885, over 28911.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3536, pruned_loss=0.1035, over 5677943.28 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3421, pruned_loss=0.08713, over 4724354.29 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3525, pruned_loss=0.1041, over 5671094.32 frames. ], batch size: 145, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:58:01,767 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-12 06:58:18,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.765e+02 1.362e+03 1.859e+03 2.311e+03 6.977e+03, threshold=3.719e+03, percent-clipped=15.0
+2023-03-12 06:58:35,420 INFO [train.py:968] (0/2) Epoch 24, batch 3100, giga_loss[loss=0.2288, simple_loss=0.3089, pruned_loss=0.07432, over 28304.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3498, pruned_loss=0.1005, over 5684310.37 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3417, pruned_loss=0.08703, over 4753215.10 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3495, pruned_loss=0.1014, over 5673911.93 frames. ], batch size: 368, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 06:58:37,750 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1052000.pt
+2023-03-12 06:58:39,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1052001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:59:19,698 INFO [train.py:968] (0/2) Epoch 24, batch 3150, giga_loss[loss=0.2335, simple_loss=0.3204, pruned_loss=0.07326, over 29076.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3464, pruned_loss=0.09773, over 5684742.33 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3417, pruned_loss=0.08705, over 4764421.79 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3462, pruned_loss=0.09855, over 5675005.38 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 06:59:44,816 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.743e+02 1.252e+03 1.514e+03 2.130e+03 1.302e+04, threshold=3.028e+03, percent-clipped=8.0
+2023-03-12 06:59:46,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1052079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:00:02,098 INFO [train.py:968] (0/2) Epoch 24, batch 3200, giga_loss[loss=0.2632, simple_loss=0.3388, pruned_loss=0.09384, over 28827.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3456, pruned_loss=0.09659, over 5670719.37 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3418, pruned_loss=0.08701, over 4783122.13 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3455, pruned_loss=0.09752, over 5670625.96 frames. ], batch size: 60, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:00:07,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4445, 4.2903, 1.5736, 1.6199], device='cuda:0'), covar=tensor([0.1034, 0.0285, 0.0914, 0.1347], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0553, 0.0392, 0.0430], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 07:00:43,851 INFO [train.py:968] (0/2) Epoch 24, batch 3250, giga_loss[loss=0.2576, simple_loss=0.3389, pruned_loss=0.08813, over 28699.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3454, pruned_loss=0.09622, over 5674550.54 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3412, pruned_loss=0.08681, over 4804993.28 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3458, pruned_loss=0.09726, over 5670926.73 frames. ], batch size: 284, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:01:10,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.975e+02 1.259e+03 1.593e+03 2.026e+03 1.489e+04, threshold=3.186e+03, percent-clipped=15.0
+2023-03-12 07:01:26,058 INFO [train.py:968] (0/2) Epoch 24, batch 3300, giga_loss[loss=0.2648, simple_loss=0.3486, pruned_loss=0.09056, over 28957.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3475, pruned_loss=0.09763, over 5682887.76 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3408, pruned_loss=0.08672, over 4847650.45 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3483, pruned_loss=0.09895, over 5672025.44 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:01:49,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1052222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:01:51,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1052225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:01:52,516 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1052227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:02:11,419 INFO [train.py:968] (0/2) Epoch 24, batch 3350, giga_loss[loss=0.2811, simple_loss=0.342, pruned_loss=0.1101, over 28676.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3505, pruned_loss=0.1002, over 5694816.94 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3402, pruned_loss=0.08637, over 4867557.61 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3518, pruned_loss=0.1017, over 5683378.84 frames. ], batch size: 85, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:02:17,071 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1052254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:02:38,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.277e+02 1.365e+03 1.738e+03 2.469e+03 5.858e+03, threshold=3.476e+03, percent-clipped=14.0
+2023-03-12 07:02:40,363 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6850, 1.7412, 1.2694, 1.3007], device='cuda:0'), covar=tensor([0.0955, 0.0611, 0.1064, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0444, 0.0519, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 07:02:54,992 INFO [train.py:968] (0/2) Epoch 24, batch 3400, giga_loss[loss=0.2983, simple_loss=0.3678, pruned_loss=0.1144, over 28777.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3524, pruned_loss=0.102, over 5694073.01 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3404, pruned_loss=0.08639, over 4882769.68 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3534, pruned_loss=0.1034, over 5682404.97 frames. ], batch size: 284, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:03:00,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6679, 4.5163, 4.3315, 2.0747], device='cuda:0'), covar=tensor([0.0646, 0.0789, 0.0926, 0.1999], device='cuda:0'), in_proj_covar=tensor([0.1238, 0.1152, 0.0966, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 07:03:37,519 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1052345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:03:39,416 INFO [train.py:968] (0/2) Epoch 24, batch 3450, giga_loss[loss=0.269, simple_loss=0.3402, pruned_loss=0.09891, over 28673.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3533, pruned_loss=0.1028, over 5691057.29 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3408, pruned_loss=0.08647, over 4904964.52 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3541, pruned_loss=0.1042, over 5679980.48 frames. ], batch size: 78, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:04:00,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7409, 1.4656, 5.2900, 3.7852], device='cuda:0'), covar=tensor([0.1765, 0.2962, 0.0352, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0658, 0.0974, 0.0930], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 07:04:04,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1052376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:04:04,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.594e+02 1.381e+03 1.768e+03 2.347e+03 4.942e+03, threshold=3.536e+03, percent-clipped=5.0
+2023-03-12 07:04:21,524 INFO [train.py:968] (0/2) Epoch 24, batch 3500, giga_loss[loss=0.2598, simple_loss=0.3424, pruned_loss=0.08861, over 28714.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3534, pruned_loss=0.1031, over 5676844.89 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3409, pruned_loss=0.08647, over 4912695.59 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3541, pruned_loss=0.1045, over 5675714.73 frames. ], batch size: 262, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:05:04,999 INFO [train.py:968] (0/2) Epoch 24, batch 3550, giga_loss[loss=0.2946, simple_loss=0.3696, pruned_loss=0.1098, over 28591.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3529, pruned_loss=0.1019, over 5687492.15 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3406, pruned_loss=0.0864, over 4937217.15 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.354, pruned_loss=0.1034, over 5681761.79 frames. ], batch size: 78, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:05:27,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.506e+02 1.265e+03 1.623e+03 2.268e+03 8.843e+03, threshold=3.246e+03, percent-clipped=8.0
+2023-03-12 07:05:44,222 INFO [train.py:968] (0/2) Epoch 24, batch 3600, libri_loss[loss=0.2307, simple_loss=0.3087, pruned_loss=0.07633, over 28511.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3531, pruned_loss=0.1011, over 5696776.89 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3408, pruned_loss=0.08651, over 4974475.03 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3543, pruned_loss=0.1028, over 5685272.33 frames. ], batch size: 63, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:06:02,923 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1052519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:06:04,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1052522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:06:21,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4625, 4.2739, 4.0462, 2.2251], device='cuda:0'), covar=tensor([0.0547, 0.0741, 0.0749, 0.1853], device='cuda:0'), in_proj_covar=tensor([0.1239, 0.1149, 0.0963, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 07:06:27,152 INFO [train.py:968] (0/2) Epoch 24, batch 3650, giga_loss[loss=0.2683, simple_loss=0.3397, pruned_loss=0.09851, over 27559.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3524, pruned_loss=0.09986, over 5697346.93 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.341, pruned_loss=0.08652, over 4999029.49 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3536, pruned_loss=0.1016, over 5685457.96 frames. ], batch size: 472, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:06:31,280 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1052551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:06:52,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.980e+02 1.185e+03 1.414e+03 1.894e+03 5.301e+03, threshold=2.828e+03, percent-clipped=5.0
+2023-03-12 07:07:11,523 INFO [train.py:968] (0/2) Epoch 24, batch 3700, giga_loss[loss=0.277, simple_loss=0.3468, pruned_loss=0.1035, over 27921.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.35, pruned_loss=0.09874, over 5698849.14 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3409, pruned_loss=0.08646, over 5010705.76 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3511, pruned_loss=0.1003, over 5688457.36 frames. ], batch size: 412, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:07:14,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1052602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:07:55,130 INFO [train.py:968] (0/2) Epoch 24, batch 3750, giga_loss[loss=0.2503, simple_loss=0.3221, pruned_loss=0.08926, over 28096.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3475, pruned_loss=0.09739, over 5704234.05 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3412, pruned_loss=0.08661, over 5040150.74 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3484, pruned_loss=0.09895, over 5690490.10 frames. ], batch size: 77, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:08:00,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2840, 1.1847, 1.1811, 1.4508], device='cuda:0'), covar=tensor([0.0793, 0.0380, 0.0350, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:0')
+2023-03-12 07:08:00,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1845, 1.7370, 1.3631, 0.4111], device='cuda:0'), covar=tensor([0.4969, 0.2998, 0.4340, 0.6411], device='cuda:0'), in_proj_covar=tensor([0.1762, 0.1652, 0.1598, 0.1429], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 07:08:17,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.968e+02 1.130e+03 1.468e+03 1.832e+03 5.282e+03, threshold=2.935e+03, percent-clipped=3.0
+2023-03-12 07:08:31,480 INFO [train.py:968] (0/2) Epoch 24, batch 3800, giga_loss[loss=0.2321, simple_loss=0.317, pruned_loss=0.07355, over 28606.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3447, pruned_loss=0.09588, over 5708809.59 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.341, pruned_loss=0.0865, over 5060186.35 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3457, pruned_loss=0.09743, over 5694777.15 frames. ], batch size: 60, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:08:52,104 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 07:08:52,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1052720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:09:12,927 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1052745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:09:14,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1534, 1.5355, 1.4643, 1.0305], device='cuda:0'), covar=tensor([0.1766, 0.2925, 0.1574, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0711, 0.0964, 0.0861], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 07:09:14,734 INFO [train.py:968] (0/2) Epoch 24, batch 3850, libri_loss[loss=0.2935, simple_loss=0.3725, pruned_loss=0.1072, over 29389.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3466, pruned_loss=0.09749, over 5704071.10 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3415, pruned_loss=0.08673, over 5069062.54 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.347, pruned_loss=0.09867, over 5693685.75 frames. ], batch size: 92, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:09:15,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1052748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:09:39,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1052777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:09:40,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.980e+02 1.091e+03 1.305e+03 1.745e+03 5.111e+03, threshold=2.609e+03, percent-clipped=7.0
+2023-03-12 07:09:56,357 INFO [train.py:968] (0/2) Epoch 24, batch 3900, giga_loss[loss=0.2535, simple_loss=0.3295, pruned_loss=0.08877, over 28676.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3469, pruned_loss=0.09732, over 5702160.32 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3419, pruned_loss=0.08709, over 5083651.61 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.347, pruned_loss=0.09826, over 5698287.02 frames. ], batch size: 92, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:10:12,744 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6234, 1.9564, 1.8292, 1.7076], device='cuda:0'), covar=tensor([0.2134, 0.1919, 0.2379, 0.2154], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0751, 0.0719, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 07:10:36,465 INFO [train.py:968] (0/2) Epoch 24, batch 3950, giga_loss[loss=0.2747, simple_loss=0.3456, pruned_loss=0.1019, over 28905.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3467, pruned_loss=0.09655, over 5707757.91 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3419, pruned_loss=0.08708, over 5100213.77 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3469, pruned_loss=0.09747, over 5700916.59 frames. ], batch size: 112, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:10:51,404 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1052863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:10:54,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1052866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:11:04,562 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.015e+02 1.107e+03 1.365e+03 1.737e+03 5.317e+03, threshold=2.730e+03, percent-clipped=11.0
+2023-03-12 07:11:16,537 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1052895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:11:18,302 INFO [train.py:968] (0/2) Epoch 24, batch 4000, giga_loss[loss=0.2334, simple_loss=0.3182, pruned_loss=0.07433, over 28647.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3452, pruned_loss=0.0953, over 5715675.92 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3417, pruned_loss=0.08695, over 5129658.52 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3457, pruned_loss=0.09648, over 5705047.06 frames. ], batch size: 85, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:11:59,937 INFO [train.py:968] (0/2) Epoch 24, batch 4050, giga_loss[loss=0.2678, simple_loss=0.3435, pruned_loss=0.096, over 28966.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3451, pruned_loss=0.09574, over 5709657.60 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3416, pruned_loss=0.08704, over 5148871.76 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3457, pruned_loss=0.09681, over 5696853.18 frames. ], batch size: 106, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:12:08,819 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1052960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:12:21,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1743, 1.7747, 1.4415, 0.3897], device='cuda:0'), covar=tensor([0.5226, 0.2843, 0.4218, 0.6287], device='cuda:0'), in_proj_covar=tensor([0.1762, 0.1647, 0.1600, 0.1431], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 07:12:24,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.096e+02 1.105e+03 1.315e+03 1.736e+03 3.761e+03, threshold=2.630e+03, percent-clipped=6.0
+2023-03-12 07:12:40,076 INFO [train.py:968] (0/2) Epoch 24, batch 4100, giga_loss[loss=0.2385, simple_loss=0.3171, pruned_loss=0.07998, over 28893.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3437, pruned_loss=0.09545, over 5719963.11 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3417, pruned_loss=0.08721, over 5164098.73 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.344, pruned_loss=0.09629, over 5706078.37 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:13:19,917 INFO [train.py:968] (0/2) Epoch 24, batch 4150, giga_loss[loss=0.243, simple_loss=0.3199, pruned_loss=0.08307, over 28878.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3404, pruned_loss=0.09368, over 5719919.82 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3422, pruned_loss=0.08744, over 5177027.46 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3404, pruned_loss=0.0943, over 5707231.72 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:13:23,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4432, 1.8571, 1.3065, 0.8385], device='cuda:0'), covar=tensor([0.5954, 0.3152, 0.4153, 0.6358], device='cuda:0'), in_proj_covar=tensor([0.1765, 0.1651, 0.1605, 0.1434], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 07:13:24,700 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-12 07:13:47,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.590e+02 1.277e+03 1.669e+03 2.606e+03 6.385e+03, threshold=3.338e+03, percent-clipped=23.0
+2023-03-12 07:13:48,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5913, 2.2588, 1.8692, 0.8377], device='cuda:0'), covar=tensor([0.6666, 0.3028, 0.3783, 0.6513], device='cuda:0'), in_proj_covar=tensor([0.1768, 0.1655, 0.1608, 0.1438], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 07:14:03,061 INFO [train.py:968] (0/2) Epoch 24, batch 4200, giga_loss[loss=0.2924, simple_loss=0.367, pruned_loss=0.1089, over 27916.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3381, pruned_loss=0.09234, over 5707783.35 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3426, pruned_loss=0.08762, over 5180242.22 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3377, pruned_loss=0.0928, over 5707489.32 frames. ], batch size: 412, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:14:15,049 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 07:14:20,589 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:14:28,797 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5385, 2.8049, 2.5558, 2.1138], device='cuda:0'), covar=tensor([0.2875, 0.2050, 0.2128, 0.2674], device='cuda:0'), in_proj_covar=tensor([0.1982, 0.1931, 0.1864, 0.2007], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 07:14:43,879 INFO [train.py:968] (0/2) Epoch 24, batch 4250, giga_loss[loss=0.2584, simple_loss=0.337, pruned_loss=0.08985, over 28855.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3386, pruned_loss=0.09297, over 5708980.47 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3428, pruned_loss=0.08771, over 5187196.41 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3381, pruned_loss=0.09332, over 5709390.60 frames. ], batch size: 145, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:15:09,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.539e+02 1.260e+03 1.393e+03 1.853e+03 5.765e+03, threshold=2.786e+03, percent-clipped=6.0
+2023-03-12 07:15:27,079 INFO [train.py:968] (0/2) Epoch 24, batch 4300, giga_loss[loss=0.2284, simple_loss=0.31, pruned_loss=0.07342, over 28997.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3371, pruned_loss=0.09277, over 5713402.55 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3429, pruned_loss=0.08798, over 5207036.35 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3365, pruned_loss=0.09297, over 5709769.61 frames. ], batch size: 213, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:15:35,308 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6224, 1.7504, 1.5095, 1.7249], device='cuda:0'), covar=tensor([0.2614, 0.2791, 0.3009, 0.2531], device='cuda:0'), in_proj_covar=tensor([0.1541, 0.1111, 0.1361, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 07:16:00,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7605, 1.1605, 5.0890, 3.7010], device='cuda:0'), covar=tensor([0.1596, 0.3047, 0.0359, 0.0766], device='cuda:0'), in_proj_covar=tensor([0.0770, 0.0654, 0.0968, 0.0926], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 07:16:06,327 INFO [train.py:968] (0/2) Epoch 24, batch 4350, giga_loss[loss=0.2385, simple_loss=0.3164, pruned_loss=0.08031, over 28929.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3362, pruned_loss=0.0928, over 5708389.27 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3436, pruned_loss=0.08847, over 5216417.96 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.335, pruned_loss=0.09266, over 5705839.11 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:16:10,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4077, 1.5894, 1.5729, 1.3048], device='cuda:0'), covar=tensor([0.3567, 0.2923, 0.2400, 0.3092], device='cuda:0'), in_proj_covar=tensor([0.1990, 0.1940, 0.1873, 0.2017], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 07:16:22,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3528, 1.8623, 1.2828, 0.6228], device='cuda:0'), covar=tensor([0.5784, 0.2860, 0.3701, 0.6558], device='cuda:0'), in_proj_covar=tensor([0.1770, 0.1663, 0.1613, 0.1442], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 07:16:30,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.137e+02 1.182e+03 1.436e+03 1.832e+03 4.589e+03, threshold=2.872e+03, percent-clipped=7.0
+2023-03-12 07:16:46,186 INFO [train.py:968] (0/2) Epoch 24, batch 4400, giga_loss[loss=0.2438, simple_loss=0.3194, pruned_loss=0.08407, over 28443.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3341, pruned_loss=0.09199, over 5711451.18 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3436, pruned_loss=0.0885, over 5236030.04 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3329, pruned_loss=0.09198, over 5704877.88 frames. ], batch size: 65, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:17:17,488 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1053335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:17:26,641 INFO [train.py:968] (0/2) Epoch 24, batch 4450, giga_loss[loss=0.2568, simple_loss=0.3403, pruned_loss=0.0867, over 28593.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3316, pruned_loss=0.09053, over 5710456.02 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3436, pruned_loss=0.08842, over 5253156.58 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3304, pruned_loss=0.09066, over 5702180.69 frames. ], batch size: 307, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:17:50,300 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.039e+02 1.144e+03 1.414e+03 1.780e+03 3.853e+03, threshold=2.828e+03, percent-clipped=4.0
+2023-03-12 07:18:07,540 INFO [train.py:968] (0/2) Epoch 24, batch 4500, giga_loss[loss=0.2576, simple_loss=0.3375, pruned_loss=0.08886, over 29056.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.332, pruned_loss=0.0901, over 5703085.40 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3436, pruned_loss=0.08844, over 5258826.49 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3307, pruned_loss=0.09025, over 5702687.91 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:18:15,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7329, 4.5674, 4.3199, 2.3277], device='cuda:0'), covar=tensor([0.0523, 0.0715, 0.0734, 0.1734], device='cuda:0'), in_proj_covar=tensor([0.1246, 0.1149, 0.0967, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 07:18:44,694 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3164, 1.5232, 1.3928, 1.6084], device='cuda:0'), covar=tensor([0.0751, 0.0313, 0.0330, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0223], device='cuda:0'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0110], device='cuda:0')
+2023-03-12 07:18:49,288 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 07:18:55,143 INFO [train.py:968] (0/2) Epoch 24, batch 4550, giga_loss[loss=0.2875, simple_loss=0.3616, pruned_loss=0.1067, over 27926.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3336, pruned_loss=0.09078, over 5707814.45 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3436, pruned_loss=0.08854, over 5268502.33 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3324, pruned_loss=0.09084, over 5704778.97 frames. ], batch size: 412, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:19:19,451 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1053478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:19:20,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.354e+02 1.171e+03 1.414e+03 1.790e+03 4.235e+03, threshold=2.828e+03, percent-clipped=11.0
+2023-03-12 07:19:21,601 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1053481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:19:35,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1053495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:19:37,962 INFO [train.py:968] (0/2) Epoch 24, batch 4600, giga_loss[loss=0.2637, simple_loss=0.3473, pruned_loss=0.09002, over 29044.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3359, pruned_loss=0.09119, over 5716386.64 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3429, pruned_loss=0.08833, over 5286532.68 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3353, pruned_loss=0.09147, over 5709746.36 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:19:45,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:19:48,113 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1053510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:20:21,605 INFO [train.py:968] (0/2) Epoch 24, batch 4650, libri_loss[loss=0.2427, simple_loss=0.3302, pruned_loss=0.07759, over 29577.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3397, pruned_loss=0.09318, over 5709017.95 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3434, pruned_loss=0.0886, over 5304999.53 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3386, pruned_loss=0.09332, over 5702083.43 frames. ], batch size: 75, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:20:28,447 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 07:20:52,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.934e+02 1.187e+03 1.463e+03 1.855e+03 6.996e+03, threshold=2.927e+03, percent-clipped=6.0
+2023-03-12 07:21:07,296 INFO [train.py:968] (0/2) Epoch 24, batch 4700, giga_loss[loss=0.247, simple_loss=0.3287, pruned_loss=0.08262, over 28800.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3393, pruned_loss=0.09246, over 5699429.50 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3436, pruned_loss=0.08887, over 5310802.85 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3381, pruned_loss=0.09237, over 5692395.14 frames. ], batch size: 186, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:21:38,221 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1053638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:21:41,217 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1053641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:21:46,429 INFO [train.py:968] (0/2) Epoch 24, batch 4750, giga_loss[loss=0.2194, simple_loss=0.3114, pruned_loss=0.06375, over 28954.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3384, pruned_loss=0.09136, over 5707882.00 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3431, pruned_loss=0.08857, over 5340072.36 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3378, pruned_loss=0.09169, over 5694628.17 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:22:03,200 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1053670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:22:10,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.618e+02 1.242e+03 1.750e+03 2.248e+03 5.002e+03, threshold=3.500e+03, percent-clipped=10.0
+2023-03-12 07:22:25,313 INFO [train.py:968] (0/2) Epoch 24, batch 4800, giga_loss[loss=0.2655, simple_loss=0.3433, pruned_loss=0.09384, over 28873.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3389, pruned_loss=0.09139, over 5714798.67 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3432, pruned_loss=0.08861, over 5361203.41 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3381, pruned_loss=0.09173, over 5699584.94 frames. ], batch size: 145, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:23:05,928 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:23:07,824 INFO [train.py:968] (0/2) Epoch 24, batch 4850, giga_loss[loss=0.263, simple_loss=0.3378, pruned_loss=0.09405, over 28970.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3391, pruned_loss=0.09133, over 5720581.57 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3436, pruned_loss=0.08888, over 5378395.30 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3379, pruned_loss=0.09147, over 5703984.72 frames. ], batch size: 213, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:23:23,870 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-12 07:23:32,738 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053777.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 07:23:35,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.195e+02 1.259e+03 1.761e+03 2.223e+03 4.856e+03, threshold=3.523e+03, percent-clipped=6.0
+2023-03-12 07:23:41,057 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 07:23:50,947 INFO [train.py:968] (0/2) Epoch 24, batch 4900, giga_loss[loss=0.3047, simple_loss=0.3757, pruned_loss=0.1169, over 28262.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3417, pruned_loss=0.0936, over 5712241.75 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3436, pruned_loss=0.08872, over 5386145.91 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3407, pruned_loss=0.09396, over 5701867.09 frames. ], batch size: 368, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:24:07,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4753, 1.8198, 1.4324, 1.6267], device='cuda:0'), covar=tensor([0.2491, 0.2503, 0.2948, 0.2263], device='cuda:0'), in_proj_covar=tensor([0.1539, 0.1108, 0.1358, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 07:24:32,492 INFO [train.py:968] (0/2) Epoch 24, batch 4950, giga_loss[loss=0.2732, simple_loss=0.3548, pruned_loss=0.09578, over 28967.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3446, pruned_loss=0.09509, over 5716902.67 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3433, pruned_loss=0.08857, over 5398859.33 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.344, pruned_loss=0.09564, over 5706011.98 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:24:51,484 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:25:01,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.126e+02 1.279e+03 1.646e+03 2.236e+03 5.593e+03, threshold=3.292e+03, percent-clipped=4.0
+2023-03-12 07:25:04,932 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1053884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:25:10,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053890.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 07:25:15,834 INFO [train.py:968] (0/2) Epoch 24, batch 5000, libri_loss[loss=0.3227, simple_loss=0.3892, pruned_loss=0.1281, over 20405.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3468, pruned_loss=0.09629, over 5710038.83 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3438, pruned_loss=0.08895, over 5399912.49 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3459, pruned_loss=0.09651, over 5706861.87 frames. ], batch size: 187, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:25:23,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8395, 1.3741, 1.2677, 1.0886], device='cuda:0'), covar=tensor([0.1989, 0.1137, 0.2179, 0.1705], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0749, 0.0716, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 07:25:57,859 INFO [train.py:968] (0/2) Epoch 24, batch 5050, giga_loss[loss=0.2567, simple_loss=0.3334, pruned_loss=0.08999, over 28476.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3486, pruned_loss=0.09736, over 5711355.22 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3436, pruned_loss=0.08886, over 5407597.70 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3482, pruned_loss=0.09775, over 5706080.06 frames. ], batch size: 60, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:26:26,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.573e+02 1.328e+03 1.628e+03 2.135e+03 6.013e+03, threshold=3.257e+03, percent-clipped=5.0
+2023-03-12 07:26:41,870 INFO [train.py:968] (0/2) Epoch 24, batch 5100, giga_loss[loss=0.3147, simple_loss=0.3793, pruned_loss=0.125, over 27722.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3496, pruned_loss=0.09817, over 5708248.68 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3438, pruned_loss=0.08884, over 5412252.05 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3492, pruned_loss=0.09855, over 5702800.70 frames. ], batch size: 472, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:26:43,592 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1054000.pt
+2023-03-12 07:26:59,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0556, 1.2496, 3.3927, 3.0625], device='cuda:0'), covar=tensor([0.1768, 0.2788, 0.0512, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0769, 0.0654, 0.0970, 0.0928], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 07:27:07,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:27:08,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 07:27:08,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:27:23,138 INFO [train.py:968] (0/2) Epoch 24, batch 5150, libri_loss[loss=0.2745, simple_loss=0.3587, pruned_loss=0.09512, over 29535.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3501, pruned_loss=0.0981, over 5702758.80 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3445, pruned_loss=0.08905, over 5416615.92 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3493, pruned_loss=0.09847, over 5702129.36 frames. ], batch size: 81, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:27:33,036 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:27:50,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.394e+02 1.184e+03 1.516e+03 1.896e+03 4.159e+03, threshold=3.031e+03, percent-clipped=3.0
+2023-03-12 07:27:54,619 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1054086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:28:04,151 INFO [train.py:968] (0/2) Epoch 24, batch 5200, giga_loss[loss=0.242, simple_loss=0.3176, pruned_loss=0.08318, over 28805.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3482, pruned_loss=0.09736, over 5709841.36 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3446, pruned_loss=0.08901, over 5425797.92 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3476, pruned_loss=0.09785, over 5706287.88 frames. ], batch size: 112, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:28:16,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6588, 1.9791, 2.0156, 1.6762], device='cuda:0'), covar=tensor([0.3315, 0.2640, 0.2528, 0.2825], device='cuda:0'), in_proj_covar=tensor([0.1983, 0.1928, 0.1866, 0.2001], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 07:28:24,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1054120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:28:46,598 INFO [train.py:968] (0/2) Epoch 24, batch 5250, giga_loss[loss=0.248, simple_loss=0.3231, pruned_loss=0.08641, over 28819.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3449, pruned_loss=0.09584, over 5711753.12 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3451, pruned_loss=0.08924, over 5437036.42 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.344, pruned_loss=0.0962, over 5704974.03 frames. ], batch size: 186, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:28:50,344 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1054152.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:29:12,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.372e+02 1.241e+03 1.506e+03 1.952e+03 4.126e+03, threshold=3.013e+03, percent-clipped=4.0
+2023-03-12 07:29:17,252 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 07:29:28,218 INFO [train.py:968] (0/2) Epoch 24, batch 5300, giga_loss[loss=0.2588, simple_loss=0.3504, pruned_loss=0.08363, over 28910.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.343, pruned_loss=0.09508, over 5715211.07 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3453, pruned_loss=0.0895, over 5446287.06 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3421, pruned_loss=0.09528, over 5706297.23 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:30:08,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1054245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:30:10,372 INFO [train.py:968] (0/2) Epoch 24, batch 5350, giga_loss[loss=0.222, simple_loss=0.3114, pruned_loss=0.06633, over 28846.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3432, pruned_loss=0.09411, over 5716298.76 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3452, pruned_loss=0.08956, over 5459355.83 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3424, pruned_loss=0.09435, over 5704636.80 frames. ], batch size: 112, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:30:24,822 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:30:26,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1054265.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 07:30:26,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:30:40,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0518, 3.1477, 1.9999, 1.0071], device='cuda:0'), covar=tensor([0.8429, 0.3028, 0.4139, 0.8027], device='cuda:0'), in_proj_covar=tensor([0.1776, 0.1668, 0.1618, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 07:30:42,344 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.166e+02 1.203e+03 1.496e+03 1.904e+03 4.541e+03, threshold=2.993e+03, percent-clipped=6.0
+2023-03-12 07:30:52,965 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:30:53,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054295.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:30:55,844 INFO [train.py:968] (0/2) Epoch 24, batch 5400, giga_loss[loss=0.2562, simple_loss=0.3451, pruned_loss=0.08362, over 28667.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3442, pruned_loss=0.09342, over 5718197.11 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3452, pruned_loss=0.08955, over 5464483.23 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3437, pruned_loss=0.09367, over 5707802.44 frames. ], batch size: 284, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:30:56,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054298.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:31:09,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6882, 2.1699, 1.9645, 1.5052], device='cuda:0'), covar=tensor([0.1951, 0.2553, 0.1635, 0.1999], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0707, 0.0961, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 07:31:20,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054327.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:31:32,170 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0276, 1.3086, 1.1331, 0.3292], device='cuda:0'), covar=tensor([0.4175, 0.3093, 0.4882, 0.6696], device='cuda:0'), in_proj_covar=tensor([0.1779, 0.1673, 0.1621, 0.1448], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 07:31:38,484 INFO [train.py:968] (0/2) Epoch 24, batch 5450, giga_loss[loss=0.2555, simple_loss=0.3366, pruned_loss=0.0872, over 28902.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3424, pruned_loss=0.09296, over 5722975.14 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.345, pruned_loss=0.08951, over 5473641.26 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3421, pruned_loss=0.09327, over 5711273.01 frames. ], batch size: 145, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:32:09,266 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.945e+02 1.327e+03 1.583e+03 2.016e+03 6.073e+03, threshold=3.166e+03, percent-clipped=7.0
+2023-03-12 07:32:13,341 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:32:15,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:32:21,250 INFO [train.py:968] (0/2) Epoch 24, batch 5500, giga_loss[loss=0.2697, simple_loss=0.341, pruned_loss=0.09927, over 28939.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3419, pruned_loss=0.09454, over 5725948.46 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.345, pruned_loss=0.08951, over 5476366.41 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3416, pruned_loss=0.09481, over 5716677.76 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:32:31,367 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054408.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 07:32:34,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054411.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 07:32:43,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:32:59,657 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054440.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:33:06,662 INFO [train.py:968] (0/2) Epoch 24, batch 5550, giga_loss[loss=0.3433, simple_loss=0.3886, pruned_loss=0.149, over 26677.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3402, pruned_loss=0.09481, over 5730503.51 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3452, pruned_loss=0.08959, over 5485009.30 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3397, pruned_loss=0.09507, over 5719883.66 frames. ], batch size: 555, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:33:17,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1054461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:33:34,832 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.549e+02 1.145e+03 1.467e+03 2.083e+03 7.432e+03, threshold=2.934e+03, percent-clipped=5.0
+2023-03-12 07:33:35,802 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 07:33:48,022 INFO [train.py:968] (0/2) Epoch 24, batch 5600, giga_loss[loss=0.2954, simple_loss=0.3581, pruned_loss=0.1163, over 27570.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3375, pruned_loss=0.09424, over 5735104.24 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3454, pruned_loss=0.08974, over 5497895.21 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3367, pruned_loss=0.09444, over 5721508.68 frames. ], batch size: 472, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:34:30,736 INFO [train.py:968] (0/2) Epoch 24, batch 5650, giga_loss[loss=0.2053, simple_loss=0.2777, pruned_loss=0.0665, over 28511.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3378, pruned_loss=0.09439, over 5729914.70 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3453, pruned_loss=0.08953, over 5514601.54 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.337, pruned_loss=0.09491, over 5713266.51 frames. ], batch size: 71, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:34:58,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.410e+02 1.196e+03 1.532e+03 1.957e+03 5.364e+03, threshold=3.065e+03, percent-clipped=10.0
+2023-03-12 07:35:09,661 INFO [train.py:968] (0/2) Epoch 24, batch 5700, giga_loss[loss=0.2447, simple_loss=0.3145, pruned_loss=0.08742, over 28856.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3338, pruned_loss=0.09226, over 5729850.60 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3451, pruned_loss=0.0895, over 5524260.73 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3331, pruned_loss=0.0928, over 5712593.08 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:35:17,142 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:35:19,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:35:42,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:35:52,847 INFO [train.py:968] (0/2) Epoch 24, batch 5750, giga_loss[loss=0.2512, simple_loss=0.3195, pruned_loss=0.09147, over 27578.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3306, pruned_loss=0.09105, over 5713443.30 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3453, pruned_loss=0.08965, over 5518792.84 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3297, pruned_loss=0.09136, over 5706044.22 frames. ], batch size: 472, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:36:22,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.938e+02 1.300e+03 1.727e+03 2.676e+03 9.538e+03, threshold=3.455e+03, percent-clipped=17.0
+2023-03-12 07:36:32,193 INFO [train.py:968] (0/2) Epoch 24, batch 5800, giga_loss[loss=0.2631, simple_loss=0.3371, pruned_loss=0.09454, over 28779.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3281, pruned_loss=0.08975, over 5712271.90 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3453, pruned_loss=0.08978, over 5522120.29 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.327, pruned_loss=0.08993, over 5709023.18 frames. ], batch size: 262, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:36:59,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5234, 2.1558, 1.6150, 0.8075], device='cuda:0'), covar=tensor([0.5575, 0.2639, 0.4208, 0.6649], device='cuda:0'), in_proj_covar=tensor([0.1777, 0.1667, 0.1613, 0.1442], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 07:37:14,983 INFO [train.py:968] (0/2) Epoch 24, batch 5850, giga_loss[loss=0.26, simple_loss=0.3339, pruned_loss=0.09309, over 28679.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3295, pruned_loss=0.09051, over 5715589.76 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3456, pruned_loss=0.09001, over 5525549.49 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3281, pruned_loss=0.09043, over 5711563.09 frames. ], batch size: 92, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:37:17,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2105, 1.7447, 1.1357, 1.2113], device='cuda:0'), covar=tensor([0.0969, 0.0436, 0.1109, 0.1001], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0447, 0.0518, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 07:37:24,485 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 07:37:28,581 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-12 07:37:45,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.486e+02 1.235e+03 1.437e+03 1.861e+03 4.861e+03, threshold=2.873e+03, percent-clipped=2.0
+2023-03-12 07:37:56,219 INFO [train.py:968] (0/2) Epoch 24, batch 5900, giga_loss[loss=0.2549, simple_loss=0.3313, pruned_loss=0.08919, over 28946.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3328, pruned_loss=0.09195, over 5716991.65 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3458, pruned_loss=0.09007, over 5534562.59 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3312, pruned_loss=0.09185, over 5709800.05 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:38:28,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8676, 1.0908, 1.0516, 0.8495], device='cuda:0'), covar=tensor([0.2611, 0.2656, 0.1701, 0.2362], device='cuda:0'), in_proj_covar=tensor([0.1981, 0.1928, 0.1865, 0.1996], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 07:38:34,484 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-12 07:38:39,541 INFO [train.py:968] (0/2) Epoch 24, batch 5950, giga_loss[loss=0.254, simple_loss=0.3378, pruned_loss=0.08514, over 28273.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3365, pruned_loss=0.09329, over 5714770.46 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3459, pruned_loss=0.09016, over 5538321.49 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3349, pruned_loss=0.09315, over 5708208.71 frames. ], batch size: 368, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:39:00,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1054872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:39:08,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8431, 1.9017, 1.4052, 1.5648], device='cuda:0'), covar=tensor([0.0993, 0.0761, 0.1096, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0447, 0.0518, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 07:39:10,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.472e+02 1.125e+03 1.343e+03 1.872e+03 4.125e+03, threshold=2.685e+03, percent-clipped=3.0
+2023-03-12 07:39:11,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3225, 3.1869, 1.5274, 1.4915], device='cuda:0'), covar=tensor([0.1004, 0.0354, 0.0966, 0.1332], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0557, 0.0393, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 07:39:20,970 INFO [train.py:968] (0/2) Epoch 24, batch 6000, giga_loss[loss=0.2736, simple_loss=0.3514, pruned_loss=0.0979, over 29039.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3397, pruned_loss=0.09452, over 5717268.14 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3456, pruned_loss=0.09002, over 5543920.34 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3386, pruned_loss=0.0946, over 5710214.11 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:39:20,975 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 07:39:29,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2821, 1.4998, 1.4545, 1.2590], device='cuda:0'), covar=tensor([0.2816, 0.2249, 0.1633, 0.2309], device='cuda:0'), in_proj_covar=tensor([0.1977, 0.1926, 0.1862, 0.1992], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 07:39:30,027 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2101, simple_loss=0.3173, pruned_loss=0.05146, over 944034.00 frames. 
+2023-03-12 07:39:30,028 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 07:39:49,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1734, 1.4635, 1.4564, 1.3159], device='cuda:0'), covar=tensor([0.1941, 0.1626, 0.2177, 0.1793], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0752, 0.0720, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 07:40:13,550 INFO [train.py:968] (0/2) Epoch 24, batch 6050, giga_loss[loss=0.2479, simple_loss=0.3266, pruned_loss=0.08463, over 28825.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3422, pruned_loss=0.09577, over 5712058.26 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3451, pruned_loss=0.08974, over 5550766.79 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3416, pruned_loss=0.09618, over 5703231.93 frames. ], batch size: 186, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:40:47,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.404e+02 1.306e+03 1.691e+03 2.150e+03 5.827e+03, threshold=3.381e+03, percent-clipped=9.0
+2023-03-12 07:40:59,682 INFO [train.py:968] (0/2) Epoch 24, batch 6100, libri_loss[loss=0.2976, simple_loss=0.3719, pruned_loss=0.1116, over 29538.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3456, pruned_loss=0.09819, over 5710956.15 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3452, pruned_loss=0.08976, over 5562808.48 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.345, pruned_loss=0.09876, over 5697590.14 frames. ], batch size: 83, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:41:45,005 INFO [train.py:968] (0/2) Epoch 24, batch 6150, giga_loss[loss=0.2693, simple_loss=0.3417, pruned_loss=0.09848, over 28888.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3524, pruned_loss=0.1043, over 5692743.60 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3454, pruned_loss=0.09015, over 5561168.20 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3519, pruned_loss=0.1047, over 5687227.12 frames. ], batch size: 106, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:41:55,821 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1055057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:42:20,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.731e+03 2.440e+03 3.327e+03 1.177e+04, threshold=4.881e+03, percent-clipped=23.0
+2023-03-12 07:42:35,112 INFO [train.py:968] (0/2) Epoch 24, batch 6200, giga_loss[loss=0.3307, simple_loss=0.3963, pruned_loss=0.1326, over 28040.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3578, pruned_loss=0.1076, over 5703395.37 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.345, pruned_loss=0.08998, over 5573130.45 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3581, pruned_loss=0.1087, over 5692947.63 frames. ], batch size: 412, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:43:20,512 INFO [train.py:968] (0/2) Epoch 24, batch 6250, giga_loss[loss=0.3255, simple_loss=0.3948, pruned_loss=0.1281, over 28807.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3635, pruned_loss=0.1116, over 5695551.14 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3455, pruned_loss=0.09018, over 5576902.21 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3638, pruned_loss=0.1129, over 5687453.45 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:43:58,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.649e+03 2.056e+03 2.895e+03 6.735e+03, threshold=4.113e+03, percent-clipped=6.0
+2023-03-12 07:44:08,699 INFO [train.py:968] (0/2) Epoch 24, batch 6300, giga_loss[loss=0.3979, simple_loss=0.4355, pruned_loss=0.1801, over 27934.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3702, pruned_loss=0.1173, over 5704396.79 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3459, pruned_loss=0.09044, over 5587444.27 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3709, pruned_loss=0.119, over 5691911.50 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:44:50,862 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1055247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:44:51,941 INFO [train.py:968] (0/2) Epoch 24, batch 6350, giga_loss[loss=0.2978, simple_loss=0.3714, pruned_loss=0.1121, over 29010.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3732, pruned_loss=0.1197, over 5703266.88 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3452, pruned_loss=0.08996, over 5601297.23 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3757, pruned_loss=0.1229, over 5685241.93 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:45:30,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.768e+03 2.277e+03 3.132e+03 8.028e+03, threshold=4.554e+03, percent-clipped=12.0
+2023-03-12 07:45:46,509 INFO [train.py:968] (0/2) Epoch 24, batch 6400, giga_loss[loss=0.2979, simple_loss=0.3626, pruned_loss=0.1166, over 28754.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3762, pruned_loss=0.1228, over 5693225.54 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3448, pruned_loss=0.08976, over 5605873.91 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3791, pruned_loss=0.1261, over 5676572.97 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:46:12,522 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 07:46:15,643 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-12 07:46:39,516 INFO [train.py:968] (0/2) Epoch 24, batch 6450, giga_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1204, over 28520.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3797, pruned_loss=0.1271, over 5675310.28 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3447, pruned_loss=0.08971, over 5610373.33 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3826, pruned_loss=0.1304, over 5658996.91 frames. ], batch size: 60, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:46:52,465 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1055360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:47:21,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.195e+03 1.832e+03 2.339e+03 3.320e+03 5.648e+03, threshold=4.678e+03, percent-clipped=7.0
+2023-03-12 07:47:26,740 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1055390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:47:30,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1055393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:47:37,631 INFO [train.py:968] (0/2) Epoch 24, batch 6500, giga_loss[loss=0.3509, simple_loss=0.4037, pruned_loss=0.1491, over 28807.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3829, pruned_loss=0.1307, over 5667553.48 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3447, pruned_loss=0.08967, over 5611485.50 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3854, pruned_loss=0.1336, over 5654496.21 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:48:04,587 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1055422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:48:16,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1055432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:48:24,460 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 07:48:31,520 INFO [train.py:968] (0/2) Epoch 24, batch 6550, giga_loss[loss=0.4688, simple_loss=0.4751, pruned_loss=0.2312, over 26609.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.387, pruned_loss=0.135, over 5638249.27 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3449, pruned_loss=0.08989, over 5602349.51 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3902, pruned_loss=0.1386, over 5637077.72 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:49:10,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.921e+03 2.913e+03 3.997e+03 8.582e+03, threshold=5.825e+03, percent-clipped=14.0
+2023-03-12 07:49:23,173 INFO [train.py:968] (0/2) Epoch 24, batch 6600, giga_loss[loss=0.2799, simple_loss=0.3604, pruned_loss=0.09974, over 28837.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3889, pruned_loss=0.1367, over 5642018.42 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3456, pruned_loss=0.09027, over 5604532.78 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3918, pruned_loss=0.1402, over 5639953.28 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:50:01,982 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1055541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:50:07,189 INFO [train.py:968] (0/2) Epoch 24, batch 6650, giga_loss[loss=0.3614, simple_loss=0.4069, pruned_loss=0.1579, over 28782.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3863, pruned_loss=0.1351, over 5647675.46 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3452, pruned_loss=0.09005, over 5616695.59 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3912, pruned_loss=0.1405, over 5636622.83 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:50:32,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4732, 1.3040, 3.8979, 3.3617], device='cuda:0'), covar=tensor([0.1564, 0.2711, 0.0479, 0.0941], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0655, 0.0978, 0.0932], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 07:50:34,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1055575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:50:39,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1055578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:50:48,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.778e+03 2.189e+03 2.891e+03 5.422e+03, threshold=4.378e+03, percent-clipped=0.0
+2023-03-12 07:50:57,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5246, 1.5707, 1.7569, 1.4290], device='cuda:0'), covar=tensor([0.1222, 0.1868, 0.1066, 0.1431], device='cuda:0'), in_proj_covar=tensor([0.0907, 0.0704, 0.0955, 0.0854], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 07:51:00,827 INFO [train.py:968] (0/2) Epoch 24, batch 6700, giga_loss[loss=0.3029, simple_loss=0.3723, pruned_loss=0.1167, over 28796.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.387, pruned_loss=0.1367, over 5627234.71 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3454, pruned_loss=0.09014, over 5609979.28 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.391, pruned_loss=0.1412, over 5624335.84 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:51:11,385 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1055607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:51:43,297 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1055636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:51:53,402 INFO [train.py:968] (0/2) Epoch 24, batch 6750, giga_loss[loss=0.3123, simple_loss=0.3834, pruned_loss=0.1206, over 29076.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3867, pruned_loss=0.1346, over 5644694.04 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3455, pruned_loss=0.0901, over 5618006.37 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.391, pruned_loss=0.1396, over 5635616.18 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:51:53,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1055648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:52:24,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2358, 1.4785, 1.5000, 1.2750], device='cuda:0'), covar=tensor([0.1943, 0.1594, 0.2125, 0.1806], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0753, 0.0720, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 07:52:32,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.170e+03 1.738e+03 2.341e+03 3.322e+03 5.566e+03, threshold=4.682e+03, percent-clipped=8.0
+2023-03-12 07:52:42,916 INFO [train.py:968] (0/2) Epoch 24, batch 6800, giga_loss[loss=0.3549, simple_loss=0.3922, pruned_loss=0.1589, over 23454.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3864, pruned_loss=0.1336, over 5645097.92 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3461, pruned_loss=0.09046, over 5623531.02 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3903, pruned_loss=0.1383, over 5633113.52 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:53:21,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1055735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:53:35,807 INFO [train.py:968] (0/2) Epoch 24, batch 6850, giga_loss[loss=0.2877, simple_loss=0.3627, pruned_loss=0.1063, over 28935.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3863, pruned_loss=0.1334, over 5638196.96 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3459, pruned_loss=0.09033, over 5629234.74 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3903, pruned_loss=0.138, over 5624060.40 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:53:36,314 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 07:54:17,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.894e+03 2.122e+03 2.834e+03 7.628e+03, threshold=4.244e+03, percent-clipped=2.0
+2023-03-12 07:54:32,643 INFO [train.py:968] (0/2) Epoch 24, batch 6900, giga_loss[loss=0.3056, simple_loss=0.3665, pruned_loss=0.1223, over 28009.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3849, pruned_loss=0.1321, over 5632643.29 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3459, pruned_loss=0.09034, over 5630846.57 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3882, pruned_loss=0.1359, over 5620349.13 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:54:56,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5558, 2.3231, 1.7043, 0.8067], device='cuda:0'), covar=tensor([0.6169, 0.3423, 0.4217, 0.7090], device='cuda:0'), in_proj_covar=tensor([0.1779, 0.1677, 0.1619, 0.1450], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 07:55:23,671 INFO [train.py:968] (0/2) Epoch 24, batch 6950, giga_loss[loss=0.3119, simple_loss=0.3625, pruned_loss=0.1306, over 24051.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3837, pruned_loss=0.1299, over 5640094.25 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3459, pruned_loss=0.09031, over 5629395.49 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3866, pruned_loss=0.1332, over 5632010.80 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:55:25,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5282, 1.0866, 4.6446, 3.5650], device='cuda:0'), covar=tensor([0.1654, 0.3007, 0.0423, 0.0989], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0653, 0.0975, 0.0930], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 07:55:53,016 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1055878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:55:56,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1055881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:56:01,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.605e+03 2.061e+03 2.615e+03 5.393e+03, threshold=4.122e+03, percent-clipped=3.0
+2023-03-12 07:56:16,056 INFO [train.py:968] (0/2) Epoch 24, batch 7000, giga_loss[loss=0.2571, simple_loss=0.3389, pruned_loss=0.08769, over 28558.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3785, pruned_loss=0.1256, over 5647644.80 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3458, pruned_loss=0.09023, over 5629005.60 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3814, pruned_loss=0.1288, over 5641901.16 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:56:27,685 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1055910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:56:33,739 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1055916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:56:54,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2902, 1.3227, 3.9636, 3.4265], device='cuda:0'), covar=tensor([0.1668, 0.2737, 0.0435, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0653, 0.0973, 0.0930], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 07:57:01,922 INFO [train.py:968] (0/2) Epoch 24, batch 7050, giga_loss[loss=0.2766, simple_loss=0.3596, pruned_loss=0.09673, over 28914.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3747, pruned_loss=0.1223, over 5652709.89 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3454, pruned_loss=0.09008, over 5637900.40 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3783, pruned_loss=0.1261, over 5640522.30 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:57:19,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6122, 1.7581, 1.8363, 1.5396], device='cuda:0'), covar=tensor([0.3162, 0.2566, 0.2146, 0.2809], device='cuda:0'), in_proj_covar=tensor([0.1993, 0.1941, 0.1875, 0.2009], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 07:57:39,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.599e+02 1.543e+03 2.030e+03 2.910e+03 1.039e+04, threshold=4.060e+03, percent-clipped=13.0
+2023-03-12 07:57:50,452 INFO [train.py:968] (0/2) Epoch 24, batch 7100, giga_loss[loss=0.2916, simple_loss=0.3603, pruned_loss=0.1114, over 28606.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3734, pruned_loss=0.1218, over 5655846.97 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.345, pruned_loss=0.08993, over 5644746.82 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3773, pruned_loss=0.1257, over 5640389.74 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:57:52,162 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1056000.pt
+2023-03-12 07:58:01,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2814, 3.1030, 1.4563, 1.4819], device='cuda:0'), covar=tensor([0.1030, 0.0380, 0.0897, 0.1354], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0561, 0.0394, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 07:58:04,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1056011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:58:16,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1056023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:58:44,822 INFO [train.py:968] (0/2) Epoch 24, batch 7150, giga_loss[loss=0.3104, simple_loss=0.3762, pruned_loss=0.1223, over 28626.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3728, pruned_loss=0.1215, over 5646501.65 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3453, pruned_loss=0.09011, over 5645108.45 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3761, pruned_loss=0.125, over 5634198.72 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:58:58,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1891, 4.0188, 3.8063, 2.1040], device='cuda:0'), covar=tensor([0.0629, 0.0765, 0.0752, 0.1999], device='cuda:0'), in_proj_covar=tensor([0.1270, 0.1174, 0.0991, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 07:58:58,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3867, 1.9646, 1.4232, 0.6978], device='cuda:0'), covar=tensor([0.7030, 0.3068, 0.3996, 0.7138], device='cuda:0'), in_proj_covar=tensor([0.1781, 0.1675, 0.1617, 0.1450], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 07:58:58,471 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1056059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:59:01,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1056062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:59:27,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.079e+02 1.588e+03 1.943e+03 2.811e+03 6.420e+03, threshold=3.886e+03, percent-clipped=6.0
+2023-03-12 07:59:31,157 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1056090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:59:32,014 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1056091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:59:40,714 INFO [train.py:968] (0/2) Epoch 24, batch 7200, giga_loss[loss=0.2654, simple_loss=0.3484, pruned_loss=0.09123, over 28996.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3702, pruned_loss=0.1188, over 5655655.61 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3453, pruned_loss=0.09006, over 5650025.94 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3733, pruned_loss=0.1222, over 5641645.38 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:00:27,689 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3386, 3.1272, 3.0088, 1.4829], device='cuda:0'), covar=tensor([0.0931, 0.1164, 0.1039, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.1268, 0.1175, 0.0992, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 08:00:41,109 INFO [train.py:968] (0/2) Epoch 24, batch 7250, libri_loss[loss=0.2635, simple_loss=0.3445, pruned_loss=0.09124, over 29749.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3685, pruned_loss=0.1156, over 5666102.58 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3447, pruned_loss=0.08977, over 5655328.71 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.372, pruned_loss=0.1191, over 5650126.78 frames. ], batch size: 87, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:00:45,787 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.46 vs. limit=5.0
+2023-03-12 08:00:47,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1056154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:00:51,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1056157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:01:00,775 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1056166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:01:06,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1056169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:01:21,245 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1056186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:01:21,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.140e+03 1.504e+03 1.842e+03 2.450e+03 5.816e+03, threshold=3.685e+03, percent-clipped=4.0
+2023-03-12 08:01:32,264 INFO [train.py:968] (0/2) Epoch 24, batch 7300, giga_loss[loss=0.3576, simple_loss=0.4125, pruned_loss=0.1513, over 27581.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3714, pruned_loss=0.1163, over 5674188.59 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3441, pruned_loss=0.08954, over 5660329.87 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3753, pruned_loss=0.1198, over 5657153.28 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:01:32,477 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1056198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:02:24,415 INFO [train.py:968] (0/2) Epoch 24, batch 7350, giga_loss[loss=0.3014, simple_loss=0.3716, pruned_loss=0.1156, over 28848.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3716, pruned_loss=0.1166, over 5673629.99 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3439, pruned_loss=0.08944, over 5668504.35 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3758, pruned_loss=0.1204, over 5652971.73 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:03:01,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.314e+02 1.727e+03 2.044e+03 2.901e+03 8.964e+03, threshold=4.088e+03, percent-clipped=14.0
+2023-03-12 08:03:12,839 INFO [train.py:968] (0/2) Epoch 24, batch 7400, giga_loss[loss=0.2836, simple_loss=0.3541, pruned_loss=0.1066, over 28887.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3706, pruned_loss=0.1165, over 5669514.89 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.344, pruned_loss=0.08957, over 5670775.67 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3746, pruned_loss=0.1201, over 5651124.78 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:03:33,375 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-12 08:04:01,573 INFO [train.py:968] (0/2) Epoch 24, batch 7450, giga_loss[loss=0.2987, simple_loss=0.3614, pruned_loss=0.118, over 28679.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3685, pruned_loss=0.1153, over 5667536.46 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3443, pruned_loss=0.0897, over 5665353.38 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3726, pruned_loss=0.1194, over 5657927.52 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:04:29,929 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2807, 1.4395, 1.5212, 1.2766], device='cuda:0'), covar=tensor([0.2884, 0.2627, 0.1825, 0.2417], device='cuda:0'), in_proj_covar=tensor([0.2000, 0.1954, 0.1883, 0.2019], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 08:04:36,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.636e+03 2.048e+03 2.714e+03 9.280e+03, threshold=4.096e+03, percent-clipped=8.0
+2023-03-12 08:04:45,474 INFO [train.py:968] (0/2) Epoch 24, batch 7500, giga_loss[loss=0.2944, simple_loss=0.3572, pruned_loss=0.1158, over 28699.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3681, pruned_loss=0.1163, over 5674365.00 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3438, pruned_loss=0.08944, over 5673137.18 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3729, pruned_loss=0.1209, over 5659962.97 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:05:00,659 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1056415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:05:26,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5575, 4.3891, 4.1595, 1.9356], device='cuda:0'), covar=tensor([0.0558, 0.0682, 0.0717, 0.1990], device='cuda:0'), in_proj_covar=tensor([0.1267, 0.1171, 0.0990, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 08:05:34,503 INFO [train.py:968] (0/2) Epoch 24, batch 7550, giga_loss[loss=0.2606, simple_loss=0.3455, pruned_loss=0.08789, over 28512.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.368, pruned_loss=0.1168, over 5673386.05 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3439, pruned_loss=0.08957, over 5673601.08 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3722, pruned_loss=0.1208, over 5661540.68 frames. ], batch size: 60, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:05:45,072 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 08:05:51,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1056465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:05:55,743 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 08:06:12,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.761e+03 2.149e+03 2.822e+03 7.470e+03, threshold=4.299e+03, percent-clipped=14.0
+2023-03-12 08:06:22,400 INFO [train.py:968] (0/2) Epoch 24, batch 7600, giga_loss[loss=0.2629, simple_loss=0.3447, pruned_loss=0.0906, over 28507.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3672, pruned_loss=0.1152, over 5671572.43 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3435, pruned_loss=0.08941, over 5680870.06 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3715, pruned_loss=0.1193, over 5655412.08 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:07:12,133 INFO [train.py:968] (0/2) Epoch 24, batch 7650, giga_loss[loss=0.2549, simple_loss=0.3335, pruned_loss=0.08812, over 28585.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3691, pruned_loss=0.116, over 5675887.97 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3441, pruned_loss=0.08973, over 5684766.30 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3724, pruned_loss=0.1195, over 5659230.59 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:07:47,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.563e+03 1.934e+03 2.680e+03 5.656e+03, threshold=3.868e+03, percent-clipped=1.0
+2023-03-12 08:07:54,217 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-12 08:07:56,326 INFO [train.py:968] (0/2) Epoch 24, batch 7700, giga_loss[loss=0.2521, simple_loss=0.3344, pruned_loss=0.08488, over 29048.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3674, pruned_loss=0.1142, over 5681944.17 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3442, pruned_loss=0.08977, over 5680502.15 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3705, pruned_loss=0.1174, over 5671688.73 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:08:05,535 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1056608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:08:09,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1056611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:08:39,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1056640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:08:46,091 INFO [train.py:968] (0/2) Epoch 24, batch 7750, libri_loss[loss=0.2249, simple_loss=0.3054, pruned_loss=0.07223, over 29498.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3656, pruned_loss=0.1132, over 5687788.17 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3439, pruned_loss=0.08947, over 5689103.27 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3692, pruned_loss=0.1171, over 5671677.45 frames. ], batch size: 70, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:09:26,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.664e+03 2.065e+03 2.823e+03 7.344e+03, threshold=4.130e+03, percent-clipped=7.0
+2023-03-12 08:09:38,416 INFO [train.py:968] (0/2) Epoch 24, batch 7800, giga_loss[loss=0.29, simple_loss=0.354, pruned_loss=0.113, over 28961.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3644, pruned_loss=0.1135, over 5671451.97 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3441, pruned_loss=0.08953, over 5692710.90 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3675, pruned_loss=0.1169, over 5655657.91 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:09:52,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3451, 1.7447, 1.4655, 1.5015], device='cuda:0'), covar=tensor([0.0788, 0.0308, 0.0311, 0.0840], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 08:10:23,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6217, 1.6900, 1.7468, 1.5195], device='cuda:0'), covar=tensor([0.1982, 0.2283, 0.2276, 0.2305], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0752, 0.0718, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 08:10:28,466 INFO [train.py:968] (0/2) Epoch 24, batch 7850, libri_loss[loss=0.3099, simple_loss=0.3827, pruned_loss=0.1185, over 29477.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3635, pruned_loss=0.1137, over 5673126.22 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.344, pruned_loss=0.08959, over 5699314.94 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.367, pruned_loss=0.1175, over 5653314.13 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:10:46,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5126, 1.7567, 1.4661, 1.4732], device='cuda:0'), covar=tensor([0.3057, 0.3143, 0.3515, 0.2696], device='cuda:0'), in_proj_covar=tensor([0.1543, 0.1114, 0.1362, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 08:11:07,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.842e+03 2.465e+03 3.363e+03 9.497e+03, threshold=4.930e+03, percent-clipped=16.0
+2023-03-12 08:11:08,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1056790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:11:16,255 INFO [train.py:968] (0/2) Epoch 24, batch 7900, giga_loss[loss=0.2723, simple_loss=0.3401, pruned_loss=0.1023, over 28933.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3637, pruned_loss=0.1146, over 5663993.85 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3443, pruned_loss=0.0897, over 5693377.96 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3666, pruned_loss=0.1181, over 5653371.59 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:12:05,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-12 08:12:06,863 INFO [train.py:968] (0/2) Epoch 24, batch 7950, giga_loss[loss=0.3057, simple_loss=0.3673, pruned_loss=0.1221, over 28494.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3626, pruned_loss=0.1146, over 5661652.67 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.08964, over 5696263.71 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3652, pruned_loss=0.1177, over 5650206.99 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:12:10,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9715, 2.1960, 1.8572, 1.9647], device='cuda:0'), covar=tensor([0.3013, 0.3024, 0.3357, 0.2813], device='cuda:0'), in_proj_covar=tensor([0.1542, 0.1112, 0.1360, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 08:12:18,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5217, 2.2451, 1.6706, 0.7408], device='cuda:0'), covar=tensor([0.6391, 0.3477, 0.4494, 0.6861], device='cuda:0'), in_proj_covar=tensor([0.1784, 0.1682, 0.1620, 0.1454], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 08:12:39,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5124, 1.9974, 1.6214, 1.6688], device='cuda:0'), covar=tensor([0.0770, 0.0287, 0.0325, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 08:12:40,015 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1056881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:12:46,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.696e+03 2.280e+03 3.023e+03 8.319e+03, threshold=4.560e+03, percent-clipped=11.0
+2023-03-12 08:12:53,881 INFO [train.py:968] (0/2) Epoch 24, batch 8000, giga_loss[loss=0.3162, simple_loss=0.3775, pruned_loss=0.1275, over 27955.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5662950.71 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3439, pruned_loss=0.08942, over 5700566.36 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3647, pruned_loss=0.1173, over 5648796.68 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:13:12,816 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1056915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:13:30,348 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1056933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:13:33,421 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1056936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:13:44,736 INFO [train.py:968] (0/2) Epoch 24, batch 8050, giga_loss[loss=0.3048, simple_loss=0.378, pruned_loss=0.1158, over 28921.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3637, pruned_loss=0.1149, over 5666265.64 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.08967, over 5702365.63 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3659, pruned_loss=0.1176, over 5653102.55 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:14:00,210 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1056964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:14:00,961 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1056965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:14:25,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 2.011e+03 2.805e+03 3.957e+03 1.061e+04, threshold=5.611e+03, percent-clipped=14.0
+2023-03-12 08:14:29,467 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.92 vs. limit=5.0
+2023-03-12 08:14:34,227 INFO [train.py:968] (0/2) Epoch 24, batch 8100, giga_loss[loss=0.2776, simple_loss=0.3591, pruned_loss=0.09805, over 28981.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3637, pruned_loss=0.1139, over 5675014.23 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3438, pruned_loss=0.08939, over 5706737.49 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3662, pruned_loss=0.1168, over 5659984.95 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:15:23,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3574, 1.4499, 1.2527, 1.6158], device='cuda:0'), covar=tensor([0.0787, 0.0351, 0.0357, 0.0870], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 08:15:23,698 INFO [train.py:968] (0/2) Epoch 24, batch 8150, giga_loss[loss=0.2689, simple_loss=0.3467, pruned_loss=0.0956, over 29016.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3636, pruned_loss=0.1129, over 5677264.38 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.08971, over 5700107.02 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3657, pruned_loss=0.1155, over 5670865.33 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:15:54,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9353, 3.7660, 3.5810, 1.7463], device='cuda:0'), covar=tensor([0.0761, 0.0882, 0.0883, 0.2121], device='cuda:0'), in_proj_covar=tensor([0.1274, 0.1178, 0.0994, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 08:16:05,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.580e+03 1.987e+03 3.109e+03 5.695e+03, threshold=3.974e+03, percent-clipped=1.0
+2023-03-12 08:16:13,524 INFO [train.py:968] (0/2) Epoch 24, batch 8200, giga_loss[loss=0.3082, simple_loss=0.3712, pruned_loss=0.1226, over 28547.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3641, pruned_loss=0.1136, over 5682112.66 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3437, pruned_loss=0.08964, over 5705218.21 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3666, pruned_loss=0.1162, over 5671884.76 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:17:05,641 INFO [train.py:968] (0/2) Epoch 24, batch 8250, giga_loss[loss=0.4019, simple_loss=0.4231, pruned_loss=0.1903, over 23621.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3679, pruned_loss=0.1176, over 5666447.64 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3437, pruned_loss=0.0896, over 5704860.11 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3703, pruned_loss=0.1202, over 5658099.70 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:17:47,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 2.022e+03 2.424e+03 3.415e+03 8.055e+03, threshold=4.849e+03, percent-clipped=17.0
+2023-03-12 08:17:49,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1057192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:17:55,127 INFO [train.py:968] (0/2) Epoch 24, batch 8300, giga_loss[loss=0.4104, simple_loss=0.4292, pruned_loss=0.1958, over 27525.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.369, pruned_loss=0.12, over 5657543.83 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3436, pruned_loss=0.08967, over 5701210.33 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3719, pruned_loss=0.1231, over 5652111.77 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:18:45,944 INFO [train.py:968] (0/2) Epoch 24, batch 8350, libri_loss[loss=0.2804, simple_loss=0.3555, pruned_loss=0.1027, over 29526.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3692, pruned_loss=0.1205, over 5666468.26 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.344, pruned_loss=0.08988, over 5702997.67 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3722, pruned_loss=0.124, over 5658420.54 frames. ], batch size: 81, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:18:53,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:19:27,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1057288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:19:27,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.192e+03 1.944e+03 2.604e+03 3.700e+03 1.021e+04, threshold=5.209e+03, percent-clipped=7.0
+2023-03-12 08:19:29,579 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:19:35,633 INFO [train.py:968] (0/2) Epoch 24, batch 8400, libri_loss[loss=0.2978, simple_loss=0.3785, pruned_loss=0.1086, over 29363.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3712, pruned_loss=0.1226, over 5659993.47 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3443, pruned_loss=0.09005, over 5704048.15 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.374, pruned_loss=0.126, over 5651788.43 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:19:35,853 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1057298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:19:51,792 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7448, 3.5826, 3.4234, 1.6429], device='cuda:0'), covar=tensor([0.0815, 0.0914, 0.0859, 0.2249], device='cuda:0'), in_proj_covar=tensor([0.1273, 0.1175, 0.0993, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 08:20:14,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:20:21,731 INFO [train.py:968] (0/2) Epoch 24, batch 8450, giga_loss[loss=0.3459, simple_loss=0.4025, pruned_loss=0.1447, over 28617.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3705, pruned_loss=0.1219, over 5664589.18 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3444, pruned_loss=0.09012, over 5701260.08 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3733, pruned_loss=0.1254, over 5659443.28 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:20:58,858 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.637e+02 1.615e+03 1.932e+03 2.482e+03 6.091e+03, threshold=3.865e+03, percent-clipped=3.0
+2023-03-12 08:21:04,872 INFO [train.py:968] (0/2) Epoch 24, batch 8500, giga_loss[loss=0.2646, simple_loss=0.3544, pruned_loss=0.08737, over 28979.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3678, pruned_loss=0.1187, over 5680975.03 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3439, pruned_loss=0.08988, over 5707952.39 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3714, pruned_loss=0.1228, over 5669811.60 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:21:07,181 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:21:11,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:21:39,914 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:21:41,576 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:21:43,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:21:55,517 INFO [train.py:968] (0/2) Epoch 24, batch 8550, giga_loss[loss=0.3211, simple_loss=0.3853, pruned_loss=0.1284, over 28853.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3659, pruned_loss=0.1158, over 5676260.95 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3439, pruned_loss=0.08994, over 5701148.46 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3689, pruned_loss=0.1193, over 5674034.71 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:22:09,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3555, 3.1708, 3.0232, 1.5376], device='cuda:0'), covar=tensor([0.0975, 0.1074, 0.0929, 0.2151], device='cuda:0'), in_proj_covar=tensor([0.1271, 0.1173, 0.0992, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 08:22:12,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:22:25,683 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:22:28,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:22:31,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.777e+02 1.716e+03 2.121e+03 2.777e+03 8.039e+03, threshold=4.241e+03, percent-clipped=9.0
+2023-03-12 08:22:40,157 INFO [train.py:968] (0/2) Epoch 24, batch 8600, giga_loss[loss=0.2971, simple_loss=0.358, pruned_loss=0.1181, over 28900.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3643, pruned_loss=0.1148, over 5679320.87 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3442, pruned_loss=0.08993, over 5706327.59 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.367, pruned_loss=0.1181, over 5672280.33 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:22:57,851 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:23:26,699 INFO [train.py:968] (0/2) Epoch 24, batch 8650, giga_loss[loss=0.2812, simple_loss=0.3433, pruned_loss=0.1095, over 28879.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5668385.54 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3439, pruned_loss=0.08981, over 5698000.70 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3648, pruned_loss=0.1175, over 5668612.89 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:23:29,784 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-12 08:23:40,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7233, 1.9057, 1.3392, 1.4585], device='cuda:0'), covar=tensor([0.0951, 0.0594, 0.1063, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0449, 0.0520, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 08:23:44,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:24:08,556 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.706e+03 2.173e+03 3.230e+03 8.663e+03, threshold=4.346e+03, percent-clipped=10.0
+2023-03-12 08:24:15,753 INFO [train.py:968] (0/2) Epoch 24, batch 8700, giga_loss[loss=0.2826, simple_loss=0.35, pruned_loss=0.1077, over 28340.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3623, pruned_loss=0.1144, over 5660024.90 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3445, pruned_loss=0.09018, over 5694636.42 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3649, pruned_loss=0.1179, over 5662171.91 frames. ], batch size: 65, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:25:08,787 INFO [train.py:968] (0/2) Epoch 24, batch 8750, giga_loss[loss=0.3225, simple_loss=0.3912, pruned_loss=0.1269, over 28737.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3658, pruned_loss=0.1169, over 5659631.29 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3445, pruned_loss=0.09023, over 5687649.45 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3682, pruned_loss=0.12, over 5666224.75 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:25:22,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:25:33,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:25:48,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.204e+03 1.615e+03 2.059e+03 2.563e+03 8.272e+03, threshold=4.118e+03, percent-clipped=6.0
+2023-03-12 08:25:57,771 INFO [train.py:968] (0/2) Epoch 24, batch 8800, libri_loss[loss=0.2634, simple_loss=0.3454, pruned_loss=0.09065, over 27837.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3694, pruned_loss=0.1171, over 5658128.38 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3451, pruned_loss=0.09058, over 5683832.18 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3716, pruned_loss=0.1201, over 5665327.48 frames. ], batch size: 116, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:26:09,482 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:26:12,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:26:40,903 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:26:44,908 INFO [train.py:968] (0/2) Epoch 24, batch 8850, giga_loss[loss=0.2901, simple_loss=0.3644, pruned_loss=0.1079, over 28999.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3722, pruned_loss=0.1173, over 5660718.17 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3454, pruned_loss=0.09078, over 5682856.00 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.374, pruned_loss=0.1199, over 5667179.68 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:27:16,064 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-12 08:27:18,379 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-12 08:27:21,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.581e+03 2.083e+03 2.901e+03 6.423e+03, threshold=4.167e+03, percent-clipped=8.0
+2023-03-12 08:27:29,689 INFO [train.py:968] (0/2) Epoch 24, batch 8900, giga_loss[loss=0.2966, simple_loss=0.3723, pruned_loss=0.1105, over 28938.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3725, pruned_loss=0.1176, over 5672760.05 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3447, pruned_loss=0.09037, over 5691787.97 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3755, pruned_loss=0.1211, over 5669493.38 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:27:36,492 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1057806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:36,555 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:36,688 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 08:27:38,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:40,843 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1057812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:44,130 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:44,927 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2739, 0.7719, 0.8491, 1.3789], device='cuda:0'), covar=tensor([0.0763, 0.0391, 0.0369, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0119, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 08:27:47,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:28:04,322 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:28:04,411 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4271, 1.7376, 1.3818, 1.2846], device='cuda:0'), covar=tensor([0.2552, 0.2617, 0.3063, 0.2434], device='cuda:0'), in_proj_covar=tensor([0.1539, 0.1108, 0.1359, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 08:28:17,248 INFO [train.py:968] (0/2) Epoch 24, batch 8950, giga_loss[loss=0.3215, simple_loss=0.3866, pruned_loss=0.1282, over 28766.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3739, pruned_loss=0.1195, over 5682899.43 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3449, pruned_loss=0.09049, over 5695962.86 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3769, pruned_loss=0.1228, over 5676260.36 frames. ], batch size: 243, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:28:17,491 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:28:27,660 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-12 08:28:56,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.796e+03 2.211e+03 2.838e+03 7.629e+03, threshold=4.422e+03, percent-clipped=7.0
+2023-03-12 08:29:03,340 INFO [train.py:968] (0/2) Epoch 24, batch 9000, giga_loss[loss=0.3234, simple_loss=0.3776, pruned_loss=0.1346, over 28854.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3735, pruned_loss=0.1201, over 5680255.97 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3451, pruned_loss=0.09092, over 5691773.69 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3763, pruned_loss=0.123, over 5678388.00 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:29:03,343 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 08:29:12,610 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2047, simple_loss=0.3121, pruned_loss=0.04866, over 944034.00 frames. 
+2023-03-12 08:29:12,611 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 08:30:00,333 INFO [train.py:968] (0/2) Epoch 24, batch 9050, giga_loss[loss=0.2814, simple_loss=0.3511, pruned_loss=0.1058, over 28634.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3709, pruned_loss=0.119, over 5689165.59 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3445, pruned_loss=0.09059, over 5699469.24 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3748, pruned_loss=0.1228, over 5680235.08 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:30:02,139 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-12 08:30:41,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.628e+03 2.039e+03 2.730e+03 7.487e+03, threshold=4.078e+03, percent-clipped=7.0
+2023-03-12 08:30:47,955 INFO [train.py:968] (0/2) Epoch 24, batch 9100, giga_loss[loss=0.2773, simple_loss=0.3457, pruned_loss=0.1045, over 28900.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.368, pruned_loss=0.1174, over 5685439.25 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3444, pruned_loss=0.09049, over 5700394.99 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3716, pruned_loss=0.121, over 5677272.88 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:30:50,283 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1058000.pt
+2023-03-12 08:31:40,499 INFO [train.py:968] (0/2) Epoch 24, batch 9150, giga_loss[loss=0.3006, simple_loss=0.3654, pruned_loss=0.1178, over 29053.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3682, pruned_loss=0.119, over 5680599.35 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3445, pruned_loss=0.09052, over 5701730.19 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3711, pruned_loss=0.1221, over 5672839.83 frames. ], batch size: 128, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:32:25,185 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.818e+03 2.618e+03 3.619e+03 7.295e+03, threshold=5.236e+03, percent-clipped=12.0
+2023-03-12 08:32:32,030 INFO [train.py:968] (0/2) Epoch 24, batch 9200, giga_loss[loss=0.339, simple_loss=0.3977, pruned_loss=0.1401, over 28884.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3681, pruned_loss=0.119, over 5677874.60 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3447, pruned_loss=0.09064, over 5695784.26 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3707, pruned_loss=0.1218, over 5676292.18 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:33:23,437 INFO [train.py:968] (0/2) Epoch 24, batch 9250, giga_loss[loss=0.2717, simple_loss=0.3443, pruned_loss=0.09951, over 28930.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3689, pruned_loss=0.1204, over 5665843.98 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3445, pruned_loss=0.09062, over 5689721.17 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3714, pruned_loss=0.1231, over 5669471.32 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:33:56,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1058181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:34:03,405 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1058187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:34:08,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.224e+03 1.707e+03 2.213e+03 3.005e+03 6.012e+03, threshold=4.426e+03, percent-clipped=4.0
+2023-03-12 08:34:13,322 INFO [train.py:968] (0/2) Epoch 24, batch 9300, giga_loss[loss=0.2825, simple_loss=0.3462, pruned_loss=0.1094, over 28102.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3671, pruned_loss=0.1193, over 5672184.87 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3446, pruned_loss=0.09061, over 5692913.42 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3694, pruned_loss=0.1218, over 5671853.34 frames. ], batch size: 77, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:34:44,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6039, 1.8739, 1.5801, 1.4200], device='cuda:0'), covar=tensor([0.2152, 0.2107, 0.2225, 0.2138], device='cuda:0'), in_proj_covar=tensor([0.1543, 0.1114, 0.1363, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 08:35:04,904 INFO [train.py:968] (0/2) Epoch 24, batch 9350, libri_loss[loss=0.248, simple_loss=0.3398, pruned_loss=0.07811, over 29538.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3657, pruned_loss=0.1181, over 5680016.37 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3445, pruned_loss=0.09044, over 5698436.57 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3682, pruned_loss=0.121, over 5673968.26 frames. ], batch size: 82, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:35:05,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3374, 1.7903, 1.3361, 0.7400], device='cuda:0'), covar=tensor([0.4897, 0.2600, 0.3241, 0.5858], device='cuda:0'), in_proj_covar=tensor([0.1774, 0.1675, 0.1609, 0.1444], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 08:35:44,138 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-12 08:35:47,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.712e+03 2.061e+03 3.084e+03 1.107e+04, threshold=4.122e+03, percent-clipped=6.0
+2023-03-12 08:35:53,244 INFO [train.py:968] (0/2) Epoch 24, batch 9400, giga_loss[loss=0.3083, simple_loss=0.3714, pruned_loss=0.1227, over 28210.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3684, pruned_loss=0.1196, over 5674901.12 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.345, pruned_loss=0.0907, over 5702605.77 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3704, pruned_loss=0.1222, over 5666143.15 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:36:18,335 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1058324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:36:20,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1058327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:36:23,779 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1058330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:36:25,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1058333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:36:40,114 INFO [train.py:968] (0/2) Epoch 24, batch 9450, giga_loss[loss=0.278, simple_loss=0.3415, pruned_loss=0.1073, over 28909.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3696, pruned_loss=0.12, over 5681105.71 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.345, pruned_loss=0.09076, over 5704900.32 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3716, pruned_loss=0.1226, over 5671708.58 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:36:52,037 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1058356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:36:58,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0328, 1.3270, 1.1150, 0.3820], device='cuda:0'), covar=tensor([0.3371, 0.2902, 0.4013, 0.4900], device='cuda:0'), in_proj_covar=tensor([0.1774, 0.1676, 0.1607, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 08:36:58,714 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1058362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:37:29,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 1.747e+03 2.213e+03 3.317e+03 1.145e+04, threshold=4.427e+03, percent-clipped=17.0
+2023-03-12 08:37:33,240 INFO [train.py:968] (0/2) Epoch 24, batch 9500, giga_loss[loss=0.3066, simple_loss=0.3712, pruned_loss=0.121, over 28748.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3688, pruned_loss=0.1199, over 5675818.55 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.345, pruned_loss=0.09073, over 5705907.82 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3705, pruned_loss=0.122, over 5667422.60 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:38:23,273 INFO [train.py:968] (0/2) Epoch 24, batch 9550, libri_loss[loss=0.2571, simple_loss=0.337, pruned_loss=0.08861, over 29572.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3696, pruned_loss=0.1176, over 5684367.53 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3447, pruned_loss=0.09059, over 5709739.58 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3716, pruned_loss=0.1199, over 5673758.24 frames. ], batch size: 74, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:38:42,712 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1058472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:38:54,915 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-12 08:39:00,190 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.924e+02 1.498e+03 1.881e+03 2.720e+03 1.277e+04, threshold=3.761e+03, percent-clipped=1.0
+2023-03-12 08:39:04,897 INFO [train.py:968] (0/2) Epoch 24, batch 9600, giga_loss[loss=0.289, simple_loss=0.3738, pruned_loss=0.1021, over 28947.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3705, pruned_loss=0.1161, over 5691116.58 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3444, pruned_loss=0.09038, over 5717592.83 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3734, pruned_loss=0.1192, over 5674584.15 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:39:52,238 INFO [train.py:968] (0/2) Epoch 24, batch 9650, giga_loss[loss=0.2783, simple_loss=0.3583, pruned_loss=0.09918, over 28908.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3723, pruned_loss=0.117, over 5680457.56 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3445, pruned_loss=0.09055, over 5715023.61 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3755, pruned_loss=0.1203, over 5667712.36 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:40:32,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2987, 4.1185, 3.9030, 1.9253], device='cuda:0'), covar=tensor([0.0647, 0.0801, 0.0849, 0.1960], device='cuda:0'), in_proj_covar=tensor([0.1275, 0.1179, 0.0995, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 08:40:37,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.643e+02 1.816e+03 2.322e+03 3.258e+03 8.140e+03, threshold=4.644e+03, percent-clipped=20.0
+2023-03-12 08:40:43,875 INFO [train.py:968] (0/2) Epoch 24, batch 9700, giga_loss[loss=0.2928, simple_loss=0.3616, pruned_loss=0.112, over 28913.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3759, pruned_loss=0.1205, over 5680717.35 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3445, pruned_loss=0.09047, over 5717614.97 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3789, pruned_loss=0.1236, over 5667861.58 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:41:32,708 INFO [train.py:968] (0/2) Epoch 24, batch 9750, libri_loss[loss=0.2614, simple_loss=0.3483, pruned_loss=0.08723, over 25635.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3772, pruned_loss=0.1228, over 5675441.41 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3444, pruned_loss=0.09033, over 5716849.93 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3802, pruned_loss=0.1259, over 5665694.50 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:42:16,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.801e+03 2.317e+03 3.397e+03 1.037e+04, threshold=4.634e+03, percent-clipped=11.0
+2023-03-12 08:42:20,486 INFO [train.py:968] (0/2) Epoch 24, batch 9800, giga_loss[loss=0.2825, simple_loss=0.3575, pruned_loss=0.1038, over 28957.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3778, pruned_loss=0.1241, over 5671752.76 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3446, pruned_loss=0.0904, over 5722699.48 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.381, pruned_loss=0.1276, over 5657631.40 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:42:23,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4400, 1.4258, 3.8295, 3.2790], device='cuda:0'), covar=tensor([0.1476, 0.2614, 0.0470, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0658, 0.0974, 0.0932], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 08:43:05,225 INFO [train.py:968] (0/2) Epoch 24, batch 9850, giga_loss[loss=0.3265, simple_loss=0.3889, pruned_loss=0.1321, over 28247.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3741, pruned_loss=0.1211, over 5669054.06 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3439, pruned_loss=0.08997, over 5727700.16 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3781, pruned_loss=0.1249, over 5652249.08 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:43:24,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5893, 1.6745, 1.8308, 1.3641], device='cuda:0'), covar=tensor([0.1996, 0.2721, 0.1649, 0.1911], device='cuda:0'), in_proj_covar=tensor([0.0909, 0.0709, 0.0957, 0.0854], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 08:43:30,613 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1058772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:43:49,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.684e+03 2.161e+03 2.795e+03 8.226e+03, threshold=4.322e+03, percent-clipped=4.0
+2023-03-12 08:43:52,931 INFO [train.py:968] (0/2) Epoch 24, batch 9900, giga_loss[loss=0.2613, simple_loss=0.3459, pruned_loss=0.08838, over 28826.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3726, pruned_loss=0.1183, over 5668466.17 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3438, pruned_loss=0.08991, over 5726725.01 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3764, pruned_loss=0.1219, over 5654620.44 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:44:05,490 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1058811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:44:20,565 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1058824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:44:38,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1058847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:44:39,308 INFO [train.py:968] (0/2) Epoch 24, batch 9950, giga_loss[loss=0.3216, simple_loss=0.3908, pruned_loss=0.1262, over 28856.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3737, pruned_loss=0.1178, over 5666811.31 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3438, pruned_loss=0.08999, over 5718899.60 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.377, pruned_loss=0.121, over 5662355.74 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:45:26,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.750e+03 2.181e+03 3.223e+03 1.192e+04, threshold=4.363e+03, percent-clipped=13.0
+2023-03-12 08:45:30,085 INFO [train.py:968] (0/2) Epoch 24, batch 10000, giga_loss[loss=0.2934, simple_loss=0.3724, pruned_loss=0.1072, over 28648.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3746, pruned_loss=0.1189, over 5657903.01 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3438, pruned_loss=0.08993, over 5713191.44 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3781, pruned_loss=0.1223, over 5658348.57 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:46:23,673 INFO [train.py:968] (0/2) Epoch 24, batch 10050, giga_loss[loss=0.2642, simple_loss=0.3456, pruned_loss=0.09144, over 28891.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3742, pruned_loss=0.119, over 5661610.62 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3437, pruned_loss=0.08985, over 5716093.52 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3777, pruned_loss=0.1223, over 5658455.64 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:46:38,579 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1058963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:46:52,062 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-12 08:47:06,444 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-12 08:47:07,956 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1058990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:47:10,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1058993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:47:10,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-12 08:47:10,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.801e+02 1.647e+03 1.919e+03 2.655e+03 5.169e+03, threshold=3.838e+03, percent-clipped=2.0
+2023-03-12 08:47:15,619 INFO [train.py:968] (0/2) Epoch 24, batch 10100, giga_loss[loss=0.4405, simple_loss=0.4423, pruned_loss=0.2194, over 26721.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3723, pruned_loss=0.1189, over 5661770.28 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3436, pruned_loss=0.08978, over 5716722.69 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3758, pruned_loss=0.1223, over 5657317.89 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:47:41,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:47:42,324 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:47:44,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1948, 1.2746, 1.0840, 0.9366], device='cuda:0'), covar=tensor([0.0911, 0.0451, 0.1003, 0.1057], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0449, 0.0521, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 08:47:47,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3081, 1.1583, 3.7496, 3.3191], device='cuda:0'), covar=tensor([0.1623, 0.2861, 0.0514, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0660, 0.0978, 0.0937], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 08:48:06,867 INFO [train.py:968] (0/2) Epoch 24, batch 10150, giga_loss[loss=0.2811, simple_loss=0.3531, pruned_loss=0.1045, over 29105.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.371, pruned_loss=0.1188, over 5662198.93 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3435, pruned_loss=0.08967, over 5719536.41 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3743, pruned_loss=0.1219, over 5655551.49 frames. ], batch size: 128, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:48:55,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.788e+02 1.796e+03 2.197e+03 3.285e+03 9.129e+03, threshold=4.394e+03, percent-clipped=16.0
+2023-03-12 08:48:58,380 INFO [train.py:968] (0/2) Epoch 24, batch 10200, giga_loss[loss=0.2635, simple_loss=0.3405, pruned_loss=0.09328, over 28953.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3691, pruned_loss=0.1181, over 5670746.73 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3435, pruned_loss=0.08964, over 5721244.12 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3719, pruned_loss=0.1208, over 5663628.16 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:49:18,587 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-12 08:49:48,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:49:49,070 INFO [train.py:968] (0/2) Epoch 24, batch 10250, giga_loss[loss=0.3411, simple_loss=0.3951, pruned_loss=0.1435, over 27928.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3685, pruned_loss=0.1189, over 5672192.47 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3431, pruned_loss=0.08932, over 5726535.52 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3718, pruned_loss=0.1221, over 5660510.96 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:50:23,695 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:50:31,884 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:50:31,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4891, 3.7722, 1.5816, 1.7363], device='cuda:0'), covar=tensor([0.0953, 0.0261, 0.0938, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0563, 0.0395, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 08:50:32,894 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.432e+02 1.770e+03 2.127e+03 3.243e+03 9.236e+03, threshold=4.254e+03, percent-clipped=10.0
+2023-03-12 08:50:35,648 INFO [train.py:968] (0/2) Epoch 24, batch 10300, giga_loss[loss=0.2591, simple_loss=0.335, pruned_loss=0.09157, over 28750.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3673, pruned_loss=0.1178, over 5678728.71 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.343, pruned_loss=0.08913, over 5732940.67 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.371, pruned_loss=0.1217, over 5661467.58 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:50:36,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:51:06,035 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-12 08:51:19,274 INFO [train.py:968] (0/2) Epoch 24, batch 10350, giga_loss[loss=0.3314, simple_loss=0.3868, pruned_loss=0.1381, over 26624.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3656, pruned_loss=0.1156, over 5664137.34 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3435, pruned_loss=0.08943, over 5727385.34 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3688, pruned_loss=0.1192, over 5654131.30 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:51:59,980 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:00,521 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:02,894 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:04,295 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.046e+02 1.553e+03 1.769e+03 2.492e+03 6.877e+03, threshold=3.538e+03, percent-clipped=4.0
+2023-03-12 08:52:06,457 INFO [train.py:968] (0/2) Epoch 24, batch 10400, giga_loss[loss=0.2801, simple_loss=0.357, pruned_loss=0.1016, over 28737.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3612, pruned_loss=0.1117, over 5669432.93 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3426, pruned_loss=0.08893, over 5732094.66 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3651, pruned_loss=0.1159, over 5655226.78 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:52:30,751 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:37,020 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:39,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:44,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:48,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:50,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:52,916 INFO [train.py:968] (0/2) Epoch 24, batch 10450, giga_loss[loss=0.2407, simple_loss=0.3226, pruned_loss=0.07942, over 28962.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3604, pruned_loss=0.111, over 5672303.96 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08887, over 5734852.70 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3644, pruned_loss=0.1151, over 5656437.16 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:53:06,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:53:18,648 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:53:40,097 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.314e+02 1.483e+03 1.973e+03 2.472e+03 6.919e+03, threshold=3.946e+03, percent-clipped=10.0
+2023-03-12 08:53:45,335 INFO [train.py:968] (0/2) Epoch 24, batch 10500, giga_loss[loss=0.4056, simple_loss=0.4186, pruned_loss=0.1963, over 23651.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3587, pruned_loss=0.111, over 5662704.94 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.08914, over 5732828.60 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3618, pruned_loss=0.1146, over 5650135.30 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:53:46,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:54:31,958 INFO [train.py:968] (0/2) Epoch 24, batch 10550, giga_loss[loss=0.2864, simple_loss=0.3544, pruned_loss=0.1092, over 28673.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3571, pruned_loss=0.1107, over 5671365.84 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08907, over 5733832.24 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3604, pruned_loss=0.1143, over 5658512.75 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:55:02,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:55:05,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:55:15,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.619e+03 2.368e+03 3.529e+03 1.142e+04, threshold=4.736e+03, percent-clipped=18.0
+2023-03-12 08:55:18,362 INFO [train.py:968] (0/2) Epoch 24, batch 10600, giga_loss[loss=0.2945, simple_loss=0.3682, pruned_loss=0.1104, over 28771.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3593, pruned_loss=0.1115, over 5677431.86 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3425, pruned_loss=0.08911, over 5738833.94 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3624, pruned_loss=0.115, over 5661267.03 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:55:31,053 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:56:00,047 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:56:02,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:56:05,146 INFO [train.py:968] (0/2) Epoch 24, batch 10650, giga_loss[loss=0.3002, simple_loss=0.3753, pruned_loss=0.1126, over 28909.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3607, pruned_loss=0.1116, over 5683122.59 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3422, pruned_loss=0.08891, over 5744003.89 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.364, pruned_loss=0.1154, over 5663428.50 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:56:25,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:56:27,504 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-12 08:56:33,107 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:56:55,096 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.548e+03 2.063e+03 3.028e+03 7.142e+03, threshold=4.125e+03, percent-clipped=5.0
+2023-03-12 08:56:57,275 INFO [train.py:968] (0/2) Epoch 24, batch 10700, giga_loss[loss=0.2905, simple_loss=0.3579, pruned_loss=0.1115, over 27934.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3632, pruned_loss=0.114, over 5650586.00 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3424, pruned_loss=0.08897, over 5743046.32 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3661, pruned_loss=0.1174, over 5634018.17 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:57:45,421 INFO [train.py:968] (0/2) Epoch 24, batch 10750, giga_loss[loss=0.2827, simple_loss=0.3523, pruned_loss=0.1066, over 28905.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3644, pruned_loss=0.1155, over 5645308.76 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3425, pruned_loss=0.08896, over 5742271.44 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3668, pruned_loss=0.1185, over 5632229.55 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:58:01,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:58:06,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1986, 1.2133, 3.4701, 3.1050], device='cuda:0'), covar=tensor([0.1630, 0.2795, 0.0502, 0.1304], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0658, 0.0973, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 08:58:11,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9056, 2.8899, 1.6239, 1.1555], device='cuda:0'), covar=tensor([0.8829, 0.3548, 0.5055, 0.7256], device='cuda:0'), in_proj_covar=tensor([0.1776, 0.1676, 0.1611, 0.1447], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 08:58:29,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.824e+02 1.664e+03 2.036e+03 3.008e+03 7.587e+03, threshold=4.072e+03, percent-clipped=7.0
+2023-03-12 08:58:33,983 INFO [train.py:968] (0/2) Epoch 24, batch 10800, giga_loss[loss=0.3191, simple_loss=0.3606, pruned_loss=0.1387, over 23203.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3656, pruned_loss=0.1166, over 5640907.70 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3426, pruned_loss=0.08895, over 5739636.46 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3678, pruned_loss=0.1194, over 5630763.09 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:58:48,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:58:50,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:59:21,254 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:59:24,934 INFO [train.py:968] (0/2) Epoch 24, batch 10850, giga_loss[loss=0.3203, simple_loss=0.365, pruned_loss=0.1378, over 23389.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3674, pruned_loss=0.1172, over 5651505.83 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.0892, over 5743051.82 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3694, pruned_loss=0.1199, over 5638157.98 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:00:01,593 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:00:11,853 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.604e+03 2.121e+03 3.106e+03 9.924e+03, threshold=4.241e+03, percent-clipped=10.0
+2023-03-12 09:00:13,312 INFO [train.py:968] (0/2) Epoch 24, batch 10900, giga_loss[loss=0.2578, simple_loss=0.3379, pruned_loss=0.08887, over 28470.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3682, pruned_loss=0.1179, over 5659502.86 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3425, pruned_loss=0.08895, over 5745256.03 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3704, pruned_loss=0.1206, over 5645871.71 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:00:23,235 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:00:23,675 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:00:26,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:00:56,338 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:00:58,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9608, 2.0718, 1.4861, 1.6383], device='cuda:0'), covar=tensor([0.1074, 0.0735, 0.1102, 0.1216], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0449, 0.0521, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 09:01:02,336 INFO [train.py:968] (0/2) Epoch 24, batch 10950, giga_loss[loss=0.2725, simple_loss=0.3468, pruned_loss=0.09906, over 28887.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3693, pruned_loss=0.1192, over 5648448.72 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3423, pruned_loss=0.08885, over 5737682.86 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3717, pruned_loss=0.1219, over 5642898.79 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:01:37,975 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 09:01:50,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.210e+03 1.794e+03 2.226e+03 3.006e+03 7.442e+03, threshold=4.453e+03, percent-clipped=4.0
+2023-03-12 09:01:53,656 INFO [train.py:968] (0/2) Epoch 24, batch 11000, giga_loss[loss=0.2667, simple_loss=0.3456, pruned_loss=0.09395, over 28697.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3716, pruned_loss=0.1204, over 5649233.91 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3427, pruned_loss=0.08891, over 5736765.57 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3736, pruned_loss=0.1229, over 5644812.79 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:02:47,644 INFO [train.py:968] (0/2) Epoch 24, batch 11050, giga_loss[loss=0.3031, simple_loss=0.375, pruned_loss=0.1156, over 28982.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3721, pruned_loss=0.1199, over 5649187.80 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08902, over 5737832.32 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3741, pruned_loss=0.1223, over 5643308.88 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:03:19,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059978.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:03:39,299 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.727e+03 2.341e+03 3.101e+03 5.712e+03, threshold=4.681e+03, percent-clipped=7.0
+2023-03-12 09:03:41,459 INFO [train.py:968] (0/2) Epoch 24, batch 11100, giga_loss[loss=0.3401, simple_loss=0.3997, pruned_loss=0.1403, over 28892.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3704, pruned_loss=0.1191, over 5651916.66 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.08919, over 5736413.99 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3723, pruned_loss=0.1213, over 5647092.09 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:03:43,224 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1060000.pt
+2023-03-12 09:04:37,547 INFO [train.py:968] (0/2) Epoch 24, batch 11150, libri_loss[loss=0.2694, simple_loss=0.3553, pruned_loss=0.09178, over 25696.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.369, pruned_loss=0.1188, over 5647970.09 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08928, over 5726181.18 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3709, pruned_loss=0.121, over 5651240.47 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:04:38,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5536, 1.6878, 1.6899, 1.4515], device='cuda:0'), covar=tensor([0.2918, 0.2587, 0.2001, 0.2435], device='cuda:0'), in_proj_covar=tensor([0.2010, 0.1967, 0.1890, 0.2027], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 09:05:28,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.768e+03 2.127e+03 2.874e+03 7.902e+03, threshold=4.253e+03, percent-clipped=8.0
+2023-03-12 09:05:28,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1612, 1.3894, 1.4026, 1.0758], device='cuda:0'), covar=tensor([0.1238, 0.1990, 0.1036, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0912, 0.0711, 0.0960, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 09:05:31,556 INFO [train.py:968] (0/2) Epoch 24, batch 11200, giga_loss[loss=0.3255, simple_loss=0.386, pruned_loss=0.1325, over 28622.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3692, pruned_loss=0.1196, over 5650306.54 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08929, over 5728876.87 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.371, pruned_loss=0.1219, over 5649385.36 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:05:38,686 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1060105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:06:20,683 INFO [train.py:968] (0/2) Epoch 24, batch 11250, libri_loss[loss=0.2523, simple_loss=0.3365, pruned_loss=0.08409, over 29510.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3679, pruned_loss=0.119, over 5662096.92 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3433, pruned_loss=0.08939, over 5733780.19 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.37, pruned_loss=0.1217, over 5654453.38 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:06:22,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8543, 1.1844, 2.8715, 2.8355], device='cuda:0'), covar=tensor([0.1680, 0.2523, 0.0618, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0657, 0.0974, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 09:06:28,933 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:06:52,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:07:01,604 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 1.787e+03 2.587e+03 3.613e+03 6.395e+03, threshold=5.173e+03, percent-clipped=16.0
+2023-03-12 09:07:04,322 INFO [train.py:968] (0/2) Epoch 24, batch 11300, giga_loss[loss=0.3534, simple_loss=0.3932, pruned_loss=0.1569, over 26627.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3678, pruned_loss=0.1191, over 5659398.29 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3435, pruned_loss=0.08958, over 5729150.17 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3699, pruned_loss=0.1219, over 5656555.61 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:07:21,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4651, 2.8417, 1.6210, 1.6344], device='cuda:0'), covar=tensor([0.0814, 0.0344, 0.0713, 0.1100], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0564, 0.0395, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 09:07:55,368 INFO [train.py:968] (0/2) Epoch 24, batch 11350, giga_loss[loss=0.3077, simple_loss=0.3721, pruned_loss=0.1217, over 27963.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3682, pruned_loss=0.1198, over 5660408.71 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3438, pruned_loss=0.08969, over 5732824.57 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.37, pruned_loss=0.1225, over 5653658.88 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:08:39,541 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1060294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:08:44,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.124e+03 1.638e+03 2.504e+03 3.344e+03 7.993e+03, threshold=5.008e+03, percent-clipped=6.0
+2023-03-12 09:08:45,552 INFO [train.py:968] (0/2) Epoch 24, batch 11400, giga_loss[loss=0.386, simple_loss=0.4255, pruned_loss=0.1732, over 27917.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3682, pruned_loss=0.12, over 5669899.35 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3432, pruned_loss=0.08936, over 5737046.95 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3708, pruned_loss=0.1231, over 5659114.55 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:08:49,690 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:08:52,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:08:54,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5113, 1.4798, 1.1890, 1.1558], device='cuda:0'), covar=tensor([0.0674, 0.0321, 0.0793, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0450, 0.0520, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 09:09:12,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1720, 4.0309, 3.8290, 2.0847], device='cuda:0'), covar=tensor([0.0684, 0.0763, 0.0799, 0.1883], device='cuda:0'), in_proj_covar=tensor([0.1280, 0.1180, 0.1000, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 09:09:14,295 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:09:18,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:09:20,240 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1060334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:09:20,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5448, 4.5184, 1.7024, 1.5905], device='cuda:0'), covar=tensor([0.0987, 0.0330, 0.0861, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0564, 0.0395, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 09:09:29,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3737, 1.6830, 1.3215, 1.3187], device='cuda:0'), covar=tensor([0.3047, 0.3082, 0.3609, 0.2645], device='cuda:0'), in_proj_covar=tensor([0.1543, 0.1112, 0.1358, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 09:09:31,064 INFO [train.py:968] (0/2) Epoch 24, batch 11450, libri_loss[loss=0.2245, simple_loss=0.3075, pruned_loss=0.0708, over 29359.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3687, pruned_loss=0.1204, over 5675942.29 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3421, pruned_loss=0.08887, over 5741587.42 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3729, pruned_loss=0.1247, over 5659795.22 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:09:36,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060353.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:09:42,988 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1060360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:10:11,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2712, 2.6103, 1.3264, 1.4812], device='cuda:0'), covar=tensor([0.0999, 0.0385, 0.0900, 0.1339], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0564, 0.0395, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 09:10:25,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.900e+03 2.825e+03 4.309e+03 1.147e+04, threshold=5.651e+03, percent-clipped=17.0
+2023-03-12 09:10:25,187 INFO [train.py:968] (0/2) Epoch 24, batch 11500, giga_loss[loss=0.4597, simple_loss=0.4666, pruned_loss=0.2264, over 26596.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3713, pruned_loss=0.1231, over 5669094.82 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08879, over 5742230.49 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.375, pruned_loss=0.1267, over 5655481.69 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 09:11:08,543 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1060439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:11:18,062 INFO [train.py:968] (0/2) Epoch 24, batch 11550, giga_loss[loss=0.2755, simple_loss=0.3499, pruned_loss=0.1006, over 28496.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3713, pruned_loss=0.1238, over 5661991.86 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3421, pruned_loss=0.08885, over 5744007.42 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3743, pruned_loss=0.127, over 5648969.80 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 09:11:19,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9401, 1.1619, 1.1128, 0.8582], device='cuda:0'), covar=tensor([0.2302, 0.2697, 0.1646, 0.2426], device='cuda:0'), in_proj_covar=tensor([0.2010, 0.1968, 0.1894, 0.2033], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 09:11:46,171 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:11:48,769 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1060483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:12:01,063 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060496.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:12:02,095 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.881e+03 2.451e+03 3.547e+03 8.236e+03, threshold=4.901e+03, percent-clipped=7.0
+2023-03-12 09:12:02,107 INFO [train.py:968] (0/2) Epoch 24, batch 11600, libri_loss[loss=0.2224, simple_loss=0.301, pruned_loss=0.07189, over 29617.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3703, pruned_loss=0.1225, over 5670263.39 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.0891, over 5749314.83 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3734, pruned_loss=0.126, over 5652489.98 frames. ], batch size: 69, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:12:04,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060499.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:12:30,864 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-12 09:12:35,526 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1060528.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:12:54,298 INFO [train.py:968] (0/2) Epoch 24, batch 11650, giga_loss[loss=0.2837, simple_loss=0.3586, pruned_loss=0.1044, over 29101.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3695, pruned_loss=0.1212, over 5674296.46 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08895, over 5752018.76 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3725, pruned_loss=0.1247, over 5656439.22 frames. ], batch size: 128, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:13:42,806 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.628e+03 2.077e+03 2.912e+03 6.188e+03, threshold=4.153e+03, percent-clipped=4.0
+2023-03-12 09:13:42,818 INFO [train.py:968] (0/2) Epoch 24, batch 11700, giga_loss[loss=0.2764, simple_loss=0.35, pruned_loss=0.1014, over 28996.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3697, pruned_loss=0.1208, over 5676689.74 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3421, pruned_loss=0.08877, over 5755780.87 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.373, pruned_loss=0.1246, over 5657084.30 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:14:06,943 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:14:10,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:14:31,465 INFO [train.py:968] (0/2) Epoch 24, batch 11750, giga_loss[loss=0.3262, simple_loss=0.391, pruned_loss=0.1307, over 28687.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3699, pruned_loss=0.1206, over 5694984.63 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3416, pruned_loss=0.08838, over 5760808.46 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3741, pruned_loss=0.1252, over 5671968.84 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:14:38,502 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1060655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:14:54,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:14:54,970 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-12 09:15:19,869 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.826e+03 2.546e+03 3.687e+03 1.160e+04, threshold=5.092e+03, percent-clipped=18.0
+2023-03-12 09:15:19,881 INFO [train.py:968] (0/2) Epoch 24, batch 11800, giga_loss[loss=0.3355, simple_loss=0.3933, pruned_loss=0.1388, over 28986.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.372, pruned_loss=0.1226, over 5680324.60 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3414, pruned_loss=0.08833, over 5753611.20 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3765, pruned_loss=0.1274, over 5666510.77 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:16:11,741 INFO [train.py:968] (0/2) Epoch 24, batch 11850, giga_loss[loss=0.3012, simple_loss=0.373, pruned_loss=0.1147, over 28982.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3717, pruned_loss=0.122, over 5686815.80 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3414, pruned_loss=0.08833, over 5753611.20 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3752, pruned_loss=0.1258, over 5676064.32 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:16:39,120 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 09:16:40,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1060776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:16:47,685 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1060783.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:16:54,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7172, 1.6760, 1.9328, 1.4856], device='cuda:0'), covar=tensor([0.1617, 0.2439, 0.1340, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.0913, 0.0710, 0.0960, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 09:17:01,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.587e+03 2.227e+03 3.032e+03 1.239e+04, threshold=4.453e+03, percent-clipped=4.0
+2023-03-12 09:17:01,262 INFO [train.py:968] (0/2) Epoch 24, batch 11900, libri_loss[loss=0.2389, simple_loss=0.3203, pruned_loss=0.07875, over 29540.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3724, pruned_loss=0.1217, over 5678225.21 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3412, pruned_loss=0.08823, over 5757423.01 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3762, pruned_loss=0.1257, over 5664343.24 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:17:16,625 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:17:17,965 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:17:18,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:17:41,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3617, 2.2141, 1.3693, 1.4744], device='cuda:0'), covar=tensor([0.0832, 0.0422, 0.0763, 0.1165], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0564, 0.0395, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 09:17:48,338 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1060844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:17:50,864 INFO [train.py:968] (0/2) Epoch 24, batch 11950, giga_loss[loss=0.2762, simple_loss=0.3486, pruned_loss=0.1019, over 28794.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3728, pruned_loss=0.1215, over 5675274.31 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3417, pruned_loss=0.08851, over 5760013.55 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3758, pruned_loss=0.1249, over 5660952.22 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:18:00,678 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:18:41,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.497e+03 1.762e+03 2.420e+03 8.846e+03, threshold=3.523e+03, percent-clipped=2.0
+2023-03-12 09:18:41,172 INFO [train.py:968] (0/2) Epoch 24, batch 12000, giga_loss[loss=0.2968, simple_loss=0.3725, pruned_loss=0.1106, over 28916.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3704, pruned_loss=0.1196, over 5675506.15 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3419, pruned_loss=0.08874, over 5749374.04 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3734, pruned_loss=0.1229, over 5672543.73 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:18:41,176 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 09:18:50,078 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2069, simple_loss=0.3143, pruned_loss=0.04977, over 944034.00 frames. 
+2023-03-12 09:18:50,079 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 09:19:07,643 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-12 09:19:36,416 INFO [train.py:968] (0/2) Epoch 24, batch 12050, giga_loss[loss=0.3362, simple_loss=0.3933, pruned_loss=0.1395, over 27916.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3707, pruned_loss=0.1204, over 5666154.03 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08881, over 5751865.42 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3734, pruned_loss=0.1234, over 5660529.28 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:19:43,612 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:19:45,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:20:14,841 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1060989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:20:24,635 INFO [train.py:968] (0/2) Epoch 24, batch 12100, giga_loss[loss=0.2938, simple_loss=0.3596, pruned_loss=0.1139, over 28939.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.371, pruned_loss=0.1203, over 5674070.48 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3414, pruned_loss=0.08845, over 5755679.92 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3746, pruned_loss=0.124, over 5663832.43 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:20:25,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.556e+02 1.857e+03 2.442e+03 3.338e+03 8.753e+03, threshold=4.884e+03, percent-clipped=18.0
+2023-03-12 09:20:27,587 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1061001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:20:29,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1061004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:20:37,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4500, 3.5122, 1.5844, 1.5750], device='cuda:0'), covar=tensor([0.1005, 0.0362, 0.0916, 0.1357], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0565, 0.0396, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 09:20:50,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1061024.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:20:50,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3655, 1.4536, 1.3141, 1.5190], device='cuda:0'), covar=tensor([0.0636, 0.0403, 0.0320, 0.0675], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 09:20:59,637 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1061033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:21:15,842 INFO [train.py:968] (0/2) Epoch 24, batch 12150, giga_loss[loss=0.2498, simple_loss=0.3287, pruned_loss=0.08546, over 28988.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3728, pruned_loss=0.1224, over 5673053.78 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3414, pruned_loss=0.0885, over 5756900.44 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3759, pruned_loss=0.1255, over 5663254.13 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:21:54,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 09:22:03,128 INFO [train.py:968] (0/2) Epoch 24, batch 12200, giga_loss[loss=0.3349, simple_loss=0.3885, pruned_loss=0.1406, over 27984.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3739, pruned_loss=0.124, over 5654965.59 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08879, over 5746468.98 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3765, pruned_loss=0.127, over 5655284.52 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:22:04,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.736e+03 2.244e+03 3.394e+03 9.100e+03, threshold=4.487e+03, percent-clipped=8.0
+2023-03-12 09:22:54,481 INFO [train.py:968] (0/2) Epoch 24, batch 12250, giga_loss[loss=0.2875, simple_loss=0.3615, pruned_loss=0.1068, over 28942.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3734, pruned_loss=0.1239, over 5661198.07 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08882, over 5748141.27 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3761, pruned_loss=0.1269, over 5658381.55 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:22:58,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1061151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:23:03,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1061158.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:23:36,719 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8516, 4.6649, 4.4675, 2.2137], device='cuda:0'), covar=tensor([0.0509, 0.0619, 0.0700, 0.1951], device='cuda:0'), in_proj_covar=tensor([0.1273, 0.1174, 0.0996, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 09:23:40,815 INFO [train.py:968] (0/2) Epoch 24, batch 12300, giga_loss[loss=0.3062, simple_loss=0.3775, pruned_loss=0.1175, over 28834.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3738, pruned_loss=0.1239, over 5662146.81 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3422, pruned_loss=0.08899, over 5752816.32 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3766, pruned_loss=0.1271, over 5653289.00 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:23:41,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.666e+03 2.108e+03 2.707e+03 4.759e+03, threshold=4.216e+03, percent-clipped=3.0
+2023-03-12 09:24:28,834 INFO [train.py:968] (0/2) Epoch 24, batch 12350, giga_loss[loss=0.3642, simple_loss=0.3946, pruned_loss=0.1669, over 23599.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3736, pruned_loss=0.1238, over 5658446.57 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3419, pruned_loss=0.08873, over 5755600.60 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.377, pruned_loss=0.1276, over 5646515.57 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:24:51,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-12 09:25:16,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1061292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:25:18,688 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1061294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:25:22,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1061297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:25:22,912 INFO [train.py:968] (0/2) Epoch 24, batch 12400, giga_loss[loss=0.2686, simple_loss=0.3459, pruned_loss=0.09562, over 28864.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3724, pruned_loss=0.1229, over 5648184.81 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08883, over 5757937.30 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3755, pruned_loss=0.1262, over 5635569.48 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:25:23,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.761e+02 1.671e+03 2.166e+03 3.257e+03 1.096e+04, threshold=4.332e+03, percent-clipped=12.0
+2023-03-12 09:25:26,072 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1061301.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:25:28,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1061304.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:25:51,090 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1061326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:25:56,935 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1061333.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:26:10,547 INFO [train.py:968] (0/2) Epoch 24, batch 12450, giga_loss[loss=0.3234, simple_loss=0.3852, pruned_loss=0.1307, over 28546.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1235, over 5650548.74 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3422, pruned_loss=0.08917, over 5759384.78 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3764, pruned_loss=0.1266, over 5636786.13 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:26:34,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0390, 2.3703, 1.1043, 1.2478], device='cuda:0'), covar=tensor([0.1190, 0.0540, 0.1003, 0.1537], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0566, 0.0397, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 09:26:57,458 INFO [train.py:968] (0/2) Epoch 24, batch 12500, libri_loss[loss=0.2691, simple_loss=0.3537, pruned_loss=0.09221, over 29525.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3727, pruned_loss=0.1226, over 5654520.08 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3425, pruned_loss=0.08929, over 5759328.87 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.375, pruned_loss=0.1253, over 5642689.67 frames. ], batch size: 82, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:26:58,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3027, 1.6009, 1.3334, 1.6465], device='cuda:0'), covar=tensor([0.0793, 0.0328, 0.0338, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0120, 0.0119, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 09:26:58,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1061399.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:26:58,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.675e+03 2.218e+03 3.055e+03 6.253e+03, threshold=4.435e+03, percent-clipped=6.0
+2023-03-12 09:27:23,529 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-12 09:27:45,448 INFO [train.py:968] (0/2) Epoch 24, batch 12550, giga_loss[loss=0.2957, simple_loss=0.3616, pruned_loss=0.1149, over 28657.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3709, pruned_loss=0.1213, over 5662603.18 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3426, pruned_loss=0.08941, over 5758726.17 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3734, pruned_loss=0.1243, over 5650671.70 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:28:21,814 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1061487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:28:36,537 INFO [train.py:968] (0/2) Epoch 24, batch 12600, giga_loss[loss=0.2862, simple_loss=0.3566, pruned_loss=0.1079, over 28933.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3682, pruned_loss=0.1192, over 5672101.00 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08941, over 5763795.20 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3709, pruned_loss=0.1225, over 5655298.22 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:28:37,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.877e+02 1.723e+03 2.183e+03 3.035e+03 7.521e+03, threshold=4.367e+03, percent-clipped=6.0
+2023-03-12 09:29:15,314 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1061542.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:29:17,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1061545.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:29:19,130 INFO [train.py:968] (0/2) Epoch 24, batch 12650, giga_loss[loss=0.3041, simple_loss=0.3696, pruned_loss=0.1193, over 28308.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3653, pruned_loss=0.1175, over 5686236.22 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3423, pruned_loss=0.08914, over 5768728.59 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.369, pruned_loss=0.1218, over 5664359.91 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:29:45,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1061574.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:30:08,235 INFO [train.py:968] (0/2) Epoch 24, batch 12700, libri_loss[loss=0.2666, simple_loss=0.3476, pruned_loss=0.09281, over 29557.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3637, pruned_loss=0.1178, over 5662426.53 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3425, pruned_loss=0.0892, over 5770222.57 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.367, pruned_loss=0.1218, over 5641245.23 frames. ], batch size: 76, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:30:10,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.198e+03 1.848e+03 2.338e+03 3.028e+03 6.673e+03, threshold=4.677e+03, percent-clipped=8.0
+2023-03-12 09:30:47,260 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-12 09:30:55,580 INFO [train.py:968] (0/2) Epoch 24, batch 12750, giga_loss[loss=0.2619, simple_loss=0.3294, pruned_loss=0.09721, over 28663.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3626, pruned_loss=0.118, over 5659525.55 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3421, pruned_loss=0.08894, over 5771834.66 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3658, pruned_loss=0.1218, over 5639829.16 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:31:12,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1061667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:31:39,082 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 09:31:42,168 INFO [train.py:968] (0/2) Epoch 24, batch 12800, giga_loss[loss=0.2762, simple_loss=0.3525, pruned_loss=0.09992, over 29026.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3621, pruned_loss=0.1175, over 5658179.12 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3424, pruned_loss=0.08913, over 5767380.54 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3653, pruned_loss=0.1214, over 5642807.17 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:31:43,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.726e+03 2.181e+03 2.794e+03 5.438e+03, threshold=4.362e+03, percent-clipped=5.0
+2023-03-12 09:32:02,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5532, 1.7507, 1.3276, 1.3166], device='cuda:0'), covar=tensor([0.0947, 0.0474, 0.0892, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0447, 0.0517, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 09:32:13,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7244, 1.7891, 1.4906, 1.8544], device='cuda:0'), covar=tensor([0.2858, 0.2950, 0.3252, 0.2580], device='cuda:0'), in_proj_covar=tensor([0.1545, 0.1114, 0.1362, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 09:32:27,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1061742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:32:33,581 INFO [train.py:968] (0/2) Epoch 24, batch 12850, giga_loss[loss=0.2673, simple_loss=0.3498, pruned_loss=0.09238, over 27961.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3615, pruned_loss=0.1152, over 5657876.08 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3424, pruned_loss=0.08909, over 5767742.86 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3642, pruned_loss=0.1187, over 5644061.81 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:32:40,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3874, 1.3900, 3.4841, 3.3668], device='cuda:0'), covar=tensor([0.1414, 0.2780, 0.0449, 0.1115], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0664, 0.0983, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 09:33:28,359 INFO [train.py:968] (0/2) Epoch 24, batch 12900, giga_loss[loss=0.2679, simple_loss=0.3235, pruned_loss=0.1061, over 24345.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3593, pruned_loss=0.1118, over 5651382.38 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3424, pruned_loss=0.08915, over 5770609.86 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3618, pruned_loss=0.115, over 5635704.60 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:33:31,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.932e+02 1.482e+03 1.888e+03 2.887e+03 6.717e+03, threshold=3.775e+03, percent-clipped=9.0
+2023-03-12 09:33:40,065 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1061810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:33:42,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1061813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:33:51,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1061822.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:34:09,292 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1061842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:34:16,835 INFO [train.py:968] (0/2) Epoch 24, batch 12950, giga_loss[loss=0.2708, simple_loss=0.3445, pruned_loss=0.09855, over 28836.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.356, pruned_loss=0.1079, over 5657001.61 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3418, pruned_loss=0.08901, over 5768330.15 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3591, pruned_loss=0.1114, over 5641791.37 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:34:23,415 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5630, 1.4103, 4.1253, 3.4821], device='cuda:0'), covar=tensor([0.1609, 0.2807, 0.0464, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0667, 0.0985, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 09:34:31,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1061862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:34:48,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4738, 1.6296, 1.6932, 1.2880], device='cuda:0'), covar=tensor([0.1943, 0.2793, 0.1657, 0.1979], device='cuda:0'), in_proj_covar=tensor([0.0910, 0.0705, 0.0957, 0.0854], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 09:35:05,476 INFO [train.py:968] (0/2) Epoch 24, batch 13000, giga_loss[loss=0.2402, simple_loss=0.3245, pruned_loss=0.0779, over 29027.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3535, pruned_loss=0.1055, over 5664878.74 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3414, pruned_loss=0.08902, over 5774396.09 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.357, pruned_loss=0.1091, over 5642926.27 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:35:07,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.680e+02 1.430e+03 1.932e+03 2.515e+03 1.290e+04, threshold=3.865e+03, percent-clipped=3.0
+2023-03-12 09:35:21,999 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1061914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:35:57,722 INFO [train.py:968] (0/2) Epoch 24, batch 13050, giga_loss[loss=0.274, simple_loss=0.3593, pruned_loss=0.09436, over 28023.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3505, pruned_loss=0.1023, over 5657088.61 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3415, pruned_loss=0.08914, over 5772397.70 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3532, pruned_loss=0.1052, over 5639962.02 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:36:49,896 INFO [train.py:968] (0/2) Epoch 24, batch 13100, giga_loss[loss=0.2701, simple_loss=0.3496, pruned_loss=0.09532, over 28033.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3499, pruned_loss=0.0996, over 5667001.79 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3417, pruned_loss=0.08931, over 5773091.46 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3521, pruned_loss=0.102, over 5650257.69 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:36:52,317 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1062000.pt
+2023-03-12 09:36:53,282 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.456e+03 1.937e+03 2.748e+03 6.060e+03, threshold=3.875e+03, percent-clipped=6.0
+2023-03-12 09:36:59,906 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:37:02,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:37:33,037 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:37:44,632 INFO [train.py:968] (0/2) Epoch 24, batch 13150, giga_loss[loss=0.291, simple_loss=0.3595, pruned_loss=0.1113, over 27635.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.351, pruned_loss=0.1006, over 5660744.21 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3417, pruned_loss=0.08947, over 5774268.14 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3529, pruned_loss=0.1025, over 5645564.59 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:38:33,556 INFO [train.py:968] (0/2) Epoch 24, batch 13200, giga_loss[loss=0.2211, simple_loss=0.3122, pruned_loss=0.06501, over 29061.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3488, pruned_loss=0.099, over 5664970.30 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3414, pruned_loss=0.0896, over 5775291.00 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3508, pruned_loss=0.1007, over 5648532.82 frames. ], batch size: 128, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:38:36,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.096e+02 1.513e+03 2.028e+03 2.770e+03 8.452e+03, threshold=4.055e+03, percent-clipped=11.0
+2023-03-12 09:38:49,997 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-12 09:38:53,437 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:38:58,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1062124.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:39:17,193 INFO [train.py:968] (0/2) Epoch 24, batch 13250, giga_loss[loss=0.2367, simple_loss=0.3217, pruned_loss=0.0758, over 28712.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.346, pruned_loss=0.09765, over 5651275.94 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3416, pruned_loss=0.09015, over 5771751.24 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3478, pruned_loss=0.09902, over 5634288.41 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:40:07,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062197.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:40:08,441 INFO [train.py:968] (0/2) Epoch 24, batch 13300, giga_loss[loss=0.2876, simple_loss=0.3594, pruned_loss=0.1079, over 28600.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3452, pruned_loss=0.09712, over 5650090.85 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3413, pruned_loss=0.09002, over 5773089.04 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.347, pruned_loss=0.09849, over 5632761.77 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:40:13,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.240e+02 1.524e+03 2.052e+03 2.755e+03 6.023e+03, threshold=4.104e+03, percent-clipped=6.0
+2023-03-12 09:40:39,890 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 09:40:59,456 INFO [train.py:968] (0/2) Epoch 24, batch 13350, giga_loss[loss=0.2552, simple_loss=0.3342, pruned_loss=0.08811, over 28647.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3449, pruned_loss=0.09657, over 5654851.53 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3409, pruned_loss=0.0898, over 5774184.41 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3468, pruned_loss=0.09799, over 5637758.03 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:41:11,430 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:41:14,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:41:19,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1080, 2.3784, 1.6473, 2.0170], device='cuda:0'), covar=tensor([0.0970, 0.0625, 0.0979, 0.1039], device='cuda:0'), in_proj_covar=tensor([0.0395, 0.0442, 0.0513, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 09:41:39,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:41:42,753 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:41:49,889 INFO [train.py:968] (0/2) Epoch 24, batch 13400, giga_loss[loss=0.2528, simple_loss=0.3344, pruned_loss=0.08556, over 28969.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3416, pruned_loss=0.09403, over 5658503.41 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3401, pruned_loss=0.08944, over 5777522.45 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3439, pruned_loss=0.09561, over 5639445.20 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:41:53,243 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.885e+02 1.441e+03 1.913e+03 2.902e+03 7.201e+03, threshold=3.826e+03, percent-clipped=8.0
+2023-03-12 09:42:06,749 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1062312.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:42:34,603 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062340.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:42:35,131 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1062341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:42:36,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062343.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:42:41,320 INFO [train.py:968] (0/2) Epoch 24, batch 13450, giga_loss[loss=0.251, simple_loss=0.3455, pruned_loss=0.07827, over 28834.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3388, pruned_loss=0.09164, over 5656704.49 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3398, pruned_loss=0.0893, over 5775678.66 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3409, pruned_loss=0.09308, over 5641012.10 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:42:45,363 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 09:42:51,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2114, 1.5426, 1.5229, 1.2310], device='cuda:0'), covar=tensor([0.2650, 0.2147, 0.1362, 0.2011], device='cuda:0'), in_proj_covar=tensor([0.1984, 0.1922, 0.1849, 0.1988], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 09:43:03,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0945, 1.3311, 2.7389, 2.6962], device='cuda:0'), covar=tensor([0.1315, 0.2297, 0.0527, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0660, 0.0974, 0.0934], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 09:43:11,404 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:43:12,708 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1062374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:43:40,391 INFO [train.py:968] (0/2) Epoch 24, batch 13500, libri_loss[loss=0.2186, simple_loss=0.2892, pruned_loss=0.07394, over 29503.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.334, pruned_loss=0.08909, over 5655252.17 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3392, pruned_loss=0.089, over 5774336.07 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3362, pruned_loss=0.09051, over 5642364.73 frames. ], batch size: 70, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:43:45,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.642e+02 1.285e+03 1.605e+03 2.250e+03 4.754e+03, threshold=3.210e+03, percent-clipped=5.0
+2023-03-12 09:43:47,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3726, 1.9180, 1.4206, 0.5895], device='cuda:0'), covar=tensor([0.5074, 0.2764, 0.3638, 0.6099], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1678, 0.1619, 0.1446], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 09:44:07,444 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-12 09:44:22,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:44:24,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:44:37,301 INFO [train.py:968] (0/2) Epoch 24, batch 13550, giga_loss[loss=0.2437, simple_loss=0.333, pruned_loss=0.07716, over 28833.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.333, pruned_loss=0.08878, over 5657529.09 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3388, pruned_loss=0.08895, over 5772066.89 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3349, pruned_loss=0.08996, over 5647500.12 frames. ], batch size: 285, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:44:53,797 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:45:00,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0555, 1.3278, 2.6917, 2.6624], device='cuda:0'), covar=tensor([0.1403, 0.2386, 0.0564, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0661, 0.0974, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 09:45:02,814 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=2.00 vs. limit=2.0
+2023-03-12 09:45:29,367 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-12 09:45:30,399 INFO [train.py:968] (0/2) Epoch 24, batch 13600, giga_loss[loss=0.2621, simple_loss=0.3256, pruned_loss=0.09925, over 26684.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3331, pruned_loss=0.08959, over 5636985.53 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3385, pruned_loss=0.08889, over 5764585.35 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3348, pruned_loss=0.09058, over 5634348.66 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:45:31,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062499.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:45:34,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.021e+02 1.438e+03 1.888e+03 2.979e+03 9.663e+03, threshold=3.775e+03, percent-clipped=20.0
+2023-03-12 09:46:24,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.23 vs. limit=5.0
+2023-03-12 09:46:28,261 INFO [train.py:968] (0/2) Epoch 24, batch 13650, giga_loss[loss=0.2871, simple_loss=0.3525, pruned_loss=0.1108, over 26781.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3346, pruned_loss=0.08968, over 5646476.79 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3379, pruned_loss=0.08865, over 5764838.06 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3364, pruned_loss=0.09073, over 5640763.60 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:47:28,660 INFO [train.py:968] (0/2) Epoch 24, batch 13700, libri_loss[loss=0.2285, simple_loss=0.289, pruned_loss=0.08404, over 29629.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3373, pruned_loss=0.0904, over 5654337.64 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3376, pruned_loss=0.08871, over 5769619.61 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.339, pruned_loss=0.09124, over 5641496.64 frames. ], batch size: 69, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:47:36,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.970e+02 1.607e+03 2.125e+03 3.102e+03 6.740e+03, threshold=4.250e+03, percent-clipped=12.0
+2023-03-12 09:48:21,096 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062642.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:48:23,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062645.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:48:27,216 INFO [train.py:968] (0/2) Epoch 24, batch 13750, giga_loss[loss=0.2819, simple_loss=0.3618, pruned_loss=0.101, over 28513.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3378, pruned_loss=0.09069, over 5664228.65 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3365, pruned_loss=0.08822, over 5770197.68 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3402, pruned_loss=0.09188, over 5649929.97 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:49:05,689 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062674.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:49:22,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062687.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:49:35,858 INFO [train.py:968] (0/2) Epoch 24, batch 13800, giga_loss[loss=0.2007, simple_loss=0.2895, pruned_loss=0.05591, over 28800.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.336, pruned_loss=0.08978, over 5665436.40 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3364, pruned_loss=0.08815, over 5770888.69 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.338, pruned_loss=0.09078, over 5653234.43 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:49:40,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.713e+02 1.371e+03 1.708e+03 2.155e+03 4.972e+03, threshold=3.415e+03, percent-clipped=2.0
+2023-03-12 09:49:44,104 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1062705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:49:55,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:50:35,455 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1833, 1.5277, 1.4824, 1.0621], device='cuda:0'), covar=tensor([0.1861, 0.2902, 0.1589, 0.1907], device='cuda:0'), in_proj_covar=tensor([0.0911, 0.0700, 0.0957, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 09:50:39,273 INFO [train.py:968] (0/2) Epoch 24, batch 13850, giga_loss[loss=0.2448, simple_loss=0.309, pruned_loss=0.09033, over 24028.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3343, pruned_loss=0.0879, over 5664265.27 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3359, pruned_loss=0.08783, over 5770678.48 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3363, pruned_loss=0.08901, over 5652294.57 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:50:41,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:51:41,179 INFO [train.py:968] (0/2) Epoch 24, batch 13900, giga_loss[loss=0.2541, simple_loss=0.3262, pruned_loss=0.09104, over 26903.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.334, pruned_loss=0.08691, over 5668467.55 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.336, pruned_loss=0.08789, over 5771774.88 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3355, pruned_loss=0.08772, over 5656004.04 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:51:47,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.094e+02 1.419e+03 1.809e+03 2.401e+03 6.015e+03, threshold=3.618e+03, percent-clipped=6.0
+2023-03-12 09:52:23,485 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062830.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:52:23,598 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 09:52:28,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062833.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:52:48,632 INFO [train.py:968] (0/2) Epoch 24, batch 13950, giga_loss[loss=0.202, simple_loss=0.2871, pruned_loss=0.05845, over 28770.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3315, pruned_loss=0.08662, over 5670781.43 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3357, pruned_loss=0.08786, over 5773266.07 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3329, pruned_loss=0.08727, over 5658922.42 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:53:00,707 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:53:04,146 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062862.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:53:04,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:53:39,990 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:53:41,048 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:53:45,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:53:48,971 INFO [train.py:968] (0/2) Epoch 24, batch 14000, giga_loss[loss=0.2306, simple_loss=0.3069, pruned_loss=0.0772, over 28952.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3305, pruned_loss=0.08678, over 5663095.44 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3358, pruned_loss=0.08803, over 5766666.21 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3314, pruned_loss=0.08712, over 5657836.31 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:53:54,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.209e+02 1.390e+03 1.744e+03 2.375e+03 4.364e+03, threshold=3.488e+03, percent-clipped=4.0
+2023-03-12 09:54:23,429 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:54:48,974 INFO [train.py:968] (0/2) Epoch 24, batch 14050, giga_loss[loss=0.3079, simple_loss=0.3639, pruned_loss=0.1259, over 26795.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.331, pruned_loss=0.08695, over 5660806.29 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3359, pruned_loss=0.08826, over 5765595.29 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3315, pruned_loss=0.08697, over 5655278.52 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:55:49,901 INFO [train.py:968] (0/2) Epoch 24, batch 14100, libri_loss[loss=0.2796, simple_loss=0.348, pruned_loss=0.1056, over 29533.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3323, pruned_loss=0.08644, over 5655263.52 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3361, pruned_loss=0.08832, over 5767563.97 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3325, pruned_loss=0.08635, over 5646537.73 frames. ], batch size: 80, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:55:55,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.768e+02 1.465e+03 1.898e+03 2.345e+03 5.922e+03, threshold=3.797e+03, percent-clipped=8.0
+2023-03-12 09:56:35,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7657, 2.0520, 1.2362, 1.6006], device='cuda:0'), covar=tensor([0.1271, 0.0801, 0.1532, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0443, 0.0514, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 09:56:50,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5848, 1.3185, 3.9383, 3.3238], device='cuda:0'), covar=tensor([0.1499, 0.2827, 0.0433, 0.1001], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0660, 0.0971, 0.0929], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 09:56:54,011 INFO [train.py:968] (0/2) Epoch 24, batch 14150, giga_loss[loss=0.2468, simple_loss=0.3282, pruned_loss=0.08266, over 28639.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3311, pruned_loss=0.08522, over 5665088.97 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3357, pruned_loss=0.08821, over 5769998.46 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3314, pruned_loss=0.08517, over 5653147.03 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:57:35,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1063080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:58:03,655 INFO [train.py:968] (0/2) Epoch 24, batch 14200, giga_loss[loss=0.3097, simple_loss=0.3818, pruned_loss=0.1188, over 28769.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3296, pruned_loss=0.08463, over 5675256.85 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3355, pruned_loss=0.08824, over 5772112.91 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3299, pruned_loss=0.08448, over 5661762.37 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:58:11,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.748e+02 1.347e+03 1.778e+03 2.427e+03 6.720e+03, threshold=3.556e+03, percent-clipped=4.0
+2023-03-12 09:59:14,658 INFO [train.py:968] (0/2) Epoch 24, batch 14250, giga_loss[loss=0.244, simple_loss=0.3328, pruned_loss=0.07755, over 28983.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3321, pruned_loss=0.08595, over 5683524.76 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3349, pruned_loss=0.08809, over 5774371.99 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3328, pruned_loss=0.08592, over 5669257.99 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:00:14,974 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1063192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:00:23,572 INFO [train.py:968] (0/2) Epoch 24, batch 14300, giga_loss[loss=0.2568, simple_loss=0.3504, pruned_loss=0.08158, over 29095.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.336, pruned_loss=0.08573, over 5671766.46 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3347, pruned_loss=0.08803, over 5766921.45 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3367, pruned_loss=0.08572, over 5665273.85 frames. ], batch size: 285, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:00:31,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.501e+02 1.460e+03 2.225e+03 3.040e+03 8.132e+03, threshold=4.451e+03, percent-clipped=17.0
+2023-03-12 10:00:40,695 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4476, 1.7183, 1.5488, 1.4681], device='cuda:0'), covar=tensor([0.0794, 0.0315, 0.0326, 0.0925], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 10:00:55,820 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1063223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:01:00,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1063226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:01:29,607 INFO [train.py:968] (0/2) Epoch 24, batch 14350, libri_loss[loss=0.2283, simple_loss=0.307, pruned_loss=0.07484, over 29574.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3386, pruned_loss=0.08552, over 5665923.10 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3349, pruned_loss=0.0882, over 5757652.17 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.339, pruned_loss=0.0853, over 5668548.58 frames. ], batch size: 79, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:01:39,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4057, 3.2466, 3.0925, 1.9088], device='cuda:0'), covar=tensor([0.0734, 0.0901, 0.0857, 0.1899], device='cuda:0'), in_proj_covar=tensor([0.1234, 0.1139, 0.0962, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 10:01:39,814 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1063255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:01:43,288 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-12 10:02:29,287 INFO [train.py:968] (0/2) Epoch 24, batch 14400, giga_loss[loss=0.2903, simple_loss=0.3529, pruned_loss=0.1139, over 26832.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3393, pruned_loss=0.08515, over 5664630.30 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3352, pruned_loss=0.08837, over 5758914.76 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3395, pruned_loss=0.0848, over 5664046.79 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:02:35,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.320e+02 1.335e+03 1.666e+03 2.132e+03 6.208e+03, threshold=3.333e+03, percent-clipped=3.0
+2023-03-12 10:03:34,702 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-12 10:03:36,752 INFO [train.py:968] (0/2) Epoch 24, batch 14450, giga_loss[loss=0.2855, simple_loss=0.3392, pruned_loss=0.1159, over 24598.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3395, pruned_loss=0.08601, over 5665048.39 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3351, pruned_loss=0.08832, over 5760997.07 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3398, pruned_loss=0.08572, over 5660984.25 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:04:38,546 INFO [train.py:968] (0/2) Epoch 24, batch 14500, giga_loss[loss=0.2587, simple_loss=0.3377, pruned_loss=0.08987, over 28109.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3385, pruned_loss=0.08668, over 5677387.15 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3355, pruned_loss=0.08862, over 5764330.59 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3385, pruned_loss=0.08612, over 5668696.52 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:04:48,396 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.686e+02 1.287e+03 1.773e+03 2.406e+03 6.304e+03, threshold=3.546e+03, percent-clipped=15.0
+2023-03-12 10:05:59,937 INFO [train.py:968] (0/2) Epoch 24, batch 14550, giga_loss[loss=0.2604, simple_loss=0.3405, pruned_loss=0.09014, over 29006.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3398, pruned_loss=0.08829, over 5688582.02 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3353, pruned_loss=0.08852, over 5765391.92 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.34, pruned_loss=0.08793, over 5679733.16 frames. ], batch size: 128, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:06:59,705 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1063485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:07:16,456 INFO [train.py:968] (0/2) Epoch 24, batch 14600, libri_loss[loss=0.2454, simple_loss=0.3224, pruned_loss=0.08418, over 29544.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3356, pruned_loss=0.08671, over 5689628.50 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3348, pruned_loss=0.08842, over 5769986.49 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3364, pruned_loss=0.08649, over 5675905.22 frames. ], batch size: 80, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:07:31,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.531e+02 1.322e+03 1.749e+03 2.685e+03 6.697e+03, threshold=3.499e+03, percent-clipped=10.0
+2023-03-12 10:08:29,865 INFO [train.py:968] (0/2) Epoch 24, batch 14650, libri_loss[loss=0.2443, simple_loss=0.3306, pruned_loss=0.07901, over 25589.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3323, pruned_loss=0.08462, over 5681074.39 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3345, pruned_loss=0.08819, over 5768673.90 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3331, pruned_loss=0.08456, over 5668856.25 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:08:57,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1063567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:09:38,582 INFO [train.py:968] (0/2) Epoch 24, batch 14700, giga_loss[loss=0.2251, simple_loss=0.3055, pruned_loss=0.07237, over 29030.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3297, pruned_loss=0.08376, over 5683957.71 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.334, pruned_loss=0.08793, over 5770981.61 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3308, pruned_loss=0.08386, over 5670647.31 frames. ], batch size: 120, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:09:43,567 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-12 10:09:47,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.904e+02 1.349e+03 1.845e+03 2.627e+03 6.694e+03, threshold=3.691e+03, percent-clipped=8.0
+2023-03-12 10:10:39,117 INFO [train.py:968] (0/2) Epoch 24, batch 14750, giga_loss[loss=0.3291, simple_loss=0.3916, pruned_loss=0.1333, over 26923.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3323, pruned_loss=0.08571, over 5675805.19 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3333, pruned_loss=0.08759, over 5764639.94 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3337, pruned_loss=0.08603, over 5668813.51 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:11:43,258 INFO [train.py:968] (0/2) Epoch 24, batch 14800, giga_loss[loss=0.2775, simple_loss=0.3564, pruned_loss=0.09927, over 28098.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3342, pruned_loss=0.0867, over 5678805.86 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3329, pruned_loss=0.08748, over 5761850.39 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3357, pruned_loss=0.08704, over 5674035.55 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:11:53,190 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.529e+02 1.445e+03 1.821e+03 2.619e+03 9.134e+03, threshold=3.643e+03, percent-clipped=12.0
+2023-03-12 10:11:59,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1063710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:12:05,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1063713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:12:21,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7581, 1.9068, 1.4895, 1.5185], device='cuda:0'), covar=tensor([0.1027, 0.0723, 0.1003, 0.1258], device='cuda:0'), in_proj_covar=tensor([0.0394, 0.0441, 0.0512, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 10:12:41,442 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1063742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:12:46,270 INFO [train.py:968] (0/2) Epoch 24, batch 14850, giga_loss[loss=0.2557, simple_loss=0.3294, pruned_loss=0.09096, over 26903.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3326, pruned_loss=0.08696, over 5671728.66 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3329, pruned_loss=0.08758, over 5756522.96 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3338, pruned_loss=0.08712, over 5670436.72 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:13:53,528 INFO [train.py:968] (0/2) Epoch 24, batch 14900, giga_loss[loss=0.2523, simple_loss=0.3286, pruned_loss=0.08806, over 26966.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3327, pruned_loss=0.0875, over 5675989.95 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3326, pruned_loss=0.08746, over 5757874.56 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3338, pruned_loss=0.08773, over 5672423.45 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:14:02,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.390e+03 1.725e+03 2.394e+03 6.736e+03, threshold=3.450e+03, percent-clipped=5.0
+2023-03-12 10:14:22,479 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 10:14:51,812 INFO [train.py:968] (0/2) Epoch 24, batch 14950, libri_loss[loss=0.2682, simple_loss=0.3512, pruned_loss=0.09267, over 28902.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3336, pruned_loss=0.08777, over 5681719.07 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3322, pruned_loss=0.08712, over 5756745.58 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.335, pruned_loss=0.08827, over 5676611.02 frames. ], batch size: 107, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:15:08,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1063860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:15:59,833 INFO [train.py:968] (0/2) Epoch 24, batch 15000, giga_loss[loss=0.3285, simple_loss=0.3929, pruned_loss=0.1321, over 28950.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3363, pruned_loss=0.08834, over 5683631.09 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3324, pruned_loss=0.08746, over 5759900.09 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3373, pruned_loss=0.08847, over 5674793.34 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:15:59,837 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 10:16:07,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1167, 1.2144, 3.4392, 3.1157], device='cuda:0'), covar=tensor([0.1964, 0.3153, 0.0563, 0.0971], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0659, 0.0967, 0.0925], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 10:16:08,944 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.1954, simple_loss=0.2963, pruned_loss=0.04726, over 944034.00 frames. 
+2023-03-12 10:16:08,945 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 10:16:17,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.883e+02 1.562e+03 1.941e+03 2.701e+03 6.340e+03, threshold=3.882e+03, percent-clipped=14.0
+2023-03-12 10:17:23,716 INFO [train.py:968] (0/2) Epoch 24, batch 15050, giga_loss[loss=0.2098, simple_loss=0.3027, pruned_loss=0.05847, over 28955.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3369, pruned_loss=0.08853, over 5673239.00 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3326, pruned_loss=0.08758, over 5759443.02 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3377, pruned_loss=0.08854, over 5665583.99 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:17:53,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7135, 4.5647, 4.2832, 2.1945], device='cuda:0'), covar=tensor([0.0529, 0.0681, 0.0798, 0.1885], device='cuda:0'), in_proj_covar=tensor([0.1229, 0.1131, 0.0956, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 10:18:37,279 INFO [train.py:968] (0/2) Epoch 24, batch 15100, giga_loss[loss=0.2718, simple_loss=0.3436, pruned_loss=0.1, over 28465.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3343, pruned_loss=0.08813, over 5664613.33 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3324, pruned_loss=0.0876, over 5753917.18 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3351, pruned_loss=0.08815, over 5660461.59 frames. ], batch size: 369, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:18:38,427 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1063998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:18:39,873 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1064000.pt
+2023-03-12 10:18:44,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1064003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:18:46,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.710e+02 1.589e+03 2.118e+03 3.074e+03 6.876e+03, threshold=4.237e+03, percent-clipped=14.0
+2023-03-12 10:18:46,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1064006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:19:29,405 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1064035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:19:37,038 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1064039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:19:46,370 INFO [train.py:968] (0/2) Epoch 24, batch 15150, giga_loss[loss=0.2161, simple_loss=0.285, pruned_loss=0.07363, over 26915.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3281, pruned_loss=0.0855, over 5662488.11 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3323, pruned_loss=0.08755, over 5754463.89 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3287, pruned_loss=0.08554, over 5656240.51 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:20:38,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4340, 1.2188, 4.1187, 3.4201], device='cuda:0'), covar=tensor([0.1597, 0.2978, 0.0421, 0.1292], device='cuda:0'), in_proj_covar=tensor([0.0771, 0.0659, 0.0965, 0.0923], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 10:20:50,800 INFO [train.py:968] (0/2) Epoch 24, batch 15200, giga_loss[loss=0.2089, simple_loss=0.3003, pruned_loss=0.05869, over 28971.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3262, pruned_loss=0.08436, over 5670653.96 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3321, pruned_loss=0.08739, over 5754222.41 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3269, pruned_loss=0.08448, over 5664011.48 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:20:59,808 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.004e+02 1.691e+03 2.343e+03 3.853e+03 1.093e+04, threshold=4.687e+03, percent-clipped=19.0
+2023-03-12 10:21:28,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4250, 1.7679, 1.4223, 1.2639], device='cuda:0'), covar=tensor([0.2518, 0.2393, 0.2736, 0.2317], device='cuda:0'), in_proj_covar=tensor([0.1542, 0.1107, 0.1362, 0.0997], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 10:21:32,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2620, 3.1580, 1.3870, 1.4336], device='cuda:0'), covar=tensor([0.1047, 0.0359, 0.0987, 0.1389], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0555, 0.0394, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 10:21:46,580 INFO [train.py:968] (0/2) Epoch 24, batch 15250, libri_loss[loss=0.2394, simple_loss=0.3264, pruned_loss=0.07616, over 29525.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3283, pruned_loss=0.08624, over 5664159.31 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3316, pruned_loss=0.08713, over 5756389.02 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3292, pruned_loss=0.08655, over 5654325.38 frames. ], batch size: 89, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:22:44,798 INFO [train.py:968] (0/2) Epoch 24, batch 15300, giga_loss[loss=0.2237, simple_loss=0.309, pruned_loss=0.06915, over 28485.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3268, pruned_loss=0.08489, over 5660086.74 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3313, pruned_loss=0.08707, over 5746949.35 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3276, pruned_loss=0.08515, over 5658358.96 frames. ], batch size: 369, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:22:53,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.628e+02 1.468e+03 1.792e+03 2.339e+03 4.321e+03, threshold=3.583e+03, percent-clipped=0.0
+2023-03-12 10:23:38,010 INFO [train.py:968] (0/2) Epoch 24, batch 15350, giga_loss[loss=0.2341, simple_loss=0.3011, pruned_loss=0.08355, over 24445.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3259, pruned_loss=0.08396, over 5645568.16 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3314, pruned_loss=0.08706, over 5739655.67 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3262, pruned_loss=0.08401, over 5645883.53 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:23:49,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5399, 1.3136, 4.4501, 3.5163], device='cuda:0'), covar=tensor([0.1675, 0.2966, 0.0441, 0.0979], device='cuda:0'), in_proj_covar=tensor([0.0769, 0.0656, 0.0964, 0.0922], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 10:24:45,488 INFO [train.py:968] (0/2) Epoch 24, batch 15400, giga_loss[loss=0.2466, simple_loss=0.3275, pruned_loss=0.08282, over 28899.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3254, pruned_loss=0.08337, over 5661444.38 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3315, pruned_loss=0.08711, over 5740571.48 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3255, pruned_loss=0.08331, over 5659645.68 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:24:59,185 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.958e+02 1.410e+03 1.797e+03 2.370e+03 5.115e+03, threshold=3.594e+03, percent-clipped=4.0
+2023-03-12 10:25:22,724 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1064321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:25:59,990 INFO [train.py:968] (0/2) Epoch 24, batch 15450, giga_loss[loss=0.2364, simple_loss=0.3013, pruned_loss=0.08577, over 24624.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3245, pruned_loss=0.08315, over 5648431.48 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3316, pruned_loss=0.08718, over 5738269.18 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3244, pruned_loss=0.083, over 5648555.64 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:26:34,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1064373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:26:46,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2343, 1.6096, 1.6040, 1.3998], device='cuda:0'), covar=tensor([0.1905, 0.1674, 0.1948, 0.1732], device='cuda:0'), in_proj_covar=tensor([0.0472, 0.0736, 0.0706, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 10:27:01,012 INFO [train.py:968] (0/2) Epoch 24, batch 15500, giga_loss[loss=0.2804, simple_loss=0.3576, pruned_loss=0.1016, over 28368.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3247, pruned_loss=0.08282, over 5645416.93 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3314, pruned_loss=0.08714, over 5733776.18 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3246, pruned_loss=0.08264, over 5646282.31 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:27:12,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.347e+02 1.385e+03 1.843e+03 2.437e+03 5.423e+03, threshold=3.685e+03, percent-clipped=7.0
+2023-03-12 10:27:25,941 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1064414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:28:08,405 INFO [train.py:968] (0/2) Epoch 24, batch 15550, giga_loss[loss=0.2884, simple_loss=0.3524, pruned_loss=0.1122, over 28961.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3258, pruned_loss=0.0839, over 5657197.23 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.331, pruned_loss=0.08696, over 5735859.44 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.326, pruned_loss=0.08386, over 5654316.01 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:29:14,533 INFO [train.py:968] (0/2) Epoch 24, batch 15600, giga_loss[loss=0.2382, simple_loss=0.3158, pruned_loss=0.08031, over 27666.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3249, pruned_loss=0.08399, over 5651559.71 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3312, pruned_loss=0.08724, over 5737148.65 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3248, pruned_loss=0.08367, over 5646593.03 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:29:27,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.402e+02 1.221e+03 1.506e+03 1.887e+03 6.586e+03, threshold=3.012e+03, percent-clipped=2.0
+2023-03-12 10:29:39,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1064516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:29:42,168 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1064519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:30:18,018 INFO [train.py:968] (0/2) Epoch 24, batch 15650, giga_loss[loss=0.2561, simple_loss=0.3376, pruned_loss=0.08732, over 27718.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3262, pruned_loss=0.08271, over 5665584.84 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3316, pruned_loss=0.08739, over 5738174.21 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3257, pruned_loss=0.08229, over 5660096.28 frames. ], batch size: 474, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:30:18,263 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1064548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:30:31,865 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1064557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:30:34,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1064560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:30:35,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8629, 1.2000, 2.8844, 2.7804], device='cuda:0'), covar=tensor([0.1684, 0.2538, 0.0587, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0769, 0.0655, 0.0963, 0.0920], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 10:31:10,257 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1064589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:31:23,509 INFO [train.py:968] (0/2) Epoch 24, batch 15700, giga_loss[loss=0.2284, simple_loss=0.3185, pruned_loss=0.06918, over 28885.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3287, pruned_loss=0.08361, over 5659503.62 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3313, pruned_loss=0.08724, over 5739594.29 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3285, pruned_loss=0.08337, over 5653409.40 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:31:37,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.907e+02 1.307e+03 1.674e+03 2.475e+03 6.755e+03, threshold=3.349e+03, percent-clipped=14.0
+2023-03-12 10:32:27,195 INFO [train.py:968] (0/2) Epoch 24, batch 15750, giga_loss[loss=0.2537, simple_loss=0.3363, pruned_loss=0.08557, over 28151.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3302, pruned_loss=0.08379, over 5662042.59 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3311, pruned_loss=0.08717, over 5740335.34 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3303, pruned_loss=0.08361, over 5655464.75 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:32:39,170 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1064657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:33:27,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1064696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:33:29,519 INFO [train.py:968] (0/2) Epoch 24, batch 15800, giga_loss[loss=0.2281, simple_loss=0.3081, pruned_loss=0.07403, over 28878.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3299, pruned_loss=0.08358, over 5674268.66 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.331, pruned_loss=0.08707, over 5739458.12 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.33, pruned_loss=0.08349, over 5668702.47 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:33:30,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3148, 1.1453, 3.7461, 3.2696], device='cuda:0'), covar=tensor([0.1671, 0.3010, 0.0452, 0.1225], device='cuda:0'), in_proj_covar=tensor([0.0767, 0.0653, 0.0960, 0.0917], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 10:33:42,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.614e+02 1.381e+03 1.753e+03 2.457e+03 8.676e+03, threshold=3.506e+03, percent-clipped=13.0
+2023-03-12 10:34:27,234 INFO [train.py:968] (0/2) Epoch 24, batch 15850, giga_loss[loss=0.1889, simple_loss=0.2796, pruned_loss=0.04911, over 28942.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3276, pruned_loss=0.08204, over 5687028.26 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3305, pruned_loss=0.08678, over 5745848.11 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3281, pruned_loss=0.08207, over 5673904.92 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:35:29,292 INFO [train.py:968] (0/2) Epoch 24, batch 15900, giga_loss[loss=0.2381, simple_loss=0.3118, pruned_loss=0.08223, over 26840.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3256, pruned_loss=0.08079, over 5692786.15 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3299, pruned_loss=0.08643, over 5751420.44 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3263, pruned_loss=0.08091, over 5674788.03 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:35:34,941 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7916, 3.6155, 3.4060, 1.6339], device='cuda:0'), covar=tensor([0.0752, 0.0891, 0.0902, 0.2327], device='cuda:0'), in_proj_covar=tensor([0.1230, 0.1132, 0.0955, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 10:35:42,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.336e+02 1.344e+03 1.773e+03 2.415e+03 9.472e+03, threshold=3.545e+03, percent-clipped=8.0
+2023-03-12 10:36:21,641 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1064839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:36:24,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1064842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:36:31,647 INFO [train.py:968] (0/2) Epoch 24, batch 15950, giga_loss[loss=0.2016, simple_loss=0.2799, pruned_loss=0.06161, over 28809.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3246, pruned_loss=0.0813, over 5681176.23 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.33, pruned_loss=0.0865, over 5742945.41 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.325, pruned_loss=0.08126, over 5673584.85 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:36:53,435 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 10:37:02,843 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1064871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:37:32,479 INFO [train.py:968] (0/2) Epoch 24, batch 16000, giga_loss[loss=0.2698, simple_loss=0.3567, pruned_loss=0.09148, over 28901.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3255, pruned_loss=0.08191, over 5681770.28 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3299, pruned_loss=0.0864, over 5747681.11 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3258, pruned_loss=0.08183, over 5669487.85 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:37:44,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.631e+02 1.369e+03 1.632e+03 2.308e+03 6.953e+03, threshold=3.264e+03, percent-clipped=7.0
+2023-03-12 10:38:33,964 INFO [train.py:968] (0/2) Epoch 24, batch 16050, giga_loss[loss=0.2663, simple_loss=0.3369, pruned_loss=0.0978, over 27631.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3285, pruned_loss=0.08341, over 5682265.87 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3302, pruned_loss=0.08659, over 5750177.36 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3284, pruned_loss=0.08306, over 5668466.27 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:38:58,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3350, 1.5564, 1.5320, 1.1642], device='cuda:0'), covar=tensor([0.1754, 0.2622, 0.1519, 0.1856], device='cuda:0'), in_proj_covar=tensor([0.0910, 0.0696, 0.0957, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 10:39:43,816 INFO [train.py:968] (0/2) Epoch 24, batch 16100, giga_loss[loss=0.2564, simple_loss=0.3395, pruned_loss=0.08667, over 28656.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3292, pruned_loss=0.0844, over 5680578.18 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.33, pruned_loss=0.08653, over 5751319.77 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3293, pruned_loss=0.08415, over 5667643.48 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:39:57,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.522e+03 1.932e+03 2.778e+03 7.648e+03, threshold=3.863e+03, percent-clipped=19.0
+2023-03-12 10:40:27,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:40:46,287 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9334, 1.2824, 5.0678, 3.6828], device='cuda:0'), covar=tensor([0.1454, 0.2925, 0.0409, 0.0809], device='cuda:0'), in_proj_covar=tensor([0.0770, 0.0656, 0.0965, 0.0922], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 10:40:47,728 INFO [train.py:968] (0/2) Epoch 24, batch 16150, giga_loss[loss=0.2794, simple_loss=0.3609, pruned_loss=0.09896, over 28943.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3318, pruned_loss=0.08542, over 5684740.35 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3298, pruned_loss=0.08638, over 5752910.69 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3321, pruned_loss=0.08535, over 5672512.31 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:41:08,090 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 10:41:45,267 INFO [train.py:968] (0/2) Epoch 24, batch 16200, giga_loss[loss=0.2693, simple_loss=0.3507, pruned_loss=0.09397, over 28973.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3343, pruned_loss=0.08595, over 5691550.79 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3297, pruned_loss=0.0863, over 5757370.68 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3347, pruned_loss=0.08595, over 5675889.25 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:41:57,650 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:41:58,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.980e+02 1.402e+03 1.930e+03 2.636e+03 5.442e+03, threshold=3.860e+03, percent-clipped=6.0
+2023-03-12 10:42:07,175 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:42:42,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5822, 1.8788, 1.4777, 1.7380], device='cuda:0'), covar=tensor([0.2827, 0.2750, 0.3195, 0.2425], device='cuda:0'), in_proj_covar=tensor([0.1542, 0.1106, 0.1362, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 10:42:54,342 INFO [train.py:968] (0/2) Epoch 24, batch 16250, giga_loss[loss=0.2987, simple_loss=0.3759, pruned_loss=0.1108, over 28507.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3351, pruned_loss=0.08614, over 5689753.15 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3296, pruned_loss=0.08622, over 5758868.27 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3355, pruned_loss=0.08621, over 5675675.49 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:43:36,278 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:43:40,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:44:04,044 INFO [train.py:968] (0/2) Epoch 24, batch 16300, giga_loss[loss=0.2151, simple_loss=0.3054, pruned_loss=0.06235, over 28886.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3333, pruned_loss=0.0854, over 5696171.40 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3293, pruned_loss=0.08613, over 5762532.14 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.334, pruned_loss=0.08551, over 5680108.02 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:44:18,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:44:19,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.568e+03 2.011e+03 2.609e+03 7.553e+03, threshold=4.022e+03, percent-clipped=6.0
+2023-03-12 10:44:24,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3731, 1.6167, 1.1697, 1.1721], device='cuda:0'), covar=tensor([0.1023, 0.0483, 0.1055, 0.1111], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0445, 0.0520, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 10:44:26,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4460, 3.5631, 1.4483, 1.6585], device='cuda:0'), covar=tensor([0.1017, 0.0377, 0.0997, 0.1357], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0557, 0.0395, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 10:45:05,760 INFO [train.py:968] (0/2) Epoch 24, batch 16350, giga_loss[loss=0.2623, simple_loss=0.3458, pruned_loss=0.08939, over 28911.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3324, pruned_loss=0.0856, over 5688367.73 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3291, pruned_loss=0.08615, over 5757489.78 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3333, pruned_loss=0.08565, over 5677425.09 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:46:10,730 INFO [train.py:968] (0/2) Epoch 24, batch 16400, giga_loss[loss=0.2427, simple_loss=0.3155, pruned_loss=0.08494, over 28892.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3319, pruned_loss=0.08567, over 5675753.43 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3289, pruned_loss=0.08604, over 5758625.94 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3327, pruned_loss=0.08581, over 5665483.91 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:46:27,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.414e+02 1.379e+03 1.655e+03 2.415e+03 5.907e+03, threshold=3.310e+03, percent-clipped=3.0
+2023-03-12 10:46:29,272 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:46:57,501 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:47:14,581 INFO [train.py:968] (0/2) Epoch 24, batch 16450, giga_loss[loss=0.2091, simple_loss=0.2964, pruned_loss=0.06087, over 28825.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3307, pruned_loss=0.08623, over 5672943.93 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3289, pruned_loss=0.08603, over 5756462.05 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3315, pruned_loss=0.08637, over 5663964.28 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:47:49,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3288, 1.2860, 3.8634, 3.3547], device='cuda:0'), covar=tensor([0.1684, 0.3013, 0.0463, 0.1706], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0658, 0.0969, 0.0924], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 10:48:14,097 INFO [train.py:968] (0/2) Epoch 24, batch 16500, giga_loss[loss=0.2552, simple_loss=0.3393, pruned_loss=0.08555, over 28923.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3306, pruned_loss=0.08629, over 5681912.83 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3287, pruned_loss=0.08592, over 5759921.49 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3315, pruned_loss=0.0865, over 5670120.57 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:48:20,327 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065402.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 10:48:32,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.551e+02 1.554e+03 2.136e+03 3.193e+03 1.259e+04, threshold=4.272e+03, percent-clipped=22.0
+2023-03-12 10:49:19,871 INFO [train.py:968] (0/2) Epoch 24, batch 16550, giga_loss[loss=0.1993, simple_loss=0.2736, pruned_loss=0.06247, over 24482.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3301, pruned_loss=0.08538, over 5674582.98 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3284, pruned_loss=0.0858, over 5761225.65 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3311, pruned_loss=0.08565, over 5663508.20 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:49:49,012 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.91 vs. limit=2.0
+2023-03-12 10:49:57,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5033, 4.3920, 4.1433, 2.1104], device='cuda:0'), covar=tensor([0.0490, 0.0624, 0.0680, 0.1862], device='cuda:0'), in_proj_covar=tensor([0.1232, 0.1133, 0.0957, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 10:50:03,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:50:13,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:50:19,711 INFO [train.py:968] (0/2) Epoch 24, batch 16600, libri_loss[loss=0.2258, simple_loss=0.3054, pruned_loss=0.07313, over 29364.00 frames. ], tot_loss[loss=0.25, simple_loss=0.331, pruned_loss=0.08452, over 5681630.29 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3279, pruned_loss=0.08552, over 5762204.16 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3323, pruned_loss=0.08497, over 5670402.25 frames. ], batch size: 67, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:50:34,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.372e+02 1.546e+03 2.145e+03 3.219e+03 1.780e+04, threshold=4.290e+03, percent-clipped=12.0
+2023-03-12 10:51:19,583 INFO [train.py:968] (0/2) Epoch 24, batch 16650, giga_loss[loss=0.2477, simple_loss=0.3369, pruned_loss=0.07921, over 28775.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3318, pruned_loss=0.08338, over 5676042.16 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3278, pruned_loss=0.08545, over 5762934.14 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3329, pruned_loss=0.08379, over 5666219.53 frames. ], batch size: 243, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:52:19,633 INFO [train.py:968] (0/2) Epoch 24, batch 16700, giga_loss[loss=0.2585, simple_loss=0.3394, pruned_loss=0.08875, over 28650.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.331, pruned_loss=0.08206, over 5690705.21 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3276, pruned_loss=0.08527, over 5765677.07 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3322, pruned_loss=0.08249, over 5678802.54 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:52:34,833 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.755e+02 1.280e+03 1.703e+03 2.370e+03 9.947e+03, threshold=3.407e+03, percent-clipped=3.0
+2023-03-12 10:52:39,535 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:52:59,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:53:02,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:53:10,814 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:53:14,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:53:30,474 INFO [train.py:968] (0/2) Epoch 24, batch 16750, giga_loss[loss=0.2599, simple_loss=0.3287, pruned_loss=0.09553, over 24328.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3317, pruned_loss=0.08291, over 5679103.82 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3277, pruned_loss=0.08532, over 5767069.51 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3325, pruned_loss=0.08319, over 5667771.09 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:53:45,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:53:56,792 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:54:09,913 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.87 vs. limit=2.0
+2023-03-12 10:54:22,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:54:41,997 INFO [train.py:968] (0/2) Epoch 24, batch 16800, giga_loss[loss=0.2365, simple_loss=0.3248, pruned_loss=0.0741, over 28442.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3312, pruned_loss=0.08243, over 5681552.08 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3271, pruned_loss=0.085, over 5770180.07 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3325, pruned_loss=0.08289, over 5667938.56 frames. ], batch size: 369, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:54:59,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.946e+02 1.561e+03 1.935e+03 2.816e+03 6.125e+03, threshold=3.870e+03, percent-clipped=12.0
+2023-03-12 10:54:59,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:55:52,995 INFO [train.py:968] (0/2) Epoch 24, batch 16850, libri_loss[loss=0.2283, simple_loss=0.3032, pruned_loss=0.07668, over 29570.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3319, pruned_loss=0.08252, over 5672295.35 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3272, pruned_loss=0.08506, over 5761940.01 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.333, pruned_loss=0.08279, over 5667170.97 frames. ], batch size: 75, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:56:20,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3819, 1.7617, 1.5139, 1.4958], device='cuda:0'), covar=tensor([0.0785, 0.0347, 0.0344, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0112], device='cuda:0')
+2023-03-12 10:56:31,587 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065777.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 10:56:58,500 INFO [train.py:968] (0/2) Epoch 24, batch 16900, giga_loss[loss=0.3083, simple_loss=0.3863, pruned_loss=0.1151, over 28974.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3326, pruned_loss=0.08303, over 5673146.02 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3274, pruned_loss=0.08531, over 5756919.21 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3334, pruned_loss=0.08288, over 5670617.03 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:57:15,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.531e+02 1.598e+03 1.955e+03 3.037e+03 7.736e+03, threshold=3.909e+03, percent-clipped=14.0
+2023-03-12 10:57:41,523 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:57:44,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:58:08,342 INFO [train.py:968] (0/2) Epoch 24, batch 16950, giga_loss[loss=0.2648, simple_loss=0.3522, pruned_loss=0.08872, over 29057.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3364, pruned_loss=0.08493, over 5682841.78 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3271, pruned_loss=0.08517, over 5761255.20 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3375, pruned_loss=0.0849, over 5674767.58 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:58:17,155 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:58:19,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:58:24,592 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:58:55,350 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:59:13,425 INFO [train.py:968] (0/2) Epoch 24, batch 17000, giga_loss[loss=0.2464, simple_loss=0.3294, pruned_loss=0.08175, over 28784.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3356, pruned_loss=0.08445, over 5682448.37 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3269, pruned_loss=0.08502, over 5763137.32 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.337, pruned_loss=0.08453, over 5672331.71 frames. ], batch size: 243, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:59:29,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.371e+02 1.452e+03 1.998e+03 2.771e+03 7.882e+03, threshold=3.995e+03, percent-clipped=11.0
+2023-03-12 10:59:42,385 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065920.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 10:59:46,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065923.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 11:00:11,258 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:00:22,125 INFO [train.py:968] (0/2) Epoch 24, batch 17050, giga_loss[loss=0.1887, simple_loss=0.275, pruned_loss=0.05115, over 28482.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3337, pruned_loss=0.08386, over 5693215.38 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.327, pruned_loss=0.08513, over 5762166.99 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3348, pruned_loss=0.08382, over 5684564.16 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:00:33,019 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065952.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 11:01:13,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9398, 1.3159, 1.1119, 0.2263], device='cuda:0'), covar=tensor([0.4374, 0.3409, 0.5073, 0.6948], device='cuda:0'), in_proj_covar=tensor([0.1770, 0.1668, 0.1615, 0.1447], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 11:01:19,420 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:01:34,735 INFO [train.py:968] (0/2) Epoch 24, batch 17100, giga_loss[loss=0.2233, simple_loss=0.3146, pruned_loss=0.06603, over 28681.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3319, pruned_loss=0.08254, over 5695854.30 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.327, pruned_loss=0.08515, over 5766058.41 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3329, pruned_loss=0.08245, over 5683814.72 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:01:36,133 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1066000.pt
+2023-03-12 11:01:44,756 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2608, 2.4544, 1.3184, 1.3589], device='cuda:0'), covar=tensor([0.1027, 0.0387, 0.0965, 0.1395], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0554, 0.0394, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 11:01:57,413 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.664e+02 1.359e+03 1.776e+03 2.929e+03 9.518e+03, threshold=3.553e+03, percent-clipped=12.0
+2023-03-12 11:01:59,803 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1066014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:02:43,803 INFO [train.py:968] (0/2) Epoch 24, batch 17150, giga_loss[loss=0.2429, simple_loss=0.3286, pruned_loss=0.0786, over 28479.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3297, pruned_loss=0.08074, over 5704549.09 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3271, pruned_loss=0.08526, over 5765486.46 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3306, pruned_loss=0.08045, over 5693883.92 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:03:42,527 INFO [train.py:968] (0/2) Epoch 24, batch 17200, giga_loss[loss=0.2244, simple_loss=0.3108, pruned_loss=0.06898, over 29093.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3299, pruned_loss=0.08164, over 5689920.26 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3269, pruned_loss=0.08533, over 5760440.56 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3309, pruned_loss=0.08123, over 5683779.40 frames. ], batch size: 200, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:03:57,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.646e+02 1.456e+03 2.016e+03 3.018e+03 1.204e+04, threshold=4.031e+03, percent-clipped=15.0
+2023-03-12 11:04:21,883 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1066129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:04:26,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1066132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:04:43,022 INFO [train.py:968] (0/2) Epoch 24, batch 17250, giga_loss[loss=0.2388, simple_loss=0.3268, pruned_loss=0.07541, over 28895.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3315, pruned_loss=0.08226, over 5692840.15 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3262, pruned_loss=0.08494, over 5764295.57 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.333, pruned_loss=0.08221, over 5682918.05 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:04:58,960 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1066161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:05:39,288 INFO [train.py:968] (0/2) Epoch 24, batch 17300, giga_loss[loss=0.238, simple_loss=0.3242, pruned_loss=0.0759, over 28855.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3318, pruned_loss=0.08283, over 5687264.48 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3259, pruned_loss=0.08474, over 5765348.52 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3334, pruned_loss=0.08294, over 5677118.25 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:05:53,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.210e+02 1.500e+03 1.921e+03 2.500e+03 6.636e+03, threshold=3.842e+03, percent-clipped=5.0
+2023-03-12 11:06:35,809 INFO [train.py:968] (0/2) Epoch 24, batch 17350, giga_loss[loss=0.2706, simple_loss=0.3424, pruned_loss=0.09938, over 27626.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3292, pruned_loss=0.08259, over 5687026.63 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3256, pruned_loss=0.08461, over 5768316.84 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3309, pruned_loss=0.08276, over 5674566.84 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:07:01,048 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-12 11:07:37,580 INFO [train.py:968] (0/2) Epoch 24, batch 17400, giga_loss[loss=0.2516, simple_loss=0.3329, pruned_loss=0.08514, over 28833.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3296, pruned_loss=0.08342, over 5688213.62 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3255, pruned_loss=0.08458, over 5769678.39 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.331, pruned_loss=0.08356, over 5676360.89 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:07:57,080 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.512e+02 1.354e+03 1.815e+03 2.549e+03 6.099e+03, threshold=3.630e+03, percent-clipped=6.0
+2023-03-12 11:07:58,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1066313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:08:09,995 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 11:08:31,380 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-12 11:08:34,646 INFO [train.py:968] (0/2) Epoch 24, batch 17450, giga_loss[loss=0.2735, simple_loss=0.3525, pruned_loss=0.09727, over 28378.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3347, pruned_loss=0.08704, over 5688731.19 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.325, pruned_loss=0.08434, over 5771673.57 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3365, pruned_loss=0.08738, over 5675400.73 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:09:05,026 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9428, 2.0914, 1.4520, 1.7158], device='cuda:0'), covar=tensor([0.0992, 0.0671, 0.1048, 0.1167], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0439, 0.0514, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 11:09:16,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1066389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:09:23,952 INFO [train.py:968] (0/2) Epoch 24, batch 17500, giga_loss[loss=0.2845, simple_loss=0.3643, pruned_loss=0.1024, over 28607.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.343, pruned_loss=0.09143, over 5685410.68 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3248, pruned_loss=0.08426, over 5762641.75 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.345, pruned_loss=0.09191, over 5680163.57 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:09:34,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.428e+03 1.922e+03 2.736e+03 7.836e+03, threshold=3.844e+03, percent-clipped=7.0
+2023-03-12 11:09:55,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1066436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:10:04,431 INFO [train.py:968] (0/2) Epoch 24, batch 17550, giga_loss[loss=0.2598, simple_loss=0.3429, pruned_loss=0.08832, over 29096.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3466, pruned_loss=0.09316, over 5699191.28 frames. ], libri_tot_loss[loss=0.246, simple_loss=0.3242, pruned_loss=0.0839, over 5767550.89 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3495, pruned_loss=0.09416, over 5688245.12 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:10:12,977 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1066456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:10:15,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1066459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:10:37,334 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 11:10:43,159 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1066488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:10:50,770 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1066496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:10:52,326 INFO [train.py:968] (0/2) Epoch 24, batch 17600, giga_loss[loss=0.2024, simple_loss=0.288, pruned_loss=0.05839, over 28366.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3431, pruned_loss=0.09217, over 5694523.09 frames. ], libri_tot_loss[loss=0.2463, simple_loss=0.3245, pruned_loss=0.08406, over 5766551.75 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3454, pruned_loss=0.09299, over 5685343.12 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:11:08,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.278e+02 1.294e+03 1.635e+03 2.319e+03 6.454e+03, threshold=3.270e+03, percent-clipped=4.0
+2023-03-12 11:11:27,231 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1066532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:11:29,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1066535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:11:30,754 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.58 vs. limit=5.0
+2023-03-12 11:11:41,262 INFO [train.py:968] (0/2) Epoch 24, batch 17650, giga_loss[loss=0.2531, simple_loss=0.3266, pruned_loss=0.08978, over 28885.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3361, pruned_loss=0.08957, over 5689793.01 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3244, pruned_loss=0.08397, over 5768628.07 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3382, pruned_loss=0.09039, over 5679574.75 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:11:56,203 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1066564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:12:26,618 INFO [train.py:968] (0/2) Epoch 24, batch 17700, giga_loss[loss=0.255, simple_loss=0.325, pruned_loss=0.0925, over 28393.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3292, pruned_loss=0.08695, over 5692363.54 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3243, pruned_loss=0.08392, over 5771684.04 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3311, pruned_loss=0.08771, over 5680143.41 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:12:37,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.068e+02 1.251e+03 1.575e+03 1.892e+03 3.608e+03, threshold=3.149e+03, percent-clipped=1.0
+2023-03-12 11:12:54,031 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3918, 2.5244, 2.3090, 1.9434], device='cuda:0'), covar=tensor([0.1858, 0.2091, 0.2035, 0.2275], device='cuda:0'), in_proj_covar=tensor([0.0467, 0.0734, 0.0706, 0.0676], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 11:13:09,075 INFO [train.py:968] (0/2) Epoch 24, batch 17750, giga_loss[loss=0.2066, simple_loss=0.2856, pruned_loss=0.06381, over 28803.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3234, pruned_loss=0.08447, over 5695622.88 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3251, pruned_loss=0.0841, over 5772350.88 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3243, pruned_loss=0.08496, over 5682311.51 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:13:50,411 INFO [train.py:968] (0/2) Epoch 24, batch 17800, libri_loss[loss=0.1937, simple_loss=0.2815, pruned_loss=0.05291, over 28506.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3184, pruned_loss=0.08255, over 5699114.84 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3251, pruned_loss=0.08397, over 5774000.96 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3188, pruned_loss=0.08305, over 5683403.42 frames. ], batch size: 63, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:14:03,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.047e+02 1.121e+03 1.342e+03 1.700e+03 4.337e+03, threshold=2.683e+03, percent-clipped=6.0
+2023-03-12 11:14:21,449 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 11:14:32,589 INFO [train.py:968] (0/2) Epoch 24, batch 17850, giga_loss[loss=0.1912, simple_loss=0.27, pruned_loss=0.05621, over 28414.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3133, pruned_loss=0.08024, over 5702717.04 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3253, pruned_loss=0.08398, over 5776785.23 frames. ], giga_tot_loss[loss=0.2372, simple_loss=0.3133, pruned_loss=0.08056, over 5686415.34 frames. ], batch size: 60, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:15:12,811 INFO [train.py:968] (0/2) Epoch 24, batch 17900, giga_loss[loss=0.2255, simple_loss=0.3006, pruned_loss=0.07525, over 28568.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3116, pruned_loss=0.07939, over 5702654.58 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3255, pruned_loss=0.08377, over 5773165.88 frames. ], giga_tot_loss[loss=0.2352, simple_loss=0.311, pruned_loss=0.07969, over 5690035.48 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:15:25,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1066811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:15:26,732 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.826e+02 1.164e+03 1.420e+03 1.981e+03 7.990e+03, threshold=2.840e+03, percent-clipped=13.0
+2023-03-12 11:15:50,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8369, 3.6708, 3.4793, 1.8930], device='cuda:0'), covar=tensor([0.0677, 0.0858, 0.0832, 0.2217], device='cuda:0'), in_proj_covar=tensor([0.1229, 0.1131, 0.0954, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 11:15:58,485 INFO [train.py:968] (0/2) Epoch 24, batch 17950, giga_loss[loss=0.272, simple_loss=0.3352, pruned_loss=0.1045, over 27959.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3084, pruned_loss=0.07816, over 5698219.66 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3262, pruned_loss=0.08408, over 5775461.91 frames. ], giga_tot_loss[loss=0.2316, simple_loss=0.3071, pruned_loss=0.07803, over 5684858.65 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:16:17,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1066871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:16:20,919 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1066875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:16:40,957 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9999, 1.1795, 1.3725, 0.9637], device='cuda:0'), covar=tensor([0.2069, 0.1664, 0.2695, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0736, 0.0708, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 11:16:41,331 INFO [train.py:968] (0/2) Epoch 24, batch 18000, giga_loss[loss=0.241, simple_loss=0.3031, pruned_loss=0.08942, over 28763.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3054, pruned_loss=0.07698, over 5709609.86 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3265, pruned_loss=0.08423, over 5776893.14 frames. ], giga_tot_loss[loss=0.2287, simple_loss=0.304, pruned_loss=0.07669, over 5697291.99 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:16:41,334 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 11:16:50,592 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.203, simple_loss=0.3098, pruned_loss=0.04807, over 944034.00 frames. 
+2023-03-12 11:16:50,593 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 11:17:03,235 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-12 11:17:03,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.868e+02 1.182e+03 1.487e+03 2.271e+03 5.176e+03, threshold=2.975e+03, percent-clipped=12.0
+2023-03-12 11:17:24,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3720, 1.6219, 1.3585, 1.5418], device='cuda:0'), covar=tensor([0.0709, 0.0407, 0.0347, 0.0785], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 11:17:34,231 INFO [train.py:968] (0/2) Epoch 24, batch 18050, giga_loss[loss=0.2048, simple_loss=0.2831, pruned_loss=0.06324, over 28004.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3024, pruned_loss=0.07549, over 5695493.39 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3265, pruned_loss=0.08416, over 5777500.07 frames. ], giga_tot_loss[loss=0.2254, simple_loss=0.3006, pruned_loss=0.07514, over 5683179.96 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:17:40,832 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1066954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:17:42,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1066957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:17:47,302 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 11:17:54,477 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 11:17:58,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7562, 5.5596, 5.2663, 2.9898], device='cuda:0'), covar=tensor([0.0366, 0.0578, 0.0660, 0.1459], device='cuda:0'), in_proj_covar=tensor([0.1230, 0.1132, 0.0955, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 11:18:09,392 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1066986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:18:20,078 INFO [train.py:968] (0/2) Epoch 24, batch 18100, giga_loss[loss=0.2018, simple_loss=0.2736, pruned_loss=0.065, over 28429.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.2991, pruned_loss=0.07423, over 5694903.60 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3265, pruned_loss=0.08415, over 5775592.59 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2974, pruned_loss=0.07384, over 5685624.84 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:18:24,074 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-12 11:18:34,403 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1067014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:18:34,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.337e+02 1.099e+03 1.369e+03 1.868e+03 4.588e+03, threshold=2.739e+03, percent-clipped=6.0
+2023-03-12 11:18:36,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1067017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:18:39,872 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1067022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:19:00,716 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1067046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:19:02,568 INFO [train.py:968] (0/2) Epoch 24, batch 18150, giga_loss[loss=0.1939, simple_loss=0.2575, pruned_loss=0.06512, over 23863.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2967, pruned_loss=0.07317, over 5677265.14 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3272, pruned_loss=0.08453, over 5757555.72 frames. ], giga_tot_loss[loss=0.2193, simple_loss=0.294, pruned_loss=0.07228, over 5682867.19 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:19:08,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5487, 1.8790, 1.5178, 1.5700], device='cuda:0'), covar=tensor([0.2930, 0.2949, 0.3543, 0.2584], device='cuda:0'), in_proj_covar=tensor([0.1546, 0.1109, 0.1365, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 11:19:11,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3893, 2.0306, 1.6229, 0.6222], device='cuda:0'), covar=tensor([0.6128, 0.3519, 0.4927, 0.7084], device='cuda:0'), in_proj_covar=tensor([0.1774, 0.1678, 0.1613, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 11:19:19,802 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-12 11:19:50,297 INFO [train.py:968] (0/2) Epoch 24, batch 18200, giga_loss[loss=0.1841, simple_loss=0.2542, pruned_loss=0.05704, over 28466.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2946, pruned_loss=0.07245, over 5672249.27 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3274, pruned_loss=0.08454, over 5757096.25 frames. ], giga_tot_loss[loss=0.2174, simple_loss=0.2918, pruned_loss=0.07152, over 5675996.64 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:20:04,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3999, 1.6371, 1.4077, 1.5959], device='cuda:0'), covar=tensor([0.0736, 0.0385, 0.0348, 0.0818], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 11:20:04,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 5.956e+02 1.120e+03 1.444e+03 1.805e+03 3.338e+03, threshold=2.887e+03, percent-clipped=6.0
+2023-03-12 11:20:34,339 INFO [train.py:968] (0/2) Epoch 24, batch 18250, giga_loss[loss=0.2725, simple_loss=0.3191, pruned_loss=0.113, over 23890.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2935, pruned_loss=0.07231, over 5678094.49 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3279, pruned_loss=0.08478, over 5759641.41 frames. ], giga_tot_loss[loss=0.2162, simple_loss=0.2901, pruned_loss=0.07112, over 5677313.88 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:20:35,313 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7214, 2.4210, 1.5073, 0.8172], device='cuda:0'), covar=tensor([0.8821, 0.6386, 0.4617, 0.8282], device='cuda:0'), in_proj_covar=tensor([0.1771, 0.1675, 0.1613, 0.1443], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 11:21:25,931 INFO [train.py:968] (0/2) Epoch 24, batch 18300, giga_loss[loss=0.2818, simple_loss=0.3539, pruned_loss=0.1048, over 28639.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3027, pruned_loss=0.07751, over 5672401.27 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3277, pruned_loss=0.08451, over 5761609.58 frames. ], giga_tot_loss[loss=0.2264, simple_loss=0.2996, pruned_loss=0.07662, over 5668530.29 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:21:40,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.727e+02 1.236e+03 1.605e+03 2.389e+03 4.868e+03, threshold=3.210e+03, percent-clipped=15.0
+2023-03-12 11:22:09,627 INFO [train.py:968] (0/2) Epoch 24, batch 18350, giga_loss[loss=0.2862, simple_loss=0.3655, pruned_loss=0.1034, over 28903.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3156, pruned_loss=0.08386, over 5684042.08 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3281, pruned_loss=0.08483, over 5766639.34 frames. ], giga_tot_loss[loss=0.2389, simple_loss=0.3122, pruned_loss=0.08275, over 5673513.48 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:22:11,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1067250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:22:47,377 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-12 11:22:50,963 INFO [train.py:968] (0/2) Epoch 24, batch 18400, giga_loss[loss=0.2898, simple_loss=0.3701, pruned_loss=0.1047, over 28583.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3275, pruned_loss=0.08964, over 5696237.07 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3285, pruned_loss=0.08494, over 5768248.98 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3244, pruned_loss=0.08874, over 5684436.36 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:23:06,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.513e+02 1.365e+03 1.741e+03 2.111e+03 5.127e+03, threshold=3.481e+03, percent-clipped=4.0
+2023-03-12 11:23:15,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2394, 1.4127, 1.2950, 1.1529], device='cuda:0'), covar=tensor([0.2599, 0.2746, 0.1927, 0.2430], device='cuda:0'), in_proj_covar=tensor([0.1982, 0.1906, 0.1830, 0.1975], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 11:23:34,455 INFO [train.py:968] (0/2) Epoch 24, batch 18450, giga_loss[loss=0.2437, simple_loss=0.3344, pruned_loss=0.07656, over 29056.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.334, pruned_loss=0.09207, over 5692768.49 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3283, pruned_loss=0.08477, over 5770333.22 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3318, pruned_loss=0.09161, over 5680526.01 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:24:02,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5541, 2.0158, 1.4868, 1.6614], device='cuda:0'), covar=tensor([0.0741, 0.0280, 0.0340, 0.0888], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 11:24:13,347 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1067393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:24:15,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1067396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:24:16,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1067397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:24:16,667 INFO [train.py:968] (0/2) Epoch 24, batch 18500, libri_loss[loss=0.2424, simple_loss=0.3308, pruned_loss=0.07697, over 29740.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3368, pruned_loss=0.09189, over 5697115.45 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3286, pruned_loss=0.0848, over 5773513.37 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3349, pruned_loss=0.09174, over 5681590.86 frames. ], batch size: 87, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:24:30,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.100e+02 1.274e+03 1.693e+03 2.726e+03 5.997e+03, threshold=3.386e+03, percent-clipped=12.0
+2023-03-12 11:24:40,025 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1067425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:25:03,501 INFO [train.py:968] (0/2) Epoch 24, batch 18550, libri_loss[loss=0.296, simple_loss=0.3704, pruned_loss=0.1108, over 26207.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3399, pruned_loss=0.09264, over 5686966.10 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3291, pruned_loss=0.08493, over 5772808.86 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3382, pruned_loss=0.09253, over 5674291.70 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:25:26,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4526, 1.6720, 1.6449, 1.5509], device='cuda:0'), covar=tensor([0.1978, 0.1824, 0.2346, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0740, 0.0710, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 11:25:47,299 INFO [train.py:968] (0/2) Epoch 24, batch 18600, libri_loss[loss=0.2515, simple_loss=0.3174, pruned_loss=0.09276, over 29673.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3418, pruned_loss=0.09418, over 5683024.31 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3292, pruned_loss=0.085, over 5773105.52 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3407, pruned_loss=0.09427, over 5669871.95 frames. ], batch size: 69, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:26:01,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.869e+02 1.182e+03 1.441e+03 1.912e+03 3.821e+03, threshold=2.881e+03, percent-clipped=1.0
+2023-03-12 11:26:13,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4763, 3.5947, 1.6378, 1.6485], device='cuda:0'), covar=tensor([0.1056, 0.0276, 0.0914, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0552, 0.0392, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 11:26:22,886 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1067540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:26:25,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1067543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:26:29,999 INFO [train.py:968] (0/2) Epoch 24, batch 18650, giga_loss[loss=0.2576, simple_loss=0.3488, pruned_loss=0.08321, over 28402.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3434, pruned_loss=0.0954, over 5686558.82 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.329, pruned_loss=0.08476, over 5777042.93 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3431, pruned_loss=0.09597, over 5670213.43 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:26:55,045 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1067572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:27:04,599 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-12 11:27:17,876 INFO [train.py:968] (0/2) Epoch 24, batch 18700, giga_loss[loss=0.2725, simple_loss=0.3549, pruned_loss=0.09504, over 28920.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3477, pruned_loss=0.0983, over 5681312.54 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.329, pruned_loss=0.08476, over 5777042.93 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3475, pruned_loss=0.09874, over 5668590.72 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:27:33,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.871e+02 1.252e+03 1.568e+03 2.004e+03 4.817e+03, threshold=3.136e+03, percent-clipped=4.0
+2023-03-12 11:28:03,331 INFO [train.py:968] (0/2) Epoch 24, batch 18750, libri_loss[loss=0.2688, simple_loss=0.3538, pruned_loss=0.09191, over 29766.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3494, pruned_loss=0.09822, over 5680138.02 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3289, pruned_loss=0.08472, over 5769547.66 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3497, pruned_loss=0.09887, over 5674012.78 frames. ], batch size: 87, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:28:20,671 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4070, 2.9946, 1.4706, 1.6051], device='cuda:0'), covar=tensor([0.1063, 0.0276, 0.0937, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0551, 0.0392, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 11:28:27,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4539, 1.5691, 1.3257, 1.4533], device='cuda:0'), covar=tensor([0.0803, 0.0335, 0.0346, 0.0944], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 11:28:46,464 INFO [train.py:968] (0/2) Epoch 24, batch 18800, giga_loss[loss=0.2718, simple_loss=0.3568, pruned_loss=0.09342, over 28876.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3525, pruned_loss=0.09947, over 5680014.50 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.329, pruned_loss=0.08462, over 5770109.91 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.353, pruned_loss=0.1003, over 5673365.82 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:28:57,810 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3924, 2.2327, 2.2670, 2.0733], device='cuda:0'), covar=tensor([0.1807, 0.2458, 0.2041, 0.2166], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0739, 0.0710, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 11:29:00,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.814e+02 1.174e+03 1.422e+03 1.781e+03 3.165e+03, threshold=2.844e+03, percent-clipped=1.0
+2023-03-12 11:29:30,414 INFO [train.py:968] (0/2) Epoch 24, batch 18850, giga_loss[loss=0.2642, simple_loss=0.3264, pruned_loss=0.101, over 23513.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.353, pruned_loss=0.09869, over 5687674.45 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3292, pruned_loss=0.08465, over 5772207.86 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3536, pruned_loss=0.09958, over 5678828.72 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:30:11,247 INFO [train.py:968] (0/2) Epoch 24, batch 18900, giga_loss[loss=0.2461, simple_loss=0.3337, pruned_loss=0.07926, over 28888.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3519, pruned_loss=0.0967, over 5702353.79 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3291, pruned_loss=0.08455, over 5775407.02 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.353, pruned_loss=0.0978, over 5690708.00 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:30:26,853 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.827e+02 1.207e+03 1.436e+03 1.996e+03 3.475e+03, threshold=2.872e+03, percent-clipped=5.0
+2023-03-12 11:30:51,914 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 11:30:52,837 INFO [train.py:968] (0/2) Epoch 24, batch 18950, giga_loss[loss=0.247, simple_loss=0.3409, pruned_loss=0.07654, over 28963.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3503, pruned_loss=0.09456, over 5709252.53 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3294, pruned_loss=0.08468, over 5777110.19 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3511, pruned_loss=0.09545, over 5697662.35 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:31:23,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6215, 1.8533, 1.7303, 1.5641], device='cuda:0'), covar=tensor([0.2131, 0.2120, 0.2310, 0.2219], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0741, 0.0712, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 11:31:31,398 INFO [train.py:968] (0/2) Epoch 24, batch 19000, giga_loss[loss=0.2775, simple_loss=0.3581, pruned_loss=0.0985, over 28679.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3489, pruned_loss=0.09363, over 5711284.48 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3298, pruned_loss=0.08477, over 5779032.53 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3495, pruned_loss=0.09439, over 5699456.28 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:31:44,329 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1067912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:31:46,027 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.619e+02 1.152e+03 1.423e+03 2.034e+03 7.593e+03, threshold=2.847e+03, percent-clipped=8.0
+2023-03-12 11:31:58,880 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-12 11:32:14,822 INFO [train.py:968] (0/2) Epoch 24, batch 19050, giga_loss[loss=0.3049, simple_loss=0.3678, pruned_loss=0.121, over 28758.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3512, pruned_loss=0.09701, over 5718036.69 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3302, pruned_loss=0.08488, over 5781326.51 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.352, pruned_loss=0.09789, over 5703973.52 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:32:31,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6419, 1.8958, 1.5980, 1.7153], device='cuda:0'), covar=tensor([0.2549, 0.2443, 0.2621, 0.2337], device='cuda:0'), in_proj_covar=tensor([0.1539, 0.1107, 0.1358, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 11:33:00,632 INFO [train.py:968] (0/2) Epoch 24, batch 19100, giga_loss[loss=0.2833, simple_loss=0.346, pruned_loss=0.1103, over 28872.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3551, pruned_loss=0.1029, over 5718365.43 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3303, pruned_loss=0.08494, over 5783386.00 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3563, pruned_loss=0.1039, over 5703816.78 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:33:01,922 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1068000.pt
+2023-03-12 11:33:15,772 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.886e+02 1.417e+03 1.761e+03 2.456e+03 7.081e+03, threshold=3.522e+03, percent-clipped=16.0
+2023-03-12 11:33:15,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1068016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:33:41,465 INFO [train.py:968] (0/2) Epoch 24, batch 19150, giga_loss[loss=0.351, simple_loss=0.3951, pruned_loss=0.1535, over 26557.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3538, pruned_loss=0.1028, over 5713700.92 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3308, pruned_loss=0.08516, over 5781813.61 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3546, pruned_loss=0.1036, over 5702826.11 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:33:56,470 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 11:34:01,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5489, 1.4484, 4.3255, 3.5378], device='cuda:0'), covar=tensor([0.1662, 0.2763, 0.0389, 0.0995], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0661, 0.0975, 0.0933], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 11:34:05,091 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 11:34:23,613 INFO [train.py:968] (0/2) Epoch 24, batch 19200, giga_loss[loss=0.2788, simple_loss=0.3483, pruned_loss=0.1046, over 28937.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.351, pruned_loss=0.1013, over 5709083.87 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3313, pruned_loss=0.08512, over 5782368.95 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3519, pruned_loss=0.1027, over 5696951.30 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:34:40,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.120e+02 1.263e+03 1.602e+03 2.199e+03 3.541e+03, threshold=3.205e+03, percent-clipped=1.0
+2023-03-12 11:34:51,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 11:35:07,233 INFO [train.py:968] (0/2) Epoch 24, batch 19250, giga_loss[loss=0.2769, simple_loss=0.3525, pruned_loss=0.1007, over 28982.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3505, pruned_loss=0.1012, over 5713989.27 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3321, pruned_loss=0.08544, over 5784688.34 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.351, pruned_loss=0.1024, over 5700517.29 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:35:25,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5166, 5.3244, 5.0895, 2.5494], device='cuda:0'), covar=tensor([0.0428, 0.0530, 0.0569, 0.1704], device='cuda:0'), in_proj_covar=tensor([0.1224, 0.1134, 0.0956, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 11:35:44,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7000, 4.5231, 4.3146, 2.1652], device='cuda:0'), covar=tensor([0.0531, 0.0627, 0.0658, 0.1811], device='cuda:0'), in_proj_covar=tensor([0.1225, 0.1134, 0.0956, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 11:35:53,832 INFO [train.py:968] (0/2) Epoch 24, batch 19300, giga_loss[loss=0.2624, simple_loss=0.3414, pruned_loss=0.09169, over 28553.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3489, pruned_loss=0.09928, over 5719201.27 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3323, pruned_loss=0.08541, over 5786198.48 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3494, pruned_loss=0.1004, over 5706171.74 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:36:08,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.694e+02 1.303e+03 1.609e+03 2.090e+03 7.590e+03, threshold=3.217e+03, percent-clipped=8.0
+2023-03-12 11:36:24,872 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.42 vs. limit=5.0
+2023-03-12 11:36:36,415 INFO [train.py:968] (0/2) Epoch 24, batch 19350, giga_loss[loss=0.2428, simple_loss=0.3225, pruned_loss=0.08158, over 28819.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3474, pruned_loss=0.09816, over 5712916.74 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3326, pruned_loss=0.08544, over 5788714.15 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3481, pruned_loss=0.0995, over 5698091.74 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:37:15,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1068287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:37:23,037 INFO [train.py:968] (0/2) Epoch 24, batch 19400, giga_loss[loss=0.2356, simple_loss=0.3161, pruned_loss=0.07758, over 29006.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3413, pruned_loss=0.09472, over 5703356.72 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3326, pruned_loss=0.08541, over 5788996.99 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3419, pruned_loss=0.09589, over 5690766.43 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:37:27,475 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1068301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:37:43,686 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.887e+02 1.155e+03 1.322e+03 1.766e+03 7.446e+03, threshold=2.644e+03, percent-clipped=2.0
+2023-03-12 11:38:09,667 INFO [train.py:968] (0/2) Epoch 24, batch 19450, giga_loss[loss=0.2439, simple_loss=0.323, pruned_loss=0.08237, over 28974.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3365, pruned_loss=0.09254, over 5678525.72 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3329, pruned_loss=0.08552, over 5778199.21 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3367, pruned_loss=0.09354, over 5676651.75 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:38:19,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1576, 2.4556, 1.9020, 1.8750], device='cuda:0'), covar=tensor([0.1077, 0.0691, 0.0978, 0.1275], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0443, 0.0519, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 11:38:51,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1068391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:38:56,215 INFO [train.py:968] (0/2) Epoch 24, batch 19500, giga_loss[loss=0.2612, simple_loss=0.3348, pruned_loss=0.09379, over 28837.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3316, pruned_loss=0.09027, over 5667259.04 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3332, pruned_loss=0.08558, over 5777496.58 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3316, pruned_loss=0.09114, over 5664087.90 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:39:15,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.402e+02 9.947e+02 1.251e+03 1.771e+03 5.079e+03, threshold=2.502e+03, percent-clipped=7.0
+2023-03-12 11:39:28,247 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1068430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:39:32,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1068433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:39:44,674 INFO [train.py:968] (0/2) Epoch 24, batch 19550, giga_loss[loss=0.2672, simple_loss=0.3472, pruned_loss=0.09355, over 28872.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.33, pruned_loss=0.08917, over 5658104.30 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3331, pruned_loss=0.08549, over 5779719.64 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.33, pruned_loss=0.09004, over 5651261.66 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:39:55,278 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1068462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:40:25,600 INFO [train.py:968] (0/2) Epoch 24, batch 19600, giga_loss[loss=0.2492, simple_loss=0.3309, pruned_loss=0.08377, over 28944.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3309, pruned_loss=0.08944, over 5665265.48 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3335, pruned_loss=0.08554, over 5776876.81 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3304, pruned_loss=0.09021, over 5658164.38 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:40:43,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.681e+02 1.176e+03 1.566e+03 2.469e+03 9.977e+03, threshold=3.132e+03, percent-clipped=23.0
+2023-03-12 11:40:54,376 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1068534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:40:57,423 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1068537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:41:06,330 INFO [train.py:968] (0/2) Epoch 24, batch 19650, giga_loss[loss=0.2133, simple_loss=0.2931, pruned_loss=0.06674, over 28509.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3307, pruned_loss=0.08924, over 5670301.24 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3338, pruned_loss=0.08559, over 5773037.08 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.33, pruned_loss=0.08999, over 5663192.38 frames. ], batch size: 60, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:41:15,164 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-12 11:41:16,152 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1068561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:41:20,985 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1068566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:41:45,037 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 11:41:46,072 INFO [train.py:968] (0/2) Epoch 24, batch 19700, libri_loss[loss=0.2241, simple_loss=0.3015, pruned_loss=0.07334, over 29518.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3302, pruned_loss=0.0888, over 5681952.78 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3344, pruned_loss=0.08579, over 5776475.19 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.329, pruned_loss=0.08934, over 5669924.79 frames. ], batch size: 70, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:42:01,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.483e+02 1.130e+03 1.434e+03 1.828e+03 6.160e+03, threshold=2.869e+03, percent-clipped=7.0
+2023-03-12 11:42:25,425 INFO [train.py:968] (0/2) Epoch 24, batch 19750, giga_loss[loss=0.2619, simple_loss=0.3339, pruned_loss=0.09501, over 28962.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3281, pruned_loss=0.08768, over 5679869.29 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3347, pruned_loss=0.0858, over 5766411.90 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3267, pruned_loss=0.08816, over 5677713.41 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:42:41,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3161, 1.3025, 3.7903, 3.0614], device='cuda:0'), covar=tensor([0.1697, 0.2846, 0.0444, 0.1053], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0656, 0.0967, 0.0927], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 11:42:49,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5696, 1.8259, 1.4751, 1.6985], device='cuda:0'), covar=tensor([0.2594, 0.2678, 0.2916, 0.2448], device='cuda:0'), in_proj_covar=tensor([0.1540, 0.1108, 0.1358, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 11:42:50,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-12 11:42:51,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1068676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:43:08,670 INFO [train.py:968] (0/2) Epoch 24, batch 19800, giga_loss[loss=0.2423, simple_loss=0.3163, pruned_loss=0.08419, over 28992.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.325, pruned_loss=0.08619, over 5691684.69 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3348, pruned_loss=0.08577, over 5767823.08 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3238, pruned_loss=0.0866, over 5688097.68 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:43:24,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.318e+02 1.060e+03 1.345e+03 1.789e+03 8.616e+03, threshold=2.690e+03, percent-clipped=10.0
+2023-03-12 11:43:50,225 INFO [train.py:968] (0/2) Epoch 24, batch 19850, giga_loss[loss=0.2464, simple_loss=0.3188, pruned_loss=0.08705, over 28810.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3236, pruned_loss=0.0854, over 5696007.81 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3353, pruned_loss=0.08568, over 5769625.54 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3219, pruned_loss=0.08581, over 5689521.28 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:44:33,184 INFO [train.py:968] (0/2) Epoch 24, batch 19900, giga_loss[loss=0.2337, simple_loss=0.3054, pruned_loss=0.08103, over 28900.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3208, pruned_loss=0.0842, over 5708291.81 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3355, pruned_loss=0.08567, over 5770874.00 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.319, pruned_loss=0.08452, over 5700817.04 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:44:35,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2276, 4.0597, 3.8457, 1.8381], device='cuda:0'), covar=tensor([0.0612, 0.0728, 0.0670, 0.2177], device='cuda:0'), in_proj_covar=tensor([0.1225, 0.1129, 0.0955, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 11:44:49,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.339e+02 1.111e+03 1.364e+03 1.987e+03 5.485e+03, threshold=2.728e+03, percent-clipped=9.0
+2023-03-12 11:44:50,136 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1068819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:44:52,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1068822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:45:14,012 INFO [train.py:968] (0/2) Epoch 24, batch 19950, giga_loss[loss=0.2331, simple_loss=0.3095, pruned_loss=0.07839, over 28952.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.319, pruned_loss=0.08359, over 5709852.62 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3361, pruned_loss=0.08589, over 5763682.46 frames. ], giga_tot_loss[loss=0.242, simple_loss=0.3168, pruned_loss=0.08364, over 5708663.49 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:45:17,696 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1068851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:45:55,749 INFO [train.py:968] (0/2) Epoch 24, batch 20000, libri_loss[loss=0.2296, simple_loss=0.3229, pruned_loss=0.06814, over 29655.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3188, pruned_loss=0.08376, over 5708438.35 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3366, pruned_loss=0.08606, over 5764101.94 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3161, pruned_loss=0.0836, over 5705346.69 frames. ], batch size: 73, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:46:11,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.504e+02 1.175e+03 1.637e+03 2.392e+03 1.080e+04, threshold=3.275e+03, percent-clipped=15.0
+2023-03-12 11:46:25,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1068936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:46:25,364 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 11:46:34,591 INFO [train.py:968] (0/2) Epoch 24, batch 20050, giga_loss[loss=0.2677, simple_loss=0.3406, pruned_loss=0.09739, over 27653.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3184, pruned_loss=0.08387, over 5709522.61 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3374, pruned_loss=0.08645, over 5764813.40 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3153, pruned_loss=0.08338, over 5705665.80 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:46:38,563 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7851, 1.9388, 1.4724, 1.4284], device='cuda:0'), covar=tensor([0.1062, 0.0671, 0.1062, 0.1218], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0447, 0.0522, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 11:47:12,653 INFO [train.py:968] (0/2) Epoch 24, batch 20100, giga_loss[loss=0.2181, simple_loss=0.2861, pruned_loss=0.07507, over 28616.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3162, pruned_loss=0.08235, over 5719780.61 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3374, pruned_loss=0.0863, over 5766624.57 frames. ], giga_tot_loss[loss=0.2388, simple_loss=0.3135, pruned_loss=0.08204, over 5714490.83 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:47:29,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.399e+02 1.073e+03 1.382e+03 1.668e+03 4.805e+03, threshold=2.764e+03, percent-clipped=4.0
+2023-03-12 11:47:30,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2146, 1.4196, 1.3136, 1.1098], device='cuda:0'), covar=tensor([0.2952, 0.2859, 0.1958, 0.2834], device='cuda:0'), in_proj_covar=tensor([0.2003, 0.1928, 0.1848, 0.2005], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 11:47:52,428 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 11:47:57,579 INFO [train.py:968] (0/2) Epoch 24, batch 20150, giga_loss[loss=0.3472, simple_loss=0.3905, pruned_loss=0.152, over 23817.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3214, pruned_loss=0.08557, over 5714040.11 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.338, pruned_loss=0.08655, over 5769474.76 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3183, pruned_loss=0.08506, over 5706565.08 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:48:26,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1069079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:48:29,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1069082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:48:37,605 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1069088.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 11:48:45,988 INFO [train.py:968] (0/2) Epoch 24, batch 20200, giga_loss[loss=0.2568, simple_loss=0.3292, pruned_loss=0.09217, over 28632.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3283, pruned_loss=0.09023, over 5705191.60 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3386, pruned_loss=0.08684, over 5770892.44 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3252, pruned_loss=0.08959, over 5697575.75 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:48:58,715 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1069111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:49:06,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2523, 1.5102, 1.5152, 1.3216], device='cuda:0'), covar=tensor([0.2108, 0.1741, 0.2463, 0.1936], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0745, 0.0716, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 11:49:08,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.552e+02 1.271e+03 1.592e+03 2.114e+03 4.655e+03, threshold=3.185e+03, percent-clipped=14.0
+2023-03-12 11:49:10,749 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1069122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:49:37,360 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4876, 1.1296, 3.9150, 3.3757], device='cuda:0'), covar=tensor([0.1523, 0.2813, 0.0453, 0.0975], device='cuda:0'), in_proj_covar=tensor([0.0770, 0.0651, 0.0965, 0.0923], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 11:49:37,776 INFO [train.py:968] (0/2) Epoch 24, batch 20250, giga_loss[loss=0.3224, simple_loss=0.3858, pruned_loss=0.1294, over 28893.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3355, pruned_loss=0.09464, over 5702307.24 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3385, pruned_loss=0.08668, over 5770192.02 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.333, pruned_loss=0.09443, over 5695357.14 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:50:23,933 INFO [train.py:968] (0/2) Epoch 24, batch 20300, giga_loss[loss=0.2506, simple_loss=0.3276, pruned_loss=0.08674, over 28868.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3406, pruned_loss=0.09732, over 5686107.47 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3386, pruned_loss=0.08678, over 5758798.45 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3385, pruned_loss=0.09722, over 5689610.39 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:50:44,130 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.823e+02 1.298e+03 1.557e+03 2.264e+03 5.891e+03, threshold=3.114e+03, percent-clipped=11.0
+2023-03-12 11:51:10,959 INFO [train.py:968] (0/2) Epoch 24, batch 20350, giga_loss[loss=0.3167, simple_loss=0.382, pruned_loss=0.1257, over 28655.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3464, pruned_loss=0.09964, over 5685820.34 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3392, pruned_loss=0.08698, over 5762542.29 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3444, pruned_loss=0.09972, over 5683215.51 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:51:12,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7866, 3.5791, 3.4232, 1.9410], device='cuda:0'), covar=tensor([0.0757, 0.0913, 0.0804, 0.2124], device='cuda:0'), in_proj_covar=tensor([0.1228, 0.1131, 0.0956, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 11:51:23,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6741, 4.3824, 1.7494, 1.8951], device='cuda:0'), covar=tensor([0.1007, 0.0224, 0.0939, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0553, 0.0393, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 11:51:56,099 INFO [train.py:968] (0/2) Epoch 24, batch 20400, giga_loss[loss=0.3325, simple_loss=0.3954, pruned_loss=0.1348, over 28922.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3505, pruned_loss=0.1012, over 5684248.90 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3393, pruned_loss=0.0872, over 5761644.81 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.349, pruned_loss=0.1015, over 5680652.39 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:51:57,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2788, 1.5883, 1.5669, 1.1539], device='cuda:0'), covar=tensor([0.1562, 0.2348, 0.1311, 0.1575], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0706, 0.0965, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 11:52:05,300 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1069306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:52:16,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.644e+02 1.280e+03 1.564e+03 2.061e+03 4.267e+03, threshold=3.128e+03, percent-clipped=7.0
+2023-03-12 11:52:43,657 INFO [train.py:968] (0/2) Epoch 24, batch 20450, giga_loss[loss=0.2701, simple_loss=0.3502, pruned_loss=0.09504, over 28778.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3547, pruned_loss=0.1036, over 5685172.09 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3395, pruned_loss=0.08731, over 5753469.74 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3535, pruned_loss=0.1038, over 5688289.82 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:52:49,607 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1069355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:53:08,367 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5693, 4.4038, 4.1503, 1.8610], device='cuda:0'), covar=tensor([0.0520, 0.0654, 0.0674, 0.2205], device='cuda:0'), in_proj_covar=tensor([0.1227, 0.1131, 0.0955, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 11:53:27,646 INFO [train.py:968] (0/2) Epoch 24, batch 20500, libri_loss[loss=0.2312, simple_loss=0.3115, pruned_loss=0.07544, over 29564.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3501, pruned_loss=0.1002, over 5689000.16 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3395, pruned_loss=0.08747, over 5758056.53 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3495, pruned_loss=0.1007, over 5685312.82 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:53:45,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.459e+02 1.302e+03 1.612e+03 2.324e+03 5.955e+03, threshold=3.224e+03, percent-clipped=12.0
+2023-03-12 11:54:08,847 INFO [train.py:968] (0/2) Epoch 24, batch 20550, giga_loss[loss=0.2176, simple_loss=0.3104, pruned_loss=0.06238, over 29040.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3475, pruned_loss=0.09787, over 5697950.43 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3394, pruned_loss=0.08759, over 5759760.05 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3473, pruned_loss=0.09833, over 5692425.17 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:54:22,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1069463.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 11:54:48,644 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1069494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:54:51,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1069497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:54:52,121 INFO [train.py:968] (0/2) Epoch 24, batch 20600, giga_loss[loss=0.231, simple_loss=0.3199, pruned_loss=0.07101, over 28920.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.347, pruned_loss=0.097, over 5694033.11 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3397, pruned_loss=0.08768, over 5763615.82 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3468, pruned_loss=0.09752, over 5684665.50 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:55:10,621 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.528e+02 1.460e+03 1.816e+03 2.811e+03 8.144e+03, threshold=3.632e+03, percent-clipped=19.0
+2023-03-12 11:55:33,073 INFO [train.py:968] (0/2) Epoch 24, batch 20650, giga_loss[loss=0.3019, simple_loss=0.3716, pruned_loss=0.1161, over 29011.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3483, pruned_loss=0.09758, over 5701410.48 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3395, pruned_loss=0.08771, over 5767383.55 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3486, pruned_loss=0.09831, over 5688203.69 frames. ], batch size: 128, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:56:15,759 INFO [train.py:968] (0/2) Epoch 24, batch 20700, giga_loss[loss=0.2644, simple_loss=0.3432, pruned_loss=0.09282, over 29032.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3513, pruned_loss=0.09995, over 5701991.08 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3395, pruned_loss=0.08759, over 5769858.20 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3517, pruned_loss=0.1009, over 5687964.39 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:56:24,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1069606.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 11:56:26,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1069609.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 11:56:30,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4321, 1.7757, 1.5425, 1.4798], device='cuda:0'), covar=tensor([0.0660, 0.0287, 0.0272, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 11:56:37,918 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.217e+02 1.262e+03 1.534e+03 2.009e+03 3.617e+03, threshold=3.069e+03, percent-clipped=0.0
+2023-03-12 11:56:51,931 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1069638.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 11:56:53,217 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1069640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:56:55,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1069643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:56:58,947 INFO [train.py:968] (0/2) Epoch 24, batch 20750, giga_loss[loss=0.2983, simple_loss=0.3752, pruned_loss=0.1107, over 28970.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3537, pruned_loss=0.1017, over 5701079.01 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.34, pruned_loss=0.08796, over 5767508.42 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3539, pruned_loss=0.1026, over 5689664.43 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:57:23,402 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1069672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:57:30,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1069681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:57:49,130 INFO [train.py:968] (0/2) Epoch 24, batch 20800, giga_loss[loss=0.301, simple_loss=0.3655, pruned_loss=0.1183, over 28667.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.354, pruned_loss=0.1021, over 5713074.24 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3406, pruned_loss=0.0884, over 5770089.25 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.354, pruned_loss=0.1027, over 5700625.32 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:58:09,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.455e+02 1.420e+03 1.785e+03 2.371e+03 4.878e+03, threshold=3.570e+03, percent-clipped=11.0
+2023-03-12 11:58:18,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1069730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:58:33,479 INFO [train.py:968] (0/2) Epoch 24, batch 20850, giga_loss[loss=0.2831, simple_loss=0.3563, pruned_loss=0.1049, over 29057.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3551, pruned_loss=0.1036, over 5709515.57 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3404, pruned_loss=0.08834, over 5769644.76 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3556, pruned_loss=0.1045, over 5698474.22 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:58:36,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3578, 1.4758, 1.4294, 1.2563], device='cuda:0'), covar=tensor([0.2437, 0.2311, 0.2123, 0.2494], device='cuda:0'), in_proj_covar=tensor([0.2010, 0.1941, 0.1865, 0.2017], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 11:58:57,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4282, 1.5310, 1.5416, 1.3733], device='cuda:0'), covar=tensor([0.2614, 0.2868, 0.2073, 0.2483], device='cuda:0'), in_proj_covar=tensor([0.2011, 0.1941, 0.1866, 0.2017], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 11:58:59,215 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 11:59:13,452 INFO [train.py:968] (0/2) Epoch 24, batch 20900, giga_loss[loss=0.2695, simple_loss=0.3487, pruned_loss=0.09511, over 28832.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3562, pruned_loss=0.1044, over 5713758.78 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3408, pruned_loss=0.08877, over 5769959.44 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3566, pruned_loss=0.1051, over 5703140.34 frames. ], batch size: 66, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:59:21,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6513, 1.7232, 1.8792, 1.4438], device='cuda:0'), covar=tensor([0.1747, 0.2499, 0.1380, 0.1640], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0705, 0.0963, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 11:59:31,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.108e+02 1.325e+03 1.761e+03 2.197e+03 6.112e+03, threshold=3.522e+03, percent-clipped=6.0
+2023-03-12 11:59:34,188 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1069824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:59:36,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1069827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:59:55,084 INFO [train.py:968] (0/2) Epoch 24, batch 20950, giga_loss[loss=0.2651, simple_loss=0.3442, pruned_loss=0.09302, over 28614.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3557, pruned_loss=0.103, over 5712333.01 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3412, pruned_loss=0.08907, over 5771978.10 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3558, pruned_loss=0.1035, over 5701289.63 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:00:01,586 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1069856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:00:12,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1069869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:00:16,025 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1069873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:00:18,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1069876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:00:18,948 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-12 12:00:35,786 INFO [train.py:968] (0/2) Epoch 24, batch 21000, giga_loss[loss=0.2765, simple_loss=0.3543, pruned_loss=0.09939, over 28623.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3573, pruned_loss=0.1032, over 5720680.42 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3421, pruned_loss=0.08969, over 5775681.86 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.357, pruned_loss=0.1034, over 5707164.05 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:00:35,791 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 12:00:44,740 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2059, simple_loss=0.314, pruned_loss=0.04892, over 944034.00 frames. 
+2023-03-12 12:00:44,741 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 12:00:51,126 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1069905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:01:04,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.597e+02 1.236e+03 1.723e+03 2.279e+03 9.411e+03, threshold=3.446e+03, percent-clipped=10.0
+2023-03-12 12:01:14,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3893, 3.3580, 1.4777, 1.4690], device='cuda:0'), covar=tensor([0.1029, 0.0273, 0.0956, 0.1396], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0551, 0.0391, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 12:01:26,415 INFO [train.py:968] (0/2) Epoch 24, batch 21050, giga_loss[loss=0.3034, simple_loss=0.3756, pruned_loss=0.1156, over 28287.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3559, pruned_loss=0.1021, over 5716907.57 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3419, pruned_loss=0.08974, over 5769340.41 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3561, pruned_loss=0.1025, over 5711385.14 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:01:39,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5402, 1.6939, 1.7843, 1.3673], device='cuda:0'), covar=tensor([0.1749, 0.2562, 0.1404, 0.1673], device='cuda:0'), in_proj_covar=tensor([0.0916, 0.0705, 0.0962, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 12:01:45,576 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 12:01:54,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2790, 1.4018, 1.4175, 1.2813], device='cuda:0'), covar=tensor([0.2408, 0.2702, 0.1714, 0.2160], device='cuda:0'), in_proj_covar=tensor([0.2000, 0.1932, 0.1855, 0.2009], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 12:02:05,375 INFO [train.py:968] (0/2) Epoch 24, batch 21100, giga_loss[loss=0.2515, simple_loss=0.3284, pruned_loss=0.08726, over 28712.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3523, pruned_loss=0.1, over 5715328.64 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.342, pruned_loss=0.08987, over 5772458.09 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3526, pruned_loss=0.1004, over 5706999.50 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:02:06,848 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1070000.pt
+2023-03-12 12:02:15,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1070012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:02:17,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1070015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:02:17,809 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1070016.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 12:02:21,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.026e+02 1.181e+03 1.639e+03 2.452e+03 9.312e+03, threshold=3.278e+03, percent-clipped=11.0
+2023-03-12 12:02:40,752 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1070044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:02:43,612 INFO [train.py:968] (0/2) Epoch 24, batch 21150, giga_loss[loss=0.2566, simple_loss=0.3398, pruned_loss=0.08664, over 28693.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3497, pruned_loss=0.09869, over 5718974.36 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3423, pruned_loss=0.09026, over 5775551.49 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3501, pruned_loss=0.09897, over 5707979.63 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:03:22,747 INFO [train.py:968] (0/2) Epoch 24, batch 21200, giga_loss[loss=0.2791, simple_loss=0.3537, pruned_loss=0.1023, over 29004.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3484, pruned_loss=0.09793, over 5716194.98 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3432, pruned_loss=0.091, over 5769472.17 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.348, pruned_loss=0.09766, over 5711248.66 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 12:03:28,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7058, 5.4865, 5.1816, 2.8730], device='cuda:0'), covar=tensor([0.0451, 0.0641, 0.0782, 0.1689], device='cuda:0'), in_proj_covar=tensor([0.1234, 0.1140, 0.0964, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 12:03:42,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.225e+02 1.052e+03 1.346e+03 1.688e+03 6.948e+03, threshold=2.693e+03, percent-clipped=4.0
+2023-03-12 12:04:05,962 INFO [train.py:968] (0/2) Epoch 24, batch 21250, giga_loss[loss=0.2825, simple_loss=0.3581, pruned_loss=0.1034, over 28856.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3496, pruned_loss=0.09939, over 5715797.53 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3432, pruned_loss=0.09107, over 5770819.84 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3493, pruned_loss=0.09919, over 5710306.22 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 12:04:08,348 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1070151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:04:43,003 INFO [train.py:968] (0/2) Epoch 24, batch 21300, giga_loss[loss=0.2484, simple_loss=0.3319, pruned_loss=0.0825, over 29104.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3494, pruned_loss=0.09949, over 5714994.61 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3435, pruned_loss=0.09178, over 5774873.04 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3491, pruned_loss=0.09902, over 5704773.38 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:04:48,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2274, 1.6899, 1.2785, 0.4773], device='cuda:0'), covar=tensor([0.4197, 0.2248, 0.3388, 0.6021], device='cuda:0'), in_proj_covar=tensor([0.1762, 0.1660, 0.1607, 0.1432], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 12:05:03,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.373e+02 1.245e+03 1.515e+03 1.927e+03 5.609e+03, threshold=3.031e+03, percent-clipped=12.0
+2023-03-12 12:05:26,079 INFO [train.py:968] (0/2) Epoch 24, batch 21350, giga_loss[loss=0.2704, simple_loss=0.3406, pruned_loss=0.1002, over 28671.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3489, pruned_loss=0.0984, over 5722582.44 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3438, pruned_loss=0.09211, over 5778856.11 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3486, pruned_loss=0.09795, over 5709543.34 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:05:36,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4034, 1.9296, 1.3943, 1.6914], device='cuda:0'), covar=tensor([0.0835, 0.0280, 0.0346, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 12:06:04,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9883, 1.1255, 3.4075, 2.8644], device='cuda:0'), covar=tensor([0.1828, 0.2890, 0.0486, 0.1058], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0656, 0.0971, 0.0930], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 12:06:07,322 INFO [train.py:968] (0/2) Epoch 24, batch 21400, giga_loss[loss=0.2565, simple_loss=0.3386, pruned_loss=0.08721, over 28967.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3483, pruned_loss=0.09769, over 5717174.74 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3441, pruned_loss=0.09238, over 5780490.45 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3479, pruned_loss=0.09723, over 5704209.76 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:06:26,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.928e+02 1.093e+03 1.285e+03 1.759e+03 3.160e+03, threshold=2.571e+03, percent-clipped=1.0
+2023-03-12 12:06:46,748 INFO [train.py:968] (0/2) Epoch 24, batch 21450, giga_loss[loss=0.2497, simple_loss=0.3274, pruned_loss=0.08599, over 28773.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3477, pruned_loss=0.09799, over 5704781.15 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3442, pruned_loss=0.09251, over 5771772.87 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3473, pruned_loss=0.09756, over 5701870.82 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:07:22,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1070391.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 12:07:26,980 INFO [train.py:968] (0/2) Epoch 24, batch 21500, libri_loss[loss=0.3269, simple_loss=0.3737, pruned_loss=0.1401, over 29657.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.348, pruned_loss=0.09891, over 5706935.13 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3452, pruned_loss=0.09336, over 5776524.10 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3469, pruned_loss=0.09803, over 5696742.99 frames. ], batch size: 73, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 12:07:50,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.208e+02 1.129e+03 1.450e+03 2.346e+03 1.159e+04, threshold=2.900e+03, percent-clipped=21.0
+2023-03-12 12:08:01,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5301, 1.8353, 1.4289, 1.7185], device='cuda:0'), covar=tensor([0.2811, 0.2784, 0.3190, 0.2427], device='cuda:0'), in_proj_covar=tensor([0.1541, 0.1111, 0.1357, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 12:08:10,113 INFO [train.py:968] (0/2) Epoch 24, batch 21550, giga_loss[loss=0.228, simple_loss=0.306, pruned_loss=0.07499, over 28686.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3442, pruned_loss=0.09707, over 5697726.45 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3458, pruned_loss=0.09406, over 5768638.45 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3427, pruned_loss=0.09578, over 5695951.92 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 12:08:47,564 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 12:08:49,465 INFO [train.py:968] (0/2) Epoch 24, batch 21600, giga_loss[loss=0.2583, simple_loss=0.3441, pruned_loss=0.08626, over 28607.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3428, pruned_loss=0.09662, over 5689504.48 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3466, pruned_loss=0.0948, over 5765675.80 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3408, pruned_loss=0.09498, over 5688836.75 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:09:06,412 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-12 12:09:07,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.766e+02 1.199e+03 1.537e+03 1.963e+03 6.336e+03, threshold=3.073e+03, percent-clipped=6.0
+2023-03-12 12:09:11,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1070526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:09:17,923 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1070534.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 12:09:20,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1070537.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 12:09:30,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6070, 4.4702, 4.2586, 1.7871], device='cuda:0'), covar=tensor([0.0529, 0.0665, 0.0732, 0.2125], device='cuda:0'), in_proj_covar=tensor([0.1234, 0.1143, 0.0965, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 12:09:30,523 INFO [train.py:968] (0/2) Epoch 24, batch 21650, giga_loss[loss=0.2515, simple_loss=0.3315, pruned_loss=0.08576, over 28982.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3416, pruned_loss=0.09611, over 5692960.83 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3474, pruned_loss=0.09532, over 5764078.61 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3394, pruned_loss=0.09438, over 5692803.72 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:09:46,286 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1070566.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 12:10:03,322 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5565, 1.7195, 1.6860, 1.4578], device='cuda:0'), covar=tensor([0.2019, 0.2269, 0.2317, 0.2505], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0747, 0.0716, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 12:10:04,561 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1070588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:10:13,500 INFO [train.py:968] (0/2) Epoch 24, batch 21700, libri_loss[loss=0.2732, simple_loss=0.3441, pruned_loss=0.1011, over 29658.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3389, pruned_loss=0.0949, over 5692505.58 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3475, pruned_loss=0.09548, over 5763138.84 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3367, pruned_loss=0.09335, over 5691059.57 frames. ], batch size: 73, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:10:35,806 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.964e+02 1.262e+03 1.634e+03 2.295e+03 8.621e+03, threshold=3.268e+03, percent-clipped=12.0
+2023-03-12 12:10:54,794 INFO [train.py:968] (0/2) Epoch 24, batch 21750, giga_loss[loss=0.2162, simple_loss=0.2912, pruned_loss=0.07055, over 28706.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.336, pruned_loss=0.09332, over 5702628.97 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3479, pruned_loss=0.09572, over 5766770.38 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3336, pruned_loss=0.0918, over 5696529.77 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:11:06,611 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-12 12:11:11,556 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1070669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:11:13,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1070672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:11:34,256 INFO [train.py:968] (0/2) Epoch 24, batch 21800, giga_loss[loss=0.2404, simple_loss=0.3201, pruned_loss=0.08033, over 28816.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3347, pruned_loss=0.09291, over 5712216.69 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3478, pruned_loss=0.09582, over 5770177.34 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3325, pruned_loss=0.09155, over 5702313.38 frames. ], batch size: 243, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:11:37,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1070701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:11:54,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.716e+02 1.155e+03 1.438e+03 1.908e+03 5.400e+03, threshold=2.876e+03, percent-clipped=4.0
+2023-03-12 12:12:12,956 INFO [train.py:968] (0/2) Epoch 24, batch 21850, giga_loss[loss=0.2435, simple_loss=0.3201, pruned_loss=0.08347, over 28855.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3324, pruned_loss=0.09189, over 5719047.52 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3482, pruned_loss=0.09622, over 5770769.25 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3301, pruned_loss=0.09037, over 5709683.72 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:12:46,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1985, 1.4844, 1.4774, 1.0939], device='cuda:0'), covar=tensor([0.1766, 0.2507, 0.1494, 0.1738], device='cuda:0'), in_proj_covar=tensor([0.0916, 0.0705, 0.0963, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 12:12:55,903 INFO [train.py:968] (0/2) Epoch 24, batch 21900, libri_loss[loss=0.3101, simple_loss=0.3777, pruned_loss=0.1213, over 28646.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3339, pruned_loss=0.09249, over 5698719.03 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3483, pruned_loss=0.09654, over 5754399.35 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3314, pruned_loss=0.09082, over 5705378.19 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:13:18,865 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.861e+02 1.178e+03 1.478e+03 2.126e+03 6.477e+03, threshold=2.956e+03, percent-clipped=8.0
+2023-03-12 12:13:39,363 INFO [train.py:968] (0/2) Epoch 24, batch 21950, giga_loss[loss=0.2521, simple_loss=0.3388, pruned_loss=0.08264, over 28261.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3363, pruned_loss=0.09309, over 5700115.17 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.348, pruned_loss=0.09646, over 5755896.05 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3344, pruned_loss=0.09177, over 5703112.91 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:14:05,235 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-12 12:14:25,015 INFO [train.py:968] (0/2) Epoch 24, batch 22000, giga_loss[loss=0.2488, simple_loss=0.3356, pruned_loss=0.08097, over 28953.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3409, pruned_loss=0.09515, over 5691049.05 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3482, pruned_loss=0.09674, over 5757868.75 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3389, pruned_loss=0.09378, over 5690080.86 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 12:14:35,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4864, 1.7066, 1.2588, 1.2871], device='cuda:0'), covar=tensor([0.1085, 0.0665, 0.1122, 0.1246], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0446, 0.0520, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 12:14:45,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.833e+02 1.112e+03 1.399e+03 1.724e+03 7.517e+03, threshold=2.798e+03, percent-clipped=3.0
+2023-03-12 12:15:07,672 INFO [train.py:968] (0/2) Epoch 24, batch 22050, giga_loss[loss=0.3402, simple_loss=0.391, pruned_loss=0.1447, over 26700.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3419, pruned_loss=0.09488, over 5700541.06 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3482, pruned_loss=0.09682, over 5759734.38 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3402, pruned_loss=0.09368, over 5696846.00 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 12:15:22,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1070963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:15:52,868 INFO [train.py:968] (0/2) Epoch 24, batch 22100, giga_loss[loss=0.2435, simple_loss=0.3333, pruned_loss=0.07687, over 28882.00 frames. ], tot_loss[loss=0.264, simple_loss=0.341, pruned_loss=0.09355, over 5705733.55 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3482, pruned_loss=0.09689, over 5760758.41 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3395, pruned_loss=0.09251, over 5701051.00 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 12:16:15,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.411e+02 1.137e+03 1.358e+03 1.583e+03 3.738e+03, threshold=2.715e+03, percent-clipped=3.0
+2023-03-12 12:16:35,457 INFO [train.py:968] (0/2) Epoch 24, batch 22150, giga_loss[loss=0.3426, simple_loss=0.3966, pruned_loss=0.1443, over 26802.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3407, pruned_loss=0.09372, over 5685973.43 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3484, pruned_loss=0.09723, over 5743956.06 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3392, pruned_loss=0.0925, over 5695501.65 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:17:20,372 INFO [train.py:968] (0/2) Epoch 24, batch 22200, giga_loss[loss=0.2741, simple_loss=0.3536, pruned_loss=0.09731, over 28879.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3412, pruned_loss=0.09428, over 5691071.46 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3487, pruned_loss=0.09738, over 5745684.50 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3397, pruned_loss=0.09314, over 5696431.03 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:17:27,188 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1071106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:17:28,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1071109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:17:41,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.710e+02 1.256e+03 1.549e+03 2.171e+03 4.625e+03, threshold=3.098e+03, percent-clipped=11.0
+2023-03-12 12:17:53,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1071138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:18:01,162 INFO [train.py:968] (0/2) Epoch 24, batch 22250, giga_loss[loss=0.2533, simple_loss=0.3289, pruned_loss=0.08884, over 28801.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3422, pruned_loss=0.09536, over 5676811.12 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3492, pruned_loss=0.09768, over 5730678.49 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3405, pruned_loss=0.09413, over 5694246.46 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:18:43,686 INFO [train.py:968] (0/2) Epoch 24, batch 22300, giga_loss[loss=0.2587, simple_loss=0.3388, pruned_loss=0.08931, over 28820.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.346, pruned_loss=0.09744, over 5692074.03 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3503, pruned_loss=0.09868, over 5734810.50 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3434, pruned_loss=0.09555, over 5701235.58 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:19:05,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.317e+02 1.293e+03 1.503e+03 2.212e+03 4.560e+03, threshold=3.006e+03, percent-clipped=4.0
+2023-03-12 12:19:09,463 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1071230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:19:20,158 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 12:19:24,462 INFO [train.py:968] (0/2) Epoch 24, batch 22350, giga_loss[loss=0.285, simple_loss=0.3628, pruned_loss=0.1036, over 28709.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3488, pruned_loss=0.09897, over 5698464.55 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3501, pruned_loss=0.09865, over 5738377.81 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3469, pruned_loss=0.09749, over 5701476.19 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:20:05,972 INFO [train.py:968] (0/2) Epoch 24, batch 22400, giga_loss[loss=0.2985, simple_loss=0.3639, pruned_loss=0.1165, over 28891.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.09964, over 5706897.12 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3508, pruned_loss=0.09908, over 5738464.01 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.348, pruned_loss=0.09808, over 5708238.47 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:20:27,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.302e+02 1.316e+03 1.785e+03 2.530e+03 6.256e+03, threshold=3.571e+03, percent-clipped=12.0
+2023-03-12 12:20:49,724 INFO [train.py:968] (0/2) Epoch 24, batch 22450, libri_loss[loss=0.427, simple_loss=0.4685, pruned_loss=0.1927, over 29669.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3503, pruned_loss=0.09959, over 5710608.98 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3515, pruned_loss=0.09978, over 5732960.22 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3479, pruned_loss=0.09768, over 5715727.12 frames. ], batch size: 91, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:21:08,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5227, 1.5676, 1.7367, 1.3422], device='cuda:0'), covar=tensor([0.1630, 0.2338, 0.1316, 0.1666], device='cuda:0'), in_proj_covar=tensor([0.0914, 0.0704, 0.0960, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 12:21:20,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4790, 3.3044, 3.1098, 2.1215], device='cuda:0'), covar=tensor([0.0744, 0.0905, 0.0835, 0.1674], device='cuda:0'), in_proj_covar=tensor([0.1238, 0.1144, 0.0967, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 12:21:32,967 INFO [train.py:968] (0/2) Epoch 24, batch 22500, giga_loss[loss=0.2784, simple_loss=0.3611, pruned_loss=0.09789, over 29014.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3505, pruned_loss=0.09942, over 5709076.55 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3516, pruned_loss=0.09996, over 5736122.93 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3484, pruned_loss=0.09772, over 5709833.22 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:21:52,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.035e+02 1.327e+03 1.645e+03 2.157e+03 5.060e+03, threshold=3.290e+03, percent-clipped=4.0
+2023-03-12 12:22:16,326 INFO [train.py:968] (0/2) Epoch 24, batch 22550, giga_loss[loss=0.3732, simple_loss=0.4088, pruned_loss=0.1688, over 26674.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3489, pruned_loss=0.09889, over 5710690.06 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3515, pruned_loss=0.09984, over 5738723.10 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3473, pruned_loss=0.09765, over 5708794.64 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:22:59,218 INFO [train.py:968] (0/2) Epoch 24, batch 22600, giga_loss[loss=0.2662, simple_loss=0.3489, pruned_loss=0.09176, over 28785.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3473, pruned_loss=0.09819, over 5708468.31 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3524, pruned_loss=0.1006, over 5730559.70 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3452, pruned_loss=0.09643, over 5713230.74 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:23:16,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2320, 1.5386, 1.4984, 1.1281], device='cuda:0'), covar=tensor([0.1732, 0.2593, 0.1488, 0.1630], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0705, 0.0961, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 12:23:20,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.552e+02 1.171e+03 1.387e+03 1.909e+03 3.616e+03, threshold=2.773e+03, percent-clipped=1.0
+2023-03-12 12:23:39,868 INFO [train.py:968] (0/2) Epoch 24, batch 22650, giga_loss[loss=0.2754, simple_loss=0.3493, pruned_loss=0.1007, over 27493.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3453, pruned_loss=0.09754, over 5711917.33 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3527, pruned_loss=0.1009, over 5734401.21 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3431, pruned_loss=0.09587, over 5712026.74 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:24:12,953 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5086, 1.7763, 1.5270, 1.4455], device='cuda:0'), covar=tensor([0.2145, 0.2139, 0.2230, 0.2093], device='cuda:0'), in_proj_covar=tensor([0.1535, 0.1106, 0.1353, 0.0991], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:0')
+2023-03-12 12:24:20,321 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-12 12:24:20,531 INFO [train.py:968] (0/2) Epoch 24, batch 22700, giga_loss[loss=0.2256, simple_loss=0.3083, pruned_loss=0.07142, over 28789.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3431, pruned_loss=0.09622, over 5713544.07 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3529, pruned_loss=0.1012, over 5733762.62 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3411, pruned_loss=0.09452, over 5713751.97 frames. ], batch size: 66, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:24:26,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1071605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:24:39,644 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-12 12:24:43,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.824e+02 1.282e+03 1.721e+03 2.690e+03 6.687e+03, threshold=3.442e+03, percent-clipped=21.0
+2023-03-12 12:25:02,460 INFO [train.py:968] (0/2) Epoch 24, batch 22750, giga_loss[loss=0.2262, simple_loss=0.3244, pruned_loss=0.06396, over 29044.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3451, pruned_loss=0.09583, over 5702272.97 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3535, pruned_loss=0.1017, over 5724434.86 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3427, pruned_loss=0.09393, over 5710296.42 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:25:10,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4535, 1.6501, 1.1882, 1.1815], device='cuda:0'), covar=tensor([0.0953, 0.0542, 0.1037, 0.1286], device='cuda:0'), in_proj_covar=tensor([0.0397, 0.0445, 0.0519, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 12:25:28,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2259, 1.3902, 1.4081, 1.2481], device='cuda:0'), covar=tensor([0.2977, 0.2143, 0.2186, 0.2460], device='cuda:0'), in_proj_covar=tensor([0.1998, 0.1933, 0.1860, 0.2009], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 12:25:43,993 INFO [train.py:968] (0/2) Epoch 24, batch 22800, giga_loss[loss=0.2539, simple_loss=0.3336, pruned_loss=0.08706, over 28945.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3457, pruned_loss=0.09627, over 5706990.55 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3532, pruned_loss=0.1019, over 5725389.58 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3438, pruned_loss=0.0943, over 5712022.74 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:26:03,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.506e+02 1.346e+03 1.678e+03 2.239e+03 7.471e+03, threshold=3.355e+03, percent-clipped=9.0
+2023-03-12 12:26:23,186 INFO [train.py:968] (0/2) Epoch 24, batch 22850, giga_loss[loss=0.3073, simple_loss=0.375, pruned_loss=0.1198, over 28156.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.345, pruned_loss=0.09657, over 5710448.49 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.354, pruned_loss=0.1026, over 5721905.51 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3424, pruned_loss=0.09413, over 5717646.17 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:26:23,429 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1071748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:26:25,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1071751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:26:49,914 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1071780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:27:04,181 INFO [train.py:968] (0/2) Epoch 24, batch 22900, libri_loss[loss=0.2985, simple_loss=0.372, pruned_loss=0.1125, over 19735.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3435, pruned_loss=0.09686, over 5703885.01 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3538, pruned_loss=0.1028, over 5714626.49 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3414, pruned_loss=0.0946, over 5716750.25 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:27:12,444 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-12 12:27:25,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.540e+02 1.338e+03 1.715e+03 2.277e+03 4.619e+03, threshold=3.430e+03, percent-clipped=6.0
+2023-03-12 12:27:42,066 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3094, 1.5423, 1.4356, 1.2508], device='cuda:0'), covar=tensor([0.3569, 0.2824, 0.2080, 0.2712], device='cuda:0'), in_proj_covar=tensor([0.2005, 0.1942, 0.1864, 0.2014], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 12:27:42,748 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-12 12:27:43,616 INFO [train.py:968] (0/2) Epoch 24, batch 22950, giga_loss[loss=0.2672, simple_loss=0.3379, pruned_loss=0.09828, over 28622.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3427, pruned_loss=0.09782, over 5707451.86 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3541, pruned_loss=0.1031, over 5716901.59 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3406, pruned_loss=0.09563, over 5715599.93 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:28:22,533 INFO [train.py:968] (0/2) Epoch 24, batch 23000, giga_loss[loss=0.2422, simple_loss=0.3233, pruned_loss=0.08052, over 28943.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3408, pruned_loss=0.09736, over 5718291.29 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3536, pruned_loss=0.1029, over 5725263.54 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.339, pruned_loss=0.09547, over 5717205.84 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:28:25,229 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-12 12:28:45,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.897e+02 1.232e+03 1.537e+03 2.075e+03 4.778e+03, threshold=3.075e+03, percent-clipped=5.0
+2023-03-12 12:29:02,873 INFO [train.py:968] (0/2) Epoch 24, batch 23050, giga_loss[loss=0.2246, simple_loss=0.2945, pruned_loss=0.07737, over 28573.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3392, pruned_loss=0.09697, over 5712561.98 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.354, pruned_loss=0.1033, over 5719832.76 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.337, pruned_loss=0.09497, over 5716988.23 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:29:20,592 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1071970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:29:43,934 INFO [train.py:968] (0/2) Epoch 24, batch 23100, giga_loss[loss=0.2774, simple_loss=0.3474, pruned_loss=0.1036, over 29051.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3359, pruned_loss=0.09517, over 5703463.40 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3543, pruned_loss=0.1035, over 5710031.57 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3339, pruned_loss=0.09334, over 5716281.61 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 12:29:46,270 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1072000.pt
+2023-03-12 12:29:58,545 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-12 12:30:05,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.107e+02 1.203e+03 1.494e+03 1.971e+03 5.300e+03, threshold=2.989e+03, percent-clipped=6.0
+2023-03-12 12:30:10,104 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1072031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:30:23,857 INFO [train.py:968] (0/2) Epoch 24, batch 23150, giga_loss[loss=0.2234, simple_loss=0.3002, pruned_loss=0.07326, over 28348.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3322, pruned_loss=0.09353, over 5708088.65 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3539, pruned_loss=0.1035, over 5714724.68 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3304, pruned_loss=0.09192, over 5713985.78 frames. ], batch size: 65, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 12:31:07,239 INFO [train.py:968] (0/2) Epoch 24, batch 23200, giga_loss[loss=0.2571, simple_loss=0.3331, pruned_loss=0.09049, over 28707.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3288, pruned_loss=0.09194, over 5715716.69 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3538, pruned_loss=0.1034, over 5715620.06 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3274, pruned_loss=0.09069, over 5719585.14 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:31:30,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.158e+02 1.160e+03 1.561e+03 2.381e+03 5.034e+03, threshold=3.121e+03, percent-clipped=15.0
+2023-03-12 12:31:48,436 INFO [train.py:968] (0/2) Epoch 24, batch 23250, giga_loss[loss=0.2355, simple_loss=0.315, pruned_loss=0.07801, over 28892.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3318, pruned_loss=0.09325, over 5710308.05 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3541, pruned_loss=0.1038, over 5719299.57 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3299, pruned_loss=0.09171, over 5710101.49 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:32:30,971 INFO [train.py:968] (0/2) Epoch 24, batch 23300, giga_loss[loss=0.2327, simple_loss=0.3186, pruned_loss=0.07345, over 29035.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3349, pruned_loss=0.09432, over 5695752.77 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.354, pruned_loss=0.1037, over 5705788.97 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.333, pruned_loss=0.09287, over 5707785.77 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:32:53,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.314e+02 1.186e+03 1.570e+03 1.994e+03 5.508e+03, threshold=3.141e+03, percent-clipped=6.0
+2023-03-12 12:33:10,486 INFO [train.py:968] (0/2) Epoch 24, batch 23350, libri_loss[loss=0.2432, simple_loss=0.3116, pruned_loss=0.08737, over 29522.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3379, pruned_loss=0.09526, over 5704271.76 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.354, pruned_loss=0.1039, over 5709211.43 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.336, pruned_loss=0.09379, over 5710689.81 frames. ], batch size: 70, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:33:10,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2439, 4.0696, 3.8537, 1.7984], device='cuda:0'), covar=tensor([0.0585, 0.0746, 0.0726, 0.2258], device='cuda:0'), in_proj_covar=tensor([0.1247, 0.1152, 0.0973, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 12:33:44,953 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5209, 1.6023, 1.4514, 1.6242], device='cuda:0'), covar=tensor([0.0723, 0.0311, 0.0320, 0.0842], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 12:33:52,237 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3669, 1.9054, 1.5507, 1.5309], device='cuda:0'), covar=tensor([0.0731, 0.0333, 0.0331, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:0')
+2023-03-12 12:33:54,698 INFO [train.py:968] (0/2) Epoch 24, batch 23400, libri_loss[loss=0.3119, simple_loss=0.3785, pruned_loss=0.1226, over 28617.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3419, pruned_loss=0.0969, over 5708455.65 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3542, pruned_loss=0.1041, over 5712111.79 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3398, pruned_loss=0.09536, over 5710878.29 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:34:19,543 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 12:34:19,695 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.249e+02 1.226e+03 1.464e+03 1.965e+03 4.105e+03, threshold=2.928e+03, percent-clipped=7.0
+2023-03-12 12:34:36,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1072345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:34:38,361 INFO [train.py:968] (0/2) Epoch 24, batch 23450, giga_loss[loss=0.2787, simple_loss=0.3453, pruned_loss=0.106, over 28865.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3439, pruned_loss=0.09765, over 5716896.14 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3543, pruned_loss=0.1043, over 5715791.73 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.342, pruned_loss=0.09612, over 5715517.84 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:35:25,192 INFO [train.py:968] (0/2) Epoch 24, batch 23500, giga_loss[loss=0.2855, simple_loss=0.3576, pruned_loss=0.1067, over 29004.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3476, pruned_loss=0.101, over 5713426.03 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3544, pruned_loss=0.1045, over 5719378.00 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3457, pruned_loss=0.09931, over 5709004.75 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:35:32,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1072406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:35:50,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.578e+03 2.181e+03 2.868e+03 7.172e+03, threshold=4.361e+03, percent-clipped=24.0
+2023-03-12 12:36:11,707 INFO [train.py:968] (0/2) Epoch 24, batch 23550, giga_loss[loss=0.3263, simple_loss=0.3881, pruned_loss=0.1323, over 28323.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3531, pruned_loss=0.1052, over 5709897.66 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3548, pruned_loss=0.1049, over 5724334.56 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3511, pruned_loss=0.1035, over 5701630.70 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:36:26,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-12 12:36:54,750 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1072488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:36:57,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1072491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:37:01,431 INFO [train.py:968] (0/2) Epoch 24, batch 23600, libri_loss[loss=0.3088, simple_loss=0.37, pruned_loss=0.1238, over 29570.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3599, pruned_loss=0.1103, over 5698576.24 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3555, pruned_loss=0.1056, over 5729852.23 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3577, pruned_loss=0.1084, over 5685847.13 frames. ], batch size: 78, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:37:23,828 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1072520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:37:28,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.217e+03 1.718e+03 2.311e+03 3.284e+03 6.094e+03, threshold=4.623e+03, percent-clipped=6.0
+2023-03-12 12:37:47,830 INFO [train.py:968] (0/2) Epoch 24, batch 23650, giga_loss[loss=0.3643, simple_loss=0.4119, pruned_loss=0.1583, over 28835.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3659, pruned_loss=0.1148, over 5692675.78 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3561, pruned_loss=0.1062, over 5730919.39 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3639, pruned_loss=0.113, over 5680151.81 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:37:48,172 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6681, 1.6086, 1.9005, 1.4557], device='cuda:0'), covar=tensor([0.1449, 0.2318, 0.1214, 0.1576], device='cuda:0'), in_proj_covar=tensor([0.0914, 0.0704, 0.0959, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 12:37:48,783 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1072549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:37:52,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1072552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:38:18,007 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1072581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:38:35,230 INFO [train.py:968] (0/2) Epoch 24, batch 23700, giga_loss[loss=0.3428, simple_loss=0.3998, pruned_loss=0.1429, over 28689.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3725, pruned_loss=0.1207, over 5689258.63 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3569, pruned_loss=0.1069, over 5733605.53 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3707, pruned_loss=0.119, over 5675007.71 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:39:07,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.218e+03 1.690e+03 2.289e+03 3.043e+03 8.044e+03, threshold=4.577e+03, percent-clipped=6.0
+2023-03-12 12:39:28,535 INFO [train.py:968] (0/2) Epoch 24, batch 23750, giga_loss[loss=0.3304, simple_loss=0.4001, pruned_loss=0.1303, over 28669.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3783, pruned_loss=0.1254, over 5683192.15 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.357, pruned_loss=0.1071, over 5725808.45 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3771, pruned_loss=0.1241, over 5677396.10 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:40:11,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-12 12:40:16,059 INFO [train.py:968] (0/2) Epoch 24, batch 23800, giga_loss[loss=0.3167, simple_loss=0.3822, pruned_loss=0.1255, over 28715.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3798, pruned_loss=0.1275, over 5667883.93 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3574, pruned_loss=0.1076, over 5712886.65 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.379, pruned_loss=0.1265, over 5672706.65 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:40:21,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1554, 4.0033, 3.8101, 1.9245], device='cuda:0'), covar=tensor([0.0690, 0.0815, 0.0838, 0.2052], device='cuda:0'), in_proj_covar=tensor([0.1252, 0.1156, 0.0977, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 12:40:39,282 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1072721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:40:46,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.853e+03 2.400e+03 3.239e+03 8.027e+03, threshold=4.799e+03, percent-clipped=14.0
+2023-03-12 12:41:07,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-12 12:41:07,262 INFO [train.py:968] (0/2) Epoch 24, batch 23850, giga_loss[loss=0.2981, simple_loss=0.3725, pruned_loss=0.1119, over 28787.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3837, pruned_loss=0.1318, over 5659853.38 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3578, pruned_loss=0.108, over 5717322.03 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3833, pruned_loss=0.1312, over 5658666.03 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:41:11,754 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3542, 1.7249, 1.2806, 0.8142], device='cuda:0'), covar=tensor([0.3420, 0.2214, 0.2074, 0.4030], device='cuda:0'), in_proj_covar=tensor([0.1776, 0.1677, 0.1623, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 12:41:23,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4516, 2.0296, 1.5435, 0.7822], device='cuda:0'), covar=tensor([0.3362, 0.2052, 0.2367, 0.4392], device='cuda:0'), in_proj_covar=tensor([0.1776, 0.1677, 0.1623, 0.1446], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 12:42:02,229 INFO [train.py:968] (0/2) Epoch 24, batch 23900, giga_loss[loss=0.4148, simple_loss=0.4383, pruned_loss=0.1957, over 28735.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3889, pruned_loss=0.1376, over 5646130.10 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3578, pruned_loss=0.1081, over 5719206.14 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3887, pruned_loss=0.1373, over 5642933.38 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:42:02,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7635, 1.8304, 1.9879, 1.5222], device='cuda:0'), covar=tensor([0.1815, 0.2514, 0.1428, 0.1747], device='cuda:0'), in_proj_covar=tensor([0.0911, 0.0704, 0.0957, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 12:42:34,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.113e+03 1.886e+03 2.364e+03 3.263e+03 6.677e+03, threshold=4.729e+03, percent-clipped=8.0
+2023-03-12 12:43:01,569 INFO [train.py:968] (0/2) Epoch 24, batch 23950, libri_loss[loss=0.273, simple_loss=0.3533, pruned_loss=0.09636, over 29475.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3915, pruned_loss=0.1398, over 5649099.62 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3575, pruned_loss=0.1082, over 5722872.63 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3925, pruned_loss=0.1402, over 5641366.84 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:43:56,374 INFO [train.py:968] (0/2) Epoch 24, batch 24000, giga_loss[loss=0.3223, simple_loss=0.3869, pruned_loss=0.1289, over 29038.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3914, pruned_loss=0.14, over 5656874.04 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3576, pruned_loss=0.1085, over 5727942.95 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3933, pruned_loss=0.1411, over 5643790.88 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:43:56,379 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 12:44:05,374 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2022, simple_loss=0.3102, pruned_loss=0.04706, over 944034.00 frames. 
+2023-03-12 12:44:05,375 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 12:44:08,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5557, 3.2973, 1.6478, 1.7017], device='cuda:0'), covar=tensor([0.0919, 0.0324, 0.0816, 0.1225], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0558, 0.0395, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 12:44:37,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.225e+03 2.024e+03 2.639e+03 3.843e+03 7.492e+03, threshold=5.279e+03, percent-clipped=15.0
+2023-03-12 12:44:56,136 INFO [train.py:968] (0/2) Epoch 24, batch 24050, giga_loss[loss=0.3349, simple_loss=0.3908, pruned_loss=0.1394, over 28322.00 frames. ], tot_loss[loss=0.336, simple_loss=0.391, pruned_loss=0.1405, over 5648323.18 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3576, pruned_loss=0.1085, over 5728825.23 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3926, pruned_loss=0.1416, over 5636862.87 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:45:05,695 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2622, 1.4834, 1.4427, 1.3422], device='cuda:0'), covar=tensor([0.1452, 0.1265, 0.1833, 0.1386], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0755, 0.0724, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 12:45:44,124 INFO [train.py:968] (0/2) Epoch 24, batch 24100, giga_loss[loss=0.3783, simple_loss=0.4006, pruned_loss=0.178, over 23518.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3901, pruned_loss=0.1397, over 5651520.89 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3583, pruned_loss=0.1091, over 5732652.98 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3913, pruned_loss=0.1406, over 5637553.27 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:46:11,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.842e+03 2.412e+03 3.742e+03 8.725e+03, threshold=4.824e+03, percent-clipped=6.0
+2023-03-12 12:46:21,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4345, 1.4700, 3.4602, 3.3517], device='cuda:0'), covar=tensor([0.1320, 0.2583, 0.0480, 0.0992], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0661, 0.0979, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 12:46:32,654 INFO [train.py:968] (0/2) Epoch 24, batch 24150, giga_loss[loss=0.3927, simple_loss=0.4346, pruned_loss=0.1755, over 28898.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.389, pruned_loss=0.1377, over 5642064.11 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3586, pruned_loss=0.1094, over 5721606.66 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3908, pruned_loss=0.1392, over 5637167.41 frames. ], batch size: 243, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:47:03,293 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2360, 1.2315, 1.2336, 1.4267], device='cuda:0'), covar=tensor([0.0808, 0.0384, 0.0344, 0.0891], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 12:47:20,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1073096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:47:22,163 INFO [train.py:968] (0/2) Epoch 24, batch 24200, giga_loss[loss=0.3712, simple_loss=0.4148, pruned_loss=0.1638, over 28600.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3904, pruned_loss=0.1385, over 5640231.33 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.359, pruned_loss=0.1099, over 5725400.11 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3921, pruned_loss=0.1397, over 5631575.02 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:47:40,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2323, 1.4971, 1.5219, 1.1114], device='cuda:0'), covar=tensor([0.1526, 0.2410, 0.1259, 0.1604], device='cuda:0'), in_proj_covar=tensor([0.0908, 0.0702, 0.0954, 0.0854], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 12:47:55,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.866e+03 2.824e+03 3.600e+03 1.021e+04, threshold=5.648e+03, percent-clipped=15.0
+2023-03-12 12:48:17,080 INFO [train.py:968] (0/2) Epoch 24, batch 24250, giga_loss[loss=0.2968, simple_loss=0.3637, pruned_loss=0.115, over 28877.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3903, pruned_loss=0.1384, over 5631730.15 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3596, pruned_loss=0.1105, over 5727781.07 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3916, pruned_loss=0.1393, over 5621298.28 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:49:08,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4921, 1.6770, 1.2247, 1.2431], device='cuda:0'), covar=tensor([0.1012, 0.0594, 0.1095, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0451, 0.0522, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 12:49:09,211 INFO [train.py:968] (0/2) Epoch 24, batch 24300, giga_loss[loss=0.3074, simple_loss=0.378, pruned_loss=0.1184, over 28690.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3859, pruned_loss=0.1334, over 5628115.45 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3597, pruned_loss=0.1106, over 5718390.57 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3876, pruned_loss=0.1347, over 5625174.60 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:49:22,264 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1073211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:49:40,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.608e+02 1.648e+03 2.058e+03 2.771e+03 5.061e+03, threshold=4.116e+03, percent-clipped=0.0
+2023-03-12 12:49:54,234 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1073239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:49:56,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1073242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:49:59,310 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1073244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:50:02,443 INFO [train.py:968] (0/2) Epoch 24, batch 24350, giga_loss[loss=0.3136, simple_loss=0.3744, pruned_loss=0.1264, over 27489.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.382, pruned_loss=0.1293, over 5638940.48 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3593, pruned_loss=0.1105, over 5720962.96 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3841, pruned_loss=0.1307, over 5633088.83 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:50:26,452 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1073271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:50:43,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5524, 4.3442, 1.6369, 1.7411], device='cuda:0'), covar=tensor([0.0982, 0.0389, 0.0930, 0.1304], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0560, 0.0395, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 12:50:51,255 INFO [train.py:968] (0/2) Epoch 24, batch 24400, giga_loss[loss=0.2831, simple_loss=0.3542, pruned_loss=0.106, over 28455.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3778, pruned_loss=0.1253, over 5650875.89 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.359, pruned_loss=0.1104, over 5714000.98 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3801, pruned_loss=0.1269, over 5651348.97 frames. ], batch size: 78, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:51:22,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.789e+03 2.272e+03 3.187e+03 7.099e+03, threshold=4.543e+03, percent-clipped=11.0
+2023-03-12 12:51:41,246 INFO [train.py:968] (0/2) Epoch 24, batch 24450, giga_loss[loss=0.3192, simple_loss=0.38, pruned_loss=0.1292, over 28643.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3767, pruned_loss=0.1246, over 5648326.96 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3592, pruned_loss=0.1105, over 5713486.89 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3787, pruned_loss=0.126, over 5648483.46 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:52:30,563 INFO [train.py:968] (0/2) Epoch 24, batch 24500, libri_loss[loss=0.2457, simple_loss=0.3143, pruned_loss=0.08856, over 29638.00 frames. ], tot_loss[loss=0.311, simple_loss=0.375, pruned_loss=0.1235, over 5655606.64 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3591, pruned_loss=0.1106, over 5707721.98 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3772, pruned_loss=0.1249, over 5659850.61 frames. ], batch size: 73, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:53:02,713 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.145e+02 1.620e+03 1.948e+03 2.629e+03 6.340e+03, threshold=3.896e+03, percent-clipped=2.0
+2023-03-12 12:53:25,187 INFO [train.py:968] (0/2) Epoch 24, batch 24550, giga_loss[loss=0.3061, simple_loss=0.3733, pruned_loss=0.1195, over 27955.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3751, pruned_loss=0.1231, over 5656440.80 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3591, pruned_loss=0.1107, over 5708881.63 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3771, pruned_loss=0.1244, over 5657747.80 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:53:26,408 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-12 12:54:15,994 INFO [train.py:968] (0/2) Epoch 24, batch 24600, giga_loss[loss=0.3038, simple_loss=0.3685, pruned_loss=0.1195, over 28686.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3727, pruned_loss=0.1209, over 5661360.90 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3595, pruned_loss=0.1111, over 5711313.07 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3742, pruned_loss=0.1217, over 5659650.79 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:54:48,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.088e+03 1.674e+03 2.232e+03 3.617e+03 1.305e+04, threshold=4.465e+03, percent-clipped=21.0
+2023-03-12 12:54:52,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3060, 2.4738, 1.8810, 2.1555], device='cuda:0'), covar=tensor([0.0911, 0.0651, 0.0935, 0.1067], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0452, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 12:55:08,749 INFO [train.py:968] (0/2) Epoch 24, batch 24650, giga_loss[loss=0.3639, simple_loss=0.413, pruned_loss=0.1574, over 27643.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3734, pruned_loss=0.1188, over 5678404.42 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3592, pruned_loss=0.1109, over 5714097.90 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3753, pruned_loss=0.1199, over 5673086.52 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 12:55:45,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1073586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:55:56,983 INFO [train.py:968] (0/2) Epoch 24, batch 24700, giga_loss[loss=0.3045, simple_loss=0.3764, pruned_loss=0.1163, over 28000.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3736, pruned_loss=0.1181, over 5664935.35 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.359, pruned_loss=0.1108, over 5719258.57 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3759, pruned_loss=0.1194, over 5654320.75 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 12:56:19,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1073619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:56:29,189 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.722e+03 2.055e+03 2.946e+03 8.093e+03, threshold=4.109e+03, percent-clipped=6.0
+2023-03-12 12:56:47,928 INFO [train.py:968] (0/2) Epoch 24, batch 24750, giga_loss[loss=0.3159, simple_loss=0.3782, pruned_loss=0.1267, over 28258.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3735, pruned_loss=0.1183, over 5669591.99 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3582, pruned_loss=0.1103, over 5722801.19 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3763, pruned_loss=0.12, over 5656986.42 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 12:57:34,226 INFO [train.py:968] (0/2) Epoch 24, batch 24800, giga_loss[loss=0.2743, simple_loss=0.3461, pruned_loss=0.1012, over 28305.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3744, pruned_loss=0.1198, over 5665924.33 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3582, pruned_loss=0.1103, over 5725054.73 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3768, pruned_loss=0.1212, over 5653253.88 frames. ], batch size: 77, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:57:59,283 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 12:58:05,396 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1073729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:58:06,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.845e+03 2.277e+03 3.123e+03 6.498e+03, threshold=4.554e+03, percent-clipped=14.0
+2023-03-12 12:58:08,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1073732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:58:22,487 INFO [train.py:968] (0/2) Epoch 24, batch 24850, libri_loss[loss=0.2442, simple_loss=0.3121, pruned_loss=0.08811, over 29564.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3709, pruned_loss=0.1187, over 5664816.17 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3576, pruned_loss=0.1101, over 5731928.02 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.374, pruned_loss=0.1204, over 5645674.88 frames. ], batch size: 74, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:58:34,597 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1073761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:58:35,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1073762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:58:39,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1073765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:59:06,575 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1073794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:59:09,078 INFO [train.py:968] (0/2) Epoch 24, batch 24900, libri_loss[loss=0.39, simple_loss=0.4306, pruned_loss=0.1747, over 29233.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.369, pruned_loss=0.118, over 5674876.75 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.358, pruned_loss=0.1104, over 5731954.38 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3713, pruned_loss=0.1192, over 5658145.72 frames. ], batch size: 94, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:59:39,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.135e+03 1.666e+03 2.164e+03 2.904e+03 5.605e+03, threshold=4.327e+03, percent-clipped=6.0
+2023-03-12 12:59:56,789 INFO [train.py:968] (0/2) Epoch 24, batch 24950, giga_loss[loss=0.3057, simple_loss=0.3877, pruned_loss=0.1118, over 28933.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3682, pruned_loss=0.1174, over 5676274.02 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3579, pruned_loss=0.1104, over 5732123.54 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3703, pruned_loss=0.1184, over 5661484.55 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:00:43,101 INFO [train.py:968] (0/2) Epoch 24, batch 25000, giga_loss[loss=0.2914, simple_loss=0.3624, pruned_loss=0.1102, over 28722.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3689, pruned_loss=0.117, over 5676705.30 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3578, pruned_loss=0.1103, over 5731810.83 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3708, pruned_loss=0.1179, over 5664962.29 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:00:46,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6616, 1.9115, 1.4155, 1.4573], device='cuda:0'), covar=tensor([0.1066, 0.0661, 0.1088, 0.1217], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0449, 0.0519, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 13:01:14,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.710e+03 2.386e+03 3.273e+03 8.707e+03, threshold=4.772e+03, percent-clipped=10.0
+2023-03-12 13:01:27,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4602, 2.1108, 1.5818, 0.8143], device='cuda:0'), covar=tensor([0.6179, 0.3106, 0.4023, 0.6404], device='cuda:0'), in_proj_covar=tensor([0.1775, 0.1675, 0.1619, 0.1439], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 13:01:32,374 INFO [train.py:968] (0/2) Epoch 24, batch 25050, giga_loss[loss=0.2931, simple_loss=0.3652, pruned_loss=0.1105, over 27974.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3695, pruned_loss=0.1174, over 5658653.14 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3582, pruned_loss=0.1107, over 5714967.11 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3708, pruned_loss=0.118, over 5662094.74 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:01:50,493 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-12 13:02:22,617 INFO [train.py:968] (0/2) Epoch 24, batch 25100, giga_loss[loss=0.2895, simple_loss=0.3552, pruned_loss=0.1119, over 28556.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3679, pruned_loss=0.1165, over 5658323.82 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3576, pruned_loss=0.1104, over 5708685.83 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3698, pruned_loss=0.1174, over 5664744.34 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:02:24,461 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1074000.pt
+2023-03-12 13:02:53,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.716e+03 2.089e+03 2.718e+03 9.887e+03, threshold=4.179e+03, percent-clipped=6.0
+2023-03-12 13:03:14,157 INFO [train.py:968] (0/2) Epoch 24, batch 25150, giga_loss[loss=0.2844, simple_loss=0.3533, pruned_loss=0.1077, over 28838.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3678, pruned_loss=0.1168, over 5668226.76 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3585, pruned_loss=0.1111, over 5702903.63 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3687, pruned_loss=0.1171, over 5677649.14 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:03:22,071 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-12 13:03:31,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7106, 1.9178, 1.7018, 1.6916], device='cuda:0'), covar=tensor([0.2251, 0.2766, 0.2579, 0.2628], device='cuda:0'), in_proj_covar=tensor([0.0490, 0.0758, 0.0726, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 13:04:05,554 INFO [train.py:968] (0/2) Epoch 24, batch 25200, giga_loss[loss=0.2836, simple_loss=0.3589, pruned_loss=0.1041, over 28850.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3668, pruned_loss=0.1171, over 5668066.44 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3583, pruned_loss=0.111, over 5705056.56 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3678, pruned_loss=0.1174, over 5673433.17 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:04:40,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.752e+03 2.235e+03 3.013e+03 6.350e+03, threshold=4.470e+03, percent-clipped=10.0
+2023-03-12 13:04:44,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2128, 3.0434, 1.3693, 1.3620], device='cuda:0'), covar=tensor([0.1101, 0.0467, 0.0945, 0.1459], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0562, 0.0396, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 13:04:55,233 INFO [train.py:968] (0/2) Epoch 24, batch 25250, giga_loss[loss=0.3242, simple_loss=0.3837, pruned_loss=0.1324, over 28958.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3648, pruned_loss=0.1161, over 5683356.65 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3585, pruned_loss=0.1111, over 5709108.01 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3656, pruned_loss=0.1164, over 5683469.32 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:05:15,092 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 13:05:46,769 INFO [train.py:968] (0/2) Epoch 24, batch 25300, giga_loss[loss=0.2457, simple_loss=0.3256, pruned_loss=0.0829, over 28910.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3633, pruned_loss=0.1155, over 5684167.78 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3591, pruned_loss=0.1115, over 5711770.53 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3635, pruned_loss=0.1154, over 5681517.00 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:06:06,624 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4898, 1.7565, 1.3953, 1.5275], device='cuda:0'), covar=tensor([0.2687, 0.2667, 0.2958, 0.2472], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1116, 0.1364, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 13:06:19,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.995e+02 1.764e+03 2.364e+03 3.162e+03 1.057e+04, threshold=4.728e+03, percent-clipped=9.0
+2023-03-12 13:06:37,509 INFO [train.py:968] (0/2) Epoch 24, batch 25350, giga_loss[loss=0.2951, simple_loss=0.3552, pruned_loss=0.1175, over 27931.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3631, pruned_loss=0.116, over 5683692.99 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3596, pruned_loss=0.1119, over 5714842.39 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3628, pruned_loss=0.1157, over 5678363.10 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:07:24,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5263, 4.3719, 1.6086, 1.7544], device='cuda:0'), covar=tensor([0.1032, 0.0328, 0.0935, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0562, 0.0396, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 13:07:25,004 INFO [train.py:968] (0/2) Epoch 24, batch 25400, giga_loss[loss=0.2713, simple_loss=0.3534, pruned_loss=0.09465, over 28882.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3643, pruned_loss=0.1166, over 5686843.69 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3597, pruned_loss=0.1119, over 5717910.25 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.364, pruned_loss=0.1164, over 5679351.34 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:07:36,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1074306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:07:59,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.863e+02 1.842e+03 2.783e+03 3.907e+03 9.596e+03, threshold=5.567e+03, percent-clipped=10.0
+2023-03-12 13:08:13,637 INFO [train.py:968] (0/2) Epoch 24, batch 25450, giga_loss[loss=0.3773, simple_loss=0.4247, pruned_loss=0.1649, over 28599.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3647, pruned_loss=0.1163, over 5690391.83 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3592, pruned_loss=0.1118, over 5720781.52 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.365, pruned_loss=0.1162, over 5681117.67 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:08:19,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5230, 2.1053, 1.5597, 0.7440], device='cuda:0'), covar=tensor([0.6278, 0.3177, 0.4198, 0.7226], device='cuda:0'), in_proj_covar=tensor([0.1787, 0.1683, 0.1623, 0.1444], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 13:08:35,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3304, 3.1399, 2.9993, 1.4283], device='cuda:0'), covar=tensor([0.0996, 0.1182, 0.1065, 0.2308], device='cuda:0'), in_proj_covar=tensor([0.1275, 0.1177, 0.0993, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 13:09:01,986 INFO [train.py:968] (0/2) Epoch 24, batch 25500, giga_loss[loss=0.2511, simple_loss=0.3402, pruned_loss=0.08102, over 28893.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3653, pruned_loss=0.1161, over 5688529.36 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.359, pruned_loss=0.1117, over 5720070.64 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3658, pruned_loss=0.1162, over 5681419.51 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:09:04,761 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1074400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:09:07,997 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-12 13:09:35,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.931e+02 1.791e+03 2.167e+03 2.870e+03 1.066e+04, threshold=4.334e+03, percent-clipped=6.0
+2023-03-12 13:09:47,760 INFO [train.py:968] (0/2) Epoch 24, batch 25550, giga_loss[loss=0.2908, simple_loss=0.3601, pruned_loss=0.1107, over 29013.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3665, pruned_loss=0.1164, over 5693192.53 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.359, pruned_loss=0.1115, over 5725371.71 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3671, pruned_loss=0.1169, over 5681659.63 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:10:35,433 INFO [train.py:968] (0/2) Epoch 24, batch 25600, giga_loss[loss=0.3406, simple_loss=0.3967, pruned_loss=0.1422, over 28906.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 5682554.16 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3593, pruned_loss=0.1117, over 5716673.20 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3687, pruned_loss=0.1185, over 5680638.30 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:10:49,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0047, 1.3156, 1.0659, 0.2279], device='cuda:0'), covar=tensor([0.4080, 0.3295, 0.4374, 0.6464], device='cuda:0'), in_proj_covar=tensor([0.1785, 0.1685, 0.1621, 0.1444], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 13:11:10,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-12 13:11:13,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.953e+03 2.494e+03 3.459e+03 5.538e+03, threshold=4.987e+03, percent-clipped=5.0
+2023-03-12 13:11:26,612 INFO [train.py:968] (0/2) Epoch 24, batch 25650, giga_loss[loss=0.2633, simple_loss=0.3386, pruned_loss=0.094, over 28536.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3706, pruned_loss=0.1211, over 5684912.46 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3593, pruned_loss=0.1117, over 5720359.31 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3713, pruned_loss=0.1215, over 5679324.43 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:12:16,956 INFO [train.py:968] (0/2) Epoch 24, batch 25700, giga_loss[loss=0.2738, simple_loss=0.3442, pruned_loss=0.1017, over 28543.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3714, pruned_loss=0.123, over 5686414.71 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3594, pruned_loss=0.1119, over 5725525.25 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3721, pruned_loss=0.1235, over 5676065.01 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:12:55,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.870e+03 2.274e+03 3.062e+03 5.531e+03, threshold=4.548e+03, percent-clipped=3.0
+2023-03-12 13:13:09,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4535, 1.8177, 1.4770, 1.5639], device='cuda:0'), covar=tensor([0.0754, 0.0304, 0.0327, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0224], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 13:13:11,531 INFO [train.py:968] (0/2) Epoch 24, batch 25750, giga_loss[loss=0.3137, simple_loss=0.3695, pruned_loss=0.1289, over 28875.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3715, pruned_loss=0.1239, over 5685663.51 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.359, pruned_loss=0.1117, over 5726248.10 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3728, pruned_loss=0.1248, over 5675501.95 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:13:37,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1074681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:13:41,489 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1074686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:13:52,770 INFO [train.py:968] (0/2) Epoch 24, batch 25800, giga_loss[loss=0.2961, simple_loss=0.3597, pruned_loss=0.1163, over 28738.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3695, pruned_loss=0.1224, over 5672678.20 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1115, over 5714490.36 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3715, pruned_loss=0.1238, over 5672901.39 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:14:25,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.869e+03 2.231e+03 3.490e+03 7.371e+03, threshold=4.462e+03, percent-clipped=12.0
+2023-03-12 13:14:41,639 INFO [train.py:968] (0/2) Epoch 24, batch 25850, giga_loss[loss=0.2525, simple_loss=0.3323, pruned_loss=0.0864, over 28894.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3688, pruned_loss=0.1219, over 5670459.96 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3584, pruned_loss=0.1114, over 5719713.70 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3708, pruned_loss=0.1233, over 5664879.63 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:15:03,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1074775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:15:26,753 INFO [train.py:968] (0/2) Epoch 24, batch 25900, giga_loss[loss=0.3499, simple_loss=0.4054, pruned_loss=0.1471, over 28874.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3691, pruned_loss=0.1203, over 5670666.64 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3583, pruned_loss=0.1114, over 5711598.13 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3708, pruned_loss=0.1215, over 5672290.17 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:15:51,848 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1074824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:15:56,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1074827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:16:03,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.808e+03 2.141e+03 3.051e+03 7.518e+03, threshold=4.282e+03, percent-clipped=7.0
+2023-03-12 13:16:15,581 INFO [train.py:968] (0/2) Epoch 24, batch 25950, giga_loss[loss=0.2918, simple_loss=0.3573, pruned_loss=0.1132, over 28972.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3668, pruned_loss=0.1187, over 5660448.56 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3582, pruned_loss=0.1115, over 5713565.87 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3684, pruned_loss=0.1197, over 5659133.16 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:16:23,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1074856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:16:56,756 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3821, 1.5601, 1.4064, 1.6221], device='cuda:0'), covar=tensor([0.0776, 0.0349, 0.0331, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 13:17:00,907 INFO [train.py:968] (0/2) Epoch 24, batch 26000, giga_loss[loss=0.2528, simple_loss=0.3278, pruned_loss=0.0889, over 28806.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3656, pruned_loss=0.1182, over 5670009.51 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3584, pruned_loss=0.1116, over 5718650.36 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3671, pruned_loss=0.1193, over 5662562.81 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:17:19,967 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1074918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:17:22,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1074921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:17:31,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.150e+03 1.739e+03 2.001e+03 2.725e+03 8.027e+03, threshold=4.002e+03, percent-clipped=4.0
+2023-03-12 13:17:46,185 INFO [train.py:968] (0/2) Epoch 24, batch 26050, giga_loss[loss=0.2919, simple_loss=0.3621, pruned_loss=0.1108, over 29088.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.364, pruned_loss=0.1174, over 5669629.53 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3583, pruned_loss=0.1114, over 5722536.38 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3656, pruned_loss=0.1187, over 5658102.62 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:17:50,100 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1074950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:18:38,043 INFO [train.py:968] (0/2) Epoch 24, batch 26100, giga_loss[loss=0.3158, simple_loss=0.3801, pruned_loss=0.1258, over 28959.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3657, pruned_loss=0.119, over 5649344.37 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3583, pruned_loss=0.1116, over 5704762.14 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3671, pruned_loss=0.1201, over 5653820.01 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:19:10,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.682e+03 2.306e+03 3.348e+03 8.498e+03, threshold=4.612e+03, percent-clipped=13.0
+2023-03-12 13:19:11,632 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-12 13:19:22,962 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-03-12 13:19:25,489 INFO [train.py:968] (0/2) Epoch 24, batch 26150, giga_loss[loss=0.3838, simple_loss=0.4269, pruned_loss=0.1703, over 27570.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3687, pruned_loss=0.1202, over 5657786.59 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3582, pruned_loss=0.1115, over 5708765.42 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3702, pruned_loss=0.1213, over 5656681.38 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:19:27,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6754, 1.8918, 1.4900, 2.0209], device='cuda:0'), covar=tensor([0.2796, 0.2926, 0.3318, 0.2521], device='cuda:0'), in_proj_covar=tensor([0.1544, 0.1112, 0.1358, 0.0997], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 13:19:33,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5829, 4.4258, 4.1337, 2.1627], device='cuda:0'), covar=tensor([0.0641, 0.0883, 0.1008, 0.1887], device='cuda:0'), in_proj_covar=tensor([0.1271, 0.1178, 0.0993, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 13:19:37,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1075061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:19:37,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1086, 1.1904, 1.0958, 0.8418], device='cuda:0'), covar=tensor([0.1113, 0.0616, 0.1159, 0.1181], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0451, 0.0521, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 13:20:13,560 INFO [train.py:968] (0/2) Epoch 24, batch 26200, giga_loss[loss=0.2712, simple_loss=0.359, pruned_loss=0.09164, over 29050.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3714, pruned_loss=0.1192, over 5663666.07 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.358, pruned_loss=0.1115, over 5711594.65 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.373, pruned_loss=0.1203, over 5659186.33 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:20:17,372 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 13:20:48,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.522e+02 1.555e+03 2.116e+03 2.854e+03 1.053e+04, threshold=4.233e+03, percent-clipped=9.0
+2023-03-12 13:21:03,696 INFO [train.py:968] (0/2) Epoch 24, batch 26250, giga_loss[loss=0.2779, simple_loss=0.352, pruned_loss=0.1019, over 28858.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3715, pruned_loss=0.1189, over 5664364.06 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3575, pruned_loss=0.1111, over 5716288.12 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3735, pruned_loss=0.1202, over 5655660.73 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:21:21,078 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:21:51,605 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-12 13:21:53,781 INFO [train.py:968] (0/2) Epoch 24, batch 26300, giga_loss[loss=0.3398, simple_loss=0.3936, pruned_loss=0.143, over 28933.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3746, pruned_loss=0.1218, over 5655786.52 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3577, pruned_loss=0.1112, over 5712982.26 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3763, pruned_loss=0.1229, over 5650895.46 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:21:59,898 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1075204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:22:02,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1075207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:22:21,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9595, 1.9846, 2.0843, 1.7312], device='cuda:0'), covar=tensor([0.1876, 0.2502, 0.1511, 0.1723], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0709, 0.0962, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 13:22:23,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2311, 1.4422, 1.3403, 1.1506], device='cuda:0'), covar=tensor([0.3052, 0.2844, 0.2065, 0.2488], device='cuda:0'), in_proj_covar=tensor([0.2018, 0.1959, 0.1875, 0.2021], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 13:22:24,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.951e+03 2.367e+03 3.305e+03 8.373e+03, threshold=4.734e+03, percent-clipped=11.0
+2023-03-12 13:22:26,660 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1075236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:22:39,310 INFO [train.py:968] (0/2) Epoch 24, batch 26350, giga_loss[loss=0.3866, simple_loss=0.4103, pruned_loss=0.1814, over 23510.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3752, pruned_loss=0.1231, over 5651366.86 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3581, pruned_loss=0.1117, over 5714033.23 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3766, pruned_loss=0.1239, over 5645514.50 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:23:13,603 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:23:32,202 INFO [train.py:968] (0/2) Epoch 24, batch 26400, giga_loss[loss=0.3269, simple_loss=0.3743, pruned_loss=0.1397, over 28909.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.374, pruned_loss=0.1232, over 5650014.84 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3579, pruned_loss=0.1116, over 5714815.06 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3753, pruned_loss=0.1239, over 5644580.33 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:24:06,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.787e+02 1.680e+03 2.259e+03 3.031e+03 9.555e+03, threshold=4.518e+03, percent-clipped=4.0
+2023-03-12 13:24:19,783 INFO [train.py:968] (0/2) Epoch 24, batch 26450, libri_loss[loss=0.3256, simple_loss=0.3878, pruned_loss=0.1317, over 25917.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.372, pruned_loss=0.1224, over 5651190.07 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3582, pruned_loss=0.1119, over 5716619.03 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3732, pruned_loss=0.123, over 5643656.93 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:24:54,128 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4926, 3.3404, 3.1559, 1.9549], device='cuda:0'), covar=tensor([0.0752, 0.0897, 0.0859, 0.1772], device='cuda:0'), in_proj_covar=tensor([0.1269, 0.1177, 0.0993, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 13:25:05,570 INFO [train.py:968] (0/2) Epoch 24, batch 26500, giga_loss[loss=0.3662, simple_loss=0.4076, pruned_loss=0.1623, over 28227.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3711, pruned_loss=0.1223, over 5661053.77 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3586, pruned_loss=0.1121, over 5719561.77 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.372, pruned_loss=0.1229, over 5650857.59 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:25:20,045 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4397, 1.0595, 4.8630, 3.7315], device='cuda:0'), covar=tensor([0.1811, 0.3132, 0.0405, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0667, 0.0985, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 13:25:39,941 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:25:42,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.233e+03 1.838e+03 2.454e+03 3.387e+03 1.029e+04, threshold=4.909e+03, percent-clipped=18.0
+2023-03-12 13:25:56,943 INFO [train.py:968] (0/2) Epoch 24, batch 26550, giga_loss[loss=0.3248, simple_loss=0.3622, pruned_loss=0.1437, over 23877.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3704, pruned_loss=0.1227, over 5648037.76 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3587, pruned_loss=0.1122, over 5723366.97 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3712, pruned_loss=0.1233, over 5635936.95 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:26:44,667 INFO [train.py:968] (0/2) Epoch 24, batch 26600, giga_loss[loss=0.2694, simple_loss=0.3344, pruned_loss=0.1022, over 28697.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3703, pruned_loss=0.1229, over 5652760.65 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3587, pruned_loss=0.1122, over 5725474.59 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3711, pruned_loss=0.1234, over 5640614.46 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:27:19,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.140e+03 1.795e+03 2.359e+03 3.119e+03 1.060e+04, threshold=4.718e+03, percent-clipped=7.0
+2023-03-12 13:27:23,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1075540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:27:24,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075541.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:27:32,748 INFO [train.py:968] (0/2) Epoch 24, batch 26650, giga_loss[loss=0.3182, simple_loss=0.3647, pruned_loss=0.1358, over 28968.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3677, pruned_loss=0.1213, over 5669110.76 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3584, pruned_loss=0.1119, over 5727288.51 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3688, pruned_loss=0.1221, over 5656697.55 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:28:19,321 INFO [train.py:968] (0/2) Epoch 24, batch 26700, giga_loss[loss=0.327, simple_loss=0.3874, pruned_loss=0.1333, over 28884.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3678, pruned_loss=0.1215, over 5666024.40 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3586, pruned_loss=0.1119, over 5718312.25 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3688, pruned_loss=0.1224, over 5663867.27 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:28:46,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3440, 1.3861, 1.2991, 1.4926], device='cuda:0'), covar=tensor([0.0777, 0.0375, 0.0347, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0120, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 13:28:54,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2995, 1.5852, 1.3439, 1.5126], device='cuda:0'), covar=tensor([0.0759, 0.0376, 0.0349, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0120, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 13:28:54,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.268e+03 1.750e+03 2.381e+03 2.832e+03 5.856e+03, threshold=4.763e+03, percent-clipped=1.0
+2023-03-12 13:29:05,975 INFO [train.py:968] (0/2) Epoch 24, batch 26750, giga_loss[loss=0.3041, simple_loss=0.3782, pruned_loss=0.1149, over 29010.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3684, pruned_loss=0.1211, over 5672074.10 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3588, pruned_loss=0.1121, over 5721198.98 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3691, pruned_loss=0.1218, over 5666868.43 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:29:14,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1075655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:29:40,776 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1075683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:29:43,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1075686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:29:53,986 INFO [train.py:968] (0/2) Epoch 24, batch 26800, giga_loss[loss=0.3114, simple_loss=0.3792, pruned_loss=0.1218, over 28817.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3708, pruned_loss=0.1219, over 5672971.09 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3591, pruned_loss=0.1121, over 5725711.81 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3712, pruned_loss=0.1226, over 5663734.65 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:30:11,777 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:30:14,383 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1075715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:30:33,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.963e+02 1.814e+03 2.379e+03 3.363e+03 9.273e+03, threshold=4.758e+03, percent-clipped=6.0
+2023-03-12 13:30:44,082 INFO [train.py:968] (0/2) Epoch 24, batch 26850, libri_loss[loss=0.2787, simple_loss=0.341, pruned_loss=0.1082, over 29577.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3715, pruned_loss=0.1231, over 5647655.70 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3592, pruned_loss=0.1122, over 5710813.42 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3722, pruned_loss=0.1239, over 5652825.79 frames. ], batch size: 77, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:30:47,323 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:30:52,787 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.51 vs. limit=5.0
+2023-03-12 13:31:31,110 INFO [train.py:968] (0/2) Epoch 24, batch 26900, giga_loss[loss=0.2929, simple_loss=0.3753, pruned_loss=0.1053, over 28733.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3746, pruned_loss=0.1232, over 5662410.51 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3592, pruned_loss=0.1122, over 5711716.36 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3752, pruned_loss=0.1239, over 5665454.94 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:31:31,380 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1075798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:31:33,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1075801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:31:39,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1075807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:32:00,626 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1075830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:32:07,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.689e+03 2.198e+03 2.955e+03 6.533e+03, threshold=4.395e+03, percent-clipped=4.0
+2023-03-12 13:32:08,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4799, 1.6918, 1.6724, 1.4436], device='cuda:0'), covar=tensor([0.3421, 0.2522, 0.2244, 0.2693], device='cuda:0'), in_proj_covar=tensor([0.2011, 0.1947, 0.1865, 0.2012], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 13:32:17,752 INFO [train.py:968] (0/2) Epoch 24, batch 26950, giga_loss[loss=0.297, simple_loss=0.375, pruned_loss=0.1095, over 28883.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3742, pruned_loss=0.1213, over 5668509.49 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3591, pruned_loss=0.1123, over 5715661.52 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3752, pruned_loss=0.1221, over 5666059.62 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:33:02,646 INFO [train.py:968] (0/2) Epoch 24, batch 27000, giga_loss[loss=0.2902, simple_loss=0.3695, pruned_loss=0.1055, over 28719.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3745, pruned_loss=0.12, over 5675928.34 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3589, pruned_loss=0.1122, over 5722035.29 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.376, pruned_loss=0.121, over 5666705.03 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:33:02,651 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 13:33:11,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3645, 2.9274, 1.4450, 1.5096], device='cuda:0'), covar=tensor([0.1044, 0.0394, 0.0987, 0.1403], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0563, 0.0396, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 13:33:11,742 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2027, simple_loss=0.3096, pruned_loss=0.04786, over 944034.00 frames. 
+2023-03-12 13:33:11,743 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 13:33:12,002 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9370, 1.3144, 1.4132, 1.0885], device='cuda:0'), covar=tensor([0.1693, 0.1102, 0.1852, 0.1400], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0753, 0.0723, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 13:33:26,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1075916.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:33:46,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.614e+03 2.201e+03 3.049e+03 9.411e+03, threshold=4.401e+03, percent-clipped=11.0
+2023-03-12 13:33:59,584 INFO [train.py:968] (0/2) Epoch 24, batch 27050, giga_loss[loss=0.4381, simple_loss=0.4677, pruned_loss=0.2042, over 27634.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3792, pruned_loss=0.1241, over 5679248.92 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5724235.68 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3803, pruned_loss=0.1248, over 5669740.41 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:34:02,978 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1075950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:34:05,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1075953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:34:32,876 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1075982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:34:51,295 INFO [train.py:968] (0/2) Epoch 24, batch 27100, giga_loss[loss=0.3233, simple_loss=0.3909, pruned_loss=0.1278, over 28870.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3803, pruned_loss=0.1259, over 5673671.87 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1126, over 5718923.36 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3816, pruned_loss=0.1266, over 5670446.60 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:34:52,662 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1076000.pt
+2023-03-12 13:35:30,523 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 2.144e+03 2.691e+03 3.370e+03 9.488e+03, threshold=5.382e+03, percent-clipped=13.0
+2023-03-12 13:35:40,659 INFO [train.py:968] (0/2) Epoch 24, batch 27150, giga_loss[loss=0.2775, simple_loss=0.3578, pruned_loss=0.09862, over 28945.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3794, pruned_loss=0.1262, over 5683360.70 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3595, pruned_loss=0.1129, over 5723908.84 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.381, pruned_loss=0.1269, over 5675287.94 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:35:43,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:35:51,652 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076059.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:35:54,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076062.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:36:04,190 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:36:17,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076085.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:36:18,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:36:23,235 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076091.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:36:31,979 INFO [train.py:968] (0/2) Epoch 24, batch 27200, giga_loss[loss=0.2919, simple_loss=0.3566, pruned_loss=0.1136, over 28870.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3793, pruned_loss=0.1264, over 5668636.80 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3597, pruned_loss=0.113, over 5719687.67 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.381, pruned_loss=0.1273, over 5664368.96 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:36:40,933 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:36:55,493 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:37:05,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.229e+03 1.744e+03 2.142e+03 2.872e+03 7.312e+03, threshold=4.283e+03, percent-clipped=3.0
+2023-03-12 13:37:15,975 INFO [train.py:968] (0/2) Epoch 24, batch 27250, giga_loss[loss=0.2872, simple_loss=0.369, pruned_loss=0.1027, over 28736.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3765, pruned_loss=0.1227, over 5669540.44 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3595, pruned_loss=0.113, over 5715128.88 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3784, pruned_loss=0.1237, over 5668706.68 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:37:54,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6644, 1.5900, 1.8756, 1.4783], device='cuda:0'), covar=tensor([0.1650, 0.2262, 0.1342, 0.1611], device='cuda:0'), in_proj_covar=tensor([0.0914, 0.0709, 0.0961, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 13:37:59,504 INFO [train.py:968] (0/2) Epoch 24, batch 27300, giga_loss[loss=0.2923, simple_loss=0.3688, pruned_loss=0.1079, over 28931.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3762, pruned_loss=0.1216, over 5664566.08 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3591, pruned_loss=0.1128, over 5718420.14 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3786, pruned_loss=0.1228, over 5659426.07 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:38:30,782 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:38:34,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:38:40,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.677e+02 1.611e+03 1.931e+03 2.827e+03 9.198e+03, threshold=3.863e+03, percent-clipped=9.0
+2023-03-12 13:38:47,884 INFO [train.py:968] (0/2) Epoch 24, batch 27350, giga_loss[loss=0.3457, simple_loss=0.4086, pruned_loss=0.1415, over 28662.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3774, pruned_loss=0.1224, over 5664339.53 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3589, pruned_loss=0.1128, over 5710436.37 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3802, pruned_loss=0.1237, over 5666262.25 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:38:57,989 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076258.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:39:03,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:39:10,758 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:39:14,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:39:35,601 INFO [train.py:968] (0/2) Epoch 24, batch 27400, giga_loss[loss=0.3105, simple_loss=0.3755, pruned_loss=0.1227, over 28955.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3783, pruned_loss=0.1237, over 5639302.69 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3591, pruned_loss=0.1131, over 5693667.10 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3811, pruned_loss=0.1249, over 5654132.56 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:39:40,215 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:39:55,409 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7174, 1.9034, 1.5050, 1.3960], device='cuda:0'), covar=tensor([0.0998, 0.0595, 0.0967, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0450, 0.0519, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 13:40:12,769 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-12 13:40:13,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.565e+03 1.955e+03 2.980e+03 6.849e+03, threshold=3.911e+03, percent-clipped=13.0
+2023-03-12 13:40:26,293 INFO [train.py:968] (0/2) Epoch 24, batch 27450, giga_loss[loss=0.3706, simple_loss=0.4239, pruned_loss=0.1586, over 27959.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3778, pruned_loss=0.1237, over 5647699.31 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3591, pruned_loss=0.1131, over 5695803.08 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3801, pruned_loss=0.1248, over 5657024.98 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:40:52,139 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-12 13:41:19,943 INFO [train.py:968] (0/2) Epoch 24, batch 27500, giga_loss[loss=0.3084, simple_loss=0.3761, pruned_loss=0.1203, over 28929.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3772, pruned_loss=0.1246, over 5646736.70 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3594, pruned_loss=0.1134, over 5686223.70 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3794, pruned_loss=0.1255, over 5662160.54 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 13:41:46,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:41:58,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.350e+02 1.843e+03 2.469e+03 3.539e+03 8.389e+03, threshold=4.938e+03, percent-clipped=21.0
+2023-03-12 13:42:00,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076440.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:42:09,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:42:11,609 INFO [train.py:968] (0/2) Epoch 24, batch 27550, giga_loss[loss=0.2717, simple_loss=0.3439, pruned_loss=0.09978, over 28311.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.374, pruned_loss=0.1229, over 5656674.78 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3588, pruned_loss=0.113, over 5691826.03 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3767, pruned_loss=0.1242, over 5663238.56 frames. ], batch size: 65, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 13:42:22,329 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5446, 1.8285, 1.4538, 1.7037], device='cuda:0'), covar=tensor([0.2592, 0.2621, 0.2933, 0.2418], device='cuda:0'), in_proj_covar=tensor([0.1546, 0.1115, 0.1363, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 13:42:22,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076460.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:42:48,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:43:01,561 INFO [train.py:968] (0/2) Epoch 24, batch 27600, giga_loss[loss=0.2911, simple_loss=0.3602, pruned_loss=0.111, over 28944.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.372, pruned_loss=0.1219, over 5654426.04 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3587, pruned_loss=0.1129, over 5694331.32 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3745, pruned_loss=0.1232, over 5656906.85 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:43:40,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.168e+03 2.093e+03 2.639e+03 3.974e+03 1.006e+04, threshold=5.278e+03, percent-clipped=18.0
+2023-03-12 13:43:49,101 INFO [train.py:968] (0/2) Epoch 24, batch 27650, giga_loss[loss=0.3048, simple_loss=0.3693, pruned_loss=0.1201, over 29055.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3726, pruned_loss=0.123, over 5664336.71 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3588, pruned_loss=0.1129, over 5700463.98 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3749, pruned_loss=0.1243, over 5659985.50 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:44:00,418 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-12 13:44:05,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6329, 1.7531, 1.4770, 1.6022], device='cuda:0'), covar=tensor([0.2857, 0.2899, 0.3266, 0.2458], device='cuda:0'), in_proj_covar=tensor([0.1550, 0.1119, 0.1365, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 13:44:12,106 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:44:14,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:44:31,591 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:44:33,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:44:37,244 INFO [train.py:968] (0/2) Epoch 24, batch 27700, giga_loss[loss=0.2709, simple_loss=0.3475, pruned_loss=0.09713, over 28750.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3714, pruned_loss=0.1223, over 5665669.86 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3592, pruned_loss=0.1132, over 5703952.53 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.373, pruned_loss=0.1232, over 5658080.73 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:44:37,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7269, 4.5590, 4.3093, 2.1319], device='cuda:0'), covar=tensor([0.0471, 0.0613, 0.0679, 0.2130], device='cuda:0'), in_proj_covar=tensor([0.1279, 0.1187, 0.1000, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 13:44:38,736 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:44:40,603 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076603.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:44:43,797 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076606.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:44:59,120 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:45:03,390 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:45:06,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:45:09,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076633.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:45:11,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076635.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:45:15,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.740e+02 1.697e+03 2.142e+03 2.948e+03 8.888e+03, threshold=4.284e+03, percent-clipped=4.0
+2023-03-12 13:45:24,608 INFO [train.py:968] (0/2) Epoch 24, batch 27750, giga_loss[loss=0.2486, simple_loss=0.3251, pruned_loss=0.08599, over 28674.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3691, pruned_loss=0.1193, over 5658312.53 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3595, pruned_loss=0.1134, over 5695991.77 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3704, pruned_loss=0.12, over 5658439.52 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:45:34,358 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:45:35,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0597, 2.0413, 1.5141, 1.7721], device='cuda:0'), covar=tensor([0.0958, 0.0752, 0.1032, 0.1188], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0451, 0.0520, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 13:46:14,701 INFO [train.py:968] (0/2) Epoch 24, batch 27800, giga_loss[loss=0.2687, simple_loss=0.3446, pruned_loss=0.09642, over 28719.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.366, pruned_loss=0.1164, over 5655201.68 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3599, pruned_loss=0.1137, over 5696015.54 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3668, pruned_loss=0.1169, over 5654958.56 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:46:52,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.626e+03 2.097e+03 2.966e+03 8.825e+03, threshold=4.195e+03, percent-clipped=6.0
+2023-03-12 13:47:02,725 INFO [train.py:968] (0/2) Epoch 24, batch 27850, giga_loss[loss=0.2764, simple_loss=0.3455, pruned_loss=0.1037, over 28864.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3651, pruned_loss=0.1162, over 5652714.86 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3595, pruned_loss=0.1135, over 5692517.90 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3663, pruned_loss=0.1168, over 5654024.23 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:47:37,209 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076776.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:47:40,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076779.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:48:00,535 INFO [train.py:968] (0/2) Epoch 24, batch 27900, giga_loss[loss=0.3088, simple_loss=0.3659, pruned_loss=0.1258, over 27489.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3636, pruned_loss=0.1166, over 5646041.91 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3598, pruned_loss=0.1135, over 5694086.88 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3644, pruned_loss=0.1171, over 5645101.15 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 13:48:11,140 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076808.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:48:17,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076815.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:48:30,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3494, 1.0967, 4.1167, 3.2783], device='cuda:0'), covar=tensor([0.1709, 0.2909, 0.0465, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0660, 0.0980, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 13:48:42,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.893e+03 2.412e+03 3.428e+03 1.191e+04, threshold=4.824e+03, percent-clipped=18.0
+2023-03-12 13:48:49,557 INFO [train.py:968] (0/2) Epoch 24, batch 27950, giga_loss[loss=0.264, simple_loss=0.3414, pruned_loss=0.09328, over 28397.00 frames. ], tot_loss[loss=0.298, simple_loss=0.363, pruned_loss=0.1165, over 5656499.43 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3599, pruned_loss=0.1135, over 5700974.45 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3637, pruned_loss=0.1169, over 5647930.17 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 13:49:39,884 INFO [train.py:968] (0/2) Epoch 24, batch 28000, libri_loss[loss=0.2541, simple_loss=0.3257, pruned_loss=0.0912, over 29684.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3655, pruned_loss=0.1171, over 5664785.42 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3597, pruned_loss=0.1134, over 5703126.83 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3663, pruned_loss=0.1178, over 5654738.63 frames. ], batch size: 73, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:49:46,550 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:50:18,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.171e+03 1.593e+03 2.116e+03 2.636e+03 6.339e+03, threshold=4.233e+03, percent-clipped=3.0
+2023-03-12 13:50:24,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5726, 1.3131, 4.3898, 3.3846], device='cuda:0'), covar=tensor([0.1583, 0.2867, 0.0380, 0.1110], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0660, 0.0979, 0.0941], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 13:50:29,151 INFO [train.py:968] (0/2) Epoch 24, batch 28050, giga_loss[loss=0.2772, simple_loss=0.3558, pruned_loss=0.09931, over 28756.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3667, pruned_loss=0.1174, over 5662360.73 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3601, pruned_loss=0.1135, over 5705656.88 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3672, pruned_loss=0.1179, over 5650946.67 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:50:37,997 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076958.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:50:40,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076961.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:51:07,843 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076990.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:51:14,186 INFO [train.py:968] (0/2) Epoch 24, batch 28100, giga_loss[loss=0.2833, simple_loss=0.3544, pruned_loss=0.1061, over 28887.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.367, pruned_loss=0.1178, over 5662304.10 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3605, pruned_loss=0.1137, over 5711485.23 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3672, pruned_loss=0.1182, over 5646650.81 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:51:32,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1077016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:51:53,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.590e+03 1.956e+03 2.870e+03 1.361e+04, threshold=3.912e+03, percent-clipped=11.0
+2023-03-12 13:52:00,140 INFO [train.py:968] (0/2) Epoch 24, batch 28150, giga_loss[loss=0.2824, simple_loss=0.3427, pruned_loss=0.1111, over 28611.00 frames. ], tot_loss[loss=0.302, simple_loss=0.367, pruned_loss=0.1185, over 5652163.18 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3605, pruned_loss=0.1137, over 5705685.16 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3673, pruned_loss=0.1188, over 5644153.20 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:52:47,513 INFO [train.py:968] (0/2) Epoch 24, batch 28200, giga_loss[loss=0.3022, simple_loss=0.3697, pruned_loss=0.1174, over 28967.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3689, pruned_loss=0.1201, over 5639860.68 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1142, over 5696061.04 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3689, pruned_loss=0.1201, over 5641300.96 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:53:30,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.999e+02 1.835e+03 2.550e+03 3.696e+03 7.098e+03, threshold=5.101e+03, percent-clipped=17.0
+2023-03-12 13:53:39,309 INFO [train.py:968] (0/2) Epoch 24, batch 28250, giga_loss[loss=0.2824, simple_loss=0.3527, pruned_loss=0.106, over 28887.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3708, pruned_loss=0.1211, over 5650205.18 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1142, over 5698932.60 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3708, pruned_loss=0.1211, over 5648242.22 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:53:47,296 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1077157.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:54:28,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5176, 1.1483, 4.7624, 3.6374], device='cuda:0'), covar=tensor([0.1736, 0.3011, 0.0399, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0664, 0.0983, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 13:54:29,840 INFO [train.py:968] (0/2) Epoch 24, batch 28300, giga_loss[loss=0.2656, simple_loss=0.3409, pruned_loss=0.09514, over 29125.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3715, pruned_loss=0.1219, over 5654538.75 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3609, pruned_loss=0.114, over 5704712.18 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.372, pruned_loss=0.1224, over 5646079.79 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:55:10,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.145e+03 1.723e+03 2.169e+03 2.669e+03 6.403e+03, threshold=4.338e+03, percent-clipped=4.0
+2023-03-12 13:55:20,004 INFO [train.py:968] (0/2) Epoch 24, batch 28350, giga_loss[loss=0.3474, simple_loss=0.3962, pruned_loss=0.1493, over 28313.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3731, pruned_loss=0.1239, over 5646667.64 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3608, pruned_loss=0.1138, over 5703543.10 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3738, pruned_loss=0.1245, over 5640138.45 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:55:54,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1077280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:56:13,440 INFO [train.py:968] (0/2) Epoch 24, batch 28400, giga_loss[loss=0.2697, simple_loss=0.3546, pruned_loss=0.09237, over 28887.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.374, pruned_loss=0.1231, over 5638560.24 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.1141, over 5695318.79 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3747, pruned_loss=0.1237, over 5639731.95 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:56:32,226 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-12 13:56:48,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6144, 1.9631, 1.9924, 1.4622], device='cuda:0'), covar=tensor([0.3164, 0.2460, 0.2559, 0.3038], device='cuda:0'), in_proj_covar=tensor([0.2009, 0.1957, 0.1867, 0.2011], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 13:56:55,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.148e+03 1.917e+03 2.495e+03 3.591e+03 7.796e+03, threshold=4.991e+03, percent-clipped=13.0
+2023-03-12 13:57:03,952 INFO [train.py:968] (0/2) Epoch 24, batch 28450, giga_loss[loss=0.3054, simple_loss=0.3792, pruned_loss=0.1158, over 28949.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3732, pruned_loss=0.1219, over 5643520.09 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.361, pruned_loss=0.1141, over 5691099.04 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.374, pruned_loss=0.1225, over 5647815.60 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:57:47,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1077391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:57:53,490 INFO [train.py:968] (0/2) Epoch 24, batch 28500, giga_loss[loss=0.3057, simple_loss=0.3689, pruned_loss=0.1212, over 28901.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3738, pruned_loss=0.1234, over 5617767.94 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3612, pruned_loss=0.1143, over 5681294.09 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3744, pruned_loss=0.1238, over 5629732.02 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:58:22,592 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1077423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:58:25,405 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1077426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:58:41,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.259e+03 1.807e+03 2.291e+03 3.121e+03 5.562e+03, threshold=4.583e+03, percent-clipped=4.0
+2023-03-12 13:58:52,796 INFO [train.py:968] (0/2) Epoch 24, batch 28550, giga_loss[loss=0.2869, simple_loss=0.3566, pruned_loss=0.1086, over 28886.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3731, pruned_loss=0.1235, over 5623987.97 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3611, pruned_loss=0.1143, over 5683613.32 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3738, pruned_loss=0.1239, over 5630810.84 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:59:02,435 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1077455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:59:15,974 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1077466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:59:33,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2103, 1.2539, 3.3795, 3.1866], device='cuda:0'), covar=tensor([0.1878, 0.2898, 0.0898, 0.2311], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0662, 0.0983, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 13:59:40,864 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 13:59:50,242 INFO [train.py:968] (0/2) Epoch 24, batch 28600, giga_loss[loss=0.3125, simple_loss=0.3714, pruned_loss=0.1268, over 28566.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3718, pruned_loss=0.1237, over 5617006.87 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1142, over 5685822.45 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3724, pruned_loss=0.1242, over 5619583.21 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:00:22,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1077532.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 14:00:24,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1077534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:00:27,094 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1077537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:00:28,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.587e+03 2.367e+03 3.209e+03 7.327e+03, threshold=4.734e+03, percent-clipped=6.0
+2023-03-12 14:00:41,308 INFO [train.py:968] (0/2) Epoch 24, batch 28650, giga_loss[loss=0.2895, simple_loss=0.353, pruned_loss=0.113, over 28582.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3708, pruned_loss=0.1233, over 5636766.36 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1143, over 5686754.34 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3716, pruned_loss=0.1238, over 5636972.51 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:00:58,651 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1077566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:01:28,838 INFO [train.py:968] (0/2) Epoch 24, batch 28700, giga_loss[loss=0.4532, simple_loss=0.4434, pruned_loss=0.2315, over 23307.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3722, pruned_loss=0.1249, over 5643075.95 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3618, pruned_loss=0.1147, over 5689277.57 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3724, pruned_loss=0.1251, over 5639998.70 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:02:14,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.822e+02 1.872e+03 2.234e+03 3.123e+03 9.325e+03, threshold=4.468e+03, percent-clipped=7.0
+2023-03-12 14:02:20,348 INFO [train.py:968] (0/2) Epoch 24, batch 28750, giga_loss[loss=0.268, simple_loss=0.3545, pruned_loss=0.09071, over 28842.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3723, pruned_loss=0.1248, over 5650788.35 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3618, pruned_loss=0.1147, over 5692713.57 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3726, pruned_loss=0.1251, over 5644544.93 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:02:26,415 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1077652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:02:31,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-12 14:02:51,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1077675.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 14:02:53,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1077678.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 14:03:03,874 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-12 14:03:14,650 INFO [train.py:968] (0/2) Epoch 24, batch 28800, giga_loss[loss=0.3062, simple_loss=0.3706, pruned_loss=0.1208, over 28856.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3726, pruned_loss=0.1249, over 5664414.78 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3619, pruned_loss=0.1147, over 5698161.87 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.373, pruned_loss=0.1255, over 5653596.42 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:03:20,989 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1077707.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 14:03:32,902 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1077720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:03:57,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.185e+03 1.889e+03 2.455e+03 3.158e+03 9.481e+03, threshold=4.909e+03, percent-clipped=10.0
+2023-03-12 14:04:03,170 INFO [train.py:968] (0/2) Epoch 24, batch 28850, giga_loss[loss=0.2658, simple_loss=0.3427, pruned_loss=0.09445, over 28783.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.374, pruned_loss=0.1259, over 5665152.13 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3616, pruned_loss=0.1144, over 5701159.12 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3749, pruned_loss=0.1268, over 5653067.34 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:04:51,839 INFO [train.py:968] (0/2) Epoch 24, batch 28900, giga_loss[loss=0.3508, simple_loss=0.402, pruned_loss=0.1498, over 28801.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3748, pruned_loss=0.1267, over 5671932.77 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3622, pruned_loss=0.1148, over 5703904.31 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3753, pruned_loss=0.1274, over 5659069.97 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:05:33,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.271e+03 2.024e+03 2.572e+03 4.006e+03 9.536e+03, threshold=5.144e+03, percent-clipped=12.0
+2023-03-12 14:05:33,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1077841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:05:39,692 INFO [train.py:968] (0/2) Epoch 24, batch 28950, giga_loss[loss=0.2749, simple_loss=0.3429, pruned_loss=0.1034, over 28756.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3742, pruned_loss=0.1263, over 5677700.08 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3626, pruned_loss=0.115, over 5706598.45 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3744, pruned_loss=0.1268, over 5664511.80 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:05:52,402 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-12 14:06:27,245 INFO [train.py:968] (0/2) Epoch 24, batch 29000, giga_loss[loss=0.2855, simple_loss=0.3594, pruned_loss=0.1058, over 28719.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3726, pruned_loss=0.1241, over 5679714.63 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3624, pruned_loss=0.115, over 5703575.50 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3732, pruned_loss=0.1248, over 5671802.58 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:06:38,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4091, 2.6401, 1.5053, 1.5264], device='cuda:0'), covar=tensor([0.0889, 0.0424, 0.0833, 0.1206], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0561, 0.0394, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 14:06:53,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5725, 1.5869, 1.2302, 1.2350], device='cuda:0'), covar=tensor([0.0712, 0.0369, 0.0785, 0.1040], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0453, 0.0525, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 14:07:08,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.209e+03 1.793e+03 2.163e+03 3.056e+03 7.531e+03, threshold=4.327e+03, percent-clipped=5.0
+2023-03-12 14:07:11,080 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-12 14:07:14,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6054, 1.8089, 1.4586, 1.7077], device='cuda:0'), covar=tensor([0.2710, 0.2894, 0.3222, 0.2482], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1119, 0.1364, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 14:07:16,338 INFO [train.py:968] (0/2) Epoch 24, batch 29050, giga_loss[loss=0.2725, simple_loss=0.3402, pruned_loss=0.1024, over 28744.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3742, pruned_loss=0.1254, over 5669081.63 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3629, pruned_loss=0.1154, over 5702188.52 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3746, pruned_loss=0.1259, over 5663098.93 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:07:37,588 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4730, 2.1397, 1.5633, 0.7362], device='cuda:0'), covar=tensor([0.6110, 0.3305, 0.4689, 0.7335], device='cuda:0'), in_proj_covar=tensor([0.1805, 0.1708, 0.1640, 0.1468], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 14:07:52,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1077984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:07:54,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1077987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:08:05,229 INFO [train.py:968] (0/2) Epoch 24, batch 29100, giga_loss[loss=0.4047, simple_loss=0.4349, pruned_loss=0.1873, over 26641.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3755, pruned_loss=0.1264, over 5663737.63 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3633, pruned_loss=0.1158, over 5693238.25 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3757, pruned_loss=0.1266, over 5667191.95 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:08:06,777 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1078000.pt
+2023-03-12 14:08:23,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:08:33,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1078027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:08:44,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.078e+03 1.761e+03 2.341e+03 3.149e+03 7.166e+03, threshold=4.683e+03, percent-clipped=10.0
+2023-03-12 14:08:51,865 INFO [train.py:968] (0/2) Epoch 24, batch 29150, giga_loss[loss=0.3462, simple_loss=0.4125, pruned_loss=0.1399, over 28875.00 frames. ], tot_loss[loss=0.316, simple_loss=0.377, pruned_loss=0.1275, over 5657822.86 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3632, pruned_loss=0.1159, over 5686732.69 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3774, pruned_loss=0.1278, over 5665891.74 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:09:17,067 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-12 14:09:22,758 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1078086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:09:31,905 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1078095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:09:33,529 INFO [train.py:968] (0/2) Epoch 24, batch 29200, giga_loss[loss=0.3028, simple_loss=0.3719, pruned_loss=0.1169, over 28950.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.377, pruned_loss=0.1276, over 5658802.73 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3635, pruned_loss=0.1162, over 5686913.10 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3777, pruned_loss=0.1281, over 5664049.96 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:09:59,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6824, 1.9409, 1.8775, 1.7527], device='cuda:0'), covar=tensor([0.2195, 0.2370, 0.2436, 0.2335], device='cuda:0'), in_proj_covar=tensor([0.0494, 0.0761, 0.0728, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 14:10:13,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.439e+02 1.753e+03 2.270e+03 2.843e+03 9.864e+03, threshold=4.540e+03, percent-clipped=4.0
+2023-03-12 14:10:17,762 INFO [train.py:968] (0/2) Epoch 24, batch 29250, libri_loss[loss=0.3165, simple_loss=0.3728, pruned_loss=0.1301, over 29566.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3762, pruned_loss=0.1264, over 5661413.48 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3629, pruned_loss=0.1159, over 5693985.31 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3778, pruned_loss=0.1275, over 5657850.68 frames. ], batch size: 76, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:10:40,205 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1078170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:10:43,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1078173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:10:46,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4452, 3.1562, 1.4692, 1.6346], device='cuda:0'), covar=tensor([0.0965, 0.0387, 0.0956, 0.1253], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0563, 0.0396, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 14:11:08,180 INFO [train.py:968] (0/2) Epoch 24, batch 29300, giga_loss[loss=0.3568, simple_loss=0.4003, pruned_loss=0.1567, over 26553.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3772, pruned_loss=0.1264, over 5650409.55 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3629, pruned_loss=0.1159, over 5697911.96 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3789, pruned_loss=0.1277, over 5642897.30 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:11:12,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:11:26,193 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.47 vs. limit=5.0
+2023-03-12 14:11:48,789 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1078238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:11:50,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1078241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:11:51,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.718e+03 2.237e+03 2.886e+03 5.495e+03, threshold=4.474e+03, percent-clipped=3.0
+2023-03-12 14:11:57,017 INFO [train.py:968] (0/2) Epoch 24, batch 29350, giga_loss[loss=0.2651, simple_loss=0.3383, pruned_loss=0.09593, over 28865.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3766, pruned_loss=0.1257, over 5650753.51 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.363, pruned_loss=0.1159, over 5699239.57 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3781, pruned_loss=0.1268, over 5642978.00 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:11:57,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1168, 1.2674, 1.0626, 0.8620], device='cuda:0'), covar=tensor([0.1043, 0.0559, 0.1119, 0.1176], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0455, 0.0526, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 14:12:15,185 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:12:26,106 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1078281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:12:40,426 INFO [train.py:968] (0/2) Epoch 24, batch 29400, giga_loss[loss=0.2964, simple_loss=0.3588, pruned_loss=0.117, over 28970.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3749, pruned_loss=0.1242, over 5651152.68 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3634, pruned_loss=0.1161, over 5686673.03 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3762, pruned_loss=0.1252, over 5653344.09 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:13:13,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5938, 1.8345, 1.4591, 1.8617], device='cuda:0'), covar=tensor([0.2665, 0.2821, 0.3126, 0.2555], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1119, 0.1365, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 14:13:17,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.809e+03 2.420e+03 3.640e+03 1.095e+04, threshold=4.840e+03, percent-clipped=12.0
+2023-03-12 14:13:22,578 INFO [train.py:968] (0/2) Epoch 24, batch 29450, giga_loss[loss=0.2789, simple_loss=0.3474, pruned_loss=0.1052, over 28499.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.373, pruned_loss=0.1231, over 5649777.53 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3633, pruned_loss=0.1161, over 5682098.74 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3745, pruned_loss=0.1242, over 5653950.67 frames. ], batch size: 78, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:13:51,395 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1078377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:14:10,843 INFO [train.py:968] (0/2) Epoch 24, batch 29500, giga_loss[loss=0.3109, simple_loss=0.3831, pruned_loss=0.1194, over 28873.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1235, over 5633506.69 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3632, pruned_loss=0.1161, over 5666318.78 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3752, pruned_loss=0.1246, over 5649514.36 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:14:53,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.536e+03 1.885e+03 2.712e+03 7.836e+03, threshold=3.769e+03, percent-clipped=4.0
+2023-03-12 14:15:01,672 INFO [train.py:968] (0/2) Epoch 24, batch 29550, giga_loss[loss=0.2804, simple_loss=0.3531, pruned_loss=0.1039, over 29048.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3737, pruned_loss=0.1238, over 5637979.60 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3632, pruned_loss=0.1161, over 5664802.41 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3754, pruned_loss=0.125, over 5651305.30 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:15:04,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6913, 1.8703, 1.3099, 1.4342], device='cuda:0'), covar=tensor([0.0967, 0.0643, 0.1071, 0.1208], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0453, 0.0523, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 14:15:12,912 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1078461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:15:50,206 INFO [train.py:968] (0/2) Epoch 24, batch 29600, giga_loss[loss=0.3391, simple_loss=0.3841, pruned_loss=0.1471, over 27583.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.373, pruned_loss=0.1242, over 5629593.24 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3633, pruned_loss=0.1163, over 5652214.03 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3744, pruned_loss=0.1252, over 5651513.04 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:15:53,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4516, 1.7277, 1.8222, 1.5209], device='cuda:0'), covar=tensor([0.1808, 0.1608, 0.1836, 0.1697], device='cuda:0'), in_proj_covar=tensor([0.0494, 0.0759, 0.0727, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 14:15:56,502 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-12 14:16:33,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 1.751e+03 2.750e+03 3.510e+03 9.679e+03, threshold=5.500e+03, percent-clipped=19.0
+2023-03-12 14:16:37,494 INFO [train.py:968] (0/2) Epoch 24, batch 29650, giga_loss[loss=0.299, simple_loss=0.3609, pruned_loss=0.1185, over 29072.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3742, pruned_loss=0.1252, over 5645637.61 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3638, pruned_loss=0.1165, over 5654656.77 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3752, pruned_loss=0.1259, over 5660532.97 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:17:25,756 INFO [train.py:968] (0/2) Epoch 24, batch 29700, giga_loss[loss=0.3039, simple_loss=0.369, pruned_loss=0.1194, over 28855.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3754, pruned_loss=0.126, over 5654076.75 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3638, pruned_loss=0.1165, over 5659421.19 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3764, pruned_loss=0.1266, over 5661645.64 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:17:32,260 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1078604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:17:37,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1078607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:18:08,494 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:18:15,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.793e+03 2.211e+03 2.956e+03 7.612e+03, threshold=4.422e+03, percent-clipped=4.0
+2023-03-12 14:18:18,628 INFO [train.py:968] (0/2) Epoch 24, batch 29750, giga_loss[loss=0.3095, simple_loss=0.3717, pruned_loss=0.1237, over 27926.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3749, pruned_loss=0.1257, over 5641720.71 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3637, pruned_loss=0.1165, over 5660387.08 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3757, pruned_loss=0.1263, over 5646766.95 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:18:27,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1078656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:19:06,794 INFO [train.py:968] (0/2) Epoch 24, batch 29800, giga_loss[loss=0.2867, simple_loss=0.3609, pruned_loss=0.1063, over 29022.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3744, pruned_loss=0.1251, over 5629841.11 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3639, pruned_loss=0.1165, over 5652650.48 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.375, pruned_loss=0.1257, over 5640791.05 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:19:09,389 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6206, 1.8354, 1.5154, 1.8090], device='cuda:0'), covar=tensor([0.2234, 0.2324, 0.2473, 0.2265], device='cuda:0'), in_proj_covar=tensor([0.1544, 0.1114, 0.1360, 0.0997], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 14:19:50,311 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.346e+02 1.564e+03 1.936e+03 2.733e+03 7.012e+03, threshold=3.872e+03, percent-clipped=3.0
+2023-03-12 14:19:55,265 INFO [train.py:968] (0/2) Epoch 24, batch 29850, giga_loss[loss=0.3474, simple_loss=0.4119, pruned_loss=0.1415, over 28352.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3732, pruned_loss=0.1233, over 5621059.72 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3635, pruned_loss=0.1163, over 5629797.58 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3743, pruned_loss=0.1241, over 5648243.12 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:19:59,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1078752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:20:00,247 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 14:20:22,274 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3595, 3.2429, 1.5038, 1.4458], device='cuda:0'), covar=tensor([0.0992, 0.0397, 0.0915, 0.1355], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0564, 0.0395, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 14:20:45,044 INFO [train.py:968] (0/2) Epoch 24, batch 29900, giga_loss[loss=0.2659, simple_loss=0.3453, pruned_loss=0.09321, over 28656.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3728, pruned_loss=0.1226, over 5629499.77 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3636, pruned_loss=0.1163, over 5633550.27 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3738, pruned_loss=0.1234, over 5647945.57 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:20:47,817 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1078799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:20:50,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1078802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:21:05,154 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-12 14:21:11,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3081, 3.1535, 2.9946, 1.3889], device='cuda:0'), covar=tensor([0.0968, 0.1039, 0.0939, 0.2261], device='cuda:0'), in_proj_covar=tensor([0.1290, 0.1194, 0.1007, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 14:21:17,135 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:21:30,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.328e+03 1.821e+03 2.449e+03 3.688e+03 1.207e+04, threshold=4.897e+03, percent-clipped=17.0
+2023-03-12 14:21:35,791 INFO [train.py:968] (0/2) Epoch 24, batch 29950, giga_loss[loss=0.2783, simple_loss=0.3445, pruned_loss=0.106, over 28267.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.372, pruned_loss=0.1218, over 5648713.69 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3636, pruned_loss=0.1163, over 5637218.97 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3728, pruned_loss=0.1225, over 5660031.29 frames. ], batch size: 77, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:22:17,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5071, 2.1230, 1.5164, 0.6694], device='cuda:0'), covar=tensor([0.5878, 0.3142, 0.4369, 0.6669], device='cuda:0'), in_proj_covar=tensor([0.1802, 0.1702, 0.1636, 0.1463], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 14:22:17,416 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1078895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:22:20,586 INFO [train.py:968] (0/2) Epoch 24, batch 30000, giga_loss[loss=0.2752, simple_loss=0.3434, pruned_loss=0.1035, over 28938.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3707, pruned_loss=0.1212, over 5661366.83 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3636, pruned_loss=0.1162, over 5644939.78 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3716, pruned_loss=0.122, over 5663689.36 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:22:20,591 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 14:22:30,353 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2045, simple_loss=0.313, pruned_loss=0.04803, over 944034.00 frames. 
+2023-03-12 14:22:30,354 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 14:22:30,737 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1078898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:22:54,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:23:03,143 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1078937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:23:09,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.871e+03 2.322e+03 3.251e+03 7.688e+03, threshold=4.645e+03, percent-clipped=10.0
+2023-03-12 14:23:12,988 INFO [train.py:968] (0/2) Epoch 24, batch 30050, giga_loss[loss=0.2868, simple_loss=0.3331, pruned_loss=0.1202, over 23539.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3675, pruned_loss=0.1199, over 5660859.40 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.363, pruned_loss=0.1158, over 5655615.12 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.369, pruned_loss=0.1211, over 5653808.16 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:23:58,367 INFO [train.py:968] (0/2) Epoch 24, batch 30100, giga_loss[loss=0.2733, simple_loss=0.3343, pruned_loss=0.1062, over 28649.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3644, pruned_loss=0.1192, over 5660273.28 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3624, pruned_loss=0.1154, over 5660134.86 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3662, pruned_loss=0.1206, over 5650724.87 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:24:15,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0157, 2.1183, 2.0413, 1.8548], device='cuda:0'), covar=tensor([0.2014, 0.2513, 0.2165, 0.2386], device='cuda:0'), in_proj_covar=tensor([0.0492, 0.0756, 0.0725, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 14:24:38,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.148e+03 1.874e+03 2.344e+03 3.198e+03 6.373e+03, threshold=4.688e+03, percent-clipped=8.0
+2023-03-12 14:24:42,798 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7142, 3.5783, 3.3982, 1.8951], device='cuda:0'), covar=tensor([0.0736, 0.0834, 0.0776, 0.2316], device='cuda:0'), in_proj_covar=tensor([0.1290, 0.1193, 0.1006, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 14:24:43,294 INFO [train.py:968] (0/2) Epoch 24, batch 30150, giga_loss[loss=0.4025, simple_loss=0.4292, pruned_loss=0.1879, over 26563.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3642, pruned_loss=0.1196, over 5643647.99 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3627, pruned_loss=0.1157, over 5640190.12 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3654, pruned_loss=0.1206, over 5653801.96 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:25:33,143 INFO [train.py:968] (0/2) Epoch 24, batch 30200, giga_loss[loss=0.2968, simple_loss=0.3495, pruned_loss=0.122, over 24296.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3638, pruned_loss=0.1196, over 5626604.17 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3629, pruned_loss=0.1159, over 5638081.67 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3647, pruned_loss=0.1204, over 5636797.38 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:25:34,601 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:25:45,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5749, 1.7228, 1.7807, 1.3644], device='cuda:0'), covar=tensor([0.1845, 0.2774, 0.1594, 0.1936], device='cuda:0'), in_proj_covar=tensor([0.0914, 0.0708, 0.0959, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 14:26:14,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.719e+03 2.244e+03 3.507e+03 8.814e+03, threshold=4.487e+03, percent-clipped=10.0
+2023-03-12 14:26:20,787 INFO [train.py:968] (0/2) Epoch 24, batch 30250, giga_loss[loss=0.2575, simple_loss=0.3214, pruned_loss=0.09677, over 24184.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3619, pruned_loss=0.1165, over 5627944.56 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3624, pruned_loss=0.1157, over 5632869.75 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3631, pruned_loss=0.1173, over 5641220.66 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:26:25,135 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 14:26:51,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2879, 1.3140, 1.2083, 1.5411], device='cuda:0'), covar=tensor([0.0792, 0.0380, 0.0365, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 14:27:14,109 INFO [train.py:968] (0/2) Epoch 24, batch 30300, giga_loss[loss=0.2541, simple_loss=0.3365, pruned_loss=0.08589, over 28324.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3591, pruned_loss=0.1127, over 5627919.14 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3622, pruned_loss=0.1158, over 5636088.49 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1132, over 5635408.68 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:27:25,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6420, 1.8868, 1.8030, 1.5354], device='cuda:0'), covar=tensor([0.1793, 0.2072, 0.2163, 0.2116], device='cuda:0'), in_proj_covar=tensor([0.0490, 0.0753, 0.0722, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 14:28:00,650 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.503e+03 1.820e+03 2.736e+03 1.484e+04, threshold=3.639e+03, percent-clipped=7.0
+2023-03-12 14:28:04,690 INFO [train.py:968] (0/2) Epoch 24, batch 30350, giga_loss[loss=0.2532, simple_loss=0.3344, pruned_loss=0.08596, over 28954.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3558, pruned_loss=0.1088, over 5649410.17 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3615, pruned_loss=0.1155, over 5645539.70 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3571, pruned_loss=0.1094, over 5646709.76 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:28:07,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8536, 5.6656, 5.4067, 3.0129], device='cuda:0'), covar=tensor([0.0558, 0.0766, 0.1005, 0.1589], device='cuda:0'), in_proj_covar=tensor([0.1285, 0.1188, 0.1002, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 14:28:15,346 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:28:17,194 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-12 14:28:23,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-12 14:28:53,841 INFO [train.py:968] (0/2) Epoch 24, batch 30400, giga_loss[loss=0.2643, simple_loss=0.3449, pruned_loss=0.09185, over 28900.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3529, pruned_loss=0.1061, over 5653663.32 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3608, pruned_loss=0.1153, over 5651920.00 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3545, pruned_loss=0.1065, over 5645744.60 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:29:07,633 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:29:24,123 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:29:32,069 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:29:40,245 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 14:29:41,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.705e+02 1.374e+03 1.781e+03 2.531e+03 5.922e+03, threshold=3.561e+03, percent-clipped=7.0
+2023-03-12 14:29:43,845 INFO [train.py:968] (0/2) Epoch 24, batch 30450, giga_loss[loss=0.2513, simple_loss=0.3395, pruned_loss=0.08156, over 28850.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3499, pruned_loss=0.1027, over 5655491.54 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3604, pruned_loss=0.1152, over 5652157.35 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3513, pruned_loss=0.103, over 5648896.37 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:29:53,646 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2704, 4.1020, 3.8825, 1.8380], device='cuda:0'), covar=tensor([0.0652, 0.0816, 0.0869, 0.2222], device='cuda:0'), in_proj_covar=tensor([0.1277, 0.1184, 0.0996, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 14:30:29,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4862, 1.6382, 1.7394, 1.3182], device='cuda:0'), covar=tensor([0.1955, 0.2767, 0.1645, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.0913, 0.0705, 0.0958, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 14:30:38,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9790, 1.2292, 2.8212, 2.8116], device='cuda:0'), covar=tensor([0.1649, 0.2677, 0.0635, 0.1344], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0661, 0.0984, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 14:30:38,946 INFO [train.py:968] (0/2) Epoch 24, batch 30500, giga_loss[loss=0.2794, simple_loss=0.3604, pruned_loss=0.09925, over 28531.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3491, pruned_loss=0.0997, over 5669808.23 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3602, pruned_loss=0.1151, over 5656774.70 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3502, pruned_loss=0.09972, over 5660643.82 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:31:30,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3180, 1.3208, 3.6217, 3.1782], device='cuda:0'), covar=tensor([0.1607, 0.2791, 0.0474, 0.1036], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0660, 0.0981, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 14:31:31,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.851e+02 1.477e+03 2.005e+03 2.755e+03 1.290e+04, threshold=4.010e+03, percent-clipped=11.0
+2023-03-12 14:31:34,365 INFO [train.py:968] (0/2) Epoch 24, batch 30550, giga_loss[loss=0.2924, simple_loss=0.3642, pruned_loss=0.1103, over 28642.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3502, pruned_loss=0.1005, over 5663703.99 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3604, pruned_loss=0.1153, over 5650686.39 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3508, pruned_loss=0.1001, over 5661392.21 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:31:41,814 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1079455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:31:47,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1079458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:32:02,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:32:14,479 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1079487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:32:25,004 INFO [train.py:968] (0/2) Epoch 24, batch 30600, libri_loss[loss=0.3531, simple_loss=0.3924, pruned_loss=0.1569, over 20069.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3493, pruned_loss=0.09998, over 5657236.93 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3601, pruned_loss=0.1154, over 5642582.04 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3496, pruned_loss=0.09902, over 5663785.66 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:32:35,279 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:32:37,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2968, 3.0689, 1.4318, 1.5538], device='cuda:0'), covar=tensor([0.1145, 0.0446, 0.1038, 0.1427], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0562, 0.0394, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 14:33:11,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.416e+02 1.485e+03 2.101e+03 2.992e+03 7.089e+03, threshold=4.202e+03, percent-clipped=6.0
+2023-03-12 14:33:14,100 INFO [train.py:968] (0/2) Epoch 24, batch 30650, giga_loss[loss=0.2321, simple_loss=0.3134, pruned_loss=0.07538, over 28609.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3468, pruned_loss=0.09837, over 5661564.89 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.36, pruned_loss=0.1154, over 5648009.86 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3469, pruned_loss=0.0973, over 5662419.02 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:33:17,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4858, 5.3500, 5.0437, 2.8052], device='cuda:0'), covar=tensor([0.0449, 0.0616, 0.0780, 0.1539], device='cuda:0'), in_proj_covar=tensor([0.1270, 0.1176, 0.0990, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 14:34:05,460 INFO [train.py:968] (0/2) Epoch 24, batch 30700, giga_loss[loss=0.3024, simple_loss=0.375, pruned_loss=0.1149, over 28197.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3459, pruned_loss=0.098, over 5652715.49 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3595, pruned_loss=0.1152, over 5645927.65 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3461, pruned_loss=0.09684, over 5655017.71 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:34:25,799 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1079618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:34:29,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1079621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:34:41,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:34:52,646 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.227e+02 1.437e+03 1.903e+03 2.380e+03 5.803e+03, threshold=3.806e+03, percent-clipped=1.0
+2023-03-12 14:34:55,410 INFO [train.py:968] (0/2) Epoch 24, batch 30750, giga_loss[loss=0.2986, simple_loss=0.3654, pruned_loss=0.1159, over 28638.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3459, pruned_loss=0.09791, over 5660330.10 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3593, pruned_loss=0.1153, over 5648103.88 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3461, pruned_loss=0.09676, over 5660384.46 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:34:57,837 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1079650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:35:45,958 INFO [train.py:968] (0/2) Epoch 24, batch 30800, giga_loss[loss=0.2589, simple_loss=0.3402, pruned_loss=0.08882, over 28505.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3441, pruned_loss=0.09643, over 5657362.98 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3594, pruned_loss=0.1153, over 5650580.83 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3438, pruned_loss=0.09504, over 5655212.48 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 14:35:48,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:36:01,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:36:08,655 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9146, 1.1877, 1.0929, 0.8554], device='cuda:0'), covar=tensor([0.2138, 0.2151, 0.1458, 0.2043], device='cuda:0'), in_proj_covar=tensor([0.1991, 0.1938, 0.1852, 0.1998], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 14:36:32,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.198e+02 1.388e+03 2.066e+03 2.667e+03 4.695e+03, threshold=4.131e+03, percent-clipped=7.0
+2023-03-12 14:36:37,478 INFO [train.py:968] (0/2) Epoch 24, batch 30850, giga_loss[loss=0.2369, simple_loss=0.3191, pruned_loss=0.07738, over 28548.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3409, pruned_loss=0.09408, over 5661992.27 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3589, pruned_loss=0.1151, over 5653032.97 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3408, pruned_loss=0.09275, over 5658005.87 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 14:37:04,861 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:37:07,025 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1079777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:37:10,840 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1079780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:37:31,928 INFO [train.py:968] (0/2) Epoch 24, batch 30900, giga_loss[loss=0.2266, simple_loss=0.3101, pruned_loss=0.07151, over 28455.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3373, pruned_loss=0.09192, over 5671763.74 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3588, pruned_loss=0.1151, over 5656911.53 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.337, pruned_loss=0.09059, over 5665496.64 frames. ], batch size: 60, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:37:43,824 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1079809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:17,496 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1079844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:18,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.021e+02 1.488e+03 2.042e+03 2.970e+03 6.774e+03, threshold=4.084e+03, percent-clipped=10.0
+2023-03-12 14:38:20,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1079847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:20,693 INFO [train.py:968] (0/2) Epoch 24, batch 30950, giga_loss[loss=0.2216, simple_loss=0.3084, pruned_loss=0.06736, over 28722.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3356, pruned_loss=0.09123, over 5666558.12 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3584, pruned_loss=0.1149, over 5656990.42 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3352, pruned_loss=0.08982, over 5661759.99 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:38:27,366 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1079853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:29,970 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1079856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:48,879 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:49,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1079876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:59,568 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:39:01,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1079885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:39:04,060 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-12 14:39:16,932 INFO [train.py:968] (0/2) Epoch 24, batch 31000, giga_loss[loss=0.2367, simple_loss=0.3221, pruned_loss=0.0756, over 28342.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3357, pruned_loss=0.09163, over 5651661.39 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3583, pruned_loss=0.1148, over 5659612.29 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3352, pruned_loss=0.09028, over 5645681.39 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:40:07,960 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.205e+02 1.534e+03 1.914e+03 2.577e+03 9.232e+03, threshold=3.828e+03, percent-clipped=12.0
+2023-03-12 14:40:10,436 INFO [train.py:968] (0/2) Epoch 24, batch 31050, giga_loss[loss=0.3082, simple_loss=0.3718, pruned_loss=0.1223, over 27665.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3382, pruned_loss=0.09297, over 5654284.40 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3576, pruned_loss=0.1145, over 5666823.54 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3378, pruned_loss=0.09151, over 5642691.77 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:40:25,685 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-12 14:40:52,285 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-12 14:41:06,531 INFO [train.py:968] (0/2) Epoch 24, batch 31100, libri_loss[loss=0.287, simple_loss=0.3551, pruned_loss=0.1094, over 29532.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3392, pruned_loss=0.09295, over 5640868.77 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3573, pruned_loss=0.1145, over 5665433.35 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3385, pruned_loss=0.09094, over 5632670.34 frames. ], batch size: 79, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:41:08,966 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1080000.pt
+2023-03-12 14:41:40,031 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:41:43,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:41:59,887 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:42:06,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.632e+03 2.022e+03 2.714e+03 8.181e+03, threshold=4.044e+03, percent-clipped=8.0
+2023-03-12 14:42:07,684 INFO [train.py:968] (0/2) Epoch 24, batch 31150, giga_loss[loss=0.2467, simple_loss=0.3257, pruned_loss=0.08386, over 28637.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3397, pruned_loss=0.09335, over 5638435.82 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3567, pruned_loss=0.1142, over 5671540.85 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3393, pruned_loss=0.09151, over 5625722.00 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:42:21,560 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:42:51,987 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:43:09,638 INFO [train.py:968] (0/2) Epoch 24, batch 31200, giga_loss[loss=0.2222, simple_loss=0.3082, pruned_loss=0.06809, over 29038.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3382, pruned_loss=0.09267, over 5651290.42 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3562, pruned_loss=0.1141, over 5677390.56 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3378, pruned_loss=0.09059, over 5634632.72 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:43:44,588 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9677, 1.1650, 1.1294, 0.9293], device='cuda:0'), covar=tensor([0.2361, 0.2543, 0.1475, 0.2081], device='cuda:0'), in_proj_covar=tensor([0.1974, 0.1918, 0.1829, 0.1974], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 14:44:06,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.796e+02 1.615e+03 2.316e+03 3.440e+03 1.650e+04, threshold=4.632e+03, percent-clipped=23.0
+2023-03-12 14:44:07,587 INFO [train.py:968] (0/2) Epoch 24, batch 31250, giga_loss[loss=0.2344, simple_loss=0.3249, pruned_loss=0.072, over 27597.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3377, pruned_loss=0.0915, over 5640455.69 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3564, pruned_loss=0.1144, over 5671740.08 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3366, pruned_loss=0.08893, over 5632263.06 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:44:11,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:44:21,510 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-12 14:45:13,257 INFO [train.py:968] (0/2) Epoch 24, batch 31300, giga_loss[loss=0.2691, simple_loss=0.3292, pruned_loss=0.1045, over 24663.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3353, pruned_loss=0.08907, over 5639766.27 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3558, pruned_loss=0.1142, over 5674297.36 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3347, pruned_loss=0.08675, over 5630241.14 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:45:48,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-12 14:45:55,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3766, 1.3712, 3.7895, 3.1540], device='cuda:0'), covar=tensor([0.1620, 0.2809, 0.0427, 0.1009], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0655, 0.0969, 0.0928], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 14:46:17,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.894e+02 1.384e+03 2.060e+03 3.075e+03 1.013e+04, threshold=4.121e+03, percent-clipped=7.0
+2023-03-12 14:46:18,976 INFO [train.py:968] (0/2) Epoch 24, batch 31350, giga_loss[loss=0.2216, simple_loss=0.306, pruned_loss=0.06855, over 28633.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3326, pruned_loss=0.08856, over 5658543.78 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3554, pruned_loss=0.1139, over 5677869.05 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3322, pruned_loss=0.08658, over 5647314.69 frames. ], batch size: 60, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:46:24,957 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:46:48,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7482, 1.9093, 1.5749, 1.9941], device='cuda:0'), covar=tensor([0.2849, 0.2858, 0.3247, 0.2581], device='cuda:0'), in_proj_covar=tensor([0.1558, 0.1119, 0.1374, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 14:46:59,746 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4671, 1.8686, 1.4631, 1.2748], device='cuda:0'), covar=tensor([0.2608, 0.2545, 0.2979, 0.2431], device='cuda:0'), in_proj_covar=tensor([0.1558, 0.1119, 0.1374, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 14:47:16,752 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:47:19,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:47:21,563 INFO [train.py:968] (0/2) Epoch 24, batch 31400, giga_loss[loss=0.2399, simple_loss=0.3266, pruned_loss=0.0766, over 28916.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3325, pruned_loss=0.0885, over 5669637.86 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3551, pruned_loss=0.1138, over 5679654.06 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.332, pruned_loss=0.08668, over 5658999.34 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:47:24,263 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 14:47:43,073 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:47:57,365 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:48:24,067 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.434e+02 1.539e+03 2.042e+03 2.869e+03 1.299e+04, threshold=4.085e+03, percent-clipped=10.0
+2023-03-12 14:48:26,085 INFO [train.py:968] (0/2) Epoch 24, batch 31450, giga_loss[loss=0.2306, simple_loss=0.3218, pruned_loss=0.06976, over 28982.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3327, pruned_loss=0.08855, over 5670159.34 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3548, pruned_loss=0.1137, over 5681887.16 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3324, pruned_loss=0.0869, over 5659666.19 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:48:45,214 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:49:21,632 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:49:24,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:49:26,819 INFO [train.py:968] (0/2) Epoch 24, batch 31500, giga_loss[loss=0.2782, simple_loss=0.3563, pruned_loss=0.1, over 27679.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3337, pruned_loss=0.08822, over 5658295.77 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3548, pruned_loss=0.1138, over 5684357.60 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3332, pruned_loss=0.08653, over 5647552.09 frames. ], batch size: 474, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:49:35,255 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1587, 1.4722, 1.4425, 1.0527], device='cuda:0'), covar=tensor([0.1698, 0.2698, 0.1422, 0.1839], device='cuda:0'), in_proj_covar=tensor([0.0916, 0.0704, 0.0961, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 14:49:49,939 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:49:54,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:50:04,099 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:50:30,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.123e+02 1.481e+03 1.919e+03 3.019e+03 7.863e+03, threshold=3.837e+03, percent-clipped=11.0
+2023-03-12 14:50:33,173 INFO [train.py:968] (0/2) Epoch 24, batch 31550, giga_loss[loss=0.2283, simple_loss=0.3141, pruned_loss=0.0713, over 28532.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3322, pruned_loss=0.08707, over 5666202.13 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3544, pruned_loss=0.1137, over 5684338.86 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3318, pruned_loss=0.08552, over 5657477.28 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:50:48,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:51:40,595 INFO [train.py:968] (0/2) Epoch 24, batch 31600, giga_loss[loss=0.28, simple_loss=0.3551, pruned_loss=0.1024, over 28715.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.332, pruned_loss=0.08757, over 5680617.64 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3538, pruned_loss=0.1135, over 5690809.39 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3315, pruned_loss=0.08562, over 5667395.86 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 14:52:28,050 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:52:45,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.971e+02 1.519e+03 1.842e+03 2.661e+03 6.221e+03, threshold=3.683e+03, percent-clipped=6.0
+2023-03-12 14:52:46,725 INFO [train.py:968] (0/2) Epoch 24, batch 31650, giga_loss[loss=0.2705, simple_loss=0.3636, pruned_loss=0.08874, over 28493.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.334, pruned_loss=0.08858, over 5676200.82 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3538, pruned_loss=0.1136, over 5695086.86 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.333, pruned_loss=0.08624, over 5661037.82 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 14:52:56,789 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.39 vs. limit=5.0
+2023-03-12 14:53:00,042 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:53:00,134 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:53:03,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:53:40,893 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:53:46,368 INFO [train.py:968] (0/2) Epoch 24, batch 31700, giga_loss[loss=0.2575, simple_loss=0.3584, pruned_loss=0.07826, over 28699.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.338, pruned_loss=0.08858, over 5682413.30 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3536, pruned_loss=0.1135, over 5696503.69 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3367, pruned_loss=0.08569, over 5668344.55 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:53:51,992 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:53:54,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:54:35,222 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:54:43,927 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8432, 1.2115, 1.2813, 1.0434], device='cuda:0'), covar=tensor([0.1988, 0.1385, 0.2181, 0.1693], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0743, 0.0711, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 14:54:51,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.731e+02 1.466e+03 1.895e+03 2.557e+03 7.105e+03, threshold=3.789e+03, percent-clipped=6.0
+2023-03-12 14:54:51,539 INFO [train.py:968] (0/2) Epoch 24, batch 31750, giga_loss[loss=0.2339, simple_loss=0.3312, pruned_loss=0.06828, over 28653.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3385, pruned_loss=0.0872, over 5667328.80 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3534, pruned_loss=0.1135, over 5698286.58 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3374, pruned_loss=0.08472, over 5654497.10 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:55:20,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0556, 1.2712, 1.0823, 0.9055], device='cuda:0'), covar=tensor([0.1098, 0.0476, 0.1101, 0.1041], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0446, 0.0519, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 14:55:28,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6256, 1.7521, 1.4852, 1.7688], device='cuda:0'), covar=tensor([0.3052, 0.3068, 0.3320, 0.2829], device='cuda:0'), in_proj_covar=tensor([0.1557, 0.1120, 0.1372, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 14:55:34,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6228, 2.0137, 1.2385, 1.5174], device='cuda:0'), covar=tensor([0.1122, 0.0583, 0.1159, 0.1154], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0446, 0.0519, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 14:55:45,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:55:52,794 INFO [train.py:968] (0/2) Epoch 24, batch 31800, giga_loss[loss=0.2265, simple_loss=0.3229, pruned_loss=0.06505, over 28621.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3389, pruned_loss=0.08638, over 5672678.56 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3531, pruned_loss=0.1133, over 5702398.82 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3381, pruned_loss=0.08412, over 5658065.35 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:55:53,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6226, 2.1724, 1.4350, 1.0445], device='cuda:0'), covar=tensor([0.8417, 0.4256, 0.4075, 0.6874], device='cuda:0'), in_proj_covar=tensor([0.1784, 0.1679, 0.1620, 0.1453], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 14:56:49,534 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:56:58,916 INFO [train.py:968] (0/2) Epoch 24, batch 31850, giga_loss[loss=0.2845, simple_loss=0.3622, pruned_loss=0.1035, over 27522.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3395, pruned_loss=0.08706, over 5675132.40 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.353, pruned_loss=0.1133, over 5697448.59 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3387, pruned_loss=0.08472, over 5667390.17 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:57:00,185 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.136e+02 1.426e+03 1.802e+03 2.612e+03 6.091e+03, threshold=3.603e+03, percent-clipped=10.0
+2023-03-12 14:57:56,451 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:58:06,232 INFO [train.py:968] (0/2) Epoch 24, batch 31900, giga_loss[loss=0.2152, simple_loss=0.2942, pruned_loss=0.06812, over 28469.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3377, pruned_loss=0.0873, over 5678837.45 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3533, pruned_loss=0.1138, over 5693421.16 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3365, pruned_loss=0.08447, over 5675823.88 frames. ], batch size: 78, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:58:12,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8224, 3.6637, 3.4826, 1.7383], device='cuda:0'), covar=tensor([0.0751, 0.0853, 0.0865, 0.2156], device='cuda:0'), in_proj_covar=tensor([0.1249, 0.1154, 0.0971, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 14:58:36,599 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:59:06,495 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:59:09,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:59:17,090 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 14:59:26,591 INFO [train.py:968] (0/2) Epoch 24, batch 31950, giga_loss[loss=0.2722, simple_loss=0.3455, pruned_loss=0.0995, over 28915.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3383, pruned_loss=0.08827, over 5680170.56 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3528, pruned_loss=0.1136, over 5697423.38 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3374, pruned_loss=0.08573, over 5673687.54 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:59:27,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.562e+02 1.368e+03 1.679e+03 2.559e+03 6.580e+03, threshold=3.359e+03, percent-clipped=8.0
+2023-03-12 14:59:56,226 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:00:08,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6584, 1.8503, 1.5393, 1.6679], device='cuda:0'), covar=tensor([0.2816, 0.2778, 0.3169, 0.2518], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1112, 0.1365, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 15:00:19,378 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:00:22,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:00:39,945 INFO [train.py:968] (0/2) Epoch 24, batch 32000, giga_loss[loss=0.2167, simple_loss=0.3038, pruned_loss=0.06479, over 28459.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3364, pruned_loss=0.08832, over 5687128.69 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3526, pruned_loss=0.1135, over 5702558.77 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3354, pruned_loss=0.08531, over 5676237.41 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:00:45,521 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:00:54,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:01:04,662 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:01:31,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:01:34,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:01:35,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:01:51,654 INFO [train.py:968] (0/2) Epoch 24, batch 32050, giga_loss[loss=0.2582, simple_loss=0.3425, pruned_loss=0.08697, over 28998.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3327, pruned_loss=0.08566, over 5682913.34 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3524, pruned_loss=0.1134, over 5703952.34 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3319, pruned_loss=0.0831, over 5672892.19 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:01:52,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.910e+02 1.433e+03 1.856e+03 2.331e+03 6.160e+03, threshold=3.711e+03, percent-clipped=7.0
+2023-03-12 15:01:56,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-12 15:02:12,509 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:02:57,656 INFO [train.py:968] (0/2) Epoch 24, batch 32100, giga_loss[loss=0.2978, simple_loss=0.3596, pruned_loss=0.1181, over 26883.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3305, pruned_loss=0.08494, over 5671446.79 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3525, pruned_loss=0.1135, over 5696326.19 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3292, pruned_loss=0.08205, over 5669563.72 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:03:16,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:01,853 INFO [train.py:968] (0/2) Epoch 24, batch 32150, giga_loss[loss=0.2642, simple_loss=0.3532, pruned_loss=0.08762, over 28967.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3317, pruned_loss=0.08583, over 5688704.95 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3513, pruned_loss=0.1129, over 5703432.52 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.331, pruned_loss=0.08315, over 5680180.53 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:04:02,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.506e+03 1.766e+03 2.256e+03 6.199e+03, threshold=3.531e+03, percent-clipped=4.0
+2023-03-12 15:04:05,413 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:10,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:40,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:43,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:44,674 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:44,699 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:05:04,358 INFO [train.py:968] (0/2) Epoch 24, batch 32200, giga_loss[loss=0.2835, simple_loss=0.3387, pruned_loss=0.1141, over 28143.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3351, pruned_loss=0.08783, over 5689305.58 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.351, pruned_loss=0.1127, over 5703836.22 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3346, pruned_loss=0.08556, over 5681966.82 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:05:13,203 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:05:17,774 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:05:40,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2560, 1.3578, 3.1868, 2.9735], device='cuda:0'), covar=tensor([0.1478, 0.2638, 0.0466, 0.1928], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0659, 0.0972, 0.0930], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 15:06:10,885 INFO [train.py:968] (0/2) Epoch 24, batch 32250, giga_loss[loss=0.2437, simple_loss=0.327, pruned_loss=0.08016, over 28470.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3337, pruned_loss=0.08862, over 5689684.20 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.35, pruned_loss=0.1121, over 5705670.55 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3338, pruned_loss=0.08664, over 5681894.78 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:06:12,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.543e+03 2.128e+03 3.012e+03 1.258e+04, threshold=4.255e+03, percent-clipped=14.0
+2023-03-12 15:07:06,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1081191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:07:08,621 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3731, 1.7831, 1.6567, 1.5376], device='cuda:0'), covar=tensor([0.2001, 0.2022, 0.2097, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0744, 0.0713, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 15:07:14,902 INFO [train.py:968] (0/2) Epoch 24, batch 32300, giga_loss[loss=0.2605, simple_loss=0.3337, pruned_loss=0.09365, over 28773.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.335, pruned_loss=0.09005, over 5675175.31 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3501, pruned_loss=0.1122, over 5697074.60 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3347, pruned_loss=0.08813, over 5675611.42 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:07:19,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6563, 1.7533, 1.3007, 1.3828], device='cuda:0'), covar=tensor([0.0966, 0.0593, 0.1020, 0.1194], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0446, 0.0520, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 15:08:26,184 INFO [train.py:968] (0/2) Epoch 24, batch 32350, giga_loss[loss=0.2308, simple_loss=0.2992, pruned_loss=0.08118, over 24960.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3356, pruned_loss=0.0904, over 5677058.22 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3497, pruned_loss=0.112, over 5701120.22 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3355, pruned_loss=0.08863, over 5673462.52 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:08:26,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.586e+03 2.115e+03 2.775e+03 6.785e+03, threshold=4.231e+03, percent-clipped=9.0
+2023-03-12 15:09:14,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1081277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:09:20,138 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4443, 1.6861, 1.7166, 1.4176], device='cuda:0'), covar=tensor([0.2388, 0.2131, 0.1653, 0.2027], device='cuda:0'), in_proj_covar=tensor([0.1971, 0.1911, 0.1819, 0.1968], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 15:09:38,770 INFO [train.py:968] (0/2) Epoch 24, batch 32400, giga_loss[loss=0.2422, simple_loss=0.3214, pruned_loss=0.08154, over 27625.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3382, pruned_loss=0.09088, over 5681538.70 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3496, pruned_loss=0.112, over 5705268.22 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3378, pruned_loss=0.08883, over 5674158.05 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 15:10:35,161 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:10:39,925 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:10:52,250 INFO [train.py:968] (0/2) Epoch 24, batch 32450, giga_loss[loss=0.257, simple_loss=0.3237, pruned_loss=0.09519, over 26803.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3377, pruned_loss=0.09061, over 5673610.07 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3491, pruned_loss=0.1117, over 5709242.90 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3375, pruned_loss=0.08874, over 5663344.14 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:10:56,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.557e+03 1.992e+03 2.860e+03 7.304e+03, threshold=3.985e+03, percent-clipped=9.0
+2023-03-12 15:11:21,689 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:11:47,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1081387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:12:01,019 INFO [train.py:968] (0/2) Epoch 24, batch 32500, giga_loss[loss=0.2327, simple_loss=0.3111, pruned_loss=0.07714, over 28755.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.333, pruned_loss=0.08887, over 5681063.86 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3484, pruned_loss=0.1114, over 5712189.11 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3331, pruned_loss=0.08696, over 5669377.43 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:12:32,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:12:37,561 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:12:40,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.46 vs. limit=5.0
+2023-03-12 15:13:10,903 INFO [train.py:968] (0/2) Epoch 24, batch 32550, giga_loss[loss=0.2247, simple_loss=0.3077, pruned_loss=0.07085, over 28903.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3276, pruned_loss=0.08662, over 5679741.29 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3482, pruned_loss=0.1113, over 5706193.35 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3275, pruned_loss=0.0846, over 5675481.98 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:13:12,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.545e+03 2.004e+03 2.765e+03 1.148e+04, threshold=4.009e+03, percent-clipped=4.0
+2023-03-12 15:13:14,869 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:13:19,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1081458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:13:25,751 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-12 15:13:53,994 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1081481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:14:10,774 INFO [train.py:968] (0/2) Epoch 24, batch 32600, giga_loss[loss=0.2938, simple_loss=0.3621, pruned_loss=0.1128, over 27535.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3266, pruned_loss=0.08678, over 5657184.26 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3478, pruned_loss=0.1112, over 5694106.17 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3259, pruned_loss=0.08416, over 5662398.31 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:14:51,897 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:14:54,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:15:12,095 INFO [train.py:968] (0/2) Epoch 24, batch 32650, giga_loss[loss=0.2514, simple_loss=0.3374, pruned_loss=0.08275, over 28874.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.328, pruned_loss=0.08752, over 5655820.70 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3473, pruned_loss=0.1109, over 5685716.31 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3276, pruned_loss=0.08538, over 5667480.33 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:15:14,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.710e+02 1.631e+03 2.147e+03 2.677e+03 6.750e+03, threshold=4.294e+03, percent-clipped=5.0
+2023-03-12 15:15:26,999 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:15:28,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2150, 4.1230, 2.3559, 2.3812], device='cuda:0'), covar=tensor([0.0714, 0.0353, 0.0643, 0.0966], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0561, 0.0396, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 15:16:10,883 INFO [train.py:968] (0/2) Epoch 24, batch 32700, giga_loss[loss=0.22, simple_loss=0.3075, pruned_loss=0.06619, over 28679.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3274, pruned_loss=0.08663, over 5666495.31 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3468, pruned_loss=0.1106, over 5685210.58 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3269, pruned_loss=0.08433, over 5675334.74 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:16:15,994 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:16:19,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:16:45,596 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:16:48,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:16:56,311 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:17:14,715 INFO [train.py:968] (0/2) Epoch 24, batch 32750, giga_loss[loss=0.2691, simple_loss=0.3486, pruned_loss=0.09474, over 28752.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3258, pruned_loss=0.08479, over 5658306.95 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3468, pruned_loss=0.1106, over 5687445.94 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.325, pruned_loss=0.08259, over 5662987.13 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:17:20,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.744e+02 1.456e+03 2.028e+03 2.690e+03 6.575e+03, threshold=4.056e+03, percent-clipped=4.0
+2023-03-12 15:17:26,514 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:18:20,713 INFO [train.py:968] (0/2) Epoch 24, batch 32800, giga_loss[loss=0.2439, simple_loss=0.3221, pruned_loss=0.08292, over 29132.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3251, pruned_loss=0.0849, over 5664111.59 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3465, pruned_loss=0.1103, over 5691660.91 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3243, pruned_loss=0.08279, over 5663321.34 frames. ], batch size: 200, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:18:55,452 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:19:00,300 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:19:30,556 INFO [train.py:968] (0/2) Epoch 24, batch 32850, giga_loss[loss=0.2545, simple_loss=0.3377, pruned_loss=0.08564, over 28376.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3249, pruned_loss=0.08458, over 5659205.04 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3466, pruned_loss=0.1104, over 5684759.84 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3239, pruned_loss=0.08244, over 5664935.20 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:19:34,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.579e+02 1.597e+03 2.081e+03 3.111e+03 1.297e+04, threshold=4.163e+03, percent-clipped=15.0
+2023-03-12 15:20:36,399 INFO [train.py:968] (0/2) Epoch 24, batch 32900, giga_loss[loss=0.2702, simple_loss=0.3534, pruned_loss=0.09346, over 28715.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3265, pruned_loss=0.08522, over 5678311.21 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.346, pruned_loss=0.1101, over 5692857.78 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3254, pruned_loss=0.08282, over 5674814.11 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:20:44,122 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:21:39,933 INFO [train.py:968] (0/2) Epoch 24, batch 32950, giga_loss[loss=0.2591, simple_loss=0.3362, pruned_loss=0.09101, over 28880.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3268, pruned_loss=0.08565, over 5682299.29 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3455, pruned_loss=0.1097, over 5697162.96 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3261, pruned_loss=0.08358, over 5675306.10 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:21:45,878 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.108e+02 1.247e+03 1.761e+03 2.194e+03 7.278e+03, threshold=3.523e+03, percent-clipped=4.0
+2023-03-12 15:22:28,169 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:22:38,829 INFO [train.py:968] (0/2) Epoch 24, batch 33000, giga_loss[loss=0.2404, simple_loss=0.3237, pruned_loss=0.07853, over 28495.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3274, pruned_loss=0.08667, over 5677204.38 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3453, pruned_loss=0.1097, over 5689628.71 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3263, pruned_loss=0.08421, over 5678009.54 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:22:38,833 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 15:22:47,606 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.1952, simple_loss=0.2957, pruned_loss=0.04734, over 944034.00 frames. 
+2023-03-12 15:22:47,607 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 15:23:46,876 INFO [train.py:968] (0/2) Epoch 24, batch 33050, giga_loss[loss=0.2566, simple_loss=0.3559, pruned_loss=0.07867, over 28964.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3266, pruned_loss=0.0848, over 5667485.54 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.345, pruned_loss=0.1094, over 5691182.61 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3256, pruned_loss=0.08253, over 5666374.05 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:23:51,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.948e+02 1.598e+03 2.136e+03 3.036e+03 1.106e+04, threshold=4.271e+03, percent-clipped=16.0
+2023-03-12 15:23:53,175 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:24:51,219 INFO [train.py:968] (0/2) Epoch 24, batch 33100, giga_loss[loss=0.3003, simple_loss=0.3729, pruned_loss=0.1139, over 28893.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3308, pruned_loss=0.08622, over 5661889.96 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3449, pruned_loss=0.1094, over 5693311.21 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.33, pruned_loss=0.08428, over 5658899.73 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:24:53,285 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1082000.pt
+2023-03-12 15:25:01,007 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 15:25:48,616 INFO [train.py:968] (0/2) Epoch 24, batch 33150, giga_loss[loss=0.2375, simple_loss=0.327, pruned_loss=0.07402, over 28879.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3335, pruned_loss=0.08756, over 5668470.70 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3447, pruned_loss=0.1093, over 5698711.07 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3325, pruned_loss=0.08512, over 5659972.05 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:25:51,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.528e+03 2.245e+03 3.641e+03 1.229e+04, threshold=4.490e+03, percent-clipped=16.0
+2023-03-12 15:26:25,959 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 15:26:53,823 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:26:54,219 INFO [train.py:968] (0/2) Epoch 24, batch 33200, libri_loss[loss=0.2476, simple_loss=0.3112, pruned_loss=0.09201, over 29650.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3344, pruned_loss=0.0882, over 5673840.59 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3451, pruned_loss=0.1095, over 5702895.74 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.333, pruned_loss=0.08558, over 5662767.35 frames. ], batch size: 73, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:26:57,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:27:56,382 INFO [train.py:968] (0/2) Epoch 24, batch 33250, giga_loss[loss=0.2166, simple_loss=0.3013, pruned_loss=0.06597, over 28933.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3335, pruned_loss=0.08756, over 5670864.89 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3453, pruned_loss=0.1098, over 5696589.19 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.332, pruned_loss=0.08484, over 5667252.44 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:28:02,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.754e+02 1.393e+03 1.733e+03 2.490e+03 7.505e+03, threshold=3.467e+03, percent-clipped=8.0
+2023-03-12 15:28:35,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:28:58,014 INFO [train.py:968] (0/2) Epoch 24, batch 33300, giga_loss[loss=0.2428, simple_loss=0.3225, pruned_loss=0.0815, over 27873.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3288, pruned_loss=0.08453, over 5667551.46 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3446, pruned_loss=0.1095, over 5691000.88 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3279, pruned_loss=0.08202, over 5669452.91 frames. ], batch size: 474, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:29:47,923 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1082240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:29:49,363 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1082242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:29:50,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1082243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:29:51,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1082245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:29:53,547 INFO [train.py:968] (0/2) Epoch 24, batch 33350, libri_loss[loss=0.25, simple_loss=0.31, pruned_loss=0.09502, over 29583.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3279, pruned_loss=0.08474, over 5675102.67 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3443, pruned_loss=0.1093, over 5695058.24 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3268, pruned_loss=0.08179, over 5671846.46 frames. ], batch size: 75, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:29:57,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.564e+02 1.342e+03 1.769e+03 2.670e+03 6.307e+03, threshold=3.537e+03, percent-clipped=9.0
+2023-03-12 15:30:14,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:30:24,976 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1082272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:30:29,425 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1082274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:30:55,365 INFO [train.py:968] (0/2) Epoch 24, batch 33400, giga_loss[loss=0.2091, simple_loss=0.2906, pruned_loss=0.06382, over 28747.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3252, pruned_loss=0.08312, over 5672413.32 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3436, pruned_loss=0.1089, over 5696796.39 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3247, pruned_loss=0.08084, over 5668077.33 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:31:18,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.1002, 5.9136, 5.6581, 2.9406], device='cuda:0'), covar=tensor([0.0534, 0.0745, 0.0944, 0.1638], device='cuda:0'), in_proj_covar=tensor([0.1256, 0.1156, 0.0975, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 15:31:26,581 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1082322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:31:31,242 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1082325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:31:33,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:32:03,346 INFO [train.py:968] (0/2) Epoch 24, batch 33450, giga_loss[loss=0.2593, simple_loss=0.3398, pruned_loss=0.08941, over 28869.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3279, pruned_loss=0.08415, over 5674537.36 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3437, pruned_loss=0.1089, over 5697245.04 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3271, pruned_loss=0.08188, over 5670335.10 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:32:08,055 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-12 15:32:10,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.357e+02 1.418e+03 1.936e+03 2.682e+03 7.152e+03, threshold=3.873e+03, percent-clipped=8.0
+2023-03-12 15:32:13,462 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1082354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:33:05,066 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4769, 1.7083, 1.7095, 1.2599], device='cuda:0'), covar=tensor([0.1590, 0.2556, 0.1361, 0.1715], device='cuda:0'), in_proj_covar=tensor([0.0913, 0.0700, 0.0959, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 15:33:14,295 INFO [train.py:968] (0/2) Epoch 24, batch 33500, libri_loss[loss=0.2889, simple_loss=0.3526, pruned_loss=0.1126, over 29741.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3293, pruned_loss=0.0852, over 5677419.31 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3439, pruned_loss=0.109, over 5699809.96 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3282, pruned_loss=0.08292, over 5671247.18 frames. ], batch size: 87, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:33:26,281 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1082409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:33:31,748 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1082412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:33:56,902 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-12 15:34:11,346 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1082441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:34:21,032 INFO [train.py:968] (0/2) Epoch 24, batch 33550, giga_loss[loss=0.2519, simple_loss=0.3181, pruned_loss=0.0928, over 26766.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.332, pruned_loss=0.08735, over 5657015.78 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3442, pruned_loss=0.1092, over 5691445.82 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3307, pruned_loss=0.08499, over 5659087.38 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:34:26,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.548e+03 2.051e+03 2.847e+03 1.126e+04, threshold=4.102e+03, percent-clipped=11.0
+2023-03-12 15:34:50,758 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1082471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:34:54,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1082474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:35:18,551 INFO [train.py:968] (0/2) Epoch 24, batch 33600, giga_loss[loss=0.2555, simple_loss=0.3413, pruned_loss=0.08484, over 28941.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3352, pruned_loss=0.08845, over 5656585.68 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3445, pruned_loss=0.1094, over 5694738.02 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3337, pruned_loss=0.08614, over 5654861.75 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:35:27,054 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1082503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:35:34,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5448, 1.7628, 1.4123, 1.5858], device='cuda:0'), covar=tensor([0.2836, 0.2758, 0.3135, 0.2437], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1115, 0.1365, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 15:36:21,822 INFO [train.py:968] (0/2) Epoch 24, batch 33650, giga_loss[loss=0.255, simple_loss=0.3351, pruned_loss=0.08742, over 29181.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3376, pruned_loss=0.09025, over 5651904.87 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3443, pruned_loss=0.1095, over 5690314.46 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3362, pruned_loss=0.08745, over 5653646.98 frames. ], batch size: 107, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:36:28,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.439e+02 1.344e+03 1.771e+03 2.724e+03 5.124e+03, threshold=3.541e+03, percent-clipped=2.0
+2023-03-12 15:37:39,924 INFO [train.py:968] (0/2) Epoch 24, batch 33700, giga_loss[loss=0.2259, simple_loss=0.3087, pruned_loss=0.07156, over 28996.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.336, pruned_loss=0.08886, over 5666323.99 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3443, pruned_loss=0.1095, over 5690314.46 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.335, pruned_loss=0.08668, over 5667679.90 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:37:53,937 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1082607.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 15:38:29,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6159, 1.8496, 1.5142, 1.6065], device='cuda:0'), covar=tensor([0.2706, 0.2527, 0.2829, 0.2501], device='cuda:0'), in_proj_covar=tensor([0.1552, 0.1117, 0.1368, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 15:38:51,790 INFO [train.py:968] (0/2) Epoch 24, batch 33750, giga_loss[loss=0.2262, simple_loss=0.3101, pruned_loss=0.07111, over 28906.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3328, pruned_loss=0.08713, over 5677342.60 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3439, pruned_loss=0.1093, over 5692501.47 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3322, pruned_loss=0.08518, over 5676111.42 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:39:00,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.993e+02 1.538e+03 1.850e+03 2.379e+03 8.286e+03, threshold=3.700e+03, percent-clipped=15.0
+2023-03-12 15:39:09,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3217, 1.4579, 1.2910, 1.5386], device='cuda:0'), covar=tensor([0.0789, 0.0370, 0.0372, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0072, 0.0065, 0.0112], device='cuda:0')
+2023-03-12 15:40:03,934 INFO [train.py:968] (0/2) Epoch 24, batch 33800, giga_loss[loss=0.2203, simple_loss=0.3098, pruned_loss=0.06537, over 28925.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3313, pruned_loss=0.08629, over 5672132.40 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3439, pruned_loss=0.1094, over 5694582.10 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3307, pruned_loss=0.08455, over 5669302.15 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:40:35,300 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1082723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:41:05,545 INFO [train.py:968] (0/2) Epoch 24, batch 33850, libri_loss[loss=0.2343, simple_loss=0.2978, pruned_loss=0.08539, over 29341.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3291, pruned_loss=0.08633, over 5671308.09 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3431, pruned_loss=0.109, over 5692371.97 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3288, pruned_loss=0.08439, over 5671256.07 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:41:11,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.567e+02 1.315e+03 1.738e+03 2.388e+03 6.974e+03, threshold=3.476e+03, percent-clipped=7.0
+2023-03-12 15:41:30,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6487, 1.9142, 1.5406, 1.6585], device='cuda:0'), covar=tensor([0.2755, 0.2771, 0.3243, 0.2466], device='cuda:0'), in_proj_covar=tensor([0.1548, 0.1114, 0.1365, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 15:41:45,943 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 15:42:10,384 INFO [train.py:968] (0/2) Epoch 24, batch 33900, giga_loss[loss=0.2467, simple_loss=0.325, pruned_loss=0.08419, over 28944.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3271, pruned_loss=0.08569, over 5678422.19 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3435, pruned_loss=0.1093, over 5693367.36 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3262, pruned_loss=0.08345, over 5677300.51 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:42:34,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1839, 1.2147, 3.6190, 3.0982], device='cuda:0'), covar=tensor([0.1675, 0.2838, 0.0473, 0.1177], device='cuda:0'), in_proj_covar=tensor([0.0775, 0.0658, 0.0971, 0.0927], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 15:43:14,063 INFO [train.py:968] (0/2) Epoch 24, batch 33950, giga_loss[loss=0.2427, simple_loss=0.3282, pruned_loss=0.07858, over 28389.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3278, pruned_loss=0.08469, over 5672338.20 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3436, pruned_loss=0.1093, over 5691018.58 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3268, pruned_loss=0.08264, over 5673579.77 frames. ], batch size: 369, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:43:19,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.832e+02 1.591e+03 2.054e+03 2.997e+03 1.296e+04, threshold=4.108e+03, percent-clipped=16.0
+2023-03-12 15:43:40,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-12 15:44:15,403 INFO [train.py:968] (0/2) Epoch 24, batch 34000, giga_loss[loss=0.2163, simple_loss=0.3166, pruned_loss=0.05804, over 29030.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3272, pruned_loss=0.08359, over 5656084.51 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3435, pruned_loss=0.1094, over 5674830.93 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3262, pruned_loss=0.08138, over 5671804.50 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 15:44:54,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2880, 1.6130, 1.2810, 1.1150], device='cuda:0'), covar=tensor([0.2812, 0.2807, 0.3338, 0.2449], device='cuda:0'), in_proj_covar=tensor([0.1550, 0.1115, 0.1367, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 15:45:08,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.7253, 1.4132, 1.1829, 1.1167], device='cuda:0'), covar=tensor([0.2181, 0.1080, 0.2437, 0.1675], device='cuda:0'), in_proj_covar=tensor([0.0469, 0.0732, 0.0703, 0.0672], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 15:45:08,809 INFO [train.py:968] (0/2) Epoch 24, batch 34050, giga_loss[loss=0.3062, simple_loss=0.3784, pruned_loss=0.117, over 28945.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.329, pruned_loss=0.08288, over 5661355.83 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3432, pruned_loss=0.1091, over 5673867.08 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3279, pruned_loss=0.08042, over 5674658.30 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:45:17,610 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.371e+02 1.434e+03 1.719e+03 2.609e+03 7.200e+03, threshold=3.438e+03, percent-clipped=9.0
+2023-03-12 15:45:17,975 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1082954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:45:18,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-12 15:45:41,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3781, 1.8752, 1.3756, 0.6827], device='cuda:0'), covar=tensor([0.5571, 0.3106, 0.4118, 0.6375], device='cuda:0'), in_proj_covar=tensor([0.1787, 0.1684, 0.1623, 0.1455], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 15:45:49,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082982.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 15:46:12,107 INFO [train.py:968] (0/2) Epoch 24, batch 34100, giga_loss[loss=0.2472, simple_loss=0.3352, pruned_loss=0.07961, over 28961.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3302, pruned_loss=0.08295, over 5669533.60 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.343, pruned_loss=0.1089, over 5678163.89 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3293, pruned_loss=0.08073, over 5676066.16 frames. ], batch size: 285, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:47:26,182 INFO [train.py:968] (0/2) Epoch 24, batch 34150, libri_loss[loss=0.2078, simple_loss=0.2829, pruned_loss=0.06642, over 29668.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3304, pruned_loss=0.08293, over 5668936.30 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3426, pruned_loss=0.1086, over 5680612.36 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.33, pruned_loss=0.08121, over 5671505.78 frames. ], batch size: 73, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:47:32,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.410e+02 1.444e+03 1.832e+03 2.253e+03 8.029e+03, threshold=3.664e+03, percent-clipped=3.0
+2023-03-12 15:48:24,987 INFO [train.py:968] (0/2) Epoch 24, batch 34200, giga_loss[loss=0.2473, simple_loss=0.3359, pruned_loss=0.07936, over 28623.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3303, pruned_loss=0.08348, over 5670221.24 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3426, pruned_loss=0.1088, over 5687958.04 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3295, pruned_loss=0.08099, over 5665105.50 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:48:25,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:49:04,270 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1083125.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 15:49:07,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1083128.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 15:49:35,692 INFO [train.py:968] (0/2) Epoch 24, batch 34250, giga_loss[loss=0.2345, simple_loss=0.3272, pruned_loss=0.07087, over 29028.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3307, pruned_loss=0.08356, over 5655308.24 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3424, pruned_loss=0.1088, over 5673486.88 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.33, pruned_loss=0.08112, over 5663636.12 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:49:45,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.234e+02 1.595e+03 1.939e+03 2.604e+03 5.942e+03, threshold=3.878e+03, percent-clipped=5.0
+2023-03-12 15:49:48,047 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1083157.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 15:49:52,409 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:50:27,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1543, 1.2952, 1.1966, 1.1459], device='cuda:0'), covar=tensor([0.2126, 0.2083, 0.1496, 0.1879], device='cuda:0'), in_proj_covar=tensor([0.1973, 0.1899, 0.1812, 0.1961], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 15:50:47,555 INFO [train.py:968] (0/2) Epoch 24, batch 34300, giga_loss[loss=0.2266, simple_loss=0.3188, pruned_loss=0.06722, over 28603.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3313, pruned_loss=0.0836, over 5654540.09 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3422, pruned_loss=0.1089, over 5669513.21 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3307, pruned_loss=0.08102, over 5664569.02 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:51:45,752 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1083241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:51:49,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1083244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:51:54,062 INFO [train.py:968] (0/2) Epoch 24, batch 34350, giga_loss[loss=0.2396, simple_loss=0.3338, pruned_loss=0.07273, over 28916.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3346, pruned_loss=0.08561, over 5660750.01 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3415, pruned_loss=0.1085, over 5673657.42 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3345, pruned_loss=0.08319, over 5664754.59 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:52:02,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.493e+03 1.843e+03 2.663e+03 1.344e+04, threshold=3.686e+03, percent-clipped=14.0
+2023-03-12 15:52:22,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7688, 2.0572, 1.6835, 2.0483], device='cuda:0'), covar=tensor([0.2497, 0.2565, 0.2876, 0.2372], device='cuda:0'), in_proj_covar=tensor([0.1550, 0.1115, 0.1365, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 15:52:26,972 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1083273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:52:28,927 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9383, 1.3008, 1.0443, 0.1921], device='cuda:0'), covar=tensor([0.4239, 0.3184, 0.4927, 0.7144], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1680, 0.1620, 0.1451], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 15:52:44,409 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:52:58,597 INFO [train.py:968] (0/2) Epoch 24, batch 34400, giga_loss[loss=0.2387, simple_loss=0.3336, pruned_loss=0.07195, over 28265.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3363, pruned_loss=0.08633, over 5662110.68 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.342, pruned_loss=0.109, over 5671171.65 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3355, pruned_loss=0.08326, over 5667568.83 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 15:53:12,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 15:53:34,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1401, 3.9762, 3.7891, 1.9788], device='cuda:0'), covar=tensor([0.0659, 0.0829, 0.0825, 0.2256], device='cuda:0'), in_proj_covar=tensor([0.1244, 0.1143, 0.0964, 0.0720], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 15:53:41,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:54:09,758 INFO [train.py:968] (0/2) Epoch 24, batch 34450, giga_loss[loss=0.244, simple_loss=0.3291, pruned_loss=0.07946, over 28173.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.335, pruned_loss=0.08626, over 5665317.95 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3417, pruned_loss=0.1088, over 5674937.70 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3345, pruned_loss=0.08361, over 5666350.99 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:54:20,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.347e+02 1.466e+03 2.176e+03 3.265e+03 7.624e+03, threshold=4.352e+03, percent-clipped=20.0
+2023-03-12 15:54:31,528 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-12 15:54:33,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3698, 1.7820, 1.6328, 1.4621], device='cuda:0'), covar=tensor([0.2025, 0.1994, 0.2343, 0.2249], device='cuda:0'), in_proj_covar=tensor([0.0466, 0.0730, 0.0700, 0.0670], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 15:55:22,104 INFO [train.py:968] (0/2) Epoch 24, batch 34500, libri_loss[loss=0.2858, simple_loss=0.352, pruned_loss=0.1099, over 27656.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3325, pruned_loss=0.08468, over 5680941.74 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3416, pruned_loss=0.1085, over 5676506.13 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3322, pruned_loss=0.08237, over 5680512.42 frames. ], batch size: 116, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:56:32,365 INFO [train.py:968] (0/2) Epoch 24, batch 34550, giga_loss[loss=0.2463, simple_loss=0.3325, pruned_loss=0.08002, over 28922.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3315, pruned_loss=0.0833, over 5679350.53 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.342, pruned_loss=0.1089, over 5677383.28 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3306, pruned_loss=0.08077, over 5678105.50 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:56:42,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.463e+02 1.382e+03 1.671e+03 2.465e+03 7.534e+03, threshold=3.342e+03, percent-clipped=5.0
+2023-03-12 15:57:00,227 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1083472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:57:06,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1083475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:57:23,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5143, 1.7042, 1.7459, 1.3327], device='cuda:0'), covar=tensor([0.1929, 0.2732, 0.1636, 0.1946], device='cuda:0'), in_proj_covar=tensor([0.0913, 0.0699, 0.0958, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 15:57:32,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1385, 1.2361, 1.0804, 0.8834], device='cuda:0'), covar=tensor([0.1036, 0.0503, 0.1115, 0.1138], device='cuda:0'), in_proj_covar=tensor([0.0398, 0.0442, 0.0517, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 15:57:34,551 INFO [train.py:968] (0/2) Epoch 24, batch 34600, giga_loss[loss=0.3161, simple_loss=0.382, pruned_loss=0.1251, over 28031.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.332, pruned_loss=0.08415, over 5670525.08 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3424, pruned_loss=0.109, over 5680349.05 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3307, pruned_loss=0.08122, over 5667159.11 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:57:41,443 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1083504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:58:07,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2223, 1.8275, 1.3081, 0.4545], device='cuda:0'), covar=tensor([0.4308, 0.2738, 0.4368, 0.5750], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1681, 0.1620, 0.1453], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 15:58:21,478 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:58:21,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:58:35,650 INFO [train.py:968] (0/2) Epoch 24, batch 34650, giga_loss[loss=0.2757, simple_loss=0.3591, pruned_loss=0.09616, over 28979.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3337, pruned_loss=0.08518, over 5671566.07 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3415, pruned_loss=0.1084, over 5685042.19 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3332, pruned_loss=0.08286, over 5664701.41 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:58:49,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.364e+02 1.449e+03 1.963e+03 2.529e+03 5.495e+03, threshold=3.925e+03, percent-clipped=10.0
+2023-03-12 15:59:18,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7879, 2.0145, 2.0605, 1.5424], device='cuda:0'), covar=tensor([0.2002, 0.2506, 0.1582, 0.1977], device='cuda:0'), in_proj_covar=tensor([0.0911, 0.0699, 0.0958, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 15:59:35,718 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:59:43,790 INFO [train.py:968] (0/2) Epoch 24, batch 34700, giga_loss[loss=0.2465, simple_loss=0.3323, pruned_loss=0.08032, over 28388.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3355, pruned_loss=0.08602, over 5678902.58 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3419, pruned_loss=0.1086, over 5684797.78 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3347, pruned_loss=0.08363, over 5673813.86 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:00:37,528 INFO [train.py:968] (0/2) Epoch 24, batch 34750, giga_loss[loss=0.2189, simple_loss=0.3008, pruned_loss=0.06854, over 28927.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3343, pruned_loss=0.08767, over 5657101.75 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.342, pruned_loss=0.1088, over 5674910.83 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3332, pruned_loss=0.08457, over 5662429.41 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:00:43,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5921, 1.9161, 1.5138, 1.6973], device='cuda:0'), covar=tensor([0.2838, 0.2689, 0.3151, 0.2513], device='cuda:0'), in_proj_covar=tensor([0.1548, 0.1114, 0.1367, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 16:00:49,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.598e+03 2.020e+03 2.846e+03 6.951e+03, threshold=4.040e+03, percent-clipped=9.0
+2023-03-12 16:00:54,969 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:01:02,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:01:12,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1083678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:01:16,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1083681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:01:34,136 INFO [train.py:968] (0/2) Epoch 24, batch 34800, giga_loss[loss=0.2256, simple_loss=0.3166, pruned_loss=0.06732, over 28864.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3331, pruned_loss=0.08754, over 5665559.44 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3419, pruned_loss=0.1087, over 5680555.21 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3321, pruned_loss=0.08447, over 5664151.27 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:01:47,918 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1083710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:02:30,604 INFO [train.py:968] (0/2) Epoch 24, batch 34850, libri_loss[loss=0.2863, simple_loss=0.3538, pruned_loss=0.1094, over 29252.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.335, pruned_loss=0.08937, over 5667341.10 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3421, pruned_loss=0.1088, over 5686272.62 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3337, pruned_loss=0.08609, over 5660396.09 frames. ], batch size: 94, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:02:39,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.608e+02 1.534e+03 2.046e+03 3.173e+03 9.843e+03, threshold=4.091e+03, percent-clipped=9.0
+2023-03-12 16:02:44,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:02:46,472 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 16:03:19,385 INFO [train.py:968] (0/2) Epoch 24, batch 34900, giga_loss[loss=0.2619, simple_loss=0.3536, pruned_loss=0.08512, over 28589.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3426, pruned_loss=0.0935, over 5677496.60 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3422, pruned_loss=0.109, over 5691876.15 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3414, pruned_loss=0.09022, over 5666528.06 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:03:23,606 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1083804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:03:26,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1083807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:03:27,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3381, 1.2732, 1.1241, 1.4689], device='cuda:0'), covar=tensor([0.0743, 0.0349, 0.0367, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0072, 0.0065, 0.0112], device='cuda:0')
+2023-03-12 16:03:54,719 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1083836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:04:00,538 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:04:06,824 INFO [train.py:968] (0/2) Epoch 24, batch 34950, giga_loss[loss=0.3089, simple_loss=0.384, pruned_loss=0.1168, over 29085.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3498, pruned_loss=0.09729, over 5672485.45 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3423, pruned_loss=0.1091, over 5687096.74 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3488, pruned_loss=0.09426, over 5667858.59 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:04:13,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.401e+03 1.718e+03 2.144e+03 4.780e+03, threshold=3.436e+03, percent-clipped=2.0
+2023-03-12 16:04:45,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5388, 1.3578, 4.4698, 3.4019], device='cuda:0'), covar=tensor([0.1745, 0.3044, 0.0420, 0.1077], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0662, 0.0974, 0.0932], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 16:04:48,444 INFO [train.py:968] (0/2) Epoch 24, batch 35000, giga_loss[loss=0.2292, simple_loss=0.3131, pruned_loss=0.0726, over 28715.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3512, pruned_loss=0.09882, over 5679107.44 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3427, pruned_loss=0.1091, over 5693961.40 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3503, pruned_loss=0.09601, over 5668479.99 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:05:00,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:05:11,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5320, 1.5510, 1.4175, 1.5908], device='cuda:0'), covar=tensor([0.0763, 0.0342, 0.0340, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0072, 0.0065, 0.0112], device='cuda:0')
+2023-03-12 16:05:31,210 INFO [train.py:968] (0/2) Epoch 24, batch 35050, giga_loss[loss=0.25, simple_loss=0.3247, pruned_loss=0.08768, over 28826.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3451, pruned_loss=0.09639, over 5694501.75 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3422, pruned_loss=0.1085, over 5703973.32 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3451, pruned_loss=0.09408, over 5676084.75 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:05:36,480 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 16:05:37,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.857e+02 1.311e+03 1.775e+03 2.378e+03 8.121e+03, threshold=3.549e+03, percent-clipped=9.0
+2023-03-12 16:05:47,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:05:55,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-12 16:06:04,959 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:06:14,095 INFO [train.py:968] (0/2) Epoch 24, batch 35100, giga_loss[loss=0.2214, simple_loss=0.2997, pruned_loss=0.07153, over 28830.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3388, pruned_loss=0.09352, over 5672328.30 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3425, pruned_loss=0.1087, over 5677473.64 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3385, pruned_loss=0.09128, over 5682205.59 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:06:15,335 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1084000.pt
+2023-03-12 16:06:34,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4275, 1.7368, 1.4591, 1.0844], device='cuda:0'), covar=tensor([0.2513, 0.2532, 0.2939, 0.2412], device='cuda:0'), in_proj_covar=tensor([0.1546, 0.1115, 0.1365, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 16:06:53,368 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-12 16:06:53,624 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1084044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:06:53,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:06:56,720 INFO [train.py:968] (0/2) Epoch 24, batch 35150, giga_loss[loss=0.2572, simple_loss=0.3212, pruned_loss=0.09662, over 28910.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3324, pruned_loss=0.09117, over 5672372.20 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3427, pruned_loss=0.1088, over 5677556.55 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3319, pruned_loss=0.08905, over 5679972.66 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:07:01,118 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:07:03,083 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:07:03,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.115e+02 1.108e+03 1.455e+03 1.895e+03 1.739e+04, threshold=2.909e+03, percent-clipped=5.0
+2023-03-12 16:07:12,861 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1084068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:07:28,420 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:07:37,910 INFO [train.py:968] (0/2) Epoch 24, batch 35200, giga_loss[loss=0.2215, simple_loss=0.294, pruned_loss=0.07448, over 28684.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3254, pruned_loss=0.08812, over 5681216.78 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3433, pruned_loss=0.109, over 5682124.14 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3241, pruned_loss=0.0858, over 5683430.38 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:07:49,816 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:07:51,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:08:10,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:08:16,044 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:08:21,114 INFO [train.py:968] (0/2) Epoch 24, batch 35250, libri_loss[loss=0.2791, simple_loss=0.354, pruned_loss=0.1021, over 29524.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3203, pruned_loss=0.08598, over 5688389.75 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3436, pruned_loss=0.1089, over 5691280.13 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3182, pruned_loss=0.08337, over 5681861.29 frames. ], batch size: 84, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:08:27,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.823e+02 1.178e+03 1.552e+03 2.196e+03 3.939e+03, threshold=3.104e+03, percent-clipped=4.0
+2023-03-12 16:08:55,158 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:08:57,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:09:03,608 INFO [train.py:968] (0/2) Epoch 24, batch 35300, giga_loss[loss=0.2378, simple_loss=0.3235, pruned_loss=0.07606, over 29083.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3159, pruned_loss=0.08402, over 5682721.88 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3436, pruned_loss=0.1088, over 5686256.47 frames. ], giga_tot_loss[loss=0.2383, simple_loss=0.3137, pruned_loss=0.08149, over 5680925.47 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:09:21,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:09:23,194 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:09:32,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1084229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:09:50,612 INFO [train.py:968] (0/2) Epoch 24, batch 35350, giga_loss[loss=0.2082, simple_loss=0.2876, pruned_loss=0.06437, over 28830.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3113, pruned_loss=0.08142, over 5692975.82 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3433, pruned_loss=0.1086, over 5690484.04 frames. ], giga_tot_loss[loss=0.2338, simple_loss=0.3093, pruned_loss=0.07911, over 5687755.48 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:09:57,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.371e+02 1.072e+03 1.395e+03 1.863e+03 6.581e+03, threshold=2.789e+03, percent-clipped=4.0
+2023-03-12 16:10:19,713 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:10:21,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:10:31,457 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1084293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:10:34,251 INFO [train.py:968] (0/2) Epoch 24, batch 35400, giga_loss[loss=0.2188, simple_loss=0.2909, pruned_loss=0.07334, over 29147.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3075, pruned_loss=0.07942, over 5701799.19 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3431, pruned_loss=0.1085, over 5692487.66 frames. ], giga_tot_loss[loss=0.2303, simple_loss=0.3057, pruned_loss=0.07742, over 5695865.34 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:10:46,615 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:11:20,403 INFO [train.py:968] (0/2) Epoch 24, batch 35450, giga_loss[loss=0.2046, simple_loss=0.274, pruned_loss=0.0676, over 28796.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3052, pruned_loss=0.07841, over 5708396.02 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3431, pruned_loss=0.1083, over 5695859.75 frames. ], giga_tot_loss[loss=0.2274, simple_loss=0.3026, pruned_loss=0.07607, over 5700986.03 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:11:28,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.565e+02 1.041e+03 1.316e+03 1.808e+03 5.081e+03, threshold=2.632e+03, percent-clipped=8.0
+2023-03-12 16:11:30,111 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:11:33,317 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:11:33,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:11:59,996 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:12:04,223 INFO [train.py:968] (0/2) Epoch 24, batch 35500, giga_loss[loss=0.2011, simple_loss=0.2783, pruned_loss=0.06201, over 28823.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3012, pruned_loss=0.07657, over 5701918.53 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3432, pruned_loss=0.1084, over 5696929.45 frames. ], giga_tot_loss[loss=0.2239, simple_loss=0.2989, pruned_loss=0.07445, over 5695152.55 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:12:20,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4964, 1.8026, 1.7888, 1.3591], device='cuda:0'), covar=tensor([0.1882, 0.2514, 0.1489, 0.1709], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0705, 0.0967, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 16:12:24,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:12:32,977 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 16:12:45,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:12:48,396 INFO [train.py:968] (0/2) Epoch 24, batch 35550, giga_loss[loss=0.22, simple_loss=0.2958, pruned_loss=0.07214, over 27982.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3005, pruned_loss=0.07668, over 5700914.11 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3439, pruned_loss=0.1087, over 5700517.74 frames. ], giga_tot_loss[loss=0.222, simple_loss=0.2966, pruned_loss=0.07371, over 5692500.42 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:12:55,388 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.790e+02 1.183e+03 1.536e+03 2.157e+03 5.579e+03, threshold=3.073e+03, percent-clipped=16.0
+2023-03-12 16:13:32,238 INFO [train.py:968] (0/2) Epoch 24, batch 35600, giga_loss[loss=0.2108, simple_loss=0.2857, pruned_loss=0.06795, over 28775.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2977, pruned_loss=0.07517, over 5690323.63 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.344, pruned_loss=0.1086, over 5693972.85 frames. ], giga_tot_loss[loss=0.2193, simple_loss=0.2938, pruned_loss=0.07237, over 5689713.75 frames. ], batch size: 285, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:13:38,764 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:13:41,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:13:41,932 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 16:14:09,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:15,815 INFO [train.py:968] (0/2) Epoch 24, batch 35650, libri_loss[loss=0.2741, simple_loss=0.349, pruned_loss=0.0996, over 29525.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2961, pruned_loss=0.07461, over 5695740.41 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3443, pruned_loss=0.1085, over 5692621.83 frames. ], giga_tot_loss[loss=0.2167, simple_loss=0.291, pruned_loss=0.07123, over 5695481.75 frames. ], batch size: 80, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:14:19,329 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-12 16:14:25,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.944e+02 1.094e+03 1.406e+03 2.162e+03 5.305e+03, threshold=2.813e+03, percent-clipped=11.0
+2023-03-12 16:14:29,785 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:33,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:48,707 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 16:14:50,175 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:52,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:59,296 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:15:02,982 INFO [train.py:968] (0/2) Epoch 24, batch 35700, giga_loss[loss=0.2115, simple_loss=0.286, pruned_loss=0.0685, over 28889.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.2968, pruned_loss=0.0755, over 5689121.93 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3447, pruned_loss=0.1088, over 5695801.93 frames. ], giga_tot_loss[loss=0.2179, simple_loss=0.2917, pruned_loss=0.07204, over 5686119.24 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:15:08,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:15:21,200 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:15:47,441 INFO [train.py:968] (0/2) Epoch 24, batch 35750, libri_loss[loss=0.314, simple_loss=0.3793, pruned_loss=0.1244, over 29269.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3078, pruned_loss=0.08108, over 5693568.71 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3449, pruned_loss=0.1087, over 5701186.69 frames. ], giga_tot_loss[loss=0.2287, simple_loss=0.3024, pruned_loss=0.07753, over 5685882.55 frames. ], batch size: 94, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:15:56,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.910e+02 1.230e+03 1.568e+03 2.074e+03 5.984e+03, threshold=3.136e+03, percent-clipped=14.0
+2023-03-12 16:16:09,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:16:22,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5312, 1.7566, 1.5615, 1.6095], device='cuda:0'), covar=tensor([0.1756, 0.2103, 0.2063, 0.1961], device='cuda:0'), in_proj_covar=tensor([0.0476, 0.0740, 0.0710, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 16:16:33,361 INFO [train.py:968] (0/2) Epoch 24, batch 35800, giga_loss[loss=0.2915, simple_loss=0.3606, pruned_loss=0.1112, over 28295.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3209, pruned_loss=0.08786, over 5687967.74 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3451, pruned_loss=0.1087, over 5697830.67 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3156, pruned_loss=0.08438, over 5684948.10 frames. ], batch size: 65, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:17:15,399 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:17:15,912 INFO [train.py:968] (0/2) Epoch 24, batch 35850, giga_loss[loss=0.2902, simple_loss=0.3615, pruned_loss=0.1094, over 28972.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3314, pruned_loss=0.0928, over 5698995.71 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.345, pruned_loss=0.1085, over 5703498.73 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3267, pruned_loss=0.08968, over 5691374.41 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:17:18,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:17:24,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.985e+02 1.387e+03 1.790e+03 2.717e+03 1.006e+04, threshold=3.579e+03, percent-clipped=17.0
+2023-03-12 16:17:26,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4068, 1.9782, 1.6068, 1.6272], device='cuda:0'), covar=tensor([0.0792, 0.0284, 0.0320, 0.0866], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0072, 0.0065, 0.0112], device='cuda:0')
+2023-03-12 16:17:42,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5197, 1.6359, 1.6107, 1.4451], device='cuda:0'), covar=tensor([0.3285, 0.2903, 0.2530, 0.2913], device='cuda:0'), in_proj_covar=tensor([0.2002, 0.1925, 0.1838, 0.1997], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 16:17:42,852 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:17:59,614 INFO [train.py:968] (0/2) Epoch 24, batch 35900, giga_loss[loss=0.242, simple_loss=0.3359, pruned_loss=0.07404, over 28922.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3367, pruned_loss=0.09427, over 5691166.70 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.345, pruned_loss=0.1084, over 5698796.58 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3325, pruned_loss=0.09144, over 5689211.98 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:18:11,098 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:18:14,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:18:44,878 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:18:48,033 INFO [train.py:968] (0/2) Epoch 24, batch 35950, giga_loss[loss=0.2846, simple_loss=0.3658, pruned_loss=0.1017, over 28506.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3382, pruned_loss=0.09344, over 5690818.09 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.345, pruned_loss=0.1084, over 5699920.59 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3349, pruned_loss=0.09118, over 5688379.99 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:18:57,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.894e+02 1.206e+03 1.553e+03 1.980e+03 5.472e+03, threshold=3.106e+03, percent-clipped=3.0
+2023-03-12 16:19:18,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6123, 5.3889, 5.0942, 2.3964], device='cuda:0'), covar=tensor([0.0402, 0.0576, 0.0648, 0.1967], device='cuda:0'), in_proj_covar=tensor([0.1243, 0.1151, 0.0969, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 16:19:35,676 INFO [train.py:968] (0/2) Epoch 24, batch 36000, giga_loss[loss=0.262, simple_loss=0.3442, pruned_loss=0.08985, over 28913.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3404, pruned_loss=0.09423, over 5691818.53 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3447, pruned_loss=0.1081, over 5705538.62 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3379, pruned_loss=0.09225, over 5684862.67 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:19:35,679 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 16:19:41,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0130, 1.2224, 3.4841, 3.0296], device='cuda:0'), covar=tensor([0.1971, 0.3109, 0.0496, 0.1029], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0659, 0.0973, 0.0933], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 16:19:44,690 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2009, simple_loss=0.3075, pruned_loss=0.0471, over 944034.00 frames. 
+2023-03-12 16:19:44,691 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 16:20:27,020 INFO [train.py:968] (0/2) Epoch 24, batch 36050, giga_loss[loss=0.3075, simple_loss=0.3767, pruned_loss=0.1191, over 28239.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3439, pruned_loss=0.09691, over 5683543.64 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3452, pruned_loss=0.1082, over 5696525.35 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3413, pruned_loss=0.09486, over 5684845.99 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:20:35,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.825e+02 1.321e+03 1.683e+03 2.255e+03 5.625e+03, threshold=3.365e+03, percent-clipped=11.0
+2023-03-12 16:21:09,719 INFO [train.py:968] (0/2) Epoch 24, batch 36100, giga_loss[loss=0.2863, simple_loss=0.3597, pruned_loss=0.1064, over 28831.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3477, pruned_loss=0.1, over 5680167.99 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3453, pruned_loss=0.1083, over 5699693.89 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3456, pruned_loss=0.09817, over 5677869.49 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:21:20,817 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 16:21:51,086 INFO [train.py:968] (0/2) Epoch 24, batch 36150, giga_loss[loss=0.2947, simple_loss=0.3544, pruned_loss=0.1175, over 23688.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3506, pruned_loss=0.1015, over 5686908.75 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.346, pruned_loss=0.1087, over 5702350.10 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3483, pruned_loss=0.09933, over 5682071.12 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:22:01,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.347e+02 1.392e+03 1.667e+03 2.280e+03 8.347e+03, threshold=3.335e+03, percent-clipped=10.0
+2023-03-12 16:22:20,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2346, 1.4383, 1.4932, 1.2651], device='cuda:0'), covar=tensor([0.2154, 0.1912, 0.2469, 0.2129], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0743, 0.0714, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 16:22:31,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4732, 4.3051, 4.0669, 1.8365], device='cuda:0'), covar=tensor([0.0548, 0.0711, 0.0716, 0.2251], device='cuda:0'), in_proj_covar=tensor([0.1240, 0.1146, 0.0965, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 16:22:31,784 INFO [train.py:968] (0/2) Epoch 24, batch 36200, giga_loss[loss=0.2954, simple_loss=0.3809, pruned_loss=0.1049, over 29058.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3538, pruned_loss=0.1022, over 5695308.95 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3461, pruned_loss=0.1087, over 5704011.39 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3521, pruned_loss=0.1005, over 5690008.67 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:22:47,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1559, 1.5180, 1.1119, 0.7255], device='cuda:0'), covar=tensor([0.4100, 0.2245, 0.3500, 0.5607], device='cuda:0'), in_proj_covar=tensor([0.1786, 0.1685, 0.1629, 0.1449], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 16:23:15,534 INFO [train.py:968] (0/2) Epoch 24, batch 36250, giga_loss[loss=0.2564, simple_loss=0.3371, pruned_loss=0.08787, over 28584.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3548, pruned_loss=0.1028, over 5691706.64 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3461, pruned_loss=0.1088, over 5710030.55 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3536, pruned_loss=0.1011, over 5681056.68 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:23:25,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.420e+02 1.341e+03 1.737e+03 2.271e+03 5.526e+03, threshold=3.475e+03, percent-clipped=7.0
+2023-03-12 16:23:56,744 INFO [train.py:968] (0/2) Epoch 24, batch 36300, giga_loss[loss=0.2288, simple_loss=0.3229, pruned_loss=0.06736, over 28812.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3549, pruned_loss=0.1017, over 5695980.45 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3461, pruned_loss=0.1087, over 5712266.66 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3542, pruned_loss=0.1003, over 5685164.92 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:24:38,507 INFO [train.py:968] (0/2) Epoch 24, batch 36350, giga_loss[loss=0.2658, simple_loss=0.3461, pruned_loss=0.09274, over 28761.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3539, pruned_loss=0.09962, over 5704998.13 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.346, pruned_loss=0.1087, over 5714334.60 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3535, pruned_loss=0.09848, over 5694637.68 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:24:47,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.170e+02 1.153e+03 1.470e+03 1.812e+03 5.059e+03, threshold=2.940e+03, percent-clipped=2.0
+2023-03-12 16:25:18,119 INFO [train.py:968] (0/2) Epoch 24, batch 36400, giga_loss[loss=0.2445, simple_loss=0.3329, pruned_loss=0.07804, over 28346.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3532, pruned_loss=0.09828, over 5699632.36 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3469, pruned_loss=0.1091, over 5706088.89 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3523, pruned_loss=0.09664, over 5698081.93 frames. ], batch size: 65, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:25:39,509 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:26:01,921 INFO [train.py:968] (0/2) Epoch 24, batch 36450, giga_loss[loss=0.2813, simple_loss=0.3575, pruned_loss=0.1026, over 28903.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3519, pruned_loss=0.0973, over 5708015.17 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.347, pruned_loss=0.109, over 5709366.01 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3512, pruned_loss=0.09595, over 5703962.86 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:26:10,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.702e+02 1.211e+03 1.438e+03 1.834e+03 6.546e+03, threshold=2.876e+03, percent-clipped=7.0
+2023-03-12 16:26:46,584 INFO [train.py:968] (0/2) Epoch 24, batch 36500, giga_loss[loss=0.3083, simple_loss=0.3669, pruned_loss=0.1248, over 28679.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3534, pruned_loss=0.1003, over 5694188.42 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3475, pruned_loss=0.1091, over 5696938.81 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3525, pruned_loss=0.09857, over 5701290.34 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:26:52,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5396, 1.8873, 1.8007, 1.7214], device='cuda:0'), covar=tensor([0.0793, 0.0296, 0.0286, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0112], device='cuda:0')
+2023-03-12 16:27:28,741 INFO [train.py:968] (0/2) Epoch 24, batch 36550, libri_loss[loss=0.2595, simple_loss=0.3314, pruned_loss=0.09379, over 29557.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3553, pruned_loss=0.1036, over 5697067.69 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3476, pruned_loss=0.1088, over 5705244.85 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3548, pruned_loss=0.1022, over 5695127.28 frames. ], batch size: 77, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:27:39,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.574e+02 1.458e+03 1.875e+03 2.737e+03 8.372e+03, threshold=3.751e+03, percent-clipped=24.0
+2023-03-12 16:27:53,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6620, 5.4286, 5.1749, 3.0187], device='cuda:0'), covar=tensor([0.0502, 0.0692, 0.0683, 0.1553], device='cuda:0'), in_proj_covar=tensor([0.1244, 0.1152, 0.0970, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 16:28:06,462 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:28:14,078 INFO [train.py:968] (0/2) Epoch 24, batch 36600, giga_loss[loss=0.2627, simple_loss=0.3328, pruned_loss=0.09632, over 28820.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3554, pruned_loss=0.1049, over 5700175.17 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3479, pruned_loss=0.109, over 5708013.99 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3549, pruned_loss=0.1036, over 5696105.96 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:29:03,864 INFO [train.py:968] (0/2) Epoch 24, batch 36650, giga_loss[loss=0.2978, simple_loss=0.3557, pruned_loss=0.1199, over 28174.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3525, pruned_loss=0.1038, over 5695868.88 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3481, pruned_loss=0.1091, over 5703314.71 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.352, pruned_loss=0.1025, over 5696970.96 frames. ], batch size: 77, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:29:12,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.996e+02 1.260e+03 1.618e+03 2.244e+03 7.260e+03, threshold=3.235e+03, percent-clipped=6.0
+2023-03-12 16:29:22,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7178, 2.7185, 2.5436, 2.3784], device='cuda:0'), covar=tensor([0.1902, 0.2385, 0.2145, 0.2172], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0746, 0.0716, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 16:29:26,784 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-12 16:29:43,188 INFO [train.py:968] (0/2) Epoch 24, batch 36700, giga_loss[loss=0.2745, simple_loss=0.3516, pruned_loss=0.09877, over 28885.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3511, pruned_loss=0.1032, over 5702391.19 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3485, pruned_loss=0.1093, over 5707255.55 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3504, pruned_loss=0.1018, over 5699696.42 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:29:46,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9584, 2.0523, 1.7473, 2.2703], device='cuda:0'), covar=tensor([0.2744, 0.2911, 0.3241, 0.2469], device='cuda:0'), in_proj_covar=tensor([0.1553, 0.1122, 0.1366, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 16:30:29,667 INFO [train.py:968] (0/2) Epoch 24, batch 36750, giga_loss[loss=0.309, simple_loss=0.3814, pruned_loss=0.1183, over 28599.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3505, pruned_loss=0.1019, over 5702909.95 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3486, pruned_loss=0.1092, over 5709005.33 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3499, pruned_loss=0.1009, over 5699111.41 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:30:40,983 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.330e+02 1.314e+03 1.559e+03 2.241e+03 5.435e+03, threshold=3.118e+03, percent-clipped=6.0
+2023-03-12 16:31:09,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8474, 1.4176, 5.0549, 3.9125], device='cuda:0'), covar=tensor([0.1888, 0.3020, 0.0626, 0.0813], device='cuda:0'), in_proj_covar=tensor([0.0771, 0.0658, 0.0972, 0.0930], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 16:31:14,637 INFO [train.py:968] (0/2) Epoch 24, batch 36800, giga_loss[loss=0.2475, simple_loss=0.3335, pruned_loss=0.08075, over 28979.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3499, pruned_loss=0.1013, over 5685456.71 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3497, pruned_loss=0.1098, over 5702351.92 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3485, pruned_loss=0.09975, over 5687556.73 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:31:15,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1085699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:31:25,203 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 16:31:42,499 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085726.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 16:32:01,178 INFO [train.py:968] (0/2) Epoch 24, batch 36850, giga_loss[loss=0.236, simple_loss=0.3133, pruned_loss=0.07934, over 28966.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3458, pruned_loss=0.099, over 5678182.70 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3498, pruned_loss=0.1098, over 5705276.12 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3445, pruned_loss=0.09754, over 5676907.17 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:32:13,619 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.854e+02 1.207e+03 1.557e+03 2.019e+03 5.557e+03, threshold=3.114e+03, percent-clipped=11.0
+2023-03-12 16:32:45,462 INFO [train.py:968] (0/2) Epoch 24, batch 36900, giga_loss[loss=0.2429, simple_loss=0.3181, pruned_loss=0.08385, over 28857.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3409, pruned_loss=0.09696, over 5671592.69 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3508, pruned_loss=0.1103, over 5706814.45 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3387, pruned_loss=0.09497, over 5668080.79 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:33:13,958 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:33:36,235 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1085842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:33:40,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1085845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:33:42,491 INFO [train.py:968] (0/2) Epoch 24, batch 36950, giga_loss[loss=0.2253, simple_loss=0.3125, pruned_loss=0.069, over 28981.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3347, pruned_loss=0.09396, over 5658537.72 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3507, pruned_loss=0.1102, over 5707864.38 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3331, pruned_loss=0.0924, over 5654802.87 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:33:49,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:33:56,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.206e+02 1.064e+03 1.397e+03 1.756e+03 5.400e+03, threshold=2.793e+03, percent-clipped=4.0
+2023-03-12 16:34:00,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1085865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:34:07,044 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1085874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:34:31,229 INFO [train.py:968] (0/2) Epoch 24, batch 37000, giga_loss[loss=0.2811, simple_loss=0.3484, pruned_loss=0.1069, over 28575.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3349, pruned_loss=0.09371, over 5662823.05 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3511, pruned_loss=0.1102, over 5709528.80 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3328, pruned_loss=0.09207, over 5657157.78 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:34:36,421 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3275, 1.3600, 3.9854, 3.3013], device='cuda:0'), covar=tensor([0.1785, 0.2764, 0.0419, 0.0850], device='cuda:0'), in_proj_covar=tensor([0.0770, 0.0656, 0.0969, 0.0927], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 16:34:43,709 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:35:03,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:35:11,109 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 16:35:13,277 INFO [train.py:968] (0/2) Epoch 24, batch 37050, giga_loss[loss=0.2485, simple_loss=0.3268, pruned_loss=0.08512, over 28763.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3356, pruned_loss=0.09351, over 5673994.06 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3522, pruned_loss=0.1107, over 5714013.91 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3324, pruned_loss=0.09122, over 5664122.16 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:35:17,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6764, 1.8017, 1.4466, 1.3236], device='cuda:0'), covar=tensor([0.1056, 0.0677, 0.1061, 0.1264], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0446, 0.0523, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 16:35:23,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.446e+02 1.287e+03 1.651e+03 2.398e+03 8.685e+03, threshold=3.303e+03, percent-clipped=14.0
+2023-03-12 16:35:56,798 INFO [train.py:968] (0/2) Epoch 24, batch 37100, giga_loss[loss=0.2469, simple_loss=0.3294, pruned_loss=0.0822, over 28682.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3341, pruned_loss=0.09207, over 5685400.81 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3524, pruned_loss=0.1106, over 5714461.44 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3309, pruned_loss=0.08982, over 5676077.47 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:35:58,288 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1086000.pt
+2023-03-12 16:36:04,643 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:36:08,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:36:22,779 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:36:33,335 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:36:38,585 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-12 16:36:40,412 INFO [train.py:968] (0/2) Epoch 24, batch 37150, giga_loss[loss=0.2702, simple_loss=0.3479, pruned_loss=0.09618, over 28636.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3301, pruned_loss=0.08994, over 5688570.97 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3527, pruned_loss=0.1108, over 5704222.98 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3271, pruned_loss=0.08782, over 5689834.11 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:36:50,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.620e+02 1.087e+03 1.299e+03 1.637e+03 4.610e+03, threshold=2.598e+03, percent-clipped=3.0
+2023-03-12 16:37:19,950 INFO [train.py:968] (0/2) Epoch 24, batch 37200, giga_loss[loss=0.2433, simple_loss=0.313, pruned_loss=0.08673, over 29132.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3292, pruned_loss=0.0896, over 5695021.88 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3533, pruned_loss=0.1109, over 5696488.90 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3254, pruned_loss=0.08705, over 5702841.15 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:37:22,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086101.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 16:37:35,769 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:37:54,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6092, 1.6117, 1.2804, 1.2735], device='cuda:0'), covar=tensor([0.0891, 0.0573, 0.0928, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0447, 0.0524, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 16:38:02,505 INFO [train.py:968] (0/2) Epoch 24, batch 37250, giga_loss[loss=0.2463, simple_loss=0.3195, pruned_loss=0.08653, over 28940.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3271, pruned_loss=0.089, over 5693268.08 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3534, pruned_loss=0.1108, over 5697256.46 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3239, pruned_loss=0.08694, over 5698704.06 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:38:11,174 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.036e+02 1.078e+03 1.283e+03 1.811e+03 9.925e+03, threshold=2.567e+03, percent-clipped=16.0
+2023-03-12 16:38:39,428 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 16:38:41,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:38:41,710 INFO [train.py:968] (0/2) Epoch 24, batch 37300, giga_loss[loss=0.2355, simple_loss=0.308, pruned_loss=0.08148, over 28894.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3249, pruned_loss=0.0882, over 5701614.36 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3541, pruned_loss=0.1113, over 5699155.71 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3215, pruned_loss=0.08602, over 5704202.11 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:38:58,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6741, 1.0205, 2.8323, 2.6182], device='cuda:0'), covar=tensor([0.1890, 0.2755, 0.0601, 0.1024], device='cuda:0'), in_proj_covar=tensor([0.0771, 0.0656, 0.0969, 0.0928], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 16:39:10,138 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4853, 1.9302, 1.4579, 1.6061], device='cuda:0'), covar=tensor([0.0793, 0.0295, 0.0332, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0065, 0.0112], device='cuda:0')
+2023-03-12 16:39:10,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:39:21,021 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086244.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:39:23,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086247.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:39:24,617 INFO [train.py:968] (0/2) Epoch 24, batch 37350, giga_loss[loss=0.2172, simple_loss=0.2908, pruned_loss=0.07182, over 28788.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3226, pruned_loss=0.08678, over 5707243.15 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3542, pruned_loss=0.1111, over 5693151.04 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3194, pruned_loss=0.08478, over 5714998.04 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:39:35,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.197e+02 1.125e+03 1.379e+03 2.077e+03 5.937e+03, threshold=2.758e+03, percent-clipped=18.0
+2023-03-12 16:39:47,022 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086276.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:39:55,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:40:04,089 INFO [train.py:968] (0/2) Epoch 24, batch 37400, libri_loss[loss=0.2364, simple_loss=0.3169, pruned_loss=0.07793, over 29549.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3229, pruned_loss=0.08701, over 5703081.03 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3547, pruned_loss=0.111, over 5689214.93 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3185, pruned_loss=0.08454, over 5713896.70 frames. ], batch size: 78, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:40:13,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:40:30,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5798, 2.3346, 1.8231, 0.7924], device='cuda:0'), covar=tensor([0.6636, 0.2630, 0.4353, 0.7102], device='cuda:0'), in_proj_covar=tensor([0.1774, 0.1671, 0.1616, 0.1442], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 16:40:36,236 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:40:38,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:40:41,132 INFO [train.py:968] (0/2) Epoch 24, batch 37450, giga_loss[loss=0.2356, simple_loss=0.313, pruned_loss=0.07908, over 29031.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3206, pruned_loss=0.08592, over 5710362.30 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3548, pruned_loss=0.1109, over 5695338.93 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3161, pruned_loss=0.08332, over 5713990.92 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:40:55,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.256e+02 1.115e+03 1.357e+03 1.792e+03 1.162e+04, threshold=2.715e+03, percent-clipped=9.0
+2023-03-12 16:41:02,476 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:03,178 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:05,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:22,945 INFO [train.py:968] (0/2) Epoch 24, batch 37500, libri_loss[loss=0.2991, simple_loss=0.3749, pruned_loss=0.1116, over 29522.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3205, pruned_loss=0.08571, over 5704194.36 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3551, pruned_loss=0.1109, over 5692043.44 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3156, pruned_loss=0.08295, over 5710567.84 frames. ], batch size: 81, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:41:28,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:28,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:48,861 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8727, 1.9894, 1.8425, 1.7358], device='cuda:0'), covar=tensor([0.2271, 0.2913, 0.2653, 0.2887], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0750, 0.0721, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 16:41:48,880 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:50,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:06,380 INFO [train.py:968] (0/2) Epoch 24, batch 37550, giga_loss[loss=0.263, simple_loss=0.3261, pruned_loss=0.09995, over 28699.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3197, pruned_loss=0.08536, over 5704261.20 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3554, pruned_loss=0.1109, over 5691383.10 frames. ], giga_tot_loss[loss=0.2404, simple_loss=0.3152, pruned_loss=0.08281, over 5709776.51 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:42:10,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:12,921 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:15,754 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:19,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.401e+02 1.071e+03 1.331e+03 1.766e+03 7.327e+03, threshold=2.661e+03, percent-clipped=12.0
+2023-03-12 16:42:40,481 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:44,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:50,285 INFO [train.py:968] (0/2) Epoch 24, batch 37600, giga_loss[loss=0.2815, simple_loss=0.3545, pruned_loss=0.1043, over 27962.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3221, pruned_loss=0.08684, over 5715998.02 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3552, pruned_loss=0.1107, over 5693965.65 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3181, pruned_loss=0.08456, over 5718431.58 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:42:59,995 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:43:21,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5753, 1.7863, 1.7780, 1.5705], device='cuda:0'), covar=tensor([0.2123, 0.2043, 0.2433, 0.2269], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0750, 0.0722, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 16:43:34,958 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:43:35,402 INFO [train.py:968] (0/2) Epoch 24, batch 37650, giga_loss[loss=0.2623, simple_loss=0.3394, pruned_loss=0.09262, over 28926.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3286, pruned_loss=0.09099, over 5704695.53 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.356, pruned_loss=0.1111, over 5689323.42 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3238, pruned_loss=0.0883, over 5711840.05 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:43:37,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:43:48,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.655e+02 1.332e+03 1.719e+03 2.135e+03 9.848e+03, threshold=3.439e+03, percent-clipped=11.0
+2023-03-12 16:44:05,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:44:26,155 INFO [train.py:968] (0/2) Epoch 24, batch 37700, giga_loss[loss=0.2921, simple_loss=0.3659, pruned_loss=0.1091, over 29114.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3375, pruned_loss=0.09691, over 5697170.12 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3562, pruned_loss=0.1111, over 5692553.82 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3332, pruned_loss=0.09447, over 5700261.50 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:44:43,503 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-12 16:44:57,668 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:45:05,018 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:45:08,799 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:45:19,121 INFO [train.py:968] (0/2) Epoch 24, batch 37750, giga_loss[loss=0.2982, simple_loss=0.3697, pruned_loss=0.1134, over 28250.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3429, pruned_loss=0.0999, over 5683036.02 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3557, pruned_loss=0.1109, over 5694543.57 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3398, pruned_loss=0.09811, over 5683623.07 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:45:34,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.548e+02 1.283e+03 1.552e+03 2.131e+03 7.718e+03, threshold=3.104e+03, percent-clipped=7.0
+2023-03-12 16:45:37,789 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:45:59,039 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 16:46:02,692 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 16:46:03,632 INFO [train.py:968] (0/2) Epoch 24, batch 37800, giga_loss[loss=0.295, simple_loss=0.3691, pruned_loss=0.1104, over 28893.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3469, pruned_loss=0.1013, over 5688384.36 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3556, pruned_loss=0.1107, over 5696633.93 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3441, pruned_loss=0.09952, over 5686679.90 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:46:13,511 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5248, 1.8350, 1.4545, 1.6598], device='cuda:0'), covar=tensor([0.2719, 0.2681, 0.3108, 0.2287], device='cuda:0'), in_proj_covar=tensor([0.1550, 0.1119, 0.1364, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 16:46:55,497 INFO [train.py:968] (0/2) Epoch 24, batch 37850, giga_loss[loss=0.3832, simple_loss=0.4122, pruned_loss=0.1771, over 23384.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3529, pruned_loss=0.1043, over 5681816.54 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3556, pruned_loss=0.1107, over 5697822.76 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3508, pruned_loss=0.1029, over 5679337.70 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:47:10,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.581e+02 1.240e+03 1.467e+03 2.103e+03 5.035e+03, threshold=2.934e+03, percent-clipped=6.0
+2023-03-12 16:47:13,199 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:47:19,187 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086774.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:47:20,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3785, 4.2163, 3.9600, 2.0131], device='cuda:0'), covar=tensor([0.0550, 0.0689, 0.0677, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.1240, 0.1147, 0.0967, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 16:47:25,804 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-12 16:47:29,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:47:39,684 INFO [train.py:968] (0/2) Epoch 24, batch 37900, giga_loss[loss=0.2589, simple_loss=0.3349, pruned_loss=0.09147, over 28469.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3552, pruned_loss=0.105, over 5688355.70 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3552, pruned_loss=0.1105, over 5698373.48 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3538, pruned_loss=0.104, over 5685571.49 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:48:17,549 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:48:21,997 INFO [train.py:968] (0/2) Epoch 24, batch 37950, giga_loss[loss=0.2685, simple_loss=0.3501, pruned_loss=0.09342, over 28715.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3517, pruned_loss=0.1023, over 5693558.08 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3555, pruned_loss=0.1107, over 5702941.15 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3503, pruned_loss=0.1012, over 5687032.84 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:48:34,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.627e+02 1.312e+03 1.655e+03 2.105e+03 4.989e+03, threshold=3.310e+03, percent-clipped=16.0
+2023-03-12 16:48:52,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:49:06,450 INFO [train.py:968] (0/2) Epoch 24, batch 38000, giga_loss[loss=0.3203, simple_loss=0.3804, pruned_loss=0.1301, over 23571.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3495, pruned_loss=0.1, over 5687447.45 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3561, pruned_loss=0.1111, over 5696561.86 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3478, pruned_loss=0.09861, over 5687986.08 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:49:35,319 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3315, 3.4713, 1.4815, 1.5459], device='cuda:0'), covar=tensor([0.1065, 0.0306, 0.0914, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0555, 0.0392, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 16:49:46,206 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3589, 4.3016, 1.5914, 1.6258], device='cuda:0'), covar=tensor([0.1101, 0.0293, 0.0923, 0.1402], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0555, 0.0392, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 16:49:47,811 INFO [train.py:968] (0/2) Epoch 24, batch 38050, giga_loss[loss=0.2468, simple_loss=0.3279, pruned_loss=0.08291, over 28945.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3501, pruned_loss=0.1003, over 5695850.80 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.357, pruned_loss=0.1117, over 5699408.81 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3478, pruned_loss=0.09827, over 5693600.97 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:49:58,167 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1758, 1.2510, 1.0979, 0.8479], device='cuda:0'), covar=tensor([0.1136, 0.0587, 0.1147, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0448, 0.0524, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 16:49:59,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 16:50:02,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.767e+02 1.253e+03 1.695e+03 2.287e+03 5.548e+03, threshold=3.390e+03, percent-clipped=8.0
+2023-03-12 16:50:04,715 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-12 16:50:14,752 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6279, 1.4860, 1.7462, 1.2964], device='cuda:0'), covar=tensor([0.2148, 0.3231, 0.1636, 0.1804], device='cuda:0'), in_proj_covar=tensor([0.0919, 0.0707, 0.0964, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 16:50:16,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2252, 4.0739, 3.8276, 1.8002], device='cuda:0'), covar=tensor([0.0607, 0.0749, 0.0722, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.1243, 0.1151, 0.0969, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 16:50:32,108 INFO [train.py:968] (0/2) Epoch 24, batch 38100, libri_loss[loss=0.2756, simple_loss=0.351, pruned_loss=0.1, over 29363.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3512, pruned_loss=0.1007, over 5697549.42 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3569, pruned_loss=0.1115, over 5702779.51 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3493, pruned_loss=0.09906, over 5692600.23 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:50:37,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:50:51,570 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-12 16:50:56,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:50:58,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:51:20,142 INFO [train.py:968] (0/2) Epoch 24, batch 38150, giga_loss[loss=0.2806, simple_loss=0.3565, pruned_loss=0.1024, over 28963.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3539, pruned_loss=0.1027, over 5701355.27 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.357, pruned_loss=0.1116, over 5699663.62 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3523, pruned_loss=0.1013, over 5700111.73 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:51:27,236 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:51:32,365 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.428e+02 1.264e+03 1.603e+03 2.130e+03 3.928e+03, threshold=3.207e+03, percent-clipped=3.0
+2023-03-12 16:51:58,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6629, 4.4019, 1.7904, 1.9622], device='cuda:0'), covar=tensor([0.0986, 0.0307, 0.0846, 0.1193], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0557, 0.0393, 0.0432], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 16:52:00,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7345, 2.0663, 1.8033, 1.8955], device='cuda:0'), covar=tensor([0.0752, 0.0279, 0.0300, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0112], device='cuda:0')
+2023-03-12 16:52:06,133 INFO [train.py:968] (0/2) Epoch 24, batch 38200, giga_loss[loss=0.2612, simple_loss=0.3419, pruned_loss=0.09027, over 29051.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3566, pruned_loss=0.105, over 5691049.99 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3571, pruned_loss=0.1118, over 5692337.57 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3552, pruned_loss=0.1037, over 5695788.25 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:52:09,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4028, 2.1613, 1.6202, 0.6028], device='cuda:0'), covar=tensor([0.4813, 0.3207, 0.3984, 0.6007], device='cuda:0'), in_proj_covar=tensor([0.1775, 0.1675, 0.1621, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 16:52:48,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:52:52,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:52:52,867 INFO [train.py:968] (0/2) Epoch 24, batch 38250, giga_loss[loss=0.2711, simple_loss=0.3425, pruned_loss=0.09982, over 29001.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3572, pruned_loss=0.1061, over 5683097.96 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3576, pruned_loss=0.1121, over 5685294.63 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3557, pruned_loss=0.1047, over 5694007.11 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:52:53,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087149.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:52:55,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:53:02,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:53:06,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.303e+02 1.354e+03 1.729e+03 2.352e+03 7.739e+03, threshold=3.458e+03, percent-clipped=12.0
+2023-03-12 16:53:11,728 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 16:53:20,128 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:53:28,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4861, 1.5033, 1.5169, 1.3187], device='cuda:0'), covar=tensor([0.2671, 0.2834, 0.2041, 0.2456], device='cuda:0'), in_proj_covar=tensor([0.2013, 0.1946, 0.1862, 0.2019], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 16:53:35,092 INFO [train.py:968] (0/2) Epoch 24, batch 38300, giga_loss[loss=0.3356, simple_loss=0.3868, pruned_loss=0.1422, over 27573.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3573, pruned_loss=0.1063, over 5690653.89 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3582, pruned_loss=0.1124, over 5688201.13 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3556, pruned_loss=0.1048, over 5696818.62 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:53:37,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3807, 1.4655, 1.5993, 1.2953], device='cuda:0'), covar=tensor([0.1333, 0.1939, 0.1117, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0705, 0.0963, 0.0861], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 16:53:52,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:54:19,105 INFO [train.py:968] (0/2) Epoch 24, batch 38350, giga_loss[loss=0.2644, simple_loss=0.3446, pruned_loss=0.09215, over 29005.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3573, pruned_loss=0.1061, over 5691139.13 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3587, pruned_loss=0.1127, over 5692434.56 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3554, pruned_loss=0.1045, over 5692152.40 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:54:32,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.732e+02 1.220e+03 1.472e+03 1.916e+03 4.757e+03, threshold=2.943e+03, percent-clipped=4.0
+2023-03-12 16:54:34,787 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.12 vs. limit=5.0
+2023-03-12 16:54:36,115 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1087269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:54:48,491 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:54:51,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:54:54,456 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087292.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:54:58,167 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087295.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:54:59,847 INFO [train.py:968] (0/2) Epoch 24, batch 38400, giga_loss[loss=0.2747, simple_loss=0.3495, pruned_loss=0.09996, over 28886.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3573, pruned_loss=0.1052, over 5687568.48 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3587, pruned_loss=0.1127, over 5684168.29 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3559, pruned_loss=0.1039, over 5695657.60 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:55:05,662 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:07,512 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:09,146 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 16:55:18,040 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:24,800 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087324.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 16:55:32,787 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:42,248 INFO [train.py:968] (0/2) Epoch 24, batch 38450, giga_loss[loss=0.2779, simple_loss=0.3525, pruned_loss=0.1016, over 28758.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3569, pruned_loss=0.1038, over 5700823.32 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3587, pruned_loss=0.1127, over 5687461.56 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3557, pruned_loss=0.1027, over 5704263.18 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:55:49,730 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1087357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:52,731 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:55,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:55,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.886e+02 1.207e+03 1.395e+03 1.831e+03 5.321e+03, threshold=2.790e+03, percent-clipped=5.0
+2023-03-12 16:56:19,755 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:56:23,595 INFO [train.py:968] (0/2) Epoch 24, batch 38500, giga_loss[loss=0.2531, simple_loss=0.3367, pruned_loss=0.0848, over 28975.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3555, pruned_loss=0.1031, over 5686149.96 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3592, pruned_loss=0.113, over 5676187.43 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3539, pruned_loss=0.1016, over 5699858.53 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:56:30,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3928, 1.6333, 1.6982, 1.4975], device='cuda:0'), covar=tensor([0.1263, 0.1121, 0.1340, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0747, 0.0717, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 16:57:03,485 INFO [train.py:968] (0/2) Epoch 24, batch 38550, giga_loss[loss=0.2803, simple_loss=0.3568, pruned_loss=0.1019, over 28643.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3538, pruned_loss=0.1026, over 5685056.82 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.359, pruned_loss=0.1128, over 5672823.15 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3527, pruned_loss=0.1012, over 5699076.30 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:57:16,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.949e+02 1.144e+03 1.432e+03 1.922e+03 8.830e+03, threshold=2.864e+03, percent-clipped=12.0
+2023-03-12 16:57:34,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0215, 2.1596, 2.0517, 1.9984], device='cuda:0'), covar=tensor([0.2204, 0.2516, 0.2355, 0.2219], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0746, 0.0717, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 16:57:42,714 INFO [train.py:968] (0/2) Epoch 24, batch 38600, giga_loss[loss=0.2584, simple_loss=0.3334, pruned_loss=0.09171, over 28641.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3516, pruned_loss=0.1014, over 5689161.29 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3593, pruned_loss=0.113, over 5672853.84 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3503, pruned_loss=0.09988, over 5701049.35 frames. ], batch size: 85, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 16:58:20,433 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1087543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:58:25,057 INFO [train.py:968] (0/2) Epoch 24, batch 38650, giga_loss[loss=0.2649, simple_loss=0.3201, pruned_loss=0.1049, over 23582.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3504, pruned_loss=0.101, over 5693914.27 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3591, pruned_loss=0.1128, over 5677486.08 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3494, pruned_loss=0.09966, over 5699505.41 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 16:58:39,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.905e+02 1.135e+03 1.425e+03 1.889e+03 6.361e+03, threshold=2.851e+03, percent-clipped=8.0
+2023-03-12 16:58:45,371 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 16:59:03,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5185, 1.6657, 1.7570, 1.3387], device='cuda:0'), covar=tensor([0.1842, 0.2660, 0.1483, 0.1764], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0708, 0.0964, 0.0861], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 16:59:06,244 INFO [train.py:968] (0/2) Epoch 24, batch 38700, giga_loss[loss=0.2585, simple_loss=0.3426, pruned_loss=0.08722, over 28915.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3506, pruned_loss=0.1011, over 5700827.60 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3591, pruned_loss=0.1127, over 5683188.83 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3495, pruned_loss=0.09987, over 5700708.39 frames. ], batch size: 213, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 16:59:26,635 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8269, 2.0898, 2.0100, 1.5907], device='cuda:0'), covar=tensor([0.1822, 0.2763, 0.1564, 0.1927], device='cuda:0'), in_proj_covar=tensor([0.0918, 0.0710, 0.0966, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 16:59:41,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1087643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:59:42,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:59:44,590 INFO [train.py:968] (0/2) Epoch 24, batch 38750, giga_loss[loss=0.2537, simple_loss=0.3359, pruned_loss=0.08578, over 28870.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.351, pruned_loss=0.101, over 5706892.91 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.359, pruned_loss=0.1126, over 5686551.52 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3501, pruned_loss=0.0999, over 5704031.37 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 16:59:57,173 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.660e+02 1.179e+03 1.409e+03 1.843e+03 4.970e+03, threshold=2.818e+03, percent-clipped=3.0
+2023-03-12 17:00:21,536 INFO [train.py:968] (0/2) Epoch 24, batch 38800, giga_loss[loss=0.2664, simple_loss=0.3451, pruned_loss=0.09385, over 28734.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3506, pruned_loss=0.1004, over 5702716.06 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3588, pruned_loss=0.1125, over 5679093.90 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3499, pruned_loss=0.09928, over 5707945.29 frames. ], batch size: 262, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:00:48,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:01:00,336 INFO [train.py:968] (0/2) Epoch 24, batch 38850, giga_loss[loss=0.2659, simple_loss=0.3427, pruned_loss=0.09455, over 29019.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3498, pruned_loss=0.09964, over 5694040.29 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3591, pruned_loss=0.1127, over 5666650.99 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3487, pruned_loss=0.09815, over 5710251.85 frames. ], batch size: 164, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:01:15,743 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.296e+02 1.135e+03 1.489e+03 2.040e+03 8.150e+03, threshold=2.978e+03, percent-clipped=14.0
+2023-03-12 17:01:33,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:01:35,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:01:43,314 INFO [train.py:968] (0/2) Epoch 24, batch 38900, giga_loss[loss=0.2555, simple_loss=0.3307, pruned_loss=0.09019, over 28783.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3483, pruned_loss=0.09916, over 5690206.07 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3591, pruned_loss=0.1127, over 5670253.22 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3473, pruned_loss=0.09782, over 5700257.80 frames. ], batch size: 66, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:01:54,798 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-12 17:01:58,917 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:02:21,281 INFO [train.py:968] (0/2) Epoch 24, batch 38950, giga_loss[loss=0.229, simple_loss=0.3056, pruned_loss=0.0762, over 28610.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.347, pruned_loss=0.09906, over 5698037.00 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3598, pruned_loss=0.1131, over 5677668.89 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3452, pruned_loss=0.09723, over 5700267.18 frames. ], batch size: 92, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:02:36,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.500e+02 1.174e+03 1.568e+03 2.195e+03 6.161e+03, threshold=3.137e+03, percent-clipped=16.0
+2023-03-12 17:02:43,089 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:02:45,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:03:00,505 INFO [train.py:968] (0/2) Epoch 24, batch 39000, giga_loss[loss=0.2288, simple_loss=0.3107, pruned_loss=0.07343, over 29012.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3437, pruned_loss=0.0974, over 5700900.87 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3601, pruned_loss=0.1134, over 5679675.40 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3415, pruned_loss=0.09511, over 5701514.35 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:03:00,509 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 17:03:08,716 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0478, 1.1911, 3.4884, 3.0792], device='cuda:0'), covar=tensor([0.1888, 0.2964, 0.0476, 0.0956], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0656, 0.0970, 0.0934], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 17:03:09,471 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2035, simple_loss=0.3115, pruned_loss=0.04776, over 944034.00 frames. 
+2023-03-12 17:03:09,472 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 17:03:16,612 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:03:24,317 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:03:24,850 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1087919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:03:49,275 INFO [train.py:968] (0/2) Epoch 24, batch 39050, giga_loss[loss=0.3133, simple_loss=0.3618, pruned_loss=0.1324, over 23436.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3415, pruned_loss=0.09607, over 5705768.75 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.36, pruned_loss=0.1133, over 5685194.11 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3393, pruned_loss=0.09396, over 5701828.97 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:04:07,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.056e+02 1.155e+03 1.491e+03 2.023e+03 6.511e+03, threshold=2.982e+03, percent-clipped=6.0
+2023-03-12 17:04:34,258 INFO [train.py:968] (0/2) Epoch 24, batch 39100, libri_loss[loss=0.2892, simple_loss=0.3591, pruned_loss=0.1097, over 29501.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3412, pruned_loss=0.0965, over 5697736.28 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3604, pruned_loss=0.1136, over 5670537.32 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3388, pruned_loss=0.09425, over 5708272.03 frames. ], batch size: 85, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:04:35,712 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1088000.pt
+2023-03-12 17:04:51,194 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:05:15,019 INFO [train.py:968] (0/2) Epoch 24, batch 39150, giga_loss[loss=0.2161, simple_loss=0.2962, pruned_loss=0.06798, over 28858.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3383, pruned_loss=0.09521, over 5693185.77 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3604, pruned_loss=0.1136, over 5661054.58 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3361, pruned_loss=0.09307, over 5711065.70 frames. ], batch size: 199, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:05:27,350 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:05:30,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:05:31,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.802e+02 1.130e+03 1.447e+03 1.871e+03 1.055e+04, threshold=2.893e+03, percent-clipped=7.0
+2023-03-12 17:05:52,154 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1465, 1.2245, 3.6383, 3.0060], device='cuda:0'), covar=tensor([0.1678, 0.2747, 0.0418, 0.1099], device='cuda:0'), in_proj_covar=tensor([0.0770, 0.0653, 0.0966, 0.0929], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 17:05:52,158 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:05:57,135 INFO [train.py:968] (0/2) Epoch 24, batch 39200, giga_loss[loss=0.2423, simple_loss=0.3219, pruned_loss=0.08135, over 28719.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3352, pruned_loss=0.09359, over 5695083.63 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3606, pruned_loss=0.1138, over 5661654.68 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3331, pruned_loss=0.0916, over 5708866.63 frames. ], batch size: 71, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:06:04,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7519, 2.0783, 1.8969, 1.8425], device='cuda:0'), covar=tensor([0.1717, 0.1620, 0.1962, 0.1661], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0746, 0.0717, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 17:06:37,994 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:06:42,412 INFO [train.py:968] (0/2) Epoch 24, batch 39250, giga_loss[loss=0.2363, simple_loss=0.3068, pruned_loss=0.08289, over 28484.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3328, pruned_loss=0.0927, over 5700818.77 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.114, over 5665384.40 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3304, pruned_loss=0.09061, over 5709037.78 frames. ], batch size: 71, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:06:52,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:06:55,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:06:56,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.556e+02 1.131e+03 1.361e+03 1.828e+03 5.734e+03, threshold=2.723e+03, percent-clipped=9.0
+2023-03-12 17:07:06,075 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:07:21,739 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:07:25,106 INFO [train.py:968] (0/2) Epoch 24, batch 39300, giga_loss[loss=0.2684, simple_loss=0.3468, pruned_loss=0.09498, over 28941.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3327, pruned_loss=0.09284, over 5693362.15 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.361, pruned_loss=0.114, over 5659786.06 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3302, pruned_loss=0.09072, over 5705690.27 frames. ], batch size: 227, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:08:10,520 INFO [train.py:968] (0/2) Epoch 24, batch 39350, giga_loss[loss=0.2643, simple_loss=0.3457, pruned_loss=0.09141, over 28829.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3352, pruned_loss=0.09369, over 5691026.49 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3611, pruned_loss=0.114, over 5654808.79 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3326, pruned_loss=0.09164, over 5706068.68 frames. ], batch size: 227, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:08:29,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.774e+02 1.235e+03 1.476e+03 2.040e+03 1.094e+04, threshold=2.951e+03, percent-clipped=13.0
+2023-03-12 17:08:47,077 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 17:08:55,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:08:59,889 INFO [train.py:968] (0/2) Epoch 24, batch 39400, libri_loss[loss=0.3436, simple_loss=0.3967, pruned_loss=0.1453, over 29535.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3383, pruned_loss=0.09469, over 5690007.47 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3614, pruned_loss=0.1142, over 5649813.00 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3355, pruned_loss=0.09259, over 5707153.30 frames. ], batch size: 80, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:09:11,984 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:09:29,577 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:09:44,316 INFO [train.py:968] (0/2) Epoch 24, batch 39450, giga_loss[loss=0.2744, simple_loss=0.3567, pruned_loss=0.09608, over 28757.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3409, pruned_loss=0.09563, over 5690651.54 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.361, pruned_loss=0.114, over 5655006.43 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3387, pruned_loss=0.09377, over 5700400.63 frames. ], batch size: 243, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:10:02,613 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.579e+02 1.180e+03 1.438e+03 1.803e+03 4.474e+03, threshold=2.876e+03, percent-clipped=8.0
+2023-03-12 17:10:28,096 INFO [train.py:968] (0/2) Epoch 24, batch 39500, giga_loss[loss=0.2355, simple_loss=0.3221, pruned_loss=0.07445, over 28865.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3425, pruned_loss=0.09611, over 5681966.20 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5649377.04 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3398, pruned_loss=0.09365, over 5696176.45 frames. ], batch size: 227, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:10:40,980 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:11:02,394 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:11:04,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:11:10,328 INFO [train.py:968] (0/2) Epoch 24, batch 39550, giga_loss[loss=0.2363, simple_loss=0.324, pruned_loss=0.07431, over 28897.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.342, pruned_loss=0.09545, over 5671924.84 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3613, pruned_loss=0.1145, over 5641736.72 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3397, pruned_loss=0.09325, over 5690017.30 frames. ], batch size: 227, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:11:27,037 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.187e+02 1.248e+03 1.569e+03 2.096e+03 6.182e+03, threshold=3.139e+03, percent-clipped=11.0
+2023-03-12 17:11:28,593 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:11:55,425 INFO [train.py:968] (0/2) Epoch 24, batch 39600, giga_loss[loss=0.3432, simple_loss=0.4049, pruned_loss=0.1407, over 28624.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3421, pruned_loss=0.09539, over 5676912.71 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3613, pruned_loss=0.1145, over 5641736.72 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3404, pruned_loss=0.09368, over 5690994.30 frames. ], batch size: 336, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:12:12,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:12:21,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4057, 1.5517, 1.5246, 1.3591], device='cuda:0'), covar=tensor([0.3540, 0.2921, 0.2310, 0.2990], device='cuda:0'), in_proj_covar=tensor([0.2007, 0.1942, 0.1859, 0.2012], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 17:12:38,697 INFO [train.py:968] (0/2) Epoch 24, batch 39650, giga_loss[loss=0.2577, simple_loss=0.3236, pruned_loss=0.09593, over 23631.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3432, pruned_loss=0.09681, over 5680162.85 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5645876.56 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3414, pruned_loss=0.09504, over 5688241.75 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:12:41,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:12:55,501 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.339e+03 1.619e+03 2.362e+03 7.391e+03, threshold=3.237e+03, percent-clipped=7.0
+2023-03-12 17:13:21,575 INFO [train.py:968] (0/2) Epoch 24, batch 39700, giga_loss[loss=0.2885, simple_loss=0.364, pruned_loss=0.1065, over 29009.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.346, pruned_loss=0.09813, over 5681877.34 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3619, pruned_loss=0.1149, over 5639960.41 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3441, pruned_loss=0.09637, over 5693732.42 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:13:22,666 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 17:14:04,072 INFO [train.py:968] (0/2) Epoch 24, batch 39750, giga_loss[loss=0.3075, simple_loss=0.3824, pruned_loss=0.1163, over 28749.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3492, pruned_loss=0.1002, over 5686669.97 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1148, over 5639706.74 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3473, pruned_loss=0.09845, over 5698132.14 frames. ], batch size: 284, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:14:16,452 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:18,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:19,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.944e+02 1.371e+03 1.720e+03 2.569e+03 1.136e+04, threshold=3.440e+03, percent-clipped=15.0
+2023-03-12 17:14:33,837 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:34,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:42,317 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:43,836 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:45,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:45,607 INFO [train.py:968] (0/2) Epoch 24, batch 39800, giga_loss[loss=0.2942, simple_loss=0.3626, pruned_loss=0.1129, over 28975.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3519, pruned_loss=0.1009, over 5703384.81 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3624, pruned_loss=0.115, over 5645804.69 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3495, pruned_loss=0.09896, over 5708588.56 frames. ], batch size: 136, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:14:52,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:15:00,082 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:15:07,178 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:15:27,563 INFO [train.py:968] (0/2) Epoch 24, batch 39850, giga_loss[loss=0.3096, simple_loss=0.368, pruned_loss=0.1256, over 28899.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3535, pruned_loss=0.1021, over 5705559.33 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3624, pruned_loss=0.115, over 5648371.32 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3513, pruned_loss=0.1002, over 5708512.44 frames. ], batch size: 106, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:15:44,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.523e+02 1.338e+03 1.631e+03 2.309e+03 6.627e+03, threshold=3.262e+03, percent-clipped=5.0
+2023-03-12 17:16:03,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:09,491 INFO [train.py:968] (0/2) Epoch 24, batch 39900, libri_loss[loss=0.2612, simple_loss=0.3454, pruned_loss=0.08852, over 29778.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3546, pruned_loss=0.103, over 5704257.47 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3633, pruned_loss=0.1155, over 5648763.17 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3518, pruned_loss=0.1006, over 5708728.14 frames. ], batch size: 87, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 17:16:33,575 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:35,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:37,743 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:48,308 INFO [train.py:968] (0/2) Epoch 24, batch 39950, giga_loss[loss=0.2683, simple_loss=0.3445, pruned_loss=0.096, over 28883.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3537, pruned_loss=0.1024, over 5712480.19 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3631, pruned_loss=0.1153, over 5655366.14 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3514, pruned_loss=0.1003, over 5711685.28 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 17:16:50,001 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:52,396 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:58,213 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:17:04,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.500e+02 1.238e+03 1.577e+03 2.019e+03 8.971e+03, threshold=3.155e+03, percent-clipped=3.0
+2023-03-12 17:17:14,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:17:25,910 INFO [train.py:968] (0/2) Epoch 24, batch 40000, giga_loss[loss=0.3085, simple_loss=0.377, pruned_loss=0.12, over 28738.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3529, pruned_loss=0.1024, over 5714393.28 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1157, over 5659390.63 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.09969, over 5712874.84 frames. ], batch size: 284, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:17:26,091 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:17:54,651 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:17:57,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:18:04,955 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:18:06,422 INFO [train.py:968] (0/2) Epoch 24, batch 40050, giga_loss[loss=0.2434, simple_loss=0.3113, pruned_loss=0.0877, over 28634.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3497, pruned_loss=0.1009, over 5713158.87 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1157, over 5664507.23 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.347, pruned_loss=0.09824, over 5709072.46 frames. ], batch size: 78, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:18:19,716 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:18:22,761 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.763e+02 1.218e+03 1.479e+03 2.026e+03 7.309e+03, threshold=2.958e+03, percent-clipped=9.0
+2023-03-12 17:18:25,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4532, 4.0951, 1.5988, 1.6789], device='cuda:0'), covar=tensor([0.0996, 0.0313, 0.0956, 0.1274], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0556, 0.0394, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 17:18:47,825 INFO [train.py:968] (0/2) Epoch 24, batch 40100, giga_loss[loss=0.223, simple_loss=0.3027, pruned_loss=0.07167, over 28716.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3453, pruned_loss=0.09826, over 5718018.71 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3632, pruned_loss=0.1153, over 5670356.30 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3431, pruned_loss=0.09614, over 5710723.36 frames. ], batch size: 99, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:19:07,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2623, 2.6031, 2.6242, 2.0064], device='cuda:0'), covar=tensor([0.3447, 0.2261, 0.2039, 0.2984], device='cuda:0'), in_proj_covar=tensor([0.2007, 0.1943, 0.1860, 0.2011], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 17:19:31,283 INFO [train.py:968] (0/2) Epoch 24, batch 40150, giga_loss[loss=0.2673, simple_loss=0.3461, pruned_loss=0.09424, over 28781.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3451, pruned_loss=0.09707, over 5716073.48 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3634, pruned_loss=0.1154, over 5670770.92 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3432, pruned_loss=0.0952, over 5710451.93 frames. ], batch size: 112, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:19:40,194 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:19:46,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.013e+02 1.211e+03 1.539e+03 1.954e+03 4.604e+03, threshold=3.078e+03, percent-clipped=6.0
+2023-03-12 17:20:08,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:20:08,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6860, 1.9276, 1.4048, 1.5503], device='cuda:0'), covar=tensor([0.0976, 0.0629, 0.1086, 0.1198], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0447, 0.0521, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 17:20:13,150 INFO [train.py:968] (0/2) Epoch 24, batch 40200, giga_loss[loss=0.2565, simple_loss=0.3452, pruned_loss=0.08387, over 28579.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.346, pruned_loss=0.09657, over 5711655.08 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3631, pruned_loss=0.1153, over 5675730.62 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3444, pruned_loss=0.09481, over 5703531.73 frames. ], batch size: 336, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:20:53,732 INFO [train.py:968] (0/2) Epoch 24, batch 40250, giga_loss[loss=0.2333, simple_loss=0.3201, pruned_loss=0.07329, over 28713.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3456, pruned_loss=0.09638, over 5714425.54 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3633, pruned_loss=0.1155, over 5679841.93 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3438, pruned_loss=0.09449, over 5705005.94 frames. ], batch size: 262, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:20:55,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:21:11,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.619e+02 1.206e+03 1.532e+03 2.183e+03 5.455e+03, threshold=3.065e+03, percent-clipped=10.0
+2023-03-12 17:21:21,056 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-12 17:21:35,452 INFO [train.py:968] (0/2) Epoch 24, batch 40300, giga_loss[loss=0.2714, simple_loss=0.3442, pruned_loss=0.09933, over 28883.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3445, pruned_loss=0.09707, over 5709488.98 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3637, pruned_loss=0.1159, over 5674549.10 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3421, pruned_loss=0.09466, over 5708600.01 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:21:38,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6794, 1.8550, 1.3750, 1.3836], device='cuda:0'), covar=tensor([0.0937, 0.0639, 0.1031, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0399, 0.0446, 0.0521, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 17:21:39,585 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:21:41,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:21:44,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:21:50,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5086, 1.5296, 1.7055, 1.3189], device='cuda:0'), covar=tensor([0.1885, 0.2774, 0.1643, 0.1867], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0705, 0.0960, 0.0861], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 17:21:55,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1423, 1.8139, 1.4141, 0.4823], device='cuda:0'), covar=tensor([0.4977, 0.2742, 0.4484, 0.6373], device='cuda:0'), in_proj_covar=tensor([0.1785, 0.1673, 0.1620, 0.1449], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 17:22:07,929 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:22:08,023 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:22:11,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:22:19,608 INFO [train.py:968] (0/2) Epoch 24, batch 40350, giga_loss[loss=0.2277, simple_loss=0.2941, pruned_loss=0.08061, over 28600.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3423, pruned_loss=0.09708, over 5717022.06 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3639, pruned_loss=0.116, over 5675551.78 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3401, pruned_loss=0.09492, over 5715766.44 frames. ], batch size: 71, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:22:28,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8222, 1.1701, 5.0197, 3.5892], device='cuda:0'), covar=tensor([0.1609, 0.3030, 0.0387, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0771, 0.0657, 0.0969, 0.0932], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 17:22:37,957 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:22:38,851 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.923e+02 1.309e+03 1.616e+03 2.271e+03 6.386e+03, threshold=3.232e+03, percent-clipped=11.0
+2023-03-12 17:22:42,971 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:23:01,228 INFO [train.py:968] (0/2) Epoch 24, batch 40400, giga_loss[loss=0.26, simple_loss=0.334, pruned_loss=0.09304, over 28997.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3418, pruned_loss=0.09831, over 5696238.38 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3648, pruned_loss=0.1166, over 5661419.60 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3386, pruned_loss=0.09548, over 5709076.55 frames. ], batch size: 128, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:23:19,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:23:42,197 INFO [train.py:968] (0/2) Epoch 24, batch 40450, giga_loss[loss=0.2693, simple_loss=0.3474, pruned_loss=0.09562, over 28703.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3418, pruned_loss=0.09903, over 5701089.26 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3649, pruned_loss=0.1167, over 5665942.30 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3388, pruned_loss=0.09633, over 5707915.76 frames. ], batch size: 242, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:23:45,128 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:23:49,003 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:23:49,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:23:58,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.938e+02 1.284e+03 1.587e+03 2.485e+03 5.724e+03, threshold=3.174e+03, percent-clipped=16.0
+2023-03-12 17:24:10,909 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:24:20,610 INFO [train.py:968] (0/2) Epoch 24, batch 40500, giga_loss[loss=0.2533, simple_loss=0.3327, pruned_loss=0.08696, over 28896.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3391, pruned_loss=0.09759, over 5687976.36 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3647, pruned_loss=0.1166, over 5657488.09 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3361, pruned_loss=0.09493, over 5702614.56 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:24:35,932 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:24:38,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:01,965 INFO [train.py:968] (0/2) Epoch 24, batch 40550, giga_loss[loss=0.2044, simple_loss=0.2902, pruned_loss=0.05933, over 28901.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3351, pruned_loss=0.09578, over 5688785.13 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3643, pruned_loss=0.1165, over 5651076.15 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3325, pruned_loss=0.09333, over 5707252.26 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:25:04,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:11,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:14,706 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:17,378 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:20,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.258e+02 1.419e+03 1.876e+03 2.636e+03 9.264e+03, threshold=3.752e+03, percent-clipped=15.0
+2023-03-12 17:25:33,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8751, 3.7007, 3.4942, 1.6313], device='cuda:0'), covar=tensor([0.0742, 0.0849, 0.0755, 0.2259], device='cuda:0'), in_proj_covar=tensor([0.1255, 0.1162, 0.0977, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 17:25:40,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2922, 3.1475, 2.9581, 1.3888], device='cuda:0'), covar=tensor([0.0924, 0.0990, 0.0842, 0.2309], device='cuda:0'), in_proj_covar=tensor([0.1256, 0.1163, 0.0977, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 17:25:40,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9907, 1.2872, 1.0661, 0.3028], device='cuda:0'), covar=tensor([0.4726, 0.3577, 0.5175, 0.7366], device='cuda:0'), in_proj_covar=tensor([0.1788, 0.1676, 0.1623, 0.1453], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 17:25:42,281 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:43,946 INFO [train.py:968] (0/2) Epoch 24, batch 40600, giga_loss[loss=0.2566, simple_loss=0.325, pruned_loss=0.09411, over 28639.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3312, pruned_loss=0.09363, over 5686058.65 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3644, pruned_loss=0.1166, over 5646459.93 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3279, pruned_loss=0.09091, over 5706329.63 frames. ], batch size: 307, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:25:44,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5472, 2.8056, 2.6794, 2.1712], device='cuda:0'), covar=tensor([0.2119, 0.1773, 0.1838, 0.2127], device='cuda:0'), in_proj_covar=tensor([0.2017, 0.1951, 0.1872, 0.2017], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 17:26:05,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:26:08,789 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:26:23,145 INFO [train.py:968] (0/2) Epoch 24, batch 40650, giga_loss[loss=0.2386, simple_loss=0.3153, pruned_loss=0.08098, over 28450.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3312, pruned_loss=0.09316, over 5700483.45 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3647, pruned_loss=0.1166, over 5652515.66 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3273, pruned_loss=0.09021, over 5712939.30 frames. ], batch size: 71, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:26:42,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.061e+02 1.350e+03 1.774e+03 2.509e+03 5.699e+03, threshold=3.548e+03, percent-clipped=8.0
+2023-03-12 17:27:05,529 INFO [train.py:968] (0/2) Epoch 24, batch 40700, giga_loss[loss=0.2812, simple_loss=0.3601, pruned_loss=0.1011, over 28986.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3332, pruned_loss=0.09385, over 5699837.92 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.364, pruned_loss=0.1163, over 5657224.26 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3299, pruned_loss=0.09117, over 5706746.76 frames. ], batch size: 164, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:27:46,823 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:27:49,372 INFO [train.py:968] (0/2) Epoch 24, batch 40750, giga_loss[loss=0.2946, simple_loss=0.359, pruned_loss=0.1151, over 28893.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3367, pruned_loss=0.09528, over 5709013.91 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3638, pruned_loss=0.116, over 5664287.21 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3334, pruned_loss=0.09276, over 5709077.63 frames. ], batch size: 106, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:28:04,861 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:28:06,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.291e+02 1.387e+03 1.690e+03 2.172e+03 7.891e+03, threshold=3.381e+03, percent-clipped=5.0
+2023-03-12 17:28:06,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:28:29,495 INFO [train.py:968] (0/2) Epoch 24, batch 40800, giga_loss[loss=0.2706, simple_loss=0.3467, pruned_loss=0.09723, over 28821.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3392, pruned_loss=0.09572, over 5706697.87 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3637, pruned_loss=0.1159, over 5666483.21 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3365, pruned_loss=0.09364, over 5705074.07 frames. ], batch size: 199, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:28:30,330 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:28:55,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:28:57,211 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-12 17:29:12,329 INFO [train.py:968] (0/2) Epoch 24, batch 40850, giga_loss[loss=0.2542, simple_loss=0.3365, pruned_loss=0.0859, over 28920.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.343, pruned_loss=0.09743, over 5707605.29 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3639, pruned_loss=0.116, over 5671626.96 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3401, pruned_loss=0.09533, over 5702661.43 frames. ], batch size: 227, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:29:28,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5030, 1.6054, 1.7077, 1.2990], device='cuda:0'), covar=tensor([0.1786, 0.2464, 0.1478, 0.1722], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0707, 0.0962, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 17:29:32,261 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.768e+02 1.214e+03 1.444e+03 1.923e+03 5.473e+03, threshold=2.887e+03, percent-clipped=2.0
+2023-03-12 17:29:53,574 INFO [train.py:968] (0/2) Epoch 24, batch 40900, giga_loss[loss=0.2552, simple_loss=0.3344, pruned_loss=0.08803, over 28748.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3454, pruned_loss=0.09882, over 5701542.55 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3631, pruned_loss=0.1155, over 5666028.94 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3429, pruned_loss=0.0967, over 5704066.63 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:30:31,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:30:51,757 INFO [train.py:968] (0/2) Epoch 24, batch 40950, giga_loss[loss=0.2947, simple_loss=0.3703, pruned_loss=0.1095, over 28897.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3513, pruned_loss=0.1044, over 5699530.38 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3631, pruned_loss=0.1155, over 5666028.94 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3494, pruned_loss=0.1027, over 5701494.90 frames. ], batch size: 227, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:31:14,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.705e+03 2.306e+03 3.226e+03 1.483e+04, threshold=4.612e+03, percent-clipped=32.0
+2023-03-12 17:31:16,132 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:31:18,109 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6506, 1.6639, 1.8664, 1.4200], device='cuda:0'), covar=tensor([0.1745, 0.2528, 0.1447, 0.1751], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0706, 0.0961, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 17:31:21,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:31:40,175 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:31:41,558 INFO [train.py:968] (0/2) Epoch 24, batch 41000, giga_loss[loss=0.3392, simple_loss=0.4089, pruned_loss=0.1347, over 28614.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3571, pruned_loss=0.1085, over 5698306.62 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.363, pruned_loss=0.1154, over 5669544.82 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3555, pruned_loss=0.1071, over 5697219.29 frames. ], batch size: 78, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:31:45,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:31:46,813 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:32:25,611 INFO [train.py:968] (0/2) Epoch 24, batch 41050, giga_loss[loss=0.3193, simple_loss=0.3842, pruned_loss=0.1272, over 28312.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.364, pruned_loss=0.1136, over 5698988.10 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.363, pruned_loss=0.1153, over 5675985.19 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3627, pruned_loss=0.1125, over 5693059.68 frames. ], batch size: 368, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:32:46,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.430e+02 1.829e+03 2.404e+03 2.946e+03 8.570e+03, threshold=4.807e+03, percent-clipped=2.0
+2023-03-12 17:32:52,563 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:32:54,345 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:32:54,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:33:02,990 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:33:10,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3769, 2.0400, 1.5535, 0.6102], device='cuda:0'), covar=tensor([0.4860, 0.2775, 0.4354, 0.6156], device='cuda:0'), in_proj_covar=tensor([0.1791, 0.1687, 0.1627, 0.1454], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 17:33:12,427 INFO [train.py:968] (0/2) Epoch 24, batch 41100, giga_loss[loss=0.3505, simple_loss=0.4084, pruned_loss=0.1463, over 28501.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3683, pruned_loss=0.117, over 5700005.64 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3625, pruned_loss=0.1149, over 5681734.50 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3677, pruned_loss=0.1165, over 5690776.39 frames. ], batch size: 336, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:33:13,999 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1090000.pt
+2023-03-12 17:33:20,705 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:33:31,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:33:55,692 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:33:56,714 INFO [train.py:968] (0/2) Epoch 24, batch 41150, giga_loss[loss=0.296, simple_loss=0.3727, pruned_loss=0.1096, over 28881.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3746, pruned_loss=0.1224, over 5698490.02 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3622, pruned_loss=0.1146, over 5686307.98 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3748, pruned_loss=0.1224, over 5687547.72 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:33:57,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:34:21,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.948e+03 2.451e+03 3.209e+03 1.015e+04, threshold=4.902e+03, percent-clipped=9.0
+2023-03-12 17:34:28,848 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:34:48,758 INFO [train.py:968] (0/2) Epoch 24, batch 41200, giga_loss[loss=0.2879, simple_loss=0.3525, pruned_loss=0.1116, over 28456.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3786, pruned_loss=0.1263, over 5661804.92 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3627, pruned_loss=0.115, over 5671720.01 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3786, pruned_loss=0.1262, over 5667116.13 frames. ], batch size: 60, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:35:03,763 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1090111.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 17:35:09,479 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 17:35:41,791 INFO [train.py:968] (0/2) Epoch 24, batch 41250, giga_loss[loss=0.3303, simple_loss=0.3867, pruned_loss=0.1369, over 28681.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3812, pruned_loss=0.1295, over 5663933.46 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3626, pruned_loss=0.115, over 5678698.09 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3818, pruned_loss=0.1298, over 5661325.59 frames. ], batch size: 242, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:35:59,210 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:36:03,591 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:36:07,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.151e+03 1.848e+03 2.315e+03 3.081e+03 7.887e+03, threshold=4.630e+03, percent-clipped=11.0
+2023-03-12 17:36:32,303 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:36:34,269 INFO [train.py:968] (0/2) Epoch 24, batch 41300, giga_loss[loss=0.3158, simple_loss=0.3713, pruned_loss=0.1301, over 28286.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3846, pruned_loss=0.1337, over 5660357.69 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3623, pruned_loss=0.1147, over 5683529.06 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3861, pruned_loss=0.1348, over 5653036.54 frames. ], batch size: 77, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:36:38,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7658, 2.0888, 2.0385, 1.7152], device='cuda:0'), covar=tensor([0.1653, 0.1591, 0.1727, 0.1808], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0750, 0.0720, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 17:37:27,061 INFO [train.py:968] (0/2) Epoch 24, batch 41350, giga_loss[loss=0.3305, simple_loss=0.3985, pruned_loss=0.1313, over 28653.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3868, pruned_loss=0.1363, over 5644848.33 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3623, pruned_loss=0.1147, over 5688173.49 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3886, pruned_loss=0.1376, over 5634206.35 frames. ], batch size: 242, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:37:50,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.417e+03 1.945e+03 2.629e+03 3.622e+03 9.919e+03, threshold=5.258e+03, percent-clipped=11.0
+2023-03-12 17:37:53,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:37:57,274 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1960, 1.7632, 1.3236, 0.4298], device='cuda:0'), covar=tensor([0.4570, 0.2877, 0.4062, 0.6252], device='cuda:0'), in_proj_covar=tensor([0.1793, 0.1690, 0.1629, 0.1454], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 17:38:17,365 INFO [train.py:968] (0/2) Epoch 24, batch 41400, giga_loss[loss=0.3237, simple_loss=0.3815, pruned_loss=0.133, over 28941.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3889, pruned_loss=0.1377, over 5624840.32 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3623, pruned_loss=0.1147, over 5674318.86 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3909, pruned_loss=0.1394, over 5627358.97 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:38:56,695 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1090333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:39:11,555 INFO [train.py:968] (0/2) Epoch 24, batch 41450, giga_loss[loss=0.2807, simple_loss=0.3536, pruned_loss=0.1039, over 28921.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3895, pruned_loss=0.1392, over 5622339.01 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3623, pruned_loss=0.1146, over 5680009.25 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3918, pruned_loss=0.1412, over 5618116.89 frames. ], batch size: 174, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:39:12,911 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 17:39:20,119 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:39:29,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:39:31,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1322, 1.2912, 3.2843, 2.9597], device='cuda:0'), covar=tensor([0.1578, 0.2581, 0.0534, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0660, 0.0978, 0.0937], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 17:39:35,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.332e+03 2.031e+03 2.714e+03 3.481e+03 8.816e+03, threshold=5.429e+03, percent-clipped=4.0
+2023-03-12 17:40:02,493 INFO [train.py:968] (0/2) Epoch 24, batch 41500, giga_loss[loss=0.2802, simple_loss=0.3472, pruned_loss=0.1066, over 28567.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3878, pruned_loss=0.1381, over 5642706.41 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3622, pruned_loss=0.1145, over 5684717.20 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3903, pruned_loss=0.1403, over 5634035.65 frames. ], batch size: 71, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:40:21,634 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:40:24,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:40:53,318 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:40:53,850 INFO [train.py:968] (0/2) Epoch 24, batch 41550, giga_loss[loss=0.3248, simple_loss=0.3949, pruned_loss=0.1273, over 28790.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3881, pruned_loss=0.1371, over 5649216.04 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3625, pruned_loss=0.1147, over 5677232.57 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3904, pruned_loss=0.1391, over 5647401.14 frames. ], batch size: 243, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 17:41:21,078 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.819e+03 2.365e+03 3.591e+03 7.433e+03, threshold=4.730e+03, percent-clipped=5.0
+2023-03-12 17:41:36,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090486.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 17:41:49,432 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:41:49,845 INFO [train.py:968] (0/2) Epoch 24, batch 41600, libri_loss[loss=0.2894, simple_loss=0.3586, pruned_loss=0.1101, over 29518.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3869, pruned_loss=0.1351, over 5648273.64 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3625, pruned_loss=0.1148, over 5671084.36 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3891, pruned_loss=0.1371, over 5651287.88 frames. ], batch size: 80, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:41:51,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:41:57,673 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:42:00,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:42:16,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3611, 2.8988, 1.3763, 1.5076], device='cuda:0'), covar=tensor([0.0972, 0.0388, 0.0913, 0.1332], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0561, 0.0395, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 17:42:22,836 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:42:36,901 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:42:44,242 INFO [train.py:968] (0/2) Epoch 24, batch 41650, giga_loss[loss=0.4509, simple_loss=0.4608, pruned_loss=0.2204, over 26352.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3889, pruned_loss=0.1363, over 5646814.11 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3626, pruned_loss=0.1148, over 5673896.25 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.391, pruned_loss=0.1382, over 5646346.61 frames. ], batch size: 555, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:42:57,005 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1090563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:43:10,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.716e+03 2.173e+03 3.200e+03 7.430e+03, threshold=4.345e+03, percent-clipped=2.0
+2023-03-12 17:43:37,001 INFO [train.py:968] (0/2) Epoch 24, batch 41700, giga_loss[loss=0.4323, simple_loss=0.4415, pruned_loss=0.2116, over 23984.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3862, pruned_loss=0.1339, over 5645741.96 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3622, pruned_loss=0.1145, over 5676721.50 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3887, pruned_loss=0.1361, over 5642095.95 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:43:46,921 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1090608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:43:57,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6363, 1.7238, 1.7381, 1.5749], device='cuda:0'), covar=tensor([0.2976, 0.2764, 0.2202, 0.2472], device='cuda:0'), in_proj_covar=tensor([0.2025, 0.1956, 0.1869, 0.2018], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 17:44:05,982 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090629.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 17:44:10,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090632.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 17:44:24,660 INFO [train.py:968] (0/2) Epoch 24, batch 41750, giga_loss[loss=0.3063, simple_loss=0.3791, pruned_loss=0.1167, over 28716.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3838, pruned_loss=0.1309, over 5650958.21 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3621, pruned_loss=0.1143, over 5683850.37 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3866, pruned_loss=0.1333, over 5640867.75 frames. ], batch size: 262, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:44:37,337 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090661.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 17:44:40,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3712, 1.9636, 1.4278, 0.6865], device='cuda:0'), covar=tensor([0.4845, 0.2596, 0.3414, 0.5818], device='cuda:0'), in_proj_covar=tensor([0.1788, 0.1684, 0.1625, 0.1452], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 17:44:49,225 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.840e+03 2.416e+03 3.328e+03 8.477e+03, threshold=4.832e+03, percent-clipped=8.0
+2023-03-12 17:44:52,357 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4793, 1.6928, 1.7130, 1.2509], device='cuda:0'), covar=tensor([0.1778, 0.2909, 0.1561, 0.1980], device='cuda:0'), in_proj_covar=tensor([0.0910, 0.0705, 0.0955, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 17:45:03,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2610, 1.8714, 1.5600, 1.4527], device='cuda:0'), covar=tensor([0.0815, 0.0315, 0.0303, 0.1023], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 17:45:16,979 INFO [train.py:968] (0/2) Epoch 24, batch 41800, giga_loss[loss=0.336, simple_loss=0.3762, pruned_loss=0.1479, over 23942.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3807, pruned_loss=0.1277, over 5662956.82 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.362, pruned_loss=0.1144, over 5687447.43 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3834, pruned_loss=0.1298, over 5651298.03 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:45:26,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:46:09,191 INFO [train.py:968] (0/2) Epoch 24, batch 41850, libri_loss[loss=0.2916, simple_loss=0.3629, pruned_loss=0.1102, over 29222.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3776, pruned_loss=0.1251, over 5657388.67 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3625, pruned_loss=0.1148, over 5682564.87 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3796, pruned_loss=0.1267, over 5651517.35 frames. ], batch size: 97, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:46:30,080 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.734e+03 2.437e+03 2.958e+03 1.251e+04, threshold=4.873e+03, percent-clipped=4.0
+2023-03-12 17:46:55,386 INFO [train.py:968] (0/2) Epoch 24, batch 41900, giga_loss[loss=0.2956, simple_loss=0.3694, pruned_loss=0.1108, over 29018.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3734, pruned_loss=0.1227, over 5651408.56 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1143, over 5690692.82 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3762, pruned_loss=0.1248, over 5638607.19 frames. ], batch size: 128, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:47:19,913 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5331, 4.3445, 4.1298, 1.8617], device='cuda:0'), covar=tensor([0.0733, 0.0930, 0.1097, 0.2160], device='cuda:0'), in_proj_covar=tensor([0.1271, 0.1176, 0.0989, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 17:47:44,256 INFO [train.py:968] (0/2) Epoch 24, batch 41950, giga_loss[loss=0.3177, simple_loss=0.3869, pruned_loss=0.1243, over 28692.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3726, pruned_loss=0.1217, over 5665775.92 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3613, pruned_loss=0.1139, over 5693595.39 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3755, pruned_loss=0.1239, over 5652290.42 frames. ], batch size: 242, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:47:46,693 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:47:48,652 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:48:06,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.526e+03 1.848e+03 2.484e+03 5.931e+03, threshold=3.697e+03, percent-clipped=1.0
+2023-03-12 17:48:18,337 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:48:36,298 INFO [train.py:968] (0/2) Epoch 24, batch 42000, giga_loss[loss=0.3083, simple_loss=0.3707, pruned_loss=0.1229, over 28871.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3723, pruned_loss=0.1212, over 5671489.69 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3613, pruned_loss=0.114, over 5696590.05 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.375, pruned_loss=0.1231, over 5657549.04 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:48:36,303 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 17:48:45,309 INFO [train.py:1012] (0/2) Epoch 24, validation: loss=0.2016, simple_loss=0.3092, pruned_loss=0.04703, over 944034.00 frames. 
+2023-03-12 17:48:45,310 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 17:48:55,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3418, 1.8726, 1.3453, 0.7338], device='cuda:0'), covar=tensor([0.5760, 0.2568, 0.3694, 0.6794], device='cuda:0'), in_proj_covar=tensor([0.1788, 0.1680, 0.1620, 0.1451], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 17:49:28,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:49:35,486 INFO [train.py:968] (0/2) Epoch 24, batch 42050, giga_loss[loss=0.2456, simple_loss=0.3266, pruned_loss=0.08235, over 28933.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3694, pruned_loss=0.1182, over 5680231.71 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3611, pruned_loss=0.1139, over 5699098.76 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.372, pruned_loss=0.12, over 5666119.84 frames. ], batch size: 112, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:50:00,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.632e+03 2.073e+03 2.717e+03 1.082e+04, threshold=4.147e+03, percent-clipped=11.0
+2023-03-12 17:50:09,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:50:25,096 INFO [train.py:968] (0/2) Epoch 24, batch 42100, giga_loss[loss=0.2769, simple_loss=0.3649, pruned_loss=0.09449, over 29015.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3717, pruned_loss=0.1177, over 5681389.86 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.362, pruned_loss=0.1146, over 5695268.28 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3734, pruned_loss=0.1188, over 5671900.90 frames. ], batch size: 136, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:50:57,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4993, 1.7215, 1.7208, 1.2783], device='cuda:0'), covar=tensor([0.1812, 0.2808, 0.1570, 0.1944], device='cuda:0'), in_proj_covar=tensor([0.0912, 0.0706, 0.0955, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 17:51:07,072 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:51:18,748 INFO [train.py:968] (0/2) Epoch 24, batch 42150, giga_loss[loss=0.2926, simple_loss=0.3446, pruned_loss=0.1203, over 23662.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3736, pruned_loss=0.1185, over 5675657.84 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3616, pruned_loss=0.1142, over 5698535.05 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3756, pruned_loss=0.1197, over 5664915.17 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:51:39,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 2.004e+03 2.417e+03 3.323e+03 1.055e+04, threshold=4.834e+03, percent-clipped=7.0
+2023-03-12 17:51:49,460 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1091081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:51:52,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1091084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:52:05,024 INFO [train.py:968] (0/2) Epoch 24, batch 42200, giga_loss[loss=0.3271, simple_loss=0.3849, pruned_loss=0.1347, over 28889.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3741, pruned_loss=0.1199, over 5669018.79 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3613, pruned_loss=0.1142, over 5692179.03 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3763, pruned_loss=0.121, over 5666395.64 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:52:05,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3022, 1.8045, 1.4124, 0.5441], device='cuda:0'), covar=tensor([0.3958, 0.2513, 0.3701, 0.5656], device='cuda:0'), in_proj_covar=tensor([0.1779, 0.1673, 0.1617, 0.1447], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 17:52:22,220 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1091113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:52:33,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1091126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:52:36,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1091129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:52:48,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3031, 1.6034, 1.2783, 0.9982], device='cuda:0'), covar=tensor([0.2492, 0.2577, 0.2987, 0.2323], device='cuda:0'), in_proj_covar=tensor([0.1549, 0.1119, 0.1369, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 17:52:53,599 INFO [train.py:968] (0/2) Epoch 24, batch 42250, giga_loss[loss=0.2854, simple_loss=0.36, pruned_loss=0.1054, over 29047.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3735, pruned_loss=0.1196, over 5674878.11 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.1141, over 5696230.20 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3755, pruned_loss=0.1207, over 5668645.12 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:53:02,742 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1091158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:53:12,055 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3384, 2.1829, 1.8613, 1.8224], device='cuda:0'), covar=tensor([0.0858, 0.0681, 0.0925, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0453, 0.0524, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 17:53:15,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.318e+03 1.813e+03 2.302e+03 3.433e+03 1.236e+04, threshold=4.605e+03, percent-clipped=7.0
+2023-03-12 17:53:39,568 INFO [train.py:968] (0/2) Epoch 24, batch 42300, giga_loss[loss=0.2733, simple_loss=0.3433, pruned_loss=0.1017, over 28821.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3714, pruned_loss=0.1193, over 5676613.28 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3611, pruned_loss=0.1139, over 5697532.31 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3733, pruned_loss=0.1204, over 5670244.30 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:54:06,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:54:28,327 INFO [train.py:968] (0/2) Epoch 24, batch 42350, giga_loss[loss=0.3206, simple_loss=0.3902, pruned_loss=0.1255, over 28543.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.37, pruned_loss=0.1194, over 5661447.82 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3615, pruned_loss=0.1144, over 5692294.08 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3716, pruned_loss=0.1201, over 5659349.70 frames. ], batch size: 85, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:54:32,054 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0668, 4.9013, 4.6737, 2.1728], device='cuda:0'), covar=tensor([0.0495, 0.0635, 0.0678, 0.1999], device='cuda:0'), in_proj_covar=tensor([0.1274, 0.1181, 0.0992, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 17:54:53,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.164e+03 1.841e+03 2.432e+03 3.197e+03 6.994e+03, threshold=4.863e+03, percent-clipped=12.0
+2023-03-12 17:55:15,388 INFO [train.py:968] (0/2) Epoch 24, batch 42400, giga_loss[loss=0.2616, simple_loss=0.3346, pruned_loss=0.09426, over 28866.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.369, pruned_loss=0.1184, over 5654749.44 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.115, over 5679530.54 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3701, pruned_loss=0.1185, over 5663444.88 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:55:58,964 INFO [train.py:968] (0/2) Epoch 24, batch 42450, giga_loss[loss=0.3028, simple_loss=0.3656, pruned_loss=0.12, over 28937.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3677, pruned_loss=0.1161, over 5677938.92 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.362, pruned_loss=0.115, over 5687978.85 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3689, pruned_loss=0.1163, over 5676944.76 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:56:25,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.545e+03 2.107e+03 2.736e+03 5.884e+03, threshold=4.213e+03, percent-clipped=3.0
+2023-03-12 17:56:50,342 INFO [train.py:968] (0/2) Epoch 24, batch 42500, giga_loss[loss=0.2847, simple_loss=0.3567, pruned_loss=0.1064, over 28285.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3683, pruned_loss=0.1158, over 5673428.66 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3622, pruned_loss=0.115, over 5682842.99 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3692, pruned_loss=0.1159, over 5676211.16 frames. ], batch size: 77, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:57:01,842 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1091411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:57:41,097 INFO [train.py:968] (0/2) Epoch 24, batch 42550, giga_loss[loss=0.3127, simple_loss=0.3746, pruned_loss=0.1254, over 27934.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.367, pruned_loss=0.1152, over 5678112.15 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3618, pruned_loss=0.1148, over 5685255.71 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3681, pruned_loss=0.1155, over 5678308.37 frames. ], batch size: 412, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:58:03,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.254e+02 1.706e+03 2.155e+03 2.721e+03 5.824e+03, threshold=4.309e+03, percent-clipped=8.0
+2023-03-12 17:58:24,619 INFO [train.py:968] (0/2) Epoch 24, batch 42600, giga_loss[loss=0.2733, simple_loss=0.3488, pruned_loss=0.09887, over 28747.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3662, pruned_loss=0.1155, over 5679222.44 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1151, over 5687326.40 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3669, pruned_loss=0.1155, over 5677431.54 frames. ], batch size: 262, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:58:25,740 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:58:41,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0352, 2.2756, 1.8148, 2.4060], device='cuda:0'), covar=tensor([0.2466, 0.2618, 0.2950, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.1551, 0.1119, 0.1369, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 17:59:10,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3730, 2.0190, 1.4896, 0.6166], device='cuda:0'), covar=tensor([0.5751, 0.3196, 0.4789, 0.6948], device='cuda:0'), in_proj_covar=tensor([0.1781, 0.1674, 0.1617, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 17:59:12,597 INFO [train.py:968] (0/2) Epoch 24, batch 42650, giga_loss[loss=0.3291, simple_loss=0.3773, pruned_loss=0.1405, over 26643.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3652, pruned_loss=0.1154, over 5673648.75 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3616, pruned_loss=0.1146, over 5692841.76 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3663, pruned_loss=0.1159, over 5666894.53 frames. ], batch size: 555, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:59:20,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1091554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:59:22,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1091557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:59:28,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-12 17:59:30,660 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091566.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 17:59:33,236 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-12 17:59:40,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.778e+03 2.775e+03 3.693e+03 7.489e+03, threshold=5.550e+03, percent-clipped=12.0
+2023-03-12 17:59:51,012 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1091586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:00:02,963 INFO [train.py:968] (0/2) Epoch 24, batch 42700, giga_loss[loss=0.2658, simple_loss=0.3417, pruned_loss=0.09495, over 29030.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3656, pruned_loss=0.1169, over 5666539.54 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3618, pruned_loss=0.1147, over 5691186.62 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3664, pruned_loss=0.1172, over 5662465.26 frames. ], batch size: 164, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:00:03,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1091599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:00:29,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4094, 1.2438, 3.8295, 3.2441], device='cuda:0'), covar=tensor([0.1537, 0.2740, 0.0497, 0.1002], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0659, 0.0977, 0.0936], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 18:00:55,240 INFO [train.py:968] (0/2) Epoch 24, batch 42750, giga_loss[loss=0.2612, simple_loss=0.3335, pruned_loss=0.0944, over 28905.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3637, pruned_loss=0.1159, over 5667123.50 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3622, pruned_loss=0.115, over 5681208.56 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3639, pruned_loss=0.1159, over 5672469.23 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:01:23,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.144e+02 1.686e+03 2.348e+03 3.097e+03 8.530e+03, threshold=4.696e+03, percent-clipped=3.0
+2023-03-12 18:01:47,505 INFO [train.py:968] (0/2) Epoch 24, batch 42800, libri_loss[loss=0.2822, simple_loss=0.3594, pruned_loss=0.1026, over 28641.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3628, pruned_loss=0.1155, over 5675039.76 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3619, pruned_loss=0.1147, over 5685845.27 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3634, pruned_loss=0.1158, over 5674523.53 frames. ], batch size: 106, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 18:02:29,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1091742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:02:31,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1091745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:02:33,466 INFO [train.py:968] (0/2) Epoch 24, batch 42850, giga_loss[loss=0.3111, simple_loss=0.3787, pruned_loss=0.1217, over 28780.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3635, pruned_loss=0.1157, over 5683906.36 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3617, pruned_loss=0.1145, over 5693033.82 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3642, pruned_loss=0.1161, over 5676644.41 frames. ], batch size: 92, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:02:45,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5326, 1.8719, 1.4769, 1.5720], device='cuda:0'), covar=tensor([0.2741, 0.2682, 0.3217, 0.2363], device='cuda:0'), in_proj_covar=tensor([0.1553, 0.1120, 0.1369, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 18:02:57,787 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:02:59,165 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1091774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:03:00,050 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.855e+03 2.490e+03 3.735e+03 8.582e+03, threshold=4.980e+03, percent-clipped=16.0
+2023-03-12 18:03:22,213 INFO [train.py:968] (0/2) Epoch 24, batch 42900, giga_loss[loss=0.3008, simple_loss=0.3725, pruned_loss=0.1146, over 28528.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3649, pruned_loss=0.116, over 5676252.61 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3619, pruned_loss=0.1147, over 5686153.30 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3653, pruned_loss=0.1162, over 5676077.38 frames. ], batch size: 336, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:04:07,943 INFO [train.py:968] (0/2) Epoch 24, batch 42950, libri_loss[loss=0.3171, simple_loss=0.3626, pruned_loss=0.1357, over 29556.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3652, pruned_loss=0.1156, over 5679788.75 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3616, pruned_loss=0.1143, over 5689746.15 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.366, pruned_loss=0.1162, over 5675871.92 frames. ], batch size: 77, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:04:18,193 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-12 18:04:29,829 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1091874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:04:31,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.905e+02 1.717e+03 2.177e+03 3.036e+03 9.870e+03, threshold=4.353e+03, percent-clipped=2.0
+2023-03-12 18:04:51,283 INFO [train.py:968] (0/2) Epoch 24, batch 43000, giga_loss[loss=0.3845, simple_loss=0.4115, pruned_loss=0.1787, over 26580.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3652, pruned_loss=0.1152, over 5679271.30 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5693805.98 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3663, pruned_loss=0.116, over 5671549.46 frames. ], batch size: 555, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:05:25,777 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 18:05:34,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1091941.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:05:40,753 INFO [train.py:968] (0/2) Epoch 24, batch 43050, giga_loss[loss=0.4258, simple_loss=0.4439, pruned_loss=0.2039, over 23805.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.367, pruned_loss=0.117, over 5660654.30 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.114, over 5685857.06 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3678, pruned_loss=0.1176, over 5661048.07 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:06:09,501 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.924e+02 1.665e+03 2.208e+03 2.858e+03 8.236e+03, threshold=4.416e+03, percent-clipped=6.0
+2023-03-12 18:06:31,342 INFO [train.py:968] (0/2) Epoch 24, batch 43100, giga_loss[loss=0.2619, simple_loss=0.3454, pruned_loss=0.08918, over 28940.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3705, pruned_loss=0.1206, over 5662205.45 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5686990.77 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3713, pruned_loss=0.1212, over 5661503.23 frames. ], batch size: 174, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:06:34,754 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1092000.pt
+2023-03-12 18:06:52,313 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1092017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:06:57,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1092020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:07:27,341 INFO [train.py:968] (0/2) Epoch 24, batch 43150, giga_loss[loss=0.3865, simple_loss=0.4207, pruned_loss=0.1761, over 27867.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3718, pruned_loss=0.1231, over 5656563.36 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.114, over 5690227.94 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3724, pruned_loss=0.1236, over 5652427.90 frames. ], batch size: 412, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:07:28,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1092049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:07:54,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 1.821e+03 2.301e+03 3.313e+03 1.156e+04, threshold=4.601e+03, percent-clipped=6.0
+2023-03-12 18:07:59,824 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1092084.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 18:08:02,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1092087.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:08:17,682 INFO [train.py:968] (0/2) Epoch 24, batch 43200, giga_loss[loss=0.3482, simple_loss=0.4039, pruned_loss=0.1462, over 28312.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3731, pruned_loss=0.1251, over 5649773.96 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3615, pruned_loss=0.1142, over 5683759.97 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3739, pruned_loss=0.1257, over 5650966.71 frames. ], batch size: 368, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:08:32,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 18:08:36,196 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1092116.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 18:08:38,085 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-12 18:09:07,413 INFO [train.py:968] (0/2) Epoch 24, batch 43250, giga_loss[loss=0.2787, simple_loss=0.3475, pruned_loss=0.105, over 28654.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3709, pruned_loss=0.1234, over 5663632.35 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5685715.61 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3719, pruned_loss=0.1243, over 5662564.46 frames. ], batch size: 307, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:09:07,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1092148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:09:20,045 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1092160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:09:37,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.979e+03 2.586e+03 3.914e+03 1.018e+04, threshold=5.172e+03, percent-clipped=14.0
+2023-03-12 18:09:54,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3837, 1.5643, 1.6175, 1.2030], device='cuda:0'), covar=tensor([0.1738, 0.2470, 0.1415, 0.1661], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0708, 0.0959, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 18:09:56,801 INFO [train.py:968] (0/2) Epoch 24, batch 43300, giga_loss[loss=0.2607, simple_loss=0.333, pruned_loss=0.09418, over 28431.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3682, pruned_loss=0.1213, over 5668713.39 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3608, pruned_loss=0.1136, over 5688106.75 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3695, pruned_loss=0.1223, over 5665450.90 frames. ], batch size: 78, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:10:42,589 INFO [train.py:968] (0/2) Epoch 24, batch 43350, giga_loss[loss=0.2492, simple_loss=0.3313, pruned_loss=0.08353, over 28924.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3683, pruned_loss=0.1195, over 5680190.31 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3611, pruned_loss=0.1138, over 5690857.65 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3693, pruned_loss=0.1204, over 5674271.76 frames. ], batch size: 174, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:11:09,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.714e+03 2.215e+03 2.972e+03 1.282e+04, threshold=4.429e+03, percent-clipped=3.0
+2023-03-12 18:11:10,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4691, 3.6069, 1.5136, 1.6650], device='cuda:0'), covar=tensor([0.1035, 0.0379, 0.0944, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0565, 0.0396, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 18:11:21,494 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1092291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:11:24,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1092294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:11:28,278 INFO [train.py:968] (0/2) Epoch 24, batch 43400, libri_loss[loss=0.2333, simple_loss=0.3102, pruned_loss=0.07821, over 29377.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.366, pruned_loss=0.1178, over 5670630.30 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3604, pruned_loss=0.1132, over 5687585.51 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3677, pruned_loss=0.1193, over 5668201.85 frames. ], batch size: 71, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:11:33,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5919, 1.7387, 1.7801, 1.4573], device='cuda:0'), covar=tensor([0.3004, 0.2758, 0.2216, 0.2620], device='cuda:0'), in_proj_covar=tensor([0.2017, 0.1952, 0.1869, 0.2015], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 18:11:51,286 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1092323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:12:11,758 INFO [train.py:968] (0/2) Epoch 24, batch 43450, giga_loss[loss=0.3454, simple_loss=0.3996, pruned_loss=0.1456, over 28794.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3645, pruned_loss=0.1173, over 5660977.28 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3602, pruned_loss=0.1132, over 5682553.25 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3662, pruned_loss=0.1186, over 5663203.24 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:12:16,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5308, 1.7318, 1.4368, 1.3354], device='cuda:0'), covar=tensor([0.2706, 0.2726, 0.3118, 0.2361], device='cuda:0'), in_proj_covar=tensor([0.1551, 0.1119, 0.1368, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 18:12:33,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3524, 3.5853, 1.5444, 1.6226], device='cuda:0'), covar=tensor([0.1086, 0.0372, 0.0920, 0.1381], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0566, 0.0396, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 18:12:41,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.695e+03 2.161e+03 2.963e+03 6.381e+03, threshold=4.321e+03, percent-clipped=8.0
+2023-03-12 18:13:00,435 INFO [train.py:968] (0/2) Epoch 24, batch 43500, giga_loss[loss=0.3871, simple_loss=0.4079, pruned_loss=0.1831, over 23474.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3639, pruned_loss=0.1177, over 5654720.16 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.1131, over 5675957.81 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3653, pruned_loss=0.1189, over 5662220.89 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:13:21,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3644, 1.6603, 1.6153, 1.4653], device='cuda:0'), covar=tensor([0.1983, 0.1895, 0.2264, 0.1985], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0754, 0.0722, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 18:13:30,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6879, 1.7884, 1.8761, 1.4657], device='cuda:0'), covar=tensor([0.1862, 0.2538, 0.1498, 0.1804], device='cuda:0'), in_proj_covar=tensor([0.0916, 0.0709, 0.0960, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 18:13:50,100 INFO [train.py:968] (0/2) Epoch 24, batch 43550, giga_loss[loss=0.3068, simple_loss=0.3785, pruned_loss=0.1176, over 29115.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3661, pruned_loss=0.1195, over 5657033.62 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3603, pruned_loss=0.1132, over 5678197.91 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3672, pruned_loss=0.1204, over 5660636.76 frames. ], batch size: 128, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:14:23,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.716e+03 2.300e+03 3.338e+03 7.839e+03, threshold=4.599e+03, percent-clipped=13.0
+2023-03-12 18:14:34,282 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1092490.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:14:41,575 INFO [train.py:968] (0/2) Epoch 24, batch 43600, giga_loss[loss=0.2971, simple_loss=0.381, pruned_loss=0.1067, over 28913.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3698, pruned_loss=0.1205, over 5661234.55 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1133, over 5679035.56 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3705, pruned_loss=0.1212, over 5662866.05 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:15:17,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1092535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:15:32,841 INFO [train.py:968] (0/2) Epoch 24, batch 43650, giga_loss[loss=0.3467, simple_loss=0.4081, pruned_loss=0.1427, over 27999.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3727, pruned_loss=0.1199, over 5657191.14 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3607, pruned_loss=0.1135, over 5681852.75 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3734, pruned_loss=0.1205, over 5655567.85 frames. ], batch size: 412, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:16:03,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.807e+03 2.319e+03 3.440e+03 8.321e+03, threshold=4.638e+03, percent-clipped=12.0
+2023-03-12 18:16:21,526 INFO [train.py:968] (0/2) Epoch 24, batch 43700, libri_loss[loss=0.2643, simple_loss=0.3284, pruned_loss=0.1002, over 29675.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3746, pruned_loss=0.1213, over 5663399.31 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3607, pruned_loss=0.1135, over 5679470.10 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3756, pruned_loss=0.122, over 5663185.56 frames. ], batch size: 73, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:16:33,188 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1092608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:17:11,411 INFO [train.py:968] (0/2) Epoch 24, batch 43750, giga_loss[loss=0.4028, simple_loss=0.437, pruned_loss=0.1843, over 27522.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3759, pruned_loss=0.1226, over 5671661.30 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3601, pruned_loss=0.1133, over 5682631.22 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3775, pruned_loss=0.1235, over 5668247.83 frames. ], batch size: 472, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:17:42,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.198e+02 1.979e+03 2.457e+03 3.554e+03 8.264e+03, threshold=4.913e+03, percent-clipped=9.0
+2023-03-12 18:17:42,466 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1092678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:17:46,829 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1092681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:18:02,263 INFO [train.py:968] (0/2) Epoch 24, batch 43800, giga_loss[loss=0.3032, simple_loss=0.3747, pruned_loss=0.1159, over 29052.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3759, pruned_loss=0.1233, over 5672358.92 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.36, pruned_loss=0.1133, over 5684311.51 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3775, pruned_loss=0.1242, over 5668052.66 frames. ], batch size: 164, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:18:16,614 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1092710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:18:18,156 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-12 18:18:32,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1934, 2.1438, 2.0944, 1.8967], device='cuda:0'), covar=tensor([0.2017, 0.2836, 0.2339, 0.2498], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0750, 0.0719, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 18:18:53,320 INFO [train.py:968] (0/2) Epoch 24, batch 43850, giga_loss[loss=0.3935, simple_loss=0.4213, pruned_loss=0.1828, over 26591.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3752, pruned_loss=0.1237, over 5668054.60 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.36, pruned_loss=0.1132, over 5686193.84 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3767, pruned_loss=0.1246, over 5662920.39 frames. ], batch size: 555, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:19:20,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.679e+03 2.136e+03 2.794e+03 6.672e+03, threshold=4.272e+03, percent-clipped=2.0
+2023-03-12 18:19:39,117 INFO [train.py:968] (0/2) Epoch 24, batch 43900, giga_loss[loss=0.2866, simple_loss=0.3536, pruned_loss=0.1098, over 28940.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3739, pruned_loss=0.123, over 5673264.45 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3603, pruned_loss=0.1133, over 5692971.19 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3753, pruned_loss=0.1241, over 5661983.67 frames. ], batch size: 227, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:20:28,007 INFO [train.py:968] (0/2) Epoch 24, batch 43950, giga_loss[loss=0.2816, simple_loss=0.3477, pruned_loss=0.1077, over 28841.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3729, pruned_loss=0.1236, over 5673349.97 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3602, pruned_loss=0.1132, over 5697222.71 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3744, pruned_loss=0.1247, over 5660303.22 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:20:39,246 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1092858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:20:44,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1092865.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 18:21:02,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.190e+03 2.086e+03 2.759e+03 4.309e+03 1.053e+04, threshold=5.518e+03, percent-clipped=25.0
+2023-03-12 18:21:04,843 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-12 18:21:21,187 INFO [train.py:968] (0/2) Epoch 24, batch 44000, giga_loss[loss=0.2717, simple_loss=0.3448, pruned_loss=0.09924, over 29010.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3712, pruned_loss=0.1228, over 5661922.66 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1129, over 5700308.18 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.373, pruned_loss=0.1242, over 5648198.49 frames. ], batch size: 128, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 18:22:10,643 INFO [train.py:968] (0/2) Epoch 24, batch 44050, giga_loss[loss=0.2632, simple_loss=0.3383, pruned_loss=0.09407, over 28929.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.371, pruned_loss=0.1232, over 5654687.04 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3605, pruned_loss=0.1134, over 5697508.91 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3722, pruned_loss=0.124, over 5644736.19 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:22:39,492 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.303e+03 1.946e+03 2.595e+03 3.490e+03 7.946e+03, threshold=5.189e+03, percent-clipped=9.0
+2023-03-12 18:22:43,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1092983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:22:55,161 INFO [train.py:968] (0/2) Epoch 24, batch 44100, giga_loss[loss=0.2555, simple_loss=0.3297, pruned_loss=0.0906, over 28678.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3686, pruned_loss=0.1216, over 5673004.70 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1128, over 5703857.74 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3706, pruned_loss=0.1232, over 5658190.26 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:23:05,225 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093008.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:23:08,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093011.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 18:23:39,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093040.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 18:23:46,250 INFO [train.py:968] (0/2) Epoch 24, batch 44150, giga_loss[loss=0.2877, simple_loss=0.3532, pruned_loss=0.1111, over 28805.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3681, pruned_loss=0.1216, over 5666164.66 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1127, over 5704912.45 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3698, pruned_loss=0.123, over 5653369.22 frames. ], batch size: 99, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:23:55,309 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:24:16,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.675e+03 2.300e+03 3.360e+03 1.192e+04, threshold=4.600e+03, percent-clipped=12.0
+2023-03-12 18:24:28,600 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:24:34,688 INFO [train.py:968] (0/2) Epoch 24, batch 44200, giga_loss[loss=0.2926, simple_loss=0.3693, pruned_loss=0.1079, over 28885.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3693, pruned_loss=0.1213, over 5665291.45 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5701516.48 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1224, over 5657229.35 frames. ], batch size: 227, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:25:05,741 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:25:09,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:25:17,272 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-12 18:25:20,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6112, 1.8600, 1.5158, 1.6662], device='cuda:0'), covar=tensor([0.2628, 0.2722, 0.3027, 0.2407], device='cuda:0'), in_proj_covar=tensor([0.1552, 0.1121, 0.1371, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 18:25:21,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4120, 2.0159, 1.4948, 0.6832], device='cuda:0'), covar=tensor([0.5810, 0.3055, 0.3819, 0.6599], device='cuda:0'), in_proj_covar=tensor([0.1791, 0.1689, 0.1620, 0.1454], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 18:25:25,990 INFO [train.py:968] (0/2) Epoch 24, batch 44250, giga_loss[loss=0.2679, simple_loss=0.3441, pruned_loss=0.09588, over 28945.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3727, pruned_loss=0.1236, over 5649507.14 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1132, over 5692953.94 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3734, pruned_loss=0.1244, over 5648808.72 frames. ], batch size: 136, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:25:35,097 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:25:37,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:25:55,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.724e+03 2.166e+03 3.393e+03 8.976e+03, threshold=4.331e+03, percent-clipped=10.0
+2023-03-12 18:26:15,990 INFO [train.py:968] (0/2) Epoch 24, batch 44300, giga_loss[loss=0.3988, simple_loss=0.4304, pruned_loss=0.1837, over 27481.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.372, pruned_loss=0.1235, over 5659630.95 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3601, pruned_loss=0.1131, over 5697023.80 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3732, pruned_loss=0.1245, over 5654593.92 frames. ], batch size: 472, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:26:31,750 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3985, 2.2086, 1.6271, 0.6089], device='cuda:0'), covar=tensor([0.6214, 0.3127, 0.4195, 0.7209], device='cuda:0'), in_proj_covar=tensor([0.1789, 0.1687, 0.1620, 0.1453], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 18:26:48,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1093233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:27:04,250 INFO [train.py:968] (0/2) Epoch 24, batch 44350, giga_loss[loss=0.2975, simple_loss=0.3778, pruned_loss=0.1086, over 28743.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3725, pruned_loss=0.122, over 5669125.47 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5701093.78 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3739, pruned_loss=0.1231, over 5660838.07 frames. ], batch size: 92, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:27:31,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.771e+03 2.284e+03 2.944e+03 5.669e+03, threshold=4.569e+03, percent-clipped=5.0
+2023-03-12 18:27:39,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4319, 1.3166, 3.9204, 3.3966], device='cuda:0'), covar=tensor([0.1649, 0.2862, 0.0527, 0.1053], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0661, 0.0982, 0.0940], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 18:27:43,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6218, 1.7047, 1.8645, 1.3765], device='cuda:0'), covar=tensor([0.1906, 0.2664, 0.1656, 0.1960], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0711, 0.0961, 0.0861], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 18:27:46,046 INFO [train.py:968] (0/2) Epoch 24, batch 44400, giga_loss[loss=0.3371, simple_loss=0.4014, pruned_loss=0.1364, over 28925.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3724, pruned_loss=0.1195, over 5674448.38 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3601, pruned_loss=0.1131, over 5706271.46 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3738, pruned_loss=0.1205, over 5662147.89 frames. ], batch size: 199, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:27:46,412 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 18:28:37,194 INFO [train.py:968] (0/2) Epoch 24, batch 44450, giga_loss[loss=0.3843, simple_loss=0.4147, pruned_loss=0.177, over 23460.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3749, pruned_loss=0.1202, over 5664143.82 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.1131, over 5700846.21 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3765, pruned_loss=0.1212, over 5657768.54 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:28:39,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-12 18:29:04,338 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:29:07,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:29:08,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.469e+02 1.499e+03 1.943e+03 2.797e+03 8.534e+03, threshold=3.886e+03, percent-clipped=4.0
+2023-03-12 18:29:27,058 INFO [train.py:968] (0/2) Epoch 24, batch 44500, giga_loss[loss=0.4139, simple_loss=0.4412, pruned_loss=0.1932, over 26574.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.377, pruned_loss=0.1224, over 5663231.82 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3599, pruned_loss=0.1129, over 5704533.35 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3786, pruned_loss=0.1234, over 5654049.99 frames. ], batch size: 555, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:29:29,170 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4216, 2.0270, 1.4225, 0.7234], device='cuda:0'), covar=tensor([0.6129, 0.3001, 0.3902, 0.6690], device='cuda:0'), in_proj_covar=tensor([0.1787, 0.1682, 0.1617, 0.1449], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 18:29:35,024 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.36 vs. limit=5.0
+2023-03-12 18:29:37,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:30:01,395 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1093432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:30:17,038 INFO [train.py:968] (0/2) Epoch 24, batch 44550, giga_loss[loss=0.3102, simple_loss=0.3796, pruned_loss=0.1204, over 28772.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3782, pruned_loss=0.1244, over 5659040.08 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1126, over 5705819.82 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3802, pruned_loss=0.1257, over 5649938.47 frames. ], batch size: 243, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:30:27,903 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9409, 1.3462, 1.0940, 0.2489], device='cuda:0'), covar=tensor([0.3481, 0.2712, 0.3801, 0.5072], device='cuda:0'), in_proj_covar=tensor([0.1793, 0.1687, 0.1620, 0.1453], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 18:30:39,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1093467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:30:48,871 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5266, 4.3696, 4.1232, 1.9361], device='cuda:0'), covar=tensor([0.0630, 0.0759, 0.0860, 0.2075], device='cuda:0'), in_proj_covar=tensor([0.1277, 0.1186, 0.0993, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 18:30:50,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.359e+03 1.871e+03 2.507e+03 3.597e+03 1.071e+04, threshold=5.015e+03, percent-clipped=19.0
+2023-03-12 18:31:08,242 INFO [train.py:968] (0/2) Epoch 24, batch 44600, giga_loss[loss=0.3206, simple_loss=0.3789, pruned_loss=0.1312, over 28899.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3766, pruned_loss=0.1237, over 5672388.49 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3589, pruned_loss=0.1122, over 5703176.82 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3793, pruned_loss=0.1255, over 5666543.21 frames. ], batch size: 199, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:31:32,269 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:31:42,781 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1093536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:31:53,390 INFO [train.py:968] (0/2) Epoch 24, batch 44650, libri_loss[loss=0.3266, simple_loss=0.3878, pruned_loss=0.1327, over 29537.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3762, pruned_loss=0.1237, over 5655768.17 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1125, over 5693884.58 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3784, pruned_loss=0.125, over 5658122.52 frames. ], batch size: 89, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:32:18,719 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:32:21,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:32:22,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.767e+03 2.153e+03 3.013e+03 1.073e+04, threshold=4.305e+03, percent-clipped=2.0
+2023-03-12 18:32:39,413 INFO [train.py:968] (0/2) Epoch 24, batch 44700, giga_loss[loss=0.2742, simple_loss=0.3464, pruned_loss=0.101, over 28634.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3749, pruned_loss=0.121, over 5672772.58 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5696526.85 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3767, pruned_loss=0.1221, over 5671786.59 frames. ], batch size: 92, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:32:49,389 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:32:52,517 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:32:54,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:33:25,298 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:33:26,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9147, 1.9927, 1.7030, 1.8781], device='cuda:0'), covar=tensor([0.2763, 0.2851, 0.3285, 0.2543], device='cuda:0'), in_proj_covar=tensor([0.1554, 0.1121, 0.1372, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 18:33:29,115 INFO [train.py:968] (0/2) Epoch 24, batch 44750, giga_loss[loss=0.349, simple_loss=0.3998, pruned_loss=0.1491, over 28026.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3763, pruned_loss=0.1209, over 5676551.37 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3593, pruned_loss=0.1126, over 5700591.21 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3781, pruned_loss=0.122, over 5671711.46 frames. ], batch size: 412, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:33:30,589 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-12 18:33:57,245 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:33:58,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.632e+02 1.765e+03 2.799e+03 4.658e+03 2.290e+04, threshold=5.599e+03, percent-clipped=26.0
+2023-03-12 18:34:00,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:34:10,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7259, 1.9489, 1.3087, 1.4561], device='cuda:0'), covar=tensor([0.1048, 0.0643, 0.1129, 0.1208], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0451, 0.0521, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 18:34:15,081 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6341, 1.7355, 1.7611, 1.4738], device='cuda:0'), covar=tensor([0.3297, 0.2817, 0.2362, 0.2833], device='cuda:0'), in_proj_covar=tensor([0.2008, 0.1943, 0.1868, 0.2006], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 18:34:16,336 INFO [train.py:968] (0/2) Epoch 24, batch 44800, giga_loss[loss=0.3293, simple_loss=0.3952, pruned_loss=0.1317, over 28963.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3772, pruned_loss=0.122, over 5625049.03 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3599, pruned_loss=0.1132, over 5655800.69 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3785, pruned_loss=0.1226, over 5661095.06 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:34:28,070 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:34:39,847 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093723.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 18:34:42,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:35:05,678 INFO [train.py:968] (0/2) Epoch 24, batch 44850, giga_loss[loss=0.3023, simple_loss=0.372, pruned_loss=0.1163, over 29087.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3784, pruned_loss=0.1238, over 5595117.51 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3602, pruned_loss=0.1134, over 5621558.36 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3794, pruned_loss=0.1242, over 5652529.83 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:35:37,113 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.775e+03 2.433e+03 3.446e+03 1.347e+04, threshold=4.865e+03, percent-clipped=6.0
+2023-03-12 18:35:55,955 INFO [train.py:968] (0/2) Epoch 24, batch 44900, giga_loss[loss=0.2819, simple_loss=0.3512, pruned_loss=0.1063, over 29074.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3753, pruned_loss=0.1219, over 5603605.53 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3605, pruned_loss=0.1137, over 5605133.46 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3759, pruned_loss=0.122, over 5662522.89 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:36:39,307 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-12 18:36:40,990 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-24.pt
+2023-03-12 18:37:51,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.927e+02 1.421e+03 1.894e+03 2.448e+03 5.922e+03, threshold=3.788e+03, percent-clipped=4.0
+2023-03-12 18:38:00,129 INFO [train.py:968] (0/2) Epoch 25, batch 50, giga_loss[loss=0.3365, simple_loss=0.391, pruned_loss=0.1409, over 26795.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3695, pruned_loss=0.1055, over 1263061.34 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3489, pruned_loss=0.09258, over 145251.79 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3716, pruned_loss=0.1069, over 1147001.26 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:38:09,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1093901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:38:49,209 INFO [train.py:968] (0/2) Epoch 25, batch 100, giga_loss[loss=0.2604, simple_loss=0.3462, pruned_loss=0.08726, over 29047.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3581, pruned_loss=0.09929, over 2240421.29 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3378, pruned_loss=0.08837, over 305844.76 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3608, pruned_loss=0.1007, over 2045651.73 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:39:02,174 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-12 18:39:26,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.599e+02 1.076e+03 1.298e+03 1.651e+03 3.388e+03, threshold=2.596e+03, percent-clipped=0.0
+2023-03-12 18:39:34,302 INFO [train.py:968] (0/2) Epoch 25, batch 150, giga_loss[loss=0.2517, simple_loss=0.3253, pruned_loss=0.08907, over 27957.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3439, pruned_loss=0.09322, over 3004429.08 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3357, pruned_loss=0.08634, over 388547.38 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3452, pruned_loss=0.09415, over 2807697.29 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:39:45,129 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1094000.pt
+2023-03-12 18:39:51,961 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:39:51,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8513, 3.6775, 3.4524, 1.6338], device='cuda:0'), covar=tensor([0.0684, 0.0892, 0.0822, 0.2279], device='cuda:0'), in_proj_covar=tensor([0.1276, 0.1184, 0.0992, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 18:40:18,872 INFO [train.py:968] (0/2) Epoch 25, batch 200, giga_loss[loss=0.2425, simple_loss=0.3174, pruned_loss=0.08374, over 28713.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3308, pruned_loss=0.08698, over 3612311.30 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3323, pruned_loss=0.08358, over 550692.19 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3315, pruned_loss=0.08781, over 3386055.42 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:40:22,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:40:24,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:40:36,414 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:40:47,872 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:40:51,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.374e+02 1.042e+03 1.234e+03 1.622e+03 3.464e+03, threshold=2.469e+03, percent-clipped=6.0
+2023-03-12 18:41:01,331 INFO [train.py:968] (0/2) Epoch 25, batch 250, giga_loss[loss=0.2201, simple_loss=0.3006, pruned_loss=0.06979, over 28664.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3211, pruned_loss=0.08247, over 4079278.11 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.334, pruned_loss=0.08416, over 657231.81 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3206, pruned_loss=0.08278, over 3862979.73 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:41:10,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094098.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:41:13,551 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:41:23,867 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-12 18:41:38,104 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:41:45,971 INFO [train.py:968] (0/2) Epoch 25, batch 300, giga_loss[loss=0.1968, simple_loss=0.2731, pruned_loss=0.06018, over 29140.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3135, pruned_loss=0.07938, over 4440913.84 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3362, pruned_loss=0.08559, over 809759.89 frames. ], giga_tot_loss[loss=0.235, simple_loss=0.3119, pruned_loss=0.0791, over 4226170.45 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:42:07,963 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:42:25,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.589e+02 1.136e+03 1.575e+03 2.027e+03 8.561e+03, threshold=3.149e+03, percent-clipped=14.0
+2023-03-12 18:42:32,911 INFO [train.py:968] (0/2) Epoch 25, batch 350, giga_loss[loss=0.2088, simple_loss=0.2784, pruned_loss=0.06964, over 28381.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3077, pruned_loss=0.07704, over 4723388.46 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3342, pruned_loss=0.08422, over 959810.73 frames. ], giga_tot_loss[loss=0.2295, simple_loss=0.3056, pruned_loss=0.07673, over 4516377.53 frames. ], batch size: 78, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:43:13,531 INFO [train.py:968] (0/2) Epoch 25, batch 400, giga_loss[loss=0.2083, simple_loss=0.277, pruned_loss=0.0698, over 28779.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3051, pruned_loss=0.07617, over 4947341.55 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3364, pruned_loss=0.08542, over 1103889.82 frames. ], giga_tot_loss[loss=0.2264, simple_loss=0.3021, pruned_loss=0.07539, over 4754906.11 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 18:43:14,398 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094241.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:43:17,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094244.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 18:43:17,708 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:43:19,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:43:41,695 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094273.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:43:43,587 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:43:46,750 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.534e+02 1.108e+03 1.348e+03 1.864e+03 5.127e+03, threshold=2.695e+03, percent-clipped=8.0
+2023-03-12 18:43:53,693 INFO [train.py:968] (0/2) Epoch 25, batch 450, libri_loss[loss=0.2031, simple_loss=0.2844, pruned_loss=0.0609, over 29466.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3026, pruned_loss=0.07511, over 5122926.54 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3346, pruned_loss=0.08495, over 1246404.50 frames. ], giga_tot_loss[loss=0.2241, simple_loss=0.2996, pruned_loss=0.07427, over 4944871.04 frames. ], batch size: 70, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 18:44:00,903 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-12 18:44:08,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9184, 2.2817, 2.2986, 1.7191], device='cuda:0'), covar=tensor([0.3783, 0.2582, 0.2342, 0.3281], device='cuda:0'), in_proj_covar=tensor([0.2010, 0.1938, 0.1863, 0.2004], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 18:44:11,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5002, 1.9094, 1.4707, 1.6090], device='cuda:0'), covar=tensor([0.2839, 0.2809, 0.3314, 0.2592], device='cuda:0'), in_proj_covar=tensor([0.1560, 0.1123, 0.1377, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 18:44:38,659 INFO [train.py:968] (0/2) Epoch 25, batch 500, giga_loss[loss=0.2162, simple_loss=0.2917, pruned_loss=0.0704, over 28850.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.3001, pruned_loss=0.07365, over 5259420.02 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3336, pruned_loss=0.08469, over 1384431.52 frames. ], giga_tot_loss[loss=0.2211, simple_loss=0.2969, pruned_loss=0.07271, over 5097679.80 frames. ], batch size: 112, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:44:47,070 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-12 18:45:17,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.692e+02 1.033e+03 1.341e+03 1.782e+03 5.993e+03, threshold=2.682e+03, percent-clipped=5.0
+2023-03-12 18:45:17,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:45:25,656 INFO [train.py:968] (0/2) Epoch 25, batch 550, giga_loss[loss=0.2029, simple_loss=0.2815, pruned_loss=0.06213, over 28671.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2976, pruned_loss=0.07242, over 5357022.79 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3335, pruned_loss=0.08445, over 1495365.70 frames. ], giga_tot_loss[loss=0.2186, simple_loss=0.2942, pruned_loss=0.07144, over 5215630.04 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:45:28,108 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:46:09,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:46:10,706 INFO [train.py:968] (0/2) Epoch 25, batch 600, giga_loss[loss=0.1851, simple_loss=0.2661, pruned_loss=0.05208, over 28474.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2957, pruned_loss=0.07165, over 5434886.00 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3337, pruned_loss=0.08418, over 1582739.77 frames. ], giga_tot_loss[loss=0.2168, simple_loss=0.2922, pruned_loss=0.07069, over 5313675.91 frames. ], batch size: 65, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:46:34,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4362, 1.5576, 1.3559, 1.6092], device='cuda:0'), covar=tensor([0.0752, 0.0352, 0.0344, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 18:46:50,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.471e+02 1.015e+03 1.260e+03 1.820e+03 4.942e+03, threshold=2.520e+03, percent-clipped=10.0
+2023-03-12 18:46:59,744 INFO [train.py:968] (0/2) Epoch 25, batch 650, giga_loss[loss=0.2192, simple_loss=0.2948, pruned_loss=0.07185, over 28285.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2944, pruned_loss=0.07139, over 5482203.34 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3333, pruned_loss=0.08371, over 1664923.19 frames. ], giga_tot_loss[loss=0.2161, simple_loss=0.2911, pruned_loss=0.07052, over 5381250.19 frames. ], batch size: 369, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:47:11,164 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:47:15,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9694, 2.2495, 2.3217, 1.8423], device='cuda:0'), covar=tensor([0.3265, 0.2429, 0.2282, 0.2889], device='cuda:0'), in_proj_covar=tensor([0.2010, 0.1936, 0.1859, 0.2002], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 18:47:28,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:47:31,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:47:42,144 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:47:43,154 INFO [train.py:968] (0/2) Epoch 25, batch 700, giga_loss[loss=0.1963, simple_loss=0.2728, pruned_loss=0.05992, over 29008.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2915, pruned_loss=0.07014, over 5531308.89 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3328, pruned_loss=0.08372, over 1788344.83 frames. ], giga_tot_loss[loss=0.213, simple_loss=0.2878, pruned_loss=0.0691, over 5441765.06 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:47:57,894 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:48:19,876 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:48:21,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.941e+02 9.985e+02 1.272e+03 1.643e+03 3.425e+03, threshold=2.543e+03, percent-clipped=3.0
+2023-03-12 18:48:24,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:48:27,550 INFO [train.py:968] (0/2) Epoch 25, batch 750, giga_loss[loss=0.1886, simple_loss=0.2638, pruned_loss=0.05668, over 28767.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2894, pruned_loss=0.06892, over 5580384.17 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3329, pruned_loss=0.08403, over 1889535.16 frames. ], giga_tot_loss[loss=0.2104, simple_loss=0.2855, pruned_loss=0.06766, over 5501907.85 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:48:47,763 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:48:58,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0139, 2.2932, 1.5902, 1.9026], device='cuda:0'), covar=tensor([0.1045, 0.0736, 0.1118, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0449, 0.0521, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 18:49:09,946 INFO [train.py:968] (0/2) Epoch 25, batch 800, giga_loss[loss=0.2159, simple_loss=0.2899, pruned_loss=0.07096, over 28878.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2872, pruned_loss=0.06811, over 5599835.98 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3323, pruned_loss=0.0837, over 1978794.93 frames. ], giga_tot_loss[loss=0.2085, simple_loss=0.2833, pruned_loss=0.06686, over 5540086.08 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 18:49:18,242 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:49:21,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:49:49,875 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:49:50,529 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:49:52,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.924e+02 1.107e+03 1.465e+03 2.113e+03 4.763e+03, threshold=2.930e+03, percent-clipped=12.0
+2023-03-12 18:49:53,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:49:57,314 INFO [train.py:968] (0/2) Epoch 25, batch 850, giga_loss[loss=0.2509, simple_loss=0.323, pruned_loss=0.08934, over 28607.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2949, pruned_loss=0.07259, over 5614375.04 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3323, pruned_loss=0.08355, over 2076386.28 frames. ], giga_tot_loss[loss=0.2168, simple_loss=0.2908, pruned_loss=0.07138, over 5559323.28 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:50:21,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:50:45,651 INFO [train.py:968] (0/2) Epoch 25, batch 900, giga_loss[loss=0.3222, simple_loss=0.3838, pruned_loss=0.1303, over 28766.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3081, pruned_loss=0.07927, over 5631282.81 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3316, pruned_loss=0.08321, over 2114732.57 frames. ], giga_tot_loss[loss=0.2309, simple_loss=0.305, pruned_loss=0.07841, over 5585436.57 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:51:11,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:51:23,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.573e+02 1.354e+03 1.665e+03 2.257e+03 6.386e+03, threshold=3.330e+03, percent-clipped=15.0
+2023-03-12 18:51:30,065 INFO [train.py:968] (0/2) Epoch 25, batch 950, giga_loss[loss=0.356, simple_loss=0.4031, pruned_loss=0.1545, over 26699.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3203, pruned_loss=0.08537, over 5652484.03 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3315, pruned_loss=0.08322, over 2189967.44 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3177, pruned_loss=0.08472, over 5611420.02 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:51:44,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4399, 1.6773, 1.5016, 1.6094], device='cuda:0'), covar=tensor([0.0795, 0.0324, 0.0326, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 18:52:09,405 INFO [train.py:968] (0/2) Epoch 25, batch 1000, giga_loss[loss=0.2832, simple_loss=0.3583, pruned_loss=0.104, over 28829.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3284, pruned_loss=0.08876, over 5667960.78 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3319, pruned_loss=0.08349, over 2281721.16 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.326, pruned_loss=0.08825, over 5629764.07 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:52:41,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.415e+02 1.308e+03 1.655e+03 2.224e+03 6.528e+03, threshold=3.310e+03, percent-clipped=5.0
+2023-03-12 18:52:47,544 INFO [train.py:968] (0/2) Epoch 25, batch 1050, giga_loss[loss=0.2574, simple_loss=0.3379, pruned_loss=0.08848, over 28736.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3326, pruned_loss=0.08961, over 5672799.39 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3331, pruned_loss=0.08447, over 2364359.57 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3302, pruned_loss=0.08899, over 5643478.12 frames. ], batch size: 78, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:53:08,531 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:53:10,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:53:12,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6844, 1.8996, 1.5806, 1.7209], device='cuda:0'), covar=tensor([0.2762, 0.2837, 0.3124, 0.2546], device='cuda:0'), in_proj_covar=tensor([0.1560, 0.1123, 0.1377, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 18:53:16,315 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 18:53:33,468 INFO [train.py:968] (0/2) Epoch 25, batch 1100, giga_loss[loss=0.2752, simple_loss=0.3582, pruned_loss=0.09615, over 28881.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3349, pruned_loss=0.0898, over 5662318.91 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3336, pruned_loss=0.08488, over 2421826.67 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3329, pruned_loss=0.08922, over 5645558.90 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:53:37,506 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:53:51,226 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-12 18:53:55,408 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:54:09,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.516e+02 1.324e+03 1.723e+03 2.363e+03 7.097e+03, threshold=3.446e+03, percent-clipped=7.0
+2023-03-12 18:54:16,051 INFO [train.py:968] (0/2) Epoch 25, batch 1150, giga_loss[loss=0.2565, simple_loss=0.3388, pruned_loss=0.08708, over 28825.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3374, pruned_loss=0.09145, over 5681542.51 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3335, pruned_loss=0.08489, over 2488986.99 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3359, pruned_loss=0.09111, over 5665826.76 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:54:24,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5503, 5.3757, 5.0596, 2.7602], device='cuda:0'), covar=tensor([0.0390, 0.0549, 0.0632, 0.1723], device='cuda:0'), in_proj_covar=tensor([0.1253, 0.1165, 0.0978, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 18:54:58,305 INFO [train.py:968] (0/2) Epoch 25, batch 1200, giga_loss[loss=0.2737, simple_loss=0.3494, pruned_loss=0.09898, over 28625.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3391, pruned_loss=0.09279, over 5676485.01 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3327, pruned_loss=0.08412, over 2639509.27 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3384, pruned_loss=0.09314, over 5654491.56 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 18:55:24,159 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-12 18:55:36,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.617e+02 1.156e+03 1.441e+03 1.892e+03 3.836e+03, threshold=2.881e+03, percent-clipped=2.0
+2023-03-12 18:55:41,781 INFO [train.py:968] (0/2) Epoch 25, batch 1250, giga_loss[loss=0.2699, simple_loss=0.3479, pruned_loss=0.09592, over 28897.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3422, pruned_loss=0.0947, over 5681741.67 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3332, pruned_loss=0.08437, over 2719827.44 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3416, pruned_loss=0.0951, over 5659607.55 frames. ], batch size: 112, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 18:56:23,470 INFO [train.py:968] (0/2) Epoch 25, batch 1300, giga_loss[loss=0.3246, simple_loss=0.3927, pruned_loss=0.1283, over 28761.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3454, pruned_loss=0.0958, over 5688433.23 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3333, pruned_loss=0.08415, over 2829846.35 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3453, pruned_loss=0.09659, over 5663635.00 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:56:44,442 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-12 18:56:56,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.777e+02 1.405e+03 1.834e+03 2.853e+03 1.394e+04, threshold=3.667e+03, percent-clipped=24.0
+2023-03-12 18:57:00,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3759, 1.9953, 1.5288, 0.5741], device='cuda:0'), covar=tensor([0.6556, 0.3592, 0.4991, 0.7260], device='cuda:0'), in_proj_covar=tensor([0.1785, 0.1675, 0.1617, 0.1445], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 18:57:00,913 INFO [train.py:968] (0/2) Epoch 25, batch 1350, giga_loss[loss=0.2649, simple_loss=0.3475, pruned_loss=0.09117, over 28432.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3479, pruned_loss=0.09607, over 5703388.63 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.334, pruned_loss=0.08446, over 2889350.48 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3477, pruned_loss=0.09676, over 5681407.53 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:57:15,211 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-12 18:57:22,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3924, 1.4734, 1.4775, 1.3907], device='cuda:0'), covar=tensor([0.2408, 0.2368, 0.2172, 0.2272], device='cuda:0'), in_proj_covar=tensor([0.2006, 0.1934, 0.1857, 0.2000], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 18:57:36,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3951, 1.6762, 1.6847, 1.4308], device='cuda:0'), covar=tensor([0.2267, 0.2043, 0.2339, 0.2161], device='cuda:0'), in_proj_covar=tensor([0.0491, 0.0751, 0.0720, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 18:57:41,815 INFO [train.py:968] (0/2) Epoch 25, batch 1400, giga_loss[loss=0.2615, simple_loss=0.3524, pruned_loss=0.08531, over 29076.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3493, pruned_loss=0.09648, over 5696087.94 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.334, pruned_loss=0.0844, over 2973605.21 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3495, pruned_loss=0.09739, over 5677695.84 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:58:13,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.781e+02 1.208e+03 1.475e+03 1.927e+03 5.846e+03, threshold=2.950e+03, percent-clipped=4.0
+2023-03-12 18:58:17,470 INFO [train.py:968] (0/2) Epoch 25, batch 1450, giga_loss[loss=0.2663, simple_loss=0.3494, pruned_loss=0.09163, over 28941.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3493, pruned_loss=0.09554, over 5695283.75 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3349, pruned_loss=0.08492, over 3106589.40 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3496, pruned_loss=0.09653, over 5679716.01 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:58:54,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1095339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:58:54,610 INFO [train.py:968] (0/2) Epoch 25, batch 1500, giga_loss[loss=0.2547, simple_loss=0.316, pruned_loss=0.09672, over 23719.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3476, pruned_loss=0.09348, over 5705618.72 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3352, pruned_loss=0.0849, over 3201427.95 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3482, pruned_loss=0.09458, over 5688442.81 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:59:05,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5642, 1.8236, 1.5141, 1.5531], device='cuda:0'), covar=tensor([0.2588, 0.2593, 0.2815, 0.2488], device='cuda:0'), in_proj_covar=tensor([0.1556, 0.1120, 0.1373, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 18:59:28,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.315e+02 1.101e+03 1.501e+03 1.958e+03 4.753e+03, threshold=3.003e+03, percent-clipped=6.0
+2023-03-12 18:59:33,502 INFO [train.py:968] (0/2) Epoch 25, batch 1550, giga_loss[loss=0.2934, simple_loss=0.368, pruned_loss=0.1094, over 29008.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3457, pruned_loss=0.09155, over 5711994.93 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3353, pruned_loss=0.08496, over 3280901.10 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3464, pruned_loss=0.0926, over 5693472.42 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:00:03,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3069, 1.1857, 1.2302, 1.5144], device='cuda:0'), covar=tensor([0.0803, 0.0390, 0.0348, 0.0890], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 19:00:17,327 INFO [train.py:968] (0/2) Epoch 25, batch 1600, giga_loss[loss=0.3009, simple_loss=0.3616, pruned_loss=0.1201, over 28590.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3468, pruned_loss=0.09311, over 5716064.42 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3351, pruned_loss=0.08479, over 3333135.44 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3477, pruned_loss=0.09417, over 5697866.26 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:00:53,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1095482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:00:55,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.747e+02 1.360e+03 1.680e+03 2.153e+03 3.685e+03, threshold=3.360e+03, percent-clipped=4.0
+2023-03-12 19:00:55,532 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1095485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:01:00,908 INFO [train.py:968] (0/2) Epoch 25, batch 1650, giga_loss[loss=0.2839, simple_loss=0.354, pruned_loss=0.1069, over 28967.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3492, pruned_loss=0.09744, over 5721511.36 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3345, pruned_loss=0.08449, over 3384290.89 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3505, pruned_loss=0.09869, over 5703483.41 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:01:21,978 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1095514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:01:41,133 INFO [train.py:968] (0/2) Epoch 25, batch 1700, libri_loss[loss=0.2406, simple_loss=0.3305, pruned_loss=0.07533, over 29518.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3503, pruned_loss=0.1001, over 5715081.34 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3346, pruned_loss=0.08457, over 3471023.38 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3519, pruned_loss=0.1015, over 5694571.60 frames. ], batch size: 82, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:02:00,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3079, 1.7046, 1.6480, 1.4551], device='cuda:0'), covar=tensor([0.1967, 0.1773, 0.2213, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.0491, 0.0750, 0.0719, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 19:02:16,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.307e+02 1.320e+03 1.639e+03 2.214e+03 4.726e+03, threshold=3.277e+03, percent-clipped=7.0
+2023-03-12 19:02:22,471 INFO [train.py:968] (0/2) Epoch 25, batch 1750, libri_loss[loss=0.2319, simple_loss=0.3097, pruned_loss=0.07708, over 29573.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.349, pruned_loss=0.1003, over 5692503.43 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3351, pruned_loss=0.08502, over 3559122.00 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3505, pruned_loss=0.1018, over 5685738.23 frames. ], batch size: 75, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:02:42,006 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1095612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:03:02,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4273, 5.2232, 4.9662, 2.6886], device='cuda:0'), covar=tensor([0.0476, 0.0683, 0.0659, 0.1798], device='cuda:0'), in_proj_covar=tensor([0.1243, 0.1155, 0.0969, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 19:03:04,193 INFO [train.py:968] (0/2) Epoch 25, batch 1800, giga_loss[loss=0.2459, simple_loss=0.3272, pruned_loss=0.0823, over 29017.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3464, pruned_loss=0.09906, over 5705391.10 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3345, pruned_loss=0.08455, over 3604654.20 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3483, pruned_loss=0.1008, over 5697410.00 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:03:06,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2313, 1.3399, 3.3099, 2.9514], device='cuda:0'), covar=tensor([0.1522, 0.2457, 0.0511, 0.1196], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0659, 0.0975, 0.0938], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 19:03:17,946 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1095655.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:03:41,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.771e+02 1.240e+03 1.595e+03 1.949e+03 4.806e+03, threshold=3.189e+03, percent-clipped=5.0
+2023-03-12 19:03:46,403 INFO [train.py:968] (0/2) Epoch 25, batch 1850, giga_loss[loss=0.236, simple_loss=0.3193, pruned_loss=0.0764, over 29081.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3449, pruned_loss=0.09781, over 5711487.65 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3342, pruned_loss=0.08454, over 3660349.65 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3468, pruned_loss=0.09955, over 5701627.37 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:04:22,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6253, 1.5543, 1.8191, 1.4386], device='cuda:0'), covar=tensor([0.1783, 0.2382, 0.1468, 0.1675], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0711, 0.0969, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 19:04:28,191 INFO [train.py:968] (0/2) Epoch 25, batch 1900, libri_loss[loss=0.2603, simple_loss=0.3492, pruned_loss=0.08565, over 29250.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3442, pruned_loss=0.09634, over 5717692.14 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3349, pruned_loss=0.08477, over 3693572.69 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3454, pruned_loss=0.09775, over 5707448.15 frames. ], batch size: 94, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:05:04,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0778, 1.3909, 1.3349, 0.9610], device='cuda:0'), covar=tensor([0.1625, 0.2641, 0.1457, 0.1712], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0712, 0.0970, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 19:05:10,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.293e+02 1.244e+03 1.562e+03 2.026e+03 4.190e+03, threshold=3.124e+03, percent-clipped=5.0
+2023-03-12 19:05:14,746 INFO [train.py:968] (0/2) Epoch 25, batch 1950, libri_loss[loss=0.267, simple_loss=0.3504, pruned_loss=0.09178, over 29534.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3417, pruned_loss=0.09471, over 5697856.57 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3358, pruned_loss=0.08537, over 3755451.93 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3424, pruned_loss=0.09578, over 5687594.81 frames. ], batch size: 84, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:05:34,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4152, 3.5523, 1.5831, 1.5716], device='cuda:0'), covar=tensor([0.1040, 0.0257, 0.0946, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0559, 0.0396, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 19:05:43,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8217, 1.8516, 1.6406, 1.9541], device='cuda:0'), covar=tensor([0.2686, 0.2925, 0.3125, 0.2634], device='cuda:0'), in_proj_covar=tensor([0.1557, 0.1120, 0.1373, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 19:05:56,241 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9055, 1.2176, 2.8871, 2.7630], device='cuda:0'), covar=tensor([0.1629, 0.2597, 0.0539, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0660, 0.0977, 0.0940], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 19:05:58,248 INFO [train.py:968] (0/2) Epoch 25, batch 2000, giga_loss[loss=0.2307, simple_loss=0.3133, pruned_loss=0.07401, over 28854.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3363, pruned_loss=0.09133, over 5701733.52 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3351, pruned_loss=0.08493, over 3840342.31 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3374, pruned_loss=0.09275, over 5685846.68 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:06:26,566 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-12 19:06:38,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.033e+02 9.229e+02 1.210e+03 1.545e+03 5.930e+03, threshold=2.419e+03, percent-clipped=4.0
+2023-03-12 19:06:43,930 INFO [train.py:968] (0/2) Epoch 25, batch 2050, giga_loss[loss=0.2335, simple_loss=0.308, pruned_loss=0.07953, over 28695.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3311, pruned_loss=0.08906, over 5691696.43 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3353, pruned_loss=0.08512, over 3881130.59 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3317, pruned_loss=0.09015, over 5675533.33 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:06:52,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0923, 1.3346, 1.0681, 0.8750], device='cuda:0'), covar=tensor([0.1213, 0.0548, 0.1264, 0.1160], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0450, 0.0524, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 19:07:27,110 INFO [train.py:968] (0/2) Epoch 25, batch 2100, giga_loss[loss=0.2342, simple_loss=0.3151, pruned_loss=0.07662, over 28695.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3273, pruned_loss=0.08706, over 5691694.95 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3356, pruned_loss=0.08516, over 3947298.96 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3274, pruned_loss=0.08798, over 5675091.92 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:08:04,071 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 19:08:04,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.088e+02 1.195e+03 1.585e+03 1.977e+03 5.282e+03, threshold=3.170e+03, percent-clipped=12.0
+2023-03-12 19:08:06,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1095987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:08:07,596 INFO [train.py:968] (0/2) Epoch 25, batch 2150, giga_loss[loss=0.3013, simple_loss=0.3675, pruned_loss=0.1176, over 28606.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3289, pruned_loss=0.08752, over 5701857.89 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3361, pruned_loss=0.08524, over 4004706.52 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3284, pruned_loss=0.08826, over 5683478.74 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:08:16,144 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1096000.pt
+2023-03-12 19:08:39,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1096030.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:08:46,222 INFO [train.py:968] (0/2) Epoch 25, batch 2200, giga_loss[loss=0.2427, simple_loss=0.3239, pruned_loss=0.08072, over 28836.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3302, pruned_loss=0.08798, over 5706752.08 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3368, pruned_loss=0.08544, over 4058222.88 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3291, pruned_loss=0.0885, over 5689149.87 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:09:16,656 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-12 19:09:24,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.087e+02 1.228e+03 1.564e+03 2.287e+03 8.779e+03, threshold=3.129e+03, percent-clipped=11.0
+2023-03-12 19:09:26,816 INFO [train.py:968] (0/2) Epoch 25, batch 2250, libri_loss[loss=0.2798, simple_loss=0.3664, pruned_loss=0.09659, over 29367.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3299, pruned_loss=0.08792, over 5705907.78 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3369, pruned_loss=0.08522, over 4138542.45 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3287, pruned_loss=0.08862, over 5684166.65 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:09:32,683 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-12 19:09:40,454 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-12 19:09:55,995 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1096130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:09:59,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1096133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:10:03,918 INFO [train.py:968] (0/2) Epoch 25, batch 2300, giga_loss[loss=0.3188, simple_loss=0.3775, pruned_loss=0.1301, over 27686.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3281, pruned_loss=0.08704, over 5720447.16 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3378, pruned_loss=0.08544, over 4182028.26 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3263, pruned_loss=0.08748, over 5699006.76 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:10:25,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1096162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:10:33,031 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1096173.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:10:35,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1096176.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:10:43,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.973e+02 1.147e+03 1.403e+03 1.800e+03 7.415e+03, threshold=2.807e+03, percent-clipped=6.0
+2023-03-12 19:10:46,271 INFO [train.py:968] (0/2) Epoch 25, batch 2350, giga_loss[loss=0.2256, simple_loss=0.2955, pruned_loss=0.0778, over 28751.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3263, pruned_loss=0.08648, over 5716118.91 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3382, pruned_loss=0.08566, over 4203672.18 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3245, pruned_loss=0.0867, over 5700386.71 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:10:51,831 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1096196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:10:59,125 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1096205.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:11:09,354 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7384, 1.8067, 1.5286, 1.9916], device='cuda:0'), covar=tensor([0.2656, 0.2884, 0.3157, 0.2548], device='cuda:0'), in_proj_covar=tensor([0.1555, 0.1121, 0.1371, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 19:11:27,479 INFO [train.py:968] (0/2) Epoch 25, batch 2400, giga_loss[loss=0.2091, simple_loss=0.2918, pruned_loss=0.06325, over 28956.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3244, pruned_loss=0.08549, over 5725411.57 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3384, pruned_loss=0.08551, over 4245175.24 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3225, pruned_loss=0.08576, over 5709247.02 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:11:40,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9701, 1.1465, 3.4449, 3.0381], device='cuda:0'), covar=tensor([0.1800, 0.2785, 0.0460, 0.0917], device='cuda:0'), in_proj_covar=tensor([0.0773, 0.0658, 0.0973, 0.0938], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 19:11:53,031 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2787, 0.8708, 1.0238, 1.4597], device='cuda:0'), covar=tensor([0.0785, 0.0406, 0.0351, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0064, 0.0112], device='cuda:0')
+2023-03-12 19:12:06,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.970e+02 1.146e+03 1.434e+03 2.063e+03 4.788e+03, threshold=2.868e+03, percent-clipped=9.0
+2023-03-12 19:12:07,612 INFO [train.py:968] (0/2) Epoch 25, batch 2450, giga_loss[loss=0.2676, simple_loss=0.33, pruned_loss=0.1026, over 28694.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3227, pruned_loss=0.08492, over 5731918.86 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3389, pruned_loss=0.08563, over 4285925.16 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3203, pruned_loss=0.085, over 5715216.65 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:12:45,465 INFO [train.py:968] (0/2) Epoch 25, batch 2500, libri_loss[loss=0.2487, simple_loss=0.3349, pruned_loss=0.0813, over 29565.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3208, pruned_loss=0.08401, over 5740618.09 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.339, pruned_loss=0.08534, over 4341857.39 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3183, pruned_loss=0.08425, over 5721613.00 frames. ], batch size: 76, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:13:21,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.144e+02 1.058e+03 1.359e+03 1.774e+03 4.288e+03, threshold=2.719e+03, percent-clipped=4.0
+2023-03-12 19:13:23,140 INFO [train.py:968] (0/2) Epoch 25, batch 2550, giga_loss[loss=0.2255, simple_loss=0.3032, pruned_loss=0.07392, over 28837.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3187, pruned_loss=0.08312, over 5733570.62 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3392, pruned_loss=0.08534, over 4365265.33 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3163, pruned_loss=0.08329, over 5716413.48 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:14:01,814 INFO [train.py:968] (0/2) Epoch 25, batch 2600, giga_loss[loss=0.2395, simple_loss=0.3181, pruned_loss=0.08045, over 28667.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3177, pruned_loss=0.08232, over 5733004.49 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3398, pruned_loss=0.08537, over 4417145.88 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3147, pruned_loss=0.08235, over 5714800.04 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:14:38,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.271e+02 1.021e+03 1.275e+03 1.592e+03 3.065e+03, threshold=2.549e+03, percent-clipped=3.0
+2023-03-12 19:14:39,494 INFO [train.py:968] (0/2) Epoch 25, batch 2650, giga_loss[loss=0.2375, simple_loss=0.3136, pruned_loss=0.08064, over 28837.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3162, pruned_loss=0.08116, over 5739443.34 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3399, pruned_loss=0.08514, over 4468314.11 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.313, pruned_loss=0.08122, over 5719655.61 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:15:19,904 INFO [train.py:968] (0/2) Epoch 25, batch 2700, giga_loss[loss=0.2856, simple_loss=0.3575, pruned_loss=0.1069, over 28704.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3172, pruned_loss=0.08184, over 5735104.75 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3404, pruned_loss=0.08516, over 4508953.29 frames. ], giga_tot_loss[loss=0.2386, simple_loss=0.3137, pruned_loss=0.08179, over 5716634.81 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:15:51,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1096571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:16:05,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.772e+02 1.206e+03 1.398e+03 1.699e+03 5.157e+03, threshold=2.795e+03, percent-clipped=6.0
+2023-03-12 19:16:07,534 INFO [train.py:968] (0/2) Epoch 25, batch 2750, giga_loss[loss=0.2374, simple_loss=0.313, pruned_loss=0.08095, over 28659.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3206, pruned_loss=0.08449, over 5727060.14 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3402, pruned_loss=0.08505, over 4515966.52 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3179, pruned_loss=0.08452, over 5711919.92 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:16:13,469 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-12 19:16:51,484 INFO [train.py:968] (0/2) Epoch 25, batch 2800, libri_loss[loss=0.269, simple_loss=0.3565, pruned_loss=0.09079, over 29635.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3259, pruned_loss=0.08823, over 5716904.80 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3402, pruned_loss=0.08513, over 4563607.01 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3231, pruned_loss=0.08822, over 5699418.51 frames. ], batch size: 88, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:17:05,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2484, 2.4092, 1.3568, 1.3216], device='cuda:0'), covar=tensor([0.0976, 0.0376, 0.0861, 0.1330], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0557, 0.0396, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 19:17:33,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.689e+02 1.368e+03 1.683e+03 2.055e+03 5.810e+03, threshold=3.365e+03, percent-clipped=6.0
+2023-03-12 19:17:37,452 INFO [train.py:968] (0/2) Epoch 25, batch 2850, giga_loss[loss=0.2873, simple_loss=0.3595, pruned_loss=0.1076, over 28801.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.334, pruned_loss=0.0938, over 5708793.23 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3402, pruned_loss=0.08513, over 4590629.42 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3316, pruned_loss=0.0939, over 5691235.18 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:17:57,542 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1096714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:17:59,675 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1096717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:18:20,080 INFO [train.py:968] (0/2) Epoch 25, batch 2900, giga_loss[loss=0.3006, simple_loss=0.3634, pruned_loss=0.1189, over 28525.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3397, pruned_loss=0.09648, over 5696528.43 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3409, pruned_loss=0.08547, over 4619768.07 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3373, pruned_loss=0.09653, over 5681418.36 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:18:25,937 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1096746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:19:06,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.558e+02 1.252e+03 1.498e+03 2.034e+03 4.292e+03, threshold=2.997e+03, percent-clipped=4.0
+2023-03-12 19:19:07,427 INFO [train.py:968] (0/2) Epoch 25, batch 2950, giga_loss[loss=0.2624, simple_loss=0.339, pruned_loss=0.09291, over 28743.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3456, pruned_loss=0.09941, over 5679101.87 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3413, pruned_loss=0.08565, over 4644923.23 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3433, pruned_loss=0.09957, over 5663296.09 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:19:51,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5376, 1.7572, 1.4283, 1.5086], device='cuda:0'), covar=tensor([0.2625, 0.2613, 0.2991, 0.2289], device='cuda:0'), in_proj_covar=tensor([0.1554, 0.1119, 0.1369, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 19:19:53,272 INFO [train.py:968] (0/2) Epoch 25, batch 3000, giga_loss[loss=0.3258, simple_loss=0.3932, pruned_loss=0.1293, over 29020.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.35, pruned_loss=0.1007, over 5695275.96 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3415, pruned_loss=0.08568, over 4667552.32 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3482, pruned_loss=0.1011, over 5681055.82 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:19:53,275 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 19:20:01,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5619, 1.9103, 1.5143, 1.4427], device='cuda:0'), covar=tensor([0.2970, 0.3085, 0.3516, 0.2742], device='cuda:0'), in_proj_covar=tensor([0.1554, 0.1119, 0.1369, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 19:20:02,036 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2087, simple_loss=0.3166, pruned_loss=0.05044, over 944034.00 frames. 
+2023-03-12 19:20:02,037 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 19:20:27,874 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1096868.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:20:42,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4301, 3.4172, 1.5444, 1.5967], device='cuda:0'), covar=tensor([0.1067, 0.0326, 0.0943, 0.1507], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0558, 0.0396, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 19:20:43,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.268e+02 1.281e+03 1.648e+03 2.196e+03 5.996e+03, threshold=3.295e+03, percent-clipped=6.0
+2023-03-12 19:20:46,508 INFO [train.py:968] (0/2) Epoch 25, batch 3050, libri_loss[loss=0.2315, simple_loss=0.3183, pruned_loss=0.07233, over 29552.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3532, pruned_loss=0.1028, over 5683111.63 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3412, pruned_loss=0.08552, over 4686057.42 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3522, pruned_loss=0.1034, over 5668757.45 frames. ], batch size: 77, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:20:46,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6059, 4.6002, 1.8218, 1.8190], device='cuda:0'), covar=tensor([0.1010, 0.0203, 0.0915, 0.1413], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0558, 0.0396, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 19:21:10,783 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1096920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:21:25,423 INFO [train.py:968] (0/2) Epoch 25, batch 3100, giga_loss[loss=0.238, simple_loss=0.3215, pruned_loss=0.07732, over 28941.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3482, pruned_loss=0.09885, over 5691207.40 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3414, pruned_loss=0.08574, over 4716315.85 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3475, pruned_loss=0.09962, over 5679344.36 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:21:51,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6562, 1.6524, 1.8866, 1.4190], device='cuda:0'), covar=tensor([0.2043, 0.2756, 0.1639, 0.2025], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0711, 0.0968, 0.0864], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 19:22:06,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.634e+02 1.357e+03 1.753e+03 2.375e+03 7.632e+03, threshold=3.505e+03, percent-clipped=11.0
+2023-03-12 19:22:09,039 INFO [train.py:968] (0/2) Epoch 25, batch 3150, giga_loss[loss=0.2611, simple_loss=0.3383, pruned_loss=0.09193, over 28861.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3464, pruned_loss=0.0973, over 5674946.09 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3424, pruned_loss=0.0867, over 4748043.82 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3452, pruned_loss=0.09764, over 5675928.18 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:22:50,778 INFO [train.py:968] (0/2) Epoch 25, batch 3200, giga_loss[loss=0.2756, simple_loss=0.3562, pruned_loss=0.09749, over 28537.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3458, pruned_loss=0.09674, over 5673045.30 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3415, pruned_loss=0.08626, over 4767625.44 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3456, pruned_loss=0.09751, over 5672763.38 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:23:18,929 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-12 19:23:22,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3313, 1.5809, 1.2771, 1.0368], device='cuda:0'), covar=tensor([0.2669, 0.2795, 0.3132, 0.2485], device='cuda:0'), in_proj_covar=tensor([0.1556, 0.1121, 0.1373, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 19:23:27,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.232e+02 1.291e+03 1.731e+03 2.193e+03 6.017e+03, threshold=3.462e+03, percent-clipped=8.0
+2023-03-12 19:23:30,261 INFO [train.py:968] (0/2) Epoch 25, batch 3250, giga_loss[loss=0.2542, simple_loss=0.3216, pruned_loss=0.09339, over 23657.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3466, pruned_loss=0.09691, over 5677665.70 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3413, pruned_loss=0.08623, over 4816618.21 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3468, pruned_loss=0.09799, over 5669432.95 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:23:55,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9695, 1.1998, 3.3719, 3.0097], device='cuda:0'), covar=tensor([0.1816, 0.2857, 0.0508, 0.0961], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0659, 0.0974, 0.0938], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 19:24:09,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5631, 1.8075, 1.4675, 1.5715], device='cuda:0'), covar=tensor([0.2614, 0.2655, 0.2996, 0.2369], device='cuda:0'), in_proj_covar=tensor([0.1557, 0.1121, 0.1372, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 19:24:11,894 INFO [train.py:968] (0/2) Epoch 25, batch 3300, giga_loss[loss=0.2936, simple_loss=0.364, pruned_loss=0.1116, over 29048.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3484, pruned_loss=0.09788, over 5689876.99 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3416, pruned_loss=0.08643, over 4853724.57 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3485, pruned_loss=0.09898, over 5676439.59 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:24:52,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.425e+03 1.709e+03 2.229e+03 7.047e+03, threshold=3.419e+03, percent-clipped=3.0
+2023-03-12 19:24:54,517 INFO [train.py:968] (0/2) Epoch 25, batch 3350, giga_loss[loss=0.247, simple_loss=0.3279, pruned_loss=0.08305, over 28539.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3512, pruned_loss=0.1004, over 5685123.23 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3414, pruned_loss=0.08646, over 4864440.18 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3516, pruned_loss=0.1014, over 5680526.89 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:25:01,005 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1097196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:25:24,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4130, 2.5564, 1.7961, 1.9711], device='cuda:0'), covar=tensor([0.0958, 0.0677, 0.1053, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0450, 0.0525, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 19:25:36,232 INFO [train.py:968] (0/2) Epoch 25, batch 3400, giga_loss[loss=0.2609, simple_loss=0.3399, pruned_loss=0.09094, over 28851.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3519, pruned_loss=0.1014, over 5688060.56 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3414, pruned_loss=0.08647, over 4871093.29 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3523, pruned_loss=0.1024, over 5685801.47 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:25:39,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1097243.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:26:19,911 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.634e+02 1.354e+03 1.614e+03 2.284e+03 5.759e+03, threshold=3.228e+03, percent-clipped=8.0
+2023-03-12 19:26:21,110 INFO [train.py:968] (0/2) Epoch 25, batch 3450, giga_loss[loss=0.3377, simple_loss=0.3918, pruned_loss=0.1418, over 28776.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3538, pruned_loss=0.1037, over 5682894.55 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3415, pruned_loss=0.08644, over 4885681.31 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3542, pruned_loss=0.1047, over 5678938.47 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:26:22,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9605, 1.3108, 1.1125, 0.2270], device='cuda:0'), covar=tensor([0.4285, 0.3081, 0.4477, 0.6578], device='cuda:0'), in_proj_covar=tensor([0.1785, 0.1677, 0.1619, 0.1446], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 19:26:24,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1097295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:26:27,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-12 19:26:28,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9171, 5.7589, 5.4824, 2.6586], device='cuda:0'), covar=tensor([0.0538, 0.0665, 0.0804, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.1255, 0.1162, 0.0977, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 19:26:46,416 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1097322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:27:01,367 INFO [train.py:968] (0/2) Epoch 25, batch 3500, giga_loss[loss=0.2765, simple_loss=0.3569, pruned_loss=0.09805, over 28567.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3539, pruned_loss=0.1034, over 5672823.54 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3417, pruned_loss=0.08648, over 4891126.34 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3542, pruned_loss=0.1043, over 5675915.88 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:27:11,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3756, 1.9825, 1.3913, 0.7521], device='cuda:0'), covar=tensor([0.6368, 0.2938, 0.4370, 0.6848], device='cuda:0'), in_proj_covar=tensor([0.1784, 0.1674, 0.1617, 0.1444], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 19:27:26,617 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0489, 2.0522, 1.7972, 2.3533], device='cuda:0'), covar=tensor([0.2389, 0.2610, 0.2836, 0.2222], device='cuda:0'), in_proj_covar=tensor([0.1553, 0.1118, 0.1369, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 19:27:37,537 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1097386.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:27:38,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.246e+02 1.237e+03 1.564e+03 1.951e+03 4.456e+03, threshold=3.127e+03, percent-clipped=7.0
+2023-03-12 19:27:39,493 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1097389.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:27:40,003 INFO [train.py:968] (0/2) Epoch 25, batch 3550, giga_loss[loss=0.2795, simple_loss=0.3604, pruned_loss=0.09925, over 28344.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.353, pruned_loss=0.1017, over 5674303.49 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3417, pruned_loss=0.08661, over 4910967.25 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3536, pruned_loss=0.1028, over 5684023.07 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:28:03,866 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1097418.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:28:22,112 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1097438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:28:23,115 INFO [train.py:968] (0/2) Epoch 25, batch 3600, giga_loss[loss=0.267, simple_loss=0.3566, pruned_loss=0.08875, over 28580.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3523, pruned_loss=0.09993, over 5671296.50 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3421, pruned_loss=0.08686, over 4919169.36 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3526, pruned_loss=0.1009, over 5685379.64 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:28:24,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1097441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:28:39,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5608, 1.8551, 1.2433, 1.3978], device='cuda:0'), covar=tensor([0.1039, 0.0526, 0.1043, 0.1132], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0449, 0.0523, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 19:28:48,854 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1097470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:28:59,032 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1097482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:29:04,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.581e+02 1.174e+03 1.382e+03 1.825e+03 4.688e+03, threshold=2.764e+03, percent-clipped=5.0
+2023-03-12 19:29:04,891 INFO [train.py:968] (0/2) Epoch 25, batch 3650, giga_loss[loss=0.2516, simple_loss=0.3241, pruned_loss=0.08953, over 28604.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3511, pruned_loss=0.0991, over 5682510.55 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3415, pruned_loss=0.08657, over 4933853.59 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3519, pruned_loss=0.1002, over 5690635.63 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:29:45,594 INFO [train.py:968] (0/2) Epoch 25, batch 3700, giga_loss[loss=0.2417, simple_loss=0.3242, pruned_loss=0.07961, over 28252.00 frames. ], tot_loss[loss=0.273, simple_loss=0.349, pruned_loss=0.09848, over 5671030.25 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.342, pruned_loss=0.08695, over 4943642.34 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3493, pruned_loss=0.09929, over 5682666.10 frames. ], batch size: 65, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:29:51,976 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-12 19:30:10,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1097571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:30:22,905 INFO [train.py:968] (0/2) Epoch 25, batch 3750, giga_loss[loss=0.2695, simple_loss=0.3422, pruned_loss=0.09833, over 28769.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3463, pruned_loss=0.09692, over 5692942.16 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3421, pruned_loss=0.08703, over 4971522.45 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3466, pruned_loss=0.09781, over 5696906.52 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:30:23,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.267e+02 1.142e+03 1.415e+03 1.965e+03 6.439e+03, threshold=2.829e+03, percent-clipped=11.0
+2023-03-12 19:31:00,443 INFO [train.py:968] (0/2) Epoch 25, batch 3800, giga_loss[loss=0.2623, simple_loss=0.3386, pruned_loss=0.09294, over 28950.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3448, pruned_loss=0.09616, over 5699800.41 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3418, pruned_loss=0.08693, over 5005025.73 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3455, pruned_loss=0.0973, over 5698329.87 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:31:19,579 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-12 19:31:42,777 INFO [train.py:968] (0/2) Epoch 25, batch 3850, giga_loss[loss=0.271, simple_loss=0.351, pruned_loss=0.09553, over 28792.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3469, pruned_loss=0.09785, over 5699209.22 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3418, pruned_loss=0.08689, over 5022981.36 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3475, pruned_loss=0.09897, over 5694284.01 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:31:45,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.252e+02 1.220e+03 1.393e+03 1.770e+03 3.863e+03, threshold=2.787e+03, percent-clipped=4.0
+2023-03-12 19:31:49,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1097697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:32:01,241 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1097714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:32:03,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1097717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:32:15,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5337, 1.8338, 1.4156, 1.7182], device='cuda:0'), covar=tensor([0.2733, 0.2670, 0.3062, 0.2302], device='cuda:0'), in_proj_covar=tensor([0.1560, 0.1124, 0.1374, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 19:32:17,913 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4509, 1.9984, 1.5291, 0.7785], device='cuda:0'), covar=tensor([0.5888, 0.2584, 0.3758, 0.6231], device='cuda:0'), in_proj_covar=tensor([0.1777, 0.1664, 0.1607, 0.1437], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 19:32:21,994 INFO [train.py:968] (0/2) Epoch 25, batch 3900, giga_loss[loss=0.2535, simple_loss=0.335, pruned_loss=0.08598, over 28842.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3473, pruned_loss=0.09769, over 5703092.19 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3421, pruned_loss=0.08725, over 5040564.22 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3476, pruned_loss=0.09849, over 5695586.59 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:32:28,460 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1097746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:32:34,893 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 19:33:01,516 INFO [train.py:968] (0/2) Epoch 25, batch 3950, giga_loss[loss=0.249, simple_loss=0.327, pruned_loss=0.08555, over 28659.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3467, pruned_loss=0.09678, over 5712598.96 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3419, pruned_loss=0.08723, over 5068070.81 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3473, pruned_loss=0.09776, over 5702392.19 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:33:02,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.799e+02 1.156e+03 1.449e+03 1.743e+03 6.133e+03, threshold=2.897e+03, percent-clipped=6.0
+2023-03-12 19:33:44,594 INFO [train.py:968] (0/2) Epoch 25, batch 4000, libri_loss[loss=0.223, simple_loss=0.3022, pruned_loss=0.07193, over 29500.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3458, pruned_loss=0.09605, over 5710930.24 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3417, pruned_loss=0.08723, over 5084726.71 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3465, pruned_loss=0.09701, over 5699087.88 frames. ], batch size: 70, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:33:44,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1097840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:33:46,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1097843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:33:55,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1097857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:34:08,575 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1097872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:34:21,607 INFO [train.py:968] (0/2) Epoch 25, batch 4050, giga_loss[loss=0.2472, simple_loss=0.3307, pruned_loss=0.08187, over 28926.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3452, pruned_loss=0.096, over 5715229.67 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3421, pruned_loss=0.08759, over 5118298.97 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3457, pruned_loss=0.09687, over 5699883.80 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:34:22,327 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.076e+02 1.202e+03 1.483e+03 1.876e+03 6.361e+03, threshold=2.967e+03, percent-clipped=10.0
+2023-03-12 19:34:50,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0524, 5.8373, 5.5593, 3.0028], device='cuda:0'), covar=tensor([0.0380, 0.0565, 0.0652, 0.1598], device='cuda:0'), in_proj_covar=tensor([0.1252, 0.1156, 0.0973, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 19:35:01,520 INFO [train.py:968] (0/2) Epoch 25, batch 4100, giga_loss[loss=0.246, simple_loss=0.3237, pruned_loss=0.08411, over 28998.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3426, pruned_loss=0.09472, over 5718150.86 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3419, pruned_loss=0.08744, over 5132001.25 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3431, pruned_loss=0.09568, over 5704252.49 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:35:07,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9068, 1.1177, 1.0855, 0.8799], device='cuda:0'), covar=tensor([0.2487, 0.2899, 0.1611, 0.2364], device='cuda:0'), in_proj_covar=tensor([0.2004, 0.1942, 0.1860, 0.2009], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 19:35:08,432 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1142, 2.2673, 2.2006, 1.7792], device='cuda:0'), covar=tensor([0.3077, 0.2523, 0.2515, 0.3045], device='cuda:0'), in_proj_covar=tensor([0.2004, 0.1942, 0.1860, 0.2009], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 19:35:09,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3262, 1.6687, 1.6102, 1.5391], device='cuda:0'), covar=tensor([0.0816, 0.0332, 0.0309, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 19:35:35,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2878, 4.0913, 3.9151, 1.7845], device='cuda:0'), covar=tensor([0.0630, 0.0789, 0.0764, 0.2112], device='cuda:0'), in_proj_covar=tensor([0.1253, 0.1156, 0.0973, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 19:35:39,700 INFO [train.py:968] (0/2) Epoch 25, batch 4150, giga_loss[loss=0.2095, simple_loss=0.2897, pruned_loss=0.0646, over 28593.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3393, pruned_loss=0.09287, over 5717013.43 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3421, pruned_loss=0.08756, over 5139656.16 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3396, pruned_loss=0.09365, over 5709328.04 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:35:41,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.200e+02 1.240e+03 1.663e+03 2.373e+03 4.711e+03, threshold=3.327e+03, percent-clipped=14.0
+2023-03-12 19:35:47,708 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1098000.pt
+2023-03-12 19:35:48,239 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1098000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:35:48,314 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1098000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:35:50,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1098003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:36:08,185 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9556, 3.7804, 3.6358, 1.6518], device='cuda:0'), covar=tensor([0.0749, 0.0876, 0.0929, 0.2125], device='cuda:0'), in_proj_covar=tensor([0.1255, 0.1158, 0.0975, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 19:36:13,584 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1098032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:36:15,489 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1098035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:36:18,737 INFO [train.py:968] (0/2) Epoch 25, batch 4200, giga_loss[loss=0.2835, simple_loss=0.3557, pruned_loss=0.1057, over 28654.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3382, pruned_loss=0.09254, over 5716911.36 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3421, pruned_loss=0.08761, over 5162628.97 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3383, pruned_loss=0.09327, over 5705581.84 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:37:01,348 INFO [train.py:968] (0/2) Epoch 25, batch 4250, giga_loss[loss=0.3072, simple_loss=0.3742, pruned_loss=0.1201, over 27698.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3372, pruned_loss=0.09224, over 5713079.98 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.342, pruned_loss=0.08757, over 5166291.83 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3374, pruned_loss=0.09287, over 5703428.88 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:37:02,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.804e+02 1.161e+03 1.383e+03 1.859e+03 4.554e+03, threshold=2.766e+03, percent-clipped=4.0
+2023-03-12 19:37:42,106 INFO [train.py:968] (0/2) Epoch 25, batch 4300, libri_loss[loss=0.2928, simple_loss=0.3671, pruned_loss=0.1092, over 19324.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3363, pruned_loss=0.09242, over 5708909.53 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3425, pruned_loss=0.08794, over 5174349.71 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3359, pruned_loss=0.09271, over 5707139.41 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:38:21,262 INFO [train.py:968] (0/2) Epoch 25, batch 4350, libri_loss[loss=0.2704, simple_loss=0.35, pruned_loss=0.09538, over 29512.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.335, pruned_loss=0.09206, over 5713789.83 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3423, pruned_loss=0.08777, over 5192109.92 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3347, pruned_loss=0.09253, over 5708263.31 frames. ], batch size: 80, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:38:22,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.566e+02 1.197e+03 1.412e+03 1.797e+03 4.951e+03, threshold=2.824e+03, percent-clipped=7.0
+2023-03-12 19:39:00,230 INFO [train.py:968] (0/2) Epoch 25, batch 4400, giga_loss[loss=0.317, simple_loss=0.3765, pruned_loss=0.1288, over 27663.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3336, pruned_loss=0.09171, over 5709473.35 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3422, pruned_loss=0.08766, over 5204622.08 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3333, pruned_loss=0.09225, over 5703124.61 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:39:28,188 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4322, 2.0603, 1.4987, 0.6395], device='cuda:0'), covar=tensor([0.6309, 0.3128, 0.4324, 0.7270], device='cuda:0'), in_proj_covar=tensor([0.1788, 0.1678, 0.1619, 0.1449], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 19:39:37,434 INFO [train.py:968] (0/2) Epoch 25, batch 4450, giga_loss[loss=0.289, simple_loss=0.3596, pruned_loss=0.1092, over 28899.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3334, pruned_loss=0.09136, over 5712166.88 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3428, pruned_loss=0.08797, over 5224551.55 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3324, pruned_loss=0.09166, over 5702665.57 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:39:38,989 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.027e+02 1.164e+03 1.373e+03 1.973e+03 4.235e+03, threshold=2.746e+03, percent-clipped=6.0
+2023-03-12 19:40:19,348 INFO [train.py:968] (0/2) Epoch 25, batch 4500, giga_loss[loss=0.2785, simple_loss=0.3388, pruned_loss=0.1091, over 28758.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3351, pruned_loss=0.09204, over 5709675.74 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3429, pruned_loss=0.08806, over 5230943.15 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3342, pruned_loss=0.09223, over 5700886.16 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:40:49,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1098375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:41:01,757 INFO [train.py:968] (0/2) Epoch 25, batch 4550, libri_loss[loss=0.2603, simple_loss=0.3473, pruned_loss=0.08666, over 29529.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3378, pruned_loss=0.09271, over 5713385.74 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.343, pruned_loss=0.08803, over 5244851.15 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3368, pruned_loss=0.09301, over 5709251.19 frames. ], batch size: 81, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:41:02,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.952e+02 1.164e+03 1.428e+03 1.752e+03 7.866e+03, threshold=2.856e+03, percent-clipped=8.0
+2023-03-12 19:41:07,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4404, 2.0712, 1.3703, 0.6102], device='cuda:0'), covar=tensor([0.6573, 0.3002, 0.4412, 0.6960], device='cuda:0'), in_proj_covar=tensor([0.1789, 0.1677, 0.1622, 0.1450], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 19:41:13,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5883, 3.4814, 1.5692, 1.7740], device='cuda:0'), covar=tensor([0.0944, 0.0288, 0.0983, 0.1238], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0553, 0.0392, 0.0431], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 19:41:13,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5511, 1.6553, 1.3741, 1.6123], device='cuda:0'), covar=tensor([0.0728, 0.0319, 0.0351, 0.0903], device='cuda:0'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0225], device='cuda:0'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:0')
+2023-03-12 19:41:18,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1098410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:41:34,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2512, 1.2507, 3.7918, 3.2063], device='cuda:0'), covar=tensor([0.1727, 0.2954, 0.0407, 0.0934], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0657, 0.0969, 0.0936], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 19:41:43,521 INFO [train.py:968] (0/2) Epoch 25, batch 4600, giga_loss[loss=0.2301, simple_loss=0.3133, pruned_loss=0.07342, over 28545.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3393, pruned_loss=0.09286, over 5717810.72 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3431, pruned_loss=0.08806, over 5261141.41 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3383, pruned_loss=0.09319, over 5710299.40 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:41:59,868 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-12 19:42:26,532 INFO [train.py:968] (0/2) Epoch 25, batch 4650, giga_loss[loss=0.2485, simple_loss=0.3337, pruned_loss=0.08168, over 28984.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3403, pruned_loss=0.09314, over 5711461.31 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3426, pruned_loss=0.0879, over 5283307.90 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3399, pruned_loss=0.09377, over 5701208.49 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:42:27,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.569e+02 1.183e+03 1.490e+03 2.027e+03 5.256e+03, threshold=2.980e+03, percent-clipped=9.0
+2023-03-12 19:42:34,700 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-12 19:42:52,345 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1098518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:42:54,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1098521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:08,197 INFO [train.py:968] (0/2) Epoch 25, batch 4700, giga_loss[loss=0.2365, simple_loss=0.3222, pruned_loss=0.07542, over 28829.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3387, pruned_loss=0.09141, over 5699261.16 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3425, pruned_loss=0.08794, over 5291800.03 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3383, pruned_loss=0.09196, over 5695186.88 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:43:17,839 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1098550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:20,370 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1098553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:22,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1098556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:33,251 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1098572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:44,402 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1098585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:48,209 INFO [train.py:968] (0/2) Epoch 25, batch 4750, giga_loss[loss=0.2474, simple_loss=0.3329, pruned_loss=0.08102, over 28689.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3394, pruned_loss=0.09158, over 5699182.35 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.343, pruned_loss=0.08833, over 5300935.69 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3387, pruned_loss=0.09177, over 5694980.26 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:43:49,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.317e+02 1.167e+03 1.488e+03 2.037e+03 5.688e+03, threshold=2.975e+03, percent-clipped=11.0
+2023-03-12 19:44:07,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6546, 1.8455, 1.5520, 1.7782], device='cuda:0'), covar=tensor([0.2590, 0.2748, 0.3076, 0.2521], device='cuda:0'), in_proj_covar=tensor([0.1558, 0.1123, 0.1374, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 19:44:12,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5246, 1.5940, 1.2785, 1.1863], device='cuda:0'), covar=tensor([0.0969, 0.0586, 0.1073, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0448, 0.0522, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 19:44:14,836 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1098625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:44:27,004 INFO [train.py:968] (0/2) Epoch 25, batch 4800, libri_loss[loss=0.2484, simple_loss=0.3407, pruned_loss=0.07803, over 26098.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3396, pruned_loss=0.09167, over 5699818.02 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3428, pruned_loss=0.08829, over 5311184.39 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3389, pruned_loss=0.092, over 5702881.64 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:45:08,281 INFO [train.py:968] (0/2) Epoch 25, batch 4850, giga_loss[loss=0.2896, simple_loss=0.3576, pruned_loss=0.1108, over 23761.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3404, pruned_loss=0.09262, over 5700950.36 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3427, pruned_loss=0.0883, over 5316188.61 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.34, pruned_loss=0.09291, over 5702384.49 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:45:10,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.870e+02 1.187e+03 1.522e+03 1.978e+03 4.449e+03, threshold=3.045e+03, percent-clipped=5.0
+2023-03-12 19:45:51,787 INFO [train.py:968] (0/2) Epoch 25, batch 4900, giga_loss[loss=0.2939, simple_loss=0.3576, pruned_loss=0.1151, over 28750.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3439, pruned_loss=0.09509, over 5708371.85 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3428, pruned_loss=0.08835, over 5322087.07 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3435, pruned_loss=0.09534, over 5707628.17 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:46:17,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.41 vs. limit=5.0
+2023-03-12 19:46:32,079 INFO [train.py:968] (0/2) Epoch 25, batch 4950, giga_loss[loss=0.2521, simple_loss=0.3372, pruned_loss=0.08348, over 28956.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.347, pruned_loss=0.09683, over 5703105.83 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3435, pruned_loss=0.08866, over 5328814.73 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3462, pruned_loss=0.09697, over 5705841.32 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:46:33,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.343e+02 1.327e+03 1.648e+03 1.972e+03 3.607e+03, threshold=3.297e+03, percent-clipped=5.0
+2023-03-12 19:47:10,403 INFO [train.py:968] (0/2) Epoch 25, batch 5000, giga_loss[loss=0.2626, simple_loss=0.3368, pruned_loss=0.09421, over 28831.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3477, pruned_loss=0.09698, over 5700166.39 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.0888, over 5339342.37 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3471, pruned_loss=0.09723, over 5704525.67 frames. ], batch size: 112, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:47:51,256 INFO [train.py:968] (0/2) Epoch 25, batch 5050, giga_loss[loss=0.324, simple_loss=0.39, pruned_loss=0.129, over 28968.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3486, pruned_loss=0.0975, over 5697830.69 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3438, pruned_loss=0.08891, over 5338697.66 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3479, pruned_loss=0.09764, over 5703289.65 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:47:52,757 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.542e+02 1.281e+03 1.572e+03 2.114e+03 4.870e+03, threshold=3.144e+03, percent-clipped=7.0
+2023-03-12 19:48:29,329 INFO [train.py:968] (0/2) Epoch 25, batch 5100, libri_loss[loss=0.2466, simple_loss=0.3289, pruned_loss=0.08209, over 29602.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3488, pruned_loss=0.09756, over 5697402.75 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3442, pruned_loss=0.08921, over 5346356.02 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.348, pruned_loss=0.09765, over 5704826.12 frames. ], batch size: 75, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:48:34,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1098947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:49:09,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1098989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:49:09,745 INFO [train.py:968] (0/2) Epoch 25, batch 5150, libri_loss[loss=0.2167, simple_loss=0.3025, pruned_loss=0.06547, over 29556.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3484, pruned_loss=0.09748, over 5704839.01 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3444, pruned_loss=0.08931, over 5361467.43 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3478, pruned_loss=0.09771, over 5706511.40 frames. ], batch size: 76, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:49:11,701 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.892e+02 1.301e+03 1.546e+03 2.051e+03 4.951e+03, threshold=3.093e+03, percent-clipped=6.0
+2023-03-12 19:49:16,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:49:49,708 INFO [train.py:968] (0/2) Epoch 25, batch 5200, giga_loss[loss=0.2477, simple_loss=0.3256, pruned_loss=0.0849, over 28840.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.345, pruned_loss=0.09601, over 5700572.30 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3445, pruned_loss=0.0894, over 5369600.51 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3444, pruned_loss=0.09622, over 5699108.63 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:49:50,117 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-12 19:50:28,584 INFO [train.py:968] (0/2) Epoch 25, batch 5250, giga_loss[loss=0.2792, simple_loss=0.337, pruned_loss=0.1107, over 24139.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.342, pruned_loss=0.09447, over 5707882.64 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3439, pruned_loss=0.08891, over 5387960.24 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3421, pruned_loss=0.09526, over 5702085.67 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:50:28,891 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:50:30,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.740e+02 1.195e+03 1.451e+03 1.802e+03 3.159e+03, threshold=2.901e+03, percent-clipped=2.0
+2023-03-12 19:50:30,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:50:32,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1268, 1.3560, 1.2992, 1.0842], device='cuda:0'), covar=tensor([0.3079, 0.2789, 0.1894, 0.2573], device='cuda:0'), in_proj_covar=tensor([0.2014, 0.1954, 0.1870, 0.2012], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 19:50:54,473 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:51:00,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3517, 1.8142, 1.4839, 1.5744], device='cuda:0'), covar=tensor([0.2339, 0.1983, 0.2488, 0.2178], device='cuda:0'), in_proj_covar=tensor([0.0495, 0.0752, 0.0725, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 19:51:08,420 INFO [train.py:968] (0/2) Epoch 25, batch 5300, giga_loss[loss=0.2574, simple_loss=0.3481, pruned_loss=0.08338, over 28674.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3405, pruned_loss=0.09286, over 5713715.47 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.08878, over 5400515.68 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3408, pruned_loss=0.09371, over 5704963.14 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:51:11,834 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:51:14,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:51:15,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6681, 1.9525, 1.4239, 1.6846], device='cuda:0'), covar=tensor([0.1064, 0.0753, 0.1048, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0450, 0.0523, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 19:51:40,112 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:51:52,482 INFO [train.py:968] (0/2) Epoch 25, batch 5350, giga_loss[loss=0.278, simple_loss=0.3571, pruned_loss=0.09944, over 28817.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3421, pruned_loss=0.09233, over 5713610.87 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3438, pruned_loss=0.08896, over 5405410.72 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3421, pruned_loss=0.09289, over 5705292.61 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:51:54,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.031e+02 1.216e+03 1.492e+03 1.952e+03 6.629e+03, threshold=2.983e+03, percent-clipped=9.0
+2023-03-12 19:52:35,638 INFO [train.py:968] (0/2) Epoch 25, batch 5400, giga_loss[loss=0.2625, simple_loss=0.3315, pruned_loss=0.09672, over 28742.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3433, pruned_loss=0.09328, over 5720744.34 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3439, pruned_loss=0.08895, over 5409633.49 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3432, pruned_loss=0.09375, over 5713260.63 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:52:47,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6066, 1.7745, 1.7729, 1.5640], device='cuda:0'), covar=tensor([0.2027, 0.2221, 0.2448, 0.2453], device='cuda:0'), in_proj_covar=tensor([0.0493, 0.0748, 0.0722, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 19:52:52,415 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:52:59,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:53:06,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099280.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:53:12,717 INFO [train.py:968] (0/2) Epoch 25, batch 5450, giga_loss[loss=0.2447, simple_loss=0.3234, pruned_loss=0.08303, over 29020.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3409, pruned_loss=0.09292, over 5723725.81 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3438, pruned_loss=0.08888, over 5425958.35 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3409, pruned_loss=0.09351, over 5713774.86 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:53:16,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.512e+02 1.289e+03 1.552e+03 2.134e+03 5.138e+03, threshold=3.104e+03, percent-clipped=5.0
+2023-03-12 19:53:36,566 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:53:56,001 INFO [train.py:968] (0/2) Epoch 25, batch 5500, giga_loss[loss=0.3109, simple_loss=0.3683, pruned_loss=0.1267, over 28842.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3394, pruned_loss=0.09329, over 5726872.57 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3434, pruned_loss=0.0886, over 5434038.63 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3396, pruned_loss=0.09412, over 5718414.67 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:54:13,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:54:34,498 INFO [train.py:968] (0/2) Epoch 25, batch 5550, giga_loss[loss=0.2407, simple_loss=0.3228, pruned_loss=0.07936, over 28881.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3371, pruned_loss=0.09325, over 5731242.45 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3429, pruned_loss=0.0883, over 5443243.22 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3376, pruned_loss=0.09426, over 5722943.10 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:54:37,008 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-12 19:54:37,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.674e+02 1.307e+03 1.532e+03 2.039e+03 5.077e+03, threshold=3.064e+03, percent-clipped=6.0
+2023-03-12 19:55:00,451 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:55:14,524 INFO [train.py:968] (0/2) Epoch 25, batch 5600, giga_loss[loss=0.3178, simple_loss=0.3876, pruned_loss=0.124, over 27690.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3364, pruned_loss=0.09337, over 5732231.92 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3427, pruned_loss=0.08824, over 5454333.98 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3368, pruned_loss=0.09435, over 5722096.85 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:55:26,278 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:55:55,492 INFO [train.py:968] (0/2) Epoch 25, batch 5650, giga_loss[loss=0.2868, simple_loss=0.3533, pruned_loss=0.1102, over 27685.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3359, pruned_loss=0.09344, over 5717406.39 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.343, pruned_loss=0.08852, over 5461652.89 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3358, pruned_loss=0.0942, over 5712356.45 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:55:58,878 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.317e+02 1.259e+03 1.650e+03 2.241e+03 5.052e+03, threshold=3.299e+03, percent-clipped=4.0
+2023-03-12 19:56:08,184 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:56:10,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:56:24,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5191, 4.3367, 4.1815, 1.9340], device='cuda:0'), covar=tensor([0.0705, 0.0887, 0.0999, 0.2072], device='cuda:0'), in_proj_covar=tensor([0.1260, 0.1165, 0.0981, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 19:56:31,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6495, 1.6954, 1.3395, 1.2678], device='cuda:0'), covar=tensor([0.0949, 0.0611, 0.1069, 0.1234], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0449, 0.0522, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 19:56:34,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:56:34,768 INFO [train.py:968] (0/2) Epoch 25, batch 5700, giga_loss[loss=0.2393, simple_loss=0.3224, pruned_loss=0.07809, over 28563.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3331, pruned_loss=0.09232, over 5715490.35 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3433, pruned_loss=0.08876, over 5471225.49 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3326, pruned_loss=0.09283, over 5708667.62 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:56:36,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-12 19:57:08,361 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:57:14,119 INFO [train.py:968] (0/2) Epoch 25, batch 5750, libri_loss[loss=0.2479, simple_loss=0.3376, pruned_loss=0.07906, over 29515.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3286, pruned_loss=0.08972, over 5718286.22 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3432, pruned_loss=0.08866, over 5477738.06 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3281, pruned_loss=0.09024, over 5710115.69 frames. ], batch size: 89, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:57:17,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.091e+02 1.276e+03 1.681e+03 2.320e+03 4.382e+03, threshold=3.362e+03, percent-clipped=6.0
+2023-03-12 19:57:35,563 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099618.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:57:51,303 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:57:53,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:57:55,282 INFO [train.py:968] (0/2) Epoch 25, batch 5800, giga_loss[loss=0.2442, simple_loss=0.3277, pruned_loss=0.08032, over 28955.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3284, pruned_loss=0.09013, over 5708331.24 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3431, pruned_loss=0.08865, over 5479329.20 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3277, pruned_loss=0.09057, over 5705279.69 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:58:00,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:58:06,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099655.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:58:28,930 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 19:58:31,190 INFO [train.py:968] (0/2) Epoch 25, batch 5850, giga_loss[loss=0.2518, simple_loss=0.3368, pruned_loss=0.08339, over 28991.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3297, pruned_loss=0.09019, over 5716788.63 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3426, pruned_loss=0.08845, over 5496151.78 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.329, pruned_loss=0.09079, over 5708274.40 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:58:34,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:58:36,177 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.577e+02 1.282e+03 1.806e+03 2.643e+03 5.541e+03, threshold=3.612e+03, percent-clipped=13.0
+2023-03-12 19:59:12,735 INFO [train.py:968] (0/2) Epoch 25, batch 5900, giga_loss[loss=0.3058, simple_loss=0.3754, pruned_loss=0.1181, over 27606.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3333, pruned_loss=0.09144, over 5702557.08 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3425, pruned_loss=0.08834, over 5493510.59 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3326, pruned_loss=0.09206, over 5702933.09 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:59:42,588 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:45,583 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:48,273 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:53,037 INFO [train.py:968] (0/2) Epoch 25, batch 5950, giga_loss[loss=0.2425, simple_loss=0.3296, pruned_loss=0.07766, over 28997.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3373, pruned_loss=0.09282, over 5713737.08 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3427, pruned_loss=0.0884, over 5505933.08 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3363, pruned_loss=0.09343, over 5710447.42 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:59:53,338 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:55,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:58,326 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.487e+02 1.207e+03 1.397e+03 1.813e+03 3.292e+03, threshold=2.794e+03, percent-clipped=0.0
+2023-03-12 19:59:59,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:59,839 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099798.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:00:01,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099801.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:00:11,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:00:19,586 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:00:23,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:00:26,363 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099830.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:00:33,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:00:35,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:00:35,589 INFO [train.py:968] (0/2) Epoch 25, batch 6000, giga_loss[loss=0.2552, simple_loss=0.3438, pruned_loss=0.08324, over 28886.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3405, pruned_loss=0.09432, over 5710101.55 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3426, pruned_loss=0.08837, over 5513725.12 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3397, pruned_loss=0.09492, over 5704369.17 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:00:35,594 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 20:00:44,414 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2087, simple_loss=0.3159, pruned_loss=0.05079, over 944034.00 frames. 
+2023-03-12 20:00:44,415 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 20:01:07,262 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:01:24,198 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:01:24,590 INFO [train.py:968] (0/2) Epoch 25, batch 6050, giga_loss[loss=0.2838, simple_loss=0.3613, pruned_loss=0.1031, over 28899.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3426, pruned_loss=0.09533, over 5713917.57 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3425, pruned_loss=0.08836, over 5524900.84 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3421, pruned_loss=0.096, over 5704331.64 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:01:28,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.474e+02 1.237e+03 1.662e+03 2.614e+03 6.423e+03, threshold=3.324e+03, percent-clipped=19.0
+2023-03-12 20:01:54,107 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-12 20:02:08,597 INFO [train.py:968] (0/2) Epoch 25, batch 6100, giga_loss[loss=0.3417, simple_loss=0.3981, pruned_loss=0.1427, over 28747.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3468, pruned_loss=0.09904, over 5708711.21 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.342, pruned_loss=0.08808, over 5534661.52 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3468, pruned_loss=0.09998, over 5696640.18 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:02:09,546 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:12,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:19,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7107, 1.8664, 1.5647, 1.9468], device='cuda:0'), covar=tensor([0.2568, 0.2777, 0.3034, 0.2413], device='cuda:0'), in_proj_covar=tensor([0.1556, 0.1123, 0.1371, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 20:02:21,065 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:34,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:36,508 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:36,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:43,170 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8353, 3.6474, 3.4680, 1.6503], device='cuda:0'), covar=tensor([0.0776, 0.0922, 0.0856, 0.2175], device='cuda:0'), in_proj_covar=tensor([0.1263, 0.1167, 0.0985, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 20:02:52,861 INFO [train.py:968] (0/2) Epoch 25, batch 6150, giga_loss[loss=0.2852, simple_loss=0.3569, pruned_loss=0.1067, over 28849.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3526, pruned_loss=0.104, over 5687935.54 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.342, pruned_loss=0.08812, over 5527484.15 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3528, pruned_loss=0.1052, over 5689634.62 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:02:56,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099993.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:02:58,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.991e+02 1.541e+03 2.075e+03 2.901e+03 7.879e+03, threshold=4.149e+03, percent-clipped=16.0
+2023-03-12 20:03:00,387 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:03:00,543 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-12 20:03:00,843 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1100000.pt
+2023-03-12 20:03:02,750 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:03:05,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7574, 5.2348, 1.9217, 2.1313], device='cuda:0'), covar=tensor([0.0964, 0.0209, 0.0938, 0.1247], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0559, 0.0395, 0.0433], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 20:03:07,078 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 20:03:10,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:03:13,412 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-12 20:03:39,561 INFO [train.py:968] (0/2) Epoch 25, batch 6200, giga_loss[loss=0.2985, simple_loss=0.3689, pruned_loss=0.1141, over 28817.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3602, pruned_loss=0.1091, over 5696077.49 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3421, pruned_loss=0.0882, over 5537495.98 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3606, pruned_loss=0.1103, over 5692097.57 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:04:29,205 INFO [train.py:968] (0/2) Epoch 25, batch 6250, giga_loss[loss=0.3517, simple_loss=0.4048, pruned_loss=0.1493, over 27977.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3649, pruned_loss=0.1132, over 5690069.59 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3419, pruned_loss=0.08817, over 5536518.64 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3659, pruned_loss=0.1147, over 5691528.95 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:04:33,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.157e+03 1.835e+03 2.333e+03 3.238e+03 7.681e+03, threshold=4.665e+03, percent-clipped=8.0
+2023-03-12 20:04:36,128 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:04:39,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:04:47,035 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-12 20:05:00,756 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:05:05,972 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100136.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:05:08,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100139.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:05:09,178 INFO [train.py:968] (0/2) Epoch 25, batch 6300, giga_loss[loss=0.4166, simple_loss=0.4322, pruned_loss=0.2005, over 23580.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3704, pruned_loss=0.118, over 5691625.77 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3418, pruned_loss=0.08817, over 5549752.88 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3723, pruned_loss=0.1203, over 5687152.69 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:05:19,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:05:21,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:05:24,087 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:05:33,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3095, 2.6059, 1.3006, 1.5099], device='cuda:0'), covar=tensor([0.1008, 0.0486, 0.0926, 0.1326], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0560, 0.0396, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 20:05:34,793 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100168.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:05:52,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:05:56,785 INFO [train.py:968] (0/2) Epoch 25, batch 6350, giga_loss[loss=0.2878, simple_loss=0.3604, pruned_loss=0.1076, over 29102.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3769, pruned_loss=0.1232, over 5686910.54 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3421, pruned_loss=0.08845, over 5555899.42 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3788, pruned_loss=0.1255, over 5680285.29 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:06:01,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.192e+03 1.745e+03 2.450e+03 3.353e+03 7.319e+03, threshold=4.899e+03, percent-clipped=6.0
+2023-03-12 20:06:47,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8901, 2.0428, 1.8709, 1.8477], device='cuda:0'), covar=tensor([0.2074, 0.2473, 0.2295, 0.2346], device='cuda:0'), in_proj_covar=tensor([0.0494, 0.0751, 0.0723, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 20:06:48,374 INFO [train.py:968] (0/2) Epoch 25, batch 6400, giga_loss[loss=0.3124, simple_loss=0.3781, pruned_loss=0.1233, over 28957.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3772, pruned_loss=0.1245, over 5670539.16 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.342, pruned_loss=0.08831, over 5558844.07 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3794, pruned_loss=0.127, over 5664624.20 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:07:13,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:07:24,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:07:39,424 INFO [train.py:968] (0/2) Epoch 25, batch 6450, giga_loss[loss=0.3194, simple_loss=0.3895, pruned_loss=0.1246, over 28873.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3806, pruned_loss=0.1287, over 5674038.19 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3415, pruned_loss=0.08803, over 5563697.85 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3834, pruned_loss=0.1316, over 5666252.68 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:07:46,819 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:07:47,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.303e+03 1.924e+03 2.352e+03 3.114e+03 6.658e+03, threshold=4.703e+03, percent-clipped=3.0
+2023-03-12 20:07:48,162 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:07:49,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:08:05,654 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:08:21,364 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:08:35,331 INFO [train.py:968] (0/2) Epoch 25, batch 6500, giga_loss[loss=0.352, simple_loss=0.3963, pruned_loss=0.1539, over 28623.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.384, pruned_loss=0.1327, over 5661145.83 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3415, pruned_loss=0.08799, over 5567260.69 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3866, pruned_loss=0.1354, over 5652755.50 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:09:10,119 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:09:11,864 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-12 20:09:25,471 INFO [train.py:968] (0/2) Epoch 25, batch 6550, giga_loss[loss=0.3298, simple_loss=0.3878, pruned_loss=0.1359, over 28963.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3875, pruned_loss=0.1357, over 5650766.16 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3421, pruned_loss=0.08829, over 5573890.28 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3901, pruned_loss=0.1386, over 5640182.09 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 20:09:33,469 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.654e+02 1.834e+03 2.879e+03 5.461e+03 3.324e+04, threshold=5.758e+03, percent-clipped=29.0
+2023-03-12 20:09:40,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-12 20:09:42,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:09:45,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:10:02,017 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-12 20:10:05,254 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 20:10:15,178 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:10:15,534 INFO [train.py:968] (0/2) Epoch 25, batch 6600, giga_loss[loss=0.4095, simple_loss=0.4373, pruned_loss=0.1908, over 26515.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3864, pruned_loss=0.1355, over 5641441.19 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3422, pruned_loss=0.08841, over 5569167.32 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3889, pruned_loss=0.1383, over 5637397.84 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 20:10:25,187 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.39 vs. limit=5.0
+2023-03-12 20:11:04,778 INFO [train.py:968] (0/2) Epoch 25, batch 6650, giga_loss[loss=0.2908, simple_loss=0.3588, pruned_loss=0.1114, over 29113.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3842, pruned_loss=0.1345, over 5647529.06 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3419, pruned_loss=0.08825, over 5574138.45 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.387, pruned_loss=0.1374, over 5641185.59 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 20:11:16,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.582e+02 1.894e+03 2.635e+03 3.448e+03 6.992e+03, threshold=5.270e+03, percent-clipped=2.0
+2023-03-12 20:11:17,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 20:11:31,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:11:32,543 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:11:34,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:11:53,554 INFO [train.py:968] (0/2) Epoch 25, batch 6700, giga_loss[loss=0.2912, simple_loss=0.3684, pruned_loss=0.107, over 28914.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3834, pruned_loss=0.1334, over 5637565.81 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3419, pruned_loss=0.08847, over 5580998.57 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3871, pruned_loss=0.1371, over 5627964.08 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 20:12:02,595 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:12:04,616 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6287, 1.6516, 1.8226, 1.3770], device='cuda:0'), covar=tensor([0.1831, 0.2693, 0.1490, 0.1794], device='cuda:0'), in_proj_covar=tensor([0.0914, 0.0707, 0.0961, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 20:12:43,297 INFO [train.py:968] (0/2) Epoch 25, batch 6750, giga_loss[loss=0.3393, simple_loss=0.3983, pruned_loss=0.1402, over 28285.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3839, pruned_loss=0.1325, over 5648584.09 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3421, pruned_loss=0.08856, over 5587363.77 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3876, pruned_loss=0.1364, over 5636484.36 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 20:12:50,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.622e+03 1.993e+03 2.567e+03 7.560e+03, threshold=3.987e+03, percent-clipped=4.0
+2023-03-12 20:12:59,333 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6705, 2.3690, 1.4790, 0.8824], device='cuda:0'), covar=tensor([0.7347, 0.3418, 0.3723, 0.6602], device='cuda:0'), in_proj_covar=tensor([0.1800, 0.1693, 0.1631, 0.1456], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 20:13:34,297 INFO [train.py:968] (0/2) Epoch 25, batch 6800, giga_loss[loss=0.3384, simple_loss=0.4014, pruned_loss=0.1376, over 28920.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3865, pruned_loss=0.1343, over 5633589.50 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3421, pruned_loss=0.08851, over 5588959.95 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3896, pruned_loss=0.1376, over 5623084.82 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:13:42,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:14:07,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:14:09,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:14:11,182 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100676.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:14:24,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:14:24,431 INFO [train.py:968] (0/2) Epoch 25, batch 6850, giga_loss[loss=0.3094, simple_loss=0.3749, pruned_loss=0.122, over 28740.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3836, pruned_loss=0.1316, over 5634044.22 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08849, over 5592289.33 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3865, pruned_loss=0.1347, over 5623217.73 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:14:36,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.181e+03 1.799e+03 2.277e+03 2.910e+03 7.345e+03, threshold=4.554e+03, percent-clipped=8.0
+2023-03-12 20:15:03,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100729.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:15:13,908 INFO [train.py:968] (0/2) Epoch 25, batch 6900, libri_loss[loss=0.2474, simple_loss=0.3344, pruned_loss=0.08021, over 29516.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3814, pruned_loss=0.1285, over 5650271.70 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08857, over 5603512.31 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3855, pruned_loss=0.1325, over 5632739.94 frames. ], batch size: 81, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:15:47,775 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:15:50,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4451, 1.5445, 1.6530, 1.2781], device='cuda:0'), covar=tensor([0.1635, 0.2496, 0.1414, 0.1726], device='cuda:0'), in_proj_covar=tensor([0.0916, 0.0706, 0.0960, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 20:16:00,685 INFO [train.py:968] (0/2) Epoch 25, batch 6950, giga_loss[loss=0.3189, simple_loss=0.3609, pruned_loss=0.1384, over 24005.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3781, pruned_loss=0.1252, over 5650820.96 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.08867, over 5606509.25 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3817, pruned_loss=0.1288, over 5634953.09 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:16:01,604 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:04,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:09,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.717e+03 2.279e+03 3.423e+03 9.182e+03, threshold=4.557e+03, percent-clipped=12.0
+2023-03-12 20:16:25,313 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:27,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:32,238 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:41,385 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:44,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:48,847 INFO [train.py:968] (0/2) Epoch 25, batch 7000, giga_loss[loss=0.2698, simple_loss=0.3471, pruned_loss=0.0963, over 28709.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3744, pruned_loss=0.1221, over 5650857.77 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08911, over 5602151.37 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3775, pruned_loss=0.1253, over 5642310.72 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:16:56,986 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:17:12,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:17:13,341 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:17:38,204 INFO [train.py:968] (0/2) Epoch 25, batch 7050, giga_loss[loss=0.3623, simple_loss=0.3943, pruned_loss=0.1652, over 23760.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3722, pruned_loss=0.1208, over 5647801.22 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08908, over 5605356.11 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3749, pruned_loss=0.1237, over 5638653.61 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:17:40,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:17:45,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.610e+03 2.080e+03 2.853e+03 6.643e+03, threshold=4.161e+03, percent-clipped=1.0
+2023-03-12 20:17:47,707 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:18:25,398 INFO [train.py:968] (0/2) Epoch 25, batch 7100, giga_loss[loss=0.3106, simple_loss=0.3753, pruned_loss=0.1229, over 28284.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3712, pruned_loss=0.1202, over 5648937.34 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3426, pruned_loss=0.08901, over 5611561.11 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3739, pruned_loss=0.1231, over 5636935.93 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:19:22,171 INFO [train.py:968] (0/2) Epoch 25, batch 7150, giga_loss[loss=0.3081, simple_loss=0.3726, pruned_loss=0.1218, over 28555.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3696, pruned_loss=0.1188, over 5652007.40 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3423, pruned_loss=0.0889, over 5614160.21 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3724, pruned_loss=0.1216, over 5640676.12 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:19:31,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.099e+03 1.664e+03 2.087e+03 2.926e+03 1.117e+04, threshold=4.173e+03, percent-clipped=9.0
+2023-03-12 20:20:10,389 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:20:13,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:20:14,344 INFO [train.py:968] (0/2) Epoch 25, batch 7200, libri_loss[loss=0.2604, simple_loss=0.3452, pruned_loss=0.08787, over 29054.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3687, pruned_loss=0.1166, over 5653658.43 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3422, pruned_loss=0.0889, over 5616436.12 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3712, pruned_loss=0.1191, over 5642927.30 frames. ], batch size: 101, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:20:22,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:20:26,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101051.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:20:43,708 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:21:05,783 INFO [train.py:968] (0/2) Epoch 25, batch 7250, giga_loss[loss=0.3388, simple_loss=0.3982, pruned_loss=0.1398, over 27576.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3691, pruned_loss=0.1145, over 5665661.24 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3417, pruned_loss=0.08871, over 5625399.57 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3725, pruned_loss=0.1176, over 5650541.81 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:21:10,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1101094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:21:13,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.595e+03 1.987e+03 2.460e+03 6.638e+03, threshold=3.974e+03, percent-clipped=4.0
+2023-03-12 20:21:17,938 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101104.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:21:52,031 INFO [train.py:968] (0/2) Epoch 25, batch 7300, giga_loss[loss=0.3021, simple_loss=0.3784, pruned_loss=0.1129, over 28483.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3706, pruned_loss=0.1149, over 5653543.30 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3418, pruned_loss=0.08893, over 5604620.93 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3741, pruned_loss=0.1179, over 5661044.43 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:22:05,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:22:40,726 INFO [train.py:968] (0/2) Epoch 25, batch 7350, giga_loss[loss=0.2712, simple_loss=0.3485, pruned_loss=0.09692, over 28818.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3716, pruned_loss=0.1166, over 5651087.09 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3418, pruned_loss=0.08894, over 5609358.51 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3747, pruned_loss=0.1194, over 5653499.16 frames. ], batch size: 112, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:22:43,672 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:22:46,445 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101194.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:22:48,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:22:50,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101197.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:22:51,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.214e+02 1.797e+03 2.232e+03 2.848e+03 1.022e+04, threshold=4.465e+03, percent-clipped=9.0
+2023-03-12 20:23:14,586 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:23:16,119 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101226.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:23:29,381 INFO [train.py:968] (0/2) Epoch 25, batch 7400, libri_loss[loss=0.2973, simple_loss=0.3742, pruned_loss=0.1102, over 29375.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3715, pruned_loss=0.1175, over 5662632.57 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08914, over 5612177.43 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.374, pruned_loss=0.1198, over 5662379.86 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:23:29,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:23:36,697 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101247.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:23:41,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101250.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:23:47,140 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 20:24:01,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:24:07,028 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101279.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:24:15,001 INFO [train.py:968] (0/2) Epoch 25, batch 7450, libri_loss[loss=0.243, simple_loss=0.3388, pruned_loss=0.07362, over 29442.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3686, pruned_loss=0.1167, over 5668650.61 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3417, pruned_loss=0.08882, over 5624313.15 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3723, pruned_loss=0.12, over 5659287.65 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:24:19,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:24:20,968 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:24:21,912 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.798e+02 1.854e+03 2.316e+03 3.425e+03 1.144e+04, threshold=4.632e+03, percent-clipped=14.0
+2023-03-12 20:24:30,678 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.80 vs. limit=5.0
+2023-03-12 20:24:45,955 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:24:57,917 INFO [train.py:968] (0/2) Epoch 25, batch 7500, giga_loss[loss=0.3062, simple_loss=0.3673, pruned_loss=0.1225, over 27523.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3666, pruned_loss=0.1157, over 5678776.13 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3413, pruned_loss=0.08856, over 5632802.69 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3705, pruned_loss=0.1193, over 5665189.94 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:25:02,076 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 20:25:02,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8376, 3.6657, 3.4728, 1.5758], device='cuda:0'), covar=tensor([0.0772, 0.0896, 0.0856, 0.2327], device='cuda:0'), in_proj_covar=tensor([0.1290, 0.1196, 0.1006, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 20:25:42,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:25:44,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:25:47,581 INFO [train.py:968] (0/2) Epoch 25, batch 7550, giga_loss[loss=0.2745, simple_loss=0.3327, pruned_loss=0.1081, over 23819.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.366, pruned_loss=0.1145, over 5669165.06 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3409, pruned_loss=0.08833, over 5641473.36 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3703, pruned_loss=0.1184, over 5651164.33 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:25:54,063 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.710e+03 2.164e+03 2.859e+03 7.738e+03, threshold=4.327e+03, percent-clipped=4.0
+2023-03-12 20:26:09,008 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:26:13,068 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:26:17,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:26:26,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5390, 1.6717, 1.2689, 1.2343], device='cuda:0'), covar=tensor([0.0966, 0.0587, 0.1047, 0.1175], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0451, 0.0522, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 20:26:29,492 INFO [train.py:968] (0/2) Epoch 25, batch 7600, giga_loss[loss=0.2818, simple_loss=0.3596, pruned_loss=0.102, over 28936.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3654, pruned_loss=0.1129, over 5682319.37 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3412, pruned_loss=0.08841, over 5651271.57 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3696, pruned_loss=0.1169, over 5659661.91 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:26:35,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5164, 1.7419, 1.6155, 1.5702], device='cuda:0'), covar=tensor([0.1893, 0.1949, 0.2284, 0.2021], device='cuda:0'), in_proj_covar=tensor([0.0494, 0.0755, 0.0726, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 20:26:41,815 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:26:57,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:27:14,021 INFO [train.py:968] (0/2) Epoch 25, batch 7650, giga_loss[loss=0.2687, simple_loss=0.3418, pruned_loss=0.09777, over 28779.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3654, pruned_loss=0.1129, over 5688096.00 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3411, pruned_loss=0.08835, over 5655746.44 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3693, pruned_loss=0.1168, over 5666506.86 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:27:21,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.104e+03 1.584e+03 1.987e+03 2.673e+03 4.824e+03, threshold=3.975e+03, percent-clipped=3.0
+2023-03-12 20:27:46,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1101528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:27:58,724 INFO [train.py:968] (0/2) Epoch 25, batch 7700, libri_loss[loss=0.2332, simple_loss=0.3129, pruned_loss=0.07676, over 28659.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.364, pruned_loss=0.1122, over 5685989.78 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3414, pruned_loss=0.08864, over 5654067.25 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3678, pruned_loss=0.1159, over 5671840.60 frames. ], batch size: 63, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:28:28,034 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 20:28:46,760 INFO [train.py:968] (0/2) Epoch 25, batch 7750, giga_loss[loss=0.2996, simple_loss=0.3651, pruned_loss=0.1171, over 28926.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3621, pruned_loss=0.1122, over 5673122.05 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3412, pruned_loss=0.08866, over 5661679.88 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3659, pruned_loss=0.1158, over 5655773.26 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:28:55,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.676e+03 2.275e+03 2.936e+03 6.768e+03, threshold=4.551e+03, percent-clipped=11.0
+2023-03-12 20:29:07,527 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:29:10,906 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:29:34,765 INFO [train.py:968] (0/2) Epoch 25, batch 7800, libri_loss[loss=0.2211, simple_loss=0.3002, pruned_loss=0.07106, over 29525.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3606, pruned_loss=0.1119, over 5681567.17 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3406, pruned_loss=0.0883, over 5666899.46 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3647, pruned_loss=0.1157, over 5663151.43 frames. ], batch size: 70, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:29:38,484 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:29:48,527 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1101654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:30:24,956 INFO [train.py:968] (0/2) Epoch 25, batch 7850, giga_loss[loss=0.2717, simple_loss=0.3441, pruned_loss=0.0997, over 28902.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3615, pruned_loss=0.1136, over 5670075.45 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3407, pruned_loss=0.08837, over 5669111.29 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3648, pruned_loss=0.1167, over 5653680.44 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:30:33,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.219e+03 1.852e+03 2.332e+03 3.064e+03 1.047e+04, threshold=4.665e+03, percent-clipped=3.0
+2023-03-12 20:30:47,011 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 20:31:10,715 INFO [train.py:968] (0/2) Epoch 25, batch 7900, giga_loss[loss=0.2866, simple_loss=0.3604, pruned_loss=0.1064, over 29019.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1129, over 5668868.09 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3404, pruned_loss=0.08812, over 5672908.68 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3632, pruned_loss=0.1163, over 5652256.12 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:31:40,350 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1101774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:31:52,978 INFO [train.py:968] (0/2) Epoch 25, batch 7950, giga_loss[loss=0.2848, simple_loss=0.353, pruned_loss=0.1082, over 28841.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3596, pruned_loss=0.1126, over 5662893.15 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3402, pruned_loss=0.08794, over 5673239.04 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3634, pruned_loss=0.1165, over 5648438.63 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:31:54,085 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-12 20:32:01,563 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.192e+03 1.728e+03 2.403e+03 3.477e+03 8.987e+03, threshold=4.805e+03, percent-clipped=9.0
+2023-03-12 20:32:24,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8906, 3.2043, 2.1267, 0.9740], device='cuda:0'), covar=tensor([0.9013, 0.3662, 0.4043, 0.7917], device='cuda:0'), in_proj_covar=tensor([0.1805, 0.1701, 0.1632, 0.1461], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 20:32:35,541 INFO [train.py:968] (0/2) Epoch 25, batch 8000, giga_loss[loss=0.2686, simple_loss=0.3469, pruned_loss=0.09519, over 28903.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.36, pruned_loss=0.1121, over 5669785.51 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3395, pruned_loss=0.08762, over 5677037.14 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3647, pruned_loss=0.1169, over 5654288.88 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:32:42,727 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0118, 5.8193, 5.5240, 2.9489], device='cuda:0'), covar=tensor([0.0527, 0.0715, 0.0920, 0.1701], device='cuda:0'), in_proj_covar=tensor([0.1291, 0.1194, 0.1007, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 20:33:20,107 INFO [train.py:968] (0/2) Epoch 25, batch 8050, giga_loss[loss=0.3296, simple_loss=0.3945, pruned_loss=0.1324, over 28556.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3605, pruned_loss=0.1113, over 5676591.23 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3396, pruned_loss=0.08755, over 5685958.61 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3652, pruned_loss=0.1163, over 5655565.19 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:33:26,497 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.52 vs. limit=5.0
+2023-03-12 20:33:29,469 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.545e+03 2.141e+03 2.766e+03 8.562e+03, threshold=4.282e+03, percent-clipped=2.0
+2023-03-12 20:33:31,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4801, 1.7798, 1.4099, 1.7295], device='cuda:0'), covar=tensor([0.2704, 0.2684, 0.3076, 0.2369], device='cuda:0'), in_proj_covar=tensor([0.1556, 0.1121, 0.1373, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 20:33:32,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:33:38,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3921, 1.6311, 1.5268, 1.3019], device='cuda:0'), covar=tensor([0.2892, 0.2532, 0.1992, 0.2519], device='cuda:0'), in_proj_covar=tensor([0.2023, 0.1955, 0.1884, 0.2026], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 20:34:03,003 INFO [train.py:968] (0/2) Epoch 25, batch 8100, giga_loss[loss=0.2393, simple_loss=0.3185, pruned_loss=0.08004, over 28158.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3594, pruned_loss=0.1095, over 5687919.48 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3392, pruned_loss=0.08737, over 5691309.64 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3644, pruned_loss=0.1147, over 5665491.50 frames. ], batch size: 77, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:34:47,635 INFO [train.py:968] (0/2) Epoch 25, batch 8150, giga_loss[loss=0.2717, simple_loss=0.3455, pruned_loss=0.09898, over 28910.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3614, pruned_loss=0.1104, over 5694032.81 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3398, pruned_loss=0.08777, over 5692511.93 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3655, pruned_loss=0.1147, over 5674908.40 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:34:57,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.994e+02 1.631e+03 1.905e+03 2.483e+03 5.681e+03, threshold=3.809e+03, percent-clipped=7.0
+2023-03-12 20:34:57,496 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1102000.pt
+2023-03-12 20:35:23,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1102029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:35:27,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5492, 1.9133, 1.5105, 1.7456], device='cuda:0'), covar=tensor([0.2602, 0.2629, 0.3000, 0.2390], device='cuda:0'), in_proj_covar=tensor([0.1555, 0.1121, 0.1373, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 20:35:34,310 INFO [train.py:968] (0/2) Epoch 25, batch 8200, giga_loss[loss=0.2901, simple_loss=0.3507, pruned_loss=0.1147, over 28782.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3632, pruned_loss=0.1125, over 5694646.77 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3395, pruned_loss=0.08764, over 5697818.16 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3672, pruned_loss=0.1166, over 5674696.10 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:35:40,596 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1102046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:35:44,588 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1102049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:36:10,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3909, 1.5724, 1.6226, 1.4834], device='cuda:0'), covar=tensor([0.1560, 0.1364, 0.1603, 0.1433], device='cuda:0'), in_proj_covar=tensor([0.0493, 0.0754, 0.0727, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 20:36:12,136 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1102078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:36:27,301 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1031, 1.3717, 2.7398, 2.6757], device='cuda:0'), covar=tensor([0.1300, 0.2191, 0.0578, 0.1193], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0666, 0.0988, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 20:36:27,710 INFO [train.py:968] (0/2) Epoch 25, batch 8250, giga_loss[loss=0.3899, simple_loss=0.4063, pruned_loss=0.1867, over 23642.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.367, pruned_loss=0.1166, over 5677503.34 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3401, pruned_loss=0.0881, over 5700011.77 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3698, pruned_loss=0.1197, over 5659698.28 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:36:36,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.923e+03 2.359e+03 3.412e+03 1.136e+04, threshold=4.719e+03, percent-clipped=22.0
+2023-03-12 20:37:04,087 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-12 20:37:16,020 INFO [train.py:968] (0/2) Epoch 25, batch 8300, giga_loss[loss=0.306, simple_loss=0.3679, pruned_loss=0.1221, over 28659.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.369, pruned_loss=0.1201, over 5666260.61 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3403, pruned_loss=0.08824, over 5699675.72 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3719, pruned_loss=0.1233, over 5651218.78 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:37:25,962 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1102149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:37:28,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3827, 3.5194, 1.5022, 1.5272], device='cuda:0'), covar=tensor([0.1004, 0.0429, 0.0871, 0.1357], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0565, 0.0398, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-12 20:37:47,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5645, 2.3281, 1.7159, 0.8313], device='cuda:0'), covar=tensor([0.6936, 0.3126, 0.4130, 0.6860], device='cuda:0'), in_proj_covar=tensor([0.1804, 0.1703, 0.1633, 0.1460], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 20:37:49,274 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1102172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:37:51,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1102175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:38:04,333 INFO [train.py:968] (0/2) Epoch 25, batch 8350, libri_loss[loss=0.2969, simple_loss=0.3878, pruned_loss=0.1029, over 27691.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3689, pruned_loss=0.1204, over 5681268.27 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3401, pruned_loss=0.08795, over 5706017.30 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3725, pruned_loss=0.1245, over 5662351.39 frames. ], batch size: 116, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:38:15,482 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.819e+03 2.372e+03 2.994e+03 8.152e+03, threshold=4.744e+03, percent-clipped=7.0
+2023-03-12 20:38:17,012 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1102204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:38:52,432 INFO [train.py:968] (0/2) Epoch 25, batch 8400, libri_loss[loss=0.3019, simple_loss=0.3506, pruned_loss=0.1266, over 29343.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3698, pruned_loss=0.1216, over 5666067.18 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3402, pruned_loss=0.08823, over 5702580.99 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3734, pruned_loss=0.1255, over 5652593.37 frames. ], batch size: 71, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:39:35,028 INFO [train.py:968] (0/2) Epoch 25, batch 8450, giga_loss[loss=0.2606, simple_loss=0.3398, pruned_loss=0.09071, over 29014.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3677, pruned_loss=0.1197, over 5675539.72 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3399, pruned_loss=0.08808, over 5706365.82 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3722, pruned_loss=0.1245, over 5659384.02 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:39:36,475 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1102292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:39:39,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1102295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:39:44,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.145e+03 1.842e+03 2.572e+03 4.130e+03 7.460e+03, threshold=5.144e+03, percent-clipped=16.0
+2023-03-12 20:40:03,743 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1102324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:40:04,657 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 20:40:06,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1619, 1.2690, 1.1850, 0.9206], device='cuda:0'), covar=tensor([0.1067, 0.0548, 0.1075, 0.1102], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0449, 0.0518, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 20:40:17,631 INFO [train.py:968] (0/2) Epoch 25, batch 8500, libri_loss[loss=0.3029, simple_loss=0.3709, pruned_loss=0.1174, over 29543.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3665, pruned_loss=0.1169, over 5683716.82 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3404, pruned_loss=0.08841, over 5705855.41 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.37, pruned_loss=0.1208, over 5670737.09 frames. ], batch size: 80, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:40:43,963 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5725, 1.8737, 1.5524, 1.5379], device='cuda:0'), covar=tensor([0.2297, 0.2217, 0.2420, 0.2145], device='cuda:0'), in_proj_covar=tensor([0.1553, 0.1119, 0.1370, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 20:41:03,340 INFO [train.py:968] (0/2) Epoch 25, batch 8550, libri_loss[loss=0.2795, simple_loss=0.3595, pruned_loss=0.09975, over 29537.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3652, pruned_loss=0.1153, over 5679061.67 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3408, pruned_loss=0.08869, over 5705229.89 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.368, pruned_loss=0.1187, over 5668890.78 frames. ], batch size: 82, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:41:11,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5497, 1.6550, 1.7609, 1.3416], device='cuda:0'), covar=tensor([0.1752, 0.2445, 0.1452, 0.1718], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0712, 0.0967, 0.0864], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 20:41:12,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.518e+03 2.027e+03 2.653e+03 6.297e+03, threshold=4.055e+03, percent-clipped=5.0
+2023-03-12 20:41:15,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9665, 3.8070, 3.6336, 1.6600], device='cuda:0'), covar=tensor([0.0762, 0.0864, 0.0827, 0.2146], device='cuda:0'), in_proj_covar=tensor([0.1289, 0.1188, 0.1005, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 20:41:31,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4044, 3.5206, 1.6302, 1.5191], device='cuda:0'), covar=tensor([0.1019, 0.0319, 0.0872, 0.1355], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0567, 0.0400, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-12 20:41:43,939 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-12 20:41:46,220 INFO [train.py:968] (0/2) Epoch 25, batch 8600, giga_loss[loss=0.3362, simple_loss=0.373, pruned_loss=0.1497, over 23620.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3634, pruned_loss=0.1148, over 5674409.72 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3407, pruned_loss=0.08855, over 5710163.92 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3665, pruned_loss=0.1184, over 5661005.43 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:42:27,319 INFO [train.py:968] (0/2) Epoch 25, batch 8650, giga_loss[loss=0.2615, simple_loss=0.3364, pruned_loss=0.09336, over 28999.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3626, pruned_loss=0.1147, over 5676689.10 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3411, pruned_loss=0.08876, over 5705560.97 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3651, pruned_loss=0.1179, over 5668855.95 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:42:43,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.731e+03 2.259e+03 3.426e+03 1.421e+04, threshold=4.517e+03, percent-clipped=15.0
+2023-03-12 20:43:19,219 INFO [train.py:968] (0/2) Epoch 25, batch 8700, giga_loss[loss=0.349, simple_loss=0.4035, pruned_loss=0.1473, over 28723.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.365, pruned_loss=0.117, over 5675589.82 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3411, pruned_loss=0.08878, over 5708662.40 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3673, pruned_loss=0.1199, over 5666212.26 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:44:07,253 INFO [train.py:968] (0/2) Epoch 25, batch 8750, giga_loss[loss=0.3022, simple_loss=0.3886, pruned_loss=0.1079, over 29007.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.368, pruned_loss=0.1182, over 5676604.85 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.341, pruned_loss=0.0887, over 5710171.85 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5667203.57 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:44:18,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.852e+02 1.665e+03 2.135e+03 3.082e+03 6.365e+03, threshold=4.270e+03, percent-clipped=10.0
+2023-03-12 20:44:49,193 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2236, 1.5675, 1.5278, 1.0779], device='cuda:0'), covar=tensor([0.1856, 0.2737, 0.1580, 0.1850], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0713, 0.0967, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 20:44:53,560 INFO [train.py:968] (0/2) Epoch 25, batch 8800, giga_loss[loss=0.304, simple_loss=0.3758, pruned_loss=0.116, over 28756.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3718, pruned_loss=0.1181, over 5678410.64 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.341, pruned_loss=0.08866, over 5713224.51 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3741, pruned_loss=0.1209, over 5668066.05 frames. ], batch size: 243, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:45:43,073 INFO [train.py:968] (0/2) Epoch 25, batch 8850, giga_loss[loss=0.3478, simple_loss=0.4036, pruned_loss=0.146, over 28302.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3749, pruned_loss=0.1201, over 5675063.11 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3412, pruned_loss=0.08879, over 5710386.38 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3767, pruned_loss=0.1223, over 5669490.77 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:45:52,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.661e+03 2.147e+03 2.776e+03 9.058e+03, threshold=4.294e+03, percent-clipped=7.0
+2023-03-12 20:46:00,005 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-12 20:46:24,171 INFO [train.py:968] (0/2) Epoch 25, batch 8900, giga_loss[loss=0.336, simple_loss=0.3938, pruned_loss=0.1391, over 28584.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3754, pruned_loss=0.1207, over 5686757.34 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3419, pruned_loss=0.08919, over 5714374.14 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.377, pruned_loss=0.1228, over 5678189.16 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:46:59,558 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1102775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:47:14,640 INFO [train.py:968] (0/2) Epoch 25, batch 8950, giga_loss[loss=0.4802, simple_loss=0.4834, pruned_loss=0.2384, over 26622.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3758, pruned_loss=0.1219, over 5682409.94 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.342, pruned_loss=0.08931, over 5715425.44 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3773, pruned_loss=0.1237, over 5674611.66 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:47:25,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.704e+02 1.801e+03 2.224e+03 3.037e+03 4.837e+03, threshold=4.447e+03, percent-clipped=6.0
+2023-03-12 20:48:03,301 INFO [train.py:968] (0/2) Epoch 25, batch 9000, giga_loss[loss=0.265, simple_loss=0.3367, pruned_loss=0.09664, over 28880.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3759, pruned_loss=0.1233, over 5685522.76 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.342, pruned_loss=0.08935, over 5716333.31 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3772, pruned_loss=0.1249, over 5678520.82 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:48:03,304 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 20:48:11,869 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2069, simple_loss=0.3152, pruned_loss=0.0493, over 944034.00 frames. 
+2023-03-12 20:48:11,870 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 20:48:44,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6383, 2.1004, 1.7658, 1.8918], device='cuda:0'), covar=tensor([0.0738, 0.0277, 0.0287, 0.0782], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0064, 0.0112], device='cuda:0')
+2023-03-12 20:48:54,320 INFO [train.py:968] (0/2) Epoch 25, batch 9050, giga_loss[loss=0.3815, simple_loss=0.4168, pruned_loss=0.1731, over 28241.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.372, pruned_loss=0.1209, over 5689542.34 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3418, pruned_loss=0.08927, over 5719805.67 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3743, pruned_loss=0.1233, over 5679741.01 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:49:06,168 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.766e+03 2.510e+03 3.840e+03 1.113e+04, threshold=5.020e+03, percent-clipped=19.0
+2023-03-12 20:49:41,929 INFO [train.py:968] (0/2) Epoch 25, batch 9100, libri_loss[loss=0.2411, simple_loss=0.3302, pruned_loss=0.07606, over 29587.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3707, pruned_loss=0.121, over 5679937.31 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3419, pruned_loss=0.08925, over 5721726.93 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3727, pruned_loss=0.1232, over 5669971.78 frames. ], batch size: 74, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:50:18,888 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-12 20:50:28,152 INFO [train.py:968] (0/2) Epoch 25, batch 9150, giga_loss[loss=0.2755, simple_loss=0.3403, pruned_loss=0.1053, over 28531.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3698, pruned_loss=0.1206, over 5675655.64 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3425, pruned_loss=0.08963, over 5715059.57 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3715, pruned_loss=0.1228, over 5672525.61 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:50:34,813 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 20:50:38,097 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.793e+03 2.342e+03 3.258e+03 8.377e+03, threshold=4.684e+03, percent-clipped=4.0
+2023-03-12 20:51:13,243 INFO [train.py:968] (0/2) Epoch 25, batch 9200, giga_loss[loss=0.3351, simple_loss=0.3879, pruned_loss=0.1411, over 27707.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3699, pruned_loss=0.1206, over 5677623.18 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08995, over 5720077.15 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3716, pruned_loss=0.1233, over 5669467.21 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:51:56,073 INFO [train.py:968] (0/2) Epoch 25, batch 9250, giga_loss[loss=0.285, simple_loss=0.3522, pruned_loss=0.1089, over 29029.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3674, pruned_loss=0.1191, over 5681419.38 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3427, pruned_loss=0.08958, over 5726300.97 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3701, pruned_loss=0.1225, over 5667804.94 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:52:09,379 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.148e+03 1.908e+03 2.287e+03 3.502e+03 8.455e+03, threshold=4.574e+03, percent-clipped=12.0
+2023-03-12 20:52:13,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1103108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:52:26,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3660, 2.8802, 1.5042, 1.4889], device='cuda:0'), covar=tensor([0.0947, 0.0373, 0.0849, 0.1291], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0567, 0.0399, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-12 20:52:43,600 INFO [train.py:968] (0/2) Epoch 25, batch 9300, giga_loss[loss=0.2441, simple_loss=0.3239, pruned_loss=0.08217, over 28896.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3664, pruned_loss=0.1187, over 5688133.72 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3424, pruned_loss=0.08944, over 5729422.69 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3695, pruned_loss=0.1224, over 5673110.21 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:52:51,684 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1103150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:52:52,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4450, 1.5736, 1.6832, 1.2451], device='cuda:0'), covar=tensor([0.1798, 0.2685, 0.1475, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.0921, 0.0711, 0.0966, 0.0864], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 20:53:23,776 INFO [train.py:968] (0/2) Epoch 25, batch 9350, giga_loss[loss=0.3016, simple_loss=0.3764, pruned_loss=0.1134, over 28585.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3671, pruned_loss=0.1185, over 5690382.95 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3428, pruned_loss=0.08975, over 5724942.68 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3701, pruned_loss=0.1223, over 5680990.32 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:53:38,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.648e+03 2.057e+03 2.746e+03 6.612e+03, threshold=4.113e+03, percent-clipped=6.0
+2023-03-12 20:54:13,998 INFO [train.py:968] (0/2) Epoch 25, batch 9400, giga_loss[loss=0.2947, simple_loss=0.3693, pruned_loss=0.1101, over 29007.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.37, pruned_loss=0.1203, over 5676018.93 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3431, pruned_loss=0.08998, over 5720261.22 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3724, pruned_loss=0.1234, over 5671987.50 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:55:00,518 INFO [train.py:968] (0/2) Epoch 25, batch 9450, giga_loss[loss=0.299, simple_loss=0.3624, pruned_loss=0.1179, over 29031.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3694, pruned_loss=0.1201, over 5677130.70 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.343, pruned_loss=0.08987, over 5724980.29 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3721, pruned_loss=0.1234, over 5668599.21 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:55:03,017 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1103293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:55:05,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1103296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:55:12,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 2.025e+03 2.617e+03 4.030e+03 9.204e+03, threshold=5.234e+03, percent-clipped=22.0
+2023-03-12 20:55:32,979 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1103325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:55:45,125 INFO [train.py:968] (0/2) Epoch 25, batch 9500, giga_loss[loss=0.2752, simple_loss=0.3611, pruned_loss=0.09469, over 28646.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3704, pruned_loss=0.1196, over 5679223.64 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3429, pruned_loss=0.08988, over 5726476.26 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3729, pruned_loss=0.1226, over 5670366.78 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:56:28,980 INFO [train.py:968] (0/2) Epoch 25, batch 9550, giga_loss[loss=0.3305, simple_loss=0.3917, pruned_loss=0.1347, over 28650.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3717, pruned_loss=0.1183, over 5678197.87 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.343, pruned_loss=0.08984, over 5724754.44 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3743, pruned_loss=0.1214, over 5671591.48 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:56:42,388 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.447e+02 1.575e+03 2.109e+03 2.890e+03 1.006e+04, threshold=4.217e+03, percent-clipped=6.0
+2023-03-12 20:57:10,451 INFO [train.py:968] (0/2) Epoch 25, batch 9600, giga_loss[loss=0.3098, simple_loss=0.3792, pruned_loss=0.1202, over 29065.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3726, pruned_loss=0.1169, over 5682421.90 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3431, pruned_loss=0.08996, over 5723959.98 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3758, pruned_loss=0.1204, over 5675707.44 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:57:52,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1103483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:57:54,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3124, 1.7414, 1.4179, 1.4877], device='cuda:0'), covar=tensor([0.0720, 0.0338, 0.0315, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0064, 0.0112], device='cuda:0')
+2023-03-12 20:57:59,536 INFO [train.py:968] (0/2) Epoch 25, batch 9650, giga_loss[loss=0.3248, simple_loss=0.3905, pruned_loss=0.1296, over 28835.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3762, pruned_loss=0.12, over 5683708.44 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.343, pruned_loss=0.08995, over 5728847.52 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3797, pruned_loss=0.1236, over 5672777.05 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:58:09,837 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1103501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:58:13,347 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 1.603e+03 2.271e+03 3.017e+03 7.152e+03, threshold=4.542e+03, percent-clipped=5.0
+2023-03-12 20:58:30,896 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2699, 1.6344, 1.2874, 0.9344], device='cuda:0'), covar=tensor([0.2616, 0.2616, 0.3131, 0.2401], device='cuda:0'), in_proj_covar=tensor([0.1554, 0.1118, 0.1371, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 20:58:32,254 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-12 20:58:40,633 INFO [train.py:968] (0/2) Epoch 25, batch 9700, giga_loss[loss=0.3109, simple_loss=0.3754, pruned_loss=0.1232, over 28745.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3779, pruned_loss=0.1224, over 5667191.40 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3433, pruned_loss=0.09025, over 5710403.79 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3812, pruned_loss=0.1258, over 5673539.42 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:59:00,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1589, 1.4305, 1.0935, 1.0241], device='cuda:0'), covar=tensor([0.1328, 0.0684, 0.1452, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0451, 0.0521, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 20:59:29,173 INFO [train.py:968] (0/2) Epoch 25, batch 9750, giga_loss[loss=0.2949, simple_loss=0.3637, pruned_loss=0.113, over 28909.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.378, pruned_loss=0.1237, over 5653656.31 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3434, pruned_loss=0.09033, over 5702037.78 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3808, pruned_loss=0.1265, over 5665393.43 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:59:43,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.202e+03 1.894e+03 2.537e+03 3.598e+03 1.241e+04, threshold=5.074e+03, percent-clipped=14.0
+2023-03-12 21:00:04,020 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1103626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:00:06,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1103629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:00:13,686 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-12 21:00:16,867 INFO [train.py:968] (0/2) Epoch 25, batch 9800, giga_loss[loss=0.2629, simple_loss=0.3354, pruned_loss=0.09519, over 29073.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3773, pruned_loss=0.124, over 5639972.99 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3435, pruned_loss=0.09038, over 5703161.59 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3796, pruned_loss=0.1263, over 5648082.97 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:00:31,854 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1103658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:01:00,175 INFO [train.py:968] (0/2) Epoch 25, batch 9850, giga_loss[loss=0.3036, simple_loss=0.378, pruned_loss=0.1147, over 28988.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3759, pruned_loss=0.1219, over 5642507.17 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3434, pruned_loss=0.09039, over 5699124.46 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3785, pruned_loss=0.1244, over 5651282.87 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:01:12,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.566e+03 2.132e+03 3.160e+03 1.213e+04, threshold=4.264e+03, percent-clipped=11.0
+2023-03-12 21:01:41,277 INFO [train.py:968] (0/2) Epoch 25, batch 9900, giga_loss[loss=0.2628, simple_loss=0.3394, pruned_loss=0.0931, over 28537.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3753, pruned_loss=0.1199, over 5653470.22 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3432, pruned_loss=0.09034, over 5695728.29 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3784, pruned_loss=0.1228, over 5661474.16 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:01:50,942 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2268, 1.3376, 1.4005, 1.1585], device='cuda:0'), covar=tensor([0.3124, 0.2714, 0.2016, 0.2614], device='cuda:0'), in_proj_covar=tensor([0.2025, 0.1959, 0.1884, 0.2031], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 21:01:57,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4094, 2.1275, 1.8205, 1.6014], device='cuda:0'), covar=tensor([0.0804, 0.0269, 0.0309, 0.0916], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0112], device='cuda:0')
+2023-03-12 21:02:27,674 INFO [train.py:968] (0/2) Epoch 25, batch 9950, giga_loss[loss=0.424, simple_loss=0.4319, pruned_loss=0.2081, over 23533.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3757, pruned_loss=0.1197, over 5656832.58 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3434, pruned_loss=0.0904, over 5698798.63 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3785, pruned_loss=0.1224, over 5659738.63 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 21:02:44,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.296e+02 1.824e+03 2.265e+03 3.206e+03 1.102e+04, threshold=4.530e+03, percent-clipped=14.0
+2023-03-12 21:03:13,690 INFO [train.py:968] (0/2) Epoch 25, batch 10000, libri_loss[loss=0.2783, simple_loss=0.3618, pruned_loss=0.09744, over 29742.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3751, pruned_loss=0.1196, over 5658383.38 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3434, pruned_loss=0.09026, over 5702670.66 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3783, pruned_loss=0.1227, over 5655682.35 frames. ], batch size: 87, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:03:45,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1103876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:03:58,128 INFO [train.py:968] (0/2) Epoch 25, batch 10050, giga_loss[loss=0.2882, simple_loss=0.3559, pruned_loss=0.1103, over 29053.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3738, pruned_loss=0.1187, over 5670162.47 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3437, pruned_loss=0.09023, over 5706666.92 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3772, pruned_loss=0.1224, over 5662704.00 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:04:10,335 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1103903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:04:14,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.311e+02 1.621e+03 2.003e+03 2.686e+03 8.581e+03, threshold=4.006e+03, percent-clipped=8.0
+2023-03-12 21:04:47,634 INFO [train.py:968] (0/2) Epoch 25, batch 10100, giga_loss[loss=0.4693, simple_loss=0.4655, pruned_loss=0.2366, over 26655.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3733, pruned_loss=0.1199, over 5665681.91 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3439, pruned_loss=0.09048, over 5712715.32 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3767, pruned_loss=0.1234, over 5653313.79 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:05:32,289 INFO [train.py:968] (0/2) Epoch 25, batch 10150, giga_loss[loss=0.2724, simple_loss=0.3538, pruned_loss=0.09549, over 28999.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3704, pruned_loss=0.1187, over 5666473.94 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3437, pruned_loss=0.0904, over 5706447.53 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3738, pruned_loss=0.122, over 5661259.10 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:05:42,740 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1104000.pt
+2023-03-12 21:05:48,711 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 21:05:50,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.864e+03 2.322e+03 3.017e+03 9.993e+03, threshold=4.645e+03, percent-clipped=14.0
+2023-03-12 21:06:01,425 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:06:04,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:06:26,599 INFO [train.py:968] (0/2) Epoch 25, batch 10200, giga_loss[loss=0.2967, simple_loss=0.3642, pruned_loss=0.1146, over 29001.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3695, pruned_loss=0.1192, over 5664505.11 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3438, pruned_loss=0.09054, over 5707473.67 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3724, pruned_loss=0.1221, over 5659056.92 frames. ], batch size: 112, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:06:37,233 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:07:02,115 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:07:11,727 INFO [train.py:968] (0/2) Epoch 25, batch 10250, giga_loss[loss=0.3944, simple_loss=0.4225, pruned_loss=0.1832, over 26623.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3689, pruned_loss=0.1198, over 5666777.21 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3433, pruned_loss=0.0903, over 5711482.43 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3722, pruned_loss=0.123, over 5657935.31 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:07:25,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.616e+02 1.816e+03 2.146e+03 2.927e+03 9.190e+03, threshold=4.292e+03, percent-clipped=7.0
+2023-03-12 21:07:41,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6447, 1.5564, 1.8561, 1.4836], device='cuda:0'), covar=tensor([0.1445, 0.2169, 0.1198, 0.1518], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0713, 0.0968, 0.0864], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 21:07:55,478 INFO [train.py:968] (0/2) Epoch 25, batch 10300, giga_loss[loss=0.3031, simple_loss=0.3729, pruned_loss=0.1167, over 28887.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3662, pruned_loss=0.1174, over 5669932.26 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.343, pruned_loss=0.09019, over 5716575.70 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3697, pruned_loss=0.1207, over 5657729.16 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:08:39,027 INFO [train.py:968] (0/2) Epoch 25, batch 10350, giga_loss[loss=0.3069, simple_loss=0.3658, pruned_loss=0.124, over 26637.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3635, pruned_loss=0.1141, over 5664291.94 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3433, pruned_loss=0.09038, over 5711580.51 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3665, pruned_loss=0.1171, over 5658070.19 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:08:46,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6431, 1.7862, 1.4926, 1.8725], device='cuda:0'), covar=tensor([0.2541, 0.2685, 0.2977, 0.2476], device='cuda:0'), in_proj_covar=tensor([0.1558, 0.1122, 0.1374, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 21:08:55,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.509e+03 2.101e+03 2.590e+03 1.122e+04, threshold=4.202e+03, percent-clipped=4.0
+2023-03-12 21:09:28,557 INFO [train.py:968] (0/2) Epoch 25, batch 10400, giga_loss[loss=0.332, simple_loss=0.3876, pruned_loss=0.1382, over 27596.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3622, pruned_loss=0.1128, over 5661712.62 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3434, pruned_loss=0.09031, over 5715362.95 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.365, pruned_loss=0.1157, over 5652294.16 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:09:32,384 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:10:03,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:10:08,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-12 21:10:14,338 INFO [train.py:968] (0/2) Epoch 25, batch 10450, libri_loss[loss=0.237, simple_loss=0.3224, pruned_loss=0.07582, over 29654.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3606, pruned_loss=0.1119, over 5670840.27 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3431, pruned_loss=0.09006, over 5719509.00 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3636, pruned_loss=0.1152, over 5657956.11 frames. ], batch size: 69, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:10:29,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.676e+03 2.088e+03 3.268e+03 9.178e+03, threshold=4.175e+03, percent-clipped=9.0
+2023-03-12 21:10:55,202 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:11:01,792 INFO [train.py:968] (0/2) Epoch 25, batch 10500, giga_loss[loss=0.274, simple_loss=0.3436, pruned_loss=0.1022, over 28867.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3584, pruned_loss=0.1118, over 5664068.60 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3433, pruned_loss=0.09022, over 5710599.86 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3608, pruned_loss=0.1147, over 5660540.19 frames. ], batch size: 263, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:11:10,840 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104349.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 21:11:36,943 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:11:47,663 INFO [train.py:968] (0/2) Epoch 25, batch 10550, giga_loss[loss=0.278, simple_loss=0.3609, pruned_loss=0.09757, over 29003.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3605, pruned_loss=0.1132, over 5668092.03 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3438, pruned_loss=0.09049, over 5709357.32 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3625, pruned_loss=0.1158, over 5665232.01 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:12:04,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.900e+03 2.353e+03 3.165e+03 5.935e+03, threshold=4.707e+03, percent-clipped=10.0
+2023-03-12 21:12:04,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2125, 4.0641, 3.8519, 1.6190], device='cuda:0'), covar=tensor([0.0624, 0.0703, 0.0746, 0.2326], device='cuda:0'), in_proj_covar=tensor([0.1287, 0.1191, 0.1004, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 21:12:12,842 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-12 21:12:14,931 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:12:17,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:12:32,171 INFO [train.py:968] (0/2) Epoch 25, batch 10600, giga_loss[loss=0.2896, simple_loss=0.3637, pruned_loss=0.1077, over 28668.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3626, pruned_loss=0.1138, over 5669434.63 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3436, pruned_loss=0.09023, over 5712541.92 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3648, pruned_loss=0.1167, over 5663198.17 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:12:42,334 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:12:44,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:13:16,524 INFO [train.py:968] (0/2) Epoch 25, batch 10650, giga_loss[loss=0.2895, simple_loss=0.362, pruned_loss=0.1085, over 28960.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3639, pruned_loss=0.1145, over 5656396.74 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3435, pruned_loss=0.0901, over 5715348.39 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3663, pruned_loss=0.1177, over 5647293.67 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:13:33,210 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5930, 1.7429, 1.2541, 1.3997], device='cuda:0'), covar=tensor([0.1061, 0.0670, 0.1036, 0.1328], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0449, 0.0520, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 21:13:35,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.929e+02 1.659e+03 2.186e+03 2.940e+03 7.457e+03, threshold=4.372e+03, percent-clipped=3.0
+2023-03-12 21:14:07,248 INFO [train.py:968] (0/2) Epoch 25, batch 10700, giga_loss[loss=0.2687, simple_loss=0.344, pruned_loss=0.0967, over 28955.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3644, pruned_loss=0.1154, over 5638649.37 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3437, pruned_loss=0.09015, over 5717981.04 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3664, pruned_loss=0.1183, over 5627844.47 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:14:32,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1448, 3.0038, 2.8455, 1.5069], device='cuda:0'), covar=tensor([0.1065, 0.1078, 0.0972, 0.2509], device='cuda:0'), in_proj_covar=tensor([0.1287, 0.1188, 0.1003, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 21:14:52,416 INFO [train.py:968] (0/2) Epoch 25, batch 10750, giga_loss[loss=0.249, simple_loss=0.3274, pruned_loss=0.08529, over 29042.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3666, pruned_loss=0.1176, over 5642574.93 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.344, pruned_loss=0.09029, over 5711153.27 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3683, pruned_loss=0.12, over 5638863.42 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:14:59,246 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:15:04,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:15:09,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.772e+03 2.248e+03 2.987e+03 7.418e+03, threshold=4.495e+03, percent-clipped=8.0
+2023-03-12 21:15:24,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:15:35,534 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:15:45,209 INFO [train.py:968] (0/2) Epoch 25, batch 10800, giga_loss[loss=0.2894, simple_loss=0.3677, pruned_loss=0.1056, over 28984.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3684, pruned_loss=0.1185, over 5640993.61 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09026, over 5711223.80 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.1209, over 5636841.58 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:16:27,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6061, 1.7246, 1.7405, 1.4990], device='cuda:0'), covar=tensor([0.3282, 0.2845, 0.2479, 0.2815], device='cuda:0'), in_proj_covar=tensor([0.2018, 0.1957, 0.1881, 0.2027], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 21:16:34,020 INFO [train.py:968] (0/2) Epoch 25, batch 10850, giga_loss[loss=0.2842, simple_loss=0.3557, pruned_loss=0.1064, over 28796.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3693, pruned_loss=0.1191, over 5649402.89 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3441, pruned_loss=0.09036, over 5715879.38 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3711, pruned_loss=0.1216, over 5640605.60 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:16:50,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:16:51,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.341e+03 1.797e+03 2.276e+03 3.028e+03 5.408e+03, threshold=4.552e+03, percent-clipped=5.0
+2023-03-12 21:16:55,592 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104713.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 21:17:02,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3651, 1.4253, 1.3261, 1.4922], device='cuda:0'), covar=tensor([0.0683, 0.0434, 0.0332, 0.0729], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-12 21:17:05,492 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104724.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 21:17:20,040 INFO [train.py:968] (0/2) Epoch 25, batch 10900, giga_loss[loss=0.3027, simple_loss=0.3769, pruned_loss=0.1142, over 28974.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3714, pruned_loss=0.1209, over 5652859.92 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09025, over 5717893.23 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3731, pruned_loss=0.1232, over 5643579.91 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:17:33,801 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:17:37,128 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-12 21:17:41,480 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4874, 1.8057, 1.4635, 1.6098], device='cuda:0'), covar=tensor([0.2630, 0.2552, 0.2994, 0.2228], device='cuda:0'), in_proj_covar=tensor([0.1559, 0.1123, 0.1375, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 21:17:42,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:17:45,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:18:08,033 INFO [train.py:968] (0/2) Epoch 25, batch 10950, giga_loss[loss=0.335, simple_loss=0.3965, pruned_loss=0.1368, over 28838.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3725, pruned_loss=0.122, over 5657071.03 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3439, pruned_loss=0.09024, over 5721884.16 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3746, pruned_loss=0.1246, over 5644725.97 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:18:14,085 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:18:18,870 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5227, 1.6264, 1.6829, 1.5053], device='cuda:0'), covar=tensor([0.2979, 0.2581, 0.2089, 0.2282], device='cuda:0'), in_proj_covar=tensor([0.2023, 0.1962, 0.1886, 0.2032], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 21:18:24,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.702e+03 2.138e+03 2.858e+03 4.972e+03, threshold=4.275e+03, percent-clipped=2.0
+2023-03-12 21:18:44,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7475, 4.6639, 1.7873, 1.9649], device='cuda:0'), covar=tensor([0.0940, 0.0379, 0.0896, 0.1240], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0566, 0.0400, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-12 21:18:55,094 INFO [train.py:968] (0/2) Epoch 25, batch 11000, libri_loss[loss=0.2798, simple_loss=0.3603, pruned_loss=0.09967, over 29191.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3715, pruned_loss=0.119, over 5667711.09 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3439, pruned_loss=0.09017, over 5728759.25 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3743, pruned_loss=0.1223, over 5649263.72 frames. ], batch size: 101, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:19:05,624 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:19:08,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:19:21,067 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104867.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 21:19:24,968 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104870.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 21:19:37,107 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:19:45,951 INFO [train.py:968] (0/2) Epoch 25, batch 11050, giga_loss[loss=0.3364, simple_loss=0.3775, pruned_loss=0.1476, over 23654.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3703, pruned_loss=0.1187, over 5652505.18 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3438, pruned_loss=0.09009, over 5730405.49 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3731, pruned_loss=0.1219, over 5635162.24 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:19:51,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:19:53,188 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104899.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 21:19:53,250 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:20:01,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.153e+03 1.757e+03 2.366e+03 3.340e+03 9.347e+03, threshold=4.733e+03, percent-clipped=13.0
+2023-03-12 21:20:21,746 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:20:33,276 INFO [train.py:968] (0/2) Epoch 25, batch 11100, giga_loss[loss=0.2695, simple_loss=0.3394, pruned_loss=0.09973, over 28844.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3684, pruned_loss=0.1178, over 5672430.65 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3435, pruned_loss=0.08985, over 5735680.49 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3717, pruned_loss=0.1215, over 5651415.65 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:21:03,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:21:04,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4984, 2.4006, 2.3034, 2.1608], device='cuda:0'), covar=tensor([0.1851, 0.2455, 0.2248, 0.2267], device='cuda:0'), in_proj_covar=tensor([0.0494, 0.0754, 0.0725, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 21:21:31,779 INFO [train.py:968] (0/2) Epoch 25, batch 11150, giga_loss[loss=0.3053, simple_loss=0.3643, pruned_loss=0.1232, over 28993.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3692, pruned_loss=0.1192, over 5662550.75 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.344, pruned_loss=0.09012, over 5737491.51 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3716, pruned_loss=0.1221, over 5643600.38 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:21:49,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.788e+03 2.329e+03 3.022e+03 8.752e+03, threshold=4.658e+03, percent-clipped=6.0
+2023-03-12 21:21:49,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5900, 1.1794, 4.8184, 3.5132], device='cuda:0'), covar=tensor([0.1680, 0.2977, 0.0399, 0.0977], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0665, 0.0983, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 21:22:19,795 INFO [train.py:968] (0/2) Epoch 25, batch 11200, giga_loss[loss=0.3068, simple_loss=0.3716, pruned_loss=0.1211, over 28671.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3681, pruned_loss=0.1189, over 5675971.95 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09003, over 5739734.38 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3707, pruned_loss=0.1217, over 5658002.22 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:23:04,561 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1105088.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 21:23:05,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5786, 1.8093, 1.4749, 1.6020], device='cuda:0'), covar=tensor([0.2570, 0.2684, 0.2966, 0.2496], device='cuda:0'), in_proj_covar=tensor([0.1555, 0.1120, 0.1372, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 21:23:05,808 INFO [train.py:968] (0/2) Epoch 25, batch 11250, giga_loss[loss=0.2856, simple_loss=0.36, pruned_loss=0.1056, over 28920.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3675, pruned_loss=0.1191, over 5671590.09 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3438, pruned_loss=0.09007, over 5736544.71 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3695, pruned_loss=0.1214, over 5660213.59 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:23:20,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6101, 1.7459, 1.6751, 1.5957], device='cuda:0'), covar=tensor([0.1823, 0.2029, 0.2228, 0.2041], device='cuda:0'), in_proj_covar=tensor([0.0494, 0.0755, 0.0726, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 21:23:21,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.094e+03 1.527e+03 2.019e+03 2.714e+03 6.246e+03, threshold=4.037e+03, percent-clipped=6.0
+2023-03-12 21:23:28,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-12 21:23:30,231 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-03-12 21:23:52,920 INFO [train.py:968] (0/2) Epoch 25, batch 11300, giga_loss[loss=0.2905, simple_loss=0.3652, pruned_loss=0.1079, over 28950.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3675, pruned_loss=0.1194, over 5668419.16 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3438, pruned_loss=0.09002, over 5736035.47 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3695, pruned_loss=0.1218, over 5658903.52 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:24:39,972 INFO [train.py:968] (0/2) Epoch 25, batch 11350, giga_loss[loss=0.3036, simple_loss=0.3846, pruned_loss=0.1113, over 28963.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3671, pruned_loss=0.1191, over 5669811.96 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.343, pruned_loss=0.08948, over 5738909.15 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.37, pruned_loss=0.1223, over 5657969.99 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:24:57,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.746e+03 2.129e+03 2.835e+03 7.860e+03, threshold=4.257e+03, percent-clipped=7.0
+2023-03-12 21:25:20,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1105231.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 21:25:22,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1105234.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 21:25:27,460 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1105239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:25:27,887 INFO [train.py:968] (0/2) Epoch 25, batch 11400, giga_loss[loss=0.3195, simple_loss=0.3821, pruned_loss=0.1284, over 28978.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3697, pruned_loss=0.1214, over 5672397.77 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3432, pruned_loss=0.08969, over 5740424.30 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3723, pruned_loss=0.1242, over 5660763.56 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:25:32,598 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.07 vs. limit=2.0
+2023-03-12 21:25:49,810 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1105263.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 21:26:13,250 INFO [train.py:968] (0/2) Epoch 25, batch 11450, giga_loss[loss=0.3545, simple_loss=0.4035, pruned_loss=0.1527, over 28924.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3701, pruned_loss=0.1216, over 5673829.13 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3431, pruned_loss=0.08975, over 5741132.59 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3731, pruned_loss=0.1248, over 5661756.92 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:26:25,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5240, 1.8148, 1.4661, 1.4150], device='cuda:0'), covar=tensor([0.2517, 0.2548, 0.2944, 0.2281], device='cuda:0'), in_proj_covar=tensor([0.1559, 0.1123, 0.1376, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 21:26:34,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.885e+02 1.854e+03 2.477e+03 3.001e+03 8.216e+03, threshold=4.954e+03, percent-clipped=12.0
+2023-03-12 21:27:03,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1105339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:27:03,515 INFO [train.py:968] (0/2) Epoch 25, batch 11500, giga_loss[loss=0.3238, simple_loss=0.3808, pruned_loss=0.1334, over 28908.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3691, pruned_loss=0.1214, over 5670772.09 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3426, pruned_loss=0.08939, over 5744402.34 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3725, pruned_loss=0.1249, over 5656899.67 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:27:25,144 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1105365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:27:35,990 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6992, 1.7519, 1.9026, 1.4630], device='cuda:0'), covar=tensor([0.1794, 0.2582, 0.1493, 0.1753], device='cuda:0'), in_proj_covar=tensor([0.0916, 0.0712, 0.0965, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 21:27:47,823 INFO [train.py:968] (0/2) Epoch 25, batch 11550, giga_loss[loss=0.3244, simple_loss=0.3884, pruned_loss=0.1303, over 28609.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3688, pruned_loss=0.1211, over 5659309.29 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3428, pruned_loss=0.08946, over 5738539.11 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3718, pruned_loss=0.1245, over 5651314.85 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:28:06,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.733e+03 2.207e+03 2.961e+03 4.766e+03, threshold=4.413e+03, percent-clipped=0.0
+2023-03-12 21:28:16,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4383, 3.5893, 1.6088, 1.6323], device='cuda:0'), covar=tensor([0.1082, 0.0409, 0.0923, 0.1429], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0566, 0.0399, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-12 21:28:26,435 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-12 21:28:37,141 INFO [train.py:968] (0/2) Epoch 25, batch 11600, giga_loss[loss=0.3021, simple_loss=0.3673, pruned_loss=0.1185, over 28617.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3683, pruned_loss=0.12, over 5663944.69 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3429, pruned_loss=0.08953, over 5732085.60 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3711, pruned_loss=0.1232, over 5662035.55 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:29:09,711 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-12 21:29:15,177 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1105482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:29:18,019 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1105485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:29:21,991 INFO [train.py:968] (0/2) Epoch 25, batch 11650, giga_loss[loss=0.2859, simple_loss=0.3599, pruned_loss=0.1059, over 28932.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3687, pruned_loss=0.1199, over 5663316.59 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08944, over 5734516.00 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3713, pruned_loss=0.1229, over 5658983.55 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:29:41,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.157e+03 1.693e+03 2.156e+03 2.953e+03 1.058e+04, threshold=4.312e+03, percent-clipped=13.0
+2023-03-12 21:29:45,324 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1105514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:30:12,999 INFO [train.py:968] (0/2) Epoch 25, batch 11700, giga_loss[loss=0.3224, simple_loss=0.3936, pruned_loss=0.1256, over 28904.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.37, pruned_loss=0.1203, over 5675478.53 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08946, over 5736669.88 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3725, pruned_loss=0.1235, over 5668386.63 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:31:01,669 INFO [train.py:968] (0/2) Epoch 25, batch 11750, giga_loss[loss=0.3513, simple_loss=0.3962, pruned_loss=0.1532, over 27607.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3734, pruned_loss=0.1235, over 5670606.96 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08942, over 5738192.20 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.376, pruned_loss=0.1267, over 5662533.03 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:31:17,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.186e+03 1.660e+03 2.030e+03 2.580e+03 5.988e+03, threshold=4.061e+03, percent-clipped=4.0
+2023-03-12 21:31:24,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1105614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:31:41,446 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8713, 3.6866, 3.5036, 1.6822], device='cuda:0'), covar=tensor([0.0767, 0.0896, 0.0899, 0.2190], device='cuda:0'), in_proj_covar=tensor([0.1291, 0.1193, 0.1005, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 21:31:48,412 INFO [train.py:968] (0/2) Epoch 25, batch 11800, giga_loss[loss=0.4081, simple_loss=0.4276, pruned_loss=0.1943, over 26747.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3738, pruned_loss=0.1235, over 5681968.62 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3434, pruned_loss=0.08968, over 5740334.21 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3759, pruned_loss=0.1262, over 5672992.38 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:31:48,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4103, 1.2403, 3.6202, 3.2585], device='cuda:0'), covar=tensor([0.1585, 0.2939, 0.0517, 0.1363], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0664, 0.0983, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 21:32:32,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2257, 1.4040, 1.3496, 1.1700], device='cuda:0'), covar=tensor([0.2428, 0.2237, 0.1547, 0.2054], device='cuda:0'), in_proj_covar=tensor([0.2026, 0.1964, 0.1884, 0.2031], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 21:32:35,011 INFO [train.py:968] (0/2) Epoch 25, batch 11850, giga_loss[loss=0.2845, simple_loss=0.3635, pruned_loss=0.1027, over 28557.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3742, pruned_loss=0.1228, over 5686086.53 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3435, pruned_loss=0.08976, over 5743646.64 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3763, pruned_loss=0.1255, over 5675198.49 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:32:52,436 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-12 21:32:54,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.863e+03 2.477e+03 3.650e+03 9.148e+03, threshold=4.954e+03, percent-clipped=15.0
+2023-03-12 21:33:13,908 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1105732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:33:22,569 INFO [train.py:968] (0/2) Epoch 25, batch 11900, giga_loss[loss=0.3607, simple_loss=0.4156, pruned_loss=0.1529, over 27996.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.373, pruned_loss=0.1213, over 5673236.26 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3431, pruned_loss=0.08955, over 5742721.34 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3757, pruned_loss=0.1244, over 5663935.32 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:33:22,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1105740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:33:23,024 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-12 21:33:38,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5796, 1.8492, 1.5189, 1.4750], device='cuda:0'), covar=tensor([0.2627, 0.2702, 0.3076, 0.2443], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1124, 0.1377, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 21:33:39,223 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1105757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:33:41,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1105760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:34:06,071 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-12 21:34:09,785 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1105789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:34:10,450 INFO [train.py:968] (0/2) Epoch 25, batch 11950, giga_loss[loss=0.2885, simple_loss=0.354, pruned_loss=0.1115, over 28653.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3722, pruned_loss=0.121, over 5677827.85 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3432, pruned_loss=0.08963, over 5745508.98 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3748, pruned_loss=0.1239, over 5666641.53 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:34:28,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.583e+03 2.016e+03 2.920e+03 5.355e+03, threshold=4.032e+03, percent-clipped=2.0
+2023-03-12 21:34:53,055 INFO [train.py:968] (0/2) Epoch 25, batch 12000, libri_loss[loss=0.2425, simple_loss=0.3241, pruned_loss=0.08048, over 29570.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3708, pruned_loss=0.12, over 5692933.24 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3432, pruned_loss=0.08961, over 5748822.34 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3738, pruned_loss=0.1232, over 5679182.64 frames. ], batch size: 76, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:34:53,058 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 21:35:02,177 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2065, simple_loss=0.3143, pruned_loss=0.04934, over 944034.00 frames. 
+2023-03-12 21:35:02,177 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 21:35:08,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5969, 3.2667, 2.9134, 2.3181], device='cuda:0'), covar=tensor([0.2694, 0.1475, 0.1703, 0.2249], device='cuda:0'), in_proj_covar=tensor([0.2030, 0.1965, 0.1888, 0.2032], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 21:35:40,973 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1105883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:35:42,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1105886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:35:46,216 INFO [train.py:968] (0/2) Epoch 25, batch 12050, libri_loss[loss=0.2817, simple_loss=0.3677, pruned_loss=0.09781, over 29492.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3714, pruned_loss=0.1208, over 5659544.32 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3431, pruned_loss=0.0896, over 5734991.40 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3748, pruned_loss=0.1245, over 5657669.27 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:36:04,570 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-03-12 21:36:06,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.632e+03 1.972e+03 2.841e+03 8.069e+03, threshold=3.944e+03, percent-clipped=12.0
+2023-03-12 21:36:09,952 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1105915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:36:25,832 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1105932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:36:33,542 INFO [train.py:968] (0/2) Epoch 25, batch 12100, giga_loss[loss=0.3126, simple_loss=0.3766, pruned_loss=0.1243, over 28896.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3729, pruned_loss=0.1215, over 5663199.09 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3434, pruned_loss=0.08967, over 5735840.51 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3756, pruned_loss=0.1247, over 5660318.02 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:37:20,628 INFO [train.py:968] (0/2) Epoch 25, batch 12150, giga_loss[loss=0.323, simple_loss=0.3844, pruned_loss=0.1308, over 29051.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3733, pruned_loss=0.1229, over 5656682.02 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3439, pruned_loss=0.08997, over 5730779.24 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3757, pruned_loss=0.1259, over 5656664.36 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:37:28,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1105999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:37:29,286 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1106000.pt
+2023-03-12 21:37:41,168 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.498e+03 2.164e+03 3.294e+03 1.514e+04, threshold=4.328e+03, percent-clipped=18.0
+2023-03-12 21:38:07,321 INFO [train.py:968] (0/2) Epoch 25, batch 12200, giga_loss[loss=0.3999, simple_loss=0.4321, pruned_loss=0.1839, over 28675.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3734, pruned_loss=0.1236, over 5659289.07 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3435, pruned_loss=0.08965, over 5731830.07 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3762, pruned_loss=0.1269, over 5657157.30 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:38:58,399 INFO [train.py:968] (0/2) Epoch 25, batch 12250, libri_loss[loss=0.2367, simple_loss=0.323, pruned_loss=0.07524, over 29549.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3742, pruned_loss=0.1246, over 5656959.53 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3434, pruned_loss=0.08958, over 5732719.92 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3767, pruned_loss=0.1274, over 5654021.45 frames. ], batch size: 78, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:39:16,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1106107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:39:18,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.616e+02 1.648e+03 2.151e+03 3.154e+03 6.076e+03, threshold=4.302e+03, percent-clipped=5.0
+2023-03-12 21:39:45,753 INFO [train.py:968] (0/2) Epoch 25, batch 12300, giga_loss[loss=0.3024, simple_loss=0.3586, pruned_loss=0.1231, over 28699.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3743, pruned_loss=0.1242, over 5653611.11 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3439, pruned_loss=0.08993, over 5733639.15 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3762, pruned_loss=0.1267, over 5649400.50 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:40:05,991 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1106163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:40:27,320 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4157, 1.7633, 1.4347, 1.2267], device='cuda:0'), covar=tensor([0.2123, 0.2048, 0.2275, 0.2158], device='cuda:0'), in_proj_covar=tensor([0.1564, 0.1125, 0.1377, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 21:40:32,430 INFO [train.py:968] (0/2) Epoch 25, batch 12350, giga_loss[loss=0.2895, simple_loss=0.3592, pruned_loss=0.1099, over 28700.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.375, pruned_loss=0.1252, over 5638917.65 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.344, pruned_loss=0.0898, over 5735975.57 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3772, pruned_loss=0.1281, over 5631754.57 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:40:36,542 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.50 vs. limit=5.0
+2023-03-12 21:40:52,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.637e+03 2.195e+03 3.046e+03 1.075e+04, threshold=4.391e+03, percent-clipped=6.0
+2023-03-12 21:41:02,096 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1106220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:41:17,683 INFO [train.py:968] (0/2) Epoch 25, batch 12400, giga_loss[loss=0.3243, simple_loss=0.3816, pruned_loss=0.1335, over 28900.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3748, pruned_loss=0.1248, over 5647804.43 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3438, pruned_loss=0.08979, over 5738166.07 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3776, pruned_loss=0.1281, over 5637284.05 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:41:26,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1106250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:41:28,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1106253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:41:52,450 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1106282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:41:59,063 INFO [train.py:968] (0/2) Epoch 25, batch 12450, giga_loss[loss=0.3156, simple_loss=0.3722, pruned_loss=0.1295, over 28827.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3739, pruned_loss=0.1233, over 5637404.63 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3441, pruned_loss=0.08997, over 5725581.01 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3768, pruned_loss=0.1268, over 5636689.22 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:42:14,171 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1106307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:42:17,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.154e+03 1.648e+03 2.094e+03 2.917e+03 6.957e+03, threshold=4.189e+03, percent-clipped=5.0
+2023-03-12 21:42:48,387 INFO [train.py:968] (0/2) Epoch 25, batch 12500, giga_loss[loss=0.3282, simple_loss=0.3689, pruned_loss=0.1438, over 23747.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3728, pruned_loss=0.1223, over 5645633.38 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3447, pruned_loss=0.09027, over 5727117.49 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3753, pruned_loss=0.1257, over 5641524.52 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:43:17,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1106374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:43:29,463 INFO [train.py:968] (0/2) Epoch 25, batch 12550, giga_loss[loss=0.2854, simple_loss=0.3466, pruned_loss=0.1121, over 28707.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3697, pruned_loss=0.1202, over 5649042.93 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3448, pruned_loss=0.09028, over 5722331.28 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3724, pruned_loss=0.1238, over 5648732.72 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:43:43,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4588, 2.0254, 1.4820, 0.8070], device='cuda:0'), covar=tensor([0.6323, 0.2927, 0.3842, 0.6611], device='cuda:0'), in_proj_covar=tensor([0.1801, 0.1696, 0.1632, 0.1457], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 21:43:53,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.839e+03 2.624e+03 3.875e+03 1.013e+04, threshold=5.249e+03, percent-clipped=16.0
+2023-03-12 21:44:18,940 INFO [train.py:968] (0/2) Epoch 25, batch 12600, giga_loss[loss=0.2563, simple_loss=0.3328, pruned_loss=0.0899, over 28946.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3651, pruned_loss=0.1172, over 5669800.63 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3442, pruned_loss=0.08996, over 5725948.06 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3681, pruned_loss=0.1208, over 5665146.20 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:44:31,627 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1106450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:44:33,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1106453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:44:45,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1227, 1.3861, 1.3494, 1.0535], device='cuda:0'), covar=tensor([0.1312, 0.1937, 0.1110, 0.1390], device='cuda:0'), in_proj_covar=tensor([0.0921, 0.0714, 0.0969, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 21:45:00,456 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1106482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:45:06,126 INFO [train.py:968] (0/2) Epoch 25, batch 12650, giga_loss[loss=0.3178, simple_loss=0.3699, pruned_loss=0.1329, over 27465.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3612, pruned_loss=0.1156, over 5653204.42 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3442, pruned_loss=0.08992, over 5729078.83 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.364, pruned_loss=0.119, over 5645463.18 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:45:17,702 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6961, 1.8464, 1.9142, 1.4655], device='cuda:0'), covar=tensor([0.1781, 0.2518, 0.1482, 0.1778], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0714, 0.0970, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 21:45:28,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.733e+03 2.237e+03 2.960e+03 7.610e+03, threshold=4.473e+03, percent-clipped=5.0
+2023-03-12 21:45:33,121 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1106517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:45:35,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1106520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:45:52,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1106538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:45:54,313 INFO [train.py:968] (0/2) Epoch 25, batch 12700, giga_loss[loss=0.3055, simple_loss=0.3669, pruned_loss=0.122, over 28373.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3621, pruned_loss=0.1173, over 5656344.30 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3447, pruned_loss=0.0902, over 5732589.77 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3642, pruned_loss=0.1202, over 5645655.73 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:46:02,470 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1106549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:46:27,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6882, 2.0177, 1.9954, 1.4757], device='cuda:0'), covar=tensor([0.2228, 0.2964, 0.1851, 0.2271], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0713, 0.0968, 0.0864], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 21:46:40,315 INFO [train.py:968] (0/2) Epoch 25, batch 12750, giga_loss[loss=0.3515, simple_loss=0.3935, pruned_loss=0.1547, over 27586.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3604, pruned_loss=0.116, over 5657434.25 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3444, pruned_loss=0.08997, over 5736746.21 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3627, pruned_loss=0.1192, over 5643266.95 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:46:46,668 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1106595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:47:00,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.852e+03 2.292e+03 3.421e+03 1.134e+04, threshold=4.584e+03, percent-clipped=12.0
+2023-03-12 21:47:30,183 INFO [train.py:968] (0/2) Epoch 25, batch 12800, giga_loss[loss=0.2792, simple_loss=0.3516, pruned_loss=0.1034, over 28908.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3589, pruned_loss=0.1134, over 5657928.48 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3442, pruned_loss=0.0899, over 5739210.72 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3612, pruned_loss=0.1165, over 5643230.86 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:48:09,713 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1106681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:48:11,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1106684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:48:16,765 INFO [train.py:968] (0/2) Epoch 25, batch 12850, giga_loss[loss=0.2862, simple_loss=0.34, pruned_loss=0.1162, over 24291.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3576, pruned_loss=0.1102, over 5658181.23 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3442, pruned_loss=0.08986, over 5742003.39 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.36, pruned_loss=0.1134, over 5641237.15 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:48:38,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4003, 3.6857, 1.5784, 1.5841], device='cuda:0'), covar=tensor([0.1028, 0.0351, 0.0963, 0.1384], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0568, 0.0400, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-12 21:48:41,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.530e+03 1.942e+03 2.813e+03 5.831e+03, threshold=3.884e+03, percent-clipped=3.0
+2023-03-12 21:48:42,220 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1106713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:49:04,833 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1106738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:49:07,838 INFO [train.py:968] (0/2) Epoch 25, batch 12900, giga_loss[loss=0.2092, simple_loss=0.2803, pruned_loss=0.06907, over 24284.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3556, pruned_loss=0.1076, over 5657322.43 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.344, pruned_loss=0.08985, over 5743669.42 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3579, pruned_loss=0.1105, over 5640659.07 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:49:08,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1106741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:49:36,276 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1106770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:49:47,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6922, 1.9347, 1.3537, 1.4948], device='cuda:0'), covar=tensor([0.0901, 0.0489, 0.0946, 0.1113], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0447, 0.0521, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 21:49:57,306 INFO [train.py:968] (0/2) Epoch 25, batch 12950, giga_loss[loss=0.2523, simple_loss=0.3343, pruned_loss=0.08513, over 28848.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3532, pruned_loss=0.1049, over 5646128.62 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.344, pruned_loss=0.09002, over 5731663.38 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3551, pruned_loss=0.1073, over 5641422.93 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:50:21,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.770e+02 1.510e+03 1.843e+03 2.454e+03 7.361e+03, threshold=3.685e+03, percent-clipped=5.0
+2023-03-12 21:50:33,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1106826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:50:45,878 INFO [train.py:968] (0/2) Epoch 25, batch 13000, giga_loss[loss=0.3345, simple_loss=0.3913, pruned_loss=0.1388, over 28308.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3503, pruned_loss=0.1018, over 5648216.59 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3441, pruned_loss=0.09013, over 5736007.79 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3519, pruned_loss=0.104, over 5638560.49 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:51:23,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3995, 3.1241, 1.3914, 1.5851], device='cuda:0'), covar=tensor([0.0976, 0.0288, 0.0963, 0.1320], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0564, 0.0398, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-12 21:51:33,258 INFO [train.py:968] (0/2) Epoch 25, batch 13050, giga_loss[loss=0.2584, simple_loss=0.348, pruned_loss=0.08439, over 29095.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3485, pruned_loss=0.09835, over 5657563.83 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.343, pruned_loss=0.08969, over 5732002.65 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.351, pruned_loss=0.1009, over 5650304.67 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:51:56,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.516e+02 1.379e+03 2.027e+03 3.176e+03 1.032e+04, threshold=4.055e+03, percent-clipped=14.0
+2023-03-12 21:52:22,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5851, 1.6319, 1.8016, 1.4295], device='cuda:0'), covar=tensor([0.1609, 0.2419, 0.1349, 0.1760], device='cuda:0'), in_proj_covar=tensor([0.0914, 0.0705, 0.0960, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 21:52:25,640 INFO [train.py:968] (0/2) Epoch 25, batch 13100, giga_loss[loss=0.2495, simple_loss=0.3329, pruned_loss=0.08304, over 29013.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3501, pruned_loss=0.09984, over 5651329.72 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3427, pruned_loss=0.08952, over 5733946.91 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3526, pruned_loss=0.1022, over 5642467.13 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:53:01,870 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 21:53:14,264 INFO [train.py:968] (0/2) Epoch 25, batch 13150, giga_loss[loss=0.2649, simple_loss=0.3386, pruned_loss=0.09558, over 28663.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3492, pruned_loss=0.09912, over 5645283.01 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3427, pruned_loss=0.08966, over 5726160.30 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3512, pruned_loss=0.101, over 5644607.54 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:53:36,155 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.475e+02 1.478e+03 1.864e+03 2.744e+03 1.023e+04, threshold=3.728e+03, percent-clipped=5.0
+2023-03-12 21:54:00,354 INFO [train.py:968] (0/2) Epoch 25, batch 13200, giga_loss[loss=0.2541, simple_loss=0.3311, pruned_loss=0.08854, over 28218.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3456, pruned_loss=0.097, over 5644229.67 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.342, pruned_loss=0.0894, over 5733183.60 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3481, pruned_loss=0.09904, over 5634346.87 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:54:21,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:54:22,145 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-12 21:54:46,101 INFO [train.py:968] (0/2) Epoch 25, batch 13250, libri_loss[loss=0.2847, simple_loss=0.365, pruned_loss=0.1021, over 29530.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3454, pruned_loss=0.0969, over 5647150.15 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3417, pruned_loss=0.08926, over 5734139.59 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3477, pruned_loss=0.09882, over 5636254.66 frames. ], batch size: 89, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:55:07,778 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-12 21:55:08,521 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.530e+02 1.543e+03 1.939e+03 2.562e+03 5.301e+03, threshold=3.877e+03, percent-clipped=6.0
+2023-03-12 21:55:08,724 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:55:23,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5123, 1.9507, 1.8247, 1.7012], device='cuda:0'), covar=tensor([0.2068, 0.1873, 0.2055, 0.1926], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0745, 0.0717, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 21:55:33,020 INFO [train.py:968] (0/2) Epoch 25, batch 13300, giga_loss[loss=0.3004, simple_loss=0.3598, pruned_loss=0.1205, over 26710.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3447, pruned_loss=0.09642, over 5649455.52 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.341, pruned_loss=0.0891, over 5737456.96 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3472, pruned_loss=0.09826, over 5635804.46 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:55:44,351 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 21:56:24,694 INFO [train.py:968] (0/2) Epoch 25, batch 13350, giga_loss[loss=0.2683, simple_loss=0.3479, pruned_loss=0.09429, over 28969.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3419, pruned_loss=0.09414, over 5652517.28 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3409, pruned_loss=0.08916, over 5739229.87 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.344, pruned_loss=0.09559, over 5639445.20 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:56:35,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1107201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:56:49,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.625e+02 1.453e+03 1.821e+03 2.513e+03 6.639e+03, threshold=3.642e+03, percent-clipped=4.0
+2023-03-12 21:57:13,419 INFO [train.py:968] (0/2) Epoch 25, batch 13400, giga_loss[loss=0.2599, simple_loss=0.3426, pruned_loss=0.08858, over 28834.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3397, pruned_loss=0.09215, over 5654126.34 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3408, pruned_loss=0.08921, over 5740752.31 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3414, pruned_loss=0.09334, over 5641012.10 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:58:08,563 INFO [train.py:968] (0/2) Epoch 25, batch 13450, giga_loss[loss=0.241, simple_loss=0.3206, pruned_loss=0.08073, over 28692.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3359, pruned_loss=0.09047, over 5655525.32 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3404, pruned_loss=0.08906, over 5742745.02 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3375, pruned_loss=0.09158, over 5641651.99 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:58:32,709 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.515e+02 1.497e+03 1.896e+03 2.486e+03 5.498e+03, threshold=3.792e+03, percent-clipped=5.0
+2023-03-12 21:58:59,520 INFO [train.py:968] (0/2) Epoch 25, batch 13500, giga_loss[loss=0.2697, simple_loss=0.3374, pruned_loss=0.101, over 27620.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3342, pruned_loss=0.08984, over 5659774.19 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3395, pruned_loss=0.08882, over 5743889.07 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3362, pruned_loss=0.09098, over 5644327.42 frames. ], batch size: 474, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:59:03,262 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1107344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:59:05,270 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1107347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:59:18,893 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0439, 3.8692, 3.6830, 1.8038], device='cuda:0'), covar=tensor([0.0717, 0.0854, 0.0930, 0.2237], device='cuda:0'), in_proj_covar=tensor([0.1264, 0.1172, 0.0984, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 21:59:33,516 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1107376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:59:46,456 INFO [train.py:968] (0/2) Epoch 25, batch 13550, libri_loss[loss=0.2401, simple_loss=0.3141, pruned_loss=0.08308, over 29591.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3335, pruned_loss=0.08994, over 5658548.02 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3391, pruned_loss=0.08883, over 5749962.43 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3353, pruned_loss=0.0909, over 5637676.38 frames. ], batch size: 75, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:00:10,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.467e+02 1.402e+03 1.879e+03 2.676e+03 4.937e+03, threshold=3.759e+03, percent-clipped=7.0
+2023-03-12 22:00:42,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1107438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:00:43,906 INFO [train.py:968] (0/2) Epoch 25, batch 13600, giga_loss[loss=0.2679, simple_loss=0.35, pruned_loss=0.09288, over 29042.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3366, pruned_loss=0.0919, over 5655236.32 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3386, pruned_loss=0.08875, over 5748464.36 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3383, pruned_loss=0.09274, over 5638480.96 frames. ], batch size: 200, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:01:37,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1107488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:01:38,160 INFO [train.py:968] (0/2) Epoch 25, batch 13650, giga_loss[loss=0.2852, simple_loss=0.3632, pruned_loss=0.1036, over 28339.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3397, pruned_loss=0.09257, over 5648413.51 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3385, pruned_loss=0.08891, over 5740760.45 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3412, pruned_loss=0.09322, over 5639393.66 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:01:49,023 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-12 22:02:07,291 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.696e+02 1.469e+03 1.891e+03 2.369e+03 5.613e+03, threshold=3.783e+03, percent-clipped=5.0
+2023-03-12 22:02:37,030 INFO [train.py:968] (0/2) Epoch 25, batch 13700, giga_loss[loss=0.2817, simple_loss=0.3587, pruned_loss=0.1023, over 29012.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3403, pruned_loss=0.09242, over 5653949.18 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3387, pruned_loss=0.08898, over 5733424.62 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3413, pruned_loss=0.09293, over 5651003.62 frames. ], batch size: 285, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:02:55,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4321, 1.8537, 1.8244, 1.5313], device='cuda:0'), covar=tensor([0.1970, 0.1495, 0.2050, 0.1783], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0741, 0.0711, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 22:02:55,399 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-12 22:03:21,306 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 22:03:31,622 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1107581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:03:35,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4035, 1.8734, 1.7875, 1.5788], device='cuda:0'), covar=tensor([0.2274, 0.1833, 0.2090, 0.2098], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0741, 0.0712, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 22:03:35,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1107584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:03:42,653 INFO [train.py:968] (0/2) Epoch 25, batch 13750, giga_loss[loss=0.2162, simple_loss=0.3026, pruned_loss=0.06483, over 28507.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3386, pruned_loss=0.09138, over 5659346.02 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3385, pruned_loss=0.08897, over 5734399.50 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3396, pruned_loss=0.0918, over 5655910.29 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:04:08,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1107613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:04:09,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.182e+02 1.581e+03 1.994e+03 2.862e+03 5.715e+03, threshold=3.987e+03, percent-clipped=10.0
+2023-03-12 22:04:28,257 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1107631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:04:32,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1107634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:04:37,853 INFO [train.py:968] (0/2) Epoch 25, batch 13800, giga_loss[loss=0.2235, simple_loss=0.3109, pruned_loss=0.06809, over 28976.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3374, pruned_loss=0.09018, over 5663075.85 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3382, pruned_loss=0.08901, over 5737777.21 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3384, pruned_loss=0.09054, over 5654903.63 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:05:06,811 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1107663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:05:11,848 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:05:35,608 INFO [train.py:968] (0/2) Epoch 25, batch 13850, giga_loss[loss=0.2299, simple_loss=0.3198, pruned_loss=0.06996, over 28621.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3351, pruned_loss=0.0874, over 5671734.88 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3378, pruned_loss=0.08879, over 5740721.63 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3362, pruned_loss=0.08787, over 5660828.96 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:06:06,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.445e+02 1.268e+03 1.889e+03 2.468e+03 9.357e+03, threshold=3.779e+03, percent-clipped=13.0
+2023-03-12 22:06:39,714 INFO [train.py:968] (0/2) Epoch 25, batch 13900, giga_loss[loss=0.2255, simple_loss=0.308, pruned_loss=0.0715, over 29152.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3332, pruned_loss=0.08784, over 5667090.56 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3376, pruned_loss=0.08879, over 5742372.20 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3342, pruned_loss=0.0882, over 5656500.45 frames. ], batch size: 113, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:07:06,630 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:07:16,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3620, 1.6557, 1.3537, 1.3232], device='cuda:0'), covar=tensor([0.2764, 0.2589, 0.2893, 0.2467], device='cuda:0'), in_proj_covar=tensor([0.1559, 0.1118, 0.1377, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 22:07:38,557 INFO [train.py:968] (0/2) Epoch 25, batch 13950, libri_loss[loss=0.2455, simple_loss=0.3191, pruned_loss=0.08597, over 29565.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3327, pruned_loss=0.08799, over 5663067.83 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3375, pruned_loss=0.0888, over 5736676.47 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3336, pruned_loss=0.08826, over 5657711.52 frames. ], batch size: 77, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:07:41,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:08:06,071 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.420e+03 1.806e+03 2.382e+03 6.171e+03, threshold=3.612e+03, percent-clipped=6.0
+2023-03-12 22:08:09,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:08:10,172 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:08:31,127 INFO [train.py:968] (0/2) Epoch 25, batch 14000, giga_loss[loss=0.2564, simple_loss=0.3262, pruned_loss=0.09333, over 28897.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3328, pruned_loss=0.08822, over 5659609.19 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3376, pruned_loss=0.08896, over 5732647.01 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3332, pruned_loss=0.08825, over 5656219.02 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:08:40,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:08:42,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8454, 2.0902, 1.4647, 1.6876], device='cuda:0'), covar=tensor([0.1037, 0.0717, 0.1022, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0443, 0.0517, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 22:09:32,941 INFO [train.py:968] (0/2) Epoch 25, batch 14050, giga_loss[loss=0.2766, simple_loss=0.3571, pruned_loss=0.09803, over 28628.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3335, pruned_loss=0.08767, over 5650892.37 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3371, pruned_loss=0.08876, over 5734523.59 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3342, pruned_loss=0.08786, over 5646079.76 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:10:01,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.479e+02 1.379e+03 1.713e+03 2.187e+03 4.712e+03, threshold=3.426e+03, percent-clipped=2.0
+2023-03-12 22:10:28,838 INFO [train.py:968] (0/2) Epoch 25, batch 14100, giga_loss[loss=0.2236, simple_loss=0.3143, pruned_loss=0.0664, over 28963.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3332, pruned_loss=0.08701, over 5664695.45 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3366, pruned_loss=0.08854, over 5741942.53 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3341, pruned_loss=0.08729, over 5651027.71 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:11:05,837 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5915, 1.8509, 1.5123, 1.7614], device='cuda:0'), covar=tensor([0.2881, 0.2754, 0.3138, 0.2444], device='cuda:0'), in_proj_covar=tensor([0.1564, 0.1124, 0.1382, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 22:11:21,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2891, 1.6638, 1.5266, 1.4039], device='cuda:0'), covar=tensor([0.2042, 0.1982, 0.2303, 0.2098], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0740, 0.0711, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 22:11:29,245 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-12 22:11:30,697 INFO [train.py:968] (0/2) Epoch 25, batch 14150, giga_loss[loss=0.3006, simple_loss=0.3699, pruned_loss=0.1157, over 28687.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3323, pruned_loss=0.08673, over 5669740.89 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3363, pruned_loss=0.08841, over 5736767.34 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3331, pruned_loss=0.08699, over 5661779.82 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:11:45,139 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1108000.pt
+2023-03-12 22:12:06,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.469e+02 1.404e+03 1.850e+03 2.705e+03 6.068e+03, threshold=3.700e+03, percent-clipped=11.0
+2023-03-12 22:12:36,590 INFO [train.py:968] (0/2) Epoch 25, batch 14200, giga_loss[loss=0.2891, simple_loss=0.3824, pruned_loss=0.09789, over 28627.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.334, pruned_loss=0.08758, over 5679782.35 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3361, pruned_loss=0.08836, over 5738254.84 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3348, pruned_loss=0.08783, over 5671300.81 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:12:44,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:13:42,403 INFO [train.py:968] (0/2) Epoch 25, batch 14250, libri_loss[loss=0.264, simple_loss=0.3425, pruned_loss=0.0928, over 27810.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3382, pruned_loss=0.0879, over 5676886.80 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3358, pruned_loss=0.08813, over 5740656.38 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3391, pruned_loss=0.08829, over 5666866.14 frames. ], batch size: 116, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:14:12,456 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4198, 1.6932, 1.3558, 1.3879], device='cuda:0'), covar=tensor([0.2960, 0.2816, 0.3284, 0.2426], device='cuda:0'), in_proj_covar=tensor([0.1565, 0.1126, 0.1383, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 22:14:13,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.309e+02 1.359e+03 1.893e+03 3.333e+03 1.240e+04, threshold=3.786e+03, percent-clipped=20.0
+2023-03-12 22:14:38,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:14:41,504 INFO [train.py:968] (0/2) Epoch 25, batch 14300, giga_loss[loss=0.2677, simple_loss=0.3574, pruned_loss=0.089, over 28712.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3392, pruned_loss=0.0863, over 5678859.42 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3351, pruned_loss=0.08786, over 5744069.22 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3406, pruned_loss=0.08684, over 5666589.49 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:15:09,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4976, 2.1358, 1.5448, 0.6454], device='cuda:0'), covar=tensor([0.5223, 0.3327, 0.4634, 0.6665], device='cuda:0'), in_proj_covar=tensor([0.1789, 0.1674, 0.1619, 0.1450], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 22:15:15,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:15:37,013 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:15:39,090 INFO [train.py:968] (0/2) Epoch 25, batch 14350, giga_loss[loss=0.2523, simple_loss=0.3426, pruned_loss=0.08102, over 28792.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3387, pruned_loss=0.08476, over 5677785.06 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3352, pruned_loss=0.08791, over 5746985.83 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3398, pruned_loss=0.08511, over 5664121.49 frames. ], batch size: 243, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:15:39,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:15:43,852 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:15:44,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:15:56,165 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6820, 2.0154, 1.8715, 1.5731], device='cuda:0'), covar=tensor([0.2109, 0.2517, 0.2199, 0.2557], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0739, 0.0711, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 22:16:08,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.177e+02 1.480e+03 1.916e+03 2.631e+03 5.915e+03, threshold=3.833e+03, percent-clipped=6.0
+2023-03-12 22:16:11,133 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:16:14,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:16:34,873 INFO [train.py:968] (0/2) Epoch 25, batch 14400, giga_loss[loss=0.2886, simple_loss=0.3649, pruned_loss=0.1061, over 28943.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3375, pruned_loss=0.08463, over 5680185.44 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3341, pruned_loss=0.08733, over 5749117.87 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3395, pruned_loss=0.08533, over 5664718.63 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:16:40,461 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1108242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:16:50,035 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-12 22:17:31,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:17:33,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:17:41,572 INFO [train.py:968] (0/2) Epoch 25, batch 14450, giga_loss[loss=0.2625, simple_loss=0.3432, pruned_loss=0.09089, over 28450.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3373, pruned_loss=0.08565, over 5680254.45 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3342, pruned_loss=0.08742, over 5749587.28 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3388, pruned_loss=0.0861, over 5667341.97 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:18:07,302 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:08,686 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:11,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:13,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.545e+03 1.913e+03 2.596e+03 5.502e+03, threshold=3.826e+03, percent-clipped=7.0
+2023-03-12 22:18:45,052 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:45,692 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:47,066 INFO [train.py:968] (0/2) Epoch 25, batch 14500, giga_loss[loss=0.2598, simple_loss=0.341, pruned_loss=0.08932, over 28958.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3379, pruned_loss=0.08721, over 5693695.84 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3341, pruned_loss=0.08758, over 5752582.25 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3393, pruned_loss=0.08741, over 5678956.94 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:18:47,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:48,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:51,180 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:19:29,915 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:19:32,143 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:19:32,201 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:19:32,921 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:19:34,054 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-12 22:19:40,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2154, 1.2129, 3.5396, 3.1434], device='cuda:0'), covar=tensor([0.1667, 0.2940, 0.0471, 0.1408], device='cuda:0'), in_proj_covar=tensor([0.0772, 0.0656, 0.0967, 0.0931], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 22:19:45,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9375, 3.7887, 3.6003, 1.9730], device='cuda:0'), covar=tensor([0.0600, 0.0719, 0.0809, 0.2296], device='cuda:0'), in_proj_covar=tensor([0.1258, 0.1161, 0.0979, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 22:20:02,947 INFO [train.py:968] (0/2) Epoch 25, batch 14550, giga_loss[loss=0.2232, simple_loss=0.3076, pruned_loss=0.06941, over 28974.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3362, pruned_loss=0.08719, over 5689381.39 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3336, pruned_loss=0.08737, over 5756230.96 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3379, pruned_loss=0.08754, over 5673027.07 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:20:20,032 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:20:44,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3833, 2.0743, 1.5083, 0.6297], device='cuda:0'), covar=tensor([0.5404, 0.2711, 0.4560, 0.6514], device='cuda:0'), in_proj_covar=tensor([0.1785, 0.1669, 0.1617, 0.1450], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 22:20:46,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.901e+02 1.389e+03 1.806e+03 2.626e+03 5.625e+03, threshold=3.612e+03, percent-clipped=8.0
+2023-03-12 22:21:18,872 INFO [train.py:968] (0/2) Epoch 25, batch 14600, giga_loss[loss=0.2525, simple_loss=0.3331, pruned_loss=0.08592, over 28786.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3319, pruned_loss=0.08426, over 5679465.13 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3334, pruned_loss=0.08728, over 5749255.96 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3334, pruned_loss=0.08458, over 5672554.61 frames. ], batch size: 243, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:22:13,688 INFO [train.py:968] (0/2) Epoch 25, batch 14650, giga_loss[loss=0.2103, simple_loss=0.2918, pruned_loss=0.06444, over 28974.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3292, pruned_loss=0.08295, over 5689274.39 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3321, pruned_loss=0.08658, over 5757157.72 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3315, pruned_loss=0.0837, over 5672628.67 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:22:47,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.824e+02 1.430e+03 1.735e+03 2.331e+03 5.872e+03, threshold=3.471e+03, percent-clipped=9.0
+2023-03-12 22:23:04,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1108528.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 22:23:16,132 INFO [train.py:968] (0/2) Epoch 25, batch 14700, giga_loss[loss=0.2386, simple_loss=0.3238, pruned_loss=0.07669, over 28839.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3301, pruned_loss=0.08403, over 5678174.08 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3322, pruned_loss=0.08663, over 5754075.93 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3318, pruned_loss=0.08456, over 5667580.57 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:23:27,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1512, 1.3755, 1.2503, 1.1210], device='cuda:0'), covar=tensor([0.2335, 0.2027, 0.1616, 0.2004], device='cuda:0'), in_proj_covar=tensor([0.1987, 0.1919, 0.1829, 0.1986], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 22:23:54,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2607, 1.6974, 1.5751, 1.5545], device='cuda:0'), covar=tensor([0.2067, 0.1792, 0.1963, 0.1870], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0738, 0.0711, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 22:24:01,424 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7338, 2.0404, 2.0052, 1.4906], device='cuda:0'), covar=tensor([0.1945, 0.2542, 0.1623, 0.1902], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0704, 0.0966, 0.0864], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 22:24:13,567 INFO [train.py:968] (0/2) Epoch 25, batch 14750, giga_loss[loss=0.2671, simple_loss=0.3612, pruned_loss=0.08655, over 28702.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3342, pruned_loss=0.08596, over 5687148.41 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3321, pruned_loss=0.08651, over 5759420.70 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3358, pruned_loss=0.08646, over 5670897.63 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:24:44,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.709e+03 2.151e+03 3.786e+03 9.101e+03, threshold=4.302e+03, percent-clipped=32.0
+2023-03-12 22:24:44,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:24:44,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4358, 2.0157, 1.5078, 0.7990], device='cuda:0'), covar=tensor([0.6375, 0.3177, 0.3985, 0.6553], device='cuda:0'), in_proj_covar=tensor([0.1790, 0.1676, 0.1620, 0.1454], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 22:25:06,643 INFO [train.py:968] (0/2) Epoch 25, batch 14800, giga_loss[loss=0.2499, simple_loss=0.3293, pruned_loss=0.08528, over 28726.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.333, pruned_loss=0.08627, over 5686186.83 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3321, pruned_loss=0.08658, over 5755260.00 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3344, pruned_loss=0.0866, over 5673351.85 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:26:12,045 INFO [train.py:968] (0/2) Epoch 25, batch 14850, giga_loss[loss=0.2796, simple_loss=0.3547, pruned_loss=0.1023, over 27598.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3322, pruned_loss=0.0866, over 5689452.60 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3322, pruned_loss=0.08674, over 5757123.98 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3332, pruned_loss=0.08671, over 5676615.28 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:26:42,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.617e+02 1.372e+03 1.721e+03 2.238e+03 7.910e+03, threshold=3.441e+03, percent-clipped=1.0
+2023-03-12 22:26:57,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.48 vs. limit=5.0
+2023-03-12 22:27:09,818 INFO [train.py:968] (0/2) Epoch 25, batch 14900, giga_loss[loss=0.3095, simple_loss=0.3747, pruned_loss=0.1221, over 28930.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3329, pruned_loss=0.08723, over 5688759.88 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3319, pruned_loss=0.08658, over 5758045.99 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.334, pruned_loss=0.08747, over 5675785.67 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:27:24,143 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1519, 1.6709, 1.5597, 1.3974], device='cuda:0'), covar=tensor([0.2429, 0.1859, 0.2382, 0.2202], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0736, 0.0709, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 22:27:37,649 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:27:41,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:28:13,584 INFO [train.py:968] (0/2) Epoch 25, batch 14950, giga_loss[loss=0.2615, simple_loss=0.3499, pruned_loss=0.08658, over 28447.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3359, pruned_loss=0.08823, over 5686063.25 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3319, pruned_loss=0.08664, over 5755664.30 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3368, pruned_loss=0.0884, over 5676250.94 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:28:17,560 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:28:54,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.910e+02 1.662e+03 2.289e+03 3.176e+03 6.008e+03, threshold=4.577e+03, percent-clipped=19.0
+2023-03-12 22:29:27,257 INFO [train.py:968] (0/2) Epoch 25, batch 15000, giga_loss[loss=0.2377, simple_loss=0.3279, pruned_loss=0.07375, over 28965.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3371, pruned_loss=0.08839, over 5681360.37 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3316, pruned_loss=0.08648, over 5756801.81 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3383, pruned_loss=0.08872, over 5671123.28 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:29:27,261 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 22:29:36,905 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.1953, simple_loss=0.2965, pruned_loss=0.04707, over 944034.00 frames. 
+2023-03-12 22:29:36,906 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 22:30:47,131 INFO [train.py:968] (0/2) Epoch 25, batch 15050, giga_loss[loss=0.2723, simple_loss=0.3473, pruned_loss=0.09871, over 28688.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3349, pruned_loss=0.08778, over 5673649.33 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3308, pruned_loss=0.08611, over 5761495.64 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3366, pruned_loss=0.08843, over 5659269.37 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:31:06,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108903.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 22:31:21,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9489, 1.9227, 1.6825, 1.9585], device='cuda:0'), covar=tensor([0.2758, 0.2908, 0.3203, 0.2548], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1119, 0.1378, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 22:31:21,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.663e+02 1.495e+03 1.852e+03 2.557e+03 5.740e+03, threshold=3.704e+03, percent-clipped=6.0
+2023-03-12 22:31:49,357 INFO [train.py:968] (0/2) Epoch 25, batch 15100, giga_loss[loss=0.2048, simple_loss=0.2911, pruned_loss=0.05922, over 28950.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3304, pruned_loss=0.08644, over 5664578.56 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3306, pruned_loss=0.08612, over 5755518.33 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3319, pruned_loss=0.08697, over 5656338.18 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 22:32:44,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3972, 1.9283, 1.8133, 1.7106], device='cuda:0'), covar=tensor([0.2156, 0.1856, 0.2128, 0.1942], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0736, 0.0709, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-12 22:32:56,046 INFO [train.py:968] (0/2) Epoch 25, batch 15150, giga_loss[loss=0.2853, simple_loss=0.356, pruned_loss=0.1073, over 28635.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3263, pruned_loss=0.08464, over 5664726.22 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3307, pruned_loss=0.0862, over 5756242.74 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3274, pruned_loss=0.08495, over 5657164.30 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 22:33:29,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.822e+02 1.518e+03 2.259e+03 3.351e+03 1.501e+04, threshold=4.517e+03, percent-clipped=20.0
+2023-03-12 22:33:42,541 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 22:33:55,049 INFO [train.py:968] (0/2) Epoch 25, batch 15200, giga_loss[loss=0.2835, simple_loss=0.3532, pruned_loss=0.1069, over 28389.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3272, pruned_loss=0.0854, over 5663233.34 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3304, pruned_loss=0.08599, over 5759287.53 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3283, pruned_loss=0.08582, over 5652973.84 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:34:02,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1109046.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 22:34:04,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1109049.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 22:34:08,617 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 22:34:34,241 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1109078.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 22:34:48,357 INFO [train.py:968] (0/2) Epoch 25, batch 15250, giga_loss[loss=0.2537, simple_loss=0.3298, pruned_loss=0.08874, over 28846.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3277, pruned_loss=0.08566, over 5672719.70 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3303, pruned_loss=0.08591, over 5762196.18 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3286, pruned_loss=0.08607, over 5660319.75 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:35:24,949 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.364e+03 1.929e+03 2.542e+03 5.769e+03, threshold=3.858e+03, percent-clipped=1.0
+2023-03-12 22:35:49,184 INFO [train.py:968] (0/2) Epoch 25, batch 15300, giga_loss[loss=0.2168, simple_loss=0.3079, pruned_loss=0.06289, over 28719.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3258, pruned_loss=0.08391, over 5654967.86 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3307, pruned_loss=0.08619, over 5755000.10 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3261, pruned_loss=0.08396, over 5649045.19 frames. ], batch size: 243, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:36:50,815 INFO [train.py:968] (0/2) Epoch 25, batch 15350, giga_loss[loss=0.2659, simple_loss=0.3399, pruned_loss=0.09592, over 28145.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3244, pruned_loss=0.08284, over 5659265.65 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3303, pruned_loss=0.08606, over 5750244.59 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3248, pruned_loss=0.0829, over 5656235.38 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:37:27,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6073, 5.4549, 5.1516, 2.5924], device='cuda:0'), covar=tensor([0.0407, 0.0547, 0.0698, 0.1666], device='cuda:0'), in_proj_covar=tensor([0.1255, 0.1158, 0.0977, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 22:37:29,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.796e+02 1.319e+03 1.735e+03 2.437e+03 5.527e+03, threshold=3.471e+03, percent-clipped=7.0
+2023-03-12 22:37:49,470 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5510, 1.6765, 1.7887, 1.3472], device='cuda:0'), covar=tensor([0.1973, 0.2779, 0.1649, 0.1954], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0703, 0.0965, 0.0864], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 22:37:55,263 INFO [train.py:968] (0/2) Epoch 25, batch 15400, libri_loss[loss=0.2501, simple_loss=0.333, pruned_loss=0.08357, over 29087.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3235, pruned_loss=0.08267, over 5652656.56 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3303, pruned_loss=0.08604, over 5745403.47 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3237, pruned_loss=0.08266, over 5651217.05 frames. ], batch size: 101, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:38:08,328 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-12 22:38:55,114 INFO [train.py:968] (0/2) Epoch 25, batch 15450, giga_loss[loss=0.224, simple_loss=0.3051, pruned_loss=0.07141, over 28282.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3238, pruned_loss=0.08245, over 5653558.82 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.33, pruned_loss=0.08597, over 5748461.20 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3241, pruned_loss=0.08243, over 5647452.39 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:39:30,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.521e+03 1.893e+03 2.635e+03 7.501e+03, threshold=3.786e+03, percent-clipped=13.0
+2023-03-12 22:39:52,981 INFO [train.py:968] (0/2) Epoch 25, batch 15500, libri_loss[loss=0.2431, simple_loss=0.3298, pruned_loss=0.07821, over 29209.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3245, pruned_loss=0.08298, over 5655164.09 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.33, pruned_loss=0.08593, over 5744171.96 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3245, pruned_loss=0.0829, over 5650194.86 frames. ], batch size: 94, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:40:56,833 INFO [train.py:968] (0/2) Epoch 25, batch 15550, giga_loss[loss=0.2462, simple_loss=0.3255, pruned_loss=0.0834, over 28959.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3247, pruned_loss=0.084, over 5646423.80 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3298, pruned_loss=0.08583, over 5737550.22 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3248, pruned_loss=0.08402, over 5646528.34 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:41:33,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.969e+02 1.413e+03 1.794e+03 2.211e+03 7.872e+03, threshold=3.588e+03, percent-clipped=7.0
+2023-03-12 22:41:58,378 INFO [train.py:968] (0/2) Epoch 25, batch 15600, giga_loss[loss=0.2606, simple_loss=0.3505, pruned_loss=0.08542, over 28968.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3254, pruned_loss=0.08287, over 5658527.20 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.33, pruned_loss=0.0859, over 5737766.41 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3254, pruned_loss=0.0828, over 5657734.56 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:42:58,351 INFO [train.py:968] (0/2) Epoch 25, batch 15650, libri_loss[loss=0.2852, simple_loss=0.3561, pruned_loss=0.1072, over 29531.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3293, pruned_loss=0.08395, over 5663453.96 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3303, pruned_loss=0.08608, over 5739879.29 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.329, pruned_loss=0.08368, over 5659689.00 frames. ], batch size: 84, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:43:33,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.692e+02 1.593e+03 2.138e+03 3.249e+03 9.643e+03, threshold=4.276e+03, percent-clipped=20.0
+2023-03-12 22:43:57,190 INFO [train.py:968] (0/2) Epoch 25, batch 15700, giga_loss[loss=0.2693, simple_loss=0.3483, pruned_loss=0.0952, over 28893.00 frames. ], tot_loss[loss=0.25, simple_loss=0.331, pruned_loss=0.08449, over 5663524.46 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3301, pruned_loss=0.08602, over 5743060.96 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3309, pruned_loss=0.08429, over 5655360.24 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:44:15,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5337, 1.9161, 1.5487, 1.5782], device='cuda:0'), covar=tensor([0.2799, 0.2703, 0.3140, 0.2373], device='cuda:0'), in_proj_covar=tensor([0.1563, 0.1121, 0.1380, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 22:44:55,523 INFO [train.py:968] (0/2) Epoch 25, batch 15750, giga_loss[loss=0.2397, simple_loss=0.3271, pruned_loss=0.07617, over 28880.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3323, pruned_loss=0.08542, over 5672560.77 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.33, pruned_loss=0.08604, over 5746261.92 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3323, pruned_loss=0.08524, over 5661504.26 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:45:28,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.871e+02 1.524e+03 1.962e+03 2.798e+03 7.477e+03, threshold=3.923e+03, percent-clipped=5.0
+2023-03-12 22:45:50,473 INFO [train.py:968] (0/2) Epoch 25, batch 15800, giga_loss[loss=0.205, simple_loss=0.2781, pruned_loss=0.06601, over 24644.00 frames. ], tot_loss[loss=0.25, simple_loss=0.331, pruned_loss=0.08452, over 5686082.23 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3297, pruned_loss=0.0857, over 5750432.55 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3314, pruned_loss=0.08465, over 5671385.14 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:46:35,632 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2982, 1.5527, 1.5283, 1.1582], device='cuda:0'), covar=tensor([0.1802, 0.2556, 0.1516, 0.1813], device='cuda:0'), in_proj_covar=tensor([0.0918, 0.0702, 0.0966, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 22:46:50,576 INFO [train.py:968] (0/2) Epoch 25, batch 15850, giga_loss[loss=0.2284, simple_loss=0.3133, pruned_loss=0.0718, over 28819.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3269, pruned_loss=0.0818, over 5695591.29 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3291, pruned_loss=0.08537, over 5754239.71 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3277, pruned_loss=0.0821, over 5678643.21 frames. ], batch size: 66, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:47:22,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.204e+02 1.368e+03 1.873e+03 2.752e+03 9.333e+03, threshold=3.745e+03, percent-clipped=11.0
+2023-03-12 22:47:43,959 INFO [train.py:968] (0/2) Epoch 25, batch 15900, giga_loss[loss=0.2364, simple_loss=0.3149, pruned_loss=0.07889, over 28859.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3256, pruned_loss=0.08166, over 5682679.32 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3288, pruned_loss=0.08527, over 5749976.68 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3264, pruned_loss=0.08185, over 5671149.90 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:48:39,756 INFO [train.py:968] (0/2) Epoch 25, batch 15950, libri_loss[loss=0.2186, simple_loss=0.292, pruned_loss=0.07256, over 29476.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.325, pruned_loss=0.08222, over 5684881.46 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3283, pruned_loss=0.08513, over 5753275.13 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.326, pruned_loss=0.08239, over 5669940.02 frames. ], batch size: 70, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:49:15,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.501e+03 1.772e+03 2.498e+03 5.352e+03, threshold=3.544e+03, percent-clipped=5.0
+2023-03-12 22:49:19,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2991, 1.6866, 1.3341, 1.0504], device='cuda:0'), covar=tensor([0.2599, 0.2386, 0.2817, 0.2349], device='cuda:0'), in_proj_covar=tensor([0.1557, 0.1114, 0.1374, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 22:49:39,649 INFO [train.py:968] (0/2) Epoch 25, batch 16000, giga_loss[loss=0.2437, simple_loss=0.3177, pruned_loss=0.08486, over 28607.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3271, pruned_loss=0.08296, over 5682549.23 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3275, pruned_loss=0.08469, over 5757080.42 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3285, pruned_loss=0.08343, over 5665492.11 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 22:50:43,149 INFO [train.py:968] (0/2) Epoch 25, batch 16050, giga_loss[loss=0.2989, simple_loss=0.3567, pruned_loss=0.1205, over 26812.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3291, pruned_loss=0.0842, over 5689059.17 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3275, pruned_loss=0.08459, over 5759764.40 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3303, pruned_loss=0.08466, over 5671498.31 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:51:18,828 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.551e+03 1.980e+03 2.997e+03 1.285e+04, threshold=3.960e+03, percent-clipped=16.0
+2023-03-12 22:51:39,257 INFO [train.py:968] (0/2) Epoch 25, batch 16100, giga_loss[loss=0.2422, simple_loss=0.3334, pruned_loss=0.07555, over 29018.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3302, pruned_loss=0.0852, over 5675380.26 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.327, pruned_loss=0.08435, over 5751972.14 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3316, pruned_loss=0.08576, over 5666721.33 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:52:37,673 INFO [train.py:968] (0/2) Epoch 25, batch 16150, libri_loss[loss=0.2681, simple_loss=0.347, pruned_loss=0.09465, over 19800.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3333, pruned_loss=0.08606, over 5674646.57 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3271, pruned_loss=0.08436, over 5743706.40 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3346, pruned_loss=0.08656, over 5673664.05 frames. ], batch size: 187, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:52:45,960 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1110000.pt
+2023-03-12 22:53:09,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.881e+02 1.729e+03 2.418e+03 3.734e+03 1.179e+04, threshold=4.836e+03, percent-clipped=22.0
+2023-03-12 22:53:32,304 INFO [train.py:968] (0/2) Epoch 25, batch 16200, giga_loss[loss=0.2899, simple_loss=0.3588, pruned_loss=0.1105, over 26818.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3349, pruned_loss=0.08652, over 5666447.09 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3273, pruned_loss=0.08462, over 5735743.75 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3359, pruned_loss=0.08673, over 5671076.40 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:54:41,078 INFO [train.py:968] (0/2) Epoch 25, batch 16250, giga_loss[loss=0.2569, simple_loss=0.3345, pruned_loss=0.08959, over 28965.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3353, pruned_loss=0.08656, over 5675394.62 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3274, pruned_loss=0.08468, over 5737941.79 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3361, pruned_loss=0.0867, over 5676193.69 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:55:25,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.412e+03 1.809e+03 2.890e+03 8.085e+03, threshold=3.618e+03, percent-clipped=4.0
+2023-03-12 22:55:42,251 INFO [train.py:968] (0/2) Epoch 25, batch 16300, giga_loss[loss=0.2908, simple_loss=0.3429, pruned_loss=0.1193, over 26882.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3328, pruned_loss=0.0857, over 5687480.81 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.327, pruned_loss=0.08454, over 5740345.33 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3339, pruned_loss=0.08596, over 5685036.76 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:56:46,396 INFO [train.py:968] (0/2) Epoch 25, batch 16350, giga_loss[loss=0.282, simple_loss=0.347, pruned_loss=0.1085, over 27599.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3331, pruned_loss=0.08655, over 5669834.68 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3268, pruned_loss=0.08454, over 5734617.30 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3343, pruned_loss=0.08677, over 5672756.27 frames. ], batch size: 473, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:57:07,542 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-12 22:57:21,970 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.515e+03 2.023e+03 2.863e+03 6.724e+03, threshold=4.047e+03, percent-clipped=15.0
+2023-03-12 22:57:42,021 INFO [train.py:968] (0/2) Epoch 25, batch 16400, giga_loss[loss=0.2695, simple_loss=0.3381, pruned_loss=0.1004, over 27728.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3324, pruned_loss=0.08685, over 5668250.55 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3269, pruned_loss=0.08449, over 5737739.34 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3336, pruned_loss=0.08717, over 5665022.44 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:58:43,597 INFO [train.py:968] (0/2) Epoch 25, batch 16450, giga_loss[loss=0.2664, simple_loss=0.3438, pruned_loss=0.09452, over 29034.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3298, pruned_loss=0.08611, over 5675022.63 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3266, pruned_loss=0.08429, over 5739725.83 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.331, pruned_loss=0.08657, over 5669405.80 frames. ], batch size: 285, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:58:58,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6290, 4.4655, 4.2789, 2.0868], device='cuda:0'), covar=tensor([0.0631, 0.0818, 0.1042, 0.2002], device='cuda:0'), in_proj_covar=tensor([0.1253, 0.1150, 0.0974, 0.0725], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 22:59:15,765 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-12 22:59:19,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.197e+02 1.416e+03 1.950e+03 2.413e+03 9.475e+03, threshold=3.900e+03, percent-clipped=5.0
+2023-03-12 22:59:42,573 INFO [train.py:968] (0/2) Epoch 25, batch 16500, giga_loss[loss=0.2652, simple_loss=0.3405, pruned_loss=0.09498, over 26824.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3296, pruned_loss=0.08518, over 5672651.88 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3267, pruned_loss=0.08437, over 5733191.52 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3305, pruned_loss=0.0855, over 5672950.20 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:00:41,536 INFO [train.py:968] (0/2) Epoch 25, batch 16550, giga_loss[loss=0.2743, simple_loss=0.364, pruned_loss=0.09236, over 28660.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3283, pruned_loss=0.08369, over 5667506.54 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3262, pruned_loss=0.08413, over 5734025.93 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3295, pruned_loss=0.08418, over 5666048.72 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:01:17,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.181e+02 1.442e+03 1.812e+03 2.563e+03 7.351e+03, threshold=3.624e+03, percent-clipped=6.0
+2023-03-12 23:01:38,253 INFO [train.py:968] (0/2) Epoch 25, batch 16600, giga_loss[loss=0.2183, simple_loss=0.3137, pruned_loss=0.06146, over 28797.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3308, pruned_loss=0.08343, over 5672214.32 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3261, pruned_loss=0.08405, over 5735989.42 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3319, pruned_loss=0.08388, over 5668482.68 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:02:35,296 INFO [train.py:968] (0/2) Epoch 25, batch 16650, giga_loss[loss=0.2062, simple_loss=0.3015, pruned_loss=0.05547, over 28445.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3322, pruned_loss=0.0833, over 5673262.95 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3262, pruned_loss=0.08405, over 5737693.89 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3331, pruned_loss=0.08365, over 5668044.68 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:02:47,353 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-12 23:02:52,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5180, 4.3674, 4.1543, 2.1004], device='cuda:0'), covar=tensor([0.0590, 0.0721, 0.0810, 0.1970], device='cuda:0'), in_proj_covar=tensor([0.1244, 0.1145, 0.0967, 0.0723], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 23:03:09,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.991e+02 1.435e+03 1.770e+03 2.408e+03 7.692e+03, threshold=3.540e+03, percent-clipped=6.0
+2023-03-12 23:03:31,639 INFO [train.py:968] (0/2) Epoch 25, batch 16700, giga_loss[loss=0.2386, simple_loss=0.3066, pruned_loss=0.08533, over 24429.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.332, pruned_loss=0.08288, over 5684460.64 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3262, pruned_loss=0.08403, over 5742138.21 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3328, pruned_loss=0.08314, over 5674691.59 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:04:33,367 INFO [train.py:968] (0/2) Epoch 25, batch 16750, giga_loss[loss=0.2144, simple_loss=0.3039, pruned_loss=0.0624, over 29027.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3319, pruned_loss=0.08314, over 5675594.54 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3266, pruned_loss=0.08433, over 5740486.30 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3324, pruned_loss=0.08307, over 5667469.59 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:04:56,810 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3278, 1.7670, 1.6641, 1.4884], device='cuda:0'), covar=tensor([0.2209, 0.2035, 0.2241, 0.2273], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0729, 0.0702, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 23:05:16,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.414e+03 1.844e+03 2.586e+03 6.726e+03, threshold=3.689e+03, percent-clipped=12.0
+2023-03-12 23:05:37,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4387, 1.8508, 1.7270, 1.4622], device='cuda:0'), covar=tensor([0.2122, 0.2056, 0.2189, 0.2278], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0729, 0.0702, 0.0675], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:0')
+2023-03-12 23:05:41,674 INFO [train.py:968] (0/2) Epoch 25, batch 16800, giga_loss[loss=0.2485, simple_loss=0.3353, pruned_loss=0.08078, over 28763.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3313, pruned_loss=0.08244, over 5676700.70 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3268, pruned_loss=0.08449, over 5742178.47 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3316, pruned_loss=0.08222, over 5667887.48 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:06:27,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3501, 1.3123, 1.1454, 1.4904], device='cuda:0'), covar=tensor([0.0807, 0.0359, 0.0383, 0.0920], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-12 23:06:51,641 INFO [train.py:968] (0/2) Epoch 25, batch 16850, giga_loss[loss=0.2363, simple_loss=0.3269, pruned_loss=0.07288, over 29037.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3313, pruned_loss=0.08196, over 5675695.45 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3265, pruned_loss=0.08449, over 5737082.95 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3319, pruned_loss=0.08173, over 5671011.37 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:07:34,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.239e+02 1.345e+03 1.665e+03 2.302e+03 5.640e+03, threshold=3.330e+03, percent-clipped=6.0
+2023-03-12 23:07:38,030 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1110725.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:07:59,735 INFO [train.py:968] (0/2) Epoch 25, batch 16900, giga_loss[loss=0.3211, simple_loss=0.3951, pruned_loss=0.1235, over 28645.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3349, pruned_loss=0.0841, over 5673351.07 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3268, pruned_loss=0.08463, over 5731755.06 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3352, pruned_loss=0.08376, over 5673543.04 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:08:55,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-03-12 23:09:08,499 INFO [train.py:968] (0/2) Epoch 25, batch 16950, giga_loss[loss=0.2381, simple_loss=0.3237, pruned_loss=0.07623, over 28795.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3375, pruned_loss=0.08518, over 5680576.80 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3268, pruned_loss=0.08468, over 5734927.85 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3379, pruned_loss=0.08486, over 5676888.66 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:09:53,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.171e+02 1.374e+03 1.898e+03 2.560e+03 4.998e+03, threshold=3.796e+03, percent-clipped=12.0
+2023-03-12 23:10:09,132 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1110835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:10:16,051 INFO [train.py:968] (0/2) Epoch 25, batch 17000, giga_loss[loss=0.2652, simple_loss=0.3437, pruned_loss=0.09333, over 28123.00 frames. ], tot_loss[loss=0.254, simple_loss=0.337, pruned_loss=0.08549, over 5681210.74 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3269, pruned_loss=0.08471, over 5734867.25 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3374, pruned_loss=0.08523, over 5677804.26 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:10:30,365 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 23:11:21,416 INFO [train.py:968] (0/2) Epoch 25, batch 17050, giga_loss[loss=0.2241, simple_loss=0.315, pruned_loss=0.06659, over 28707.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3342, pruned_loss=0.08429, over 5683381.05 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3269, pruned_loss=0.0847, over 5731148.88 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3348, pruned_loss=0.08411, over 5682322.17 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:11:41,884 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 23:12:05,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.956e+02 1.407e+03 1.866e+03 2.509e+03 6.665e+03, threshold=3.733e+03, percent-clipped=5.0
+2023-03-12 23:12:23,834 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1110932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:12:31,100 INFO [train.py:968] (0/2) Epoch 25, batch 17100, giga_loss[loss=0.2282, simple_loss=0.3178, pruned_loss=0.06929, over 28943.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3324, pruned_loss=0.08242, over 5690995.68 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3269, pruned_loss=0.0847, over 5732536.19 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3329, pruned_loss=0.08224, over 5688037.41 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:13:29,552 INFO [train.py:968] (0/2) Epoch 25, batch 17150, libri_loss[loss=0.2301, simple_loss=0.2996, pruned_loss=0.0803, over 29356.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3318, pruned_loss=0.08273, over 5689520.94 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3271, pruned_loss=0.08489, over 5729581.02 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3323, pruned_loss=0.08229, over 5688330.53 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:14:06,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.776e+02 1.249e+03 1.650e+03 2.288e+03 6.096e+03, threshold=3.299e+03, percent-clipped=3.0
+2023-03-12 23:14:07,589 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1111024.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:14:25,369 INFO [train.py:968] (0/2) Epoch 25, batch 17200, giga_loss[loss=0.2778, simple_loss=0.3589, pruned_loss=0.09833, over 28109.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3322, pruned_loss=0.08275, over 5687789.35 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3267, pruned_loss=0.08465, over 5733906.93 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.333, pruned_loss=0.08257, over 5681921.92 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:14:41,276 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 23:15:20,768 INFO [train.py:968] (0/2) Epoch 25, batch 17250, giga_loss[loss=0.2896, simple_loss=0.3646, pruned_loss=0.1073, over 28698.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3344, pruned_loss=0.08445, over 5682162.84 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3266, pruned_loss=0.08464, over 5730873.50 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3354, pruned_loss=0.08428, over 5679286.33 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:15:33,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1111100.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:15:59,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.991e+02 1.386e+03 1.919e+03 2.495e+03 5.835e+03, threshold=3.839e+03, percent-clipped=11.0
+2023-03-12 23:16:14,768 INFO [train.py:968] (0/2) Epoch 25, batch 17300, giga_loss[loss=0.2933, simple_loss=0.3552, pruned_loss=0.1157, over 28667.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3341, pruned_loss=0.08492, over 5679515.39 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3268, pruned_loss=0.08472, over 5733161.57 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3348, pruned_loss=0.08472, over 5674162.43 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:16:50,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7023, 1.9316, 1.5960, 1.7559], device='cuda:0'), covar=tensor([0.2606, 0.2531, 0.2796, 0.2533], device='cuda:0'), in_proj_covar=tensor([0.1556, 0.1116, 0.1377, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 23:17:15,388 INFO [train.py:968] (0/2) Epoch 25, batch 17350, giga_loss[loss=0.2636, simple_loss=0.3477, pruned_loss=0.08978, over 28903.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3318, pruned_loss=0.08459, over 5679923.25 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3267, pruned_loss=0.08475, over 5737630.98 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3327, pruned_loss=0.08441, over 5670731.95 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:17:39,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1111210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:17:53,277 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.575e+02 1.471e+03 2.158e+03 2.931e+03 7.291e+03, threshold=4.315e+03, percent-clipped=8.0
+2023-03-12 23:18:11,573 INFO [train.py:968] (0/2) Epoch 25, batch 17400, giga_loss[loss=0.262, simple_loss=0.3225, pruned_loss=0.1008, over 24455.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3312, pruned_loss=0.08453, over 5688908.22 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3265, pruned_loss=0.08464, over 5741443.81 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3321, pruned_loss=0.08447, over 5676840.87 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:18:15,452 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1111243.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:18:19,529 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1111246.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:18:45,678 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1111275.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:19:05,279 INFO [train.py:968] (0/2) Epoch 25, batch 17450, libri_loss[loss=0.2491, simple_loss=0.3305, pruned_loss=0.08384, over 29556.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3374, pruned_loss=0.08872, over 5687152.81 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3259, pruned_loss=0.08433, over 5746956.31 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.339, pruned_loss=0.08902, over 5670444.66 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:19:07,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3442, 1.4637, 1.2969, 1.5726], device='cuda:0'), covar=tensor([0.0822, 0.0359, 0.0355, 0.0903], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-12 23:19:08,641 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-12 23:19:19,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1111307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:19:33,571 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.136e+02 1.371e+03 1.672e+03 2.204e+03 5.257e+03, threshold=3.345e+03, percent-clipped=1.0
+2023-03-12 23:19:48,533 INFO [train.py:968] (0/2) Epoch 25, batch 17500, giga_loss[loss=0.2979, simple_loss=0.3572, pruned_loss=0.1193, over 23693.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3466, pruned_loss=0.09369, over 5694807.10 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3257, pruned_loss=0.08422, over 5749575.36 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3483, pruned_loss=0.09412, over 5678663.04 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:19:57,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3474, 2.0256, 1.5315, 0.5879], device='cuda:0'), covar=tensor([0.6310, 0.3325, 0.4573, 0.6911], device='cuda:0'), in_proj_covar=tensor([0.1787, 0.1675, 0.1621, 0.1448], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 23:20:01,613 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1111353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:20:03,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1111356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:20:24,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4452, 1.5857, 1.2999, 1.1055], device='cuda:0'), covar=tensor([0.1044, 0.0585, 0.1087, 0.1152], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0443, 0.0518, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-12 23:20:30,205 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1111385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:20:33,343 INFO [train.py:968] (0/2) Epoch 25, batch 17550, giga_loss[loss=0.3008, simple_loss=0.3737, pruned_loss=0.114, over 28909.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3479, pruned_loss=0.09448, over 5697229.06 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3254, pruned_loss=0.08412, over 5751551.70 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3499, pruned_loss=0.09514, over 5681510.47 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:20:41,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1111399.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:21:02,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.029e+02 1.281e+03 1.680e+03 2.166e+03 4.836e+03, threshold=3.360e+03, percent-clipped=6.0
+2023-03-12 23:21:19,290 INFO [train.py:968] (0/2) Epoch 25, batch 17600, giga_loss[loss=0.2138, simple_loss=0.2966, pruned_loss=0.06546, over 28779.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.343, pruned_loss=0.09285, over 5700849.07 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3256, pruned_loss=0.08413, over 5755559.58 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3449, pruned_loss=0.09359, over 5683654.04 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:21:27,700 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1111450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:21:29,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1111453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:21:55,170 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1111482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:21:56,037 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 23:22:05,223 INFO [train.py:968] (0/2) Epoch 25, batch 17650, giga_loss[loss=0.2392, simple_loss=0.3133, pruned_loss=0.08258, over 28451.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3355, pruned_loss=0.0899, over 5691574.59 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3255, pruned_loss=0.08409, over 5757128.03 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3372, pruned_loss=0.09056, over 5676160.57 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:22:35,156 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.726e+02 1.128e+03 1.508e+03 1.861e+03 4.884e+03, threshold=3.016e+03, percent-clipped=3.0
+2023-03-12 23:22:45,721 INFO [train.py:968] (0/2) Epoch 25, batch 17700, giga_loss[loss=0.1982, simple_loss=0.2841, pruned_loss=0.05615, over 28953.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.328, pruned_loss=0.08645, over 5699239.88 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3252, pruned_loss=0.08389, over 5759794.41 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3298, pruned_loss=0.08725, over 5682987.98 frames. ], batch size: 164, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:22:48,042 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1111542.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:22:52,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1111545.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:23:19,939 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1111574.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:23:26,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7476, 1.9011, 1.6172, 1.8427], device='cuda:0'), covar=tensor([0.2627, 0.2807, 0.3116, 0.2702], device='cuda:0'), in_proj_covar=tensor([0.1555, 0.1117, 0.1375, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 23:23:32,875 INFO [train.py:968] (0/2) Epoch 25, batch 17750, giga_loss[loss=0.2088, simple_loss=0.2825, pruned_loss=0.06755, over 28675.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3208, pruned_loss=0.0837, over 5693470.64 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3253, pruned_loss=0.0839, over 5761200.24 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.322, pruned_loss=0.08432, over 5678896.34 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:24:02,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.541e+02 1.075e+03 1.295e+03 1.688e+03 6.944e+03, threshold=2.589e+03, percent-clipped=3.0
+2023-03-12 23:24:15,393 INFO [train.py:968] (0/2) Epoch 25, batch 17800, giga_loss[loss=0.2146, simple_loss=0.288, pruned_loss=0.07064, over 27745.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3155, pruned_loss=0.08137, over 5700811.68 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3255, pruned_loss=0.08395, over 5764048.96 frames. ], giga_tot_loss[loss=0.2398, simple_loss=0.3161, pruned_loss=0.08177, over 5685343.68 frames. ], batch size: 474, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:24:16,870 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1111642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:24:54,796 INFO [train.py:968] (0/2) Epoch 25, batch 17850, giga_loss[loss=0.2288, simple_loss=0.307, pruned_loss=0.07529, over 28864.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3137, pruned_loss=0.08079, over 5692879.90 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3262, pruned_loss=0.08419, over 5755750.92 frames. ], giga_tot_loss[loss=0.2374, simple_loss=0.3132, pruned_loss=0.08079, over 5686699.03 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:25:21,817 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.015e+02 1.160e+03 1.445e+03 1.847e+03 4.704e+03, threshold=2.889e+03, percent-clipped=9.0
+2023-03-12 23:25:34,149 INFO [train.py:968] (0/2) Epoch 25, batch 17900, giga_loss[loss=0.1931, simple_loss=0.2757, pruned_loss=0.05519, over 28486.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3104, pruned_loss=0.07927, over 5681586.28 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3267, pruned_loss=0.08459, over 5740697.45 frames. ], giga_tot_loss[loss=0.2334, simple_loss=0.3092, pruned_loss=0.0788, over 5689132.11 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:26:14,497 INFO [train.py:968] (0/2) Epoch 25, batch 17950, libri_loss[loss=0.2359, simple_loss=0.318, pruned_loss=0.07689, over 29328.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3084, pruned_loss=0.07838, over 5682035.83 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3273, pruned_loss=0.08473, over 5736449.79 frames. ], giga_tot_loss[loss=0.2308, simple_loss=0.3062, pruned_loss=0.07767, over 5690184.24 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:26:30,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 23:26:40,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.795e+02 1.053e+03 1.332e+03 1.847e+03 6.088e+03, threshold=2.665e+03, percent-clipped=12.0
+2023-03-12 23:26:54,195 INFO [train.py:968] (0/2) Epoch 25, batch 18000, giga_loss[loss=0.2119, simple_loss=0.2889, pruned_loss=0.06745, over 29040.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3046, pruned_loss=0.07625, over 5689042.40 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3273, pruned_loss=0.08474, over 5739527.49 frames. ], giga_tot_loss[loss=0.2266, simple_loss=0.3023, pruned_loss=0.07546, over 5691489.81 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:26:54,199 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-12 23:27:03,034 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2015, simple_loss=0.3074, pruned_loss=0.04783, over 944034.00 frames. 
+2023-03-12 23:27:03,036 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-12 23:27:08,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2714, 1.5093, 1.3437, 1.1374], device='cuda:0'), covar=tensor([0.3035, 0.2868, 0.1873, 0.2631], device='cuda:0'), in_proj_covar=tensor([0.1999, 0.1923, 0.1827, 0.1985], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 23:27:44,235 INFO [train.py:968] (0/2) Epoch 25, batch 18050, giga_loss[loss=0.1966, simple_loss=0.2761, pruned_loss=0.05851, over 28766.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3023, pruned_loss=0.07507, over 5688080.48 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3273, pruned_loss=0.08453, over 5742072.67 frames. ], giga_tot_loss[loss=0.2242, simple_loss=0.2997, pruned_loss=0.07433, over 5686392.98 frames. ], batch size: 243, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:27:46,554 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2752, 2.4761, 1.3275, 1.4467], device='cuda:0'), covar=tensor([0.1008, 0.0368, 0.0931, 0.1353], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0556, 0.0397, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 23:28:15,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.399e+02 1.063e+03 1.243e+03 1.687e+03 3.808e+03, threshold=2.485e+03, percent-clipped=2.0
+2023-03-12 23:28:30,114 INFO [train.py:968] (0/2) Epoch 25, batch 18100, giga_loss[loss=0.2044, simple_loss=0.281, pruned_loss=0.0639, over 28892.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3003, pruned_loss=0.0745, over 5681926.58 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3279, pruned_loss=0.08495, over 5734930.82 frames. ], giga_tot_loss[loss=0.2221, simple_loss=0.2974, pruned_loss=0.07346, over 5686308.98 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:28:30,532 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-12 23:28:42,314 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 23:28:51,794 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-12 23:29:09,561 INFO [train.py:968] (0/2) Epoch 25, batch 18150, giga_loss[loss=0.2125, simple_loss=0.2892, pruned_loss=0.06791, over 28619.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2978, pruned_loss=0.07344, over 5686486.86 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3278, pruned_loss=0.08471, over 5736475.21 frames. ], giga_tot_loss[loss=0.2199, simple_loss=0.2948, pruned_loss=0.07252, over 5687328.21 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:29:09,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6332, 2.2773, 1.8947, 0.7610], device='cuda:0'), covar=tensor([0.6706, 0.3410, 0.3960, 0.7469], device='cuda:0'), in_proj_covar=tensor([0.1788, 0.1684, 0.1620, 0.1448], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 23:29:19,673 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1112000.pt
+2023-03-12 23:29:23,666 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:29:35,962 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1112017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:29:42,575 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.996e+02 1.113e+03 1.517e+03 1.831e+03 7.018e+03, threshold=3.035e+03, percent-clipped=10.0
+2023-03-12 23:29:55,048 INFO [train.py:968] (0/2) Epoch 25, batch 18200, giga_loss[loss=0.1762, simple_loss=0.2555, pruned_loss=0.04848, over 29085.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2953, pruned_loss=0.07241, over 5678696.28 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3281, pruned_loss=0.08466, over 5740813.45 frames. ], giga_tot_loss[loss=0.2174, simple_loss=0.2919, pruned_loss=0.07144, over 5674065.10 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:30:31,830 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-12 23:30:37,278 INFO [train.py:968] (0/2) Epoch 25, batch 18250, giga_loss[loss=0.2707, simple_loss=0.3372, pruned_loss=0.1021, over 27679.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2955, pruned_loss=0.07297, over 5679440.24 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3282, pruned_loss=0.08464, over 5743216.72 frames. ], giga_tot_loss[loss=0.2179, simple_loss=0.2919, pruned_loss=0.07192, over 5672361.10 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:31:14,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.574e+02 1.289e+03 1.650e+03 2.061e+03 6.916e+03, threshold=3.300e+03, percent-clipped=8.0
+2023-03-12 23:31:26,725 INFO [train.py:968] (0/2) Epoch 25, batch 18300, giga_loss[loss=0.2946, simple_loss=0.3646, pruned_loss=0.1123, over 28736.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3081, pruned_loss=0.07991, over 5677286.88 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3286, pruned_loss=0.08482, over 5741900.51 frames. ], giga_tot_loss[loss=0.2311, simple_loss=0.3045, pruned_loss=0.07881, over 5671967.28 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:31:46,768 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1112160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:31:50,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1112163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:31:53,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4067, 2.0293, 1.5330, 0.6701], device='cuda:0'), covar=tensor([0.6869, 0.3518, 0.4506, 0.7436], device='cuda:0'), in_proj_covar=tensor([0.1788, 0.1682, 0.1619, 0.1448], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-12 23:32:10,091 INFO [train.py:968] (0/2) Epoch 25, batch 18350, giga_loss[loss=0.2804, simple_loss=0.3572, pruned_loss=0.1018, over 28788.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3215, pruned_loss=0.08683, over 5686321.82 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3286, pruned_loss=0.0847, over 5744949.12 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3186, pruned_loss=0.08606, over 5678126.10 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:32:11,938 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1112192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:32:37,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.403e+02 1.526e+03 1.805e+03 2.573e+03 6.686e+03, threshold=3.611e+03, percent-clipped=11.0
+2023-03-12 23:32:50,111 INFO [train.py:968] (0/2) Epoch 25, batch 18400, giga_loss[loss=0.2372, simple_loss=0.3239, pruned_loss=0.07525, over 29011.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3299, pruned_loss=0.09037, over 5689993.63 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3285, pruned_loss=0.08463, over 5748663.65 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3275, pruned_loss=0.08992, over 5678995.50 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:32:53,980 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 23:33:24,409 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:33:29,326 INFO [train.py:968] (0/2) Epoch 25, batch 18450, giga_loss[loss=0.2185, simple_loss=0.3077, pruned_loss=0.06464, over 28619.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3349, pruned_loss=0.09139, over 5688084.92 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3288, pruned_loss=0.08462, over 5748107.27 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3329, pruned_loss=0.09121, over 5678261.16 frames. ], batch size: 60, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:33:58,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.729e+02 1.282e+03 1.603e+03 2.118e+03 7.518e+03, threshold=3.207e+03, percent-clipped=3.0
+2023-03-12 23:34:06,841 INFO [train.py:968] (0/2) Epoch 25, batch 18500, giga_loss[loss=0.26, simple_loss=0.3482, pruned_loss=0.08586, over 28917.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3367, pruned_loss=0.09091, over 5697636.10 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3292, pruned_loss=0.08481, over 5752653.96 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.335, pruned_loss=0.0908, over 5684119.22 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:34:19,273 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:34:32,783 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5929, 1.5754, 1.8262, 1.4332], device='cuda:0'), covar=tensor([0.1544, 0.2131, 0.1319, 0.1629], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0705, 0.0973, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 23:34:46,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1112379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:34:56,223 INFO [train.py:968] (0/2) Epoch 25, batch 18550, giga_loss[loss=0.3516, simple_loss=0.3966, pruned_loss=0.1533, over 26724.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3398, pruned_loss=0.0927, over 5677242.14 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3291, pruned_loss=0.08478, over 5753240.69 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3385, pruned_loss=0.09267, over 5665933.51 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:35:28,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.965e+02 1.180e+03 1.575e+03 2.096e+03 3.811e+03, threshold=3.150e+03, percent-clipped=4.0
+2023-03-12 23:35:37,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5011, 1.7365, 1.4082, 1.5241], device='cuda:0'), covar=tensor([0.2681, 0.2639, 0.2885, 0.2393], device='cuda:0'), in_proj_covar=tensor([0.1560, 0.1123, 0.1378, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 23:35:38,850 INFO [train.py:968] (0/2) Epoch 25, batch 18600, libri_loss[loss=0.2364, simple_loss=0.3311, pruned_loss=0.07087, over 29749.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.341, pruned_loss=0.09378, over 5674328.09 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3291, pruned_loss=0.08467, over 5748446.85 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3403, pruned_loss=0.09414, over 5667578.65 frames. ], batch size: 87, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:36:12,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7621, 2.0277, 1.9290, 1.5197], device='cuda:0'), covar=tensor([0.3271, 0.2545, 0.2699, 0.3131], device='cuda:0'), in_proj_covar=tensor([0.2000, 0.1929, 0.1832, 0.1991], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-12 23:36:22,938 INFO [train.py:968] (0/2) Epoch 25, batch 18650, giga_loss[loss=0.3074, simple_loss=0.372, pruned_loss=0.1214, over 28636.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3449, pruned_loss=0.09655, over 5678198.37 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3291, pruned_loss=0.08459, over 5749766.20 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3445, pruned_loss=0.09701, over 5671017.69 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:36:50,145 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1112522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:36:52,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1112525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:36:53,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.833e+02 1.347e+03 1.737e+03 2.265e+03 5.248e+03, threshold=3.474e+03, percent-clipped=11.0
+2023-03-12 23:37:04,519 INFO [train.py:968] (0/2) Epoch 25, batch 18700, giga_loss[loss=0.3086, simple_loss=0.3827, pruned_loss=0.1172, over 28890.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.348, pruned_loss=0.09789, over 5683278.40 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3296, pruned_loss=0.08462, over 5752860.64 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3477, pruned_loss=0.09856, over 5673030.19 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:37:17,500 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1112554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:37:37,736 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-12 23:37:39,506 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-12 23:37:43,087 INFO [train.py:968] (0/2) Epoch 25, batch 18750, giga_loss[loss=0.2552, simple_loss=0.3476, pruned_loss=0.08143, over 28906.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3506, pruned_loss=0.09859, over 5684327.19 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3299, pruned_loss=0.08478, over 5748205.64 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3507, pruned_loss=0.09936, over 5678848.14 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:37:53,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5501, 1.8196, 1.4558, 1.6679], device='cuda:0'), covar=tensor([0.2794, 0.2914, 0.3175, 0.2492], device='cuda:0'), in_proj_covar=tensor([0.1557, 0.1122, 0.1376, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 23:38:13,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.173e+02 1.233e+03 1.457e+03 1.926e+03 3.465e+03, threshold=2.915e+03, percent-clipped=0.0
+2023-03-12 23:38:25,994 INFO [train.py:968] (0/2) Epoch 25, batch 18800, giga_loss[loss=0.2789, simple_loss=0.3679, pruned_loss=0.09495, over 28481.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3526, pruned_loss=0.09901, over 5676788.53 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3301, pruned_loss=0.08487, over 5740094.80 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3526, pruned_loss=0.09963, over 5679041.21 frames. ], batch size: 65, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:38:39,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1112659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:38:42,559 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 23:39:05,624 INFO [train.py:968] (0/2) Epoch 25, batch 18850, giga_loss[loss=0.304, simple_loss=0.3733, pruned_loss=0.1173, over 28660.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3538, pruned_loss=0.09902, over 5680094.12 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3308, pruned_loss=0.08521, over 5739980.68 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3538, pruned_loss=0.09959, over 5680405.53 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:39:35,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.846e+02 1.191e+03 1.490e+03 1.885e+03 4.886e+03, threshold=2.981e+03, percent-clipped=12.0
+2023-03-12 23:39:35,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1112727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:39:45,870 INFO [train.py:968] (0/2) Epoch 25, batch 18900, giga_loss[loss=0.2786, simple_loss=0.3553, pruned_loss=0.101, over 28921.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3527, pruned_loss=0.09692, over 5699572.18 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3307, pruned_loss=0.08506, over 5744037.93 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3533, pruned_loss=0.09782, over 5695010.33 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:40:11,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:40:23,380 INFO [train.py:968] (0/2) Epoch 25, batch 18950, giga_loss[loss=0.3003, simple_loss=0.3694, pruned_loss=0.1156, over 28785.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3497, pruned_loss=0.09414, over 5705862.39 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3306, pruned_loss=0.08498, over 5746350.47 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3506, pruned_loss=0.09505, over 5699690.31 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:40:35,117 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1112802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:40:37,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1112805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:40:39,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-12 23:40:54,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.187e+02 1.226e+03 1.514e+03 1.936e+03 4.059e+03, threshold=3.027e+03, percent-clipped=6.0
+2023-03-12 23:41:00,415 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1112834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:41:02,445 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:41:05,803 INFO [train.py:968] (0/2) Epoch 25, batch 19000, giga_loss[loss=0.3096, simple_loss=0.3785, pruned_loss=0.1203, over 28901.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3503, pruned_loss=0.09498, over 5709086.83 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3305, pruned_loss=0.08488, over 5748076.01 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3513, pruned_loss=0.09588, over 5702287.58 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:41:30,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:41:31,337 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-12 23:41:34,674 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1112870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:41:37,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1112873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:41:48,172 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3456, 1.3672, 1.2755, 1.5578], device='cuda:0'), covar=tensor([0.0801, 0.0363, 0.0343, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-12 23:41:53,424 INFO [train.py:968] (0/2) Epoch 25, batch 19050, giga_loss[loss=0.2557, simple_loss=0.3393, pruned_loss=0.08609, over 28492.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3529, pruned_loss=0.09927, over 5712309.11 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3305, pruned_loss=0.08486, over 5748877.63 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3537, pruned_loss=0.1001, over 5706041.27 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:42:06,588 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1112902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:42:12,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2655, 2.8394, 1.3980, 1.3962], device='cuda:0'), covar=tensor([0.1024, 0.0341, 0.0877, 0.1392], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0555, 0.0397, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-12 23:42:25,818 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:42:26,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.789e+02 1.407e+03 1.753e+03 2.309e+03 4.841e+03, threshold=3.507e+03, percent-clipped=11.0
+2023-03-12 23:42:37,195 INFO [train.py:968] (0/2) Epoch 25, batch 19100, giga_loss[loss=0.2756, simple_loss=0.3467, pruned_loss=0.1022, over 28346.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3547, pruned_loss=0.1027, over 5702042.68 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.331, pruned_loss=0.08509, over 5743064.89 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3555, pruned_loss=0.1035, over 5701190.41 frames. ], batch size: 65, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:42:40,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4637, 1.5967, 1.7162, 1.3125], device='cuda:0'), covar=tensor([0.1705, 0.2694, 0.1422, 0.1764], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0705, 0.0970, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-12 23:43:17,412 INFO [train.py:968] (0/2) Epoch 25, batch 19150, giga_loss[loss=0.2399, simple_loss=0.3198, pruned_loss=0.08001, over 28977.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3531, pruned_loss=0.1028, over 5701704.92 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3312, pruned_loss=0.08516, over 5745950.56 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3541, pruned_loss=0.1038, over 5697715.29 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:43:48,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.023e+02 1.338e+03 1.675e+03 2.483e+03 5.661e+03, threshold=3.350e+03, percent-clipped=8.0
+2023-03-12 23:43:58,899 INFO [train.py:968] (0/2) Epoch 25, batch 19200, giga_loss[loss=0.3436, simple_loss=0.3909, pruned_loss=0.1481, over 28688.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3507, pruned_loss=0.102, over 5696213.69 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3314, pruned_loss=0.08519, over 5743877.68 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3515, pruned_loss=0.103, over 5694152.09 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:44:12,719 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-12 23:44:24,026 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-12 23:44:40,381 INFO [train.py:968] (0/2) Epoch 25, batch 19250, giga_loss[loss=0.2423, simple_loss=0.3257, pruned_loss=0.07951, over 29071.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3495, pruned_loss=0.1005, over 5702234.87 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3313, pruned_loss=0.08509, over 5737259.68 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3507, pruned_loss=0.1018, over 5705500.48 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:45:09,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.597e+02 1.315e+03 1.560e+03 2.206e+03 6.484e+03, threshold=3.120e+03, percent-clipped=9.0
+2023-03-12 23:45:17,236 INFO [train.py:968] (0/2) Epoch 25, batch 19300, giga_loss[loss=0.2487, simple_loss=0.3317, pruned_loss=0.08286, over 28982.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3493, pruned_loss=0.09982, over 5708905.77 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.332, pruned_loss=0.08532, over 5740417.63 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3503, pruned_loss=0.1012, over 5707617.83 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:45:23,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:46:00,836 INFO [train.py:968] (0/2) Epoch 25, batch 19350, giga_loss[loss=0.2316, simple_loss=0.3149, pruned_loss=0.07412, over 28834.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3463, pruned_loss=0.09782, over 5680836.73 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3324, pruned_loss=0.08549, over 5726432.57 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3472, pruned_loss=0.09912, over 5690137.96 frames. ], batch size: 243, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:46:22,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:46:35,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.416e+02 1.177e+03 1.500e+03 2.100e+03 8.063e+03, threshold=3.000e+03, percent-clipped=13.0
+2023-03-12 23:46:43,072 INFO [train.py:968] (0/2) Epoch 25, batch 19400, giga_loss[loss=0.2276, simple_loss=0.3037, pruned_loss=0.0757, over 28569.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3413, pruned_loss=0.09487, over 5687249.15 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3328, pruned_loss=0.08558, over 5729919.76 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.342, pruned_loss=0.09624, over 5690055.54 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:46:45,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:47:22,783 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113286.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:47:25,264 INFO [train.py:968] (0/2) Epoch 25, batch 19450, giga_loss[loss=0.2445, simple_loss=0.3276, pruned_loss=0.08072, over 28896.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3361, pruned_loss=0.09261, over 5672860.45 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.333, pruned_loss=0.08567, over 5724372.34 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3367, pruned_loss=0.09381, over 5678286.62 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:47:26,423 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1113291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:47:30,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1113294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:47:37,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:47:59,030 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1113323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:03,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.960e+02 1.079e+03 1.416e+03 1.796e+03 4.469e+03, threshold=2.831e+03, percent-clipped=4.0
+2023-03-12 23:48:15,407 INFO [train.py:968] (0/2) Epoch 25, batch 19500, giga_loss[loss=0.2332, simple_loss=0.3144, pruned_loss=0.07601, over 28846.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3315, pruned_loss=0.09099, over 5645194.51 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3331, pruned_loss=0.08575, over 5717647.14 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3318, pruned_loss=0.09194, over 5654534.36 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:48:28,209 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1113355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:31,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1113358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:55,163 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1113385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:56,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1113387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:57,963 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1113388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:58,904 INFO [train.py:968] (0/2) Epoch 25, batch 19550, giga_loss[loss=0.2663, simple_loss=0.3403, pruned_loss=0.09617, over 28723.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3327, pruned_loss=0.09147, over 5640442.09 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3335, pruned_loss=0.08589, over 5709379.64 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3326, pruned_loss=0.09216, over 5654062.69 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:49:21,871 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1113417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:49:31,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.297e+02 1.135e+03 1.353e+03 1.710e+03 4.374e+03, threshold=2.706e+03, percent-clipped=6.0
+2023-03-12 23:49:43,034 INFO [train.py:968] (0/2) Epoch 25, batch 19600, giga_loss[loss=0.2708, simple_loss=0.3348, pruned_loss=0.1034, over 28158.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.332, pruned_loss=0.09076, over 5647091.87 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3333, pruned_loss=0.08568, over 5710839.78 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3321, pruned_loss=0.09155, over 5655702.82 frames. ], batch size: 77, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:49:46,798 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1113444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:49:48,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1113447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:50:11,591 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1113476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:50:17,265 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:50:22,284 INFO [train.py:968] (0/2) Epoch 25, batch 19650, giga_loss[loss=0.2226, simple_loss=0.2998, pruned_loss=0.0727, over 28717.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3311, pruned_loss=0.08991, over 5657539.66 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3338, pruned_loss=0.08582, over 5703281.65 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3306, pruned_loss=0.0905, over 5669258.71 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:50:54,435 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.349e+02 1.169e+03 1.450e+03 1.979e+03 7.810e+03, threshold=2.900e+03, percent-clipped=15.0
+2023-03-12 23:51:03,935 INFO [train.py:968] (0/2) Epoch 25, batch 19700, giga_loss[loss=0.2351, simple_loss=0.3075, pruned_loss=0.08131, over 28086.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.329, pruned_loss=0.08894, over 5657398.21 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3338, pruned_loss=0.08577, over 5696124.15 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3286, pruned_loss=0.0895, over 5672609.91 frames. ], batch size: 77, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:51:26,217 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-12 23:51:43,794 INFO [train.py:968] (0/2) Epoch 25, batch 19750, giga_loss[loss=0.2439, simple_loss=0.3211, pruned_loss=0.08332, over 29010.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3258, pruned_loss=0.08698, over 5675306.89 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3336, pruned_loss=0.08549, over 5698964.95 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3257, pruned_loss=0.08772, over 5684284.10 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:52:01,536 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:52:04,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6056, 1.7637, 1.4650, 1.6052], device='cuda:0'), covar=tensor([0.2622, 0.2843, 0.3010, 0.2555], device='cuda:0'), in_proj_covar=tensor([0.1557, 0.1123, 0.1375, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 23:52:14,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.677e+02 1.022e+03 1.265e+03 1.686e+03 5.913e+03, threshold=2.531e+03, percent-clipped=4.0
+2023-03-12 23:52:22,809 INFO [train.py:968] (0/2) Epoch 25, batch 19800, giga_loss[loss=0.2276, simple_loss=0.3054, pruned_loss=0.07487, over 28251.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3249, pruned_loss=0.08624, over 5685060.79 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3344, pruned_loss=0.08573, over 5704849.69 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3238, pruned_loss=0.08667, over 5686504.20 frames. ], batch size: 77, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:52:39,259 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:52:40,559 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113661.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:53:03,289 INFO [train.py:968] (0/2) Epoch 25, batch 19850, giga_loss[loss=0.2132, simple_loss=0.294, pruned_loss=0.06616, over 28858.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3229, pruned_loss=0.08555, over 5698433.96 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3344, pruned_loss=0.08564, over 5709319.49 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3217, pruned_loss=0.08597, over 5695298.32 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:53:20,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3418, 1.3854, 3.8437, 3.1859], device='cuda:0'), covar=tensor([0.1708, 0.2743, 0.0446, 0.0983], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0661, 0.0973, 0.0936], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-12 23:53:21,600 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 23:53:33,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.509e+02 1.028e+03 1.204e+03 1.502e+03 5.458e+03, threshold=2.409e+03, percent-clipped=7.0
+2023-03-12 23:53:41,909 INFO [train.py:968] (0/2) Epoch 25, batch 19900, giga_loss[loss=0.2059, simple_loss=0.2881, pruned_loss=0.06187, over 28635.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.32, pruned_loss=0.08408, over 5707366.77 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3348, pruned_loss=0.08574, over 5709866.89 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3184, pruned_loss=0.08431, over 5704453.00 frames. ], batch size: 60, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:54:08,722 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-12 23:54:23,303 INFO [train.py:968] (0/2) Epoch 25, batch 19950, giga_loss[loss=0.2237, simple_loss=0.2997, pruned_loss=0.07388, over 28972.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3188, pruned_loss=0.08374, over 5714282.82 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3351, pruned_loss=0.08571, over 5713838.48 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.317, pruned_loss=0.0839, over 5708346.30 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:54:33,317 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1113804.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:54:36,737 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1113807.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:54:53,608 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.684e+02 1.120e+03 1.377e+03 1.839e+03 3.897e+03, threshold=2.755e+03, percent-clipped=13.0
+2023-03-12 23:54:57,844 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1113836.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:55:00,660 INFO [train.py:968] (0/2) Epoch 25, batch 20000, giga_loss[loss=0.2164, simple_loss=0.2899, pruned_loss=0.07144, over 28555.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3176, pruned_loss=0.08293, over 5719362.27 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3356, pruned_loss=0.08566, over 5719733.92 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3152, pruned_loss=0.08305, over 5709352.21 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:55:08,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5864, 1.6272, 1.4047, 1.6968], device='cuda:0'), covar=tensor([0.0775, 0.0338, 0.0341, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-12 23:55:15,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:55:20,351 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6732, 1.7704, 1.5115, 1.6358], device='cuda:0'), covar=tensor([0.2895, 0.3016, 0.3305, 0.2664], device='cuda:0'), in_proj_covar=tensor([0.1559, 0.1126, 0.1375, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-12 23:55:27,013 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:55:40,682 INFO [train.py:968] (0/2) Epoch 25, batch 20050, giga_loss[loss=0.2725, simple_loss=0.333, pruned_loss=0.1061, over 28943.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3172, pruned_loss=0.08283, over 5722457.03 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.336, pruned_loss=0.08576, over 5722232.07 frames. ], giga_tot_loss[loss=0.2401, simple_loss=0.3147, pruned_loss=0.08278, over 5712139.93 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:56:11,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.028e+02 1.100e+03 1.311e+03 1.631e+03 4.424e+03, threshold=2.621e+03, percent-clipped=5.0
+2023-03-12 23:56:20,980 INFO [train.py:968] (0/2) Epoch 25, batch 20100, libri_loss[loss=0.2576, simple_loss=0.3557, pruned_loss=0.07973, over 29681.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3192, pruned_loss=0.08391, over 5708815.26 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3361, pruned_loss=0.08579, over 5709371.44 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.3166, pruned_loss=0.0838, over 5710832.89 frames. ], batch size: 88, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:56:57,640 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113982.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:57:05,299 INFO [train.py:968] (0/2) Epoch 25, batch 20150, giga_loss[loss=0.3007, simple_loss=0.3661, pruned_loss=0.1176, over 29018.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3247, pruned_loss=0.08756, over 5711783.37 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3363, pruned_loss=0.08581, over 5711340.75 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3224, pruned_loss=0.08744, over 5711800.38 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:57:05,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:57:13,812 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1114000.pt
+2023-03-12 23:57:16,086 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:57:18,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:57:45,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.089e+02 1.323e+03 1.590e+03 2.051e+03 4.777e+03, threshold=3.179e+03, percent-clipped=13.0
+2023-03-12 23:57:46,237 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:57:47,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:57:53,177 INFO [train.py:968] (0/2) Epoch 25, batch 20200, libri_loss[loss=0.2983, simple_loss=0.3824, pruned_loss=0.1071, over 29226.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3306, pruned_loss=0.09143, over 5699112.93 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3366, pruned_loss=0.08588, over 5715792.79 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3283, pruned_loss=0.09136, over 5694851.34 frames. ], batch size: 97, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:57:53,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3965, 3.2990, 1.5561, 1.4287], device='cuda:0'), covar=tensor([0.1010, 0.0313, 0.0895, 0.1406], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0554, 0.0398, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0026, 0.0030], device='cuda:0')
+2023-03-12 23:58:17,895 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3224, 3.1687, 3.0127, 1.5431], device='cuda:0'), covar=tensor([0.0929, 0.1027, 0.0905, 0.2176], device='cuda:0'), in_proj_covar=tensor([0.1243, 0.1144, 0.0967, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-12 23:58:37,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:58:43,190 INFO [train.py:968] (0/2) Epoch 25, batch 20250, giga_loss[loss=0.2359, simple_loss=0.3267, pruned_loss=0.07252, over 28963.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3379, pruned_loss=0.09595, over 5699019.29 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3365, pruned_loss=0.08574, over 5718435.53 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3362, pruned_loss=0.09617, over 5692881.09 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:59:23,847 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.894e+02 1.302e+03 1.874e+03 2.821e+03 8.772e+03, threshold=3.748e+03, percent-clipped=22.0
+2023-03-12 23:59:25,462 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:59:27,241 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:59:30,231 INFO [train.py:968] (0/2) Epoch 25, batch 20300, giga_loss[loss=0.287, simple_loss=0.3638, pruned_loss=0.1051, over 28917.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3442, pruned_loss=0.09928, over 5689842.19 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3365, pruned_loss=0.08584, over 5719408.35 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3429, pruned_loss=0.09942, over 5683946.25 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:59:46,296 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-12 23:59:52,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:00:02,338 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:00:04,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:00:13,517 INFO [train.py:968] (0/2) Epoch 25, batch 20350, giga_loss[loss=0.2724, simple_loss=0.351, pruned_loss=0.09694, over 28757.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3483, pruned_loss=0.1004, over 5690957.35 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3367, pruned_loss=0.08592, over 5721138.74 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3473, pruned_loss=0.1009, over 5684029.09 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:00:21,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3899, 2.1115, 1.5021, 0.6107], device='cuda:0'), covar=tensor([0.5413, 0.3327, 0.4783, 0.6560], device='cuda:0'), in_proj_covar=tensor([0.1780, 0.1675, 0.1616, 0.1444], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 00:00:28,691 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:00:47,771 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:00:48,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.548e+02 1.227e+03 1.502e+03 2.184e+03 6.973e+03, threshold=3.004e+03, percent-clipped=5.0
+2023-03-13 00:00:55,154 INFO [train.py:968] (0/2) Epoch 25, batch 20400, giga_loss[loss=0.2865, simple_loss=0.364, pruned_loss=0.1046, over 28369.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3516, pruned_loss=0.1018, over 5700196.17 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.336, pruned_loss=0.08549, over 5728227.22 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.352, pruned_loss=0.1032, over 5686976.23 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:01:00,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:01:03,754 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:01:37,147 INFO [train.py:968] (0/2) Epoch 25, batch 20450, libri_loss[loss=0.2322, simple_loss=0.3083, pruned_loss=0.07803, over 28082.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3506, pruned_loss=0.1006, over 5697097.19 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3361, pruned_loss=0.08564, over 5731970.93 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3513, pruned_loss=0.1021, over 5682243.69 frames. ], batch size: 62, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:02:13,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.117e+02 1.331e+03 1.741e+03 2.244e+03 5.156e+03, threshold=3.482e+03, percent-clipped=10.0
+2023-03-13 00:02:19,372 INFO [train.py:968] (0/2) Epoch 25, batch 20500, giga_loss[loss=0.2558, simple_loss=0.3328, pruned_loss=0.08934, over 28766.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3464, pruned_loss=0.0971, over 5705170.36 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3363, pruned_loss=0.08571, over 5733854.12 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3469, pruned_loss=0.09828, over 5691533.78 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:02:20,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6606, 1.9241, 1.3362, 1.4794], device='cuda:0'), covar=tensor([0.1096, 0.0581, 0.1024, 0.1130], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0447, 0.0524, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 00:02:33,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114357.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:02:48,048 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:03:01,032 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:03:02,227 INFO [train.py:968] (0/2) Epoch 25, batch 20550, libri_loss[loss=0.2319, simple_loss=0.3115, pruned_loss=0.07613, over 28133.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3454, pruned_loss=0.09609, over 5698337.67 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.336, pruned_loss=0.0856, over 5734237.95 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3462, pruned_loss=0.09722, over 5687096.57 frames. ], batch size: 62, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:03:03,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:03:27,795 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:03:38,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.623e+02 1.223e+03 1.596e+03 2.178e+03 5.588e+03, threshold=3.193e+03, percent-clipped=3.0
+2023-03-13 00:03:44,074 INFO [train.py:968] (0/2) Epoch 25, batch 20600, giga_loss[loss=0.2829, simple_loss=0.3576, pruned_loss=0.1041, over 28839.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3452, pruned_loss=0.09523, over 5707497.51 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3367, pruned_loss=0.08604, over 5737559.00 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3455, pruned_loss=0.09612, over 5693587.88 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:04:00,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:04:26,725 INFO [train.py:968] (0/2) Epoch 25, batch 20650, giga_loss[loss=0.2971, simple_loss=0.3621, pruned_loss=0.116, over 29033.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3476, pruned_loss=0.09684, over 5701436.45 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.337, pruned_loss=0.08622, over 5737227.25 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3477, pruned_loss=0.09746, over 5690433.57 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:04:37,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114500.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:04:38,732 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-13 00:04:39,116 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114503.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:04:51,069 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3334, 1.3684, 1.3059, 1.5099], device='cuda:0'), covar=tensor([0.0798, 0.0360, 0.0333, 0.0867], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 00:05:03,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.492e+03 1.719e+03 2.052e+03 3.690e+03, threshold=3.438e+03, percent-clipped=2.0
+2023-03-13 00:05:03,435 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114532.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:05:10,563 INFO [train.py:968] (0/2) Epoch 25, batch 20700, giga_loss[loss=0.2796, simple_loss=0.3553, pruned_loss=0.102, over 28597.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3501, pruned_loss=0.09901, over 5697147.83 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3371, pruned_loss=0.08617, over 5740528.90 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3503, pruned_loss=0.09979, over 5684667.35 frames. ], batch size: 60, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:05:49,023 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2897, 1.4676, 1.3453, 1.2403], device='cuda:0'), covar=tensor([0.2449, 0.2350, 0.2015, 0.2296], device='cuda:0'), in_proj_covar=tensor([0.2011, 0.1938, 0.1849, 0.2010], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 00:05:55,751 INFO [train.py:968] (0/2) Epoch 25, batch 20750, giga_loss[loss=0.2629, simple_loss=0.341, pruned_loss=0.09241, over 29032.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3509, pruned_loss=0.09967, over 5711323.56 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3374, pruned_loss=0.08627, over 5743841.19 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3511, pruned_loss=0.1005, over 5697639.23 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:05:56,718 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:05,010 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:08,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:08,678 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:24,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:28,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:31,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.902e+02 1.310e+03 1.774e+03 2.298e+03 4.850e+03, threshold=3.548e+03, percent-clipped=7.0
+2023-03-13 00:06:32,903 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:36,407 INFO [train.py:968] (0/2) Epoch 25, batch 20800, giga_loss[loss=0.2751, simple_loss=0.3488, pruned_loss=0.1006, over 28782.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3525, pruned_loss=0.1007, over 5718585.97 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3379, pruned_loss=0.08632, over 5750101.15 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3526, pruned_loss=0.1018, over 5700982.40 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:07:04,806 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:07:16,510 INFO [train.py:968] (0/2) Epoch 25, batch 20850, giga_loss[loss=0.353, simple_loss=0.4147, pruned_loss=0.1456, over 28913.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3542, pruned_loss=0.1023, over 5715237.50 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3384, pruned_loss=0.08649, over 5752248.14 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3542, pruned_loss=0.1033, over 5698851.74 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:07:48,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.030e+02 1.239e+03 1.464e+03 2.069e+03 4.021e+03, threshold=2.928e+03, percent-clipped=3.0
+2023-03-13 00:07:55,640 INFO [train.py:968] (0/2) Epoch 25, batch 20900, giga_loss[loss=0.257, simple_loss=0.3415, pruned_loss=0.08625, over 28388.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3532, pruned_loss=0.1014, over 5722016.82 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3379, pruned_loss=0.0862, over 5756161.22 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3541, pruned_loss=0.1028, over 5704717.59 frames. ], batch size: 65, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:08:01,091 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:02,319 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:06,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:19,679 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:23,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:30,472 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:38,054 INFO [train.py:968] (0/2) Epoch 25, batch 20950, giga_loss[loss=0.283, simple_loss=0.359, pruned_loss=0.1035, over 29005.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.353, pruned_loss=0.0999, over 5717864.03 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.338, pruned_loss=0.08625, over 5756902.41 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3536, pruned_loss=0.1011, over 5703429.04 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:08:44,997 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:09:12,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.517e+02 1.122e+03 1.405e+03 1.663e+03 4.327e+03, threshold=2.809e+03, percent-clipped=3.0
+2023-03-13 00:09:19,041 INFO [train.py:968] (0/2) Epoch 25, batch 21000, giga_loss[loss=0.2612, simple_loss=0.3397, pruned_loss=0.09135, over 29044.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3537, pruned_loss=0.1001, over 5720078.69 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3383, pruned_loss=0.08653, over 5755974.90 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3541, pruned_loss=0.101, over 5708695.20 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:09:19,045 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 00:09:28,061 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2071, simple_loss=0.3152, pruned_loss=0.04952, over 944034.00 frames. 
+2023-03-13 00:09:28,062 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 00:10:05,356 INFO [train.py:968] (0/2) Epoch 25, batch 21050, giga_loss[loss=0.2534, simple_loss=0.3126, pruned_loss=0.09715, over 23589.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3522, pruned_loss=0.09979, over 5714767.40 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3384, pruned_loss=0.08675, over 5756342.31 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3529, pruned_loss=0.1007, over 5704288.14 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:10:05,648 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:10:07,579 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:10:13,160 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4556, 1.7510, 1.3969, 1.5355], device='cuda:0'), covar=tensor([0.2713, 0.2708, 0.3001, 0.2530], device='cuda:0'), in_proj_covar=tensor([0.1555, 0.1123, 0.1374, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 00:10:19,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4973, 1.6570, 1.2045, 1.2192], device='cuda:0'), covar=tensor([0.0906, 0.0479, 0.0906, 0.1186], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0443, 0.0521, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 00:10:28,196 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:10:37,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.715e+02 1.218e+03 1.497e+03 1.996e+03 3.855e+03, threshold=2.993e+03, percent-clipped=5.0
+2023-03-13 00:10:41,834 INFO [train.py:968] (0/2) Epoch 25, batch 21100, giga_loss[loss=0.303, simple_loss=0.3668, pruned_loss=0.1196, over 27970.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3502, pruned_loss=0.0988, over 5710225.92 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3392, pruned_loss=0.08725, over 5751466.67 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3505, pruned_loss=0.09947, over 5704269.32 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:11:02,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:11:20,117 INFO [train.py:968] (0/2) Epoch 25, batch 21150, giga_loss[loss=0.2542, simple_loss=0.3363, pruned_loss=0.08606, over 28849.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3484, pruned_loss=0.09798, over 5714793.92 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3395, pruned_loss=0.08739, over 5752848.99 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3485, pruned_loss=0.09849, over 5708663.44 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:11:31,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1115003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:11:57,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.257e+02 1.115e+03 1.511e+03 1.963e+03 3.461e+03, threshold=3.021e+03, percent-clipped=3.0
+2023-03-13 00:12:01,863 INFO [train.py:968] (0/2) Epoch 25, batch 21200, giga_loss[loss=0.3064, simple_loss=0.3721, pruned_loss=0.1203, over 27909.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3488, pruned_loss=0.09908, over 5715712.39 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.339, pruned_loss=0.08727, over 5754580.40 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3495, pruned_loss=0.09973, over 5708816.44 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:12:10,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1115050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:12:30,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 00:12:41,339 INFO [train.py:968] (0/2) Epoch 25, batch 21250, giga_loss[loss=0.2374, simple_loss=0.3252, pruned_loss=0.07477, over 28429.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3486, pruned_loss=0.09887, over 5694416.35 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3392, pruned_loss=0.08744, over 5736757.21 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3492, pruned_loss=0.09948, over 5703374.27 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:12:56,082 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1115109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:12:58,785 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1115112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:13:15,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.728e+02 1.123e+03 1.410e+03 1.941e+03 7.068e+03, threshold=2.821e+03, percent-clipped=6.0
+2023-03-13 00:13:21,133 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-13 00:13:21,362 INFO [train.py:968] (0/2) Epoch 25, batch 21300, giga_loss[loss=0.3053, simple_loss=0.3761, pruned_loss=0.1173, over 28546.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3478, pruned_loss=0.09755, over 5706219.43 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3393, pruned_loss=0.08751, over 5739572.45 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3483, pruned_loss=0.09817, over 5710293.80 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:13:22,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1115141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:13:26,055 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1115146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:13:28,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1115149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:13:36,163 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9512, 2.7185, 1.7837, 1.2114], device='cuda:0'), covar=tensor([0.7785, 0.3192, 0.4168, 0.6637], device='cuda:0'), in_proj_covar=tensor([0.1785, 0.1667, 0.1616, 0.1446], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 00:13:39,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5092, 1.7652, 1.6793, 1.4709], device='cuda:0'), covar=tensor([0.3533, 0.2721, 0.2414, 0.2917], device='cuda:0'), in_proj_covar=tensor([0.2002, 0.1941, 0.1850, 0.2007], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 00:13:40,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6404, 1.7476, 1.8789, 1.4437], device='cuda:0'), covar=tensor([0.1848, 0.2695, 0.1498, 0.1792], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0712, 0.0973, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 00:13:50,808 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1115178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:13:59,656 INFO [train.py:968] (0/2) Epoch 25, batch 21350, giga_loss[loss=0.2994, simple_loss=0.3613, pruned_loss=0.1187, over 26537.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3467, pruned_loss=0.09646, over 5702992.54 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.34, pruned_loss=0.08826, over 5742578.29 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3467, pruned_loss=0.09654, over 5702805.14 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:14:02,758 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1115193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:14:04,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1115196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:14:11,179 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-13 00:14:12,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1115206.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:14:13,138 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 00:14:28,363 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1115225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:14:34,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.424e+02 1.083e+03 1.305e+03 1.790e+03 4.060e+03, threshold=2.610e+03, percent-clipped=7.0
+2023-03-13 00:14:38,680 INFO [train.py:968] (0/2) Epoch 25, batch 21400, giga_loss[loss=0.2492, simple_loss=0.3266, pruned_loss=0.08588, over 28766.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3469, pruned_loss=0.09728, over 5702132.44 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3403, pruned_loss=0.08836, over 5743543.41 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3468, pruned_loss=0.09737, over 5700657.21 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:15:04,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2882, 2.7775, 1.4306, 1.3940], device='cuda:0'), covar=tensor([0.1033, 0.0338, 0.0917, 0.1404], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0554, 0.0396, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:0')
+2023-03-13 00:15:18,822 INFO [train.py:968] (0/2) Epoch 25, batch 21450, giga_loss[loss=0.2725, simple_loss=0.3488, pruned_loss=0.09813, over 27960.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3449, pruned_loss=0.09638, over 5701357.57 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3407, pruned_loss=0.08876, over 5744738.63 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3445, pruned_loss=0.09624, over 5698389.94 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:15:51,848 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.124e+02 1.169e+03 1.444e+03 2.120e+03 4.754e+03, threshold=2.888e+03, percent-clipped=10.0
+2023-03-13 00:15:57,421 INFO [train.py:968] (0/2) Epoch 25, batch 21500, giga_loss[loss=0.2855, simple_loss=0.3568, pruned_loss=0.1071, over 28726.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3427, pruned_loss=0.09536, over 5694372.85 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3417, pruned_loss=0.08943, over 5736941.50 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3415, pruned_loss=0.09488, over 5697509.00 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:16:25,411 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-13 00:16:36,954 INFO [train.py:968] (0/2) Epoch 25, batch 21550, libri_loss[loss=0.2691, simple_loss=0.357, pruned_loss=0.0906, over 25427.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3412, pruned_loss=0.09485, over 5677640.93 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3424, pruned_loss=0.08991, over 5727720.84 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3397, pruned_loss=0.09418, over 5687322.17 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:16:52,563 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 00:17:00,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9001, 2.1256, 1.4898, 1.6620], device='cuda:0'), covar=tensor([0.0802, 0.0427, 0.0878, 0.0963], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0446, 0.0523, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 00:17:08,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.474e+02 1.239e+03 1.531e+03 1.896e+03 6.792e+03, threshold=3.062e+03, percent-clipped=7.0
+2023-03-13 00:17:13,794 INFO [train.py:968] (0/2) Epoch 25, batch 21600, libri_loss[loss=0.3308, simple_loss=0.3978, pruned_loss=0.1319, over 27788.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3416, pruned_loss=0.09537, over 5671749.14 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3429, pruned_loss=0.09032, over 5715510.18 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3398, pruned_loss=0.09461, over 5690313.08 frames. ], batch size: 116, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:17:53,127 INFO [train.py:968] (0/2) Epoch 25, batch 21650, giga_loss[loss=0.3062, simple_loss=0.3739, pruned_loss=0.1192, over 28347.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3387, pruned_loss=0.09415, over 5682698.10 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.343, pruned_loss=0.09081, over 5719779.74 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3371, pruned_loss=0.09321, over 5693000.70 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:17:56,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5667, 1.7524, 1.7566, 1.3329], device='cuda:0'), covar=tensor([0.1799, 0.2925, 0.1597, 0.1978], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0711, 0.0973, 0.0868], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 00:18:15,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0527, 2.3896, 2.0962, 2.2322], device='cuda:0'), covar=tensor([0.2198, 0.2088, 0.2360, 0.1896], device='cuda:0'), in_proj_covar=tensor([0.1555, 0.1124, 0.1374, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 00:18:29,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.691e+02 1.197e+03 1.541e+03 2.393e+03 6.817e+03, threshold=3.082e+03, percent-clipped=13.0
+2023-03-13 00:18:32,316 INFO [train.py:968] (0/2) Epoch 25, batch 21700, giga_loss[loss=0.239, simple_loss=0.3158, pruned_loss=0.08108, over 28587.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3377, pruned_loss=0.09405, over 5687367.94 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3442, pruned_loss=0.09204, over 5714180.88 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.335, pruned_loss=0.09228, over 5699405.73 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:19:03,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1115581.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:19:09,836 INFO [train.py:968] (0/2) Epoch 25, batch 21750, giga_loss[loss=0.2942, simple_loss=0.3551, pruned_loss=0.1166, over 27634.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3356, pruned_loss=0.09325, over 5692950.42 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3448, pruned_loss=0.0926, over 5716087.90 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3327, pruned_loss=0.09134, over 5700038.73 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:19:47,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.000e+02 1.133e+03 1.412e+03 1.898e+03 1.040e+04, threshold=2.825e+03, percent-clipped=5.0
+2023-03-13 00:19:50,927 INFO [train.py:968] (0/2) Epoch 25, batch 21800, libri_loss[loss=0.3108, simple_loss=0.3828, pruned_loss=0.1193, over 29380.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3333, pruned_loss=0.09212, over 5701155.11 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3452, pruned_loss=0.09296, over 5716120.84 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3304, pruned_loss=0.09029, over 5706267.49 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:19:51,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4144, 3.5765, 1.5927, 1.5436], device='cuda:0'), covar=tensor([0.1005, 0.0415, 0.0956, 0.1354], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0556, 0.0398, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 00:19:54,176 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 00:20:30,536 INFO [train.py:968] (0/2) Epoch 25, batch 21850, giga_loss[loss=0.2613, simple_loss=0.3418, pruned_loss=0.09038, over 28993.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3338, pruned_loss=0.09251, over 5705227.40 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3458, pruned_loss=0.09353, over 5717961.18 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3308, pruned_loss=0.09054, over 5707382.42 frames. ], batch size: 164, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:20:36,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3794, 3.3170, 1.4239, 1.5493], device='cuda:0'), covar=tensor([0.0986, 0.0342, 0.0998, 0.1333], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0556, 0.0398, 0.0436], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 00:20:59,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1115724.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:21:03,421 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1115727.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:21:05,378 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1115730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:21:08,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4579, 1.6059, 1.2289, 1.1766], device='cuda:0'), covar=tensor([0.1022, 0.0618, 0.1066, 0.1276], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0445, 0.0522, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 00:21:09,910 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.776e+02 1.127e+03 1.368e+03 1.986e+03 4.840e+03, threshold=2.737e+03, percent-clipped=7.0
+2023-03-13 00:21:14,027 INFO [train.py:968] (0/2) Epoch 25, batch 21900, giga_loss[loss=0.2585, simple_loss=0.3418, pruned_loss=0.08763, over 29004.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3364, pruned_loss=0.09329, over 5695727.14 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3458, pruned_loss=0.09366, over 5710601.84 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3339, pruned_loss=0.09161, over 5703249.74 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:21:19,986 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-13 00:21:27,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1115756.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:21:38,561 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1115770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:21:55,076 INFO [train.py:968] (0/2) Epoch 25, batch 21950, giga_loss[loss=0.2851, simple_loss=0.36, pruned_loss=0.1051, over 28853.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3398, pruned_loss=0.09454, over 5689214.36 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3461, pruned_loss=0.09403, over 5711894.70 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3373, pruned_loss=0.09288, over 5693525.81 frames. ], batch size: 243, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:22:00,202 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1115794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:22:32,380 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.866e+02 1.134e+03 1.469e+03 1.864e+03 6.652e+03, threshold=2.937e+03, percent-clipped=5.0
+2023-03-13 00:22:38,063 INFO [train.py:968] (0/2) Epoch 25, batch 22000, giga_loss[loss=0.2612, simple_loss=0.3457, pruned_loss=0.08834, over 28912.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3398, pruned_loss=0.09349, over 5695308.27 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3458, pruned_loss=0.09396, over 5713033.41 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3381, pruned_loss=0.09224, over 5697475.51 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:23:19,103 INFO [train.py:968] (0/2) Epoch 25, batch 22050, giga_loss[loss=0.2403, simple_loss=0.3212, pruned_loss=0.07973, over 28887.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3392, pruned_loss=0.09241, over 5698769.85 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3458, pruned_loss=0.09409, over 5713102.02 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3378, pruned_loss=0.09128, over 5700178.91 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:24:00,291 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.781e+02 1.130e+03 1.517e+03 1.876e+03 6.064e+03, threshold=3.034e+03, percent-clipped=4.0
+2023-03-13 00:24:01,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1950, 4.0135, 3.7999, 1.8451], device='cuda:0'), covar=tensor([0.0710, 0.0873, 0.0792, 0.2075], device='cuda:0'), in_proj_covar=tensor([0.1248, 0.1156, 0.0972, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 00:24:02,781 INFO [train.py:968] (0/2) Epoch 25, batch 22100, giga_loss[loss=0.235, simple_loss=0.3096, pruned_loss=0.08021, over 28914.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3378, pruned_loss=0.09168, over 5698164.33 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3456, pruned_loss=0.09423, over 5717146.96 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3366, pruned_loss=0.0906, over 5695500.82 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:24:22,946 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8662, 1.1640, 2.8735, 2.6847], device='cuda:0'), covar=tensor([0.1669, 0.2566, 0.0613, 0.0968], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0656, 0.0969, 0.0934], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 00:24:41,177 INFO [train.py:968] (0/2) Epoch 25, batch 22150, libri_loss[loss=0.3011, simple_loss=0.3689, pruned_loss=0.1166, over 25936.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3391, pruned_loss=0.09277, over 5702199.43 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3464, pruned_loss=0.09504, over 5718777.54 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.337, pruned_loss=0.09104, over 5697874.59 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:24:48,388 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1116000.pt
+2023-03-13 00:24:56,059 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:25:04,677 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:25:18,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.304e+02 1.390e+03 1.781e+03 2.397e+03 5.729e+03, threshold=3.562e+03, percent-clipped=12.0
+2023-03-13 00:25:22,879 INFO [train.py:968] (0/2) Epoch 25, batch 22200, giga_loss[loss=0.2729, simple_loss=0.3402, pruned_loss=0.1028, over 29024.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3403, pruned_loss=0.09405, over 5698083.58 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3465, pruned_loss=0.09522, over 5717601.49 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3384, pruned_loss=0.09246, over 5695164.64 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:25:29,331 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-13 00:26:03,503 INFO [train.py:968] (0/2) Epoch 25, batch 22250, giga_loss[loss=0.2909, simple_loss=0.3654, pruned_loss=0.1082, over 29014.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3428, pruned_loss=0.09549, over 5702483.55 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3467, pruned_loss=0.0955, over 5720406.85 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.341, pruned_loss=0.09396, over 5697454.58 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:26:17,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:26:40,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.386e+02 1.330e+03 1.676e+03 2.135e+03 6.877e+03, threshold=3.351e+03, percent-clipped=6.0
+2023-03-13 00:26:43,995 INFO [train.py:968] (0/2) Epoch 25, batch 22300, giga_loss[loss=0.3093, simple_loss=0.3727, pruned_loss=0.123, over 28820.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3457, pruned_loss=0.09663, over 5712321.91 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3467, pruned_loss=0.09569, over 5723580.35 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3441, pruned_loss=0.09526, over 5705356.72 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:26:47,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:27:09,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:27:12,006 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:27:23,896 INFO [train.py:968] (0/2) Epoch 25, batch 22350, giga_loss[loss=0.2763, simple_loss=0.3553, pruned_loss=0.0986, over 28951.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3491, pruned_loss=0.09867, over 5716033.36 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3474, pruned_loss=0.09625, over 5727533.11 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3473, pruned_loss=0.09712, over 5706479.73 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:28:01,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.188e+02 1.303e+03 1.666e+03 2.256e+03 7.276e+03, threshold=3.333e+03, percent-clipped=6.0
+2023-03-13 00:28:03,716 INFO [train.py:968] (0/2) Epoch 25, batch 22400, giga_loss[loss=0.3369, simple_loss=0.3909, pruned_loss=0.1414, over 28880.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3505, pruned_loss=0.09933, over 5721570.57 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3474, pruned_loss=0.09632, over 5729431.70 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3491, pruned_loss=0.09807, over 5712152.99 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:28:11,592 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:28:13,544 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:28:38,841 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-13 00:28:39,457 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:28:45,604 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:28:46,610 INFO [train.py:968] (0/2) Epoch 25, batch 22450, giga_loss[loss=0.3099, simple_loss=0.3795, pruned_loss=0.1201, over 29079.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3511, pruned_loss=0.09985, over 5720539.73 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3481, pruned_loss=0.09689, over 5731773.06 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3493, pruned_loss=0.09836, over 5710835.91 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:28:47,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:29:05,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:29:07,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:29:08,489 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116316.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:29:10,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9451, 1.2005, 2.8169, 2.7588], device='cuda:0'), covar=tensor([0.1618, 0.2609, 0.0567, 0.1216], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0658, 0.0972, 0.0940], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 00:29:11,090 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:29:23,252 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.097e+02 1.243e+03 1.541e+03 2.015e+03 5.248e+03, threshold=3.082e+03, percent-clipped=6.0
+2023-03-13 00:29:28,238 INFO [train.py:968] (0/2) Epoch 25, batch 22500, giga_loss[loss=0.2747, simple_loss=0.3525, pruned_loss=0.09846, over 28971.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3505, pruned_loss=0.09973, over 5714956.34 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3483, pruned_loss=0.09701, over 5726576.80 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.349, pruned_loss=0.09849, over 5712274.77 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:29:31,899 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:30:00,869 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:30:05,512 INFO [train.py:968] (0/2) Epoch 25, batch 22550, giga_loss[loss=0.2337, simple_loss=0.3165, pruned_loss=0.07546, over 28699.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3491, pruned_loss=0.09898, over 5714360.43 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3489, pruned_loss=0.09752, over 5722538.82 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3474, pruned_loss=0.09759, over 5715068.78 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:30:10,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:30:25,251 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-13 00:30:47,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.008e+02 1.217e+03 1.465e+03 2.026e+03 5.028e+03, threshold=2.930e+03, percent-clipped=9.0
+2023-03-13 00:30:50,486 INFO [train.py:968] (0/2) Epoch 25, batch 22600, giga_loss[loss=0.2644, simple_loss=0.3452, pruned_loss=0.09177, over 28712.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3458, pruned_loss=0.09745, over 5722146.56 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.349, pruned_loss=0.09784, over 5727204.98 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3442, pruned_loss=0.09602, over 5718058.28 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:31:28,146 INFO [train.py:968] (0/2) Epoch 25, batch 22650, giga_loss[loss=0.2497, simple_loss=0.3344, pruned_loss=0.08244, over 28895.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3432, pruned_loss=0.09586, over 5701806.33 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3495, pruned_loss=0.09832, over 5709762.20 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3413, pruned_loss=0.09423, over 5714117.24 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:31:30,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5273, 1.8548, 1.4663, 1.5775], device='cuda:0'), covar=tensor([0.2663, 0.2713, 0.3116, 0.2647], device='cuda:0'), in_proj_covar=tensor([0.1554, 0.1120, 0.1372, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 00:31:57,551 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:31:59,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:05,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.662e+02 1.285e+03 1.569e+03 2.082e+03 8.759e+03, threshold=3.138e+03, percent-clipped=13.0
+2023-03-13 00:32:06,095 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:08,557 INFO [train.py:968] (0/2) Epoch 25, batch 22700, giga_loss[loss=0.2616, simple_loss=0.3486, pruned_loss=0.08725, over 28920.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3446, pruned_loss=0.0959, over 5699085.78 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3502, pruned_loss=0.09903, over 5704895.57 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3422, pruned_loss=0.09388, over 5713772.67 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:32:08,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:15,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:17,152 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116550.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:32:22,510 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:33,513 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:50,915 INFO [train.py:968] (0/2) Epoch 25, batch 22750, libri_loss[loss=0.2556, simple_loss=0.3249, pruned_loss=0.09314, over 29361.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3454, pruned_loss=0.09542, over 5698431.61 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3506, pruned_loss=0.09939, over 5704357.48 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3431, pruned_loss=0.09344, over 5710244.95 frames. ], batch size: 67, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:33:07,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5700, 4.9978, 1.7774, 1.8638], device='cuda:0'), covar=tensor([0.0982, 0.0298, 0.0957, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0557, 0.0398, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 00:33:20,429 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-13 00:33:28,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.240e+02 1.237e+03 1.601e+03 2.175e+03 4.931e+03, threshold=3.203e+03, percent-clipped=4.0
+2023-03-13 00:33:30,470 INFO [train.py:968] (0/2) Epoch 25, batch 22800, libri_loss[loss=0.2821, simple_loss=0.3577, pruned_loss=0.1033, over 19520.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3447, pruned_loss=0.09505, over 5703749.55 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.351, pruned_loss=0.09967, over 5699733.97 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3425, pruned_loss=0.09314, over 5718461.48 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:34:12,329 INFO [train.py:968] (0/2) Epoch 25, batch 22850, giga_loss[loss=0.2662, simple_loss=0.3389, pruned_loss=0.09682, over 28907.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3435, pruned_loss=0.0954, over 5709861.57 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3506, pruned_loss=0.09957, over 5704499.44 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3419, pruned_loss=0.09386, over 5717443.46 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:34:12,641 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:34:13,301 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116691.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:34:14,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:34:32,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:34:38,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:34:50,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.833e+02 1.291e+03 1.604e+03 2.205e+03 5.854e+03, threshold=3.209e+03, percent-clipped=10.0
+2023-03-13 00:34:52,188 INFO [train.py:968] (0/2) Epoch 25, batch 22900, giga_loss[loss=0.265, simple_loss=0.3194, pruned_loss=0.1053, over 28653.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3417, pruned_loss=0.09597, over 5712603.16 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3508, pruned_loss=0.09987, over 5708958.94 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.34, pruned_loss=0.09435, over 5714991.52 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:34:52,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0544, 1.9616, 1.6351, 1.5742], device='cuda:0'), covar=tensor([0.0982, 0.0812, 0.1014, 0.1300], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0446, 0.0520, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 00:35:26,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6542, 1.8874, 1.8330, 1.4460], device='cuda:0'), covar=tensor([0.1766, 0.2689, 0.1542, 0.1855], device='cuda:0'), in_proj_covar=tensor([0.0921, 0.0706, 0.0967, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 00:35:30,075 INFO [train.py:968] (0/2) Epoch 25, batch 22950, libri_loss[loss=0.282, simple_loss=0.3552, pruned_loss=0.1044, over 29521.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3415, pruned_loss=0.09686, over 5717515.24 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3514, pruned_loss=0.1004, over 5712224.44 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3392, pruned_loss=0.09495, over 5716537.08 frames. ], batch size: 82, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:35:47,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2847, 1.1379, 4.1095, 3.4681], device='cuda:0'), covar=tensor([0.1801, 0.3014, 0.0411, 0.0785], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0659, 0.0974, 0.0940], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 00:36:06,391 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116834.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:36:08,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.062e+02 1.193e+03 1.430e+03 1.861e+03 4.766e+03, threshold=2.861e+03, percent-clipped=3.0
+2023-03-13 00:36:08,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116837.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:36:10,748 INFO [train.py:968] (0/2) Epoch 25, batch 23000, giga_loss[loss=0.2743, simple_loss=0.349, pruned_loss=0.09981, over 28689.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3412, pruned_loss=0.09726, over 5722843.38 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.352, pruned_loss=0.1009, over 5716239.12 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3385, pruned_loss=0.09516, over 5718474.91 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:36:30,850 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116866.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:36:49,437 INFO [train.py:968] (0/2) Epoch 25, batch 23050, giga_loss[loss=0.2419, simple_loss=0.3169, pruned_loss=0.08347, over 28877.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3372, pruned_loss=0.09532, over 5712693.06 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3517, pruned_loss=0.1008, over 5711023.90 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3351, pruned_loss=0.09363, over 5713644.52 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:37:18,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116925.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:37:29,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.621e+02 1.291e+03 1.645e+03 2.172e+03 6.428e+03, threshold=3.291e+03, percent-clipped=12.0
+2023-03-13 00:37:30,434 INFO [train.py:968] (0/2) Epoch 25, batch 23100, giga_loss[loss=0.2593, simple_loss=0.3337, pruned_loss=0.09251, over 27855.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3333, pruned_loss=0.09323, over 5719443.69 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.352, pruned_loss=0.1009, over 5713780.19 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3312, pruned_loss=0.09173, over 5717954.54 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:37:50,177 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116963.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:38:05,069 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:38:11,051 INFO [train.py:968] (0/2) Epoch 25, batch 23150, giga_loss[loss=0.2109, simple_loss=0.2973, pruned_loss=0.06221, over 29004.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3294, pruned_loss=0.09147, over 5710621.02 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3524, pruned_loss=0.1013, over 5704427.82 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.327, pruned_loss=0.08973, over 5717020.03 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:38:46,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.761e+02 1.267e+03 1.671e+03 2.209e+03 4.664e+03, threshold=3.341e+03, percent-clipped=7.0
+2023-03-13 00:38:48,781 INFO [train.py:968] (0/2) Epoch 25, batch 23200, giga_loss[loss=0.2813, simple_loss=0.3419, pruned_loss=0.1103, over 23679.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3292, pruned_loss=0.09095, over 5708745.89 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3529, pruned_loss=0.1017, over 5705002.63 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3262, pruned_loss=0.08892, over 5713787.23 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:39:11,335 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117068.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:39:13,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117071.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:39:27,842 INFO [train.py:968] (0/2) Epoch 25, batch 23250, giga_loss[loss=0.2289, simple_loss=0.3145, pruned_loss=0.07164, over 28425.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3327, pruned_loss=0.09279, over 5701961.60 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3529, pruned_loss=0.102, over 5698790.15 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3296, pruned_loss=0.09063, over 5710997.88 frames. ], batch size: 60, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:39:28,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:39:36,188 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117100.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:39:49,055 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1117115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:40:08,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.889e+02 1.234e+03 1.609e+03 2.164e+03 5.168e+03, threshold=3.219e+03, percent-clipped=9.0
+2023-03-13 00:40:09,271 INFO [train.py:968] (0/2) Epoch 25, batch 23300, giga_loss[loss=0.2615, simple_loss=0.3446, pruned_loss=0.08918, over 28932.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3363, pruned_loss=0.09438, over 5706808.93 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3532, pruned_loss=0.1023, over 5702844.90 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3333, pruned_loss=0.09227, over 5710242.51 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:40:09,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 00:40:16,833 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1117149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:40:37,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5204, 1.7673, 1.4275, 1.8837], device='cuda:0'), covar=tensor([0.2173, 0.2258, 0.2439, 0.2170], device='cuda:0'), in_proj_covar=tensor([0.1556, 0.1122, 0.1371, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 00:40:47,506 INFO [train.py:968] (0/2) Epoch 25, batch 23350, giga_loss[loss=0.263, simple_loss=0.3388, pruned_loss=0.09363, over 28813.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3395, pruned_loss=0.09558, over 5711851.03 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3529, pruned_loss=0.1024, over 5710359.44 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3368, pruned_loss=0.09353, over 5707934.71 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:41:26,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:41:28,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:41:29,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.183e+02 1.306e+03 1.682e+03 2.211e+03 5.052e+03, threshold=3.363e+03, percent-clipped=4.0
+2023-03-13 00:41:30,486 INFO [train.py:968] (0/2) Epoch 25, batch 23400, giga_loss[loss=0.2892, simple_loss=0.3638, pruned_loss=0.1073, over 28891.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3419, pruned_loss=0.0965, over 5719937.73 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.353, pruned_loss=0.1025, over 5713085.90 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3396, pruned_loss=0.09468, over 5714430.36 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:41:53,503 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:41:56,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3058, 0.8595, 1.0145, 1.4274], device='cuda:0'), covar=tensor([0.0720, 0.0368, 0.0329, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 00:42:01,987 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1117278.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:42:09,742 INFO [train.py:968] (0/2) Epoch 25, batch 23450, giga_loss[loss=0.2942, simple_loss=0.3594, pruned_loss=0.1145, over 28901.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3433, pruned_loss=0.09721, over 5722707.59 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3531, pruned_loss=0.1028, over 5714963.22 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.341, pruned_loss=0.09532, over 5716861.20 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:42:57,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117338.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:42:58,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.491e+03 1.901e+03 3.143e+03 1.342e+04, threshold=3.802e+03, percent-clipped=21.0
+2023-03-13 00:42:58,413 INFO [train.py:968] (0/2) Epoch 25, batch 23500, libri_loss[loss=0.2833, simple_loss=0.3629, pruned_loss=0.1019, over 28612.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3488, pruned_loss=0.1022, over 5707743.61 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3528, pruned_loss=0.1028, over 5716225.85 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3471, pruned_loss=0.1006, over 5701551.60 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 00:43:13,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:43:49,182 INFO [train.py:968] (0/2) Epoch 25, batch 23550, giga_loss[loss=0.3614, simple_loss=0.3923, pruned_loss=0.1653, over 23639.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3562, pruned_loss=0.1078, over 5688986.71 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3529, pruned_loss=0.1029, over 5710247.84 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3547, pruned_loss=0.1065, over 5689725.30 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 00:43:55,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1117397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:44:32,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.792e+03 2.601e+03 3.879e+03 7.122e+03, threshold=5.202e+03, percent-clipped=27.0
+2023-03-13 00:44:32,593 INFO [train.py:968] (0/2) Epoch 25, batch 23600, giga_loss[loss=0.3572, simple_loss=0.4061, pruned_loss=0.1541, over 27579.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3628, pruned_loss=0.1129, over 5688922.62 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3534, pruned_loss=0.1036, over 5716967.13 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3614, pruned_loss=0.1116, over 5682639.34 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:45:03,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4191, 1.1962, 1.1992, 1.5891], device='cuda:0'), covar=tensor([0.0714, 0.0358, 0.0337, 0.0820], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 00:45:11,255 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117481.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:45:14,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117484.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:45:20,004 INFO [train.py:968] (0/2) Epoch 25, batch 23650, giga_loss[loss=0.3678, simple_loss=0.4216, pruned_loss=0.157, over 28664.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3701, pruned_loss=0.1188, over 5682353.05 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3535, pruned_loss=0.1036, over 5709439.56 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3691, pruned_loss=0.1178, over 5684609.25 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:45:20,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:45:29,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3793, 2.0608, 1.6071, 1.3648], device='cuda:0'), covar=tensor([0.0794, 0.0275, 0.0291, 0.1061], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 00:45:31,489 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:45:34,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:45:39,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2515, 1.6146, 1.5586, 1.1021], device='cuda:0'), covar=tensor([0.1481, 0.2516, 0.1317, 0.1618], device='cuda:0'), in_proj_covar=tensor([0.0918, 0.0706, 0.0965, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 00:45:46,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117513.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:45:56,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:46:03,573 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:46:12,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.201e+03 1.819e+03 2.199e+03 3.240e+03 8.758e+03, threshold=4.398e+03, percent-clipped=6.0
+2023-03-13 00:46:12,864 INFO [train.py:968] (0/2) Epoch 25, batch 23700, giga_loss[loss=0.4373, simple_loss=0.4433, pruned_loss=0.2157, over 23411.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3775, pruned_loss=0.125, over 5674982.62 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3537, pruned_loss=0.1038, over 5710531.17 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3766, pruned_loss=0.1243, over 5675515.63 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:46:26,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4782, 1.6887, 1.5551, 1.5775], device='cuda:0'), covar=tensor([0.1575, 0.1662, 0.1986, 0.1691], device='cuda:0'), in_proj_covar=tensor([0.0490, 0.0750, 0.0720, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 00:46:59,077 INFO [train.py:968] (0/2) Epoch 25, batch 23750, giga_loss[loss=0.291, simple_loss=0.3545, pruned_loss=0.1138, over 28737.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.38, pruned_loss=0.1275, over 5675838.42 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3535, pruned_loss=0.1039, over 5712898.08 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3804, pruned_loss=0.1275, over 5672945.43 frames. ], batch size: 66, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:47:34,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4452, 1.6944, 1.2252, 1.3145], device='cuda:0'), covar=tensor([0.0991, 0.0510, 0.1106, 0.1026], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0449, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 00:47:38,766 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:47:42,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:47:46,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 1.994e+03 2.431e+03 3.257e+03 8.302e+03, threshold=4.862e+03, percent-clipped=11.0
+2023-03-13 00:47:46,271 INFO [train.py:968] (0/2) Epoch 25, batch 23800, giga_loss[loss=0.3744, simple_loss=0.4167, pruned_loss=0.166, over 28571.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3811, pruned_loss=0.1291, over 5670838.72 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3536, pruned_loss=0.1041, over 5715742.27 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.382, pruned_loss=0.1296, over 5665002.30 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:47:59,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117653.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:48:09,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7954, 1.5466, 1.7922, 1.3647], device='cuda:0'), covar=tensor([0.2026, 0.3165, 0.1546, 0.1759], device='cuda:0'), in_proj_covar=tensor([0.0916, 0.0706, 0.0963, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 00:48:10,974 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:48:12,841 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:48:18,293 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:48:37,932 INFO [train.py:968] (0/2) Epoch 25, batch 23850, giga_loss[loss=0.2883, simple_loss=0.3487, pruned_loss=0.1139, over 28787.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3862, pruned_loss=0.1348, over 5656708.24 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3534, pruned_loss=0.104, over 5717373.38 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3873, pruned_loss=0.1355, over 5650362.12 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:48:47,035 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:49:27,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.809e+03 2.431e+03 3.019e+03 6.600e+03, threshold=4.861e+03, percent-clipped=7.0
+2023-03-13 00:49:27,573 INFO [train.py:968] (0/2) Epoch 25, batch 23900, libri_loss[loss=0.2265, simple_loss=0.2994, pruned_loss=0.07676, over 28566.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3901, pruned_loss=0.139, over 5649477.92 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3532, pruned_loss=0.104, over 5720078.98 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.392, pruned_loss=0.1402, over 5640835.06 frames. ], batch size: 63, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:50:02,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:50:22,025 INFO [train.py:968] (0/2) Epoch 25, batch 23950, giga_loss[loss=0.3338, simple_loss=0.3922, pruned_loss=0.1377, over 28943.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3916, pruned_loss=0.1397, over 5655691.83 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3534, pruned_loss=0.1043, over 5724975.36 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3941, pruned_loss=0.1415, over 5642505.34 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:50:24,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7568, 1.9872, 1.5905, 2.0360], device='cuda:0'), covar=tensor([0.2724, 0.2697, 0.3140, 0.2246], device='cuda:0'), in_proj_covar=tensor([0.1555, 0.1123, 0.1374, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 00:50:30,157 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117796.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:50:36,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117799.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:51:01,641 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117828.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:51:14,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.088e+03 1.818e+03 2.518e+03 3.185e+03 1.120e+04, threshold=5.036e+03, percent-clipped=6.0
+2023-03-13 00:51:14,980 INFO [train.py:968] (0/2) Epoch 25, batch 24000, giga_loss[loss=0.3523, simple_loss=0.4156, pruned_loss=0.1446, over 28571.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3917, pruned_loss=0.1404, over 5639799.84 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3536, pruned_loss=0.1044, over 5726669.28 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3946, pruned_loss=0.1428, over 5625609.13 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:51:14,984 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 00:51:23,955 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2031, simple_loss=0.3108, pruned_loss=0.04772, over 944034.00 frames. 
+2023-03-13 00:51:23,956 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 00:52:06,951 INFO [train.py:968] (0/2) Epoch 25, batch 24050, giga_loss[loss=0.361, simple_loss=0.4065, pruned_loss=0.1577, over 28970.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3904, pruned_loss=0.1399, over 5649861.25 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3537, pruned_loss=0.1047, over 5722954.54 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3937, pruned_loss=0.1427, over 5640049.86 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:52:28,622 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 00:52:29,227 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:52:32,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:52:35,935 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-13 00:52:50,565 INFO [train.py:968] (0/2) Epoch 25, batch 24100, giga_loss[loss=0.4156, simple_loss=0.4478, pruned_loss=0.1917, over 26536.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3889, pruned_loss=0.138, over 5653140.14 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3531, pruned_loss=0.1044, over 5726574.21 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3933, pruned_loss=0.1415, over 5639790.98 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:52:51,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.308e+03 1.792e+03 2.344e+03 2.949e+03 6.548e+03, threshold=4.688e+03, percent-clipped=3.0
+2023-03-13 00:52:59,046 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:53:46,806 INFO [train.py:968] (0/2) Epoch 25, batch 24150, giga_loss[loss=0.3288, simple_loss=0.3847, pruned_loss=0.1365, over 28858.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3897, pruned_loss=0.1375, over 5650471.55 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3534, pruned_loss=0.1046, over 5727722.15 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3932, pruned_loss=0.1405, over 5638168.70 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:53:57,080 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1118000.pt
+2023-03-13 00:54:38,274 INFO [train.py:968] (0/2) Epoch 25, batch 24200, giga_loss[loss=0.4784, simple_loss=0.4869, pruned_loss=0.235, over 26469.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3918, pruned_loss=0.1393, over 5634578.72 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3538, pruned_loss=0.105, over 5729899.15 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3951, pruned_loss=0.142, over 5621063.49 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:54:38,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.851e+03 2.296e+03 3.116e+03 6.203e+03, threshold=4.592e+03, percent-clipped=8.0
+2023-03-13 00:54:39,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7305, 1.9307, 1.7576, 1.6665], device='cuda:0'), covar=tensor([0.2151, 0.2519, 0.2297, 0.2365], device='cuda:0'), in_proj_covar=tensor([0.0493, 0.0755, 0.0727, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 00:55:27,301 INFO [train.py:968] (0/2) Epoch 25, batch 24250, giga_loss[loss=0.4125, simple_loss=0.423, pruned_loss=0.201, over 23787.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3875, pruned_loss=0.1354, over 5614978.03 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3537, pruned_loss=0.1051, over 5712429.14 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3912, pruned_loss=0.1383, over 5616884.12 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:55:44,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9738, 2.1534, 1.7595, 1.9782], device='cuda:0'), covar=tensor([0.2587, 0.2716, 0.3183, 0.2574], device='cuda:0'), in_proj_covar=tensor([0.1553, 0.1123, 0.1372, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 00:56:14,263 INFO [train.py:968] (0/2) Epoch 25, batch 24300, giga_loss[loss=0.2811, simple_loss=0.3581, pruned_loss=0.1021, over 28874.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3842, pruned_loss=0.1311, over 5633759.00 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3541, pruned_loss=0.1055, over 5716469.60 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3876, pruned_loss=0.1338, over 5629478.33 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:56:17,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.525e+02 1.753e+03 2.417e+03 3.683e+03 8.741e+03, threshold=4.833e+03, percent-clipped=16.0
+2023-03-13 00:56:19,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3981, 1.2562, 3.5195, 3.2160], device='cuda:0'), covar=tensor([0.1427, 0.2697, 0.0478, 0.1311], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0661, 0.0980, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 00:56:32,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0197, 1.3280, 3.4092, 3.0327], device='cuda:0'), covar=tensor([0.1946, 0.2709, 0.0885, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0661, 0.0981, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 00:57:03,876 INFO [train.py:968] (0/2) Epoch 25, batch 24350, giga_loss[loss=0.3267, simple_loss=0.3842, pruned_loss=0.1346, over 28731.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.38, pruned_loss=0.1271, over 5651570.36 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3541, pruned_loss=0.1056, over 5721291.43 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3836, pruned_loss=0.1299, over 5641666.60 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:57:19,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:57:43,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6354, 1.6632, 1.9072, 1.4312], device='cuda:0'), covar=tensor([0.1802, 0.2645, 0.1458, 0.1783], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0706, 0.0961, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 00:57:49,893 INFO [train.py:968] (0/2) Epoch 25, batch 24400, giga_loss[loss=0.2767, simple_loss=0.3538, pruned_loss=0.09981, over 28598.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3768, pruned_loss=0.124, over 5661153.50 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3538, pruned_loss=0.1054, over 5714805.45 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3802, pruned_loss=0.1266, over 5657529.18 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:57:50,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.786e+02 1.677e+03 2.096e+03 2.679e+03 7.119e+03, threshold=4.192e+03, percent-clipped=3.0
+2023-03-13 00:57:55,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9166, 3.7681, 3.5797, 1.6339], device='cuda:0'), covar=tensor([0.0713, 0.0803, 0.0750, 0.2063], device='cuda:0'), in_proj_covar=tensor([0.1279, 0.1186, 0.0997, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 00:58:38,436 INFO [train.py:968] (0/2) Epoch 25, batch 24450, giga_loss[loss=0.2964, simple_loss=0.3637, pruned_loss=0.1146, over 28695.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3742, pruned_loss=0.1226, over 5653251.89 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.354, pruned_loss=0.1056, over 5714807.84 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3771, pruned_loss=0.1249, over 5649609.69 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:58:48,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7593, 4.8932, 2.0160, 2.0694], device='cuda:0'), covar=tensor([0.0986, 0.0393, 0.0842, 0.1236], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0564, 0.0400, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 00:59:27,618 INFO [train.py:968] (0/2) Epoch 25, batch 24500, libri_loss[loss=0.2907, simple_loss=0.3673, pruned_loss=0.1071, over 29180.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3743, pruned_loss=0.1227, over 5654216.69 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3543, pruned_loss=0.1059, over 5708253.10 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3767, pruned_loss=0.1245, over 5655841.63 frames. ], batch size: 97, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:59:29,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.729e+03 2.209e+03 2.777e+03 7.780e+03, threshold=4.418e+03, percent-clipped=4.0
+2023-03-13 00:59:39,269 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 01:00:10,448 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-13 01:00:19,307 INFO [train.py:968] (0/2) Epoch 25, batch 24550, giga_loss[loss=0.3068, simple_loss=0.3759, pruned_loss=0.1189, over 28875.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3734, pruned_loss=0.1216, over 5658791.05 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3542, pruned_loss=0.106, over 5701118.84 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3759, pruned_loss=0.1234, over 5665445.86 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:01:08,322 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:01:09,503 INFO [train.py:968] (0/2) Epoch 25, batch 24600, giga_loss[loss=0.2851, simple_loss=0.3631, pruned_loss=0.1036, over 28677.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3717, pruned_loss=0.1194, over 5662487.25 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3541, pruned_loss=0.106, over 5704750.24 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3743, pruned_loss=0.1212, over 5663475.79 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:01:11,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.843e+02 1.552e+03 1.960e+03 2.829e+03 6.834e+03, threshold=3.920e+03, percent-clipped=7.0
+2023-03-13 01:01:21,539 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:01:54,607 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 01:02:03,053 INFO [train.py:968] (0/2) Epoch 25, batch 24650, giga_loss[loss=0.2939, simple_loss=0.3772, pruned_loss=0.1053, over 28869.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3724, pruned_loss=0.1175, over 5668099.33 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3538, pruned_loss=0.1058, over 5708581.83 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.375, pruned_loss=0.1193, over 5664851.24 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:02:30,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:02:34,724 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:02:53,315 INFO [train.py:968] (0/2) Epoch 25, batch 24700, giga_loss[loss=0.4012, simple_loss=0.4364, pruned_loss=0.183, over 28840.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3735, pruned_loss=0.1184, over 5664946.17 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3542, pruned_loss=0.1062, over 5711674.46 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3756, pruned_loss=0.1197, over 5658256.12 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:02:54,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.650e+03 2.182e+03 2.918e+03 7.892e+03, threshold=4.364e+03, percent-clipped=8.0
+2023-03-13 01:03:06,099 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:03:36,344 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:03:40,338 INFO [train.py:968] (0/2) Epoch 25, batch 24750, giga_loss[loss=0.3059, simple_loss=0.3825, pruned_loss=0.1147, over 28838.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3735, pruned_loss=0.1189, over 5646751.28 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3544, pruned_loss=0.1065, over 5694719.41 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3752, pruned_loss=0.1199, over 5655472.37 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:03:55,980 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:04:28,288 INFO [train.py:968] (0/2) Epoch 25, batch 24800, giga_loss[loss=0.2757, simple_loss=0.3494, pruned_loss=0.101, over 28894.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3728, pruned_loss=0.1195, over 5641127.44 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3546, pruned_loss=0.1067, over 5698930.75 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3744, pruned_loss=0.1204, over 5643240.01 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:04:29,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.152e+03 1.780e+03 2.329e+03 2.947e+03 6.316e+03, threshold=4.658e+03, percent-clipped=7.0
+2023-03-13 01:04:41,295 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-13 01:05:14,213 INFO [train.py:968] (0/2) Epoch 25, batch 24850, giga_loss[loss=0.3063, simple_loss=0.3681, pruned_loss=0.1223, over 28671.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3708, pruned_loss=0.1192, over 5656617.32 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3548, pruned_loss=0.1069, over 5700296.27 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3721, pruned_loss=0.1199, over 5656122.53 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:05:36,508 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:05:45,142 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1118726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:05:50,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1118729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:05:52,019 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-13 01:05:57,146 INFO [train.py:968] (0/2) Epoch 25, batch 24900, giga_loss[loss=0.2868, simple_loss=0.3642, pruned_loss=0.1047, over 28880.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3695, pruned_loss=0.1188, over 5661231.31 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3547, pruned_loss=0.1069, over 5703052.36 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3711, pruned_loss=0.1197, over 5657209.19 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:05:59,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.787e+03 2.440e+03 3.144e+03 6.247e+03, threshold=4.881e+03, percent-clipped=4.0
+2023-03-13 01:06:00,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-13 01:06:08,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 01:06:13,961 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1118758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:06:36,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4200, 1.5254, 1.5979, 1.4141], device='cuda:0'), covar=tensor([0.2777, 0.2499, 0.2250, 0.2480], device='cuda:0'), in_proj_covar=tensor([0.2025, 0.1969, 0.1882, 0.2027], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 01:06:38,497 INFO [train.py:968] (0/2) Epoch 25, batch 24950, giga_loss[loss=0.3316, simple_loss=0.3805, pruned_loss=0.1414, over 27621.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3683, pruned_loss=0.1169, over 5680442.91 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.354, pruned_loss=0.1066, over 5709500.42 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3706, pruned_loss=0.1182, over 5670296.17 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:07:01,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:07:04,596 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6050, 4.6439, 1.6029, 1.7764], device='cuda:0'), covar=tensor([0.1037, 0.0409, 0.0998, 0.1357], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0566, 0.0401, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 01:07:14,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:07:28,763 INFO [train.py:968] (0/2) Epoch 25, batch 25000, giga_loss[loss=0.3721, simple_loss=0.4086, pruned_loss=0.1679, over 26788.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3685, pruned_loss=0.1171, over 5666754.46 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3545, pruned_loss=0.107, over 5709598.35 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3702, pruned_loss=0.1179, over 5657854.58 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:07:30,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2992, 2.5969, 2.0624, 2.0908], device='cuda:0'), covar=tensor([0.1025, 0.0708, 0.0946, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0451, 0.0523, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 01:07:30,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.161e+03 1.819e+03 2.115e+03 2.820e+03 9.724e+03, threshold=4.230e+03, percent-clipped=6.0
+2023-03-13 01:08:11,476 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-13 01:08:13,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:08:15,538 INFO [train.py:968] (0/2) Epoch 25, batch 25050, giga_loss[loss=0.3113, simple_loss=0.3742, pruned_loss=0.1242, over 27977.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3691, pruned_loss=0.1169, over 5670339.55 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3551, pruned_loss=0.1074, over 5702166.81 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3701, pruned_loss=0.1173, over 5668381.35 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:08:18,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:08:27,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6856, 2.4709, 1.8340, 0.8745], device='cuda:0'), covar=tensor([0.4644, 0.2908, 0.3561, 0.6021], device='cuda:0'), in_proj_covar=tensor([0.1809, 0.1699, 0.1638, 0.1466], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 01:08:47,201 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:08:56,602 INFO [train.py:968] (0/2) Epoch 25, batch 25100, giga_loss[loss=0.2683, simple_loss=0.3429, pruned_loss=0.09685, over 28951.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3683, pruned_loss=0.1171, over 5684976.77 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3552, pruned_loss=0.1078, over 5711286.80 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3696, pruned_loss=0.1175, over 5673934.90 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:09:00,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.738e+03 2.244e+03 2.944e+03 6.126e+03, threshold=4.489e+03, percent-clipped=6.0
+2023-03-13 01:09:14,017 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1118956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:16,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1118959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:29,002 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1118970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:30,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1118973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:37,135 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:43,335 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1118988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:45,329 INFO [train.py:968] (0/2) Epoch 25, batch 25150, giga_loss[loss=0.3325, simple_loss=0.3868, pruned_loss=0.1391, over 27905.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3667, pruned_loss=0.1165, over 5679778.74 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3549, pruned_loss=0.1077, over 5709484.69 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3685, pruned_loss=0.1173, over 5671869.34 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:09:57,899 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:23,131 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:25,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:27,396 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:29,045 INFO [train.py:968] (0/2) Epoch 25, batch 25200, giga_loss[loss=0.3277, simple_loss=0.3923, pruned_loss=0.1316, over 28925.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3665, pruned_loss=0.1171, over 5686854.36 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3553, pruned_loss=0.1081, over 5711492.57 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3681, pruned_loss=0.1177, over 5677568.86 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:10:29,325 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:33,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.835e+03 2.343e+03 3.268e+03 7.767e+03, threshold=4.686e+03, percent-clipped=10.0
+2023-03-13 01:10:52,452 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:56,227 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:57,046 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:59,163 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:11:16,620 INFO [train.py:968] (0/2) Epoch 25, batch 25250, giga_loss[loss=0.3493, simple_loss=0.4097, pruned_loss=0.1445, over 28618.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3645, pruned_loss=0.1159, over 5697203.37 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3551, pruned_loss=0.1083, over 5715170.70 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3662, pruned_loss=0.1166, over 5685835.04 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:11:18,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1119092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:11:19,347 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1119093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:11:28,721 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:11:51,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:11:54,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:12:04,192 INFO [train.py:968] (0/2) Epoch 25, batch 25300, giga_loss[loss=0.2512, simple_loss=0.3212, pruned_loss=0.09059, over 28631.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3617, pruned_loss=0.1145, over 5693466.45 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.355, pruned_loss=0.1083, over 5717115.27 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3632, pruned_loss=0.1151, over 5682648.61 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:12:05,348 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 01:12:07,613 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.880e+02 1.643e+03 2.227e+03 2.955e+03 8.692e+03, threshold=4.455e+03, percent-clipped=5.0
+2023-03-13 01:12:19,935 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:12:53,752 INFO [train.py:968] (0/2) Epoch 25, batch 25350, giga_loss[loss=0.2854, simple_loss=0.3567, pruned_loss=0.107, over 28583.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3626, pruned_loss=0.1159, over 5691405.35 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3554, pruned_loss=0.1086, over 5720475.43 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3637, pruned_loss=0.1163, over 5679083.72 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:13:35,382 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:13:37,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:13:38,957 INFO [train.py:968] (0/2) Epoch 25, batch 25400, giga_loss[loss=0.2968, simple_loss=0.3662, pruned_loss=0.1137, over 28943.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3629, pruned_loss=0.1156, over 5691910.56 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3552, pruned_loss=0.1084, over 5720145.96 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3642, pruned_loss=0.1162, over 5681693.52 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:13:40,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3283, 1.5334, 1.5962, 1.4444], device='cuda:0'), covar=tensor([0.1582, 0.1210, 0.1632, 0.1389], device='cuda:0'), in_proj_covar=tensor([0.0491, 0.0752, 0.0724, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 01:13:46,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.184e+03 1.802e+03 2.228e+03 3.056e+03 7.805e+03, threshold=4.455e+03, percent-clipped=9.0
+2023-03-13 01:14:05,403 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:14:26,063 INFO [train.py:968] (0/2) Epoch 25, batch 25450, giga_loss[loss=0.2884, simple_loss=0.3589, pruned_loss=0.109, over 28979.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3635, pruned_loss=0.1152, over 5693821.34 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3553, pruned_loss=0.1085, over 5723661.51 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3646, pruned_loss=0.1158, over 5682168.98 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:14:56,536 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3637, 1.9395, 1.4415, 0.6411], device='cuda:0'), covar=tensor([0.5749, 0.2811, 0.3740, 0.6637], device='cuda:0'), in_proj_covar=tensor([0.1813, 0.1701, 0.1641, 0.1467], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 01:15:13,247 INFO [train.py:968] (0/2) Epoch 25, batch 25500, giga_loss[loss=0.2536, simple_loss=0.3382, pruned_loss=0.08455, over 29042.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.364, pruned_loss=0.1148, over 5693384.54 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3556, pruned_loss=0.1087, over 5722330.91 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3646, pruned_loss=0.1151, over 5684866.22 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:15:17,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.664e+03 2.130e+03 3.522e+03 1.510e+04, threshold=4.261e+03, percent-clipped=12.0
+2023-03-13 01:15:58,656 INFO [train.py:968] (0/2) Epoch 25, batch 25550, giga_loss[loss=0.3349, simple_loss=0.3893, pruned_loss=0.1403, over 29020.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.365, pruned_loss=0.1153, over 5684499.71 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3558, pruned_loss=0.1089, over 5716957.91 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3655, pruned_loss=0.1154, over 5681505.64 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:16:33,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8728, 2.0368, 1.5586, 1.6786], device='cuda:0'), covar=tensor([0.0785, 0.0445, 0.0845, 0.0949], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0451, 0.0522, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 01:16:44,792 INFO [train.py:968] (0/2) Epoch 25, batch 25600, giga_loss[loss=0.344, simple_loss=0.4014, pruned_loss=0.1433, over 28289.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 5684361.60 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3563, pruned_loss=0.1093, over 5715563.79 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3684, pruned_loss=0.1181, over 5682552.38 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:16:51,044 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.826e+03 2.550e+03 4.167e+03 2.835e+04, threshold=5.100e+03, percent-clipped=22.0
+2023-03-13 01:17:12,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1119468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:17:33,075 INFO [train.py:968] (0/2) Epoch 25, batch 25650, giga_loss[loss=0.2703, simple_loss=0.3473, pruned_loss=0.09665, over 28961.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3703, pruned_loss=0.1211, over 5684272.97 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3564, pruned_loss=0.1093, over 5720465.67 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3707, pruned_loss=0.1213, over 5677414.37 frames. ], batch size: 164, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:18:22,188 INFO [train.py:968] (0/2) Epoch 25, batch 25700, giga_loss[loss=0.2774, simple_loss=0.3479, pruned_loss=0.1034, over 28909.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3714, pruned_loss=0.1235, over 5678306.52 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3562, pruned_loss=0.1092, over 5721961.22 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3721, pruned_loss=0.1241, over 5670963.28 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:18:27,742 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.207e+03 2.177e+03 3.140e+03 4.771e+03 1.215e+04, threshold=6.280e+03, percent-clipped=22.0
+2023-03-13 01:19:10,615 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1119587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:19:12,409 INFO [train.py:968] (0/2) Epoch 25, batch 25750, giga_loss[loss=0.2823, simple_loss=0.3508, pruned_loss=0.1069, over 28309.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3721, pruned_loss=0.1244, over 5684761.02 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3564, pruned_loss=0.1093, over 5722829.05 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3726, pruned_loss=0.1248, over 5677857.69 frames. ], batch size: 77, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:19:28,960 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:19:30,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:19:54,928 INFO [train.py:968] (0/2) Epoch 25, batch 25800, giga_loss[loss=0.3649, simple_loss=0.4069, pruned_loss=0.1614, over 28573.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3694, pruned_loss=0.1225, over 5674344.68 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3564, pruned_loss=0.1093, over 5723825.04 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3702, pruned_loss=0.1232, over 5667048.75 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:19:57,434 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:19:58,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.170e+03 1.652e+03 2.337e+03 3.025e+03 7.137e+03, threshold=4.675e+03, percent-clipped=1.0
+2023-03-13 01:20:43,262 INFO [train.py:968] (0/2) Epoch 25, batch 25850, giga_loss[loss=0.283, simple_loss=0.3652, pruned_loss=0.1004, over 28991.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3685, pruned_loss=0.1207, over 5680841.11 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3564, pruned_loss=0.1094, over 5724853.83 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3693, pruned_loss=0.1213, over 5673440.73 frames. ], batch size: 164, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:20:55,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5447, 2.1711, 1.5602, 0.8110], device='cuda:0'), covar=tensor([0.6866, 0.4333, 0.4326, 0.6986], device='cuda:0'), in_proj_covar=tensor([0.1820, 0.1712, 0.1647, 0.1474], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 01:21:12,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3306, 3.1127, 2.9719, 1.5339], device='cuda:0'), covar=tensor([0.1045, 0.1274, 0.1013, 0.2236], device='cuda:0'), in_proj_covar=tensor([0.1294, 0.1191, 0.1006, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 01:21:28,213 INFO [train.py:968] (0/2) Epoch 25, batch 25900, giga_loss[loss=0.245, simple_loss=0.3021, pruned_loss=0.09397, over 23825.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3681, pruned_loss=0.1194, over 5668105.05 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3566, pruned_loss=0.1095, over 5722522.33 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3686, pruned_loss=0.1199, over 5663978.21 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:21:29,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3787, 1.6705, 1.4774, 1.5271], device='cuda:0'), covar=tensor([0.0801, 0.0330, 0.0328, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 01:21:33,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.955e+02 1.770e+03 2.238e+03 3.128e+03 7.147e+03, threshold=4.477e+03, percent-clipped=11.0
+2023-03-13 01:22:12,663 INFO [train.py:968] (0/2) Epoch 25, batch 25950, giga_loss[loss=0.351, simple_loss=0.3853, pruned_loss=0.1583, over 26414.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3664, pruned_loss=0.1184, over 5667048.16 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3563, pruned_loss=0.1093, over 5720365.04 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3676, pruned_loss=0.1194, over 5662770.41 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:22:56,602 INFO [train.py:968] (0/2) Epoch 25, batch 26000, giga_loss[loss=0.2789, simple_loss=0.3482, pruned_loss=0.1048, over 28984.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3651, pruned_loss=0.1178, over 5672953.36 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3569, pruned_loss=0.1098, over 5720517.13 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3658, pruned_loss=0.1185, over 5667606.94 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:23:00,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.196e+03 1.634e+03 2.146e+03 2.713e+03 5.706e+03, threshold=4.291e+03, percent-clipped=1.0
+2023-03-13 01:23:42,785 INFO [train.py:968] (0/2) Epoch 25, batch 26050, giga_loss[loss=0.2998, simple_loss=0.3626, pruned_loss=0.1184, over 28947.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3666, pruned_loss=0.1201, over 5650440.78 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3571, pruned_loss=0.1099, over 5714705.22 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3673, pruned_loss=0.1208, over 5649307.54 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:24:20,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4243, 1.2798, 4.2053, 3.4691], device='cuda:0'), covar=tensor([0.1660, 0.2929, 0.0461, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0668, 0.0990, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 01:24:30,025 INFO [train.py:968] (0/2) Epoch 25, batch 26100, giga_loss[loss=0.324, simple_loss=0.3832, pruned_loss=0.1324, over 28643.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3688, pruned_loss=0.1214, over 5656868.90 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3568, pruned_loss=0.1097, over 5717633.61 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3697, pruned_loss=0.1223, over 5652636.37 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:24:35,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.788e+03 2.167e+03 2.815e+03 6.115e+03, threshold=4.333e+03, percent-clipped=7.0
+2023-03-13 01:24:49,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1119962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:25:14,226 INFO [train.py:968] (0/2) Epoch 25, batch 26150, giga_loss[loss=0.2954, simple_loss=0.3814, pruned_loss=0.1047, over 28898.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3733, pruned_loss=0.1224, over 5653286.66 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3572, pruned_loss=0.1101, over 5709228.63 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.374, pruned_loss=0.123, over 5655759.61 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:25:23,402 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1120000.pt
+2023-03-13 01:25:59,401 INFO [train.py:968] (0/2) Epoch 25, batch 26200, giga_loss[loss=0.3047, simple_loss=0.3762, pruned_loss=0.1166, over 28727.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3749, pruned_loss=0.1214, over 5663704.11 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3576, pruned_loss=0.1105, over 5712226.31 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3756, pruned_loss=0.1219, over 5661694.63 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:26:05,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.414e+02 1.690e+03 2.128e+03 2.971e+03 9.739e+03, threshold=4.256e+03, percent-clipped=7.0
+2023-03-13 01:26:43,971 INFO [train.py:968] (0/2) Epoch 25, batch 26250, giga_loss[loss=0.3979, simple_loss=0.4341, pruned_loss=0.1809, over 27676.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3759, pruned_loss=0.1226, over 5663411.41 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3579, pruned_loss=0.1109, over 5716526.43 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3768, pruned_loss=0.123, over 5656384.49 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:26:57,055 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4571, 1.8603, 1.2645, 0.9564], device='cuda:0'), covar=tensor([0.5372, 0.3412, 0.2837, 0.5483], device='cuda:0'), in_proj_covar=tensor([0.1812, 0.1702, 0.1640, 0.1469], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 01:27:00,412 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1120105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:27:02,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1120108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:27:10,615 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3827, 1.4825, 1.5557, 1.3006], device='cuda:0'), covar=tensor([0.3160, 0.2677, 0.2106, 0.2763], device='cuda:0'), in_proj_covar=tensor([0.2033, 0.1975, 0.1887, 0.2034], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 01:27:26,797 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1120137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:27:29,242 INFO [train.py:968] (0/2) Epoch 25, batch 26300, libri_loss[loss=0.251, simple_loss=0.3173, pruned_loss=0.09239, over 29340.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3776, pruned_loss=0.1249, over 5642536.39 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3582, pruned_loss=0.1113, over 5701890.81 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3786, pruned_loss=0.1252, over 5649207.08 frames. ], batch size: 67, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:27:33,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.825e+03 2.479e+03 3.298e+03 8.507e+03, threshold=4.957e+03, percent-clipped=12.0
+2023-03-13 01:27:46,497 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-13 01:28:16,037 INFO [train.py:968] (0/2) Epoch 25, batch 26350, giga_loss[loss=0.299, simple_loss=0.3597, pruned_loss=0.1192, over 28789.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3761, pruned_loss=0.1244, over 5639891.00 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3577, pruned_loss=0.1109, over 5705739.89 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3776, pruned_loss=0.1252, over 5640590.82 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:29:04,943 INFO [train.py:968] (0/2) Epoch 25, batch 26400, giga_loss[loss=0.2785, simple_loss=0.3457, pruned_loss=0.1056, over 28728.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3754, pruned_loss=0.1247, over 5637963.34 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.358, pruned_loss=0.1112, over 5704476.53 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3767, pruned_loss=0.1254, over 5638735.70 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:29:10,210 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.610e+03 1.984e+03 2.742e+03 6.247e+03, threshold=3.969e+03, percent-clipped=3.0
+2023-03-13 01:29:50,287 INFO [train.py:968] (0/2) Epoch 25, batch 26450, giga_loss[loss=0.374, simple_loss=0.4055, pruned_loss=0.1713, over 26739.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3733, pruned_loss=0.1236, over 5647551.69 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1114, over 5707241.42 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3741, pruned_loss=0.1241, over 5644307.15 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:29:56,937 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-13 01:30:37,030 INFO [train.py:968] (0/2) Epoch 25, batch 26500, giga_loss[loss=0.2607, simple_loss=0.3321, pruned_loss=0.09463, over 28577.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3721, pruned_loss=0.1235, over 5650609.22 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3586, pruned_loss=0.1114, over 5710020.84 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3731, pruned_loss=0.1243, over 5643887.38 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:30:43,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.364e+02 1.668e+03 2.357e+03 3.303e+03 9.755e+03, threshold=4.715e+03, percent-clipped=13.0
+2023-03-13 01:31:18,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0047, 1.3234, 1.0793, 0.2675], device='cuda:0'), covar=tensor([0.4544, 0.3565, 0.4980, 0.6981], device='cuda:0'), in_proj_covar=tensor([0.1814, 0.1704, 0.1640, 0.1469], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 01:31:24,325 INFO [train.py:968] (0/2) Epoch 25, batch 26550, giga_loss[loss=0.2819, simple_loss=0.3443, pruned_loss=0.1098, over 28748.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3717, pruned_loss=0.1236, over 5651405.82 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3586, pruned_loss=0.1114, over 5714123.02 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3728, pruned_loss=0.1245, over 5641094.60 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:31:48,932 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1120417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:32:04,317 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-13 01:32:08,675 INFO [train.py:968] (0/2) Epoch 25, batch 26600, giga_loss[loss=0.3117, simple_loss=0.3688, pruned_loss=0.1273, over 28789.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3707, pruned_loss=0.123, over 5659569.23 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3587, pruned_loss=0.1115, over 5716474.30 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3717, pruned_loss=0.1238, over 5648152.19 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:32:13,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.238e+03 2.110e+03 3.005e+03 4.517e+03 1.052e+04, threshold=6.011e+03, percent-clipped=25.0
+2023-03-13 01:32:43,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7929, 1.8714, 1.7242, 1.5816], device='cuda:0'), covar=tensor([0.2136, 0.2838, 0.2626, 0.2772], device='cuda:0'), in_proj_covar=tensor([0.0493, 0.0758, 0.0726, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 01:32:52,562 INFO [train.py:968] (0/2) Epoch 25, batch 26650, giga_loss[loss=0.3041, simple_loss=0.3705, pruned_loss=0.1189, over 28401.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3685, pruned_loss=0.1216, over 5673701.95 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3588, pruned_loss=0.1118, over 5720340.63 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3693, pruned_loss=0.1222, over 5660371.92 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:33:38,559 INFO [train.py:968] (0/2) Epoch 25, batch 26700, libri_loss[loss=0.271, simple_loss=0.3487, pruned_loss=0.09665, over 27965.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3684, pruned_loss=0.1213, over 5670177.95 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3593, pruned_loss=0.1119, over 5712392.45 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.369, pruned_loss=0.122, over 5663810.05 frames. ], batch size: 116, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:33:44,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.552e+02 1.773e+03 2.495e+03 3.297e+03 8.020e+03, threshold=4.989e+03, percent-clipped=2.0
+2023-03-13 01:34:12,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3280, 1.4760, 1.3534, 1.5020], device='cuda:0'), covar=tensor([0.0730, 0.0413, 0.0335, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 01:34:25,393 INFO [train.py:968] (0/2) Epoch 25, batch 26750, giga_loss[loss=0.4112, simple_loss=0.4376, pruned_loss=0.1924, over 26644.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3695, pruned_loss=0.1213, over 5664667.35 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3593, pruned_loss=0.1118, over 5713321.68 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3701, pruned_loss=0.122, over 5658546.78 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:34:47,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1824, 3.4733, 2.1765, 1.3275], device='cuda:0'), covar=tensor([0.7703, 0.3159, 0.4149, 0.6756], device='cuda:0'), in_proj_covar=tensor([0.1809, 0.1699, 0.1636, 0.1466], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 01:34:59,710 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-13 01:35:10,309 INFO [train.py:968] (0/2) Epoch 25, batch 26800, libri_loss[loss=0.2408, simple_loss=0.3085, pruned_loss=0.0865, over 29533.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3697, pruned_loss=0.1209, over 5673778.43 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3583, pruned_loss=0.1114, over 5720217.67 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3715, pruned_loss=0.1223, over 5660852.11 frames. ], batch size: 70, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:35:18,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.667e+03 2.077e+03 2.529e+03 1.155e+04, threshold=4.155e+03, percent-clipped=1.0
+2023-03-13 01:35:23,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1963, 1.2451, 3.4318, 3.0457], device='cuda:0'), covar=tensor([0.1621, 0.2736, 0.0462, 0.1645], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0669, 0.0992, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 01:35:56,824 INFO [train.py:968] (0/2) Epoch 25, batch 26850, giga_loss[loss=0.3242, simple_loss=0.3813, pruned_loss=0.1336, over 28625.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3717, pruned_loss=0.1235, over 5664153.28 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3584, pruned_loss=0.1113, over 5719597.20 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3734, pruned_loss=0.1249, over 5653284.87 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:36:40,835 INFO [train.py:968] (0/2) Epoch 25, batch 26900, giga_loss[loss=0.3139, simple_loss=0.4031, pruned_loss=0.1123, over 28945.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3725, pruned_loss=0.121, over 5670806.87 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3583, pruned_loss=0.1114, over 5712032.93 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3742, pruned_loss=0.1223, over 5667330.08 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:36:48,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.626e+03 2.155e+03 2.773e+03 8.571e+03, threshold=4.310e+03, percent-clipped=3.0
+2023-03-13 01:37:28,036 INFO [train.py:968] (0/2) Epoch 25, batch 26950, giga_loss[loss=0.3707, simple_loss=0.4025, pruned_loss=0.1695, over 23738.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3746, pruned_loss=0.1211, over 5671624.41 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3586, pruned_loss=0.1116, over 5715967.14 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3759, pruned_loss=0.122, over 5664748.32 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:37:30,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1120792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:37:42,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2703, 1.7807, 1.3406, 0.4867], device='cuda:0'), covar=tensor([0.4779, 0.3229, 0.4449, 0.6600], device='cuda:0'), in_proj_covar=tensor([0.1807, 0.1697, 0.1633, 0.1464], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 01:38:10,838 INFO [train.py:968] (0/2) Epoch 25, batch 27000, giga_loss[loss=0.3342, simple_loss=0.4066, pruned_loss=0.1309, over 28869.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3772, pruned_loss=0.1215, over 5674484.92 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3589, pruned_loss=0.1118, over 5718673.61 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3782, pruned_loss=0.1223, over 5666189.07 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:38:10,842 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 01:38:19,495 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2037, simple_loss=0.311, pruned_loss=0.04819, over 944034.00 frames. 
+2023-03-13 01:38:19,495 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 01:38:25,721 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.034e+02 1.578e+03 1.989e+03 2.750e+03 6.546e+03, threshold=3.979e+03, percent-clipped=7.0
+2023-03-13 01:39:06,918 INFO [train.py:968] (0/2) Epoch 25, batch 27050, giga_loss[loss=0.4092, simple_loss=0.4341, pruned_loss=0.1922, over 26548.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3791, pruned_loss=0.1244, over 5672786.67 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3584, pruned_loss=0.1115, over 5722391.98 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3808, pruned_loss=0.1255, over 5661922.47 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:39:43,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6251, 1.9190, 1.5176, 2.0264], device='cuda:0'), covar=tensor([0.2529, 0.2714, 0.3013, 0.2387], device='cuda:0'), in_proj_covar=tensor([0.1556, 0.1121, 0.1374, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 01:39:53,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1120935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:39:55,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1120938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:39:57,462 INFO [train.py:968] (0/2) Epoch 25, batch 27100, giga_loss[loss=0.3304, simple_loss=0.3831, pruned_loss=0.1389, over 27531.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3801, pruned_loss=0.1259, over 5680922.10 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3587, pruned_loss=0.1118, over 5722818.59 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3814, pruned_loss=0.1266, over 5671306.58 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:40:09,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.750e+03 2.259e+03 3.122e+03 9.929e+03, threshold=4.519e+03, percent-clipped=20.0
+2023-03-13 01:40:26,169 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1120967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:40:50,742 INFO [train.py:968] (0/2) Epoch 25, batch 27150, giga_loss[loss=0.3544, simple_loss=0.3869, pruned_loss=0.1609, over 23516.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.38, pruned_loss=0.1265, over 5673516.56 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3587, pruned_loss=0.1117, over 5723744.49 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3812, pruned_loss=0.1272, over 5665100.23 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:41:35,690 INFO [train.py:968] (0/2) Epoch 25, batch 27200, giga_loss[loss=0.3339, simple_loss=0.3873, pruned_loss=0.1402, over 27549.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3772, pruned_loss=0.1238, over 5679764.16 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3591, pruned_loss=0.112, over 5723632.79 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3784, pruned_loss=0.1246, over 5671378.35 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:41:40,395 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:41:40,686 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-13 01:41:45,046 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.729e+03 2.401e+03 3.261e+03 5.594e+03, threshold=4.802e+03, percent-clipped=4.0
+2023-03-13 01:41:53,185 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9462, 1.2166, 5.1839, 3.6007], device='cuda:0'), covar=tensor([0.1640, 0.3194, 0.0473, 0.0916], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0669, 0.0994, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 01:42:21,100 INFO [train.py:968] (0/2) Epoch 25, batch 27250, giga_loss[loss=0.2714, simple_loss=0.3626, pruned_loss=0.09014, over 29010.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3757, pruned_loss=0.1212, over 5668348.36 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.359, pruned_loss=0.112, over 5717745.16 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3772, pruned_loss=0.1221, over 5664972.49 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:43:06,121 INFO [train.py:968] (0/2) Epoch 25, batch 27300, libri_loss[loss=0.2363, simple_loss=0.307, pruned_loss=0.08282, over 29687.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3761, pruned_loss=0.1213, over 5665339.23 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3585, pruned_loss=0.1117, over 5720044.87 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3783, pruned_loss=0.1225, over 5658989.02 frames. ], batch size: 73, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:43:14,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.664e+02 1.644e+03 2.163e+03 3.082e+03 8.261e+03, threshold=4.326e+03, percent-clipped=7.0
+2023-03-13 01:43:41,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0389, 5.8665, 5.5665, 2.7907], device='cuda:0'), covar=tensor([0.0499, 0.0628, 0.0747, 0.1796], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.1194, 0.1006, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 01:43:53,791 INFO [train.py:968] (0/2) Epoch 25, batch 27350, giga_loss[loss=0.3819, simple_loss=0.4264, pruned_loss=0.1687, over 27965.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3774, pruned_loss=0.1224, over 5660893.53 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1118, over 5720808.22 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3794, pruned_loss=0.1235, over 5654183.11 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:44:15,606 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 01:44:22,449 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 01:44:39,300 INFO [train.py:968] (0/2) Epoch 25, batch 27400, giga_loss[loss=0.3146, simple_loss=0.3784, pruned_loss=0.1254, over 28615.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3759, pruned_loss=0.1217, over 5663754.13 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1118, over 5724445.64 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3778, pruned_loss=0.1228, over 5654225.68 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:44:46,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.241e+03 1.784e+03 2.203e+03 3.266e+03 5.663e+03, threshold=4.406e+03, percent-clipped=9.0
+2023-03-13 01:45:08,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2842, 1.2203, 3.9328, 3.1791], device='cuda:0'), covar=tensor([0.1820, 0.2902, 0.0486, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0669, 0.0995, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 01:45:21,299 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:45:28,755 INFO [train.py:968] (0/2) Epoch 25, batch 27450, giga_loss[loss=0.3131, simple_loss=0.3764, pruned_loss=0.1249, over 28574.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3754, pruned_loss=0.1223, over 5669212.62 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3586, pruned_loss=0.1117, over 5728346.45 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3773, pruned_loss=0.1235, over 5656574.26 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:46:13,195 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:46:13,662 INFO [train.py:968] (0/2) Epoch 25, batch 27500, giga_loss[loss=0.378, simple_loss=0.4153, pruned_loss=0.1704, over 26710.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3742, pruned_loss=0.1222, over 5666899.63 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3587, pruned_loss=0.1119, over 5721303.92 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3762, pruned_loss=0.1233, over 5661530.98 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:46:19,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:46:23,311 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.847e+03 2.344e+03 2.935e+03 5.935e+03, threshold=4.689e+03, percent-clipped=2.0
+2023-03-13 01:47:01,320 INFO [train.py:968] (0/2) Epoch 25, batch 27550, giga_loss[loss=0.3116, simple_loss=0.3851, pruned_loss=0.119, over 28519.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3718, pruned_loss=0.121, over 5673023.71 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3584, pruned_loss=0.1118, over 5725695.34 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3741, pruned_loss=0.1223, over 5663336.46 frames. ], batch size: 60, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:47:30,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1121419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:47:54,067 INFO [train.py:968] (0/2) Epoch 25, batch 27600, giga_loss[loss=0.3146, simple_loss=0.3554, pruned_loss=0.1369, over 23273.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3717, pruned_loss=0.1222, over 5664893.72 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3585, pruned_loss=0.1118, over 5726613.86 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3735, pruned_loss=0.1232, over 5656256.86 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:47:54,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4588, 1.0846, 4.2434, 3.3931], device='cuda:0'), covar=tensor([0.1774, 0.3151, 0.0493, 0.0866], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0667, 0.0994, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 01:48:03,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.581e+02 1.739e+03 2.388e+03 3.003e+03 6.189e+03, threshold=4.776e+03, percent-clipped=3.0
+2023-03-13 01:48:38,277 INFO [train.py:968] (0/2) Epoch 25, batch 27650, giga_loss[loss=0.2876, simple_loss=0.3644, pruned_loss=0.1054, over 28957.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3712, pruned_loss=0.1223, over 5653574.25 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3588, pruned_loss=0.112, over 5715959.63 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3726, pruned_loss=0.1231, over 5654826.02 frames. ], batch size: 164, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:48:54,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8360, 2.0885, 1.5989, 1.7182], device='cuda:0'), covar=tensor([0.0824, 0.0444, 0.0852, 0.0954], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0454, 0.0523, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 01:49:21,539 INFO [train.py:968] (0/2) Epoch 25, batch 27700, giga_loss[loss=0.3386, simple_loss=0.3898, pruned_loss=0.1437, over 26562.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3685, pruned_loss=0.119, over 5660392.35 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3585, pruned_loss=0.1118, over 5717181.89 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3701, pruned_loss=0.12, over 5659073.90 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:49:32,177 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.736e+02 1.632e+03 2.024e+03 2.626e+03 7.815e+03, threshold=4.049e+03, percent-clipped=4.0
+2023-03-13 01:49:41,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9616, 1.3401, 5.1914, 3.7148], device='cuda:0'), covar=tensor([0.1560, 0.2869, 0.0467, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0666, 0.0993, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 01:49:41,972 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1121562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:49:44,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1121565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:49:44,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7486, 2.2668, 1.4741, 1.0589], device='cuda:0'), covar=tensor([0.7792, 0.3866, 0.3880, 0.6887], device='cuda:0'), in_proj_covar=tensor([0.1811, 0.1703, 0.1636, 0.1468], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 01:50:02,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1121587.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 01:50:05,426 INFO [train.py:968] (0/2) Epoch 25, batch 27750, giga_loss[loss=0.2764, simple_loss=0.3537, pruned_loss=0.09951, over 28853.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3653, pruned_loss=0.1161, over 5661472.72 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3588, pruned_loss=0.1122, over 5714421.67 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3666, pruned_loss=0.1168, over 5661301.33 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:50:08,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1121594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:50:57,203 INFO [train.py:968] (0/2) Epoch 25, batch 27800, giga_loss[loss=0.293, simple_loss=0.3579, pruned_loss=0.1141, over 29052.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3639, pruned_loss=0.1147, over 5658484.98 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.359, pruned_loss=0.1121, over 5718350.40 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3649, pruned_loss=0.1153, over 5654191.09 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:51:06,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.650e+02 1.447e+03 1.908e+03 2.486e+03 5.770e+03, threshold=3.816e+03, percent-clipped=7.0
+2023-03-13 01:51:11,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1121656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:51:12,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 01:51:18,350 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-13 01:51:44,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6157, 1.9211, 1.8435, 1.4995], device='cuda:0'), covar=tensor([0.2458, 0.1941, 0.1531, 0.2104], device='cuda:0'), in_proj_covar=tensor([0.2030, 0.1973, 0.1890, 0.2033], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 01:51:47,499 INFO [train.py:968] (0/2) Epoch 25, batch 27850, giga_loss[loss=0.2681, simple_loss=0.3357, pruned_loss=0.1002, over 28857.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3628, pruned_loss=0.1152, over 5639266.56 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3593, pruned_loss=0.1124, over 5702608.64 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3633, pruned_loss=0.1155, over 5648566.94 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:51:52,199 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:52:13,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1121714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:52:19,808 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1121721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:52:30,079 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1121730.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 01:52:33,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1121733.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 01:52:38,905 INFO [train.py:968] (0/2) Epoch 25, batch 27900, giga_loss[loss=0.3977, simple_loss=0.4292, pruned_loss=0.1831, over 27509.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3618, pruned_loss=0.1153, over 5646185.14 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5705915.20 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.362, pruned_loss=0.1154, over 5649296.59 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:52:49,319 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.178e+03 1.984e+03 2.379e+03 3.432e+03 7.455e+03, threshold=4.759e+03, percent-clipped=13.0
+2023-03-13 01:52:57,681 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1121762.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 01:53:14,417 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 01:53:23,954 INFO [train.py:968] (0/2) Epoch 25, batch 27950, giga_loss[loss=0.3203, simple_loss=0.3645, pruned_loss=0.1381, over 23457.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1156, over 5652700.57 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5709660.10 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3639, pruned_loss=0.1158, over 5651054.76 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:53:34,759 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1121799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:53:38,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1121802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:02,685 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1121831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:10,463 INFO [train.py:968] (0/2) Epoch 25, batch 28000, giga_loss[loss=0.3526, simple_loss=0.4065, pruned_loss=0.1494, over 27525.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3662, pruned_loss=0.117, over 5642473.00 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3597, pruned_loss=0.1126, over 5705075.25 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3664, pruned_loss=0.1171, over 5643648.54 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:54:22,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.536e+02 1.558e+03 2.068e+03 3.012e+03 5.818e+03, threshold=4.136e+03, percent-clipped=4.0
+2023-03-13 01:54:22,460 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:28,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1121857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:31,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1121860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:35,666 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1121864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:37,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1121867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:39,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7109, 1.7446, 1.8379, 1.6834], device='cuda:0'), covar=tensor([0.2446, 0.2600, 0.2077, 0.2217], device='cuda:0'), in_proj_covar=tensor([0.2036, 0.1981, 0.1895, 0.2039], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 01:54:58,627 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1121889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:59,026 INFO [train.py:968] (0/2) Epoch 25, batch 28050, giga_loss[loss=0.2841, simple_loss=0.3556, pruned_loss=0.1063, over 28926.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3663, pruned_loss=0.1171, over 5643343.51 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1127, over 5705906.87 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3663, pruned_loss=0.1172, over 5643195.52 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:55:04,892 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1121896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:55:05,078 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.46 vs. limit=5.0
+2023-03-13 01:55:45,345 INFO [train.py:968] (0/2) Epoch 25, batch 28100, giga_loss[loss=0.2564, simple_loss=0.3372, pruned_loss=0.08778, over 29055.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3664, pruned_loss=0.1177, over 5642191.67 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3597, pruned_loss=0.1125, over 5705976.39 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3667, pruned_loss=0.118, over 5641200.86 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:55:52,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.207e+03 1.618e+03 2.064e+03 2.766e+03 1.110e+04, threshold=4.128e+03, percent-clipped=6.0
+2023-03-13 01:56:08,212 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-13 01:56:26,034 INFO [train.py:968] (0/2) Epoch 25, batch 28150, libri_loss[loss=0.3326, simple_loss=0.3931, pruned_loss=0.1361, over 29512.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3685, pruned_loss=0.1195, over 5636167.10 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3606, pruned_loss=0.1131, over 5703354.17 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3681, pruned_loss=0.1194, over 5636235.94 frames. ], batch size: 89, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:56:33,664 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1122000.pt
+2023-03-13 01:56:36,959 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 01:57:11,602 INFO [train.py:968] (0/2) Epoch 25, batch 28200, giga_loss[loss=0.3079, simple_loss=0.3723, pruned_loss=0.1218, over 28793.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3703, pruned_loss=0.1205, over 5646046.90 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3608, pruned_loss=0.1132, over 5703089.97 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 5645188.61 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:57:24,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.491e+02 1.857e+03 2.441e+03 3.358e+03 1.266e+04, threshold=4.881e+03, percent-clipped=13.0
+2023-03-13 01:57:41,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1122070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:57:59,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4493, 1.7528, 1.6939, 1.4511], device='cuda:0'), covar=tensor([0.2625, 0.2152, 0.1620, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.2035, 0.1982, 0.1898, 0.2042], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 01:58:02,226 INFO [train.py:968] (0/2) Epoch 25, batch 28250, giga_loss[loss=0.2796, simple_loss=0.35, pruned_loss=0.1046, over 28848.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3721, pruned_loss=0.1223, over 5646212.89 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3607, pruned_loss=0.1131, over 5704153.52 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3719, pruned_loss=0.1223, over 5644203.35 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:58:37,870 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1122128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:58:49,361 INFO [train.py:968] (0/2) Epoch 25, batch 28300, giga_loss[loss=0.3442, simple_loss=0.382, pruned_loss=0.1532, over 23602.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.373, pruned_loss=0.1236, over 5639376.41 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3603, pruned_loss=0.1128, over 5697852.88 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3735, pruned_loss=0.1241, over 5641869.84 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:58:51,291 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1241, 1.1535, 3.6573, 3.1846], device='cuda:0'), covar=tensor([0.1779, 0.2862, 0.0519, 0.0974], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0666, 0.0989, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 01:59:03,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.009e+03 1.822e+03 2.182e+03 2.792e+03 5.582e+03, threshold=4.363e+03, percent-clipped=3.0
+2023-03-13 01:59:40,829 INFO [train.py:968] (0/2) Epoch 25, batch 28350, giga_loss[loss=0.2828, simple_loss=0.3617, pruned_loss=0.102, over 29018.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3721, pruned_loss=0.1215, over 5647392.59 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3602, pruned_loss=0.1127, over 5699889.97 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3728, pruned_loss=0.1221, over 5646857.97 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:00:05,379 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1122213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:00:08,064 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1122216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:00:19,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1122226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:00:26,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5433, 1.7919, 1.4911, 1.5354], device='cuda:0'), covar=tensor([0.2564, 0.2539, 0.2804, 0.2342], device='cuda:0'), in_proj_covar=tensor([0.1564, 0.1126, 0.1381, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 02:00:32,465 INFO [train.py:968] (0/2) Epoch 25, batch 28400, giga_loss[loss=0.4129, simple_loss=0.4433, pruned_loss=0.1913, over 26755.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3719, pruned_loss=0.1206, over 5639538.46 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3602, pruned_loss=0.1127, over 5692173.90 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3725, pruned_loss=0.1212, over 5645875.84 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:00:37,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1122245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:00:42,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.279e+03 1.958e+03 2.717e+03 3.674e+03 1.745e+04, threshold=5.435e+03, percent-clipped=19.0
+2023-03-13 02:01:16,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6821, 1.3513, 4.8349, 3.5885], device='cuda:0'), covar=tensor([0.1629, 0.2858, 0.0428, 0.0907], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0666, 0.0989, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 02:01:21,399 INFO [train.py:968] (0/2) Epoch 25, batch 28450, giga_loss[loss=0.2903, simple_loss=0.3613, pruned_loss=0.1096, over 28968.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3716, pruned_loss=0.1214, over 5624807.81 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3601, pruned_loss=0.1126, over 5692100.72 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3723, pruned_loss=0.122, over 5629531.56 frames. ], batch size: 164, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:02:11,947 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1122339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:02:14,402 INFO [train.py:968] (0/2) Epoch 25, batch 28500, giga_loss[loss=0.3377, simple_loss=0.3914, pruned_loss=0.142, over 27507.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3708, pruned_loss=0.1216, over 5631457.12 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3597, pruned_loss=0.1124, over 5697330.23 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3721, pruned_loss=0.1226, over 5627922.05 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:02:24,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.991e+02 1.800e+03 2.414e+03 3.319e+03 6.615e+03, threshold=4.827e+03, percent-clipped=3.0
+2023-03-13 02:02:46,431 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1122369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:02:49,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1122372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:03:10,513 INFO [train.py:968] (0/2) Epoch 25, batch 28550, giga_loss[loss=0.2631, simple_loss=0.3358, pruned_loss=0.09518, over 28850.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.369, pruned_loss=0.1214, over 5614689.80 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3594, pruned_loss=0.1123, over 5690124.30 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5616909.55 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:03:20,468 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1122401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:03:54,036 INFO [train.py:968] (0/2) Epoch 25, batch 28600, giga_loss[loss=0.3137, simple_loss=0.3781, pruned_loss=0.1247, over 28945.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.368, pruned_loss=0.1209, over 5628901.90 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3596, pruned_loss=0.1124, over 5687627.48 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3693, pruned_loss=0.1219, over 5631341.00 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:03:57,310 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3350, 1.5012, 1.4349, 1.2868], device='cuda:0'), covar=tensor([0.2962, 0.2816, 0.2190, 0.2578], device='cuda:0'), in_proj_covar=tensor([0.2034, 0.1984, 0.1894, 0.2036], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 02:03:59,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2931, 1.8858, 5.5007, 3.9378], device='cuda:0'), covar=tensor([0.1411, 0.2427, 0.0416, 0.0774], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0667, 0.0991, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 02:04:07,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.785e+03 2.366e+03 3.217e+03 8.366e+03, threshold=4.733e+03, percent-clipped=9.0
+2023-03-13 02:04:42,646 INFO [train.py:968] (0/2) Epoch 25, batch 28650, giga_loss[loss=0.2775, simple_loss=0.351, pruned_loss=0.102, over 28977.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3678, pruned_loss=0.1208, over 5642404.73 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3595, pruned_loss=0.1123, over 5690647.54 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.369, pruned_loss=0.1218, over 5640827.59 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:04:55,421 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1122503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:05:26,193 INFO [train.py:968] (0/2) Epoch 25, batch 28700, giga_loss[loss=0.3298, simple_loss=0.3903, pruned_loss=0.1347, over 28846.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3673, pruned_loss=0.1203, over 5655141.45 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3596, pruned_loss=0.1124, over 5699291.12 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3684, pruned_loss=0.1213, over 5644510.21 frames. ], batch size: 285, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:05:27,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6912, 1.9362, 1.3627, 1.4545], device='cuda:0'), covar=tensor([0.1015, 0.0628, 0.1089, 0.1127], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0451, 0.0519, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 02:05:33,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1122549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:05:35,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7987, 1.0563, 2.9003, 2.7135], device='cuda:0'), covar=tensor([0.1789, 0.2724, 0.0638, 0.1011], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0665, 0.0990, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 02:05:37,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.603e+03 1.994e+03 2.667e+03 8.139e+03, threshold=3.988e+03, percent-clipped=3.0
+2023-03-13 02:06:16,260 INFO [train.py:968] (0/2) Epoch 25, batch 28750, giga_loss[loss=0.3212, simple_loss=0.3877, pruned_loss=0.1274, over 28892.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3687, pruned_loss=0.1216, over 5658828.15 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1127, over 5701106.32 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3694, pruned_loss=0.1222, over 5648651.50 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:06:34,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5452, 1.9387, 1.5246, 1.5583], device='cuda:0'), covar=tensor([0.2524, 0.2579, 0.2926, 0.2356], device='cuda:0'), in_proj_covar=tensor([0.1563, 0.1126, 0.1379, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 02:06:44,881 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-13 02:06:49,270 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1122629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:07:01,632 INFO [train.py:968] (0/2) Epoch 25, batch 28800, giga_loss[loss=0.3631, simple_loss=0.4117, pruned_loss=0.1572, over 28561.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3709, pruned_loss=0.1229, over 5658313.59 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3606, pruned_loss=0.1132, over 5695834.26 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3713, pruned_loss=0.1234, over 5653041.25 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:07:08,481 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1122646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:07:10,576 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1122649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:07:13,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.972e+03 2.341e+03 3.265e+03 7.746e+03, threshold=4.682e+03, percent-clipped=14.0
+2023-03-13 02:07:40,362 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1122678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:07:50,715 INFO [train.py:968] (0/2) Epoch 25, batch 28850, giga_loss[loss=0.3004, simple_loss=0.3648, pruned_loss=0.118, over 28847.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3708, pruned_loss=0.123, over 5664457.23 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.113, over 5697801.52 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3715, pruned_loss=0.1236, over 5658404.29 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:08:11,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1122714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:08:36,354 INFO [train.py:968] (0/2) Epoch 25, batch 28900, giga_loss[loss=0.3262, simple_loss=0.3706, pruned_loss=0.1409, over 23471.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3714, pruned_loss=0.1239, over 5664954.67 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3603, pruned_loss=0.113, over 5699526.94 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3722, pruned_loss=0.1246, over 5657942.17 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:08:49,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.756e+03 2.303e+03 3.325e+03 9.194e+03, threshold=4.607e+03, percent-clipped=10.0
+2023-03-13 02:09:05,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4181, 1.5600, 1.4561, 1.3318], device='cuda:0'), covar=tensor([0.2535, 0.2254, 0.2219, 0.2464], device='cuda:0'), in_proj_covar=tensor([0.2034, 0.1985, 0.1895, 0.2035], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 02:09:22,523 INFO [train.py:968] (0/2) Epoch 25, batch 28950, giga_loss[loss=0.2665, simple_loss=0.3425, pruned_loss=0.09531, over 28818.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.371, pruned_loss=0.1231, over 5677996.87 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3608, pruned_loss=0.1133, over 5701706.34 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3714, pruned_loss=0.1236, over 5669918.95 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:10:10,792 INFO [train.py:968] (0/2) Epoch 25, batch 29000, giga_loss[loss=0.2947, simple_loss=0.3644, pruned_loss=0.1125, over 28660.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3725, pruned_loss=0.1239, over 5664024.37 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3614, pruned_loss=0.1138, over 5694862.04 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3724, pruned_loss=0.124, over 5663617.93 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:10:15,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9894, 1.2336, 0.9558, 0.3483], device='cuda:0'), covar=tensor([0.2668, 0.2106, 0.2534, 0.5316], device='cuda:0'), in_proj_covar=tensor([0.1820, 0.1704, 0.1643, 0.1474], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 02:10:22,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.877e+03 2.570e+03 3.353e+03 7.093e+03, threshold=5.140e+03, percent-clipped=13.0
+2023-03-13 02:10:26,710 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1122857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:10:29,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1122860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:10:50,626 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1122882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:10:56,237 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1122889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:10:56,686 INFO [train.py:968] (0/2) Epoch 25, batch 29050, libri_loss[loss=0.2965, simple_loss=0.3678, pruned_loss=0.1126, over 29636.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3727, pruned_loss=0.124, over 5665399.46 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.1139, over 5690308.43 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3728, pruned_loss=0.1242, over 5668538.08 frames. ], batch size: 91, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:11:29,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1122924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:11:41,224 INFO [train.py:968] (0/2) Epoch 25, batch 29100, libri_loss[loss=0.2986, simple_loss=0.3643, pruned_loss=0.1165, over 29499.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3743, pruned_loss=0.1254, over 5657377.71 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.362, pruned_loss=0.1144, over 5684697.10 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3741, pruned_loss=0.1254, over 5664925.85 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:11:51,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.173e+03 1.686e+03 2.207e+03 2.749e+03 9.578e+03, threshold=4.413e+03, percent-clipped=4.0
+2023-03-13 02:12:24,623 INFO [train.py:968] (0/2) Epoch 25, batch 29150, giga_loss[loss=0.3, simple_loss=0.3613, pruned_loss=0.1193, over 28286.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3755, pruned_loss=0.1267, over 5659958.43 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1143, over 5688636.64 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3756, pruned_loss=0.1269, over 5661964.73 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:12:36,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:13:10,869 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:13:11,347 INFO [train.py:968] (0/2) Epoch 25, batch 29200, giga_loss[loss=0.3423, simple_loss=0.3982, pruned_loss=0.1431, over 28835.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3748, pruned_loss=0.1253, over 5652590.20 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3617, pruned_loss=0.1142, over 5682278.55 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3753, pruned_loss=0.1258, over 5658950.39 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:13:22,019 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 02:13:25,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.212e+02 1.806e+03 2.404e+03 3.348e+03 8.172e+03, threshold=4.808e+03, percent-clipped=12.0
+2023-03-13 02:13:38,081 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:13:43,848 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:13:50,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2617, 1.5380, 1.4956, 1.2731], device='cuda:0'), covar=tensor([0.3383, 0.2579, 0.2381, 0.2868], device='cuda:0'), in_proj_covar=tensor([0.2030, 0.1977, 0.1890, 0.2030], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 02:14:02,255 INFO [train.py:968] (0/2) Epoch 25, batch 29250, giga_loss[loss=0.3023, simple_loss=0.3725, pruned_loss=0.1161, over 28889.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3741, pruned_loss=0.1242, over 5641625.50 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3615, pruned_loss=0.114, over 5682433.51 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.375, pruned_loss=0.125, over 5645427.24 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:14:10,077 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:14:21,670 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-13 02:14:38,693 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:14:41,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6473, 4.5057, 4.2570, 2.0844], device='cuda:0'), covar=tensor([0.0533, 0.0663, 0.0732, 0.1902], device='cuda:0'), in_proj_covar=tensor([0.1294, 0.1196, 0.1007, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 02:14:47,547 INFO [train.py:968] (0/2) Epoch 25, batch 29300, giga_loss[loss=0.2967, simple_loss=0.3706, pruned_loss=0.1114, over 28947.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3739, pruned_loss=0.1238, over 5629654.23 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3617, pruned_loss=0.1143, over 5668643.50 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3747, pruned_loss=0.1244, over 5643710.11 frames. ], batch size: 164, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:14:53,128 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:14:55,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:15:00,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.676e+03 2.180e+03 3.335e+03 1.236e+04, threshold=4.361e+03, percent-clipped=13.0
+2023-03-13 02:15:26,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:15:34,575 INFO [train.py:968] (0/2) Epoch 25, batch 29350, giga_loss[loss=0.2891, simple_loss=0.358, pruned_loss=0.1101, over 28834.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3719, pruned_loss=0.1222, over 5646519.81 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.1141, over 5669658.28 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3729, pruned_loss=0.1229, over 5656402.53 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:15:57,834 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 02:16:09,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:16:09,713 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 02:16:19,171 INFO [train.py:968] (0/2) Epoch 25, batch 29400, giga_loss[loss=0.2994, simple_loss=0.3721, pruned_loss=0.1133, over 28678.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3712, pruned_loss=0.1216, over 5650491.39 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1144, over 5673900.35 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3718, pruned_loss=0.1221, over 5654302.71 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:16:35,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.612e+03 2.046e+03 2.979e+03 8.466e+03, threshold=4.092e+03, percent-clipped=8.0
+2023-03-13 02:16:37,463 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:17:07,060 INFO [train.py:968] (0/2) Epoch 25, batch 29450, giga_loss[loss=0.2719, simple_loss=0.3477, pruned_loss=0.09802, over 28836.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3723, pruned_loss=0.1223, over 5640408.51 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3623, pruned_loss=0.1148, over 5661315.66 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3726, pruned_loss=0.1225, over 5653930.47 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:17:39,849 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 02:17:45,957 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:17:54,119 INFO [train.py:968] (0/2) Epoch 25, batch 29500, giga_loss[loss=0.3666, simple_loss=0.394, pruned_loss=0.1696, over 23556.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3738, pruned_loss=0.1242, over 5645128.14 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3624, pruned_loss=0.1148, over 5666343.18 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3743, pruned_loss=0.1246, over 5650863.91 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:18:05,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2452, 1.5847, 1.5423, 1.1090], device='cuda:0'), covar=tensor([0.1592, 0.2732, 0.1428, 0.1731], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0713, 0.0967, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 02:18:06,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.787e+03 2.439e+03 3.144e+03 8.187e+03, threshold=4.878e+03, percent-clipped=11.0
+2023-03-13 02:18:19,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6843, 1.5898, 1.8140, 1.4326], device='cuda:0'), covar=tensor([0.1519, 0.2419, 0.1266, 0.1575], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0713, 0.0967, 0.0864], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 02:18:35,980 INFO [train.py:968] (0/2) Epoch 25, batch 29550, giga_loss[loss=0.2738, simple_loss=0.3513, pruned_loss=0.09815, over 28863.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3724, pruned_loss=0.1239, over 5660993.09 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3614, pruned_loss=0.1143, over 5675568.54 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.374, pruned_loss=0.1249, over 5656566.76 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:18:44,667 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:18:47,859 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:18:57,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:19:11,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2354, 2.2643, 2.0031, 2.0383], device='cuda:0'), covar=tensor([0.1862, 0.2351, 0.2287, 0.2185], device='cuda:0'), in_proj_covar=tensor([0.0496, 0.0759, 0.0728, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 02:19:13,375 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:19:19,955 INFO [train.py:968] (0/2) Epoch 25, batch 29600, giga_loss[loss=0.3256, simple_loss=0.3962, pruned_loss=0.1275, over 28727.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3731, pruned_loss=0.1244, over 5663545.36 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1144, over 5673634.17 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3748, pruned_loss=0.1255, over 5660981.03 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:19:33,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.690e+02 1.771e+03 2.423e+03 3.449e+03 6.968e+03, threshold=4.846e+03, percent-clipped=7.0
+2023-03-13 02:20:05,420 INFO [train.py:968] (0/2) Epoch 25, batch 29650, giga_loss[loss=0.3102, simple_loss=0.3762, pruned_loss=0.1221, over 28548.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.376, pruned_loss=0.1267, over 5658535.26 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3616, pruned_loss=0.1143, over 5678565.04 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3775, pruned_loss=0.1279, over 5651847.77 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:20:19,478 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:20:36,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5925, 2.0738, 1.3862, 0.9067], device='cuda:0'), covar=tensor([0.6962, 0.3702, 0.3031, 0.6751], device='cuda:0'), in_proj_covar=tensor([0.1821, 0.1712, 0.1644, 0.1477], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 02:20:51,688 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-13 02:20:54,790 INFO [train.py:968] (0/2) Epoch 25, batch 29700, giga_loss[loss=0.3899, simple_loss=0.4288, pruned_loss=0.1756, over 27570.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3762, pruned_loss=0.1267, over 5646606.25 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3618, pruned_loss=0.1145, over 5670567.01 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3773, pruned_loss=0.1276, over 5647598.84 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:21:07,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.219e+03 1.745e+03 2.154e+03 2.981e+03 6.436e+03, threshold=4.309e+03, percent-clipped=4.0
+2023-03-13 02:21:09,153 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:21:12,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:21:41,700 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:21:42,080 INFO [train.py:968] (0/2) Epoch 25, batch 29750, giga_loss[loss=0.3115, simple_loss=0.3774, pruned_loss=0.1228, over 28268.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3755, pruned_loss=0.1259, over 5643630.27 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3615, pruned_loss=0.1142, over 5672911.72 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3768, pruned_loss=0.1269, over 5642302.73 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:21:44,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8522, 2.1227, 2.0467, 1.6365], device='cuda:0'), covar=tensor([0.1904, 0.2730, 0.1670, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.0918, 0.0715, 0.0968, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 02:21:54,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:22:27,358 INFO [train.py:968] (0/2) Epoch 25, batch 29800, giga_loss[loss=0.2905, simple_loss=0.3669, pruned_loss=0.107, over 28686.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3753, pruned_loss=0.1249, over 5648539.78 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3612, pruned_loss=0.114, over 5667719.71 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3772, pruned_loss=0.1264, over 5650396.06 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:22:35,734 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:22:37,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:22:40,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.213e+03 1.612e+03 2.132e+03 2.748e+03 8.464e+03, threshold=4.265e+03, percent-clipped=6.0
+2023-03-13 02:22:49,474 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:23:03,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:23:12,907 INFO [train.py:968] (0/2) Epoch 25, batch 29850, giga_loss[loss=0.2906, simple_loss=0.366, pruned_loss=0.1076, over 28958.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3754, pruned_loss=0.1249, over 5628002.55 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3616, pruned_loss=0.1143, over 5649116.84 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3769, pruned_loss=0.1262, over 5644603.79 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:23:26,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:24:00,918 INFO [train.py:968] (0/2) Epoch 25, batch 29900, giga_loss[loss=0.2821, simple_loss=0.3553, pruned_loss=0.1044, over 28918.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.373, pruned_loss=0.1229, over 5652080.19 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3617, pruned_loss=0.1143, over 5653080.55 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3743, pruned_loss=0.124, over 5661722.27 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:24:06,779 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:24:09,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:24:13,351 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 2.016e+03 2.819e+03 4.107e+03 1.600e+04, threshold=5.638e+03, percent-clipped=23.0
+2023-03-13 02:24:30,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6712, 1.6064, 1.8251, 1.4485], device='cuda:0'), covar=tensor([0.1904, 0.2550, 0.1536, 0.1801], device='cuda:0'), in_proj_covar=tensor([0.0918, 0.0713, 0.0967, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 02:24:34,727 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:24:46,194 INFO [train.py:968] (0/2) Epoch 25, batch 29950, giga_loss[loss=0.344, simple_loss=0.4067, pruned_loss=0.1407, over 28599.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.372, pruned_loss=0.1225, over 5646196.13 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3616, pruned_loss=0.1142, over 5638706.10 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3733, pruned_loss=0.1237, over 5665572.24 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:25:34,030 INFO [train.py:968] (0/2) Epoch 25, batch 30000, libri_loss[loss=0.2267, simple_loss=0.3012, pruned_loss=0.07611, over 29351.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3674, pruned_loss=0.1203, over 5646740.44 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3613, pruned_loss=0.114, over 5646730.31 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.369, pruned_loss=0.1216, over 5654928.86 frames. ], batch size: 67, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:25:34,034 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 02:25:42,599 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2049, simple_loss=0.3135, pruned_loss=0.04814, over 944034.00 frames. 
+2023-03-13 02:25:42,600 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 02:25:51,456 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:25:53,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:25:56,811 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.208e+03 2.007e+03 2.668e+03 3.687e+03 9.415e+03, threshold=5.336e+03, percent-clipped=13.0
+2023-03-13 02:26:17,929 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:26:24,955 INFO [train.py:968] (0/2) Epoch 25, batch 30050, giga_loss[loss=0.2725, simple_loss=0.3408, pruned_loss=0.1021, over 28900.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3646, pruned_loss=0.1196, over 5652396.50 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3615, pruned_loss=0.1142, over 5652430.55 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3658, pruned_loss=0.1205, over 5653815.25 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:27:10,488 INFO [train.py:968] (0/2) Epoch 25, batch 30100, giga_loss[loss=0.2977, simple_loss=0.3599, pruned_loss=0.1177, over 28880.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.363, pruned_loss=0.1189, over 5648463.56 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3611, pruned_loss=0.1137, over 5657607.89 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3645, pruned_loss=0.1204, over 5644737.34 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:27:22,390 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-13 02:27:25,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.275e+03 1.726e+03 2.339e+03 3.328e+03 1.121e+04, threshold=4.677e+03, percent-clipped=5.0
+2023-03-13 02:27:34,108 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123965.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:27:56,533 INFO [train.py:968] (0/2) Epoch 25, batch 30150, giga_loss[loss=0.2766, simple_loss=0.3508, pruned_loss=0.1012, over 28866.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3614, pruned_loss=0.1173, over 5645452.03 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5661302.08 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3625, pruned_loss=0.1184, over 5638941.66 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:28:05,682 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1124000.pt
+2023-03-13 02:28:44,258 INFO [train.py:968] (0/2) Epoch 25, batch 30200, giga_loss[loss=0.2579, simple_loss=0.3433, pruned_loss=0.08625, over 28684.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3603, pruned_loss=0.1145, over 5638848.31 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3614, pruned_loss=0.1142, over 5657928.10 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3609, pruned_loss=0.1152, over 5636722.24 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:28:45,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1124041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:29:00,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.748e+03 2.212e+03 2.756e+03 1.064e+04, threshold=4.424e+03, percent-clipped=3.0
+2023-03-13 02:29:36,532 INFO [train.py:968] (0/2) Epoch 25, batch 30250, giga_loss[loss=0.3248, simple_loss=0.377, pruned_loss=0.1363, over 26664.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3589, pruned_loss=0.1118, over 5635680.42 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3615, pruned_loss=0.1143, over 5653623.17 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3594, pruned_loss=0.1122, over 5638007.05 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:29:52,476 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:30:22,890 INFO [train.py:968] (0/2) Epoch 25, batch 30300, giga_loss[loss=0.2546, simple_loss=0.3412, pruned_loss=0.08397, over 28713.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3563, pruned_loss=0.1083, over 5651282.82 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3612, pruned_loss=0.1142, over 5656471.25 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3567, pruned_loss=0.1086, over 5650158.09 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:30:42,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.842e+02 1.643e+03 2.010e+03 2.897e+03 4.871e+03, threshold=4.020e+03, percent-clipped=3.0
+2023-03-13 02:31:09,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1124184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:31:11,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1124187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:31:12,713 INFO [train.py:968] (0/2) Epoch 25, batch 30350, giga_loss[loss=0.2179, simple_loss=0.2884, pruned_loss=0.07371, over 24086.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3525, pruned_loss=0.105, over 5646051.27 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3608, pruned_loss=0.114, over 5658237.16 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3531, pruned_loss=0.1053, over 5643370.38 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:31:38,248 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1124216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:31:57,366 INFO [train.py:968] (0/2) Epoch 25, batch 30400, giga_loss[loss=0.2431, simple_loss=0.3369, pruned_loss=0.07464, over 28588.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3512, pruned_loss=0.1026, over 5643599.24 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3606, pruned_loss=0.114, over 5644376.31 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3516, pruned_loss=0.1025, over 5653945.96 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:32:15,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.538e+02 1.381e+03 1.906e+03 3.038e+03 7.832e+03, threshold=3.813e+03, percent-clipped=6.0
+2023-03-13 02:32:50,129 INFO [train.py:968] (0/2) Epoch 25, batch 30450, libri_loss[loss=0.2814, simple_loss=0.3543, pruned_loss=0.1043, over 28524.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3504, pruned_loss=0.1005, over 5658678.88 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1136, over 5649841.93 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3513, pruned_loss=0.1004, over 5662055.26 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:33:45,500 INFO [train.py:968] (0/2) Epoch 25, batch 30500, giga_loss[loss=0.3483, simple_loss=0.4005, pruned_loss=0.148, over 27972.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3512, pruned_loss=0.1008, over 5657751.67 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1136, over 5649841.93 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3519, pruned_loss=0.1008, over 5660379.55 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:33:45,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1124340.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:34:01,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.441e+02 1.526e+03 2.106e+03 3.361e+03 9.866e+03, threshold=4.212e+03, percent-clipped=17.0
+2023-03-13 02:34:36,658 INFO [train.py:968] (0/2) Epoch 25, batch 30550, giga_loss[loss=0.2613, simple_loss=0.3456, pruned_loss=0.08848, over 28817.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3484, pruned_loss=0.09923, over 5663620.91 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3592, pruned_loss=0.1136, over 5656618.95 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3491, pruned_loss=0.0987, over 5659662.38 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:35:05,466 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4663, 1.1367, 4.4062, 3.5097], device='cuda:0'), covar=tensor([0.1690, 0.3109, 0.0394, 0.1051], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0664, 0.0983, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 02:35:20,932 INFO [train.py:968] (0/2) Epoch 25, batch 30600, libri_loss[loss=0.2922, simple_loss=0.3408, pruned_loss=0.1218, over 29569.00 frames. ], tot_loss[loss=0.271, simple_loss=0.346, pruned_loss=0.09799, over 5663231.78 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3589, pruned_loss=0.1137, over 5663147.46 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3464, pruned_loss=0.09695, over 5654234.17 frames. ], batch size: 75, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:35:27,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4255, 1.3600, 1.5599, 1.1680], device='cuda:0'), covar=tensor([0.1804, 0.3116, 0.1455, 0.1717], device='cuda:0'), in_proj_covar=tensor([0.0915, 0.0707, 0.0962, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 02:35:32,721 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 02:35:39,452 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.587e+02 1.343e+03 1.685e+03 2.395e+03 4.656e+03, threshold=3.370e+03, percent-clipped=1.0
+2023-03-13 02:36:02,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1124482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:36:03,008 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1124483.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:36:07,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1124486.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:36:09,730 INFO [train.py:968] (0/2) Epoch 25, batch 30650, libri_loss[loss=0.2722, simple_loss=0.3453, pruned_loss=0.09961, over 29538.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3465, pruned_loss=0.09818, over 5668254.01 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3589, pruned_loss=0.1138, over 5667129.76 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3465, pruned_loss=0.09693, over 5657405.75 frames. ], batch size: 82, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:36:32,754 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:36:32,769 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1124515.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:36:57,934 INFO [train.py:968] (0/2) Epoch 25, batch 30700, giga_loss[loss=0.261, simple_loss=0.3452, pruned_loss=0.08842, over 28625.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3451, pruned_loss=0.09698, over 5660790.99 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3586, pruned_loss=0.1136, over 5664913.83 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3452, pruned_loss=0.09598, over 5654314.95 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:36:59,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4888, 3.3198, 1.6125, 1.5059], device='cuda:0'), covar=tensor([0.0978, 0.0322, 0.0919, 0.1379], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0564, 0.0400, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 02:36:59,936 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:37:15,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.420e+02 1.445e+03 1.805e+03 2.400e+03 4.498e+03, threshold=3.610e+03, percent-clipped=4.0
+2023-03-13 02:37:47,713 INFO [train.py:968] (0/2) Epoch 25, batch 30750, giga_loss[loss=0.2351, simple_loss=0.3288, pruned_loss=0.07073, over 28987.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3428, pruned_loss=0.09508, over 5650456.98 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3586, pruned_loss=0.1138, over 5651391.34 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3426, pruned_loss=0.0938, over 5656893.43 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:37:54,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3908, 3.2093, 3.0934, 1.5297], device='cuda:0'), covar=tensor([0.1068, 0.1228, 0.1193, 0.2157], device='cuda:0'), in_proj_covar=tensor([0.1273, 0.1178, 0.0989, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 02:38:21,995 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1124625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:38:24,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1124628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:38:36,193 INFO [train.py:968] (0/2) Epoch 25, batch 30800, libri_loss[loss=0.2799, simple_loss=0.3319, pruned_loss=0.114, over 28602.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.34, pruned_loss=0.09329, over 5660264.66 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3582, pruned_loss=0.1137, over 5656298.99 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3397, pruned_loss=0.09181, over 5661053.85 frames. ], batch size: 63, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:38:53,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.710e+02 1.491e+03 1.928e+03 2.788e+03 7.331e+03, threshold=3.856e+03, percent-clipped=13.0
+2023-03-13 02:38:54,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1124657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:39:02,279 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:39:06,449 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 02:39:24,755 INFO [train.py:968] (0/2) Epoch 25, batch 30850, libri_loss[loss=0.3077, simple_loss=0.361, pruned_loss=0.1272, over 26110.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3368, pruned_loss=0.09212, over 5664932.87 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3576, pruned_loss=0.1136, over 5659299.46 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3365, pruned_loss=0.09034, over 5663357.89 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:39:55,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6318, 1.8458, 1.5116, 1.7603], device='cuda:0'), covar=tensor([0.2902, 0.2890, 0.3390, 0.2425], device='cuda:0'), in_proj_covar=tensor([0.1569, 0.1127, 0.1389, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 02:40:04,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7331, 2.0324, 1.3603, 1.5549], device='cuda:0'), covar=tensor([0.1001, 0.0520, 0.0980, 0.1113], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0447, 0.0518, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 02:40:06,673 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124736.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:40:09,052 INFO [train.py:968] (0/2) Epoch 25, batch 30900, libri_loss[loss=0.2682, simple_loss=0.3389, pruned_loss=0.09875, over 29743.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3365, pruned_loss=0.09265, over 5664864.38 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3577, pruned_loss=0.1138, over 5659780.38 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3354, pruned_loss=0.09019, over 5663789.60 frames. ], batch size: 87, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:40:24,819 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.201e+02 1.454e+03 1.882e+03 2.683e+03 3.988e+03, threshold=3.764e+03, percent-clipped=1.0
+2023-03-13 02:41:01,987 INFO [train.py:968] (0/2) Epoch 25, batch 30950, giga_loss[loss=0.2761, simple_loss=0.3363, pruned_loss=0.1079, over 26677.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3362, pruned_loss=0.0929, over 5648215.87 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.357, pruned_loss=0.1134, over 5663780.37 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3355, pruned_loss=0.09092, over 5643802.29 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:41:07,558 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5099, 1.7827, 1.4523, 1.5839], device='cuda:0'), covar=tensor([0.3078, 0.2869, 0.3362, 0.2514], device='cuda:0'), in_proj_covar=tensor([0.1570, 0.1127, 0.1388, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 02:41:51,072 INFO [train.py:968] (0/2) Epoch 25, batch 31000, giga_loss[loss=0.2686, simple_loss=0.3363, pruned_loss=0.1004, over 26650.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3388, pruned_loss=0.09401, over 5644592.50 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3561, pruned_loss=0.1129, over 5664097.19 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3384, pruned_loss=0.09201, over 5640361.82 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:41:56,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4136, 1.6839, 1.3611, 1.3283], device='cuda:0'), covar=tensor([0.2781, 0.2693, 0.3149, 0.2481], device='cuda:0'), in_proj_covar=tensor([0.1568, 0.1127, 0.1388, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 02:42:11,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.890e+02 1.472e+03 2.052e+03 2.661e+03 6.087e+03, threshold=4.105e+03, percent-clipped=6.0
+2023-03-13 02:42:32,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3731, 1.3734, 3.4773, 3.3565], device='cuda:0'), covar=tensor([0.1429, 0.2812, 0.0420, 0.1052], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0664, 0.0980, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 02:42:44,419 INFO [train.py:968] (0/2) Epoch 25, batch 31050, libri_loss[loss=0.3095, simple_loss=0.357, pruned_loss=0.131, over 29556.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3391, pruned_loss=0.09323, over 5644664.94 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3553, pruned_loss=0.1127, over 5670288.43 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.339, pruned_loss=0.09124, over 5634794.63 frames. ], batch size: 77, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:42:45,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1124890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:43:15,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1124917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:43:15,583 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 02:43:19,147 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:43:39,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8916, 1.3002, 1.3264, 1.1094], device='cuda:0'), covar=tensor([0.1777, 0.1049, 0.1778, 0.1322], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0740, 0.0712, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 02:43:41,997 INFO [train.py:968] (0/2) Epoch 25, batch 31100, giga_loss[loss=0.2128, simple_loss=0.3024, pruned_loss=0.06157, over 28734.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3387, pruned_loss=0.09297, over 5637897.57 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.355, pruned_loss=0.1126, over 5673618.91 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3384, pruned_loss=0.09096, over 5625858.36 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:44:04,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.852e+02 1.438e+03 1.981e+03 2.550e+03 8.636e+03, threshold=3.961e+03, percent-clipped=5.0
+2023-03-13 02:44:42,646 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9585, 2.6376, 2.3744, 2.0103], device='cuda:0'), covar=tensor([0.2663, 0.2133, 0.2131, 0.2499], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0743, 0.0714, 0.0683], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 02:44:43,654 INFO [train.py:968] (0/2) Epoch 25, batch 31150, libri_loss[loss=0.3273, simple_loss=0.3799, pruned_loss=0.1373, over 25762.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.337, pruned_loss=0.09168, over 5647179.57 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3543, pruned_loss=0.1123, over 5674265.54 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3371, pruned_loss=0.08996, over 5636729.71 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:45:10,369 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125015.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:45:31,756 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:45:35,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:45:37,643 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:45:40,712 INFO [train.py:968] (0/2) Epoch 25, batch 31200, giga_loss[loss=0.2066, simple_loss=0.2822, pruned_loss=0.06549, over 24365.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3359, pruned_loss=0.09023, over 5635168.49 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3539, pruned_loss=0.1121, over 5672326.21 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3358, pruned_loss=0.08836, over 5627040.28 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:45:41,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:02,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.010e+02 1.365e+03 1.730e+03 2.512e+03 4.594e+03, threshold=3.460e+03, percent-clipped=4.0
+2023-03-13 02:46:04,755 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:06,644 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:08,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:09,245 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.87 vs. limit=2.0
+2023-03-13 02:46:09,804 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:37,209 INFO [train.py:968] (0/2) Epoch 25, batch 31250, giga_loss[loss=0.2251, simple_loss=0.3095, pruned_loss=0.07031, over 28994.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3338, pruned_loss=0.08807, over 5642000.36 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3539, pruned_loss=0.1123, over 5674350.32 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3333, pruned_loss=0.08588, over 5633111.42 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:46:37,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5240, 2.1166, 1.7648, 1.9604], device='cuda:0'), covar=tensor([0.0790, 0.0265, 0.0328, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 02:46:41,071 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:47:01,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125111.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:47:32,474 INFO [train.py:968] (0/2) Epoch 25, batch 31300, giga_loss[loss=0.2292, simple_loss=0.3102, pruned_loss=0.07414, over 28633.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3326, pruned_loss=0.08886, over 5659973.28 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3543, pruned_loss=0.1129, over 5679199.44 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.331, pruned_loss=0.08566, over 5647314.69 frames. ], batch size: 60, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:47:44,329 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-13 02:47:57,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.670e+02 1.363e+03 1.864e+03 2.603e+03 5.336e+03, threshold=3.727e+03, percent-clipped=10.0
+2023-03-13 02:48:03,633 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-13 02:48:22,179 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:48:26,777 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:48:30,089 INFO [train.py:968] (0/2) Epoch 25, batch 31350, libri_loss[loss=0.2836, simple_loss=0.3477, pruned_loss=0.1097, over 29506.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3326, pruned_loss=0.08921, over 5674231.16 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3534, pruned_loss=0.1123, over 5685986.29 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3316, pruned_loss=0.08652, over 5657376.86 frames. ], batch size: 89, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:48:42,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5190, 2.2622, 1.6508, 0.6947], device='cuda:0'), covar=tensor([0.6587, 0.3122, 0.4535, 0.6829], device='cuda:0'), in_proj_covar=tensor([0.1806, 0.1688, 0.1629, 0.1464], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 02:48:59,573 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:49:23,724 INFO [train.py:968] (0/2) Epoch 25, batch 31400, libri_loss[loss=0.2386, simple_loss=0.3031, pruned_loss=0.08709, over 29368.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3333, pruned_loss=0.08943, over 5661436.26 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3533, pruned_loss=0.1123, over 5668983.84 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.332, pruned_loss=0.08661, over 5662420.13 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:49:40,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:49:44,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125257.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:49:44,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-13 02:49:45,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.195e+02 1.401e+03 1.829e+03 2.905e+03 7.628e+03, threshold=3.657e+03, percent-clipped=13.0
+2023-03-13 02:50:10,046 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:50:16,308 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125286.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:50:20,555 INFO [train.py:968] (0/2) Epoch 25, batch 31450, giga_loss[loss=0.3041, simple_loss=0.3804, pruned_loss=0.1139, over 28865.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3344, pruned_loss=0.08942, over 5657689.73 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3528, pruned_loss=0.112, over 5674222.21 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3333, pruned_loss=0.0869, over 5653712.79 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:50:29,080 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:51:20,825 INFO [train.py:968] (0/2) Epoch 25, batch 31500, giga_loss[loss=0.273, simple_loss=0.3497, pruned_loss=0.09811, over 28715.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3335, pruned_loss=0.08844, over 5663856.37 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3521, pruned_loss=0.1117, over 5680457.17 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3327, pruned_loss=0.08594, over 5654617.46 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:51:41,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.887e+02 1.361e+03 1.835e+03 2.652e+03 6.771e+03, threshold=3.670e+03, percent-clipped=9.0
+2023-03-13 02:51:46,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:52:25,165 INFO [train.py:968] (0/2) Epoch 25, batch 31550, giga_loss[loss=0.222, simple_loss=0.3037, pruned_loss=0.0702, over 28189.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3317, pruned_loss=0.08743, over 5678860.83 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.352, pruned_loss=0.1116, over 5685799.90 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3306, pruned_loss=0.085, over 5666522.12 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:52:25,378 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125390.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:52:32,503 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-13 02:52:54,716 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:53:25,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:53:29,389 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:53:29,695 INFO [train.py:968] (0/2) Epoch 25, batch 31600, giga_loss[loss=0.2485, simple_loss=0.3329, pruned_loss=0.08204, over 28952.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3327, pruned_loss=0.08793, over 5676625.56 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3514, pruned_loss=0.1112, over 5689288.90 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3322, pruned_loss=0.08598, over 5663741.73 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:53:30,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2908, 1.2737, 1.2870, 1.4644], device='cuda:0'), covar=tensor([0.0789, 0.0386, 0.0351, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 02:53:33,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:53:54,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.800e+02 1.469e+03 1.980e+03 3.022e+03 8.169e+03, threshold=3.961e+03, percent-clipped=16.0
+2023-03-13 02:54:10,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:54:29,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1006, 2.4890, 2.1541, 2.3205], device='cuda:0'), covar=tensor([0.1931, 0.1890, 0.2123, 0.1733], device='cuda:0'), in_proj_covar=tensor([0.1564, 0.1124, 0.1383, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 02:54:34,699 INFO [train.py:968] (0/2) Epoch 25, batch 31650, giga_loss[loss=0.2433, simple_loss=0.3245, pruned_loss=0.08107, over 28677.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3369, pruned_loss=0.08779, over 5670222.55 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3515, pruned_loss=0.1114, over 5682180.54 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3362, pruned_loss=0.08583, over 5666917.28 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:55:29,766 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125533.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:55:31,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125536.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:55:34,805 INFO [train.py:968] (0/2) Epoch 25, batch 31700, giga_loss[loss=0.2614, simple_loss=0.3407, pruned_loss=0.091, over 27559.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3375, pruned_loss=0.08681, over 5661066.74 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3513, pruned_loss=0.1113, over 5688807.44 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3367, pruned_loss=0.08455, over 5651639.86 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:55:52,639 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:55:56,556 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:55:56,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.438e+02 1.467e+03 1.759e+03 2.370e+03 4.762e+03, threshold=3.519e+03, percent-clipped=8.0
+2023-03-13 02:56:04,227 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125565.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:56:19,326 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:56:23,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:56:31,007 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:56:32,272 INFO [train.py:968] (0/2) Epoch 25, batch 31750, giga_loss[loss=0.2592, simple_loss=0.3343, pruned_loss=0.09199, over 26888.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3381, pruned_loss=0.0857, over 5666166.47 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3511, pruned_loss=0.1113, over 5691203.92 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3373, pruned_loss=0.08349, over 5656158.40 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:56:56,863 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:57:30,675 INFO [train.py:968] (0/2) Epoch 25, batch 31800, giga_loss[loss=0.2708, simple_loss=0.3525, pruned_loss=0.09454, over 28684.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3389, pruned_loss=0.08655, over 5681623.15 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3509, pruned_loss=0.1113, over 5698672.96 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3381, pruned_loss=0.08394, over 5666361.14 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:57:52,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:57:55,100 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-13 02:57:55,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.033e+02 1.537e+03 1.903e+03 2.707e+03 7.992e+03, threshold=3.805e+03, percent-clipped=11.0
+2023-03-13 02:58:33,657 INFO [train.py:968] (0/2) Epoch 25, batch 31850, giga_loss[loss=0.238, simple_loss=0.3163, pruned_loss=0.07983, over 29121.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3387, pruned_loss=0.08777, over 5693304.92 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3506, pruned_loss=0.1111, over 5702135.78 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3381, pruned_loss=0.08544, over 5677768.15 frames. ], batch size: 200, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:58:54,799 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-13 02:59:30,741 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:59:41,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:59:42,623 INFO [train.py:968] (0/2) Epoch 25, batch 31900, giga_loss[loss=0.2521, simple_loss=0.3375, pruned_loss=0.08333, over 28712.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3386, pruned_loss=0.08908, over 5686061.73 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3499, pruned_loss=0.1107, over 5707361.15 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3385, pruned_loss=0.08699, over 5668568.42 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:00:16,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.979e+02 1.378e+03 2.001e+03 2.653e+03 8.068e+03, threshold=4.002e+03, percent-clipped=8.0
+2023-03-13 03:00:21,428 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-13 03:00:51,233 INFO [train.py:968] (0/2) Epoch 25, batch 31950, giga_loss[loss=0.2266, simple_loss=0.3137, pruned_loss=0.06975, over 28502.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3383, pruned_loss=0.08955, over 5691647.75 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3501, pruned_loss=0.111, over 5710117.32 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3377, pruned_loss=0.08705, over 5674527.37 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:01:07,538 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:01:10,362 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:01:45,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:01:55,331 INFO [train.py:968] (0/2) Epoch 25, batch 32000, giga_loss[loss=0.2137, simple_loss=0.3088, pruned_loss=0.05929, over 28955.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3332, pruned_loss=0.08666, over 5688777.43 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3494, pruned_loss=0.1106, over 5714686.42 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.333, pruned_loss=0.08433, over 5670154.28 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:02:06,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6888, 1.9149, 1.3019, 1.5148], device='cuda:0'), covar=tensor([0.1032, 0.0569, 0.1079, 0.1107], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0445, 0.0517, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 03:02:18,942 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.319e+02 1.366e+03 1.712e+03 2.309e+03 7.041e+03, threshold=3.425e+03, percent-clipped=5.0
+2023-03-13 03:02:42,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4136, 1.3083, 4.1702, 3.4023], device='cuda:0'), covar=tensor([0.1630, 0.2880, 0.0420, 0.1242], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0662, 0.0979, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 03:02:46,786 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:02:52,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:03:01,037 INFO [train.py:968] (0/2) Epoch 25, batch 32050, giga_loss[loss=0.2482, simple_loss=0.3275, pruned_loss=0.08441, over 29003.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3315, pruned_loss=0.0856, over 5688835.51 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3495, pruned_loss=0.1106, over 5715743.98 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3308, pruned_loss=0.08319, over 5672407.05 frames. ], batch size: 285, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:03:13,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5965, 1.8922, 1.6035, 1.6613], device='cuda:0'), covar=tensor([0.2536, 0.2356, 0.2585, 0.2321], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1119, 0.1380, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 03:03:30,187 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:04:02,936 INFO [train.py:968] (0/2) Epoch 25, batch 32100, giga_loss[loss=0.2763, simple_loss=0.3615, pruned_loss=0.09555, over 28742.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3317, pruned_loss=0.08598, over 5692611.59 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3493, pruned_loss=0.1104, over 5716260.43 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.331, pruned_loss=0.08366, over 5678551.49 frames. ], batch size: 243, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:04:26,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.862e+02 1.438e+03 1.812e+03 2.498e+03 5.680e+03, threshold=3.624e+03, percent-clipped=14.0
+2023-03-13 03:04:56,567 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:05:01,478 INFO [train.py:968] (0/2) Epoch 25, batch 32150, giga_loss[loss=0.2774, simple_loss=0.3542, pruned_loss=0.1003, over 29056.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3354, pruned_loss=0.08764, over 5695355.29 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3493, pruned_loss=0.1104, over 5720223.20 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3345, pruned_loss=0.08532, over 5680232.58 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:05:14,292 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1126000.pt
+2023-03-13 03:06:02,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4244, 3.2328, 1.5150, 1.6457], device='cuda:0'), covar=tensor([0.0981, 0.0382, 0.1004, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0558, 0.0398, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 03:06:02,492 INFO [train.py:968] (0/2) Epoch 25, batch 32200, giga_loss[loss=0.2849, simple_loss=0.3551, pruned_loss=0.1073, over 28381.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3339, pruned_loss=0.08845, over 5697706.84 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3486, pruned_loss=0.11, over 5724678.46 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3333, pruned_loss=0.08622, over 5680792.34 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:06:02,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4915, 1.8137, 1.4736, 1.4460], device='cuda:0'), covar=tensor([0.2850, 0.2749, 0.3204, 0.2506], device='cuda:0'), in_proj_covar=tensor([0.1564, 0.1122, 0.1383, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 03:06:25,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.819e+02 1.491e+03 1.805e+03 2.437e+03 4.876e+03, threshold=3.609e+03, percent-clipped=4.0
+2023-03-13 03:07:00,525 INFO [train.py:968] (0/2) Epoch 25, batch 32250, giga_loss[loss=0.2723, simple_loss=0.3374, pruned_loss=0.1037, over 28629.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3346, pruned_loss=0.08981, over 5696571.63 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3486, pruned_loss=0.11, over 5727038.46 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3337, pruned_loss=0.08744, over 5680670.62 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:07:18,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1126105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:07:56,053 INFO [train.py:968] (0/2) Epoch 25, batch 32300, giga_loss[loss=0.2518, simple_loss=0.3359, pruned_loss=0.08382, over 28841.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3343, pruned_loss=0.0901, over 5699988.82 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3477, pruned_loss=0.1096, over 5733398.59 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3338, pruned_loss=0.08776, over 5679941.75 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:08:27,326 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.225e+02 1.513e+03 2.077e+03 3.009e+03 9.072e+03, threshold=4.154e+03, percent-clipped=14.0
+2023-03-13 03:09:07,394 INFO [train.py:968] (0/2) Epoch 25, batch 32350, giga_loss[loss=0.2899, simple_loss=0.3703, pruned_loss=0.1048, over 28951.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.337, pruned_loss=0.09101, over 5687652.04 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3476, pruned_loss=0.1097, over 5727141.11 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3364, pruned_loss=0.08864, over 5676661.76 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:10:11,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8099, 2.0278, 1.4083, 1.6633], device='cuda:0'), covar=tensor([0.0983, 0.0624, 0.0982, 0.1196], device='cuda:0'), in_proj_covar=tensor([0.0400, 0.0444, 0.0516, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 03:10:18,081 INFO [train.py:968] (0/2) Epoch 25, batch 32400, giga_loss[loss=0.249, simple_loss=0.3377, pruned_loss=0.08015, over 29009.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3388, pruned_loss=0.09157, over 5680926.35 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3475, pruned_loss=0.1097, over 5730203.42 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3383, pruned_loss=0.08925, over 5668485.65 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:10:30,694 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1126248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:10:34,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1126251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:10:49,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.156e+03 1.605e+03 2.149e+03 2.972e+03 5.597e+03, threshold=4.297e+03, percent-clipped=11.0
+2023-03-13 03:10:53,447 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:11:14,763 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1126280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:11:17,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7470, 1.9810, 1.6554, 1.7776], device='cuda:0'), covar=tensor([0.2886, 0.2796, 0.3168, 0.2742], device='cuda:0'), in_proj_covar=tensor([0.1566, 0.1123, 0.1384, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 03:11:17,926 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 03:11:26,574 INFO [train.py:968] (0/2) Epoch 25, batch 32450, giga_loss[loss=0.2602, simple_loss=0.3344, pruned_loss=0.09302, over 28461.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3353, pruned_loss=0.08986, over 5672964.22 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3474, pruned_loss=0.1097, over 5722604.17 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3348, pruned_loss=0.08775, over 5670149.79 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:11:57,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3676, 1.7011, 1.6684, 1.4580], device='cuda:0'), covar=tensor([0.1981, 0.2104, 0.2064, 0.1999], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0742, 0.0714, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 03:12:31,978 INFO [train.py:968] (0/2) Epoch 25, batch 32500, giga_loss[loss=0.2093, simple_loss=0.2872, pruned_loss=0.06569, over 27731.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3302, pruned_loss=0.08799, over 5675156.14 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3474, pruned_loss=0.1097, over 5721343.79 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3295, pruned_loss=0.08585, over 5672905.34 frames. ], batch size: 474, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:12:59,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1126360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:13:02,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.369e+02 1.511e+03 1.938e+03 2.533e+03 4.515e+03, threshold=3.876e+03, percent-clipped=3.0
+2023-03-13 03:13:34,573 INFO [train.py:968] (0/2) Epoch 25, batch 32550, giga_loss[loss=0.2302, simple_loss=0.3126, pruned_loss=0.07395, over 28974.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3266, pruned_loss=0.08661, over 5673340.46 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3466, pruned_loss=0.1096, over 5727758.29 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.326, pruned_loss=0.08412, over 5663803.94 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:14:28,522 INFO [train.py:968] (0/2) Epoch 25, batch 32600, giga_loss[loss=0.292, simple_loss=0.3608, pruned_loss=0.1116, over 28741.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3282, pruned_loss=0.08795, over 5682146.76 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3463, pruned_loss=0.1094, over 5732376.82 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3274, pruned_loss=0.08537, over 5668367.61 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:14:32,968 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-13 03:14:48,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3545, 1.6728, 1.1115, 1.2426], device='cuda:0'), covar=tensor([0.1121, 0.0508, 0.1220, 0.1057], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0444, 0.0517, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 03:14:52,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.668e+03 2.119e+03 2.714e+03 6.485e+03, threshold=4.238e+03, percent-clipped=6.0
+2023-03-13 03:15:25,589 INFO [train.py:968] (0/2) Epoch 25, batch 32650, giga_loss[loss=0.2391, simple_loss=0.3207, pruned_loss=0.07876, over 28932.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3293, pruned_loss=0.08815, over 5681634.85 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3464, pruned_loss=0.1095, over 5729906.68 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3283, pruned_loss=0.08568, over 5671952.41 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:15:43,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1126503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:15:43,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:15:45,947 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-13 03:15:46,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1126506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:15:50,818 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6919, 1.9870, 1.6525, 1.7006], device='cuda:0'), covar=tensor([0.2609, 0.2361, 0.2480, 0.2282], device='cuda:0'), in_proj_covar=tensor([0.1563, 0.1121, 0.1382, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 03:16:24,405 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1126535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:16:32,055 INFO [train.py:968] (0/2) Epoch 25, batch 32700, giga_loss[loss=0.2096, simple_loss=0.2815, pruned_loss=0.06889, over 24224.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3259, pruned_loss=0.0849, over 5668252.03 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3462, pruned_loss=0.1093, over 5729340.89 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.325, pruned_loss=0.08285, over 5660654.23 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:16:57,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.486e+02 1.494e+03 1.782e+03 2.472e+03 7.074e+03, threshold=3.563e+03, percent-clipped=4.0
+2023-03-13 03:17:19,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4079, 1.1881, 3.6809, 3.2121], device='cuda:0'), covar=tensor([0.1586, 0.2756, 0.0545, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0660, 0.0973, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 03:17:31,949 INFO [train.py:968] (0/2) Epoch 25, batch 32750, giga_loss[loss=0.2365, simple_loss=0.3206, pruned_loss=0.07615, over 28606.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3248, pruned_loss=0.08415, over 5664105.28 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3463, pruned_loss=0.1094, over 5722935.75 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3236, pruned_loss=0.0819, over 5662059.36 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 03:17:45,900 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 03:18:24,057 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-13 03:18:36,297 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1126639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:18:36,626 INFO [train.py:968] (0/2) Epoch 25, batch 32800, giga_loss[loss=0.2027, simple_loss=0.2746, pruned_loss=0.06541, over 24410.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3245, pruned_loss=0.08438, over 5664713.66 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3465, pruned_loss=0.1095, over 5722847.69 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3227, pruned_loss=0.08178, over 5661913.44 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:19:08,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.007e+02 1.418e+03 1.965e+03 2.805e+03 7.258e+03, threshold=3.930e+03, percent-clipped=14.0
+2023-03-13 03:19:41,351 INFO [train.py:968] (0/2) Epoch 25, batch 32850, giga_loss[loss=0.2444, simple_loss=0.3317, pruned_loss=0.07856, over 28415.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3268, pruned_loss=0.08518, over 5676248.21 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3464, pruned_loss=0.1094, over 5722234.73 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3252, pruned_loss=0.08276, over 5673588.37 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:20:14,025 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-13 03:20:32,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:20:43,524 INFO [train.py:968] (0/2) Epoch 25, batch 32900, giga_loss[loss=0.2013, simple_loss=0.2861, pruned_loss=0.05826, over 28538.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3277, pruned_loss=0.08595, over 5679963.48 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3467, pruned_loss=0.1098, over 5724724.67 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3258, pruned_loss=0.08335, over 5674931.90 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:21:08,103 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.595e+02 1.356e+03 1.848e+03 2.684e+03 1.048e+04, threshold=3.695e+03, percent-clipped=12.0
+2023-03-13 03:21:12,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-13 03:21:27,520 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1126782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:21:30,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1126785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:21:35,126 INFO [train.py:968] (0/2) Epoch 25, batch 32950, giga_loss[loss=0.2282, simple_loss=0.313, pruned_loss=0.0717, over 28425.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3278, pruned_loss=0.08689, over 5676281.55 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3463, pruned_loss=0.1096, over 5719479.73 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3258, pruned_loss=0.08393, over 5674330.07 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:21:37,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 03:22:03,406 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1126814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:22:11,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1348, 1.4949, 1.4347, 1.0410], device='cuda:0'), covar=tensor([0.1777, 0.2664, 0.1438, 0.1805], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0702, 0.0964, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 03:22:24,679 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-13 03:22:31,251 INFO [train.py:968] (0/2) Epoch 25, batch 33000, giga_loss[loss=0.21, simple_loss=0.2789, pruned_loss=0.07054, over 24513.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3277, pruned_loss=0.08597, over 5673489.65 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3464, pruned_loss=0.1097, over 5723305.03 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3255, pruned_loss=0.08279, over 5667341.94 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:22:31,255 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 03:22:39,891 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.195, simple_loss=0.296, pruned_loss=0.047, over 944034.00 frames. 
+2023-03-13 03:22:39,892 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 03:22:56,825 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:23:05,326 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.816e+02 1.421e+03 1.925e+03 2.954e+03 6.048e+03, threshold=3.851e+03, percent-clipped=18.0
+2023-03-13 03:23:05,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:23:24,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1126879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:23:34,669 INFO [train.py:968] (0/2) Epoch 25, batch 33050, giga_loss[loss=0.2432, simple_loss=0.3314, pruned_loss=0.07749, over 28897.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3301, pruned_loss=0.08597, over 5671179.31 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3465, pruned_loss=0.1097, over 5726886.73 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3279, pruned_loss=0.083, over 5661946.02 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:23:58,473 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 03:24:12,682 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 03:24:31,973 INFO [train.py:968] (0/2) Epoch 25, batch 33100, giga_loss[loss=0.2314, simple_loss=0.3172, pruned_loss=0.07284, over 28954.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.332, pruned_loss=0.08631, over 5664120.45 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3462, pruned_loss=0.1094, over 5720413.71 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3302, pruned_loss=0.08367, over 5662243.63 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:24:36,632 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:24:59,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.149e+02 1.458e+03 1.843e+03 2.367e+03 4.806e+03, threshold=3.685e+03, percent-clipped=5.0
+2023-03-13 03:25:36,945 INFO [train.py:968] (0/2) Epoch 25, batch 33150, giga_loss[loss=0.2198, simple_loss=0.3072, pruned_loss=0.06626, over 28956.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.332, pruned_loss=0.08588, over 5659751.97 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3459, pruned_loss=0.1093, over 5718111.20 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3306, pruned_loss=0.08354, over 5659905.70 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:26:16,417 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:26:19,392 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:26:28,630 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 03:26:33,550 INFO [train.py:968] (0/2) Epoch 25, batch 33200, giga_loss[loss=0.2186, simple_loss=0.3044, pruned_loss=0.06638, over 28772.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3325, pruned_loss=0.08688, over 5672259.74 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3455, pruned_loss=0.1091, over 5725451.43 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3312, pruned_loss=0.08432, over 5663668.21 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:26:47,027 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:26:50,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:26:58,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.277e+02 1.563e+03 1.967e+03 2.596e+03 5.813e+03, threshold=3.933e+03, percent-clipped=11.0
+2023-03-13 03:27:34,275 INFO [train.py:968] (0/2) Epoch 25, batch 33250, giga_loss[loss=0.2433, simple_loss=0.3242, pruned_loss=0.08119, over 27638.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3286, pruned_loss=0.08396, over 5668288.29 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3456, pruned_loss=0.1091, over 5716031.53 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3272, pruned_loss=0.08157, over 5668350.22 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:27:51,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:28:10,400 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6591, 1.8603, 1.2806, 1.5504], device='cuda:0'), covar=tensor([0.1128, 0.0735, 0.1104, 0.1269], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0444, 0.0518, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 03:28:31,760 INFO [train.py:968] (0/2) Epoch 25, batch 33300, giga_loss[loss=0.2569, simple_loss=0.333, pruned_loss=0.09037, over 28402.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.328, pruned_loss=0.08418, over 5672133.73 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3455, pruned_loss=0.109, over 5716447.82 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3267, pruned_loss=0.08191, over 5670752.60 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:28:55,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.148e+02 1.400e+03 1.885e+03 2.473e+03 5.417e+03, threshold=3.769e+03, percent-clipped=4.0
+2023-03-13 03:29:29,866 INFO [train.py:968] (0/2) Epoch 25, batch 33350, giga_loss[loss=0.2533, simple_loss=0.3356, pruned_loss=0.08549, over 28973.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3271, pruned_loss=0.08389, over 5671786.52 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3453, pruned_loss=0.1088, over 5720357.17 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3258, pruned_loss=0.08179, over 5666684.75 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:30:16,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:30:26,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:30:28,245 INFO [train.py:968] (0/2) Epoch 25, batch 33400, giga_loss[loss=0.2353, simple_loss=0.3264, pruned_loss=0.07204, over 28459.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.33, pruned_loss=0.0856, over 5666195.05 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3456, pruned_loss=0.1091, over 5707694.03 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3282, pruned_loss=0.08274, over 5672054.25 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:30:37,824 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:30:42,948 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:30:56,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.357e+02 1.365e+03 1.837e+03 2.660e+03 5.818e+03, threshold=3.675e+03, percent-clipped=13.0
+2023-03-13 03:31:01,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0334, 1.1495, 3.3458, 3.1157], device='cuda:0'), covar=tensor([0.1774, 0.2948, 0.0509, 0.0987], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0661, 0.0972, 0.0932], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 03:31:14,445 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 03:31:15,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:31:26,257 INFO [train.py:968] (0/2) Epoch 25, batch 33450, giga_loss[loss=0.2758, simple_loss=0.3461, pruned_loss=0.1027, over 28362.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3309, pruned_loss=0.08661, over 5662391.92 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3454, pruned_loss=0.109, over 5704253.25 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3291, pruned_loss=0.08369, over 5668286.40 frames. ], batch size: 369, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:31:53,050 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:32:04,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:32:28,186 INFO [train.py:968] (0/2) Epoch 25, batch 33500, giga_loss[loss=0.2601, simple_loss=0.3417, pruned_loss=0.0892, over 28944.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3316, pruned_loss=0.08769, over 5659940.19 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3453, pruned_loss=0.1089, over 5705637.11 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3301, pruned_loss=0.08502, over 5662506.57 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:32:30,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3299, 1.3861, 3.1626, 3.1557], device='cuda:0'), covar=tensor([0.1388, 0.2663, 0.0486, 0.1010], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0662, 0.0973, 0.0933], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 03:32:54,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3809, 1.5930, 1.6531, 1.2194], device='cuda:0'), covar=tensor([0.1734, 0.2684, 0.1470, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.0914, 0.0700, 0.0961, 0.0862], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 03:32:58,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.606e+02 1.530e+03 2.032e+03 2.843e+03 1.166e+04, threshold=4.065e+03, percent-clipped=15.0
+2023-03-13 03:33:08,958 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:33:11,046 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:33:17,072 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:33:19,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:33:23,578 INFO [train.py:968] (0/2) Epoch 25, batch 33550, giga_loss[loss=0.2679, simple_loss=0.3528, pruned_loss=0.09145, over 29060.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3355, pruned_loss=0.08988, over 5661162.48 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3449, pruned_loss=0.1088, over 5713175.25 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3341, pruned_loss=0.08713, over 5654432.54 frames. ], batch size: 285, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:33:38,765 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:33:48,595 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:34:03,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:34:14,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3992, 4.2501, 1.5456, 1.6377], device='cuda:0'), covar=tensor([0.1022, 0.0342, 0.1010, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0558, 0.0398, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 03:34:16,717 INFO [train.py:968] (0/2) Epoch 25, batch 33600, giga_loss[loss=0.2413, simple_loss=0.3339, pruned_loss=0.07433, over 28480.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3368, pruned_loss=0.08966, over 5667224.90 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.345, pruned_loss=0.1088, over 5715491.05 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3355, pruned_loss=0.087, over 5658570.34 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:34:38,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3766, 3.4035, 1.5548, 1.5064], device='cuda:0'), covar=tensor([0.0983, 0.0356, 0.0945, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0558, 0.0398, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 03:34:41,114 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:34:47,399 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:34:51,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.176e+02 1.466e+03 1.931e+03 2.621e+03 6.940e+03, threshold=3.863e+03, percent-clipped=7.0
+2023-03-13 03:34:52,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:35:26,395 INFO [train.py:968] (0/2) Epoch 25, batch 33650, giga_loss[loss=0.2067, simple_loss=0.297, pruned_loss=0.0582, over 29223.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3356, pruned_loss=0.08879, over 5674581.64 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3443, pruned_loss=0.1084, over 5719072.61 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3351, pruned_loss=0.08668, over 5663671.96 frames. ], batch size: 113, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:35:32,829 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:35:32,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5440, 2.0672, 1.3685, 0.9543], device='cuda:0'), covar=tensor([0.7742, 0.3691, 0.3787, 0.6648], device='cuda:0'), in_proj_covar=tensor([0.1800, 0.1691, 0.1628, 0.1465], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 03:36:08,560 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 03:36:19,306 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:36:20,148 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:36:30,151 INFO [train.py:968] (0/2) Epoch 25, batch 33700, giga_loss[loss=0.2395, simple_loss=0.3257, pruned_loss=0.07661, over 28905.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3328, pruned_loss=0.08741, over 5673875.44 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3443, pruned_loss=0.1085, over 5715210.90 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3321, pruned_loss=0.0851, over 5668209.95 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:36:59,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.825e+02 1.473e+03 1.830e+03 2.614e+03 1.118e+04, threshold=3.660e+03, percent-clipped=8.0
+2023-03-13 03:37:05,198 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:37:07,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:37:29,218 INFO [train.py:968] (0/2) Epoch 25, batch 33750, giga_loss[loss=0.2442, simple_loss=0.331, pruned_loss=0.07874, over 28452.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.332, pruned_loss=0.08682, over 5669657.95 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3442, pruned_loss=0.1084, over 5708249.96 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3312, pruned_loss=0.08447, over 5669610.15 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:37:30,812 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-13 03:37:42,728 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:38:30,894 INFO [train.py:968] (0/2) Epoch 25, batch 33800, giga_loss[loss=0.2431, simple_loss=0.3233, pruned_loss=0.08143, over 28609.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3316, pruned_loss=0.08716, over 5676887.50 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3441, pruned_loss=0.1083, over 5713764.12 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3306, pruned_loss=0.08476, over 5671195.77 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:39:03,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.820e+02 1.582e+03 2.103e+03 2.881e+03 5.887e+03, threshold=4.206e+03, percent-clipped=15.0
+2023-03-13 03:39:32,960 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-13 03:39:33,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:39:35,989 INFO [train.py:968] (0/2) Epoch 25, batch 33850, giga_loss[loss=0.2323, simple_loss=0.316, pruned_loss=0.0743, over 28216.00 frames. ], tot_loss[loss=0.252, simple_loss=0.33, pruned_loss=0.08704, over 5673020.27 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3442, pruned_loss=0.1085, over 5707337.31 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.329, pruned_loss=0.08478, over 5673300.80 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:40:33,140 INFO [train.py:968] (0/2) Epoch 25, batch 33900, giga_loss[loss=0.2176, simple_loss=0.3117, pruned_loss=0.06172, over 28952.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3294, pruned_loss=0.08612, over 5673153.22 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.344, pruned_loss=0.1084, over 5700124.99 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3283, pruned_loss=0.0838, over 5679672.89 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:41:02,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.153e+02 1.398e+03 1.822e+03 2.703e+03 5.560e+03, threshold=3.644e+03, percent-clipped=3.0
+2023-03-13 03:41:33,068 INFO [train.py:968] (0/2) Epoch 25, batch 33950, giga_loss[loss=0.2552, simple_loss=0.3329, pruned_loss=0.08877, over 28823.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3274, pruned_loss=0.08433, over 5668519.91 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3436, pruned_loss=0.108, over 5703367.48 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3266, pruned_loss=0.08231, over 5670401.56 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:41:43,334 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:42:06,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5273, 1.7014, 1.7792, 1.3356], device='cuda:0'), covar=tensor([0.2038, 0.2956, 0.1756, 0.2078], device='cuda:0'), in_proj_covar=tensor([0.0918, 0.0701, 0.0964, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 03:42:19,231 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:42:20,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:42:21,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:42:29,298 INFO [train.py:968] (0/2) Epoch 25, batch 34000, giga_loss[loss=0.2574, simple_loss=0.3301, pruned_loss=0.09235, over 24522.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3291, pruned_loss=0.08328, over 5651844.76 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3436, pruned_loss=0.1081, over 5685212.86 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3283, pruned_loss=0.08131, over 5669939.50 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:42:51,692 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:42:54,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.129e+02 1.394e+03 1.809e+03 2.332e+03 7.003e+03, threshold=3.618e+03, percent-clipped=6.0
+2023-03-13 03:43:23,669 INFO [train.py:968] (0/2) Epoch 25, batch 34050, giga_loss[loss=0.2455, simple_loss=0.3341, pruned_loss=0.07841, over 28961.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3301, pruned_loss=0.08291, over 5668644.38 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3429, pruned_loss=0.1078, over 5692948.44 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3296, pruned_loss=0.0807, over 5675398.17 frames. ], batch size: 285, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:43:31,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5912, 1.8330, 1.4986, 1.5662], device='cuda:0'), covar=tensor([0.2797, 0.2752, 0.3184, 0.2580], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1119, 0.1379, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 03:43:42,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:43:43,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:43:44,261 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 03:44:06,996 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:44:21,377 INFO [train.py:968] (0/2) Epoch 25, batch 34100, giga_loss[loss=0.2406, simple_loss=0.3299, pruned_loss=0.07563, over 28656.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3314, pruned_loss=0.08368, over 5670128.10 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3431, pruned_loss=0.1078, over 5692766.07 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3303, pruned_loss=0.08082, over 5675071.62 frames. ], batch size: 242, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:44:58,595 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.816e+02 1.495e+03 2.037e+03 2.587e+03 5.481e+03, threshold=4.074e+03, percent-clipped=9.0
+2023-03-13 03:45:12,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:45:16,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:45:31,380 INFO [train.py:968] (0/2) Epoch 25, batch 34150, giga_loss[loss=0.324, simple_loss=0.379, pruned_loss=0.1346, over 26805.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.332, pruned_loss=0.08421, over 5668210.90 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.343, pruned_loss=0.1078, over 5693868.24 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3312, pruned_loss=0.0819, over 5671054.64 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:45:45,576 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1128000.pt
+2023-03-13 03:45:53,330 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:46:05,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3657, 1.5400, 1.5223, 1.3636], device='cuda:0'), covar=tensor([0.2840, 0.2494, 0.2087, 0.2514], device='cuda:0'), in_proj_covar=tensor([0.1984, 0.1911, 0.1818, 0.1977], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 03:46:37,346 INFO [train.py:968] (0/2) Epoch 25, batch 34200, giga_loss[loss=0.2457, simple_loss=0.3358, pruned_loss=0.07782, over 27547.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3311, pruned_loss=0.0837, over 5667343.56 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3427, pruned_loss=0.1075, over 5699196.62 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3304, pruned_loss=0.08158, over 5664225.84 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:46:48,670 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:46:50,095 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:46:51,876 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1128051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:46:52,743 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1128052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:47:14,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.698e+02 1.412e+03 1.827e+03 2.534e+03 5.282e+03, threshold=3.654e+03, percent-clipped=5.0
+2023-03-13 03:47:34,045 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:47:35,612 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:47:50,486 INFO [train.py:968] (0/2) Epoch 25, batch 34250, giga_loss[loss=0.2, simple_loss=0.2767, pruned_loss=0.06165, over 24651.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3313, pruned_loss=0.08279, over 5659385.20 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3429, pruned_loss=0.1076, over 5690765.56 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3305, pruned_loss=0.08081, over 5663366.21 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:48:52,704 INFO [train.py:968] (0/2) Epoch 25, batch 34300, giga_loss[loss=0.2819, simple_loss=0.3486, pruned_loss=0.1076, over 28215.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3329, pruned_loss=0.08393, over 5668148.45 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3426, pruned_loss=0.1075, over 5695306.60 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3323, pruned_loss=0.08178, over 5666390.85 frames. ], batch size: 77, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:49:26,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.546e+02 1.557e+03 2.020e+03 2.618e+03 6.187e+03, threshold=4.039e+03, percent-clipped=6.0
+2023-03-13 03:49:39,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1128173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:49:57,545 INFO [train.py:968] (0/2) Epoch 25, batch 34350, giga_loss[loss=0.2848, simple_loss=0.368, pruned_loss=0.1008, over 28460.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3365, pruned_loss=0.08552, over 5674427.85 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3427, pruned_loss=0.1074, over 5697543.23 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3357, pruned_loss=0.08342, over 5670418.29 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:50:10,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4380, 1.6405, 1.3390, 1.7309], device='cuda:0'), covar=tensor([0.0719, 0.0308, 0.0322, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 03:50:27,212 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 03:51:00,191 INFO [train.py:968] (0/2) Epoch 25, batch 34400, giga_loss[loss=0.2561, simple_loss=0.3365, pruned_loss=0.08785, over 28723.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3354, pruned_loss=0.08546, over 5679951.63 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3423, pruned_loss=0.1071, over 5704057.91 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.335, pruned_loss=0.08349, over 5670286.31 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:51:33,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.371e+02 1.289e+03 1.882e+03 2.584e+03 6.465e+03, threshold=3.763e+03, percent-clipped=10.0
+2023-03-13 03:51:56,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1603, 1.4694, 1.3703, 1.1455], device='cuda:0'), covar=tensor([0.3033, 0.2352, 0.1618, 0.2487], device='cuda:0'), in_proj_covar=tensor([0.1981, 0.1910, 0.1812, 0.1972], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 03:52:09,080 INFO [train.py:968] (0/2) Epoch 25, batch 34450, giga_loss[loss=0.2513, simple_loss=0.343, pruned_loss=0.07985, over 28563.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.333, pruned_loss=0.08442, over 5678632.76 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3421, pruned_loss=0.107, over 5695463.03 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3328, pruned_loss=0.08274, over 5677426.32 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:52:28,943 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1128304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:52:45,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:52:49,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1128319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:53:15,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3422, 3.4354, 1.5972, 1.5066], device='cuda:0'), covar=tensor([0.1062, 0.0307, 0.0979, 0.1435], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0557, 0.0400, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 03:53:18,255 INFO [train.py:968] (0/2) Epoch 25, batch 34500, giga_loss[loss=0.2468, simple_loss=0.3318, pruned_loss=0.0809, over 28893.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3317, pruned_loss=0.08338, over 5667251.82 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3422, pruned_loss=0.1069, over 5686569.43 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3312, pruned_loss=0.08154, over 5673774.92 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:53:19,364 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-13 03:53:28,252 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:53:50,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.823e+02 1.273e+03 1.746e+03 2.344e+03 6.770e+03, threshold=3.493e+03, percent-clipped=8.0
+2023-03-13 03:54:12,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4612, 1.3685, 3.8819, 3.2730], device='cuda:0'), covar=tensor([0.1897, 0.3123, 0.0800, 0.1352], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0665, 0.0974, 0.0937], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 03:54:22,463 INFO [train.py:968] (0/2) Epoch 25, batch 34550, giga_loss[loss=0.255, simple_loss=0.3385, pruned_loss=0.08579, over 28681.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3304, pruned_loss=0.08246, over 5668248.01 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.342, pruned_loss=0.1068, over 5688583.28 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3301, pruned_loss=0.08095, over 5671449.72 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:55:20,045 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1128439.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 03:55:20,359 INFO [train.py:968] (0/2) Epoch 25, batch 34600, giga_loss[loss=0.278, simple_loss=0.3384, pruned_loss=0.1088, over 24426.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3325, pruned_loss=0.08416, over 5660766.56 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.342, pruned_loss=0.1069, over 5683698.55 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3321, pruned_loss=0.08248, over 5666609.11 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 03:55:20,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1128440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:55:23,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2116, 0.8541, 0.8731, 1.3779], device='cuda:0'), covar=tensor([0.0774, 0.0453, 0.0404, 0.0878], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 03:55:30,500 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:55:32,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1128450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:55:54,757 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.393e+02 1.684e+03 2.169e+03 2.992e+03 7.488e+03, threshold=4.338e+03, percent-clipped=16.0
+2023-03-13 03:56:06,721 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:56:18,321 INFO [train.py:968] (0/2) Epoch 25, batch 34650, giga_loss[loss=0.244, simple_loss=0.3302, pruned_loss=0.07891, over 28849.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3345, pruned_loss=0.08516, over 5669368.53 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3417, pruned_loss=0.1066, over 5687340.77 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3341, pruned_loss=0.08324, over 5669826.63 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 03:57:09,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1756, 4.0401, 3.8127, 1.8616], device='cuda:0'), covar=tensor([0.0657, 0.0770, 0.0764, 0.2054], device='cuda:0'), in_proj_covar=tensor([0.1256, 0.1153, 0.0974, 0.0726], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 03:57:13,984 INFO [train.py:968] (0/2) Epoch 25, batch 34700, giga_loss[loss=0.322, simple_loss=0.3798, pruned_loss=0.1321, over 28948.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.334, pruned_loss=0.08576, over 5680004.63 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3422, pruned_loss=0.1068, over 5692610.68 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3331, pruned_loss=0.08355, over 5675419.44 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 03:57:42,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.503e+02 1.453e+03 1.810e+03 2.497e+03 5.775e+03, threshold=3.621e+03, percent-clipped=4.0
+2023-03-13 03:58:04,408 INFO [train.py:968] (0/2) Epoch 25, batch 34750, giga_loss[loss=0.2192, simple_loss=0.2899, pruned_loss=0.07422, over 24310.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3322, pruned_loss=0.08631, over 5664304.66 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3421, pruned_loss=0.1069, over 5689336.06 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3312, pruned_loss=0.08379, over 5663410.60 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 03:58:57,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4145, 1.7932, 1.4023, 1.4487], device='cuda:0'), covar=tensor([0.2693, 0.2661, 0.3179, 0.2291], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1120, 0.1382, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 03:59:02,883 INFO [train.py:968] (0/2) Epoch 25, batch 34800, giga_loss[loss=0.2322, simple_loss=0.3171, pruned_loss=0.07359, over 28587.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3318, pruned_loss=0.08619, over 5669206.08 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3421, pruned_loss=0.107, over 5690449.38 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.331, pruned_loss=0.08404, over 5667354.22 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:59:32,665 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.476e+03 1.854e+03 2.841e+03 8.071e+03, threshold=3.708e+03, percent-clipped=17.0
+2023-03-13 03:59:52,776 INFO [train.py:968] (0/2) Epoch 25, batch 34850, giga_loss[loss=0.2874, simple_loss=0.364, pruned_loss=0.1054, over 27634.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3396, pruned_loss=0.09033, over 5660695.12 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3421, pruned_loss=0.107, over 5682969.73 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3389, pruned_loss=0.08842, over 5665653.65 frames. ], batch size: 474, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:00:40,392 INFO [train.py:968] (0/2) Epoch 25, batch 34900, giga_loss[loss=0.2995, simple_loss=0.359, pruned_loss=0.12, over 23961.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3486, pruned_loss=0.09567, over 5660589.08 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3421, pruned_loss=0.107, over 5684040.88 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.348, pruned_loss=0.0941, over 5663480.30 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:00:42,988 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-13 04:00:51,877 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1128755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:01:03,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.532e+02 1.378e+03 1.681e+03 2.440e+03 6.183e+03, threshold=3.363e+03, percent-clipped=5.0
+2023-03-13 04:01:21,637 INFO [train.py:968] (0/2) Epoch 25, batch 34950, giga_loss[loss=0.2745, simple_loss=0.3469, pruned_loss=0.101, over 28762.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3501, pruned_loss=0.09751, over 5662186.79 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3425, pruned_loss=0.1073, over 5678498.42 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3494, pruned_loss=0.09579, over 5669927.25 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:01:41,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4757, 2.1263, 1.5961, 0.7607], device='cuda:0'), covar=tensor([0.7886, 0.3591, 0.4381, 0.7540], device='cuda:0'), in_proj_covar=tensor([0.1806, 0.1699, 0.1633, 0.1470], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 04:01:43,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1128814.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:01:44,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1128815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:02:00,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3135, 3.6253, 2.4721, 1.3025], device='cuda:0'), covar=tensor([0.8954, 0.3380, 0.4087, 0.7584], device='cuda:0'), in_proj_covar=tensor([0.1806, 0.1700, 0.1633, 0.1470], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 04:02:03,238 INFO [train.py:968] (0/2) Epoch 25, batch 35000, giga_loss[loss=0.2171, simple_loss=0.2954, pruned_loss=0.06938, over 28613.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3434, pruned_loss=0.09509, over 5674937.67 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3426, pruned_loss=0.1074, over 5679907.82 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.343, pruned_loss=0.0933, over 5679246.94 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:02:29,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.432e+02 1.180e+03 1.708e+03 2.097e+03 9.408e+03, threshold=3.415e+03, percent-clipped=11.0
+2023-03-13 04:02:50,084 INFO [train.py:968] (0/2) Epoch 25, batch 35050, giga_loss[loss=0.2552, simple_loss=0.3215, pruned_loss=0.09444, over 28545.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3366, pruned_loss=0.0922, over 5667344.93 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3427, pruned_loss=0.1076, over 5671052.28 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3362, pruned_loss=0.09066, over 5678136.05 frames. ], batch size: 65, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:03:26,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4500, 1.6676, 1.4170, 1.6749], device='cuda:0'), covar=tensor([0.0790, 0.0348, 0.0356, 0.0882], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 04:03:30,519 INFO [train.py:968] (0/2) Epoch 25, batch 35100, giga_loss[loss=0.2039, simple_loss=0.2775, pruned_loss=0.06517, over 28604.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3292, pruned_loss=0.089, over 5671462.29 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3429, pruned_loss=0.1076, over 5672742.13 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3284, pruned_loss=0.08736, over 5678896.77 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:03:44,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128957.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 04:03:45,272 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:03:45,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4278, 1.2673, 3.6802, 3.0737], device='cuda:0'), covar=tensor([0.1581, 0.2957, 0.0456, 0.1258], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0667, 0.0981, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 04:03:46,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1128960.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 04:03:47,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1128961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:03:52,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.524e+02 1.148e+03 1.426e+03 2.029e+03 5.770e+03, threshold=2.853e+03, percent-clipped=6.0
+2023-03-13 04:03:52,528 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4299, 1.2061, 4.0291, 3.2972], device='cuda:0'), covar=tensor([0.1709, 0.3014, 0.0423, 0.0933], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0666, 0.0980, 0.0943], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 04:04:08,216 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128989.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:04:08,643 INFO [train.py:968] (0/2) Epoch 25, batch 35150, giga_loss[loss=0.2856, simple_loss=0.3325, pruned_loss=0.1194, over 26632.00 frames. ], tot_loss[loss=0.248, simple_loss=0.323, pruned_loss=0.08654, over 5684234.56 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3429, pruned_loss=0.1074, over 5680586.46 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3218, pruned_loss=0.08474, over 5683238.44 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:04:08,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:04:47,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9443, 3.7748, 3.6174, 1.5657], device='cuda:0'), covar=tensor([0.0769, 0.0998, 0.0877, 0.2248], device='cuda:0'), in_proj_covar=tensor([0.1258, 0.1159, 0.0975, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 04:04:51,259 INFO [train.py:968] (0/2) Epoch 25, batch 35200, giga_loss[loss=0.2327, simple_loss=0.3002, pruned_loss=0.08257, over 28698.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3179, pruned_loss=0.08449, over 5684020.39 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3432, pruned_loss=0.1075, over 5680416.04 frames. ], giga_tot_loss[loss=0.2405, simple_loss=0.3161, pruned_loss=0.08245, over 5683589.12 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:05:14,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.524e+02 1.144e+03 1.422e+03 1.944e+03 8.111e+03, threshold=2.843e+03, percent-clipped=14.0
+2023-03-13 04:05:30,070 INFO [train.py:968] (0/2) Epoch 25, batch 35250, giga_loss[loss=0.2272, simple_loss=0.3067, pruned_loss=0.0739, over 28857.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3144, pruned_loss=0.0829, over 5679704.95 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3433, pruned_loss=0.1075, over 5674847.60 frames. ], giga_tot_loss[loss=0.2371, simple_loss=0.3125, pruned_loss=0.08085, over 5685062.69 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:06:07,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1129130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:06:14,750 INFO [train.py:968] (0/2) Epoch 25, batch 35300, giga_loss[loss=0.2359, simple_loss=0.3095, pruned_loss=0.08115, over 29116.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3101, pruned_loss=0.08054, over 5689120.39 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3432, pruned_loss=0.1074, over 5675988.36 frames. ], giga_tot_loss[loss=0.2332, simple_loss=0.3085, pruned_loss=0.07889, over 5692282.55 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:06:38,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.386e+02 1.161e+03 1.499e+03 2.066e+03 4.806e+03, threshold=2.999e+03, percent-clipped=10.0
+2023-03-13 04:06:54,027 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1129189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:06:54,434 INFO [train.py:968] (0/2) Epoch 25, batch 35350, giga_loss[loss=0.2278, simple_loss=0.3012, pruned_loss=0.07718, over 28883.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3077, pruned_loss=0.07954, over 5689867.64 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3437, pruned_loss=0.1077, over 5669398.21 frames. ], giga_tot_loss[loss=0.2297, simple_loss=0.305, pruned_loss=0.07718, over 5698687.39 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:07:28,774 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1129227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:07:39,852 INFO [train.py:968] (0/2) Epoch 25, batch 35400, giga_loss[loss=0.2605, simple_loss=0.316, pruned_loss=0.1025, over 26486.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3042, pruned_loss=0.07784, over 5689931.65 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3438, pruned_loss=0.1077, over 5668834.18 frames. ], giga_tot_loss[loss=0.2264, simple_loss=0.3015, pruned_loss=0.07562, over 5697443.05 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:07:46,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1775, 2.5745, 1.2351, 1.3099], device='cuda:0'), covar=tensor([0.1071, 0.0416, 0.0992, 0.1482], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0558, 0.0399, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 04:07:53,150 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1129257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:08:01,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.632e+02 1.061e+03 1.299e+03 1.729e+03 3.260e+03, threshold=2.599e+03, percent-clipped=2.0
+2023-03-13 04:08:06,100 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1129273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:08:08,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1129276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:08:20,959 INFO [train.py:968] (0/2) Epoch 25, batch 35450, giga_loss[loss=0.1934, simple_loss=0.2692, pruned_loss=0.05883, over 28789.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3025, pruned_loss=0.07727, over 5688079.87 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3446, pruned_loss=0.1083, over 5664762.13 frames. ], giga_tot_loss[loss=0.2237, simple_loss=0.2987, pruned_loss=0.0743, over 5697705.84 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:08:36,188 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1129305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:08:54,679 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8466, 1.0570, 2.8535, 2.7839], device='cuda:0'), covar=tensor([0.1777, 0.2766, 0.0659, 0.1028], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0667, 0.0983, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 04:09:01,646 INFO [train.py:968] (0/2) Epoch 25, batch 35500, giga_loss[loss=0.2191, simple_loss=0.2919, pruned_loss=0.07311, over 27886.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3003, pruned_loss=0.07634, over 5693077.46 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3449, pruned_loss=0.1084, over 5670507.42 frames. ], giga_tot_loss[loss=0.2212, simple_loss=0.296, pruned_loss=0.07319, over 5695933.95 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:09:25,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.221e+02 1.073e+03 1.446e+03 1.985e+03 5.240e+03, threshold=2.891e+03, percent-clipped=20.0
+2023-03-13 04:09:44,246 INFO [train.py:968] (0/2) Epoch 25, batch 35550, giga_loss[loss=0.1875, simple_loss=0.2608, pruned_loss=0.05708, over 28892.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.2974, pruned_loss=0.07539, over 5689373.75 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3452, pruned_loss=0.1087, over 5674341.46 frames. ], giga_tot_loss[loss=0.2188, simple_loss=0.2932, pruned_loss=0.07219, over 5688613.71 frames. ], batch size: 93, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:10:10,975 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1129416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:10:32,718 INFO [train.py:968] (0/2) Epoch 25, batch 35600, giga_loss[loss=0.2385, simple_loss=0.3052, pruned_loss=0.08593, over 28873.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2948, pruned_loss=0.07449, over 5699019.61 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3453, pruned_loss=0.1087, over 5677871.92 frames. ], giga_tot_loss[loss=0.2168, simple_loss=0.2907, pruned_loss=0.0715, over 5695522.40 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:10:58,309 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.718e+02 1.113e+03 1.300e+03 1.741e+03 4.184e+03, threshold=2.601e+03, percent-clipped=3.0
+2023-03-13 04:11:04,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2649, 3.0888, 2.9333, 1.4762], device='cuda:0'), covar=tensor([0.0942, 0.1069, 0.0940, 0.2290], device='cuda:0'), in_proj_covar=tensor([0.1256, 0.1159, 0.0975, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 04:11:13,318 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-13 04:11:13,552 INFO [train.py:968] (0/2) Epoch 25, batch 35650, giga_loss[loss=0.2542, simple_loss=0.3357, pruned_loss=0.08635, over 28969.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3008, pruned_loss=0.07763, over 5698742.62 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3456, pruned_loss=0.1085, over 5686951.29 frames. ], giga_tot_loss[loss=0.2218, simple_loss=0.2953, pruned_loss=0.0741, over 5688243.09 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:12:03,054 INFO [train.py:968] (0/2) Epoch 25, batch 35700, giga_loss[loss=0.3061, simple_loss=0.3761, pruned_loss=0.118, over 28485.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3125, pruned_loss=0.08336, over 5693274.91 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3457, pruned_loss=0.1085, over 5688051.54 frames. ], giga_tot_loss[loss=0.2345, simple_loss=0.308, pruned_loss=0.08049, over 5684161.74 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:12:25,561 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1129564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:12:25,744 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-13 04:12:30,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.786e+02 1.374e+03 1.768e+03 2.544e+03 6.664e+03, threshold=3.537e+03, percent-clipped=23.0
+2023-03-13 04:12:44,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3141, 1.5609, 1.4125, 1.4225], device='cuda:0'), covar=tensor([0.1888, 0.1915, 0.2394, 0.1885], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0749, 0.0720, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 04:12:48,082 INFO [train.py:968] (0/2) Epoch 25, batch 35750, giga_loss[loss=0.3418, simple_loss=0.4027, pruned_loss=0.1405, over 27947.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3258, pruned_loss=0.09022, over 5699294.87 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3462, pruned_loss=0.1088, over 5691866.84 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3212, pruned_loss=0.08722, over 5688825.00 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:12:58,328 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1129602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:13:25,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1129632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:13:30,899 INFO [train.py:968] (0/2) Epoch 25, batch 35800, giga_loss[loss=0.3301, simple_loss=0.3861, pruned_loss=0.1371, over 26515.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3346, pruned_loss=0.0942, over 5695054.96 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3464, pruned_loss=0.1088, over 5694875.64 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3306, pruned_loss=0.09156, over 5684296.18 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:13:56,869 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.485e+02 1.332e+03 1.700e+03 2.351e+03 7.236e+03, threshold=3.399e+03, percent-clipped=7.0
+2023-03-13 04:14:06,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5601, 1.7963, 1.4589, 1.5185], device='cuda:0'), covar=tensor([0.2811, 0.2913, 0.3316, 0.2541], device='cuda:0'), in_proj_covar=tensor([0.1564, 0.1125, 0.1384, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 04:14:14,299 INFO [train.py:968] (0/2) Epoch 25, batch 35850, giga_loss[loss=0.2597, simple_loss=0.3376, pruned_loss=0.09092, over 28683.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3381, pruned_loss=0.09462, over 5676025.40 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3463, pruned_loss=0.1088, over 5679155.01 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3348, pruned_loss=0.09231, over 5681047.66 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:14:27,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1129707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:14:32,214 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1129710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:14:51,075 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3796, 1.9603, 1.5742, 1.4830], device='cuda:0'), covar=tensor([0.0819, 0.0307, 0.0343, 0.0927], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 04:14:58,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1129739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:14:59,418 INFO [train.py:968] (0/2) Epoch 25, batch 35900, giga_loss[loss=0.2666, simple_loss=0.3476, pruned_loss=0.09284, over 28706.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3403, pruned_loss=0.09442, over 5686535.20 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3468, pruned_loss=0.1089, over 5683262.76 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3371, pruned_loss=0.09225, over 5687055.62 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:14:59,601 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1129740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:04,105 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1129745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:06,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1129748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:27,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.997e+02 1.201e+03 1.490e+03 2.013e+03 7.692e+03, threshold=2.981e+03, percent-clipped=3.0
+2023-03-13 04:15:32,588 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1129775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:33,860 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1129777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:34,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1129778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:44,207 INFO [train.py:968] (0/2) Epoch 25, batch 35950, giga_loss[loss=0.3531, simple_loss=0.3969, pruned_loss=0.1546, over 26578.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3437, pruned_loss=0.09641, over 5691395.25 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3475, pruned_loss=0.1093, over 5686529.55 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3405, pruned_loss=0.09411, over 5688988.78 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:15:45,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1129791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:58,554 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1129807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:16:26,787 INFO [train.py:968] (0/2) Epoch 25, batch 36000, giga_loss[loss=0.3282, simple_loss=0.3761, pruned_loss=0.1402, over 23526.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3472, pruned_loss=0.09922, over 5679696.79 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3476, pruned_loss=0.1094, over 5686212.67 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3445, pruned_loss=0.09721, over 5678434.28 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:16:26,792 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 04:16:35,860 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2036, simple_loss=0.3113, pruned_loss=0.04791, over 944034.00 frames. 
+2023-03-13 04:16:35,860 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 04:16:56,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4953, 1.7385, 1.4340, 1.4776], device='cuda:0'), covar=tensor([0.2794, 0.2740, 0.3039, 0.2429], device='cuda:0'), in_proj_covar=tensor([0.1565, 0.1127, 0.1384, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 04:16:59,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.245e+02 1.248e+03 1.500e+03 1.903e+03 4.537e+03, threshold=2.999e+03, percent-clipped=8.0
+2023-03-13 04:17:16,799 INFO [train.py:968] (0/2) Epoch 25, batch 36050, giga_loss[loss=0.296, simple_loss=0.3705, pruned_loss=0.1108, over 28995.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3512, pruned_loss=0.1019, over 5687742.99 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3483, pruned_loss=0.1096, over 5691347.66 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3485, pruned_loss=0.09985, over 5681968.59 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:17:50,428 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5772, 2.0680, 1.4911, 0.9109], device='cuda:0'), covar=tensor([0.6738, 0.3651, 0.4651, 0.6902], device='cuda:0'), in_proj_covar=tensor([0.1807, 0.1698, 0.1634, 0.1466], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 04:17:51,819 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1129934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:17:54,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1129937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:17:56,439 INFO [train.py:968] (0/2) Epoch 25, batch 36100, giga_loss[loss=0.2753, simple_loss=0.3571, pruned_loss=0.09678, over 28593.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3533, pruned_loss=0.1023, over 5681331.31 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3487, pruned_loss=0.1099, over 5677131.31 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3508, pruned_loss=0.1003, over 5687756.18 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:18:05,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5416, 1.5193, 1.5504, 1.1842], device='cuda:0'), covar=tensor([0.1678, 0.3055, 0.1435, 0.1579], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0706, 0.0972, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 04:18:09,534 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3296, 1.9066, 1.4080, 0.6045], device='cuda:0'), covar=tensor([0.6488, 0.3407, 0.5046, 0.7230], device='cuda:0'), in_proj_covar=tensor([0.1808, 0.1698, 0.1635, 0.1467], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 04:18:16,451 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1129966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:18:19,141 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6779, 0.9888, 2.8026, 2.7090], device='cuda:0'), covar=tensor([0.2342, 0.3243, 0.1041, 0.1430], device='cuda:0'), in_proj_covar=tensor([0.0779, 0.0661, 0.0973, 0.0939], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 04:18:21,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.137e+02 1.340e+03 1.742e+03 2.218e+03 8.016e+03, threshold=3.484e+03, percent-clipped=9.0
+2023-03-13 04:18:21,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6592, 1.8919, 1.8290, 1.7692], device='cuda:0'), covar=tensor([0.2033, 0.2049, 0.2286, 0.2016], device='cuda:0'), in_proj_covar=tensor([0.0481, 0.0746, 0.0719, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 04:18:37,566 INFO [train.py:968] (0/2) Epoch 25, batch 36150, giga_loss[loss=0.2931, simple_loss=0.3699, pruned_loss=0.1082, over 28310.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3554, pruned_loss=0.103, over 5678339.46 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3489, pruned_loss=0.1099, over 5678558.72 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3534, pruned_loss=0.1012, over 5682172.64 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:18:46,086 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1130000.pt
+2023-03-13 04:19:15,267 INFO [train.py:968] (0/2) Epoch 25, batch 36200, giga_loss[loss=0.2541, simple_loss=0.3423, pruned_loss=0.083, over 28890.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.356, pruned_loss=0.1027, over 5663121.21 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3497, pruned_loss=0.1104, over 5661777.34 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3539, pruned_loss=0.1006, over 5680763.29 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:19:40,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.672e+02 1.222e+03 1.524e+03 1.911e+03 5.458e+03, threshold=3.049e+03, percent-clipped=3.0
+2023-03-13 04:19:54,458 INFO [train.py:968] (0/2) Epoch 25, batch 36250, giga_loss[loss=0.2787, simple_loss=0.3606, pruned_loss=0.09834, over 27931.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3556, pruned_loss=0.1011, over 5675600.33 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3494, pruned_loss=0.1101, over 5663041.94 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3543, pruned_loss=0.09949, over 5688689.12 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:20:13,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1130115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:20:24,299 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:20:26,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:20:33,198 INFO [train.py:968] (0/2) Epoch 25, batch 36300, giga_loss[loss=0.2493, simple_loss=0.3356, pruned_loss=0.08155, over 28245.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3535, pruned_loss=0.09868, over 5683011.44 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3495, pruned_loss=0.11, over 5666047.82 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3526, pruned_loss=0.09728, over 5691386.63 frames. ], batch size: 77, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:20:58,340 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.329e+02 1.189e+03 1.466e+03 1.947e+03 5.806e+03, threshold=2.933e+03, percent-clipped=7.0
+2023-03-13 04:21:04,694 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:21:11,555 INFO [train.py:968] (0/2) Epoch 25, batch 36350, giga_loss[loss=0.3023, simple_loss=0.3679, pruned_loss=0.1184, over 28860.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3518, pruned_loss=0.09697, over 5698977.16 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3498, pruned_loss=0.1101, over 5671621.45 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3508, pruned_loss=0.09548, over 5701375.20 frames. ], batch size: 199, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:21:11,849 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0073, 1.5687, 1.3768, 1.3916], device='cuda:0'), covar=tensor([0.2845, 0.1939, 0.2931, 0.2231], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0751, 0.0723, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 04:21:42,938 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-13 04:21:50,687 INFO [train.py:968] (0/2) Epoch 25, batch 36400, giga_loss[loss=0.3064, simple_loss=0.378, pruned_loss=0.1174, over 28852.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3526, pruned_loss=0.09813, over 5711499.59 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3498, pruned_loss=0.1099, over 5678945.14 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3518, pruned_loss=0.09652, over 5708062.71 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:22:04,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2682, 2.3093, 1.9628, 1.6406], device='cuda:0'), covar=tensor([0.0886, 0.0259, 0.0274, 0.1081], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 04:22:07,386 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1130258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:22:09,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1130261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:22:20,347 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.461e+02 1.317e+03 1.638e+03 2.012e+03 3.921e+03, threshold=3.276e+03, percent-clipped=7.0
+2023-03-13 04:22:28,935 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-13 04:22:36,619 INFO [train.py:968] (0/2) Epoch 25, batch 36450, giga_loss[loss=0.2833, simple_loss=0.3518, pruned_loss=0.1074, over 28923.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3546, pruned_loss=0.1022, over 5709453.07 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3496, pruned_loss=0.1096, over 5687137.04 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3543, pruned_loss=0.1008, over 5700195.31 frames. ], batch size: 66, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:22:36,813 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1130290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:23:19,532 INFO [train.py:968] (0/2) Epoch 25, batch 36500, libri_loss[loss=0.3092, simple_loss=0.3707, pruned_loss=0.1238, over 29547.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3557, pruned_loss=0.1047, over 5703245.02 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3498, pruned_loss=0.1096, over 5688478.51 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3554, pruned_loss=0.1036, over 5694798.03 frames. ], batch size: 89, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:23:47,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.493e+03 1.826e+03 2.702e+03 9.835e+03, threshold=3.652e+03, percent-clipped=16.0
+2023-03-13 04:24:03,303 INFO [train.py:968] (0/2) Epoch 25, batch 36550, giga_loss[loss=0.2759, simple_loss=0.3499, pruned_loss=0.101, over 28279.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3542, pruned_loss=0.1044, over 5713622.31 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.35, pruned_loss=0.1098, over 5696249.27 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.354, pruned_loss=0.1031, over 5700251.74 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:24:44,744 INFO [train.py:968] (0/2) Epoch 25, batch 36600, giga_loss[loss=0.2434, simple_loss=0.3247, pruned_loss=0.08104, over 28605.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3517, pruned_loss=0.1035, over 5710831.67 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3504, pruned_loss=0.1099, over 5699088.09 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3512, pruned_loss=0.1023, over 5697888.54 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:25:08,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.876e+02 1.416e+03 1.699e+03 2.465e+03 7.487e+03, threshold=3.398e+03, percent-clipped=11.0
+2023-03-13 04:25:24,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7196, 2.3390, 1.4120, 0.8310], device='cuda:0'), covar=tensor([0.8791, 0.4003, 0.4573, 0.8064], device='cuda:0'), in_proj_covar=tensor([0.1815, 0.1706, 0.1638, 0.1471], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 04:25:25,893 INFO [train.py:968] (0/2) Epoch 25, batch 36650, giga_loss[loss=0.2581, simple_loss=0.3447, pruned_loss=0.08576, over 28730.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3509, pruned_loss=0.1027, over 5708962.48 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3507, pruned_loss=0.11, over 5699771.39 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3503, pruned_loss=0.1015, over 5698116.03 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:25:37,964 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1130503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:25:40,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1130506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:26:07,225 INFO [train.py:968] (0/2) Epoch 25, batch 36700, giga_loss[loss=0.2237, simple_loss=0.3104, pruned_loss=0.06853, over 28478.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3498, pruned_loss=0.1014, over 5703495.12 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3512, pruned_loss=0.1103, over 5699311.37 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3489, pruned_loss=0.09986, over 5695479.02 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:26:22,569 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1130556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:26:34,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6224, 1.8220, 1.4679, 2.0385], device='cuda:0'), covar=tensor([0.2821, 0.2978, 0.3295, 0.2534], device='cuda:0'), in_proj_covar=tensor([0.1559, 0.1122, 0.1378, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 04:26:35,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.102e+02 1.305e+03 1.711e+03 2.049e+03 5.081e+03, threshold=3.422e+03, percent-clipped=2.0
+2023-03-13 04:26:51,420 INFO [train.py:968] (0/2) Epoch 25, batch 36750, giga_loss[loss=0.2499, simple_loss=0.3109, pruned_loss=0.09445, over 23579.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3475, pruned_loss=0.1002, over 5695162.40 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3511, pruned_loss=0.1102, over 5707127.95 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3466, pruned_loss=0.09866, over 5681331.06 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:27:17,111 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:27:35,504 INFO [train.py:968] (0/2) Epoch 25, batch 36800, giga_loss[loss=0.2272, simple_loss=0.3048, pruned_loss=0.07485, over 28966.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3418, pruned_loss=0.09707, over 5687786.29 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3511, pruned_loss=0.1102, over 5710230.04 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3411, pruned_loss=0.09573, over 5673929.33 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:27:43,382 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1130646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:27:45,226 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:27:45,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1130649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:27:45,326 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1130649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:27:47,149 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1130652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:28:04,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.608e+02 1.043e+03 1.391e+03 1.794e+03 6.155e+03, threshold=2.782e+03, percent-clipped=2.0
+2023-03-13 04:28:13,559 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1130678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:28:16,848 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1130681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:28:26,959 INFO [train.py:968] (0/2) Epoch 25, batch 36850, giga_loss[loss=0.2582, simple_loss=0.3299, pruned_loss=0.09329, over 28291.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3359, pruned_loss=0.0943, over 5669792.43 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3513, pruned_loss=0.1103, over 5711854.69 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.335, pruned_loss=0.09305, over 5657222.58 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:28:38,090 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1130699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:28:41,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1130702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:29:04,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1643, 5.0102, 4.7484, 2.3254], device='cuda:0'), covar=tensor([0.0421, 0.0547, 0.0602, 0.1962], device='cuda:0'), in_proj_covar=tensor([0.1248, 0.1156, 0.0969, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 04:29:05,376 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1130731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:29:13,146 INFO [train.py:968] (0/2) Epoch 25, batch 36900, giga_loss[loss=0.2686, simple_loss=0.3462, pruned_loss=0.09554, over 28942.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3339, pruned_loss=0.09405, over 5658802.94 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.352, pruned_loss=0.1107, over 5707135.29 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3321, pruned_loss=0.09207, over 5651607.87 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:29:44,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.725e+02 1.073e+03 1.435e+03 2.073e+03 6.967e+03, threshold=2.869e+03, percent-clipped=13.0
+2023-03-13 04:29:55,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3019, 1.5994, 1.5590, 1.1433], device='cuda:0'), covar=tensor([0.1461, 0.2637, 0.1355, 0.1618], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0707, 0.0970, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 04:29:57,947 INFO [train.py:968] (0/2) Epoch 25, batch 36950, giga_loss[loss=0.2431, simple_loss=0.3251, pruned_loss=0.08055, over 28396.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3339, pruned_loss=0.09287, over 5673819.57 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.352, pruned_loss=0.1107, over 5709275.28 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3323, pruned_loss=0.0912, over 5665874.85 frames. ], batch size: 65, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:30:35,367 INFO [train.py:968] (0/2) Epoch 25, batch 37000, giga_loss[loss=0.2369, simple_loss=0.3134, pruned_loss=0.08017, over 28633.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3343, pruned_loss=0.09271, over 5669133.65 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3524, pruned_loss=0.1107, over 5703803.41 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.332, pruned_loss=0.09073, over 5665959.07 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:30:59,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6725, 4.5351, 4.3113, 1.9537], device='cuda:0'), covar=tensor([0.0619, 0.0760, 0.0720, 0.2170], device='cuda:0'), in_proj_covar=tensor([0.1248, 0.1156, 0.0968, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 04:31:04,243 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.442e+02 1.146e+03 1.430e+03 2.050e+03 5.824e+03, threshold=2.860e+03, percent-clipped=9.0
+2023-03-13 04:31:17,793 INFO [train.py:968] (0/2) Epoch 25, batch 37050, giga_loss[loss=0.2237, simple_loss=0.3027, pruned_loss=0.07238, over 28714.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.332, pruned_loss=0.09111, over 5686438.82 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3527, pruned_loss=0.1108, over 5704638.81 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3297, pruned_loss=0.08933, over 5682935.63 frames. ], batch size: 99, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:31:53,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:31:58,841 INFO [train.py:968] (0/2) Epoch 25, batch 37100, giga_loss[loss=0.2447, simple_loss=0.3159, pruned_loss=0.08677, over 28907.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3293, pruned_loss=0.08968, over 5699377.43 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3531, pruned_loss=0.1109, over 5703142.38 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3269, pruned_loss=0.08792, over 5698109.28 frames. ], batch size: 112, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:32:05,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130947.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 04:32:23,348 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.137e+02 1.142e+03 1.433e+03 1.997e+03 5.385e+03, threshold=2.866e+03, percent-clipped=11.0
+2023-03-13 04:32:36,012 INFO [train.py:968] (0/2) Epoch 25, batch 37150, giga_loss[loss=0.2382, simple_loss=0.3067, pruned_loss=0.08484, over 28543.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3286, pruned_loss=0.08964, over 5707420.38 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3541, pruned_loss=0.1111, over 5712469.88 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3246, pruned_loss=0.08707, over 5697891.96 frames. ], batch size: 78, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:32:36,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4293, 2.1423, 1.6983, 0.6866], device='cuda:0'), covar=tensor([0.6154, 0.2822, 0.4403, 0.6687], device='cuda:0'), in_proj_covar=tensor([0.1804, 0.1693, 0.1629, 0.1460], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 04:32:39,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1130994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:33:02,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1131024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:33:14,638 INFO [train.py:968] (0/2) Epoch 25, batch 37200, giga_loss[loss=0.2428, simple_loss=0.3139, pruned_loss=0.08583, over 28800.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3256, pruned_loss=0.08825, over 5714776.03 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3541, pruned_loss=0.1109, over 5716665.55 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3218, pruned_loss=0.08584, over 5703296.03 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:33:40,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.679e+02 1.046e+03 1.191e+03 1.832e+03 4.775e+03, threshold=2.382e+03, percent-clipped=8.0
+2023-03-13 04:33:46,319 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 04:33:49,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1131083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:33:53,964 INFO [train.py:968] (0/2) Epoch 25, batch 37250, giga_loss[loss=0.2072, simple_loss=0.2887, pruned_loss=0.06286, over 28929.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3244, pruned_loss=0.08786, over 5717853.66 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3544, pruned_loss=0.111, over 5716824.39 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3206, pruned_loss=0.08544, over 5708573.42 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:33:54,848 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1131091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:34:33,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 04:34:34,283 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:34:35,914 INFO [train.py:968] (0/2) Epoch 25, batch 37300, giga_loss[loss=0.2519, simple_loss=0.333, pruned_loss=0.0854, over 28321.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3235, pruned_loss=0.08764, over 5711847.40 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3553, pruned_loss=0.1116, over 5709841.22 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.319, pruned_loss=0.08479, over 5711670.40 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:34:36,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:34:57,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:34:58,875 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:34:59,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:35:03,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.138e+02 1.122e+03 1.391e+03 1.773e+03 4.844e+03, threshold=2.781e+03, percent-clipped=11.0
+2023-03-13 04:35:14,349 INFO [train.py:968] (0/2) Epoch 25, batch 37350, giga_loss[loss=0.2332, simple_loss=0.3062, pruned_loss=0.08008, over 28842.00 frames. ], tot_loss[loss=0.248, simple_loss=0.322, pruned_loss=0.08698, over 5716892.48 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3555, pruned_loss=0.1114, over 5715406.82 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3174, pruned_loss=0.08424, over 5712015.77 frames. ], batch size: 199, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:35:21,150 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:35:33,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3978, 1.8004, 1.6001, 1.5745], device='cuda:0'), covar=tensor([0.2557, 0.2294, 0.2767, 0.2566], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0752, 0.0725, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 04:35:53,299 INFO [train.py:968] (0/2) Epoch 25, batch 37400, giga_loss[loss=0.2294, simple_loss=0.31, pruned_loss=0.0744, over 28747.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3211, pruned_loss=0.08617, over 5713967.59 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3561, pruned_loss=0.1114, over 5714234.67 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3161, pruned_loss=0.08338, over 5711458.04 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:36:19,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.654e+02 1.068e+03 1.336e+03 1.722e+03 4.359e+03, threshold=2.672e+03, percent-clipped=5.0
+2023-03-13 04:36:29,135 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-13 04:36:33,235 INFO [train.py:968] (0/2) Epoch 25, batch 37450, giga_loss[loss=0.2501, simple_loss=0.3164, pruned_loss=0.09188, over 24175.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3211, pruned_loss=0.08615, over 5708799.30 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3564, pruned_loss=0.1113, over 5714846.37 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3156, pruned_loss=0.08315, over 5706377.19 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:36:45,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1131307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:36:58,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1131322.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:37:12,208 INFO [train.py:968] (0/2) Epoch 25, batch 37500, giga_loss[loss=0.2472, simple_loss=0.328, pruned_loss=0.0832, over 28969.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3211, pruned_loss=0.08592, over 5718018.91 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3563, pruned_loss=0.1111, over 5718194.37 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.3161, pruned_loss=0.0832, over 5713157.43 frames. ], batch size: 164, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:37:40,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.898e+02 1.329e+03 1.823e+03 2.726e+03 7.974e+03, threshold=3.645e+03, percent-clipped=25.0
+2023-03-13 04:37:53,211 INFO [train.py:968] (0/2) Epoch 25, batch 37550, giga_loss[loss=0.2917, simple_loss=0.3608, pruned_loss=0.1113, over 27636.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3263, pruned_loss=0.08918, over 5722426.75 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3571, pruned_loss=0.1113, over 5723674.39 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3203, pruned_loss=0.08594, over 5713673.33 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:38:37,485 INFO [train.py:968] (0/2) Epoch 25, batch 37600, giga_loss[loss=0.3165, simple_loss=0.3812, pruned_loss=0.1259, over 28587.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3322, pruned_loss=0.09298, over 5709277.42 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3575, pruned_loss=0.1115, over 5716893.76 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3266, pruned_loss=0.08984, over 5708192.41 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:38:48,998 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:38:51,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:38:55,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1131458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:39:03,269 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131465.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 04:39:03,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1131466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:39:06,487 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131468.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:39:10,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.189e+02 1.402e+03 1.680e+03 2.196e+03 4.862e+03, threshold=3.360e+03, percent-clipped=9.0
+2023-03-13 04:39:18,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:39:25,052 INFO [train.py:968] (0/2) Epoch 25, batch 37650, giga_loss[loss=0.2555, simple_loss=0.3343, pruned_loss=0.08838, over 28839.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3403, pruned_loss=0.09857, over 5703213.01 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3571, pruned_loss=0.1112, over 5719759.60 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.336, pruned_loss=0.09616, over 5699746.36 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:39:33,070 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131497.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:40:07,525 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 04:40:14,180 INFO [train.py:968] (0/2) Epoch 25, batch 37700, giga_loss[loss=0.3445, simple_loss=0.4033, pruned_loss=0.1428, over 28720.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3449, pruned_loss=0.1006, over 5686952.95 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3572, pruned_loss=0.1112, over 5718428.61 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3411, pruned_loss=0.09846, over 5685081.76 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:40:39,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.448e+02 1.358e+03 1.726e+03 2.253e+03 7.853e+03, threshold=3.453e+03, percent-clipped=8.0
+2023-03-13 04:40:55,245 INFO [train.py:968] (0/2) Epoch 25, batch 37750, giga_loss[loss=0.2703, simple_loss=0.3523, pruned_loss=0.09414, over 28717.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3498, pruned_loss=0.1026, over 5692910.92 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3574, pruned_loss=0.1115, over 5722641.69 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3461, pruned_loss=0.1002, over 5686327.07 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:41:05,287 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:41:08,004 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:41:11,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:41:15,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:41:34,697 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:41:39,343 INFO [train.py:968] (0/2) Epoch 25, batch 37800, giga_loss[loss=0.337, simple_loss=0.402, pruned_loss=0.136, over 27883.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3561, pruned_loss=0.1064, over 5689081.19 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3573, pruned_loss=0.1114, over 5725642.38 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3533, pruned_loss=0.1045, over 5680639.34 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:41:40,263 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:42:01,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0442, 1.0996, 3.2766, 3.0458], device='cuda:0'), covar=tensor([0.1788, 0.3063, 0.0466, 0.0973], device='cuda:0'), in_proj_covar=tensor([0.0776, 0.0656, 0.0968, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 04:42:07,000 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-13 04:42:09,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.596e+02 1.343e+03 1.693e+03 2.295e+03 9.931e+03, threshold=3.387e+03, percent-clipped=10.0
+2023-03-13 04:42:20,850 INFO [train.py:968] (0/2) Epoch 25, batch 37850, giga_loss[loss=0.2572, simple_loss=0.3399, pruned_loss=0.08729, over 28693.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.356, pruned_loss=0.1056, over 5688422.34 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3573, pruned_loss=0.1116, over 5722994.71 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3537, pruned_loss=0.1039, over 5683535.63 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:42:32,237 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.24 vs. limit=5.0
+2023-03-13 04:43:02,928 INFO [train.py:968] (0/2) Epoch 25, batch 37900, giga_loss[loss=0.2677, simple_loss=0.3468, pruned_loss=0.09435, over 28838.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3525, pruned_loss=0.1025, over 5696124.25 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3574, pruned_loss=0.1116, over 5724960.54 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3506, pruned_loss=0.1011, over 5690369.53 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:43:32,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.976e+02 1.256e+03 1.524e+03 2.032e+03 5.142e+03, threshold=3.047e+03, percent-clipped=1.0
+2023-03-13 04:43:45,075 INFO [train.py:968] (0/2) Epoch 25, batch 37950, giga_loss[loss=0.2901, simple_loss=0.3731, pruned_loss=0.1036, over 28711.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3502, pruned_loss=0.1003, over 5689387.35 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3581, pruned_loss=0.1121, over 5718032.55 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3481, pruned_loss=0.09856, over 5691175.72 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:44:13,563 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3436, 2.0633, 1.5134, 0.5814], device='cuda:0'), covar=tensor([0.5580, 0.2714, 0.4421, 0.6459], device='cuda:0'), in_proj_covar=tensor([0.1804, 0.1692, 0.1632, 0.1464], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 04:44:20,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3792, 1.2451, 4.1678, 3.3713], device='cuda:0'), covar=tensor([0.1503, 0.2503, 0.0428, 0.0928], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0657, 0.0968, 0.0936], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 04:44:23,956 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1131835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:44:26,985 INFO [train.py:968] (0/2) Epoch 25, batch 38000, giga_loss[loss=0.2333, simple_loss=0.3177, pruned_loss=0.07446, over 28470.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3501, pruned_loss=0.1001, over 5698086.66 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3579, pruned_loss=0.112, over 5721011.67 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3485, pruned_loss=0.09863, over 5696480.99 frames. ], batch size: 71, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:44:36,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6230, 4.4645, 4.2118, 2.1770], device='cuda:0'), covar=tensor([0.0510, 0.0625, 0.0625, 0.2081], device='cuda:0'), in_proj_covar=tensor([0.1260, 0.1166, 0.0977, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 04:44:57,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.583e+02 1.416e+03 1.751e+03 2.755e+03 1.093e+04, threshold=3.502e+03, percent-clipped=19.0
+2023-03-13 04:45:03,021 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-13 04:45:08,149 INFO [train.py:968] (0/2) Epoch 25, batch 38050, giga_loss[loss=0.2953, simple_loss=0.3645, pruned_loss=0.1131, over 28978.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3518, pruned_loss=0.1009, over 5691733.76 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3577, pruned_loss=0.112, over 5715873.04 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3504, pruned_loss=0.0995, over 5694672.90 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:45:24,598 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1131910.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:45:50,237 INFO [train.py:968] (0/2) Epoch 25, batch 38100, giga_loss[loss=0.2535, simple_loss=0.3358, pruned_loss=0.08555, over 28456.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3544, pruned_loss=0.1031, over 5696329.21 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3583, pruned_loss=0.1124, over 5718225.69 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3528, pruned_loss=0.1014, over 5696363.02 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:46:20,097 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.664e+02 1.326e+03 1.819e+03 2.635e+03 1.014e+04, threshold=3.638e+03, percent-clipped=13.0
+2023-03-13 04:46:33,045 INFO [train.py:968] (0/2) Epoch 25, batch 38150, giga_loss[loss=0.2944, simple_loss=0.3487, pruned_loss=0.1201, over 23726.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3572, pruned_loss=0.1054, over 5694436.22 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3591, pruned_loss=0.1127, over 5723428.39 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3551, pruned_loss=0.1035, over 5688915.23 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:46:43,152 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1132000.pt
+2023-03-13 04:47:19,304 INFO [train.py:968] (0/2) Epoch 25, batch 38200, giga_loss[loss=0.3032, simple_loss=0.3703, pruned_loss=0.1181, over 28818.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3561, pruned_loss=0.1048, over 5700735.90 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3593, pruned_loss=0.1127, over 5724088.25 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3542, pruned_loss=0.1033, over 5695754.96 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:47:39,778 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 04:47:47,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.582e+02 1.322e+03 1.653e+03 2.124e+03 4.676e+03, threshold=3.306e+03, percent-clipped=5.0
+2023-03-13 04:47:59,755 INFO [train.py:968] (0/2) Epoch 25, batch 38250, giga_loss[loss=0.277, simple_loss=0.3515, pruned_loss=0.1013, over 28904.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3566, pruned_loss=0.1056, over 5694114.36 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1128, over 5724216.83 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3549, pruned_loss=0.1042, over 5689902.97 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:48:01,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1440, 1.2359, 3.8288, 3.1700], device='cuda:0'), covar=tensor([0.1818, 0.2917, 0.0434, 0.0977], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0657, 0.0972, 0.0939], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 04:48:41,236 INFO [train.py:968] (0/2) Epoch 25, batch 38300, giga_loss[loss=0.2767, simple_loss=0.3541, pruned_loss=0.09966, over 28382.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3565, pruned_loss=0.1051, over 5701294.23 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5727105.50 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3553, pruned_loss=0.104, over 5694945.69 frames. ], batch size: 65, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:49:08,563 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1132173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:49:10,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.343e+02 1.240e+03 1.574e+03 2.060e+03 6.423e+03, threshold=3.148e+03, percent-clipped=4.0
+2023-03-13 04:49:21,522 INFO [train.py:968] (0/2) Epoch 25, batch 38350, giga_loss[loss=0.2374, simple_loss=0.3246, pruned_loss=0.07514, over 28804.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3559, pruned_loss=0.1038, over 5705347.02 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3597, pruned_loss=0.1129, over 5731625.15 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3546, pruned_loss=0.1025, over 5695372.87 frames. ], batch size: 99, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:49:39,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1132210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:50:01,780 INFO [train.py:968] (0/2) Epoch 25, batch 38400, giga_loss[loss=0.2643, simple_loss=0.3457, pruned_loss=0.09151, over 29014.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3563, pruned_loss=0.1031, over 5713711.76 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1129, over 5732169.02 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3553, pruned_loss=0.1021, over 5705356.51 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:50:32,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.244e+02 1.264e+03 1.556e+03 2.161e+03 7.468e+03, threshold=3.113e+03, percent-clipped=10.0
+2023-03-13 04:50:39,095 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1132285.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 04:50:42,172 INFO [train.py:968] (0/2) Epoch 25, batch 38450, giga_loss[loss=0.2746, simple_loss=0.3513, pruned_loss=0.09902, over 28950.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3544, pruned_loss=0.1028, over 5707241.01 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3595, pruned_loss=0.1128, over 5728621.86 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3534, pruned_loss=0.1015, over 5702290.69 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:51:18,419 INFO [train.py:968] (0/2) Epoch 25, batch 38500, giga_loss[loss=0.2446, simple_loss=0.3325, pruned_loss=0.07835, over 28970.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.353, pruned_loss=0.1024, over 5697963.31 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3601, pruned_loss=0.1133, over 5721000.85 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3515, pruned_loss=0.1005, over 5700344.55 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:51:27,327 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1132353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:51:29,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1132356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:51:47,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.202e+02 1.334e+03 1.684e+03 2.718e+03 2.249e+04, threshold=3.367e+03, percent-clipped=18.0
+2023-03-13 04:51:52,538 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1132385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:51:55,610 INFO [train.py:968] (0/2) Epoch 25, batch 38550, libri_loss[loss=0.2921, simple_loss=0.3559, pruned_loss=0.1141, over 25898.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3511, pruned_loss=0.1016, over 5700143.81 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3599, pruned_loss=0.1133, over 5720818.23 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3498, pruned_loss=0.09969, over 5701672.49 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:52:07,699 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2442, 1.6832, 1.7100, 1.4777], device='cuda:0'), covar=tensor([0.1829, 0.1269, 0.1939, 0.1490], device='cuda:0'), in_proj_covar=tensor([0.0490, 0.0752, 0.0726, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 04:52:24,392 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1132428.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 04:52:28,489 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1132431.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:52:33,006 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1132436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:52:35,520 INFO [train.py:968] (0/2) Epoch 25, batch 38600, giga_loss[loss=0.3364, simple_loss=0.376, pruned_loss=0.1484, over 23582.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3512, pruned_loss=0.1018, over 5704622.60 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3598, pruned_loss=0.1132, over 5727377.55 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3499, pruned_loss=0.1, over 5699505.41 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:52:51,138 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1132460.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:53:06,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.542e+02 1.174e+03 1.415e+03 1.936e+03 5.985e+03, threshold=2.831e+03, percent-clipped=3.0
+2023-03-13 04:53:15,862 INFO [train.py:968] (0/2) Epoch 25, batch 38650, giga_loss[loss=0.274, simple_loss=0.3473, pruned_loss=0.1004, over 28392.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3509, pruned_loss=0.1014, over 5696058.43 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3598, pruned_loss=0.1132, over 5716708.63 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3497, pruned_loss=0.09977, over 5700596.85 frames. ], batch size: 65, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:53:53,690 INFO [train.py:968] (0/2) Epoch 25, batch 38700, giga_loss[loss=0.2841, simple_loss=0.3575, pruned_loss=0.1054, over 28492.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3502, pruned_loss=0.1003, over 5702338.58 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3597, pruned_loss=0.113, over 5718512.42 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3492, pruned_loss=0.09905, over 5704036.89 frames. ], batch size: 71, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:53:59,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1132548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:54:10,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6030, 2.4432, 2.3264, 2.1356], device='cuda:0'), covar=tensor([0.1903, 0.2331, 0.2089, 0.2323], device='cuda:0'), in_proj_covar=tensor([0.0490, 0.0752, 0.0724, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 04:54:23,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.022e+02 1.060e+03 1.236e+03 1.737e+03 5.080e+03, threshold=2.472e+03, percent-clipped=6.0
+2023-03-13 04:54:31,179 INFO [train.py:968] (0/2) Epoch 25, batch 38750, giga_loss[loss=0.2523, simple_loss=0.3347, pruned_loss=0.08494, over 28941.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3499, pruned_loss=0.09967, over 5711133.64 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3601, pruned_loss=0.1134, over 5722113.17 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3484, pruned_loss=0.09788, over 5708730.90 frames. ], batch size: 164, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:55:09,825 INFO [train.py:968] (0/2) Epoch 25, batch 38800, giga_loss[loss=0.2906, simple_loss=0.3604, pruned_loss=0.1104, over 28871.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3491, pruned_loss=0.09881, over 5717742.42 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3599, pruned_loss=0.1131, over 5726860.99 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3478, pruned_loss=0.09719, over 5711376.41 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:55:29,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5213, 4.9494, 1.7089, 1.9433], device='cuda:0'), covar=tensor([0.0975, 0.0370, 0.0874, 0.1214], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0555, 0.0396, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 04:55:41,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.468e+02 1.082e+03 1.378e+03 1.831e+03 9.327e+03, threshold=2.756e+03, percent-clipped=9.0
+2023-03-13 04:55:51,465 INFO [train.py:968] (0/2) Epoch 25, batch 38850, giga_loss[loss=0.2702, simple_loss=0.341, pruned_loss=0.09967, over 28783.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.348, pruned_loss=0.09873, over 5704201.73 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.1131, over 5725263.87 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3464, pruned_loss=0.09703, over 5700257.80 frames. ], batch size: 66, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:55:52,472 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1132691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:55:54,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1132694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:56:15,872 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1132723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:56:29,217 INFO [train.py:968] (0/2) Epoch 25, batch 38900, giga_loss[loss=0.247, simple_loss=0.327, pruned_loss=0.08352, over 29038.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3449, pruned_loss=0.0974, over 5695123.75 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1128, over 5713681.12 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3438, pruned_loss=0.09601, over 5701021.41 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:57:00,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.966e+02 1.178e+03 1.483e+03 2.322e+03 9.512e+03, threshold=2.966e+03, percent-clipped=16.0
+2023-03-13 04:57:11,465 INFO [train.py:968] (0/2) Epoch 25, batch 38950, giga_loss[loss=0.2514, simple_loss=0.3301, pruned_loss=0.08636, over 28698.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3412, pruned_loss=0.09532, over 5700543.86 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3598, pruned_loss=0.1129, over 5715769.35 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.34, pruned_loss=0.09399, over 5703179.33 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:57:26,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1132811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:57:32,694 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 04:57:49,962 INFO [train.py:968] (0/2) Epoch 25, batch 39000, giga_loss[loss=0.3497, simple_loss=0.3972, pruned_loss=0.151, over 26588.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3414, pruned_loss=0.09616, over 5700161.48 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1128, over 5717907.45 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3404, pruned_loss=0.09494, over 5700057.46 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:57:49,966 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 04:57:58,920 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2055, simple_loss=0.3136, pruned_loss=0.0487, over 944034.00 frames. 
+2023-03-13 04:57:58,921 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 04:58:16,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3462, 1.2891, 1.2096, 1.4886], device='cuda:0'), covar=tensor([0.0725, 0.0356, 0.0351, 0.0831], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 04:58:24,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1132875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:58:27,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.224e+02 1.258e+03 1.569e+03 1.959e+03 6.459e+03, threshold=3.137e+03, percent-clipped=7.0
+2023-03-13 04:58:27,352 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6151, 1.7584, 1.7447, 1.5421], device='cuda:0'), covar=tensor([0.3077, 0.2605, 0.2065, 0.2594], device='cuda:0'), in_proj_covar=tensor([0.2007, 0.1944, 0.1856, 0.2023], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 04:58:36,680 INFO [train.py:968] (0/2) Epoch 25, batch 39050, giga_loss[loss=0.2426, simple_loss=0.3155, pruned_loss=0.08482, over 28847.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3408, pruned_loss=0.09649, over 5707978.49 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3589, pruned_loss=0.1125, over 5720998.00 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.34, pruned_loss=0.09524, over 5704759.22 frames. ], batch size: 112, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:58:39,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1132894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:58:47,093 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1132902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:59:15,250 INFO [train.py:968] (0/2) Epoch 25, batch 39100, giga_loss[loss=0.2485, simple_loss=0.3252, pruned_loss=0.08596, over 28691.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3384, pruned_loss=0.09544, over 5692598.52 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3592, pruned_loss=0.1127, over 5703469.17 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3372, pruned_loss=0.09404, over 5706780.34 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:59:28,083 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1132954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:59:30,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1132957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:59:39,821 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4739, 1.4808, 3.7294, 3.1946], device='cuda:0'), covar=tensor([0.1483, 0.2590, 0.0439, 0.1083], device='cuda:0'), in_proj_covar=tensor([0.0774, 0.0653, 0.0966, 0.0935], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 04:59:44,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.979e+02 1.171e+03 1.514e+03 1.861e+03 8.349e+03, threshold=3.027e+03, percent-clipped=4.0
+2023-03-13 04:59:51,981 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1132986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:59:54,504 INFO [train.py:968] (0/2) Epoch 25, batch 39150, giga_loss[loss=0.253, simple_loss=0.3187, pruned_loss=0.09358, over 28614.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3356, pruned_loss=0.09441, over 5693980.48 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3588, pruned_loss=0.1124, over 5706472.55 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3348, pruned_loss=0.09332, over 5702399.77 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:00:05,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4120, 3.3792, 1.5320, 1.6305], device='cuda:0'), covar=tensor([0.0972, 0.0414, 0.0920, 0.1303], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0555, 0.0396, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-13 05:00:14,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6801, 1.6485, 1.8760, 1.4551], device='cuda:0'), covar=tensor([0.2068, 0.2429, 0.1576, 0.1812], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0712, 0.0972, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 05:00:34,177 INFO [train.py:968] (0/2) Epoch 25, batch 39200, giga_loss[loss=0.2857, simple_loss=0.3626, pruned_loss=0.1044, over 28755.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3331, pruned_loss=0.09295, over 5696219.81 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3591, pruned_loss=0.1125, over 5700735.86 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3317, pruned_loss=0.09155, over 5708411.16 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:01:05,758 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.033e+02 1.146e+03 1.487e+03 2.382e+03 6.360e+03, threshold=2.973e+03, percent-clipped=13.0
+2023-03-13 05:01:14,592 INFO [train.py:968] (0/2) Epoch 25, batch 39250, giga_loss[loss=0.2439, simple_loss=0.3247, pruned_loss=0.08153, over 29054.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.333, pruned_loss=0.09311, over 5701941.69 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1125, over 5707442.51 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.331, pruned_loss=0.09133, over 5705561.79 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:01:14,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3892, 3.5747, 1.4940, 1.5842], device='cuda:0'), covar=tensor([0.0997, 0.0372, 0.0953, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0555, 0.0396, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:0')
+2023-03-13 05:01:52,784 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133137.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 05:01:54,598 INFO [train.py:968] (0/2) Epoch 25, batch 39300, giga_loss[loss=0.2467, simple_loss=0.3263, pruned_loss=0.08355, over 28963.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.334, pruned_loss=0.09314, over 5708075.02 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3581, pruned_loss=0.112, over 5713326.82 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3322, pruned_loss=0.09153, over 5705768.52 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:02:27,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.048e+02 1.202e+03 1.483e+03 2.100e+03 6.185e+03, threshold=2.966e+03, percent-clipped=6.0
+2023-03-13 05:02:37,660 INFO [train.py:968] (0/2) Epoch 25, batch 39350, giga_loss[loss=0.2659, simple_loss=0.3508, pruned_loss=0.09047, over 28655.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.336, pruned_loss=0.09324, over 5714688.73 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3579, pruned_loss=0.1119, over 5719376.64 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3342, pruned_loss=0.09149, over 5707408.50 frames. ], batch size: 242, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:03:16,875 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0864, 2.3240, 1.9342, 1.9786], device='cuda:0'), covar=tensor([0.2182, 0.1948, 0.2078, 0.1963], device='cuda:0'), in_proj_covar=tensor([0.0492, 0.0754, 0.0726, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 05:03:17,939 INFO [train.py:968] (0/2) Epoch 25, batch 39400, giga_loss[loss=0.2457, simple_loss=0.335, pruned_loss=0.07823, over 28680.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3391, pruned_loss=0.09481, over 5692127.12 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3582, pruned_loss=0.1122, over 5702876.69 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3371, pruned_loss=0.09279, over 5701013.73 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:03:22,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6264, 1.8975, 1.5129, 1.8647], device='cuda:0'), covar=tensor([0.2837, 0.2926, 0.3299, 0.2731], device='cuda:0'), in_proj_covar=tensor([0.1561, 0.1124, 0.1379, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 05:03:25,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1133250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:03:42,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1133269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:03:49,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1133277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:03:50,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.918e+02 1.245e+03 1.531e+03 1.962e+03 4.266e+03, threshold=3.061e+03, percent-clipped=7.0
+2023-03-13 05:03:59,512 INFO [train.py:968] (0/2) Epoch 25, batch 39450, libri_loss[loss=0.3169, simple_loss=0.3746, pruned_loss=0.1295, over 29540.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3404, pruned_loss=0.09444, over 5696084.30 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3589, pruned_loss=0.1126, over 5707270.72 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3376, pruned_loss=0.09197, over 5698809.82 frames. ], batch size: 79, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:04:39,761 INFO [train.py:968] (0/2) Epoch 25, batch 39500, giga_loss[loss=0.3008, simple_loss=0.3743, pruned_loss=0.1136, over 28332.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.0947, over 5689889.30 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3586, pruned_loss=0.1124, over 5712084.29 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3383, pruned_loss=0.09217, over 5686801.61 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:05:08,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.480e+02 1.298e+03 1.710e+03 2.734e+03 5.763e+03, threshold=3.420e+03, percent-clipped=19.0
+2023-03-13 05:05:08,947 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:16,776 INFO [train.py:968] (0/2) Epoch 25, batch 39550, giga_loss[loss=0.2889, simple_loss=0.3551, pruned_loss=0.1114, over 24244.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.341, pruned_loss=0.09438, over 5701712.85 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3583, pruned_loss=0.1122, over 5716281.56 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3382, pruned_loss=0.09175, over 5694939.03 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:05:19,200 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1133393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:21,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1133396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:29,320 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 05:05:34,371 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1133412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:36,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1133415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:39,595 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1133420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:41,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1133423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:43,895 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1133425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:53,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7942, 2.0120, 1.7039, 1.9925], device='cuda:0'), covar=tensor([0.2629, 0.2748, 0.3103, 0.2547], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1126, 0.1380, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 05:05:56,392 INFO [train.py:968] (0/2) Epoch 25, batch 39600, giga_loss[loss=0.2788, simple_loss=0.3536, pruned_loss=0.102, over 27673.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3431, pruned_loss=0.09652, over 5701787.27 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3586, pruned_loss=0.1124, over 5721810.64 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3402, pruned_loss=0.09378, over 5691144.87 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:05:59,382 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1133444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:06:04,434 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1133452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:06:28,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.010e+02 1.354e+03 1.777e+03 2.330e+03 5.312e+03, threshold=3.554e+03, percent-clipped=5.0
+2023-03-13 05:06:36,754 INFO [train.py:968] (0/2) Epoch 25, batch 39650, giga_loss[loss=0.2733, simple_loss=0.3474, pruned_loss=0.09956, over 28485.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3437, pruned_loss=0.09673, over 5695805.30 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3584, pruned_loss=0.1122, over 5720644.34 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3413, pruned_loss=0.09451, over 5688041.30 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:06:53,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3984, 1.6078, 1.6736, 1.3931], device='cuda:0'), covar=tensor([0.3286, 0.2566, 0.1832, 0.2650], device='cuda:0'), in_proj_covar=tensor([0.2030, 0.1960, 0.1880, 0.2040], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 05:06:54,764 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1133512.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:07:17,475 INFO [train.py:968] (0/2) Epoch 25, batch 39700, giga_loss[loss=0.2982, simple_loss=0.3627, pruned_loss=0.1169, over 28786.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3481, pruned_loss=0.09976, over 5705946.26 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3588, pruned_loss=0.1123, over 5723381.51 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3454, pruned_loss=0.09726, over 5696494.33 frames. ], batch size: 99, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:07:46,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.985e+02 1.477e+03 1.876e+03 2.691e+03 7.923e+03, threshold=3.752e+03, percent-clipped=8.0
+2023-03-13 05:07:55,614 INFO [train.py:968] (0/2) Epoch 25, batch 39750, giga_loss[loss=0.3002, simple_loss=0.3687, pruned_loss=0.1159, over 28942.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3499, pruned_loss=0.1003, over 5704116.54 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1124, over 5716604.79 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3474, pruned_loss=0.09799, over 5701815.75 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:08:15,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6997, 1.9899, 1.6012, 1.6938], device='cuda:0'), covar=tensor([0.2563, 0.2652, 0.3109, 0.2547], device='cuda:0'), in_proj_covar=tensor([0.1564, 0.1127, 0.1382, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 05:08:36,325 INFO [train.py:968] (0/2) Epoch 25, batch 39800, giga_loss[loss=0.2948, simple_loss=0.3807, pruned_loss=0.1045, over 28970.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3504, pruned_loss=0.09978, over 5710958.03 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3588, pruned_loss=0.1124, over 5717566.16 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3484, pruned_loss=0.09787, over 5707985.97 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:08:48,184 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1133655.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:08:51,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1133658.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:09:10,069 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.282e+02 1.371e+03 1.704e+03 2.481e+03 6.247e+03, threshold=3.408e+03, percent-clipped=7.0
+2023-03-13 05:09:15,482 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1133687.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 05:09:17,031 INFO [train.py:968] (0/2) Epoch 25, batch 39850, libri_loss[loss=0.2944, simple_loss=0.37, pruned_loss=0.1094, over 29396.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3517, pruned_loss=0.1008, over 5717883.35 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3588, pruned_loss=0.1122, over 5721906.38 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.35, pruned_loss=0.09918, over 5711435.64 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:09:17,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2495, 1.2423, 3.8164, 3.2079], device='cuda:0'), covar=tensor([0.1712, 0.2819, 0.0457, 0.0949], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0655, 0.0970, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 05:09:56,743 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:09:57,224 INFO [train.py:968] (0/2) Epoch 25, batch 39900, giga_loss[loss=0.2925, simple_loss=0.3541, pruned_loss=0.1154, over 28645.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3513, pruned_loss=0.1006, over 5718904.98 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.359, pruned_loss=0.1121, over 5723645.05 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3495, pruned_loss=0.09893, over 5711781.11 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:10:09,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1133754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:10:27,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.987e+02 1.315e+03 1.618e+03 2.515e+03 9.029e+03, threshold=3.235e+03, percent-clipped=12.0
+2023-03-13 05:10:35,167 INFO [train.py:968] (0/2) Epoch 25, batch 39950, giga_loss[loss=0.3088, simple_loss=0.3752, pruned_loss=0.1212, over 27657.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3502, pruned_loss=0.09989, over 5717355.04 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3594, pruned_loss=0.1124, over 5725311.45 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3483, pruned_loss=0.09825, over 5710110.01 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:10:35,991 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:10:40,322 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6372, 1.7244, 1.8314, 1.3607], device='cuda:0'), covar=tensor([0.1750, 0.2582, 0.1506, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0710, 0.0969, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 05:11:03,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-13 05:11:12,754 INFO [train.py:968] (0/2) Epoch 25, batch 40000, giga_loss[loss=0.311, simple_loss=0.3624, pruned_loss=0.1298, over 23927.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3479, pruned_loss=0.09917, over 5710279.54 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3594, pruned_loss=0.1123, over 5722130.68 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3461, pruned_loss=0.09745, over 5706405.08 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:11:45,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.795e+02 1.188e+03 1.376e+03 2.044e+03 7.213e+03, threshold=2.751e+03, percent-clipped=8.0
+2023-03-13 05:11:47,406 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6273, 1.6580, 1.8602, 1.4126], device='cuda:0'), covar=tensor([0.2071, 0.2720, 0.1729, 0.1979], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0710, 0.0969, 0.0868], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 05:11:53,320 INFO [train.py:968] (0/2) Epoch 25, batch 40050, giga_loss[loss=0.2652, simple_loss=0.3402, pruned_loss=0.09509, over 28911.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3442, pruned_loss=0.09702, over 5713872.19 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3594, pruned_loss=0.1124, over 5717890.07 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3422, pruned_loss=0.09509, over 5714887.66 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:12:00,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1133897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:12:02,155 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1133900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:12:10,287 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4315, 4.4957, 1.6041, 1.6037], device='cuda:0'), covar=tensor([0.1015, 0.0260, 0.0981, 0.1344], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0555, 0.0397, 0.0434], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 05:12:20,402 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:12:23,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1133929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:12:30,806 INFO [train.py:968] (0/2) Epoch 25, batch 40100, giga_loss[loss=0.2469, simple_loss=0.3218, pruned_loss=0.08598, over 28523.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3433, pruned_loss=0.09683, over 5716296.13 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3594, pruned_loss=0.1124, over 5725055.28 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3411, pruned_loss=0.09467, over 5710185.28 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:12:32,234 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:12:45,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9893, 1.3007, 1.1249, 0.2475], device='cuda:0'), covar=tensor([0.4511, 0.3473, 0.5159, 0.7189], device='cuda:0'), in_proj_covar=tensor([0.1796, 0.1682, 0.1625, 0.1459], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 05:13:02,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.147e+02 1.194e+03 1.503e+03 2.025e+03 5.609e+03, threshold=3.007e+03, percent-clipped=14.0
+2023-03-13 05:13:05,164 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.88 vs. limit=2.0
+2023-03-13 05:13:11,708 INFO [train.py:968] (0/2) Epoch 25, batch 40150, giga_loss[loss=0.3142, simple_loss=0.3874, pruned_loss=0.1205, over 28596.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3449, pruned_loss=0.09632, over 5716010.62 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3594, pruned_loss=0.1124, over 5724822.85 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3429, pruned_loss=0.09428, over 5711266.73 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:13:20,556 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:13:21,840 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1134000.pt
+2023-03-13 05:13:53,232 INFO [train.py:968] (0/2) Epoch 25, batch 40200, giga_loss[loss=0.2842, simple_loss=0.3399, pruned_loss=0.1143, over 28626.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3473, pruned_loss=0.09753, over 5702176.13 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3597, pruned_loss=0.1125, over 5717991.68 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3452, pruned_loss=0.09558, over 5703720.44 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:14:26,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.920e+02 1.314e+03 1.651e+03 2.386e+03 7.824e+03, threshold=3.302e+03, percent-clipped=12.0
+2023-03-13 05:14:34,037 INFO [train.py:968] (0/2) Epoch 25, batch 40250, libri_loss[loss=0.2977, simple_loss=0.3723, pruned_loss=0.1116, over 29555.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3455, pruned_loss=0.09714, over 5714021.83 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3601, pruned_loss=0.1127, over 5722994.54 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3431, pruned_loss=0.09508, over 5710459.96 frames. ], batch size: 83, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:14:36,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6343, 5.4916, 5.1868, 2.9030], device='cuda:0'), covar=tensor([0.0475, 0.0627, 0.0674, 0.1614], device='cuda:0'), in_proj_covar=tensor([0.1257, 0.1159, 0.0977, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 05:14:53,720 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:15:13,191 INFO [train.py:968] (0/2) Epoch 25, batch 40300, giga_loss[loss=0.236, simple_loss=0.3123, pruned_loss=0.07987, over 28795.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.344, pruned_loss=0.09733, over 5711230.01 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3605, pruned_loss=0.1128, over 5717042.17 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3413, pruned_loss=0.0951, over 5713707.98 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:15:17,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7362, 2.0879, 1.6657, 1.8597], device='cuda:0'), covar=tensor([0.2530, 0.2633, 0.3046, 0.2503], device='cuda:0'), in_proj_covar=tensor([0.1561, 0.1126, 0.1380, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 05:15:34,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:15:43,735 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6069, 2.0075, 1.5191, 1.8322], device='cuda:0'), covar=tensor([0.2580, 0.2551, 0.2986, 0.2221], device='cuda:0'), in_proj_covar=tensor([0.1560, 0.1126, 0.1380, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 05:15:46,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.311e+02 1.342e+03 1.593e+03 2.158e+03 5.410e+03, threshold=3.187e+03, percent-clipped=8.0
+2023-03-13 05:15:53,603 INFO [train.py:968] (0/2) Epoch 25, batch 40350, giga_loss[loss=0.2813, simple_loss=0.3495, pruned_loss=0.1066, over 29113.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3427, pruned_loss=0.09814, over 5703866.64 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3606, pruned_loss=0.113, over 5712082.50 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3399, pruned_loss=0.0957, over 5710241.58 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:16:33,903 INFO [train.py:968] (0/2) Epoch 25, batch 40400, giga_loss[loss=0.2842, simple_loss=0.3432, pruned_loss=0.1126, over 28779.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3407, pruned_loss=0.09821, over 5691560.67 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3606, pruned_loss=0.113, over 5704433.01 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3382, pruned_loss=0.09597, over 5703804.78 frames. ], batch size: 99, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:16:47,676 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:16:49,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:00,177 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4615, 3.4210, 1.4794, 1.6238], device='cuda:0'), covar=tensor([0.0956, 0.0365, 0.0983, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0556, 0.0397, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 05:17:07,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.058e+02 1.285e+03 1.613e+03 2.375e+03 5.821e+03, threshold=3.226e+03, percent-clipped=7.0
+2023-03-13 05:17:14,739 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:15,083 INFO [train.py:968] (0/2) Epoch 25, batch 40450, giga_loss[loss=0.3159, simple_loss=0.3701, pruned_loss=0.1308, over 23917.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3395, pruned_loss=0.09757, over 5694006.35 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3602, pruned_loss=0.1127, over 5708245.84 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3374, pruned_loss=0.09577, over 5700266.43 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:17:22,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:24,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1134302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:29,317 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:31,275 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:37,423 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:53,386 INFO [train.py:968] (0/2) Epoch 25, batch 40500, libri_loss[loss=0.2621, simple_loss=0.3377, pruned_loss=0.09323, over 29569.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3357, pruned_loss=0.09543, over 5686427.33 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3595, pruned_loss=0.1125, over 5691276.97 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.334, pruned_loss=0.09375, over 5705892.47 frames. ], batch size: 76, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:17:54,293 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:18:18,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:18:24,328 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1134381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:18:24,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.931e+02 1.289e+03 1.750e+03 2.553e+03 8.678e+03, threshold=3.500e+03, percent-clipped=11.0
+2023-03-13 05:18:32,407 INFO [train.py:968] (0/2) Epoch 25, batch 40550, giga_loss[loss=0.262, simple_loss=0.3223, pruned_loss=0.1009, over 29038.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3318, pruned_loss=0.09383, over 5694105.77 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.36, pruned_loss=0.1129, over 5695563.48 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3295, pruned_loss=0.09174, over 5705824.54 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:18:53,266 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9509, 1.0953, 2.8612, 2.7628], device='cuda:0'), covar=tensor([0.1606, 0.2676, 0.0562, 0.1127], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0658, 0.0970, 0.0942], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 05:19:13,318 INFO [train.py:968] (0/2) Epoch 25, batch 40600, giga_loss[loss=0.2224, simple_loss=0.3138, pruned_loss=0.06548, over 29050.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3302, pruned_loss=0.09269, over 5703043.87 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3599, pruned_loss=0.1128, over 5697656.37 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3281, pruned_loss=0.09098, over 5710357.09 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:19:16,066 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:19:18,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:19:29,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:19:31,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:19:41,236 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:19:45,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.467e+02 1.270e+03 1.564e+03 2.155e+03 5.178e+03, threshold=3.127e+03, percent-clipped=5.0
+2023-03-13 05:19:53,346 INFO [train.py:968] (0/2) Epoch 25, batch 40650, giga_loss[loss=0.2955, simple_loss=0.3655, pruned_loss=0.1128, over 28514.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3315, pruned_loss=0.09304, over 5709561.78 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5704160.51 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3295, pruned_loss=0.09131, over 5709834.66 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:19:54,797 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:20:15,690 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:20:18,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:20:35,114 INFO [train.py:968] (0/2) Epoch 25, batch 40700, giga_loss[loss=0.3815, simple_loss=0.426, pruned_loss=0.1686, over 26768.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3352, pruned_loss=0.09429, over 5700795.13 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5697910.64 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.333, pruned_loss=0.09241, over 5707489.21 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:20:42,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:21:05,944 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0771, 2.3192, 2.2796, 2.0397], device='cuda:0'), covar=tensor([0.2099, 0.1714, 0.1635, 0.1831], device='cuda:0'), in_proj_covar=tensor([0.2029, 0.1959, 0.1881, 0.2034], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 05:21:07,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.338e+02 1.229e+03 1.549e+03 2.136e+03 6.353e+03, threshold=3.099e+03, percent-clipped=10.0
+2023-03-13 05:21:13,680 INFO [train.py:968] (0/2) Epoch 25, batch 40750, giga_loss[loss=0.2813, simple_loss=0.346, pruned_loss=0.1082, over 28554.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3387, pruned_loss=0.09605, over 5688464.62 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1129, over 5685074.62 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3361, pruned_loss=0.09378, over 5706171.39 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:21:21,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2563, 1.2958, 3.9337, 3.2742], device='cuda:0'), covar=tensor([0.1652, 0.2655, 0.0484, 0.1040], device='cuda:0'), in_proj_covar=tensor([0.0778, 0.0659, 0.0972, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 05:21:43,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3282, 1.2244, 3.7735, 3.2094], device='cuda:0'), covar=tensor([0.1604, 0.2893, 0.0433, 0.0996], device='cuda:0'), in_proj_covar=tensor([0.0777, 0.0658, 0.0971, 0.0943], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 05:21:51,028 INFO [train.py:968] (0/2) Epoch 25, batch 40800, giga_loss[loss=0.2776, simple_loss=0.359, pruned_loss=0.09806, over 28879.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3428, pruned_loss=0.09788, over 5692164.60 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5692637.41 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3398, pruned_loss=0.09537, over 5699665.89 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:22:20,879 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:22:25,025 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.130e+02 1.263e+03 1.722e+03 2.257e+03 4.999e+03, threshold=3.445e+03, percent-clipped=7.0
+2023-03-13 05:22:29,035 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-13 05:22:33,178 INFO [train.py:968] (0/2) Epoch 25, batch 40850, giga_loss[loss=0.2703, simple_loss=0.3419, pruned_loss=0.09935, over 28692.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3453, pruned_loss=0.09906, over 5698258.67 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5694433.56 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3424, pruned_loss=0.09666, over 5702922.86 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:22:55,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3449, 4.1765, 3.9752, 1.9141], device='cuda:0'), covar=tensor([0.0595, 0.0719, 0.0726, 0.1978], device='cuda:0'), in_proj_covar=tensor([0.1259, 0.1161, 0.0976, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 05:23:09,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2832, 2.9400, 1.4671, 1.4372], device='cuda:0'), covar=tensor([0.1006, 0.0403, 0.0948, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0557, 0.0397, 0.0435], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 05:23:18,186 INFO [train.py:968] (0/2) Epoch 25, batch 40900, giga_loss[loss=0.5399, simple_loss=0.5312, pruned_loss=0.2743, over 26528.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3508, pruned_loss=0.1036, over 5696240.53 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3603, pruned_loss=0.1131, over 5695864.12 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.348, pruned_loss=0.1014, over 5698622.73 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:23:37,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:23:57,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.640e+02 1.673e+03 2.172e+03 3.094e+03 6.636e+03, threshold=4.344e+03, percent-clipped=17.0
+2023-03-13 05:24:04,712 INFO [train.py:968] (0/2) Epoch 25, batch 40950, giga_loss[loss=0.312, simple_loss=0.3746, pruned_loss=0.1248, over 28961.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3557, pruned_loss=0.1074, over 5684668.15 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3608, pruned_loss=0.1134, over 5679633.14 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3528, pruned_loss=0.1051, over 5701371.22 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:24:28,924 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:24:32,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:24:46,677 INFO [train.py:968] (0/2) Epoch 25, batch 41000, giga_loss[loss=0.3788, simple_loss=0.427, pruned_loss=0.1653, over 28027.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.361, pruned_loss=0.1113, over 5679517.30 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.1129, over 5677072.16 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.359, pruned_loss=0.1096, over 5695544.31 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:24:57,113 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:25:22,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.690e+03 2.335e+03 3.484e+03 7.635e+03, threshold=4.670e+03, percent-clipped=10.0
+2023-03-13 05:25:27,507 INFO [train.py:968] (0/2) Epoch 25, batch 41050, giga_loss[loss=0.3895, simple_loss=0.4338, pruned_loss=0.1726, over 29024.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3657, pruned_loss=0.1149, over 5667779.95 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5667463.91 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3644, pruned_loss=0.1135, over 5689835.27 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:25:36,922 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:25:38,874 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:26:03,667 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:26:10,621 INFO [train.py:968] (0/2) Epoch 25, batch 41100, giga_loss[loss=0.3553, simple_loss=0.4058, pruned_loss=0.1524, over 28766.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3723, pruned_loss=0.1205, over 5677932.49 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1128, over 5671665.28 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3717, pruned_loss=0.1196, over 5691857.00 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:26:36,928 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-13 05:26:48,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.295e+03 1.864e+03 2.378e+03 3.035e+03 5.871e+03, threshold=4.756e+03, percent-clipped=4.0
+2023-03-13 05:26:54,766 INFO [train.py:968] (0/2) Epoch 25, batch 41150, giga_loss[loss=0.3433, simple_loss=0.405, pruned_loss=0.1409, over 28267.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3776, pruned_loss=0.1248, over 5681053.28 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5676659.14 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3776, pruned_loss=0.1245, over 5688017.73 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:27:44,631 INFO [train.py:968] (0/2) Epoch 25, batch 41200, giga_loss[loss=0.3215, simple_loss=0.3806, pruned_loss=0.1312, over 28973.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3809, pruned_loss=0.1288, over 5654032.08 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1128, over 5673439.90 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3813, pruned_loss=0.1289, over 5662202.69 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:27:47,555 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1135042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:28:33,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.182e+03 1.768e+03 2.465e+03 3.655e+03 1.243e+04, threshold=4.931e+03, percent-clipped=10.0
+2023-03-13 05:28:39,249 INFO [train.py:968] (0/2) Epoch 25, batch 41250, giga_loss[loss=0.2989, simple_loss=0.3663, pruned_loss=0.1158, over 28835.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3814, pruned_loss=0.1301, over 5659889.09 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5677042.18 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3822, pruned_loss=0.1305, over 5663045.58 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:28:48,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6194, 1.7392, 1.2744, 1.3829], device='cuda:0'), covar=tensor([0.0983, 0.0645, 0.0999, 0.1325], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0452, 0.0523, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 05:29:27,781 INFO [train.py:968] (0/2) Epoch 25, batch 41300, giga_loss[loss=0.3425, simple_loss=0.4028, pruned_loss=0.141, over 28862.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.386, pruned_loss=0.1353, over 5645437.02 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1128, over 5683338.02 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.387, pruned_loss=0.136, over 5641523.47 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:30:13,604 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.914e+03 2.410e+03 3.094e+03 8.611e+03, threshold=4.821e+03, percent-clipped=7.0
+2023-03-13 05:30:15,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4637, 2.8723, 1.5306, 1.6353], device='cuda:0'), covar=tensor([0.0815, 0.0336, 0.0749, 0.1104], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0562, 0.0400, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 05:30:22,914 INFO [train.py:968] (0/2) Epoch 25, batch 41350, libri_loss[loss=0.2583, simple_loss=0.3284, pruned_loss=0.09405, over 29557.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.389, pruned_loss=0.1381, over 5623428.99 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5675561.45 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3902, pruned_loss=0.139, over 5626469.95 frames. ], batch size: 77, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:31:09,758 INFO [train.py:968] (0/2) Epoch 25, batch 41400, giga_loss[loss=0.3864, simple_loss=0.4137, pruned_loss=0.1795, over 23342.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3889, pruned_loss=0.1384, over 5612395.92 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5669362.70 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3909, pruned_loss=0.1399, over 5618866.64 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 05:31:15,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4574, 1.9926, 1.4708, 0.7379], device='cuda:0'), covar=tensor([0.5587, 0.2832, 0.3552, 0.6312], device='cuda:0'), in_proj_covar=tensor([0.1811, 0.1702, 0.1638, 0.1480], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 05:31:53,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.202e+03 2.172e+03 2.929e+03 3.867e+03 9.561e+03, threshold=5.858e+03, percent-clipped=15.0
+2023-03-13 05:31:58,680 INFO [train.py:968] (0/2) Epoch 25, batch 41450, giga_loss[loss=0.3872, simple_loss=0.4357, pruned_loss=0.1694, over 28716.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3885, pruned_loss=0.1387, over 5622839.56 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1127, over 5673755.27 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3905, pruned_loss=0.1404, over 5623167.79 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 05:32:49,703 INFO [train.py:968] (0/2) Epoch 25, batch 41500, giga_loss[loss=0.2959, simple_loss=0.3715, pruned_loss=0.1101, over 28997.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3857, pruned_loss=0.136, over 5641268.10 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5675984.75 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3877, pruned_loss=0.1376, over 5639156.51 frames. ], batch size: 164, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 05:33:14,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4011, 1.2782, 1.3312, 1.5423], device='cuda:0'), covar=tensor([0.0711, 0.0380, 0.0303, 0.0780], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 05:33:36,514 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.726e+03 2.300e+03 2.893e+03 9.502e+03, threshold=4.600e+03, percent-clipped=2.0
+2023-03-13 05:33:39,823 INFO [train.py:968] (0/2) Epoch 25, batch 41550, giga_loss[loss=0.3104, simple_loss=0.3824, pruned_loss=0.1192, over 28892.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3858, pruned_loss=0.1348, over 5647646.37 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5679070.66 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.388, pruned_loss=0.1366, over 5642763.23 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 05:34:02,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2962, 1.3482, 4.2094, 3.4071], device='cuda:0'), covar=tensor([0.1789, 0.2868, 0.0452, 0.1196], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0663, 0.0979, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 05:34:07,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1135417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:34:20,052 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 05:34:25,339 INFO [train.py:968] (0/2) Epoch 25, batch 41600, giga_loss[loss=0.3404, simple_loss=0.3993, pruned_loss=0.1407, over 28644.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3865, pruned_loss=0.1344, over 5669155.93 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3595, pruned_loss=0.1125, over 5689421.32 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3897, pruned_loss=0.1371, over 5654266.60 frames. ], batch size: 242, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:34:32,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5515, 4.3718, 4.1724, 2.2618], device='cuda:0'), covar=tensor([0.0538, 0.0679, 0.0664, 0.2017], device='cuda:0'), in_proj_covar=tensor([0.1282, 0.1181, 0.0996, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 05:35:07,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.275e+03 1.869e+03 2.299e+03 3.708e+03 9.222e+03, threshold=4.597e+03, percent-clipped=13.0
+2023-03-13 05:35:12,769 INFO [train.py:968] (0/2) Epoch 25, batch 41650, giga_loss[loss=0.2869, simple_loss=0.3619, pruned_loss=0.1059, over 28937.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3868, pruned_loss=0.1346, over 5651378.71 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3594, pruned_loss=0.1124, over 5691225.87 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3905, pruned_loss=0.1377, over 5637088.97 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:36:01,543 INFO [train.py:968] (0/2) Epoch 25, batch 41700, giga_loss[loss=0.2807, simple_loss=0.3641, pruned_loss=0.09869, over 28789.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3838, pruned_loss=0.1315, over 5655296.51 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5697719.26 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3876, pruned_loss=0.1345, over 5636984.96 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:36:13,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5409, 1.6654, 1.2925, 1.3440], device='cuda:0'), covar=tensor([0.0880, 0.0515, 0.0872, 0.1190], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0452, 0.0522, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 05:36:20,862 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1135560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:36:24,415 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1135563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:36:45,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.188e+03 1.815e+03 2.352e+03 2.898e+03 8.896e+03, threshold=4.704e+03, percent-clipped=3.0
+2023-03-13 05:36:50,596 INFO [train.py:968] (0/2) Epoch 25, batch 41750, giga_loss[loss=0.3223, simple_loss=0.3849, pruned_loss=0.1299, over 28307.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3816, pruned_loss=0.1287, over 5665294.84 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1125, over 5700716.96 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3851, pruned_loss=0.1313, over 5647672.77 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:36:52,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3803, 1.6096, 1.6484, 1.5028], device='cuda:0'), covar=tensor([0.1639, 0.1434, 0.1533, 0.1469], device='cuda:0'), in_proj_covar=tensor([0.0493, 0.0758, 0.0725, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 05:36:53,074 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1135592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:37:25,564 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 05:37:37,227 INFO [train.py:968] (0/2) Epoch 25, batch 41800, giga_loss[loss=0.2523, simple_loss=0.338, pruned_loss=0.0833, over 28866.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3778, pruned_loss=0.1251, over 5666706.23 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3591, pruned_loss=0.1125, over 5696516.59 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3813, pruned_loss=0.1276, over 5655757.43 frames. ], batch size: 66, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:38:19,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.598e+02 1.874e+03 2.330e+03 3.216e+03 9.962e+03, threshold=4.660e+03, percent-clipped=9.0
+2023-03-13 05:38:23,328 INFO [train.py:968] (0/2) Epoch 25, batch 41850, giga_loss[loss=0.3066, simple_loss=0.3656, pruned_loss=0.1239, over 27969.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3734, pruned_loss=0.1219, over 5663562.72 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3585, pruned_loss=0.1122, over 5699945.84 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3776, pruned_loss=0.1248, over 5650038.35 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:39:07,138 INFO [train.py:968] (0/2) Epoch 25, batch 41900, libri_loss[loss=0.3044, simple_loss=0.3706, pruned_loss=0.1191, over 29601.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3716, pruned_loss=0.1213, over 5653693.41 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3584, pruned_loss=0.112, over 5701949.50 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3756, pruned_loss=0.1242, over 5639202.76 frames. ], batch size: 75, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:39:15,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2991, 1.3763, 3.5013, 3.2277], device='cuda:0'), covar=tensor([0.1527, 0.2726, 0.0461, 0.1913], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0664, 0.0981, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 05:39:20,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2710, 3.1082, 1.3535, 1.4570], device='cuda:0'), covar=tensor([0.1024, 0.0454, 0.0925, 0.1378], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0563, 0.0399, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 05:39:27,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9916, 3.8427, 3.6337, 1.8052], device='cuda:0'), covar=tensor([0.0690, 0.0770, 0.0759, 0.2149], device='cuda:0'), in_proj_covar=tensor([0.1285, 0.1187, 0.0998, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 05:39:49,124 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.913e+03 2.362e+03 3.239e+03 1.028e+04, threshold=4.723e+03, percent-clipped=11.0
+2023-03-13 05:39:53,424 INFO [train.py:968] (0/2) Epoch 25, batch 41950, giga_loss[loss=0.2839, simple_loss=0.3593, pruned_loss=0.1042, over 28853.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3722, pruned_loss=0.1213, over 5673954.15 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3585, pruned_loss=0.1121, over 5706563.90 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3756, pruned_loss=0.1237, over 5657363.32 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:40:42,470 INFO [train.py:968] (0/2) Epoch 25, batch 42000, giga_loss[loss=0.3326, simple_loss=0.3933, pruned_loss=0.136, over 28027.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3708, pruned_loss=0.1202, over 5676456.00 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3582, pruned_loss=0.112, over 5709820.85 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3743, pruned_loss=0.1225, over 5659176.88 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:40:42,475 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 05:40:51,196 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2007, simple_loss=0.308, pruned_loss=0.04671, over 944034.00 frames. 
+2023-03-13 05:40:51,196 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 05:41:38,752 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1135884.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 05:41:39,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.540e+03 2.025e+03 2.855e+03 6.054e+03, threshold=4.051e+03, percent-clipped=4.0
+2023-03-13 05:41:43,187 INFO [train.py:968] (0/2) Epoch 25, batch 42050, giga_loss[loss=0.2685, simple_loss=0.3486, pruned_loss=0.0942, over 29094.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3689, pruned_loss=0.117, over 5686972.77 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3578, pruned_loss=0.1117, over 5713411.25 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3721, pruned_loss=0.1192, over 5669540.99 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:41:53,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1135896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:42:27,836 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-13 05:42:34,531 INFO [train.py:968] (0/2) Epoch 25, batch 42100, giga_loss[loss=0.3043, simple_loss=0.3569, pruned_loss=0.1258, over 23619.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3705, pruned_loss=0.1159, over 5689527.21 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3575, pruned_loss=0.1114, over 5718910.19 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3737, pruned_loss=0.1181, over 5670020.67 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:43:00,265 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9715, 1.1612, 1.0608, 0.9245], device='cuda:0'), covar=tensor([0.2392, 0.2784, 0.1789, 0.2255], device='cuda:0'), in_proj_covar=tensor([0.2025, 0.1959, 0.1878, 0.2034], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 05:43:10,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4451, 1.7915, 1.4051, 1.3175], device='cuda:0'), covar=tensor([0.2691, 0.2695, 0.3219, 0.2365], device='cuda:0'), in_proj_covar=tensor([0.1563, 0.1126, 0.1381, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 05:43:18,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.832e+03 2.255e+03 3.182e+03 1.124e+04, threshold=4.510e+03, percent-clipped=9.0
+2023-03-13 05:43:20,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0946, 1.3565, 1.3865, 1.0493], device='cuda:0'), covar=tensor([0.3469, 0.2880, 0.2019, 0.2941], device='cuda:0'), in_proj_covar=tensor([0.2023, 0.1956, 0.1875, 0.2031], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 05:43:23,421 INFO [train.py:968] (0/2) Epoch 25, batch 42150, giga_loss[loss=0.3025, simple_loss=0.3662, pruned_loss=0.1194, over 28411.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3715, pruned_loss=0.1166, over 5687706.52 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3576, pruned_loss=0.1114, over 5720959.22 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3741, pruned_loss=0.1184, over 5670041.98 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:43:34,261 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1136000.pt
+2023-03-13 05:44:04,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0206, 3.8681, 3.6939, 1.7817], device='cuda:0'), covar=tensor([0.0718, 0.0804, 0.0784, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.1285, 0.1188, 0.1000, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 05:44:07,967 INFO [train.py:968] (0/2) Epoch 25, batch 42200, giga_loss[loss=0.2771, simple_loss=0.3488, pruned_loss=0.1027, over 28666.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3719, pruned_loss=0.1177, over 5687993.61 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3574, pruned_loss=0.1112, over 5724395.88 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3747, pruned_loss=0.1196, over 5669392.86 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:44:33,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7095, 4.5444, 4.3352, 2.0671], device='cuda:0'), covar=tensor([0.0588, 0.0758, 0.0787, 0.2002], device='cuda:0'), in_proj_covar=tensor([0.1287, 0.1191, 0.1001, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 05:44:48,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:44:48,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.775e+03 2.214e+03 2.950e+03 6.679e+03, threshold=4.428e+03, percent-clipped=2.0
+2023-03-13 05:44:51,349 INFO [train.py:968] (0/2) Epoch 25, batch 42250, giga_loss[loss=0.2656, simple_loss=0.3326, pruned_loss=0.09928, over 28936.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3713, pruned_loss=0.1182, over 5683336.59 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3575, pruned_loss=0.1114, over 5720730.57 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.374, pruned_loss=0.1197, over 5671186.20 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:45:07,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-13 05:45:37,615 INFO [train.py:968] (0/2) Epoch 25, batch 42300, giga_loss[loss=0.3401, simple_loss=0.396, pruned_loss=0.1421, over 28711.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3711, pruned_loss=0.1195, over 5681234.29 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3579, pruned_loss=0.1117, over 5724270.78 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3731, pruned_loss=0.1205, over 5667656.28 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:46:21,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.686e+03 2.160e+03 3.207e+03 8.104e+03, threshold=4.321e+03, percent-clipped=10.0
+2023-03-13 05:46:26,127 INFO [train.py:968] (0/2) Epoch 25, batch 42350, giga_loss[loss=0.2626, simple_loss=0.3311, pruned_loss=0.09705, over 28834.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3685, pruned_loss=0.1182, over 5672382.33 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3576, pruned_loss=0.1114, over 5724442.84 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3705, pruned_loss=0.1194, over 5660804.11 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:46:57,821 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:47:16,829 INFO [train.py:968] (0/2) Epoch 25, batch 42400, giga_loss[loss=0.28, simple_loss=0.3577, pruned_loss=0.1012, over 28759.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3675, pruned_loss=0.116, over 5683442.61 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3573, pruned_loss=0.1112, over 5726346.60 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3695, pruned_loss=0.1173, over 5672081.93 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:47:24,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4508, 2.5105, 2.4826, 2.1562], device='cuda:0'), covar=tensor([0.1931, 0.2408, 0.2133, 0.2406], device='cuda:0'), in_proj_covar=tensor([0.0495, 0.0760, 0.0728, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 05:47:34,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136259.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:47:45,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:48:00,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.623e+03 2.013e+03 2.789e+03 5.941e+03, threshold=4.025e+03, percent-clipped=4.0
+2023-03-13 05:48:03,447 INFO [train.py:968] (0/2) Epoch 25, batch 42450, giga_loss[loss=0.3232, simple_loss=0.3837, pruned_loss=0.1314, over 28634.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3672, pruned_loss=0.1151, over 5685672.11 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3568, pruned_loss=0.1109, over 5726188.92 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3695, pruned_loss=0.1164, over 5675747.55 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:48:50,053 INFO [train.py:968] (0/2) Epoch 25, batch 42500, giga_loss[loss=0.2789, simple_loss=0.36, pruned_loss=0.09887, over 28958.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.368, pruned_loss=0.1157, over 5694604.39 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3577, pruned_loss=0.1117, over 5729453.78 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3692, pruned_loss=0.1162, over 5682579.60 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:48:51,355 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:49:15,656 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5813, 1.7584, 1.8335, 1.5539], device='cuda:0'), covar=tensor([0.2630, 0.2411, 0.2492, 0.2567], device='cuda:0'), in_proj_covar=tensor([0.2019, 0.1955, 0.1875, 0.2031], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 05:49:32,305 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.601e+02 1.751e+03 2.159e+03 3.178e+03 1.230e+04, threshold=4.319e+03, percent-clipped=10.0
+2023-03-13 05:49:35,279 INFO [train.py:968] (0/2) Epoch 25, batch 42550, giga_loss[loss=0.3089, simple_loss=0.3596, pruned_loss=0.1291, over 28729.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3665, pruned_loss=0.1153, over 5688724.17 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3577, pruned_loss=0.1114, over 5730345.34 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3677, pruned_loss=0.116, over 5677827.44 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:49:47,799 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1136402.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 05:49:50,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1622, 1.2327, 3.2669, 3.0006], device='cuda:0'), covar=tensor([0.1578, 0.2743, 0.0540, 0.1403], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0666, 0.0982, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 05:49:50,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1136405.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 05:49:58,311 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1136414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:50:00,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7048, 1.8953, 1.6557, 1.7914], device='cuda:0'), covar=tensor([0.2104, 0.2140, 0.2145, 0.2129], device='cuda:0'), in_proj_covar=tensor([0.1562, 0.1125, 0.1380, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 05:50:01,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1136417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:50:19,057 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1136434.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:50:23,411 INFO [train.py:968] (0/2) Epoch 25, batch 42600, giga_loss[loss=0.2983, simple_loss=0.3716, pruned_loss=0.1125, over 28988.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3667, pruned_loss=0.1162, over 5681094.42 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3579, pruned_loss=0.1115, over 5723162.24 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3676, pruned_loss=0.1167, over 5677297.99 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:50:29,333 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1136446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:50:43,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:51:06,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.161e+03 1.856e+03 2.486e+03 3.580e+03 9.197e+03, threshold=4.972e+03, percent-clipped=12.0
+2023-03-13 05:51:10,681 INFO [train.py:968] (0/2) Epoch 25, batch 42650, giga_loss[loss=0.2939, simple_loss=0.3652, pruned_loss=0.1113, over 28838.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3664, pruned_loss=0.1171, over 5672132.05 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3583, pruned_loss=0.1117, over 5718473.21 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3669, pruned_loss=0.1174, over 5671187.77 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:51:25,204 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6314, 1.1212, 4.3509, 3.4932], device='cuda:0'), covar=tensor([0.1509, 0.2994, 0.0416, 0.1112], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0665, 0.0981, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 05:51:58,216 INFO [train.py:968] (0/2) Epoch 25, batch 42700, giga_loss[loss=0.3294, simple_loss=0.3791, pruned_loss=0.1399, over 28029.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3657, pruned_loss=0.1176, over 5670383.34 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3585, pruned_loss=0.1119, over 5721376.88 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3661, pruned_loss=0.1178, over 5666133.59 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:52:36,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:52:43,014 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.755e+03 2.309e+03 2.941e+03 5.995e+03, threshold=4.617e+03, percent-clipped=5.0
+2023-03-13 05:52:45,274 INFO [train.py:968] (0/2) Epoch 25, batch 42750, giga_loss[loss=0.2773, simple_loss=0.3461, pruned_loss=0.1043, over 28819.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3637, pruned_loss=0.1163, over 5675738.66 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3589, pruned_loss=0.1122, over 5721747.29 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3639, pruned_loss=0.1164, over 5670712.61 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:52:50,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:52:58,979 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1136603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:53:05,117 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1136606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:53:07,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2354, 2.6347, 1.1886, 1.5269], device='cuda:0'), covar=tensor([0.1043, 0.0439, 0.0889, 0.1304], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0567, 0.0401, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 05:53:33,800 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1136635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:53:36,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-13 05:53:39,242 INFO [train.py:968] (0/2) Epoch 25, batch 42800, giga_loss[loss=0.3685, simple_loss=0.4225, pruned_loss=0.1573, over 28911.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3638, pruned_loss=0.1165, over 5682159.46 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3588, pruned_loss=0.1123, over 5722712.55 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.364, pruned_loss=0.1165, over 5677272.75 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:53:44,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5650, 1.7670, 1.4714, 1.5269], device='cuda:0'), covar=tensor([0.2823, 0.2754, 0.3144, 0.2481], device='cuda:0'), in_proj_covar=tensor([0.1564, 0.1127, 0.1382, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 05:53:47,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6633, 4.4496, 4.2351, 2.2205], device='cuda:0'), covar=tensor([0.0709, 0.0913, 0.0997, 0.1913], device='cuda:0'), in_proj_covar=tensor([0.1287, 0.1188, 0.0998, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 05:54:19,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.197e+03 1.696e+03 2.291e+03 3.064e+03 5.612e+03, threshold=4.582e+03, percent-clipped=7.0
+2023-03-13 05:54:23,126 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 05:54:23,283 INFO [train.py:968] (0/2) Epoch 25, batch 42850, giga_loss[loss=0.2764, simple_loss=0.3585, pruned_loss=0.0972, over 28869.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3644, pruned_loss=0.116, over 5688656.54 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.359, pruned_loss=0.1122, over 5728875.04 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3646, pruned_loss=0.1162, over 5677577.13 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:54:44,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:55:04,464 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1136738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:55:05,443 INFO [train.py:968] (0/2) Epoch 25, batch 42900, giga_loss[loss=0.2679, simple_loss=0.3463, pruned_loss=0.09475, over 28940.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3644, pruned_loss=0.1155, over 5685992.06 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1125, over 5726251.32 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3649, pruned_loss=0.1156, over 5677633.03 frames. ], batch size: 199, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:55:06,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1136741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:55:32,682 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1136770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:55:47,228 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.806e+03 2.240e+03 3.282e+03 7.768e+03, threshold=4.479e+03, percent-clipped=8.0
+2023-03-13 05:55:49,787 INFO [train.py:968] (0/2) Epoch 25, batch 42950, libri_loss[loss=0.2355, simple_loss=0.3075, pruned_loss=0.08178, over 28596.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3639, pruned_loss=0.1144, over 5683561.25 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3579, pruned_loss=0.1119, over 5729110.68 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3654, pruned_loss=0.1152, over 5672977.50 frames. ], batch size: 63, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:56:38,463 INFO [train.py:968] (0/2) Epoch 25, batch 43000, giga_loss[loss=0.3611, simple_loss=0.3962, pruned_loss=0.163, over 23264.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3652, pruned_loss=0.1159, over 5676420.23 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3579, pruned_loss=0.1119, over 5731551.27 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3666, pruned_loss=0.1164, over 5665062.07 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:56:45,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5517, 1.7976, 1.2834, 1.3229], device='cuda:0'), covar=tensor([0.1017, 0.0582, 0.1035, 0.1173], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0449, 0.0519, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 05:56:58,286 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1136859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:57:01,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1136862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:57:25,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.250e+03 1.778e+03 2.186e+03 3.081e+03 1.026e+04, threshold=4.372e+03, percent-clipped=10.0
+2023-03-13 05:57:26,704 INFO [train.py:968] (0/2) Epoch 25, batch 43050, giga_loss[loss=0.4397, simple_loss=0.4793, pruned_loss=0.2, over 26484.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3678, pruned_loss=0.1182, over 5673964.32 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3582, pruned_loss=0.1121, over 5735534.28 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3688, pruned_loss=0.1186, over 5660303.18 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:57:27,615 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1136891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:57:53,146 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:57:53,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6777, 4.5987, 1.7454, 1.9010], device='cuda:0'), covar=tensor([0.0944, 0.0417, 0.0856, 0.1233], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0565, 0.0400, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 05:58:02,312 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136927.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:58:16,430 INFO [train.py:968] (0/2) Epoch 25, batch 43100, giga_loss[loss=0.3122, simple_loss=0.3768, pruned_loss=0.1238, over 28691.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3713, pruned_loss=0.1221, over 5665305.86 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3585, pruned_loss=0.1124, over 5732560.56 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3721, pruned_loss=0.1224, over 5655753.42 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:58:19,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3611, 1.6594, 1.3553, 1.4662], device='cuda:0'), covar=tensor([0.1776, 0.1770, 0.2018, 0.1846], device='cuda:0'), in_proj_covar=tensor([0.0495, 0.0760, 0.0728, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 05:58:33,903 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:59:08,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.336e+03 2.063e+03 2.514e+03 3.469e+03 9.104e+03, threshold=5.029e+03, percent-clipped=10.0
+2023-03-13 05:59:11,660 INFO [train.py:968] (0/2) Epoch 25, batch 43150, giga_loss[loss=0.4213, simple_loss=0.45, pruned_loss=0.1963, over 26328.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3728, pruned_loss=0.1248, over 5659886.24 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3585, pruned_loss=0.1123, over 5733447.76 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3735, pruned_loss=0.1251, over 5651346.06 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:59:59,441 INFO [train.py:968] (0/2) Epoch 25, batch 43200, giga_loss[loss=0.295, simple_loss=0.3598, pruned_loss=0.115, over 28611.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3739, pruned_loss=0.126, over 5671353.07 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3585, pruned_loss=0.1124, over 5737688.43 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3748, pruned_loss=0.1265, over 5659286.91 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 06:00:16,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6632, 1.8353, 1.5266, 1.8913], device='cuda:0'), covar=tensor([0.2576, 0.2806, 0.3054, 0.2555], device='cuda:0'), in_proj_covar=tensor([0.1565, 0.1128, 0.1381, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 06:00:42,003 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.177e+03 1.837e+03 2.662e+03 3.504e+03 8.304e+03, threshold=5.324e+03, percent-clipped=6.0
+2023-03-13 06:00:42,597 INFO [train.py:968] (0/2) Epoch 25, batch 43250, giga_loss[loss=0.2747, simple_loss=0.3499, pruned_loss=0.09971, over 28937.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.371, pruned_loss=0.1236, over 5666568.48 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3588, pruned_loss=0.1127, over 5730755.02 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3718, pruned_loss=0.1241, over 5661444.97 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:00:48,419 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1137097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:00:51,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1137100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:01:07,014 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-13 06:01:16,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1137129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:01:27,858 INFO [train.py:968] (0/2) Epoch 25, batch 43300, giga_loss[loss=0.2842, simple_loss=0.3545, pruned_loss=0.107, over 28691.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3703, pruned_loss=0.1224, over 5655374.81 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3591, pruned_loss=0.1129, over 5711911.63 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.371, pruned_loss=0.1229, over 5665768.93 frames. ], batch size: 242, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:02:08,033 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.768e+03 2.123e+03 2.886e+03 1.161e+04, threshold=4.246e+03, percent-clipped=3.0
+2023-03-13 06:02:08,577 INFO [train.py:968] (0/2) Epoch 25, batch 43350, giga_loss[loss=0.2891, simple_loss=0.3548, pruned_loss=0.1117, over 28850.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3692, pruned_loss=0.1198, over 5666963.35 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3592, pruned_loss=0.1127, over 5710306.85 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3701, pruned_loss=0.1207, over 5675986.30 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:02:19,513 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-13 06:02:55,177 INFO [train.py:968] (0/2) Epoch 25, batch 43400, giga_loss[loss=0.2952, simple_loss=0.3386, pruned_loss=0.1259, over 23892.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3672, pruned_loss=0.1193, over 5654631.17 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3592, pruned_loss=0.1129, over 5713073.11 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.368, pruned_loss=0.1199, over 5658680.98 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:03:40,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.209e+03 1.862e+03 2.349e+03 3.272e+03 7.489e+03, threshold=4.698e+03, percent-clipped=6.0
+2023-03-13 06:03:41,971 INFO [train.py:968] (0/2) Epoch 25, batch 43450, giga_loss[loss=0.2881, simple_loss=0.3572, pruned_loss=0.1094, over 28643.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3655, pruned_loss=0.1189, over 5661867.20 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3589, pruned_loss=0.1127, over 5713673.38 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3664, pruned_loss=0.1196, over 5663994.10 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:03:45,175 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1137293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:03:52,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1137302.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:04:00,538 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1137312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:04:25,018 INFO [train.py:968] (0/2) Epoch 25, batch 43500, giga_loss[loss=0.3094, simple_loss=0.3706, pruned_loss=0.1241, over 28959.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3657, pruned_loss=0.1197, over 5662838.22 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3594, pruned_loss=0.1131, over 5719902.17 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3662, pruned_loss=0.1202, over 5657303.03 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:04:37,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6120, 1.8733, 1.5614, 1.5312], device='cuda:0'), covar=tensor([0.2370, 0.2362, 0.2646, 0.2254], device='cuda:0'), in_proj_covar=tensor([0.1563, 0.1126, 0.1379, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 06:05:13,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.970e+02 1.861e+03 2.481e+03 3.924e+03 7.299e+03, threshold=4.962e+03, percent-clipped=9.0
+2023-03-13 06:05:14,157 INFO [train.py:968] (0/2) Epoch 25, batch 43550, giga_loss[loss=0.3051, simple_loss=0.3916, pruned_loss=0.1093, over 28954.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3692, pruned_loss=0.1216, over 5673742.58 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3597, pruned_loss=0.1134, over 5722202.12 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3695, pruned_loss=0.1218, over 5666443.09 frames. ], batch size: 164, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:05:45,058 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4832, 1.6278, 1.6624, 1.2646], device='cuda:0'), covar=tensor([0.2004, 0.2741, 0.1669, 0.1970], device='cuda:0'), in_proj_covar=tensor([0.0918, 0.0710, 0.0965, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 06:05:57,593 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1137436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:05:59,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1137439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:06:00,129 INFO [train.py:968] (0/2) Epoch 25, batch 43600, giga_loss[loss=0.3014, simple_loss=0.3763, pruned_loss=0.1133, over 28593.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3715, pruned_loss=0.1207, over 5659918.92 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3596, pruned_loss=0.1132, over 5712871.92 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3721, pruned_loss=0.1212, over 5661901.87 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 06:06:05,705 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1137445.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:06:10,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1137448.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 06:06:28,177 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1137468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:06:35,055 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1137477.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:06:39,819 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-13 06:06:50,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.532e+03 2.163e+03 2.865e+03 5.541e+03, threshold=4.325e+03, percent-clipped=3.0
+2023-03-13 06:06:50,808 INFO [train.py:968] (0/2) Epoch 25, batch 43650, giga_loss[loss=0.3018, simple_loss=0.3751, pruned_loss=0.1142, over 28929.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3739, pruned_loss=0.1207, over 5657580.33 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3598, pruned_loss=0.1134, over 5715905.28 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3744, pruned_loss=0.1211, over 5655242.36 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:07:38,277 INFO [train.py:968] (0/2) Epoch 25, batch 43700, giga_loss[loss=0.266, simple_loss=0.3372, pruned_loss=0.09741, over 28392.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3752, pruned_loss=0.1217, over 5670855.04 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3595, pruned_loss=0.1132, over 5718828.91 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3761, pruned_loss=0.1223, over 5665758.46 frames. ], batch size: 78, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:08:23,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.898e+03 2.620e+03 3.530e+03 1.095e+04, threshold=5.240e+03, percent-clipped=15.0
+2023-03-13 06:08:23,996 INFO [train.py:968] (0/2) Epoch 25, batch 43750, giga_loss[loss=0.2968, simple_loss=0.3675, pruned_loss=0.1131, over 28898.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3774, pruned_loss=0.1238, over 5671679.52 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3597, pruned_loss=0.1133, over 5720260.82 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3781, pruned_loss=0.1243, over 5665647.99 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:08:47,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-13 06:09:09,533 INFO [train.py:968] (0/2) Epoch 25, batch 43800, giga_loss[loss=0.2971, simple_loss=0.3639, pruned_loss=0.1152, over 28987.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3767, pruned_loss=0.124, over 5678501.34 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3596, pruned_loss=0.1131, over 5722617.13 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3777, pruned_loss=0.1247, over 5670994.98 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:09:49,085 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1137681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:09:53,546 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-13 06:09:54,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1137687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:09:57,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.845e+03 2.215e+03 2.909e+03 7.927e+03, threshold=4.429e+03, percent-clipped=2.0
+2023-03-13 06:09:57,224 INFO [train.py:968] (0/2) Epoch 25, batch 43850, libri_loss[loss=0.2686, simple_loss=0.3332, pruned_loss=0.1019, over 29571.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3757, pruned_loss=0.1246, over 5668266.98 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3598, pruned_loss=0.1132, over 5725632.07 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3768, pruned_loss=0.1254, over 5658505.16 frames. ], batch size: 74, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:09:58,755 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1137692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:10:14,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4591, 1.6878, 1.5840, 1.5401], device='cuda:0'), covar=tensor([0.2087, 0.2355, 0.2611, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.0492, 0.0756, 0.0725, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 06:10:30,331 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1137729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:10:42,952 INFO [train.py:968] (0/2) Epoch 25, batch 43900, giga_loss[loss=0.3299, simple_loss=0.3686, pruned_loss=0.1455, over 23520.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3745, pruned_loss=0.1246, over 5665504.72 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3604, pruned_loss=0.1137, over 5722983.92 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3751, pruned_loss=0.1249, over 5658587.08 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:11:03,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6817, 3.5463, 3.3865, 1.7657], device='cuda:0'), covar=tensor([0.0802, 0.0881, 0.0800, 0.2277], device='cuda:0'), in_proj_covar=tensor([0.1290, 0.1192, 0.1004, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 06:11:28,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.222e+02 1.781e+03 2.211e+03 2.934e+03 1.030e+04, threshold=4.421e+03, percent-clipped=7.0
+2023-03-13 06:11:28,562 INFO [train.py:968] (0/2) Epoch 25, batch 43950, giga_loss[loss=0.3146, simple_loss=0.3776, pruned_loss=0.1258, over 28214.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3735, pruned_loss=0.1242, over 5665871.99 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3605, pruned_loss=0.1138, over 5725898.52 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.374, pruned_loss=0.1246, over 5656957.05 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:12:09,111 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1137830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:12:12,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1137833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:12:17,538 INFO [train.py:968] (0/2) Epoch 25, batch 44000, giga_loss[loss=0.2726, simple_loss=0.3404, pruned_loss=0.1024, over 28633.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3741, pruned_loss=0.1251, over 5652014.63 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1142, over 5720068.02 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3743, pruned_loss=0.1252, over 5647958.30 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:12:40,110 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1137862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:13:06,233 INFO [train.py:968] (0/2) Epoch 25, batch 44050, giga_loss[loss=0.322, simple_loss=0.3654, pruned_loss=0.1393, over 28777.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3718, pruned_loss=0.1243, over 5655198.48 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3607, pruned_loss=0.1139, over 5721914.89 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3725, pruned_loss=0.1248, over 5649841.75 frames. ], batch size: 99, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:13:07,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.903e+03 2.375e+03 3.552e+03 8.946e+03, threshold=4.750e+03, percent-clipped=14.0
+2023-03-13 06:13:53,574 INFO [train.py:968] (0/2) Epoch 25, batch 44100, giga_loss[loss=0.2569, simple_loss=0.3385, pruned_loss=0.08763, over 28821.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3702, pruned_loss=0.1231, over 5664352.28 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3607, pruned_loss=0.1139, over 5723692.39 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3708, pruned_loss=0.1236, over 5657926.24 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:14:34,898 INFO [train.py:968] (0/2) Epoch 25, batch 44150, libri_loss[loss=0.3311, simple_loss=0.3924, pruned_loss=0.1349, over 29376.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3682, pruned_loss=0.1212, over 5666322.82 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3605, pruned_loss=0.1136, over 5727809.75 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3692, pruned_loss=0.1222, over 5654708.22 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:14:35,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.740e+03 2.153e+03 2.897e+03 7.476e+03, threshold=4.306e+03, percent-clipped=6.0
+2023-03-13 06:14:44,413 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1138000.pt
+2023-03-13 06:15:26,937 INFO [train.py:968] (0/2) Epoch 25, batch 44200, giga_loss[loss=0.3557, simple_loss=0.4148, pruned_loss=0.1483, over 28704.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3701, pruned_loss=0.1219, over 5657091.68 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3599, pruned_loss=0.1133, over 5730538.24 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3716, pruned_loss=0.1232, over 5644394.37 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:15:28,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:15:38,669 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-13 06:15:41,916 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:15:42,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7586, 1.7721, 1.9747, 1.5189], device='cuda:0'), covar=tensor([0.1948, 0.2562, 0.1497, 0.1825], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0713, 0.0969, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 06:15:49,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5674, 1.7212, 1.1912, 1.3172], device='cuda:0'), covar=tensor([0.0995, 0.0592, 0.1129, 0.1168], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0451, 0.0521, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 06:15:50,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:16:12,192 INFO [train.py:968] (0/2) Epoch 25, batch 44250, giga_loss[loss=0.2498, simple_loss=0.3328, pruned_loss=0.08343, over 28930.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3718, pruned_loss=0.1231, over 5642979.12 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3604, pruned_loss=0.1136, over 5712646.67 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3729, pruned_loss=0.124, over 5645925.11 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:16:13,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.669e+03 2.111e+03 2.791e+03 1.166e+04, threshold=4.223e+03, percent-clipped=9.0
+2023-03-13 06:16:26,351 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:16:49,554 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 06:17:01,906 INFO [train.py:968] (0/2) Epoch 25, batch 44300, giga_loss[loss=0.3032, simple_loss=0.3888, pruned_loss=0.1088, over 28463.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3718, pruned_loss=0.1228, over 5661048.14 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3603, pruned_loss=0.1137, over 5715821.84 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3729, pruned_loss=0.1236, over 5659643.81 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:17:11,996 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:17:45,440 INFO [train.py:968] (0/2) Epoch 25, batch 44350, giga_loss[loss=0.3497, simple_loss=0.3984, pruned_loss=0.1505, over 27926.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3725, pruned_loss=0.1201, over 5664687.44 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3606, pruned_loss=0.1137, over 5714716.18 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3733, pruned_loss=0.1209, over 5663062.10 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:17:45,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.688e+02 1.619e+03 2.223e+03 2.744e+03 5.488e+03, threshold=4.447e+03, percent-clipped=7.0
+2023-03-13 06:17:53,657 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:17:56,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:04,216 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:05,011 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 06:18:06,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:21,604 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:28,253 INFO [train.py:968] (0/2) Epoch 25, batch 44400, giga_loss[loss=0.2998, simple_loss=0.3812, pruned_loss=0.1092, over 29135.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3736, pruned_loss=0.1189, over 5669425.76 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3605, pruned_loss=0.1135, over 5719066.87 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3747, pruned_loss=0.1198, over 5662991.83 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 06:18:31,633 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:36,331 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-13 06:18:36,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4381, 1.2046, 1.1261, 1.6290], device='cuda:0'), covar=tensor([0.0729, 0.0362, 0.0367, 0.0849], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 06:18:36,772 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:38,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:54,368 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:19:08,390 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:19:18,616 INFO [train.py:968] (0/2) Epoch 25, batch 44450, giga_loss[loss=0.3971, simple_loss=0.4253, pruned_loss=0.1844, over 26628.00 frames. ], tot_loss[loss=0.309, simple_loss=0.376, pruned_loss=0.121, over 5657391.51 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3604, pruned_loss=0.1134, over 5719459.07 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3773, pruned_loss=0.122, over 5650564.76 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:19:19,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.607e+03 2.089e+03 2.635e+03 6.469e+03, threshold=4.178e+03, percent-clipped=4.0
+2023-03-13 06:19:28,281 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:19:46,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3717, 1.7561, 1.6572, 1.5283], device='cuda:0'), covar=tensor([0.2212, 0.1885, 0.2486, 0.2164], device='cuda:0'), in_proj_covar=tensor([0.0492, 0.0756, 0.0724, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 06:20:03,078 INFO [train.py:968] (0/2) Epoch 25, batch 44500, giga_loss[loss=0.3025, simple_loss=0.3663, pruned_loss=0.1194, over 29007.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3776, pruned_loss=0.1234, over 5657259.79 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3602, pruned_loss=0.1134, over 5712674.31 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3794, pruned_loss=0.1245, over 5655843.81 frames. ], batch size: 120, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:20:27,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:20:53,500 INFO [train.py:968] (0/2) Epoch 25, batch 44550, giga_loss[loss=0.2918, simple_loss=0.3631, pruned_loss=0.1102, over 28985.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3778, pruned_loss=0.1243, over 5664080.71 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3599, pruned_loss=0.1132, over 5715569.28 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3798, pruned_loss=0.1256, over 5659279.87 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:20:55,080 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.088e+03 1.762e+03 2.383e+03 3.454e+03 7.781e+03, threshold=4.765e+03, percent-clipped=15.0
+2023-03-13 06:21:03,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2597, 1.4699, 1.4904, 1.2310], device='cuda:0'), covar=tensor([0.1769, 0.1632, 0.2198, 0.1855], device='cuda:0'), in_proj_covar=tensor([0.0492, 0.0758, 0.0726, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 06:21:15,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:21:38,406 INFO [train.py:968] (0/2) Epoch 25, batch 44600, giga_loss[loss=0.2833, simple_loss=0.3571, pruned_loss=0.1048, over 28944.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3766, pruned_loss=0.1237, over 5671987.71 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3597, pruned_loss=0.1129, over 5718885.76 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3788, pruned_loss=0.1253, over 5664298.17 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:21:43,019 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-13 06:22:23,572 INFO [train.py:968] (0/2) Epoch 25, batch 44650, giga_loss[loss=0.2799, simple_loss=0.356, pruned_loss=0.1019, over 28574.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3753, pruned_loss=0.1225, over 5672566.41 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3596, pruned_loss=0.1129, over 5721977.90 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3773, pruned_loss=0.1238, over 5663277.76 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:22:26,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.769e+03 2.475e+03 4.102e+03 1.365e+04, threshold=4.951e+03, percent-clipped=18.0
+2023-03-13 06:22:37,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 06:22:49,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2876, 3.0184, 1.4045, 1.4736], device='cuda:0'), covar=tensor([0.1039, 0.0387, 0.0968, 0.1379], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0566, 0.0401, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 06:22:58,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:23:10,468 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-13 06:23:11,968 INFO [train.py:968] (0/2) Epoch 25, batch 44700, giga_loss[loss=0.3446, simple_loss=0.4104, pruned_loss=0.1394, over 28873.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.375, pruned_loss=0.1203, over 5681657.78 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1129, over 5724302.51 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3768, pruned_loss=0.1215, over 5671633.20 frames. ], batch size: 112, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:23:23,620 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:23:31,010 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:23:32,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:23:57,425 INFO [train.py:968] (0/2) Epoch 25, batch 44750, giga_loss[loss=0.2618, simple_loss=0.3384, pruned_loss=0.0926, over 28482.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3756, pruned_loss=0.1199, over 5690163.12 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5726733.01 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3771, pruned_loss=0.1209, over 5678852.17 frames. ], batch size: 78, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:23:57,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:23:58,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.532e+03 1.919e+03 2.813e+03 6.354e+03, threshold=3.837e+03, percent-clipped=2.0
+2023-03-13 06:24:09,420 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-13 06:24:18,611 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 06:24:43,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:24:43,727 INFO [train.py:968] (0/2) Epoch 25, batch 44800, giga_loss[loss=0.3119, simple_loss=0.3791, pruned_loss=0.1223, over 29017.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3772, pruned_loss=0.122, over 5667385.82 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5726733.89 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3787, pruned_loss=0.1231, over 5657366.57 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 06:24:46,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:25:00,708 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:25:10,585 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:25:12,852 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:25:17,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:25:30,068 INFO [train.py:968] (0/2) Epoch 25, batch 44850, giga_loss[loss=0.2791, simple_loss=0.3515, pruned_loss=0.1034, over 29074.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3746, pruned_loss=0.1206, over 5659400.48 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5719609.27 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3764, pruned_loss=0.1217, over 5656876.15 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:25:33,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.256e+03 1.847e+03 2.583e+03 3.719e+03 1.241e+04, threshold=5.166e+03, percent-clipped=22.0
+2023-03-13 06:25:40,485 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:26:08,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 06:26:10,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4076, 1.3567, 3.8004, 3.2078], device='cuda:0'), covar=tensor([0.1638, 0.2724, 0.0492, 0.1145], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0666, 0.0986, 0.0956], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 06:26:17,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:26:18,403 INFO [train.py:968] (0/2) Epoch 25, batch 44900, giga_loss[loss=0.2652, simple_loss=0.3433, pruned_loss=0.09352, over 29088.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3731, pruned_loss=0.1207, over 5662621.94 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3599, pruned_loss=0.1129, over 5713386.79 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3746, pruned_loss=0.1216, over 5665094.50 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:26:29,791 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-13 06:26:58,961 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:27:01,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:27:06,072 INFO [train.py:968] (0/2) Epoch 25, batch 44950, giga_loss[loss=0.2345, simple_loss=0.3188, pruned_loss=0.0751, over 29115.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3716, pruned_loss=0.1206, over 5653094.13 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3601, pruned_loss=0.113, over 5710166.31 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3727, pruned_loss=0.1213, over 5657043.38 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:27:07,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.149e+03 1.783e+03 2.249e+03 3.289e+03 1.049e+04, threshold=4.498e+03, percent-clipped=7.0
+2023-03-13 06:27:27,586 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:27:32,613 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:27:35,245 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:27:51,422 INFO [train.py:968] (0/2) Epoch 25, batch 45000, giga_loss[loss=0.2807, simple_loss=0.3527, pruned_loss=0.1043, over 28948.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3701, pruned_loss=0.1202, over 5660380.57 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.1131, over 5716316.46 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3712, pruned_loss=0.1209, over 5656681.24 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:27:51,425 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 06:28:00,388 INFO [train.py:1012] (0/2) Epoch 25, validation: loss=0.2048, simple_loss=0.3141, pruned_loss=0.04768, over 944034.00 frames. 
+2023-03-13 06:28:00,389 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 06:28:11,971 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:28:23,131 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138866.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:28:36,395 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:28:38,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:28:41,288 INFO [train.py:968] (0/2) Epoch 25, batch 45050, giga_loss[loss=0.2936, simple_loss=0.3707, pruned_loss=0.1083, over 28999.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3697, pruned_loss=0.1207, over 5644125.67 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3604, pruned_loss=0.1132, over 5701529.06 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3707, pruned_loss=0.1214, over 5653245.30 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:28:43,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8325, 4.6719, 4.4361, 2.2523], device='cuda:0'), covar=tensor([0.0656, 0.0808, 0.0893, 0.1858], device='cuda:0'), in_proj_covar=tensor([0.1298, 0.1200, 0.1011, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 06:28:45,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.716e+03 2.215e+03 3.177e+03 9.351e+03, threshold=4.431e+03, percent-clipped=11.0
+2023-03-13 06:28:57,674 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:29:03,720 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:29:16,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:29:28,464 INFO [train.py:968] (0/2) Epoch 25, batch 45100, giga_loss[loss=0.2949, simple_loss=0.3656, pruned_loss=0.1121, over 28566.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3684, pruned_loss=0.1202, over 5619607.96 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3604, pruned_loss=0.1134, over 5686517.23 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3693, pruned_loss=0.1207, over 5638878.20 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 06:29:30,053 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-13 06:30:05,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3107, 2.4073, 1.3408, 1.3847], device='cuda:0'), covar=tensor([0.0743, 0.0472, 0.0721, 0.1106], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0563, 0.0399, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 06:30:09,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3308, 1.2896, 3.9344, 3.0547], device='cuda:0'), covar=tensor([0.1790, 0.2891, 0.0491, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0668, 0.0987, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 06:30:10,863 INFO [train.py:968] (0/2) Epoch 25, batch 45150, giga_loss[loss=0.2496, simple_loss=0.3345, pruned_loss=0.08236, over 28557.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3661, pruned_loss=0.1166, over 5574261.52 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3613, pruned_loss=0.1142, over 5614130.11 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3663, pruned_loss=0.1164, over 5652196.11 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 06:30:13,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.659e+03 2.379e+03 3.562e+03 1.255e+04, threshold=4.758e+03, percent-clipped=13.0
+2023-03-13 06:30:30,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:30:38,905 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1139022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:30:47,305 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:30:55,378 INFO [train.py:968] (0/2) Epoch 25, batch 45200, giga_loss[loss=0.3023, simple_loss=0.3711, pruned_loss=0.1167, over 28698.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3623, pruned_loss=0.1134, over 5566407.55 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3617, pruned_loss=0.1146, over 5589816.37 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3622, pruned_loss=0.1131, over 5647360.24 frames. ], batch size: 242, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:31:12,841 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-13 06:31:14,020 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-25.pt
+2023-03-13 06:31:55,577 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:31:57,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:32:15,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.818e+02 1.427e+03 1.728e+03 2.366e+03 6.487e+03, threshold=3.456e+03, percent-clipped=3.0
+2023-03-13 06:32:15,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6361, 1.8298, 1.4927, 1.9092], device='cuda:0'), covar=tensor([0.2814, 0.2960, 0.3273, 0.2562], device='cuda:0'), in_proj_covar=tensor([0.1572, 0.1134, 0.1390, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 06:32:17,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7798, 2.0569, 1.4863, 1.6879], device='cuda:0'), covar=tensor([0.1208, 0.0752, 0.1156, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0451, 0.0521, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 06:32:20,942 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1139102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:32:25,293 INFO [train.py:968] (0/2) Epoch 26, batch 50, giga_loss[loss=0.2466, simple_loss=0.3445, pruned_loss=0.0744, over 28851.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3657, pruned_loss=0.1032, over 1262430.71 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3311, pruned_loss=0.08552, over 198319.34 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3711, pruned_loss=0.1059, over 1102653.71 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:32:34,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3201, 3.0073, 1.4062, 1.5187], device='cuda:0'), covar=tensor([0.1060, 0.0310, 0.0968, 0.1419], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0562, 0.0399, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 06:33:09,643 INFO [train.py:968] (0/2) Epoch 26, batch 100, giga_loss[loss=0.2804, simple_loss=0.3501, pruned_loss=0.1054, over 27552.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3571, pruned_loss=0.09874, over 2244510.17 frames. ], libri_tot_loss[loss=0.2441, simple_loss=0.3259, pruned_loss=0.08116, over 390966.98 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3624, pruned_loss=0.1017, over 1989437.40 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:33:09,881 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:33:13,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:33:24,046 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:33:26,922 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:33:37,270 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1139189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:33:40,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.885e+02 1.206e+03 1.418e+03 1.770e+03 6.136e+03, threshold=2.837e+03, percent-clipped=4.0
+2023-03-13 06:33:50,721 INFO [train.py:968] (0/2) Epoch 26, batch 150, giga_loss[loss=0.2772, simple_loss=0.3446, pruned_loss=0.1048, over 28623.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3458, pruned_loss=0.0942, over 3011747.23 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3315, pruned_loss=0.08408, over 527095.95 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3482, pruned_loss=0.09576, over 2735646.86 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:33:50,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1139207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:34:20,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139241.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:34:29,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6160, 1.8691, 1.5759, 1.4185], device='cuda:0'), covar=tensor([0.2784, 0.2845, 0.3298, 0.2569], device='cuda:0'), in_proj_covar=tensor([0.1574, 0.1137, 0.1393, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 06:34:31,230 INFO [train.py:968] (0/2) Epoch 26, batch 200, giga_loss[loss=0.2144, simple_loss=0.2973, pruned_loss=0.06576, over 29098.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3321, pruned_loss=0.08766, over 3613879.21 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3335, pruned_loss=0.0854, over 606534.83 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3325, pruned_loss=0.08817, over 3362064.81 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:34:52,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:34:59,891 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.178e+02 1.006e+03 1.228e+03 1.554e+03 2.729e+03, threshold=2.457e+03, percent-clipped=0.0
+2023-03-13 06:35:11,252 INFO [train.py:968] (0/2) Epoch 26, batch 250, giga_loss[loss=0.2175, simple_loss=0.2973, pruned_loss=0.06889, over 28610.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3216, pruned_loss=0.08279, over 4080842.95 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.333, pruned_loss=0.08457, over 711947.43 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3212, pruned_loss=0.08298, over 3844097.14 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:35:31,292 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 06:35:54,546 INFO [train.py:968] (0/2) Epoch 26, batch 300, giga_loss[loss=0.1885, simple_loss=0.2733, pruned_loss=0.05182, over 29085.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3135, pruned_loss=0.07922, over 4439912.32 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3338, pruned_loss=0.08441, over 836622.06 frames. ], giga_tot_loss[loss=0.2352, simple_loss=0.3122, pruned_loss=0.07912, over 4218121.06 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:36:16,525 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139384.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:36:21,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139387.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:36:26,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.207e+02 1.044e+03 1.261e+03 1.546e+03 9.669e+03, threshold=2.523e+03, percent-clipped=7.0
+2023-03-13 06:36:30,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:36:38,358 INFO [train.py:968] (0/2) Epoch 26, batch 350, giga_loss[loss=0.2131, simple_loss=0.2831, pruned_loss=0.07153, over 28510.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.308, pruned_loss=0.07735, over 4720603.82 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3346, pruned_loss=0.08547, over 983030.46 frames. ], giga_tot_loss[loss=0.2297, simple_loss=0.3058, pruned_loss=0.07678, over 4510549.28 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:36:46,108 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1139416.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 06:36:52,998 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:36:56,740 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:37:19,515 INFO [train.py:968] (0/2) Epoch 26, batch 400, libri_loss[loss=0.233, simple_loss=0.3204, pruned_loss=0.07279, over 29538.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3034, pruned_loss=0.07543, over 4943672.85 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3351, pruned_loss=0.08541, over 1032126.18 frames. ], giga_tot_loss[loss=0.2254, simple_loss=0.3012, pruned_loss=0.07485, over 4768999.64 frames. ], batch size: 80, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:37:21,836 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1139458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:37:50,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.882e+02 1.173e+03 1.380e+03 1.926e+03 4.933e+03, threshold=2.760e+03, percent-clipped=14.0
+2023-03-13 06:38:00,147 INFO [train.py:968] (0/2) Epoch 26, batch 450, giga_loss[loss=0.2312, simple_loss=0.2891, pruned_loss=0.08658, over 24070.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3011, pruned_loss=0.0742, over 5112961.43 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3355, pruned_loss=0.08564, over 1173056.62 frames. ], giga_tot_loss[loss=0.2224, simple_loss=0.2982, pruned_loss=0.07334, over 4952628.80 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:38:29,426 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:38:31,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:38:42,380 INFO [train.py:968] (0/2) Epoch 26, batch 500, giga_loss[loss=0.2126, simple_loss=0.2878, pruned_loss=0.06866, over 28640.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2987, pruned_loss=0.07305, over 5250380.88 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3361, pruned_loss=0.08611, over 1263600.12 frames. ], giga_tot_loss[loss=0.2198, simple_loss=0.2956, pruned_loss=0.07202, over 5113078.85 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:38:45,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9266, 1.2924, 1.3724, 1.1500], device='cuda:0'), covar=tensor([0.2248, 0.1524, 0.2658, 0.1906], device='cuda:0'), in_proj_covar=tensor([0.0491, 0.0751, 0.0721, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 06:38:55,921 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1139572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:38:59,442 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1139576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:39:16,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.776e+02 1.082e+03 1.371e+03 2.019e+03 5.871e+03, threshold=2.742e+03, percent-clipped=12.0
+2023-03-13 06:39:26,943 INFO [train.py:968] (0/2) Epoch 26, batch 550, giga_loss[loss=0.2089, simple_loss=0.2859, pruned_loss=0.06594, over 28695.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.297, pruned_loss=0.07235, over 5351008.20 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3368, pruned_loss=0.08628, over 1332880.52 frames. ], giga_tot_loss[loss=0.2182, simple_loss=0.2938, pruned_loss=0.07128, over 5234405.56 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:40:12,447 INFO [train.py:968] (0/2) Epoch 26, batch 600, giga_loss[loss=0.2172, simple_loss=0.2892, pruned_loss=0.07261, over 27617.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2958, pruned_loss=0.07204, over 5423623.51 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3387, pruned_loss=0.08717, over 1442246.47 frames. ], giga_tot_loss[loss=0.2165, simple_loss=0.2917, pruned_loss=0.07061, over 5322488.41 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:40:50,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.327e+02 1.073e+03 1.459e+03 1.937e+03 1.169e+04, threshold=2.919e+03, percent-clipped=13.0
+2023-03-13 06:40:59,665 INFO [train.py:968] (0/2) Epoch 26, batch 650, giga_loss[loss=0.2063, simple_loss=0.2834, pruned_loss=0.06461, over 28878.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2935, pruned_loss=0.07105, over 5479953.49 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3384, pruned_loss=0.08719, over 1508378.14 frames. ], giga_tot_loss[loss=0.2145, simple_loss=0.2897, pruned_loss=0.06967, over 5394392.33 frames. ], batch size: 112, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 06:41:43,254 INFO [train.py:968] (0/2) Epoch 26, batch 700, giga_loss[loss=0.2231, simple_loss=0.2919, pruned_loss=0.07718, over 28885.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.292, pruned_loss=0.07067, over 5528969.23 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3391, pruned_loss=0.08773, over 1594638.86 frames. ], giga_tot_loss[loss=0.213, simple_loss=0.2878, pruned_loss=0.06909, over 5454594.58 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 06:42:19,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.954e+02 1.012e+03 1.349e+03 1.842e+03 5.044e+03, threshold=2.698e+03, percent-clipped=7.0
+2023-03-13 06:42:22,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4302, 1.6238, 1.6982, 1.2659], device='cuda:0'), covar=tensor([0.2050, 0.2819, 0.1672, 0.1882], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0720, 0.0985, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 06:42:29,947 INFO [train.py:968] (0/2) Epoch 26, batch 750, giga_loss[loss=0.1808, simple_loss=0.26, pruned_loss=0.05084, over 28819.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2892, pruned_loss=0.0692, over 5574755.94 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3389, pruned_loss=0.08744, over 1659084.84 frames. ], giga_tot_loss[loss=0.2104, simple_loss=0.2853, pruned_loss=0.06777, over 5511209.68 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 06:43:02,007 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1139846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:43:12,332 INFO [train.py:968] (0/2) Epoch 26, batch 800, giga_loss[loss=0.2413, simple_loss=0.3092, pruned_loss=0.08674, over 29001.00 frames. ], tot_loss[loss=0.213, simple_loss=0.288, pruned_loss=0.06894, over 5600199.69 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3389, pruned_loss=0.08727, over 1743505.21 frames. ], giga_tot_loss[loss=0.2095, simple_loss=0.284, pruned_loss=0.0675, over 5544030.48 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:43:31,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2123, 1.1820, 3.8656, 3.1656], device='cuda:0'), covar=tensor([0.1786, 0.3018, 0.0461, 0.1089], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0667, 0.0986, 0.0954], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 06:43:37,263 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1139881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:43:49,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.512e+02 1.148e+03 1.520e+03 2.004e+03 5.846e+03, threshold=3.041e+03, percent-clipped=4.0
+2023-03-13 06:43:49,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5109, 1.8471, 1.5091, 1.3409], device='cuda:0'), covar=tensor([0.2781, 0.2884, 0.3300, 0.2610], device='cuda:0'), in_proj_covar=tensor([0.1577, 0.1135, 0.1395, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 06:44:02,160 INFO [train.py:968] (0/2) Epoch 26, batch 850, giga_loss[loss=0.2421, simple_loss=0.3165, pruned_loss=0.08384, over 29008.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.2997, pruned_loss=0.07542, over 5612521.05 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3401, pruned_loss=0.08813, over 1804852.77 frames. ], giga_tot_loss[loss=0.2217, simple_loss=0.2956, pruned_loss=0.07389, over 5565105.78 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:44:41,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:44:48,291 INFO [train.py:968] (0/2) Epoch 26, batch 900, giga_loss[loss=0.273, simple_loss=0.3505, pruned_loss=0.09769, over 28645.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3143, pruned_loss=0.08275, over 5625032.05 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3403, pruned_loss=0.08816, over 1877089.13 frames. ], giga_tot_loss[loss=0.2367, simple_loss=0.3106, pruned_loss=0.08143, over 5590625.51 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:45:11,303 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1139985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:45:16,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4763, 4.2914, 4.1125, 1.9422], device='cuda:0'), covar=tensor([0.0737, 0.0890, 0.0974, 0.2076], device='cuda:0'), in_proj_covar=tensor([0.1271, 0.1179, 0.0991, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 06:45:22,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.557e+02 1.318e+03 1.799e+03 2.305e+03 4.002e+03, threshold=3.599e+03, percent-clipped=9.0
+2023-03-13 06:45:25,005 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1140000.pt
+2023-03-13 06:45:30,439 INFO [train.py:968] (0/2) Epoch 26, batch 950, giga_loss[loss=0.2898, simple_loss=0.3488, pruned_loss=0.1153, over 23658.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3247, pruned_loss=0.08768, over 5647479.26 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.339, pruned_loss=0.08745, over 1996070.35 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3217, pruned_loss=0.08683, over 5612934.39 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:45:34,344 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:45:40,181 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:45:56,362 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:46:11,873 INFO [train.py:968] (0/2) Epoch 26, batch 1000, giga_loss[loss=0.277, simple_loss=0.3544, pruned_loss=0.09979, over 28738.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3314, pruned_loss=0.09024, over 5660427.06 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3394, pruned_loss=0.088, over 2070114.18 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3288, pruned_loss=0.08945, over 5631752.59 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:46:15,059 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:46:40,862 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:46:43,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.321e+02 1.262e+03 1.546e+03 2.045e+03 3.977e+03, threshold=3.091e+03, percent-clipped=2.0
+2023-03-13 06:46:44,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:46:54,074 INFO [train.py:968] (0/2) Epoch 26, batch 1050, giga_loss[loss=0.2609, simple_loss=0.3359, pruned_loss=0.09301, over 28575.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3349, pruned_loss=0.09065, over 5661327.50 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3393, pruned_loss=0.08819, over 2108201.04 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3328, pruned_loss=0.09, over 5636897.19 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:47:10,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5292, 1.9176, 1.6719, 1.7133], device='cuda:0'), covar=tensor([0.2338, 0.2552, 0.2453, 0.2302], device='cuda:0'), in_proj_covar=tensor([0.0493, 0.0754, 0.0724, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 06:47:15,283 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:47:38,096 INFO [train.py:968] (0/2) Epoch 26, batch 1100, giga_loss[loss=0.2434, simple_loss=0.3273, pruned_loss=0.07976, over 28958.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3353, pruned_loss=0.08959, over 5677573.32 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3391, pruned_loss=0.08792, over 2219287.73 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3336, pruned_loss=0.08922, over 5652616.65 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:47:38,623 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 06:47:49,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5523, 1.7793, 1.1955, 1.2741], device='cuda:0'), covar=tensor([0.1172, 0.0703, 0.1259, 0.1330], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0450, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 06:48:12,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.138e+02 1.252e+03 1.556e+03 2.173e+03 3.965e+03, threshold=3.111e+03, percent-clipped=4.0
+2023-03-13 06:48:22,879 INFO [train.py:968] (0/2) Epoch 26, batch 1150, giga_loss[loss=0.2761, simple_loss=0.3353, pruned_loss=0.1085, over 23705.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3381, pruned_loss=0.09173, over 5680090.52 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3389, pruned_loss=0.08756, over 2288753.97 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3368, pruned_loss=0.09163, over 5659564.34 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:48:36,592 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:49:06,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:49:07,308 INFO [train.py:968] (0/2) Epoch 26, batch 1200, giga_loss[loss=0.3167, simple_loss=0.38, pruned_loss=0.1267, over 27918.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3423, pruned_loss=0.0948, over 5678190.14 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3389, pruned_loss=0.08749, over 2342523.32 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3412, pruned_loss=0.09482, over 5659301.71 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:49:15,960 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:49:41,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.768e+02 1.324e+03 1.612e+03 2.143e+03 3.871e+03, threshold=3.224e+03, percent-clipped=7.0
+2023-03-13 06:49:49,665 INFO [train.py:968] (0/2) Epoch 26, batch 1250, giga_loss[loss=0.2663, simple_loss=0.3488, pruned_loss=0.09186, over 28776.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3457, pruned_loss=0.09695, over 5672593.26 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3396, pruned_loss=0.08806, over 2383538.54 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3447, pruned_loss=0.09686, over 5665755.56 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:50:35,391 INFO [train.py:968] (0/2) Epoch 26, batch 1300, giga_loss[loss=0.2504, simple_loss=0.3416, pruned_loss=0.07961, over 28735.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3477, pruned_loss=0.09698, over 5680581.91 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.34, pruned_loss=0.08832, over 2435254.36 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3469, pruned_loss=0.09694, over 5672899.16 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:50:35,564 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:50:38,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:50:41,550 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:50:44,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:50:56,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:06,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:07,578 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:07,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.668e+02 1.238e+03 1.523e+03 1.921e+03 3.986e+03, threshold=3.046e+03, percent-clipped=3.0
+2023-03-13 06:51:09,393 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:11,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:14,405 INFO [train.py:968] (0/2) Epoch 26, batch 1350, giga_loss[loss=0.2808, simple_loss=0.3642, pruned_loss=0.09871, over 28174.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3488, pruned_loss=0.09649, over 5682148.10 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.34, pruned_loss=0.08812, over 2559094.33 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3485, pruned_loss=0.09686, over 5679481.27 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:51:20,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:32,861 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:36,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:52,672 INFO [train.py:968] (0/2) Epoch 26, batch 1400, giga_loss[loss=0.273, simple_loss=0.3477, pruned_loss=0.09913, over 27928.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3498, pruned_loss=0.09632, over 5686572.95 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3412, pruned_loss=0.08879, over 2622677.22 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3492, pruned_loss=0.09649, over 5682918.49 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:52:22,736 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-13 06:52:26,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.680e+02 1.237e+03 1.571e+03 2.212e+03 6.065e+03, threshold=3.143e+03, percent-clipped=9.0
+2023-03-13 06:52:31,744 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:52:34,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:52:35,026 INFO [train.py:968] (0/2) Epoch 26, batch 1450, libri_loss[loss=0.2575, simple_loss=0.3408, pruned_loss=0.08712, over 29546.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3489, pruned_loss=0.09477, over 5698517.16 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.341, pruned_loss=0.08865, over 2704023.60 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3487, pruned_loss=0.09512, over 5690589.64 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:52:49,981 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:52:53,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:52:57,087 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:52:58,588 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:01,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:13,238 INFO [train.py:968] (0/2) Epoch 26, batch 1500, giga_loss[loss=0.224, simple_loss=0.3163, pruned_loss=0.06585, over 28929.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3474, pruned_loss=0.09307, over 5701449.04 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.341, pruned_loss=0.08876, over 2783070.97 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3474, pruned_loss=0.09344, over 5690790.78 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:53:14,287 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:15,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:16,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:23,343 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:31,399 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:33,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:40,352 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:46,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.553e+02 1.171e+03 1.379e+03 1.816e+03 3.162e+03, threshold=2.758e+03, percent-clipped=1.0
+2023-03-13 06:53:55,082 INFO [train.py:968] (0/2) Epoch 26, batch 1550, giga_loss[loss=0.2781, simple_loss=0.3586, pruned_loss=0.09879, over 29037.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3464, pruned_loss=0.09254, over 5700746.20 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3418, pruned_loss=0.08906, over 2809004.38 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3461, pruned_loss=0.09272, over 5695333.02 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:53:57,686 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:54:25,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:54:25,826 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 06:54:38,313 INFO [train.py:968] (0/2) Epoch 26, batch 1600, giga_loss[loss=0.3244, simple_loss=0.3798, pruned_loss=0.1345, over 28475.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3491, pruned_loss=0.09697, over 5706114.46 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3412, pruned_loss=0.08867, over 2854804.33 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3492, pruned_loss=0.09739, over 5699498.15 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:55:15,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.027e+02 1.345e+03 1.716e+03 2.161e+03 4.630e+03, threshold=3.432e+03, percent-clipped=10.0
+2023-03-13 06:55:24,119 INFO [train.py:968] (0/2) Epoch 26, batch 1650, giga_loss[loss=0.2892, simple_loss=0.359, pruned_loss=0.1097, over 28648.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3506, pruned_loss=0.1, over 5703538.22 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3411, pruned_loss=0.08848, over 2973662.40 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3511, pruned_loss=0.1008, over 5691931.08 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:55:32,858 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-13 06:55:44,338 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140730.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:55:46,481 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:55:52,279 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-13 06:56:06,241 INFO [train.py:968] (0/2) Epoch 26, batch 1700, giga_loss[loss=0.2403, simple_loss=0.3199, pruned_loss=0.08039, over 28991.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3501, pruned_loss=0.1013, over 5697536.99 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3407, pruned_loss=0.08831, over 3027923.38 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3508, pruned_loss=0.1022, over 5688085.25 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:56:31,687 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:56:33,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:56:33,610 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.41 vs. limit=5.0
+2023-03-13 06:56:41,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3253, 4.1354, 3.9515, 1.9773], device='cuda:0'), covar=tensor([0.0637, 0.0795, 0.0730, 0.2183], device='cuda:0'), in_proj_covar=tensor([0.1257, 0.1168, 0.0980, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 06:56:42,808 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.707e+02 1.373e+03 1.790e+03 2.417e+03 5.282e+03, threshold=3.580e+03, percent-clipped=7.0
+2023-03-13 06:56:50,918 INFO [train.py:968] (0/2) Epoch 26, batch 1750, giga_loss[loss=0.2656, simple_loss=0.3396, pruned_loss=0.09581, over 28521.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3475, pruned_loss=0.09998, over 5707157.19 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3411, pruned_loss=0.0884, over 3084133.70 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3481, pruned_loss=0.1009, over 5696927.51 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:56:59,709 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:57:11,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2550, 1.6852, 1.2619, 0.7883], device='cuda:0'), covar=tensor([0.5247, 0.2612, 0.2991, 0.6034], device='cuda:0'), in_proj_covar=tensor([0.1802, 0.1693, 0.1630, 0.1461], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 06:57:16,437 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:57:33,051 INFO [train.py:968] (0/2) Epoch 26, batch 1800, giga_loss[loss=0.2384, simple_loss=0.3258, pruned_loss=0.07545, over 28988.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3462, pruned_loss=0.09907, over 5709918.45 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3409, pruned_loss=0.08824, over 3126069.92 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3469, pruned_loss=0.1001, over 5699369.12 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:57:47,251 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:57:50,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:57:50,960 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7530, 2.1199, 1.6776, 2.1046], device='cuda:0'), covar=tensor([0.2916, 0.2897, 0.3332, 0.2564], device='cuda:0'), in_proj_covar=tensor([0.1570, 0.1130, 0.1387, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 06:58:07,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.555e+02 1.230e+03 1.517e+03 2.125e+03 5.801e+03, threshold=3.033e+03, percent-clipped=7.0
+2023-03-13 06:58:14,387 INFO [train.py:968] (0/2) Epoch 26, batch 1850, giga_loss[loss=0.2523, simple_loss=0.3281, pruned_loss=0.08826, over 28678.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3448, pruned_loss=0.09731, over 5718658.77 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.34, pruned_loss=0.08774, over 3205241.59 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3459, pruned_loss=0.09864, over 5707901.42 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:58:14,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:58:55,026 INFO [train.py:968] (0/2) Epoch 26, batch 1900, libri_loss[loss=0.283, simple_loss=0.3658, pruned_loss=0.1001, over 29635.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3425, pruned_loss=0.0951, over 5714184.61 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3401, pruned_loss=0.0876, over 3361142.90 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3435, pruned_loss=0.09674, over 5696163.04 frames. ], batch size: 91, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:59:01,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2809, 1.3472, 1.2076, 1.2539], device='cuda:0'), covar=tensor([0.2222, 0.2137, 0.2076, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.2019, 0.1961, 0.1883, 0.2023], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 06:59:14,128 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2394, 1.5162, 1.5301, 1.1104], device='cuda:0'), covar=tensor([0.1847, 0.2735, 0.1493, 0.1769], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0716, 0.0978, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:0')
+2023-03-13 06:59:14,844 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140977.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:59:34,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.871e+02 1.247e+03 1.487e+03 2.092e+03 5.777e+03, threshold=2.974e+03, percent-clipped=6.0
+2023-03-13 06:59:40,975 INFO [train.py:968] (0/2) Epoch 26, batch 1950, giga_loss[loss=0.2665, simple_loss=0.3386, pruned_loss=0.09726, over 28717.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3393, pruned_loss=0.09329, over 5707369.90 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3407, pruned_loss=0.08785, over 3411291.63 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3398, pruned_loss=0.09457, over 5689822.88 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:00:29,882 INFO [train.py:968] (0/2) Epoch 26, batch 2000, giga_loss[loss=0.2219, simple_loss=0.3044, pruned_loss=0.06966, over 28558.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3338, pruned_loss=0.09066, over 5694719.28 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3406, pruned_loss=0.0877, over 3444702.76 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3342, pruned_loss=0.09181, over 5681457.76 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:01:05,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.082e+02 1.037e+03 1.321e+03 1.910e+03 6.600e+03, threshold=2.643e+03, percent-clipped=4.0
+2023-03-13 07:01:14,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1141105.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:01:17,441 INFO [train.py:968] (0/2) Epoch 26, batch 2050, giga_loss[loss=0.2313, simple_loss=0.3017, pruned_loss=0.08044, over 28642.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3291, pruned_loss=0.0885, over 5685673.78 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3406, pruned_loss=0.08758, over 3514477.23 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3292, pruned_loss=0.08953, over 5672365.28 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:01:57,232 INFO [train.py:968] (0/2) Epoch 26, batch 2100, giga_loss[loss=0.2238, simple_loss=0.3111, pruned_loss=0.06824, over 28494.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3305, pruned_loss=0.08902, over 5697853.23 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.341, pruned_loss=0.08782, over 3585164.79 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.33, pruned_loss=0.08971, over 5681437.85 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:02:07,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5852, 4.8903, 1.8995, 1.9320], device='cuda:0'), covar=tensor([0.1125, 0.0251, 0.0923, 0.1400], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0560, 0.0400, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 07:02:09,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3075, 3.1580, 2.9952, 1.2849], device='cuda:0'), covar=tensor([0.0899, 0.1061, 0.0819, 0.2368], device='cuda:0'), in_proj_covar=tensor([0.1252, 0.1166, 0.0978, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 07:02:22,318 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1141188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:02:31,070 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.272e+02 1.213e+03 1.502e+03 2.128e+03 5.472e+03, threshold=3.004e+03, percent-clipped=15.0
+2023-03-13 07:02:36,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1307, 2.2093, 2.0718, 1.9356], device='cuda:0'), covar=tensor([0.2090, 0.2542, 0.2486, 0.2522], device='cuda:0'), in_proj_covar=tensor([0.0496, 0.0755, 0.0726, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 07:02:37,364 INFO [train.py:968] (0/2) Epoch 26, batch 2150, giga_loss[loss=0.2704, simple_loss=0.3343, pruned_loss=0.1032, over 28851.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3306, pruned_loss=0.08863, over 5704225.74 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3408, pruned_loss=0.08762, over 3653010.79 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.33, pruned_loss=0.08929, over 5686538.11 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:02:41,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1141212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:03:09,709 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1141248.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:03:12,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1141251.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:03:16,636 INFO [train.py:968] (0/2) Epoch 26, batch 2200, giga_loss[loss=0.2596, simple_loss=0.3395, pruned_loss=0.0898, over 28682.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3297, pruned_loss=0.08793, over 5698811.22 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3415, pruned_loss=0.08786, over 3698058.80 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3286, pruned_loss=0.08833, over 5690158.13 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:03:24,421 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-13 07:03:35,530 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1141280.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 07:03:50,595 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.329e+02 1.110e+03 1.453e+03 1.844e+03 3.636e+03, threshold=2.906e+03, percent-clipped=5.0
+2023-03-13 07:03:55,467 INFO [train.py:968] (0/2) Epoch 26, batch 2250, giga_loss[loss=0.2306, simple_loss=0.3127, pruned_loss=0.07421, over 29000.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3291, pruned_loss=0.08789, over 5703362.58 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3417, pruned_loss=0.0878, over 3783735.65 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3276, pruned_loss=0.08828, over 5697515.78 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:04:13,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5105, 1.7651, 1.4437, 1.4440], device='cuda:0'), covar=tensor([0.2769, 0.2716, 0.3059, 0.2551], device='cuda:0'), in_proj_covar=tensor([0.1572, 0.1130, 0.1387, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 07:04:36,222 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1141352.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:04:38,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1141355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:04:39,180 INFO [train.py:968] (0/2) Epoch 26, batch 2300, giga_loss[loss=0.263, simple_loss=0.3364, pruned_loss=0.09475, over 28590.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3261, pruned_loss=0.08667, over 5704767.93 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3418, pruned_loss=0.08778, over 3794339.97 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3248, pruned_loss=0.08698, over 5699214.79 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:04:40,206 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1141358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:04:45,271 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1141364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:05:03,671 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1141387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:05:13,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.353e+02 1.093e+03 1.272e+03 1.689e+03 3.349e+03, threshold=2.545e+03, percent-clipped=4.0
+2023-03-13 07:05:20,174 INFO [train.py:968] (0/2) Epoch 26, batch 2350, giga_loss[loss=0.2496, simple_loss=0.3174, pruned_loss=0.09084, over 28885.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3242, pruned_loss=0.08568, over 5714532.88 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3418, pruned_loss=0.08752, over 3836009.93 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3229, pruned_loss=0.08605, over 5707061.41 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:06:00,846 INFO [train.py:968] (0/2) Epoch 26, batch 2400, giga_loss[loss=0.2369, simple_loss=0.3068, pruned_loss=0.08346, over 28520.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3229, pruned_loss=0.08498, over 5725047.73 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3426, pruned_loss=0.08749, over 3906059.81 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3207, pruned_loss=0.0852, over 5714337.21 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:06:29,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2088, 0.7684, 0.8457, 1.3911], device='cuda:0'), covar=tensor([0.0768, 0.0404, 0.0385, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 07:06:30,780 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1141495.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:06:31,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1475, 2.3242, 2.4446, 1.9806], device='cuda:0'), covar=tensor([0.2679, 0.2116, 0.1830, 0.2487], device='cuda:0'), in_proj_covar=tensor([0.2013, 0.1948, 0.1876, 0.2021], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 07:06:32,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1141498.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:06:33,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.968e+02 1.079e+03 1.334e+03 1.797e+03 7.702e+03, threshold=2.667e+03, percent-clipped=11.0
+2023-03-13 07:06:38,251 INFO [train.py:968] (0/2) Epoch 26, batch 2450, giga_loss[loss=0.2335, simple_loss=0.313, pruned_loss=0.07706, over 29011.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3212, pruned_loss=0.08425, over 5732011.66 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.343, pruned_loss=0.08764, over 3944980.74 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3188, pruned_loss=0.08429, over 5721006.54 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:06:53,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1141527.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:07:11,707 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1141550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:07:16,114 INFO [train.py:968] (0/2) Epoch 26, batch 2500, giga_loss[loss=0.2492, simple_loss=0.3267, pruned_loss=0.08583, over 28742.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3208, pruned_loss=0.08403, over 5722307.28 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3438, pruned_loss=0.08782, over 4020148.76 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3174, pruned_loss=0.08378, over 5715891.97 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:07:20,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1141563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:07:49,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.237e+02 1.199e+03 1.512e+03 2.158e+03 9.137e+03, threshold=3.024e+03, percent-clipped=15.0
+2023-03-13 07:07:54,045 INFO [train.py:968] (0/2) Epoch 26, batch 2550, giga_loss[loss=0.2136, simple_loss=0.2948, pruned_loss=0.06618, over 28722.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3199, pruned_loss=0.08348, over 5725974.79 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3436, pruned_loss=0.08747, over 4093971.34 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3164, pruned_loss=0.08335, over 5714264.56 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:08:34,114 INFO [train.py:968] (0/2) Epoch 26, batch 2600, giga_loss[loss=0.2407, simple_loss=0.3282, pruned_loss=0.07663, over 29008.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3176, pruned_loss=0.08212, over 5729047.28 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3432, pruned_loss=0.0872, over 4112203.70 frames. ], giga_tot_loss[loss=0.2395, simple_loss=0.3148, pruned_loss=0.08215, over 5717939.66 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:08:39,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2546, 1.5599, 1.5203, 1.1298], device='cuda:0'), covar=tensor([0.1875, 0.2694, 0.1509, 0.1708], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0718, 0.0981, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 07:08:54,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8407, 1.1397, 2.8673, 2.8477], device='cuda:0'), covar=tensor([0.1781, 0.2674, 0.0581, 0.0951], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0662, 0.0975, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:09:09,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.323e+02 1.079e+03 1.389e+03 1.721e+03 3.246e+03, threshold=2.778e+03, percent-clipped=1.0
+2023-03-13 07:09:11,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9296, 1.2880, 1.0641, 0.1994], device='cuda:0'), covar=tensor([0.3649, 0.2591, 0.3865, 0.5441], device='cuda:0'), in_proj_covar=tensor([0.1800, 0.1689, 0.1627, 0.1463], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 07:09:13,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1141706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:09:13,710 INFO [train.py:968] (0/2) Epoch 26, batch 2650, libri_loss[loss=0.2456, simple_loss=0.3392, pruned_loss=0.07594, over 29516.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.317, pruned_loss=0.08195, over 5727460.38 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3432, pruned_loss=0.08689, over 4155538.14 frames. ], giga_tot_loss[loss=0.2392, simple_loss=0.3142, pruned_loss=0.08207, over 5715665.00 frames. ], batch size: 82, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:09:15,374 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1141709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:09:22,464 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6447, 2.7200, 2.5891, 2.3449], device='cuda:0'), covar=tensor([0.2102, 0.2439, 0.2227, 0.2502], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0759, 0.0728, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 07:09:39,632 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1141738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:09:40,264 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1141739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:09:56,782 INFO [train.py:968] (0/2) Epoch 26, batch 2700, giga_loss[loss=0.2302, simple_loss=0.3074, pruned_loss=0.07652, over 28861.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3191, pruned_loss=0.08336, over 5715945.07 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.343, pruned_loss=0.08669, over 4188783.36 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3164, pruned_loss=0.08351, over 5711028.92 frames. ], batch size: 66, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:10:21,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9549, 2.0628, 2.1153, 1.7184], device='cuda:0'), covar=tensor([0.1777, 0.2233, 0.1417, 0.1642], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0716, 0.0981, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 07:10:34,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.810e+02 1.198e+03 1.461e+03 1.925e+03 6.396e+03, threshold=2.923e+03, percent-clipped=10.0
+2023-03-13 07:10:39,256 INFO [train.py:968] (0/2) Epoch 26, batch 2750, giga_loss[loss=0.2663, simple_loss=0.3378, pruned_loss=0.09739, over 28804.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3247, pruned_loss=0.08708, over 5712236.01 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3431, pruned_loss=0.08667, over 4222881.95 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3221, pruned_loss=0.08719, over 5704919.10 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:11:27,117 INFO [train.py:968] (0/2) Epoch 26, batch 2800, giga_loss[loss=0.2675, simple_loss=0.3356, pruned_loss=0.09974, over 28541.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3323, pruned_loss=0.09259, over 5701531.32 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3428, pruned_loss=0.08652, over 4231118.54 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3305, pruned_loss=0.09278, over 5695120.11 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:11:35,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8144, 1.0574, 2.8499, 2.7805], device='cuda:0'), covar=tensor([0.2082, 0.2946, 0.1025, 0.1416], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0662, 0.0974, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:11:50,960 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1141882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:11:53,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1141885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:12:04,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.879e+02 1.437e+03 1.787e+03 2.265e+03 6.934e+03, threshold=3.574e+03, percent-clipped=11.0
+2023-03-13 07:12:10,664 INFO [train.py:968] (0/2) Epoch 26, batch 2850, giga_loss[loss=0.2906, simple_loss=0.37, pruned_loss=0.1056, over 28871.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3384, pruned_loss=0.09569, over 5690769.64 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3424, pruned_loss=0.0865, over 4268929.53 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3371, pruned_loss=0.09605, over 5684365.14 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:12:17,848 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1141914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:12:27,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1141925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:12:57,837 INFO [train.py:968] (0/2) Epoch 26, batch 2900, giga_loss[loss=0.2933, simple_loss=0.3677, pruned_loss=0.1095, over 28865.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3442, pruned_loss=0.09868, over 5670581.92 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3421, pruned_loss=0.08628, over 4308837.33 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3433, pruned_loss=0.09933, over 5661364.15 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:13:33,905 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1142000.pt
+2023-03-13 07:13:34,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.108e+02 1.223e+03 1.492e+03 1.827e+03 3.920e+03, threshold=2.984e+03, percent-clipped=1.0
+2023-03-13 07:13:39,758 INFO [train.py:968] (0/2) Epoch 26, batch 2950, giga_loss[loss=0.2811, simple_loss=0.358, pruned_loss=0.1021, over 28705.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3481, pruned_loss=0.09952, over 5692825.58 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3421, pruned_loss=0.08621, over 4356175.46 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3474, pruned_loss=0.1004, over 5679455.29 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:14:27,128 INFO [train.py:968] (0/2) Epoch 26, batch 3000, giga_loss[loss=0.2783, simple_loss=0.3573, pruned_loss=0.09961, over 28875.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3526, pruned_loss=0.1025, over 5685934.73 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.342, pruned_loss=0.08626, over 4386903.71 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3523, pruned_loss=0.1035, over 5671688.08 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:14:27,132 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 07:14:35,948 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2114, simple_loss=0.3175, pruned_loss=0.0527, over 944034.00 frames. 
+2023-03-13 07:14:35,949 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 07:14:45,264 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1142068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:14:47,056 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1142071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:15:12,218 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1142100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:15:12,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.816e+02 1.401e+03 1.681e+03 2.269e+03 7.431e+03, threshold=3.363e+03, percent-clipped=17.0
+2023-03-13 07:15:16,707 INFO [train.py:968] (0/2) Epoch 26, batch 3050, giga_loss[loss=0.2868, simple_loss=0.3583, pruned_loss=0.1077, over 28694.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3476, pruned_loss=0.09872, over 5691168.80 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3417, pruned_loss=0.08618, over 4407910.27 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3477, pruned_loss=0.09972, over 5678478.24 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:15:52,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4429, 2.1760, 1.5940, 0.6334], device='cuda:0'), covar=tensor([0.7100, 0.3290, 0.4953, 0.7646], device='cuda:0'), in_proj_covar=tensor([0.1807, 0.1694, 0.1636, 0.1470], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 07:15:59,215 INFO [train.py:968] (0/2) Epoch 26, batch 3100, giga_loss[loss=0.2566, simple_loss=0.3362, pruned_loss=0.08852, over 28861.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3453, pruned_loss=0.09679, over 5689556.24 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3415, pruned_loss=0.08613, over 4437641.89 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3455, pruned_loss=0.09782, over 5675928.18 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:16:18,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7837, 1.0297, 2.8496, 2.7134], device='cuda:0'), covar=tensor([0.1856, 0.2864, 0.0606, 0.1159], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0662, 0.0973, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:16:20,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1273, 3.9516, 3.7642, 2.0317], device='cuda:0'), covar=tensor([0.0629, 0.0804, 0.0750, 0.2196], device='cuda:0'), in_proj_covar=tensor([0.1262, 0.1172, 0.0983, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 07:16:26,564 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1137, 1.3027, 1.2088, 1.0887], device='cuda:0'), covar=tensor([0.2635, 0.2587, 0.1881, 0.2376], device='cuda:0'), in_proj_covar=tensor([0.2018, 0.1956, 0.1886, 0.2030], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 07:16:36,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.267e+02 1.251e+03 1.496e+03 2.021e+03 7.294e+03, threshold=2.991e+03, percent-clipped=2.0
+2023-03-13 07:16:37,554 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-13 07:16:41,764 INFO [train.py:968] (0/2) Epoch 26, batch 3150, giga_loss[loss=0.2712, simple_loss=0.354, pruned_loss=0.09425, over 28537.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3444, pruned_loss=0.09581, over 5678974.97 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3413, pruned_loss=0.08605, over 4457071.32 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3449, pruned_loss=0.09687, over 5672763.38 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:17:23,847 INFO [train.py:968] (0/2) Epoch 26, batch 3200, giga_loss[loss=0.3017, simple_loss=0.3714, pruned_loss=0.116, over 27621.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3444, pruned_loss=0.09547, over 5667999.15 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3411, pruned_loss=0.08606, over 4479251.59 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3449, pruned_loss=0.09651, over 5670477.55 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:17:27,052 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1142260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:17:41,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2116, 1.3420, 3.6843, 3.1784], device='cuda:0'), covar=tensor([0.1722, 0.2743, 0.0459, 0.0900], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0661, 0.0973, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:17:42,695 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6882, 1.7003, 1.8864, 1.4672], device='cuda:0'), covar=tensor([0.1953, 0.2527, 0.1557, 0.1838], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0715, 0.0977, 0.0873], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:0')
+2023-03-13 07:17:55,469 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2938, 1.6539, 1.4608, 1.4859], device='cuda:0'), covar=tensor([0.0826, 0.0335, 0.0337, 0.0919], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 07:17:59,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.663e+02 1.346e+03 1.710e+03 2.204e+03 5.788e+03, threshold=3.419e+03, percent-clipped=8.0
+2023-03-13 07:17:59,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5087, 1.1226, 4.2967, 3.5793], device='cuda:0'), covar=tensor([0.1571, 0.2781, 0.0394, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0660, 0.0972, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:18:03,601 INFO [train.py:968] (0/2) Epoch 26, batch 3250, giga_loss[loss=0.3033, simple_loss=0.374, pruned_loss=0.1163, over 28686.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3472, pruned_loss=0.09742, over 5681101.67 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.341, pruned_loss=0.08614, over 4512568.71 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3478, pruned_loss=0.09846, over 5679395.04 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:18:14,186 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-13 07:18:17,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2319, 4.0605, 3.8317, 1.8405], device='cuda:0'), covar=tensor([0.0631, 0.0775, 0.0727, 0.2118], device='cuda:0'), in_proj_covar=tensor([0.1260, 0.1171, 0.0983, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 07:18:45,630 INFO [train.py:968] (0/2) Epoch 26, batch 3300, giga_loss[loss=0.3144, simple_loss=0.3795, pruned_loss=0.1247, over 27645.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3498, pruned_loss=0.09948, over 5681196.36 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3411, pruned_loss=0.08624, over 4550831.60 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3505, pruned_loss=0.1006, over 5676217.46 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:19:20,529 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.998e+02 1.493e+03 1.783e+03 2.584e+03 7.703e+03, threshold=3.566e+03, percent-clipped=11.0
+2023-03-13 07:19:24,766 INFO [train.py:968] (0/2) Epoch 26, batch 3350, giga_loss[loss=0.3019, simple_loss=0.3761, pruned_loss=0.1139, over 28851.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.351, pruned_loss=0.1004, over 5694105.46 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3413, pruned_loss=0.08629, over 4589179.63 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3516, pruned_loss=0.1016, over 5685801.47 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:19:55,821 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.97 vs. limit=2.0
+2023-03-13 07:20:02,063 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-13 07:20:06,567 INFO [train.py:968] (0/2) Epoch 26, batch 3400, libri_loss[loss=0.2354, simple_loss=0.311, pruned_loss=0.07989, over 29657.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3514, pruned_loss=0.1012, over 5689203.00 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3411, pruned_loss=0.08611, over 4626156.36 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3524, pruned_loss=0.1028, over 5678388.40 frames. ], batch size: 69, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:20:41,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.583e+02 1.257e+03 1.571e+03 1.925e+03 4.252e+03, threshold=3.142e+03, percent-clipped=1.0
+2023-03-13 07:20:47,421 INFO [train.py:968] (0/2) Epoch 26, batch 3450, giga_loss[loss=0.2673, simple_loss=0.3491, pruned_loss=0.09268, over 28877.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3518, pruned_loss=0.1013, over 5678986.62 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3413, pruned_loss=0.08623, over 4649018.43 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3526, pruned_loss=0.1028, over 5673582.47 frames. ], batch size: 112, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:21:27,047 INFO [train.py:968] (0/2) Epoch 26, batch 3500, giga_loss[loss=0.2706, simple_loss=0.3553, pruned_loss=0.09293, over 29012.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3515, pruned_loss=0.09982, over 5690457.19 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3415, pruned_loss=0.08622, over 4666874.31 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3522, pruned_loss=0.1012, over 5684099.56 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:21:43,919 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8486, 2.0766, 1.8443, 1.7783], device='cuda:0'), covar=tensor([0.2275, 0.2620, 0.2459, 0.2399], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0758, 0.0729, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 07:22:02,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2232, 1.2333, 1.2603, 1.4422], device='cuda:0'), covar=tensor([0.0790, 0.0398, 0.0353, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 07:22:03,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.303e+02 1.268e+03 1.612e+03 2.113e+03 5.522e+03, threshold=3.224e+03, percent-clipped=7.0
+2023-03-13 07:22:04,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-13 07:22:07,254 INFO [train.py:968] (0/2) Epoch 26, batch 3550, giga_loss[loss=0.2711, simple_loss=0.3547, pruned_loss=0.09371, over 28899.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3527, pruned_loss=0.0999, over 5690920.19 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3416, pruned_loss=0.08611, over 4698439.75 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3534, pruned_loss=0.1015, over 5684884.69 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:22:27,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4261, 1.6553, 1.3455, 1.3788], device='cuda:0'), covar=tensor([0.2829, 0.2747, 0.3181, 0.2321], device='cuda:0'), in_proj_covar=tensor([0.1575, 0.1135, 0.1392, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 07:22:30,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1142635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:22:49,337 INFO [train.py:968] (0/2) Epoch 26, batch 3600, libri_loss[loss=0.2669, simple_loss=0.3421, pruned_loss=0.09581, over 29544.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3516, pruned_loss=0.09913, over 5696952.93 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3418, pruned_loss=0.08639, over 4710450.41 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3522, pruned_loss=0.1003, over 5690318.65 frames. ], batch size: 79, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:23:25,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.266e+02 1.159e+03 1.562e+03 2.154e+03 5.200e+03, threshold=3.125e+03, percent-clipped=8.0
+2023-03-13 07:23:29,554 INFO [train.py:968] (0/2) Epoch 26, batch 3650, libri_loss[loss=0.2609, simple_loss=0.3477, pruned_loss=0.08705, over 29534.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3491, pruned_loss=0.09806, over 5693012.72 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3418, pruned_loss=0.08633, over 4740360.50 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3498, pruned_loss=0.09932, over 5682361.42 frames. ], batch size: 84, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:23:55,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4570, 1.7114, 1.4079, 1.4715], device='cuda:0'), covar=tensor([0.2660, 0.2703, 0.3035, 0.2374], device='cuda:0'), in_proj_covar=tensor([0.1574, 0.1134, 0.1389, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 07:24:06,204 INFO [train.py:968] (0/2) Epoch 26, batch 3700, giga_loss[loss=0.245, simple_loss=0.3278, pruned_loss=0.08109, over 28908.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3465, pruned_loss=0.09633, over 5709281.72 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3416, pruned_loss=0.08639, over 4793525.40 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3474, pruned_loss=0.09781, over 5694697.21 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:24:22,155 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1142778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:24:24,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1142781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:24:39,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.403e+02 1.130e+03 1.314e+03 1.631e+03 3.697e+03, threshold=2.627e+03, percent-clipped=3.0
+2023-03-13 07:24:43,324 INFO [train.py:968] (0/2) Epoch 26, batch 3750, giga_loss[loss=0.3082, simple_loss=0.3723, pruned_loss=0.122, over 28264.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3439, pruned_loss=0.09488, over 5711847.96 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3412, pruned_loss=0.0861, over 4814693.95 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.345, pruned_loss=0.09645, over 5697536.93 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:24:45,656 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1142810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:25:01,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1948, 1.1736, 3.5335, 3.1333], device='cuda:0'), covar=tensor([0.1667, 0.2952, 0.0440, 0.1721], device='cuda:0'), in_proj_covar=tensor([0.0786, 0.0663, 0.0975, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:25:28,072 INFO [train.py:968] (0/2) Epoch 26, batch 3800, giga_loss[loss=0.2757, simple_loss=0.3518, pruned_loss=0.0998, over 27989.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3457, pruned_loss=0.09651, over 5706827.97 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.341, pruned_loss=0.08599, over 4825396.11 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3468, pruned_loss=0.09792, over 5693961.82 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:26:01,883 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 07:26:02,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.214e+02 1.192e+03 1.485e+03 1.782e+03 7.303e+03, threshold=2.969e+03, percent-clipped=10.0
+2023-03-13 07:26:06,507 INFO [train.py:968] (0/2) Epoch 26, batch 3850, giga_loss[loss=0.2784, simple_loss=0.3629, pruned_loss=0.09696, over 28951.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3451, pruned_loss=0.09542, over 5709504.19 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.341, pruned_loss=0.08594, over 4851975.60 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.346, pruned_loss=0.09682, over 5694899.19 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:26:15,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8736, 5.7062, 5.3980, 3.1981], device='cuda:0'), covar=tensor([0.0413, 0.0572, 0.0663, 0.1508], device='cuda:0'), in_proj_covar=tensor([0.1270, 0.1174, 0.0989, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 07:26:45,464 INFO [train.py:968] (0/2) Epoch 26, batch 3900, giga_loss[loss=0.2531, simple_loss=0.3356, pruned_loss=0.08529, over 28601.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3438, pruned_loss=0.09406, over 5719399.61 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3408, pruned_loss=0.08598, over 4883781.53 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3448, pruned_loss=0.0954, over 5702244.41 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:27:09,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1142986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:27:16,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1394, 1.4388, 1.1739, 0.6119], device='cuda:0'), covar=tensor([0.3095, 0.2006, 0.2696, 0.5240], device='cuda:0'), in_proj_covar=tensor([0.1792, 0.1669, 0.1614, 0.1457], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 07:27:23,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.134e+02 1.103e+03 1.255e+03 1.564e+03 3.648e+03, threshold=2.511e+03, percent-clipped=1.0
+2023-03-13 07:27:26,202 INFO [train.py:968] (0/2) Epoch 26, batch 3950, giga_loss[loss=0.2933, simple_loss=0.3654, pruned_loss=0.1106, over 28587.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.344, pruned_loss=0.09435, over 5711940.47 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3405, pruned_loss=0.08591, over 4899922.64 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3451, pruned_loss=0.09562, over 5698546.61 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:28:06,531 INFO [train.py:968] (0/2) Epoch 26, batch 4000, giga_loss[loss=0.3218, simple_loss=0.3779, pruned_loss=0.1328, over 27896.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3436, pruned_loss=0.09475, over 5707581.79 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3402, pruned_loss=0.08569, over 4916491.78 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3449, pruned_loss=0.09614, over 5700027.94 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:28:26,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3030, 1.3903, 1.2287, 1.5434], device='cuda:0'), covar=tensor([0.0810, 0.0357, 0.0357, 0.0890], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 07:28:44,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.841e+02 1.142e+03 1.432e+03 1.916e+03 4.198e+03, threshold=2.863e+03, percent-clipped=7.0
+2023-03-13 07:28:47,579 INFO [train.py:968] (0/2) Epoch 26, batch 4050, giga_loss[loss=0.2538, simple_loss=0.3243, pruned_loss=0.09167, over 28620.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3413, pruned_loss=0.09371, over 5701099.14 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3406, pruned_loss=0.08599, over 4923349.13 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.342, pruned_loss=0.09473, over 5702926.40 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:29:02,343 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-13 07:29:24,943 INFO [train.py:968] (0/2) Epoch 26, batch 4100, giga_loss[loss=0.2359, simple_loss=0.3159, pruned_loss=0.07796, over 28707.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3382, pruned_loss=0.09218, over 5713393.79 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3403, pruned_loss=0.08601, over 4952613.12 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.339, pruned_loss=0.09318, over 5709488.40 frames. ], batch size: 66, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:29:47,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3732, 1.4042, 3.9958, 3.2509], device='cuda:0'), covar=tensor([0.1621, 0.2705, 0.0419, 0.0964], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0661, 0.0973, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:30:00,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.817e+02 1.221e+03 1.650e+03 2.241e+03 4.562e+03, threshold=3.299e+03, percent-clipped=7.0
+2023-03-13 07:30:03,502 INFO [train.py:968] (0/2) Epoch 26, batch 4150, giga_loss[loss=0.2994, simple_loss=0.3679, pruned_loss=0.1155, over 28654.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3372, pruned_loss=0.09174, over 5714756.96 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3402, pruned_loss=0.08598, over 4985437.99 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3379, pruned_loss=0.09276, over 5705581.84 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:30:42,206 INFO [train.py:968] (0/2) Epoch 26, batch 4200, giga_loss[loss=0.2291, simple_loss=0.3096, pruned_loss=0.07424, over 28820.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3366, pruned_loss=0.0916, over 5714573.00 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3399, pruned_loss=0.08593, over 5011671.29 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3373, pruned_loss=0.09269, over 5709598.80 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:30:51,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-13 07:31:21,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.959e+02 1.179e+03 1.575e+03 2.127e+03 9.707e+03, threshold=3.149e+03, percent-clipped=3.0
+2023-03-13 07:31:23,611 INFO [train.py:968] (0/2) Epoch 26, batch 4250, giga_loss[loss=0.2497, simple_loss=0.3256, pruned_loss=0.08693, over 28833.00 frames. ], tot_loss[loss=0.26, simple_loss=0.336, pruned_loss=0.09196, over 5712688.95 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3401, pruned_loss=0.08602, over 5024264.99 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3363, pruned_loss=0.09283, over 5706709.40 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:31:54,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2226, 1.2469, 1.1935, 1.4925], device='cuda:0'), covar=tensor([0.0735, 0.0383, 0.0355, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 07:32:00,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:32:03,473 INFO [train.py:968] (0/2) Epoch 26, batch 4300, giga_loss[loss=0.2559, simple_loss=0.3297, pruned_loss=0.09105, over 28634.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3337, pruned_loss=0.09101, over 5713112.11 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3403, pruned_loss=0.08607, over 5031871.66 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3337, pruned_loss=0.09171, over 5707720.58 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:32:09,006 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1143361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:32:41,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.322e+02 1.198e+03 1.511e+03 1.989e+03 4.977e+03, threshold=3.023e+03, percent-clipped=9.0
+2023-03-13 07:32:43,294 INFO [train.py:968] (0/2) Epoch 26, batch 4350, giga_loss[loss=0.2164, simple_loss=0.3027, pruned_loss=0.06505, over 28972.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3308, pruned_loss=0.08975, over 5707802.37 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3401, pruned_loss=0.08614, over 5039013.31 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3308, pruned_loss=0.09034, over 5708717.10 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:33:20,439 INFO [train.py:968] (0/2) Epoch 26, batch 4400, giga_loss[loss=0.3406, simple_loss=0.39, pruned_loss=0.1456, over 26774.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3307, pruned_loss=0.08967, over 5703551.20 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3409, pruned_loss=0.08655, over 5061048.94 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3299, pruned_loss=0.08992, over 5702174.83 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:33:26,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6220, 3.4621, 3.2622, 1.6033], device='cuda:0'), covar=tensor([0.0760, 0.0873, 0.0817, 0.2596], device='cuda:0'), in_proj_covar=tensor([0.1261, 0.1165, 0.0983, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 07:33:42,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3943, 1.8349, 1.5530, 1.5920], device='cuda:0'), covar=tensor([0.0648, 0.0263, 0.0295, 0.0734], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 07:33:51,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:33:57,977 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1143504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:33:59,279 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.208e+02 1.142e+03 1.426e+03 1.819e+03 5.926e+03, threshold=2.853e+03, percent-clipped=5.0
+2023-03-13 07:34:00,502 INFO [train.py:968] (0/2) Epoch 26, batch 4450, giga_loss[loss=0.2847, simple_loss=0.3654, pruned_loss=0.102, over 28880.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3337, pruned_loss=0.09113, over 5706363.38 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.341, pruned_loss=0.08666, over 5090241.32 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3326, pruned_loss=0.09134, over 5698917.66 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:34:00,745 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1143507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:34:02,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0820, 2.5637, 2.1310, 1.9738], device='cuda:0'), covar=tensor([0.1877, 0.1663, 0.2043, 0.2157], device='cuda:0'), in_proj_covar=tensor([0.0498, 0.0754, 0.0725, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 07:34:26,351 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1143536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:34:45,208 INFO [train.py:968] (0/2) Epoch 26, batch 4500, giga_loss[loss=0.2889, simple_loss=0.3601, pruned_loss=0.1089, over 28870.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3353, pruned_loss=0.09122, over 5716658.65 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3411, pruned_loss=0.08676, over 5105447.17 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3342, pruned_loss=0.09138, over 5708286.99 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:35:08,768 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:35:23,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.151e+02 1.095e+03 1.274e+03 1.728e+03 4.047e+03, threshold=2.548e+03, percent-clipped=4.0
+2023-03-13 07:35:25,395 INFO [train.py:968] (0/2) Epoch 26, batch 4550, giga_loss[loss=0.2708, simple_loss=0.3407, pruned_loss=0.1005, over 28668.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3372, pruned_loss=0.09146, over 5721322.31 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.341, pruned_loss=0.08675, over 5117503.87 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3364, pruned_loss=0.09165, over 5712298.43 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:35:44,145 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-03-13 07:36:07,750 INFO [train.py:968] (0/2) Epoch 26, batch 4600, giga_loss[loss=0.2686, simple_loss=0.3512, pruned_loss=0.09303, over 28801.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3389, pruned_loss=0.09225, over 5694337.22 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3413, pruned_loss=0.08703, over 5123236.03 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.338, pruned_loss=0.09234, over 5700728.13 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:36:11,454 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:36:15,764 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:36:45,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.529e+02 1.109e+03 1.409e+03 2.089e+03 5.908e+03, threshold=2.818e+03, percent-clipped=10.0
+2023-03-13 07:36:46,665 INFO [train.py:968] (0/2) Epoch 26, batch 4650, giga_loss[loss=0.2388, simple_loss=0.3259, pruned_loss=0.07588, over 28576.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3373, pruned_loss=0.09061, over 5694149.71 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3412, pruned_loss=0.08712, over 5147995.94 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3366, pruned_loss=0.09078, over 5694652.30 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:37:03,666 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:37:03,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1143727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:37:18,444 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:37:25,732 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143756.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:37:26,141 INFO [train.py:968] (0/2) Epoch 26, batch 4700, giga_loss[loss=0.2706, simple_loss=0.3424, pruned_loss=0.09944, over 28883.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3383, pruned_loss=0.09077, over 5701254.22 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3412, pruned_loss=0.08704, over 5164908.32 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3377, pruned_loss=0.09106, over 5698485.85 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:37:53,089 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:38:06,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.786e+02 1.374e+03 1.741e+03 2.410e+03 1.089e+04, threshold=3.481e+03, percent-clipped=14.0
+2023-03-13 07:38:07,908 INFO [train.py:968] (0/2) Epoch 26, batch 4750, giga_loss[loss=0.2596, simple_loss=0.3379, pruned_loss=0.09067, over 29001.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3393, pruned_loss=0.09181, over 5700110.35 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3409, pruned_loss=0.08698, over 5166551.10 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.339, pruned_loss=0.09212, over 5704053.07 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:38:34,023 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 07:38:49,178 INFO [train.py:968] (0/2) Epoch 26, batch 4800, giga_loss[loss=0.36, simple_loss=0.4058, pruned_loss=0.157, over 26722.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3403, pruned_loss=0.09292, over 5700409.47 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.341, pruned_loss=0.08704, over 5177985.96 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3399, pruned_loss=0.09318, over 5700415.54 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:39:00,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1143870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:39:00,886 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1143870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:39:03,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1143873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:39:27,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1143902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:39:28,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.213e+02 1.305e+03 1.525e+03 2.046e+03 6.460e+03, threshold=3.050e+03, percent-clipped=5.0
+2023-03-13 07:39:30,368 INFO [train.py:968] (0/2) Epoch 26, batch 4850, giga_loss[loss=0.2475, simple_loss=0.3281, pruned_loss=0.08347, over 28781.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3424, pruned_loss=0.09405, over 5710213.40 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.341, pruned_loss=0.08697, over 5194384.08 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3421, pruned_loss=0.09447, over 5707415.25 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:40:10,610 INFO [train.py:968] (0/2) Epoch 26, batch 4900, libri_loss[loss=0.2251, simple_loss=0.3084, pruned_loss=0.07091, over 29355.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3445, pruned_loss=0.09476, over 5713896.69 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3408, pruned_loss=0.08676, over 5221817.44 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3445, pruned_loss=0.09557, over 5704862.40 frames. ], batch size: 67, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:40:14,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1143962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:40:14,933 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-13 07:40:45,509 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1144000.pt
+2023-03-13 07:40:51,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.981e+02 1.445e+03 1.897e+03 2.429e+03 5.643e+03, threshold=3.795e+03, percent-clipped=15.0
+2023-03-13 07:40:52,276 INFO [train.py:968] (0/2) Epoch 26, batch 4950, giga_loss[loss=0.2485, simple_loss=0.3282, pruned_loss=0.0844, over 28831.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3468, pruned_loss=0.09607, over 5714824.19 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3407, pruned_loss=0.08676, over 5235232.81 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.347, pruned_loss=0.09685, over 5704525.67 frames. ], batch size: 112, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:40:56,953 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:40:57,504 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:40:58,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:41:14,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:41:18,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:41:20,178 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:41:30,950 INFO [train.py:968] (0/2) Epoch 26, batch 5000, giga_loss[loss=0.2467, simple_loss=0.3257, pruned_loss=0.08382, over 28541.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3478, pruned_loss=0.09679, over 5713675.53 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3407, pruned_loss=0.08676, over 5248313.88 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.348, pruned_loss=0.09758, over 5702359.57 frames. ], batch size: 60, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:41:38,290 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:05,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:07,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.107e+02 1.339e+03 1.647e+03 2.211e+03 3.688e+03, threshold=3.294e+03, percent-clipped=0.0
+2023-03-13 07:42:08,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:09,142 INFO [train.py:968] (0/2) Epoch 26, batch 5050, giga_loss[loss=0.2611, simple_loss=0.3433, pruned_loss=0.08948, over 28940.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.349, pruned_loss=0.09745, over 5717674.15 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3415, pruned_loss=0.0872, over 5267496.53 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3486, pruned_loss=0.098, over 5703646.88 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:42:10,083 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:19,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:21,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:29,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144131.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:42:34,835 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:50,406 INFO [train.py:968] (0/2) Epoch 26, batch 5100, giga_loss[loss=0.2451, simple_loss=0.3304, pruned_loss=0.07991, over 28655.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3486, pruned_loss=0.09744, over 5720008.90 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3421, pruned_loss=0.08759, over 5276761.05 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3479, pruned_loss=0.09767, over 5706511.40 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:42:57,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:08,387 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:12,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:15,704 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:18,259 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:31,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.353e+02 1.275e+03 1.727e+03 2.234e+03 7.799e+03, threshold=3.455e+03, percent-clipped=13.0
+2023-03-13 07:43:32,068 INFO [train.py:968] (0/2) Epoch 26, batch 5150, giga_loss[loss=0.2232, simple_loss=0.3107, pruned_loss=0.06782, over 28975.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3451, pruned_loss=0.09587, over 5704245.10 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3421, pruned_loss=0.08764, over 5278782.59 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3447, pruned_loss=0.09616, over 5698762.44 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:43:36,272 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:40,176 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:49,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8888, 2.8147, 1.8593, 0.9372], device='cuda:0'), covar=tensor([0.9582, 0.3600, 0.4677, 0.8787], device='cuda:0'), in_proj_covar=tensor([0.1804, 0.1684, 0.1631, 0.1468], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 07:44:00,943 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:03,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:11,428 INFO [train.py:968] (0/2) Epoch 26, batch 5200, giga_loss[loss=0.2527, simple_loss=0.3314, pruned_loss=0.08702, over 29084.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3425, pruned_loss=0.09465, over 5715858.25 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3427, pruned_loss=0.088, over 5301336.78 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3417, pruned_loss=0.09485, over 5706479.06 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:44:17,548 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:20,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:25,854 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144274.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 07:44:27,940 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:27,995 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144277.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:44:31,395 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 07:44:32,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6349, 1.7831, 1.8767, 1.4193], device='cuda:0'), covar=tensor([0.1904, 0.2503, 0.1574, 0.1783], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0712, 0.0973, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 07:44:42,141 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:50,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.692e+02 1.165e+03 1.489e+03 2.044e+03 6.270e+03, threshold=2.977e+03, percent-clipped=5.0
+2023-03-13 07:44:51,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144306.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:44:51,489 INFO [train.py:968] (0/2) Epoch 26, batch 5250, giga_loss[loss=0.2935, simple_loss=0.3711, pruned_loss=0.1079, over 28674.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.341, pruned_loss=0.09293, over 5713876.02 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3424, pruned_loss=0.08784, over 5311003.33 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3406, pruned_loss=0.09334, over 5704963.14 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:44:52,316 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:54,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:45:17,906 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:45:32,814 INFO [train.py:968] (0/2) Epoch 26, batch 5300, giga_loss[loss=0.2554, simple_loss=0.3464, pruned_loss=0.08221, over 28918.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3429, pruned_loss=0.09278, over 5720065.50 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3427, pruned_loss=0.08812, over 5325162.27 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3424, pruned_loss=0.09297, over 5709253.28 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:45:47,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4517, 1.7626, 1.6645, 1.4963], device='cuda:0'), covar=tensor([0.1986, 0.1912, 0.2353, 0.2233], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0752, 0.0723, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 07:45:59,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:46:14,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.102e+02 1.272e+03 1.507e+03 2.035e+03 5.949e+03, threshold=3.015e+03, percent-clipped=7.0
+2023-03-13 07:46:14,613 INFO [train.py:968] (0/2) Epoch 26, batch 5350, giga_loss[loss=0.2599, simple_loss=0.3354, pruned_loss=0.09225, over 28894.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3432, pruned_loss=0.09275, over 5725178.09 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3424, pruned_loss=0.08794, over 5339382.09 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3429, pruned_loss=0.09316, over 5712424.84 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:46:39,163 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:46:40,812 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:46:52,264 INFO [train.py:968] (0/2) Epoch 26, batch 5400, giga_loss[loss=0.2215, simple_loss=0.2958, pruned_loss=0.0736, over 28299.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3412, pruned_loss=0.09288, over 5717437.42 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3432, pruned_loss=0.08856, over 5341886.08 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3403, pruned_loss=0.0928, over 5712658.63 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:47:15,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8376, 2.0633, 1.8662, 1.7685], device='cuda:0'), covar=tensor([0.2192, 0.2568, 0.2447, 0.2603], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0752, 0.0723, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 07:47:24,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:47:36,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.419e+02 1.277e+03 1.497e+03 2.156e+03 4.625e+03, threshold=2.994e+03, percent-clipped=6.0
+2023-03-13 07:47:36,334 INFO [train.py:968] (0/2) Epoch 26, batch 5450, giga_loss[loss=0.2573, simple_loss=0.3309, pruned_loss=0.09188, over 28842.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3402, pruned_loss=0.09392, over 5723798.86 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3433, pruned_loss=0.08878, over 5347574.85 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3394, pruned_loss=0.09372, over 5718414.67 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:47:54,630 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:47:57,829 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:48:17,312 INFO [train.py:968] (0/2) Epoch 26, batch 5500, giga_loss[loss=0.2044, simple_loss=0.2764, pruned_loss=0.06621, over 28447.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3376, pruned_loss=0.09379, over 5727091.66 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3435, pruned_loss=0.08892, over 5353266.19 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3368, pruned_loss=0.09358, over 5721263.89 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:48:22,981 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:48:39,580 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:48:41,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:48:41,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:48:59,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.459e+02 1.208e+03 1.432e+03 1.813e+03 4.694e+03, threshold=2.865e+03, percent-clipped=4.0
+2023-03-13 07:48:59,676 INFO [train.py:968] (0/2) Epoch 26, batch 5550, giga_loss[loss=0.2392, simple_loss=0.3205, pruned_loss=0.07898, over 28842.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3373, pruned_loss=0.09427, over 5725334.00 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3437, pruned_loss=0.08896, over 5356176.86 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3365, pruned_loss=0.09408, over 5719879.52 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:49:05,934 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:49:13,291 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2295, 4.0777, 3.8287, 1.8673], device='cuda:0'), covar=tensor([0.0671, 0.0817, 0.0804, 0.2125], device='cuda:0'), in_proj_covar=tensor([0.1265, 0.1172, 0.0987, 0.0740], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 07:49:24,109 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:49:26,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:49:39,293 INFO [train.py:968] (0/2) Epoch 26, batch 5600, libri_loss[loss=0.2674, simple_loss=0.3508, pruned_loss=0.09195, over 29773.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3361, pruned_loss=0.09375, over 5720902.20 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3443, pruned_loss=0.0892, over 5371419.22 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3347, pruned_loss=0.09356, over 5714337.52 frames. ], batch size: 87, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:49:49,178 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:50:15,362 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:50:19,316 INFO [train.py:968] (0/2) Epoch 26, batch 5650, giga_loss[loss=0.2424, simple_loss=0.3188, pruned_loss=0.08305, over 28679.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3323, pruned_loss=0.09203, over 5718238.80 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3446, pruned_loss=0.08934, over 5376535.77 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3309, pruned_loss=0.0918, over 5711582.17 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:50:19,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.050e+02 1.301e+03 1.746e+03 2.240e+03 4.691e+03, threshold=3.492e+03, percent-clipped=11.0
+2023-03-13 07:50:53,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5138, 1.6897, 1.7370, 1.2853], device='cuda:0'), covar=tensor([0.1794, 0.2591, 0.1523, 0.1748], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0711, 0.0972, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 07:50:59,612 INFO [train.py:968] (0/2) Epoch 26, batch 5700, giga_loss[loss=0.2486, simple_loss=0.3181, pruned_loss=0.08954, over 28757.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3287, pruned_loss=0.09021, over 5707244.37 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3445, pruned_loss=0.08947, over 5377589.93 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3274, pruned_loss=0.08994, over 5706466.95 frames. ], batch size: 66, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:51:14,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2599, 1.5559, 1.5277, 1.1193], device='cuda:0'), covar=tensor([0.1776, 0.2543, 0.1488, 0.1677], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0711, 0.0972, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 07:51:33,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:51:38,196 INFO [train.py:968] (0/2) Epoch 26, batch 5750, giga_loss[loss=0.2354, simple_loss=0.3181, pruned_loss=0.07638, over 28700.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3282, pruned_loss=0.08971, over 5711801.50 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3449, pruned_loss=0.08974, over 5392132.74 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3264, pruned_loss=0.08927, over 5707092.96 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:51:38,358 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:51:38,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.566e+02 1.220e+03 1.601e+03 2.215e+03 7.466e+03, threshold=3.202e+03, percent-clipped=5.0
+2023-03-13 07:51:43,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:51:48,289 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 07:52:18,270 INFO [train.py:968] (0/2) Epoch 26, batch 5800, giga_loss[loss=0.2553, simple_loss=0.33, pruned_loss=0.09034, over 28935.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3306, pruned_loss=0.09068, over 5713621.44 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3443, pruned_loss=0.08943, over 5406322.95 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3293, pruned_loss=0.09061, over 5705590.89 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:52:36,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3709, 1.4353, 3.1619, 3.0329], device='cuda:0'), covar=tensor([0.1314, 0.2585, 0.0446, 0.1136], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0663, 0.0975, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:52:57,935 INFO [train.py:968] (0/2) Epoch 26, batch 5850, giga_loss[loss=0.2645, simple_loss=0.341, pruned_loss=0.09397, over 28833.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3334, pruned_loss=0.09147, over 5714507.08 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3435, pruned_loss=0.08914, over 5419977.07 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3326, pruned_loss=0.09172, over 5703964.61 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:52:58,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.595e+02 1.218e+03 1.432e+03 2.039e+03 5.677e+03, threshold=2.865e+03, percent-clipped=7.0
+2023-03-13 07:53:04,398 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-13 07:53:36,976 INFO [train.py:968] (0/2) Epoch 26, batch 5900, giga_loss[loss=0.3078, simple_loss=0.3836, pruned_loss=0.116, over 29003.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3372, pruned_loss=0.09276, over 5720210.12 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3435, pruned_loss=0.08933, over 5431975.73 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3365, pruned_loss=0.09285, over 5707696.03 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:53:38,851 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:53:40,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:53:44,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:53:47,838 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 07:54:07,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0432, 1.1496, 3.4011, 3.0526], device='cuda:0'), covar=tensor([0.1724, 0.2813, 0.0472, 0.0890], device='cuda:0'), in_proj_covar=tensor([0.0783, 0.0663, 0.0976, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:54:07,666 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:54:07,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2498, 1.8483, 1.3984, 0.4701], device='cuda:0'), covar=tensor([0.4945, 0.3013, 0.4480, 0.6594], device='cuda:0'), in_proj_covar=tensor([0.1796, 0.1677, 0.1621, 0.1461], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 07:54:22,676 INFO [train.py:968] (0/2) Epoch 26, batch 5950, giga_loss[loss=0.273, simple_loss=0.3455, pruned_loss=0.1002, over 28939.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3403, pruned_loss=0.09422, over 5718640.23 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3435, pruned_loss=0.08935, over 5440504.74 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3396, pruned_loss=0.09438, over 5705911.17 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:54:23,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.553e+02 1.249e+03 1.589e+03 2.096e+03 4.095e+03, threshold=3.178e+03, percent-clipped=5.0
+2023-03-13 07:55:03,109 INFO [train.py:968] (0/2) Epoch 26, batch 6000, giga_loss[loss=0.2626, simple_loss=0.3299, pruned_loss=0.09759, over 28468.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3429, pruned_loss=0.09555, over 5709538.92 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3439, pruned_loss=0.08955, over 5445629.94 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3419, pruned_loss=0.09571, over 5704195.85 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:55:03,112 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 07:55:11,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2191, 1.3700, 3.3282, 3.1256], device='cuda:0'), covar=tensor([0.1719, 0.2811, 0.0507, 0.1004], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0661, 0.0975, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:55:11,781 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2071, simple_loss=0.3137, pruned_loss=0.05024, over 944034.00 frames. 
+2023-03-13 07:55:11,782 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 07:55:29,132 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1145077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:55:55,064 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1145106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:55:55,468 INFO [train.py:968] (0/2) Epoch 26, batch 6050, giga_loss[loss=0.2691, simple_loss=0.3433, pruned_loss=0.09743, over 28956.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3491, pruned_loss=0.101, over 5702515.58 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.344, pruned_loss=0.08958, over 5452211.69 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3482, pruned_loss=0.1013, over 5696874.06 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:55:56,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.875e+02 1.530e+03 1.853e+03 2.486e+03 7.039e+03, threshold=3.707e+03, percent-clipped=16.0
+2023-03-13 07:55:58,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:56:19,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0714, 1.0806, 3.4433, 3.0646], device='cuda:0'), covar=tensor([0.2029, 0.3054, 0.0852, 0.1143], device='cuda:0'), in_proj_covar=tensor([0.0782, 0.0661, 0.0974, 0.0943], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 07:56:26,874 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1145138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:56:43,959 INFO [train.py:968] (0/2) Epoch 26, batch 6100, giga_loss[loss=0.2697, simple_loss=0.3456, pruned_loss=0.09695, over 28776.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3552, pruned_loss=0.1061, over 5700012.77 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.344, pruned_loss=0.08979, over 5459128.73 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3546, pruned_loss=0.1063, over 5692444.16 frames. ], batch size: 66, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:57:03,437 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1145176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:57:07,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1145182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:57:30,712 INFO [train.py:968] (0/2) Epoch 26, batch 6150, giga_loss[loss=0.297, simple_loss=0.3715, pruned_loss=0.1113, over 28820.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3622, pruned_loss=0.1108, over 5694460.86 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3444, pruned_loss=0.08998, over 5463773.47 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3617, pruned_loss=0.1111, over 5688737.45 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:57:33,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.246e+03 1.678e+03 2.154e+03 2.932e+03 5.294e+03, threshold=4.308e+03, percent-clipped=11.0
+2023-03-13 07:57:44,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1145220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:57:47,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:58:14,414 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1145252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:58:20,022 INFO [train.py:968] (0/2) Epoch 26, batch 6200, giga_loss[loss=0.366, simple_loss=0.417, pruned_loss=0.1575, over 28693.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3671, pruned_loss=0.115, over 5699370.58 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3444, pruned_loss=0.09005, over 5474487.60 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3671, pruned_loss=0.1157, over 5690406.99 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:59:03,679 INFO [train.py:968] (0/2) Epoch 26, batch 6250, libri_loss[loss=0.2518, simple_loss=0.3342, pruned_loss=0.08474, over 29649.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3725, pruned_loss=0.1193, over 5693248.26 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3446, pruned_loss=0.0901, over 5481553.81 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3729, pruned_loss=0.1204, over 5684696.85 frames. ], batch size: 73, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:59:05,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.871e+03 2.479e+03 3.094e+03 9.218e+03, threshold=4.958e+03, percent-clipped=11.0
+2023-03-13 07:59:09,064 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-13 07:59:15,566 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1145319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:59:16,946 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-13 07:59:17,692 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:59:19,842 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1145325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:59:22,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:59:44,682 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1145351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:59:52,031 INFO [train.py:968] (0/2) Epoch 26, batch 6300, giga_loss[loss=0.3598, simple_loss=0.4155, pruned_loss=0.1521, over 28647.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3754, pruned_loss=0.1222, over 5687967.10 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3445, pruned_loss=0.09003, over 5494175.54 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3767, pruned_loss=0.1241, over 5675474.48 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:59:52,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1145357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:00:43,457 INFO [train.py:968] (0/2) Epoch 26, batch 6350, giga_loss[loss=0.3189, simple_loss=0.3821, pruned_loss=0.1279, over 28926.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3774, pruned_loss=0.1249, over 5671813.71 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3446, pruned_loss=0.08999, over 5500030.10 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3788, pruned_loss=0.1268, over 5659050.04 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:00:44,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.762e+03 2.391e+03 3.237e+03 1.050e+04, threshold=4.781e+03, percent-clipped=6.0
+2023-03-13 08:01:15,673 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 08:01:36,521 INFO [train.py:968] (0/2) Epoch 26, batch 6400, giga_loss[loss=0.2694, simple_loss=0.3391, pruned_loss=0.09988, over 28548.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3805, pruned_loss=0.1283, over 5673908.63 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3444, pruned_loss=0.08984, over 5506510.72 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3824, pruned_loss=0.1307, over 5661064.19 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:02:30,405 INFO [train.py:968] (0/2) Epoch 26, batch 6450, giga_loss[loss=0.3297, simple_loss=0.3943, pruned_loss=0.1326, over 28649.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3841, pruned_loss=0.1325, over 5654999.46 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3442, pruned_loss=0.08982, over 5516381.29 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3866, pruned_loss=0.1353, over 5639735.16 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:02:32,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.296e+03 1.980e+03 2.684e+03 3.480e+03 9.505e+03, threshold=5.369e+03, percent-clipped=10.0
+2023-03-13 08:02:32,487 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1145510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:02:48,713 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0065, 1.2474, 2.8775, 2.8824], device='cuda:0'), covar=tensor([0.1597, 0.2430, 0.0627, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0787, 0.0664, 0.0980, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 08:03:01,742 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1145541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:03:19,808 INFO [train.py:968] (0/2) Epoch 26, batch 6500, libri_loss[loss=0.2547, simple_loss=0.3365, pruned_loss=0.08644, over 29562.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3858, pruned_loss=0.1338, over 5650105.21 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3443, pruned_loss=0.0898, over 5519455.39 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3884, pruned_loss=0.1368, over 5637454.30 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:03:39,910 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-13 08:04:07,347 INFO [train.py:968] (0/2) Epoch 26, batch 6550, giga_loss[loss=0.3061, simple_loss=0.3661, pruned_loss=0.123, over 28418.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.385, pruned_loss=0.1341, over 5650308.86 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.0897, over 5527222.82 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3882, pruned_loss=0.1376, over 5636020.93 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:04:09,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.317e+03 1.923e+03 2.470e+03 3.167e+03 5.415e+03, threshold=4.940e+03, percent-clipped=1.0
+2023-03-13 08:04:53,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3955, 1.5480, 1.6677, 1.4576], device='cuda:0'), covar=tensor([0.1705, 0.1549, 0.1662, 0.1554], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0759, 0.0728, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 08:04:57,368 INFO [train.py:968] (0/2) Epoch 26, batch 6600, giga_loss[loss=0.3686, simple_loss=0.3982, pruned_loss=0.1695, over 23563.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3843, pruned_loss=0.1341, over 5638140.85 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3443, pruned_loss=0.08968, over 5535467.10 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3877, pruned_loss=0.138, over 5621923.85 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 1.0
+2023-03-13 08:04:57,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4867, 1.7660, 1.6085, 1.5471], device='cuda:0'), covar=tensor([0.2140, 0.2185, 0.2628, 0.2277], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0758, 0.0728, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 08:05:28,910 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 08:05:48,563 INFO [train.py:968] (0/2) Epoch 26, batch 6650, giga_loss[loss=0.2846, simple_loss=0.3651, pruned_loss=0.102, over 29076.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3858, pruned_loss=0.1342, over 5650718.70 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3445, pruned_loss=0.08978, over 5539301.56 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3887, pruned_loss=0.1376, over 5635616.18 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 1.0
+2023-03-13 08:05:53,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.154e+03 1.755e+03 2.223e+03 3.756e+03 1.923e+04, threshold=4.447e+03, percent-clipped=16.0
+2023-03-13 08:06:34,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6862, 1.9997, 1.4073, 1.5215], device='cuda:0'), covar=tensor([0.1013, 0.0598, 0.1029, 0.1087], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0452, 0.0522, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 08:06:34,831 INFO [train.py:968] (0/2) Epoch 26, batch 6700, giga_loss[loss=0.3168, simple_loss=0.3874, pruned_loss=0.1231, over 28993.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3847, pruned_loss=0.1322, over 5648375.19 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3445, pruned_loss=0.08963, over 5540185.35 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3881, pruned_loss=0.1363, over 5637848.77 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 1.0
+2023-03-13 08:06:50,245 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5716, 1.7819, 1.6853, 1.5978], device='cuda:0'), covar=tensor([0.2035, 0.2096, 0.2337, 0.2004], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0759, 0.0728, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 08:07:21,350 INFO [train.py:968] (0/2) Epoch 26, batch 6750, giga_loss[loss=0.3724, simple_loss=0.4237, pruned_loss=0.1606, over 28560.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3836, pruned_loss=0.1312, over 5637086.57 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3439, pruned_loss=0.08927, over 5552272.14 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3881, pruned_loss=0.1361, over 5621007.17 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 1.0
+2023-03-13 08:07:25,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.913e+02 1.707e+03 2.188e+03 3.082e+03 6.864e+03, threshold=4.376e+03, percent-clipped=6.0
+2023-03-13 08:08:01,570 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 08:08:12,514 INFO [train.py:968] (0/2) Epoch 26, batch 6800, giga_loss[loss=0.2799, simple_loss=0.3531, pruned_loss=0.1033, over 28635.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3804, pruned_loss=0.1281, over 5634604.24 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08893, over 5558143.94 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3854, pruned_loss=0.1333, over 5618165.20 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:08:26,305 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-13 08:08:39,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1145885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:08:57,699 INFO [train.py:968] (0/2) Epoch 26, batch 6850, libri_loss[loss=0.3032, simple_loss=0.3828, pruned_loss=0.1119, over 29212.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3787, pruned_loss=0.125, over 5655126.71 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3437, pruned_loss=0.08911, over 5568216.25 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3834, pruned_loss=0.13, over 5634911.68 frames. ], batch size: 101, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:09:02,678 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-13 08:09:02,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.601e+03 1.988e+03 2.831e+03 8.642e+03, threshold=3.975e+03, percent-clipped=9.0
+2023-03-13 08:09:05,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1145916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:09:44,814 INFO [train.py:968] (0/2) Epoch 26, batch 6900, giga_loss[loss=0.2767, simple_loss=0.3531, pruned_loss=0.1001, over 27917.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3741, pruned_loss=0.1213, over 5660370.62 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3432, pruned_loss=0.0888, over 5580067.98 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3794, pruned_loss=0.1267, over 5636121.24 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:10:14,095 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-13 08:10:21,973 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6247, 1.6318, 1.8248, 1.4252], device='cuda:0'), covar=tensor([0.1513, 0.2224, 0.1277, 0.1517], device='cuda:0'), in_proj_covar=tensor([0.0921, 0.0708, 0.0966, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 08:10:25,256 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1146000.pt
+2023-03-13 08:10:32,236 INFO [train.py:968] (0/2) Epoch 26, batch 6950, giga_loss[loss=0.2696, simple_loss=0.3417, pruned_loss=0.0988, over 28791.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3718, pruned_loss=0.1194, over 5662389.93 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3433, pruned_loss=0.08886, over 5581429.20 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3764, pruned_loss=0.1242, over 5643504.16 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:10:37,565 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.646e+03 2.323e+03 2.902e+03 1.020e+04, threshold=4.646e+03, percent-clipped=8.0
+2023-03-13 08:10:37,780 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1146012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:10:52,525 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1146028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:10:54,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1146031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:11:19,088 INFO [train.py:968] (0/2) Epoch 26, batch 7000, giga_loss[loss=0.3189, simple_loss=0.3832, pruned_loss=0.1273, over 28992.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.369, pruned_loss=0.118, over 5645408.49 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3433, pruned_loss=0.08893, over 5577014.75 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3735, pruned_loss=0.1226, over 5636182.18 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:11:21,157 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1146059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:11:21,823 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1146060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:11:25,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1146062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:11:50,024 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1146091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:12:06,077 INFO [train.py:968] (0/2) Epoch 26, batch 7050, giga_loss[loss=0.3703, simple_loss=0.4163, pruned_loss=0.1621, over 27843.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.37, pruned_loss=0.1188, over 5645178.28 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3434, pruned_loss=0.08894, over 5577637.69 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3737, pruned_loss=0.1228, over 5638374.02 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:12:12,692 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.696e+03 2.108e+03 2.909e+03 8.862e+03, threshold=4.216e+03, percent-clipped=6.0
+2023-03-13 08:12:13,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6893, 1.7954, 1.3103, 1.4825], device='cuda:0'), covar=tensor([0.1030, 0.0764, 0.1075, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0451, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 08:13:03,603 INFO [train.py:968] (0/2) Epoch 26, batch 7100, giga_loss[loss=0.2963, simple_loss=0.3557, pruned_loss=0.1184, over 27555.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3691, pruned_loss=0.1183, over 5649158.42 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3429, pruned_loss=0.08868, over 5582250.22 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3728, pruned_loss=0.122, over 5640647.57 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:13:48,952 INFO [train.py:968] (0/2) Epoch 26, batch 7150, libri_loss[loss=0.292, simple_loss=0.3694, pruned_loss=0.1073, over 28083.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3686, pruned_loss=0.1165, over 5644613.03 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3432, pruned_loss=0.08902, over 5580471.25 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3723, pruned_loss=0.1203, over 5642321.91 frames. ], batch size: 116, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:13:56,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.527e+02 1.655e+03 2.110e+03 3.082e+03 1.097e+04, threshold=4.219e+03, percent-clipped=13.0
+2023-03-13 08:14:43,174 INFO [train.py:968] (0/2) Epoch 26, batch 7200, giga_loss[loss=0.2897, simple_loss=0.3694, pruned_loss=0.105, over 29064.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3691, pruned_loss=0.1145, over 5658108.00 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3427, pruned_loss=0.08879, over 5587061.86 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.373, pruned_loss=0.1183, over 5651884.72 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:15:30,699 INFO [train.py:968] (0/2) Epoch 26, batch 7250, giga_loss[loss=0.2838, simple_loss=0.3575, pruned_loss=0.105, over 28757.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.37, pruned_loss=0.1146, over 5674621.59 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3423, pruned_loss=0.08865, over 5599099.39 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3746, pruned_loss=0.1187, over 5661496.21 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:15:34,063 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.969e+02 1.727e+03 2.235e+03 3.270e+03 6.843e+03, threshold=4.470e+03, percent-clipped=15.0
+2023-03-13 08:15:34,476 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7993, 2.1208, 2.1202, 1.6553], device='cuda:0'), covar=tensor([0.3187, 0.2637, 0.2709, 0.3042], device='cuda:0'), in_proj_covar=tensor([0.2039, 0.1982, 0.1914, 0.2046], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 08:15:57,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4813, 1.5397, 1.6793, 1.2995], device='cuda:0'), covar=tensor([0.1635, 0.2429, 0.1342, 0.1670], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0709, 0.0967, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 08:16:17,544 INFO [train.py:968] (0/2) Epoch 26, batch 7300, giga_loss[loss=0.2845, simple_loss=0.3573, pruned_loss=0.1058, over 28913.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3702, pruned_loss=0.1155, over 5667290.33 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08846, over 5605018.83 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3748, pruned_loss=0.1197, over 5652945.89 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:16:17,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3678, 3.6469, 1.6789, 1.5660], device='cuda:0'), covar=tensor([0.1048, 0.0431, 0.0882, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0566, 0.0401, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 08:16:49,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1146387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:17:05,355 INFO [train.py:968] (0/2) Epoch 26, batch 7350, giga_loss[loss=0.2621, simple_loss=0.3336, pruned_loss=0.09531, over 28922.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3688, pruned_loss=0.1155, over 5678171.81 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3414, pruned_loss=0.08815, over 5612380.37 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3738, pruned_loss=0.1197, over 5661263.35 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:17:10,732 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.141e+03 1.722e+03 2.181e+03 2.942e+03 8.345e+03, threshold=4.362e+03, percent-clipped=8.0
+2023-03-13 08:17:25,765 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1146425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:17:55,594 INFO [train.py:968] (0/2) Epoch 26, batch 7400, giga_loss[loss=0.3269, simple_loss=0.3799, pruned_loss=0.1369, over 27634.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3674, pruned_loss=0.116, over 5675349.11 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3415, pruned_loss=0.08828, over 5615303.22 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3715, pruned_loss=0.1195, over 5660173.45 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:18:44,601 INFO [train.py:968] (0/2) Epoch 26, batch 7450, giga_loss[loss=0.3069, simple_loss=0.3772, pruned_loss=0.1183, over 28576.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3678, pruned_loss=0.1169, over 5673146.76 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3415, pruned_loss=0.08828, over 5615303.22 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.371, pruned_loss=0.1196, over 5661335.36 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:18:52,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.699e+03 2.028e+03 2.582e+03 7.453e+03, threshold=4.056e+03, percent-clipped=3.0
+2023-03-13 08:19:08,115 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1146530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:19:11,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1146533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:19:32,712 INFO [train.py:968] (0/2) Epoch 26, batch 7500, giga_loss[loss=0.265, simple_loss=0.3453, pruned_loss=0.09232, over 28507.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.368, pruned_loss=0.1159, over 5667826.46 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3416, pruned_loss=0.0883, over 5619783.34 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.371, pruned_loss=0.1187, over 5655412.08 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:19:37,355 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1146562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:19:59,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3673, 1.6761, 1.3655, 1.2966], device='cuda:0'), covar=tensor([0.2552, 0.2553, 0.2923, 0.2252], device='cuda:0'), in_proj_covar=tensor([0.1574, 0.1135, 0.1389, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 08:20:18,658 INFO [train.py:968] (0/2) Epoch 26, batch 7550, giga_loss[loss=0.2859, simple_loss=0.3708, pruned_loss=0.1005, over 28976.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3692, pruned_loss=0.116, over 5664405.54 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3416, pruned_loss=0.08825, over 5613643.39 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3719, pruned_loss=0.1187, over 5659910.44 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:20:23,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.657e+02 1.694e+03 2.023e+03 2.736e+03 1.307e+04, threshold=4.047e+03, percent-clipped=5.0
+2023-03-13 08:20:38,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0692, 1.1740, 3.7536, 3.1998], device='cuda:0'), covar=tensor([0.1835, 0.2931, 0.0544, 0.0913], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0668, 0.0988, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 08:20:58,079 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1146651.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:21:01,658 INFO [train.py:968] (0/2) Epoch 26, batch 7600, giga_loss[loss=0.3175, simple_loss=0.3848, pruned_loss=0.1251, over 29048.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3683, pruned_loss=0.1148, over 5680980.96 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3418, pruned_loss=0.08835, over 5621715.00 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3711, pruned_loss=0.1176, over 5671688.73 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:21:09,120 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 08:21:31,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2874, 0.8598, 0.9543, 1.4797], device='cuda:0'), covar=tensor([0.0762, 0.0408, 0.0355, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 08:21:52,107 INFO [train.py:968] (0/2) Epoch 26, batch 7650, giga_loss[loss=0.3316, simple_loss=0.3994, pruned_loss=0.1319, over 28698.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3678, pruned_loss=0.1155, over 5671663.16 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3418, pruned_loss=0.08827, over 5618257.88 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3704, pruned_loss=0.118, over 5668400.09 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:21:56,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.871e+03 2.193e+03 3.231e+03 5.997e+03, threshold=4.386e+03, percent-clipped=14.0
+2023-03-13 08:22:42,539 INFO [train.py:968] (0/2) Epoch 26, batch 7700, giga_loss[loss=0.2797, simple_loss=0.3557, pruned_loss=0.1018, over 28956.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3661, pruned_loss=0.1151, over 5662564.78 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3418, pruned_loss=0.08823, over 5621247.15 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3685, pruned_loss=0.1177, over 5657939.80 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:23:24,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1146800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:23:30,149 INFO [train.py:968] (0/2) Epoch 26, batch 7750, giga_loss[loss=0.3116, simple_loss=0.3783, pruned_loss=0.1225, over 28578.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3654, pruned_loss=0.1155, over 5663572.26 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3421, pruned_loss=0.08831, over 5628452.69 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3676, pruned_loss=0.1181, over 5654509.59 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:23:36,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.250e+03 1.696e+03 2.131e+03 3.020e+03 8.666e+03, threshold=4.262e+03, percent-clipped=7.0
+2023-03-13 08:24:17,631 INFO [train.py:968] (0/2) Epoch 26, batch 7800, giga_loss[loss=0.2776, simple_loss=0.3486, pruned_loss=0.1033, over 28919.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.364, pruned_loss=0.1152, over 5663202.22 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3424, pruned_loss=0.0883, over 5631803.25 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.366, pruned_loss=0.1178, over 5653680.22 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:25:02,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-13 08:25:03,936 INFO [train.py:968] (0/2) Epoch 26, batch 7850, giga_loss[loss=0.2672, simple_loss=0.3374, pruned_loss=0.0985, over 28598.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3631, pruned_loss=0.1152, over 5647819.29 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3426, pruned_loss=0.08845, over 5626461.25 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3649, pruned_loss=0.1175, over 5645335.75 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:25:07,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.791e+03 2.294e+03 3.036e+03 7.364e+03, threshold=4.587e+03, percent-clipped=4.0
+2023-03-13 08:25:34,000 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 08:25:35,524 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1146943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:25:38,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1146946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:25:48,737 INFO [train.py:968] (0/2) Epoch 26, batch 7900, giga_loss[loss=0.3144, simple_loss=0.3818, pruned_loss=0.1235, over 28554.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3629, pruned_loss=0.1145, over 5661201.04 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3428, pruned_loss=0.08852, over 5631521.70 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3645, pruned_loss=0.1168, over 5655338.75 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:26:01,653 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1146969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:26:07,939 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1146975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:26:27,896 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-13 08:26:35,873 INFO [train.py:968] (0/2) Epoch 26, batch 7950, giga_loss[loss=0.2814, simple_loss=0.3627, pruned_loss=0.1, over 28946.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.364, pruned_loss=0.1145, over 5665371.02 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3426, pruned_loss=0.08853, over 5635685.55 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3658, pruned_loss=0.1169, over 5657254.51 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:26:42,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.240e+03 1.702e+03 2.113e+03 2.747e+03 9.077e+03, threshold=4.226e+03, percent-clipped=8.0
+2023-03-13 08:26:55,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1147026.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:27:22,356 INFO [train.py:968] (0/2) Epoch 26, batch 8000, giga_loss[loss=0.2582, simple_loss=0.3426, pruned_loss=0.08691, over 28963.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3645, pruned_loss=0.1141, over 5665558.82 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3426, pruned_loss=0.08852, over 5629413.95 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3662, pruned_loss=0.1162, over 5664954.82 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:27:33,095 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 08:27:58,775 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 08:28:06,921 INFO [train.py:968] (0/2) Epoch 26, batch 8050, giga_loss[loss=0.3304, simple_loss=0.3924, pruned_loss=0.1342, over 28862.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3646, pruned_loss=0.1134, over 5680422.34 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3433, pruned_loss=0.08902, over 5637789.59 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.366, pruned_loss=0.1155, over 5673699.24 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:28:14,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.550e+03 1.907e+03 2.541e+03 9.251e+03, threshold=3.814e+03, percent-clipped=5.0
+2023-03-13 08:28:38,022 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5024, 1.8036, 1.4363, 1.6069], device='cuda:0'), covar=tensor([0.2389, 0.2448, 0.2696, 0.2194], device='cuda:0'), in_proj_covar=tensor([0.1570, 0.1133, 0.1386, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 08:28:38,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6758, 4.7902, 1.8276, 1.8361], device='cuda:0'), covar=tensor([0.0999, 0.0308, 0.0861, 0.1277], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0566, 0.0402, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 08:28:51,633 INFO [train.py:968] (0/2) Epoch 26, batch 8100, giga_loss[loss=0.2977, simple_loss=0.3727, pruned_loss=0.1113, over 28948.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3646, pruned_loss=0.1135, over 5684496.20 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3437, pruned_loss=0.08916, over 5643626.48 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.366, pruned_loss=0.1158, over 5674967.24 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:29:03,421 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1147169.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:29:08,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1147172.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:29:35,616 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1147201.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:29:36,812 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1147203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:29:40,470 INFO [train.py:968] (0/2) Epoch 26, batch 8150, giga_loss[loss=0.3067, simple_loss=0.3694, pruned_loss=0.122, over 28631.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3678, pruned_loss=0.1172, over 5669312.61 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3433, pruned_loss=0.08893, over 5646225.01 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.37, pruned_loss=0.12, over 5659846.40 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:29:49,282 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.875e+03 2.415e+03 3.610e+03 9.342e+03, threshold=4.830e+03, percent-clipped=19.0
+2023-03-13 08:29:58,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6588, 2.0232, 1.3979, 1.6280], device='cuda:0'), covar=tensor([0.0991, 0.0537, 0.1006, 0.1012], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0450, 0.0523, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 08:30:23,629 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-13 08:30:29,963 INFO [train.py:968] (0/2) Epoch 26, batch 8200, giga_loss[loss=0.2965, simple_loss=0.3638, pruned_loss=0.1146, over 28891.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3697, pruned_loss=0.1204, over 5664446.92 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3436, pruned_loss=0.089, over 5654139.58 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3718, pruned_loss=0.1234, over 5650334.22 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:31:18,286 INFO [train.py:968] (0/2) Epoch 26, batch 8250, giga_loss[loss=0.3973, simple_loss=0.4286, pruned_loss=0.183, over 26452.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3697, pruned_loss=0.1211, over 5675143.35 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08884, over 5658247.65 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3722, pruned_loss=0.1246, over 5660465.08 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:31:24,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.355e+02 1.913e+03 2.425e+03 3.361e+03 8.140e+03, threshold=4.850e+03, percent-clipped=9.0
+2023-03-13 08:31:39,564 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-13 08:31:55,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1147344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:32:07,309 INFO [train.py:968] (0/2) Epoch 26, batch 8300, giga_loss[loss=0.3367, simple_loss=0.387, pruned_loss=0.1432, over 28260.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3707, pruned_loss=0.1221, over 5664658.55 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.344, pruned_loss=0.08923, over 5653812.76 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3727, pruned_loss=0.1251, over 5656724.69 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:32:51,041 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 08:32:51,362 INFO [train.py:968] (0/2) Epoch 26, batch 8350, giga_loss[loss=0.2678, simple_loss=0.3364, pruned_loss=0.09959, over 28828.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3685, pruned_loss=0.1205, over 5667882.85 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3438, pruned_loss=0.08908, over 5658041.03 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.371, pruned_loss=0.1239, over 5658050.58 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:32:56,862 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.565e+02 1.732e+03 2.196e+03 2.918e+03 6.525e+03, threshold=4.392e+03, percent-clipped=2.0
+2023-03-13 08:33:28,943 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-13 08:33:31,260 INFO [train.py:968] (0/2) Epoch 26, batch 8400, giga_loss[loss=0.2975, simple_loss=0.3693, pruned_loss=0.1128, over 28570.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3672, pruned_loss=0.118, over 5682516.38 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3432, pruned_loss=0.08884, over 5663924.88 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3704, pruned_loss=0.1217, over 5669798.58 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:34:00,200 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1147485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:34:02,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4547, 1.7807, 1.6416, 1.5614], device='cuda:0'), covar=tensor([0.2220, 0.2102, 0.2394, 0.2143], device='cuda:0'), in_proj_covar=tensor([0.0496, 0.0751, 0.0721, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 08:34:02,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1147487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:34:05,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1147490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:34:19,815 INFO [train.py:968] (0/2) Epoch 26, batch 8450, libri_loss[loss=0.2575, simple_loss=0.3383, pruned_loss=0.0883, over 29581.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3662, pruned_loss=0.1168, over 5679866.75 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.343, pruned_loss=0.08876, over 5667708.73 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3692, pruned_loss=0.1202, over 5666588.43 frames. ], batch size: 76, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:34:26,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.236e+03 1.725e+03 2.306e+03 3.477e+03 1.241e+04, threshold=4.613e+03, percent-clipped=20.0
+2023-03-13 08:34:29,080 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1147519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:34:42,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0980, 2.3062, 1.6608, 1.9201], device='cuda:0'), covar=tensor([0.0985, 0.0647, 0.1005, 0.1099], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0449, 0.0521, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 08:35:01,286 INFO [train.py:968] (0/2) Epoch 26, batch 8500, giga_loss[loss=0.3029, simple_loss=0.3452, pruned_loss=0.1303, over 23828.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3638, pruned_loss=0.1145, over 5675701.77 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.344, pruned_loss=0.08937, over 5662204.74 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3664, pruned_loss=0.118, over 5669801.40 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:35:19,859 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 08:35:21,937 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1147578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:35:34,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3945, 1.9907, 1.5653, 1.5738], device='cuda:0'), covar=tensor([0.0764, 0.0305, 0.0309, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 08:35:47,141 INFO [train.py:968] (0/2) Epoch 26, batch 8550, giga_loss[loss=0.3285, simple_loss=0.3639, pruned_loss=0.1466, over 24009.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3626, pruned_loss=0.1145, over 5666774.64 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3441, pruned_loss=0.08943, over 5657332.93 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.365, pruned_loss=0.1178, over 5666652.33 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:35:54,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.648e+03 2.157e+03 2.820e+03 7.616e+03, threshold=4.315e+03, percent-clipped=6.0
+2023-03-13 08:36:37,746 INFO [train.py:968] (0/2) Epoch 26, batch 8600, giga_loss[loss=0.3451, simple_loss=0.4038, pruned_loss=0.1432, over 28765.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3642, pruned_loss=0.1163, over 5666963.40 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3444, pruned_loss=0.08968, over 5661220.72 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.366, pruned_loss=0.119, over 5663433.14 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:36:58,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4313, 1.7330, 1.3790, 1.4341], device='cuda:0'), covar=tensor([0.2525, 0.2536, 0.2889, 0.2222], device='cuda:0'), in_proj_covar=tensor([0.1570, 0.1131, 0.1385, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 08:37:05,799 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5218, 1.5445, 1.7312, 1.4142], device='cuda:0'), covar=tensor([0.1279, 0.1991, 0.1109, 0.1454], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0713, 0.0969, 0.0868], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 08:37:26,566 INFO [train.py:968] (0/2) Epoch 26, batch 8650, giga_loss[loss=0.2945, simple_loss=0.367, pruned_loss=0.111, over 28533.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3668, pruned_loss=0.1172, over 5674551.66 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3447, pruned_loss=0.08979, over 5668290.92 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3686, pruned_loss=0.12, over 5665699.36 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:37:32,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.688e+03 2.121e+03 2.800e+03 6.600e+03, threshold=4.242e+03, percent-clipped=8.0
+2023-03-13 08:37:40,972 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1147721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:37:43,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1147724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:38:12,034 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1147753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:38:15,722 INFO [train.py:968] (0/2) Epoch 26, batch 8700, giga_loss[loss=0.3175, simple_loss=0.3879, pruned_loss=0.1236, over 28210.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3716, pruned_loss=0.1177, over 5669764.27 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3449, pruned_loss=0.08982, over 5662241.73 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3731, pruned_loss=0.1203, over 5667415.13 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:39:02,069 INFO [train.py:968] (0/2) Epoch 26, batch 8750, giga_loss[loss=0.298, simple_loss=0.3683, pruned_loss=0.1139, over 28932.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3736, pruned_loss=0.1182, over 5675389.56 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3448, pruned_loss=0.08973, over 5663808.90 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3754, pruned_loss=0.1208, over 5672132.21 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:39:09,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.572e+03 1.905e+03 2.571e+03 7.601e+03, threshold=3.810e+03, percent-clipped=4.0
+2023-03-13 08:39:43,758 INFO [train.py:968] (0/2) Epoch 26, batch 8800, giga_loss[loss=0.3588, simple_loss=0.4023, pruned_loss=0.1576, over 28724.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3745, pruned_loss=0.1195, over 5681621.48 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3448, pruned_loss=0.08974, over 5668287.07 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3766, pruned_loss=0.1221, over 5675117.43 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:39:48,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1147860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:40:15,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6731, 1.9515, 1.3728, 1.5911], device='cuda:0'), covar=tensor([0.0963, 0.0598, 0.0977, 0.1176], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0450, 0.0522, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 08:40:28,998 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-13 08:40:29,414 INFO [train.py:968] (0/2) Epoch 26, batch 8850, giga_loss[loss=0.2447, simple_loss=0.321, pruned_loss=0.08426, over 28446.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3742, pruned_loss=0.1202, over 5686184.73 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3446, pruned_loss=0.08966, over 5673250.09 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3765, pruned_loss=0.1229, over 5676934.14 frames. ], batch size: 60, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:40:39,323 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.730e+03 2.240e+03 3.019e+03 7.001e+03, threshold=4.481e+03, percent-clipped=11.0
+2023-03-13 08:40:48,659 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 08:40:56,369 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1147935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:41:14,861 INFO [train.py:968] (0/2) Epoch 26, batch 8900, giga_loss[loss=0.2939, simple_loss=0.3562, pruned_loss=0.1158, over 28777.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3738, pruned_loss=0.1208, over 5686474.54 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3446, pruned_loss=0.08967, over 5678557.42 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3764, pruned_loss=0.1237, over 5674713.85 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:41:56,562 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1148000.pt
+2023-03-13 08:42:01,612 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1148003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:42:03,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1148006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:42:04,191 INFO [train.py:968] (0/2) Epoch 26, batch 8950, giga_loss[loss=0.3151, simple_loss=0.3816, pruned_loss=0.1244, over 28667.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3722, pruned_loss=0.1201, over 5688168.69 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3446, pruned_loss=0.0897, over 5683349.73 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3748, pruned_loss=0.1231, over 5675037.37 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:42:11,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.686e+03 2.047e+03 2.562e+03 5.489e+03, threshold=4.095e+03, percent-clipped=3.0
+2023-03-13 08:42:28,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1148035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:42:49,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3537, 1.4198, 1.2450, 1.4871], device='cuda:0'), covar=tensor([0.0646, 0.0426, 0.0343, 0.0715], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:0')
+2023-03-13 08:42:49,616 INFO [train.py:968] (0/2) Epoch 26, batch 9000, giga_loss[loss=0.4626, simple_loss=0.4603, pruned_loss=0.2324, over 26603.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3714, pruned_loss=0.1208, over 5682731.96 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3452, pruned_loss=0.09007, over 5682108.84 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3739, pruned_loss=0.1238, over 5672895.31 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:42:49,620 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 08:42:58,105 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2061, simple_loss=0.3142, pruned_loss=0.04894, over 944034.00 frames. 
+2023-03-13 08:42:58,106 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 08:43:41,558 INFO [train.py:968] (0/2) Epoch 26, batch 9050, giga_loss[loss=0.3224, simple_loss=0.3609, pruned_loss=0.1419, over 23609.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3683, pruned_loss=0.1186, over 5688726.19 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3449, pruned_loss=0.08975, over 5691209.79 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3717, pruned_loss=0.1226, over 5672269.85 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:43:48,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.827e+03 2.608e+03 3.962e+03 1.800e+04, threshold=5.216e+03, percent-clipped=22.0
+2023-03-13 08:44:28,460 INFO [train.py:968] (0/2) Epoch 26, batch 9100, giga_loss[loss=0.2938, simple_loss=0.3627, pruned_loss=0.1124, over 28977.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3674, pruned_loss=0.1177, over 5690470.69 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3453, pruned_loss=0.0901, over 5695750.94 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3704, pruned_loss=0.1216, over 5673091.71 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:45:14,672 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 08:45:15,676 INFO [train.py:968] (0/2) Epoch 26, batch 9150, giga_loss[loss=0.3908, simple_loss=0.4263, pruned_loss=0.1777, over 27638.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3691, pruned_loss=0.1199, over 5683101.07 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.345, pruned_loss=0.08987, over 5698761.13 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3725, pruned_loss=0.1241, over 5665956.58 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:45:24,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.145e+03 1.813e+03 2.258e+03 2.936e+03 5.934e+03, threshold=4.517e+03, percent-clipped=2.0
+2023-03-13 08:45:25,328 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:45:58,073 INFO [train.py:968] (0/2) Epoch 26, batch 9200, giga_loss[loss=0.3203, simple_loss=0.3662, pruned_loss=0.1372, over 28521.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3662, pruned_loss=0.1179, over 5691318.83 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3445, pruned_loss=0.08946, over 5703379.31 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3702, pruned_loss=0.1227, over 5672670.37 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:46:47,856 INFO [train.py:968] (0/2) Epoch 26, batch 9250, giga_loss[loss=0.3082, simple_loss=0.3719, pruned_loss=0.1223, over 28864.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3673, pruned_loss=0.1193, over 5692290.54 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3441, pruned_loss=0.08921, over 5706729.93 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3712, pruned_loss=0.1238, over 5674334.79 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:46:50,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1148310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:46:54,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.255e+03 1.774e+03 2.281e+03 3.066e+03 7.809e+03, threshold=4.561e+03, percent-clipped=7.0
+2023-03-13 08:47:17,969 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:47:33,924 INFO [train.py:968] (0/2) Epoch 26, batch 9300, libri_loss[loss=0.2973, simple_loss=0.3739, pruned_loss=0.1104, over 26138.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3697, pruned_loss=0.1201, over 5674755.59 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3446, pruned_loss=0.08938, over 5700693.73 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3732, pruned_loss=0.1245, over 5665406.43 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:48:20,996 INFO [train.py:968] (0/2) Epoch 26, batch 9350, libri_loss[loss=0.3162, simple_loss=0.3927, pruned_loss=0.1198, over 27960.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3723, pruned_loss=0.122, over 5675426.44 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3448, pruned_loss=0.08953, over 5700150.26 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3749, pruned_loss=0.1255, over 5668532.53 frames. ], batch size: 116, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:48:30,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.100e+02 1.755e+03 2.297e+03 3.049e+03 1.167e+04, threshold=4.593e+03, percent-clipped=8.0
+2023-03-13 08:49:02,006 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1148453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:49:03,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0490, 2.7271, 2.3149, 2.2312], device='cuda:0'), covar=tensor([0.2605, 0.2180, 0.2514, 0.2613], device='cuda:0'), in_proj_covar=tensor([0.0498, 0.0754, 0.0724, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 08:49:04,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1148456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:49:04,699 INFO [train.py:968] (0/2) Epoch 26, batch 9400, giga_loss[loss=0.2951, simple_loss=0.3641, pruned_loss=0.113, over 28748.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3707, pruned_loss=0.1211, over 5670679.87 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3452, pruned_loss=0.0897, over 5696965.69 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.373, pruned_loss=0.1244, over 5667422.60 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:49:27,838 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1148485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:49:50,064 INFO [train.py:968] (0/2) Epoch 26, batch 9450, giga_loss[loss=0.2827, simple_loss=0.3588, pruned_loss=0.1033, over 28852.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3717, pruned_loss=0.119, over 5681898.20 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3453, pruned_loss=0.08988, over 5701822.75 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3741, pruned_loss=0.1223, over 5674546.97 frames. ], batch size: 66, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:49:59,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.877e+02 1.581e+03 1.996e+03 2.663e+03 1.048e+04, threshold=3.993e+03, percent-clipped=6.0
+2023-03-13 08:50:01,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2238, 2.5946, 1.8683, 2.4038], device='cuda:0'), covar=tensor([0.0897, 0.0527, 0.0921, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0448, 0.0519, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 08:50:34,428 INFO [train.py:968] (0/2) Epoch 26, batch 9500, giga_loss[loss=0.2896, simple_loss=0.3747, pruned_loss=0.1022, over 28874.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3723, pruned_loss=0.1176, over 5681034.44 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3448, pruned_loss=0.08969, over 5705767.89 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3751, pruned_loss=0.1208, over 5671155.60 frames. ], batch size: 112, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:51:11,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1148593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:51:20,940 INFO [train.py:968] (0/2) Epoch 26, batch 9550, giga_loss[loss=0.357, simple_loss=0.4169, pruned_loss=0.1486, over 28996.00 frames. ], tot_loss[loss=0.308, simple_loss=0.376, pruned_loss=0.12, over 5671070.26 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3452, pruned_loss=0.09001, over 5699490.76 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3783, pruned_loss=0.1228, over 5669102.19 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:51:31,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.867e+02 1.644e+03 2.192e+03 3.299e+03 7.538e+03, threshold=4.383e+03, percent-clipped=17.0
+2023-03-13 08:52:06,553 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148653.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 08:52:08,929 INFO [train.py:968] (0/2) Epoch 26, batch 9600, giga_loss[loss=0.316, simple_loss=0.382, pruned_loss=0.1251, over 28740.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3781, pruned_loss=0.1227, over 5673684.90 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3454, pruned_loss=0.09008, over 5700648.30 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.38, pruned_loss=0.1249, over 5671073.20 frames. ], batch size: 243, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:52:56,127 INFO [train.py:968] (0/2) Epoch 26, batch 9650, giga_loss[loss=0.2991, simple_loss=0.3635, pruned_loss=0.1173, over 28173.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3788, pruned_loss=0.1243, over 5671234.59 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3454, pruned_loss=0.09022, over 5702257.69 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.381, pruned_loss=0.1267, over 5667278.55 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:53:06,200 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1148717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:53:06,618 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.168e+03 1.968e+03 2.932e+03 3.963e+03 1.476e+04, threshold=5.865e+03, percent-clipped=22.0
+2023-03-13 08:53:26,520 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1148736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:53:28,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1148739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:53:31,921 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148744.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 08:53:42,826 INFO [train.py:968] (0/2) Epoch 26, batch 9700, giga_loss[loss=0.2977, simple_loss=0.3742, pruned_loss=0.1106, over 28782.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3782, pruned_loss=0.1245, over 5655929.58 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3458, pruned_loss=0.09046, over 5697339.70 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3804, pruned_loss=0.1271, over 5656584.35 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:53:43,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4231, 1.7789, 1.4100, 1.3236], device='cuda:0'), covar=tensor([0.2637, 0.2693, 0.3165, 0.2405], device='cuda:0'), in_proj_covar=tensor([0.1573, 0.1133, 0.1391, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 08:53:46,793 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-13 08:53:51,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1148768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:54:16,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-13 08:54:24,241 INFO [train.py:968] (0/2) Epoch 26, batch 9750, giga_loss[loss=0.4037, simple_loss=0.4444, pruned_loss=0.1815, over 28616.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3758, pruned_loss=0.1223, over 5659667.48 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3452, pruned_loss=0.09032, over 5700793.78 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3788, pruned_loss=0.1252, over 5656037.05 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:54:34,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.420e+02 1.688e+03 2.422e+03 3.367e+03 7.171e+03, threshold=4.844e+03, percent-clipped=5.0
+2023-03-13 08:54:50,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6155, 2.0359, 1.8732, 1.7814], device='cuda:0'), covar=tensor([0.2299, 0.2246, 0.2581, 0.2542], device='cuda:0'), in_proj_covar=tensor([0.0498, 0.0755, 0.0724, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 08:55:08,354 INFO [train.py:968] (0/2) Epoch 26, batch 9800, giga_loss[loss=0.309, simple_loss=0.3821, pruned_loss=0.118, over 28942.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.374, pruned_loss=0.1192, over 5668353.07 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3445, pruned_loss=0.09004, over 5706662.13 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3779, pruned_loss=0.1227, over 5658844.73 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:55:12,201 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1148860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:55:14,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1148863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:55:23,918 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9935, 1.2991, 1.0478, 0.1935], device='cuda:0'), covar=tensor([0.3889, 0.3180, 0.4258, 0.6905], device='cuda:0'), in_proj_covar=tensor([0.1806, 0.1693, 0.1629, 0.1466], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 08:55:39,216 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1148892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:55:51,910 INFO [train.py:968] (0/2) Epoch 26, batch 9850, giga_loss[loss=0.3188, simple_loss=0.3828, pruned_loss=0.1274, over 28880.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3745, pruned_loss=0.1188, over 5679095.45 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3446, pruned_loss=0.09022, over 5712586.25 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3783, pruned_loss=0.1222, over 5665064.10 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:56:00,418 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:56:00,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.435e+02 1.543e+03 1.946e+03 2.386e+03 4.474e+03, threshold=3.892e+03, percent-clipped=0.0
+2023-03-13 08:56:37,799 INFO [train.py:968] (0/2) Epoch 26, batch 9900, giga_loss[loss=0.3325, simple_loss=0.3896, pruned_loss=0.1377, over 28642.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3748, pruned_loss=0.1191, over 5664770.18 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3448, pruned_loss=0.09017, over 5704477.40 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3789, pruned_loss=0.123, over 5659368.72 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:57:00,714 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:57:25,128 INFO [train.py:968] (0/2) Epoch 26, batch 9950, giga_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1125, over 28627.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3735, pruned_loss=0.1186, over 5657730.13 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3443, pruned_loss=0.09001, over 5698192.64 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3779, pruned_loss=0.1226, over 5658790.37 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:57:35,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.670e+03 2.206e+03 2.847e+03 8.345e+03, threshold=4.411e+03, percent-clipped=10.0
+2023-03-13 08:57:44,070 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149028.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:58:11,941 INFO [train.py:968] (0/2) Epoch 26, batch 10000, giga_loss[loss=0.3022, simple_loss=0.373, pruned_loss=0.1157, over 28576.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3717, pruned_loss=0.1184, over 5657413.12 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3443, pruned_loss=0.09016, over 5695752.77 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3761, pruned_loss=0.1223, over 5658891.35 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:58:53,588 INFO [train.py:968] (0/2) Epoch 26, batch 10050, giga_loss[loss=0.2991, simple_loss=0.364, pruned_loss=0.1171, over 28749.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3702, pruned_loss=0.1178, over 5662139.72 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3441, pruned_loss=0.08998, over 5700316.62 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3751, pruned_loss=0.1224, over 5657680.66 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:58:53,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9063, 2.2752, 2.1045, 1.6989], device='cuda:0'), covar=tensor([0.3094, 0.2316, 0.2749, 0.3116], device='cuda:0'), in_proj_covar=tensor([0.2043, 0.1990, 0.1923, 0.2057], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 08:59:05,948 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-13 08:59:06,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.639e+03 2.245e+03 3.339e+03 1.216e+04, threshold=4.490e+03, percent-clipped=11.0
+2023-03-13 08:59:07,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149119.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:59:29,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4423, 1.5748, 1.2328, 1.1476], device='cuda:0'), covar=tensor([0.0992, 0.0556, 0.1073, 0.1056], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0451, 0.0523, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 08:59:41,192 INFO [train.py:968] (0/2) Epoch 26, batch 10100, giga_loss[loss=0.2749, simple_loss=0.3474, pruned_loss=0.1011, over 28756.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3683, pruned_loss=0.1168, over 5664079.20 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3443, pruned_loss=0.09014, over 5696766.99 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3728, pruned_loss=0.1213, over 5662525.13 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:59:53,524 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149171.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:59:55,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149174.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:00:25,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149203.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:00:28,826 INFO [train.py:968] (0/2) Epoch 26, batch 10150, giga_loss[loss=0.2833, simple_loss=0.3463, pruned_loss=0.1102, over 28986.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3677, pruned_loss=0.1172, over 5670506.95 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.09043, over 5701490.63 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3715, pruned_loss=0.1212, over 5664225.93 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:00:38,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.746e+03 2.259e+03 2.784e+03 5.990e+03, threshold=4.519e+03, percent-clipped=6.0
+2023-03-13 09:01:02,463 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 09:01:15,758 INFO [train.py:968] (0/2) Epoch 26, batch 10200, giga_loss[loss=0.2792, simple_loss=0.3494, pruned_loss=0.1045, over 28905.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3682, pruned_loss=0.1184, over 5670481.55 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.09043, over 5701490.63 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3712, pruned_loss=0.1215, over 5665592.95 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:01:22,650 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149262.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:01:25,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149265.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:01:34,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8639, 1.1877, 1.2812, 1.0350], device='cuda:0'), covar=tensor([0.2100, 0.1457, 0.2410, 0.1747], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0755, 0.0725, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 09:01:49,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:01:51,162 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149294.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:01:55,296 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1149299.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:02:01,806 INFO [train.py:968] (0/2) Epoch 26, batch 10250, giga_loss[loss=0.2775, simple_loss=0.351, pruned_loss=0.102, over 29024.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3655, pruned_loss=0.1157, over 5663141.54 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.09041, over 5704297.03 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3687, pruned_loss=0.1191, over 5655697.01 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:02:12,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.140e+03 1.608e+03 1.961e+03 2.583e+03 6.467e+03, threshold=3.922e+03, percent-clipped=4.0
+2023-03-13 09:02:48,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:02:50,260 INFO [train.py:968] (0/2) Epoch 26, batch 10300, giga_loss[loss=0.3012, simple_loss=0.3764, pruned_loss=0.113, over 28822.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3637, pruned_loss=0.1139, over 5663658.69 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3449, pruned_loss=0.09052, over 5707170.08 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3663, pruned_loss=0.1168, over 5654763.60 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:03:01,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1856, 3.2686, 1.3800, 1.3969], device='cuda:0'), covar=tensor([0.1106, 0.0440, 0.0981, 0.1520], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0567, 0.0403, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 09:03:07,594 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 09:03:32,448 INFO [train.py:968] (0/2) Epoch 26, batch 10350, giga_loss[loss=0.272, simple_loss=0.3517, pruned_loss=0.09611, over 28884.00 frames. ], tot_loss[loss=0.292, simple_loss=0.361, pruned_loss=0.1115, over 5672297.79 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3441, pruned_loss=0.09011, over 5715254.66 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3648, pruned_loss=0.1153, over 5655615.38 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:03:42,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.759e+02 1.574e+03 1.902e+03 2.509e+03 7.142e+03, threshold=3.803e+03, percent-clipped=8.0
+2023-03-13 09:03:48,028 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1149425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:03:57,723 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:03:59,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:04:19,987 INFO [train.py:968] (0/2) Epoch 26, batch 10400, giga_loss[loss=0.295, simple_loss=0.3566, pruned_loss=0.1167, over 28225.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3592, pruned_loss=0.1108, over 5659926.20 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3444, pruned_loss=0.09036, over 5696860.75 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3626, pruned_loss=0.1145, over 5660711.95 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:04:31,974 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:05:03,515 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:05:07,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:05:12,705 INFO [train.py:968] (0/2) Epoch 26, batch 10450, giga_loss[loss=0.3976, simple_loss=0.4321, pruned_loss=0.1815, over 27497.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3584, pruned_loss=0.1117, over 5657518.84 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3444, pruned_loss=0.09036, over 5696860.75 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.361, pruned_loss=0.1146, over 5658130.40 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:05:26,423 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.313e+03 1.890e+03 2.300e+03 3.303e+03 8.582e+03, threshold=4.599e+03, percent-clipped=13.0
+2023-03-13 09:05:33,173 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:05:55,905 INFO [train.py:968] (0/2) Epoch 26, batch 10500, giga_loss[loss=0.294, simple_loss=0.367, pruned_loss=0.1105, over 28737.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3607, pruned_loss=0.1125, over 5662476.77 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3447, pruned_loss=0.09048, over 5699299.12 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3632, pruned_loss=0.1157, over 5659227.55 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:06:41,674 INFO [train.py:968] (0/2) Epoch 26, batch 10550, libri_loss[loss=0.2541, simple_loss=0.3429, pruned_loss=0.0827, over 28735.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3625, pruned_loss=0.1129, over 5669480.80 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3449, pruned_loss=0.09054, over 5704391.76 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3648, pruned_loss=0.116, over 5661580.97 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:06:53,128 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.597e+03 2.002e+03 2.828e+03 9.920e+03, threshold=4.005e+03, percent-clipped=12.0
+2023-03-13 09:06:57,529 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-13 09:07:16,898 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1149645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:07:29,432 INFO [train.py:968] (0/2) Epoch 26, batch 10600, giga_loss[loss=0.2789, simple_loss=0.3498, pruned_loss=0.104, over 28840.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3642, pruned_loss=0.1145, over 5644136.65 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3452, pruned_loss=0.09058, over 5708657.52 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3662, pruned_loss=0.1174, over 5633542.51 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:07:32,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1149660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:07:46,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149674.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:07:46,571 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 09:08:17,253 INFO [train.py:968] (0/2) Epoch 26, batch 10650, giga_loss[loss=0.3305, simple_loss=0.3873, pruned_loss=0.1368, over 28732.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3656, pruned_loss=0.1159, over 5645711.28 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3454, pruned_loss=0.09054, over 5713836.55 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3675, pruned_loss=0.1189, over 5630728.50 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:08:29,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.543e+03 2.010e+03 2.890e+03 9.130e+03, threshold=4.020e+03, percent-clipped=13.0
+2023-03-13 09:09:05,334 INFO [train.py:968] (0/2) Epoch 26, batch 10700, giga_loss[loss=0.3454, simple_loss=0.4026, pruned_loss=0.1441, over 28737.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3672, pruned_loss=0.1174, over 5639128.53 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3459, pruned_loss=0.0907, over 5706108.28 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3685, pruned_loss=0.12, over 5632670.89 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:09:48,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:09:55,239 INFO [train.py:968] (0/2) Epoch 26, batch 10750, giga_loss[loss=0.284, simple_loss=0.3529, pruned_loss=0.1075, over 28393.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3675, pruned_loss=0.1171, over 5647408.72 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3461, pruned_loss=0.09096, over 5706608.29 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3689, pruned_loss=0.1195, over 5640064.19 frames. ], batch size: 65, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:10:05,519 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149817.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:10:08,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149820.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:10:08,540 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.824e+02 1.816e+03 2.350e+03 3.483e+03 8.103e+03, threshold=4.699e+03, percent-clipped=18.0
+2023-03-13 09:10:36,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149849.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:10:42,956 INFO [train.py:968] (0/2) Epoch 26, batch 10800, giga_loss[loss=0.3197, simple_loss=0.3802, pruned_loss=0.1296, over 28564.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3695, pruned_loss=0.1187, over 5652457.90 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.346, pruned_loss=0.09083, over 5708769.35 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3708, pruned_loss=0.1209, over 5644245.98 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:10:54,101 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1149868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:11:32,053 INFO [train.py:968] (0/2) Epoch 26, batch 10850, giga_loss[loss=0.3172, simple_loss=0.3765, pruned_loss=0.1289, over 28969.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3719, pruned_loss=0.1214, over 5654362.92 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3458, pruned_loss=0.09074, over 5710504.05 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3735, pruned_loss=0.1236, over 5645694.56 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:11:37,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3918, 1.5881, 1.5734, 1.4051], device='cuda:0'), covar=tensor([0.2058, 0.1985, 0.2463, 0.2118], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0757, 0.0726, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 09:11:45,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.129e+03 1.750e+03 2.370e+03 3.162e+03 7.048e+03, threshold=4.740e+03, percent-clipped=9.0
+2023-03-13 09:12:08,095 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:12:10,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:12:11,441 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-13 09:12:16,563 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3362, 1.7324, 1.2618, 1.4840], device='cuda:0'), covar=tensor([0.2752, 0.2735, 0.3267, 0.2427], device='cuda:0'), in_proj_covar=tensor([0.1568, 0.1131, 0.1385, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 09:12:20,456 INFO [train.py:968] (0/2) Epoch 26, batch 10900, giga_loss[loss=0.3398, simple_loss=0.394, pruned_loss=0.1428, over 28624.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3724, pruned_loss=0.1217, over 5645219.21 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3457, pruned_loss=0.09066, over 5703906.78 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3741, pruned_loss=0.124, over 5643231.29 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:12:36,425 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:13:01,021 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1150000.pt
+2023-03-13 09:13:08,235 INFO [train.py:968] (0/2) Epoch 26, batch 10950, giga_loss[loss=0.2945, simple_loss=0.3641, pruned_loss=0.1125, over 28693.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3713, pruned_loss=0.1193, over 5646716.38 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3458, pruned_loss=0.09059, over 5707793.66 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3733, pruned_loss=0.122, over 5640283.34 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:13:21,173 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1150020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:13:22,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.158e+03 1.708e+03 2.207e+03 2.784e+03 6.824e+03, threshold=4.415e+03, percent-clipped=9.0
+2023-03-13 09:13:33,890 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1150035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:13:56,812 INFO [train.py:968] (0/2) Epoch 26, batch 11000, libri_loss[loss=0.2512, simple_loss=0.3315, pruned_loss=0.08551, over 29525.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3689, pruned_loss=0.1179, over 5652123.82 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3453, pruned_loss=0.09034, over 5711333.18 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3716, pruned_loss=0.1209, over 5642442.61 frames. ], batch size: 81, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:14:48,128 INFO [train.py:968] (0/2) Epoch 26, batch 11050, giga_loss[loss=0.3069, simple_loss=0.3679, pruned_loss=0.1229, over 28692.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3674, pruned_loss=0.1174, over 5665527.49 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3455, pruned_loss=0.09052, over 5714001.45 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3697, pruned_loss=0.1201, over 5654633.76 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:15:06,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.134e+03 1.771e+03 2.281e+03 3.519e+03 6.986e+03, threshold=4.563e+03, percent-clipped=10.0
+2023-03-13 09:15:31,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5027, 1.6045, 1.7417, 1.3209], device='cuda:0'), covar=tensor([0.1713, 0.2549, 0.1394, 0.1715], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0719, 0.0975, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:0')
+2023-03-13 09:15:43,675 INFO [train.py:968] (0/2) Epoch 26, batch 11100, giga_loss[loss=0.3023, simple_loss=0.3691, pruned_loss=0.1177, over 27880.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3668, pruned_loss=0.1176, over 5660335.66 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3454, pruned_loss=0.09043, over 5717532.03 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3692, pruned_loss=0.1204, over 5647362.15 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:15:48,881 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1150163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:15:52,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1150166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:16:00,380 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1150178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:16:03,468 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1150181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:16:16,192 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1150195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:16:18,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.94 vs. limit=5.0
+2023-03-13 09:16:28,140 INFO [train.py:968] (0/2) Epoch 26, batch 11150, giga_loss[loss=0.2869, simple_loss=0.3554, pruned_loss=0.1092, over 28890.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3653, pruned_loss=0.1167, over 5660071.49 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3459, pruned_loss=0.09079, over 5703068.68 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3674, pruned_loss=0.1194, over 5660962.74 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:16:31,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1150210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:16:40,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.209e+03 1.796e+03 2.251e+03 3.758e+03 1.040e+04, threshold=4.501e+03, percent-clipped=13.0
+2023-03-13 09:16:59,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1150243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:17:11,522 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4207, 1.4853, 3.4785, 3.3139], device='cuda:0'), covar=tensor([0.1384, 0.2575, 0.0509, 0.1057], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0671, 0.0991, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 09:17:12,540 INFO [train.py:968] (0/2) Epoch 26, batch 11200, giga_loss[loss=0.2661, simple_loss=0.3396, pruned_loss=0.09629, over 28274.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3656, pruned_loss=0.1175, over 5658252.80 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3461, pruned_loss=0.09094, over 5706197.31 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3675, pruned_loss=0.1201, over 5655176.95 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:17:25,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6292, 1.9044, 1.6244, 1.5600], device='cuda:0'), covar=tensor([0.2070, 0.1999, 0.2088, 0.1979], device='cuda:0'), in_proj_covar=tensor([0.1566, 0.1129, 0.1384, 0.1004], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 09:17:51,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4499, 1.1656, 3.7107, 3.2298], device='cuda:0'), covar=tensor([0.1643, 0.2953, 0.0592, 0.1026], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0670, 0.0990, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 09:17:56,438 INFO [train.py:968] (0/2) Epoch 26, batch 11250, giga_loss[loss=0.2945, simple_loss=0.3618, pruned_loss=0.1136, over 28590.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3646, pruned_loss=0.1165, over 5672071.74 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3455, pruned_loss=0.09055, over 5714720.09 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3675, pruned_loss=0.12, over 5660057.13 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:18:08,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.562e+03 2.063e+03 3.175e+03 1.327e+04, threshold=4.126e+03, percent-clipped=11.0
+2023-03-13 09:18:43,436 INFO [train.py:968] (0/2) Epoch 26, batch 11300, giga_loss[loss=0.2998, simple_loss=0.3688, pruned_loss=0.1154, over 28601.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3661, pruned_loss=0.118, over 5664140.17 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3455, pruned_loss=0.09061, over 5707229.44 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3687, pruned_loss=0.121, over 5660013.43 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:19:10,843 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1150386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:19:15,918 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1150389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:19:29,235 INFO [train.py:968] (0/2) Epoch 26, batch 11350, giga_loss[loss=0.3604, simple_loss=0.4162, pruned_loss=0.1524, over 28951.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3696, pruned_loss=0.1206, over 5671818.37 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3457, pruned_loss=0.0906, over 5712391.19 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3719, pruned_loss=0.1237, over 5662966.37 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:19:39,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1150418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:19:43,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.220e+03 1.942e+03 2.523e+03 3.606e+03 7.675e+03, threshold=5.047e+03, percent-clipped=17.0
+2023-03-13 09:19:56,913 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-13 09:20:17,077 INFO [train.py:968] (0/2) Epoch 26, batch 11400, giga_loss[loss=0.2896, simple_loss=0.3587, pruned_loss=0.1102, over 28972.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3709, pruned_loss=0.1216, over 5667300.03 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3456, pruned_loss=0.09055, over 5711206.06 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3733, pruned_loss=0.1247, over 5660702.75 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:20:38,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6026, 5.1769, 1.8543, 1.9583], device='cuda:0'), covar=tensor([0.1002, 0.0206, 0.0936, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0568, 0.0403, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 09:21:06,636 INFO [train.py:968] (0/2) Epoch 26, batch 11450, giga_loss[loss=0.2818, simple_loss=0.3534, pruned_loss=0.1051, over 29076.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3695, pruned_loss=0.1217, over 5658695.63 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3446, pruned_loss=0.09006, over 5716082.96 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3728, pruned_loss=0.1253, over 5647949.72 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:21:15,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6421, 1.7917, 1.5040, 1.8181], device='cuda:0'), covar=tensor([0.2617, 0.2792, 0.3082, 0.2330], device='cuda:0'), in_proj_covar=tensor([0.1570, 0.1134, 0.1387, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 09:21:19,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.152e+03 1.744e+03 2.473e+03 3.257e+03 9.113e+03, threshold=4.947e+03, percent-clipped=5.0
+2023-03-13 09:21:53,308 INFO [train.py:968] (0/2) Epoch 26, batch 11500, giga_loss[loss=0.2541, simple_loss=0.3303, pruned_loss=0.08894, over 28846.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3692, pruned_loss=0.1216, over 5653046.23 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3449, pruned_loss=0.09028, over 5708964.24 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3721, pruned_loss=0.1248, over 5649631.86 frames. ], batch size: 112, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:22:41,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5559, 1.8412, 1.5752, 1.4217], device='cuda:0'), covar=tensor([0.2020, 0.1988, 0.2055, 0.1986], device='cuda:0'), in_proj_covar=tensor([0.1572, 0.1136, 0.1389, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 09:22:42,162 INFO [train.py:968] (0/2) Epoch 26, batch 11550, giga_loss[loss=0.3609, simple_loss=0.4049, pruned_loss=0.1585, over 27463.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.37, pruned_loss=0.1217, over 5654716.39 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.345, pruned_loss=0.09043, over 5700179.62 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3727, pruned_loss=0.1247, over 5657543.88 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:22:49,050 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-13 09:22:55,420 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.189e+03 1.996e+03 2.628e+03 4.072e+03 7.817e+03, threshold=5.255e+03, percent-clipped=13.0
+2023-03-13 09:23:03,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1150629.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:23:26,302 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-13 09:23:26,514 INFO [train.py:968] (0/2) Epoch 26, batch 11600, giga_loss[loss=0.3003, simple_loss=0.366, pruned_loss=0.1173, over 28768.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.371, pruned_loss=0.1218, over 5645285.72 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3457, pruned_loss=0.09088, over 5685415.12 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3733, pruned_loss=0.1248, over 5659567.16 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:23:46,206 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1150677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:23:56,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6013, 1.6336, 1.8612, 1.3782], device='cuda:0'), covar=tensor([0.1834, 0.2849, 0.1494, 0.1838], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0717, 0.0973, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 09:23:59,488 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1150691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:24:17,719 INFO [train.py:968] (0/2) Epoch 26, batch 11650, giga_loss[loss=0.3307, simple_loss=0.393, pruned_loss=0.1342, over 28288.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.371, pruned_loss=0.1218, over 5663170.93 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.345, pruned_loss=0.09063, over 5690354.09 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3741, pruned_loss=0.1252, over 5669304.33 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:24:30,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.251e+03 1.789e+03 2.263e+03 3.189e+03 6.793e+03, threshold=4.525e+03, percent-clipped=6.0
+2023-03-13 09:25:05,420 INFO [train.py:968] (0/2) Epoch 26, batch 11700, giga_loss[loss=0.2634, simple_loss=0.3416, pruned_loss=0.0926, over 29036.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3737, pruned_loss=0.1243, over 5663299.72 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.345, pruned_loss=0.09062, over 5693682.76 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3764, pruned_loss=0.1274, over 5664652.94 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:25:50,630 INFO [train.py:968] (0/2) Epoch 26, batch 11750, giga_loss[loss=0.3333, simple_loss=0.3887, pruned_loss=0.139, over 28238.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.372, pruned_loss=0.1226, over 5673989.35 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3446, pruned_loss=0.09036, over 5695739.53 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3754, pruned_loss=0.1262, over 5672865.41 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:26:06,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.797e+03 2.289e+03 3.364e+03 6.207e+03, threshold=4.577e+03, percent-clipped=8.0
+2023-03-13 09:26:39,234 INFO [train.py:968] (0/2) Epoch 26, batch 11800, giga_loss[loss=0.3846, simple_loss=0.4154, pruned_loss=0.1769, over 23552.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3722, pruned_loss=0.1217, over 5674295.51 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3446, pruned_loss=0.09037, over 5698643.86 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3754, pruned_loss=0.1252, over 5670374.50 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:27:26,800 INFO [train.py:968] (0/2) Epoch 26, batch 11850, giga_loss[loss=0.2977, simple_loss=0.3429, pruned_loss=0.1262, over 23547.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3709, pruned_loss=0.1202, over 5664403.22 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3444, pruned_loss=0.09024, over 5700827.09 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.374, pruned_loss=0.1235, over 5658921.37 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:27:39,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.714e+03 2.287e+03 3.460e+03 6.207e+03, threshold=4.574e+03, percent-clipped=12.0
+2023-03-13 09:28:12,768 INFO [train.py:968] (0/2) Epoch 26, batch 11900, libri_loss[loss=0.2565, simple_loss=0.3403, pruned_loss=0.08637, over 29521.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3686, pruned_loss=0.1186, over 5669463.32 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.344, pruned_loss=0.09013, over 5697023.47 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.372, pruned_loss=0.1219, over 5667950.51 frames. ], batch size: 81, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:28:50,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151004.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:28:53,547 INFO [train.py:968] (0/2) Epoch 26, batch 11950, libri_loss[loss=0.2222, simple_loss=0.3122, pruned_loss=0.06605, over 29546.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3665, pruned_loss=0.1168, over 5687652.95 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3436, pruned_loss=0.0899, over 5705030.57 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3709, pruned_loss=0.1211, over 5678231.75 frames. ], batch size: 76, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:29:07,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.139e+03 1.569e+03 2.082e+03 2.723e+03 6.522e+03, threshold=4.164e+03, percent-clipped=3.0
+2023-03-13 09:29:15,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 09:29:16,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5926, 1.8507, 1.7658, 1.5090], device='cuda:0'), covar=tensor([0.3047, 0.2406, 0.1959, 0.2478], device='cuda:0'), in_proj_covar=tensor([0.2048, 0.2001, 0.1925, 0.2060], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 09:29:33,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:29:39,510 INFO [train.py:968] (0/2) Epoch 26, batch 12000, giga_loss[loss=0.31, simple_loss=0.384, pruned_loss=0.118, over 28794.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3685, pruned_loss=0.1187, over 5661267.14 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3434, pruned_loss=0.08979, over 5699742.75 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3728, pruned_loss=0.123, over 5656885.70 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:29:39,514 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 09:29:48,477 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2056, simple_loss=0.313, pruned_loss=0.04906, over 944034.00 frames. 
+2023-03-13 09:29:48,478 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 09:29:49,482 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3446, 1.6019, 1.3236, 0.9689], device='cuda:0'), covar=tensor([0.2637, 0.2683, 0.3118, 0.2437], device='cuda:0'), in_proj_covar=tensor([0.1568, 0.1133, 0.1387, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 09:29:55,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:30:13,006 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1151083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:30:33,050 INFO [train.py:968] (0/2) Epoch 26, batch 12050, libri_loss[loss=0.2929, simple_loss=0.3763, pruned_loss=0.1048, over 29529.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3702, pruned_loss=0.1191, over 5677562.20 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3432, pruned_loss=0.08968, over 5708034.99 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3749, pruned_loss=0.1238, over 5665221.17 frames. ], batch size: 84, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:30:48,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.688e+03 2.279e+03 2.771e+03 6.133e+03, threshold=4.557e+03, percent-clipped=6.0
+2023-03-13 09:31:11,006 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1151147.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:31:14,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1151150.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:31:19,965 INFO [train.py:968] (0/2) Epoch 26, batch 12100, giga_loss[loss=0.2749, simple_loss=0.3513, pruned_loss=0.09928, over 28911.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3707, pruned_loss=0.1205, over 5675695.19 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3432, pruned_loss=0.08958, over 5711657.83 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3749, pruned_loss=0.1249, over 5662130.33 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:31:34,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0321, 2.3393, 2.2463, 1.7895], device='cuda:0'), covar=tensor([0.2787, 0.2416, 0.2548, 0.2861], device='cuda:0'), in_proj_covar=tensor([0.2045, 0.1999, 0.1921, 0.2058], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 09:31:37,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9336, 1.1393, 1.0741, 0.8562], device='cuda:0'), covar=tensor([0.2468, 0.2817, 0.1787, 0.2451], device='cuda:0'), in_proj_covar=tensor([0.2046, 0.1999, 0.1921, 0.2058], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 09:31:39,221 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1151179.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:31:55,548 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1151195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:31:57,936 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1151198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:32:06,836 INFO [train.py:968] (0/2) Epoch 26, batch 12150, giga_loss[loss=0.3151, simple_loss=0.3827, pruned_loss=0.1237, over 28530.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1235, over 5669780.63 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08965, over 5715915.58 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3772, pruned_loss=0.1275, over 5654840.14 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:32:08,673 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1151209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:32:10,823 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1151212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:32:25,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.861e+03 2.363e+03 3.614e+03 7.377e+03, threshold=4.725e+03, percent-clipped=14.0
+2023-03-13 09:32:30,192 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1151227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:32:43,116 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1151241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:32:57,776 INFO [train.py:968] (0/2) Epoch 26, batch 12200, giga_loss[loss=0.2863, simple_loss=0.3583, pruned_loss=0.1071, over 28966.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3749, pruned_loss=0.1249, over 5665624.67 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3432, pruned_loss=0.08961, over 5717772.70 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3783, pruned_loss=0.1284, over 5651875.29 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:33:45,255 INFO [train.py:968] (0/2) Epoch 26, batch 12250, giga_loss[loss=0.3101, simple_loss=0.3736, pruned_loss=0.1233, over 28908.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3749, pruned_loss=0.1245, over 5671790.06 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.343, pruned_loss=0.08951, over 5721720.17 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3783, pruned_loss=0.128, over 5656387.73 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:33:59,221 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.719e+03 2.122e+03 2.776e+03 8.046e+03, threshold=4.244e+03, percent-clipped=5.0
+2023-03-13 09:34:32,456 INFO [train.py:968] (0/2) Epoch 26, batch 12300, giga_loss[loss=0.2957, simple_loss=0.3621, pruned_loss=0.1146, over 28298.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3743, pruned_loss=0.1245, over 5644500.45 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3432, pruned_loss=0.08947, over 5721064.63 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3774, pruned_loss=0.1279, over 5631639.82 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:34:46,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9655, 1.3377, 1.1453, 0.2141], device='cuda:0'), covar=tensor([0.4478, 0.3471, 0.5052, 0.7016], device='cuda:0'), in_proj_covar=tensor([0.1820, 0.1709, 0.1644, 0.1474], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 09:35:18,263 INFO [train.py:968] (0/2) Epoch 26, batch 12350, giga_loss[loss=0.2627, simple_loss=0.3411, pruned_loss=0.09217, over 28327.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3747, pruned_loss=0.1244, over 5650545.88 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3436, pruned_loss=0.08967, over 5723682.19 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3779, pruned_loss=0.1282, over 5635579.05 frames. ], batch size: 65, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:35:33,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.539e+03 2.165e+03 2.884e+03 7.851e+03, threshold=4.329e+03, percent-clipped=6.0
+2023-03-13 09:35:56,792 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5752, 1.8039, 1.5242, 1.4707], device='cuda:0'), covar=tensor([0.2631, 0.2813, 0.3128, 0.2468], device='cuda:0'), in_proj_covar=tensor([0.1572, 0.1136, 0.1389, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 09:36:01,484 INFO [train.py:968] (0/2) Epoch 26, batch 12400, libri_loss[loss=0.2972, simple_loss=0.3756, pruned_loss=0.1094, over 29287.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3744, pruned_loss=0.1231, over 5658555.15 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3438, pruned_loss=0.08962, over 5724093.73 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3775, pruned_loss=0.1269, over 5644845.18 frames. ], batch size: 94, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:36:02,453 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:36:38,865 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1151497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:36:42,646 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2335, 1.3557, 3.7101, 3.1699], device='cuda:0'), covar=tensor([0.1772, 0.2743, 0.0509, 0.0959], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0671, 0.0991, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 09:36:49,677 INFO [train.py:968] (0/2) Epoch 26, batch 12450, giga_loss[loss=0.3323, simple_loss=0.3938, pruned_loss=0.1353, over 28587.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3726, pruned_loss=0.1222, over 5659028.03 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3436, pruned_loss=0.0895, over 5725934.81 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3756, pruned_loss=0.1258, over 5645485.99 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:37:03,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.679e+03 2.347e+03 3.126e+03 7.357e+03, threshold=4.694e+03, percent-clipped=11.0
+2023-03-13 09:37:30,550 INFO [train.py:968] (0/2) Epoch 26, batch 12500, giga_loss[loss=0.2636, simple_loss=0.3367, pruned_loss=0.09523, over 28348.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3692, pruned_loss=0.1199, over 5654447.57 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3436, pruned_loss=0.08967, over 5720319.09 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3724, pruned_loss=0.1236, over 5647036.61 frames. ], batch size: 65, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:37:36,715 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1151562.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:38:12,443 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7007, 2.5843, 1.6740, 0.9005], device='cuda:0'), covar=tensor([0.8749, 0.4203, 0.4294, 0.8202], device='cuda:0'), in_proj_covar=tensor([0.1820, 0.1710, 0.1645, 0.1478], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 09:38:13,014 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1151601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:38:15,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1151604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:38:19,001 INFO [train.py:968] (0/2) Epoch 26, batch 12550, giga_loss[loss=0.2839, simple_loss=0.3546, pruned_loss=0.1066, over 28714.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.367, pruned_loss=0.1184, over 5664828.93 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.08998, over 5715549.67 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3696, pruned_loss=0.1219, over 5662078.57 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:38:36,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.261e+03 1.885e+03 2.389e+03 3.379e+03 1.060e+04, threshold=4.777e+03, percent-clipped=13.0
+2023-03-13 09:38:45,006 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1151633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:39:06,345 INFO [train.py:968] (0/2) Epoch 26, batch 12600, giga_loss[loss=0.2833, simple_loss=0.3495, pruned_loss=0.1085, over 28644.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3629, pruned_loss=0.1165, over 5651724.28 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3446, pruned_loss=0.09026, over 5718417.84 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3649, pruned_loss=0.1194, over 5646154.28 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:39:54,580 INFO [train.py:968] (0/2) Epoch 26, batch 12650, giga_loss[loss=0.336, simple_loss=0.3798, pruned_loss=0.146, over 26632.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3631, pruned_loss=0.1179, over 5641604.49 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3445, pruned_loss=0.09019, over 5708941.63 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3651, pruned_loss=0.1208, over 5644051.16 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:40:09,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.210e+03 1.722e+03 2.211e+03 2.983e+03 9.598e+03, threshold=4.422e+03, percent-clipped=4.0
+2023-03-13 09:40:40,910 INFO [train.py:968] (0/2) Epoch 26, batch 12700, libri_loss[loss=0.2799, simple_loss=0.3584, pruned_loss=0.1007, over 29072.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3616, pruned_loss=0.1171, over 5642936.23 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3441, pruned_loss=0.0899, over 5710887.78 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3641, pruned_loss=0.1204, over 5641167.71 frames. ], batch size: 101, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:41:24,893 INFO [train.py:968] (0/2) Epoch 26, batch 12750, giga_loss[loss=0.2517, simple_loss=0.3418, pruned_loss=0.08081, over 28963.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3606, pruned_loss=0.1152, over 5637636.20 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3442, pruned_loss=0.09003, over 5703577.92 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.363, pruned_loss=0.1185, over 5640867.60 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:41:44,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.161e+03 1.851e+03 2.275e+03 3.013e+03 5.593e+03, threshold=4.550e+03, percent-clipped=2.0
+2023-03-13 09:42:14,232 INFO [train.py:968] (0/2) Epoch 26, batch 12800, giga_loss[loss=0.3105, simple_loss=0.3836, pruned_loss=0.1187, over 28776.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3587, pruned_loss=0.1117, over 5644647.39 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3439, pruned_loss=0.08993, over 5707181.69 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3612, pruned_loss=0.1149, over 5642900.64 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:42:15,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6456, 1.8370, 1.4663, 1.8934], device='cuda:0'), covar=tensor([0.2793, 0.2887, 0.3225, 0.2560], device='cuda:0'), in_proj_covar=tensor([0.1573, 0.1134, 0.1391, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 09:42:32,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:43:04,791 INFO [train.py:968] (0/2) Epoch 26, batch 12850, giga_loss[loss=0.3016, simple_loss=0.375, pruned_loss=0.1141, over 28768.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3563, pruned_loss=0.1085, over 5645060.71 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3434, pruned_loss=0.08985, over 5707280.11 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3591, pruned_loss=0.1118, over 5641562.14 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:43:25,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.043e+02 1.505e+03 1.919e+03 2.392e+03 5.428e+03, threshold=3.838e+03, percent-clipped=1.0
+2023-03-13 09:43:37,786 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151937.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:43:57,093 INFO [train.py:968] (0/2) Epoch 26, batch 12900, giga_loss[loss=0.3005, simple_loss=0.364, pruned_loss=0.1185, over 28899.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3541, pruned_loss=0.106, over 5644736.21 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3434, pruned_loss=0.08983, over 5708337.21 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3565, pruned_loss=0.1088, over 5640330.78 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:44:11,549 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1257, 1.5054, 1.5595, 1.2753], device='cuda:0'), covar=tensor([0.1781, 0.1269, 0.1710, 0.1528], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0750, 0.0718, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 09:44:30,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4793, 5.2851, 5.0516, 2.6849], device='cuda:0'), covar=tensor([0.0465, 0.0635, 0.0830, 0.1734], device='cuda:0'), in_proj_covar=tensor([0.1293, 0.1195, 0.1009, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 09:44:31,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3266, 1.9185, 1.3894, 0.5016], device='cuda:0'), covar=tensor([0.5146, 0.2835, 0.4110, 0.6615], device='cuda:0'), in_proj_covar=tensor([0.1810, 0.1698, 0.1639, 0.1472], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 09:44:35,488 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2986, 1.2849, 1.1981, 1.5352], device='cuda:0'), covar=tensor([0.0719, 0.0466, 0.0373, 0.0785], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 09:44:38,048 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1152000.pt
+2023-03-13 09:44:44,193 INFO [train.py:968] (0/2) Epoch 26, batch 12950, giga_loss[loss=0.2622, simple_loss=0.3427, pruned_loss=0.09091, over 28897.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3496, pruned_loss=0.1021, over 5646681.67 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3428, pruned_loss=0.08968, over 5715609.73 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3524, pruned_loss=0.105, over 5634399.11 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:44:51,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:44:53,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:45:02,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.763e+02 1.390e+03 1.747e+03 2.230e+03 5.368e+03, threshold=3.493e+03, percent-clipped=4.0
+2023-03-13 09:45:07,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-13 09:45:22,941 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:45:34,632 INFO [train.py:968] (0/2) Epoch 26, batch 13000, giga_loss[loss=0.2479, simple_loss=0.339, pruned_loss=0.0784, over 28930.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3489, pruned_loss=0.09904, over 5660214.95 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3425, pruned_loss=0.08953, over 5717007.65 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3516, pruned_loss=0.1016, over 5648451.75 frames. ], batch size: 112, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:45:57,540 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152080.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:46:00,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152083.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:46:22,663 INFO [train.py:968] (0/2) Epoch 26, batch 13050, giga_loss[loss=0.2551, simple_loss=0.3418, pruned_loss=0.08416, over 28939.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3506, pruned_loss=0.1002, over 5658746.24 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3426, pruned_loss=0.0898, over 5721996.51 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3528, pruned_loss=0.1023, over 5642880.24 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:46:26,977 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152112.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:46:38,742 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.157e+02 1.771e+03 2.379e+03 3.570e+03 8.023e+03, threshold=4.758e+03, percent-clipped=25.0
+2023-03-13 09:46:39,010 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:46:51,966 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:47:09,014 INFO [train.py:968] (0/2) Epoch 26, batch 13100, giga_loss[loss=0.313, simple_loss=0.3613, pruned_loss=0.1324, over 26599.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3497, pruned_loss=0.0997, over 5659307.12 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3425, pruned_loss=0.08998, over 5725748.43 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3518, pruned_loss=0.1015, over 5641652.11 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:47:53,303 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:47:57,606 INFO [train.py:968] (0/2) Epoch 26, batch 13150, giga_loss[loss=0.2787, simple_loss=0.3395, pruned_loss=0.109, over 26778.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3456, pruned_loss=0.09697, over 5656295.20 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3417, pruned_loss=0.08968, over 5729618.44 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3481, pruned_loss=0.09888, over 5637488.90 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:48:16,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.954e+02 1.455e+03 1.940e+03 2.676e+03 7.106e+03, threshold=3.880e+03, percent-clipped=3.0
+2023-03-13 09:48:23,486 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:48:34,679 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2044, 1.8364, 1.6337, 1.4274], device='cuda:0'), covar=tensor([0.2509, 0.1827, 0.2258, 0.2121], device='cuda:0'), in_proj_covar=tensor([0.0495, 0.0747, 0.0716, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 09:48:42,149 INFO [train.py:968] (0/2) Epoch 26, batch 13200, giga_loss[loss=0.2379, simple_loss=0.325, pruned_loss=0.0754, over 28659.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3449, pruned_loss=0.09664, over 5657944.39 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3415, pruned_loss=0.08965, over 5733665.32 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3473, pruned_loss=0.09841, over 5636807.24 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:49:18,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4817, 2.0891, 1.4792, 0.7751], device='cuda:0'), covar=tensor([0.5937, 0.3301, 0.4216, 0.6585], device='cuda:0'), in_proj_covar=tensor([0.1809, 0.1698, 0.1639, 0.1472], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 09:49:28,100 INFO [train.py:968] (0/2) Epoch 26, batch 13250, giga_loss[loss=0.2393, simple_loss=0.3226, pruned_loss=0.07804, over 28762.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3448, pruned_loss=0.09665, over 5640179.83 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3411, pruned_loss=0.08966, over 5718632.76 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3471, pruned_loss=0.09826, over 5634340.55 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:49:48,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.425e+03 2.074e+03 2.747e+03 8.421e+03, threshold=4.148e+03, percent-clipped=10.0
+2023-03-13 09:50:16,901 INFO [train.py:968] (0/2) Epoch 26, batch 13300, giga_loss[loss=0.2507, simple_loss=0.3318, pruned_loss=0.08478, over 28766.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.343, pruned_loss=0.09482, over 5651754.76 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3409, pruned_loss=0.08953, over 5719654.55 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.345, pruned_loss=0.09626, over 5645584.91 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:50:42,120 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-13 09:50:47,170 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-13 09:51:06,279 INFO [train.py:968] (0/2) Epoch 26, batch 13350, giga_loss[loss=0.2211, simple_loss=0.3063, pruned_loss=0.06799, over 28758.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3407, pruned_loss=0.09334, over 5647833.20 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3411, pruned_loss=0.08993, over 5721620.28 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3421, pruned_loss=0.0942, over 5639803.74 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:51:27,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.364e+02 1.366e+03 1.681e+03 2.296e+03 6.914e+03, threshold=3.362e+03, percent-clipped=7.0
+2023-03-13 09:51:59,250 INFO [train.py:968] (0/2) Epoch 26, batch 13400, giga_loss[loss=0.237, simple_loss=0.3164, pruned_loss=0.07883, over 28925.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3366, pruned_loss=0.09101, over 5649381.48 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3413, pruned_loss=0.09026, over 5725048.67 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3375, pruned_loss=0.09146, over 5638781.53 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:52:07,207 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:52:14,378 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 09:52:48,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1152501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:52:53,520 INFO [train.py:968] (0/2) Epoch 26, batch 13450, giga_loss[loss=0.2884, simple_loss=0.3564, pruned_loss=0.1102, over 27886.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3349, pruned_loss=0.09024, over 5663002.52 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3405, pruned_loss=0.08991, over 5729694.21 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3362, pruned_loss=0.09091, over 5648668.97 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:53:00,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1152514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:53:12,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.101e+02 1.453e+03 1.901e+03 2.808e+03 6.248e+03, threshold=3.802e+03, percent-clipped=14.0
+2023-03-13 09:53:21,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1580, 1.0252, 3.7160, 3.1484], device='cuda:0'), covar=tensor([0.1808, 0.3039, 0.0527, 0.1038], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0665, 0.0980, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 09:53:40,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4270, 1.6230, 1.2780, 1.1879], device='cuda:0'), covar=tensor([0.0990, 0.0479, 0.0937, 0.1086], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0446, 0.0517, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 09:53:41,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4404, 1.1400, 3.9588, 3.3469], device='cuda:0'), covar=tensor([0.1656, 0.3055, 0.0465, 0.0924], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0665, 0.0979, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 09:53:41,748 INFO [train.py:968] (0/2) Epoch 26, batch 13500, giga_loss[loss=0.2235, simple_loss=0.3138, pruned_loss=0.06657, over 29016.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3337, pruned_loss=0.09019, over 5645450.83 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3398, pruned_loss=0.08954, over 5725037.43 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3353, pruned_loss=0.09109, over 5635839.94 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:54:02,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1152577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:54:35,950 INFO [train.py:968] (0/2) Epoch 26, batch 13550, giga_loss[loss=0.2808, simple_loss=0.363, pruned_loss=0.09926, over 28831.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3346, pruned_loss=0.09032, over 5655422.86 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3394, pruned_loss=0.08943, over 5731542.64 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.336, pruned_loss=0.09117, over 5639257.89 frames. ], batch size: 243, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:54:38,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1152608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:54:43,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0662, 1.5129, 1.5343, 1.2660], device='cuda:0'), covar=tensor([0.2298, 0.1640, 0.2114, 0.1868], device='cuda:0'), in_proj_covar=tensor([0.0490, 0.0742, 0.0710, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 09:54:46,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2058, 1.7366, 1.2759, 0.4979], device='cuda:0'), covar=tensor([0.4555, 0.2852, 0.4405, 0.6431], device='cuda:0'), in_proj_covar=tensor([0.1809, 0.1697, 0.1639, 0.1472], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 09:54:55,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.992e+02 1.482e+03 2.038e+03 2.892e+03 5.349e+03, threshold=4.076e+03, percent-clipped=13.0
+2023-03-13 09:55:14,604 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:55:20,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:55:31,058 INFO [train.py:968] (0/2) Epoch 26, batch 13600, giga_loss[loss=0.2911, simple_loss=0.3745, pruned_loss=0.1039, over 28867.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3386, pruned_loss=0.09148, over 5654219.60 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3391, pruned_loss=0.08915, over 5734228.56 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3399, pruned_loss=0.09244, over 5637361.52 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:55:31,408 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:55:35,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:55:56,536 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:56:10,842 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:56:31,076 INFO [train.py:968] (0/2) Epoch 26, batch 13650, giga_loss[loss=0.2624, simple_loss=0.3441, pruned_loss=0.09028, over 28993.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3391, pruned_loss=0.09125, over 5666433.68 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3387, pruned_loss=0.08906, over 5736641.96 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3405, pruned_loss=0.09211, over 5650051.77 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:56:47,438 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:56:51,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:56:55,048 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 09:56:58,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.472e+03 2.088e+03 3.206e+03 6.438e+03, threshold=4.176e+03, percent-clipped=12.0
+2023-03-13 09:57:25,352 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:57:27,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:57:27,718 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:57:29,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:57:31,683 INFO [train.py:968] (0/2) Epoch 26, batch 13700, giga_loss[loss=0.231, simple_loss=0.3157, pruned_loss=0.07311, over 28709.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.338, pruned_loss=0.09089, over 5673112.66 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3387, pruned_loss=0.0893, over 5742991.52 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3391, pruned_loss=0.09144, over 5652016.22 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:57:59,947 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:58:32,071 INFO [train.py:968] (0/2) Epoch 26, batch 13750, giga_loss[loss=0.2779, simple_loss=0.3468, pruned_loss=0.1045, over 26782.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3358, pruned_loss=0.0892, over 5668119.60 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3381, pruned_loss=0.08901, over 5733719.94 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3372, pruned_loss=0.08989, over 5657864.07 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:58:37,428 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1014, 3.9453, 3.7262, 1.7069], device='cuda:0'), covar=tensor([0.0677, 0.0770, 0.0839, 0.2184], device='cuda:0'), in_proj_covar=tensor([0.1265, 0.1168, 0.0984, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 09:58:57,147 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.319e+03 1.757e+03 2.389e+03 7.265e+03, threshold=3.514e+03, percent-clipped=2.0
+2023-03-13 09:59:11,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1152839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:59:29,224 INFO [train.py:968] (0/2) Epoch 26, batch 13800, libri_loss[loss=0.258, simple_loss=0.3375, pruned_loss=0.08928, over 29249.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3347, pruned_loss=0.08736, over 5665091.45 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3376, pruned_loss=0.08893, over 5728742.60 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3362, pruned_loss=0.08797, over 5658745.61 frames. ], batch size: 94, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:59:52,869 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:00:09,130 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:00:31,679 INFO [train.py:968] (0/2) Epoch 26, batch 13850, giga_loss[loss=0.2202, simple_loss=0.2997, pruned_loss=0.07038, over 28957.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3329, pruned_loss=0.08736, over 5662717.94 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3375, pruned_loss=0.08889, over 5731563.58 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3341, pruned_loss=0.08785, over 5654481.50 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:00:57,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.567e+02 1.349e+03 1.720e+03 2.443e+03 5.373e+03, threshold=3.440e+03, percent-clipped=7.0
+2023-03-13 10:01:08,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5896, 2.1788, 1.7349, 1.7098], device='cuda:0'), covar=tensor([0.0764, 0.0258, 0.0308, 0.0891], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 10:01:15,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4122, 1.8453, 1.0721, 1.4109], device='cuda:0'), covar=tensor([0.1350, 0.0729, 0.1598, 0.1390], device='cuda:0'), in_proj_covar=tensor([0.0401, 0.0444, 0.0515, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 10:01:31,086 INFO [train.py:968] (0/2) Epoch 26, batch 13900, libri_loss[loss=0.279, simple_loss=0.3515, pruned_loss=0.1033, over 19955.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.332, pruned_loss=0.08742, over 5658305.63 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3374, pruned_loss=0.08887, over 5724614.65 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3329, pruned_loss=0.08779, over 5657796.18 frames. ], batch size: 187, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:01:36,603 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:02:03,126 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:02:05,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:02:29,401 INFO [train.py:968] (0/2) Epoch 26, batch 13950, giga_loss[loss=0.3124, simple_loss=0.3859, pruned_loss=0.1194, over 28671.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3307, pruned_loss=0.08685, over 5659016.19 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3374, pruned_loss=0.0889, over 5728675.91 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3313, pruned_loss=0.08706, over 5653368.73 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:02:31,080 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1153009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:02:32,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3645, 1.5237, 1.3065, 1.4304], device='cuda:0'), covar=tensor([0.0772, 0.0340, 0.0355, 0.0921], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 10:02:35,736 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:02:45,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5787, 1.6032, 1.8005, 1.3719], device='cuda:0'), covar=tensor([0.2040, 0.2814, 0.1717, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0706, 0.0967, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 10:02:52,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.833e+02 1.510e+03 2.236e+03 3.096e+03 7.719e+03, threshold=4.473e+03, percent-clipped=15.0
+2023-03-13 10:03:24,338 INFO [train.py:968] (0/2) Epoch 26, batch 14000, giga_loss[loss=0.2784, simple_loss=0.3509, pruned_loss=0.1029, over 27616.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3313, pruned_loss=0.08648, over 5660564.87 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3367, pruned_loss=0.0886, over 5734192.59 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3322, pruned_loss=0.08687, over 5649284.64 frames. ], batch size: 474, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:04:05,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9591, 4.8009, 4.5810, 2.3608], device='cuda:0'), covar=tensor([0.0467, 0.0592, 0.0690, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.1263, 0.1166, 0.0983, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 10:04:26,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-13 10:04:26,625 INFO [train.py:968] (0/2) Epoch 26, batch 14050, giga_loss[loss=0.2366, simple_loss=0.3181, pruned_loss=0.07751, over 28928.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3322, pruned_loss=0.08619, over 5660669.85 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3366, pruned_loss=0.08854, over 5736334.65 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.333, pruned_loss=0.08652, over 5648668.20 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:04:51,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:04:53,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.405e+02 1.350e+03 1.931e+03 2.669e+03 6.669e+03, threshold=3.862e+03, percent-clipped=3.0
+2023-03-13 10:05:18,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6124, 1.8775, 1.2225, 1.4192], device='cuda:0'), covar=tensor([0.1005, 0.0527, 0.1038, 0.1080], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0446, 0.0519, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 10:05:34,666 INFO [train.py:968] (0/2) Epoch 26, batch 14100, giga_loss[loss=0.3355, simple_loss=0.3818, pruned_loss=0.1445, over 26793.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3302, pruned_loss=0.08537, over 5672616.63 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3366, pruned_loss=0.0886, over 5738600.62 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3307, pruned_loss=0.08552, over 5659841.16 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:06:39,109 INFO [train.py:968] (0/2) Epoch 26, batch 14150, giga_loss[loss=0.2388, simple_loss=0.3289, pruned_loss=0.07439, over 28455.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3314, pruned_loss=0.08588, over 5684733.04 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3363, pruned_loss=0.08843, over 5741436.86 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.332, pruned_loss=0.08611, over 5670915.57 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:07:05,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.198e+02 1.364e+03 1.732e+03 2.452e+03 4.879e+03, threshold=3.464e+03, percent-clipped=6.0
+2023-03-13 10:07:33,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:07:34,910 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 10:07:37,332 INFO [train.py:968] (0/2) Epoch 26, batch 14200, giga_loss[loss=0.2666, simple_loss=0.3627, pruned_loss=0.08525, over 28983.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3366, pruned_loss=0.08746, over 5663201.19 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3366, pruned_loss=0.08876, over 5721748.84 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3367, pruned_loss=0.08729, over 5666736.25 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:07:44,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:07:51,609 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:07:52,600 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 10:07:57,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:08:31,592 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:08:36,936 INFO [train.py:968] (0/2) Epoch 26, batch 14250, giga_loss[loss=0.2563, simple_loss=0.3465, pruned_loss=0.0831, over 29018.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3403, pruned_loss=0.0876, over 5666979.66 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3365, pruned_loss=0.08871, over 5723978.59 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3405, pruned_loss=0.08749, over 5666988.46 frames. ], batch size: 285, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:08:52,193 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1153321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:09:00,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.552e+03 1.928e+03 2.790e+03 8.634e+03, threshold=3.856e+03, percent-clipped=13.0
+2023-03-13 10:09:12,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:09:34,804 INFO [train.py:968] (0/2) Epoch 26, batch 14300, giga_loss[loss=0.2633, simple_loss=0.3564, pruned_loss=0.08511, over 28758.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3411, pruned_loss=0.08684, over 5667859.85 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3362, pruned_loss=0.08849, over 5726509.57 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3417, pruned_loss=0.08691, over 5664603.52 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:10:09,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:10:20,389 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:10:24,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:10:36,369 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:10:37,106 INFO [train.py:968] (0/2) Epoch 26, batch 14350, giga_loss[loss=0.2694, simple_loss=0.3495, pruned_loss=0.09467, over 28918.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3408, pruned_loss=0.08654, over 5667950.48 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3363, pruned_loss=0.0886, over 5728999.51 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3411, pruned_loss=0.08648, over 5662486.88 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:10:40,143 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:11:05,006 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:11:07,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.245e+02 1.401e+03 1.855e+03 2.667e+03 6.221e+03, threshold=3.711e+03, percent-clipped=4.0
+2023-03-13 10:11:19,096 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:11:42,234 INFO [train.py:968] (0/2) Epoch 26, batch 14400, giga_loss[loss=0.2442, simple_loss=0.3285, pruned_loss=0.07993, over 28396.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3401, pruned_loss=0.08722, over 5669495.17 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3363, pruned_loss=0.0886, over 5728999.51 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3404, pruned_loss=0.08717, over 5665242.78 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:12:05,453 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1153475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:12:13,666 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:12:17,827 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:12:52,993 INFO [train.py:968] (0/2) Epoch 26, batch 14450, giga_loss[loss=0.2575, simple_loss=0.3404, pruned_loss=0.08727, over 28958.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3401, pruned_loss=0.08804, over 5684086.51 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3363, pruned_loss=0.08868, over 5730445.47 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3404, pruned_loss=0.08791, over 5678956.94 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:12:58,428 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:13:08,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 10:13:23,437 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:13:25,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.388e+02 1.402e+03 1.789e+03 2.338e+03 5.059e+03, threshold=3.577e+03, percent-clipped=4.0
+2023-03-13 10:13:27,908 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:14:09,216 INFO [train.py:968] (0/2) Epoch 26, batch 14500, libri_loss[loss=0.2815, simple_loss=0.3595, pruned_loss=0.1018, over 29523.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3379, pruned_loss=0.08765, over 5682414.77 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3358, pruned_loss=0.08845, over 5735089.73 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3387, pruned_loss=0.08774, over 5673027.07 frames. ], batch size: 89, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:14:15,713 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:15:19,554 INFO [train.py:968] (0/2) Epoch 26, batch 14550, giga_loss[loss=0.2314, simple_loss=0.3172, pruned_loss=0.07282, over 28754.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.334, pruned_loss=0.08537, over 5684736.38 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3358, pruned_loss=0.0886, over 5740002.92 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3345, pruned_loss=0.0852, over 5671106.68 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:15:49,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.391e+02 1.280e+03 1.780e+03 2.520e+03 4.681e+03, threshold=3.559e+03, percent-clipped=8.0
+2023-03-13 10:15:57,989 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 10:16:23,543 INFO [train.py:968] (0/2) Epoch 26, batch 14600, giga_loss[loss=0.205, simple_loss=0.2945, pruned_loss=0.0577, over 28763.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3318, pruned_loss=0.08414, over 5689055.79 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.336, pruned_loss=0.08868, over 5743424.78 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3319, pruned_loss=0.08385, over 5674059.41 frames. ], batch size: 263, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:16:59,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3117, 1.2420, 3.3956, 3.1407], device='cuda:0'), covar=tensor([0.1527, 0.2902, 0.0494, 0.1967], device='cuda:0'), in_proj_covar=tensor([0.0785, 0.0665, 0.0977, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 10:17:11,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:17:22,872 INFO [train.py:968] (0/2) Epoch 26, batch 14650, giga_loss[loss=0.2904, simple_loss=0.3683, pruned_loss=0.1062, over 28925.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.333, pruned_loss=0.08576, over 5684421.22 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.336, pruned_loss=0.08878, over 5746622.80 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3331, pruned_loss=0.08538, over 5668894.11 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:17:50,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.977e+02 1.522e+03 2.060e+03 2.946e+03 8.984e+03, threshold=4.121e+03, percent-clipped=15.0
+2023-03-13 10:18:25,691 INFO [train.py:968] (0/2) Epoch 26, batch 14700, giga_loss[loss=0.2688, simple_loss=0.3293, pruned_loss=0.1042, over 26879.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3364, pruned_loss=0.0874, over 5686894.43 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3354, pruned_loss=0.08845, over 5750869.19 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.337, pruned_loss=0.08733, over 5669397.37 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:19:22,841 INFO [train.py:968] (0/2) Epoch 26, batch 14750, giga_loss[loss=0.2461, simple_loss=0.3242, pruned_loss=0.08401, over 28932.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3346, pruned_loss=0.08745, over 5679012.82 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3355, pruned_loss=0.0886, over 5741751.69 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3349, pruned_loss=0.08723, over 5672259.83 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:19:46,042 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1153825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:19:52,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.494e+02 1.381e+03 1.916e+03 2.759e+03 8.684e+03, threshold=3.831e+03, percent-clipped=5.0
+2023-03-13 10:20:03,209 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:20:08,839 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:20:19,760 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:20:20,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3183, 1.1606, 3.9515, 3.2737], device='cuda:0'), covar=tensor([0.1753, 0.2849, 0.0497, 0.1003], device='cuda:0'), in_proj_covar=tensor([0.0791, 0.0671, 0.0985, 0.0953], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 10:20:22,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8460, 2.7631, 1.7258, 1.2088], device='cuda:0'), covar=tensor([0.8347, 0.4226, 0.4396, 0.6556], device='cuda:0'), in_proj_covar=tensor([0.1814, 0.1698, 0.1639, 0.1475], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 10:20:26,560 INFO [train.py:968] (0/2) Epoch 26, batch 14800, giga_loss[loss=0.2368, simple_loss=0.3062, pruned_loss=0.08372, over 24427.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3339, pruned_loss=0.08782, over 5682788.05 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.335, pruned_loss=0.08833, over 5745943.70 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3346, pruned_loss=0.08789, over 5672249.48 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:20:42,190 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:21:25,719 INFO [train.py:968] (0/2) Epoch 26, batch 14850, giga_loss[loss=0.2703, simple_loss=0.3561, pruned_loss=0.09224, over 28973.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3342, pruned_loss=0.0879, over 5688649.62 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3348, pruned_loss=0.08826, over 5747176.44 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3349, pruned_loss=0.08801, over 5677849.64 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:21:40,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8390, 2.1083, 1.6329, 2.0060], device='cuda:0'), covar=tensor([0.2789, 0.2736, 0.3266, 0.2575], device='cuda:0'), in_proj_covar=tensor([0.1567, 0.1126, 0.1384, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 10:21:55,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.909e+02 1.616e+03 2.097e+03 3.112e+03 6.861e+03, threshold=4.194e+03, percent-clipped=17.0
+2023-03-13 10:22:31,098 INFO [train.py:968] (0/2) Epoch 26, batch 14900, giga_loss[loss=0.2645, simple_loss=0.348, pruned_loss=0.09048, over 28690.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.337, pruned_loss=0.08855, over 5689587.79 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3348, pruned_loss=0.08828, over 5749439.40 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3377, pruned_loss=0.08864, over 5678210.46 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:23:29,021 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:23:33,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:23:38,819 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1154000.pt
+2023-03-13 10:23:46,793 INFO [train.py:968] (0/2) Epoch 26, batch 14950, giga_loss[loss=0.2423, simple_loss=0.3331, pruned_loss=0.07568, over 28609.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3364, pruned_loss=0.08814, over 5676271.23 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3344, pruned_loss=0.08819, over 5751163.98 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3373, pruned_loss=0.08829, over 5664953.76 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:24:15,895 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1154025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:24:20,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.460e+03 1.807e+03 2.365e+03 4.511e+03, threshold=3.614e+03, percent-clipped=2.0
+2023-03-13 10:24:55,775 INFO [train.py:968] (0/2) Epoch 26, batch 15000, giga_loss[loss=0.2074, simple_loss=0.2901, pruned_loss=0.06239, over 28834.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3332, pruned_loss=0.08729, over 5673735.78 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3339, pruned_loss=0.08787, over 5754456.22 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3343, pruned_loss=0.08769, over 5659502.59 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:24:55,779 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 10:25:04,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3977, 1.8272, 1.6784, 1.2173], device='cuda:0'), covar=tensor([0.1907, 0.2909, 0.1657, 0.1995], device='cuda:0'), in_proj_covar=tensor([0.0918, 0.0704, 0.0967, 0.0866], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 10:25:05,779 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.1951, simple_loss=0.2963, pruned_loss=0.04693, over 944034.00 frames. 
+2023-03-13 10:25:05,780 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 10:25:24,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5605, 1.9527, 1.5317, 1.6370], device='cuda:0'), covar=tensor([0.0775, 0.0279, 0.0331, 0.0898], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0120, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 10:25:42,752 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1154088.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 10:25:57,070 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3620, 1.5827, 1.4887, 1.2948], device='cuda:0'), covar=tensor([0.2699, 0.2142, 0.1800, 0.2371], device='cuda:0'), in_proj_covar=tensor([0.2001, 0.1939, 0.1856, 0.2000], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 10:26:11,075 INFO [train.py:968] (0/2) Epoch 26, batch 15050, giga_loss[loss=0.2096, simple_loss=0.2935, pruned_loss=0.06284, over 28325.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3274, pruned_loss=0.0847, over 5673553.71 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3338, pruned_loss=0.08771, over 5758615.85 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3283, pruned_loss=0.08512, over 5656402.51 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:26:39,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.517e+02 1.599e+03 2.024e+03 3.070e+03 6.783e+03, threshold=4.048e+03, percent-clipped=15.0
+2023-03-13 10:26:55,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2406, 1.8474, 1.5885, 1.4714], device='cuda:0'), covar=tensor([0.2545, 0.2112, 0.2203, 0.2278], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0740, 0.0710, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 10:27:11,300 INFO [train.py:968] (0/2) Epoch 26, batch 15100, giga_loss[loss=0.2379, simple_loss=0.3243, pruned_loss=0.07576, over 28623.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3258, pruned_loss=0.08411, over 5662995.17 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3339, pruned_loss=0.08796, over 5741697.52 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3264, pruned_loss=0.08417, over 5663571.12 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:28:03,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1154200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:28:09,060 INFO [train.py:968] (0/2) Epoch 26, batch 15150, giga_loss[loss=0.2764, simple_loss=0.3478, pruned_loss=0.1025, over 28523.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3286, pruned_loss=0.08628, over 5656839.09 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3335, pruned_loss=0.08778, over 5743490.25 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3293, pruned_loss=0.08645, over 5654951.24 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:28:33,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.513e+02 1.699e+03 2.411e+03 3.242e+03 9.620e+03, threshold=4.822e+03, percent-clipped=13.0
+2023-03-13 10:28:55,121 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-13 10:28:56,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4280, 1.6821, 1.2132, 1.2339], device='cuda:0'), covar=tensor([0.1055, 0.0450, 0.1018, 0.1051], device='cuda:0'), in_proj_covar=tensor([0.0403, 0.0445, 0.0518, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 10:29:05,248 INFO [train.py:968] (0/2) Epoch 26, batch 15200, giga_loss[loss=0.2395, simple_loss=0.3185, pruned_loss=0.0803, over 28973.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3265, pruned_loss=0.08476, over 5662631.87 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3331, pruned_loss=0.08764, over 5744336.31 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3273, pruned_loss=0.08498, over 5658489.69 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:30:04,165 INFO [train.py:968] (0/2) Epoch 26, batch 15250, giga_loss[loss=0.236, simple_loss=0.3188, pruned_loss=0.0766, over 28928.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3254, pruned_loss=0.08357, over 5650620.64 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3326, pruned_loss=0.08731, over 5738571.15 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3263, pruned_loss=0.08398, over 5650771.35 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:30:05,721 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-13 10:30:30,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2394, 0.8845, 0.9106, 1.3845], device='cuda:0'), covar=tensor([0.0800, 0.0414, 0.0385, 0.0890], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0120, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 10:30:30,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.679e+02 1.370e+03 1.935e+03 2.726e+03 7.023e+03, threshold=3.871e+03, percent-clipped=2.0
+2023-03-13 10:30:49,023 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1154343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:30:52,523 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1154346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:30:59,464 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4130, 2.9537, 2.7777, 2.0454], device='cuda:0'), covar=tensor([0.2931, 0.1748, 0.1913, 0.2626], device='cuda:0'), in_proj_covar=tensor([0.1992, 0.1932, 0.1849, 0.1993], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 10:31:03,047 INFO [train.py:968] (0/2) Epoch 26, batch 15300, giga_loss[loss=0.2017, simple_loss=0.2913, pruned_loss=0.05602, over 28919.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3254, pruned_loss=0.08402, over 5655595.10 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3327, pruned_loss=0.08755, over 5731163.48 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3258, pruned_loss=0.08401, over 5659548.21 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:31:27,700 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1154375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:31:43,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1154386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:32:08,056 INFO [train.py:968] (0/2) Epoch 26, batch 15350, giga_loss[loss=0.2281, simple_loss=0.3082, pruned_loss=0.074, over 28611.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3237, pruned_loss=0.08292, over 5651752.47 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3321, pruned_loss=0.08722, over 5735717.77 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3244, pruned_loss=0.08312, over 5649004.67 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:32:11,338 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1154409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:32:41,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.837e+02 1.389e+03 1.843e+03 2.580e+03 6.221e+03, threshold=3.686e+03, percent-clipped=5.0
+2023-03-13 10:33:09,191 INFO [train.py:968] (0/2) Epoch 26, batch 15400, giga_loss[loss=0.2633, simple_loss=0.3406, pruned_loss=0.09306, over 28083.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3245, pruned_loss=0.08306, over 5651194.81 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3321, pruned_loss=0.08722, over 5735674.87 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.325, pruned_loss=0.08313, over 5647516.41 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:33:16,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1154463.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 10:34:10,934 INFO [train.py:968] (0/2) Epoch 26, batch 15450, giga_loss[loss=0.2674, simple_loss=0.3384, pruned_loss=0.09813, over 28371.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3258, pruned_loss=0.08436, over 5656667.29 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3321, pruned_loss=0.08731, over 5733572.75 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.326, pruned_loss=0.08428, over 5654097.11 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:34:41,178 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.671e+02 1.468e+03 1.955e+03 2.893e+03 9.823e+03, threshold=3.910e+03, percent-clipped=14.0
+2023-03-13 10:35:11,567 INFO [train.py:968] (0/2) Epoch 26, batch 15500, giga_loss[loss=0.2246, simple_loss=0.3187, pruned_loss=0.06524, over 28835.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3245, pruned_loss=0.08385, over 5657532.88 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3316, pruned_loss=0.08703, over 5739885.87 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3249, pruned_loss=0.08391, over 5647195.06 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:35:21,299 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-13 10:35:21,866 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-13 10:36:09,890 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1154606.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:36:10,158 INFO [train.py:968] (0/2) Epoch 26, batch 15550, giga_loss[loss=0.2448, simple_loss=0.3322, pruned_loss=0.07872, over 27718.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3254, pruned_loss=0.08225, over 5666533.55 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3315, pruned_loss=0.08701, over 5736827.98 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3256, pruned_loss=0.08226, over 5660096.28 frames. ], batch size: 474, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:36:12,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1154609.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:36:36,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.974e+02 1.468e+03 1.743e+03 2.366e+03 6.495e+03, threshold=3.486e+03, percent-clipped=8.0
+2023-03-13 10:36:46,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1154638.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:37:04,789 INFO [train.py:968] (0/2) Epoch 26, batch 15600, giga_loss[loss=0.2489, simple_loss=0.3206, pruned_loss=0.08856, over 26896.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3287, pruned_loss=0.08392, over 5659440.50 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3309, pruned_loss=0.08679, over 5730769.46 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3293, pruned_loss=0.084, over 5657194.84 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:37:07,099 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-13 10:38:00,312 INFO [train.py:968] (0/2) Epoch 26, batch 15650, giga_loss[loss=0.2816, simple_loss=0.3435, pruned_loss=0.1099, over 26870.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3308, pruned_loss=0.08487, over 5662612.94 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3309, pruned_loss=0.08683, over 5734962.68 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3312, pruned_loss=0.0848, over 5654414.52 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:38:29,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4788, 1.6699, 1.7106, 1.3103], device='cuda:0'), covar=tensor([0.1872, 0.2726, 0.1585, 0.2003], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0703, 0.0968, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 10:38:32,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.347e+02 1.609e+03 2.099e+03 2.973e+03 7.699e+03, threshold=4.198e+03, percent-clipped=16.0
+2023-03-13 10:38:36,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5691, 1.9031, 1.4508, 1.6329], device='cuda:0'), covar=tensor([0.2922, 0.2694, 0.3258, 0.2394], device='cuda:0'), in_proj_covar=tensor([0.1573, 0.1131, 0.1391, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 10:38:55,416 INFO [train.py:968] (0/2) Epoch 26, batch 15700, libri_loss[loss=0.2867, simple_loss=0.3559, pruned_loss=0.1087, over 29536.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3324, pruned_loss=0.08585, over 5674398.94 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3314, pruned_loss=0.08724, over 5738775.67 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3323, pruned_loss=0.08534, over 5661907.74 frames. ], batch size: 81, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:38:59,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1154761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:39:15,386 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1154773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:39:29,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1154784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:39:41,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1154794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:39:54,138 INFO [train.py:968] (0/2) Epoch 26, batch 15750, giga_loss[loss=0.2505, simple_loss=0.3278, pruned_loss=0.08657, over 28648.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3314, pruned_loss=0.08528, over 5684922.33 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3311, pruned_loss=0.08712, over 5741426.58 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3316, pruned_loss=0.08495, over 5671270.38 frames. ], batch size: 60, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:40:22,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.887e+02 1.515e+03 1.954e+03 2.794e+03 5.220e+03, threshold=3.909e+03, percent-clipped=6.0
+2023-03-13 10:40:30,306 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1154839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:40:50,482 INFO [train.py:968] (0/2) Epoch 26, batch 15800, libri_loss[loss=0.2288, simple_loss=0.3104, pruned_loss=0.0736, over 29557.00 frames. ], tot_loss[loss=0.248, simple_loss=0.329, pruned_loss=0.08347, over 5687363.69 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.331, pruned_loss=0.08702, over 5736496.90 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3292, pruned_loss=0.08317, over 5678215.28 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:41:48,446 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1154904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:41:51,421 INFO [train.py:968] (0/2) Epoch 26, batch 15850, giga_loss[loss=0.2325, simple_loss=0.3092, pruned_loss=0.07793, over 29019.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.327, pruned_loss=0.08294, over 5675856.33 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.331, pruned_loss=0.08702, over 5736496.90 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3272, pruned_loss=0.08271, over 5668736.01 frames. ], batch size: 285, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:41:51,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1154907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:42:15,030 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1154927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:42:18,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1154930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:42:22,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.701e+02 1.448e+03 1.818e+03 2.661e+03 7.338e+03, threshold=3.635e+03, percent-clipped=7.0
+2023-03-13 10:42:25,857 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1154936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:42:39,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3618, 3.6978, 1.5753, 1.5279], device='cuda:0'), covar=tensor([0.1049, 0.0299, 0.0933, 0.1403], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0563, 0.0403, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 10:42:48,853 INFO [train.py:968] (0/2) Epoch 26, batch 15900, giga_loss[loss=0.2899, simple_loss=0.3625, pruned_loss=0.1086, over 28862.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3262, pruned_loss=0.08269, over 5680938.89 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3303, pruned_loss=0.08665, over 5740735.71 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3269, pruned_loss=0.08272, over 5669842.94 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:42:51,100 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1154959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:43:50,235 INFO [train.py:968] (0/2) Epoch 26, batch 15950, libri_loss[loss=0.1992, simple_loss=0.277, pruned_loss=0.06071, over 28170.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3286, pruned_loss=0.08363, over 5678621.98 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3295, pruned_loss=0.08629, over 5741781.99 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3298, pruned_loss=0.08394, over 5668268.25 frames. ], batch size: 62, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:44:28,235 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.818e+02 1.452e+03 1.975e+03 2.404e+03 7.563e+03, threshold=3.950e+03, percent-clipped=6.0
+2023-03-13 10:44:56,560 INFO [train.py:968] (0/2) Epoch 26, batch 16000, giga_loss[loss=0.2407, simple_loss=0.3225, pruned_loss=0.07947, over 28910.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3289, pruned_loss=0.08415, over 5678187.22 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3292, pruned_loss=0.08611, over 5745146.18 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3301, pruned_loss=0.0845, over 5665841.68 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 10:45:51,681 INFO [train.py:968] (0/2) Epoch 26, batch 16050, giga_loss[loss=0.2747, simple_loss=0.3521, pruned_loss=0.09864, over 28917.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3307, pruned_loss=0.08503, over 5681154.89 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.329, pruned_loss=0.086, over 5745993.26 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3319, pruned_loss=0.08537, over 5669088.84 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 10:46:22,905 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.977e+02 1.456e+03 1.674e+03 2.336e+03 6.195e+03, threshold=3.348e+03, percent-clipped=5.0
+2023-03-13 10:46:37,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:46:41,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5199, 1.7006, 1.2457, 1.2529], device='cuda:0'), covar=tensor([0.1032, 0.0586, 0.1061, 0.1212], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0443, 0.0517, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 10:46:50,006 INFO [train.py:968] (0/2) Epoch 26, batch 16100, giga_loss[loss=0.2621, simple_loss=0.346, pruned_loss=0.0891, over 28498.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.334, pruned_loss=0.08611, over 5687236.58 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3288, pruned_loss=0.08593, over 5749283.15 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3351, pruned_loss=0.08643, over 5673338.33 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:47:00,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:47:27,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155192.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:47:42,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4432, 1.8833, 1.3101, 0.9220], device='cuda:0'), covar=tensor([0.5921, 0.3231, 0.4287, 0.6317], device='cuda:0'), in_proj_covar=tensor([0.1808, 0.1697, 0.1634, 0.1474], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 10:47:45,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:47:45,893 INFO [train.py:968] (0/2) Epoch 26, batch 16150, giga_loss[loss=0.2378, simple_loss=0.3249, pruned_loss=0.07537, over 28995.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3354, pruned_loss=0.08676, over 5679901.76 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.329, pruned_loss=0.08611, over 5740508.86 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3363, pruned_loss=0.08685, over 5674659.19 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:47:55,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:48:23,135 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.733e+02 1.558e+03 2.035e+03 3.059e+03 1.045e+04, threshold=4.070e+03, percent-clipped=21.0
+2023-03-13 10:48:52,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2418, 1.4554, 1.4240, 1.0915], device='cuda:0'), covar=tensor([0.1454, 0.2145, 0.1242, 0.1524], device='cuda:0'), in_proj_covar=tensor([0.0918, 0.0702, 0.0966, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 10:48:55,292 INFO [train.py:968] (0/2) Epoch 26, batch 16200, giga_loss[loss=0.2238, simple_loss=0.3092, pruned_loss=0.0692, over 28641.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3355, pruned_loss=0.08688, over 5676074.90 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3296, pruned_loss=0.08651, over 5733323.96 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3358, pruned_loss=0.0866, over 5678161.74 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:49:14,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7969, 4.6499, 4.4035, 2.1267], device='cuda:0'), covar=tensor([0.0491, 0.0660, 0.0732, 0.2049], device='cuda:0'), in_proj_covar=tensor([0.1259, 0.1157, 0.0978, 0.0727], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 10:49:38,320 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1155291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:49:42,163 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1155294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:49:58,140 INFO [train.py:968] (0/2) Epoch 26, batch 16250, giga_loss[loss=0.2673, simple_loss=0.352, pruned_loss=0.09126, over 29036.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.335, pruned_loss=0.0872, over 5687717.95 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.33, pruned_loss=0.08677, over 5736130.17 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3349, pruned_loss=0.08674, over 5686069.69 frames. ], batch size: 285, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:50:05,994 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1155312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:50:08,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1155315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:50:18,601 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1155323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:50:32,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.486e+02 1.401e+03 1.749e+03 2.702e+03 7.346e+03, threshold=3.499e+03, percent-clipped=11.0
+2023-03-13 10:50:43,219 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1155344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:50:57,147 INFO [train.py:968] (0/2) Epoch 26, batch 16300, giga_loss[loss=0.2532, simple_loss=0.3304, pruned_loss=0.08806, over 28899.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3331, pruned_loss=0.08655, over 5672660.58 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3293, pruned_loss=0.0865, over 5741309.78 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3338, pruned_loss=0.08646, over 5665012.56 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:50:57,681 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1155357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:51:01,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1155360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:51:37,162 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1155389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:51:58,152 INFO [train.py:968] (0/2) Epoch 26, batch 16350, libri_loss[loss=0.2277, simple_loss=0.3003, pruned_loss=0.0776, over 29573.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.332, pruned_loss=0.08682, over 5674707.43 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3294, pruned_loss=0.08675, over 5737833.54 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3326, pruned_loss=0.08655, over 5668899.28 frames. ], batch size: 74, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:52:30,033 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.714e+02 1.466e+03 1.785e+03 2.498e+03 6.010e+03, threshold=3.571e+03, percent-clipped=7.0
+2023-03-13 10:52:59,864 INFO [train.py:968] (0/2) Epoch 26, batch 16400, giga_loss[loss=0.2398, simple_loss=0.3264, pruned_loss=0.07659, over 28764.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3305, pruned_loss=0.08686, over 5673446.36 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.329, pruned_loss=0.08656, over 5740497.27 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3314, pruned_loss=0.08681, over 5665584.69 frames. ], batch size: 263, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 10:53:57,544 INFO [train.py:968] (0/2) Epoch 26, batch 16450, giga_loss[loss=0.2527, simple_loss=0.3384, pruned_loss=0.08352, over 28721.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3295, pruned_loss=0.08482, over 5675136.77 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3288, pruned_loss=0.0864, over 5740595.66 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3303, pruned_loss=0.08491, over 5667644.66 frames. ], batch size: 263, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 10:54:30,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.312e+02 1.354e+03 1.765e+03 2.404e+03 5.399e+03, threshold=3.530e+03, percent-clipped=10.0
+2023-03-13 10:54:52,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1194, 1.4417, 1.4143, 1.1118], device='cuda:0'), covar=tensor([0.2809, 0.2237, 0.1665, 0.2314], device='cuda:0'), in_proj_covar=tensor([0.1995, 0.1932, 0.1851, 0.1993], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 10:54:57,200 INFO [train.py:968] (0/2) Epoch 26, batch 16500, giga_loss[loss=0.2503, simple_loss=0.3469, pruned_loss=0.07686, over 28747.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3294, pruned_loss=0.08394, over 5676959.88 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3286, pruned_loss=0.08627, over 5742546.90 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3303, pruned_loss=0.08409, over 5668120.56 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 10:55:05,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155567.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 10:55:19,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:55:43,444 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:55:46,577 INFO [train.py:968] (0/2) Epoch 26, batch 16550, giga_loss[loss=0.2517, simple_loss=0.3433, pruned_loss=0.08009, over 28471.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3311, pruned_loss=0.08335, over 5659279.59 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3288, pruned_loss=0.08655, over 5718917.24 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3317, pruned_loss=0.08316, over 5669157.57 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:55:59,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-13 10:56:16,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.386e+03 1.837e+03 2.431e+03 7.936e+03, threshold=3.673e+03, percent-clipped=4.0
+2023-03-13 10:56:40,724 INFO [train.py:968] (0/2) Epoch 26, batch 16600, giga_loss[loss=0.242, simple_loss=0.33, pruned_loss=0.07697, over 29189.00 frames. ], tot_loss[loss=0.249, simple_loss=0.332, pruned_loss=0.08299, over 5664581.57 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3285, pruned_loss=0.08645, over 5718361.18 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3328, pruned_loss=0.08285, over 5671848.40 frames. ], batch size: 113, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:56:43,599 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-13 10:57:43,083 INFO [train.py:968] (0/2) Epoch 26, batch 16650, giga_loss[loss=0.3507, simple_loss=0.3875, pruned_loss=0.1569, over 26796.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.333, pruned_loss=0.08387, over 5671872.52 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3287, pruned_loss=0.08648, over 5719347.76 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3335, pruned_loss=0.08363, over 5675618.53 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:57:47,531 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1155710.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 10:57:50,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1155713.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:58:02,934 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1155724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:58:07,970 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1155727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:58:18,841 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.916e+02 1.535e+03 2.064e+03 3.042e+03 8.850e+03, threshold=4.128e+03, percent-clipped=12.0
+2023-03-13 10:58:24,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:58:27,524 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1155742.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 10:58:42,703 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1155756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:58:43,034 INFO [train.py:968] (0/2) Epoch 26, batch 16700, giga_loss[loss=0.2874, simple_loss=0.3726, pruned_loss=0.1011, over 28828.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3327, pruned_loss=0.08388, over 5671678.41 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3281, pruned_loss=0.08611, over 5724840.28 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3339, pruned_loss=0.08393, over 5667643.97 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:59:26,621 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 10:59:52,699 INFO [train.py:968] (0/2) Epoch 26, batch 16750, giga_loss[loss=0.2436, simple_loss=0.3339, pruned_loss=0.0766, over 28501.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3328, pruned_loss=0.08399, over 5670182.71 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3273, pruned_loss=0.08569, over 5723748.16 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3346, pruned_loss=0.0844, over 5666583.40 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:00:24,877 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.756e+02 1.417e+03 1.795e+03 2.395e+03 5.830e+03, threshold=3.590e+03, percent-clipped=5.0
+2023-03-13 11:00:36,477 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:00:59,406 INFO [train.py:968] (0/2) Epoch 26, batch 16800, giga_loss[loss=0.2508, simple_loss=0.3372, pruned_loss=0.0822, over 28693.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3321, pruned_loss=0.08272, over 5680195.69 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3271, pruned_loss=0.08559, over 5727802.66 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.334, pruned_loss=0.08308, over 5672157.31 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:01:19,825 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:01:51,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2994, 1.4715, 1.4166, 1.2865], device='cuda:0'), covar=tensor([0.2470, 0.2131, 0.1703, 0.2147], device='cuda:0'), in_proj_covar=tensor([0.1994, 0.1931, 0.1842, 0.1989], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 11:02:09,778 INFO [train.py:968] (0/2) Epoch 26, batch 16850, giga_loss[loss=0.335, simple_loss=0.3833, pruned_loss=0.1434, over 26835.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3347, pruned_loss=0.0843, over 5677433.83 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.327, pruned_loss=0.08545, over 5731329.02 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3364, pruned_loss=0.08467, over 5667103.32 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:02:29,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7240, 1.8444, 1.3031, 1.4586], device='cuda:0'), covar=tensor([0.0964, 0.0640, 0.0990, 0.1277], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0442, 0.0517, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 11:02:52,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.804e+02 1.470e+03 2.044e+03 2.841e+03 1.082e+04, threshold=4.087e+03, percent-clipped=19.0
+2023-03-13 11:03:14,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4648, 4.3118, 4.1104, 1.7864], device='cuda:0'), covar=tensor([0.0549, 0.0678, 0.0725, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.1251, 0.1149, 0.0969, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 11:03:17,251 INFO [train.py:968] (0/2) Epoch 26, batch 16900, libri_loss[loss=0.3021, simple_loss=0.3633, pruned_loss=0.1205, over 19413.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3367, pruned_loss=0.08439, over 5680048.12 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3272, pruned_loss=0.08559, over 5723949.61 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3379, pruned_loss=0.08454, over 5678863.87 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:03:18,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3208, 1.2242, 3.6720, 3.1581], device='cuda:0'), covar=tensor([0.1605, 0.2810, 0.0453, 0.1220], device='cuda:0'), in_proj_covar=tensor([0.0781, 0.0662, 0.0970, 0.0940], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 11:03:48,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:04:17,199 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1156000.pt
+2023-03-13 11:04:30,268 INFO [train.py:968] (0/2) Epoch 26, batch 16950, giga_loss[loss=0.2643, simple_loss=0.3429, pruned_loss=0.09278, over 29000.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3344, pruned_loss=0.08345, over 5681781.95 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3271, pruned_loss=0.08555, over 5724963.86 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3355, pruned_loss=0.0836, over 5679875.90 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:05:14,155 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.135e+02 1.335e+03 1.701e+03 2.379e+03 5.212e+03, threshold=3.401e+03, percent-clipped=5.0
+2023-03-13 11:05:41,912 INFO [train.py:968] (0/2) Epoch 26, batch 17000, giga_loss[loss=0.248, simple_loss=0.3323, pruned_loss=0.08183, over 28933.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3322, pruned_loss=0.08272, over 5686005.04 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3271, pruned_loss=0.08551, over 5726635.58 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3332, pruned_loss=0.08285, over 5682607.58 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:06:47,871 INFO [train.py:968] (0/2) Epoch 26, batch 17050, giga_loss[loss=0.2238, simple_loss=0.3128, pruned_loss=0.0674, over 27652.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3305, pruned_loss=0.08097, over 5688546.77 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3274, pruned_loss=0.08577, over 5721312.60 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3311, pruned_loss=0.08069, over 5688923.97 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 1.0
+2023-03-13 11:06:59,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:07:05,363 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:07:09,378 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1156124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:07:22,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.222e+02 1.325e+03 1.930e+03 2.747e+03 9.445e+03, threshold=3.861e+03, percent-clipped=12.0
+2023-03-13 11:07:37,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1242, 2.8469, 1.2259, 1.3893], device='cuda:0'), covar=tensor([0.1287, 0.0496, 0.1136, 0.1588], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0561, 0.0402, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 11:07:39,627 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1156149.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 11:07:43,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:07:47,139 INFO [train.py:968] (0/2) Epoch 26, batch 17100, giga_loss[loss=0.2606, simple_loss=0.343, pruned_loss=0.08907, over 28885.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3298, pruned_loss=0.08091, over 5687585.28 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3271, pruned_loss=0.08557, over 5717889.78 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3305, pruned_loss=0.08077, over 5690310.19 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 1.0
+2023-03-13 11:08:47,741 INFO [train.py:968] (0/2) Epoch 26, batch 17150, giga_loss[loss=0.2897, simple_loss=0.3662, pruned_loss=0.1066, over 28913.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3322, pruned_loss=0.08225, over 5681088.35 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3272, pruned_loss=0.08566, over 5716782.93 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3327, pruned_loss=0.08201, over 5683720.89 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 1.0
+2023-03-13 11:08:59,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:09:22,704 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.899e+02 1.455e+03 1.902e+03 2.979e+03 1.156e+04, threshold=3.804e+03, percent-clipped=17.0
+2023-03-13 11:09:33,897 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:09:44,553 INFO [train.py:968] (0/2) Epoch 26, batch 17200, libri_loss[loss=0.3043, simple_loss=0.3753, pruned_loss=0.1166, over 29553.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3343, pruned_loss=0.08395, over 5677539.65 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3272, pruned_loss=0.08565, over 5720199.67 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.335, pruned_loss=0.0837, over 5675417.56 frames. ], batch size: 89, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:09:44,927 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:09:46,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1156260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:10:18,428 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:10:41,150 INFO [train.py:968] (0/2) Epoch 26, batch 17250, giga_loss[loss=0.2256, simple_loss=0.3074, pruned_loss=0.07185, over 28393.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.332, pruned_loss=0.08378, over 5679724.70 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3271, pruned_loss=0.08563, over 5723078.11 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3327, pruned_loss=0.08357, over 5674873.74 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:11:15,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.572e+02 1.469e+03 1.870e+03 2.709e+03 6.991e+03, threshold=3.741e+03, percent-clipped=7.0
+2023-03-13 11:11:34,336 INFO [train.py:968] (0/2) Epoch 26, batch 17300, giga_loss[loss=0.2585, simple_loss=0.3369, pruned_loss=0.09001, over 28678.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3322, pruned_loss=0.08469, over 5683505.93 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3272, pruned_loss=0.08581, over 5726425.05 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3328, pruned_loss=0.08433, over 5675539.80 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:11:38,715 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:11:42,195 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2429, 0.8404, 0.8742, 1.4442], device='cuda:0'), covar=tensor([0.0783, 0.0421, 0.0393, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0120, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0073, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 11:11:42,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1156362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:11:59,436 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1156379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:12:11,787 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:12:11,850 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:12:15,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1156394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:12:30,069 INFO [train.py:968] (0/2) Epoch 26, batch 17350, giga_loss[loss=0.274, simple_loss=0.3479, pruned_loss=0.1001, over 27498.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3343, pruned_loss=0.08639, over 5692669.13 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3275, pruned_loss=0.086, over 5731391.90 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3346, pruned_loss=0.08589, over 5680759.60 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:12:47,351 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:13:02,556 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4660, 2.8034, 1.5641, 1.6034], device='cuda:0'), covar=tensor([0.0861, 0.0350, 0.0785, 0.1154], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0561, 0.0402, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 11:13:04,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.370e+02 1.436e+03 1.844e+03 2.414e+03 4.938e+03, threshold=3.687e+03, percent-clipped=7.0
+2023-03-13 11:13:22,623 INFO [train.py:968] (0/2) Epoch 26, batch 17400, giga_loss[loss=0.3056, simple_loss=0.3841, pruned_loss=0.1135, over 28883.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3424, pruned_loss=0.09098, over 5688864.24 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3275, pruned_loss=0.08587, over 5732403.79 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3428, pruned_loss=0.09075, over 5677715.30 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:14:04,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6483, 1.6928, 1.8279, 1.4167], device='cuda:0'), covar=tensor([0.1750, 0.2604, 0.1493, 0.1750], device='cuda:0'), in_proj_covar=tensor([0.0917, 0.0700, 0.0966, 0.0867], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 11:14:08,025 INFO [train.py:968] (0/2) Epoch 26, batch 17450, giga_loss[loss=0.2567, simple_loss=0.3435, pruned_loss=0.08501, over 28559.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3491, pruned_loss=0.09464, over 5694414.17 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3272, pruned_loss=0.08575, over 5733463.05 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3499, pruned_loss=0.09464, over 5684212.71 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:14:21,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156524.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 11:14:32,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5698, 5.0918, 1.9275, 2.0030], device='cuda:0'), covar=tensor([0.1016, 0.0204, 0.0918, 0.1310], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0561, 0.0402, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 11:14:33,805 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.81 vs. limit=5.0
+2023-03-13 11:14:33,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.086e+02 1.333e+03 1.713e+03 2.151e+03 5.944e+03, threshold=3.425e+03, percent-clipped=3.0
+2023-03-13 11:14:48,572 INFO [train.py:968] (0/2) Epoch 26, batch 17500, giga_loss[loss=0.2976, simple_loss=0.3747, pruned_loss=0.1103, over 28590.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3469, pruned_loss=0.09401, over 5694838.25 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3271, pruned_loss=0.0856, over 5731200.30 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3486, pruned_loss=0.09453, over 5686666.92 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:15:03,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1156571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:15:22,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-13 11:15:32,080 INFO [train.py:968] (0/2) Epoch 26, batch 17550, libri_loss[loss=0.2526, simple_loss=0.3277, pruned_loss=0.08872, over 29577.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3414, pruned_loss=0.09234, over 5693935.62 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3273, pruned_loss=0.08569, over 5734779.20 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3431, pruned_loss=0.09293, over 5682598.78 frames. ], batch size: 74, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:15:52,443 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5671, 2.2165, 1.6596, 0.8689], device='cuda:0'), covar=tensor([0.6694, 0.3393, 0.4448, 0.7141], device='cuda:0'), in_proj_covar=tensor([0.1810, 0.1701, 0.1641, 0.1476], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 11:15:56,904 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.813e+02 1.178e+03 1.528e+03 2.010e+03 5.691e+03, threshold=3.056e+03, percent-clipped=6.0
+2023-03-13 11:16:14,570 INFO [train.py:968] (0/2) Epoch 26, batch 17600, libri_loss[loss=0.2128, simple_loss=0.2901, pruned_loss=0.06775, over 29418.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3339, pruned_loss=0.08909, over 5689193.99 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.327, pruned_loss=0.08551, over 5736744.56 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3358, pruned_loss=0.08984, over 5677260.35 frames. ], batch size: 67, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:16:24,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156667.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 11:16:26,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1156670.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 11:16:50,413 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156699.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 11:16:58,549 INFO [train.py:968] (0/2) Epoch 26, batch 17650, giga_loss[loss=0.2391, simple_loss=0.3062, pruned_loss=0.08602, over 28848.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3258, pruned_loss=0.08556, over 5687546.34 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3272, pruned_loss=0.08565, over 5739068.47 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3271, pruned_loss=0.08603, over 5675363.46 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:17:27,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.073e+02 1.189e+03 1.667e+03 2.455e+03 5.292e+03, threshold=3.334e+03, percent-clipped=11.0
+2023-03-13 11:17:38,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:17:40,670 INFO [train.py:968] (0/2) Epoch 26, batch 17700, giga_loss[loss=0.2299, simple_loss=0.3024, pruned_loss=0.07873, over 28922.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3193, pruned_loss=0.08259, over 5697521.10 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3272, pruned_loss=0.08554, over 5743908.23 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3202, pruned_loss=0.08301, over 5681942.87 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:17:45,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4681, 1.5774, 1.4351, 1.6744], device='cuda:0'), covar=tensor([0.0762, 0.0360, 0.0345, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0073, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 11:18:22,169 INFO [train.py:968] (0/2) Epoch 26, batch 17750, giga_loss[loss=0.2207, simple_loss=0.3054, pruned_loss=0.06802, over 28534.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3142, pruned_loss=0.08029, over 5701061.19 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3271, pruned_loss=0.08541, over 5745372.89 frames. ], giga_tot_loss[loss=0.238, simple_loss=0.3147, pruned_loss=0.08065, over 5686307.48 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:18:47,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.928e+02 1.205e+03 1.539e+03 2.150e+03 4.846e+03, threshold=3.078e+03, percent-clipped=5.0
+2023-03-13 11:19:01,191 INFO [train.py:968] (0/2) Epoch 26, batch 17800, giga_loss[loss=0.1965, simple_loss=0.2804, pruned_loss=0.05629, over 29106.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3116, pruned_loss=0.07918, over 5685814.69 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3276, pruned_loss=0.08555, over 5728533.74 frames. ], giga_tot_loss[loss=0.2348, simple_loss=0.3112, pruned_loss=0.07917, over 5688237.43 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:19:35,602 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:19:37,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1156900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:19:41,685 INFO [train.py:968] (0/2) Epoch 26, batch 17850, giga_loss[loss=0.2142, simple_loss=0.2848, pruned_loss=0.07184, over 28611.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3094, pruned_loss=0.07852, over 5680734.14 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3277, pruned_loss=0.08553, over 5719944.82 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.3084, pruned_loss=0.07831, over 5689861.95 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:19:43,450 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1156908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:20:03,317 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:20:10,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.837e+02 1.089e+03 1.468e+03 1.993e+03 4.919e+03, threshold=2.936e+03, percent-clipped=8.0
+2023-03-13 11:20:14,645 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 11:20:16,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:20:25,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2981, 1.1619, 4.0228, 3.3389], device='cuda:0'), covar=tensor([0.1685, 0.3004, 0.0417, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0666, 0.0975, 0.0945], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 11:20:28,841 INFO [train.py:968] (0/2) Epoch 26, batch 17900, giga_loss[loss=0.185, simple_loss=0.2641, pruned_loss=0.05299, over 28744.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3062, pruned_loss=0.07717, over 5687007.37 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3277, pruned_loss=0.08555, over 5721761.67 frames. ], giga_tot_loss[loss=0.2295, simple_loss=0.3052, pruned_loss=0.07693, over 5692110.25 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:21:09,107 INFO [train.py:968] (0/2) Epoch 26, batch 17950, giga_loss[loss=0.2214, simple_loss=0.2831, pruned_loss=0.07988, over 23894.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3036, pruned_loss=0.07601, over 5689984.36 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3279, pruned_loss=0.08548, over 5726421.88 frames. ], giga_tot_loss[loss=0.2267, simple_loss=0.3021, pruned_loss=0.07568, over 5689309.47 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:21:35,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.746e+02 1.208e+03 1.703e+03 2.389e+03 6.738e+03, threshold=3.406e+03, percent-clipped=9.0
+2023-03-13 11:21:50,266 INFO [train.py:968] (0/2) Epoch 26, batch 18000, giga_loss[loss=0.2134, simple_loss=0.2926, pruned_loss=0.06708, over 28762.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3021, pruned_loss=0.07558, over 5687834.53 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3279, pruned_loss=0.08531, over 5725974.04 frames. ], giga_tot_loss[loss=0.2252, simple_loss=0.3001, pruned_loss=0.07515, over 5686691.33 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:21:50,270 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 11:21:54,474 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2633, 1.5598, 1.4822, 1.2369], device='cuda:0'), covar=tensor([0.3066, 0.2613, 0.1684, 0.2501], device='cuda:0'), in_proj_covar=tensor([0.2013, 0.1947, 0.1865, 0.2010], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 11:21:57,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3083, 3.0831, 1.3818, 1.5015], device='cuda:0'), covar=tensor([0.1125, 0.0413, 0.1030, 0.1483], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0559, 0.0401, 0.0438], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 11:21:58,990 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2012, simple_loss=0.3073, pruned_loss=0.04753, over 944034.00 frames. 
+2023-03-13 11:21:58,991 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 11:22:26,725 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1157089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:22:29,478 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1157092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:22:42,123 INFO [train.py:968] (0/2) Epoch 26, batch 18050, giga_loss[loss=0.193, simple_loss=0.2565, pruned_loss=0.06475, over 23833.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2991, pruned_loss=0.07411, over 5688546.87 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3283, pruned_loss=0.0853, over 5728996.71 frames. ], giga_tot_loss[loss=0.2219, simple_loss=0.2966, pruned_loss=0.07356, over 5684102.23 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:22:53,427 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1157121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:23:08,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.433e+02 1.066e+03 1.338e+03 1.750e+03 5.963e+03, threshold=2.675e+03, percent-clipped=2.0
+2023-03-13 11:23:23,470 INFO [train.py:968] (0/2) Epoch 26, batch 18100, giga_loss[loss=0.2203, simple_loss=0.2879, pruned_loss=0.07642, over 27684.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2976, pruned_loss=0.07351, over 5688836.71 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3289, pruned_loss=0.08548, over 5730943.29 frames. ], giga_tot_loss[loss=0.2196, simple_loss=0.2941, pruned_loss=0.07253, over 5682102.89 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:24:06,013 INFO [train.py:968] (0/2) Epoch 26, batch 18150, giga_loss[loss=0.201, simple_loss=0.2693, pruned_loss=0.06633, over 27739.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2958, pruned_loss=0.0728, over 5659340.09 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3288, pruned_loss=0.08535, over 5707904.25 frames. ], giga_tot_loss[loss=0.2173, simple_loss=0.2915, pruned_loss=0.07157, over 5674046.37 frames. ], batch size: 474, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:24:14,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5111, 3.2809, 1.6340, 1.6165], device='cuda:0'), covar=tensor([0.0984, 0.0364, 0.0885, 0.1323], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0560, 0.0401, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 11:24:23,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4985, 1.8627, 1.6993, 1.3276], device='cuda:0'), covar=tensor([0.1963, 0.2781, 0.1731, 0.1997], device='cuda:0'), in_proj_covar=tensor([0.0929, 0.0710, 0.0978, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 11:24:32,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.898e+02 1.159e+03 1.409e+03 1.785e+03 3.603e+03, threshold=2.819e+03, percent-clipped=4.0
+2023-03-13 11:24:47,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7118, 1.9388, 1.6249, 1.8110], device='cuda:0'), covar=tensor([0.2559, 0.2677, 0.2871, 0.2703], device='cuda:0'), in_proj_covar=tensor([0.1572, 0.1129, 0.1389, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 11:24:47,974 INFO [train.py:968] (0/2) Epoch 26, batch 18200, giga_loss[loss=0.2405, simple_loss=0.3149, pruned_loss=0.08306, over 27679.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2968, pruned_loss=0.07359, over 5666841.98 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3289, pruned_loss=0.08523, over 5713302.28 frames. ], giga_tot_loss[loss=0.2183, simple_loss=0.2921, pruned_loss=0.07229, over 5672361.10 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:25:13,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1157283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:25:36,110 INFO [train.py:968] (0/2) Epoch 26, batch 18250, libri_loss[loss=0.2843, simple_loss=0.3592, pruned_loss=0.1047, over 28649.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3078, pruned_loss=0.07938, over 5663071.48 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3295, pruned_loss=0.0856, over 5706344.37 frames. ], giga_tot_loss[loss=0.2293, simple_loss=0.303, pruned_loss=0.0778, over 5672400.23 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:26:04,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.557e+02 1.453e+03 1.828e+03 2.533e+03 4.748e+03, threshold=3.656e+03, percent-clipped=17.0
+2023-03-13 11:26:17,792 INFO [train.py:968] (0/2) Epoch 26, batch 18300, giga_loss[loss=0.3148, simple_loss=0.3849, pruned_loss=0.1224, over 28344.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3218, pruned_loss=0.0868, over 5672414.11 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3291, pruned_loss=0.08527, over 5708833.94 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.318, pruned_loss=0.08579, over 5676707.44 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:26:32,711 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-13 11:26:43,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2952, 1.3363, 1.2386, 1.5474], device='cuda:0'), covar=tensor([0.0805, 0.0379, 0.0337, 0.1039], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 11:26:56,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7124, 1.8964, 1.7725, 1.6983], device='cuda:0'), covar=tensor([0.2036, 0.2277, 0.2404, 0.2187], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0744, 0.0713, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 11:27:00,659 INFO [train.py:968] (0/2) Epoch 26, batch 18350, giga_loss[loss=0.2496, simple_loss=0.3369, pruned_loss=0.0812, over 29066.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3311, pruned_loss=0.09141, over 5678231.55 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3288, pruned_loss=0.08506, over 5712515.90 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3283, pruned_loss=0.09089, over 5677763.19 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:27:05,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6078, 1.7249, 1.6600, 1.4282], device='cuda:0'), covar=tensor([0.3177, 0.2829, 0.2451, 0.2900], device='cuda:0'), in_proj_covar=tensor([0.2018, 0.1954, 0.1872, 0.2014], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 11:27:15,471 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1157426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:27:17,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1157429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:27:26,700 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.719e+02 1.306e+03 1.740e+03 2.118e+03 4.720e+03, threshold=3.479e+03, percent-clipped=7.0
+2023-03-13 11:27:41,644 INFO [train.py:968] (0/2) Epoch 26, batch 18400, giga_loss[loss=0.2544, simple_loss=0.3426, pruned_loss=0.08313, over 28619.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3365, pruned_loss=0.09284, over 5682249.72 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3288, pruned_loss=0.08492, over 5715869.54 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3344, pruned_loss=0.09269, over 5678261.16 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:27:42,448 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1157458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:28:23,013 INFO [train.py:968] (0/2) Epoch 26, batch 18450, giga_loss[loss=0.2841, simple_loss=0.3637, pruned_loss=0.1022, over 27683.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3397, pruned_loss=0.0933, over 5678024.88 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3293, pruned_loss=0.08504, over 5706038.81 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3378, pruned_loss=0.09321, over 5683381.62 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:28:43,486 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1704, 1.6361, 1.2662, 0.4394], device='cuda:0'), covar=tensor([0.5121, 0.2933, 0.4184, 0.6539], device='cuda:0'), in_proj_covar=tensor([0.1805, 0.1696, 0.1639, 0.1473], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 11:28:51,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.219e+02 1.251e+03 1.492e+03 1.852e+03 4.302e+03, threshold=2.985e+03, percent-clipped=5.0
+2023-03-13 11:29:02,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4869, 1.7291, 1.4679, 1.3435], device='cuda:0'), covar=tensor([0.2487, 0.2393, 0.2592, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.1565, 0.1127, 0.1384, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 11:29:05,475 INFO [train.py:968] (0/2) Epoch 26, batch 18500, giga_loss[loss=0.2365, simple_loss=0.3204, pruned_loss=0.07635, over 28370.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3416, pruned_loss=0.09402, over 5652356.45 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3297, pruned_loss=0.08522, over 5693680.57 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3399, pruned_loss=0.09404, over 5666248.33 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:29:20,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7036, 1.5923, 1.9238, 1.5192], device='cuda:0'), covar=tensor([0.1533, 0.2051, 0.1255, 0.1605], device='cuda:0'), in_proj_covar=tensor([0.0925, 0.0707, 0.0973, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 11:29:47,775 INFO [train.py:968] (0/2) Epoch 26, batch 18550, giga_loss[loss=0.3065, simple_loss=0.3738, pruned_loss=0.1196, over 28990.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3431, pruned_loss=0.09522, over 5664223.12 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3298, pruned_loss=0.08503, over 5700329.86 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3422, pruned_loss=0.09579, over 5667794.41 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:29:49,745 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 11:29:57,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7691, 1.0829, 2.9301, 2.8016], device='cuda:0'), covar=tensor([0.1758, 0.2674, 0.0576, 0.1065], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0663, 0.0973, 0.0944], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 11:30:15,973 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.275e+03 1.584e+03 2.044e+03 4.657e+03, threshold=3.168e+03, percent-clipped=7.0
+2023-03-13 11:30:31,907 INFO [train.py:968] (0/2) Epoch 26, batch 18600, libri_loss[loss=0.2819, simple_loss=0.3557, pruned_loss=0.104, over 19983.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3457, pruned_loss=0.09737, over 5660885.58 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3302, pruned_loss=0.08521, over 5696902.99 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.345, pruned_loss=0.09795, over 5666947.09 frames. ], batch size: 187, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:30:57,655 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2445, 1.5648, 1.5749, 1.4450], device='cuda:0'), covar=tensor([0.2157, 0.1680, 0.2244, 0.1856], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0747, 0.0717, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 11:31:13,280 INFO [train.py:968] (0/2) Epoch 26, batch 18650, giga_loss[loss=0.2802, simple_loss=0.3613, pruned_loss=0.09952, over 28786.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3493, pruned_loss=0.09907, over 5658993.05 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3306, pruned_loss=0.08536, over 5689491.33 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3486, pruned_loss=0.09955, over 5669623.43 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:31:40,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.461e+02 1.310e+03 1.725e+03 2.741e+03 8.140e+03, threshold=3.450e+03, percent-clipped=15.0
+2023-03-13 11:31:53,618 INFO [train.py:968] (0/2) Epoch 26, batch 18700, giga_loss[loss=0.2512, simple_loss=0.3428, pruned_loss=0.07979, over 28383.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.351, pruned_loss=0.09891, over 5671529.21 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3303, pruned_loss=0.08519, over 5692749.60 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.351, pruned_loss=0.09972, over 5676433.63 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:32:01,463 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-13 11:32:16,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5312, 1.7423, 1.4452, 1.4854], device='cuda:0'), covar=tensor([0.2891, 0.2958, 0.3360, 0.2475], device='cuda:0'), in_proj_covar=tensor([0.1567, 0.1128, 0.1383, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 11:32:18,783 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6895, 1.3725, 4.5320, 3.8265], device='cuda:0'), covar=tensor([0.1958, 0.3254, 0.0648, 0.0910], device='cuda:0'), in_proj_covar=tensor([0.0780, 0.0665, 0.0973, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 11:32:20,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1157788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:32:35,373 INFO [train.py:968] (0/2) Epoch 26, batch 18750, libri_loss[loss=0.2712, simple_loss=0.3458, pruned_loss=0.09833, over 19923.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3529, pruned_loss=0.09955, over 5667065.38 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3304, pruned_loss=0.08512, over 5687493.02 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3534, pruned_loss=0.1006, over 5675861.20 frames. ], batch size: 187, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:32:39,033 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-13 11:32:58,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.095e+02 1.325e+03 1.530e+03 1.900e+03 3.972e+03, threshold=3.060e+03, percent-clipped=4.0
+2023-03-13 11:33:14,686 INFO [train.py:968] (0/2) Epoch 26, batch 18800, giga_loss[loss=0.354, simple_loss=0.3865, pruned_loss=0.1607, over 23513.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3535, pruned_loss=0.099, over 5672841.19 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.331, pruned_loss=0.08532, over 5688784.74 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3544, pruned_loss=0.1003, over 5678828.72 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:33:53,025 INFO [train.py:968] (0/2) Epoch 26, batch 18850, giga_loss[loss=0.2766, simple_loss=0.3622, pruned_loss=0.0955, over 28639.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3538, pruned_loss=0.09774, over 5690008.31 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3317, pruned_loss=0.08566, over 5692989.64 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3543, pruned_loss=0.09869, over 5690893.46 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:34:21,669 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.537e+02 1.321e+03 1.740e+03 2.400e+03 7.480e+03, threshold=3.480e+03, percent-clipped=14.0
+2023-03-13 11:34:34,901 INFO [train.py:968] (0/2) Epoch 26, batch 18900, giga_loss[loss=0.26, simple_loss=0.3465, pruned_loss=0.08681, over 28398.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3518, pruned_loss=0.0954, over 5698136.53 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3316, pruned_loss=0.0856, over 5695011.14 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3525, pruned_loss=0.0963, over 5697040.40 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:35:08,981 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1158000.pt
+2023-03-13 11:35:14,960 INFO [train.py:968] (0/2) Epoch 26, batch 18950, giga_loss[loss=0.3006, simple_loss=0.3642, pruned_loss=0.1185, over 28884.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3514, pruned_loss=0.0955, over 5703123.37 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3317, pruned_loss=0.08564, over 5698356.58 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.352, pruned_loss=0.09629, over 5699267.61 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:35:39,869 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.1319, 5.8892, 5.5718, 3.0412], device='cuda:0'), covar=tensor([0.0386, 0.0571, 0.0651, 0.1629], device='cuda:0'), in_proj_covar=tensor([0.1250, 0.1155, 0.0974, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 11:35:44,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.150e+02 1.251e+03 1.566e+03 2.021e+03 7.181e+03, threshold=3.133e+03, percent-clipped=6.0
+2023-03-13 11:35:58,664 INFO [train.py:968] (0/2) Epoch 26, batch 19000, giga_loss[loss=0.3087, simple_loss=0.3773, pruned_loss=0.1201, over 28594.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.354, pruned_loss=0.09907, over 5708438.28 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.332, pruned_loss=0.08567, over 5700442.92 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3546, pruned_loss=0.09987, over 5703744.08 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:36:07,768 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-13 11:36:11,934 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:36:42,576 INFO [train.py:968] (0/2) Epoch 26, batch 19050, giga_loss[loss=0.2423, simple_loss=0.3205, pruned_loss=0.08203, over 28550.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3568, pruned_loss=0.1036, over 5713874.69 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3321, pruned_loss=0.08565, over 5706737.73 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3578, pruned_loss=0.1047, over 5704774.07 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:37:03,735 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 11:37:08,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.145e+03 1.593e+03 1.966e+03 2.590e+03 5.805e+03, threshold=3.932e+03, percent-clipped=10.0
+2023-03-13 11:37:20,127 INFO [train.py:968] (0/2) Epoch 26, batch 19100, giga_loss[loss=0.3109, simple_loss=0.3781, pruned_loss=0.1219, over 28887.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3565, pruned_loss=0.1045, over 5707584.15 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3332, pruned_loss=0.08632, over 5702813.60 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.357, pruned_loss=0.1052, over 5703440.63 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:37:27,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1158163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:37:35,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4345, 1.6758, 1.6563, 1.2437], device='cuda:0'), covar=tensor([0.1994, 0.3011, 0.1751, 0.2059], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0706, 0.0970, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 11:38:03,287 INFO [train.py:968] (0/2) Epoch 26, batch 19150, giga_loss[loss=0.2381, simple_loss=0.3192, pruned_loss=0.07849, over 28368.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3531, pruned_loss=0.1032, over 5700116.99 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3333, pruned_loss=0.08629, over 5704100.97 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.354, pruned_loss=0.1043, over 5695515.67 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:38:20,919 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7535, 1.8992, 1.4294, 1.4109], device='cuda:0'), covar=tensor([0.1007, 0.0625, 0.1035, 0.1181], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0445, 0.0523, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 11:38:30,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.351e+03 1.640e+03 2.138e+03 5.529e+03, threshold=3.280e+03, percent-clipped=3.0
+2023-03-13 11:38:45,747 INFO [train.py:968] (0/2) Epoch 26, batch 19200, giga_loss[loss=0.2612, simple_loss=0.3333, pruned_loss=0.09461, over 28637.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3507, pruned_loss=0.1016, over 5709476.24 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3334, pruned_loss=0.08618, over 5709589.21 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3518, pruned_loss=0.103, over 5700808.06 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:39:06,419 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-13 11:39:28,498 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1158306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:39:28,838 INFO [train.py:968] (0/2) Epoch 26, batch 19250, giga_loss[loss=0.2897, simple_loss=0.363, pruned_loss=0.1082, over 28581.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3496, pruned_loss=0.1002, over 5714397.84 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3335, pruned_loss=0.08617, over 5710992.99 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3506, pruned_loss=0.1015, over 5706252.05 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:39:30,480 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1158309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:39:53,726 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1158338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:39:56,533 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:39:57,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.657e+02 1.312e+03 1.703e+03 2.292e+03 5.125e+03, threshold=3.407e+03, percent-clipped=8.0
+2023-03-13 11:40:11,869 INFO [train.py:968] (0/2) Epoch 26, batch 19300, libri_loss[loss=0.2477, simple_loss=0.3357, pruned_loss=0.07991, over 29510.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3472, pruned_loss=0.09823, over 5711417.69 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3341, pruned_loss=0.08629, over 5718200.68 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3481, pruned_loss=0.09964, over 5698239.71 frames. ], batch size: 81, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:40:39,933 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:40:51,445 INFO [train.py:968] (0/2) Epoch 26, batch 19350, giga_loss[loss=0.2232, simple_loss=0.2985, pruned_loss=0.0739, over 28848.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3424, pruned_loss=0.0953, over 5698473.91 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3347, pruned_loss=0.08652, over 5714476.38 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3428, pruned_loss=0.09656, over 5690244.18 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:41:26,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.779e+02 1.091e+03 1.302e+03 1.729e+03 6.244e+03, threshold=2.603e+03, percent-clipped=3.0
+2023-03-13 11:41:28,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1158445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:41:37,886 INFO [train.py:968] (0/2) Epoch 26, batch 19400, giga_loss[loss=0.2292, simple_loss=0.2888, pruned_loss=0.08481, over 23421.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.336, pruned_loss=0.09207, over 5676819.77 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3347, pruned_loss=0.08643, over 5708391.81 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3364, pruned_loss=0.09326, over 5675845.27 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:42:06,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2143, 5.0935, 4.8015, 2.3647], device='cuda:0'), covar=tensor([0.0408, 0.0490, 0.0564, 0.1848], device='cuda:0'), in_proj_covar=tensor([0.1251, 0.1156, 0.0971, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 11:42:11,146 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.33 vs. limit=5.0
+2023-03-13 11:42:23,086 INFO [train.py:968] (0/2) Epoch 26, batch 19450, giga_loss[loss=0.2108, simple_loss=0.2733, pruned_loss=0.07416, over 23393.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3321, pruned_loss=0.09012, over 5658810.67 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3355, pruned_loss=0.08671, over 5697128.13 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3317, pruned_loss=0.09105, over 5666838.12 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:42:31,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2378, 4.0962, 3.8747, 2.1019], device='cuda:0'), covar=tensor([0.0610, 0.0761, 0.0706, 0.1891], device='cuda:0'), in_proj_covar=tensor([0.1251, 0.1156, 0.0970, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 11:42:44,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5996, 1.8228, 1.5001, 1.7267], device='cuda:0'), covar=tensor([0.2792, 0.2864, 0.3174, 0.2576], device='cuda:0'), in_proj_covar=tensor([0.1563, 0.1127, 0.1381, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 11:42:50,642 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:42:55,181 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.828e+02 1.143e+03 1.493e+03 2.006e+03 5.851e+03, threshold=2.986e+03, percent-clipped=14.0
+2023-03-13 11:43:07,445 INFO [train.py:968] (0/2) Epoch 26, batch 19500, giga_loss[loss=0.2528, simple_loss=0.335, pruned_loss=0.08529, over 28881.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3307, pruned_loss=0.08974, over 5645240.61 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3358, pruned_loss=0.08684, over 5701008.13 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.33, pruned_loss=0.09041, over 5647234.71 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:43:28,515 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-13 11:43:35,564 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1158588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:43:37,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1158591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:43:51,109 INFO [train.py:968] (0/2) Epoch 26, batch 19550, giga_loss[loss=0.2845, simple_loss=0.3459, pruned_loss=0.1115, over 28700.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3313, pruned_loss=0.08988, over 5656367.91 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3358, pruned_loss=0.08677, over 5701697.07 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3307, pruned_loss=0.09052, over 5656691.15 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:44:04,864 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1158620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:44:15,070 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:44:23,060 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.918e+02 1.132e+03 1.376e+03 1.768e+03 5.456e+03, threshold=2.753e+03, percent-clipped=6.0
+2023-03-13 11:44:32,228 INFO [train.py:968] (0/2) Epoch 26, batch 19600, giga_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1115, over 28352.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3321, pruned_loss=0.09009, over 5670074.41 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3362, pruned_loss=0.0868, over 5708026.33 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3311, pruned_loss=0.09069, over 5662998.37 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:44:42,882 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-13 11:44:48,509 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 11:45:03,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4244, 1.6112, 1.3787, 1.5022], device='cuda:0'), covar=tensor([0.0795, 0.0335, 0.0344, 0.0894], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 11:45:13,977 INFO [train.py:968] (0/2) Epoch 26, batch 19650, giga_loss[loss=0.2386, simple_loss=0.3087, pruned_loss=0.08423, over 28735.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3313, pruned_loss=0.09015, over 5677475.83 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3364, pruned_loss=0.08702, over 5709974.96 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3303, pruned_loss=0.09045, over 5669831.95 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:45:20,165 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1158715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:45:42,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.354e+02 1.160e+03 1.415e+03 1.844e+03 4.337e+03, threshold=2.831e+03, percent-clipped=7.0
+2023-03-13 11:45:54,291 INFO [train.py:968] (0/2) Epoch 26, batch 19700, giga_loss[loss=0.2134, simple_loss=0.2925, pruned_loss=0.06718, over 28778.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3287, pruned_loss=0.08904, over 5683474.80 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3364, pruned_loss=0.08704, over 5710824.08 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3278, pruned_loss=0.08926, over 5676556.86 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:46:00,901 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1158765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:46:34,774 INFO [train.py:968] (0/2) Epoch 26, batch 19750, giga_loss[loss=0.2159, simple_loss=0.2944, pruned_loss=0.06869, over 28982.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3261, pruned_loss=0.08743, over 5697608.19 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3368, pruned_loss=0.08708, over 5712244.62 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.325, pruned_loss=0.08758, over 5690738.34 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:47:05,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.460e+02 1.110e+03 1.411e+03 1.855e+03 7.629e+03, threshold=2.821e+03, percent-clipped=7.0
+2023-03-13 11:47:15,783 INFO [train.py:968] (0/2) Epoch 26, batch 19800, giga_loss[loss=0.2387, simple_loss=0.3116, pruned_loss=0.0829, over 28986.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3234, pruned_loss=0.08618, over 5700423.85 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3369, pruned_loss=0.08703, over 5714818.53 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3222, pruned_loss=0.08635, over 5692395.33 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:47:16,862 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1158858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:47:17,026 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-13 11:47:19,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1158861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:47:42,770 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1158890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:47:46,667 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1627, 1.2658, 1.1540, 0.8989], device='cuda:0'), covar=tensor([0.1115, 0.0559, 0.1097, 0.1151], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0448, 0.0524, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 11:47:56,069 INFO [train.py:968] (0/2) Epoch 26, batch 19850, giga_loss[loss=0.2235, simple_loss=0.2941, pruned_loss=0.0764, over 28125.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3212, pruned_loss=0.08523, over 5711242.89 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3376, pruned_loss=0.08721, over 5718899.86 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3194, pruned_loss=0.08517, over 5701047.73 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:47:56,566 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 11:47:57,032 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1158908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:47:59,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1158911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:47:59,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1158911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:48:10,992 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-13 11:48:13,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:48:23,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2327, 1.8322, 1.4971, 0.4295], device='cuda:0'), covar=tensor([0.4520, 0.2617, 0.4297, 0.6281], device='cuda:0'), in_proj_covar=tensor([0.1805, 0.1697, 0.1634, 0.1470], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 11:48:23,643 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1158940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:48:24,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3990, 1.3057, 1.3082, 1.6319], device='cuda:0'), covar=tensor([0.0792, 0.0373, 0.0342, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 11:48:25,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.089e+02 1.080e+03 1.255e+03 1.597e+03 3.278e+03, threshold=2.509e+03, percent-clipped=2.0
+2023-03-13 11:48:36,380 INFO [train.py:968] (0/2) Epoch 26, batch 19900, giga_loss[loss=0.2671, simple_loss=0.3381, pruned_loss=0.09804, over 28835.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3198, pruned_loss=0.08441, over 5721266.62 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.338, pruned_loss=0.08722, over 5722046.75 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3176, pruned_loss=0.0843, over 5710239.33 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:49:16,633 INFO [train.py:968] (0/2) Epoch 26, batch 19950, giga_loss[loss=0.2341, simple_loss=0.3082, pruned_loss=0.08001, over 28869.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3178, pruned_loss=0.08342, over 5718866.68 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3377, pruned_loss=0.08704, over 5726602.31 frames. ], giga_tot_loss[loss=0.2413, simple_loss=0.3159, pruned_loss=0.0834, over 5706104.24 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:49:16,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:49:46,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.280e+02 1.036e+03 1.320e+03 1.885e+03 6.164e+03, threshold=2.639e+03, percent-clipped=12.0
+2023-03-13 11:49:54,989 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:49:56,659 INFO [train.py:968] (0/2) Epoch 26, batch 20000, giga_loss[loss=0.2164, simple_loss=0.2908, pruned_loss=0.07101, over 28645.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3164, pruned_loss=0.0825, over 5722645.64 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3379, pruned_loss=0.08704, over 5729050.48 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3143, pruned_loss=0.08241, over 5710125.92 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:49:56,954 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:50:15,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 11:50:17,750 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:50:34,053 INFO [train.py:968] (0/2) Epoch 26, batch 20050, giga_loss[loss=0.2851, simple_loss=0.3558, pruned_loss=0.1072, over 27928.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3166, pruned_loss=0.08265, over 5724500.09 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.338, pruned_loss=0.087, over 5731804.93 frames. ], giga_tot_loss[loss=0.2398, simple_loss=0.3146, pruned_loss=0.08255, over 5712062.04 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:51:06,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.487e+02 1.108e+03 1.334e+03 1.736e+03 5.198e+03, threshold=2.667e+03, percent-clipped=9.0
+2023-03-13 11:51:11,162 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:51:13,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:51:17,593 INFO [train.py:968] (0/2) Epoch 26, batch 20100, giga_loss[loss=0.2582, simple_loss=0.337, pruned_loss=0.08966, over 29001.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3216, pruned_loss=0.08536, over 5720813.82 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3384, pruned_loss=0.08711, over 5732622.12 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3193, pruned_loss=0.08516, over 5709852.31 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:51:38,903 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:51:40,813 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:52:05,275 INFO [train.py:968] (0/2) Epoch 26, batch 20150, giga_loss[loss=0.2692, simple_loss=0.3458, pruned_loss=0.09637, over 28879.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3284, pruned_loss=0.09002, over 5702338.32 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3387, pruned_loss=0.08714, over 5732924.44 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3262, pruned_loss=0.08983, over 5693236.08 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:52:14,614 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6016, 1.7508, 1.8036, 1.5927], device='cuda:0'), covar=tensor([0.2057, 0.2218, 0.2255, 0.2137], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0754, 0.0723, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 11:52:42,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.879e+02 1.401e+03 1.702e+03 2.410e+03 5.603e+03, threshold=3.405e+03, percent-clipped=14.0
+2023-03-13 11:52:55,184 INFO [train.py:968] (0/2) Epoch 26, batch 20200, giga_loss[loss=0.2755, simple_loss=0.3517, pruned_loss=0.09963, over 28879.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3383, pruned_loss=0.09659, over 5700243.18 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3392, pruned_loss=0.08727, over 5735262.00 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3359, pruned_loss=0.09643, over 5690449.08 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:53:36,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:53:38,962 INFO [train.py:968] (0/2) Epoch 26, batch 20250, giga_loss[loss=0.2626, simple_loss=0.3389, pruned_loss=0.09318, over 28892.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3428, pruned_loss=0.09832, over 5695425.77 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3394, pruned_loss=0.08728, over 5733352.05 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3408, pruned_loss=0.09835, over 5688616.54 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:53:44,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3639, 1.4961, 1.5972, 1.2039], device='cuda:0'), covar=tensor([0.1707, 0.2470, 0.1448, 0.1718], device='cuda:0'), in_proj_covar=tensor([0.0925, 0.0708, 0.0973, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 11:53:50,913 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4835, 1.5722, 1.7305, 1.2924], device='cuda:0'), covar=tensor([0.1687, 0.2595, 0.1469, 0.1778], device='cuda:0'), in_proj_covar=tensor([0.0925, 0.0707, 0.0972, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 11:54:13,177 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.101e+02 1.359e+03 1.781e+03 2.424e+03 9.084e+03, threshold=3.562e+03, percent-clipped=12.0
+2023-03-13 11:54:24,435 INFO [train.py:968] (0/2) Epoch 26, batch 20300, giga_loss[loss=0.285, simple_loss=0.3615, pruned_loss=0.1042, over 28807.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3475, pruned_loss=0.1, over 5684533.04 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3392, pruned_loss=0.08713, over 5725232.70 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3462, pruned_loss=0.1005, over 5686065.70 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:54:34,425 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:54:39,656 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:55:07,500 INFO [train.py:968] (0/2) Epoch 26, batch 20350, giga_loss[loss=0.25, simple_loss=0.3373, pruned_loss=0.08134, over 29011.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3507, pruned_loss=0.1009, over 5687839.50 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3391, pruned_loss=0.08698, over 5727245.71 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3501, pruned_loss=0.1019, over 5685758.54 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:55:18,451 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-13 11:55:40,231 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.211e+02 1.251e+03 1.619e+03 2.204e+03 1.102e+04, threshold=3.238e+03, percent-clipped=5.0
+2023-03-13 11:55:42,109 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:55:43,445 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159449.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 11:55:44,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:55:48,944 INFO [train.py:968] (0/2) Epoch 26, batch 20400, libri_loss[loss=0.2284, simple_loss=0.307, pruned_loss=0.07489, over 29655.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.352, pruned_loss=0.1013, over 5694635.34 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.339, pruned_loss=0.08697, over 5729175.68 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3517, pruned_loss=0.1022, over 5690733.62 frames. ], batch size: 69, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:56:00,591 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.63 vs. limit=5.0
+2023-03-13 11:56:07,132 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:56:10,955 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:56:30,618 INFO [train.py:968] (0/2) Epoch 26, batch 20450, giga_loss[loss=0.2598, simple_loss=0.3402, pruned_loss=0.08969, over 28231.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3469, pruned_loss=0.0972, over 5696526.61 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3393, pruned_loss=0.0871, over 5732695.68 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3467, pruned_loss=0.09821, over 5689072.95 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:56:59,319 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.230e+02 1.278e+03 1.713e+03 2.084e+03 6.022e+03, threshold=3.426e+03, percent-clipped=6.0
+2023-03-13 11:57:07,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:57:09,975 INFO [train.py:968] (0/2) Epoch 26, batch 20500, giga_loss[loss=0.2906, simple_loss=0.3674, pruned_loss=0.1069, over 27962.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3463, pruned_loss=0.09627, over 5693482.52 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3396, pruned_loss=0.0872, over 5731470.09 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.346, pruned_loss=0.09719, over 5687806.94 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:57:50,897 INFO [train.py:968] (0/2) Epoch 26, batch 20550, libri_loss[loss=0.2147, simple_loss=0.3049, pruned_loss=0.06226, over 29562.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3467, pruned_loss=0.09605, over 5698483.12 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3399, pruned_loss=0.08739, over 5737195.49 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3465, pruned_loss=0.09692, over 5687505.93 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:58:23,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.107e+02 1.324e+03 1.770e+03 2.221e+03 5.122e+03, threshold=3.539e+03, percent-clipped=8.0
+2023-03-13 11:58:32,081 INFO [train.py:968] (0/2) Epoch 26, batch 20600, giga_loss[loss=0.2558, simple_loss=0.3333, pruned_loss=0.08912, over 28540.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3471, pruned_loss=0.09628, over 5699935.31 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3394, pruned_loss=0.08718, over 5742368.28 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3477, pruned_loss=0.09744, over 5685202.86 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:59:05,005 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2974, 1.5702, 1.3741, 1.3680], device='cuda:0'), covar=tensor([0.2109, 0.2119, 0.2427, 0.2116], device='cuda:0'), in_proj_covar=tensor([0.0496, 0.0752, 0.0721, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 11:59:07,966 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:59:09,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:59:14,308 INFO [train.py:968] (0/2) Epoch 26, batch 20650, giga_loss[loss=0.2905, simple_loss=0.3693, pruned_loss=0.1058, over 28871.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3491, pruned_loss=0.09794, over 5698204.87 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3392, pruned_loss=0.08709, over 5741564.54 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3499, pruned_loss=0.09915, over 5686672.72 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:59:33,501 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:59:43,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:59:46,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.465e+03 1.757e+03 2.521e+03 5.702e+03, threshold=3.515e+03, percent-clipped=7.0
+2023-03-13 11:59:48,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:59:59,920 INFO [train.py:968] (0/2) Epoch 26, batch 20700, giga_loss[loss=0.2941, simple_loss=0.3771, pruned_loss=0.1055, over 28816.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3511, pruned_loss=0.09958, over 5704125.26 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3397, pruned_loss=0.08728, over 5742432.73 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3514, pruned_loss=0.1005, over 5694005.32 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:00:45,389 INFO [train.py:968] (0/2) Epoch 26, batch 20750, giga_loss[loss=0.2857, simple_loss=0.3682, pruned_loss=0.1016, over 29067.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3532, pruned_loss=0.1016, over 5702850.10 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3401, pruned_loss=0.08753, over 5730442.30 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3532, pruned_loss=0.1022, over 5704368.12 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:00:58,003 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159824.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:01:15,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.378e+03 1.757e+03 2.426e+03 4.184e+03, threshold=3.514e+03, percent-clipped=3.0
+2023-03-13 12:01:23,949 INFO [train.py:968] (0/2) Epoch 26, batch 20800, giga_loss[loss=0.2607, simple_loss=0.3407, pruned_loss=0.09038, over 28897.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3537, pruned_loss=0.1021, over 5702813.52 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3406, pruned_loss=0.08742, over 5735844.39 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3539, pruned_loss=0.1033, over 5697777.25 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:01:25,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:01:43,731 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:01:45,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8731, 1.1381, 2.8282, 2.7946], device='cuda:0'), covar=tensor([0.1624, 0.2594, 0.0589, 0.1242], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0663, 0.0974, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 12:01:45,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:01:47,628 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:01:49,621 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:02:00,709 INFO [train.py:968] (0/2) Epoch 26, batch 20850, giga_loss[loss=0.2802, simple_loss=0.3626, pruned_loss=0.09896, over 28709.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3534, pruned_loss=0.1014, over 5711267.21 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3404, pruned_loss=0.08726, over 5741811.59 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3541, pruned_loss=0.103, over 5700886.66 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:02:08,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:02:12,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:02:42,272 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:02:43,421 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.369e+02 1.278e+03 1.720e+03 2.496e+03 6.351e+03, threshold=3.441e+03, percent-clipped=11.0
+2023-03-13 12:02:51,366 INFO [train.py:968] (0/2) Epoch 26, batch 20900, giga_loss[loss=0.2757, simple_loss=0.3509, pruned_loss=0.1002, over 28695.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3533, pruned_loss=0.1003, over 5706201.42 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.341, pruned_loss=0.0876, over 5736387.24 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3536, pruned_loss=0.1016, over 5701595.34 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:03:00,957 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159967.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:03:02,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159970.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:03:25,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159999.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:03:26,165 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1160000.pt
+2023-03-13 12:03:28,541 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1160002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:03:31,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1160005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:03:32,524 INFO [train.py:968] (0/2) Epoch 26, batch 20950, giga_loss[loss=0.2766, simple_loss=0.3578, pruned_loss=0.09768, over 29045.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3539, pruned_loss=0.09989, over 5716183.37 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3412, pruned_loss=0.08776, over 5740492.14 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3543, pruned_loss=0.1011, over 5708175.61 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:03:54,693 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1160034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:04:01,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5361, 5.3741, 5.0552, 2.7485], device='cuda:0'), covar=tensor([0.0379, 0.0556, 0.0604, 0.1689], device='cuda:0'), in_proj_covar=tensor([0.1253, 0.1160, 0.0977, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 12:04:04,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.744e+02 1.195e+03 1.423e+03 1.963e+03 6.898e+03, threshold=2.846e+03, percent-clipped=6.0
+2023-03-13 12:04:12,377 INFO [train.py:968] (0/2) Epoch 26, batch 21000, giga_loss[loss=0.2874, simple_loss=0.3561, pruned_loss=0.1094, over 28690.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3526, pruned_loss=0.09881, over 5723544.91 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3416, pruned_loss=0.0879, over 5743882.01 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3527, pruned_loss=0.09983, over 5713734.96 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:04:12,382 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 12:04:20,545 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.206, simple_loss=0.3137, pruned_loss=0.04914, over 944034.00 frames. 
+2023-03-13 12:04:20,546 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 12:04:56,802 INFO [train.py:968] (0/2) Epoch 26, batch 21050, giga_loss[loss=0.2484, simple_loss=0.3244, pruned_loss=0.08617, over 28634.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.351, pruned_loss=0.09895, over 5708088.50 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3417, pruned_loss=0.08804, over 5735628.25 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3512, pruned_loss=0.09982, over 5707132.79 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:05:30,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.238e+02 1.163e+03 1.479e+03 2.045e+03 7.553e+03, threshold=2.958e+03, percent-clipped=7.0
+2023-03-13 12:05:39,522 INFO [train.py:968] (0/2) Epoch 26, batch 21100, giga_loss[loss=0.2469, simple_loss=0.3272, pruned_loss=0.08327, over 29035.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3492, pruned_loss=0.09811, over 5708707.21 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3422, pruned_loss=0.08836, over 5737619.50 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.349, pruned_loss=0.09864, over 5705764.53 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:06:15,956 INFO [train.py:968] (0/2) Epoch 26, batch 21150, giga_loss[loss=0.2337, simple_loss=0.3166, pruned_loss=0.07541, over 28284.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3485, pruned_loss=0.09793, over 5711728.44 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3426, pruned_loss=0.08854, over 5735652.14 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3482, pruned_loss=0.09858, over 5710531.47 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:06:49,174 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.301e+02 1.212e+03 1.445e+03 1.907e+03 5.294e+03, threshold=2.890e+03, percent-clipped=9.0
+2023-03-13 12:06:56,549 INFO [train.py:968] (0/2) Epoch 26, batch 21200, giga_loss[loss=0.2887, simple_loss=0.364, pruned_loss=0.1067, over 29044.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3488, pruned_loss=0.09858, over 5710289.83 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3425, pruned_loss=0.08856, over 5737269.14 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3488, pruned_loss=0.09922, over 5707466.72 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:07:07,914 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1160266.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:07:28,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1160293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:07:29,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7857, 2.6560, 1.6347, 1.1124], device='cuda:0'), covar=tensor([0.8047, 0.3195, 0.4033, 0.6808], device='cuda:0'), in_proj_covar=tensor([0.1803, 0.1683, 0.1629, 0.1465], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 12:07:40,207 INFO [train.py:968] (0/2) Epoch 26, batch 21250, giga_loss[loss=0.2749, simple_loss=0.3516, pruned_loss=0.09913, over 29057.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3489, pruned_loss=0.09824, over 5712942.54 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3427, pruned_loss=0.08871, over 5738115.79 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3487, pruned_loss=0.09866, over 5709865.55 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:07:51,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1160320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:08:07,842 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1160341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:08:12,369 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.833e+02 1.157e+03 1.316e+03 1.600e+03 4.526e+03, threshold=2.632e+03, percent-clipped=4.0
+2023-03-13 12:08:19,134 INFO [train.py:968] (0/2) Epoch 26, batch 21300, giga_loss[loss=0.2461, simple_loss=0.3293, pruned_loss=0.08144, over 28609.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3488, pruned_loss=0.09746, over 5708693.96 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3431, pruned_loss=0.0891, over 5730550.73 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3485, pruned_loss=0.09777, over 5712537.44 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:09:00,157 INFO [train.py:968] (0/2) Epoch 26, batch 21350, giga_loss[loss=0.2623, simple_loss=0.344, pruned_loss=0.09028, over 28698.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3473, pruned_loss=0.09703, over 5703791.63 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.08915, over 5733945.64 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3473, pruned_loss=0.09738, over 5703076.70 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:09:34,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.401e+02 1.102e+03 1.392e+03 1.679e+03 6.788e+03, threshold=2.785e+03, percent-clipped=9.0
+2023-03-13 12:09:42,361 INFO [train.py:968] (0/2) Epoch 26, batch 21400, giga_loss[loss=0.2306, simple_loss=0.3094, pruned_loss=0.07589, over 28606.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3458, pruned_loss=0.09686, over 5690122.98 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08925, over 5726473.63 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3458, pruned_loss=0.0971, over 5696207.20 frames. ], batch size: 78, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:09:48,466 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1160463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:09:50,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1160466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:10:12,080 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1160495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:10:21,655 INFO [train.py:968] (0/2) Epoch 26, batch 21450, giga_loss[loss=0.2219, simple_loss=0.3018, pruned_loss=0.07099, over 28569.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3429, pruned_loss=0.09546, over 5697248.03 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08928, over 5729419.29 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3429, pruned_loss=0.09572, over 5698958.97 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:10:37,877 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-13 12:10:51,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4276, 1.7560, 1.5941, 1.6228], device='cuda:0'), covar=tensor([0.0823, 0.0327, 0.0324, 0.0878], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 12:10:52,295 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.544e+02 1.156e+03 1.468e+03 2.045e+03 3.795e+03, threshold=2.937e+03, percent-clipped=8.0
+2023-03-13 12:10:59,797 INFO [train.py:968] (0/2) Epoch 26, batch 21500, giga_loss[loss=0.2572, simple_loss=0.3275, pruned_loss=0.09346, over 28760.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3395, pruned_loss=0.09373, over 5699495.88 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08922, over 5734138.87 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3395, pruned_loss=0.09412, over 5695726.67 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:11:36,484 INFO [train.py:968] (0/2) Epoch 26, batch 21550, giga_loss[loss=0.2985, simple_loss=0.3714, pruned_loss=0.1127, over 29111.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3401, pruned_loss=0.09425, over 5699638.48 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3433, pruned_loss=0.08952, over 5737291.20 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3396, pruned_loss=0.09448, over 5692481.02 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:12:05,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1160641.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:12:09,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.624e+02 1.292e+03 1.549e+03 2.014e+03 6.098e+03, threshold=3.097e+03, percent-clipped=9.0
+2023-03-13 12:12:18,396 INFO [train.py:968] (0/2) Epoch 26, batch 21600, giga_loss[loss=0.2294, simple_loss=0.3129, pruned_loss=0.07299, over 28909.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3382, pruned_loss=0.09347, over 5703229.19 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3433, pruned_loss=0.08953, over 5740819.26 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3379, pruned_loss=0.09373, over 5693721.56 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:12:27,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1160668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:12:55,352 INFO [train.py:968] (0/2) Epoch 26, batch 21650, giga_loss[loss=0.2433, simple_loss=0.3227, pruned_loss=0.08197, over 28632.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3366, pruned_loss=0.09301, over 5712914.08 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3435, pruned_loss=0.09, over 5748798.61 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3358, pruned_loss=0.09294, over 5696143.22 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:13:00,118 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 12:13:03,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1160716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:13:26,971 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-13 12:13:28,226 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.884e+02 1.194e+03 1.441e+03 2.185e+03 5.783e+03, threshold=2.882e+03, percent-clipped=8.0
+2023-03-13 12:13:36,280 INFO [train.py:968] (0/2) Epoch 26, batch 21700, giga_loss[loss=0.2385, simple_loss=0.313, pruned_loss=0.08197, over 28934.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3341, pruned_loss=0.09176, over 5715075.25 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3435, pruned_loss=0.08998, over 5750393.76 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3333, pruned_loss=0.09175, over 5700075.05 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:13:49,716 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-13 12:13:58,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1160784.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:14:00,959 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1160787.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:14:18,554 INFO [train.py:968] (0/2) Epoch 26, batch 21750, giga_loss[loss=0.3072, simple_loss=0.3654, pruned_loss=0.1245, over 26567.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3321, pruned_loss=0.09124, over 5716895.35 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3439, pruned_loss=0.09025, over 5751064.90 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3312, pruned_loss=0.09102, over 5704272.17 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:14:23,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1160811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:14:25,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1160814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:14:26,224 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1160816.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:14:45,657 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1160843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:14:51,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.555e+02 1.162e+03 1.395e+03 1.838e+03 4.423e+03, threshold=2.790e+03, percent-clipped=4.0
+2023-03-13 12:14:59,381 INFO [train.py:968] (0/2) Epoch 26, batch 21800, giga_loss[loss=0.2874, simple_loss=0.367, pruned_loss=0.1039, over 27917.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3315, pruned_loss=0.09089, over 5718685.30 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3436, pruned_loss=0.09015, over 5752897.41 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3308, pruned_loss=0.09081, over 5706461.40 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:15:01,119 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1160859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:15:03,585 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1160862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:15:24,538 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1160891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:15:37,476 INFO [train.py:968] (0/2) Epoch 26, batch 21850, giga_loss[loss=0.2616, simple_loss=0.3427, pruned_loss=0.09024, over 29026.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3349, pruned_loss=0.09222, over 5696651.89 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3442, pruned_loss=0.09073, over 5735577.44 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3334, pruned_loss=0.09166, over 5702187.02 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:16:13,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.174e+02 1.227e+03 1.567e+03 2.161e+03 5.800e+03, threshold=3.135e+03, percent-clipped=16.0
+2023-03-13 12:16:21,539 INFO [train.py:968] (0/2) Epoch 26, batch 21900, giga_loss[loss=0.2693, simple_loss=0.3496, pruned_loss=0.09448, over 27931.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.338, pruned_loss=0.09352, over 5695053.10 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3438, pruned_loss=0.09063, over 5739071.51 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3371, pruned_loss=0.0932, over 5695634.68 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:16:57,599 INFO [train.py:968] (0/2) Epoch 26, batch 21950, giga_loss[loss=0.2697, simple_loss=0.342, pruned_loss=0.09872, over 28889.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3402, pruned_loss=0.09389, over 5700783.14 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3441, pruned_loss=0.09129, over 5743576.02 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3388, pruned_loss=0.09321, over 5694772.12 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:17:07,562 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2850, 1.2918, 3.6901, 3.2978], device='cuda:0'), covar=tensor([0.1641, 0.2895, 0.0419, 0.1157], device='cuda:0'), in_proj_covar=tensor([0.0784, 0.0664, 0.0976, 0.0946], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 12:17:35,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.964e+02 1.144e+03 1.473e+03 2.090e+03 5.972e+03, threshold=2.946e+03, percent-clipped=9.0
+2023-03-13 12:17:42,457 INFO [train.py:968] (0/2) Epoch 26, batch 22000, giga_loss[loss=0.2269, simple_loss=0.3112, pruned_loss=0.07129, over 28738.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3422, pruned_loss=0.09471, over 5702299.12 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3441, pruned_loss=0.09135, over 5743399.43 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3411, pruned_loss=0.09415, over 5697334.92 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:18:24,606 INFO [train.py:968] (0/2) Epoch 26, batch 22050, giga_loss[loss=0.2455, simple_loss=0.3351, pruned_loss=0.07796, over 28925.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3415, pruned_loss=0.09391, over 5702238.80 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3449, pruned_loss=0.09215, over 5744952.28 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3398, pruned_loss=0.0928, over 5695591.78 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:18:29,675 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6449, 1.7448, 1.8495, 1.4179], device='cuda:0'), covar=tensor([0.1963, 0.2576, 0.1638, 0.1792], device='cuda:0'), in_proj_covar=tensor([0.0925, 0.0709, 0.0972, 0.0873], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 12:19:01,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.466e+02 1.302e+03 1.690e+03 2.216e+03 6.701e+03, threshold=3.380e+03, percent-clipped=9.0
+2023-03-13 12:19:07,821 INFO [train.py:968] (0/2) Epoch 26, batch 22100, giga_loss[loss=0.2386, simple_loss=0.3241, pruned_loss=0.07649, over 28859.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3422, pruned_loss=0.09468, over 5699598.78 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3454, pruned_loss=0.09247, over 5745136.08 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3403, pruned_loss=0.09354, over 5693072.61 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:19:46,988 INFO [train.py:968] (0/2) Epoch 26, batch 22150, giga_loss[loss=0.267, simple_loss=0.3456, pruned_loss=0.09417, over 28695.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3418, pruned_loss=0.09462, over 5704296.47 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3453, pruned_loss=0.09261, over 5746191.42 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3403, pruned_loss=0.09367, over 5697185.26 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:20:14,921 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:20:23,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.814e+02 1.286e+03 1.570e+03 2.040e+03 4.804e+03, threshold=3.140e+03, percent-clipped=5.0
+2023-03-13 12:20:23,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3604, 1.4600, 1.3462, 1.5595], device='cuda:0'), covar=tensor([0.0744, 0.0372, 0.0350, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 12:20:30,064 INFO [train.py:968] (0/2) Epoch 26, batch 22200, giga_loss[loss=0.2729, simple_loss=0.3475, pruned_loss=0.09918, over 29011.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3429, pruned_loss=0.09534, over 5701322.31 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3452, pruned_loss=0.09253, over 5747039.46 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3418, pruned_loss=0.09468, over 5694774.83 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:21:11,304 INFO [train.py:968] (0/2) Epoch 26, batch 22250, giga_loss[loss=0.279, simple_loss=0.3506, pruned_loss=0.1037, over 28694.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3458, pruned_loss=0.09663, over 5702366.58 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3454, pruned_loss=0.09274, over 5740209.72 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3448, pruned_loss=0.09596, over 5703406.37 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:21:46,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.225e+02 1.373e+03 1.537e+03 2.095e+03 5.697e+03, threshold=3.074e+03, percent-clipped=6.0
+2023-03-13 12:21:52,287 INFO [train.py:968] (0/2) Epoch 26, batch 22300, giga_loss[loss=0.2799, simple_loss=0.3525, pruned_loss=0.1037, over 28865.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3486, pruned_loss=0.09827, over 5708297.81 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3461, pruned_loss=0.0934, over 5744679.68 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3471, pruned_loss=0.09719, over 5704124.80 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:22:31,790 INFO [train.py:968] (0/2) Epoch 26, batch 22350, giga_loss[loss=0.256, simple_loss=0.3411, pruned_loss=0.08543, over 29034.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3504, pruned_loss=0.09926, over 5715130.52 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3465, pruned_loss=0.09371, over 5745940.21 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.349, pruned_loss=0.09823, over 5710043.55 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:22:34,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3149, 1.9486, 1.6194, 1.5934], device='cuda:0'), covar=tensor([0.0780, 0.0279, 0.0321, 0.0893], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 12:22:56,097 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-13 12:23:02,042 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-13 12:23:04,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.372e+03 1.795e+03 2.240e+03 7.336e+03, threshold=3.590e+03, percent-clipped=6.0
+2023-03-13 12:23:07,453 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:23:13,012 INFO [train.py:968] (0/2) Epoch 26, batch 22400, giga_loss[loss=0.2749, simple_loss=0.3476, pruned_loss=0.1011, over 28519.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3497, pruned_loss=0.09868, over 5716775.10 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3468, pruned_loss=0.09404, over 5749313.63 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3484, pruned_loss=0.09767, over 5709129.62 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:23:52,509 INFO [train.py:968] (0/2) Epoch 26, batch 22450, giga_loss[loss=0.2501, simple_loss=0.3224, pruned_loss=0.08885, over 28609.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3508, pruned_loss=0.09953, over 5706071.29 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3474, pruned_loss=0.09444, over 5736380.42 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3492, pruned_loss=0.09852, over 5710027.71 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:24:14,473 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-13 12:24:27,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.634e+02 1.330e+03 1.701e+03 2.441e+03 5.792e+03, threshold=3.402e+03, percent-clipped=12.0
+2023-03-13 12:24:29,105 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:24:34,051 INFO [train.py:968] (0/2) Epoch 26, batch 22500, libri_loss[loss=0.2423, simple_loss=0.3249, pruned_loss=0.07987, over 29583.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3491, pruned_loss=0.0988, over 5715332.43 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3476, pruned_loss=0.09499, over 5744611.67 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3478, pruned_loss=0.0977, over 5709783.94 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:25:04,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6770, 1.7855, 1.8386, 1.4412], device='cuda:0'), covar=tensor([0.1972, 0.2632, 0.1651, 0.1865], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0709, 0.0973, 0.0873], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 12:25:14,126 INFO [train.py:968] (0/2) Epoch 26, batch 22550, giga_loss[loss=0.276, simple_loss=0.346, pruned_loss=0.103, over 28818.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.347, pruned_loss=0.09793, over 5719310.01 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.348, pruned_loss=0.09552, over 5743933.61 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3456, pruned_loss=0.09665, over 5714473.36 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:25:21,278 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1161614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:25:28,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6317, 1.8552, 1.7877, 1.5841], device='cuda:0'), covar=tensor([0.2518, 0.2011, 0.1640, 0.2127], device='cuda:0'), in_proj_covar=tensor([0.2043, 0.1992, 0.1918, 0.2049], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 12:25:49,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.820e+02 1.223e+03 1.484e+03 1.807e+03 7.752e+03, threshold=2.969e+03, percent-clipped=5.0
+2023-03-13 12:25:55,995 INFO [train.py:968] (0/2) Epoch 26, batch 22600, giga_loss[loss=0.3389, simple_loss=0.3909, pruned_loss=0.1435, over 26740.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3432, pruned_loss=0.09604, over 5713320.41 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3483, pruned_loss=0.09594, over 5741957.31 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3417, pruned_loss=0.09467, over 5710820.15 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:26:36,941 INFO [train.py:968] (0/2) Epoch 26, batch 22650, giga_loss[loss=0.2514, simple_loss=0.3436, pruned_loss=0.07954, over 27963.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3425, pruned_loss=0.09541, over 5716811.14 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3487, pruned_loss=0.09648, over 5745134.95 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3408, pruned_loss=0.09378, over 5711069.69 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:27:09,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.401e+02 1.168e+03 1.606e+03 2.292e+03 7.730e+03, threshold=3.212e+03, percent-clipped=10.0
+2023-03-13 12:27:14,427 INFO [train.py:968] (0/2) Epoch 26, batch 22700, giga_loss[loss=0.2816, simple_loss=0.3635, pruned_loss=0.0999, over 27625.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3438, pruned_loss=0.09518, over 5720394.84 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3489, pruned_loss=0.0971, over 5749912.34 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3421, pruned_loss=0.09322, over 5709837.32 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:27:15,660 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1161757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:27:18,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1161760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:27:41,035 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-13 12:27:43,081 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1161789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:27:58,323 INFO [train.py:968] (0/2) Epoch 26, batch 22750, giga_loss[loss=0.2832, simple_loss=0.3561, pruned_loss=0.1052, over 28885.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3449, pruned_loss=0.09577, over 5722678.13 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3493, pruned_loss=0.09762, over 5749558.29 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.343, pruned_loss=0.09373, over 5714237.64 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:28:05,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2995, 2.8738, 1.4878, 1.3776], device='cuda:0'), covar=tensor([0.0967, 0.0370, 0.0934, 0.1346], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0561, 0.0401, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 12:28:13,754 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1161825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:28:17,111 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6676, 1.8916, 1.5273, 1.7469], device='cuda:0'), covar=tensor([0.2912, 0.3011, 0.3427, 0.2887], device='cuda:0'), in_proj_covar=tensor([0.1571, 0.1132, 0.1387, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 12:28:18,247 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161832.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:28:22,599 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:28:33,614 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.134e+02 1.304e+03 1.569e+03 2.208e+03 6.872e+03, threshold=3.138e+03, percent-clipped=7.0
+2023-03-13 12:28:39,917 INFO [train.py:968] (0/2) Epoch 26, batch 22800, giga_loss[loss=0.2858, simple_loss=0.3446, pruned_loss=0.1135, over 28686.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3435, pruned_loss=0.09546, over 5728790.18 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3492, pruned_loss=0.09763, over 5751150.84 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.342, pruned_loss=0.09383, over 5720611.76 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:28:41,673 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-13 12:29:04,462 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:29:19,419 INFO [train.py:968] (0/2) Epoch 26, batch 22850, giga_loss[loss=0.2427, simple_loss=0.3123, pruned_loss=0.08657, over 28411.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3431, pruned_loss=0.09673, over 5727634.83 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3492, pruned_loss=0.09774, over 5754378.11 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3418, pruned_loss=0.09531, over 5717959.18 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:29:36,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1161926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:29:55,280 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.914e+02 1.257e+03 1.458e+03 2.024e+03 5.925e+03, threshold=2.915e+03, percent-clipped=4.0
+2023-03-13 12:30:03,251 INFO [train.py:968] (0/2) Epoch 26, batch 22900, giga_loss[loss=0.2933, simple_loss=0.3565, pruned_loss=0.115, over 28735.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3407, pruned_loss=0.09664, over 5725048.48 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3489, pruned_loss=0.0977, over 5757507.19 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.34, pruned_loss=0.09554, over 5713963.86 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:30:10,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1161968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:30:11,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1866, 1.4981, 1.3723, 1.1203], device='cuda:0'), covar=tensor([0.3415, 0.2756, 0.2122, 0.2871], device='cuda:0'), in_proj_covar=tensor([0.2046, 0.1990, 0.1916, 0.2050], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 12:30:12,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1161971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:30:37,150 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1162000.pt
+2023-03-13 12:30:37,667 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:30:41,878 INFO [train.py:968] (0/2) Epoch 26, batch 22950, giga_loss[loss=0.2897, simple_loss=0.3648, pruned_loss=0.1073, over 28552.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3399, pruned_loss=0.09716, over 5721913.47 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.349, pruned_loss=0.09791, over 5750544.91 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.339, pruned_loss=0.09609, over 5718173.09 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:31:16,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.110e+02 1.245e+03 1.567e+03 2.181e+03 6.393e+03, threshold=3.135e+03, percent-clipped=10.0
+2023-03-13 12:31:22,113 INFO [train.py:968] (0/2) Epoch 26, batch 23000, giga_loss[loss=0.2709, simple_loss=0.3528, pruned_loss=0.09445, over 28748.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.339, pruned_loss=0.09682, over 5719054.41 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3498, pruned_loss=0.09862, over 5754301.20 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3373, pruned_loss=0.09531, over 5712042.41 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:31:32,350 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:31:35,846 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1162072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:31:58,338 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:32:02,789 INFO [train.py:968] (0/2) Epoch 26, batch 23050, giga_loss[loss=0.2568, simple_loss=0.3123, pruned_loss=0.1006, over 28699.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3342, pruned_loss=0.09433, over 5727139.34 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3499, pruned_loss=0.09878, over 5757357.99 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3325, pruned_loss=0.09293, over 5718230.12 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:32:19,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.1895, 5.0594, 4.7462, 2.3758], device='cuda:0'), covar=tensor([0.0432, 0.0528, 0.0604, 0.1782], device='cuda:0'), in_proj_covar=tensor([0.1268, 0.1168, 0.0986, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 12:32:24,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 12:32:38,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.845e+02 1.298e+03 1.736e+03 2.264e+03 1.371e+04, threshold=3.473e+03, percent-clipped=10.0
+2023-03-13 12:32:42,422 INFO [train.py:968] (0/2) Epoch 26, batch 23100, libri_loss[loss=0.3054, simple_loss=0.3718, pruned_loss=0.1195, over 29397.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3306, pruned_loss=0.09255, over 5725332.88 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3499, pruned_loss=0.09903, over 5759723.61 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3285, pruned_loss=0.09097, over 5714680.89 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:32:46,927 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1162162.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:32:57,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9084, 1.2396, 1.3772, 1.0533], device='cuda:0'), covar=tensor([0.2190, 0.1468, 0.2544, 0.1858], device='cuda:0'), in_proj_covar=tensor([0.0498, 0.0755, 0.0723, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 12:33:21,446 INFO [train.py:968] (0/2) Epoch 26, batch 23150, libri_loss[loss=0.2799, simple_loss=0.3437, pruned_loss=0.108, over 29553.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3289, pruned_loss=0.09141, over 5733459.00 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3502, pruned_loss=0.09936, over 5763536.31 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3265, pruned_loss=0.08963, over 5720650.79 frames. ], batch size: 76, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:33:21,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1162207.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:33:25,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1162213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:33:29,675 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1162219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:33:56,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.975e+02 1.215e+03 1.560e+03 2.258e+03 5.227e+03, threshold=3.120e+03, percent-clipped=7.0
+2023-03-13 12:34:01,311 INFO [train.py:968] (0/2) Epoch 26, batch 23200, giga_loss[loss=0.2476, simple_loss=0.3272, pruned_loss=0.08399, over 29059.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.332, pruned_loss=0.09244, over 5727340.98 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3502, pruned_loss=0.09943, over 5767785.38 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3294, pruned_loss=0.09073, over 5711960.17 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:34:05,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1162263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:34:16,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.9717, 5.8185, 5.4869, 3.2288], device='cuda:0'), covar=tensor([0.0408, 0.0593, 0.0638, 0.1369], device='cuda:0'), in_proj_covar=tensor([0.1268, 0.1169, 0.0987, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 12:34:43,596 INFO [train.py:968] (0/2) Epoch 26, batch 23250, libri_loss[loss=0.2772, simple_loss=0.3541, pruned_loss=0.1001, over 27323.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3351, pruned_loss=0.09397, over 5717854.75 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3498, pruned_loss=0.0994, over 5764399.46 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3329, pruned_loss=0.09244, over 5707519.09 frames. ], batch size: 115, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:35:07,049 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-13 12:35:18,166 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-13 12:35:20,073 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162350.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:35:20,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.944e+02 1.306e+03 1.561e+03 1.852e+03 6.305e+03, threshold=3.123e+03, percent-clipped=2.0
+2023-03-13 12:35:23,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1162353.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:35:25,524 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:35:25,901 INFO [train.py:968] (0/2) Epoch 26, batch 23300, giga_loss[loss=0.2571, simple_loss=0.3257, pruned_loss=0.09426, over 23815.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3391, pruned_loss=0.09529, over 5717389.12 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3505, pruned_loss=0.09991, over 5766049.69 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3365, pruned_loss=0.09352, over 5706954.04 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:35:27,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1162359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:35:46,909 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162382.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:35:52,060 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:36:07,944 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:36:08,309 INFO [train.py:968] (0/2) Epoch 26, batch 23350, giga_loss[loss=0.2606, simple_loss=0.3364, pruned_loss=0.09238, over 28728.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3422, pruned_loss=0.09666, over 5722540.21 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3505, pruned_loss=0.09998, over 5768526.38 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.34, pruned_loss=0.09514, over 5711257.77 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:36:10,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1162409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:36:35,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:36:45,817 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.732e+02 1.313e+03 1.510e+03 2.095e+03 8.108e+03, threshold=3.021e+03, percent-clipped=14.0
+2023-03-13 12:36:50,228 INFO [train.py:968] (0/2) Epoch 26, batch 23400, giga_loss[loss=0.2663, simple_loss=0.3356, pruned_loss=0.09855, over 28897.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3437, pruned_loss=0.09714, over 5720084.99 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3507, pruned_loss=0.1002, over 5761684.37 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3415, pruned_loss=0.09557, over 5715551.79 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:36:58,980 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 12:37:34,961 INFO [train.py:968] (0/2) Epoch 26, batch 23450, giga_loss[loss=0.3088, simple_loss=0.3729, pruned_loss=0.1223, over 28927.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.348, pruned_loss=0.1008, over 5699237.86 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3508, pruned_loss=0.1005, over 5746677.34 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3459, pruned_loss=0.09925, over 5708218.74 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:38:04,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1162537.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:38:18,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.588e+02 1.398e+03 1.877e+03 2.922e+03 5.740e+03, threshold=3.754e+03, percent-clipped=23.0
+2023-03-13 12:38:24,310 INFO [train.py:968] (0/2) Epoch 26, batch 23500, giga_loss[loss=0.3331, simple_loss=0.3971, pruned_loss=0.1346, over 28252.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3544, pruned_loss=0.1058, over 5693486.16 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3512, pruned_loss=0.1008, over 5749090.17 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3524, pruned_loss=0.1043, over 5697702.76 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:39:01,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1162594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:39:15,327 INFO [train.py:968] (0/2) Epoch 26, batch 23550, giga_loss[loss=0.3932, simple_loss=0.4188, pruned_loss=0.1838, over 23616.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3615, pruned_loss=0.1113, over 5680756.39 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.351, pruned_loss=0.101, over 5751043.82 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3602, pruned_loss=0.1101, over 5681609.67 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:39:30,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4322, 1.5643, 1.1729, 1.1897], device='cuda:0'), covar=tensor([0.0965, 0.0562, 0.1095, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0452, 0.0524, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 12:39:58,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.226e+03 1.800e+03 2.247e+03 3.033e+03 9.017e+03, threshold=4.494e+03, percent-clipped=12.0
+2023-03-13 12:40:03,195 INFO [train.py:968] (0/2) Epoch 26, batch 23600, giga_loss[loss=0.3499, simple_loss=0.4075, pruned_loss=0.1462, over 28894.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3676, pruned_loss=0.1162, over 5683602.75 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3514, pruned_loss=0.1013, over 5753134.06 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3663, pruned_loss=0.1151, over 5681552.02 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:40:10,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0616, 1.2525, 1.2466, 1.0106], device='cuda:0'), covar=tensor([0.1989, 0.2371, 0.1323, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.2043, 0.1992, 0.1916, 0.2042], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 12:40:26,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7747, 5.6152, 5.3382, 2.7593], device='cuda:0'), covar=tensor([0.0466, 0.0601, 0.0697, 0.1608], device='cuda:0'), in_proj_covar=tensor([0.1269, 0.1170, 0.0987, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 12:40:27,217 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162680.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:40:29,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1162683.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:40:55,500 INFO [train.py:968] (0/2) Epoch 26, batch 23650, giga_loss[loss=0.335, simple_loss=0.3933, pruned_loss=0.1384, over 28578.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3738, pruned_loss=0.1212, over 5684980.67 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3513, pruned_loss=0.1013, over 5756293.70 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3733, pruned_loss=0.1207, over 5679391.19 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:41:00,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162712.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:41:22,880 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3906, 1.5525, 1.5226, 1.2779], device='cuda:0'), covar=tensor([0.2691, 0.2572, 0.1853, 0.2498], device='cuda:0'), in_proj_covar=tensor([0.2044, 0.1994, 0.1917, 0.2043], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 12:41:26,293 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:41:30,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1162740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:41:41,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.293e+03 1.917e+03 2.313e+03 3.534e+03 6.593e+03, threshold=4.625e+03, percent-clipped=9.0
+2023-03-13 12:41:44,970 INFO [train.py:968] (0/2) Epoch 26, batch 23700, giga_loss[loss=0.3105, simple_loss=0.3682, pruned_loss=0.1264, over 28827.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.378, pruned_loss=0.1252, over 5676845.06 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3516, pruned_loss=0.1016, over 5757809.43 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3777, pruned_loss=0.1249, over 5669538.88 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:41:55,726 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:42:31,013 INFO [train.py:968] (0/2) Epoch 26, batch 23750, giga_loss[loss=0.2898, simple_loss=0.3663, pruned_loss=0.1067, over 29054.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3791, pruned_loss=0.1269, over 5674330.22 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3516, pruned_loss=0.1016, over 5761888.17 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3797, pruned_loss=0.1274, over 5662593.20 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:42:43,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5525, 1.6429, 1.1761, 1.2233], device='cuda:0'), covar=tensor([0.0818, 0.0452, 0.0952, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0453, 0.0525, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 12:42:47,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7013, 1.8175, 1.8310, 1.5370], device='cuda:0'), covar=tensor([0.3222, 0.2787, 0.2439, 0.2988], device='cuda:0'), in_proj_covar=tensor([0.2045, 0.1994, 0.1917, 0.2043], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 12:43:13,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.844e+03 2.269e+03 3.068e+03 6.435e+03, threshold=4.537e+03, percent-clipped=5.0
+2023-03-13 12:43:16,740 INFO [train.py:968] (0/2) Epoch 26, batch 23800, giga_loss[loss=0.2967, simple_loss=0.369, pruned_loss=0.1122, over 28768.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3821, pruned_loss=0.1308, over 5672572.43 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3514, pruned_loss=0.1018, over 5765271.65 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.384, pruned_loss=0.1321, over 5656632.13 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:44:04,618 INFO [train.py:968] (0/2) Epoch 26, batch 23850, giga_loss[loss=0.3825, simple_loss=0.4205, pruned_loss=0.1722, over 27509.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3855, pruned_loss=0.1345, over 5666130.95 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3514, pruned_loss=0.1019, over 5769094.39 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3879, pruned_loss=0.1363, over 5647217.61 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:44:54,350 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.412e+03 2.064e+03 2.654e+03 3.540e+03 8.291e+03, threshold=5.309e+03, percent-clipped=14.0
+2023-03-13 12:44:58,031 INFO [train.py:968] (0/2) Epoch 26, batch 23900, giga_loss[loss=0.3115, simple_loss=0.3829, pruned_loss=0.1201, over 28978.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3875, pruned_loss=0.1358, over 5667999.07 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3513, pruned_loss=0.102, over 5774295.79 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3915, pruned_loss=0.139, over 5642329.08 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:45:18,904 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-13 12:45:24,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1162979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:45:53,973 INFO [train.py:968] (0/2) Epoch 26, batch 23950, libri_loss[loss=0.2527, simple_loss=0.328, pruned_loss=0.08868, over 29706.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3885, pruned_loss=0.1371, over 5665857.06 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3515, pruned_loss=0.1021, over 5776047.45 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3921, pruned_loss=0.1401, over 5642244.74 frames. ], batch size: 73, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:46:24,038 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:46:36,920 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.231e+03 2.045e+03 2.829e+03 3.725e+03 1.367e+04, threshold=5.659e+03, percent-clipped=11.0
+2023-03-13 12:46:40,581 INFO [train.py:968] (0/2) Epoch 26, batch 24000, giga_loss[loss=0.341, simple_loss=0.394, pruned_loss=0.144, over 28766.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3866, pruned_loss=0.1363, over 5665638.92 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3514, pruned_loss=0.1021, over 5780005.16 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3905, pruned_loss=0.1397, over 5640131.14 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:46:40,585 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 12:46:48,698 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2031, simple_loss=0.3111, pruned_loss=0.04751, over 944034.00 frames. 
+2023-03-13 12:46:48,699 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 12:47:02,122 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:47:33,111 INFO [train.py:968] (0/2) Epoch 26, batch 24050, giga_loss[loss=0.396, simple_loss=0.4236, pruned_loss=0.1841, over 26532.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3854, pruned_loss=0.1355, over 5662203.07 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3511, pruned_loss=0.102, over 5781941.29 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3898, pruned_loss=0.1392, over 5636847.50 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:48:22,163 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.194e+03 1.766e+03 2.384e+03 3.565e+03 8.049e+03, threshold=4.768e+03, percent-clipped=2.0
+2023-03-13 12:48:25,669 INFO [train.py:968] (0/2) Epoch 26, batch 24100, giga_loss[loss=0.38, simple_loss=0.4214, pruned_loss=0.1693, over 27641.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3851, pruned_loss=0.1345, over 5657682.53 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3505, pruned_loss=0.1017, over 5783858.21 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3897, pruned_loss=0.1382, over 5634316.74 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:49:03,197 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:49:18,296 INFO [train.py:968] (0/2) Epoch 26, batch 24150, giga_loss[loss=0.3318, simple_loss=0.3912, pruned_loss=0.1362, over 27908.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3868, pruned_loss=0.1352, over 5652302.14 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3509, pruned_loss=0.102, over 5787258.18 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3909, pruned_loss=0.1387, over 5627843.66 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:49:56,842 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2307, 2.3091, 1.9799, 2.2671], device='cuda:0'), covar=tensor([0.2331, 0.2637, 0.2942, 0.2465], device='cuda:0'), in_proj_covar=tensor([0.1567, 0.1129, 0.1387, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 12:50:06,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.194e+03 1.756e+03 2.173e+03 2.673e+03 7.142e+03, threshold=4.345e+03, percent-clipped=8.0
+2023-03-13 12:50:08,190 INFO [train.py:968] (0/2) Epoch 26, batch 24200, giga_loss[loss=0.2906, simple_loss=0.3643, pruned_loss=0.1085, over 28777.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3844, pruned_loss=0.1328, over 5650546.73 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.351, pruned_loss=0.1021, over 5789503.88 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3882, pruned_loss=0.1361, over 5626654.31 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:50:33,401 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.42 vs. limit=5.0
+2023-03-13 12:50:59,205 INFO [train.py:968] (0/2) Epoch 26, batch 24250, giga_loss[loss=0.3647, simple_loss=0.3984, pruned_loss=0.1655, over 24189.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3813, pruned_loss=0.1289, over 5639375.63 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3515, pruned_loss=0.1025, over 5780612.55 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3843, pruned_loss=0.1316, over 5625866.82 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:51:03,055 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1578, 1.5125, 1.4929, 1.2784], device='cuda:0'), covar=tensor([0.1922, 0.1543, 0.2088, 0.1717], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0763, 0.0729, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 12:51:45,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:51:45,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.212e+03 1.720e+03 2.052e+03 2.908e+03 9.260e+03, threshold=4.103e+03, percent-clipped=7.0
+2023-03-13 12:51:47,375 INFO [train.py:968] (0/2) Epoch 26, batch 24300, giga_loss[loss=0.2843, simple_loss=0.3551, pruned_loss=0.1068, over 28709.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3789, pruned_loss=0.1259, over 5646950.73 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3521, pruned_loss=0.1031, over 5774801.01 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3815, pruned_loss=0.1281, over 5637320.15 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:51:58,500 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:52:17,491 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:52:37,557 INFO [train.py:968] (0/2) Epoch 26, batch 24350, giga_loss[loss=0.2809, simple_loss=0.3589, pruned_loss=0.1014, over 28902.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3764, pruned_loss=0.1234, over 5664515.25 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3522, pruned_loss=0.1033, over 5776759.84 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3787, pruned_loss=0.1253, over 5653641.18 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:52:44,129 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:53:16,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:53:24,059 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.815e+03 2.636e+03 3.790e+03 1.405e+04, threshold=5.271e+03, percent-clipped=22.0
+2023-03-13 12:53:27,384 INFO [train.py:968] (0/2) Epoch 26, batch 24400, giga_loss[loss=0.2958, simple_loss=0.359, pruned_loss=0.1163, over 28923.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3737, pruned_loss=0.1219, over 5663856.58 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3526, pruned_loss=0.1037, over 5779005.62 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3756, pruned_loss=0.1234, over 5651291.60 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:54:00,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 12:54:05,706 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:54:07,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:54:14,797 INFO [train.py:968] (0/2) Epoch 26, batch 24450, giga_loss[loss=0.2718, simple_loss=0.3457, pruned_loss=0.09891, over 28247.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3725, pruned_loss=0.121, over 5677056.02 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3529, pruned_loss=0.1042, over 5781752.18 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3743, pruned_loss=0.1222, over 5661457.55 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:54:40,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:08,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.642e+03 2.166e+03 2.850e+03 5.055e+03, threshold=4.332e+03, percent-clipped=0.0
+2023-03-13 12:55:10,860 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:11,228 INFO [train.py:968] (0/2) Epoch 26, batch 24500, giga_loss[loss=0.2822, simple_loss=0.353, pruned_loss=0.1057, over 28767.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3729, pruned_loss=0.1213, over 5672893.20 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3529, pruned_loss=0.1044, over 5781138.15 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3747, pruned_loss=0.1224, over 5659111.23 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:55:12,976 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:20,559 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:40,478 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:42,473 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:45,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:53,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2199, 1.7902, 1.4067, 0.4368], device='cuda:0'), covar=tensor([0.5274, 0.3372, 0.4448, 0.7475], device='cuda:0'), in_proj_covar=tensor([0.1823, 0.1717, 0.1654, 0.1488], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 12:55:59,388 INFO [train.py:968] (0/2) Epoch 26, batch 24550, giga_loss[loss=0.2686, simple_loss=0.3496, pruned_loss=0.09377, over 28792.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3704, pruned_loss=0.1187, over 5668429.80 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3527, pruned_loss=0.1044, over 5774855.33 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3725, pruned_loss=0.1201, over 5660775.89 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:56:13,980 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:56:20,516 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 12:56:32,509 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:56:47,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.674e+03 2.221e+03 2.876e+03 8.541e+03, threshold=4.442e+03, percent-clipped=5.0
+2023-03-13 12:56:48,789 INFO [train.py:968] (0/2) Epoch 26, batch 24600, giga_loss[loss=0.315, simple_loss=0.3912, pruned_loss=0.1193, over 28725.00 frames. ], tot_loss[loss=0.302, simple_loss=0.371, pruned_loss=0.1166, over 5684897.34 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3525, pruned_loss=0.1043, over 5776011.27 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3734, pruned_loss=0.1181, over 5674911.28 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:57:04,225 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:57:39,750 INFO [train.py:968] (0/2) Epoch 26, batch 24650, giga_loss[loss=0.2938, simple_loss=0.3721, pruned_loss=0.1078, over 28944.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3711, pruned_loss=0.1163, over 5655438.39 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3525, pruned_loss=0.1045, over 5766281.68 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3736, pruned_loss=0.1176, over 5653002.85 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:57:43,965 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:57:47,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:57:58,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3868, 1.5306, 1.5404, 1.3694], device='cuda:0'), covar=tensor([0.2848, 0.2584, 0.2030, 0.2492], device='cuda:0'), in_proj_covar=tensor([0.2040, 0.1989, 0.1917, 0.2041], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 12:58:15,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:58:17,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:58:29,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 2.072e+03 2.588e+03 3.459e+03 7.698e+03, threshold=5.176e+03, percent-clipped=14.0
+2023-03-13 12:58:31,521 INFO [train.py:968] (0/2) Epoch 26, batch 24700, giga_loss[loss=0.2645, simple_loss=0.3457, pruned_loss=0.09162, over 28882.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3726, pruned_loss=0.1179, over 5658875.48 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3527, pruned_loss=0.1047, over 5767079.27 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3744, pruned_loss=0.1188, over 5655779.11 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:58:36,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:59:20,191 INFO [train.py:968] (0/2) Epoch 26, batch 24750, giga_loss[loss=0.2983, simple_loss=0.3677, pruned_loss=0.1144, over 28513.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.372, pruned_loss=0.1184, over 5645637.98 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3525, pruned_loss=0.1045, over 5765455.81 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3739, pruned_loss=0.1195, over 5643152.33 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:59:33,360 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8112, 1.8105, 2.0726, 1.5877], device='cuda:0'), covar=tensor([0.1938, 0.2569, 0.1495, 0.1765], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0709, 0.0966, 0.0868], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 12:59:53,333 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9512, 3.7658, 3.5994, 1.7407], device='cuda:0'), covar=tensor([0.0765, 0.0892, 0.0893, 0.2001], device='cuda:0'), in_proj_covar=tensor([0.1283, 0.1186, 0.0999, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 13:00:08,849 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.211e+03 2.012e+03 2.648e+03 3.549e+03 9.576e+03, threshold=5.296e+03, percent-clipped=5.0
+2023-03-13 13:00:10,099 INFO [train.py:968] (0/2) Epoch 26, batch 24800, giga_loss[loss=0.3091, simple_loss=0.3733, pruned_loss=0.1225, over 28244.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3704, pruned_loss=0.1186, over 5655866.48 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3524, pruned_loss=0.1044, over 5766515.49 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3721, pruned_loss=0.1197, over 5651978.78 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 13:00:33,439 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:00:35,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:00:45,055 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 13:00:50,877 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:00:54,886 INFO [train.py:968] (0/2) Epoch 26, batch 24850, libri_loss[loss=0.3258, simple_loss=0.3849, pruned_loss=0.1333, over 26086.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3694, pruned_loss=0.1187, over 5654597.35 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3529, pruned_loss=0.1048, over 5755992.38 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3708, pruned_loss=0.1195, over 5658175.70 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:00:55,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:01:01,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:01:12,895 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4613, 1.5808, 1.2491, 1.0864], device='cuda:0'), covar=tensor([0.0832, 0.0415, 0.0882, 0.1199], device='cuda:0'), in_proj_covar=tensor([0.0404, 0.0451, 0.0521, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 13:01:19,249 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:01:36,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.718e+03 2.173e+03 2.829e+03 7.148e+03, threshold=4.346e+03, percent-clipped=3.0
+2023-03-13 13:01:37,340 INFO [train.py:968] (0/2) Epoch 26, batch 24900, giga_loss[loss=0.2736, simple_loss=0.3618, pruned_loss=0.09272, over 29007.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.369, pruned_loss=0.1178, over 5667725.12 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3535, pruned_loss=0.1054, over 5757446.34 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3701, pruned_loss=0.1184, over 5666519.12 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:02:18,420 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1164000.pt
+2023-03-13 13:02:24,743 INFO [train.py:968] (0/2) Epoch 26, batch 24950, giga_loss[loss=0.3, simple_loss=0.3697, pruned_loss=0.1152, over 28819.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.369, pruned_loss=0.1171, over 5662923.90 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3539, pruned_loss=0.1057, over 5755606.49 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3697, pruned_loss=0.1174, over 5662283.99 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:02:32,523 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1164014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:03:00,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1164046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:03:11,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.722e+03 2.315e+03 3.294e+03 9.618e+03, threshold=4.630e+03, percent-clipped=6.0
+2023-03-13 13:03:11,850 INFO [train.py:968] (0/2) Epoch 26, batch 25000, giga_loss[loss=0.271, simple_loss=0.3443, pruned_loss=0.09889, over 28632.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3689, pruned_loss=0.1168, over 5666017.41 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3538, pruned_loss=0.1056, over 5756601.62 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3699, pruned_loss=0.1173, over 5662939.93 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:03:41,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5478, 1.9062, 1.9032, 1.4221], device='cuda:0'), covar=tensor([0.3646, 0.2986, 0.2969, 0.3422], device='cuda:0'), in_proj_covar=tensor([0.2039, 0.1987, 0.1917, 0.2040], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 13:03:59,601 INFO [train.py:968] (0/2) Epoch 26, batch 25050, giga_loss[loss=0.2932, simple_loss=0.3613, pruned_loss=0.1126, over 29151.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3669, pruned_loss=0.1156, over 5676520.12 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3536, pruned_loss=0.1056, over 5757956.66 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3679, pruned_loss=0.1161, over 5672235.96 frames. ], batch size: 113, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:04:45,443 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2336, 1.4392, 1.4186, 1.1962], device='cuda:0'), covar=tensor([0.3090, 0.2878, 0.1828, 0.2670], device='cuda:0'), in_proj_covar=tensor([0.2040, 0.1990, 0.1919, 0.2041], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 13:04:47,673 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5748, 1.8678, 1.4806, 1.6861], device='cuda:0'), covar=tensor([0.0762, 0.0299, 0.0336, 0.0852], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 13:04:50,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.667e+03 2.264e+03 3.057e+03 7.539e+03, threshold=4.529e+03, percent-clipped=9.0
+2023-03-13 13:04:50,917 INFO [train.py:968] (0/2) Epoch 26, batch 25100, giga_loss[loss=0.2874, simple_loss=0.3602, pruned_loss=0.1073, over 28762.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.366, pruned_loss=0.1157, over 5679974.00 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3536, pruned_loss=0.1057, over 5760169.36 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3672, pruned_loss=0.1164, over 5672325.97 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:04:51,413 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1164157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:04:53,646 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1164160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:05:20,916 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1164189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:05:20,985 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1164189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:05:23,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1164192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:05:36,294 INFO [train.py:968] (0/2) Epoch 26, batch 25150, giga_loss[loss=0.3186, simple_loss=0.3844, pruned_loss=0.1264, over 28961.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3656, pruned_loss=0.116, over 5681472.03 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3542, pruned_loss=0.1062, over 5753905.31 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3664, pruned_loss=0.1163, over 5678142.01 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:05:51,948 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1164221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:06:27,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.136e+03 1.801e+03 2.427e+03 3.463e+03 8.351e+03, threshold=4.854e+03, percent-clipped=11.0
+2023-03-13 13:06:27,279 INFO [train.py:968] (0/2) Epoch 26, batch 25200, giga_loss[loss=0.2894, simple_loss=0.3628, pruned_loss=0.108, over 28857.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3648, pruned_loss=0.1159, over 5692807.96 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3544, pruned_loss=0.1064, over 5756347.96 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3654, pruned_loss=0.1162, over 5687046.99 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:07:03,381 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-13 13:07:10,731 INFO [train.py:968] (0/2) Epoch 26, batch 25250, giga_loss[loss=0.2501, simple_loss=0.3274, pruned_loss=0.08641, over 28615.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3618, pruned_loss=0.1142, over 5678173.75 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3544, pruned_loss=0.1065, over 5744057.78 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3626, pruned_loss=0.1147, over 5682429.76 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:07:15,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3853, 3.7046, 1.4723, 1.8074], device='cuda:0'), covar=tensor([0.1053, 0.0410, 0.0892, 0.1248], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0569, 0.0404, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 13:08:02,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.192e+03 1.742e+03 2.247e+03 2.902e+03 8.071e+03, threshold=4.493e+03, percent-clipped=3.0
+2023-03-13 13:08:02,353 INFO [train.py:968] (0/2) Epoch 26, batch 25300, libri_loss[loss=0.2647, simple_loss=0.331, pruned_loss=0.09923, over 29336.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3616, pruned_loss=0.1149, over 5680161.90 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3542, pruned_loss=0.1065, over 5747487.34 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3626, pruned_loss=0.1154, over 5678952.31 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:08:52,116 INFO [train.py:968] (0/2) Epoch 26, batch 25350, libri_loss[loss=0.3292, simple_loss=0.3875, pruned_loss=0.1354, over 29036.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3631, pruned_loss=0.1155, over 5681445.52 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3545, pruned_loss=0.1067, over 5749339.43 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3638, pruned_loss=0.1159, over 5677877.52 frames. ], batch size: 101, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:08:56,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4538, 4.3004, 4.0922, 1.9992], device='cuda:0'), covar=tensor([0.0604, 0.0743, 0.0790, 0.2024], device='cuda:0'), in_proj_covar=tensor([0.1292, 0.1192, 0.1007, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 13:09:35,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.588e+03 2.014e+03 2.753e+03 5.336e+03, threshold=4.028e+03, percent-clipped=5.0
+2023-03-13 13:09:35,544 INFO [train.py:968] (0/2) Epoch 26, batch 25400, giga_loss[loss=0.2803, simple_loss=0.3565, pruned_loss=0.102, over 28987.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3644, pruned_loss=0.1157, over 5689303.35 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3542, pruned_loss=0.1067, over 5751707.25 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3653, pruned_loss=0.1162, over 5682974.63 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:10:21,308 INFO [train.py:968] (0/2) Epoch 26, batch 25450, giga_loss[loss=0.257, simple_loss=0.3448, pruned_loss=0.0846, over 28976.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3636, pruned_loss=0.1143, over 5691809.67 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.354, pruned_loss=0.1066, over 5751800.59 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3647, pruned_loss=0.115, over 5685417.88 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:11:08,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.117e+02 1.793e+03 2.476e+03 3.588e+03 1.264e+04, threshold=4.953e+03, percent-clipped=14.0
+2023-03-13 13:11:08,366 INFO [train.py:968] (0/2) Epoch 26, batch 25500, giga_loss[loss=0.2847, simple_loss=0.3518, pruned_loss=0.1088, over 29008.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3649, pruned_loss=0.1153, over 5676382.46 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3544, pruned_loss=0.1068, over 5738328.17 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3657, pruned_loss=0.1158, over 5682106.11 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:11:10,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2336, 1.6026, 1.2241, 0.7425], device='cuda:0'), covar=tensor([0.4630, 0.2703, 0.2845, 0.5679], device='cuda:0'), in_proj_covar=tensor([0.1826, 0.1723, 0.1656, 0.1486], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 13:11:56,409 INFO [train.py:968] (0/2) Epoch 26, batch 25550, giga_loss[loss=0.4102, simple_loss=0.43, pruned_loss=0.1952, over 23866.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3667, pruned_loss=0.1168, over 5666561.41 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3546, pruned_loss=0.107, over 5728406.15 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3674, pruned_loss=0.1172, over 5678005.62 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:11:59,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8747, 5.7285, 5.4680, 2.6604], device='cuda:0'), covar=tensor([0.0493, 0.0645, 0.0724, 0.1658], device='cuda:0'), in_proj_covar=tensor([0.1292, 0.1191, 0.1005, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 13:12:22,237 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2231, 0.7346, 0.8108, 1.3618], device='cuda:0'), covar=tensor([0.0753, 0.0411, 0.0385, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 13:12:44,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.253e+03 1.927e+03 2.385e+03 3.410e+03 6.823e+03, threshold=4.769e+03, percent-clipped=7.0
+2023-03-13 13:12:44,318 INFO [train.py:968] (0/2) Epoch 26, batch 25600, giga_loss[loss=0.3201, simple_loss=0.3707, pruned_loss=0.1348, over 27962.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3691, pruned_loss=0.1196, over 5670933.58 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3551, pruned_loss=0.1072, over 5730401.64 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3695, pruned_loss=0.12, over 5676989.30 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 13:13:32,427 INFO [train.py:968] (0/2) Epoch 26, batch 25650, giga_loss[loss=0.2643, simple_loss=0.3388, pruned_loss=0.09485, over 28909.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3693, pruned_loss=0.1215, over 5671473.88 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.355, pruned_loss=0.1075, over 5735063.34 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.37, pruned_loss=0.1218, over 5670963.28 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 13:14:17,106 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5455, 1.3938, 4.0848, 3.2602], device='cuda:0'), covar=tensor([0.1593, 0.2739, 0.0477, 0.0972], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0672, 0.0993, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 13:14:23,220 INFO [train.py:968] (0/2) Epoch 26, batch 25700, giga_loss[loss=0.2811, simple_loss=0.3562, pruned_loss=0.103, over 28867.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3712, pruned_loss=0.1233, over 5681916.94 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3551, pruned_loss=0.1075, over 5738531.85 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3719, pruned_loss=0.1237, over 5677513.96 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:14:23,825 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.318e+03 1.833e+03 2.548e+03 3.552e+03 5.927e+03, threshold=5.095e+03, percent-clipped=6.0
+2023-03-13 13:14:59,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6994, 1.8094, 1.2859, 1.4059], device='cuda:0'), covar=tensor([0.0943, 0.0604, 0.1081, 0.1163], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0452, 0.0523, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 13:15:04,834 INFO [train.py:968] (0/2) Epoch 26, batch 25750, giga_loss[loss=0.2569, simple_loss=0.3273, pruned_loss=0.09318, over 28561.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3683, pruned_loss=0.1216, over 5678136.49 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3546, pruned_loss=0.1076, over 5742850.90 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3698, pruned_loss=0.1225, over 5668396.08 frames. ], batch size: 78, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:15:51,610 INFO [train.py:968] (0/2) Epoch 26, batch 25800, giga_loss[loss=0.3001, simple_loss=0.3727, pruned_loss=0.1138, over 28751.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3674, pruned_loss=0.1198, over 5682087.17 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3547, pruned_loss=0.1077, over 5745766.54 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3688, pruned_loss=0.1208, over 5670586.55 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:15:52,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.624e+03 2.249e+03 3.166e+03 8.933e+03, threshold=4.497e+03, percent-clipped=6.0
+2023-03-13 13:16:33,037 INFO [train.py:968] (0/2) Epoch 26, batch 25850, giga_loss[loss=0.3306, simple_loss=0.3729, pruned_loss=0.1442, over 26627.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.368, pruned_loss=0.1194, over 5680593.37 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3552, pruned_loss=0.1081, over 5750171.93 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3691, pruned_loss=0.12, over 5666071.25 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:17:08,982 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-13 13:17:19,090 INFO [train.py:968] (0/2) Epoch 26, batch 25900, giga_loss[loss=0.2812, simple_loss=0.3535, pruned_loss=0.1044, over 29002.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3665, pruned_loss=0.1184, over 5679210.10 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3556, pruned_loss=0.1085, over 5753927.59 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3672, pruned_loss=0.1188, over 5662472.85 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:17:20,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.721e+03 2.076e+03 2.580e+03 9.553e+03, threshold=4.152e+03, percent-clipped=6.0
+2023-03-13 13:18:05,579 INFO [train.py:968] (0/2) Epoch 26, batch 25950, giga_loss[loss=0.2925, simple_loss=0.3563, pruned_loss=0.1144, over 28638.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3646, pruned_loss=0.1174, over 5679342.99 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3564, pruned_loss=0.1091, over 5752701.99 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3647, pruned_loss=0.1173, over 5665650.25 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:18:35,555 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6463, 1.8506, 1.7552, 1.5246], device='cuda:0'), covar=tensor([0.2860, 0.2538, 0.2310, 0.2682], device='cuda:0'), in_proj_covar=tensor([0.2046, 0.1998, 0.1926, 0.2048], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 13:18:52,738 INFO [train.py:968] (0/2) Epoch 26, batch 26000, giga_loss[loss=0.3402, simple_loss=0.3895, pruned_loss=0.1454, over 27894.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3647, pruned_loss=0.1183, over 5661810.41 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3567, pruned_loss=0.1093, over 5753670.97 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3647, pruned_loss=0.1184, over 5647542.05 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:18:56,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.654e+03 2.242e+03 3.184e+03 1.256e+04, threshold=4.484e+03, percent-clipped=10.0
+2023-03-13 13:19:20,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7192, 1.9827, 1.3980, 1.5956], device='cuda:0'), covar=tensor([0.1006, 0.0591, 0.1043, 0.1099], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0452, 0.0522, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 13:19:26,069 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-13 13:19:36,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3129, 3.2104, 1.5385, 1.4535], device='cuda:0'), covar=tensor([0.1038, 0.0388, 0.0919, 0.1397], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0569, 0.0403, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 13:19:39,812 INFO [train.py:968] (0/2) Epoch 26, batch 26050, giga_loss[loss=0.2944, simple_loss=0.3626, pruned_loss=0.1131, over 28630.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3664, pruned_loss=0.1196, over 5667485.63 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3571, pruned_loss=0.1096, over 5758846.08 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3664, pruned_loss=0.1197, over 5648427.78 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:19:41,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.23 vs. limit=5.0
+2023-03-13 13:19:45,735 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6983, 1.6901, 1.8655, 1.4657], device='cuda:0'), covar=tensor([0.1909, 0.2716, 0.1561, 0.1840], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0712, 0.0969, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 13:20:25,714 INFO [train.py:968] (0/2) Epoch 26, batch 26100, giga_loss[loss=0.2647, simple_loss=0.3485, pruned_loss=0.09044, over 28764.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3702, pruned_loss=0.12, over 5667110.40 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.357, pruned_loss=0.1096, over 5750692.36 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3704, pruned_loss=0.1202, over 5658082.72 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:20:27,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.685e+03 2.181e+03 2.543e+03 8.388e+03, threshold=4.363e+03, percent-clipped=6.0
+2023-03-13 13:21:08,624 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 13:21:13,548 INFO [train.py:968] (0/2) Epoch 26, batch 26150, giga_loss[loss=0.4172, simple_loss=0.447, pruned_loss=0.1937, over 26684.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3725, pruned_loss=0.1195, over 5670787.56 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3571, pruned_loss=0.1097, over 5753871.47 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.373, pruned_loss=0.1198, over 5658938.78 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:22:01,478 INFO [train.py:968] (0/2) Epoch 26, batch 26200, giga_loss[loss=0.3652, simple_loss=0.4133, pruned_loss=0.1586, over 27676.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3737, pruned_loss=0.1206, over 5667234.01 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3569, pruned_loss=0.1096, over 5754311.18 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3745, pruned_loss=0.1211, over 5656384.49 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:22:04,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.677e+03 2.244e+03 3.036e+03 5.986e+03, threshold=4.488e+03, percent-clipped=4.0
+2023-03-13 13:22:15,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8826, 1.2274, 1.2998, 1.0677], device='cuda:0'), covar=tensor([0.2031, 0.1396, 0.2391, 0.1655], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0762, 0.0728, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 13:22:48,236 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:22:49,325 INFO [train.py:968] (0/2) Epoch 26, batch 26250, giga_loss[loss=0.3381, simple_loss=0.3988, pruned_loss=0.1387, over 28553.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3751, pruned_loss=0.1224, over 5660093.70 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3569, pruned_loss=0.1097, over 5754354.90 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.376, pruned_loss=0.1229, over 5649514.05 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:23:04,597 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-13 13:23:37,592 INFO [train.py:968] (0/2) Epoch 26, batch 26300, giga_loss[loss=0.344, simple_loss=0.3983, pruned_loss=0.1449, over 28896.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3751, pruned_loss=0.1231, over 5651128.18 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.357, pruned_loss=0.1097, over 5753958.41 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.376, pruned_loss=0.1238, over 5641283.87 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:23:41,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.822e+03 2.409e+03 3.247e+03 7.041e+03, threshold=4.817e+03, percent-clipped=12.0
+2023-03-13 13:24:12,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:24:23,564 INFO [train.py:968] (0/2) Epoch 26, batch 26350, giga_loss[loss=0.3771, simple_loss=0.4088, pruned_loss=0.1727, over 23665.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3734, pruned_loss=0.1227, over 5648626.63 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3576, pruned_loss=0.1103, over 5754425.67 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3743, pruned_loss=0.1231, over 5637091.49 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:24:38,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1801, 1.4784, 1.5275, 1.3251], device='cuda:0'), covar=tensor([0.2259, 0.2002, 0.2662, 0.2126], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0763, 0.0729, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 13:25:08,610 INFO [train.py:968] (0/2) Epoch 26, batch 26400, giga_loss[loss=0.3263, simple_loss=0.382, pruned_loss=0.1354, over 28584.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3713, pruned_loss=0.1218, over 5654397.43 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3577, pruned_loss=0.1104, over 5752978.26 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3722, pruned_loss=0.1223, over 5645116.72 frames. ], batch size: 78, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:25:09,617 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-13 13:25:11,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.184e+03 1.817e+03 2.453e+03 3.886e+03 9.636e+03, threshold=4.905e+03, percent-clipped=15.0
+2023-03-13 13:25:24,031 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4055, 1.5949, 1.6458, 1.3909], device='cuda:0'), covar=tensor([0.2155, 0.2155, 0.2623, 0.2294], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0762, 0.0729, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 13:25:25,500 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-13 13:25:59,148 INFO [train.py:968] (0/2) Epoch 26, batch 26450, giga_loss[loss=0.278, simple_loss=0.3419, pruned_loss=0.1071, over 28577.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3694, pruned_loss=0.1214, over 5655658.97 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3571, pruned_loss=0.11, over 5756042.58 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3708, pruned_loss=0.1223, over 5643887.38 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:26:19,553 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-13 13:26:45,722 INFO [train.py:968] (0/2) Epoch 26, batch 26500, libri_loss[loss=0.2997, simple_loss=0.3703, pruned_loss=0.1146, over 28595.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3691, pruned_loss=0.1216, over 5651008.20 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.357, pruned_loss=0.1099, over 5754335.41 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3707, pruned_loss=0.1227, over 5640549.34 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:26:47,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.732e+03 2.178e+03 3.196e+03 1.175e+04, threshold=4.355e+03, percent-clipped=3.0
+2023-03-13 13:27:24,463 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 13:27:25,505 INFO [train.py:968] (0/2) Epoch 26, batch 26550, giga_loss[loss=0.3272, simple_loss=0.3862, pruned_loss=0.1341, over 28614.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3687, pruned_loss=0.1215, over 5651557.80 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.357, pruned_loss=0.11, over 5749475.04 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3704, pruned_loss=0.1229, over 5643680.95 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:28:08,885 INFO [train.py:968] (0/2) Epoch 26, batch 26600, giga_loss[loss=0.2834, simple_loss=0.3471, pruned_loss=0.1099, over 28665.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3666, pruned_loss=0.1201, over 5668886.37 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3573, pruned_loss=0.1101, over 5752230.56 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3679, pruned_loss=0.1212, over 5658195.70 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:28:11,042 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.150e+03 1.689e+03 2.287e+03 3.247e+03 1.041e+04, threshold=4.575e+03, percent-clipped=8.0
+2023-03-13 13:28:22,274 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4810, 1.5208, 1.3245, 1.5486], device='cuda:0'), covar=tensor([0.0779, 0.0324, 0.0340, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 13:28:34,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1165680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:28:54,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3886, 1.5279, 1.2006, 1.0908], device='cuda:0'), covar=tensor([0.0991, 0.0556, 0.1057, 0.1178], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0453, 0.0523, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 13:29:00,308 INFO [train.py:968] (0/2) Epoch 26, batch 26650, giga_loss[loss=0.2606, simple_loss=0.336, pruned_loss=0.09263, over 28803.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.366, pruned_loss=0.1197, over 5672130.73 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3573, pruned_loss=0.1101, over 5752230.56 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.367, pruned_loss=0.1206, over 5663810.05 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:29:10,941 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:29:46,114 INFO [train.py:968] (0/2) Epoch 26, batch 26700, giga_loss[loss=0.2771, simple_loss=0.3568, pruned_loss=0.09874, over 28620.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3678, pruned_loss=0.1201, over 5660913.35 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3573, pruned_loss=0.1101, over 5744992.84 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3687, pruned_loss=0.1209, over 5659869.14 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:29:48,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.700e+03 2.113e+03 2.921e+03 6.896e+03, threshold=4.227e+03, percent-clipped=9.0
+2023-03-13 13:29:59,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1165774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:30:31,230 INFO [train.py:968] (0/2) Epoch 26, batch 26750, giga_loss[loss=0.331, simple_loss=0.3886, pruned_loss=0.1367, over 27636.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3704, pruned_loss=0.1217, over 5657645.92 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3582, pruned_loss=0.1109, over 5739190.19 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3707, pruned_loss=0.1219, over 5660336.79 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:30:31,419 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:30:48,075 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1165823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:30:49,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5603, 4.4264, 4.2153, 2.1713], device='cuda:0'), covar=tensor([0.0615, 0.0715, 0.0777, 0.1878], device='cuda:0'), in_proj_covar=tensor([0.1296, 0.1194, 0.1010, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 13:30:49,997 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1165826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:31:16,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1165855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:31:17,399 INFO [train.py:968] (0/2) Epoch 26, batch 26800, giga_loss[loss=0.2866, simple_loss=0.3559, pruned_loss=0.1087, over 28884.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3702, pruned_loss=0.1224, over 5649090.56 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3585, pruned_loss=0.1112, over 5733886.10 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3705, pruned_loss=0.1226, over 5652624.61 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:31:20,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.749e+03 2.374e+03 3.918e+03 1.014e+04, threshold=4.747e+03, percent-clipped=20.0
+2023-03-13 13:31:42,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165886.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:32:01,861 INFO [train.py:968] (0/2) Epoch 26, batch 26850, giga_loss[loss=0.2676, simple_loss=0.3616, pruned_loss=0.08677, over 28945.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3708, pruned_loss=0.1196, over 5657016.40 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3584, pruned_loss=0.1112, over 5724297.03 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3713, pruned_loss=0.12, over 5667330.08 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:32:12,964 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1165917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:32:15,322 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1165920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:32:43,007 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1165949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:32:48,427 INFO [train.py:968] (0/2) Epoch 26, batch 26900, libri_loss[loss=0.2847, simple_loss=0.3556, pruned_loss=0.1069, over 29663.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3713, pruned_loss=0.1186, over 5666420.86 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3578, pruned_loss=0.1108, over 5728844.41 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3725, pruned_loss=0.1194, over 5668883.52 frames. ], batch size: 91, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:32:51,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.847e+03 2.763e+03 3.831e+03 1.003e+04, threshold=5.526e+03, percent-clipped=15.0
+2023-03-13 13:32:52,689 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:33:26,529 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1166000.pt
+2023-03-13 13:33:31,091 INFO [train.py:968] (0/2) Epoch 26, batch 26950, giga_loss[loss=0.2994, simple_loss=0.3714, pruned_loss=0.1136, over 28724.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3734, pruned_loss=0.119, over 5665287.37 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.358, pruned_loss=0.1112, over 5728643.20 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3748, pruned_loss=0.1196, over 5664763.96 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:34:19,222 INFO [train.py:968] (0/2) Epoch 26, batch 27000, giga_loss[loss=0.2621, simple_loss=0.3417, pruned_loss=0.09126, over 28985.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3761, pruned_loss=0.1217, over 5669979.64 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3579, pruned_loss=0.1111, over 5730550.00 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3776, pruned_loss=0.1225, over 5666498.80 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:34:19,226 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 13:34:27,998 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2025, simple_loss=0.3096, pruned_loss=0.04771, over 944034.00 frames. 
+2023-03-13 13:34:27,999 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 13:34:30,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.651e+03 2.290e+03 2.792e+03 8.455e+03, threshold=4.581e+03, percent-clipped=3.0
+2023-03-13 13:35:01,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:35:04,860 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1166097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:35:14,198 INFO [train.py:968] (0/2) Epoch 26, batch 27050, libri_loss[loss=0.2825, simple_loss=0.3612, pruned_loss=0.1019, over 29513.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3776, pruned_loss=0.1235, over 5666850.26 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3586, pruned_loss=0.1115, over 5719006.57 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3788, pruned_loss=0.1242, over 5672694.28 frames. ], batch size: 80, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:35:16,626 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1166110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:36:05,179 INFO [train.py:968] (0/2) Epoch 26, batch 27100, giga_loss[loss=0.3299, simple_loss=0.3866, pruned_loss=0.1366, over 27952.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.378, pruned_loss=0.1249, over 5668113.80 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3588, pruned_loss=0.1117, over 5718244.49 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3791, pruned_loss=0.1254, over 5672789.39 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:36:09,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.954e+03 2.351e+03 3.207e+03 1.024e+04, threshold=4.702e+03, percent-clipped=12.0
+2023-03-13 13:36:25,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3223, 3.1415, 1.4662, 1.5025], device='cuda:0'), covar=tensor([0.1018, 0.0384, 0.0914, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0569, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 13:36:30,376 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:36:57,009 INFO [train.py:968] (0/2) Epoch 26, batch 27150, giga_loss[loss=0.2646, simple_loss=0.3422, pruned_loss=0.09352, over 29082.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3764, pruned_loss=0.1239, over 5669694.82 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3586, pruned_loss=0.1117, over 5721962.05 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3778, pruned_loss=0.1246, over 5669137.95 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:37:25,294 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:37:27,354 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:37:41,184 INFO [train.py:968] (0/2) Epoch 26, batch 27200, giga_loss[loss=0.2603, simple_loss=0.3551, pruned_loss=0.08273, over 28915.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3747, pruned_loss=0.1209, over 5669169.07 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3582, pruned_loss=0.1114, over 5723833.40 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3765, pruned_loss=0.1219, over 5666180.70 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 13:37:47,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.700e+03 2.387e+03 3.308e+03 1.164e+04, threshold=4.774e+03, percent-clipped=9.0
+2023-03-13 13:37:47,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166261.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:37:54,230 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:38:08,046 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-13 13:38:30,353 INFO [train.py:968] (0/2) Epoch 26, batch 27250, giga_loss[loss=0.2902, simple_loss=0.368, pruned_loss=0.1062, over 28926.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3761, pruned_loss=0.1213, over 5652783.40 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.358, pruned_loss=0.1114, over 5716735.59 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3779, pruned_loss=0.1222, over 5656414.15 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 13:38:47,744 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:38:50,156 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:38:57,979 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:39:09,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6573, 2.3591, 1.6325, 0.9365], device='cuda:0'), covar=tensor([0.7303, 0.3606, 0.4733, 0.7422], device='cuda:0'), in_proj_covar=tensor([0.1820, 0.1716, 0.1647, 0.1480], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 13:39:19,039 INFO [train.py:968] (0/2) Epoch 26, batch 27300, giga_loss[loss=0.3244, simple_loss=0.3849, pruned_loss=0.1319, over 27674.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3776, pruned_loss=0.1223, over 5650264.11 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3585, pruned_loss=0.1117, over 5710902.97 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.379, pruned_loss=0.123, over 5657589.26 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:39:19,296 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:39:24,851 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.597e+03 1.995e+03 2.574e+03 6.063e+03, threshold=3.990e+03, percent-clipped=5.0
+2023-03-13 13:40:02,883 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166404.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 13:40:05,232 INFO [train.py:968] (0/2) Epoch 26, batch 27350, giga_loss[loss=0.2862, simple_loss=0.3684, pruned_loss=0.102, over 28982.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.377, pruned_loss=0.1229, over 5644439.57 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3583, pruned_loss=0.1117, over 5708665.23 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3791, pruned_loss=0.1238, over 5649780.64 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:40:05,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166407.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:40:31,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1166436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:40:31,508 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166436.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:40:52,181 INFO [train.py:968] (0/2) Epoch 26, batch 27400, giga_loss[loss=0.2959, simple_loss=0.3683, pruned_loss=0.1118, over 28978.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3757, pruned_loss=0.1228, over 5651075.47 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3581, pruned_loss=0.1115, over 5710009.70 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3779, pruned_loss=0.1239, over 5653369.46 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:41:00,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.285e+03 1.879e+03 2.326e+03 3.271e+03 1.083e+04, threshold=4.652e+03, percent-clipped=16.0
+2023-03-13 13:41:08,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:41:15,015 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:41:19,678 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:41:23,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166485.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:41:44,370 INFO [train.py:968] (0/2) Epoch 26, batch 27450, giga_loss[loss=0.3363, simple_loss=0.3947, pruned_loss=0.139, over 27855.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3738, pruned_loss=0.1219, over 5657167.16 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3581, pruned_loss=0.1116, over 5702724.03 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3757, pruned_loss=0.1229, over 5663976.46 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:41:50,583 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:42:34,488 INFO [train.py:968] (0/2) Epoch 26, batch 27500, giga_loss[loss=0.2378, simple_loss=0.3176, pruned_loss=0.079, over 28440.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3726, pruned_loss=0.1218, over 5651033.20 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3589, pruned_loss=0.1121, over 5696179.26 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.374, pruned_loss=0.1225, over 5660645.87 frames. ], batch size: 78, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:42:38,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.711e+03 2.116e+03 3.178e+03 9.958e+03, threshold=4.232e+03, percent-clipped=11.0
+2023-03-13 13:43:21,627 INFO [train.py:968] (0/2) Epoch 26, batch 27550, giga_loss[loss=0.3163, simple_loss=0.3906, pruned_loss=0.121, over 28694.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3715, pruned_loss=0.1216, over 5657524.68 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3589, pruned_loss=0.112, over 5701245.89 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3729, pruned_loss=0.1225, over 5659615.87 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:43:29,752 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:43:32,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:43:42,747 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166628.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:43:44,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166631.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:43:59,975 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:44:07,233 INFO [train.py:968] (0/2) Epoch 26, batch 27600, giga_loss[loss=0.3085, simple_loss=0.3644, pruned_loss=0.1263, over 26512.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3708, pruned_loss=0.1218, over 5660056.35 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3586, pruned_loss=0.1117, over 5704521.44 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3725, pruned_loss=0.123, over 5657527.48 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:44:09,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166660.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 13:44:11,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.863e+03 2.347e+03 3.256e+03 1.623e+04, threshold=4.694e+03, percent-clipped=6.0
+2023-03-13 13:44:40,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1166700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:44:47,105 INFO [train.py:968] (0/2) Epoch 26, batch 27650, giga_loss[loss=0.3239, simple_loss=0.393, pruned_loss=0.1274, over 28293.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3683, pruned_loss=0.119, over 5651335.91 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3586, pruned_loss=0.1116, over 5692310.95 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 5659014.42 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:44:50,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2187, 1.5232, 1.4801, 1.0774], device='cuda:0'), covar=tensor([0.1739, 0.2646, 0.1470, 0.1787], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0714, 0.0973, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:0')
+2023-03-13 13:45:00,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6527, 1.7876, 1.4582, 1.8019], device='cuda:0'), covar=tensor([0.2772, 0.3043, 0.3367, 0.2879], device='cuda:0'), in_proj_covar=tensor([0.1577, 0.1135, 0.1395, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 13:45:32,092 INFO [train.py:968] (0/2) Epoch 26, batch 27700, giga_loss[loss=0.2855, simple_loss=0.3534, pruned_loss=0.1088, over 27627.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3649, pruned_loss=0.1155, over 5658156.03 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3582, pruned_loss=0.1115, over 5693779.51 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3669, pruned_loss=0.1168, over 5662103.29 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:45:39,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.506e+03 2.067e+03 2.777e+03 5.470e+03, threshold=4.134e+03, percent-clipped=4.0
+2023-03-13 13:46:23,778 INFO [train.py:968] (0/2) Epoch 26, batch 27750, giga_loss[loss=0.2677, simple_loss=0.3403, pruned_loss=0.09751, over 29014.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.363, pruned_loss=0.1142, over 5647015.19 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3582, pruned_loss=0.1116, over 5694605.70 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3647, pruned_loss=0.1153, over 5648502.91 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:46:28,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:47:13,026 INFO [train.py:968] (0/2) Epoch 26, batch 27800, giga_loss[loss=0.2708, simple_loss=0.3379, pruned_loss=0.1018, over 28896.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3622, pruned_loss=0.1141, over 5658284.29 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3583, pruned_loss=0.1117, over 5700694.55 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3636, pruned_loss=0.115, over 5652802.17 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:47:19,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.185e+03 1.714e+03 2.390e+03 3.068e+03 1.105e+04, threshold=4.780e+03, percent-clipped=9.0
+2023-03-13 13:47:19,469 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9480, 2.1033, 1.7291, 2.0918], device='cuda:0'), covar=tensor([0.2509, 0.2672, 0.3008, 0.2554], device='cuda:0'), in_proj_covar=tensor([0.1582, 0.1138, 0.1398, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 13:47:34,100 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1166877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:47:44,396 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=5.12 vs. limit=5.0
+2023-03-13 13:48:05,620 INFO [train.py:968] (0/2) Epoch 26, batch 27850, giga_loss[loss=0.2723, simple_loss=0.3464, pruned_loss=0.09909, over 28396.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.359, pruned_loss=0.1131, over 5658084.77 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3577, pruned_loss=0.1113, over 5704514.76 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3607, pruned_loss=0.1142, over 5649556.44 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:48:32,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0332, 5.8763, 5.5728, 3.2008], device='cuda:0'), covar=tensor([0.0469, 0.0581, 0.0705, 0.1428], device='cuda:0'), in_proj_covar=tensor([0.1305, 0.1204, 0.1016, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 13:48:52,103 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:48:53,742 INFO [train.py:968] (0/2) Epoch 26, batch 27900, giga_loss[loss=0.2632, simple_loss=0.3489, pruned_loss=0.08872, over 28534.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3606, pruned_loss=0.1139, over 5658353.34 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3578, pruned_loss=0.1114, over 5705177.53 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3619, pruned_loss=0.1147, over 5649771.91 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:48:54,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:48:59,202 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.774e+03 2.270e+03 2.927e+03 5.711e+03, threshold=4.540e+03, percent-clipped=2.0
+2023-03-13 13:49:21,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3698, 1.5474, 1.1865, 1.1197], device='cuda:0'), covar=tensor([0.1006, 0.0515, 0.1022, 0.1165], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0455, 0.0526, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 13:49:24,859 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:49:45,130 INFO [train.py:968] (0/2) Epoch 26, batch 27950, giga_loss[loss=0.3612, simple_loss=0.4104, pruned_loss=0.156, over 27668.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3631, pruned_loss=0.1152, over 5646946.47 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3576, pruned_loss=0.1112, over 5706199.64 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3644, pruned_loss=0.116, over 5639074.90 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:50:16,175 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:50:25,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7969, 1.1600, 5.0252, 3.6842], device='cuda:0'), covar=tensor([0.1526, 0.2912, 0.0367, 0.0843], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0670, 0.0996, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 13:50:34,702 INFO [train.py:968] (0/2) Epoch 26, batch 28000, giga_loss[loss=0.3165, simple_loss=0.3736, pruned_loss=0.1297, over 28679.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3636, pruned_loss=0.1149, over 5658219.01 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3578, pruned_loss=0.1113, over 5708208.52 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3644, pruned_loss=0.1155, over 5649810.99 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:50:40,701 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.094e+03 1.485e+03 1.839e+03 2.487e+03 6.429e+03, threshold=3.678e+03, percent-clipped=7.0
+2023-03-13 13:50:49,481 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:51:20,404 INFO [train.py:968] (0/2) Epoch 26, batch 28050, giga_loss[loss=0.29, simple_loss=0.3623, pruned_loss=0.1089, over 28887.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3652, pruned_loss=0.1165, over 5643506.95 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3583, pruned_loss=0.1117, over 5702796.55 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3656, pruned_loss=0.1168, over 5640882.38 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:52:01,363 INFO [train.py:968] (0/2) Epoch 26, batch 28100, giga_loss[loss=0.327, simple_loss=0.3859, pruned_loss=0.134, over 28006.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3651, pruned_loss=0.1169, over 5651919.18 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.358, pruned_loss=0.1116, over 5710231.25 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.366, pruned_loss=0.1174, over 5641113.59 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:52:07,845 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.113e+03 1.738e+03 2.143e+03 2.929e+03 5.555e+03, threshold=4.285e+03, percent-clipped=14.0
+2023-03-13 13:52:14,255 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-13 13:52:47,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2055, 4.0296, 3.8402, 1.8749], device='cuda:0'), covar=tensor([0.0690, 0.0811, 0.0868, 0.1988], device='cuda:0'), in_proj_covar=tensor([0.1308, 0.1205, 0.1019, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 13:52:49,628 INFO [train.py:968] (0/2) Epoch 26, batch 28150, giga_loss[loss=0.3173, simple_loss=0.3794, pruned_loss=0.1276, over 28916.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3677, pruned_loss=0.1185, over 5650659.04 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.358, pruned_loss=0.1116, over 5703077.05 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3685, pruned_loss=0.1189, over 5648361.99 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:53:00,817 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:53:04,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:53:32,964 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:53:34,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:53:38,251 INFO [train.py:968] (0/2) Epoch 26, batch 28200, libri_loss[loss=0.314, simple_loss=0.3843, pruned_loss=0.1219, over 29307.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3695, pruned_loss=0.12, over 5647846.43 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.358, pruned_loss=0.1117, over 5704996.19 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3702, pruned_loss=0.1204, over 5643552.74 frames. ], batch size: 94, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:53:47,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.294e+03 2.106e+03 2.686e+03 3.711e+03 9.369e+03, threshold=5.372e+03, percent-clipped=20.0
+2023-03-13 13:54:26,870 INFO [train.py:968] (0/2) Epoch 26, batch 28250, giga_loss[loss=0.3682, simple_loss=0.4139, pruned_loss=0.1612, over 28538.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3703, pruned_loss=0.121, over 5641613.14 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3581, pruned_loss=0.1118, over 5698635.34 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3712, pruned_loss=0.1216, over 5641716.08 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:54:33,219 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:54:33,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:54:56,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:55:19,313 INFO [train.py:968] (0/2) Epoch 26, batch 28300, giga_loss[loss=0.3091, simple_loss=0.3729, pruned_loss=0.1227, over 28986.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.1209, over 5650388.61 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3578, pruned_loss=0.1116, over 5702880.75 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3713, pruned_loss=0.1216, over 5645822.59 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:55:19,862 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-13 13:55:26,469 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.225e+03 1.701e+03 2.031e+03 3.278e+03 1.132e+04, threshold=4.062e+03, percent-clipped=4.0
+2023-03-13 13:55:58,392 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:56:00,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:56:08,522 INFO [train.py:968] (0/2) Epoch 26, batch 28350, giga_loss[loss=0.2924, simple_loss=0.3632, pruned_loss=0.1108, over 27969.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3704, pruned_loss=0.1195, over 5641909.61 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3583, pruned_loss=0.112, over 5692428.22 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.1199, over 5645806.12 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:56:17,042 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:56:27,881 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:56:28,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2238, 1.4375, 1.4013, 1.1513], device='cuda:0'), covar=tensor([0.3370, 0.2934, 0.1991, 0.2751], device='cuda:0'), in_proj_covar=tensor([0.2048, 0.1996, 0.1924, 0.2045], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 13:56:38,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5693, 3.6910, 1.6879, 1.7875], device='cuda:0'), covar=tensor([0.0987, 0.0333, 0.0886, 0.1299], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0569, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 13:56:43,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167444.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:56:54,265 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-13 13:56:55,840 INFO [train.py:968] (0/2) Epoch 26, batch 28400, giga_loss[loss=0.2987, simple_loss=0.3631, pruned_loss=0.1171, over 28899.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3713, pruned_loss=0.1209, over 5629074.62 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3588, pruned_loss=0.1125, over 5688864.00 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3718, pruned_loss=0.121, over 5633357.60 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:57:04,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.918e+03 2.426e+03 2.971e+03 7.080e+03, threshold=4.852e+03, percent-clipped=14.0
+2023-03-13 13:57:43,093 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6590, 1.8358, 1.5338, 1.7217], device='cuda:0'), covar=tensor([0.2561, 0.2706, 0.2826, 0.2523], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1139, 0.1395, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 13:57:43,447 INFO [train.py:968] (0/2) Epoch 26, batch 28450, giga_loss[loss=0.3202, simple_loss=0.3835, pruned_loss=0.1284, over 27989.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3717, pruned_loss=0.1216, over 5632455.30 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5694006.44 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3719, pruned_loss=0.1218, over 5629276.66 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:57:57,723 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:58:44,602 INFO [train.py:968] (0/2) Epoch 26, batch 28500, giga_loss[loss=0.366, simple_loss=0.4029, pruned_loss=0.1646, over 26492.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3705, pruned_loss=0.1216, over 5632754.32 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5698277.97 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3709, pruned_loss=0.1219, over 5625299.04 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:58:45,034 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:58:48,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:58:53,049 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.194e+03 1.785e+03 2.356e+03 3.203e+03 6.859e+03, threshold=4.711e+03, percent-clipped=7.0
+2023-03-13 13:59:18,612 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:59:33,874 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8647, 1.8287, 2.1021, 1.6352], device='cuda:0'), covar=tensor([0.1739, 0.2435, 0.1370, 0.1684], device='cuda:0'), in_proj_covar=tensor([0.0921, 0.0713, 0.0971, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 13:59:34,213 INFO [train.py:968] (0/2) Epoch 26, batch 28550, giga_loss[loss=0.2808, simple_loss=0.3536, pruned_loss=0.104, over 29085.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3687, pruned_loss=0.1208, over 5630371.95 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3594, pruned_loss=0.1126, over 5696938.05 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3693, pruned_loss=0.1213, over 5623842.35 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:59:38,091 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-13 14:00:17,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3470, 1.8923, 1.4408, 0.6520], device='cuda:0'), covar=tensor([0.5859, 0.3032, 0.3861, 0.6498], device='cuda:0'), in_proj_covar=tensor([0.1828, 0.1726, 0.1654, 0.1489], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 14:00:20,799 INFO [train.py:968] (0/2) Epoch 26, batch 28600, giga_loss[loss=0.3209, simple_loss=0.3753, pruned_loss=0.1333, over 27969.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.367, pruned_loss=0.1197, over 5651052.18 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5700267.04 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3674, pruned_loss=0.1202, over 5642182.41 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:00:24,643 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:00:29,873 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.216e+03 1.861e+03 2.347e+03 3.114e+03 6.030e+03, threshold=4.695e+03, percent-clipped=5.0
+2023-03-13 14:00:35,688 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:00:49,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5055, 2.8030, 1.6629, 1.5585], device='cuda:0'), covar=tensor([0.0794, 0.0342, 0.0685, 0.1120], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0568, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 14:00:50,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:00:54,083 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:00:54,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:01:13,069 INFO [train.py:968] (0/2) Epoch 26, batch 28650, giga_loss[loss=0.283, simple_loss=0.343, pruned_loss=0.1116, over 28576.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3667, pruned_loss=0.1199, over 5646233.34 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5702283.38 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.367, pruned_loss=0.1203, over 5637134.27 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:01:20,700 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:01:23,580 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4142, 1.6527, 1.7100, 1.4744], device='cuda:0'), covar=tensor([0.2454, 0.2506, 0.2510, 0.2521], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0764, 0.0730, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 14:01:48,337 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-13 14:01:59,646 INFO [train.py:968] (0/2) Epoch 26, batch 28700, giga_loss[loss=0.2841, simple_loss=0.3453, pruned_loss=0.1114, over 28832.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.367, pruned_loss=0.1199, over 5653231.43 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5697287.09 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3673, pruned_loss=0.1203, over 5649681.28 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:02:07,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.754e+03 2.269e+03 2.878e+03 5.883e+03, threshold=4.538e+03, percent-clipped=3.0
+2023-03-13 14:02:28,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:02:46,838 INFO [train.py:968] (0/2) Epoch 26, batch 28750, giga_loss[loss=0.3051, simple_loss=0.3766, pruned_loss=0.1168, over 29051.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3682, pruned_loss=0.1208, over 5652092.98 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3598, pruned_loss=0.1127, over 5691945.49 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1214, over 5653024.43 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:02:59,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167819.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 14:03:13,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:14,531 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:15,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:17,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:37,879 INFO [train.py:968] (0/2) Epoch 26, batch 28800, giga_loss[loss=0.2743, simple_loss=0.3483, pruned_loss=0.1001, over 28222.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3692, pruned_loss=0.1217, over 5658127.29 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3603, pruned_loss=0.1132, over 5693974.83 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3692, pruned_loss=0.1218, over 5656567.61 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:03:38,134 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:40,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:44,867 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:45,500 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:45,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.977e+02 1.814e+03 2.160e+03 3.214e+03 1.120e+04, threshold=4.321e+03, percent-clipped=14.0
+2023-03-13 14:04:00,134 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:04:02,584 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:04:05,380 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:04:09,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:04:20,271 INFO [train.py:968] (0/2) Epoch 26, batch 28850, giga_loss[loss=0.3083, simple_loss=0.3711, pruned_loss=0.1227, over 28924.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3708, pruned_loss=0.1233, over 5667070.68 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5695534.05 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3712, pruned_loss=0.1238, over 5663615.96 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:04:25,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3924, 1.4603, 1.2575, 1.5262], device='cuda:0'), covar=tensor([0.0731, 0.0345, 0.0349, 0.0850], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 14:04:39,561 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-13 14:04:49,680 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-13 14:05:07,212 INFO [train.py:968] (0/2) Epoch 26, batch 28900, giga_loss[loss=0.2768, simple_loss=0.3487, pruned_loss=0.1025, over 28641.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3694, pruned_loss=0.1224, over 5666017.59 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5694276.42 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3704, pruned_loss=0.1233, over 5663791.15 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:05:07,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5502, 4.3841, 4.1786, 2.0747], device='cuda:0'), covar=tensor([0.0589, 0.0740, 0.0771, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.1310, 0.1207, 0.1021, 0.0752], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 14:05:12,837 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167962.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 14:05:16,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167965.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 14:05:17,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.958e+03 2.751e+03 3.979e+03 1.016e+04, threshold=5.502e+03, percent-clipped=23.0
+2023-03-13 14:05:41,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5427, 1.7173, 1.6864, 1.5166], device='cuda:0'), covar=tensor([0.2161, 0.2306, 0.2540, 0.2363], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0762, 0.0729, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 14:05:43,576 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167994.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 14:05:49,494 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1168000.pt
+2023-03-13 14:05:58,901 INFO [train.py:968] (0/2) Epoch 26, batch 28950, giga_loss[loss=0.3177, simple_loss=0.3778, pruned_loss=0.1289, over 28656.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3704, pruned_loss=0.1225, over 5666700.74 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3598, pruned_loss=0.1127, over 5696475.02 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3711, pruned_loss=0.1232, over 5662787.48 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:06:17,730 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:23,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:27,195 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:29,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:33,424 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:43,095 INFO [train.py:968] (0/2) Epoch 26, batch 29000, giga_loss[loss=0.2441, simple_loss=0.3341, pruned_loss=0.07706, over 28901.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3709, pruned_loss=0.1225, over 5670171.41 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.113, over 5696217.87 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3713, pruned_loss=0.1231, over 5666772.71 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:06:44,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:52,029 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:53,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.730e+03 2.213e+03 3.042e+03 9.635e+03, threshold=4.427e+03, percent-clipped=2.0
+2023-03-13 14:06:55,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3372, 0.8970, 1.0362, 1.4597], device='cuda:0'), covar=tensor([0.0757, 0.0390, 0.0343, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 14:06:56,224 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:07:19,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:07:22,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4747, 2.0402, 1.4405, 1.9478], device='cuda:0'), covar=tensor([0.0741, 0.0272, 0.0336, 0.0822], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 14:07:29,579 INFO [train.py:968] (0/2) Epoch 26, batch 29050, giga_loss[loss=0.2742, simple_loss=0.3369, pruned_loss=0.1058, over 28723.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.372, pruned_loss=0.1238, over 5659170.85 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3603, pruned_loss=0.113, over 5682902.54 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3726, pruned_loss=0.1245, over 5667395.20 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:08:15,509 INFO [train.py:968] (0/2) Epoch 26, batch 29100, libri_loss[loss=0.3009, simple_loss=0.3582, pruned_loss=0.1218, over 29569.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3735, pruned_loss=0.125, over 5660109.79 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3601, pruned_loss=0.1128, over 5687866.45 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3744, pruned_loss=0.1261, over 5661442.57 frames. ], batch size: 78, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:08:17,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:24,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.796e+03 2.372e+03 3.441e+03 1.498e+04, threshold=4.745e+03, percent-clipped=14.0
+2023-03-13 14:08:34,714 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:38,654 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:44,306 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:44,405 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:46,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:57,228 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:59,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:01,252 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-13 14:09:01,441 INFO [train.py:968] (0/2) Epoch 26, batch 29150, giga_loss[loss=0.3005, simple_loss=0.3763, pruned_loss=0.1124, over 28868.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3733, pruned_loss=0.1248, over 5649854.94 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3604, pruned_loss=0.113, over 5680936.78 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3739, pruned_loss=0.1256, over 5656293.87 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:09:04,489 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:12,588 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:15,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5209, 1.7526, 1.2844, 1.2765], device='cuda:0'), covar=tensor([0.1143, 0.0652, 0.1157, 0.1230], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0455, 0.0525, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 14:09:25,758 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:27,836 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:38,817 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:49,272 INFO [train.py:968] (0/2) Epoch 26, batch 29200, giga_loss[loss=0.2446, simple_loss=0.3304, pruned_loss=0.07939, over 28377.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3726, pruned_loss=0.1236, over 5646499.80 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3601, pruned_loss=0.1131, over 5686464.99 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3738, pruned_loss=0.1245, over 5645834.73 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:09:52,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:55,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:10:01,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.664e+03 2.048e+03 2.695e+03 5.908e+03, threshold=4.097e+03, percent-clipped=6.0
+2023-03-13 14:10:35,326 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:10:38,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:10:39,056 INFO [train.py:968] (0/2) Epoch 26, batch 29250, giga_loss[loss=0.2719, simple_loss=0.3447, pruned_loss=0.09954, over 28265.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3729, pruned_loss=0.1232, over 5642891.61 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.113, over 5689176.63 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3742, pruned_loss=0.1243, over 5638781.30 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:10:46,692 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:11:03,515 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:11:22,511 INFO [train.py:968] (0/2) Epoch 26, batch 29300, giga_loss[loss=0.2939, simple_loss=0.3594, pruned_loss=0.1142, over 28907.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3723, pruned_loss=0.1226, over 5651409.58 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3603, pruned_loss=0.1133, over 5689761.02 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3733, pruned_loss=0.1234, over 5647034.40 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:11:32,082 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.614e+02 1.727e+03 2.032e+03 2.754e+03 1.035e+04, threshold=4.065e+03, percent-clipped=9.0
+2023-03-13 14:12:02,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:04,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:04,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:05,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:06,765 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:07,126 INFO [train.py:968] (0/2) Epoch 26, batch 29350, giga_loss[loss=0.3078, simple_loss=0.377, pruned_loss=0.1193, over 28944.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3707, pruned_loss=0.1215, over 5650253.42 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3604, pruned_loss=0.1134, over 5679769.07 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3718, pruned_loss=0.1223, over 5654360.08 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:12:30,617 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:32,013 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:37,549 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:54,711 INFO [train.py:968] (0/2) Epoch 26, batch 29400, libri_loss[loss=0.3081, simple_loss=0.3832, pruned_loss=0.1165, over 29485.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3721, pruned_loss=0.1226, over 5637674.53 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3605, pruned_loss=0.1134, over 5672485.48 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.373, pruned_loss=0.1233, over 5647240.35 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:13:04,208 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.197e+03 1.757e+03 2.416e+03 3.308e+03 1.763e+04, threshold=4.832e+03, percent-clipped=19.0
+2023-03-13 14:13:05,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:13:07,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5293, 1.7858, 1.4166, 1.4795], device='cuda:0'), covar=tensor([0.2740, 0.2799, 0.3155, 0.2468], device='cuda:0'), in_proj_covar=tensor([0.1578, 0.1138, 0.1395, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 14:13:08,733 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:13:13,916 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:13:31,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-13 14:13:39,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3406, 1.2827, 3.7006, 3.1868], device='cuda:0'), covar=tensor([0.1648, 0.2807, 0.0481, 0.1127], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0669, 0.0997, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 14:13:45,294 INFO [train.py:968] (0/2) Epoch 26, batch 29450, giga_loss[loss=0.2949, simple_loss=0.3668, pruned_loss=0.1115, over 29058.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3732, pruned_loss=0.1233, over 5642384.33 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3608, pruned_loss=0.1136, over 5667957.26 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.374, pruned_loss=0.1239, over 5653132.81 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:14:12,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3855, 1.5183, 1.5784, 1.4441], device='cuda:0'), covar=tensor([0.1590, 0.1506, 0.1798, 0.1522], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0761, 0.0728, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 14:14:22,058 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:14:25,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:14:33,085 INFO [train.py:968] (0/2) Epoch 26, batch 29500, giga_loss[loss=0.3342, simple_loss=0.3902, pruned_loss=0.1391, over 28623.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3719, pruned_loss=0.1233, over 5649183.26 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1133, over 5672646.48 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3734, pruned_loss=0.1244, over 5653165.55 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:14:37,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:14:42,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 1.675e+03 2.189e+03 2.866e+03 5.596e+03, threshold=4.379e+03, percent-clipped=4.0
+2023-03-13 14:14:50,363 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:14:54,060 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:14:56,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:18,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:20,143 INFO [train.py:968] (0/2) Epoch 26, batch 29550, giga_loss[loss=0.3501, simple_loss=0.4081, pruned_loss=0.146, over 28894.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3731, pruned_loss=0.1241, over 5663617.17 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3607, pruned_loss=0.1136, over 5677123.30 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.374, pruned_loss=0.1248, over 5662585.18 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:15:24,506 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:25,096 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:27,708 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:31,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:42,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 14:15:49,232 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2469, 1.3732, 3.1896, 2.8976], device='cuda:0'), covar=tensor([0.1450, 0.2503, 0.0506, 0.2026], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0670, 0.0997, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 14:15:52,804 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:16:03,789 INFO [train.py:968] (0/2) Epoch 26, batch 29600, giga_loss[loss=0.287, simple_loss=0.3609, pruned_loss=0.1065, over 28753.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3755, pruned_loss=0.1262, over 5659004.01 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5680019.29 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3762, pruned_loss=0.1269, over 5655229.16 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:16:11,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.687e+03 2.259e+03 3.215e+03 6.287e+03, threshold=4.519e+03, percent-clipped=11.0
+2023-03-13 14:16:20,072 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4510, 3.8583, 1.5567, 1.5401], device='cuda:0'), covar=tensor([0.0974, 0.0367, 0.0934, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0568, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 14:16:31,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3581, 1.6237, 1.4801, 1.5365], device='cuda:0'), covar=tensor([0.0803, 0.0334, 0.0329, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 14:16:37,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:16:48,389 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:16:50,323 INFO [train.py:968] (0/2) Epoch 26, batch 29650, giga_loss[loss=0.3495, simple_loss=0.4143, pruned_loss=0.1423, over 28578.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3757, pruned_loss=0.1264, over 5653129.04 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3609, pruned_loss=0.1137, over 5684131.15 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3768, pruned_loss=0.1273, over 5645699.15 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:16:53,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:16,970 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:26,813 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:28,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:34,004 INFO [train.py:968] (0/2) Epoch 26, batch 29700, giga_loss[loss=0.3169, simple_loss=0.3777, pruned_loss=0.128, over 28012.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3766, pruned_loss=0.1271, over 5646262.34 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5678748.18 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3776, pruned_loss=0.1281, over 5644948.22 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:17:40,118 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:42,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:44,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.942e+03 2.492e+03 3.432e+03 7.659e+03, threshold=4.985e+03, percent-clipped=12.0
+2023-03-13 14:17:59,636 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:18:10,283 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:18:20,352 INFO [train.py:968] (0/2) Epoch 26, batch 29750, giga_loss[loss=0.2653, simple_loss=0.3458, pruned_loss=0.09246, over 28491.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3758, pruned_loss=0.1256, over 5647405.55 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3613, pruned_loss=0.1138, over 5675406.65 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3769, pruned_loss=0.1268, over 5648288.42 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:18:48,190 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:18:50,307 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 14:18:52,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:18:59,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:19:04,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:19:05,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3940, 3.3780, 1.5227, 1.5225], device='cuda:0'), covar=tensor([0.0999, 0.0391, 0.0914, 0.1346], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0568, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 14:19:07,127 INFO [train.py:968] (0/2) Epoch 26, batch 29800, libri_loss[loss=0.2775, simple_loss=0.3316, pruned_loss=0.1117, over 29392.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3761, pruned_loss=0.1256, over 5650587.19 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5677655.66 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3777, pruned_loss=0.1269, over 5648470.51 frames. ], batch size: 67, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:19:17,110 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:19:17,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.739e+03 2.445e+03 3.561e+03 6.259e+03, threshold=4.890e+03, percent-clipped=7.0
+2023-03-13 14:19:54,925 INFO [train.py:968] (0/2) Epoch 26, batch 29850, libri_loss[loss=0.2914, simple_loss=0.3583, pruned_loss=0.1123, over 29560.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3743, pruned_loss=0.1243, over 5658218.54 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3611, pruned_loss=0.1138, over 5682219.23 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3759, pruned_loss=0.1256, over 5651853.67 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:20:42,401 INFO [train.py:968] (0/2) Epoch 26, batch 29900, giga_loss[loss=0.3539, simple_loss=0.4061, pruned_loss=0.1508, over 28779.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3731, pruned_loss=0.1234, over 5667827.56 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.361, pruned_loss=0.1137, over 5686774.89 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3747, pruned_loss=0.1247, over 5658322.83 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:20:52,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 2.046e+03 2.596e+03 3.799e+03 1.031e+04, threshold=5.192e+03, percent-clipped=12.0
+2023-03-13 14:21:14,153 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:16,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:19,579 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:22,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:22,537 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:28,080 INFO [train.py:968] (0/2) Epoch 26, batch 29950, giga_loss[loss=0.2732, simple_loss=0.3405, pruned_loss=0.1029, over 28557.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3701, pruned_loss=0.1218, over 5667217.05 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3615, pruned_loss=0.1141, over 5684756.95 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1226, over 5660836.46 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:21:43,100 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1169022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:52,170 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1169028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:22:03,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6381, 1.7932, 1.6571, 1.5547], device='cuda:0'), covar=tensor([0.2061, 0.2646, 0.2519, 0.2571], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0762, 0.0728, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 14:22:14,328 INFO [train.py:968] (0/2) Epoch 26, batch 30000, giga_loss[loss=0.2451, simple_loss=0.3087, pruned_loss=0.09074, over 28690.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3657, pruned_loss=0.1197, over 5662976.38 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3617, pruned_loss=0.1142, over 5691285.97 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3666, pruned_loss=0.1205, over 5651138.40 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:22:14,332 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 14:22:22,589 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2037, simple_loss=0.3124, pruned_loss=0.0475, over 944034.00 frames. 
+2023-03-13 14:22:22,589 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 14:22:33,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.909e+03 2.391e+03 3.184e+03 9.684e+03, threshold=4.782e+03, percent-clipped=5.0
+2023-03-13 14:22:33,646 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 14:22:53,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5712, 1.7944, 1.7782, 1.3389], device='cuda:0'), covar=tensor([0.1679, 0.2725, 0.1479, 0.1816], device='cuda:0'), in_proj_covar=tensor([0.0925, 0.0716, 0.0975, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 14:23:01,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4607, 3.3764, 1.6143, 1.5766], device='cuda:0'), covar=tensor([0.1013, 0.0425, 0.0913, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0571, 0.0405, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 14:23:05,940 INFO [train.py:968] (0/2) Epoch 26, batch 30050, giga_loss[loss=0.2865, simple_loss=0.3551, pruned_loss=0.109, over 28918.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3644, pruned_loss=0.1195, over 5671113.10 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.114, over 5694789.03 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3655, pruned_loss=0.1205, over 5657508.64 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:23:38,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6746, 1.9443, 1.5592, 1.7662], device='cuda:0'), covar=tensor([0.2685, 0.2837, 0.3291, 0.2477], device='cuda:0'), in_proj_covar=tensor([0.1581, 0.1140, 0.1398, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 14:23:50,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-13 14:23:54,749 INFO [train.py:968] (0/2) Epoch 26, batch 30100, giga_loss[loss=0.3398, simple_loss=0.3907, pruned_loss=0.1445, over 27890.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3626, pruned_loss=0.1188, over 5653316.70 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5695307.85 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3637, pruned_loss=0.1199, over 5641513.95 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:24:08,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.903e+03 2.303e+03 3.950e+03 8.851e+03, threshold=4.607e+03, percent-clipped=12.0
+2023-03-13 14:24:20,347 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8285, 2.1400, 1.6187, 2.1547], device='cuda:0'), covar=tensor([0.2487, 0.2652, 0.3078, 0.2403], device='cuda:0'), in_proj_covar=tensor([0.1581, 0.1140, 0.1399, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 14:24:38,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4609, 4.2958, 4.0591, 1.9996], device='cuda:0'), covar=tensor([0.0640, 0.0766, 0.0821, 0.2074], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.1212, 0.1024, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 14:24:42,842 INFO [train.py:968] (0/2) Epoch 26, batch 30150, giga_loss[loss=0.3247, simple_loss=0.3871, pruned_loss=0.1312, over 28608.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3625, pruned_loss=0.1173, over 5643438.44 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1138, over 5682172.29 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3633, pruned_loss=0.1183, over 5644444.28 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:25:36,040 INFO [train.py:968] (0/2) Epoch 26, batch 30200, giga_loss[loss=0.3063, simple_loss=0.3868, pruned_loss=0.1129, over 28589.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3594, pruned_loss=0.1132, over 5632779.10 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3609, pruned_loss=0.1137, over 5681695.16 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3604, pruned_loss=0.1141, over 5632948.88 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:25:49,511 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.757e+03 2.305e+03 2.945e+03 7.755e+03, threshold=4.611e+03, percent-clipped=5.0
+2023-03-13 14:26:01,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2494, 3.4795, 1.4193, 1.4827], device='cuda:0'), covar=tensor([0.1299, 0.0472, 0.1119, 0.1645], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0570, 0.0404, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 14:26:23,486 INFO [train.py:968] (0/2) Epoch 26, batch 30250, giga_loss[loss=0.2913, simple_loss=0.3625, pruned_loss=0.1101, over 28547.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3575, pruned_loss=0.1104, over 5643701.02 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3598, pruned_loss=0.113, over 5682426.75 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3592, pruned_loss=0.1117, over 5642024.53 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:26:27,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1169312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:27:13,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5445, 1.5604, 1.7593, 1.3763], device='cuda:0'), covar=tensor([0.1732, 0.2580, 0.1511, 0.1905], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0712, 0.0971, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 14:27:14,591 INFO [train.py:968] (0/2) Epoch 26, batch 30300, giga_loss[loss=0.2923, simple_loss=0.3583, pruned_loss=0.1132, over 28562.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3551, pruned_loss=0.1075, over 5631635.94 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3599, pruned_loss=0.1133, over 5670527.68 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3563, pruned_loss=0.1082, over 5639729.61 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:27:24,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.290e+02 1.472e+03 2.071e+03 3.209e+03 5.814e+03, threshold=4.142e+03, percent-clipped=2.0
+2023-03-13 14:27:28,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1169374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:27:56,783 INFO [train.py:968] (0/2) Epoch 26, batch 30350, giga_loss[loss=0.292, simple_loss=0.3635, pruned_loss=0.1103, over 27973.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3519, pruned_loss=0.1046, over 5635687.71 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3592, pruned_loss=0.1132, over 5666124.30 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3532, pruned_loss=0.105, over 5645359.59 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:28:48,394 INFO [train.py:968] (0/2) Epoch 26, batch 30400, giga_loss[loss=0.2402, simple_loss=0.3371, pruned_loss=0.07164, over 28901.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3496, pruned_loss=0.1007, over 5646212.40 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3589, pruned_loss=0.1131, over 5659630.04 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3509, pruned_loss=0.1009, over 5659021.83 frames. ], batch size: 285, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:28:58,514 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 14:29:01,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.304e+02 1.397e+03 1.802e+03 2.406e+03 5.153e+03, threshold=3.604e+03, percent-clipped=4.0
+2023-03-13 14:29:03,898 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1169471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:29:38,886 INFO [train.py:968] (0/2) Epoch 26, batch 30450, libri_loss[loss=0.2665, simple_loss=0.3264, pruned_loss=0.1033, over 29568.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3491, pruned_loss=0.0997, over 5661071.14 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3576, pruned_loss=0.1125, over 5668908.64 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3509, pruned_loss=0.0999, over 5662366.46 frames. ], batch size: 75, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:29:46,942 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1169517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:29:49,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1169520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:30:20,829 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1169549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:30:28,358 INFO [train.py:968] (0/2) Epoch 26, batch 30500, giga_loss[loss=0.2874, simple_loss=0.3621, pruned_loss=0.1063, over 28701.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3495, pruned_loss=0.09981, over 5666050.30 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3577, pruned_loss=0.1127, over 5675826.17 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3506, pruned_loss=0.09943, over 5660644.43 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:30:41,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.696e+02 1.390e+03 1.946e+03 2.870e+03 5.641e+03, threshold=3.892e+03, percent-clipped=12.0
+2023-03-13 14:31:21,119 INFO [train.py:968] (0/2) Epoch 26, batch 30550, giga_loss[loss=0.2543, simple_loss=0.3338, pruned_loss=0.08746, over 28314.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3468, pruned_loss=0.09768, over 5665384.64 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3576, pruned_loss=0.1127, over 5677019.04 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3477, pruned_loss=0.0973, over 5660100.32 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:32:13,711 INFO [train.py:968] (0/2) Epoch 26, batch 30600, giga_loss[loss=0.2669, simple_loss=0.3432, pruned_loss=0.09535, over 27640.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3443, pruned_loss=0.09659, over 5657011.20 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3569, pruned_loss=0.1125, over 5673644.09 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3454, pruned_loss=0.09608, over 5655240.67 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:32:23,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.584e+02 1.520e+03 1.896e+03 2.513e+03 6.946e+03, threshold=3.791e+03, percent-clipped=4.0
+2023-03-13 14:32:39,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1169687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:32:56,825 INFO [train.py:968] (0/2) Epoch 26, batch 30650, giga_loss[loss=0.2525, simple_loss=0.338, pruned_loss=0.08347, over 28224.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3446, pruned_loss=0.0963, over 5664237.53 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3567, pruned_loss=0.1123, over 5676840.31 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3452, pruned_loss=0.09553, over 5659382.66 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:33:35,679 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-13 14:33:44,273 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1169752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:33:51,195 INFO [train.py:968] (0/2) Epoch 26, batch 30700, libri_loss[loss=0.2721, simple_loss=0.332, pruned_loss=0.1061, over 29530.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3439, pruned_loss=0.09606, over 5657931.39 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3566, pruned_loss=0.1123, over 5678761.01 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3444, pruned_loss=0.09529, over 5651984.13 frames. ], batch size: 81, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:34:00,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.707e+02 1.548e+03 2.265e+03 3.266e+03 1.088e+04, threshold=4.531e+03, percent-clipped=15.0
+2023-03-13 14:34:38,615 INFO [train.py:968] (0/2) Epoch 26, batch 30750, giga_loss[loss=0.2153, simple_loss=0.3035, pruned_loss=0.06357, over 28625.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3414, pruned_loss=0.09394, over 5664819.44 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3556, pruned_loss=0.1118, over 5684629.75 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3423, pruned_loss=0.09322, over 5654313.40 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:34:51,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3872, 1.8624, 1.7103, 1.6202], device='cuda:0'), covar=tensor([0.2252, 0.2224, 0.1951, 0.2184], device='cuda:0'), in_proj_covar=tensor([0.0495, 0.0751, 0.0720, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 14:35:02,149 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1169830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:35:04,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1169833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:35:16,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1169846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:35:27,346 INFO [train.py:968] (0/2) Epoch 26, batch 30800, giga_loss[loss=0.2647, simple_loss=0.3336, pruned_loss=0.09789, over 27523.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3389, pruned_loss=0.09244, over 5672859.46 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3553, pruned_loss=0.1117, over 5686471.92 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3395, pruned_loss=0.09151, over 5662104.98 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:35:31,926 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1169862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:35:40,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.438e+02 1.413e+03 1.790e+03 2.574e+03 1.071e+04, threshold=3.580e+03, percent-clipped=1.0
+2023-03-13 14:36:11,646 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1169900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:36:13,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5216, 1.7756, 1.6067, 1.3830], device='cuda:0'), covar=tensor([0.2222, 0.1901, 0.1725, 0.2091], device='cuda:0'), in_proj_covar=tensor([0.2019, 0.1969, 0.1892, 0.2018], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 14:36:16,832 INFO [train.py:968] (0/2) Epoch 26, batch 30850, giga_loss[loss=0.2259, simple_loss=0.2901, pruned_loss=0.08083, over 23872.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3367, pruned_loss=0.09199, over 5670550.53 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3557, pruned_loss=0.1122, over 5689150.78 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3362, pruned_loss=0.09026, over 5659384.39 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:36:24,356 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1169915.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 14:37:05,177 INFO [train.py:968] (0/2) Epoch 26, batch 30900, libri_loss[loss=0.2403, simple_loss=0.3063, pruned_loss=0.08711, over 28179.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3351, pruned_loss=0.0916, over 5667186.66 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3547, pruned_loss=0.1116, over 5692891.37 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3353, pruned_loss=0.09023, over 5654251.13 frames. ], batch size: 62, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:37:17,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.177e+02 1.583e+03 2.126e+03 3.041e+03 5.464e+03, threshold=4.252e+03, percent-clipped=15.0
+2023-03-13 14:37:40,214 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1169989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:37:44,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2422, 1.6105, 1.6123, 1.4226], device='cuda:0'), covar=tensor([0.1964, 0.1695, 0.1800, 0.1813], device='cuda:0'), in_proj_covar=tensor([0.0494, 0.0750, 0.0718, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 14:37:44,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1169992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:37:47,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5048, 1.7774, 1.5785, 1.5476], device='cuda:0'), covar=tensor([0.0784, 0.0314, 0.0337, 0.0923], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 14:37:51,374 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1170000.pt
+2023-03-13 14:37:58,115 INFO [train.py:968] (0/2) Epoch 26, batch 30950, giga_loss[loss=0.2535, simple_loss=0.3335, pruned_loss=0.08674, over 28563.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.335, pruned_loss=0.09138, over 5662629.79 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3543, pruned_loss=0.1114, over 5698007.79 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3351, pruned_loss=0.09003, over 5646957.94 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:38:14,494 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:38:17,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 14:38:39,988 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 14:38:54,539 INFO [train.py:968] (0/2) Epoch 26, batch 31000, giga_loss[loss=0.2354, simple_loss=0.325, pruned_loss=0.07288, over 28925.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.337, pruned_loss=0.09163, over 5656618.76 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3541, pruned_loss=0.1114, over 5702152.48 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3369, pruned_loss=0.09018, over 5639770.72 frames. ], batch size: 285, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:38:57,773 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1170058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:39:11,442 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.592e+03 2.080e+03 2.779e+03 1.096e+04, threshold=4.161e+03, percent-clipped=8.0
+2023-03-13 14:39:46,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2130, 1.5110, 1.3967, 1.1672], device='cuda:0'), covar=tensor([0.2593, 0.2332, 0.1678, 0.2276], device='cuda:0'), in_proj_covar=tensor([0.2018, 0.1971, 0.1890, 0.2016], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 14:39:53,600 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1170106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:39:53,969 INFO [train.py:968] (0/2) Epoch 26, batch 31050, giga_loss[loss=0.2799, simple_loss=0.3501, pruned_loss=0.1048, over 28034.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3373, pruned_loss=0.09103, over 5646569.21 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3543, pruned_loss=0.1116, over 5702624.26 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3368, pruned_loss=0.08944, over 5632103.51 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:40:21,973 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1170127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:41:00,169 INFO [train.py:968] (0/2) Epoch 26, batch 31100, giga_loss[loss=0.2504, simple_loss=0.3311, pruned_loss=0.08482, over 28956.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3384, pruned_loss=0.09196, over 5645393.05 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3545, pruned_loss=0.1118, over 5702352.55 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3376, pruned_loss=0.09025, over 5633167.37 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:41:15,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.854e+02 1.533e+03 2.124e+03 3.075e+03 1.106e+04, threshold=4.248e+03, percent-clipped=12.0
+2023-03-13 14:41:59,896 INFO [train.py:968] (0/2) Epoch 26, batch 31150, giga_loss[loss=0.2366, simple_loss=0.3267, pruned_loss=0.07322, over 28731.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3362, pruned_loss=0.09027, over 5653587.47 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3544, pruned_loss=0.1118, over 5704102.03 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3353, pruned_loss=0.08842, over 5640990.27 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:43:03,506 INFO [train.py:968] (0/2) Epoch 26, batch 31200, giga_loss[loss=0.1974, simple_loss=0.2946, pruned_loss=0.05016, over 28874.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3332, pruned_loss=0.08718, over 5643465.76 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3534, pruned_loss=0.1114, over 5708113.18 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.333, pruned_loss=0.0857, over 5628857.90 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:43:16,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1170270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:43:16,992 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.704e+02 1.384e+03 1.647e+03 2.291e+03 5.338e+03, threshold=3.294e+03, percent-clipped=3.0
+2023-03-13 14:43:21,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1170273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:43:23,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1170275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:43:44,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1170290.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 14:43:58,122 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:44:01,934 INFO [train.py:968] (0/2) Epoch 26, batch 31250, libri_loss[loss=0.2681, simple_loss=0.3447, pruned_loss=0.09575, over 29523.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3326, pruned_loss=0.08738, over 5638288.18 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3535, pruned_loss=0.1115, over 5693601.88 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3318, pruned_loss=0.0855, over 5637792.18 frames. ], batch size: 84, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:45:06,032 INFO [train.py:968] (0/2) Epoch 26, batch 31300, giga_loss[loss=0.249, simple_loss=0.3188, pruned_loss=0.08958, over 28849.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3326, pruned_loss=0.0881, over 5651845.57 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3538, pruned_loss=0.1116, over 5691753.39 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3317, pruned_loss=0.08627, over 5652731.24 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:45:15,309 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 14:45:20,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.553e+02 1.542e+03 1.979e+03 2.528e+03 6.073e+03, threshold=3.958e+03, percent-clipped=12.0
+2023-03-13 14:46:03,072 INFO [train.py:968] (0/2) Epoch 26, batch 31350, giga_loss[loss=0.3168, simple_loss=0.3849, pruned_loss=0.1244, over 28623.00 frames. ], tot_loss[loss=0.254, simple_loss=0.332, pruned_loss=0.08799, over 5662779.36 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3534, pruned_loss=0.1115, over 5695368.24 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.331, pruned_loss=0.08612, over 5659421.40 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:46:17,323 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1170418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:46:20,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1170421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:46:34,928 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1170433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:46:34,976 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1170433.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 14:46:35,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5328, 2.2325, 1.5570, 0.7888], device='cuda:0'), covar=tensor([0.7535, 0.3658, 0.5138, 0.7175], device='cuda:0'), in_proj_covar=tensor([0.1824, 0.1720, 0.1649, 0.1492], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 14:46:38,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1170436.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 14:46:42,827 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 14:46:54,151 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:47:01,217 INFO [train.py:968] (0/2) Epoch 26, batch 31400, giga_loss[loss=0.2917, simple_loss=0.3685, pruned_loss=0.1075, over 28996.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3334, pruned_loss=0.08779, over 5662544.80 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3534, pruned_loss=0.1115, over 5695145.56 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3325, pruned_loss=0.08609, over 5659736.58 frames. ], batch size: 285, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:47:13,652 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170465.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 14:47:19,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.815e+02 1.545e+03 2.267e+03 2.901e+03 6.220e+03, threshold=4.535e+03, percent-clipped=7.0
+2023-03-13 14:47:34,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1170481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:47:51,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5364, 1.5914, 1.7528, 1.3704], device='cuda:0'), covar=tensor([0.1778, 0.2600, 0.1477, 0.1859], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0707, 0.0969, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 14:48:08,878 INFO [train.py:968] (0/2) Epoch 26, batch 31450, giga_loss[loss=0.2305, simple_loss=0.3175, pruned_loss=0.07176, over 28847.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3352, pruned_loss=0.0886, over 5654826.09 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3532, pruned_loss=0.1114, over 5697454.17 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3344, pruned_loss=0.08703, over 5650013.84 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:48:29,621 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-13 14:49:14,165 INFO [train.py:968] (0/2) Epoch 26, batch 31500, libri_loss[loss=0.3143, simple_loss=0.3696, pruned_loss=0.1295, over 29272.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3324, pruned_loss=0.08681, over 5664239.93 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3535, pruned_loss=0.1117, over 5690411.58 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3311, pruned_loss=0.08486, over 5666940.45 frames. ], batch size: 94, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:49:29,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.499e+03 1.885e+03 2.560e+03 5.585e+03, threshold=3.769e+03, percent-clipped=6.0
+2023-03-13 14:49:33,896 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.94 vs. limit=5.0
+2023-03-13 14:49:38,023 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1170576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:49:42,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1170579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:50:07,122 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-13 14:50:13,129 INFO [train.py:968] (0/2) Epoch 26, batch 31550, giga_loss[loss=0.2591, simple_loss=0.3369, pruned_loss=0.09066, over 28466.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3341, pruned_loss=0.08887, over 5657049.17 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3532, pruned_loss=0.1118, over 5684543.55 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3326, pruned_loss=0.08638, over 5663428.89 frames. ], batch size: 369, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:50:16,486 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:50:38,745 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1170624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:50:40,925 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1170627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:50:50,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3433, 1.4806, 1.5690, 1.2110], device='cuda:0'), covar=tensor([0.1886, 0.2927, 0.1627, 0.2083], device='cuda:0'), in_proj_covar=tensor([0.0921, 0.0705, 0.0968, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 14:51:13,035 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:51:14,788 INFO [train.py:968] (0/2) Epoch 26, batch 31600, giga_loss[loss=0.2369, simple_loss=0.3324, pruned_loss=0.07072, over 28137.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3369, pruned_loss=0.08861, over 5666059.51 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3529, pruned_loss=0.1116, over 5689638.33 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3356, pruned_loss=0.08626, over 5665876.23 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:51:31,745 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1170670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:51:33,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.173e+02 1.559e+03 2.279e+03 3.294e+03 8.029e+03, threshold=4.558e+03, percent-clipped=16.0
+2023-03-13 14:51:58,826 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1170691.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 14:51:58,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8473, 3.6720, 3.5153, 1.7569], device='cuda:0'), covar=tensor([0.0682, 0.0810, 0.0750, 0.2442], device='cuda:0'), in_proj_covar=tensor([0.1283, 0.1181, 0.0997, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 14:52:16,760 INFO [train.py:968] (0/2) Epoch 26, batch 31650, giga_loss[loss=0.2503, simple_loss=0.3421, pruned_loss=0.07924, over 28893.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3387, pruned_loss=0.08733, over 5663536.90 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3529, pruned_loss=0.1117, over 5691528.36 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3373, pruned_loss=0.0848, over 5661105.81 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:53:17,498 INFO [train.py:968] (0/2) Epoch 26, batch 31700, giga_loss[loss=0.2453, simple_loss=0.3394, pruned_loss=0.07555, over 28923.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3385, pruned_loss=0.08582, over 5664472.09 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3523, pruned_loss=0.1113, over 5695821.58 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3376, pruned_loss=0.08367, over 5658171.89 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:53:36,441 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.167e+02 1.437e+03 2.169e+03 2.721e+03 6.428e+03, threshold=4.338e+03, percent-clipped=2.0
+2023-03-13 14:54:20,492 INFO [train.py:968] (0/2) Epoch 26, batch 31750, giga_loss[loss=0.2424, simple_loss=0.3321, pruned_loss=0.0763, over 28369.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3366, pruned_loss=0.08375, over 5669542.00 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3521, pruned_loss=0.1111, over 5697642.56 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.336, pruned_loss=0.08199, over 5662733.75 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:55:28,145 INFO [train.py:968] (0/2) Epoch 26, batch 31800, giga_loss[loss=0.2395, simple_loss=0.3267, pruned_loss=0.07616, over 28900.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3379, pruned_loss=0.08556, over 5682888.14 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3519, pruned_loss=0.1111, over 5700458.44 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3373, pruned_loss=0.08384, over 5674708.96 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:55:49,679 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.656e+02 1.272e+03 1.758e+03 2.623e+03 5.701e+03, threshold=3.516e+03, percent-clipped=5.0
+2023-03-13 14:56:42,294 INFO [train.py:968] (0/2) Epoch 26, batch 31850, giga_loss[loss=0.2263, simple_loss=0.317, pruned_loss=0.06777, over 28593.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.338, pruned_loss=0.08726, over 5674339.72 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3521, pruned_loss=0.1113, over 5693815.34 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3372, pruned_loss=0.0853, over 5672732.24 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:57:55,070 INFO [train.py:968] (0/2) Epoch 26, batch 31900, giga_loss[loss=0.2688, simple_loss=0.3422, pruned_loss=0.09769, over 28694.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.338, pruned_loss=0.08833, over 5668422.92 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3516, pruned_loss=0.1111, over 5686861.51 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3375, pruned_loss=0.08642, over 5673742.78 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:58:20,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.702e+02 1.550e+03 1.890e+03 2.763e+03 9.306e+03, threshold=3.780e+03, percent-clipped=11.0
+2023-03-13 14:58:39,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1660, 1.3210, 3.3562, 3.0575], device='cuda:0'), covar=tensor([0.1613, 0.2659, 0.0514, 0.0985], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0665, 0.0985, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 14:59:03,957 INFO [train.py:968] (0/2) Epoch 26, batch 31950, giga_loss[loss=0.2271, simple_loss=0.3169, pruned_loss=0.06868, over 29053.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3342, pruned_loss=0.08637, over 5666840.80 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3522, pruned_loss=0.1117, over 5688598.13 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3328, pruned_loss=0.08378, over 5669150.44 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:59:47,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3581, 2.2203, 1.3457, 1.5127], device='cuda:0'), covar=tensor([0.0827, 0.0437, 0.0806, 0.1186], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0567, 0.0406, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 14:59:53,549 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1171045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:00:13,439 INFO [train.py:968] (0/2) Epoch 26, batch 32000, giga_loss[loss=0.2292, simple_loss=0.3104, pruned_loss=0.07403, over 28787.00 frames. ], tot_loss[loss=0.25, simple_loss=0.331, pruned_loss=0.08451, over 5674652.83 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3518, pruned_loss=0.1115, over 5691883.40 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.33, pruned_loss=0.08228, over 5673338.55 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:00:23,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1171066.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 15:00:31,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.347e+02 1.333e+03 1.675e+03 2.025e+03 5.664e+03, threshold=3.349e+03, percent-clipped=4.0
+2023-03-13 15:01:17,747 INFO [train.py:968] (0/2) Epoch 26, batch 32050, giga_loss[loss=0.2658, simple_loss=0.3523, pruned_loss=0.08967, over 28386.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3303, pruned_loss=0.08453, over 5676083.81 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3513, pruned_loss=0.1112, over 5687244.69 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3296, pruned_loss=0.08245, over 5677921.10 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:02:16,512 INFO [train.py:968] (0/2) Epoch 26, batch 32100, giga_loss[loss=0.2732, simple_loss=0.3577, pruned_loss=0.09435, over 28735.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3341, pruned_loss=0.08657, over 5681294.31 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3512, pruned_loss=0.1112, over 5691727.34 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.333, pruned_loss=0.084, over 5678577.09 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:02:36,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.173e+03 1.647e+03 2.246e+03 3.109e+03 7.748e+03, threshold=4.491e+03, percent-clipped=15.0
+2023-03-13 15:02:49,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1171188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:02:54,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1171191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:03:16,164 INFO [train.py:968] (0/2) Epoch 26, batch 32150, libri_loss[loss=0.2958, simple_loss=0.3635, pruned_loss=0.1141, over 29068.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3328, pruned_loss=0.08739, over 5688450.70 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3507, pruned_loss=0.1109, over 5698063.00 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3318, pruned_loss=0.08484, over 5680280.13 frames. ], batch size: 101, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:03:18,502 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1171209.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 15:03:23,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1171212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 15:03:35,573 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1171220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:03:35,845 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 15:04:00,362 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1171241.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 15:04:12,535 INFO [train.py:968] (0/2) Epoch 26, batch 32200, giga_loss[loss=0.2697, simple_loss=0.3512, pruned_loss=0.09407, over 28701.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3334, pruned_loss=0.08868, over 5674845.75 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3506, pruned_loss=0.1109, over 5683848.49 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3321, pruned_loss=0.0859, over 5681441.06 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:04:32,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.307e+02 1.485e+03 1.922e+03 2.805e+03 6.259e+03, threshold=3.844e+03, percent-clipped=8.0
+2023-03-13 15:04:55,447 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3880, 3.6197, 1.6773, 1.5552], device='cuda:0'), covar=tensor([0.1043, 0.0449, 0.0958, 0.1426], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0565, 0.0404, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 15:05:12,973 INFO [train.py:968] (0/2) Epoch 26, batch 32250, libri_loss[loss=0.2376, simple_loss=0.3003, pruned_loss=0.08745, over 29477.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3331, pruned_loss=0.08897, over 5666882.27 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3501, pruned_loss=0.1108, over 5674955.46 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3323, pruned_loss=0.08656, over 5679498.24 frames. ], batch size: 70, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:05:49,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2393, 3.2227, 1.3925, 1.4874], device='cuda:0'), covar=tensor([0.1167, 0.0446, 0.1011, 0.1462], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0564, 0.0403, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 15:06:23,845 INFO [train.py:968] (0/2) Epoch 26, batch 32300, giga_loss[loss=0.2536, simple_loss=0.3459, pruned_loss=0.08067, over 28843.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3346, pruned_loss=0.08899, over 5672237.25 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3495, pruned_loss=0.1105, over 5681857.68 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.334, pruned_loss=0.08672, over 5675839.41 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:06:50,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.158e+02 1.458e+03 1.974e+03 2.544e+03 6.803e+03, threshold=3.947e+03, percent-clipped=9.0
+2023-03-13 15:07:39,661 INFO [train.py:968] (0/2) Epoch 26, batch 32350, giga_loss[loss=0.2482, simple_loss=0.3365, pruned_loss=0.07988, over 29009.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3363, pruned_loss=0.08925, over 5668808.57 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3494, pruned_loss=0.1104, over 5684941.22 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3357, pruned_loss=0.08721, over 5668485.65 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:07:52,964 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1171415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:08:08,420 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-13 15:08:46,102 INFO [train.py:968] (0/2) Epoch 26, batch 32400, giga_loss[loss=0.2369, simple_loss=0.3143, pruned_loss=0.07976, over 28828.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3336, pruned_loss=0.08787, over 5672281.25 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3496, pruned_loss=0.1106, over 5686959.61 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3325, pruned_loss=0.08548, over 5669928.78 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:09:09,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.895e+02 1.547e+03 2.239e+03 3.143e+03 5.566e+03, threshold=4.479e+03, percent-clipped=10.0
+2023-03-13 15:09:50,039 INFO [train.py:968] (0/2) Epoch 26, batch 32450, libri_loss[loss=0.2836, simple_loss=0.3513, pruned_loss=0.1079, over 29543.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3295, pruned_loss=0.087, over 5661305.00 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3492, pruned_loss=0.1105, over 5672582.35 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3285, pruned_loss=0.08457, over 5672347.63 frames. ], batch size: 80, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:10:50,281 INFO [train.py:968] (0/2) Epoch 26, batch 32500, libri_loss[loss=0.2907, simple_loss=0.3569, pruned_loss=0.1123, over 28679.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3268, pruned_loss=0.08627, over 5664064.58 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3497, pruned_loss=0.1109, over 5679110.98 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3248, pruned_loss=0.08294, over 5666573.33 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:10:58,869 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 15:11:11,493 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.511e+02 1.594e+03 2.125e+03 2.781e+03 7.501e+03, threshold=4.249e+03, percent-clipped=6.0
+2023-03-13 15:11:47,447 INFO [train.py:968] (0/2) Epoch 26, batch 32550, giga_loss[loss=0.2544, simple_loss=0.333, pruned_loss=0.08794, over 28698.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3279, pruned_loss=0.08732, over 5651621.56 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3492, pruned_loss=0.1108, over 5664347.34 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.326, pruned_loss=0.08408, over 5666779.68 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:12:45,003 INFO [train.py:968] (0/2) Epoch 26, batch 32600, giga_loss[loss=0.2483, simple_loss=0.3311, pruned_loss=0.08272, over 28642.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3281, pruned_loss=0.08713, over 5663262.96 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3485, pruned_loss=0.1104, over 5668589.13 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3268, pruned_loss=0.08431, over 5671336.22 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:13:05,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.701e+02 1.533e+03 1.888e+03 2.372e+03 4.932e+03, threshold=3.776e+03, percent-clipped=2.0
+2023-03-13 15:13:45,126 INFO [train.py:968] (0/2) Epoch 26, batch 32650, giga_loss[loss=0.2161, simple_loss=0.3055, pruned_loss=0.06335, over 28917.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3252, pruned_loss=0.08441, over 5660254.32 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3483, pruned_loss=0.1102, over 5671374.76 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.324, pruned_loss=0.08194, over 5663965.36 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:14:37,443 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5111, 1.3172, 4.0036, 3.3158], device='cuda:0'), covar=tensor([0.1511, 0.2724, 0.0480, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0665, 0.0984, 0.0948], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 15:14:43,258 INFO [train.py:968] (0/2) Epoch 26, batch 32700, giga_loss[loss=0.2012, simple_loss=0.2916, pruned_loss=0.05536, over 28452.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3257, pruned_loss=0.08478, over 5651546.01 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3487, pruned_loss=0.1105, over 5665634.18 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3235, pruned_loss=0.08146, over 5659889.69 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:15:09,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.090e+02 1.586e+03 2.088e+03 2.575e+03 4.857e+03, threshold=4.175e+03, percent-clipped=5.0
+2023-03-13 15:15:29,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1171790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:15:51,374 INFO [train.py:968] (0/2) Epoch 26, batch 32750, giga_loss[loss=0.2245, simple_loss=0.3068, pruned_loss=0.0711, over 28701.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.324, pruned_loss=0.08427, over 5652540.37 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3484, pruned_loss=0.1105, over 5660158.40 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.322, pruned_loss=0.08115, over 5664303.16 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:16:06,078 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-13 15:16:50,667 INFO [train.py:968] (0/2) Epoch 26, batch 32800, giga_loss[loss=0.3255, simple_loss=0.3952, pruned_loss=0.1279, over 28928.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3258, pruned_loss=0.08473, over 5666830.01 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3481, pruned_loss=0.1105, over 5664162.41 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3234, pruned_loss=0.08113, over 5672462.40 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:17:10,460 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.639e+02 1.597e+03 2.137e+03 2.942e+03 6.906e+03, threshold=4.273e+03, percent-clipped=7.0
+2023-03-13 15:17:56,304 INFO [train.py:968] (0/2) Epoch 26, batch 32850, giga_loss[loss=0.2689, simple_loss=0.3424, pruned_loss=0.09774, over 28037.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3256, pruned_loss=0.08445, over 5673319.19 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3476, pruned_loss=0.1102, over 5669945.37 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3236, pruned_loss=0.08122, over 5673130.84 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:18:29,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1171933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:18:31,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1171936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:18:55,505 INFO [train.py:968] (0/2) Epoch 26, batch 32900, giga_loss[loss=0.2393, simple_loss=0.3156, pruned_loss=0.08151, over 29006.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3254, pruned_loss=0.08502, over 5683245.14 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.347, pruned_loss=0.1099, over 5676658.59 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3238, pruned_loss=0.082, over 5677151.59 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:19:01,027 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1171962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:19:04,921 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1171965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:19:16,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.163e+02 1.309e+03 1.707e+03 2.270e+03 3.758e+03, threshold=3.415e+03, percent-clipped=0.0
+2023-03-13 15:19:47,921 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1172000.pt
+2023-03-13 15:19:58,777 INFO [train.py:968] (0/2) Epoch 26, batch 32950, giga_loss[loss=0.2367, simple_loss=0.3098, pruned_loss=0.0818, over 26872.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3235, pruned_loss=0.08299, over 5676650.08 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3467, pruned_loss=0.1097, over 5679668.71 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3222, pruned_loss=0.08041, over 5669435.44 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:20:53,275 INFO [train.py:968] (0/2) Epoch 26, batch 33000, giga_loss[loss=0.2717, simple_loss=0.3467, pruned_loss=0.09832, over 26914.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3259, pruned_loss=0.08318, over 5660723.82 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3469, pruned_loss=0.1098, over 5673708.97 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3241, pruned_loss=0.0803, over 5660616.05 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:20:53,280 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 15:21:02,129 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.1933, simple_loss=0.2948, pruned_loss=0.04585, over 944034.00 frames. 
+2023-03-13 15:21:02,130 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 15:21:22,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.465e+03 2.010e+03 3.031e+03 7.195e+03, threshold=4.021e+03, percent-clipped=20.0
+2023-03-13 15:21:51,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5498, 1.2930, 4.2404, 3.4395], device='cuda:0'), covar=tensor([0.1524, 0.2709, 0.0507, 0.0966], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0666, 0.0984, 0.0947], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 15:21:54,830 INFO [train.py:968] (0/2) Epoch 26, batch 33050, giga_loss[loss=0.2952, simple_loss=0.3675, pruned_loss=0.1114, over 28881.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3291, pruned_loss=0.08497, over 5660497.82 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3461, pruned_loss=0.1094, over 5674368.51 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3274, pruned_loss=0.08169, over 5659719.77 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:22:48,966 INFO [train.py:968] (0/2) Epoch 26, batch 33100, giga_loss[loss=0.2174, simple_loss=0.3094, pruned_loss=0.06267, over 29138.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3309, pruned_loss=0.08622, over 5651782.09 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3459, pruned_loss=0.1097, over 5661591.34 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.329, pruned_loss=0.08223, over 5662891.75 frames. ], batch size: 113, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:23:16,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.991e+02 1.593e+03 1.999e+03 2.881e+03 8.416e+03, threshold=3.998e+03, percent-clipped=6.0
+2023-03-13 15:23:52,728 INFO [train.py:968] (0/2) Epoch 26, batch 33150, giga_loss[loss=0.2359, simple_loss=0.3234, pruned_loss=0.0742, over 28754.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3318, pruned_loss=0.08727, over 5636655.22 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3463, pruned_loss=0.1102, over 5646429.35 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3295, pruned_loss=0.08307, over 5659110.67 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:23:59,563 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5425, 4.3958, 4.1355, 2.0167], device='cuda:0'), covar=tensor([0.0569, 0.0715, 0.0859, 0.2006], device='cuda:0'), in_proj_covar=tensor([0.1276, 0.1170, 0.0989, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 15:24:46,116 INFO [train.py:968] (0/2) Epoch 26, batch 33200, giga_loss[loss=0.2205, simple_loss=0.3026, pruned_loss=0.06917, over 28820.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3284, pruned_loss=0.08494, over 5651428.50 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3457, pruned_loss=0.1099, over 5644887.74 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3267, pruned_loss=0.08112, over 5670194.14 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:25:08,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.934e+02 1.346e+03 1.854e+03 2.539e+03 1.266e+04, threshold=3.708e+03, percent-clipped=11.0
+2023-03-13 15:25:46,703 INFO [train.py:968] (0/2) Epoch 26, batch 33250, giga_loss[loss=0.2283, simple_loss=0.3175, pruned_loss=0.06958, over 28465.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3266, pruned_loss=0.08374, over 5655635.24 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3455, pruned_loss=0.1099, over 5641687.09 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.325, pruned_loss=0.08014, over 5673133.32 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:26:01,590 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2026, 1.5529, 1.5176, 1.0940], device='cuda:0'), covar=tensor([0.1779, 0.2764, 0.1502, 0.1822], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0703, 0.0969, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 15:26:19,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1172337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:26:40,750 INFO [train.py:968] (0/2) Epoch 26, batch 33300, libri_loss[loss=0.3278, simple_loss=0.3756, pruned_loss=0.14, over 25785.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3246, pruned_loss=0.08332, over 5660847.04 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3448, pruned_loss=0.1095, over 5642599.92 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3234, pruned_loss=0.08001, over 5674836.12 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:27:06,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.940e+02 1.443e+03 1.887e+03 2.355e+03 5.508e+03, threshold=3.774e+03, percent-clipped=1.0
+2023-03-13 15:27:13,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1880, 3.2398, 1.3473, 1.4726], device='cuda:0'), covar=tensor([0.1254, 0.0394, 0.1079, 0.1509], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0565, 0.0404, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 15:27:40,571 INFO [train.py:968] (0/2) Epoch 26, batch 33350, libri_loss[loss=0.3194, simple_loss=0.3699, pruned_loss=0.1344, over 29523.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3253, pruned_loss=0.08348, over 5664223.06 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3446, pruned_loss=0.1095, over 5647546.51 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3239, pruned_loss=0.08018, over 5671009.50 frames. ], batch size: 83, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:28:04,964 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1172426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:28:09,655 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7970, 4.6369, 4.4126, 2.1795], device='cuda:0'), covar=tensor([0.0485, 0.0637, 0.0772, 0.2079], device='cuda:0'), in_proj_covar=tensor([0.1277, 0.1171, 0.0989, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 15:28:11,636 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1172431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:28:43,520 INFO [train.py:968] (0/2) Epoch 26, batch 33400, giga_loss[loss=0.2615, simple_loss=0.3421, pruned_loss=0.09042, over 28696.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3278, pruned_loss=0.08437, over 5660176.52 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3446, pruned_loss=0.1095, over 5643087.87 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3265, pruned_loss=0.08135, over 5670237.46 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:29:06,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.858e+02 1.401e+03 1.858e+03 2.519e+03 5.755e+03, threshold=3.716e+03, percent-clipped=4.0
+2023-03-13 15:29:11,169 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1172480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:29:11,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6942, 2.0179, 1.6281, 1.9729], device='cuda:0'), covar=tensor([0.2995, 0.2999, 0.3400, 0.2668], device='cuda:0'), in_proj_covar=tensor([0.1582, 0.1133, 0.1397, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 15:29:16,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1172483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:29:45,030 INFO [train.py:968] (0/2) Epoch 26, batch 33450, giga_loss[loss=0.2222, simple_loss=0.2919, pruned_loss=0.0762, over 24585.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3289, pruned_loss=0.0855, over 5664621.84 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3446, pruned_loss=0.1093, over 5649675.08 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3275, pruned_loss=0.08266, over 5667276.81 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:29:54,813 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1172512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:30:04,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1281, 1.4020, 1.4100, 1.1477], device='cuda:0'), covar=tensor([0.3219, 0.2549, 0.1684, 0.2533], device='cuda:0'), in_proj_covar=tensor([0.2008, 0.1948, 0.1857, 0.1996], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 15:30:18,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1412, 1.4771, 1.1541, 0.5250], device='cuda:0'), covar=tensor([0.3662, 0.2239, 0.3292, 0.5685], device='cuda:0'), in_proj_covar=tensor([0.1821, 0.1721, 0.1648, 0.1491], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 15:30:47,408 INFO [train.py:968] (0/2) Epoch 26, batch 33500, giga_loss[loss=0.256, simple_loss=0.3221, pruned_loss=0.09489, over 24340.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.331, pruned_loss=0.08727, over 5653583.88 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3439, pruned_loss=0.1089, over 5655652.59 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3299, pruned_loss=0.08448, over 5650280.47 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:31:10,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.497e+03 2.049e+03 2.776e+03 6.662e+03, threshold=4.097e+03, percent-clipped=4.0
+2023-03-13 15:31:40,367 INFO [train.py:968] (0/2) Epoch 26, batch 33550, giga_loss[loss=0.2521, simple_loss=0.3389, pruned_loss=0.08267, over 28872.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3338, pruned_loss=0.08804, over 5667058.41 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3438, pruned_loss=0.1088, over 5663322.64 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3327, pruned_loss=0.08517, over 5657423.72 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:31:42,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6976, 1.8409, 1.5614, 1.8486], device='cuda:0'), covar=tensor([0.2711, 0.2835, 0.3123, 0.2680], device='cuda:0'), in_proj_covar=tensor([0.1579, 0.1131, 0.1394, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 15:32:09,267 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 15:32:34,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6452, 1.7891, 1.4425, 1.7794], device='cuda:0'), covar=tensor([0.2848, 0.2891, 0.3310, 0.2492], device='cuda:0'), in_proj_covar=tensor([0.1578, 0.1131, 0.1393, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 15:32:39,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2462, 1.5906, 0.9466, 1.1684], device='cuda:0'), covar=tensor([0.1272, 0.0707, 0.1545, 0.1485], device='cuda:0'), in_proj_covar=tensor([0.0402, 0.0443, 0.0517, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 15:32:39,918 INFO [train.py:968] (0/2) Epoch 26, batch 33600, giga_loss[loss=0.2965, simple_loss=0.3764, pruned_loss=0.1083, over 28471.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3353, pruned_loss=0.08863, over 5666641.38 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3431, pruned_loss=0.1084, over 5667407.60 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3348, pruned_loss=0.08605, over 5655170.64 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:32:59,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.343e+02 1.512e+03 1.906e+03 2.617e+03 6.436e+03, threshold=3.811e+03, percent-clipped=5.0
+2023-03-13 15:33:42,512 INFO [train.py:968] (0/2) Epoch 26, batch 33650, giga_loss[loss=0.2599, simple_loss=0.3381, pruned_loss=0.09083, over 28996.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3345, pruned_loss=0.08812, over 5683696.32 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3426, pruned_loss=0.1081, over 5675361.55 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3342, pruned_loss=0.08571, over 5667679.90 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:33:51,926 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1172712.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 15:34:16,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5155, 3.5566, 1.6169, 1.7222], device='cuda:0'), covar=tensor([0.0996, 0.0384, 0.0946, 0.1331], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0564, 0.0404, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 15:34:27,962 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 15:34:51,908 INFO [train.py:968] (0/2) Epoch 26, batch 33700, giga_loss[loss=0.2299, simple_loss=0.3109, pruned_loss=0.07449, over 28906.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3332, pruned_loss=0.08774, over 5683218.25 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3427, pruned_loss=0.1081, over 5668911.42 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3327, pruned_loss=0.08545, over 5676111.42 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:35:15,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.547e+03 2.096e+03 2.763e+03 7.696e+03, threshold=4.192e+03, percent-clipped=7.0
+2023-03-13 15:35:46,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1172801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:35:53,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1172806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:35:53,396 INFO [train.py:968] (0/2) Epoch 26, batch 33750, giga_loss[loss=0.2459, simple_loss=0.3289, pruned_loss=0.08146, over 28569.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3325, pruned_loss=0.08764, over 5666360.19 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3427, pruned_loss=0.1084, over 5655823.70 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3318, pruned_loss=0.08494, over 5672660.30 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:36:58,335 INFO [train.py:968] (0/2) Epoch 26, batch 33800, giga_loss[loss=0.2618, simple_loss=0.3334, pruned_loss=0.09514, over 28847.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3308, pruned_loss=0.08764, over 5667608.03 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3428, pruned_loss=0.1084, over 5657489.56 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.33, pruned_loss=0.08508, over 5671316.65 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:37:28,513 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.530e+02 1.539e+03 2.005e+03 2.930e+03 1.096e+04, threshold=4.010e+03, percent-clipped=13.0
+2023-03-13 15:38:03,433 INFO [train.py:968] (0/2) Epoch 26, batch 33850, giga_loss[loss=0.2734, simple_loss=0.3458, pruned_loss=0.1005, over 28944.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3284, pruned_loss=0.08668, over 5678055.41 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3424, pruned_loss=0.1081, over 5661872.07 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3279, pruned_loss=0.08454, over 5677300.51 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:38:47,700 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1172944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:38:50,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1172947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:38:53,525 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1172949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:38:55,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1172952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:38:59,769 INFO [train.py:968] (0/2) Epoch 26, batch 33900, giga_loss[loss=0.2637, simple_loss=0.3475, pruned_loss=0.08993, over 28635.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3291, pruned_loss=0.08613, over 5679661.03 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3425, pruned_loss=0.1083, over 5665288.20 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3283, pruned_loss=0.08361, over 5676412.97 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:39:23,350 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1172976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:39:25,750 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.909e+02 1.440e+03 1.938e+03 3.093e+03 8.463e+03, threshold=3.877e+03, percent-clipped=12.0
+2023-03-13 15:39:29,827 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1172981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:39:50,455 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 15:39:51,829 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1173001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:40:00,041 INFO [train.py:968] (0/2) Epoch 26, batch 33950, giga_loss[loss=0.256, simple_loss=0.3422, pruned_loss=0.0849, over 27562.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3268, pruned_loss=0.08399, over 5674098.58 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3421, pruned_loss=0.1081, over 5669234.77 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.326, pruned_loss=0.08146, over 5668261.47 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:40:09,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4501, 1.6017, 1.6773, 1.2767], device='cuda:0'), covar=tensor([0.1954, 0.2895, 0.1695, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0705, 0.0972, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0015], device='cuda:0')
+2023-03-13 15:40:55,926 INFO [train.py:968] (0/2) Epoch 26, batch 34000, giga_loss[loss=0.2406, simple_loss=0.3337, pruned_loss=0.07371, over 28461.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3295, pruned_loss=0.08342, over 5678963.83 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3423, pruned_loss=0.1082, over 5670160.27 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3284, pruned_loss=0.08088, over 5673719.32 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:41:08,108 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1173067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:41:21,022 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.706e+02 1.474e+03 1.787e+03 2.457e+03 5.593e+03, threshold=3.575e+03, percent-clipped=3.0
+2023-03-13 15:41:25,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3162, 1.6762, 1.6554, 1.4248], device='cuda:0'), covar=tensor([0.2155, 0.2057, 0.2242, 0.2163], device='cuda:0'), in_proj_covar=tensor([0.0480, 0.0734, 0.0706, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 15:41:28,585 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1173087.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 15:41:49,984 INFO [train.py:968] (0/2) Epoch 26, batch 34050, giga_loss[loss=0.24, simple_loss=0.3293, pruned_loss=0.07534, over 28757.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3304, pruned_loss=0.08347, over 5676503.62 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3426, pruned_loss=0.1084, over 5664600.67 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.329, pruned_loss=0.08056, over 5677193.33 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:41:58,822 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5259, 1.5395, 1.6980, 1.3286], device='cuda:0'), covar=tensor([0.1950, 0.2810, 0.1644, 0.1962], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0706, 0.0972, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0015], device='cuda:0')
+2023-03-13 15:42:43,777 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 15:42:56,192 INFO [train.py:968] (0/2) Epoch 26, batch 34100, giga_loss[loss=0.2549, simple_loss=0.3388, pruned_loss=0.08552, over 28604.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3311, pruned_loss=0.08404, over 5674991.38 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3424, pruned_loss=0.1084, over 5668283.53 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3298, pruned_loss=0.08107, over 5672481.31 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:43:25,114 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.459e+03 1.840e+03 2.364e+03 5.454e+03, threshold=3.681e+03, percent-clipped=8.0
+2023-03-13 15:44:00,111 INFO [train.py:968] (0/2) Epoch 26, batch 34150, giga_loss[loss=0.2325, simple_loss=0.3178, pruned_loss=0.07354, over 28917.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3309, pruned_loss=0.08415, over 5669378.51 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3424, pruned_loss=0.1084, over 5672056.19 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3296, pruned_loss=0.08111, over 5663987.22 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:44:06,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-13 15:44:30,449 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1173230.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 15:44:33,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1173233.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 15:45:03,602 INFO [train.py:968] (0/2) Epoch 26, batch 34200, giga_loss[loss=0.2433, simple_loss=0.337, pruned_loss=0.07476, over 28737.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3315, pruned_loss=0.08487, over 5671234.02 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3425, pruned_loss=0.1087, over 5678314.81 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3298, pruned_loss=0.08105, over 5660990.05 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:45:10,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1173262.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 15:45:34,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.814e+02 1.483e+03 1.855e+03 2.511e+03 1.253e+04, threshold=3.709e+03, percent-clipped=7.0
+2023-03-13 15:46:14,186 INFO [train.py:968] (0/2) Epoch 26, batch 34250, giga_loss[loss=0.2239, simple_loss=0.3204, pruned_loss=0.06369, over 28888.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.332, pruned_loss=0.08427, over 5673863.18 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3424, pruned_loss=0.1087, over 5681786.30 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3303, pruned_loss=0.08053, over 5662161.91 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:47:15,701 INFO [train.py:968] (0/2) Epoch 26, batch 34300, giga_loss[loss=0.2366, simple_loss=0.3304, pruned_loss=0.07142, over 28152.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3336, pruned_loss=0.08525, over 5673431.92 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3422, pruned_loss=0.1086, over 5678967.14 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3322, pruned_loss=0.08155, over 5666074.99 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:47:37,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1173376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:47:40,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.197e+02 1.449e+03 1.957e+03 2.730e+03 1.230e+04, threshold=3.914e+03, percent-clipped=12.0
+2023-03-13 15:48:19,518 INFO [train.py:968] (0/2) Epoch 26, batch 34350, giga_loss[loss=0.2661, simple_loss=0.3534, pruned_loss=0.08937, over 28699.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3366, pruned_loss=0.08634, over 5671695.40 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3418, pruned_loss=0.1084, over 5673403.51 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3357, pruned_loss=0.08314, over 5670134.70 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:48:25,911 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-13 15:49:06,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1173442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:49:23,354 INFO [train.py:968] (0/2) Epoch 26, batch 34400, giga_loss[loss=0.2333, simple_loss=0.3226, pruned_loss=0.07197, over 28458.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3352, pruned_loss=0.08598, over 5667841.10 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3412, pruned_loss=0.108, over 5670619.44 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3349, pruned_loss=0.08317, over 5668374.67 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:49:50,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.195e+02 1.502e+03 2.114e+03 2.784e+03 5.376e+03, threshold=4.228e+03, percent-clipped=10.0
+2023-03-13 15:50:31,410 INFO [train.py:968] (0/2) Epoch 26, batch 34450, libri_loss[loss=0.274, simple_loss=0.343, pruned_loss=0.1025, over 29473.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3332, pruned_loss=0.08536, over 5683237.94 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3411, pruned_loss=0.1078, over 5674911.48 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3329, pruned_loss=0.08288, over 5679841.40 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:50:48,477 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1173519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:50:52,317 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1173522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:51:31,414 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1173551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:51:36,500 INFO [train.py:968] (0/2) Epoch 26, batch 34500, giga_loss[loss=0.2175, simple_loss=0.3153, pruned_loss=0.05981, over 28986.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3322, pruned_loss=0.08464, over 5673946.27 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3408, pruned_loss=0.1076, over 5671273.55 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.332, pruned_loss=0.08212, over 5674994.15 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:51:39,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5277, 1.8063, 1.5039, 1.5152], device='cuda:0'), covar=tensor([0.2720, 0.2587, 0.2957, 0.2383], device='cuda:0'), in_proj_covar=tensor([0.1577, 0.1130, 0.1393, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 15:51:52,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2552, 1.1573, 3.8706, 3.3182], device='cuda:0'), covar=tensor([0.1739, 0.3038, 0.0444, 0.0997], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0669, 0.0984, 0.0950], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 15:52:03,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.892e+02 1.388e+03 1.994e+03 2.740e+03 7.532e+03, threshold=3.989e+03, percent-clipped=12.0
+2023-03-13 15:52:08,274 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1173585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:52:11,114 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1173588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:52:27,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5543, 1.5794, 1.7885, 1.4577], device='cuda:0'), covar=tensor([0.1785, 0.2492, 0.1543, 0.1947], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0703, 0.0968, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 15:52:37,970 INFO [train.py:968] (0/2) Epoch 26, batch 34550, giga_loss[loss=0.2331, simple_loss=0.3216, pruned_loss=0.07232, over 28681.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3311, pruned_loss=0.08379, over 5674004.49 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3409, pruned_loss=0.1078, over 5675275.06 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3307, pruned_loss=0.08124, over 5671449.72 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:52:47,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1173617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:53:20,707 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1745, 4.0058, 3.8289, 1.8542], device='cuda:0'), covar=tensor([0.0694, 0.0820, 0.0896, 0.2131], device='cuda:0'), in_proj_covar=tensor([0.1272, 0.1166, 0.0985, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 15:53:26,664 INFO [train.py:968] (0/2) Epoch 26, batch 34600, giga_loss[loss=0.2444, simple_loss=0.3365, pruned_loss=0.07613, over 28970.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3325, pruned_loss=0.08534, over 5683798.12 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3401, pruned_loss=0.1071, over 5688791.68 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3324, pruned_loss=0.08245, over 5669361.04 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:53:38,829 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1173666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:53:44,431 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2165, 1.3760, 3.6918, 3.1821], device='cuda:0'), covar=tensor([0.2035, 0.3150, 0.0859, 0.1532], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0669, 0.0985, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 15:53:54,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.120e+02 1.410e+03 1.771e+03 2.337e+03 1.581e+04, threshold=3.543e+03, percent-clipped=7.0
+2023-03-13 15:54:26,938 INFO [train.py:968] (0/2) Epoch 26, batch 34650, libri_loss[loss=0.2805, simple_loss=0.3333, pruned_loss=0.1139, over 29352.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3355, pruned_loss=0.08702, over 5673402.97 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3402, pruned_loss=0.1072, over 5683386.84 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3351, pruned_loss=0.08415, over 5666956.18 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:54:54,485 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1173729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:55:16,532 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 15:55:26,870 INFO [train.py:968] (0/2) Epoch 26, batch 34700, giga_loss[loss=0.227, simple_loss=0.3113, pruned_loss=0.07136, over 28940.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3351, pruned_loss=0.08707, over 5675264.37 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3402, pruned_loss=0.1072, over 5675997.82 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3348, pruned_loss=0.08446, over 5677340.73 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:55:48,937 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1656, 2.4779, 1.2097, 1.3888], device='cuda:0'), covar=tensor([0.1061, 0.0541, 0.1016, 0.1436], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0561, 0.0404, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 15:55:55,254 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.846e+02 1.440e+03 1.773e+03 2.478e+03 4.731e+03, threshold=3.545e+03, percent-clipped=8.0
+2023-03-13 15:56:25,216 INFO [train.py:968] (0/2) Epoch 26, batch 34750, giga_loss[loss=0.2188, simple_loss=0.2887, pruned_loss=0.07443, over 24310.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3331, pruned_loss=0.08714, over 5663880.31 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.34, pruned_loss=0.107, over 5678320.98 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.333, pruned_loss=0.08502, over 5663410.60 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:56:44,467 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1173823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:56:48,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4337, 1.7548, 1.5482, 1.6234], device='cuda:0'), covar=tensor([0.0688, 0.0302, 0.0318, 0.0735], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0120, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0073, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 15:57:22,401 INFO [train.py:968] (0/2) Epoch 26, batch 34800, giga_loss[loss=0.2464, simple_loss=0.3266, pruned_loss=0.08312, over 28699.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3316, pruned_loss=0.08649, over 5672019.70 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3397, pruned_loss=0.1069, over 5683686.33 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3315, pruned_loss=0.08436, over 5666412.64 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:57:46,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.767e+02 1.576e+03 2.014e+03 2.849e+03 6.445e+03, threshold=4.027e+03, percent-clipped=11.0
+2023-03-13 15:58:11,216 INFO [train.py:968] (0/2) Epoch 26, batch 34850, giga_loss[loss=0.2616, simple_loss=0.3562, pruned_loss=0.08351, over 29035.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3389, pruned_loss=0.09077, over 5671479.57 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3398, pruned_loss=0.1069, over 5685420.94 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3387, pruned_loss=0.08859, over 5664964.94 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:58:54,454 INFO [train.py:968] (0/2) Epoch 26, batch 34900, giga_loss[loss=0.3274, simple_loss=0.3915, pruned_loss=0.1316, over 26743.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3468, pruned_loss=0.09536, over 5668346.56 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3393, pruned_loss=0.1066, over 5682201.44 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3471, pruned_loss=0.09354, over 5666518.29 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:59:04,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3164, 3.2288, 1.4236, 1.5414], device='cuda:0'), covar=tensor([0.1057, 0.0322, 0.0969, 0.1365], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0563, 0.0404, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 15:59:14,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.122e+03 1.502e+03 1.868e+03 2.931e+03 8.112e+03, threshold=3.736e+03, percent-clipped=10.0
+2023-03-13 15:59:31,829 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1174000.pt
+2023-03-13 15:59:37,623 INFO [train.py:968] (0/2) Epoch 26, batch 34950, giga_loss[loss=0.2313, simple_loss=0.3176, pruned_loss=0.07255, over 28414.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3494, pruned_loss=0.09734, over 5664361.99 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3393, pruned_loss=0.1067, over 5680320.88 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.35, pruned_loss=0.09551, over 5663419.09 frames. ], batch size: 65, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:00:07,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:00:19,327 INFO [train.py:968] (0/2) Epoch 26, batch 35000, giga_loss[loss=0.2781, simple_loss=0.3542, pruned_loss=0.101, over 28912.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3457, pruned_loss=0.09627, over 5677134.49 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3392, pruned_loss=0.1066, over 5683208.00 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3464, pruned_loss=0.09469, over 5673695.23 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:00:32,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6751, 4.5216, 4.2541, 1.8338], device='cuda:0'), covar=tensor([0.0488, 0.0667, 0.0691, 0.2131], device='cuda:0'), in_proj_covar=tensor([0.1277, 0.1169, 0.0987, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 16:00:38,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.087e+02 1.284e+03 1.544e+03 2.026e+03 3.746e+03, threshold=3.089e+03, percent-clipped=1.0
+2023-03-13 16:00:49,512 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3881, 3.1285, 1.5445, 1.5301], device='cuda:0'), covar=tensor([0.0986, 0.0354, 0.0922, 0.1305], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0563, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 16:00:50,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7411, 1.9568, 1.2306, 1.7248], device='cuda:0'), covar=tensor([0.1005, 0.0613, 0.1219, 0.1073], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0445, 0.0520, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 16:00:55,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:00:57,565 INFO [train.py:968] (0/2) Epoch 26, batch 35050, giga_loss[loss=0.2704, simple_loss=0.3454, pruned_loss=0.09773, over 27681.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.34, pruned_loss=0.09419, over 5681202.64 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3397, pruned_loss=0.1069, over 5679671.11 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3402, pruned_loss=0.0922, over 5681048.36 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:01:40,148 INFO [train.py:968] (0/2) Epoch 26, batch 35100, giga_loss[loss=0.2257, simple_loss=0.3046, pruned_loss=0.07341, over 28646.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3334, pruned_loss=0.09148, over 5683336.81 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3396, pruned_loss=0.1067, over 5685788.69 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3336, pruned_loss=0.08969, over 5677752.28 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:02:01,621 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.156e+02 1.212e+03 1.581e+03 2.133e+03 7.006e+03, threshold=3.161e+03, percent-clipped=10.0
+2023-03-13 16:02:04,057 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:06,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:15,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:24,186 INFO [train.py:968] (0/2) Epoch 26, batch 35150, giga_loss[loss=0.2035, simple_loss=0.2765, pruned_loss=0.06526, over 28863.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3253, pruned_loss=0.08744, over 5688342.58 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3396, pruned_loss=0.1065, over 5688840.33 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3253, pruned_loss=0.08591, over 5681198.10 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:02:30,174 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:32,090 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:53,572 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:57,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:03:02,926 INFO [train.py:968] (0/2) Epoch 26, batch 35200, giga_loss[loss=0.2, simple_loss=0.2823, pruned_loss=0.05883, over 28932.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3201, pruned_loss=0.08576, over 5680749.32 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.34, pruned_loss=0.1067, over 5682096.44 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3194, pruned_loss=0.08393, over 5681080.68 frames. ], batch size: 227, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:03:22,144 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:03:23,242 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.412e+02 1.198e+03 1.461e+03 2.111e+03 5.759e+03, threshold=2.922e+03, percent-clipped=7.0
+2023-03-13 16:03:46,237 INFO [train.py:968] (0/2) Epoch 26, batch 35250, giga_loss[loss=0.2581, simple_loss=0.329, pruned_loss=0.09362, over 28858.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3162, pruned_loss=0.08418, over 5671687.86 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3407, pruned_loss=0.1071, over 5673994.61 frames. ], giga_tot_loss[loss=0.2394, simple_loss=0.3148, pruned_loss=0.08205, over 5679312.05 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:04:00,187 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:04:16,180 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:04:18,089 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:04:30,402 INFO [train.py:968] (0/2) Epoch 26, batch 35300, giga_loss[loss=0.2037, simple_loss=0.28, pruned_loss=0.06372, over 28997.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3132, pruned_loss=0.08252, over 5683283.73 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.341, pruned_loss=0.1071, over 5678345.40 frames. ], giga_tot_loss[loss=0.2359, simple_loss=0.3112, pruned_loss=0.08033, over 5685336.46 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:04:42,460 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:04:48,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5346, 1.8117, 1.5021, 1.2883], device='cuda:0'), covar=tensor([0.2791, 0.2926, 0.3342, 0.2604], device='cuda:0'), in_proj_covar=tensor([0.1578, 0.1134, 0.1395, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 16:04:49,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.394e+02 1.123e+03 1.418e+03 1.986e+03 5.749e+03, threshold=2.835e+03, percent-clipped=9.0
+2023-03-13 16:04:56,690 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:05:07,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5334, 3.6685, 1.6683, 1.6732], device='cuda:0'), covar=tensor([0.1025, 0.0307, 0.0903, 0.1377], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0564, 0.0404, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 16:05:12,118 INFO [train.py:968] (0/2) Epoch 26, batch 35350, giga_loss[loss=0.2336, simple_loss=0.3016, pruned_loss=0.08283, over 27723.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3098, pruned_loss=0.08085, over 5685585.43 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3409, pruned_loss=0.107, over 5672478.77 frames. ], giga_tot_loss[loss=0.2327, simple_loss=0.3078, pruned_loss=0.07877, over 5692901.70 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:05:56,448 INFO [train.py:968] (0/2) Epoch 26, batch 35400, giga_loss[loss=0.2023, simple_loss=0.2807, pruned_loss=0.06189, over 28568.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3082, pruned_loss=0.08019, over 5678973.64 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3416, pruned_loss=0.1074, over 5656547.72 frames. ], giga_tot_loss[loss=0.23, simple_loss=0.3051, pruned_loss=0.07747, over 5699789.61 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:06:15,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.444e+02 1.132e+03 1.395e+03 1.862e+03 6.413e+03, threshold=2.789e+03, percent-clipped=9.0
+2023-03-13 16:06:27,878 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 16:06:36,842 INFO [train.py:968] (0/2) Epoch 26, batch 35450, giga_loss[loss=0.2137, simple_loss=0.2844, pruned_loss=0.07149, over 28571.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3051, pruned_loss=0.07862, over 5687713.69 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3417, pruned_loss=0.1072, over 5662425.34 frames. ], giga_tot_loss[loss=0.227, simple_loss=0.3018, pruned_loss=0.07604, over 5699664.08 frames. ], batch size: 60, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:06:59,979 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4149, 4.2678, 4.0338, 1.9637], device='cuda:0'), covar=tensor([0.0556, 0.0718, 0.0734, 0.2067], device='cuda:0'), in_proj_covar=tensor([0.1278, 0.1173, 0.0987, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 16:07:02,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:07:19,116 INFO [train.py:968] (0/2) Epoch 26, batch 35500, giga_loss[loss=0.2088, simple_loss=0.2774, pruned_loss=0.07013, over 28578.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3021, pruned_loss=0.07735, over 5678897.34 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3424, pruned_loss=0.1076, over 5657360.73 frames. ], giga_tot_loss[loss=0.2236, simple_loss=0.2984, pruned_loss=0.07446, over 5693727.18 frames. ], batch size: 60, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:07:39,669 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.969e+02 1.143e+03 1.498e+03 1.834e+03 5.697e+03, threshold=2.995e+03, percent-clipped=9.0
+2023-03-13 16:07:52,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:08:01,130 INFO [train.py:968] (0/2) Epoch 26, batch 35550, giga_loss[loss=0.2108, simple_loss=0.2951, pruned_loss=0.06323, over 28786.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.2992, pruned_loss=0.07586, over 5681049.74 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3422, pruned_loss=0.1073, over 5662542.27 frames. ], giga_tot_loss[loss=0.221, simple_loss=0.2956, pruned_loss=0.07325, over 5688618.79 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:08:11,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:08:12,771 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:08:47,407 INFO [train.py:968] (0/2) Epoch 26, batch 35600, giga_loss[loss=0.1951, simple_loss=0.2748, pruned_loss=0.0577, over 28879.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2967, pruned_loss=0.07499, over 5692345.64 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3423, pruned_loss=0.1072, over 5667285.81 frames. ], giga_tot_loss[loss=0.2186, simple_loss=0.2927, pruned_loss=0.07227, over 5694527.84 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:09:06,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.971e+02 1.107e+03 1.629e+03 2.088e+03 8.290e+03, threshold=3.257e+03, percent-clipped=10.0
+2023-03-13 16:09:21,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:09:27,645 INFO [train.py:968] (0/2) Epoch 26, batch 35650, giga_loss[loss=0.3569, simple_loss=0.3858, pruned_loss=0.164, over 26711.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.2973, pruned_loss=0.0759, over 5683979.82 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3427, pruned_loss=0.1073, over 5663994.18 frames. ], giga_tot_loss[loss=0.2188, simple_loss=0.2923, pruned_loss=0.07269, over 5688993.69 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:09:39,935 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 16:09:56,974 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:10:00,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:10:10,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2324, 3.0652, 2.9218, 1.4635], device='cuda:0'), covar=tensor([0.1032, 0.1120, 0.0939, 0.2455], device='cuda:0'), in_proj_covar=tensor([0.1272, 0.1172, 0.0986, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 16:10:15,748 INFO [train.py:968] (0/2) Epoch 26, batch 35700, giga_loss[loss=0.2904, simple_loss=0.3664, pruned_loss=0.1072, over 29050.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3072, pruned_loss=0.081, over 5682354.60 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3427, pruned_loss=0.1072, over 5665073.21 frames. ], giga_tot_loss[loss=0.2299, simple_loss=0.3031, pruned_loss=0.07838, over 5685428.87 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:10:23,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:10:26,743 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:10:41,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.461e+02 1.340e+03 1.841e+03 2.423e+03 5.565e+03, threshold=3.682e+03, percent-clipped=7.0
+2023-03-13 16:10:51,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3850, 1.8050, 1.3461, 0.7607], device='cuda:0'), covar=tensor([0.4470, 0.2421, 0.3714, 0.5828], device='cuda:0'), in_proj_covar=tensor([0.1813, 0.1714, 0.1646, 0.1482], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 16:11:03,584 INFO [train.py:968] (0/2) Epoch 26, batch 35750, giga_loss[loss=0.3152, simple_loss=0.382, pruned_loss=0.1242, over 27918.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3208, pruned_loss=0.08782, over 5674743.52 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3434, pruned_loss=0.1076, over 5656435.85 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3166, pruned_loss=0.08513, over 5685078.49 frames. ], batch size: 412, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:11:06,630 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:11:15,684 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174822.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:11:27,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3182, 2.7133, 1.3997, 1.4235], device='cuda:0'), covar=tensor([0.1004, 0.0369, 0.0940, 0.1378], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0562, 0.0403, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 16:11:33,005 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:11:34,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:11:46,974 INFO [train.py:968] (0/2) Epoch 26, batch 35800, giga_loss[loss=0.3003, simple_loss=0.3728, pruned_loss=0.1139, over 28559.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3323, pruned_loss=0.09334, over 5680733.74 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3439, pruned_loss=0.1079, over 5658722.85 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3284, pruned_loss=0.09081, over 5687096.19 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:12:02,436 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:12:09,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.349e+03 1.722e+03 2.516e+03 6.265e+03, threshold=3.444e+03, percent-clipped=10.0
+2023-03-13 16:12:29,095 INFO [train.py:968] (0/2) Epoch 26, batch 35850, giga_loss[loss=0.2638, simple_loss=0.3392, pruned_loss=0.09417, over 28821.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3366, pruned_loss=0.09415, over 5680326.86 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3439, pruned_loss=0.1077, over 5654179.97 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3333, pruned_loss=0.09188, over 5690161.26 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:12:30,108 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:12:32,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:12:32,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:12:54,556 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:13:14,996 INFO [train.py:968] (0/2) Epoch 26, batch 35900, giga_loss[loss=0.2641, simple_loss=0.3472, pruned_loss=0.09043, over 28542.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3397, pruned_loss=0.09493, over 5685513.93 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3443, pruned_loss=0.1081, over 5660690.23 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3366, pruned_loss=0.09252, over 5688315.57 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:13:35,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.419e+02 1.417e+03 1.910e+03 2.824e+03 9.629e+03, threshold=3.819e+03, percent-clipped=18.0
+2023-03-13 16:13:45,055 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:13:46,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:13:56,637 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3620, 1.4706, 1.5564, 1.2034], device='cuda:0'), covar=tensor([0.1737, 0.2643, 0.1434, 0.1755], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0708, 0.0977, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 16:13:59,063 INFO [train.py:968] (0/2) Epoch 26, batch 35950, giga_loss[loss=0.2683, simple_loss=0.3567, pruned_loss=0.09, over 29012.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3425, pruned_loss=0.09609, over 5679497.69 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.345, pruned_loss=0.1085, over 5658195.38 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3392, pruned_loss=0.09344, over 5684371.52 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:14:18,700 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-13 16:14:19,940 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1175033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:14:32,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3854, 3.2009, 1.5060, 1.5402], device='cuda:0'), covar=tensor([0.1032, 0.0307, 0.0916, 0.1337], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0561, 0.0403, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 16:14:37,859 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:14:38,808 INFO [train.py:968] (0/2) Epoch 26, batch 36000, giga_loss[loss=0.2717, simple_loss=0.3438, pruned_loss=0.09984, over 28946.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3453, pruned_loss=0.0983, over 5682684.61 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3459, pruned_loss=0.1091, over 5660793.80 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3419, pruned_loss=0.09525, over 5685032.15 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:14:38,813 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 16:14:42,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5133, 1.8633, 1.4872, 1.4469], device='cuda:0'), covar=tensor([0.3242, 0.3138, 0.3687, 0.2627], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1137, 0.1397, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 16:14:47,089 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2007, simple_loss=0.3075, pruned_loss=0.047, over 944034.00 frames. 
+2023-03-13 16:14:47,090 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 16:14:48,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:15:09,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.993e+02 1.348e+03 1.829e+03 2.460e+03 5.201e+03, threshold=3.658e+03, percent-clipped=7.0
+2023-03-13 16:15:11,803 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:15:28,684 INFO [train.py:968] (0/2) Epoch 26, batch 36050, giga_loss[loss=0.2869, simple_loss=0.3554, pruned_loss=0.1092, over 28626.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3491, pruned_loss=0.101, over 5680632.93 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3463, pruned_loss=0.1093, over 5666820.91 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3459, pruned_loss=0.09818, over 5677425.62 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:15:52,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:15:52,969 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:15:54,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:15:55,624 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:16:08,298 INFO [train.py:968] (0/2) Epoch 26, batch 36100, giga_loss[loss=0.2691, simple_loss=0.3498, pruned_loss=0.09416, over 28949.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3515, pruned_loss=0.1022, over 5683747.43 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3467, pruned_loss=0.1096, over 5660212.83 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3487, pruned_loss=0.09944, over 5686817.21 frames. ], batch size: 213, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:16:19,022 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:16:19,953 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:16:31,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.806e+02 1.322e+03 1.728e+03 2.423e+03 7.344e+03, threshold=3.455e+03, percent-clipped=10.0
+2023-03-13 16:16:33,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1175186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:16:34,512 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-13 16:16:41,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1175197.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:16:48,143 INFO [train.py:968] (0/2) Epoch 26, batch 36150, giga_loss[loss=0.2692, simple_loss=0.3523, pruned_loss=0.09302, over 28180.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3549, pruned_loss=0.1034, over 5689583.61 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3468, pruned_loss=0.1096, over 5663900.04 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3527, pruned_loss=0.1011, over 5688965.48 frames. ], batch size: 77, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:16:56,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1583, 1.3347, 1.1941, 1.0499], device='cuda:0'), covar=tensor([0.2692, 0.2971, 0.2219, 0.2809], device='cuda:0'), in_proj_covar=tensor([0.2020, 0.1961, 0.1868, 0.2020], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 16:17:11,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4574, 1.6408, 1.6747, 1.4344], device='cuda:0'), covar=tensor([0.2187, 0.2408, 0.2333, 0.2348], device='cuda:0'), in_proj_covar=tensor([0.0485, 0.0744, 0.0716, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 16:17:31,477 INFO [train.py:968] (0/2) Epoch 26, batch 36200, giga_loss[loss=0.2972, simple_loss=0.373, pruned_loss=0.1108, over 28904.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3567, pruned_loss=0.1039, over 5685846.02 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3475, pruned_loss=0.1098, over 5668580.77 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3545, pruned_loss=0.1018, over 5681555.40 frames. ], batch size: 227, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:17:43,406 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-13 16:17:53,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.735e+02 1.351e+03 1.650e+03 2.277e+03 4.605e+03, threshold=3.300e+03, percent-clipped=7.0
+2023-03-13 16:18:10,862 INFO [train.py:968] (0/2) Epoch 26, batch 36250, giga_loss[loss=0.2634, simple_loss=0.3451, pruned_loss=0.09084, over 28801.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3574, pruned_loss=0.103, over 5689472.60 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3485, pruned_loss=0.1102, over 5670879.35 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3551, pruned_loss=0.1008, over 5684373.67 frames. ], batch size: 119, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:18:15,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2058, 1.3864, 1.3082, 1.1195], device='cuda:0'), covar=tensor([0.3069, 0.3377, 0.2309, 0.2782], device='cuda:0'), in_proj_covar=tensor([0.2021, 0.1964, 0.1870, 0.2020], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 16:18:28,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:18:31,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:18:38,208 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175340.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:18:40,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175343.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:18:51,490 INFO [train.py:968] (0/2) Epoch 26, batch 36300, giga_loss[loss=0.2497, simple_loss=0.3395, pruned_loss=0.07995, over 28511.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3565, pruned_loss=0.1012, over 5693522.69 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3487, pruned_loss=0.1102, over 5667285.33 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3546, pruned_loss=0.09924, over 5692471.42 frames. ], batch size: 60, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:18:54,061 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:18:55,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3106, 1.4443, 1.3372, 1.1951], device='cuda:0'), covar=tensor([0.3340, 0.3248, 0.2650, 0.3146], device='cuda:0'), in_proj_covar=tensor([0.2024, 0.1968, 0.1873, 0.2023], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 16:19:01,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175372.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:19:10,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.745e+02 1.286e+03 1.491e+03 2.114e+03 6.046e+03, threshold=2.982e+03, percent-clipped=7.0
+2023-03-13 16:19:29,582 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1175406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:19:30,062 INFO [train.py:968] (0/2) Epoch 26, batch 36350, libri_loss[loss=0.2876, simple_loss=0.3642, pruned_loss=0.1055, over 29544.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3536, pruned_loss=0.09833, over 5702075.57 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3484, pruned_loss=0.1099, over 5672913.07 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3526, pruned_loss=0.09673, over 5697004.95 frames. ], batch size: 89, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:19:30,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1175408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:20:10,664 INFO [train.py:968] (0/2) Epoch 26, batch 36400, giga_loss[loss=0.2554, simple_loss=0.3421, pruned_loss=0.08435, over 28968.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3516, pruned_loss=0.09667, over 5702608.76 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3481, pruned_loss=0.1097, over 5668095.55 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3511, pruned_loss=0.09539, over 5704206.01 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:20:32,915 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.640e+02 1.260e+03 1.578e+03 2.032e+03 7.743e+03, threshold=3.157e+03, percent-clipped=12.0
+2023-03-13 16:20:42,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5144, 1.3990, 4.3396, 3.6196], device='cuda:0'), covar=tensor([0.2060, 0.3040, 0.0846, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0665, 0.0986, 0.0949], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 16:20:56,073 INFO [train.py:968] (0/2) Epoch 26, batch 36450, giga_loss[loss=0.3005, simple_loss=0.3635, pruned_loss=0.1187, over 28679.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3544, pruned_loss=0.1005, over 5704845.12 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3483, pruned_loss=0.1097, over 5674332.23 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.354, pruned_loss=0.09918, over 5701290.34 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:21:33,531 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:21:37,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:21:38,973 INFO [train.py:968] (0/2) Epoch 26, batch 36500, giga_loss[loss=0.2694, simple_loss=0.3448, pruned_loss=0.09702, over 28572.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3573, pruned_loss=0.1048, over 5702568.46 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3486, pruned_loss=0.1098, over 5679063.97 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3567, pruned_loss=0.1035, over 5696062.83 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:21:59,279 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:22:00,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.569e+02 1.454e+03 1.765e+03 2.459e+03 1.100e+04, threshold=3.530e+03, percent-clipped=10.0
+2023-03-13 16:22:22,089 INFO [train.py:968] (0/2) Epoch 26, batch 36550, giga_loss[loss=0.2354, simple_loss=0.3157, pruned_loss=0.07758, over 28748.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3567, pruned_loss=0.1057, over 5700196.11 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3486, pruned_loss=0.1097, over 5678721.58 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3563, pruned_loss=0.1047, over 5695823.32 frames. ], batch size: 66, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:22:38,903 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-13 16:23:04,251 INFO [train.py:968] (0/2) Epoch 26, batch 36600, giga_loss[loss=0.2837, simple_loss=0.3523, pruned_loss=0.1075, over 28226.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3546, pruned_loss=0.1051, over 5701522.55 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.349, pruned_loss=0.1098, over 5678762.74 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3542, pruned_loss=0.1042, over 5698394.83 frames. ], batch size: 77, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:23:26,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.431e+03 1.855e+03 2.340e+03 5.167e+03, threshold=3.710e+03, percent-clipped=6.0
+2023-03-13 16:23:46,282 INFO [train.py:968] (0/2) Epoch 26, batch 36650, giga_loss[loss=0.2592, simple_loss=0.3253, pruned_loss=0.0965, over 23582.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3539, pruned_loss=0.105, over 5699138.44 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3492, pruned_loss=0.11, over 5679838.06 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3534, pruned_loss=0.104, over 5695968.70 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:23:46,552 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3495, 1.5355, 1.4830, 1.2466], device='cuda:0'), covar=tensor([0.3404, 0.3092, 0.2278, 0.2989], device='cuda:0'), in_proj_covar=tensor([0.2034, 0.1974, 0.1885, 0.2037], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 16:24:02,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2071, 1.4387, 1.5565, 1.2611], device='cuda:0'), covar=tensor([0.2628, 0.2329, 0.1430, 0.1994], device='cuda:0'), in_proj_covar=tensor([0.2034, 0.1973, 0.1884, 0.2036], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 16:24:27,754 INFO [train.py:968] (0/2) Epoch 26, batch 36700, giga_loss[loss=0.2524, simple_loss=0.3389, pruned_loss=0.08296, over 28306.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3524, pruned_loss=0.1035, over 5704720.09 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3495, pruned_loss=0.1103, over 5684109.69 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3519, pruned_loss=0.1023, over 5698820.10 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:24:53,688 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1175781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:24:56,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.674e+02 1.252e+03 1.553e+03 1.937e+03 7.742e+03, threshold=3.107e+03, percent-clipped=5.0
+2023-03-13 16:24:58,321 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-13 16:25:15,582 INFO [train.py:968] (0/2) Epoch 26, batch 36750, giga_loss[loss=0.2722, simple_loss=0.343, pruned_loss=0.1007, over 28952.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3489, pruned_loss=0.1006, over 5689870.51 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3493, pruned_loss=0.1101, over 5682922.14 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3487, pruned_loss=0.09984, over 5686401.68 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:26:04,420 INFO [train.py:968] (0/2) Epoch 26, batch 36800, giga_loss[loss=0.2172, simple_loss=0.2983, pruned_loss=0.06811, over 28958.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3434, pruned_loss=0.0975, over 5678214.38 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3496, pruned_loss=0.1102, over 5686970.42 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3429, pruned_loss=0.09663, over 5672035.49 frames. ], batch size: 213, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:26:30,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.920e+02 1.154e+03 1.519e+03 1.990e+03 5.903e+03, threshold=3.038e+03, percent-clipped=9.0
+2023-03-13 16:26:51,494 INFO [train.py:968] (0/2) Epoch 26, batch 36850, giga_loss[loss=0.226, simple_loss=0.2896, pruned_loss=0.08123, over 23433.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.338, pruned_loss=0.09523, over 5647859.56 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3502, pruned_loss=0.1106, over 5667476.02 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3369, pruned_loss=0.0939, over 5659989.99 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:27:08,917 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:27:11,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:27:40,524 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:27:40,924 INFO [train.py:968] (0/2) Epoch 26, batch 36900, giga_loss[loss=0.2364, simple_loss=0.3126, pruned_loss=0.08009, over 28844.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3337, pruned_loss=0.0933, over 5644732.15 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3504, pruned_loss=0.1105, over 5672431.64 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3323, pruned_loss=0.09195, over 5649702.55 frames. ], batch size: 199, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:28:07,461 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.836e+02 1.071e+03 1.441e+03 1.904e+03 9.116e+03, threshold=2.883e+03, percent-clipped=8.0
+2023-03-13 16:28:19,904 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1176000.pt
+2023-03-13 16:28:26,548 INFO [train.py:968] (0/2) Epoch 26, batch 36950, giga_loss[loss=0.219, simple_loss=0.3045, pruned_loss=0.06671, over 28908.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3335, pruned_loss=0.09235, over 5643599.22 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3507, pruned_loss=0.1106, over 5655158.80 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3317, pruned_loss=0.09081, over 5661811.15 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:29:07,409 INFO [train.py:968] (0/2) Epoch 26, batch 37000, libri_loss[loss=0.3008, simple_loss=0.3619, pruned_loss=0.1199, over 29660.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3344, pruned_loss=0.0925, over 5652040.79 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3514, pruned_loss=0.1111, over 5658150.80 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.332, pruned_loss=0.09039, over 5663639.88 frames. ], batch size: 73, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:29:29,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.560e+02 1.143e+03 1.390e+03 1.826e+03 5.594e+03, threshold=2.779e+03, percent-clipped=8.0
+2023-03-13 16:29:49,363 INFO [train.py:968] (0/2) Epoch 26, batch 37050, giga_loss[loss=0.2445, simple_loss=0.3276, pruned_loss=0.08068, over 28737.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3343, pruned_loss=0.09221, over 5671527.62 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3521, pruned_loss=0.1113, over 5660035.08 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3313, pruned_loss=0.08986, over 5678943.93 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:30:29,571 INFO [train.py:968] (0/2) Epoch 26, batch 37100, giga_loss[loss=0.2429, simple_loss=0.323, pruned_loss=0.08137, over 28848.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3321, pruned_loss=0.09116, over 5683296.72 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3528, pruned_loss=0.1113, over 5656680.71 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3287, pruned_loss=0.08876, over 5692259.56 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:30:53,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.512e+02 1.125e+03 1.300e+03 1.747e+03 5.929e+03, threshold=2.601e+03, percent-clipped=10.0
+2023-03-13 16:31:08,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5254, 1.7424, 1.7107, 1.5826], device='cuda:0'), covar=tensor([0.2322, 0.2328, 0.2684, 0.2479], device='cuda:0'), in_proj_covar=tensor([0.0493, 0.0754, 0.0725, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 16:31:11,571 INFO [train.py:968] (0/2) Epoch 26, batch 37150, giga_loss[loss=0.2621, simple_loss=0.331, pruned_loss=0.09662, over 23974.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3298, pruned_loss=0.09008, over 5690676.23 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3527, pruned_loss=0.1111, over 5660559.19 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3269, pruned_loss=0.08812, over 5694683.92 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:31:36,318 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-13 16:31:45,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3229, 1.1304, 3.9657, 3.3125], device='cuda:0'), covar=tensor([0.1490, 0.2677, 0.0432, 0.1030], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0664, 0.0985, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 16:31:50,277 INFO [train.py:968] (0/2) Epoch 26, batch 37200, libri_loss[loss=0.2431, simple_loss=0.3189, pruned_loss=0.08364, over 29652.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3277, pruned_loss=0.08929, over 5699275.02 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.353, pruned_loss=0.1111, over 5662817.96 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3246, pruned_loss=0.08732, over 5700779.08 frames. ], batch size: 69, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:32:12,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.764e+02 1.171e+03 1.389e+03 1.759e+03 8.881e+03, threshold=2.778e+03, percent-clipped=10.0
+2023-03-13 16:32:30,469 INFO [train.py:968] (0/2) Epoch 26, batch 37250, giga_loss[loss=0.2649, simple_loss=0.3321, pruned_loss=0.09882, over 28866.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3258, pruned_loss=0.08872, over 5707024.99 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3529, pruned_loss=0.1109, over 5665550.74 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3231, pruned_loss=0.08707, over 5706154.17 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:33:09,567 INFO [train.py:968] (0/2) Epoch 26, batch 37300, giga_loss[loss=0.2384, simple_loss=0.3066, pruned_loss=0.08513, over 28509.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3239, pruned_loss=0.08767, over 5712931.46 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3532, pruned_loss=0.1108, over 5668681.45 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3207, pruned_loss=0.0858, over 5710607.71 frames. ], batch size: 78, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:33:16,018 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-13 16:33:23,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:33:31,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.366e+02 1.105e+03 1.357e+03 1.760e+03 6.306e+03, threshold=2.714e+03, percent-clipped=5.0
+2023-03-13 16:33:47,870 INFO [train.py:968] (0/2) Epoch 26, batch 37350, giga_loss[loss=0.2457, simple_loss=0.3172, pruned_loss=0.08713, over 28089.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3219, pruned_loss=0.08638, over 5714152.43 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3539, pruned_loss=0.111, over 5670684.44 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3181, pruned_loss=0.08413, over 5711580.22 frames. ], batch size: 77, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:34:28,506 INFO [train.py:968] (0/2) Epoch 26, batch 37400, libri_loss[loss=0.3777, simple_loss=0.4319, pruned_loss=0.1618, over 28646.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3211, pruned_loss=0.08605, over 5718052.99 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3549, pruned_loss=0.1116, over 5674141.61 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3165, pruned_loss=0.08326, over 5713678.52 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:34:44,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4480, 4.2941, 4.0699, 1.8947], device='cuda:0'), covar=tensor([0.0560, 0.0694, 0.0634, 0.2048], device='cuda:0'), in_proj_covar=tensor([0.1264, 0.1167, 0.0983, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 16:34:53,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.192e+02 1.015e+03 1.198e+03 1.680e+03 7.416e+03, threshold=2.395e+03, percent-clipped=7.0
+2023-03-13 16:35:00,295 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-13 16:35:08,703 INFO [train.py:968] (0/2) Epoch 26, batch 37450, giga_loss[loss=0.2223, simple_loss=0.3003, pruned_loss=0.07215, over 28691.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3197, pruned_loss=0.08517, over 5716076.04 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3551, pruned_loss=0.1117, over 5676421.73 frames. ], giga_tot_loss[loss=0.2403, simple_loss=0.3154, pruned_loss=0.08263, over 5710851.72 frames. ], batch size: 242, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:35:51,425 INFO [train.py:968] (0/2) Epoch 26, batch 37500, giga_loss[loss=0.2341, simple_loss=0.3206, pruned_loss=0.07384, over 28887.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3203, pruned_loss=0.0857, over 5710256.69 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3556, pruned_loss=0.1117, over 5670500.00 frames. ], giga_tot_loss[loss=0.2411, simple_loss=0.3159, pruned_loss=0.08314, over 5712716.78 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:36:17,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.647e+02 1.272e+03 1.502e+03 2.129e+03 8.291e+03, threshold=3.003e+03, percent-clipped=20.0
+2023-03-13 16:36:27,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-13 16:36:33,224 INFO [train.py:968] (0/2) Epoch 26, batch 37550, giga_loss[loss=0.2396, simple_loss=0.3205, pruned_loss=0.07934, over 28922.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3247, pruned_loss=0.08847, over 5704492.41 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3563, pruned_loss=0.1121, over 5663385.94 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3199, pruned_loss=0.08561, over 5714026.96 frames. ], batch size: 213, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:37:17,351 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:37:17,704 INFO [train.py:968] (0/2) Epoch 26, batch 37600, giga_loss[loss=0.3068, simple_loss=0.3788, pruned_loss=0.1174, over 28587.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3312, pruned_loss=0.09217, over 5705290.80 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3571, pruned_loss=0.1123, over 5669618.70 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.326, pruned_loss=0.08919, over 5708192.41 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:37:25,042 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:37:47,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.229e+02 1.385e+03 1.867e+03 2.775e+03 7.925e+03, threshold=3.733e+03, percent-clipped=24.0
+2023-03-13 16:38:07,569 INFO [train.py:968] (0/2) Epoch 26, batch 37650, giga_loss[loss=0.3231, simple_loss=0.3855, pruned_loss=0.1303, over 28844.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3393, pruned_loss=0.09755, over 5699975.97 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3572, pruned_loss=0.1124, over 5673423.20 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3347, pruned_loss=0.0948, over 5699404.38 frames. ], batch size: 199, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:38:15,556 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:38:52,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1176751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:38:55,806 INFO [train.py:968] (0/2) Epoch 26, batch 37700, giga_loss[loss=0.3146, simple_loss=0.3819, pruned_loss=0.1236, over 28302.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.344, pruned_loss=0.0998, over 5671806.44 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3576, pruned_loss=0.1126, over 5657764.99 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3395, pruned_loss=0.09709, over 5686036.11 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:38:58,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9920, 1.3462, 1.1160, 0.2131], device='cuda:0'), covar=tensor([0.4678, 0.3648, 0.5203, 0.7414], device='cuda:0'), in_proj_covar=tensor([0.1812, 0.1709, 0.1646, 0.1481], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 16:39:09,441 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:39:19,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.039e+02 1.342e+03 1.845e+03 2.321e+03 8.550e+03, threshold=3.691e+03, percent-clipped=3.0
+2023-03-13 16:39:22,410 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:39:39,326 INFO [train.py:968] (0/2) Epoch 26, batch 37750, giga_loss[loss=0.2792, simple_loss=0.3535, pruned_loss=0.1024, over 28858.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3481, pruned_loss=0.1009, over 5677603.10 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3577, pruned_loss=0.1124, over 5660334.76 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3442, pruned_loss=0.09861, over 5686753.44 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:40:26,530 INFO [train.py:968] (0/2) Epoch 26, batch 37800, giga_loss[loss=0.2639, simple_loss=0.3397, pruned_loss=0.09403, over 28546.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3541, pruned_loss=0.1042, over 5673898.39 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3581, pruned_loss=0.1127, over 5660588.92 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3507, pruned_loss=0.1021, over 5680922.24 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:40:31,639 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-13 16:40:52,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.867e+02 1.235e+03 1.468e+03 1.961e+03 7.392e+03, threshold=2.937e+03, percent-clipped=2.0
+2023-03-13 16:40:54,297 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-13 16:40:57,203 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1176894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:40:59,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1176897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:41:04,196 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176904.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:41:05,795 INFO [train.py:968] (0/2) Epoch 26, batch 37850, giga_loss[loss=0.2496, simple_loss=0.3297, pruned_loss=0.08473, over 28693.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3538, pruned_loss=0.1034, over 5683166.19 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.358, pruned_loss=0.1126, over 5666795.64 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3509, pruned_loss=0.1016, over 5683535.63 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:41:21,427 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1176926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:41:46,685 INFO [train.py:968] (0/2) Epoch 26, batch 37900, giga_loss[loss=0.2188, simple_loss=0.3035, pruned_loss=0.06704, over 28596.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3495, pruned_loss=0.1002, over 5692920.95 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3574, pruned_loss=0.1123, over 5671580.40 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3477, pruned_loss=0.09879, over 5689443.66 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:41:55,453 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:42:11,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.503e+02 1.235e+03 1.585e+03 2.035e+03 6.112e+03, threshold=3.169e+03, percent-clipped=3.0
+2023-03-13 16:42:17,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3580, 1.7149, 1.4155, 1.5509], device='cuda:0'), covar=tensor([0.0730, 0.0324, 0.0328, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0114], device='cuda:0')
+2023-03-13 16:42:27,089 INFO [train.py:968] (0/2) Epoch 26, batch 37950, giga_loss[loss=0.2941, simple_loss=0.3656, pruned_loss=0.1113, over 28878.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3471, pruned_loss=0.09819, over 5698680.47 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3575, pruned_loss=0.1125, over 5677507.21 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3453, pruned_loss=0.09646, over 5691086.54 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:42:30,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6111, 1.5962, 1.7894, 1.3827], device='cuda:0'), covar=tensor([0.1796, 0.2521, 0.1427, 0.1725], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0710, 0.0975, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 16:42:48,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:42:54,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:43:10,786 INFO [train.py:968] (0/2) Epoch 26, batch 38000, giga_loss[loss=0.2639, simple_loss=0.3431, pruned_loss=0.09235, over 28961.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3467, pruned_loss=0.09761, over 5700359.47 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3576, pruned_loss=0.1125, over 5675971.35 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3451, pruned_loss=0.09614, over 5696248.31 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:43:24,845 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3662, 1.3820, 1.4784, 1.2487], device='cuda:0'), covar=tensor([0.3582, 0.3266, 0.2385, 0.3168], device='cuda:0'), in_proj_covar=tensor([0.2040, 0.1976, 0.1895, 0.2051], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 16:43:36,903 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1177087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:43:37,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.738e+02 1.480e+03 1.683e+03 2.301e+03 7.286e+03, threshold=3.366e+03, percent-clipped=10.0
+2023-03-13 16:43:40,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:43:54,396 INFO [train.py:968] (0/2) Epoch 26, batch 38050, giga_loss[loss=0.3249, simple_loss=0.397, pruned_loss=0.1264, over 28704.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3494, pruned_loss=0.09927, over 5697924.44 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3573, pruned_loss=0.1125, over 5676853.95 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3482, pruned_loss=0.09794, over 5694165.73 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:44:27,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:44:35,351 INFO [train.py:968] (0/2) Epoch 26, batch 38100, giga_loss[loss=0.2817, simple_loss=0.3449, pruned_loss=0.1092, over 23645.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3515, pruned_loss=0.101, over 5686392.71 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3575, pruned_loss=0.1124, over 5666249.33 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.35, pruned_loss=0.0994, over 5694878.90 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:44:41,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:44:50,363 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:44:55,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:44:57,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:45:00,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:45:05,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.958e+02 1.445e+03 1.873e+03 2.669e+03 1.123e+04, threshold=3.747e+03, percent-clipped=17.0
+2023-03-13 16:45:19,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:45:19,506 INFO [train.py:968] (0/2) Epoch 26, batch 38150, giga_loss[loss=0.2915, simple_loss=0.3575, pruned_loss=0.1128, over 29019.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3542, pruned_loss=0.103, over 5684400.14 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3582, pruned_loss=0.1128, over 5665139.41 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3524, pruned_loss=0.1013, over 5692543.48 frames. ], batch size: 113, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:45:24,680 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:45:35,211 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1177225.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:45:39,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4089, 1.6994, 1.3708, 1.3236], device='cuda:0'), covar=tensor([0.2654, 0.2698, 0.3015, 0.2383], device='cuda:0'), in_proj_covar=tensor([0.1579, 0.1139, 0.1392, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 16:45:41,789 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:45:44,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:02,975 INFO [train.py:968] (0/2) Epoch 26, batch 38200, giga_loss[loss=0.2592, simple_loss=0.3362, pruned_loss=0.09107, over 28636.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3551, pruned_loss=0.104, over 5689548.95 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3583, pruned_loss=0.1126, over 5667386.21 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3535, pruned_loss=0.1026, over 5694618.80 frames. ], batch size: 78, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:46:11,539 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:23,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177279.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:46:31,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.584e+02 1.352e+03 1.692e+03 2.205e+03 5.773e+03, threshold=3.384e+03, percent-clipped=8.0
+2023-03-13 16:46:31,861 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:34,200 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:43,921 INFO [train.py:968] (0/2) Epoch 26, batch 38250, giga_loss[loss=0.3358, simple_loss=0.3876, pruned_loss=0.142, over 27573.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3562, pruned_loss=0.1051, over 5696134.86 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3588, pruned_loss=0.1129, over 5671569.05 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3544, pruned_loss=0.1036, over 5696818.62 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:46:46,949 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:49,849 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:57,077 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:47:12,962 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:47:15,189 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:47:26,868 INFO [train.py:968] (0/2) Epoch 26, batch 38300, giga_loss[loss=0.258, simple_loss=0.3412, pruned_loss=0.08742, over 28775.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3563, pruned_loss=0.1051, over 5687095.38 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3593, pruned_loss=0.1132, over 5666863.32 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3544, pruned_loss=0.1034, over 5692466.64 frames. ], batch size: 119, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:47:53,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.102e+02 1.305e+03 1.832e+03 2.785e+03 1.498e+04, threshold=3.663e+03, percent-clipped=11.0
+2023-03-13 16:48:05,985 INFO [train.py:968] (0/2) Epoch 26, batch 38350, libri_loss[loss=0.234, simple_loss=0.303, pruned_loss=0.08255, over 29681.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3554, pruned_loss=0.1034, over 5694577.31 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3592, pruned_loss=0.1132, over 5671363.03 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3539, pruned_loss=0.1019, over 5695102.35 frames. ], batch size: 69, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:48:18,276 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177422.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:48:21,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177425.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:48:43,796 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177454.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:48:45,988 INFO [train.py:968] (0/2) Epoch 26, batch 38400, giga_loss[loss=0.266, simple_loss=0.3434, pruned_loss=0.09432, over 28630.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3558, pruned_loss=0.1031, over 5698149.62 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3594, pruned_loss=0.1135, over 5666602.87 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3543, pruned_loss=0.1013, over 5702975.90 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:48:50,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:49:04,503 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3620, 1.7631, 1.7525, 1.5389], device='cuda:0'), covar=tensor([0.2001, 0.1634, 0.2224, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.0496, 0.0755, 0.0725, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 16:49:09,849 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:49:12,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:49:13,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.626e+02 1.267e+03 1.536e+03 2.046e+03 5.021e+03, threshold=3.073e+03, percent-clipped=3.0
+2023-03-13 16:49:27,751 INFO [train.py:968] (0/2) Epoch 26, batch 38450, giga_loss[loss=0.2695, simple_loss=0.3485, pruned_loss=0.09531, over 28884.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3542, pruned_loss=0.1023, over 5698932.56 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3596, pruned_loss=0.1137, over 5671044.72 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3527, pruned_loss=0.1005, over 5699380.43 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:49:35,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:50:02,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9729, 1.2397, 1.0476, 0.2323], device='cuda:0'), covar=tensor([0.4374, 0.2970, 0.3796, 0.6717], device='cuda:0'), in_proj_covar=tensor([0.1801, 0.1694, 0.1634, 0.1473], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 16:50:05,088 INFO [train.py:968] (0/2) Epoch 26, batch 38500, giga_loss[loss=0.2667, simple_loss=0.3457, pruned_loss=0.09386, over 28919.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3525, pruned_loss=0.1016, over 5688447.72 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3599, pruned_loss=0.114, over 5659284.00 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3509, pruned_loss=0.09973, over 5699499.92 frames. ], batch size: 199, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:50:32,118 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.091e+02 1.174e+03 1.487e+03 2.028e+03 5.542e+03, threshold=2.975e+03, percent-clipped=4.0
+2023-03-13 16:50:39,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177600.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:50:43,742 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:50:45,614 INFO [train.py:968] (0/2) Epoch 26, batch 38550, giga_loss[loss=0.2393, simple_loss=0.3176, pruned_loss=0.08045, over 28258.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3506, pruned_loss=0.1008, over 5695925.04 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3601, pruned_loss=0.114, over 5663835.01 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.349, pruned_loss=0.09893, over 5701236.67 frames. ], batch size: 77, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:50:46,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:51:07,369 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:51:08,292 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-13 16:51:24,076 INFO [train.py:968] (0/2) Epoch 26, batch 38600, libri_loss[loss=0.3864, simple_loss=0.4441, pruned_loss=0.1644, over 29641.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.351, pruned_loss=0.1016, over 5701153.17 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3602, pruned_loss=0.1141, over 5671319.47 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3493, pruned_loss=0.09962, over 5699505.41 frames. ], batch size: 91, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:51:43,886 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4108, 1.2353, 1.2111, 1.4888], device='cuda:0'), covar=tensor([0.0822, 0.0384, 0.0359, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 16:51:49,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.464e+02 1.212e+03 1.557e+03 1.994e+03 4.375e+03, threshold=3.114e+03, percent-clipped=13.0
+2023-03-13 16:52:03,921 INFO [train.py:968] (0/2) Epoch 26, batch 38650, giga_loss[loss=0.2701, simple_loss=0.3453, pruned_loss=0.09747, over 28392.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3504, pruned_loss=0.1011, over 5696455.37 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3602, pruned_loss=0.1142, over 5666014.15 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3488, pruned_loss=0.09926, over 5700596.85 frames. ], batch size: 65, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:52:07,798 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1177712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:52:32,993 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177743.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:52:34,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177746.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:52:42,294 INFO [train.py:968] (0/2) Epoch 26, batch 38700, giga_loss[loss=0.2529, simple_loss=0.3325, pruned_loss=0.08664, over 28422.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3505, pruned_loss=0.1005, over 5695546.11 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3607, pruned_loss=0.1143, over 5662393.81 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3485, pruned_loss=0.09848, over 5703341.48 frames. ], batch size: 77, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:52:57,107 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177775.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:53:07,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.123e+02 1.115e+03 1.404e+03 1.745e+03 7.059e+03, threshold=2.807e+03, percent-clipped=5.0
+2023-03-13 16:53:21,341 INFO [train.py:968] (0/2) Epoch 26, batch 38750, giga_loss[loss=0.291, simple_loss=0.3604, pruned_loss=0.1108, over 28884.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3506, pruned_loss=0.09984, over 5687094.90 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3609, pruned_loss=0.1144, over 5644781.72 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3489, pruned_loss=0.09807, over 5709071.25 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:53:31,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3631, 1.8351, 1.5227, 1.5845], device='cuda:0'), covar=tensor([0.0704, 0.0283, 0.0315, 0.0838], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 16:53:57,523 INFO [train.py:968] (0/2) Epoch 26, batch 38800, giga_loss[loss=0.2831, simple_loss=0.3506, pruned_loss=0.1077, over 28839.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3506, pruned_loss=0.09999, over 5696266.41 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3609, pruned_loss=0.1145, over 5649990.17 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3487, pruned_loss=0.09797, over 5711060.71 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:54:15,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9391, 1.2616, 1.1357, 0.2136], device='cuda:0'), covar=tensor([0.4747, 0.3433, 0.4942, 0.7427], device='cuda:0'), in_proj_covar=tensor([0.1800, 0.1693, 0.1632, 0.1474], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 16:54:24,500 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1177888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:54:26,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.139e+02 1.251e+03 1.565e+03 2.235e+03 7.200e+03, threshold=3.129e+03, percent-clipped=16.0
+2023-03-13 16:54:38,912 INFO [train.py:968] (0/2) Epoch 26, batch 38850, giga_loss[loss=0.2478, simple_loss=0.3264, pruned_loss=0.08462, over 28783.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3495, pruned_loss=0.1004, over 5691672.66 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3611, pruned_loss=0.1148, over 5654736.03 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3476, pruned_loss=0.0981, over 5700257.80 frames. ], batch size: 66, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:54:40,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1719, 1.2668, 3.5871, 3.1552], device='cuda:0'), covar=tensor([0.1705, 0.2748, 0.0472, 0.1064], device='cuda:0'), in_proj_covar=tensor([0.0789, 0.0665, 0.0984, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 16:55:18,204 INFO [train.py:968] (0/2) Epoch 26, batch 38900, giga_loss[loss=0.2305, simple_loss=0.308, pruned_loss=0.07646, over 28720.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3461, pruned_loss=0.09835, over 5696898.98 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.361, pruned_loss=0.1145, over 5659605.74 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3443, pruned_loss=0.09641, over 5700485.84 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:55:44,924 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.331e+02 1.184e+03 1.441e+03 1.774e+03 6.923e+03, threshold=2.882e+03, percent-clipped=5.0
+2023-03-13 16:55:52,074 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1178000.pt
+2023-03-13 16:55:59,398 INFO [train.py:968] (0/2) Epoch 26, batch 38950, giga_loss[loss=0.2317, simple_loss=0.3049, pruned_loss=0.07929, over 28367.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3418, pruned_loss=0.09603, over 5702712.50 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3606, pruned_loss=0.1143, over 5663256.30 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3405, pruned_loss=0.09447, over 5702884.74 frames. ], batch size: 77, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:56:05,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1942, 2.2806, 1.7143, 1.8772], device='cuda:0'), covar=tensor([0.1060, 0.0778, 0.1096, 0.1259], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0449, 0.0525, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 16:56:39,431 INFO [train.py:968] (0/2) Epoch 26, batch 39000, giga_loss[loss=0.3386, simple_loss=0.3956, pruned_loss=0.1408, over 28315.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3414, pruned_loss=0.09613, over 5703695.03 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3607, pruned_loss=0.1142, over 5667661.08 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3399, pruned_loss=0.09455, over 5700974.20 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:56:39,435 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 16:56:48,186 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2051, simple_loss=0.3132, pruned_loss=0.04853, over 944034.00 frames. 
+2023-03-13 16:56:48,186 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 16:57:13,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1178087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:57:16,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.537e+02 1.354e+03 1.652e+03 2.051e+03 6.326e+03, threshold=3.304e+03, percent-clipped=9.0
+2023-03-13 16:57:27,660 INFO [train.py:968] (0/2) Epoch 26, batch 39050, giga_loss[loss=0.2479, simple_loss=0.3283, pruned_loss=0.08377, over 28904.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3413, pruned_loss=0.09681, over 5705300.82 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3609, pruned_loss=0.1145, over 5663676.07 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3389, pruned_loss=0.09445, over 5708634.67 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:57:47,548 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-13 16:58:07,489 INFO [train.py:968] (0/2) Epoch 26, batch 39100, giga_loss[loss=0.2797, simple_loss=0.3464, pruned_loss=0.1065, over 28203.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3393, pruned_loss=0.09591, over 5704823.19 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3613, pruned_loss=0.1147, over 5666141.72 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3369, pruned_loss=0.09363, over 5705860.42 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:58:33,791 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.094e+02 1.229e+03 1.514e+03 2.110e+03 6.226e+03, threshold=3.028e+03, percent-clipped=6.0
+2023-03-13 16:58:47,264 INFO [train.py:968] (0/2) Epoch 26, batch 39150, giga_loss[loss=0.2231, simple_loss=0.2989, pruned_loss=0.07363, over 28719.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3372, pruned_loss=0.09508, over 5707646.46 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3615, pruned_loss=0.1148, over 5666240.45 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3348, pruned_loss=0.09285, over 5708866.63 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:59:00,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3213, 3.1454, 3.0123, 1.3267], device='cuda:0'), covar=tensor([0.0953, 0.1072, 0.0903, 0.2313], device='cuda:0'), in_proj_covar=tensor([0.1270, 0.1169, 0.0986, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 16:59:05,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1178230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:59:07,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1178233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:59:20,195 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-13 16:59:26,062 INFO [train.py:968] (0/2) Epoch 26, batch 39200, giga_loss[loss=0.2499, simple_loss=0.3269, pruned_loss=0.08644, over 28836.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3338, pruned_loss=0.09354, over 5709250.08 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3612, pruned_loss=0.1148, over 5668315.88 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3315, pruned_loss=0.09136, over 5709099.27 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:59:32,008 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1178262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:59:32,563 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1178263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:59:35,477 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-13 16:59:51,176 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-13 16:59:57,660 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.513e+02 1.196e+03 1.451e+03 2.173e+03 6.678e+03, threshold=2.902e+03, percent-clipped=10.0
+2023-03-13 17:00:10,396 INFO [train.py:968] (0/2) Epoch 26, batch 39250, giga_loss[loss=0.2567, simple_loss=0.3387, pruned_loss=0.08735, over 28784.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3325, pruned_loss=0.09266, over 5696224.37 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3613, pruned_loss=0.1149, over 5658565.86 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3303, pruned_loss=0.09068, over 5705032.95 frames. ], batch size: 243, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:00:51,886 INFO [train.py:968] (0/2) Epoch 26, batch 39300, giga_loss[loss=0.2631, simple_loss=0.3459, pruned_loss=0.09017, over 28829.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3352, pruned_loss=0.09385, over 5687029.78 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3615, pruned_loss=0.115, over 5645179.41 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3326, pruned_loss=0.09163, over 5706068.68 frames. ], batch size: 227, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:01:21,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.001e+02 1.162e+03 1.412e+03 2.172e+03 8.273e+03, threshold=2.823e+03, percent-clipped=11.0
+2023-03-13 17:01:36,472 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1178406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:01:36,863 INFO [train.py:968] (0/2) Epoch 26, batch 39350, giga_loss[loss=0.2376, simple_loss=0.3262, pruned_loss=0.07449, over 28922.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3376, pruned_loss=0.09418, over 5694626.76 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3618, pruned_loss=0.1152, over 5648767.11 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3349, pruned_loss=0.09186, over 5707153.30 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:01:39,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1178409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:01:53,867 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1178428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:02:00,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2921, 4.1222, 3.9261, 1.8329], device='cuda:0'), covar=tensor([0.0647, 0.0850, 0.0746, 0.2103], device='cuda:0'), in_proj_covar=tensor([0.1269, 0.1169, 0.0987, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 17:02:01,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1178438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:02:01,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1570, 1.6438, 1.1896, 0.5320], device='cuda:0'), covar=tensor([0.5093, 0.2453, 0.3493, 0.6903], device='cuda:0'), in_proj_covar=tensor([0.1799, 0.1694, 0.1635, 0.1472], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 17:02:16,258 INFO [train.py:968] (0/2) Epoch 26, batch 39400, giga_loss[loss=0.2571, simple_loss=0.3427, pruned_loss=0.08572, over 28680.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3417, pruned_loss=0.09631, over 5688564.82 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3624, pruned_loss=0.1157, over 5648509.57 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3381, pruned_loss=0.09327, over 5701013.73 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:02:27,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8041, 2.0997, 2.0304, 1.5852], device='cuda:0'), covar=tensor([0.1804, 0.2544, 0.1531, 0.1801], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0711, 0.0974, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 17:02:34,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4013, 2.0092, 1.4430, 0.6732], device='cuda:0'), covar=tensor([0.6075, 0.3221, 0.4886, 0.6922], device='cuda:0'), in_proj_covar=tensor([0.1796, 0.1690, 0.1633, 0.1470], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 17:02:45,896 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.634e+02 1.111e+03 1.481e+03 2.108e+03 5.788e+03, threshold=2.962e+03, percent-clipped=12.0
+2023-03-13 17:02:57,447 INFO [train.py:968] (0/2) Epoch 26, batch 39450, giga_loss[loss=0.2392, simple_loss=0.3177, pruned_loss=0.08041, over 28479.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3428, pruned_loss=0.09617, over 5693300.10 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3626, pruned_loss=0.1158, over 5655302.69 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3391, pruned_loss=0.09298, over 5698397.81 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:03:00,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3493, 2.0596, 1.4979, 0.6866], device='cuda:0'), covar=tensor([0.6479, 0.3042, 0.4342, 0.5877], device='cuda:0'), in_proj_covar=tensor([0.1798, 0.1693, 0.1635, 0.1473], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 17:03:38,753 INFO [train.py:968] (0/2) Epoch 26, batch 39500, giga_loss[loss=0.2886, simple_loss=0.3623, pruned_loss=0.1075, over 28725.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3421, pruned_loss=0.09547, over 5684546.37 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3621, pruned_loss=0.1155, over 5657599.29 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.339, pruned_loss=0.09266, over 5687558.26 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:03:55,992 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-13 17:04:06,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.480e+02 1.301e+03 1.651e+03 2.259e+03 9.306e+03, threshold=3.303e+03, percent-clipped=12.0
+2023-03-13 17:04:17,029 INFO [train.py:968] (0/2) Epoch 26, batch 39550, giga_loss[loss=0.2554, simple_loss=0.3373, pruned_loss=0.08681, over 28837.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3426, pruned_loss=0.09564, over 5684484.50 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3625, pruned_loss=0.1159, over 5647996.20 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3395, pruned_loss=0.09268, over 5696997.61 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:04:22,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7197, 1.7461, 1.9054, 1.4898], device='cuda:0'), covar=tensor([0.1908, 0.2477, 0.1563, 0.1725], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0710, 0.0974, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 17:04:58,957 INFO [train.py:968] (0/2) Epoch 26, batch 39600, libri_loss[loss=0.3127, simple_loss=0.3838, pruned_loss=0.1208, over 29250.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.344, pruned_loss=0.09689, over 5678391.57 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3628, pruned_loss=0.1159, over 5644534.53 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3405, pruned_loss=0.09384, over 5693246.97 frames. ], batch size: 97, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:05:27,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.812e+02 1.491e+03 1.810e+03 2.402e+03 6.741e+03, threshold=3.620e+03, percent-clipped=12.0
+2023-03-13 17:05:39,207 INFO [train.py:968] (0/2) Epoch 26, batch 39650, giga_loss[loss=0.3306, simple_loss=0.3973, pruned_loss=0.132, over 27588.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3459, pruned_loss=0.09828, over 5670930.08 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.363, pruned_loss=0.116, over 5639421.87 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3425, pruned_loss=0.09527, over 5688183.10 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:06:22,173 INFO [train.py:968] (0/2) Epoch 26, batch 39700, giga_loss[loss=0.2728, simple_loss=0.3464, pruned_loss=0.09962, over 28620.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3478, pruned_loss=0.09909, over 5686933.23 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3628, pruned_loss=0.1159, over 5641889.01 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3452, pruned_loss=0.09669, over 5698596.51 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:06:28,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9031, 1.1800, 1.3469, 1.0184], device='cuda:0'), covar=tensor([0.2248, 0.1658, 0.2514, 0.2035], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0757, 0.0728, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 17:06:49,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.282e+02 1.363e+03 1.671e+03 2.283e+03 4.550e+03, threshold=3.342e+03, percent-clipped=5.0
+2023-03-13 17:06:57,121 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1178803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:06:59,707 INFO [train.py:968] (0/2) Epoch 26, batch 39750, giga_loss[loss=0.2737, simple_loss=0.3477, pruned_loss=0.09987, over 28987.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.351, pruned_loss=0.1006, over 5702666.79 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3628, pruned_loss=0.1156, over 5653381.12 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3483, pruned_loss=0.09823, over 5703789.13 frames. ], batch size: 213, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:07:06,177 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-13 17:07:26,316 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1178843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:07:38,167 INFO [train.py:968] (0/2) Epoch 26, batch 39800, giga_loss[loss=0.277, simple_loss=0.3571, pruned_loss=0.09847, over 28925.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3527, pruned_loss=0.1015, over 5694635.22 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3634, pruned_loss=0.1161, over 5640166.04 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3497, pruned_loss=0.0987, over 5708371.04 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:08:10,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.680e+02 1.344e+03 1.592e+03 2.070e+03 8.064e+03, threshold=3.184e+03, percent-clipped=6.0
+2023-03-13 17:08:22,202 INFO [train.py:968] (0/2) Epoch 26, batch 39850, libri_loss[loss=0.3142, simple_loss=0.3864, pruned_loss=0.121, over 29484.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3532, pruned_loss=0.1017, over 5703355.84 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3636, pruned_loss=0.1161, over 5643647.55 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3505, pruned_loss=0.09932, over 5711774.90 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:08:48,444 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8271, 1.9814, 2.1067, 1.7183], device='cuda:0'), covar=tensor([0.3178, 0.2743, 0.2474, 0.3053], device='cuda:0'), in_proj_covar=tensor([0.2042, 0.1985, 0.1907, 0.2043], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 17:08:53,201 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1178946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:08:55,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1178949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:09:01,549 INFO [train.py:968] (0/2) Epoch 26, batch 39900, giga_loss[loss=0.2522, simple_loss=0.3412, pruned_loss=0.08163, over 29034.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3521, pruned_loss=0.1009, over 5700190.97 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.116, over 5637608.95 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3498, pruned_loss=0.0989, over 5713935.10 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:09:19,896 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1178978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:09:30,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.893e+02 1.303e+03 1.627e+03 2.080e+03 6.848e+03, threshold=3.254e+03, percent-clipped=4.0
+2023-03-13 17:09:41,021 INFO [train.py:968] (0/2) Epoch 26, batch 39950, giga_loss[loss=0.3305, simple_loss=0.3808, pruned_loss=0.1401, over 28726.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3505, pruned_loss=0.1001, over 5703424.70 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3635, pruned_loss=0.1158, over 5643552.77 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3483, pruned_loss=0.09824, over 5710683.53 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:09:43,093 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:10:10,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6329, 1.8586, 1.5521, 1.6318], device='cuda:0'), covar=tensor([0.2665, 0.2821, 0.3156, 0.2549], device='cuda:0'), in_proj_covar=tensor([0.1568, 0.1131, 0.1384, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 17:10:20,999 INFO [train.py:968] (0/2) Epoch 26, batch 40000, giga_loss[loss=0.207, simple_loss=0.2892, pruned_loss=0.06239, over 28861.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.347, pruned_loss=0.09839, over 5705590.44 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3629, pruned_loss=0.1154, over 5648726.02 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3455, pruned_loss=0.097, over 5707955.29 frames. ], batch size: 66, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:10:32,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0285, 1.2628, 1.2487, 1.0252], device='cuda:0'), covar=tensor([0.2108, 0.2368, 0.1430, 0.1890], device='cuda:0'), in_proj_covar=tensor([0.2043, 0.1988, 0.1910, 0.2046], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 17:10:50,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.341e+02 1.238e+03 1.462e+03 1.890e+03 7.468e+03, threshold=2.924e+03, percent-clipped=7.0
+2023-03-13 17:11:01,573 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:11:03,561 INFO [train.py:968] (0/2) Epoch 26, batch 40050, giga_loss[loss=0.247, simple_loss=0.3185, pruned_loss=0.08775, over 28759.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3434, pruned_loss=0.09648, over 5704380.14 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3632, pruned_loss=0.1156, over 5644449.27 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3416, pruned_loss=0.0948, over 5711019.92 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:11:39,916 INFO [train.py:968] (0/2) Epoch 26, batch 40100, giga_loss[loss=0.4364, simple_loss=0.4601, pruned_loss=0.2064, over 26663.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3427, pruned_loss=0.09557, over 5703598.93 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3626, pruned_loss=0.1152, over 5646108.03 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3411, pruned_loss=0.094, over 5709416.73 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:12:15,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.462e+02 1.192e+03 1.444e+03 1.875e+03 4.244e+03, threshold=2.887e+03, percent-clipped=5.0
+2023-03-13 17:12:26,638 INFO [train.py:968] (0/2) Epoch 26, batch 40150, giga_loss[loss=0.2812, simple_loss=0.3535, pruned_loss=0.1044, over 24007.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3444, pruned_loss=0.09546, over 5700787.99 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.363, pruned_loss=0.1154, over 5652665.77 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3423, pruned_loss=0.09356, over 5700823.65 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:12:34,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:13:07,422 INFO [train.py:968] (0/2) Epoch 26, batch 40200, giga_loss[loss=0.2666, simple_loss=0.3479, pruned_loss=0.09266, over 28858.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3458, pruned_loss=0.0963, over 5708960.67 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3632, pruned_loss=0.1155, over 5658947.12 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3436, pruned_loss=0.09429, over 5704491.16 frames. ], batch size: 227, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:13:11,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9883, 1.1978, 2.8314, 2.7944], device='cuda:0'), covar=tensor([0.1577, 0.2584, 0.0606, 0.1130], device='cuda:0'), in_proj_covar=tensor([0.0788, 0.0661, 0.0984, 0.0951], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 17:13:38,108 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.342e+02 1.284e+03 1.578e+03 2.249e+03 7.777e+03, threshold=3.156e+03, percent-clipped=10.0
+2023-03-13 17:13:48,576 INFO [train.py:968] (0/2) Epoch 26, batch 40250, giga_loss[loss=0.2454, simple_loss=0.3365, pruned_loss=0.07717, over 28896.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3438, pruned_loss=0.09595, over 5703493.41 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3634, pruned_loss=0.1155, over 5652060.40 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3417, pruned_loss=0.09406, over 5707239.00 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:14:22,996 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:14:31,985 INFO [train.py:968] (0/2) Epoch 26, batch 40300, giga_loss[loss=0.2624, simple_loss=0.3394, pruned_loss=0.09269, over 28564.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3425, pruned_loss=0.09626, over 5713148.07 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3635, pruned_loss=0.1155, over 5653094.10 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3406, pruned_loss=0.09472, over 5715379.67 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:14:35,299 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1179361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:14:37,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1179364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:14:56,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:15:05,204 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1179393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:15:05,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.807e+02 1.241e+03 1.498e+03 1.793e+03 7.537e+03, threshold=2.995e+03, percent-clipped=4.0
+2023-03-13 17:15:14,085 INFO [train.py:968] (0/2) Epoch 26, batch 40350, giga_loss[loss=0.3033, simple_loss=0.3639, pruned_loss=0.1214, over 28997.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3399, pruned_loss=0.09653, over 5707339.66 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3635, pruned_loss=0.1155, over 5653094.10 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3385, pruned_loss=0.09533, over 5709076.55 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:15:22,331 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5081, 4.3670, 4.1374, 1.8764], device='cuda:0'), covar=tensor([0.0661, 0.0783, 0.0740, 0.2118], device='cuda:0'), in_proj_covar=tensor([0.1268, 0.1169, 0.0985, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 17:15:54,276 INFO [train.py:968] (0/2) Epoch 26, batch 40400, giga_loss[loss=0.2829, simple_loss=0.3391, pruned_loss=0.1134, over 28638.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3395, pruned_loss=0.09677, over 5710641.59 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3637, pruned_loss=0.1154, over 5659271.11 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3377, pruned_loss=0.09547, over 5707751.51 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:16:13,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:16:25,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.183e+02 1.304e+03 1.583e+03 2.149e+03 5.378e+03, threshold=3.167e+03, percent-clipped=5.0
+2023-03-13 17:16:34,390 INFO [train.py:968] (0/2) Epoch 26, batch 40450, giga_loss[loss=0.2877, simple_loss=0.3562, pruned_loss=0.1096, over 28896.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3373, pruned_loss=0.09571, over 5706005.96 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3634, pruned_loss=0.1151, over 5661650.39 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3355, pruned_loss=0.09444, over 5702614.56 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:16:34,834 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-13 17:16:38,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4941, 2.1534, 1.6544, 0.8501], device='cuda:0'), covar=tensor([0.7659, 0.3428, 0.4648, 0.7613], device='cuda:0'), in_proj_covar=tensor([0.1805, 0.1695, 0.1640, 0.1474], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 17:16:44,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7883, 1.9817, 1.7696, 1.6864], device='cuda:0'), covar=tensor([0.2145, 0.2748, 0.2588, 0.2717], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0757, 0.0730, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 17:16:52,020 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1179527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:16:54,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1179530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:17:15,249 INFO [train.py:968] (0/2) Epoch 26, batch 40500, giga_loss[loss=0.2681, simple_loss=0.3327, pruned_loss=0.1018, over 28876.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3343, pruned_loss=0.09447, over 5712988.93 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3637, pruned_loss=0.1153, over 5666727.56 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3318, pruned_loss=0.09274, over 5706885.69 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:17:17,505 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1179559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:17:36,605 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:17:43,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.693e+02 1.244e+03 1.488e+03 1.960e+03 4.744e+03, threshold=2.976e+03, percent-clipped=6.0
+2023-03-13 17:17:46,046 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:17:52,727 INFO [train.py:968] (0/2) Epoch 26, batch 40550, libri_loss[loss=0.3234, simple_loss=0.393, pruned_loss=0.1269, over 28575.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3302, pruned_loss=0.09217, over 5704856.06 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3638, pruned_loss=0.1152, over 5661133.84 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.327, pruned_loss=0.09011, over 5706221.74 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:17:57,713 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 17:17:59,061 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:18:04,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1179622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:18:06,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1179625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:18:29,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1179654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:18:31,995 INFO [train.py:968] (0/2) Epoch 26, batch 40600, giga_loss[loss=0.2288, simple_loss=0.3126, pruned_loss=0.07246, over 28957.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.329, pruned_loss=0.09096, over 5712383.97 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3636, pruned_loss=0.1151, over 5662787.54 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3259, pruned_loss=0.08885, over 5713331.60 frames. ], batch size: 213, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:18:56,756 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6981, 1.9581, 1.5829, 1.8472], device='cuda:0'), covar=tensor([0.2870, 0.2889, 0.3327, 0.2665], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1139, 0.1392, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 17:19:01,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.588e+02 1.390e+03 1.724e+03 2.435e+03 7.985e+03, threshold=3.448e+03, percent-clipped=13.0
+2023-03-13 17:19:11,893 INFO [train.py:968] (0/2) Epoch 26, batch 40650, giga_loss[loss=0.3035, simple_loss=0.363, pruned_loss=0.122, over 29023.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3326, pruned_loss=0.09307, over 5713275.51 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3633, pruned_loss=0.1151, over 5673905.32 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3289, pruned_loss=0.0904, over 5706400.05 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:19:14,359 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2477, 2.2786, 2.1615, 1.9805], device='cuda:0'), covar=tensor([0.2183, 0.2694, 0.2440, 0.2639], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0758, 0.0730, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 17:19:26,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:19:53,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4623, 1.8792, 1.4546, 1.5158], device='cuda:0'), covar=tensor([0.2719, 0.2713, 0.3131, 0.2529], device='cuda:0'), in_proj_covar=tensor([0.1577, 0.1136, 0.1388, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 17:19:54,351 INFO [train.py:968] (0/2) Epoch 26, batch 40700, giga_loss[loss=0.2685, simple_loss=0.3473, pruned_loss=0.09484, over 28826.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3363, pruned_loss=0.09489, over 5715931.82 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3634, pruned_loss=0.1152, over 5675058.80 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3332, pruned_loss=0.09263, over 5709660.33 frames. ], batch size: 243, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:20:22,831 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.773e+02 1.390e+03 1.729e+03 2.496e+03 9.217e+03, threshold=3.457e+03, percent-clipped=8.0
+2023-03-13 17:20:28,418 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6627, 1.5019, 4.8029, 3.5181], device='cuda:0'), covar=tensor([0.1625, 0.2857, 0.0388, 0.0900], device='cuda:0'), in_proj_covar=tensor([0.0790, 0.0665, 0.0988, 0.0955], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 17:20:32,127 INFO [train.py:968] (0/2) Epoch 26, batch 40750, giga_loss[loss=0.2668, simple_loss=0.3448, pruned_loss=0.09435, over 28913.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3399, pruned_loss=0.09631, over 5714437.51 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3632, pruned_loss=0.1151, over 5681089.12 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.337, pruned_loss=0.09414, over 5704776.96 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:20:46,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5522, 1.5890, 1.6837, 1.5098], device='cuda:0'), covar=tensor([0.3216, 0.2950, 0.2407, 0.2815], device='cuda:0'), in_proj_covar=tensor([0.2046, 0.1988, 0.1913, 0.2045], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 17:21:14,080 INFO [train.py:968] (0/2) Epoch 26, batch 40800, giga_loss[loss=0.2892, simple_loss=0.348, pruned_loss=0.1152, over 28818.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3432, pruned_loss=0.09771, over 5713124.09 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3629, pruned_loss=0.1148, over 5684280.55 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3409, pruned_loss=0.096, over 5702966.13 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:21:21,289 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1179867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:21:24,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1179870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:21:24,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6506, 1.9930, 1.9555, 1.4779], device='cuda:0'), covar=tensor([0.3741, 0.2817, 0.2918, 0.3497], device='cuda:0'), in_proj_covar=tensor([0.2047, 0.1990, 0.1914, 0.2045], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 17:21:46,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.679e+02 1.391e+03 1.682e+03 2.294e+03 4.775e+03, threshold=3.363e+03, percent-clipped=6.0
+2023-03-13 17:21:46,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4208, 3.4147, 1.4788, 1.6330], device='cuda:0'), covar=tensor([0.0964, 0.0437, 0.0899, 0.1305], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0562, 0.0403, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 17:21:48,226 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1179899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:21:54,159 INFO [train.py:968] (0/2) Epoch 26, batch 40850, giga_loss[loss=0.2691, simple_loss=0.3458, pruned_loss=0.09619, over 28890.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3455, pruned_loss=0.09883, over 5708463.11 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3632, pruned_loss=0.1149, over 5679551.84 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3428, pruned_loss=0.09684, over 5705369.36 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:22:34,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2663, 1.8249, 1.4484, 0.4373], device='cuda:0'), covar=tensor([0.4828, 0.3037, 0.4210, 0.6361], device='cuda:0'), in_proj_covar=tensor([0.1814, 0.1705, 0.1645, 0.1479], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 17:22:43,096 INFO [train.py:968] (0/2) Epoch 26, batch 40900, giga_loss[loss=0.3527, simple_loss=0.3904, pruned_loss=0.1575, over 28712.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.351, pruned_loss=0.1037, over 5685449.32 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3632, pruned_loss=0.1148, over 5659757.99 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3484, pruned_loss=0.1018, over 5700219.52 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:22:45,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:22:56,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:23:15,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:23:22,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.668e+03 2.102e+03 3.043e+03 7.211e+03, threshold=4.205e+03, percent-clipped=23.0
+2023-03-13 17:23:24,980 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1180000.pt
+2023-03-13 17:23:30,053 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 17:23:32,024 INFO [train.py:968] (0/2) Epoch 26, batch 40950, giga_loss[loss=0.3694, simple_loss=0.4181, pruned_loss=0.1604, over 28635.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3573, pruned_loss=0.1086, over 5686418.75 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3632, pruned_loss=0.1148, over 5660981.20 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3552, pruned_loss=0.1071, over 5697090.74 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:24:14,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:24:15,573 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:24:18,766 INFO [train.py:968] (0/2) Epoch 26, batch 41000, giga_loss[loss=0.3322, simple_loss=0.398, pruned_loss=0.1332, over 28799.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3633, pruned_loss=0.1126, over 5679564.83 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.363, pruned_loss=0.1146, over 5655868.10 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3618, pruned_loss=0.1116, over 5693504.53 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:24:37,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5004, 1.7678, 1.4203, 1.5985], device='cuda:0'), covar=tensor([0.2472, 0.2545, 0.2895, 0.2224], device='cuda:0'), in_proj_covar=tensor([0.1576, 0.1138, 0.1391, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 17:24:52,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.536e+02 1.833e+03 2.374e+03 3.158e+03 6.560e+03, threshold=4.747e+03, percent-clipped=12.0
+2023-03-13 17:24:58,627 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:01,279 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:01,641 INFO [train.py:968] (0/2) Epoch 26, batch 41050, giga_loss[loss=0.3477, simple_loss=0.4106, pruned_loss=0.1424, over 28815.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3692, pruned_loss=0.1177, over 5681601.87 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3629, pruned_loss=0.1144, over 5657707.34 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3681, pruned_loss=0.117, over 5691430.16 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:25:09,657 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:11,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:25,243 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:28,497 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:29,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:39,287 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:47,886 INFO [train.py:968] (0/2) Epoch 26, batch 41100, giga_loss[loss=0.3185, simple_loss=0.3816, pruned_loss=0.1277, over 28632.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3754, pruned_loss=0.1227, over 5681066.00 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3629, pruned_loss=0.1144, over 5658537.76 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3747, pruned_loss=0.1223, over 5688867.86 frames. ], batch size: 242, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:25:54,956 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:55,634 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:58,180 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:26:31,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.318e+03 2.036e+03 2.627e+03 3.718e+03 1.547e+04, threshold=5.253e+03, percent-clipped=16.0
+2023-03-13 17:26:42,635 INFO [train.py:968] (0/2) Epoch 26, batch 41150, giga_loss[loss=0.3809, simple_loss=0.4065, pruned_loss=0.1777, over 23323.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3795, pruned_loss=0.1271, over 5658254.43 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3628, pruned_loss=0.1142, over 5660945.54 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3792, pruned_loss=0.127, over 5662252.57 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:27:08,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4977, 2.9110, 1.7095, 1.5578], device='cuda:0'), covar=tensor([0.0805, 0.0352, 0.0677, 0.1133], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0564, 0.0403, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 17:27:34,702 INFO [train.py:968] (0/2) Epoch 26, batch 41200, giga_loss[loss=0.3564, simple_loss=0.3952, pruned_loss=0.1588, over 28752.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.382, pruned_loss=0.1303, over 5655010.54 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3629, pruned_loss=0.1144, over 5657064.84 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3821, pruned_loss=0.1304, over 5661960.95 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:27:38,494 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.48 vs. limit=5.0
+2023-03-13 17:28:10,031 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5100, 3.5342, 1.5981, 1.7363], device='cuda:0'), covar=tensor([0.0961, 0.0340, 0.0880, 0.1238], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0564, 0.0403, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 17:28:17,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.213e+03 1.890e+03 2.735e+03 3.886e+03 9.238e+03, threshold=5.470e+03, percent-clipped=14.0
+2023-03-13 17:28:27,676 INFO [train.py:968] (0/2) Epoch 26, batch 41250, libri_loss[loss=0.266, simple_loss=0.3293, pruned_loss=0.1014, over 29392.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3861, pruned_loss=0.1353, over 5641144.51 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3628, pruned_loss=0.1143, over 5661660.33 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3867, pruned_loss=0.1358, over 5642202.71 frames. ], batch size: 67, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:28:37,712 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180318.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 17:29:07,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3604, 1.7510, 1.1586, 1.2854], device='cuda:0'), covar=tensor([0.1169, 0.0656, 0.1228, 0.1224], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0453, 0.0527, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 17:29:19,052 INFO [train.py:968] (0/2) Epoch 26, batch 41300, giga_loss[loss=0.3471, simple_loss=0.4002, pruned_loss=0.147, over 28615.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3891, pruned_loss=0.1378, over 5642965.36 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3626, pruned_loss=0.1141, over 5666710.62 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3901, pruned_loss=0.1388, over 5638757.78 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:29:52,320 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-13 17:29:55,876 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 17:30:00,175 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.346e+03 1.897e+03 2.331e+03 3.222e+03 6.618e+03, threshold=4.662e+03, percent-clipped=3.0
+2023-03-13 17:30:09,692 INFO [train.py:968] (0/2) Epoch 26, batch 41350, giga_loss[loss=0.3315, simple_loss=0.384, pruned_loss=0.1395, over 28311.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3909, pruned_loss=0.1396, over 5629100.14 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3628, pruned_loss=0.1141, over 5670787.15 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3922, pruned_loss=0.1408, over 5621332.52 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:30:32,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:30:33,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:31:00,932 INFO [train.py:968] (0/2) Epoch 26, batch 41400, giga_loss[loss=0.3056, simple_loss=0.3687, pruned_loss=0.1212, over 28707.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3907, pruned_loss=0.1405, over 5633672.16 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3627, pruned_loss=0.1141, over 5674307.54 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3921, pruned_loss=0.1419, over 5623792.39 frames. ], batch size: 242, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:31:08,871 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:31:23,504 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:31:37,477 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 17:31:39,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.277e+03 1.892e+03 2.472e+03 2.926e+03 4.954e+03, threshold=4.944e+03, percent-clipped=3.0
+2023-03-13 17:31:49,995 INFO [train.py:968] (0/2) Epoch 26, batch 41450, libri_loss[loss=0.3143, simple_loss=0.3738, pruned_loss=0.1274, over 29080.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.388, pruned_loss=0.1381, over 5654148.27 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3629, pruned_loss=0.1141, over 5679058.07 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3896, pruned_loss=0.1398, over 5641203.28 frames. ], batch size: 101, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:32:21,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:22,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:33,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6786, 1.5779, 1.8467, 1.4584], device='cuda:0'), covar=tensor([0.1433, 0.2162, 0.1177, 0.1543], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0711, 0.0968, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 17:32:34,285 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:34,422 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5676, 1.8089, 1.7664, 1.3030], device='cuda:0'), covar=tensor([0.1648, 0.2749, 0.1542, 0.1904], device='cuda:0'), in_proj_covar=tensor([0.0920, 0.0711, 0.0968, 0.0871], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 17:32:42,272 INFO [train.py:968] (0/2) Epoch 26, batch 41500, giga_loss[loss=0.3292, simple_loss=0.3904, pruned_loss=0.134, over 28293.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.387, pruned_loss=0.136, over 5649544.82 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3627, pruned_loss=0.114, over 5679744.74 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.389, pruned_loss=0.1378, over 5638223.32 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:32:52,556 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:54,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:57,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:57,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:33:08,131 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-13 17:33:20,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.204e+03 1.782e+03 2.304e+03 2.973e+03 7.370e+03, threshold=4.607e+03, percent-clipped=4.0
+2023-03-13 17:33:23,510 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:33:24,201 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:33:28,235 INFO [train.py:968] (0/2) Epoch 26, batch 41550, giga_loss[loss=0.3009, simple_loss=0.3714, pruned_loss=0.1152, over 28987.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3853, pruned_loss=0.1332, over 5673104.13 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3624, pruned_loss=0.1137, over 5689405.69 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3882, pruned_loss=0.1359, over 5654203.04 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:34:17,665 INFO [train.py:968] (0/2) Epoch 26, batch 41600, giga_loss[loss=0.3352, simple_loss=0.3867, pruned_loss=0.1418, over 27559.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3874, pruned_loss=0.135, over 5663917.48 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3625, pruned_loss=0.1139, over 5698302.21 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3906, pruned_loss=0.1378, over 5639655.15 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:34:43,784 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:34:44,303 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:34:45,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:34:46,548 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:34:53,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7363, 1.9844, 1.4240, 1.6214], device='cuda:0'), covar=tensor([0.0882, 0.0506, 0.0958, 0.1007], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0454, 0.0528, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 17:34:55,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180693.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 17:34:58,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.192e+03 1.883e+03 2.326e+03 3.339e+03 6.942e+03, threshold=4.652e+03, percent-clipped=9.0
+2023-03-13 17:35:07,235 INFO [train.py:968] (0/2) Epoch 26, batch 41650, giga_loss[loss=0.3007, simple_loss=0.3698, pruned_loss=0.1158, over 28616.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3857, pruned_loss=0.1331, over 5653073.66 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3628, pruned_loss=0.1142, over 5690553.12 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3885, pruned_loss=0.1355, over 5640395.01 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:35:11,381 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:35:12,212 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:35:53,690 INFO [train.py:968] (0/2) Epoch 26, batch 41700, giga_loss[loss=0.3204, simple_loss=0.3805, pruned_loss=0.1301, over 27749.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3836, pruned_loss=0.1305, over 5630557.04 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3631, pruned_loss=0.1145, over 5664197.15 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3863, pruned_loss=0.1326, over 5644223.29 frames. ], batch size: 474, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:36:35,201 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.861e+02 1.780e+03 2.628e+03 3.775e+03 1.436e+04, threshold=5.256e+03, percent-clipped=16.0
+2023-03-13 17:36:44,644 INFO [train.py:968] (0/2) Epoch 26, batch 41750, giga_loss[loss=0.381, simple_loss=0.4285, pruned_loss=0.1668, over 27883.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3798, pruned_loss=0.1267, over 5645729.73 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.363, pruned_loss=0.1145, over 5667664.40 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3822, pruned_loss=0.1286, over 5653170.22 frames. ], batch size: 412, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:36:44,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1976, 3.4752, 1.5021, 1.4165], device='cuda:0'), covar=tensor([0.1263, 0.0461, 0.1010, 0.1612], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0566, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 17:36:50,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:37:05,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4206, 1.3605, 3.7949, 3.2013], device='cuda:0'), covar=tensor([0.1781, 0.2764, 0.0915, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0665, 0.0991, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 17:37:11,253 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180836.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 17:37:14,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180839.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 17:37:14,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:37:21,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1186, 1.4355, 1.3475, 1.0387], device='cuda:0'), covar=tensor([0.2803, 0.2482, 0.1860, 0.2568], device='cuda:0'), in_proj_covar=tensor([0.2049, 0.1989, 0.1914, 0.2046], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 17:37:30,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:37:32,758 INFO [train.py:968] (0/2) Epoch 26, batch 41800, giga_loss[loss=0.2917, simple_loss=0.3641, pruned_loss=0.1097, over 28954.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3762, pruned_loss=0.1242, over 5650890.46 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3628, pruned_loss=0.1146, over 5674351.67 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3787, pruned_loss=0.1259, over 5650320.96 frames. ], batch size: 227, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:37:42,021 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180868.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 17:37:47,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2057, 1.3940, 1.3123, 1.1790], device='cuda:0'), covar=tensor([0.2628, 0.2467, 0.1766, 0.2284], device='cuda:0'), in_proj_covar=tensor([0.2049, 0.1989, 0.1914, 0.2045], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 17:38:13,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.808e+03 2.406e+03 3.155e+03 9.955e+03, threshold=4.812e+03, percent-clipped=5.0
+2023-03-13 17:38:19,135 INFO [train.py:968] (0/2) Epoch 26, batch 41850, giga_loss[loss=0.3032, simple_loss=0.3797, pruned_loss=0.1134, over 28862.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3747, pruned_loss=0.1237, over 5634878.44 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.363, pruned_loss=0.1148, over 5665598.28 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3767, pruned_loss=0.1251, over 5641357.05 frames. ], batch size: 119, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:38:36,587 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:38:50,106 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:39:04,812 INFO [train.py:968] (0/2) Epoch 26, batch 41900, giga_loss[loss=0.3065, simple_loss=0.3781, pruned_loss=0.1174, over 28981.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3744, pruned_loss=0.1232, over 5657993.72 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3629, pruned_loss=0.1147, over 5670276.85 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3764, pruned_loss=0.1246, over 5658374.77 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:39:32,494 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:39:36,387 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:39:49,736 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.668e+03 2.052e+03 3.113e+03 7.047e+03, threshold=4.104e+03, percent-clipped=2.0
+2023-03-13 17:39:49,998 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:39:51,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:39:57,050 INFO [train.py:968] (0/2) Epoch 26, batch 41950, giga_loss[loss=0.2874, simple_loss=0.365, pruned_loss=0.1049, over 28983.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3731, pruned_loss=0.1218, over 5666695.97 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3629, pruned_loss=0.1147, over 5676362.66 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.375, pruned_loss=0.1231, over 5661178.54 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:40:10,264 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:40:27,279 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:40:50,097 INFO [train.py:968] (0/2) Epoch 26, batch 42000, giga_loss[loss=0.287, simple_loss=0.3539, pruned_loss=0.11, over 28567.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3717, pruned_loss=0.1192, over 5676622.50 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3629, pruned_loss=0.1149, over 5679194.11 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3734, pruned_loss=0.1202, over 5669520.36 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:40:50,101 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 17:40:58,163 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.202, simple_loss=0.3083, pruned_loss=0.04787, over 944034.00 frames. 
+2023-03-13 17:40:58,163 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 17:41:08,801 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1181068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:41:10,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:41:36,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.619e+03 2.036e+03 2.531e+03 1.186e+04, threshold=4.072e+03, percent-clipped=5.0
+2023-03-13 17:41:38,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:41:47,769 INFO [train.py:968] (0/2) Epoch 26, batch 42050, giga_loss[loss=0.2763, simple_loss=0.3673, pruned_loss=0.09265, over 28976.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3716, pruned_loss=0.117, over 5678628.77 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3623, pruned_loss=0.1145, over 5685379.40 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3737, pruned_loss=0.1182, over 5667133.68 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:42:33,511 INFO [train.py:968] (0/2) Epoch 26, batch 42100, giga_loss[loss=0.3267, simple_loss=0.3841, pruned_loss=0.1347, over 27582.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3718, pruned_loss=0.1172, over 5677494.61 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3618, pruned_loss=0.1142, over 5684360.89 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3741, pruned_loss=0.1185, over 5669235.75 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:42:33,791 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4691, 1.7350, 1.5090, 1.5576], device='cuda:0'), covar=tensor([0.1747, 0.2049, 0.2181, 0.1977], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0758, 0.0730, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 17:42:56,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5193, 1.9575, 1.2383, 1.5562], device='cuda:0'), covar=tensor([0.1192, 0.0646, 0.1198, 0.1138], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0453, 0.0525, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 17:43:05,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1181189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:43:10,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5222, 1.6007, 1.5205, 1.4528], device='cuda:0'), covar=tensor([0.2386, 0.2496, 0.2112, 0.2292], device='cuda:0'), in_proj_covar=tensor([0.2045, 0.1983, 0.1910, 0.2042], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 17:43:13,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.150e+03 1.860e+03 2.449e+03 3.240e+03 6.212e+03, threshold=4.899e+03, percent-clipped=11.0
+2023-03-13 17:43:21,242 INFO [train.py:968] (0/2) Epoch 26, batch 42150, giga_loss[loss=0.3169, simple_loss=0.3863, pruned_loss=0.1237, over 29027.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3723, pruned_loss=0.1182, over 5676227.55 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.114, over 5684270.06 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3746, pruned_loss=0.1195, over 5669561.19 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:44:07,485 INFO [train.py:968] (0/2) Epoch 26, batch 42200, libri_loss[loss=0.2395, simple_loss=0.3096, pruned_loss=0.08476, over 29369.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3709, pruned_loss=0.118, over 5681745.19 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.361, pruned_loss=0.1137, over 5687549.71 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3733, pruned_loss=0.1193, over 5673329.97 frames. ], batch size: 67, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:44:49,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.281e+03 1.760e+03 2.748e+03 3.923e+03 1.086e+04, threshold=5.497e+03, percent-clipped=15.0
+2023-03-13 17:44:57,243 INFO [train.py:968] (0/2) Epoch 26, batch 42250, giga_loss[loss=0.2794, simple_loss=0.3439, pruned_loss=0.1074, over 28655.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3698, pruned_loss=0.1186, over 5675016.22 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3611, pruned_loss=0.1138, over 5689353.67 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3718, pruned_loss=0.1196, over 5666795.09 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:45:02,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1181313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:45:03,597 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-13 17:45:05,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1181317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:45:06,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1181319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:45:19,630 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1181332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:45:23,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:45:43,816 INFO [train.py:968] (0/2) Epoch 26, batch 42300, giga_loss[loss=0.3051, simple_loss=0.3781, pruned_loss=0.1161, over 28945.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3694, pruned_loss=0.1185, over 5676530.20 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3615, pruned_loss=0.1141, over 5694554.72 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3706, pruned_loss=0.1191, over 5664756.82 frames. ], batch size: 213, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:45:52,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:46:08,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6394, 1.6207, 1.8298, 1.4356], device='cuda:0'), covar=tensor([0.1715, 0.2461, 0.1465, 0.1728], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0712, 0.0969, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 17:46:25,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.597e+03 1.980e+03 2.864e+03 1.425e+04, threshold=3.960e+03, percent-clipped=3.0
+2023-03-13 17:46:30,470 INFO [train.py:968] (0/2) Epoch 26, batch 42350, giga_loss[loss=0.2536, simple_loss=0.3399, pruned_loss=0.08366, over 29025.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3681, pruned_loss=0.1158, over 5687326.91 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.114, over 5697807.02 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3693, pruned_loss=0.1164, over 5674901.16 frames. ], batch size: 113, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:46:46,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9889, 1.1076, 5.2543, 3.7788], device='cuda:0'), covar=tensor([0.1580, 0.3080, 0.0513, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0668, 0.0995, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 17:47:15,236 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1181456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:47:16,805 INFO [train.py:968] (0/2) Epoch 26, batch 42400, giga_loss[loss=0.2636, simple_loss=0.3367, pruned_loss=0.09522, over 28845.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3684, pruned_loss=0.1155, over 5689134.79 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3616, pruned_loss=0.1141, over 5703139.29 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3693, pruned_loss=0.116, over 5674211.96 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:47:20,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:47:47,614 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:47:49,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7560, 5.6121, 5.3131, 3.1737], device='cuda:0'), covar=tensor([0.0476, 0.0597, 0.0717, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.1196, 0.1005, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 17:48:00,469 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.338e+02 1.610e+03 2.076e+03 3.308e+03 9.612e+03, threshold=4.153e+03, percent-clipped=16.0
+2023-03-13 17:48:07,332 INFO [train.py:968] (0/2) Epoch 26, batch 42450, giga_loss[loss=0.3159, simple_loss=0.3687, pruned_loss=0.1315, over 28829.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3688, pruned_loss=0.1159, over 5675905.58 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3619, pruned_loss=0.1144, over 5683671.50 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3693, pruned_loss=0.116, over 5680811.63 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:48:09,469 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5528, 1.5392, 1.7214, 1.4238], device='cuda:0'), covar=tensor([0.1389, 0.1988, 0.1194, 0.1471], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0714, 0.0971, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 17:48:50,776 INFO [train.py:968] (0/2) Epoch 26, batch 42500, giga_loss[loss=0.3236, simple_loss=0.3822, pruned_loss=0.1325, over 28991.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3675, pruned_loss=0.1159, over 5670838.61 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3622, pruned_loss=0.1146, over 5680530.91 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3679, pruned_loss=0.1159, over 5676704.40 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:48:59,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5908, 1.8346, 1.4647, 1.6853], device='cuda:0'), covar=tensor([0.2698, 0.2873, 0.3268, 0.2463], device='cuda:0'), in_proj_covar=tensor([0.1577, 0.1138, 0.1393, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 17:49:21,557 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9856, 3.8369, 3.6906, 1.8443], device='cuda:0'), covar=tensor([0.0782, 0.0861, 0.0847, 0.1975], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.1196, 0.1005, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 17:49:26,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5460, 1.6834, 1.7188, 1.4871], device='cuda:0'), covar=tensor([0.3030, 0.2823, 0.2205, 0.2559], device='cuda:0'), in_proj_covar=tensor([0.2053, 0.1993, 0.1919, 0.2048], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 17:49:29,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.156e+03 1.774e+03 2.239e+03 3.220e+03 5.939e+03, threshold=4.478e+03, percent-clipped=11.0
+2023-03-13 17:49:36,890 INFO [train.py:968] (0/2) Epoch 26, batch 42550, giga_loss[loss=0.3096, simple_loss=0.3702, pruned_loss=0.1245, over 28693.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3673, pruned_loss=0.1166, over 5666742.07 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3624, pruned_loss=0.1145, over 5682406.04 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3676, pruned_loss=0.1167, over 5669455.05 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:50:20,548 INFO [train.py:968] (0/2) Epoch 26, batch 42600, giga_loss[loss=0.29, simple_loss=0.3529, pruned_loss=0.1135, over 28350.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3668, pruned_loss=0.1172, over 5664360.04 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3624, pruned_loss=0.1143, over 5686349.28 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3672, pruned_loss=0.1175, over 5662783.40 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:50:21,544 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1181658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:50:50,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1181692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:50:52,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1181694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:50:56,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.264e+03 1.785e+03 2.104e+03 2.612e+03 6.741e+03, threshold=4.208e+03, percent-clipped=3.0
+2023-03-13 17:51:03,625 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-13 17:51:05,662 INFO [train.py:968] (0/2) Epoch 26, batch 42650, giga_loss[loss=0.2866, simple_loss=0.3562, pruned_loss=0.1085, over 28515.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3668, pruned_loss=0.1178, over 5654453.71 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.363, pruned_loss=0.1148, over 5669620.09 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3666, pruned_loss=0.1177, over 5667155.55 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:51:51,057 INFO [train.py:968] (0/2) Epoch 26, batch 42700, giga_loss[loss=0.2556, simple_loss=0.3328, pruned_loss=0.08917, over 28970.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3639, pruned_loss=0.1159, over 5669889.46 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3628, pruned_loss=0.1148, over 5676745.48 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.364, pruned_loss=0.1159, over 5673457.37 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:52:28,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-13 17:52:34,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.580e+02 1.885e+03 2.605e+03 4.281e+03 1.333e+04, threshold=5.210e+03, percent-clipped=26.0
+2023-03-13 17:52:40,245 INFO [train.py:968] (0/2) Epoch 26, batch 42750, libri_loss[loss=0.2919, simple_loss=0.3709, pruned_loss=0.1065, over 28572.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.364, pruned_loss=0.1164, over 5671346.65 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3628, pruned_loss=0.1149, over 5672483.01 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3641, pruned_loss=0.1164, over 5677272.75 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:52:53,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8042, 1.7945, 1.9886, 1.5576], device='cuda:0'), covar=tensor([0.1959, 0.2636, 0.1582, 0.1845], device='cuda:0'), in_proj_covar=tensor([0.0921, 0.0712, 0.0969, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 17:53:05,414 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1181835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:53:07,403 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1181837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:53:08,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:53:09,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:53:25,797 INFO [train.py:968] (0/2) Epoch 26, batch 42800, giga_loss[loss=0.2758, simple_loss=0.3481, pruned_loss=0.1018, over 28769.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3648, pruned_loss=0.1162, over 5679334.97 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3625, pruned_loss=0.1145, over 5677725.93 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3653, pruned_loss=0.1165, over 5679503.55 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:53:35,941 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:53:37,211 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:54:02,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4619, 2.0928, 1.4437, 0.6404], device='cuda:0'), covar=tensor([0.6844, 0.3150, 0.4577, 0.7500], device='cuda:0'), in_proj_covar=tensor([0.1821, 0.1718, 0.1654, 0.1489], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 17:54:07,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3704, 1.2961, 3.4828, 3.1132], device='cuda:0'), covar=tensor([0.1539, 0.2788, 0.0509, 0.1427], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0668, 0.0998, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 17:54:08,003 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.744e+03 2.114e+03 3.070e+03 6.643e+03, threshold=4.228e+03, percent-clipped=3.0
+2023-03-13 17:54:12,049 INFO [train.py:968] (0/2) Epoch 26, batch 42850, giga_loss[loss=0.3295, simple_loss=0.3862, pruned_loss=0.1364, over 28613.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3658, pruned_loss=0.1163, over 5673098.16 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3627, pruned_loss=0.1147, over 5669380.69 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3661, pruned_loss=0.1165, over 5679845.48 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:54:20,667 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-13 17:54:55,885 INFO [train.py:968] (0/2) Epoch 26, batch 42900, giga_loss[loss=0.3072, simple_loss=0.381, pruned_loss=0.1168, over 28833.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3653, pruned_loss=0.1154, over 5661316.95 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3623, pruned_loss=0.1145, over 5664700.05 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3658, pruned_loss=0.1157, over 5671711.50 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:55:17,352 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1181979.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 17:55:35,684 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1182000.pt
+2023-03-13 17:55:36,664 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.686e+03 2.179e+03 3.128e+03 7.228e+03, threshold=4.359e+03, percent-clipped=9.0
+2023-03-13 17:55:40,975 INFO [train.py:968] (0/2) Epoch 26, batch 42950, giga_loss[loss=0.2816, simple_loss=0.3613, pruned_loss=0.1009, over 29036.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3662, pruned_loss=0.1165, over 5655196.83 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.362, pruned_loss=0.1142, over 5663313.63 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3671, pruned_loss=0.1171, over 5664487.21 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:56:08,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1182033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:56:29,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7737, 1.9662, 1.4155, 1.4899], device='cuda:0'), covar=tensor([0.1007, 0.0623, 0.1084, 0.1143], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0453, 0.0526, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 17:56:32,047 INFO [train.py:968] (0/2) Epoch 26, batch 43000, giga_loss[loss=0.3583, simple_loss=0.3989, pruned_loss=0.1589, over 27962.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3685, pruned_loss=0.1191, over 5655008.23 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3621, pruned_loss=0.1143, over 5664138.09 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3692, pruned_loss=0.1195, over 5661609.79 frames. ], batch size: 412, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:56:47,324 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1182073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:57:14,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 2.069e+03 2.882e+03 4.540e+03 1.078e+04, threshold=5.763e+03, percent-clipped=26.0
+2023-03-13 17:57:16,060 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-13 17:57:19,663 INFO [train.py:968] (0/2) Epoch 26, batch 43050, giga_loss[loss=0.2783, simple_loss=0.3469, pruned_loss=0.1048, over 28837.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3714, pruned_loss=0.1225, over 5648959.12 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3628, pruned_loss=0.1148, over 5661257.12 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3717, pruned_loss=0.1227, over 5656074.64 frames. ], batch size: 199, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:57:59,567 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4989, 2.1482, 1.5921, 0.7732], device='cuda:0'), covar=tensor([0.5914, 0.3129, 0.4075, 0.7222], device='cuda:0'), in_proj_covar=tensor([0.1823, 0.1721, 0.1654, 0.1488], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 17:58:07,931 INFO [train.py:968] (0/2) Epoch 26, batch 43100, giga_loss[loss=0.2759, simple_loss=0.3499, pruned_loss=0.1009, over 28977.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3721, pruned_loss=0.1243, over 5655038.80 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3624, pruned_loss=0.1146, over 5669951.49 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3729, pruned_loss=0.1249, over 5652289.10 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:58:28,488 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1182176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:58:32,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1182179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:58:53,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.298e+03 2.043e+03 2.709e+03 3.845e+03 1.005e+04, threshold=5.417e+03, percent-clipped=7.0
+2023-03-13 17:58:58,062 INFO [train.py:968] (0/2) Epoch 26, batch 43150, giga_loss[loss=0.2823, simple_loss=0.3454, pruned_loss=0.1096, over 28611.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3734, pruned_loss=0.1257, over 5661889.08 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3626, pruned_loss=0.1146, over 5670370.04 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.374, pruned_loss=0.1263, over 5659286.91 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:58:58,923 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1182208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:59:09,987 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8933, 2.0569, 1.7118, 2.0261], device='cuda:0'), covar=tensor([0.2480, 0.2775, 0.3052, 0.2594], device='cuda:0'), in_proj_covar=tensor([0.1576, 0.1136, 0.1393, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 17:59:37,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3534, 2.0776, 1.5527, 0.5965], device='cuda:0'), covar=tensor([0.6241, 0.3503, 0.4756, 0.7151], device='cuda:0'), in_proj_covar=tensor([0.1826, 0.1725, 0.1658, 0.1491], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 17:59:41,702 INFO [train.py:968] (0/2) Epoch 26, batch 43200, libri_loss[loss=0.2348, simple_loss=0.3122, pruned_loss=0.07867, over 28109.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3705, pruned_loss=0.1235, over 5655355.14 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3621, pruned_loss=0.1143, over 5661564.34 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3718, pruned_loss=0.1246, over 5661444.97 frames. ], batch size: 62, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:00:25,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.798e+03 2.422e+03 3.163e+03 9.742e+03, threshold=4.844e+03, percent-clipped=7.0
+2023-03-13 18:00:29,607 INFO [train.py:968] (0/2) Epoch 26, batch 43250, giga_loss[loss=0.2441, simple_loss=0.3354, pruned_loss=0.07645, over 28540.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3701, pruned_loss=0.1221, over 5665393.26 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3623, pruned_loss=0.1144, over 5664092.86 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.371, pruned_loss=0.1228, over 5668106.13 frames. ], batch size: 65, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:00:29,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9967, 1.3374, 1.1009, 0.2397], device='cuda:0'), covar=tensor([0.4581, 0.4100, 0.5209, 0.7637], device='cuda:0'), in_proj_covar=tensor([0.1824, 0.1725, 0.1656, 0.1490], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 18:01:13,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1182354.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:01:15,202 INFO [train.py:968] (0/2) Epoch 26, batch 43300, giga_loss[loss=0.3073, simple_loss=0.3519, pruned_loss=0.1314, over 26539.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3681, pruned_loss=0.1196, over 5674940.46 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.1139, over 5669489.33 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3698, pruned_loss=0.1208, over 5672167.69 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:01:24,115 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1182365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:01:45,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-13 18:01:59,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.751e+03 2.181e+03 2.754e+03 5.964e+03, threshold=4.362e+03, percent-clipped=3.0
+2023-03-13 18:02:03,397 INFO [train.py:968] (0/2) Epoch 26, batch 43350, giga_loss[loss=0.3276, simple_loss=0.3855, pruned_loss=0.1348, over 27785.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3664, pruned_loss=0.119, over 5664726.05 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3617, pruned_loss=0.1143, over 5670169.30 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3676, pruned_loss=0.1198, over 5661931.20 frames. ], batch size: 412, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:02:39,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1182448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:02:47,772 INFO [train.py:968] (0/2) Epoch 26, batch 43400, giga_loss[loss=0.3153, simple_loss=0.3781, pruned_loss=0.1262, over 28895.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3642, pruned_loss=0.1176, over 5670828.00 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3617, pruned_loss=0.1142, over 5672710.34 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3652, pruned_loss=0.1183, over 5666417.46 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:03:12,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7288, 2.1777, 1.6418, 1.9321], device='cuda:0'), covar=tensor([0.2577, 0.2524, 0.3003, 0.2500], device='cuda:0'), in_proj_covar=tensor([0.1576, 0.1134, 0.1391, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 18:03:23,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1182497.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:03:26,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1182500.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:03:29,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.867e+02 1.689e+03 2.004e+03 2.631e+03 5.796e+03, threshold=4.008e+03, percent-clipped=6.0
+2023-03-13 18:03:34,358 INFO [train.py:968] (0/2) Epoch 26, batch 43450, giga_loss[loss=0.2989, simple_loss=0.3719, pruned_loss=0.113, over 28877.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3647, pruned_loss=0.1187, over 5659241.19 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.1141, over 5667239.69 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3658, pruned_loss=0.1195, over 5660216.91 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:03:54,194 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1182529.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 18:04:12,737 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2204, 1.2602, 3.3690, 3.1030], device='cuda:0'), covar=tensor([0.1650, 0.2764, 0.0581, 0.1007], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0672, 0.1001, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 18:04:22,665 INFO [train.py:968] (0/2) Epoch 26, batch 43500, giga_loss[loss=0.2759, simple_loss=0.3557, pruned_loss=0.09804, over 28962.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3685, pruned_loss=0.1204, over 5660358.51 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.114, over 5668562.13 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3695, pruned_loss=0.1213, over 5659902.54 frames. ], batch size: 199, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:04:53,917 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1182591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:04:57,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1182594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:05:05,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.178e+03 1.665e+03 2.146e+03 2.920e+03 1.525e+04, threshold=4.292e+03, percent-clipped=9.0
+2023-03-13 18:05:09,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6245, 1.8972, 1.7943, 1.6962], device='cuda:0'), covar=tensor([0.2277, 0.2711, 0.2330, 0.2455], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0761, 0.0732, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 18:05:11,303 INFO [train.py:968] (0/2) Epoch 26, batch 43550, giga_loss[loss=0.3133, simple_loss=0.3955, pruned_loss=0.1156, over 28593.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3716, pruned_loss=0.1196, over 5658633.75 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3613, pruned_loss=0.1139, over 5670123.90 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3726, pruned_loss=0.1204, over 5656986.35 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:05:28,781 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1182623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:05:59,050 INFO [train.py:968] (0/2) Epoch 26, batch 43600, giga_loss[loss=0.3176, simple_loss=0.3913, pruned_loss=0.122, over 29016.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3732, pruned_loss=0.1199, over 5658813.69 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5667349.88 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3744, pruned_loss=0.1208, over 5659750.27 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:06:39,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.571e+02 1.741e+03 2.389e+03 3.330e+03 1.077e+04, threshold=4.778e+03, percent-clipped=15.0
+2023-03-13 18:06:43,274 INFO [train.py:968] (0/2) Epoch 26, batch 43650, giga_loss[loss=0.3286, simple_loss=0.3935, pruned_loss=0.1318, over 28310.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3753, pruned_loss=0.1215, over 5644333.02 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3615, pruned_loss=0.1142, over 5642820.15 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3764, pruned_loss=0.1221, over 5665919.87 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:07:14,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1182740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:07:30,977 INFO [train.py:968] (0/2) Epoch 26, batch 43700, giga_loss[loss=0.4048, simple_loss=0.4392, pruned_loss=0.1851, over 26767.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3766, pruned_loss=0.123, over 5650924.26 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.114, over 5646448.96 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3778, pruned_loss=0.1238, over 5665084.21 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:07:32,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4008, 1.6042, 1.2913, 1.1402], device='cuda:0'), covar=tensor([0.1083, 0.0549, 0.1048, 0.1139], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0454, 0.0527, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 18:08:09,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.298e+03 1.709e+03 2.124e+03 2.754e+03 4.312e+03, threshold=4.248e+03, percent-clipped=1.0
+2023-03-13 18:08:12,198 INFO [train.py:968] (0/2) Epoch 26, batch 43750, libri_loss[loss=0.2901, simple_loss=0.3642, pruned_loss=0.108, over 29536.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3751, pruned_loss=0.1226, over 5665340.05 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5654304.06 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3768, pruned_loss=0.1238, over 5669855.05 frames. ], batch size: 83, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:08:13,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2565, 1.2259, 3.5982, 3.2282], device='cuda:0'), covar=tensor([0.1649, 0.2774, 0.0497, 0.1020], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0671, 0.1001, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 18:08:58,419 INFO [train.py:968] (0/2) Epoch 26, batch 43800, giga_loss[loss=0.2504, simple_loss=0.3306, pruned_loss=0.0851, over 28939.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3734, pruned_loss=0.1225, over 5657291.86 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3606, pruned_loss=0.1135, over 5657418.88 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3755, pruned_loss=0.1239, over 5658032.43 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:09:23,185 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1182883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:09:25,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1182886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:09:39,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.829e+03 2.191e+03 3.011e+03 8.462e+03, threshold=4.383e+03, percent-clipped=7.0
+2023-03-13 18:09:41,560 INFO [train.py:968] (0/2) Epoch 26, batch 43850, giga_loss[loss=0.3279, simple_loss=0.383, pruned_loss=0.1364, over 28460.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3715, pruned_loss=0.1219, over 5669405.31 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3604, pruned_loss=0.1132, over 5666255.02 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.374, pruned_loss=0.1236, over 5661956.65 frames. ], batch size: 78, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:09:50,006 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1182915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:10:30,009 INFO [train.py:968] (0/2) Epoch 26, batch 43900, giga_loss[loss=0.2924, simple_loss=0.3567, pruned_loss=0.114, over 28520.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3702, pruned_loss=0.1216, over 5676394.68 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3608, pruned_loss=0.1135, over 5671298.78 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.372, pruned_loss=0.1228, over 5666145.74 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:10:34,966 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2470, 1.4517, 1.5538, 1.3551], device='cuda:0'), covar=tensor([0.1492, 0.1141, 0.1799, 0.1330], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0764, 0.0733, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 18:10:49,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1182975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:11:17,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.207e+03 2.058e+03 2.687e+03 4.242e+03 9.809e+03, threshold=5.374e+03, percent-clipped=23.0
+2023-03-13 18:11:20,832 INFO [train.py:968] (0/2) Epoch 26, batch 43950, giga_loss[loss=0.3299, simple_loss=0.3875, pruned_loss=0.1361, over 28286.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3717, pruned_loss=0.1235, over 5646615.15 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5662972.11 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3729, pruned_loss=0.1244, over 5645636.58 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:11:44,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5892, 1.7563, 1.3253, 1.2694], device='cuda:0'), covar=tensor([0.1021, 0.0586, 0.1057, 0.1193], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0454, 0.0527, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 18:12:11,460 INFO [train.py:968] (0/2) Epoch 26, batch 44000, giga_loss[loss=0.2946, simple_loss=0.3592, pruned_loss=0.1149, over 28960.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3702, pruned_loss=0.1232, over 5650186.49 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3613, pruned_loss=0.1139, over 5664194.25 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3712, pruned_loss=0.124, over 5648388.56 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:12:28,129 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-13 18:12:39,998 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-13 18:12:53,328 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.754e+03 2.537e+03 3.207e+03 6.456e+03, threshold=5.075e+03, percent-clipped=3.0
+2023-03-13 18:12:56,031 INFO [train.py:968] (0/2) Epoch 26, batch 44050, giga_loss[loss=0.2867, simple_loss=0.3583, pruned_loss=0.1076, over 28881.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3685, pruned_loss=0.1218, over 5656441.68 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3616, pruned_loss=0.1141, over 5659116.55 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3691, pruned_loss=0.1223, over 5659379.42 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:13:19,745 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1183134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:13:41,566 INFO [train.py:968] (0/2) Epoch 26, batch 44100, giga_loss[loss=0.3185, simple_loss=0.3873, pruned_loss=0.1249, over 28601.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.367, pruned_loss=0.1204, over 5655790.36 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3615, pruned_loss=0.114, over 5663373.11 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3677, pruned_loss=0.121, over 5654325.55 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:14:22,829 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.86 vs. limit=2.0
+2023-03-13 18:14:24,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.736e+03 2.127e+03 2.859e+03 7.857e+03, threshold=4.254e+03, percent-clipped=4.0
+2023-03-13 18:14:27,757 INFO [train.py:968] (0/2) Epoch 26, batch 44150, giga_loss[loss=0.3912, simple_loss=0.4232, pruned_loss=0.1796, over 26767.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3679, pruned_loss=0.1202, over 5650800.85 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3609, pruned_loss=0.1137, over 5661611.30 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3693, pruned_loss=0.1212, over 5650540.63 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:14:37,799 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1183217.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 18:14:46,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7687, 1.9795, 1.6855, 1.7666], device='cuda:0'), covar=tensor([0.2110, 0.2041, 0.2079, 0.2025], device='cuda:0'), in_proj_covar=tensor([0.1578, 0.1138, 0.1396, 0.1010], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 18:15:13,999 INFO [train.py:968] (0/2) Epoch 26, batch 44200, giga_loss[loss=0.2689, simple_loss=0.3426, pruned_loss=0.09756, over 28649.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3686, pruned_loss=0.1204, over 5653870.87 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3607, pruned_loss=0.1135, over 5666744.35 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.37, pruned_loss=0.1216, over 5648771.24 frames. ], batch size: 242, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:15:56,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.154e+02 1.795e+03 2.332e+03 3.070e+03 8.373e+03, threshold=4.663e+03, percent-clipped=11.0
+2023-03-13 18:15:56,511 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-13 18:15:59,381 INFO [train.py:968] (0/2) Epoch 26, batch 44250, giga_loss[loss=0.3389, simple_loss=0.3872, pruned_loss=0.1453, over 27542.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3676, pruned_loss=0.1196, over 5656192.92 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3609, pruned_loss=0.1136, over 5662220.39 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3688, pruned_loss=0.1207, over 5655050.77 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:16:38,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1183350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:16:42,959 INFO [train.py:968] (0/2) Epoch 26, batch 44300, giga_loss[loss=0.2847, simple_loss=0.3687, pruned_loss=0.1003, over 28914.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3694, pruned_loss=0.1177, over 5672444.16 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3606, pruned_loss=0.1134, over 5668173.38 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3709, pruned_loss=0.1189, over 5666108.51 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:17:24,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.467e+03 1.908e+03 2.523e+03 5.310e+03, threshold=3.816e+03, percent-clipped=2.0
+2023-03-13 18:17:27,339 INFO [train.py:968] (0/2) Epoch 26, batch 44350, giga_loss[loss=0.3445, simple_loss=0.4186, pruned_loss=0.1352, over 29023.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3713, pruned_loss=0.1172, over 5667251.57 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3605, pruned_loss=0.1133, over 5670676.81 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3727, pruned_loss=0.1183, over 5660080.66 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:17:33,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5375, 5.3579, 5.0897, 2.5281], device='cuda:0'), covar=tensor([0.0547, 0.0769, 0.0887, 0.1797], device='cuda:0'), in_proj_covar=tensor([0.1310, 0.1211, 0.1019, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 18:18:10,652 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-13 18:18:18,527 INFO [train.py:968] (0/2) Epoch 26, batch 44400, giga_loss[loss=0.326, simple_loss=0.4013, pruned_loss=0.1253, over 28869.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3743, pruned_loss=0.1196, over 5665795.22 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5675491.51 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3761, pruned_loss=0.1208, over 5655710.45 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 18:18:24,776 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1183462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:18:54,664 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1183493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:18:56,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1183496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:19:05,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.299e+03 1.890e+03 2.383e+03 3.080e+03 6.427e+03, threshold=4.765e+03, percent-clipped=11.0
+2023-03-13 18:19:07,164 INFO [train.py:968] (0/2) Epoch 26, batch 44450, giga_loss[loss=0.2787, simple_loss=0.3564, pruned_loss=0.1005, over 29051.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3791, pruned_loss=0.1242, over 5670154.80 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3602, pruned_loss=0.1132, over 5678651.99 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3807, pruned_loss=0.1253, over 5658849.09 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:19:09,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1183509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:19:26,428 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1183525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:19:54,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5831, 1.8066, 1.7827, 1.3323], device='cuda:0'), covar=tensor([0.1810, 0.2851, 0.1625, 0.1954], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0715, 0.0972, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 18:19:56,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3717, 1.8970, 1.5747, 1.5755], device='cuda:0'), covar=tensor([0.0788, 0.0305, 0.0321, 0.0884], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 18:19:58,060 INFO [train.py:968] (0/2) Epoch 26, batch 44500, giga_loss[loss=0.3437, simple_loss=0.3989, pruned_loss=0.1443, over 29084.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3789, pruned_loss=0.125, over 5670924.52 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1127, over 5680979.92 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3813, pruned_loss=0.1266, over 5659071.96 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:19:59,538 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1183559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:20:01,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6138, 1.5565, 1.7809, 1.4327], device='cuda:0'), covar=tensor([0.1442, 0.2251, 0.1223, 0.1516], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0715, 0.0972, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 18:20:32,277 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1183592.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 18:20:42,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.189e+03 1.742e+03 2.566e+03 3.369e+03 9.372e+03, threshold=5.132e+03, percent-clipped=7.0
+2023-03-13 18:20:44,434 INFO [train.py:968] (0/2) Epoch 26, batch 44550, giga_loss[loss=0.3933, simple_loss=0.4271, pruned_loss=0.1797, over 26591.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3771, pruned_loss=0.124, over 5681240.60 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1128, over 5685249.87 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3793, pruned_loss=0.1255, over 5667781.49 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:21:17,921 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-13 18:21:26,544 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1183652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:21:29,215 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1183655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:21:30,362 INFO [train.py:968] (0/2) Epoch 26, batch 44600, giga_loss[loss=0.2826, simple_loss=0.3571, pruned_loss=0.104, over 28903.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3752, pruned_loss=0.1224, over 5674002.38 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5687468.21 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3772, pruned_loss=0.1238, over 5661335.11 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:21:54,376 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1183684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:22:14,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.567e+02 1.665e+03 2.028e+03 2.807e+03 1.106e+04, threshold=4.056e+03, percent-clipped=3.0
+2023-03-13 18:22:15,257 INFO [train.py:968] (0/2) Epoch 26, batch 44650, giga_loss[loss=0.2965, simple_loss=0.3748, pruned_loss=0.1091, over 28925.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3754, pruned_loss=0.1208, over 5671958.28 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5680966.01 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3772, pruned_loss=0.1219, over 5668265.75 frames. ], batch size: 227, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:22:43,643 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1183735.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 18:22:45,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1183738.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:23:02,533 INFO [train.py:968] (0/2) Epoch 26, batch 44700, giga_loss[loss=0.3809, simple_loss=0.4272, pruned_loss=0.1673, over 28302.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3752, pruned_loss=0.1196, over 5669415.03 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3599, pruned_loss=0.1129, over 5666946.76 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3768, pruned_loss=0.1205, over 5678594.22 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:23:14,760 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1183767.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:23:47,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9651, 3.7850, 3.5921, 1.7490], device='cuda:0'), covar=tensor([0.0897, 0.0991, 0.1097, 0.2184], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.1214, 0.1023, 0.0758], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 18:23:48,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.745e+03 2.388e+03 3.338e+03 8.929e+03, threshold=4.775e+03, percent-clipped=16.0
+2023-03-13 18:23:49,644 INFO [train.py:968] (0/2) Epoch 26, batch 44750, giga_loss[loss=0.3138, simple_loss=0.3793, pruned_loss=0.1242, over 28831.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3751, pruned_loss=0.1204, over 5649729.25 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5663321.76 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3772, pruned_loss=0.1216, over 5660000.69 frames. ], batch size: 199, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:24:20,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1183837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:24:28,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5921, 1.8133, 1.7495, 1.6823], device='cuda:0'), covar=tensor([0.2146, 0.2154, 0.2516, 0.2168], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0762, 0.0730, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 18:24:37,677 INFO [train.py:968] (0/2) Epoch 26, batch 44800, giga_loss[loss=0.2687, simple_loss=0.3378, pruned_loss=0.09977, over 28600.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3737, pruned_loss=0.12, over 5646162.83 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3593, pruned_loss=0.1125, over 5662446.26 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3756, pruned_loss=0.1211, over 5654827.91 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 18:24:44,464 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8429, 3.6849, 3.5133, 1.7776], device='cuda:0'), covar=tensor([0.0785, 0.0873, 0.0760, 0.2057], device='cuda:0'), in_proj_covar=tensor([0.1312, 0.1213, 0.1023, 0.0758], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 18:25:26,155 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.172e+03 1.795e+03 2.344e+03 3.447e+03 8.639e+03, threshold=4.687e+03, percent-clipped=10.0
+2023-03-13 18:25:26,827 INFO [train.py:968] (0/2) Epoch 26, batch 44850, libri_loss[loss=0.2978, simple_loss=0.3591, pruned_loss=0.1183, over 29580.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3721, pruned_loss=0.12, over 5659501.13 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3589, pruned_loss=0.1123, over 5666248.73 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3743, pruned_loss=0.1213, over 5662683.73 frames. ], batch size: 74, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:25:55,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1183934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:26:15,628 INFO [train.py:968] (0/2) Epoch 26, batch 44900, giga_loss[loss=0.2747, simple_loss=0.348, pruned_loss=0.1008, over 28845.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3708, pruned_loss=0.1202, over 5654061.37 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3586, pruned_loss=0.112, over 5667811.62 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.373, pruned_loss=0.1215, over 5655016.60 frames. ], batch size: 199, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:26:22,310 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 18:26:36,217 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1183980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:26:38,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1183983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:26:57,132 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1184000.pt
+2023-03-13 18:27:03,443 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.764e+03 2.285e+03 2.874e+03 7.242e+03, threshold=4.570e+03, percent-clipped=5.0
+2023-03-13 18:27:04,110 INFO [train.py:968] (0/2) Epoch 26, batch 44950, libri_loss[loss=0.2855, simple_loss=0.3489, pruned_loss=0.1111, over 29378.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.37, pruned_loss=0.1203, over 5653953.39 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3591, pruned_loss=0.1124, over 5668631.72 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3716, pruned_loss=0.1213, over 5653491.81 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:27:07,780 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1184012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:27:22,800 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-13 18:27:50,374 INFO [train.py:968] (0/2) Epoch 26, batch 45000, libri_loss[loss=0.3171, simple_loss=0.384, pruned_loss=0.1251, over 29518.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3706, pruned_loss=0.1222, over 5649337.02 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.359, pruned_loss=0.1123, over 5665953.48 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3722, pruned_loss=0.1232, over 5650687.31 frames. ], batch size: 89, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:27:50,378 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 18:27:58,374 INFO [train.py:1012] (0/2) Epoch 26, validation: loss=0.2047, simple_loss=0.3142, pruned_loss=0.04765, over 944034.00 frames. 
+2023-03-13 18:27:58,375 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 18:28:17,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1184077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:28:20,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1184080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:28:43,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.133e+03 1.909e+03 2.475e+03 3.207e+03 7.855e+03, threshold=4.950e+03, percent-clipped=11.0
+2023-03-13 18:28:45,011 INFO [train.py:968] (0/2) Epoch 26, batch 45050, giga_loss[loss=0.2967, simple_loss=0.3703, pruned_loss=0.1115, over 28275.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3699, pruned_loss=0.1218, over 5644296.09 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.359, pruned_loss=0.1123, over 5668406.68 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3712, pruned_loss=0.1227, over 5642986.66 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:28:47,295 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1184109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:29:31,346 INFO [train.py:968] (0/2) Epoch 26, batch 45100, giga_loss[loss=0.2581, simple_loss=0.3344, pruned_loss=0.09092, over 28647.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.365, pruned_loss=0.1164, over 5657592.31 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3585, pruned_loss=0.1119, over 5672211.98 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3668, pruned_loss=0.1176, over 5652582.13 frames. ], batch size: 119, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:29:34,592 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-13 18:29:44,141 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6846, 1.9914, 1.3870, 1.4322], device='cuda:0'), covar=tensor([0.1082, 0.0628, 0.1111, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0450, 0.0523, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 18:30:13,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.343e+03 1.794e+03 2.258e+03 6.723e+03, threshold=3.587e+03, percent-clipped=1.0
+2023-03-13 18:30:13,935 INFO [train.py:968] (0/2) Epoch 26, batch 45150, libri_loss[loss=0.259, simple_loss=0.3415, pruned_loss=0.08822, over 29489.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3619, pruned_loss=0.1136, over 5657425.48 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3584, pruned_loss=0.1119, over 5679754.86 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3636, pruned_loss=0.1147, over 5645550.16 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:30:14,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2654, 4.1174, 3.9138, 2.0533], device='cuda:0'), covar=tensor([0.0633, 0.0722, 0.0732, 0.1876], device='cuda:0'), in_proj_covar=tensor([0.1311, 0.1214, 0.1021, 0.0757], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 18:30:42,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4565, 1.6398, 1.5490, 1.3295], device='cuda:0'), covar=tensor([0.2777, 0.2386, 0.2325, 0.2811], device='cuda:0'), in_proj_covar=tensor([0.2056, 0.2007, 0.1925, 0.2063], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 18:31:04,707 INFO [train.py:968] (0/2) Epoch 26, batch 45200, libri_loss[loss=0.2917, simple_loss=0.3609, pruned_loss=0.1112, over 29552.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5664453.62 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3583, pruned_loss=0.1119, over 5683591.89 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3633, pruned_loss=0.1148, over 5650935.03 frames. ], batch size: 83, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:31:10,262 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1184262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:31:12,673 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-13 18:32:00,333 INFO [train.py:968] (0/2) Epoch 26, batch 45250, giga_loss[loss=0.2513, simple_loss=0.3263, pruned_loss=0.08813, over 28805.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3586, pruned_loss=0.1124, over 5676340.40 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3581, pruned_loss=0.1118, over 5684554.93 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.36, pruned_loss=0.1133, over 5664799.08 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:32:00,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 2.025e+03 2.481e+03 3.638e+03 1.028e+04, threshold=4.961e+03, percent-clipped=25.0
+2023-03-13 18:32:45,799 INFO [train.py:968] (0/2) Epoch 26, batch 45300, giga_loss[loss=0.3106, simple_loss=0.3827, pruned_loss=0.1193, over 28876.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3596, pruned_loss=0.1136, over 5665777.45 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3586, pruned_loss=0.1122, over 5667229.68 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3603, pruned_loss=0.1138, over 5672269.10 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:33:20,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9039, 2.1881, 1.5586, 1.6429], device='cuda:0'), covar=tensor([0.1134, 0.0726, 0.1085, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0450, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 18:33:32,762 INFO [train.py:968] (0/2) Epoch 26, batch 45350, giga_loss[loss=0.3049, simple_loss=0.3764, pruned_loss=0.1167, over 28881.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3616, pruned_loss=0.1141, over 5672642.84 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3585, pruned_loss=0.1122, over 5670476.87 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3622, pruned_loss=0.1143, over 5674886.68 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:33:33,347 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.237e+03 1.714e+03 2.273e+03 2.997e+03 5.629e+03, threshold=4.547e+03, percent-clipped=2.0
+2023-03-13 18:34:22,458 INFO [train.py:968] (0/2) Epoch 26, batch 45400, giga_loss[loss=0.3327, simple_loss=0.3887, pruned_loss=0.1383, over 28273.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3649, pruned_loss=0.1162, over 5669393.28 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3589, pruned_loss=0.1124, over 5675312.99 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3651, pruned_loss=0.1162, over 5666851.88 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:34:45,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5445, 1.8131, 1.5019, 1.4582], device='cuda:0'), covar=tensor([0.2523, 0.2362, 0.2593, 0.2331], device='cuda:0'), in_proj_covar=tensor([0.1581, 0.1139, 0.1396, 0.1010], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 18:35:08,030 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-13 18:35:08,811 INFO [train.py:968] (0/2) Epoch 26, batch 45450, giga_loss[loss=0.3179, simple_loss=0.3837, pruned_loss=0.1261, over 28857.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3663, pruned_loss=0.117, over 5675911.51 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3592, pruned_loss=0.1123, over 5681062.45 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3663, pruned_loss=0.1172, over 5668840.14 frames. ], batch size: 227, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:35:09,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.672e+03 2.145e+03 2.749e+03 8.607e+03, threshold=4.289e+03, percent-clipped=3.0
+2023-03-13 18:35:26,441 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 18:35:50,124 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 18:35:56,127 INFO [train.py:968] (0/2) Epoch 26, batch 45500, giga_loss[loss=0.3961, simple_loss=0.4239, pruned_loss=0.1842, over 26560.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3676, pruned_loss=0.119, over 5659994.62 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3591, pruned_loss=0.1123, over 5683173.59 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3678, pruned_loss=0.1193, over 5652384.82 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:36:19,027 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1184585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:36:42,003 INFO [train.py:968] (0/2) Epoch 26, batch 45550, giga_loss[loss=0.3083, simple_loss=0.3713, pruned_loss=0.1226, over 28784.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3702, pruned_loss=0.1206, over 5634225.64 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3594, pruned_loss=0.1124, over 5665807.79 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3703, pruned_loss=0.1208, over 5643144.67 frames. ], batch size: 119, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:36:43,687 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.779e+03 2.301e+03 3.030e+03 8.381e+03, threshold=4.603e+03, percent-clipped=9.0
+2023-03-13 18:36:45,620 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.27 vs. limit=5.0
+2023-03-13 18:36:47,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5232, 1.6023, 1.5081, 1.4259], device='cuda:0'), covar=tensor([0.1582, 0.1951, 0.2069, 0.1894], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0760, 0.0728, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 18:37:09,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1184637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:37:28,205 INFO [train.py:968] (0/2) Epoch 26, batch 45600, giga_loss[loss=0.3655, simple_loss=0.4036, pruned_loss=0.1637, over 26696.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3718, pruned_loss=0.1216, over 5647630.47 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3593, pruned_loss=0.1122, over 5671638.28 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3722, pruned_loss=0.1221, over 5648958.85 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:37:47,339 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1184678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:38:15,961 INFO [train.py:968] (0/2) Epoch 26, batch 45650, giga_loss[loss=0.327, simple_loss=0.3854, pruned_loss=0.1342, over 28524.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3732, pruned_loss=0.1228, over 5624147.10 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3595, pruned_loss=0.1123, over 5646683.20 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3735, pruned_loss=0.1233, over 5646271.80 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:38:16,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.999e+03 2.399e+03 3.259e+03 9.363e+03, threshold=4.799e+03, percent-clipped=10.0
+2023-03-13 18:39:06,107 INFO [train.py:968] (0/2) Epoch 26, batch 45700, giga_loss[loss=0.2812, simple_loss=0.354, pruned_loss=0.1042, over 28966.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3746, pruned_loss=0.124, over 5603754.80 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3601, pruned_loss=0.1128, over 5610256.26 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3745, pruned_loss=0.1241, over 5652980.95 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:39:30,055 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1184780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:39:32,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1184783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:39:57,176 INFO [train.py:968] (0/2) Epoch 26, batch 45750, giga_loss[loss=0.2989, simple_loss=0.3535, pruned_loss=0.1222, over 23876.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3754, pruned_loss=0.1236, over 5557711.07 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3608, pruned_loss=0.1134, over 5559296.41 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.375, pruned_loss=0.1232, over 5643664.25 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:39:59,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.246e+03 2.057e+03 2.891e+03 4.400e+03 2.830e+04, threshold=5.781e+03, percent-clipped=18.0
+2023-03-13 18:40:03,594 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1184812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:40:13,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0941, 3.9526, 3.7226, 1.7350], device='cuda:0'), covar=tensor([0.0723, 0.0780, 0.0910, 0.2183], device='cuda:0'), in_proj_covar=tensor([0.1305, 0.1210, 0.1020, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 18:40:27,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8089, 2.1125, 2.1348, 1.6881], device='cuda:0'), covar=tensor([0.3091, 0.2415, 0.2424, 0.2880], device='cuda:0'), in_proj_covar=tensor([0.2047, 0.1994, 0.1916, 0.2051], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 18:40:48,265 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-13 18:40:50,172 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-26.pt
+2023-03-13 18:42:03,608 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1184899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:42:08,745 INFO [train.py:968] (0/2) Epoch 27, batch 50, giga_loss[loss=0.2676, simple_loss=0.3542, pruned_loss=0.09053, over 28855.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3721, pruned_loss=0.1081, over 1263311.41 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3385, pruned_loss=0.08976, over 116791.14 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3753, pruned_loss=0.1098, over 1170121.25 frames. ], batch size: 199, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:42:13,759 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.145e+02 1.395e+03 1.780e+03 2.334e+03 4.741e+03, threshold=3.560e+03, percent-clipped=0.0
+2023-03-13 18:42:59,629 INFO [train.py:968] (0/2) Epoch 27, batch 100, giga_loss[loss=0.282, simple_loss=0.3687, pruned_loss=0.09762, over 28884.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3631, pruned_loss=0.1029, over 2242463.02 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3409, pruned_loss=0.09138, over 171162.11 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3645, pruned_loss=0.1037, over 2135433.21 frames. ], batch size: 199, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:43:04,002 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1184959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:43:04,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1184960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:43:43,584 INFO [train.py:968] (0/2) Epoch 27, batch 150, giga_loss[loss=0.2493, simple_loss=0.3188, pruned_loss=0.08987, over 27651.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3483, pruned_loss=0.0967, over 3010035.53 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3412, pruned_loss=0.09308, over 368743.15 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3494, pruned_loss=0.09723, over 2821309.81 frames. ], batch size: 472, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:43:46,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.768e+02 1.158e+03 1.467e+03 1.921e+03 8.552e+03, threshold=2.935e+03, percent-clipped=2.0
+2023-03-13 18:43:59,733 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1628, 1.2031, 5.2701, 3.8554], device='cuda:0'), covar=tensor([0.1515, 0.3154, 0.0463, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0671, 0.1001, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 18:44:17,144 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 18:44:22,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1185053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:44:23,518 INFO [train.py:968] (0/2) Epoch 27, batch 200, giga_loss[loss=0.2208, simple_loss=0.3016, pruned_loss=0.06997, over 28294.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3343, pruned_loss=0.08972, over 3608000.65 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3401, pruned_loss=0.09262, over 468398.91 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3345, pruned_loss=0.08982, over 3421120.30 frames. ], batch size: 368, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:44:36,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8409, 2.0173, 1.6889, 1.9835], device='cuda:0'), covar=tensor([0.2963, 0.3094, 0.3483, 0.2799], device='cuda:0'), in_proj_covar=tensor([0.1583, 0.1141, 0.1399, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 18:45:07,360 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1185103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:45:08,250 INFO [train.py:968] (0/2) Epoch 27, batch 250, giga_loss[loss=0.2339, simple_loss=0.3053, pruned_loss=0.08124, over 28979.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3242, pruned_loss=0.08534, over 4071636.71 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.342, pruned_loss=0.0936, over 495157.92 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3237, pruned_loss=0.08512, over 3916763.21 frames. ], batch size: 112, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:45:09,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1185106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:45:10,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.536e+02 1.114e+03 1.382e+03 1.970e+03 4.535e+03, threshold=2.764e+03, percent-clipped=5.0
+2023-03-13 18:45:29,626 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1185135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:45:44,874 INFO [train.py:968] (0/2) Epoch 27, batch 300, giga_loss[loss=0.1924, simple_loss=0.2763, pruned_loss=0.05423, over 28879.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3153, pruned_loss=0.08103, over 4432208.55 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3414, pruned_loss=0.09166, over 672673.08 frames. ], giga_tot_loss[loss=0.2375, simple_loss=0.3138, pruned_loss=0.08056, over 4263994.96 frames. ], batch size: 174, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:46:22,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7365, 1.8916, 1.5478, 1.8929], device='cuda:0'), covar=tensor([0.0761, 0.0311, 0.0344, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 18:46:22,320 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1185196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:46:25,591 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1185199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:46:29,461 INFO [train.py:968] (0/2) Epoch 27, batch 350, giga_loss[loss=0.2121, simple_loss=0.2881, pruned_loss=0.06801, over 28820.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3086, pruned_loss=0.07806, over 4717508.47 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3389, pruned_loss=0.08978, over 827917.66 frames. ], giga_tot_loss[loss=0.2307, simple_loss=0.3065, pruned_loss=0.07746, over 4546605.49 frames. ], batch size: 186, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:46:32,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.596e+02 1.073e+03 1.359e+03 1.664e+03 3.503e+03, threshold=2.718e+03, percent-clipped=4.0
+2023-03-13 18:46:49,805 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1185228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:47:09,772 INFO [train.py:968] (0/2) Epoch 27, batch 400, giga_loss[loss=0.2353, simple_loss=0.3068, pruned_loss=0.08188, over 28543.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.305, pruned_loss=0.07637, over 4936801.02 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3397, pruned_loss=0.08933, over 975115.35 frames. ], giga_tot_loss[loss=0.2266, simple_loss=0.3021, pruned_loss=0.07552, over 4773986.20 frames. ], batch size: 71, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:47:26,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1185274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:47:50,695 INFO [train.py:968] (0/2) Epoch 27, batch 450, giga_loss[loss=0.22, simple_loss=0.2945, pruned_loss=0.07277, over 28704.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3021, pruned_loss=0.07457, over 5110081.55 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3406, pruned_loss=0.08902, over 1093084.78 frames. ], giga_tot_loss[loss=0.2229, simple_loss=0.2987, pruned_loss=0.07357, over 4964504.58 frames. ], batch size: 262, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:47:55,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.287e+02 1.207e+03 1.405e+03 1.852e+03 3.805e+03, threshold=2.811e+03, percent-clipped=8.0
+2023-03-13 18:48:15,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1185334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:48:19,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3066, 3.1341, 2.9757, 1.2580], device='cuda:0'), covar=tensor([0.0938, 0.1121, 0.0973, 0.2456], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.1199, 0.1010, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 18:48:33,729 INFO [train.py:968] (0/2) Epoch 27, batch 500, giga_loss[loss=0.2023, simple_loss=0.2829, pruned_loss=0.0609, over 28509.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3, pruned_loss=0.07382, over 5240447.30 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3412, pruned_loss=0.08941, over 1183335.09 frames. ], giga_tot_loss[loss=0.2209, simple_loss=0.2964, pruned_loss=0.07266, over 5116812.86 frames. ], batch size: 336, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:49:13,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-13 18:49:16,674 INFO [train.py:968] (0/2) Epoch 27, batch 550, giga_loss[loss=0.2299, simple_loss=0.2993, pruned_loss=0.08025, over 28711.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.298, pruned_loss=0.07279, over 5347879.22 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3405, pruned_loss=0.08866, over 1276024.04 frames. ], giga_tot_loss[loss=0.2189, simple_loss=0.2944, pruned_loss=0.0717, over 5239698.66 frames. ], batch size: 262, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:49:22,186 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.121e+02 1.149e+03 1.419e+03 1.783e+03 8.043e+03, threshold=2.838e+03, percent-clipped=6.0
+2023-03-13 18:49:29,609 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1185417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:49:31,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1185420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:49:56,399 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1185449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:50:01,280 INFO [train.py:968] (0/2) Epoch 27, batch 600, giga_loss[loss=0.1995, simple_loss=0.2767, pruned_loss=0.06114, over 28519.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2957, pruned_loss=0.07209, over 5421598.63 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3423, pruned_loss=0.08996, over 1343545.73 frames. ], giga_tot_loss[loss=0.2166, simple_loss=0.2919, pruned_loss=0.07069, over 5330712.36 frames. ], batch size: 336, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:50:21,109 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1185477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:50:23,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1185479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:50:24,477 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1185480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:50:48,404 INFO [train.py:968] (0/2) Epoch 27, batch 650, giga_loss[loss=0.2354, simple_loss=0.3068, pruned_loss=0.08202, over 28835.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2953, pruned_loss=0.07212, over 5472530.72 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3437, pruned_loss=0.0908, over 1452568.85 frames. ], giga_tot_loss[loss=0.2158, simple_loss=0.2908, pruned_loss=0.07039, over 5393211.71 frames. ], batch size: 99, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:50:52,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.602e+02 1.003e+03 1.267e+03 1.565e+03 3.705e+03, threshold=2.535e+03, percent-clipped=2.0
+2023-03-13 18:50:52,313 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1185509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:50:58,414 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6171, 3.3323, 1.6547, 1.7765], device='cuda:0'), covar=tensor([0.0973, 0.0309, 0.0897, 0.1293], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0564, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 18:51:20,711 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2265, 1.1927, 3.4664, 3.0605], device='cuda:0'), covar=tensor([0.1652, 0.2919, 0.0517, 0.1489], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0670, 0.1001, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 18:51:31,290 INFO [train.py:968] (0/2) Epoch 27, batch 700, giga_loss[loss=0.2013, simple_loss=0.2729, pruned_loss=0.0648, over 28660.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2916, pruned_loss=0.07036, over 5526141.09 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3435, pruned_loss=0.0907, over 1496110.76 frames. ], giga_tot_loss[loss=0.2127, simple_loss=0.2877, pruned_loss=0.06885, over 5460996.47 frames. ], batch size: 85, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:51:38,388 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 18:51:46,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1649, 2.5116, 2.3295, 2.1428], device='cuda:0'), covar=tensor([0.2422, 0.2269, 0.2195, 0.2402], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0760, 0.0727, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 18:52:17,464 INFO [train.py:968] (0/2) Epoch 27, batch 750, giga_loss[loss=0.1785, simple_loss=0.2559, pruned_loss=0.05055, over 28875.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.288, pruned_loss=0.06812, over 5573674.30 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3405, pruned_loss=0.08868, over 1605113.22 frames. ], giga_tot_loss[loss=0.2091, simple_loss=0.2843, pruned_loss=0.06692, over 5513707.66 frames. ], batch size: 119, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:52:21,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.638e+02 1.057e+03 1.414e+03 1.839e+03 4.282e+03, threshold=2.829e+03, percent-clipped=14.0
+2023-03-13 18:53:00,262 INFO [train.py:968] (0/2) Epoch 27, batch 800, giga_loss[loss=0.2159, simple_loss=0.2845, pruned_loss=0.07363, over 28780.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2881, pruned_loss=0.06872, over 5588277.11 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3411, pruned_loss=0.0889, over 1699340.33 frames. ], giga_tot_loss[loss=0.2092, simple_loss=0.2838, pruned_loss=0.06725, over 5545090.33 frames. ], batch size: 99, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 18:53:47,917 INFO [train.py:968] (0/2) Epoch 27, batch 850, giga_loss[loss=0.2804, simple_loss=0.3596, pruned_loss=0.1006, over 28268.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2986, pruned_loss=0.07444, over 5602461.07 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3399, pruned_loss=0.08829, over 1819995.11 frames. ], giga_tot_loss[loss=0.2202, simple_loss=0.2943, pruned_loss=0.07301, over 5563002.38 frames. ], batch size: 368, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:53:52,190 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.767e+02 1.313e+03 1.938e+03 2.634e+03 6.214e+03, threshold=3.877e+03, percent-clipped=23.0
+2023-03-13 18:54:27,794 INFO [train.py:968] (0/2) Epoch 27, batch 900, giga_loss[loss=0.2574, simple_loss=0.3426, pruned_loss=0.08611, over 28390.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3115, pruned_loss=0.08049, over 5606011.97 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3412, pruned_loss=0.08907, over 1989566.41 frames. ], giga_tot_loss[loss=0.2317, simple_loss=0.3061, pruned_loss=0.07863, over 5588378.90 frames. ], batch size: 71, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:54:32,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5495, 1.7629, 1.4640, 1.6287], device='cuda:0'), covar=tensor([0.2790, 0.2715, 0.2940, 0.2424], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1141, 0.1402, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 18:54:38,955 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3228, 1.6797, 1.6023, 1.1563], device='cuda:0'), covar=tensor([0.1896, 0.2903, 0.1645, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0721, 0.0985, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 18:55:10,277 INFO [train.py:968] (0/2) Epoch 27, batch 950, giga_loss[loss=0.3242, simple_loss=0.3773, pruned_loss=0.1356, over 26699.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.322, pruned_loss=0.08538, over 5636003.98 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3411, pruned_loss=0.08897, over 2143368.07 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.3168, pruned_loss=0.08371, over 5611420.02 frames. ], batch size: 555, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:55:14,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.330e+03 1.538e+03 2.070e+03 3.978e+03, threshold=3.076e+03, percent-clipped=1.0
+2023-03-13 18:55:51,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1185854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:55:51,455 INFO [train.py:968] (0/2) Epoch 27, batch 1000, giga_loss[loss=0.2745, simple_loss=0.3518, pruned_loss=0.09853, over 28829.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3292, pruned_loss=0.08815, over 5653556.44 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.34, pruned_loss=0.08814, over 2234629.24 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3252, pruned_loss=0.0871, over 5629764.07 frames. ], batch size: 119, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:56:29,707 INFO [train.py:968] (0/2) Epoch 27, batch 1050, giga_loss[loss=0.2412, simple_loss=0.3287, pruned_loss=0.07686, over 28939.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3333, pruned_loss=0.08892, over 5669486.96 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3405, pruned_loss=0.08841, over 2355879.25 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3296, pruned_loss=0.088, over 5647422.01 frames. ], batch size: 164, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:56:34,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.759e+02 1.269e+03 1.601e+03 2.049e+03 6.638e+03, threshold=3.201e+03, percent-clipped=8.0
+2023-03-13 18:57:14,022 INFO [train.py:968] (0/2) Epoch 27, batch 1100, giga_loss[loss=0.2509, simple_loss=0.3409, pruned_loss=0.08047, over 28648.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3354, pruned_loss=0.0891, over 5658965.64 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3419, pruned_loss=0.0893, over 2450143.80 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3318, pruned_loss=0.08804, over 5645046.59 frames. ], batch size: 242, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:57:27,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1185971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:57:51,447 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1185997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:57:53,156 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1186000.pt
+2023-03-13 18:57:54,647 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1186000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:57:57,656 INFO [train.py:968] (0/2) Epoch 27, batch 1150, giga_loss[loss=0.2456, simple_loss=0.3316, pruned_loss=0.07977, over 28868.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3375, pruned_loss=0.09046, over 5679816.09 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3416, pruned_loss=0.08932, over 2551276.80 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3345, pruned_loss=0.08959, over 5663357.04 frames. ], batch size: 174, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:58:00,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.603e+02 1.213e+03 1.384e+03 1.771e+03 4.983e+03, threshold=2.768e+03, percent-clipped=3.0
+2023-03-13 18:58:06,050 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3406, 1.6119, 1.3166, 1.0558], device='cuda:0'), covar=tensor([0.2749, 0.2807, 0.3201, 0.2437], device='cuda:0'), in_proj_covar=tensor([0.1588, 0.1142, 0.1403, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 18:58:20,633 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1186029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:58:41,924 INFO [train.py:968] (0/2) Epoch 27, batch 1200, giga_loss[loss=0.2747, simple_loss=0.3502, pruned_loss=0.09962, over 28625.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3402, pruned_loss=0.09307, over 5667924.24 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3411, pruned_loss=0.08894, over 2598247.22 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3381, pruned_loss=0.09259, over 5654491.56 frames. ], batch size: 242, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 18:58:42,126 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1186055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:59:24,463 INFO [train.py:968] (0/2) Epoch 27, batch 1250, giga_loss[loss=0.2874, simple_loss=0.357, pruned_loss=0.1089, over 28897.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3433, pruned_loss=0.09511, over 5673658.50 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3408, pruned_loss=0.08861, over 2678089.76 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3418, pruned_loss=0.09499, over 5659607.55 frames. ], batch size: 112, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 18:59:28,413 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.392e+02 1.290e+03 1.602e+03 2.212e+03 5.747e+03, threshold=3.203e+03, percent-clipped=12.0
+2023-03-13 19:00:06,731 INFO [train.py:968] (0/2) Epoch 27, batch 1300, giga_loss[loss=0.2758, simple_loss=0.3592, pruned_loss=0.09616, over 28761.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3462, pruned_loss=0.09607, over 5671240.11 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3412, pruned_loss=0.08865, over 2777343.78 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3448, pruned_loss=0.09616, over 5663635.00 frames. ], batch size: 119, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:00:27,716 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 19:00:44,005 INFO [train.py:968] (0/2) Epoch 27, batch 1350, giga_loss[loss=0.2687, simple_loss=0.3513, pruned_loss=0.09309, over 28750.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3468, pruned_loss=0.09528, over 5691516.77 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3406, pruned_loss=0.08864, over 2853680.17 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3462, pruned_loss=0.09551, over 5681382.44 frames. ], batch size: 284, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:00:47,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.484e+02 1.269e+03 1.631e+03 2.136e+03 5.191e+03, threshold=3.263e+03, percent-clipped=5.0
+2023-03-13 19:01:25,251 INFO [train.py:968] (0/2) Epoch 27, batch 1400, giga_loss[loss=0.2824, simple_loss=0.369, pruned_loss=0.09791, over 29076.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3477, pruned_loss=0.09556, over 5688246.07 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3387, pruned_loss=0.08771, over 2926681.24 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3482, pruned_loss=0.09634, over 5677695.84 frames. ], batch size: 155, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:02:06,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1186302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:02:07,869 INFO [train.py:968] (0/2) Epoch 27, batch 1450, giga_loss[loss=0.2683, simple_loss=0.3476, pruned_loss=0.09449, over 28315.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3478, pruned_loss=0.09466, over 5692691.36 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3378, pruned_loss=0.08709, over 2969806.53 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3488, pruned_loss=0.09566, over 5682543.55 frames. ], batch size: 368, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:02:11,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.725e+02 1.291e+03 1.730e+03 2.241e+03 5.684e+03, threshold=3.461e+03, percent-clipped=6.0
+2023-03-13 19:02:38,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1186346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:02:45,299 INFO [train.py:968] (0/2) Epoch 27, batch 1500, giga_loss[loss=0.255, simple_loss=0.3406, pruned_loss=0.08469, over 28434.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3458, pruned_loss=0.09238, over 5705030.84 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3368, pruned_loss=0.08635, over 3085569.11 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3473, pruned_loss=0.09379, over 5689312.73 frames. ], batch size: 85, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:02:45,525 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4986, 1.9613, 1.7302, 1.6688], device='cuda:0'), covar=tensor([0.0799, 0.0315, 0.0318, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 19:02:51,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9136, 1.2793, 1.0901, 0.1804], device='cuda:0'), covar=tensor([0.5416, 0.4064, 0.5705, 0.8074], device='cuda:0'), in_proj_covar=tensor([0.1814, 0.1701, 0.1643, 0.1477], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 19:02:58,857 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-13 19:03:24,336 INFO [train.py:968] (0/2) Epoch 27, batch 1550, giga_loss[loss=0.2483, simple_loss=0.3321, pruned_loss=0.0823, over 28676.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3441, pruned_loss=0.09079, over 5706821.39 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3371, pruned_loss=0.08633, over 3141666.66 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3454, pruned_loss=0.092, over 5690844.74 frames. ], batch size: 262, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:03:29,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.819e+02 1.131e+03 1.415e+03 1.784e+03 4.761e+03, threshold=2.830e+03, percent-clipped=4.0
+2023-03-13 19:03:41,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1186425.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:03:45,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1186430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:04:07,739 INFO [train.py:968] (0/2) Epoch 27, batch 1600, giga_loss[loss=0.3004, simple_loss=0.3624, pruned_loss=0.1192, over 28938.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3457, pruned_loss=0.09296, over 5700396.21 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3371, pruned_loss=0.08634, over 3190826.58 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3469, pruned_loss=0.09404, over 5699417.08 frames. ], batch size: 112, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:04:33,827 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1186489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:04:38,366 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1186492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:04:49,787 INFO [train.py:968] (0/2) Epoch 27, batch 1650, libri_loss[loss=0.2301, simple_loss=0.3108, pruned_loss=0.0747, over 29488.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3482, pruned_loss=0.09729, over 5707992.49 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3368, pruned_loss=0.0866, over 3269884.35 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3496, pruned_loss=0.09827, over 5703052.07 frames. ], batch size: 70, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:04:53,206 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.960e+02 1.333e+03 1.845e+03 2.292e+03 4.898e+03, threshold=3.691e+03, percent-clipped=12.0
+2023-03-13 19:05:04,601 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1186521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:05:06,636 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1186524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:05:18,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1783, 5.0397, 2.3389, 2.4382], device='cuda:0'), covar=tensor([0.0918, 0.0193, 0.0795, 0.1149], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0563, 0.0404, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 19:05:30,697 INFO [train.py:968] (0/2) Epoch 27, batch 1700, giga_loss[loss=0.2673, simple_loss=0.3322, pruned_loss=0.1012, over 28767.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3494, pruned_loss=0.1, over 5700036.57 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3364, pruned_loss=0.08643, over 3345500.64 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3511, pruned_loss=0.1012, over 5692843.73 frames. ], batch size: 92, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:05:45,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1186573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:05:47,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1186576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:06:12,093 INFO [train.py:968] (0/2) Epoch 27, batch 1750, giga_loss[loss=0.2911, simple_loss=0.3622, pruned_loss=0.11, over 28888.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3486, pruned_loss=0.1004, over 5697012.13 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3367, pruned_loss=0.08654, over 3393629.27 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.35, pruned_loss=0.1016, over 5690035.17 frames. ], batch size: 145, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:06:12,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1186605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:06:16,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.415e+02 1.412e+03 1.864e+03 2.573e+03 6.782e+03, threshold=3.727e+03, percent-clipped=10.0
+2023-03-13 19:06:48,308 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 19:06:51,609 INFO [train.py:968] (0/2) Epoch 27, batch 1800, giga_loss[loss=0.2514, simple_loss=0.3293, pruned_loss=0.08679, over 28937.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3472, pruned_loss=0.09971, over 5710682.41 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3369, pruned_loss=0.08649, over 3443891.17 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3485, pruned_loss=0.1009, over 5701076.03 frames. ], batch size: 106, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:06:55,058 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4140, 1.2638, 3.7930, 3.2943], device='cuda:0'), covar=tensor([0.1617, 0.2855, 0.0456, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0666, 0.0991, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 19:07:10,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1186677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:07:32,916 INFO [train.py:968] (0/2) Epoch 27, batch 1850, giga_loss[loss=0.2734, simple_loss=0.3596, pruned_loss=0.09361, over 28330.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3475, pruned_loss=0.09941, over 5712893.19 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3366, pruned_loss=0.08623, over 3480825.70 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3487, pruned_loss=0.1007, over 5702451.43 frames. ], batch size: 368, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:07:36,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.565e+02 1.494e+03 1.881e+03 2.548e+03 4.680e+03, threshold=3.762e+03, percent-clipped=3.0
+2023-03-13 19:08:10,397 INFO [train.py:968] (0/2) Epoch 27, batch 1900, giga_loss[loss=0.2898, simple_loss=0.3514, pruned_loss=0.1141, over 26558.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3456, pruned_loss=0.09758, over 5713937.78 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.337, pruned_loss=0.08643, over 3595235.56 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3469, pruned_loss=0.09907, over 5700532.32 frames. ], batch size: 555, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:08:54,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1186800.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:08:58,093 INFO [train.py:968] (0/2) Epoch 27, batch 1950, giga_loss[loss=0.234, simple_loss=0.3126, pruned_loss=0.0777, over 28967.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3412, pruned_loss=0.09452, over 5705308.93 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3364, pruned_loss=0.08624, over 3662912.01 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3427, pruned_loss=0.0961, over 5689911.82 frames. ], batch size: 213, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:09:01,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4541, 1.8857, 1.9121, 1.5660], device='cuda:0'), covar=tensor([0.2098, 0.1699, 0.2008, 0.1999], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0760, 0.0729, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 19:09:02,125 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.910e+02 1.214e+03 1.614e+03 2.211e+03 4.355e+03, threshold=3.229e+03, percent-clipped=4.0
+2023-03-13 19:09:10,333 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1186820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:09:12,407 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1186823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:09:20,239 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-13 19:09:41,082 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1186852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:09:44,254 INFO [train.py:968] (0/2) Epoch 27, batch 2000, libri_loss[loss=0.2745, simple_loss=0.364, pruned_loss=0.09251, over 28475.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3366, pruned_loss=0.09205, over 5699530.67 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3362, pruned_loss=0.0861, over 3717106.86 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.338, pruned_loss=0.09355, over 5684195.27 frames. ], batch size: 106, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:10:23,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1186899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:10:29,328 INFO [train.py:968] (0/2) Epoch 27, batch 2050, giga_loss[loss=0.245, simple_loss=0.3192, pruned_loss=0.08544, over 28710.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3311, pruned_loss=0.08975, over 5691341.49 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.336, pruned_loss=0.08591, over 3749816.12 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3323, pruned_loss=0.09111, over 5676602.22 frames. ], batch size: 242, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:10:34,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.187e+02 1.065e+03 1.342e+03 1.592e+03 6.194e+03, threshold=2.684e+03, percent-clipped=5.0
+2023-03-13 19:11:07,888 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1186943.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:11:09,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1186946.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:11:16,391 INFO [train.py:968] (0/2) Epoch 27, batch 2100, libri_loss[loss=0.2619, simple_loss=0.3514, pruned_loss=0.0862, over 29654.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3278, pruned_loss=0.08757, over 5694463.36 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3363, pruned_loss=0.08584, over 3812920.91 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3284, pruned_loss=0.08878, over 5677407.79 frames. ], batch size: 88, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:11:31,565 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1186975.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:11:57,007 INFO [train.py:968] (0/2) Epoch 27, batch 2150, giga_loss[loss=0.238, simple_loss=0.3211, pruned_loss=0.07752, over 28768.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3288, pruned_loss=0.08766, over 5697650.33 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3367, pruned_loss=0.08616, over 3823529.31 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.329, pruned_loss=0.08843, over 5683274.71 frames. ], batch size: 284, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:12:02,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.804e+02 1.148e+03 1.433e+03 1.930e+03 1.170e+04, threshold=2.867e+03, percent-clipped=11.0
+2023-03-13 19:12:24,976 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1187042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:12:27,129 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-13 19:12:28,437 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1187045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:12:36,419 INFO [train.py:968] (0/2) Epoch 27, batch 2200, libri_loss[loss=0.2472, simple_loss=0.3345, pruned_loss=0.07994, over 29575.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3285, pruned_loss=0.08705, over 5704411.27 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3372, pruned_loss=0.08631, over 3875294.50 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3281, pruned_loss=0.08761, over 5688001.46 frames. ], batch size: 74, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:12:51,466 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1187074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:13:05,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2466, 1.4419, 1.3192, 1.1884], device='cuda:0'), covar=tensor([0.2945, 0.2776, 0.1852, 0.2620], device='cuda:0'), in_proj_covar=tensor([0.2018, 0.1962, 0.1885, 0.2025], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 19:13:17,970 INFO [train.py:968] (0/2) Epoch 27, batch 2250, giga_loss[loss=0.233, simple_loss=0.3097, pruned_loss=0.07813, over 28916.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.327, pruned_loss=0.0866, over 5708704.08 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3371, pruned_loss=0.08616, over 3895614.55 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3267, pruned_loss=0.08712, over 5694126.22 frames. ], batch size: 112, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:13:24,313 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.873e+02 1.038e+03 1.263e+03 1.630e+03 3.451e+03, threshold=2.525e+03, percent-clipped=1.0
+2023-03-13 19:13:25,199 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2339, 3.0750, 2.9148, 1.3875], device='cuda:0'), covar=tensor([0.0995, 0.1094, 0.0857, 0.2580], device='cuda:0'), in_proj_covar=tensor([0.1270, 0.1176, 0.0992, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 19:14:00,900 INFO [train.py:968] (0/2) Epoch 27, batch 2300, giga_loss[loss=0.2493, simple_loss=0.3233, pruned_loss=0.08763, over 29016.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3266, pruned_loss=0.08688, over 5701150.63 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3378, pruned_loss=0.08644, over 3936797.59 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3256, pruned_loss=0.08711, over 5693510.34 frames. ], batch size: 164, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:14:42,149 INFO [train.py:968] (0/2) Epoch 27, batch 2350, libri_loss[loss=0.2847, simple_loss=0.3807, pruned_loss=0.09432, over 29767.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3242, pruned_loss=0.08557, over 5715070.05 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3377, pruned_loss=0.08625, over 3974832.35 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3232, pruned_loss=0.08586, over 5706330.71 frames. ], batch size: 87, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:14:47,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.385e+02 1.127e+03 1.363e+03 1.809e+03 4.048e+03, threshold=2.726e+03, percent-clipped=6.0
+2023-03-13 19:15:23,781 INFO [train.py:968] (0/2) Epoch 27, batch 2400, giga_loss[loss=0.2102, simple_loss=0.2968, pruned_loss=0.06186, over 28863.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3214, pruned_loss=0.08474, over 5720213.16 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3376, pruned_loss=0.08617, over 3984536.19 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3206, pruned_loss=0.085, over 5712492.18 frames. ], batch size: 174, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:15:25,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6706, 1.7012, 1.9006, 1.4691], device='cuda:0'), covar=tensor([0.2085, 0.2523, 0.1617, 0.1870], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0716, 0.0979, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 19:15:55,101 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-13 19:15:59,300 INFO [train.py:968] (0/2) Epoch 27, batch 2450, giga_loss[loss=0.2144, simple_loss=0.2845, pruned_loss=0.07208, over 28797.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.32, pruned_loss=0.08376, over 5728851.77 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3389, pruned_loss=0.08646, over 4049421.60 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3181, pruned_loss=0.08373, over 5718374.91 frames. ], batch size: 99, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:16:04,803 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.998e+02 1.192e+03 1.522e+03 1.982e+03 5.230e+03, threshold=3.044e+03, percent-clipped=13.0
+2023-03-13 19:16:09,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1187318.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:16:35,584 INFO [train.py:968] (0/2) Epoch 27, batch 2500, giga_loss[loss=0.2158, simple_loss=0.3009, pruned_loss=0.06531, over 28943.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3179, pruned_loss=0.0822, over 5723616.98 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3394, pruned_loss=0.08635, over 4094419.59 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.3156, pruned_loss=0.08216, over 5719176.27 frames. ], batch size: 164, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:17:00,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2566, 0.7946, 0.8713, 1.4035], device='cuda:0'), covar=tensor([0.0796, 0.0394, 0.0380, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 19:17:00,942 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9828, 1.2976, 1.1182, 0.2849], device='cuda:0'), covar=tensor([0.4333, 0.3368, 0.5384, 0.7213], device='cuda:0'), in_proj_covar=tensor([0.1810, 0.1697, 0.1638, 0.1477], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 19:17:05,959 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1187392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:17:14,539 INFO [train.py:968] (0/2) Epoch 27, batch 2550, giga_loss[loss=0.217, simple_loss=0.2939, pruned_loss=0.07, over 28884.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3156, pruned_loss=0.08115, over 5719579.80 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3394, pruned_loss=0.08619, over 4119710.08 frames. ], giga_tot_loss[loss=0.2379, simple_loss=0.3134, pruned_loss=0.08117, over 5715161.76 frames. ], batch size: 112, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:17:21,523 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.338e+02 1.145e+03 1.311e+03 1.843e+03 4.482e+03, threshold=2.622e+03, percent-clipped=6.0
+2023-03-13 19:17:51,937 INFO [train.py:968] (0/2) Epoch 27, batch 2600, giga_loss[loss=0.2539, simple_loss=0.3239, pruned_loss=0.09194, over 28984.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3164, pruned_loss=0.08124, over 5722068.69 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3409, pruned_loss=0.08645, over 4185947.54 frames. ], giga_tot_loss[loss=0.2372, simple_loss=0.3127, pruned_loss=0.08086, over 5715219.03 frames. ], batch size: 227, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:18:28,540 INFO [train.py:968] (0/2) Epoch 27, batch 2650, giga_loss[loss=0.2361, simple_loss=0.3188, pruned_loss=0.07669, over 28277.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3163, pruned_loss=0.08114, over 5728093.74 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3421, pruned_loss=0.08703, over 4236392.12 frames. ], giga_tot_loss[loss=0.2362, simple_loss=0.3119, pruned_loss=0.08029, over 5718234.98 frames. ], batch size: 368, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:18:35,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.763e+02 1.134e+03 1.341e+03 1.709e+03 3.697e+03, threshold=2.682e+03, percent-clipped=5.0
+2023-03-13 19:19:06,797 INFO [train.py:968] (0/2) Epoch 27, batch 2700, giga_loss[loss=0.2283, simple_loss=0.3119, pruned_loss=0.07235, over 28920.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3172, pruned_loss=0.08186, over 5720507.31 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3418, pruned_loss=0.08663, over 4284111.35 frames. ], giga_tot_loss[loss=0.2379, simple_loss=0.3131, pruned_loss=0.0813, over 5709194.07 frames. ], batch size: 145, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:19:18,138 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-13 19:19:49,210 INFO [train.py:968] (0/2) Epoch 27, batch 2750, giga_loss[loss=0.2741, simple_loss=0.3488, pruned_loss=0.09969, over 28572.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3222, pruned_loss=0.08493, over 5721507.96 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.342, pruned_loss=0.08657, over 4316315.28 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3184, pruned_loss=0.08447, over 5709194.48 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:19:55,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.894e+02 1.244e+03 1.533e+03 1.924e+03 1.216e+04, threshold=3.066e+03, percent-clipped=8.0
+2023-03-13 19:20:31,689 INFO [train.py:968] (0/2) Epoch 27, batch 2800, giga_loss[loss=0.317, simple_loss=0.3838, pruned_loss=0.1251, over 28033.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3297, pruned_loss=0.09018, over 5711922.72 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.342, pruned_loss=0.0865, over 4377657.29 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.326, pruned_loss=0.08985, over 5696392.82 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:20:46,175 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-13 19:21:07,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1187693.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:21:16,630 INFO [train.py:968] (0/2) Epoch 27, batch 2850, giga_loss[loss=0.2473, simple_loss=0.3346, pruned_loss=0.08004, over 28858.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3364, pruned_loss=0.09446, over 5700372.60 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.342, pruned_loss=0.08648, over 4385250.01 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3335, pruned_loss=0.09426, over 5687310.37 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:21:18,234 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1748, 1.4766, 1.4808, 1.2803], device='cuda:0'), covar=tensor([0.1933, 0.1576, 0.2189, 0.1824], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0758, 0.0729, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 19:21:24,579 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.280e+02 1.369e+03 1.719e+03 2.281e+03 4.097e+03, threshold=3.438e+03, percent-clipped=5.0
+2023-03-13 19:21:47,452 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8932, 3.0382, 2.0871, 1.0071], device='cuda:0'), covar=tensor([0.8668, 0.3105, 0.3703, 0.7425], device='cuda:0'), in_proj_covar=tensor([0.1817, 0.1705, 0.1644, 0.1483], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 19:22:04,654 INFO [train.py:968] (0/2) Epoch 27, batch 2900, giga_loss[loss=0.298, simple_loss=0.3842, pruned_loss=0.106, over 28962.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3415, pruned_loss=0.09704, over 5674802.25 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3415, pruned_loss=0.08621, over 4400246.38 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3395, pruned_loss=0.09715, over 5662894.89 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:22:08,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7644, 1.9588, 2.0520, 1.6963], device='cuda:0'), covar=tensor([0.2971, 0.2631, 0.2679, 0.2861], device='cuda:0'), in_proj_covar=tensor([0.2021, 0.1965, 0.1896, 0.2030], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 19:22:13,863 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1187767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:22:34,533 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-13 19:22:43,749 INFO [train.py:968] (0/2) Epoch 27, batch 2950, giga_loss[loss=0.2557, simple_loss=0.3421, pruned_loss=0.08465, over 29067.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3474, pruned_loss=0.09955, over 5687512.63 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.341, pruned_loss=0.08599, over 4422866.46 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3462, pruned_loss=0.09997, over 5674972.23 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:22:52,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.145e+02 1.163e+03 1.511e+03 1.987e+03 3.605e+03, threshold=3.023e+03, percent-clipped=1.0
+2023-03-13 19:22:58,621 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1187821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:23:11,292 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1187836.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:23:14,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1187839.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:23:27,250 INFO [train.py:968] (0/2) Epoch 27, batch 3000, giga_loss[loss=0.2823, simple_loss=0.3522, pruned_loss=0.1062, over 28621.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3511, pruned_loss=0.1012, over 5694432.20 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3405, pruned_loss=0.08558, over 4493271.95 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3509, pruned_loss=0.1025, over 5676605.98 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:23:27,255 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 19:23:34,794 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5309, 1.7802, 1.3053, 1.3591], device='cuda:0'), covar=tensor([0.0996, 0.0413, 0.0970, 0.0934], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0449, 0.0524, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 19:23:36,365 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2096, simple_loss=0.3155, pruned_loss=0.05187, over 944034.00 frames. 
+2023-03-13 19:23:36,366 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 19:23:37,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7149, 1.9892, 1.3914, 1.4983], device='cuda:0'), covar=tensor([0.1021, 0.0540, 0.1005, 0.1084], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0449, 0.0524, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 19:23:45,035 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1187868.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:24:03,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2001, 3.1508, 1.4105, 1.3960], device='cuda:0'), covar=tensor([0.1107, 0.0319, 0.0972, 0.1547], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0562, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 19:24:15,177 INFO [train.py:968] (0/2) Epoch 27, batch 3050, giga_loss[loss=0.2677, simple_loss=0.3471, pruned_loss=0.09411, over 28896.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3489, pruned_loss=0.09938, over 5687187.02 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3401, pruned_loss=0.08542, over 4529583.40 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3492, pruned_loss=0.1009, over 5672086.87 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:24:21,025 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1187910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:24:22,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1187913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:24:23,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.788e+02 1.377e+03 1.677e+03 2.387e+03 6.152e+03, threshold=3.354e+03, percent-clipped=16.0
+2023-03-13 19:24:47,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1187942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:24:49,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-13 19:24:57,006 INFO [train.py:968] (0/2) Epoch 27, batch 3100, libri_loss[loss=0.2666, simple_loss=0.3546, pruned_loss=0.08929, over 28644.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3457, pruned_loss=0.09676, over 5692353.74 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3408, pruned_loss=0.08587, over 4562907.08 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3455, pruned_loss=0.09795, over 5675946.16 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:25:32,254 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1188000.pt
+2023-03-13 19:25:37,082 INFO [train.py:968] (0/2) Epoch 27, batch 3150, giga_loss[loss=0.2685, simple_loss=0.3434, pruned_loss=0.09682, over 28871.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3448, pruned_loss=0.09609, over 5678402.86 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3407, pruned_loss=0.0861, over 4586670.33 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3449, pruned_loss=0.09708, over 5669068.72 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:25:43,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.765e+02 1.246e+03 1.613e+03 1.978e+03 6.863e+03, threshold=3.226e+03, percent-clipped=4.0
+2023-03-13 19:26:00,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3016, 1.4635, 3.1056, 3.0049], device='cuda:0'), covar=tensor([0.1349, 0.2479, 0.0447, 0.1081], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0668, 0.0994, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 19:26:15,179 INFO [train.py:968] (0/2) Epoch 27, batch 3200, giga_loss[loss=0.2513, simple_loss=0.3352, pruned_loss=0.08374, over 28905.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3448, pruned_loss=0.09584, over 5673471.73 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3411, pruned_loss=0.08663, over 4628809.70 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3447, pruned_loss=0.09659, over 5666095.50 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:26:42,745 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-13 19:26:53,439 INFO [train.py:968] (0/2) Epoch 27, batch 3250, giga_loss[loss=0.2673, simple_loss=0.346, pruned_loss=0.09426, over 28882.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3465, pruned_loss=0.09659, over 5680925.78 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.341, pruned_loss=0.08655, over 4654607.64 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3466, pruned_loss=0.09742, over 5670564.01 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:26:55,705 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1188108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:27:00,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.943e+02 1.457e+03 1.788e+03 2.639e+03 7.500e+03, threshold=3.576e+03, percent-clipped=13.0
+2023-03-13 19:27:10,058 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4168, 1.4729, 1.2561, 1.7092], device='cuda:0'), covar=tensor([0.0789, 0.0363, 0.0353, 0.0846], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 19:27:27,638 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1188148.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:27:31,990 INFO [train.py:968] (0/2) Epoch 27, batch 3300, giga_loss[loss=0.3038, simple_loss=0.3781, pruned_loss=0.1147, over 29004.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3486, pruned_loss=0.09779, over 5693241.75 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3404, pruned_loss=0.0862, over 4681251.74 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3492, pruned_loss=0.09895, over 5684031.05 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:28:04,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1188196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:28:11,400 INFO [train.py:968] (0/2) Epoch 27, batch 3350, libri_loss[loss=0.2753, simple_loss=0.3628, pruned_loss=0.09387, over 29523.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.351, pruned_loss=0.1002, over 5691861.64 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3403, pruned_loss=0.08605, over 4715781.75 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3518, pruned_loss=0.1015, over 5680421.42 frames. ], batch size: 83, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:28:18,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.481e+03 2.075e+03 3.072e+03 1.044e+04, threshold=4.150e+03, percent-clipped=17.0
+2023-03-13 19:28:51,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9279, 3.7686, 3.5603, 1.8279], device='cuda:0'), covar=tensor([0.0749, 0.0878, 0.0770, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.1278, 0.1184, 0.0997, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 19:28:53,560 INFO [train.py:968] (0/2) Epoch 27, batch 3400, giga_loss[loss=0.3505, simple_loss=0.4012, pruned_loss=0.1499, over 26722.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1011, over 5686753.34 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3401, pruned_loss=0.08588, over 4727468.11 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3524, pruned_loss=0.1025, over 5676007.64 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:29:32,768 INFO [train.py:968] (0/2) Epoch 27, batch 3450, giga_loss[loss=0.2711, simple_loss=0.3528, pruned_loss=0.09472, over 28787.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3524, pruned_loss=0.1019, over 5685465.45 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3401, pruned_loss=0.08574, over 4756642.46 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3534, pruned_loss=0.1035, over 5672120.44 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:29:39,126 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.090e+02 1.373e+03 1.778e+03 2.426e+03 4.094e+03, threshold=3.555e+03, percent-clipped=0.0
+2023-03-13 19:29:58,554 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1188339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:30:00,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1188342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:30:11,655 INFO [train.py:968] (0/2) Epoch 27, batch 3500, giga_loss[loss=0.2921, simple_loss=0.3752, pruned_loss=0.1045, over 28812.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3529, pruned_loss=0.1016, over 5685702.68 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3411, pruned_loss=0.08639, over 4775016.79 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3533, pruned_loss=0.1027, over 5678743.95 frames. ], batch size: 66, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:30:24,296 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1188371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:30:51,708 INFO [train.py:968] (0/2) Epoch 27, batch 3550, giga_loss[loss=0.2381, simple_loss=0.3196, pruned_loss=0.07837, over 28699.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3522, pruned_loss=0.1002, over 5690962.49 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3407, pruned_loss=0.08622, over 4786290.68 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3529, pruned_loss=0.1013, over 5683637.72 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:31:00,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.787e+02 1.243e+03 1.551e+03 2.456e+03 1.115e+04, threshold=3.101e+03, percent-clipped=6.0
+2023-03-13 19:31:32,914 INFO [train.py:968] (0/2) Epoch 27, batch 3600, giga_loss[loss=0.2988, simple_loss=0.3696, pruned_loss=0.114, over 29003.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3529, pruned_loss=0.0999, over 5688965.34 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3411, pruned_loss=0.08639, over 4808954.12 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3535, pruned_loss=0.1011, over 5685921.94 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:31:54,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1188483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:32:09,080 INFO [train.py:968] (0/2) Epoch 27, batch 3650, giga_loss[loss=0.258, simple_loss=0.3432, pruned_loss=0.08637, over 28247.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3504, pruned_loss=0.09826, over 5703608.29 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3409, pruned_loss=0.0863, over 4851132.99 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3513, pruned_loss=0.09974, over 5694371.63 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:32:10,141 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-13 19:32:17,263 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.876e+02 1.299e+03 1.573e+03 1.991e+03 4.761e+03, threshold=3.146e+03, percent-clipped=6.0
+2023-03-13 19:32:24,770 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1188523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:32:36,235 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1188537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:32:48,030 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3375, 1.5326, 1.2547, 1.5333], device='cuda:0'), covar=tensor([0.0828, 0.0351, 0.0359, 0.0901], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 19:32:50,701 INFO [train.py:968] (0/2) Epoch 27, batch 3700, giga_loss[loss=0.2434, simple_loss=0.3278, pruned_loss=0.07952, over 29073.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3475, pruned_loss=0.09721, over 5691434.15 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.341, pruned_loss=0.08633, over 4857013.98 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3484, pruned_loss=0.0985, over 5689392.07 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:32:56,296 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4825, 2.2408, 1.5547, 0.6410], device='cuda:0'), covar=tensor([0.6149, 0.3386, 0.5068, 0.7488], device='cuda:0'), in_proj_covar=tensor([0.1808, 0.1692, 0.1633, 0.1471], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 19:33:27,245 INFO [train.py:968] (0/2) Epoch 27, batch 3750, libri_loss[loss=0.2897, simple_loss=0.3689, pruned_loss=0.1052, over 29491.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.345, pruned_loss=0.09576, over 5702968.57 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3415, pruned_loss=0.0867, over 4893228.83 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3455, pruned_loss=0.09683, over 5694750.76 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:33:34,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.010e+02 1.068e+03 1.294e+03 1.621e+03 4.046e+03, threshold=2.587e+03, percent-clipped=1.0
+2023-03-13 19:33:42,452 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1188626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:33:44,341 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1188629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:34:07,246 INFO [train.py:968] (0/2) Epoch 27, batch 3800, giga_loss[loss=0.2573, simple_loss=0.3398, pruned_loss=0.08735, over 28965.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3454, pruned_loss=0.0966, over 5692870.51 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.341, pruned_loss=0.08654, over 4914945.32 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3462, pruned_loss=0.09786, over 5690911.05 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:34:09,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1188658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:34:14,712 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1188666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:34:16,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1188669.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:34:39,340 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1188698.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:34:43,752 INFO [train.py:968] (0/2) Epoch 27, batch 3850, giga_loss[loss=0.2783, simple_loss=0.3544, pruned_loss=0.1011, over 28897.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3451, pruned_loss=0.09624, over 5698781.70 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3406, pruned_loss=0.08639, over 4945263.38 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3462, pruned_loss=0.09767, over 5694359.33 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:34:53,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.737e+02 1.214e+03 1.379e+03 1.798e+03 4.874e+03, threshold=2.758e+03, percent-clipped=9.0
+2023-03-13 19:35:25,153 INFO [train.py:968] (0/2) Epoch 27, batch 3900, giga_loss[loss=0.2782, simple_loss=0.3587, pruned_loss=0.09881, over 28296.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3454, pruned_loss=0.09562, over 5706039.12 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3409, pruned_loss=0.08659, over 4959884.98 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3462, pruned_loss=0.09672, over 5699664.97 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:35:34,684 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5110, 2.0840, 1.5984, 0.8156], device='cuda:0'), covar=tensor([0.6821, 0.3071, 0.4354, 0.7261], device='cuda:0'), in_proj_covar=tensor([0.1807, 0.1689, 0.1630, 0.1470], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 19:36:08,461 INFO [train.py:968] (0/2) Epoch 27, batch 3950, giga_loss[loss=0.3115, simple_loss=0.3687, pruned_loss=0.1272, over 26674.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3451, pruned_loss=0.09526, over 5709663.32 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3408, pruned_loss=0.08668, over 4964641.55 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3457, pruned_loss=0.0961, over 5703690.27 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:36:16,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.903e+02 1.037e+03 1.285e+03 1.556e+03 3.606e+03, threshold=2.571e+03, percent-clipped=3.0
+2023-03-13 19:36:36,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6093, 1.7704, 1.8165, 1.3893], device='cuda:0'), covar=tensor([0.2017, 0.2789, 0.1633, 0.1877], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0718, 0.0980, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 19:36:47,726 INFO [train.py:968] (0/2) Epoch 27, batch 4000, giga_loss[loss=0.2718, simple_loss=0.3485, pruned_loss=0.09754, over 28712.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3449, pruned_loss=0.09561, over 5706013.55 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3404, pruned_loss=0.08644, over 4987426.34 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3458, pruned_loss=0.09665, over 5697365.47 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:37:12,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5826, 1.8158, 1.8056, 1.3984], device='cuda:0'), covar=tensor([0.1801, 0.2569, 0.1479, 0.1790], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0718, 0.0981, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 19:37:28,131 INFO [train.py:968] (0/2) Epoch 27, batch 4050, giga_loss[loss=0.2737, simple_loss=0.3509, pruned_loss=0.09827, over 28358.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3424, pruned_loss=0.09439, over 5715468.61 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.34, pruned_loss=0.08628, over 5008728.36 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3436, pruned_loss=0.09557, over 5705429.11 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:37:33,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1188912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:37:35,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.617e+02 1.130e+03 1.379e+03 1.931e+03 4.587e+03, threshold=2.757e+03, percent-clipped=9.0
+2023-03-13 19:37:58,759 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-13 19:38:05,564 INFO [train.py:968] (0/2) Epoch 27, batch 4100, giga_loss[loss=0.2334, simple_loss=0.3108, pruned_loss=0.07797, over 29099.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3395, pruned_loss=0.09277, over 5715962.98 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.34, pruned_loss=0.08621, over 5028810.92 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3405, pruned_loss=0.09394, over 5705599.92 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:38:18,631 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1188973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:38:43,663 INFO [train.py:968] (0/2) Epoch 27, batch 4150, libri_loss[loss=0.2444, simple_loss=0.3252, pruned_loss=0.08184, over 28487.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3364, pruned_loss=0.09102, over 5719133.00 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3398, pruned_loss=0.0861, over 5049583.94 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3373, pruned_loss=0.09217, over 5707466.19 frames. ], batch size: 63, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:38:44,659 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6725, 2.0272, 1.4225, 1.4489], device='cuda:0'), covar=tensor([0.1098, 0.0634, 0.1081, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0446, 0.0524, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 19:38:53,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 19:38:53,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.467e+02 1.182e+03 1.569e+03 1.961e+03 4.815e+03, threshold=3.138e+03, percent-clipped=8.0
+2023-03-13 19:39:23,393 INFO [train.py:968] (0/2) Epoch 27, batch 4200, giga_loss[loss=0.2235, simple_loss=0.3074, pruned_loss=0.06979, over 28550.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3359, pruned_loss=0.09089, over 5717080.54 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3404, pruned_loss=0.08652, over 5063216.91 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3361, pruned_loss=0.09157, over 5707321.27 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:39:23,708 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1189055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:39:25,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1189058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:39:49,837 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1189087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:39:51,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6805, 1.8980, 1.2924, 1.5043], device='cuda:0'), covar=tensor([0.0983, 0.0672, 0.1107, 0.1206], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0447, 0.0526, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 19:40:03,043 INFO [train.py:968] (0/2) Epoch 27, batch 4250, giga_loss[loss=0.2627, simple_loss=0.3383, pruned_loss=0.09352, over 28661.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3343, pruned_loss=0.09048, over 5717294.63 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3403, pruned_loss=0.08648, over 5070537.07 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3344, pruned_loss=0.09108, over 5708832.17 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:40:13,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.615e+02 1.263e+03 1.601e+03 2.033e+03 5.240e+03, threshold=3.203e+03, percent-clipped=7.0
+2023-03-13 19:40:42,297 INFO [train.py:968] (0/2) Epoch 27, batch 4300, giga_loss[loss=0.2785, simple_loss=0.3474, pruned_loss=0.1048, over 28933.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3338, pruned_loss=0.09082, over 5714618.38 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3408, pruned_loss=0.08673, over 5092858.61 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3333, pruned_loss=0.09123, over 5704762.03 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:41:20,047 INFO [train.py:968] (0/2) Epoch 27, batch 4350, giga_loss[loss=0.2391, simple_loss=0.3133, pruned_loss=0.08248, over 28816.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3318, pruned_loss=0.08997, over 5720198.23 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3406, pruned_loss=0.08658, over 5120808.95 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3314, pruned_loss=0.09053, over 5706619.72 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:41:25,206 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-13 19:41:28,793 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.102e+02 1.161e+03 1.324e+03 1.915e+03 6.659e+03, threshold=2.648e+03, percent-clipped=3.0
+2023-03-13 19:41:56,826 INFO [train.py:968] (0/2) Epoch 27, batch 4400, giga_loss[loss=0.2599, simple_loss=0.3322, pruned_loss=0.09381, over 28859.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.331, pruned_loss=0.08986, over 5715366.66 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3407, pruned_loss=0.08678, over 5142778.61 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3303, pruned_loss=0.09023, over 5700553.26 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:42:03,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7139, 1.8035, 1.8812, 1.4805], device='cuda:0'), covar=tensor([0.2000, 0.2622, 0.1614, 0.1795], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0718, 0.0982, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 19:42:07,749 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3729, 1.6202, 1.7079, 1.3669], device='cuda:0'), covar=tensor([0.2416, 0.2297, 0.2583, 0.2537], device='cuda:0'), in_proj_covar=tensor([0.0506, 0.0759, 0.0727, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 19:42:37,401 INFO [train.py:968] (0/2) Epoch 27, batch 4450, giga_loss[loss=0.2375, simple_loss=0.3128, pruned_loss=0.0811, over 28535.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3321, pruned_loss=0.09021, over 5712691.08 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3408, pruned_loss=0.08675, over 5144555.72 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3314, pruned_loss=0.09053, over 5702142.63 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:42:38,292 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1189306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:42:48,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.750e+02 1.127e+03 1.330e+03 1.650e+03 4.628e+03, threshold=2.660e+03, percent-clipped=4.0
+2023-03-13 19:42:49,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3643, 1.6065, 1.6180, 1.2155], device='cuda:0'), covar=tensor([0.1947, 0.2700, 0.1658, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0717, 0.0980, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 19:43:11,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4956, 1.8208, 1.6565, 1.4918], device='cuda:0'), covar=tensor([0.2153, 0.2392, 0.2182, 0.2444], device='cuda:0'), in_proj_covar=tensor([0.0505, 0.0758, 0.0726, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 19:43:13,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1189348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:43:17,612 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1189350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:43:21,445 INFO [train.py:968] (0/2) Epoch 27, batch 4500, libri_loss[loss=0.2247, simple_loss=0.3055, pruned_loss=0.07194, over 29654.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3332, pruned_loss=0.09064, over 5716375.31 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3403, pruned_loss=0.0866, over 5156017.80 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.333, pruned_loss=0.09108, over 5705379.87 frames. ], batch size: 73, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:44:03,124 INFO [train.py:968] (0/2) Epoch 27, batch 4550, giga_loss[loss=0.2411, simple_loss=0.3153, pruned_loss=0.08343, over 28739.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3354, pruned_loss=0.0908, over 5720358.82 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3403, pruned_loss=0.08664, over 5159771.71 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3352, pruned_loss=0.09112, over 5710941.72 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:44:14,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.418e+02 1.190e+03 1.399e+03 1.834e+03 3.307e+03, threshold=2.797e+03, percent-clipped=7.0
+2023-03-13 19:44:49,774 INFO [train.py:968] (0/2) Epoch 27, batch 4600, giga_loss[loss=0.2356, simple_loss=0.3065, pruned_loss=0.08233, over 28369.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3376, pruned_loss=0.09168, over 5707626.62 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3404, pruned_loss=0.08667, over 5167190.59 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3374, pruned_loss=0.09196, over 5698526.13 frames. ], batch size: 65, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:45:21,104 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1189491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:45:23,121 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1189494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:45:30,897 INFO [train.py:968] (0/2) Epoch 27, batch 4650, giga_loss[loss=0.3072, simple_loss=0.3703, pruned_loss=0.122, over 27494.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3368, pruned_loss=0.0907, over 5702636.26 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3405, pruned_loss=0.08674, over 5176879.41 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3365, pruned_loss=0.09092, over 5694085.48 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:45:31,936 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 19:45:35,466 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 19:45:41,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.722e+02 1.125e+03 1.367e+03 1.704e+03 5.579e+03, threshold=2.734e+03, percent-clipped=2.0
+2023-03-13 19:45:46,665 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1189523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:46:10,976 INFO [train.py:968] (0/2) Epoch 27, batch 4700, giga_loss[loss=0.3066, simple_loss=0.3795, pruned_loss=0.1168, over 28517.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3376, pruned_loss=0.0906, over 5701730.35 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3403, pruned_loss=0.08664, over 5192519.68 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3374, pruned_loss=0.09099, over 5696864.03 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:46:30,933 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-13 19:46:41,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5355, 2.1211, 1.6412, 0.8880], device='cuda:0'), covar=tensor([0.7243, 0.3220, 0.4163, 0.7205], device='cuda:0'), in_proj_covar=tensor([0.1806, 0.1690, 0.1633, 0.1469], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 19:46:41,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-13 19:46:49,485 INFO [train.py:968] (0/2) Epoch 27, batch 4750, giga_loss[loss=0.2662, simple_loss=0.3409, pruned_loss=0.09575, over 28801.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3386, pruned_loss=0.09139, over 5707353.32 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3403, pruned_loss=0.08666, over 5202913.21 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3384, pruned_loss=0.09176, over 5703329.38 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:46:50,332 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3226, 1.4451, 1.4864, 1.2953], device='cuda:0'), covar=tensor([0.3055, 0.2669, 0.1960, 0.2613], device='cuda:0'), in_proj_covar=tensor([0.2025, 0.1975, 0.1900, 0.2034], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 19:46:55,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8373, 4.6721, 4.4411, 2.0400], device='cuda:0'), covar=tensor([0.0611, 0.0759, 0.0734, 0.2014], device='cuda:0'), in_proj_covar=tensor([0.1269, 0.1174, 0.0990, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 19:47:00,603 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.590e+02 1.292e+03 1.653e+03 2.099e+03 6.396e+03, threshold=3.306e+03, percent-clipped=7.0
+2023-03-13 19:47:28,333 INFO [train.py:968] (0/2) Epoch 27, batch 4800, giga_loss[loss=0.2566, simple_loss=0.3401, pruned_loss=0.08658, over 28661.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3388, pruned_loss=0.09188, over 5713231.59 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3406, pruned_loss=0.08697, over 5217143.25 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3384, pruned_loss=0.09202, over 5706366.72 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:47:52,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2234, 2.6264, 1.2232, 1.4678], device='cuda:0'), covar=tensor([0.0997, 0.0457, 0.0935, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0557, 0.0401, 0.0437], device='cuda:0'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 19:47:53,035 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1189681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:48:11,864 INFO [train.py:968] (0/2) Epoch 27, batch 4850, giga_loss[loss=0.2803, simple_loss=0.3577, pruned_loss=0.1015, over 28782.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3429, pruned_loss=0.09499, over 5713458.05 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3408, pruned_loss=0.08711, over 5223656.57 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3424, pruned_loss=0.09505, over 5706636.33 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:48:22,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.874e+02 1.368e+03 1.719e+03 2.334e+03 4.662e+03, threshold=3.437e+03, percent-clipped=7.0
+2023-03-13 19:48:28,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1189725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:48:51,425 INFO [train.py:968] (0/2) Epoch 27, batch 4900, giga_loss[loss=0.2589, simple_loss=0.3421, pruned_loss=0.0879, over 28956.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3453, pruned_loss=0.09618, over 5712834.03 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.341, pruned_loss=0.08723, over 5230294.48 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3447, pruned_loss=0.09621, over 5705841.32 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:49:29,891 INFO [train.py:968] (0/2) Epoch 27, batch 4950, giga_loss[loss=0.2547, simple_loss=0.3382, pruned_loss=0.08564, over 28997.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.346, pruned_loss=0.09584, over 5714189.49 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3412, pruned_loss=0.0871, over 5248524.61 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3456, pruned_loss=0.0962, over 5705000.05 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:49:40,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.241e+02 1.277e+03 1.471e+03 1.707e+03 3.037e+03, threshold=2.942e+03, percent-clipped=0.0
+2023-03-13 19:49:45,315 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1189824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:49:47,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1189827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:50:08,538 INFO [train.py:968] (0/2) Epoch 27, batch 5000, giga_loss[loss=0.3236, simple_loss=0.3989, pruned_loss=0.1242, over 28942.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3484, pruned_loss=0.09755, over 5713353.25 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3411, pruned_loss=0.08691, over 5261456.75 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3483, pruned_loss=0.09818, over 5702789.78 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:50:09,490 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1189856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:50:18,517 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1189868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:50:20,561 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1189871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:50:41,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4072, 1.4538, 3.8066, 3.2510], device='cuda:0'), covar=tensor([0.1856, 0.2834, 0.0827, 0.1304], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0666, 0.0990, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 19:50:43,915 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1189900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:50:48,107 INFO [train.py:968] (0/2) Epoch 27, batch 5050, giga_loss[loss=0.3063, simple_loss=0.3731, pruned_loss=0.1197, over 28940.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3495, pruned_loss=0.09857, over 5710424.71 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3414, pruned_loss=0.08716, over 5267931.83 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3492, pruned_loss=0.09898, over 5700330.54 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:50:51,308 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 19:50:57,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.389e+03 1.698e+03 2.127e+03 4.498e+03, threshold=3.397e+03, percent-clipped=10.0
+2023-03-13 19:51:30,786 INFO [train.py:968] (0/2) Epoch 27, batch 5100, giga_loss[loss=0.2634, simple_loss=0.3389, pruned_loss=0.09392, over 28747.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3482, pruned_loss=0.09806, over 5712234.82 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3414, pruned_loss=0.08716, over 5267931.83 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.348, pruned_loss=0.09837, over 5704378.40 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:52:07,512 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1190000.pt
+2023-03-13 19:52:11,102 INFO [train.py:968] (0/2) Epoch 27, batch 5150, giga_loss[loss=0.257, simple_loss=0.3305, pruned_loss=0.09169, over 28920.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3446, pruned_loss=0.09645, over 5708167.59 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3414, pruned_loss=0.0872, over 5274050.34 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3444, pruned_loss=0.09675, over 5700595.12 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:52:22,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.030e+02 1.213e+03 1.535e+03 2.002e+03 3.866e+03, threshold=3.070e+03, percent-clipped=3.0
+2023-03-13 19:52:54,619 INFO [train.py:968] (0/2) Epoch 27, batch 5200, giga_loss[loss=0.2493, simple_loss=0.3318, pruned_loss=0.08338, over 28878.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3418, pruned_loss=0.09501, over 5709883.74 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3416, pruned_loss=0.0872, over 5276376.42 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3416, pruned_loss=0.09533, over 5705511.80 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:52:57,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4369, 1.6592, 1.3454, 1.5638], device='cuda:0'), covar=tensor([0.0762, 0.0310, 0.0359, 0.0923], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 19:53:06,586 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-13 19:53:31,889 INFO [train.py:968] (0/2) Epoch 27, batch 5250, giga_loss[loss=0.2507, simple_loss=0.3361, pruned_loss=0.08264, over 29020.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3427, pruned_loss=0.09454, over 5710877.44 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3419, pruned_loss=0.0876, over 5288962.28 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3423, pruned_loss=0.0946, over 5703865.21 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:53:41,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.679e+02 1.292e+03 1.526e+03 2.060e+03 6.194e+03, threshold=3.052e+03, percent-clipped=11.0
+2023-03-13 19:53:43,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-13 19:54:11,318 INFO [train.py:968] (0/2) Epoch 27, batch 5300, giga_loss[loss=0.2639, simple_loss=0.3496, pruned_loss=0.08911, over 28018.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3449, pruned_loss=0.09436, over 5715426.33 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3419, pruned_loss=0.08755, over 5304071.90 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3446, pruned_loss=0.09462, over 5705819.99 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:54:13,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4090, 1.7430, 1.3559, 1.4022], device='cuda:0'), covar=tensor([0.2799, 0.2883, 0.3371, 0.2498], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1136, 0.1393, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 19:54:49,003 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1190204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:54:49,489 INFO [train.py:968] (0/2) Epoch 27, batch 5350, giga_loss[loss=0.2487, simple_loss=0.3291, pruned_loss=0.08419, over 28935.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3447, pruned_loss=0.09449, over 5724476.25 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3421, pruned_loss=0.08774, over 5325341.22 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3444, pruned_loss=0.09478, over 5710313.17 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:54:59,685 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.439e+02 1.213e+03 1.519e+03 1.901e+03 4.640e+03, threshold=3.038e+03, percent-clipped=6.0
+2023-03-13 19:55:27,129 INFO [train.py:968] (0/2) Epoch 27, batch 5400, giga_loss[loss=0.2529, simple_loss=0.3307, pruned_loss=0.0876, over 28609.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3422, pruned_loss=0.09423, over 5732154.70 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3422, pruned_loss=0.08787, over 5345182.52 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.342, pruned_loss=0.09455, over 5715245.73 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:56:06,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5901, 1.7909, 1.2361, 1.3418], device='cuda:0'), covar=tensor([0.0988, 0.0604, 0.1119, 0.1188], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0448, 0.0524, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 19:56:07,236 INFO [train.py:968] (0/2) Epoch 27, batch 5450, giga_loss[loss=0.2635, simple_loss=0.3362, pruned_loss=0.09538, over 28521.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3413, pruned_loss=0.09518, over 5734640.41 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3423, pruned_loss=0.08803, over 5351953.94 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.341, pruned_loss=0.0954, over 5720536.24 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:56:13,396 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 19:56:17,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.750e+02 1.349e+03 1.613e+03 2.061e+03 5.675e+03, threshold=3.227e+03, percent-clipped=11.0
+2023-03-13 19:56:46,311 INFO [train.py:968] (0/2) Epoch 27, batch 5500, giga_loss[loss=0.2502, simple_loss=0.3213, pruned_loss=0.08951, over 29010.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3396, pruned_loss=0.09507, over 5738963.74 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3428, pruned_loss=0.08827, over 5370720.59 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3388, pruned_loss=0.09529, over 5722608.82 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:57:29,165 INFO [train.py:968] (0/2) Epoch 27, batch 5550, giga_loss[loss=0.3575, simple_loss=0.3986, pruned_loss=0.1582, over 26826.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3382, pruned_loss=0.09478, over 5722107.37 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.343, pruned_loss=0.08846, over 5363112.98 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3374, pruned_loss=0.0948, over 5718687.67 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:57:29,374 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1190405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:57:32,843 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-13 19:57:39,435 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 19:57:39,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.439e+02 1.242e+03 1.612e+03 1.933e+03 5.520e+03, threshold=3.224e+03, percent-clipped=6.0
+2023-03-13 19:57:47,066 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1190425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:57:51,339 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9057, 1.2603, 2.8673, 2.8140], device='cuda:0'), covar=tensor([0.1644, 0.2574, 0.0562, 0.0951], device='cuda:0'), in_proj_covar=tensor([0.0792, 0.0664, 0.0987, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 19:58:10,755 INFO [train.py:968] (0/2) Epoch 27, batch 5600, giga_loss[loss=0.2616, simple_loss=0.3228, pruned_loss=0.1002, over 28791.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3364, pruned_loss=0.09425, over 5716152.78 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.08886, over 5371302.67 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3352, pruned_loss=0.09401, over 5710871.59 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:58:43,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3930, 1.6719, 1.4134, 1.5297], device='cuda:0'), covar=tensor([0.0709, 0.0385, 0.0350, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 19:58:48,868 INFO [train.py:968] (0/2) Epoch 27, batch 5650, libri_loss[loss=0.2824, simple_loss=0.3673, pruned_loss=0.09871, over 29495.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3328, pruned_loss=0.09248, over 5720252.14 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3435, pruned_loss=0.08898, over 5384346.96 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3316, pruned_loss=0.09227, over 5711692.07 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:59:00,517 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.673e+02 1.346e+03 1.688e+03 2.425e+03 7.822e+03, threshold=3.377e+03, percent-clipped=8.0
+2023-03-13 19:59:25,775 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.7574, 5.6177, 5.2920, 2.7669], device='cuda:0'), covar=tensor([0.0469, 0.0623, 0.0603, 0.1734], device='cuda:0'), in_proj_covar=tensor([0.1273, 0.1178, 0.0993, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 19:59:27,480 INFO [train.py:968] (0/2) Epoch 27, batch 5700, giga_loss[loss=0.2633, simple_loss=0.3372, pruned_loss=0.0947, over 28931.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3289, pruned_loss=0.09024, over 5718214.41 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3436, pruned_loss=0.08902, over 5399654.35 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3276, pruned_loss=0.09011, over 5706554.27 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:59:48,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1190579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:00:05,810 INFO [train.py:968] (0/2) Epoch 27, batch 5750, giga_loss[loss=0.2329, simple_loss=0.3129, pruned_loss=0.07647, over 28800.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3277, pruned_loss=0.08888, over 5721654.06 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3436, pruned_loss=0.08896, over 5413464.27 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3262, pruned_loss=0.08884, over 5708177.58 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:00:16,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.933e+02 1.248e+03 1.524e+03 1.984e+03 4.879e+03, threshold=3.048e+03, percent-clipped=7.0
+2023-03-13 20:00:29,296 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1190636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:00:46,020 INFO [train.py:968] (0/2) Epoch 27, batch 5800, giga_loss[loss=0.2591, simple_loss=0.3375, pruned_loss=0.09038, over 29016.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3314, pruned_loss=0.0907, over 5721135.73 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3432, pruned_loss=0.08887, over 5426221.79 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3301, pruned_loss=0.09078, over 5706837.67 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:01:19,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7090, 2.0337, 1.6427, 1.9032], device='cuda:0'), covar=tensor([0.2695, 0.2719, 0.3192, 0.2534], device='cuda:0'), in_proj_covar=tensor([0.1582, 0.1138, 0.1395, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 20:01:23,655 INFO [train.py:968] (0/2) Epoch 27, batch 5850, giga_loss[loss=0.2564, simple_loss=0.3281, pruned_loss=0.09234, over 28836.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3351, pruned_loss=0.09222, over 5720217.67 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3436, pruned_loss=0.0891, over 5436067.59 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3336, pruned_loss=0.09214, over 5705210.53 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:01:36,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.768e+02 1.236e+03 1.536e+03 1.956e+03 5.505e+03, threshold=3.073e+03, percent-clipped=4.0
+2023-03-13 20:01:37,732 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1190722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:01:39,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1190725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:02:05,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1190754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:02:05,436 INFO [train.py:968] (0/2) Epoch 27, batch 5900, giga_loss[loss=0.2523, simple_loss=0.332, pruned_loss=0.08631, over 28840.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3378, pruned_loss=0.09291, over 5723996.36 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3436, pruned_loss=0.08909, over 5445072.34 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3365, pruned_loss=0.09291, over 5709125.85 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:02:26,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1190780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:02:29,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4562, 1.3604, 4.1110, 3.2832], device='cuda:0'), covar=tensor([0.1640, 0.2767, 0.0416, 0.0954], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0667, 0.0990, 0.0965], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 20:02:40,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1190800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:02:44,337 INFO [train.py:968] (0/2) Epoch 27, batch 5950, giga_loss[loss=0.2556, simple_loss=0.3409, pruned_loss=0.08518, over 28661.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3405, pruned_loss=0.09387, over 5721158.77 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3434, pruned_loss=0.08885, over 5457140.09 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3395, pruned_loss=0.09423, over 5706545.34 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:02:58,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.712e+02 1.299e+03 1.528e+03 2.046e+03 4.470e+03, threshold=3.056e+03, percent-clipped=6.0
+2023-03-13 20:03:06,800 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-13 20:03:30,662 INFO [train.py:968] (0/2) Epoch 27, batch 6000, giga_loss[loss=0.2771, simple_loss=0.3553, pruned_loss=0.09939, over 28961.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3442, pruned_loss=0.09681, over 5710390.01 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3434, pruned_loss=0.08883, over 5461876.26 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3434, pruned_loss=0.09718, over 5696879.48 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:03:30,666 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 20:03:39,123 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2094, simple_loss=0.3169, pruned_loss=0.05094, over 944034.00 frames. 
+2023-03-13 20:03:39,123 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 20:04:24,921 INFO [train.py:968] (0/2) Epoch 27, batch 6050, giga_loss[loss=0.3107, simple_loss=0.3775, pruned_loss=0.122, over 29008.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3502, pruned_loss=0.102, over 5701978.37 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3435, pruned_loss=0.0889, over 5461156.60 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3495, pruned_loss=0.1023, over 5696148.06 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:04:42,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.657e+03 2.158e+03 3.132e+03 5.864e+03, threshold=4.316e+03, percent-clipped=26.0
+2023-03-13 20:04:43,706 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1190923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:04:45,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1190926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:04:57,947 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1190943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:05:00,203 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1190946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:05:09,758 INFO [train.py:968] (0/2) Epoch 27, batch 6100, giga_loss[loss=0.2935, simple_loss=0.3665, pruned_loss=0.1103, over 29086.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3543, pruned_loss=0.1049, over 5704021.29 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3431, pruned_loss=0.08894, over 5477559.99 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3545, pruned_loss=0.1058, over 5693770.87 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:05:10,010 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1190955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:05:10,269 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-13 20:05:27,158 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1190975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:05:53,468 INFO [train.py:968] (0/2) Epoch 27, batch 6150, giga_loss[loss=0.2864, simple_loss=0.3585, pruned_loss=0.1072, over 28867.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3618, pruned_loss=0.1105, over 5701341.60 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3428, pruned_loss=0.08869, over 5489615.94 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3626, pruned_loss=0.112, over 5689340.18 frames. ], batch size: 243, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:05:53,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4980, 3.0862, 1.6182, 1.6926], device='cuda:0'), covar=tensor([0.0811, 0.0348, 0.0734, 0.1084], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0561, 0.0403, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 20:06:00,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1191011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:06:10,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.180e+03 1.890e+03 2.335e+03 3.267e+03 6.930e+03, threshold=4.670e+03, percent-clipped=9.0
+2023-03-13 20:06:21,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2143, 1.4431, 1.4434, 1.2954], device='cuda:0'), covar=tensor([0.1715, 0.1499, 0.2172, 0.1526], device='cuda:0'), in_proj_covar=tensor([0.0506, 0.0760, 0.0730, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 20:06:43,928 INFO [train.py:968] (0/2) Epoch 27, batch 6200, giga_loss[loss=0.3311, simple_loss=0.3807, pruned_loss=0.1407, over 28473.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3679, pruned_loss=0.1159, over 5701302.70 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3428, pruned_loss=0.08869, over 5489615.94 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3686, pruned_loss=0.1171, over 5691961.84 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:06:51,290 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1191063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:07:27,978 INFO [train.py:968] (0/2) Epoch 27, batch 6250, giga_loss[loss=0.3191, simple_loss=0.3882, pruned_loss=0.125, over 28877.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3737, pruned_loss=0.1205, over 5692556.02 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3431, pruned_loss=0.08895, over 5492954.26 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3745, pruned_loss=0.1218, over 5685600.97 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:07:29,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8850, 1.0846, 1.0324, 0.8604], device='cuda:0'), covar=tensor([0.1688, 0.1934, 0.1467, 0.1725], device='cuda:0'), in_proj_covar=tensor([0.2037, 0.1994, 0.1911, 0.2042], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 20:07:44,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.256e+03 1.960e+03 2.523e+03 3.433e+03 5.891e+03, threshold=5.046e+03, percent-clipped=11.0
+2023-03-13 20:08:16,979 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1191154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:08:17,300 INFO [train.py:968] (0/2) Epoch 27, batch 6300, giga_loss[loss=0.2914, simple_loss=0.3676, pruned_loss=0.1076, over 29013.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3757, pruned_loss=0.1229, over 5686837.62 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3426, pruned_loss=0.08863, over 5501087.45 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3775, pruned_loss=0.1248, over 5677432.09 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:08:19,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1191157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:08:45,897 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1191186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:08:45,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3578, 3.3748, 1.4587, 1.5387], device='cuda:0'), covar=tensor([0.1033, 0.0372, 0.0921, 0.1373], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0564, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 20:09:04,613 INFO [train.py:968] (0/2) Epoch 27, batch 6350, giga_loss[loss=0.2747, simple_loss=0.3494, pruned_loss=0.1, over 28520.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.379, pruned_loss=0.1268, over 5666597.30 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3427, pruned_loss=0.08868, over 5505672.19 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.381, pruned_loss=0.129, over 5658996.91 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:09:11,872 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1191211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:09:21,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.885e+03 2.728e+03 3.625e+03 1.124e+04, threshold=5.456e+03, percent-clipped=11.0
+2023-03-13 20:09:56,832 INFO [train.py:968] (0/2) Epoch 27, batch 6400, giga_loss[loss=0.2974, simple_loss=0.3652, pruned_loss=0.1148, over 28807.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.382, pruned_loss=0.1303, over 5662175.68 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3428, pruned_loss=0.0888, over 5509213.06 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3839, pruned_loss=0.1323, over 5654496.21 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:10:45,212 INFO [train.py:968] (0/2) Epoch 27, batch 6450, giga_loss[loss=0.3266, simple_loss=0.3878, pruned_loss=0.1327, over 28477.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3861, pruned_loss=0.1342, over 5647633.49 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3433, pruned_loss=0.08915, over 5514816.73 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3879, pruned_loss=0.1364, over 5638917.74 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:11:01,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.392e+03 1.809e+03 2.269e+03 3.208e+03 7.011e+03, threshold=4.538e+03, percent-clipped=1.0
+2023-03-13 20:11:28,175 INFO [train.py:968] (0/2) Epoch 27, batch 6500, giga_loss[loss=0.3151, simple_loss=0.3831, pruned_loss=0.1235, over 28837.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3863, pruned_loss=0.1344, over 5644780.80 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3437, pruned_loss=0.08958, over 5517542.75 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3889, pruned_loss=0.1374, over 5639953.28 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:12:12,813 INFO [train.py:968] (0/2) Epoch 27, batch 6550, libri_loss[loss=0.2497, simple_loss=0.3442, pruned_loss=0.07762, over 29662.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3861, pruned_loss=0.135, over 5643377.25 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3436, pruned_loss=0.08929, over 5519961.64 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3896, pruned_loss=0.1389, over 5639618.54 frames. ], batch size: 91, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:12:17,705 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1191410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:12:26,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5788, 1.5386, 1.7012, 1.3520], device='cuda:0'), covar=tensor([0.1339, 0.2281, 0.1206, 0.1525], device='cuda:0'), in_proj_covar=tensor([0.0922, 0.0710, 0.0970, 0.0869], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 20:12:29,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.192e+03 2.040e+03 2.570e+03 3.302e+03 1.171e+04, threshold=5.139e+03, percent-clipped=9.0
+2023-03-13 20:12:40,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1191438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:12:56,074 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 20:12:57,027 INFO [train.py:968] (0/2) Epoch 27, batch 6600, giga_loss[loss=0.3489, simple_loss=0.4047, pruned_loss=0.1465, over 28559.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3857, pruned_loss=0.1349, over 5634209.60 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.08964, over 5534602.15 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3897, pruned_loss=0.1396, over 5623132.80 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:13:17,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4810, 1.7770, 1.4767, 1.3470], device='cuda:0'), covar=tensor([0.2119, 0.2087, 0.2202, 0.2075], device='cuda:0'), in_proj_covar=tensor([0.1583, 0.1144, 0.1400, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 20:13:24,713 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0968, 2.2405, 1.8396, 2.5464], device='cuda:0'), covar=tensor([0.2532, 0.2623, 0.2989, 0.2101], device='cuda:0'), in_proj_covar=tensor([0.1585, 0.1144, 0.1401, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 20:13:41,641 INFO [train.py:968] (0/2) Epoch 27, batch 6650, giga_loss[loss=0.3655, simple_loss=0.4162, pruned_loss=0.1575, over 28312.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3849, pruned_loss=0.1329, over 5649569.69 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.344, pruned_loss=0.08947, over 5543786.61 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3893, pruned_loss=0.1378, over 5634713.75 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:13:59,538 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.696e+03 2.255e+03 3.033e+03 1.082e+04, threshold=4.511e+03, percent-clipped=6.0
+2023-03-13 20:14:27,109 INFO [train.py:968] (0/2) Epoch 27, batch 6700, giga_loss[loss=0.3318, simple_loss=0.3924, pruned_loss=0.1356, over 29066.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3841, pruned_loss=0.1316, over 5658333.12 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3435, pruned_loss=0.08899, over 5557736.98 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3896, pruned_loss=0.1376, over 5637587.61 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:14:52,015 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1191581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:14:55,451 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1191584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:14:56,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1191586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:15:12,253 INFO [train.py:968] (0/2) Epoch 27, batch 6750, giga_loss[loss=0.3036, simple_loss=0.3668, pruned_loss=0.1202, over 28595.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3856, pruned_loss=0.1327, over 5633198.17 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3438, pruned_loss=0.08918, over 5556085.01 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3909, pruned_loss=0.1385, over 5619923.54 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:15:19,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1191613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:15:26,652 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1191621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:15:27,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.841e+03 2.389e+03 3.164e+03 6.816e+03, threshold=4.778e+03, percent-clipped=9.0
+2023-03-13 20:16:00,049 INFO [train.py:968] (0/2) Epoch 27, batch 6800, giga_loss[loss=0.3869, simple_loss=0.4201, pruned_loss=0.1768, over 26475.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.382, pruned_loss=0.1292, over 5632912.55 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3438, pruned_loss=0.08927, over 5562376.59 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3872, pruned_loss=0.135, over 5618697.70 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:16:44,132 INFO [train.py:968] (0/2) Epoch 27, batch 6850, giga_loss[loss=0.3128, simple_loss=0.3807, pruned_loss=0.1225, over 28814.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3811, pruned_loss=0.1273, over 5640996.98 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3439, pruned_loss=0.08941, over 5560403.49 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3862, pruned_loss=0.1328, over 5632902.39 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:16:54,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3662, 1.5775, 1.3482, 1.5466], device='cuda:0'), covar=tensor([0.0747, 0.0350, 0.0343, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:0')
+2023-03-13 20:17:00,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.861e+03 2.327e+03 3.554e+03 6.411e+03, threshold=4.655e+03, percent-clipped=8.0
+2023-03-13 20:17:01,929 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1906, 1.2958, 3.3446, 3.1263], device='cuda:0'), covar=tensor([0.1608, 0.2699, 0.0544, 0.1054], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0665, 0.0992, 0.0964], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 20:17:07,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1191729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:17:09,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1191732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:17:34,483 INFO [train.py:968] (0/2) Epoch 27, batch 6900, giga_loss[loss=0.2773, simple_loss=0.3481, pruned_loss=0.1032, over 28467.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3776, pruned_loss=0.1247, over 5645572.21 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3438, pruned_loss=0.08941, over 5563855.87 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.382, pruned_loss=0.1293, over 5637049.80 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:17:39,794 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1191761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:18:06,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1191785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:18:24,347 INFO [train.py:968] (0/2) Epoch 27, batch 6950, giga_loss[loss=0.293, simple_loss=0.3616, pruned_loss=0.1122, over 28524.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3745, pruned_loss=0.1221, over 5644884.30 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3437, pruned_loss=0.08939, over 5568597.26 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3785, pruned_loss=0.1263, over 5634911.49 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:18:38,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.086e+03 1.857e+03 2.538e+03 3.606e+03 9.907e+03, threshold=5.075e+03, percent-clipped=12.0
+2023-03-13 20:19:05,335 INFO [train.py:968] (0/2) Epoch 27, batch 7000, giga_loss[loss=0.2757, simple_loss=0.352, pruned_loss=0.09974, over 28834.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3708, pruned_loss=0.1196, over 5661203.21 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3428, pruned_loss=0.08906, over 5587002.90 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3764, pruned_loss=0.125, over 5639950.77 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:19:26,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4320, 1.5731, 1.6693, 1.2648], device='cuda:0'), covar=tensor([0.1752, 0.2578, 0.1443, 0.1671], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0711, 0.0971, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 20:19:46,818 INFO [train.py:968] (0/2) Epoch 27, batch 7050, giga_loss[loss=0.3388, simple_loss=0.3809, pruned_loss=0.1484, over 26629.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3693, pruned_loss=0.1188, over 5659280.97 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3416, pruned_loss=0.08854, over 5595782.07 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3757, pruned_loss=0.1245, over 5636836.11 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:20:04,511 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.641e+03 2.111e+03 2.733e+03 6.304e+03, threshold=4.221e+03, percent-clipped=3.0
+2023-03-13 20:20:10,908 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1191928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:20:13,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1191931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:20:29,167 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.53 vs. limit=5.0
+2023-03-13 20:20:41,797 INFO [train.py:968] (0/2) Epoch 27, batch 7100, giga_loss[loss=0.3457, simple_loss=0.3992, pruned_loss=0.1461, over 28555.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3696, pruned_loss=0.1188, over 5660584.55 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3416, pruned_loss=0.08848, over 5598935.41 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.375, pruned_loss=0.1237, over 5640676.12 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:20:47,640 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1191960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:20:50,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1191963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:21:22,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1191996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:21:25,564 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1192000.pt
+2023-03-13 20:21:29,358 INFO [train.py:968] (0/2) Epoch 27, batch 7150, libri_loss[loss=0.2569, simple_loss=0.3429, pruned_loss=0.08542, over 29501.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3678, pruned_loss=0.1162, over 5668064.59 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3412, pruned_loss=0.08836, over 5606627.85 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3731, pruned_loss=0.1209, over 5646352.25 frames. ], batch size: 81, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:21:54,113 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.474e+03 2.089e+03 2.816e+03 5.113e+03, threshold=4.177e+03, percent-clipped=5.0
+2023-03-13 20:22:25,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5520, 1.8627, 1.6135, 1.5847], device='cuda:0'), covar=tensor([0.2162, 0.2509, 0.2397, 0.2468], device='cuda:0'), in_proj_covar=tensor([0.0505, 0.0761, 0.0729, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 20:22:26,218 INFO [train.py:968] (0/2) Epoch 27, batch 7200, giga_loss[loss=0.2947, simple_loss=0.369, pruned_loss=0.1102, over 28281.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3689, pruned_loss=0.1145, over 5670872.95 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3411, pruned_loss=0.0883, over 5609375.50 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3734, pruned_loss=0.1184, over 5651906.30 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:22:26,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0300, 2.4255, 2.5071, 1.7924], device='cuda:0'), covar=tensor([0.3151, 0.2066, 0.1876, 0.2694], device='cuda:0'), in_proj_covar=tensor([0.2050, 0.2007, 0.1925, 0.2057], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 20:23:16,052 INFO [train.py:968] (0/2) Epoch 27, batch 7250, giga_loss[loss=0.3369, simple_loss=0.3955, pruned_loss=0.1391, over 28574.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.371, pruned_loss=0.1152, over 5679503.62 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3415, pruned_loss=0.08834, over 5613910.09 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3751, pruned_loss=0.119, over 5661989.85 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:23:37,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.186e+03 1.795e+03 2.275e+03 2.765e+03 6.345e+03, threshold=4.551e+03, percent-clipped=9.0
+2023-03-13 20:23:49,117 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1192139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:23:50,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1192142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:23:53,761 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1192146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:24:02,065 INFO [train.py:968] (0/2) Epoch 27, batch 7300, giga_loss[loss=0.2652, simple_loss=0.3455, pruned_loss=0.09248, over 28818.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3701, pruned_loss=0.1156, over 5673994.39 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3406, pruned_loss=0.08803, over 5622734.10 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3752, pruned_loss=0.1199, over 5653499.16 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:24:18,057 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1192171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:24:49,149 INFO [train.py:968] (0/2) Epoch 27, batch 7350, giga_loss[loss=0.2672, simple_loss=0.3428, pruned_loss=0.0958, over 28657.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3691, pruned_loss=0.1154, over 5682431.61 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3407, pruned_loss=0.08791, over 5628146.07 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3736, pruned_loss=0.1195, over 5662162.96 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:25:09,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.240e+03 1.699e+03 2.067e+03 2.771e+03 6.682e+03, threshold=4.134e+03, percent-clipped=6.0
+2023-03-13 20:25:34,081 INFO [train.py:968] (0/2) Epoch 27, batch 7400, giga_loss[loss=0.2869, simple_loss=0.3531, pruned_loss=0.1104, over 29116.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3681, pruned_loss=0.1162, over 5682253.19 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3404, pruned_loss=0.08779, over 5636543.23 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3727, pruned_loss=0.1203, over 5660107.21 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:26:16,224 INFO [train.py:968] (0/2) Epoch 27, batch 7450, giga_loss[loss=0.2765, simple_loss=0.3515, pruned_loss=0.1008, over 28360.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3675, pruned_loss=0.1165, over 5684376.86 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3402, pruned_loss=0.08766, over 5639232.84 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3722, pruned_loss=0.1207, over 5665223.99 frames. ], batch size: 65, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:26:37,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.211e+03 1.807e+03 2.248e+03 2.808e+03 7.366e+03, threshold=4.496e+03, percent-clipped=5.0
+2023-03-13 20:26:49,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1192338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:27:04,514 INFO [train.py:968] (0/2) Epoch 27, batch 7500, giga_loss[loss=0.2793, simple_loss=0.3343, pruned_loss=0.1122, over 23819.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3666, pruned_loss=0.1147, over 5669085.33 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3406, pruned_loss=0.08785, over 5643218.52 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.371, pruned_loss=0.1189, over 5651164.33 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:27:07,636 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1192359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:27:44,476 INFO [train.py:968] (0/2) Epoch 27, batch 7550, giga_loss[loss=0.3007, simple_loss=0.3719, pruned_loss=0.1147, over 28649.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3658, pruned_loss=0.113, over 5680070.45 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3402, pruned_loss=0.08757, over 5650365.15 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3705, pruned_loss=0.1175, over 5660012.60 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:27:48,368 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.20 vs. limit=5.0
+2023-03-13 20:28:04,469 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.001e+03 1.614e+03 1.901e+03 2.656e+03 4.714e+03, threshold=3.803e+03, percent-clipped=1.0
+2023-03-13 20:28:21,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2102, 4.0612, 3.8391, 1.8130], device='cuda:0'), covar=tensor([0.0661, 0.0815, 0.0823, 0.2092], device='cuda:0'), in_proj_covar=tensor([0.1304, 0.1203, 0.1014, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 20:28:29,387 INFO [train.py:968] (0/2) Epoch 27, batch 7600, giga_loss[loss=0.2644, simple_loss=0.3457, pruned_loss=0.09155, over 29035.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.365, pruned_loss=0.1123, over 5688293.23 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3403, pruned_loss=0.08773, over 5656928.35 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3692, pruned_loss=0.1162, over 5667209.33 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:28:34,776 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-13 20:28:52,822 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1192481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:28:56,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1192484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:29:19,560 INFO [train.py:968] (0/2) Epoch 27, batch 7650, giga_loss[loss=0.2803, simple_loss=0.3496, pruned_loss=0.1055, over 28968.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3643, pruned_loss=0.1123, over 5688470.97 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3404, pruned_loss=0.08774, over 5658969.69 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3677, pruned_loss=0.1156, over 5670500.12 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:29:25,842 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1192513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:29:31,824 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1192521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:29:34,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.655e+03 2.093e+03 2.883e+03 6.027e+03, threshold=4.186e+03, percent-clipped=12.0
+2023-03-13 20:30:05,273 INFO [train.py:968] (0/2) Epoch 27, batch 7700, giga_loss[loss=0.252, simple_loss=0.3263, pruned_loss=0.08881, over 28899.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3621, pruned_loss=0.1118, over 5676683.53 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3405, pruned_loss=0.08782, over 5663730.61 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3652, pruned_loss=0.1149, over 5658828.89 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:30:49,375 INFO [train.py:968] (0/2) Epoch 27, batch 7750, giga_loss[loss=0.2725, simple_loss=0.3445, pruned_loss=0.1003, over 28857.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.361, pruned_loss=0.1118, over 5671925.39 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3406, pruned_loss=0.08777, over 5668895.55 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3639, pruned_loss=0.1148, over 5653407.74 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:31:08,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.798e+03 2.194e+03 2.810e+03 7.554e+03, threshold=4.389e+03, percent-clipped=8.0
+2023-03-13 20:31:13,311 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1197, 3.3318, 1.3230, 1.4964], device='cuda:0'), covar=tensor([0.1219, 0.0474, 0.0975, 0.1493], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0568, 0.0405, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 20:31:34,929 INFO [train.py:968] (0/2) Epoch 27, batch 7800, giga_loss[loss=0.2605, simple_loss=0.3262, pruned_loss=0.09742, over 28906.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3613, pruned_loss=0.1128, over 5670043.19 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3403, pruned_loss=0.08767, over 5668900.83 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3648, pruned_loss=0.1165, over 5654318.04 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:31:44,297 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1192664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:31:48,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1192667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:32:13,568 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1192696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:32:21,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2476, 2.4684, 1.8628, 2.2189], device='cuda:0'), covar=tensor([0.0976, 0.0617, 0.0947, 0.1024], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0451, 0.0524, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 20:32:21,471 INFO [train.py:968] (0/2) Epoch 27, batch 7850, libri_loss[loss=0.2221, simple_loss=0.3066, pruned_loss=0.06884, over 29549.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3595, pruned_loss=0.1119, over 5663589.83 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3402, pruned_loss=0.0875, over 5672280.72 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1157, over 5647661.84 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:32:29,092 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1192716.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:32:36,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.314e+03 1.894e+03 2.332e+03 3.088e+03 7.141e+03, threshold=4.665e+03, percent-clipped=8.0
+2023-03-13 20:32:43,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1192734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:32:49,797 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1192743.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:33:00,456 INFO [train.py:968] (0/2) Epoch 27, batch 7900, giga_loss[loss=0.2612, simple_loss=0.3427, pruned_loss=0.08988, over 29007.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1127, over 5668928.40 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.34, pruned_loss=0.08757, over 5679289.41 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3637, pruned_loss=0.1166, over 5649505.73 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:33:46,805 INFO [train.py:968] (0/2) Epoch 27, batch 7950, giga_loss[loss=0.3016, simple_loss=0.3702, pruned_loss=0.1165, over 28851.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3623, pruned_loss=0.1143, over 5662698.42 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.34, pruned_loss=0.08749, over 5676362.04 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3657, pruned_loss=0.1179, over 5649955.52 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:34:05,365 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.820e+03 2.460e+03 3.436e+03 8.650e+03, threshold=4.920e+03, percent-clipped=8.0
+2023-03-13 20:34:32,198 INFO [train.py:968] (0/2) Epoch 27, batch 8000, giga_loss[loss=0.2972, simple_loss=0.3675, pruned_loss=0.1134, over 28609.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3628, pruned_loss=0.1134, over 5672216.43 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.34, pruned_loss=0.0875, over 5680776.11 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3661, pruned_loss=0.117, over 5657870.18 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:34:49,888 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1192877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:34:52,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1192880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:34:55,641 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0953, 4.9093, 4.6892, 2.5612], device='cuda:0'), covar=tensor([0.0513, 0.0664, 0.0744, 0.1696], device='cuda:0'), in_proj_covar=tensor([0.1311, 0.1207, 0.1017, 0.0758], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 20:35:10,311 INFO [train.py:968] (0/2) Epoch 27, batch 8050, giga_loss[loss=0.2511, simple_loss=0.3365, pruned_loss=0.08283, over 28377.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3619, pruned_loss=0.1116, over 5687695.19 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3398, pruned_loss=0.08754, over 5687992.83 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3655, pruned_loss=0.1154, over 5669246.39 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:35:16,607 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1192909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:35:30,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.382e+02 1.726e+03 2.261e+03 3.409e+03 1.062e+04, threshold=4.523e+03, percent-clipped=12.0
+2023-03-13 20:35:35,224 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4457, 4.7870, 1.6983, 1.9044], device='cuda:0'), covar=tensor([0.1086, 0.0353, 0.0949, 0.1310], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0568, 0.0405, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 20:35:41,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3446, 1.6123, 1.5799, 1.4594], device='cuda:0'), covar=tensor([0.1906, 0.1728, 0.2137, 0.1801], device='cuda:0'), in_proj_covar=tensor([0.0505, 0.0763, 0.0729, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 20:35:55,512 INFO [train.py:968] (0/2) Epoch 27, batch 8100, giga_loss[loss=0.3122, simple_loss=0.3763, pruned_loss=0.1241, over 29027.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3632, pruned_loss=0.1126, over 5692625.56 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3403, pruned_loss=0.0879, over 5693637.80 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3663, pruned_loss=0.116, over 5672764.75 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:36:01,214 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1192960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:36:26,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6435, 1.9658, 1.5833, 1.7173], device='cuda:0'), covar=tensor([0.2596, 0.2694, 0.3154, 0.2491], device='cuda:0'), in_proj_covar=tensor([0.1589, 0.1146, 0.1403, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 20:36:43,461 INFO [train.py:968] (0/2) Epoch 27, batch 8150, giga_loss[loss=0.3011, simple_loss=0.3687, pruned_loss=0.1167, over 28906.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3661, pruned_loss=0.1159, over 5679636.85 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.34, pruned_loss=0.08775, over 5696410.82 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3692, pruned_loss=0.1191, over 5661414.25 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:37:04,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.774e+03 2.273e+03 3.145e+03 7.127e+03, threshold=4.546e+03, percent-clipped=11.0
+2023-03-13 20:37:33,666 INFO [train.py:968] (0/2) Epoch 27, batch 8200, giga_loss[loss=0.2812, simple_loss=0.3504, pruned_loss=0.106, over 28691.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3682, pruned_loss=0.1187, over 5668512.36 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3404, pruned_loss=0.088, over 5697003.76 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3708, pruned_loss=0.1215, over 5653190.67 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:38:08,573 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1193091.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:38:21,688 INFO [train.py:968] (0/2) Epoch 27, batch 8250, giga_loss[loss=0.3311, simple_loss=0.3857, pruned_loss=0.1382, over 28333.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.37, pruned_loss=0.1215, over 5665370.66 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3404, pruned_loss=0.08798, over 5695620.53 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.124, over 5654322.87 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:38:37,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1193118.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 20:38:44,263 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.966e+03 2.464e+03 3.434e+03 8.589e+03, threshold=4.927e+03, percent-clipped=12.0
+2023-03-13 20:38:56,862 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3522, 1.1960, 4.2685, 3.4423], device='cuda:0'), covar=tensor([0.1717, 0.2969, 0.0456, 0.1235], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0672, 0.1002, 0.0974], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 20:39:10,536 INFO [train.py:968] (0/2) Epoch 27, batch 8300, giga_loss[loss=0.3842, simple_loss=0.4195, pruned_loss=0.1744, over 24189.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3718, pruned_loss=0.1234, over 5662922.79 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3409, pruned_loss=0.08828, over 5699530.59 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3741, pruned_loss=0.1261, over 5649396.17 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:39:53,334 INFO [train.py:968] (0/2) Epoch 27, batch 8350, giga_loss[loss=0.3456, simple_loss=0.397, pruned_loss=0.1471, over 27902.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3709, pruned_loss=0.1226, over 5661359.30 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3412, pruned_loss=0.08836, over 5693128.34 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3729, pruned_loss=0.1252, over 5656641.71 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:39:57,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 20:40:05,181 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 20:40:09,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.204e+03 1.793e+03 2.519e+03 3.247e+03 6.406e+03, threshold=5.037e+03, percent-clipped=6.0
+2023-03-13 20:40:16,430 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1193234.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:40:18,307 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1193237.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:40:32,963 INFO [train.py:968] (0/2) Epoch 27, batch 8400, giga_loss[loss=0.2813, simple_loss=0.3613, pruned_loss=0.1006, over 28962.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.37, pruned_loss=0.1211, over 5666058.36 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3413, pruned_loss=0.08853, over 5686487.28 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3721, pruned_loss=0.1238, over 5667558.98 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:40:38,807 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1193261.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:40:41,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1193264.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:40:43,366 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1193266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:40:43,381 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1193266.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 20:41:05,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1193293.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 20:41:17,291 INFO [train.py:968] (0/2) Epoch 27, batch 8450, giga_loss[loss=0.2815, simple_loss=0.3513, pruned_loss=0.1058, over 28892.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3692, pruned_loss=0.1184, over 5675014.22 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3415, pruned_loss=0.0885, over 5687931.02 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3713, pruned_loss=0.1212, over 5674763.32 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:41:33,165 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.623e+03 2.093e+03 3.904e+03 7.574e+03, threshold=4.185e+03, percent-clipped=3.0
+2023-03-13 20:41:40,442 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1193335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:41:44,630 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 20:41:55,784 INFO [train.py:968] (0/2) Epoch 27, batch 8500, giga_loss[loss=0.2926, simple_loss=0.3539, pruned_loss=0.1156, over 28617.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3671, pruned_loss=0.117, over 5673178.25 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3416, pruned_loss=0.08852, over 5692603.53 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3696, pruned_loss=0.1203, over 5668254.19 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:42:17,308 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6766, 1.3904, 1.6976, 1.2611], device='cuda:0'), covar=tensor([0.1968, 0.3290, 0.1552, 0.1770], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0714, 0.0973, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-13 20:42:42,300 INFO [train.py:968] (0/2) Epoch 27, batch 8550, giga_loss[loss=0.2741, simple_loss=0.3435, pruned_loss=0.1023, over 28904.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3643, pruned_loss=0.1159, over 5674540.76 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3415, pruned_loss=0.08851, over 5698158.70 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.367, pruned_loss=0.1191, over 5665466.08 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:43:01,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.726e+03 2.650e+03 3.661e+03 1.176e+04, threshold=5.299e+03, percent-clipped=18.0
+2023-03-13 20:43:29,980 INFO [train.py:968] (0/2) Epoch 27, batch 8600, libri_loss[loss=0.2752, simple_loss=0.3578, pruned_loss=0.09628, over 25800.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3633, pruned_loss=0.1156, over 5674660.33 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3419, pruned_loss=0.08872, over 5699389.35 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3655, pruned_loss=0.1186, over 5665983.00 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:43:53,696 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1193478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:43:54,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8489, 4.6953, 4.4754, 2.3710], device='cuda:0'), covar=tensor([0.0483, 0.0571, 0.0641, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.1307, 0.1201, 0.1015, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 20:43:55,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1193481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:44:18,591 INFO [train.py:968] (0/2) Epoch 27, batch 8650, giga_loss[loss=0.324, simple_loss=0.3908, pruned_loss=0.1286, over 28804.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3665, pruned_loss=0.1176, over 5677718.65 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3419, pruned_loss=0.08872, over 5703053.31 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3687, pruned_loss=0.1205, over 5667049.51 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:44:22,704 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1193510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:44:35,990 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.196e+03 1.622e+03 2.145e+03 3.244e+03 7.346e+03, threshold=4.290e+03, percent-clipped=9.0
+2023-03-13 20:45:01,905 INFO [train.py:968] (0/2) Epoch 27, batch 8700, giga_loss[loss=0.3038, simple_loss=0.3817, pruned_loss=0.1129, over 28830.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3696, pruned_loss=0.1176, over 5683321.76 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.08876, over 5707504.11 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3723, pruned_loss=0.121, over 5669613.38 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:45:46,458 INFO [train.py:968] (0/2) Epoch 27, batch 8750, libri_loss[loss=0.2944, simple_loss=0.3719, pruned_loss=0.1085, over 29379.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3708, pruned_loss=0.1166, over 5679993.35 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3416, pruned_loss=0.08858, over 5710097.73 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3739, pruned_loss=0.12, over 5666005.49 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:46:07,265 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.622e+03 2.147e+03 2.690e+03 8.726e+03, threshold=4.293e+03, percent-clipped=7.0
+2023-03-13 20:46:20,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1193641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:46:32,887 INFO [train.py:968] (0/2) Epoch 27, batch 8800, libri_loss[loss=0.2774, simple_loss=0.3617, pruned_loss=0.09657, over 29514.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.375, pruned_loss=0.12, over 5682190.14 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3417, pruned_loss=0.08862, over 5711061.24 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3776, pruned_loss=0.1229, over 5670139.92 frames. ], batch size: 89, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:46:35,174 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 20:47:16,250 INFO [train.py:968] (0/2) Epoch 27, batch 8850, giga_loss[loss=0.268, simple_loss=0.3427, pruned_loss=0.09661, over 28823.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3751, pruned_loss=0.1204, over 5688881.83 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.08874, over 5713106.54 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3774, pruned_loss=0.1231, over 5677097.77 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:47:35,484 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.345e+03 1.788e+03 2.338e+03 3.218e+03 6.233e+03, threshold=4.676e+03, percent-clipped=7.0
+2023-03-13 20:47:58,320 INFO [train.py:968] (0/2) Epoch 27, batch 8900, giga_loss[loss=0.3144, simple_loss=0.3718, pruned_loss=0.1285, over 28854.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3741, pruned_loss=0.1203, over 5695666.99 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08909, over 5720279.77 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3768, pruned_loss=0.1234, over 5678388.00 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:48:24,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1193784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:48:25,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1193786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:48:26,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1193787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:48:42,940 INFO [train.py:968] (0/2) Epoch 27, batch 8950, giga_loss[loss=0.3325, simple_loss=0.3995, pruned_loss=0.1327, over 28738.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3732, pruned_loss=0.1203, over 5698365.25 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.08921, over 5722981.53 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3758, pruned_loss=0.1234, over 5681506.12 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:48:54,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1193816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:49:04,847 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.244e+03 1.791e+03 2.170e+03 2.725e+03 6.142e+03, threshold=4.340e+03, percent-clipped=2.0
+2023-03-13 20:49:23,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3750, 1.5654, 1.3857, 1.4649], device='cuda:0'), covar=tensor([0.0763, 0.0362, 0.0330, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0121, 0.0120, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0066, 0.0113], device='cuda:0')
+2023-03-13 20:49:24,333 INFO [train.py:968] (0/2) Epoch 27, batch 9000, giga_loss[loss=0.2938, simple_loss=0.3657, pruned_loss=0.1109, over 28780.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3708, pruned_loss=0.1192, over 5696005.62 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3429, pruned_loss=0.08933, over 5722398.77 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3742, pruned_loss=0.1231, over 5681713.40 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:49:24,338 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 20:49:28,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2825, 1.9007, 1.4638, 0.4858], device='cuda:0'), covar=tensor([0.5473, 0.3953, 0.5189, 0.7481], device='cuda:0'), in_proj_covar=tensor([0.1823, 0.1721, 0.1648, 0.1487], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 20:49:31,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3937, 1.2899, 1.1703, 1.5805], device='cuda:0'), covar=tensor([0.0854, 0.0378, 0.0379, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0121, 0.0120, 0.0227], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0066, 0.0113], device='cuda:0')
+2023-03-13 20:49:32,819 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2034, simple_loss=0.3104, pruned_loss=0.04823, over 944034.00 frames. 
+2023-03-13 20:49:32,820 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 20:50:14,506 INFO [train.py:968] (0/2) Epoch 27, batch 9050, giga_loss[loss=0.269, simple_loss=0.3446, pruned_loss=0.09666, over 28614.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3686, pruned_loss=0.1187, over 5676892.27 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08938, over 5717078.62 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3722, pruned_loss=0.1226, over 5669971.78 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:50:36,811 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.770e+03 2.307e+03 3.407e+03 7.960e+03, threshold=4.614e+03, percent-clipped=12.0
+2023-03-13 20:51:04,082 INFO [train.py:968] (0/2) Epoch 27, batch 9100, giga_loss[loss=0.2766, simple_loss=0.3477, pruned_loss=0.1028, over 28976.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3671, pruned_loss=0.1182, over 5682205.19 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3426, pruned_loss=0.08926, over 5719993.90 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3704, pruned_loss=0.1217, over 5673679.44 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:51:48,333 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1194000.pt
+2023-03-13 20:51:53,224 INFO [train.py:968] (0/2) Epoch 27, batch 9150, libri_loss[loss=0.284, simple_loss=0.3659, pruned_loss=0.1011, over 29656.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3681, pruned_loss=0.1193, over 5678433.88 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08937, over 5723788.62 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.371, pruned_loss=0.1226, over 5667190.95 frames. ], batch size: 88, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:52:10,201 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1194024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:52:13,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.224e+03 1.766e+03 2.265e+03 3.135e+03 7.740e+03, threshold=4.530e+03, percent-clipped=10.0
+2023-03-13 20:52:38,432 INFO [train.py:968] (0/2) Epoch 27, batch 9200, giga_loss[loss=0.3436, simple_loss=0.4006, pruned_loss=0.1433, over 28179.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3681, pruned_loss=0.1199, over 5679953.22 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08962, over 5725087.57 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3703, pruned_loss=0.1227, over 5669023.67 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:53:08,153 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4674, 1.7121, 1.6715, 1.4926], device='cuda:0'), covar=tensor([0.2117, 0.2140, 0.2497, 0.2309], device='cuda:0'), in_proj_covar=tensor([0.0506, 0.0764, 0.0730, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 20:53:19,718 INFO [train.py:968] (0/2) Epoch 27, batch 9250, giga_loss[loss=0.2874, simple_loss=0.3613, pruned_loss=0.1067, over 28627.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3674, pruned_loss=0.1194, over 5689091.86 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3435, pruned_loss=0.08971, over 5730401.79 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3697, pruned_loss=0.1224, over 5674564.72 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:53:40,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.799e+03 2.498e+03 3.325e+03 6.483e+03, threshold=4.997e+03, percent-clipped=8.0
+2023-03-13 20:53:58,795 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 20:54:07,780 INFO [train.py:968] (0/2) Epoch 27, batch 9300, giga_loss[loss=0.3311, simple_loss=0.3922, pruned_loss=0.135, over 28707.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3689, pruned_loss=0.12, over 5682131.96 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3434, pruned_loss=0.08969, over 5731302.78 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3708, pruned_loss=0.1225, over 5669889.23 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:54:14,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1194161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:54:24,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7706, 1.9504, 1.9622, 1.7194], device='cuda:0'), covar=tensor([0.2921, 0.2521, 0.2097, 0.2510], device='cuda:0'), in_proj_covar=tensor([0.2047, 0.1999, 0.1928, 0.2057], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 20:54:49,467 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.36 vs. limit=5.0
+2023-03-13 20:54:49,663 INFO [train.py:968] (0/2) Epoch 27, batch 9350, giga_loss[loss=0.3309, simple_loss=0.3875, pruned_loss=0.1372, over 27574.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3696, pruned_loss=0.1196, over 5684248.97 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3437, pruned_loss=0.08976, over 5735671.36 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3717, pruned_loss=0.1224, over 5668887.60 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:54:53,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3765, 2.8682, 1.4665, 1.4877], device='cuda:0'), covar=tensor([0.0995, 0.0380, 0.0888, 0.1349], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0569, 0.0406, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 20:55:12,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.721e+03 2.407e+03 2.877e+03 5.331e+03, threshold=4.814e+03, percent-clipped=2.0
+2023-03-13 20:55:34,496 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.31 vs. limit=5.0
+2023-03-13 20:55:35,228 INFO [train.py:968] (0/2) Epoch 27, batch 9400, giga_loss[loss=0.4163, simple_loss=0.4486, pruned_loss=0.192, over 26623.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3691, pruned_loss=0.12, over 5683359.91 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3437, pruned_loss=0.08975, over 5737830.87 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1227, over 5668554.66 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:56:23,185 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1194304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:56:23,757 INFO [train.py:968] (0/2) Epoch 27, batch 9450, giga_loss[loss=0.2737, simple_loss=0.3624, pruned_loss=0.09249, over 28519.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3712, pruned_loss=0.1191, over 5683390.46 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3438, pruned_loss=0.08972, over 5738401.71 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3728, pruned_loss=0.1214, over 5671115.63 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:56:25,234 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1194307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:56:42,619 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.601e+03 1.990e+03 2.760e+03 6.456e+03, threshold=3.979e+03, percent-clipped=2.0
+2023-03-13 20:56:47,837 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1194336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:57:02,490 INFO [train.py:968] (0/2) Epoch 27, batch 9500, giga_loss[loss=0.449, simple_loss=0.4571, pruned_loss=0.2204, over 23600.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3718, pruned_loss=0.1175, over 5688376.34 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3443, pruned_loss=0.09007, over 5744739.90 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3738, pruned_loss=0.1203, over 5670337.52 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:57:38,751 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1194399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:57:42,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1194404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:57:43,441 INFO [train.py:968] (0/2) Epoch 27, batch 9550, giga_loss[loss=0.2916, simple_loss=0.3665, pruned_loss=0.1083, over 28908.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3727, pruned_loss=0.1171, over 5693275.49 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3435, pruned_loss=0.08978, over 5749588.16 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.376, pruned_loss=0.1205, over 5672493.53 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:58:05,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.525e+03 1.957e+03 2.571e+03 1.028e+04, threshold=3.913e+03, percent-clipped=9.0
+2023-03-13 20:58:06,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3260, 2.5730, 1.2987, 1.4464], device='cuda:0'), covar=tensor([0.0963, 0.0377, 0.0964, 0.1374], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0571, 0.0408, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 20:58:28,337 INFO [train.py:968] (0/2) Epoch 27, batch 9600, giga_loss[loss=0.3222, simple_loss=0.3844, pruned_loss=0.13, over 28938.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3759, pruned_loss=0.12, over 5684773.00 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3442, pruned_loss=0.09022, over 5744574.73 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3789, pruned_loss=0.1232, over 5670445.16 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:59:06,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3350, 2.7923, 1.4375, 1.4657], device='cuda:0'), covar=tensor([0.0940, 0.0426, 0.0871, 0.1250], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0571, 0.0407, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 20:59:08,599 INFO [train.py:968] (0/2) Epoch 27, batch 9650, giga_loss[loss=0.3833, simple_loss=0.4197, pruned_loss=0.1734, over 27638.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3762, pruned_loss=0.1213, over 5688605.37 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3435, pruned_loss=0.08977, over 5750532.40 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3805, pruned_loss=0.1254, over 5669549.79 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:59:17,437 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1194514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:59:32,199 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.785e+03 2.215e+03 2.982e+03 1.264e+04, threshold=4.431e+03, percent-clipped=12.0
+2023-03-13 20:59:43,443 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1194542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:59:47,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1194545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:59:57,028 INFO [train.py:968] (0/2) Epoch 27, batch 9700, giga_loss[loss=0.3424, simple_loss=0.3775, pruned_loss=0.1536, over 23526.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3761, pruned_loss=0.1225, over 5674932.09 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3435, pruned_loss=0.08979, over 5748440.97 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3798, pruned_loss=0.126, over 5661222.43 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:00:05,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7106, 1.4113, 4.1017, 3.4682], device='cuda:0'), covar=tensor([0.1498, 0.2829, 0.0482, 0.1028], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0671, 0.1002, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 21:00:13,155 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1194574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:00:39,726 INFO [train.py:968] (0/2) Epoch 27, batch 9750, giga_loss[loss=0.3042, simple_loss=0.3688, pruned_loss=0.1198, over 28710.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3749, pruned_loss=0.1217, over 5661417.03 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3434, pruned_loss=0.08979, over 5749029.37 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3786, pruned_loss=0.1253, over 5648050.20 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:01:01,203 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.160e+03 1.667e+03 2.261e+03 2.929e+03 7.077e+03, threshold=4.523e+03, percent-clipped=6.0
+2023-03-13 21:01:15,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3599, 1.6976, 1.0873, 1.2211], device='cuda:0'), covar=tensor([0.1385, 0.0793, 0.1483, 0.1422], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0453, 0.0526, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 21:01:22,421 INFO [train.py:968] (0/2) Epoch 27, batch 9800, giga_loss[loss=0.265, simple_loss=0.3485, pruned_loss=0.09077, over 28815.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.373, pruned_loss=0.1186, over 5669241.25 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3434, pruned_loss=0.08982, over 5750584.31 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3762, pruned_loss=0.1216, over 5656561.84 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:01:59,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3534, 4.1747, 3.9665, 2.0441], device='cuda:0'), covar=tensor([0.0656, 0.0827, 0.0829, 0.2074], device='cuda:0'), in_proj_covar=tensor([0.1310, 0.1206, 0.1017, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 21:02:03,886 INFO [train.py:968] (0/2) Epoch 27, batch 9850, giga_loss[loss=0.3016, simple_loss=0.3749, pruned_loss=0.1142, over 29041.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3721, pruned_loss=0.1166, over 5678602.98 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3432, pruned_loss=0.08956, over 5755394.23 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3758, pruned_loss=0.1202, over 5661909.46 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:02:17,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5432, 4.4330, 1.7631, 1.6987], device='cuda:0'), covar=tensor([0.1032, 0.0432, 0.0948, 0.1372], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0570, 0.0407, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 21:02:25,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.548e+03 1.936e+03 2.762e+03 9.638e+03, threshold=3.873e+03, percent-clipped=5.0
+2023-03-13 21:02:50,878 INFO [train.py:968] (0/2) Epoch 27, batch 9900, giga_loss[loss=0.2637, simple_loss=0.3454, pruned_loss=0.09106, over 28158.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3726, pruned_loss=0.1168, over 5675674.60 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3431, pruned_loss=0.08962, over 5754176.57 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.376, pruned_loss=0.12, over 5662106.57 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:03:13,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1194779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:03:40,099 INFO [train.py:968] (0/2) Epoch 27, batch 9950, giga_loss[loss=0.2968, simple_loss=0.3682, pruned_loss=0.1128, over 28777.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3735, pruned_loss=0.1183, over 5668029.77 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3431, pruned_loss=0.08961, over 5756872.99 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3768, pruned_loss=0.1214, over 5653287.00 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:04:02,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.973e+02 1.868e+03 2.434e+03 3.340e+03 7.690e+03, threshold=4.868e+03, percent-clipped=18.0
+2023-03-13 21:04:26,197 INFO [train.py:968] (0/2) Epoch 27, batch 10000, giga_loss[loss=0.3915, simple_loss=0.4179, pruned_loss=0.1825, over 26657.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3729, pruned_loss=0.1188, over 5681059.75 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3433, pruned_loss=0.08972, over 5759091.49 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3758, pruned_loss=0.1217, over 5665924.91 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:04:58,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1194889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:05:12,534 INFO [train.py:968] (0/2) Epoch 27, batch 10050, giga_loss[loss=0.3253, simple_loss=0.39, pruned_loss=0.1303, over 28738.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3715, pruned_loss=0.1194, over 5672627.38 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08942, over 5762102.73 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3749, pruned_loss=0.1226, over 5656294.87 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:05:29,525 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1194922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:05:34,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1194925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:05:38,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.744e+02 1.741e+03 2.225e+03 3.025e+03 7.359e+03, threshold=4.449e+03, percent-clipped=5.0
+2023-03-13 21:06:01,458 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1194954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:06:01,989 INFO [train.py:968] (0/2) Epoch 27, batch 10100, giga_loss[loss=0.2911, simple_loss=0.3573, pruned_loss=0.1125, over 28747.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3693, pruned_loss=0.1184, over 5675414.80 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.343, pruned_loss=0.08959, over 5760139.51 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3722, pruned_loss=0.1211, over 5662959.50 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:06:49,544 INFO [train.py:968] (0/2) Epoch 27, batch 10150, libri_loss[loss=0.2891, simple_loss=0.3602, pruned_loss=0.109, over 29551.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3691, pruned_loss=0.1194, over 5671870.64 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3435, pruned_loss=0.08973, over 5760259.15 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3715, pruned_loss=0.1219, over 5660317.63 frames. ], batch size: 81, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:07:12,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.179e+03 1.836e+03 2.467e+03 3.647e+03 7.720e+03, threshold=4.935e+03, percent-clipped=14.0
+2023-03-13 21:07:13,609 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1195032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:07:16,163 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1195035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:07:16,670 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1195036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:07:31,968 INFO [train.py:968] (0/2) Epoch 27, batch 10200, giga_loss[loss=0.2656, simple_loss=0.3358, pruned_loss=0.09769, over 28615.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3681, pruned_loss=0.1187, over 5677273.12 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3432, pruned_loss=0.08955, over 5761948.62 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3708, pruned_loss=0.1216, over 5664481.42 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:07:43,124 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1195064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:08:21,726 INFO [train.py:968] (0/2) Epoch 27, batch 10250, giga_loss[loss=0.2668, simple_loss=0.3442, pruned_loss=0.09473, over 28238.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3656, pruned_loss=0.116, over 5670037.99 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08961, over 5764681.63 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3681, pruned_loss=0.1188, over 5655656.52 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:08:44,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.626e+03 1.983e+03 2.702e+03 1.519e+04, threshold=3.967e+03, percent-clipped=5.0
+2023-03-13 21:09:07,209 INFO [train.py:968] (0/2) Epoch 27, batch 10300, giga_loss[loss=0.3529, simple_loss=0.3949, pruned_loss=0.1554, over 26561.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3622, pruned_loss=0.113, over 5672363.81 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.08922, over 5768630.77 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3654, pruned_loss=0.1163, over 5654616.74 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:09:51,830 INFO [train.py:968] (0/2) Epoch 27, batch 10350, giga_loss[loss=0.3797, simple_loss=0.4334, pruned_loss=0.163, over 27943.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3599, pruned_loss=0.1107, over 5673504.90 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3422, pruned_loss=0.08886, over 5768849.19 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3636, pruned_loss=0.1146, over 5655280.30 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:10:15,250 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.523e+03 1.944e+03 2.378e+03 5.212e+03, threshold=3.889e+03, percent-clipped=4.0
+2023-03-13 21:10:38,154 INFO [train.py:968] (0/2) Epoch 27, batch 10400, giga_loss[loss=0.3489, simple_loss=0.3879, pruned_loss=0.155, over 26666.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3584, pruned_loss=0.1105, over 5671627.98 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3424, pruned_loss=0.08902, over 5761853.13 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3618, pruned_loss=0.1141, over 5660596.46 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:11:24,413 INFO [train.py:968] (0/2) Epoch 27, batch 10450, giga_loss[loss=0.2604, simple_loss=0.3389, pruned_loss=0.09092, over 28979.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3569, pruned_loss=0.1106, over 5669574.57 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3422, pruned_loss=0.08892, over 5762583.87 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1136, over 5659779.70 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:11:34,430 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-13 21:11:45,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.396e+02 1.822e+03 2.586e+03 3.450e+03 8.098e+03, threshold=5.172e+03, percent-clipped=17.0
+2023-03-13 21:12:05,305 INFO [train.py:968] (0/2) Epoch 27, batch 10500, giga_loss[loss=0.2887, simple_loss=0.3652, pruned_loss=0.1061, over 28771.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3602, pruned_loss=0.1121, over 5666832.30 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3426, pruned_loss=0.08901, over 5757446.61 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3627, pruned_loss=0.1152, over 5661267.03 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:12:45,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9606, 2.2612, 1.4921, 1.7009], device='cuda:0'), covar=tensor([0.1168, 0.0752, 0.1169, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0453, 0.0526, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 21:12:50,257 INFO [train.py:968] (0/2) Epoch 27, batch 10550, giga_loss[loss=0.3169, simple_loss=0.3808, pruned_loss=0.1265, over 28894.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.363, pruned_loss=0.1133, over 5674318.35 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08925, over 5761329.30 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3653, pruned_loss=0.1162, over 5664005.36 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:12:54,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1195411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:13:13,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.610e+03 2.078e+03 3.213e+03 7.665e+03, threshold=4.155e+03, percent-clipped=5.0
+2023-03-13 21:13:35,548 INFO [train.py:968] (0/2) Epoch 27, batch 10600, giga_loss[loss=0.2848, simple_loss=0.3549, pruned_loss=0.1074, over 27934.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3645, pruned_loss=0.1147, over 5639243.73 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3432, pruned_loss=0.08932, over 5754844.71 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3665, pruned_loss=0.1175, over 5634018.17 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:13:53,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2506, 1.0780, 3.7955, 3.1588], device='cuda:0'), covar=tensor([0.1731, 0.2981, 0.0469, 0.1093], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0673, 0.1005, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 21:14:21,266 INFO [train.py:968] (0/2) Epoch 27, batch 10650, giga_loss[loss=0.3666, simple_loss=0.4021, pruned_loss=0.1656, over 26633.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3638, pruned_loss=0.1145, over 5638945.36 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08919, over 5756690.86 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3659, pruned_loss=0.1173, over 5631046.19 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:14:42,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.640e+03 2.149e+03 3.153e+03 9.058e+03, threshold=4.298e+03, percent-clipped=11.0
+2023-03-13 21:15:01,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1195554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:15:01,823 INFO [train.py:968] (0/2) Epoch 27, batch 10700, giga_loss[loss=0.2811, simple_loss=0.3621, pruned_loss=0.1, over 28877.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3655, pruned_loss=0.1161, over 5638171.42 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08922, over 5750362.58 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3677, pruned_loss=0.1189, over 5634292.91 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:15:04,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1195557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:15:35,541 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1195586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:15:50,702 INFO [train.py:968] (0/2) Epoch 27, batch 10750, giga_loss[loss=0.2711, simple_loss=0.3464, pruned_loss=0.0979, over 28486.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3671, pruned_loss=0.1166, over 5647440.94 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3435, pruned_loss=0.08946, over 5750500.33 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.369, pruned_loss=0.1193, over 5641733.43 frames. ], batch size: 65, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:16:15,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.150e+03 1.695e+03 2.122e+03 2.616e+03 7.112e+03, threshold=4.244e+03, percent-clipped=9.0
+2023-03-13 21:16:24,633 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-13 21:16:33,792 INFO [train.py:968] (0/2) Epoch 27, batch 10800, giga_loss[loss=0.3061, simple_loss=0.3768, pruned_loss=0.1177, over 28701.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3691, pruned_loss=0.1184, over 5645820.34 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.08919, over 5744323.86 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3714, pruned_loss=0.1212, over 5644725.75 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:17:18,429 INFO [train.py:968] (0/2) Epoch 27, batch 10850, giga_loss[loss=0.3509, simple_loss=0.4041, pruned_loss=0.1489, over 28301.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3683, pruned_loss=0.1181, over 5649828.62 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08886, over 5745824.79 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3717, pruned_loss=0.1217, over 5643834.34 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:17:44,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.136e+03 1.759e+03 2.153e+03 3.014e+03 6.479e+03, threshold=4.306e+03, percent-clipped=9.0
+2023-03-13 21:17:56,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-13 21:18:06,783 INFO [train.py:968] (0/2) Epoch 27, batch 10900, giga_loss[loss=0.3297, simple_loss=0.4012, pruned_loss=0.1291, over 28913.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3691, pruned_loss=0.1186, over 5642382.18 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3423, pruned_loss=0.08869, over 5738892.74 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3724, pruned_loss=0.1223, over 5641736.35 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:18:55,009 INFO [train.py:968] (0/2) Epoch 27, batch 10950, giga_loss[loss=0.3038, simple_loss=0.3671, pruned_loss=0.1202, over 28924.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3702, pruned_loss=0.1181, over 5644369.03 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3426, pruned_loss=0.0888, over 5740320.53 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.373, pruned_loss=0.1213, over 5641376.24 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:19:12,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5403, 3.1117, 1.6594, 1.6396], device='cuda:0'), covar=tensor([0.0833, 0.0368, 0.0727, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0570, 0.0407, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 21:19:20,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.198e+03 1.819e+03 2.294e+03 3.089e+03 5.686e+03, threshold=4.587e+03, percent-clipped=8.0
+2023-03-13 21:19:42,513 INFO [train.py:968] (0/2) Epoch 27, batch 11000, giga_loss[loss=0.2629, simple_loss=0.3397, pruned_loss=0.09307, over 29052.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3686, pruned_loss=0.1177, over 5640101.71 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3421, pruned_loss=0.08872, over 5734645.66 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.372, pruned_loss=0.1212, over 5640078.63 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:20:36,632 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6669, 1.7620, 1.7227, 1.5214], device='cuda:0'), covar=tensor([0.3062, 0.2780, 0.2600, 0.2885], device='cuda:0'), in_proj_covar=tensor([0.2048, 0.2004, 0.1933, 0.2058], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 21:20:36,943 INFO [train.py:968] (0/2) Epoch 27, batch 11050, giga_loss[loss=0.2847, simple_loss=0.3493, pruned_loss=0.11, over 28727.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3675, pruned_loss=0.1173, over 5655968.68 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3421, pruned_loss=0.08867, over 5736329.17 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3704, pruned_loss=0.1203, over 5653604.85 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:21:12,067 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.164e+03 1.841e+03 2.307e+03 2.932e+03 6.611e+03, threshold=4.615e+03, percent-clipped=4.0
+2023-03-13 21:21:32,608 INFO [train.py:968] (0/2) Epoch 27, batch 11100, giga_loss[loss=0.2865, simple_loss=0.3578, pruned_loss=0.1076, over 28650.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3673, pruned_loss=0.118, over 5649557.76 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3422, pruned_loss=0.08872, over 5735079.41 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3699, pruned_loss=0.1208, over 5647720.75 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:22:13,285 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1196000.pt
+2023-03-13 21:22:17,917 INFO [train.py:968] (0/2) Epoch 27, batch 11150, libri_loss[loss=0.2322, simple_loss=0.3151, pruned_loss=0.0746, over 29330.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3655, pruned_loss=0.1169, over 5659551.13 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08856, over 5729260.96 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3684, pruned_loss=0.12, over 5660962.74 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:22:39,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.113e+03 1.739e+03 2.098e+03 2.646e+03 6.638e+03, threshold=4.197e+03, percent-clipped=6.0
+2023-03-13 21:22:57,852 INFO [train.py:968] (0/2) Epoch 27, batch 11200, giga_loss[loss=0.3483, simple_loss=0.395, pruned_loss=0.1508, over 27969.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3658, pruned_loss=0.1173, over 5668876.31 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08886, over 5738130.35 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.369, pruned_loss=0.1209, over 5658697.27 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:23:15,576 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 21:23:42,000 INFO [train.py:968] (0/2) Epoch 27, batch 11250, libri_loss[loss=0.2075, simple_loss=0.2909, pruned_loss=0.06207, over 29668.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3651, pruned_loss=0.117, over 5675791.90 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3417, pruned_loss=0.08852, over 5745557.12 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1213, over 5658244.74 frames. ], batch size: 73, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:24:07,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.169e+03 1.651e+03 2.172e+03 3.094e+03 7.132e+03, threshold=4.344e+03, percent-clipped=14.0
+2023-03-13 21:24:27,771 INFO [train.py:968] (0/2) Epoch 27, batch 11300, giga_loss[loss=0.3007, simple_loss=0.3633, pruned_loss=0.1191, over 28816.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3662, pruned_loss=0.1176, over 5674236.81 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3419, pruned_loss=0.08862, over 5747055.91 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.37, pruned_loss=0.1219, over 5656639.65 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:25:16,108 INFO [train.py:968] (0/2) Epoch 27, batch 11350, giga_loss[loss=0.3091, simple_loss=0.3727, pruned_loss=0.1228, over 28614.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3676, pruned_loss=0.1188, over 5675802.59 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3421, pruned_loss=0.08872, over 5748797.70 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3708, pruned_loss=0.1225, over 5659416.07 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:25:40,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2475, 4.0940, 3.8898, 1.8203], device='cuda:0'), covar=tensor([0.0606, 0.0714, 0.0762, 0.2178], device='cuda:0'), in_proj_covar=tensor([0.1320, 0.1215, 0.1026, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 21:25:42,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.176e+03 1.771e+03 2.251e+03 3.118e+03 8.927e+03, threshold=4.502e+03, percent-clipped=7.0
+2023-03-13 21:25:59,466 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4854, 1.7730, 1.4482, 1.2591], device='cuda:0'), covar=tensor([0.2471, 0.2550, 0.2866, 0.2375], device='cuda:0'), in_proj_covar=tensor([0.1583, 0.1144, 0.1399, 0.1010], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 21:26:02,812 INFO [train.py:968] (0/2) Epoch 27, batch 11400, giga_loss[loss=0.3147, simple_loss=0.3734, pruned_loss=0.1281, over 28763.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3679, pruned_loss=0.1187, over 5675599.78 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3421, pruned_loss=0.08876, over 5748083.08 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3709, pruned_loss=0.1221, over 5662211.14 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:26:51,038 INFO [train.py:968] (0/2) Epoch 27, batch 11450, giga_loss[loss=0.2795, simple_loss=0.349, pruned_loss=0.105, over 28918.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.369, pruned_loss=0.1202, over 5662482.65 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3421, pruned_loss=0.08877, over 5740162.22 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3722, pruned_loss=0.1239, over 5655682.44 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:27:17,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.872e+03 2.304e+03 3.311e+03 6.200e+03, threshold=4.607e+03, percent-clipped=7.0
+2023-03-13 21:27:35,401 INFO [train.py:968] (0/2) Epoch 27, batch 11500, giga_loss[loss=0.2767, simple_loss=0.3482, pruned_loss=0.1026, over 28922.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1212, over 5664436.70 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3425, pruned_loss=0.08888, over 5745012.10 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.373, pruned_loss=0.1254, over 5651517.44 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:27:54,309 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1196374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:28:23,894 INFO [train.py:968] (0/2) Epoch 27, batch 11550, giga_loss[loss=0.3014, simple_loss=0.3715, pruned_loss=0.1157, over 28881.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3694, pruned_loss=0.1204, over 5675035.57 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08908, over 5743372.10 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3725, pruned_loss=0.1243, over 5664412.29 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:28:45,507 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-13 21:28:51,976 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.189e+03 1.854e+03 2.335e+03 3.543e+03 1.153e+04, threshold=4.670e+03, percent-clipped=13.0
+2023-03-13 21:29:11,736 INFO [train.py:968] (0/2) Epoch 27, batch 11600, giga_loss[loss=0.3536, simple_loss=0.4146, pruned_loss=0.1463, over 28865.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3703, pruned_loss=0.1207, over 5667409.01 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3428, pruned_loss=0.08902, over 5744302.89 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3729, pruned_loss=0.1241, over 5657450.46 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:29:31,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6147, 2.2839, 1.6414, 0.8442], device='cuda:0'), covar=tensor([0.6895, 0.3541, 0.4616, 0.7387], device='cuda:0'), in_proj_covar=tensor([0.1838, 0.1733, 0.1660, 0.1497], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 21:29:51,069 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8802, 2.1421, 2.0184, 1.6021], device='cuda:0'), covar=tensor([0.3168, 0.2663, 0.3109, 0.3383], device='cuda:0'), in_proj_covar=tensor([0.2050, 0.2010, 0.1936, 0.2058], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 21:29:59,093 INFO [train.py:968] (0/2) Epoch 27, batch 11650, giga_loss[loss=0.3232, simple_loss=0.3829, pruned_loss=0.1318, over 28934.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3703, pruned_loss=0.1204, over 5662335.68 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3428, pruned_loss=0.089, over 5728107.73 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3734, pruned_loss=0.1242, over 5666354.41 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:30:11,370 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1196516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:30:13,036 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1196518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:30:27,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3939, 1.9033, 1.3421, 0.7665], device='cuda:0'), covar=tensor([0.5619, 0.2973, 0.3515, 0.6159], device='cuda:0'), in_proj_covar=tensor([0.1837, 0.1732, 0.1658, 0.1495], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 21:30:28,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.752e+02 1.808e+03 2.376e+03 3.113e+03 8.173e+03, threshold=4.752e+03, percent-clipped=5.0
+2023-03-13 21:30:48,334 INFO [train.py:968] (0/2) Epoch 27, batch 11700, giga_loss[loss=0.3181, simple_loss=0.3805, pruned_loss=0.1279, over 28887.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3728, pruned_loss=0.1225, over 5662205.93 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3432, pruned_loss=0.08912, over 5729634.43 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3753, pruned_loss=0.126, over 5662841.71 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:31:32,559 INFO [train.py:968] (0/2) Epoch 27, batch 11750, giga_loss[loss=0.2851, simple_loss=0.3514, pruned_loss=0.1094, over 28541.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3716, pruned_loss=0.1211, over 5678913.24 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.343, pruned_loss=0.08896, over 5734038.07 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3749, pruned_loss=0.1252, over 5673271.66 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:31:59,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.649e+03 2.140e+03 2.631e+03 4.877e+03, threshold=4.279e+03, percent-clipped=1.0
+2023-03-13 21:32:17,599 INFO [train.py:968] (0/2) Epoch 27, batch 11800, giga_loss[loss=0.2936, simple_loss=0.3657, pruned_loss=0.1107, over 28776.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3731, pruned_loss=0.1217, over 5682773.25 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.343, pruned_loss=0.08892, over 5736687.18 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3763, pruned_loss=0.1256, over 5674969.60 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:33:05,965 INFO [train.py:968] (0/2) Epoch 27, batch 11850, giga_loss[loss=0.2988, simple_loss=0.3743, pruned_loss=0.1116, over 28929.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3725, pruned_loss=0.1202, over 5667037.00 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3432, pruned_loss=0.08904, over 5723675.01 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3753, pruned_loss=0.1238, over 5671566.70 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:33:13,975 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-13 21:33:34,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.652e+03 2.162e+03 2.911e+03 6.756e+03, threshold=4.324e+03, percent-clipped=4.0
+2023-03-13 21:33:47,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1196749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:33:53,746 INFO [train.py:968] (0/2) Epoch 27, batch 11900, giga_loss[loss=0.2805, simple_loss=0.3488, pruned_loss=0.1061, over 29087.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3716, pruned_loss=0.12, over 5656295.72 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3434, pruned_loss=0.0892, over 5715955.79 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3738, pruned_loss=0.1228, over 5666367.82 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:34:36,393 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1196801.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 21:34:38,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4251, 1.7372, 1.5223, 1.5360], device='cuda:0'), covar=tensor([0.0786, 0.0320, 0.0323, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 21:34:38,771 INFO [train.py:968] (0/2) Epoch 27, batch 11950, giga_loss[loss=0.2541, simple_loss=0.3338, pruned_loss=0.08723, over 28874.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3697, pruned_loss=0.1186, over 5671200.87 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3434, pruned_loss=0.08918, over 5717040.24 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3717, pruned_loss=0.1212, over 5677781.66 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:34:41,506 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 21:35:05,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6648, 1.8524, 1.6071, 1.8514], device='cuda:0'), covar=tensor([0.2251, 0.2377, 0.2449, 0.2260], device='cuda:0'), in_proj_covar=tensor([0.1585, 0.1146, 0.1399, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 21:35:09,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.134e+03 1.860e+03 2.313e+03 2.996e+03 6.539e+03, threshold=4.626e+03, percent-clipped=7.0
+2023-03-13 21:35:28,178 INFO [train.py:968] (0/2) Epoch 27, batch 12000, giga_loss[loss=0.3203, simple_loss=0.3876, pruned_loss=0.1265, over 28794.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3706, pruned_loss=0.1196, over 5656320.11 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3435, pruned_loss=0.08908, over 5720432.02 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3728, pruned_loss=0.1225, over 5656885.70 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:35:28,184 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 21:35:35,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4142, 3.1303, 1.5120, 1.6172], device='cuda:0'), covar=tensor([0.1117, 0.0442, 0.1005, 0.1470], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0572, 0.0407, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 21:35:35,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2481, 1.2456, 3.4391, 3.1719], device='cuda:0'), covar=tensor([0.1872, 0.3074, 0.0623, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0669, 0.1002, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 21:35:36,334 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2058, simple_loss=0.3139, pruned_loss=0.04886, over 944034.00 frames. 
+2023-03-13 21:35:36,334 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 21:35:53,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5286, 3.6901, 1.5714, 1.8065], device='cuda:0'), covar=tensor([0.1028, 0.0402, 0.0906, 0.1300], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0572, 0.0407, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 21:36:08,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1196891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:36:10,006 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1196892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:36:10,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1196893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:36:12,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1196895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:36:22,747 INFO [train.py:968] (0/2) Epoch 27, batch 12050, giga_loss[loss=0.3411, simple_loss=0.3938, pruned_loss=0.1442, over 28272.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3725, pruned_loss=0.1207, over 5653857.92 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3435, pruned_loss=0.08918, over 5714798.97 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3748, pruned_loss=0.1236, over 5658458.19 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:36:39,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1196924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:36:49,007 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.666e+03 2.235e+03 2.915e+03 9.029e+03, threshold=4.469e+03, percent-clipped=8.0
+2023-03-13 21:37:04,782 INFO [train.py:968] (0/2) Epoch 27, batch 12100, giga_loss[loss=0.3088, simple_loss=0.3723, pruned_loss=0.1227, over 28877.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.37, pruned_loss=0.1196, over 5652699.20 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.08912, over 5706701.76 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3739, pruned_loss=0.1236, over 5660321.72 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:37:52,443 INFO [train.py:968] (0/2) Epoch 27, batch 12150, libri_loss[loss=0.2785, simple_loss=0.3632, pruned_loss=0.09691, over 27890.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3709, pruned_loss=0.1208, over 5655052.97 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.08912, over 5708729.87 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3744, pruned_loss=0.1246, over 5658076.95 frames. ], batch size: 116, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:38:16,016 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:17,815 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.584e+03 1.978e+03 2.743e+03 5.786e+03, threshold=3.956e+03, percent-clipped=3.0
+2023-03-13 21:38:18,257 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:18,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:20,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:36,079 INFO [train.py:968] (0/2) Epoch 27, batch 12200, giga_loss[loss=0.2795, simple_loss=0.3538, pruned_loss=0.1026, over 28794.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3712, pruned_loss=0.1207, over 5660028.37 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3434, pruned_loss=0.08916, over 5710541.30 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3746, pruned_loss=0.1247, over 5659360.82 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:38:42,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2866, 1.3632, 1.2595, 1.2354], device='cuda:0'), covar=tensor([0.1618, 0.1740, 0.1457, 0.1542], device='cuda:0'), in_proj_covar=tensor([0.2047, 0.2007, 0.1927, 0.2058], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 21:38:45,292 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:47,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:49,406 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:53,694 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3659, 1.3778, 1.3441, 1.3224], device='cuda:0'), covar=tensor([0.1417, 0.1628, 0.1382, 0.1456], device='cuda:0'), in_proj_covar=tensor([0.2048, 0.2008, 0.1927, 0.2059], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 21:39:24,751 INFO [train.py:968] (0/2) Epoch 27, batch 12250, giga_loss[loss=0.4152, simple_loss=0.4511, pruned_loss=0.1896, over 28733.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3734, pruned_loss=0.1226, over 5654973.58 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3436, pruned_loss=0.0893, over 5711833.25 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3762, pruned_loss=0.1259, over 5652800.20 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:39:32,976 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:39:54,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.176e+03 1.642e+03 2.153e+03 2.759e+03 6.649e+03, threshold=4.305e+03, percent-clipped=4.0
+2023-03-13 21:40:14,310 INFO [train.py:968] (0/2) Epoch 27, batch 12300, giga_loss[loss=0.286, simple_loss=0.3578, pruned_loss=0.1071, over 28845.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3735, pruned_loss=0.1233, over 5638915.07 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3437, pruned_loss=0.0894, over 5713893.49 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3759, pruned_loss=0.1262, over 5634688.15 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:40:15,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:40:33,200 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197176.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 21:40:58,324 INFO [train.py:968] (0/2) Epoch 27, batch 12350, giga_loss[loss=0.3732, simple_loss=0.4104, pruned_loss=0.168, over 26521.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3734, pruned_loss=0.1232, over 5646380.78 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3433, pruned_loss=0.08923, over 5716561.06 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3768, pruned_loss=0.1269, over 5638073.43 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:41:24,119 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5491, 1.8483, 1.4850, 1.4768], device='cuda:0'), covar=tensor([0.2665, 0.2729, 0.3108, 0.2454], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1147, 0.1401, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 21:41:27,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.399e+02 1.839e+03 2.237e+03 3.203e+03 1.030e+04, threshold=4.475e+03, percent-clipped=11.0
+2023-03-13 21:41:42,635 INFO [train.py:968] (0/2) Epoch 27, batch 12400, giga_loss[loss=0.2794, simple_loss=0.3527, pruned_loss=0.103, over 28753.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3736, pruned_loss=0.1226, over 5646829.36 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3434, pruned_loss=0.08918, over 5711250.48 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3769, pruned_loss=0.1263, over 5642880.34 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:42:14,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4541, 1.6847, 1.5982, 1.5252], device='cuda:0'), covar=tensor([0.1912, 0.2124, 0.2231, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0764, 0.0733, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 21:42:26,131 INFO [train.py:968] (0/2) Epoch 27, batch 12450, giga_loss[loss=0.2689, simple_loss=0.3413, pruned_loss=0.09825, over 29044.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3711, pruned_loss=0.1206, over 5647977.31 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.343, pruned_loss=0.08896, over 5711024.87 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3751, pruned_loss=0.125, over 5642637.02 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:42:38,002 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 21:42:42,088 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197319.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 21:42:44,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197322.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 21:42:52,217 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:42:57,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.683e+03 2.305e+03 3.328e+03 1.030e+04, threshold=4.610e+03, percent-clipped=8.0
+2023-03-13 21:43:11,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197351.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 21:43:14,092 INFO [train.py:968] (0/2) Epoch 27, batch 12500, giga_loss[loss=0.2622, simple_loss=0.3315, pruned_loss=0.0965, over 28736.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3682, pruned_loss=0.1189, over 5655982.78 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.343, pruned_loss=0.08893, over 5712454.54 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3717, pruned_loss=0.1227, over 5650046.26 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:44:02,535 INFO [train.py:968] (0/2) Epoch 27, batch 12550, giga_loss[loss=0.298, simple_loss=0.3531, pruned_loss=0.1214, over 28908.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3651, pruned_loss=0.117, over 5671634.85 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3429, pruned_loss=0.0888, over 5716128.04 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3684, pruned_loss=0.1207, over 5662613.73 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:44:25,643 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 21:44:34,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.709e+03 2.192e+03 3.154e+03 7.462e+03, threshold=4.385e+03, percent-clipped=9.0
+2023-03-13 21:44:34,608 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:44:51,573 INFO [train.py:968] (0/2) Epoch 27, batch 12600, giga_loss[loss=0.2834, simple_loss=0.3522, pruned_loss=0.1073, over 28722.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3618, pruned_loss=0.1159, over 5653278.56 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.343, pruned_loss=0.08883, over 5718019.04 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3645, pruned_loss=0.119, over 5644189.25 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:44:53,477 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:45:20,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:45:21,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6433, 1.5888, 1.8170, 1.4457], device='cuda:0'), covar=tensor([0.1535, 0.2362, 0.1280, 0.1574], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0718, 0.0976, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 21:45:23,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6220, 2.1150, 1.6100, 1.7395], device='cuda:0'), covar=tensor([0.0756, 0.0283, 0.0332, 0.0874], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 21:45:38,464 INFO [train.py:968] (0/2) Epoch 27, batch 12650, giga_loss[loss=0.3933, simple_loss=0.4181, pruned_loss=0.1842, over 26574.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3615, pruned_loss=0.1167, over 5655272.96 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3429, pruned_loss=0.0887, over 5720838.18 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3642, pruned_loss=0.1199, over 5644031.16 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:46:03,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:46:09,660 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.210e+03 1.872e+03 2.565e+03 3.868e+03 1.015e+04, threshold=5.130e+03, percent-clipped=15.0
+2023-03-13 21:46:26,817 INFO [train.py:968] (0/2) Epoch 27, batch 12700, giga_loss[loss=0.2781, simple_loss=0.348, pruned_loss=0.1041, over 28936.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3606, pruned_loss=0.1162, over 5654682.48 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08892, over 5724514.98 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3627, pruned_loss=0.1191, over 5641493.44 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:46:52,631 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:46:54,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:47:14,780 INFO [train.py:968] (0/2) Epoch 27, batch 12750, giga_loss[loss=0.2767, simple_loss=0.3453, pruned_loss=0.104, over 28586.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3601, pruned_loss=0.1149, over 5647145.56 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3432, pruned_loss=0.08887, over 5717657.72 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3622, pruned_loss=0.1177, over 5641249.27 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:47:24,666 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:47:38,926 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:47:44,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:47:49,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.908e+02 1.525e+03 1.992e+03 2.581e+03 6.266e+03, threshold=3.984e+03, percent-clipped=3.0
+2023-03-13 21:48:05,787 INFO [train.py:968] (0/2) Epoch 27, batch 12800, giga_loss[loss=0.293, simple_loss=0.3722, pruned_loss=0.1069, over 28996.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.36, pruned_loss=0.1124, over 5650433.70 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3434, pruned_loss=0.08902, over 5718524.19 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1147, over 5644448.25 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:48:13,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:48:25,868 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:48:28,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:48:54,623 INFO [train.py:968] (0/2) Epoch 27, batch 12850, giga_loss[loss=0.2679, simple_loss=0.3454, pruned_loss=0.09517, over 28407.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3574, pruned_loss=0.1091, over 5654962.54 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3432, pruned_loss=0.08897, over 5724041.64 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3593, pruned_loss=0.1115, over 5643502.66 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:48:56,433 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:48:56,446 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:49:29,160 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.044e+02 1.669e+03 2.126e+03 3.092e+03 5.860e+03, threshold=4.251e+03, percent-clipped=10.0
+2023-03-13 21:49:45,679 INFO [train.py:968] (0/2) Epoch 27, batch 12900, giga_loss[loss=0.2483, simple_loss=0.3353, pruned_loss=0.08059, over 28801.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.354, pruned_loss=0.1056, over 5654789.42 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3428, pruned_loss=0.08882, over 5727375.55 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3561, pruned_loss=0.1081, over 5640930.13 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:49:55,553 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197766.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 21:50:25,344 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-13 21:50:32,324 INFO [train.py:968] (0/2) Epoch 27, batch 12950, giga_loss[loss=0.2264, simple_loss=0.3145, pruned_loss=0.06916, over 28470.00 frames. ], tot_loss[loss=0.276, simple_loss=0.349, pruned_loss=0.1015, over 5651709.87 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3414, pruned_loss=0.08817, over 5730653.10 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3524, pruned_loss=0.1047, over 5634697.12 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:50:57,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:51:03,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.870e+02 1.391e+03 1.758e+03 2.459e+03 8.237e+03, threshold=3.515e+03, percent-clipped=4.0
+2023-03-13 21:51:12,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7988, 2.7027, 1.6038, 1.0238], device='cuda:0'), covar=tensor([0.9069, 0.4104, 0.5127, 0.7450], device='cuda:0'), in_proj_covar=tensor([0.1825, 0.1717, 0.1646, 0.1485], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 21:51:12,885 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:51:16,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:51:19,822 INFO [train.py:968] (0/2) Epoch 27, batch 13000, giga_loss[loss=0.2456, simple_loss=0.3383, pruned_loss=0.07646, over 28988.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3483, pruned_loss=0.09857, over 5666845.38 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3412, pruned_loss=0.08818, over 5735115.54 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3513, pruned_loss=0.1014, over 5647760.56 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:51:36,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-13 21:51:47,413 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:51:57,159 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4576, 1.7117, 1.3920, 1.3629], device='cuda:0'), covar=tensor([0.2782, 0.2680, 0.3093, 0.2317], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1144, 0.1403, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 21:52:10,207 INFO [train.py:968] (0/2) Epoch 27, batch 13050, giga_loss[loss=0.2542, simple_loss=0.3356, pruned_loss=0.08637, over 28643.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3491, pruned_loss=0.09885, over 5659753.68 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3416, pruned_loss=0.08867, over 5735337.58 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3514, pruned_loss=0.101, over 5642152.00 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:52:11,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4823, 1.6884, 1.6900, 1.4462], device='cuda:0'), covar=tensor([0.2674, 0.2300, 0.1648, 0.2183], device='cuda:0'), in_proj_covar=tensor([0.2039, 0.1996, 0.1921, 0.2050], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 21:52:25,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.9160, 4.7482, 4.5004, 2.2430], device='cuda:0'), covar=tensor([0.0472, 0.0609, 0.0737, 0.1981], device='cuda:0'), in_proj_covar=tensor([0.1296, 0.1197, 0.1008, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 21:52:31,973 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-13 21:52:38,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.533e+03 1.958e+03 2.776e+03 6.001e+03, threshold=3.917e+03, percent-clipped=10.0
+2023-03-13 21:52:54,984 INFO [train.py:968] (0/2) Epoch 27, batch 13100, giga_loss[loss=0.2501, simple_loss=0.3321, pruned_loss=0.08402, over 28951.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3477, pruned_loss=0.09781, over 5664635.04 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.341, pruned_loss=0.08853, over 5742500.00 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3504, pruned_loss=0.1001, over 5641144.09 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:53:13,307 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:53:17,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:53:40,824 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1198000.pt
+2023-03-13 21:53:45,064 INFO [train.py:968] (0/2) Epoch 27, batch 13150, giga_loss[loss=0.2306, simple_loss=0.2997, pruned_loss=0.08069, over 24298.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3445, pruned_loss=0.09566, over 5649951.99 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3409, pruned_loss=0.08857, over 5734263.45 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3468, pruned_loss=0.09756, over 5638241.38 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:53:46,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1198006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:54:14,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.464e+03 1.793e+03 2.664e+03 1.057e+04, threshold=3.587e+03, percent-clipped=11.0
+2023-03-13 21:54:27,739 INFO [train.py:968] (0/2) Epoch 27, batch 13200, giga_loss[loss=0.2522, simple_loss=0.3377, pruned_loss=0.08339, over 28853.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3433, pruned_loss=0.09522, over 5645959.56 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3404, pruned_loss=0.08843, over 5730875.03 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3458, pruned_loss=0.09715, over 5636080.40 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:54:32,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1198059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:54:57,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2618, 1.5959, 0.9348, 1.2218], device='cuda:0'), covar=tensor([0.1352, 0.0812, 0.1544, 0.1586], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0446, 0.0520, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 21:55:16,076 INFO [train.py:968] (0/2) Epoch 27, batch 13250, giga_loss[loss=0.2418, simple_loss=0.3292, pruned_loss=0.07723, over 28978.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3438, pruned_loss=0.09548, over 5642578.16 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.34, pruned_loss=0.0883, over 5730753.57 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3462, pruned_loss=0.09727, over 5633490.67 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:55:17,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3379, 1.2910, 3.8985, 3.2638], device='cuda:0'), covar=tensor([0.1683, 0.2841, 0.0448, 0.1022], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0668, 0.0995, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 21:55:28,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1818, 2.6451, 2.7123, 2.2375], device='cuda:0'), covar=tensor([0.2395, 0.2152, 0.1855, 0.2171], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0751, 0.0722, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 21:55:46,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.016e+02 1.658e+03 2.241e+03 3.121e+03 6.772e+03, threshold=4.483e+03, percent-clipped=16.0
+2023-03-13 21:55:50,978 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1198141.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 21:56:03,525 INFO [train.py:968] (0/2) Epoch 27, batch 13300, giga_loss[loss=0.2624, simple_loss=0.347, pruned_loss=0.08889, over 28818.00 frames. ], tot_loss[loss=0.265, simple_loss=0.342, pruned_loss=0.09398, over 5653139.69 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3395, pruned_loss=0.08819, over 5733930.96 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3444, pruned_loss=0.09568, over 5640870.39 frames. ], batch size: 285, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:56:50,819 INFO [train.py:968] (0/2) Epoch 27, batch 13350, libri_loss[loss=0.247, simple_loss=0.328, pruned_loss=0.08306, over 29530.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3394, pruned_loss=0.0919, over 5647785.39 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3391, pruned_loss=0.08816, over 5730920.74 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3418, pruned_loss=0.09347, over 5637860.09 frames. ], batch size: 82, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:57:25,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.110e+02 1.388e+03 1.745e+03 2.455e+03 6.135e+03, threshold=3.490e+03, percent-clipped=2.0
+2023-03-13 21:57:40,233 INFO [train.py:968] (0/2) Epoch 27, batch 13400, giga_loss[loss=0.2469, simple_loss=0.3259, pruned_loss=0.08396, over 28648.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3366, pruned_loss=0.09026, over 5649628.20 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3387, pruned_loss=0.08818, over 5732778.61 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3388, pruned_loss=0.09154, over 5638364.11 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:57:58,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3065, 1.3401, 3.7054, 3.1936], device='cuda:0'), covar=tensor([0.1903, 0.2914, 0.0876, 0.1306], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0666, 0.0992, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 21:57:58,387 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 21:58:12,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1198284.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 21:58:14,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1198287.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 21:58:37,060 INFO [train.py:968] (0/2) Epoch 27, batch 13450, giga_loss[loss=0.238, simple_loss=0.3236, pruned_loss=0.07626, over 28979.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3336, pruned_loss=0.08886, over 5658679.94 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3386, pruned_loss=0.08816, over 5734534.88 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3353, pruned_loss=0.0899, over 5646759.93 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:58:48,673 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1198316.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 21:59:05,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4248, 1.5571, 1.6235, 1.3035], device='cuda:0'), covar=tensor([0.1527, 0.2394, 0.1342, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.0925, 0.0710, 0.0971, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0015], device='cuda:0')
+2023-03-13 21:59:08,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.162e+02 1.335e+03 1.812e+03 2.364e+03 9.105e+03, threshold=3.624e+03, percent-clipped=8.0
+2023-03-13 21:59:19,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3591, 2.1282, 1.4948, 0.5628], device='cuda:0'), covar=tensor([0.5635, 0.3372, 0.4717, 0.6999], device='cuda:0'), in_proj_covar=tensor([0.1822, 0.1714, 0.1644, 0.1484], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 21:59:23,611 INFO [train.py:968] (0/2) Epoch 27, batch 13500, giga_loss[loss=0.2301, simple_loss=0.3171, pruned_loss=0.07155, over 28728.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.332, pruned_loss=0.08874, over 5663003.76 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3372, pruned_loss=0.08761, over 5738234.04 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3345, pruned_loss=0.09012, over 5646962.75 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:59:47,216 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-13 22:00:23,369 INFO [train.py:968] (0/2) Epoch 27, batch 13550, giga_loss[loss=0.2693, simple_loss=0.3524, pruned_loss=0.09308, over 28278.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3334, pruned_loss=0.08967, over 5644803.30 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3373, pruned_loss=0.08775, over 5730010.39 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3352, pruned_loss=0.09064, over 5638268.63 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:00:25,455 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1198407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:00:52,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1198434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:00:59,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.372e+02 1.431e+03 1.795e+03 2.360e+03 6.534e+03, threshold=3.590e+03, percent-clipped=4.0
+2023-03-13 22:01:02,996 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 22:01:18,291 INFO [train.py:968] (0/2) Epoch 27, batch 13600, giga_loss[loss=0.2786, simple_loss=0.3613, pruned_loss=0.09799, over 28427.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3358, pruned_loss=0.09016, over 5640549.61 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3364, pruned_loss=0.08738, over 5729457.76 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.338, pruned_loss=0.09131, over 5634065.65 frames. ], batch size: 369, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:02:16,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5038, 2.2898, 1.7350, 0.7945], device='cuda:0'), covar=tensor([0.6478, 0.3488, 0.4358, 0.7007], device='cuda:0'), in_proj_covar=tensor([0.1818, 0.1714, 0.1643, 0.1482], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 22:02:17,232 INFO [train.py:968] (0/2) Epoch 27, batch 13650, libri_loss[loss=0.2034, simple_loss=0.2802, pruned_loss=0.06326, over 29353.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.336, pruned_loss=0.08917, over 5656013.71 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3364, pruned_loss=0.08739, over 5732045.12 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3377, pruned_loss=0.09012, over 5646866.45 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:02:36,672 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1198520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:03:00,032 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.501e+02 1.548e+03 1.809e+03 2.261e+03 7.853e+03, threshold=3.618e+03, percent-clipped=9.0
+2023-03-13 22:03:21,334 INFO [train.py:968] (0/2) Epoch 27, batch 13700, giga_loss[loss=0.2367, simple_loss=0.3023, pruned_loss=0.08553, over 24484.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.336, pruned_loss=0.0891, over 5660805.76 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3362, pruned_loss=0.08731, over 5733743.53 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3376, pruned_loss=0.08994, over 5651448.67 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:03:46,245 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1198577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:03:48,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1198580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:04:18,924 INFO [train.py:968] (0/2) Epoch 27, batch 13750, giga_loss[loss=0.2428, simple_loss=0.328, pruned_loss=0.07874, over 28787.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3333, pruned_loss=0.08707, over 5672890.71 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3353, pruned_loss=0.08696, over 5736339.32 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3354, pruned_loss=0.08808, over 5661495.19 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:04:24,293 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1198609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:04:54,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9226, 2.9713, 1.8710, 1.0699], device='cuda:0'), covar=tensor([0.9385, 0.3839, 0.4796, 0.7838], device='cuda:0'), in_proj_covar=tensor([0.1822, 0.1718, 0.1647, 0.1485], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 22:05:02,922 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.196e+02 1.462e+03 2.001e+03 2.641e+03 5.254e+03, threshold=4.002e+03, percent-clipped=10.0
+2023-03-13 22:05:04,324 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1198641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:05:18,275 INFO [train.py:968] (0/2) Epoch 27, batch 13800, libri_loss[loss=0.2261, simple_loss=0.2989, pruned_loss=0.07667, over 29582.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3326, pruned_loss=0.08536, over 5662515.99 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3351, pruned_loss=0.087, over 5729998.99 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3344, pruned_loss=0.0861, over 5658673.57 frames. ], batch size: 74, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:06:22,537 INFO [train.py:968] (0/2) Epoch 27, batch 13850, giga_loss[loss=0.2436, simple_loss=0.32, pruned_loss=0.08356, over 28895.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3305, pruned_loss=0.08495, over 5658337.64 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3351, pruned_loss=0.08701, over 5731013.47 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3319, pruned_loss=0.08551, over 5653793.47 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:07:01,562 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.556e+02 1.383e+03 1.884e+03 2.568e+03 4.791e+03, threshold=3.767e+03, percent-clipped=3.0
+2023-03-13 22:07:20,420 INFO [train.py:968] (0/2) Epoch 27, batch 13900, giga_loss[loss=0.2333, simple_loss=0.3153, pruned_loss=0.07569, over 28951.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3309, pruned_loss=0.08614, over 5660969.03 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3354, pruned_loss=0.08733, over 5725443.89 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3316, pruned_loss=0.08628, over 5661162.22 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:07:54,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1198782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:08:20,634 INFO [train.py:968] (0/2) Epoch 27, batch 13950, giga_loss[loss=0.2415, simple_loss=0.3343, pruned_loss=0.07429, over 28891.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3308, pruned_loss=0.08662, over 5654407.94 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3355, pruned_loss=0.08734, over 5725365.04 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3312, pruned_loss=0.0867, over 5653992.89 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:08:38,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1283, 1.1328, 3.7025, 3.1170], device='cuda:0'), covar=tensor([0.1837, 0.2954, 0.0555, 0.1128], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0667, 0.0988, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 22:08:56,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.727e+02 1.499e+03 1.979e+03 2.469e+03 7.349e+03, threshold=3.959e+03, percent-clipped=10.0
+2023-03-13 22:09:15,406 INFO [train.py:968] (0/2) Epoch 27, batch 14000, giga_loss[loss=0.2914, simple_loss=0.3696, pruned_loss=0.1066, over 28767.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3323, pruned_loss=0.08683, over 5645339.64 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3352, pruned_loss=0.08729, over 5717518.63 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3328, pruned_loss=0.08694, over 5649805.21 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:09:18,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3936, 1.4503, 3.3435, 3.2091], device='cuda:0'), covar=tensor([0.1398, 0.2579, 0.0451, 0.1125], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0667, 0.0988, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 22:09:58,388 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7467, 5.1527, 1.8687, 2.1434], device='cuda:0'), covar=tensor([0.0942, 0.0210, 0.0927, 0.1228], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0563, 0.0405, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 22:10:00,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1198895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:10:11,306 INFO [train.py:968] (0/2) Epoch 27, batch 14050, giga_loss[loss=0.2294, simple_loss=0.3186, pruned_loss=0.07005, over 29040.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3334, pruned_loss=0.08696, over 5647246.99 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3346, pruned_loss=0.08694, over 5714198.82 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3343, pruned_loss=0.08738, over 5651704.94 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:10:39,146 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1198925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:10:42,080 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1198928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:10:59,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.240e+02 1.624e+03 1.992e+03 2.778e+03 7.396e+03, threshold=3.984e+03, percent-clipped=10.0
+2023-03-13 22:11:15,457 INFO [train.py:968] (0/2) Epoch 27, batch 14100, giga_loss[loss=0.2716, simple_loss=0.3491, pruned_loss=0.09708, over 28451.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3298, pruned_loss=0.08462, over 5655589.71 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3342, pruned_loss=0.0868, over 5710959.87 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3307, pruned_loss=0.08504, over 5660712.67 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:11:18,184 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1198957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:12:17,299 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-13 22:12:22,860 INFO [train.py:968] (0/2) Epoch 27, batch 14150, giga_loss[loss=0.2532, simple_loss=0.3319, pruned_loss=0.08722, over 28182.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3321, pruned_loss=0.08646, over 5666414.12 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3343, pruned_loss=0.08682, over 5712763.32 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3328, pruned_loss=0.08676, over 5668460.23 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:12:29,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8592, 3.6640, 3.5329, 1.7451], device='cuda:0'), covar=tensor([0.0737, 0.0862, 0.0778, 0.2183], device='cuda:0'), in_proj_covar=tensor([0.1278, 0.1178, 0.0993, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 22:12:36,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1199016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:13:11,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1199038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:13:13,495 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.638e+03 2.264e+03 3.342e+03 1.177e+04, threshold=4.527e+03, percent-clipped=17.0
+2023-03-13 22:13:14,002 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1199041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:13:30,834 INFO [train.py:968] (0/2) Epoch 27, batch 14200, giga_loss[loss=0.265, simple_loss=0.3605, pruned_loss=0.08473, over 28983.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3354, pruned_loss=0.08671, over 5667111.21 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3338, pruned_loss=0.08658, over 5714522.62 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3363, pruned_loss=0.08713, over 5666736.25 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:13:50,243 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1199070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:14:02,410 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 22:14:18,857 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.43 vs. limit=5.0
+2023-03-13 22:14:32,078 INFO [train.py:968] (0/2) Epoch 27, batch 14250, giga_loss[loss=0.2519, simple_loss=0.3459, pruned_loss=0.07892, over 29018.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3375, pruned_loss=0.08556, over 5671013.53 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3334, pruned_loss=0.08637, over 5717986.37 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3387, pruned_loss=0.08608, over 5666988.46 frames. ], batch size: 285, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:15:18,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.718e+02 1.633e+03 2.170e+03 2.945e+03 7.593e+03, threshold=4.340e+03, percent-clipped=3.0
+2023-03-13 22:15:33,205 INFO [train.py:968] (0/2) Epoch 27, batch 14300, giga_loss[loss=0.2225, simple_loss=0.3183, pruned_loss=0.06334, over 28333.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3382, pruned_loss=0.08459, over 5669965.66 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3332, pruned_loss=0.08626, over 5719546.80 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3394, pruned_loss=0.08508, over 5664613.50 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:15:38,413 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1199159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:15:41,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1199162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:16:15,379 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1199191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:16:34,000 INFO [train.py:968] (0/2) Epoch 27, batch 14350, giga_loss[loss=0.2801, simple_loss=0.3565, pruned_loss=0.1018, over 28918.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3366, pruned_loss=0.08363, over 5661797.52 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.333, pruned_loss=0.08628, over 5713287.40 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3378, pruned_loss=0.08398, over 5662486.88 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:16:51,753 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1199218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:17:17,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.522e+03 2.013e+03 2.866e+03 6.790e+03, threshold=4.026e+03, percent-clipped=6.0
+2023-03-13 22:17:19,720 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1199243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:17:38,604 INFO [train.py:968] (0/2) Epoch 27, batch 14400, giga_loss[loss=0.254, simple_loss=0.3374, pruned_loss=0.08534, over 28549.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3373, pruned_loss=0.08534, over 5651280.70 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3328, pruned_loss=0.08638, over 5699564.76 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3386, pruned_loss=0.08547, over 5662871.26 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:17:39,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4279, 5.2701, 4.9952, 2.6413], device='cuda:0'), covar=tensor([0.0443, 0.0547, 0.0715, 0.1626], device='cuda:0'), in_proj_covar=tensor([0.1277, 0.1174, 0.0992, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 22:18:01,228 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1199274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:18:43,669 INFO [train.py:968] (0/2) Epoch 27, batch 14450, giga_loss[loss=0.2368, simple_loss=0.3284, pruned_loss=0.07262, over 28927.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3369, pruned_loss=0.08598, over 5661915.00 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3327, pruned_loss=0.08646, over 5692739.10 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.338, pruned_loss=0.086, over 5676296.07 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:19:36,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.805e+02 1.427e+03 1.809e+03 2.375e+03 6.435e+03, threshold=3.617e+03, percent-clipped=8.0
+2023-03-13 22:20:00,280 INFO [train.py:968] (0/2) Epoch 27, batch 14500, giga_loss[loss=0.2602, simple_loss=0.3269, pruned_loss=0.09673, over 27131.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3366, pruned_loss=0.08703, over 5668619.09 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3323, pruned_loss=0.08645, over 5697035.46 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3379, pruned_loss=0.08705, over 5675915.17 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:20:04,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2371, 1.6982, 1.2687, 0.4468], device='cuda:0'), covar=tensor([0.4644, 0.2574, 0.4154, 0.6641], device='cuda:0'), in_proj_covar=tensor([0.1823, 0.1716, 0.1648, 0.1488], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 22:20:57,189 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8286, 1.3063, 1.2549, 1.0845], device='cuda:0'), covar=tensor([0.2290, 0.1360, 0.2419, 0.1814], device='cuda:0'), in_proj_covar=tensor([0.0491, 0.0743, 0.0712, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 22:21:04,639 INFO [train.py:968] (0/2) Epoch 27, batch 14550, libri_loss[loss=0.2788, simple_loss=0.3601, pruned_loss=0.09875, over 29120.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3325, pruned_loss=0.08469, over 5675789.08 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3319, pruned_loss=0.08639, over 5704510.78 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.334, pruned_loss=0.08473, over 5673372.37 frames. ], batch size: 101, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:21:51,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.147e+02 1.395e+03 1.776e+03 2.486e+03 7.867e+03, threshold=3.551e+03, percent-clipped=11.0
+2023-03-13 22:22:08,105 INFO [train.py:968] (0/2) Epoch 27, batch 14600, giga_loss[loss=0.2703, simple_loss=0.3478, pruned_loss=0.0964, over 28692.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.33, pruned_loss=0.08304, over 5682857.89 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3312, pruned_loss=0.08598, over 5708892.61 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3319, pruned_loss=0.08338, over 5676434.98 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:23:13,854 INFO [train.py:968] (0/2) Epoch 27, batch 14650, giga_loss[loss=0.2591, simple_loss=0.3221, pruned_loss=0.09801, over 24531.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3288, pruned_loss=0.08288, over 5666206.33 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.331, pruned_loss=0.08585, over 5701027.86 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3304, pruned_loss=0.08322, over 5667043.19 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:23:54,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.453e+03 1.869e+03 2.653e+03 5.129e+03, threshold=3.738e+03, percent-clipped=10.0
+2023-03-13 22:24:09,880 INFO [train.py:968] (0/2) Epoch 27, batch 14700, giga_loss[loss=0.2637, simple_loss=0.3411, pruned_loss=0.09315, over 28808.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3331, pruned_loss=0.08528, over 5668989.48 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3304, pruned_loss=0.0855, over 5703101.26 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.335, pruned_loss=0.08584, over 5666571.10 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:24:57,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1199593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:25:12,458 INFO [train.py:968] (0/2) Epoch 27, batch 14750, giga_loss[loss=0.2314, simple_loss=0.3171, pruned_loss=0.07285, over 28944.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.333, pruned_loss=0.08624, over 5677437.62 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.33, pruned_loss=0.08531, over 5705010.99 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3349, pruned_loss=0.08686, over 5673537.49 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:25:26,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2920, 1.0886, 3.7671, 3.2937], device='cuda:0'), covar=tensor([0.1677, 0.2977, 0.0456, 0.0986], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0668, 0.0988, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 22:25:29,010 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1199618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:25:58,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.919e+02 1.569e+03 1.908e+03 2.685e+03 6.116e+03, threshold=3.816e+03, percent-clipped=11.0
+2023-03-13 22:26:09,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1199649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:26:16,731 INFO [train.py:968] (0/2) Epoch 27, batch 14800, giga_loss[loss=0.2673, simple_loss=0.3435, pruned_loss=0.09558, over 28918.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3327, pruned_loss=0.08691, over 5675360.44 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.33, pruned_loss=0.08531, over 5699801.52 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3343, pruned_loss=0.08741, over 5676813.72 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:26:24,094 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 22:27:09,583 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1199700.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 22:27:15,432 INFO [train.py:968] (0/2) Epoch 27, batch 14850, giga_loss[loss=0.2721, simple_loss=0.3419, pruned_loss=0.1011, over 27035.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3332, pruned_loss=0.08753, over 5669734.41 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.33, pruned_loss=0.08537, over 5694519.70 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3345, pruned_loss=0.08793, over 5675231.83 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:27:54,838 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1199736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:27:59,807 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1199739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:28:03,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.287e+02 1.494e+03 1.814e+03 2.386e+03 5.640e+03, threshold=3.628e+03, percent-clipped=5.0
+2023-03-13 22:28:16,694 INFO [train.py:968] (0/2) Epoch 27, batch 14900, giga_loss[loss=0.2757, simple_loss=0.3614, pruned_loss=0.09494, over 28748.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3359, pruned_loss=0.0884, over 5655671.44 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.33, pruned_loss=0.08556, over 5675444.93 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.337, pruned_loss=0.0886, over 5675816.94 frames. ], batch size: 263, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:28:26,469 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1199761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:28:30,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1199764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:28:36,596 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1199768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:28:52,767 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1199782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:29:07,602 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4066, 3.2699, 3.0506, 1.8517], device='cuda:0'), covar=tensor([0.0779, 0.0888, 0.0866, 0.1832], device='cuda:0'), in_proj_covar=tensor([0.1267, 0.1169, 0.0985, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 22:29:08,744 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1199792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:29:09,278 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1199793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:29:11,925 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1199795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:29:23,684 INFO [train.py:968] (0/2) Epoch 27, batch 14950, giga_loss[loss=0.2313, simple_loss=0.3236, pruned_loss=0.0695, over 28958.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3364, pruned_loss=0.08832, over 5644489.50 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3296, pruned_loss=0.08551, over 5665905.91 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3379, pruned_loss=0.08865, over 5668718.67 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:29:49,461 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1199824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:30:03,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4372, 1.6015, 1.2831, 1.1682], device='cuda:0'), covar=tensor([0.0993, 0.0511, 0.0952, 0.1101], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0443, 0.0520, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 22:30:12,781 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.834e+02 1.494e+03 1.865e+03 2.306e+03 6.653e+03, threshold=3.730e+03, percent-clipped=6.0
+2023-03-13 22:30:29,809 INFO [train.py:968] (0/2) Epoch 27, batch 15000, libri_loss[loss=0.2459, simple_loss=0.3303, pruned_loss=0.0807, over 25763.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3343, pruned_loss=0.08709, over 5647669.71 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3291, pruned_loss=0.08535, over 5673172.39 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3363, pruned_loss=0.08763, over 5660137.29 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:30:29,813 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 22:30:38,196 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.1933, simple_loss=0.2948, pruned_loss=0.04593, over 944034.00 frames. 
+2023-03-13 22:30:38,197 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 22:31:42,045 INFO [train.py:968] (0/2) Epoch 27, batch 15050, giga_loss[loss=0.2606, simple_loss=0.3421, pruned_loss=0.08953, over 28974.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3306, pruned_loss=0.08633, over 5646434.14 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3288, pruned_loss=0.08523, over 5667285.04 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3326, pruned_loss=0.08691, over 5660445.26 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 22:32:10,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1971, 1.7389, 1.6151, 1.3827], device='cuda:0'), covar=tensor([0.2531, 0.2135, 0.2198, 0.2260], device='cuda:0'), in_proj_covar=tensor([0.0490, 0.0742, 0.0711, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 22:32:29,266 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-13 22:32:32,474 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.761e+02 1.520e+03 1.877e+03 2.449e+03 5.989e+03, threshold=3.754e+03, percent-clipped=10.0
+2023-03-13 22:32:44,267 INFO [train.py:968] (0/2) Epoch 27, batch 15100, libri_loss[loss=0.2255, simple_loss=0.3032, pruned_loss=0.07397, over 29586.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.325, pruned_loss=0.08373, over 5654589.43 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3287, pruned_loss=0.08533, over 5674639.37 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3266, pruned_loss=0.08409, over 5658518.02 frames. ], batch size: 75, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 22:33:39,703 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1200000.pt
+2023-03-13 22:33:46,148 INFO [train.py:968] (0/2) Epoch 27, batch 15150, giga_loss[loss=0.254, simple_loss=0.3213, pruned_loss=0.0934, over 24363.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3256, pruned_loss=0.08453, over 5648443.20 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3283, pruned_loss=0.08518, over 5675359.95 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3272, pruned_loss=0.08492, over 5650831.25 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 22:34:03,319 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 22:34:28,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.963e+02 1.434e+03 1.864e+03 2.724e+03 4.695e+03, threshold=3.728e+03, percent-clipped=4.0
+2023-03-13 22:34:38,724 INFO [train.py:968] (0/2) Epoch 27, batch 15200, giga_loss[loss=0.2399, simple_loss=0.3193, pruned_loss=0.08029, over 28898.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3267, pruned_loss=0.08517, over 5660914.21 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3282, pruned_loss=0.0852, over 5679088.73 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.328, pruned_loss=0.08547, over 5658954.17 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:35:03,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1200075.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 22:35:37,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1200103.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:35:37,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4431, 1.7363, 1.4553, 1.6228], device='cuda:0'), covar=tensor([0.0797, 0.0316, 0.0340, 0.0921], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 22:35:40,887 INFO [train.py:968] (0/2) Epoch 27, batch 15250, giga_loss[loss=0.2751, simple_loss=0.3379, pruned_loss=0.1062, over 26850.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3244, pruned_loss=0.08353, over 5657446.30 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3279, pruned_loss=0.08513, over 5684951.58 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3258, pruned_loss=0.08383, over 5650547.19 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:35:48,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0735, 1.5549, 1.5657, 1.2842], device='cuda:0'), covar=tensor([0.2333, 0.1671, 0.2236, 0.2041], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0740, 0.0708, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 22:35:57,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2264, 1.6909, 1.5699, 1.4639], device='cuda:0'), covar=tensor([0.1984, 0.1517, 0.1903, 0.1701], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0739, 0.0708, 0.0681], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 22:36:23,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.949e+02 1.430e+03 1.802e+03 2.672e+03 5.898e+03, threshold=3.604e+03, percent-clipped=8.0
+2023-03-13 22:36:36,455 INFO [train.py:968] (0/2) Epoch 27, batch 15300, giga_loss[loss=0.2518, simple_loss=0.3262, pruned_loss=0.08873, over 28105.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3247, pruned_loss=0.08304, over 5665734.23 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3281, pruned_loss=0.08542, over 5688003.53 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3254, pruned_loss=0.08294, over 5656386.55 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:36:40,227 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1200157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:36:59,117 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-13 22:37:45,140 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 22:37:45,311 INFO [train.py:968] (0/2) Epoch 27, batch 15350, giga_loss[loss=0.203, simple_loss=0.3031, pruned_loss=0.05147, over 29003.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3236, pruned_loss=0.08305, over 5659384.89 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3282, pruned_loss=0.08552, over 5688420.54 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3241, pruned_loss=0.08286, over 5651412.97 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:38:01,430 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1200218.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:38:05,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1200221.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:38:31,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.646e+02 1.424e+03 1.938e+03 3.083e+03 8.837e+03, threshold=3.875e+03, percent-clipped=15.0
+2023-03-13 22:38:39,746 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1200250.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:38:45,005 INFO [train.py:968] (0/2) Epoch 27, batch 15400, giga_loss[loss=0.2233, simple_loss=0.2933, pruned_loss=0.07672, over 24331.00 frames. ], tot_loss[loss=0.245, simple_loss=0.324, pruned_loss=0.08298, over 5661491.93 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3276, pruned_loss=0.08535, over 5697026.11 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3247, pruned_loss=0.08288, over 5645910.34 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:39:43,355 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1200300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:39:46,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1200303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:39:48,003 INFO [train.py:968] (0/2) Epoch 27, batch 15450, giga_loss[loss=0.2154, simple_loss=0.2957, pruned_loss=0.06755, over 28572.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.324, pruned_loss=0.08273, over 5660282.37 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3279, pruned_loss=0.08546, over 5697734.48 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3244, pruned_loss=0.08253, over 5647166.10 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:40:20,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0084, 1.3835, 1.0977, 0.1942], device='cuda:0'), covar=tensor([0.3728, 0.3146, 0.4711, 0.6469], device='cuda:0'), in_proj_covar=tensor([0.1825, 0.1721, 0.1649, 0.1496], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 22:40:21,814 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1200332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:40:36,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.280e+02 1.338e+03 1.964e+03 2.654e+03 5.848e+03, threshold=3.927e+03, percent-clipped=5.0
+2023-03-13 22:40:51,141 INFO [train.py:968] (0/2) Epoch 27, batch 15500, giga_loss[loss=0.2409, simple_loss=0.3198, pruned_loss=0.08104, over 28931.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3252, pruned_loss=0.08435, over 5662730.32 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3276, pruned_loss=0.08539, over 5702948.48 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3256, pruned_loss=0.0842, over 5647057.66 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:41:13,829 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1200374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:41:50,159 INFO [train.py:968] (0/2) Epoch 27, batch 15550, giga_loss[loss=0.2313, simple_loss=0.3229, pruned_loss=0.06984, over 28698.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3249, pruned_loss=0.08354, over 5671501.65 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3276, pruned_loss=0.08549, over 5706760.61 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3252, pruned_loss=0.08331, over 5654956.63 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:42:33,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.827e+02 1.375e+03 1.749e+03 2.355e+03 7.007e+03, threshold=3.498e+03, percent-clipped=8.0
+2023-03-13 22:42:47,833 INFO [train.py:968] (0/2) Epoch 27, batch 15600, giga_loss[loss=0.2595, simple_loss=0.331, pruned_loss=0.09406, over 26764.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3277, pruned_loss=0.08375, over 5675387.17 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3273, pruned_loss=0.08533, over 5707450.04 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3282, pruned_loss=0.08366, over 5660631.03 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:43:15,575 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1200478.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 22:43:18,638 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.67 vs. limit=5.0
+2023-03-13 22:43:46,829 INFO [train.py:968] (0/2) Epoch 27, batch 15650, giga_loss[loss=0.2374, simple_loss=0.3248, pruned_loss=0.07502, over 28943.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.33, pruned_loss=0.08457, over 5669198.13 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.327, pruned_loss=0.08516, over 5711578.10 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3307, pruned_loss=0.08463, over 5652908.82 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:44:32,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.412e+02 1.577e+03 2.061e+03 2.944e+03 7.856e+03, threshold=4.122e+03, percent-clipped=10.0
+2023-03-13 22:44:42,300 INFO [train.py:968] (0/2) Epoch 27, batch 15700, giga_loss[loss=0.226, simple_loss=0.3146, pruned_loss=0.0687, over 28868.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3311, pruned_loss=0.08478, over 5676005.67 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3269, pruned_loss=0.08508, over 5714050.25 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3318, pruned_loss=0.08488, over 5658765.68 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:45:15,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5317, 1.8346, 1.4927, 1.3304], device='cuda:0'), covar=tensor([0.2890, 0.2850, 0.3354, 0.2619], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1142, 0.1403, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 22:45:27,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8533, 2.0885, 1.6918, 2.1705], device='cuda:0'), covar=tensor([0.2821, 0.2975, 0.3321, 0.2514], device='cuda:0'), in_proj_covar=tensor([0.1585, 0.1141, 0.1402, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 22:45:39,918 INFO [train.py:968] (0/2) Epoch 27, batch 15750, giga_loss[loss=0.2163, simple_loss=0.3029, pruned_loss=0.0648, over 28994.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3306, pruned_loss=0.08418, over 5692840.94 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.327, pruned_loss=0.08508, over 5721695.17 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3313, pruned_loss=0.08425, over 5670793.37 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:45:51,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0747, 1.2487, 3.4206, 2.9699], device='cuda:0'), covar=tensor([0.1667, 0.2626, 0.0565, 0.0984], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0665, 0.0981, 0.0952], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 22:45:56,282 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1200621.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:45:59,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1200624.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:46:22,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.547e+02 1.539e+03 2.004e+03 2.724e+03 7.780e+03, threshold=4.009e+03, percent-clipped=11.0
+2023-03-13 22:46:34,600 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1200653.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 22:46:36,386 INFO [train.py:968] (0/2) Epoch 27, batch 15800, giga_loss[loss=0.2267, simple_loss=0.3137, pruned_loss=0.0698, over 28887.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3272, pruned_loss=0.08195, over 5695414.19 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3266, pruned_loss=0.08482, over 5724899.22 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3282, pruned_loss=0.08219, over 5674711.82 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:46:37,030 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 22:47:11,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2921, 1.4952, 1.3372, 1.5947], device='cuda:0'), covar=tensor([0.0821, 0.0357, 0.0367, 0.0934], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:0')
+2023-03-13 22:47:23,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3854, 1.7956, 1.7208, 1.6126], device='cuda:0'), covar=tensor([0.2145, 0.1765, 0.2275, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0737, 0.0706, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 22:47:34,502 INFO [train.py:968] (0/2) Epoch 27, batch 15850, giga_loss[loss=0.2294, simple_loss=0.3099, pruned_loss=0.07446, over 27724.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3263, pruned_loss=0.08139, over 5696130.30 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3269, pruned_loss=0.08516, over 5726968.79 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3267, pruned_loss=0.08116, over 5676664.63 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:48:15,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.015e+02 1.238e+03 1.726e+03 2.251e+03 7.649e+03, threshold=3.452e+03, percent-clipped=2.0
+2023-03-13 22:48:19,428 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1200749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:48:26,921 INFO [train.py:968] (0/2) Epoch 27, batch 15900, giga_loss[loss=0.2278, simple_loss=0.3151, pruned_loss=0.07025, over 28158.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3245, pruned_loss=0.08108, over 5679039.28 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.327, pruned_loss=0.08518, over 5713376.95 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3247, pruned_loss=0.08074, over 5674845.04 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:49:02,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5954, 2.0144, 1.3238, 1.5636], device='cuda:0'), covar=tensor([0.1083, 0.0575, 0.0994, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0446, 0.0523, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 22:49:24,781 INFO [train.py:968] (0/2) Epoch 27, batch 15950, giga_loss[loss=0.2507, simple_loss=0.3401, pruned_loss=0.08068, over 29081.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3266, pruned_loss=0.08245, over 5667199.09 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3272, pruned_loss=0.08529, over 5707296.19 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3266, pruned_loss=0.08197, over 5667939.67 frames. ], batch size: 285, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:50:12,211 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.495e+02 1.590e+03 1.966e+03 2.577e+03 4.705e+03, threshold=3.933e+03, percent-clipped=10.0
+2023-03-13 22:50:21,335 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-13 22:50:23,839 INFO [train.py:968] (0/2) Epoch 27, batch 16000, giga_loss[loss=0.2291, simple_loss=0.3141, pruned_loss=0.07203, over 28933.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3293, pruned_loss=0.08393, over 5677103.12 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3273, pruned_loss=0.08539, over 5712146.96 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3292, pruned_loss=0.08337, over 5671863.32 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:51:13,090 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1200892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:51:16,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1200895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:51:26,730 INFO [train.py:968] (0/2) Epoch 27, batch 16050, giga_loss[loss=0.2331, simple_loss=0.3203, pruned_loss=0.07297, over 28834.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.331, pruned_loss=0.0857, over 5675743.80 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3268, pruned_loss=0.08508, over 5717446.92 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3313, pruned_loss=0.08549, over 5666098.75 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:51:47,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1200924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:52:11,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.482e+03 1.935e+03 2.563e+03 8.826e+03, threshold=3.869e+03, percent-clipped=6.0
+2023-03-13 22:52:11,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3490, 1.4844, 1.4003, 1.5328], device='cuda:0'), covar=tensor([0.0793, 0.0354, 0.0349, 0.0911], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0115], device='cuda:0')
+2023-03-13 22:52:15,869 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1200950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:52:20,391 INFO [train.py:968] (0/2) Epoch 27, batch 16100, giga_loss[loss=0.2925, simple_loss=0.3739, pruned_loss=0.1055, over 28296.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3334, pruned_loss=0.08704, over 5678737.08 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3262, pruned_loss=0.08481, over 5715917.31 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3346, pruned_loss=0.0872, over 5671372.81 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:52:38,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3549, 1.8594, 1.3974, 0.6854], device='cuda:0'), covar=tensor([0.5257, 0.2879, 0.4323, 0.6336], device='cuda:0'), in_proj_covar=tensor([0.1825, 0.1726, 0.1650, 0.1496], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 22:53:17,402 INFO [train.py:968] (0/2) Epoch 27, batch 16150, giga_loss[loss=0.3072, simple_loss=0.3807, pruned_loss=0.1169, over 28882.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3347, pruned_loss=0.08649, over 5685230.26 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3258, pruned_loss=0.0846, over 5718770.88 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3361, pruned_loss=0.08683, over 5676391.80 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:53:23,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5050, 1.6208, 1.2602, 1.2330], device='cuda:0'), covar=tensor([0.0779, 0.0305, 0.0777, 0.0947], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0445, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 22:54:08,475 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.153e+02 1.565e+03 2.024e+03 2.551e+03 7.898e+03, threshold=4.049e+03, percent-clipped=9.0
+2023-03-13 22:54:18,971 INFO [train.py:968] (0/2) Epoch 27, batch 16200, giga_loss[loss=0.2218, simple_loss=0.3094, pruned_loss=0.06711, over 29234.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3353, pruned_loss=0.08679, over 5688446.42 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3258, pruned_loss=0.08472, over 5724581.09 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3366, pruned_loss=0.087, over 5675131.54 frames. ], batch size: 113, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:55:30,435 INFO [train.py:968] (0/2) Epoch 27, batch 16250, giga_loss[loss=0.2586, simple_loss=0.3228, pruned_loss=0.09723, over 26791.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3333, pruned_loss=0.08577, over 5693088.48 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3257, pruned_loss=0.0846, over 5727365.46 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3346, pruned_loss=0.08607, over 5679620.12 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:55:33,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5407, 2.1385, 1.7745, 1.8847], device='cuda:0'), covar=tensor([0.0734, 0.0266, 0.0292, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:0')
+2023-03-13 22:56:19,569 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.492e+02 1.424e+03 1.798e+03 2.347e+03 7.904e+03, threshold=3.597e+03, percent-clipped=5.0
+2023-03-13 22:56:31,314 INFO [train.py:968] (0/2) Epoch 27, batch 16300, giga_loss[loss=0.2509, simple_loss=0.3325, pruned_loss=0.08464, over 29072.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3326, pruned_loss=0.08589, over 5693387.53 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3255, pruned_loss=0.0845, over 5731708.81 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3339, pruned_loss=0.08626, over 5678109.96 frames. ], batch size: 285, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:56:53,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4581, 1.3616, 3.8021, 3.3010], device='cuda:0'), covar=tensor([0.1590, 0.2774, 0.0501, 0.1041], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0670, 0.0988, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 22:57:31,335 INFO [train.py:968] (0/2) Epoch 27, batch 16350, giga_loss[loss=0.2449, simple_loss=0.3282, pruned_loss=0.08078, over 28340.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3313, pruned_loss=0.08541, over 5681533.38 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3256, pruned_loss=0.08457, over 5736174.94 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3325, pruned_loss=0.08567, over 5664703.01 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:58:22,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.554e+02 1.443e+03 2.244e+03 2.874e+03 9.406e+03, threshold=4.488e+03, percent-clipped=10.0
+2023-03-13 22:58:31,351 INFO [train.py:968] (0/2) Epoch 27, batch 16400, giga_loss[loss=0.297, simple_loss=0.3541, pruned_loss=0.1199, over 26791.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3299, pruned_loss=0.08558, over 5671239.25 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3259, pruned_loss=0.08473, over 5723633.45 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3307, pruned_loss=0.08567, over 5667473.93 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:59:30,899 INFO [train.py:968] (0/2) Epoch 27, batch 16450, giga_loss[loss=0.2395, simple_loss=0.3236, pruned_loss=0.07771, over 28892.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3291, pruned_loss=0.08539, over 5668987.87 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.326, pruned_loss=0.08482, over 5718588.04 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3297, pruned_loss=0.08539, over 5668901.81 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:59:56,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1201325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:00:18,483 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.286e+02 1.479e+03 1.955e+03 2.700e+03 5.866e+03, threshold=3.909e+03, percent-clipped=3.0
+2023-03-13 23:00:28,134 INFO [train.py:968] (0/2) Epoch 27, batch 16500, giga_loss[loss=0.2158, simple_loss=0.3092, pruned_loss=0.06116, over 28951.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3282, pruned_loss=0.08387, over 5667728.55 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3258, pruned_loss=0.08464, over 5719920.31 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3289, pruned_loss=0.08403, over 5665337.67 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:01:06,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2032, 4.0685, 3.8092, 1.9588], device='cuda:0'), covar=tensor([0.0654, 0.0728, 0.0822, 0.1990], device='cuda:0'), in_proj_covar=tensor([0.1276, 0.1171, 0.0991, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 23:01:24,999 INFO [train.py:968] (0/2) Epoch 27, batch 16550, giga_loss[loss=0.2584, simple_loss=0.3438, pruned_loss=0.08644, over 28926.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3282, pruned_loss=0.08278, over 5674421.97 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3254, pruned_loss=0.08438, over 5721643.60 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3292, pruned_loss=0.0831, over 5669652.51 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:02:13,115 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-13 23:02:13,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.318e+02 1.591e+03 2.124e+03 2.996e+03 1.033e+04, threshold=4.248e+03, percent-clipped=13.0
+2023-03-13 23:02:23,888 INFO [train.py:968] (0/2) Epoch 27, batch 16600, giga_loss[loss=0.248, simple_loss=0.3391, pruned_loss=0.0785, over 27608.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3304, pruned_loss=0.08238, over 5666597.03 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3256, pruned_loss=0.08448, over 5722602.38 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.331, pruned_loss=0.08253, over 5661889.03 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:02:27,170 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-13 23:02:35,240 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1201468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:02:40,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1201471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:02:54,337 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1201485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:03:12,331 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1201500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:03:16,819 INFO [train.py:968] (0/2) Epoch 27, batch 16650, giga_loss[loss=0.2541, simple_loss=0.3429, pruned_loss=0.08264, over 28822.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3311, pruned_loss=0.08235, over 5674364.18 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3255, pruned_loss=0.08458, over 5716322.67 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3318, pruned_loss=0.08235, over 5674360.18 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:03:43,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4237, 1.9016, 1.4057, 0.7676], device='cuda:0'), covar=tensor([0.5539, 0.3178, 0.4378, 0.6345], device='cuda:0'), in_proj_covar=tensor([0.1819, 0.1715, 0.1643, 0.1486], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 23:04:08,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.686e+02 1.443e+03 1.723e+03 2.292e+03 4.898e+03, threshold=3.446e+03, percent-clipped=6.0
+2023-03-13 23:04:17,497 INFO [train.py:968] (0/2) Epoch 27, batch 16700, giga_loss[loss=0.2755, simple_loss=0.3526, pruned_loss=0.09917, over 28920.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3305, pruned_loss=0.08196, over 5680821.35 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3253, pruned_loss=0.08436, over 5717769.99 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3314, pruned_loss=0.08209, over 5678721.62 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:04:57,937 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-13 23:05:22,311 INFO [train.py:968] (0/2) Epoch 27, batch 16750, giga_loss[loss=0.2339, simple_loss=0.3275, pruned_loss=0.07015, over 28028.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3302, pruned_loss=0.08175, over 5670174.10 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.325, pruned_loss=0.08421, over 5711744.21 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3313, pruned_loss=0.08195, over 5672266.65 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:05:53,863 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-13 23:06:21,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.824e+02 1.490e+03 1.944e+03 2.583e+03 7.199e+03, threshold=3.888e+03, percent-clipped=11.0
+2023-03-13 23:06:31,282 INFO [train.py:968] (0/2) Epoch 27, batch 16800, giga_loss[loss=0.2491, simple_loss=0.3298, pruned_loss=0.08419, over 28051.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.331, pruned_loss=0.08201, over 5669664.26 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3249, pruned_loss=0.08425, over 5714751.23 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3321, pruned_loss=0.0821, over 5667804.76 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:06:52,753 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1201670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:07:40,348 INFO [train.py:968] (0/2) Epoch 27, batch 16850, giga_loss[loss=0.2492, simple_loss=0.3403, pruned_loss=0.07903, over 28896.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3316, pruned_loss=0.08209, over 5679646.89 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3245, pruned_loss=0.0839, over 5719404.22 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.333, pruned_loss=0.08241, over 5672879.77 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:08:38,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.254e+02 1.481e+03 2.038e+03 2.982e+03 6.728e+03, threshold=4.076e+03, percent-clipped=11.0
+2023-03-13 23:08:49,207 INFO [train.py:968] (0/2) Epoch 27, batch 16900, giga_loss[loss=0.2618, simple_loss=0.3565, pruned_loss=0.08354, over 28942.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3356, pruned_loss=0.08392, over 5678251.36 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3248, pruned_loss=0.08409, over 5718637.86 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3366, pruned_loss=0.08398, over 5672948.87 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:09:26,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6347, 1.8092, 1.5210, 1.8000], device='cuda:0'), covar=tensor([0.2763, 0.2847, 0.3265, 0.2672], device='cuda:0'), in_proj_covar=tensor([0.1584, 0.1137, 0.1402, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 23:09:54,693 INFO [train.py:968] (0/2) Epoch 27, batch 16950, giga_loss[loss=0.2504, simple_loss=0.3301, pruned_loss=0.08538, over 28589.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3352, pruned_loss=0.08325, over 5685610.56 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3245, pruned_loss=0.0838, over 5724315.87 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3365, pruned_loss=0.08356, over 5675043.46 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:10:46,425 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1201842.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 23:10:52,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.467e+03 1.828e+03 2.550e+03 6.815e+03, threshold=3.656e+03, percent-clipped=7.0
+2023-03-13 23:11:01,858 INFO [train.py:968] (0/2) Epoch 27, batch 17000, giga_loss[loss=0.2579, simple_loss=0.3357, pruned_loss=0.09004, over 28221.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3337, pruned_loss=0.08325, over 5693252.66 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3246, pruned_loss=0.08377, over 5726628.42 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3348, pruned_loss=0.0835, over 5681956.65 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:11:06,361 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1201860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:11:41,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3804, 3.2249, 1.5293, 1.4782], device='cuda:0'), covar=tensor([0.0986, 0.0398, 0.0949, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0560, 0.0405, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 23:12:01,189 INFO [train.py:968] (0/2) Epoch 27, batch 17050, giga_loss[loss=0.2634, simple_loss=0.3427, pruned_loss=0.09208, over 28125.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3305, pruned_loss=0.08164, over 5701685.11 frames. ], libri_tot_loss[loss=0.2456, simple_loss=0.3242, pruned_loss=0.08345, over 5734229.73 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3321, pruned_loss=0.08207, over 5684424.08 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:12:07,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2326, 1.6388, 1.6123, 1.4012], device='cuda:0'), covar=tensor([0.2138, 0.1891, 0.2228, 0.2156], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0736, 0.0708, 0.0680], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 23:12:09,484 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6029, 1.8767, 1.2908, 1.5330], device='cuda:0'), covar=tensor([0.1048, 0.0646, 0.1031, 0.1222], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0442, 0.0519, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 23:13:01,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.766e+02 1.292e+03 1.636e+03 2.178e+03 4.367e+03, threshold=3.272e+03, percent-clipped=4.0
+2023-03-13 23:13:10,134 INFO [train.py:968] (0/2) Epoch 27, batch 17100, giga_loss[loss=0.205, simple_loss=0.2958, pruned_loss=0.05711, over 28974.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3282, pruned_loss=0.07936, over 5709102.14 frames. ], libri_tot_loss[loss=0.2452, simple_loss=0.3238, pruned_loss=0.08325, over 5737265.68 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.33, pruned_loss=0.0798, over 5691554.74 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:13:44,229 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1201985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:14:01,805 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1202000.pt
+2023-03-13 23:14:05,335 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:14:06,215 INFO [train.py:968] (0/2) Epoch 27, batch 17150, giga_loss[loss=0.2231, simple_loss=0.3116, pruned_loss=0.0673, over 29024.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3282, pruned_loss=0.07969, over 5704292.61 frames. ], libri_tot_loss[loss=0.2448, simple_loss=0.3236, pruned_loss=0.08305, over 5741441.27 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3299, pruned_loss=0.08013, over 5685965.98 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:14:09,276 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:14:29,457 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:14:38,635 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:14:49,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1202045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:14:51,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.593e+02 1.474e+03 1.855e+03 2.665e+03 8.305e+03, threshold=3.709e+03, percent-clipped=14.0
+2023-03-13 23:14:59,437 INFO [train.py:968] (0/2) Epoch 27, batch 17200, giga_loss[loss=0.2454, simple_loss=0.3342, pruned_loss=0.07828, over 28913.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3295, pruned_loss=0.08058, over 5687035.72 frames. ], libri_tot_loss[loss=0.2448, simple_loss=0.3235, pruned_loss=0.08311, over 5725991.08 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3312, pruned_loss=0.08079, over 5683720.89 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:15:08,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4603, 1.5651, 1.6450, 1.2866], device='cuda:0'), covar=tensor([0.1737, 0.2676, 0.1476, 0.1836], device='cuda:0'), in_proj_covar=tensor([0.0925, 0.0706, 0.0973, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0015], device='cuda:0')
+2023-03-13 23:15:37,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2926, 0.9849, 1.0382, 1.4132], device='cuda:0'), covar=tensor([0.0698, 0.0366, 0.0347, 0.0764], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:0')
+2023-03-13 23:15:54,803 INFO [train.py:968] (0/2) Epoch 27, batch 17250, giga_loss[loss=0.2093, simple_loss=0.3023, pruned_loss=0.05814, over 28900.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3311, pruned_loss=0.08187, over 5679220.70 frames. ], libri_tot_loss[loss=0.2445, simple_loss=0.3231, pruned_loss=0.0829, over 5725643.74 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3331, pruned_loss=0.08219, over 5675940.47 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:16:41,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.578e+03 1.882e+03 2.987e+03 5.416e+03, threshold=3.765e+03, percent-clipped=18.0
+2023-03-13 23:16:49,140 INFO [train.py:968] (0/2) Epoch 27, batch 17300, giga_loss[loss=0.2678, simple_loss=0.3381, pruned_loss=0.09873, over 27648.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3291, pruned_loss=0.08178, over 5681030.65 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3235, pruned_loss=0.08313, over 5728352.93 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3305, pruned_loss=0.08179, over 5674855.01 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:17:26,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5571, 1.5940, 1.7635, 1.3684], device='cuda:0'), covar=tensor([0.1869, 0.2705, 0.1566, 0.1947], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0705, 0.0972, 0.0873], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0015], device='cuda:0')
+2023-03-13 23:17:28,382 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:17:29,607 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:17:31,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:17:45,113 INFO [train.py:968] (0/2) Epoch 27, batch 17350, giga_loss[loss=0.2783, simple_loss=0.3528, pruned_loss=0.1019, over 28678.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3295, pruned_loss=0.08254, over 5682979.33 frames. ], libri_tot_loss[loss=0.2451, simple_loss=0.3238, pruned_loss=0.08319, over 5729990.15 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3304, pruned_loss=0.08248, over 5675539.80 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:18:01,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1202217.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 23:18:04,118 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:18:36,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.545e+03 1.932e+03 2.544e+03 5.682e+03, threshold=3.865e+03, percent-clipped=10.0
+2023-03-13 23:18:43,558 INFO [train.py:968] (0/2) Epoch 27, batch 17400, giga_loss[loss=0.2647, simple_loss=0.3508, pruned_loss=0.08926, over 28770.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3325, pruned_loss=0.08486, over 5686925.26 frames. ], libri_tot_loss[loss=0.2452, simple_loss=0.3239, pruned_loss=0.08324, over 5730021.20 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3332, pruned_loss=0.08478, over 5680684.67 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:19:35,514 INFO [train.py:968] (0/2) Epoch 27, batch 17450, giga_loss[loss=0.2723, simple_loss=0.3628, pruned_loss=0.09095, over 28993.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3412, pruned_loss=0.08993, over 5686653.63 frames. ], libri_tot_loss[loss=0.2452, simple_loss=0.3239, pruned_loss=0.08325, over 5731756.44 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3419, pruned_loss=0.08989, over 5679851.28 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:20:15,633 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.349e+03 1.683e+03 2.231e+03 5.156e+03, threshold=3.367e+03, percent-clipped=6.0
+2023-03-13 23:20:19,917 INFO [train.py:968] (0/2) Epoch 27, batch 17500, giga_loss[loss=0.2477, simple_loss=0.3256, pruned_loss=0.08494, over 28871.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3483, pruned_loss=0.09359, over 5684628.44 frames. ], libri_tot_loss[loss=0.2454, simple_loss=0.3241, pruned_loss=0.08333, over 5723192.99 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3488, pruned_loss=0.09357, over 5686249.62 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:20:24,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1202360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:20:24,554 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202360.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 23:20:27,954 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202363.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 23:20:45,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6092, 1.8051, 1.7825, 1.3828], device='cuda:0'), covar=tensor([0.1963, 0.2877, 0.1694, 0.2003], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0703, 0.0972, 0.0873], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0015], device='cuda:0')
+2023-03-13 23:20:52,307 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202392.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 23:21:02,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1202402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:21:04,663 INFO [train.py:968] (0/2) Epoch 27, batch 17550, giga_loss[loss=0.263, simple_loss=0.3387, pruned_loss=0.0937, over 28344.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3461, pruned_loss=0.0935, over 5679284.37 frames. ], libri_tot_loss[loss=0.2456, simple_loss=0.3243, pruned_loss=0.08339, over 5717870.58 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3469, pruned_loss=0.0937, over 5683910.32 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:21:43,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.156e+02 1.213e+03 1.488e+03 2.099e+03 6.238e+03, threshold=2.975e+03, percent-clipped=5.0
+2023-03-13 23:21:49,343 INFO [train.py:968] (0/2) Epoch 27, batch 17600, giga_loss[loss=0.208, simple_loss=0.2831, pruned_loss=0.06649, over 28583.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3386, pruned_loss=0.09061, over 5681743.25 frames. ], libri_tot_loss[loss=0.2456, simple_loss=0.3244, pruned_loss=0.08341, over 5720651.74 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3395, pruned_loss=0.09088, over 5682420.93 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:22:28,979 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:22:30,145 INFO [train.py:968] (0/2) Epoch 27, batch 17650, libri_loss[loss=0.216, simple_loss=0.2905, pruned_loss=0.07077, over 29502.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.332, pruned_loss=0.08788, over 5676968.23 frames. ], libri_tot_loss[loss=0.246, simple_loss=0.3249, pruned_loss=0.08361, over 5721670.07 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3327, pruned_loss=0.08816, over 5674847.90 frames. ], batch size: 70, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:22:30,990 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:22:55,770 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:23:06,017 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:23:08,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:23:09,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.358e+02 1.167e+03 1.711e+03 2.276e+03 8.732e+03, threshold=3.422e+03, percent-clipped=12.0
+2023-03-13 23:23:14,737 INFO [train.py:968] (0/2) Epoch 27, batch 17700, giga_loss[loss=0.2069, simple_loss=0.2879, pruned_loss=0.06293, over 28592.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3242, pruned_loss=0.08433, over 5685077.99 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3249, pruned_loss=0.08338, over 5725468.46 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3249, pruned_loss=0.08481, over 5679112.71 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:23:22,446 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1202564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:23:34,731 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:23:56,328 INFO [train.py:968] (0/2) Epoch 27, batch 17750, giga_loss[loss=0.2475, simple_loss=0.3048, pruned_loss=0.09513, over 26599.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3175, pruned_loss=0.08151, over 5690625.94 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3249, pruned_loss=0.08335, over 5728190.04 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3178, pruned_loss=0.0819, over 5682831.73 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:24:31,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.129e+02 1.164e+03 1.465e+03 1.890e+03 6.226e+03, threshold=2.931e+03, percent-clipped=4.0
+2023-03-13 23:24:35,703 INFO [train.py:968] (0/2) Epoch 27, batch 17800, giga_loss[loss=0.2395, simple_loss=0.3036, pruned_loss=0.08766, over 28593.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3125, pruned_loss=0.07936, over 5695504.55 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3252, pruned_loss=0.08353, over 5732494.22 frames. ], giga_tot_loss[loss=0.2356, simple_loss=0.3123, pruned_loss=0.07945, over 5684683.79 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:24:47,267 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:14,720 INFO [train.py:968] (0/2) Epoch 27, batch 17850, libri_loss[loss=0.2255, simple_loss=0.309, pruned_loss=0.07098, over 29578.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3096, pruned_loss=0.07775, over 5702979.35 frames. ], libri_tot_loss[loss=0.2457, simple_loss=0.3249, pruned_loss=0.08329, over 5736222.23 frames. ], giga_tot_loss[loss=0.2326, simple_loss=0.3093, pruned_loss=0.07791, over 5689981.53 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:25:17,108 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:18,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:24,231 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:32,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:42,486 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 23:25:45,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:55,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.405e+02 1.086e+03 1.342e+03 1.825e+03 5.428e+03, threshold=2.683e+03, percent-clipped=6.0
+2023-03-13 23:25:59,918 INFO [train.py:968] (0/2) Epoch 27, batch 17900, giga_loss[loss=0.2154, simple_loss=0.2897, pruned_loss=0.07054, over 28785.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.306, pruned_loss=0.07652, over 5697271.29 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.325, pruned_loss=0.08331, over 5738647.71 frames. ], giga_tot_loss[loss=0.2292, simple_loss=0.3054, pruned_loss=0.07653, over 5684265.82 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:26:39,985 INFO [train.py:968] (0/2) Epoch 27, batch 17950, giga_loss[loss=0.2076, simple_loss=0.2823, pruned_loss=0.06647, over 29053.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3049, pruned_loss=0.07583, over 5711565.02 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3255, pruned_loss=0.08337, over 5742619.57 frames. ], giga_tot_loss[loss=0.227, simple_loss=0.3031, pruned_loss=0.07548, over 5695577.72 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:26:40,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7429, 1.9903, 1.6533, 1.7970], device='cuda:0'), covar=tensor([0.2776, 0.2880, 0.3393, 0.2693], device='cuda:0'), in_proj_covar=tensor([0.1581, 0.1137, 0.1399, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 23:27:03,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5796, 1.8003, 1.7304, 1.5259], device='cuda:0'), covar=tensor([0.3144, 0.2711, 0.2112, 0.2784], device='cuda:0'), in_proj_covar=tensor([0.2026, 0.1954, 0.1862, 0.2013], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 23:27:14,567 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:27:16,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.630e+02 1.197e+03 1.454e+03 2.065e+03 5.728e+03, threshold=2.908e+03, percent-clipped=10.0
+2023-03-13 23:27:23,563 INFO [train.py:968] (0/2) Epoch 27, batch 18000, giga_loss[loss=0.2069, simple_loss=0.2691, pruned_loss=0.07237, over 23777.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3019, pruned_loss=0.07472, over 5696798.45 frames. ], libri_tot_loss[loss=0.2463, simple_loss=0.3257, pruned_loss=0.08346, over 5744719.67 frames. ], giga_tot_loss[loss=0.2243, simple_loss=0.3, pruned_loss=0.07426, over 5681978.40 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:27:23,567 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-13 23:27:31,990 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2015, simple_loss=0.308, pruned_loss=0.04746, over 944034.00 frames. 
+2023-03-13 23:27:31,991 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-13 23:27:46,998 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:28:05,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7218, 1.8084, 1.9188, 1.5027], device='cuda:0'), covar=tensor([0.1934, 0.2600, 0.1533, 0.1762], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0710, 0.0981, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 23:28:12,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7877, 1.7988, 2.0013, 1.5599], device='cuda:0'), covar=tensor([0.2000, 0.2648, 0.1620, 0.1886], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0710, 0.0981, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 23:28:14,540 INFO [train.py:968] (0/2) Epoch 27, batch 18050, giga_loss[loss=0.2005, simple_loss=0.2785, pruned_loss=0.06122, over 28849.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2989, pruned_loss=0.07325, over 5702251.19 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3258, pruned_loss=0.08344, over 5746013.12 frames. ], giga_tot_loss[loss=0.2214, simple_loss=0.2971, pruned_loss=0.07281, over 5689069.00 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:28:52,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.180e+02 1.155e+03 1.443e+03 1.992e+03 5.485e+03, threshold=2.886e+03, percent-clipped=11.0
+2023-03-13 23:28:55,895 INFO [train.py:968] (0/2) Epoch 27, batch 18100, giga_loss[loss=0.1937, simple_loss=0.2705, pruned_loss=0.05844, over 29073.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2969, pruned_loss=0.07265, over 5697014.69 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3264, pruned_loss=0.08359, over 5743052.39 frames. ], giga_tot_loss[loss=0.2188, simple_loss=0.294, pruned_loss=0.07183, over 5687731.40 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:29:01,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4257, 1.5472, 1.5300, 1.3566], device='cuda:0'), covar=tensor([0.3170, 0.2922, 0.2295, 0.2803], device='cuda:0'), in_proj_covar=tensor([0.2034, 0.1963, 0.1870, 0.2021], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 23:29:40,228 INFO [train.py:968] (0/2) Epoch 27, batch 18150, giga_loss[loss=0.1697, simple_loss=0.2382, pruned_loss=0.05064, over 28552.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2938, pruned_loss=0.07146, over 5690716.62 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3266, pruned_loss=0.08368, over 5745405.86 frames. ], giga_tot_loss[loss=0.2161, simple_loss=0.291, pruned_loss=0.07059, over 5680717.50 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:30:09,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5816, 1.7391, 1.7822, 1.3828], device='cuda:0'), covar=tensor([0.1866, 0.2915, 0.1632, 0.1936], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0712, 0.0984, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 23:30:17,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:30:21,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.791e+02 1.073e+03 1.297e+03 1.897e+03 4.355e+03, threshold=2.594e+03, percent-clipped=6.0
+2023-03-13 23:30:25,593 INFO [train.py:968] (0/2) Epoch 27, batch 18200, giga_loss[loss=0.2291, simple_loss=0.301, pruned_loss=0.07857, over 29049.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2907, pruned_loss=0.0702, over 5690561.00 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3267, pruned_loss=0.08377, over 5746986.89 frames. ], giga_tot_loss[loss=0.2134, simple_loss=0.2881, pruned_loss=0.06932, over 5680737.84 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:31:00,054 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:31:06,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:31:12,399 INFO [train.py:968] (0/2) Epoch 27, batch 18250, giga_loss[loss=0.2873, simple_loss=0.3517, pruned_loss=0.1114, over 26737.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2985, pruned_loss=0.07474, over 5680230.43 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3269, pruned_loss=0.08381, over 5752235.00 frames. ], giga_tot_loss[loss=0.2211, simple_loss=0.295, pruned_loss=0.0736, over 5665544.23 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:31:48,790 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 23:31:49,066 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.075e+02 1.289e+03 1.628e+03 2.091e+03 6.046e+03, threshold=3.256e+03, percent-clipped=15.0
+2023-03-13 23:31:52,784 INFO [train.py:968] (0/2) Epoch 27, batch 18300, giga_loss[loss=0.3107, simple_loss=0.3837, pruned_loss=0.1189, over 27909.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3116, pruned_loss=0.08107, over 5687646.79 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3267, pruned_loss=0.08356, over 5755689.94 frames. ], giga_tot_loss[loss=0.2346, simple_loss=0.3087, pruned_loss=0.0803, over 5671149.24 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:32:01,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2369, 1.4348, 1.3129, 1.0850], device='cuda:0'), covar=tensor([0.2834, 0.2900, 0.2085, 0.2870], device='cuda:0'), in_proj_covar=tensor([0.2040, 0.1969, 0.1875, 0.2028], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 23:32:15,883 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1203186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:32:18,570 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1203189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:32:30,879 INFO [train.py:968] (0/2) Epoch 27, batch 18350, giga_loss[loss=0.2985, simple_loss=0.364, pruned_loss=0.1165, over 29098.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3233, pruned_loss=0.0867, over 5702229.09 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3266, pruned_loss=0.08341, over 5759551.71 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3208, pruned_loss=0.08623, over 5684353.30 frames. ], batch size: 113, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:32:41,511 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:32:45,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:32:57,010 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1203234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:32:58,825 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1203237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:02,864 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1203242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:04,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1203245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:07,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:08,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.114e+02 1.342e+03 1.655e+03 2.137e+03 5.058e+03, threshold=3.310e+03, percent-clipped=11.0
+2023-03-13 23:33:13,864 INFO [train.py:968] (0/2) Epoch 27, batch 18400, giga_loss[loss=0.32, simple_loss=0.3814, pruned_loss=0.1293, over 26436.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3311, pruned_loss=0.09015, over 5694576.73 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3269, pruned_loss=0.0835, over 5761030.98 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3289, pruned_loss=0.08977, over 5678384.01 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:33:20,831 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-13 23:33:23,081 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:27,063 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=4.54 vs. limit=5.0
+2023-03-13 23:33:30,020 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:55,957 INFO [train.py:968] (0/2) Epoch 27, batch 18450, giga_loss[loss=0.2667, simple_loss=0.3395, pruned_loss=0.09695, over 28541.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3349, pruned_loss=0.09071, over 5695454.43 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3267, pruned_loss=0.08333, over 5762269.24 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3334, pruned_loss=0.09063, over 5681075.24 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:34:37,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.249e+02 1.354e+03 1.688e+03 2.169e+03 4.592e+03, threshold=3.376e+03, percent-clipped=10.0
+2023-03-13 23:34:39,395 INFO [train.py:968] (0/2) Epoch 27, batch 18500, giga_loss[loss=0.2878, simple_loss=0.3574, pruned_loss=0.1091, over 28661.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3369, pruned_loss=0.09096, over 5691507.10 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3266, pruned_loss=0.08315, over 5762701.02 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3361, pruned_loss=0.09132, over 5677235.81 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:34:50,666 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1203365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:34:50,789 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-13 23:34:52,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1203368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:35:14,694 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1203392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:35:16,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1203395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:35:18,020 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:35:23,759 INFO [train.py:968] (0/2) Epoch 27, batch 18550, giga_loss[loss=0.2735, simple_loss=0.3531, pruned_loss=0.09697, over 28829.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3387, pruned_loss=0.09224, over 5681459.11 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3266, pruned_loss=0.08315, over 5762701.02 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3381, pruned_loss=0.09252, over 5670351.59 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:35:41,114 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:35:51,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4840, 1.3769, 3.9671, 3.3331], device='cuda:0'), covar=tensor([0.1674, 0.2907, 0.0460, 0.0900], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0663, 0.0985, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 23:36:03,520 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5638, 1.8283, 1.5096, 1.5447], device='cuda:0'), covar=tensor([0.2476, 0.2454, 0.2648, 0.2209], device='cuda:0'), in_proj_covar=tensor([0.1578, 0.1135, 0.1396, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-13 23:36:04,501 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.828e+02 1.257e+03 1.545e+03 2.106e+03 4.946e+03, threshold=3.090e+03, percent-clipped=5.0
+2023-03-13 23:36:07,125 INFO [train.py:968] (0/2) Epoch 27, batch 18600, giga_loss[loss=0.2899, simple_loss=0.3615, pruned_loss=0.1092, over 28994.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3415, pruned_loss=0.09416, over 5683285.87 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3269, pruned_loss=0.08309, over 5764407.64 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3411, pruned_loss=0.09469, over 5670856.36 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:36:48,743 INFO [train.py:968] (0/2) Epoch 27, batch 18650, giga_loss[loss=0.3473, simple_loss=0.3992, pruned_loss=0.1477, over 27617.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.345, pruned_loss=0.09634, over 5681000.35 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3271, pruned_loss=0.0831, over 5767272.82 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3448, pruned_loss=0.097, over 5667155.92 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:37:23,791 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1203547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:37:27,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.521e+02 1.338e+03 1.657e+03 2.334e+03 4.378e+03, threshold=3.314e+03, percent-clipped=9.0
+2023-03-13 23:37:29,836 INFO [train.py:968] (0/2) Epoch 27, batch 18700, giga_loss[loss=0.2597, simple_loss=0.3461, pruned_loss=0.08662, over 28929.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3474, pruned_loss=0.09699, over 5689372.37 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3273, pruned_loss=0.08324, over 5771262.61 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3475, pruned_loss=0.09776, over 5672440.92 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 23:38:07,765 INFO [train.py:968] (0/2) Epoch 27, batch 18750, giga_loss[loss=0.2674, simple_loss=0.3486, pruned_loss=0.0931, over 28763.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3501, pruned_loss=0.09767, over 5692609.12 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3281, pruned_loss=0.08358, over 5773713.00 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3502, pruned_loss=0.09844, over 5674350.08 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 23:38:20,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6332, 4.4464, 4.2018, 1.8180], device='cuda:0'), covar=tensor([0.0521, 0.0715, 0.0679, 0.2292], device='cuda:0'), in_proj_covar=tensor([0.1266, 0.1166, 0.0986, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 23:38:46,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.194e+02 1.276e+03 1.735e+03 2.181e+03 5.737e+03, threshold=3.470e+03, percent-clipped=5.0
+2023-03-13 23:38:47,343 INFO [train.py:968] (0/2) Epoch 27, batch 18800, giga_loss[loss=0.2615, simple_loss=0.3459, pruned_loss=0.0886, over 28858.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3518, pruned_loss=0.09774, over 5699767.10 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3283, pruned_loss=0.08366, over 5775290.06 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.352, pruned_loss=0.09846, over 5683027.97 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:38:59,802 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-13 23:39:14,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4595, 1.5648, 1.5651, 1.3312], device='cuda:0'), covar=tensor([0.3511, 0.3265, 0.2453, 0.3215], device='cuda:0'), in_proj_covar=tensor([0.2041, 0.1976, 0.1879, 0.2031], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 23:39:26,348 INFO [train.py:968] (0/2) Epoch 27, batch 18850, giga_loss[loss=0.2651, simple_loss=0.3464, pruned_loss=0.09193, over 28876.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3515, pruned_loss=0.09673, over 5688807.55 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3282, pruned_loss=0.0836, over 5760611.60 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3526, pruned_loss=0.09785, over 5686570.43 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:39:53,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5091, 2.1264, 1.5702, 0.8616], device='cuda:0'), covar=tensor([0.7955, 0.3826, 0.5409, 0.7281], device='cuda:0'), in_proj_covar=tensor([0.1820, 0.1716, 0.1648, 0.1484], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 23:40:08,223 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.821e+02 1.167e+03 1.449e+03 1.884e+03 3.628e+03, threshold=2.897e+03, percent-clipped=1.0
+2023-03-13 23:40:09,659 INFO [train.py:968] (0/2) Epoch 27, batch 18900, giga_loss[loss=0.2936, simple_loss=0.3772, pruned_loss=0.105, over 28926.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3508, pruned_loss=0.09486, over 5697474.80 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3282, pruned_loss=0.0836, over 5760611.60 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3517, pruned_loss=0.09573, over 5695733.62 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:40:46,606 INFO [train.py:968] (0/2) Epoch 27, batch 18950, giga_loss[loss=0.2654, simple_loss=0.3423, pruned_loss=0.09421, over 28614.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3498, pruned_loss=0.09402, over 5693964.39 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3288, pruned_loss=0.08372, over 5754461.71 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3506, pruned_loss=0.0949, over 5696362.87 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:41:03,933 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 23:41:25,439 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.102e+02 1.245e+03 1.489e+03 1.981e+03 4.083e+03, threshold=2.979e+03, percent-clipped=9.0
+2023-03-13 23:41:26,730 INFO [train.py:968] (0/2) Epoch 27, batch 19000, giga_loss[loss=0.3351, simple_loss=0.4064, pruned_loss=0.1318, over 29026.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3514, pruned_loss=0.09613, over 5702093.02 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3291, pruned_loss=0.08365, over 5755426.60 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3524, pruned_loss=0.09724, over 5701258.14 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:41:33,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4329, 1.5974, 1.5941, 1.4022], device='cuda:0'), covar=tensor([0.1992, 0.1969, 0.2286, 0.2157], device='cuda:0'), in_proj_covar=tensor([0.0494, 0.0751, 0.0721, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 23:42:11,488 INFO [train.py:968] (0/2) Epoch 27, batch 19050, giga_loss[loss=0.3324, simple_loss=0.3885, pruned_loss=0.1382, over 28864.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3558, pruned_loss=0.1022, over 5694078.95 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3295, pruned_loss=0.08382, over 5744883.36 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3567, pruned_loss=0.1033, over 5702103.31 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:42:25,974 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3503, 1.1963, 1.2043, 1.5886], device='cuda:0'), covar=tensor([0.0786, 0.0391, 0.0359, 0.0867], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 23:42:25,978 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:42:26,494 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1203923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:42:47,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.562e+02 1.476e+03 1.887e+03 2.491e+03 5.322e+03, threshold=3.774e+03, percent-clipped=14.0
+2023-03-13 23:42:48,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9505, 3.0256, 2.0482, 1.2772], device='cuda:0'), covar=tensor([0.8380, 0.2963, 0.4205, 0.6311], device='cuda:0'), in_proj_covar=tensor([0.1819, 0.1712, 0.1643, 0.1482], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-13 23:42:50,144 INFO [train.py:968] (0/2) Epoch 27, batch 19100, giga_loss[loss=0.2422, simple_loss=0.3232, pruned_loss=0.08062, over 28687.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3541, pruned_loss=0.1022, over 5704472.38 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3297, pruned_loss=0.08374, over 5750778.35 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3555, pruned_loss=0.1037, over 5703991.49 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:42:57,540 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-13 23:42:58,434 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1203966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:43:26,583 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1204000.pt
+2023-03-13 23:43:28,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7353, 4.5930, 4.3878, 2.1664], device='cuda:0'), covar=tensor([0.0531, 0.0695, 0.0699, 0.1974], device='cuda:0'), in_proj_covar=tensor([0.1269, 0.1171, 0.0990, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 23:43:30,845 INFO [train.py:968] (0/2) Epoch 27, batch 19150, giga_loss[loss=0.249, simple_loss=0.3295, pruned_loss=0.08425, over 28974.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3516, pruned_loss=0.1014, over 5703928.25 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3298, pruned_loss=0.08374, over 5752841.40 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.353, pruned_loss=0.1028, over 5701226.88 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:43:57,553 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8296, 1.9397, 1.8596, 1.8133], device='cuda:0'), covar=tensor([0.1976, 0.2131, 0.2293, 0.2148], device='cuda:0'), in_proj_covar=tensor([0.0496, 0.0752, 0.0723, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-13 23:44:05,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3521, 4.1982, 3.9837, 1.9715], device='cuda:0'), covar=tensor([0.0620, 0.0714, 0.0731, 0.2025], device='cuda:0'), in_proj_covar=tensor([0.1270, 0.1171, 0.0990, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 23:44:08,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.517e+02 1.340e+03 1.592e+03 2.131e+03 5.229e+03, threshold=3.183e+03, percent-clipped=4.0
+2023-03-13 23:44:10,016 INFO [train.py:968] (0/2) Epoch 27, batch 19200, giga_loss[loss=0.2789, simple_loss=0.3521, pruned_loss=0.1029, over 28270.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3492, pruned_loss=0.1004, over 5702379.90 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3305, pruned_loss=0.08419, over 5754353.77 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3506, pruned_loss=0.1019, over 5696254.67 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:44:20,403 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1204065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:44:22,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1204068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:44:49,940 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1204097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:44:56,009 INFO [train.py:968] (0/2) Epoch 27, batch 19250, giga_loss[loss=0.2597, simple_loss=0.3382, pruned_loss=0.09065, over 28508.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3487, pruned_loss=0.0993, over 5712096.34 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3305, pruned_loss=0.0842, over 5757017.71 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3502, pruned_loss=0.1009, over 5703717.66 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:45:34,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.409e+02 1.261e+03 1.397e+03 1.744e+03 3.642e+03, threshold=2.795e+03, percent-clipped=2.0
+2023-03-13 23:45:35,367 INFO [train.py:968] (0/2) Epoch 27, batch 19300, giga_loss[loss=0.2371, simple_loss=0.3206, pruned_loss=0.0768, over 28952.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3469, pruned_loss=0.09748, over 5709271.60 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3309, pruned_loss=0.08438, over 5757074.84 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.348, pruned_loss=0.09878, over 5701944.31 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:45:58,448 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-13 23:45:59,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1960, 1.1342, 4.0362, 3.3646], device='cuda:0'), covar=tensor([0.1722, 0.2792, 0.0439, 0.0914], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0665, 0.0989, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-13 23:46:22,434 INFO [train.py:968] (0/2) Epoch 27, batch 19350, libri_loss[loss=0.2543, simple_loss=0.3458, pruned_loss=0.08133, over 29215.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3423, pruned_loss=0.09485, over 5693284.15 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3308, pruned_loss=0.08416, over 5757951.63 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3435, pruned_loss=0.09629, over 5685591.73 frames. ], batch size: 97, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:46:53,700 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-13 23:47:05,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.899e+02 1.163e+03 1.610e+03 2.092e+03 1.009e+04, threshold=3.221e+03, percent-clipped=11.0
+2023-03-13 23:47:05,718 INFO [train.py:968] (0/2) Epoch 27, batch 19400, giga_loss[loss=0.2176, simple_loss=0.3045, pruned_loss=0.06539, over 28853.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3368, pruned_loss=0.09212, over 5688638.66 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3311, pruned_loss=0.08417, over 5761982.46 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3378, pruned_loss=0.09353, over 5677311.56 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:47:21,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4806, 1.6703, 1.6852, 1.2615], device='cuda:0'), covar=tensor([0.1814, 0.2817, 0.1545, 0.1872], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0710, 0.0977, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 23:47:42,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1204298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:47:48,696 INFO [train.py:968] (0/2) Epoch 27, batch 19450, giga_loss[loss=0.2465, simple_loss=0.3141, pruned_loss=0.08944, over 27577.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3323, pruned_loss=0.08987, over 5687108.61 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3317, pruned_loss=0.0845, over 5754715.72 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3327, pruned_loss=0.09092, over 5682227.27 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:48:01,650 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 23:48:10,636 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9915, 1.0967, 1.0952, 0.9484], device='cuda:0'), covar=tensor([0.2128, 0.2703, 0.1556, 0.2232], device='cuda:0'), in_proj_covar=tensor([0.2041, 0.1977, 0.1887, 0.2041], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-13 23:48:23,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6825, 1.6614, 1.8766, 1.4485], device='cuda:0'), covar=tensor([0.1805, 0.2571, 0.1464, 0.1789], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0711, 0.0978, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 23:48:24,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1204341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:48:32,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9024, 3.7512, 3.5300, 1.5980], device='cuda:0'), covar=tensor([0.0699, 0.0825, 0.0748, 0.2237], device='cuda:0'), in_proj_covar=tensor([0.1268, 0.1171, 0.0988, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-13 23:48:35,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.306e+02 1.066e+03 1.354e+03 2.010e+03 5.245e+03, threshold=2.708e+03, percent-clipped=6.0
+2023-03-13 23:48:36,685 INFO [train.py:968] (0/2) Epoch 27, batch 19500, libri_loss[loss=0.2565, simple_loss=0.3292, pruned_loss=0.09194, over 29659.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3282, pruned_loss=0.08819, over 5656696.20 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3314, pruned_loss=0.08442, over 5757330.94 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3287, pruned_loss=0.08914, over 5649262.37 frames. ], batch size: 73, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:49:14,540 INFO [train.py:968] (0/2) Epoch 27, batch 19550, giga_loss[loss=0.2566, simple_loss=0.337, pruned_loss=0.08815, over 28691.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3294, pruned_loss=0.08832, over 5668024.78 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3321, pruned_loss=0.08469, over 5762887.92 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3291, pruned_loss=0.08901, over 5653419.17 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:49:15,333 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1204406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:49:48,593 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1204441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:49:50,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1204444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:49:58,775 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.235e+02 1.201e+03 1.519e+03 2.024e+03 6.761e+03, threshold=3.039e+03, percent-clipped=10.0
+2023-03-13 23:49:59,600 INFO [train.py:968] (0/2) Epoch 27, batch 19600, giga_loss[loss=0.2377, simple_loss=0.3146, pruned_loss=0.08037, over 28833.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3304, pruned_loss=0.08869, over 5677525.40 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3326, pruned_loss=0.08482, over 5766888.53 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3296, pruned_loss=0.08922, over 5659726.44 frames. ], batch size: 66, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:50:12,643 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1204473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:50:22,873 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1204484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:50:25,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1204487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:50:40,975 INFO [train.py:968] (0/2) Epoch 27, batch 19650, giga_loss[loss=0.2321, simple_loss=0.3024, pruned_loss=0.08092, over 28450.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3288, pruned_loss=0.08789, over 5683853.95 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3328, pruned_loss=0.08485, over 5768830.39 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.328, pruned_loss=0.08834, over 5666970.51 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:50:42,805 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-13 23:50:49,704 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1204516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:50:58,207 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-13 23:51:18,569 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.903e+02 1.197e+03 1.466e+03 1.919e+03 3.909e+03, threshold=2.933e+03, percent-clipped=4.0
+2023-03-13 23:51:19,264 INFO [train.py:968] (0/2) Epoch 27, batch 19700, giga_loss[loss=0.2413, simple_loss=0.323, pruned_loss=0.07975, over 28775.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3278, pruned_loss=0.08769, over 5691818.64 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3328, pruned_loss=0.08468, over 5770241.33 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.327, pruned_loss=0.08827, over 5675312.52 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:51:37,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-13 23:52:00,472 INFO [train.py:968] (0/2) Epoch 27, batch 19750, giga_loss[loss=0.2241, simple_loss=0.307, pruned_loss=0.07066, over 28597.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.326, pruned_loss=0.08649, over 5703955.79 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3334, pruned_loss=0.08486, over 5774638.44 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3246, pruned_loss=0.08686, over 5684699.59 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:52:21,451 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1204631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:52:40,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.688e+02 1.175e+03 1.430e+03 2.171e+03 6.124e+03, threshold=2.860e+03, percent-clipped=8.0
+2023-03-13 23:52:40,820 INFO [train.py:968] (0/2) Epoch 27, batch 19800, giga_loss[loss=0.2034, simple_loss=0.2829, pruned_loss=0.06191, over 28239.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3239, pruned_loss=0.0856, over 5701598.89 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3335, pruned_loss=0.08488, over 5772958.97 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3226, pruned_loss=0.08589, over 5686983.68 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:53:16,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1266, 2.1500, 1.6308, 1.8325], device='cuda:0'), covar=tensor([0.1012, 0.0747, 0.1057, 0.1191], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0446, 0.0524, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 23:53:17,271 INFO [train.py:968] (0/2) Epoch 27, batch 19850, giga_loss[loss=0.2294, simple_loss=0.3091, pruned_loss=0.07485, over 28831.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3221, pruned_loss=0.08492, over 5703873.45 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3344, pruned_loss=0.08517, over 5764852.01 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3198, pruned_loss=0.08487, over 5697202.55 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:53:34,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3438, 1.4546, 1.5243, 1.1070], device='cuda:0'), covar=tensor([0.2038, 0.3218, 0.1539, 0.1683], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0710, 0.0978, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-13 23:53:36,523 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.95 vs. limit=2.0
+2023-03-13 23:53:58,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.490e+02 1.133e+03 1.378e+03 1.796e+03 4.462e+03, threshold=2.756e+03, percent-clipped=7.0
+2023-03-13 23:53:58,396 INFO [train.py:968] (0/2) Epoch 27, batch 19900, giga_loss[loss=0.2354, simple_loss=0.3153, pruned_loss=0.07781, over 29005.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3203, pruned_loss=0.08431, over 5713456.77 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3348, pruned_loss=0.08525, over 5767315.19 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3179, pruned_loss=0.08418, over 5705334.71 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:54:17,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1204781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:54:36,416 INFO [train.py:968] (0/2) Epoch 27, batch 19950, giga_loss[loss=0.2215, simple_loss=0.3044, pruned_loss=0.06926, over 28770.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3188, pruned_loss=0.08396, over 5718406.52 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3349, pruned_loss=0.08514, over 5769304.27 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3167, pruned_loss=0.08394, over 5709542.12 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:55:04,904 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-13 23:55:16,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.495e+02 1.038e+03 1.242e+03 1.654e+03 4.454e+03, threshold=2.485e+03, percent-clipped=6.0
+2023-03-13 23:55:16,785 INFO [train.py:968] (0/2) Epoch 27, batch 20000, giga_loss[loss=0.2229, simple_loss=0.3049, pruned_loss=0.07047, over 28918.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3164, pruned_loss=0.08275, over 5715318.76 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3352, pruned_loss=0.08516, over 5766994.06 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.3144, pruned_loss=0.08271, over 5709886.36 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:55:31,449 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4477, 3.6250, 1.6359, 1.6229], device='cuda:0'), covar=tensor([0.0998, 0.0361, 0.0882, 0.1299], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0559, 0.0403, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-13 23:55:55,434 INFO [train.py:968] (0/2) Epoch 27, batch 20050, giga_loss[loss=0.252, simple_loss=0.3254, pruned_loss=0.08931, over 28813.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3156, pruned_loss=0.08176, over 5720461.14 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3354, pruned_loss=0.08507, over 5769144.54 frames. ], giga_tot_loss[loss=0.2385, simple_loss=0.3135, pruned_loss=0.08177, over 5713620.01 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:56:09,864 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1204924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:56:12,051 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1204927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:56:33,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.816e+02 1.113e+03 1.338e+03 2.088e+03 6.500e+03, threshold=2.676e+03, percent-clipped=15.0
+2023-03-13 23:56:33,896 INFO [train.py:968] (0/2) Epoch 27, batch 20100, libri_loss[loss=0.2892, simple_loss=0.373, pruned_loss=0.1027, over 29531.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3195, pruned_loss=0.08404, over 5717322.13 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3361, pruned_loss=0.08534, over 5769251.42 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.3168, pruned_loss=0.08374, over 5710708.68 frames. ], batch size: 82, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:56:34,878 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1204956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:57:17,344 INFO [train.py:968] (0/2) Epoch 27, batch 20150, giga_loss[loss=0.2848, simple_loss=0.3423, pruned_loss=0.1136, over 23690.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3254, pruned_loss=0.08754, over 5716972.14 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3359, pruned_loss=0.08504, over 5773390.82 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3228, pruned_loss=0.08757, over 5706161.09 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:57:18,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:58:04,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.351e+02 1.436e+03 1.842e+03 2.559e+03 6.201e+03, threshold=3.685e+03, percent-clipped=22.0
+2023-03-13 23:58:04,177 INFO [train.py:968] (0/2) Epoch 27, batch 20200, giga_loss[loss=0.3203, simple_loss=0.3937, pruned_loss=0.1234, over 29006.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3326, pruned_loss=0.09199, over 5707192.20 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3367, pruned_loss=0.08533, over 5776964.97 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3296, pruned_loss=0.09187, over 5693773.04 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:58:24,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:58:35,939 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8032, 2.0890, 1.3597, 1.6910], device='cuda:0'), covar=tensor([0.0968, 0.0524, 0.1083, 0.0980], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0447, 0.0525, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-13 23:58:40,322 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4041, 3.2791, 1.6135, 1.5487], device='cuda:0'), covar=tensor([0.1022, 0.0296, 0.0881, 0.1377], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0561, 0.0403, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-13 23:58:50,024 INFO [train.py:968] (0/2) Epoch 27, batch 20250, giga_loss[loss=0.2941, simple_loss=0.3663, pruned_loss=0.1109, over 27911.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3387, pruned_loss=0.09585, over 5708862.49 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3367, pruned_loss=0.08537, over 5780451.84 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3363, pruned_loss=0.09601, over 5692798.82 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:59:06,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4919, 1.9720, 1.5566, 1.5548], device='cuda:0'), covar=tensor([0.0808, 0.0293, 0.0333, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-13 23:59:29,926 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:59:31,378 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1205149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:59:33,436 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1205152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:59:35,182 INFO [train.py:968] (0/2) Epoch 27, batch 20300, giga_loss[loss=0.2962, simple_loss=0.3749, pruned_loss=0.1087, over 28907.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.344, pruned_loss=0.09827, over 5699188.33 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.337, pruned_loss=0.08546, over 5781712.26 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.342, pruned_loss=0.09858, over 5683446.48 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:59:35,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.974e+02 1.457e+03 1.771e+03 2.275e+03 5.530e+03, threshold=3.542e+03, percent-clipped=6.0
+2023-03-13 23:59:49,871 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:59:58,785 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1205181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:00:19,142 INFO [train.py:968] (0/2) Epoch 27, batch 20350, libri_loss[loss=0.2514, simple_loss=0.3282, pruned_loss=0.08732, over 29566.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3484, pruned_loss=0.1, over 5683840.19 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3374, pruned_loss=0.08583, over 5764654.20 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3466, pruned_loss=0.1002, over 5684029.09 frames. ], batch size: 75, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:01:04,752 INFO [train.py:968] (0/2) Epoch 27, batch 20400, giga_loss[loss=0.2794, simple_loss=0.362, pruned_loss=0.09846, over 28913.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3539, pruned_loss=0.1027, over 5691057.89 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3378, pruned_loss=0.08587, over 5765031.28 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3524, pruned_loss=0.1033, over 5688798.24 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:01:05,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.478e+02 1.282e+03 1.528e+03 1.917e+03 4.995e+03, threshold=3.057e+03, percent-clipped=2.0
+2023-03-14 00:01:45,213 INFO [train.py:968] (0/2) Epoch 27, batch 20450, giga_loss[loss=0.2246, simple_loss=0.3115, pruned_loss=0.06887, over 28969.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3537, pruned_loss=0.1027, over 5689338.07 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3386, pruned_loss=0.08665, over 5769673.27 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3523, pruned_loss=0.103, over 5681307.50 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:02:26,018 INFO [train.py:968] (0/2) Epoch 27, batch 20500, giga_loss[loss=0.2268, simple_loss=0.3119, pruned_loss=0.07086, over 28917.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3503, pruned_loss=0.09991, over 5701037.41 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3392, pruned_loss=0.08728, over 5771288.35 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3489, pruned_loss=0.09993, over 5691223.90 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:02:27,378 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.401e+02 1.342e+03 1.704e+03 2.365e+03 4.672e+03, threshold=3.408e+03, percent-clipped=9.0
+2023-03-14 00:03:07,102 INFO [train.py:968] (0/2) Epoch 27, batch 20550, giga_loss[loss=0.2491, simple_loss=0.3351, pruned_loss=0.08159, over 28795.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3487, pruned_loss=0.09839, over 5689693.98 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3395, pruned_loss=0.08746, over 5759771.62 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3475, pruned_loss=0.09836, over 5691987.79 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:03:07,716 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-14 00:03:45,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:03:50,383 INFO [train.py:968] (0/2) Epoch 27, batch 20600, giga_loss[loss=0.2619, simple_loss=0.344, pruned_loss=0.08987, over 28839.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3472, pruned_loss=0.09641, over 5691965.59 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3399, pruned_loss=0.08762, over 5758341.45 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3462, pruned_loss=0.09652, over 5693587.88 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:03:52,815 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.288e+03 1.568e+03 2.085e+03 4.467e+03, threshold=3.135e+03, percent-clipped=4.0
+2023-03-14 00:04:00,306 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:04:23,236 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:04:30,587 INFO [train.py:968] (0/2) Epoch 27, batch 20650, giga_loss[loss=0.2872, simple_loss=0.3596, pruned_loss=0.1074, over 28614.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3498, pruned_loss=0.09847, over 5693369.36 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3396, pruned_loss=0.08749, over 5763571.70 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3494, pruned_loss=0.09896, over 5688136.55 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:04:45,090 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:05:05,106 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:05:12,292 INFO [train.py:968] (0/2) Epoch 27, batch 20700, giga_loss[loss=0.2842, simple_loss=0.3592, pruned_loss=0.1046, over 28689.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3511, pruned_loss=0.09961, over 5690667.46 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3391, pruned_loss=0.08705, over 5766035.69 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3516, pruned_loss=0.1007, over 5682437.72 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:05:13,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.716e+02 1.368e+03 1.700e+03 2.268e+03 8.674e+03, threshold=3.400e+03, percent-clipped=12.0
+2023-03-14 00:05:39,771 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1205592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:05:42,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1205595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:05:50,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2761, 3.1227, 1.4198, 1.5280], device='cuda:0'), covar=tensor([0.1076, 0.0344, 0.0882, 0.1345], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0560, 0.0402, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-14 00:05:50,554 INFO [train.py:968] (0/2) Epoch 27, batch 20750, giga_loss[loss=0.2671, simple_loss=0.3428, pruned_loss=0.09572, over 28791.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.351, pruned_loss=0.09943, over 5709417.88 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3394, pruned_loss=0.08724, over 5771258.73 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3516, pruned_loss=0.1007, over 5695171.40 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:05:53,194 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:06:04,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4196, 1.8033, 1.4060, 1.4029], device='cuda:0'), covar=tensor([0.2626, 0.2595, 0.3032, 0.2366], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1139, 0.1394, 0.1003], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 00:06:07,388 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1205624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:06:23,429 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205641.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 00:06:34,594 INFO [train.py:968] (0/2) Epoch 27, batch 20800, giga_loss[loss=0.2982, simple_loss=0.3699, pruned_loss=0.1133, over 28866.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3538, pruned_loss=0.1019, over 5709112.23 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3396, pruned_loss=0.08739, over 5761332.04 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3542, pruned_loss=0.1029, over 5705459.78 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:06:37,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.385e+02 1.366e+03 1.654e+03 2.034e+03 6.016e+03, threshold=3.307e+03, percent-clipped=5.0
+2023-03-14 00:06:44,598 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1205665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:06:48,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1205668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:02,363 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1205688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:05,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1205691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:09,262 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1205697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:14,550 INFO [train.py:968] (0/2) Epoch 27, batch 20850, giga_loss[loss=0.242, simple_loss=0.3347, pruned_loss=0.07468, over 28610.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.355, pruned_loss=0.1031, over 5705370.55 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3405, pruned_loss=0.08795, over 5762002.64 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3549, pruned_loss=0.1039, over 5700222.14 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:07:26,778 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1205720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:34,458 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:35,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4233, 3.4127, 1.4689, 1.5204], device='cuda:0'), covar=tensor([0.1065, 0.0289, 0.0935, 0.1406], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0560, 0.0403, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-14 00:07:52,920 INFO [train.py:968] (0/2) Epoch 27, batch 20900, giga_loss[loss=0.2652, simple_loss=0.3437, pruned_loss=0.09341, over 28990.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3533, pruned_loss=0.1014, over 5711457.06 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3405, pruned_loss=0.08821, over 5765347.10 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3537, pruned_loss=0.1023, over 5702253.66 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:07:54,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7328, 1.6807, 1.8925, 1.5470], device='cuda:0'), covar=tensor([0.1996, 0.2654, 0.1587, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0713, 0.0977, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 00:07:55,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.146e+02 1.326e+03 1.694e+03 2.443e+03 7.302e+03, threshold=3.389e+03, percent-clipped=7.0
+2023-03-14 00:08:32,402 INFO [train.py:968] (0/2) Epoch 27, batch 20950, giga_loss[loss=0.2767, simple_loss=0.3481, pruned_loss=0.1027, over 28505.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3538, pruned_loss=0.1013, over 5716887.41 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3406, pruned_loss=0.08853, over 5771181.74 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3546, pruned_loss=0.1022, over 5701966.40 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:09:02,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:09:03,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2191, 4.0590, 3.8170, 2.0143], device='cuda:0'), covar=tensor([0.0637, 0.0803, 0.0744, 0.1949], device='cuda:0'), in_proj_covar=tensor([0.1267, 0.1168, 0.0986, 0.0736], device='cuda:0'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 00:09:13,271 INFO [train.py:968] (0/2) Epoch 27, batch 21000, giga_loss[loss=0.2787, simple_loss=0.3604, pruned_loss=0.09853, over 28671.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.355, pruned_loss=0.1009, over 5723827.26 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3406, pruned_loss=0.08852, over 5773058.74 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3559, pruned_loss=0.1019, over 5709602.70 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:09:13,275 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 00:09:21,777 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2045, simple_loss=0.3123, pruned_loss=0.04838, over 944034.00 frames. 
+2023-03-14 00:09:21,778 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 00:09:23,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.278e+02 1.394e+03 1.685e+03 2.590e+03 7.991e+03, threshold=3.371e+03, percent-clipped=7.0
+2023-03-14 00:09:36,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:10:01,707 INFO [train.py:968] (0/2) Epoch 27, batch 21050, giga_loss[loss=0.2492, simple_loss=0.3289, pruned_loss=0.08476, over 28914.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3539, pruned_loss=0.1006, over 5719153.25 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3407, pruned_loss=0.08863, over 5763572.11 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3545, pruned_loss=0.1013, over 5715697.09 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:10:11,519 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:10:39,122 INFO [train.py:968] (0/2) Epoch 27, batch 21100, giga_loss[loss=0.2791, simple_loss=0.3469, pruned_loss=0.1056, over 28653.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3503, pruned_loss=0.09883, over 5706619.67 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3407, pruned_loss=0.08872, over 5757753.18 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3511, pruned_loss=0.09958, over 5708475.65 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 2.0
+2023-03-14 00:10:41,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.353e+02 1.223e+03 1.485e+03 1.992e+03 4.021e+03, threshold=2.969e+03, percent-clipped=3.0
+2023-03-14 00:11:01,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:03,166 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1205984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:05,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1205987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:16,096 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1206000.pt
+2023-03-14 00:11:19,379 INFO [train.py:968] (0/2) Epoch 27, batch 21150, giga_loss[loss=0.2854, simple_loss=0.3646, pruned_loss=0.1031, over 28879.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3494, pruned_loss=0.0983, over 5707645.51 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.341, pruned_loss=0.08889, over 5759731.84 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3499, pruned_loss=0.09891, over 5706301.68 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 2.0
+2023-03-14 00:11:27,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:27,648 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:27,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:11:29,169 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:46,140 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-14 00:11:51,168 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:59,438 INFO [train.py:968] (0/2) Epoch 27, batch 21200, giga_loss[loss=0.254, simple_loss=0.3333, pruned_loss=0.08733, over 28794.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3497, pruned_loss=0.09886, over 5714177.52 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3412, pruned_loss=0.08906, over 5758745.71 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.35, pruned_loss=0.09935, over 5713189.19 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:12:03,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.426e+02 1.274e+03 1.604e+03 2.233e+03 7.094e+03, threshold=3.208e+03, percent-clipped=10.0
+2023-03-14 00:12:07,996 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 00:12:39,629 INFO [train.py:968] (0/2) Epoch 27, batch 21250, giga_loss[loss=0.2909, simple_loss=0.3664, pruned_loss=0.1076, over 28819.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3506, pruned_loss=0.09998, over 5705857.14 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3416, pruned_loss=0.0894, over 5761613.39 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3507, pruned_loss=0.1003, over 5700952.66 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:12:40,424 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:12:54,709 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:12:56,567 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:13:18,375 INFO [train.py:968] (0/2) Epoch 27, batch 21300, giga_loss[loss=0.2487, simple_loss=0.3322, pruned_loss=0.0826, over 28757.00 frames. ], tot_loss[loss=0.274, simple_loss=0.35, pruned_loss=0.099, over 5708002.22 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3422, pruned_loss=0.08994, over 5752204.28 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3497, pruned_loss=0.09904, over 5710668.82 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:13:20,469 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:13:22,051 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.671e+02 1.188e+03 1.434e+03 1.739e+03 8.710e+03, threshold=2.868e+03, percent-clipped=6.0
+2023-03-14 00:13:22,344 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206159.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 00:13:24,289 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206162.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:13:38,028 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3819, 1.8850, 1.8605, 1.4694], device='cuda:0'), covar=tensor([0.0848, 0.0303, 0.0277, 0.1119], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0065, 0.0114], device='cuda:0')
+2023-03-14 00:13:46,228 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206191.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 00:13:57,051 INFO [train.py:968] (0/2) Epoch 27, batch 21350, giga_loss[loss=0.2744, simple_loss=0.3551, pruned_loss=0.0969, over 28951.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3498, pruned_loss=0.09834, over 5692491.26 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3425, pruned_loss=0.09023, over 5736026.03 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3495, pruned_loss=0.09825, over 5707280.25 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:14:30,079 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-14 00:14:31,153 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:14:33,499 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:14:36,108 INFO [train.py:968] (0/2) Epoch 27, batch 21400, giga_loss[loss=0.2508, simple_loss=0.3324, pruned_loss=0.08457, over 28558.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3484, pruned_loss=0.09806, over 5694248.84 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3424, pruned_loss=0.09035, over 5738710.54 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3484, pruned_loss=0.09806, over 5702589.11 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:14:39,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.953e+02 1.097e+03 1.308e+03 1.644e+03 5.800e+03, threshold=2.615e+03, percent-clipped=4.0
+2023-03-14 00:14:48,002 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4828, 4.3233, 4.0760, 1.9824], device='cuda:0'), covar=tensor([0.0562, 0.0704, 0.0743, 0.2055], device='cuda:0'), in_proj_covar=tensor([0.1273, 0.1168, 0.0986, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 00:14:57,569 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:15:06,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2912, 1.4567, 1.3071, 1.4878], device='cuda:0'), covar=tensor([0.0798, 0.0386, 0.0358, 0.0910], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 00:15:07,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:15:16,008 INFO [train.py:968] (0/2) Epoch 27, batch 21450, giga_loss[loss=0.2708, simple_loss=0.3499, pruned_loss=0.09586, over 28697.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3465, pruned_loss=0.09747, over 5683460.50 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3426, pruned_loss=0.09047, over 5731369.67 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3465, pruned_loss=0.09746, over 5696527.18 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:15:42,690 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3654, 1.7458, 1.6433, 1.5903], device='cuda:0'), covar=tensor([0.2463, 0.2154, 0.2572, 0.2253], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0753, 0.0721, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 00:15:54,358 INFO [train.py:968] (0/2) Epoch 27, batch 21500, giga_loss[loss=0.256, simple_loss=0.336, pruned_loss=0.08804, over 29057.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3444, pruned_loss=0.09656, over 5694246.42 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3426, pruned_loss=0.09064, over 5736544.06 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3444, pruned_loss=0.09659, over 5698874.63 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:15:57,168 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.276e+02 1.180e+03 1.521e+03 2.311e+03 7.419e+03, threshold=3.042e+03, percent-clipped=19.0
+2023-03-14 00:16:32,699 INFO [train.py:968] (0/2) Epoch 27, batch 21550, giga_loss[loss=0.2325, simple_loss=0.3153, pruned_loss=0.07487, over 28751.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3413, pruned_loss=0.09501, over 5697008.44 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3434, pruned_loss=0.0912, over 5740359.70 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3406, pruned_loss=0.09463, over 5696241.90 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:16:43,415 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1206415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:16:57,358 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1206436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:16:57,432 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:16:59,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:17:11,994 INFO [train.py:968] (0/2) Epoch 27, batch 21600, giga_loss[loss=0.2595, simple_loss=0.3377, pruned_loss=0.09069, over 28775.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3418, pruned_loss=0.09589, over 5692030.25 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3438, pruned_loss=0.09146, over 5742173.16 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3409, pruned_loss=0.09543, over 5688938.97 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:17:14,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.286e+02 1.231e+03 1.568e+03 2.040e+03 3.594e+03, threshold=3.137e+03, percent-clipped=6.0
+2023-03-14 00:17:23,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:17:24,166 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1206469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:17:51,511 INFO [train.py:968] (0/2) Epoch 27, batch 21650, giga_loss[loss=0.2275, simple_loss=0.3083, pruned_loss=0.07334, over 28971.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3395, pruned_loss=0.09489, over 5687238.32 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.344, pruned_loss=0.09172, over 5734399.47 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3385, pruned_loss=0.09436, over 5691466.47 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:18:09,261 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1206525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:18:32,993 INFO [train.py:968] (0/2) Epoch 27, batch 21700, giga_loss[loss=0.244, simple_loss=0.3169, pruned_loss=0.08558, over 28145.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3371, pruned_loss=0.09389, over 5695006.33 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3448, pruned_loss=0.09229, over 5732991.80 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3355, pruned_loss=0.09299, over 5699053.70 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:18:34,653 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-14 00:18:38,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.244e+02 1.199e+03 1.466e+03 1.958e+03 4.316e+03, threshold=2.932e+03, percent-clipped=2.0
+2023-03-14 00:19:15,413 INFO [train.py:968] (0/2) Epoch 27, batch 21750, giga_loss[loss=0.2181, simple_loss=0.2997, pruned_loss=0.0683, over 28948.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3334, pruned_loss=0.09193, over 5696891.92 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3451, pruned_loss=0.09255, over 5733523.84 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3318, pruned_loss=0.09096, over 5699345.96 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:19:29,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7167, 1.9207, 1.7893, 1.6042], device='cuda:0'), covar=tensor([0.2177, 0.2711, 0.2435, 0.2797], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0750, 0.0718, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 00:19:45,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4178, 3.5620, 1.5430, 1.5889], device='cuda:0'), covar=tensor([0.0991, 0.0402, 0.0935, 0.1320], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0560, 0.0402, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-14 00:19:49,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7144, 1.7972, 1.9116, 1.4682], device='cuda:0'), covar=tensor([0.2098, 0.2560, 0.1611, 0.1803], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0713, 0.0978, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 00:19:50,413 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-14 00:19:51,867 INFO [train.py:968] (0/2) Epoch 27, batch 21800, giga_loss[loss=0.2209, simple_loss=0.3039, pruned_loss=0.06897, over 29105.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3313, pruned_loss=0.09068, over 5708185.55 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3454, pruned_loss=0.09303, over 5735330.72 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3292, pruned_loss=0.08941, over 5707261.70 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:19:52,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5062, 2.7241, 2.6748, 2.1789], device='cuda:0'), covar=tensor([0.2851, 0.2079, 0.2238, 0.2912], device='cuda:0'), in_proj_covar=tensor([0.2037, 0.1982, 0.1904, 0.2045], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 00:19:55,335 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.311e+02 1.265e+03 1.469e+03 1.874e+03 4.755e+03, threshold=2.938e+03, percent-clipped=4.0
+2023-03-14 00:20:17,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6837, 1.6246, 1.8444, 1.4423], device='cuda:0'), covar=tensor([0.1994, 0.2533, 0.1581, 0.1820], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0713, 0.0978, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 00:20:24,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1206694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:20:28,700 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7533, 1.8983, 1.9308, 1.5051], device='cuda:0'), covar=tensor([0.1983, 0.2532, 0.1597, 0.1767], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0713, 0.0978, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 00:20:33,487 INFO [train.py:968] (0/2) Epoch 27, batch 21850, giga_loss[loss=0.3135, simple_loss=0.3701, pruned_loss=0.1284, over 23821.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3318, pruned_loss=0.09103, over 5696518.65 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3454, pruned_loss=0.0931, over 5727525.16 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3301, pruned_loss=0.08995, over 5702809.60 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:20:33,968 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-14 00:20:50,937 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 00:21:04,426 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-14 00:21:14,269 INFO [train.py:968] (0/2) Epoch 27, batch 21900, giga_loss[loss=0.2575, simple_loss=0.3333, pruned_loss=0.09082, over 28879.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3346, pruned_loss=0.09214, over 5699517.42 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3457, pruned_loss=0.09348, over 5727883.34 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3327, pruned_loss=0.09091, over 5703676.92 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:21:18,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.723e+02 1.235e+03 1.423e+03 1.819e+03 3.557e+03, threshold=2.846e+03, percent-clipped=1.0
+2023-03-14 00:21:46,237 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:21:56,880 INFO [train.py:968] (0/2) Epoch 27, batch 21950, giga_loss[loss=0.2975, simple_loss=0.3695, pruned_loss=0.1128, over 27603.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3394, pruned_loss=0.09404, over 5694104.46 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3462, pruned_loss=0.0939, over 5731795.28 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3371, pruned_loss=0.09264, over 5692760.06 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:21:59,035 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3015, 1.4994, 1.5040, 1.4019], device='cuda:0'), covar=tensor([0.1602, 0.1185, 0.1687, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0750, 0.0720, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 00:22:02,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:22:30,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:22:39,301 INFO [train.py:968] (0/2) Epoch 27, batch 22000, giga_loss[loss=0.2842, simple_loss=0.3564, pruned_loss=0.106, over 28770.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3409, pruned_loss=0.09401, over 5700041.26 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3461, pruned_loss=0.0939, over 5733418.71 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3392, pruned_loss=0.09291, over 5697248.77 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:22:43,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.760e+02 1.210e+03 1.586e+03 2.158e+03 8.404e+03, threshold=3.171e+03, percent-clipped=11.0
+2023-03-14 00:22:53,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1549, 0.9319, 1.0614, 1.3394], device='cuda:0'), covar=tensor([0.0765, 0.0385, 0.0367, 0.0930], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0114], device='cuda:0')
+2023-03-14 00:23:15,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:23:19,165 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 00:23:19,980 INFO [train.py:968] (0/2) Epoch 27, batch 22050, giga_loss[loss=0.2653, simple_loss=0.3556, pruned_loss=0.08744, over 28631.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3409, pruned_loss=0.09338, over 5708309.38 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3464, pruned_loss=0.09425, over 5738655.44 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3391, pruned_loss=0.09219, over 5700700.16 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:23:43,237 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:23:45,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:00,066 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:00,183 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-14 00:24:00,398 INFO [train.py:968] (0/2) Epoch 27, batch 22100, giga_loss[loss=0.2749, simple_loss=0.3564, pruned_loss=0.09675, over 27704.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3413, pruned_loss=0.09388, over 5703610.08 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3477, pruned_loss=0.09534, over 5740055.48 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3384, pruned_loss=0.09186, over 5694558.92 frames. ], batch size: 474, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:24:02,914 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:04,757 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.043e+02 1.305e+03 1.830e+03 3.001e+03 5.986e+03, threshold=3.660e+03, percent-clipped=22.0
+2023-03-14 00:24:09,539 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:26,299 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:27,976 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:30,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:43,153 INFO [train.py:968] (0/2) Epoch 27, batch 22150, giga_loss[loss=0.2649, simple_loss=0.3499, pruned_loss=0.08999, over 28350.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3421, pruned_loss=0.09485, over 5705133.74 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3474, pruned_loss=0.09534, over 5743681.02 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3399, pruned_loss=0.09321, over 5693490.17 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:24:46,112 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2736, 2.0399, 1.5644, 0.4981], device='cuda:0'), covar=tensor([0.6140, 0.2888, 0.4491, 0.7344], device='cuda:0'), in_proj_covar=tensor([0.1820, 0.1703, 0.1642, 0.1482], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 00:24:53,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1207019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:25:12,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1207043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:25:14,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1207046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:25:18,744 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9111, 4.6621, 2.0349, 2.1097], device='cuda:0'), covar=tensor([0.0846, 0.0357, 0.0786, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0562, 0.0404, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-14 00:25:22,460 INFO [train.py:968] (0/2) Epoch 27, batch 22200, giga_loss[loss=0.2686, simple_loss=0.3458, pruned_loss=0.09566, over 29024.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3426, pruned_loss=0.0956, over 5689969.50 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.348, pruned_loss=0.09582, over 5726714.59 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3401, pruned_loss=0.0938, over 5695164.64 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:25:25,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.883e+02 1.368e+03 1.687e+03 2.377e+03 6.305e+03, threshold=3.374e+03, percent-clipped=7.0
+2023-03-14 00:25:33,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1207069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:25:39,467 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1207075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:25:43,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9157, 1.0749, 2.8822, 2.7357], device='cuda:0'), covar=tensor([0.1637, 0.2666, 0.0596, 0.0970], device='cuda:0'), in_proj_covar=tensor([0.0795, 0.0664, 0.0989, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 00:26:03,881 INFO [train.py:968] (0/2) Epoch 27, batch 22250, giga_loss[loss=0.2849, simple_loss=0.3613, pruned_loss=0.1042, over 28525.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3446, pruned_loss=0.09701, over 5693034.07 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3481, pruned_loss=0.09609, over 5726605.73 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3424, pruned_loss=0.09533, over 5696925.21 frames. ], batch size: 65, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:26:45,744 INFO [train.py:968] (0/2) Epoch 27, batch 22300, giga_loss[loss=0.2758, simple_loss=0.3556, pruned_loss=0.09797, over 28890.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3481, pruned_loss=0.09864, over 5705521.65 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3487, pruned_loss=0.09668, over 5729909.18 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3457, pruned_loss=0.09679, over 5705062.03 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:26:49,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.710e+02 1.337e+03 1.614e+03 2.228e+03 7.952e+03, threshold=3.228e+03, percent-clipped=7.0
+2023-03-14 00:27:22,764 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 00:27:26,357 INFO [train.py:968] (0/2) Epoch 27, batch 22350, giga_loss[loss=0.3012, simple_loss=0.3753, pruned_loss=0.1135, over 28979.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3498, pruned_loss=0.09923, over 5711742.24 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3489, pruned_loss=0.09689, over 5734348.86 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3478, pruned_loss=0.09763, over 5706992.60 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:27:31,995 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1207212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:27:34,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1207215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:27:56,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1207244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:28:03,765 INFO [train.py:968] (0/2) Epoch 27, batch 22400, giga_loss[loss=0.287, simple_loss=0.3518, pruned_loss=0.1111, over 28746.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3518, pruned_loss=0.1006, over 5719270.60 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3497, pruned_loss=0.09761, over 5738002.50 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3494, pruned_loss=0.09869, over 5711832.15 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:28:09,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.994e+02 1.325e+03 1.598e+03 2.073e+03 4.273e+03, threshold=3.197e+03, percent-clipped=5.0
+2023-03-14 00:28:42,236 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-14 00:28:47,172 INFO [train.py:968] (0/2) Epoch 27, batch 22450, giga_loss[loss=0.2609, simple_loss=0.3403, pruned_loss=0.09074, over 28972.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3523, pruned_loss=0.101, over 5717188.84 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3499, pruned_loss=0.09786, over 5739478.23 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3503, pruned_loss=0.09935, over 5709345.92 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:29:10,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9536, 1.1778, 0.8752, 0.7978], device='cuda:0'), covar=tensor([0.0998, 0.0525, 0.1303, 0.1247], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0445, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 00:29:20,735 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3183, 3.4907, 1.4726, 1.4144], device='cuda:0'), covar=tensor([0.1031, 0.0356, 0.0977, 0.1398], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0565, 0.0405, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 00:29:28,265 INFO [train.py:968] (0/2) Epoch 27, batch 22500, giga_loss[loss=0.2379, simple_loss=0.3255, pruned_loss=0.07512, over 28648.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3526, pruned_loss=0.1015, over 5719638.18 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3501, pruned_loss=0.09817, over 5741723.36 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3509, pruned_loss=0.09995, over 5711034.88 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:29:35,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.125e+02 1.388e+03 1.810e+03 2.345e+03 7.224e+03, threshold=3.621e+03, percent-clipped=8.0
+2023-03-14 00:30:11,013 INFO [train.py:968] (0/2) Epoch 27, batch 22550, giga_loss[loss=0.2773, simple_loss=0.3533, pruned_loss=0.1007, over 28699.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3497, pruned_loss=0.0998, over 5723510.47 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.35, pruned_loss=0.09815, over 5743576.51 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3485, pruned_loss=0.09863, over 5715068.78 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:30:53,791 INFO [train.py:968] (0/2) Epoch 27, batch 22600, giga_loss[loss=0.2912, simple_loss=0.3567, pruned_loss=0.1128, over 27618.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3466, pruned_loss=0.09819, over 5716590.30 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3502, pruned_loss=0.0983, over 5735884.09 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3453, pruned_loss=0.09712, over 5717085.99 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:30:56,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1207459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:30:58,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.353e+02 1.173e+03 1.477e+03 1.963e+03 4.171e+03, threshold=2.954e+03, percent-clipped=3.0
+2023-03-14 00:31:16,235 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5488, 1.8809, 1.4811, 1.7261], device='cuda:0'), covar=tensor([0.2801, 0.2873, 0.3308, 0.2550], device='cuda:0'), in_proj_covar=tensor([0.1572, 0.1135, 0.1390, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 00:31:25,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5953, 1.4222, 1.5959, 1.2165], device='cuda:0'), covar=tensor([0.2257, 0.3471, 0.1800, 0.1917], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0711, 0.0975, 0.0876], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 00:31:31,125 INFO [train.py:968] (0/2) Epoch 27, batch 22650, libri_loss[loss=0.271, simple_loss=0.3508, pruned_loss=0.09556, over 29572.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3445, pruned_loss=0.09747, over 5717608.43 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3507, pruned_loss=0.09892, over 5739651.17 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.343, pruned_loss=0.09605, over 5713987.04 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:32:09,872 INFO [train.py:968] (0/2) Epoch 27, batch 22700, giga_loss[loss=0.3156, simple_loss=0.3838, pruned_loss=0.1237, over 28922.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3458, pruned_loss=0.0973, over 5710098.38 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3511, pruned_loss=0.09943, over 5731172.65 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3439, pruned_loss=0.09559, over 5714125.81 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:32:10,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3235, 1.1868, 3.6215, 3.2491], device='cuda:0'), covar=tensor([0.1580, 0.2972, 0.0421, 0.1500], device='cuda:0'), in_proj_covar=tensor([0.0796, 0.0663, 0.0991, 0.0963], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 00:32:17,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.702e+02 1.208e+03 1.607e+03 2.306e+03 8.313e+03, threshold=3.214e+03, percent-clipped=11.0
+2023-03-14 00:32:50,865 INFO [train.py:968] (0/2) Epoch 27, batch 22750, giga_loss[loss=0.2813, simple_loss=0.3584, pruned_loss=0.1021, over 28636.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3478, pruned_loss=0.09771, over 5701969.82 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3517, pruned_loss=0.09998, over 5725018.48 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3456, pruned_loss=0.09575, over 5710244.95 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:32:59,486 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1207616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:33:29,103 INFO [train.py:968] (0/2) Epoch 27, batch 22800, giga_loss[loss=0.2934, simple_loss=0.3692, pruned_loss=0.1088, over 28379.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.348, pruned_loss=0.09811, over 5718516.43 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3524, pruned_loss=0.101, over 5733842.58 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3454, pruned_loss=0.09549, over 5716375.69 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:33:33,705 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.507e+02 1.370e+03 1.822e+03 2.817e+03 1.527e+04, threshold=3.643e+03, percent-clipped=22.0
+2023-03-14 00:34:10,769 INFO [train.py:968] (0/2) Epoch 27, batch 22850, libri_loss[loss=0.2625, simple_loss=0.3353, pruned_loss=0.09482, over 29566.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3464, pruned_loss=0.09797, over 5721607.17 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3525, pruned_loss=0.1012, over 5737246.66 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3442, pruned_loss=0.09567, over 5716540.80 frames. ], batch size: 79, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:34:11,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8992, 2.1123, 1.7435, 2.0194], device='cuda:0'), covar=tensor([0.2723, 0.2942, 0.3385, 0.2765], device='cuda:0'), in_proj_covar=tensor([0.1575, 0.1136, 0.1391, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 00:34:50,022 INFO [train.py:968] (0/2) Epoch 27, batch 22900, giga_loss[loss=0.2821, simple_loss=0.348, pruned_loss=0.1081, over 28809.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3441, pruned_loss=0.09752, over 5724294.98 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3525, pruned_loss=0.1013, over 5738168.47 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3422, pruned_loss=0.09559, over 5719386.01 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:34:54,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.394e+02 1.333e+03 1.534e+03 1.854e+03 6.217e+03, threshold=3.069e+03, percent-clipped=1.0
+2023-03-14 00:35:11,258 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.55 vs. limit=5.0
+2023-03-14 00:35:29,616 INFO [train.py:968] (0/2) Epoch 27, batch 22950, giga_loss[loss=0.2954, simple_loss=0.3468, pruned_loss=0.122, over 28706.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3438, pruned_loss=0.09874, over 5721469.51 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.353, pruned_loss=0.1017, over 5738568.62 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3416, pruned_loss=0.09675, over 5716660.39 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:35:31,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3183, 3.4241, 1.5437, 1.4738], device='cuda:0'), covar=tensor([0.1045, 0.0427, 0.0982, 0.1429], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0564, 0.0405, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 00:35:50,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7894, 2.1310, 1.9130, 1.7785], device='cuda:0'), covar=tensor([0.1797, 0.1686, 0.2014, 0.1845], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0757, 0.0726, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 00:35:52,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1207834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:36:07,422 INFO [train.py:968] (0/2) Epoch 27, batch 23000, libri_loss[loss=0.2921, simple_loss=0.3598, pruned_loss=0.1122, over 29255.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3424, pruned_loss=0.09834, over 5729150.31 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3533, pruned_loss=0.102, over 5745256.42 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.34, pruned_loss=0.09623, over 5718471.57 frames. ], batch size: 94, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:36:13,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.067e+02 1.270e+03 1.574e+03 2.368e+03 7.852e+03, threshold=3.148e+03, percent-clipped=11.0
+2023-03-14 00:36:14,365 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2493, 1.0919, 4.0326, 3.1189], device='cuda:0'), covar=tensor([0.1747, 0.2965, 0.0452, 0.0999], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0662, 0.0990, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 00:36:33,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4105, 1.8125, 1.6212, 1.6974], device='cuda:0'), covar=tensor([0.0732, 0.0288, 0.0316, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0191, 0.0120, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0103, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 00:36:45,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6795, 1.8532, 1.4147, 1.4751], device='cuda:0'), covar=tensor([0.1028, 0.0750, 0.1041, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0446, 0.0521, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 00:36:45,670 INFO [train.py:968] (0/2) Epoch 27, batch 23050, giga_loss[loss=0.2507, simple_loss=0.3303, pruned_loss=0.08555, over 28322.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3397, pruned_loss=0.09721, over 5722515.34 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3536, pruned_loss=0.1024, over 5746044.51 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3373, pruned_loss=0.09511, over 5712997.17 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:37:14,096 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-14 00:37:24,611 INFO [train.py:968] (0/2) Epoch 27, batch 23100, giga_loss[loss=0.2525, simple_loss=0.3272, pruned_loss=0.0889, over 28761.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3364, pruned_loss=0.09574, over 5721678.82 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3543, pruned_loss=0.1031, over 5741561.78 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3332, pruned_loss=0.09316, over 5718132.74 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 00:37:28,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6423, 4.5181, 4.2292, 2.4067], device='cuda:0'), covar=tensor([0.0533, 0.0615, 0.0678, 0.1659], device='cuda:0'), in_proj_covar=tensor([0.1280, 0.1176, 0.0995, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 00:37:31,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.954e+02 1.220e+03 1.458e+03 1.864e+03 4.327e+03, threshold=2.916e+03, percent-clipped=3.0
+2023-03-14 00:37:43,096 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1207977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:37:44,382 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-14 00:37:44,961 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1207980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:37:46,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9939, 1.4999, 5.0728, 3.5908], device='cuda:0'), covar=tensor([0.1432, 0.2759, 0.0343, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0793, 0.0663, 0.0990, 0.0962], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 00:37:54,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1207991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:37:56,777 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1207995.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:38:00,959 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1208000.pt
+2023-03-14 00:38:04,274 INFO [train.py:968] (0/2) Epoch 27, batch 23150, libri_loss[loss=0.3343, simple_loss=0.3882, pruned_loss=0.1402, over 29201.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3332, pruned_loss=0.09417, over 5724762.66 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3549, pruned_loss=0.1037, over 5747153.62 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3295, pruned_loss=0.09124, over 5716172.77 frames. ], batch size: 97, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 00:38:07,110 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:38:12,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9619, 1.1809, 1.1603, 0.9196], device='cuda:0'), covar=tensor([0.2748, 0.3130, 0.1794, 0.2479], device='cuda:0'), in_proj_covar=tensor([0.2062, 0.2009, 0.1924, 0.2058], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 00:38:35,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3230, 4.1731, 3.9252, 1.8055], device='cuda:0'), covar=tensor([0.0589, 0.0716, 0.0714, 0.2229], device='cuda:0'), in_proj_covar=tensor([0.1276, 0.1172, 0.0992, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 00:38:42,980 INFO [train.py:968] (0/2) Epoch 27, batch 23200, giga_loss[loss=0.2252, simple_loss=0.3026, pruned_loss=0.07384, over 28649.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3317, pruned_loss=0.09331, over 5713674.03 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3547, pruned_loss=0.1039, over 5729958.79 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3281, pruned_loss=0.09049, over 5721475.47 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:38:50,309 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.480e+02 1.370e+03 1.797e+03 2.316e+03 5.930e+03, threshold=3.595e+03, percent-clipped=13.0
+2023-03-14 00:39:22,623 INFO [train.py:968] (0/2) Epoch 27, batch 23250, giga_loss[loss=0.2449, simple_loss=0.3175, pruned_loss=0.08617, over 28902.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.334, pruned_loss=0.09392, over 5703228.57 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3548, pruned_loss=0.1042, over 5724385.18 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3304, pruned_loss=0.0911, over 5713244.73 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:39:24,133 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1208107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:39:39,257 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1208126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:39:46,426 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1208134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:39:50,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1208137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:40:05,006 INFO [train.py:968] (0/2) Epoch 27, batch 23300, giga_loss[loss=0.2396, simple_loss=0.3157, pruned_loss=0.08175, over 28487.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3376, pruned_loss=0.09508, over 5703093.57 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3553, pruned_loss=0.1044, over 5727944.59 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.334, pruned_loss=0.09247, over 5707560.32 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:40:12,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.749e+02 1.275e+03 1.558e+03 2.028e+03 4.357e+03, threshold=3.115e+03, percent-clipped=3.0
+2023-03-14 00:40:13,584 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:40:29,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1208186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:40:43,829 INFO [train.py:968] (0/2) Epoch 27, batch 23350, giga_loss[loss=0.2425, simple_loss=0.3285, pruned_loss=0.07824, over 28971.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3398, pruned_loss=0.09545, over 5706681.75 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3557, pruned_loss=0.1049, over 5731755.72 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.336, pruned_loss=0.09269, over 5706479.85 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:41:26,544 INFO [train.py:968] (0/2) Epoch 27, batch 23400, giga_loss[loss=0.2805, simple_loss=0.3548, pruned_loss=0.1031, over 28780.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3416, pruned_loss=0.09569, over 5714469.78 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3552, pruned_loss=0.1047, over 5732803.95 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3389, pruned_loss=0.09357, over 5713076.49 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:41:33,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.780e+02 1.368e+03 1.632e+03 2.165e+03 5.530e+03, threshold=3.264e+03, percent-clipped=6.0
+2023-03-14 00:42:08,600 INFO [train.py:968] (0/2) Epoch 27, batch 23450, giga_loss[loss=0.259, simple_loss=0.3408, pruned_loss=0.08864, over 28445.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3433, pruned_loss=0.09675, over 5712258.37 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3554, pruned_loss=0.1048, over 5726701.10 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3407, pruned_loss=0.09469, over 5716912.62 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:42:15,985 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3602, 3.6813, 1.6446, 1.6068], device='cuda:0'), covar=tensor([0.1035, 0.0362, 0.0904, 0.1343], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0565, 0.0405, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 00:42:19,790 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-14 00:42:57,007 INFO [train.py:968] (0/2) Epoch 27, batch 23500, giga_loss[loss=0.396, simple_loss=0.4182, pruned_loss=0.1869, over 23514.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3499, pruned_loss=0.1025, over 5700035.28 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3558, pruned_loss=0.1052, over 5729156.77 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3472, pruned_loss=0.1004, over 5700946.41 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:43:03,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.674e+03 2.287e+03 3.326e+03 1.088e+04, threshold=4.575e+03, percent-clipped=25.0
+2023-03-14 00:43:12,267 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1208370.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 00:43:14,878 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6625, 4.5082, 4.2610, 2.1552], device='cuda:0'), covar=tensor([0.0543, 0.0656, 0.0678, 0.1870], device='cuda:0'), in_proj_covar=tensor([0.1282, 0.1180, 0.0997, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 00:43:45,569 INFO [train.py:968] (0/2) Epoch 27, batch 23550, libri_loss[loss=0.292, simple_loss=0.3612, pruned_loss=0.1114, over 29648.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3552, pruned_loss=0.1064, over 5697407.61 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3556, pruned_loss=0.1052, over 5732892.99 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3533, pruned_loss=0.1048, over 5694134.65 frames. ], batch size: 91, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:44:00,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7566, 1.7318, 1.8769, 1.4740], device='cuda:0'), covar=tensor([0.1849, 0.2612, 0.1517, 0.1874], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0710, 0.0974, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 00:44:31,564 INFO [train.py:968] (0/2) Epoch 27, batch 23600, giga_loss[loss=0.2881, simple_loss=0.3636, pruned_loss=0.1063, over 29035.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3619, pruned_loss=0.1117, over 5689601.65 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3556, pruned_loss=0.1054, over 5736001.33 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3604, pruned_loss=0.1103, over 5682673.44 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 00:44:39,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.758e+03 2.173e+03 2.901e+03 6.344e+03, threshold=4.347e+03, percent-clipped=6.0
+2023-03-14 00:44:57,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1208482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:45:14,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1208501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:45:17,054 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4196, 1.7556, 1.6195, 1.4578], device='cuda:0'), covar=tensor([0.2181, 0.2048, 0.2377, 0.2229], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0757, 0.0727, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 00:45:19,049 INFO [train.py:968] (0/2) Epoch 27, batch 23650, giga_loss[loss=0.3028, simple_loss=0.3735, pruned_loss=0.116, over 28849.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3675, pruned_loss=0.1162, over 5692718.56 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3554, pruned_loss=0.1053, over 5739542.63 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3667, pruned_loss=0.1154, over 5683195.46 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 00:45:26,996 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1208513.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:45:31,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1208516.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:45:59,314 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208545.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:46:07,745 INFO [train.py:968] (0/2) Epoch 27, batch 23700, giga_loss[loss=0.3288, simple_loss=0.392, pruned_loss=0.1328, over 28730.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3745, pruned_loss=0.122, over 5681573.63 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3554, pruned_loss=0.1054, over 5732557.89 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3742, pruned_loss=0.1216, over 5678819.75 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:46:15,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1208561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:46:18,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.176e+03 1.839e+03 2.233e+03 3.016e+03 9.087e+03, threshold=4.465e+03, percent-clipped=9.0
+2023-03-14 00:46:35,694 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3014, 2.3768, 1.9901, 2.0936], device='cuda:0'), covar=tensor([0.0684, 0.0452, 0.0703, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0447, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 00:46:54,382 INFO [train.py:968] (0/2) Epoch 27, batch 23750, giga_loss[loss=0.2917, simple_loss=0.3641, pruned_loss=0.1097, over 29015.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3774, pruned_loss=0.1248, over 5671289.56 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3555, pruned_loss=0.1054, over 5729211.68 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3778, pruned_loss=0.125, over 5670206.19 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:47:05,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4867, 1.5466, 1.6454, 1.4795], device='cuda:0'), covar=tensor([0.2642, 0.2771, 0.2004, 0.2343], device='cuda:0'), in_proj_covar=tensor([0.2061, 0.2014, 0.1928, 0.2059], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 00:47:09,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4647, 2.0096, 1.4060, 0.8294], device='cuda:0'), covar=tensor([0.6167, 0.3277, 0.3830, 0.6345], device='cuda:0'), in_proj_covar=tensor([0.1829, 0.1728, 0.1658, 0.1497], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 00:47:12,246 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1208625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:47:14,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1208628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:47:30,710 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1208644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:47:33,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1208647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:47:39,337 INFO [train.py:968] (0/2) Epoch 27, batch 23800, giga_loss[loss=0.329, simple_loss=0.3894, pruned_loss=0.1343, over 28767.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3777, pruned_loss=0.126, over 5670633.35 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3547, pruned_loss=0.1051, over 5734334.53 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3793, pruned_loss=0.127, over 5663574.62 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:47:40,412 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 00:47:40,824 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:47:51,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.198e+03 1.880e+03 2.631e+03 3.646e+03 9.280e+03, threshold=5.261e+03, percent-clipped=16.0
+2023-03-14 00:48:03,635 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:48:32,889 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1208704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:48:33,504 INFO [train.py:968] (0/2) Epoch 27, batch 23850, giga_loss[loss=0.3065, simple_loss=0.3745, pruned_loss=0.1193, over 29009.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3833, pruned_loss=0.1322, over 5657494.86 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3547, pruned_loss=0.105, over 5735204.86 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3848, pruned_loss=0.1331, over 5650836.48 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:48:36,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1208707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:48:39,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5751, 1.9588, 1.3236, 1.5353], device='cuda:0'), covar=tensor([0.0891, 0.0451, 0.0968, 0.0959], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0447, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 00:48:59,076 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8995, 5.7432, 5.4877, 3.3635], device='cuda:0'), covar=tensor([0.0495, 0.0607, 0.0732, 0.1301], device='cuda:0'), in_proj_covar=tensor([0.1287, 0.1185, 0.1002, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 00:49:05,369 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:49:24,278 INFO [train.py:968] (0/2) Epoch 27, batch 23900, libri_loss[loss=0.3246, simple_loss=0.3899, pruned_loss=0.1296, over 27852.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3874, pruned_loss=0.1368, over 5648614.33 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3548, pruned_loss=0.1052, over 5735116.07 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3889, pruned_loss=0.138, over 5642157.75 frames. ], batch size: 116, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:49:36,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.298e+03 1.979e+03 2.765e+03 3.969e+03 9.657e+03, threshold=5.529e+03, percent-clipped=12.0
+2023-03-14 00:49:43,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5324, 1.7068, 1.3091, 1.2612], device='cuda:0'), covar=tensor([0.0933, 0.0518, 0.0935, 0.1087], device='cuda:0'), in_proj_covar=tensor([0.0405, 0.0447, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 00:50:12,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6525, 1.6901, 1.7854, 1.3978], device='cuda:0'), covar=tensor([0.1806, 0.2659, 0.1513, 0.1848], device='cuda:0'), in_proj_covar=tensor([0.0924, 0.0708, 0.0969, 0.0870], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:0')
+2023-03-14 00:50:22,870 INFO [train.py:968] (0/2) Epoch 27, batch 23950, giga_loss[loss=0.3021, simple_loss=0.3704, pruned_loss=0.1169, over 28671.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3904, pruned_loss=0.1383, over 5654703.56 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3555, pruned_loss=0.1057, over 5736572.76 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3916, pruned_loss=0.1393, over 5646889.54 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:51:14,837 INFO [train.py:968] (0/2) Epoch 27, batch 24000, giga_loss[loss=0.4056, simple_loss=0.4285, pruned_loss=0.1913, over 23728.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3916, pruned_loss=0.1403, over 5642787.72 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3558, pruned_loss=0.106, over 5736562.98 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3933, pruned_loss=0.1417, over 5634295.67 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 00:51:14,843 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 00:51:19,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5092, 1.8201, 1.4541, 1.4260], device='cuda:0'), covar=tensor([0.3064, 0.3009, 0.3436, 0.2590], device='cuda:0'), in_proj_covar=tensor([0.1582, 0.1142, 0.1398, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 00:51:23,048 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2026, simple_loss=0.3107, pruned_loss=0.04727, over 944034.00 frames. 
+2023-03-14 00:51:23,049 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 00:51:30,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.151e+03 1.985e+03 2.422e+03 3.390e+03 8.830e+03, threshold=4.845e+03, percent-clipped=7.0
+2023-03-14 00:52:06,668 INFO [train.py:968] (0/2) Epoch 27, batch 24050, giga_loss[loss=0.2707, simple_loss=0.342, pruned_loss=0.09971, over 28960.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3901, pruned_loss=0.1395, over 5642872.96 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3559, pruned_loss=0.1062, over 5730475.56 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3921, pruned_loss=0.1413, over 5639276.24 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:52:27,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6886, 2.4974, 1.7857, 0.8841], device='cuda:0'), covar=tensor([0.6457, 0.3658, 0.4391, 0.6925], device='cuda:0'), in_proj_covar=tensor([0.1826, 0.1724, 0.1656, 0.1492], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 00:52:37,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5240, 1.6015, 1.3663, 1.6482], device='cuda:0'), covar=tensor([0.0572, 0.0314, 0.0275, 0.0634], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 00:52:44,068 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-14 00:52:52,375 INFO [train.py:968] (0/2) Epoch 27, batch 24100, giga_loss[loss=0.375, simple_loss=0.4179, pruned_loss=0.1661, over 28714.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3882, pruned_loss=0.1371, over 5640208.81 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3557, pruned_loss=0.106, over 5723123.58 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3912, pruned_loss=0.1398, over 5640892.75 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:53:03,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.151e+03 1.781e+03 2.232e+03 2.790e+03 8.074e+03, threshold=4.465e+03, percent-clipped=6.0
+2023-03-14 00:53:10,002 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1208971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:53:23,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3715, 1.7535, 1.5983, 1.5678], device='cuda:0'), covar=tensor([0.0829, 0.0329, 0.0314, 0.0923], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 00:53:45,324 INFO [train.py:968] (0/2) Epoch 27, batch 24150, giga_loss[loss=0.3334, simple_loss=0.3965, pruned_loss=0.1351, over 28884.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3879, pruned_loss=0.1363, over 5639400.33 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3555, pruned_loss=0.106, over 5726021.02 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3909, pruned_loss=0.1389, over 5636009.73 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:54:36,900 INFO [train.py:968] (0/2) Epoch 27, batch 24200, giga_loss[loss=0.4641, simple_loss=0.4751, pruned_loss=0.2266, over 26294.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3905, pruned_loss=0.1382, over 5628791.57 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.356, pruned_loss=0.1064, over 5728146.94 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.393, pruned_loss=0.1405, over 5622860.33 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:54:47,471 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 2.000e+03 2.623e+03 3.678e+03 8.818e+03, threshold=5.245e+03, percent-clipped=13.0
+2023-03-14 00:55:22,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5022, 3.2201, 2.8317, 2.1699], device='cuda:0'), covar=tensor([0.2807, 0.1596, 0.2061, 0.2515], device='cuda:0'), in_proj_covar=tensor([0.2066, 0.2019, 0.1936, 0.2066], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 00:55:27,524 INFO [train.py:968] (0/2) Epoch 27, batch 24250, giga_loss[loss=0.2957, simple_loss=0.3745, pruned_loss=0.1084, over 28774.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.386, pruned_loss=0.1343, over 5630230.42 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3559, pruned_loss=0.1066, over 5730435.77 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3889, pruned_loss=0.1366, over 5621203.14 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:56:15,458 INFO [train.py:968] (0/2) Epoch 27, batch 24300, libri_loss[loss=0.3034, simple_loss=0.3675, pruned_loss=0.1196, over 29546.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3822, pruned_loss=0.1297, over 5634706.12 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3558, pruned_loss=0.1067, over 5725310.41 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3853, pruned_loss=0.1321, over 5629478.33 frames. ], batch size: 89, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:56:26,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.661e+03 2.100e+03 2.698e+03 7.221e+03, threshold=4.200e+03, percent-clipped=3.0
+2023-03-14 00:56:49,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 00:56:56,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1209196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:57:07,008 INFO [train.py:968] (0/2) Epoch 27, batch 24350, giga_loss[loss=0.2801, simple_loss=0.3562, pruned_loss=0.102, over 28936.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.379, pruned_loss=0.1263, over 5641981.31 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3557, pruned_loss=0.1067, over 5717973.73 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3818, pruned_loss=0.1284, over 5643615.46 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:57:15,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3938, 1.2498, 3.5664, 3.2087], device='cuda:0'), covar=tensor([0.1535, 0.2817, 0.0477, 0.1400], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0669, 0.1000, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 00:57:39,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1033, 2.2420, 1.8135, 2.4732], device='cuda:0'), covar=tensor([0.2381, 0.2614, 0.2947, 0.2222], device='cuda:0'), in_proj_covar=tensor([0.1580, 0.1141, 0.1396, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 00:57:54,387 INFO [train.py:968] (0/2) Epoch 27, batch 24400, giga_loss[loss=0.2904, simple_loss=0.3654, pruned_loss=0.1077, over 28523.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3762, pruned_loss=0.1234, over 5657576.13 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3559, pruned_loss=0.107, over 5717157.40 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3784, pruned_loss=0.125, over 5658650.71 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 00:58:08,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.135e+03 2.023e+03 2.608e+03 3.492e+03 1.193e+04, threshold=5.216e+03, percent-clipped=10.0
+2023-03-14 00:58:44,110 INFO [train.py:968] (0/2) Epoch 27, batch 24450, giga_loss[loss=0.2748, simple_loss=0.354, pruned_loss=0.09779, over 28526.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3743, pruned_loss=0.1229, over 5652330.30 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3561, pruned_loss=0.1071, over 5719080.80 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3761, pruned_loss=0.1242, over 5650832.57 frames. ], batch size: 65, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:59:20,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1209346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:59:31,394 INFO [train.py:968] (0/2) Epoch 27, batch 24500, giga_loss[loss=0.2804, simple_loss=0.3628, pruned_loss=0.099, over 29008.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.374, pruned_loss=0.1225, over 5655697.19 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3558, pruned_loss=0.1071, over 5714066.14 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3762, pruned_loss=0.1241, over 5656855.15 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:59:42,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.223e+03 1.701e+03 2.274e+03 2.989e+03 5.824e+03, threshold=4.549e+03, percent-clipped=4.0
+2023-03-14 00:59:55,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4483, 1.7061, 1.3107, 1.5086], device='cuda:0'), covar=tensor([0.0724, 0.0323, 0.0349, 0.0870], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 01:00:20,881 INFO [train.py:968] (0/2) Epoch 27, batch 24550, giga_loss[loss=0.2594, simple_loss=0.342, pruned_loss=0.08838, over 28919.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3727, pruned_loss=0.1215, over 5662089.53 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3552, pruned_loss=0.107, over 5708227.44 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3755, pruned_loss=0.1233, over 5666531.92 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:01:06,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3418, 1.4480, 1.4855, 1.2971], device='cuda:0'), covar=tensor([0.2525, 0.2610, 0.1949, 0.2375], device='cuda:0'), in_proj_covar=tensor([0.2069, 0.2019, 0.1938, 0.2072], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 01:01:12,786 INFO [train.py:968] (0/2) Epoch 27, batch 24600, giga_loss[loss=0.2821, simple_loss=0.3622, pruned_loss=0.101, over 28907.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3706, pruned_loss=0.1185, over 5665944.11 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3552, pruned_loss=0.107, over 5712091.84 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3732, pruned_loss=0.1202, over 5664882.44 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:01:24,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.685e+03 2.131e+03 3.065e+03 6.744e+03, threshold=4.262e+03, percent-clipped=8.0
+2023-03-14 01:01:47,206 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1209489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:01:49,015 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1209492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:02:02,239 INFO [train.py:968] (0/2) Epoch 27, batch 24650, giga_loss[loss=0.3297, simple_loss=0.4129, pruned_loss=0.1233, over 28889.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3733, pruned_loss=0.1181, over 5662059.04 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3554, pruned_loss=0.1073, over 5706410.37 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3755, pruned_loss=0.1194, over 5666003.77 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:02:09,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6304, 5.4695, 5.1756, 2.8575], device='cuda:0'), covar=tensor([0.0522, 0.0714, 0.0798, 0.1611], device='cuda:0'), in_proj_covar=tensor([0.1299, 0.1198, 0.1010, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 01:02:20,051 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1209521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:02:25,763 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2128, 1.7180, 5.3410, 3.7274], device='cuda:0'), covar=tensor([0.1508, 0.2635, 0.0442, 0.0938], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0670, 0.1001, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 01:02:40,734 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9907, 1.2513, 1.1758, 0.9101], device='cuda:0'), covar=tensor([0.2129, 0.2452, 0.1546, 0.2180], device='cuda:0'), in_proj_covar=tensor([0.2072, 0.2023, 0.1943, 0.2076], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 01:02:51,039 INFO [train.py:968] (0/2) Epoch 27, batch 24700, giga_loss[loss=0.2978, simple_loss=0.3691, pruned_loss=0.1132, over 28716.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3742, pruned_loss=0.119, over 5649549.01 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3557, pruned_loss=0.1077, over 5699160.94 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3764, pruned_loss=0.12, over 5657704.64 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:03:00,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.855e+03 2.414e+03 3.259e+03 1.188e+04, threshold=4.828e+03, percent-clipped=11.0
+2023-03-14 01:03:04,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1209571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:03:09,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3750, 2.0465, 1.7234, 1.4707], device='cuda:0'), covar=tensor([0.0663, 0.0252, 0.0255, 0.0801], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 01:03:11,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5719, 1.8743, 1.3511, 1.6447], device='cuda:0'), covar=tensor([0.0721, 0.0296, 0.0343, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 01:03:35,605 INFO [train.py:968] (0/2) Epoch 27, batch 24750, giga_loss[loss=0.3557, simple_loss=0.4018, pruned_loss=0.1548, over 28278.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3739, pruned_loss=0.1195, over 5650491.58 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.355, pruned_loss=0.1073, over 5702297.45 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3769, pruned_loss=0.1211, over 5652958.72 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:04:01,829 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-14 01:04:15,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-14 01:04:23,912 INFO [train.py:968] (0/2) Epoch 27, batch 24800, giga_loss[loss=0.3352, simple_loss=0.3867, pruned_loss=0.1418, over 28067.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3735, pruned_loss=0.1202, over 5639859.88 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3551, pruned_loss=0.1076, over 5704288.75 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3762, pruned_loss=0.1216, over 5638467.52 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 01:04:26,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4061, 3.6770, 1.6703, 1.6099], device='cuda:0'), covar=tensor([0.1027, 0.0342, 0.0893, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0570, 0.0408, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 01:04:31,840 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.230e+03 1.901e+03 2.686e+03 3.552e+03 8.605e+03, threshold=5.372e+03, percent-clipped=8.0
+2023-03-14 01:05:08,076 INFO [train.py:968] (0/2) Epoch 27, batch 24850, giga_loss[loss=0.3051, simple_loss=0.3634, pruned_loss=0.1234, over 29006.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3714, pruned_loss=0.1196, over 5661564.21 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3557, pruned_loss=0.1082, over 5711030.14 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3736, pruned_loss=0.1206, over 5652724.88 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 01:05:15,411 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1209714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:05:17,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1209717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:05:44,877 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1209746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:05:53,806 INFO [train.py:968] (0/2) Epoch 27, batch 24900, giga_loss[loss=0.2995, simple_loss=0.3655, pruned_loss=0.1168, over 28875.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3704, pruned_loss=0.1197, over 5663486.00 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.356, pruned_loss=0.1085, over 5713231.90 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3721, pruned_loss=0.1203, over 5654007.51 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 01:06:03,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.671e+03 2.038e+03 2.721e+03 6.146e+03, threshold=4.076e+03, percent-clipped=1.0
+2023-03-14 01:06:29,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1209798.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 01:06:34,462 INFO [train.py:968] (0/2) Epoch 27, batch 24950, giga_loss[loss=0.2977, simple_loss=0.3752, pruned_loss=0.11, over 29013.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3701, pruned_loss=0.1179, over 5678650.54 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3562, pruned_loss=0.1085, over 5713658.05 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3715, pruned_loss=0.1186, over 5669895.38 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 01:06:47,609 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3717, 1.4474, 1.5001, 1.3542], device='cuda:0'), covar=tensor([0.2654, 0.2876, 0.2099, 0.2429], device='cuda:0'), in_proj_covar=tensor([0.2073, 0.2028, 0.1947, 0.2078], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 01:07:21,760 INFO [train.py:968] (0/2) Epoch 27, batch 25000, giga_loss[loss=0.2749, simple_loss=0.351, pruned_loss=0.09944, over 28602.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3698, pruned_loss=0.1175, over 5671516.17 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3567, pruned_loss=0.1091, over 5712370.32 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3709, pruned_loss=0.1177, over 5664704.48 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:07:34,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.159e+03 1.751e+03 2.281e+03 3.155e+03 7.765e+03, threshold=4.562e+03, percent-clipped=12.0
+2023-03-14 01:07:41,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7863, 1.1295, 2.8881, 2.7869], device='cuda:0'), covar=tensor([0.1773, 0.2558, 0.0630, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0669, 0.1002, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 01:08:09,257 INFO [train.py:968] (0/2) Epoch 27, batch 25050, giga_loss[loss=0.3143, simple_loss=0.379, pruned_loss=0.1249, over 28887.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3699, pruned_loss=0.1171, over 5674222.98 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3569, pruned_loss=0.1092, over 5715139.55 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3707, pruned_loss=0.1173, over 5665899.42 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:08:47,265 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-14 01:08:57,109 INFO [train.py:968] (0/2) Epoch 27, batch 25100, giga_loss[loss=0.3124, simple_loss=0.3745, pruned_loss=0.1251, over 27964.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3683, pruned_loss=0.1163, over 5672723.97 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3572, pruned_loss=0.1094, over 5706725.61 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3688, pruned_loss=0.1163, over 5674201.17 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:09:05,329 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1209960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:09:11,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.760e+03 2.345e+03 3.122e+03 9.415e+03, threshold=4.689e+03, percent-clipped=4.0
+2023-03-14 01:09:43,660 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1210000.pt
+2023-03-14 01:09:48,695 INFO [train.py:968] (0/2) Epoch 27, batch 25150, giga_loss[loss=0.2592, simple_loss=0.3259, pruned_loss=0.09622, over 28562.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3662, pruned_loss=0.1156, over 5677449.67 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3573, pruned_loss=0.1095, over 5710763.03 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3667, pruned_loss=0.1156, over 5674726.40 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:10:02,160 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3817, 1.5836, 1.4635, 1.3075], device='cuda:0'), covar=tensor([0.3085, 0.2813, 0.2272, 0.2811], device='cuda:0'), in_proj_covar=tensor([0.2074, 0.2027, 0.1947, 0.2079], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 01:10:07,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4466, 1.7137, 1.5776, 1.6373], device='cuda:0'), covar=tensor([0.0635, 0.0289, 0.0274, 0.0661], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0122, 0.0121, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 01:10:34,545 INFO [train.py:968] (0/2) Epoch 27, batch 25200, giga_loss[loss=0.2657, simple_loss=0.3377, pruned_loss=0.09685, over 28902.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3663, pruned_loss=0.1164, over 5687403.45 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3577, pruned_loss=0.1098, over 5713209.43 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3665, pruned_loss=0.1163, over 5682252.44 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 01:10:49,784 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.973e+03 2.923e+03 4.193e+03 1.144e+04, threshold=5.847e+03, percent-clipped=18.0
+2023-03-14 01:10:58,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5779, 1.6230, 1.7783, 1.3607], device='cuda:0'), covar=tensor([0.1679, 0.2592, 0.1414, 0.1725], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0713, 0.0973, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 01:11:21,372 INFO [train.py:968] (0/2) Epoch 27, batch 25250, giga_loss[loss=0.3726, simple_loss=0.4081, pruned_loss=0.1685, over 27984.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3648, pruned_loss=0.1159, over 5694722.62 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3574, pruned_loss=0.1098, over 5714677.19 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3653, pruned_loss=0.116, over 5688929.97 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:11:50,167 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3798, 1.6683, 1.3508, 0.9152], device='cuda:0'), covar=tensor([0.2603, 0.2648, 0.3084, 0.2522], device='cuda:0'), in_proj_covar=tensor([0.1583, 0.1143, 0.1401, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 01:12:06,289 INFO [train.py:968] (0/2) Epoch 27, batch 25300, giga_loss[loss=0.3213, simple_loss=0.3846, pruned_loss=0.129, over 29017.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3627, pruned_loss=0.1151, over 5688131.59 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3581, pruned_loss=0.1103, over 5713706.38 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3626, pruned_loss=0.1148, over 5683889.61 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:12:21,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.231e+03 1.857e+03 2.412e+03 3.102e+03 9.833e+03, threshold=4.825e+03, percent-clipped=6.0
+2023-03-14 01:12:25,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1210173.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 01:12:59,172 INFO [train.py:968] (0/2) Epoch 27, batch 25350, giga_loss[loss=0.2978, simple_loss=0.3674, pruned_loss=0.1141, over 28876.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3639, pruned_loss=0.1168, over 5687333.76 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3584, pruned_loss=0.1106, over 5716472.12 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3637, pruned_loss=0.1163, over 5680965.82 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:13:36,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-14 01:13:44,193 INFO [train.py:968] (0/2) Epoch 27, batch 25400, libri_loss[loss=0.3025, simple_loss=0.3727, pruned_loss=0.1162, over 29168.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3641, pruned_loss=0.116, over 5687862.38 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3582, pruned_loss=0.1103, over 5719469.95 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3643, pruned_loss=0.1161, over 5679130.21 frames. ], batch size: 101, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:13:56,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 2.001e+03 3.117e+03 4.237e+03 9.804e+03, threshold=6.233e+03, percent-clipped=19.0
+2023-03-14 01:14:29,016 INFO [train.py:968] (0/2) Epoch 27, batch 25450, giga_loss[loss=0.286, simple_loss=0.3561, pruned_loss=0.108, over 28822.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3646, pruned_loss=0.1157, over 5688107.14 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.358, pruned_loss=0.1103, over 5713320.80 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3651, pruned_loss=0.116, over 5685542.71 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:14:42,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1210316.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 01:14:44,444 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1210319.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 01:15:00,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1210335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:15:10,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1210348.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 01:15:18,068 INFO [train.py:968] (0/2) Epoch 27, batch 25500, giga_loss[loss=0.3049, simple_loss=0.3794, pruned_loss=0.1152, over 28831.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3646, pruned_loss=0.1151, over 5690164.15 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3579, pruned_loss=0.1103, over 5716056.74 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3652, pruned_loss=0.1153, over 5685257.38 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:15:33,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.244e+03 1.632e+03 2.101e+03 2.826e+03 6.358e+03, threshold=4.203e+03, percent-clipped=2.0
+2023-03-14 01:15:35,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2929, 1.8143, 1.4542, 1.4756], device='cuda:0'), covar=tensor([0.0799, 0.0324, 0.0346, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0122, 0.0121, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 01:16:03,455 INFO [train.py:968] (0/2) Epoch 27, batch 25550, giga_loss[loss=0.3024, simple_loss=0.3666, pruned_loss=0.1191, over 28649.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3662, pruned_loss=0.1165, over 5689082.64 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3583, pruned_loss=0.1106, over 5719593.75 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3665, pruned_loss=0.1165, over 5681346.08 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:16:54,483 INFO [train.py:968] (0/2) Epoch 27, batch 25600, giga_loss[loss=0.2924, simple_loss=0.3644, pruned_loss=0.1102, over 28868.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3687, pruned_loss=0.1187, over 5681691.79 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3581, pruned_loss=0.1106, over 5717898.75 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3692, pruned_loss=0.1189, over 5676373.01 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:17:07,166 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.876e+03 2.365e+03 3.235e+03 7.435e+03, threshold=4.730e+03, percent-clipped=10.0
+2023-03-14 01:17:15,995 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1210478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:17:18,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1210481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:17:42,969 INFO [train.py:968] (0/2) Epoch 27, batch 25650, giga_loss[loss=0.272, simple_loss=0.3333, pruned_loss=0.1053, over 28807.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3706, pruned_loss=0.1219, over 5679838.22 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.358, pruned_loss=0.1106, over 5717891.49 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3714, pruned_loss=0.1222, over 5675059.47 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:17:47,140 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1210510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:18:09,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2633, 2.3264, 1.3349, 1.3369], device='cuda:0'), covar=tensor([0.0930, 0.0477, 0.0866, 0.1304], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0571, 0.0408, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 01:18:32,140 INFO [train.py:968] (0/2) Epoch 27, batch 25700, giga_loss[loss=0.3458, simple_loss=0.3965, pruned_loss=0.1476, over 28867.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3724, pruned_loss=0.1243, over 5679650.20 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3579, pruned_loss=0.1105, over 5720619.11 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3733, pruned_loss=0.125, over 5672378.85 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:18:49,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.196e+03 1.905e+03 2.553e+03 3.547e+03 7.693e+03, threshold=5.107e+03, percent-clipped=11.0
+2023-03-14 01:19:22,731 INFO [train.py:968] (0/2) Epoch 27, batch 25750, libri_loss[loss=0.2458, simple_loss=0.3129, pruned_loss=0.08935, over 29649.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3728, pruned_loss=0.1248, over 5686503.70 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3579, pruned_loss=0.1106, over 5723564.12 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3738, pruned_loss=0.1255, over 5677590.60 frames. ], batch size: 69, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:20:01,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5312, 1.7091, 1.6573, 1.4806], device='cuda:0'), covar=tensor([0.2021, 0.2336, 0.2433, 0.2392], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0760, 0.0731, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 01:20:05,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4241, 1.5276, 1.5820, 1.4116], device='cuda:0'), covar=tensor([0.2591, 0.2440, 0.2013, 0.2408], device='cuda:0'), in_proj_covar=tensor([0.2079, 0.2030, 0.1947, 0.2082], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 01:20:05,858 INFO [train.py:968] (0/2) Epoch 27, batch 25800, giga_loss[loss=0.3223, simple_loss=0.3886, pruned_loss=0.128, over 28909.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3721, pruned_loss=0.1247, over 5679365.79 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3584, pruned_loss=0.1109, over 5728789.69 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3729, pruned_loss=0.1254, over 5665625.99 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:20:20,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.204e+03 1.748e+03 2.359e+03 3.021e+03 8.308e+03, threshold=4.718e+03, percent-clipped=5.0
+2023-03-14 01:20:30,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3702, 2.8505, 1.4216, 1.4695], device='cuda:0'), covar=tensor([0.0998, 0.0396, 0.0912, 0.1390], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0571, 0.0408, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 01:20:48,977 INFO [train.py:968] (0/2) Epoch 27, batch 25850, giga_loss[loss=0.3149, simple_loss=0.385, pruned_loss=0.1224, over 28755.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3708, pruned_loss=0.122, over 5691383.08 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3586, pruned_loss=0.111, over 5733280.05 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3717, pruned_loss=0.1228, over 5675261.15 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:21:12,995 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5401, 1.8442, 1.5375, 1.4864], device='cuda:0'), covar=tensor([0.2245, 0.2219, 0.2411, 0.2206], device='cuda:0'), in_proj_covar=tensor([0.1583, 0.1142, 0.1400, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 01:21:25,452 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4751, 1.6793, 1.7063, 1.4253], device='cuda:0'), covar=tensor([0.3835, 0.3236, 0.2451, 0.3260], device='cuda:0'), in_proj_covar=tensor([0.2077, 0.2027, 0.1944, 0.2078], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 01:21:38,076 INFO [train.py:968] (0/2) Epoch 27, batch 25900, giga_loss[loss=0.276, simple_loss=0.3467, pruned_loss=0.1026, over 28409.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.369, pruned_loss=0.1197, over 5681188.35 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3587, pruned_loss=0.1111, over 5734724.47 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3697, pruned_loss=0.1204, over 5666930.69 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:21:54,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.642e+03 2.188e+03 2.797e+03 1.140e+04, threshold=4.377e+03, percent-clipped=3.0
+2023-03-14 01:22:26,762 INFO [train.py:968] (0/2) Epoch 27, batch 25950, giga_loss[loss=0.2462, simple_loss=0.3235, pruned_loss=0.08442, over 28925.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3677, pruned_loss=0.1194, over 5676182.96 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3585, pruned_loss=0.111, over 5735688.85 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3685, pruned_loss=0.12, over 5663893.07 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:22:46,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4444, 1.5665, 1.6422, 1.3823], device='cuda:0'), covar=tensor([0.3523, 0.3210, 0.2442, 0.3143], device='cuda:0'), in_proj_covar=tensor([0.2073, 0.2025, 0.1942, 0.2076], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 01:22:55,175 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1210836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:23:11,578 INFO [train.py:968] (0/2) Epoch 27, batch 26000, libri_loss[loss=0.3342, simple_loss=0.3926, pruned_loss=0.1379, over 19050.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3668, pruned_loss=0.1196, over 5666611.73 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3588, pruned_loss=0.1114, over 5727770.46 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3672, pruned_loss=0.1199, over 5663601.54 frames. ], batch size: 187, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:23:26,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.195e+03 2.032e+03 2.502e+03 3.363e+03 7.800e+03, threshold=5.004e+03, percent-clipped=15.0
+2023-03-14 01:23:44,240 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 01:24:01,433 INFO [train.py:968] (0/2) Epoch 27, batch 26050, giga_loss[loss=0.2792, simple_loss=0.3524, pruned_loss=0.103, over 28895.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3679, pruned_loss=0.1213, over 5650771.40 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3587, pruned_loss=0.1114, over 5725062.99 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3687, pruned_loss=0.1218, over 5649050.72 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:24:18,976 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6894, 4.5292, 4.2875, 1.8647], device='cuda:0'), covar=tensor([0.0628, 0.0759, 0.0895, 0.2140], device='cuda:0'), in_proj_covar=tensor([0.1304, 0.1201, 0.1012, 0.0752], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 01:24:48,746 INFO [train.py:968] (0/2) Epoch 27, batch 26100, giga_loss[loss=0.2665, simple_loss=0.3559, pruned_loss=0.08851, over 29042.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3707, pruned_loss=0.122, over 5658150.50 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3588, pruned_loss=0.1114, over 5725820.68 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3712, pruned_loss=0.1224, over 5655896.55 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:25:05,044 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.167e+03 1.876e+03 2.443e+03 3.719e+03 1.032e+04, threshold=4.886e+03, percent-clipped=14.0
+2023-03-14 01:25:37,284 INFO [train.py:968] (0/2) Epoch 27, batch 26150, giga_loss[loss=0.2849, simple_loss=0.3615, pruned_loss=0.1042, over 28722.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3747, pruned_loss=0.1218, over 5661029.97 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3589, pruned_loss=0.1115, over 5724885.34 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3752, pruned_loss=0.1223, over 5659150.22 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:26:22,522 INFO [train.py:968] (0/2) Epoch 27, batch 26200, giga_loss[loss=0.2866, simple_loss=0.3643, pruned_loss=0.1044, over 28903.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3748, pruned_loss=0.1214, over 5650475.67 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3593, pruned_loss=0.1118, over 5714749.42 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3755, pruned_loss=0.1218, over 5655691.13 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:26:26,801 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-14 01:26:27,442 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:26:37,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.847e+03 2.273e+03 3.060e+03 6.014e+03, threshold=4.547e+03, percent-clipped=5.0
+2023-03-14 01:27:01,008 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4195, 1.8114, 1.3606, 1.3387], device='cuda:0'), covar=tensor([0.2385, 0.2375, 0.2843, 0.2166], device='cuda:0'), in_proj_covar=tensor([0.1583, 0.1141, 0.1399, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 01:27:10,361 INFO [train.py:968] (0/2) Epoch 27, batch 26250, giga_loss[loss=0.3159, simple_loss=0.3789, pruned_loss=0.1265, over 28449.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3767, pruned_loss=0.1234, over 5649112.79 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.359, pruned_loss=0.1116, over 5718267.82 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3778, pruned_loss=0.1241, over 5649073.37 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:27:16,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2186, 1.1490, 3.5784, 3.0590], device='cuda:0'), covar=tensor([0.1700, 0.2839, 0.0526, 0.1125], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0673, 0.1006, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 01:27:20,699 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:27:51,043 INFO [train.py:968] (0/2) Epoch 27, batch 26300, giga_loss[loss=0.3276, simple_loss=0.3916, pruned_loss=0.1318, over 28551.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3755, pruned_loss=0.1228, over 5664402.48 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3584, pruned_loss=0.1112, over 5723787.72 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3773, pruned_loss=0.124, over 5657970.04 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:28:08,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0231, 5.8612, 5.5460, 3.1639], device='cuda:0'), covar=tensor([0.0474, 0.0618, 0.0768, 0.1524], device='cuda:0'), in_proj_covar=tensor([0.1305, 0.1204, 0.1014, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 01:28:09,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.203e+03 1.862e+03 2.448e+03 3.791e+03 1.408e+04, threshold=4.896e+03, percent-clipped=14.0
+2023-03-14 01:28:35,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3449, 1.2969, 3.8939, 3.2719], device='cuda:0'), covar=tensor([0.1685, 0.2831, 0.0478, 0.1115], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0673, 0.1005, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 01:28:41,141 INFO [train.py:968] (0/2) Epoch 27, batch 26350, giga_loss[loss=0.2963, simple_loss=0.3643, pruned_loss=0.1141, over 29040.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3749, pruned_loss=0.1231, over 5656309.16 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3589, pruned_loss=0.1116, over 5722431.19 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3761, pruned_loss=0.1239, over 5651637.71 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:28:46,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:29:26,243 INFO [train.py:968] (0/2) Epoch 27, batch 26400, giga_loss[loss=0.2538, simple_loss=0.3257, pruned_loss=0.09096, over 29002.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3728, pruned_loss=0.1224, over 5644195.18 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3589, pruned_loss=0.1116, over 5717543.35 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3741, pruned_loss=0.1234, over 5643340.84 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:29:41,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.099e+03 1.865e+03 2.284e+03 3.034e+03 6.400e+03, threshold=4.568e+03, percent-clipped=3.0
+2023-03-14 01:30:12,322 INFO [train.py:968] (0/2) Epoch 27, batch 26450, giga_loss[loss=0.3011, simple_loss=0.3736, pruned_loss=0.1144, over 28471.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3706, pruned_loss=0.1217, over 5653230.63 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.359, pruned_loss=0.1118, over 5722264.03 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3719, pruned_loss=0.1225, over 5646879.47 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:30:22,078 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:30:41,177 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-14 01:30:59,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1211354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:30:59,931 INFO [train.py:968] (0/2) Epoch 27, batch 26500, libri_loss[loss=0.2672, simple_loss=0.3393, pruned_loss=0.09753, over 29578.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3697, pruned_loss=0.1218, over 5655293.75 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3584, pruned_loss=0.1115, over 5728308.91 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3717, pruned_loss=0.1231, over 5641944.41 frames. ], batch size: 76, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:31:01,656 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1211357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:31:17,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.665e+03 2.183e+03 3.152e+03 8.536e+03, threshold=4.367e+03, percent-clipped=8.0
+2023-03-14 01:31:30,116 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1211386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:31:31,511 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:31:44,400 INFO [train.py:968] (0/2) Epoch 27, batch 26550, giga_loss[loss=0.2716, simple_loss=0.3428, pruned_loss=0.1002, over 29010.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3705, pruned_loss=0.1226, over 5662310.82 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3586, pruned_loss=0.1117, over 5734116.73 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3724, pruned_loss=0.1239, over 5643617.38 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:32:12,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:32:30,669 INFO [train.py:968] (0/2) Epoch 27, batch 26600, giga_loss[loss=0.3016, simple_loss=0.3488, pruned_loss=0.1272, over 23643.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3677, pruned_loss=0.1205, over 5665505.25 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3583, pruned_loss=0.1114, over 5735279.23 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3697, pruned_loss=0.122, over 5647888.60 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:32:47,269 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.134e+03 2.007e+03 2.652e+03 3.548e+03 8.409e+03, threshold=5.303e+03, percent-clipped=12.0
+2023-03-14 01:33:07,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:33:20,678 INFO [train.py:968] (0/2) Epoch 27, batch 26650, giga_loss[loss=0.29, simple_loss=0.3516, pruned_loss=0.1142, over 28340.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.366, pruned_loss=0.1197, over 5676867.20 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3582, pruned_loss=0.1113, over 5736166.83 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3678, pruned_loss=0.121, over 5661965.15 frames. ], batch size: 65, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:34:06,149 INFO [train.py:968] (0/2) Epoch 27, batch 26700, giga_loss[loss=0.3676, simple_loss=0.4048, pruned_loss=0.1652, over 26517.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3659, pruned_loss=0.1191, over 5675014.41 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3582, pruned_loss=0.1113, over 5733236.69 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3675, pruned_loss=0.1204, over 5663829.58 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:34:23,035 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.830e+03 2.387e+03 3.443e+03 7.969e+03, threshold=4.773e+03, percent-clipped=5.0
+2023-03-14 01:34:29,199 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1211578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:34:31,892 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1211581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:34:49,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211600.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 01:34:53,359 INFO [train.py:968] (0/2) Epoch 27, batch 26750, giga_loss[loss=0.3277, simple_loss=0.3892, pruned_loss=0.1331, over 28954.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.368, pruned_loss=0.1197, over 5671945.77 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3581, pruned_loss=0.1113, over 5735870.86 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3695, pruned_loss=0.1208, over 5659758.37 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:34:59,700 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1211610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:35:04,406 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:35:14,529 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:35:27,294 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1211636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:35:30,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1211639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:35:31,997 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 01:35:42,871 INFO [train.py:968] (0/2) Epoch 27, batch 26800, giga_loss[loss=0.2928, simple_loss=0.3669, pruned_loss=0.1093, over 29017.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3676, pruned_loss=0.1197, over 5663746.66 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.358, pruned_loss=0.1112, over 5736562.93 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3692, pruned_loss=0.121, over 5651332.73 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:35:56,972 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1211668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:36:00,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.739e+03 2.383e+03 3.343e+03 1.892e+04, threshold=4.766e+03, percent-clipped=9.0
+2023-03-14 01:36:15,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:36:29,216 INFO [train.py:968] (0/2) Epoch 27, batch 26850, giga_loss[loss=0.3024, simple_loss=0.3724, pruned_loss=0.1162, over 28582.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3692, pruned_loss=0.1199, over 5661455.38 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3581, pruned_loss=0.1112, over 5727970.00 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3705, pruned_loss=0.121, over 5659458.77 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:36:48,842 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:36:55,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1826, 2.5886, 1.2527, 1.3316], device='cuda:0'), covar=tensor([0.1098, 0.0492, 0.1021, 0.1483], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0572, 0.0407, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 01:37:13,373 INFO [train.py:968] (0/2) Epoch 27, batch 26900, giga_loss[loss=0.296, simple_loss=0.3626, pruned_loss=0.1147, over 28031.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3708, pruned_loss=0.1185, over 5674077.91 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3579, pruned_loss=0.111, over 5731420.92 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3722, pruned_loss=0.1197, over 5668190.36 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:37:21,790 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:37:30,307 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:37:30,358 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6080, 1.3733, 3.9895, 3.5532], device='cuda:0'), covar=tensor([0.1549, 0.2862, 0.0532, 0.0950], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0673, 0.1006, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 01:37:32,818 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.035e+03 1.546e+03 1.966e+03 2.763e+03 7.532e+03, threshold=3.932e+03, percent-clipped=7.0
+2023-03-14 01:37:58,532 INFO [train.py:968] (0/2) Epoch 27, batch 26950, giga_loss[loss=0.2979, simple_loss=0.3699, pruned_loss=0.1129, over 28642.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3721, pruned_loss=0.1176, over 5670936.31 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3582, pruned_loss=0.1113, over 5726092.08 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3733, pruned_loss=0.1185, over 5669316.84 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:38:04,032 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-03-14 01:38:18,114 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-14 01:38:21,896 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1211833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:38:24,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1211836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:38:32,517 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 01:38:43,777 INFO [train.py:968] (0/2) Epoch 27, batch 27000, giga_loss[loss=0.2691, simple_loss=0.3463, pruned_loss=0.09596, over 29062.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3753, pruned_loss=0.1199, over 5675408.42 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3583, pruned_loss=0.1113, over 5727817.75 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3763, pruned_loss=0.1207, over 5672266.83 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:38:43,782 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 01:38:52,020 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2037, simple_loss=0.3111, pruned_loss=0.04816, over 944034.00 frames. 
+2023-03-14 01:38:52,020 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 01:38:59,732 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1211865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:39:09,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.099e+03 1.863e+03 2.379e+03 3.180e+03 6.137e+03, threshold=4.758e+03, percent-clipped=13.0
+2023-03-14 01:39:40,226 INFO [train.py:968] (0/2) Epoch 27, batch 27050, giga_loss[loss=0.3498, simple_loss=0.4083, pruned_loss=0.1456, over 28573.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3781, pruned_loss=0.1234, over 5677280.28 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3582, pruned_loss=0.1113, over 5731156.40 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3793, pruned_loss=0.1241, over 5670979.09 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:39:40,514 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1211905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:39:43,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1211908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:39:56,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5477, 4.4043, 4.1943, 2.0644], device='cuda:0'), covar=tensor([0.0516, 0.0614, 0.0683, 0.2112], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.1208, 0.1018, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 01:39:59,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-14 01:40:17,917 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1211937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:40:33,059 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 01:40:34,362 INFO [train.py:968] (0/2) Epoch 27, batch 27100, giga_loss[loss=0.3652, simple_loss=0.4138, pruned_loss=0.1583, over 28574.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3782, pruned_loss=0.1245, over 5679130.00 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3584, pruned_loss=0.1116, over 5731995.88 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3792, pruned_loss=0.1249, over 5672810.44 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:40:51,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.212e+03 2.010e+03 2.517e+03 3.306e+03 5.620e+03, threshold=5.034e+03, percent-clipped=5.0
+2023-03-14 01:40:52,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211975.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 01:41:05,495 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:41:10,121 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3899, 1.5254, 1.3826, 1.5067], device='cuda:0'), covar=tensor([0.0794, 0.0354, 0.0348, 0.0908], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 01:41:14,400 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1212000.pt
+2023-03-14 01:41:17,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1212002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:41:19,897 INFO [train.py:968] (0/2) Epoch 27, batch 27150, giga_loss[loss=0.2708, simple_loss=0.3484, pruned_loss=0.09658, over 28510.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.378, pruned_loss=0.1247, over 5665165.45 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.359, pruned_loss=0.1121, over 5722603.21 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3789, pruned_loss=0.125, over 5666665.68 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:41:37,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6841, 1.6630, 1.8993, 1.4533], device='cuda:0'), covar=tensor([0.1818, 0.2632, 0.1538, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0715, 0.0975, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 01:42:05,285 INFO [train.py:968] (0/2) Epoch 27, batch 27200, giga_loss[loss=0.2813, simple_loss=0.3693, pruned_loss=0.09662, over 29132.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3757, pruned_loss=0.1216, over 5674306.57 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3594, pruned_loss=0.1123, over 5724791.52 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3763, pruned_loss=0.1218, over 5672436.24 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:42:25,158 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.691e+03 2.261e+03 2.910e+03 6.631e+03, threshold=4.521e+03, percent-clipped=6.0
+2023-03-14 01:42:51,964 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1212103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:42:55,375 INFO [train.py:968] (0/2) Epoch 27, batch 27250, giga_loss[loss=0.3236, simple_loss=0.3648, pruned_loss=0.1412, over 23445.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3762, pruned_loss=0.1211, over 5658320.79 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3593, pruned_loss=0.1124, over 5726700.60 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3769, pruned_loss=0.1214, over 5654578.78 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:43:07,120 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212118.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 01:43:08,903 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212121.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 01:43:14,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3828, 1.4882, 3.7241, 3.2353], device='cuda:0'), covar=tensor([0.1611, 0.2576, 0.0568, 0.0910], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0673, 0.1005, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 01:43:17,759 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:43:20,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:43:30,923 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-14 01:43:32,198 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1212145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:43:32,227 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:43:36,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:43:37,448 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212150.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 01:43:42,075 INFO [train.py:968] (0/2) Epoch 27, batch 27300, giga_loss[loss=0.3267, simple_loss=0.3921, pruned_loss=0.1306, over 28966.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3766, pruned_loss=0.1211, over 5669374.89 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5730496.04 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3779, pruned_loss=0.1216, over 5661976.91 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:43:50,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:44:00,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.206e+03 1.700e+03 2.315e+03 2.999e+03 7.112e+03, threshold=4.629e+03, percent-clipped=7.0
+2023-03-14 01:44:02,993 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:44:18,770 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1212193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:44:32,051 INFO [train.py:968] (0/2) Epoch 27, batch 27350, giga_loss[loss=0.3243, simple_loss=0.385, pruned_loss=0.1318, over 28772.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3775, pruned_loss=0.1226, over 5652672.64 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3587, pruned_loss=0.1121, over 5724734.16 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3789, pruned_loss=0.1232, over 5650480.21 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:44:53,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4606, 2.0430, 1.4383, 0.8843], device='cuda:0'), covar=tensor([0.5417, 0.2570, 0.3632, 0.5064], device='cuda:0'), in_proj_covar=tensor([0.1834, 0.1729, 0.1654, 0.1495], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 01:44:56,061 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-14 01:45:07,975 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:45:11,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:45:17,903 INFO [train.py:968] (0/2) Epoch 27, batch 27400, giga_loss[loss=0.3014, simple_loss=0.3505, pruned_loss=0.1261, over 23309.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3756, pruned_loss=0.1218, over 5654599.13 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3587, pruned_loss=0.112, over 5727725.31 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3771, pruned_loss=0.1225, over 5649461.42 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:45:21,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4091, 2.7261, 2.5444, 1.9977], device='cuda:0'), covar=tensor([0.2885, 0.2364, 0.2484, 0.3117], device='cuda:0'), in_proj_covar=tensor([0.2057, 0.2015, 0.1938, 0.2068], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 01:45:37,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.823e+03 2.359e+03 2.896e+03 1.010e+04, threshold=4.719e+03, percent-clipped=6.0
+2023-03-14 01:45:41,131 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:45:50,221 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:45:54,892 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:46:05,827 INFO [train.py:968] (0/2) Epoch 27, batch 27450, giga_loss[loss=0.2798, simple_loss=0.3567, pruned_loss=0.1015, over 28978.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3736, pruned_loss=0.121, over 5676224.13 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5731899.46 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.375, pruned_loss=0.1217, over 5667054.63 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:46:11,356 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 01:46:20,945 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1212320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:46:20,960 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:46:55,980 INFO [train.py:968] (0/2) Epoch 27, batch 27500, giga_loss[loss=0.2864, simple_loss=0.355, pruned_loss=0.1089, over 28618.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.372, pruned_loss=0.1205, over 5665901.03 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3596, pruned_loss=0.1125, over 5726579.06 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3728, pruned_loss=0.1209, over 5662110.42 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:47:14,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.737e+03 2.285e+03 3.053e+03 7.544e+03, threshold=4.570e+03, percent-clipped=4.0
+2023-03-14 01:47:43,284 INFO [train.py:968] (0/2) Epoch 27, batch 27550, giga_loss[loss=0.2785, simple_loss=0.3464, pruned_loss=0.1053, over 28940.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3698, pruned_loss=0.1196, over 5668760.48 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3594, pruned_loss=0.1124, over 5729370.02 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3709, pruned_loss=0.1203, over 5661641.03 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:48:13,549 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-14 01:48:25,326 INFO [train.py:968] (0/2) Epoch 27, batch 27600, giga_loss[loss=0.3185, simple_loss=0.3767, pruned_loss=0.1302, over 28835.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3702, pruned_loss=0.1206, over 5658267.59 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3595, pruned_loss=0.1124, over 5716910.18 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3713, pruned_loss=0.1214, over 5662056.84 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:48:45,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.173e+03 1.634e+03 2.175e+03 3.034e+03 6.285e+03, threshold=4.349e+03, percent-clipped=6.0
+2023-03-14 01:49:11,270 INFO [train.py:968] (0/2) Epoch 27, batch 27650, giga_loss[loss=0.3445, simple_loss=0.3822, pruned_loss=0.1534, over 23638.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3681, pruned_loss=0.119, over 5655397.72 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3593, pruned_loss=0.1123, over 5717989.63 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3691, pruned_loss=0.1197, over 5657244.52 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:49:55,583 INFO [train.py:968] (0/2) Epoch 27, batch 27700, giga_loss[loss=0.306, simple_loss=0.3764, pruned_loss=0.1178, over 28834.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3654, pruned_loss=0.1159, over 5655362.25 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.359, pruned_loss=0.1121, over 5710708.54 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3667, pruned_loss=0.1167, over 5662153.01 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:50:07,587 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1212568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:50:14,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.535e+03 1.962e+03 2.757e+03 6.111e+03, threshold=3.924e+03, percent-clipped=5.0
+2023-03-14 01:50:45,181 INFO [train.py:968] (0/2) Epoch 27, batch 27750, giga_loss[loss=0.286, simple_loss=0.3594, pruned_loss=0.1063, over 28563.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3636, pruned_loss=0.1144, over 5641494.04 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3592, pruned_loss=0.1122, over 5704704.41 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3645, pruned_loss=0.1151, over 5652120.15 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:51:32,731 INFO [train.py:968] (0/2) Epoch 27, batch 27800, giga_loss[loss=0.3156, simple_loss=0.3734, pruned_loss=0.1289, over 28928.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3627, pruned_loss=0.1143, over 5647476.07 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3589, pruned_loss=0.112, over 5707288.03 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3637, pruned_loss=0.1151, over 5652167.01 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:51:54,609 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.763e+03 2.326e+03 3.210e+03 1.005e+04, threshold=4.653e+03, percent-clipped=14.0
+2023-03-14 01:52:13,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1212695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:52:17,573 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.35 vs. limit=5.0
+2023-03-14 01:52:23,177 INFO [train.py:968] (0/2) Epoch 27, batch 27850, giga_loss[loss=0.2579, simple_loss=0.3299, pruned_loss=0.093, over 28557.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3593, pruned_loss=0.1131, over 5651679.82 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3591, pruned_loss=0.112, over 5713067.97 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.36, pruned_loss=0.1137, over 5648405.93 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:52:27,772 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:52:32,830 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:52:43,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3755, 1.6880, 1.5527, 1.4934], device='cuda:0'), covar=tensor([0.1789, 0.1573, 0.2225, 0.1816], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0757, 0.0727, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 01:52:54,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3145, 3.7699, 1.5021, 1.5059], device='cuda:0'), covar=tensor([0.1022, 0.0357, 0.0895, 0.1326], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0569, 0.0406, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 01:52:58,755 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:53:07,612 INFO [train.py:968] (0/2) Epoch 27, batch 27900, libri_loss[loss=0.2893, simple_loss=0.3624, pruned_loss=0.1081, over 29609.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3599, pruned_loss=0.1134, over 5658821.84 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3588, pruned_loss=0.1118, over 5720337.19 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1142, over 5646882.51 frames. ], batch size: 91, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:53:24,693 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.195e+03 2.039e+03 2.835e+03 3.804e+03 8.996e+03, threshold=5.669e+03, percent-clipped=14.0
+2023-03-14 01:53:52,911 INFO [train.py:968] (0/2) Epoch 27, batch 27950, giga_loss[loss=0.366, simple_loss=0.4042, pruned_loss=0.1639, over 26566.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3627, pruned_loss=0.1144, over 5667405.64 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3588, pruned_loss=0.1116, over 5725863.82 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3636, pruned_loss=0.1152, over 5650771.47 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:53:55,094 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4834, 4.2524, 1.6160, 1.7157], device='cuda:0'), covar=tensor([0.1017, 0.0239, 0.0930, 0.1333], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0570, 0.0407, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 01:54:23,725 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:54:25,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:54:40,125 INFO [train.py:968] (0/2) Epoch 27, batch 28000, giga_loss[loss=0.3075, simple_loss=0.3768, pruned_loss=0.1191, over 28905.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3641, pruned_loss=0.1148, over 5653138.27 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3593, pruned_loss=0.112, over 5715554.25 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3644, pruned_loss=0.1152, over 5647753.53 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:54:55,333 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:55:02,334 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.616e+03 1.978e+03 2.703e+03 8.501e+03, threshold=3.955e+03, percent-clipped=3.0
+2023-03-14 01:55:25,915 INFO [train.py:968] (0/2) Epoch 27, batch 28050, giga_loss[loss=0.3129, simple_loss=0.373, pruned_loss=0.1264, over 28698.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3655, pruned_loss=0.1165, over 5646962.74 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3596, pruned_loss=0.1122, over 5712105.26 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3656, pruned_loss=0.1167, over 5643354.46 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:55:32,711 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3345, 1.5393, 3.1394, 2.9382], device='cuda:0'), covar=tensor([0.1289, 0.2319, 0.0450, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0673, 0.1005, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 01:56:11,127 INFO [train.py:968] (0/2) Epoch 27, batch 28100, giga_loss[loss=0.2619, simple_loss=0.3385, pruned_loss=0.09268, over 28439.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3651, pruned_loss=0.1165, over 5648347.19 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3596, pruned_loss=0.1121, over 5709537.73 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3652, pruned_loss=0.1168, over 5646708.84 frames. ], batch size: 65, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:56:29,936 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.129e+03 1.631e+03 1.976e+03 2.466e+03 8.288e+03, threshold=3.952e+03, percent-clipped=4.0
+2023-03-14 01:56:45,796 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1212992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:56:57,494 INFO [train.py:968] (0/2) Epoch 27, batch 28150, libri_loss[loss=0.2902, simple_loss=0.3656, pruned_loss=0.1074, over 29528.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3667, pruned_loss=0.1176, over 5646252.64 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3598, pruned_loss=0.1121, over 5709646.96 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3668, pruned_loss=0.1179, over 5643354.24 frames. ], batch size: 82, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:57:43,228 INFO [train.py:968] (0/2) Epoch 27, batch 28200, giga_loss[loss=0.2887, simple_loss=0.3575, pruned_loss=0.11, over 28450.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3692, pruned_loss=0.1189, over 5655641.30 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3602, pruned_loss=0.1124, over 5710810.74 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.369, pruned_loss=0.119, over 5651440.44 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:58:00,468 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1213072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:58:05,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.224e+03 1.929e+03 2.385e+03 3.470e+03 6.519e+03, threshold=4.771e+03, percent-clipped=13.0
+2023-03-14 01:58:27,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6477, 2.6147, 2.4593, 2.3046], device='cuda:0'), covar=tensor([0.1742, 0.2029, 0.1921, 0.1939], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0756, 0.0728, 0.0697], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 01:58:33,165 INFO [train.py:968] (0/2) Epoch 27, batch 28250, giga_loss[loss=0.3407, simple_loss=0.4031, pruned_loss=0.1391, over 29011.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3708, pruned_loss=0.1209, over 5649188.83 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3598, pruned_loss=0.1122, over 5712309.07 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3714, pruned_loss=0.1214, over 5642263.25 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:59:21,296 INFO [train.py:968] (0/2) Epoch 27, batch 28300, giga_loss[loss=0.3103, simple_loss=0.3751, pruned_loss=0.1227, over 28697.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.373, pruned_loss=0.1233, over 5641443.23 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3603, pruned_loss=0.1126, over 5703210.34 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3732, pruned_loss=0.1236, over 5642275.51 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:59:34,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8211, 1.2951, 1.2558, 1.0265], device='cuda:0'), covar=tensor([0.2221, 0.1310, 0.2466, 0.1865], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0756, 0.0726, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 01:59:41,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.133e+03 1.952e+03 2.602e+03 3.778e+03 1.110e+04, threshold=5.204e+03, percent-clipped=13.0
+2023-03-14 01:59:50,673 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 02:00:03,959 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1213199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:00:07,515 INFO [train.py:968] (0/2) Epoch 27, batch 28350, giga_loss[loss=0.2672, simple_loss=0.3565, pruned_loss=0.08892, over 28941.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3721, pruned_loss=0.1214, over 5639290.05 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3595, pruned_loss=0.1122, over 5698852.05 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3734, pruned_loss=0.1223, over 5642555.31 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:00:43,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9555, 3.7862, 3.6217, 1.7725], device='cuda:0'), covar=tensor([0.0790, 0.0917, 0.0922, 0.2151], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.1212, 0.1022, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 02:00:58,481 INFO [train.py:968] (0/2) Epoch 27, batch 28400, giga_loss[loss=0.3746, simple_loss=0.4095, pruned_loss=0.1699, over 26630.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.371, pruned_loss=0.1201, over 5645483.67 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3595, pruned_loss=0.1121, over 5702904.40 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3722, pruned_loss=0.121, over 5643291.79 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:01:17,796 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.179e+03 1.768e+03 2.354e+03 3.191e+03 1.132e+04, threshold=4.707e+03, percent-clipped=7.0
+2023-03-14 02:01:34,104 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-14 02:01:38,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-14 02:01:39,402 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5680, 1.8234, 1.4799, 1.5497], device='cuda:0'), covar=tensor([0.2664, 0.2830, 0.3224, 0.2401], device='cuda:0'), in_proj_covar=tensor([0.1585, 0.1143, 0.1403, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 02:01:45,489 INFO [train.py:968] (0/2) Epoch 27, batch 28450, giga_loss[loss=0.3069, simple_loss=0.3724, pruned_loss=0.1207, over 29066.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3715, pruned_loss=0.1215, over 5639211.34 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3593, pruned_loss=0.112, over 5707911.34 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.373, pruned_loss=0.1226, over 5631091.77 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:02:39,915 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1213353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:02:41,448 INFO [train.py:968] (0/2) Epoch 27, batch 28500, giga_loss[loss=0.2867, simple_loss=0.3546, pruned_loss=0.1094, over 28932.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3704, pruned_loss=0.1214, over 5638956.83 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3592, pruned_loss=0.1121, over 5708521.90 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3721, pruned_loss=0.1225, over 5630075.22 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:02:56,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1213367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:03:07,129 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.182e+03 1.845e+03 2.351e+03 3.174e+03 7.248e+03, threshold=4.703e+03, percent-clipped=9.0
+2023-03-14 02:03:36,078 INFO [train.py:968] (0/2) Epoch 27, batch 28550, giga_loss[loss=0.3097, simple_loss=0.3716, pruned_loss=0.1239, over 27501.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3691, pruned_loss=0.1214, over 5629637.41 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3591, pruned_loss=0.1119, over 5711118.28 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3707, pruned_loss=0.1225, over 5619112.77 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:04:04,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4883, 2.2478, 1.6797, 0.7092], device='cuda:0'), covar=tensor([0.7371, 0.3680, 0.4876, 0.8047], device='cuda:0'), in_proj_covar=tensor([0.1831, 0.1732, 0.1655, 0.1495], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 02:04:16,004 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1213447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:04:25,915 INFO [train.py:968] (0/2) Epoch 27, batch 28600, giga_loss[loss=0.2584, simple_loss=0.3325, pruned_loss=0.09215, over 28582.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3685, pruned_loss=0.121, over 5648871.04 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3592, pruned_loss=0.1119, over 5714098.57 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3698, pruned_loss=0.122, over 5636972.51 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:04:26,930 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3374, 1.4658, 1.3551, 1.4302], device='cuda:0'), covar=tensor([0.0743, 0.0370, 0.0339, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 02:04:45,174 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.740e+03 2.504e+03 3.372e+03 1.111e+04, threshold=5.008e+03, percent-clipped=8.0
+2023-03-14 02:05:12,489 INFO [train.py:968] (0/2) Epoch 27, batch 28650, giga_loss[loss=0.2823, simple_loss=0.3533, pruned_loss=0.1056, over 28943.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3686, pruned_loss=0.1215, over 5659576.55 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3594, pruned_loss=0.1121, over 5718313.23 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3697, pruned_loss=0.1224, over 5644916.28 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:05:18,456 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1213510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:05:21,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1213513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:05:47,345 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1213542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:05:59,710 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1213553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:06:00,940 INFO [train.py:968] (0/2) Epoch 27, batch 28700, giga_loss[loss=0.2822, simple_loss=0.3641, pruned_loss=0.1001, over 28842.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3695, pruned_loss=0.1224, over 5658276.70 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3593, pruned_loss=0.1121, over 5720163.23 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3705, pruned_loss=0.1232, over 5644544.93 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:06:19,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1213574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:06:22,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.691e+03 2.214e+03 3.458e+03 7.463e+03, threshold=4.429e+03, percent-clipped=4.0
+2023-03-14 02:06:34,175 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5108, 1.8533, 1.4108, 1.7846], device='cuda:0'), covar=tensor([0.2559, 0.2605, 0.2899, 0.2516], device='cuda:0'), in_proj_covar=tensor([0.1584, 0.1142, 0.1402, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 02:06:35,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1213590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:06:37,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1213593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:06:50,694 INFO [train.py:968] (0/2) Epoch 27, batch 28750, giga_loss[loss=0.3378, simple_loss=0.3919, pruned_loss=0.1418, over 28856.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3686, pruned_loss=0.1216, over 5669711.14 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3588, pruned_loss=0.1118, over 5724371.82 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3701, pruned_loss=0.1228, over 5653596.42 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:07:06,646 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1213622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:07:32,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8904, 2.0260, 1.5262, 1.6707], device='cuda:0'), covar=tensor([0.1088, 0.0789, 0.1096, 0.1242], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0453, 0.0524, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 02:07:38,854 INFO [train.py:968] (0/2) Epoch 27, batch 28800, giga_loss[loss=0.2778, simple_loss=0.3467, pruned_loss=0.1045, over 28488.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.371, pruned_loss=0.1232, over 5669413.34 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3593, pruned_loss=0.1121, over 5726748.43 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3719, pruned_loss=0.124, over 5653816.03 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:08:02,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.335e+03 1.756e+03 2.214e+03 2.878e+03 6.082e+03, threshold=4.428e+03, percent-clipped=10.0
+2023-03-14 02:08:24,768 INFO [train.py:968] (0/2) Epoch 27, batch 28850, giga_loss[loss=0.2768, simple_loss=0.3396, pruned_loss=0.107, over 28621.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3721, pruned_loss=0.1245, over 5672041.20 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3595, pruned_loss=0.1124, over 5725519.38 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3731, pruned_loss=0.1254, over 5657910.33 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:08:35,297 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1213717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:08:37,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1213720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:08:44,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1213728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:08:57,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5661, 5.4178, 5.1629, 2.9223], device='cuda:0'), covar=tensor([0.0469, 0.0598, 0.0645, 0.1451], device='cuda:0'), in_proj_covar=tensor([0.1314, 0.1212, 0.1022, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 02:09:02,541 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-14 02:09:03,141 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1213749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:09:03,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3320, 1.6478, 1.3734, 1.4923], device='cuda:0'), covar=tensor([0.0725, 0.0392, 0.0350, 0.0798], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0122, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 02:09:04,455 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5378, 2.1512, 1.9242, 1.7147], device='cuda:0'), covar=tensor([0.0757, 0.0255, 0.0281, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0122, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 02:09:08,846 INFO [train.py:968] (0/2) Epoch 27, batch 28900, giga_loss[loss=0.2957, simple_loss=0.3657, pruned_loss=0.1128, over 28980.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3714, pruned_loss=0.1243, over 5679717.09 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.359, pruned_loss=0.112, over 5730771.16 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3731, pruned_loss=0.1256, over 5662563.76 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:09:28,867 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.200e+03 1.776e+03 2.346e+03 2.854e+03 8.695e+03, threshold=4.692e+03, percent-clipped=8.0
+2023-03-14 02:09:52,728 INFO [train.py:968] (0/2) Epoch 27, batch 28950, giga_loss[loss=0.2915, simple_loss=0.3641, pruned_loss=0.1095, over 28342.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3698, pruned_loss=0.122, over 5688723.16 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3589, pruned_loss=0.1119, over 5732793.38 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3714, pruned_loss=0.1235, over 5671949.52 frames. ], batch size: 65, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:09:59,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2899, 1.9218, 1.3241, 0.5356], device='cuda:0'), covar=tensor([0.5378, 0.2924, 0.4186, 0.6884], device='cuda:0'), in_proj_covar=tensor([0.1843, 0.1741, 0.1666, 0.1503], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 02:10:38,659 INFO [train.py:968] (0/2) Epoch 27, batch 29000, giga_loss[loss=0.3789, simple_loss=0.4114, pruned_loss=0.1732, over 26507.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3725, pruned_loss=0.1238, over 5671834.01 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3597, pruned_loss=0.1126, over 5726207.54 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3735, pruned_loss=0.1247, over 5662922.63 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:10:51,292 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1213871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:10:54,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1213874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:10:56,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.318e+03 1.777e+03 2.231e+03 3.273e+03 6.173e+03, threshold=4.463e+03, percent-clipped=4.0
+2023-03-14 02:11:20,358 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1213903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:11:21,505 INFO [train.py:968] (0/2) Epoch 27, batch 29050, giga_loss[loss=0.3431, simple_loss=0.3937, pruned_loss=0.1462, over 28679.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3733, pruned_loss=0.1245, over 5671863.23 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3594, pruned_loss=0.1124, over 5719955.14 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3749, pruned_loss=0.1258, over 5668874.39 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:11:31,739 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1213916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:11:42,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1213928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:11:45,610 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1213930.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 02:12:04,809 INFO [train.py:968] (0/2) Epoch 27, batch 29100, giga_loss[loss=0.3419, simple_loss=0.4042, pruned_loss=0.1398, over 28540.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3738, pruned_loss=0.1251, over 5671276.22 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3593, pruned_loss=0.1123, over 5723076.13 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3755, pruned_loss=0.1265, over 5665141.22 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:12:25,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.325e+03 1.896e+03 2.346e+03 3.002e+03 1.158e+04, threshold=4.692e+03, percent-clipped=10.0
+2023-03-14 02:12:45,462 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1214000.pt
+2023-03-14 02:12:49,944 INFO [train.py:968] (0/2) Epoch 27, batch 29150, giga_loss[loss=0.3309, simple_loss=0.3956, pruned_loss=0.1331, over 29117.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3757, pruned_loss=0.1268, over 5662685.46 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3592, pruned_loss=0.1123, over 5716551.51 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3774, pruned_loss=0.1282, over 5662771.90 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:13:33,169 INFO [train.py:968] (0/2) Epoch 27, batch 29200, libri_loss[loss=0.3166, simple_loss=0.3837, pruned_loss=0.1247, over 29673.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3756, pruned_loss=0.126, over 5664800.98 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1124, over 5720463.19 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3776, pruned_loss=0.1276, over 5658527.08 frames. ], batch size: 88, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:13:48,825 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:13:50,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1214074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:13:53,004 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.182e+03 1.807e+03 2.390e+03 3.250e+03 7.020e+03, threshold=4.780e+03, percent-clipped=3.0
+2023-03-14 02:14:18,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1214103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:14:20,719 INFO [train.py:968] (0/2) Epoch 27, batch 29250, giga_loss[loss=0.3303, simple_loss=0.3866, pruned_loss=0.137, over 27937.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3749, pruned_loss=0.125, over 5652112.35 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.359, pruned_loss=0.1123, over 5722148.42 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3772, pruned_loss=0.1268, over 5643504.78 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:15:05,127 INFO [train.py:968] (0/2) Epoch 27, batch 29300, giga_loss[loss=0.2804, simple_loss=0.3558, pruned_loss=0.1025, over 29027.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3739, pruned_loss=0.1239, over 5649622.89 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3588, pruned_loss=0.1122, over 5718289.12 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3764, pruned_loss=0.1258, over 5644890.22 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:15:05,524 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1214155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:15:23,023 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.665e+03 2.387e+03 3.616e+03 7.428e+03, threshold=4.775e+03, percent-clipped=5.0
+2023-03-14 02:15:49,680 INFO [train.py:968] (0/2) Epoch 27, batch 29350, giga_loss[loss=0.3197, simple_loss=0.3834, pruned_loss=0.128, over 27845.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3723, pruned_loss=0.1225, over 5645145.68 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1125, over 5704293.97 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3745, pruned_loss=0.124, over 5650942.31 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:16:32,474 INFO [train.py:968] (0/2) Epoch 27, batch 29400, giga_loss[loss=0.2987, simple_loss=0.3664, pruned_loss=0.1154, over 28711.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3708, pruned_loss=0.1212, over 5644623.23 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3593, pruned_loss=0.1126, over 5698107.63 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3726, pruned_loss=0.1226, over 5653720.28 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:16:55,999 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.179e+03 1.746e+03 2.129e+03 2.901e+03 8.746e+03, threshold=4.258e+03, percent-clipped=4.0
+2023-03-14 02:17:07,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1214291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:17:20,329 INFO [train.py:968] (0/2) Epoch 27, batch 29450, giga_loss[loss=0.2742, simple_loss=0.359, pruned_loss=0.09467, over 28873.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3709, pruned_loss=0.1213, over 5645197.19 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3588, pruned_loss=0.1123, over 5699272.29 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3731, pruned_loss=0.1229, over 5649514.36 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:17:20,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1214305.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 02:17:25,332 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-14 02:18:07,753 INFO [train.py:968] (0/2) Epoch 27, batch 29500, giga_loss[loss=0.2583, simple_loss=0.3284, pruned_loss=0.09408, over 28384.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3726, pruned_loss=0.123, over 5633529.90 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5682166.78 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3741, pruned_loss=0.124, over 5651483.35 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:18:29,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.177e+03 1.778e+03 2.494e+03 3.742e+03 7.938e+03, threshold=4.988e+03, percent-clipped=16.0
+2023-03-14 02:18:49,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5352, 1.8037, 1.4719, 1.6197], device='cuda:0'), covar=tensor([0.2354, 0.2376, 0.2535, 0.2195], device='cuda:0'), in_proj_covar=tensor([0.1587, 0.1146, 0.1405, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 02:18:53,526 INFO [train.py:968] (0/2) Epoch 27, batch 29550, giga_loss[loss=0.2987, simple_loss=0.3543, pruned_loss=0.1215, over 29023.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3725, pruned_loss=0.1239, over 5643440.19 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5685574.16 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3737, pruned_loss=0.1249, over 5653886.39 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:19:21,443 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:19:24,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1214437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:19:30,671 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-14 02:19:34,823 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214448.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 02:19:37,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1214451.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 02:19:40,914 INFO [train.py:968] (0/2) Epoch 27, batch 29600, giga_loss[loss=0.3087, simple_loss=0.3788, pruned_loss=0.1193, over 28567.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3732, pruned_loss=0.1242, over 5648960.24 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1127, over 5679883.44 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3744, pruned_loss=0.1252, over 5661960.30 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:19:52,743 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1214466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:19:55,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1545, 1.2974, 1.1158, 0.8902], device='cuda:0'), covar=tensor([0.0976, 0.0469, 0.0998, 0.1111], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0452, 0.0524, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 02:20:04,864 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.716e+03 2.490e+03 3.505e+03 9.929e+03, threshold=4.981e+03, percent-clipped=10.0
+2023-03-14 02:20:05,064 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1214480.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 02:20:24,996 INFO [train.py:968] (0/2) Epoch 27, batch 29650, libri_loss[loss=0.2824, simple_loss=0.3567, pruned_loss=0.104, over 29284.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3742, pruned_loss=0.1251, over 5653329.15 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3595, pruned_loss=0.1127, over 5685670.65 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3757, pruned_loss=0.1264, over 5657167.63 frames. ], batch size: 94, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:20:27,449 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1214508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:20:49,091 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1214530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:21:12,821 INFO [train.py:968] (0/2) Epoch 27, batch 29700, libri_loss[loss=0.2909, simple_loss=0.3688, pruned_loss=0.1065, over 29663.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3748, pruned_loss=0.1254, over 5647214.49 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5688813.74 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3759, pruned_loss=0.1266, over 5646766.95 frames. ], batch size: 88, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:21:33,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.162e+03 1.748e+03 2.176e+03 2.931e+03 8.984e+03, threshold=4.351e+03, percent-clipped=2.0
+2023-03-14 02:21:36,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1544, 1.3593, 5.4456, 3.7411], device='cuda:0'), covar=tensor([0.1460, 0.2866, 0.0452, 0.1030], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0671, 0.1004, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 02:21:40,646 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1214588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:21:56,770 INFO [train.py:968] (0/2) Epoch 27, batch 29750, giga_loss[loss=0.3058, simple_loss=0.3739, pruned_loss=0.1188, over 28937.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3746, pruned_loss=0.1255, over 5636436.94 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3594, pruned_loss=0.1126, over 5681754.34 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3763, pruned_loss=0.1269, over 5640704.95 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:22:46,466 INFO [train.py:968] (0/2) Epoch 27, batch 29800, giga_loss[loss=0.3385, simple_loss=0.3955, pruned_loss=0.1407, over 27859.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3757, pruned_loss=0.1259, over 5646306.45 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1125, over 5682727.48 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3775, pruned_loss=0.1272, over 5648131.78 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:23:04,741 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:23:08,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1214676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:23:10,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.322e+03 2.008e+03 2.463e+03 3.488e+03 1.086e+04, threshold=4.926e+03, percent-clipped=15.0
+2023-03-14 02:23:37,066 INFO [train.py:968] (0/2) Epoch 27, batch 29850, giga_loss[loss=0.2992, simple_loss=0.3688, pruned_loss=0.1148, over 28958.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3747, pruned_loss=0.1248, over 5643183.12 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1125, over 5682727.48 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3761, pruned_loss=0.1259, over 5644603.79 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:23:37,294 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1214705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:23:57,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3175, 1.9132, 1.4064, 0.6213], device='cuda:0'), covar=tensor([0.6562, 0.3233, 0.4063, 0.7197], device='cuda:0'), in_proj_covar=tensor([0.1844, 0.1746, 0.1668, 0.1506], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 02:24:11,172 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-14 02:24:21,829 INFO [train.py:968] (0/2) Epoch 27, batch 29900, libri_loss[loss=0.3322, simple_loss=0.3901, pruned_loss=0.1371, over 29528.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3732, pruned_loss=0.1235, over 5662194.80 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3593, pruned_loss=0.1126, over 5684860.66 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3745, pruned_loss=0.1246, over 5660462.13 frames. ], batch size: 89, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:24:45,895 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.145e+03 1.904e+03 2.408e+03 3.323e+03 8.663e+03, threshold=4.816e+03, percent-clipped=9.0
+2023-03-14 02:24:54,068 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1214790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:25:06,862 INFO [train.py:968] (0/2) Epoch 27, batch 29950, giga_loss[loss=0.2991, simple_loss=0.3616, pruned_loss=0.1183, over 28799.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3705, pruned_loss=0.1217, over 5670035.11 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3598, pruned_loss=0.1127, over 5688045.52 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3714, pruned_loss=0.1225, over 5665299.74 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:25:54,473 INFO [train.py:968] (0/2) Epoch 27, batch 30000, giga_loss[loss=0.2661, simple_loss=0.3322, pruned_loss=0.09998, over 29002.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3658, pruned_loss=0.1192, over 5662878.94 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3594, pruned_loss=0.1126, over 5693100.73 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.367, pruned_loss=0.1203, over 5654341.57 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:25:54,477 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 02:26:02,679 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2043, simple_loss=0.313, pruned_loss=0.04778, over 944034.00 frames. 
+2023-03-14 02:26:02,680 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 02:26:23,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.839e+03 2.543e+03 3.350e+03 9.840e+03, threshold=5.087e+03, percent-clipped=11.0
+2023-03-14 02:26:26,119 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1214883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:26:33,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2543, 1.2279, 3.3761, 3.1510], device='cuda:0'), covar=tensor([0.1598, 0.2730, 0.0623, 0.1084], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0675, 0.1009, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 02:26:44,447 INFO [train.py:968] (0/2) Epoch 27, batch 30050, libri_loss[loss=0.2629, simple_loss=0.3303, pruned_loss=0.09771, over 29590.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3624, pruned_loss=0.118, over 5649009.42 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3599, pruned_loss=0.113, over 5678720.21 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3631, pruned_loss=0.1186, over 5653181.15 frames. ], batch size: 74, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:26:49,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9266, 2.1626, 2.1494, 1.6940], device='cuda:0'), covar=tensor([0.1854, 0.2782, 0.1577, 0.1993], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0719, 0.0977, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 02:27:30,954 INFO [train.py:968] (0/2) Epoch 27, batch 30100, giga_loss[loss=0.2881, simple_loss=0.3479, pruned_loss=0.1142, over 28756.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3624, pruned_loss=0.1188, over 5645406.47 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3599, pruned_loss=0.1131, over 5681575.98 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3629, pruned_loss=0.1194, over 5645269.65 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:27:37,591 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1214963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:27:55,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.758e+03 2.320e+03 3.844e+03 8.561e+03, threshold=4.640e+03, percent-clipped=10.0
+2023-03-14 02:28:17,303 INFO [train.py:968] (0/2) Epoch 27, batch 30150, giga_loss[loss=0.2796, simple_loss=0.3601, pruned_loss=0.09956, over 28927.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3619, pruned_loss=0.1177, over 5643411.87 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3604, pruned_loss=0.1135, over 5684924.83 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.362, pruned_loss=0.1178, over 5639673.95 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:28:32,490 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:28:34,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:28:37,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-14 02:28:58,278 INFO [train.py:968] (0/2) Epoch 27, batch 30200, giga_loss[loss=0.2475, simple_loss=0.3375, pruned_loss=0.07881, over 28622.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.36, pruned_loss=0.1144, over 5649833.56 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3597, pruned_loss=0.1132, over 5689966.16 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3607, pruned_loss=0.1149, over 5640410.63 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:29:02,644 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:29:26,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.612e+03 2.242e+03 3.079e+03 6.833e+03, threshold=4.483e+03, percent-clipped=3.0
+2023-03-14 02:29:50,311 INFO [train.py:968] (0/2) Epoch 27, batch 30250, libri_loss[loss=0.3071, simple_loss=0.3739, pruned_loss=0.1201, over 29644.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3578, pruned_loss=0.1113, over 5633808.25 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3598, pruned_loss=0.1135, over 5677029.91 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3583, pruned_loss=0.1114, over 5637367.61 frames. ], batch size: 88, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:29:51,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:29:54,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:30:18,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4033, 1.5875, 1.5698, 1.4291], device='cuda:0'), covar=tensor([0.2796, 0.2113, 0.1952, 0.2264], device='cuda:0'), in_proj_covar=tensor([0.2047, 0.2003, 0.1926, 0.2057], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 02:30:21,508 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:30:22,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3839, 1.5573, 1.2204, 1.1055], device='cuda:0'), covar=tensor([0.1022, 0.0536, 0.1024, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0451, 0.0524, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 02:30:23,692 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-14 02:30:31,295 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 02:30:38,800 INFO [train.py:968] (0/2) Epoch 27, batch 30300, giga_loss[loss=0.2749, simple_loss=0.3547, pruned_loss=0.09753, over 28905.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3568, pruned_loss=0.1086, over 5648329.66 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3598, pruned_loss=0.1135, over 5678869.98 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3571, pruned_loss=0.1086, over 5649113.36 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:30:51,177 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1215165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:31:04,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.353e+02 1.451e+03 1.833e+03 2.480e+03 5.919e+03, threshold=3.665e+03, percent-clipped=2.0
+2023-03-14 02:31:06,698 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:31:23,648 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:31:28,555 INFO [train.py:968] (0/2) Epoch 27, batch 30350, giga_loss[loss=0.2457, simple_loss=0.3272, pruned_loss=0.0821, over 27943.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3542, pruned_loss=0.1059, over 5650030.51 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3596, pruned_loss=0.1134, over 5680260.06 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3547, pruned_loss=0.1059, over 5649168.83 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:32:16,302 INFO [train.py:968] (0/2) Epoch 27, batch 30400, giga_loss[loss=0.2566, simple_loss=0.3468, pruned_loss=0.08316, over 28971.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3521, pruned_loss=0.1027, over 5658763.82 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3591, pruned_loss=0.1132, over 5682798.46 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3528, pruned_loss=0.1028, over 5655261.41 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:32:43,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.592e+03 1.892e+03 2.769e+03 7.335e+03, threshold=3.785e+03, percent-clipped=13.0
+2023-03-14 02:33:07,752 INFO [train.py:968] (0/2) Epoch 27, batch 30450, giga_loss[loss=0.3049, simple_loss=0.3736, pruned_loss=0.1181, over 28836.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3523, pruned_loss=0.1013, over 5667587.57 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3594, pruned_loss=0.1136, over 5684227.21 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3523, pruned_loss=0.1008, over 5663163.78 frames. ], batch size: 285, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:33:11,233 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:33:15,313 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:33:41,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:33:53,009 INFO [train.py:968] (0/2) Epoch 27, batch 30500, libri_loss[loss=0.2967, simple_loss=0.36, pruned_loss=0.1167, over 28616.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3523, pruned_loss=0.1021, over 5645211.45 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3587, pruned_loss=0.1135, over 5665420.78 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3526, pruned_loss=0.1011, over 5658167.27 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:34:17,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.544e+03 1.809e+03 2.350e+03 5.251e+03, threshold=3.619e+03, percent-clipped=6.0
+2023-03-14 02:34:40,552 INFO [train.py:968] (0/2) Epoch 27, batch 30550, giga_loss[loss=0.2383, simple_loss=0.324, pruned_loss=0.07631, over 28834.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3501, pruned_loss=0.1006, over 5658993.97 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3583, pruned_loss=0.1134, over 5671844.16 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3505, pruned_loss=0.09957, over 5663087.22 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:35:09,718 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:35:31,602 INFO [train.py:968] (0/2) Epoch 27, batch 30600, giga_loss[loss=0.2489, simple_loss=0.3057, pruned_loss=0.09608, over 24437.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3463, pruned_loss=0.09786, over 5653820.61 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3583, pruned_loss=0.1133, over 5671847.79 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3466, pruned_loss=0.09694, over 5656943.24 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:35:59,012 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3121, 1.4228, 3.0797, 2.9985], device='cuda:0'), covar=tensor([0.1349, 0.2573, 0.0543, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0673, 0.1007, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 02:35:59,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.580e+02 1.431e+03 1.723e+03 2.621e+03 4.823e+03, threshold=3.445e+03, percent-clipped=6.0
+2023-03-14 02:36:21,173 INFO [train.py:968] (0/2) Epoch 27, batch 30650, libri_loss[loss=0.3126, simple_loss=0.3523, pruned_loss=0.1364, over 29562.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3455, pruned_loss=0.0976, over 5659789.58 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3577, pruned_loss=0.1133, over 5679104.85 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3459, pruned_loss=0.09644, over 5654793.04 frames. ], batch size: 77, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:36:51,260 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6251, 3.4791, 3.2918, 1.7931], device='cuda:0'), covar=tensor([0.0849, 0.0953, 0.0935, 0.2362], device='cuda:0'), in_proj_covar=tensor([0.1300, 0.1201, 0.1011, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 02:37:09,018 INFO [train.py:968] (0/2) Epoch 27, batch 30700, giga_loss[loss=0.2745, simple_loss=0.3533, pruned_loss=0.09786, over 28382.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3453, pruned_loss=0.09721, over 5663113.08 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3577, pruned_loss=0.1133, over 5680891.52 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3456, pruned_loss=0.09611, over 5657255.76 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:37:11,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1215558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:37:20,520 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:37:28,148 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1215575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:37:34,623 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.780e+02 1.625e+03 2.261e+03 2.975e+03 8.923e+03, threshold=4.522e+03, percent-clipped=19.0
+2023-03-14 02:37:44,543 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6617, 1.9297, 1.2956, 1.5935], device='cuda:0'), covar=tensor([0.1016, 0.0605, 0.0976, 0.1183], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0449, 0.0522, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 02:37:58,381 INFO [train.py:968] (0/2) Epoch 27, batch 30750, giga_loss[loss=0.2542, simple_loss=0.3404, pruned_loss=0.08399, over 28479.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3426, pruned_loss=0.09514, over 5655504.67 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3572, pruned_loss=0.1132, over 5677032.17 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3429, pruned_loss=0.09402, over 5653642.76 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:38:47,220 INFO [train.py:968] (0/2) Epoch 27, batch 30800, giga_loss[loss=0.2363, simple_loss=0.3165, pruned_loss=0.078, over 28590.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3396, pruned_loss=0.09299, over 5667142.01 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.357, pruned_loss=0.1132, over 5680361.77 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3399, pruned_loss=0.09186, over 5662340.55 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:39:16,104 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.132e+02 1.616e+03 2.019e+03 3.239e+03 1.232e+04, threshold=4.039e+03, percent-clipped=11.0
+2023-03-14 02:39:32,012 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:39:35,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:39:36,117 INFO [train.py:968] (0/2) Epoch 27, batch 30850, giga_loss[loss=0.2775, simple_loss=0.348, pruned_loss=0.1035, over 28719.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3359, pruned_loss=0.09111, over 5669051.27 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3572, pruned_loss=0.1134, over 5683711.99 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3356, pruned_loss=0.0897, over 5662116.57 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:39:49,210 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:39:51,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:40:02,478 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:40:17,968 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:40:22,729 INFO [train.py:968] (0/2) Epoch 27, batch 30900, giga_loss[loss=0.252, simple_loss=0.3323, pruned_loss=0.08579, over 28622.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.335, pruned_loss=0.09106, over 5659276.23 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3573, pruned_loss=0.1135, over 5674917.50 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3344, pruned_loss=0.08955, over 5661227.59 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:40:23,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:40:52,756 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.551e+02 1.548e+03 2.137e+03 2.880e+03 6.695e+03, threshold=4.275e+03, percent-clipped=11.0
+2023-03-14 02:40:56,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3866, 2.6550, 1.5581, 1.5986], device='cuda:0'), covar=tensor([0.0890, 0.0430, 0.0848, 0.1228], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0567, 0.0407, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 02:41:12,597 INFO [train.py:968] (0/2) Epoch 27, batch 30950, giga_loss[loss=0.2187, simple_loss=0.3013, pruned_loss=0.06804, over 28585.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3351, pruned_loss=0.09181, over 5632388.72 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3561, pruned_loss=0.113, over 5661774.12 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.335, pruned_loss=0.09034, over 5645396.85 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:41:18,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1215809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:42:10,840 INFO [train.py:968] (0/2) Epoch 27, batch 31000, giga_loss[loss=0.2298, simple_loss=0.3013, pruned_loss=0.07912, over 24197.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3375, pruned_loss=0.09201, over 5631910.85 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3558, pruned_loss=0.1128, over 5664239.30 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3375, pruned_loss=0.09081, over 5639471.28 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:42:44,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.561e+03 1.893e+03 2.810e+03 8.763e+03, threshold=3.787e+03, percent-clipped=8.0
+2023-03-14 02:42:52,991 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:43:06,425 INFO [train.py:968] (0/2) Epoch 27, batch 31050, giga_loss[loss=0.2848, simple_loss=0.3611, pruned_loss=0.1043, over 28535.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3385, pruned_loss=0.09188, over 5635723.87 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3554, pruned_loss=0.1126, over 5670242.56 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3385, pruned_loss=0.09061, over 5635239.02 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:43:52,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1215943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:44:03,125 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:44:07,382 INFO [train.py:968] (0/2) Epoch 27, batch 31100, giga_loss[loss=0.2589, simple_loss=0.3298, pruned_loss=0.09405, over 26979.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3397, pruned_loss=0.09328, over 5636478.70 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3558, pruned_loss=0.1132, over 5676171.28 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3389, pruned_loss=0.09133, over 5630238.63 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:44:08,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:44:11,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6033, 1.7895, 1.4390, 1.6897], device='cuda:0'), covar=tensor([0.2841, 0.2905, 0.3450, 0.2514], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1141, 0.1408, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 02:44:43,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.604e+02 1.761e+03 2.131e+03 2.973e+03 8.765e+03, threshold=4.262e+03, percent-clipped=10.0
+2023-03-14 02:44:44,827 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:45:02,729 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1216000.pt
+2023-03-14 02:45:07,302 INFO [train.py:968] (0/2) Epoch 27, batch 31150, giga_loss[loss=0.2469, simple_loss=0.3318, pruned_loss=0.08098, over 28930.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3365, pruned_loss=0.0909, over 5628494.38 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3557, pruned_loss=0.1132, over 5660231.24 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3358, pruned_loss=0.08903, over 5636637.06 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:46:07,547 INFO [train.py:968] (0/2) Epoch 27, batch 31200, giga_loss[loss=0.2498, simple_loss=0.3342, pruned_loss=0.08271, over 28678.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.336, pruned_loss=0.08953, over 5634003.29 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3553, pruned_loss=0.113, over 5669021.03 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.335, pruned_loss=0.08734, over 5631110.32 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:46:41,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.980e+02 1.325e+03 1.714e+03 2.570e+03 5.995e+03, threshold=3.428e+03, percent-clipped=4.0
+2023-03-14 02:46:46,333 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1216086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:46:49,913 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1216089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:46:51,910 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-14 02:47:07,294 INFO [train.py:968] (0/2) Epoch 27, batch 31250, giga_loss[loss=0.24, simple_loss=0.3194, pruned_loss=0.08035, over 28579.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3341, pruned_loss=0.08807, over 5641853.58 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3551, pruned_loss=0.1128, over 5672382.06 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3332, pruned_loss=0.08616, over 5635915.45 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:47:24,918 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1216118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:47:26,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1250, 1.3660, 1.1400, 0.9272], device='cuda:0'), covar=tensor([0.1110, 0.0439, 0.1105, 0.1082], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0448, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 02:47:36,046 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1216130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:48:07,728 INFO [train.py:968] (0/2) Epoch 27, batch 31300, giga_loss[loss=0.2262, simple_loss=0.3135, pruned_loss=0.06943, over 28898.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3327, pruned_loss=0.0881, over 5658471.28 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3551, pruned_loss=0.1128, over 5675016.46 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3317, pruned_loss=0.08623, over 5651034.74 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:48:39,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.238e+02 1.459e+03 1.990e+03 2.649e+03 6.250e+03, threshold=3.981e+03, percent-clipped=13.0
+2023-03-14 02:49:00,777 INFO [train.py:968] (0/2) Epoch 27, batch 31350, giga_loss[loss=0.2388, simple_loss=0.3144, pruned_loss=0.08164, over 27718.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3332, pruned_loss=0.08937, over 5675871.13 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3547, pruned_loss=0.1128, over 5685817.56 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3318, pruned_loss=0.08677, over 5659398.16 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:49:58,400 INFO [train.py:968] (0/2) Epoch 27, batch 31400, giga_loss[loss=0.2195, simple_loss=0.2916, pruned_loss=0.07367, over 24568.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.333, pruned_loss=0.08835, over 5669998.09 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3544, pruned_loss=0.1125, over 5688025.85 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3319, pruned_loss=0.0863, over 5655034.74 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:50:10,550 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1216266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:50:18,125 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1216273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:50:21,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1216276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:50:31,062 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.505e+02 1.526e+03 2.127e+03 3.172e+03 9.391e+03, threshold=4.254e+03, percent-clipped=12.0
+2023-03-14 02:50:42,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5257, 1.6422, 1.7483, 1.3256], device='cuda:0'), covar=tensor([0.1928, 0.2842, 0.1589, 0.1962], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0713, 0.0976, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 02:50:45,008 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1216295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:50:55,871 INFO [train.py:968] (0/2) Epoch 27, batch 31450, giga_loss[loss=0.2635, simple_loss=0.3404, pruned_loss=0.09332, over 28913.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3343, pruned_loss=0.08838, over 5664382.33 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3539, pruned_loss=0.1123, over 5690004.67 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3335, pruned_loss=0.08643, over 5650338.46 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:50:56,730 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1216305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:51:38,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5538, 1.6638, 1.7616, 1.3639], device='cuda:0'), covar=tensor([0.2033, 0.3013, 0.1649, 0.1965], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0713, 0.0976, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 02:51:54,456 INFO [train.py:968] (0/2) Epoch 27, batch 31500, giga_loss[loss=0.2727, simple_loss=0.3475, pruned_loss=0.09893, over 28654.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3326, pruned_loss=0.08734, over 5669560.57 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3533, pruned_loss=0.112, over 5686244.92 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3318, pruned_loss=0.08528, over 5660358.16 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:52:31,054 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.454e+02 1.559e+03 1.978e+03 2.618e+03 5.135e+03, threshold=3.957e+03, percent-clipped=3.0
+2023-03-14 02:52:59,706 INFO [train.py:968] (0/2) Epoch 27, batch 31550, giga_loss[loss=0.2579, simple_loss=0.3305, pruned_loss=0.09267, over 27777.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3332, pruned_loss=0.08818, over 5674248.44 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.353, pruned_loss=0.1118, over 5687581.08 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3323, pruned_loss=0.08605, over 5665289.84 frames. ], batch size: 474, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:53:04,043 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1216409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:53:10,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1216412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:53:46,026 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1216441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:54:00,136 INFO [train.py:968] (0/2) Epoch 27, batch 31600, giga_loss[loss=0.2647, simple_loss=0.3557, pruned_loss=0.08678, over 28623.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3345, pruned_loss=0.08763, over 5674137.86 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3528, pruned_loss=0.1117, over 5690532.80 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3337, pruned_loss=0.0857, over 5664138.06 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:54:40,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.373e+02 1.458e+03 1.950e+03 2.760e+03 6.404e+03, threshold=3.901e+03, percent-clipped=8.0
+2023-03-14 02:55:07,014 INFO [train.py:968] (0/2) Epoch 27, batch 31650, giga_loss[loss=0.266, simple_loss=0.3508, pruned_loss=0.09057, over 27781.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3374, pruned_loss=0.08666, over 5667588.22 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3525, pruned_loss=0.1115, over 5693524.52 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3368, pruned_loss=0.08498, over 5656779.16 frames. ], batch size: 474, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:56:06,650 INFO [train.py:968] (0/2) Epoch 27, batch 31700, giga_loss[loss=0.2473, simple_loss=0.3404, pruned_loss=0.07711, over 29010.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3388, pruned_loss=0.08633, over 5665452.57 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3523, pruned_loss=0.1116, over 5695854.99 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3382, pruned_loss=0.08446, over 5654280.77 frames. ], batch size: 285, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:56:10,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5327, 1.9422, 1.2471, 1.4900], device='cuda:0'), covar=tensor([0.1118, 0.0591, 0.1131, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0448, 0.0522, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 02:56:20,058 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1216566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:56:41,576 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.605e+03 2.056e+03 2.663e+03 7.753e+03, threshold=4.111e+03, percent-clipped=10.0
+2023-03-14 02:57:06,699 INFO [train.py:968] (0/2) Epoch 27, batch 31750, giga_loss[loss=0.2604, simple_loss=0.3517, pruned_loss=0.08455, over 28962.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3385, pruned_loss=0.08523, over 5664909.79 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3524, pruned_loss=0.1118, over 5688981.52 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3376, pruned_loss=0.08297, over 5662470.21 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:57:27,849 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1216622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:58:07,837 INFO [train.py:968] (0/2) Epoch 27, batch 31800, giga_loss[loss=0.2779, simple_loss=0.3608, pruned_loss=0.09754, over 28041.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3386, pruned_loss=0.08626, over 5677388.22 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3516, pruned_loss=0.1115, over 5693235.45 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3384, pruned_loss=0.08425, over 5671207.39 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:58:11,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 02:58:29,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1216670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:58:36,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8193, 2.4029, 1.8348, 1.0893], device='cuda:0'), covar=tensor([0.5770, 0.3087, 0.4179, 0.6210], device='cuda:0'), in_proj_covar=tensor([0.1831, 0.1723, 0.1653, 0.1499], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 02:58:40,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1828, 1.4187, 1.1762, 1.0139], device='cuda:0'), covar=tensor([0.0917, 0.0392, 0.0965, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0406, 0.0446, 0.0520, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 02:58:46,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.454e+02 1.362e+03 1.778e+03 2.675e+03 5.055e+03, threshold=3.556e+03, percent-clipped=6.0
+2023-03-14 02:58:46,433 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1216685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:58:53,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5717, 1.7271, 1.5094, 1.5760], device='cuda:0'), covar=tensor([0.0728, 0.0302, 0.0326, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 02:59:15,634 INFO [train.py:968] (0/2) Epoch 27, batch 31850, giga_loss[loss=0.2381, simple_loss=0.3211, pruned_loss=0.07752, over 28471.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3365, pruned_loss=0.08632, over 5675921.22 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3518, pruned_loss=0.1116, over 5693896.71 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3359, pruned_loss=0.08412, over 5670149.01 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:00:05,262 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-14 03:00:30,556 INFO [train.py:968] (0/2) Epoch 27, batch 31900, giga_loss[loss=0.2261, simple_loss=0.317, pruned_loss=0.06758, over 28906.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3377, pruned_loss=0.08815, over 5673512.82 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3511, pruned_loss=0.1111, over 5694288.38 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3376, pruned_loss=0.08632, over 5668402.50 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:01:11,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.352e+02 1.380e+03 1.717e+03 2.387e+03 7.204e+03, threshold=3.434e+03, percent-clipped=12.0
+2023-03-14 03:01:37,658 INFO [train.py:968] (0/2) Epoch 27, batch 31950, giga_loss[loss=0.2447, simple_loss=0.3249, pruned_loss=0.08226, over 29027.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3334, pruned_loss=0.08592, over 5673427.23 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3509, pruned_loss=0.1111, over 5688670.59 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.333, pruned_loss=0.08368, over 5672950.16 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:01:47,210 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1216813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:01:51,081 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1216816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:02:04,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3217, 1.3951, 1.3136, 1.3004], device='cuda:0'), covar=tensor([0.2153, 0.2017, 0.1806, 0.1838], device='cuda:0'), in_proj_covar=tensor([0.2011, 0.1969, 0.1882, 0.2020], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 03:02:26,976 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 03:02:29,473 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1216845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:02:40,051 INFO [train.py:968] (0/2) Epoch 27, batch 32000, giga_loss[loss=0.1955, simple_loss=0.2903, pruned_loss=0.05038, over 28628.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3311, pruned_loss=0.08466, over 5656048.19 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3508, pruned_loss=0.1112, over 5670412.76 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3304, pruned_loss=0.08217, over 5670885.18 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:03:19,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.406e+02 1.418e+03 1.819e+03 2.650e+03 1.079e+04, threshold=3.638e+03, percent-clipped=12.0
+2023-03-14 03:03:45,099 INFO [train.py:968] (0/2) Epoch 27, batch 32050, giga_loss[loss=0.2279, simple_loss=0.3111, pruned_loss=0.0724, over 28471.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3288, pruned_loss=0.08417, over 5665988.59 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3499, pruned_loss=0.1107, over 5674963.11 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3286, pruned_loss=0.08217, over 5673860.78 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:04:28,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1216941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:04:44,293 INFO [train.py:968] (0/2) Epoch 27, batch 32100, giga_loss[loss=0.2546, simple_loss=0.3406, pruned_loss=0.08433, over 28927.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3322, pruned_loss=0.08586, over 5663628.38 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3492, pruned_loss=0.1104, over 5670062.39 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3322, pruned_loss=0.0839, over 5674842.33 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:05:21,467 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.874e+02 1.681e+03 1.937e+03 2.848e+03 6.644e+03, threshold=3.874e+03, percent-clipped=11.0
+2023-03-14 03:05:33,161 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1216997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:05:40,264 INFO [train.py:968] (0/2) Epoch 27, batch 32150, libri_loss[loss=0.2838, simple_loss=0.3492, pruned_loss=0.1092, over 28818.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.333, pruned_loss=0.08677, over 5679187.32 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3489, pruned_loss=0.1102, over 5674045.20 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.333, pruned_loss=0.08487, over 5684499.58 frames. ], batch size: 107, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:06:44,262 INFO [train.py:968] (0/2) Epoch 27, batch 32200, giga_loss[loss=0.3001, simple_loss=0.3618, pruned_loss=0.1192, over 28044.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3318, pruned_loss=0.08728, over 5683645.88 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3486, pruned_loss=0.11, over 5677263.36 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3317, pruned_loss=0.08527, over 5684837.11 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:06:49,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1217060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:07:18,829 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1217084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:07:20,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.052e+02 1.646e+03 2.084e+03 2.954e+03 7.275e+03, threshold=4.169e+03, percent-clipped=11.0
+2023-03-14 03:07:22,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1217087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:07:42,613 INFO [train.py:968] (0/2) Epoch 27, batch 32250, giga_loss[loss=0.2335, simple_loss=0.3158, pruned_loss=0.07557, over 28625.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3326, pruned_loss=0.08847, over 5670853.83 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3488, pruned_loss=0.1104, over 5668617.88 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.332, pruned_loss=0.08621, over 5679403.81 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:07:56,375 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1217116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:08:09,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3946, 3.1794, 1.4987, 1.5508], device='cuda:0'), covar=tensor([0.0962, 0.0328, 0.0919, 0.1303], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0563, 0.0406, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 03:08:23,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5244, 1.5961, 1.7687, 1.4044], device='cuda:0'), covar=tensor([0.1715, 0.2366, 0.1423, 0.1876], device='cuda:0'), in_proj_covar=tensor([0.0923, 0.0709, 0.0970, 0.0872], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0015], device='cuda:0')
+2023-03-14 03:08:26,712 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1217140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:08:32,685 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1217143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:08:48,683 INFO [train.py:968] (0/2) Epoch 27, batch 32300, giga_loss[loss=0.2226, simple_loss=0.2941, pruned_loss=0.07552, over 24646.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3331, pruned_loss=0.08825, over 5666268.30 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3486, pruned_loss=0.1103, over 5669533.14 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3325, pruned_loss=0.08608, over 5672332.75 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:09:11,324 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1217172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:09:26,038 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1217183.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:09:28,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.055e+02 1.548e+03 2.181e+03 2.732e+03 7.800e+03, threshold=4.363e+03, percent-clipped=9.0
+2023-03-14 03:09:29,841 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2310, 0.9476, 1.1011, 1.4302], device='cuda:0'), covar=tensor([0.0764, 0.0448, 0.0362, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 03:09:34,124 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6190, 1.9056, 1.5950, 1.5630], device='cuda:0'), covar=tensor([0.2607, 0.2439, 0.2684, 0.2409], device='cuda:0'), in_proj_covar=tensor([0.1590, 0.1141, 0.1407, 0.1010], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 03:09:56,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1217203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:09:57,864 INFO [train.py:968] (0/2) Epoch 27, batch 32350, giga_loss[loss=0.2357, simple_loss=0.3215, pruned_loss=0.07494, over 28613.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3354, pruned_loss=0.08874, over 5658595.41 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3485, pruned_loss=0.1104, over 5664098.47 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3347, pruned_loss=0.08651, over 5667851.88 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:10:00,625 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1217206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:10:43,172 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1217235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:11:06,456 INFO [train.py:968] (0/2) Epoch 27, batch 32400, giga_loss[loss=0.2413, simple_loss=0.3192, pruned_loss=0.08168, over 27673.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3348, pruned_loss=0.08873, over 5660867.43 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3484, pruned_loss=0.1105, over 5671222.04 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3339, pruned_loss=0.08614, over 5661412.50 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:11:48,265 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.794e+02 1.444e+03 1.884e+03 2.339e+03 4.615e+03, threshold=3.767e+03, percent-clipped=2.0
+2023-03-14 03:12:12,991 INFO [train.py:968] (0/2) Epoch 27, batch 32450, libri_loss[loss=0.3038, simple_loss=0.3725, pruned_loss=0.1176, over 29267.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3309, pruned_loss=0.08738, over 5673890.47 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3487, pruned_loss=0.1107, over 5672941.97 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3296, pruned_loss=0.08474, over 5672476.00 frames. ], batch size: 94, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:13:08,950 INFO [train.py:968] (0/2) Epoch 27, batch 32500, giga_loss[loss=0.2521, simple_loss=0.3277, pruned_loss=0.0882, over 28061.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3251, pruned_loss=0.08466, over 5674093.19 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3484, pruned_loss=0.1105, over 5673784.81 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3236, pruned_loss=0.08175, over 5672349.77 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:13:17,833 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7224, 4.7112, 1.8428, 1.9056], device='cuda:0'), covar=tensor([0.0943, 0.0297, 0.0887, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0565, 0.0406, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 03:13:51,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.309e+02 1.517e+03 1.868e+03 2.617e+03 5.798e+03, threshold=3.736e+03, percent-clipped=10.0
+2023-03-14 03:14:14,385 INFO [train.py:968] (0/2) Epoch 27, batch 32550, giga_loss[loss=0.2498, simple_loss=0.3136, pruned_loss=0.09301, over 24039.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3258, pruned_loss=0.08513, over 5666324.47 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3483, pruned_loss=0.1104, over 5676127.33 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3244, pruned_loss=0.08269, over 5662915.89 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:15:12,704 INFO [train.py:968] (0/2) Epoch 27, batch 32600, giga_loss[loss=0.2509, simple_loss=0.3353, pruned_loss=0.08325, over 28627.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3281, pruned_loss=0.08664, over 5663049.34 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3483, pruned_loss=0.1106, over 5665700.69 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3268, pruned_loss=0.0843, over 5670100.98 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:15:50,825 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.920e+02 1.598e+03 2.028e+03 3.011e+03 9.225e+03, threshold=4.057e+03, percent-clipped=12.0
+2023-03-14 03:16:01,482 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2274, 1.3740, 1.3338, 1.2094], device='cuda:0'), covar=tensor([0.2313, 0.2296, 0.1564, 0.2126], device='cuda:0'), in_proj_covar=tensor([0.2009, 0.1959, 0.1872, 0.2013], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 03:16:08,522 INFO [train.py:968] (0/2) Epoch 27, batch 32650, libri_loss[loss=0.3002, simple_loss=0.3576, pruned_loss=0.1214, over 29518.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3265, pruned_loss=0.08553, over 5668990.22 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3482, pruned_loss=0.1106, over 5674090.62 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3246, pruned_loss=0.08265, over 5666832.54 frames. ], batch size: 82, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:16:20,235 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1217513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:17:09,652 INFO [train.py:968] (0/2) Epoch 27, batch 32700, giga_loss[loss=0.2189, simple_loss=0.3093, pruned_loss=0.06424, over 28845.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3251, pruned_loss=0.08396, over 5668332.15 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3481, pruned_loss=0.1105, over 5677607.51 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3235, pruned_loss=0.08137, over 5663451.42 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:17:16,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1217558.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:17:50,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.243e+02 1.516e+03 1.919e+03 2.748e+03 6.161e+03, threshold=3.838e+03, percent-clipped=8.0
+2023-03-14 03:18:11,136 INFO [train.py:968] (0/2) Epoch 27, batch 32750, giga_loss[loss=0.2815, simple_loss=0.3503, pruned_loss=0.1063, over 28931.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3241, pruned_loss=0.08438, over 5670277.24 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3473, pruned_loss=0.1102, over 5684875.90 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3227, pruned_loss=0.08167, over 5659299.67 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:19:18,459 INFO [train.py:968] (0/2) Epoch 27, batch 32800, libri_loss[loss=0.3261, simple_loss=0.3814, pruned_loss=0.1354, over 26126.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3235, pruned_loss=0.08292, over 5681565.58 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3468, pruned_loss=0.1099, over 5686929.54 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3223, pruned_loss=0.08038, over 5671045.06 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:19:44,221 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-14 03:19:58,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5903, 1.9838, 1.5444, 1.7392], device='cuda:0'), covar=tensor([0.0774, 0.0281, 0.0339, 0.0898], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 03:19:58,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.741e+02 1.474e+03 1.854e+03 2.747e+03 7.082e+03, threshold=3.707e+03, percent-clipped=8.0
+2023-03-14 03:20:16,669 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1217701.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:20:22,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1217704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:20:22,323 INFO [train.py:968] (0/2) Epoch 27, batch 32850, giga_loss[loss=0.2647, simple_loss=0.3377, pruned_loss=0.09587, over 28962.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3246, pruned_loss=0.08375, over 5677629.88 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3468, pruned_loss=0.1101, over 5681300.89 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3232, pruned_loss=0.08106, over 5673785.15 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:20:56,530 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1217733.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:21:08,118 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3000, 1.5525, 1.5217, 1.1730], device='cuda:0'), covar=tensor([0.1541, 0.2417, 0.1351, 0.1698], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0708, 0.0972, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 03:21:12,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1217748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:21:21,060 INFO [train.py:968] (0/2) Epoch 27, batch 32900, giga_loss[loss=0.2309, simple_loss=0.3113, pruned_loss=0.07528, over 28817.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3254, pruned_loss=0.08487, over 5682352.80 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3461, pruned_loss=0.1097, over 5684970.17 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3245, pruned_loss=0.08264, over 5676015.26 frames. ], batch size: 263, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:21:30,683 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4477, 1.3242, 1.2373, 1.6103], device='cuda:0'), covar=tensor([0.0802, 0.0356, 0.0358, 0.0889], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 03:22:01,271 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1010, 1.4237, 1.3253, 1.0054], device='cuda:0'), covar=tensor([0.1651, 0.2536, 0.1387, 0.1753], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0707, 0.0972, 0.0873], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 03:22:01,836 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1217785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:22:05,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.843e+02 1.411e+03 1.860e+03 2.734e+03 9.355e+03, threshold=3.721e+03, percent-clipped=12.0
+2023-03-14 03:22:18,415 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1217798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:22:24,431 INFO [train.py:968] (0/2) Epoch 27, batch 32950, giga_loss[loss=0.2227, simple_loss=0.3126, pruned_loss=0.06635, over 28667.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3248, pruned_loss=0.08369, over 5670613.66 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3461, pruned_loss=0.1097, over 5678070.89 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3237, pruned_loss=0.08159, over 5670905.52 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:23:16,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3215, 1.0462, 4.5265, 3.4237], device='cuda:0'), covar=tensor([0.1890, 0.3225, 0.0449, 0.1158], device='cuda:0'), in_proj_covar=tensor([0.0794, 0.0667, 0.0991, 0.0957], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 03:23:23,104 INFO [train.py:968] (0/2) Epoch 27, batch 33000, libri_loss[loss=0.3089, simple_loss=0.366, pruned_loss=0.1259, over 25760.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3259, pruned_loss=0.08322, over 5662600.18 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3457, pruned_loss=0.1095, over 5678722.51 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3252, pruned_loss=0.08122, over 5662012.11 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:23:23,108 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 03:23:31,647 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.1936, simple_loss=0.295, pruned_loss=0.04608, over 944034.00 frames. 
+2023-03-14 03:23:31,648 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 03:24:09,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.565e+03 2.149e+03 3.460e+03 1.006e+04, threshold=4.297e+03, percent-clipped=20.0
+2023-03-14 03:24:09,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1217888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:24:27,103 INFO [train.py:968] (0/2) Epoch 27, batch 33050, giga_loss[loss=0.2689, simple_loss=0.3506, pruned_loss=0.09359, over 28935.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3288, pruned_loss=0.08435, over 5657268.13 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3454, pruned_loss=0.1093, over 5675036.09 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3278, pruned_loss=0.08205, over 5659737.68 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:25:13,610 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1217943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:25:29,690 INFO [train.py:968] (0/2) Epoch 27, batch 33100, giga_loss[loss=0.2787, simple_loss=0.3481, pruned_loss=0.1046, over 26979.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3304, pruned_loss=0.08484, over 5647759.29 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3456, pruned_loss=0.1095, over 5659635.91 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3291, pruned_loss=0.08232, over 5663785.33 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:26:08,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.488e+03 1.892e+03 2.298e+03 7.438e+03, threshold=3.784e+03, percent-clipped=4.0
+2023-03-14 03:26:20,926 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1218000.pt
+2023-03-14 03:26:27,319 INFO [train.py:968] (0/2) Epoch 27, batch 33150, giga_loss[loss=0.2537, simple_loss=0.3376, pruned_loss=0.08491, over 28482.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3308, pruned_loss=0.08578, over 5656989.78 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.345, pruned_loss=0.1093, over 5668004.77 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3297, pruned_loss=0.08313, over 5662120.17 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:26:34,951 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1218012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:26:40,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3551, 1.7932, 1.6465, 1.6311], device='cuda:0'), covar=tensor([0.2065, 0.1620, 0.1584, 0.1603], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0743, 0.0717, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 03:26:54,472 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:26:58,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:27:22,487 INFO [train.py:968] (0/2) Epoch 27, batch 33200, giga_loss[loss=0.229, simple_loss=0.3143, pruned_loss=0.07185, over 28469.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3277, pruned_loss=0.08377, over 5657788.73 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3452, pruned_loss=0.1094, over 5661261.26 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3263, pruned_loss=0.08091, over 5666882.93 frames. ], batch size: 65, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:27:30,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4483, 3.0082, 2.7623, 2.2619], device='cuda:0'), covar=tensor([0.2960, 0.1744, 0.1778, 0.2352], device='cuda:0'), in_proj_covar=tensor([0.2017, 0.1965, 0.1873, 0.2014], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 03:27:31,863 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:28:03,611 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.230e+02 1.402e+03 2.024e+03 2.973e+03 1.089e+04, threshold=4.049e+03, percent-clipped=15.0
+2023-03-14 03:28:25,909 INFO [train.py:968] (0/2) Epoch 27, batch 33250, giga_loss[loss=0.2087, simple_loss=0.298, pruned_loss=0.05969, over 28871.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3266, pruned_loss=0.08285, over 5661352.77 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3452, pruned_loss=0.1094, over 5661261.26 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3256, pruned_loss=0.08063, over 5668430.90 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:28:45,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:28:55,156 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-03-14 03:29:25,142 INFO [train.py:968] (0/2) Epoch 27, batch 33300, giga_loss[loss=0.2722, simple_loss=0.3492, pruned_loss=0.09757, over 28638.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3247, pruned_loss=0.08222, over 5666413.68 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.345, pruned_loss=0.1093, over 5663843.13 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.3237, pruned_loss=0.0801, over 5669883.11 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:29:33,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:29:34,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3835, 3.4418, 1.4992, 1.6475], device='cuda:0'), covar=tensor([0.1018, 0.0312, 0.0938, 0.1362], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0564, 0.0406, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 03:29:47,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:30:06,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.584e+02 1.460e+03 1.755e+03 2.411e+03 5.143e+03, threshold=3.510e+03, percent-clipped=2.0
+2023-03-14 03:30:22,988 INFO [train.py:968] (0/2) Epoch 27, batch 33350, giga_loss[loss=0.2424, simple_loss=0.3296, pruned_loss=0.0776, over 28910.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3247, pruned_loss=0.08177, over 5663745.92 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3442, pruned_loss=0.1088, over 5659272.53 frames. ], giga_tot_loss[loss=0.242, simple_loss=0.3243, pruned_loss=0.07991, over 5671188.47 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:30:31,079 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 03:31:23,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6426, 1.9854, 1.3339, 1.4666], device='cuda:0'), covar=tensor([0.1023, 0.0524, 0.1049, 0.1183], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0445, 0.0522, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 03:31:28,677 INFO [train.py:968] (0/2) Epoch 27, batch 33400, giga_loss[loss=0.2485, simple_loss=0.3341, pruned_loss=0.08143, over 28973.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3272, pruned_loss=0.08315, over 5668872.29 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3439, pruned_loss=0.1086, over 5662767.37 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3268, pruned_loss=0.08147, over 5671486.63 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 03:31:45,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:31:49,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:32:13,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.622e+02 1.634e+03 2.125e+03 3.018e+03 1.232e+04, threshold=4.250e+03, percent-clipped=17.0
+2023-03-14 03:32:25,054 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:32:30,980 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:32:32,969 INFO [train.py:968] (0/2) Epoch 27, batch 33450, giga_loss[loss=0.2258, simple_loss=0.3207, pruned_loss=0.06548, over 28808.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3277, pruned_loss=0.08392, over 5663051.67 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3439, pruned_loss=0.1086, over 5665598.07 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3273, pruned_loss=0.08232, over 5662633.48 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 03:32:34,009 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:32:44,829 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:32:46,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:32:46,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:33:08,040 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:33:21,763 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:33:26,781 INFO [train.py:968] (0/2) Epoch 27, batch 33500, giga_loss[loss=0.2761, simple_loss=0.3555, pruned_loss=0.09838, over 27954.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3318, pruned_loss=0.08658, over 5662907.63 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3433, pruned_loss=0.1081, over 5676790.94 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3312, pruned_loss=0.08456, over 5652037.79 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 03:33:37,668 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1218365.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:34:02,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:34:04,225 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.655e+03 2.306e+03 3.137e+03 1.971e+04, threshold=4.613e+03, percent-clipped=17.0
+2023-03-14 03:34:21,671 INFO [train.py:968] (0/2) Epoch 27, batch 33550, libri_loss[loss=0.2933, simple_loss=0.3522, pruned_loss=0.1172, over 29668.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3351, pruned_loss=0.08815, over 5671987.30 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3429, pruned_loss=0.1079, over 5680482.23 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3347, pruned_loss=0.08607, over 5659243.87 frames. ], batch size: 88, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 03:34:40,371 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1218421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:34:40,473 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4745, 1.7341, 1.4222, 1.3810], device='cuda:0'), covar=tensor([0.2716, 0.2670, 0.3041, 0.2505], device='cuda:0'), in_proj_covar=tensor([0.1589, 0.1141, 0.1407, 0.1010], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 03:35:12,558 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-14 03:35:22,495 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 03:35:23,466 INFO [train.py:968] (0/2) Epoch 27, batch 33600, giga_loss[loss=0.2409, simple_loss=0.3284, pruned_loss=0.0767, over 28624.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.336, pruned_loss=0.08868, over 5674334.99 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3431, pruned_loss=0.108, over 5684805.28 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3355, pruned_loss=0.08651, over 5659951.02 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:35:34,815 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:35:40,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:36:13,199 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.480e+03 2.088e+03 2.920e+03 6.012e+03, threshold=4.177e+03, percent-clipped=7.0
+2023-03-14 03:36:15,839 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:36:30,761 INFO [train.py:968] (0/2) Epoch 27, batch 33650, giga_loss[loss=0.2207, simple_loss=0.3046, pruned_loss=0.06845, over 28922.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3327, pruned_loss=0.0871, over 5681138.54 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3427, pruned_loss=0.1078, over 5691020.51 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3323, pruned_loss=0.08492, over 5663711.76 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:36:49,486 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-14 03:37:02,995 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:37:07,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:37:14,311 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-14 03:37:22,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0082, 3.8500, 3.6401, 1.9610], device='cuda:0'), covar=tensor([0.0685, 0.0811, 0.0910, 0.2150], device='cuda:0'), in_proj_covar=tensor([0.1271, 0.1171, 0.0988, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 03:37:30,287 INFO [train.py:968] (0/2) Epoch 27, batch 33700, giga_loss[loss=0.2476, simple_loss=0.3321, pruned_loss=0.08152, over 28896.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3325, pruned_loss=0.08699, over 5688130.95 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3424, pruned_loss=0.1075, over 5692557.38 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3321, pruned_loss=0.08498, over 5672865.54 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:37:39,605 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:38:17,113 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.319e+02 1.540e+03 2.126e+03 3.266e+03 7.135e+03, threshold=4.252e+03, percent-clipped=11.0
+2023-03-14 03:38:22,498 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-14 03:38:33,543 INFO [train.py:968] (0/2) Epoch 27, batch 33750, giga_loss[loss=0.2406, simple_loss=0.3179, pruned_loss=0.08159, over 29080.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3323, pruned_loss=0.08749, over 5679334.83 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3424, pruned_loss=0.1076, over 5688827.26 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3318, pruned_loss=0.08529, over 5670424.13 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:39:25,860 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3299, 1.2084, 3.9112, 3.3703], device='cuda:0'), covar=tensor([0.1639, 0.3000, 0.0473, 0.1672], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0668, 0.0993, 0.0959], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 03:39:37,831 INFO [train.py:968] (0/2) Epoch 27, batch 33800, giga_loss[loss=0.229, simple_loss=0.3113, pruned_loss=0.07339, over 28556.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3295, pruned_loss=0.08709, over 5682981.20 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3418, pruned_loss=0.1072, over 5695233.92 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3293, pruned_loss=0.08508, over 5669458.38 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:40:19,681 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 03:40:21,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.569e+02 1.573e+03 2.161e+03 2.748e+03 7.346e+03, threshold=4.323e+03, percent-clipped=6.0
+2023-03-14 03:40:40,935 INFO [train.py:968] (0/2) Epoch 27, batch 33850, giga_loss[loss=0.2776, simple_loss=0.3532, pruned_loss=0.1011, over 28817.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3291, pruned_loss=0.08676, over 5684436.73 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3421, pruned_loss=0.1073, over 5693681.68 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3283, pruned_loss=0.08456, over 5675163.15 frames. ], batch size: 243, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:40:43,429 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0295, 4.8744, 4.6482, 2.3252], device='cuda:0'), covar=tensor([0.0538, 0.0675, 0.0931, 0.1989], device='cuda:0'), in_proj_covar=tensor([0.1276, 0.1176, 0.0992, 0.0732], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 03:40:48,867 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1218712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:41:03,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3654, 1.7087, 1.3752, 1.1707], device='cuda:0'), covar=tensor([0.2993, 0.2926, 0.3465, 0.2523], device='cuda:0'), in_proj_covar=tensor([0.1591, 0.1143, 0.1408, 0.1010], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 03:41:20,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218740.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:41:39,038 INFO [train.py:968] (0/2) Epoch 27, batch 33900, giga_loss[loss=0.2221, simple_loss=0.3117, pruned_loss=0.06625, over 28959.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.328, pruned_loss=0.08477, over 5672718.50 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3422, pruned_loss=0.1073, over 5687892.48 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.327, pruned_loss=0.08252, over 5670262.95 frames. ], batch size: 175, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:42:08,535 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1745, 1.5614, 1.5702, 1.3736], device='cuda:0'), covar=tensor([0.2171, 0.1869, 0.2160, 0.1911], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0740, 0.0714, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 03:42:21,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.555e+02 1.449e+03 1.979e+03 2.567e+03 5.704e+03, threshold=3.957e+03, percent-clipped=2.0
+2023-03-14 03:42:26,933 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:42:29,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2461, 1.8112, 1.4058, 0.4579], device='cuda:0'), covar=tensor([0.5042, 0.3293, 0.4774, 0.7182], device='cuda:0'), in_proj_covar=tensor([0.1835, 0.1730, 0.1656, 0.1502], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 03:42:37,114 INFO [train.py:968] (0/2) Epoch 27, batch 33950, giga_loss[loss=0.2484, simple_loss=0.3375, pruned_loss=0.07962, over 28743.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3272, pruned_loss=0.08285, over 5668992.17 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3419, pruned_loss=0.1071, over 5681820.53 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3265, pruned_loss=0.0809, over 5672170.86 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:43:14,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1138, 1.2570, 1.1433, 0.8771], device='cuda:0'), covar=tensor([0.1121, 0.0522, 0.1115, 0.1150], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0444, 0.0521, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 03:43:36,731 INFO [train.py:968] (0/2) Epoch 27, batch 34000, giga_loss[loss=0.2516, simple_loss=0.3401, pruned_loss=0.08154, over 28876.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3286, pruned_loss=0.08169, over 5676570.89 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3418, pruned_loss=0.107, over 5684925.05 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3279, pruned_loss=0.07992, over 5676040.90 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:44:05,506 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218883.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:44:08,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:44:14,054 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.496e+02 1.466e+03 1.924e+03 2.770e+03 9.095e+03, threshold=3.848e+03, percent-clipped=5.0
+2023-03-14 03:44:28,346 INFO [train.py:968] (0/2) Epoch 27, batch 34050, giga_loss[loss=0.217, simple_loss=0.2869, pruned_loss=0.07356, over 24364.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3298, pruned_loss=0.08253, over 5679914.63 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3417, pruned_loss=0.107, over 5690290.52 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3288, pruned_loss=0.08016, over 5674533.19 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:44:43,906 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218915.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:44:53,165 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-14 03:45:16,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:45:21,894 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:45:36,372 INFO [train.py:968] (0/2) Epoch 27, batch 34100, giga_loss[loss=0.2379, simple_loss=0.3215, pruned_loss=0.07718, over 27760.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3308, pruned_loss=0.08367, over 5677263.03 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3419, pruned_loss=0.1073, over 5692581.93 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3297, pruned_loss=0.08108, over 5670719.75 frames. ], batch size: 474, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:45:58,191 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:46:22,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.878e+02 1.516e+03 2.042e+03 2.595e+03 7.559e+03, threshold=4.084e+03, percent-clipped=10.0
+2023-03-14 03:46:44,257 INFO [train.py:968] (0/2) Epoch 27, batch 34150, libri_loss[loss=0.2984, simple_loss=0.36, pruned_loss=0.1183, over 29542.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3306, pruned_loss=0.08326, over 5671092.35 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.342, pruned_loss=0.1073, over 5693661.02 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3296, pruned_loss=0.08104, over 5664781.54 frames. ], batch size: 81, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:47:11,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9990, 3.8759, 3.6385, 2.0969], device='cuda:0'), covar=tensor([0.0630, 0.0722, 0.0751, 0.1980], device='cuda:0'), in_proj_covar=tensor([0.1271, 0.1174, 0.0988, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 03:47:13,789 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219025.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:47:28,352 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3225, 4.1731, 4.0225, 1.7724], device='cuda:0'), covar=tensor([0.0652, 0.0710, 0.0854, 0.2153], device='cuda:0'), in_proj_covar=tensor([0.1271, 0.1174, 0.0988, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 03:47:57,609 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:47:57,962 INFO [train.py:968] (0/2) Epoch 27, batch 34200, giga_loss[loss=0.195, simple_loss=0.2743, pruned_loss=0.05788, over 24677.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3301, pruned_loss=0.08197, over 5673310.96 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3417, pruned_loss=0.1072, over 5695953.75 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3294, pruned_loss=0.08008, over 5666074.85 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:48:29,445 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5389, 1.6864, 1.6091, 1.5249], device='cuda:0'), covar=tensor([0.2392, 0.2375, 0.1893, 0.2188], device='cuda:0'), in_proj_covar=tensor([0.2017, 0.1966, 0.1874, 0.2015], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 03:48:30,711 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3501, 1.2208, 3.7948, 3.0836], device='cuda:0'), covar=tensor([0.1624, 0.2845, 0.0485, 0.1115], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0669, 0.0993, 0.0961], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 03:48:41,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5846, 1.6993, 1.2696, 1.3009], device='cuda:0'), covar=tensor([0.1015, 0.0582, 0.0992, 0.1216], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0445, 0.0521, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 03:48:43,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:48:49,621 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.097e+02 1.534e+03 2.024e+03 2.720e+03 7.091e+03, threshold=4.047e+03, percent-clipped=6.0
+2023-03-14 03:49:06,524 INFO [train.py:968] (0/2) Epoch 27, batch 34250, giga_loss[loss=0.25, simple_loss=0.337, pruned_loss=0.08149, over 28697.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3326, pruned_loss=0.08381, over 5677058.87 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3419, pruned_loss=0.1073, over 5700741.31 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3317, pruned_loss=0.08149, over 5666227.20 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:49:40,779 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:50:09,285 INFO [train.py:968] (0/2) Epoch 27, batch 34300, giga_loss[loss=0.2452, simple_loss=0.3321, pruned_loss=0.0792, over 28877.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.337, pruned_loss=0.08608, over 5679260.05 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3422, pruned_loss=0.1076, over 5701889.33 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3358, pruned_loss=0.08353, over 5668955.52 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:50:55,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.656e+02 1.670e+03 2.294e+03 3.212e+03 1.169e+04, threshold=4.588e+03, percent-clipped=14.0
+2023-03-14 03:51:14,703 INFO [train.py:968] (0/2) Epoch 27, batch 34350, giga_loss[loss=0.2328, simple_loss=0.3222, pruned_loss=0.07172, over 28757.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3357, pruned_loss=0.08536, over 5672451.75 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3419, pruned_loss=0.1073, over 5695049.10 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3349, pruned_loss=0.08306, over 5670008.00 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:51:49,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1219230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:51:53,248 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1219233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:52:22,096 INFO [train.py:968] (0/2) Epoch 27, batch 34400, giga_loss[loss=0.2209, simple_loss=0.3081, pruned_loss=0.06688, over 28968.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3348, pruned_loss=0.0857, over 5670904.43 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3421, pruned_loss=0.1076, over 5688155.18 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3339, pruned_loss=0.08341, over 5674272.80 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:52:34,689 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1219262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:53:19,688 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.956e+02 1.456e+03 1.715e+03 2.351e+03 5.167e+03, threshold=3.430e+03, percent-clipped=3.0
+2023-03-14 03:53:35,289 INFO [train.py:968] (0/2) Epoch 27, batch 34450, giga_loss[loss=0.2585, simple_loss=0.345, pruned_loss=0.08605, over 29049.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3322, pruned_loss=0.08351, over 5675892.56 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3422, pruned_loss=0.1077, over 5691527.64 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3313, pruned_loss=0.08129, over 5675447.60 frames. ], batch size: 285, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:54:35,551 INFO [train.py:968] (0/2) Epoch 27, batch 34500, giga_loss[loss=0.2142, simple_loss=0.3074, pruned_loss=0.06051, over 28890.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3308, pruned_loss=0.08251, over 5676824.40 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3425, pruned_loss=0.108, over 5689558.23 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3295, pruned_loss=0.07973, over 5677235.44 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:55:27,115 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.750e+02 1.337e+03 1.746e+03 2.480e+03 8.605e+03, threshold=3.492e+03, percent-clipped=9.0
+2023-03-14 03:55:37,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219400.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:55:41,040 INFO [train.py:968] (0/2) Epoch 27, batch 34550, giga_loss[loss=0.242, simple_loss=0.3215, pruned_loss=0.08126, over 28515.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.332, pruned_loss=0.08337, over 5669289.54 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3422, pruned_loss=0.1079, over 5690798.43 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3312, pruned_loss=0.08119, over 5668291.81 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:56:11,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:56:45,784 INFO [train.py:968] (0/2) Epoch 27, batch 34600, giga_loss[loss=0.2351, simple_loss=0.3202, pruned_loss=0.07495, over 28854.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3341, pruned_loss=0.08406, over 5672233.81 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3422, pruned_loss=0.1078, over 5691882.44 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3335, pruned_loss=0.08226, over 5670331.50 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:57:28,215 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.556e+03 1.934e+03 2.745e+03 8.999e+03, threshold=3.867e+03, percent-clipped=14.0
+2023-03-14 03:57:46,776 INFO [train.py:968] (0/2) Epoch 27, batch 34650, giga_loss[loss=0.2214, simple_loss=0.2899, pruned_loss=0.07647, over 24246.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3328, pruned_loss=0.08437, over 5674658.19 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3423, pruned_loss=0.1079, over 5695228.22 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3322, pruned_loss=0.08253, over 5670084.81 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:57:48,480 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:58:22,867 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1219543.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:58:25,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1219546.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:58:32,161 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:58:37,734 INFO [train.py:968] (0/2) Epoch 27, batch 34700, giga_loss[loss=0.2397, simple_loss=0.3291, pruned_loss=0.07516, over 29002.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3316, pruned_loss=0.08493, over 5656475.34 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3422, pruned_loss=0.108, over 5680564.27 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3309, pruned_loss=0.08283, over 5664742.56 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:58:57,303 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:58:58,063 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1219572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:59:02,106 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1219575.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:59:02,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1219575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:59:19,142 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.192e+02 1.484e+03 1.770e+03 2.371e+03 5.638e+03, threshold=3.540e+03, percent-clipped=6.0
+2023-03-14 03:59:26,204 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-14 03:59:34,790 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1219604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:59:35,178 INFO [train.py:968] (0/2) Epoch 27, batch 34750, giga_loss[loss=0.2633, simple_loss=0.3494, pruned_loss=0.08857, over 28644.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3311, pruned_loss=0.08493, over 5657268.67 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3422, pruned_loss=0.1081, over 5674167.05 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3304, pruned_loss=0.08296, over 5668361.16 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:00:00,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6653, 1.9705, 1.7919, 1.6767], device='cuda:0'), covar=tensor([0.2111, 0.2270, 0.2093, 0.2220], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0739, 0.0713, 0.0682], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 04:00:18,898 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1219650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:00:21,902 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219653.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:00:21,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1219653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:00:22,991 INFO [train.py:968] (0/2) Epoch 27, batch 34800, giga_loss[loss=0.3102, simple_loss=0.3874, pruned_loss=0.1165, over 28652.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3407, pruned_loss=0.09108, over 5654051.17 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3428, pruned_loss=0.1085, over 5669568.07 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3394, pruned_loss=0.08854, over 5666351.40 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:00:49,605 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1219682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:00:57,443 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.573e+03 2.008e+03 2.642e+03 1.008e+04, threshold=4.016e+03, percent-clipped=15.0
+2023-03-14 04:01:06,438 INFO [train.py:968] (0/2) Epoch 27, batch 34850, giga_loss[loss=0.2808, simple_loss=0.3728, pruned_loss=0.09441, over 28658.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3486, pruned_loss=0.09575, over 5660937.74 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3431, pruned_loss=0.1087, over 5674098.87 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3473, pruned_loss=0.09322, over 5666201.84 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:01:34,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3937, 3.4538, 1.5493, 1.4890], device='cuda:0'), covar=tensor([0.1087, 0.0333, 0.0954, 0.1511], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0564, 0.0407, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 04:01:49,665 INFO [train.py:968] (0/2) Epoch 27, batch 34900, giga_loss[loss=0.267, simple_loss=0.3437, pruned_loss=0.09512, over 28953.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3499, pruned_loss=0.09722, over 5660124.46 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3429, pruned_loss=0.1085, over 5668487.23 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3493, pruned_loss=0.09516, over 5668105.52 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:02:21,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.844e+02 1.224e+03 1.599e+03 2.361e+03 6.220e+03, threshold=3.199e+03, percent-clipped=3.0
+2023-03-14 04:02:30,306 INFO [train.py:968] (0/2) Epoch 27, batch 34950, giga_loss[loss=0.2205, simple_loss=0.3029, pruned_loss=0.06902, over 28943.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3435, pruned_loss=0.09459, over 5670416.98 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3427, pruned_loss=0.1084, over 5666861.37 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3432, pruned_loss=0.09286, over 5678389.03 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:03:11,982 INFO [train.py:968] (0/2) Epoch 27, batch 35000, giga_loss[loss=0.2047, simple_loss=0.2887, pruned_loss=0.06039, over 29092.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3371, pruned_loss=0.09195, over 5669186.20 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3428, pruned_loss=0.1084, over 5660038.59 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3367, pruned_loss=0.09041, over 5681347.17 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:03:40,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.200e+02 1.088e+03 1.479e+03 2.041e+03 6.495e+03, threshold=2.957e+03, percent-clipped=9.0
+2023-03-14 04:03:52,008 INFO [train.py:968] (0/2) Epoch 27, batch 35050, giga_loss[loss=0.2041, simple_loss=0.2838, pruned_loss=0.0622, over 28917.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3298, pruned_loss=0.089, over 5672652.37 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3425, pruned_loss=0.1082, over 5660468.82 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3295, pruned_loss=0.08756, over 5682262.57 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:04:08,113 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:04:26,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:04:31,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3943, 1.2760, 3.4738, 3.2131], device='cuda:0'), covar=tensor([0.1494, 0.2813, 0.0482, 0.1515], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0671, 0.0999, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 04:04:32,075 INFO [train.py:968] (0/2) Epoch 27, batch 35100, giga_loss[loss=0.23, simple_loss=0.2996, pruned_loss=0.08016, over 28838.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.322, pruned_loss=0.08552, over 5679398.40 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3427, pruned_loss=0.1082, over 5662938.27 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3214, pruned_loss=0.08402, over 5684864.93 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:05:05,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.146e+02 1.088e+03 1.300e+03 1.830e+03 4.887e+03, threshold=2.601e+03, percent-clipped=7.0
+2023-03-14 04:05:10,925 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1220000.pt
+2023-03-14 04:05:15,770 INFO [train.py:968] (0/2) Epoch 27, batch 35150, giga_loss[loss=0.2166, simple_loss=0.2828, pruned_loss=0.07525, over 28698.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3161, pruned_loss=0.08306, over 5679541.52 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3429, pruned_loss=0.1082, over 5663279.32 frames. ], giga_tot_loss[loss=0.2392, simple_loss=0.3152, pruned_loss=0.08159, over 5683589.12 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:05:36,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1220028.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:05:57,781 INFO [train.py:968] (0/2) Epoch 27, batch 35200, giga_loss[loss=0.2321, simple_loss=0.3058, pruned_loss=0.07919, over 28802.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3122, pruned_loss=0.0811, over 5671282.73 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3431, pruned_loss=0.1082, over 5656277.78 frames. ], giga_tot_loss[loss=0.2352, simple_loss=0.311, pruned_loss=0.07967, over 5680825.25 frames. ], batch size: 243, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:06:09,351 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:06:11,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:06:21,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5317, 3.6539, 1.6168, 1.7840], device='cuda:0'), covar=tensor([0.0977, 0.0336, 0.0916, 0.1237], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0565, 0.0407, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 04:06:26,793 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:06:28,986 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:06:29,388 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.643e+02 1.180e+03 1.389e+03 1.780e+03 5.611e+03, threshold=2.778e+03, percent-clipped=10.0
+2023-03-14 04:06:35,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:06:38,642 INFO [train.py:968] (0/2) Epoch 27, batch 35250, giga_loss[loss=0.2166, simple_loss=0.2955, pruned_loss=0.06889, over 29116.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3085, pruned_loss=0.07921, over 5687613.47 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3432, pruned_loss=0.1082, over 5659818.20 frames. ], giga_tot_loss[loss=0.2312, simple_loss=0.307, pruned_loss=0.07771, over 5692282.55 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:06:51,936 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:07:19,585 INFO [train.py:968] (0/2) Epoch 27, batch 35300, giga_loss[loss=0.2236, simple_loss=0.2992, pruned_loss=0.07394, over 28846.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3078, pruned_loss=0.07923, over 5699422.31 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3441, pruned_loss=0.1084, over 5667653.71 frames. ], giga_tot_loss[loss=0.2287, simple_loss=0.3043, pruned_loss=0.07659, over 5697375.97 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:07:35,310 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220171.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 04:07:37,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220174.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 04:07:55,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.442e+02 1.168e+03 1.413e+03 2.049e+03 4.252e+03, threshold=2.826e+03, percent-clipped=8.0
+2023-03-14 04:08:02,234 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220203.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:08:03,231 INFO [train.py:968] (0/2) Epoch 27, batch 35350, libri_loss[loss=0.3648, simple_loss=0.4148, pruned_loss=0.1574, over 25948.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3053, pruned_loss=0.0782, over 5700994.06 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3441, pruned_loss=0.1084, over 5668987.05 frames. ], giga_tot_loss[loss=0.2263, simple_loss=0.3017, pruned_loss=0.07549, over 5698836.96 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:08:39,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-14 04:08:43,918 INFO [train.py:968] (0/2) Epoch 27, batch 35400, giga_loss[loss=0.1998, simple_loss=0.2816, pruned_loss=0.059, over 28811.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.3032, pruned_loss=0.07731, over 5696860.68 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3442, pruned_loss=0.1084, over 5669428.61 frames. ], giga_tot_loss[loss=0.2233, simple_loss=0.2987, pruned_loss=0.07396, over 5695682.58 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:09:12,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.845e+02 1.153e+03 1.476e+03 2.092e+03 7.949e+03, threshold=2.952e+03, percent-clipped=12.0
+2023-03-14 04:09:20,948 INFO [train.py:968] (0/2) Epoch 27, batch 35450, libri_loss[loss=0.216, simple_loss=0.288, pruned_loss=0.07195, over 29700.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3027, pruned_loss=0.07773, over 5692886.85 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3444, pruned_loss=0.1084, over 5669082.37 frames. ], giga_tot_loss[loss=0.2222, simple_loss=0.297, pruned_loss=0.07367, over 5692793.92 frames. ], batch size: 69, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:09:26,686 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3897, 1.2702, 3.9109, 3.1508], device='cuda:0'), covar=tensor([0.1593, 0.2751, 0.0540, 0.1125], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0673, 0.1004, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 04:09:46,281 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1220335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:10:01,028 INFO [train.py:968] (0/2) Epoch 27, batch 35500, libri_loss[loss=0.3421, simple_loss=0.4109, pruned_loss=0.1366, over 29118.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3004, pruned_loss=0.07645, over 5683713.90 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.345, pruned_loss=0.1086, over 5663508.53 frames. ], giga_tot_loss[loss=0.219, simple_loss=0.294, pruned_loss=0.07197, over 5688988.62 frames. ], batch size: 101, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:10:05,336 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1220360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:10:36,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.314e+02 1.091e+03 1.443e+03 2.273e+03 1.096e+04, threshold=2.885e+03, percent-clipped=12.0
+2023-03-14 04:10:46,595 INFO [train.py:968] (0/2) Epoch 27, batch 35550, giga_loss[loss=0.225, simple_loss=0.2981, pruned_loss=0.07596, over 28670.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2958, pruned_loss=0.0742, over 5692089.43 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3451, pruned_loss=0.1087, over 5664680.99 frames. ], giga_tot_loss[loss=0.2157, simple_loss=0.2904, pruned_loss=0.07045, over 5695266.91 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:10:57,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1220416.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 04:11:02,839 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1220422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:11:28,556 INFO [train.py:968] (0/2) Epoch 27, batch 35600, giga_loss[loss=0.2348, simple_loss=0.3098, pruned_loss=0.07996, over 28798.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2979, pruned_loss=0.0759, over 5688012.76 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3451, pruned_loss=0.1085, over 5669351.65 frames. ], giga_tot_loss[loss=0.2183, simple_loss=0.2922, pruned_loss=0.07218, over 5686890.02 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:12:04,003 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.134e+02 1.258e+03 1.672e+03 2.183e+03 4.972e+03, threshold=3.343e+03, percent-clipped=7.0
+2023-03-14 04:12:13,516 INFO [train.py:968] (0/2) Epoch 27, batch 35650, giga_loss[loss=0.2626, simple_loss=0.3405, pruned_loss=0.09231, over 28437.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.3093, pruned_loss=0.08172, over 5689932.10 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3457, pruned_loss=0.1088, over 5671524.57 frames. ], giga_tot_loss[loss=0.2297, simple_loss=0.3035, pruned_loss=0.07795, over 5687385.36 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:12:18,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2403, 1.3330, 1.2381, 0.9866], device='cuda:0'), covar=tensor([0.0919, 0.0473, 0.0999, 0.1000], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0443, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 04:12:25,131 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-14 04:12:57,515 INFO [train.py:968] (0/2) Epoch 27, batch 35700, giga_loss[loss=0.287, simple_loss=0.365, pruned_loss=0.1045, over 28553.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3218, pruned_loss=0.08808, over 5693228.22 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3459, pruned_loss=0.1087, over 5678429.68 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3162, pruned_loss=0.08445, over 5685621.98 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:13:28,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.492e+03 1.859e+03 2.384e+03 5.991e+03, threshold=3.718e+03, percent-clipped=8.0
+2023-03-14 04:13:37,430 INFO [train.py:968] (0/2) Epoch 27, batch 35750, giga_loss[loss=0.3202, simple_loss=0.3833, pruned_loss=0.1286, over 28697.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3325, pruned_loss=0.09324, over 5697994.19 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.346, pruned_loss=0.1086, over 5683638.40 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3275, pruned_loss=0.09001, over 5687620.90 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:13:56,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3821, 3.5250, 1.6808, 1.4465], device='cuda:0'), covar=tensor([0.1071, 0.0357, 0.0922, 0.1467], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0564, 0.0406, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 04:14:08,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1220642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:14:18,648 INFO [train.py:968] (0/2) Epoch 27, batch 35800, giga_loss[loss=0.2778, simple_loss=0.3624, pruned_loss=0.09661, over 29066.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3372, pruned_loss=0.0942, over 5703231.15 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.346, pruned_loss=0.1084, over 5688416.81 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3329, pruned_loss=0.09139, over 5690965.70 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:14:27,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4721, 1.4956, 1.6898, 1.2938], device='cuda:0'), covar=tensor([0.1763, 0.2542, 0.1461, 0.1751], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0714, 0.0983, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 04:14:43,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9684, 1.1301, 1.0966, 0.9173], device='cuda:0'), covar=tensor([0.2460, 0.3014, 0.1817, 0.2373], device='cuda:0'), in_proj_covar=tensor([0.2032, 0.1976, 0.1889, 0.2032], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 04:14:50,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.033e+02 1.316e+03 1.618e+03 1.972e+03 4.112e+03, threshold=3.236e+03, percent-clipped=2.0
+2023-03-14 04:14:59,346 INFO [train.py:968] (0/2) Epoch 27, batch 35850, giga_loss[loss=0.2533, simple_loss=0.3427, pruned_loss=0.08195, over 28677.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3389, pruned_loss=0.09388, over 5702513.90 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3465, pruned_loss=0.1086, over 5693631.37 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3346, pruned_loss=0.0909, over 5688214.64 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:15:03,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1220710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:15:25,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1220735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:15:46,696 INFO [train.py:968] (0/2) Epoch 27, batch 35900, giga_loss[loss=0.2824, simple_loss=0.3598, pruned_loss=0.1026, over 29070.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3408, pruned_loss=0.09402, over 5698625.76 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3463, pruned_loss=0.1084, over 5695630.54 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3376, pruned_loss=0.09172, over 5685841.37 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:15:49,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2790, 1.4859, 3.7739, 3.2295], device='cuda:0'), covar=tensor([0.1719, 0.2717, 0.0468, 0.0995], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0671, 0.1002, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 04:15:49,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1492, 1.2497, 3.6736, 3.0975], device='cuda:0'), covar=tensor([0.1827, 0.2893, 0.0510, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0671, 0.1002, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 04:16:16,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1220791.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 04:16:19,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.928e+02 1.324e+03 1.703e+03 2.526e+03 7.159e+03, threshold=3.406e+03, percent-clipped=12.0
+2023-03-14 04:16:22,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1220797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:16:29,206 INFO [train.py:968] (0/2) Epoch 27, batch 35950, giga_loss[loss=0.3032, simple_loss=0.3757, pruned_loss=0.1153, over 28727.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3442, pruned_loss=0.097, over 5690685.85 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3467, pruned_loss=0.1086, over 5697594.37 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3412, pruned_loss=0.09481, over 5678920.50 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:16:50,343 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3596, 1.5630, 1.5089, 1.5191], device='cuda:0'), covar=tensor([0.0698, 0.0310, 0.0297, 0.0726], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 04:17:09,447 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:17:10,459 INFO [train.py:968] (0/2) Epoch 27, batch 36000, giga_loss[loss=0.2454, simple_loss=0.3377, pruned_loss=0.07655, over 28869.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3475, pruned_loss=0.09918, over 5685774.10 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3467, pruned_loss=0.1085, over 5693207.41 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3451, pruned_loss=0.09727, over 5679369.72 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:17:10,464 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 04:17:18,231 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5338, 1.7297, 1.3833, 1.3732], device='cuda:0'), covar=tensor([0.0944, 0.0470, 0.1035, 0.0950], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0444, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 04:17:20,036 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2013, simple_loss=0.3091, pruned_loss=0.04675, over 944034.00 frames. 
+2023-03-14 04:17:20,037 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 04:17:20,978 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:17:38,413 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:17:39,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:17:42,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:17:50,738 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.010e+02 1.355e+03 1.705e+03 2.469e+03 5.993e+03, threshold=3.411e+03, percent-clipped=7.0
+2023-03-14 04:17:57,486 INFO [train.py:968] (0/2) Epoch 27, batch 36050, giga_loss[loss=0.2737, simple_loss=0.3533, pruned_loss=0.09706, over 28949.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3514, pruned_loss=0.1015, over 5676676.49 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3484, pruned_loss=0.1093, over 5676246.92 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3481, pruned_loss=0.09875, over 5686817.21 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:18:02,119 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:18:02,237 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3758, 1.6359, 1.6228, 1.2053], device='cuda:0'), covar=tensor([0.1690, 0.2626, 0.1507, 0.1768], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0713, 0.0981, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 04:18:21,059 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220934.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:18:23,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220937.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 04:18:25,881 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:18:27,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:18:36,585 INFO [train.py:968] (0/2) Epoch 27, batch 36100, giga_loss[loss=0.2586, simple_loss=0.3457, pruned_loss=0.08577, over 28180.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3551, pruned_loss=0.1027, over 5683389.99 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3484, pruned_loss=0.1093, over 5679064.50 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3524, pruned_loss=0.1005, over 5688965.48 frames. ], batch size: 77, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:18:47,344 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220966.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 04:18:51,868 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:19:13,241 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.369e+02 1.372e+03 1.768e+03 2.353e+03 7.390e+03, threshold=3.536e+03, percent-clipped=6.0
+2023-03-14 04:19:19,222 INFO [train.py:968] (0/2) Epoch 27, batch 36150, giga_loss[loss=0.2724, simple_loss=0.3537, pruned_loss=0.09554, over 28805.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3562, pruned_loss=0.1033, over 5679175.44 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3485, pruned_loss=0.1092, over 5682311.79 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3542, pruned_loss=0.1014, over 5680877.07 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:19:31,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1221017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:19:43,938 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-14 04:20:01,774 INFO [train.py:968] (0/2) Epoch 27, batch 36200, giga_loss[loss=0.2572, simple_loss=0.3437, pruned_loss=0.08537, over 28993.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3563, pruned_loss=0.1017, over 5685482.62 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3486, pruned_loss=0.1092, over 5683090.23 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3547, pruned_loss=0.1002, over 5686188.97 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:20:23,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9877, 3.1034, 2.0470, 1.0594], device='cuda:0'), covar=tensor([1.0000, 0.3661, 0.4922, 0.8831], device='cuda:0'), in_proj_covar=tensor([0.1829, 0.1730, 0.1656, 0.1498], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 04:20:34,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.230e+02 1.228e+03 1.413e+03 1.732e+03 4.000e+03, threshold=2.826e+03, percent-clipped=2.0
+2023-03-14 04:20:41,309 INFO [train.py:968] (0/2) Epoch 27, batch 36250, giga_loss[loss=0.2387, simple_loss=0.3254, pruned_loss=0.07599, over 28761.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3547, pruned_loss=0.09918, over 5696337.45 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3491, pruned_loss=0.1094, over 5685768.73 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3531, pruned_loss=0.09765, over 5694637.68 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:21:00,768 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9438, 1.2636, 3.3743, 2.9465], device='cuda:0'), covar=tensor([0.1849, 0.2818, 0.0494, 0.0976], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0668, 0.0998, 0.0966], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 04:21:21,825 INFO [train.py:968] (0/2) Epoch 27, batch 36300, giga_loss[loss=0.2566, simple_loss=0.3452, pruned_loss=0.08403, over 28563.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.354, pruned_loss=0.09786, over 5701767.33 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3498, pruned_loss=0.1097, over 5687593.65 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3522, pruned_loss=0.09615, over 5699114.48 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:21:26,648 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1221160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:21:28,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1221163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:21:51,698 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1221192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:21:56,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.600e+02 1.175e+03 1.512e+03 1.947e+03 6.526e+03, threshold=3.025e+03, percent-clipped=4.0
+2023-03-14 04:22:03,147 INFO [train.py:968] (0/2) Epoch 27, batch 36350, giga_loss[loss=0.2608, simple_loss=0.3396, pruned_loss=0.09093, over 28740.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3522, pruned_loss=0.09683, over 5708248.57 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3498, pruned_loss=0.1096, over 5689595.11 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3509, pruned_loss=0.09541, over 5704601.87 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:22:39,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5624, 1.8225, 1.3319, 1.3262], device='cuda:0'), covar=tensor([0.1159, 0.0590, 0.1049, 0.1184], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0444, 0.0522, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 04:22:43,584 INFO [train.py:968] (0/2) Epoch 27, batch 36400, giga_loss[loss=0.2939, simple_loss=0.367, pruned_loss=0.1104, over 28912.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3551, pruned_loss=0.1012, over 5711331.29 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3499, pruned_loss=0.1095, over 5696101.60 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3539, pruned_loss=0.09974, over 5703257.42 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:22:47,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6063, 1.6137, 1.5893, 1.5285], device='cuda:0'), covar=tensor([0.2610, 0.2716, 0.2400, 0.2551], device='cuda:0'), in_proj_covar=tensor([0.2039, 0.1983, 0.1898, 0.2043], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 04:23:07,353 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5325, 1.6654, 1.5253, 1.3702], device='cuda:0'), covar=tensor([0.2854, 0.2646, 0.2272, 0.2717], device='cuda:0'), in_proj_covar=tensor([0.2041, 0.1984, 0.1899, 0.2044], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 04:23:20,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.445e+03 1.795e+03 2.295e+03 4.862e+03, threshold=3.590e+03, percent-clipped=6.0
+2023-03-14 04:23:26,137 INFO [train.py:968] (0/2) Epoch 27, batch 36450, giga_loss[loss=0.2834, simple_loss=0.3524, pruned_loss=0.1072, over 28738.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3568, pruned_loss=0.1045, over 5705159.97 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3498, pruned_loss=0.1094, over 5698177.71 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3561, pruned_loss=0.1033, over 5696943.74 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:24:09,294 INFO [train.py:968] (0/2) Epoch 27, batch 36500, giga_loss[loss=0.2564, simple_loss=0.3364, pruned_loss=0.08825, over 28317.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3565, pruned_loss=0.1055, over 5697810.29 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3503, pruned_loss=0.1097, over 5690947.55 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3557, pruned_loss=0.1043, over 5697431.52 frames. ], batch size: 65, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:24:32,049 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1221382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:24:32,948 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-14 04:24:45,215 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-14 04:24:45,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.752e+02 1.368e+03 1.711e+03 2.423e+03 1.399e+04, threshold=3.422e+03, percent-clipped=11.0
+2023-03-14 04:24:51,480 INFO [train.py:968] (0/2) Epoch 27, batch 36550, giga_loss[loss=0.2519, simple_loss=0.3224, pruned_loss=0.09074, over 28508.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3538, pruned_loss=0.1044, over 5702074.92 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3507, pruned_loss=0.1099, over 5694054.98 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3529, pruned_loss=0.1032, over 5699188.35 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:25:23,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7035, 1.7043, 1.8858, 1.4695], device='cuda:0'), covar=tensor([0.1919, 0.2555, 0.1548, 0.1813], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0711, 0.0978, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 04:25:33,081 INFO [train.py:968] (0/2) Epoch 27, batch 36600, giga_loss[loss=0.3287, simple_loss=0.3832, pruned_loss=0.1371, over 26836.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3533, pruned_loss=0.1044, over 5700221.23 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3512, pruned_loss=0.1099, over 5697452.85 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3522, pruned_loss=0.1032, over 5694847.38 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:25:43,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2875, 1.6364, 1.3320, 1.1291], device='cuda:0'), covar=tensor([0.2792, 0.2802, 0.3215, 0.2543], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1143, 0.1402, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 04:25:46,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4049, 1.5259, 1.2141, 1.0830], device='cuda:0'), covar=tensor([0.0983, 0.0515, 0.0993, 0.1159], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0444, 0.0521, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 04:26:07,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.843e+02 1.260e+03 1.581e+03 2.095e+03 3.953e+03, threshold=3.162e+03, percent-clipped=1.0
+2023-03-14 04:26:16,762 INFO [train.py:968] (0/2) Epoch 27, batch 36650, giga_loss[loss=0.3526, simple_loss=0.4021, pruned_loss=0.1515, over 26590.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3523, pruned_loss=0.1031, over 5699418.01 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3515, pruned_loss=0.1101, over 5700647.13 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3512, pruned_loss=0.1019, over 5692453.34 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:26:58,325 INFO [train.py:968] (0/2) Epoch 27, batch 36700, giga_loss[loss=0.2745, simple_loss=0.3346, pruned_loss=0.1072, over 23626.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3497, pruned_loss=0.1009, over 5693735.84 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3518, pruned_loss=0.1101, over 5703489.46 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3485, pruned_loss=0.09973, over 5685343.79 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:27:37,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.527e+02 1.171e+03 1.481e+03 2.056e+03 6.632e+03, threshold=2.962e+03, percent-clipped=6.0
+2023-03-14 04:27:42,739 INFO [train.py:968] (0/2) Epoch 27, batch 36750, giga_loss[loss=0.2535, simple_loss=0.327, pruned_loss=0.08998, over 27993.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3449, pruned_loss=0.09833, over 5681639.48 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3519, pruned_loss=0.11, over 5698935.99 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3437, pruned_loss=0.0971, over 5678684.89 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:28:33,871 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-14 04:28:34,069 INFO [train.py:968] (0/2) Epoch 27, batch 36800, giga_loss[loss=0.3132, simple_loss=0.3651, pruned_loss=0.1306, over 26496.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3389, pruned_loss=0.09564, over 5668778.03 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3522, pruned_loss=0.1103, over 5701120.80 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3376, pruned_loss=0.0943, over 5664303.91 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:29:19,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.655e+02 1.078e+03 1.284e+03 1.695e+03 7.183e+03, threshold=2.567e+03, percent-clipped=6.0
+2023-03-14 04:29:28,452 INFO [train.py:968] (0/2) Epoch 27, batch 36850, giga_loss[loss=0.2591, simple_loss=0.3339, pruned_loss=0.09216, over 28844.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3344, pruned_loss=0.09369, over 5652010.69 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3524, pruned_loss=0.1103, over 5699085.25 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.333, pruned_loss=0.09239, over 5649702.55 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:30:13,434 INFO [train.py:968] (0/2) Epoch 27, batch 36900, giga_loss[loss=0.2252, simple_loss=0.3101, pruned_loss=0.07012, over 28853.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3341, pruned_loss=0.09245, over 5668955.25 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3526, pruned_loss=0.1103, over 5703169.84 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3324, pruned_loss=0.09109, over 5662676.42 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:30:15,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1221757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:30:48,004 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.686e+02 1.134e+03 1.385e+03 1.823e+03 5.005e+03, threshold=2.771e+03, percent-clipped=7.0
+2023-03-14 04:30:54,134 INFO [train.py:968] (0/2) Epoch 27, batch 36950, giga_loss[loss=0.2584, simple_loss=0.3362, pruned_loss=0.09036, over 28971.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3352, pruned_loss=0.093, over 5662501.26 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3527, pruned_loss=0.1104, over 5696553.90 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3336, pruned_loss=0.09166, over 5663666.03 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:31:13,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3227, 1.5440, 1.2934, 1.6199], device='cuda:0'), covar=tensor([0.0818, 0.0354, 0.0363, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 04:31:33,328 INFO [train.py:968] (0/2) Epoch 27, batch 37000, giga_loss[loss=0.2678, simple_loss=0.3317, pruned_loss=0.1019, over 28614.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3345, pruned_loss=0.09237, over 5683314.59 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3538, pruned_loss=0.1108, over 5700622.40 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3319, pruned_loss=0.09054, over 5680139.00 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:31:39,410 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1221864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:32:07,519 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.326e+02 1.214e+03 1.474e+03 2.071e+03 7.848e+03, threshold=2.948e+03, percent-clipped=12.0
+2023-03-14 04:32:08,509 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1221900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:32:12,147 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1221903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:32:13,496 INFO [train.py:968] (0/2) Epoch 27, batch 37050, giga_loss[loss=0.2733, simple_loss=0.3473, pruned_loss=0.09968, over 27940.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3327, pruned_loss=0.09182, over 5697291.58 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3541, pruned_loss=0.1109, over 5701377.41 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.33, pruned_loss=0.08996, over 5693809.28 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:32:27,482 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5334, 2.1818, 1.6391, 0.8064], device='cuda:0'), covar=tensor([0.6460, 0.2697, 0.4191, 0.7225], device='cuda:0'), in_proj_covar=tensor([0.1818, 0.1719, 0.1649, 0.1490], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 04:32:33,487 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1221932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:32:50,362 INFO [train.py:968] (0/2) Epoch 27, batch 37100, libri_loss[loss=0.2807, simple_loss=0.3561, pruned_loss=0.1027, over 29525.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3302, pruned_loss=0.09051, over 5702531.20 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3544, pruned_loss=0.1108, over 5705439.36 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3271, pruned_loss=0.08858, over 5695891.99 frames. ], batch size: 82, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:33:11,021 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6104, 1.7755, 1.4688, 1.8044], device='cuda:0'), covar=tensor([0.2893, 0.3047, 0.3238, 0.2619], device='cuda:0'), in_proj_covar=tensor([0.1581, 0.1141, 0.1399, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 04:33:22,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.633e+02 1.065e+03 1.261e+03 1.571e+03 6.701e+03, threshold=2.523e+03, percent-clipped=4.0
+2023-03-14 04:33:24,192 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1222000.pt
+2023-03-14 04:33:27,894 INFO [train.py:968] (0/2) Epoch 27, batch 37150, giga_loss[loss=0.2486, simple_loss=0.3173, pruned_loss=0.08993, over 28862.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3291, pruned_loss=0.09043, over 5710797.11 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.356, pruned_loss=0.1119, over 5709730.44 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3245, pruned_loss=0.0873, over 5701566.14 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:34:07,678 INFO [train.py:968] (0/2) Epoch 27, batch 37200, giga_loss[loss=0.217, simple_loss=0.2945, pruned_loss=0.06981, over 28957.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3283, pruned_loss=0.09031, over 5715522.56 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3569, pruned_loss=0.1123, over 5711598.01 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3233, pruned_loss=0.08702, over 5706580.40 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:34:40,050 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-14 04:34:41,416 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-14 04:34:41,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.851e+02 1.133e+03 1.391e+03 2.004e+03 9.182e+03, threshold=2.783e+03, percent-clipped=16.0
+2023-03-14 04:34:46,522 INFO [train.py:968] (0/2) Epoch 27, batch 37250, giga_loss[loss=0.2205, simple_loss=0.2921, pruned_loss=0.0744, over 28818.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3255, pruned_loss=0.08894, over 5706591.54 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3574, pruned_loss=0.1126, over 5697001.11 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3205, pruned_loss=0.08566, over 5711143.28 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:35:23,315 INFO [train.py:968] (0/2) Epoch 27, batch 37300, libri_loss[loss=0.2958, simple_loss=0.3692, pruned_loss=0.1112, over 28589.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3265, pruned_loss=0.08919, over 5688830.99 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3589, pruned_loss=0.113, over 5677831.72 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3194, pruned_loss=0.08488, over 5711276.14 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:35:46,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4518, 1.5181, 1.1838, 1.5354], device='cuda:0'), covar=tensor([0.0714, 0.0336, 0.0365, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 04:35:56,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.996e+02 1.131e+03 1.416e+03 1.982e+03 8.636e+03, threshold=2.832e+03, percent-clipped=14.0
+2023-03-14 04:36:01,877 INFO [train.py:968] (0/2) Epoch 27, batch 37350, giga_loss[loss=0.2233, simple_loss=0.3038, pruned_loss=0.07138, over 28993.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3232, pruned_loss=0.08714, over 5692882.72 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3585, pruned_loss=0.1126, over 5674008.10 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3167, pruned_loss=0.08323, over 5714609.57 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:36:12,967 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1222217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:36:15,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3804, 1.5053, 1.5841, 1.2152], device='cuda:0'), covar=tensor([0.1777, 0.2517, 0.1452, 0.1733], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0713, 0.0981, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 04:36:31,828 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1222239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:36:42,391 INFO [train.py:968] (0/2) Epoch 27, batch 37400, giga_loss[loss=0.2198, simple_loss=0.2982, pruned_loss=0.07068, over 28730.00 frames. ], tot_loss[loss=0.246, simple_loss=0.321, pruned_loss=0.0855, over 5697345.28 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3588, pruned_loss=0.1126, over 5676129.88 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.3154, pruned_loss=0.08222, over 5712631.82 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:37:09,025 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1222286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:37:19,669 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.005e+02 1.117e+03 1.306e+03 1.806e+03 5.344e+03, threshold=2.612e+03, percent-clipped=9.0
+2023-03-14 04:37:24,973 INFO [train.py:968] (0/2) Epoch 27, batch 37450, giga_loss[loss=0.2436, simple_loss=0.3203, pruned_loss=0.08345, over 28969.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3195, pruned_loss=0.08483, over 5700756.11 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3587, pruned_loss=0.1125, over 5677173.84 frames. ], giga_tot_loss[loss=0.2395, simple_loss=0.3148, pruned_loss=0.08208, over 5712018.14 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:37:37,923 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 04:38:05,942 INFO [train.py:968] (0/2) Epoch 27, batch 37500, giga_loss[loss=0.2775, simple_loss=0.3525, pruned_loss=0.1013, over 28601.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3233, pruned_loss=0.08687, over 5707682.98 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3592, pruned_loss=0.1126, over 5680324.30 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3187, pruned_loss=0.08427, over 5714196.02 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:38:28,584 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1222382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:38:31,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1222385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:38:42,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.411e+02 1.334e+03 1.646e+03 2.188e+03 9.262e+03, threshold=3.292e+03, percent-clipped=16.0
+2023-03-14 04:38:48,322 INFO [train.py:968] (0/2) Epoch 27, batch 37550, giga_loss[loss=0.4029, simple_loss=0.4504, pruned_loss=0.1777, over 27536.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3309, pruned_loss=0.09181, over 5705206.06 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3599, pruned_loss=0.1127, over 5684656.88 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3256, pruned_loss=0.08898, over 5707059.34 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:38:57,032 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1222414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:39:24,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3485, 3.5399, 1.5229, 1.5914], device='cuda:0'), covar=tensor([0.1039, 0.0317, 0.0902, 0.1325], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0563, 0.0405, 0.0442], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 04:39:34,366 INFO [train.py:968] (0/2) Epoch 27, batch 37600, giga_loss[loss=0.2814, simple_loss=0.3588, pruned_loss=0.102, over 28906.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3391, pruned_loss=0.09751, over 5696506.26 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3598, pruned_loss=0.1127, over 5685275.15 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3345, pruned_loss=0.09498, over 5697593.78 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:39:57,492 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 04:40:17,198 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.217e+02 1.458e+03 1.789e+03 2.294e+03 6.526e+03, threshold=3.577e+03, percent-clipped=13.0
+2023-03-14 04:40:17,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4137, 1.5915, 1.6479, 1.2500], device='cuda:0'), covar=tensor([0.1820, 0.2611, 0.1532, 0.1789], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0712, 0.0979, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 04:40:21,917 INFO [train.py:968] (0/2) Epoch 27, batch 37650, giga_loss[loss=0.2824, simple_loss=0.3634, pruned_loss=0.1007, over 29065.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3437, pruned_loss=0.09958, over 5685698.79 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3604, pruned_loss=0.113, over 5686031.14 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3393, pruned_loss=0.09698, over 5685681.84 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:40:22,378 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5414, 1.7273, 1.7843, 1.3533], device='cuda:0'), covar=tensor([0.1833, 0.2712, 0.1521, 0.1828], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0712, 0.0979, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 04:41:01,221 INFO [train.py:968] (0/2) Epoch 27, batch 37700, giga_loss[loss=0.2802, simple_loss=0.3599, pruned_loss=0.1003, over 28869.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3489, pruned_loss=0.1018, over 5691089.41 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3612, pruned_loss=0.1136, over 5687416.02 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3441, pruned_loss=0.09875, over 5689662.68 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:41:17,601 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1222572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:41:34,968 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1222592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:41:43,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.147e+02 1.270e+03 1.609e+03 2.417e+03 8.894e+03, threshold=3.218e+03, percent-clipped=10.0
+2023-03-14 04:41:47,428 INFO [train.py:968] (0/2) Epoch 27, batch 37750, giga_loss[loss=0.2816, simple_loss=0.3562, pruned_loss=0.1036, over 28926.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3541, pruned_loss=0.1046, over 5684305.67 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.361, pruned_loss=0.1135, over 5691576.22 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3503, pruned_loss=0.1022, over 5679643.30 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:42:28,289 INFO [train.py:968] (0/2) Epoch 27, batch 37800, giga_loss[loss=0.2348, simple_loss=0.3144, pruned_loss=0.07759, over 28906.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3541, pruned_loss=0.1039, over 5686888.73 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3608, pruned_loss=0.1134, over 5692745.27 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3513, pruned_loss=0.102, over 5682148.86 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:42:34,565 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1222661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:42:54,774 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9339, 3.7613, 3.5859, 1.7458], device='cuda:0'), covar=tensor([0.0762, 0.0866, 0.0793, 0.2162], device='cuda:0'), in_proj_covar=tensor([0.1282, 0.1183, 0.0993, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 04:43:03,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.571e+02 1.349e+03 1.791e+03 2.554e+03 5.093e+03, threshold=3.581e+03, percent-clipped=13.0
+2023-03-14 04:43:06,649 INFO [train.py:968] (0/2) Epoch 27, batch 37850, libri_loss[loss=0.3435, simple_loss=0.402, pruned_loss=0.1425, over 28525.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3519, pruned_loss=0.1023, over 5688667.11 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3616, pruned_loss=0.1141, over 5688517.96 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3485, pruned_loss=0.09957, over 5688026.99 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:43:31,859 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1222735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:43:34,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1222738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:43:47,373 INFO [train.py:968] (0/2) Epoch 27, batch 37900, giga_loss[loss=0.2464, simple_loss=0.3291, pruned_loss=0.08183, over 28592.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.349, pruned_loss=0.09989, over 5682331.82 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3613, pruned_loss=0.1141, over 5682309.38 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3463, pruned_loss=0.09745, over 5688138.15 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:43:48,534 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1222756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:43:58,833 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1222767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:43:58,847 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1222767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:44:00,482 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8337, 1.1115, 2.8262, 2.7242], device='cuda:0'), covar=tensor([0.1798, 0.2776, 0.0615, 0.1386], device='cuda:0'), in_proj_covar=tensor([0.0800, 0.0670, 0.0997, 0.0968], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 04:44:20,546 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 04:44:26,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.126e+02 1.306e+03 1.655e+03 2.441e+03 8.006e+03, threshold=3.310e+03, percent-clipped=12.0
+2023-03-14 04:44:28,887 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1222804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:44:29,418 INFO [train.py:968] (0/2) Epoch 27, batch 37950, giga_loss[loss=0.2698, simple_loss=0.344, pruned_loss=0.09785, over 28662.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3487, pruned_loss=0.09943, over 5691801.05 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3617, pruned_loss=0.1144, over 5684536.96 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3458, pruned_loss=0.09683, over 5694364.12 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:44:31,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1222807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:44:53,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1222836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:45:06,675 INFO [train.py:968] (0/2) Epoch 27, batch 38000, libri_loss[loss=0.3001, simple_loss=0.3635, pruned_loss=0.1184, over 29530.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3504, pruned_loss=0.1003, over 5687717.75 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3622, pruned_loss=0.1148, over 5678954.02 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3472, pruned_loss=0.09737, over 5694478.62 frames. ], batch size: 80, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:45:19,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7567, 4.6047, 4.3393, 2.2135], device='cuda:0'), covar=tensor([0.0589, 0.0726, 0.0740, 0.1958], device='cuda:0'), in_proj_covar=tensor([0.1283, 0.1187, 0.0995, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 04:45:39,439 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2838, 3.0340, 1.5251, 1.4538], device='cuda:0'), covar=tensor([0.1080, 0.0447, 0.0840, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0563, 0.0405, 0.0440], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 04:45:46,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.458e+02 1.429e+03 1.769e+03 2.216e+03 5.753e+03, threshold=3.538e+03, percent-clipped=5.0
+2023-03-14 04:45:50,863 INFO [train.py:968] (0/2) Epoch 27, batch 38050, giga_loss[loss=0.2937, simple_loss=0.3656, pruned_loss=0.1109, over 28567.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3525, pruned_loss=0.1016, over 5686603.74 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3623, pruned_loss=0.1149, over 5670608.77 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3498, pruned_loss=0.09911, over 5699223.08 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:46:27,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1222947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:46:33,396 INFO [train.py:968] (0/2) Epoch 27, batch 38100, giga_loss[loss=0.2998, simple_loss=0.3654, pruned_loss=0.1172, over 28197.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3547, pruned_loss=0.1035, over 5683673.18 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3624, pruned_loss=0.1149, over 5668011.51 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3523, pruned_loss=0.1012, over 5696656.27 frames. ], batch size: 77, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:47:11,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.006e+02 1.392e+03 1.744e+03 2.147e+03 7.750e+03, threshold=3.487e+03, percent-clipped=6.0
+2023-03-14 04:47:15,429 INFO [train.py:968] (0/2) Epoch 27, batch 38150, giga_loss[loss=0.2656, simple_loss=0.347, pruned_loss=0.09206, over 29040.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3554, pruned_loss=0.1045, over 5685658.66 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3627, pruned_loss=0.1151, over 5670451.65 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3531, pruned_loss=0.1023, over 5694307.58 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:47:15,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4755, 1.0969, 4.5306, 3.4571], device='cuda:0'), covar=tensor([0.1662, 0.3034, 0.0418, 0.1069], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0671, 0.1001, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 04:47:58,669 INFO [train.py:968] (0/2) Epoch 27, batch 38200, giga_loss[loss=0.2484, simple_loss=0.3284, pruned_loss=0.08426, over 28414.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3548, pruned_loss=0.1041, over 5692263.50 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.115, over 5672897.63 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3531, pruned_loss=0.1024, over 5696918.97 frames. ], batch size: 65, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:48:15,342 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-14 04:48:22,874 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1223081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:48:29,756 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:48:31,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:48:38,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.472e+02 1.261e+03 1.560e+03 2.033e+03 6.007e+03, threshold=3.121e+03, percent-clipped=3.0
+2023-03-14 04:48:41,232 INFO [train.py:968] (0/2) Epoch 27, batch 38250, giga_loss[loss=0.2621, simple_loss=0.3439, pruned_loss=0.09013, over 29005.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.355, pruned_loss=0.104, over 5690316.05 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5674895.67 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3535, pruned_loss=0.1027, over 5692152.40 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:48:51,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6637, 1.9383, 1.2979, 1.5425], device='cuda:0'), covar=tensor([0.1083, 0.0691, 0.1092, 0.1247], device='cuda:0'), in_proj_covar=tensor([0.0407, 0.0444, 0.0521, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 04:48:54,280 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3443, 1.7286, 1.4506, 1.4278], device='cuda:0'), covar=tensor([0.0741, 0.0437, 0.0332, 0.0942], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 04:48:55,014 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1223122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:49:02,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1223131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:49:07,957 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1223139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:49:10,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1223142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:49:19,188 INFO [train.py:968] (0/2) Epoch 27, batch 38300, libri_loss[loss=0.2553, simple_loss=0.324, pruned_loss=0.09329, over 28068.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3554, pruned_loss=0.1034, over 5694750.54 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5671685.07 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3539, pruned_loss=0.1019, over 5699248.01 frames. ], batch size: 62, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:49:57,555 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.371e+02 1.396e+03 1.703e+03 2.356e+03 7.188e+03, threshold=3.406e+03, percent-clipped=8.0
+2023-03-14 04:50:00,274 INFO [train.py:968] (0/2) Epoch 27, batch 38350, giga_loss[loss=0.2473, simple_loss=0.329, pruned_loss=0.08276, over 28706.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3558, pruned_loss=0.1031, over 5702520.79 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3623, pruned_loss=0.1145, over 5677729.69 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3547, pruned_loss=0.1019, over 5701321.85 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:50:40,499 INFO [train.py:968] (0/2) Epoch 27, batch 38400, giga_loss[loss=0.2587, simple_loss=0.3412, pruned_loss=0.08811, over 28400.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3544, pruned_loss=0.1025, over 5702979.68 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3626, pruned_loss=0.1148, over 5682021.33 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.353, pruned_loss=0.101, over 5698492.25 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:50:56,556 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:50:58,620 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:51:04,751 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:51:07,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:51:13,922 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1223295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:51:17,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.710e+02 1.259e+03 1.534e+03 2.333e+03 6.825e+03, threshold=3.067e+03, percent-clipped=12.0
+2023-03-14 04:51:21,550 INFO [train.py:968] (0/2) Epoch 27, batch 38450, giga_loss[loss=0.2783, simple_loss=0.3547, pruned_loss=0.1009, over 28623.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3517, pruned_loss=0.1011, over 5705232.28 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.362, pruned_loss=0.1145, over 5686858.38 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3507, pruned_loss=0.0997, over 5697919.38 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:51:22,418 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1223306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:51:30,712 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1223317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:52:01,090 INFO [train.py:968] (0/2) Epoch 27, batch 38500, giga_loss[loss=0.2295, simple_loss=0.3104, pruned_loss=0.07427, over 28514.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3501, pruned_loss=0.1003, over 5708148.14 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1143, over 5689148.25 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3493, pruned_loss=0.09908, over 5700526.48 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:52:35,822 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.774e+02 1.139e+03 1.516e+03 2.033e+03 6.585e+03, threshold=3.032e+03, percent-clipped=9.0
+2023-03-14 04:52:38,520 INFO [train.py:968] (0/2) Epoch 27, batch 38550, giga_loss[loss=0.3002, simple_loss=0.37, pruned_loss=0.1152, over 28803.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3491, pruned_loss=0.09987, over 5714722.03 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3615, pruned_loss=0.114, over 5694876.09 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3484, pruned_loss=0.09872, over 5703995.77 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:52:49,666 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3530, 1.1742, 4.0606, 3.4253], device='cuda:0'), covar=tensor([0.1696, 0.2895, 0.0453, 0.0851], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0668, 0.0996, 0.0970], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 04:52:59,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5968, 4.3294, 1.7873, 1.7223], device='cuda:0'), covar=tensor([0.1020, 0.0244, 0.0902, 0.1334], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0561, 0.0404, 0.0439], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:0')
+2023-03-14 04:53:18,329 INFO [train.py:968] (0/2) Epoch 27, batch 38600, giga_loss[loss=0.2334, simple_loss=0.3126, pruned_loss=0.07708, over 28428.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3492, pruned_loss=0.09985, over 5712398.00 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3617, pruned_loss=0.1141, over 5694649.17 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3482, pruned_loss=0.09861, over 5704510.72 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:53:19,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1223456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:53:55,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.242e+02 1.172e+03 1.406e+03 1.906e+03 1.030e+04, threshold=2.813e+03, percent-clipped=7.0
+2023-03-14 04:53:56,790 INFO [train.py:968] (0/2) Epoch 27, batch 38650, giga_loss[loss=0.259, simple_loss=0.3331, pruned_loss=0.09246, over 28435.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.35, pruned_loss=0.1002, over 5706595.98 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3623, pruned_loss=0.1144, over 5690375.83 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3484, pruned_loss=0.09854, over 5703995.02 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:54:00,780 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.37 vs. limit=5.0
+2023-03-14 04:54:03,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1223514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:54:35,480 INFO [train.py:968] (0/2) Epoch 27, batch 38700, giga_loss[loss=0.2469, simple_loss=0.3282, pruned_loss=0.08275, over 28642.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.349, pruned_loss=0.09891, over 5712856.87 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.362, pruned_loss=0.1143, over 5695435.74 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3477, pruned_loss=0.09735, over 5706612.59 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:54:48,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7320, 1.9978, 2.0417, 1.7673], device='cuda:0'), covar=tensor([0.2925, 0.2218, 0.1717, 0.2137], device='cuda:0'), in_proj_covar=tensor([0.2048, 0.2002, 0.1915, 0.2060], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 04:55:09,227 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:55:12,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:55:12,425 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.431e+02 1.107e+03 1.315e+03 1.771e+03 6.146e+03, threshold=2.629e+03, percent-clipped=9.0
+2023-03-14 04:55:13,635 INFO [train.py:968] (0/2) Epoch 27, batch 38750, giga_loss[loss=0.295, simple_loss=0.3683, pruned_loss=0.1108, over 28573.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.348, pruned_loss=0.09813, over 5710372.31 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3614, pruned_loss=0.1139, over 5690877.67 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3472, pruned_loss=0.0969, over 5709483.32 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:55:34,115 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1223631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:55:52,297 INFO [train.py:968] (0/2) Epoch 27, batch 38800, giga_loss[loss=0.2569, simple_loss=0.3304, pruned_loss=0.09166, over 28921.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3471, pruned_loss=0.09806, over 5707319.84 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1143, over 5692433.48 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3458, pruned_loss=0.0965, over 5705317.84 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:55:56,816 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:55:58,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:56:06,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1223670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:56:21,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1223689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:56:32,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.799e+02 1.148e+03 1.358e+03 2.026e+03 7.207e+03, threshold=2.717e+03, percent-clipped=11.0
+2023-03-14 04:56:34,129 INFO [train.py:968] (0/2) Epoch 27, batch 38850, giga_loss[loss=0.2378, simple_loss=0.3094, pruned_loss=0.08309, over 28610.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3449, pruned_loss=0.0973, over 5704436.65 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3621, pruned_loss=0.1144, over 5695302.14 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3435, pruned_loss=0.09573, over 5700267.18 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:56:49,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7948, 1.9205, 2.0214, 1.5720], device='cuda:0'), covar=tensor([0.1996, 0.2810, 0.1600, 0.1933], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0717, 0.0982, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 04:57:11,839 INFO [train.py:968] (0/2) Epoch 27, batch 38900, giga_loss[loss=0.2478, simple_loss=0.3261, pruned_loss=0.08474, over 28980.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3419, pruned_loss=0.09583, over 5709805.80 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3626, pruned_loss=0.1147, over 5699970.04 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3399, pruned_loss=0.09388, over 5702517.23 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:57:49,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.437e+03 1.785e+03 2.807e+03 1.162e+04, threshold=3.571e+03, percent-clipped=26.0
+2023-03-14 04:57:50,831 INFO [train.py:968] (0/2) Epoch 27, batch 38950, giga_loss[loss=0.261, simple_loss=0.3317, pruned_loss=0.09509, over 28740.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3406, pruned_loss=0.09508, over 5705454.59 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3628, pruned_loss=0.1148, over 5696106.20 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3383, pruned_loss=0.09298, over 5702573.10 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:57:59,786 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:58:02,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:58:26,952 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1223845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:58:26,967 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1223845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:58:33,324 INFO [train.py:968] (0/2) Epoch 27, batch 39000, giga_loss[loss=0.2659, simple_loss=0.3401, pruned_loss=0.09587, over 28904.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3393, pruned_loss=0.09455, over 5707203.85 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3628, pruned_loss=0.1148, over 5692082.47 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3368, pruned_loss=0.09237, over 5708634.67 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:58:33,328 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 04:58:41,968 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2028, simple_loss=0.3112, pruned_loss=0.04724, over 944034.00 frames. 
+2023-03-14 04:58:41,969 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 04:58:48,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6507, 1.9095, 1.6990, 1.6217], device='cuda:0'), covar=tensor([0.2180, 0.2322, 0.2594, 0.2479], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0756, 0.0725, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 04:58:54,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1720, 1.3976, 1.3657, 1.1387], device='cuda:0'), covar=tensor([0.3298, 0.2942, 0.1936, 0.2661], device='cuda:0'), in_proj_covar=tensor([0.2038, 0.1991, 0.1903, 0.2051], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 04:59:18,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.406e+02 1.198e+03 1.555e+03 2.065e+03 5.640e+03, threshold=3.109e+03, percent-clipped=4.0
+2023-03-14 04:59:19,071 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 04:59:19,300 INFO [train.py:968] (0/2) Epoch 27, batch 39050, giga_loss[loss=0.2607, simple_loss=0.3359, pruned_loss=0.09277, over 27939.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.338, pruned_loss=0.09444, over 5708851.95 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3629, pruned_loss=0.1149, over 5692211.98 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3352, pruned_loss=0.092, over 5710598.50 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:59:31,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1529, 1.2859, 1.1173, 1.1535], device='cuda:0'), covar=tensor([0.1931, 0.1974, 0.1602, 0.1678], device='cuda:0'), in_proj_covar=tensor([0.2040, 0.1990, 0.1904, 0.2051], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 04:59:57,717 INFO [train.py:968] (0/2) Epoch 27, batch 39100, giga_loss[loss=0.2099, simple_loss=0.2942, pruned_loss=0.0628, over 28574.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3361, pruned_loss=0.0939, over 5708582.22 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3629, pruned_loss=0.1147, over 5695677.14 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3333, pruned_loss=0.09162, over 5707306.66 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:00:08,469 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1809, 2.2878, 1.7748, 1.8304], device='cuda:0'), covar=tensor([0.0935, 0.0741, 0.1074, 0.1211], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0445, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 05:00:32,947 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1224000.pt
+2023-03-14 05:00:35,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.075e+02 1.174e+03 1.492e+03 1.835e+03 6.192e+03, threshold=2.984e+03, percent-clipped=3.0
+2023-03-14 05:00:36,536 INFO [train.py:968] (0/2) Epoch 27, batch 39150, giga_loss[loss=0.3009, simple_loss=0.3588, pruned_loss=0.1215, over 28657.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3336, pruned_loss=0.09289, over 5713388.61 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3628, pruned_loss=0.1146, over 5701173.49 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3308, pruned_loss=0.09065, over 5707638.59 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:00:41,239 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.41 vs. limit=5.0
+2023-03-14 05:00:58,535 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1224031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:01:17,857 INFO [train.py:968] (0/2) Epoch 27, batch 39200, giga_loss[loss=0.2549, simple_loss=0.326, pruned_loss=0.09189, over 28595.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3322, pruned_loss=0.09235, over 5710944.04 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3622, pruned_loss=0.1142, over 5704621.76 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3298, pruned_loss=0.09045, over 5703549.26 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:01:42,143 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 05:02:00,409 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.210e+02 1.099e+03 1.332e+03 1.696e+03 3.921e+03, threshold=2.664e+03, percent-clipped=5.0
+2023-03-14 05:02:01,043 INFO [train.py:968] (0/2) Epoch 27, batch 39250, libri_loss[loss=0.3321, simple_loss=0.3943, pruned_loss=0.1349, over 29529.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3336, pruned_loss=0.09284, over 5705234.32 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3622, pruned_loss=0.1142, over 5699307.69 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3312, pruned_loss=0.09086, over 5704756.46 frames. ], batch size: 83, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:02:34,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2665, 1.3892, 1.4606, 1.3024], device='cuda:0'), covar=tensor([0.2757, 0.2593, 0.2010, 0.2309], device='cuda:0'), in_proj_covar=tensor([0.2052, 0.2003, 0.1920, 0.2064], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 05:02:35,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1096, 1.4662, 1.4860, 1.2999], device='cuda:0'), covar=tensor([0.2197, 0.1680, 0.2425, 0.1996], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0752, 0.0723, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 05:02:42,903 INFO [train.py:968] (0/2) Epoch 27, batch 39300, giga_loss[loss=0.235, simple_loss=0.3179, pruned_loss=0.07606, over 28729.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3371, pruned_loss=0.09386, over 5709935.30 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3626, pruned_loss=0.1145, over 5700801.84 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3342, pruned_loss=0.0916, over 5708563.50 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:02:51,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3406, 1.1850, 1.0851, 1.4211], device='cuda:0'), covar=tensor([0.0794, 0.0361, 0.0396, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 05:03:24,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.324e+02 1.209e+03 1.548e+03 2.309e+03 7.846e+03, threshold=3.097e+03, percent-clipped=21.0
+2023-03-14 05:03:25,112 INFO [train.py:968] (0/2) Epoch 27, batch 39350, giga_loss[loss=0.3211, simple_loss=0.371, pruned_loss=0.1356, over 23777.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3403, pruned_loss=0.09522, over 5695272.65 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3629, pruned_loss=0.1147, over 5694882.50 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.337, pruned_loss=0.09264, over 5699138.87 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:03:36,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:03:54,250 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1224240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:04:05,699 INFO [train.py:968] (0/2) Epoch 27, batch 39400, giga_loss[loss=0.2457, simple_loss=0.3305, pruned_loss=0.08046, over 28931.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3423, pruned_loss=0.09575, over 5698602.45 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3624, pruned_loss=0.1146, over 5699525.11 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3393, pruned_loss=0.09312, over 5697270.94 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:04:15,902 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2510, 3.4814, 1.4351, 1.4979], device='cuda:0'), covar=tensor([0.1028, 0.0283, 0.0984, 0.1326], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0562, 0.0405, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 05:04:27,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4070, 1.6277, 1.6844, 1.2354], device='cuda:0'), covar=tensor([0.1962, 0.2643, 0.1674, 0.1867], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0711, 0.0976, 0.0876], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 05:04:48,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.218e+02 1.168e+03 1.478e+03 1.941e+03 6.445e+03, threshold=2.956e+03, percent-clipped=6.0
+2023-03-14 05:04:48,879 INFO [train.py:968] (0/2) Epoch 27, batch 39450, giga_loss[loss=0.2211, simple_loss=0.3087, pruned_loss=0.06677, over 29068.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3415, pruned_loss=0.09492, over 5687431.00 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3627, pruned_loss=0.115, over 5693342.49 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3385, pruned_loss=0.09215, over 5691453.40 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:05:07,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1569, 1.7531, 1.2879, 0.4022], device='cuda:0'), covar=tensor([0.5512, 0.3091, 0.4739, 0.7454], device='cuda:0'), in_proj_covar=tensor([0.1823, 0.1712, 0.1647, 0.1494], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 05:05:15,859 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-14 05:05:28,415 INFO [train.py:968] (0/2) Epoch 27, batch 39500, giga_loss[loss=0.2612, simple_loss=0.3314, pruned_loss=0.09549, over 28830.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.341, pruned_loss=0.09415, over 5683456.71 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3631, pruned_loss=0.1151, over 5683816.57 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3378, pruned_loss=0.09136, over 5695702.59 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:05:36,362 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1224363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:05:38,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1224366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:06:03,503 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1224395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:06:10,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.221e+02 1.362e+03 1.787e+03 2.392e+03 9.188e+03, threshold=3.573e+03, percent-clipped=17.0
+2023-03-14 05:06:10,195 INFO [train.py:968] (0/2) Epoch 27, batch 39550, giga_loss[loss=0.3124, simple_loss=0.3673, pruned_loss=0.1287, over 28524.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3429, pruned_loss=0.09586, over 5686373.96 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3631, pruned_loss=0.1151, over 5689224.22 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3396, pruned_loss=0.09294, over 5691213.14 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:06:11,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:06:23,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7841, 1.9742, 2.0373, 1.6233], device='cuda:0'), covar=tensor([0.3659, 0.2850, 0.2922, 0.3581], device='cuda:0'), in_proj_covar=tensor([0.2054, 0.2001, 0.1920, 0.2061], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 05:06:53,875 INFO [train.py:968] (0/2) Epoch 27, batch 39600, giga_loss[loss=0.275, simple_loss=0.3462, pruned_loss=0.1018, over 28925.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3439, pruned_loss=0.09646, over 5684908.86 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3632, pruned_loss=0.1152, over 5688741.99 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3411, pruned_loss=0.09398, over 5689040.31 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:07:36,408 INFO [train.py:968] (0/2) Epoch 27, batch 39650, libri_loss[loss=0.2564, simple_loss=0.3248, pruned_loss=0.09402, over 29642.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3469, pruned_loss=0.09827, over 5682326.44 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3634, pruned_loss=0.1153, over 5674387.29 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3441, pruned_loss=0.09578, over 5697780.35 frames. ], batch size: 69, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:07:37,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.513e+02 1.332e+03 1.651e+03 2.126e+03 6.743e+03, threshold=3.302e+03, percent-clipped=7.0
+2023-03-14 05:07:54,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5523, 1.9420, 1.9656, 1.6935], device='cuda:0'), covar=tensor([0.1861, 0.1731, 0.1790, 0.1834], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0759, 0.0728, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 05:07:54,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-14 05:08:13,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1224549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:08:15,786 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1224552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:08:17,589 INFO [train.py:968] (0/2) Epoch 27, batch 39700, giga_loss[loss=0.294, simple_loss=0.3767, pruned_loss=0.1057, over 28962.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3503, pruned_loss=0.09971, over 5694341.84 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3636, pruned_loss=0.1154, over 5676709.68 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3477, pruned_loss=0.0975, over 5704453.13 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:08:38,061 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1224581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:08:56,309 INFO [train.py:968] (0/2) Epoch 27, batch 39750, libri_loss[loss=0.3064, simple_loss=0.3653, pruned_loss=0.1237, over 29559.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3512, pruned_loss=0.09995, over 5702707.54 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3635, pruned_loss=0.1152, over 5681570.82 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3489, pruned_loss=0.09785, over 5706974.03 frames. ], batch size: 77, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:08:56,885 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.264e+02 1.307e+03 1.533e+03 2.084e+03 4.826e+03, threshold=3.066e+03, percent-clipped=4.0
+2023-03-14 05:09:06,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:09:07,639 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1224617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:09:38,388 INFO [train.py:968] (0/2) Epoch 27, batch 39800, giga_loss[loss=0.2819, simple_loss=0.3522, pruned_loss=0.1058, over 28925.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3523, pruned_loss=0.1006, over 5706745.32 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3635, pruned_loss=0.1151, over 5684771.92 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3502, pruned_loss=0.09878, over 5707612.80 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:10:16,142 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3229, 1.9555, 1.4118, 0.6216], device='cuda:0'), covar=tensor([0.6688, 0.3519, 0.5131, 0.7692], device='cuda:0'), in_proj_covar=tensor([0.1831, 0.1721, 0.1651, 0.1496], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 05:10:17,958 INFO [train.py:968] (0/2) Epoch 27, batch 39850, giga_loss[loss=0.3172, simple_loss=0.3743, pruned_loss=0.13, over 28779.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3511, pruned_loss=0.09988, over 5707580.55 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3637, pruned_loss=0.1153, over 5677608.84 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3492, pruned_loss=0.09824, over 5714220.29 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:10:18,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.564e+02 1.345e+03 1.698e+03 2.116e+03 1.101e+04, threshold=3.395e+03, percent-clipped=9.0
+2023-03-14 05:10:28,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-14 05:10:44,203 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-14 05:10:56,375 INFO [train.py:968] (0/2) Epoch 27, batch 39900, giga_loss[loss=0.2888, simple_loss=0.3644, pruned_loss=0.1066, over 28617.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3491, pruned_loss=0.09888, over 5706465.25 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3637, pruned_loss=0.1153, over 5677752.59 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3473, pruned_loss=0.09726, over 5712471.70 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:10:59,488 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1224758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:11:01,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1224761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:11:15,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3445, 1.5078, 1.4382, 1.2580], device='cuda:0'), covar=tensor([0.3442, 0.2871, 0.2523, 0.3081], device='cuda:0'), in_proj_covar=tensor([0.2060, 0.2014, 0.1926, 0.2065], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 05:11:25,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1224790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:11:37,089 INFO [train.py:968] (0/2) Epoch 27, batch 39950, giga_loss[loss=0.2574, simple_loss=0.3375, pruned_loss=0.08869, over 28946.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3462, pruned_loss=0.09772, over 5700811.20 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3636, pruned_loss=0.1151, over 5675100.84 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3445, pruned_loss=0.09613, over 5708983.38 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:11:37,769 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.568e+02 1.265e+03 1.585e+03 2.280e+03 7.843e+03, threshold=3.170e+03, percent-clipped=8.0
+2023-03-14 05:11:46,499 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1224817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:11:57,158 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1224829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:12:18,700 INFO [train.py:968] (0/2) Epoch 27, batch 40000, giga_loss[loss=0.2796, simple_loss=0.3557, pruned_loss=0.1017, over 28999.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3438, pruned_loss=0.09666, over 5707391.31 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3642, pruned_loss=0.1155, over 5678619.34 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3417, pruned_loss=0.09479, over 5711168.74 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:12:55,662 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-14 05:13:00,998 INFO [train.py:968] (0/2) Epoch 27, batch 40050, giga_loss[loss=0.2693, simple_loss=0.3491, pruned_loss=0.09473, over 29088.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3437, pruned_loss=0.0958, over 5709297.78 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3639, pruned_loss=0.1153, over 5680515.85 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3421, pruned_loss=0.09436, over 5710987.67 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:13:01,344 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-14 05:13:01,496 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.439e+02 1.212e+03 1.576e+03 2.101e+03 7.582e+03, threshold=3.152e+03, percent-clipped=8.0
+2023-03-14 05:13:42,328 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0621, 1.4531, 1.4826, 1.2703], device='cuda:0'), covar=tensor([0.2197, 0.1758, 0.2392, 0.2032], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0759, 0.0728, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 05:13:42,669 INFO [train.py:968] (0/2) Epoch 27, batch 40100, giga_loss[loss=0.2878, simple_loss=0.3703, pruned_loss=0.1027, over 28705.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3445, pruned_loss=0.09517, over 5704826.37 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3639, pruned_loss=0.1154, over 5682948.29 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3431, pruned_loss=0.09383, over 5704432.09 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:14:09,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:14:19,383 INFO [train.py:968] (0/2) Epoch 27, batch 40150, giga_loss[loss=0.2969, simple_loss=0.3665, pruned_loss=0.1137, over 28704.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3464, pruned_loss=0.09679, over 5710464.41 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3646, pruned_loss=0.1157, over 5690367.33 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3436, pruned_loss=0.09439, over 5704576.95 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:14:20,763 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.211e+02 1.409e+03 1.878e+03 2.636e+03 1.189e+04, threshold=3.756e+03, percent-clipped=17.0
+2023-03-14 05:14:57,519 INFO [train.py:968] (0/2) Epoch 27, batch 40200, giga_loss[loss=0.2784, simple_loss=0.3533, pruned_loss=0.1017, over 27955.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3452, pruned_loss=0.0972, over 5706386.52 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3651, pruned_loss=0.116, over 5684325.84 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.342, pruned_loss=0.09444, over 5706831.77 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:15:09,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4782, 2.1876, 1.6859, 0.7041], device='cuda:0'), covar=tensor([0.7156, 0.3203, 0.4305, 0.8314], device='cuda:0'), in_proj_covar=tensor([0.1823, 0.1712, 0.1646, 0.1492], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 05:15:12,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4913, 2.0664, 1.7889, 1.7095], device='cuda:0'), covar=tensor([0.0753, 0.0262, 0.0298, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 05:15:36,432 INFO [train.py:968] (0/2) Epoch 27, batch 40250, giga_loss[loss=0.2767, simple_loss=0.3487, pruned_loss=0.1024, over 28740.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3436, pruned_loss=0.09735, over 5707864.15 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3656, pruned_loss=0.1162, over 5677780.80 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.34, pruned_loss=0.09446, over 5715392.64 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:15:37,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.935e+02 1.332e+03 1.771e+03 2.222e+03 4.924e+03, threshold=3.542e+03, percent-clipped=5.0
+2023-03-14 05:16:02,010 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1225135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:16:04,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1225138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:16:11,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4077, 1.8041, 1.5113, 1.5421], device='cuda:0'), covar=tensor([0.0620, 0.0270, 0.0301, 0.0710], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 05:16:18,285 INFO [train.py:968] (0/2) Epoch 27, batch 40300, giga_loss[loss=0.2166, simple_loss=0.2985, pruned_loss=0.06737, over 29023.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3408, pruned_loss=0.09711, over 5702453.95 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3658, pruned_loss=0.1163, over 5678278.45 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3374, pruned_loss=0.09439, over 5708402.75 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:16:27,734 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1225167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:16:36,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4500, 2.2360, 1.7005, 0.7425], device='cuda:0'), covar=tensor([0.6870, 0.3060, 0.4858, 0.7879], device='cuda:0'), in_proj_covar=tensor([0.1825, 0.1713, 0.1650, 0.1494], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 05:16:48,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1225192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:16:53,128 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4188, 1.7449, 1.3949, 1.3565], device='cuda:0'), covar=tensor([0.2681, 0.2695, 0.3162, 0.2437], device='cuda:0'), in_proj_covar=tensor([0.1586, 0.1142, 0.1402, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 05:16:55,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1225204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:16:56,401 INFO [train.py:968] (0/2) Epoch 27, batch 40350, giga_loss[loss=0.26, simple_loss=0.3294, pruned_loss=0.09523, over 28948.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3403, pruned_loss=0.09758, over 5705846.83 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3656, pruned_loss=0.1162, over 5683796.62 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3373, pruned_loss=0.09513, over 5706153.44 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:16:57,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.756e+02 1.274e+03 1.607e+03 2.130e+03 9.663e+03, threshold=3.214e+03, percent-clipped=6.0
+2023-03-14 05:17:30,999 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1225248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:17:36,507 INFO [train.py:968] (0/2) Epoch 27, batch 40400, giga_loss[loss=0.2752, simple_loss=0.3462, pruned_loss=0.1021, over 28715.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.338, pruned_loss=0.09634, over 5696771.00 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3656, pruned_loss=0.1161, over 5679641.24 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3348, pruned_loss=0.09388, over 5701483.25 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:18:15,451 INFO [train.py:968] (0/2) Epoch 27, batch 40450, giga_loss[loss=0.2532, simple_loss=0.3206, pruned_loss=0.09293, over 28884.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3353, pruned_loss=0.09514, over 5706236.64 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3657, pruned_loss=0.1161, over 5684785.24 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3319, pruned_loss=0.09266, over 5705860.45 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:18:15,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3429, 1.2163, 3.6627, 3.1435], device='cuda:0'), covar=tensor([0.1632, 0.2951, 0.0463, 0.1303], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0669, 0.0996, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 05:18:17,104 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.950e+02 1.303e+03 1.598e+03 1.999e+03 8.995e+03, threshold=3.197e+03, percent-clipped=8.0
+2023-03-14 05:18:38,582 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1225335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:18:40,491 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1225338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:18:48,182 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1225347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:18:50,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1225350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:18:55,523 INFO [train.py:968] (0/2) Epoch 27, batch 40500, giga_loss[loss=0.2237, simple_loss=0.3083, pruned_loss=0.06951, over 28805.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3314, pruned_loss=0.09343, over 5711244.88 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3658, pruned_loss=0.1162, over 5692548.83 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3276, pruned_loss=0.09067, over 5704707.11 frames. ], batch size: 285, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:18:59,481 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4337, 1.9499, 1.5628, 1.6816], device='cuda:0'), covar=tensor([0.0782, 0.0274, 0.0340, 0.0899], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 05:19:04,397 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1225367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:19:05,224 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-14 05:19:13,593 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1225379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:19:32,929 INFO [train.py:968] (0/2) Epoch 27, batch 40550, giga_loss[loss=0.2634, simple_loss=0.336, pruned_loss=0.09541, over 28590.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3298, pruned_loss=0.09192, over 5720780.13 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3659, pruned_loss=0.1163, over 5696471.98 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.326, pruned_loss=0.08931, over 5712611.63 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:19:35,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.791e+02 1.281e+03 1.751e+03 2.257e+03 6.062e+03, threshold=3.502e+03, percent-clipped=6.0
+2023-03-14 05:20:15,924 INFO [train.py:968] (0/2) Epoch 27, batch 40600, libri_loss[loss=0.2433, simple_loss=0.3121, pruned_loss=0.08727, over 28679.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3339, pruned_loss=0.09393, over 5714469.77 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3659, pruned_loss=0.1162, over 5698889.84 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3302, pruned_loss=0.0914, over 5706166.05 frames. ], batch size: 63, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:20:24,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6060, 1.8205, 1.5130, 1.7369], device='cuda:0'), covar=tensor([0.2642, 0.2773, 0.3136, 0.2411], device='cuda:0'), in_proj_covar=tensor([0.1589, 0.1144, 0.1403, 0.1010], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 05:20:58,493 INFO [train.py:968] (0/2) Epoch 27, batch 40650, giga_loss[loss=0.2805, simple_loss=0.3464, pruned_loss=0.1073, over 28756.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3369, pruned_loss=0.09514, over 5719484.82 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3658, pruned_loss=0.1161, over 5704670.19 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3331, pruned_loss=0.09259, over 5708278.66 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:21:00,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.850e+02 1.306e+03 1.613e+03 2.087e+03 6.766e+03, threshold=3.226e+03, percent-clipped=5.0
+2023-03-14 05:21:07,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8346, 5.6532, 5.3558, 2.9577], device='cuda:0'), covar=tensor([0.0430, 0.0577, 0.0629, 0.1522], device='cuda:0'), in_proj_covar=tensor([0.1284, 0.1186, 0.0996, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 05:21:35,842 INFO [train.py:968] (0/2) Epoch 27, batch 40700, giga_loss[loss=0.2924, simple_loss=0.3666, pruned_loss=0.1091, over 28658.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3397, pruned_loss=0.09604, over 5722302.26 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3655, pruned_loss=0.1158, over 5710356.84 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3362, pruned_loss=0.09365, over 5708371.72 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:21:37,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3597, 1.6258, 1.6494, 1.3998], device='cuda:0'), covar=tensor([0.2100, 0.1842, 0.1450, 0.1689], device='cuda:0'), in_proj_covar=tensor([0.2052, 0.2009, 0.1921, 0.2052], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 05:21:46,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7982, 1.3079, 4.8060, 3.5146], device='cuda:0'), covar=tensor([0.1608, 0.3039, 0.0408, 0.0948], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0670, 0.1000, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 05:22:17,007 INFO [train.py:968] (0/2) Epoch 27, batch 40750, giga_loss[loss=0.2647, simple_loss=0.343, pruned_loss=0.09322, over 28982.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3437, pruned_loss=0.09789, over 5713988.40 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3658, pruned_loss=0.1161, over 5712565.45 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3397, pruned_loss=0.0951, over 5700611.20 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:22:18,260 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.118e+02 1.283e+03 1.803e+03 2.638e+03 1.040e+04, threshold=3.606e+03, percent-clipped=16.0
+2023-03-14 05:22:31,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1225623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:22:54,155 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9308, 1.1201, 1.0713, 0.9071], device='cuda:0'), covar=tensor([0.2244, 0.2548, 0.1652, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.2053, 0.2009, 0.1922, 0.2051], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 05:22:56,807 INFO [train.py:968] (0/2) Epoch 27, batch 40800, giga_loss[loss=0.287, simple_loss=0.3615, pruned_loss=0.1063, over 28737.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3454, pruned_loss=0.09883, over 5712703.55 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3657, pruned_loss=0.116, over 5707860.83 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3417, pruned_loss=0.09609, over 5705496.98 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 05:23:09,795 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-14 05:23:44,488 INFO [train.py:968] (0/2) Epoch 27, batch 40850, giga_loss[loss=0.2606, simple_loss=0.3391, pruned_loss=0.0911, over 29025.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.351, pruned_loss=0.1037, over 5705410.37 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3659, pruned_loss=0.1161, over 5709533.43 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3477, pruned_loss=0.1012, over 5698290.15 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:23:49,050 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.406e+03 1.744e+03 2.314e+03 8.960e+03, threshold=3.489e+03, percent-clipped=6.0
+2023-03-14 05:24:15,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4277, 1.5458, 1.2877, 1.1586], device='cuda:0'), covar=tensor([0.1082, 0.0614, 0.1077, 0.1141], device='cuda:0'), in_proj_covar=tensor([0.0408, 0.0448, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 05:24:26,677 INFO [train.py:968] (0/2) Epoch 27, batch 40900, giga_loss[loss=0.2724, simple_loss=0.3527, pruned_loss=0.096, over 28972.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3558, pruned_loss=0.1076, over 5703599.23 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3659, pruned_loss=0.1163, over 5708554.05 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3524, pruned_loss=0.1049, over 5698536.37 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:24:37,634 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1225766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:24:39,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1225769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:24:50,821 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 05:25:06,360 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1225798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:25:11,527 INFO [train.py:968] (0/2) Epoch 27, batch 40950, giga_loss[loss=0.3206, simple_loss=0.3967, pruned_loss=0.1223, over 28880.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3631, pruned_loss=0.1126, over 5687036.38 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3658, pruned_loss=0.1162, over 5698752.48 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3604, pruned_loss=0.1104, over 5692119.82 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:25:14,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.109e+03 1.849e+03 2.619e+03 3.464e+03 9.310e+03, threshold=5.237e+03, percent-clipped=24.0
+2023-03-14 05:25:15,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3240, 1.4454, 1.2618, 1.4959], device='cuda:0'), covar=tensor([0.0802, 0.0361, 0.0356, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 05:25:56,684 INFO [train.py:968] (0/2) Epoch 27, batch 41000, giga_loss[loss=0.3014, simple_loss=0.3742, pruned_loss=0.1143, over 28864.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3692, pruned_loss=0.1175, over 5692489.92 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3657, pruned_loss=0.1161, over 5701076.67 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3671, pruned_loss=0.1158, over 5694349.13 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:26:11,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6497, 1.7459, 1.3634, 1.3501], device='cuda:0'), covar=tensor([0.0983, 0.0610, 0.0997, 0.1122], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0448, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 05:26:22,108 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4218, 1.6652, 1.6156, 1.4122], device='cuda:0'), covar=tensor([0.2415, 0.2025, 0.1704, 0.2079], device='cuda:0'), in_proj_covar=tensor([0.2053, 0.2011, 0.1924, 0.2049], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 05:26:42,335 INFO [train.py:968] (0/2) Epoch 27, batch 41050, giga_loss[loss=0.3572, simple_loss=0.4129, pruned_loss=0.1507, over 28939.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3763, pruned_loss=0.1237, over 5688530.69 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3656, pruned_loss=0.116, over 5699469.73 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3748, pruned_loss=0.1225, over 5691419.83 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:26:44,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.197e+03 1.838e+03 2.234e+03 2.966e+03 9.171e+03, threshold=4.467e+03, percent-clipped=3.0
+2023-03-14 05:26:50,995 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1225916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:27:32,319 INFO [train.py:968] (0/2) Epoch 27, batch 41100, giga_loss[loss=0.3107, simple_loss=0.3801, pruned_loss=0.1207, over 28927.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3808, pruned_loss=0.1281, over 5655590.76 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.366, pruned_loss=0.1163, over 5685524.92 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3796, pruned_loss=0.127, over 5669114.64 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:27:40,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3870, 1.5375, 1.2095, 1.1132], device='cuda:0'), covar=tensor([0.1056, 0.0592, 0.1068, 0.1143], device='cuda:0'), in_proj_covar=tensor([0.0409, 0.0448, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 05:28:12,101 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1226000.pt
+2023-03-14 05:28:19,119 INFO [train.py:968] (0/2) Epoch 27, batch 41150, giga_loss[loss=0.3942, simple_loss=0.4164, pruned_loss=0.1859, over 23452.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3811, pruned_loss=0.1292, over 5644673.15 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3652, pruned_loss=0.1159, over 5681698.93 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3812, pruned_loss=0.1292, over 5657367.32 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:28:20,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.253e+03 1.957e+03 2.385e+03 2.960e+03 7.229e+03, threshold=4.771e+03, percent-clipped=7.0
+2023-03-14 05:29:13,929 INFO [train.py:968] (0/2) Epoch 27, batch 41200, giga_loss[loss=0.2972, simple_loss=0.3689, pruned_loss=0.1127, over 28868.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.384, pruned_loss=0.1327, over 5647389.30 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3651, pruned_loss=0.1158, over 5684701.03 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3845, pruned_loss=0.1329, over 5654268.19 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 05:29:54,590 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:29:55,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3637, 1.6154, 1.5958, 1.1763], device='cuda:0'), covar=tensor([0.1663, 0.2801, 0.1499, 0.1780], device='cuda:0'), in_proj_covar=tensor([0.0929, 0.0713, 0.0975, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 05:30:03,606 INFO [train.py:968] (0/2) Epoch 27, batch 41250, giga_loss[loss=0.3157, simple_loss=0.3737, pruned_loss=0.1289, over 28547.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3868, pruned_loss=0.1362, over 5636354.47 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3648, pruned_loss=0.1156, over 5689024.24 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3882, pruned_loss=0.1373, over 5636293.94 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:30:07,601 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.911e+03 2.522e+03 3.199e+03 8.236e+03, threshold=5.043e+03, percent-clipped=6.0
+2023-03-14 05:30:53,357 INFO [train.py:968] (0/2) Epoch 27, batch 41300, giga_loss[loss=0.4001, simple_loss=0.4181, pruned_loss=0.191, over 23624.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3892, pruned_loss=0.1381, over 5634514.86 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3643, pruned_loss=0.1154, over 5695101.03 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3914, pruned_loss=0.1397, over 5627739.22 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:31:44,225 INFO [train.py:968] (0/2) Epoch 27, batch 41350, libri_loss[loss=0.3241, simple_loss=0.3899, pruned_loss=0.1292, over 29702.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3903, pruned_loss=0.1399, over 5620354.44 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3642, pruned_loss=0.1154, over 5690924.36 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3929, pruned_loss=0.142, over 5615810.95 frames. ], batch size: 91, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:31:47,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.419e+03 2.055e+03 2.730e+03 3.446e+03 8.547e+03, threshold=5.461e+03, percent-clipped=11.0
+2023-03-14 05:31:53,145 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 05:32:30,356 INFO [train.py:968] (0/2) Epoch 27, batch 41400, giga_loss[loss=0.3493, simple_loss=0.3921, pruned_loss=0.1533, over 28684.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3889, pruned_loss=0.1393, over 5635894.93 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.364, pruned_loss=0.1152, over 5691775.72 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3913, pruned_loss=0.1413, over 5631029.07 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:33:10,268 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1226291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:33:21,385 INFO [train.py:968] (0/2) Epoch 27, batch 41450, libri_loss[loss=0.3004, simple_loss=0.3682, pruned_loss=0.1164, over 29754.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3876, pruned_loss=0.1369, over 5652421.94 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3641, pruned_loss=0.1151, over 5694961.22 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3901, pruned_loss=0.1391, over 5644454.86 frames. ], batch size: 87, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:33:26,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.113e+03 1.836e+03 2.369e+03 3.547e+03 8.467e+03, threshold=4.737e+03, percent-clipped=3.0
+2023-03-14 05:33:48,273 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-14 05:34:13,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6820, 1.7901, 1.7747, 1.5739], device='cuda:0'), covar=tensor([0.2914, 0.2853, 0.2312, 0.2793], device='cuda:0'), in_proj_covar=tensor([0.2066, 0.2021, 0.1933, 0.2060], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 05:34:15,734 INFO [train.py:968] (0/2) Epoch 27, batch 41500, giga_loss[loss=0.2769, simple_loss=0.3496, pruned_loss=0.1021, over 28316.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3872, pruned_loss=0.1352, over 5657493.27 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3645, pruned_loss=0.1154, over 5695946.84 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3889, pruned_loss=0.1368, over 5649958.40 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:35:05,835 INFO [train.py:968] (0/2) Epoch 27, batch 41550, giga_loss[loss=0.3468, simple_loss=0.4048, pruned_loss=0.1444, over 28908.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3889, pruned_loss=0.1362, over 5646801.21 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3648, pruned_loss=0.1156, over 5688965.16 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3907, pruned_loss=0.1379, over 5645325.84 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:35:10,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.281e+03 1.906e+03 2.564e+03 3.478e+03 1.017e+04, threshold=5.129e+03, percent-clipped=14.0
+2023-03-14 05:35:37,994 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1226434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:35:41,011 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1226437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:35:56,299 INFO [train.py:968] (0/2) Epoch 27, batch 41600, giga_loss[loss=0.2904, simple_loss=0.3691, pruned_loss=0.1058, over 29019.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3872, pruned_loss=0.1346, over 5650399.11 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3644, pruned_loss=0.1155, over 5692121.31 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3896, pruned_loss=0.1365, over 5645193.22 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:36:08,928 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1226466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:36:11,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1226469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:36:15,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3716, 1.5714, 1.2328, 1.1632], device='cuda:0'), covar=tensor([0.1157, 0.0633, 0.1126, 0.1202], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0449, 0.0524, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 05:36:45,885 INFO [train.py:968] (0/2) Epoch 27, batch 41650, giga_loss[loss=0.3036, simple_loss=0.3729, pruned_loss=0.1172, over 28345.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3853, pruned_loss=0.1321, over 5643335.16 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3644, pruned_loss=0.1156, over 5692033.09 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3874, pruned_loss=0.1339, over 5638910.45 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:36:51,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.792e+03 2.086e+03 2.902e+03 7.124e+03, threshold=4.172e+03, percent-clipped=1.0
+2023-03-14 05:37:38,699 INFO [train.py:968] (0/2) Epoch 27, batch 41700, giga_loss[loss=0.2851, simple_loss=0.3556, pruned_loss=0.1073, over 28941.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3805, pruned_loss=0.1272, over 5661463.21 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3644, pruned_loss=0.1156, over 5694197.43 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3824, pruned_loss=0.1286, over 5655634.20 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:38:13,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 05:38:21,763 INFO [train.py:968] (0/2) Epoch 27, batch 41750, libri_loss[loss=0.2839, simple_loss=0.3551, pruned_loss=0.1063, over 29142.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3776, pruned_loss=0.125, over 5664514.81 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3645, pruned_loss=0.1158, over 5703536.69 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.38, pruned_loss=0.1267, over 5648822.88 frames. ], batch size: 101, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:38:27,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.924e+03 2.695e+03 3.362e+03 1.022e+04, threshold=5.391e+03, percent-clipped=11.0
+2023-03-14 05:38:28,946 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1226612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:38:30,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1226615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:38:39,098 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:38:43,326 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:38:55,907 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1226644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:39:05,834 INFO [train.py:968] (0/2) Epoch 27, batch 41800, libri_loss[loss=0.2866, simple_loss=0.3593, pruned_loss=0.1069, over 29284.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3747, pruned_loss=0.1235, over 5659199.31 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.364, pruned_loss=0.1157, over 5705174.54 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3776, pruned_loss=0.1253, over 5643345.13 frames. ], batch size: 94, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:39:25,535 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 05:39:38,271 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-14 05:39:57,044 INFO [train.py:968] (0/2) Epoch 27, batch 41850, giga_loss[loss=0.2961, simple_loss=0.3586, pruned_loss=0.1168, over 28755.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3746, pruned_loss=0.1234, over 5664073.72 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3638, pruned_loss=0.1156, over 5706221.67 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3771, pruned_loss=0.1249, over 5650647.33 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:40:01,302 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.111e+03 1.722e+03 2.294e+03 3.039e+03 6.515e+03, threshold=4.588e+03, percent-clipped=4.0
+2023-03-14 05:40:44,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3788, 1.5596, 1.2914, 0.9708], device='cuda:0'), covar=tensor([0.2822, 0.2724, 0.3206, 0.2409], device='cuda:0'), in_proj_covar=tensor([0.1590, 0.1146, 0.1405, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 05:40:45,044 INFO [train.py:968] (0/2) Epoch 27, batch 41900, giga_loss[loss=0.2822, simple_loss=0.3612, pruned_loss=0.1016, over 28995.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3746, pruned_loss=0.1227, over 5663820.92 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.364, pruned_loss=0.1157, over 5698146.67 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3766, pruned_loss=0.124, over 5659584.66 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:41:17,398 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:41:30,852 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 05:41:33,781 INFO [train.py:968] (0/2) Epoch 27, batch 41950, giga_loss[loss=0.256, simple_loss=0.3513, pruned_loss=0.08033, over 28982.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3713, pruned_loss=0.1196, over 5662336.31 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.364, pruned_loss=0.116, over 5690656.51 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3732, pruned_loss=0.1206, over 5664745.28 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:41:38,144 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:41:39,376 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.628e+02 1.667e+03 2.114e+03 3.098e+03 7.852e+03, threshold=4.229e+03, percent-clipped=11.0
+2023-03-14 05:42:24,668 INFO [train.py:968] (0/2) Epoch 27, batch 42000, giga_loss[loss=0.2819, simple_loss=0.3658, pruned_loss=0.09904, over 28673.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3711, pruned_loss=0.117, over 5673440.51 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3638, pruned_loss=0.1159, over 5695189.08 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.373, pruned_loss=0.1179, over 5670720.71 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 05:42:24,674 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 05:42:33,252 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.1998, simple_loss=0.3058, pruned_loss=0.04693, over 944034.00 frames. 
+2023-03-14 05:42:33,253 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 05:43:17,779 INFO [train.py:968] (0/2) Epoch 27, batch 42050, giga_loss[loss=0.371, simple_loss=0.4325, pruned_loss=0.1547, over 28943.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3723, pruned_loss=0.1173, over 5659863.85 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3638, pruned_loss=0.1161, over 5682148.06 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3743, pruned_loss=0.1179, over 5667381.59 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 05:43:21,720 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.734e+03 2.190e+03 3.109e+03 7.798e+03, threshold=4.379e+03, percent-clipped=7.0
+2023-03-14 05:43:45,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:43:55,491 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9646, 2.3219, 1.7035, 1.7565], device='cuda:0'), covar=tensor([0.1054, 0.0640, 0.0973, 0.1212], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0450, 0.0524, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 05:44:05,843 INFO [train.py:968] (0/2) Epoch 27, batch 42100, libri_loss[loss=0.303, simple_loss=0.3702, pruned_loss=0.1179, over 27649.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3728, pruned_loss=0.1183, over 5640272.07 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3637, pruned_loss=0.1162, over 5661945.62 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3747, pruned_loss=0.1188, over 5663767.95 frames. ], batch size: 115, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:44:08,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7826, 2.1178, 1.9885, 1.5433], device='cuda:0'), covar=tensor([0.3542, 0.2651, 0.2912, 0.3551], device='cuda:0'), in_proj_covar=tensor([0.2071, 0.2023, 0.1937, 0.2069], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 05:44:43,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1227000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:44:47,187 INFO [train.py:968] (0/2) Epoch 27, batch 42150, giga_loss[loss=0.2949, simple_loss=0.3679, pruned_loss=0.1109, over 28509.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3721, pruned_loss=0.1183, over 5647530.82 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3631, pruned_loss=0.1159, over 5659159.43 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3745, pruned_loss=0.1191, over 5668070.19 frames. ], batch size: 65, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:44:49,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1227006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:44:52,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2904, 1.6927, 1.3820, 1.6442], device='cuda:0'), covar=tensor([0.0822, 0.0323, 0.0347, 0.0893], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 05:44:53,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 2.020e+03 2.846e+03 3.697e+03 8.253e+03, threshold=5.693e+03, percent-clipped=15.0
+2023-03-14 05:45:31,636 INFO [train.py:968] (0/2) Epoch 27, batch 42200, giga_loss[loss=0.2776, simple_loss=0.3407, pruned_loss=0.1072, over 28669.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3711, pruned_loss=0.1186, over 5659170.01 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.363, pruned_loss=0.1158, over 5663824.04 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3733, pruned_loss=0.1193, over 5671278.29 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:46:18,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2598, 0.8316, 0.9419, 1.3878], device='cuda:0'), covar=tensor([0.0795, 0.0381, 0.0357, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 05:46:20,820 INFO [train.py:968] (0/2) Epoch 27, batch 42250, giga_loss[loss=0.2664, simple_loss=0.3547, pruned_loss=0.08901, over 28503.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3691, pruned_loss=0.1182, over 5645090.61 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3628, pruned_loss=0.1156, over 5658463.29 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3713, pruned_loss=0.1191, over 5659171.91 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:46:25,510 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.273e+03 1.883e+03 2.379e+03 3.418e+03 1.002e+04, threshold=4.759e+03, percent-clipped=6.0
+2023-03-14 05:46:59,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1227143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:02,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1227146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:07,024 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1227149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:09,263 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1227152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:12,292 INFO [train.py:968] (0/2) Epoch 27, batch 42300, giga_loss[loss=0.2546, simple_loss=0.3345, pruned_loss=0.08737, over 28864.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3684, pruned_loss=0.1175, over 5651053.58 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3629, pruned_loss=0.1156, over 5658658.97 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3701, pruned_loss=0.1182, over 5661850.67 frames. ], batch size: 66, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:47:18,158 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1227162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:20,538 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-14 05:47:30,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1227175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:36,832 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1227181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:39,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1227184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:57,120 INFO [train.py:968] (0/2) Epoch 27, batch 42350, giga_loss[loss=0.293, simple_loss=0.3723, pruned_loss=0.1069, over 28927.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3682, pruned_loss=0.1159, over 5672061.67 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3632, pruned_loss=0.1158, over 5662175.42 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3692, pruned_loss=0.1162, over 5677487.03 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:48:06,068 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.680e+03 2.190e+03 3.342e+03 1.015e+04, threshold=4.380e+03, percent-clipped=10.0
+2023-03-14 05:48:45,651 INFO [train.py:968] (0/2) Epoch 27, batch 42400, giga_loss[loss=0.3367, simple_loss=0.3988, pruned_loss=0.1373, over 28726.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3673, pruned_loss=0.1147, over 5676436.59 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3627, pruned_loss=0.1154, over 5666596.27 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3687, pruned_loss=0.1153, over 5676902.31 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:49:11,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7246, 1.8216, 1.9175, 1.4669], device='cuda:0'), covar=tensor([0.1881, 0.2560, 0.1538, 0.1774], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0713, 0.0974, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 05:49:24,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9601, 3.8110, 3.6587, 1.7257], device='cuda:0'), covar=tensor([0.0766, 0.0853, 0.0902, 0.2032], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.1210, 0.1015, 0.0752], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 05:49:35,039 INFO [train.py:968] (0/2) Epoch 27, batch 42450, giga_loss[loss=0.2585, simple_loss=0.3302, pruned_loss=0.0934, over 28412.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.367, pruned_loss=0.115, over 5679186.52 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3628, pruned_loss=0.1155, over 5667909.29 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.368, pruned_loss=0.1154, over 5678429.19 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:49:35,330 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1227305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:49:38,450 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1227308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:49:38,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1227309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:49:40,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.778e+03 2.325e+03 3.017e+03 7.296e+03, threshold=4.650e+03, percent-clipped=13.0
+2023-03-14 05:49:54,471 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1227327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:49:57,728 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1227330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:50:02,692 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1227337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:50:17,202 INFO [train.py:968] (0/2) Epoch 27, batch 42500, giga_loss[loss=0.2613, simple_loss=0.3359, pruned_loss=0.09338, over 28843.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3658, pruned_loss=0.1151, over 5681169.00 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3627, pruned_loss=0.1154, over 5670852.65 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3669, pruned_loss=0.1154, over 5677887.39 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:50:20,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1227359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:50:59,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3557, 1.2648, 3.4422, 3.1618], device='cuda:0'), covar=tensor([0.1491, 0.2760, 0.0494, 0.2067], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0674, 0.1008, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 05:51:05,456 INFO [train.py:968] (0/2) Epoch 27, batch 42550, giga_loss[loss=0.2503, simple_loss=0.3252, pruned_loss=0.08774, over 28511.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.366, pruned_loss=0.1163, over 5666185.47 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3626, pruned_loss=0.1154, over 5664589.85 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3669, pruned_loss=0.1166, over 5668431.06 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:51:11,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8100, 3.6538, 3.4943, 1.7270], device='cuda:0'), covar=tensor([0.0755, 0.0823, 0.0786, 0.2124], device='cuda:0'), in_proj_covar=tensor([0.1310, 0.1208, 0.1014, 0.0752], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 05:51:11,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.941e+03 2.545e+03 3.581e+03 1.359e+04, threshold=5.090e+03, percent-clipped=10.0
+2023-03-14 05:51:20,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7663, 4.6182, 4.3967, 2.1552], device='cuda:0'), covar=tensor([0.0503, 0.0588, 0.0648, 0.1933], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.1207, 0.1014, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 05:51:49,875 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1227452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:51:51,507 INFO [train.py:968] (0/2) Epoch 27, batch 42600, giga_loss[loss=0.2574, simple_loss=0.3371, pruned_loss=0.08882, over 28947.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3656, pruned_loss=0.117, over 5658974.84 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3628, pruned_loss=0.1155, over 5660907.13 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3662, pruned_loss=0.1172, over 5663409.05 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:51:51,871 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1227455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:52:20,797 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1227484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:52:32,743 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 05:52:39,870 INFO [train.py:968] (0/2) Epoch 27, batch 42650, giga_loss[loss=0.2386, simple_loss=0.3167, pruned_loss=0.08022, over 28905.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3623, pruned_loss=0.1151, over 5674788.79 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3624, pruned_loss=0.1151, over 5667137.78 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3632, pruned_loss=0.1156, over 5672469.23 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:52:45,535 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.226e+03 1.765e+03 2.456e+03 3.758e+03 8.794e+03, threshold=4.913e+03, percent-clipped=9.0
+2023-03-14 05:53:26,089 INFO [train.py:968] (0/2) Epoch 27, batch 42700, giga_loss[loss=0.2803, simple_loss=0.3458, pruned_loss=0.1074, over 28942.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1152, over 5650823.09 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3624, pruned_loss=0.1154, over 5639393.84 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3628, pruned_loss=0.1154, over 5675197.58 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:53:33,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1828, 0.8930, 0.9874, 1.3828], device='cuda:0'), covar=tensor([0.0793, 0.0382, 0.0359, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 05:54:13,296 INFO [train.py:968] (0/2) Epoch 27, batch 42750, giga_loss[loss=0.3593, simple_loss=0.3903, pruned_loss=0.1642, over 23967.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3629, pruned_loss=0.1154, over 5659822.43 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3624, pruned_loss=0.1151, over 5643878.36 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3634, pruned_loss=0.1158, over 5675622.75 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:54:20,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.270e+03 1.951e+03 2.702e+03 3.561e+03 1.113e+04, threshold=5.404e+03, percent-clipped=8.0
+2023-03-14 05:54:28,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9050, 3.7239, 3.5449, 1.9325], device='cuda:0'), covar=tensor([0.0752, 0.0833, 0.0825, 0.1919], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.1210, 0.1016, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 05:54:58,099 INFO [train.py:968] (0/2) Epoch 27, batch 42800, giga_loss[loss=0.2895, simple_loss=0.3702, pruned_loss=0.1044, over 28986.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3633, pruned_loss=0.1148, over 5665189.86 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3623, pruned_loss=0.1149, over 5646153.37 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3638, pruned_loss=0.1152, over 5676498.63 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 05:55:42,750 INFO [train.py:968] (0/2) Epoch 27, batch 42850, giga_loss[loss=0.3581, simple_loss=0.3842, pruned_loss=0.166, over 23572.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3638, pruned_loss=0.1146, over 5669627.62 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3619, pruned_loss=0.1145, over 5650511.42 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3647, pruned_loss=0.1153, over 5674975.66 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:55:50,219 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.053e+03 1.860e+03 2.251e+03 2.978e+03 7.800e+03, threshold=4.501e+03, percent-clipped=4.0
+2023-03-14 05:55:50,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5712, 2.6768, 2.6973, 2.3920], device='cuda:0'), covar=tensor([0.1888, 0.2110, 0.1695, 0.1883], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0760, 0.0730, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 05:56:25,534 INFO [train.py:968] (0/2) Epoch 27, batch 42900, giga_loss[loss=0.3036, simple_loss=0.376, pruned_loss=0.1157, over 28622.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3641, pruned_loss=0.1142, over 5676216.47 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.361, pruned_loss=0.114, over 5659162.48 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3656, pruned_loss=0.1153, over 5673336.14 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:57:18,405 INFO [train.py:968] (0/2) Epoch 27, batch 42950, giga_loss[loss=0.2844, simple_loss=0.3546, pruned_loss=0.1071, over 28948.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3652, pruned_loss=0.1158, over 5667120.33 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.361, pruned_loss=0.1141, over 5663045.85 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3666, pruned_loss=0.1165, over 5661690.83 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:57:23,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.717e+03 2.194e+03 3.043e+03 8.273e+03, threshold=4.387e+03, percent-clipped=6.0
+2023-03-14 05:58:02,528 INFO [train.py:968] (0/2) Epoch 27, batch 43000, giga_loss[loss=0.3553, simple_loss=0.399, pruned_loss=0.1558, over 28527.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3673, pruned_loss=0.118, over 5656508.95 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3604, pruned_loss=0.1137, over 5651507.11 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3691, pruned_loss=0.119, over 5662321.12 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:58:56,852 INFO [train.py:968] (0/2) Epoch 27, batch 43050, giga_loss[loss=0.3396, simple_loss=0.3903, pruned_loss=0.1445, over 27600.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3695, pruned_loss=0.1213, over 5646631.97 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3602, pruned_loss=0.1137, over 5651702.32 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3711, pruned_loss=0.1223, over 5651094.05 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:59:05,375 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.321e+03 2.054e+03 2.765e+03 3.787e+03 9.861e+03, threshold=5.531e+03, percent-clipped=16.0
+2023-03-14 05:59:46,566 INFO [train.py:968] (0/2) Epoch 27, batch 43100, giga_loss[loss=0.341, simple_loss=0.3952, pruned_loss=0.1434, over 28608.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3713, pruned_loss=0.1237, over 5643371.20 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3607, pruned_loss=0.1139, over 5643238.94 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3724, pruned_loss=0.1244, over 5654830.40 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:00:01,152 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1981, 1.3743, 1.2873, 1.1712], device='cuda:0'), covar=tensor([0.2308, 0.2292, 0.1698, 0.2019], device='cuda:0'), in_proj_covar=tensor([0.2077, 0.2034, 0.1942, 0.2074], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 06:00:24,873 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1227996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:00:28,281 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1228000.pt
+2023-03-14 06:00:31,895 INFO [train.py:968] (0/2) Epoch 27, batch 43150, giga_loss[loss=0.2956, simple_loss=0.361, pruned_loss=0.1151, over 28903.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3712, pruned_loss=0.1237, over 5647137.53 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3606, pruned_loss=0.1139, over 5638467.15 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3723, pruned_loss=0.1244, over 5660953.73 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:00:40,372 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.388e+03 2.051e+03 2.606e+03 3.529e+03 8.258e+03, threshold=5.212e+03, percent-clipped=9.0
+2023-03-14 06:01:20,989 INFO [train.py:968] (0/2) Epoch 27, batch 43200, giga_loss[loss=0.2946, simple_loss=0.3632, pruned_loss=0.113, over 28657.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3685, pruned_loss=0.1216, over 5659474.66 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3606, pruned_loss=0.1139, over 5641452.94 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3696, pruned_loss=0.1223, over 5668190.78 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 06:01:49,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:02:06,833 INFO [train.py:968] (0/2) Epoch 27, batch 43250, giga_loss[loss=0.2847, simple_loss=0.3673, pruned_loss=0.1011, over 29067.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3676, pruned_loss=0.1187, over 5668740.36 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3601, pruned_loss=0.1136, over 5643544.24 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3691, pruned_loss=0.1197, over 5674122.69 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:02:14,031 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.256e+03 1.669e+03 2.277e+03 2.704e+03 5.660e+03, threshold=4.554e+03, percent-clipped=2.0
+2023-03-14 06:02:52,278 INFO [train.py:968] (0/2) Epoch 27, batch 43300, giga_loss[loss=0.2557, simple_loss=0.3318, pruned_loss=0.08979, over 28712.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3662, pruned_loss=0.1175, over 5667469.98 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3603, pruned_loss=0.1137, over 5648383.79 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3674, pruned_loss=0.1183, over 5667920.45 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:03:21,361 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:03:32,441 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:03:35,340 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-14 06:03:36,223 INFO [train.py:968] (0/2) Epoch 27, batch 43350, libri_loss[loss=0.3086, simple_loss=0.3654, pruned_loss=0.1258, over 29561.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3653, pruned_loss=0.1176, over 5663995.82 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3607, pruned_loss=0.1139, over 5649877.05 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.366, pruned_loss=0.1181, over 5662885.22 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:03:41,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.864e+03 2.344e+03 2.995e+03 1.547e+04, threshold=4.688e+03, percent-clipped=7.0
+2023-03-14 06:04:20,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2145, 2.4957, 2.3211, 2.1425], device='cuda:0'), covar=tensor([0.2395, 0.2417, 0.2343, 0.2514], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0761, 0.0731, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 06:04:20,713 INFO [train.py:968] (0/2) Epoch 27, batch 43400, giga_loss[loss=0.2608, simple_loss=0.3308, pruned_loss=0.09544, over 28644.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3638, pruned_loss=0.1172, over 5663580.71 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.1141, over 5646352.30 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3643, pruned_loss=0.1176, over 5667082.30 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:05:05,625 INFO [train.py:968] (0/2) Epoch 27, batch 43450, giga_loss[loss=0.3231, simple_loss=0.3838, pruned_loss=0.1311, over 28258.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3659, pruned_loss=0.1194, over 5651716.94 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1141, over 5642451.30 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3664, pruned_loss=0.1198, over 5658258.14 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:05:12,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.154e+03 1.839e+03 2.396e+03 3.325e+03 8.775e+03, threshold=4.792e+03, percent-clipped=6.0
+2023-03-14 06:05:49,565 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:05:52,628 INFO [train.py:968] (0/2) Epoch 27, batch 43500, giga_loss[loss=0.2744, simple_loss=0.3645, pruned_loss=0.09212, over 28868.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3689, pruned_loss=0.1199, over 5659622.01 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3606, pruned_loss=0.1139, over 5645229.67 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3696, pruned_loss=0.1205, over 5662393.33 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:06:07,225 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:06:23,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3699, 1.7586, 1.5086, 1.5693], device='cuda:0'), covar=tensor([0.0692, 0.0297, 0.0301, 0.0715], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 06:06:43,538 INFO [train.py:968] (0/2) Epoch 27, batch 43550, giga_loss[loss=0.2777, simple_loss=0.3592, pruned_loss=0.0981, over 28975.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3704, pruned_loss=0.118, over 5656885.36 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3602, pruned_loss=0.1136, over 5649104.80 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3714, pruned_loss=0.1188, over 5655848.09 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:06:52,354 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.497e+03 1.988e+03 2.588e+03 5.004e+03, threshold=3.975e+03, percent-clipped=1.0
+2023-03-14 06:07:32,818 INFO [train.py:968] (0/2) Epoch 27, batch 43600, giga_loss[loss=0.3167, simple_loss=0.3861, pruned_loss=0.1236, over 28944.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3722, pruned_loss=0.1192, over 5666762.02 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3603, pruned_loss=0.1139, over 5650674.73 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3731, pruned_loss=0.1197, over 5664421.65 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:07:41,549 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:07:41,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:08:15,899 INFO [train.py:968] (0/2) Epoch 27, batch 43650, libri_loss[loss=0.2912, simple_loss=0.3591, pruned_loss=0.1116, over 29251.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3729, pruned_loss=0.1198, over 5678075.69 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3596, pruned_loss=0.1133, over 5659420.74 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3747, pruned_loss=0.1208, over 5669071.18 frames. ], batch size: 97, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:08:16,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6950, 1.6114, 1.9007, 1.4652], device='cuda:0'), covar=tensor([0.1652, 0.2503, 0.1387, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0715, 0.0975, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 06:08:25,169 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:08:26,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.298e+03 2.000e+03 2.587e+03 3.927e+03 1.383e+04, threshold=5.174e+03, percent-clipped=23.0
+2023-03-14 06:08:28,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:08:55,510 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1228546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:00,587 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 06:09:04,275 INFO [train.py:968] (0/2) Epoch 27, batch 43700, giga_loss[loss=0.3416, simple_loss=0.3985, pruned_loss=0.1423, over 28622.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3744, pruned_loss=0.122, over 5677463.70 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3596, pruned_loss=0.1133, over 5663460.84 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3762, pruned_loss=0.1229, over 5666990.30 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:09:10,039 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:10,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:20,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:41,714 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:45,164 INFO [train.py:968] (0/2) Epoch 27, batch 43750, giga_loss[loss=0.2878, simple_loss=0.352, pruned_loss=0.1118, over 28484.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3717, pruned_loss=0.1203, over 5685814.75 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3597, pruned_loss=0.1133, over 5671408.48 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3734, pruned_loss=0.1214, over 5670633.79 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:09:47,103 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:50,883 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:57,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.871e+03 2.429e+03 3.601e+03 1.265e+04, threshold=4.859e+03, percent-clipped=5.0
+2023-03-14 06:10:17,062 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1228638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:10:33,649 INFO [train.py:968] (0/2) Epoch 27, batch 43800, giga_loss[loss=0.2746, simple_loss=0.3503, pruned_loss=0.09944, over 28614.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3702, pruned_loss=0.1201, over 5667188.24 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3591, pruned_loss=0.1129, over 5665301.20 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3723, pruned_loss=0.1214, over 5660392.95 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:11:21,180 INFO [train.py:968] (0/2) Epoch 27, batch 43850, giga_loss[loss=0.3139, simple_loss=0.3746, pruned_loss=0.1266, over 28322.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3696, pruned_loss=0.1208, over 5665179.34 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3597, pruned_loss=0.1134, over 5665523.91 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3709, pruned_loss=0.1216, over 5659576.99 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:11:21,535 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:11:24,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:11:32,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.208e+03 1.830e+03 2.664e+03 3.772e+03 8.259e+03, threshold=5.328e+03, percent-clipped=4.0
+2023-03-14 06:11:35,520 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:11:37,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:11:43,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:11:56,176 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1228737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:12:01,809 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-14 06:12:07,171 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1228749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:12:11,511 INFO [train.py:968] (0/2) Epoch 27, batch 43900, giga_loss[loss=0.3759, simple_loss=0.4192, pruned_loss=0.1663, over 28561.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1212, over 5654713.32 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3591, pruned_loss=0.1129, over 5670070.99 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3711, pruned_loss=0.1226, over 5645574.45 frames. ], batch size: 85, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:12:32,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2475, 1.4052, 1.3284, 1.1340], device='cuda:0'), covar=tensor([0.2539, 0.2532, 0.1903, 0.2335], device='cuda:0'), in_proj_covar=tensor([0.2065, 0.2022, 0.1935, 0.2070], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 06:13:01,112 INFO [train.py:968] (0/2) Epoch 27, batch 43950, giga_loss[loss=0.2885, simple_loss=0.3527, pruned_loss=0.1122, over 28612.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3683, pruned_loss=0.1206, over 5644498.29 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3593, pruned_loss=0.1131, over 5662166.12 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3697, pruned_loss=0.1215, over 5644027.33 frames. ], batch size: 85, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:13:10,773 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.209e+03 1.886e+03 2.346e+03 3.038e+03 1.327e+04, threshold=4.691e+03, percent-clipped=2.0
+2023-03-14 06:13:28,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:13:41,225 INFO [train.py:968] (0/2) Epoch 27, batch 44000, giga_loss[loss=0.2977, simple_loss=0.3621, pruned_loss=0.1166, over 28666.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3672, pruned_loss=0.1201, over 5664370.38 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3596, pruned_loss=0.1132, over 5669918.11 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3685, pruned_loss=0.121, over 5656384.43 frames. ], batch size: 262, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:13:56,836 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:13:58,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:14:26,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1228901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:14:29,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6914, 1.9262, 2.0756, 1.6738], device='cuda:0'), covar=tensor([0.3116, 0.2681, 0.2501, 0.2731], device='cuda:0'), in_proj_covar=tensor([0.2067, 0.2022, 0.1936, 0.2070], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 06:14:30,288 INFO [train.py:968] (0/2) Epoch 27, batch 44050, giga_loss[loss=0.2964, simple_loss=0.3609, pruned_loss=0.1159, over 27968.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3663, pruned_loss=0.1195, over 5658604.27 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3596, pruned_loss=0.1132, over 5663160.69 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3674, pruned_loss=0.1203, over 5657474.77 frames. ], batch size: 412, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:14:40,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.652e+03 2.124e+03 2.695e+03 8.310e+03, threshold=4.249e+03, percent-clipped=5.0
+2023-03-14 06:14:56,925 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:15:12,380 INFO [train.py:968] (0/2) Epoch 27, batch 44100, giga_loss[loss=0.3087, simple_loss=0.3837, pruned_loss=0.1169, over 28939.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3661, pruned_loss=0.1189, over 5654261.13 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3594, pruned_loss=0.113, over 5658147.24 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3674, pruned_loss=0.12, over 5657217.65 frames. ], batch size: 145, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:15:25,184 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7702, 1.9475, 1.6967, 1.7000], device='cuda:0'), covar=tensor([0.2139, 0.2089, 0.2102, 0.2089], device='cuda:0'), in_proj_covar=tensor([0.1587, 0.1144, 0.1403, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 06:15:30,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6229, 1.8536, 1.2916, 1.4126], device='cuda:0'), covar=tensor([0.1062, 0.0580, 0.1096, 0.1189], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0454, 0.0527, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 06:15:34,433 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:15:37,520 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:15:40,244 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:16:02,087 INFO [train.py:968] (0/2) Epoch 27, batch 44150, giga_loss[loss=0.3998, simple_loss=0.4205, pruned_loss=0.1895, over 23529.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3683, pruned_loss=0.1198, over 5638968.72 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3593, pruned_loss=0.1128, over 5653663.65 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3696, pruned_loss=0.1211, over 5644760.98 frames. ], batch size: 705, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:16:08,801 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:16:10,528 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.174e+03 1.579e+03 1.954e+03 2.443e+03 7.896e+03, threshold=3.908e+03, percent-clipped=4.0
+2023-03-14 06:16:10,775 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 06:16:21,300 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:16:45,995 INFO [train.py:968] (0/2) Epoch 27, batch 44200, giga_loss[loss=0.2986, simple_loss=0.3588, pruned_loss=0.1192, over 28655.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3693, pruned_loss=0.1207, over 5648518.75 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3596, pruned_loss=0.1131, over 5656221.60 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3703, pruned_loss=0.1217, over 5650871.19 frames. ], batch size: 85, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:17:10,518 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:17:14,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:17:33,688 INFO [train.py:968] (0/2) Epoch 27, batch 44250, giga_loss[loss=0.2685, simple_loss=0.3589, pruned_loss=0.08907, over 29119.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3699, pruned_loss=0.1203, over 5645581.93 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3599, pruned_loss=0.1133, over 5643054.49 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3707, pruned_loss=0.121, over 5659460.82 frames. ], batch size: 128, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:17:40,652 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:17:43,987 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.136e+03 1.776e+03 2.518e+03 3.287e+03 8.754e+03, threshold=5.036e+03, percent-clipped=17.0
+2023-03-14 06:17:44,854 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:17:46,597 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:17:50,185 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:18:12,866 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:18:18,084 INFO [train.py:968] (0/2) Epoch 27, batch 44300, giga_loss[loss=0.2748, simple_loss=0.3681, pruned_loss=0.09078, over 28476.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3705, pruned_loss=0.1184, over 5656710.29 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3598, pruned_loss=0.1134, over 5651229.03 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3715, pruned_loss=0.1189, over 5660500.07 frames. ], batch size: 71, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:18:19,762 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3266, 1.6660, 1.6138, 1.4094], device='cuda:0'), covar=tensor([0.2121, 0.1945, 0.2270, 0.2303], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0762, 0.0730, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 06:18:37,022 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3177, 1.3404, 3.4870, 3.1456], device='cuda:0'), covar=tensor([0.1518, 0.2613, 0.0524, 0.1567], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0675, 0.1011, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 06:19:00,286 INFO [train.py:968] (0/2) Epoch 27, batch 44350, giga_loss[loss=0.3242, simple_loss=0.39, pruned_loss=0.1292, over 28767.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3729, pruned_loss=0.1184, over 5667222.81 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.36, pruned_loss=0.1136, over 5656050.23 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3738, pruned_loss=0.1187, over 5666068.14 frames. ], batch size: 307, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:19:15,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.961e+02 1.501e+03 1.927e+03 2.773e+03 5.127e+03, threshold=3.854e+03, percent-clipped=2.0
+2023-03-14 06:19:21,605 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4363, 3.1848, 1.4943, 1.5469], device='cuda:0'), covar=tensor([0.0935, 0.0424, 0.0884, 0.1276], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0574, 0.0409, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 06:19:47,620 INFO [train.py:968] (0/2) Epoch 27, batch 44400, giga_loss[loss=0.3717, simple_loss=0.4163, pruned_loss=0.1635, over 28597.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3762, pruned_loss=0.1217, over 5650984.96 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.1141, over 5653661.68 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3771, pruned_loss=0.1217, over 5652191.34 frames. ], batch size: 307, lr: 1.16e-03, grad_scale: 8.0
+2023-03-14 06:19:51,719 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:20:06,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6989, 1.5767, 1.8896, 1.4991], device='cuda:0'), covar=tensor([0.1230, 0.2007, 0.1042, 0.1329], device='cuda:0'), in_proj_covar=tensor([0.0929, 0.0716, 0.0975, 0.0875], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 06:20:37,406 INFO [train.py:968] (0/2) Epoch 27, batch 44450, giga_loss[loss=0.4545, simple_loss=0.4638, pruned_loss=0.2226, over 26685.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3793, pruned_loss=0.1256, over 5642328.42 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3604, pruned_loss=0.1141, over 5649316.49 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3802, pruned_loss=0.1258, over 5647839.93 frames. ], batch size: 555, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:20:43,359 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-14 06:20:46,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.285e+03 1.864e+03 2.501e+03 3.276e+03 9.267e+03, threshold=5.002e+03, percent-clipped=11.0
+2023-03-14 06:21:10,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:21:22,263 INFO [train.py:968] (0/2) Epoch 27, batch 44500, giga_loss[loss=0.2926, simple_loss=0.3613, pruned_loss=0.112, over 28847.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3777, pruned_loss=0.1251, over 5663087.88 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3597, pruned_loss=0.1138, over 5656294.69 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3797, pruned_loss=0.1259, over 5661446.01 frames. ], batch size: 199, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:21:24,632 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:21:55,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229391.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:22:06,674 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:22:09,110 INFO [train.py:968] (0/2) Epoch 27, batch 44550, giga_loss[loss=0.3086, simple_loss=0.3735, pruned_loss=0.1219, over 28974.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3772, pruned_loss=0.1249, over 5664982.69 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3599, pruned_loss=0.114, over 5658745.47 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3787, pruned_loss=0.1255, over 5661679.01 frames. ], batch size: 213, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:22:19,511 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.301e+03 1.790e+03 2.136e+03 2.920e+03 8.372e+03, threshold=4.273e+03, percent-clipped=2.0
+2023-03-14 06:22:52,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2241, 0.9259, 0.9118, 1.4575], device='cuda:0'), covar=tensor([0.0834, 0.0377, 0.0387, 0.0950], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 06:22:53,561 INFO [train.py:968] (0/2) Epoch 27, batch 44600, giga_loss[loss=0.2726, simple_loss=0.3598, pruned_loss=0.09272, over 28650.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3757, pruned_loss=0.1226, over 5671642.89 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3597, pruned_loss=0.1139, over 5659938.74 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3772, pruned_loss=0.1232, over 5667947.12 frames. ], batch size: 78, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:22:55,034 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:23:29,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:23:34,576 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.08 vs. limit=2.0
+2023-03-14 06:23:42,738 INFO [train.py:968] (0/2) Epoch 27, batch 44650, giga_loss[loss=0.339, simple_loss=0.4049, pruned_loss=0.1365, over 28902.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.376, pruned_loss=0.1211, over 5675842.54 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3599, pruned_loss=0.114, over 5662107.11 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3772, pruned_loss=0.1215, over 5671002.79 frames. ], batch size: 186, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:23:52,663 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.250e+02 1.659e+03 2.199e+03 2.778e+03 1.009e+04, threshold=4.398e+03, percent-clipped=8.0
+2023-03-14 06:24:06,795 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229534.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 06:24:10,479 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229537.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:24:19,975 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4825, 1.9551, 1.6024, 1.7693], device='cuda:0'), covar=tensor([0.0785, 0.0292, 0.0325, 0.0875], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 06:24:20,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:24:24,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:24:28,942 INFO [train.py:968] (0/2) Epoch 27, batch 44700, giga_loss[loss=0.3018, simple_loss=0.3752, pruned_loss=0.1142, over 28884.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3765, pruned_loss=0.1215, over 5666030.95 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3598, pruned_loss=0.1139, over 5664550.89 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3777, pruned_loss=0.122, over 5660136.91 frames. ], batch size: 227, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:24:39,383 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229566.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:24:51,031 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:24:56,898 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-14 06:25:16,421 INFO [train.py:968] (0/2) Epoch 27, batch 44750, giga_loss[loss=0.3543, simple_loss=0.4062, pruned_loss=0.1512, over 28588.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3773, pruned_loss=0.1227, over 5661644.46 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3598, pruned_loss=0.1137, over 5670134.86 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3787, pruned_loss=0.1234, over 5651934.17 frames. ], batch size: 336, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:25:23,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7304, 1.9416, 1.3375, 1.4406], device='cuda:0'), covar=tensor([0.1026, 0.0628, 0.1069, 0.1185], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0452, 0.0523, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 06:25:27,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.261e+03 2.076e+03 2.694e+03 3.774e+03 1.274e+04, threshold=5.388e+03, percent-clipped=15.0
+2023-03-14 06:25:44,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:25:44,332 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:25:46,162 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:25:58,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2226, 1.5208, 1.6520, 1.3531], device='cuda:0'), covar=tensor([0.2314, 0.1807, 0.2477, 0.2036], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0762, 0.0731, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 06:26:02,506 INFO [train.py:968] (0/2) Epoch 27, batch 44800, giga_loss[loss=0.2909, simple_loss=0.3522, pruned_loss=0.1148, over 28760.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3759, pruned_loss=0.1218, over 5672267.63 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3603, pruned_loss=0.1139, over 5674787.62 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3768, pruned_loss=0.1224, over 5660120.39 frames. ], batch size: 119, lr: 1.16e-03, grad_scale: 8.0
+2023-03-14 06:26:14,740 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:26:55,577 INFO [train.py:968] (0/2) Epoch 27, batch 44850, giga_loss[loss=0.3395, simple_loss=0.3822, pruned_loss=0.1484, over 26676.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3749, pruned_loss=0.1227, over 5666971.88 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3603, pruned_loss=0.1139, over 5676176.68 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3757, pruned_loss=0.1232, over 5656296.09 frames. ], batch size: 555, lr: 1.16e-03, grad_scale: 8.0
+2023-03-14 06:27:05,309 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.805e+02 1.728e+03 2.237e+03 2.996e+03 4.932e+03, threshold=4.474e+03, percent-clipped=0.0
+2023-03-14 06:27:06,438 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-14 06:27:06,957 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:27:20,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:27:39,760 INFO [train.py:968] (0/2) Epoch 27, batch 44900, giga_loss[loss=0.3155, simple_loss=0.3785, pruned_loss=0.1263, over 28929.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3728, pruned_loss=0.1216, over 5669865.62 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3602, pruned_loss=0.1136, over 5681174.95 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3738, pruned_loss=0.1224, over 5656524.20 frames. ], batch size: 213, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:27:40,038 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:27:44,925 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-14 06:27:58,968 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-14 06:28:02,932 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:28:05,146 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:28:26,115 INFO [train.py:968] (0/2) Epoch 27, batch 44950, giga_loss[loss=0.3793, simple_loss=0.4145, pruned_loss=0.1721, over 26687.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3727, pruned_loss=0.1225, over 5651680.21 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.1141, over 5668401.34 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3731, pruned_loss=0.1229, over 5653098.55 frames. ], batch size: 555, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:28:30,498 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:28:37,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.181e+03 1.860e+03 2.478e+03 3.382e+03 8.388e+03, threshold=4.956e+03, percent-clipped=12.0
+2023-03-14 06:28:42,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5265, 1.8239, 1.4312, 1.7556], device='cuda:0'), covar=tensor([0.2629, 0.2705, 0.3079, 0.2364], device='cuda:0'), in_proj_covar=tensor([0.1591, 0.1147, 0.1404, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 06:28:50,253 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:28:54,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-14 06:29:12,116 INFO [train.py:968] (0/2) Epoch 27, batch 45000, giga_loss[loss=0.2743, simple_loss=0.3469, pruned_loss=0.1009, over 28908.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3727, pruned_loss=0.1231, over 5646591.49 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3608, pruned_loss=0.1139, over 5673114.31 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3732, pruned_loss=0.1238, over 5643177.43 frames. ], batch size: 145, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:29:12,121 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 06:29:20,636 INFO [train.py:1012] (0/2) Epoch 27, validation: loss=0.2053, simple_loss=0.3148, pruned_loss=0.04794, over 944034.00 frames. 
+2023-03-14 06:29:20,637 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 06:29:25,879 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:29:28,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:29:39,045 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:29:43,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:29:54,089 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:30:06,955 INFO [train.py:968] (0/2) Epoch 27, batch 45050, giga_loss[loss=0.2552, simple_loss=0.3429, pruned_loss=0.08373, over 28905.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3701, pruned_loss=0.1206, over 5649166.58 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3606, pruned_loss=0.1138, over 5675157.14 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3709, pruned_loss=0.1212, over 5644132.38 frames. ], batch size: 174, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:30:08,611 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:30:19,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.711e+02 1.606e+03 1.984e+03 2.578e+03 7.092e+03, threshold=3.968e+03, percent-clipped=3.0
+2023-03-14 06:30:49,182 INFO [train.py:968] (0/2) Epoch 27, batch 45100, giga_loss[loss=0.2948, simple_loss=0.3692, pruned_loss=0.1102, over 28236.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3656, pruned_loss=0.116, over 5648979.63 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3611, pruned_loss=0.1141, over 5672179.68 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3659, pruned_loss=0.1164, over 5647279.71 frames. ], batch size: 368, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:31:05,597 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:31:07,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:31:30,254 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1230000.pt
+2023-03-14 06:31:36,815 INFO [train.py:968] (0/2) Epoch 27, batch 45150, giga_loss[loss=0.2919, simple_loss=0.3596, pruned_loss=0.1121, over 28292.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3647, pruned_loss=0.1153, over 5648547.71 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3609, pruned_loss=0.1138, over 5676546.24 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3652, pruned_loss=0.1159, over 5642805.13 frames. ], batch size: 368, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:31:40,381 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1230007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:31:51,276 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.109e+03 1.621e+03 2.024e+03 2.466e+03 6.724e+03, threshold=4.049e+03, percent-clipped=4.0
+2023-03-14 06:32:23,343 INFO [train.py:968] (0/2) Epoch 27, batch 45200, giga_loss[loss=0.2954, simple_loss=0.3598, pruned_loss=0.1155, over 28723.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3639, pruned_loss=0.1153, over 5668298.18 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3617, pruned_loss=0.1143, over 5680411.34 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3636, pruned_loss=0.1153, over 5659930.68 frames. ], batch size: 242, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:33:14,169 INFO [train.py:968] (0/2) Epoch 27, batch 45250, giga_loss[loss=0.2682, simple_loss=0.3526, pruned_loss=0.09195, over 28960.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3617, pruned_loss=0.1146, over 5676682.40 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1146, over 5684919.84 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1144, over 5665875.78 frames. ], batch size: 145, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:33:25,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.836e+03 2.629e+03 3.577e+03 1.262e+04, threshold=5.258e+03, percent-clipped=19.0
+2023-03-14 06:33:34,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1230130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:33:54,031 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1230153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:33:55,286 INFO [train.py:968] (0/2) Epoch 27, batch 45300, giga_loss[loss=0.3369, simple_loss=0.4036, pruned_loss=0.1351, over 28561.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3626, pruned_loss=0.1147, over 5685395.63 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3617, pruned_loss=0.1143, over 5684912.09 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3625, pruned_loss=0.1149, over 5676238.72 frames. ], batch size: 336, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:34:32,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-14 06:34:42,375 INFO [train.py:968] (0/2) Epoch 27, batch 45350, giga_loss[loss=0.2618, simple_loss=0.3398, pruned_loss=0.09186, over 28638.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3648, pruned_loss=0.1158, over 5677253.84 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1143, over 5685269.60 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3647, pruned_loss=0.1159, over 5669689.51 frames. ], batch size: 60, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:34:56,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.628e+03 2.020e+03 2.935e+03 7.067e+03, threshold=4.040e+03, percent-clipped=4.0
+2023-03-14 06:35:32,280 INFO [train.py:968] (0/2) Epoch 27, batch 45400, giga_loss[loss=0.2868, simple_loss=0.3564, pruned_loss=0.1086, over 28497.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3662, pruned_loss=0.1167, over 5669063.52 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1143, over 5681919.00 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.366, pruned_loss=0.1168, over 5666046.52 frames. ], batch size: 85, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:35:47,856 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1230273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:35:50,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1230276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:35:54,578 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6111, 1.8736, 1.5767, 1.7111], device='cuda:0'), covar=tensor([0.2311, 0.2257, 0.2375, 0.2216], device='cuda:0'), in_proj_covar=tensor([0.1598, 0.1151, 0.1409, 0.1015], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 06:36:16,938 INFO [train.py:968] (0/2) Epoch 27, batch 45450, giga_loss[loss=0.3015, simple_loss=0.3671, pruned_loss=0.1179, over 27894.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3676, pruned_loss=0.1185, over 5651848.13 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3624, pruned_loss=0.1146, over 5667668.93 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3671, pruned_loss=0.1184, over 5662018.82 frames. ], batch size: 412, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:36:17,164 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1230305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:36:30,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.651e+03 2.030e+03 2.625e+03 6.971e+03, threshold=4.059e+03, percent-clipped=5.0
+2023-03-14 06:37:01,160 INFO [train.py:968] (0/2) Epoch 27, batch 45500, giga_loss[loss=0.2852, simple_loss=0.3541, pruned_loss=0.1082, over 29123.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3677, pruned_loss=0.1186, over 5650696.42 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3624, pruned_loss=0.1145, over 5668991.78 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3674, pruned_loss=0.1185, over 5657330.67 frames. ], batch size: 113, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:37:13,782 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7686, 1.0420, 5.0533, 3.7024], device='cuda:0'), covar=tensor([0.1673, 0.3098, 0.0398, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0677, 0.1010, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 06:37:50,164 INFO [train.py:968] (0/2) Epoch 27, batch 45550, giga_loss[loss=0.2656, simple_loss=0.343, pruned_loss=0.09412, over 28484.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3692, pruned_loss=0.1195, over 5640698.97 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3626, pruned_loss=0.1147, over 5674975.79 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3689, pruned_loss=0.1195, over 5639939.08 frames. ], batch size: 85, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:38:03,541 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.214e+03 1.943e+03 2.523e+03 3.492e+03 1.065e+04, threshold=5.047e+03, percent-clipped=18.0
+2023-03-14 06:38:34,157 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-14 06:38:35,213 INFO [train.py:968] (0/2) Epoch 27, batch 45600, giga_loss[loss=0.3681, simple_loss=0.4149, pruned_loss=0.1606, over 28002.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3704, pruned_loss=0.1206, over 5646687.47 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3628, pruned_loss=0.1148, over 5674745.63 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3701, pruned_loss=0.1205, over 5645707.82 frames. ], batch size: 412, lr: 1.16e-03, grad_scale: 8.0
+2023-03-14 06:38:52,649 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8990, 1.0730, 0.9019, 0.3528], device='cuda:0'), covar=tensor([0.2775, 0.2347, 0.2644, 0.4563], device='cuda:0'), in_proj_covar=tensor([0.1837, 0.1735, 0.1653, 0.1502], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 06:39:26,326 INFO [train.py:968] (0/2) Epoch 27, batch 45650, giga_loss[loss=0.2847, simple_loss=0.3577, pruned_loss=0.1059, over 29041.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3715, pruned_loss=0.1216, over 5654692.68 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.363, pruned_loss=0.115, over 5680068.01 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3713, pruned_loss=0.1215, over 5648311.11 frames. ], batch size: 155, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:39:40,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.131e+03 1.788e+03 2.130e+03 3.003e+03 8.013e+03, threshold=4.260e+03, percent-clipped=3.0
+2023-03-14 06:39:47,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1230528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:39:57,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5602, 1.8034, 1.3391, 1.3321], device='cuda:0'), covar=tensor([0.1082, 0.0659, 0.1084, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0457, 0.0528, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 06:40:05,830 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5662, 1.8122, 1.4225, 1.7036], device='cuda:0'), covar=tensor([0.2747, 0.2860, 0.3271, 0.2475], device='cuda:0'), in_proj_covar=tensor([0.1599, 0.1151, 0.1410, 0.1015], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 06:40:14,121 INFO [train.py:968] (0/2) Epoch 27, batch 45700, giga_loss[loss=0.3496, simple_loss=0.4056, pruned_loss=0.1468, over 26645.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3726, pruned_loss=0.122, over 5643581.07 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3629, pruned_loss=0.1149, over 5675234.36 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3728, pruned_loss=0.1222, over 5643029.60 frames. ], batch size: 555, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:40:23,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3211, 1.4979, 1.4772, 1.2267], device='cuda:0'), covar=tensor([0.2826, 0.2814, 0.1822, 0.2492], device='cuda:0'), in_proj_covar=tensor([0.2068, 0.2026, 0.1942, 0.2080], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 06:41:06,154 INFO [train.py:968] (0/2) Epoch 27, batch 45750, giga_loss[loss=0.2686, simple_loss=0.3479, pruned_loss=0.09465, over 28904.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3718, pruned_loss=0.1201, over 5657322.74 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3629, pruned_loss=0.1148, over 5679498.81 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3721, pruned_loss=0.1205, over 5652791.10 frames. ], batch size: 174, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:41:20,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.172e+03 2.012e+03 2.768e+03 3.656e+03 1.016e+04, threshold=5.536e+03, percent-clipped=16.0
+2023-03-14 06:41:36,633 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1230635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:41:41,625 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3993, 1.7653, 1.3551, 1.3543], device='cuda:0'), covar=tensor([0.2789, 0.2897, 0.3318, 0.2422], device='cuda:0'), in_proj_covar=tensor([0.1599, 0.1151, 0.1410, 0.1015], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 06:41:56,115 INFO [train.py:968] (0/2) Epoch 27, batch 45800, giga_loss[loss=0.3574, simple_loss=0.3987, pruned_loss=0.1581, over 26650.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3716, pruned_loss=0.1209, over 5658584.50 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3628, pruned_loss=0.1147, over 5683469.94 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3721, pruned_loss=0.1214, over 5651104.32 frames. ], batch size: 555, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:42:03,280 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-14 06:42:13,272 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1230671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:42:15,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1230674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:42:16,298 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1230675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:42:38,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1411, 3.1205, 1.2340, 1.5250], device='cuda:0'), covar=tensor([0.1259, 0.0465, 0.1016, 0.1490], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0575, 0.0410, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 06:42:47,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1230703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:42:48,273 INFO [train.py:968] (0/2) Epoch 27, batch 45850, giga_loss[loss=0.3515, simple_loss=0.3786, pruned_loss=0.1622, over 23490.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3702, pruned_loss=0.1203, over 5652641.58 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3629, pruned_loss=0.1147, over 5675729.68 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3705, pruned_loss=0.1207, over 5654031.33 frames. ], batch size: 705, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:43:03,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7136, 1.9436, 1.3279, 1.5864], device='cuda:0'), covar=tensor([0.1101, 0.0692, 0.1074, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0456, 0.0528, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 06:43:03,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.164e+03 1.855e+03 2.398e+03 3.401e+03 1.008e+04, threshold=4.796e+03, percent-clipped=10.0
+2023-03-14 06:43:08,144 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1230722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:43:15,420 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4812, 4.3299, 4.1225, 2.2196], device='cuda:0'), covar=tensor([0.0629, 0.0734, 0.0815, 0.1782], device='cuda:0'), in_proj_covar=tensor([0.1322, 0.1218, 0.1028, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 06:43:26,750 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1230739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:43:43,350 INFO [train.py:968] (0/2) Epoch 27, batch 45900, giga_loss[loss=0.34, simple_loss=0.3833, pruned_loss=0.1484, over 26455.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.368, pruned_loss=0.1192, over 5653602.86 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3628, pruned_loss=0.1147, over 5669363.35 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3684, pruned_loss=0.1197, over 5659571.55 frames. ], batch size: 555, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:44:31,235 INFO [train.py:968] (0/2) Epoch 27, batch 45950, giga_loss[loss=0.2583, simple_loss=0.3322, pruned_loss=0.09223, over 28707.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3675, pruned_loss=0.1194, over 5608745.30 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3636, pruned_loss=0.1154, over 5617621.41 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3672, pruned_loss=0.1192, over 5659875.16 frames. ], batch size: 119, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:44:45,742 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.172e+03 1.786e+03 2.330e+03 3.482e+03 9.966e+03, threshold=4.660e+03, percent-clipped=5.0
+2023-03-14 06:45:14,381 INFO [train.py:968] (0/2) Epoch 27, batch 46000, giga_loss[loss=0.2918, simple_loss=0.3569, pruned_loss=0.1134, over 27984.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3686, pruned_loss=0.1207, over 5583321.61 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3644, pruned_loss=0.116, over 5575142.96 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3677, pruned_loss=0.12, over 5661104.13 frames. ], batch size: 412, lr: 1.16e-03, grad_scale: 8.0
+2023-03-14 06:45:53,409 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-14 06:45:54,714 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-27.pt
+2023-03-14 06:46:46,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.492e+03 1.886e+03 3.140e+03 1.624e+04, threshold=3.773e+03, percent-clipped=5.0
+2023-03-14 06:46:55,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5571, 1.6978, 1.7796, 1.3562], device='cuda:0'), covar=tensor([0.2075, 0.3024, 0.1740, 0.2174], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0717, 0.0977, 0.0876], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 06:47:13,128 INFO [train.py:968] (0/2) Epoch 28, batch 50, libri_loss[loss=0.2739, simple_loss=0.36, pruned_loss=0.0939, over 27792.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3684, pruned_loss=0.1043, over 1262126.11 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3318, pruned_loss=0.08034, over 115575.23 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3717, pruned_loss=0.1064, over 1170121.25 frames. ], batch size: 116, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:47:16,863 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-14 06:47:59,893 INFO [train.py:968] (0/2) Epoch 28, batch 100, giga_loss[loss=0.2378, simple_loss=0.3208, pruned_loss=0.07743, over 28712.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3592, pruned_loss=0.09979, over 2247893.08 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3325, pruned_loss=0.0812, over 230741.61 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3617, pruned_loss=0.1015, over 2100967.72 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:48:09,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:48:19,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.600e+02 1.307e+03 1.509e+03 2.151e+03 7.287e+03, threshold=3.018e+03, percent-clipped=2.0
+2023-03-14 06:48:21,039 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1231022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:48:23,000 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1231025.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:48:45,283 INFO [train.py:968] (0/2) Epoch 28, batch 150, giga_loss[loss=0.2489, simple_loss=0.3251, pruned_loss=0.08633, over 28636.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.345, pruned_loss=0.09422, over 3009060.76 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3344, pruned_loss=0.08311, over 287155.48 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3461, pruned_loss=0.09511, over 2864868.63 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:48:46,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:48:55,770 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3639, 1.3817, 3.6170, 3.2440], device='cuda:0'), covar=tensor([0.1585, 0.2824, 0.0460, 0.1456], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0674, 0.1009, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 06:49:13,998 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 06:49:24,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:49:25,208 INFO [train.py:968] (0/2) Epoch 28, batch 200, libri_loss[loss=0.23, simple_loss=0.3128, pruned_loss=0.07358, over 28064.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3327, pruned_loss=0.08821, over 3613683.88 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3318, pruned_loss=0.08271, over 450102.55 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3334, pruned_loss=0.08897, over 3431913.70 frames. ], batch size: 62, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:49:36,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:49:38,337 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-14 06:49:40,345 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.820e+02 1.156e+03 1.371e+03 1.821e+03 5.019e+03, threshold=2.743e+03, percent-clipped=2.0
+2023-03-14 06:49:59,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3323, 3.4998, 1.5082, 1.4678], device='cuda:0'), covar=tensor([0.1123, 0.0360, 0.1026, 0.1519], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0570, 0.0407, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 06:50:04,670 INFO [train.py:968] (0/2) Epoch 28, batch 250, giga_loss[loss=0.2115, simple_loss=0.2919, pruned_loss=0.06552, over 28664.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3228, pruned_loss=0.08363, over 4083510.43 frames. ], libri_tot_loss[loss=0.2448, simple_loss=0.3273, pruned_loss=0.08118, over 664010.28 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3233, pruned_loss=0.08437, over 3862979.73 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:50:10,155 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:50:12,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:50:35,047 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:50:42,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:50:45,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:50:47,629 INFO [train.py:968] (0/2) Epoch 28, batch 300, giga_loss[loss=0.217, simple_loss=0.2903, pruned_loss=0.07183, over 28893.00 frames. ], tot_loss[loss=0.238, simple_loss=0.315, pruned_loss=0.08053, over 4445225.41 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3291, pruned_loss=0.08223, over 767619.86 frames. ], giga_tot_loss[loss=0.238, simple_loss=0.3145, pruned_loss=0.08076, over 4241305.70 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:51:03,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3184, 3.1633, 2.9820, 1.5016], device='cuda:0'), covar=tensor([0.0975, 0.1152, 0.1039, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.1312, 0.1212, 0.1020, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 06:51:04,330 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.176e+02 1.170e+03 1.482e+03 2.327e+03 5.341e+03, threshold=2.964e+03, percent-clipped=13.0
+2023-03-14 06:51:08,210 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:51:23,866 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:51:25,873 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:51:29,963 INFO [train.py:968] (0/2) Epoch 28, batch 350, libri_loss[loss=0.2335, simple_loss=0.3162, pruned_loss=0.07535, over 29596.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3089, pruned_loss=0.07777, over 4728141.52 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3299, pruned_loss=0.08167, over 944444.60 frames. ], giga_tot_loss[loss=0.2315, simple_loss=0.3074, pruned_loss=0.07781, over 4522798.64 frames. ], batch size: 75, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:51:36,761 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:51:38,673 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:51:48,758 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:52:02,666 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:52:10,039 INFO [train.py:968] (0/2) Epoch 28, batch 400, giga_loss[loss=0.2178, simple_loss=0.2916, pruned_loss=0.07197, over 27650.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3056, pruned_loss=0.0764, over 4948189.11 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3316, pruned_loss=0.08273, over 1063377.94 frames. ], giga_tot_loss[loss=0.2278, simple_loss=0.3034, pruned_loss=0.07606, over 4763918.23 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 06:52:27,383 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.990e+02 1.139e+03 1.458e+03 1.882e+03 5.273e+03, threshold=2.917e+03, percent-clipped=8.0
+2023-03-14 06:52:42,211 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1231340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:52:48,638 INFO [train.py:968] (0/2) Epoch 28, batch 450, giga_loss[loss=0.2125, simple_loss=0.2979, pruned_loss=0.06358, over 28931.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3044, pruned_loss=0.07571, over 5115676.51 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3339, pruned_loss=0.08453, over 1242964.61 frames. ], giga_tot_loss[loss=0.2253, simple_loss=0.3011, pruned_loss=0.07475, over 4944871.04 frames. ], batch size: 174, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 06:53:31,295 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:53:32,548 INFO [train.py:968] (0/2) Epoch 28, batch 500, giga_loss[loss=0.2131, simple_loss=0.2936, pruned_loss=0.06631, over 28913.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3012, pruned_loss=0.07404, over 5249957.86 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3348, pruned_loss=0.0849, over 1354477.19 frames. ], giga_tot_loss[loss=0.2217, simple_loss=0.2976, pruned_loss=0.07292, over 5101104.40 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:53:33,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231400.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 06:53:54,133 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.248e+02 1.251e+03 1.498e+03 2.109e+03 4.917e+03, threshold=2.995e+03, percent-clipped=11.0
+2023-03-14 06:54:11,898 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7147, 2.0070, 1.9621, 1.7635], device='cuda:0'), covar=tensor([0.2445, 0.2339, 0.2473, 0.2477], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0761, 0.0729, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 06:54:17,826 INFO [train.py:968] (0/2) Epoch 28, batch 550, giga_loss[loss=0.1986, simple_loss=0.2768, pruned_loss=0.06017, over 29060.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2991, pruned_loss=0.07319, over 5350231.04 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3348, pruned_loss=0.08517, over 1444707.32 frames. ], giga_tot_loss[loss=0.2197, simple_loss=0.2954, pruned_loss=0.07199, over 5221352.97 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:54:34,965 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5140, 1.8029, 1.5414, 1.0591], device='cuda:0'), covar=tensor([0.2650, 0.2677, 0.3077, 0.2622], device='cuda:0'), in_proj_covar=tensor([0.1603, 0.1153, 0.1415, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 06:55:02,637 INFO [train.py:968] (0/2) Epoch 28, batch 600, giga_loss[loss=0.214, simple_loss=0.2875, pruned_loss=0.07028, over 28269.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2974, pruned_loss=0.07275, over 5428419.57 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3348, pruned_loss=0.0852, over 1533378.18 frames. ], giga_tot_loss[loss=0.2185, simple_loss=0.2939, pruned_loss=0.07157, over 5316845.77 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:55:23,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.709e+02 1.133e+03 1.423e+03 2.065e+03 8.759e+03, threshold=2.846e+03, percent-clipped=13.0
+2023-03-14 06:55:40,526 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:55:44,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:55:44,499 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231543.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:55:46,475 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231546.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:55:48,007 INFO [train.py:968] (0/2) Epoch 28, batch 650, giga_loss[loss=0.2196, simple_loss=0.2891, pruned_loss=0.07507, over 28851.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2965, pruned_loss=0.07248, over 5480785.91 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3356, pruned_loss=0.08566, over 1639847.24 frames. ], giga_tot_loss[loss=0.2173, simple_loss=0.2925, pruned_loss=0.07109, over 5383194.94 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:56:05,206 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:56:08,791 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231575.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 06:56:30,629 INFO [train.py:968] (0/2) Epoch 28, batch 700, giga_loss[loss=0.1911, simple_loss=0.2633, pruned_loss=0.05947, over 28059.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2954, pruned_loss=0.0721, over 5524552.83 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.337, pruned_loss=0.08601, over 1797200.72 frames. ], giga_tot_loss[loss=0.2156, simple_loss=0.2903, pruned_loss=0.0704, over 5439956.85 frames. ], batch size: 77, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:56:45,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5463, 3.6307, 1.6921, 1.8034], device='cuda:0'), covar=tensor([0.1033, 0.0316, 0.0940, 0.1326], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0568, 0.0408, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 06:56:49,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.609e+02 1.111e+03 1.328e+03 2.007e+03 5.457e+03, threshold=2.657e+03, percent-clipped=5.0
+2023-03-14 06:57:12,647 INFO [train.py:968] (0/2) Epoch 28, batch 750, giga_loss[loss=0.1859, simple_loss=0.2612, pruned_loss=0.05524, over 28973.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2931, pruned_loss=0.07089, over 5575080.51 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3356, pruned_loss=0.08603, over 1953991.16 frames. ], giga_tot_loss[loss=0.2128, simple_loss=0.2878, pruned_loss=0.06894, over 5497854.06 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:57:56,039 INFO [train.py:968] (0/2) Epoch 28, batch 800, giga_loss[loss=0.229, simple_loss=0.3062, pruned_loss=0.07585, over 28284.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2914, pruned_loss=0.07057, over 5603771.66 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3359, pruned_loss=0.08603, over 2032169.03 frames. ], giga_tot_loss[loss=0.2118, simple_loss=0.2863, pruned_loss=0.06871, over 5536991.45 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 06:57:57,739 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1231700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:58:09,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.24 vs. limit=5.0
+2023-03-14 06:58:10,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:58:16,509 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.685e+02 1.226e+03 1.631e+03 2.283e+03 1.001e+04, threshold=3.261e+03, percent-clipped=15.0
+2023-03-14 06:58:42,964 INFO [train.py:968] (0/2) Epoch 28, batch 850, giga_loss[loss=0.2431, simple_loss=0.3289, pruned_loss=0.07859, over 29041.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.2978, pruned_loss=0.07444, over 5615384.40 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3356, pruned_loss=0.08605, over 2127231.64 frames. ], giga_tot_loss[loss=0.219, simple_loss=0.2928, pruned_loss=0.07265, over 5556266.46 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 06:58:44,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5576, 4.0905, 1.6785, 1.7357], device='cuda:0'), covar=tensor([0.0998, 0.0300, 0.0952, 0.1298], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0568, 0.0408, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 06:59:21,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-14 06:59:28,533 INFO [train.py:968] (0/2) Epoch 28, batch 900, libri_loss[loss=0.2127, simple_loss=0.3003, pruned_loss=0.06251, over 29546.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3092, pruned_loss=0.07986, over 5633327.57 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3345, pruned_loss=0.08519, over 2236953.52 frames. ], giga_tot_loss[loss=0.231, simple_loss=0.3049, pruned_loss=0.07856, over 5580179.14 frames. ], batch size: 76, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 06:59:47,172 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.886e+02 1.380e+03 1.679e+03 2.317e+03 5.828e+03, threshold=3.357e+03, percent-clipped=10.0
+2023-03-14 07:00:11,759 INFO [train.py:968] (0/2) Epoch 28, batch 950, giga_loss[loss=0.2761, simple_loss=0.3473, pruned_loss=0.1025, over 28515.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3209, pruned_loss=0.08555, over 5645789.84 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3338, pruned_loss=0.08488, over 2281229.06 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3176, pruned_loss=0.08464, over 5609240.82 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:00:18,552 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:00:21,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:00:26,099 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.1668, 5.9783, 5.7052, 3.5101], device='cuda:0'), covar=tensor([0.0536, 0.0631, 0.0896, 0.1371], device='cuda:0'), in_proj_covar=tensor([0.1296, 0.1199, 0.1010, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 07:00:42,816 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:00:51,249 INFO [train.py:968] (0/2) Epoch 28, batch 1000, giga_loss[loss=0.2463, simple_loss=0.3323, pruned_loss=0.08013, over 28366.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3281, pruned_loss=0.08857, over 5664908.91 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3324, pruned_loss=0.08433, over 2441366.74 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3256, pruned_loss=0.08815, over 5624891.06 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:00:55,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2610, 1.2359, 1.1276, 1.4771], device='cuda:0'), covar=tensor([0.0815, 0.0384, 0.0377, 0.0931], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 07:01:11,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.837e+02 1.339e+03 1.640e+03 2.303e+03 6.215e+03, threshold=3.280e+03, percent-clipped=8.0
+2023-03-14 07:01:31,460 INFO [train.py:968] (0/2) Epoch 28, batch 1050, giga_loss[loss=0.232, simple_loss=0.3256, pruned_loss=0.06921, over 28799.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3321, pruned_loss=0.08943, over 5681234.97 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3333, pruned_loss=0.08526, over 2561085.32 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3297, pruned_loss=0.08889, over 5641366.62 frames. ], batch size: 285, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:01:42,437 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-14 07:02:17,943 INFO [train.py:968] (0/2) Epoch 28, batch 1100, giga_loss[loss=0.2548, simple_loss=0.3417, pruned_loss=0.08394, over 28841.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3342, pruned_loss=0.08922, over 5674407.60 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.333, pruned_loss=0.08488, over 2654525.94 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3325, pruned_loss=0.08907, over 5641066.79 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:02:18,741 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1232000.pt
+2023-03-14 07:02:36,055 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.099e+02 1.302e+03 1.491e+03 1.944e+03 5.683e+03, threshold=2.982e+03, percent-clipped=5.0
+2023-03-14 07:02:53,038 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5301, 2.0444, 1.6161, 1.7264], device='cuda:0'), covar=tensor([0.0767, 0.0282, 0.0333, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 07:02:59,828 INFO [train.py:968] (0/2) Epoch 28, batch 1150, giga_loss[loss=0.2554, simple_loss=0.3224, pruned_loss=0.09417, over 28706.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3359, pruned_loss=0.08986, over 5688625.51 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3339, pruned_loss=0.08561, over 2731618.08 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3341, pruned_loss=0.08951, over 5659953.36 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:03:19,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1232075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:03:43,946 INFO [train.py:968] (0/2) Epoch 28, batch 1200, giga_loss[loss=0.3221, simple_loss=0.387, pruned_loss=0.1286, over 28675.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3393, pruned_loss=0.09239, over 5680136.93 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3342, pruned_loss=0.08556, over 2791038.51 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3379, pruned_loss=0.09222, over 5656691.22 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:04:02,254 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3913, 2.3083, 2.3920, 2.0237], device='cuda:0'), covar=tensor([0.2164, 0.2505, 0.2168, 0.2418], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0761, 0.0731, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 07:04:04,365 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.071e+02 1.352e+03 1.572e+03 2.288e+03 1.028e+04, threshold=3.144e+03, percent-clipped=14.0
+2023-03-14 07:04:26,128 INFO [train.py:968] (0/2) Epoch 28, batch 1250, giga_loss[loss=0.2802, simple_loss=0.3578, pruned_loss=0.1013, over 28848.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.342, pruned_loss=0.09438, over 5684293.58 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3346, pruned_loss=0.08587, over 2853025.33 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3408, pruned_loss=0.09426, over 5661589.86 frames. ], batch size: 174, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:04:42,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8735, 1.7768, 2.1362, 1.6250], device='cuda:0'), covar=tensor([0.1837, 0.2767, 0.1464, 0.1874], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0720, 0.0989, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 07:04:56,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8480, 3.6790, 3.4912, 2.1052], device='cuda:0'), covar=tensor([0.0632, 0.0781, 0.0792, 0.1801], device='cuda:0'), in_proj_covar=tensor([0.1284, 0.1190, 0.1001, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 07:05:08,593 INFO [train.py:968] (0/2) Epoch 28, batch 1300, giga_loss[loss=0.2531, simple_loss=0.3367, pruned_loss=0.08473, over 29040.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3458, pruned_loss=0.096, over 5673422.98 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3358, pruned_loss=0.08641, over 2918080.79 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3444, pruned_loss=0.09585, over 5659842.22 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:05:23,268 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.33 vs. limit=2.0
+2023-03-14 07:05:25,367 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1232218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:05:28,514 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1232221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:05:29,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.168e+02 1.374e+03 1.600e+03 2.038e+03 3.454e+03, threshold=3.199e+03, percent-clipped=4.0
+2023-03-14 07:05:49,333 INFO [train.py:968] (0/2) Epoch 28, batch 1350, giga_loss[loss=0.3513, simple_loss=0.4102, pruned_loss=0.1461, over 28695.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3481, pruned_loss=0.09619, over 5692329.89 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.336, pruned_loss=0.08641, over 2947840.57 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3471, pruned_loss=0.09616, over 5679683.88 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:05:51,338 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1232250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:06:32,199 INFO [train.py:968] (0/2) Epoch 28, batch 1400, giga_loss[loss=0.2629, simple_loss=0.3489, pruned_loss=0.08851, over 28565.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3479, pruned_loss=0.09549, over 5690167.94 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3347, pruned_loss=0.08567, over 3033650.21 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.348, pruned_loss=0.09608, over 5676027.44 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:06:50,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.701e+02 1.389e+03 1.638e+03 2.209e+03 4.457e+03, threshold=3.276e+03, percent-clipped=7.0
+2023-03-14 07:07:11,261 INFO [train.py:968] (0/2) Epoch 28, batch 1450, giga_loss[loss=0.2675, simple_loss=0.3522, pruned_loss=0.09134, over 28961.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3482, pruned_loss=0.09468, over 5694014.07 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3345, pruned_loss=0.08542, over 3129616.80 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3488, pruned_loss=0.09557, over 5678682.28 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:07:27,492 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1232367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:07:52,901 INFO [train.py:968] (0/2) Epoch 28, batch 1500, libri_loss[loss=0.2649, simple_loss=0.3551, pruned_loss=0.08733, over 29538.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3459, pruned_loss=0.09205, over 5703253.52 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3337, pruned_loss=0.08492, over 3183752.84 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.347, pruned_loss=0.09313, over 5688238.60 frames. ], batch size: 79, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:07:54,500 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1232401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:08:11,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.107e+02 1.189e+03 1.423e+03 1.878e+03 4.229e+03, threshold=2.845e+03, percent-clipped=4.0
+2023-03-14 07:08:20,310 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3757, 1.5338, 1.3422, 1.6886], device='cuda:0'), covar=tensor([0.0809, 0.0374, 0.0347, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 07:08:29,826 INFO [train.py:968] (0/2) Epoch 28, batch 1550, giga_loss[loss=0.2534, simple_loss=0.3309, pruned_loss=0.08792, over 28806.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3443, pruned_loss=0.0902, over 5709304.71 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3344, pruned_loss=0.08518, over 3249648.69 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3451, pruned_loss=0.09107, over 5694167.90 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:08:41,642 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1232461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:09:17,398 INFO [train.py:968] (0/2) Epoch 28, batch 1600, giga_loss[loss=0.2891, simple_loss=0.3614, pruned_loss=0.1084, over 28953.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3443, pruned_loss=0.09095, over 5713795.67 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3343, pruned_loss=0.08524, over 3289391.76 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3451, pruned_loss=0.09169, over 5699101.83 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:09:37,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.551e+02 1.333e+03 1.731e+03 2.445e+03 4.013e+03, threshold=3.461e+03, percent-clipped=11.0
+2023-03-14 07:10:00,147 INFO [train.py:968] (0/2) Epoch 28, batch 1650, giga_loss[loss=0.3066, simple_loss=0.3716, pruned_loss=0.1208, over 28673.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3476, pruned_loss=0.09606, over 5717748.70 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3346, pruned_loss=0.08544, over 3328617.69 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3482, pruned_loss=0.09668, over 5703338.77 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:10:05,561 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4361, 2.0132, 1.4767, 0.8067], device='cuda:0'), covar=tensor([0.7359, 0.3459, 0.4384, 0.7023], device='cuda:0'), in_proj_covar=tensor([0.1843, 0.1731, 0.1657, 0.1504], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 07:10:23,154 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3926, 1.6439, 1.2385, 1.1896], device='cuda:0'), covar=tensor([0.1085, 0.0506, 0.1037, 0.1099], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0454, 0.0525, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 07:10:40,380 INFO [train.py:968] (0/2) Epoch 28, batch 1700, giga_loss[loss=0.3495, simple_loss=0.3899, pruned_loss=0.1546, over 26505.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3485, pruned_loss=0.0987, over 5709839.10 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3346, pruned_loss=0.08548, over 3417127.04 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3494, pruned_loss=0.09957, over 5692300.93 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:11:01,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.833e+02 1.427e+03 1.981e+03 2.850e+03 6.639e+03, threshold=3.962e+03, percent-clipped=13.0
+2023-03-14 07:11:27,770 INFO [train.py:968] (0/2) Epoch 28, batch 1750, giga_loss[loss=0.2874, simple_loss=0.3495, pruned_loss=0.1126, over 28932.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3476, pruned_loss=0.09917, over 5705718.49 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3342, pruned_loss=0.08516, over 3454059.46 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3488, pruned_loss=0.1002, over 5689595.15 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:11:30,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6691, 1.5235, 1.7841, 1.3301], device='cuda:0'), covar=tensor([0.1997, 0.3679, 0.1547, 0.1920], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0717, 0.0985, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 07:11:38,755 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1232662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:11:51,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6610, 1.9313, 1.5634, 1.8324], device='cuda:0'), covar=tensor([0.2671, 0.2755, 0.3053, 0.2368], device='cuda:0'), in_proj_covar=tensor([0.1597, 0.1151, 0.1407, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 07:12:09,053 INFO [train.py:968] (0/2) Epoch 28, batch 1800, giga_loss[loss=0.3069, simple_loss=0.3746, pruned_loss=0.1196, over 28937.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3464, pruned_loss=0.09879, over 5714328.99 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3341, pruned_loss=0.08501, over 3522614.97 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3477, pruned_loss=0.1001, over 5699775.85 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:12:31,808 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.853e+02 1.326e+03 1.601e+03 2.143e+03 5.264e+03, threshold=3.202e+03, percent-clipped=3.0
+2023-03-14 07:12:45,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1232742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:12:50,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6315, 1.7875, 1.8518, 1.4156], device='cuda:0'), covar=tensor([0.2008, 0.2785, 0.1679, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0717, 0.0986, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 07:12:51,187 INFO [train.py:968] (0/2) Epoch 28, batch 1850, giga_loss[loss=0.2401, simple_loss=0.331, pruned_loss=0.07463, over 28887.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.346, pruned_loss=0.09833, over 5717500.65 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3344, pruned_loss=0.08517, over 3570165.74 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.347, pruned_loss=0.0995, over 5702124.43 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:13:15,562 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1232776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:13:34,645 INFO [train.py:968] (0/2) Epoch 28, batch 1900, giga_loss[loss=0.4091, simple_loss=0.4298, pruned_loss=0.1942, over 26558.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3455, pruned_loss=0.09748, over 5712537.37 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3348, pruned_loss=0.08539, over 3633591.89 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3463, pruned_loss=0.09861, over 5700532.32 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:13:59,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.775e+02 1.317e+03 1.549e+03 2.128e+03 6.129e+03, threshold=3.098e+03, percent-clipped=4.0
+2023-03-14 07:14:12,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1232836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:14:22,555 INFO [train.py:968] (0/2) Epoch 28, batch 1950, libri_loss[loss=0.2072, simple_loss=0.2905, pruned_loss=0.06191, over 29386.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3418, pruned_loss=0.09517, over 5701447.60 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3344, pruned_loss=0.08517, over 3678119.32 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.343, pruned_loss=0.09641, over 5688617.81 frames. ], batch size: 67, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:14:50,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3069, 1.2186, 4.1972, 3.3777], device='cuda:0'), covar=tensor([0.1700, 0.2884, 0.0490, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.0798, 0.0671, 0.1002, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 07:14:54,854 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1232885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:14:56,945 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1232888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:15:07,475 INFO [train.py:968] (0/2) Epoch 28, batch 2000, giga_loss[loss=0.2645, simple_loss=0.3149, pruned_loss=0.1071, over 23579.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3357, pruned_loss=0.09148, over 5685523.84 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3347, pruned_loss=0.08516, over 3722419.08 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3365, pruned_loss=0.09264, over 5679708.53 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:15:09,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3449, 1.5990, 1.2096, 1.1366], device='cuda:0'), covar=tensor([0.1051, 0.0503, 0.1027, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0451, 0.0523, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 07:15:25,084 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1232917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:15:26,494 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1232919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:15:28,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1232922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:15:30,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.130e+02 1.063e+03 1.362e+03 2.020e+03 6.123e+03, threshold=2.724e+03, percent-clipped=12.0
+2023-03-14 07:15:46,524 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7271, 4.4328, 1.7787, 1.7037], device='cuda:0'), covar=tensor([0.0962, 0.0226, 0.0935, 0.1358], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0564, 0.0407, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 07:15:50,323 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2182, 1.5746, 1.6424, 1.3727], device='cuda:0'), covar=tensor([0.2422, 0.1889, 0.2541, 0.2107], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0763, 0.0732, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 07:15:51,911 INFO [train.py:968] (0/2) Epoch 28, batch 2050, giga_loss[loss=0.2206, simple_loss=0.3007, pruned_loss=0.07022, over 28880.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3297, pruned_loss=0.08831, over 5677773.15 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3343, pruned_loss=0.08495, over 3785475.89 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3305, pruned_loss=0.08951, over 5676408.63 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:15:53,341 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1232951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:16:20,071 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1232979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:16:22,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1232982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:16:38,073 INFO [train.py:968] (0/2) Epoch 28, batch 2100, giga_loss[loss=0.2548, simple_loss=0.3321, pruned_loss=0.08874, over 28058.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3265, pruned_loss=0.08637, over 5680783.67 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3342, pruned_loss=0.08464, over 3844614.50 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3271, pruned_loss=0.0876, over 5676931.04 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:16:48,219 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1233011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:16:58,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.631e+02 1.140e+03 1.440e+03 1.881e+03 4.272e+03, threshold=2.879e+03, percent-clipped=9.0
+2023-03-14 07:17:06,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1233037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:17:07,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 07:17:15,780 INFO [train.py:968] (0/2) Epoch 28, batch 2150, giga_loss[loss=0.2202, simple_loss=0.3025, pruned_loss=0.06898, over 28592.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3274, pruned_loss=0.08628, over 5689621.99 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3342, pruned_loss=0.08449, over 3895088.23 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3276, pruned_loss=0.08738, over 5682125.06 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:17:26,767 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-14 07:17:37,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4965, 1.7555, 1.7185, 1.2772], device='cuda:0'), covar=tensor([0.1971, 0.3004, 0.1738, 0.2027], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0718, 0.0986, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 07:17:52,979 INFO [train.py:968] (0/2) Epoch 28, batch 2200, giga_loss[loss=0.2755, simple_loss=0.3447, pruned_loss=0.1032, over 28859.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3275, pruned_loss=0.08565, over 5697914.32 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3347, pruned_loss=0.08444, over 3954259.07 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3272, pruned_loss=0.08661, over 5686705.46 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:18:16,406 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.052e+02 1.160e+03 1.320e+03 1.975e+03 6.913e+03, threshold=2.641e+03, percent-clipped=9.0
+2023-03-14 07:18:17,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4652, 1.7310, 1.4284, 1.3659], device='cuda:0'), covar=tensor([0.2690, 0.2850, 0.3175, 0.2517], device='cuda:0'), in_proj_covar=tensor([0.1597, 0.1150, 0.1407, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 07:18:36,000 INFO [train.py:968] (0/2) Epoch 28, batch 2250, libri_loss[loss=0.2828, simple_loss=0.3699, pruned_loss=0.09788, over 29540.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3276, pruned_loss=0.08584, over 5704804.44 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3353, pruned_loss=0.08445, over 4021074.79 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3268, pruned_loss=0.08666, over 5689630.04 frames. ], batch size: 83, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:18:46,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5277, 1.7777, 1.6106, 1.5317], device='cuda:0'), covar=tensor([0.0802, 0.0335, 0.0318, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 07:18:59,836 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1233180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:19:02,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1233183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:19:18,411 INFO [train.py:968] (0/2) Epoch 28, batch 2300, giga_loss[loss=0.2207, simple_loss=0.2975, pruned_loss=0.07194, over 28243.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.325, pruned_loss=0.08489, over 5705408.91 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3351, pruned_loss=0.08426, over 4037392.44 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3243, pruned_loss=0.08565, over 5693925.35 frames. ], batch size: 65, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:19:31,312 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1233212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:19:41,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.812e+02 1.200e+03 1.422e+03 1.789e+03 4.356e+03, threshold=2.845e+03, percent-clipped=5.0
+2023-03-14 07:19:50,923 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1233238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:19:55,806 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 07:19:59,661 INFO [train.py:968] (0/2) Epoch 28, batch 2350, giga_loss[loss=0.2331, simple_loss=0.3048, pruned_loss=0.08067, over 28345.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3235, pruned_loss=0.08445, over 5715947.97 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.336, pruned_loss=0.08453, over 4064655.35 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3223, pruned_loss=0.0849, over 5704880.94 frames. ], batch size: 65, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:20:10,426 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1233264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:20:41,330 INFO [train.py:968] (0/2) Epoch 28, batch 2400, giga_loss[loss=0.2134, simple_loss=0.2872, pruned_loss=0.06982, over 28876.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3223, pruned_loss=0.08443, over 5721898.87 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3363, pruned_loss=0.08469, over 4083375.19 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3211, pruned_loss=0.08469, over 5711382.13 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:21:01,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.530e+02 1.145e+03 1.344e+03 1.657e+03 6.722e+03, threshold=2.688e+03, percent-clipped=4.0
+2023-03-14 07:21:18,753 INFO [train.py:968] (0/2) Epoch 28, batch 2450, giga_loss[loss=0.2308, simple_loss=0.3103, pruned_loss=0.07571, over 28839.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3208, pruned_loss=0.08405, over 5727985.59 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3367, pruned_loss=0.08472, over 4119496.26 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3193, pruned_loss=0.08422, over 5716427.54 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:21:58,203 INFO [train.py:968] (0/2) Epoch 28, batch 2500, giga_loss[loss=0.2494, simple_loss=0.3224, pruned_loss=0.0882, over 28625.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3178, pruned_loss=0.08252, over 5727300.76 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3367, pruned_loss=0.08469, over 4134728.93 frames. ], giga_tot_loss[loss=0.2409, simple_loss=0.3165, pruned_loss=0.08267, over 5718893.88 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:22:18,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.677e+02 1.127e+03 1.332e+03 1.631e+03 4.714e+03, threshold=2.664e+03, percent-clipped=3.0
+2023-03-14 07:22:36,665 INFO [train.py:968] (0/2) Epoch 28, batch 2550, giga_loss[loss=0.2177, simple_loss=0.2905, pruned_loss=0.07249, over 28616.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3172, pruned_loss=0.08211, over 5707953.46 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3379, pruned_loss=0.08519, over 4183242.29 frames. ], giga_tot_loss[loss=0.2392, simple_loss=0.3147, pruned_loss=0.08182, over 5713343.06 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:23:14,067 INFO [train.py:968] (0/2) Epoch 28, batch 2600, giga_loss[loss=0.2263, simple_loss=0.3107, pruned_loss=0.07097, over 29074.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.317, pruned_loss=0.08147, over 5706224.11 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3385, pruned_loss=0.0854, over 4263999.00 frames. ], giga_tot_loss[loss=0.2376, simple_loss=0.3134, pruned_loss=0.08092, over 5718323.23 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:23:28,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0257, 1.3021, 1.0846, 0.3660], device='cuda:0'), covar=tensor([0.4679, 0.3689, 0.5636, 0.7173], device='cuda:0'), in_proj_covar=tensor([0.1824, 0.1715, 0.1644, 0.1492], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 07:23:37,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.086e+02 1.207e+03 1.420e+03 2.152e+03 1.029e+04, threshold=2.840e+03, percent-clipped=20.0
+2023-03-14 07:23:53,349 INFO [train.py:968] (0/2) Epoch 28, batch 2650, giga_loss[loss=0.252, simple_loss=0.3165, pruned_loss=0.09376, over 24292.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3158, pruned_loss=0.08091, over 5703810.93 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3383, pruned_loss=0.08508, over 4311164.16 frames. ], giga_tot_loss[loss=0.2367, simple_loss=0.3124, pruned_loss=0.08055, over 5715989.41 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:24:31,891 INFO [train.py:968] (0/2) Epoch 28, batch 2700, giga_loss[loss=0.3351, simple_loss=0.3851, pruned_loss=0.1426, over 26527.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3168, pruned_loss=0.08146, over 5708095.50 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3385, pruned_loss=0.0849, over 4358422.13 frames. ], giga_tot_loss[loss=0.2378, simple_loss=0.3132, pruned_loss=0.08118, over 5712340.23 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:24:36,880 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1233604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:24:43,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1233613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:24:56,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.661e+02 1.150e+03 1.527e+03 2.049e+03 6.161e+03, threshold=3.055e+03, percent-clipped=8.0
+2023-03-14 07:25:05,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1233639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:25:12,676 INFO [train.py:968] (0/2) Epoch 28, batch 2750, libri_loss[loss=0.3221, simple_loss=0.3908, pruned_loss=0.1267, over 29548.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3207, pruned_loss=0.08399, over 5704221.33 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3392, pruned_loss=0.08517, over 4392929.94 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.3168, pruned_loss=0.08349, over 5713339.73 frames. ], batch size: 82, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:25:18,454 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3047, 4.1388, 3.9820, 1.7166], device='cuda:0'), covar=tensor([0.0746, 0.0914, 0.1063, 0.2186], device='cuda:0'), in_proj_covar=tensor([0.1280, 0.1182, 0.0998, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 07:25:46,621 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6922, 2.0280, 1.8854, 1.3813], device='cuda:0'), covar=tensor([0.3899, 0.2776, 0.2975, 0.3741], device='cuda:0'), in_proj_covar=tensor([0.2040, 0.1996, 0.1915, 0.2054], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 07:25:57,517 INFO [train.py:968] (0/2) Epoch 28, batch 2800, giga_loss[loss=0.3624, simple_loss=0.4151, pruned_loss=0.1549, over 28604.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3263, pruned_loss=0.08783, over 5691928.10 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3394, pruned_loss=0.08512, over 4415285.08 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3229, pruned_loss=0.08747, over 5696763.50 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:26:23,486 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.160e+02 1.332e+03 1.667e+03 2.247e+03 4.681e+03, threshold=3.334e+03, percent-clipped=9.0
+2023-03-14 07:26:43,622 INFO [train.py:968] (0/2) Epoch 28, batch 2850, giga_loss[loss=0.2955, simple_loss=0.3662, pruned_loss=0.1124, over 28730.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3348, pruned_loss=0.09349, over 5686490.08 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3395, pruned_loss=0.08518, over 4422697.65 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.332, pruned_loss=0.09318, over 5689453.17 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:26:48,056 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4924, 1.8354, 1.4213, 1.6200], device='cuda:0'), covar=tensor([0.0776, 0.0305, 0.0332, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 07:26:50,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1233756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:26:52,388 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1233759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:13,253 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1233782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:15,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1233785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:16,897 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1233788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:24,545 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0307, 1.9539, 2.1627, 1.7689], device='cuda:0'), covar=tensor([0.1784, 0.2409, 0.1459, 0.1718], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0716, 0.0986, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 07:27:27,733 INFO [train.py:968] (0/2) Epoch 28, batch 2900, giga_loss[loss=0.258, simple_loss=0.3446, pruned_loss=0.08566, over 28724.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.34, pruned_loss=0.09544, over 5685755.10 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3397, pruned_loss=0.08515, over 4469871.82 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3375, pruned_loss=0.09549, over 5685091.84 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:27:43,522 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1233814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:55,469 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1233827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:55,957 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.120e+02 1.275e+03 1.574e+03 2.267e+03 5.153e+03, threshold=3.149e+03, percent-clipped=6.0
+2023-03-14 07:28:12,637 INFO [train.py:968] (0/2) Epoch 28, batch 2950, giga_loss[loss=0.2539, simple_loss=0.3446, pruned_loss=0.08157, over 28397.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3453, pruned_loss=0.098, over 5672766.98 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3399, pruned_loss=0.08539, over 4497333.42 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3433, pruned_loss=0.09809, over 5669008.35 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:28:56,901 INFO [train.py:968] (0/2) Epoch 28, batch 3000, giga_loss[loss=0.4153, simple_loss=0.455, pruned_loss=0.1877, over 27540.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3497, pruned_loss=0.09974, over 5684781.16 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3399, pruned_loss=0.08528, over 4541650.91 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3483, pruned_loss=0.1003, over 5678573.56 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:28:56,906 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 07:29:04,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2447, 1.8710, 1.4508, 0.4384], device='cuda:0'), covar=tensor([0.4961, 0.2965, 0.4988, 0.6890], device='cuda:0'), in_proj_covar=tensor([0.1829, 0.1720, 0.1647, 0.1496], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 07:29:05,375 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2082, simple_loss=0.3154, pruned_loss=0.05053, over 944034.00 frames. 
+2023-03-14 07:29:05,376 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 07:29:07,679 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9277, 4.9538, 2.0014, 2.3222], device='cuda:0'), covar=tensor([0.1004, 0.0315, 0.0907, 0.1210], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0566, 0.0407, 0.0444], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 07:29:30,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.979e+02 1.350e+03 1.745e+03 2.410e+03 4.625e+03, threshold=3.490e+03, percent-clipped=9.0
+2023-03-14 07:29:47,547 INFO [train.py:968] (0/2) Epoch 28, batch 3050, giga_loss[loss=0.2397, simple_loss=0.3253, pruned_loss=0.07707, over 28794.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3501, pruned_loss=0.0999, over 5676172.08 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3403, pruned_loss=0.08565, over 4568285.87 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.349, pruned_loss=0.1004, over 5667369.40 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:30:11,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1233979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:30:27,032 INFO [train.py:968] (0/2) Epoch 28, batch 3100, giga_loss[loss=0.2739, simple_loss=0.3521, pruned_loss=0.09787, over 28659.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3454, pruned_loss=0.09648, over 5679546.93 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.34, pruned_loss=0.08581, over 4606349.73 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3449, pruned_loss=0.09716, over 5672877.68 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:30:27,732 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1234000.pt
+2023-03-14 07:30:50,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.718e+02 1.358e+03 1.687e+03 2.221e+03 6.726e+03, threshold=3.374e+03, percent-clipped=6.0
+2023-03-14 07:31:09,861 INFO [train.py:968] (0/2) Epoch 28, batch 3150, giga_loss[loss=0.2556, simple_loss=0.3348, pruned_loss=0.08822, over 28667.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3443, pruned_loss=0.09532, over 5668487.55 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3402, pruned_loss=0.08608, over 4607441.66 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3437, pruned_loss=0.09573, over 5671150.34 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:31:51,509 INFO [train.py:968] (0/2) Epoch 28, batch 3200, giga_loss[loss=0.2696, simple_loss=0.341, pruned_loss=0.09906, over 28090.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3437, pruned_loss=0.09482, over 5671322.82 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.34, pruned_loss=0.08592, over 4633084.79 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3435, pruned_loss=0.09542, over 5669476.29 frames. ], batch size: 77, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:31:56,624 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:32:02,995 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3145, 1.1669, 1.1003, 1.4449], device='cuda:0'), covar=tensor([0.0820, 0.0399, 0.0368, 0.0900], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 07:32:07,826 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1234122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:32:11,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1234125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:32:13,081 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.011e+02 1.391e+03 1.811e+03 2.353e+03 9.473e+03, threshold=3.623e+03, percent-clipped=13.0
+2023-03-14 07:32:29,965 INFO [train.py:968] (0/2) Epoch 28, batch 3250, giga_loss[loss=0.2812, simple_loss=0.3622, pruned_loss=0.1001, over 28928.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3449, pruned_loss=0.09508, over 5682494.51 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3393, pruned_loss=0.08558, over 4676911.82 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3454, pruned_loss=0.09609, over 5674173.47 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:32:35,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1234154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:32:58,588 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4762, 1.7056, 1.6295, 1.4625], device='cuda:0'), covar=tensor([0.2146, 0.2364, 0.2427, 0.2300], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0761, 0.0732, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 07:33:11,484 INFO [train.py:968] (0/2) Epoch 28, batch 3300, giga_loss[loss=0.304, simple_loss=0.3719, pruned_loss=0.1181, over 28677.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3475, pruned_loss=0.09705, over 5682300.71 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3396, pruned_loss=0.08579, over 4690608.98 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3478, pruned_loss=0.09787, over 5681828.67 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:33:14,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1234202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:33:34,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.894e+02 1.369e+03 1.690e+03 2.392e+03 4.442e+03, threshold=3.380e+03, percent-clipped=4.0
+2023-03-14 07:33:50,637 INFO [train.py:968] (0/2) Epoch 28, batch 3350, giga_loss[loss=0.2905, simple_loss=0.3584, pruned_loss=0.1113, over 28589.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3484, pruned_loss=0.09805, over 5682076.36 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3392, pruned_loss=0.08564, over 4728163.99 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3492, pruned_loss=0.0992, over 5678708.03 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:34:02,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2839, 1.3281, 1.3338, 1.2134], device='cuda:0'), covar=tensor([0.2424, 0.2451, 0.2360, 0.2644], device='cuda:0'), in_proj_covar=tensor([0.2043, 0.2004, 0.1918, 0.2059], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 07:34:03,972 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234265.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 07:34:25,977 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6860, 1.8718, 1.7423, 1.6424], device='cuda:0'), covar=tensor([0.2279, 0.2446, 0.2580, 0.2277], device='cuda:0'), in_proj_covar=tensor([0.0507, 0.0761, 0.0731, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 07:34:33,854 INFO [train.py:968] (0/2) Epoch 28, batch 3400, giga_loss[loss=0.2646, simple_loss=0.3407, pruned_loss=0.09428, over 28852.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3493, pruned_loss=0.09922, over 5686054.47 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3393, pruned_loss=0.08564, over 4745879.06 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3501, pruned_loss=0.1003, over 5680170.94 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:34:35,631 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4800, 4.3428, 4.0994, 2.0536], device='cuda:0'), covar=tensor([0.0609, 0.0750, 0.0705, 0.2006], device='cuda:0'), in_proj_covar=tensor([0.1279, 0.1185, 0.0998, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 07:34:58,083 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.969e+02 1.377e+03 1.616e+03 2.171e+03 6.622e+03, threshold=3.231e+03, percent-clipped=4.0
+2023-03-14 07:35:12,060 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1234345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:35:14,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1234348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:35:14,427 INFO [train.py:968] (0/2) Epoch 28, batch 3450, libri_loss[loss=0.2713, simple_loss=0.3596, pruned_loss=0.09148, over 26107.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3508, pruned_loss=0.1007, over 5680041.49 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3404, pruned_loss=0.08621, over 4773681.35 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3508, pruned_loss=0.1016, over 5675560.98 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:35:35,374 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1234377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:35:44,481 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 07:35:51,549 INFO [train.py:968] (0/2) Epoch 28, batch 3500, giga_loss[loss=0.3082, simple_loss=0.383, pruned_loss=0.1167, over 28591.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3514, pruned_loss=0.1006, over 5682796.41 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3406, pruned_loss=0.08641, over 4800087.51 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3515, pruned_loss=0.1015, over 5679491.18 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:35:59,588 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4769, 1.6458, 1.5031, 1.3044], device='cuda:0'), covar=tensor([0.2787, 0.2696, 0.2159, 0.2773], device='cuda:0'), in_proj_covar=tensor([0.2044, 0.2004, 0.1918, 0.2060], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 07:36:14,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.756e+02 1.309e+03 1.638e+03 2.340e+03 5.761e+03, threshold=3.275e+03, percent-clipped=5.0
+2023-03-14 07:36:30,202 INFO [train.py:968] (0/2) Epoch 28, batch 3550, giga_loss[loss=0.269, simple_loss=0.3495, pruned_loss=0.09429, over 28522.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3502, pruned_loss=0.09867, over 5695995.82 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3401, pruned_loss=0.08594, over 4851685.55 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3512, pruned_loss=0.1004, over 5685598.01 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:36:57,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1234481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:37:11,874 INFO [train.py:968] (0/2) Epoch 28, batch 3600, giga_loss[loss=0.2556, simple_loss=0.338, pruned_loss=0.08661, over 28679.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3502, pruned_loss=0.09797, over 5698068.46 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3403, pruned_loss=0.0862, over 4881796.22 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3511, pruned_loss=0.0995, over 5685259.65 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:37:12,092 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8590, 1.0601, 2.8585, 2.7775], device='cuda:0'), covar=tensor([0.1756, 0.2784, 0.0589, 0.1040], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0667, 0.0993, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 07:37:35,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.309e+02 1.185e+03 1.425e+03 1.908e+03 5.453e+03, threshold=2.851e+03, percent-clipped=6.0
+2023-03-14 07:37:50,298 INFO [train.py:968] (0/2) Epoch 28, batch 3650, giga_loss[loss=0.2545, simple_loss=0.3324, pruned_loss=0.08833, over 29013.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3491, pruned_loss=0.09721, over 5704618.56 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3404, pruned_loss=0.08619, over 4896474.18 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3499, pruned_loss=0.09858, over 5692388.04 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:38:17,167 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-14 07:38:29,245 INFO [train.py:968] (0/2) Epoch 28, batch 3700, giga_loss[loss=0.2657, simple_loss=0.3383, pruned_loss=0.09659, over 28789.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.347, pruned_loss=0.09672, over 5698834.98 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3403, pruned_loss=0.08618, over 4925619.32 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.348, pruned_loss=0.09814, over 5684214.32 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:38:49,179 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1234624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:38:49,759 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:38:51,229 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1234627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:38:51,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.842e+02 1.173e+03 1.423e+03 1.761e+03 4.453e+03, threshold=2.847e+03, percent-clipped=6.0
+2023-03-14 07:38:59,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1234640.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 07:39:01,161 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:39:07,482 INFO [train.py:968] (0/2) Epoch 28, batch 3750, giga_loss[loss=0.2565, simple_loss=0.3144, pruned_loss=0.09933, over 23501.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3448, pruned_loss=0.09561, over 5708847.67 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3409, pruned_loss=0.08659, over 4944824.83 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3452, pruned_loss=0.0966, over 5693943.03 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:39:13,319 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1234656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:39:40,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6152, 1.6252, 1.8335, 1.4111], device='cuda:0'), covar=tensor([0.1831, 0.2719, 0.1494, 0.1831], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0718, 0.0986, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 07:39:48,929 INFO [train.py:968] (0/2) Epoch 28, batch 3800, giga_loss[loss=0.2608, simple_loss=0.3372, pruned_loss=0.09223, over 28902.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3441, pruned_loss=0.09556, over 5702926.11 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3409, pruned_loss=0.08656, over 4949673.70 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3445, pruned_loss=0.09639, over 5690350.10 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:40:10,584 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.949e+02 1.138e+03 1.394e+03 1.704e+03 4.316e+03, threshold=2.787e+03, percent-clipped=6.0
+2023-03-14 07:40:12,172 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:40:19,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4385, 2.0772, 1.5926, 0.7064], device='cuda:0'), covar=tensor([0.6714, 0.3508, 0.4689, 0.7436], device='cuda:0'), in_proj_covar=tensor([0.1821, 0.1708, 0.1638, 0.1489], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 07:40:27,392 INFO [train.py:968] (0/2) Epoch 28, batch 3850, libri_loss[loss=0.2544, simple_loss=0.341, pruned_loss=0.0839, over 29401.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3453, pruned_loss=0.09621, over 5706505.36 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3413, pruned_loss=0.08673, over 4979800.82 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3454, pruned_loss=0.09708, over 5692680.77 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:40:53,000 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1234783.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 07:40:55,048 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1234786.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 07:41:06,923 INFO [train.py:968] (0/2) Epoch 28, batch 3900, giga_loss[loss=0.2866, simple_loss=0.3615, pruned_loss=0.1059, over 28367.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.345, pruned_loss=0.09517, over 5713048.73 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3415, pruned_loss=0.08691, over 5007302.89 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.345, pruned_loss=0.09597, over 5696814.55 frames. ], batch size: 77, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:41:14,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7124, 1.9885, 1.6195, 1.8713], device='cuda:0'), covar=tensor([0.2745, 0.2664, 0.3051, 0.2607], device='cuda:0'), in_proj_covar=tensor([0.1594, 0.1150, 0.1405, 0.1010], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 07:41:20,740 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1234815.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 07:41:30,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.166e+02 1.130e+03 1.366e+03 1.834e+03 3.987e+03, threshold=2.732e+03, percent-clipped=5.0
+2023-03-14 07:41:49,217 INFO [train.py:968] (0/2) Epoch 28, batch 3950, giga_loss[loss=0.3053, simple_loss=0.375, pruned_loss=0.1177, over 28397.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3439, pruned_loss=0.09393, over 5717428.57 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3415, pruned_loss=0.08693, over 5020950.45 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3439, pruned_loss=0.09465, over 5701935.22 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:42:13,840 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:42:14,763 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-14 07:42:28,996 INFO [train.py:968] (0/2) Epoch 28, batch 4000, giga_loss[loss=0.2253, simple_loss=0.3196, pruned_loss=0.06549, over 28626.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3433, pruned_loss=0.09397, over 5709716.38 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3413, pruned_loss=0.08687, over 5034454.27 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3436, pruned_loss=0.09471, over 5697571.33 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:42:51,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.102e+02 1.053e+03 1.284e+03 1.643e+03 5.581e+03, threshold=2.567e+03, percent-clipped=4.0
+2023-03-14 07:42:58,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7804, 2.0935, 1.4968, 1.6358], device='cuda:0'), covar=tensor([0.1104, 0.0617, 0.1023, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0452, 0.0525, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 07:42:59,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9481, 1.2987, 1.0977, 0.2204], device='cuda:0'), covar=tensor([0.5024, 0.3467, 0.5359, 0.7445], device='cuda:0'), in_proj_covar=tensor([0.1814, 0.1699, 0.1630, 0.1484], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 07:43:08,898 INFO [train.py:968] (0/2) Epoch 28, batch 4050, giga_loss[loss=0.2727, simple_loss=0.3453, pruned_loss=0.1, over 28908.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.343, pruned_loss=0.09437, over 5715166.40 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3418, pruned_loss=0.0872, over 5047436.18 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3428, pruned_loss=0.09478, over 5702941.06 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:43:26,974 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4244, 1.7388, 1.6639, 1.3473], device='cuda:0'), covar=tensor([0.3169, 0.2476, 0.1778, 0.2635], device='cuda:0'), in_proj_covar=tensor([0.2037, 0.2001, 0.1909, 0.2052], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 07:43:33,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6201, 1.8026, 1.2607, 1.4633], device='cuda:0'), covar=tensor([0.0919, 0.0540, 0.0938, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0451, 0.0524, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 07:43:46,630 INFO [train.py:968] (0/2) Epoch 28, batch 4100, giga_loss[loss=0.257, simple_loss=0.3274, pruned_loss=0.09324, over 28857.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3398, pruned_loss=0.09239, over 5714454.23 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3416, pruned_loss=0.08705, over 5076563.34 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3398, pruned_loss=0.09309, over 5704802.44 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:43:47,574 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:43:48,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5302, 1.5856, 1.7547, 1.3475], device='cuda:0'), covar=tensor([0.1955, 0.2711, 0.1620, 0.1937], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0720, 0.0987, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 07:44:00,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:44:07,894 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.702e+02 1.250e+03 1.650e+03 2.191e+03 4.400e+03, threshold=3.301e+03, percent-clipped=18.0
+2023-03-14 07:44:25,357 INFO [train.py:968] (0/2) Epoch 28, batch 4150, giga_loss[loss=0.2299, simple_loss=0.2983, pruned_loss=0.08076, over 23260.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3365, pruned_loss=0.09083, over 5714004.16 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.341, pruned_loss=0.08676, over 5092340.00 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3369, pruned_loss=0.09174, over 5706369.83 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:45:06,759 INFO [train.py:968] (0/2) Epoch 28, batch 4200, giga_loss[loss=0.2233, simple_loss=0.3073, pruned_loss=0.06964, over 28339.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3366, pruned_loss=0.09136, over 5713388.58 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3408, pruned_loss=0.08667, over 5100509.07 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.337, pruned_loss=0.09218, over 5705612.94 frames. ], batch size: 65, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:45:11,714 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:45:15,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1396, 3.2493, 1.3270, 1.4050], device='cuda:0'), covar=tensor([0.1171, 0.0355, 0.1015, 0.1576], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0563, 0.0406, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 07:45:30,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.941e+02 1.250e+03 1.569e+03 2.028e+03 5.293e+03, threshold=3.137e+03, percent-clipped=5.0
+2023-03-14 07:45:42,627 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:45:43,899 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9896, 5.0687, 2.1623, 2.1426], device='cuda:0'), covar=tensor([0.0927, 0.0429, 0.0846, 0.1237], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0564, 0.0406, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 07:45:44,581 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:45:47,258 INFO [train.py:968] (0/2) Epoch 28, batch 4250, giga_loss[loss=0.2364, simple_loss=0.3128, pruned_loss=0.07999, over 28573.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3354, pruned_loss=0.09095, over 5714981.93 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3405, pruned_loss=0.08639, over 5122940.83 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3358, pruned_loss=0.09197, over 5705043.08 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:45:55,938 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:45:59,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:46:10,775 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-14 07:46:11,277 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:46:15,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4694, 4.3175, 4.0978, 1.9418], device='cuda:0'), covar=tensor([0.0570, 0.0730, 0.0768, 0.2070], device='cuda:0'), in_proj_covar=tensor([0.1285, 0.1189, 0.1000, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 07:46:24,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:46:30,924 INFO [train.py:968] (0/2) Epoch 28, batch 4300, giga_loss[loss=0.2967, simple_loss=0.362, pruned_loss=0.1157, over 28312.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3343, pruned_loss=0.09107, over 5711157.30 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3405, pruned_loss=0.08644, over 5134785.62 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3345, pruned_loss=0.09188, over 5700677.35 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:46:40,671 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1235210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:46:54,320 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.457e+02 1.109e+03 1.401e+03 1.808e+03 3.522e+03, threshold=2.803e+03, percent-clipped=3.0
+2023-03-14 07:47:07,932 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-14 07:47:09,799 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:47:11,663 INFO [train.py:968] (0/2) Epoch 28, batch 4350, giga_loss[loss=0.218, simple_loss=0.2938, pruned_loss=0.07105, over 28646.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3311, pruned_loss=0.08943, over 5715781.48 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3407, pruned_loss=0.08652, over 5149681.46 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3311, pruned_loss=0.09009, over 5704372.60 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:47:11,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:47:15,634 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:47:23,490 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 07:47:34,041 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:47:36,161 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1235281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:47:50,659 INFO [train.py:968] (0/2) Epoch 28, batch 4400, giga_loss[loss=0.2455, simple_loss=0.3158, pruned_loss=0.08763, over 28739.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3292, pruned_loss=0.0887, over 5717350.08 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3405, pruned_loss=0.0864, over 5164756.06 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3291, pruned_loss=0.08936, over 5704854.13 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:47:56,639 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1235306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:48:08,819 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 07:48:15,326 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.211e+02 1.131e+03 1.393e+03 1.912e+03 4.464e+03, threshold=2.785e+03, percent-clipped=5.0
+2023-03-14 07:48:29,684 INFO [train.py:968] (0/2) Epoch 28, batch 4450, giga_loss[loss=0.2577, simple_loss=0.3364, pruned_loss=0.08952, over 28890.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3302, pruned_loss=0.08904, over 5711839.09 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3408, pruned_loss=0.0865, over 5176768.65 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3295, pruned_loss=0.08955, over 5706227.21 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:49:07,495 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:49:08,636 INFO [train.py:968] (0/2) Epoch 28, batch 4500, giga_loss[loss=0.3335, simple_loss=0.3969, pruned_loss=0.135, over 27599.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3326, pruned_loss=0.09, over 5712456.50 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3408, pruned_loss=0.0865, over 5207146.24 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3317, pruned_loss=0.09053, over 5701448.63 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:49:11,139 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:49:18,664 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6750, 1.9922, 1.5326, 1.6080], device='cuda:0'), covar=tensor([0.0708, 0.0259, 0.0339, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 07:49:23,565 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.44 vs. limit=5.0
+2023-03-14 07:49:36,148 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:49:37,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.267e+02 1.146e+03 1.499e+03 1.871e+03 7.159e+03, threshold=2.998e+03, percent-clipped=7.0
+2023-03-14 07:49:51,832 INFO [train.py:968] (0/2) Epoch 28, batch 4550, giga_loss[loss=0.2767, simple_loss=0.3595, pruned_loss=0.09692, over 28765.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3346, pruned_loss=0.09022, over 5724145.12 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3408, pruned_loss=0.08671, over 5229804.37 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3335, pruned_loss=0.09056, over 5710678.83 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:49:55,783 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9797, 1.2935, 1.0858, 0.2614], device='cuda:0'), covar=tensor([0.4829, 0.3600, 0.5275, 0.7663], device='cuda:0'), in_proj_covar=tensor([0.1830, 0.1715, 0.1648, 0.1498], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 07:50:28,936 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1235496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:50:30,698 INFO [train.py:968] (0/2) Epoch 28, batch 4600, libri_loss[loss=0.2797, simple_loss=0.3574, pruned_loss=0.1009, over 19771.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3367, pruned_loss=0.09104, over 5706008.21 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3407, pruned_loss=0.08689, over 5230033.04 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3358, pruned_loss=0.09129, over 5711171.50 frames. ], batch size: 187, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:50:58,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.808e+02 1.258e+03 1.600e+03 2.516e+03 4.978e+03, threshold=3.199e+03, percent-clipped=16.0
+2023-03-14 07:51:12,505 INFO [train.py:968] (0/2) Epoch 28, batch 4650, giga_loss[loss=0.3047, simple_loss=0.3707, pruned_loss=0.1194, over 28964.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3383, pruned_loss=0.09143, over 5702317.27 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3408, pruned_loss=0.08699, over 5249463.03 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3374, pruned_loss=0.09169, over 5703311.68 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:51:44,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:51:56,339 INFO [train.py:968] (0/2) Epoch 28, batch 4700, giga_loss[loss=0.255, simple_loss=0.3323, pruned_loss=0.08885, over 28874.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3383, pruned_loss=0.09102, over 5692813.91 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.341, pruned_loss=0.08715, over 5259076.29 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3374, pruned_loss=0.09115, over 5691012.10 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:52:21,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.628e+02 1.237e+03 1.503e+03 1.917e+03 3.907e+03, threshold=3.007e+03, percent-clipped=6.0
+2023-03-14 07:52:35,785 INFO [train.py:968] (0/2) Epoch 28, batch 4750, giga_loss[loss=0.2599, simple_loss=0.3362, pruned_loss=0.09177, over 29075.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3389, pruned_loss=0.09129, over 5698266.57 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3416, pruned_loss=0.08744, over 5268564.97 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3377, pruned_loss=0.09125, over 5697800.28 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:52:41,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:53:03,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:53:17,052 INFO [train.py:968] (0/2) Epoch 28, batch 4800, giga_loss[loss=0.2313, simple_loss=0.3179, pruned_loss=0.07233, over 29094.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3395, pruned_loss=0.09193, over 5703975.61 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3417, pruned_loss=0.08755, over 5271894.14 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3384, pruned_loss=0.09182, over 5702495.70 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:53:33,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4207, 1.6646, 1.3552, 1.3165], device='cuda:0'), covar=tensor([0.1047, 0.0451, 0.0999, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0449, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 07:53:40,580 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:53:43,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.585e+02 1.241e+03 1.458e+03 1.854e+03 4.287e+03, threshold=2.916e+03, percent-clipped=1.0
+2023-03-14 07:53:43,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:53:44,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1291, 5.1306, 2.1543, 2.1983], device='cuda:0'), covar=tensor([0.0832, 0.0340, 0.0834, 0.1113], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0563, 0.0406, 0.0441], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 07:54:00,756 INFO [train.py:968] (0/2) Epoch 28, batch 4850, giga_loss[loss=0.2684, simple_loss=0.3361, pruned_loss=0.1003, over 28712.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.341, pruned_loss=0.09364, over 5703498.55 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3418, pruned_loss=0.0877, over 5279445.66 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3401, pruned_loss=0.09351, over 5701380.06 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:54:07,811 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:54:39,114 INFO [train.py:968] (0/2) Epoch 28, batch 4900, giga_loss[loss=0.2496, simple_loss=0.3369, pruned_loss=0.08114, over 28967.00 frames. ], tot_loss[loss=0.267, simple_loss=0.344, pruned_loss=0.09504, over 5707412.58 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3419, pruned_loss=0.08773, over 5285351.48 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3432, pruned_loss=0.09499, over 5706011.98 frames. ], batch size: 174, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:54:39,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3307, 1.8220, 1.4332, 1.4792], device='cuda:0'), covar=tensor([0.0757, 0.0296, 0.0339, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 07:54:39,539 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:54:41,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:54:59,105 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:55:01,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:55:03,826 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.393e+03 1.724e+03 2.258e+03 9.441e+03, threshold=3.449e+03, percent-clipped=12.0
+2023-03-14 07:55:04,113 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:55:14,325 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-14 07:55:17,327 INFO [train.py:968] (0/2) Epoch 28, batch 4950, giga_loss[loss=0.2799, simple_loss=0.3608, pruned_loss=0.0995, over 28929.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3467, pruned_loss=0.09618, over 5714346.74 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3421, pruned_loss=0.08778, over 5309142.59 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3459, pruned_loss=0.0964, over 5706538.97 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:55:23,459 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:55:33,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:55:56,175 INFO [train.py:968] (0/2) Epoch 28, batch 5000, giga_loss[loss=0.3729, simple_loss=0.4136, pruned_loss=0.1661, over 26646.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3493, pruned_loss=0.09783, over 5715100.62 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3426, pruned_loss=0.08795, over 5322079.58 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3484, pruned_loss=0.09807, over 5706306.69 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:55:58,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5060, 1.7782, 1.4749, 1.5759], device='cuda:0'), covar=tensor([0.2408, 0.2529, 0.2739, 0.2382], device='cuda:0'), in_proj_covar=tensor([0.1592, 0.1147, 0.1402, 0.1010], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 07:56:20,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.416e+03 1.784e+03 2.246e+03 3.661e+03, threshold=3.567e+03, percent-clipped=5.0
+2023-03-14 07:56:35,023 INFO [train.py:968] (0/2) Epoch 28, batch 5050, giga_loss[loss=0.2747, simple_loss=0.344, pruned_loss=0.1027, over 28982.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3493, pruned_loss=0.09766, over 5710407.18 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3427, pruned_loss=0.08792, over 5331878.61 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3487, pruned_loss=0.09804, over 5701664.27 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:57:16,684 INFO [train.py:968] (0/2) Epoch 28, batch 5100, giga_loss[loss=0.2372, simple_loss=0.3317, pruned_loss=0.0713, over 28984.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3484, pruned_loss=0.09691, over 5715128.20 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3426, pruned_loss=0.08783, over 5346076.99 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3481, pruned_loss=0.09751, over 5703757.54 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:57:17,424 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1236000.pt
+2023-03-14 07:57:29,575 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1236014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:57:32,000 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1236017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:57:43,879 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.914e+02 1.313e+03 1.623e+03 2.189e+03 5.136e+03, threshold=3.247e+03, percent-clipped=5.0
+2023-03-14 07:57:56,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1236046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:57:58,159 INFO [train.py:968] (0/2) Epoch 28, batch 5150, giga_loss[loss=0.2311, simple_loss=0.31, pruned_loss=0.07608, over 28550.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3476, pruned_loss=0.09686, over 5716622.48 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.343, pruned_loss=0.08801, over 5354454.08 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3472, pruned_loss=0.0973, over 5705146.77 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:58:29,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5584, 1.6534, 1.4632, 1.6264], device='cuda:0'), covar=tensor([0.0734, 0.0318, 0.0337, 0.0894], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 07:58:35,273 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4558, 1.7775, 1.4700, 1.5880], device='cuda:0'), covar=tensor([0.0743, 0.0300, 0.0331, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 07:58:40,399 INFO [train.py:968] (0/2) Epoch 28, batch 5200, giga_loss[loss=0.2234, simple_loss=0.3126, pruned_loss=0.06711, over 29072.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3444, pruned_loss=0.09581, over 5711779.79 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3431, pruned_loss=0.08805, over 5359033.00 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3441, pruned_loss=0.09625, over 5703387.78 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:59:01,352 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5720, 4.7867, 1.7261, 1.8943], device='cuda:0'), covar=tensor([0.0966, 0.0314, 0.0929, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0564, 0.0408, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 07:59:05,121 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.889e+02 1.221e+03 1.446e+03 1.734e+03 3.726e+03, threshold=2.893e+03, percent-clipped=2.0
+2023-03-14 07:59:19,757 INFO [train.py:968] (0/2) Epoch 28, batch 5250, giga_loss[loss=0.2411, simple_loss=0.3262, pruned_loss=0.07799, over 28878.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.342, pruned_loss=0.09431, over 5708149.56 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3435, pruned_loss=0.08832, over 5369047.43 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3412, pruned_loss=0.09465, over 5705511.80 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:59:56,511 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-14 07:59:56,540 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-14 07:59:59,220 INFO [train.py:968] (0/2) Epoch 28, batch 5300, giga_loss[loss=0.3225, simple_loss=0.3951, pruned_loss=0.125, over 27706.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.343, pruned_loss=0.09394, over 5696869.21 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3442, pruned_loss=0.08877, over 5374777.68 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3417, pruned_loss=0.09402, over 5703861.16 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:00:02,432 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1236203.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 08:00:27,906 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.438e+02 1.243e+03 1.593e+03 2.295e+03 6.347e+03, threshold=3.185e+03, percent-clipped=11.0
+2023-03-14 08:00:40,982 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1236245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:00:44,080 INFO [train.py:968] (0/2) Epoch 28, batch 5350, giga_loss[loss=0.2626, simple_loss=0.3527, pruned_loss=0.08623, over 27936.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3448, pruned_loss=0.09353, over 5700467.24 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.344, pruned_loss=0.08878, over 5375926.79 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3439, pruned_loss=0.09361, over 5706333.66 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:00:56,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6803, 1.7088, 1.8750, 1.4419], device='cuda:0'), covar=tensor([0.2122, 0.2442, 0.1731, 0.1896], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0715, 0.0981, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 08:01:24,077 INFO [train.py:968] (0/2) Epoch 28, batch 5400, giga_loss[loss=0.2681, simple_loss=0.3449, pruned_loss=0.09566, over 28935.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3447, pruned_loss=0.09357, over 5705516.20 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3447, pruned_loss=0.08902, over 5385902.52 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3434, pruned_loss=0.09355, over 5710313.17 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:01:48,243 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-14 08:01:48,878 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-14 08:01:49,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.053e+02 1.301e+03 1.567e+03 1.914e+03 6.575e+03, threshold=3.134e+03, percent-clipped=5.0
+2023-03-14 08:01:59,022 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1236343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:02:03,782 INFO [train.py:968] (0/2) Epoch 28, batch 5450, giga_loss[loss=0.2618, simple_loss=0.3395, pruned_loss=0.09205, over 28700.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3427, pruned_loss=0.09393, over 5708031.96 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3448, pruned_loss=0.08913, over 5388900.97 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3415, pruned_loss=0.09392, over 5715487.68 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:02:32,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.64 vs. limit=5.0
+2023-03-14 08:02:41,983 INFO [train.py:968] (0/2) Epoch 28, batch 5500, libri_loss[loss=0.2771, simple_loss=0.3593, pruned_loss=0.09746, over 29661.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3417, pruned_loss=0.09453, over 5712154.70 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3452, pruned_loss=0.0893, over 5399166.99 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3403, pruned_loss=0.09455, over 5720059.17 frames. ], batch size: 88, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:03:07,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.878e+02 1.383e+03 1.642e+03 2.295e+03 8.602e+03, threshold=3.284e+03, percent-clipped=8.0
+2023-03-14 08:03:21,253 INFO [train.py:968] (0/2) Epoch 28, batch 5550, giga_loss[loss=0.307, simple_loss=0.3704, pruned_loss=0.1218, over 28697.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3393, pruned_loss=0.09441, over 5718077.93 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3453, pruned_loss=0.08945, over 5409421.36 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.338, pruned_loss=0.09444, over 5722209.87 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:04:03,526 INFO [train.py:968] (0/2) Epoch 28, batch 5600, libri_loss[loss=0.2905, simple_loss=0.3651, pruned_loss=0.1079, over 20008.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3377, pruned_loss=0.0941, over 5709187.92 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3453, pruned_loss=0.08955, over 5404728.36 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3366, pruned_loss=0.09407, over 5720314.56 frames. ], batch size: 188, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 08:04:16,063 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3172, 1.1806, 1.1041, 1.5027], device='cuda:0'), covar=tensor([0.0759, 0.0369, 0.0368, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 08:04:32,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.797e+02 1.300e+03 1.504e+03 2.077e+03 3.909e+03, threshold=3.009e+03, percent-clipped=4.0
+2023-03-14 08:04:48,677 INFO [train.py:968] (0/2) Epoch 28, batch 5650, giga_loss[loss=0.2699, simple_loss=0.3333, pruned_loss=0.1032, over 28791.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3344, pruned_loss=0.09271, over 5702970.67 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3452, pruned_loss=0.08951, over 5407276.72 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3336, pruned_loss=0.09274, over 5710871.59 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 08:04:55,451 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7867, 2.7925, 1.8859, 1.0745], device='cuda:0'), covar=tensor([1.1097, 0.4184, 0.4743, 0.8766], device='cuda:0'), in_proj_covar=tensor([0.1826, 0.1710, 0.1644, 0.1494], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 08:05:12,557 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1236578.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:05:15,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1236583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:05:27,474 INFO [train.py:968] (0/2) Epoch 28, batch 5700, libri_loss[loss=0.262, simple_loss=0.3496, pruned_loss=0.08718, over 29650.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3305, pruned_loss=0.09073, over 5710330.98 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.345, pruned_loss=0.08935, over 5419922.06 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3298, pruned_loss=0.09095, over 5711692.07 frames. ], batch size: 88, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:05:45,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1236620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:05:54,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.964e+02 1.322e+03 1.662e+03 2.195e+03 1.003e+04, threshold=3.325e+03, percent-clipped=10.0
+2023-03-14 08:06:07,870 INFO [train.py:968] (0/2) Epoch 28, batch 5750, giga_loss[loss=0.2554, simple_loss=0.3275, pruned_loss=0.09164, over 28821.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.328, pruned_loss=0.08976, over 5709974.36 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3445, pruned_loss=0.08901, over 5431901.88 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3275, pruned_loss=0.09022, over 5706842.50 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:06:14,167 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0539, 2.2685, 1.5955, 1.9114], device='cuda:0'), covar=tensor([0.1118, 0.0827, 0.1435, 0.1322], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0450, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 08:06:24,239 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1236670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:06:46,443 INFO [train.py:968] (0/2) Epoch 28, batch 5800, giga_loss[loss=0.2525, simple_loss=0.3358, pruned_loss=0.08456, over 28641.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3285, pruned_loss=0.08971, over 5712908.08 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3447, pruned_loss=0.08904, over 5440562.79 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3275, pruned_loss=0.09008, over 5708717.60 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:07:01,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1236718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:07:03,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1236721.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:07:05,721 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1236724.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:07:13,190 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.251e+02 1.283e+03 1.534e+03 2.080e+03 4.810e+03, threshold=3.067e+03, percent-clipped=6.0
+2023-03-14 08:07:27,571 INFO [train.py:968] (0/2) Epoch 28, batch 5850, giga_loss[loss=0.2745, simple_loss=0.3501, pruned_loss=0.09942, over 29058.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3311, pruned_loss=0.09065, over 5713996.60 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3444, pruned_loss=0.08886, over 5447472.18 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3304, pruned_loss=0.09111, over 5708246.69 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:07:30,290 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1236753.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:07:37,776 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1236763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:07:40,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1236766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:08:04,867 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1236795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:08:07,239 INFO [train.py:968] (0/2) Epoch 28, batch 5900, giga_loss[loss=0.3153, simple_loss=0.3773, pruned_loss=0.1266, over 28503.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3346, pruned_loss=0.09204, over 5705670.81 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3442, pruned_loss=0.08877, over 5452328.46 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3339, pruned_loss=0.09256, over 5705529.71 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:08:34,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.836e+02 1.456e+03 1.783e+03 2.492e+03 9.061e+03, threshold=3.567e+03, percent-clipped=16.0
+2023-03-14 08:08:48,565 INFO [train.py:968] (0/2) Epoch 28, batch 5950, giga_loss[loss=0.2524, simple_loss=0.3386, pruned_loss=0.08314, over 29040.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3382, pruned_loss=0.09322, over 5712268.34 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3442, pruned_loss=0.08877, over 5461302.35 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3374, pruned_loss=0.09373, over 5709620.22 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:08:49,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4534, 1.7269, 1.1897, 1.2795], device='cuda:0'), covar=tensor([0.1024, 0.0542, 0.1064, 0.1203], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0451, 0.0523, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 08:08:58,731 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1236861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:09:01,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1236864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:09:26,175 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1236893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:09:30,181 INFO [train.py:968] (0/2) Epoch 28, batch 6000, giga_loss[loss=0.2719, simple_loss=0.3505, pruned_loss=0.09666, over 28820.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.341, pruned_loss=0.09462, over 5708317.76 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3444, pruned_loss=0.08897, over 5465016.46 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3401, pruned_loss=0.09496, over 5707148.85 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 08:09:30,185 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 08:09:38,311 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2075, simple_loss=0.3151, pruned_loss=0.04996, over 944034.00 frames. 
+2023-03-14 08:09:38,312 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 08:10:09,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.996e+02 1.338e+03 1.687e+03 2.220e+03 5.324e+03, threshold=3.373e+03, percent-clipped=8.0
+2023-03-14 08:10:24,670 INFO [train.py:968] (0/2) Epoch 28, batch 6050, giga_loss[loss=0.2468, simple_loss=0.3268, pruned_loss=0.08346, over 28347.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3441, pruned_loss=0.09682, over 5699391.16 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3447, pruned_loss=0.08911, over 5468135.09 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3432, pruned_loss=0.09703, over 5697836.70 frames. ], batch size: 65, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:10:33,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1236958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:10:57,521 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-14 08:11:10,148 INFO [train.py:968] (0/2) Epoch 28, batch 6100, giga_loss[loss=0.3201, simple_loss=0.3875, pruned_loss=0.1264, over 28998.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3513, pruned_loss=0.1033, over 5692004.77 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3446, pruned_loss=0.08917, over 5471361.18 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3507, pruned_loss=0.1035, over 5690762.54 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:11:40,782 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.823e+03 2.318e+03 3.276e+03 1.509e+04, threshold=4.637e+03, percent-clipped=22.0
+2023-03-14 08:11:53,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1237045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:11:56,143 INFO [train.py:968] (0/2) Epoch 28, batch 6150, giga_loss[loss=0.294, simple_loss=0.3636, pruned_loss=0.1122, over 28939.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3575, pruned_loss=0.1077, over 5701797.32 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3442, pruned_loss=0.08893, over 5484375.78 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3575, pruned_loss=0.1086, over 5694982.12 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:12:31,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5247, 1.6994, 1.2174, 1.2988], device='cuda:0'), covar=tensor([0.0995, 0.0582, 0.1067, 0.1131], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0451, 0.0523, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 08:12:42,697 INFO [train.py:968] (0/2) Epoch 28, batch 6200, giga_loss[loss=0.3794, simple_loss=0.4188, pruned_loss=0.17, over 28607.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3632, pruned_loss=0.1119, over 5693247.81 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3441, pruned_loss=0.08884, over 5490278.49 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3637, pruned_loss=0.1131, over 5687081.86 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:12:45,485 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1237101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:12:47,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1237104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:13:13,553 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1237133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:13:14,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.178e+03 1.710e+03 2.190e+03 3.036e+03 8.096e+03, threshold=4.380e+03, percent-clipped=5.0
+2023-03-14 08:13:21,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6143, 1.8655, 1.5064, 1.6671], device='cuda:0'), covar=tensor([0.2427, 0.2580, 0.2812, 0.2603], device='cuda:0'), in_proj_covar=tensor([0.1596, 0.1150, 0.1406, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 08:13:26,461 INFO [train.py:968] (0/2) Epoch 28, batch 6250, giga_loss[loss=0.3336, simple_loss=0.3966, pruned_loss=0.1353, over 28319.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3678, pruned_loss=0.1157, over 5704971.09 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3437, pruned_loss=0.08878, over 5507148.67 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3693, pruned_loss=0.1177, over 5692161.51 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:14:03,501 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1237188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:14:06,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1237191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:14:14,667 INFO [train.py:968] (0/2) Epoch 28, batch 6300, giga_loss[loss=0.3252, simple_loss=0.3834, pruned_loss=0.1334, over 28578.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3747, pruned_loss=0.1218, over 5694942.60 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3438, pruned_loss=0.08887, over 5510907.04 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.376, pruned_loss=0.1235, over 5683431.29 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:14:37,386 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1237220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:14:51,573 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.195e+03 1.816e+03 2.435e+03 3.163e+03 7.213e+03, threshold=4.869e+03, percent-clipped=10.0
+2023-03-14 08:15:08,448 INFO [train.py:968] (0/2) Epoch 28, batch 6350, giga_loss[loss=0.5052, simple_loss=0.4876, pruned_loss=0.2614, over 26462.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3782, pruned_loss=0.1254, over 5680697.47 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.344, pruned_loss=0.089, over 5513213.23 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3794, pruned_loss=0.1269, over 5671133.06 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:15:57,207 INFO [train.py:968] (0/2) Epoch 28, batch 6400, giga_loss[loss=0.3017, simple_loss=0.3673, pruned_loss=0.118, over 28615.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3797, pruned_loss=0.1275, over 5673582.41 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3435, pruned_loss=0.0886, over 5522339.56 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.382, pruned_loss=0.13, over 5661047.90 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 08:16:31,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.728e+03 2.268e+03 3.041e+03 8.450e+03, threshold=4.536e+03, percent-clipped=8.0
+2023-03-14 08:16:49,321 INFO [train.py:968] (0/2) Epoch 28, batch 6450, giga_loss[loss=0.3259, simple_loss=0.3865, pruned_loss=0.1326, over 28719.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3826, pruned_loss=0.1304, over 5654674.27 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.344, pruned_loss=0.08896, over 5516275.32 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.385, pruned_loss=0.1331, over 5654629.82 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:17:39,196 INFO [train.py:968] (0/2) Epoch 28, batch 6500, giga_loss[loss=0.3676, simple_loss=0.4164, pruned_loss=0.1594, over 28477.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3852, pruned_loss=0.1334, over 5645667.90 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3433, pruned_loss=0.0886, over 5527679.47 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3888, pruned_loss=0.1371, over 5638917.74 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:17:47,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1237407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:17:52,214 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 08:17:55,704 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5737, 2.1143, 1.7900, 1.7285], device='cuda:0'), covar=tensor([0.0776, 0.0270, 0.0297, 0.0864], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 08:18:03,739 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7590, 1.9614, 1.4132, 1.6433], device='cuda:0'), covar=tensor([0.1130, 0.0784, 0.1131, 0.1218], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0454, 0.0526, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 08:18:14,357 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 2.027e+03 3.021e+03 4.086e+03 1.753e+04, threshold=6.043e+03, percent-clipped=21.0
+2023-03-14 08:18:26,605 INFO [train.py:968] (0/2) Epoch 28, batch 6550, giga_loss[loss=0.2588, simple_loss=0.3361, pruned_loss=0.0908, over 28902.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3866, pruned_loss=0.1349, over 5645887.91 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3436, pruned_loss=0.08872, over 5534635.93 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3901, pruned_loss=0.1386, over 5636338.23 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:19:07,372 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-14 08:19:15,866 INFO [train.py:968] (0/2) Epoch 28, batch 6600, giga_loss[loss=0.3535, simple_loss=0.4047, pruned_loss=0.1512, over 27831.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3857, pruned_loss=0.1351, over 5632942.52 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.08879, over 5530180.86 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3893, pruned_loss=0.139, over 5631148.20 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:19:51,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.955e+03 2.520e+03 3.500e+03 1.007e+04, threshold=5.041e+03, percent-clipped=6.0
+2023-03-14 08:20:03,008 INFO [train.py:968] (0/2) Epoch 28, batch 6650, giga_loss[loss=0.3421, simple_loss=0.3996, pruned_loss=0.1423, over 28688.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3849, pruned_loss=0.1347, over 5634088.62 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.08879, over 5541098.72 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3889, pruned_loss=0.139, over 5626003.26 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:20:13,130 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1237560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:20:46,588 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1237597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:20:48,554 INFO [train.py:968] (0/2) Epoch 28, batch 6700, giga_loss[loss=0.2716, simple_loss=0.3497, pruned_loss=0.09681, over 29035.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3847, pruned_loss=0.1329, over 5650880.28 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3443, pruned_loss=0.08922, over 5552003.72 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3888, pruned_loss=0.1376, over 5638120.37 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:21:17,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.675e+03 2.148e+03 3.463e+03 9.198e+03, threshold=4.297e+03, percent-clipped=11.0
+2023-03-14 08:21:32,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4157, 1.4662, 3.4735, 3.2268], device='cuda:0'), covar=tensor([0.1358, 0.2542, 0.0468, 0.1055], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0670, 0.1005, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 08:21:32,671 INFO [train.py:968] (0/2) Epoch 28, batch 6750, giga_loss[loss=0.3634, simple_loss=0.3927, pruned_loss=0.167, over 23454.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3845, pruned_loss=0.1317, over 5648438.76 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3444, pruned_loss=0.0892, over 5562049.52 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3893, pruned_loss=0.1373, over 5633113.52 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:22:19,915 INFO [train.py:968] (0/2) Epoch 28, batch 6800, giga_loss[loss=0.3029, simple_loss=0.3723, pruned_loss=0.1167, over 28794.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3846, pruned_loss=0.1317, over 5641809.25 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3441, pruned_loss=0.08903, over 5573353.20 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3901, pruned_loss=0.1378, over 5621696.14 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:22:58,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.214e+03 1.823e+03 2.286e+03 2.935e+03 7.571e+03, threshold=4.572e+03, percent-clipped=7.0
+2023-03-14 08:23:13,857 INFO [train.py:968] (0/2) Epoch 28, batch 6850, giga_loss[loss=0.2737, simple_loss=0.344, pruned_loss=0.1016, over 28540.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3816, pruned_loss=0.1291, over 5639646.17 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3438, pruned_loss=0.08896, over 5578469.41 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3868, pruned_loss=0.1346, over 5619896.86 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:23:36,665 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3667, 1.5219, 1.2749, 1.5386], device='cuda:0'), covar=tensor([0.0805, 0.0343, 0.0366, 0.0911], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:0')
+2023-03-14 08:23:47,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1237782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:24:01,081 INFO [train.py:968] (0/2) Epoch 28, batch 6900, giga_loss[loss=0.3898, simple_loss=0.437, pruned_loss=0.1713, over 28606.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3801, pruned_loss=0.1265, over 5659503.77 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3439, pruned_loss=0.08917, over 5588040.41 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3852, pruned_loss=0.1317, over 5636848.65 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:24:09,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5615, 1.6476, 1.2383, 1.2914], device='cuda:0'), covar=tensor([0.0968, 0.0588, 0.0972, 0.1235], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0453, 0.0525, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 08:24:32,452 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-14 08:24:37,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.034e+03 1.887e+03 2.398e+03 3.526e+03 1.110e+04, threshold=4.797e+03, percent-clipped=18.0
+2023-03-14 08:24:48,333 INFO [train.py:968] (0/2) Epoch 28, batch 6950, libri_loss[loss=0.2798, simple_loss=0.3645, pruned_loss=0.09751, over 27822.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3763, pruned_loss=0.1231, over 5657738.52 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3442, pruned_loss=0.08924, over 5591672.98 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3807, pruned_loss=0.1279, over 5637721.00 frames. ], batch size: 116, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:25:34,730 INFO [train.py:968] (0/2) Epoch 28, batch 7000, giga_loss[loss=0.3132, simple_loss=0.3759, pruned_loss=0.1253, over 28872.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3731, pruned_loss=0.1208, over 5657455.40 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3438, pruned_loss=0.08909, over 5597987.78 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3776, pruned_loss=0.1253, over 5637053.50 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:25:56,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2651, 1.5381, 1.2750, 1.4854], device='cuda:0'), covar=tensor([0.0744, 0.0394, 0.0350, 0.0830], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 08:25:59,675 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1237925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:26:02,345 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1237928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:26:08,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1237935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:26:11,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.155e+03 1.892e+03 2.591e+03 3.446e+03 9.811e+03, threshold=5.181e+03, percent-clipped=13.0
+2023-03-14 08:26:22,558 INFO [train.py:968] (0/2) Epoch 28, batch 7050, giga_loss[loss=0.3012, simple_loss=0.3708, pruned_loss=0.1158, over 28630.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3726, pruned_loss=0.1208, over 5656761.87 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3439, pruned_loss=0.08906, over 5602822.47 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3765, pruned_loss=0.125, over 5637381.55 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:26:30,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1237957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:26:44,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1237972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:26:53,524 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-14 08:27:00,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2948, 3.4104, 2.2495, 1.2845], device='cuda:0'), covar=tensor([0.7227, 0.2669, 0.3632, 0.7144], device='cuda:0'), in_proj_covar=tensor([0.1840, 0.1729, 0.1653, 0.1501], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 08:27:11,890 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 08:27:12,143 INFO [train.py:968] (0/2) Epoch 28, batch 7100, libri_loss[loss=0.2359, simple_loss=0.3172, pruned_loss=0.07733, over 29579.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3716, pruned_loss=0.1202, over 5656645.49 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.08883, over 5608673.12 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3757, pruned_loss=0.1245, over 5636909.92 frames. ], batch size: 74, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:27:12,489 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1237999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:27:12,939 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1238000.pt
+2023-03-14 08:27:37,711 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:27:42,659 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-14 08:27:49,831 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.127e+03 1.604e+03 2.056e+03 2.792e+03 8.726e+03, threshold=4.112e+03, percent-clipped=6.0
+2023-03-14 08:28:00,858 INFO [train.py:968] (0/2) Epoch 28, batch 7150, libri_loss[loss=0.27, simple_loss=0.3513, pruned_loss=0.09434, over 29550.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3689, pruned_loss=0.1179, over 5665732.82 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3434, pruned_loss=0.08872, over 5620629.69 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3734, pruned_loss=0.1227, over 5640450.33 frames. ], batch size: 89, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:28:29,700 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:28:32,970 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:28:53,160 INFO [train.py:968] (0/2) Epoch 28, batch 7200, giga_loss[loss=0.2822, simple_loss=0.3643, pruned_loss=0.1001, over 29078.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3695, pruned_loss=0.1168, over 5659504.00 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3433, pruned_loss=0.08874, over 5615235.64 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3736, pruned_loss=0.1209, over 5645698.64 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:28:54,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:28:56,197 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:29:05,837 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:29:11,400 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:29:14,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:29:32,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.945e+02 1.620e+03 2.084e+03 3.097e+03 1.042e+04, threshold=4.169e+03, percent-clipped=14.0
+2023-03-14 08:29:40,618 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:29:43,151 INFO [train.py:968] (0/2) Epoch 28, batch 7250, giga_loss[loss=0.2926, simple_loss=0.3695, pruned_loss=0.1078, over 28472.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3692, pruned_loss=0.1142, over 5667936.71 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3435, pruned_loss=0.0888, over 5619979.70 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3728, pruned_loss=0.118, over 5653671.99 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:29:45,805 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6458, 1.7827, 1.7865, 1.5783], device='cuda:0'), covar=tensor([0.3606, 0.2992, 0.2428, 0.2874], device='cuda:0'), in_proj_covar=tensor([0.2054, 0.2023, 0.1930, 0.2063], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 08:30:08,981 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:30:25,310 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5303, 1.6178, 1.6028, 1.4614], device='cuda:0'), covar=tensor([0.3075, 0.2838, 0.2472, 0.2739], device='cuda:0'), in_proj_covar=tensor([0.2051, 0.2019, 0.1927, 0.2059], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 08:30:37,141 INFO [train.py:968] (0/2) Epoch 28, batch 7300, giga_loss[loss=0.3024, simple_loss=0.3715, pruned_loss=0.1167, over 28290.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3704, pruned_loss=0.1151, over 5658973.98 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3436, pruned_loss=0.08888, over 5611853.80 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3732, pruned_loss=0.1181, over 5655148.70 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:31:13,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.862e+03 2.419e+03 3.378e+03 6.235e+03, threshold=4.838e+03, percent-clipped=14.0
+2023-03-14 08:31:23,979 INFO [train.py:968] (0/2) Epoch 28, batch 7350, giga_loss[loss=0.3433, simple_loss=0.399, pruned_loss=0.1438, over 28301.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3702, pruned_loss=0.1159, over 5657810.19 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08886, over 5619979.83 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3735, pruned_loss=0.1192, over 5648293.68 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:31:58,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2525, 1.5007, 1.3332, 1.1903], device='cuda:0'), covar=tensor([0.2874, 0.2919, 0.1966, 0.2483], device='cuda:0'), in_proj_covar=tensor([0.2057, 0.2026, 0.1932, 0.2067], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 08:32:00,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1833, 1.4971, 1.5157, 1.2596], device='cuda:0'), covar=tensor([0.2116, 0.1757, 0.2409, 0.2118], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0763, 0.0733, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 08:32:06,524 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:32:10,952 INFO [train.py:968] (0/2) Epoch 28, batch 7400, giga_loss[loss=0.294, simple_loss=0.3581, pruned_loss=0.115, over 28911.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3687, pruned_loss=0.1155, over 5672174.74 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3432, pruned_loss=0.08876, over 5628019.36 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3725, pruned_loss=0.1192, over 5658578.68 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:32:12,217 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:32:18,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3907, 1.5463, 1.5743, 1.3761], device='cuda:0'), covar=tensor([0.2554, 0.2176, 0.2208, 0.2346], device='cuda:0'), in_proj_covar=tensor([0.2057, 0.2026, 0.1932, 0.2068], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 08:32:20,276 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 08:32:30,389 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-14 08:32:38,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4990, 2.7794, 1.5761, 1.6131], device='cuda:0'), covar=tensor([0.0760, 0.0333, 0.0692, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0572, 0.0409, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 08:32:38,995 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2878, 1.3611, 1.2511, 1.2547], device='cuda:0'), covar=tensor([0.2096, 0.2080, 0.1901, 0.1889], device='cuda:0'), in_proj_covar=tensor([0.2058, 0.2028, 0.1934, 0.2069], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 08:32:46,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.163e+03 1.853e+03 2.364e+03 3.086e+03 6.972e+03, threshold=4.728e+03, percent-clipped=7.0
+2023-03-14 08:32:55,148 INFO [train.py:968] (0/2) Epoch 28, batch 7450, giga_loss[loss=0.2928, simple_loss=0.3613, pruned_loss=0.1122, over 28766.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3673, pruned_loss=0.1156, over 5677452.51 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08891, over 5637068.56 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.371, pruned_loss=0.1193, over 5659561.78 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:33:00,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 08:33:17,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:33:42,552 INFO [train.py:968] (0/2) Epoch 28, batch 7500, giga_loss[loss=0.2965, simple_loss=0.3419, pruned_loss=0.1256, over 23713.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.367, pruned_loss=0.116, over 5672972.11 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08891, over 5639777.95 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3701, pruned_loss=0.1191, over 5656945.98 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:33:42,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:34:09,638 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1788, 1.0331, 3.7592, 3.3231], device='cuda:0'), covar=tensor([0.2208, 0.3520, 0.0932, 0.1227], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0673, 0.1008, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 08:34:21,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.681e+03 2.224e+03 3.406e+03 7.980e+03, threshold=4.448e+03, percent-clipped=6.0
+2023-03-14 08:34:31,768 INFO [train.py:968] (0/2) Epoch 28, batch 7550, giga_loss[loss=0.2839, simple_loss=0.3608, pruned_loss=0.1035, over 28665.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3682, pruned_loss=0.1158, over 5667736.80 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3434, pruned_loss=0.08885, over 5638922.19 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3709, pruned_loss=0.1186, over 5656113.47 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:34:57,926 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:35:00,762 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:35:20,722 INFO [train.py:968] (0/2) Epoch 28, batch 7600, giga_loss[loss=0.3072, simple_loss=0.379, pruned_loss=0.1177, over 29042.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3685, pruned_loss=0.1156, over 5666294.32 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.343, pruned_loss=0.08863, over 5639270.91 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3713, pruned_loss=0.1184, over 5657278.53 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:35:32,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3262, 3.3211, 1.4650, 1.4589], device='cuda:0'), covar=tensor([0.1035, 0.0363, 0.0950, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0574, 0.0410, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 08:35:37,959 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:35:40,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:35:56,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.773e+03 2.166e+03 3.058e+03 5.172e+03, threshold=4.333e+03, percent-clipped=5.0
+2023-03-14 08:35:59,648 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:36:01,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:36:05,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:36:05,870 INFO [train.py:968] (0/2) Epoch 28, batch 7650, giga_loss[loss=0.2774, simple_loss=0.3552, pruned_loss=0.09976, over 28687.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3669, pruned_loss=0.1141, over 5682981.36 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3431, pruned_loss=0.08865, over 5641710.81 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3692, pruned_loss=0.1165, over 5674003.00 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:36:06,586 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:36:32,249 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:36:55,027 INFO [train.py:968] (0/2) Epoch 28, batch 7700, giga_loss[loss=0.2401, simple_loss=0.3118, pruned_loss=0.08416, over 28631.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3653, pruned_loss=0.1138, over 5678702.39 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3431, pruned_loss=0.08868, over 5648318.92 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3678, pruned_loss=0.1163, over 5666282.53 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:37:15,371 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:37:18,334 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:37:18,350 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:37:20,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:37:32,446 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.520e+03 2.098e+03 2.597e+03 1.287e+04, threshold=4.195e+03, percent-clipped=3.0
+2023-03-14 08:37:41,753 INFO [train.py:968] (0/2) Epoch 28, batch 7750, libri_loss[loss=0.2629, simple_loss=0.3462, pruned_loss=0.08979, over 29519.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3641, pruned_loss=0.1136, over 5677660.28 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.343, pruned_loss=0.08859, over 5659888.97 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3672, pruned_loss=0.1168, over 5657844.27 frames. ], batch size: 83, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:37:43,664 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:37:45,638 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:38:01,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:38:06,357 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:38:21,243 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:38:23,228 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:38:27,691 INFO [train.py:968] (0/2) Epoch 28, batch 7800, giga_loss[loss=0.2311, simple_loss=0.3047, pruned_loss=0.0787, over 28340.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3624, pruned_loss=0.1132, over 5666727.27 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3431, pruned_loss=0.08862, over 5656530.41 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3653, pruned_loss=0.1164, over 5654577.05 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:38:49,223 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:39:05,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.276e+03 1.838e+03 2.232e+03 3.012e+03 9.115e+03, threshold=4.464e+03, percent-clipped=7.0
+2023-03-14 08:39:13,308 INFO [train.py:968] (0/2) Epoch 28, batch 7850, giga_loss[loss=0.2822, simple_loss=0.3531, pruned_loss=0.1057, over 28968.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3617, pruned_loss=0.1131, over 5668548.72 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3434, pruned_loss=0.08887, over 5663982.09 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3643, pruned_loss=0.1162, over 5652559.15 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:39:14,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3554, 1.0801, 4.3594, 3.6308], device='cuda:0'), covar=tensor([0.1942, 0.3218, 0.0762, 0.1253], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0676, 0.1011, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 08:39:15,992 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238752.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:39:58,716 INFO [train.py:968] (0/2) Epoch 28, batch 7900, giga_loss[loss=0.3239, simple_loss=0.3618, pruned_loss=0.143, over 23440.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3613, pruned_loss=0.1135, over 5652185.90 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3433, pruned_loss=0.08893, over 5659021.21 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3638, pruned_loss=0.1164, over 5643736.79 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:40:08,626 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:40:10,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:40:11,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:40:14,186 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:40:30,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.103e+03 1.671e+03 2.414e+03 3.714e+03 9.818e+03, threshold=4.828e+03, percent-clipped=12.0
+2023-03-14 08:40:35,614 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:40:39,719 INFO [train.py:968] (0/2) Epoch 28, batch 7950, giga_loss[loss=0.2749, simple_loss=0.3525, pruned_loss=0.09866, over 29003.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3618, pruned_loss=0.1135, over 5654273.42 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3437, pruned_loss=0.08917, over 5655144.26 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3641, pruned_loss=0.1166, over 5651116.82 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:40:39,928 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:41:22,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4146, 3.3829, 1.5643, 1.5922], device='cuda:0'), covar=tensor([0.1040, 0.0381, 0.0892, 0.1370], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0572, 0.0409, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 08:41:29,332 INFO [train.py:968] (0/2) Epoch 28, batch 8000, libri_loss[loss=0.1992, simple_loss=0.2843, pruned_loss=0.05703, over 29378.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3626, pruned_loss=0.1136, over 5660520.46 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3432, pruned_loss=0.0889, over 5658811.65 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3652, pruned_loss=0.1166, over 5654563.62 frames. ], batch size: 67, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:41:37,356 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:41:53,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0183, 2.3648, 1.8512, 2.2243], device='cuda:0'), covar=tensor([0.2673, 0.2691, 0.3199, 0.2527], device='cuda:0'), in_proj_covar=tensor([0.1595, 0.1151, 0.1408, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 08:42:04,814 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.227e+03 1.755e+03 2.177e+03 2.964e+03 9.420e+03, threshold=4.353e+03, percent-clipped=8.0
+2023-03-14 08:42:13,730 INFO [train.py:968] (0/2) Epoch 28, batch 8050, libri_loss[loss=0.2692, simple_loss=0.3544, pruned_loss=0.09203, over 29760.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3634, pruned_loss=0.1135, over 5660767.20 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.343, pruned_loss=0.08885, over 5655172.55 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3661, pruned_loss=0.1166, over 5660024.02 frames. ], batch size: 87, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:42:34,609 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-14 08:42:59,801 INFO [train.py:968] (0/2) Epoch 28, batch 8100, giga_loss[loss=0.2946, simple_loss=0.367, pruned_loss=0.1111, over 28790.00 frames. ], tot_loss[loss=0.294, simple_loss=0.363, pruned_loss=0.1124, over 5673152.61 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3428, pruned_loss=0.08893, over 5656966.98 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3659, pruned_loss=0.1154, over 5671301.01 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:43:24,313 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1239028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:43:34,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.290e+02 1.685e+03 2.186e+03 2.883e+03 1.254e+04, threshold=4.373e+03, percent-clipped=6.0
+2023-03-14 08:43:44,128 INFO [train.py:968] (0/2) Epoch 28, batch 8150, giga_loss[loss=0.2824, simple_loss=0.3548, pruned_loss=0.1051, over 28980.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3626, pruned_loss=0.112, over 5682141.61 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3427, pruned_loss=0.08885, over 5668056.30 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3659, pruned_loss=0.1155, over 5671399.58 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:44:34,724 INFO [train.py:968] (0/2) Epoch 28, batch 8200, giga_loss[loss=0.2454, simple_loss=0.3248, pruned_loss=0.08299, over 28310.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3653, pruned_loss=0.1147, over 5675573.32 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3431, pruned_loss=0.08908, over 5672330.27 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3679, pruned_loss=0.1177, over 5663232.67 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:44:53,357 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1239118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:45:01,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1239127.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:45:11,716 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4617, 1.5331, 3.4939, 3.2897], device='cuda:0'), covar=tensor([0.1313, 0.2429, 0.0502, 0.1111], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0677, 0.1012, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 08:45:15,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.356e+03 1.980e+03 2.387e+03 3.728e+03 1.267e+04, threshold=4.774e+03, percent-clipped=14.0
+2023-03-14 08:45:21,266 INFO [train.py:968] (0/2) Epoch 28, batch 8250, giga_loss[loss=0.2844, simple_loss=0.3535, pruned_loss=0.1077, over 29094.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3664, pruned_loss=0.1171, over 5666554.03 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3429, pruned_loss=0.08903, over 5676894.48 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3695, pruned_loss=0.1204, over 5652524.65 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:46:05,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5264, 1.7685, 1.1899, 1.3652], device='cuda:0'), covar=tensor([0.1119, 0.0685, 0.1219, 0.1219], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0452, 0.0524, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 08:46:10,887 INFO [train.py:968] (0/2) Epoch 28, batch 8300, giga_loss[loss=0.3332, simple_loss=0.3908, pruned_loss=0.1378, over 28645.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.367, pruned_loss=0.1187, over 5674711.91 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3426, pruned_loss=0.0889, over 5680609.56 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3701, pruned_loss=0.122, over 5660066.86 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:46:11,968 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1239200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:46:51,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.220e+03 1.989e+03 2.585e+03 3.475e+03 7.445e+03, threshold=5.169e+03, percent-clipped=8.0
+2023-03-14 08:46:58,879 INFO [train.py:968] (0/2) Epoch 28, batch 8350, giga_loss[loss=0.295, simple_loss=0.363, pruned_loss=0.1134, over 28683.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3687, pruned_loss=0.1209, over 5669824.48 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08903, over 5685674.59 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3718, pruned_loss=0.1242, over 5653263.13 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:47:20,696 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1239269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:47:21,438 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239270.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 08:47:23,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239273.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 08:47:32,174 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1239284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:47:46,129 INFO [train.py:968] (0/2) Epoch 28, batch 8400, giga_loss[loss=0.3049, simple_loss=0.3631, pruned_loss=0.1233, over 29079.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3671, pruned_loss=0.1192, over 5675050.45 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3427, pruned_loss=0.08901, over 5689950.87 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3701, pruned_loss=0.1225, over 5657616.95 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:47:48,546 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239302.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 08:48:17,884 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.288e+03 1.876e+03 2.392e+03 3.394e+03 8.471e+03, threshold=4.783e+03, percent-clipped=5.0
+2023-03-14 08:48:26,751 INFO [train.py:968] (0/2) Epoch 28, batch 8450, giga_loss[loss=0.2853, simple_loss=0.3682, pruned_loss=0.1012, over 28953.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3674, pruned_loss=0.118, over 5684207.02 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08938, over 5691840.51 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3701, pruned_loss=0.121, over 5668159.60 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:48:31,970 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 08:49:09,088 INFO [train.py:968] (0/2) Epoch 28, batch 8500, libri_loss[loss=0.305, simple_loss=0.3792, pruned_loss=0.1154, over 29774.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3651, pruned_loss=0.1148, over 5695410.24 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.08921, over 5700972.82 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3684, pruned_loss=0.1186, over 5674034.71 frames. ], batch size: 87, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:49:13,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1239403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:49:17,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7284, 1.1018, 2.8731, 2.7391], device='cuda:0'), covar=tensor([0.1780, 0.2630, 0.0641, 0.1008], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0678, 0.1014, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 08:49:33,143 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:49:35,137 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:49:42,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.683e+03 2.306e+03 3.369e+03 5.731e+03, threshold=4.612e+03, percent-clipped=4.0
+2023-03-14 08:49:50,850 INFO [train.py:968] (0/2) Epoch 28, batch 8550, giga_loss[loss=0.2671, simple_loss=0.3404, pruned_loss=0.09692, over 28891.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3627, pruned_loss=0.1135, over 5692749.87 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08909, over 5703442.29 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3657, pruned_loss=0.1169, over 5673549.81 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:50:00,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3803, 1.8636, 1.4227, 0.5660], device='cuda:0'), covar=tensor([0.5035, 0.3261, 0.3582, 0.6720], device='cuda:0'), in_proj_covar=tensor([0.1843, 0.1743, 0.1659, 0.1507], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 08:50:01,292 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:50:30,933 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1239493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:50:34,733 INFO [train.py:968] (0/2) Epoch 28, batch 8600, giga_loss[loss=0.246, simple_loss=0.3234, pruned_loss=0.08431, over 28900.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3615, pruned_loss=0.113, over 5675415.49 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08929, over 5694423.61 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3645, pruned_loss=0.1166, over 5668074.26 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:51:14,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.833e+03 2.318e+03 3.132e+03 9.408e+03, threshold=4.635e+03, percent-clipped=9.0
+2023-03-14 08:51:22,228 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:51:24,001 INFO [train.py:968] (0/2) Epoch 28, batch 8650, giga_loss[loss=0.2904, simple_loss=0.3624, pruned_loss=0.1092, over 28716.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3612, pruned_loss=0.1134, over 5674110.97 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3431, pruned_loss=0.08936, over 5699030.92 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3638, pruned_loss=0.1166, over 5663946.30 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:51:24,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:51:42,367 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4420, 1.7460, 1.4337, 1.3110], device='cuda:0'), covar=tensor([0.2601, 0.2584, 0.2926, 0.2313], device='cuda:0'), in_proj_covar=tensor([0.1593, 0.1149, 0.1406, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 08:51:53,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1239575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:51:55,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:52:14,280 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4562, 4.3041, 4.1244, 2.1268], device='cuda:0'), covar=tensor([0.0579, 0.0678, 0.0724, 0.1888], device='cuda:0'), in_proj_covar=tensor([0.1318, 0.1216, 0.1026, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 08:52:14,739 INFO [train.py:968] (0/2) Epoch 28, batch 8700, giga_loss[loss=0.3179, simple_loss=0.3842, pruned_loss=0.1258, over 28844.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3651, pruned_loss=0.116, over 5679617.91 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08962, over 5702154.59 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3672, pruned_loss=0.1187, over 5668319.77 frames. ], batch size: 243, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:52:47,902 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-14 08:52:49,804 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:52:53,147 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:52:53,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.705e+03 2.128e+03 2.720e+03 5.517e+03, threshold=4.256e+03, percent-clipped=8.0
+2023-03-14 08:52:59,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1239644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:53:02,836 INFO [train.py:968] (0/2) Epoch 28, batch 8750, giga_loss[loss=0.3257, simple_loss=0.3958, pruned_loss=0.1278, over 28492.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3681, pruned_loss=0.1157, over 5669088.31 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3437, pruned_loss=0.0899, over 5696826.90 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3701, pruned_loss=0.1183, over 5664401.29 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:53:20,254 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:53:48,526 INFO [train.py:968] (0/2) Epoch 28, batch 8800, giga_loss[loss=0.3303, simple_loss=0.4015, pruned_loss=0.1296, over 28997.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3696, pruned_loss=0.1153, over 5669201.71 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3432, pruned_loss=0.08974, over 5693794.38 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3723, pruned_loss=0.1182, over 5668149.33 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 08:53:58,006 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 08:54:07,113 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:54:09,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:54:16,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 08:54:27,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.880e+03 2.630e+03 4.085e+03 1.046e+04, threshold=5.259e+03, percent-clipped=23.0
+2023-03-14 08:54:33,474 INFO [train.py:968] (0/2) Epoch 28, batch 8850, giga_loss[loss=0.328, simple_loss=0.391, pruned_loss=0.1325, over 29030.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3728, pruned_loss=0.1177, over 5671592.16 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3432, pruned_loss=0.0897, over 5693700.55 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3751, pruned_loss=0.1202, over 5670646.24 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:54:34,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:54:35,957 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1239752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:55:05,503 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:55:09,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:55:17,417 INFO [train.py:968] (0/2) Epoch 28, batch 8900, giga_loss[loss=0.3449, simple_loss=0.402, pruned_loss=0.1439, over 28902.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3719, pruned_loss=0.1174, over 5684108.22 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3435, pruned_loss=0.08996, over 5699361.30 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3745, pruned_loss=0.1203, over 5677720.70 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:55:34,040 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:55:53,522 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.227e+03 1.791e+03 2.356e+03 2.918e+03 8.958e+03, threshold=4.712e+03, percent-clipped=1.0
+2023-03-14 08:55:58,742 INFO [train.py:968] (0/2) Epoch 28, batch 8950, libri_loss[loss=0.2271, simple_loss=0.3142, pruned_loss=0.07001, over 29590.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3716, pruned_loss=0.1184, over 5688293.75 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08989, over 5704954.98 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.375, pruned_loss=0.1218, over 5677434.56 frames. ], batch size: 74, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:56:07,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1239859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:56:27,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.7225, 0.9231, 0.7376, 0.2201], device='cuda:0'), covar=tensor([0.3169, 0.3068, 0.3564, 0.5265], device='cuda:0'), in_proj_covar=tensor([0.1846, 0.1746, 0.1661, 0.1508], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 08:56:45,586 INFO [train.py:968] (0/2) Epoch 28, batch 9000, giga_loss[loss=0.2752, simple_loss=0.3431, pruned_loss=0.1037, over 28471.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3708, pruned_loss=0.1188, over 5685268.65 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3435, pruned_loss=0.09023, over 5698807.07 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3739, pruned_loss=0.1218, over 5682135.00 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:56:45,590 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 08:56:53,054 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4079, 1.7788, 1.6939, 1.2147], device='cuda:0'), covar=tensor([0.1944, 0.3056, 0.1752, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0717, 0.0980, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 08:56:53,879 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2036, simple_loss=0.3114, pruned_loss=0.04786, over 944034.00 frames. 
+2023-03-14 08:56:53,880 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 08:57:33,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.689e+03 2.334e+03 3.380e+03 8.038e+03, threshold=4.668e+03, percent-clipped=9.0
+2023-03-14 08:57:40,927 INFO [train.py:968] (0/2) Epoch 28, batch 9050, giga_loss[loss=0.2806, simple_loss=0.3495, pruned_loss=0.1059, over 28894.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.368, pruned_loss=0.1174, over 5682573.67 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3434, pruned_loss=0.09008, over 5699494.28 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.371, pruned_loss=0.1203, over 5679135.70 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:57:47,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3342, 1.6188, 1.3197, 0.9991], device='cuda:0'), covar=tensor([0.2516, 0.2630, 0.2968, 0.2427], device='cuda:0'), in_proj_covar=tensor([0.1593, 0.1150, 0.1408, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 08:58:24,788 INFO [train.py:968] (0/2) Epoch 28, batch 9100, giga_loss[loss=0.3159, simple_loss=0.3766, pruned_loss=0.1276, over 28696.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3662, pruned_loss=0.1168, over 5681343.37 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3431, pruned_loss=0.08984, over 5704918.17 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3697, pruned_loss=0.1203, over 5673171.63 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:58:26,942 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1240000.pt
+2023-03-14 08:59:06,476 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.801e+03 2.272e+03 3.025e+03 5.054e+03, threshold=4.545e+03, percent-clipped=2.0
+2023-03-14 08:59:15,965 INFO [train.py:968] (0/2) Epoch 28, batch 9150, giga_loss[loss=0.3342, simple_loss=0.3883, pruned_loss=0.14, over 27959.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3666, pruned_loss=0.1176, over 5677708.92 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.09, over 5698698.65 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5675869.72 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:00:01,610 INFO [train.py:968] (0/2) Epoch 28, batch 9200, giga_loss[loss=0.3302, simple_loss=0.3907, pruned_loss=0.1349, over 28679.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3674, pruned_loss=0.1188, over 5675367.38 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3431, pruned_loss=0.0899, over 5702709.47 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3704, pruned_loss=0.1219, over 5669802.96 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:00:27,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1240127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:00:40,952 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.111e+03 1.878e+03 2.640e+03 3.600e+03 8.870e+03, threshold=5.279e+03, percent-clipped=15.0
+2023-03-14 09:00:46,524 INFO [train.py:968] (0/2) Epoch 28, batch 9250, giga_loss[loss=0.2924, simple_loss=0.3645, pruned_loss=0.1102, over 28917.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3651, pruned_loss=0.1172, over 5678652.12 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.343, pruned_loss=0.08972, over 5704070.28 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3681, pruned_loss=0.1204, over 5672111.90 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:01:33,842 INFO [train.py:968] (0/2) Epoch 28, batch 9300, giga_loss[loss=0.3131, simple_loss=0.3725, pruned_loss=0.1268, over 28262.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3652, pruned_loss=0.1175, over 5681191.13 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.343, pruned_loss=0.08961, over 5706000.46 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3679, pruned_loss=0.1207, over 5673968.26 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:01:49,385 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-14 09:02:00,597 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.1578, 6.0016, 5.6714, 3.4391], device='cuda:0'), covar=tensor([0.0528, 0.0643, 0.0849, 0.1426], device='cuda:0'), in_proj_covar=tensor([0.1316, 0.1215, 0.1025, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 09:02:04,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1240234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:02:12,249 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.086e+03 1.740e+03 2.457e+03 3.550e+03 7.770e+03, threshold=4.914e+03, percent-clipped=3.0
+2023-03-14 09:02:13,688 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3526, 3.4180, 1.5130, 1.4888], device='cuda:0'), covar=tensor([0.1035, 0.0347, 0.0932, 0.1411], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0572, 0.0409, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 09:02:18,945 INFO [train.py:968] (0/2) Epoch 28, batch 9350, giga_loss[loss=0.2836, simple_loss=0.3627, pruned_loss=0.1022, over 28608.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.366, pruned_loss=0.1171, over 5676748.62 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3426, pruned_loss=0.08936, over 5704944.07 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3694, pruned_loss=0.1209, over 5670147.78 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:02:36,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1240270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:02:38,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1240273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:02:44,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7068, 2.1475, 1.7713, 1.8515], device='cuda:0'), covar=tensor([0.0723, 0.0256, 0.0295, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 09:03:02,365 INFO [train.py:968] (0/2) Epoch 28, batch 9400, giga_loss[loss=0.355, simple_loss=0.4055, pruned_loss=0.1522, over 27778.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3682, pruned_loss=0.1183, over 5678611.17 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3428, pruned_loss=0.0895, over 5708777.25 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3712, pruned_loss=0.1217, over 5669544.93 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:03:06,910 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1240302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:03:30,056 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1240327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:03:43,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.789e+03 2.358e+03 3.488e+03 1.014e+04, threshold=4.715e+03, percent-clipped=13.0
+2023-03-14 09:03:50,294 INFO [train.py:968] (0/2) Epoch 28, batch 9450, giga_loss[loss=0.3557, simple_loss=0.3826, pruned_loss=0.1644, over 23533.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3682, pruned_loss=0.1192, over 5675000.54 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.343, pruned_loss=0.08968, over 5712193.15 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.371, pruned_loss=0.1225, over 5663882.42 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:04:15,611 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1240377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:04:18,364 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1240380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:04:38,653 INFO [train.py:968] (0/2) Epoch 28, batch 9500, giga_loss[loss=0.2729, simple_loss=0.3659, pruned_loss=0.08997, over 28958.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3699, pruned_loss=0.1179, over 5681580.45 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.343, pruned_loss=0.08968, over 5713232.12 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3724, pruned_loss=0.1207, over 5671718.05 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:04:46,703 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1240409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:05:01,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4729, 1.8209, 1.4178, 1.7985], device='cuda:0'), covar=tensor([0.2584, 0.2562, 0.2954, 0.2184], device='cuda:0'), in_proj_covar=tensor([0.1595, 0.1150, 0.1410, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 09:05:06,421 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1240431.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 09:05:14,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-14 09:05:15,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.519e+03 1.949e+03 2.779e+03 5.377e+03, threshold=3.899e+03, percent-clipped=1.0
+2023-03-14 09:05:23,000 INFO [train.py:968] (0/2) Epoch 28, batch 9550, giga_loss[loss=0.2703, simple_loss=0.3595, pruned_loss=0.09053, over 28977.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3712, pruned_loss=0.1167, over 5681526.54 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3432, pruned_loss=0.08974, over 5712774.78 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3734, pruned_loss=0.1193, over 5673450.44 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:06:07,439 INFO [train.py:968] (0/2) Epoch 28, batch 9600, giga_loss[loss=0.2933, simple_loss=0.367, pruned_loss=0.1098, over 28952.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3739, pruned_loss=0.1182, over 5676823.61 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3432, pruned_loss=0.08965, over 5716215.62 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3762, pruned_loss=0.1209, over 5666638.71 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:06:21,771 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1240514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:06:47,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.669e+03 2.080e+03 2.690e+03 5.239e+03, threshold=4.159e+03, percent-clipped=4.0
+2023-03-14 09:06:53,551 INFO [train.py:968] (0/2) Epoch 28, batch 9650, giga_loss[loss=0.3231, simple_loss=0.3814, pruned_loss=0.1324, over 28913.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3758, pruned_loss=0.1203, over 5678588.62 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.0897, over 5718117.74 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3783, pruned_loss=0.1231, over 5667861.58 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:07:00,141 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7491, 4.5612, 4.3417, 2.1617], device='cuda:0'), covar=tensor([0.0686, 0.0856, 0.0985, 0.2027], device='cuda:0'), in_proj_covar=tensor([0.1319, 0.1219, 0.1026, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 09:07:38,341 INFO [train.py:968] (0/2) Epoch 28, batch 9700, giga_loss[loss=0.299, simple_loss=0.3681, pruned_loss=0.115, over 28713.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.377, pruned_loss=0.1223, over 5678731.43 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3431, pruned_loss=0.08958, over 5722389.64 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.38, pruned_loss=0.1254, over 5665406.53 frames. ], batch size: 243, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:08:19,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 2.011e+03 2.651e+03 3.765e+03 9.020e+03, threshold=5.302e+03, percent-clipped=21.0
+2023-03-14 09:08:24,850 INFO [train.py:968] (0/2) Epoch 28, batch 9750, giga_loss[loss=0.2673, simple_loss=0.3449, pruned_loss=0.09482, over 28739.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3771, pruned_loss=0.1236, over 5671142.12 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3433, pruned_loss=0.08972, over 5723725.93 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3798, pruned_loss=0.1265, over 5658638.29 frames. ], batch size: 66, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:09:07,146 INFO [train.py:968] (0/2) Epoch 28, batch 9800, giga_loss[loss=0.3262, simple_loss=0.4035, pruned_loss=0.1244, over 28546.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3752, pruned_loss=0.1216, over 5669551.40 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3435, pruned_loss=0.08982, over 5729305.51 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3782, pruned_loss=0.1249, over 5652533.83 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:09:10,513 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1240702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:09:41,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6571, 1.7314, 1.8345, 1.4028], device='cuda:0'), covar=tensor([0.2113, 0.2712, 0.1766, 0.1997], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0719, 0.0982, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 09:09:46,384 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.692e+02 1.701e+03 2.151e+03 2.764e+03 7.441e+03, threshold=4.302e+03, percent-clipped=2.0
+2023-03-14 09:09:50,898 INFO [train.py:968] (0/2) Epoch 28, batch 9850, giga_loss[loss=0.2864, simple_loss=0.3691, pruned_loss=0.1018, over 28917.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3741, pruned_loss=0.119, over 5668318.43 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3437, pruned_loss=0.08985, over 5726839.19 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3765, pruned_loss=0.1218, over 5656599.37 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:10:37,242 INFO [train.py:968] (0/2) Epoch 28, batch 9900, giga_loss[loss=0.2915, simple_loss=0.3705, pruned_loss=0.1062, over 28951.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3741, pruned_loss=0.1178, over 5675338.77 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3441, pruned_loss=0.09008, over 5729057.19 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3763, pruned_loss=0.1204, over 5662994.96 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:10:44,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1240806.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 09:10:58,952 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7253, 5.1513, 1.9340, 2.0325], device='cuda:0'), covar=tensor([0.0954, 0.0221, 0.0872, 0.1258], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0574, 0.0410, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 09:11:04,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5117, 1.7842, 1.4692, 1.3633], device='cuda:0'), covar=tensor([0.3008, 0.2838, 0.3361, 0.2531], device='cuda:0'), in_proj_covar=tensor([0.1596, 0.1151, 0.1410, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 09:11:15,823 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1240839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:11:18,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.797e+03 2.169e+03 2.678e+03 8.044e+03, threshold=4.338e+03, percent-clipped=10.0
+2023-03-14 09:11:20,698 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1240845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:11:22,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1240848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:11:23,069 INFO [train.py:968] (0/2) Epoch 28, batch 9950, libri_loss[loss=0.3189, simple_loss=0.383, pruned_loss=0.1274, over 19541.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3749, pruned_loss=0.1187, over 5659973.72 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3443, pruned_loss=0.0903, over 5717382.49 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3778, pruned_loss=0.1218, over 5659587.46 frames. ], batch size: 187, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:11:49,082 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1240877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:12:01,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1240889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:12:11,714 INFO [train.py:968] (0/2) Epoch 28, batch 10000, giga_loss[loss=0.2827, simple_loss=0.3573, pruned_loss=0.1041, over 28987.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3739, pruned_loss=0.1184, over 5662439.75 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09018, over 5720602.90 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3771, pruned_loss=0.1217, over 5657813.88 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:12:37,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1240927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:12:55,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.247e+03 1.805e+03 2.204e+03 3.010e+03 7.233e+03, threshold=4.409e+03, percent-clipped=5.0
+2023-03-14 09:12:56,459 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 09:12:59,844 INFO [train.py:968] (0/2) Epoch 28, batch 10050, giga_loss[loss=0.264, simple_loss=0.3325, pruned_loss=0.0977, over 28587.00 frames. ], tot_loss[loss=0.304, simple_loss=0.372, pruned_loss=0.118, over 5672428.56 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3441, pruned_loss=0.09008, over 5724871.45 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3754, pruned_loss=0.1214, over 5663501.90 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:13:00,404 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1240949.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 09:13:04,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1240952.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 09:13:34,843 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1240981.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 09:13:48,382 INFO [train.py:968] (0/2) Epoch 28, batch 10100, libri_loss[loss=0.291, simple_loss=0.362, pruned_loss=0.11, over 19777.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3711, pruned_loss=0.1185, over 5660673.57 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3442, pruned_loss=0.09019, over 5718668.66 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3741, pruned_loss=0.1216, over 5658928.72 frames. ], batch size: 187, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:14:18,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1241032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:14:19,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8663, 5.6972, 5.4563, 2.8716], device='cuda:0'), covar=tensor([0.0470, 0.0587, 0.0697, 0.1622], device='cuda:0'), in_proj_covar=tensor([0.1325, 0.1223, 0.1032, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 09:14:20,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1241035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:14:28,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.785e+03 2.281e+03 3.135e+03 8.029e+03, threshold=4.563e+03, percent-clipped=11.0
+2023-03-14 09:14:32,233 INFO [train.py:968] (0/2) Epoch 28, batch 10150, giga_loss[loss=0.2733, simple_loss=0.3464, pruned_loss=0.1001, over 28747.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3668, pruned_loss=0.1157, over 5670584.06 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3434, pruned_loss=0.08971, over 5722591.06 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.371, pruned_loss=0.1198, over 5662959.50 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:14:46,168 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1241064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:15:11,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2058, 1.4771, 1.4945, 1.3015], device='cuda:0'), covar=tensor([0.2070, 0.1843, 0.2427, 0.2005], device='cuda:0'), in_proj_covar=tensor([0.0507, 0.0762, 0.0730, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 09:15:21,262 INFO [train.py:968] (0/2) Epoch 28, batch 10200, giga_loss[loss=0.3619, simple_loss=0.3899, pruned_loss=0.167, over 23372.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.366, pruned_loss=0.1161, over 5664145.95 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3432, pruned_loss=0.08954, over 5719216.14 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3702, pruned_loss=0.1202, over 5659276.80 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:16:03,533 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.236e+03 1.772e+03 2.344e+03 3.467e+03 7.171e+03, threshold=4.689e+03, percent-clipped=12.0
+2023-03-14 09:16:06,674 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6451, 1.8910, 1.5181, 2.0421], device='cuda:0'), covar=tensor([0.2677, 0.2863, 0.3151, 0.2521], device='cuda:0'), in_proj_covar=tensor([0.1599, 0.1153, 0.1412, 0.1015], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 09:16:07,105 INFO [train.py:968] (0/2) Epoch 28, batch 10250, giga_loss[loss=0.2645, simple_loss=0.3414, pruned_loss=0.09376, over 28717.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3665, pruned_loss=0.1169, over 5673244.32 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08943, over 5723975.44 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3706, pruned_loss=0.1209, over 5663846.59 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:16:26,630 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5244, 2.0406, 1.5003, 1.7322], device='cuda:0'), covar=tensor([0.0796, 0.0288, 0.0346, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 09:16:45,392 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 09:16:52,799 INFO [train.py:968] (0/2) Epoch 28, batch 10300, giga_loss[loss=0.2753, simple_loss=0.3565, pruned_loss=0.097, over 28945.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.365, pruned_loss=0.1153, over 5666472.64 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08914, over 5727476.28 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.369, pruned_loss=0.1192, over 5654947.74 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:17:07,056 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1241214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:17:36,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.875e+02 1.648e+03 2.202e+03 2.903e+03 7.141e+03, threshold=4.403e+03, percent-clipped=6.0
+2023-03-14 09:17:39,792 INFO [train.py:968] (0/2) Epoch 28, batch 10350, giga_loss[loss=0.2875, simple_loss=0.362, pruned_loss=0.1065, over 28217.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3615, pruned_loss=0.1121, over 5673083.21 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3426, pruned_loss=0.08944, over 5731921.30 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3653, pruned_loss=0.1157, over 5658132.09 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:17:44,047 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-14 09:18:05,460 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1241275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:18:25,869 INFO [train.py:968] (0/2) Epoch 28, batch 10400, libri_loss[loss=0.2798, simple_loss=0.3666, pruned_loss=0.09654, over 29181.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3616, pruned_loss=0.1119, over 5665330.63 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3427, pruned_loss=0.08934, over 5726936.09 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.365, pruned_loss=0.1155, over 5655280.30 frames. ], batch size: 101, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:18:28,805 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1241302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:18:38,665 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-14 09:18:51,934 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8016, 1.9145, 1.6443, 1.6165], device='cuda:0'), covar=tensor([0.2400, 0.2570, 0.2575, 0.2538], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0764, 0.0732, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 09:19:06,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.630e+03 2.184e+03 3.729e+03 1.027e+04, threshold=4.369e+03, percent-clipped=15.0
+2023-03-14 09:19:10,598 INFO [train.py:968] (0/2) Epoch 28, batch 10450, giga_loss[loss=0.3072, simple_loss=0.3677, pruned_loss=0.1233, over 28841.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3591, pruned_loss=0.1107, over 5678042.79 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3425, pruned_loss=0.08916, over 5733636.18 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3628, pruned_loss=0.1146, over 5661540.37 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:19:20,936 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1241357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:19:23,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1241360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:19:50,053 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1241389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:19:59,513 INFO [train.py:968] (0/2) Epoch 28, batch 10500, giga_loss[loss=0.3038, simple_loss=0.3643, pruned_loss=0.1216, over 28613.00 frames. ], tot_loss[loss=0.289, simple_loss=0.357, pruned_loss=0.1105, over 5673712.46 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3424, pruned_loss=0.0891, over 5734999.45 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3602, pruned_loss=0.1141, over 5658356.82 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:20:09,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5868, 2.0701, 1.3377, 0.8161], device='cuda:0'), covar=tensor([0.7747, 0.4038, 0.3458, 0.7420], device='cuda:0'), in_proj_covar=tensor([0.1844, 0.1736, 0.1655, 0.1501], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 09:20:30,210 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1241430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:20:44,764 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1241445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:20:45,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.134e+03 1.948e+03 2.705e+03 4.042e+03 1.437e+04, threshold=5.411e+03, percent-clipped=18.0
+2023-03-14 09:20:46,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1241448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:20:47,306 INFO [train.py:968] (0/2) Epoch 28, batch 10550, giga_loss[loss=0.2924, simple_loss=0.3581, pruned_loss=0.1134, over 28782.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3601, pruned_loss=0.1125, over 5677414.48 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.342, pruned_loss=0.08887, over 5738614.76 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3634, pruned_loss=0.116, over 5660800.03 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:21:03,363 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7647, 2.5406, 1.5833, 1.0625], device='cuda:0'), covar=tensor([0.8970, 0.3791, 0.4573, 0.7489], device='cuda:0'), in_proj_covar=tensor([0.1842, 0.1734, 0.1654, 0.1501], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 09:21:12,571 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0829, 1.0467, 3.3977, 2.9483], device='cuda:0'), covar=tensor([0.1776, 0.2972, 0.0570, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0677, 0.1013, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 09:21:13,928 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1241477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:21:32,953 INFO [train.py:968] (0/2) Epoch 28, batch 10600, giga_loss[loss=0.2742, simple_loss=0.3519, pruned_loss=0.09821, over 28936.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3616, pruned_loss=0.1125, over 5678129.79 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.08874, over 5737821.51 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3646, pruned_loss=0.1158, over 5664621.33 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:21:42,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6965, 2.0869, 1.5216, 1.6856], device='cuda:0'), covar=tensor([0.1007, 0.0455, 0.0976, 0.0951], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0450, 0.0519, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 09:22:17,610 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.104e+03 1.603e+03 2.135e+03 3.065e+03 7.220e+03, threshold=4.269e+03, percent-clipped=7.0
+2023-03-14 09:22:19,558 INFO [train.py:968] (0/2) Epoch 28, batch 10650, giga_loss[loss=0.2621, simple_loss=0.3447, pruned_loss=0.08973, over 28882.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3627, pruned_loss=0.1135, over 5647291.71 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3423, pruned_loss=0.08896, over 5738065.18 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3653, pruned_loss=0.1165, over 5634730.08 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:22:20,367 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1241550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:22:56,726 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-14 09:23:00,286 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-14 09:23:02,835 INFO [train.py:968] (0/2) Epoch 28, batch 10700, giga_loss[loss=0.295, simple_loss=0.3643, pruned_loss=0.1128, over 29115.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3632, pruned_loss=0.1144, over 5633036.07 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3426, pruned_loss=0.08921, over 5725444.13 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3658, pruned_loss=0.1175, over 5631060.37 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:23:31,007 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.46 vs. limit=5.0
+2023-03-14 09:23:45,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.256e+03 1.711e+03 2.187e+03 2.668e+03 5.564e+03, threshold=4.374e+03, percent-clipped=3.0
+2023-03-14 09:23:49,351 INFO [train.py:968] (0/2) Epoch 28, batch 10750, giga_loss[loss=0.3104, simple_loss=0.3807, pruned_loss=0.1201, over 28737.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3649, pruned_loss=0.1159, over 5630793.68 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08969, over 5719164.81 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3668, pruned_loss=0.1185, over 5632670.89 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:23:50,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1241650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:24:15,208 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-14 09:24:38,858 INFO [train.py:968] (0/2) Epoch 28, batch 10800, giga_loss[loss=0.2906, simple_loss=0.3573, pruned_loss=0.1119, over 28769.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3669, pruned_loss=0.1169, over 5643048.55 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3432, pruned_loss=0.08973, over 5721925.05 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3689, pruned_loss=0.1194, over 5640632.87 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:25:21,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.134e+03 1.902e+03 2.347e+03 3.286e+03 6.818e+03, threshold=4.694e+03, percent-clipped=13.0
+2023-03-14 09:25:25,285 INFO [train.py:968] (0/2) Epoch 28, batch 10850, giga_loss[loss=0.3361, simple_loss=0.3949, pruned_loss=0.1387, over 28316.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3678, pruned_loss=0.1177, over 5650121.01 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3429, pruned_loss=0.08955, over 5725414.95 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3702, pruned_loss=0.1203, over 5643901.49 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:25:31,796 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 09:25:37,866 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-14 09:26:02,687 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1241793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:26:05,933 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1241796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:26:08,489 INFO [train.py:968] (0/2) Epoch 28, batch 10900, giga_loss[loss=0.3291, simple_loss=0.3877, pruned_loss=0.1353, over 28301.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3698, pruned_loss=0.1193, over 5655066.04 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3431, pruned_loss=0.08959, over 5730105.22 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3722, pruned_loss=0.1222, over 5643834.34 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:26:10,665 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1241800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:26:16,308 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1241805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:26:30,799 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.00 vs. limit=2.0
+2023-03-14 09:26:34,043 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1241825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:26:53,827 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.223e+03 1.671e+03 2.294e+03 3.149e+03 7.114e+03, threshold=4.587e+03, percent-clipped=5.0
+2023-03-14 09:26:57,349 INFO [train.py:968] (0/2) Epoch 28, batch 10950, giga_loss[loss=0.267, simple_loss=0.3512, pruned_loss=0.09139, over 28913.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3708, pruned_loss=0.12, over 5640483.38 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3434, pruned_loss=0.08972, over 5716752.31 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3731, pruned_loss=0.1229, over 5641736.35 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:27:35,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4496, 1.5203, 1.2432, 1.0956], device='cuda:0'), covar=tensor([0.0853, 0.0436, 0.0870, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0453, 0.0522, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 09:27:47,617 INFO [train.py:968] (0/2) Epoch 28, batch 11000, giga_loss[loss=0.2866, simple_loss=0.358, pruned_loss=0.1076, over 28924.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3709, pruned_loss=0.1187, over 5634178.41 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3437, pruned_loss=0.08996, over 5709133.45 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3727, pruned_loss=0.1211, over 5641376.24 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:28:13,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1241925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:28:35,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.122e+03 1.840e+03 2.222e+03 3.143e+03 7.644e+03, threshold=4.444e+03, percent-clipped=7.0
+2023-03-14 09:28:37,508 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1241948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:28:39,366 INFO [train.py:968] (0/2) Epoch 28, batch 11050, giga_loss[loss=0.2802, simple_loss=0.3536, pruned_loss=0.1034, over 28871.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3704, pruned_loss=0.1191, over 5628811.29 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.09015, over 5700019.53 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.372, pruned_loss=0.1211, over 5641907.15 frames. ], batch size: 174, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:28:43,188 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1241951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:29:09,465 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1241980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:29:29,334 INFO [train.py:968] (0/2) Epoch 28, batch 11100, libri_loss[loss=0.2274, simple_loss=0.307, pruned_loss=0.07388, over 29382.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3691, pruned_loss=0.1188, over 5648761.68 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3441, pruned_loss=0.09052, over 5704616.07 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3709, pruned_loss=0.1209, over 5653143.57 frames. ], batch size: 67, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:29:30,055 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1242000.pt
+2023-03-14 09:29:31,286 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1364, 2.6508, 1.8514, 2.2784], device='cuda:0'), covar=tensor([0.1052, 0.0588, 0.0989, 0.1046], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0453, 0.0522, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 09:30:17,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.159e+03 1.810e+03 2.420e+03 3.348e+03 7.222e+03, threshold=4.839e+03, percent-clipped=11.0
+2023-03-14 09:30:19,465 INFO [train.py:968] (0/2) Epoch 28, batch 11150, giga_loss[loss=0.2816, simple_loss=0.3569, pruned_loss=0.1032, over 28734.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3678, pruned_loss=0.1181, over 5649036.69 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3441, pruned_loss=0.09041, over 5711335.60 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3701, pruned_loss=0.121, over 5644511.91 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:30:35,128 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1242068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:30:37,209 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1242071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:30:53,431 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6217, 1.8106, 1.6777, 1.4949], device='cuda:0'), covar=tensor([0.3261, 0.2707, 0.2594, 0.3049], device='cuda:0'), in_proj_covar=tensor([0.2072, 0.2038, 0.1950, 0.2085], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 09:31:02,449 INFO [train.py:968] (0/2) Epoch 28, batch 11200, giga_loss[loss=0.284, simple_loss=0.3491, pruned_loss=0.1094, over 28495.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3661, pruned_loss=0.1169, over 5666089.41 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3435, pruned_loss=0.08999, over 5715425.05 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3692, pruned_loss=0.1204, over 5657164.94 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:31:03,363 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1242100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:31:20,198 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-14 09:31:45,998 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.680e+03 1.993e+03 3.028e+03 5.736e+03, threshold=3.986e+03, percent-clipped=3.0
+2023-03-14 09:31:47,329 INFO [train.py:968] (0/2) Epoch 28, batch 11250, libri_loss[loss=0.2262, simple_loss=0.3046, pruned_loss=0.07392, over 29617.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3658, pruned_loss=0.1174, over 5673354.17 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.08985, over 5720338.84 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3691, pruned_loss=0.121, over 5660826.52 frames. ], batch size: 69, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:32:13,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1242175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:32:14,025 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-14 09:32:24,670 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1242187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:32:36,959 INFO [train.py:968] (0/2) Epoch 28, batch 11300, giga_loss[loss=0.3058, simple_loss=0.3672, pruned_loss=0.1222, over 28684.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3664, pruned_loss=0.1182, over 5671075.61 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3432, pruned_loss=0.08981, over 5720948.14 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3691, pruned_loss=0.1212, over 5660440.84 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:33:22,691 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.756e+03 2.311e+03 3.073e+03 5.752e+03, threshold=4.621e+03, percent-clipped=12.0
+2023-03-14 09:33:23,806 INFO [train.py:968] (0/2) Epoch 28, batch 11350, giga_loss[loss=0.3091, simple_loss=0.3726, pruned_loss=0.1228, over 28904.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.367, pruned_loss=0.1184, over 5674323.01 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3436, pruned_loss=0.08999, over 5726491.32 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3694, pruned_loss=0.1214, over 5659298.48 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:33:41,699 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7515, 1.7963, 1.9301, 1.5033], device='cuda:0'), covar=tensor([0.1737, 0.2633, 0.1455, 0.1716], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0721, 0.0983, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 09:33:59,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2433, 1.4815, 1.4989, 1.1065], device='cuda:0'), covar=tensor([0.1722, 0.2724, 0.1444, 0.1695], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0721, 0.0983, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 09:34:10,304 INFO [train.py:968] (0/2) Epoch 28, batch 11400, giga_loss[loss=0.3242, simple_loss=0.3768, pruned_loss=0.1358, over 28677.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3683, pruned_loss=0.1197, over 5670634.60 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.0902, over 5721897.99 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3706, pruned_loss=0.1225, over 5661658.30 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:34:27,786 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1242318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:34:31,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1242321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:34:56,053 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.948e+03 2.310e+03 2.940e+03 5.878e+03, threshold=4.621e+03, percent-clipped=3.0
+2023-03-14 09:34:57,496 INFO [train.py:968] (0/2) Epoch 28, batch 11450, giga_loss[loss=0.3674, simple_loss=0.4094, pruned_loss=0.1627, over 26568.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3689, pruned_loss=0.1197, over 5665243.49 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09026, over 5717715.15 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1227, over 5659794.41 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:34:58,453 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1242350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:35:24,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6909, 4.5431, 4.3286, 1.8779], device='cuda:0'), covar=tensor([0.0536, 0.0652, 0.0778, 0.2015], device='cuda:0'), in_proj_covar=tensor([0.1318, 0.1215, 0.1026, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 09:35:50,041 INFO [train.py:968] (0/2) Epoch 28, batch 11500, giga_loss[loss=0.3296, simple_loss=0.3949, pruned_loss=0.1321, over 28295.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3701, pruned_loss=0.1221, over 5653921.38 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.344, pruned_loss=0.09014, over 5719877.57 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3722, pruned_loss=0.1248, over 5647109.27 frames. ], batch size: 77, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:35:55,728 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 09:35:58,596 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1242407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:36:00,336 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 09:36:34,544 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.179e+03 1.798e+03 2.394e+03 2.941e+03 6.659e+03, threshold=4.788e+03, percent-clipped=7.0
+2023-03-14 09:36:36,124 INFO [train.py:968] (0/2) Epoch 28, batch 11550, giga_loss[loss=0.2737, simple_loss=0.3414, pruned_loss=0.103, over 28846.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3697, pruned_loss=0.1221, over 5656495.19 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3439, pruned_loss=0.09021, over 5720731.72 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3718, pruned_loss=0.1246, over 5649631.86 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:37:25,133 INFO [train.py:968] (0/2) Epoch 28, batch 11600, giga_loss[loss=0.3167, simple_loss=0.3828, pruned_loss=0.1253, over 28649.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3699, pruned_loss=0.1217, over 5665923.53 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.09022, over 5722585.68 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3721, pruned_loss=0.1243, over 5657905.16 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:38:11,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.251e+03 1.790e+03 2.542e+03 4.244e+03 1.115e+04, threshold=5.084e+03, percent-clipped=18.0
+2023-03-14 09:38:12,476 INFO [train.py:968] (0/2) Epoch 28, batch 11650, giga_loss[loss=0.2712, simple_loss=0.3436, pruned_loss=0.09942, over 29065.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3704, pruned_loss=0.1214, over 5664145.71 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3441, pruned_loss=0.09037, over 5723035.49 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3723, pruned_loss=0.1239, over 5656298.48 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:38:23,114 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6648, 1.6414, 1.8343, 1.4473], device='cuda:0'), covar=tensor([0.1826, 0.2715, 0.1478, 0.1781], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0722, 0.0984, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 09:38:25,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1242562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:39:01,688 INFO [train.py:968] (0/2) Epoch 28, batch 11700, giga_loss[loss=0.3486, simple_loss=0.404, pruned_loss=0.1467, over 28628.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3697, pruned_loss=0.1205, over 5682446.61 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3436, pruned_loss=0.09019, over 5727940.55 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3724, pruned_loss=0.1235, over 5670345.03 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:39:35,089 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 09:39:40,408 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3559, 1.2226, 4.0757, 3.4156], device='cuda:0'), covar=tensor([0.1734, 0.2941, 0.0509, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0677, 0.1013, 0.0990], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 09:39:49,913 INFO [train.py:968] (0/2) Epoch 28, batch 11750, giga_loss[loss=0.3304, simple_loss=0.3888, pruned_loss=0.136, over 28684.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3721, pruned_loss=0.1227, over 5676732.99 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3435, pruned_loss=0.09014, over 5729272.83 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3747, pruned_loss=0.1256, over 5665345.77 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:39:51,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.262e+03 1.835e+03 2.401e+03 3.141e+03 1.315e+04, threshold=4.803e+03, percent-clipped=7.0
+2023-03-14 09:39:56,338 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-14 09:40:35,935 INFO [train.py:968] (0/2) Epoch 28, batch 11800, giga_loss[loss=0.3465, simple_loss=0.4039, pruned_loss=0.1445, over 28947.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3707, pruned_loss=0.1211, over 5686892.54 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09011, over 5731990.79 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3737, pruned_loss=0.1246, over 5673448.09 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:40:42,202 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1242705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:40:45,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1242708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:40:57,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7848, 1.9227, 1.4764, 1.5886], device='cuda:0'), covar=tensor([0.1127, 0.0801, 0.1032, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0452, 0.0521, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 09:41:13,582 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1242737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:41:25,654 INFO [train.py:968] (0/2) Epoch 28, batch 11850, giga_loss[loss=0.2904, simple_loss=0.3417, pruned_loss=0.1196, over 23599.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3714, pruned_loss=0.1207, over 5678790.41 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.09018, over 5733320.52 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3739, pruned_loss=0.1237, over 5666532.66 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:41:26,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.164e+03 1.665e+03 2.086e+03 2.992e+03 7.333e+03, threshold=4.172e+03, percent-clipped=6.0
+2023-03-14 09:41:57,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1242782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:42:14,770 INFO [train.py:968] (0/2) Epoch 28, batch 11900, giga_loss[loss=0.2935, simple_loss=0.3709, pruned_loss=0.108, over 28947.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.371, pruned_loss=0.1199, over 5672911.46 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3437, pruned_loss=0.09012, over 5736284.56 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3736, pruned_loss=0.1228, over 5659621.91 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:43:00,342 INFO [train.py:968] (0/2) Epoch 28, batch 11950, giga_loss[loss=0.2763, simple_loss=0.3482, pruned_loss=0.1022, over 29055.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3686, pruned_loss=0.118, over 5685879.90 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3432, pruned_loss=0.08974, over 5742753.22 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3721, pruned_loss=0.1217, over 5667363.33 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:43:01,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.725e+03 2.346e+03 3.321e+03 7.712e+03, threshold=4.693e+03, percent-clipped=12.0
+2023-03-14 09:43:26,494 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1242877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:43:46,809 INFO [train.py:968] (0/2) Epoch 28, batch 12000, libri_loss[loss=0.3134, simple_loss=0.3795, pruned_loss=0.1237, over 20233.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3688, pruned_loss=0.1186, over 5672168.82 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3435, pruned_loss=0.08997, over 5734750.74 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3715, pruned_loss=0.1217, over 5664866.46 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:43:46,813 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 09:43:55,222 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2067, simple_loss=0.3148, pruned_loss=0.04926, over 944034.00 frames. 
+2023-03-14 09:43:55,222 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 09:44:19,560 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1242925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:44:21,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1242928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:44:40,462 INFO [train.py:968] (0/2) Epoch 28, batch 12050, libri_loss[loss=0.2839, simple_loss=0.3721, pruned_loss=0.09779, over 29474.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3712, pruned_loss=0.1203, over 5671878.68 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3437, pruned_loss=0.08995, over 5738725.76 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3739, pruned_loss=0.1235, over 5660803.53 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:44:41,078 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.664e+03 2.127e+03 2.763e+03 5.488e+03, threshold=4.255e+03, percent-clipped=2.0
+2023-03-14 09:44:47,367 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1242957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:45:25,309 INFO [train.py:968] (0/2) Epoch 28, batch 12100, libri_loss[loss=0.2528, simple_loss=0.3347, pruned_loss=0.08541, over 29481.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3722, pruned_loss=0.121, over 5669601.01 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3434, pruned_loss=0.08991, over 5735268.46 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3758, pruned_loss=0.125, over 5660318.02 frames. ], batch size: 70, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:46:14,284 INFO [train.py:968] (0/2) Epoch 28, batch 12150, giga_loss[loss=0.2715, simple_loss=0.3506, pruned_loss=0.09616, over 28995.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.372, pruned_loss=0.1218, over 5669212.36 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.09022, over 5738321.78 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.375, pruned_loss=0.1253, over 5657859.28 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:46:15,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 1.772e+03 2.396e+03 3.329e+03 8.690e+03, threshold=4.793e+03, percent-clipped=12.0
+2023-03-14 09:46:42,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3627, 1.2578, 3.9801, 3.2158], device='cuda:0'), covar=tensor([0.1716, 0.2900, 0.0470, 0.1002], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0677, 0.1013, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 09:46:58,120 INFO [train.py:968] (0/2) Epoch 28, batch 12200, giga_loss[loss=0.3068, simple_loss=0.3793, pruned_loss=0.1172, over 28923.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3715, pruned_loss=0.1216, over 5667625.59 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3439, pruned_loss=0.0902, over 5737236.92 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3745, pruned_loss=0.1251, over 5657794.51 frames. ], batch size: 174, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:47:35,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0473, 2.4522, 1.7083, 1.9387], device='cuda:0'), covar=tensor([0.1167, 0.0674, 0.1118, 0.1152], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0453, 0.0524, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 09:47:49,932 INFO [train.py:968] (0/2) Epoch 28, batch 12250, giga_loss[loss=0.2966, simple_loss=0.3615, pruned_loss=0.1158, over 28849.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3734, pruned_loss=0.1234, over 5664692.98 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09005, over 5739816.38 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3763, pruned_loss=0.1267, over 5653585.38 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:47:50,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.697e+03 2.221e+03 3.069e+03 1.635e+04, threshold=4.441e+03, percent-clipped=12.0
+2023-03-14 09:48:15,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5227, 1.8323, 1.5532, 1.5834], device='cuda:0'), covar=tensor([0.0749, 0.0320, 0.0327, 0.0861], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 09:48:28,350 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:48:36,403 INFO [train.py:968] (0/2) Epoch 28, batch 12300, giga_loss[loss=0.2855, simple_loss=0.3452, pruned_loss=0.1129, over 28963.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3728, pruned_loss=0.1227, over 5657853.10 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3436, pruned_loss=0.09014, over 5733610.53 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3759, pruned_loss=0.126, over 5653126.69 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:49:04,795 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:49:14,780 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:49:25,809 INFO [train.py:968] (0/2) Epoch 28, batch 12350, giga_loss[loss=0.3159, simple_loss=0.3758, pruned_loss=0.128, over 28700.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3731, pruned_loss=0.1235, over 5636913.45 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3439, pruned_loss=0.0903, over 5734427.75 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3756, pruned_loss=0.1262, over 5631754.57 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:49:26,491 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.747e+03 2.092e+03 2.862e+03 7.287e+03, threshold=4.184e+03, percent-clipped=6.0
+2023-03-14 09:49:28,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:50:08,548 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 09:50:11,983 INFO [train.py:968] (0/2) Epoch 28, batch 12400, giga_loss[loss=0.3004, simple_loss=0.3747, pruned_loss=0.113, over 28589.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3731, pruned_loss=0.123, over 5637281.17 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.09019, over 5727793.06 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3763, pruned_loss=0.1265, over 5636002.90 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:50:13,988 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:50:29,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7775, 1.8329, 1.9426, 1.5424], device='cuda:0'), covar=tensor([0.1706, 0.2430, 0.1414, 0.1663], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0720, 0.0982, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 09:50:55,950 INFO [train.py:968] (0/2) Epoch 28, batch 12450, giga_loss[loss=0.2603, simple_loss=0.3385, pruned_loss=0.09105, over 28964.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3738, pruned_loss=0.123, over 5640809.61 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3441, pruned_loss=0.09038, over 5728465.46 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3763, pruned_loss=0.126, over 5638532.98 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:50:57,498 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.154e+03 1.689e+03 2.147e+03 3.012e+03 7.509e+03, threshold=4.294e+03, percent-clipped=11.0
+2023-03-14 09:51:09,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6360, 1.6353, 1.7343, 1.3926], device='cuda:0'), covar=tensor([0.1535, 0.2599, 0.1399, 0.1696], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0721, 0.0983, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 09:51:10,425 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:51:43,173 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:51:45,208 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1243398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:51:45,633 INFO [train.py:968] (0/2) Epoch 28, batch 12500, giga_loss[loss=0.2457, simple_loss=0.33, pruned_loss=0.08066, over 28950.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.372, pruned_loss=0.1219, over 5637818.53 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3444, pruned_loss=0.09053, over 5717508.56 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3741, pruned_loss=0.1246, over 5645079.17 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:51:49,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8834, 3.7260, 3.5511, 1.9328], device='cuda:0'), covar=tensor([0.0764, 0.0872, 0.0888, 0.1928], device='cuda:0'), in_proj_covar=tensor([0.1325, 0.1221, 0.1032, 0.0761], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 09:52:12,504 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:52:36,361 INFO [train.py:968] (0/2) Epoch 28, batch 12550, giga_loss[loss=0.3605, simple_loss=0.4058, pruned_loss=0.1576, over 28589.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.37, pruned_loss=0.1207, over 5647464.59 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3447, pruned_loss=0.09073, over 5718045.02 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3715, pruned_loss=0.1228, over 5652287.02 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:52:37,998 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.164e+03 2.184e+03 2.880e+03 4.442e+03 1.658e+04, threshold=5.760e+03, percent-clipped=25.0
+2023-03-14 09:53:20,265 INFO [train.py:968] (0/2) Epoch 28, batch 12600, giga_loss[loss=0.2888, simple_loss=0.3576, pruned_loss=0.11, over 28624.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3668, pruned_loss=0.1185, over 5666404.65 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3449, pruned_loss=0.09071, over 5724033.81 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3686, pruned_loss=0.1211, over 5663092.91 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:54:04,421 INFO [train.py:968] (0/2) Epoch 28, batch 12650, giga_loss[loss=0.3253, simple_loss=0.3749, pruned_loss=0.1378, over 28792.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3625, pruned_loss=0.1166, over 5652617.10 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3443, pruned_loss=0.09041, over 5730769.40 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3653, pruned_loss=0.12, over 5641271.01 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:54:05,955 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.198e+03 1.863e+03 2.366e+03 3.176e+03 9.543e+03, threshold=4.733e+03, percent-clipped=3.0
+2023-03-14 09:54:18,896 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243563.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 09:54:19,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:54:30,241 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.39 vs. limit=5.0
+2023-03-14 09:54:50,788 INFO [train.py:968] (0/2) Epoch 28, batch 12700, giga_loss[loss=0.3234, simple_loss=0.3776, pruned_loss=0.1346, over 27942.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3628, pruned_loss=0.1176, over 5657821.49 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3442, pruned_loss=0.09028, over 5732420.14 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3656, pruned_loss=0.1212, over 5645369.45 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:54:53,951 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:55:02,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:55:37,393 INFO [train.py:968] (0/2) Epoch 28, batch 12750, giga_loss[loss=0.313, simple_loss=0.3617, pruned_loss=0.1321, over 28532.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3614, pruned_loss=0.1167, over 5657689.45 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3442, pruned_loss=0.09022, over 5735051.11 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.364, pruned_loss=0.1202, over 5643582.62 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:55:38,093 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.760e+03 2.252e+03 2.885e+03 8.596e+03, threshold=4.504e+03, percent-clipped=5.0
+2023-03-14 09:55:58,981 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 09:56:05,768 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:56:25,944 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3383, 1.8167, 1.5181, 1.4704], device='cuda:0'), covar=tensor([0.0782, 0.0356, 0.0343, 0.0866], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 09:56:27,927 INFO [train.py:968] (0/2) Epoch 28, batch 12800, giga_loss[loss=0.2933, simple_loss=0.3687, pruned_loss=0.1089, over 28787.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3603, pruned_loss=0.1143, over 5657739.02 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3442, pruned_loss=0.09018, over 5737391.19 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3626, pruned_loss=0.1175, over 5643495.10 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:56:36,122 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:56:39,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1243710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:05,691 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:08,185 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:14,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:17,482 INFO [train.py:968] (0/2) Epoch 28, batch 12850, giga_loss[loss=0.2622, simple_loss=0.3336, pruned_loss=0.09543, over 28506.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3581, pruned_loss=0.1105, over 5652564.76 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.09017, over 5732588.12 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3605, pruned_loss=0.1137, over 5642791.01 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:57:17,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1243749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:18,246 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.212e+03 1.607e+03 1.993e+03 2.871e+03 8.993e+03, threshold=3.986e+03, percent-clipped=7.0
+2023-03-14 09:57:26,719 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:28,580 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1243759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:48,191 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:57,046 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:58:08,279 INFO [train.py:968] (0/2) Epoch 28, batch 12900, giga_loss[loss=0.2504, simple_loss=0.3356, pruned_loss=0.08264, over 28843.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3558, pruned_loss=0.1076, over 5651727.12 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3441, pruned_loss=0.09032, over 5734865.07 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3578, pruned_loss=0.1102, over 5640885.60 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:58:23,975 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-14 09:58:28,622 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:58:30,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1243822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:58:58,236 INFO [train.py:968] (0/2) Epoch 28, batch 12950, giga_loss[loss=0.2633, simple_loss=0.34, pruned_loss=0.09331, over 29027.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.353, pruned_loss=0.105, over 5649913.10 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3431, pruned_loss=0.08998, over 5729605.56 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3557, pruned_loss=0.1079, over 5642926.27 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:58:58,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.728e+03 2.266e+03 3.221e+03 6.578e+03, threshold=4.533e+03, percent-clipped=10.0
+2023-03-14 09:58:59,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:59:34,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:59:38,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1243887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:59:43,442 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0538, 1.3542, 1.3140, 1.0836], device='cuda:0'), covar=tensor([0.3039, 0.2341, 0.1672, 0.2332], device='cuda:0'), in_proj_covar=tensor([0.2049, 0.2014, 0.1923, 0.2064], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 09:59:49,566 INFO [train.py:968] (0/2) Epoch 28, batch 13000, giga_loss[loss=0.2544, simple_loss=0.344, pruned_loss=0.08238, over 28023.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.35, pruned_loss=0.1018, over 5648635.99 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3432, pruned_loss=0.0901, over 5732250.87 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3523, pruned_loss=0.1041, over 5639962.02 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:00:06,475 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:00:11,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6463, 1.9134, 1.5559, 1.5632], device='cuda:0'), covar=tensor([0.2682, 0.2629, 0.3131, 0.2407], device='cuda:0'), in_proj_covar=tensor([0.1602, 0.1152, 0.1417, 0.1015], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 10:00:28,733 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243938.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:00:40,123 INFO [train.py:968] (0/2) Epoch 28, batch 13050, libri_loss[loss=0.2625, simple_loss=0.3414, pruned_loss=0.09175, over 28009.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3497, pruned_loss=0.09933, over 5661612.96 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3431, pruned_loss=0.09008, over 5735234.16 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3517, pruned_loss=0.1014, over 5650257.69 frames. ], batch size: 116, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:00:41,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.550e+03 2.055e+03 2.732e+03 1.000e+04, threshold=4.109e+03, percent-clipped=5.0
+2023-03-14 10:01:30,641 INFO [train.py:968] (0/2) Epoch 28, batch 13100, libri_loss[loss=0.2073, simple_loss=0.2843, pruned_loss=0.06517, over 28083.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3499, pruned_loss=0.09964, over 5657975.98 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.342, pruned_loss=0.08962, over 5737168.37 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3527, pruned_loss=0.1019, over 5645952.15 frames. ], batch size: 62, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:01:31,290 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1244000.pt
+2023-03-14 10:02:13,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244041.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:02:22,663 INFO [train.py:968] (0/2) Epoch 28, batch 13150, giga_loss[loss=0.2745, simple_loss=0.3525, pruned_loss=0.09827, over 28534.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3491, pruned_loss=0.09897, over 5659240.38 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3421, pruned_loss=0.0897, over 5737702.53 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3512, pruned_loss=0.1007, over 5649049.82 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:02:22,898 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:02:24,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.981e+02 1.490e+03 1.866e+03 2.487e+03 8.050e+03, threshold=3.733e+03, percent-clipped=5.0
+2023-03-14 10:02:35,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:02:55,616 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244081.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:02:58,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244084.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:03:13,668 INFO [train.py:968] (0/2) Epoch 28, batch 13200, giga_loss[loss=0.2761, simple_loss=0.3425, pruned_loss=0.1048, over 27847.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3459, pruned_loss=0.09737, over 5643205.96 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3419, pruned_loss=0.08965, over 5740469.57 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.348, pruned_loss=0.09893, over 5631592.39 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:03:28,753 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244113.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:04:02,385 INFO [train.py:968] (0/2) Epoch 28, batch 13250, giga_loss[loss=0.2801, simple_loss=0.3549, pruned_loss=0.1027, over 27906.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.346, pruned_loss=0.09743, over 5646843.85 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.341, pruned_loss=0.08921, over 5743611.45 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3485, pruned_loss=0.09922, over 5633118.71 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:04:04,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.911e+02 1.438e+03 1.849e+03 2.461e+03 6.119e+03, threshold=3.698e+03, percent-clipped=6.0
+2023-03-14 10:04:12,934 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:04:51,901 INFO [train.py:968] (0/2) Epoch 28, batch 13300, giga_loss[loss=0.2853, simple_loss=0.3491, pruned_loss=0.1107, over 27607.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3452, pruned_loss=0.09662, over 5650338.74 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3406, pruned_loss=0.08906, over 5747769.74 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3475, pruned_loss=0.09836, over 5633642.95 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:05:25,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4535, 1.9841, 1.6714, 1.6937], device='cuda:0'), covar=tensor([0.0714, 0.0278, 0.0313, 0.0752], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 10:05:43,794 INFO [train.py:968] (0/2) Epoch 28, batch 13350, giga_loss[loss=0.2417, simple_loss=0.3282, pruned_loss=0.07759, over 28602.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3423, pruned_loss=0.09416, over 5653787.59 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3402, pruned_loss=0.08886, over 5751015.50 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3447, pruned_loss=0.09583, over 5635849.13 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:05:48,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.822e+02 1.576e+03 2.037e+03 2.926e+03 2.244e+04, threshold=4.073e+03, percent-clipped=14.0
+2023-03-14 10:06:32,936 INFO [train.py:968] (0/2) Epoch 28, batch 13400, libri_loss[loss=0.2493, simple_loss=0.3312, pruned_loss=0.08365, over 29516.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3392, pruned_loss=0.09177, over 5654201.85 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3402, pruned_loss=0.08896, over 5748960.63 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.341, pruned_loss=0.09316, over 5638435.65 frames. ], batch size: 84, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:07:28,438 INFO [train.py:968] (0/2) Epoch 28, batch 13450, giga_loss[loss=0.2337, simple_loss=0.3149, pruned_loss=0.07626, over 27936.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3357, pruned_loss=0.09016, over 5647245.90 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3401, pruned_loss=0.08886, over 5742066.03 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3372, pruned_loss=0.09136, over 5639504.36 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:07:31,867 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.350e+02 1.412e+03 1.816e+03 2.515e+03 7.822e+03, threshold=3.632e+03, percent-clipped=4.0
+2023-03-14 10:07:43,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9201, 3.0457, 1.9140, 1.0763], device='cuda:0'), covar=tensor([0.8372, 0.3217, 0.3682, 0.6202], device='cuda:0'), in_proj_covar=tensor([0.1837, 0.1721, 0.1653, 0.1496], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 10:07:55,810 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:08:19,313 INFO [train.py:968] (0/2) Epoch 28, batch 13500, giga_loss[loss=0.3038, simple_loss=0.3698, pruned_loss=0.1188, over 28608.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.336, pruned_loss=0.09089, over 5658898.94 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3396, pruned_loss=0.0888, over 5745533.19 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3376, pruned_loss=0.09197, over 5646197.67 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:08:38,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244416.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:08:46,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:08:59,308 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:09:13,521 INFO [train.py:968] (0/2) Epoch 28, batch 13550, giga_loss[loss=0.2755, simple_loss=0.3495, pruned_loss=0.1008, over 27572.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3362, pruned_loss=0.09198, over 5648584.01 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3393, pruned_loss=0.08874, over 5747928.09 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3377, pruned_loss=0.09294, over 5635192.11 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:09:17,677 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.477e+02 1.507e+03 1.939e+03 2.899e+03 1.114e+04, threshold=3.878e+03, percent-clipped=14.0
+2023-03-14 10:10:12,999 INFO [train.py:968] (0/2) Epoch 28, batch 13600, giga_loss[loss=0.2352, simple_loss=0.3082, pruned_loss=0.08114, over 24211.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3377, pruned_loss=0.09152, over 5654363.34 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.339, pruned_loss=0.0886, over 5750239.63 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3391, pruned_loss=0.09246, over 5639970.20 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:10:52,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:11:12,586 INFO [train.py:968] (0/2) Epoch 28, batch 13650, giga_loss[loss=0.251, simple_loss=0.3396, pruned_loss=0.08124, over 28623.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3396, pruned_loss=0.0916, over 5662062.27 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3387, pruned_loss=0.08847, over 5753997.62 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3409, pruned_loss=0.09252, over 5645262.79 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:11:14,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4993, 5.3246, 5.1076, 2.6763], device='cuda:0'), covar=tensor([0.0528, 0.0698, 0.0849, 0.1684], device='cuda:0'), in_proj_covar=tensor([0.1298, 0.1200, 0.1009, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 10:11:16,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.857e+02 1.432e+03 1.983e+03 2.710e+03 1.027e+04, threshold=3.967e+03, percent-clipped=7.0
+2023-03-14 10:11:24,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244559.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:11:26,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244562.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:11:33,859 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:11:37,750 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:11:42,282 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:11:48,112 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:11:51,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:12:05,512 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244591.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:12:13,210 INFO [train.py:968] (0/2) Epoch 28, batch 13700, giga_loss[loss=0.2599, simple_loss=0.3413, pruned_loss=0.08931, over 28662.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3391, pruned_loss=0.09101, over 5673885.56 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3386, pruned_loss=0.0884, over 5756286.13 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3402, pruned_loss=0.09183, over 5657355.40 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:12:13,798 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:12:30,434 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:13:11,902 INFO [train.py:968] (0/2) Epoch 28, batch 13750, giga_loss[loss=0.2704, simple_loss=0.3436, pruned_loss=0.09861, over 26767.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3361, pruned_loss=0.08921, over 5664410.07 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3383, pruned_loss=0.08833, over 5749499.94 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3372, pruned_loss=0.08996, over 5655274.01 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:13:15,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.483e+03 1.878e+03 2.470e+03 8.204e+03, threshold=3.755e+03, percent-clipped=4.0
+2023-03-14 10:13:32,183 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-14 10:13:34,535 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:13:47,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:13:51,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:14:13,930 INFO [train.py:968] (0/2) Epoch 28, batch 13800, giga_loss[loss=0.2335, simple_loss=0.3233, pruned_loss=0.07185, over 28496.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3352, pruned_loss=0.0874, over 5656196.17 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3382, pruned_loss=0.08843, over 5742691.90 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3361, pruned_loss=0.08789, over 5654275.55 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:14:21,865 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4441, 1.1384, 4.0585, 3.3934], device='cuda:0'), covar=tensor([0.1621, 0.3198, 0.0433, 0.1368], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0672, 0.1000, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 10:14:23,483 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:14:23,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3736, 1.3011, 1.2734, 1.6175], device='cuda:0'), covar=tensor([0.0803, 0.0362, 0.0362, 0.0916], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 10:14:24,995 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-14 10:14:30,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4467, 1.7253, 1.6704, 1.3604], device='cuda:0'), covar=tensor([0.2708, 0.2310, 0.1866, 0.2459], device='cuda:0'), in_proj_covar=tensor([0.2027, 0.1993, 0.1895, 0.2041], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 10:14:41,162 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2003, 1.5910, 1.1864, 0.4426], device='cuda:0'), covar=tensor([0.3695, 0.2215, 0.3344, 0.6152], device='cuda:0'), in_proj_covar=tensor([0.1843, 0.1725, 0.1658, 0.1501], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 10:15:10,311 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:15:11,780 INFO [train.py:968] (0/2) Epoch 28, batch 13850, giga_loss[loss=0.2435, simple_loss=0.3157, pruned_loss=0.08561, over 27597.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3334, pruned_loss=0.08613, over 5658629.13 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3383, pruned_loss=0.08848, over 5747566.67 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3339, pruned_loss=0.08641, over 5650226.05 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:15:15,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.651e+02 1.285e+03 1.850e+03 2.296e+03 5.531e+03, threshold=3.699e+03, percent-clipped=6.0
+2023-03-14 10:15:57,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3582, 1.5926, 1.6623, 1.3755], device='cuda:0'), covar=tensor([0.3180, 0.2295, 0.1816, 0.2411], device='cuda:0'), in_proj_covar=tensor([0.2026, 0.1992, 0.1893, 0.2039], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 10:16:08,637 INFO [train.py:968] (0/2) Epoch 28, batch 13900, giga_loss[loss=0.3147, simple_loss=0.3665, pruned_loss=0.1314, over 26864.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3315, pruned_loss=0.08636, over 5674005.38 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3375, pruned_loss=0.08819, over 5753420.75 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3324, pruned_loss=0.08675, over 5658258.01 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:16:21,168 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:16:48,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6739, 1.8318, 1.5134, 1.9582], device='cuda:0'), covar=tensor([0.2820, 0.2858, 0.3243, 0.2518], device='cuda:0'), in_proj_covar=tensor([0.1599, 0.1147, 0.1414, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 10:17:03,016 INFO [train.py:968] (0/2) Epoch 28, batch 13950, giga_loss[loss=0.2503, simple_loss=0.3228, pruned_loss=0.08893, over 28953.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3307, pruned_loss=0.08636, over 5675326.27 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3375, pruned_loss=0.08827, over 5754891.82 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3313, pruned_loss=0.08654, over 5659108.89 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:17:08,125 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.646e+02 1.415e+03 2.035e+03 2.950e+03 7.301e+03, threshold=4.070e+03, percent-clipped=8.0
+2023-03-14 10:17:48,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:17:52,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:17:58,127 INFO [train.py:968] (0/2) Epoch 28, batch 14000, giga_loss[loss=0.2475, simple_loss=0.3361, pruned_loss=0.07947, over 28984.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3308, pruned_loss=0.08642, over 5675963.33 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.337, pruned_loss=0.0881, over 5758498.67 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3315, pruned_loss=0.08668, over 5656767.36 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:18:25,865 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:19:01,201 INFO [train.py:968] (0/2) Epoch 28, batch 14050, giga_loss[loss=0.2757, simple_loss=0.3473, pruned_loss=0.102, over 27666.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3335, pruned_loss=0.08748, over 5664052.93 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.337, pruned_loss=0.08822, over 5760701.76 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.334, pruned_loss=0.08755, over 5645971.04 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:19:02,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:19:06,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.221e+02 1.535e+03 2.098e+03 3.026e+03 7.968e+03, threshold=4.197e+03, percent-clipped=11.0
+2023-03-14 10:20:02,852 INFO [train.py:968] (0/2) Epoch 28, batch 14100, giga_loss[loss=0.2346, simple_loss=0.3196, pruned_loss=0.07475, over 28634.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3324, pruned_loss=0.08629, over 5665209.45 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3369, pruned_loss=0.08829, over 5754400.66 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3328, pruned_loss=0.08626, over 5653515.29 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:20:13,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6910, 1.8258, 1.5536, 1.8936], device='cuda:0'), covar=tensor([0.2811, 0.2836, 0.3193, 0.2416], device='cuda:0'), in_proj_covar=tensor([0.1601, 0.1149, 0.1415, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 10:20:14,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6437, 1.9616, 1.5928, 1.7590], device='cuda:0'), covar=tensor([0.0732, 0.0267, 0.0326, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 10:20:37,201 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6335, 2.3838, 1.7671, 0.8440], device='cuda:0'), covar=tensor([0.6915, 0.3782, 0.4857, 0.7517], device='cuda:0'), in_proj_covar=tensor([0.1839, 0.1724, 0.1655, 0.1497], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 10:20:54,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:21:04,884 INFO [train.py:968] (0/2) Epoch 28, batch 14150, giga_loss[loss=0.2431, simple_loss=0.3317, pruned_loss=0.07726, over 29000.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3301, pruned_loss=0.08512, over 5678548.45 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3363, pruned_loss=0.08823, over 5760660.95 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3307, pruned_loss=0.08504, over 5660410.92 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:21:11,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.937e+02 1.535e+03 2.012e+03 2.834e+03 6.724e+03, threshold=4.023e+03, percent-clipped=9.0
+2023-03-14 10:21:54,240 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:21:56,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1451, 1.5492, 1.4978, 1.3578], device='cuda:0'), covar=tensor([0.2227, 0.1813, 0.2134, 0.1897], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0751, 0.0721, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 10:21:58,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:22:01,040 INFO [train.py:968] (0/2) Epoch 28, batch 14200, giga_loss[loss=0.2642, simple_loss=0.3563, pruned_loss=0.08603, over 28636.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3309, pruned_loss=0.08536, over 5694895.37 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3355, pruned_loss=0.08788, over 5766142.64 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.332, pruned_loss=0.0855, over 5671580.31 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:22:37,754 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:22:37,770 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:23:09,695 INFO [train.py:968] (0/2) Epoch 28, batch 14250, giga_loss[loss=0.1794, simple_loss=0.2588, pruned_loss=0.05004, over 24487.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3356, pruned_loss=0.08601, over 5677392.17 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3353, pruned_loss=0.08783, over 5758867.01 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3365, pruned_loss=0.08614, over 5664106.74 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:23:17,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.585e+03 2.173e+03 3.265e+03 9.656e+03, threshold=4.347e+03, percent-clipped=17.0
+2023-03-14 10:23:47,376 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:23:49,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:23:50,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:24:06,937 INFO [train.py:968] (0/2) Epoch 28, batch 14300, giga_loss[loss=0.253, simple_loss=0.3385, pruned_loss=0.08374, over 27760.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3379, pruned_loss=0.08576, over 5683112.69 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3351, pruned_loss=0.08785, over 5762886.58 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3388, pruned_loss=0.08577, over 5666208.53 frames. ], batch size: 474, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:24:29,401 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:24:42,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:25:05,660 INFO [train.py:968] (0/2) Epoch 28, batch 14350, giga_loss[loss=0.2359, simple_loss=0.3225, pruned_loss=0.07464, over 27702.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3377, pruned_loss=0.0843, over 5678254.80 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.335, pruned_loss=0.08777, over 5762743.28 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3386, pruned_loss=0.08432, over 5663205.76 frames. ], batch size: 474, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:25:12,167 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.017e+02 1.529e+03 1.930e+03 2.868e+03 1.217e+04, threshold=3.860e+03, percent-clipped=8.0
+2023-03-14 10:25:21,965 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-14 10:26:04,840 INFO [train.py:968] (0/2) Epoch 28, batch 14400, giga_loss[loss=0.2672, simple_loss=0.3513, pruned_loss=0.09159, over 28378.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3367, pruned_loss=0.08359, over 5670613.57 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.335, pruned_loss=0.08773, over 5753154.99 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3375, pruned_loss=0.08353, over 5664096.11 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:26:23,081 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:26:44,215 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:26:47,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:27:08,570 INFO [train.py:968] (0/2) Epoch 28, batch 14450, giga_loss[loss=0.2474, simple_loss=0.3251, pruned_loss=0.08483, over 28883.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3358, pruned_loss=0.08386, over 5673636.43 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3348, pruned_loss=0.08761, over 5755013.60 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3366, pruned_loss=0.08387, over 5665799.56 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:27:15,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.167e+02 1.406e+03 1.733e+03 2.446e+03 4.684e+03, threshold=3.466e+03, percent-clipped=4.0
+2023-03-14 10:27:23,468 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:27:40,644 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:28:21,889 INFO [train.py:968] (0/2) Epoch 28, batch 14500, giga_loss[loss=0.3022, simple_loss=0.3631, pruned_loss=0.1206, over 26887.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3366, pruned_loss=0.08554, over 5685802.13 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3345, pruned_loss=0.08753, over 5756678.50 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3376, pruned_loss=0.08561, over 5677449.16 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:29:34,236 INFO [train.py:968] (0/2) Epoch 28, batch 14550, giga_loss[loss=0.2446, simple_loss=0.3063, pruned_loss=0.0915, over 24298.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3347, pruned_loss=0.08555, over 5686686.45 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3332, pruned_loss=0.08702, over 5762793.77 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3368, pruned_loss=0.08602, over 5671509.86 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:29:42,072 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.570e+02 1.440e+03 1.751e+03 2.321e+03 4.055e+03, threshold=3.501e+03, percent-clipped=4.0
+2023-03-14 10:30:21,995 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:30:42,490 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 10:30:46,242 INFO [train.py:968] (0/2) Epoch 28, batch 14600, giga_loss[loss=0.2562, simple_loss=0.3419, pruned_loss=0.08529, over 28850.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3318, pruned_loss=0.08423, over 5690400.63 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3329, pruned_loss=0.08702, over 5764337.54 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3337, pruned_loss=0.08453, over 5675385.13 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:30:48,722 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:31:55,252 INFO [train.py:968] (0/2) Epoch 28, batch 14650, giga_loss[loss=0.2408, simple_loss=0.332, pruned_loss=0.07482, over 28973.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3304, pruned_loss=0.08372, over 5681632.26 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3331, pruned_loss=0.08718, over 5758139.97 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3318, pruned_loss=0.08377, over 5673664.73 frames. ], batch size: 175, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:32:04,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.136e+02 1.343e+03 1.913e+03 2.603e+03 6.675e+03, threshold=3.825e+03, percent-clipped=11.0
+2023-03-14 10:32:27,907 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:32:57,834 INFO [train.py:968] (0/2) Epoch 28, batch 14700, giga_loss[loss=0.2325, simple_loss=0.3254, pruned_loss=0.06985, over 28925.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3305, pruned_loss=0.08416, over 5674671.74 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.333, pruned_loss=0.08714, over 5756459.78 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3316, pruned_loss=0.08418, over 5668894.11 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:32:58,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4931, 1.4075, 3.6888, 3.3340], device='cuda:0'), covar=tensor([0.1479, 0.2628, 0.0584, 0.1039], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0673, 0.1000, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 10:32:59,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:33:10,743 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:33:52,536 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:33:56,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:33:59,939 INFO [train.py:968] (0/2) Epoch 28, batch 14750, giga_loss[loss=0.2604, simple_loss=0.3463, pruned_loss=0.08726, over 28985.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3345, pruned_loss=0.0861, over 5671184.61 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3327, pruned_loss=0.08711, over 5748982.00 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3356, pruned_loss=0.08611, over 5670872.73 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:34:06,919 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.480e+03 1.911e+03 2.559e+03 6.191e+03, threshold=3.821e+03, percent-clipped=5.0
+2023-03-14 10:34:29,708 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:34:43,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:34:58,505 INFO [train.py:968] (0/2) Epoch 28, batch 14800, giga_loss[loss=0.2433, simple_loss=0.3194, pruned_loss=0.08353, over 28717.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3332, pruned_loss=0.08656, over 5675636.58 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.333, pruned_loss=0.08726, over 5751698.00 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3339, pruned_loss=0.08639, over 5671686.26 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:35:14,996 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 10:35:15,357 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:35:54,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6635, 1.9679, 1.3737, 1.5311], device='cuda:0'), covar=tensor([0.1065, 0.0561, 0.0990, 0.1114], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0449, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 10:35:57,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:36:00,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245747.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:36:00,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:36:01,294 INFO [train.py:968] (0/2) Epoch 28, batch 14850, giga_loss[loss=0.2838, simple_loss=0.3518, pruned_loss=0.108, over 28948.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3333, pruned_loss=0.0875, over 5682568.17 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3325, pruned_loss=0.08712, over 5753677.75 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3343, pruned_loss=0.08749, over 5675323.67 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:36:06,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.607e+03 2.148e+03 3.006e+03 9.556e+03, threshold=4.296e+03, percent-clipped=13.0
+2023-03-14 10:36:19,681 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-14 10:36:32,512 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:36:41,604 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7059, 2.0177, 1.3423, 1.5476], device='cuda:0'), covar=tensor([0.1045, 0.0572, 0.1077, 0.1167], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0448, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 10:36:58,102 INFO [train.py:968] (0/2) Epoch 28, batch 14900, giga_loss[loss=0.252, simple_loss=0.3352, pruned_loss=0.08442, over 28872.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3328, pruned_loss=0.08723, over 5679483.65 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.332, pruned_loss=0.08703, over 5748136.66 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.334, pruned_loss=0.08731, over 5676779.81 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:37:37,505 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:37:40,496 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:38:06,277 INFO [train.py:968] (0/2) Epoch 28, batch 14950, libri_loss[loss=0.2296, simple_loss=0.2969, pruned_loss=0.08119, over 29655.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3346, pruned_loss=0.08724, over 5682379.98 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3318, pruned_loss=0.08698, over 5751113.01 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3359, pruned_loss=0.08735, over 5676039.04 frames. ], batch size: 69, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:38:13,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.094e+03 1.469e+03 1.872e+03 2.518e+03 5.429e+03, threshold=3.743e+03, percent-clipped=6.0
+2023-03-14 10:38:16,438 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:38:16,788 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-14 10:38:20,083 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:39:06,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245890.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:39:10,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245893.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:39:18,963 INFO [train.py:968] (0/2) Epoch 28, batch 15000, giga_loss[loss=0.2427, simple_loss=0.3295, pruned_loss=0.07793, over 28043.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3344, pruned_loss=0.08644, over 5681829.07 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3316, pruned_loss=0.0869, over 5755635.62 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3356, pruned_loss=0.0866, over 5671092.12 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:39:18,969 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 10:39:27,332 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.1949, simple_loss=0.2957, pruned_loss=0.04704, over 944034.00 frames. 
+2023-03-14 10:39:27,333 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 10:40:06,109 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245922.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:40:15,721 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:40:43,123 INFO [train.py:968] (0/2) Epoch 28, batch 15050, giga_loss[loss=0.3211, simple_loss=0.368, pruned_loss=0.1371, over 26922.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3323, pruned_loss=0.08638, over 5666496.61 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3317, pruned_loss=0.08701, over 5755690.50 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3332, pruned_loss=0.08639, over 5656884.62 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:40:43,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:40:52,109 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.780e+02 1.607e+03 1.998e+03 3.152e+03 5.295e+03, threshold=3.997e+03, percent-clipped=20.0
+2023-03-14 10:41:29,960 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:41:47,794 INFO [train.py:968] (0/2) Epoch 28, batch 15100, giga_loss[loss=0.2053, simple_loss=0.2898, pruned_loss=0.06039, over 29108.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3279, pruned_loss=0.08493, over 5673067.36 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3319, pruned_loss=0.08719, over 5759172.52 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3284, pruned_loss=0.08472, over 5659780.28 frames. ], batch size: 285, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:41:48,855 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1246000.pt
+2023-03-14 10:41:50,520 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:41:53,310 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2731, 1.5349, 1.6248, 1.4168], device='cuda:0'), covar=tensor([0.1734, 0.1293, 0.1583, 0.1432], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0743, 0.0715, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 10:41:54,613 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1246003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:42:27,762 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1246032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:42:48,326 INFO [train.py:968] (0/2) Epoch 28, batch 15150, giga_loss[loss=0.241, simple_loss=0.3184, pruned_loss=0.08181, over 28947.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3258, pruned_loss=0.0839, over 5670800.51 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3323, pruned_loss=0.0874, over 5756827.58 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3258, pruned_loss=0.08353, over 5661623.62 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:42:58,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.644e+02 1.533e+03 2.122e+03 2.840e+03 8.211e+03, threshold=4.243e+03, percent-clipped=7.0
+2023-03-14 10:43:33,844 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:43:38,491 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:43:41,795 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1246095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:43:44,460 INFO [train.py:968] (0/2) Epoch 28, batch 15200, giga_loss[loss=0.26, simple_loss=0.3378, pruned_loss=0.09104, over 29103.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3271, pruned_loss=0.08493, over 5657573.72 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3324, pruned_loss=0.08748, over 5747513.31 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3269, pruned_loss=0.08449, over 5655820.97 frames. ], batch size: 200, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:44:12,286 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1246124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:44:20,192 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:44:22,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1246133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:44:37,064 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2247, 0.7433, 0.8389, 1.4697], device='cuda:0'), covar=tensor([0.0791, 0.0428, 0.0408, 0.0913], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 10:44:47,234 INFO [train.py:968] (0/2) Epoch 28, batch 15250, giga_loss[loss=0.1982, simple_loss=0.2782, pruned_loss=0.0591, over 24273.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3265, pruned_loss=0.08445, over 5660619.69 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3323, pruned_loss=0.08752, over 5748347.74 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3264, pruned_loss=0.08406, over 5658164.78 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:44:54,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.632e+03 2.125e+03 2.936e+03 7.727e+03, threshold=4.250e+03, percent-clipped=10.0
+2023-03-14 10:45:02,841 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1246162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:45:28,748 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.75 vs. limit=5.0
+2023-03-14 10:45:45,007 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4838, 1.9856, 1.7668, 1.7464], device='cuda:0'), covar=tensor([0.2172, 0.1972, 0.2266, 0.1984], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0743, 0.0715, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 10:45:46,684 INFO [train.py:968] (0/2) Epoch 28, batch 15300, giga_loss[loss=0.2571, simple_loss=0.3471, pruned_loss=0.08357, over 28635.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3245, pruned_loss=0.08246, over 5656228.98 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3323, pruned_loss=0.08743, over 5751121.81 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3243, pruned_loss=0.08214, over 5650143.27 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:46:19,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9662, 2.7050, 1.8587, 1.2375], device='cuda:0'), covar=tensor([0.8017, 0.3906, 0.4123, 0.6521], device='cuda:0'), in_proj_covar=tensor([0.1840, 0.1731, 0.1656, 0.1504], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 10:46:26,476 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:46:31,531 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1246235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:46:47,663 INFO [train.py:968] (0/2) Epoch 28, batch 15350, libri_loss[loss=0.1962, simple_loss=0.2775, pruned_loss=0.05748, over 29666.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3243, pruned_loss=0.08256, over 5672706.48 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3319, pruned_loss=0.08721, over 5757442.61 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3242, pruned_loss=0.08234, over 5658874.27 frames. ], batch size: 73, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:46:58,169 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.280e+02 1.472e+03 1.902e+03 2.236e+03 7.102e+03, threshold=3.805e+03, percent-clipped=4.0
+2023-03-14 10:47:10,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1246264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:47:55,854 INFO [train.py:968] (0/2) Epoch 28, batch 15400, giga_loss[loss=0.263, simple_loss=0.3457, pruned_loss=0.09014, over 28324.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3228, pruned_loss=0.08164, over 5662468.63 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3317, pruned_loss=0.08714, over 5759297.93 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3228, pruned_loss=0.08143, over 5648636.86 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:48:04,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:48:18,875 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1246318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:48:48,982 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3797, 1.7529, 1.7387, 1.4866], device='cuda:0'), covar=tensor([0.2185, 0.2122, 0.2232, 0.2235], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0743, 0.0715, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 10:48:57,647 INFO [train.py:968] (0/2) Epoch 28, batch 15450, giga_loss[loss=0.2355, simple_loss=0.3234, pruned_loss=0.07384, over 28709.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3233, pruned_loss=0.08126, over 5663559.39 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3317, pruned_loss=0.08698, over 5762703.19 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.323, pruned_loss=0.08109, over 5646759.90 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:49:00,811 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5538, 2.0036, 1.8175, 1.7335], device='cuda:0'), covar=tensor([0.2170, 0.2402, 0.2268, 0.2286], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0743, 0.0715, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 10:49:08,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.206e+02 1.306e+03 1.631e+03 2.386e+03 6.485e+03, threshold=3.262e+03, percent-clipped=1.0
+2023-03-14 10:50:04,565 INFO [train.py:968] (0/2) Epoch 28, batch 15500, giga_loss[loss=0.2759, simple_loss=0.3481, pruned_loss=0.1019, over 28982.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3241, pruned_loss=0.08235, over 5669946.82 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3313, pruned_loss=0.08679, over 5764867.71 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3241, pruned_loss=0.08231, over 5653592.05 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:50:06,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2387, 3.1022, 2.9402, 1.5105], device='cuda:0'), covar=tensor([0.1036, 0.1066, 0.1048, 0.2276], device='cuda:0'), in_proj_covar=tensor([0.1283, 0.1181, 0.0995, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 10:51:09,385 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:51:11,664 INFO [train.py:968] (0/2) Epoch 28, batch 15550, giga_loss[loss=0.2675, simple_loss=0.3435, pruned_loss=0.09578, over 28955.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3233, pruned_loss=0.08219, over 5664033.77 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3314, pruned_loss=0.08681, over 5765979.41 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3231, pruned_loss=0.0821, over 5649318.69 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:51:13,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1246451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:51:22,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.531e+03 2.099e+03 2.871e+03 1.182e+04, threshold=4.199e+03, percent-clipped=21.0
+2023-03-14 10:51:48,442 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1246480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:52:10,021 INFO [train.py:968] (0/2) Epoch 28, batch 15600, giga_loss[loss=0.2744, simple_loss=0.351, pruned_loss=0.09896, over 27604.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3243, pruned_loss=0.08085, over 5675322.04 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3312, pruned_loss=0.08672, over 5768421.21 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3242, pruned_loss=0.08076, over 5659862.99 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:52:48,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1246533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:53:08,077 INFO [train.py:968] (0/2) Epoch 28, batch 15650, giga_loss[loss=0.2876, simple_loss=0.3514, pruned_loss=0.1119, over 26880.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3279, pruned_loss=0.0827, over 5676172.03 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3305, pruned_loss=0.0863, over 5773210.97 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3283, pruned_loss=0.08285, over 5656450.57 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:53:19,649 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.411e+02 1.419e+03 2.113e+03 2.815e+03 7.200e+03, threshold=4.225e+03, percent-clipped=5.0
+2023-03-14 10:53:20,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3862, 2.0182, 1.4479, 0.6927], device='cuda:0'), covar=tensor([0.6098, 0.3594, 0.5096, 0.6862], device='cuda:0'), in_proj_covar=tensor([0.1840, 0.1731, 0.1657, 0.1505], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 10:53:59,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2260, 4.0811, 3.8917, 2.0617], device='cuda:0'), covar=tensor([0.0593, 0.0690, 0.0826, 0.2420], device='cuda:0'), in_proj_covar=tensor([0.1283, 0.1182, 0.0997, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 10:54:08,069 INFO [train.py:968] (0/2) Epoch 28, batch 15700, giga_loss[loss=0.2432, simple_loss=0.3308, pruned_loss=0.07786, over 28660.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3304, pruned_loss=0.08398, over 5671454.73 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3305, pruned_loss=0.08638, over 5771652.83 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3307, pruned_loss=0.08397, over 5655456.96 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:54:50,315 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 10:55:06,427 INFO [train.py:968] (0/2) Epoch 28, batch 15750, giga_loss[loss=0.3121, simple_loss=0.384, pruned_loss=0.1202, over 28980.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3306, pruned_loss=0.0841, over 5675556.79 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3303, pruned_loss=0.08628, over 5765452.23 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3311, pruned_loss=0.08413, over 5666288.17 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:55:12,973 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1246654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:55:14,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-14 10:55:21,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.253e+02 1.712e+03 2.040e+03 2.883e+03 6.089e+03, threshold=4.080e+03, percent-clipped=7.0
+2023-03-14 10:56:01,358 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:56:06,538 INFO [train.py:968] (0/2) Epoch 28, batch 15800, giga_loss[loss=0.2234, simple_loss=0.3035, pruned_loss=0.07166, over 28942.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3287, pruned_loss=0.08293, over 5684504.27 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.33, pruned_loss=0.08617, over 5767796.70 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3292, pruned_loss=0.08301, over 5673904.92 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:57:08,255 INFO [train.py:968] (0/2) Epoch 28, batch 15850, giga_loss[loss=0.2177, simple_loss=0.3091, pruned_loss=0.06313, over 28724.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3266, pruned_loss=0.08144, over 5672654.80 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3298, pruned_loss=0.08598, over 5752872.45 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3271, pruned_loss=0.08156, over 5675382.53 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:57:20,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.684e+02 1.411e+03 1.779e+03 2.693e+03 8.438e+03, threshold=3.558e+03, percent-clipped=9.0
+2023-03-14 10:57:43,541 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-14 10:58:02,792 INFO [train.py:968] (0/2) Epoch 28, batch 15900, giga_loss[loss=0.2281, simple_loss=0.3167, pruned_loss=0.06978, over 28877.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3249, pruned_loss=0.08167, over 5668678.32 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3296, pruned_loss=0.08597, over 5748091.98 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3255, pruned_loss=0.08164, over 5673141.56 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:58:47,261 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:58:51,783 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1246839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:59:01,996 INFO [train.py:968] (0/2) Epoch 28, batch 15950, giga_loss[loss=0.2662, simple_loss=0.3541, pruned_loss=0.08917, over 28901.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3252, pruned_loss=0.08181, over 5667807.70 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3292, pruned_loss=0.08574, over 5748190.17 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3258, pruned_loss=0.08189, over 5669487.85 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:59:08,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8505, 1.2038, 1.4057, 0.9178], device='cuda:0'), covar=tensor([0.2302, 0.1529, 0.2316, 0.2062], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0743, 0.0716, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 10:59:15,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.842e+02 1.485e+03 2.102e+03 2.715e+03 6.109e+03, threshold=4.203e+03, percent-clipped=14.0
+2023-03-14 10:59:27,959 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1246868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:59:27,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1246868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:59:33,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1246872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:59:50,041 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8449, 2.1419, 1.4047, 1.5651], device='cuda:0'), covar=tensor([0.0992, 0.0505, 0.1019, 0.1109], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0445, 0.0520, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 11:00:06,103 INFO [train.py:968] (0/2) Epoch 28, batch 16000, giga_loss[loss=0.2398, simple_loss=0.3197, pruned_loss=0.07991, over 28889.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3285, pruned_loss=0.08357, over 5654348.41 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3296, pruned_loss=0.08617, over 5731310.78 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3286, pruned_loss=0.08319, over 5669502.14 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:00:12,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4561, 1.2637, 4.1764, 3.3172], device='cuda:0'), covar=tensor([0.1638, 0.2953, 0.0456, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0671, 0.0996, 0.0969], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 11:00:12,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5153, 1.7723, 1.5827, 1.6762], device='cuda:0'), covar=tensor([0.0786, 0.0310, 0.0327, 0.0896], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0121, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 11:00:18,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:00:31,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8599, 1.2853, 1.3680, 1.0432], device='cuda:0'), covar=tensor([0.2193, 0.1255, 0.2270, 0.1718], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0742, 0.0715, 0.0685], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 11:01:08,647 INFO [train.py:968] (0/2) Epoch 28, batch 16050, giga_loss[loss=0.2598, simple_loss=0.3512, pruned_loss=0.08424, over 28965.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3279, pruned_loss=0.08357, over 5661166.21 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3285, pruned_loss=0.08561, over 5735804.53 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.329, pruned_loss=0.0837, over 5666954.87 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:01:18,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.314e+02 1.451e+03 2.049e+03 2.634e+03 5.848e+03, threshold=4.098e+03, percent-clipped=5.0
+2023-03-14 11:02:07,885 INFO [train.py:968] (0/2) Epoch 28, batch 16100, giga_loss[loss=0.2618, simple_loss=0.3495, pruned_loss=0.08707, over 28688.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3311, pruned_loss=0.08522, over 5668742.71 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3286, pruned_loss=0.08572, over 5737436.85 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3318, pruned_loss=0.08519, over 5670857.27 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:02:39,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1247029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:03:01,749 INFO [train.py:968] (0/2) Epoch 28, batch 16150, giga_loss[loss=0.2493, simple_loss=0.3377, pruned_loss=0.08045, over 29027.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3335, pruned_loss=0.08589, over 5668622.08 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3284, pruned_loss=0.08573, over 5729683.90 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3344, pruned_loss=0.08586, over 5674549.18 frames. ], batch size: 285, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:03:03,516 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:03:07,120 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1247054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:03:10,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.812e+02 1.378e+03 2.025e+03 2.622e+03 8.043e+03, threshold=4.050e+03, percent-clipped=5.0
+2023-03-14 11:03:40,859 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:04:03,410 INFO [train.py:968] (0/2) Epoch 28, batch 16200, giga_loss[loss=0.2482, simple_loss=0.3279, pruned_loss=0.08424, over 28975.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3347, pruned_loss=0.08652, over 5674419.35 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3277, pruned_loss=0.08528, over 5732915.05 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3363, pruned_loss=0.08692, over 5674767.76 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:04:19,908 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1247113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:05:10,273 INFO [train.py:968] (0/2) Epoch 28, batch 16250, giga_loss[loss=0.2418, simple_loss=0.3292, pruned_loss=0.07715, over 28641.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3339, pruned_loss=0.08621, over 5681415.48 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3274, pruned_loss=0.08497, over 5735062.95 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3356, pruned_loss=0.08683, over 5678161.74 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:05:23,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.838e+02 1.584e+03 1.989e+03 2.709e+03 5.615e+03, threshold=3.977e+03, percent-clipped=5.0
+2023-03-14 11:05:23,369 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1247159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:05:40,173 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:05:42,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1247175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:05:48,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2139, 1.8408, 1.4085, 0.4075], device='cuda:0'), covar=tensor([0.5588, 0.3176, 0.5124, 0.7476], device='cuda:0'), in_proj_covar=tensor([0.1835, 0.1726, 0.1648, 0.1501], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 11:06:10,362 INFO [train.py:968] (0/2) Epoch 28, batch 16300, giga_loss[loss=0.2371, simple_loss=0.322, pruned_loss=0.07609, over 28990.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.332, pruned_loss=0.08572, over 5693697.05 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3268, pruned_loss=0.08467, over 5739902.38 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3341, pruned_loss=0.08652, over 5685442.25 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:06:18,676 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:07:07,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1247243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:07:13,116 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1247247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:07:14,050 INFO [train.py:968] (0/2) Epoch 28, batch 16350, giga_loss[loss=0.2523, simple_loss=0.3343, pruned_loss=0.08522, over 29051.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3316, pruned_loss=0.08559, over 5662835.22 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.327, pruned_loss=0.08488, over 5730111.64 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3331, pruned_loss=0.08604, over 5664435.74 frames. ], batch size: 285, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:07:25,778 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.428e+03 1.711e+03 2.349e+03 6.866e+03, threshold=3.423e+03, percent-clipped=5.0
+2023-03-14 11:08:19,746 INFO [train.py:968] (0/2) Epoch 28, batch 16400, giga_loss[loss=0.2521, simple_loss=0.3267, pruned_loss=0.08873, over 28936.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3307, pruned_loss=0.08599, over 5669797.43 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3272, pruned_loss=0.08499, over 5731392.27 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3318, pruned_loss=0.08624, over 5669122.95 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 11:09:16,154 INFO [train.py:968] (0/2) Epoch 28, batch 16450, giga_loss[loss=0.3136, simple_loss=0.3788, pruned_loss=0.1243, over 28347.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3295, pruned_loss=0.08618, over 5668898.28 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3268, pruned_loss=0.08487, over 5732920.42 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.331, pruned_loss=0.08656, over 5663673.90 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:09:28,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.529e+02 1.471e+03 1.936e+03 2.609e+03 9.810e+03, threshold=3.872e+03, percent-clipped=11.0
+2023-03-14 11:09:57,755 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:01,853 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1247389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:02,567 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:02,712 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-14 11:10:05,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1247393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:09,552 INFO [train.py:968] (0/2) Epoch 28, batch 16500, giga_loss[loss=0.2417, simple_loss=0.3362, pruned_loss=0.07363, over 28848.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3293, pruned_loss=0.08511, over 5673256.37 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3267, pruned_loss=0.08493, over 5728038.75 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3306, pruned_loss=0.08538, over 5670646.33 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:10:33,206 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:36,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:11:11,286 INFO [train.py:968] (0/2) Epoch 28, batch 16550, giga_loss[loss=0.2576, simple_loss=0.3334, pruned_loss=0.09088, over 26812.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3276, pruned_loss=0.0832, over 5669155.89 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3265, pruned_loss=0.08487, over 5727855.81 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3289, pruned_loss=0.08345, over 5666043.38 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:11:23,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.922e+02 1.445e+03 1.848e+03 2.389e+03 7.332e+03, threshold=3.696e+03, percent-clipped=8.0
+2023-03-14 11:11:55,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1247488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:12:06,879 INFO [train.py:968] (0/2) Epoch 28, batch 16600, giga_loss[loss=0.2353, simple_loss=0.3336, pruned_loss=0.0685, over 28393.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3299, pruned_loss=0.08289, over 5665658.22 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3267, pruned_loss=0.0851, over 5721953.19 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3309, pruned_loss=0.08282, over 5667372.64 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:12:51,989 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1247534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:12:52,785 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-14 11:13:05,991 INFO [train.py:968] (0/2) Epoch 28, batch 16650, giga_loss[loss=0.2335, simple_loss=0.3272, pruned_loss=0.06993, over 28847.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3315, pruned_loss=0.08261, over 5670760.15 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3267, pruned_loss=0.08512, over 5722871.42 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3322, pruned_loss=0.08253, over 5670965.29 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:13:18,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.971e+02 1.368e+03 1.839e+03 2.561e+03 5.290e+03, threshold=3.679e+03, percent-clipped=3.0
+2023-03-14 11:13:58,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3203, 1.5152, 1.4567, 1.3021], device='cuda:0'), covar=tensor([0.2774, 0.2113, 0.2402, 0.2229], device='cuda:0'), in_proj_covar=tensor([0.2026, 0.1982, 0.1881, 0.2028], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 11:14:12,632 INFO [train.py:968] (0/2) Epoch 28, batch 16700, giga_loss[loss=0.2609, simple_loss=0.3515, pruned_loss=0.08519, over 28907.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3323, pruned_loss=0.08322, over 5674699.58 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3266, pruned_loss=0.08505, over 5720980.96 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3331, pruned_loss=0.08319, over 5676147.43 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:14:57,601 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:15:00,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1247634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:15:17,497 INFO [train.py:968] (0/2) Epoch 28, batch 16750, libri_loss[loss=0.2644, simple_loss=0.3401, pruned_loss=0.09436, over 29660.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3319, pruned_loss=0.0831, over 5672536.76 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3263, pruned_loss=0.085, over 5724407.39 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3329, pruned_loss=0.08308, over 5668810.91 frames. ], batch size: 88, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:15:30,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.486e+03 2.109e+03 3.274e+03 1.248e+04, threshold=4.218e+03, percent-clipped=18.0
+2023-03-14 11:15:34,455 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:15:57,601 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:16:01,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1247680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:16:28,199 INFO [train.py:968] (0/2) Epoch 28, batch 16800, giga_loss[loss=0.2686, simple_loss=0.338, pruned_loss=0.09962, over 27019.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3326, pruned_loss=0.0835, over 5668292.79 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3266, pruned_loss=0.08497, over 5725991.92 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3334, pruned_loss=0.08349, over 5662927.90 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:16:38,818 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:16:38,866 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6217, 1.9285, 1.8334, 1.5454], device='cuda:0'), covar=tensor([0.2237, 0.2663, 0.2171, 0.2612], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0743, 0.0715, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 11:17:38,165 INFO [train.py:968] (0/2) Epoch 28, batch 16850, giga_loss[loss=0.266, simple_loss=0.3293, pruned_loss=0.1013, over 26853.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3328, pruned_loss=0.08271, over 5675266.94 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3263, pruned_loss=0.08491, over 5726156.71 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3337, pruned_loss=0.08272, over 5670243.43 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:17:54,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.133e+02 1.450e+03 1.660e+03 2.325e+03 6.064e+03, threshold=3.320e+03, percent-clipped=5.0
+2023-03-14 11:18:34,381 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1972, 1.3248, 1.2209, 1.1955], device='cuda:0'), covar=tensor([0.1801, 0.1808, 0.1433, 0.1536], device='cuda:0'), in_proj_covar=tensor([0.2023, 0.1977, 0.1873, 0.2024], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 11:18:46,172 INFO [train.py:968] (0/2) Epoch 28, batch 16900, giga_loss[loss=0.2741, simple_loss=0.3617, pruned_loss=0.09327, over 28698.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3362, pruned_loss=0.08474, over 5678652.15 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3266, pruned_loss=0.08513, over 5731573.53 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.337, pruned_loss=0.08449, over 5668307.60 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:19:54,438 INFO [train.py:968] (0/2) Epoch 28, batch 16950, giga_loss[loss=0.2423, simple_loss=0.315, pruned_loss=0.08478, over 24434.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.338, pruned_loss=0.08519, over 5679101.99 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3264, pruned_loss=0.08511, over 5724606.98 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3388, pruned_loss=0.08501, over 5676155.71 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:20:08,646 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.514e+03 2.130e+03 2.900e+03 7.052e+03, threshold=4.260e+03, percent-clipped=17.0
+2023-03-14 11:20:33,383 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-14 11:21:01,377 INFO [train.py:968] (0/2) Epoch 28, batch 17000, giga_loss[loss=0.2689, simple_loss=0.3469, pruned_loss=0.0955, over 29000.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3368, pruned_loss=0.08521, over 5686338.47 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3267, pruned_loss=0.08513, over 5727878.14 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3375, pruned_loss=0.08505, over 5679875.90 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:21:31,371 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1247921.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 11:21:36,206 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3233, 1.9516, 1.3566, 0.6301], device='cuda:0'), covar=tensor([0.4662, 0.2454, 0.3903, 0.5357], device='cuda:0'), in_proj_covar=tensor([0.1836, 0.1729, 0.1650, 0.1503], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 11:22:11,380 INFO [train.py:968] (0/2) Epoch 28, batch 17050, giga_loss[loss=0.2152, simple_loss=0.3022, pruned_loss=0.0641, over 29119.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3356, pruned_loss=0.08545, over 5690556.72 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3271, pruned_loss=0.08537, over 5731819.86 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.336, pruned_loss=0.08511, over 5680977.24 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:22:24,411 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.235e+02 1.692e+03 2.365e+03 3.001e+03 5.600e+03, threshold=4.729e+03, percent-clipped=11.0
+2023-03-14 11:23:25,008 INFO [train.py:968] (0/2) Epoch 28, batch 17100, giga_loss[loss=0.2271, simple_loss=0.3182, pruned_loss=0.06795, over 28976.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3337, pruned_loss=0.08325, over 5700626.60 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3269, pruned_loss=0.08526, over 5735136.15 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3344, pruned_loss=0.08304, over 5689455.35 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:23:25,679 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1248000.pt
+2023-03-14 11:24:22,725 INFO [train.py:968] (0/2) Epoch 28, batch 17150, giga_loss[loss=0.2887, simple_loss=0.3706, pruned_loss=0.1034, over 28113.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3326, pruned_loss=0.08285, over 5692159.26 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.327, pruned_loss=0.0853, over 5727694.10 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3332, pruned_loss=0.08259, over 5689466.89 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:24:36,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.777e+02 1.356e+03 1.591e+03 2.220e+03 8.633e+03, threshold=3.182e+03, percent-clipped=5.0
+2023-03-14 11:24:48,104 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5939, 1.8992, 1.5683, 1.7037], device='cuda:0'), covar=tensor([0.2704, 0.2576, 0.2962, 0.2455], device='cuda:0'), in_proj_covar=tensor([0.1598, 0.1149, 0.1414, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 11:24:48,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8166, 2.8081, 1.7107, 1.0371], device='cuda:0'), covar=tensor([0.8369, 0.3331, 0.4362, 0.6182], device='cuda:0'), in_proj_covar=tensor([0.1832, 0.1722, 0.1646, 0.1499], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 11:25:13,997 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1248094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:25:19,850 INFO [train.py:968] (0/2) Epoch 28, batch 17200, giga_loss[loss=0.2137, simple_loss=0.3083, pruned_loss=0.05953, over 28570.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3333, pruned_loss=0.08295, over 5688574.19 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3271, pruned_loss=0.08526, over 5731040.14 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3339, pruned_loss=0.0827, over 5682342.90 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:26:18,174 INFO [train.py:968] (0/2) Epoch 28, batch 17250, giga_loss[loss=0.2739, simple_loss=0.3562, pruned_loss=0.0958, over 28436.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3364, pruned_loss=0.08537, over 5684916.70 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.327, pruned_loss=0.08515, over 5734181.23 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3371, pruned_loss=0.08525, over 5676393.56 frames. ], batch size: 369, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:26:32,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.090e+03 1.570e+03 1.972e+03 2.490e+03 6.020e+03, threshold=3.944e+03, percent-clipped=11.0
+2023-03-14 11:27:15,595 INFO [train.py:968] (0/2) Epoch 28, batch 17300, giga_loss[loss=0.2588, simple_loss=0.3198, pruned_loss=0.09889, over 26779.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3334, pruned_loss=0.08444, over 5684622.15 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3272, pruned_loss=0.0852, over 5736386.45 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3339, pruned_loss=0.0843, over 5675055.52 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:28:13,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 11:28:13,969 INFO [train.py:968] (0/2) Epoch 28, batch 17350, giga_loss[loss=0.2201, simple_loss=0.3108, pruned_loss=0.06465, over 29038.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3323, pruned_loss=0.08445, over 5687609.61 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.327, pruned_loss=0.08504, over 5740871.64 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3331, pruned_loss=0.08448, over 5674560.82 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:28:26,230 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 11:28:27,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.528e+02 1.586e+03 2.193e+03 2.876e+03 6.413e+03, threshold=4.386e+03, percent-clipped=9.0
+2023-03-14 11:29:08,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1248296.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 11:29:11,796 INFO [train.py:968] (0/2) Epoch 28, batch 17400, giga_loss[loss=0.2892, simple_loss=0.3697, pruned_loss=0.1044, over 28887.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3336, pruned_loss=0.08568, over 5694812.48 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3268, pruned_loss=0.08493, over 5743031.75 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3345, pruned_loss=0.0858, over 5681286.38 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:29:20,297 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9565, 2.1324, 2.1510, 1.7268], device='cuda:0'), covar=tensor([0.1543, 0.2008, 0.1292, 0.1611], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0706, 0.0979, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 11:29:25,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5832, 4.4287, 4.2116, 2.0146], device='cuda:0'), covar=tensor([0.0516, 0.0630, 0.0739, 0.2287], device='cuda:0'), in_proj_covar=tensor([0.1275, 0.1170, 0.0988, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 11:29:36,614 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1938, 2.2929, 2.3823, 1.8773], device='cuda:0'), covar=tensor([0.1697, 0.2226, 0.1387, 0.1795], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0706, 0.0979, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 11:30:03,874 INFO [train.py:968] (0/2) Epoch 28, batch 17450, giga_loss[loss=0.3129, simple_loss=0.3902, pruned_loss=0.1178, over 28659.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3418, pruned_loss=0.09081, over 5669698.64 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3269, pruned_loss=0.08509, over 5720157.16 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3427, pruned_loss=0.09085, over 5676417.01 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:30:15,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.098e+03 1.565e+03 1.953e+03 2.575e+03 1.026e+04, threshold=3.906e+03, percent-clipped=8.0
+2023-03-14 11:30:46,915 INFO [train.py:968] (0/2) Epoch 28, batch 17500, libri_loss[loss=0.2481, simple_loss=0.3337, pruned_loss=0.08122, over 29536.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3478, pruned_loss=0.09366, over 5682038.72 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3265, pruned_loss=0.08466, over 5724921.91 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3496, pruned_loss=0.09441, over 5681272.57 frames. ], batch size: 84, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:31:07,579 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1248426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:31:21,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1248439.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 11:31:24,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1248442.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 11:31:29,215 INFO [train.py:968] (0/2) Epoch 28, batch 17550, giga_loss[loss=0.2785, simple_loss=0.352, pruned_loss=0.1025, over 28914.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3484, pruned_loss=0.09441, over 5682178.59 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3268, pruned_loss=0.08474, over 5720423.74 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3502, pruned_loss=0.09518, over 5684606.21 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:31:42,604 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.892e+02 1.371e+03 1.724e+03 2.306e+03 4.272e+03, threshold=3.449e+03, percent-clipped=1.0
+2023-03-14 11:31:47,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6080, 1.5950, 1.7924, 1.4373], device='cuda:0'), covar=tensor([0.1844, 0.2587, 0.1533, 0.1807], device='cuda:0'), in_proj_covar=tensor([0.0929, 0.0707, 0.0980, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 11:31:48,548 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1248469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:31:50,067 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1248471.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 11:31:56,576 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1248476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:32:16,192 INFO [train.py:968] (0/2) Epoch 28, batch 17600, libri_loss[loss=0.2158, simple_loss=0.2933, pruned_loss=0.06911, over 29519.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3424, pruned_loss=0.09231, over 5680522.74 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3266, pruned_loss=0.0846, over 5720019.10 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3443, pruned_loss=0.09318, over 5682267.02 frames. ], batch size: 70, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:32:42,540 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1248530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:32:58,325 INFO [train.py:968] (0/2) Epoch 28, batch 17650, giga_loss[loss=0.2173, simple_loss=0.3003, pruned_loss=0.0672, over 28724.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3356, pruned_loss=0.08959, over 5679974.69 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3267, pruned_loss=0.0845, over 5724430.87 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3374, pruned_loss=0.09058, over 5676128.16 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:33:11,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.105e+02 1.204e+03 1.527e+03 1.943e+03 5.412e+03, threshold=3.053e+03, percent-clipped=2.0
+2023-03-14 11:33:45,409 INFO [train.py:968] (0/2) Epoch 28, batch 17700, giga_loss[loss=0.2063, simple_loss=0.2844, pruned_loss=0.0641, over 29002.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3263, pruned_loss=0.08542, over 5679870.70 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3265, pruned_loss=0.08435, over 5726350.74 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3279, pruned_loss=0.08636, over 5674711.47 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:33:55,013 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1248612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:33:58,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1248615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:34:25,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1248644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:34:28,022 INFO [train.py:968] (0/2) Epoch 28, batch 17750, giga_loss[loss=0.1996, simple_loss=0.2766, pruned_loss=0.06127, over 28710.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.319, pruned_loss=0.08204, over 5689149.81 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3265, pruned_loss=0.08422, over 5729753.82 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3202, pruned_loss=0.0829, over 5680909.56 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:34:40,271 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.710e+02 1.151e+03 1.497e+03 1.824e+03 4.216e+03, threshold=2.993e+03, percent-clipped=6.0
+2023-03-14 11:35:10,648 INFO [train.py:968] (0/2) Epoch 28, batch 17800, giga_loss[loss=0.1983, simple_loss=0.2759, pruned_loss=0.06032, over 28919.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3129, pruned_loss=0.07899, over 5695498.03 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3268, pruned_loss=0.08427, over 5731679.72 frames. ], giga_tot_loss[loss=0.2363, simple_loss=0.3134, pruned_loss=0.07957, over 5686794.94 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:35:43,166 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 11:35:50,566 INFO [train.py:968] (0/2) Epoch 28, batch 17850, giga_loss[loss=0.2137, simple_loss=0.3023, pruned_loss=0.06251, over 28862.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3103, pruned_loss=0.07794, over 5701566.13 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3271, pruned_loss=0.08437, over 5736221.49 frames. ], giga_tot_loss[loss=0.2332, simple_loss=0.3101, pruned_loss=0.07815, over 5689588.46 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:36:00,746 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.423e+02 1.164e+03 1.534e+03 2.128e+03 5.924e+03, threshold=3.068e+03, percent-clipped=15.0
+2023-03-14 11:36:33,759 INFO [train.py:968] (0/2) Epoch 28, batch 17900, giga_loss[loss=0.2457, simple_loss=0.3161, pruned_loss=0.08765, over 28669.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.308, pruned_loss=0.07736, over 5696219.75 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3275, pruned_loss=0.08459, over 5730424.00 frames. ], giga_tot_loss[loss=0.2307, simple_loss=0.307, pruned_loss=0.0772, over 5691232.84 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:36:35,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1248801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:37:02,622 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5084, 3.2970, 1.6462, 1.5465], device='cuda:0'), covar=tensor([0.0989, 0.0405, 0.0928, 0.1364], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0564, 0.0408, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 11:37:15,636 INFO [train.py:968] (0/2) Epoch 28, batch 17950, libri_loss[loss=0.2578, simple_loss=0.3456, pruned_loss=0.08503, over 29503.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3061, pruned_loss=0.07649, over 5703168.61 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3282, pruned_loss=0.08478, over 5736880.63 frames. ], giga_tot_loss[loss=0.228, simple_loss=0.304, pruned_loss=0.07594, over 5692110.25 frames. ], batch size: 81, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:37:18,730 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1248851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:37:18,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1248851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:37:26,636 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.349e+02 1.216e+03 1.631e+03 2.087e+03 4.317e+03, threshold=3.261e+03, percent-clipped=6.0
+2023-03-14 11:37:57,386 INFO [train.py:968] (0/2) Epoch 28, batch 18000, giga_loss[loss=0.2201, simple_loss=0.2963, pruned_loss=0.072, over 27885.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3031, pruned_loss=0.0752, over 5700382.06 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3285, pruned_loss=0.08487, over 5739440.63 frames. ], giga_tot_loss[loss=0.2248, simple_loss=0.3007, pruned_loss=0.07449, over 5688747.93 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:37:57,391 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 11:38:05,501 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2019, simple_loss=0.3081, pruned_loss=0.04781, over 944034.00 frames. 
+2023-03-14 11:38:05,502 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 11:38:11,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1248905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:38:34,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2110, 5.0455, 4.7717, 2.3524], device='cuda:0'), covar=tensor([0.0391, 0.0553, 0.0643, 0.1867], device='cuda:0'), in_proj_covar=tensor([0.1286, 0.1182, 0.0998, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 11:38:41,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5808, 1.9127, 1.5481, 1.4911], device='cuda:0'), covar=tensor([0.2719, 0.2605, 0.3076, 0.2584], device='cuda:0'), in_proj_covar=tensor([0.1599, 0.1151, 0.1413, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 11:38:44,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1248944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:38:47,001 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1248947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:38:47,933 INFO [train.py:968] (0/2) Epoch 28, batch 18050, giga_loss[loss=0.1926, simple_loss=0.2758, pruned_loss=0.05469, over 29028.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3006, pruned_loss=0.07419, over 5674900.34 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3288, pruned_loss=0.08506, over 5713618.34 frames. ], giga_tot_loss[loss=0.2223, simple_loss=0.2981, pruned_loss=0.07331, over 5688568.66 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:39:02,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.209e+02 1.128e+03 1.373e+03 1.815e+03 5.531e+03, threshold=2.745e+03, percent-clipped=5.0
+2023-03-14 11:39:13,455 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1248976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:39:24,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4587, 2.1430, 1.7554, 0.7863], device='cuda:0'), covar=tensor([0.6599, 0.3811, 0.4734, 0.7395], device='cuda:0'), in_proj_covar=tensor([0.1841, 0.1739, 0.1652, 0.1501], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 11:39:30,376 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1248994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:39:33,199 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1248997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:39:34,086 INFO [train.py:968] (0/2) Epoch 28, batch 18100, giga_loss[loss=0.2043, simple_loss=0.2865, pruned_loss=0.06107, over 28987.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.297, pruned_loss=0.07256, over 5676389.05 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3288, pruned_loss=0.085, over 5716324.40 frames. ], giga_tot_loss[loss=0.2191, simple_loss=0.2946, pruned_loss=0.07177, over 5684193.03 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:39:56,079 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1249026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:40:16,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1249048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:40:16,900 INFO [train.py:968] (0/2) Epoch 28, batch 18150, giga_loss[loss=0.2124, simple_loss=0.2872, pruned_loss=0.06877, over 28699.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2949, pruned_loss=0.07143, over 5683465.19 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.329, pruned_loss=0.08488, over 5721833.79 frames. ], giga_tot_loss[loss=0.2164, simple_loss=0.2917, pruned_loss=0.07049, over 5683794.83 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:40:19,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1249051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:40:30,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.053e+02 1.059e+03 1.257e+03 1.548e+03 3.570e+03, threshold=2.514e+03, percent-clipped=4.0
+2023-03-14 11:40:46,551 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1249080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:41:00,035 INFO [train.py:968] (0/2) Epoch 28, batch 18200, libri_loss[loss=0.3208, simple_loss=0.3816, pruned_loss=0.13, over 19824.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2932, pruned_loss=0.07053, over 5668859.55 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3293, pruned_loss=0.08485, over 5711700.61 frames. ], giga_tot_loss[loss=0.2134, simple_loss=0.2886, pruned_loss=0.06914, over 5676796.73 frames. ], batch size: 187, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:41:47,368 INFO [train.py:968] (0/2) Epoch 28, batch 18250, libri_loss[loss=0.2216, simple_loss=0.2956, pruned_loss=0.07374, over 29386.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2958, pruned_loss=0.07288, over 5659172.58 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3295, pruned_loss=0.08495, over 5703062.98 frames. ], giga_tot_loss[loss=0.2172, simple_loss=0.2915, pruned_loss=0.07145, over 5672639.46 frames. ], batch size: 67, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:42:01,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.524e+02 1.218e+03 1.517e+03 2.050e+03 4.524e+03, threshold=3.033e+03, percent-clipped=10.0
+2023-03-14 11:42:09,961 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 11:42:18,841 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-14 11:42:36,074 INFO [train.py:968] (0/2) Epoch 28, batch 18300, giga_loss[loss=0.2679, simple_loss=0.3509, pruned_loss=0.09251, over 29054.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3074, pruned_loss=0.07862, over 5650415.68 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3294, pruned_loss=0.08481, over 5688884.10 frames. ], giga_tot_loss[loss=0.229, simple_loss=0.3032, pruned_loss=0.07734, over 5672931.49 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:42:40,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3214, 3.1693, 2.9855, 1.4128], device='cuda:0'), covar=tensor([0.0970, 0.1097, 0.0953, 0.2299], device='cuda:0'), in_proj_covar=tensor([0.1286, 0.1183, 0.0998, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 11:43:00,373 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1249226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:43:20,582 INFO [train.py:968] (0/2) Epoch 28, batch 18350, giga_loss[loss=0.2891, simple_loss=0.3671, pruned_loss=0.1056, over 28899.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3197, pruned_loss=0.08496, over 5664142.39 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3291, pruned_loss=0.08464, over 5691319.94 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3164, pruned_loss=0.08403, over 5679400.12 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:43:32,644 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.412e+02 1.591e+03 2.012e+03 2.489e+03 7.247e+03, threshold=4.024e+03, percent-clipped=11.0
+2023-03-14 11:43:48,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2025, 4.0404, 3.7962, 2.1084], device='cuda:0'), covar=tensor([0.0577, 0.0702, 0.0753, 0.2311], device='cuda:0'), in_proj_covar=tensor([0.1283, 0.1183, 0.0998, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 11:44:02,098 INFO [train.py:968] (0/2) Epoch 28, batch 18400, giga_loss[loss=0.2673, simple_loss=0.3565, pruned_loss=0.089, over 28869.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3293, pruned_loss=0.08944, over 5665079.72 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3293, pruned_loss=0.0847, over 5684315.42 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3264, pruned_loss=0.08869, over 5682164.69 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:44:23,325 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5257, 1.7817, 1.2500, 1.3173], device='cuda:0'), covar=tensor([0.1109, 0.0596, 0.1128, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0410, 0.0446, 0.0520, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 11:44:45,835 INFO [train.py:968] (0/2) Epoch 28, batch 18450, giga_loss[loss=0.2673, simple_loss=0.3431, pruned_loss=0.09569, over 28623.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3341, pruned_loss=0.09054, over 5667009.23 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3293, pruned_loss=0.0846, over 5688377.49 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3318, pruned_loss=0.09014, over 5676474.01 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:44:58,751 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.677e+02 1.361e+03 1.767e+03 2.479e+03 5.877e+03, threshold=3.535e+03, percent-clipped=3.0
+2023-03-14 11:45:02,630 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1249369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:45:05,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1249372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:45:27,783 INFO [train.py:968] (0/2) Epoch 28, batch 18500, giga_loss[loss=0.2725, simple_loss=0.3552, pruned_loss=0.09489, over 27706.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.336, pruned_loss=0.09033, over 5671469.19 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3295, pruned_loss=0.08454, over 5693661.03 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3341, pruned_loss=0.09023, over 5673728.46 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:45:31,661 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1249401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:45:36,775 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1249407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:45:46,061 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 11:46:16,037 INFO [train.py:968] (0/2) Epoch 28, batch 18550, giga_loss[loss=0.2754, simple_loss=0.3497, pruned_loss=0.1005, over 28717.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3375, pruned_loss=0.09096, over 5669894.63 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3292, pruned_loss=0.0844, over 5695982.79 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3364, pruned_loss=0.09108, over 5669315.74 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:46:29,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.017e+02 1.217e+03 1.520e+03 2.030e+03 4.828e+03, threshold=3.040e+03, percent-clipped=4.0
+2023-03-14 11:46:57,673 INFO [train.py:968] (0/2) Epoch 28, batch 18600, giga_loss[loss=0.2671, simple_loss=0.3478, pruned_loss=0.0932, over 28959.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3406, pruned_loss=0.09339, over 5673109.49 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3298, pruned_loss=0.08468, over 5698480.62 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3395, pruned_loss=0.09345, over 5669897.08 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:47:39,292 INFO [train.py:968] (0/2) Epoch 28, batch 18650, giga_loss[loss=0.2865, simple_loss=0.3578, pruned_loss=0.1076, over 28903.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3436, pruned_loss=0.09526, over 5679273.39 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3308, pruned_loss=0.08508, over 5703587.34 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3424, pruned_loss=0.09546, over 5670735.36 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:47:47,196 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-14 11:47:53,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.147e+02 1.335e+03 1.598e+03 2.331e+03 7.365e+03, threshold=3.195e+03, percent-clipped=7.0
+2023-03-14 11:48:20,854 INFO [train.py:968] (0/2) Epoch 28, batch 18700, giga_loss[loss=0.3028, simple_loss=0.3781, pruned_loss=0.1137, over 28764.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3461, pruned_loss=0.09647, over 5676603.42 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3307, pruned_loss=0.08496, over 5700322.10 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3456, pruned_loss=0.09704, over 5672502.70 frames. ], batch size: 66, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:48:55,013 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-14 11:49:01,024 INFO [train.py:968] (0/2) Epoch 28, batch 18750, giga_loss[loss=0.2607, simple_loss=0.3464, pruned_loss=0.08745, over 28776.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3497, pruned_loss=0.09752, over 5685117.84 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3311, pruned_loss=0.08501, over 5702713.04 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3493, pruned_loss=0.09815, over 5679229.90 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:49:03,581 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1249651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:49:15,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.025e+03 1.316e+03 1.581e+03 2.258e+03 6.432e+03, threshold=3.161e+03, percent-clipped=11.0
+2023-03-14 11:49:44,460 INFO [train.py:968] (0/2) Epoch 28, batch 18800, giga_loss[loss=0.2683, simple_loss=0.3559, pruned_loss=0.09042, over 28605.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3513, pruned_loss=0.09769, over 5678868.14 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3312, pruned_loss=0.08494, over 5697781.80 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3514, pruned_loss=0.0986, over 5677844.23 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:50:27,193 INFO [train.py:968] (0/2) Epoch 28, batch 18850, giga_loss[loss=0.2472, simple_loss=0.3411, pruned_loss=0.07663, over 28686.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3519, pruned_loss=0.09721, over 5684743.91 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3312, pruned_loss=0.08493, over 5701598.83 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3523, pruned_loss=0.09819, over 5680146.26 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:50:30,957 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1249753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:50:42,653 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.374e+02 1.230e+03 1.498e+03 2.037e+03 3.495e+03, threshold=2.995e+03, percent-clipped=3.0
+2023-03-14 11:50:54,956 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1249782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:51:09,390 INFO [train.py:968] (0/2) Epoch 28, batch 18900, giga_loss[loss=0.2326, simple_loss=0.3229, pruned_loss=0.0712, over 29062.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3512, pruned_loss=0.09534, over 5693970.23 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3315, pruned_loss=0.08501, over 5696577.43 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3518, pruned_loss=0.09628, over 5694562.14 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:51:44,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8533, 1.0689, 0.9634, 0.7640], device='cuda:0'), covar=tensor([0.2821, 0.3197, 0.2034, 0.2784], device='cuda:0'), in_proj_covar=tensor([0.2041, 0.1999, 0.1894, 0.2047], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 11:51:48,832 INFO [train.py:968] (0/2) Epoch 28, batch 18950, libri_loss[loss=0.2602, simple_loss=0.3469, pruned_loss=0.08677, over 25890.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3499, pruned_loss=0.0938, over 5698213.50 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3316, pruned_loss=0.08498, over 5697032.09 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3506, pruned_loss=0.09478, over 5698521.13 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:52:05,015 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.604e+02 1.200e+03 1.606e+03 2.220e+03 4.459e+03, threshold=3.212e+03, percent-clipped=8.0
+2023-03-14 11:52:06,753 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-14 11:52:29,903 INFO [train.py:968] (0/2) Epoch 28, batch 19000, giga_loss[loss=0.3291, simple_loss=0.3883, pruned_loss=0.135, over 27642.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3512, pruned_loss=0.09534, over 5701860.04 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.332, pruned_loss=0.08509, over 5699413.94 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3518, pruned_loss=0.09619, over 5699930.66 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:52:54,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6134, 1.9633, 1.5738, 1.7832], device='cuda:0'), covar=tensor([0.2685, 0.2731, 0.3053, 0.2349], device='cuda:0'), in_proj_covar=tensor([0.1597, 0.1150, 0.1412, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 11:52:54,638 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1249925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:52:57,292 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1249928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:53:16,624 INFO [train.py:968] (0/2) Epoch 28, batch 19050, giga_loss[loss=0.3112, simple_loss=0.3777, pruned_loss=0.1223, over 29010.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3537, pruned_loss=0.09944, over 5706988.63 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3323, pruned_loss=0.08519, over 5698954.46 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3541, pruned_loss=0.1002, over 5705717.69 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:53:22,806 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1249957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:53:32,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.083e+03 1.511e+03 1.825e+03 2.711e+03 4.867e+03, threshold=3.651e+03, percent-clipped=13.0
+2023-03-14 11:54:00,206 INFO [train.py:968] (0/2) Epoch 28, batch 19100, giga_loss[loss=0.2861, simple_loss=0.3528, pruned_loss=0.1097, over 28814.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3538, pruned_loss=0.1018, over 5709838.43 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.332, pruned_loss=0.08501, over 5702742.00 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3548, pruned_loss=0.1028, over 5705648.37 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:54:01,932 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1250000.pt
+2023-03-14 11:54:20,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1250026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:54:41,296 INFO [train.py:968] (0/2) Epoch 28, batch 19150, giga_loss[loss=0.3007, simple_loss=0.372, pruned_loss=0.1147, over 28494.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3531, pruned_loss=0.1025, over 5703685.54 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3325, pruned_loss=0.08517, over 5708505.48 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.354, pruned_loss=0.1037, over 5695547.11 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:54:54,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.501e+03 1.915e+03 2.555e+03 4.895e+03, threshold=3.829e+03, percent-clipped=7.0
+2023-03-14 11:55:23,888 INFO [train.py:968] (0/2) Epoch 28, batch 19200, giga_loss[loss=0.2582, simple_loss=0.3171, pruned_loss=0.09968, over 23928.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3492, pruned_loss=0.1007, over 5697885.09 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3325, pruned_loss=0.08506, over 5708817.14 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3504, pruned_loss=0.1021, over 5691086.72 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:55:47,606 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1250128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:56:04,394 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-14 11:56:06,202 INFO [train.py:968] (0/2) Epoch 28, batch 19250, giga_loss[loss=0.2709, simple_loss=0.3453, pruned_loss=0.09828, over 28501.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3476, pruned_loss=0.09894, over 5711589.75 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3328, pruned_loss=0.08491, over 5711978.00 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3489, pruned_loss=0.1007, over 5702978.95 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:56:21,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.341e+03 1.683e+03 2.385e+03 8.268e+03, threshold=3.365e+03, percent-clipped=6.0
+2023-03-14 11:56:23,170 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1250169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:56:25,336 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1250172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:56:46,473 INFO [train.py:968] (0/2) Epoch 28, batch 19300, libri_loss[loss=0.2684, simple_loss=0.3566, pruned_loss=0.09013, over 29236.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3463, pruned_loss=0.09734, over 5710860.99 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3331, pruned_loss=0.08504, over 5707098.80 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3474, pruned_loss=0.09903, over 5707886.42 frames. ], batch size: 97, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:56:47,985 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1250201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:57:31,171 INFO [train.py:968] (0/2) Epoch 28, batch 19350, giga_loss[loss=0.2389, simple_loss=0.3269, pruned_loss=0.07544, over 28902.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3444, pruned_loss=0.09622, over 5697091.59 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3335, pruned_loss=0.08524, over 5713926.26 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3453, pruned_loss=0.09776, over 5688052.17 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:57:45,597 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.183e+02 1.210e+03 1.477e+03 1.846e+03 4.290e+03, threshold=2.954e+03, percent-clipped=1.0
+2023-03-14 11:57:49,046 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1250271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:57:52,257 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1250274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:58:14,477 INFO [train.py:968] (0/2) Epoch 28, batch 19400, giga_loss[loss=0.2256, simple_loss=0.3065, pruned_loss=0.07238, over 28772.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3393, pruned_loss=0.09348, over 5685139.34 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3335, pruned_loss=0.08532, over 5698613.60 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3401, pruned_loss=0.09476, over 5691153.89 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:58:18,266 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1250303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:58:41,248 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9300, 1.2082, 1.3043, 1.1130], device='cuda:0'), covar=tensor([0.2005, 0.1455, 0.2298, 0.1580], device='cuda:0'), in_proj_covar=tensor([0.0496, 0.0751, 0.0724, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 11:58:58,746 INFO [train.py:968] (0/2) Epoch 28, batch 19450, giga_loss[loss=0.2296, simple_loss=0.3036, pruned_loss=0.07774, over 28752.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3349, pruned_loss=0.09153, over 5677260.63 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.334, pruned_loss=0.08543, over 5703034.68 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3353, pruned_loss=0.09266, over 5677557.76 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:59:17,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.310e+02 1.097e+03 1.408e+03 2.008e+03 1.180e+04, threshold=2.816e+03, percent-clipped=10.0
+2023-03-14 11:59:46,119 INFO [train.py:968] (0/2) Epoch 28, batch 19500, giga_loss[loss=0.243, simple_loss=0.3229, pruned_loss=0.08158, over 28779.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3302, pruned_loss=0.08958, over 5662896.74 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3339, pruned_loss=0.08549, over 5705116.19 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3305, pruned_loss=0.09056, over 5660326.04 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:00:31,253 INFO [train.py:968] (0/2) Epoch 28, batch 19550, giga_loss[loss=0.2319, simple_loss=0.3147, pruned_loss=0.07449, over 28872.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3312, pruned_loss=0.09001, over 5661770.73 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3343, pruned_loss=0.08554, over 5712293.85 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.331, pruned_loss=0.09091, over 5651261.66 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:00:47,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.811e+02 1.165e+03 1.544e+03 2.132e+03 4.687e+03, threshold=3.088e+03, percent-clipped=13.0
+2023-03-14 12:01:14,650 INFO [train.py:968] (0/2) Epoch 28, batch 19600, giga_loss[loss=0.3023, simple_loss=0.368, pruned_loss=0.1183, over 27652.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3321, pruned_loss=0.09056, over 5662776.05 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3341, pruned_loss=0.0854, over 5707059.14 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3321, pruned_loss=0.0915, over 5657922.96 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:01:27,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4691, 3.4695, 1.5392, 1.6696], device='cuda:0'), covar=tensor([0.1051, 0.0347, 0.0915, 0.1355], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0563, 0.0408, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 12:01:58,771 INFO [train.py:968] (0/2) Epoch 28, batch 19650, giga_loss[loss=0.2845, simple_loss=0.356, pruned_loss=0.1065, over 28766.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3307, pruned_loss=0.08969, over 5668643.29 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3343, pruned_loss=0.08545, over 5707208.85 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3305, pruned_loss=0.0904, over 5664526.43 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:02:07,158 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 12:02:15,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.851e+02 1.156e+03 1.413e+03 2.017e+03 6.128e+03, threshold=2.825e+03, percent-clipped=8.0
+2023-03-14 12:02:39,497 INFO [train.py:968] (0/2) Epoch 28, batch 19700, giga_loss[loss=0.2466, simple_loss=0.3154, pruned_loss=0.08888, over 28616.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3297, pruned_loss=0.08869, over 5682979.92 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3347, pruned_loss=0.0853, over 5714077.08 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.329, pruned_loss=0.08953, over 5672224.42 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:03:19,923 INFO [train.py:968] (0/2) Epoch 28, batch 19750, giga_loss[loss=0.2494, simple_loss=0.326, pruned_loss=0.08636, over 28879.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3279, pruned_loss=0.08795, over 5689062.37 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3347, pruned_loss=0.0852, over 5716001.34 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3273, pruned_loss=0.08874, over 5678594.41 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:03:31,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-14 12:03:34,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.502e+02 1.146e+03 1.461e+03 1.915e+03 3.994e+03, threshold=2.922e+03, percent-clipped=7.0
+2023-03-14 12:03:49,019 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6849, 1.8756, 1.5339, 1.9409], device='cuda:0'), covar=tensor([0.2697, 0.2963, 0.3264, 0.2678], device='cuda:0'), in_proj_covar=tensor([0.1594, 0.1146, 0.1408, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 12:03:58,518 INFO [train.py:968] (0/2) Epoch 28, batch 19800, libri_loss[loss=0.2243, simple_loss=0.3146, pruned_loss=0.06695, over 29557.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3249, pruned_loss=0.08586, over 5703823.20 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3344, pruned_loss=0.08487, over 5720765.46 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3245, pruned_loss=0.08684, over 5690738.34 frames. ], batch size: 75, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:04:11,726 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-14 12:04:38,664 INFO [train.py:968] (0/2) Epoch 28, batch 19850, giga_loss[loss=0.2604, simple_loss=0.3353, pruned_loss=0.09276, over 28986.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3236, pruned_loss=0.08569, over 5705623.03 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3344, pruned_loss=0.08478, over 5723686.85 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3231, pruned_loss=0.08656, over 5692395.33 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:04:54,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.850e+02 1.133e+03 1.429e+03 2.112e+03 5.052e+03, threshold=2.858e+03, percent-clipped=15.0
+2023-03-14 12:05:20,641 INFO [train.py:968] (0/2) Epoch 28, batch 19900, giga_loss[loss=0.2409, simple_loss=0.3065, pruned_loss=0.08766, over 28715.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.321, pruned_loss=0.08497, over 5713552.89 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3346, pruned_loss=0.08478, over 5725498.47 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3204, pruned_loss=0.08565, over 5701562.72 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:05:58,573 INFO [train.py:968] (0/2) Epoch 28, batch 19950, giga_loss[loss=0.2372, simple_loss=0.3185, pruned_loss=0.078, over 28927.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3206, pruned_loss=0.08427, over 5725043.02 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3361, pruned_loss=0.08518, over 5730640.85 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3184, pruned_loss=0.08446, over 5710615.13 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:06:14,321 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.746e+02 1.093e+03 1.270e+03 1.677e+03 3.113e+03, threshold=2.540e+03, percent-clipped=2.0
+2023-03-14 12:06:33,933 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-14 12:06:40,207 INFO [train.py:968] (0/2) Epoch 28, batch 20000, giga_loss[loss=0.2173, simple_loss=0.2915, pruned_loss=0.07152, over 28439.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3191, pruned_loss=0.0837, over 5719133.94 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3367, pruned_loss=0.08537, over 5732000.14 frames. ], giga_tot_loss[loss=0.242, simple_loss=0.3166, pruned_loss=0.08368, over 5706366.14 frames. ], batch size: 65, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:07:01,868 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4902, 1.6654, 1.7012, 1.2956], device='cuda:0'), covar=tensor([0.1834, 0.2894, 0.1652, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0713, 0.0984, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 12:07:18,837 INFO [train.py:968] (0/2) Epoch 28, batch 20050, libri_loss[loss=0.2684, simple_loss=0.3704, pruned_loss=0.08318, over 29534.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3179, pruned_loss=0.08261, over 5713929.04 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3375, pruned_loss=0.08553, over 5725236.96 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3147, pruned_loss=0.08238, over 5709990.29 frames. ], batch size: 84, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:07:32,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.000e+02 1.169e+03 1.412e+03 1.762e+03 5.180e+03, threshold=2.824e+03, percent-clipped=8.0
+2023-03-14 12:07:57,196 INFO [train.py:968] (0/2) Epoch 28, batch 20100, giga_loss[loss=0.3141, simple_loss=0.3807, pruned_loss=0.1237, over 27928.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3186, pruned_loss=0.08322, over 5709628.14 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3377, pruned_loss=0.08562, over 5718183.82 frames. ], giga_tot_loss[loss=0.2407, simple_loss=0.3156, pruned_loss=0.08291, over 5712062.04 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:08:33,209 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5764, 1.8685, 1.6811, 1.6000], device='cuda:0'), covar=tensor([0.2193, 0.2051, 0.2498, 0.2269], device='cuda:0'), in_proj_covar=tensor([0.0505, 0.0761, 0.0734, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 12:08:40,250 INFO [train.py:968] (0/2) Epoch 28, batch 20150, giga_loss[loss=0.2738, simple_loss=0.3446, pruned_loss=0.1016, over 28883.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3239, pruned_loss=0.08648, over 5713200.09 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3375, pruned_loss=0.08524, over 5724922.87 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.321, pruned_loss=0.08654, over 5708935.89 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:08:55,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.978e+02 1.269e+03 1.699e+03 2.494e+03 8.628e+03, threshold=3.399e+03, percent-clipped=16.0
+2023-03-14 12:09:16,159 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-14 12:09:23,054 INFO [train.py:968] (0/2) Epoch 28, batch 20200, giga_loss[loss=0.2524, simple_loss=0.3252, pruned_loss=0.08974, over 28751.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3285, pruned_loss=0.0894, over 5697515.19 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3379, pruned_loss=0.08542, over 5719322.46 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3255, pruned_loss=0.08939, over 5698127.55 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:09:39,968 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:09:52,162 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-14 12:10:09,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4912, 1.8332, 1.5071, 1.7450], device='cuda:0'), covar=tensor([0.2553, 0.2549, 0.2865, 0.2194], device='cuda:0'), in_proj_covar=tensor([0.1598, 0.1149, 0.1413, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 12:10:12,452 INFO [train.py:968] (0/2) Epoch 28, batch 20250, giga_loss[loss=0.3814, simple_loss=0.4249, pruned_loss=0.169, over 26758.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3356, pruned_loss=0.09408, over 5697288.34 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.338, pruned_loss=0.08537, over 5722914.01 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.333, pruned_loss=0.09427, over 5694060.45 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:10:31,921 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.491e+03 1.965e+03 2.841e+03 9.239e+03, threshold=3.930e+03, percent-clipped=17.0
+2023-03-14 12:10:55,251 INFO [train.py:968] (0/2) Epoch 28, batch 20300, giga_loss[loss=0.288, simple_loss=0.3575, pruned_loss=0.1092, over 28868.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3401, pruned_loss=0.09626, over 5697588.06 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3375, pruned_loss=0.08502, over 5728260.15 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3384, pruned_loss=0.09699, over 5689610.39 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:11:10,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4875, 1.8430, 1.6733, 1.6386], device='cuda:0'), covar=tensor([0.2141, 0.1957, 0.2442, 0.2099], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0758, 0.0730, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 12:11:14,737 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:11:24,292 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:11:42,984 INFO [train.py:968] (0/2) Epoch 28, batch 20350, giga_loss[loss=0.3012, simple_loss=0.3745, pruned_loss=0.1139, over 28747.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.345, pruned_loss=0.09838, over 5690976.09 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3376, pruned_loss=0.08511, over 5728385.82 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3436, pruned_loss=0.09909, over 5684024.95 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:11:59,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.790e+02 1.273e+03 1.542e+03 1.887e+03 7.043e+03, threshold=3.083e+03, percent-clipped=2.0
+2023-03-14 12:12:05,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8691, 1.9599, 1.5093, 1.5788], device='cuda:0'), covar=tensor([0.1096, 0.0745, 0.1062, 0.1279], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0450, 0.0524, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 12:12:28,634 INFO [train.py:968] (0/2) Epoch 28, batch 20400, giga_loss[loss=0.3194, simple_loss=0.3925, pruned_loss=0.1231, over 28957.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3495, pruned_loss=0.09984, over 5690960.99 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3378, pruned_loss=0.08514, over 5730782.88 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3483, pruned_loss=0.1006, over 5682685.10 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:13:12,157 INFO [train.py:968] (0/2) Epoch 28, batch 20450, giga_loss[loss=0.2895, simple_loss=0.3636, pruned_loss=0.1077, over 28946.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3524, pruned_loss=0.101, over 5697953.01 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3382, pruned_loss=0.08532, over 5733300.71 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3514, pruned_loss=0.1017, over 5688697.61 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:13:31,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.898e+02 1.329e+03 1.649e+03 2.220e+03 5.803e+03, threshold=3.297e+03, percent-clipped=11.0
+2023-03-14 12:13:52,354 INFO [train.py:968] (0/2) Epoch 28, batch 20500, giga_loss[loss=0.2418, simple_loss=0.3319, pruned_loss=0.07585, over 28954.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3471, pruned_loss=0.09736, over 5692803.89 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.338, pruned_loss=0.08543, over 5731146.51 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3469, pruned_loss=0.09839, over 5685840.25 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:13:54,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5827, 1.6601, 1.7785, 1.3849], device='cuda:0'), covar=tensor([0.2003, 0.2741, 0.1732, 0.2078], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0714, 0.0984, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 12:14:05,884 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2505, 1.9927, 1.4820, 0.4921], device='cuda:0'), covar=tensor([0.4835, 0.2443, 0.3665, 0.5266], device='cuda:0'), in_proj_covar=tensor([0.1843, 0.1732, 0.1657, 0.1500], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 12:14:34,284 INFO [train.py:968] (0/2) Epoch 28, batch 20550, giga_loss[loss=0.2368, simple_loss=0.3269, pruned_loss=0.07334, over 28947.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3461, pruned_loss=0.09598, over 5701040.17 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.338, pruned_loss=0.08539, over 5733595.37 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3461, pruned_loss=0.09698, over 5692981.56 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:14:53,184 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.553e+02 1.421e+03 1.932e+03 2.859e+03 5.065e+03, threshold=3.864e+03, percent-clipped=12.0
+2023-03-14 12:15:10,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1251492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:15:17,008 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 12:15:17,244 INFO [train.py:968] (0/2) Epoch 28, batch 20600, giga_loss[loss=0.2688, simple_loss=0.3556, pruned_loss=0.09101, over 28663.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3458, pruned_loss=0.09536, over 5697186.54 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3381, pruned_loss=0.08546, over 5738814.52 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3459, pruned_loss=0.09639, over 5684905.17 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:15:25,129 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:15:48,702 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:15:48,795 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4988, 2.1129, 1.5334, 0.8041], device='cuda:0'), covar=tensor([0.6939, 0.3205, 0.4737, 0.7161], device='cuda:0'), in_proj_covar=tensor([0.1841, 0.1731, 0.1656, 0.1500], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 12:15:57,895 INFO [train.py:968] (0/2) Epoch 28, batch 20650, giga_loss[loss=0.2649, simple_loss=0.3477, pruned_loss=0.09105, over 28972.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3459, pruned_loss=0.095, over 5704039.81 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.338, pruned_loss=0.08545, over 5742736.29 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3463, pruned_loss=0.09611, over 5689431.02 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:16:17,321 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.355e+03 1.824e+03 2.674e+03 5.840e+03, threshold=3.649e+03, percent-clipped=5.0
+2023-03-14 12:16:38,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1251594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:16:42,935 INFO [train.py:968] (0/2) Epoch 28, batch 20700, giga_loss[loss=0.3339, simple_loss=0.3959, pruned_loss=0.1359, over 27634.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3498, pruned_loss=0.09792, over 5694643.12 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3388, pruned_loss=0.08608, over 5737130.12 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3497, pruned_loss=0.09851, over 5687754.75 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:16:47,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1251605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:17:16,820 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1251635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:17:18,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1251638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:17:27,207 INFO [train.py:968] (0/2) Epoch 28, batch 20750, giga_loss[loss=0.2748, simple_loss=0.35, pruned_loss=0.09982, over 28999.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.352, pruned_loss=0.1001, over 5701076.09 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.339, pruned_loss=0.08625, over 5741272.81 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3521, pruned_loss=0.1007, over 5690968.99 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:17:47,282 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1251667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:17:49,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.487e+03 1.957e+03 2.590e+03 7.291e+03, threshold=3.913e+03, percent-clipped=8.0
+2023-03-14 12:18:02,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4712, 1.3759, 3.8657, 3.3686], device='cuda:0'), covar=tensor([0.1570, 0.2736, 0.0450, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0670, 0.0996, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 12:18:16,219 INFO [train.py:968] (0/2) Epoch 28, batch 20800, giga_loss[loss=0.2597, simple_loss=0.338, pruned_loss=0.09073, over 28831.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3527, pruned_loss=0.1008, over 5713929.44 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.339, pruned_loss=0.08636, over 5745361.44 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3531, pruned_loss=0.1016, over 5700953.20 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:18:50,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1251737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:18:53,281 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1251740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:18:59,816 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1251748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:19:00,172 INFO [train.py:968] (0/2) Epoch 28, batch 20850, giga_loss[loss=0.3863, simple_loss=0.4194, pruned_loss=0.1766, over 26605.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3555, pruned_loss=0.1036, over 5707575.85 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3394, pruned_loss=0.08659, over 5744976.83 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3556, pruned_loss=0.1042, over 5697392.60 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:19:02,041 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1251751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:19:19,451 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1251769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:19:20,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.754e+02 1.328e+03 1.620e+03 2.120e+03 4.092e+03, threshold=3.240e+03, percent-clipped=1.0
+2023-03-14 12:19:20,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5006, 5.3240, 5.0561, 2.5591], device='cuda:0'), covar=tensor([0.0468, 0.0608, 0.0695, 0.1821], device='cuda:0'), in_proj_covar=tensor([0.1276, 0.1181, 0.0995, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 12:19:27,992 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1251780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:19:43,069 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1662, 1.4683, 1.4716, 1.3336], device='cuda:0'), covar=tensor([0.2054, 0.1631, 0.2259, 0.1736], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0758, 0.0730, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 12:19:43,400 INFO [train.py:968] (0/2) Epoch 28, batch 20900, giga_loss[loss=0.3057, simple_loss=0.3786, pruned_loss=0.1164, over 27822.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3559, pruned_loss=0.1038, over 5714795.89 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3395, pruned_loss=0.08674, over 5746644.19 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3561, pruned_loss=0.1043, over 5704971.10 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:19:43,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4206, 1.5055, 3.2942, 3.1684], device='cuda:0'), covar=tensor([0.1332, 0.2474, 0.0447, 0.1071], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0667, 0.0993, 0.0972], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 12:20:15,634 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-14 12:20:21,506 INFO [train.py:968] (0/2) Epoch 28, batch 20950, giga_loss[loss=0.2955, simple_loss=0.3718, pruned_loss=0.1096, over 28972.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3541, pruned_loss=0.1018, over 5716476.67 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3397, pruned_loss=0.08711, over 5751941.05 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3546, pruned_loss=0.1024, over 5702784.17 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:20:38,681 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.200e+02 1.366e+03 1.788e+03 2.622e+03 7.303e+03, threshold=3.576e+03, percent-clipped=8.0
+2023-03-14 12:20:44,976 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:20:49,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1251884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:21:01,998 INFO [train.py:968] (0/2) Epoch 28, batch 21000, giga_loss[loss=0.2786, simple_loss=0.3557, pruned_loss=0.1007, over 28901.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3548, pruned_loss=0.1011, over 5721371.28 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3398, pruned_loss=0.08718, over 5754521.79 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3554, pruned_loss=0.1019, over 5707446.67 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:21:02,003 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 12:21:11,421 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2054, simple_loss=0.3134, pruned_loss=0.04863, over 944034.00 frames. 
+2023-03-14 12:21:11,422 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 12:21:19,972 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1995, 2.3533, 2.3888, 1.9057], device='cuda:0'), covar=tensor([0.1698, 0.2245, 0.1354, 0.1680], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0713, 0.0980, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 12:21:21,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1251911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:21:45,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5879, 1.8550, 1.4563, 1.8611], device='cuda:0'), covar=tensor([0.2658, 0.2608, 0.3110, 0.2241], device='cuda:0'), in_proj_covar=tensor([0.1600, 0.1153, 0.1414, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 12:21:51,704 INFO [train.py:968] (0/2) Epoch 28, batch 21050, giga_loss[loss=0.2417, simple_loss=0.3256, pruned_loss=0.07896, over 28897.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3535, pruned_loss=0.1002, over 5727820.17 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3398, pruned_loss=0.08736, over 5757994.47 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3543, pruned_loss=0.1009, over 5712755.18 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:22:07,883 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.254e+02 1.155e+03 1.427e+03 1.853e+03 5.563e+03, threshold=2.855e+03, percent-clipped=7.0
+2023-03-14 12:22:08,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9938, 1.2432, 1.0191, 0.2804], device='cuda:0'), covar=tensor([0.3814, 0.2750, 0.3941, 0.6494], device='cuda:0'), in_proj_covar=tensor([0.1821, 0.1709, 0.1638, 0.1485], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 12:22:13,171 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2646, 2.3518, 2.3182, 2.0208], device='cuda:0'), covar=tensor([0.2597, 0.2413, 0.2407, 0.2726], device='cuda:0'), in_proj_covar=tensor([0.2046, 0.2002, 0.1903, 0.2055], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 12:22:29,154 INFO [train.py:968] (0/2) Epoch 28, batch 21100, giga_loss[loss=0.2988, simple_loss=0.3555, pruned_loss=0.1211, over 26605.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3508, pruned_loss=0.09905, over 5722015.65 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3401, pruned_loss=0.08761, over 5761749.51 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3515, pruned_loss=0.09971, over 5705882.52 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:22:29,853 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1252000.pt
+2023-03-14 12:22:36,040 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252008.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 12:22:53,172 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:22:54,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:23:10,027 INFO [train.py:968] (0/2) Epoch 28, batch 21150, giga_loss[loss=0.242, simple_loss=0.3286, pruned_loss=0.07771, over 28727.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3479, pruned_loss=0.09739, over 5717312.89 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3398, pruned_loss=0.08759, over 5760739.95 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3489, pruned_loss=0.09816, over 5704506.17 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:23:15,083 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:23:17,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:23:18,525 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:23:26,818 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.200e+02 1.204e+03 1.450e+03 2.041e+03 6.342e+03, threshold=2.900e+03, percent-clipped=12.0
+2023-03-14 12:23:27,866 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 12:23:37,754 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:23:42,291 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 12:23:44,519 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.30 vs. limit=5.0
+2023-03-14 12:23:48,776 INFO [train.py:968] (0/2) Epoch 28, batch 21200, giga_loss[loss=0.2998, simple_loss=0.3772, pruned_loss=0.1112, over 28933.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3473, pruned_loss=0.09708, over 5713422.77 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3407, pruned_loss=0.08815, over 5748687.83 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3476, pruned_loss=0.09757, over 5712166.41 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:24:28,110 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:24:29,288 INFO [train.py:968] (0/2) Epoch 28, batch 21250, giga_loss[loss=0.2811, simple_loss=0.3664, pruned_loss=0.09794, over 29048.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3477, pruned_loss=0.09766, over 5714180.91 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.341, pruned_loss=0.08849, over 5750721.20 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3478, pruned_loss=0.09803, over 5710223.34 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:24:51,915 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.317e+02 1.239e+03 1.649e+03 2.114e+03 4.959e+03, threshold=3.299e+03, percent-clipped=8.0
+2023-03-14 12:25:12,824 INFO [train.py:968] (0/2) Epoch 28, batch 21300, giga_loss[loss=0.3103, simple_loss=0.3752, pruned_loss=0.1228, over 28570.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3481, pruned_loss=0.09802, over 5711088.51 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3413, pruned_loss=0.08879, over 5752412.76 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.348, pruned_loss=0.09817, over 5705818.16 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:25:36,309 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:25:55,057 INFO [train.py:968] (0/2) Epoch 28, batch 21350, giga_loss[loss=0.2568, simple_loss=0.3381, pruned_loss=0.08777, over 28846.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3479, pruned_loss=0.09699, over 5718361.13 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3415, pruned_loss=0.089, over 5755734.36 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3478, pruned_loss=0.09706, over 5710663.05 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:25:57,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:26:01,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4541, 1.6789, 1.4935, 1.6548], device='cuda:0'), covar=tensor([0.0812, 0.0330, 0.0327, 0.0885], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0120, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 12:26:10,288 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.032e+02 1.107e+03 1.440e+03 1.901e+03 4.365e+03, threshold=2.879e+03, percent-clipped=2.0
+2023-03-14 12:26:13,933 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:26:22,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5272, 2.7798, 1.5947, 1.6376], device='cuda:0'), covar=tensor([0.0825, 0.0287, 0.0766, 0.1130], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0562, 0.0407, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 12:26:31,752 INFO [train.py:968] (0/2) Epoch 28, batch 21400, giga_loss[loss=0.298, simple_loss=0.363, pruned_loss=0.1165, over 27593.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3477, pruned_loss=0.09686, over 5715121.44 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3422, pruned_loss=0.08968, over 5759513.38 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3472, pruned_loss=0.09658, over 5704145.50 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:27:15,753 INFO [train.py:968] (0/2) Epoch 28, batch 21450, giga_loss[loss=0.2445, simple_loss=0.3234, pruned_loss=0.08284, over 28475.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3471, pruned_loss=0.09736, over 5706283.19 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3422, pruned_loss=0.08971, over 5760239.81 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3467, pruned_loss=0.09715, over 5696801.37 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:27:28,738 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-14 12:27:33,433 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.178e+02 1.133e+03 1.395e+03 1.695e+03 4.416e+03, threshold=2.791e+03, percent-clipped=4.0
+2023-03-14 12:27:41,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252383.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 12:27:42,810 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4852, 1.5077, 1.6040, 1.5053], device='cuda:0'), covar=tensor([0.3119, 0.3121, 0.2332, 0.2699], device='cuda:0'), in_proj_covar=tensor([0.2045, 0.2002, 0.1901, 0.2054], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 12:27:51,312 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:27:54,030 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:27:54,444 INFO [train.py:968] (0/2) Epoch 28, batch 21500, giga_loss[loss=0.2406, simple_loss=0.3131, pruned_loss=0.08405, over 28854.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3453, pruned_loss=0.09704, over 5705693.75 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3422, pruned_loss=0.0899, over 5759691.16 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3451, pruned_loss=0.09689, over 5697494.78 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:28:16,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:28:17,684 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:28:34,460 INFO [train.py:968] (0/2) Epoch 28, batch 21550, libri_loss[loss=0.294, simple_loss=0.3683, pruned_loss=0.1099, over 29530.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3419, pruned_loss=0.09525, over 5703392.76 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3427, pruned_loss=0.09036, over 5762501.69 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3413, pruned_loss=0.09484, over 5692654.95 frames. ], batch size: 89, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:28:50,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.617e+02 1.380e+03 1.715e+03 2.347e+03 5.459e+03, threshold=3.431e+03, percent-clipped=16.0
+2023-03-14 12:29:11,633 INFO [train.py:968] (0/2) Epoch 28, batch 21600, giga_loss[loss=0.2414, simple_loss=0.3139, pruned_loss=0.08443, over 28975.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3407, pruned_loss=0.09487, over 5706030.18 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3426, pruned_loss=0.09046, over 5766808.20 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3402, pruned_loss=0.0946, over 5691869.42 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:29:32,021 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:29:36,119 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252526.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 12:29:38,167 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252529.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 12:29:51,999 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1672, 1.3564, 3.4011, 3.0193], device='cuda:0'), covar=tensor([0.1686, 0.2685, 0.0503, 0.1003], device='cuda:0'), in_proj_covar=tensor([0.0797, 0.0666, 0.0990, 0.0971], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 12:29:52,431 INFO [train.py:968] (0/2) Epoch 28, batch 21650, giga_loss[loss=0.2408, simple_loss=0.3173, pruned_loss=0.08215, over 28258.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3396, pruned_loss=0.09468, over 5708086.56 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3426, pruned_loss=0.0906, over 5768999.16 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3392, pruned_loss=0.09441, over 5694023.63 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:29:57,637 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 12:30:00,135 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252558.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 12:30:05,256 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 12:30:11,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.369e+02 1.284e+03 1.571e+03 2.074e+03 3.481e+03, threshold=3.141e+03, percent-clipped=1.0
+2023-03-14 12:30:22,417 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:30:36,176 INFO [train.py:968] (0/2) Epoch 28, batch 21700, giga_loss[loss=0.2798, simple_loss=0.3479, pruned_loss=0.1059, over 28817.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3377, pruned_loss=0.09428, over 5707956.06 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3424, pruned_loss=0.09067, over 5771221.45 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3375, pruned_loss=0.09407, over 5693893.94 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:30:40,639 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:30:47,731 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-14 12:31:16,925 INFO [train.py:968] (0/2) Epoch 28, batch 21750, giga_loss[loss=0.2283, simple_loss=0.3057, pruned_loss=0.07549, over 28893.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3349, pruned_loss=0.09296, over 5710528.16 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3426, pruned_loss=0.09097, over 5772762.24 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3345, pruned_loss=0.09256, over 5697199.74 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:31:19,319 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:31:29,877 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:31:33,031 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:31:37,296 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.262e+02 1.118e+03 1.317e+03 1.658e+03 4.887e+03, threshold=2.635e+03, percent-clipped=2.0
+2023-03-14 12:31:57,419 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:31:58,658 INFO [train.py:968] (0/2) Epoch 28, batch 21800, giga_loss[loss=0.24, simple_loss=0.3133, pruned_loss=0.08335, over 28746.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3321, pruned_loss=0.0915, over 5719530.58 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3425, pruned_loss=0.0909, over 5774147.19 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3317, pruned_loss=0.09126, over 5707308.43 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:32:36,149 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:32:37,410 INFO [train.py:968] (0/2) Epoch 28, batch 21850, giga_loss[loss=0.2379, simple_loss=0.3061, pruned_loss=0.08486, over 28575.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3304, pruned_loss=0.09053, over 5719424.92 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3421, pruned_loss=0.09075, over 5776017.49 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3303, pruned_loss=0.09048, over 5707270.90 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:32:38,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:32:53,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5220, 1.5935, 1.6113, 1.5112], device='cuda:0'), covar=tensor([0.3201, 0.2942, 0.2428, 0.2716], device='cuda:0'), in_proj_covar=tensor([0.2045, 0.2005, 0.1907, 0.2058], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 12:32:57,484 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-14 12:32:57,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.388e+02 1.233e+03 1.533e+03 2.106e+03 1.250e+04, threshold=3.066e+03, percent-clipped=17.0
+2023-03-14 12:33:02,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:33:16,224 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5992, 1.4014, 4.6407, 3.4969], device='cuda:0'), covar=tensor([0.1685, 0.2947, 0.0383, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0799, 0.0668, 0.0995, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 12:33:16,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:33:19,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:33:19,433 INFO [train.py:968] (0/2) Epoch 28, batch 21900, giga_loss[loss=0.297, simple_loss=0.3591, pruned_loss=0.1175, over 28653.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3323, pruned_loss=0.09055, over 5719919.23 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3422, pruned_loss=0.0909, over 5777760.41 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.332, pruned_loss=0.09038, over 5708008.82 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:33:22,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:33:42,934 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:34:02,999 INFO [train.py:968] (0/2) Epoch 28, batch 21950, giga_loss[loss=0.2732, simple_loss=0.3563, pruned_loss=0.09504, over 28305.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3359, pruned_loss=0.09221, over 5708515.26 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3425, pruned_loss=0.09122, over 5780401.08 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3352, pruned_loss=0.09179, over 5695112.85 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:34:19,209 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:34:23,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.049e+02 1.329e+03 1.543e+03 1.991e+03 5.161e+03, threshold=3.087e+03, percent-clipped=7.0
+2023-03-14 12:34:45,022 INFO [train.py:968] (0/2) Epoch 28, batch 22000, giga_loss[loss=0.3361, simple_loss=0.3901, pruned_loss=0.1411, over 26620.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3388, pruned_loss=0.0933, over 5704730.52 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3428, pruned_loss=0.09169, over 5779253.92 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3379, pruned_loss=0.09254, over 5693654.57 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:35:07,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4783, 2.2653, 1.6609, 0.6101], device='cuda:0'), covar=tensor([0.5664, 0.3381, 0.4030, 0.6733], device='cuda:0'), in_proj_covar=tensor([0.1827, 0.1709, 0.1645, 0.1487], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 12:35:22,508 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:35:25,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:35:27,312 INFO [train.py:968] (0/2) Epoch 28, batch 22050, giga_loss[loss=0.2828, simple_loss=0.3601, pruned_loss=0.1028, over 28286.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3404, pruned_loss=0.09344, over 5710223.14 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3435, pruned_loss=0.09225, over 5782087.24 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3389, pruned_loss=0.09236, over 5697134.24 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:35:35,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:35:36,866 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:35:42,401 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 12:35:46,045 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.111e+02 1.152e+03 1.412e+03 1.798e+03 5.360e+03, threshold=2.824e+03, percent-clipped=5.0
+2023-03-14 12:35:48,146 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:35:55,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8008, 5.2398, 2.1007, 1.8888], device='cuda:0'), covar=tensor([0.0872, 0.0310, 0.0831, 0.1280], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0564, 0.0408, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 12:36:09,156 INFO [train.py:968] (0/2) Epoch 28, batch 22100, giga_loss[loss=0.2535, simple_loss=0.3348, pruned_loss=0.08611, over 28901.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3391, pruned_loss=0.0921, over 5708522.48 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3437, pruned_loss=0.09246, over 5782821.07 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3377, pruned_loss=0.09105, over 5696124.95 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:36:19,587 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-14 12:36:50,601 INFO [train.py:968] (0/2) Epoch 28, batch 22150, giga_loss[loss=0.2705, simple_loss=0.3401, pruned_loss=0.1004, over 28658.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3396, pruned_loss=0.09284, over 5706869.55 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.344, pruned_loss=0.09274, over 5783775.30 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3381, pruned_loss=0.09177, over 5695165.08 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:37:08,605 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1253070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:37:11,537 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.721e+02 1.380e+03 1.820e+03 2.448e+03 5.650e+03, threshold=3.640e+03, percent-clipped=13.0
+2023-03-14 12:37:30,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6771, 2.1397, 1.6801, 1.1381], device='cuda:0'), covar=tensor([0.5518, 0.3599, 0.3773, 0.5681], device='cuda:0'), in_proj_covar=tensor([0.1830, 0.1712, 0.1647, 0.1489], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 12:37:32,345 INFO [train.py:968] (0/2) Epoch 28, batch 22200, giga_loss[loss=0.2633, simple_loss=0.3281, pruned_loss=0.0993, over 23728.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3405, pruned_loss=0.09405, over 5696131.37 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3442, pruned_loss=0.09289, over 5775871.77 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3391, pruned_loss=0.09306, over 5693782.62 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:37:36,195 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1253103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:37:38,074 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1253106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:38:01,593 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1253135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:38:12,237 INFO [train.py:968] (0/2) Epoch 28, batch 22250, giga_loss[loss=0.265, simple_loss=0.3423, pruned_loss=0.09385, over 28888.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3422, pruned_loss=0.09487, over 5702311.26 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3443, pruned_loss=0.093, over 5778914.95 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3409, pruned_loss=0.09401, over 5696047.87 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:38:33,757 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.340e+03 1.563e+03 2.237e+03 8.439e+03, threshold=3.126e+03, percent-clipped=2.0
+2023-03-14 12:38:52,948 INFO [train.py:968] (0/2) Epoch 28, batch 22300, giga_loss[loss=0.2826, simple_loss=0.3615, pruned_loss=0.1019, over 28954.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3452, pruned_loss=0.09622, over 5708355.73 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3445, pruned_loss=0.09317, over 5777789.38 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.344, pruned_loss=0.0954, over 5704198.63 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:39:26,615 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5330, 2.2260, 1.5646, 0.8022], device='cuda:0'), covar=tensor([0.7146, 0.3054, 0.4528, 0.7418], device='cuda:0'), in_proj_covar=tensor([0.1831, 0.1714, 0.1649, 0.1490], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 12:39:29,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1253243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:39:34,005 INFO [train.py:968] (0/2) Epoch 28, batch 22350, giga_loss[loss=0.2871, simple_loss=0.3654, pruned_loss=0.1044, over 28951.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3489, pruned_loss=0.0984, over 5712680.80 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3454, pruned_loss=0.09371, over 5779609.26 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3472, pruned_loss=0.09733, over 5706479.73 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:39:52,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.569e+02 1.385e+03 1.759e+03 2.670e+03 7.163e+03, threshold=3.519e+03, percent-clipped=18.0
+2023-03-14 12:40:10,445 INFO [train.py:968] (0/2) Epoch 28, batch 22400, giga_loss[loss=0.2701, simple_loss=0.3477, pruned_loss=0.09618, over 28913.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3504, pruned_loss=0.09914, over 5720417.47 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3463, pruned_loss=0.09443, over 5782110.95 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3483, pruned_loss=0.09778, over 5711223.58 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:40:41,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1253337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:40:51,142 INFO [train.py:968] (0/2) Epoch 28, batch 22450, libri_loss[loss=0.3359, simple_loss=0.4047, pruned_loss=0.1335, over 29064.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3495, pruned_loss=0.09826, over 5717148.33 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3467, pruned_loss=0.09482, over 5781113.08 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3475, pruned_loss=0.09692, over 5709010.87 frames. ], batch size: 101, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:41:06,240 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2367, 0.7986, 0.8276, 1.4208], device='cuda:0'), covar=tensor([0.0763, 0.0391, 0.0397, 0.0907], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 12:41:10,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.892e+02 1.456e+03 1.760e+03 2.611e+03 6.720e+03, threshold=3.519e+03, percent-clipped=10.0
+2023-03-14 12:41:21,046 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1253386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:41:23,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1253389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:41:31,203 INFO [train.py:968] (0/2) Epoch 28, batch 22500, giga_loss[loss=0.3408, simple_loss=0.3906, pruned_loss=0.1455, over 27699.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3503, pruned_loss=0.09921, over 5720038.84 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3473, pruned_loss=0.09529, over 5782852.73 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3483, pruned_loss=0.09779, over 5711213.69 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:41:47,962 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1253418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:41:58,075 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-14 12:42:10,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1253445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:42:12,002 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 12:42:12,842 INFO [train.py:968] (0/2) Epoch 28, batch 22550, giga_loss[loss=0.2456, simple_loss=0.3285, pruned_loss=0.08139, over 28682.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3473, pruned_loss=0.09737, over 5722352.09 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3472, pruned_loss=0.09526, over 5783613.47 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3458, pruned_loss=0.09629, over 5714469.85 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:42:35,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.064e+02 1.232e+03 1.454e+03 2.109e+03 4.668e+03, threshold=2.909e+03, percent-clipped=3.0
+2023-03-14 12:42:41,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1253480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:42:43,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1253483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:42:54,062 INFO [train.py:968] (0/2) Epoch 28, batch 22600, giga_loss[loss=0.24, simple_loss=0.3148, pruned_loss=0.08257, over 28777.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3449, pruned_loss=0.0967, over 5726828.10 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3476, pruned_loss=0.09578, over 5785257.22 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3433, pruned_loss=0.09537, over 5717600.96 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:43:02,606 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1253512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:43:31,459 INFO [train.py:968] (0/2) Epoch 28, batch 22650, giga_loss[loss=0.213, simple_loss=0.2934, pruned_loss=0.06626, over 28613.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3427, pruned_loss=0.09575, over 5716323.77 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.348, pruned_loss=0.09629, over 5777319.14 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.341, pruned_loss=0.09424, over 5714443.42 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:43:50,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.018e+02 1.253e+03 1.579e+03 2.386e+03 1.080e+04, threshold=3.158e+03, percent-clipped=17.0
+2023-03-14 12:43:59,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8258, 1.2357, 1.2794, 1.0569], device='cuda:0'), covar=tensor([0.2159, 0.1349, 0.2368, 0.1741], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0754, 0.0726, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 12:44:02,342 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1253588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:44:04,540 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1253591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:44:07,377 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-14 12:44:10,921 INFO [train.py:968] (0/2) Epoch 28, batch 22700, giga_loss[loss=0.2578, simple_loss=0.336, pruned_loss=0.08983, over 28922.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3434, pruned_loss=0.0951, over 5717609.87 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3481, pruned_loss=0.09648, over 5778184.78 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3418, pruned_loss=0.09372, over 5714125.81 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:44:30,268 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1253620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:44:30,367 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6841, 1.8176, 1.5196, 1.7240], device='cuda:0'), covar=tensor([0.2895, 0.3055, 0.3511, 0.2681], device='cuda:0'), in_proj_covar=tensor([0.1598, 0.1151, 0.1411, 0.1010], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 12:44:53,312 INFO [train.py:968] (0/2) Epoch 28, batch 22750, giga_loss[loss=0.2957, simple_loss=0.3746, pruned_loss=0.1084, over 28541.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3452, pruned_loss=0.09542, over 5712712.83 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.348, pruned_loss=0.09651, over 5776119.85 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3439, pruned_loss=0.09424, over 5710234.73 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:45:13,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.389e+03 1.688e+03 2.254e+03 7.357e+03, threshold=3.376e+03, percent-clipped=11.0
+2023-03-14 12:45:31,339 INFO [train.py:968] (0/2) Epoch 28, batch 22800, giga_loss[loss=0.2877, simple_loss=0.3587, pruned_loss=0.1084, over 28939.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3452, pruned_loss=0.09581, over 5725110.77 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3478, pruned_loss=0.09669, over 5779669.94 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3442, pruned_loss=0.09469, over 5718808.17 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:45:51,532 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4555, 1.6536, 1.6363, 1.4908], device='cuda:0'), covar=tensor([0.4052, 0.3265, 0.2729, 0.3219], device='cuda:0'), in_proj_covar=tensor([0.2040, 0.2003, 0.1907, 0.2052], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 12:46:04,444 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1253737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:46:13,276 INFO [train.py:968] (0/2) Epoch 28, batch 22850, giga_loss[loss=0.2776, simple_loss=0.3464, pruned_loss=0.1044, over 29058.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3441, pruned_loss=0.09608, over 5715175.58 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3485, pruned_loss=0.09733, over 5769527.05 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3426, pruned_loss=0.09456, over 5717914.24 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:46:32,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.459e+03 1.833e+03 2.586e+03 4.956e+03, threshold=3.666e+03, percent-clipped=11.0
+2023-03-14 12:46:48,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-14 12:46:51,302 INFO [train.py:968] (0/2) Epoch 28, batch 22900, giga_loss[loss=0.2525, simple_loss=0.315, pruned_loss=0.09502, over 28653.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3428, pruned_loss=0.09708, over 5711409.27 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3487, pruned_loss=0.09759, over 5766441.42 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3414, pruned_loss=0.09563, over 5714991.52 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:47:30,845 INFO [train.py:968] (0/2) Epoch 28, batch 22950, giga_loss[loss=0.265, simple_loss=0.335, pruned_loss=0.09757, over 28847.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3407, pruned_loss=0.09683, over 5718857.20 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3483, pruned_loss=0.09771, over 5770093.79 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3396, pruned_loss=0.0955, over 5716924.09 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:47:51,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.150e+02 1.389e+03 1.758e+03 2.402e+03 8.093e+03, threshold=3.516e+03, percent-clipped=7.0
+2023-03-14 12:48:10,280 INFO [train.py:968] (0/2) Epoch 28, batch 23000, giga_loss[loss=0.2615, simple_loss=0.3346, pruned_loss=0.09423, over 28999.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3389, pruned_loss=0.09637, over 5723295.25 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3484, pruned_loss=0.09804, over 5771642.75 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3378, pruned_loss=0.09501, over 5719480.60 frames. ], batch size: 120, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:48:15,215 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-14 12:48:32,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5295, 2.2476, 1.7177, 0.9148], device='cuda:0'), covar=tensor([0.7652, 0.3397, 0.4564, 0.7478], device='cuda:0'), in_proj_covar=tensor([0.1828, 0.1714, 0.1649, 0.1490], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 12:48:48,999 INFO [train.py:968] (0/2) Epoch 28, batch 23050, giga_loss[loss=0.2791, simple_loss=0.334, pruned_loss=0.1122, over 28607.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3368, pruned_loss=0.0955, over 5711637.58 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3485, pruned_loss=0.09818, over 5763609.46 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3356, pruned_loss=0.09421, over 5713683.29 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:49:07,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.852e+02 1.413e+03 1.724e+03 2.668e+03 5.011e+03, threshold=3.448e+03, percent-clipped=10.0
+2023-03-14 12:49:12,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8511, 1.1317, 1.0329, 0.8117], device='cuda:0'), covar=tensor([0.2723, 0.2603, 0.1775, 0.2452], device='cuda:0'), in_proj_covar=tensor([0.2046, 0.2007, 0.1908, 0.2056], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 12:49:24,955 INFO [train.py:968] (0/2) Epoch 28, batch 23100, giga_loss[loss=0.2592, simple_loss=0.3335, pruned_loss=0.0924, over 28217.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3339, pruned_loss=0.09426, over 5713100.45 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3486, pruned_loss=0.09853, over 5758926.58 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3323, pruned_loss=0.09279, over 5717930.89 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:49:25,674 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1254000.pt
+2023-03-14 12:49:31,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5095, 1.2664, 4.5282, 3.4716], device='cuda:0'), covar=tensor([0.1668, 0.2946, 0.0400, 0.1042], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0671, 0.1002, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 12:49:45,543 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1254023.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 12:50:07,090 INFO [train.py:968] (0/2) Epoch 28, batch 23150, giga_loss[loss=0.2132, simple_loss=0.296, pruned_loss=0.06523, over 29014.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3297, pruned_loss=0.09205, over 5711437.43 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.349, pruned_loss=0.0989, over 5756611.30 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3278, pruned_loss=0.09045, over 5716487.46 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:50:27,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.396e+02 1.306e+03 1.710e+03 2.417e+03 8.216e+03, threshold=3.421e+03, percent-clipped=6.0
+2023-03-14 12:50:39,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6409, 2.0676, 1.6748, 1.6767], device='cuda:0'), covar=tensor([0.2651, 0.2669, 0.3008, 0.2438], device='cuda:0'), in_proj_covar=tensor([0.1598, 0.1151, 0.1409, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 12:50:44,622 INFO [train.py:968] (0/2) Epoch 28, batch 23200, giga_loss[loss=0.2758, simple_loss=0.3439, pruned_loss=0.1039, over 29059.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3291, pruned_loss=0.09124, over 5719416.28 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3489, pruned_loss=0.099, over 5760826.78 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3271, pruned_loss=0.08967, over 5718567.58 frames. ], batch size: 113, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:50:56,885 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1254112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:51:02,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5132, 1.8646, 1.8263, 1.5737], device='cuda:0'), covar=tensor([0.1947, 0.1774, 0.1961, 0.1929], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0757, 0.0728, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 12:51:27,205 INFO [train.py:968] (0/2) Epoch 28, batch 23250, giga_loss[loss=0.2879, simple_loss=0.3654, pruned_loss=0.1052, over 27973.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.333, pruned_loss=0.09312, over 5708832.77 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3495, pruned_loss=0.09955, over 5755286.93 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3305, pruned_loss=0.09117, over 5711128.52 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:51:49,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.953e+02 1.303e+03 1.688e+03 2.239e+03 4.291e+03, threshold=3.376e+03, percent-clipped=3.0
+2023-03-14 12:52:08,250 INFO [train.py:968] (0/2) Epoch 28, batch 23300, giga_loss[loss=0.2397, simple_loss=0.3193, pruned_loss=0.0801, over 28619.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.336, pruned_loss=0.09419, over 5702859.90 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3496, pruned_loss=0.09983, over 5748542.65 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3336, pruned_loss=0.09226, over 5709422.92 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:52:42,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1254242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:52:49,725 INFO [train.py:968] (0/2) Epoch 28, batch 23350, giga_loss[loss=0.2588, simple_loss=0.3342, pruned_loss=0.0917, over 28907.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3407, pruned_loss=0.09649, over 5704939.47 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3507, pruned_loss=0.1008, over 5749423.37 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3374, pruned_loss=0.09389, over 5708302.03 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:52:54,653 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1254255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:52:56,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1254258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:53:10,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.951e+02 1.337e+03 1.640e+03 2.344e+03 6.265e+03, threshold=3.279e+03, percent-clipped=8.0
+2023-03-14 12:53:19,588 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1254287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:53:27,538 INFO [train.py:968] (0/2) Epoch 28, batch 23400, giga_loss[loss=0.2408, simple_loss=0.3271, pruned_loss=0.07723, over 28932.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3431, pruned_loss=0.09696, over 5713570.35 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3512, pruned_loss=0.1013, over 5751040.75 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3397, pruned_loss=0.0943, over 5713800.43 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:54:10,805 INFO [train.py:968] (0/2) Epoch 28, batch 23450, giga_loss[loss=0.2838, simple_loss=0.3528, pruned_loss=0.1074, over 28901.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3455, pruned_loss=0.09829, over 5719019.62 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3514, pruned_loss=0.1015, over 5753134.42 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3425, pruned_loss=0.09595, over 5716861.20 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:54:24,947 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3936, 1.6022, 1.5921, 1.4111], device='cuda:0'), covar=tensor([0.2138, 0.2027, 0.2438, 0.2189], device='cuda:0'), in_proj_covar=tensor([0.0505, 0.0758, 0.0730, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 12:54:38,232 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.731e+02 1.507e+03 1.916e+03 2.641e+03 5.884e+03, threshold=3.832e+03, percent-clipped=16.0
+2023-03-14 12:54:57,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1254398.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 12:54:58,448 INFO [train.py:968] (0/2) Epoch 28, batch 23500, libri_loss[loss=0.2982, simple_loss=0.3698, pruned_loss=0.1133, over 27845.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3512, pruned_loss=0.1032, over 5693596.43 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3519, pruned_loss=0.102, over 5741803.42 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3482, pruned_loss=0.1008, over 5701551.60 frames. ], batch size: 116, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:55:45,460 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.44 vs. limit=5.0
+2023-03-14 12:55:49,319 INFO [train.py:968] (0/2) Epoch 28, batch 23550, libri_loss[loss=0.2803, simple_loss=0.3451, pruned_loss=0.1077, over 29557.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3568, pruned_loss=0.1075, over 5685685.96 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3522, pruned_loss=0.1023, over 5743625.49 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3542, pruned_loss=0.1054, over 5689776.67 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:56:15,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.795e+03 2.320e+03 3.172e+03 7.077e+03, threshold=4.641e+03, percent-clipped=15.0
+2023-03-14 12:56:39,257 INFO [train.py:968] (0/2) Epoch 28, batch 23600, giga_loss[loss=0.3045, simple_loss=0.3747, pruned_loss=0.1171, over 28842.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3639, pruned_loss=0.1128, over 5684902.40 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3528, pruned_loss=0.1028, over 5745942.82 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3615, pruned_loss=0.1108, over 5684891.01 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:57:07,112 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1254531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:57:15,824 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1254541.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 12:57:19,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1254544.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 12:57:23,506 INFO [train.py:968] (0/2) Epoch 28, batch 23650, giga_loss[loss=0.3455, simple_loss=0.3816, pruned_loss=0.1547, over 23438.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3689, pruned_loss=0.117, over 5676931.91 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3533, pruned_loss=0.1032, over 5739817.46 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.367, pruned_loss=0.1154, over 5679624.21 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:57:44,178 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3871, 1.2022, 3.8927, 3.2813], device='cuda:0'), covar=tensor([0.1666, 0.2881, 0.0486, 0.1081], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0673, 0.1007, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 12:57:44,864 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1254573.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 12:57:50,194 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.214e+03 1.862e+03 2.466e+03 3.338e+03 9.004e+03, threshold=4.932e+03, percent-clipped=7.0
+2023-03-14 12:58:00,811 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-14 12:58:10,499 INFO [train.py:968] (0/2) Epoch 28, batch 23700, giga_loss[loss=0.3846, simple_loss=0.4247, pruned_loss=0.1722, over 28969.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.375, pruned_loss=0.1219, over 5677932.99 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3537, pruned_loss=0.1035, over 5738188.56 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3735, pruned_loss=0.1209, over 5679641.31 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:58:28,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4063, 3.5299, 1.5246, 1.5620], device='cuda:0'), covar=tensor([0.0936, 0.0389, 0.0886, 0.1238], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0570, 0.0411, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 12:58:28,561 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1254617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:58:59,948 INFO [train.py:968] (0/2) Epoch 28, batch 23750, giga_loss[loss=0.2704, simple_loss=0.3375, pruned_loss=0.1017, over 28737.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3778, pruned_loss=0.1247, over 5672195.78 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3536, pruned_loss=0.1035, over 5738246.09 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3771, pruned_loss=0.1241, over 5672945.43 frames. ], batch size: 66, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:59:15,709 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 12:59:29,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.823e+03 2.316e+03 3.562e+03 7.800e+03, threshold=4.633e+03, percent-clipped=10.0
+2023-03-14 12:59:49,404 INFO [train.py:968] (0/2) Epoch 28, batch 23800, giga_loss[loss=0.4302, simple_loss=0.4414, pruned_loss=0.2095, over 26578.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3807, pruned_loss=0.1285, over 5660442.63 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3539, pruned_loss=0.1038, over 5739688.14 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3802, pruned_loss=0.1281, over 5658871.14 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:00:41,678 INFO [train.py:968] (0/2) Epoch 28, batch 23850, giga_loss[loss=0.3898, simple_loss=0.429, pruned_loss=0.1753, over 27854.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3857, pruned_loss=0.1338, over 5643768.56 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3543, pruned_loss=0.1041, over 5730435.82 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3853, pruned_loss=0.1335, over 5650218.22 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:00:53,915 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1254760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:00:55,955 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1254763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:01:09,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.231e+03 1.807e+03 2.555e+03 3.740e+03 8.284e+03, threshold=5.109e+03, percent-clipped=13.0
+2023-03-14 13:01:26,306 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1254792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:01:33,526 INFO [train.py:968] (0/2) Epoch 28, batch 23900, giga_loss[loss=0.4257, simple_loss=0.4449, pruned_loss=0.2032, over 26575.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3891, pruned_loss=0.1377, over 5639781.65 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3543, pruned_loss=0.1044, over 5734733.30 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3895, pruned_loss=0.138, over 5639205.88 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:01:34,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7275, 4.8094, 1.8518, 1.8045], device='cuda:0'), covar=tensor([0.0979, 0.0288, 0.0846, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0571, 0.0411, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 13:02:18,953 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2181, 1.5222, 1.4728, 1.3222], device='cuda:0'), covar=tensor([0.1932, 0.1629, 0.2233, 0.1858], device='cuda:0'), in_proj_covar=tensor([0.0506, 0.0760, 0.0732, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 13:02:29,429 INFO [train.py:968] (0/2) Epoch 28, batch 23950, giga_loss[loss=0.319, simple_loss=0.3803, pruned_loss=0.1288, over 28874.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3895, pruned_loss=0.1374, over 5639277.11 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3545, pruned_loss=0.1047, over 5728342.61 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3903, pruned_loss=0.1379, over 5642918.67 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:02:40,356 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2255, 1.5776, 1.4317, 1.3258], device='cuda:0'), covar=tensor([0.2085, 0.1688, 0.2319, 0.1950], device='cuda:0'), in_proj_covar=tensor([0.0506, 0.0760, 0.0732, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 13:03:02,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.249e+03 1.977e+03 2.635e+03 3.462e+03 1.177e+04, threshold=5.270e+03, percent-clipped=9.0
+2023-03-14 13:03:08,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6466, 1.5704, 1.8068, 1.4527], device='cuda:0'), covar=tensor([0.1584, 0.2356, 0.1315, 0.1625], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0713, 0.0976, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 13:03:12,896 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.85 vs. limit=2.0
+2023-03-14 13:03:22,315 INFO [train.py:968] (0/2) Epoch 28, batch 24000, giga_loss[loss=0.2954, simple_loss=0.355, pruned_loss=0.1179, over 28498.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3885, pruned_loss=0.1376, over 5630759.23 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3544, pruned_loss=0.1047, over 5731584.37 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3899, pruned_loss=0.1387, over 5628944.50 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:03:22,319 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 13:03:29,827 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4076, 1.7735, 1.6762, 1.2046], device='cuda:0'), covar=tensor([0.1971, 0.3093, 0.1756, 0.2066], device='cuda:0'), in_proj_covar=tensor([0.0927, 0.0713, 0.0976, 0.0877], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 13:03:30,591 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2029, simple_loss=0.3112, pruned_loss=0.04725, over 944034.00 frames. 
+2023-03-14 13:03:30,591 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 13:03:36,409 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1254906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:04:15,862 INFO [train.py:968] (0/2) Epoch 28, batch 24050, giga_loss[loss=0.2846, simple_loss=0.3571, pruned_loss=0.1061, over 28784.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3869, pruned_loss=0.1367, over 5627411.31 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3547, pruned_loss=0.105, over 5715706.96 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3884, pruned_loss=0.1378, over 5639641.12 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:04:44,760 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 1.907e+03 2.390e+03 3.123e+03 6.631e+03, threshold=4.780e+03, percent-clipped=3.0
+2023-03-14 13:05:02,357 INFO [train.py:968] (0/2) Epoch 28, batch 24100, giga_loss[loss=0.3092, simple_loss=0.3822, pruned_loss=0.1181, over 28866.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3859, pruned_loss=0.1343, over 5638784.76 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3549, pruned_loss=0.1051, over 5717921.95 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3872, pruned_loss=0.1355, over 5645229.34 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:05:56,148 INFO [train.py:968] (0/2) Epoch 28, batch 24150, giga_loss[loss=0.3722, simple_loss=0.4055, pruned_loss=0.1695, over 26668.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3873, pruned_loss=0.1356, over 5623636.45 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3547, pruned_loss=0.105, over 5718476.10 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3889, pruned_loss=0.137, over 5627556.67 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:05:56,378 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1255049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:05:59,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1255052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:06:25,938 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1255076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:06:27,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.209e+03 1.813e+03 2.342e+03 3.588e+03 8.325e+03, threshold=4.684e+03, percent-clipped=9.0
+2023-03-14 13:06:29,432 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1255081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:06:51,605 INFO [train.py:968] (0/2) Epoch 28, batch 24200, giga_loss[loss=0.2974, simple_loss=0.3632, pruned_loss=0.1158, over 28564.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3879, pruned_loss=0.1358, over 5620700.14 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 5720540.64 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.39, pruned_loss=0.1375, over 5619693.48 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:06:59,269 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5786, 1.8145, 1.5759, 1.4068], device='cuda:0'), covar=tensor([0.1942, 0.1866, 0.1967, 0.1805], device='cuda:0'), in_proj_covar=tensor([0.1598, 0.1152, 0.1411, 0.1011], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 13:07:41,830 INFO [train.py:968] (0/2) Epoch 28, batch 24250, giga_loss[loss=0.2878, simple_loss=0.3647, pruned_loss=0.1055, over 28975.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3834, pruned_loss=0.1313, over 5621910.91 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 5714232.78 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3857, pruned_loss=0.1332, over 5625249.54 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:08:10,559 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4417, 4.3011, 4.0683, 2.1422], device='cuda:0'), covar=tensor([0.0642, 0.0780, 0.0852, 0.1907], device='cuda:0'), in_proj_covar=tensor([0.1307, 0.1206, 0.1017, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 13:08:10,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.732e+02 1.691e+03 2.149e+03 2.983e+03 7.276e+03, threshold=4.297e+03, percent-clipped=4.0
+2023-03-14 13:08:31,318 INFO [train.py:968] (0/2) Epoch 28, batch 24300, giga_loss[loss=0.3009, simple_loss=0.3746, pruned_loss=0.1136, over 28882.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3809, pruned_loss=0.1277, over 5629433.92 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 5712245.49 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3829, pruned_loss=0.1294, over 5633374.78 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:09:16,698 INFO [train.py:968] (0/2) Epoch 28, batch 24350, libri_loss[loss=0.3006, simple_loss=0.3701, pruned_loss=0.1156, over 29113.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3771, pruned_loss=0.1244, over 5652901.48 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3549, pruned_loss=0.1054, over 5716805.48 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3792, pruned_loss=0.1261, over 5649660.29 frames. ], batch size: 101, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:09:47,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.111e+03 1.690e+03 2.255e+03 2.918e+03 8.193e+03, threshold=4.509e+03, percent-clipped=7.0
+2023-03-14 13:09:47,630 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1255279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:09:51,756 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1255283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:10:08,654 INFO [train.py:968] (0/2) Epoch 28, batch 24400, giga_loss[loss=0.2959, simple_loss=0.3594, pruned_loss=0.1162, over 28742.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3745, pruned_loss=0.1224, over 5652109.32 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3549, pruned_loss=0.1055, over 5715842.46 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3763, pruned_loss=0.1237, over 5650201.00 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:10:15,950 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-14 13:10:17,272 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3806, 1.6638, 1.5581, 1.4834], device='cuda:0'), covar=tensor([0.1960, 0.2051, 0.2270, 0.2108], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0760, 0.0731, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 13:10:42,008 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-14 13:10:54,622 INFO [train.py:968] (0/2) Epoch 28, batch 24450, giga_loss[loss=0.2734, simple_loss=0.3531, pruned_loss=0.09683, over 28863.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3724, pruned_loss=0.1211, over 5666424.22 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3552, pruned_loss=0.106, over 5721078.31 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.374, pruned_loss=0.1221, over 5658294.36 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:11:26,597 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.274e+03 2.051e+03 2.828e+03 4.190e+03 1.047e+04, threshold=5.657e+03, percent-clipped=19.0
+2023-03-14 13:11:50,076 INFO [train.py:968] (0/2) Epoch 28, batch 24500, giga_loss[loss=0.3749, simple_loss=0.4228, pruned_loss=0.1635, over 28331.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.373, pruned_loss=0.1214, over 5665610.51 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3552, pruned_loss=0.106, over 5722856.26 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3744, pruned_loss=0.1223, over 5657321.83 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:11:51,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-14 13:12:01,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5112, 1.5908, 1.7305, 1.3050], device='cuda:0'), covar=tensor([0.1915, 0.2756, 0.1607, 0.1860], device='cuda:0'), in_proj_covar=tensor([0.0926, 0.0712, 0.0974, 0.0874], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 13:12:12,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1747, 1.4169, 1.3164, 1.0998], device='cuda:0'), covar=tensor([0.3108, 0.2980, 0.2073, 0.2981], device='cuda:0'), in_proj_covar=tensor([0.2058, 0.2018, 0.1921, 0.2071], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 13:12:41,261 INFO [train.py:968] (0/2) Epoch 28, batch 24550, giga_loss[loss=0.2856, simple_loss=0.3568, pruned_loss=0.1072, over 28548.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3714, pruned_loss=0.1196, over 5659607.68 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3557, pruned_loss=0.1064, over 5714759.04 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3723, pruned_loss=0.1202, over 5659494.57 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:12:44,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1255451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:12:55,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4644, 1.8010, 1.2322, 1.4093], device='cuda:0'), covar=tensor([0.1153, 0.0618, 0.1113, 0.1226], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0450, 0.0522, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 13:13:13,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.560e+03 2.080e+03 2.691e+03 6.208e+03, threshold=4.160e+03, percent-clipped=2.0
+2023-03-14 13:13:35,679 INFO [train.py:968] (0/2) Epoch 28, batch 24600, giga_loss[loss=0.2866, simple_loss=0.3686, pruned_loss=0.1022, over 29138.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3717, pruned_loss=0.1172, over 5672459.21 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3556, pruned_loss=0.1065, over 5713111.69 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3727, pruned_loss=0.1178, over 5673500.45 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:14:27,872 INFO [train.py:968] (0/2) Epoch 28, batch 24650, giga_loss[loss=0.3071, simple_loss=0.3527, pruned_loss=0.1308, over 23597.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.373, pruned_loss=0.1177, over 5647274.09 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3559, pruned_loss=0.1067, over 5708455.80 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3741, pruned_loss=0.1183, over 5649838.57 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:14:55,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.191e+03 1.954e+03 2.847e+03 3.739e+03 1.005e+04, threshold=5.694e+03, percent-clipped=19.0
+2023-03-14 13:15:12,643 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1255594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:15:15,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1255597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:15:16,097 INFO [train.py:968] (0/2) Epoch 28, batch 24700, giga_loss[loss=0.3041, simple_loss=0.3712, pruned_loss=0.1185, over 28415.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3731, pruned_loss=0.1177, over 5658270.93 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3562, pruned_loss=0.1069, over 5709505.74 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3739, pruned_loss=0.1182, over 5658311.08 frames. ], batch size: 78, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:15:40,844 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1255626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:16:02,514 INFO [train.py:968] (0/2) Epoch 28, batch 24750, giga_loss[loss=0.3383, simple_loss=0.3883, pruned_loss=0.1441, over 27986.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3735, pruned_loss=0.1189, over 5652648.22 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3561, pruned_loss=0.107, over 5706487.45 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3748, pruned_loss=0.1195, over 5653214.96 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:16:06,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1255654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:16:13,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1255658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:16:34,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.139e+03 1.908e+03 2.556e+03 3.621e+03 1.006e+04, threshold=5.111e+03, percent-clipped=10.0
+2023-03-14 13:16:53,431 INFO [train.py:968] (0/2) Epoch 28, batch 24800, giga_loss[loss=0.3136, simple_loss=0.3832, pruned_loss=0.1221, over 29063.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3727, pruned_loss=0.1199, over 5646594.74 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3563, pruned_loss=0.1072, over 5705765.01 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3736, pruned_loss=0.1203, over 5647397.81 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:17:25,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3877, 1.5832, 1.3761, 1.5519], device='cuda:0'), covar=tensor([0.0799, 0.0355, 0.0345, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 13:17:35,191 INFO [train.py:968] (0/2) Epoch 28, batch 24850, giga_loss[loss=0.3075, simple_loss=0.3724, pruned_loss=0.1213, over 29038.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3713, pruned_loss=0.1195, over 5656092.07 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3569, pruned_loss=0.1076, over 5699043.43 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3719, pruned_loss=0.1197, over 5660995.55 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:18:01,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.148e+03 1.748e+03 2.293e+03 2.890e+03 7.141e+03, threshold=4.586e+03, percent-clipped=1.0
+2023-03-14 13:18:17,624 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1255797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:18:18,833 INFO [train.py:968] (0/2) Epoch 28, batch 24900, giga_loss[loss=0.2885, simple_loss=0.3614, pruned_loss=0.1078, over 28681.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3696, pruned_loss=0.1182, over 5664268.46 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3566, pruned_loss=0.1075, over 5704846.12 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3707, pruned_loss=0.1187, over 5661840.97 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:18:20,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1255800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:18:21,122 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1255801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:18:23,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1255804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:18:30,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4189, 1.6849, 1.3431, 1.2465], device='cuda:0'), covar=tensor([0.1013, 0.0483, 0.0971, 0.1015], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0453, 0.0525, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 13:18:44,392 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1255829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:18:47,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1255833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:19:01,717 INFO [train.py:968] (0/2) Epoch 28, batch 24950, giga_loss[loss=0.3432, simple_loss=0.3936, pruned_loss=0.1464, over 27603.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3688, pruned_loss=0.1166, over 5678624.58 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3567, pruned_loss=0.1075, over 5711784.14 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.37, pruned_loss=0.1174, over 5669342.82 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:19:04,962 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1255852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:19:15,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1087, 1.4664, 1.5197, 1.1890], device='cuda:0'), covar=tensor([0.2380, 0.2008, 0.2744, 0.2536], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0765, 0.0736, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-14 13:19:31,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.550e+03 1.900e+03 2.564e+03 5.443e+03, threshold=3.799e+03, percent-clipped=3.0
+2023-03-14 13:19:49,629 INFO [train.py:968] (0/2) Epoch 28, batch 25000, giga_loss[loss=0.2971, simple_loss=0.3694, pruned_loss=0.1125, over 29030.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3693, pruned_loss=0.1169, over 5674397.13 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3571, pruned_loss=0.1078, over 5716299.35 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3703, pruned_loss=0.1175, over 5661594.94 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:20:33,535 INFO [train.py:968] (0/2) Epoch 28, batch 25050, giga_loss[loss=0.3727, simple_loss=0.4054, pruned_loss=0.17, over 26505.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3694, pruned_loss=0.1172, over 5682069.63 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.357, pruned_loss=0.1078, over 5722527.93 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3708, pruned_loss=0.1181, over 5664592.58 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:21:03,227 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.291e+03 1.841e+03 2.413e+03 3.367e+03 7.128e+03, threshold=4.826e+03, percent-clipped=17.0
+2023-03-14 13:21:21,973 INFO [train.py:968] (0/2) Epoch 28, batch 25100, giga_loss[loss=0.2602, simple_loss=0.3324, pruned_loss=0.09403, over 28845.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3668, pruned_loss=0.1158, over 5696256.87 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3564, pruned_loss=0.1075, over 5726093.30 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3685, pruned_loss=0.1169, over 5678607.42 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:21:22,630 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1256000.pt
+2023-03-14 13:21:50,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2624, 1.5492, 1.5730, 1.1382], device='cuda:0'), covar=tensor([0.1729, 0.2550, 0.1409, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.0930, 0.0714, 0.0978, 0.0878], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 13:22:03,963 INFO [train.py:968] (0/2) Epoch 28, batch 25150, giga_loss[loss=0.2626, simple_loss=0.3384, pruned_loss=0.09336, over 29004.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3667, pruned_loss=0.1165, over 5697386.60 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3565, pruned_loss=0.1077, over 5734521.27 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3686, pruned_loss=0.1176, over 5673599.57 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:22:24,274 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:22:36,095 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.054e+03 1.883e+03 2.400e+03 3.267e+03 9.305e+03, threshold=4.800e+03, percent-clipped=10.0
+2023-03-14 13:22:52,377 INFO [train.py:968] (0/2) Epoch 28, batch 25200, giga_loss[loss=0.2701, simple_loss=0.3299, pruned_loss=0.1052, over 28252.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3652, pruned_loss=0.116, over 5704073.79 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3565, pruned_loss=0.1077, over 5734521.27 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3666, pruned_loss=0.1169, over 5685560.04 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:23:40,153 INFO [train.py:968] (0/2) Epoch 28, batch 25250, giga_loss[loss=0.2702, simple_loss=0.3384, pruned_loss=0.101, over 28990.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3627, pruned_loss=0.1145, over 5706440.00 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3569, pruned_loss=0.1079, over 5737914.60 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3636, pruned_loss=0.1151, over 5688283.55 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:23:51,361 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5388, 4.3855, 4.1786, 2.1438], device='cuda:0'), covar=tensor([0.0660, 0.0748, 0.0808, 0.1792], device='cuda:0'), in_proj_covar=tensor([0.1313, 0.1211, 0.1021, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 13:23:54,805 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256164.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:24:09,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.708e+02 1.757e+03 2.550e+03 3.309e+03 7.379e+03, threshold=5.099e+03, percent-clipped=5.0
+2023-03-14 13:24:14,392 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:24:26,918 INFO [train.py:968] (0/2) Epoch 28, batch 25300, giga_loss[loss=0.2924, simple_loss=0.3479, pruned_loss=0.1184, over 28996.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.362, pruned_loss=0.1146, over 5696590.82 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3571, pruned_loss=0.108, over 5736796.13 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3627, pruned_loss=0.1151, over 5682776.92 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:24:56,057 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1256227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:25:12,942 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5253, 1.6531, 1.7296, 1.3070], device='cuda:0'), covar=tensor([0.1666, 0.2795, 0.1519, 0.1833], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0715, 0.0979, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 13:25:15,794 INFO [train.py:968] (0/2) Epoch 28, batch 25350, giga_loss[loss=0.2859, simple_loss=0.3681, pruned_loss=0.1018, over 28877.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1156, over 5688595.93 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3574, pruned_loss=0.1082, over 5736916.20 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.1161, over 5676354.36 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:25:44,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.907e+03 2.311e+03 3.368e+03 8.812e+03, threshold=4.621e+03, percent-clipped=5.0
+2023-03-14 13:25:59,171 INFO [train.py:968] (0/2) Epoch 28, batch 25400, giga_loss[loss=0.2707, simple_loss=0.3556, pruned_loss=0.0929, over 28951.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3622, pruned_loss=0.1137, over 5696878.36 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3569, pruned_loss=0.1079, over 5739416.04 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3631, pruned_loss=0.1145, over 5683937.91 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:26:40,646 INFO [train.py:968] (0/2) Epoch 28, batch 25450, libri_loss[loss=0.3036, simple_loss=0.3656, pruned_loss=0.1208, over 19280.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3623, pruned_loss=0.1137, over 5684590.00 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3566, pruned_loss=0.108, over 5736283.65 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3635, pruned_loss=0.1145, over 5675708.45 frames. ], batch size: 187, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:26:59,063 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1256370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:27:00,883 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1256373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:27:08,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.727e+03 2.102e+03 2.993e+03 1.076e+04, threshold=4.204e+03, percent-clipped=5.0
+2023-03-14 13:27:11,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:27:18,916 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-14 13:27:25,283 INFO [train.py:968] (0/2) Epoch 28, batch 25500, giga_loss[loss=0.3067, simple_loss=0.3814, pruned_loss=0.116, over 28982.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3633, pruned_loss=0.1139, over 5679062.60 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3565, pruned_loss=0.1081, over 5728578.05 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3645, pruned_loss=0.1146, over 5677967.02 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:27:29,461 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1256402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:27:44,241 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-14 13:28:07,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1256445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:28:10,882 INFO [train.py:968] (0/2) Epoch 28, batch 25550, giga_loss[loss=0.435, simple_loss=0.4687, pruned_loss=0.2006, over 26735.00 frames. ], tot_loss[loss=0.298, simple_loss=0.365, pruned_loss=0.1155, over 5680727.28 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3568, pruned_loss=0.1085, over 5731672.25 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3658, pruned_loss=0.1158, over 5676318.95 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:28:24,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3211, 1.7998, 1.7166, 1.5390], device='cuda:0'), covar=tensor([0.0784, 0.0309, 0.0295, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 13:28:38,478 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3339, 3.0768, 1.4195, 1.4671], device='cuda:0'), covar=tensor([0.0993, 0.0567, 0.0911, 0.1416], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0574, 0.0411, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 13:28:39,532 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-14 13:28:44,460 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.910e+03 2.450e+03 3.448e+03 1.184e+04, threshold=4.900e+03, percent-clipped=15.0
+2023-03-14 13:28:58,973 INFO [train.py:968] (0/2) Epoch 28, batch 25600, giga_loss[loss=0.3113, simple_loss=0.3786, pruned_loss=0.122, over 28867.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3687, pruned_loss=0.1189, over 5682917.73 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3576, pruned_loss=0.1091, over 5732551.20 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3688, pruned_loss=0.1187, over 5677729.67 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:29:13,522 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 13:29:39,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1256539.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:29:49,419 INFO [train.py:968] (0/2) Epoch 28, batch 25650, giga_loss[loss=0.305, simple_loss=0.372, pruned_loss=0.119, over 28961.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3713, pruned_loss=0.1225, over 5665079.65 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3578, pruned_loss=0.1093, over 5714281.16 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3713, pruned_loss=0.1223, over 5675845.64 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:30:01,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1256562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:30:21,136 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.317e+03 2.063e+03 2.969e+03 4.272e+03 8.740e+03, threshold=5.937e+03, percent-clipped=18.0
+2023-03-14 13:30:26,626 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1256588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:30:28,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1256591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:30:37,468 INFO [train.py:968] (0/2) Epoch 28, batch 25700, giga_loss[loss=0.2826, simple_loss=0.3482, pruned_loss=0.1085, over 28919.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3713, pruned_loss=0.1237, over 5664019.58 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3575, pruned_loss=0.1092, over 5711610.53 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3721, pruned_loss=0.1241, over 5673559.43 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:30:49,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 13:30:58,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1256620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:31:23,574 INFO [train.py:968] (0/2) Epoch 28, batch 25750, giga_loss[loss=0.3171, simple_loss=0.3744, pruned_loss=0.1299, over 27524.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3718, pruned_loss=0.1242, over 5674478.21 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3579, pruned_loss=0.1096, over 5714666.97 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3722, pruned_loss=0.1244, over 5678698.35 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:31:54,843 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1256682.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:31:55,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.620e+02 1.803e+03 2.185e+03 2.996e+03 6.514e+03, threshold=4.369e+03, percent-clipped=2.0
+2023-03-14 13:31:57,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1256685.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:32:07,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3214, 1.3219, 1.2530, 1.5456], device='cuda:0'), covar=tensor([0.0745, 0.0400, 0.0349, 0.0806], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 13:32:11,523 INFO [train.py:968] (0/2) Epoch 28, batch 25800, giga_loss[loss=0.2737, simple_loss=0.3487, pruned_loss=0.09934, over 28857.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3709, pruned_loss=0.1237, over 5666830.31 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3582, pruned_loss=0.1098, over 5717157.15 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3712, pruned_loss=0.1239, over 5667125.35 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:32:16,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1256705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:32:19,020 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1256708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:32:23,876 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1256714.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:32:27,275 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256716.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:32:42,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1256737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:32:53,043 INFO [train.py:968] (0/2) Epoch 28, batch 25850, giga_loss[loss=0.27, simple_loss=0.355, pruned_loss=0.0925, over 28884.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3706, pruned_loss=0.1214, over 5677775.76 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3582, pruned_loss=0.1097, over 5721082.61 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3711, pruned_loss=0.1218, over 5673736.24 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:32:53,369 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3005, 1.7203, 1.5960, 1.4691], device='cuda:0'), covar=tensor([0.2324, 0.1899, 0.2485, 0.2167], device='cuda:0'), in_proj_covar=tensor([0.0506, 0.0762, 0.0734, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 13:33:04,998 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1256760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:33:25,151 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.873e+03 2.531e+03 3.348e+03 1.283e+04, threshold=5.062e+03, percent-clipped=13.0
+2023-03-14 13:33:41,819 INFO [train.py:968] (0/2) Epoch 28, batch 25900, giga_loss[loss=0.3003, simple_loss=0.3708, pruned_loss=0.1149, over 28484.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3693, pruned_loss=0.1205, over 5654628.17 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3584, pruned_loss=0.1099, over 5715059.65 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3698, pruned_loss=0.121, over 5654786.98 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:34:16,072 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:34:26,099 INFO [train.py:968] (0/2) Epoch 28, batch 25950, giga_loss[loss=0.296, simple_loss=0.354, pruned_loss=0.119, over 29036.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3659, pruned_loss=0.1183, over 5656448.12 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3585, pruned_loss=0.1101, over 5709098.85 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3663, pruned_loss=0.1186, over 5660614.28 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:34:57,753 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.174e+03 1.593e+03 2.085e+03 3.126e+03 7.320e+03, threshold=4.170e+03, percent-clipped=8.0
+2023-03-14 13:35:12,924 INFO [train.py:968] (0/2) Epoch 28, batch 26000, giga_loss[loss=0.3141, simple_loss=0.3673, pruned_loss=0.1305, over 28679.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3669, pruned_loss=0.1202, over 5652245.61 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3589, pruned_loss=0.1105, over 5709225.74 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.367, pruned_loss=0.1203, over 5654189.21 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:35:18,654 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1256903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:35:22,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1256906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:35:52,540 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1256935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:36:03,827 INFO [train.py:968] (0/2) Epoch 28, batch 26050, giga_loss[loss=0.2941, simple_loss=0.3596, pruned_loss=0.1143, over 28792.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3677, pruned_loss=0.1208, over 5660289.02 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3587, pruned_loss=0.1104, over 5714287.15 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3682, pruned_loss=0.1211, over 5655900.20 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:36:34,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.798e+03 2.105e+03 3.063e+03 9.293e+03, threshold=4.210e+03, percent-clipped=9.0
+2023-03-14 13:36:39,826 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5888, 2.0611, 1.3114, 0.9099], device='cuda:0'), covar=tensor([0.7509, 0.4411, 0.3415, 0.7200], device='cuda:0'), in_proj_covar=tensor([0.1854, 0.1751, 0.1666, 0.1509], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 13:36:46,327 INFO [train.py:968] (0/2) Epoch 28, batch 26100, giga_loss[loss=0.3165, simple_loss=0.3856, pruned_loss=0.1237, over 28638.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3707, pruned_loss=0.1215, over 5660985.54 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3586, pruned_loss=0.1104, over 5711792.72 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3716, pruned_loss=0.1223, over 5657398.37 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:37:32,327 INFO [train.py:968] (0/2) Epoch 28, batch 26150, giga_loss[loss=0.3373, simple_loss=0.396, pruned_loss=0.1392, over 28944.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3738, pruned_loss=0.1213, over 5656892.29 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3589, pruned_loss=0.1108, over 5705248.75 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3747, pruned_loss=0.1219, over 5658721.83 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:38:04,673 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.139e+03 1.660e+03 2.212e+03 3.026e+03 8.752e+03, threshold=4.424e+03, percent-clipped=11.0
+2023-03-14 13:38:13,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1257091.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:38:19,811 INFO [train.py:968] (0/2) Epoch 28, batch 26200, giga_loss[loss=0.2896, simple_loss=0.3602, pruned_loss=0.1095, over 28907.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3742, pruned_loss=0.121, over 5661581.58 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3585, pruned_loss=0.1107, over 5710091.01 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3757, pruned_loss=0.1218, over 5657526.52 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:38:24,653 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1257106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:39:01,325 INFO [train.py:968] (0/2) Epoch 28, batch 26250, giga_loss[loss=0.261, simple_loss=0.335, pruned_loss=0.09351, over 28783.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3761, pruned_loss=0.1227, over 5652394.39 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3584, pruned_loss=0.1107, over 5708071.44 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3781, pruned_loss=0.1238, over 5648543.43 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:39:31,775 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 13:39:34,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.259e+03 1.904e+03 2.636e+03 3.348e+03 6.602e+03, threshold=5.273e+03, percent-clipped=10.0
+2023-03-14 13:39:47,535 INFO [train.py:968] (0/2) Epoch 28, batch 26300, giga_loss[loss=0.2871, simple_loss=0.3597, pruned_loss=0.1072, over 29052.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3771, pruned_loss=0.124, over 5663091.23 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3587, pruned_loss=0.111, over 5709106.08 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3785, pruned_loss=0.1247, over 5658981.89 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:40:03,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1257213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:40:21,372 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4197, 1.6968, 1.6198, 1.4912], device='cuda:0'), covar=tensor([0.2157, 0.1959, 0.2444, 0.2124], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0758, 0.0731, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 13:40:23,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3662, 1.3737, 3.7401, 3.2089], device='cuda:0'), covar=tensor([0.1873, 0.2768, 0.0959, 0.1232], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0677, 0.1011, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 13:40:24,102 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1257234.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:40:25,913 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1257237.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:40:27,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1421, 3.0050, 2.8640, 1.4898], device='cuda:0'), covar=tensor([0.1083, 0.1081, 0.0956, 0.2457], device='cuda:0'), in_proj_covar=tensor([0.1319, 0.1216, 0.1028, 0.0761], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 13:40:37,048 INFO [train.py:968] (0/2) Epoch 28, batch 26350, giga_loss[loss=0.423, simple_loss=0.4324, pruned_loss=0.2068, over 23313.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3767, pruned_loss=0.1249, over 5653671.53 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3591, pruned_loss=0.1113, over 5711803.43 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3777, pruned_loss=0.1254, over 5647319.18 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:40:51,959 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1257266.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:41:05,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6143, 1.8971, 1.5352, 1.8333], device='cuda:0'), covar=tensor([0.2550, 0.2731, 0.2971, 0.2552], device='cuda:0'), in_proj_covar=tensor([0.1600, 0.1155, 0.1415, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 13:41:08,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.266e+03 1.706e+03 2.365e+03 3.309e+03 9.446e+03, threshold=4.730e+03, percent-clipped=7.0
+2023-03-14 13:41:12,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6422, 5.4791, 5.2517, 2.9444], device='cuda:0'), covar=tensor([0.0444, 0.0576, 0.0627, 0.1508], device='cuda:0'), in_proj_covar=tensor([0.1318, 0.1216, 0.1028, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 13:41:20,567 INFO [train.py:968] (0/2) Epoch 28, batch 26400, giga_loss[loss=0.2879, simple_loss=0.3572, pruned_loss=0.1092, over 28689.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3734, pruned_loss=0.1229, over 5653053.41 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3593, pruned_loss=0.1116, over 5714073.93 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3746, pruned_loss=0.1235, over 5643813.13 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:42:05,677 INFO [train.py:968] (0/2) Epoch 28, batch 26450, giga_loss[loss=0.2892, simple_loss=0.3531, pruned_loss=0.1127, over 28905.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3712, pruned_loss=0.1222, over 5662008.42 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3589, pruned_loss=0.1114, over 5720990.62 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.373, pruned_loss=0.1232, over 5646641.68 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:42:13,149 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1257356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:42:15,617 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1257359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:42:41,239 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4266, 1.6187, 1.6008, 1.4689], device='cuda:0'), covar=tensor([0.2457, 0.2051, 0.1745, 0.1986], device='cuda:0'), in_proj_covar=tensor([0.2062, 0.2026, 0.1932, 0.2076], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 13:42:43,292 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.790e+03 2.610e+03 3.444e+03 6.109e+03, threshold=5.221e+03, percent-clipped=9.0
+2023-03-14 13:42:46,661 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1257388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:42:55,237 INFO [train.py:968] (0/2) Epoch 28, batch 26500, libri_loss[loss=0.2982, simple_loss=0.3715, pruned_loss=0.1124, over 29558.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3703, pruned_loss=0.1222, over 5650739.78 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3591, pruned_loss=0.1115, over 5714551.20 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3718, pruned_loss=0.1231, over 5643056.88 frames. ], batch size: 89, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:43:37,829 INFO [train.py:968] (0/2) Epoch 28, batch 26550, giga_loss[loss=0.2854, simple_loss=0.3532, pruned_loss=0.1088, over 28661.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3706, pruned_loss=0.1224, over 5655421.18 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.359, pruned_loss=0.1115, over 5717374.96 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3723, pruned_loss=0.1236, over 5644830.34 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:44:06,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1257481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:44:11,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.334e+03 1.929e+03 2.495e+03 3.609e+03 1.065e+04, threshold=4.990e+03, percent-clipped=11.0
+2023-03-14 13:44:12,766 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1257486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:44:24,796 INFO [train.py:968] (0/2) Epoch 28, batch 26600, giga_loss[loss=0.2704, simple_loss=0.334, pruned_loss=0.1034, over 28722.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3684, pruned_loss=0.1213, over 5651537.43 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.359, pruned_loss=0.1116, over 5709691.23 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3698, pruned_loss=0.1222, over 5649630.40 frames. ], batch size: 66, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:44:41,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5882, 1.7542, 1.2695, 1.4165], device='cuda:0'), covar=tensor([0.1078, 0.0731, 0.1106, 0.1329], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0456, 0.0528, 0.0466], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 13:44:52,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0658, 2.5918, 1.9933, 1.5737], device='cuda:0'), covar=tensor([0.5823, 0.3735, 0.3240, 0.5370], device='cuda:0'), in_proj_covar=tensor([0.1857, 0.1748, 0.1664, 0.1506], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 13:45:06,994 INFO [train.py:968] (0/2) Epoch 28, batch 26650, giga_loss[loss=0.3264, simple_loss=0.3851, pruned_loss=0.1338, over 27599.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3667, pruned_loss=0.1198, over 5655616.20 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3594, pruned_loss=0.1119, over 5698117.87 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3677, pruned_loss=0.1206, over 5663203.49 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:45:33,795 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 13:45:42,903 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.660e+03 2.038e+03 2.451e+03 5.035e+03, threshold=4.076e+03, percent-clipped=1.0
+2023-03-14 13:45:54,636 INFO [train.py:968] (0/2) Epoch 28, batch 26700, libri_loss[loss=0.2453, simple_loss=0.3139, pruned_loss=0.08831, over 29654.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3659, pruned_loss=0.119, over 5661870.65 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3589, pruned_loss=0.1116, over 5699832.07 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3674, pruned_loss=0.12, over 5665002.80 frames. ], batch size: 69, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:46:18,905 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1257624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:46:21,038 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1257627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:46:39,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3184, 1.7062, 1.5091, 1.4703], device='cuda:0'), covar=tensor([0.2412, 0.2258, 0.2656, 0.2423], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0764, 0.0735, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 13:46:39,754 INFO [train.py:968] (0/2) Epoch 28, batch 26750, giga_loss[loss=0.335, simple_loss=0.3949, pruned_loss=0.1376, over 28633.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.368, pruned_loss=0.1195, over 5662123.92 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3587, pruned_loss=0.1115, over 5700537.78 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3695, pruned_loss=0.1205, over 5663450.62 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:46:48,076 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1257656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:47:18,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.274e+03 1.748e+03 2.080e+03 2.782e+03 6.627e+03, threshold=4.160e+03, percent-clipped=8.0
+2023-03-14 13:47:30,729 INFO [train.py:968] (0/2) Epoch 28, batch 26800, libri_loss[loss=0.2937, simple_loss=0.3521, pruned_loss=0.1176, over 29575.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3679, pruned_loss=0.1198, over 5646334.51 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1115, over 5696722.89 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3695, pruned_loss=0.1209, over 5650225.08 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:47:42,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5615, 2.0782, 1.4331, 0.8885], device='cuda:0'), covar=tensor([0.8271, 0.4895, 0.3207, 0.7134], device='cuda:0'), in_proj_covar=tensor([0.1853, 0.1745, 0.1658, 0.1503], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 13:48:03,670 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5059, 1.6516, 1.2738, 1.3091], device='cuda:0'), covar=tensor([0.1054, 0.0618, 0.1096, 0.1080], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0455, 0.0526, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 13:48:12,272 INFO [train.py:968] (0/2) Epoch 28, batch 26850, giga_loss[loss=0.2874, simple_loss=0.3732, pruned_loss=0.1009, over 28992.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3686, pruned_loss=0.1203, over 5660307.86 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3587, pruned_loss=0.1115, over 5704109.88 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3701, pruned_loss=0.1215, over 5655084.59 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:48:22,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0345, 1.3142, 1.1424, 0.2891], device='cuda:0'), covar=tensor([0.4930, 0.3983, 0.4894, 0.7994], device='cuda:0'), in_proj_covar=tensor([0.1854, 0.1746, 0.1659, 0.1505], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 13:48:46,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.630e+03 2.037e+03 2.622e+03 5.611e+03, threshold=4.075e+03, percent-clipped=3.0
+2023-03-14 13:48:56,472 INFO [train.py:968] (0/2) Epoch 28, batch 26900, giga_loss[loss=0.2821, simple_loss=0.3655, pruned_loss=0.09934, over 28956.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3693, pruned_loss=0.1177, over 5672413.09 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3584, pruned_loss=0.1114, over 5706073.06 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3708, pruned_loss=0.1188, over 5666150.08 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:49:43,453 INFO [train.py:968] (0/2) Epoch 28, batch 26950, libri_loss[loss=0.2995, simple_loss=0.3752, pruned_loss=0.1119, over 29375.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3706, pruned_loss=0.1166, over 5674812.85 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1114, over 5708502.15 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.372, pruned_loss=0.1176, over 5666455.34 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:49:53,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1257861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:49:56,222 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2404, 1.4129, 1.3161, 1.2346], device='cuda:0'), covar=tensor([0.2218, 0.2137, 0.1635, 0.1970], device='cuda:0'), in_proj_covar=tensor([0.2066, 0.2026, 0.1940, 0.2082], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 13:50:13,804 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.334e+03 1.752e+03 2.087e+03 2.921e+03 6.730e+03, threshold=4.174e+03, percent-clipped=13.0
+2023-03-14 13:50:25,409 INFO [train.py:968] (0/2) Epoch 28, batch 27000, giga_loss[loss=0.319, simple_loss=0.3866, pruned_loss=0.1257, over 28898.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3734, pruned_loss=0.1184, over 5681154.51 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3588, pruned_loss=0.1117, over 5712328.82 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3745, pruned_loss=0.119, over 5670570.33 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:50:25,413 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 13:50:34,438 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2021, simple_loss=0.3093, pruned_loss=0.04742, over 944034.00 frames. 
+2023-03-14 13:50:34,439 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 13:51:01,316 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1257927.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:51:12,513 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1257938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:51:24,081 INFO [train.py:968] (0/2) Epoch 28, batch 27050, giga_loss[loss=0.3381, simple_loss=0.4024, pruned_loss=0.1369, over 28627.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3781, pruned_loss=0.1235, over 5676909.95 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3589, pruned_loss=0.1118, over 5712671.18 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.379, pruned_loss=0.1239, over 5668191.98 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:51:53,349 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-14 13:51:57,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.156e+03 2.072e+03 2.850e+03 4.017e+03 1.309e+04, threshold=5.699e+03, percent-clipped=23.0
+2023-03-14 13:52:12,873 INFO [train.py:968] (0/2) Epoch 28, batch 27100, giga_loss[loss=0.2869, simple_loss=0.3491, pruned_loss=0.1123, over 28431.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3781, pruned_loss=0.1242, over 5689262.73 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3588, pruned_loss=0.1118, over 5718566.75 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3794, pruned_loss=0.1249, over 5676246.87 frames. ], batch size: 65, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:52:13,878 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1258000.pt
+2023-03-14 13:52:19,534 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:52:21,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:52:44,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8449, 2.2333, 1.4329, 1.8206], device='cuda:0'), covar=tensor([0.1105, 0.0688, 0.1129, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0455, 0.0526, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 13:52:48,468 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:53:00,338 INFO [train.py:968] (0/2) Epoch 28, batch 27150, giga_loss[loss=0.2922, simple_loss=0.3624, pruned_loss=0.111, over 28801.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3783, pruned_loss=0.1254, over 5679383.23 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.359, pruned_loss=0.112, over 5719030.38 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3797, pruned_loss=0.1261, over 5667636.23 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:53:17,355 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-14 13:53:36,285 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.789e+03 2.333e+03 3.051e+03 1.044e+04, threshold=4.667e+03, percent-clipped=4.0
+2023-03-14 13:53:40,974 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3412, 1.2361, 3.7685, 3.2246], device='cuda:0'), covar=tensor([0.1715, 0.2886, 0.0553, 0.1052], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0675, 0.1011, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 13:53:46,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2731, 1.3800, 1.2733, 1.4716], device='cuda:0'), covar=tensor([0.0777, 0.0414, 0.0361, 0.0872], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 13:53:49,348 INFO [train.py:968] (0/2) Epoch 28, batch 27200, giga_loss[loss=0.3374, simple_loss=0.3944, pruned_loss=0.1402, over 27941.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3759, pruned_loss=0.1225, over 5689575.06 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3589, pruned_loss=0.1119, over 5723971.72 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3775, pruned_loss=0.1235, over 5674970.91 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:54:29,442 INFO [train.py:968] (0/2) Epoch 28, batch 27250, giga_loss[loss=0.2917, simple_loss=0.3794, pruned_loss=0.102, over 28952.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3754, pruned_loss=0.1208, over 5667379.87 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3591, pruned_loss=0.1121, over 5710165.49 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.377, pruned_loss=0.1217, over 5666789.69 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:54:49,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1258167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:54:52,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0177, 3.8614, 3.6708, 1.9086], device='cuda:0'), covar=tensor([0.0679, 0.0796, 0.0829, 0.2280], device='cuda:0'), in_proj_covar=tensor([0.1321, 0.1218, 0.1028, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 13:55:07,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.843e+03 2.344e+03 3.072e+03 1.170e+04, threshold=4.688e+03, percent-clipped=8.0
+2023-03-14 13:55:16,358 INFO [train.py:968] (0/2) Epoch 28, batch 27300, giga_loss[loss=0.3255, simple_loss=0.3939, pruned_loss=0.1286, over 28227.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3757, pruned_loss=0.1208, over 5666351.92 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5713808.52 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3774, pruned_loss=0.1215, over 5661373.15 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:55:28,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1897, 2.2720, 1.7626, 1.9057], device='cuda:0'), covar=tensor([0.1019, 0.0744, 0.1021, 0.1149], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0455, 0.0524, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 13:56:07,050 INFO [train.py:968] (0/2) Epoch 28, batch 27350, giga_loss[loss=0.2978, simple_loss=0.3704, pruned_loss=0.1126, over 28707.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3761, pruned_loss=0.1213, over 5640491.97 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1125, over 5696285.89 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3776, pruned_loss=0.122, over 5650918.54 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:56:24,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6672, 1.5792, 1.8544, 1.4408], device='cuda:0'), covar=tensor([0.1437, 0.2280, 0.1202, 0.1542], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0720, 0.0981, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 13:56:33,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1258275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:56:44,777 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.653e+03 2.188e+03 2.973e+03 7.151e+03, threshold=4.377e+03, percent-clipped=8.0
+2023-03-14 13:56:53,999 INFO [train.py:968] (0/2) Epoch 28, batch 27400, giga_loss[loss=0.3689, simple_loss=0.4163, pruned_loss=0.1607, over 27676.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3757, pruned_loss=0.1213, over 5651211.06 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1127, over 5697642.31 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3769, pruned_loss=0.1218, over 5657452.75 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:56:56,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258302.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:57:07,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:57:46,188 INFO [train.py:968] (0/2) Epoch 28, batch 27450, giga_loss[loss=0.2856, simple_loss=0.3543, pruned_loss=0.1084, over 28846.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3744, pruned_loss=0.1216, over 5657048.83 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5698913.97 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3757, pruned_loss=0.1222, over 5660383.14 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:58:23,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.346e+03 1.799e+03 2.599e+03 3.581e+03 7.095e+03, threshold=5.198e+03, percent-clipped=14.0
+2023-03-14 13:58:35,934 INFO [train.py:968] (0/2) Epoch 28, batch 27500, libri_loss[loss=0.3029, simple_loss=0.3824, pruned_loss=0.1118, over 29661.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3731, pruned_loss=0.1216, over 5662644.54 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5701753.49 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3743, pruned_loss=0.1222, over 5662096.11 frames. ], batch size: 88, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:59:16,773 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1258440.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:59:20,506 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258445.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:59:24,686 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258448.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:59:26,638 INFO [train.py:968] (0/2) Epoch 28, batch 27550, giga_loss[loss=0.3624, simple_loss=0.415, pruned_loss=0.1549, over 28611.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3719, pruned_loss=0.1212, over 5669586.22 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1127, over 5707223.80 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.373, pruned_loss=0.1218, over 5663012.89 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:59:31,728 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:59:33,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:59:45,097 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1258473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:59:49,133 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258477.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:59:56,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.153e+03 2.035e+03 2.569e+03 3.363e+03 9.156e+03, threshold=5.137e+03, percent-clipped=10.0
+2023-03-14 13:59:56,332 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:59:56,565 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-14 14:00:06,087 INFO [train.py:968] (0/2) Epoch 28, batch 27600, giga_loss[loss=0.3098, simple_loss=0.3704, pruned_loss=0.1246, over 29011.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3706, pruned_loss=0.1211, over 5661727.97 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3599, pruned_loss=0.1129, over 5701046.23 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3719, pruned_loss=0.1218, over 5660272.50 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:00:26,589 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4705, 2.0994, 1.5603, 0.7203], device='cuda:0'), covar=tensor([0.6555, 0.3558, 0.4727, 0.7280], device='cuda:0'), in_proj_covar=tensor([0.1857, 0.1749, 0.1662, 0.1508], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 14:00:41,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:00:47,376 INFO [train.py:968] (0/2) Epoch 28, batch 27650, giga_loss[loss=0.3022, simple_loss=0.3672, pruned_loss=0.1186, over 28912.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3706, pruned_loss=0.1216, over 5656077.53 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.113, over 5698221.65 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3718, pruned_loss=0.1224, over 5656030.52 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:01:02,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6199, 4.5026, 4.2841, 2.0286], device='cuda:0'), covar=tensor([0.0583, 0.0686, 0.0724, 0.1930], device='cuda:0'), in_proj_covar=tensor([0.1329, 0.1225, 0.1035, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 14:01:21,585 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.669e+03 2.249e+03 3.521e+03 8.863e+03, threshold=4.498e+03, percent-clipped=5.0
+2023-03-14 14:01:24,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1258592.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:01:29,899 INFO [train.py:968] (0/2) Epoch 28, batch 27700, giga_loss[loss=0.2491, simple_loss=0.3353, pruned_loss=0.08146, over 28894.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3674, pruned_loss=0.118, over 5665227.84 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3601, pruned_loss=0.1131, over 5703378.57 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3688, pruned_loss=0.1189, over 5659672.53 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:02:09,707 INFO [train.py:968] (0/2) Epoch 28, batch 27750, giga_loss[loss=0.2623, simple_loss=0.3405, pruned_loss=0.09203, over 28798.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3637, pruned_loss=0.115, over 5674629.04 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1124, over 5712407.74 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3662, pruned_loss=0.1165, over 5659856.38 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:02:10,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:02:46,079 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:02:49,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.750e+03 2.183e+03 3.412e+03 6.934e+03, threshold=4.366e+03, percent-clipped=7.0
+2023-03-14 14:02:50,059 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:03:01,021 INFO [train.py:968] (0/2) Epoch 28, batch 27800, giga_loss[loss=0.2786, simple_loss=0.3447, pruned_loss=0.1062, over 28793.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3628, pruned_loss=0.1142, over 5665317.47 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3587, pruned_loss=0.1122, over 5713022.18 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.365, pruned_loss=0.1155, over 5652167.79 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:03:18,066 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:03:53,070 INFO [train.py:968] (0/2) Epoch 28, batch 27850, giga_loss[loss=0.2739, simple_loss=0.346, pruned_loss=0.1009, over 28757.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3614, pruned_loss=0.1142, over 5651173.20 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1125, over 5703445.64 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3629, pruned_loss=0.1151, over 5647824.82 frames. ], batch size: 243, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:04:08,839 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-14 14:04:32,550 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.347e+03 1.898e+03 2.589e+03 3.881e+03 8.303e+03, threshold=5.178e+03, percent-clipped=19.0
+2023-03-14 14:04:39,426 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:04:42,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:04:43,723 INFO [train.py:968] (0/2) Epoch 28, batch 27900, giga_loss[loss=0.2842, simple_loss=0.3495, pruned_loss=0.1094, over 28881.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3594, pruned_loss=0.1137, over 5657802.47 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3586, pruned_loss=0.1122, over 5708586.38 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.361, pruned_loss=0.1147, over 5648879.76 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:04:58,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258815.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:05:07,268 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:05:29,412 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:05:29,887 INFO [train.py:968] (0/2) Epoch 28, batch 27950, giga_loss[loss=0.2435, simple_loss=0.3252, pruned_loss=0.08088, over 28237.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.114, over 5655392.86 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3586, pruned_loss=0.1121, over 5699469.33 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3628, pruned_loss=0.115, over 5654579.49 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:05:42,466 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5445, 1.7896, 1.6474, 1.6415], device='cuda:0'), covar=tensor([0.1836, 0.1949, 0.2030, 0.1879], device='cuda:0'), in_proj_covar=tensor([0.0509, 0.0761, 0.0734, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 14:06:05,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.719e+03 2.548e+03 3.437e+03 1.383e+04, threshold=5.096e+03, percent-clipped=9.0
+2023-03-14 14:06:14,675 INFO [train.py:968] (0/2) Epoch 28, batch 28000, giga_loss[loss=0.2873, simple_loss=0.3643, pruned_loss=0.1052, over 28855.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3648, pruned_loss=0.1162, over 5637933.13 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3596, pruned_loss=0.1129, over 5693676.55 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3651, pruned_loss=0.1163, over 5641227.94 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:06:58,597 INFO [train.py:968] (0/2) Epoch 28, batch 28050, giga_loss[loss=0.3285, simple_loss=0.3943, pruned_loss=0.1314, over 28529.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3654, pruned_loss=0.1162, over 5638678.87 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3596, pruned_loss=0.1129, over 5682822.44 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3658, pruned_loss=0.1164, over 5648714.45 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:07:07,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258958.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:07:09,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258961.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:07:14,485 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258967.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:07:35,748 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.085e+03 1.706e+03 2.218e+03 3.102e+03 8.185e+03, threshold=4.437e+03, percent-clipped=7.0
+2023-03-14 14:07:36,843 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258990.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:07:37,736 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:07:41,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:07:45,155 INFO [train.py:968] (0/2) Epoch 28, batch 28100, giga_loss[loss=0.3189, simple_loss=0.3601, pruned_loss=0.1389, over 23488.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3665, pruned_loss=0.1176, over 5627261.26 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.36, pruned_loss=0.1132, over 5676033.95 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3666, pruned_loss=0.1176, over 5640196.36 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:08:06,320 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1259023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:08:29,224 INFO [train.py:968] (0/2) Epoch 28, batch 28150, giga_loss[loss=0.2996, simple_loss=0.3671, pruned_loss=0.116, over 29116.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3666, pruned_loss=0.1178, over 5631308.95 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3603, pruned_loss=0.1133, over 5673595.18 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3665, pruned_loss=0.1177, over 5643246.71 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:09:08,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.260e+03 1.982e+03 3.112e+03 4.507e+03 1.031e+04, threshold=6.224e+03, percent-clipped=25.0
+2023-03-14 14:09:12,983 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-14 14:09:15,836 INFO [train.py:968] (0/2) Epoch 28, batch 28200, libri_loss[loss=0.3171, simple_loss=0.3845, pruned_loss=0.1249, over 29757.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3678, pruned_loss=0.1183, over 5645590.34 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3603, pruned_loss=0.1134, over 5679836.07 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.368, pruned_loss=0.1184, over 5648361.99 frames. ], batch size: 87, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:09:28,141 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1259110.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:09:30,463 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1259113.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:09:59,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1259142.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:10:01,404 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-14 14:10:05,552 INFO [train.py:968] (0/2) Epoch 28, batch 28250, giga_loss[loss=0.3489, simple_loss=0.4004, pruned_loss=0.1487, over 27908.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3692, pruned_loss=0.1196, over 5640465.72 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3598, pruned_loss=0.113, over 5678302.82 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.37, pruned_loss=0.12, over 5643515.32 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:10:40,097 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 14:10:45,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.277e+03 1.711e+03 2.296e+03 3.053e+03 7.786e+03, threshold=4.591e+03, percent-clipped=2.0
+2023-03-14 14:10:53,531 INFO [train.py:968] (0/2) Epoch 28, batch 28300, giga_loss[loss=0.2855, simple_loss=0.3622, pruned_loss=0.1044, over 28938.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3708, pruned_loss=0.1212, over 5647028.35 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3597, pruned_loss=0.1129, over 5683767.13 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3718, pruned_loss=0.1219, over 5643248.88 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:10:54,066 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-14 14:11:26,580 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1259233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:11:43,407 INFO [train.py:968] (0/2) Epoch 28, batch 28350, giga_loss[loss=0.2885, simple_loss=0.3677, pruned_loss=0.1047, over 28597.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3707, pruned_loss=0.1215, over 5643851.11 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3597, pruned_loss=0.1129, over 5683520.34 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3716, pruned_loss=0.1221, over 5640331.80 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:12:24,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.172e+03 1.935e+03 2.349e+03 3.307e+03 9.519e+03, threshold=4.697e+03, percent-clipped=9.0
+2023-03-14 14:12:30,590 INFO [train.py:968] (0/2) Epoch 28, batch 28400, giga_loss[loss=0.3078, simple_loss=0.3685, pruned_loss=0.1236, over 28831.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3696, pruned_loss=0.119, over 5655411.37 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5688991.92 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.371, pruned_loss=0.12, over 5646408.08 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:13:08,534 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-14 14:13:19,210 INFO [train.py:968] (0/2) Epoch 28, batch 28450, giga_loss[loss=0.3056, simple_loss=0.372, pruned_loss=0.1196, over 28988.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3697, pruned_loss=0.1195, over 5635190.63 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3592, pruned_loss=0.1126, over 5680868.74 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3712, pruned_loss=0.1205, over 5635350.46 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:13:24,939 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-14 14:13:58,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.194e+03 1.866e+03 2.473e+03 3.165e+03 7.000e+03, threshold=4.947e+03, percent-clipped=10.0
+2023-03-14 14:14:07,330 INFO [train.py:968] (0/2) Epoch 28, batch 28500, giga_loss[loss=0.3281, simple_loss=0.388, pruned_loss=0.1341, over 27905.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3692, pruned_loss=0.1195, over 5636195.02 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3587, pruned_loss=0.1123, over 5686414.29 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3711, pruned_loss=0.1208, over 5630044.19 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:14:17,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-14 14:14:54,347 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 14:15:05,763 INFO [train.py:968] (0/2) Epoch 28, batch 28550, libri_loss[loss=0.2715, simple_loss=0.3418, pruned_loss=0.1006, over 29517.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.368, pruned_loss=0.1195, over 5636750.86 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3585, pruned_loss=0.1122, over 5691208.05 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3699, pruned_loss=0.1209, over 5625743.55 frames. ], batch size: 81, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:15:49,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.696e+03 2.187e+03 2.901e+03 7.954e+03, threshold=4.375e+03, percent-clipped=5.0
+2023-03-14 14:15:56,697 INFO [train.py:968] (0/2) Epoch 28, batch 28600, libri_loss[loss=0.3022, simple_loss=0.3826, pruned_loss=0.1109, over 29507.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3682, pruned_loss=0.1205, over 5638009.59 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3589, pruned_loss=0.1122, over 5694528.88 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3697, pruned_loss=0.1217, over 5625105.56 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:16:22,877 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-14 14:16:40,891 INFO [train.py:968] (0/2) Epoch 28, batch 28650, giga_loss[loss=0.3716, simple_loss=0.4151, pruned_loss=0.1641, over 28570.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3673, pruned_loss=0.1197, over 5658221.58 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5698685.64 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3686, pruned_loss=0.1209, over 5643406.24 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:16:52,931 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1259559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:17:22,004 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.265e+03 1.747e+03 2.238e+03 3.122e+03 6.977e+03, threshold=4.477e+03, percent-clipped=11.0
+2023-03-14 14:17:31,282 INFO [train.py:968] (0/2) Epoch 28, batch 28700, giga_loss[loss=0.3045, simple_loss=0.371, pruned_loss=0.119, over 29004.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3664, pruned_loss=0.1197, over 5652916.47 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3589, pruned_loss=0.1121, over 5701698.37 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3675, pruned_loss=0.1207, over 5637952.60 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:17:33,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3481, 1.4744, 1.3437, 1.4823], device='cuda:0'), covar=tensor([0.0745, 0.0386, 0.0339, 0.0835], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 14:17:40,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1259608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:18:17,948 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5458, 1.7803, 1.2257, 1.2745], device='cuda:0'), covar=tensor([0.1102, 0.0631, 0.1119, 0.1244], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0458, 0.0529, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 14:18:18,289 INFO [train.py:968] (0/2) Epoch 28, batch 28750, giga_loss[loss=0.2816, simple_loss=0.3518, pruned_loss=0.1057, over 28896.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3657, pruned_loss=0.1189, over 5666237.14 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3584, pruned_loss=0.1119, over 5704905.96 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3672, pruned_loss=0.1201, over 5650880.86 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:18:23,281 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3701, 1.2239, 4.1012, 3.3717], device='cuda:0'), covar=tensor([0.1703, 0.2809, 0.0469, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0676, 0.1014, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 14:18:41,333 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1259671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:18:57,208 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.093e+03 1.819e+03 2.230e+03 2.870e+03 6.162e+03, threshold=4.459e+03, percent-clipped=5.0
+2023-03-14 14:19:03,784 INFO [train.py:968] (0/2) Epoch 28, batch 28800, giga_loss[loss=0.3556, simple_loss=0.4074, pruned_loss=0.1519, over 28398.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3677, pruned_loss=0.1202, over 5667485.05 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1118, over 5705196.24 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3689, pruned_loss=0.1214, over 5653750.21 frames. ], batch size: 369, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 14:19:10,628 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1259706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:19:12,972 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-14 14:19:52,552 INFO [train.py:968] (0/2) Epoch 28, batch 28850, giga_loss[loss=0.2882, simple_loss=0.3564, pruned_loss=0.11, over 28945.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3689, pruned_loss=0.1211, over 5668922.12 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3588, pruned_loss=0.1119, over 5705625.74 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1222, over 5656628.75 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:19:54,459 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1259751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:19:56,598 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1259754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:20:00,033 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1259756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:20:23,810 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1259783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:20:31,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.257e+03 1.777e+03 2.271e+03 3.219e+03 7.555e+03, threshold=4.543e+03, percent-clipped=10.0
+2023-03-14 14:20:37,339 INFO [train.py:968] (0/2) Epoch 28, batch 28900, libri_loss[loss=0.3028, simple_loss=0.374, pruned_loss=0.1158, over 29237.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3693, pruned_loss=0.1217, over 5671360.76 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3589, pruned_loss=0.1119, over 5701231.47 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3703, pruned_loss=0.1228, over 5664169.88 frames. ], batch size: 94, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:21:04,407 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 14:21:20,620 INFO [train.py:968] (0/2) Epoch 28, batch 28950, giga_loss[loss=0.3021, simple_loss=0.3757, pruned_loss=0.1143, over 29026.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3688, pruned_loss=0.1212, over 5670825.35 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1118, over 5701314.80 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.37, pruned_loss=0.1223, over 5664498.19 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:22:00,893 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.820e+03 2.232e+03 3.025e+03 6.848e+03, threshold=4.464e+03, percent-clipped=7.0
+2023-03-14 14:22:03,130 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2533, 3.1080, 1.4159, 1.4003], device='cuda:0'), covar=tensor([0.1089, 0.0430, 0.0935, 0.1500], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0574, 0.0412, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 14:22:11,590 INFO [train.py:968] (0/2) Epoch 28, batch 29000, giga_loss[loss=0.3584, simple_loss=0.4045, pruned_loss=0.1562, over 28656.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3705, pruned_loss=0.1224, over 5671196.87 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3581, pruned_loss=0.1115, over 5704548.84 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3721, pruned_loss=0.1238, over 5662787.48 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:22:45,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1259934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:23:00,111 INFO [train.py:968] (0/2) Epoch 28, batch 29050, giga_loss[loss=0.29, simple_loss=0.3624, pruned_loss=0.1088, over 28848.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3709, pruned_loss=0.1224, over 5675192.28 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3581, pruned_loss=0.1113, over 5707672.86 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3723, pruned_loss=0.1238, over 5665500.96 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:23:36,384 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3828, 3.2391, 3.0606, 1.9891], device='cuda:0'), covar=tensor([0.0871, 0.0965, 0.0886, 0.1773], device='cuda:0'), in_proj_covar=tensor([0.1330, 0.1225, 0.1033, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 14:23:37,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.754e+03 2.223e+03 3.237e+03 8.566e+03, threshold=4.445e+03, percent-clipped=8.0
+2023-03-14 14:23:44,196 INFO [train.py:968] (0/2) Epoch 28, batch 29100, giga_loss[loss=0.3154, simple_loss=0.3803, pruned_loss=0.1253, over 28706.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3729, pruned_loss=0.1241, over 5677000.33 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3584, pruned_loss=0.1115, over 5709483.34 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.374, pruned_loss=0.1252, over 5667091.05 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:23:45,680 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1260000.pt
+2023-03-14 14:24:27,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:24:28,618 INFO [train.py:968] (0/2) Epoch 28, batch 29150, giga_loss[loss=0.3091, simple_loss=0.3737, pruned_loss=0.1222, over 28607.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3746, pruned_loss=0.1255, over 5672543.62 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3586, pruned_loss=0.1116, over 5709779.91 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3756, pruned_loss=0.1266, over 5663554.81 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:24:54,019 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:24:57,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:24:58,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:25:07,598 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.145e+03 1.945e+03 2.457e+03 3.760e+03 8.887e+03, threshold=4.913e+03, percent-clipped=15.0
+2023-03-14 14:25:13,366 INFO [train.py:968] (0/2) Epoch 28, batch 29200, giga_loss[loss=0.3137, simple_loss=0.3784, pruned_loss=0.1245, over 28473.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3732, pruned_loss=0.1241, over 5667515.80 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3584, pruned_loss=0.1115, over 5712830.01 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3744, pruned_loss=0.1252, over 5657259.10 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:25:23,170 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:25:44,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3191, 4.1364, 3.9371, 2.0987], device='cuda:0'), covar=tensor([0.0671, 0.0852, 0.0881, 0.1957], device='cuda:0'), in_proj_covar=tensor([0.1329, 0.1224, 0.1032, 0.0761], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 14:25:45,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:26:02,694 INFO [train.py:968] (0/2) Epoch 28, batch 29250, giga_loss[loss=0.3737, simple_loss=0.4131, pruned_loss=0.1671, over 27636.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3734, pruned_loss=0.1237, over 5646692.23 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3583, pruned_loss=0.1115, over 5708498.14 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3747, pruned_loss=0.1249, over 5640927.62 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:26:38,435 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9467, 1.4329, 5.2743, 4.1045], device='cuda:0'), covar=tensor([0.1927, 0.3145, 0.0756, 0.0858], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0678, 0.1014, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 14:26:40,374 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:26:45,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:26:46,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.808e+03 2.146e+03 2.881e+03 1.271e+04, threshold=4.292e+03, percent-clipped=6.0
+2023-03-14 14:26:50,338 INFO [train.py:968] (0/2) Epoch 28, batch 29300, giga_loss[loss=0.2834, simple_loss=0.3432, pruned_loss=0.1118, over 28619.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3726, pruned_loss=0.1225, over 5642109.09 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3584, pruned_loss=0.1116, over 5702733.78 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.374, pruned_loss=0.1237, over 5640971.67 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:26:58,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:08,585 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:11,324 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:13,418 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:33,743 INFO [train.py:968] (0/2) Epoch 28, batch 29350, giga_loss[loss=0.29, simple_loss=0.355, pruned_loss=0.1125, over 28725.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.371, pruned_loss=0.121, over 5652355.86 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3582, pruned_loss=0.1115, over 5705056.94 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3726, pruned_loss=0.1222, over 5648189.76 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:27:42,983 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:56,606 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:59,179 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:28:12,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.170e+03 2.008e+03 2.601e+03 3.460e+03 8.485e+03, threshold=5.202e+03, percent-clipped=13.0
+2023-03-14 14:28:14,845 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:28:17,952 INFO [train.py:968] (0/2) Epoch 28, batch 29400, giga_loss[loss=0.2749, simple_loss=0.3497, pruned_loss=0.09999, over 28319.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3699, pruned_loss=0.1204, over 5644198.72 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3588, pruned_loss=0.112, over 5690429.39 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3711, pruned_loss=0.1212, over 5652721.95 frames. ], batch size: 65, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:28:21,526 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4898, 3.3606, 1.5751, 1.5891], device='cuda:0'), covar=tensor([0.0996, 0.0403, 0.0925, 0.1354], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0575, 0.0412, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 14:28:23,295 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:29:00,500 INFO [train.py:968] (0/2) Epoch 28, batch 29450, giga_loss[loss=0.2744, simple_loss=0.3561, pruned_loss=0.09629, over 28923.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3704, pruned_loss=0.1208, over 5639523.68 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.359, pruned_loss=0.1121, over 5687241.61 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3715, pruned_loss=0.1217, over 5647240.35 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:29:47,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.239e+03 1.755e+03 2.270e+03 3.172e+03 9.811e+03, threshold=4.541e+03, percent-clipped=3.0
+2023-03-14 14:29:52,478 INFO [train.py:968] (0/2) Epoch 28, batch 29500, giga_loss[loss=0.2694, simple_loss=0.3434, pruned_loss=0.09769, over 28957.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3721, pruned_loss=0.1217, over 5649479.67 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3591, pruned_loss=0.1121, over 5688129.95 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3731, pruned_loss=0.1225, over 5653824.15 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:30:09,651 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:30:41,051 INFO [train.py:968] (0/2) Epoch 28, batch 29550, giga_loss[loss=0.2989, simple_loss=0.361, pruned_loss=0.1184, over 29109.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3733, pruned_loss=0.124, over 5647864.54 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1127, over 5689307.44 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3736, pruned_loss=0.1243, over 5649959.75 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:30:59,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4199, 1.7856, 1.4306, 1.3076], device='cuda:0'), covar=tensor([0.2379, 0.2402, 0.2658, 0.2241], device='cuda:0'), in_proj_covar=tensor([0.1609, 0.1157, 0.1419, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 14:31:24,209 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.212e+03 1.880e+03 2.385e+03 3.172e+03 9.026e+03, threshold=4.770e+03, percent-clipped=10.0
+2023-03-14 14:31:25,022 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4749, 2.0200, 1.3920, 0.7951], device='cuda:0'), covar=tensor([0.6288, 0.3134, 0.4029, 0.6801], device='cuda:0'), in_proj_covar=tensor([0.1858, 0.1751, 0.1665, 0.1511], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 14:31:28,491 INFO [train.py:968] (0/2) Epoch 28, batch 29600, giga_loss[loss=0.3273, simple_loss=0.3861, pruned_loss=0.1343, over 29043.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3752, pruned_loss=0.1256, over 5664340.49 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3597, pruned_loss=0.1126, over 5690452.90 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3756, pruned_loss=0.126, over 5664842.32 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:31:43,626 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:32:17,993 INFO [train.py:968] (0/2) Epoch 28, batch 29650, giga_loss[loss=0.332, simple_loss=0.3988, pruned_loss=0.1326, over 28873.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3768, pruned_loss=0.127, over 5662156.63 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3596, pruned_loss=0.1125, over 5692657.89 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3774, pruned_loss=0.1275, over 5660316.70 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:32:54,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:33:03,231 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.151e+03 1.704e+03 2.343e+03 3.244e+03 9.742e+03, threshold=4.685e+03, percent-clipped=12.0
+2023-03-14 14:33:06,981 INFO [train.py:968] (0/2) Epoch 28, batch 29700, giga_loss[loss=0.302, simple_loss=0.3716, pruned_loss=0.1162, over 28949.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3757, pruned_loss=0.1263, over 5642973.20 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3596, pruned_loss=0.1124, over 5686725.20 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3765, pruned_loss=0.127, over 5645646.32 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:33:08,498 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:33:23,562 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:33:52,011 INFO [train.py:968] (0/2) Epoch 28, batch 29750, libri_loss[loss=0.274, simple_loss=0.3444, pruned_loss=0.1018, over 29484.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3759, pruned_loss=0.1264, over 5638680.97 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3598, pruned_loss=0.1126, over 5682446.71 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3769, pruned_loss=0.1273, over 5642840.34 frames. ], batch size: 70, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:34:00,371 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:34:10,498 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:34:11,878 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260669.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:34:12,399 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:34:30,619 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.734e+03 2.109e+03 2.815e+03 5.575e+03, threshold=4.218e+03, percent-clipped=4.0
+2023-03-14 14:34:37,544 INFO [train.py:968] (0/2) Epoch 28, batch 29800, giga_loss[loss=0.3588, simple_loss=0.3927, pruned_loss=0.1625, over 23550.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3754, pruned_loss=0.1251, over 5648379.87 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3601, pruned_loss=0.1128, over 5686915.32 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3763, pruned_loss=0.1259, over 5646942.76 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:35:05,651 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:35:08,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:35:28,181 INFO [train.py:968] (0/2) Epoch 28, batch 29850, giga_loss[loss=0.3192, simple_loss=0.386, pruned_loss=0.1262, over 28603.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3734, pruned_loss=0.123, over 5652246.90 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3601, pruned_loss=0.1128, over 5686639.85 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3741, pruned_loss=0.1238, over 5650982.14 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:35:38,837 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:36:08,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.217e+03 1.745e+03 2.342e+03 3.110e+03 6.968e+03, threshold=4.685e+03, percent-clipped=10.0
+2023-03-14 14:36:09,238 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:36:15,555 INFO [train.py:968] (0/2) Epoch 28, batch 29900, giga_loss[loss=0.2584, simple_loss=0.3353, pruned_loss=0.09076, over 28852.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3708, pruned_loss=0.121, over 5663492.82 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3594, pruned_loss=0.1123, over 5692616.81 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3724, pruned_loss=0.1224, over 5655950.31 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:36:27,887 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:36:29,667 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:36:58,116 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:37:00,487 INFO [train.py:968] (0/2) Epoch 28, batch 29950, giga_loss[loss=0.2588, simple_loss=0.3309, pruned_loss=0.09333, over 28762.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3691, pruned_loss=0.12, over 5668862.94 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3592, pruned_loss=0.112, over 5694492.20 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3708, pruned_loss=0.1214, over 5661055.32 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:37:38,930 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:37:41,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.319e+03 1.910e+03 2.286e+03 3.494e+03 9.924e+03, threshold=4.573e+03, percent-clipped=9.0
+2023-03-14 14:37:49,656 INFO [train.py:968] (0/2) Epoch 28, batch 30000, giga_loss[loss=0.2971, simple_loss=0.3387, pruned_loss=0.1277, over 23435.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3648, pruned_loss=0.1176, over 5667532.71 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3591, pruned_loss=0.1119, over 5697259.26 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3663, pruned_loss=0.1189, over 5658561.97 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 14:37:49,661 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 14:37:58,751 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2044, simple_loss=0.3133, pruned_loss=0.04777, over 944034.00 frames. 
+2023-03-14 14:37:58,752 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 14:38:30,316 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:38:32,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:38:42,397 INFO [train.py:968] (0/2) Epoch 28, batch 30050, giga_loss[loss=0.2756, simple_loss=0.3427, pruned_loss=0.1042, over 28638.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3629, pruned_loss=0.1178, over 5660653.33 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3592, pruned_loss=0.112, over 5700838.00 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3642, pruned_loss=0.119, over 5649261.26 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 14:38:51,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4965, 2.1525, 1.5715, 0.7402], device='cuda:0'), covar=tensor([0.7301, 0.3463, 0.4715, 0.7736], device='cuda:0'), in_proj_covar=tensor([0.1859, 0.1753, 0.1665, 0.1510], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 14:38:57,826 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:39:03,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:39:19,836 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:39:22,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.050e+03 1.754e+03 2.189e+03 2.919e+03 6.740e+03, threshold=4.378e+03, percent-clipped=3.0
+2023-03-14 14:39:27,608 INFO [train.py:968] (0/2) Epoch 28, batch 30100, libri_loss[loss=0.3339, simple_loss=0.3894, pruned_loss=0.1392, over 25946.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3627, pruned_loss=0.1183, over 5656499.33 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3592, pruned_loss=0.1121, over 5692180.52 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3637, pruned_loss=0.1193, over 5653801.96 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:39:39,111 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1261012.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:39:59,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:39:59,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:40:00,668 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5606, 2.2846, 1.8127, 0.7043], device='cuda:0'), covar=tensor([0.7135, 0.3854, 0.4315, 0.8081], device='cuda:0'), in_proj_covar=tensor([0.1860, 0.1753, 0.1664, 0.1510], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 14:40:01,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:40:09,607 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:40:11,079 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261044.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:40:17,060 INFO [train.py:968] (0/2) Epoch 28, batch 30150, giga_loss[loss=0.2394, simple_loss=0.3254, pruned_loss=0.07674, over 28825.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3629, pruned_loss=0.1186, over 5641908.36 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3597, pruned_loss=0.1124, over 5692493.93 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3634, pruned_loss=0.1193, over 5638704.26 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:40:31,445 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:40:34,583 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-14 14:40:56,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 2.108e+03 2.881e+03 4.256e+03 1.205e+04, threshold=5.761e+03, percent-clipped=22.0
+2023-03-14 14:41:00,758 INFO [train.py:968] (0/2) Epoch 28, batch 30200, giga_loss[loss=0.2419, simple_loss=0.3111, pruned_loss=0.08635, over 24250.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3601, pruned_loss=0.1151, over 5647582.76 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5698810.78 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3614, pruned_loss=0.116, over 5637264.56 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:41:18,956 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:41:21,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:41:34,623 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:41:37,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:41:44,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4569, 1.5231, 1.6465, 1.3147], device='cuda:0'), covar=tensor([0.1662, 0.2526, 0.1437, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0721, 0.0985, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 14:41:51,777 INFO [train.py:968] (0/2) Epoch 28, batch 30250, giga_loss[loss=0.2506, simple_loss=0.3346, pruned_loss=0.08331, over 28324.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3581, pruned_loss=0.1117, over 5645158.12 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3589, pruned_loss=0.1123, over 5699102.75 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.359, pruned_loss=0.1122, over 5635408.68 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:41:52,681 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:09,009 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:17,735 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:22,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:26,082 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:27,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261187.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:42:28,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:30,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261190.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:42:34,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.722e+03 2.014e+03 2.689e+03 1.097e+04, threshold=4.028e+03, percent-clipped=1.0
+2023-03-14 14:42:36,705 INFO [train.py:968] (0/2) Epoch 28, batch 30300, giga_loss[loss=0.2653, simple_loss=0.3445, pruned_loss=0.09304, over 28675.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3565, pruned_loss=0.1093, over 5662907.93 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3586, pruned_loss=0.1124, over 5707061.09 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3576, pruned_loss=0.1097, over 5645499.18 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 1.0
+2023-03-14 14:42:47,243 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:56,836 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:58,684 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261219.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:43:22,182 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-14 14:43:24,953 INFO [train.py:968] (0/2) Epoch 28, batch 30350, giga_loss[loss=0.2634, simple_loss=0.3394, pruned_loss=0.09373, over 28475.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3537, pruned_loss=0.1066, over 5653375.64 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3581, pruned_loss=0.1123, over 5699076.57 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3549, pruned_loss=0.1068, over 5644985.88 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 1.0
+2023-03-14 14:43:28,661 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1261251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:43:48,450 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1261273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:44:01,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5145, 1.6522, 1.7338, 1.3326], device='cuda:0'), covar=tensor([0.1947, 0.2843, 0.1650, 0.1990], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0719, 0.0983, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 14:44:11,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.717e+02 1.582e+03 1.965e+03 2.868e+03 1.375e+04, threshold=3.930e+03, percent-clipped=14.0
+2023-03-14 14:44:15,589 INFO [train.py:968] (0/2) Epoch 28, batch 30400, giga_loss[loss=0.2879, simple_loss=0.374, pruned_loss=0.1009, over 28937.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3511, pruned_loss=0.1036, over 5657744.19 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3579, pruned_loss=0.1123, over 5702453.78 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3521, pruned_loss=0.1037, over 5647783.73 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:45:00,634 INFO [train.py:968] (0/2) Epoch 28, batch 30450, giga_loss[loss=0.2609, simple_loss=0.3407, pruned_loss=0.09055, over 28789.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3501, pruned_loss=0.1007, over 5673014.75 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3576, pruned_loss=0.1124, over 5700678.53 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3509, pruned_loss=0.1003, over 5664448.85 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:45:06,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6366, 1.8633, 1.3004, 1.4478], device='cuda:0'), covar=tensor([0.1071, 0.0551, 0.1011, 0.1126], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0454, 0.0525, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 14:45:39,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261387.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:45:47,602 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.911e+02 1.520e+03 1.998e+03 2.806e+03 7.866e+03, threshold=3.995e+03, percent-clipped=6.0
+2023-03-14 14:45:50,601 INFO [train.py:968] (0/2) Epoch 28, batch 30500, giga_loss[loss=0.3198, simple_loss=0.3717, pruned_loss=0.134, over 26621.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3511, pruned_loss=0.1013, over 5672920.57 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3574, pruned_loss=0.1124, over 5704668.96 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3517, pruned_loss=0.1007, over 5661947.69 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:46:12,503 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-14 14:46:14,757 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1261422.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:46:40,925 INFO [train.py:968] (0/2) Epoch 28, batch 30550, giga_loss[loss=0.2355, simple_loss=0.3247, pruned_loss=0.07316, over 28958.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3489, pruned_loss=0.0997, over 5673097.62 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3567, pruned_loss=0.112, over 5707478.66 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.35, pruned_loss=0.09934, over 5661204.55 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:47:27,344 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.561e+02 1.569e+03 2.088e+03 3.429e+03 8.459e+03, threshold=4.177e+03, percent-clipped=14.0
+2023-03-14 14:47:31,704 INFO [train.py:968] (0/2) Epoch 28, batch 30600, giga_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09027, over 28864.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3454, pruned_loss=0.09729, over 5672692.18 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3559, pruned_loss=0.1117, over 5709702.66 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3468, pruned_loss=0.09711, over 5661075.49 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:47:41,225 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6858, 1.8586, 1.2940, 1.4740], device='cuda:0'), covar=tensor([0.1012, 0.0573, 0.1054, 0.1141], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0454, 0.0526, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 14:48:03,362 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261530.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:48:05,933 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261533.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:48:19,298 INFO [train.py:968] (0/2) Epoch 28, batch 30650, giga_loss[loss=0.261, simple_loss=0.3434, pruned_loss=0.08933, over 28730.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3443, pruned_loss=0.0968, over 5666449.04 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3553, pruned_loss=0.1116, over 5711486.65 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3456, pruned_loss=0.09635, over 5653933.92 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:48:31,506 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261562.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:49:04,339 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.245e+02 1.419e+03 1.821e+03 2.403e+03 9.844e+03, threshold=3.642e+03, percent-clipped=7.0
+2023-03-14 14:49:07,347 INFO [train.py:968] (0/2) Epoch 28, batch 30700, giga_loss[loss=0.2553, simple_loss=0.3351, pruned_loss=0.08774, over 28024.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3447, pruned_loss=0.0967, over 5672395.53 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3554, pruned_loss=0.1118, over 5714647.40 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3456, pruned_loss=0.09593, over 5659060.12 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:49:26,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2788, 4.1436, 3.9312, 1.8840], device='cuda:0'), covar=tensor([0.0632, 0.0726, 0.0801, 0.2114], device='cuda:0'), in_proj_covar=tensor([0.1310, 0.1210, 0.1018, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 14:49:31,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5023, 1.9392, 1.8897, 1.5630], device='cuda:0'), covar=tensor([0.2277, 0.1986, 0.2240, 0.2284], device='cuda:0'), in_proj_covar=tensor([0.0505, 0.0756, 0.0729, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 14:49:33,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:49:50,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5498, 4.4361, 4.1478, 2.0520], device='cuda:0'), covar=tensor([0.0582, 0.0698, 0.0775, 0.1976], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.1208, 0.1016, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 14:49:53,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:49:53,985 INFO [train.py:968] (0/2) Epoch 28, batch 30750, giga_loss[loss=0.2377, simple_loss=0.3023, pruned_loss=0.08648, over 24282.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3425, pruned_loss=0.09547, over 5671978.53 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3548, pruned_loss=0.1119, over 5720865.64 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3432, pruned_loss=0.09409, over 5653657.00 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:50:40,041 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.502e+02 1.464e+03 2.072e+03 3.053e+03 1.067e+04, threshold=4.144e+03, percent-clipped=16.0
+2023-03-14 14:50:42,397 INFO [train.py:968] (0/2) Epoch 28, batch 30800, giga_loss[loss=0.2521, simple_loss=0.3292, pruned_loss=0.08752, over 27930.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3397, pruned_loss=0.09299, over 5674676.21 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3546, pruned_loss=0.1118, over 5724682.05 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3403, pruned_loss=0.09167, over 5656166.96 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:51:32,096 INFO [train.py:968] (0/2) Epoch 28, batch 30850, giga_loss[loss=0.2299, simple_loss=0.3176, pruned_loss=0.07113, over 29052.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3366, pruned_loss=0.09101, over 5685279.71 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3545, pruned_loss=0.1118, over 5727434.18 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3368, pruned_loss=0.08952, over 5666956.64 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:51:54,924 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:51:57,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:52:12,806 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:52:15,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:52:16,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.426e+03 1.975e+03 2.583e+03 5.869e+03, threshold=3.950e+03, percent-clipped=5.0
+2023-03-14 14:52:17,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261797.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:52:18,917 INFO [train.py:968] (0/2) Epoch 28, batch 30900, giga_loss[loss=0.2451, simple_loss=0.3109, pruned_loss=0.0896, over 23948.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.336, pruned_loss=0.09147, over 5671203.82 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3546, pruned_loss=0.1119, over 5721601.78 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3356, pruned_loss=0.08968, over 5661399.40 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:52:20,953 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:52:41,341 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:52:52,961 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1261833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:53:06,791 INFO [train.py:968] (0/2) Epoch 28, batch 30950, giga_loss[loss=0.281, simple_loss=0.3544, pruned_loss=0.1038, over 28533.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3363, pruned_loss=0.09229, over 5656254.17 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3543, pruned_loss=0.1119, over 5713733.17 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3357, pruned_loss=0.09023, over 5654647.72 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:53:58,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.261e+02 1.709e+03 2.226e+03 3.161e+03 9.260e+03, threshold=4.451e+03, percent-clipped=13.0
+2023-03-14 14:54:03,298 INFO [train.py:968] (0/2) Epoch 28, batch 31000, giga_loss[loss=0.2758, simple_loss=0.3555, pruned_loss=0.09804, over 28341.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3376, pruned_loss=0.09259, over 5640236.79 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3544, pruned_loss=0.112, over 5704793.50 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.337, pruned_loss=0.09083, over 5646236.96 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:54:34,692 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-14 14:54:45,870 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261940.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:54:47,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261943.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:54:54,946 INFO [train.py:968] (0/2) Epoch 28, batch 31050, giga_loss[loss=0.251, simple_loss=0.3418, pruned_loss=0.08009, over 28952.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3399, pruned_loss=0.0934, over 5638781.54 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3543, pruned_loss=0.1122, over 5704879.00 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3388, pruned_loss=0.09095, over 5641296.71 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:54:59,887 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5599, 1.9063, 1.9367, 1.6162], device='cuda:0'), covar=tensor([0.2862, 0.2202, 0.2332, 0.2425], device='cuda:0'), in_proj_covar=tensor([0.2038, 0.2003, 0.1910, 0.2052], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 14:55:09,152 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-14 14:55:19,361 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261972.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:55:22,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7976, 2.2457, 1.9593, 1.9075], device='cuda:0'), covar=tensor([0.2317, 0.2299, 0.2181, 0.2303], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0751, 0.0724, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 14:55:47,132 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.790e+02 1.677e+03 2.196e+03 3.100e+03 8.975e+03, threshold=4.393e+03, percent-clipped=8.0
+2023-03-14 14:55:49,609 INFO [train.py:968] (0/2) Epoch 28, batch 31100, giga_loss[loss=0.2232, simple_loss=0.3043, pruned_loss=0.07106, over 28662.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3383, pruned_loss=0.09245, over 5631580.68 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3531, pruned_loss=0.1117, over 5704792.34 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3377, pruned_loss=0.08997, over 5630492.05 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:55:51,365 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1262000.pt
+2023-03-14 14:56:55,785 INFO [train.py:968] (0/2) Epoch 28, batch 31150, giga_loss[loss=0.2011, simple_loss=0.2902, pruned_loss=0.05597, over 28837.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.337, pruned_loss=0.09122, over 5637449.92 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3528, pruned_loss=0.1115, over 5705627.63 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3367, pruned_loss=0.08911, over 5634681.24 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:57:49,004 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.559e+03 2.051e+03 2.733e+03 8.145e+03, threshold=4.103e+03, percent-clipped=3.0
+2023-03-14 14:57:53,327 INFO [train.py:968] (0/2) Epoch 28, batch 31200, giga_loss[loss=0.2326, simple_loss=0.3213, pruned_loss=0.07195, over 28937.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3348, pruned_loss=0.08936, over 5643420.73 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3523, pruned_loss=0.1112, over 5708509.29 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3346, pruned_loss=0.0874, over 5636943.41 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 14:58:50,516 INFO [train.py:968] (0/2) Epoch 28, batch 31250, giga_loss[loss=0.2169, simple_loss=0.3083, pruned_loss=0.06275, over 28669.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3329, pruned_loss=0.08704, over 5633571.57 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3517, pruned_loss=0.111, over 5704413.98 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3325, pruned_loss=0.08485, over 5629620.12 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:59:43,566 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:59:50,591 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.362e+02 1.503e+03 2.112e+03 3.002e+03 7.956e+03, threshold=4.224e+03, percent-clipped=11.0
+2023-03-14 14:59:51,877 INFO [train.py:968] (0/2) Epoch 28, batch 31300, libri_loss[loss=0.2997, simple_loss=0.3662, pruned_loss=0.1166, over 27819.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3311, pruned_loss=0.08656, over 5646890.97 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3515, pruned_loss=0.1109, over 5706814.81 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3305, pruned_loss=0.08433, over 5640001.99 frames. ], batch size: 115, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:00:01,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:00:49,749 INFO [train.py:968] (0/2) Epoch 28, batch 31350, giga_loss[loss=0.2837, simple_loss=0.3535, pruned_loss=0.107, over 27742.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3308, pruned_loss=0.08714, over 5651811.88 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3514, pruned_loss=0.111, over 5698987.22 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3301, pruned_loss=0.08491, over 5652209.58 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:01:47,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.203e+02 1.477e+03 1.999e+03 2.678e+03 7.890e+03, threshold=3.997e+03, percent-clipped=5.0
+2023-03-14 15:01:49,831 INFO [train.py:968] (0/2) Epoch 28, batch 31400, giga_loss[loss=0.2384, simple_loss=0.3229, pruned_loss=0.0769, over 28987.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3309, pruned_loss=0.08689, over 5662365.85 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3514, pruned_loss=0.111, over 5699157.28 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3303, pruned_loss=0.08502, over 5662354.91 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:01:54,961 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:02:42,834 INFO [train.py:968] (0/2) Epoch 28, batch 31450, giga_loss[loss=0.2347, simple_loss=0.3265, pruned_loss=0.07144, over 28957.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3327, pruned_loss=0.08766, over 5654915.44 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3513, pruned_loss=0.1111, over 5691819.76 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3318, pruned_loss=0.08551, over 5660394.90 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:02:47,960 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1262351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:02:51,457 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1262354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:03:00,392 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262362.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:03:25,166 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1262383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:03:43,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.663e+02 1.512e+03 1.943e+03 2.377e+03 5.342e+03, threshold=3.886e+03, percent-clipped=5.0
+2023-03-14 15:03:46,539 INFO [train.py:968] (0/2) Epoch 28, batch 31500, giga_loss[loss=0.2283, simple_loss=0.3086, pruned_loss=0.07399, over 28951.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3341, pruned_loss=0.08808, over 5642386.07 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3511, pruned_loss=0.111, over 5687395.17 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3332, pruned_loss=0.08591, over 5649413.91 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:04:13,031 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:04:47,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6106, 1.7376, 1.2180, 1.3546], device='cuda:0'), covar=tensor([0.1092, 0.0670, 0.1130, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0449, 0.0523, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 15:04:49,888 INFO [train.py:968] (0/2) Epoch 28, batch 31550, giga_loss[loss=0.2526, simple_loss=0.3321, pruned_loss=0.08651, over 28912.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3308, pruned_loss=0.08584, over 5663497.57 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3508, pruned_loss=0.111, over 5689116.18 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3301, pruned_loss=0.08383, over 5666940.45 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:05:17,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3847, 1.6903, 1.5323, 1.3263], device='cuda:0'), covar=tensor([0.2734, 0.2286, 0.1854, 0.2407], device='cuda:0'), in_proj_covar=tensor([0.2041, 0.2005, 0.1911, 0.2059], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 15:05:46,923 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:05:54,575 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2492, 1.7113, 1.2943, 0.8462], device='cuda:0'), covar=tensor([0.5927, 0.3487, 0.3482, 0.6619], device='cuda:0'), in_proj_covar=tensor([0.1856, 0.1746, 0.1661, 0.1510], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 15:05:55,707 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.534e+03 2.094e+03 2.911e+03 8.891e+03, threshold=4.188e+03, percent-clipped=5.0
+2023-03-14 15:05:59,469 INFO [train.py:968] (0/2) Epoch 28, batch 31600, giga_loss[loss=0.2707, simple_loss=0.3587, pruned_loss=0.09141, over 28750.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3329, pruned_loss=0.08787, over 5660115.44 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3502, pruned_loss=0.1107, over 5690068.71 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3326, pruned_loss=0.08609, over 5661393.84 frames. ], batch size: 243, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:07:00,156 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:07:01,880 INFO [train.py:968] (0/2) Epoch 28, batch 31650, giga_loss[loss=0.2949, simple_loss=0.3813, pruned_loss=0.1043, over 28100.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3363, pruned_loss=0.08727, over 5667993.09 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3497, pruned_loss=0.1104, over 5692983.11 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3362, pruned_loss=0.0858, over 5666201.36 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:07:02,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:07:25,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:07:26,929 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:08:01,453 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.277e+02 1.503e+03 1.884e+03 2.493e+03 4.913e+03, threshold=3.768e+03, percent-clipped=1.0
+2023-03-14 15:08:04,419 INFO [train.py:968] (0/2) Epoch 28, batch 31700, giga_loss[loss=0.2459, simple_loss=0.3392, pruned_loss=0.07631, over 28612.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3373, pruned_loss=0.08618, over 5652154.32 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3492, pruned_loss=0.1102, over 5686513.15 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3374, pruned_loss=0.08469, over 5655738.71 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:08:04,994 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 15:08:05,732 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 15:08:13,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3842, 1.8496, 1.6277, 1.6452], device='cuda:0'), covar=tensor([0.2361, 0.2441, 0.2382, 0.2385], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0751, 0.0724, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 15:08:14,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6269, 2.1889, 1.3563, 0.8782], device='cuda:0'), covar=tensor([0.9137, 0.4510, 0.5165, 0.7669], device='cuda:0'), in_proj_covar=tensor([0.1851, 0.1739, 0.1658, 0.1505], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 15:08:44,524 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 15:09:00,902 INFO [train.py:968] (0/2) Epoch 28, batch 31750, giga_loss[loss=0.2364, simple_loss=0.3327, pruned_loss=0.07001, over 28758.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3381, pruned_loss=0.08574, over 5660726.05 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3486, pruned_loss=0.1099, over 5690110.41 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3385, pruned_loss=0.08415, over 5659333.50 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:09:17,343 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262663.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 15:09:23,179 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 15:09:26,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7082, 1.9178, 2.0084, 1.4760], device='cuda:0'), covar=tensor([0.2114, 0.2942, 0.1692, 0.2046], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0715, 0.0985, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 15:09:37,044 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:09:59,310 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.266e+02 1.453e+03 1.667e+03 2.618e+03 6.798e+03, threshold=3.335e+03, percent-clipped=9.0
+2023-03-14 15:10:00,158 INFO [train.py:968] (0/2) Epoch 28, batch 31800, giga_loss[loss=0.2699, simple_loss=0.3497, pruned_loss=0.09509, over 28369.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3367, pruned_loss=0.08465, over 5670336.87 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3481, pruned_loss=0.1098, over 5696396.08 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3371, pruned_loss=0.08273, over 5662733.75 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:10:13,049 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1262710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:10:16,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1262713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:10:45,525 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262737.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:10:53,826 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1262742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:11:00,502 INFO [train.py:968] (0/2) Epoch 28, batch 31850, libri_loss[loss=0.2923, simple_loss=0.3525, pruned_loss=0.116, over 19366.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3379, pruned_loss=0.08667, over 5675248.40 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.348, pruned_loss=0.1099, over 5691276.67 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.338, pruned_loss=0.08449, over 5674253.12 frames. ], batch size: 187, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:11:06,907 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-14 15:12:03,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:12:06,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.798e+02 1.662e+03 2.156e+03 2.690e+03 5.640e+03, threshold=4.313e+03, percent-clipped=12.0
+2023-03-14 15:12:08,087 INFO [train.py:968] (0/2) Epoch 28, batch 31900, libri_loss[loss=0.2438, simple_loss=0.3085, pruned_loss=0.08955, over 29540.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3364, pruned_loss=0.08722, over 5677721.78 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3472, pruned_loss=0.1095, over 5696686.51 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3369, pruned_loss=0.08522, over 5671507.93 frames. ], batch size: 70, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:12:36,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1262820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:12:38,862 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1262823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:13:15,769 INFO [train.py:968] (0/2) Epoch 28, batch 31950, giga_loss[loss=0.22, simple_loss=0.307, pruned_loss=0.06646, over 29140.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3372, pruned_loss=0.0886, over 5675534.98 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.347, pruned_loss=0.1093, over 5692716.16 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3374, pruned_loss=0.08627, over 5673319.67 frames. ], batch size: 120, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:13:20,086 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1262852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:13:38,309 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:13:59,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1262880.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:14:02,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1262883.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 15:14:21,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.560e+02 1.440e+03 1.799e+03 2.499e+03 5.319e+03, threshold=3.597e+03, percent-clipped=3.0
+2023-03-14 15:14:22,524 INFO [train.py:968] (0/2) Epoch 28, batch 32000, giga_loss[loss=0.2408, simple_loss=0.3228, pruned_loss=0.07939, over 28978.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3327, pruned_loss=0.08608, over 5675859.05 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3468, pruned_loss=0.1092, over 5695417.02 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3328, pruned_loss=0.08395, over 5671523.43 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:14:42,296 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1262912.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:14:55,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:14:55,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:14:55,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4893, 3.0420, 2.8643, 2.0767], device='cuda:0'), covar=tensor([0.2440, 0.1640, 0.1906, 0.2451], device='cuda:0'), in_proj_covar=tensor([0.2022, 0.1988, 0.1893, 0.2042], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 15:15:09,804 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1262938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:15:12,362 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1262941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:15:15,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:15:20,999 INFO [train.py:968] (0/2) Epoch 28, batch 32050, giga_loss[loss=0.229, simple_loss=0.3152, pruned_loss=0.07137, over 28480.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3303, pruned_loss=0.0849, over 5669531.40 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3464, pruned_loss=0.1089, over 5691165.76 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3301, pruned_loss=0.08231, over 5668961.42 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:15:48,896 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1262970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:16:22,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.472e+02 1.582e+03 1.918e+03 2.740e+03 6.446e+03, threshold=3.835e+03, percent-clipped=11.0
+2023-03-14 15:16:24,411 INFO [train.py:968] (0/2) Epoch 28, batch 32100, giga_loss[loss=0.2818, simple_loss=0.356, pruned_loss=0.1038, over 28895.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3294, pruned_loss=0.0848, over 5678455.77 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3467, pruned_loss=0.1093, over 5690704.34 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3287, pruned_loss=0.08203, over 5677875.78 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:16:38,039 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:16:42,183 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:16:45,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3308, 1.4230, 1.2296, 1.5239], device='cuda:0'), covar=tensor([0.0779, 0.0397, 0.0377, 0.0860], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 15:17:15,604 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1263038.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 15:17:18,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1263040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:17:27,275 INFO [train.py:968] (0/2) Epoch 28, batch 32150, giga_loss[loss=0.2219, simple_loss=0.3141, pruned_loss=0.06485, over 28744.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3337, pruned_loss=0.08701, over 5678931.07 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.347, pruned_loss=0.1096, over 5694160.50 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3327, pruned_loss=0.08424, over 5675014.25 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:17:48,781 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:17:49,682 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:17:52,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:17:52,913 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:18:10,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:18:14,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:18:28,244 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1263098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:18:28,726 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.765e+03 2.290e+03 3.222e+03 6.517e+03, threshold=4.580e+03, percent-clipped=16.0
+2023-03-14 15:18:28,739 INFO [train.py:968] (0/2) Epoch 28, batch 32200, libri_loss[loss=0.253, simple_loss=0.3218, pruned_loss=0.09214, over 29568.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3329, pruned_loss=0.0873, over 5682759.36 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.347, pruned_loss=0.1097, over 5687088.63 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3318, pruned_loss=0.08453, over 5685464.07 frames. ], batch size: 76, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:18:29,325 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1263099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:18:36,894 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4897, 1.6785, 1.3245, 1.6220], device='cuda:0'), covar=tensor([0.0778, 0.0320, 0.0363, 0.0920], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 15:18:51,150 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-14 15:18:57,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1263119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:19:28,634 INFO [train.py:968] (0/2) Epoch 28, batch 32250, giga_loss[loss=0.2853, simple_loss=0.355, pruned_loss=0.1078, over 27518.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3328, pruned_loss=0.08815, over 5682676.18 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3468, pruned_loss=0.1096, over 5692607.39 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3318, pruned_loss=0.08551, over 5679643.62 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:20:02,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7051, 4.9232, 1.8546, 2.0775], device='cuda:0'), covar=tensor([0.0985, 0.0340, 0.0937, 0.1241], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0570, 0.0410, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 15:20:05,022 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263181.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:20:08,418 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263184.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 15:20:28,628 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.561e+03 2.102e+03 3.012e+03 1.035e+04, threshold=4.203e+03, percent-clipped=10.0
+2023-03-14 15:20:28,640 INFO [train.py:968] (0/2) Epoch 28, batch 32300, giga_loss[loss=0.2389, simple_loss=0.3252, pruned_loss=0.07623, over 28992.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.334, pruned_loss=0.08957, over 5682583.19 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3472, pruned_loss=0.1099, over 5695607.03 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3325, pruned_loss=0.08676, over 5677216.21 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:20:50,011 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1263213.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:21:23,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1941, 1.6664, 1.6168, 1.4787], device='cuda:0'), covar=tensor([0.2371, 0.1956, 0.2242, 0.2059], device='cuda:0'), in_proj_covar=tensor([0.0495, 0.0749, 0.0722, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 15:21:23,566 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-14 15:21:38,395 INFO [train.py:968] (0/2) Epoch 28, batch 32350, giga_loss[loss=0.2417, simple_loss=0.3327, pruned_loss=0.07532, over 28715.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3357, pruned_loss=0.08946, over 5677956.71 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3474, pruned_loss=0.11, over 5694375.41 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3343, pruned_loss=0.08689, over 5674861.58 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:22:36,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6053, 1.7976, 1.7780, 1.5807], device='cuda:0'), covar=tensor([0.2632, 0.2232, 0.1924, 0.2320], device='cuda:0'), in_proj_covar=tensor([0.2027, 0.1986, 0.1889, 0.2042], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 15:22:50,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.670e+03 2.217e+03 3.233e+03 1.074e+04, threshold=4.435e+03, percent-clipped=13.0
+2023-03-14 15:22:50,291 INFO [train.py:968] (0/2) Epoch 28, batch 32400, giga_loss[loss=0.2627, simple_loss=0.3509, pruned_loss=0.08723, over 28941.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3376, pruned_loss=0.09014, over 5670445.05 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3471, pruned_loss=0.1099, over 5695993.13 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3365, pruned_loss=0.08782, over 5666185.80 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:23:58,025 INFO [train.py:968] (0/2) Epoch 28, batch 32450, giga_loss[loss=0.2148, simple_loss=0.2969, pruned_loss=0.06635, over 29120.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3335, pruned_loss=0.08759, over 5676226.71 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.346, pruned_loss=0.1092, over 5700803.70 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3332, pruned_loss=0.08576, over 5667834.76 frames. ], batch size: 113, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:24:22,975 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.61 vs. limit=5.0
+2023-03-14 15:24:58,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.865e+02 1.502e+03 1.948e+03 3.027e+03 5.732e+03, threshold=3.895e+03, percent-clipped=6.0
+2023-03-14 15:24:58,414 INFO [train.py:968] (0/2) Epoch 28, batch 32500, giga_loss[loss=0.2074, simple_loss=0.2927, pruned_loss=0.0611, over 28840.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3297, pruned_loss=0.08675, over 5672749.17 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3462, pruned_loss=0.1092, over 5694055.92 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.329, pruned_loss=0.08456, over 5671398.04 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:26:02,381 INFO [train.py:968] (0/2) Epoch 28, batch 32550, giga_loss[loss=0.2501, simple_loss=0.3271, pruned_loss=0.08658, over 28725.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3256, pruned_loss=0.0854, over 5669036.02 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3457, pruned_loss=0.1091, over 5696490.57 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.325, pruned_loss=0.08318, over 5665431.21 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:26:14,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1263458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:26:38,177 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1263478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:26:46,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5728, 1.7683, 1.7527, 1.3560], device='cuda:0'), covar=tensor([0.1600, 0.2578, 0.1433, 0.1767], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0712, 0.0982, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 15:26:53,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4705, 4.1646, 1.6587, 1.7889], device='cuda:0'), covar=tensor([0.1011, 0.0365, 0.0973, 0.1317], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0570, 0.0411, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 15:26:55,532 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 15:26:59,443 INFO [train.py:968] (0/2) Epoch 28, batch 32600, libri_loss[loss=0.2373, simple_loss=0.3089, pruned_loss=0.08287, over 29586.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3263, pruned_loss=0.08602, over 5670666.77 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3453, pruned_loss=0.1087, over 5698186.99 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3255, pruned_loss=0.0838, over 5665285.05 frames. ], batch size: 74, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:27:00,258 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.595e+03 1.981e+03 2.441e+03 4.932e+03, threshold=3.962e+03, percent-clipped=7.0
+2023-03-14 15:27:14,564 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1263514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:27:39,721 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1263538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:27:55,928 INFO [train.py:968] (0/2) Epoch 28, batch 32650, giga_loss[loss=0.2474, simple_loss=0.3279, pruned_loss=0.08349, over 28885.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3275, pruned_loss=0.08629, over 5680691.27 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3445, pruned_loss=0.1082, over 5703287.79 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3271, pruned_loss=0.08435, over 5671323.60 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:27:57,315 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 15:28:06,413 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3869, 1.3207, 1.2028, 1.5437], device='cuda:0'), covar=tensor([0.0778, 0.0365, 0.0378, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0120, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 15:28:58,459 INFO [train.py:968] (0/2) Epoch 28, batch 32700, giga_loss[loss=0.2507, simple_loss=0.3176, pruned_loss=0.09192, over 26856.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3252, pruned_loss=0.08413, over 5674568.55 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3447, pruned_loss=0.1083, over 5705548.79 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3245, pruned_loss=0.08229, over 5664935.86 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:28:59,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.849e+02 1.784e+03 2.453e+03 3.220e+03 7.948e+03, threshold=4.905e+03, percent-clipped=11.0
+2023-03-14 15:29:59,288 INFO [train.py:968] (0/2) Epoch 28, batch 32750, giga_loss[loss=0.2514, simple_loss=0.3303, pruned_loss=0.08629, over 28042.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3251, pruned_loss=0.08392, over 5662212.59 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3452, pruned_loss=0.1086, over 5698573.04 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3238, pruned_loss=0.08181, over 5660083.53 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:31:08,741 INFO [train.py:968] (0/2) Epoch 28, batch 32800, giga_loss[loss=0.2533, simple_loss=0.3288, pruned_loss=0.08892, over 27625.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3245, pruned_loss=0.08428, over 5664332.46 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3452, pruned_loss=0.1088, over 5696845.39 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3231, pruned_loss=0.08213, over 5663593.38 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:31:09,794 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.869e+02 1.539e+03 2.126e+03 3.020e+03 9.810e+03, threshold=4.252e+03, percent-clipped=8.0
+2023-03-14 15:32:12,379 INFO [train.py:968] (0/2) Epoch 28, batch 32850, giga_loss[loss=0.2772, simple_loss=0.3424, pruned_loss=0.106, over 26902.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3252, pruned_loss=0.08402, over 5663374.83 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3451, pruned_loss=0.1089, over 5688443.98 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3236, pruned_loss=0.08148, over 5670365.12 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:32:21,975 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 15:33:13,771 INFO [train.py:968] (0/2) Epoch 28, batch 32900, giga_loss[loss=0.2714, simple_loss=0.3434, pruned_loss=0.09965, over 28109.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3253, pruned_loss=0.08402, over 5674237.14 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3448, pruned_loss=0.1087, over 5693762.80 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3238, pruned_loss=0.08158, over 5674438.78 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:33:16,197 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.546e+02 1.288e+03 1.768e+03 2.743e+03 1.049e+04, threshold=3.536e+03, percent-clipped=11.0
+2023-03-14 15:33:53,740 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1263833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:34:13,756 INFO [train.py:968] (0/2) Epoch 28, batch 32950, giga_loss[loss=0.2131, simple_loss=0.3048, pruned_loss=0.06071, over 28587.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3258, pruned_loss=0.08496, over 5680798.02 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3445, pruned_loss=0.1086, over 5699129.89 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3244, pruned_loss=0.08246, over 5676004.92 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:34:18,980 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1263853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:34:58,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1263889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:35:00,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3984, 1.5548, 1.4505, 1.4185], device='cuda:0'), covar=tensor([0.2014, 0.2010, 0.1576, 0.1823], device='cuda:0'), in_proj_covar=tensor([0.2025, 0.1982, 0.1886, 0.2039], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 15:35:11,398 INFO [train.py:968] (0/2) Epoch 28, batch 33000, giga_loss[loss=0.2272, simple_loss=0.3088, pruned_loss=0.07283, over 27661.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3241, pruned_loss=0.08345, over 5675858.34 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3437, pruned_loss=0.1081, over 5700991.85 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3231, pruned_loss=0.08117, over 5669405.79 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:35:11,403 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 15:35:19,554 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.1926, simple_loss=0.2944, pruned_loss=0.04546, over 944034.00 frames. 
+2023-03-14 15:35:19,555 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 15:35:20,954 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.759e+02 1.506e+03 2.033e+03 2.850e+03 1.132e+04, threshold=4.066e+03, percent-clipped=13.0
+2023-03-14 15:35:36,060 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1263913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:35:46,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4979, 1.6535, 1.6983, 1.4869], device='cuda:0'), covar=tensor([0.2869, 0.2444, 0.2066, 0.2505], device='cuda:0'), in_proj_covar=tensor([0.2025, 0.1980, 0.1885, 0.2037], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 15:36:16,361 INFO [train.py:968] (0/2) Epoch 28, batch 33050, giga_loss[loss=0.2566, simple_loss=0.3469, pruned_loss=0.08311, over 28919.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3257, pruned_loss=0.08297, over 5670562.71 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3431, pruned_loss=0.1078, over 5704968.89 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3249, pruned_loss=0.08075, over 5661216.97 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:36:26,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1263958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:36:48,051 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:36:51,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:06,480 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:08,179 INFO [train.py:968] (0/2) Epoch 28, batch 33100, giga_loss[loss=0.249, simple_loss=0.3347, pruned_loss=0.08163, over 28997.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3295, pruned_loss=0.08515, over 5673328.46 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3426, pruned_loss=0.1075, over 5707887.19 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3286, pruned_loss=0.08263, over 5661043.25 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:37:08,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:08,815 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1264000.pt
+2023-03-14 15:37:09,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.101e+03 1.519e+03 1.984e+03 2.622e+03 7.544e+03, threshold=3.967e+03, percent-clipped=6.0
+2023-03-14 15:37:18,208 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:25,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8699, 2.4266, 2.1034, 1.9041], device='cuda:0'), covar=tensor([0.2189, 0.1973, 0.1987, 0.2209], device='cuda:0'), in_proj_covar=tensor([0.0493, 0.0746, 0.0717, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 15:37:28,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9210, 2.8857, 2.2031, 1.0036], device='cuda:0'), covar=tensor([0.9557, 0.4222, 0.3826, 0.7990], device='cuda:0'), in_proj_covar=tensor([0.1839, 0.1730, 0.1652, 0.1499], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 15:37:43,445 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:48,217 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1264032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:52,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1264035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:56,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4692, 1.6184, 1.3119, 1.5802], device='cuda:0'), covar=tensor([0.0699, 0.0315, 0.0325, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 15:38:09,358 INFO [train.py:968] (0/2) Epoch 28, batch 33150, giga_loss[loss=0.2474, simple_loss=0.331, pruned_loss=0.08193, over 28670.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.331, pruned_loss=0.08562, over 5670111.41 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3431, pruned_loss=0.1079, over 5701832.19 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3296, pruned_loss=0.08278, over 5664818.05 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:38:18,270 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1264056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:38:22,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1264059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:38:29,891 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:39:00,474 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:39:11,050 INFO [train.py:968] (0/2) Epoch 28, batch 33200, giga_loss[loss=0.2268, simple_loss=0.3065, pruned_loss=0.07352, over 28930.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3304, pruned_loss=0.08531, over 5664649.92 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3428, pruned_loss=0.1077, over 5696146.38 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3293, pruned_loss=0.08269, over 5663992.32 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:39:13,390 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.994e+02 1.435e+03 1.920e+03 2.812e+03 1.445e+04, threshold=3.840e+03, percent-clipped=6.0
+2023-03-14 15:39:23,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4136, 4.2679, 4.0137, 2.1933], device='cuda:0'), covar=tensor([0.0586, 0.0691, 0.0784, 0.2203], device='cuda:0'), in_proj_covar=tensor([0.1291, 0.1186, 0.1001, 0.0738], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 15:40:03,638 INFO [train.py:968] (0/2) Epoch 28, batch 33250, giga_loss[loss=0.2484, simple_loss=0.3291, pruned_loss=0.0839, over 28965.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3266, pruned_loss=0.08291, over 5675506.87 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3419, pruned_loss=0.1072, over 5702594.13 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.326, pruned_loss=0.08041, over 5668074.28 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:40:35,369 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-14 15:41:01,181 INFO [train.py:968] (0/2) Epoch 28, batch 33300, giga_loss[loss=0.2844, simple_loss=0.3579, pruned_loss=0.1055, over 28604.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3257, pruned_loss=0.08241, over 5676286.44 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3417, pruned_loss=0.1071, over 5699542.30 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.325, pruned_loss=0.07982, over 5671950.23 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:41:04,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.890e+02 1.523e+03 1.937e+03 2.896e+03 6.354e+03, threshold=3.874e+03, percent-clipped=9.0
+2023-03-14 15:42:01,807 INFO [train.py:968] (0/2) Epoch 28, batch 33350, giga_loss[loss=0.2363, simple_loss=0.3184, pruned_loss=0.07714, over 27825.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.323, pruned_loss=0.08133, over 5678554.55 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3409, pruned_loss=0.1065, over 5704468.13 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3228, pruned_loss=0.07916, over 5669973.30 frames. ], batch size: 474, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:42:09,480 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1264257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:42:17,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5786, 1.9209, 1.3360, 1.4174], device='cuda:0'), covar=tensor([0.1087, 0.0529, 0.1047, 0.1178], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0447, 0.0523, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 15:43:00,657 INFO [train.py:968] (0/2) Epoch 28, batch 33400, giga_loss[loss=0.2609, simple_loss=0.3415, pruned_loss=0.09019, over 28141.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3244, pruned_loss=0.0819, over 5679450.48 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3405, pruned_loss=0.1063, over 5708228.43 frames. ], giga_tot_loss[loss=0.242, simple_loss=0.3242, pruned_loss=0.0799, over 5668830.51 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:43:03,419 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.949e+02 1.422e+03 1.942e+03 2.607e+03 4.992e+03, threshold=3.884e+03, percent-clipped=7.0
+2023-03-14 15:43:10,459 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1264305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:43:44,262 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1264333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:44:00,963 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 15:44:01,283 INFO [train.py:968] (0/2) Epoch 28, batch 33450, giga_loss[loss=0.2507, simple_loss=0.3295, pruned_loss=0.08588, over 28685.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3261, pruned_loss=0.08309, over 5682960.13 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3397, pruned_loss=0.1059, over 5713374.47 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3263, pruned_loss=0.08121, over 5669188.87 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:44:15,081 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-14 15:45:08,968 INFO [train.py:968] (0/2) Epoch 28, batch 33500, giga_loss[loss=0.2458, simple_loss=0.3196, pruned_loss=0.08595, over 28700.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3266, pruned_loss=0.08394, over 5676293.98 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3394, pruned_loss=0.1057, over 5716349.09 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3269, pruned_loss=0.08235, over 5662392.61 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:45:11,669 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.548e+03 1.924e+03 2.599e+03 9.397e+03, threshold=3.848e+03, percent-clipped=13.0
+2023-03-14 15:46:10,408 INFO [train.py:968] (0/2) Epoch 28, batch 33550, giga_loss[loss=0.2461, simple_loss=0.3321, pruned_loss=0.08006, over 28895.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3319, pruned_loss=0.08717, over 5656260.75 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3396, pruned_loss=0.1059, over 5708137.70 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3317, pruned_loss=0.08524, over 5652345.51 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:46:37,804 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1264476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:46:40,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1264479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:47:04,341 INFO [train.py:968] (0/2) Epoch 28, batch 33600, giga_loss[loss=0.2382, simple_loss=0.3323, pruned_loss=0.07203, over 28437.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3337, pruned_loss=0.08692, over 5664418.82 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3391, pruned_loss=0.1054, over 5709235.66 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3338, pruned_loss=0.08536, over 5659243.87 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:47:08,216 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.086e+03 1.460e+03 1.888e+03 2.557e+03 6.778e+03, threshold=3.775e+03, percent-clipped=7.0
+2023-03-14 15:47:15,520 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:47:55,688 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 15:48:04,584 INFO [train.py:968] (0/2) Epoch 28, batch 33650, giga_loss[loss=0.2496, simple_loss=0.3323, pruned_loss=0.0835, over 28940.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3352, pruned_loss=0.08857, over 5667470.82 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3391, pruned_loss=0.1056, over 5713271.47 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3351, pruned_loss=0.08649, over 5657928.54 frames. ], batch size: 228, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:48:19,091 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0784, 1.4941, 1.5246, 1.2964], device='cuda:0'), covar=tensor([0.2425, 0.1900, 0.2461, 0.2069], device='cuda:0'), in_proj_covar=tensor([0.0491, 0.0745, 0.0716, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 15:48:48,966 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-14 15:49:06,386 INFO [train.py:968] (0/2) Epoch 28, batch 33700, giga_loss[loss=0.2131, simple_loss=0.3027, pruned_loss=0.06176, over 28826.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3331, pruned_loss=0.08761, over 5679600.65 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.339, pruned_loss=0.1054, over 5721586.00 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3329, pruned_loss=0.08521, over 5662517.48 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:49:10,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0441, 3.9040, 3.6932, 1.9915], device='cuda:0'), covar=tensor([0.0645, 0.0817, 0.0842, 0.2313], device='cuda:0'), in_proj_covar=tensor([0.1302, 0.1194, 0.1008, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 15:49:12,490 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.794e+03 2.351e+03 3.266e+03 8.962e+03, threshold=4.703e+03, percent-clipped=16.0
+2023-03-14 15:49:49,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1264632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:50:08,435 INFO [train.py:968] (0/2) Epoch 28, batch 33750, giga_loss[loss=0.2441, simple_loss=0.3252, pruned_loss=0.08151, over 27740.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3322, pruned_loss=0.08735, over 5686951.07 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3392, pruned_loss=0.1056, over 5722194.00 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3316, pruned_loss=0.08463, over 5671613.61 frames. ], batch size: 474, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:50:49,209 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 15:50:49,850 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1264680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:51:13,416 INFO [train.py:968] (0/2) Epoch 28, batch 33800, giga_loss[loss=0.2287, simple_loss=0.3097, pruned_loss=0.07387, over 28866.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3313, pruned_loss=0.08681, over 5683180.11 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3393, pruned_loss=0.1057, over 5724090.13 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3307, pruned_loss=0.08441, over 5669022.67 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:51:19,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.590e+03 2.247e+03 3.035e+03 5.213e+03, threshold=4.494e+03, percent-clipped=3.0
+2023-03-14 15:51:31,328 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-14 15:52:20,608 INFO [train.py:968] (0/2) Epoch 28, batch 33850, giga_loss[loss=0.2398, simple_loss=0.3156, pruned_loss=0.08199, over 28845.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3291, pruned_loss=0.08654, over 5683546.05 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3393, pruned_loss=0.1057, over 5727708.13 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3284, pruned_loss=0.08429, over 5668404.53 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:52:40,182 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1264762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:52:47,430 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1264769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:52:53,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1264775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:52:57,984 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1264778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:53:01,708 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-14 15:53:06,655 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2662, 1.8506, 1.7057, 1.4820], device='cuda:0'), covar=tensor([0.2396, 0.1791, 0.2200, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.0491, 0.0746, 0.0716, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 15:53:24,752 INFO [train.py:968] (0/2) Epoch 28, batch 33900, giga_loss[loss=0.2879, simple_loss=0.3644, pruned_loss=0.1057, over 28932.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3276, pruned_loss=0.08568, over 5680559.01 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.339, pruned_loss=0.1055, over 5718876.76 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3271, pruned_loss=0.08364, over 5675276.97 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:53:29,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.245e+02 1.431e+03 1.857e+03 2.424e+03 6.388e+03, threshold=3.715e+03, percent-clipped=2.0
+2023-03-14 15:53:33,378 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:53:45,370 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 15:53:52,590 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1264823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:53:55,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1264826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:54:25,743 INFO [train.py:968] (0/2) Epoch 28, batch 33950, libri_loss[loss=0.2785, simple_loss=0.3486, pruned_loss=0.1042, over 29536.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3271, pruned_loss=0.08435, over 5678169.96 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.339, pruned_loss=0.1054, over 5722878.11 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3266, pruned_loss=0.08239, over 5669545.94 frames. ], batch size: 81, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:54:33,512 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:54:40,718 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-14 15:55:21,284 INFO [train.py:968] (0/2) Epoch 28, batch 34000, giga_loss[loss=0.2176, simple_loss=0.3121, pruned_loss=0.06159, over 28944.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3273, pruned_loss=0.08302, over 5679354.72 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.339, pruned_loss=0.1055, over 5717288.52 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3266, pruned_loss=0.08095, over 5676434.92 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:55:25,916 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.996e+02 1.495e+03 1.921e+03 2.744e+03 5.887e+03, threshold=3.843e+03, percent-clipped=10.0
+2023-03-14 15:56:08,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4559, 1.6624, 1.7224, 1.2677], device='cuda:0'), covar=tensor([0.2066, 0.3123, 0.1697, 0.2110], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0710, 0.0980, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 15:56:17,634 INFO [train.py:968] (0/2) Epoch 28, batch 34050, giga_loss[loss=0.2179, simple_loss=0.3148, pruned_loss=0.06047, over 28764.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3293, pruned_loss=0.08269, over 5685656.39 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.339, pruned_loss=0.1056, over 5721619.55 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3283, pruned_loss=0.08022, over 5678116.48 frames. ], batch size: 243, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:56:51,224 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0332, 2.4227, 1.7484, 1.9284], device='cuda:0'), covar=tensor([0.1119, 0.0687, 0.1007, 0.1220], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0446, 0.0521, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 15:57:12,980 INFO [train.py:968] (0/2) Epoch 28, batch 34100, giga_loss[loss=0.2331, simple_loss=0.322, pruned_loss=0.07208, over 28945.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3308, pruned_loss=0.08324, over 5686352.20 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3391, pruned_loss=0.1056, over 5722185.10 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3297, pruned_loss=0.08059, over 5679150.92 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:57:17,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.488e+03 1.908e+03 2.446e+03 5.352e+03, threshold=3.817e+03, percent-clipped=4.0
+2023-03-14 15:58:21,482 INFO [train.py:968] (0/2) Epoch 28, batch 34150, giga_loss[loss=0.2215, simple_loss=0.31, pruned_loss=0.06652, over 28865.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3305, pruned_loss=0.08338, over 5680072.50 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3389, pruned_loss=0.1055, over 5726788.16 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3297, pruned_loss=0.08088, over 5669435.83 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:59:20,523 INFO [train.py:968] (0/2) Epoch 28, batch 34200, giga_loss[loss=0.241, simple_loss=0.3266, pruned_loss=0.07767, over 28458.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.331, pruned_loss=0.08414, over 5671532.52 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3388, pruned_loss=0.1056, over 5721273.89 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3301, pruned_loss=0.08112, over 5665890.46 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:59:27,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.315e+02 1.575e+03 2.172e+03 3.257e+03 1.078e+04, threshold=4.345e+03, percent-clipped=14.0
+2023-03-14 16:00:14,481 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1265137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:00:20,420 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-14 16:00:22,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1265144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:00:27,924 INFO [train.py:968] (0/2) Epoch 28, batch 34250, giga_loss[loss=0.2547, simple_loss=0.3495, pruned_loss=0.07997, over 28747.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3315, pruned_loss=0.08399, over 5674878.26 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3385, pruned_loss=0.1053, over 5725410.44 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3309, pruned_loss=0.08136, over 5665434.53 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:01:27,313 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1265196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:01:29,924 INFO [train.py:968] (0/2) Epoch 28, batch 34300, giga_loss[loss=0.2272, simple_loss=0.317, pruned_loss=0.0687, over 28716.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3317, pruned_loss=0.08442, over 5681047.34 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3382, pruned_loss=0.1052, over 5734049.42 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.331, pruned_loss=0.08111, over 5662761.31 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:01:35,995 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.917e+02 1.366e+03 1.785e+03 2.471e+03 5.926e+03, threshold=3.570e+03, percent-clipped=1.0
+2023-03-14 16:02:00,693 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5359, 5.3672, 5.1396, 2.5118], device='cuda:0'), covar=tensor([0.0519, 0.0651, 0.0742, 0.1736], device='cuda:0'), in_proj_covar=tensor([0.1297, 0.1192, 0.1004, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 16:02:02,915 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5891, 1.7135, 1.7059, 1.5930], device='cuda:0'), covar=tensor([0.2815, 0.2510, 0.2205, 0.2392], device='cuda:0'), in_proj_covar=tensor([0.2025, 0.1979, 0.1882, 0.2032], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 16:02:10,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5368, 4.3729, 4.1935, 1.9964], device='cuda:0'), covar=tensor([0.0634, 0.0757, 0.0835, 0.1948], device='cuda:0'), in_proj_covar=tensor([0.1296, 0.1191, 0.1003, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 16:02:36,864 INFO [train.py:968] (0/2) Epoch 28, batch 34350, giga_loss[loss=0.2473, simple_loss=0.3382, pruned_loss=0.07813, over 28916.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.335, pruned_loss=0.08562, over 5680874.72 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3383, pruned_loss=0.1054, over 5735410.98 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3344, pruned_loss=0.08271, over 5664754.59 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:02:46,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3296, 3.1379, 1.4195, 1.5217], device='cuda:0'), covar=tensor([0.1012, 0.0348, 0.0983, 0.1360], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0567, 0.0410, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 16:02:52,901 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-14 16:03:14,436 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1265280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:03:18,992 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1265283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:03:24,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1265287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:03:29,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1265290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:03:38,434 INFO [train.py:968] (0/2) Epoch 28, batch 34400, libri_loss[loss=0.3066, simple_loss=0.3628, pruned_loss=0.1252, over 29541.00 frames. ], tot_loss[loss=0.255, simple_loss=0.337, pruned_loss=0.0865, over 5678332.22 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3389, pruned_loss=0.1058, over 5731234.26 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3358, pruned_loss=0.08302, over 5667642.04 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:03:44,758 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.025e+02 1.563e+03 2.085e+03 2.897e+03 7.316e+03, threshold=4.169e+03, percent-clipped=15.0
+2023-03-14 16:03:56,722 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1265312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:04:08,123 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1265319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:04:36,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1645, 1.2911, 1.1301, 0.9122], device='cuda:0'), covar=tensor([0.1069, 0.0526, 0.1162, 0.1136], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0446, 0.0521, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 16:04:47,124 INFO [train.py:968] (0/2) Epoch 28, batch 34450, giga_loss[loss=0.2656, simple_loss=0.3426, pruned_loss=0.09432, over 28173.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3372, pruned_loss=0.0877, over 5677971.09 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3391, pruned_loss=0.1061, over 5733715.98 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3361, pruned_loss=0.08445, over 5666350.99 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:05:57,780 INFO [train.py:968] (0/2) Epoch 28, batch 34500, giga_loss[loss=0.2381, simple_loss=0.3154, pruned_loss=0.0804, over 24610.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3349, pruned_loss=0.08636, over 5688764.15 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3392, pruned_loss=0.106, over 5736305.49 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.334, pruned_loss=0.0836, over 5676719.86 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:06:07,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.280e+02 1.508e+03 2.056e+03 3.161e+03 6.856e+03, threshold=4.112e+03, percent-clipped=12.0
+2023-03-14 16:07:05,463 INFO [train.py:968] (0/2) Epoch 28, batch 34550, giga_loss[loss=0.2338, simple_loss=0.3193, pruned_loss=0.07413, over 28922.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3335, pruned_loss=0.08495, over 5682917.40 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3394, pruned_loss=0.1061, over 5730451.36 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3324, pruned_loss=0.08214, over 5678105.50 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:08:09,244 INFO [train.py:968] (0/2) Epoch 28, batch 34600, giga_loss[loss=0.2333, simple_loss=0.3236, pruned_loss=0.0715, over 28880.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3328, pruned_loss=0.08454, over 5673547.65 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3392, pruned_loss=0.106, over 5730952.64 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3319, pruned_loss=0.08197, over 5667989.08 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:08:16,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.324e+02 1.308e+03 1.797e+03 2.421e+03 8.404e+03, threshold=3.595e+03, percent-clipped=8.0
+2023-03-14 16:08:20,534 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3976, 1.5563, 1.5925, 1.2323], device='cuda:0'), covar=tensor([0.1650, 0.2593, 0.1438, 0.1791], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0710, 0.0981, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 16:08:23,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2815, 1.6591, 1.5367, 1.5169], device='cuda:0'), covar=tensor([0.2111, 0.2089, 0.2290, 0.2046], device='cuda:0'), in_proj_covar=tensor([0.0489, 0.0745, 0.0717, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 16:09:07,574 INFO [train.py:968] (0/2) Epoch 28, batch 34650, giga_loss[loss=0.2496, simple_loss=0.3356, pruned_loss=0.08175, over 28891.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.335, pruned_loss=0.08622, over 5666368.47 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3392, pruned_loss=0.1062, over 5725315.58 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3342, pruned_loss=0.08349, over 5665757.94 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:09:33,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1265571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:10:00,331 INFO [train.py:968] (0/2) Epoch 28, batch 34700, libri_loss[loss=0.2718, simple_loss=0.3367, pruned_loss=0.1035, over 29766.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3369, pruned_loss=0.08767, over 5682877.14 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.339, pruned_loss=0.106, over 5732386.82 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3363, pruned_loss=0.08472, over 5673136.15 frames. ], batch size: 87, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:10:05,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.835e+02 1.661e+03 2.049e+03 2.847e+03 7.986e+03, threshold=4.098e+03, percent-clipped=17.0
+2023-03-14 16:10:43,071 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1265638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:10:55,467 INFO [train.py:968] (0/2) Epoch 28, batch 34750, giga_loss[loss=0.2232, simple_loss=0.3058, pruned_loss=0.0703, over 28929.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3341, pruned_loss=0.08733, over 5678770.07 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3388, pruned_loss=0.1059, over 5738341.58 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3337, pruned_loss=0.08433, over 5663809.97 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:11:28,159 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-14 16:11:45,771 INFO [train.py:968] (0/2) Epoch 28, batch 34800, giga_loss[loss=0.2778, simple_loss=0.3558, pruned_loss=0.09992, over 28368.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.334, pruned_loss=0.08815, over 5667860.08 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3391, pruned_loss=0.1059, over 5729335.78 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3333, pruned_loss=0.0851, over 5661023.71 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:11:53,459 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.629e+02 1.653e+03 2.133e+03 2.989e+03 8.942e+03, threshold=4.265e+03, percent-clipped=8.0
+2023-03-14 16:12:02,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6082, 2.0345, 1.5974, 1.8794], device='cuda:0'), covar=tensor([0.0753, 0.0263, 0.0331, 0.0853], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 16:12:05,558 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1265714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:12:08,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1265717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:12:36,874 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1265746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:12:39,684 INFO [train.py:968] (0/2) Epoch 28, batch 34850, libri_loss[loss=0.2907, simple_loss=0.3584, pruned_loss=0.1115, over 29183.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3334, pruned_loss=0.08776, over 5680280.48 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3388, pruned_loss=0.1056, over 5734338.59 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3329, pruned_loss=0.08503, over 5668361.16 frames. ], batch size: 101, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:13:26,702 INFO [train.py:968] (0/2) Epoch 28, batch 34900, giga_loss[loss=0.2663, simple_loss=0.3541, pruned_loss=0.08921, over 28154.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3406, pruned_loss=0.09192, over 5674290.47 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3384, pruned_loss=0.1053, over 5730342.79 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3405, pruned_loss=0.08949, over 5666029.55 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:13:31,915 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.066e+03 1.591e+03 2.127e+03 3.064e+03 1.007e+04, threshold=4.255e+03, percent-clipped=6.0
+2023-03-14 16:13:58,573 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1265833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:14:14,509 INFO [train.py:968] (0/2) Epoch 28, batch 34950, giga_loss[loss=0.3281, simple_loss=0.4067, pruned_loss=0.1248, over 28879.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3487, pruned_loss=0.09655, over 5677988.14 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3386, pruned_loss=0.1055, over 5734961.02 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3486, pruned_loss=0.09423, over 5665873.21 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:14:56,988 INFO [train.py:968] (0/2) Epoch 28, batch 35000, giga_loss[loss=0.2591, simple_loss=0.3405, pruned_loss=0.08886, over 28871.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3511, pruned_loss=0.09874, over 5677413.03 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3387, pruned_loss=0.1054, over 5734932.02 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3513, pruned_loss=0.09665, over 5666476.37 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:15:04,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.285e+02 1.424e+03 1.790e+03 2.212e+03 4.736e+03, threshold=3.580e+03, percent-clipped=3.0
+2023-03-14 16:15:39,513 INFO [train.py:968] (0/2) Epoch 28, batch 35050, giga_loss[loss=0.236, simple_loss=0.3128, pruned_loss=0.07963, over 28966.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3456, pruned_loss=0.09641, over 5687775.59 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.339, pruned_loss=0.1056, over 5736302.95 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3456, pruned_loss=0.09449, over 5676926.62 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:16:18,838 INFO [train.py:968] (0/2) Epoch 28, batch 35100, libri_loss[loss=0.2576, simple_loss=0.3249, pruned_loss=0.09513, over 29572.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3389, pruned_loss=0.09355, over 5697205.29 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3386, pruned_loss=0.1052, over 5742114.80 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3393, pruned_loss=0.09194, over 5681784.51 frames. ], batch size: 75, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:16:19,507 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1266000.pt
+2023-03-14 16:16:24,301 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.587e+02 1.259e+03 1.585e+03 2.080e+03 1.043e+04, threshold=3.169e+03, percent-clipped=4.0
+2023-03-14 16:16:29,080 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8404, 1.1942, 2.8741, 2.8530], device='cuda:0'), covar=tensor([0.1763, 0.2662, 0.0606, 0.0963], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0679, 0.1008, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 16:16:29,761 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1266013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:16:59,399 INFO [train.py:968] (0/2) Epoch 28, batch 35150, giga_loss[loss=0.2149, simple_loss=0.2912, pruned_loss=0.06936, over 28750.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3328, pruned_loss=0.09133, over 5695722.64 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3387, pruned_loss=0.1053, over 5745086.57 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.333, pruned_loss=0.0897, over 5679556.54 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:17:31,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4572, 4.3096, 4.0664, 1.9627], device='cuda:0'), covar=tensor([0.0549, 0.0731, 0.0746, 0.2056], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.1190, 0.1004, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 16:17:38,101 INFO [train.py:968] (0/2) Epoch 28, batch 35200, giga_loss[loss=0.1896, simple_loss=0.2721, pruned_loss=0.05354, over 28869.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3263, pruned_loss=0.08861, over 5694224.95 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3393, pruned_loss=0.1055, over 5741317.47 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3255, pruned_loss=0.08648, over 5682871.22 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:17:40,371 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-14 16:17:44,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.530e+02 1.175e+03 1.542e+03 2.038e+03 6.038e+03, threshold=3.084e+03, percent-clipped=8.0
+2023-03-14 16:18:20,717 INFO [train.py:968] (0/2) Epoch 28, batch 35250, giga_loss[loss=0.2044, simple_loss=0.2836, pruned_loss=0.06257, over 29003.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3194, pruned_loss=0.08559, over 5694085.71 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3391, pruned_loss=0.1053, over 5744534.29 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3187, pruned_loss=0.08372, over 5681372.15 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:18:21,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7597, 4.5941, 4.3432, 2.2090], device='cuda:0'), covar=tensor([0.0454, 0.0643, 0.0688, 0.2076], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.1190, 0.1002, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 16:18:25,825 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1266156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:18:28,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1266159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:18:55,072 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1266188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:19:03,649 INFO [train.py:968] (0/2) Epoch 28, batch 35300, giga_loss[loss=0.2015, simple_loss=0.2782, pruned_loss=0.06241, over 29029.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3153, pruned_loss=0.08387, over 5685419.68 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3394, pruned_loss=0.1056, over 5737801.04 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.3139, pruned_loss=0.08173, over 5680243.69 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:19:09,725 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.399e+02 1.205e+03 1.607e+03 2.093e+03 7.633e+03, threshold=3.215e+03, percent-clipped=9.0
+2023-03-14 16:19:10,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1266208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:19:44,998 INFO [train.py:968] (0/2) Epoch 28, batch 35350, giga_loss[loss=0.2056, simple_loss=0.2808, pruned_loss=0.06518, over 28997.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3135, pruned_loss=0.08299, over 5667872.87 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3399, pruned_loss=0.1057, over 5711780.13 frames. ], giga_tot_loss[loss=0.2358, simple_loss=0.3111, pruned_loss=0.0803, over 5685336.46 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:19:57,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9236, 2.2955, 2.0016, 1.9423], device='cuda:0'), covar=tensor([0.2254, 0.2279, 0.2384, 0.2389], device='cuda:0'), in_proj_covar=tensor([0.0495, 0.0753, 0.0722, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 16:20:26,429 INFO [train.py:968] (0/2) Epoch 28, batch 35400, giga_loss[loss=0.2113, simple_loss=0.2842, pruned_loss=0.0692, over 27723.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3096, pruned_loss=0.08066, over 5675487.94 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.34, pruned_loss=0.1056, over 5706090.35 frames. ], giga_tot_loss[loss=0.2317, simple_loss=0.3071, pruned_loss=0.07819, over 5692901.70 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:20:31,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.068e+02 1.154e+03 1.426e+03 1.966e+03 5.335e+03, threshold=2.853e+03, percent-clipped=5.0
+2023-03-14 16:20:41,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9586, 1.1422, 1.1124, 0.9522], device='cuda:0'), covar=tensor([0.2706, 0.2936, 0.1693, 0.2424], device='cuda:0'), in_proj_covar=tensor([0.2032, 0.1983, 0.1881, 0.2036], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 16:20:45,395 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1266322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:21:10,102 INFO [train.py:968] (0/2) Epoch 28, batch 35450, giga_loss[loss=0.1777, simple_loss=0.2599, pruned_loss=0.04777, over 28554.00 frames. ], tot_loss[loss=0.233, simple_loss=0.307, pruned_loss=0.0795, over 5682274.88 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3404, pruned_loss=0.1059, over 5701532.64 frames. ], giga_tot_loss[loss=0.2286, simple_loss=0.3038, pruned_loss=0.07669, over 5699844.67 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:21:11,739 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1266351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:21:13,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1266354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:21:19,800 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9042, 3.7617, 3.5430, 1.8617], device='cuda:0'), covar=tensor([0.0652, 0.0823, 0.0790, 0.2354], device='cuda:0'), in_proj_covar=tensor([0.1293, 0.1189, 0.1002, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 16:21:38,301 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1266383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:21:48,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-14 16:21:50,806 INFO [train.py:968] (0/2) Epoch 28, batch 35500, giga_loss[loss=0.2065, simple_loss=0.273, pruned_loss=0.06999, over 28513.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3036, pruned_loss=0.07779, over 5685449.50 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3404, pruned_loss=0.1058, over 5702801.70 frames. ], giga_tot_loss[loss=0.2257, simple_loss=0.3007, pruned_loss=0.0753, over 5698033.73 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:21:53,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4430, 1.6601, 1.3082, 1.1769], device='cuda:0'), covar=tensor([0.1037, 0.0526, 0.1052, 0.1175], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0446, 0.0523, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 16:21:58,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.635e+02 1.134e+03 1.629e+03 2.101e+03 1.040e+04, threshold=3.258e+03, percent-clipped=14.0
+2023-03-14 16:22:16,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-14 16:22:31,702 INFO [train.py:968] (0/2) Epoch 28, batch 35550, giga_loss[loss=0.1937, simple_loss=0.274, pruned_loss=0.05665, over 28873.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3014, pruned_loss=0.07678, over 5685066.35 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3406, pruned_loss=0.1058, over 5703814.90 frames. ], giga_tot_loss[loss=0.223, simple_loss=0.2979, pruned_loss=0.07403, over 5694131.54 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:22:34,130 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-14 16:23:13,405 INFO [train.py:968] (0/2) Epoch 28, batch 35600, giga_loss[loss=0.2093, simple_loss=0.2807, pruned_loss=0.06896, over 28720.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3003, pruned_loss=0.07638, over 5687944.60 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3415, pruned_loss=0.1061, over 5710776.59 frames. ], giga_tot_loss[loss=0.2204, simple_loss=0.2952, pruned_loss=0.07286, over 5688130.81 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:23:20,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.411e+02 1.114e+03 1.494e+03 1.917e+03 6.314e+03, threshold=2.988e+03, percent-clipped=7.0
+2023-03-14 16:23:28,201 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3240, 1.2220, 3.8358, 3.3658], device='cuda:0'), covar=tensor([0.1686, 0.3085, 0.0464, 0.1686], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0677, 0.1008, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 16:23:55,362 INFO [train.py:968] (0/2) Epoch 28, batch 35650, giga_loss[loss=0.1826, simple_loss=0.256, pruned_loss=0.05459, over 28577.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2973, pruned_loss=0.07501, over 5699149.08 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3417, pruned_loss=0.1061, over 5716417.85 frames. ], giga_tot_loss[loss=0.2172, simple_loss=0.2918, pruned_loss=0.07127, over 5693412.12 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:24:38,340 INFO [train.py:968] (0/2) Epoch 28, batch 35700, giga_loss[loss=0.3256, simple_loss=0.3749, pruned_loss=0.1382, over 28202.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.2975, pruned_loss=0.07595, over 5688992.24 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3419, pruned_loss=0.1063, over 5710197.42 frames. ], giga_tot_loss[loss=0.2178, simple_loss=0.2917, pruned_loss=0.07199, over 5690447.61 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:24:40,544 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1266602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:24:46,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.683e+02 1.165e+03 1.607e+03 2.081e+03 4.718e+03, threshold=3.214e+03, percent-clipped=12.0
+2023-03-14 16:24:55,610 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9058, 3.1472, 2.0606, 1.0039], device='cuda:0'), covar=tensor([0.8747, 0.3027, 0.3651, 0.6489], device='cuda:0'), in_proj_covar=tensor([0.1856, 0.1747, 0.1663, 0.1510], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 16:25:27,260 INFO [train.py:968] (0/2) Epoch 28, batch 35750, giga_loss[loss=0.2801, simple_loss=0.3527, pruned_loss=0.1037, over 28746.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3068, pruned_loss=0.08065, over 5684742.42 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3418, pruned_loss=0.1062, over 5712281.89 frames. ], giga_tot_loss[loss=0.2283, simple_loss=0.302, pruned_loss=0.07732, over 5684015.55 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:25:29,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7101, 1.7397, 1.9150, 1.4668], device='cuda:0'), covar=tensor([0.1855, 0.2541, 0.1552, 0.1835], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0715, 0.0990, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 16:25:45,079 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-14 16:25:52,036 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-14 16:26:10,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1266697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:26:11,949 INFO [train.py:968] (0/2) Epoch 28, batch 35800, giga_loss[loss=0.2633, simple_loss=0.343, pruned_loss=0.09182, over 28536.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3208, pruned_loss=0.08827, over 5687895.07 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3422, pruned_loss=0.1065, over 5715553.19 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3161, pruned_loss=0.08498, over 5683950.14 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:26:20,553 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.396e+03 1.666e+03 2.354e+03 6.801e+03, threshold=3.331e+03, percent-clipped=12.0
+2023-03-14 16:26:55,645 INFO [train.py:968] (0/2) Epoch 28, batch 35850, libri_loss[loss=0.2554, simple_loss=0.3287, pruned_loss=0.09104, over 29579.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3326, pruned_loss=0.09394, over 5696988.91 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.342, pruned_loss=0.1064, over 5717846.21 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.329, pruned_loss=0.0913, over 5691362.51 frames. ], batch size: 76, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:27:03,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4320, 3.3003, 1.6198, 1.4801], device='cuda:0'), covar=tensor([0.1015, 0.0334, 0.0910, 0.1395], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0566, 0.0410, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 16:27:14,625 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 16:27:39,465 INFO [train.py:968] (0/2) Epoch 28, batch 35900, giga_loss[loss=0.2491, simple_loss=0.3427, pruned_loss=0.07781, over 28922.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3381, pruned_loss=0.09533, over 5693405.52 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.343, pruned_loss=0.107, over 5717337.56 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3341, pruned_loss=0.09249, over 5689211.98 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:27:46,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.500e+03 1.966e+03 2.685e+03 7.582e+03, threshold=3.933e+03, percent-clipped=14.0
+2023-03-14 16:27:50,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7438, 1.9494, 1.8282, 1.6984], device='cuda:0'), covar=tensor([0.2002, 0.2102, 0.2315, 0.2165], device='cuda:0'), in_proj_covar=tensor([0.0498, 0.0754, 0.0725, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 16:28:10,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5353, 2.5824, 2.4383, 2.3497], device='cuda:0'), covar=tensor([0.2246, 0.2574, 0.2381, 0.2254], device='cuda:0'), in_proj_covar=tensor([0.0498, 0.0755, 0.0725, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 16:28:12,201 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1266840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:28:14,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1266843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:28:21,248 INFO [train.py:968] (0/2) Epoch 28, batch 35950, giga_loss[loss=0.2736, simple_loss=0.3556, pruned_loss=0.09583, over 27823.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3389, pruned_loss=0.09417, over 5688144.86 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3428, pruned_loss=0.1067, over 5714112.92 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3358, pruned_loss=0.09167, over 5686825.94 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:28:44,456 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1266872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:29:08,719 INFO [train.py:968] (0/2) Epoch 28, batch 36000, giga_loss[loss=0.2837, simple_loss=0.3541, pruned_loss=0.1067, over 28908.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3414, pruned_loss=0.09486, over 5685120.35 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3431, pruned_loss=0.1068, over 5714276.27 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3386, pruned_loss=0.0926, over 5683536.88 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:29:08,723 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 16:29:16,932 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.199, simple_loss=0.3055, pruned_loss=0.04625, over 944034.00 frames. 
+2023-03-14 16:29:16,933 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 16:29:23,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.634e+02 1.417e+03 1.760e+03 2.205e+03 5.195e+03, threshold=3.521e+03, percent-clipped=2.0
+2023-03-14 16:29:55,394 INFO [train.py:968] (0/2) Epoch 28, batch 36050, giga_loss[loss=0.2635, simple_loss=0.3463, pruned_loss=0.09035, over 29001.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.344, pruned_loss=0.09689, over 5695721.99 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3429, pruned_loss=0.1066, over 5719767.84 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3419, pruned_loss=0.09496, over 5688966.07 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:30:19,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1266977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:30:37,691 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-14 16:30:39,364 INFO [train.py:968] (0/2) Epoch 28, batch 36100, giga_loss[loss=0.3277, simple_loss=0.3911, pruned_loss=0.1321, over 28849.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3472, pruned_loss=0.09934, over 5677855.53 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3433, pruned_loss=0.1068, over 5713815.24 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3452, pruned_loss=0.09743, over 5676594.94 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:30:48,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.440e+03 1.789e+03 2.301e+03 4.810e+03, threshold=3.579e+03, percent-clipped=6.0
+2023-03-14 16:30:55,231 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-14 16:31:19,531 INFO [train.py:968] (0/2) Epoch 28, batch 36150, giga_loss[loss=0.2602, simple_loss=0.3448, pruned_loss=0.08777, over 28947.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3501, pruned_loss=0.1009, over 5691265.24 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3435, pruned_loss=0.1068, over 5718803.70 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3485, pruned_loss=0.09919, over 5684923.02 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:31:58,419 INFO [train.py:968] (0/2) Epoch 28, batch 36200, libri_loss[loss=0.2986, simple_loss=0.3715, pruned_loss=0.1128, over 29475.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3522, pruned_loss=0.101, over 5696148.52 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3433, pruned_loss=0.1065, over 5721601.67 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3512, pruned_loss=0.09972, over 5688005.68 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:32:05,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.612e+02 1.357e+03 1.672e+03 2.411e+03 6.176e+03, threshold=3.343e+03, percent-clipped=7.0
+2023-03-14 16:32:15,649 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1267120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:32:18,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1267123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:32:35,210 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2733, 1.3423, 3.6554, 3.1932], device='cuda:0'), covar=tensor([0.1721, 0.2872, 0.0456, 0.1271], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0675, 0.1002, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 16:32:40,424 INFO [train.py:968] (0/2) Epoch 28, batch 36250, libri_loss[loss=0.2373, simple_loss=0.3065, pruned_loss=0.08408, over 29348.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3537, pruned_loss=0.1012, over 5696872.91 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3437, pruned_loss=0.1066, over 5725720.60 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3528, pruned_loss=0.09998, over 5685968.88 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:32:42,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1267152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:32:45,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4147, 2.0105, 1.4287, 0.6810], device='cuda:0'), covar=tensor([0.7354, 0.3764, 0.5028, 0.7363], device='cuda:0'), in_proj_covar=tensor([0.1853, 0.1744, 0.1663, 0.1506], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 16:32:56,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4706, 1.3675, 1.5862, 1.2067], device='cuda:0'), covar=tensor([0.1754, 0.3181, 0.1488, 0.1764], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0714, 0.0988, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 16:33:19,257 INFO [train.py:968] (0/2) Epoch 28, batch 36300, giga_loss[loss=0.2969, simple_loss=0.3728, pruned_loss=0.1105, over 28404.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3544, pruned_loss=0.1006, over 5695680.76 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3437, pruned_loss=0.1064, over 5728721.83 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.354, pruned_loss=0.09958, over 5683364.52 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:33:28,404 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.469e+02 1.337e+03 1.751e+03 2.277e+03 1.033e+04, threshold=3.503e+03, percent-clipped=10.0
+2023-03-14 16:33:59,670 INFO [train.py:968] (0/2) Epoch 28, batch 36350, giga_loss[loss=0.286, simple_loss=0.3621, pruned_loss=0.1049, over 28778.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.354, pruned_loss=0.09909, over 5705235.52 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3438, pruned_loss=0.1065, over 5730860.84 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3538, pruned_loss=0.0981, over 5693026.54 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:34:04,462 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 16:34:28,468 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-14 16:34:33,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8656, 1.3083, 1.0624, 0.2515], device='cuda:0'), covar=tensor([0.5190, 0.3282, 0.4721, 0.6913], device='cuda:0'), in_proj_covar=tensor([0.1851, 0.1741, 0.1662, 0.1504], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 16:34:41,044 INFO [train.py:968] (0/2) Epoch 28, batch 36400, giga_loss[loss=0.2787, simple_loss=0.3547, pruned_loss=0.1014, over 27977.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3522, pruned_loss=0.09696, over 5700466.16 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.344, pruned_loss=0.1067, over 5724867.70 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.352, pruned_loss=0.09587, over 5696100.44 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:34:49,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.256e+03 1.628e+03 2.398e+03 6.991e+03, threshold=3.256e+03, percent-clipped=11.0
+2023-03-14 16:34:52,412 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1267313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:35:23,192 INFO [train.py:968] (0/2) Epoch 28, batch 36450, giga_loss[loss=0.2716, simple_loss=0.3534, pruned_loss=0.09485, over 28760.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3498, pruned_loss=0.09504, over 5708921.93 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.344, pruned_loss=0.1067, over 5726742.78 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3499, pruned_loss=0.09408, over 5703540.49 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:36:06,319 INFO [train.py:968] (0/2) Epoch 28, batch 36500, libri_loss[loss=0.3141, simple_loss=0.3865, pruned_loss=0.1208, over 29769.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3524, pruned_loss=0.09843, over 5707976.22 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3447, pruned_loss=0.107, over 5728888.43 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3519, pruned_loss=0.09721, over 5701116.93 frames. ], batch size: 87, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:36:15,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.837e+02 1.262e+03 1.760e+03 2.455e+03 4.647e+03, threshold=3.520e+03, percent-clipped=8.0
+2023-03-14 16:36:46,783 INFO [train.py:968] (0/2) Epoch 28, batch 36550, giga_loss[loss=0.2899, simple_loss=0.3604, pruned_loss=0.1097, over 28773.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3562, pruned_loss=0.1035, over 5702183.29 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3451, pruned_loss=0.107, over 5729443.62 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3556, pruned_loss=0.1024, over 5695544.41 frames. ], batch size: 243, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:36:52,004 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-14 16:37:26,585 INFO [train.py:968] (0/2) Epoch 28, batch 36600, giga_loss[loss=0.2695, simple_loss=0.3362, pruned_loss=0.1014, over 28925.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3552, pruned_loss=0.1044, over 5705994.61 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.345, pruned_loss=0.107, over 5735445.56 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3553, pruned_loss=0.1034, over 5693653.81 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:37:37,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.923e+02 1.533e+03 1.851e+03 2.340e+03 6.974e+03, threshold=3.702e+03, percent-clipped=5.0
+2023-03-14 16:38:08,620 INFO [train.py:968] (0/2) Epoch 28, batch 36650, giga_loss[loss=0.2654, simple_loss=0.3362, pruned_loss=0.09731, over 28290.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3534, pruned_loss=0.104, over 5709232.10 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.345, pruned_loss=0.1068, over 5731077.66 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3538, pruned_loss=0.1031, over 5702587.95 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:38:16,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4726, 4.2660, 4.0554, 2.0460], device='cuda:0'), covar=tensor([0.0663, 0.0827, 0.0789, 0.2181], device='cuda:0'), in_proj_covar=tensor([0.1289, 0.1192, 0.1002, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 16:38:29,153 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-14 16:38:50,658 INFO [train.py:968] (0/2) Epoch 28, batch 36700, giga_loss[loss=0.273, simple_loss=0.348, pruned_loss=0.09901, over 29099.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3507, pruned_loss=0.1027, over 5697759.08 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3453, pruned_loss=0.1069, over 5723822.46 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.351, pruned_loss=0.102, over 5697978.24 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:38:59,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.544e+02 1.320e+03 1.732e+03 2.314e+03 6.391e+03, threshold=3.465e+03, percent-clipped=4.0
+2023-03-14 16:39:32,289 INFO [train.py:968] (0/2) Epoch 28, batch 36750, libri_loss[loss=0.2791, simple_loss=0.3513, pruned_loss=0.1035, over 26270.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3497, pruned_loss=0.1014, over 5695727.31 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3458, pruned_loss=0.1071, over 5720576.78 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3496, pruned_loss=0.1005, over 5698447.04 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:39:35,604 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 16:40:04,889 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1267688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:40:14,245 INFO [train.py:968] (0/2) Epoch 28, batch 36800, giga_loss[loss=0.2352, simple_loss=0.3161, pruned_loss=0.07709, over 28286.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3484, pruned_loss=0.1002, over 5678804.36 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3457, pruned_loss=0.107, over 5713912.29 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3484, pruned_loss=0.09938, over 5685482.62 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:40:23,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.942e+02 1.269e+03 1.559e+03 2.072e+03 4.256e+03, threshold=3.119e+03, percent-clipped=3.0
+2023-03-14 16:40:24,329 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 16:40:37,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1267726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:40:58,902 INFO [train.py:968] (0/2) Epoch 28, batch 36850, giga_loss[loss=0.2835, simple_loss=0.3295, pruned_loss=0.1187, over 23626.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3444, pruned_loss=0.0978, over 5671143.35 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3456, pruned_loss=0.1068, over 5708184.81 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3445, pruned_loss=0.09716, over 5680634.62 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:41:41,664 INFO [train.py:968] (0/2) Epoch 28, batch 36900, giga_loss[loss=0.2104, simple_loss=0.2904, pruned_loss=0.06523, over 28354.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3381, pruned_loss=0.09466, over 5670777.11 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3458, pruned_loss=0.1068, over 5714115.07 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.338, pruned_loss=0.09392, over 5672176.96 frames. ], batch size: 78, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:41:53,362 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.458e+02 1.089e+03 1.464e+03 1.820e+03 8.577e+03, threshold=2.928e+03, percent-clipped=11.0
+2023-03-14 16:42:04,018 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1267821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:42:04,197 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 16:42:17,412 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1267831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:42:19,553 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1267834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:42:33,813 INFO [train.py:968] (0/2) Epoch 28, batch 36950, giga_loss[loss=0.222, simple_loss=0.2973, pruned_loss=0.07332, over 28686.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3343, pruned_loss=0.09333, over 5656783.97 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3465, pruned_loss=0.1073, over 5714139.39 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3333, pruned_loss=0.09203, over 5656927.90 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:42:34,947 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-14 16:42:49,374 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1267863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:43:19,391 INFO [train.py:968] (0/2) Epoch 28, batch 37000, giga_loss[loss=0.2348, simple_loss=0.3162, pruned_loss=0.07672, over 29005.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3322, pruned_loss=0.09232, over 5656526.32 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3463, pruned_loss=0.107, over 5719092.36 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3313, pruned_loss=0.09124, over 5650890.93 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:43:29,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.018e+02 1.174e+03 1.513e+03 2.157e+03 1.087e+04, threshold=3.026e+03, percent-clipped=11.0
+2023-03-14 16:43:34,841 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1267916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:43:58,881 INFO [train.py:968] (0/2) Epoch 28, batch 37050, giga_loss[loss=0.2361, simple_loss=0.3191, pruned_loss=0.07654, over 28891.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3331, pruned_loss=0.09197, over 5656926.12 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.347, pruned_loss=0.1073, over 5701380.64 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3315, pruned_loss=0.09039, over 5665658.61 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:44:29,615 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-14 16:44:39,785 INFO [train.py:968] (0/2) Epoch 28, batch 37100, giga_loss[loss=0.2744, simple_loss=0.3392, pruned_loss=0.1048, over 29023.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3326, pruned_loss=0.09119, over 5670009.49 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.347, pruned_loss=0.1072, over 5705301.60 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.331, pruned_loss=0.08978, over 5672707.16 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:44:40,474 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1268000.pt
+2023-03-14 16:44:43,257 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4831, 2.3601, 2.4183, 2.0744], device='cuda:0'), covar=tensor([0.2201, 0.2718, 0.2432, 0.2722], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0757, 0.0729, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 16:44:51,274 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.893e+02 1.196e+03 1.547e+03 2.070e+03 6.675e+03, threshold=3.094e+03, percent-clipped=13.0
+2023-03-14 16:45:19,781 INFO [train.py:968] (0/2) Epoch 28, batch 37150, giga_loss[loss=0.2405, simple_loss=0.3168, pruned_loss=0.08207, over 28619.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3315, pruned_loss=0.09059, over 5678547.12 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3478, pruned_loss=0.1075, over 5700590.39 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3289, pruned_loss=0.08853, over 5684459.08 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:45:59,982 INFO [train.py:968] (0/2) Epoch 28, batch 37200, giga_loss[loss=0.2257, simple_loss=0.3071, pruned_loss=0.07217, over 28957.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3296, pruned_loss=0.08991, over 5698830.28 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3486, pruned_loss=0.108, over 5703637.65 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3266, pruned_loss=0.08754, over 5700502.60 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:46:01,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:46:10,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.236e+02 1.104e+03 1.443e+03 2.099e+03 5.772e+03, threshold=2.885e+03, percent-clipped=10.0
+2023-03-14 16:46:36,237 INFO [train.py:968] (0/2) Epoch 28, batch 37250, giga_loss[loss=0.2046, simple_loss=0.2907, pruned_loss=0.05929, over 28578.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3281, pruned_loss=0.0894, over 5695744.32 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3495, pruned_loss=0.1086, over 5699705.98 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3243, pruned_loss=0.08649, over 5699745.51 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:46:37,903 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-14 16:47:13,949 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:47:15,688 INFO [train.py:968] (0/2) Epoch 28, batch 37300, giga_loss[loss=0.2281, simple_loss=0.308, pruned_loss=0.07406, over 28873.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3249, pruned_loss=0.08796, over 5696771.48 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3498, pruned_loss=0.1087, over 5696982.45 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3215, pruned_loss=0.08547, over 5702455.16 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:47:27,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.398e+02 1.114e+03 1.404e+03 1.831e+03 7.195e+03, threshold=2.808e+03, percent-clipped=10.0
+2023-03-14 16:47:53,541 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:47:55,559 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:47:56,608 INFO [train.py:968] (0/2) Epoch 28, batch 37350, giga_loss[loss=0.2365, simple_loss=0.3119, pruned_loss=0.08061, over 29043.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3237, pruned_loss=0.0875, over 5698141.39 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3506, pruned_loss=0.1092, over 5685442.24 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3202, pruned_loss=0.08494, over 5713258.69 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:48:17,473 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:48:18,337 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-14 16:48:29,852 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:48:35,401 INFO [train.py:968] (0/2) Epoch 28, batch 37400, libri_loss[loss=0.2665, simple_loss=0.3413, pruned_loss=0.09581, over 29558.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3226, pruned_loss=0.08649, over 5704186.51 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3516, pruned_loss=0.1092, over 5688483.16 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3179, pruned_loss=0.0836, over 5714055.36 frames. ], batch size: 75, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:48:46,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.833e+02 1.176e+03 1.460e+03 2.156e+03 1.288e+04, threshold=2.920e+03, percent-clipped=18.0
+2023-03-14 16:48:57,801 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.37 vs. limit=5.0
+2023-03-14 16:49:06,748 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:49:08,676 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:49:12,988 INFO [train.py:968] (0/2) Epoch 28, batch 37450, giga_loss[loss=0.2369, simple_loss=0.3086, pruned_loss=0.08262, over 28961.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3216, pruned_loss=0.0861, over 5707305.09 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3523, pruned_loss=0.1092, over 5694819.36 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3159, pruned_loss=0.08285, over 5709900.70 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:49:29,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:49:40,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1164, 2.7769, 1.2530, 1.3658], device='cuda:0'), covar=tensor([0.1324, 0.0427, 0.1085, 0.1573], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0565, 0.0409, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 16:49:51,771 INFO [train.py:968] (0/2) Epoch 28, batch 37500, giga_loss[loss=0.2535, simple_loss=0.3323, pruned_loss=0.08733, over 29010.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3205, pruned_loss=0.08512, over 5714453.82 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3525, pruned_loss=0.1092, over 5697808.10 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3152, pruned_loss=0.08213, over 5714069.02 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:50:03,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.510e+02 1.157e+03 1.474e+03 1.910e+03 9.917e+03, threshold=2.947e+03, percent-clipped=8.0
+2023-03-14 16:50:05,352 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1268415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:50:14,635 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1268426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:50:21,545 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:50:23,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:50:26,374 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1268440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:50:33,277 INFO [train.py:968] (0/2) Epoch 28, batch 37550, giga_loss[loss=0.2477, simple_loss=0.3212, pruned_loss=0.08711, over 28857.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3189, pruned_loss=0.08434, over 5701146.56 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3526, pruned_loss=0.1093, over 5690450.15 frames. ], giga_tot_loss[loss=0.2389, simple_loss=0.3143, pruned_loss=0.0817, over 5707764.92 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:50:36,374 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4925, 3.4853, 1.5011, 1.5863], device='cuda:0'), covar=tensor([0.1042, 0.0321, 0.0944, 0.1352], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0564, 0.0409, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 16:50:39,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1281, 1.3071, 1.1268, 0.8769], device='cuda:0'), covar=tensor([0.1087, 0.0493, 0.1031, 0.1183], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0450, 0.0526, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 16:50:47,655 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:50:52,972 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1268472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:51:14,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4294, 1.5455, 1.6370, 1.2355], device='cuda:0'), covar=tensor([0.1645, 0.2575, 0.1398, 0.1673], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0715, 0.0987, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 16:51:15,700 INFO [train.py:968] (0/2) Epoch 28, batch 37600, giga_loss[loss=0.2302, simple_loss=0.3095, pruned_loss=0.0755, over 28809.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3211, pruned_loss=0.0859, over 5708412.35 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3532, pruned_loss=0.1095, over 5686589.31 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.3162, pruned_loss=0.08305, over 5716684.42 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:51:28,043 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.602e+02 1.202e+03 1.481e+03 2.025e+03 5.252e+03, threshold=2.963e+03, percent-clipped=9.0
+2023-03-14 16:51:58,602 INFO [train.py:968] (0/2) Epoch 28, batch 37650, giga_loss[loss=0.2631, simple_loss=0.3357, pruned_loss=0.09522, over 28795.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3258, pruned_loss=0.08875, over 5707290.71 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3533, pruned_loss=0.1094, over 5688985.22 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3215, pruned_loss=0.08629, over 5711976.30 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:52:48,293 INFO [train.py:968] (0/2) Epoch 28, batch 37700, giga_loss[loss=0.3476, simple_loss=0.4155, pruned_loss=0.1398, over 28501.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.335, pruned_loss=0.09499, over 5698977.94 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3536, pruned_loss=0.1094, over 5693083.42 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3308, pruned_loss=0.09271, over 5699377.84 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:52:58,593 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.715e+02 1.382e+03 1.753e+03 2.100e+03 7.284e+03, threshold=3.506e+03, percent-clipped=5.0
+2023-03-14 16:53:22,060 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7300, 1.7881, 1.8264, 1.6518], device='cuda:0'), covar=tensor([0.2410, 0.2724, 0.1969, 0.2385], device='cuda:0'), in_proj_covar=tensor([0.2046, 0.2012, 0.1913, 0.2067], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 16:53:34,324 INFO [train.py:968] (0/2) Epoch 28, batch 37750, giga_loss[loss=0.249, simple_loss=0.3383, pruned_loss=0.07984, over 28970.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3405, pruned_loss=0.09814, over 5693416.96 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3534, pruned_loss=0.1093, over 5695150.05 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3372, pruned_loss=0.09623, over 5692074.85 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:54:22,488 INFO [train.py:968] (0/2) Epoch 28, batch 37800, libri_loss[loss=0.2968, simple_loss=0.3698, pruned_loss=0.1119, over 29532.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3449, pruned_loss=0.09943, over 5681121.00 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3538, pruned_loss=0.1095, over 5689191.11 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3417, pruned_loss=0.0976, over 5684943.49 frames. ], batch size: 89, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:54:33,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.350e+02 1.221e+03 1.572e+03 2.082e+03 7.399e+03, threshold=3.143e+03, percent-clipped=6.0
+2023-03-14 16:55:05,353 INFO [train.py:968] (0/2) Epoch 28, batch 37850, giga_loss[loss=0.2856, simple_loss=0.3614, pruned_loss=0.1049, over 28387.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3495, pruned_loss=0.1016, over 5676396.30 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3537, pruned_loss=0.1094, over 5684562.59 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3468, pruned_loss=0.09994, over 5682650.63 frames. ], batch size: 65, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:55:41,246 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:55:42,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3369, 3.1796, 2.9973, 1.3675], device='cuda:0'), covar=tensor([0.0951, 0.1059, 0.0930, 0.2382], device='cuda:0'), in_proj_covar=tensor([0.1287, 0.1186, 0.0999, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 16:55:49,115 INFO [train.py:968] (0/2) Epoch 28, batch 37900, giga_loss[loss=0.2701, simple_loss=0.3489, pruned_loss=0.09562, over 28811.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3551, pruned_loss=0.1051, over 5673351.59 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3541, pruned_loss=0.1097, over 5677551.95 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3526, pruned_loss=0.1034, over 5684473.23 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:55:50,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:55:59,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4260, 1.3707, 1.1225, 1.4605], device='cuda:0'), covar=tensor([0.0731, 0.0339, 0.0354, 0.0873], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 16:56:00,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.001e+02 1.520e+03 2.036e+03 3.172e+03 1.011e+04, threshold=4.071e+03, percent-clipped=27.0
+2023-03-14 16:56:02,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:56:26,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:56:28,500 INFO [train.py:968] (0/2) Epoch 28, batch 37950, giga_loss[loss=0.2473, simple_loss=0.3339, pruned_loss=0.08031, over 28861.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3521, pruned_loss=0.1027, over 5678017.36 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3544, pruned_loss=0.1098, over 5682470.65 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3497, pruned_loss=0.1011, over 5682475.15 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:57:08,425 INFO [train.py:968] (0/2) Epoch 28, batch 38000, giga_loss[loss=0.2437, simple_loss=0.3256, pruned_loss=0.08091, over 28680.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3486, pruned_loss=0.09955, over 5690330.62 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3542, pruned_loss=0.1099, over 5688169.59 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3468, pruned_loss=0.09799, over 5688901.65 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:57:10,789 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1268901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:18,790 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.849e+02 1.299e+03 1.689e+03 2.466e+03 5.935e+03, threshold=3.378e+03, percent-clipped=9.0
+2023-03-14 16:57:35,000 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:37,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:38,823 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-14 16:57:44,907 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:47,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:48,947 INFO [train.py:968] (0/2) Epoch 28, batch 38050, giga_loss[loss=0.3224, simple_loss=0.3809, pruned_loss=0.132, over 26748.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3471, pruned_loss=0.09807, over 5693718.97 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.354, pruned_loss=0.1096, over 5689791.60 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3457, pruned_loss=0.09676, over 5690827.86 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:57:55,703 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:58,315 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:01,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:10,305 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:20,542 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:20,615 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:23,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:28,221 INFO [train.py:968] (0/2) Epoch 28, batch 38100, giga_loss[loss=0.2635, simple_loss=0.3449, pruned_loss=0.09107, over 28384.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.347, pruned_loss=0.09763, over 5683420.39 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3546, pruned_loss=0.1101, over 5676420.06 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3453, pruned_loss=0.0959, over 5692914.99 frames. ], batch size: 65, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:58:42,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.653e+02 1.339e+03 1.593e+03 2.098e+03 8.031e+03, threshold=3.186e+03, percent-clipped=8.0
+2023-03-14 16:58:49,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1269022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:59:10,925 INFO [train.py:968] (0/2) Epoch 28, batch 38150, giga_loss[loss=0.2603, simple_loss=0.3422, pruned_loss=0.08923, over 28781.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3499, pruned_loss=0.09923, over 5693882.94 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3547, pruned_loss=0.1102, over 5680658.20 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3483, pruned_loss=0.09752, over 5697755.86 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:59:30,403 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 16:59:36,851 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4245, 1.2488, 4.2107, 3.3862], device='cuda:0'), covar=tensor([0.2012, 0.3132, 0.0784, 0.1106], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0672, 0.1002, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 16:59:57,473 INFO [train.py:968] (0/2) Epoch 28, batch 38200, giga_loss[loss=0.2871, simple_loss=0.3582, pruned_loss=0.108, over 28200.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3515, pruned_loss=0.1006, over 5692881.83 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3549, pruned_loss=0.1103, over 5683760.31 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.35, pruned_loss=0.09907, over 5693446.93 frames. ], batch size: 65, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:00:09,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.938e+02 1.343e+03 1.745e+03 2.463e+03 4.150e+03, threshold=3.489e+03, percent-clipped=7.0
+2023-03-14 17:00:41,183 INFO [train.py:968] (0/2) Epoch 28, batch 38250, giga_loss[loss=0.3157, simple_loss=0.3823, pruned_loss=0.1245, over 29015.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3535, pruned_loss=0.1025, over 5694066.81 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3551, pruned_loss=0.1103, over 5683323.39 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3521, pruned_loss=0.1011, over 5694787.69 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:01:24,103 INFO [train.py:968] (0/2) Epoch 28, batch 38300, giga_loss[loss=0.2841, simple_loss=0.3588, pruned_loss=0.1047, over 28949.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3552, pruned_loss=0.1042, over 5687949.59 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3555, pruned_loss=0.1104, over 5678169.06 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3537, pruned_loss=0.1029, over 5693490.90 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:01:35,641 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.525e+02 1.317e+03 1.628e+03 2.097e+03 5.144e+03, threshold=3.256e+03, percent-clipped=5.0
+2023-03-14 17:02:00,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3197, 2.3443, 1.3311, 1.4133], device='cuda:0'), covar=tensor([0.0945, 0.0416, 0.0857, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0566, 0.0409, 0.0445], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 17:02:05,453 INFO [train.py:968] (0/2) Epoch 28, batch 38350, giga_loss[loss=0.2814, simple_loss=0.3529, pruned_loss=0.1049, over 28580.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3549, pruned_loss=0.1042, over 5690423.98 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3554, pruned_loss=0.1103, over 5683722.59 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3537, pruned_loss=0.1031, over 5690224.31 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:02:28,329 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1269276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:02:46,586 INFO [train.py:968] (0/2) Epoch 28, batch 38400, giga_loss[loss=0.3294, simple_loss=0.3918, pruned_loss=0.1334, over 28324.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3564, pruned_loss=0.1045, over 5695353.84 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3554, pruned_loss=0.1104, over 5683009.98 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3555, pruned_loss=0.1036, over 5695878.04 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 17:02:58,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.351e+02 1.336e+03 1.727e+03 2.444e+03 7.465e+03, threshold=3.454e+03, percent-clipped=9.0
+2023-03-14 17:02:59,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3589, 1.3628, 3.8309, 3.2481], device='cuda:0'), covar=tensor([0.1618, 0.2788, 0.0420, 0.1094], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0670, 0.0999, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 17:03:17,436 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1269334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:03:20,761 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4802, 1.6211, 1.2597, 1.1764], device='cuda:0'), covar=tensor([0.1114, 0.0589, 0.1082, 0.1193], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0448, 0.0523, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 17:03:28,903 INFO [train.py:968] (0/2) Epoch 28, batch 38450, giga_loss[loss=0.2871, simple_loss=0.3637, pruned_loss=0.1053, over 28918.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3571, pruned_loss=0.1042, over 5686663.19 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3561, pruned_loss=0.1109, over 5670454.80 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3557, pruned_loss=0.1028, over 5699801.05 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:04:08,349 INFO [train.py:968] (0/2) Epoch 28, batch 38500, giga_loss[loss=0.2809, simple_loss=0.3632, pruned_loss=0.09933, over 28870.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3566, pruned_loss=0.1036, over 5694996.52 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3562, pruned_loss=0.1109, over 5673849.92 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3555, pruned_loss=0.1025, over 5702634.87 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:04:08,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3416, 1.1481, 3.6342, 3.2389], device='cuda:0'), covar=tensor([0.1681, 0.3015, 0.0477, 0.0943], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0670, 0.0998, 0.0973], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 17:04:22,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.332e+02 1.283e+03 1.609e+03 2.383e+03 9.377e+03, threshold=3.218e+03, percent-clipped=6.0
+2023-03-14 17:04:25,730 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1269419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:04:28,005 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1269422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:04:47,598 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3198, 1.6034, 1.6345, 1.3022], device='cuda:0'), covar=tensor([0.2648, 0.2150, 0.1668, 0.2271], device='cuda:0'), in_proj_covar=tensor([0.2048, 0.2013, 0.1920, 0.2072], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 17:04:49,408 INFO [train.py:968] (0/2) Epoch 28, batch 38550, giga_loss[loss=0.2883, simple_loss=0.362, pruned_loss=0.1073, over 28259.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3542, pruned_loss=0.1026, over 5692755.62 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3567, pruned_loss=0.1111, over 5672381.10 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3527, pruned_loss=0.1012, over 5700243.54 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 17:04:50,767 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1269451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:04:54,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4150, 3.7779, 1.4864, 1.5696], device='cuda:0'), covar=tensor([0.0951, 0.0363, 0.0887, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0563, 0.0406, 0.0443], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 17:05:12,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2437, 1.5141, 1.5353, 1.1030], device='cuda:0'), covar=tensor([0.1713, 0.2908, 0.1454, 0.1807], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0718, 0.0985, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 17:05:19,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3943, 1.5131, 1.6259, 1.2803], device='cuda:0'), covar=tensor([0.1330, 0.2127, 0.1172, 0.1521], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0718, 0.0985, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 17:05:28,125 INFO [train.py:968] (0/2) Epoch 28, batch 38600, giga_loss[loss=0.2341, simple_loss=0.3179, pruned_loss=0.07517, over 28513.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3503, pruned_loss=0.1002, over 5696235.98 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.356, pruned_loss=0.1106, over 5676916.36 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3497, pruned_loss=0.09928, over 5698802.76 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 17:05:35,776 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1269507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:05:42,594 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.789e+02 1.169e+03 1.489e+03 1.674e+03 5.284e+03, threshold=2.978e+03, percent-clipped=3.0
+2023-03-14 17:06:07,869 INFO [train.py:968] (0/2) Epoch 28, batch 38650, giga_loss[loss=0.2932, simple_loss=0.367, pruned_loss=0.1097, over 28777.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3486, pruned_loss=0.09937, over 5700918.65 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3561, pruned_loss=0.1106, over 5679180.34 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.348, pruned_loss=0.09841, over 5701438.71 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 17:06:49,769 INFO [train.py:968] (0/2) Epoch 28, batch 38700, giga_loss[loss=0.2784, simple_loss=0.3567, pruned_loss=0.1, over 29032.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3487, pruned_loss=0.09934, over 5703447.08 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3563, pruned_loss=0.1107, over 5680137.44 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.348, pruned_loss=0.09846, over 5703049.69 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 17:07:01,219 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 17:07:03,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.355e+02 1.165e+03 1.511e+03 2.038e+03 4.763e+03, threshold=3.023e+03, percent-clipped=8.0
+2023-03-14 17:07:25,174 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-14 17:07:29,001 INFO [train.py:968] (0/2) Epoch 28, batch 38750, giga_loss[loss=0.2639, simple_loss=0.3426, pruned_loss=0.09262, over 28919.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3483, pruned_loss=0.09861, over 5706490.56 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3558, pruned_loss=0.1104, over 5683385.08 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3481, pruned_loss=0.09807, over 5703375.61 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 17:07:59,682 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1269688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:08:08,494 INFO [train.py:968] (0/2) Epoch 28, batch 38800, giga_loss[loss=0.2825, simple_loss=0.3569, pruned_loss=0.104, over 28700.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3489, pruned_loss=0.09841, over 5701431.70 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3559, pruned_loss=0.1104, over 5675188.64 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3485, pruned_loss=0.09786, over 5706103.63 frames. ], batch size: 85, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:08:15,129 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1269709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:08:21,570 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.845e+02 1.138e+03 1.397e+03 1.964e+03 4.906e+03, threshold=2.794e+03, percent-clipped=7.0
+2023-03-14 17:08:43,732 INFO [train.py:968] (0/2) Epoch 28, batch 38850, giga_loss[loss=0.2528, simple_loss=0.3351, pruned_loss=0.08526, over 28474.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3476, pruned_loss=0.09765, over 5702332.76 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3559, pruned_loss=0.1106, over 5672435.27 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.347, pruned_loss=0.09662, over 5709887.29 frames. ], batch size: 60, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:09:22,972 INFO [train.py:968] (0/2) Epoch 28, batch 38900, giga_loss[loss=0.2655, simple_loss=0.3425, pruned_loss=0.09429, over 28399.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3474, pruned_loss=0.09798, over 5701607.02 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3561, pruned_loss=0.1106, over 5675810.65 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3467, pruned_loss=0.09695, over 5704765.29 frames. ], batch size: 60, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:09:39,146 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.212e+02 1.261e+03 1.637e+03 2.218e+03 1.853e+04, threshold=3.274e+03, percent-clipped=17.0
+2023-03-14 17:09:51,220 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-14 17:10:03,158 INFO [train.py:968] (0/2) Epoch 28, batch 38950, giga_loss[loss=0.2761, simple_loss=0.3484, pruned_loss=0.1019, over 28791.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3463, pruned_loss=0.09804, over 5689417.65 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3566, pruned_loss=0.111, over 5669011.00 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3451, pruned_loss=0.09661, over 5698761.31 frames. ], batch size: 242, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:10:05,300 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1269852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:10:07,335 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1269855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:10:26,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1269882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:10:28,772 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1269884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:10:30,132 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3785, 1.4670, 1.5964, 1.2505], device='cuda:0'), covar=tensor([0.1451, 0.2194, 0.1200, 0.1501], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0719, 0.0987, 0.0884], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 17:10:31,318 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0730, 2.2912, 1.6134, 1.9413], device='cuda:0'), covar=tensor([0.1100, 0.0760, 0.1142, 0.1206], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0450, 0.0525, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 17:10:40,095 INFO [train.py:968] (0/2) Epoch 28, batch 39000, giga_loss[loss=0.2678, simple_loss=0.3379, pruned_loss=0.09882, over 28608.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3432, pruned_loss=0.09668, over 5681892.51 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.357, pruned_loss=0.1113, over 5658181.69 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3414, pruned_loss=0.09483, over 5699930.74 frames. ], batch size: 71, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:10:40,100 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 17:10:48,944 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2032, simple_loss=0.3111, pruned_loss=0.0476, over 944034.00 frames. 
+2023-03-14 17:10:48,945 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 17:11:02,225 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.785e+02 1.187e+03 1.474e+03 2.130e+03 6.373e+03, threshold=2.949e+03, percent-clipped=8.0
+2023-03-14 17:11:25,444 INFO [train.py:968] (0/2) Epoch 28, batch 39050, giga_loss[loss=0.244, simple_loss=0.3235, pruned_loss=0.0822, over 28889.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3421, pruned_loss=0.09604, over 5691608.14 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3573, pruned_loss=0.1114, over 5662130.83 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3398, pruned_loss=0.09385, over 5704077.31 frames. ], batch size: 112, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:11:59,913 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1269990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:12:05,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4058, 1.7260, 1.6787, 1.5743], device='cuda:0'), covar=tensor([0.2335, 0.2005, 0.2550, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0764, 0.0734, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 17:12:07,257 INFO [train.py:968] (0/2) Epoch 28, batch 39100, giga_loss[loss=0.2848, simple_loss=0.347, pruned_loss=0.1113, over 28982.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3414, pruned_loss=0.09625, over 5689869.22 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3575, pruned_loss=0.1114, over 5656624.93 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3391, pruned_loss=0.09414, over 5706138.97 frames. ], batch size: 227, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:12:08,139 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1270000.pt
+2023-03-14 17:12:22,566 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.324e+02 1.320e+03 1.615e+03 2.128e+03 9.026e+03, threshold=3.230e+03, percent-clipped=11.0
+2023-03-14 17:12:28,273 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1270025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:12:30,598 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1270028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:12:36,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3039, 1.4895, 1.3035, 1.5471], device='cuda:0'), covar=tensor([0.0782, 0.0315, 0.0354, 0.0924], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0120, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 17:12:44,950 INFO [train.py:968] (0/2) Epoch 28, batch 39150, libri_loss[loss=0.3075, simple_loss=0.3762, pruned_loss=0.1194, over 29229.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3392, pruned_loss=0.09527, over 5692066.43 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.358, pruned_loss=0.1118, over 5650380.23 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3365, pruned_loss=0.09293, over 5711145.26 frames. ], batch size: 97, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:12:51,202 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1270057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:12:56,138 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1270063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:13:08,977 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1270079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:13:20,277 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-14 17:13:23,489 INFO [train.py:968] (0/2) Epoch 28, batch 39200, giga_loss[loss=0.247, simple_loss=0.3198, pruned_loss=0.08712, over 28507.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3372, pruned_loss=0.09473, over 5685546.96 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3583, pruned_loss=0.112, over 5646683.29 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3341, pruned_loss=0.09201, over 5705039.67 frames. ], batch size: 71, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:13:38,413 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.653e+02 1.284e+03 1.558e+03 2.495e+03 1.329e+04, threshold=3.115e+03, percent-clipped=12.0
+2023-03-14 17:14:02,039 INFO [train.py:968] (0/2) Epoch 28, batch 39250, giga_loss[loss=0.2529, simple_loss=0.3289, pruned_loss=0.08846, over 28935.00 frames. ], tot_loss[loss=0.259, simple_loss=0.333, pruned_loss=0.09251, over 5688834.24 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3582, pruned_loss=0.1119, over 5644995.60 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3303, pruned_loss=0.09016, over 5706592.66 frames. ], batch size: 145, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:14:43,306 INFO [train.py:968] (0/2) Epoch 28, batch 39300, giga_loss[loss=0.235, simple_loss=0.3093, pruned_loss=0.0804, over 28591.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3314, pruned_loss=0.09184, over 5696075.31 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3586, pruned_loss=0.1122, over 5649899.06 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3284, pruned_loss=0.08933, over 5706518.48 frames. ], batch size: 85, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:14:51,242 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1270206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:14:53,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1270209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:14:57,333 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-14 17:14:59,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.384e+02 1.123e+03 1.322e+03 1.851e+03 7.809e+03, threshold=2.645e+03, percent-clipped=6.0
+2023-03-14 17:15:17,800 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1270238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:15:26,971 INFO [train.py:968] (0/2) Epoch 28, batch 39350, giga_loss[loss=0.2536, simple_loss=0.3424, pruned_loss=0.0824, over 28685.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3338, pruned_loss=0.09307, over 5703989.10 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1125, over 5655010.91 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3304, pruned_loss=0.09037, over 5709005.15 frames. ], batch size: 242, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:16:05,651 INFO [train.py:968] (0/2) Epoch 28, batch 39400, giga_loss[loss=0.2462, simple_loss=0.3305, pruned_loss=0.08098, over 28865.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3361, pruned_loss=0.09384, over 5697457.10 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3586, pruned_loss=0.1123, over 5652345.98 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3327, pruned_loss=0.09106, over 5706449.95 frames. ], batch size: 186, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:16:22,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.648e+02 1.174e+03 1.366e+03 1.873e+03 6.331e+03, threshold=2.732e+03, percent-clipped=10.0
+2023-03-14 17:16:48,107 INFO [train.py:968] (0/2) Epoch 28, batch 39450, giga_loss[loss=0.2853, simple_loss=0.361, pruned_loss=0.1048, over 27536.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3392, pruned_loss=0.09469, over 5705095.55 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3582, pruned_loss=0.112, over 5662141.30 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3359, pruned_loss=0.09197, over 5704998.92 frames. ], batch size: 472, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:16:50,052 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-14 17:17:02,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1270365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:17:10,102 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 17:17:29,455 INFO [train.py:968] (0/2) Epoch 28, batch 39500, giga_loss[loss=0.2624, simple_loss=0.3458, pruned_loss=0.08947, over 28371.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3419, pruned_loss=0.09564, over 5702864.74 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3584, pruned_loss=0.1121, over 5668410.85 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3386, pruned_loss=0.09291, over 5698060.23 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:17:45,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.746e+02 1.202e+03 1.531e+03 2.188e+03 8.536e+03, threshold=3.063e+03, percent-clipped=15.0
+2023-03-14 17:18:09,126 INFO [train.py:968] (0/2) Epoch 28, batch 39550, giga_loss[loss=0.2627, simple_loss=0.3419, pruned_loss=0.09173, over 28669.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3408, pruned_loss=0.09418, over 5702714.88 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3587, pruned_loss=0.1122, over 5674873.73 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3374, pruned_loss=0.09136, over 5693628.62 frames. ], batch size: 242, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:18:16,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1270454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:18:51,636 INFO [train.py:968] (0/2) Epoch 28, batch 39600, giga_loss[loss=0.2416, simple_loss=0.3223, pruned_loss=0.08046, over 29041.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3399, pruned_loss=0.09273, over 5706420.07 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3588, pruned_loss=0.1123, over 5677417.05 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.337, pruned_loss=0.09025, over 5697166.96 frames. ], batch size: 136, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:18:59,277 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1270508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:19:02,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1270511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:19:07,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.324e+02 1.269e+03 1.581e+03 2.082e+03 6.340e+03, threshold=3.161e+03, percent-clipped=12.0
+2023-03-14 17:19:18,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1435, 2.2539, 1.7421, 2.0833], device='cuda:0'), covar=tensor([0.0971, 0.0632, 0.0923, 0.1028], device='cuda:0'), in_proj_covar=tensor([0.0412, 0.0448, 0.0523, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 17:19:26,433 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1270540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:19:39,985 INFO [train.py:968] (0/2) Epoch 28, batch 39650, giga_loss[loss=0.4068, simple_loss=0.4388, pruned_loss=0.1874, over 26821.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3418, pruned_loss=0.09484, over 5687038.08 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.359, pruned_loss=0.1124, over 5667752.68 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.339, pruned_loss=0.09245, over 5689124.91 frames. ], batch size: 555, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:20:00,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3029, 1.6818, 1.6659, 1.4824], device='cuda:0'), covar=tensor([0.2242, 0.1865, 0.2326, 0.2155], device='cuda:0'), in_proj_covar=tensor([0.0507, 0.0763, 0.0732, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 17:20:11,547 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 17:20:21,527 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1270597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:20:22,372 INFO [train.py:968] (0/2) Epoch 28, batch 39700, giga_loss[loss=0.3288, simple_loss=0.3908, pruned_loss=0.1334, over 28825.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3422, pruned_loss=0.09542, over 5688451.01 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1125, over 5671383.71 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3398, pruned_loss=0.09326, over 5687318.20 frames. ], batch size: 119, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:20:23,455 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1270600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:20:37,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.191e+02 1.326e+03 1.778e+03 2.138e+03 8.468e+03, threshold=3.555e+03, percent-clipped=7.0
+2023-03-14 17:20:48,105 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1270629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:20:50,348 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1270633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:21:02,608 INFO [train.py:968] (0/2) Epoch 28, batch 39750, giga_loss[loss=0.3117, simple_loss=0.3809, pruned_loss=0.1212, over 27684.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3459, pruned_loss=0.09785, over 5687884.82 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3594, pruned_loss=0.1128, over 5665754.36 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3431, pruned_loss=0.09528, over 5693708.32 frames. ], batch size: 472, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:21:30,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3225, 1.5749, 1.3065, 0.9826], device='cuda:0'), covar=tensor([0.2942, 0.2914, 0.3412, 0.2527], device='cuda:0'), in_proj_covar=tensor([0.1610, 0.1158, 0.1419, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 17:21:33,978 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4882, 2.1222, 1.5467, 0.6816], device='cuda:0'), covar=tensor([0.7930, 0.4025, 0.5097, 0.8133], device='cuda:0'), in_proj_covar=tensor([0.1845, 0.1725, 0.1658, 0.1497], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 17:21:42,504 INFO [train.py:968] (0/2) Epoch 28, batch 39800, libri_loss[loss=0.2317, simple_loss=0.3048, pruned_loss=0.07932, over 29671.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3474, pruned_loss=0.09812, over 5689795.34 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.359, pruned_loss=0.1124, over 5659969.07 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3452, pruned_loss=0.09602, over 5701435.20 frames. ], batch size: 73, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:21:49,429 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 17:21:56,831 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.571e+02 1.357e+03 1.552e+03 2.277e+03 6.210e+03, threshold=3.104e+03, percent-clipped=6.0
+2023-03-14 17:22:19,487 INFO [train.py:968] (0/2) Epoch 28, batch 39850, giga_loss[loss=0.2753, simple_loss=0.3569, pruned_loss=0.09686, over 28875.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3493, pruned_loss=0.09868, over 5697699.77 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3599, pruned_loss=0.113, over 5656959.78 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3463, pruned_loss=0.09602, over 5710231.24 frames. ], batch size: 164, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:22:36,338 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1270767.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:22:47,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7486, 1.8759, 1.3823, 1.4589], device='cuda:0'), covar=tensor([0.1025, 0.0698, 0.1110, 0.1204], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0449, 0.0524, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 17:22:48,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4138, 1.5206, 1.2583, 1.4621], device='cuda:0'), covar=tensor([0.0715, 0.0316, 0.0335, 0.0867], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 17:22:59,505 INFO [train.py:968] (0/2) Epoch 28, batch 39900, libri_loss[loss=0.3574, simple_loss=0.4157, pruned_loss=0.1496, over 29363.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3513, pruned_loss=0.1001, over 5699531.71 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3605, pruned_loss=0.1132, over 5657539.33 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3479, pruned_loss=0.09722, over 5710256.83 frames. ], batch size: 92, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:23:04,152 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1270803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:23:07,416 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 17:23:16,892 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.460e+02 1.401e+03 1.841e+03 2.325e+03 4.057e+03, threshold=3.682e+03, percent-clipped=11.0
+2023-03-14 17:23:30,423 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5405, 3.7692, 1.6329, 1.5923], device='cuda:0'), covar=tensor([0.0964, 0.0385, 0.0922, 0.1350], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0567, 0.0409, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 17:23:41,568 INFO [train.py:968] (0/2) Epoch 28, batch 39950, giga_loss[loss=0.2622, simple_loss=0.3394, pruned_loss=0.09252, over 28262.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3511, pruned_loss=0.1006, over 5704781.85 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3606, pruned_loss=0.1133, over 5660991.08 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3483, pruned_loss=0.098, over 5710633.30 frames. ], batch size: 71, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:24:03,253 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3374, 1.6484, 1.5862, 1.4532], device='cuda:0'), covar=tensor([0.2255, 0.2031, 0.2602, 0.2217], device='cuda:0'), in_proj_covar=tensor([0.0507, 0.0762, 0.0732, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 17:24:21,407 INFO [train.py:968] (0/2) Epoch 28, batch 40000, giga_loss[loss=0.2628, simple_loss=0.3439, pruned_loss=0.09088, over 29016.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3498, pruned_loss=0.09952, over 5710573.04 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3603, pruned_loss=0.1131, over 5665794.46 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3475, pruned_loss=0.09734, over 5712252.44 frames. ], batch size: 128, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:24:37,379 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.265e+02 1.294e+03 1.551e+03 2.114e+03 6.008e+03, threshold=3.101e+03, percent-clipped=5.0
+2023-03-14 17:25:02,914 INFO [train.py:968] (0/2) Epoch 28, batch 40050, giga_loss[loss=0.2541, simple_loss=0.333, pruned_loss=0.0876, over 28838.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3481, pruned_loss=0.0987, over 5710032.12 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3603, pruned_loss=0.113, over 5669338.26 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.346, pruned_loss=0.09674, over 5708705.52 frames. ], batch size: 199, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:25:09,572 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1270959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:25:40,426 INFO [train.py:968] (0/2) Epoch 28, batch 40100, giga_loss[loss=0.2073, simple_loss=0.2953, pruned_loss=0.0596, over 28909.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3444, pruned_loss=0.09666, over 5715362.34 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3606, pruned_loss=0.1131, over 5672040.36 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3423, pruned_loss=0.09472, over 5712686.03 frames. ], batch size: 174, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:25:48,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1271008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:25:58,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.918e+02 1.306e+03 1.625e+03 2.246e+03 6.737e+03, threshold=3.250e+03, percent-clipped=9.0
+2023-03-14 17:26:21,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7644, 4.9104, 1.9592, 1.9612], device='cuda:0'), covar=tensor([0.0915, 0.0285, 0.0886, 0.1252], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0566, 0.0408, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 17:26:22,683 INFO [train.py:968] (0/2) Epoch 28, batch 40150, giga_loss[loss=0.2351, simple_loss=0.3092, pruned_loss=0.08047, over 28437.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3418, pruned_loss=0.09534, over 5711883.94 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3606, pruned_loss=0.1132, over 5673975.44 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3399, pruned_loss=0.09354, over 5708467.11 frames. ], batch size: 71, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:26:59,428 INFO [train.py:968] (0/2) Epoch 28, batch 40200, giga_loss[loss=0.2294, simple_loss=0.3132, pruned_loss=0.07279, over 28533.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.343, pruned_loss=0.0948, over 5720141.19 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3606, pruned_loss=0.1132, over 5681920.10 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3409, pruned_loss=0.09279, over 5711526.69 frames. ], batch size: 78, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:27:04,485 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 17:27:13,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7036, 4.9980, 1.8823, 1.9765], device='cuda:0'), covar=tensor([0.0936, 0.0285, 0.0905, 0.1224], device='cuda:0'), in_proj_covar=tensor([0.0423, 0.0567, 0.0409, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:0')
+2023-03-14 17:27:18,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.728e+02 1.271e+03 1.607e+03 2.072e+03 5.074e+03, threshold=3.215e+03, percent-clipped=9.0
+2023-03-14 17:27:30,224 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2063, 1.2911, 3.3688, 3.0354], device='cuda:0'), covar=tensor([0.1624, 0.2694, 0.0525, 0.1091], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0671, 0.0999, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 17:27:38,222 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1271142.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:27:43,603 INFO [train.py:968] (0/2) Epoch 28, batch 40250, giga_loss[loss=0.2883, simple_loss=0.3613, pruned_loss=0.1076, over 28209.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3446, pruned_loss=0.09511, over 5709457.33 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3608, pruned_loss=0.1132, over 5683040.50 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3428, pruned_loss=0.09338, over 5701865.41 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:27:45,402 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1271151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:27:48,924 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1271154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:28:04,967 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1271178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:28:07,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-14 17:28:08,556 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1271183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:28:09,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4006, 1.5776, 1.6520, 1.3325], device='cuda:0'), covar=tensor([0.3971, 0.3206, 0.2438, 0.3364], device='cuda:0'), in_proj_covar=tensor([0.2072, 0.2037, 0.1944, 0.2090], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 17:28:22,065 INFO [train.py:968] (0/2) Epoch 28, batch 40300, giga_loss[loss=0.298, simple_loss=0.3592, pruned_loss=0.1184, over 26635.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3451, pruned_loss=0.09646, over 5706235.52 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3607, pruned_loss=0.1133, over 5675235.35 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3432, pruned_loss=0.09456, over 5708543.92 frames. ], batch size: 555, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:28:37,797 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.252e+03 1.443e+03 1.834e+03 4.370e+03, threshold=2.886e+03, percent-clipped=4.0
+2023-03-14 17:29:02,147 INFO [train.py:968] (0/2) Epoch 28, batch 40350, giga_loss[loss=0.3409, simple_loss=0.389, pruned_loss=0.1464, over 26796.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3432, pruned_loss=0.09634, over 5712886.18 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3609, pruned_loss=0.1135, over 5677229.04 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3414, pruned_loss=0.09453, over 5713265.10 frames. ], batch size: 555, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:29:31,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1271285.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:29:34,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1271288.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:29:43,405 INFO [train.py:968] (0/2) Epoch 28, batch 40400, giga_loss[loss=0.2448, simple_loss=0.3302, pruned_loss=0.07973, over 29113.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3413, pruned_loss=0.09677, over 5712839.17 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3612, pruned_loss=0.1136, over 5681307.86 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3392, pruned_loss=0.09491, over 5710241.58 frames. ], batch size: 155, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:29:58,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1271317.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:29:59,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.307e+02 1.330e+03 1.641e+03 2.094e+03 7.080e+03, threshold=3.281e+03, percent-clipped=9.0
+2023-03-14 17:30:01,510 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1271321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:30:03,296 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1271324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:30:12,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1271334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:30:25,204 INFO [train.py:968] (0/2) Epoch 28, batch 40450, giga_loss[loss=0.2858, simple_loss=0.3468, pruned_loss=0.1124, over 28779.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3403, pruned_loss=0.0975, over 5709404.42 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3611, pruned_loss=0.1136, over 5685906.62 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3384, pruned_loss=0.0958, over 5703804.78 frames. ], batch size: 99, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:30:27,925 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1271353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:31:02,897 INFO [train.py:968] (0/2) Epoch 28, batch 40500, giga_loss[loss=0.2421, simple_loss=0.3187, pruned_loss=0.08275, over 28427.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3404, pruned_loss=0.09813, over 5706264.14 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3613, pruned_loss=0.1139, over 5681956.87 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3382, pruned_loss=0.09602, over 5705146.53 frames. ], batch size: 85, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:31:19,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.800e+02 1.293e+03 1.599e+03 2.057e+03 5.881e+03, threshold=3.199e+03, percent-clipped=4.0
+2023-03-14 17:31:41,681 INFO [train.py:968] (0/2) Epoch 28, batch 40550, giga_loss[loss=0.2464, simple_loss=0.3195, pruned_loss=0.0866, over 28905.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3379, pruned_loss=0.09703, over 5705429.81 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.362, pruned_loss=0.1144, over 5681588.92 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3348, pruned_loss=0.0945, over 5705465.80 frames. ], batch size: 186, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:32:05,000 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1271477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:32:06,974 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1271480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:32:21,026 INFO [train.py:968] (0/2) Epoch 28, batch 40600, giga_loss[loss=0.2142, simple_loss=0.2889, pruned_loss=0.06972, over 28331.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3311, pruned_loss=0.09326, over 5707301.45 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1145, over 5683565.53 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3284, pruned_loss=0.09099, over 5705798.43 frames. ], batch size: 65, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:32:30,623 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1271509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:32:38,497 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.098e+02 1.255e+03 1.743e+03 2.201e+03 7.212e+03, threshold=3.487e+03, percent-clipped=13.0
+2023-03-14 17:32:51,851 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1271538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:33:00,785 INFO [train.py:968] (0/2) Epoch 28, batch 40650, giga_loss[loss=0.2641, simple_loss=0.3432, pruned_loss=0.09251, over 28302.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3288, pruned_loss=0.09146, over 5708210.38 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3616, pruned_loss=0.1141, over 5681249.32 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3262, pruned_loss=0.0894, over 5709087.93 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:33:03,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8189, 3.6643, 3.4583, 1.7518], device='cuda:0'), covar=tensor([0.0750, 0.0863, 0.0745, 0.2151], device='cuda:0'), in_proj_covar=tensor([0.1293, 0.1194, 0.1004, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 17:33:41,915 INFO [train.py:968] (0/2) Epoch 28, batch 40700, libri_loss[loss=0.3064, simple_loss=0.3729, pruned_loss=0.1199, over 29263.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3307, pruned_loss=0.09221, over 5702327.77 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3617, pruned_loss=0.1143, over 5674998.91 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3278, pruned_loss=0.08988, over 5709870.01 frames. ], batch size: 97, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:33:59,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.491e+02 1.273e+03 1.703e+03 2.290e+03 6.077e+03, threshold=3.406e+03, percent-clipped=7.0
+2023-03-14 17:34:21,165 INFO [train.py:968] (0/2) Epoch 28, batch 40750, giga_loss[loss=0.2704, simple_loss=0.3508, pruned_loss=0.09502, over 28319.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3342, pruned_loss=0.09337, over 5707014.79 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3612, pruned_loss=0.114, over 5680079.82 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3317, pruned_loss=0.09137, over 5709073.92 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:34:35,064 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-14 17:35:00,477 INFO [train.py:968] (0/2) Epoch 28, batch 40800, giga_loss[loss=0.2537, simple_loss=0.3249, pruned_loss=0.09126, over 28553.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3371, pruned_loss=0.0946, over 5708414.70 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3611, pruned_loss=0.1139, over 5684500.60 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3348, pruned_loss=0.09269, over 5706727.99 frames. ], batch size: 71, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:35:17,132 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.684e+02 1.304e+03 1.661e+03 2.249e+03 8.956e+03, threshold=3.323e+03, percent-clipped=7.0
+2023-03-14 17:35:21,614 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1271726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:35:21,741 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8936, 2.0940, 1.7233, 2.0906], device='cuda:0'), covar=tensor([0.2682, 0.2812, 0.3128, 0.2542], device='cuda:0'), in_proj_covar=tensor([0.1608, 0.1156, 0.1417, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 17:35:41,448 INFO [train.py:968] (0/2) Epoch 28, batch 40850, giga_loss[loss=0.2751, simple_loss=0.3542, pruned_loss=0.09799, over 28647.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3414, pruned_loss=0.09674, over 5696717.81 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3611, pruned_loss=0.1137, over 5681180.63 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.339, pruned_loss=0.09488, over 5698257.31 frames. ], batch size: 92, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:36:22,032 INFO [train.py:968] (0/2) Epoch 28, batch 40900, giga_loss[loss=0.2532, simple_loss=0.3311, pruned_loss=0.08764, over 28898.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3438, pruned_loss=0.0975, over 5704538.84 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.361, pruned_loss=0.1136, over 5683019.43 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3415, pruned_loss=0.09581, over 5704565.17 frames. ], batch size: 186, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:36:22,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0994, 1.4978, 1.4278, 1.2936], device='cuda:0'), covar=tensor([0.2082, 0.1631, 0.2421, 0.1904], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0765, 0.0734, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 17:36:39,812 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.739e+02 1.328e+03 1.699e+03 2.211e+03 8.176e+03, threshold=3.398e+03, percent-clipped=6.0
+2023-03-14 17:37:06,248 INFO [train.py:968] (0/2) Epoch 28, batch 40950, giga_loss[loss=0.2786, simple_loss=0.3483, pruned_loss=0.1044, over 28757.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3491, pruned_loss=0.1024, over 5703874.48 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3605, pruned_loss=0.1134, over 5689336.89 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3473, pruned_loss=0.1008, over 5698886.62 frames. ], batch size: 119, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:37:51,134 INFO [train.py:968] (0/2) Epoch 28, batch 41000, giga_loss[loss=0.3311, simple_loss=0.3917, pruned_loss=0.1352, over 28760.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3544, pruned_loss=0.1069, over 5711756.92 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1132, over 5696718.43 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3529, pruned_loss=0.1054, over 5701624.37 frames. ], batch size: 99, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:38:00,597 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-14 17:38:04,921 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1271913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:38:11,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.776e+03 2.276e+03 3.231e+03 6.296e+03, threshold=4.552e+03, percent-clipped=20.0
+2023-03-14 17:38:29,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3211, 1.5785, 1.5568, 1.4220], device='cuda:0'), covar=tensor([0.1879, 0.1734, 0.2165, 0.1821], device='cuda:0'), in_proj_covar=tensor([0.0506, 0.0762, 0.0732, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 17:38:34,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4505, 2.7286, 1.5977, 1.5693], device='cuda:0'), covar=tensor([0.0825, 0.0357, 0.0727, 0.1162], device='cuda:0'), in_proj_covar=tensor([0.0424, 0.0568, 0.0410, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 17:38:34,511 INFO [train.py:968] (0/2) Epoch 28, batch 41050, giga_loss[loss=0.3669, simple_loss=0.399, pruned_loss=0.1674, over 23736.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3615, pruned_loss=0.1119, over 5699471.73 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1133, over 5696898.08 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3602, pruned_loss=0.1106, over 5691542.72 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:39:20,022 INFO [train.py:968] (0/2) Epoch 28, batch 41100, giga_loss[loss=0.2912, simple_loss=0.3673, pruned_loss=0.1076, over 28656.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3676, pruned_loss=0.1165, over 5699469.44 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3603, pruned_loss=0.1134, over 5699760.10 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3665, pruned_loss=0.1153, over 5690681.39 frames. ], batch size: 307, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:39:20,801 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1272000.pt
+2023-03-14 17:39:36,706 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.164e+03 1.758e+03 2.165e+03 2.888e+03 5.216e+03, threshold=4.329e+03, percent-clipped=2.0
+2023-03-14 17:40:00,678 INFO [train.py:968] (0/2) Epoch 28, batch 41150, giga_loss[loss=0.2971, simple_loss=0.3589, pruned_loss=0.1176, over 28812.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3731, pruned_loss=0.1214, over 5701709.92 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3597, pruned_loss=0.1129, over 5704106.14 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3732, pruned_loss=0.1211, over 5690984.50 frames. ], batch size: 99, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:40:08,093 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 17:40:08,570 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1272056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:40:11,032 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1272059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:40:38,137 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1272088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:40:49,292 INFO [train.py:968] (0/2) Epoch 28, batch 41200, giga_loss[loss=0.2684, simple_loss=0.3418, pruned_loss=0.09745, over 28346.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.378, pruned_loss=0.1252, over 5680553.21 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3601, pruned_loss=0.1131, over 5697250.61 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.378, pruned_loss=0.125, over 5678286.54 frames. ], batch size: 65, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:40:53,367 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1272101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:41:12,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.166e+03 2.003e+03 2.444e+03 3.005e+03 7.011e+03, threshold=4.888e+03, percent-clipped=6.0
+2023-03-14 17:41:14,449 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:41:39,656 INFO [train.py:968] (0/2) Epoch 28, batch 41250, giga_loss[loss=0.3731, simple_loss=0.4147, pruned_loss=0.1658, over 28207.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3807, pruned_loss=0.1284, over 5652479.11 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3601, pruned_loss=0.1131, over 5682237.29 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3809, pruned_loss=0.1284, over 5663696.20 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:42:36,630 INFO [train.py:968] (0/2) Epoch 28, batch 41300, giga_loss[loss=0.3275, simple_loss=0.3897, pruned_loss=0.1326, over 29015.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3835, pruned_loss=0.1318, over 5656634.56 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.36, pruned_loss=0.1129, over 5685594.90 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3841, pruned_loss=0.1322, over 5662174.91 frames. ], batch size: 164, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:43:01,719 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.238e+03 1.924e+03 2.576e+03 3.439e+03 7.405e+03, threshold=5.153e+03, percent-clipped=7.0
+2023-03-14 17:43:21,866 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1272244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:43:23,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1272247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:43:24,949 INFO [train.py:968] (0/2) Epoch 28, batch 41350, giga_loss[loss=0.3839, simple_loss=0.4269, pruned_loss=0.1704, over 28886.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.386, pruned_loss=0.1351, over 5636619.68 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5688441.76 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3872, pruned_loss=0.1359, over 5637189.89 frames. ], batch size: 199, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:43:29,144 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3267, 3.1604, 3.0285, 1.5014], device='cuda:0'), covar=tensor([0.0999, 0.1104, 0.0963, 0.2166], device='cuda:0'), in_proj_covar=tensor([0.1304, 0.1205, 0.1015, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 17:43:52,954 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1272276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:44:15,661 INFO [train.py:968] (0/2) Epoch 28, batch 41400, libri_loss[loss=0.2777, simple_loss=0.3448, pruned_loss=0.1052, over 29573.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3887, pruned_loss=0.1373, over 5633751.32 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.36, pruned_loss=0.1129, over 5689757.76 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3903, pruned_loss=0.1386, over 5631983.19 frames. ], batch size: 75, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:44:36,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.365e+03 2.193e+03 2.817e+03 4.358e+03 1.307e+04, threshold=5.634e+03, percent-clipped=17.0
+2023-03-14 17:45:02,863 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3649, 1.4391, 1.2752, 1.5539], device='cuda:0'), covar=tensor([0.0684, 0.0399, 0.0334, 0.0750], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 17:45:03,592 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 17:45:03,857 INFO [train.py:968] (0/2) Epoch 28, batch 41450, giga_loss[loss=0.3655, simple_loss=0.4104, pruned_loss=0.1603, over 27962.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3892, pruned_loss=0.1386, over 5621697.97 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.1132, over 5686301.41 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3914, pruned_loss=0.1403, over 5620982.84 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:45:11,726 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-14 17:45:18,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272360.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:45:53,146 INFO [train.py:968] (0/2) Epoch 28, batch 41500, giga_loss[loss=0.2942, simple_loss=0.3645, pruned_loss=0.112, over 29082.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3878, pruned_loss=0.1381, over 5615103.17 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3602, pruned_loss=0.1132, over 5670796.01 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3901, pruned_loss=0.14, over 5626891.65 frames. ], batch size: 155, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:45:57,594 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:46:14,366 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.350e+03 1.955e+03 2.526e+03 3.468e+03 9.578e+03, threshold=5.053e+03, percent-clipped=4.0
+2023-03-14 17:46:17,878 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 17:46:39,320 INFO [train.py:968] (0/2) Epoch 28, batch 41550, giga_loss[loss=0.3337, simple_loss=0.3938, pruned_loss=0.1368, over 28861.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3856, pruned_loss=0.1356, over 5634791.79 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3597, pruned_loss=0.1127, over 5674544.36 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3887, pruned_loss=0.1383, over 5639821.73 frames. ], batch size: 199, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:47:26,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1272497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:47:27,701 INFO [train.py:968] (0/2) Epoch 28, batch 41600, libri_loss[loss=0.2608, simple_loss=0.3361, pruned_loss=0.09277, over 29548.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3841, pruned_loss=0.1328, over 5644555.96 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3592, pruned_loss=0.1123, over 5680111.64 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3876, pruned_loss=0.1358, over 5642763.23 frames. ], batch size: 79, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:47:51,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.266e+03 2.078e+03 2.489e+03 3.110e+03 9.058e+03, threshold=4.977e+03, percent-clipped=5.0
+2023-03-14 17:47:56,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3820, 3.1109, 1.5723, 1.4872], device='cuda:0'), covar=tensor([0.0928, 0.0326, 0.0844, 0.1308], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0573, 0.0412, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 17:48:06,684 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 17:48:14,516 INFO [train.py:968] (0/2) Epoch 28, batch 41650, giga_loss[loss=0.335, simple_loss=0.3892, pruned_loss=0.1404, over 27857.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3854, pruned_loss=0.1329, over 5659359.89 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3591, pruned_loss=0.1122, over 5681370.43 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3888, pruned_loss=0.1358, over 5656250.03 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:49:05,642 INFO [train.py:968] (0/2) Epoch 28, batch 41700, giga_loss[loss=0.2782, simple_loss=0.3591, pruned_loss=0.09863, over 28845.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.386, pruned_loss=0.1337, over 5648830.16 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3591, pruned_loss=0.1122, over 5689605.50 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3898, pruned_loss=0.1368, over 5638184.37 frames. ], batch size: 174, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:49:14,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9197, 1.1442, 1.0797, 0.8743], device='cuda:0'), covar=tensor([0.2624, 0.3058, 0.1887, 0.2580], device='cuda:0'), in_proj_covar=tensor([0.2072, 0.2042, 0.1945, 0.2091], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 17:49:27,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.317e+03 1.961e+03 2.655e+03 3.425e+03 7.400e+03, threshold=5.310e+03, percent-clipped=8.0
+2023-03-14 17:49:45,392 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1272640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:49:47,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1272643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:49:52,332 INFO [train.py:968] (0/2) Epoch 28, batch 41750, giga_loss[loss=0.3441, simple_loss=0.4077, pruned_loss=0.1403, over 28675.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3823, pruned_loss=0.1298, over 5646804.85 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5685842.05 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3864, pruned_loss=0.1331, over 5640253.81 frames. ], batch size: 307, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:50:16,701 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1272672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:50:27,531 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4201, 4.2583, 4.0382, 2.2173], device='cuda:0'), covar=tensor([0.0570, 0.0682, 0.0769, 0.2056], device='cuda:0'), in_proj_covar=tensor([0.1315, 0.1214, 0.1023, 0.0757], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 17:50:28,345 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4961, 1.6187, 1.6863, 1.2882], device='cuda:0'), covar=tensor([0.1951, 0.2797, 0.1636, 0.1904], device='cuda:0'), in_proj_covar=tensor([0.0928, 0.0713, 0.0976, 0.0876], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 17:50:42,538 INFO [train.py:968] (0/2) Epoch 28, batch 41800, giga_loss[loss=0.3346, simple_loss=0.3913, pruned_loss=0.139, over 28726.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.38, pruned_loss=0.1269, over 5651914.44 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5686368.27 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3834, pruned_loss=0.1296, over 5645884.92 frames. ], batch size: 92, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:51:06,427 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.942e+02 1.803e+03 2.211e+03 3.185e+03 9.662e+03, threshold=4.422e+03, percent-clipped=5.0
+2023-03-14 17:51:16,589 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1272735.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:51:31,439 INFO [train.py:968] (0/2) Epoch 28, batch 41850, giga_loss[loss=0.3258, simple_loss=0.3873, pruned_loss=0.1322, over 28611.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3753, pruned_loss=0.1228, over 5661336.18 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3585, pruned_loss=0.1119, over 5688750.81 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3785, pruned_loss=0.1253, over 5654072.82 frames. ], batch size: 336, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:51:47,736 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:51:48,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3250, 1.5571, 1.2245, 1.1559], device='cuda:0'), covar=tensor([0.1136, 0.0585, 0.1122, 0.1151], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0455, 0.0528, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 17:51:55,843 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272773.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:52:00,516 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1272778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:52:05,237 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 17:52:16,624 INFO [train.py:968] (0/2) Epoch 28, batch 41900, giga_loss[loss=0.281, simple_loss=0.3548, pruned_loss=0.1037, over 27505.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3726, pruned_loss=0.1211, over 5654671.32 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3585, pruned_loss=0.112, over 5684989.74 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3756, pruned_loss=0.1233, over 5650951.76 frames. ], batch size: 472, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:52:41,614 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.773e+03 2.178e+03 3.549e+03 1.357e+04, threshold=4.357e+03, percent-clipped=12.0
+2023-03-14 17:53:01,937 INFO [train.py:968] (0/2) Epoch 28, batch 41950, giga_loss[loss=0.2732, simple_loss=0.346, pruned_loss=0.1002, over 28283.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3715, pruned_loss=0.1208, over 5645194.71 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3584, pruned_loss=0.1121, over 5681558.40 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3746, pruned_loss=0.1229, over 5644243.45 frames. ], batch size: 77, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:53:07,636 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.86 vs. limit=2.0
+2023-03-14 17:53:26,899 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1272878.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:53:28,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1272881.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:53:47,931 INFO [train.py:968] (0/2) Epoch 28, batch 42000, giga_loss[loss=0.3801, simple_loss=0.4115, pruned_loss=0.1743, over 26742.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3713, pruned_loss=0.1208, over 5659016.92 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.358, pruned_loss=0.1119, over 5686396.33 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3744, pruned_loss=0.1228, over 5653548.38 frames. ], batch size: 555, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:53:47,937 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 17:53:56,247 INFO [train.py:1012] (0/2) Epoch 28, validation: loss=0.2009, simple_loss=0.3082, pruned_loss=0.04679, over 944034.00 frames. 
+2023-03-14 17:53:56,248 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 17:54:08,702 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1272910.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:54:20,642 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1272921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:54:21,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.063e+03 1.629e+03 2.192e+03 3.176e+03 6.770e+03, threshold=4.384e+03, percent-clipped=8.0
+2023-03-14 17:54:22,734 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1272924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:54:46,362 INFO [train.py:968] (0/2) Epoch 28, batch 42050, libri_loss[loss=0.311, simple_loss=0.3674, pruned_loss=0.1273, over 19353.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3697, pruned_loss=0.1187, over 5645338.39 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3582, pruned_loss=0.1121, over 5659653.88 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3723, pruned_loss=0.1203, over 5665453.38 frames. ], batch size: 187, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:54:50,022 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1272953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:55:00,304 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272962.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:55:03,259 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:55:39,160 INFO [train.py:968] (0/2) Epoch 28, batch 42100, giga_loss[loss=0.2631, simple_loss=0.3413, pruned_loss=0.09239, over 28165.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3688, pruned_loss=0.1161, over 5652411.00 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3579, pruned_loss=0.1119, over 5663250.97 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3713, pruned_loss=0.1176, over 5664966.65 frames. ], batch size: 77, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:55:39,662 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 17:56:00,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.651e+03 2.130e+03 3.095e+03 5.846e+03, threshold=4.260e+03, percent-clipped=11.0
+2023-03-14 17:56:12,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3773, 1.4783, 1.4046, 1.6294], device='cuda:0'), covar=tensor([0.0648, 0.0311, 0.0299, 0.0700], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 17:56:18,513 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-14 17:56:20,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3831, 1.2329, 4.0951, 3.3502], device='cuda:0'), covar=tensor([0.1747, 0.2972, 0.0515, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0678, 0.1015, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 17:56:26,204 INFO [train.py:968] (0/2) Epoch 28, batch 42150, giga_loss[loss=0.3478, simple_loss=0.3975, pruned_loss=0.1491, over 27578.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.37, pruned_loss=0.1152, over 5662990.45 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3575, pruned_loss=0.1118, over 5667920.94 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3726, pruned_loss=0.1166, over 5668697.18 frames. ], batch size: 472, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:56:49,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2346, 0.8237, 0.9157, 1.4428], device='cuda:0'), covar=tensor([0.0776, 0.0419, 0.0376, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 17:57:13,355 INFO [train.py:968] (0/2) Epoch 28, batch 42200, libri_loss[loss=0.301, simple_loss=0.3659, pruned_loss=0.118, over 29524.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3718, pruned_loss=0.1174, over 5669337.29 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3577, pruned_loss=0.112, over 5674184.25 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3741, pruned_loss=0.1185, over 5667824.56 frames. ], batch size: 82, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:57:33,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2683, 1.8552, 1.3727, 0.5416], device='cuda:0'), covar=tensor([0.5086, 0.2655, 0.3698, 0.6635], device='cuda:0'), in_proj_covar=tensor([0.1847, 0.1733, 0.1657, 0.1500], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 17:57:36,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.136e+03 1.826e+03 2.360e+03 3.117e+03 5.075e+03, threshold=4.720e+03, percent-clipped=10.0
+2023-03-14 17:57:49,439 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:57:56,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273148.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:57:56,938 INFO [train.py:968] (0/2) Epoch 28, batch 42250, giga_loss[loss=0.3254, simple_loss=0.3789, pruned_loss=0.1359, over 28823.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3717, pruned_loss=0.1182, over 5673596.00 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3571, pruned_loss=0.1117, over 5679659.20 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3744, pruned_loss=0.1195, over 5667538.39 frames. ], batch size: 99, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:58:39,273 INFO [train.py:968] (0/2) Epoch 28, batch 42300, giga_loss[loss=0.262, simple_loss=0.3332, pruned_loss=0.09542, over 28900.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3697, pruned_loss=0.1175, over 5683064.39 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3569, pruned_loss=0.1115, over 5682734.21 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3723, pruned_loss=0.1188, over 5675480.76 frames. ], batch size: 106, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:58:51,424 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1273212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:59:04,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.247e+03 1.931e+03 2.593e+03 3.348e+03 8.747e+03, threshold=5.186e+03, percent-clipped=10.0
+2023-03-14 17:59:25,851 INFO [train.py:968] (0/2) Epoch 28, batch 42350, giga_loss[loss=0.3184, simple_loss=0.3569, pruned_loss=0.1399, over 23506.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3696, pruned_loss=0.1191, over 5670948.94 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3568, pruned_loss=0.1114, over 5690166.88 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3721, pruned_loss=0.1205, over 5657745.23 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:59:57,172 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1273282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:59:59,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1273285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:00:05,406 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1273291.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:00:08,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1273294.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:00:12,968 INFO [train.py:968] (0/2) Epoch 28, batch 42400, giga_loss[loss=0.2854, simple_loss=0.3627, pruned_loss=0.1041, over 28921.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3683, pruned_loss=0.1181, over 5680130.64 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3567, pruned_loss=0.1113, over 5695521.27 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3707, pruned_loss=0.1194, over 5664189.47 frames. ], batch size: 186, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:00:24,358 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 18:00:27,692 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1273314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:00:35,133 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1273323.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:00:36,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.956e+02 1.905e+03 2.462e+03 3.464e+03 1.299e+04, threshold=4.925e+03, percent-clipped=9.0
+2023-03-14 18:00:47,304 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273337.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:00:51,528 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:00:56,786 INFO [train.py:968] (0/2) Epoch 28, batch 42450, giga_loss[loss=0.2459, simple_loss=0.3284, pruned_loss=0.08172, over 28675.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3665, pruned_loss=0.1152, over 5689927.50 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3565, pruned_loss=0.1112, over 5697549.41 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3689, pruned_loss=0.1165, over 5675201.66 frames. ], batch size: 99, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:00:58,457 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4964, 1.7120, 1.4093, 1.4086], device='cuda:0'), covar=tensor([0.0923, 0.0440, 0.0903, 0.0920], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0452, 0.0525, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 18:01:05,259 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4334, 1.5945, 1.6373, 1.2527], device='cuda:0'), covar=tensor([0.1783, 0.2617, 0.1526, 0.1742], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0717, 0.0981, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 18:01:45,251 INFO [train.py:968] (0/2) Epoch 28, batch 42500, giga_loss[loss=0.3014, simple_loss=0.3794, pruned_loss=0.1117, over 28949.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3664, pruned_loss=0.1141, over 5687739.96 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3566, pruned_loss=0.1111, over 5697934.86 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3684, pruned_loss=0.1152, over 5675488.34 frames. ], batch size: 164, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:02:10,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.071e+03 1.734e+03 2.130e+03 2.584e+03 7.376e+03, threshold=4.259e+03, percent-clipped=3.0
+2023-03-14 18:02:12,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5769, 1.6472, 1.7726, 1.3405], device='cuda:0'), covar=tensor([0.1864, 0.2636, 0.1566, 0.1789], device='cuda:0'), in_proj_covar=tensor([0.0932, 0.0716, 0.0980, 0.0879], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 18:02:23,173 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 18:02:31,514 INFO [train.py:968] (0/2) Epoch 28, batch 42550, giga_loss[loss=0.2526, simple_loss=0.3347, pruned_loss=0.08528, over 29013.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3662, pruned_loss=0.114, over 5693589.30 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3567, pruned_loss=0.1112, over 5700442.37 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3679, pruned_loss=0.115, over 5681420.57 frames. ], batch size: 155, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:02:37,696 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-14 18:02:40,970 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0931, 3.0039, 2.0755, 1.3165], device='cuda:0'), covar=tensor([0.7679, 0.3890, 0.4093, 0.6520], device='cuda:0'), in_proj_covar=tensor([0.1849, 0.1737, 0.1659, 0.1502], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 18:02:57,763 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1273480.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:03:00,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1273483.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:03:01,244 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1273484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:03:03,852 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1273487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:03:15,698 INFO [train.py:968] (0/2) Epoch 28, batch 42600, giga_loss[loss=0.3272, simple_loss=0.3844, pruned_loss=0.135, over 28010.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.364, pruned_loss=0.1133, over 5686418.16 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3567, pruned_loss=0.1111, over 5702465.81 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3656, pruned_loss=0.1141, over 5674462.75 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:03:28,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1273512.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:03:33,071 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1273516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:03:40,948 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.744e+03 2.348e+03 3.131e+03 8.859e+03, threshold=4.695e+03, percent-clipped=11.0
+2023-03-14 18:04:00,855 INFO [train.py:968] (0/2) Epoch 28, batch 42650, giga_loss[loss=0.2747, simple_loss=0.3477, pruned_loss=0.1008, over 29011.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3645, pruned_loss=0.1145, over 5684825.56 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.357, pruned_loss=0.1112, over 5707145.50 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3657, pruned_loss=0.1152, over 5670523.47 frames. ], batch size: 164, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:04:33,330 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7937, 1.7162, 1.9573, 1.5373], device='cuda:0'), covar=tensor([0.1860, 0.2587, 0.1513, 0.1791], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0717, 0.0981, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 18:04:37,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:04:45,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1273592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:04:46,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3779, 1.8390, 1.5296, 1.5730], device='cuda:0'), covar=tensor([0.0781, 0.0328, 0.0328, 0.0881], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 18:04:49,845 INFO [train.py:968] (0/2) Epoch 28, batch 42700, giga_loss[loss=0.297, simple_loss=0.3604, pruned_loss=0.1168, over 28997.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3646, pruned_loss=0.1156, over 5667301.79 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3571, pruned_loss=0.1112, over 5695485.03 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3655, pruned_loss=0.1162, over 5665912.19 frames. ], batch size: 128, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:05:14,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.212e+03 1.920e+03 2.249e+03 2.982e+03 6.980e+03, threshold=4.499e+03, percent-clipped=8.0
+2023-03-14 18:05:37,243 INFO [train.py:968] (0/2) Epoch 28, batch 42750, libri_loss[loss=0.2636, simple_loss=0.3365, pruned_loss=0.09533, over 29582.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3634, pruned_loss=0.1154, over 5676486.63 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3569, pruned_loss=0.1109, over 5700454.20 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3645, pruned_loss=0.1162, over 5670223.21 frames. ], batch size: 75, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:06:28,255 INFO [train.py:968] (0/2) Epoch 28, batch 42800, giga_loss[loss=0.3456, simple_loss=0.3974, pruned_loss=0.1469, over 27584.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3627, pruned_loss=0.1152, over 5679343.76 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3568, pruned_loss=0.1109, over 5701516.93 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3636, pruned_loss=0.1159, over 5673514.07 frames. ], batch size: 472, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 18:06:34,827 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-14 18:06:52,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.160e+03 1.748e+03 2.226e+03 3.202e+03 8.355e+03, threshold=4.452e+03, percent-clipped=9.0
+2023-03-14 18:06:56,771 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1273730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:07:00,283 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1273733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:07:16,037 INFO [train.py:968] (0/2) Epoch 28, batch 42850, giga_loss[loss=0.2981, simple_loss=0.3779, pruned_loss=0.1092, over 29001.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3635, pruned_loss=0.1158, over 5682043.71 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3572, pruned_loss=0.1111, over 5702698.38 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3641, pruned_loss=0.1162, over 5676245.64 frames. ], batch size: 155, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:07:26,091 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1273762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:07:56,197 INFO [train.py:968] (0/2) Epoch 28, batch 42900, giga_loss[loss=0.3125, simple_loss=0.3578, pruned_loss=0.1336, over 23537.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3639, pruned_loss=0.1154, over 5688035.30 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3569, pruned_loss=0.1109, over 5710511.51 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.365, pruned_loss=0.1161, over 5675260.58 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:08:15,794 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1273820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:08:21,164 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.196e+03 1.926e+03 2.425e+03 3.855e+03 9.691e+03, threshold=4.850e+03, percent-clipped=18.0
+2023-03-14 18:08:38,950 INFO [train.py:968] (0/2) Epoch 28, batch 42950, giga_loss[loss=0.2981, simple_loss=0.3628, pruned_loss=0.1167, over 28034.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3646, pruned_loss=0.1151, over 5690426.02 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3568, pruned_loss=0.1108, over 5710805.21 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3656, pruned_loss=0.1158, over 5679701.17 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:09:02,495 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3982, 1.5169, 1.5447, 1.3699], device='cuda:0'), covar=tensor([0.2654, 0.2550, 0.2271, 0.2606], device='cuda:0'), in_proj_covar=tensor([0.2096, 0.2058, 0.1972, 0.2111], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 18:09:23,340 INFO [train.py:968] (0/2) Epoch 28, batch 43000, giga_loss[loss=0.3086, simple_loss=0.3775, pruned_loss=0.1199, over 28622.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3643, pruned_loss=0.1142, over 5686142.05 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3572, pruned_loss=0.1111, over 5714630.05 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3649, pruned_loss=0.1147, over 5673336.14 frames. ], batch size: 242, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:09:47,981 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.670e+03 2.116e+03 3.162e+03 8.527e+03, threshold=4.232e+03, percent-clipped=9.0
+2023-03-14 18:10:02,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-14 18:10:06,858 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-14 18:10:11,951 INFO [train.py:968] (0/2) Epoch 28, batch 43050, libri_loss[loss=0.2433, simple_loss=0.3093, pruned_loss=0.08859, over 29646.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3641, pruned_loss=0.1149, over 5682124.15 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3565, pruned_loss=0.1107, over 5721023.49 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3656, pruned_loss=0.1157, over 5665061.23 frames. ], batch size: 69, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:10:26,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:10:30,658 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1273971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:10:37,002 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8871, 3.7213, 3.5630, 1.6276], device='cuda:0'), covar=tensor([0.0764, 0.0857, 0.0821, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.1323, 0.1219, 0.1027, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 18:10:56,117 INFO [train.py:968] (0/2) Epoch 28, batch 43100, giga_loss[loss=0.2934, simple_loss=0.3609, pruned_loss=0.113, over 28807.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.367, pruned_loss=0.118, over 5679025.72 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3557, pruned_loss=0.1102, over 5723629.37 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.369, pruned_loss=0.1192, over 5662104.65 frames. ], batch size: 199, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:10:58,645 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1274000.pt
+2023-03-14 18:11:25,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.992e+03 2.633e+03 3.895e+03 9.040e+03, threshold=5.266e+03, percent-clipped=17.0
+2023-03-14 18:11:48,020 INFO [train.py:968] (0/2) Epoch 28, batch 43150, giga_loss[loss=0.3221, simple_loss=0.3719, pruned_loss=0.1361, over 28586.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3689, pruned_loss=0.1209, over 5668523.03 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3559, pruned_loss=0.1103, over 5726691.77 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3706, pruned_loss=0.122, over 5651274.18 frames. ], batch size: 71, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:12:41,867 INFO [train.py:968] (0/2) Epoch 28, batch 43200, giga_loss[loss=0.4292, simple_loss=0.4635, pruned_loss=0.1974, over 24353.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3704, pruned_loss=0.123, over 5665511.22 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3558, pruned_loss=0.1103, over 5727696.31 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3719, pruned_loss=0.1239, over 5650909.25 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:12:52,634 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1274110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:12:55,072 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1274113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:13:08,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.252e+03 1.865e+03 2.257e+03 3.003e+03 7.105e+03, threshold=4.514e+03, percent-clipped=3.0
+2023-03-14 18:13:23,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1274142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:13:28,159 INFO [train.py:968] (0/2) Epoch 28, batch 43250, giga_loss[loss=0.268, simple_loss=0.3415, pruned_loss=0.09721, over 28654.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3696, pruned_loss=0.1221, over 5676191.78 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3559, pruned_loss=0.1103, over 5729528.58 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3709, pruned_loss=0.123, over 5662564.46 frames. ], batch size: 307, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:13:59,158 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6935, 1.7079, 1.8409, 1.4308], device='cuda:0'), covar=tensor([0.1712, 0.2458, 0.1395, 0.1650], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0719, 0.0982, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 18:14:12,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1274195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:14:14,631 INFO [train.py:968] (0/2) Epoch 28, batch 43300, giga_loss[loss=0.2602, simple_loss=0.3331, pruned_loss=0.09363, over 28431.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3683, pruned_loss=0.1212, over 5677544.63 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3558, pruned_loss=0.1102, over 5730448.21 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3695, pruned_loss=0.1221, over 5665450.90 frames. ], batch size: 78, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:14:39,766 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.126e+03 1.829e+03 2.537e+03 3.748e+03 7.827e+03, threshold=5.074e+03, percent-clipped=16.0
+2023-03-14 18:14:58,298 INFO [train.py:968] (0/2) Epoch 28, batch 43350, giga_loss[loss=0.2721, simple_loss=0.3393, pruned_loss=0.1024, over 27868.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.368, pruned_loss=0.1191, over 5686458.00 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3559, pruned_loss=0.1102, over 5731879.20 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3691, pruned_loss=0.12, over 5673768.41 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:15:41,385 INFO [train.py:968] (0/2) Epoch 28, batch 43400, giga_loss[loss=0.2935, simple_loss=0.3607, pruned_loss=0.1131, over 28860.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3673, pruned_loss=0.1185, over 5664211.55 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3564, pruned_loss=0.1105, over 5713192.17 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3682, pruned_loss=0.1193, over 5668720.84 frames. ], batch size: 199, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:15:41,899 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-14 18:16:05,282 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.879e+03 2.554e+03 3.229e+03 9.950e+03, threshold=5.108e+03, percent-clipped=8.0
+2023-03-14 18:16:15,875 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1274338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:16:18,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1274341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:16:22,701 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1274346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:16:24,984 INFO [train.py:968] (0/2) Epoch 28, batch 43450, giga_loss[loss=0.2936, simple_loss=0.3613, pruned_loss=0.113, over 28923.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3667, pruned_loss=0.1191, over 5654545.83 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3569, pruned_loss=0.1108, over 5706062.17 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3672, pruned_loss=0.1195, over 5663519.24 frames. ], batch size: 145, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:16:44,167 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1274370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:16:54,047 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3803, 1.5715, 1.5848, 1.2083], device='cuda:0'), covar=tensor([0.1628, 0.2499, 0.1412, 0.1653], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0719, 0.0983, 0.0881], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 18:17:09,579 INFO [train.py:968] (0/2) Epoch 28, batch 43500, libri_loss[loss=0.2922, simple_loss=0.3672, pruned_loss=0.1086, over 27640.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3651, pruned_loss=0.1187, over 5643979.13 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3571, pruned_loss=0.1111, over 5691590.50 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3654, pruned_loss=0.1191, over 5662220.89 frames. ], batch size: 115, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:17:09,879 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1274399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:17:21,725 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1274410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:17:37,385 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.284e+03 1.756e+03 2.272e+03 2.992e+03 8.927e+03, threshold=4.544e+03, percent-clipped=3.0
+2023-03-14 18:17:54,491 INFO [train.py:968] (0/2) Epoch 28, batch 43550, giga_loss[loss=0.3471, simple_loss=0.4036, pruned_loss=0.1453, over 28550.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.367, pruned_loss=0.1203, over 5649859.40 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3568, pruned_loss=0.1107, over 5696172.31 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3679, pruned_loss=0.1211, over 5659065.80 frames. ], batch size: 336, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:18:30,871 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1274489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:18:33,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1274492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:18:40,339 INFO [train.py:968] (0/2) Epoch 28, batch 43600, giga_loss[loss=0.2981, simple_loss=0.3704, pruned_loss=0.1129, over 28899.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3692, pruned_loss=0.1206, over 5659384.50 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3562, pruned_loss=0.1104, over 5700698.98 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3706, pruned_loss=0.1218, over 5661834.50 frames. ], batch size: 66, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:18:57,991 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-14 18:19:01,309 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1274521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:19:06,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.672e+03 2.150e+03 2.895e+03 6.030e+03, threshold=4.300e+03, percent-clipped=5.0
+2023-03-14 18:19:30,423 INFO [train.py:968] (0/2) Epoch 28, batch 43650, giga_loss[loss=0.2638, simple_loss=0.3506, pruned_loss=0.08844, over 28975.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3714, pruned_loss=0.119, over 5655778.29 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3562, pruned_loss=0.1104, over 5702483.33 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3726, pruned_loss=0.12, over 5655848.09 frames. ], batch size: 199, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:19:32,090 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4794, 4.9050, 1.6555, 1.9142], device='cuda:0'), covar=tensor([0.1062, 0.0429, 0.0988, 0.1270], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0577, 0.0414, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 18:20:19,713 INFO [train.py:968] (0/2) Epoch 28, batch 43700, giga_loss[loss=0.3162, simple_loss=0.381, pruned_loss=0.1257, over 28911.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3731, pruned_loss=0.1199, over 5666392.09 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3566, pruned_loss=0.1108, over 5702821.88 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3738, pruned_loss=0.1204, over 5665612.41 frames. ], batch size: 106, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:20:40,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-14 18:20:49,464 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.255e+03 1.902e+03 2.208e+03 3.143e+03 6.895e+03, threshold=4.415e+03, percent-clipped=9.0
+2023-03-14 18:21:09,237 INFO [train.py:968] (0/2) Epoch 28, batch 43750, giga_loss[loss=0.3233, simple_loss=0.3681, pruned_loss=0.1393, over 23554.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.375, pruned_loss=0.1217, over 5666877.00 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3567, pruned_loss=0.1108, over 5703461.77 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3756, pruned_loss=0.1221, over 5665577.32 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:21:13,994 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-14 18:21:36,207 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 18:21:55,803 INFO [train.py:968] (0/2) Epoch 28, batch 43800, giga_loss[loss=0.3422, simple_loss=0.3941, pruned_loss=0.1451, over 27879.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3746, pruned_loss=0.1219, over 5674259.22 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3566, pruned_loss=0.1107, over 5705726.83 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3755, pruned_loss=0.1225, over 5670453.24 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:22:00,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4054, 1.6658, 1.6761, 1.3988], device='cuda:0'), covar=tensor([0.2188, 0.2072, 0.2378, 0.2224], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0764, 0.0733, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 18:22:12,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-14 18:22:18,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.176e+03 1.818e+03 2.263e+03 2.990e+03 1.266e+04, threshold=4.527e+03, percent-clipped=6.0
+2023-03-14 18:22:32,178 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1274741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:22:33,917 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-14 18:22:38,639 INFO [train.py:968] (0/2) Epoch 28, batch 43850, giga_loss[loss=0.3283, simple_loss=0.3926, pruned_loss=0.132, over 28273.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3742, pruned_loss=0.1226, over 5670239.44 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3569, pruned_loss=0.111, over 5708768.41 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3752, pruned_loss=0.1233, over 5663331.91 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:22:55,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5701, 1.8768, 1.4972, 1.6488], device='cuda:0'), covar=tensor([0.2715, 0.2780, 0.3213, 0.2461], device='cuda:0'), in_proj_covar=tensor([0.1609, 0.1158, 0.1422, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 18:23:02,393 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1274774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:23:12,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1274785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:23:24,284 INFO [train.py:968] (0/2) Epoch 28, batch 43900, giga_loss[loss=0.333, simple_loss=0.3783, pruned_loss=0.1439, over 28806.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3739, pruned_loss=0.1232, over 5663477.93 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3577, pruned_loss=0.1116, over 5702475.19 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3743, pruned_loss=0.1234, over 5661678.03 frames. ], batch size: 119, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 18:23:25,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5765, 1.8107, 1.4586, 1.7687], device='cuda:0'), covar=tensor([0.2721, 0.2889, 0.3232, 0.2520], device='cuda:0'), in_proj_covar=tensor([0.1613, 0.1161, 0.1424, 0.1021], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 18:23:27,219 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-14 18:23:52,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.201e+03 1.937e+03 2.447e+03 3.952e+03 1.290e+04, threshold=4.894e+03, percent-clipped=21.0
+2023-03-14 18:24:08,825 INFO [train.py:968] (0/2) Epoch 28, batch 43950, giga_loss[loss=0.2583, simple_loss=0.338, pruned_loss=0.08932, over 29049.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3725, pruned_loss=0.1233, over 5660345.11 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3579, pruned_loss=0.1118, over 5697921.98 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3731, pruned_loss=0.1236, over 5662091.35 frames. ], batch size: 155, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 18:24:21,264 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1274861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:24:58,438 INFO [train.py:968] (0/2) Epoch 28, batch 44000, giga_loss[loss=0.3628, simple_loss=0.4171, pruned_loss=0.1543, over 28221.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3727, pruned_loss=0.1243, over 5652740.38 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3574, pruned_loss=0.1115, over 5703293.47 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3739, pruned_loss=0.1251, over 5648178.50 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:25:17,109 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1274917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:25:18,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1274920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:25:26,465 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1274928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:25:27,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.176e+03 1.891e+03 2.255e+03 3.107e+03 9.337e+03, threshold=4.510e+03, percent-clipped=5.0
+2023-03-14 18:25:30,372 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1274931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:25:40,089 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4943, 1.2157, 4.1752, 3.4443], device='cuda:0'), covar=tensor([0.1599, 0.2833, 0.0476, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0677, 0.1016, 0.0991], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 18:25:47,204 INFO [train.py:968] (0/2) Epoch 28, batch 44050, giga_loss[loss=0.3125, simple_loss=0.3645, pruned_loss=0.1303, over 28654.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3719, pruned_loss=0.1241, over 5644279.46 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3577, pruned_loss=0.1115, over 5695657.01 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3729, pruned_loss=0.1249, over 5647160.90 frames. ], batch size: 85, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:25:47,506 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1274949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:25:57,731 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1274960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:26:29,762 INFO [train.py:968] (0/2) Epoch 28, batch 44100, giga_loss[loss=0.3111, simple_loss=0.3772, pruned_loss=0.1225, over 28835.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3701, pruned_loss=0.1233, over 5659551.17 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3579, pruned_loss=0.1118, over 5697330.69 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.371, pruned_loss=0.1239, over 5659159.11 frames. ], batch size: 186, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:26:40,646 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1389, 4.0253, 1.4778, 1.4865], device='cuda:0'), covar=tensor([0.1313, 0.0414, 0.0998, 0.1571], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0577, 0.0413, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 18:26:48,078 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3274, 1.9153, 1.4375, 0.5193], device='cuda:0'), covar=tensor([0.4381, 0.3412, 0.4340, 0.6623], device='cuda:0'), in_proj_covar=tensor([0.1853, 0.1744, 0.1662, 0.1508], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 18:26:52,926 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5171, 1.7702, 1.6470, 1.5717], device='cuda:0'), covar=tensor([0.1827, 0.2108, 0.2084, 0.2034], device='cuda:0'), in_proj_covar=tensor([0.0510, 0.0767, 0.0737, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 18:26:56,398 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.327e+03 1.814e+03 2.453e+03 3.413e+03 7.398e+03, threshold=4.906e+03, percent-clipped=9.0
+2023-03-14 18:27:12,351 INFO [train.py:968] (0/2) Epoch 28, batch 44150, giga_loss[loss=0.3325, simple_loss=0.385, pruned_loss=0.14, over 28805.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3688, pruned_loss=0.1221, over 5651355.40 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3582, pruned_loss=0.112, over 5692974.38 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3696, pruned_loss=0.1228, over 5653369.22 frames. ], batch size: 99, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:27:33,222 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-14 18:27:57,367 INFO [train.py:968] (0/2) Epoch 28, batch 44200, giga_loss[loss=0.2726, simple_loss=0.3487, pruned_loss=0.09827, over 28565.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3689, pruned_loss=0.1215, over 5649541.63 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3581, pruned_loss=0.1121, over 5686046.12 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1222, over 5656627.49 frames. ], batch size: 78, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:28:17,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1275116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:28:31,493 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.281e+03 1.916e+03 2.480e+03 3.212e+03 1.196e+04, threshold=4.961e+03, percent-clipped=5.0
+2023-03-14 18:28:34,075 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1275131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:28:48,289 INFO [train.py:968] (0/2) Epoch 28, batch 44250, giga_loss[loss=0.3322, simple_loss=0.3948, pruned_loss=0.1347, over 27876.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3727, pruned_loss=0.1237, over 5637697.30 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3585, pruned_loss=0.1124, over 5679130.05 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3733, pruned_loss=0.1241, over 5649777.59 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:28:50,085 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4900, 1.1970, 4.4844, 3.3469], device='cuda:0'), covar=tensor([0.1758, 0.3088, 0.0435, 0.1247], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0677, 0.1015, 0.0990], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 18:29:17,964 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0524, 1.2740, 1.2447, 1.0442], device='cuda:0'), covar=tensor([0.2340, 0.2765, 0.1542, 0.2063], device='cuda:0'), in_proj_covar=tensor([0.2096, 0.2065, 0.1980, 0.2118], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 18:29:35,608 INFO [train.py:968] (0/2) Epoch 28, batch 44300, giga_loss[loss=0.304, simple_loss=0.3745, pruned_loss=0.1167, over 28867.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3713, pruned_loss=0.1231, over 5650058.17 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3587, pruned_loss=0.1126, over 5683165.68 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.372, pruned_loss=0.1236, over 5655124.51 frames. ], batch size: 174, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:30:03,546 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.807e+03 2.223e+03 3.020e+03 6.830e+03, threshold=4.447e+03, percent-clipped=1.0
+2023-03-14 18:30:07,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1275236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:30:20,123 INFO [train.py:968] (0/2) Epoch 28, batch 44350, giga_loss[loss=0.294, simple_loss=0.3658, pruned_loss=0.1112, over 28693.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3717, pruned_loss=0.1212, over 5661457.13 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3584, pruned_loss=0.1124, over 5686655.40 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3728, pruned_loss=0.1219, over 5661790.86 frames. ], batch size: 307, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:30:31,262 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1275259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:30:33,389 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1275262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:30:57,459 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1275291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:31:03,324 INFO [train.py:968] (0/2) Epoch 28, batch 44400, giga_loss[loss=0.2681, simple_loss=0.3597, pruned_loss=0.08824, over 28756.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3721, pruned_loss=0.1191, over 5656608.16 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3587, pruned_loss=0.1125, over 5683258.01 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3731, pruned_loss=0.1198, over 5658749.09 frames. ], batch size: 99, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:31:32,504 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.740e+02 1.598e+03 2.343e+03 3.094e+03 9.885e+03, threshold=4.685e+03, percent-clipped=10.0
+2023-03-14 18:31:44,070 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1275340.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:31:48,318 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8373, 2.8886, 1.8055, 0.9573], device='cuda:0'), covar=tensor([0.8049, 0.3576, 0.3810, 0.7496], device='cuda:0'), in_proj_covar=tensor([0.1853, 0.1743, 0.1660, 0.1509], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 18:31:53,328 INFO [train.py:968] (0/2) Epoch 28, batch 44450, giga_loss[loss=0.3649, simple_loss=0.434, pruned_loss=0.1479, over 28877.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3761, pruned_loss=0.1209, over 5657429.99 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3588, pruned_loss=0.1126, over 5683131.16 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.377, pruned_loss=0.1214, over 5658730.38 frames. ], batch size: 199, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:31:56,300 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-14 18:32:11,226 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1275367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:32:22,240 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1275379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:32:25,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1275382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:32:44,358 INFO [train.py:968] (0/2) Epoch 28, batch 44500, giga_loss[loss=0.3559, simple_loss=0.4098, pruned_loss=0.151, over 28556.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3784, pruned_loss=0.1235, over 5657828.13 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3584, pruned_loss=0.1122, over 5687385.89 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3798, pruned_loss=0.1245, over 5654384.83 frames. ], batch size: 336, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:32:55,153 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1275411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:33:12,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.076e+03 1.796e+03 2.331e+03 3.000e+03 7.690e+03, threshold=4.662e+03, percent-clipped=8.0
+2023-03-14 18:33:15,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6188, 1.8369, 1.2863, 1.4461], device='cuda:0'), covar=tensor([0.1126, 0.0694, 0.1092, 0.1316], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0455, 0.0527, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 18:33:30,307 INFO [train.py:968] (0/2) Epoch 28, batch 44550, giga_loss[loss=0.3323, simple_loss=0.3899, pruned_loss=0.1374, over 28631.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3797, pruned_loss=0.126, over 5639357.19 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3585, pruned_loss=0.1124, over 5671020.62 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3811, pruned_loss=0.1268, over 5650978.97 frames. ], batch size: 65, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:34:18,645 INFO [train.py:968] (0/2) Epoch 28, batch 44600, giga_loss[loss=0.2937, simple_loss=0.3661, pruned_loss=0.1107, over 28796.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3794, pruned_loss=0.1263, over 5659751.37 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3586, pruned_loss=0.1125, over 5674350.62 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3807, pruned_loss=0.1271, over 5665549.22 frames. ], batch size: 119, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:34:25,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1275506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:34:45,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.682e+02 1.645e+03 2.085e+03 2.739e+03 7.779e+03, threshold=4.170e+03, percent-clipped=3.0
+2023-03-14 18:35:01,510 INFO [train.py:968] (0/2) Epoch 28, batch 44650, giga_loss[loss=0.2607, simple_loss=0.3455, pruned_loss=0.08794, over 28857.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3771, pruned_loss=0.1244, over 5658629.17 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3586, pruned_loss=0.1126, over 5677532.56 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3787, pruned_loss=0.1253, over 5659982.60 frames. ], batch size: 145, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:35:40,809 INFO [train.py:968] (0/2) Epoch 28, batch 44700, giga_loss[loss=0.3269, simple_loss=0.3998, pruned_loss=0.127, over 28687.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3744, pruned_loss=0.1209, over 5681788.78 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3577, pruned_loss=0.112, over 5688865.34 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3774, pruned_loss=0.1227, over 5672114.66 frames. ], batch size: 262, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:35:49,134 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-14 18:36:09,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.281e+03 1.800e+03 2.351e+03 3.100e+03 1.021e+04, threshold=4.702e+03, percent-clipped=12.0
+2023-03-14 18:36:26,394 INFO [train.py:968] (0/2) Epoch 28, batch 44750, giga_loss[loss=0.3371, simple_loss=0.402, pruned_loss=0.1361, over 28828.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3753, pruned_loss=0.1205, over 5664600.36 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3577, pruned_loss=0.1121, over 5672217.96 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3781, pruned_loss=0.122, over 5671736.37 frames. ], batch size: 119, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:36:26,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1275649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:36:28,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1275652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:36:51,594 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1275681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:37:09,609 INFO [train.py:968] (0/2) Epoch 28, batch 44800, giga_loss[loss=0.3406, simple_loss=0.3971, pruned_loss=0.142, over 28297.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3766, pruned_loss=0.1219, over 5591527.42 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3587, pruned_loss=0.1131, over 5602785.30 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3784, pruned_loss=0.1225, over 5659550.67 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:37:23,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1275715.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:37:37,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.948e+03 2.504e+03 4.033e+03 1.293e+04, threshold=5.008e+03, percent-clipped=16.0
+2023-03-14 18:37:44,364 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 18:37:50,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1275742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:37:56,489 INFO [train.py:968] (0/2) Epoch 28, batch 44850, giga_loss[loss=0.3442, simple_loss=0.3888, pruned_loss=0.1498, over 23679.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3784, pruned_loss=0.1241, over 5560326.35 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3594, pruned_loss=0.1137, over 5560112.39 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3796, pruned_loss=0.1241, over 5651604.19 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:38:42,349 INFO [train.py:968] (0/2) Epoch 28, batch 44900, giga_loss[loss=0.2522, simple_loss=0.3285, pruned_loss=0.08795, over 28760.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3762, pruned_loss=0.1227, over 5567055.62 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3598, pruned_loss=0.114, over 5536319.98 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.377, pruned_loss=0.1226, over 5660120.39 frames. ], batch size: 119, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:38:45,969 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9265, 2.0890, 1.7125, 1.9767], device='cuda:0'), covar=tensor([0.2877, 0.3024, 0.3417, 0.2885], device='cuda:0'), in_proj_covar=tensor([0.1610, 0.1160, 0.1425, 0.1021], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 18:39:03,326 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-14 18:39:05,414 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-28.pt
+2023-03-14 18:39:45,795 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.636e+03 2.029e+03 2.857e+03 7.185e+03, threshold=4.058e+03, percent-clipped=5.0
+2023-03-14 18:40:08,209 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1275858.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:40:11,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1275861.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:40:20,511 INFO [train.py:968] (0/2) Epoch 29, batch 50, giga_loss[loss=0.273, simple_loss=0.3556, pruned_loss=0.09516, over 28855.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.372, pruned_loss=0.1057, over 1262133.69 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3535, pruned_loss=0.08813, over 115631.69 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3736, pruned_loss=0.1073, over 1170121.25 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:40:33,470 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1275885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:40:36,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1275888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:40:37,695 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1275890.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:41:01,812 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1275917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:41:05,393 INFO [train.py:968] (0/2) Epoch 29, batch 100, giga_loss[loss=0.2505, simple_loss=0.3355, pruned_loss=0.08278, over 28857.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3616, pruned_loss=0.101, over 2248317.09 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3482, pruned_loss=0.08796, over 286389.74 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3635, pruned_loss=0.1026, over 2064280.47 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:41:07,566 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1275923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:41:13,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.978e+02 1.310e+03 1.513e+03 2.092e+03 6.253e+03, threshold=3.025e+03, percent-clipped=4.0
+2023-03-14 18:41:49,594 INFO [train.py:968] (0/2) Epoch 29, batch 150, giga_loss[loss=0.2732, simple_loss=0.3308, pruned_loss=0.1078, over 26635.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3474, pruned_loss=0.09506, over 3011835.08 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3463, pruned_loss=0.0869, over 423574.83 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.348, pruned_loss=0.09624, over 2793708.84 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:42:12,213 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1276000.pt
+2023-03-14 18:42:28,653 INFO [train.py:968] (0/2) Epoch 29, batch 200, giga_loss[loss=0.2217, simple_loss=0.3058, pruned_loss=0.06881, over 29098.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3358, pruned_loss=0.08967, over 3619302.92 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3454, pruned_loss=0.08766, over 612803.81 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3353, pruned_loss=0.09028, over 3362064.81 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:42:36,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.547e+02 1.215e+03 1.410e+03 1.901e+03 5.531e+03, threshold=2.821e+03, percent-clipped=5.0
+2023-03-14 18:42:57,696 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:43:10,199 INFO [train.py:968] (0/2) Epoch 29, batch 250, giga_loss[loss=0.1938, simple_loss=0.277, pruned_loss=0.05527, over 28610.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.325, pruned_loss=0.08478, over 4084744.14 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3436, pruned_loss=0.08663, over 717777.46 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.324, pruned_loss=0.08509, over 3844097.14 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:43:19,828 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:43:29,187 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1318, 1.1674, 3.5140, 3.1319], device='cuda:0'), covar=tensor([0.2084, 0.3321, 0.0919, 0.1419], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0680, 0.1021, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 18:43:46,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5341, 4.3764, 4.1199, 2.1204], device='cuda:0'), covar=tensor([0.0512, 0.0691, 0.0718, 0.1901], device='cuda:0'), in_proj_covar=tensor([0.1321, 0.1219, 0.1028, 0.0758], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 18:43:47,816 INFO [train.py:968] (0/2) Epoch 29, batch 300, giga_loss[loss=0.1997, simple_loss=0.2744, pruned_loss=0.06248, over 28591.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3174, pruned_loss=0.08103, over 4447892.03 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3413, pruned_loss=0.08554, over 991444.04 frames. ], giga_tot_loss[loss=0.2386, simple_loss=0.3151, pruned_loss=0.08103, over 4172874.00 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:43:55,580 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.944e+02 1.108e+03 1.337e+03 1.806e+03 4.212e+03, threshold=2.674e+03, percent-clipped=8.0
+2023-03-14 18:44:25,110 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:44:30,404 INFO [train.py:968] (0/2) Epoch 29, batch 350, giga_loss[loss=0.2573, simple_loss=0.3109, pruned_loss=0.1019, over 26622.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3101, pruned_loss=0.07787, over 4727225.29 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3414, pruned_loss=0.08568, over 1088984.51 frames. ], giga_tot_loss[loss=0.2312, simple_loss=0.3073, pruned_loss=0.07752, over 4485483.08 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:44:44,814 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:44:56,287 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6109, 1.7951, 1.8798, 1.4047], device='cuda:0'), covar=tensor([0.1967, 0.2774, 0.1609, 0.1772], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0722, 0.0990, 0.0888], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 18:45:08,117 INFO [train.py:968] (0/2) Epoch 29, batch 400, giga_loss[loss=0.2399, simple_loss=0.3065, pruned_loss=0.08671, over 28798.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.306, pruned_loss=0.07644, over 4951372.58 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3411, pruned_loss=0.08626, over 1208587.50 frames. ], giga_tot_loss[loss=0.2271, simple_loss=0.3028, pruned_loss=0.07572, over 4734399.70 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:45:17,739 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.917e+02 1.099e+03 1.408e+03 1.844e+03 5.261e+03, threshold=2.816e+03, percent-clipped=7.0
+2023-03-14 18:45:44,919 INFO [train.py:968] (0/2) Epoch 29, batch 450, giga_loss[loss=0.1945, simple_loss=0.2708, pruned_loss=0.05905, over 28568.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3056, pruned_loss=0.07664, over 5118851.22 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3414, pruned_loss=0.0876, over 1426114.81 frames. ], giga_tot_loss[loss=0.2258, simple_loss=0.3012, pruned_loss=0.07525, over 4913036.29 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:45:51,573 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8081, 5.6078, 5.3402, 2.8876], device='cuda:0'), covar=tensor([0.0507, 0.0692, 0.0869, 0.1651], device='cuda:0'), in_proj_covar=tensor([0.1314, 0.1215, 0.1024, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 18:45:59,087 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:46:09,069 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:46:26,530 INFO [train.py:968] (0/2) Epoch 29, batch 500, giga_loss[loss=0.2234, simple_loss=0.295, pruned_loss=0.07593, over 27666.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3022, pruned_loss=0.07449, over 5257195.42 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3395, pruned_loss=0.08634, over 1578539.14 frames. ], giga_tot_loss[loss=0.2222, simple_loss=0.2979, pruned_loss=0.07328, over 5070831.64 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:46:35,583 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.665e+02 1.160e+03 1.417e+03 1.966e+03 4.728e+03, threshold=2.835e+03, percent-clipped=10.0
+2023-03-14 18:47:05,748 INFO [train.py:968] (0/2) Epoch 29, batch 550, giga_loss[loss=0.225, simple_loss=0.3062, pruned_loss=0.07195, over 28598.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3001, pruned_loss=0.07371, over 5348807.00 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3401, pruned_loss=0.08675, over 1676012.72 frames. ], giga_tot_loss[loss=0.2202, simple_loss=0.2956, pruned_loss=0.07234, over 5194959.99 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:47:47,967 INFO [train.py:968] (0/2) Epoch 29, batch 600, giga_loss[loss=0.2308, simple_loss=0.3099, pruned_loss=0.07586, over 28841.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.2992, pruned_loss=0.07354, over 5422237.32 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3392, pruned_loss=0.08642, over 1788619.91 frames. ], giga_tot_loss[loss=0.2196, simple_loss=0.2948, pruned_loss=0.07216, over 5296089.36 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:47:53,238 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5748, 1.8837, 1.5410, 1.2997], device='cuda:0'), covar=tensor([0.2802, 0.2960, 0.3391, 0.2783], device='cuda:0'), in_proj_covar=tensor([0.1614, 0.1161, 0.1427, 0.1021], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 18:47:55,647 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1350, 2.5446, 2.4862, 1.8660], device='cuda:0'), covar=tensor([0.4000, 0.2599, 0.2438, 0.3632], device='cuda:0'), in_proj_covar=tensor([0.2079, 0.2040, 0.1955, 0.2097], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 18:47:57,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.056e+02 1.194e+03 1.636e+03 2.130e+03 5.015e+03, threshold=3.272e+03, percent-clipped=8.0
+2023-03-14 18:47:59,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:48:06,639 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:48:12,272 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:48:22,541 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:48:35,501 INFO [train.py:968] (0/2) Epoch 29, batch 650, giga_loss[loss=0.2152, simple_loss=0.2881, pruned_loss=0.07114, over 28561.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2958, pruned_loss=0.07218, over 5472056.22 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.339, pruned_loss=0.08626, over 1809221.81 frames. ], giga_tot_loss[loss=0.2172, simple_loss=0.2922, pruned_loss=0.07107, over 5371253.03 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:48:37,794 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:49:06,956 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2755, 3.2825, 2.2890, 1.4326], device='cuda:0'), covar=tensor([0.8066, 0.3076, 0.4264, 0.7091], device='cuda:0'), in_proj_covar=tensor([0.1843, 0.1736, 0.1657, 0.1501], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 18:49:22,068 INFO [train.py:968] (0/2) Epoch 29, batch 700, giga_loss[loss=0.269, simple_loss=0.3232, pruned_loss=0.1074, over 26581.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2937, pruned_loss=0.07149, over 5523262.54 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3384, pruned_loss=0.08593, over 1869553.06 frames. ], giga_tot_loss[loss=0.2157, simple_loss=0.2903, pruned_loss=0.0705, over 5438797.86 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:49:32,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.196e+02 1.106e+03 1.372e+03 1.685e+03 4.039e+03, threshold=2.745e+03, percent-clipped=3.0
+2023-03-14 18:49:39,379 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:49:53,179 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:04,622 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:07,376 INFO [train.py:968] (0/2) Epoch 29, batch 750, giga_loss[loss=0.2114, simple_loss=0.2829, pruned_loss=0.06997, over 28855.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2916, pruned_loss=0.07023, over 5570614.47 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3376, pruned_loss=0.08519, over 1968038.99 frames. ], giga_tot_loss[loss=0.2133, simple_loss=0.2881, pruned_loss=0.0693, over 5496362.88 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:50:11,232 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:14,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:32,702 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:34,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:38,103 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:41,152 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:48,588 INFO [train.py:968] (0/2) Epoch 29, batch 800, libri_loss[loss=0.2845, simple_loss=0.3669, pruned_loss=0.1011, over 29531.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2903, pruned_loss=0.06994, over 5608736.37 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3375, pruned_loss=0.08549, over 2104172.47 frames. ], giga_tot_loss[loss=0.2114, simple_loss=0.2857, pruned_loss=0.06854, over 5538283.00 frames. ], batch size: 84, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:50:50,340 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:56,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.533e+02 1.122e+03 1.421e+03 1.900e+03 6.600e+03, threshold=2.843e+03, percent-clipped=8.0
+2023-03-14 18:50:56,752 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:51:24,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:51:26,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5090, 1.7242, 1.2278, 1.3092], device='cuda:0'), covar=tensor([0.1095, 0.0610, 0.1098, 0.1249], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0452, 0.0527, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 18:51:29,328 INFO [train.py:968] (0/2) Epoch 29, batch 850, libri_loss[loss=0.256, simple_loss=0.3397, pruned_loss=0.08617, over 29061.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2966, pruned_loss=0.07375, over 5620078.52 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3372, pruned_loss=0.08532, over 2304992.48 frames. ], giga_tot_loss[loss=0.2173, simple_loss=0.2907, pruned_loss=0.07198, over 5550182.68 frames. ], batch size: 101, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:51:39,011 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:51:40,895 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:52:02,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:52:06,551 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:52:07,795 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:52:12,514 INFO [train.py:968] (0/2) Epoch 29, batch 900, giga_loss[loss=0.3254, simple_loss=0.3876, pruned_loss=0.1316, over 28752.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.3083, pruned_loss=0.07934, over 5639062.83 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3371, pruned_loss=0.08543, over 2390524.22 frames. ], giga_tot_loss[loss=0.2292, simple_loss=0.3029, pruned_loss=0.07774, over 5580331.51 frames. ], batch size: 99, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:52:15,606 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 18:52:24,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.454e+03 1.886e+03 2.472e+03 6.420e+03, threshold=3.772e+03, percent-clipped=19.0
+2023-03-14 18:52:32,625 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:52:33,736 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9221, 3.7012, 3.5172, 1.8948], device='cuda:0'), covar=tensor([0.0774, 0.1009, 0.0971, 0.2131], device='cuda:0'), in_proj_covar=tensor([0.1301, 0.1203, 0.1015, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 18:52:33,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1493, 1.1771, 3.3208, 3.0045], device='cuda:0'), covar=tensor([0.1767, 0.2993, 0.0605, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0676, 0.1013, 0.0986], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 18:52:34,000 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-14 18:52:54,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7241, 1.8451, 1.5546, 1.8190], device='cuda:0'), covar=tensor([0.2654, 0.2893, 0.3173, 0.2456], device='cuda:0'), in_proj_covar=tensor([0.1613, 0.1161, 0.1425, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 18:52:57,106 INFO [train.py:968] (0/2) Epoch 29, batch 950, giga_loss[loss=0.3223, simple_loss=0.3891, pruned_loss=0.1278, over 28776.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3213, pruned_loss=0.08592, over 5645110.28 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3372, pruned_loss=0.08552, over 2398888.60 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3171, pruned_loss=0.08468, over 5605778.41 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:53:27,748 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 18:53:29,693 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:53:30,440 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-14 18:53:31,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:53:40,503 INFO [train.py:968] (0/2) Epoch 29, batch 1000, giga_loss[loss=0.2459, simple_loss=0.3292, pruned_loss=0.08127, over 28557.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3298, pruned_loss=0.09004, over 5649070.92 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3371, pruned_loss=0.08553, over 2440295.69 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3265, pruned_loss=0.08911, over 5624648.30 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:53:49,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.301e+03 1.688e+03 2.127e+03 6.508e+03, threshold=3.377e+03, percent-clipped=4.0
+2023-03-14 18:53:55,946 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:54:19,354 INFO [train.py:968] (0/2) Epoch 29, batch 1050, giga_loss[loss=0.2641, simple_loss=0.3563, pruned_loss=0.08595, over 28624.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3345, pruned_loss=0.09111, over 5653674.54 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3378, pruned_loss=0.08593, over 2494558.27 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3316, pruned_loss=0.09031, over 5642219.77 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:55:04,333 INFO [train.py:968] (0/2) Epoch 29, batch 1100, giga_loss[loss=0.2511, simple_loss=0.3361, pruned_loss=0.08306, over 28995.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.337, pruned_loss=0.09111, over 5656309.45 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3386, pruned_loss=0.08628, over 2562248.45 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3343, pruned_loss=0.09042, over 5643043.24 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:55:04,880 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-14 18:55:11,463 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:55:14,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.318e+02 1.309e+03 1.735e+03 2.501e+03 6.323e+03, threshold=3.469e+03, percent-clipped=9.0
+2023-03-14 18:55:47,840 INFO [train.py:968] (0/2) Epoch 29, batch 1150, giga_loss[loss=0.2482, simple_loss=0.333, pruned_loss=0.08176, over 28868.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3375, pruned_loss=0.091, over 5673682.30 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3386, pruned_loss=0.08628, over 2562248.45 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3354, pruned_loss=0.09047, over 5663357.04 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:55:52,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2658, 2.5275, 1.2850, 1.4475], device='cuda:0'), covar=tensor([0.1081, 0.0371, 0.0962, 0.1454], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0571, 0.0413, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 18:56:00,248 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:56:10,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:56:12,990 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4270, 1.6231, 1.1343, 1.1853], device='cuda:0'), covar=tensor([0.1157, 0.0591, 0.1147, 0.1262], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0451, 0.0526, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 18:56:18,213 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-14 18:56:29,597 INFO [train.py:968] (0/2) Epoch 29, batch 1200, giga_loss[loss=0.2527, simple_loss=0.3305, pruned_loss=0.08747, over 28830.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3394, pruned_loss=0.09293, over 5663690.28 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3382, pruned_loss=0.08637, over 2625241.68 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3379, pruned_loss=0.09258, over 5654137.25 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:56:39,297 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.767e+02 1.294e+03 1.552e+03 1.855e+03 4.017e+03, threshold=3.103e+03, percent-clipped=2.0
+2023-03-14 18:56:55,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3337, 3.4140, 1.5745, 1.4968], device='cuda:0'), covar=tensor([0.1120, 0.0269, 0.0957, 0.1490], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0571, 0.0413, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 18:57:08,359 INFO [train.py:968] (0/2) Epoch 29, batch 1250, giga_loss[loss=0.2418, simple_loss=0.3256, pruned_loss=0.07903, over 28897.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3421, pruned_loss=0.09439, over 5660549.82 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3387, pruned_loss=0.08656, over 2677911.46 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3407, pruned_loss=0.09417, over 5659607.55 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:57:10,235 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:57:12,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:57:39,385 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:57:51,852 INFO [train.py:968] (0/2) Epoch 29, batch 1300, giga_loss[loss=0.3289, simple_loss=0.3928, pruned_loss=0.1325, over 28382.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3448, pruned_loss=0.09514, over 5669946.06 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3386, pruned_loss=0.0865, over 2742083.88 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3438, pruned_loss=0.09515, over 5664631.57 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:57:58,208 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:01,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:01,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.648e+02 1.237e+03 1.646e+03 2.063e+03 4.540e+03, threshold=3.292e+03, percent-clipped=8.0
+2023-03-14 18:58:06,268 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:08,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:22,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:28,774 INFO [train.py:968] (0/2) Epoch 29, batch 1350, giga_loss[loss=0.2703, simple_loss=0.3515, pruned_loss=0.0946, over 28971.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3463, pruned_loss=0.09482, over 5691039.27 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3383, pruned_loss=0.08654, over 2819959.44 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3459, pruned_loss=0.095, over 5682035.14 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:58:29,630 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:44,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2623, 1.8697, 1.5634, 1.5992], device='cuda:0'), covar=tensor([0.0881, 0.0310, 0.0321, 0.1092], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 18:59:10,092 INFO [train.py:968] (0/2) Epoch 29, batch 1400, giga_loss[loss=0.2974, simple_loss=0.3763, pruned_loss=0.1092, over 29036.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3485, pruned_loss=0.09562, over 5691393.76 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3395, pruned_loss=0.08717, over 2896594.37 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3478, pruned_loss=0.09568, over 5679253.03 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:59:19,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.611e+02 1.378e+03 1.656e+03 2.020e+03 4.710e+03, threshold=3.311e+03, percent-clipped=3.0
+2023-03-14 18:59:26,555 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1831, 1.2612, 3.6289, 3.0680], device='cuda:0'), covar=tensor([0.1798, 0.2993, 0.0477, 0.1374], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0673, 0.1006, 0.0984], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 18:59:26,607 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4423, 1.4851, 1.5405, 1.4088], device='cuda:0'), covar=tensor([0.3039, 0.3240, 0.2535, 0.2993], device='cuda:0'), in_proj_covar=tensor([0.2071, 0.2031, 0.1944, 0.2088], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 18:59:51,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-14 18:59:52,003 INFO [train.py:968] (0/2) Epoch 29, batch 1450, giga_loss[loss=0.2525, simple_loss=0.3452, pruned_loss=0.07987, over 28987.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3491, pruned_loss=0.09504, over 5696614.80 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3399, pruned_loss=0.08746, over 2956346.50 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3485, pruned_loss=0.09511, over 5683341.00 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:59:55,294 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-14 19:00:27,147 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:00:29,686 INFO [train.py:968] (0/2) Epoch 29, batch 1500, giga_loss[loss=0.2644, simple_loss=0.3494, pruned_loss=0.08971, over 29063.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3477, pruned_loss=0.09318, over 5702366.79 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3401, pruned_loss=0.08746, over 3011191.38 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3473, pruned_loss=0.09336, over 5691579.50 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:00:37,886 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.932e+02 1.290e+03 1.583e+03 2.167e+03 6.047e+03, threshold=3.166e+03, percent-clipped=7.0
+2023-03-14 19:00:47,569 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:01:02,999 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:01:06,507 INFO [train.py:968] (0/2) Epoch 29, batch 1550, giga_loss[loss=0.25, simple_loss=0.3408, pruned_loss=0.07962, over 29019.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3461, pruned_loss=0.09158, over 5705359.73 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3409, pruned_loss=0.08778, over 3096724.66 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3456, pruned_loss=0.09167, over 5691472.87 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:01:21,001 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5282, 1.5573, 1.4309, 1.6663], device='cuda:0'), covar=tensor([0.0603, 0.0327, 0.0292, 0.0669], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 19:01:45,116 INFO [train.py:968] (0/2) Epoch 29, batch 1600, giga_loss[loss=0.2726, simple_loss=0.3441, pruned_loss=0.1006, over 28938.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3469, pruned_loss=0.09324, over 5706658.27 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3408, pruned_loss=0.08775, over 3207038.12 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3467, pruned_loss=0.0935, over 5699417.08 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:01:54,957 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.679e+02 1.269e+03 1.571e+03 1.904e+03 4.185e+03, threshold=3.142e+03, percent-clipped=4.0
+2023-03-14 19:02:10,176 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 19:02:23,211 INFO [train.py:968] (0/2) Epoch 29, batch 1650, giga_loss[loss=0.2846, simple_loss=0.3529, pruned_loss=0.1082, over 28230.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3485, pruned_loss=0.09637, over 5711694.64 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3403, pruned_loss=0.08738, over 3284341.48 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3488, pruned_loss=0.09696, over 5703052.07 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:03:04,358 INFO [train.py:968] (0/2) Epoch 29, batch 1700, giga_loss[loss=0.3053, simple_loss=0.3705, pruned_loss=0.12, over 28700.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3494, pruned_loss=0.0991, over 5705925.06 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3403, pruned_loss=0.08787, over 3374703.56 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3499, pruned_loss=0.09967, over 5692539.42 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:03:13,741 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.435e+03 1.721e+03 2.184e+03 5.745e+03, threshold=3.443e+03, percent-clipped=7.0
+2023-03-14 19:03:40,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3908, 3.0393, 1.5130, 1.5346], device='cuda:0'), covar=tensor([0.1034, 0.0347, 0.0902, 0.1395], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0571, 0.0412, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 19:03:44,465 INFO [train.py:968] (0/2) Epoch 29, batch 1750, giga_loss[loss=0.2568, simple_loss=0.3338, pruned_loss=0.08985, over 29013.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3477, pruned_loss=0.09866, over 5703303.17 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3404, pruned_loss=0.0877, over 3435753.30 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3483, pruned_loss=0.09948, over 5689108.69 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:03:56,576 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 19:04:26,752 INFO [train.py:968] (0/2) Epoch 29, batch 1800, giga_loss[loss=0.3007, simple_loss=0.3716, pruned_loss=0.1149, over 28624.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3461, pruned_loss=0.09827, over 5715054.60 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.08797, over 3507720.22 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3465, pruned_loss=0.09908, over 5699335.53 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:04:36,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.889e+02 1.348e+03 1.751e+03 2.712e+03 7.556e+03, threshold=3.501e+03, percent-clipped=12.0
+2023-03-14 19:05:04,918 INFO [train.py:968] (0/2) Epoch 29, batch 1850, giga_loss[loss=0.3038, simple_loss=0.3717, pruned_loss=0.1179, over 28681.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3449, pruned_loss=0.09718, over 5718089.66 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3403, pruned_loss=0.08751, over 3566578.70 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3456, pruned_loss=0.09833, over 5701549.91 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:05:21,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1277692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:05:44,086 INFO [train.py:968] (0/2) Epoch 29, batch 1900, giga_loss[loss=0.2783, simple_loss=0.3502, pruned_loss=0.1032, over 27665.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3439, pruned_loss=0.09592, over 5719458.44 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3397, pruned_loss=0.0871, over 3634533.27 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.345, pruned_loss=0.09733, over 5702471.11 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:05:44,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1277721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:05:53,224 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:05:55,807 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.693e+02 1.192e+03 1.470e+03 1.940e+03 3.773e+03, threshold=2.939e+03, percent-clipped=1.0
+2023-03-14 19:06:01,435 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1277740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:06:31,611 INFO [train.py:968] (0/2) Epoch 29, batch 1950, giga_loss[loss=0.2359, simple_loss=0.3157, pruned_loss=0.07806, over 28753.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3418, pruned_loss=0.09444, over 5697500.67 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3401, pruned_loss=0.08714, over 3679673.55 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3425, pruned_loss=0.09573, over 5689391.78 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:07:03,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277808.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:07:08,238 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:14,504 INFO [train.py:968] (0/2) Epoch 29, batch 2000, giga_loss[loss=0.2135, simple_loss=0.2744, pruned_loss=0.0763, over 23579.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3359, pruned_loss=0.09155, over 5687716.30 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3396, pruned_loss=0.08685, over 3701778.92 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3367, pruned_loss=0.09277, over 5679708.53 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:07:26,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.888e+02 1.140e+03 1.409e+03 2.068e+03 5.365e+03, threshold=2.818e+03, percent-clipped=10.0
+2023-03-14 19:07:27,081 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-14 19:07:27,440 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:29,494 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:51,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:53,599 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:53,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:56,794 INFO [train.py:968] (0/2) Epoch 29, batch 2050, giga_loss[loss=0.2216, simple_loss=0.3034, pruned_loss=0.0699, over 28710.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.331, pruned_loss=0.08921, over 5685366.42 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3403, pruned_loss=0.08718, over 3753321.14 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3311, pruned_loss=0.09006, over 5676602.22 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:08:06,711 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:08:09,008 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:08:17,844 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:08:34,103 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:08:38,669 INFO [train.py:968] (0/2) Epoch 29, batch 2100, giga_loss[loss=0.2593, simple_loss=0.3473, pruned_loss=0.08562, over 28874.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3277, pruned_loss=0.0869, over 5685071.29 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3399, pruned_loss=0.0867, over 3857111.12 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3274, pruned_loss=0.08792, over 5677813.41 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:08:49,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.667e+02 1.073e+03 1.365e+03 1.880e+03 3.829e+03, threshold=2.729e+03, percent-clipped=7.0
+2023-03-14 19:09:19,251 INFO [train.py:968] (0/2) Epoch 29, batch 2150, giga_loss[loss=0.2327, simple_loss=0.3024, pruned_loss=0.08156, over 23915.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3279, pruned_loss=0.08649, over 5691260.34 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3387, pruned_loss=0.08599, over 3926314.43 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.328, pruned_loss=0.08775, over 5679946.37 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:09:21,925 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5182, 1.6298, 1.7144, 1.3274], device='cuda:0'), covar=tensor([0.1980, 0.2701, 0.1627, 0.1860], device='cuda:0'), in_proj_covar=tensor([0.0943, 0.0722, 0.0993, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 19:09:25,452 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:09:25,565 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3171, 2.8620, 2.6409, 2.0531], device='cuda:0'), covar=tensor([0.2906, 0.1913, 0.2120, 0.2598], device='cuda:0'), in_proj_covar=tensor([0.2075, 0.2037, 0.1949, 0.2097], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 19:09:42,014 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1278000.pt
+2023-03-14 19:09:59,729 INFO [train.py:968] (0/2) Epoch 29, batch 2200, libri_loss[loss=0.2811, simple_loss=0.3583, pruned_loss=0.1019, over 19688.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3291, pruned_loss=0.08657, over 5691833.47 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3394, pruned_loss=0.08617, over 3997875.61 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3284, pruned_loss=0.0875, over 5689190.78 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:10:07,099 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9374, 2.0841, 1.4883, 1.6035], device='cuda:0'), covar=tensor([0.1142, 0.0790, 0.1143, 0.1340], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0449, 0.0525, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 19:10:10,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.070e+02 1.223e+03 1.416e+03 1.747e+03 3.920e+03, threshold=2.831e+03, percent-clipped=4.0
+2023-03-14 19:10:42,018 INFO [train.py:968] (0/2) Epoch 29, batch 2250, giga_loss[loss=0.2386, simple_loss=0.3255, pruned_loss=0.07584, over 28959.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3288, pruned_loss=0.08687, over 5684946.43 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3396, pruned_loss=0.08626, over 4017162.35 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.328, pruned_loss=0.08755, over 5687895.33 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:10:51,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6273, 2.2800, 1.6487, 0.9325], device='cuda:0'), covar=tensor([0.7275, 0.3450, 0.5195, 0.7232], device='cuda:0'), in_proj_covar=tensor([0.1835, 0.1719, 0.1647, 0.1490], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 19:11:09,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1278106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:11:21,188 INFO [train.py:968] (0/2) Epoch 29, batch 2300, giga_loss[loss=0.2704, simple_loss=0.3368, pruned_loss=0.102, over 28918.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3265, pruned_loss=0.08605, over 5699051.49 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3402, pruned_loss=0.08654, over 4054478.10 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3252, pruned_loss=0.08642, over 5697797.39 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:11:33,614 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-14 19:11:34,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.146e+02 1.163e+03 1.346e+03 2.011e+03 5.674e+03, threshold=2.691e+03, percent-clipped=8.0
+2023-03-14 19:12:04,960 INFO [train.py:968] (0/2) Epoch 29, batch 2350, giga_loss[loss=0.2262, simple_loss=0.31, pruned_loss=0.07126, over 29042.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3242, pruned_loss=0.08505, over 5705706.86 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3403, pruned_loss=0.08652, over 4062879.71 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3231, pruned_loss=0.08535, over 5704562.79 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:12:15,915 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1278183.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:12:20,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1278189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:12:45,744 INFO [train.py:968] (0/2) Epoch 29, batch 2400, giga_loss[loss=0.2319, simple_loss=0.3104, pruned_loss=0.07666, over 28567.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.323, pruned_loss=0.08489, over 5702348.17 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3403, pruned_loss=0.0864, over 4095399.04 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3218, pruned_loss=0.08516, over 5709597.67 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:12:55,994 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.830e+02 1.099e+03 1.348e+03 1.824e+03 3.686e+03, threshold=2.696e+03, percent-clipped=4.0
+2023-03-14 19:13:07,239 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1278249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:13:09,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1278252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:13:24,375 INFO [train.py:968] (0/2) Epoch 29, batch 2450, giga_loss[loss=0.2713, simple_loss=0.3291, pruned_loss=0.1067, over 28663.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3203, pruned_loss=0.08348, over 5711456.62 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3402, pruned_loss=0.08614, over 4136856.34 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3189, pruned_loss=0.08381, over 5715756.56 frames. ], batch size: 85, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:13:31,326 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1278281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:13:37,948 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-14 19:13:50,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5386, 1.8211, 1.4853, 1.3407], device='cuda:0'), covar=tensor([0.2841, 0.2966, 0.3389, 0.2686], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1163, 0.1428, 0.1020], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 19:13:58,868 INFO [train.py:968] (0/2) Epoch 29, batch 2500, giga_loss[loss=0.2156, simple_loss=0.2979, pruned_loss=0.06662, over 28951.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3195, pruned_loss=0.08279, over 5723651.00 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3408, pruned_loss=0.08618, over 4220532.92 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.317, pruned_loss=0.08289, over 5721302.61 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:14:02,958 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1278326.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:14:05,187 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1278329.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:14:08,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1278332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:14:10,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.424e+02 1.125e+03 1.361e+03 1.896e+03 5.086e+03, threshold=2.722e+03, percent-clipped=9.0
+2023-03-14 19:14:11,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1278335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:14:12,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1865, 1.3635, 1.3017, 1.1087], device='cuda:0'), covar=tensor([0.3722, 0.3180, 0.1958, 0.3040], device='cuda:0'), in_proj_covar=tensor([0.2065, 0.2027, 0.1938, 0.2091], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 19:14:25,651 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1278352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:14:27,001 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 19:14:29,465 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3040, 1.1215, 3.6628, 3.2205], device='cuda:0'), covar=tensor([0.1654, 0.2893, 0.0454, 0.0938], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0669, 0.1001, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 19:14:29,990 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1278358.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:14:35,419 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1278364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:14:40,434 INFO [train.py:968] (0/2) Epoch 29, batch 2550, giga_loss[loss=0.2044, simple_loss=0.2822, pruned_loss=0.06335, over 28559.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3174, pruned_loss=0.0821, over 5714553.60 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3411, pruned_loss=0.08637, over 4228987.25 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3151, pruned_loss=0.08203, over 5711806.30 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:14:45,213 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-14 19:15:20,351 INFO [train.py:968] (0/2) Epoch 29, batch 2600, giga_loss[loss=0.2104, simple_loss=0.2899, pruned_loss=0.06552, over 28957.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.316, pruned_loss=0.08122, over 5722363.88 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3416, pruned_loss=0.08639, over 4260996.29 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.3133, pruned_loss=0.08106, over 5717838.42 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:15:20,594 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6410, 4.8518, 1.7051, 1.9738], device='cuda:0'), covar=tensor([0.0957, 0.0213, 0.0902, 0.1232], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0568, 0.0412, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 19:15:32,399 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.704e+02 1.095e+03 1.261e+03 1.492e+03 2.794e+03, threshold=2.521e+03, percent-clipped=1.0
+2023-03-14 19:15:43,758 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9197, 2.1450, 2.2711, 1.7687], device='cuda:0'), covar=tensor([0.3546, 0.2883, 0.2673, 0.3654], device='cuda:0'), in_proj_covar=tensor([0.2061, 0.2022, 0.1933, 0.2088], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 19:15:58,178 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-14 19:15:59,717 INFO [train.py:968] (0/2) Epoch 29, batch 2650, giga_loss[loss=0.2091, simple_loss=0.2919, pruned_loss=0.06309, over 28549.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3154, pruned_loss=0.0808, over 5721566.26 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3423, pruned_loss=0.08661, over 4299233.34 frames. ], giga_tot_loss[loss=0.2365, simple_loss=0.3122, pruned_loss=0.0804, over 5715958.46 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:16:08,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7720, 2.1271, 2.0480, 1.7319], device='cuda:0'), covar=tensor([0.2783, 0.1959, 0.1761, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.2060, 0.2020, 0.1932, 0.2087], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 19:16:19,229 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1278495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:16:21,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1278498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:16:38,726 INFO [train.py:968] (0/2) Epoch 29, batch 2700, giga_loss[loss=0.246, simple_loss=0.3255, pruned_loss=0.08328, over 28956.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3158, pruned_loss=0.08108, over 5724157.65 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3415, pruned_loss=0.08598, over 4360818.63 frames. ], giga_tot_loss[loss=0.2374, simple_loss=0.3128, pruned_loss=0.08098, over 5714385.15 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:16:43,815 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1278527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:16:50,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.307e+02 1.180e+03 1.456e+03 1.941e+03 4.336e+03, threshold=2.913e+03, percent-clipped=8.0
+2023-03-14 19:17:17,499 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4636, 4.2195, 1.6959, 1.5773], device='cuda:0'), covar=tensor([0.1048, 0.0297, 0.0928, 0.1414], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0569, 0.0412, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 19:17:21,710 INFO [train.py:968] (0/2) Epoch 29, batch 2750, giga_loss[loss=0.2487, simple_loss=0.3155, pruned_loss=0.09093, over 28637.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3206, pruned_loss=0.08415, over 5715516.62 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3426, pruned_loss=0.08664, over 4395896.87 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.3168, pruned_loss=0.08351, over 5713329.38 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:18:04,357 INFO [train.py:968] (0/2) Epoch 29, batch 2800, libri_loss[loss=0.2425, simple_loss=0.3189, pruned_loss=0.08303, over 29355.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3272, pruned_loss=0.08869, over 5709807.04 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3432, pruned_loss=0.08697, over 4454990.05 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3231, pruned_loss=0.08796, over 5700987.38 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:18:17,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.239e+02 1.416e+03 1.756e+03 2.267e+03 3.536e+03, threshold=3.512e+03, percent-clipped=12.0
+2023-03-14 19:18:50,487 INFO [train.py:968] (0/2) Epoch 29, batch 2850, giga_loss[loss=0.3579, simple_loss=0.4082, pruned_loss=0.1538, over 27621.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.336, pruned_loss=0.0948, over 5702207.69 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3427, pruned_loss=0.08686, over 4483063.57 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3328, pruned_loss=0.0944, over 5692262.07 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:19:14,327 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1278699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:19:24,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0380, 1.1079, 3.4358, 2.9660], device='cuda:0'), covar=tensor([0.1697, 0.2775, 0.0525, 0.1126], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0671, 0.1001, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 19:19:30,815 INFO [train.py:968] (0/2) Epoch 29, batch 2900, giga_loss[loss=0.2389, simple_loss=0.3287, pruned_loss=0.07449, over 28908.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3404, pruned_loss=0.09652, over 5695630.77 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3423, pruned_loss=0.08654, over 4554250.47 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.338, pruned_loss=0.09691, over 5681919.26 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:19:41,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.371e+02 1.437e+03 1.706e+03 2.468e+03 5.731e+03, threshold=3.413e+03, percent-clipped=8.0
+2023-03-14 19:20:17,012 INFO [train.py:968] (0/2) Epoch 29, batch 2950, giga_loss[loss=0.2801, simple_loss=0.3622, pruned_loss=0.09901, over 28668.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3444, pruned_loss=0.09817, over 5668884.84 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3411, pruned_loss=0.08588, over 4582419.20 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3433, pruned_loss=0.09926, over 5662867.43 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:20:39,197 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-14 19:20:57,945 INFO [train.py:968] (0/2) Epoch 29, batch 3000, giga_loss[loss=0.3651, simple_loss=0.4187, pruned_loss=0.1558, over 28205.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3489, pruned_loss=0.09948, over 5690640.06 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3412, pruned_loss=0.08581, over 4626875.59 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3481, pruned_loss=0.1008, over 5679744.89 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 19:20:57,952 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 19:21:07,154 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.2057, simple_loss=0.3132, pruned_loss=0.04906, over 944034.00 frames. 
+2023-03-14 19:21:07,155 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 19:21:09,490 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1278824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:21:20,078 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.737e+02 1.428e+03 1.693e+03 2.112e+03 4.652e+03, threshold=3.386e+03, percent-clipped=4.0
+2023-03-14 19:21:41,028 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9649, 1.1952, 2.7749, 2.6736], device='cuda:0'), covar=tensor([0.1606, 0.2673, 0.0566, 0.1551], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0669, 0.0999, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 19:21:47,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9427, 2.0704, 2.2191, 1.7083], device='cuda:0'), covar=tensor([0.1977, 0.2506, 0.1570, 0.1943], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0721, 0.0991, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 19:21:47,351 INFO [train.py:968] (0/2) Epoch 29, batch 3050, giga_loss[loss=0.2455, simple_loss=0.3296, pruned_loss=0.08073, over 28711.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3519, pruned_loss=0.1015, over 5677319.99 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3407, pruned_loss=0.08573, over 4654227.63 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3519, pruned_loss=0.103, over 5672040.64 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 19:22:15,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1278904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:22:17,149 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1278905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:22:29,986 INFO [train.py:968] (0/2) Epoch 29, batch 3100, giga_loss[loss=0.2415, simple_loss=0.3177, pruned_loss=0.08267, over 27620.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3475, pruned_loss=0.09805, over 5681069.80 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3409, pruned_loss=0.08591, over 4669003.47 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3474, pruned_loss=0.09929, over 5677705.85 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 19:22:42,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.925e+02 1.348e+03 1.656e+03 2.269e+03 4.569e+03, threshold=3.311e+03, percent-clipped=8.0
+2023-03-14 19:23:11,409 INFO [train.py:968] (0/2) Epoch 29, batch 3150, giga_loss[loss=0.2759, simple_loss=0.3554, pruned_loss=0.09822, over 28679.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3451, pruned_loss=0.0961, over 5682862.00 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3408, pruned_loss=0.08591, over 4693626.58 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3452, pruned_loss=0.09729, over 5676227.54 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 19:23:17,955 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1278977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:23:55,488 INFO [train.py:968] (0/2) Epoch 29, batch 3200, giga_loss[loss=0.2893, simple_loss=0.3619, pruned_loss=0.1083, over 28827.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3444, pruned_loss=0.09538, over 5674547.79 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3408, pruned_loss=0.08588, over 4709099.23 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3446, pruned_loss=0.09653, over 5673226.56 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:24:13,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.755e+02 1.326e+03 1.618e+03 2.029e+03 5.625e+03, threshold=3.237e+03, percent-clipped=8.0
+2023-03-14 19:24:39,791 INFO [train.py:968] (0/2) Epoch 29, batch 3250, giga_loss[loss=0.3476, simple_loss=0.3982, pruned_loss=0.1485, over 26688.00 frames. ], tot_loss[loss=0.269, simple_loss=0.346, pruned_loss=0.09603, over 5673536.68 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3412, pruned_loss=0.08612, over 4737062.60 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3459, pruned_loss=0.09701, over 5668813.16 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:24:41,932 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:25:21,883 INFO [train.py:968] (0/2) Epoch 29, batch 3300, giga_loss[loss=0.2485, simple_loss=0.333, pruned_loss=0.08198, over 29118.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3487, pruned_loss=0.09795, over 5684736.13 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3415, pruned_loss=0.08626, over 4748503.88 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3485, pruned_loss=0.09873, over 5679077.23 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:25:35,012 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.690e+02 1.348e+03 1.779e+03 2.532e+03 6.188e+03, threshold=3.558e+03, percent-clipped=15.0
+2023-03-14 19:26:05,078 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-14 19:26:05,282 INFO [train.py:968] (0/2) Epoch 29, batch 3350, giga_loss[loss=0.2704, simple_loss=0.3439, pruned_loss=0.09843, over 28651.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3509, pruned_loss=0.1, over 5676212.27 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3419, pruned_loss=0.08655, over 4762701.34 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3506, pruned_loss=0.1006, over 5676161.35 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:26:27,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:26:42,253 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:26:43,957 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:26:44,324 INFO [train.py:968] (0/2) Epoch 29, batch 3400, libri_loss[loss=0.2593, simple_loss=0.3404, pruned_loss=0.08906, over 29566.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3523, pruned_loss=0.1011, over 5679065.50 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3423, pruned_loss=0.08671, over 4781564.41 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.352, pruned_loss=0.1018, over 5683122.53 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:26:57,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.997e+02 1.448e+03 1.700e+03 2.081e+03 5.699e+03, threshold=3.399e+03, percent-clipped=5.0
+2023-03-14 19:27:08,578 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:27:28,019 INFO [train.py:968] (0/2) Epoch 29, batch 3450, libri_loss[loss=0.2504, simple_loss=0.3438, pruned_loss=0.07855, over 29562.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3531, pruned_loss=0.1025, over 5681185.60 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3421, pruned_loss=0.08664, over 4809167.23 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3532, pruned_loss=0.1035, over 5679486.81 frames. ], batch size: 83, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:27:28,880 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1279272.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:27:34,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:27:35,220 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:28:07,940 INFO [train.py:968] (0/2) Epoch 29, batch 3500, giga_loss[loss=0.2404, simple_loss=0.3295, pruned_loss=0.07565, over 28672.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3523, pruned_loss=0.1016, over 5681532.35 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3417, pruned_loss=0.08637, over 4820294.95 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3528, pruned_loss=0.1027, over 5678041.75 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:28:12,307 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5303, 1.4273, 4.6817, 3.5072], device='cuda:0'), covar=tensor([0.1715, 0.2887, 0.0421, 0.1022], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0671, 0.1001, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 19:28:22,412 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.711e+02 1.345e+03 1.658e+03 2.095e+03 8.057e+03, threshold=3.316e+03, percent-clipped=3.0
+2023-03-14 19:28:26,675 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:28:28,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:28:34,800 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:28:48,678 INFO [train.py:968] (0/2) Epoch 29, batch 3550, giga_loss[loss=0.2543, simple_loss=0.3438, pruned_loss=0.08243, over 28946.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.352, pruned_loss=0.1002, over 5688425.04 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3418, pruned_loss=0.0864, over 4824759.47 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3523, pruned_loss=0.1011, over 5685571.33 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:28:50,926 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:12,449 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1279399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:30,839 INFO [train.py:968] (0/2) Epoch 29, batch 3600, giga_loss[loss=0.2726, simple_loss=0.354, pruned_loss=0.09563, over 28861.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3514, pruned_loss=0.09918, over 5689984.61 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3413, pruned_loss=0.08618, over 4842513.36 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3522, pruned_loss=0.1003, over 5687412.61 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:29:31,932 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:32,621 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:35,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:35,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:42,837 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.955e+02 1.270e+03 1.482e+03 1.915e+03 4.100e+03, threshold=2.964e+03, percent-clipped=5.0
+2023-03-14 19:29:49,816 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4339, 3.1624, 1.5235, 1.5333], device='cuda:0'), covar=tensor([0.0999, 0.0319, 0.0943, 0.1391], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0566, 0.0410, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 19:29:55,439 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1279453.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:29:55,997 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:56,690 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:57,461 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-14 19:30:00,116 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5380, 4.3797, 4.1183, 2.2075], device='cuda:0'), covar=tensor([0.0501, 0.0619, 0.0691, 0.2126], device='cuda:0'), in_proj_covar=tensor([0.1287, 0.1192, 0.1003, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 19:30:01,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1531, 1.5344, 1.5452, 1.3161], device='cuda:0'), covar=tensor([0.2366, 0.1772, 0.2519, 0.2095], device='cuda:0'), in_proj_covar=tensor([0.0512, 0.0762, 0.0735, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 19:30:08,497 INFO [train.py:968] (0/2) Epoch 29, batch 3650, giga_loss[loss=0.3117, simple_loss=0.3637, pruned_loss=0.1298, over 26746.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3503, pruned_loss=0.09844, over 5686426.14 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3417, pruned_loss=0.08637, over 4855187.36 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3508, pruned_loss=0.09945, over 5692101.83 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:30:27,127 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:30:29,204 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:30:46,027 INFO [train.py:968] (0/2) Epoch 29, batch 3700, giga_loss[loss=0.2454, simple_loss=0.3274, pruned_loss=0.08164, over 28603.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3478, pruned_loss=0.09734, over 5683288.62 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3419, pruned_loss=0.0864, over 4877678.69 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3482, pruned_loss=0.09839, over 5686037.28 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:30:51,386 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:30:52,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1279529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:30:55,821 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-14 19:30:57,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6378, 1.7132, 1.8054, 1.4155], device='cuda:0'), covar=tensor([0.1748, 0.2715, 0.1494, 0.1764], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0721, 0.0990, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 19:30:59,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.053e+02 1.270e+03 1.622e+03 2.048e+03 4.889e+03, threshold=3.245e+03, percent-clipped=6.0
+2023-03-14 19:31:23,502 INFO [train.py:968] (0/2) Epoch 29, batch 3750, giga_loss[loss=0.2519, simple_loss=0.3258, pruned_loss=0.089, over 28506.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3453, pruned_loss=0.09602, over 5696564.48 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3417, pruned_loss=0.0863, over 4898336.26 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3459, pruned_loss=0.09714, over 5694588.62 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:31:43,534 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1279600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:32:03,281 INFO [train.py:968] (0/2) Epoch 29, batch 3800, giga_loss[loss=0.2962, simple_loss=0.3476, pruned_loss=0.1224, over 23739.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3443, pruned_loss=0.09563, over 5696250.71 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3413, pruned_loss=0.08609, over 4931456.26 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3452, pruned_loss=0.09697, over 5689897.58 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:32:16,095 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.184e+02 1.211e+03 1.469e+03 2.022e+03 4.851e+03, threshold=2.937e+03, percent-clipped=6.0
+2023-03-14 19:32:22,534 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279647.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:32:42,254 INFO [train.py:968] (0/2) Epoch 29, batch 3850, giga_loss[loss=0.2738, simple_loss=0.3465, pruned_loss=0.1005, over 28497.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3446, pruned_loss=0.09607, over 5704130.87 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3402, pruned_loss=0.08556, over 4965346.22 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3464, pruned_loss=0.09792, over 5692135.45 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:33:21,443 INFO [train.py:968] (0/2) Epoch 29, batch 3900, giga_loss[loss=0.2859, simple_loss=0.3645, pruned_loss=0.1037, over 28724.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.345, pruned_loss=0.09534, over 5710952.65 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3408, pruned_loss=0.08591, over 4992397.15 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.346, pruned_loss=0.09681, over 5696932.21 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:33:34,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.166e+02 1.166e+03 1.422e+03 1.742e+03 6.112e+03, threshold=2.844e+03, percent-clipped=2.0
+2023-03-14 19:33:59,294 INFO [train.py:968] (0/2) Epoch 29, batch 3950, giga_loss[loss=0.2868, simple_loss=0.3609, pruned_loss=0.1064, over 28968.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3446, pruned_loss=0.09455, over 5716917.05 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3406, pruned_loss=0.08585, over 5015762.36 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3457, pruned_loss=0.09601, over 5703899.89 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:34:02,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:34:16,310 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279790.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:34:18,171 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279793.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:34:38,497 INFO [train.py:968] (0/2) Epoch 29, batch 4000, libri_loss[loss=0.2194, simple_loss=0.2969, pruned_loss=0.07101, over 29329.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3432, pruned_loss=0.09382, over 5719909.55 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3399, pruned_loss=0.08575, over 5059680.46 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3449, pruned_loss=0.09555, over 5699961.89 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:34:39,386 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279822.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:34:43,016 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279828.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:34:50,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.279e+02 1.165e+03 1.413e+03 1.995e+03 5.598e+03, threshold=2.826e+03, percent-clipped=10.0
+2023-03-14 19:34:59,409 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1302, 2.9668, 2.8090, 1.5421], device='cuda:0'), covar=tensor([0.1078, 0.1181, 0.1032, 0.2582], device='cuda:0'), in_proj_covar=tensor([0.1288, 0.1192, 0.1005, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 19:35:15,758 INFO [train.py:968] (0/2) Epoch 29, batch 4050, giga_loss[loss=0.2667, simple_loss=0.3375, pruned_loss=0.09796, over 28538.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3432, pruned_loss=0.09426, over 5720330.35 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3401, pruned_loss=0.08607, over 5080169.28 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3445, pruned_loss=0.09558, over 5700836.39 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:35:26,206 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-14 19:35:39,247 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:35:49,810 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:35:51,697 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:35:53,253 INFO [train.py:968] (0/2) Epoch 29, batch 4100, giga_loss[loss=0.2553, simple_loss=0.3291, pruned_loss=0.09075, over 28952.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3407, pruned_loss=0.09306, over 5719446.94 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3401, pruned_loss=0.08597, over 5093077.55 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3418, pruned_loss=0.09432, over 5703665.43 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:36:04,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.809e+02 1.203e+03 1.503e+03 2.112e+03 5.860e+03, threshold=3.005e+03, percent-clipped=13.0
+2023-03-14 19:36:12,289 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:36:27,490 INFO [train.py:968] (0/2) Epoch 29, batch 4150, libri_loss[loss=0.2468, simple_loss=0.3337, pruned_loss=0.07997, over 29756.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3373, pruned_loss=0.09098, over 5727321.30 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3398, pruned_loss=0.08573, over 5116983.73 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3383, pruned_loss=0.09235, over 5709864.26 frames. ], batch size: 87, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:36:27,836 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279971.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:36:30,282 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279974.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:36:30,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 19:36:30,871 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:36:44,696 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7154, 1.9155, 1.8355, 1.7598], device='cuda:0'), covar=tensor([0.2388, 0.2427, 0.2753, 0.2370], device='cuda:0'), in_proj_covar=tensor([0.0512, 0.0762, 0.0735, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 19:36:49,871 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1280000.pt
+2023-03-14 19:36:52,358 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280003.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:37:06,569 INFO [train.py:968] (0/2) Epoch 29, batch 4200, giga_loss[loss=0.2311, simple_loss=0.3176, pruned_loss=0.0723, over 29088.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3366, pruned_loss=0.09103, over 5721525.88 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3396, pruned_loss=0.08569, over 5136194.01 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3375, pruned_loss=0.09227, over 5703872.35 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:37:19,627 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.661e+02 1.316e+03 1.644e+03 2.288e+03 8.873e+03, threshold=3.287e+03, percent-clipped=11.0
+2023-03-14 19:37:25,562 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1280047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:37:27,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1280050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:37:42,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7447, 1.3370, 5.2774, 3.6732], device='cuda:0'), covar=tensor([0.1609, 0.2868, 0.0385, 0.1033], device='cuda:0'), in_proj_covar=tensor([0.0802, 0.0669, 0.0997, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 19:37:43,133 INFO [train.py:968] (0/2) Epoch 29, batch 4250, giga_loss[loss=0.2616, simple_loss=0.338, pruned_loss=0.09255, over 28968.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3369, pruned_loss=0.09187, over 5720205.03 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3396, pruned_loss=0.08584, over 5150255.03 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3376, pruned_loss=0.09284, over 5704014.91 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:37:48,543 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:38:13,493 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-14 19:38:14,163 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-14 19:38:18,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1280118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:38:20,611 INFO [train.py:968] (0/2) Epoch 29, batch 4300, libri_loss[loss=0.2352, simple_loss=0.3205, pruned_loss=0.07493, over 29565.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3353, pruned_loss=0.0916, over 5722061.68 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3398, pruned_loss=0.08597, over 5161270.24 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3356, pruned_loss=0.09234, over 5706863.91 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:38:22,338 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1280121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:38:36,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.311e+02 1.233e+03 1.555e+03 2.028e+03 6.185e+03, threshold=3.110e+03, percent-clipped=8.0
+2023-03-14 19:38:44,463 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:39:00,112 INFO [train.py:968] (0/2) Epoch 29, batch 4350, libri_loss[loss=0.2539, simple_loss=0.3386, pruned_loss=0.08458, over 29527.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3338, pruned_loss=0.0908, over 5728205.94 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3401, pruned_loss=0.08612, over 5193931.03 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3335, pruned_loss=0.09151, over 5708263.31 frames. ], batch size: 81, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:39:15,522 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1280193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:39:36,557 INFO [train.py:968] (0/2) Epoch 29, batch 4400, giga_loss[loss=0.2192, simple_loss=0.3024, pruned_loss=0.06796, over 28972.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3313, pruned_loss=0.08973, over 5726027.81 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3398, pruned_loss=0.08595, over 5211075.73 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3311, pruned_loss=0.09056, over 5708717.10 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:39:42,231 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1280226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:39:51,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.934e+02 1.242e+03 1.528e+03 1.909e+03 4.179e+03, threshold=3.056e+03, percent-clipped=6.0
+2023-03-14 19:40:16,266 INFO [train.py:968] (0/2) Epoch 29, batch 4450, libri_loss[loss=0.2326, simple_loss=0.3106, pruned_loss=0.07727, over 29630.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3295, pruned_loss=0.0887, over 5723638.91 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3392, pruned_loss=0.08575, over 5234998.25 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3295, pruned_loss=0.08965, over 5703920.44 frames. ], batch size: 69, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:40:55,559 INFO [train.py:968] (0/2) Epoch 29, batch 4500, libri_loss[loss=0.2711, simple_loss=0.3537, pruned_loss=0.09423, over 29372.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3322, pruned_loss=0.09011, over 5708715.72 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3391, pruned_loss=0.08572, over 5247835.30 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3321, pruned_loss=0.09105, over 5699050.57 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:41:11,061 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.652e+02 1.209e+03 1.414e+03 1.819e+03 4.419e+03, threshold=2.828e+03, percent-clipped=4.0
+2023-03-14 19:41:35,747 INFO [train.py:968] (0/2) Epoch 29, batch 4550, giga_loss[loss=0.2804, simple_loss=0.358, pruned_loss=0.1014, over 28363.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3347, pruned_loss=0.09093, over 5715966.72 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3396, pruned_loss=0.08605, over 5260150.50 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3341, pruned_loss=0.0915, over 5707604.10 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:42:01,422 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1280402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:42:16,955 INFO [train.py:968] (0/2) Epoch 29, batch 4600, giga_loss[loss=0.2383, simple_loss=0.3244, pruned_loss=0.07613, over 29011.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3372, pruned_loss=0.09167, over 5722322.53 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3396, pruned_loss=0.08615, over 5272700.57 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3366, pruned_loss=0.09213, over 5712620.38 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:42:23,498 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9588, 3.7623, 3.6023, 1.6721], device='cuda:0'), covar=tensor([0.0705, 0.0868, 0.0798, 0.2087], device='cuda:0'), in_proj_covar=tensor([0.1288, 0.1190, 0.1002, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 19:42:31,752 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.998e+02 1.359e+03 1.728e+03 2.198e+03 5.932e+03, threshold=3.457e+03, percent-clipped=13.0
+2023-03-14 19:42:57,978 INFO [train.py:968] (0/2) Epoch 29, batch 4650, giga_loss[loss=0.2681, simple_loss=0.3383, pruned_loss=0.099, over 28774.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3399, pruned_loss=0.09304, over 5715344.64 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3398, pruned_loss=0.08643, over 5306160.94 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3392, pruned_loss=0.09352, over 5698611.44 frames. ], batch size: 99, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:43:26,376 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1280507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:43:37,442 INFO [train.py:968] (0/2) Epoch 29, batch 4700, giga_loss[loss=0.234, simple_loss=0.3214, pruned_loss=0.07328, over 28954.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3392, pruned_loss=0.09213, over 5705690.24 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3406, pruned_loss=0.08719, over 5318127.01 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.338, pruned_loss=0.09206, over 5693406.44 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:43:54,240 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.598e+02 1.242e+03 1.624e+03 2.400e+03 8.090e+03, threshold=3.249e+03, percent-clipped=10.0
+2023-03-14 19:44:13,181 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-14 19:44:14,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1280568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:44:16,296 INFO [train.py:968] (0/2) Epoch 29, batch 4750, giga_loss[loss=0.277, simple_loss=0.3381, pruned_loss=0.1079, over 28603.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3399, pruned_loss=0.0922, over 5704281.82 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3411, pruned_loss=0.08743, over 5328561.30 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3384, pruned_loss=0.09207, over 5695200.56 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:44:39,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1280601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:44:54,262 INFO [train.py:968] (0/2) Epoch 29, batch 4800, giga_loss[loss=0.2803, simple_loss=0.3503, pruned_loss=0.1052, over 28931.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3403, pruned_loss=0.09266, over 5714329.69 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3409, pruned_loss=0.0874, over 5344934.98 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3393, pruned_loss=0.09274, over 5702198.26 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:45:08,231 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.519e+02 1.402e+03 1.782e+03 2.405e+03 5.055e+03, threshold=3.564e+03, percent-clipped=7.0
+2023-03-14 19:45:23,838 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6122, 2.2028, 1.3462, 0.9552], device='cuda:0'), covar=tensor([0.8244, 0.3737, 0.3869, 0.7544], device='cuda:0'), in_proj_covar=tensor([0.1841, 0.1723, 0.1659, 0.1502], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 19:45:33,037 INFO [train.py:968] (0/2) Epoch 29, batch 4850, giga_loss[loss=0.2596, simple_loss=0.3366, pruned_loss=0.09133, over 28753.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3407, pruned_loss=0.09298, over 5717500.69 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3406, pruned_loss=0.08715, over 5355374.75 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3401, pruned_loss=0.09333, over 5705149.90 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:46:04,708 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1280708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:06,728 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1280711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:09,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1280714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:14,598 INFO [train.py:968] (0/2) Epoch 29, batch 4900, giga_loss[loss=0.2511, simple_loss=0.3277, pruned_loss=0.08724, over 28632.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3426, pruned_loss=0.09411, over 5710443.00 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3408, pruned_loss=0.0873, over 5354428.54 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.342, pruned_loss=0.09434, over 5706725.43 frames. ], batch size: 85, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:46:29,218 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2101, 1.4094, 1.5280, 1.3677], device='cuda:0'), covar=tensor([0.1567, 0.1144, 0.1660, 0.1265], device='cuda:0'), in_proj_covar=tensor([0.0511, 0.0759, 0.0732, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 19:46:29,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.674e+02 1.337e+03 1.553e+03 2.327e+03 4.559e+03, threshold=3.106e+03, percent-clipped=5.0
+2023-03-14 19:46:31,083 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:31,731 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1280744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:33,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1280747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:36,579 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1280750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:41,067 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1528, 1.8250, 1.4077, 0.4041], device='cuda:0'), covar=tensor([0.4763, 0.2993, 0.3851, 0.6007], device='cuda:0'), in_proj_covar=tensor([0.1841, 0.1724, 0.1662, 0.1501], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 19:46:51,828 INFO [train.py:968] (0/2) Epoch 29, batch 4950, giga_loss[loss=0.2906, simple_loss=0.3656, pruned_loss=0.1078, over 28567.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3454, pruned_loss=0.09562, over 5706877.21 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3408, pruned_loss=0.0873, over 5367158.66 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.345, pruned_loss=0.09609, over 5706220.25 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:46:55,875 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:57,500 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1280777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:57,595 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5238, 2.2614, 1.6941, 0.7960], device='cuda:0'), covar=tensor([0.6704, 0.3114, 0.4864, 0.7280], device='cuda:0'), in_proj_covar=tensor([0.1841, 0.1725, 0.1663, 0.1502], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 19:47:29,490 INFO [train.py:968] (0/2) Epoch 29, batch 5000, libri_loss[loss=0.327, simple_loss=0.3986, pruned_loss=0.1277, over 20530.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3477, pruned_loss=0.0967, over 5698940.64 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.341, pruned_loss=0.08736, over 5369082.40 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3474, pruned_loss=0.09733, over 5708758.74 frames. ], batch size: 187, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:47:44,404 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.883e+02 1.423e+03 1.640e+03 2.200e+03 5.614e+03, threshold=3.280e+03, percent-clipped=7.0
+2023-03-14 19:47:49,445 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 19:48:06,151 INFO [train.py:968] (0/2) Epoch 29, batch 5050, giga_loss[loss=0.2957, simple_loss=0.3684, pruned_loss=0.1115, over 28850.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3493, pruned_loss=0.09759, over 5703651.87 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3412, pruned_loss=0.08749, over 5381174.07 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3491, pruned_loss=0.09822, over 5707928.17 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:48:17,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1280882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:48:44,447 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1280920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:48:44,885 INFO [train.py:968] (0/2) Epoch 29, batch 5100, giga_loss[loss=0.257, simple_loss=0.3375, pruned_loss=0.08826, over 28951.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3505, pruned_loss=0.09857, over 5698052.53 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3415, pruned_loss=0.08775, over 5389688.23 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3501, pruned_loss=0.09904, over 5699384.34 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:48:46,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1280923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:49:00,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.666e+02 1.341e+03 1.684e+03 2.059e+03 4.733e+03, threshold=3.369e+03, percent-clipped=4.0
+2023-03-14 19:49:08,822 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:49:23,128 INFO [train.py:968] (0/2) Epoch 29, batch 5150, libri_loss[loss=0.2682, simple_loss=0.3576, pruned_loss=0.08936, over 29522.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3504, pruned_loss=0.09851, over 5704572.36 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3417, pruned_loss=0.08778, over 5399143.15 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3503, pruned_loss=0.09914, over 5703854.45 frames. ], batch size: 81, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:49:38,542 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0992, 1.3021, 5.4349, 4.0364], device='cuda:0'), covar=tensor([0.1466, 0.2833, 0.0366, 0.0762], device='cuda:0'), in_proj_covar=tensor([0.0801, 0.0669, 0.1001, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 19:49:57,537 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3042, 1.5615, 1.3980, 1.5216], device='cuda:0'), covar=tensor([0.0715, 0.0398, 0.0351, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 19:50:02,707 INFO [train.py:968] (0/2) Epoch 29, batch 5200, giga_loss[loss=0.2681, simple_loss=0.3451, pruned_loss=0.09557, over 28649.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3483, pruned_loss=0.09755, over 5707016.73 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3417, pruned_loss=0.08785, over 5413599.82 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3484, pruned_loss=0.09827, over 5701684.77 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:50:06,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1281025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:50:07,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3836, 1.3920, 1.5194, 1.1290], device='cuda:0'), covar=tensor([0.1994, 0.3225, 0.1670, 0.1740], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0718, 0.0987, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 19:50:08,483 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1281028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:50:19,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.158e+02 1.242e+03 1.756e+03 2.278e+03 6.519e+03, threshold=3.512e+03, percent-clipped=9.0
+2023-03-14 19:50:20,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6047, 1.9370, 1.5525, 1.8230], device='cuda:0'), covar=tensor([0.2829, 0.2900, 0.3311, 0.2483], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1162, 0.1423, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 19:50:31,064 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1281057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:50:40,818 INFO [train.py:968] (0/2) Epoch 29, batch 5250, giga_loss[loss=0.2582, simple_loss=0.3283, pruned_loss=0.09398, over 28435.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3447, pruned_loss=0.0959, over 5711451.35 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3417, pruned_loss=0.08788, over 5423133.08 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3449, pruned_loss=0.09665, over 5705143.55 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:50:51,717 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1281083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:50:51,939 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-14 19:51:20,991 INFO [train.py:968] (0/2) Epoch 29, batch 5300, giga_loss[loss=0.2891, simple_loss=0.3684, pruned_loss=0.1049, over 28782.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3428, pruned_loss=0.09465, over 5713669.05 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3417, pruned_loss=0.08785, over 5425563.41 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.343, pruned_loss=0.09529, over 5707784.25 frames. ], batch size: 285, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:51:24,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1281125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:51:37,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.114e+02 1.229e+03 1.539e+03 1.957e+03 3.622e+03, threshold=3.078e+03, percent-clipped=1.0
+2023-03-14 19:52:01,130 INFO [train.py:968] (0/2) Epoch 29, batch 5350, giga_loss[loss=0.2295, simple_loss=0.316, pruned_loss=0.07144, over 29050.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3435, pruned_loss=0.09408, over 5715827.86 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3414, pruned_loss=0.08776, over 5437246.48 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3439, pruned_loss=0.09483, over 5706663.55 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:52:39,158 INFO [train.py:968] (0/2) Epoch 29, batch 5400, giga_loss[loss=0.2399, simple_loss=0.3276, pruned_loss=0.07611, over 28922.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3439, pruned_loss=0.09318, over 5723580.27 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3412, pruned_loss=0.08778, over 5455992.24 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3445, pruned_loss=0.09397, over 5709446.83 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:52:44,132 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1281226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:52:47,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1281229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:52:57,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.903e+02 1.280e+03 1.584e+03 2.007e+03 4.658e+03, threshold=3.167e+03, percent-clipped=7.0
+2023-03-14 19:53:00,129 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-14 19:53:04,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5595, 1.7655, 1.7933, 1.3289], device='cuda:0'), covar=tensor([0.1880, 0.2806, 0.1638, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0717, 0.0986, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 19:53:08,902 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1281258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:53:15,348 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6624, 2.0116, 1.5218, 1.5124], device='cuda:0'), covar=tensor([0.1030, 0.0607, 0.0961, 0.1206], device='cuda:0'), in_proj_covar=tensor([0.0411, 0.0446, 0.0522, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 19:53:17,071 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1281268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:53:18,811 INFO [train.py:968] (0/2) Epoch 29, batch 5450, giga_loss[loss=0.2404, simple_loss=0.3117, pruned_loss=0.08459, over 28587.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3426, pruned_loss=0.09362, over 5724436.10 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3411, pruned_loss=0.08777, over 5466676.43 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3433, pruned_loss=0.0944, over 5709641.92 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:53:19,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1281271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:53:21,005 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1281272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:53:36,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7015, 1.7358, 1.8706, 1.4482], device='cuda:0'), covar=tensor([0.1587, 0.2424, 0.1386, 0.1637], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0717, 0.0985, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 19:53:44,134 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1281300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:54:02,405 INFO [train.py:968] (0/2) Epoch 29, batch 5500, giga_loss[loss=0.252, simple_loss=0.3366, pruned_loss=0.0837, over 28811.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3419, pruned_loss=0.09449, over 5727677.43 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3413, pruned_loss=0.08788, over 5470690.63 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3423, pruned_loss=0.09507, over 5714784.49 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:54:18,619 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.773e+02 1.331e+03 1.736e+03 2.430e+03 1.035e+04, threshold=3.473e+03, percent-clipped=9.0
+2023-03-14 19:54:24,018 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3585, 1.2823, 1.2514, 1.4483], device='cuda:0'), covar=tensor([0.0731, 0.0360, 0.0349, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 19:54:41,199 INFO [train.py:968] (0/2) Epoch 29, batch 5550, giga_loss[loss=0.2007, simple_loss=0.2818, pruned_loss=0.05982, over 28508.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3403, pruned_loss=0.0947, over 5720429.74 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.342, pruned_loss=0.08833, over 5465879.96 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3399, pruned_loss=0.09492, over 5720308.65 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:55:18,034 INFO [train.py:968] (0/2) Epoch 29, batch 5600, giga_loss[loss=0.2048, simple_loss=0.2802, pruned_loss=0.06463, over 28672.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3375, pruned_loss=0.09394, over 5724701.29 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3415, pruned_loss=0.08805, over 5481203.81 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3375, pruned_loss=0.09465, over 5722924.06 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:55:33,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.060e+02 1.399e+03 1.667e+03 2.028e+03 4.483e+03, threshold=3.335e+03, percent-clipped=3.0
+2023-03-14 19:56:00,817 INFO [train.py:968] (0/2) Epoch 29, batch 5650, giga_loss[loss=0.2281, simple_loss=0.313, pruned_loss=0.07155, over 28921.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3378, pruned_loss=0.09426, over 5718793.71 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3414, pruned_loss=0.08805, over 5489597.51 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3379, pruned_loss=0.09494, over 5714208.13 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:56:39,671 INFO [train.py:968] (0/2) Epoch 29, batch 5700, libri_loss[loss=0.2177, simple_loss=0.3007, pruned_loss=0.06733, over 29627.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3344, pruned_loss=0.09254, over 5722727.96 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3415, pruned_loss=0.08829, over 5505520.72 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3341, pruned_loss=0.09311, over 5712593.08 frames. ], batch size: 69, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:56:58,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.594e+02 1.308e+03 1.464e+03 1.741e+03 4.279e+03, threshold=2.928e+03, percent-clipped=2.0
+2023-03-14 19:57:19,946 INFO [train.py:968] (0/2) Epoch 29, batch 5750, giga_loss[loss=0.2181, simple_loss=0.3005, pruned_loss=0.06792, over 29024.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3304, pruned_loss=0.09082, over 5717275.80 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3412, pruned_loss=0.08811, over 5511262.16 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3303, pruned_loss=0.09146, over 5707001.23 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:57:58,016 INFO [train.py:968] (0/2) Epoch 29, batch 5800, giga_loss[loss=0.2991, simple_loss=0.3608, pruned_loss=0.1187, over 28717.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3291, pruned_loss=0.09031, over 5719010.22 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3416, pruned_loss=0.08826, over 5520493.67 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3283, pruned_loss=0.09075, over 5708788.12 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:58:17,389 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.311e+02 1.516e+03 1.755e+03 2.237e+03 5.103e+03, threshold=3.510e+03, percent-clipped=13.0
+2023-03-14 19:58:20,326 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1281647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:58:36,539 INFO [train.py:968] (0/2) Epoch 29, batch 5850, libri_loss[loss=0.2074, simple_loss=0.2913, pruned_loss=0.06177, over 29418.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3287, pruned_loss=0.08968, over 5721423.95 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3415, pruned_loss=0.08823, over 5525781.02 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.328, pruned_loss=0.09009, over 5711150.05 frames. ], batch size: 67, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:58:41,840 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1281677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:58:53,974 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3122, 2.4110, 1.8721, 1.9854], device='cuda:0'), covar=tensor([0.0930, 0.0733, 0.0966, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0450, 0.0525, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 19:59:09,805 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 19:59:13,937 INFO [train.py:968] (0/2) Epoch 29, batch 5900, giga_loss[loss=0.2743, simple_loss=0.3597, pruned_loss=0.09451, over 28606.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3332, pruned_loss=0.0919, over 5719306.17 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3423, pruned_loss=0.08878, over 5535326.30 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3315, pruned_loss=0.09183, over 5708267.36 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:59:30,424 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.009e+02 1.294e+03 1.675e+03 2.220e+03 4.687e+03, threshold=3.350e+03, percent-clipped=7.0
+2023-03-14 19:59:52,884 INFO [train.py:968] (0/2) Epoch 29, batch 5950, libri_loss[loss=0.2663, simple_loss=0.3529, pruned_loss=0.08981, over 29182.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3363, pruned_loss=0.09265, over 5714729.21 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3429, pruned_loss=0.089, over 5541179.82 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3343, pruned_loss=0.0925, over 5705951.63 frames. ], batch size: 97, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:59:56,425 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8916, 2.9098, 1.8214, 1.1594], device='cuda:0'), covar=tensor([0.9614, 0.3747, 0.4943, 0.7699], device='cuda:0'), in_proj_covar=tensor([0.1845, 0.1726, 0.1663, 0.1505], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 20:00:07,845 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1281790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:00:09,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1281793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:00:32,739 INFO [train.py:968] (0/2) Epoch 29, batch 6000, giga_loss[loss=0.2941, simple_loss=0.3734, pruned_loss=0.1074, over 29040.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3401, pruned_loss=0.09408, over 5724265.90 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.08912, over 5557333.16 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3381, pruned_loss=0.09404, over 5709620.22 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:00:32,743 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 20:00:40,809 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.2055, simple_loss=0.312, pruned_loss=0.04951, over 944034.00 frames. 
+2023-03-14 20:00:40,810 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 20:00:41,683 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1281822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:01:00,581 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.512e+02 1.312e+03 1.471e+03 1.933e+03 5.170e+03, threshold=2.942e+03, percent-clipped=2.0
+2023-03-14 20:01:21,533 INFO [train.py:968] (0/2) Epoch 29, batch 6050, giga_loss[loss=0.2616, simple_loss=0.3377, pruned_loss=0.09275, over 28709.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3425, pruned_loss=0.09541, over 5712767.68 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3431, pruned_loss=0.08921, over 5555721.27 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3409, pruned_loss=0.09544, over 5706863.17 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:02:06,308 INFO [train.py:968] (0/2) Epoch 29, batch 6100, giga_loss[loss=0.2624, simple_loss=0.3435, pruned_loss=0.09067, over 28975.00 frames. ], tot_loss[loss=0.27, simple_loss=0.345, pruned_loss=0.09748, over 5703998.78 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3426, pruned_loss=0.08888, over 5559913.39 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3441, pruned_loss=0.09793, over 5697494.07 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:02:23,163 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3682, 1.5972, 1.6856, 1.4225], device='cuda:0'), covar=tensor([0.2841, 0.2455, 0.2570, 0.2572], device='cuda:0'), in_proj_covar=tensor([0.2065, 0.2036, 0.1938, 0.2082], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 20:02:24,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3816, 4.0186, 1.4803, 1.5647], device='cuda:0'), covar=tensor([0.1007, 0.0391, 0.0941, 0.1330], device='cuda:0'), in_proj_covar=tensor([0.0425, 0.0569, 0.0411, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 20:02:25,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.118e+02 1.394e+03 1.739e+03 2.220e+03 4.853e+03, threshold=3.478e+03, percent-clipped=6.0
+2023-03-14 20:02:49,703 INFO [train.py:968] (0/2) Epoch 29, batch 6150, giga_loss[loss=0.3167, simple_loss=0.3831, pruned_loss=0.1251, over 28930.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3503, pruned_loss=0.1019, over 5697753.04 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3422, pruned_loss=0.08872, over 5562010.87 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3503, pruned_loss=0.1028, over 5695618.15 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:02:56,883 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5323, 1.9290, 1.7491, 1.7820], device='cuda:0'), covar=tensor([0.0762, 0.0281, 0.0309, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:0')
+2023-03-14 20:03:11,275 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3939, 3.1320, 1.4242, 1.4110], device='cuda:0'), covar=tensor([0.1012, 0.0393, 0.0927, 0.1439], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0570, 0.0412, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 20:03:15,047 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1282000.pt
+2023-03-14 20:03:37,127 INFO [train.py:968] (0/2) Epoch 29, batch 6200, libri_loss[loss=0.3089, simple_loss=0.3801, pruned_loss=0.1188, over 29745.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3574, pruned_loss=0.1073, over 5699229.96 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08891, over 5569038.19 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3573, pruned_loss=0.1082, over 5693468.63 frames. ], batch size: 87, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:03:56,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.767e+03 2.358e+03 3.283e+03 1.165e+04, threshold=4.716e+03, percent-clipped=20.0
+2023-03-14 20:03:58,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4212, 1.7326, 1.4137, 1.3982], device='cuda:0'), covar=tensor([0.2530, 0.2577, 0.2871, 0.2248], device='cuda:0'), in_proj_covar=tensor([0.1611, 0.1161, 0.1423, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 20:04:03,898 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1282052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:04:21,027 INFO [train.py:968] (0/2) Epoch 29, batch 6250, giga_loss[loss=0.3007, simple_loss=0.3663, pruned_loss=0.1175, over 28556.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3632, pruned_loss=0.1117, over 5687459.60 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3424, pruned_loss=0.08898, over 5570088.54 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3637, pruned_loss=0.1129, over 5685265.87 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:05:05,315 INFO [train.py:968] (0/2) Epoch 29, batch 6300, giga_loss[loss=0.3115, simple_loss=0.3715, pruned_loss=0.1257, over 28473.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3676, pruned_loss=0.1153, over 5697183.12 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3419, pruned_loss=0.08869, over 5577171.78 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.369, pruned_loss=0.117, over 5691961.84 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:05:27,332 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.883e+03 2.457e+03 3.544e+03 1.774e+04, threshold=4.914e+03, percent-clipped=12.0
+2023-03-14 20:05:49,721 INFO [train.py:968] (0/2) Epoch 29, batch 6350, giga_loss[loss=0.3532, simple_loss=0.4188, pruned_loss=0.1438, over 28877.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.374, pruned_loss=0.1205, over 5693638.81 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.08874, over 5583798.26 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3755, pruned_loss=0.1224, over 5685600.97 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:06:05,949 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-14 20:06:12,422 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1282195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:06:15,493 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1282198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:06:37,664 INFO [train.py:968] (0/2) Epoch 29, batch 6400, giga_loss[loss=0.3005, simple_loss=0.3591, pruned_loss=0.121, over 28529.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.376, pruned_loss=0.1227, over 5687419.21 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3419, pruned_loss=0.08873, over 5592572.24 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3784, pruned_loss=0.1254, over 5676199.97 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:06:44,064 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1282227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:07:01,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.176e+03 2.026e+03 2.754e+03 3.639e+03 6.793e+03, threshold=5.508e+03, percent-clipped=12.0
+2023-03-14 20:07:27,543 INFO [train.py:968] (0/2) Epoch 29, batch 6450, giga_loss[loss=0.3252, simple_loss=0.3891, pruned_loss=0.1307, over 28767.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3796, pruned_loss=0.1268, over 5671473.70 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.0887, over 5599053.56 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3821, pruned_loss=0.1297, over 5658150.51 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:07:30,874 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-14 20:08:13,171 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-14 20:08:16,188 INFO [train.py:968] (0/2) Epoch 29, batch 6500, giga_loss[loss=0.2805, simple_loss=0.3473, pruned_loss=0.1068, over 29015.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3814, pruned_loss=0.1291, over 5678120.44 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3415, pruned_loss=0.08841, over 5607238.04 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3851, pruned_loss=0.1329, over 5662514.65 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:08:42,407 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.328e+03 2.160e+03 2.747e+03 4.260e+03 9.178e+03, threshold=5.495e+03, percent-clipped=11.0
+2023-03-14 20:09:11,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6667, 1.8865, 1.8506, 1.4356], device='cuda:0'), covar=tensor([0.1739, 0.2629, 0.1552, 0.1926], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0717, 0.0983, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 20:09:11,427 INFO [train.py:968] (0/2) Epoch 29, batch 6550, libri_loss[loss=0.2931, simple_loss=0.3751, pruned_loss=0.1056, over 25704.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3845, pruned_loss=0.1325, over 5656443.09 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.08871, over 5610891.76 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.388, pruned_loss=0.1364, over 5641694.79 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:09:12,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9977, 2.4011, 2.2570, 1.6500], device='cuda:0'), covar=tensor([0.3691, 0.2484, 0.2678, 0.3476], device='cuda:0'), in_proj_covar=tensor([0.2068, 0.2046, 0.1942, 0.2085], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 20:09:24,952 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-14 20:10:01,071 INFO [train.py:968] (0/2) Epoch 29, batch 6600, giga_loss[loss=0.4372, simple_loss=0.4491, pruned_loss=0.2127, over 27598.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3872, pruned_loss=0.1349, over 5650078.07 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3425, pruned_loss=0.08894, over 5613080.40 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3904, pruned_loss=0.1385, over 5637460.94 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:10:23,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.397e+03 2.052e+03 2.687e+03 3.714e+03 7.605e+03, threshold=5.373e+03, percent-clipped=6.0
+2023-03-14 20:10:49,159 INFO [train.py:968] (0/2) Epoch 29, batch 6650, giga_loss[loss=0.3007, simple_loss=0.3688, pruned_loss=0.1163, over 28798.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3864, pruned_loss=0.1353, over 5651690.15 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08905, over 5616073.03 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3892, pruned_loss=0.1385, over 5639618.54 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:11:08,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4339, 1.6088, 1.5243, 1.3969], device='cuda:0'), covar=tensor([0.3038, 0.2509, 0.2199, 0.2399], device='cuda:0'), in_proj_covar=tensor([0.2073, 0.2050, 0.1946, 0.2089], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 20:11:42,105 INFO [train.py:968] (0/2) Epoch 29, batch 6700, giga_loss[loss=0.2879, simple_loss=0.3545, pruned_loss=0.1107, over 28887.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3849, pruned_loss=0.1348, over 5639147.12 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.342, pruned_loss=0.08864, over 5623081.92 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3887, pruned_loss=0.1389, over 5623583.11 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:11:58,284 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1282538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:12:05,933 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.245e+03 1.962e+03 2.821e+03 4.035e+03 1.088e+04, threshold=5.642e+03, percent-clipped=7.0
+2023-03-14 20:12:20,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9233, 1.1041, 1.1103, 0.9506], device='cuda:0'), covar=tensor([0.2210, 0.2588, 0.1621, 0.2001], device='cuda:0'), in_proj_covar=tensor([0.2075, 0.2054, 0.1949, 0.2093], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 20:12:30,354 INFO [train.py:968] (0/2) Epoch 29, batch 6750, libri_loss[loss=0.3156, simple_loss=0.373, pruned_loss=0.1291, over 29570.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.386, pruned_loss=0.1341, over 5654867.60 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08927, over 5629929.58 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3895, pruned_loss=0.138, over 5636657.01 frames. ], batch size: 74, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:12:31,826 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3761, 3.1747, 1.5008, 1.5060], device='cuda:0'), covar=tensor([0.0999, 0.0334, 0.0881, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0574, 0.0413, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 20:12:32,438 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2567, 1.4957, 1.5023, 1.3388], device='cuda:0'), covar=tensor([0.1595, 0.1338, 0.1897, 0.1439], device='cuda:0'), in_proj_covar=tensor([0.0513, 0.0763, 0.0737, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-14 20:12:53,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6994, 1.9249, 1.8232, 1.6841], device='cuda:0'), covar=tensor([0.2301, 0.2339, 0.2547, 0.2330], device='cuda:0'), in_proj_covar=tensor([0.0514, 0.0763, 0.0737, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-14 20:12:54,705 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3946, 1.7734, 1.5447, 1.4772], device='cuda:0'), covar=tensor([0.0793, 0.0331, 0.0318, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 20:13:17,345 INFO [train.py:968] (0/2) Epoch 29, batch 6800, giga_loss[loss=0.3514, simple_loss=0.4116, pruned_loss=0.1457, over 28760.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3865, pruned_loss=0.1337, over 5652029.01 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.08917, over 5633835.57 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3899, pruned_loss=0.1376, over 5634396.17 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:13:41,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.171e+03 1.877e+03 2.552e+03 3.727e+03 6.905e+03, threshold=5.103e+03, percent-clipped=4.0
+2023-03-14 20:13:55,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7032, 1.8959, 1.5155, 1.8398], device='cuda:0'), covar=tensor([0.2591, 0.2778, 0.3059, 0.2558], device='cuda:0'), in_proj_covar=tensor([0.1604, 0.1157, 0.1418, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 20:14:02,506 INFO [train.py:968] (0/2) Epoch 29, batch 6850, giga_loss[loss=0.293, simple_loss=0.3633, pruned_loss=0.1113, over 28192.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3861, pruned_loss=0.1333, over 5624706.01 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08942, over 5619850.79 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3899, pruned_loss=0.1375, over 5624132.09 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:14:47,792 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1282714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:14:53,681 INFO [train.py:968] (0/2) Epoch 29, batch 6900, giga_loss[loss=0.2873, simple_loss=0.3639, pruned_loss=0.1053, over 28873.00 frames. ], tot_loss[loss=0.322, simple_loss=0.383, pruned_loss=0.1305, over 5618224.28 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08942, over 5615683.97 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3869, pruned_loss=0.1347, over 5621065.90 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:15:12,388 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1282739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:15:12,450 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3345, 1.3165, 1.2424, 1.5696], device='cuda:0'), covar=tensor([0.0738, 0.0412, 0.0362, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0228], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 20:15:13,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5924, 3.5652, 1.6304, 1.7653], device='cuda:0'), covar=tensor([0.0973, 0.0374, 0.0914, 0.1293], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0573, 0.0413, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 20:15:18,869 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.243e+03 1.902e+03 2.595e+03 3.721e+03 9.067e+03, threshold=5.191e+03, percent-clipped=11.0
+2023-03-14 20:15:42,739 INFO [train.py:968] (0/2) Epoch 29, batch 6950, giga_loss[loss=0.3581, simple_loss=0.4078, pruned_loss=0.1542, over 28566.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3806, pruned_loss=0.1272, over 5636230.64 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08932, over 5618644.69 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3842, pruned_loss=0.131, over 5636005.35 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:16:30,716 INFO [train.py:968] (0/2) Epoch 29, batch 7000, giga_loss[loss=0.2664, simple_loss=0.3489, pruned_loss=0.09189, over 28936.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3771, pruned_loss=0.1245, over 5638881.42 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3426, pruned_loss=0.0893, over 5616778.79 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3812, pruned_loss=0.1285, over 5640243.61 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:16:54,977 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.177e+03 1.751e+03 2.377e+03 3.317e+03 7.840e+03, threshold=4.753e+03, percent-clipped=9.0
+2023-03-14 20:17:16,663 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-14 20:17:16,794 INFO [train.py:968] (0/2) Epoch 29, batch 7050, giga_loss[loss=0.3074, simple_loss=0.3758, pruned_loss=0.1195, over 28948.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.374, pruned_loss=0.1219, over 5641948.28 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08916, over 5620606.34 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3778, pruned_loss=0.1257, over 5639622.45 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:17:39,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6795, 5.5401, 5.3024, 2.6203], device='cuda:0'), covar=tensor([0.0412, 0.0499, 0.0597, 0.1642], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.1209, 0.1017, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 20:17:59,911 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1282913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:18:05,925 INFO [train.py:968] (0/2) Epoch 29, batch 7100, giga_loss[loss=0.3075, simple_loss=0.3713, pruned_loss=0.1218, over 28606.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3727, pruned_loss=0.1213, over 5645317.84 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3427, pruned_loss=0.08923, over 5624195.06 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3761, pruned_loss=0.1248, over 5640389.74 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:18:28,449 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.714e+03 2.103e+03 2.748e+03 4.819e+03, threshold=4.205e+03, percent-clipped=1.0
+2023-03-14 20:18:29,541 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 20:18:32,711 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2966, 3.6097, 1.4408, 1.5595], device='cuda:0'), covar=tensor([0.1096, 0.0330, 0.0969, 0.1442], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0574, 0.0412, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 20:18:43,538 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7854, 2.0895, 1.4175, 1.5879], device='cuda:0'), covar=tensor([0.1103, 0.0694, 0.1097, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0453, 0.0527, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 20:18:51,787 INFO [train.py:968] (0/2) Epoch 29, batch 7150, giga_loss[loss=0.3209, simple_loss=0.3785, pruned_loss=0.1317, over 27947.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3716, pruned_loss=0.1203, over 5649893.60 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.08922, over 5635854.15 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3755, pruned_loss=0.1244, over 5636118.61 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:19:18,157 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1282998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:19:38,123 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 20:19:43,712 INFO [train.py:968] (0/2) Epoch 29, batch 7200, giga_loss[loss=0.2698, simple_loss=0.3478, pruned_loss=0.09588, over 29092.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3697, pruned_loss=0.1187, over 5651573.01 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08933, over 5637501.47 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3729, pruned_loss=0.1221, over 5639436.24 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:20:06,945 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.178e+03 1.698e+03 2.082e+03 2.546e+03 8.284e+03, threshold=4.165e+03, percent-clipped=4.0
+2023-03-14 20:20:16,015 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1283056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:20:19,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1283059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:20:36,386 INFO [train.py:968] (0/2) Epoch 29, batch 7250, giga_loss[loss=0.3168, simple_loss=0.3908, pruned_loss=0.1214, over 28524.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3689, pruned_loss=0.1155, over 5659486.68 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3431, pruned_loss=0.08934, over 5640398.08 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3719, pruned_loss=0.1189, over 5647831.40 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:20:56,091 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1283088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:20:56,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4387, 3.3602, 1.5450, 1.5785], device='cuda:0'), covar=tensor([0.0987, 0.0352, 0.0926, 0.1333], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0573, 0.0412, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 20:20:56,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1283089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:21:19,076 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1283114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:21:25,182 INFO [train.py:968] (0/2) Epoch 29, batch 7300, giga_loss[loss=0.326, simple_loss=0.3964, pruned_loss=0.1278, over 28970.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3713, pruned_loss=0.1154, over 5668018.57 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08927, over 5641585.09 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3739, pruned_loss=0.1183, over 5657861.59 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:21:32,871 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-14 20:21:52,561 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.888e+03 2.501e+03 3.891e+03 1.383e+04, threshold=5.001e+03, percent-clipped=20.0
+2023-03-14 20:21:52,900 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6154, 1.7972, 1.3031, 1.3005], device='cuda:0'), covar=tensor([0.1113, 0.0655, 0.1111, 0.1206], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0452, 0.0525, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 20:21:54,931 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1283149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:22:16,237 INFO [train.py:968] (0/2) Epoch 29, batch 7350, giga_loss[loss=0.3015, simple_loss=0.3773, pruned_loss=0.1128, over 28485.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3714, pruned_loss=0.116, over 5664984.52 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3425, pruned_loss=0.08915, over 5646292.71 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3746, pruned_loss=0.119, over 5653191.87 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:22:52,049 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-14 20:23:00,781 INFO [train.py:968] (0/2) Epoch 29, batch 7400, giga_loss[loss=0.3006, simple_loss=0.3676, pruned_loss=0.1168, over 28558.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3717, pruned_loss=0.1169, over 5657112.50 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3427, pruned_loss=0.08927, over 5642880.75 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3748, pruned_loss=0.1199, over 5651427.16 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:23:09,466 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1283232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:23:12,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1283235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:23:22,589 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.275e+03 1.865e+03 2.467e+03 3.102e+03 6.092e+03, threshold=4.934e+03, percent-clipped=4.0
+2023-03-14 20:23:32,385 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1283257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:23:37,308 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1283260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:23:42,521 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1283264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:23:48,511 INFO [train.py:968] (0/2) Epoch 29, batch 7450, giga_loss[loss=0.2961, simple_loss=0.3561, pruned_loss=0.118, over 28899.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3695, pruned_loss=0.1163, over 5668208.59 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3424, pruned_loss=0.08923, over 5648178.91 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3727, pruned_loss=0.1192, over 5659489.87 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:24:01,382 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 20:24:03,149 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1283289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:24:32,085 INFO [train.py:968] (0/2) Epoch 29, batch 7500, giga_loss[loss=0.3033, simple_loss=0.3679, pruned_loss=0.1194, over 28527.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3689, pruned_loss=0.1175, over 5658537.92 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3426, pruned_loss=0.08953, over 5645141.59 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.372, pruned_loss=0.1203, over 5655236.23 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:24:54,523 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.836e+03 2.305e+03 2.812e+03 9.475e+03, threshold=4.610e+03, percent-clipped=6.0
+2023-03-14 20:25:22,652 INFO [train.py:968] (0/2) Epoch 29, batch 7550, giga_loss[loss=0.333, simple_loss=0.3913, pruned_loss=0.1373, over 28856.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3686, pruned_loss=0.117, over 5662333.99 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.343, pruned_loss=0.08978, over 5647845.82 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.371, pruned_loss=0.1194, over 5657431.03 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:25:24,886 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1283373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:25:35,319 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2630, 1.6230, 1.6519, 1.3762], device='cuda:0'), covar=tensor([0.2358, 0.2099, 0.2608, 0.2474], device='cuda:0'), in_proj_covar=tensor([0.0513, 0.0764, 0.0738, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-14 20:26:08,690 INFO [train.py:968] (0/2) Epoch 29, batch 7600, giga_loss[loss=0.3182, simple_loss=0.3846, pruned_loss=0.1259, over 29030.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3675, pruned_loss=0.115, over 5668013.92 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.343, pruned_loss=0.08968, over 5652952.67 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3699, pruned_loss=0.1175, over 5659800.74 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:26:33,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.102e+03 1.500e+03 1.960e+03 2.490e+03 4.896e+03, threshold=3.921e+03, percent-clipped=2.0
+2023-03-14 20:26:54,459 INFO [train.py:968] (0/2) Epoch 29, batch 7650, giga_loss[loss=0.3418, simple_loss=0.3939, pruned_loss=0.1449, over 28878.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3672, pruned_loss=0.1145, over 5670096.86 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3426, pruned_loss=0.0895, over 5656258.14 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3698, pruned_loss=0.117, over 5660756.09 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:27:16,085 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1283495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:27:26,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2607, 3.5261, 2.4364, 1.4606], device='cuda:0'), covar=tensor([0.7851, 0.2684, 0.3858, 0.6720], device='cuda:0'), in_proj_covar=tensor([0.1852, 0.1740, 0.1666, 0.1509], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 20:27:36,532 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1283516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:27:39,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1283519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:27:42,099 INFO [train.py:968] (0/2) Epoch 29, batch 7700, giga_loss[loss=0.2742, simple_loss=0.3455, pruned_loss=0.1015, over 28881.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3657, pruned_loss=0.1133, over 5685288.50 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3428, pruned_loss=0.08955, over 5657100.85 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3677, pruned_loss=0.1153, over 5677358.15 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:27:45,657 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1283524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:28:06,745 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.735e+03 2.114e+03 2.897e+03 5.760e+03, threshold=4.229e+03, percent-clipped=9.0
+2023-03-14 20:28:06,986 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1283548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:28:26,632 INFO [train.py:968] (0/2) Epoch 29, batch 7750, giga_loss[loss=0.3116, simple_loss=0.3785, pruned_loss=0.1223, over 28994.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.364, pruned_loss=0.1128, over 5680796.93 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3429, pruned_loss=0.08955, over 5662958.03 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3662, pruned_loss=0.115, over 5669686.29 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:28:36,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1316, 1.4091, 1.3632, 1.0470], device='cuda:0'), covar=tensor([0.1422, 0.2102, 0.1194, 0.1414], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0719, 0.0985, 0.0884], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 20:29:19,010 INFO [train.py:968] (0/2) Epoch 29, batch 7800, giga_loss[loss=0.3349, simple_loss=0.3966, pruned_loss=0.1366, over 28950.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3632, pruned_loss=0.1135, over 5672476.91 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3425, pruned_loss=0.08943, over 5666576.81 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3656, pruned_loss=0.1158, over 5660413.92 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:29:38,255 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.0051, 2.0382, 3.9967, 3.7003], device='cuda:0'), covar=tensor([0.1253, 0.2398, 0.0508, 0.1155], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0677, 0.1015, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 20:29:44,076 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.145e+03 1.748e+03 2.210e+03 3.032e+03 9.643e+03, threshold=4.420e+03, percent-clipped=9.0
+2023-03-14 20:29:58,151 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.8696, 5.6235, 5.3547, 2.8518], device='cuda:0'), covar=tensor([0.0563, 0.0701, 0.0975, 0.1859], device='cuda:0'), in_proj_covar=tensor([0.1316, 0.1217, 0.1022, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 20:29:59,472 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1283667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:30:01,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1283670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:30:02,107 INFO [train.py:968] (0/2) Epoch 29, batch 7850, giga_loss[loss=0.2682, simple_loss=0.3483, pruned_loss=0.09402, over 28894.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3631, pruned_loss=0.1142, over 5649564.57 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3428, pruned_loss=0.08945, over 5649729.37 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3654, pruned_loss=0.1168, over 5655523.31 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:30:27,810 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1283699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:30:47,614 INFO [train.py:968] (0/2) Epoch 29, batch 7900, giga_loss[loss=0.3101, simple_loss=0.3796, pruned_loss=0.1202, over 28605.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3626, pruned_loss=0.1147, over 5645939.93 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.343, pruned_loss=0.08947, over 5652595.26 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3645, pruned_loss=0.117, over 5648181.58 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:31:12,468 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.184e+03 1.800e+03 2.524e+03 3.561e+03 6.343e+03, threshold=5.048e+03, percent-clipped=10.0
+2023-03-14 20:31:31,066 INFO [train.py:968] (0/2) Epoch 29, batch 7950, giga_loss[loss=0.3055, simple_loss=0.3725, pruned_loss=0.1192, over 28594.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3613, pruned_loss=0.1139, over 5646980.33 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3424, pruned_loss=0.08906, over 5649939.38 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.364, pruned_loss=0.117, over 5650792.90 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:32:15,029 INFO [train.py:968] (0/2) Epoch 29, batch 8000, giga_loss[loss=0.266, simple_loss=0.3453, pruned_loss=0.09336, over 28505.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3614, pruned_loss=0.1134, over 5655735.27 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3418, pruned_loss=0.08872, over 5656340.81 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3646, pruned_loss=0.1169, over 5653005.89 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:32:42,036 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.208e+03 1.841e+03 2.386e+03 3.772e+03 1.155e+04, threshold=4.772e+03, percent-clipped=13.0
+2023-03-14 20:32:56,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1283868.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:32:57,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1283870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:32:58,787 INFO [train.py:968] (0/2) Epoch 29, batch 8050, giga_loss[loss=0.3218, simple_loss=0.3832, pruned_loss=0.1302, over 28556.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3616, pruned_loss=0.1127, over 5660606.92 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3415, pruned_loss=0.08855, over 5659740.54 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.365, pruned_loss=0.1163, over 5655565.19 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:33:08,952 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1283881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:33:10,166 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1283882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:33:40,944 INFO [train.py:968] (0/2) Epoch 29, batch 8100, giga_loss[loss=0.2738, simple_loss=0.3552, pruned_loss=0.09621, over 29038.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3632, pruned_loss=0.1127, over 5674313.51 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3418, pruned_loss=0.08862, over 5662567.08 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3661, pruned_loss=0.1159, over 5667775.69 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:34:06,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.033e+03 1.504e+03 1.868e+03 2.422e+03 5.228e+03, threshold=3.737e+03, percent-clipped=1.0
+2023-03-14 20:34:21,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4290, 2.7189, 1.5586, 1.6245], device='cuda:0'), covar=tensor([0.0826, 0.0338, 0.0711, 0.1085], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0572, 0.0411, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 20:34:25,518 INFO [train.py:968] (0/2) Epoch 29, batch 8150, giga_loss[loss=0.2775, simple_loss=0.3533, pruned_loss=0.1008, over 28860.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3631, pruned_loss=0.1122, over 5682246.99 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08878, over 5668789.68 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3657, pruned_loss=0.1152, over 5671940.16 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:34:53,246 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1284000.pt
+2023-03-14 20:35:04,399 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1284013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:35:07,840 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1284016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:35:12,588 INFO [train.py:968] (0/2) Epoch 29, batch 8200, giga_loss[loss=0.3356, simple_loss=0.3922, pruned_loss=0.1395, over 28568.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3649, pruned_loss=0.1143, over 5681857.66 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3421, pruned_loss=0.0889, over 5672286.37 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.367, pruned_loss=0.1169, over 5670786.13 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:35:38,304 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1284045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:35:42,422 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.780e+03 2.298e+03 3.337e+03 6.876e+03, threshold=4.597e+03, percent-clipped=19.0
+2023-03-14 20:36:03,157 INFO [train.py:968] (0/2) Epoch 29, batch 8250, giga_loss[loss=0.2966, simple_loss=0.3609, pruned_loss=0.1161, over 28877.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3676, pruned_loss=0.1174, over 5667731.38 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3421, pruned_loss=0.08871, over 5674015.22 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3701, pruned_loss=0.1204, over 5657349.40 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:36:09,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5022, 1.5597, 1.6868, 1.3953], device='cuda:0'), covar=tensor([0.1263, 0.1997, 0.1116, 0.1466], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0719, 0.0983, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 20:36:48,922 INFO [train.py:968] (0/2) Epoch 29, batch 8300, giga_loss[loss=0.4382, simple_loss=0.4591, pruned_loss=0.2086, over 23781.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3684, pruned_loss=0.1198, over 5661182.66 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3414, pruned_loss=0.08829, over 5677414.84 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3715, pruned_loss=0.1233, over 5649631.01 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:37:15,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.120e+03 1.860e+03 2.294e+03 3.068e+03 8.251e+03, threshold=4.588e+03, percent-clipped=8.0
+2023-03-14 20:37:33,687 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6166, 2.0929, 1.8156, 1.9740], device='cuda:0'), covar=tensor([0.0761, 0.0284, 0.0299, 0.0813], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 20:37:34,114 INFO [train.py:968] (0/2) Epoch 29, batch 8350, giga_loss[loss=0.2474, simple_loss=0.3233, pruned_loss=0.0858, over 28636.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3687, pruned_loss=0.1205, over 5664086.24 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3415, pruned_loss=0.0883, over 5676932.65 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3722, pruned_loss=0.1245, over 5655309.73 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:37:41,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7842, 1.6194, 1.8619, 1.3826], device='cuda:0'), covar=tensor([0.1692, 0.2810, 0.1318, 0.1542], device='cuda:0'), in_proj_covar=tensor([0.0935, 0.0718, 0.0982, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 20:38:05,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4231, 3.1292, 1.4718, 1.6076], device='cuda:0'), covar=tensor([0.0991, 0.0394, 0.0885, 0.1324], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0573, 0.0411, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 20:38:18,117 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4778, 2.0031, 1.6096, 1.6568], device='cuda:0'), covar=tensor([0.0786, 0.0298, 0.0322, 0.0876], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 20:38:18,276 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-14 20:38:19,174 INFO [train.py:968] (0/2) Epoch 29, batch 8400, libri_loss[loss=0.2465, simple_loss=0.3375, pruned_loss=0.07772, over 29368.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3698, pruned_loss=0.1212, over 5659170.71 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3419, pruned_loss=0.08843, over 5679153.04 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3732, pruned_loss=0.1254, over 5649377.94 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:38:32,034 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3553, 1.6003, 1.2549, 1.1486], device='cuda:0'), covar=tensor([0.1106, 0.0563, 0.1119, 0.1165], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0454, 0.0528, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 20:38:36,314 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1284243.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:38:41,067 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.242e+03 1.707e+03 2.419e+03 3.381e+03 1.348e+04, threshold=4.838e+03, percent-clipped=12.0
+2023-03-14 20:38:46,603 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1284256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:38:47,160 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1284257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:38:56,517 INFO [train.py:968] (0/2) Epoch 29, batch 8450, giga_loss[loss=0.297, simple_loss=0.3626, pruned_loss=0.1157, over 28506.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3667, pruned_loss=0.1184, over 5668956.42 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.0885, over 5677561.32 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3704, pruned_loss=0.1228, over 5661375.64 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:39:34,471 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-14 20:39:34,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3625, 1.9634, 1.4551, 0.6068], device='cuda:0'), covar=tensor([0.5891, 0.4295, 0.4364, 0.7282], device='cuda:0'), in_proj_covar=tensor([0.1852, 0.1741, 0.1663, 0.1506], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 20:39:35,221 INFO [train.py:968] (0/2) Epoch 29, batch 8500, giga_loss[loss=0.2859, simple_loss=0.3636, pruned_loss=0.1041, over 28630.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3653, pruned_loss=0.1153, over 5679496.68 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.342, pruned_loss=0.0884, over 5680468.71 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3692, pruned_loss=0.1202, over 5670737.09 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:39:59,614 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.795e+03 2.327e+03 3.071e+03 1.055e+04, threshold=4.653e+03, percent-clipped=7.0
+2023-03-14 20:40:14,012 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1284367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:40:16,407 INFO [train.py:968] (0/2) Epoch 29, batch 8550, giga_loss[loss=0.2763, simple_loss=0.3493, pruned_loss=0.1016, over 28824.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3641, pruned_loss=0.1137, over 5678728.61 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.342, pruned_loss=0.08829, over 5686095.01 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3681, pruned_loss=0.1188, over 5666588.43 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:40:27,964 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1284386.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:40:30,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1284389.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:40:38,320 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1284399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:40:38,992 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1284400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:40:40,219 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1284402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:40:40,804 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1284403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:40:55,328 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1284418.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:40:57,976 INFO [train.py:968] (0/2) Epoch 29, batch 8600, giga_loss[loss=0.3117, simple_loss=0.3698, pruned_loss=0.1268, over 28228.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3634, pruned_loss=0.1138, over 5684233.78 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3424, pruned_loss=0.08858, over 5691984.81 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3668, pruned_loss=0.1183, over 5669193.40 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:41:08,988 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1284431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:41:09,669 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1284432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:41:23,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.152e+03 1.700e+03 2.232e+03 2.821e+03 9.548e+03, threshold=4.464e+03, percent-clipped=3.0
+2023-03-14 20:41:37,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 20:41:42,738 INFO [train.py:968] (0/2) Epoch 29, batch 8650, giga_loss[loss=0.2565, simple_loss=0.3343, pruned_loss=0.08936, over 28999.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3615, pruned_loss=0.1135, over 5681622.11 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3423, pruned_loss=0.08856, over 5693096.89 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3643, pruned_loss=0.1171, over 5668855.95 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:41:48,858 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2124, 1.4575, 1.4727, 1.1082], device='cuda:0'), covar=tensor([0.1653, 0.2519, 0.1388, 0.1628], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0719, 0.0983, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 20:42:29,031 INFO [train.py:968] (0/2) Epoch 29, batch 8700, giga_loss[loss=0.2817, simple_loss=0.3592, pruned_loss=0.1021, over 28975.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3619, pruned_loss=0.1139, over 5679941.43 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3419, pruned_loss=0.0883, over 5697380.46 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.365, pruned_loss=0.1176, over 5665517.94 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:42:57,316 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.153e+03 1.777e+03 2.317e+03 3.203e+03 9.405e+03, threshold=4.634e+03, percent-clipped=9.0
+2023-03-14 20:43:15,982 INFO [train.py:968] (0/2) Epoch 29, batch 8750, giga_loss[loss=0.2962, simple_loss=0.3747, pruned_loss=0.1088, over 28879.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.365, pruned_loss=0.1153, over 5682747.67 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3413, pruned_loss=0.08801, over 5702496.32 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3686, pruned_loss=0.1192, over 5666083.63 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:43:59,961 INFO [train.py:968] (0/2) Epoch 29, batch 8800, giga_loss[loss=0.3102, simple_loss=0.3787, pruned_loss=0.1208, over 27509.00 frames. ], tot_loss[loss=0.3, simple_loss=0.369, pruned_loss=0.1155, over 5684197.36 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3417, pruned_loss=0.08842, over 5706212.41 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3721, pruned_loss=0.1189, over 5667182.59 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:44:28,130 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.765e+02 1.801e+03 2.316e+03 3.450e+03 1.271e+04, threshold=4.633e+03, percent-clipped=9.0
+2023-03-14 20:44:44,504 INFO [train.py:968] (0/2) Epoch 29, batch 8850, giga_loss[loss=0.3435, simple_loss=0.4, pruned_loss=0.1435, over 28685.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3704, pruned_loss=0.1158, over 5690140.83 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3416, pruned_loss=0.08835, over 5708807.01 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3734, pruned_loss=0.1189, over 5673996.02 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:44:49,781 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1284674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:44:53,014 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3435, 1.2266, 3.8938, 3.3921], device='cuda:0'), covar=tensor([0.1614, 0.2929, 0.0490, 0.1674], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0679, 0.1015, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 20:45:19,355 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 20:45:26,682 INFO [train.py:968] (0/2) Epoch 29, batch 8900, giga_loss[loss=0.2791, simple_loss=0.3534, pruned_loss=0.1024, over 28878.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.372, pruned_loss=0.1176, over 5681572.93 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3416, pruned_loss=0.08843, over 5699295.39 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3748, pruned_loss=0.1205, over 5676673.56 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:45:40,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1284736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:45:45,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1284742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:45:52,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.211e+03 1.723e+03 2.444e+03 3.058e+03 6.252e+03, threshold=4.889e+03, percent-clipped=6.0
+2023-03-14 20:45:53,315 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-14 20:46:09,293 INFO [train.py:968] (0/2) Epoch 29, batch 8950, libri_loss[loss=0.265, simple_loss=0.351, pruned_loss=0.08953, over 29217.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3718, pruned_loss=0.1179, over 5681377.14 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3416, pruned_loss=0.08839, over 5703224.14 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3749, pruned_loss=0.1211, over 5673459.88 frames. ], batch size: 97, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:46:19,326 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1284783.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:46:29,060 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1284795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:46:48,724 INFO [train.py:968] (0/2) Epoch 29, batch 9000, giga_loss[loss=0.3096, simple_loss=0.3783, pruned_loss=0.1205, over 28811.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3713, pruned_loss=0.1182, over 5691131.94 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3418, pruned_loss=0.08832, over 5712537.53 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3749, pruned_loss=0.1222, over 5675151.28 frames. ], batch size: 285, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:46:48,729 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 20:46:57,132 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.2024, simple_loss=0.3099, pruned_loss=0.04746, over 944034.00 frames. 
+2023-03-14 20:46:57,132 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 20:47:25,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.192e+03 1.872e+03 2.596e+03 3.768e+03 8.498e+03, threshold=5.192e+03, percent-clipped=11.0
+2023-03-14 20:47:40,288 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7762, 1.8061, 1.9975, 1.5457], device='cuda:0'), covar=tensor([0.1651, 0.2334, 0.1322, 0.1596], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0718, 0.0982, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 20:47:45,167 INFO [train.py:968] (0/2) Epoch 29, batch 9050, giga_loss[loss=0.2687, simple_loss=0.3385, pruned_loss=0.09941, over 28995.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3697, pruned_loss=0.118, over 5689028.56 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3418, pruned_loss=0.0883, over 5713367.58 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3731, pruned_loss=0.1215, over 5675377.44 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:47:55,786 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1284885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:47:57,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1284888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:48:15,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5809, 1.8343, 1.5153, 1.5855], device='cuda:0'), covar=tensor([0.2721, 0.2638, 0.2978, 0.2346], device='cuda:0'), in_proj_covar=tensor([0.1609, 0.1159, 0.1421, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 20:48:25,874 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1284917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:48:28,197 INFO [train.py:968] (0/2) Epoch 29, batch 9100, giga_loss[loss=0.2854, simple_loss=0.3514, pruned_loss=0.1097, over 28703.00 frames. ], tot_loss[loss=0.302, simple_loss=0.368, pruned_loss=0.118, over 5667932.53 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3416, pruned_loss=0.08822, over 5697700.51 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3713, pruned_loss=0.1215, over 5670291.29 frames. ], batch size: 66, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:48:55,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.195e+03 1.733e+03 2.190e+03 2.937e+03 7.494e+03, threshold=4.381e+03, percent-clipped=3.0
+2023-03-14 20:49:11,067 INFO [train.py:968] (0/2) Epoch 29, batch 9150, giga_loss[loss=0.2802, simple_loss=0.3465, pruned_loss=0.1069, over 28344.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3663, pruned_loss=0.1174, over 5663129.58 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3421, pruned_loss=0.08865, over 5692817.83 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3691, pruned_loss=0.1204, over 5669153.21 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:49:57,268 INFO [train.py:968] (0/2) Epoch 29, batch 9200, giga_loss[loss=0.255, simple_loss=0.3278, pruned_loss=0.09111, over 28905.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3676, pruned_loss=0.1186, over 5650547.61 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3427, pruned_loss=0.08902, over 5677202.99 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.37, pruned_loss=0.1214, over 5668055.54 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:50:00,401 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4428, 1.3830, 3.8842, 3.2481], device='cuda:0'), covar=tensor([0.1556, 0.2621, 0.0513, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0678, 0.1016, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 20:50:22,862 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:50:26,855 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.798e+03 2.393e+03 3.416e+03 6.487e+03, threshold=4.786e+03, percent-clipped=10.0
+2023-03-14 20:50:37,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-14 20:50:41,801 INFO [train.py:968] (0/2) Epoch 29, batch 9250, libri_loss[loss=0.315, simple_loss=0.3846, pruned_loss=0.1228, over 29553.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3661, pruned_loss=0.1181, over 5660082.09 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3426, pruned_loss=0.089, over 5681655.14 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3685, pruned_loss=0.1209, over 5669371.17 frames. ], batch size: 89, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:51:15,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:51:16,509 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285112.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:51:25,364 INFO [train.py:968] (0/2) Epoch 29, batch 9300, giga_loss[loss=0.2725, simple_loss=0.3423, pruned_loss=0.1013, over 28293.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3659, pruned_loss=0.118, over 5659701.22 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.08917, over 5676906.92 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3679, pruned_loss=0.1207, over 5670344.27 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:51:26,999 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:51:51,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.883e+03 2.167e+03 2.734e+03 5.623e+03, threshold=4.335e+03, percent-clipped=3.0
+2023-03-14 20:51:54,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1805, 1.2363, 3.2578, 2.8818], device='cuda:0'), covar=tensor([0.1628, 0.2758, 0.0561, 0.1439], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0680, 0.1019, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 20:51:54,904 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:51:55,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285158.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:52:05,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:52:05,967 INFO [train.py:968] (0/2) Epoch 29, batch 9350, giga_loss[loss=0.2742, simple_loss=0.3446, pruned_loss=0.1019, over 29030.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3645, pruned_loss=0.1162, over 5675997.40 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3421, pruned_loss=0.08865, over 5684262.40 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3678, pruned_loss=0.1199, over 5677769.33 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:52:19,294 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:52:28,198 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:52:31,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:52:53,243 INFO [train.py:968] (0/2) Epoch 29, batch 9400, giga_loss[loss=0.3262, simple_loss=0.3883, pruned_loss=0.1321, over 28910.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3662, pruned_loss=0.1171, over 5672932.36 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08859, over 5687588.33 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3693, pruned_loss=0.1205, over 5671127.79 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:52:56,388 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:53:19,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.307e+03 1.917e+03 2.416e+03 3.629e+03 7.548e+03, threshold=4.833e+03, percent-clipped=15.0
+2023-03-14 20:53:21,726 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:53:23,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:53:36,603 INFO [train.py:968] (0/2) Epoch 29, batch 9450, giga_loss[loss=0.2429, simple_loss=0.3188, pruned_loss=0.08355, over 28585.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3679, pruned_loss=0.1186, over 5669570.77 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3418, pruned_loss=0.08848, over 5690997.65 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3708, pruned_loss=0.1218, over 5664881.10 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:53:47,566 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:53:48,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8980, 1.0726, 0.8929, 0.2576], device='cuda:0'), covar=tensor([0.3872, 0.3314, 0.3306, 0.6751], device='cuda:0'), in_proj_covar=tensor([0.1854, 0.1742, 0.1665, 0.1508], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 20:54:01,032 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285301.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:54:03,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:54:12,470 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:54:15,515 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:54:15,712 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-14 20:54:20,375 INFO [train.py:968] (0/2) Epoch 29, batch 9500, giga_loss[loss=0.2983, simple_loss=0.385, pruned_loss=0.1058, over 28760.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3678, pruned_loss=0.1183, over 5674965.05 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08854, over 5696331.84 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3707, pruned_loss=0.1217, over 5666167.06 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:54:30,650 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285333.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:54:41,859 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:54:43,264 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-14 20:54:47,990 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.758e+03 2.140e+03 2.914e+03 7.572e+03, threshold=4.280e+03, percent-clipped=4.0
+2023-03-14 20:54:50,320 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-14 20:55:02,212 INFO [train.py:968] (0/2) Epoch 29, batch 9550, giga_loss[loss=0.2542, simple_loss=0.3469, pruned_loss=0.08072, over 29042.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3691, pruned_loss=0.1161, over 5684390.07 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3425, pruned_loss=0.08881, over 5697628.99 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3716, pruned_loss=0.1192, over 5675715.88 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:55:42,557 INFO [train.py:968] (0/2) Epoch 29, batch 9600, giga_loss[loss=0.2952, simple_loss=0.364, pruned_loss=0.1132, over 28266.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3714, pruned_loss=0.1159, over 5663146.63 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.343, pruned_loss=0.08909, over 5679171.09 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3733, pruned_loss=0.1185, over 5672129.22 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:56:10,953 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.141e+02 1.568e+03 2.016e+03 3.020e+03 6.397e+03, threshold=4.032e+03, percent-clipped=6.0
+2023-03-14 20:56:28,134 INFO [train.py:968] (0/2) Epoch 29, batch 9650, giga_loss[loss=0.2726, simple_loss=0.3569, pruned_loss=0.09411, over 29011.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3756, pruned_loss=0.1189, over 5660469.24 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3435, pruned_loss=0.08941, over 5674117.25 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3772, pruned_loss=0.1212, over 5671128.49 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:56:45,425 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285487.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:56:54,441 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:56:58,560 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:57:13,011 INFO [train.py:968] (0/2) Epoch 29, batch 9700, giga_loss[loss=0.2746, simple_loss=0.3444, pruned_loss=0.1024, over 28528.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3759, pruned_loss=0.1203, over 5658969.22 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3434, pruned_loss=0.0893, over 5668742.51 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3781, pruned_loss=0.123, over 5672268.79 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:57:22,775 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:57:43,356 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.161e+03 1.847e+03 2.381e+03 2.925e+03 4.786e+03, threshold=4.761e+03, percent-clipped=5.0
+2023-03-14 20:57:49,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:57:59,210 INFO [train.py:968] (0/2) Epoch 29, batch 9750, giga_loss[loss=0.35, simple_loss=0.3866, pruned_loss=0.1567, over 23333.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.376, pruned_loss=0.1218, over 5648252.35 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.08918, over 5662362.95 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3785, pruned_loss=0.1245, over 5664270.44 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:57:59,582 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-14 20:58:29,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9897, 2.3091, 1.8085, 2.2017], device='cuda:0'), covar=tensor([0.2482, 0.2580, 0.3042, 0.2453], device='cuda:0'), in_proj_covar=tensor([0.1610, 0.1161, 0.1425, 0.1020], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 20:58:41,423 INFO [train.py:968] (0/2) Epoch 29, batch 9800, giga_loss[loss=0.3816, simple_loss=0.4076, pruned_loss=0.1778, over 23749.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.375, pruned_loss=0.1216, over 5645028.73 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.0892, over 5666407.91 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3784, pruned_loss=0.125, over 5654066.56 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:58:50,175 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285630.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:58:52,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285633.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:58:59,999 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:01,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:07,672 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.763e+03 2.261e+03 2.937e+03 5.734e+03, threshold=4.523e+03, percent-clipped=6.0
+2023-03-14 20:59:16,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285662.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:59:23,031 INFO [train.py:968] (0/2) Epoch 29, batch 9850, giga_loss[loss=0.2906, simple_loss=0.3743, pruned_loss=0.1035, over 28863.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3731, pruned_loss=0.1194, over 5652998.64 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3428, pruned_loss=0.08917, over 5668754.76 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3763, pruned_loss=0.1225, over 5657793.50 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:59:25,861 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:27,271 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:30,109 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:51,890 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:53,792 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:54,355 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:00:05,047 INFO [train.py:968] (0/2) Epoch 29, batch 9900, giga_loss[loss=0.3435, simple_loss=0.4122, pruned_loss=0.1374, over 28712.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3705, pruned_loss=0.1156, over 5664772.18 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3422, pruned_loss=0.08885, over 5677722.46 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3747, pruned_loss=0.1195, over 5659894.38 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:00:19,488 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:00:33,342 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.646e+03 1.941e+03 2.907e+03 5.181e+03, threshold=3.882e+03, percent-clipped=3.0
+2023-03-14 21:00:48,177 INFO [train.py:968] (0/2) Epoch 29, batch 9950, giga_loss[loss=0.2939, simple_loss=0.3647, pruned_loss=0.1115, over 28570.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3705, pruned_loss=0.1152, over 5674983.92 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.342, pruned_loss=0.08885, over 5682348.08 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3745, pruned_loss=0.1187, over 5666848.41 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:01:10,853 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-14 21:01:19,161 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3999, 1.5645, 1.6294, 1.2381], device='cuda:0'), covar=tensor([0.1722, 0.2577, 0.1401, 0.1677], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0721, 0.0986, 0.0885], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 21:01:35,786 INFO [train.py:968] (0/2) Epoch 29, batch 10000, giga_loss[loss=0.2496, simple_loss=0.3279, pruned_loss=0.08562, over 29036.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3715, pruned_loss=0.1163, over 5670411.90 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3423, pruned_loss=0.08892, over 5688752.33 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3753, pruned_loss=0.1199, over 5657476.46 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:02:06,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.065e+03 1.819e+03 2.297e+03 3.411e+03 8.618e+03, threshold=4.595e+03, percent-clipped=16.0
+2023-03-14 21:02:06,653 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:02:20,266 INFO [train.py:968] (0/2) Epoch 29, batch 10050, giga_loss[loss=0.3172, simple_loss=0.3833, pruned_loss=0.1255, over 28574.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3702, pruned_loss=0.1155, over 5674547.28 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08891, over 5692575.21 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.374, pruned_loss=0.1191, over 5660195.64 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:02:27,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:03:11,499 INFO [train.py:968] (0/2) Epoch 29, batch 10100, libri_loss[loss=0.226, simple_loss=0.3061, pruned_loss=0.07301, over 29346.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3702, pruned_loss=0.1173, over 5669687.82 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3422, pruned_loss=0.08883, over 5693458.34 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3734, pruned_loss=0.1203, over 5657459.06 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:03:34,594 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:03:39,107 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.154e+03 1.718e+03 2.154e+03 3.256e+03 8.895e+03, threshold=4.307e+03, percent-clipped=9.0
+2023-03-14 21:03:46,029 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:03:57,310 INFO [train.py:968] (0/2) Epoch 29, batch 10150, giga_loss[loss=0.2914, simple_loss=0.361, pruned_loss=0.1109, over 28710.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3688, pruned_loss=0.1172, over 5672280.05 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.342, pruned_loss=0.08865, over 5699416.33 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3724, pruned_loss=0.1206, over 5656561.44 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:04:25,484 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1286000.pt
+2023-03-14 21:04:49,861 INFO [train.py:968] (0/2) Epoch 29, batch 10200, giga_loss[loss=0.2743, simple_loss=0.341, pruned_loss=0.1037, over 28783.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3669, pruned_loss=0.116, over 5676279.41 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3422, pruned_loss=0.08879, over 5695311.13 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3701, pruned_loss=0.1191, over 5666558.41 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:04:50,213 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:04:53,152 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:05:02,420 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-14 21:05:19,956 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:05:20,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.795e+02 1.721e+03 2.161e+03 2.949e+03 6.845e+03, threshold=4.322e+03, percent-clipped=10.0
+2023-03-14 21:05:24,514 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2347, 1.5275, 1.5715, 1.3808], device='cuda:0'), covar=tensor([0.2320, 0.1932, 0.2664, 0.2072], device='cuda:0'), in_proj_covar=tensor([0.0511, 0.0762, 0.0736, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 21:05:26,864 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-14 21:05:38,594 INFO [train.py:968] (0/2) Epoch 29, batch 10250, giga_loss[loss=0.2826, simple_loss=0.3581, pruned_loss=0.1036, over 28944.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3667, pruned_loss=0.1173, over 5669229.33 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3422, pruned_loss=0.08879, over 5699622.67 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3697, pruned_loss=0.1203, over 5657357.61 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:05:54,979 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2768, 1.1992, 3.7254, 3.2760], device='cuda:0'), covar=tensor([0.1676, 0.2901, 0.0492, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0680, 0.1020, 0.0997], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 21:06:04,193 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9145, 1.2355, 1.3046, 1.0620], device='cuda:0'), covar=tensor([0.2054, 0.1384, 0.2329, 0.1701], device='cuda:0'), in_proj_covar=tensor([0.0512, 0.0763, 0.0737, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 21:06:23,048 INFO [train.py:968] (0/2) Epoch 29, batch 10300, giga_loss[loss=0.2614, simple_loss=0.3421, pruned_loss=0.0903, over 28518.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3653, pruned_loss=0.1164, over 5668531.87 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3425, pruned_loss=0.08892, over 5699098.51 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3677, pruned_loss=0.119, over 5659238.33 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:06:29,882 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1286126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:06:33,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5320, 1.7324, 1.2967, 1.3177], device='cuda:0'), covar=tensor([0.1072, 0.0648, 0.1020, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0455, 0.0526, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 21:06:53,410 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.135e+03 1.731e+03 2.240e+03 3.067e+03 8.925e+03, threshold=4.480e+03, percent-clipped=8.0
+2023-03-14 21:07:05,568 INFO [train.py:968] (0/2) Epoch 29, batch 10350, giga_loss[loss=0.2626, simple_loss=0.3442, pruned_loss=0.09051, over 28838.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3636, pruned_loss=0.1138, over 5646684.38 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08933, over 5681354.23 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.366, pruned_loss=0.1166, over 5654556.72 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:07:29,985 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6767, 1.9580, 1.4274, 1.5416], device='cuda:0'), covar=tensor([0.1181, 0.0715, 0.1107, 0.1232], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0456, 0.0527, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 21:07:53,346 INFO [train.py:968] (0/2) Epoch 29, batch 10400, giga_loss[loss=0.3233, simple_loss=0.3854, pruned_loss=0.1306, over 28895.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3614, pruned_loss=0.1118, over 5654442.50 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3432, pruned_loss=0.08954, over 5686796.37 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3636, pruned_loss=0.1144, over 5655477.03 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:07:57,859 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-14 21:08:01,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5138, 2.6036, 1.6058, 1.5950], device='cuda:0'), covar=tensor([0.0812, 0.0350, 0.0720, 0.1137], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0575, 0.0413, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 21:08:01,987 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1286229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:08:22,796 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1286252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:08:25,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.515e+02 1.598e+03 2.190e+03 3.023e+03 7.075e+03, threshold=4.380e+03, percent-clipped=7.0
+2023-03-14 21:08:40,434 INFO [train.py:968] (0/2) Epoch 29, batch 10450, giga_loss[loss=0.2624, simple_loss=0.3367, pruned_loss=0.094, over 28628.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.36, pruned_loss=0.1113, over 5663712.71 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08931, over 5689747.83 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3624, pruned_loss=0.114, over 5661355.53 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:08:45,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5045, 1.5820, 1.2184, 1.1638], device='cuda:0'), covar=tensor([0.0888, 0.0495, 0.0940, 0.1227], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0456, 0.0527, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 21:09:17,772 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7578, 2.2189, 2.1946, 1.5168], device='cuda:0'), covar=tensor([0.4663, 0.2967, 0.3049, 0.4109], device='cuda:0'), in_proj_covar=tensor([0.2078, 0.2054, 0.1953, 0.2106], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 21:09:29,981 INFO [train.py:968] (0/2) Epoch 29, batch 10500, giga_loss[loss=0.2997, simple_loss=0.3711, pruned_loss=0.1142, over 28949.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3577, pruned_loss=0.111, over 5655584.73 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08949, over 5685239.45 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3597, pruned_loss=0.1134, over 5657099.10 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:09:30,927 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1286322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:09:39,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1286332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:10:05,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.347e+03 2.176e+03 2.749e+03 3.487e+03 8.825e+03, threshold=5.498e+03, percent-clipped=16.0
+2023-03-14 21:10:12,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9464, 1.1714, 1.1765, 0.9462], device='cuda:0'), covar=tensor([0.2685, 0.2964, 0.1751, 0.2429], device='cuda:0'), in_proj_covar=tensor([0.2080, 0.2056, 0.1957, 0.2107], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 21:10:17,863 INFO [train.py:968] (0/2) Epoch 29, batch 10550, giga_loss[loss=0.3056, simple_loss=0.3739, pruned_loss=0.1186, over 28566.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3581, pruned_loss=0.1117, over 5665265.14 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08935, over 5687402.22 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3602, pruned_loss=0.114, over 5664029.32 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:10:18,880 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:10:20,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:10:43,571 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:10:59,069 INFO [train.py:968] (0/2) Epoch 29, batch 10600, giga_loss[loss=0.3181, simple_loss=0.3806, pruned_loss=0.1278, over 28921.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3616, pruned_loss=0.1125, over 5674921.67 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3433, pruned_loss=0.08948, over 5694480.33 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3633, pruned_loss=0.115, over 5667129.68 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:11:09,942 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 21:11:33,501 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.252e+03 1.702e+03 2.175e+03 3.448e+03 8.631e+03, threshold=4.351e+03, percent-clipped=7.0
+2023-03-14 21:11:41,253 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:11:43,150 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:11:44,879 INFO [train.py:968] (0/2) Epoch 29, batch 10650, giga_loss[loss=0.3317, simple_loss=0.4001, pruned_loss=0.1317, over 28838.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3634, pruned_loss=0.1133, over 5667318.62 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3435, pruned_loss=0.08935, over 5701331.37 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3653, pruned_loss=0.1163, over 5653670.36 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:11:49,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:11:52,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:12:10,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:12:13,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4287, 1.5021, 1.2110, 1.0789], device='cuda:0'), covar=tensor([0.1048, 0.0558, 0.1065, 0.1195], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0456, 0.0528, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 21:12:14,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1286501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:12:18,379 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:12:32,300 INFO [train.py:968] (0/2) Epoch 29, batch 10700, giga_loss[loss=0.2829, simple_loss=0.3502, pruned_loss=0.1078, over 28237.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3638, pruned_loss=0.1143, over 5646044.16 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3434, pruned_loss=0.0893, over 5705211.49 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3657, pruned_loss=0.1171, over 5630790.46 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:12:44,633 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5563, 1.7826, 1.8391, 1.5327], device='cuda:0'), covar=tensor([0.2438, 0.2470, 0.2589, 0.2475], device='cuda:0'), in_proj_covar=tensor([0.0510, 0.0759, 0.0733, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 21:13:04,682 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.575e+03 2.013e+03 2.795e+03 6.523e+03, threshold=4.025e+03, percent-clipped=6.0
+2023-03-14 21:13:19,304 INFO [train.py:968] (0/2) Epoch 29, batch 10750, giga_loss[loss=0.3052, simple_loss=0.3689, pruned_loss=0.1207, over 28751.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3644, pruned_loss=0.1154, over 5650448.30 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3433, pruned_loss=0.08916, over 5708117.63 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3663, pruned_loss=0.1181, over 5635321.55 frames. ], batch size: 99, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:14:13,049 INFO [train.py:968] (0/2) Epoch 29, batch 10800, giga_loss[loss=0.2752, simple_loss=0.3552, pruned_loss=0.09759, over 28988.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3657, pruned_loss=0.1162, over 5646758.73 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3431, pruned_loss=0.08897, over 5711129.96 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3677, pruned_loss=0.1189, over 5631474.38 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:14:19,791 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1286627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:14:33,717 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:14:37,119 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:14:45,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.352e+03 1.760e+03 2.352e+03 3.650e+03 9.997e+03, threshold=4.704e+03, percent-clipped=19.0
+2023-03-14 21:14:57,771 INFO [train.py:968] (0/2) Epoch 29, batch 10850, giga_loss[loss=0.2738, simple_loss=0.3483, pruned_loss=0.09971, over 28697.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3665, pruned_loss=0.1162, over 5655180.67 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3433, pruned_loss=0.08898, over 5717582.83 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3687, pruned_loss=0.1193, over 5634977.00 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:15:02,440 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:15:43,307 INFO [train.py:968] (0/2) Epoch 29, batch 10900, libri_loss[loss=0.2492, simple_loss=0.3287, pruned_loss=0.08485, over 29577.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3679, pruned_loss=0.1174, over 5663443.69 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3431, pruned_loss=0.08885, over 5723309.30 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3706, pruned_loss=0.1208, over 5640182.96 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:16:06,986 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.49 vs. limit=5.0
+2023-03-14 21:16:13,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1286755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:16:13,897 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.238e+03 1.873e+03 2.458e+03 3.151e+03 8.247e+03, threshold=4.915e+03, percent-clipped=6.0
+2023-03-14 21:16:27,371 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:16:27,726 INFO [train.py:968] (0/2) Epoch 29, batch 10950, libri_loss[loss=0.2439, simple_loss=0.3259, pruned_loss=0.08096, over 29551.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3708, pruned_loss=0.1199, over 5665871.86 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3435, pruned_loss=0.08911, over 5724451.29 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3734, pruned_loss=0.1234, over 5644115.89 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:16:29,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:16:39,672 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 21:16:49,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3557, 1.4065, 1.2287, 1.5212], device='cuda:0'), covar=tensor([0.0773, 0.0359, 0.0359, 0.0891], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 21:16:54,995 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:17:15,403 INFO [train.py:968] (0/2) Epoch 29, batch 11000, giga_loss[loss=0.3019, simple_loss=0.3549, pruned_loss=0.1245, over 23761.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3724, pruned_loss=0.1201, over 5665741.53 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3437, pruned_loss=0.08926, over 5727950.41 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3748, pruned_loss=0.1233, over 5644532.85 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:17:52,404 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.133e+03 1.826e+03 2.350e+03 3.343e+03 5.437e+03, threshold=4.701e+03, percent-clipped=3.0
+2023-03-14 21:18:05,901 INFO [train.py:968] (0/2) Epoch 29, batch 11050, giga_loss[loss=0.3099, simple_loss=0.3744, pruned_loss=0.1227, over 28642.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3713, pruned_loss=0.1191, over 5659176.73 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3433, pruned_loss=0.08905, over 5730575.55 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.374, pruned_loss=0.1222, over 5639138.32 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:18:55,202 INFO [train.py:968] (0/2) Epoch 29, batch 11100, libri_loss[loss=0.2505, simple_loss=0.3317, pruned_loss=0.08468, over 29544.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.369, pruned_loss=0.1179, over 5672347.79 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3436, pruned_loss=0.08922, over 5733056.54 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3716, pruned_loss=0.121, over 5652353.25 frames. ], batch size: 79, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:19:09,375 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5772, 1.7711, 1.6304, 1.4529], device='cuda:0'), covar=tensor([0.3671, 0.3058, 0.2927, 0.3254], device='cuda:0'), in_proj_covar=tensor([0.2083, 0.2063, 0.1960, 0.2109], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 21:19:28,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.324e+03 1.896e+03 2.533e+03 3.427e+03 8.467e+03, threshold=5.066e+03, percent-clipped=11.0
+2023-03-14 21:19:51,328 INFO [train.py:968] (0/2) Epoch 29, batch 11150, giga_loss[loss=0.4125, simple_loss=0.431, pruned_loss=0.197, over 26462.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3686, pruned_loss=0.1187, over 5668023.90 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3427, pruned_loss=0.0888, over 5737695.71 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3721, pruned_loss=0.1223, over 5646571.26 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:20:20,984 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1287001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:20:39,277 INFO [train.py:968] (0/2) Epoch 29, batch 11200, libri_loss[loss=0.272, simple_loss=0.3547, pruned_loss=0.09466, over 29652.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.367, pruned_loss=0.1174, over 5671154.46 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3429, pruned_loss=0.08892, over 5738755.94 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3702, pruned_loss=0.121, over 5650953.11 frames. ], batch size: 91, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:20:45,774 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-14 21:21:11,280 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.749e+03 2.234e+03 2.921e+03 8.867e+03, threshold=4.469e+03, percent-clipped=3.0
+2023-03-14 21:21:24,847 INFO [train.py:968] (0/2) Epoch 29, batch 11250, giga_loss[loss=0.2842, simple_loss=0.3501, pruned_loss=0.1091, over 28626.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3662, pruned_loss=0.1177, over 5677280.34 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3432, pruned_loss=0.08897, over 5740629.75 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3688, pruned_loss=0.1208, over 5658947.43 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:22:12,404 INFO [train.py:968] (0/2) Epoch 29, batch 11300, giga_loss[loss=0.2631, simple_loss=0.3429, pruned_loss=0.09167, over 28865.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3658, pruned_loss=0.1174, over 5672296.94 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.343, pruned_loss=0.08871, over 5741504.26 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3687, pruned_loss=0.1208, over 5655464.12 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:22:20,445 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1287130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:22:26,638 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4888, 1.5102, 1.6722, 1.3256], device='cuda:0'), covar=tensor([0.1571, 0.2438, 0.1312, 0.1572], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0721, 0.0986, 0.0884], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 21:22:26,731 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 21:22:27,085 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1287139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:22:44,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.786e+03 2.233e+03 3.043e+03 7.513e+03, threshold=4.466e+03, percent-clipped=7.0
+2023-03-14 21:22:59,187 INFO [train.py:968] (0/2) Epoch 29, batch 11350, giga_loss[loss=0.26, simple_loss=0.3363, pruned_loss=0.09188, over 28767.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3662, pruned_loss=0.1176, over 5660418.91 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3438, pruned_loss=0.08918, over 5735465.76 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3685, pruned_loss=0.121, over 5650111.29 frames. ], batch size: 99, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:23:37,935 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3431, 1.7055, 1.2729, 0.8756], device='cuda:0'), covar=tensor([0.4532, 0.2856, 0.2763, 0.5567], device='cuda:0'), in_proj_covar=tensor([0.1856, 0.1747, 0.1669, 0.1511], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 21:23:43,592 INFO [train.py:968] (0/2) Epoch 29, batch 11400, libri_loss[loss=0.2461, simple_loss=0.3352, pruned_loss=0.07855, over 29511.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.367, pruned_loss=0.1181, over 5665454.50 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3438, pruned_loss=0.0891, over 5730868.03 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3698, pruned_loss=0.1221, over 5657825.37 frames. ], batch size: 82, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:24:13,001 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.280e+03 1.717e+03 2.108e+03 2.815e+03 7.182e+03, threshold=4.217e+03, percent-clipped=3.0
+2023-03-14 21:24:27,610 INFO [train.py:968] (0/2) Epoch 29, batch 11450, giga_loss[loss=0.2627, simple_loss=0.3424, pruned_loss=0.09152, over 28911.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3685, pruned_loss=0.119, over 5675221.35 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3435, pruned_loss=0.08885, over 5737142.61 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3718, pruned_loss=0.1233, over 5661710.50 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:24:29,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1287273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:24:32,551 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1287276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:24:55,125 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-14 21:25:00,234 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1287305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:25:15,407 INFO [train.py:968] (0/2) Epoch 29, batch 11500, giga_loss[loss=0.2795, simple_loss=0.3449, pruned_loss=0.1071, over 28998.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3695, pruned_loss=0.1206, over 5666542.61 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3432, pruned_loss=0.08868, over 5740692.70 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3729, pruned_loss=0.1247, over 5651507.46 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:25:28,540 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6469, 1.8475, 1.6865, 1.5196], device='cuda:0'), covar=tensor([0.3095, 0.2721, 0.2730, 0.3008], device='cuda:0'), in_proj_covar=tensor([0.2082, 0.2062, 0.1958, 0.2109], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 21:25:31,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7078, 4.5711, 4.3165, 2.0212], device='cuda:0'), covar=tensor([0.0607, 0.0739, 0.0823, 0.2000], device='cuda:0'), in_proj_covar=tensor([0.1332, 0.1227, 0.1031, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 21:25:51,787 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.130e+03 1.822e+03 2.422e+03 3.140e+03 7.718e+03, threshold=4.845e+03, percent-clipped=11.0
+2023-03-14 21:26:05,309 INFO [train.py:968] (0/2) Epoch 29, batch 11550, giga_loss[loss=0.2827, simple_loss=0.3609, pruned_loss=0.1023, over 28932.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1224, over 5658957.58 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3433, pruned_loss=0.08873, over 5737538.23 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3733, pruned_loss=0.1257, over 5649456.51 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:26:10,881 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1287376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:26:33,157 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 21:26:52,952 INFO [train.py:968] (0/2) Epoch 29, batch 11600, giga_loss[loss=0.2467, simple_loss=0.3191, pruned_loss=0.08715, over 28651.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3693, pruned_loss=0.1211, over 5667054.13 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3431, pruned_loss=0.08856, over 5740039.83 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3719, pruned_loss=0.1244, over 5656366.48 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:27:23,560 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.214e+03 1.657e+03 2.023e+03 2.728e+03 5.665e+03, threshold=4.045e+03, percent-clipped=2.0
+2023-03-14 21:27:24,451 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-14 21:27:36,940 INFO [train.py:968] (0/2) Epoch 29, batch 11650, giga_loss[loss=0.2823, simple_loss=0.358, pruned_loss=0.1033, over 29101.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3683, pruned_loss=0.1191, over 5662482.90 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3441, pruned_loss=0.0891, over 5734142.99 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3708, pruned_loss=0.1228, over 5656439.22 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:28:17,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1287514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:28:22,911 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1287519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:28:24,010 INFO [train.py:968] (0/2) Epoch 29, batch 11700, giga_loss[loss=0.2871, simple_loss=0.3585, pruned_loss=0.1078, over 29007.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3688, pruned_loss=0.1192, over 5666791.12 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3438, pruned_loss=0.08895, over 5737327.72 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3715, pruned_loss=0.1228, over 5657805.88 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:28:25,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1287522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:28:50,717 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1287551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:28:50,896 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 21:28:58,355 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 2.032e+03 2.651e+03 3.374e+03 7.312e+03, threshold=5.301e+03, percent-clipped=15.0
+2023-03-14 21:29:08,474 INFO [train.py:968] (0/2) Epoch 29, batch 11750, giga_loss[loss=0.3179, simple_loss=0.3854, pruned_loss=0.1252, over 28687.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3698, pruned_loss=0.1193, over 5683174.89 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3438, pruned_loss=0.0889, over 5739096.22 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.373, pruned_loss=0.1235, over 5671968.84 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:29:58,743 INFO [train.py:968] (0/2) Epoch 29, batch 11800, giga_loss[loss=0.3353, simple_loss=0.3849, pruned_loss=0.1429, over 28669.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3718, pruned_loss=0.1215, over 5682014.24 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3433, pruned_loss=0.0886, over 5744065.73 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3756, pruned_loss=0.1261, over 5666847.22 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:30:32,828 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1287657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:30:33,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.736e+03 2.446e+03 3.600e+03 1.025e+04, threshold=4.893e+03, percent-clipped=6.0
+2023-03-14 21:30:36,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1287660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:30:45,643 INFO [train.py:968] (0/2) Epoch 29, batch 11850, libri_loss[loss=0.2474, simple_loss=0.322, pruned_loss=0.08645, over 29655.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3707, pruned_loss=0.1205, over 5691410.65 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3433, pruned_loss=0.0886, over 5746592.08 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3741, pruned_loss=0.1246, over 5676354.39 frames. ], batch size: 73, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:31:01,843 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1287689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:31:33,808 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4324, 3.5569, 1.4100, 1.6621], device='cuda:0'), covar=tensor([0.1012, 0.0286, 0.0975, 0.1314], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0577, 0.0414, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 21:31:34,209 INFO [train.py:968] (0/2) Epoch 29, batch 11900, libri_loss[loss=0.2693, simple_loss=0.353, pruned_loss=0.0928, over 29529.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3719, pruned_loss=0.1206, over 5678883.49 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3431, pruned_loss=0.08849, over 5748157.30 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.375, pruned_loss=0.1242, over 5665041.52 frames. ], batch size: 89, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:31:56,975 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3593, 1.3711, 3.5051, 3.2087], device='cuda:0'), covar=tensor([0.1565, 0.2664, 0.0529, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0677, 0.1016, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 21:32:11,230 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.695e+03 2.091e+03 3.332e+03 9.722e+03, threshold=4.182e+03, percent-clipped=11.0
+2023-03-14 21:32:22,520 INFO [train.py:968] (0/2) Epoch 29, batch 11950, giga_loss[loss=0.3109, simple_loss=0.3784, pruned_loss=0.1217, over 28737.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3711, pruned_loss=0.1196, over 5675076.80 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.343, pruned_loss=0.08842, over 5749753.64 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3739, pruned_loss=0.1228, over 5662194.63 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:32:23,608 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7918, 1.6924, 1.9779, 1.5628], device='cuda:0'), covar=tensor([0.1843, 0.2552, 0.1462, 0.1784], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0721, 0.0986, 0.0885], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 21:32:55,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1287804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:33:08,933 INFO [train.py:968] (0/2) Epoch 29, batch 12000, giga_loss[loss=0.3123, simple_loss=0.3789, pruned_loss=0.1229, over 29067.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3693, pruned_loss=0.1186, over 5688567.43 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.343, pruned_loss=0.08851, over 5753298.66 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3722, pruned_loss=0.1218, over 5673491.94 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:33:08,938 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 21:33:17,848 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.2048, simple_loss=0.3128, pruned_loss=0.04841, over 944034.00 frames. 
+2023-03-14 21:33:17,848 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 21:33:48,477 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.716e+03 2.179e+03 3.211e+03 7.073e+03, threshold=4.359e+03, percent-clipped=10.0
+2023-03-14 21:33:53,962 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 21:34:01,788 INFO [train.py:968] (0/2) Epoch 29, batch 12050, giga_loss[loss=0.3728, simple_loss=0.4212, pruned_loss=0.1622, over 28589.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3701, pruned_loss=0.1196, over 5665945.62 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3436, pruned_loss=0.08888, over 5742342.74 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3726, pruned_loss=0.1226, over 5660917.87 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:34:50,227 INFO [train.py:968] (0/2) Epoch 29, batch 12100, giga_loss[loss=0.2851, simple_loss=0.3552, pruned_loss=0.1075, over 28580.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3715, pruned_loss=0.1205, over 5671776.25 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3433, pruned_loss=0.08875, over 5744191.36 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3742, pruned_loss=0.1234, over 5665221.93 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:34:56,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2832, 1.6847, 1.5027, 1.5073], device='cuda:0'), covar=tensor([0.2036, 0.1718, 0.2182, 0.1780], device='cuda:0'), in_proj_covar=tensor([0.0513, 0.0764, 0.0736, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-14 21:35:11,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4755, 1.5823, 1.1891, 1.1886], device='cuda:0'), covar=tensor([0.0871, 0.0446, 0.0908, 0.1181], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0456, 0.0528, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 21:35:26,338 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.109e+03 1.701e+03 2.392e+03 3.353e+03 8.779e+03, threshold=4.785e+03, percent-clipped=10.0
+2023-03-14 21:35:36,118 INFO [train.py:968] (0/2) Epoch 29, batch 12150, giga_loss[loss=0.2581, simple_loss=0.3299, pruned_loss=0.09311, over 28952.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3715, pruned_loss=0.1205, over 5674967.51 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.0888, over 5747503.17 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3745, pruned_loss=0.1241, over 5663830.61 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:36:02,858 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1288000.pt
+2023-03-14 21:36:13,033 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 21:36:21,253 INFO [train.py:968] (0/2) Epoch 29, batch 12200, giga_loss[loss=0.3088, simple_loss=0.3792, pruned_loss=0.1192, over 28737.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3717, pruned_loss=0.121, over 5675516.91 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3438, pruned_loss=0.08882, over 5750748.74 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3747, pruned_loss=0.1249, over 5661187.32 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:36:21,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5603, 1.8168, 1.4690, 1.4811], device='cuda:0'), covar=tensor([0.2632, 0.2725, 0.3126, 0.2344], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1163, 0.1426, 0.1020], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 21:36:24,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3908, 2.9505, 1.5114, 1.4624], device='cuda:0'), covar=tensor([0.0889, 0.0400, 0.0864, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0578, 0.0414, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 21:36:27,073 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5115, 1.7888, 1.4388, 1.6765], device='cuda:0'), covar=tensor([0.2440, 0.2499, 0.2696, 0.2185], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1164, 0.1427, 0.1021], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 21:36:56,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.244e+03 1.758e+03 2.264e+03 2.887e+03 7.498e+03, threshold=4.528e+03, percent-clipped=5.0
+2023-03-14 21:36:58,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1837, 3.4637, 1.3502, 1.3615], device='cuda:0'), covar=tensor([0.1275, 0.0411, 0.1064, 0.1609], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0577, 0.0414, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 21:37:09,505 INFO [train.py:968] (0/2) Epoch 29, batch 12250, giga_loss[loss=0.3123, simple_loss=0.3768, pruned_loss=0.1239, over 28570.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3727, pruned_loss=0.1223, over 5670661.98 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.344, pruned_loss=0.08887, over 5751843.92 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3755, pruned_loss=0.1259, over 5656834.28 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:37:19,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3816, 1.9271, 1.3898, 0.7636], device='cuda:0'), covar=tensor([0.6353, 0.3314, 0.3657, 0.6932], device='cuda:0'), in_proj_covar=tensor([0.1859, 0.1752, 0.1671, 0.1515], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 21:37:36,393 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5219, 1.8838, 1.4520, 1.3390], device='cuda:0'), covar=tensor([0.2606, 0.2643, 0.3109, 0.2451], device='cuda:0'), in_proj_covar=tensor([0.1614, 0.1164, 0.1426, 0.1021], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 21:37:41,867 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5315, 1.8327, 1.4788, 1.4927], device='cuda:0'), covar=tensor([0.2714, 0.2723, 0.3110, 0.2403], device='cuda:0'), in_proj_covar=tensor([0.1613, 0.1163, 0.1425, 0.1020], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 21:37:57,423 INFO [train.py:968] (0/2) Epoch 29, batch 12300, giga_loss[loss=0.3207, simple_loss=0.3833, pruned_loss=0.1291, over 28620.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3733, pruned_loss=0.1226, over 5668195.18 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3441, pruned_loss=0.08885, over 5756748.75 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3763, pruned_loss=0.1265, over 5650497.49 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:38:32,754 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.190e+03 1.933e+03 2.521e+03 3.340e+03 1.274e+04, threshold=5.043e+03, percent-clipped=12.0
+2023-03-14 21:38:43,527 INFO [train.py:968] (0/2) Epoch 29, batch 12350, giga_loss[loss=0.3038, simple_loss=0.3705, pruned_loss=0.1185, over 28694.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3733, pruned_loss=0.1224, over 5667215.04 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3443, pruned_loss=0.08902, over 5758192.73 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3757, pruned_loss=0.1257, over 5651172.44 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:38:52,419 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1288179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:39:37,109 INFO [train.py:968] (0/2) Epoch 29, batch 12400, giga_loss[loss=0.3003, simple_loss=0.3661, pruned_loss=0.1173, over 28864.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3735, pruned_loss=0.123, over 5650184.69 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3442, pruned_loss=0.089, over 5759196.91 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3757, pruned_loss=0.1259, over 5635569.48 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:40:04,508 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1288250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:40:12,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 1.817e+03 2.325e+03 3.463e+03 1.062e+04, threshold=4.650e+03, percent-clipped=3.0
+2023-03-14 21:40:21,511 INFO [train.py:968] (0/2) Epoch 29, batch 12450, giga_loss[loss=0.3027, simple_loss=0.3631, pruned_loss=0.1212, over 28205.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3734, pruned_loss=0.1224, over 5642819.34 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3444, pruned_loss=0.08906, over 5750402.20 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.376, pruned_loss=0.1258, over 5635718.84 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:40:46,245 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1288302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:41:02,161 INFO [train.py:968] (0/2) Epoch 29, batch 12500, giga_loss[loss=0.2946, simple_loss=0.3592, pruned_loss=0.115, over 28870.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3716, pruned_loss=0.1206, over 5653598.80 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3444, pruned_loss=0.08934, over 5753331.09 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3751, pruned_loss=0.1246, over 5640343.95 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:41:03,152 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1288322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:41:05,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1288325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:41:36,890 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1288354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:41:42,325 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.156e+03 1.751e+03 2.395e+03 3.441e+03 1.112e+04, threshold=4.790e+03, percent-clipped=12.0
+2023-03-14 21:41:53,831 INFO [train.py:968] (0/2) Epoch 29, batch 12550, giga_loss[loss=0.3121, simple_loss=0.3668, pruned_loss=0.1287, over 28609.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3715, pruned_loss=0.1211, over 5659934.56 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3445, pruned_loss=0.08942, over 5754654.16 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3744, pruned_loss=0.1244, over 5647648.37 frames. ], batch size: 85, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:42:01,325 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1288379.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 21:42:41,107 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.6150, 5.4616, 5.1760, 2.7702], device='cuda:0'), covar=tensor([0.0477, 0.0635, 0.0721, 0.1588], device='cuda:0'), in_proj_covar=tensor([0.1333, 0.1232, 0.1035, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-14 21:42:43,595 INFO [train.py:968] (0/2) Epoch 29, batch 12600, giga_loss[loss=0.3088, simple_loss=0.3763, pruned_loss=0.1207, over 28969.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3689, pruned_loss=0.1196, over 5667541.33 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3442, pruned_loss=0.08926, over 5757130.60 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3718, pruned_loss=0.1228, over 5654327.04 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:43:16,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.187e+03 1.720e+03 2.269e+03 3.112e+03 1.181e+04, threshold=4.537e+03, percent-clipped=8.0
+2023-03-14 21:43:27,088 INFO [train.py:968] (0/2) Epoch 29, batch 12650, giga_loss[loss=0.2774, simple_loss=0.3355, pruned_loss=0.1096, over 28689.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3648, pruned_loss=0.1174, over 5674686.60 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3435, pruned_loss=0.08902, over 5759754.07 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3682, pruned_loss=0.1208, over 5660042.85 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:44:10,148 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5424, 1.8392, 1.5997, 1.6428], device='cuda:0'), covar=tensor([0.1965, 0.2202, 0.2238, 0.2168], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0766, 0.0737, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-14 21:44:13,752 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0000, 1.1291, 1.1270, 0.9808], device='cuda:0'), covar=tensor([0.2327, 0.2853, 0.1635, 0.2178], device='cuda:0'), in_proj_covar=tensor([0.2091, 0.2067, 0.1961, 0.2111], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 21:44:15,584 INFO [train.py:968] (0/2) Epoch 29, batch 12700, giga_loss[loss=0.3481, simple_loss=0.3945, pruned_loss=0.1508, over 28590.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3622, pruned_loss=0.117, over 5635462.40 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3442, pruned_loss=0.08961, over 5738698.90 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3646, pruned_loss=0.1196, over 5640594.24 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:44:52,576 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.184e+03 1.949e+03 2.497e+03 3.552e+03 1.390e+04, threshold=4.994e+03, percent-clipped=15.0
+2023-03-14 21:45:00,129 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2946, 1.1756, 1.1364, 1.5356], device='cuda:0'), covar=tensor([0.0747, 0.0419, 0.0361, 0.0811], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 21:45:01,903 INFO [train.py:968] (0/2) Epoch 29, batch 12750, giga_loss[loss=0.2577, simple_loss=0.3264, pruned_loss=0.0945, over 29131.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3604, pruned_loss=0.1163, over 5640289.41 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3442, pruned_loss=0.08959, over 5741270.37 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3625, pruned_loss=0.1187, over 5640746.07 frames. ], batch size: 113, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:45:22,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5820, 1.9387, 1.5419, 1.5742], device='cuda:0'), covar=tensor([0.2610, 0.2714, 0.3136, 0.2495], device='cuda:0'), in_proj_covar=tensor([0.1617, 0.1164, 0.1430, 0.1021], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 21:45:32,926 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-14 21:45:50,267 INFO [train.py:968] (0/2) Epoch 29, batch 12800, giga_loss[loss=0.341, simple_loss=0.3793, pruned_loss=0.1513, over 24019.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3599, pruned_loss=0.1158, over 5635370.56 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3441, pruned_loss=0.08948, over 5735487.12 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.362, pruned_loss=0.1185, over 5639624.67 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:45:54,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1288625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:46:27,405 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.171e+03 1.711e+03 2.402e+03 3.987e+03 1.023e+04, threshold=4.805e+03, percent-clipped=14.0
+2023-03-14 21:46:35,317 INFO [train.py:968] (0/2) Epoch 29, batch 12850, giga_loss[loss=0.256, simple_loss=0.3392, pruned_loss=0.08642, over 28928.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3581, pruned_loss=0.1122, over 5641926.79 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3439, pruned_loss=0.08935, over 5732011.23 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3605, pruned_loss=0.1152, over 5645135.35 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:46:41,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1288677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:46:51,677 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-14 21:47:27,017 INFO [train.py:968] (0/2) Epoch 29, batch 12900, giga_loss[loss=0.2629, simple_loss=0.3476, pruned_loss=0.08907, over 28813.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3558, pruned_loss=0.1088, over 5639239.36 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3438, pruned_loss=0.08935, over 5735563.46 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3581, pruned_loss=0.1116, over 5636986.30 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:47:31,368 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3520, 1.7807, 1.6449, 1.5631], device='cuda:0'), covar=tensor([0.2133, 0.2039, 0.1879, 0.1958], device='cuda:0'), in_proj_covar=tensor([0.0512, 0.0762, 0.0733, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 21:48:01,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1288754.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 21:48:07,996 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.099e+03 1.576e+03 2.030e+03 2.690e+03 4.851e+03, threshold=4.060e+03, percent-clipped=1.0
+2023-03-14 21:48:17,355 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1288768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:48:21,071 INFO [train.py:968] (0/2) Epoch 29, batch 12950, giga_loss[loss=0.215, simple_loss=0.2875, pruned_loss=0.0712, over 23976.00 frames. ], tot_loss[loss=0.282, simple_loss=0.353, pruned_loss=0.1055, over 5637387.08 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3438, pruned_loss=0.08943, over 5736557.64 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3547, pruned_loss=0.1077, over 5634361.67 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:48:21,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1288771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:48:51,024 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1288800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:49:10,378 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1288820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:49:10,794 INFO [train.py:968] (0/2) Epoch 29, batch 13000, giga_loss[loss=0.2622, simple_loss=0.3412, pruned_loss=0.09156, over 28389.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3508, pruned_loss=0.1035, over 5635571.06 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3436, pruned_loss=0.08952, over 5731563.77 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3526, pruned_loss=0.1055, over 5635592.20 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:49:12,832 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1288823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:49:40,426 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1288852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:49:49,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.726e+02 1.362e+03 1.908e+03 2.581e+03 1.003e+04, threshold=3.816e+03, percent-clipped=8.0
+2023-03-14 21:49:53,494 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5064, 1.6892, 1.2171, 1.2638], device='cuda:0'), covar=tensor([0.0980, 0.0497, 0.0956, 0.1221], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0451, 0.0523, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 21:50:01,432 INFO [train.py:968] (0/2) Epoch 29, batch 13050, giga_loss[loss=0.2473, simple_loss=0.3433, pruned_loss=0.07565, over 29033.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3489, pruned_loss=0.1003, over 5640398.94 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3432, pruned_loss=0.08931, over 5734034.84 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3508, pruned_loss=0.1022, over 5637170.15 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:50:14,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4244, 1.9259, 1.8485, 1.6255], device='cuda:0'), covar=tensor([0.2494, 0.2487, 0.2075, 0.2390], device='cuda:0'), in_proj_covar=tensor([0.0511, 0.0759, 0.0730, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 21:50:25,970 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1288897.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 21:50:28,402 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1288900.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 21:50:47,635 INFO [train.py:968] (0/2) Epoch 29, batch 13100, giga_loss[loss=0.2864, simple_loss=0.3412, pruned_loss=0.1158, over 24091.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3492, pruned_loss=0.09882, over 5656012.70 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08942, over 5738536.52 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.351, pruned_loss=0.1006, over 5646777.35 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:50:56,566 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1288929.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 21:51:30,293 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.581e+02 1.366e+03 1.813e+03 2.472e+03 4.035e+03, threshold=3.627e+03, percent-clipped=2.0
+2023-03-14 21:51:39,939 INFO [train.py:968] (0/2) Epoch 29, batch 13150, giga_loss[loss=0.2362, simple_loss=0.3207, pruned_loss=0.0758, over 28740.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3496, pruned_loss=0.09922, over 5650090.49 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3427, pruned_loss=0.08932, over 5739716.52 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3514, pruned_loss=0.1008, over 5641000.54 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:52:14,319 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:52:31,999 INFO [train.py:968] (0/2) Epoch 29, batch 13200, giga_loss[loss=0.2391, simple_loss=0.3264, pruned_loss=0.07586, over 28872.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3467, pruned_loss=0.09731, over 5647039.65 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3424, pruned_loss=0.08918, over 5738667.94 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3484, pruned_loss=0.09869, over 5640412.58 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:53:12,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.994e+02 1.422e+03 1.896e+03 2.905e+03 6.052e+03, threshold=3.792e+03, percent-clipped=11.0
+2023-03-14 21:53:22,273 INFO [train.py:968] (0/2) Epoch 29, batch 13250, giga_loss[loss=0.2795, simple_loss=0.3481, pruned_loss=0.1054, over 28314.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3451, pruned_loss=0.09671, over 5646489.81 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08915, over 5741758.34 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3467, pruned_loss=0.09804, over 5636310.07 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:54:13,734 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:54:14,115 INFO [train.py:968] (0/2) Epoch 29, batch 13300, giga_loss[loss=0.2391, simple_loss=0.3274, pruned_loss=0.07542, over 28817.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3451, pruned_loss=0.09642, over 5641277.36 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08915, over 5741758.34 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3465, pruned_loss=0.09745, over 5633354.33 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:54:54,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.029e+03 1.590e+03 2.207e+03 3.139e+03 6.653e+03, threshold=4.413e+03, percent-clipped=19.0
+2023-03-14 21:55:03,307 INFO [train.py:968] (0/2) Epoch 29, batch 13350, giga_loss[loss=0.2713, simple_loss=0.3498, pruned_loss=0.09638, over 29025.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3448, pruned_loss=0.09617, over 5633533.64 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3423, pruned_loss=0.08948, over 5724959.47 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3457, pruned_loss=0.09683, over 5639763.47 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:55:51,767 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:55:55,734 INFO [train.py:968] (0/2) Epoch 29, batch 13400, giga_loss[loss=0.2299, simple_loss=0.3141, pruned_loss=0.07281, over 28652.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3422, pruned_loss=0.09394, over 5638034.37 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3424, pruned_loss=0.08964, over 5727502.65 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3429, pruned_loss=0.0944, over 5639808.31 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:55:57,349 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:56:21,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5237, 1.7660, 1.2418, 1.3104], device='cuda:0'), covar=tensor([0.1066, 0.0517, 0.0980, 0.1257], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0451, 0.0522, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 21:56:27,708 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7553, 1.9411, 1.9699, 1.5114], device='cuda:0'), covar=tensor([0.1999, 0.2854, 0.1699, 0.2138], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0717, 0.0983, 0.0884], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 21:56:37,455 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.054e+02 1.551e+03 1.981e+03 2.814e+03 6.771e+03, threshold=3.961e+03, percent-clipped=6.0
+2023-03-14 21:56:44,437 INFO [train.py:968] (0/2) Epoch 29, batch 13450, libri_loss[loss=0.2544, simple_loss=0.3405, pruned_loss=0.08409, over 28607.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3383, pruned_loss=0.09131, over 5641911.11 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3419, pruned_loss=0.08954, over 5731257.95 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3392, pruned_loss=0.09185, over 5637522.46 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:56:44,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3863, 1.3273, 3.7063, 3.2649], device='cuda:0'), covar=tensor([0.1520, 0.2782, 0.0459, 0.1152], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0675, 0.1011, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-14 21:57:34,166 INFO [train.py:968] (0/2) Epoch 29, batch 13500, giga_loss[loss=0.2575, simple_loss=0.3384, pruned_loss=0.08833, over 28926.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3361, pruned_loss=0.09061, over 5648367.58 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3416, pruned_loss=0.08951, over 5728027.84 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3369, pruned_loss=0.09114, over 5644633.45 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:58:07,767 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:58:14,840 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.586e+03 2.127e+03 2.755e+03 6.783e+03, threshold=4.255e+03, percent-clipped=9.0
+2023-03-14 21:58:18,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6161, 1.8276, 1.8170, 1.5931], device='cuda:0'), covar=tensor([0.2804, 0.2186, 0.1797, 0.2318], device='cuda:0'), in_proj_covar=tensor([0.2043, 0.2013, 0.1909, 0.2055], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 21:58:24,231 INFO [train.py:968] (0/2) Epoch 29, batch 13550, giga_loss[loss=0.2262, simple_loss=0.3083, pruned_loss=0.07206, over 28725.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3348, pruned_loss=0.09029, over 5649868.50 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3415, pruned_loss=0.08946, over 5730790.64 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3355, pruned_loss=0.09077, over 5643141.79 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:58:33,125 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:58:50,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6570, 2.3170, 1.8766, 1.9334], device='cuda:0'), covar=tensor([0.0714, 0.0241, 0.0296, 0.0804], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-14 21:59:02,091 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3922, 1.8176, 1.4794, 1.5549], device='cuda:0'), covar=tensor([0.0784, 0.0336, 0.0361, 0.0901], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 21:59:14,398 INFO [train.py:968] (0/2) Epoch 29, batch 13600, giga_loss[loss=0.2924, simple_loss=0.3627, pruned_loss=0.111, over 28669.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3355, pruned_loss=0.09096, over 5651766.81 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3407, pruned_loss=0.08933, over 5737154.87 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3365, pruned_loss=0.0915, over 5637011.61 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:59:14,587 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:59:59,921 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.862e+02 1.565e+03 2.115e+03 3.438e+03 1.253e+04, threshold=4.231e+03, percent-clipped=19.0
+2023-03-14 22:00:10,107 INFO [train.py:968] (0/2) Epoch 29, batch 13650, giga_loss[loss=0.2491, simple_loss=0.334, pruned_loss=0.08214, over 27904.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3374, pruned_loss=0.09083, over 5652358.49 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3405, pruned_loss=0.08928, over 5740380.03 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3383, pruned_loss=0.09134, over 5636456.55 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:00:38,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:00:39,788 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:01:06,548 INFO [train.py:968] (0/2) Epoch 29, batch 13700, giga_loss[loss=0.2579, simple_loss=0.3347, pruned_loss=0.0906, over 28944.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.339, pruned_loss=0.09107, over 5669113.12 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.34, pruned_loss=0.08927, over 5745425.13 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3401, pruned_loss=0.09153, over 5649342.95 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:01:08,790 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:01:10,817 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:01:44,854 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:01:54,592 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.573e+03 1.766e+03 2.293e+03 4.243e+03, threshold=3.532e+03, percent-clipped=1.0
+2023-03-14 22:02:02,176 INFO [train.py:968] (0/2) Epoch 29, batch 13750, giga_loss[loss=0.2469, simple_loss=0.3256, pruned_loss=0.08406, over 28879.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3395, pruned_loss=0.09133, over 5673794.01 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3395, pruned_loss=0.08902, over 5742085.60 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3409, pruned_loss=0.09201, over 5658956.65 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:02:12,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2683, 1.4993, 1.4472, 1.3185], device='cuda:0'), covar=tensor([0.2306, 0.1904, 0.1552, 0.1917], device='cuda:0'), in_proj_covar=tensor([0.2050, 0.2019, 0.1912, 0.2061], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 22:02:18,391 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-14 22:02:25,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:02:31,744 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:02:46,017 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-14 22:02:59,023 INFO [train.py:968] (0/2) Epoch 29, batch 13800, giga_loss[loss=0.3027, simple_loss=0.3847, pruned_loss=0.1104, over 28631.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3369, pruned_loss=0.08979, over 5672804.58 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3391, pruned_loss=0.0889, over 5745756.10 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3383, pruned_loss=0.09048, over 5656289.25 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:03:22,013 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:03:26,645 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:03:43,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3609, 1.2620, 1.2091, 1.6274], device='cuda:0'), covar=tensor([0.0760, 0.0364, 0.0367, 0.0880], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 22:03:49,191 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.425e+02 1.565e+03 1.896e+03 2.676e+03 7.554e+03, threshold=3.792e+03, percent-clipped=12.0
+2023-03-14 22:03:59,261 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:03:59,593 INFO [train.py:968] (0/2) Epoch 29, batch 13850, giga_loss[loss=0.2554, simple_loss=0.342, pruned_loss=0.08444, over 28650.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3358, pruned_loss=0.08805, over 5673185.54 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3384, pruned_loss=0.08866, over 5749217.29 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3375, pruned_loss=0.08881, over 5655061.74 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:04:11,930 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1626, 1.2494, 3.4054, 3.1039], device='cuda:0'), covar=tensor([0.1688, 0.2810, 0.0567, 0.1053], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0672, 0.1003, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 22:04:59,339 INFO [train.py:968] (0/2) Epoch 29, batch 13900, giga_loss[loss=0.2595, simple_loss=0.3354, pruned_loss=0.09179, over 28892.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.335, pruned_loss=0.08793, over 5669418.65 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3384, pruned_loss=0.08864, over 5749544.97 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3363, pruned_loss=0.08854, over 5653372.17 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:05:10,310 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:16,367 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:19,824 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:20,624 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289738.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:05:25,094 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:28,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:50,436 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.686e+02 1.488e+03 1.979e+03 2.944e+03 1.069e+04, threshold=3.958e+03, percent-clipped=13.0
+2023-03-14 22:05:55,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:59,084 INFO [train.py:968] (0/2) Epoch 29, batch 13950, giga_loss[loss=0.2585, simple_loss=0.3358, pruned_loss=0.09061, over 29002.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.332, pruned_loss=0.08709, over 5673364.39 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3381, pruned_loss=0.08865, over 5751440.73 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3332, pruned_loss=0.08752, over 5657915.88 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:06:01,634 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:06:15,029 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5337, 1.9824, 1.8451, 1.6947], device='cuda:0'), covar=tensor([0.2234, 0.2428, 0.2095, 0.2387], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0753, 0.0724, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 22:06:27,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:06:45,895 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-14 22:06:53,249 INFO [train.py:968] (0/2) Epoch 29, batch 14000, giga_loss[loss=0.2198, simple_loss=0.3022, pruned_loss=0.06868, over 28994.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3311, pruned_loss=0.08705, over 5680255.40 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.338, pruned_loss=0.08867, over 5755075.17 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3319, pruned_loss=0.08732, over 5662007.98 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 22:07:40,190 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.986e+02 1.457e+03 1.744e+03 2.650e+03 5.209e+03, threshold=3.487e+03, percent-clipped=5.0
+2023-03-14 22:07:50,131 INFO [train.py:968] (0/2) Epoch 29, batch 14050, giga_loss[loss=0.274, simple_loss=0.3572, pruned_loss=0.09542, over 28431.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3312, pruned_loss=0.08663, over 5664030.10 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3377, pruned_loss=0.08862, over 5753677.91 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.332, pruned_loss=0.08687, over 5649891.45 frames. ], batch size: 369, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:07:50,498 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:07:53,358 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:07:57,197 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:08:31,106 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:08:34,652 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-14 22:08:48,801 INFO [train.py:968] (0/2) Epoch 29, batch 14100, giga_loss[loss=0.2528, simple_loss=0.3358, pruned_loss=0.08485, over 29011.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3328, pruned_loss=0.08662, over 5668460.91 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.337, pruned_loss=0.08838, over 5757379.76 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.334, pruned_loss=0.08699, over 5651732.54 frames. ], batch size: 285, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:08:57,468 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5640, 1.7506, 1.2686, 1.3733], device='cuda:0'), covar=tensor([0.1028, 0.0529, 0.0981, 0.1119], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0449, 0.0522, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 22:08:57,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3881, 1.5720, 1.6241, 1.2389], device='cuda:0'), covar=tensor([0.1935, 0.2792, 0.1591, 0.1934], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0715, 0.0984, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 22:09:11,114 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:09:15,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:09:17,892 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-14 22:09:36,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.592e+02 1.447e+03 1.858e+03 2.496e+03 6.093e+03, threshold=3.717e+03, percent-clipped=6.0
+2023-03-14 22:09:47,319 INFO [train.py:968] (0/2) Epoch 29, batch 14150, giga_loss[loss=0.206, simple_loss=0.2788, pruned_loss=0.06657, over 24488.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3296, pruned_loss=0.08467, over 5678256.16 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3359, pruned_loss=0.08781, over 5763706.64 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3313, pruned_loss=0.08539, over 5655861.65 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:09:47,556 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:10:01,502 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9919, 1.2417, 2.8614, 2.8241], device='cuda:0'), covar=tensor([0.1560, 0.2600, 0.0562, 0.1114], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0677, 0.1008, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 22:10:26,737 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1290000.pt
+2023-03-14 22:10:44,345 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:10:46,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:10:52,211 INFO [train.py:968] (0/2) Epoch 29, batch 14200, giga_loss[loss=0.2954, simple_loss=0.3741, pruned_loss=0.1084, over 28913.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3309, pruned_loss=0.08547, over 5689253.91 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3355, pruned_loss=0.08759, over 5766460.11 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3326, pruned_loss=0.08618, over 5667500.15 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:11:03,837 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290030.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:11:23,154 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:11:28,858 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:11:48,917 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.573e+03 1.919e+03 2.815e+03 7.379e+03, threshold=3.839e+03, percent-clipped=6.0
+2023-03-14 22:11:58,010 INFO [train.py:968] (0/2) Epoch 29, batch 14250, giga_loss[loss=0.26, simple_loss=0.3524, pruned_loss=0.08383, over 28157.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.334, pruned_loss=0.0865, over 5690101.84 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3353, pruned_loss=0.08754, over 5769880.25 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3354, pruned_loss=0.08707, over 5668005.45 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:11:58,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6945, 1.9880, 1.3445, 1.6376], device='cuda:0'), covar=tensor([0.1109, 0.0651, 0.1039, 0.1171], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0450, 0.0523, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 22:12:14,679 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 22:12:51,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290113.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:13:01,556 INFO [train.py:968] (0/2) Epoch 29, batch 14300, giga_loss[loss=0.2348, simple_loss=0.3342, pruned_loss=0.06769, over 28938.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3377, pruned_loss=0.08614, over 5685963.43 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3351, pruned_loss=0.08752, over 5772654.58 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.339, pruned_loss=0.08659, over 5664563.84 frames. ], batch size: 120, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:13:13,956 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:13:23,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3197, 1.8838, 1.3539, 0.5570], device='cuda:0'), covar=tensor([0.5810, 0.3068, 0.4629, 0.7019], device='cuda:0'), in_proj_covar=tensor([0.1849, 0.1737, 0.1665, 0.1507], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 22:13:26,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3078, 2.0213, 1.6230, 1.6086], device='cuda:0'), covar=tensor([0.0809, 0.0284, 0.0327, 0.0901], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 22:13:55,257 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.421e+02 1.526e+03 1.971e+03 2.463e+03 4.695e+03, threshold=3.942e+03, percent-clipped=5.0
+2023-03-14 22:14:02,816 INFO [train.py:968] (0/2) Epoch 29, batch 14350, giga_loss[loss=0.2384, simple_loss=0.3299, pruned_loss=0.07345, over 28372.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3398, pruned_loss=0.08586, over 5679149.75 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3351, pruned_loss=0.08753, over 5773337.31 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3408, pruned_loss=0.08619, over 5661319.26 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:14:58,833 INFO [train.py:968] (0/2) Epoch 29, batch 14400, giga_loss[loss=0.2667, simple_loss=0.3452, pruned_loss=0.09413, over 28954.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3387, pruned_loss=0.08445, over 5685064.03 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3345, pruned_loss=0.08714, over 5775125.60 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3402, pruned_loss=0.08502, over 5666378.46 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 22:15:45,457 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290256.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:15:50,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290259.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:15:53,680 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.725e+03 2.339e+03 3.227e+03 1.341e+04, threshold=4.678e+03, percent-clipped=15.0
+2023-03-14 22:16:01,181 INFO [train.py:968] (0/2) Epoch 29, batch 14450, giga_loss[loss=0.2988, simple_loss=0.3731, pruned_loss=0.1122, over 29002.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3384, pruned_loss=0.08507, over 5683656.92 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3342, pruned_loss=0.08701, over 5777706.99 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.34, pruned_loss=0.0856, over 5664712.47 frames. ], batch size: 285, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 22:16:23,963 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290288.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:16:33,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9239, 1.1474, 2.8423, 2.6922], device='cuda:0'), covar=tensor([0.1577, 0.2599, 0.0580, 0.1607], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0673, 0.1003, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 22:16:36,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5880, 1.7625, 1.2582, 1.3809], device='cuda:0'), covar=tensor([0.0983, 0.0557, 0.0976, 0.1166], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0447, 0.0520, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 22:16:43,651 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-14 22:17:04,860 INFO [train.py:968] (0/2) Epoch 29, batch 14500, giga_loss[loss=0.3023, simple_loss=0.3694, pruned_loss=0.1176, over 28962.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3376, pruned_loss=0.08576, over 5674981.67 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3339, pruned_loss=0.08702, over 5761539.25 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3392, pruned_loss=0.08616, over 5672743.89 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:17:16,721 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:17:24,691 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3853, 3.4209, 1.4387, 1.6116], device='cuda:0'), covar=tensor([0.0989, 0.0312, 0.0926, 0.1333], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0572, 0.0413, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 22:17:34,652 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 22:18:02,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4300, 1.7409, 1.4143, 1.2883], device='cuda:0'), covar=tensor([0.2621, 0.2643, 0.3071, 0.2493], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1159, 0.1428, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 22:18:11,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.865e+02 1.433e+03 1.967e+03 2.584e+03 5.707e+03, threshold=3.935e+03, percent-clipped=1.0
+2023-03-14 22:18:12,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6827, 2.0157, 1.6009, 1.8646], device='cuda:0'), covar=tensor([0.2799, 0.2751, 0.3163, 0.2706], device='cuda:0'), in_proj_covar=tensor([0.1611, 0.1158, 0.1427, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 22:18:22,366 INFO [train.py:968] (0/2) Epoch 29, batch 14550, giga_loss[loss=0.2489, simple_loss=0.3347, pruned_loss=0.08157, over 28673.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3381, pruned_loss=0.08686, over 5680364.51 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3337, pruned_loss=0.08685, over 5763722.07 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3396, pruned_loss=0.0873, over 5675645.29 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:18:45,046 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:19:16,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290405.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:19:37,216 INFO [train.py:968] (0/2) Epoch 29, batch 14600, giga_loss[loss=0.2239, simple_loss=0.3126, pruned_loss=0.0676, over 28979.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.334, pruned_loss=0.08515, over 5681402.52 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3339, pruned_loss=0.08692, over 5767271.20 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3352, pruned_loss=0.08543, over 5672216.69 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:19:42,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:20:30,621 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.794e+02 1.324e+03 1.763e+03 2.185e+03 5.202e+03, threshold=3.527e+03, percent-clipped=1.0
+2023-03-14 22:20:32,083 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2752, 3.5814, 1.4230, 1.4362], device='cuda:0'), covar=tensor([0.1227, 0.0497, 0.1044, 0.1631], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0572, 0.0413, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 22:20:35,958 INFO [train.py:968] (0/2) Epoch 29, batch 14650, giga_loss[loss=0.2266, simple_loss=0.3196, pruned_loss=0.06682, over 29018.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3318, pruned_loss=0.08366, over 5684719.19 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3336, pruned_loss=0.08685, over 5767911.47 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3329, pruned_loss=0.08387, over 5674589.05 frames. ], batch size: 285, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:21:27,693 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:21:47,181 INFO [train.py:968] (0/2) Epoch 29, batch 14700, giga_loss[loss=0.2603, simple_loss=0.3436, pruned_loss=0.0885, over 28922.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3302, pruned_loss=0.08361, over 5678693.99 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3336, pruned_loss=0.08685, over 5767911.47 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3311, pruned_loss=0.08377, over 5670809.57 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:22:13,300 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 22:22:16,250 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290548.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:22:18,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290551.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:22:36,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-14 22:22:36,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.241e+02 1.567e+03 1.977e+03 2.963e+03 8.091e+03, threshold=3.954e+03, percent-clipped=16.0
+2023-03-14 22:22:41,591 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:22:42,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8024, 2.4608, 1.4592, 0.9706], device='cuda:0'), covar=tensor([0.8314, 0.4358, 0.4772, 0.7550], device='cuda:0'), in_proj_covar=tensor([0.1847, 0.1733, 0.1665, 0.1509], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 22:22:43,865 INFO [train.py:968] (0/2) Epoch 29, batch 14750, giga_loss[loss=0.3243, simple_loss=0.3876, pruned_loss=0.1305, over 28970.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3343, pruned_loss=0.086, over 5671359.86 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.333, pruned_loss=0.0866, over 5760989.99 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3354, pruned_loss=0.0863, over 5667743.76 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:22:44,110 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:22:54,091 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290580.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:23:00,564 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.57 vs. limit=5.0
+2023-03-14 22:23:16,186 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:23:37,838 INFO [train.py:968] (0/2) Epoch 29, batch 14800, giga_loss[loss=0.2541, simple_loss=0.3302, pruned_loss=0.08899, over 28963.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3348, pruned_loss=0.08707, over 5679957.51 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3326, pruned_loss=0.0864, over 5762512.66 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3362, pruned_loss=0.08748, over 5672544.59 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 22:24:11,492 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:24:15,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:24:29,253 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.028e+03 1.541e+03 2.030e+03 2.730e+03 5.906e+03, threshold=4.061e+03, percent-clipped=7.0
+2023-03-14 22:24:35,311 INFO [train.py:968] (0/2) Epoch 29, batch 14850, giga_loss[loss=0.2429, simple_loss=0.3255, pruned_loss=0.08018, over 28394.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3343, pruned_loss=0.08816, over 5675869.73 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3326, pruned_loss=0.08633, over 5758288.68 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3355, pruned_loss=0.08859, over 5671511.56 frames. ], batch size: 369, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:24:47,982 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:25:14,489 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:25:31,865 INFO [train.py:968] (0/2) Epoch 29, batch 14900, giga_loss[loss=0.251, simple_loss=0.3265, pruned_loss=0.08778, over 28917.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3344, pruned_loss=0.08887, over 5687239.32 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3324, pruned_loss=0.08638, over 5764777.76 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3357, pruned_loss=0.08929, over 5674779.96 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:26:22,713 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:26:28,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.583e+03 2.016e+03 3.151e+03 1.642e+04, threshold=4.033e+03, percent-clipped=14.0
+2023-03-14 22:26:35,264 INFO [train.py:968] (0/2) Epoch 29, batch 14950, giga_loss[loss=0.3023, simple_loss=0.3872, pruned_loss=0.1087, over 28511.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3362, pruned_loss=0.08945, over 5689904.25 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3323, pruned_loss=0.08638, over 5767636.74 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3373, pruned_loss=0.08983, over 5675716.37 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:26:50,872 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290784.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:27:09,081 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:27:29,926 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:27:41,301 INFO [train.py:968] (0/2) Epoch 29, batch 15000, giga_loss[loss=0.2591, simple_loss=0.3377, pruned_loss=0.09029, over 27628.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.337, pruned_loss=0.08893, over 5686213.63 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3315, pruned_loss=0.08592, over 5770393.29 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3387, pruned_loss=0.08974, over 5670590.18 frames. ], batch size: 474, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:27:41,307 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 22:27:50,053 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.1935, simple_loss=0.2949, pruned_loss=0.04606, over 944034.00 frames. 
+2023-03-14 22:27:50,053 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 22:27:55,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0562, 4.9222, 4.6539, 2.3062], device='cuda:0'), covar=tensor([0.0456, 0.0565, 0.0673, 0.1891], device='cuda:0'), in_proj_covar=tensor([0.1292, 0.1194, 0.1002, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 22:28:29,984 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:28:34,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:28:41,787 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3431, 1.7031, 1.3986, 0.9615], device='cuda:0'), covar=tensor([0.2749, 0.2741, 0.3119, 0.2507], device='cuda:0'), in_proj_covar=tensor([0.1616, 0.1161, 0.1430, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 22:28:44,944 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8607, 2.9450, 1.8845, 1.0686], device='cuda:0'), covar=tensor([0.9909, 0.3579, 0.4981, 0.8372], device='cuda:0'), in_proj_covar=tensor([0.1854, 0.1740, 0.1670, 0.1513], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 22:29:00,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.484e+02 1.516e+03 1.980e+03 2.916e+03 9.653e+03, threshold=3.960e+03, percent-clipped=10.0
+2023-03-14 22:29:11,088 INFO [train.py:968] (0/2) Epoch 29, batch 15050, giga_loss[loss=0.2683, simple_loss=0.341, pruned_loss=0.09784, over 28879.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.336, pruned_loss=0.08809, over 5671374.40 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3317, pruned_loss=0.08597, over 5769788.34 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3372, pruned_loss=0.0887, over 5658786.31 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:29:19,105 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:29:55,420 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:29:59,055 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4295, 1.4916, 3.6092, 3.2756], device='cuda:0'), covar=tensor([0.1455, 0.2593, 0.0497, 0.1220], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0677, 0.1008, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 22:29:59,091 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:30:17,320 INFO [train.py:968] (0/2) Epoch 29, batch 15100, giga_loss[loss=0.2225, simple_loss=0.3029, pruned_loss=0.07101, over 28662.00 frames. ], tot_loss[loss=0.253, simple_loss=0.332, pruned_loss=0.08697, over 5673131.35 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3314, pruned_loss=0.08581, over 5772902.88 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3333, pruned_loss=0.08763, over 5658436.48 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:30:22,353 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:30:35,347 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:31:10,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.435e+03 1.758e+03 2.377e+03 6.787e+03, threshold=3.516e+03, percent-clipped=7.0
+2023-03-14 22:31:16,682 INFO [train.py:968] (0/2) Epoch 29, batch 15150, giga_loss[loss=0.245, simple_loss=0.3255, pruned_loss=0.08223, over 28092.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3264, pruned_loss=0.08463, over 5674746.12 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3312, pruned_loss=0.08586, over 5775341.31 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3276, pruned_loss=0.08511, over 5658139.71 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:32:10,947 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1291014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:32:15,870 INFO [train.py:968] (0/2) Epoch 29, batch 15200, giga_loss[loss=0.2615, simple_loss=0.3398, pruned_loss=0.09155, over 29065.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3266, pruned_loss=0.08462, over 5674553.89 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3311, pruned_loss=0.08576, over 5776529.40 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3275, pruned_loss=0.08508, over 5659481.68 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:33:05,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 1.787e+03 2.348e+03 3.367e+03 7.435e+03, threshold=4.696e+03, percent-clipped=20.0
+2023-03-14 22:33:09,903 INFO [train.py:968] (0/2) Epoch 29, batch 15250, giga_loss[loss=0.2088, simple_loss=0.2922, pruned_loss=0.06271, over 28587.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3278, pruned_loss=0.08591, over 5675646.13 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3299, pruned_loss=0.08525, over 5780882.02 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3295, pruned_loss=0.08674, over 5655633.14 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:34:06,485 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 22:34:08,281 INFO [train.py:968] (0/2) Epoch 29, batch 15300, giga_loss[loss=0.2395, simple_loss=0.3265, pruned_loss=0.0762, over 28843.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3258, pruned_loss=0.08413, over 5676839.51 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.33, pruned_loss=0.08527, over 5782080.57 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.327, pruned_loss=0.08474, over 5658060.09 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:34:56,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291159.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:35:04,785 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.683e+02 1.515e+03 1.892e+03 2.529e+03 9.564e+03, threshold=3.783e+03, percent-clipped=6.0
+2023-03-14 22:35:11,036 INFO [train.py:968] (0/2) Epoch 29, batch 15350, giga_loss[loss=0.2471, simple_loss=0.3258, pruned_loss=0.08416, over 28167.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3257, pruned_loss=0.08346, over 5673529.08 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3296, pruned_loss=0.08512, over 5785038.94 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3269, pruned_loss=0.08407, over 5654090.18 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:35:12,671 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:35:34,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:36:17,508 INFO [train.py:968] (0/2) Epoch 29, batch 15400, giga_loss[loss=0.2269, simple_loss=0.3124, pruned_loss=0.07069, over 28894.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3245, pruned_loss=0.08336, over 5669038.44 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3299, pruned_loss=0.08543, over 5786983.52 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3252, pruned_loss=0.08352, over 5650157.98 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:37:09,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0697, 1.4177, 1.3970, 1.0600], device='cuda:0'), covar=tensor([0.2631, 0.2152, 0.1564, 0.2179], device='cuda:0'), in_proj_covar=tensor([0.2025, 0.1989, 0.1888, 0.2036], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 22:37:16,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.848e+02 1.444e+03 1.936e+03 2.603e+03 6.567e+03, threshold=3.873e+03, percent-clipped=13.0
+2023-03-14 22:37:23,347 INFO [train.py:968] (0/2) Epoch 29, batch 15450, giga_loss[loss=0.2418, simple_loss=0.3293, pruned_loss=0.07715, over 28980.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3253, pruned_loss=0.0834, over 5668681.39 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3297, pruned_loss=0.08537, over 5789162.39 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3259, pruned_loss=0.08356, over 5649828.49 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:37:57,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:37:57,511 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291302.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:38:00,052 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291305.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:38:10,479 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4805, 1.7867, 1.7139, 1.4646], device='cuda:0'), covar=tensor([0.3350, 0.2363, 0.2142, 0.2750], device='cuda:0'), in_proj_covar=tensor([0.2026, 0.1989, 0.1887, 0.2035], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 22:38:10,489 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:38:18,470 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:38:20,631 INFO [train.py:968] (0/2) Epoch 29, batch 15500, giga_loss[loss=0.2443, simple_loss=0.3249, pruned_loss=0.08189, over 28896.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3249, pruned_loss=0.08296, over 5664404.09 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3295, pruned_loss=0.0853, over 5783793.72 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3253, pruned_loss=0.08303, over 5648107.56 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:38:31,766 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:38:31,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 22:38:34,932 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:38:35,451 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291334.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:38:52,179 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:39:10,281 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:39:18,275 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.665e+02 1.363e+03 1.721e+03 2.444e+03 5.359e+03, threshold=3.441e+03, percent-clipped=6.0
+2023-03-14 22:39:22,131 INFO [train.py:968] (0/2) Epoch 29, batch 15550, giga_loss[loss=0.2653, simple_loss=0.334, pruned_loss=0.09826, over 27636.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3256, pruned_loss=0.084, over 5661221.64 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3297, pruned_loss=0.08536, over 5781954.40 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3257, pruned_loss=0.08396, over 5646860.30 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:39:44,737 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:40:22,286 INFO [train.py:968] (0/2) Epoch 29, batch 15600, giga_loss[loss=0.2456, simple_loss=0.3342, pruned_loss=0.07849, over 28953.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3244, pruned_loss=0.08295, over 5670124.69 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3294, pruned_loss=0.08535, over 5785231.64 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3246, pruned_loss=0.08289, over 5653156.14 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:40:36,470 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1291434.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:40:49,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:40:53,405 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:41:13,617 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.930e+02 1.450e+03 1.921e+03 2.716e+03 1.288e+04, threshold=3.842e+03, percent-clipped=15.0
+2023-03-14 22:41:17,261 INFO [train.py:968] (0/2) Epoch 29, batch 15650, giga_loss[loss=0.2561, simple_loss=0.3416, pruned_loss=0.08529, over 28914.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3258, pruned_loss=0.08213, over 5680700.00 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3287, pruned_loss=0.08503, over 5787562.78 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3266, pruned_loss=0.08229, over 5661861.05 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:41:17,681 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1291471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:41:24,349 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:41:42,193 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0510, 1.3439, 2.7156, 2.7432], device='cuda:0'), covar=tensor([0.1358, 0.2329, 0.0558, 0.1238], device='cuda:0'), in_proj_covar=tensor([0.0806, 0.0676, 0.1004, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 22:42:13,670 INFO [train.py:968] (0/2) Epoch 29, batch 15700, giga_loss[loss=0.2367, simple_loss=0.3343, pruned_loss=0.06956, over 28691.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3283, pruned_loss=0.08308, over 5675162.00 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3285, pruned_loss=0.08483, over 5790356.96 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3291, pruned_loss=0.08332, over 5654239.42 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:42:25,826 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:42:28,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:42:43,101 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1291548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:42:59,823 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:43:03,531 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.320e+02 1.592e+03 1.877e+03 2.597e+03 6.026e+03, threshold=3.754e+03, percent-clipped=8.0
+2023-03-14 22:43:06,931 INFO [train.py:968] (0/2) Epoch 29, batch 15750, giga_loss[loss=0.2829, simple_loss=0.3571, pruned_loss=0.1043, over 28151.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3314, pruned_loss=0.0848, over 5669655.71 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.329, pruned_loss=0.0853, over 5783932.01 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3316, pruned_loss=0.08453, over 5655464.75 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:44:00,634 INFO [train.py:968] (0/2) Epoch 29, batch 15800, giga_loss[loss=0.2472, simple_loss=0.3274, pruned_loss=0.08349, over 28980.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3315, pruned_loss=0.08504, over 5670226.52 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3285, pruned_loss=0.08505, over 5770784.92 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3322, pruned_loss=0.08506, over 5666288.17 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:44:16,016 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-14 22:44:44,446 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4003, 1.3622, 4.4487, 3.6397], device='cuda:0'), covar=tensor([0.1746, 0.2939, 0.0381, 0.0931], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0675, 0.1002, 0.0976], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 22:44:54,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.031e+03 1.489e+03 2.049e+03 3.041e+03 1.064e+04, threshold=4.098e+03, percent-clipped=14.0
+2023-03-14 22:44:55,866 INFO [train.py:968] (0/2) Epoch 29, batch 15850, giga_loss[loss=0.2254, simple_loss=0.3112, pruned_loss=0.0698, over 28698.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3294, pruned_loss=0.08357, over 5680109.44 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3281, pruned_loss=0.08498, over 5772773.33 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3304, pruned_loss=0.08365, over 5673051.82 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:45:55,288 INFO [train.py:968] (0/2) Epoch 29, batch 15900, giga_loss[loss=0.2279, simple_loss=0.3134, pruned_loss=0.07119, over 28449.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3259, pruned_loss=0.0813, over 5691898.56 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3275, pruned_loss=0.08463, over 5779759.35 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3273, pruned_loss=0.08157, over 5675530.99 frames. ], batch size: 369, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:46:14,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7022, 2.2344, 1.4389, 0.8699], device='cuda:0'), covar=tensor([0.8007, 0.4345, 0.4098, 0.7400], device='cuda:0'), in_proj_covar=tensor([0.1844, 0.1738, 0.1661, 0.1507], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 22:46:43,969 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1291763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:46:49,353 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.349e+03 1.679e+03 2.003e+03 6.038e+03, threshold=3.358e+03, percent-clipped=1.0
+2023-03-14 22:46:52,405 INFO [train.py:968] (0/2) Epoch 29, batch 15950, giga_loss[loss=0.2011, simple_loss=0.2803, pruned_loss=0.0609, over 28925.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3245, pruned_loss=0.08126, over 5679256.58 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3275, pruned_loss=0.08459, over 5772178.49 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3256, pruned_loss=0.08147, over 5672070.05 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:47:34,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291809.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:47:50,548 INFO [train.py:968] (0/2) Epoch 29, batch 16000, giga_loss[loss=0.2874, simple_loss=0.3634, pruned_loss=0.1056, over 28917.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3247, pruned_loss=0.08149, over 5676985.34 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3271, pruned_loss=0.08448, over 5773302.07 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3258, pruned_loss=0.08169, over 5668222.62 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:47:59,283 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 22:48:23,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:48:46,701 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.672e+02 1.412e+03 2.010e+03 2.979e+03 6.722e+03, threshold=4.020e+03, percent-clipped=14.0
+2023-03-14 22:48:48,219 INFO [train.py:968] (0/2) Epoch 29, batch 16050, giga_loss[loss=0.2326, simple_loss=0.3206, pruned_loss=0.07227, over 29025.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3269, pruned_loss=0.08245, over 5683145.82 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3264, pruned_loss=0.08407, over 5777375.60 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3285, pruned_loss=0.08296, over 5669511.73 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:49:10,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4190, 3.3306, 1.5632, 1.5654], device='cuda:0'), covar=tensor([0.1017, 0.0362, 0.0959, 0.1354], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0570, 0.0413, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 22:49:49,692 INFO [train.py:968] (0/2) Epoch 29, batch 16100, giga_loss[loss=0.2449, simple_loss=0.3238, pruned_loss=0.08295, over 27604.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3274, pruned_loss=0.0833, over 5681958.41 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3263, pruned_loss=0.08402, over 5780809.84 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3289, pruned_loss=0.08371, over 5665116.47 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:49:52,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:50:21,720 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291952.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:50:24,855 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291955.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:50:40,119 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.372e+02 1.419e+03 1.831e+03 2.508e+03 5.116e+03, threshold=3.661e+03, percent-clipped=7.0
+2023-03-14 22:50:41,988 INFO [train.py:968] (0/2) Epoch 29, batch 16150, giga_loss[loss=0.2214, simple_loss=0.3142, pruned_loss=0.06437, over 28854.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3305, pruned_loss=0.08523, over 5682996.89 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3262, pruned_loss=0.08409, over 5778425.73 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3318, pruned_loss=0.0855, over 5668514.41 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:50:47,382 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5045, 1.7814, 1.6689, 1.4519], device='cuda:0'), covar=tensor([0.3243, 0.2596, 0.2405, 0.2783], device='cuda:0'), in_proj_covar=tensor([0.2033, 0.1989, 0.1884, 0.2037], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 22:50:57,436 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291984.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:50:57,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4537, 1.4828, 1.5696, 1.1626], device='cuda:0'), covar=tensor([0.1931, 0.3486, 0.1636, 0.1790], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0708, 0.0979, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 22:51:03,138 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:51:06,182 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:51:15,332 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1292000.pt
+2023-03-14 22:51:40,929 INFO [train.py:968] (0/2) Epoch 29, batch 16200, giga_loss[loss=0.2875, simple_loss=0.3679, pruned_loss=0.1035, over 28498.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3332, pruned_loss=0.08556, over 5685863.79 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3261, pruned_loss=0.08405, over 5777809.06 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3343, pruned_loss=0.08584, over 5673338.33 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:51:41,285 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1292021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:52:29,962 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1292066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:52:32,844 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.758e+02 1.510e+03 1.839e+03 2.561e+03 7.904e+03, threshold=3.677e+03, percent-clipped=5.0
+2023-03-14 22:52:33,786 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1292069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:52:35,378 INFO [train.py:968] (0/2) Epoch 29, batch 16250, giga_loss[loss=0.2542, simple_loss=0.3371, pruned_loss=0.08566, over 28995.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3345, pruned_loss=0.08614, over 5691864.07 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.326, pruned_loss=0.08398, over 5782411.00 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3359, pruned_loss=0.08647, over 5674659.19 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:53:07,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1292098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:53:11,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5206, 1.8688, 1.3186, 1.0925], device='cuda:0'), covar=tensor([0.6545, 0.4127, 0.4402, 0.5937], device='cuda:0'), in_proj_covar=tensor([0.1843, 0.1737, 0.1661, 0.1504], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 22:53:39,100 INFO [train.py:968] (0/2) Epoch 29, batch 16300, giga_loss[loss=0.2335, simple_loss=0.3179, pruned_loss=0.07457, over 28961.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3346, pruned_loss=0.08636, over 5692186.85 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3268, pruned_loss=0.08451, over 5780977.75 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3352, pruned_loss=0.08623, over 5675921.38 frames. ], batch size: 120, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:54:02,226 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1292138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:54:36,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.655e+03 2.228e+03 3.292e+03 1.823e+04, threshold=4.456e+03, percent-clipped=18.0
+2023-03-14 22:54:39,920 INFO [train.py:968] (0/2) Epoch 29, batch 16350, giga_loss[loss=0.2816, simple_loss=0.3401, pruned_loss=0.1116, over 26865.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3339, pruned_loss=0.08664, over 5699823.20 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3267, pruned_loss=0.0845, over 5782448.00 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3347, pruned_loss=0.0866, over 5684023.49 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:55:08,907 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3191, 0.7630, 0.8438, 1.4406], device='cuda:0'), covar=tensor([0.0760, 0.0410, 0.0381, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0121, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 22:55:19,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3155, 1.4700, 1.5326, 1.1701], device='cuda:0'), covar=tensor([0.1688, 0.2547, 0.1392, 0.1823], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0709, 0.0979, 0.0882], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 22:55:40,710 INFO [train.py:968] (0/2) Epoch 29, batch 16400, giga_loss[loss=0.2751, simple_loss=0.3467, pruned_loss=0.1018, over 28397.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.333, pruned_loss=0.08607, over 5684990.04 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3268, pruned_loss=0.08442, over 5781229.86 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3338, pruned_loss=0.08616, over 5670769.40 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:55:41,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4851, 1.5987, 3.2525, 3.2309], device='cuda:0'), covar=tensor([0.1276, 0.2467, 0.0478, 0.1015], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0674, 0.1002, 0.0975], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 22:56:21,724 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-14 22:56:42,077 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.068e+02 1.437e+03 1.839e+03 2.571e+03 1.274e+04, threshold=3.678e+03, percent-clipped=4.0
+2023-03-14 22:56:42,918 INFO [train.py:968] (0/2) Epoch 29, batch 16450, giga_loss[loss=0.2111, simple_loss=0.295, pruned_loss=0.06359, over 28840.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3307, pruned_loss=0.08552, over 5681078.38 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3267, pruned_loss=0.08436, over 5779619.10 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3315, pruned_loss=0.08567, over 5669887.49 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:56:54,670 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1292281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:56:58,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1292284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:57:03,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3511, 1.8494, 1.5879, 1.4729], device='cuda:0'), covar=tensor([0.0669, 0.0275, 0.0295, 0.0839], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 22:57:19,461 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:57:31,693 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1292313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:57:41,590 INFO [train.py:968] (0/2) Epoch 29, batch 16500, giga_loss[loss=0.3023, simple_loss=0.369, pruned_loss=0.1178, over 28978.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3298, pruned_loss=0.08594, over 5674948.14 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.327, pruned_loss=0.08444, over 5779187.03 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3302, pruned_loss=0.08602, over 5663645.13 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:58:08,644 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1456, 1.1545, 3.2912, 2.9036], device='cuda:0'), covar=tensor([0.1663, 0.2877, 0.0530, 0.1334], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0675, 0.1003, 0.0977], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 22:58:35,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2454, 1.7438, 1.2727, 0.4407], device='cuda:0'), covar=tensor([0.4721, 0.3031, 0.4642, 0.6753], device='cuda:0'), in_proj_covar=tensor([0.1845, 0.1737, 0.1662, 0.1506], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 22:58:43,387 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.863e+02 1.454e+03 1.878e+03 2.853e+03 6.837e+03, threshold=3.756e+03, percent-clipped=13.0
+2023-03-14 22:58:43,399 INFO [train.py:968] (0/2) Epoch 29, batch 16550, giga_loss[loss=0.2326, simple_loss=0.3044, pruned_loss=0.08036, over 24484.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.329, pruned_loss=0.08427, over 5677856.59 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3271, pruned_loss=0.0845, over 5780632.80 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3293, pruned_loss=0.08429, over 5665266.33 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:58:52,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6929, 1.8643, 1.2927, 1.4216], device='cuda:0'), covar=tensor([0.1067, 0.0590, 0.1003, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0447, 0.0523, 0.0458], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 22:59:34,513 INFO [train.py:968] (0/2) Epoch 29, batch 16600, giga_loss[loss=0.2319, simple_loss=0.3314, pruned_loss=0.06626, over 28933.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3276, pruned_loss=0.08289, over 5682337.31 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3269, pruned_loss=0.08446, over 5781327.17 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.328, pruned_loss=0.08289, over 5667261.53 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 23:00:29,685 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-14 23:00:32,741 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.674e+02 1.464e+03 1.863e+03 2.528e+03 5.282e+03, threshold=3.726e+03, percent-clipped=10.0
+2023-03-14 23:00:32,753 INFO [train.py:968] (0/2) Epoch 29, batch 16650, giga_loss[loss=0.2295, simple_loss=0.3229, pruned_loss=0.06811, over 28997.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3302, pruned_loss=0.08257, over 5682697.19 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3269, pruned_loss=0.08445, over 5782926.84 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3306, pruned_loss=0.08256, over 5668032.77 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 23:00:52,311 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-14 23:01:22,178 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 23:01:30,318 INFO [train.py:968] (0/2) Epoch 29, batch 16700, giga_loss[loss=0.2078, simple_loss=0.305, pruned_loss=0.0553, over 29055.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3301, pruned_loss=0.08157, over 5685436.84 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3266, pruned_loss=0.08433, over 5785320.07 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3308, pruned_loss=0.08163, over 5669926.28 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 23:02:25,762 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.530e+02 1.523e+03 1.879e+03 2.566e+03 7.120e+03, threshold=3.757e+03, percent-clipped=9.0
+2023-03-14 23:02:25,775 INFO [train.py:968] (0/2) Epoch 29, batch 16750, giga_loss[loss=0.2487, simple_loss=0.3309, pruned_loss=0.0832, over 27994.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3295, pruned_loss=0.08126, over 5692409.41 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3262, pruned_loss=0.08414, over 5787226.10 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3305, pruned_loss=0.08139, over 5674824.10 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 23:03:35,921 INFO [train.py:968] (0/2) Epoch 29, batch 16800, giga_loss[loss=0.2233, simple_loss=0.3115, pruned_loss=0.06754, over 28840.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3294, pruned_loss=0.08094, over 5681832.62 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3262, pruned_loss=0.08414, over 5787226.10 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3301, pruned_loss=0.08104, over 5668145.75 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:04:44,774 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.533e+03 1.967e+03 2.526e+03 5.897e+03, threshold=3.934e+03, percent-clipped=6.0
+2023-03-14 23:04:44,786 INFO [train.py:968] (0/2) Epoch 29, batch 16850, giga_loss[loss=0.238, simple_loss=0.3259, pruned_loss=0.07504, over 29012.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3307, pruned_loss=0.08207, over 5682662.97 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3261, pruned_loss=0.08414, over 5789200.76 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3316, pruned_loss=0.0821, over 5667980.95 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:04:45,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4666, 1.5191, 1.2588, 1.1282], device='cuda:0'), covar=tensor([0.0860, 0.0362, 0.0820, 0.1048], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0446, 0.0522, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 23:04:53,884 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1292679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:05:46,545 INFO [train.py:968] (0/2) Epoch 29, batch 16900, giga_loss[loss=0.2168, simple_loss=0.3092, pruned_loss=0.06216, over 28038.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3304, pruned_loss=0.08101, over 5689347.27 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3259, pruned_loss=0.08396, over 5789977.75 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3315, pruned_loss=0.08107, over 5671823.42 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:06:03,455 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-14 23:06:50,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.586e+03 2.044e+03 3.062e+03 7.182e+03, threshold=4.089e+03, percent-clipped=13.0
+2023-03-14 23:06:50,723 INFO [train.py:968] (0/2) Epoch 29, batch 16950, giga_loss[loss=0.2628, simple_loss=0.3371, pruned_loss=0.09422, over 24530.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3325, pruned_loss=0.08242, over 5677817.17 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3255, pruned_loss=0.0839, over 5781375.79 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3339, pruned_loss=0.08248, over 5669705.33 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:07:04,813 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:07:32,301 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:07:57,239 INFO [train.py:968] (0/2) Epoch 29, batch 17000, libri_loss[loss=0.248, simple_loss=0.3305, pruned_loss=0.08275, over 29520.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3356, pruned_loss=0.08373, over 5688043.80 frames. ], libri_tot_loss[loss=0.2463, simple_loss=0.3251, pruned_loss=0.08372, over 5783942.24 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3373, pruned_loss=0.08395, over 5676888.66 frames. ], batch size: 81, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:07:59,668 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1292822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:08:02,118 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1292825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:08:39,413 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1292854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:08:56,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:08:59,360 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.514e+02 1.528e+03 2.005e+03 2.575e+03 7.122e+03, threshold=4.010e+03, percent-clipped=6.0
+2023-03-14 23:08:59,373 INFO [train.py:968] (0/2) Epoch 29, batch 17050, giga_loss[loss=0.2285, simple_loss=0.3136, pruned_loss=0.07167, over 28551.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3345, pruned_loss=0.0835, over 5691528.15 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3251, pruned_loss=0.0836, over 5787382.09 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3362, pruned_loss=0.08377, over 5676871.14 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:09:45,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4878, 1.6966, 1.2000, 1.2861], device='cuda:0'), covar=tensor([0.1035, 0.0575, 0.1065, 0.1209], device='cuda:0'), in_proj_covar=tensor([0.0414, 0.0446, 0.0523, 0.0457], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 23:10:07,910 INFO [train.py:968] (0/2) Epoch 29, batch 17100, giga_loss[loss=0.2367, simple_loss=0.3259, pruned_loss=0.07377, over 28733.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.332, pruned_loss=0.0829, over 5696975.90 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3247, pruned_loss=0.08343, over 5789278.63 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3339, pruned_loss=0.08327, over 5681588.24 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:10:14,086 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3669, 1.5986, 1.3341, 1.5199], device='cuda:0'), covar=tensor([0.0762, 0.0386, 0.0366, 0.0862], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0121, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 23:10:26,869 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:10:49,001 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:11:16,656 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.627e+02 1.556e+03 2.086e+03 3.014e+03 1.017e+04, threshold=4.172e+03, percent-clipped=11.0
+2023-03-14 23:11:16,669 INFO [train.py:968] (0/2) Epoch 29, batch 17150, giga_loss[loss=0.1943, simple_loss=0.2902, pruned_loss=0.0492, over 29073.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.331, pruned_loss=0.08139, over 5699938.37 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3249, pruned_loss=0.08347, over 5788102.27 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3325, pruned_loss=0.08161, over 5687526.70 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:11:20,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3285, 1.7533, 1.7588, 1.5283], device='cuda:0'), covar=tensor([0.2251, 0.2101, 0.2264, 0.2182], device='cuda:0'), in_proj_covar=tensor([0.0488, 0.0739, 0.0712, 0.0679], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 23:12:14,623 INFO [train.py:968] (0/2) Epoch 29, batch 17200, libri_loss[loss=0.2551, simple_loss=0.3328, pruned_loss=0.08872, over 29527.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3297, pruned_loss=0.08075, over 5701288.37 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3249, pruned_loss=0.08344, over 5786129.38 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3311, pruned_loss=0.08084, over 5689161.34 frames. ], batch size: 84, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 23:13:10,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.661e+02 1.326e+03 1.804e+03 2.634e+03 4.718e+03, threshold=3.608e+03, percent-clipped=2.0
+2023-03-14 23:13:10,013 INFO [train.py:968] (0/2) Epoch 29, batch 17250, giga_loss[loss=0.2685, simple_loss=0.3591, pruned_loss=0.08893, over 28712.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3311, pruned_loss=0.08172, over 5692654.88 frames. ], libri_tot_loss[loss=0.2456, simple_loss=0.3246, pruned_loss=0.08325, over 5784810.95 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3326, pruned_loss=0.08193, over 5682224.04 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:13:30,685 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1293091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:14:04,956 INFO [train.py:968] (0/2) Epoch 29, batch 17300, giga_loss[loss=0.2293, simple_loss=0.3164, pruned_loss=0.07109, over 28994.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3345, pruned_loss=0.08405, over 5682313.37 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3254, pruned_loss=0.08378, over 5776214.28 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3352, pruned_loss=0.08372, over 5678983.25 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:14:20,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5426, 1.5901, 1.7428, 1.3771], device='cuda:0'), covar=tensor([0.1869, 0.2656, 0.1597, 0.1910], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0708, 0.0981, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 23:14:21,192 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6993, 2.0442, 2.0068, 1.7113], device='cuda:0'), covar=tensor([0.2216, 0.2381, 0.1950, 0.2299], device='cuda:0'), in_proj_covar=tensor([0.0486, 0.0737, 0.0710, 0.0677], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 23:14:41,078 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1293156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:14:57,824 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8191, 2.2081, 2.0232, 1.8547], device='cuda:0'), covar=tensor([0.2222, 0.2303, 0.2077, 0.2291], device='cuda:0'), in_proj_covar=tensor([0.0487, 0.0737, 0.0711, 0.0678], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 23:14:58,156 INFO [train.py:968] (0/2) Epoch 29, batch 17350, giga_loss[loss=0.2317, simple_loss=0.3119, pruned_loss=0.07573, over 28928.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3328, pruned_loss=0.08356, over 5679949.12 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3253, pruned_loss=0.08372, over 5777857.74 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3337, pruned_loss=0.08336, over 5673334.93 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:15:00,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.122e+03 1.702e+03 2.039e+03 2.759e+03 1.172e+04, threshold=4.078e+03, percent-clipped=11.0
+2023-03-14 23:15:04,753 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1293176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:15:52,301 INFO [train.py:968] (0/2) Epoch 29, batch 17400, giga_loss[loss=0.2282, simple_loss=0.3185, pruned_loss=0.06891, over 28984.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3308, pruned_loss=0.08379, over 5676269.31 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3248, pruned_loss=0.08355, over 5772688.88 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3322, pruned_loss=0.08381, over 5671936.12 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:16:18,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1293244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:16:47,183 INFO [train.py:968] (0/2) Epoch 29, batch 17450, giga_loss[loss=0.2357, simple_loss=0.3186, pruned_loss=0.07642, over 28897.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3312, pruned_loss=0.08463, over 5687713.80 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3246, pruned_loss=0.08355, over 5776007.35 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3326, pruned_loss=0.08466, over 5679084.86 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:16:48,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.139e+03 1.652e+03 2.022e+03 2.584e+03 7.059e+03, threshold=4.045e+03, percent-clipped=4.0
+2023-03-14 23:17:18,302 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:17:21,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:17:34,050 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1293311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:17:44,375 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:17:45,263 INFO [train.py:968] (0/2) Epoch 29, batch 17500, giga_loss[loss=0.29, simple_loss=0.3768, pruned_loss=0.1016, over 28944.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3388, pruned_loss=0.08931, over 5680487.45 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3246, pruned_loss=0.08354, over 5777559.37 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3401, pruned_loss=0.08938, over 5671511.83 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:17:46,920 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:17:49,294 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1293325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:17:55,017 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:18:11,341 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:18:29,224 INFO [train.py:968] (0/2) Epoch 29, batch 17550, giga_loss[loss=0.2931, simple_loss=0.3518, pruned_loss=0.1172, over 23693.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3461, pruned_loss=0.09306, over 5690071.45 frames. ], libri_tot_loss[loss=0.2457, simple_loss=0.3245, pruned_loss=0.08347, over 5780131.77 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3475, pruned_loss=0.09334, over 5678663.04 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:18:29,824 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.059e+03 1.665e+03 2.003e+03 2.809e+03 8.240e+03, threshold=4.006e+03, percent-clipped=9.0
+2023-03-14 23:18:44,031 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:18:46,944 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:12,840 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:13,890 INFO [train.py:968] (0/2) Epoch 29, batch 17600, giga_loss[loss=0.2416, simple_loss=0.32, pruned_loss=0.08162, over 28874.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3479, pruned_loss=0.09404, over 5692375.36 frames. ], libri_tot_loss[loss=0.246, simple_loss=0.3248, pruned_loss=0.08362, over 5780328.91 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3491, pruned_loss=0.09427, over 5681976.92 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:19:37,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5695, 1.6796, 1.2855, 1.2055], device='cuda:0'), covar=tensor([0.1142, 0.0697, 0.1125, 0.1368], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0445, 0.0522, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 23:19:45,454 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:49,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:56,306 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1293466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:57,497 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:59,870 INFO [train.py:968] (0/2) Epoch 29, batch 17650, giga_loss[loss=0.2372, simple_loss=0.3161, pruned_loss=0.07918, over 28643.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3421, pruned_loss=0.09185, over 5691413.74 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3247, pruned_loss=0.08352, over 5782589.14 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3434, pruned_loss=0.09226, over 5679928.36 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:20:00,134 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:20:00,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.992e+02 1.229e+03 1.491e+03 2.123e+03 8.595e+03, threshold=2.981e+03, percent-clipped=1.0
+2023-03-14 23:20:10,107 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:20:23,053 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:20:40,219 INFO [train.py:968] (0/2) Epoch 29, batch 17700, giga_loss[loss=0.2605, simple_loss=0.3321, pruned_loss=0.09443, over 27913.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3346, pruned_loss=0.08842, over 5696100.03 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3248, pruned_loss=0.08347, over 5786309.21 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.336, pruned_loss=0.08902, over 5680155.12 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:20:40,544 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7861, 2.0589, 1.7220, 1.7925], device='cuda:0'), covar=tensor([0.2730, 0.2874, 0.3330, 0.2617], device='cuda:0'), in_proj_covar=tensor([0.1604, 0.1155, 0.1423, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 23:21:19,249 INFO [train.py:968] (0/2) Epoch 29, batch 17750, giga_loss[loss=0.2159, simple_loss=0.2955, pruned_loss=0.06811, over 28718.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3287, pruned_loss=0.08603, over 5693806.24 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.325, pruned_loss=0.0834, over 5780664.68 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.33, pruned_loss=0.08674, over 5681685.93 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:21:20,572 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.152e+02 1.273e+03 1.652e+03 2.151e+03 6.218e+03, threshold=3.304e+03, percent-clipped=10.0
+2023-03-14 23:21:52,544 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:21:55,519 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:22:04,483 INFO [train.py:968] (0/2) Epoch 29, batch 17800, giga_loss[loss=0.2055, simple_loss=0.283, pruned_loss=0.06402, over 29031.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3215, pruned_loss=0.0829, over 5692150.49 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3251, pruned_loss=0.08326, over 5782804.19 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3224, pruned_loss=0.08363, over 5677556.72 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:22:15,592 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-14 23:22:18,655 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:22:44,884 INFO [train.py:968] (0/2) Epoch 29, batch 17850, giga_loss[loss=0.2297, simple_loss=0.2948, pruned_loss=0.08231, over 28003.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3159, pruned_loss=0.08061, over 5692906.71 frames. ], libri_tot_loss[loss=0.2455, simple_loss=0.3248, pruned_loss=0.08305, over 5780037.66 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3167, pruned_loss=0.08135, over 5680823.39 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:22:46,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.752e+02 1.105e+03 1.334e+03 1.786e+03 6.754e+03, threshold=2.667e+03, percent-clipped=2.0
+2023-03-14 23:23:26,781 INFO [train.py:968] (0/2) Epoch 29, batch 17900, giga_loss[loss=0.2351, simple_loss=0.32, pruned_loss=0.07515, over 28914.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3119, pruned_loss=0.07856, over 5700637.74 frames. ], libri_tot_loss[loss=0.2454, simple_loss=0.3248, pruned_loss=0.08294, over 5782515.50 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.3122, pruned_loss=0.07916, over 5687191.49 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:24:08,439 INFO [train.py:968] (0/2) Epoch 29, batch 17950, giga_loss[loss=0.2122, simple_loss=0.2877, pruned_loss=0.06836, over 28615.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3087, pruned_loss=0.07748, over 5700548.03 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3245, pruned_loss=0.0827, over 5784033.11 frames. ], giga_tot_loss[loss=0.2326, simple_loss=0.3091, pruned_loss=0.07811, over 5687472.70 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:24:09,650 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.625e+02 1.169e+03 1.437e+03 1.840e+03 5.627e+03, threshold=2.875e+03, percent-clipped=4.0
+2023-03-14 23:24:14,965 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5980, 2.2027, 1.6295, 0.7724], device='cuda:0'), covar=tensor([0.8186, 0.3703, 0.5084, 0.8597], device='cuda:0'), in_proj_covar=tensor([0.1842, 0.1743, 0.1665, 0.1505], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 23:24:42,481 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-14 23:24:49,889 INFO [train.py:968] (0/2) Epoch 29, batch 18000, giga_loss[loss=0.2406, simple_loss=0.3167, pruned_loss=0.08227, over 27944.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3065, pruned_loss=0.07656, over 5687434.07 frames. ], libri_tot_loss[loss=0.2457, simple_loss=0.3251, pruned_loss=0.08309, over 5769100.68 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.3057, pruned_loss=0.07654, over 5687091.77 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:24:49,893 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-14 23:24:56,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3102, 1.2406, 1.0986, 1.5092], device='cuda:0'), covar=tensor([0.0848, 0.0387, 0.0413, 0.0942], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 23:24:58,148 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.2, simple_loss=0.306, pruned_loss=0.047, over 944034.00 frames. 
+2023-03-14 23:24:58,148 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-14 23:25:23,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3975, 1.6010, 1.5634, 1.5769], device='cuda:0'), covar=tensor([0.0782, 0.0337, 0.0323, 0.0904], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 23:25:39,575 INFO [train.py:968] (0/2) Epoch 29, batch 18050, libri_loss[loss=0.2756, simple_loss=0.3597, pruned_loss=0.09574, over 27733.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3042, pruned_loss=0.07578, over 5691935.66 frames. ], libri_tot_loss[loss=0.246, simple_loss=0.3255, pruned_loss=0.08328, over 5770728.50 frames. ], giga_tot_loss[loss=0.2269, simple_loss=0.3028, pruned_loss=0.07545, over 5688638.94 frames. ], batch size: 116, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:25:40,964 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.259e+02 1.159e+03 1.476e+03 2.016e+03 6.773e+03, threshold=2.952e+03, percent-clipped=9.0
+2023-03-14 23:25:59,497 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1293900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:26:18,205 INFO [train.py:968] (0/2) Epoch 29, batch 18100, giga_loss[loss=0.1986, simple_loss=0.2792, pruned_loss=0.05901, over 29021.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3028, pruned_loss=0.07527, over 5683598.33 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3262, pruned_loss=0.08358, over 5766027.82 frames. ], giga_tot_loss[loss=0.2246, simple_loss=0.3003, pruned_loss=0.07443, over 5682842.50 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:26:55,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7838, 1.7332, 1.9106, 1.3521], device='cuda:0'), covar=tensor([0.2104, 0.3515, 0.1727, 0.1940], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0714, 0.0990, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 23:26:58,365 INFO [train.py:968] (0/2) Epoch 29, batch 18150, giga_loss[loss=0.2159, simple_loss=0.2885, pruned_loss=0.07167, over 28885.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3012, pruned_loss=0.07477, over 5682423.49 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3268, pruned_loss=0.08378, over 5758111.68 frames. ], giga_tot_loss[loss=0.2227, simple_loss=0.298, pruned_loss=0.07368, over 5686205.96 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:27:00,415 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.206e+02 1.242e+03 1.521e+03 2.287e+03 7.589e+03, threshold=3.041e+03, percent-clipped=15.0
+2023-03-14 23:27:24,988 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1294000.pt
+2023-03-14 23:27:43,264 INFO [train.py:968] (0/2) Epoch 29, batch 18200, giga_loss[loss=0.2022, simple_loss=0.2792, pruned_loss=0.0626, over 28865.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2973, pruned_loss=0.07337, over 5684502.42 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3267, pruned_loss=0.08365, over 5759849.36 frames. ], giga_tot_loss[loss=0.2197, simple_loss=0.2945, pruned_loss=0.07248, over 5684975.00 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:28:09,115 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7908, 1.7442, 1.9934, 1.5954], device='cuda:0'), covar=tensor([0.2168, 0.2773, 0.1688, 0.1929], device='cuda:0'), in_proj_covar=tensor([0.0946, 0.0717, 0.0994, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 23:28:27,677 INFO [train.py:968] (0/2) Epoch 29, batch 18250, giga_loss[loss=0.199, simple_loss=0.2793, pruned_loss=0.05934, over 28825.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2952, pruned_loss=0.07224, over 5682376.53 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3272, pruned_loss=0.08379, over 5764313.98 frames. ], giga_tot_loss[loss=0.2168, simple_loss=0.2915, pruned_loss=0.07107, over 5676450.14 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:28:29,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.559e+02 1.065e+03 1.308e+03 1.869e+03 4.813e+03, threshold=2.615e+03, percent-clipped=2.0
+2023-03-14 23:28:49,148 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:29:08,122 INFO [train.py:968] (0/2) Epoch 29, batch 18300, giga_loss[loss=0.1933, simple_loss=0.2674, pruned_loss=0.05957, over 28567.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2945, pruned_loss=0.072, over 5683880.74 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3274, pruned_loss=0.08379, over 5758735.54 frames. ], giga_tot_loss[loss=0.2155, simple_loss=0.29, pruned_loss=0.07056, over 5681711.60 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:29:58,301 INFO [train.py:968] (0/2) Epoch 29, batch 18350, libri_loss[loss=0.2817, simple_loss=0.3598, pruned_loss=0.1018, over 29247.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3027, pruned_loss=0.07683, over 5673475.64 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3275, pruned_loss=0.08378, over 5761400.66 frames. ], giga_tot_loss[loss=0.2247, simple_loss=0.2985, pruned_loss=0.0755, over 5667921.06 frames. ], batch size: 94, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:30:01,375 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.925e+02 1.274e+03 1.587e+03 2.112e+03 4.915e+03, threshold=3.173e+03, percent-clipped=13.0
+2023-03-14 23:30:31,167 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294206.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 23:30:32,022 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:30:42,261 INFO [train.py:968] (0/2) Epoch 29, batch 18400, giga_loss[loss=0.2773, simple_loss=0.3585, pruned_loss=0.09804, over 28944.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3157, pruned_loss=0.08321, over 5684803.36 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3277, pruned_loss=0.08379, over 5766376.97 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.3114, pruned_loss=0.082, over 5672667.97 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:30:47,986 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8466, 2.0143, 1.8737, 1.7501], device='cuda:0'), covar=tensor([0.1931, 0.2054, 0.2232, 0.2211], device='cuda:0'), in_proj_covar=tensor([0.0495, 0.0750, 0.0720, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 23:31:22,621 INFO [train.py:968] (0/2) Epoch 29, batch 18450, giga_loss[loss=0.2715, simple_loss=0.3527, pruned_loss=0.09522, over 28874.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3268, pruned_loss=0.08857, over 5695373.23 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3279, pruned_loss=0.0839, over 5766767.71 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3231, pruned_loss=0.08756, over 5684274.73 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:31:24,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.115e+03 1.493e+03 1.826e+03 2.526e+03 5.057e+03, threshold=3.651e+03, percent-clipped=14.0
+2023-03-14 23:31:25,763 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1294275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:32:01,572 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5172, 2.1060, 1.5449, 0.8456], device='cuda:0'), covar=tensor([0.7935, 0.3782, 0.4700, 0.7536], device='cuda:0'), in_proj_covar=tensor([0.1842, 0.1738, 0.1663, 0.1502], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 23:32:04,567 INFO [train.py:968] (0/2) Epoch 29, batch 18500, giga_loss[loss=0.2401, simple_loss=0.3334, pruned_loss=0.0734, over 29056.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3339, pruned_loss=0.09134, over 5689615.05 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3283, pruned_loss=0.08405, over 5766426.97 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3307, pruned_loss=0.09047, over 5680526.01 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:32:30,942 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1475, 1.5245, 1.4629, 1.3491], device='cuda:0'), covar=tensor([0.2288, 0.1795, 0.2621, 0.2007], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0752, 0.0722, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 23:32:46,977 INFO [train.py:968] (0/2) Epoch 29, batch 18550, libri_loss[loss=0.2357, simple_loss=0.3171, pruned_loss=0.0772, over 29561.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3367, pruned_loss=0.0914, over 5688790.75 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3285, pruned_loss=0.08405, over 5765633.64 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3342, pruned_loss=0.09089, over 5680415.53 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:32:48,967 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.635e+02 1.241e+03 1.683e+03 2.163e+03 4.164e+03, threshold=3.366e+03, percent-clipped=4.0
+2023-03-14 23:33:24,490 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1294418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:33:27,943 INFO [train.py:968] (0/2) Epoch 29, batch 18600, libri_loss[loss=0.252, simple_loss=0.3385, pruned_loss=0.08273, over 29476.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3376, pruned_loss=0.09067, over 5685093.71 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3285, pruned_loss=0.0838, over 5768195.31 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3358, pruned_loss=0.09079, over 5672492.63 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:33:28,115 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1294421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:33:30,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1085, 3.9126, 3.7110, 1.8650], device='cuda:0'), covar=tensor([0.0661, 0.0820, 0.0756, 0.2171], device='cuda:0'), in_proj_covar=tensor([0.1281, 0.1184, 0.0995, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-14 23:33:55,096 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1294450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:34:11,131 INFO [train.py:968] (0/2) Epoch 29, batch 18650, giga_loss[loss=0.268, simple_loss=0.3453, pruned_loss=0.09534, over 28954.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3393, pruned_loss=0.09184, over 5681933.95 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3286, pruned_loss=0.08379, over 5769648.61 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.338, pruned_loss=0.09203, over 5669871.95 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:34:12,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1294473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:34:13,331 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.571e+02 1.278e+03 1.603e+03 2.051e+03 5.276e+03, threshold=3.205e+03, percent-clipped=4.0
+2023-03-14 23:34:55,369 INFO [train.py:968] (0/2) Epoch 29, batch 18700, giga_loss[loss=0.3036, simple_loss=0.379, pruned_loss=0.1141, over 28952.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3418, pruned_loss=0.09397, over 5683830.34 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3282, pruned_loss=0.08361, over 5771589.46 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3412, pruned_loss=0.09439, over 5671597.08 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:35:29,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5702, 3.5750, 1.6907, 1.6856], device='cuda:0'), covar=tensor([0.1089, 0.0278, 0.0922, 0.1438], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0566, 0.0412, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-14 23:35:38,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0164, 1.3783, 3.3284, 2.9940], device='cuda:0'), covar=tensor([0.1839, 0.2771, 0.0519, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.0803, 0.0673, 0.1004, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 23:35:39,608 INFO [train.py:968] (0/2) Epoch 29, batch 18750, giga_loss[loss=0.2839, simple_loss=0.3654, pruned_loss=0.1012, over 28920.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3451, pruned_loss=0.09603, over 5671666.03 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3284, pruned_loss=0.08372, over 5763720.72 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3447, pruned_loss=0.09639, over 5668590.72 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:35:43,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.736e+02 1.402e+03 1.689e+03 2.161e+03 3.651e+03, threshold=3.378e+03, percent-clipped=5.0
+2023-03-14 23:35:47,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1294581.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 23:35:47,950 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1294582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:36:16,489 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1294616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:36:18,726 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1294619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:36:19,842 INFO [train.py:968] (0/2) Epoch 29, batch 18800, giga_loss[loss=0.2621, simple_loss=0.3459, pruned_loss=0.08911, over 28893.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3481, pruned_loss=0.09679, over 5675737.73 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.329, pruned_loss=0.08392, over 5760939.48 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3476, pruned_loss=0.09717, over 5673792.75 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:36:41,235 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:36:42,449 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1294648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:36:53,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2106, 1.4871, 1.4999, 1.3464], device='cuda:0'), covar=tensor([0.2117, 0.1756, 0.2464, 0.1996], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0754, 0.0725, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 23:37:02,531 INFO [train.py:968] (0/2) Epoch 29, batch 18850, giga_loss[loss=0.2901, simple_loss=0.3715, pruned_loss=0.1044, over 28702.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3511, pruned_loss=0.09791, over 5676155.70 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3289, pruned_loss=0.08387, over 5761677.78 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3509, pruned_loss=0.09831, over 5673700.99 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:37:06,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.786e+02 1.354e+03 1.566e+03 1.997e+03 4.309e+03, threshold=3.132e+03, percent-clipped=5.0
+2023-03-14 23:37:10,740 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294681.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:37:42,984 INFO [train.py:968] (0/2) Epoch 29, batch 18900, giga_loss[loss=0.2984, simple_loss=0.3468, pruned_loss=0.125, over 23513.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3518, pruned_loss=0.09743, over 5682986.15 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.329, pruned_loss=0.08366, over 5762101.89 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.352, pruned_loss=0.09824, over 5678828.72 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:37:45,636 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1294724.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 23:37:46,329 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1294725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:37:46,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7163, 1.9029, 1.5295, 1.9014], device='cuda:0'), covar=tensor([0.2778, 0.2967, 0.3381, 0.2733], device='cuda:0'), in_proj_covar=tensor([0.1608, 0.1160, 0.1423, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 23:37:48,030 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1294727.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 23:37:49,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1294728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:38:11,159 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1294756.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 23:38:11,670 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1294757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:38:21,541 INFO [train.py:968] (0/2) Epoch 29, batch 18950, giga_loss[loss=0.2529, simple_loss=0.3422, pruned_loss=0.08176, over 28716.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3511, pruned_loss=0.09544, over 5696842.84 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3295, pruned_loss=0.08393, over 5763521.06 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3511, pruned_loss=0.09593, over 5691542.58 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:38:24,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.292e+02 1.180e+03 1.474e+03 1.850e+03 4.006e+03, threshold=2.947e+03, percent-clipped=4.0
+2023-03-14 23:38:53,330 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:39:01,749 INFO [train.py:968] (0/2) Epoch 29, batch 19000, giga_loss[loss=0.255, simple_loss=0.3471, pruned_loss=0.08144, over 28398.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3499, pruned_loss=0.09372, over 5702464.04 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3296, pruned_loss=0.08388, over 5763805.01 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3501, pruned_loss=0.09433, over 5697040.40 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:39:11,201 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3257, 1.5672, 1.1844, 1.1326], device='cuda:0'), covar=tensor([0.1246, 0.0614, 0.1185, 0.1223], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0447, 0.0525, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 23:39:38,913 INFO [train.py:968] (0/2) Epoch 29, batch 19050, giga_loss[loss=0.2856, simple_loss=0.3631, pruned_loss=0.1041, over 28679.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3497, pruned_loss=0.09367, over 5699593.79 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3306, pruned_loss=0.08437, over 5756754.71 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3494, pruned_loss=0.09392, over 5699456.28 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:39:43,214 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.464e+02 1.378e+03 1.765e+03 2.463e+03 6.458e+03, threshold=3.529e+03, percent-clipped=18.0
+2023-03-14 23:40:19,197 INFO [train.py:968] (0/2) Epoch 29, batch 19100, libri_loss[loss=0.242, simple_loss=0.3262, pruned_loss=0.07894, over 29533.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3518, pruned_loss=0.09697, over 5709870.53 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3311, pruned_loss=0.08452, over 5761111.57 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3519, pruned_loss=0.09746, over 5703973.52 frames. ], batch size: 80, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:41:04,581 INFO [train.py:968] (0/2) Epoch 29, batch 19150, giga_loss[loss=0.2638, simple_loss=0.3424, pruned_loss=0.09261, over 28924.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3547, pruned_loss=0.1022, over 5712154.57 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3309, pruned_loss=0.08448, over 5763850.49 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3553, pruned_loss=0.1029, over 5704221.70 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:41:08,497 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.949e+02 1.498e+03 1.846e+03 2.595e+03 6.690e+03, threshold=3.693e+03, percent-clipped=8.0
+2023-03-14 23:41:41,327 INFO [train.py:968] (0/2) Epoch 29, batch 19200, giga_loss[loss=0.2649, simple_loss=0.3464, pruned_loss=0.09173, over 28385.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3527, pruned_loss=0.1016, over 5716467.36 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.331, pruned_loss=0.08435, over 5767051.67 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3538, pruned_loss=0.1028, over 5705766.91 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:41:41,501 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:41:41,558 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:42:11,202 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295056.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:42:11,912 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6344, 1.9459, 1.6181, 1.7440], device='cuda:0'), covar=tensor([0.2191, 0.2393, 0.2494, 0.2233], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0753, 0.0723, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 23:42:26,373 INFO [train.py:968] (0/2) Epoch 29, batch 19250, giga_loss[loss=0.253, simple_loss=0.3245, pruned_loss=0.09072, over 28689.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3502, pruned_loss=0.1009, over 5701897.24 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3313, pruned_loss=0.08445, over 5761239.32 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3513, pruned_loss=0.1021, over 5697155.54 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:42:30,073 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.106e+02 1.304e+03 1.590e+03 2.058e+03 4.057e+03, threshold=3.180e+03, percent-clipped=1.0
+2023-03-14 23:43:06,521 INFO [train.py:968] (0/2) Epoch 29, batch 19300, giga_loss[loss=0.2725, simple_loss=0.3509, pruned_loss=0.097, over 28240.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3484, pruned_loss=0.09974, over 5708445.59 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3315, pruned_loss=0.08439, over 5764702.47 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3495, pruned_loss=0.1011, over 5700254.70 frames. ], batch size: 77, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:43:26,365 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-14 23:43:26,375 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-14 23:43:35,315 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5787, 1.7695, 1.4777, 1.6470], device='cuda:0'), covar=tensor([0.2613, 0.2677, 0.2862, 0.2478], device='cuda:0'), in_proj_covar=tensor([0.1607, 0.1159, 0.1421, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 23:43:41,346 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:43:45,473 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:43:47,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1295167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:43:50,572 INFO [train.py:968] (0/2) Epoch 29, batch 19350, giga_loss[loss=0.2816, simple_loss=0.3586, pruned_loss=0.1023, over 28524.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.347, pruned_loss=0.09804, over 5714232.34 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3314, pruned_loss=0.08429, over 5765819.71 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3481, pruned_loss=0.09942, over 5706164.88 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:43:54,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.979e+02 1.321e+03 1.646e+03 2.093e+03 4.286e+03, threshold=3.292e+03, percent-clipped=6.0
+2023-03-14 23:44:01,860 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:44:10,761 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1295196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:44:13,990 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295199.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:44:15,243 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4570, 1.4580, 4.1104, 3.3639], device='cuda:0'), covar=tensor([0.1595, 0.2809, 0.0399, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0674, 0.1007, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 23:44:15,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1295202.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:44:34,036 INFO [train.py:968] (0/2) Epoch 29, batch 19400, giga_loss[loss=0.2239, simple_loss=0.309, pruned_loss=0.06943, over 28352.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3453, pruned_loss=0.0966, over 5695836.01 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3318, pruned_loss=0.08437, over 5754797.84 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3462, pruned_loss=0.09791, over 5698100.51 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 2.0
+2023-03-14 23:44:43,414 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1295231.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:45:15,244 INFO [train.py:968] (0/2) Epoch 29, batch 19450, giga_loss[loss=0.2456, simple_loss=0.3263, pruned_loss=0.08241, over 29006.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3398, pruned_loss=0.09354, over 5694042.87 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.332, pruned_loss=0.0846, over 5758799.98 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3405, pruned_loss=0.09462, over 5690766.43 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 2.0
+2023-03-14 23:45:21,255 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.643e+02 1.150e+03 1.500e+03 2.092e+03 5.165e+03, threshold=3.000e+03, percent-clipped=6.0
+2023-03-14 23:46:00,057 INFO [train.py:968] (0/2) Epoch 29, batch 19500, giga_loss[loss=0.2386, simple_loss=0.3171, pruned_loss=0.08003, over 28974.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3352, pruned_loss=0.09143, over 5684793.71 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3323, pruned_loss=0.08456, over 5762851.77 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3357, pruned_loss=0.09253, over 5676651.75 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 2.0
+2023-03-14 23:46:06,237 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:46:09,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6088, 1.9761, 1.8292, 1.7453], device='cuda:0'), covar=tensor([0.2480, 0.2276, 0.2548, 0.2320], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0754, 0.0726, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-14 23:46:11,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1295332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:46:39,313 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1295361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:46:46,347 INFO [train.py:968] (0/2) Epoch 29, batch 19550, giga_loss[loss=0.24, simple_loss=0.3189, pruned_loss=0.0805, over 28837.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3305, pruned_loss=0.08961, over 5675385.39 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.332, pruned_loss=0.08442, over 5765984.44 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3311, pruned_loss=0.09079, over 5664087.90 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 2.0
+2023-03-14 23:46:53,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.295e+02 1.055e+03 1.373e+03 1.812e+03 5.268e+03, threshold=2.745e+03, percent-clipped=3.0
+2023-03-14 23:47:10,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:47:29,544 INFO [train.py:968] (0/2) Epoch 29, batch 19600, libri_loss[loss=0.2551, simple_loss=0.3418, pruned_loss=0.08423, over 28637.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3296, pruned_loss=0.08899, over 5666278.76 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3317, pruned_loss=0.08419, over 5771037.61 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3303, pruned_loss=0.09039, over 5647872.33 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:47:47,806 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4719, 1.7938, 1.5316, 1.6522], device='cuda:0'), covar=tensor([0.0767, 0.0324, 0.0342, 0.0870], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-14 23:47:55,568 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-14 23:48:09,839 INFO [train.py:968] (0/2) Epoch 29, batch 19650, giga_loss[loss=0.2249, simple_loss=0.312, pruned_loss=0.06891, over 28628.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3306, pruned_loss=0.08896, over 5674657.53 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.332, pruned_loss=0.08417, over 5772407.78 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3308, pruned_loss=0.09023, over 5655667.86 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:48:15,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.485e+02 1.231e+03 1.597e+03 1.993e+03 5.896e+03, threshold=3.195e+03, percent-clipped=5.0
+2023-03-14 23:48:37,921 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-14 23:48:53,513 INFO [train.py:968] (0/2) Epoch 29, batch 19700, giga_loss[loss=0.2294, simple_loss=0.3054, pruned_loss=0.07665, over 29095.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3312, pruned_loss=0.08967, over 5672482.01 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3326, pruned_loss=0.08451, over 5764107.21 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3308, pruned_loss=0.09039, over 5664369.46 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:48:53,675 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:48:55,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5895, 2.1881, 1.7065, 0.8423], device='cuda:0'), covar=tensor([0.6489, 0.2772, 0.4283, 0.7192], device='cuda:0'), in_proj_covar=tensor([0.1842, 0.1734, 0.1661, 0.1501], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-14 23:49:01,375 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:49:07,554 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:49:09,715 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:49:11,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1295542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:49:34,747 INFO [train.py:968] (0/2) Epoch 29, batch 19750, giga_loss[loss=0.2359, simple_loss=0.3175, pruned_loss=0.07714, over 29058.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3306, pruned_loss=0.08932, over 5681170.16 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3331, pruned_loss=0.08456, over 5766919.63 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3299, pruned_loss=0.08995, over 5670540.79 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:49:34,920 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1295571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:49:40,122 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.848e+02 1.223e+03 1.487e+03 2.313e+03 7.843e+03, threshold=2.974e+03, percent-clipped=10.0
+2023-03-14 23:50:16,112 INFO [train.py:968] (0/2) Epoch 29, batch 19800, giga_loss[loss=0.2353, simple_loss=0.316, pruned_loss=0.07736, over 28846.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3282, pruned_loss=0.08825, over 5682884.30 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3333, pruned_loss=0.0847, over 5763704.03 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3274, pruned_loss=0.08863, over 5677020.07 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:50:32,892 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:50:57,691 INFO [train.py:968] (0/2) Epoch 29, batch 19850, giga_loss[loss=0.3057, simple_loss=0.3539, pruned_loss=0.1287, over 24047.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3256, pruned_loss=0.08691, over 5692646.28 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3339, pruned_loss=0.08497, over 5764813.77 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3244, pruned_loss=0.08699, over 5686331.65 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:51:02,349 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.186e+02 1.187e+03 1.337e+03 1.717e+03 7.758e+03, threshold=2.674e+03, percent-clipped=3.0
+2023-03-14 23:51:03,451 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:51:06,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1295682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:51:28,169 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1295711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:51:34,983 INFO [train.py:968] (0/2) Epoch 29, batch 19900, giga_loss[loss=0.2354, simple_loss=0.3096, pruned_loss=0.08056, over 28581.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3225, pruned_loss=0.08523, over 5690160.20 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3343, pruned_loss=0.08505, over 5754835.81 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.321, pruned_loss=0.08525, over 5691868.67 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:52:13,670 INFO [train.py:968] (0/2) Epoch 29, batch 19950, giga_loss[loss=0.2252, simple_loss=0.2989, pruned_loss=0.07578, over 28768.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3206, pruned_loss=0.08442, over 5696038.61 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3349, pruned_loss=0.08528, over 5747210.32 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3185, pruned_loss=0.08421, over 5703228.45 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:52:18,437 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.634e+02 1.243e+03 1.585e+03 2.069e+03 6.702e+03, threshold=3.169e+03, percent-clipped=12.0
+2023-03-14 23:52:49,242 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:52:52,698 INFO [train.py:968] (0/2) Epoch 29, batch 20000, giga_loss[loss=0.2432, simple_loss=0.3227, pruned_loss=0.08189, over 28755.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3186, pruned_loss=0.08321, over 5708280.21 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3348, pruned_loss=0.08503, over 5750902.88 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3166, pruned_loss=0.08323, over 5709588.45 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:53:08,197 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6596, 1.7166, 1.8506, 1.4347], device='cuda:0'), covar=tensor([0.1913, 0.2659, 0.1557, 0.1743], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0714, 0.0987, 0.0888], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-14 23:53:32,602 INFO [train.py:968] (0/2) Epoch 29, batch 20050, giga_loss[loss=0.2025, simple_loss=0.2877, pruned_loss=0.05869, over 28853.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3195, pruned_loss=0.08374, over 5707206.03 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.336, pruned_loss=0.08555, over 5753185.06 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3163, pruned_loss=0.08325, over 5704894.07 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:53:36,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.175e+02 1.155e+03 1.399e+03 1.690e+03 4.875e+03, threshold=2.797e+03, percent-clipped=2.0
+2023-03-14 23:53:51,135 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:53:56,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:54:09,299 INFO [train.py:968] (0/2) Epoch 29, batch 20100, giga_loss[loss=0.2307, simple_loss=0.3067, pruned_loss=0.07735, over 28376.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3181, pruned_loss=0.08293, over 5713859.66 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3364, pruned_loss=0.08557, over 5756979.71 frames. ], giga_tot_loss[loss=0.2398, simple_loss=0.3147, pruned_loss=0.08246, over 5707544.50 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:54:45,681 INFO [train.py:968] (0/2) Epoch 29, batch 20150, giga_loss[loss=0.2085, simple_loss=0.2861, pruned_loss=0.06541, over 28531.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3167, pruned_loss=0.08189, over 5716521.33 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3366, pruned_loss=0.08566, over 5754898.48 frames. ], giga_tot_loss[loss=0.2383, simple_loss=0.3137, pruned_loss=0.08141, over 5713156.47 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:54:52,690 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.376e+02 1.123e+03 1.282e+03 1.681e+03 4.330e+03, threshold=2.564e+03, percent-clipped=3.0
+2023-03-14 23:55:10,744 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1296000.pt
+2023-03-14 23:55:22,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1296015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:55:28,805 INFO [train.py:968] (0/2) Epoch 29, batch 20200, giga_loss[loss=0.2855, simple_loss=0.3526, pruned_loss=0.1092, over 28961.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3228, pruned_loss=0.0856, over 5712330.14 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3379, pruned_loss=0.08622, over 5755915.92 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3185, pruned_loss=0.0846, over 5706993.25 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:55:43,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1296039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:55:45,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1296042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:55:49,075 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1296048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:55:51,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1296051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:56:13,111 INFO [train.py:968] (0/2) Epoch 29, batch 20250, giga_loss[loss=0.3535, simple_loss=0.4004, pruned_loss=0.1533, over 26598.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3291, pruned_loss=0.0896, over 5705633.73 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3388, pruned_loss=0.08669, over 5759318.18 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3244, pruned_loss=0.08838, over 5696968.63 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:56:13,329 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1296071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:56:19,587 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.566e+02 1.420e+03 1.664e+03 2.167e+03 6.550e+03, threshold=3.329e+03, percent-clipped=18.0
+2023-03-14 23:56:20,510 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1296080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:56:42,161 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9240, 1.1484, 2.8218, 2.6949], device='cuda:0'), covar=tensor([0.1569, 0.2585, 0.0555, 0.1929], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0672, 0.1002, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-14 23:57:00,252 INFO [train.py:968] (0/2) Epoch 29, batch 20300, libri_loss[loss=0.2474, simple_loss=0.3185, pruned_loss=0.08822, over 29346.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3364, pruned_loss=0.09453, over 5704305.89 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3387, pruned_loss=0.08668, over 5761275.27 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3325, pruned_loss=0.09369, over 5694143.35 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:57:12,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6875, 1.9760, 1.2861, 1.4762], device='cuda:0'), covar=tensor([0.1077, 0.0579, 0.1136, 0.1132], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0448, 0.0527, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-14 23:57:32,467 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1296158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:57:34,418 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1296161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:57:42,177 INFO [train.py:968] (0/2) Epoch 29, batch 20350, giga_loss[loss=0.3475, simple_loss=0.396, pruned_loss=0.1495, over 26666.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3413, pruned_loss=0.09723, over 5700095.65 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3392, pruned_loss=0.08709, over 5763335.18 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3377, pruned_loss=0.09661, over 5686899.66 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:57:47,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.065e+02 1.548e+03 1.990e+03 2.872e+03 7.849e+03, threshold=3.980e+03, percent-clipped=17.0
+2023-03-14 23:57:56,218 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1296190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:57:57,571 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1296192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:58:04,226 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6266, 1.8433, 1.4827, 1.7798], device='cuda:0'), covar=tensor([0.2748, 0.2908, 0.3232, 0.2409], device='cuda:0'), in_proj_covar=tensor([0.1610, 0.1161, 0.1424, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-14 23:58:24,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4771, 1.7498, 1.8015, 1.5064], device='cuda:0'), covar=tensor([0.2860, 0.2303, 0.1772, 0.2279], device='cuda:0'), in_proj_covar=tensor([0.2062, 0.2019, 0.1922, 0.2073], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-14 23:58:24,885 INFO [train.py:968] (0/2) Epoch 29, batch 20400, libri_loss[loss=0.2604, simple_loss=0.3322, pruned_loss=0.09427, over 29557.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3452, pruned_loss=0.09863, over 5701349.97 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3394, pruned_loss=0.08741, over 5766240.47 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3422, pruned_loss=0.09817, over 5686196.59 frames. ], batch size: 75, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:58:32,170 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-14 23:59:05,531 INFO [train.py:968] (0/2) Epoch 29, batch 20450, giga_loss[loss=0.3194, simple_loss=0.3963, pruned_loss=0.1212, over 28931.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3487, pruned_loss=0.09963, over 5700698.34 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3397, pruned_loss=0.08771, over 5767773.19 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3463, pruned_loss=0.09936, over 5684903.87 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:59:12,480 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.447e+02 1.287e+03 1.493e+03 1.931e+03 6.527e+03, threshold=2.986e+03, percent-clipped=5.0
+2023-03-14 23:59:49,784 INFO [train.py:968] (0/2) Epoch 29, batch 20500, giga_loss[loss=0.2718, simple_loss=0.3586, pruned_loss=0.09253, over 29126.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3536, pruned_loss=0.1023, over 5705105.40 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3401, pruned_loss=0.08789, over 5767258.71 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3516, pruned_loss=0.1021, over 5691654.80 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:00:01,628 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1296335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:00:03,382 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1296338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:00:28,895 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1296367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:00:31,853 INFO [train.py:968] (0/2) Epoch 29, batch 20550, giga_loss[loss=0.2686, simple_loss=0.347, pruned_loss=0.09504, over 28563.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3513, pruned_loss=0.1007, over 5696395.44 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3403, pruned_loss=0.08813, over 5769402.91 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3497, pruned_loss=0.1006, over 5682824.78 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:00:39,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.936e+02 1.489e+03 1.910e+03 2.986e+03 7.808e+03, threshold=3.820e+03, percent-clipped=25.0
+2023-03-15 00:01:14,672 INFO [train.py:968] (0/2) Epoch 29, batch 20600, giga_loss[loss=0.2431, simple_loss=0.3275, pruned_loss=0.07933, over 28616.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3481, pruned_loss=0.09766, over 5706322.74 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3403, pruned_loss=0.08823, over 5770819.26 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3469, pruned_loss=0.0976, over 5693904.66 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:01:48,268 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3221, 1.7289, 1.4780, 1.4255], device='cuda:0'), covar=tensor([0.0804, 0.0350, 0.0332, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 00:01:58,691 INFO [train.py:968] (0/2) Epoch 29, batch 20650, giga_loss[loss=0.2656, simple_loss=0.3486, pruned_loss=0.0913, over 28666.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3474, pruned_loss=0.09713, over 5693867.69 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3405, pruned_loss=0.08833, over 5771646.16 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3463, pruned_loss=0.09703, over 5683065.49 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:02:04,924 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1296479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:02:05,629 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.991e+02 1.334e+03 1.771e+03 2.285e+03 6.288e+03, threshold=3.542e+03, percent-clipped=6.0
+2023-03-15 00:02:41,499 INFO [train.py:968] (0/2) Epoch 29, batch 20700, giga_loss[loss=0.2536, simple_loss=0.339, pruned_loss=0.08412, over 29011.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3476, pruned_loss=0.09666, over 5698612.40 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3404, pruned_loss=0.0883, over 5773237.24 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3468, pruned_loss=0.09669, over 5688203.69 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:03:15,123 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2392, 1.3648, 1.2725, 1.1164], device='cuda:0'), covar=tensor([0.2628, 0.2810, 0.2107, 0.2520], device='cuda:0'), in_proj_covar=tensor([0.2068, 0.2027, 0.1929, 0.2082], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 00:03:22,867 INFO [train.py:968] (0/2) Epoch 29, batch 20750, giga_loss[loss=0.2651, simple_loss=0.3398, pruned_loss=0.09519, over 28892.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3499, pruned_loss=0.0985, over 5697796.61 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3406, pruned_loss=0.08858, over 5773819.05 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3493, pruned_loss=0.09848, over 5687369.24 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:03:29,363 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.474e+02 1.454e+03 1.975e+03 2.716e+03 6.524e+03, threshold=3.950e+03, percent-clipped=15.0
+2023-03-15 00:03:33,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 00:04:05,298 INFO [train.py:968] (0/2) Epoch 29, batch 20800, giga_loss[loss=0.2851, simple_loss=0.3545, pruned_loss=0.1078, over 28999.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3517, pruned_loss=0.1003, over 5699685.45 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3408, pruned_loss=0.08872, over 5773587.98 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3512, pruned_loss=0.1003, over 5690968.99 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:04:53,289 INFO [train.py:968] (0/2) Epoch 29, batch 20850, giga_loss[loss=0.3767, simple_loss=0.4202, pruned_loss=0.1666, over 28674.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3531, pruned_loss=0.1015, over 5702154.33 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3408, pruned_loss=0.08875, over 5765510.77 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3527, pruned_loss=0.1015, over 5702096.67 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:05:01,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.370e+03 1.677e+03 2.110e+03 8.193e+03, threshold=3.355e+03, percent-clipped=6.0
+2023-03-15 00:05:23,783 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-15 00:05:36,576 INFO [train.py:968] (0/2) Epoch 29, batch 20900, giga_loss[loss=0.2567, simple_loss=0.3318, pruned_loss=0.09074, over 28683.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3541, pruned_loss=0.1028, over 5694816.05 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3412, pruned_loss=0.08922, over 5760146.00 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3537, pruned_loss=0.1027, over 5697401.62 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:06:00,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.7244, 1.1516, 1.1594, 0.9640], device='cuda:0'), covar=tensor([0.2339, 0.1532, 0.2409, 0.1811], device='cuda:0'), in_proj_covar=tensor([0.0507, 0.0758, 0.0731, 0.0696], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 00:06:05,065 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2778, 1.2760, 4.0029, 3.3680], device='cuda:0'), covar=tensor([0.2300, 0.3325, 0.0853, 0.1106], device='cuda:0'), in_proj_covar=tensor([0.0805, 0.0672, 0.1003, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 00:06:13,696 INFO [train.py:968] (0/2) Epoch 29, batch 20950, giga_loss[loss=0.3048, simple_loss=0.3747, pruned_loss=0.1175, over 28703.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3543, pruned_loss=0.1026, over 5700329.95 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3417, pruned_loss=0.08961, over 5755699.90 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.354, pruned_loss=0.1026, over 5704483.02 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:06:21,313 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.170e+02 1.405e+03 1.912e+03 2.791e+03 5.267e+03, threshold=3.824e+03, percent-clipped=13.0
+2023-03-15 00:06:31,670 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-15 00:06:55,620 INFO [train.py:968] (0/2) Epoch 29, batch 21000, giga_loss[loss=0.256, simple_loss=0.3413, pruned_loss=0.08538, over 28525.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3541, pruned_loss=0.1016, over 5690160.60 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3427, pruned_loss=0.09025, over 5745250.16 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3533, pruned_loss=0.1014, over 5701308.18 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:06:55,624 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 00:07:05,058 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.2038, simple_loss=0.3125, pruned_loss=0.04751, over 944034.00 frames. 
+2023-03-15 00:07:05,059 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 00:07:30,699 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1296854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:07:43,045 INFO [train.py:968] (0/2) Epoch 29, batch 21050, giga_loss[loss=0.2887, simple_loss=0.365, pruned_loss=0.1062, over 28623.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.354, pruned_loss=0.1005, over 5702853.67 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3427, pruned_loss=0.09046, over 5748973.52 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3536, pruned_loss=0.1004, over 5707164.05 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:07:51,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.745e+02 1.267e+03 1.679e+03 2.225e+03 6.770e+03, threshold=3.358e+03, percent-clipped=7.0
+2023-03-15 00:08:02,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2578, 1.7876, 1.4470, 0.6060], device='cuda:0'), covar=tensor([0.5476, 0.3243, 0.4883, 0.6470], device='cuda:0'), in_proj_covar=tensor([0.1834, 0.1722, 0.1652, 0.1494], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 00:08:24,738 INFO [train.py:968] (0/2) Epoch 29, batch 21100, giga_loss[loss=0.2726, simple_loss=0.3423, pruned_loss=0.1015, over 28889.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3537, pruned_loss=0.1004, over 5710766.54 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.343, pruned_loss=0.09076, over 5750518.12 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3534, pruned_loss=0.1002, over 5712158.08 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:09:00,974 INFO [train.py:968] (0/2) Epoch 29, batch 21150, libri_loss[loss=0.2601, simple_loss=0.3448, pruned_loss=0.08771, over 29543.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3502, pruned_loss=0.09878, over 5710014.30 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3432, pruned_loss=0.091, over 5753308.97 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.35, pruned_loss=0.09862, over 5707469.95 frames. ], batch size: 83, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:09:05,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4166, 1.6583, 1.4195, 1.5235], device='cuda:0'), covar=tensor([0.0805, 0.0333, 0.0344, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 00:09:08,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.951e+02 1.171e+03 1.353e+03 1.761e+03 4.216e+03, threshold=2.706e+03, percent-clipped=1.0
+2023-03-15 00:09:13,426 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2519, 1.4136, 1.3479, 1.1860], device='cuda:0'), covar=tensor([0.3285, 0.2940, 0.2445, 0.3010], device='cuda:0'), in_proj_covar=tensor([0.2070, 0.2027, 0.1927, 0.2078], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 00:09:22,634 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1296997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:09:24,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1297000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:09:39,660 INFO [train.py:968] (0/2) Epoch 29, batch 21200, giga_loss[loss=0.2628, simple_loss=0.3424, pruned_loss=0.0916, over 28727.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.348, pruned_loss=0.09759, over 5709276.44 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3433, pruned_loss=0.09113, over 5755716.31 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3477, pruned_loss=0.09745, over 5704506.17 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:09:46,113 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1297029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:09:57,571 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1297043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:10:13,989 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4897, 1.7075, 1.4870, 1.5600], device='cuda:0'), covar=tensor([0.0804, 0.0331, 0.0332, 0.0909], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0121, 0.0120, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 00:10:18,643 INFO [train.py:968] (0/2) Epoch 29, batch 21250, giga_loss[loss=0.2836, simple_loss=0.3581, pruned_loss=0.1045, over 28605.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3473, pruned_loss=0.09754, over 5718963.63 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3434, pruned_loss=0.09131, over 5758776.12 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3471, pruned_loss=0.09738, over 5711513.32 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:10:28,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6866, 2.1296, 1.7393, 1.1518], device='cuda:0'), covar=tensor([0.5308, 0.3395, 0.3107, 0.5611], device='cuda:0'), in_proj_covar=tensor([0.1832, 0.1717, 0.1647, 0.1491], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 00:10:29,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.514e+02 1.283e+03 1.517e+03 2.005e+03 6.715e+03, threshold=3.034e+03, percent-clipped=12.0
+2023-03-15 00:11:01,370 INFO [train.py:968] (0/2) Epoch 29, batch 21300, giga_loss[loss=0.2551, simple_loss=0.3242, pruned_loss=0.09305, over 28425.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3478, pruned_loss=0.09812, over 5709508.40 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.344, pruned_loss=0.09197, over 5748826.48 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3472, pruned_loss=0.09754, over 5710732.82 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:11:04,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2119, 1.2504, 3.8273, 3.1874], device='cuda:0'), covar=tensor([0.1837, 0.2965, 0.0454, 0.1162], device='cuda:0'), in_proj_covar=tensor([0.0804, 0.0671, 0.1001, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 00:11:43,298 INFO [train.py:968] (0/2) Epoch 29, batch 21350, giga_loss[loss=0.2878, simple_loss=0.3703, pruned_loss=0.1027, over 28966.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3483, pruned_loss=0.09824, over 5706119.30 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3445, pruned_loss=0.09235, over 5748655.95 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3474, pruned_loss=0.09747, over 5706995.65 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:11:52,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.709e+02 1.231e+03 1.465e+03 2.244e+03 8.876e+03, threshold=2.929e+03, percent-clipped=15.0
+2023-03-15 00:12:11,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4428, 1.7403, 1.4148, 1.2814], device='cuda:0'), covar=tensor([0.2375, 0.2304, 0.2551, 0.2280], device='cuda:0'), in_proj_covar=tensor([0.1614, 0.1164, 0.1425, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 00:12:24,110 INFO [train.py:968] (0/2) Epoch 29, batch 21400, giga_loss[loss=0.2695, simple_loss=0.3462, pruned_loss=0.09638, over 28914.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3485, pruned_loss=0.09767, over 5705539.60 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3444, pruned_loss=0.09236, over 5740725.92 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3479, pruned_loss=0.09711, over 5712632.06 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:13:03,891 INFO [train.py:968] (0/2) Epoch 29, batch 21450, giga_loss[loss=0.2346, simple_loss=0.3296, pruned_loss=0.06975, over 28965.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3472, pruned_loss=0.09713, over 5689151.33 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3449, pruned_loss=0.09304, over 5732492.61 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3463, pruned_loss=0.09622, over 5700547.16 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:13:11,698 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.385e+02 1.191e+03 1.514e+03 2.058e+03 7.833e+03, threshold=3.027e+03, percent-clipped=12.0
+2023-03-15 00:13:43,048 INFO [train.py:968] (0/2) Epoch 29, batch 21500, giga_loss[loss=0.3342, simple_loss=0.3796, pruned_loss=0.1444, over 26636.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3469, pruned_loss=0.09789, over 5689575.01 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3449, pruned_loss=0.09315, over 5736748.20 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3464, pruned_loss=0.09718, over 5693886.01 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:14:12,754 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-15 00:14:20,084 INFO [train.py:968] (0/2) Epoch 29, batch 21550, giga_loss[loss=0.2591, simple_loss=0.3325, pruned_loss=0.0929, over 28345.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3451, pruned_loss=0.09723, over 5692487.53 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3454, pruned_loss=0.09356, over 5731864.70 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3442, pruned_loss=0.09636, over 5698509.77 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:14:25,073 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-15 00:14:29,333 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.643e+02 1.320e+03 1.612e+03 2.101e+03 6.664e+03, threshold=3.224e+03, percent-clipped=13.0
+2023-03-15 00:14:36,039 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 00:14:58,564 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1297418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:15:00,646 INFO [train.py:968] (0/2) Epoch 29, batch 21600, giga_loss[loss=0.2512, simple_loss=0.3237, pruned_loss=0.08936, over 28365.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3407, pruned_loss=0.09467, over 5692134.26 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3457, pruned_loss=0.09388, over 5733644.88 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3396, pruned_loss=0.09374, over 5694818.95 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:15:29,099 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1297456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:15:32,247 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8712, 1.1697, 1.3233, 0.9872], device='cuda:0'), covar=tensor([0.2270, 0.1616, 0.2459, 0.1955], device='cuda:0'), in_proj_covar=tensor([0.0510, 0.0762, 0.0734, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 00:15:40,588 INFO [train.py:968] (0/2) Epoch 29, batch 21650, giga_loss[loss=0.2858, simple_loss=0.352, pruned_loss=0.1098, over 28784.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3408, pruned_loss=0.09533, over 5689739.64 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3457, pruned_loss=0.09388, over 5733644.88 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3399, pruned_loss=0.09461, over 5691829.17 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:15:50,367 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.995e+02 1.237e+03 1.526e+03 1.875e+03 6.393e+03, threshold=3.052e+03, percent-clipped=5.0
+2023-03-15 00:16:03,744 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-15 00:16:17,778 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-15 00:16:20,473 INFO [train.py:968] (0/2) Epoch 29, batch 21700, libri_loss[loss=0.272, simple_loss=0.3506, pruned_loss=0.09668, over 25906.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3397, pruned_loss=0.09494, over 5690178.20 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3462, pruned_loss=0.09438, over 5732054.82 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3384, pruned_loss=0.09393, over 5692239.84 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:16:52,347 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1297561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:16:54,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1297564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:16:59,502 INFO [train.py:968] (0/2) Epoch 29, batch 21750, libri_loss[loss=0.3187, simple_loss=0.3883, pruned_loss=0.1245, over 29639.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3376, pruned_loss=0.09417, over 5700888.71 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.347, pruned_loss=0.09495, over 5737687.88 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3355, pruned_loss=0.09279, over 5695991.18 frames. ], batch size: 91, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:17:08,882 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.082e+02 1.220e+03 1.625e+03 2.220e+03 5.082e+03, threshold=3.251e+03, percent-clipped=8.0
+2023-03-15 00:17:17,198 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1297593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:17:23,777 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2474, 1.5630, 1.6362, 1.3428], device='cuda:0'), covar=tensor([0.2273, 0.1824, 0.2438, 0.2080], device='cuda:0'), in_proj_covar=tensor([0.0510, 0.0761, 0.0733, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 00:17:38,128 INFO [train.py:968] (0/2) Epoch 29, batch 21800, giga_loss[loss=0.2566, simple_loss=0.331, pruned_loss=0.0911, over 28794.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3347, pruned_loss=0.09268, over 5690501.36 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3468, pruned_loss=0.09495, over 5722558.10 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.333, pruned_loss=0.09156, over 5699333.63 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:18:11,217 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1297663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:18:17,251 INFO [train.py:968] (0/2) Epoch 29, batch 21850, giga_loss[loss=0.3736, simple_loss=0.415, pruned_loss=0.166, over 27703.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3323, pruned_loss=0.09142, over 5700512.36 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3471, pruned_loss=0.09521, over 5723205.88 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3305, pruned_loss=0.09027, over 5706629.43 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:18:26,876 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.151e+02 1.171e+03 1.343e+03 1.727e+03 5.445e+03, threshold=2.686e+03, percent-clipped=5.0
+2023-03-15 00:18:57,996 INFO [train.py:968] (0/2) Epoch 29, batch 21900, giga_loss[loss=0.2498, simple_loss=0.3284, pruned_loss=0.08557, over 29071.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3318, pruned_loss=0.09149, over 5704359.00 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3469, pruned_loss=0.09538, over 5726101.84 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3303, pruned_loss=0.09035, over 5706374.02 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:19:27,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2765, 1.4640, 1.3826, 1.2267], device='cuda:0'), covar=tensor([0.3185, 0.2724, 0.2302, 0.2742], device='cuda:0'), in_proj_covar=tensor([0.2074, 0.2034, 0.1932, 0.2082], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 00:19:38,782 INFO [train.py:968] (0/2) Epoch 29, batch 21950, libri_loss[loss=0.3204, simple_loss=0.3883, pruned_loss=0.1262, over 29496.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3345, pruned_loss=0.09218, over 5704340.90 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3479, pruned_loss=0.09618, over 5730138.73 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3321, pruned_loss=0.09045, over 5701669.37 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:19:44,397 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1297779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:19:47,088 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.450e+02 1.204e+03 1.480e+03 1.851e+03 4.750e+03, threshold=2.959e+03, percent-clipped=11.0
+2023-03-15 00:20:20,741 INFO [train.py:968] (0/2) Epoch 29, batch 22000, giga_loss[loss=0.3162, simple_loss=0.3932, pruned_loss=0.1196, over 28647.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.338, pruned_loss=0.09387, over 5698927.41 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3481, pruned_loss=0.09667, over 5731405.06 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3355, pruned_loss=0.0919, over 5695284.28 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:20:28,559 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1297831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:20:41,304 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-15 00:21:02,228 INFO [train.py:968] (0/2) Epoch 29, batch 22050, giga_loss[loss=0.2625, simple_loss=0.3491, pruned_loss=0.08792, over 28803.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.34, pruned_loss=0.09411, over 5700340.97 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.348, pruned_loss=0.09676, over 5734387.60 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3379, pruned_loss=0.09241, over 5693989.30 frames. ], batch size: 243, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:21:11,352 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.989e+02 1.177e+03 1.501e+03 1.944e+03 6.129e+03, threshold=3.003e+03, percent-clipped=5.0
+2023-03-15 00:21:33,979 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1297909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:21:43,229 INFO [train.py:968] (0/2) Epoch 29, batch 22100, giga_loss[loss=0.267, simple_loss=0.3534, pruned_loss=0.0903, over 28695.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3401, pruned_loss=0.093, over 5697844.66 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.348, pruned_loss=0.0968, over 5727128.38 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3384, pruned_loss=0.09159, over 5698139.11 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:22:28,350 INFO [train.py:968] (0/2) Epoch 29, batch 22150, giga_loss[loss=0.2578, simple_loss=0.3396, pruned_loss=0.08793, over 28974.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3389, pruned_loss=0.09197, over 5693219.26 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3485, pruned_loss=0.09738, over 5721762.29 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3369, pruned_loss=0.09028, over 5696982.87 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:22:31,789 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1297974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:22:33,614 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1297977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:22:40,174 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.117e+02 1.306e+03 1.622e+03 2.821e+03 1.380e+04, threshold=3.244e+03, percent-clipped=22.0
+2023-03-15 00:22:52,121 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1298000.pt
+2023-03-15 00:22:57,788 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1298006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:23:10,370 INFO [train.py:968] (0/2) Epoch 29, batch 22200, giga_loss[loss=0.2609, simple_loss=0.3418, pruned_loss=0.09001, over 28699.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3395, pruned_loss=0.09286, over 5692598.61 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3486, pruned_loss=0.09759, over 5721357.62 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3377, pruned_loss=0.09127, over 5695388.25 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:23:23,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1298038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:23:30,901 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5731, 1.6900, 1.7688, 1.3705], device='cuda:0'), covar=tensor([0.1924, 0.2585, 0.1587, 0.1717], device='cuda:0'), in_proj_covar=tensor([0.0941, 0.0719, 0.0989, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 00:23:50,095 INFO [train.py:968] (0/2) Epoch 29, batch 22250, giga_loss[loss=0.3045, simple_loss=0.3708, pruned_loss=0.1191, over 27687.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.34, pruned_loss=0.09362, over 5699171.97 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3484, pruned_loss=0.09749, over 5723854.79 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3385, pruned_loss=0.09234, over 5698547.35 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:23:58,582 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-15 00:24:00,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.695e+02 1.291e+03 1.561e+03 2.042e+03 4.151e+03, threshold=3.122e+03, percent-clipped=6.0
+2023-03-15 00:24:09,131 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-15 00:24:30,056 INFO [train.py:968] (0/2) Epoch 29, batch 22300, libri_loss[loss=0.2672, simple_loss=0.3478, pruned_loss=0.09329, over 27722.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3424, pruned_loss=0.09506, over 5695653.39 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3488, pruned_loss=0.09773, over 5722951.95 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3407, pruned_loss=0.09374, over 5695384.98 frames. ], batch size: 116, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:24:30,443 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-15 00:24:32,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1773, 1.8031, 1.4550, 0.4831], device='cuda:0'), covar=tensor([0.5543, 0.3124, 0.4767, 0.7080], device='cuda:0'), in_proj_covar=tensor([0.1843, 0.1726, 0.1662, 0.1499], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 00:24:56,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3256, 4.1670, 3.9658, 1.6494], device='cuda:0'), covar=tensor([0.0647, 0.0842, 0.0776, 0.2139], device='cuda:0'), in_proj_covar=tensor([0.1283, 0.1186, 0.1000, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 00:24:57,321 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1298154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:25:09,225 INFO [train.py:968] (0/2) Epoch 29, batch 22350, giga_loss[loss=0.3015, simple_loss=0.3712, pruned_loss=0.1159, over 28493.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3461, pruned_loss=0.09729, over 5707336.00 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3497, pruned_loss=0.09859, over 5726501.98 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3439, pruned_loss=0.09541, over 5703382.34 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:25:16,730 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1298181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:25:20,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1298184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:25:21,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.397e+03 1.669e+03 2.104e+03 5.981e+03, threshold=3.337e+03, percent-clipped=7.0
+2023-03-15 00:25:29,670 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-15 00:25:44,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1298213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:25:51,090 INFO [train.py:968] (0/2) Epoch 29, batch 22400, giga_loss[loss=0.275, simple_loss=0.3578, pruned_loss=0.09608, over 28871.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3492, pruned_loss=0.09889, over 5711364.66 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3502, pruned_loss=0.09902, over 5729983.79 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3469, pruned_loss=0.097, over 5704894.44 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:26:27,899 INFO [train.py:968] (0/2) Epoch 29, batch 22450, libri_loss[loss=0.2569, simple_loss=0.3248, pruned_loss=0.09452, over 29663.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3503, pruned_loss=0.09925, over 5720545.97 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3503, pruned_loss=0.09918, over 5735152.35 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3484, pruned_loss=0.09759, over 5710043.55 frames. ], batch size: 73, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:26:36,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1298284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:26:38,551 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.545e+02 1.433e+03 1.698e+03 2.221e+03 4.046e+03, threshold=3.396e+03, percent-clipped=3.0
+2023-03-15 00:26:47,782 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1298297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:26:49,662 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1298300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:27:05,443 INFO [train.py:968] (0/2) Epoch 29, batch 22500, giga_loss[loss=0.3078, simple_loss=0.3765, pruned_loss=0.1196, over 28921.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3507, pruned_loss=0.09944, over 5721843.48 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.351, pruned_loss=0.09977, over 5739649.68 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3485, pruned_loss=0.09753, over 5708705.94 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:27:13,418 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1298329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:27:14,736 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-15 00:27:45,269 INFO [train.py:968] (0/2) Epoch 29, batch 22550, libri_loss[loss=0.303, simple_loss=0.3744, pruned_loss=0.1158, over 29194.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3515, pruned_loss=0.1002, over 5714591.43 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3516, pruned_loss=0.1004, over 5733075.45 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3491, pruned_loss=0.09811, over 5709910.10 frames. ], batch size: 97, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:27:48,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4403, 1.6340, 1.3063, 1.1778], device='cuda:0'), covar=tensor([0.1030, 0.0576, 0.1072, 0.1198], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0449, 0.0526, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 00:27:55,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.302e+02 1.505e+03 1.865e+03 2.658e+03 7.682e+03, threshold=3.729e+03, percent-clipped=16.0
+2023-03-15 00:28:26,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1298420.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 00:28:27,269 INFO [train.py:968] (0/2) Epoch 29, batch 22600, giga_loss[loss=0.2821, simple_loss=0.3545, pruned_loss=0.1048, over 28356.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3486, pruned_loss=0.09862, over 5715819.13 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3514, pruned_loss=0.1003, over 5734547.24 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3469, pruned_loss=0.09701, over 5710770.08 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:28:31,212 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1298427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:28:34,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1298430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:28:58,626 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1298459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:29:08,722 INFO [train.py:968] (0/2) Epoch 29, batch 22650, giga_loss[loss=0.2363, simple_loss=0.3059, pruned_loss=0.08331, over 29046.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3473, pruned_loss=0.09829, over 5722902.26 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3524, pruned_loss=0.1011, over 5737378.35 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.345, pruned_loss=0.09626, over 5715762.62 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:29:18,493 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.222e+02 1.257e+03 1.579e+03 2.014e+03 6.064e+03, threshold=3.158e+03, percent-clipped=4.0
+2023-03-15 00:29:46,285 INFO [train.py:968] (0/2) Epoch 29, batch 22700, giga_loss[loss=0.2376, simple_loss=0.3141, pruned_loss=0.08055, over 28899.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.345, pruned_loss=0.09733, over 5715725.04 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3529, pruned_loss=0.1014, over 5734407.98 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3426, pruned_loss=0.09534, over 5712310.64 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:30:04,409 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4507, 1.7477, 1.3852, 1.5353], device='cuda:0'), covar=tensor([0.0769, 0.0291, 0.0361, 0.0923], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 00:30:27,483 INFO [train.py:968] (0/2) Epoch 29, batch 22750, giga_loss[loss=0.2666, simple_loss=0.3458, pruned_loss=0.09364, over 28832.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3433, pruned_loss=0.09535, over 5716275.45 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3527, pruned_loss=0.1014, over 5735312.94 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3415, pruned_loss=0.0938, over 5712617.68 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:30:41,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.220e+03 1.512e+03 2.080e+03 5.111e+03, threshold=3.023e+03, percent-clipped=2.0
+2023-03-15 00:31:09,836 INFO [train.py:968] (0/2) Epoch 29, batch 22800, giga_loss[loss=0.2282, simple_loss=0.3034, pruned_loss=0.07644, over 28417.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3446, pruned_loss=0.09516, over 5707690.78 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3529, pruned_loss=0.1016, over 5730570.89 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3428, pruned_loss=0.09355, over 5708836.10 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:31:46,243 INFO [train.py:968] (0/2) Epoch 29, batch 22850, giga_loss[loss=0.2632, simple_loss=0.3409, pruned_loss=0.09279, over 28307.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3451, pruned_loss=0.09564, over 5715745.59 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3534, pruned_loss=0.1023, over 5731533.13 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3429, pruned_loss=0.09354, over 5715066.99 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:31:57,965 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.987e+02 1.292e+03 1.667e+03 2.297e+03 6.037e+03, threshold=3.335e+03, percent-clipped=10.0
+2023-03-15 00:32:03,810 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1495, 3.3099, 2.2216, 1.1872], device='cuda:0'), covar=tensor([0.9143, 0.3128, 0.4083, 0.8470], device='cuda:0'), in_proj_covar=tensor([0.1848, 0.1730, 0.1664, 0.1502], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 00:32:28,196 INFO [train.py:968] (0/2) Epoch 29, batch 22900, giga_loss[loss=0.2856, simple_loss=0.3521, pruned_loss=0.1096, over 28870.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3441, pruned_loss=0.09577, over 5712983.67 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3536, pruned_loss=0.1027, over 5725963.52 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3421, pruned_loss=0.09366, over 5717070.09 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:32:38,598 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-15 00:33:04,985 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1298767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:33:07,396 INFO [train.py:968] (0/2) Epoch 29, batch 22950, giga_loss[loss=0.2577, simple_loss=0.3334, pruned_loss=0.09097, over 28734.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3436, pruned_loss=0.09718, over 5719340.92 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3536, pruned_loss=0.1028, over 5730481.56 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3417, pruned_loss=0.09518, over 5718081.25 frames. ], batch size: 243, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:33:18,626 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1579, 1.5868, 1.6101, 1.3955], device='cuda:0'), covar=tensor([0.1829, 0.1404, 0.2082, 0.1641], device='cuda:0'), in_proj_covar=tensor([0.0509, 0.0761, 0.0734, 0.0698], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 00:33:18,762 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-15 00:33:18,988 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.811e+02 1.357e+03 1.658e+03 2.222e+03 7.456e+03, threshold=3.316e+03, percent-clipped=10.0
+2023-03-15 00:33:26,772 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1298795.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 00:33:47,085 INFO [train.py:968] (0/2) Epoch 29, batch 23000, giga_loss[loss=0.3024, simple_loss=0.3617, pruned_loss=0.1216, over 28735.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3421, pruned_loss=0.0975, over 5721386.61 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3534, pruned_loss=0.1029, over 5737053.86 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3403, pruned_loss=0.09563, over 5713963.86 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:33:54,682 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2611, 1.2099, 1.2109, 1.4175], device='cuda:0'), covar=tensor([0.0760, 0.0376, 0.0360, 0.0893], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-15 00:34:06,504 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7114, 1.9980, 1.5744, 1.7797], device='cuda:0'), covar=tensor([0.2964, 0.2938, 0.3441, 0.2879], device='cuda:0'), in_proj_covar=tensor([0.1607, 0.1159, 0.1421, 0.1012], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 00:34:23,985 INFO [train.py:968] (0/2) Epoch 29, batch 23050, giga_loss[loss=0.2489, simple_loss=0.3217, pruned_loss=0.08806, over 28897.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3415, pruned_loss=0.09782, over 5718987.18 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3543, pruned_loss=0.1036, over 5731399.17 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3389, pruned_loss=0.09548, over 5717982.37 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:34:34,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.899e+02 1.366e+03 1.694e+03 2.331e+03 8.292e+03, threshold=3.387e+03, percent-clipped=12.0
+2023-03-15 00:34:57,588 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1298917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:35:00,816 INFO [train.py:968] (0/2) Epoch 29, batch 23100, giga_loss[loss=0.2429, simple_loss=0.3186, pruned_loss=0.08362, over 28950.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3404, pruned_loss=0.09765, over 5703603.30 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3548, pruned_loss=0.1041, over 5721275.15 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3375, pruned_loss=0.0952, over 5711224.73 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:35:13,537 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1298938.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 00:35:16,345 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1298941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:35:16,413 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1298941.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 00:35:25,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9329, 1.9513, 4.0430, 3.6810], device='cuda:0'), covar=tensor([0.1213, 0.2247, 0.0436, 0.1093], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0673, 0.1009, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 00:35:37,511 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1298970.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 00:35:38,016 INFO [train.py:968] (0/2) Epoch 29, batch 23150, giga_loss[loss=0.2195, simple_loss=0.296, pruned_loss=0.07148, over 29021.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3351, pruned_loss=0.09461, over 5705052.79 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3549, pruned_loss=0.1042, over 5713568.35 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3324, pruned_loss=0.0924, over 5717945.39 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:35:44,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1298981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:35:45,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7112, 2.0789, 2.0089, 1.5115], device='cuda:0'), covar=tensor([0.3990, 0.2719, 0.2926, 0.3882], device='cuda:0'), in_proj_covar=tensor([0.2083, 0.2039, 0.1936, 0.2084], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 00:35:48,402 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.764e+02 1.370e+03 1.689e+03 2.594e+03 6.258e+03, threshold=3.379e+03, percent-clipped=12.0
+2023-03-15 00:36:16,794 INFO [train.py:968] (0/2) Epoch 29, batch 23200, giga_loss[loss=0.2079, simple_loss=0.2822, pruned_loss=0.06679, over 28449.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3311, pruned_loss=0.09261, over 5701654.31 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.355, pruned_loss=0.1046, over 5711955.77 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.328, pruned_loss=0.08998, over 5713879.40 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:36:39,462 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1299052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:36:54,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4134, 2.1247, 1.5946, 0.7936], device='cuda:0'), covar=tensor([0.7398, 0.3661, 0.5218, 0.7508], device='cuda:0'), in_proj_covar=tensor([0.1848, 0.1731, 0.1663, 0.1503], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 00:36:54,378 INFO [train.py:968] (0/2) Epoch 29, batch 23250, giga_loss[loss=0.2327, simple_loss=0.3113, pruned_loss=0.07701, over 29141.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3281, pruned_loss=0.09066, over 5712786.03 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.355, pruned_loss=0.1046, over 5713862.41 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3255, pruned_loss=0.08849, over 5720650.79 frames. ], batch size: 113, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:37:06,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.918e+02 1.257e+03 1.556e+03 2.025e+03 4.886e+03, threshold=3.111e+03, percent-clipped=6.0
+2023-03-15 00:37:20,074 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 00:37:34,828 INFO [train.py:968] (0/2) Epoch 29, batch 23300, giga_loss[loss=0.2579, simple_loss=0.3357, pruned_loss=0.09011, over 29059.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3323, pruned_loss=0.09276, over 5698963.73 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3552, pruned_loss=0.105, over 5705698.62 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3292, pruned_loss=0.09028, over 5711960.17 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:37:51,731 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:38:12,764 INFO [train.py:968] (0/2) Epoch 29, batch 23350, giga_loss[loss=0.2534, simple_loss=0.3338, pruned_loss=0.08655, over 28667.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3358, pruned_loss=0.09452, over 5690899.80 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.355, pruned_loss=0.105, over 5699194.88 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3327, pruned_loss=0.09209, over 5707576.97 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:38:24,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.410e+03 1.781e+03 2.361e+03 7.594e+03, threshold=3.563e+03, percent-clipped=16.0
+2023-03-15 00:38:51,739 INFO [train.py:968] (0/2) Epoch 29, batch 23400, libri_loss[loss=0.2776, simple_loss=0.3418, pruned_loss=0.1067, over 29594.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.339, pruned_loss=0.09554, over 5702335.56 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3551, pruned_loss=0.1053, over 5704198.48 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3361, pruned_loss=0.09316, over 5711079.71 frames. ], batch size: 75, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:39:08,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3141, 1.5631, 1.3999, 1.5558], device='cuda:0'), covar=tensor([0.0764, 0.0329, 0.0344, 0.0892], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:0')
+2023-03-15 00:39:31,333 INFO [train.py:968] (0/2) Epoch 29, batch 23450, giga_loss[loss=0.3119, simple_loss=0.3926, pruned_loss=0.1156, over 28702.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3423, pruned_loss=0.09689, over 5708560.10 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3557, pruned_loss=0.1057, over 5708243.56 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.339, pruned_loss=0.09427, over 5711936.26 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:39:34,044 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-15 00:39:42,393 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1299285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:39:43,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.932e+02 1.352e+03 1.668e+03 2.158e+03 6.657e+03, threshold=3.335e+03, percent-clipped=6.0
+2023-03-15 00:39:44,482 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1299288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:39:48,108 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:40:10,658 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:40:11,223 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1299317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:40:13,601 INFO [train.py:968] (0/2) Epoch 29, batch 23500, giga_loss[loss=0.2513, simple_loss=0.3328, pruned_loss=0.08488, over 28541.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3454, pruned_loss=0.09862, over 5714214.38 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3562, pruned_loss=0.1061, over 5709706.15 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.342, pruned_loss=0.09594, over 5715702.55 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:40:42,506 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:40:57,779 INFO [train.py:968] (0/2) Epoch 29, batch 23550, giga_loss[loss=0.2924, simple_loss=0.3523, pruned_loss=0.1163, over 28563.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3488, pruned_loss=0.1016, over 5697874.71 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3565, pruned_loss=0.1065, over 5700028.48 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3456, pruned_loss=0.09905, over 5707830.89 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:41:13,007 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.423e+02 1.471e+03 1.833e+03 2.891e+03 6.853e+03, threshold=3.665e+03, percent-clipped=18.0
+2023-03-15 00:41:29,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4786, 1.7915, 1.4277, 1.3736], device='cuda:0'), covar=tensor([0.2744, 0.2808, 0.3270, 0.2550], device='cuda:0'), in_proj_covar=tensor([0.1612, 0.1164, 0.1425, 0.1016], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 00:41:34,254 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3593, 2.8169, 1.4737, 1.5011], device='cuda:0'), covar=tensor([0.0938, 0.0409, 0.0899, 0.1289], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0572, 0.0412, 0.0446], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 00:41:45,164 INFO [train.py:968] (0/2) Epoch 29, batch 23600, giga_loss[loss=0.3282, simple_loss=0.3958, pruned_loss=0.1303, over 28863.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3548, pruned_loss=0.1061, over 5686282.34 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3572, pruned_loss=0.107, over 5689640.42 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3515, pruned_loss=0.1035, over 5702869.81 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:41:54,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:41:59,655 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1299435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:42:01,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1299438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:42:22,218 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1299459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:42:24,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1299462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:42:28,318 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1299467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:42:32,867 INFO [train.py:968] (0/2) Epoch 29, batch 23650, giga_loss[loss=0.2984, simple_loss=0.3741, pruned_loss=0.1113, over 28607.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3623, pruned_loss=0.1117, over 5677324.66 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3579, pruned_loss=0.1077, over 5693001.75 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.359, pruned_loss=0.1091, over 5687419.24 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:42:51,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.177e+03 1.844e+03 2.158e+03 2.685e+03 6.307e+03, threshold=4.317e+03, percent-clipped=11.0
+2023-03-15 00:42:54,654 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1299491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:43:03,601 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1299499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:43:05,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1299502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:43:06,293 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1299503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:43:20,653 INFO [train.py:968] (0/2) Epoch 29, batch 23700, giga_loss[loss=0.292, simple_loss=0.3615, pruned_loss=0.1112, over 28942.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3676, pruned_loss=0.1161, over 5674664.84 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3581, pruned_loss=0.1078, over 5697471.41 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.365, pruned_loss=0.1141, over 5678045.93 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:43:30,303 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1299531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:43:48,003 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-15 00:44:07,357 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1299570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:44:07,733 INFO [train.py:968] (0/2) Epoch 29, batch 23750, giga_loss[loss=0.3434, simple_loss=0.3992, pruned_loss=0.1438, over 28957.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3736, pruned_loss=0.1213, over 5672675.97 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3586, pruned_loss=0.1082, over 5692288.74 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3715, pruned_loss=0.1196, over 5678883.47 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:44:10,086 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1299573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:44:26,245 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.235e+03 1.802e+03 2.290e+03 3.149e+03 6.459e+03, threshold=4.581e+03, percent-clipped=8.0
+2023-03-15 00:44:38,526 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1299602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:44:57,563 INFO [train.py:968] (0/2) Epoch 29, batch 23800, libri_loss[loss=0.3301, simple_loss=0.3918, pruned_loss=0.1342, over 29302.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.38, pruned_loss=0.1266, over 5672978.97 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3588, pruned_loss=0.1085, over 5695313.11 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3785, pruned_loss=0.1253, over 5674682.26 frames. ], batch size: 94, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:45:43,284 INFO [train.py:968] (0/2) Epoch 29, batch 23850, giga_loss[loss=0.3905, simple_loss=0.4201, pruned_loss=0.1805, over 23548.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3802, pruned_loss=0.1276, over 5668329.13 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3585, pruned_loss=0.1084, over 5701412.82 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3797, pruned_loss=0.1271, over 5663772.58 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:45:44,597 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1299672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:46:00,871 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.178e+03 2.103e+03 2.604e+03 3.440e+03 8.003e+03, threshold=5.208e+03, percent-clipped=11.0
+2023-03-15 00:46:32,900 INFO [train.py:968] (0/2) Epoch 29, batch 23900, giga_loss[loss=0.3169, simple_loss=0.3832, pruned_loss=0.1253, over 28942.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3835, pruned_loss=0.1314, over 5663182.93 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.359, pruned_loss=0.1087, over 5704718.10 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3833, pruned_loss=0.1313, over 5655668.95 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:47:15,901 INFO [train.py:968] (0/2) Epoch 29, batch 23950, giga_loss[loss=0.4218, simple_loss=0.4402, pruned_loss=0.2017, over 23458.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3856, pruned_loss=0.1345, over 5657188.22 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3582, pruned_loss=0.1086, over 5711027.92 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3878, pruned_loss=0.1361, over 5642410.43 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:47:29,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.336e+03 1.961e+03 2.777e+03 3.720e+03 6.347e+03, threshold=5.553e+03, percent-clipped=6.0
+2023-03-15 00:48:07,153 INFO [train.py:968] (0/2) Epoch 29, batch 24000, giga_loss[loss=0.2982, simple_loss=0.3736, pruned_loss=0.1114, over 28877.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3898, pruned_loss=0.138, over 5651175.17 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3581, pruned_loss=0.1087, over 5707712.91 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3925, pruned_loss=0.14, over 5640303.48 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:48:07,158 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 00:48:15,434 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.2002, simple_loss=0.3074, pruned_loss=0.04648, over 944034.00 frames. 
+2023-03-15 00:48:15,435 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 00:48:28,364 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5138, 1.7463, 1.2262, 1.2668], device='cuda:0'), covar=tensor([0.1192, 0.0701, 0.1198, 0.1285], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0453, 0.0528, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 00:49:03,991 INFO [train.py:968] (0/2) Epoch 29, batch 24050, giga_loss[loss=0.4331, simple_loss=0.4409, pruned_loss=0.2126, over 23687.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3894, pruned_loss=0.138, over 5657215.71 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3579, pruned_loss=0.1089, over 5713073.59 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3929, pruned_loss=0.1404, over 5641939.92 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:49:11,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:49:23,237 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.204e+03 2.140e+03 2.832e+03 3.828e+03 1.094e+04, threshold=5.664e+03, percent-clipped=4.0
+2023-03-15 00:49:39,076 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-15 00:49:53,173 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4852, 1.8927, 1.6912, 1.7681], device='cuda:0'), covar=tensor([0.0795, 0.0288, 0.0305, 0.0850], device='cuda:0'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 00:49:54,227 INFO [train.py:968] (0/2) Epoch 29, batch 24100, giga_loss[loss=0.3067, simple_loss=0.3759, pruned_loss=0.1187, over 29043.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3882, pruned_loss=0.1375, over 5648752.51 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3582, pruned_loss=0.1091, over 5714874.22 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3913, pruned_loss=0.1397, over 5633748.77 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:50:41,236 INFO [train.py:968] (0/2) Epoch 29, batch 24150, giga_loss[loss=0.2609, simple_loss=0.3381, pruned_loss=0.09182, over 29023.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3861, pruned_loss=0.1356, over 5655403.46 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3582, pruned_loss=0.109, over 5719288.24 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3891, pruned_loss=0.138, over 5638433.59 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:50:55,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3459, 1.5072, 1.4180, 1.2939], device='cuda:0'), covar=tensor([0.2537, 0.2430, 0.2166, 0.2407], device='cuda:0'), in_proj_covar=tensor([0.2098, 0.2057, 0.1955, 0.2105], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 00:50:59,200 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.306e+03 1.923e+03 2.507e+03 3.320e+03 8.933e+03, threshold=5.014e+03, percent-clipped=4.0
+2023-03-15 00:50:59,752 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.33 vs. limit=5.0
+2023-03-15 00:51:06,953 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1300000.pt
+2023-03-15 00:51:21,920 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7143, 1.6006, 1.9329, 1.5378], device='cuda:0'), covar=tensor([0.1465, 0.2041, 0.1192, 0.1540], device='cuda:0'), in_proj_covar=tensor([0.0931, 0.0715, 0.0979, 0.0880], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 00:51:27,783 INFO [train.py:968] (0/2) Epoch 29, batch 24200, giga_loss[loss=0.2882, simple_loss=0.3638, pruned_loss=0.1063, over 28692.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.386, pruned_loss=0.1346, over 5648438.44 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3586, pruned_loss=0.1093, over 5715120.95 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3889, pruned_loss=0.1369, over 5637755.98 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:51:28,105 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:51:32,631 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:51:56,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:51:59,797 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:52:06,693 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:52:17,004 INFO [train.py:968] (0/2) Epoch 29, batch 24250, libri_loss[loss=0.2838, simple_loss=0.3601, pruned_loss=0.1037, over 29529.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3862, pruned_loss=0.1345, over 5636805.06 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3586, pruned_loss=0.1095, over 5711173.05 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3896, pruned_loss=0.1372, over 5629033.84 frames. ], batch size: 84, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:52:34,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.324e+03 2.057e+03 2.668e+03 3.404e+03 7.574e+03, threshold=5.335e+03, percent-clipped=5.0
+2023-03-15 00:53:11,714 INFO [train.py:968] (0/2) Epoch 29, batch 24300, giga_loss[loss=0.2783, simple_loss=0.3505, pruned_loss=0.103, over 28879.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3866, pruned_loss=0.1347, over 5630239.43 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3579, pruned_loss=0.1093, over 5714184.57 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3902, pruned_loss=0.1374, over 5620474.01 frames. ], batch size: 285, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:53:21,160 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:53:27,654 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 00:54:02,085 INFO [train.py:968] (0/2) Epoch 29, batch 24350, giga_loss[loss=0.3354, simple_loss=0.4088, pruned_loss=0.131, over 28744.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3831, pruned_loss=0.1308, over 5626047.31 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3583, pruned_loss=0.1096, over 5704663.40 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3861, pruned_loss=0.133, over 5625867.29 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:54:17,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.225e+03 1.708e+03 2.371e+03 3.078e+03 7.687e+03, threshold=4.742e+03, percent-clipped=6.0
+2023-03-15 00:54:19,006 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:54:20,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:54:45,472 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5254, 3.0711, 1.5252, 1.6923], device='cuda:0'), covar=tensor([0.0966, 0.0471, 0.0892, 0.1302], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0575, 0.0414, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 00:54:50,584 INFO [train.py:968] (0/2) Epoch 29, batch 24400, giga_loss[loss=0.2944, simple_loss=0.3724, pruned_loss=0.1082, over 28882.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3797, pruned_loss=0.1268, over 5635542.56 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3581, pruned_loss=0.1094, over 5705932.13 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3825, pruned_loss=0.1289, over 5633374.78 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:54:52,858 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:55:25,441 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.46 vs. limit=5.0
+2023-03-15 00:55:37,934 INFO [train.py:968] (0/2) Epoch 29, batch 24450, giga_loss[loss=0.3415, simple_loss=0.4003, pruned_loss=0.1413, over 28731.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3765, pruned_loss=0.1239, over 5655503.88 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3579, pruned_loss=0.1093, over 5707192.04 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3791, pruned_loss=0.1259, over 5651715.74 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:55:56,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.131e+03 1.925e+03 2.296e+03 2.844e+03 6.996e+03, threshold=4.592e+03, percent-clipped=10.0
+2023-03-15 00:56:18,560 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:56:22,194 INFO [train.py:968] (0/2) Epoch 29, batch 24500, giga_loss[loss=0.2841, simple_loss=0.3572, pruned_loss=0.1055, over 28800.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3731, pruned_loss=0.1215, over 5651341.49 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3576, pruned_loss=0.1093, over 5703958.21 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3762, pruned_loss=0.1236, over 5648930.32 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:56:33,270 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:57:06,529 INFO [train.py:968] (0/2) Epoch 29, batch 24550, giga_loss[loss=0.3047, simple_loss=0.3744, pruned_loss=0.1175, over 28877.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3711, pruned_loss=0.1201, over 5664969.61 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3573, pruned_loss=0.1092, over 5709491.32 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3743, pruned_loss=0.1221, over 5656782.20 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:57:22,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.274e+03 1.919e+03 2.671e+03 4.370e+03 1.426e+04, threshold=5.342e+03, percent-clipped=20.0
+2023-03-15 00:57:54,604 INFO [train.py:968] (0/2) Epoch 29, batch 24600, giga_loss[loss=0.3432, simple_loss=0.4016, pruned_loss=0.1424, over 28618.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3722, pruned_loss=0.121, over 5663798.39 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3573, pruned_loss=0.1092, over 5711864.49 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3751, pruned_loss=0.1231, over 5654042.32 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:58:10,946 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:58:19,978 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:58:24,419 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7933, 2.0589, 2.0904, 1.7605], device='cuda:0'), covar=tensor([0.3513, 0.2875, 0.2893, 0.3227], device='cuda:0'), in_proj_covar=tensor([0.2091, 0.2052, 0.1947, 0.2099], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 00:58:34,399 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-15 00:58:45,936 INFO [train.py:968] (0/2) Epoch 29, batch 24650, giga_loss[loss=0.2689, simple_loss=0.3471, pruned_loss=0.0953, over 28137.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3704, pruned_loss=0.1193, over 5670625.55 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3574, pruned_loss=0.1092, over 5713424.91 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3728, pruned_loss=0.121, over 5661060.51 frames. ], batch size: 77, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:59:07,112 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.179e+03 1.781e+03 2.325e+03 3.210e+03 8.811e+03, threshold=4.649e+03, percent-clipped=5.0
+2023-03-15 00:59:23,724 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:59:36,913 INFO [train.py:968] (0/2) Epoch 29, batch 24700, giga_loss[loss=0.2657, simple_loss=0.3427, pruned_loss=0.09438, over 28740.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3701, pruned_loss=0.1166, over 5682063.81 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3566, pruned_loss=0.1089, over 5716224.15 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3728, pruned_loss=0.1184, over 5671606.30 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:59:59,185 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-15 01:00:27,717 INFO [train.py:968] (0/2) Epoch 29, batch 24750, giga_loss[loss=0.3091, simple_loss=0.3813, pruned_loss=0.1185, over 28930.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3709, pruned_loss=0.1162, over 5663561.96 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3565, pruned_loss=0.1089, over 5717166.95 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3735, pruned_loss=0.1178, over 5653366.44 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:00:35,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:00:38,962 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:00:43,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1763, 0.8038, 0.9008, 1.4226], device='cuda:0'), covar=tensor([0.0804, 0.0404, 0.0379, 0.0903], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 01:00:47,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.148e+03 1.781e+03 2.205e+03 2.941e+03 8.147e+03, threshold=4.409e+03, percent-clipped=7.0
+2023-03-15 01:01:03,424 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:01:12,465 INFO [train.py:968] (0/2) Epoch 29, batch 24800, giga_loss[loss=0.3211, simple_loss=0.3802, pruned_loss=0.131, over 28919.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3701, pruned_loss=0.1161, over 5669689.66 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3555, pruned_loss=0.1084, over 5716524.89 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.374, pruned_loss=0.1182, over 5659614.25 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:01:42,812 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:01:44,774 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:02:01,482 INFO [train.py:968] (0/2) Epoch 29, batch 24850, giga_loss[loss=0.3316, simple_loss=0.3945, pruned_loss=0.1344, over 28473.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3714, pruned_loss=0.1176, over 5662315.54 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3558, pruned_loss=0.1086, over 5719600.36 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3745, pruned_loss=0.1193, over 5650705.90 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:02:11,864 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:02:21,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:02:21,758 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.237e+03 1.835e+03 2.191e+03 3.492e+03 7.659e+03, threshold=4.381e+03, percent-clipped=8.0
+2023-03-15 01:02:38,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:02:40,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:02:49,548 INFO [train.py:968] (0/2) Epoch 29, batch 24900, giga_loss[loss=0.2884, simple_loss=0.3503, pruned_loss=0.1132, over 28933.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.37, pruned_loss=0.1177, over 5659316.37 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3559, pruned_loss=0.1087, over 5720630.77 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3727, pruned_loss=0.1192, over 5647566.38 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:03:22,855 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9422, 1.4598, 1.3791, 1.3385], device='cuda:0'), covar=tensor([0.2447, 0.1736, 0.2396, 0.1940], device='cuda:0'), in_proj_covar=tensor([0.0510, 0.0762, 0.0735, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 01:03:25,489 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2364, 1.8271, 1.3847, 0.4781], device='cuda:0'), covar=tensor([0.5380, 0.3476, 0.4859, 0.7272], device='cuda:0'), in_proj_covar=tensor([0.1854, 0.1740, 0.1664, 0.1510], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 01:03:31,841 INFO [train.py:968] (0/2) Epoch 29, batch 24950, giga_loss[loss=0.3158, simple_loss=0.379, pruned_loss=0.1263, over 28884.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3689, pruned_loss=0.1176, over 5668402.50 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3563, pruned_loss=0.1089, over 5717657.31 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3711, pruned_loss=0.1188, over 5659433.05 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:03:48,456 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 2.025e+03 2.534e+03 3.487e+03 1.350e+04, threshold=5.068e+03, percent-clipped=13.0
+2023-03-15 01:04:17,633 INFO [train.py:968] (0/2) Epoch 29, batch 25000, giga_loss[loss=0.297, simple_loss=0.3711, pruned_loss=0.1115, over 28986.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3677, pruned_loss=0.1171, over 5673278.57 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.356, pruned_loss=0.1087, over 5723916.53 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3703, pruned_loss=0.1187, over 5658236.88 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:04:17,929 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:04:28,427 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:04:30,692 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:04:43,712 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:04:46,673 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:04:55,094 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:05:03,602 INFO [train.py:968] (0/2) Epoch 29, batch 25050, giga_loss[loss=0.3011, simple_loss=0.3521, pruned_loss=0.125, over 23784.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3682, pruned_loss=0.1162, over 5680678.38 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3566, pruned_loss=0.1091, over 5725758.81 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3698, pruned_loss=0.1172, over 5666821.43 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:05:12,288 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:05:12,355 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6158, 1.8582, 1.4985, 1.5884], device='cuda:0'), covar=tensor([0.0767, 0.0298, 0.0339, 0.0912], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 01:05:20,733 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.750e+03 2.098e+03 2.794e+03 6.135e+03, threshold=4.196e+03, percent-clipped=6.0
+2023-03-15 01:05:50,605 INFO [train.py:968] (0/2) Epoch 29, batch 25100, giga_loss[loss=0.2973, simple_loss=0.3731, pruned_loss=0.1108, over 28651.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3687, pruned_loss=0.1166, over 5669783.54 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3568, pruned_loss=0.1094, over 5727725.51 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.37, pruned_loss=0.1172, over 5656648.53 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:06:11,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-15 01:06:30,247 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:06:33,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:06:37,077 INFO [train.py:968] (0/2) Epoch 29, batch 25150, giga_loss[loss=0.3135, simple_loss=0.38, pruned_loss=0.1235, over 28763.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3674, pruned_loss=0.1153, over 5682385.36 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3562, pruned_loss=0.1091, over 5732014.80 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3693, pruned_loss=0.1162, over 5666903.76 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:06:58,100 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.240e+03 1.703e+03 2.122e+03 3.034e+03 9.353e+03, threshold=4.244e+03, percent-clipped=10.0
+2023-03-15 01:07:02,507 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:07:25,597 INFO [train.py:968] (0/2) Epoch 29, batch 25200, giga_loss[loss=0.2997, simple_loss=0.3685, pruned_loss=0.1155, over 28242.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3673, pruned_loss=0.1159, over 5690432.75 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3566, pruned_loss=0.1093, over 5732697.13 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3686, pruned_loss=0.1165, over 5676620.51 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:07:50,102 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2516, 1.5520, 1.5470, 1.1279], device='cuda:0'), covar=tensor([0.1728, 0.2589, 0.1456, 0.1678], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0719, 0.0985, 0.0885], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 01:07:56,631 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1301054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:08:10,450 INFO [train.py:968] (0/2) Epoch 29, batch 25250, giga_loss[loss=0.3043, simple_loss=0.3683, pruned_loss=0.1202, over 27987.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3659, pruned_loss=0.1157, over 5691090.61 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3564, pruned_loss=0.1094, over 5738671.52 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3675, pruned_loss=0.1164, over 5672869.72 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:08:17,220 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3160, 1.5654, 1.5366, 1.2723], device='cuda:0'), covar=tensor([0.3275, 0.2875, 0.2236, 0.3026], device='cuda:0'), in_proj_covar=tensor([0.2102, 0.2065, 0.1963, 0.2113], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 01:08:22,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1301085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:08:28,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.214e+03 1.770e+03 2.359e+03 3.164e+03 7.846e+03, threshold=4.718e+03, percent-clipped=10.0
+2023-03-15 01:08:54,911 INFO [train.py:968] (0/2) Epoch 29, batch 25300, giga_loss[loss=0.3063, simple_loss=0.3697, pruned_loss=0.1215, over 28855.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.365, pruned_loss=0.1156, over 5694677.89 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3566, pruned_loss=0.1095, over 5732732.78 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3665, pruned_loss=0.1163, over 5684275.80 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:09:39,609 INFO [train.py:968] (0/2) Epoch 29, batch 25350, giga_loss[loss=0.3803, simple_loss=0.3984, pruned_loss=0.1811, over 23534.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3645, pruned_loss=0.1157, over 5691389.15 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3569, pruned_loss=0.1098, over 5731088.59 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3656, pruned_loss=0.1162, over 5683438.25 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:09:48,782 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1301179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:09:48,889 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6780, 1.8815, 1.5505, 1.9291], device='cuda:0'), covar=tensor([0.2697, 0.2949, 0.3226, 0.2589], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1166, 0.1430, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 01:10:01,024 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.329e+03 1.886e+03 2.643e+03 3.597e+03 5.740e+03, threshold=5.286e+03, percent-clipped=7.0
+2023-03-15 01:10:11,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-15 01:10:27,275 INFO [train.py:968] (0/2) Epoch 29, batch 25400, libri_loss[loss=0.3201, simple_loss=0.3782, pruned_loss=0.131, over 26143.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1157, over 5688486.23 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3571, pruned_loss=0.1099, over 5729988.67 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3639, pruned_loss=0.116, over 5682936.80 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:10:33,430 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8369, 2.1320, 1.7136, 2.0633], device='cuda:0'), covar=tensor([0.2839, 0.2746, 0.3236, 0.2326], device='cuda:0'), in_proj_covar=tensor([0.1614, 0.1164, 0.1429, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 01:10:34,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1301228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:10:36,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1301231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:11:04,771 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1301260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:11:15,019 INFO [train.py:968] (0/2) Epoch 29, batch 25450, libri_loss[loss=0.306, simple_loss=0.3702, pruned_loss=0.1209, over 25683.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3623, pruned_loss=0.1156, over 5682754.25 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3571, pruned_loss=0.1099, over 5731079.49 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3631, pruned_loss=0.116, over 5676715.57 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:11:31,744 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.397e+03 1.900e+03 2.391e+03 2.884e+03 7.530e+03, threshold=4.783e+03, percent-clipped=6.0
+2023-03-15 01:11:57,099 INFO [train.py:968] (0/2) Epoch 29, batch 25500, libri_loss[loss=0.2567, simple_loss=0.3235, pruned_loss=0.09493, over 29332.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3626, pruned_loss=0.1152, over 5682927.55 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3567, pruned_loss=0.1099, over 5727494.64 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.364, pruned_loss=0.1158, over 5678486.45 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:12:11,715 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9860, 3.8544, 3.6609, 1.8921], device='cuda:0'), covar=tensor([0.0680, 0.0785, 0.0781, 0.2269], device='cuda:0'), in_proj_covar=tensor([0.1325, 0.1221, 0.1026, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 01:12:39,848 INFO [train.py:968] (0/2) Epoch 29, batch 25550, giga_loss[loss=0.3554, simple_loss=0.3826, pruned_loss=0.1641, over 23679.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3631, pruned_loss=0.1147, over 5677914.75 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3567, pruned_loss=0.11, over 5720064.26 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3642, pruned_loss=0.1151, over 5680794.00 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:12:42,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4835, 2.0161, 1.5774, 1.7480], device='cuda:0'), covar=tensor([0.0782, 0.0288, 0.0329, 0.0860], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 01:12:58,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.138e+03 1.691e+03 2.152e+03 2.781e+03 7.027e+03, threshold=4.304e+03, percent-clipped=5.0
+2023-03-15 01:13:21,277 INFO [train.py:968] (0/2) Epoch 29, batch 25600, libri_loss[loss=0.2867, simple_loss=0.3587, pruned_loss=0.1073, over 29521.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3623, pruned_loss=0.1133, over 5691648.74 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3568, pruned_loss=0.1102, over 5727027.93 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3634, pruned_loss=0.1135, over 5686162.08 frames. ], batch size: 82, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 01:13:29,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1301429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:13:54,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2001, 1.4668, 0.9888, 1.0131], device='cuda:0'), covar=tensor([0.1143, 0.0572, 0.1380, 0.1297], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0453, 0.0527, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 01:14:06,462 INFO [train.py:968] (0/2) Epoch 29, batch 25650, giga_loss[loss=0.2792, simple_loss=0.3527, pruned_loss=0.1029, over 29054.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.363, pruned_loss=0.114, over 5687919.44 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3567, pruned_loss=0.1103, over 5730624.15 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3639, pruned_loss=0.1143, over 5679641.76 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:14:28,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.795e+03 2.295e+03 3.008e+03 6.847e+03, threshold=4.589e+03, percent-clipped=8.0
+2023-03-15 01:14:57,226 INFO [train.py:968] (0/2) Epoch 29, batch 25700, giga_loss[loss=0.3192, simple_loss=0.3747, pruned_loss=0.1318, over 28769.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3664, pruned_loss=0.117, over 5679836.63 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3569, pruned_loss=0.1104, over 5722714.70 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3671, pruned_loss=0.1172, over 5680359.20 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:15:06,779 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1301529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:15:26,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4863, 1.7026, 1.6953, 1.4926], device='cuda:0'), covar=tensor([0.2368, 0.2427, 0.2474, 0.2437], device='cuda:0'), in_proj_covar=tensor([0.0509, 0.0761, 0.0735, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 01:15:28,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1301554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:15:47,283 INFO [train.py:968] (0/2) Epoch 29, batch 25750, giga_loss[loss=0.4034, simple_loss=0.4217, pruned_loss=0.1926, over 26686.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3689, pruned_loss=0.1206, over 5678734.26 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3565, pruned_loss=0.1101, over 5726582.26 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.37, pruned_loss=0.1211, over 5674949.26 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:15:48,191 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1301572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:15:50,290 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1301575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:16:06,284 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.200e+03 1.989e+03 2.461e+03 3.115e+03 9.324e+03, threshold=4.921e+03, percent-clipped=9.0
+2023-03-15 01:16:12,538 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 01:16:15,259 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1301604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:16:32,884 INFO [train.py:968] (0/2) Epoch 29, batch 25800, giga_loss[loss=0.3292, simple_loss=0.3856, pruned_loss=0.1364, over 28238.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3686, pruned_loss=0.1215, over 5680354.68 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3561, pruned_loss=0.1099, over 5732022.86 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3703, pruned_loss=0.1224, over 5671012.11 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:17:11,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3426, 1.2248, 3.7454, 3.2441], device='cuda:0'), covar=tensor([0.1643, 0.2846, 0.0490, 0.1233], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0680, 0.1019, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 01:17:16,938 INFO [train.py:968] (0/2) Epoch 29, batch 25850, giga_loss[loss=0.35, simple_loss=0.3952, pruned_loss=0.1523, over 28309.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3703, pruned_loss=0.1227, over 5676588.95 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3565, pruned_loss=0.1101, over 5720696.89 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3718, pruned_loss=0.1238, over 5677857.69 frames. ], batch size: 77, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:17:28,593 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4693, 3.3978, 1.5433, 1.5743], device='cuda:0'), covar=tensor([0.1014, 0.0402, 0.0919, 0.1382], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0577, 0.0414, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 01:17:35,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3550, 1.6385, 1.3400, 0.9250], device='cuda:0'), covar=tensor([0.2619, 0.2773, 0.3159, 0.2542], device='cuda:0'), in_proj_covar=tensor([0.1616, 0.1164, 0.1430, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 01:17:36,120 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.413e+03 1.869e+03 2.384e+03 3.142e+03 2.043e+04, threshold=4.769e+03, percent-clipped=9.0
+2023-03-15 01:17:38,948 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1301697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:17:42,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1301700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:17:58,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2816, 1.2011, 3.5532, 3.1490], device='cuda:0'), covar=tensor([0.1591, 0.2855, 0.0504, 0.1727], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0681, 0.1020, 0.0995], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 01:18:00,422 INFO [train.py:968] (0/2) Epoch 29, batch 25900, giga_loss[loss=0.3231, simple_loss=0.3795, pruned_loss=0.1334, over 28573.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.369, pruned_loss=0.122, over 5672335.66 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3568, pruned_loss=0.1103, over 5726389.39 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3702, pruned_loss=0.123, over 5667048.75 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:18:02,933 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3716, 1.1975, 1.1439, 1.4695], device='cuda:0'), covar=tensor([0.0788, 0.0376, 0.0360, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 01:18:06,967 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1301729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:18:42,904 INFO [train.py:968] (0/2) Epoch 29, batch 25950, giga_loss[loss=0.302, simple_loss=0.388, pruned_loss=0.108, over 29001.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3685, pruned_loss=0.1211, over 5652194.71 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.357, pruned_loss=0.1106, over 5699361.98 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3698, pruned_loss=0.1221, over 5670186.48 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:19:02,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.080e+03 1.932e+03 2.484e+03 3.532e+03 6.365e+03, threshold=4.968e+03, percent-clipped=12.0
+2023-03-15 01:19:28,114 INFO [train.py:968] (0/2) Epoch 29, batch 26000, giga_loss[loss=0.2787, simple_loss=0.3476, pruned_loss=0.1049, over 28635.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3677, pruned_loss=0.119, over 5654631.36 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.357, pruned_loss=0.1106, over 5701166.74 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3689, pruned_loss=0.12, over 5666799.40 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:20:16,756 INFO [train.py:968] (0/2) Epoch 29, batch 26050, giga_loss[loss=0.2825, simple_loss=0.3524, pruned_loss=0.1063, over 28484.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3658, pruned_loss=0.1177, over 5654401.65 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3571, pruned_loss=0.1106, over 5702185.91 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3667, pruned_loss=0.1184, over 5662940.56 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:20:40,498 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-15 01:20:40,659 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.265e+03 1.768e+03 2.184e+03 2.985e+03 5.998e+03, threshold=4.368e+03, percent-clipped=2.0
+2023-03-15 01:20:48,741 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1301904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:21:04,937 INFO [train.py:968] (0/2) Epoch 29, batch 26100, giga_loss[loss=0.2931, simple_loss=0.3638, pruned_loss=0.1112, over 28694.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3633, pruned_loss=0.1163, over 5663585.20 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3571, pruned_loss=0.1105, over 5704217.88 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3641, pruned_loss=0.1171, over 5668066.63 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:21:18,625 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-15 01:21:53,697 INFO [train.py:968] (0/2) Epoch 29, batch 26150, giga_loss[loss=0.3337, simple_loss=0.387, pruned_loss=0.1402, over 27919.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3633, pruned_loss=0.1172, over 5655643.30 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3567, pruned_loss=0.1102, over 5710847.56 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3644, pruned_loss=0.1182, over 5652125.23 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:22:16,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.779e+03 2.406e+03 3.225e+03 6.815e+03, threshold=4.811e+03, percent-clipped=10.0
+2023-03-15 01:22:21,107 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1302000.pt
+2023-03-15 01:22:41,362 INFO [train.py:968] (0/2) Epoch 29, batch 26200, giga_loss[loss=0.4138, simple_loss=0.4406, pruned_loss=0.1935, over 26591.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3662, pruned_loss=0.1188, over 5650647.99 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3567, pruned_loss=0.1101, over 5703368.44 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3672, pruned_loss=0.1198, over 5653123.74 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:23:03,473 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1302047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:23:07,193 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1302050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:23:25,425 INFO [train.py:968] (0/2) Epoch 29, batch 26250, giga_loss[loss=0.2936, simple_loss=0.3797, pruned_loss=0.1037, over 28683.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3711, pruned_loss=0.1195, over 5653419.76 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3574, pruned_loss=0.1106, over 5699065.95 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3716, pruned_loss=0.1201, over 5657658.52 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:23:33,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1302079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:23:47,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.169e+03 1.663e+03 2.022e+03 2.864e+03 8.344e+03, threshold=4.044e+03, percent-clipped=2.0
+2023-03-15 01:24:10,064 INFO [train.py:968] (0/2) Epoch 29, batch 26300, giga_loss[loss=0.2958, simple_loss=0.3647, pruned_loss=0.1134, over 28850.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3725, pruned_loss=0.1192, over 5661383.68 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3574, pruned_loss=0.1107, over 5704013.76 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3733, pruned_loss=0.1197, over 5659190.94 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:24:55,812 INFO [train.py:968] (0/2) Epoch 29, batch 26350, giga_loss[loss=0.3239, simple_loss=0.3854, pruned_loss=0.1312, over 28730.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3735, pruned_loss=0.1205, over 5659331.27 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3577, pruned_loss=0.1112, over 5702074.82 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3746, pruned_loss=0.121, over 5656464.49 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:25:16,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.188e+03 1.808e+03 2.232e+03 3.115e+03 9.988e+03, threshold=4.464e+03, percent-clipped=7.0
+2023-03-15 01:25:39,124 INFO [train.py:968] (0/2) Epoch 29, batch 26400, giga_loss[loss=0.3447, simple_loss=0.4079, pruned_loss=0.1407, over 28881.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.375, pruned_loss=0.1223, over 5655205.58 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3574, pruned_loss=0.111, over 5703958.41 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3766, pruned_loss=0.1231, over 5650160.84 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 01:26:11,904 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1302255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:26:26,821 INFO [train.py:968] (0/2) Epoch 29, batch 26450, giga_loss[loss=0.2798, simple_loss=0.3523, pruned_loss=0.1037, over 28546.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3761, pruned_loss=0.1241, over 5641564.85 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.358, pruned_loss=0.1113, over 5698702.76 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3772, pruned_loss=0.1247, over 5641623.45 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:26:35,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8848, 2.6265, 1.7404, 1.0365], device='cuda:0'), covar=tensor([0.8395, 0.3995, 0.4143, 0.7823], device='cuda:0'), in_proj_covar=tensor([0.1855, 0.1750, 0.1663, 0.1512], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 01:26:46,494 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.216e+03 1.932e+03 2.408e+03 3.298e+03 1.331e+04, threshold=4.817e+03, percent-clipped=13.0
+2023-03-15 01:27:12,824 INFO [train.py:968] (0/2) Epoch 29, batch 26500, giga_loss[loss=0.2772, simple_loss=0.342, pruned_loss=0.1062, over 28728.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.375, pruned_loss=0.1238, over 5636663.73 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3583, pruned_loss=0.1115, over 5694331.03 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3761, pruned_loss=0.1245, over 5638735.70 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:27:27,985 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3339, 3.1864, 3.0126, 1.4820], device='cuda:0'), covar=tensor([0.0952, 0.1055, 0.0923, 0.2212], device='cuda:0'), in_proj_covar=tensor([0.1331, 0.1228, 0.1034, 0.0766], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 01:27:58,170 INFO [train.py:968] (0/2) Epoch 29, batch 26550, libri_loss[loss=0.2518, simple_loss=0.3178, pruned_loss=0.09293, over 29369.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3712, pruned_loss=0.1217, over 5646530.37 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3579, pruned_loss=0.1112, over 5697572.87 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3727, pruned_loss=0.1227, over 5644307.15 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:28:13,676 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3684, 1.2908, 3.6880, 3.2255], device='cuda:0'), covar=tensor([0.1613, 0.2763, 0.0490, 0.1245], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0684, 0.1024, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 01:28:17,140 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1302394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:28:18,239 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.165e+03 1.925e+03 2.613e+03 3.504e+03 9.574e+03, threshold=5.227e+03, percent-clipped=8.0
+2023-03-15 01:28:48,218 INFO [train.py:968] (0/2) Epoch 29, batch 26600, giga_loss[loss=0.2755, simple_loss=0.3412, pruned_loss=0.1049, over 28577.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3695, pruned_loss=0.1213, over 5642878.41 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3577, pruned_loss=0.1111, over 5693043.87 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3711, pruned_loss=0.1225, over 5643887.38 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:28:55,804 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7213, 1.6832, 1.9185, 1.5180], device='cuda:0'), covar=tensor([0.1767, 0.2417, 0.1421, 0.1722], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0722, 0.0986, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 01:29:32,243 INFO [train.py:968] (0/2) Epoch 29, batch 26650, libri_loss[loss=0.2615, simple_loss=0.335, pruned_loss=0.09403, over 29509.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3686, pruned_loss=0.121, over 5638849.48 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3576, pruned_loss=0.1111, over 5690823.16 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5639272.86 frames. ], batch size: 81, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:29:54,393 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.882e+03 2.274e+03 3.085e+03 6.370e+03, threshold=4.549e+03, percent-clipped=2.0
+2023-03-15 01:30:16,878 INFO [train.py:968] (0/2) Epoch 29, batch 26700, giga_loss[loss=0.2362, simple_loss=0.3182, pruned_loss=0.07705, over 28950.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3684, pruned_loss=0.1211, over 5640662.09 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3579, pruned_loss=0.1114, over 5683261.38 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3697, pruned_loss=0.122, over 5647168.04 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:30:41,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5079, 1.7372, 1.6407, 1.4022], device='cuda:0'), covar=tensor([0.3282, 0.2821, 0.2501, 0.2874], device='cuda:0'), in_proj_covar=tensor([0.2084, 0.2057, 0.1958, 0.2102], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 01:31:00,217 INFO [train.py:968] (0/2) Epoch 29, batch 26750, giga_loss[loss=0.3057, simple_loss=0.3669, pruned_loss=0.1222, over 28186.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3658, pruned_loss=0.1195, over 5657706.89 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3573, pruned_loss=0.1111, over 5685931.55 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3676, pruned_loss=0.1208, over 5659179.04 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:31:00,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1463, 1.3505, 3.3536, 3.0666], device='cuda:0'), covar=tensor([0.1680, 0.2653, 0.0526, 0.1061], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0685, 0.1024, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 01:31:25,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.152e+03 1.798e+03 2.206e+03 3.328e+03 7.065e+03, threshold=4.413e+03, percent-clipped=8.0
+2023-03-15 01:31:48,312 INFO [train.py:968] (0/2) Epoch 29, batch 26800, giga_loss[loss=0.3031, simple_loss=0.3816, pruned_loss=0.1124, over 28997.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3664, pruned_loss=0.1198, over 5653400.89 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3576, pruned_loss=0.1114, over 5676594.43 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3677, pruned_loss=0.1207, over 5663323.67 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:31:58,256 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1302630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:32:00,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5467, 1.6154, 1.7075, 1.3358], device='cuda:0'), covar=tensor([0.1727, 0.2615, 0.1452, 0.1750], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0723, 0.0987, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 01:32:09,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3501, 3.1784, 3.0179, 1.4723], device='cuda:0'), covar=tensor([0.0981, 0.1107, 0.0972, 0.2188], device='cuda:0'), in_proj_covar=tensor([0.1330, 0.1228, 0.1033, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 01:32:34,355 INFO [train.py:968] (0/2) Epoch 29, batch 26850, giga_loss[loss=0.3049, simple_loss=0.3738, pruned_loss=0.118, over 28620.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3686, pruned_loss=0.1206, over 5654356.13 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3581, pruned_loss=0.1117, over 5678964.48 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3694, pruned_loss=0.1212, over 5659869.14 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:32:58,259 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.900e+03 2.310e+03 3.291e+03 8.186e+03, threshold=4.621e+03, percent-clipped=15.0
+2023-03-15 01:33:21,746 INFO [train.py:968] (0/2) Epoch 29, batch 26900, giga_loss[loss=0.3532, simple_loss=0.4001, pruned_loss=0.1531, over 28308.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3716, pruned_loss=0.1223, over 5652114.23 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3581, pruned_loss=0.1117, over 5673458.39 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3724, pruned_loss=0.1229, over 5660855.85 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:34:07,988 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1302769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:34:09,090 INFO [train.py:968] (0/2) Epoch 29, batch 26950, giga_loss[loss=0.3207, simple_loss=0.3764, pruned_loss=0.1325, over 28911.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3711, pruned_loss=0.123, over 5651160.25 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3574, pruned_loss=0.1113, over 5677554.19 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3726, pruned_loss=0.1241, over 5653929.44 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:34:11,103 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1302773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:34:13,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1302776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:34:13,542 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-15 01:34:29,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2369, 4.0389, 3.8134, 1.7902], device='cuda:0'), covar=tensor([0.0760, 0.0871, 0.0881, 0.2300], device='cuda:0'), in_proj_covar=tensor([0.1332, 0.1229, 0.1033, 0.0766], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 01:34:31,289 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.728e+03 2.293e+03 2.991e+03 5.849e+03, threshold=4.585e+03, percent-clipped=3.0
+2023-03-15 01:34:38,807 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1302805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:34:52,572 INFO [train.py:968] (0/2) Epoch 29, batch 27000, giga_loss[loss=0.2803, simple_loss=0.3637, pruned_loss=0.09838, over 28907.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.372, pruned_loss=0.1202, over 5665838.39 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3575, pruned_loss=0.1113, over 5677129.96 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3733, pruned_loss=0.1212, over 5668390.20 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:34:52,578 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 01:34:58,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1994, 1.2719, 3.4359, 3.1631], device='cuda:0'), covar=tensor([0.1980, 0.3113, 0.0556, 0.0945], device='cuda:0'), in_proj_covar=tensor([0.0817, 0.0682, 0.1021, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 01:35:02,465 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.2026, simple_loss=0.3106, pruned_loss=0.0473, over 944034.00 frames. 
+2023-03-15 01:35:02,466 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 01:35:51,330 INFO [train.py:968] (0/2) Epoch 29, batch 27050, giga_loss[loss=0.2853, simple_loss=0.3649, pruned_loss=0.1028, over 28917.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3724, pruned_loss=0.1188, over 5660837.74 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3575, pruned_loss=0.1112, over 5678481.99 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3735, pruned_loss=0.1197, over 5661641.87 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:36:12,269 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.452e+03 1.838e+03 2.450e+03 4.439e+03, threshold=3.676e+03, percent-clipped=0.0
+2023-03-15 01:36:25,070 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1302912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:36:27,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1302915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:36:32,528 INFO [train.py:968] (0/2) Epoch 29, batch 27100, giga_loss[loss=0.3441, simple_loss=0.4, pruned_loss=0.1441, over 27922.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3738, pruned_loss=0.1189, over 5653442.87 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3574, pruned_loss=0.1113, over 5663970.05 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3751, pruned_loss=0.1197, over 5666276.58 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:36:53,398 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1302944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:37:19,308 INFO [train.py:968] (0/2) Epoch 29, batch 27150, giga_loss[loss=0.2945, simple_loss=0.364, pruned_loss=0.1125, over 28933.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3768, pruned_loss=0.1221, over 5658772.07 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3574, pruned_loss=0.1114, over 5670048.70 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3782, pruned_loss=0.1229, over 5663443.93 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:37:41,472 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-15 01:37:45,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.785e+03 2.528e+03 3.929e+03 7.011e+03, threshold=5.055e+03, percent-clipped=27.0
+2023-03-15 01:38:05,578 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-15 01:38:09,338 INFO [train.py:968] (0/2) Epoch 29, batch 27200, giga_loss[loss=0.3408, simple_loss=0.3965, pruned_loss=0.1425, over 28603.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.377, pruned_loss=0.1229, over 5671111.78 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3575, pruned_loss=0.1113, over 5673249.63 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3784, pruned_loss=0.1237, over 5671968.43 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:38:32,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:38:40,115 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5790, 1.7160, 1.7591, 1.3362], device='cuda:0'), covar=tensor([0.1542, 0.2744, 0.1390, 0.1723], device='cuda:0'), in_proj_covar=tensor([0.0933, 0.0720, 0.0983, 0.0883], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 01:38:59,018 INFO [train.py:968] (0/2) Epoch 29, batch 27250, giga_loss[loss=0.2968, simple_loss=0.3665, pruned_loss=0.1135, over 28851.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3771, pruned_loss=0.1238, over 5673765.83 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3579, pruned_loss=0.1118, over 5678677.11 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3783, pruned_loss=0.1244, over 5669598.90 frames. ], batch size: 285, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:39:06,419 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-15 01:39:26,078 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.297e+03 1.872e+03 2.210e+03 3.099e+03 7.720e+03, threshold=4.421e+03, percent-clipped=3.0
+2023-03-15 01:39:45,817 INFO [train.py:968] (0/2) Epoch 29, batch 27300, giga_loss[loss=0.3354, simple_loss=0.4045, pruned_loss=0.1332, over 28698.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3748, pruned_loss=0.1217, over 5677136.72 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3577, pruned_loss=0.1117, over 5680315.56 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3762, pruned_loss=0.1224, over 5672358.35 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:40:25,089 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:40:28,762 INFO [train.py:968] (0/2) Epoch 29, batch 27350, libri_loss[loss=0.2911, simple_loss=0.362, pruned_loss=0.1101, over 29641.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3735, pruned_loss=0.1191, over 5665102.78 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3574, pruned_loss=0.1116, over 5674963.44 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3754, pruned_loss=0.1199, over 5665676.63 frames. ], batch size: 91, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:40:32,340 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-15 01:40:56,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.094e+03 1.604e+03 2.082e+03 3.002e+03 1.010e+04, threshold=4.164e+03, percent-clipped=6.0
+2023-03-15 01:41:09,866 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:41:15,801 INFO [train.py:968] (0/2) Epoch 29, batch 27400, libri_loss[loss=0.3034, simple_loss=0.3729, pruned_loss=0.117, over 29542.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3742, pruned_loss=0.1192, over 5663224.60 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3576, pruned_loss=0.1117, over 5679542.65 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.376, pruned_loss=0.1201, over 5658989.02 frames. ], batch size: 83, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:41:33,441 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:42:05,729 INFO [train.py:968] (0/2) Epoch 29, batch 27450, giga_loss[loss=0.3224, simple_loss=0.3808, pruned_loss=0.132, over 28552.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3765, pruned_loss=0.121, over 5654267.16 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3581, pruned_loss=0.1119, over 5675504.47 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.378, pruned_loss=0.1217, over 5653756.34 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:42:32,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.893e+03 2.880e+03 3.819e+03 9.900e+03, threshold=5.759e+03, percent-clipped=20.0
+2023-03-15 01:42:50,586 INFO [train.py:968] (0/2) Epoch 29, batch 27500, giga_loss[loss=0.2891, simple_loss=0.3623, pruned_loss=0.1079, over 28750.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3747, pruned_loss=0.12, over 5659948.48 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3584, pruned_loss=0.112, over 5681617.17 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3761, pruned_loss=0.1207, over 5653477.90 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:43:02,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0705, 4.6813, 2.0852, 2.3010], device='cuda:0'), covar=tensor([0.0834, 0.0373, 0.0788, 0.1089], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0578, 0.0415, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 01:43:25,441 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 01:43:40,826 INFO [train.py:968] (0/2) Epoch 29, batch 27550, giga_loss[loss=0.3477, simple_loss=0.4024, pruned_loss=0.1465, over 28831.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3733, pruned_loss=0.12, over 5660924.48 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3584, pruned_loss=0.1121, over 5680970.59 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3747, pruned_loss=0.1207, over 5655859.97 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:43:45,742 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303376.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:43:59,162 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3031, 1.2222, 1.2040, 1.4751], device='cuda:0'), covar=tensor([0.0714, 0.0430, 0.0347, 0.0781], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 01:44:05,736 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.291e+03 1.819e+03 2.241e+03 2.983e+03 4.614e+03, threshold=4.482e+03, percent-clipped=0.0
+2023-03-15 01:44:27,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:44:27,557 INFO [train.py:968] (0/2) Epoch 29, batch 27600, libri_loss[loss=0.3038, simple_loss=0.371, pruned_loss=0.1183, over 29144.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3712, pruned_loss=0.1191, over 5673750.04 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3588, pruned_loss=0.1123, over 5687826.68 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3724, pruned_loss=0.1196, over 5663136.66 frames. ], batch size: 101, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:45:16,452 INFO [train.py:968] (0/2) Epoch 29, batch 27650, giga_loss[loss=0.3106, simple_loss=0.3726, pruned_loss=0.1242, over 28535.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3694, pruned_loss=0.1184, over 5671109.97 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1125, over 5686474.94 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3702, pruned_loss=0.1188, over 5663554.78 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:45:38,793 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:45:41,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.242e+03 1.804e+03 2.353e+03 3.272e+03 8.067e+03, threshold=4.706e+03, percent-clipped=8.0
+2023-03-15 01:46:02,383 INFO [train.py:968] (0/2) Epoch 29, batch 27700, libri_loss[loss=0.2654, simple_loss=0.3315, pruned_loss=0.09965, over 29571.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3685, pruned_loss=0.1187, over 5672452.02 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5694105.59 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3692, pruned_loss=0.1191, over 5658674.59 frames. ], batch size: 75, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:46:19,887 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:46:38,575 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1303563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:46:41,398 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1303566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:46:46,167 INFO [train.py:968] (0/2) Epoch 29, batch 27750, giga_loss[loss=0.2847, simple_loss=0.3355, pruned_loss=0.117, over 23727.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3685, pruned_loss=0.1196, over 5660576.55 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3597, pruned_loss=0.1127, over 5688335.74 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3693, pruned_loss=0.1201, over 5653636.84 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:46:59,583 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:47:06,884 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1303595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:47:08,549 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.728e+02 2.038e+03 2.594e+03 3.448e+03 9.662e+03, threshold=5.189e+03, percent-clipped=7.0
+2023-03-15 01:47:22,689 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:47:31,659 INFO [train.py:968] (0/2) Epoch 29, batch 27800, giga_loss[loss=0.2669, simple_loss=0.3486, pruned_loss=0.09255, over 28313.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3664, pruned_loss=0.1171, over 5668503.62 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3599, pruned_loss=0.1129, over 5692703.93 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3669, pruned_loss=0.1175, over 5658876.05 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:48:12,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:48:17,377 INFO [train.py:968] (0/2) Epoch 29, batch 27850, giga_loss[loss=0.259, simple_loss=0.339, pruned_loss=0.08947, over 28798.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3633, pruned_loss=0.1143, over 5670433.64 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3599, pruned_loss=0.1129, over 5695594.59 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3638, pruned_loss=0.1146, over 5659856.38 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:48:30,077 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1303683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:48:32,346 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1303686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:48:38,407 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303691.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:48:42,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.170e+03 1.572e+03 2.269e+03 3.086e+03 1.489e+04, threshold=4.538e+03, percent-clipped=9.0
+2023-03-15 01:49:02,298 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1303715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:06,381 INFO [train.py:968] (0/2) Epoch 29, batch 27900, giga_loss[loss=0.2886, simple_loss=0.3534, pruned_loss=0.1119, over 28291.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3615, pruned_loss=0.1128, over 5661150.87 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3599, pruned_loss=0.1128, over 5696826.23 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3619, pruned_loss=0.1132, over 5650634.55 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:49:08,693 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4418, 1.6183, 1.5980, 1.3774], device='cuda:0'), covar=tensor([0.3090, 0.2892, 0.2364, 0.2849], device='cuda:0'), in_proj_covar=tensor([0.2092, 0.2065, 0.1956, 0.2108], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 01:49:15,490 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1303730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:18,837 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1303733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:36,340 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303751.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 01:49:43,170 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1303757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:46,894 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1303760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:48,603 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1303762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:56,758 INFO [train.py:968] (0/2) Epoch 29, batch 27950, giga_loss[loss=0.2673, simple_loss=0.3347, pruned_loss=0.09994, over 28792.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.359, pruned_loss=0.1119, over 5662857.21 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3597, pruned_loss=0.1126, over 5699138.58 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3596, pruned_loss=0.1123, over 5651789.86 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:50:08,849 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.43 vs. limit=5.0
+2023-03-15 01:50:19,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1303789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:50:26,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.196e+03 1.686e+03 2.285e+03 3.444e+03 9.120e+03, threshold=4.569e+03, percent-clipped=9.0
+2023-03-15 01:50:27,699 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:50:30,846 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.48 vs. limit=5.0
+2023-03-15 01:50:51,237 INFO [train.py:968] (0/2) Epoch 29, batch 28000, giga_loss[loss=0.2741, simple_loss=0.3445, pruned_loss=0.1018, over 28727.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3563, pruned_loss=0.1107, over 5657147.28 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3594, pruned_loss=0.1123, over 5693150.62 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3569, pruned_loss=0.1112, over 5652361.66 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 01:51:05,997 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6068, 1.8001, 1.4478, 1.7506], device='cuda:0'), covar=tensor([0.2630, 0.2865, 0.3114, 0.2702], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1164, 0.1430, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 01:51:32,922 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:51:34,132 INFO [train.py:968] (0/2) Epoch 29, batch 28050, libri_loss[loss=0.2984, simple_loss=0.3667, pruned_loss=0.115, over 29518.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.358, pruned_loss=0.1117, over 5661933.43 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.359, pruned_loss=0.1121, over 5697754.17 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3588, pruned_loss=0.1123, over 5653084.59 frames. ], batch size: 81, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 01:51:43,550 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3524, 2.5472, 1.3603, 1.5786], device='cuda:0'), covar=tensor([0.0954, 0.0420, 0.0815, 0.1237], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0578, 0.0414, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 01:51:45,722 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-15 01:51:58,220 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1303894.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:52:01,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1303897.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:52:01,386 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5578, 1.5534, 1.7866, 1.3340], device='cuda:0'), covar=tensor([0.1828, 0.3067, 0.1544, 0.1977], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0724, 0.0989, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 01:52:01,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.820e+03 2.286e+03 3.163e+03 8.417e+03, threshold=4.572e+03, percent-clipped=7.0
+2023-03-15 01:52:23,021 INFO [train.py:968] (0/2) Epoch 29, batch 28100, giga_loss[loss=0.2644, simple_loss=0.3424, pruned_loss=0.09316, over 28789.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3613, pruned_loss=0.1135, over 5652655.28 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1117, over 5701508.64 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3623, pruned_loss=0.1143, over 5641810.80 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:52:23,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3412, 1.5982, 1.5182, 1.4636], device='cuda:0'), covar=tensor([0.2106, 0.1932, 0.2421, 0.1897], device='cuda:0'), in_proj_covar=tensor([0.0511, 0.0764, 0.0736, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 01:52:28,686 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1303926.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 01:52:50,773 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1613, 1.6300, 1.6224, 1.3209], device='cuda:0'), covar=tensor([0.2016, 0.1408, 0.2058, 0.1811], device='cuda:0'), in_proj_covar=tensor([0.0511, 0.0764, 0.0737, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 01:52:52,815 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:52:54,891 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3356, 1.7345, 1.3653, 1.3982], device='cuda:0'), covar=tensor([0.2605, 0.2526, 0.2832, 0.2315], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1164, 0.1431, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 01:53:07,665 INFO [train.py:968] (0/2) Epoch 29, batch 28150, giga_loss[loss=0.2628, simple_loss=0.3416, pruned_loss=0.09203, over 28908.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3614, pruned_loss=0.1131, over 5664789.40 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3587, pruned_loss=0.1116, over 5706893.79 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3622, pruned_loss=0.1139, over 5649378.88 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:53:12,086 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303974.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 01:53:32,912 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.962e+02 1.576e+03 2.019e+03 2.491e+03 6.139e+03, threshold=4.039e+03, percent-clipped=3.0
+2023-03-15 01:53:33,650 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1304000.pt
+2023-03-15 01:53:45,695 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:53:49,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:53:53,783 INFO [train.py:968] (0/2) Epoch 29, batch 28200, giga_loss[loss=0.2992, simple_loss=0.3701, pruned_loss=0.1141, over 28887.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3628, pruned_loss=0.1146, over 5654929.64 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.359, pruned_loss=0.1118, over 5707791.16 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3632, pruned_loss=0.1152, over 5640882.38 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:54:05,604 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:54:13,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:54:15,353 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:54:18,809 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-15 01:54:34,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304066.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:54:34,881 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4995, 2.9399, 1.5729, 1.6321], device='cuda:0'), covar=tensor([0.0832, 0.0333, 0.0740, 0.1162], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0578, 0.0415, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 01:54:37,295 INFO [train.py:968] (0/2) Epoch 29, batch 28250, giga_loss[loss=0.4152, simple_loss=0.4416, pruned_loss=0.1944, over 28006.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3641, pruned_loss=0.1157, over 5654041.30 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3596, pruned_loss=0.112, over 5706921.02 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3641, pruned_loss=0.1161, over 5641113.59 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:54:59,302 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:55:03,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.959e+03 2.634e+03 3.377e+03 8.451e+03, threshold=5.267e+03, percent-clipped=16.0
+2023-03-15 01:55:24,276 INFO [train.py:968] (0/2) Epoch 29, batch 28300, giga_loss[loss=0.3318, simple_loss=0.39, pruned_loss=0.1368, over 27698.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3657, pruned_loss=0.1166, over 5662959.44 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3587, pruned_loss=0.1115, over 5711925.63 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3666, pruned_loss=0.1175, over 5646992.38 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:55:32,308 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:55:57,407 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3763, 1.7313, 1.3412, 1.4774], device='cuda:0'), covar=tensor([0.2578, 0.2662, 0.3028, 0.2343], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1163, 0.1429, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 01:56:12,415 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:56:12,776 INFO [train.py:968] (0/2) Epoch 29, batch 28350, giga_loss[loss=0.3033, simple_loss=0.3736, pruned_loss=0.1165, over 28878.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3684, pruned_loss=0.1187, over 5659716.19 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3588, pruned_loss=0.1115, over 5714849.17 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3691, pruned_loss=0.1195, over 5643826.45 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:56:17,233 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:56:29,701 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:56:32,301 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:56:44,554 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.877e+03 2.195e+03 3.041e+03 7.295e+03, threshold=4.390e+03, percent-clipped=6.0
+2023-03-15 01:56:54,424 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304209.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 01:56:57,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304212.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:57:00,867 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:57:04,035 INFO [train.py:968] (0/2) Epoch 29, batch 28400, giga_loss[loss=0.305, simple_loss=0.3736, pruned_loss=0.1182, over 28858.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3707, pruned_loss=0.121, over 5657536.62 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.359, pruned_loss=0.1116, over 5714082.92 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3712, pruned_loss=0.1215, over 5645569.58 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 01:57:26,858 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:57:26,873 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304241.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:57:39,196 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3968, 1.5452, 1.6085, 1.2210], device='cuda:0'), covar=tensor([0.1636, 0.2557, 0.1391, 0.1672], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0724, 0.0989, 0.0888], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 01:57:55,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8608, 4.6902, 4.4723, 2.0944], device='cuda:0'), covar=tensor([0.0535, 0.0696, 0.0834, 0.2094], device='cuda:0'), in_proj_covar=tensor([0.1334, 0.1235, 0.1039, 0.0766], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 01:57:56,263 INFO [train.py:968] (0/2) Epoch 29, batch 28450, giga_loss[loss=0.2951, simple_loss=0.3705, pruned_loss=0.1098, over 28597.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3722, pruned_loss=0.1225, over 5656954.79 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3595, pruned_loss=0.1119, over 5719848.98 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3726, pruned_loss=0.123, over 5640331.80 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:58:24,111 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.270e+03 1.843e+03 2.436e+03 3.284e+03 9.299e+03, threshold=4.872e+03, percent-clipped=10.0
+2023-03-15 01:58:39,005 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:58:41,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:58:42,100 INFO [train.py:968] (0/2) Epoch 29, batch 28500, giga_loss[loss=0.2911, simple_loss=0.3583, pruned_loss=0.1119, over 28831.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3704, pruned_loss=0.1197, over 5665058.38 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3589, pruned_loss=0.1115, over 5723323.12 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3716, pruned_loss=0.1207, over 5646408.08 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:58:49,384 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:59:11,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:59:11,782 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304349.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:59:31,953 INFO [train.py:968] (0/2) Epoch 29, batch 28550, giga_loss[loss=0.3029, simple_loss=0.3717, pruned_loss=0.117, over 28646.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3704, pruned_loss=0.1202, over 5645383.53 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3587, pruned_loss=0.1114, over 5716360.85 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3718, pruned_loss=0.1213, over 5635819.71 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:59:51,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.4220, 3.2842, 3.1339, 1.8049], device='cuda:0'), covar=tensor([0.0814, 0.0906, 0.0869, 0.1846], device='cuda:0'), in_proj_covar=tensor([0.1334, 0.1234, 0.1039, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 01:59:57,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.859e+03 2.329e+03 3.391e+03 6.060e+03, threshold=4.659e+03, percent-clipped=8.0
+2023-03-15 01:59:59,580 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:00:03,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:00:18,510 INFO [train.py:968] (0/2) Epoch 29, batch 28600, giga_loss[loss=0.3138, simple_loss=0.3865, pruned_loss=0.1205, over 28713.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3718, pruned_loss=0.1219, over 5637372.95 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3588, pruned_loss=0.1115, over 5712742.74 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3733, pruned_loss=0.123, over 5630290.65 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:01:06,488 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9580, 2.1528, 1.9342, 1.9301], device='cuda:0'), covar=tensor([0.2168, 0.2513, 0.2449, 0.2378], device='cuda:0'), in_proj_covar=tensor([0.0509, 0.0763, 0.0734, 0.0699], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 02:01:11,664 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:01:14,725 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:01:15,613 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:01:17,554 INFO [train.py:968] (0/2) Epoch 29, batch 28650, giga_loss[loss=0.346, simple_loss=0.3776, pruned_loss=0.1572, over 23595.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3702, pruned_loss=0.122, over 5631105.85 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3584, pruned_loss=0.1113, over 5715768.26 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3719, pruned_loss=0.1232, over 5621720.55 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:01:19,868 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:01:20,000 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-15 02:01:41,364 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304492.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 02:01:44,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304495.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 02:01:48,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.164e+03 1.733e+03 2.397e+03 3.131e+03 8.559e+03, threshold=4.794e+03, percent-clipped=11.0
+2023-03-15 02:01:49,181 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:01:51,580 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:02:00,079 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3230, 1.8980, 1.5852, 1.5200], device='cuda:0'), covar=tensor([0.0772, 0.0357, 0.0320, 0.0833], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 02:02:09,849 INFO [train.py:968] (0/2) Epoch 29, batch 28700, giga_loss[loss=0.3172, simple_loss=0.3757, pruned_loss=0.1294, over 27961.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3683, pruned_loss=0.121, over 5624849.68 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3584, pruned_loss=0.1114, over 5704597.14 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3698, pruned_loss=0.1221, over 5624941.03 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:02:12,273 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304524.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 02:02:29,767 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:02:32,577 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:02:34,975 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:02:55,757 INFO [train.py:968] (0/2) Epoch 29, batch 28750, giga_loss[loss=0.3276, simple_loss=0.3775, pruned_loss=0.1389, over 27578.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3674, pruned_loss=0.1206, over 5639986.50 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3583, pruned_loss=0.1113, over 5705809.83 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3689, pruned_loss=0.1217, over 5638061.82 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:03:05,326 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:03:23,913 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.169e+03 1.703e+03 2.124e+03 3.005e+03 7.643e+03, threshold=4.248e+03, percent-clipped=3.0
+2023-03-15 02:03:34,868 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:03:37,180 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:03:38,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:03:42,412 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:03:43,436 INFO [train.py:968] (0/2) Epoch 29, batch 28800, giga_loss[loss=0.283, simple_loss=0.3597, pruned_loss=0.1031, over 29004.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3672, pruned_loss=0.1207, over 5642602.53 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3585, pruned_loss=0.1114, over 5707447.96 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3684, pruned_loss=0.1217, over 5637952.60 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:04:03,216 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:04:07,154 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:04:10,397 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:04:33,085 INFO [train.py:968] (0/2) Epoch 29, batch 28850, giga_loss[loss=0.275, simple_loss=0.3513, pruned_loss=0.09933, over 28950.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3673, pruned_loss=0.1206, over 5657315.03 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3586, pruned_loss=0.1116, over 5709403.93 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3683, pruned_loss=0.1213, over 5651576.45 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:04:40,316 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:04:49,683 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:04:52,028 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:05:04,089 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 2.192e+03 2.807e+03 3.913e+03 7.030e+03, threshold=5.614e+03, percent-clipped=18.0
+2023-03-15 02:05:11,248 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-15 02:05:20,405 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:05:20,777 INFO [train.py:968] (0/2) Epoch 29, batch 28900, giga_loss[loss=0.3944, simple_loss=0.4242, pruned_loss=0.1823, over 26535.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3703, pruned_loss=0.123, over 5660486.90 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1114, over 5711411.75 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3712, pruned_loss=0.1238, over 5653722.70 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:05:59,402 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:06:01,381 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:06:02,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7577, 2.6111, 1.6507, 0.9281], device='cuda:0'), covar=tensor([0.9451, 0.4017, 0.4575, 0.8082], device='cuda:0'), in_proj_covar=tensor([0.1859, 0.1752, 0.1670, 0.1517], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 02:06:09,275 INFO [train.py:968] (0/2) Epoch 29, batch 28950, giga_loss[loss=0.349, simple_loss=0.3988, pruned_loss=0.1496, over 28734.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.371, pruned_loss=0.1232, over 5662434.51 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3584, pruned_loss=0.1113, over 5707737.05 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3722, pruned_loss=0.1243, over 5658263.63 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:06:14,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:06:27,472 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:06:36,008 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.015e+03 1.864e+03 2.391e+03 3.305e+03 5.961e+03, threshold=4.783e+03, percent-clipped=2.0
+2023-03-15 02:06:55,259 INFO [train.py:968] (0/2) Epoch 29, batch 29000, giga_loss[loss=0.3368, simple_loss=0.3828, pruned_loss=0.1453, over 28611.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3729, pruned_loss=0.1251, over 5661124.13 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1114, over 5696893.43 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3739, pruned_loss=0.126, over 5666015.22 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:07:14,893 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:07:44,065 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:07:44,480 INFO [train.py:968] (0/2) Epoch 29, batch 29050, giga_loss[loss=0.3061, simple_loss=0.3795, pruned_loss=0.1163, over 28962.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3721, pruned_loss=0.1243, over 5663181.87 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3583, pruned_loss=0.1114, over 5697764.96 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3731, pruned_loss=0.1251, over 5666025.23 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:08:13,351 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.243e+03 1.885e+03 2.332e+03 3.198e+03 8.480e+03, threshold=4.664e+03, percent-clipped=8.0
+2023-03-15 02:08:22,912 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:08:31,172 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:08:32,143 INFO [train.py:968] (0/2) Epoch 29, batch 29100, giga_loss[loss=0.3274, simple_loss=0.389, pruned_loss=0.1329, over 28904.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3736, pruned_loss=0.1255, over 5664060.06 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3588, pruned_loss=0.112, over 5700506.67 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3742, pruned_loss=0.1259, over 5663191.45 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:08:33,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:09:01,209 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:09:16,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4608, 1.8633, 1.4477, 1.5413], device='cuda:0'), covar=tensor([0.0813, 0.0306, 0.0356, 0.0937], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 02:09:19,444 INFO [train.py:968] (0/2) Epoch 29, batch 29150, giga_loss[loss=0.3328, simple_loss=0.3826, pruned_loss=0.1414, over 27611.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.374, pruned_loss=0.1253, over 5668987.90 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3589, pruned_loss=0.1121, over 5702084.31 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3749, pruned_loss=0.1259, over 5666041.31 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:09:33,010 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:09:35,771 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:09:42,043 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:09:49,526 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.315e+03 1.937e+03 2.425e+03 3.528e+03 1.213e+04, threshold=4.851e+03, percent-clipped=12.0
+2023-03-15 02:09:55,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4661, 4.2965, 4.0898, 1.9394], device='cuda:0'), covar=tensor([0.0709, 0.0889, 0.0985, 0.2046], device='cuda:0'), in_proj_covar=tensor([0.1335, 0.1236, 0.1038, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 02:10:03,121 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:10:07,838 INFO [train.py:968] (0/2) Epoch 29, batch 29200, giga_loss[loss=0.2616, simple_loss=0.3401, pruned_loss=0.0916, over 29007.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3746, pruned_loss=0.1258, over 5668808.85 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3592, pruned_loss=0.1122, over 5705020.47 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3753, pruned_loss=0.1264, over 5663472.88 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:10:53,610 INFO [train.py:968] (0/2) Epoch 29, batch 29250, libri_loss[loss=0.3667, simple_loss=0.4199, pruned_loss=0.1567, over 25526.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3751, pruned_loss=0.1261, over 5667119.60 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3594, pruned_loss=0.1123, over 5703968.66 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3756, pruned_loss=0.1266, over 5663658.94 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:11:21,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.203e+03 1.830e+03 2.224e+03 3.175e+03 9.072e+03, threshold=4.448e+03, percent-clipped=4.0
+2023-03-15 02:11:38,292 INFO [train.py:968] (0/2) Epoch 29, batch 29300, giga_loss[loss=0.3635, simple_loss=0.4161, pruned_loss=0.1554, over 28780.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3748, pruned_loss=0.1255, over 5658449.66 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3595, pruned_loss=0.1124, over 5697873.63 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3753, pruned_loss=0.126, over 5659896.76 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:11:55,541 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1305137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:11:59,027 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1305140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:12:10,409 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-15 02:12:28,869 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:12:30,003 INFO [train.py:968] (0/2) Epoch 29, batch 29350, giga_loss[loss=0.2986, simple_loss=0.3683, pruned_loss=0.1145, over 28489.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3737, pruned_loss=0.124, over 5646854.28 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3588, pruned_loss=0.1119, over 5702970.45 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3751, pruned_loss=0.1252, over 5642465.37 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:13:04,418 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.708e+03 2.000e+03 2.953e+03 5.841e+03, threshold=4.000e+03, percent-clipped=2.0
+2023-03-15 02:13:18,915 INFO [train.py:968] (0/2) Epoch 29, batch 29400, giga_loss[loss=0.2643, simple_loss=0.3427, pruned_loss=0.09292, over 28823.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3734, pruned_loss=0.1232, over 5651775.94 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.359, pruned_loss=0.112, over 5708068.42 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3747, pruned_loss=0.1243, over 5642465.08 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:13:37,926 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2959, 1.2924, 3.7102, 3.2718], device='cuda:0'), covar=tensor([0.1660, 0.2801, 0.0491, 0.1120], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0682, 0.1024, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 02:13:39,660 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1305245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:13:39,677 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6554, 1.8978, 1.3680, 1.3630], device='cuda:0'), covar=tensor([0.1128, 0.0636, 0.1062, 0.1260], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0455, 0.0527, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 02:14:00,042 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1305266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:14:04,595 INFO [train.py:968] (0/2) Epoch 29, batch 29450, giga_loss[loss=0.2706, simple_loss=0.3453, pruned_loss=0.09795, over 28521.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3716, pruned_loss=0.1215, over 5664356.46 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3591, pruned_loss=0.112, over 5712412.41 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3729, pruned_loss=0.1226, over 5651839.93 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:14:15,365 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1305285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:14:31,285 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.355e+03 1.916e+03 2.500e+03 3.049e+03 1.222e+04, threshold=5.000e+03, percent-clipped=13.0
+2023-03-15 02:14:47,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3713, 1.3004, 3.5878, 3.1489], device='cuda:0'), covar=tensor([0.1575, 0.2780, 0.0526, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0681, 0.1022, 0.0997], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 02:14:48,457 INFO [train.py:968] (0/2) Epoch 29, batch 29500, libri_loss[loss=0.2795, simple_loss=0.3409, pruned_loss=0.109, over 29369.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.371, pruned_loss=0.1213, over 5656053.22 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.359, pruned_loss=0.112, over 5702456.80 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3723, pruned_loss=0.1224, over 5653249.90 frames. ], batch size: 67, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:15:05,598 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1305341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:15:32,370 INFO [train.py:968] (0/2) Epoch 29, batch 29550, giga_loss[loss=0.3225, simple_loss=0.3804, pruned_loss=0.1323, over 28938.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3724, pruned_loss=0.1228, over 5652207.13 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3587, pruned_loss=0.1119, over 5703505.26 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3742, pruned_loss=0.1241, over 5647628.35 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:15:52,146 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1305388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:15:55,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1305391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:16:05,444 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-15 02:16:06,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.203e+03 1.865e+03 2.365e+03 3.348e+03 1.445e+04, threshold=4.729e+03, percent-clipped=6.0
+2023-03-15 02:16:19,398 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:16:20,086 INFO [train.py:968] (0/2) Epoch 29, batch 29600, giga_loss[loss=0.3485, simple_loss=0.3999, pruned_loss=0.1485, over 28735.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3735, pruned_loss=0.1236, over 5656038.02 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3584, pruned_loss=0.1117, over 5699513.99 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3756, pruned_loss=0.1251, over 5654813.33 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:16:28,531 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1305428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:16:31,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1305431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:16:58,663 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:17:00,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2029, 1.2521, 1.0717, 0.9137], device='cuda:0'), covar=tensor([0.0807, 0.0380, 0.0800, 0.1055], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0456, 0.0529, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 02:17:01,718 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-15 02:17:08,991 INFO [train.py:968] (0/2) Epoch 29, batch 29650, giga_loss[loss=0.3065, simple_loss=0.3687, pruned_loss=0.1222, over 28599.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3738, pruned_loss=0.1248, over 5653548.73 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3591, pruned_loss=0.1123, over 5698960.49 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.375, pruned_loss=0.1257, over 5652190.99 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:17:33,898 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.10 vs. limit=5.0
+2023-03-15 02:17:38,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.211e+03 1.863e+03 2.309e+03 3.162e+03 5.740e+03, threshold=4.618e+03, percent-clipped=3.0
+2023-03-15 02:17:54,167 INFO [train.py:968] (0/2) Epoch 29, batch 29700, giga_loss[loss=0.2849, simple_loss=0.3662, pruned_loss=0.1018, over 28993.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3745, pruned_loss=0.125, over 5662709.51 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3603, pruned_loss=0.113, over 5695233.85 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.375, pruned_loss=0.1256, over 5664119.92 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:17:59,440 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6398, 1.7716, 1.7411, 1.4903], device='cuda:0'), covar=tensor([0.3023, 0.2785, 0.2405, 0.2830], device='cuda:0'), in_proj_covar=tensor([0.2082, 0.2056, 0.1944, 0.2099], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 02:18:39,188 INFO [train.py:968] (0/2) Epoch 29, batch 29750, giga_loss[loss=0.291, simple_loss=0.3638, pruned_loss=0.1091, over 28937.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3751, pruned_loss=0.1256, over 5663105.01 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3597, pruned_loss=0.1125, over 5700201.80 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3764, pruned_loss=0.1268, over 5658960.14 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:18:41,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5526, 1.5936, 1.7328, 1.3492], device='cuda:0'), covar=tensor([0.1778, 0.2806, 0.1532, 0.1839], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0724, 0.0989, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 02:19:07,289 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1305600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:19:09,174 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.874e+03 2.273e+03 3.140e+03 7.553e+03, threshold=4.546e+03, percent-clipped=9.0
+2023-03-15 02:19:27,676 INFO [train.py:968] (0/2) Epoch 29, batch 29800, giga_loss[loss=0.2924, simple_loss=0.3635, pruned_loss=0.1107, over 28729.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3754, pruned_loss=0.1261, over 5636083.86 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.36, pruned_loss=0.1126, over 5688459.85 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3765, pruned_loss=0.1272, over 5642646.45 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:19:44,466 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1305641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:19:51,261 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3309, 1.5277, 1.5912, 1.4828], device='cuda:0'), covar=tensor([0.1713, 0.1201, 0.1735, 0.1341], device='cuda:0'), in_proj_covar=tensor([0.0514, 0.0766, 0.0741, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 02:20:10,836 INFO [train.py:968] (0/2) Epoch 29, batch 29850, giga_loss[loss=0.331, simple_loss=0.3914, pruned_loss=0.1352, over 28561.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3765, pruned_loss=0.127, over 5644623.92 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3606, pruned_loss=0.1129, over 5693188.63 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3773, pruned_loss=0.128, over 5644456.54 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:20:41,956 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.686e+03 2.230e+03 3.126e+03 1.160e+04, threshold=4.459e+03, percent-clipped=6.0
+2023-03-15 02:20:53,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1305716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:20:56,701 INFO [train.py:968] (0/2) Epoch 29, batch 29900, giga_loss[loss=0.3078, simple_loss=0.3787, pruned_loss=0.1185, over 28922.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3758, pruned_loss=0.1255, over 5657513.01 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3606, pruned_loss=0.1129, over 5697684.75 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3767, pruned_loss=0.1266, over 5652380.64 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:21:45,467 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4593, 3.2010, 1.5941, 1.5662], device='cuda:0'), covar=tensor([0.0965, 0.0378, 0.0865, 0.1288], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0578, 0.0415, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 02:21:48,749 INFO [train.py:968] (0/2) Epoch 29, batch 29950, giga_loss[loss=0.2737, simple_loss=0.3461, pruned_loss=0.1006, over 28939.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3749, pruned_loss=0.1244, over 5654626.90 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3606, pruned_loss=0.1129, over 5697684.75 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3756, pruned_loss=0.1253, over 5650632.30 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:22:00,606 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1305784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:22:03,012 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1305787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:22:09,809 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-15 02:22:15,050 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.232e+03 1.593e+03 1.993e+03 2.715e+03 4.994e+03, threshold=3.986e+03, percent-clipped=2.0
+2023-03-15 02:22:28,786 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:22:32,412 INFO [train.py:968] (0/2) Epoch 29, batch 30000, giga_loss[loss=0.3259, simple_loss=0.3895, pruned_loss=0.1312, over 28988.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1227, over 5654282.79 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1125, over 5691416.52 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3742, pruned_loss=0.1241, over 5654445.26 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:22:32,416 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 02:22:40,684 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.2031, simple_loss=0.3116, pruned_loss=0.04732, over 944034.00 frames. 
+2023-03-15 02:22:40,684 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 02:23:19,743 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1305859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:23:21,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1305862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:23:27,768 INFO [train.py:968] (0/2) Epoch 29, batch 30050, giga_loss[loss=0.2788, simple_loss=0.3425, pruned_loss=0.1076, over 28762.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3714, pruned_loss=0.1221, over 5660928.87 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1125, over 5691416.52 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3727, pruned_loss=0.1232, over 5661055.32 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:23:48,500 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:23:58,982 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.240e+03 1.829e+03 2.551e+03 3.113e+03 1.280e+04, threshold=5.103e+03, percent-clipped=7.0
+2023-03-15 02:24:16,804 INFO [train.py:968] (0/2) Epoch 29, batch 30100, giga_loss[loss=0.2859, simple_loss=0.3315, pruned_loss=0.1202, over 23447.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1192, over 5667227.97 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.36, pruned_loss=0.1123, over 5696071.76 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3681, pruned_loss=0.1204, over 5662841.40 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:24:43,119 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-15 02:25:03,925 INFO [train.py:968] (0/2) Epoch 29, batch 30150, giga_loss[loss=0.2914, simple_loss=0.3535, pruned_loss=0.1147, over 28458.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3644, pruned_loss=0.1188, over 5647079.31 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.36, pruned_loss=0.1122, over 5688814.34 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3654, pruned_loss=0.12, over 5648867.60 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:25:06,691 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1305975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:25:11,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.6618, 2.4173, 2.4767, 2.2690], device='cuda:0'), covar=tensor([0.1885, 0.2638, 0.2188, 0.2390], device='cuda:0'), in_proj_covar=tensor([0.0516, 0.0770, 0.0744, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 02:25:26,049 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1306000.pt
+2023-03-15 02:25:30,961 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.834e+03 2.272e+03 3.089e+03 6.553e+03, threshold=4.544e+03, percent-clipped=5.0
+2023-03-15 02:25:48,649 INFO [train.py:968] (0/2) Epoch 29, batch 30200, giga_loss[loss=0.3123, simple_loss=0.3688, pruned_loss=0.1278, over 28941.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.363, pruned_loss=0.1184, over 5660771.02 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3602, pruned_loss=0.1124, over 5695329.96 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3639, pruned_loss=0.1194, over 5655516.54 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:26:37,990 INFO [train.py:968] (0/2) Epoch 29, batch 30250, giga_loss[loss=0.2614, simple_loss=0.3466, pruned_loss=0.0881, over 28943.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3624, pruned_loss=0.1181, over 5646762.59 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3603, pruned_loss=0.1126, over 5699298.18 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.363, pruned_loss=0.1189, over 5637556.39 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:27:09,161 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.267e+03 1.900e+03 2.432e+03 3.256e+03 1.116e+04, threshold=4.865e+03, percent-clipped=11.0
+2023-03-15 02:27:25,133 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1306118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:27:27,300 INFO [train.py:968] (0/2) Epoch 29, batch 30300, giga_loss[loss=0.2417, simple_loss=0.3297, pruned_loss=0.07683, over 28900.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3614, pruned_loss=0.1156, over 5640519.99 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3603, pruned_loss=0.1125, over 5692896.31 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.362, pruned_loss=0.1163, over 5638084.00 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:27:27,627 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1306121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:27:33,033 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5348, 2.0317, 1.5047, 1.4712], device='cuda:0'), covar=tensor([0.2752, 0.2823, 0.3223, 0.2597], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1162, 0.1431, 0.1016], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 02:27:35,968 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0361, 1.3478, 1.2462, 0.9744], device='cuda:0'), covar=tensor([0.1470, 0.2257, 0.1265, 0.1712], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0723, 0.0990, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 02:27:53,747 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-15 02:27:57,295 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1306150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:28:20,272 INFO [train.py:968] (0/2) Epoch 29, batch 30350, giga_loss[loss=0.2951, simple_loss=0.3736, pruned_loss=0.1083, over 28599.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3588, pruned_loss=0.1118, over 5639831.41 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3599, pruned_loss=0.1124, over 5693674.65 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3596, pruned_loss=0.1125, over 5636745.33 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:28:49,907 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.117e+03 1.870e+03 2.279e+03 3.581e+03 9.021e+03, threshold=4.558e+03, percent-clipped=11.0
+2023-03-15 02:29:04,276 INFO [train.py:968] (0/2) Epoch 29, batch 30400, giga_loss[loss=0.2874, simple_loss=0.3605, pruned_loss=0.1072, over 28908.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3561, pruned_loss=0.1087, over 5660243.80 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3585, pruned_loss=0.1118, over 5701902.90 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3579, pruned_loss=0.1098, over 5648246.96 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:29:50,742 INFO [train.py:968] (0/2) Epoch 29, batch 30450, giga_loss[loss=0.2514, simple_loss=0.3335, pruned_loss=0.08465, over 28825.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3547, pruned_loss=0.1072, over 5660977.23 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1125, over 5705508.48 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3558, pruned_loss=0.1073, over 5646790.89 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:30:01,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6239, 2.0893, 1.3911, 0.9993], device='cuda:0'), covar=tensor([0.7509, 0.3893, 0.3507, 0.6776], device='cuda:0'), in_proj_covar=tensor([0.1852, 0.1743, 0.1667, 0.1511], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 02:30:21,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.616e+03 2.170e+03 3.720e+03 1.143e+04, threshold=4.340e+03, percent-clipped=17.0
+2023-03-15 02:30:34,947 INFO [train.py:968] (0/2) Epoch 29, batch 30500, libri_loss[loss=0.2645, simple_loss=0.3351, pruned_loss=0.09696, over 29512.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.351, pruned_loss=0.104, over 5650630.51 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3579, pruned_loss=0.1122, over 5694334.50 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3525, pruned_loss=0.104, over 5646536.82 frames. ], batch size: 84, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:31:22,201 INFO [train.py:968] (0/2) Epoch 29, batch 30550, libri_loss[loss=0.3175, simple_loss=0.3738, pruned_loss=0.1306, over 29382.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3492, pruned_loss=0.09997, over 5666185.63 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3574, pruned_loss=0.1121, over 5690558.44 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3506, pruned_loss=0.0998, over 5664448.85 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:31:57,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.908e+02 1.616e+03 2.202e+03 3.340e+03 8.826e+03, threshold=4.403e+03, percent-clipped=8.0
+2023-03-15 02:32:13,433 INFO [train.py:968] (0/2) Epoch 29, batch 30600, giga_loss[loss=0.3112, simple_loss=0.3768, pruned_loss=0.1228, over 28541.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3501, pruned_loss=0.1007, over 5665022.20 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.357, pruned_loss=0.112, over 5691876.11 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3514, pruned_loss=0.1003, over 5662178.96 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:32:19,352 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5578, 1.6617, 1.7395, 1.3430], device='cuda:0'), covar=tensor([0.1935, 0.2879, 0.1671, 0.1978], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0718, 0.0984, 0.0884], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 02:33:05,548 INFO [train.py:968] (0/2) Epoch 29, batch 30650, giga_loss[loss=0.2649, simple_loss=0.3473, pruned_loss=0.09122, over 28754.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3489, pruned_loss=0.09923, over 5666590.23 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.357, pruned_loss=0.1121, over 5692944.73 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3499, pruned_loss=0.09884, over 5663347.40 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:33:37,525 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.340e+02 1.490e+03 1.810e+03 2.630e+03 9.098e+03, threshold=3.621e+03, percent-clipped=5.0
+2023-03-15 02:33:48,942 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.63 vs. limit=5.0
+2023-03-15 02:33:52,390 INFO [train.py:968] (0/2) Epoch 29, batch 30700, giga_loss[loss=0.2658, simple_loss=0.344, pruned_loss=0.09377, over 28648.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.346, pruned_loss=0.09735, over 5660699.57 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3565, pruned_loss=0.112, over 5687180.83 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.347, pruned_loss=0.09692, over 5662120.63 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:34:39,564 INFO [train.py:968] (0/2) Epoch 29, batch 30750, giga_loss[loss=0.2602, simple_loss=0.3417, pruned_loss=0.08938, over 28509.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3449, pruned_loss=0.09702, over 5659767.42 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3565, pruned_loss=0.1122, over 5692855.24 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3454, pruned_loss=0.09607, over 5655096.19 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:35:08,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.696e+03 2.174e+03 3.120e+03 9.955e+03, threshold=4.348e+03, percent-clipped=16.0
+2023-03-15 02:35:21,886 INFO [train.py:968] (0/2) Epoch 29, batch 30800, libri_loss[loss=0.2257, simple_loss=0.2924, pruned_loss=0.0795, over 29376.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3447, pruned_loss=0.09694, over 5662771.74 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3557, pruned_loss=0.112, over 5691157.02 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3453, pruned_loss=0.09569, over 5659452.41 frames. ], batch size: 67, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:35:39,724 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6328, 1.9416, 1.6360, 1.4407], device='cuda:0'), covar=tensor([0.2429, 0.2282, 0.2534, 0.2421], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1164, 0.1434, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 02:36:10,089 INFO [train.py:968] (0/2) Epoch 29, batch 30850, giga_loss[loss=0.2272, simple_loss=0.3175, pruned_loss=0.06847, over 28920.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3427, pruned_loss=0.09512, over 5658371.38 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3554, pruned_loss=0.1117, over 5690258.23 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3433, pruned_loss=0.09414, over 5655797.53 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:36:15,221 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8733, 1.1307, 2.8543, 2.7180], device='cuda:0'), covar=tensor([0.1692, 0.2661, 0.0608, 0.1162], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0678, 0.1016, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 02:36:15,786 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1306677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:36:34,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9745, 2.3319, 1.8051, 2.2912], device='cuda:0'), covar=tensor([0.2610, 0.2679, 0.3054, 0.2376], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1162, 0.1431, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 02:36:36,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7216, 2.4429, 1.5229, 0.8061], device='cuda:0'), covar=tensor([0.8593, 0.4361, 0.4279, 0.7917], device='cuda:0'), in_proj_covar=tensor([0.1851, 0.1741, 0.1664, 0.1509], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 02:36:41,202 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-15 02:36:46,312 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.628e+02 1.461e+03 1.863e+03 2.487e+03 7.489e+03, threshold=3.726e+03, percent-clipped=4.0
+2023-03-15 02:36:59,578 INFO [train.py:968] (0/2) Epoch 29, batch 30900, libri_loss[loss=0.3251, simple_loss=0.3829, pruned_loss=0.1337, over 28557.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3398, pruned_loss=0.09299, over 5654724.38 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3548, pruned_loss=0.1115, over 5685784.92 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3404, pruned_loss=0.09192, over 5655927.13 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:37:08,992 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9482, 1.1749, 1.0815, 0.9412], device='cuda:0'), covar=tensor([0.2192, 0.2314, 0.1461, 0.1951], device='cuda:0'), in_proj_covar=tensor([0.2059, 0.2034, 0.1920, 0.2072], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 02:37:51,009 INFO [train.py:968] (0/2) Epoch 29, batch 30950, giga_loss[loss=0.273, simple_loss=0.335, pruned_loss=0.1055, over 26604.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.337, pruned_loss=0.0913, over 5666699.80 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3545, pruned_loss=0.1113, over 5689145.29 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3375, pruned_loss=0.09025, over 5664233.23 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:37:51,352 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1306771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:38:17,874 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1306798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:38:19,842 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-15 02:38:25,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.014e+02 1.401e+03 1.814e+03 2.691e+03 7.110e+03, threshold=3.628e+03, percent-clipped=9.0
+2023-03-15 02:38:38,531 INFO [train.py:968] (0/2) Epoch 29, batch 31000, giga_loss[loss=0.2483, simple_loss=0.3271, pruned_loss=0.08474, over 28706.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3354, pruned_loss=0.09088, over 5668730.64 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3538, pruned_loss=0.111, over 5693684.57 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3361, pruned_loss=0.09001, over 5662056.97 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:38:42,980 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1306826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:39:15,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4842, 1.8383, 1.5751, 1.7208], device='cuda:0'), covar=tensor([0.0682, 0.0303, 0.0315, 0.0740], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 02:39:18,338 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 02:39:21,390 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1306867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:39:25,507 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6283, 2.3904, 1.7779, 0.8083], device='cuda:0'), covar=tensor([0.7363, 0.3698, 0.4332, 0.7194], device='cuda:0'), in_proj_covar=tensor([0.1853, 0.1745, 0.1667, 0.1513], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 02:39:26,623 INFO [train.py:968] (0/2) Epoch 29, batch 31050, giga_loss[loss=0.226, simple_loss=0.3109, pruned_loss=0.07055, over 29007.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3346, pruned_loss=0.0911, over 5661102.24 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.353, pruned_loss=0.1107, over 5696205.91 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3353, pruned_loss=0.09003, over 5652348.36 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:40:04,532 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.205e+02 1.549e+03 1.971e+03 2.718e+03 6.732e+03, threshold=3.943e+03, percent-clipped=14.0
+2023-03-15 02:40:19,453 INFO [train.py:968] (0/2) Epoch 29, batch 31100, giga_loss[loss=0.2618, simple_loss=0.3502, pruned_loss=0.08671, over 28453.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3368, pruned_loss=0.09192, over 5654300.59 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3532, pruned_loss=0.111, over 5696185.28 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3369, pruned_loss=0.09052, over 5646921.26 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:41:13,298 INFO [train.py:968] (0/2) Epoch 29, batch 31150, giga_loss[loss=0.2347, simple_loss=0.3189, pruned_loss=0.07522, over 28618.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3385, pruned_loss=0.09236, over 5653527.29 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3531, pruned_loss=0.1114, over 5699737.12 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3382, pruned_loss=0.0902, over 5642510.45 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:41:45,392 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4350, 1.5739, 1.3796, 1.5956], device='cuda:0'), covar=tensor([0.0720, 0.0432, 0.0364, 0.0794], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 02:41:52,811 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.539e+03 2.246e+03 2.751e+03 6.953e+03, threshold=4.492e+03, percent-clipped=10.0
+2023-03-15 02:42:10,774 INFO [train.py:968] (0/2) Epoch 29, batch 31200, giga_loss[loss=0.2739, simple_loss=0.352, pruned_loss=0.09787, over 28763.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3382, pruned_loss=0.09248, over 5643878.33 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3528, pruned_loss=0.1114, over 5705716.68 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3378, pruned_loss=0.09023, over 5628471.32 frames. ], batch size: 243, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:42:50,490 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:43:09,094 INFO [train.py:968] (0/2) Epoch 29, batch 31250, giga_loss[loss=0.2425, simple_loss=0.3246, pruned_loss=0.08024, over 28887.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.338, pruned_loss=0.09244, over 5654855.52 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3525, pruned_loss=0.1112, over 5712555.94 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3373, pruned_loss=0.09006, over 5633783.94 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:43:43,577 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.026e+03 1.493e+03 2.068e+03 2.990e+03 5.370e+03, threshold=4.136e+03, percent-clipped=9.0
+2023-03-15 02:44:01,824 INFO [train.py:968] (0/2) Epoch 29, batch 31300, giga_loss[loss=0.2682, simple_loss=0.3601, pruned_loss=0.08819, over 28992.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3366, pruned_loss=0.09142, over 5656562.12 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3518, pruned_loss=0.111, over 5711727.77 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3362, pruned_loss=0.08891, over 5637750.69 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:44:33,934 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:45:01,266 INFO [train.py:968] (0/2) Epoch 29, batch 31350, giga_loss[loss=0.2212, simple_loss=0.308, pruned_loss=0.06717, over 29029.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3343, pruned_loss=0.08859, over 5654036.92 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3516, pruned_loss=0.1109, over 5717490.19 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3337, pruned_loss=0.08612, over 5632136.67 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:45:04,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:45:31,917 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:45:34,616 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:45:38,464 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:45:45,318 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.294e+02 1.377e+03 1.809e+03 2.268e+03 6.226e+03, threshold=3.619e+03, percent-clipped=3.0
+2023-03-15 02:46:00,745 INFO [train.py:968] (0/2) Epoch 29, batch 31400, giga_loss[loss=0.2166, simple_loss=0.2927, pruned_loss=0.07026, over 28993.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3319, pruned_loss=0.08772, over 5664949.09 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3514, pruned_loss=0.1109, over 5720988.40 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3312, pruned_loss=0.08529, over 5643140.51 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:46:07,715 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:46:28,239 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:46:41,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2897, 4.1083, 3.9165, 2.0961], device='cuda:0'), covar=tensor([0.0636, 0.0813, 0.0797, 0.1966], device='cuda:0'), in_proj_covar=tensor([0.1307, 0.1212, 0.1014, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 02:46:59,211 INFO [train.py:968] (0/2) Epoch 29, batch 31450, libri_loss[loss=0.2929, simple_loss=0.3553, pruned_loss=0.1153, over 28643.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3307, pruned_loss=0.08752, over 5671357.09 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3506, pruned_loss=0.1105, over 5723366.97 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3304, pruned_loss=0.08541, over 5650671.84 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:47:20,486 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:47:22,370 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:47:42,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.487e+02 1.390e+03 1.770e+03 2.320e+03 1.010e+04, threshold=3.539e+03, percent-clipped=10.0
+2023-03-15 02:47:49,335 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3874, 2.0290, 1.5021, 0.6661], device='cuda:0'), covar=tensor([0.6065, 0.3533, 0.4845, 0.7216], device='cuda:0'), in_proj_covar=tensor([0.1848, 0.1740, 0.1661, 0.1509], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 02:47:51,646 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:47:53,909 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:47:55,619 INFO [train.py:968] (0/2) Epoch 29, batch 31500, giga_loss[loss=0.2801, simple_loss=0.3452, pruned_loss=0.1075, over 26821.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3296, pruned_loss=0.08646, over 5679232.18 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3501, pruned_loss=0.1102, over 5725447.15 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3297, pruned_loss=0.08486, over 5660617.60 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:47:56,009 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:48:23,170 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1307344.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 02:48:23,221 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:48:25,353 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:48:26,018 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:48:52,762 INFO [train.py:968] (0/2) Epoch 29, batch 31550, giga_loss[loss=0.2387, simple_loss=0.3094, pruned_loss=0.08401, over 24342.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3309, pruned_loss=0.08646, over 5665206.00 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3503, pruned_loss=0.1104, over 5718527.42 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3305, pruned_loss=0.08463, over 5656719.78 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:48:59,840 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:49:10,831 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3391, 4.2496, 1.4363, 1.6563], device='cuda:0'), covar=tensor([0.1254, 0.0398, 0.1090, 0.1534], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0575, 0.0415, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 02:49:11,416 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:49:15,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:49:37,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.541e+03 2.223e+03 3.027e+03 9.004e+03, threshold=4.445e+03, percent-clipped=16.0
+2023-03-15 02:49:51,676 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:49:55,851 INFO [train.py:968] (0/2) Epoch 29, batch 31600, giga_loss[loss=0.2342, simple_loss=0.3196, pruned_loss=0.07441, over 29182.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3321, pruned_loss=0.08687, over 5659656.69 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3501, pruned_loss=0.1103, over 5717908.12 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3317, pruned_loss=0.08516, over 5652783.06 frames. ], batch size: 113, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:51:01,187 INFO [train.py:968] (0/2) Epoch 29, batch 31650, giga_loss[loss=0.2618, simple_loss=0.3412, pruned_loss=0.09121, over 28393.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3295, pruned_loss=0.085, over 5675262.55 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3499, pruned_loss=0.1102, over 5720783.20 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3291, pruned_loss=0.08342, over 5666666.45 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:51:16,698 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-15 02:51:43,766 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2283, 4.0867, 3.8812, 2.6097], device='cuda:0'), covar=tensor([0.0738, 0.0934, 0.0970, 0.1700], device='cuda:0'), in_proj_covar=tensor([0.1306, 0.1210, 0.1013, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 02:51:45,087 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.357e+02 1.504e+03 1.959e+03 2.403e+03 7.727e+03, threshold=3.919e+03, percent-clipped=3.0
+2023-03-15 02:52:01,544 INFO [train.py:968] (0/2) Epoch 29, batch 31700, giga_loss[loss=0.2542, simple_loss=0.344, pruned_loss=0.08221, over 28984.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3316, pruned_loss=0.08675, over 5641908.21 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3495, pruned_loss=0.1102, over 5688366.66 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3311, pruned_loss=0.0848, over 5663455.72 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:52:20,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5060, 1.6695, 1.7295, 1.3021], device='cuda:0'), covar=tensor([0.1986, 0.2973, 0.1684, 0.2071], device='cuda:0'), in_proj_covar=tensor([0.0941, 0.0717, 0.0989, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 02:52:59,864 INFO [train.py:968] (0/2) Epoch 29, batch 31750, giga_loss[loss=0.2599, simple_loss=0.3579, pruned_loss=0.08098, over 29029.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3366, pruned_loss=0.08744, over 5654984.55 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3492, pruned_loss=0.11, over 5693397.75 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3361, pruned_loss=0.08553, over 5666899.36 frames. ], batch size: 285, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:53:39,402 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.860e+03 2.314e+03 3.598e+03 9.867e+03, threshold=4.628e+03, percent-clipped=22.0
+2023-03-15 02:53:56,298 INFO [train.py:968] (0/2) Epoch 29, batch 31800, giga_loss[loss=0.2778, simple_loss=0.3647, pruned_loss=0.09543, over 28503.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.34, pruned_loss=0.08855, over 5655645.95 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3498, pruned_loss=0.1107, over 5697918.79 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3384, pruned_loss=0.0854, over 5659651.84 frames. ], batch size: 369, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:54:04,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1307627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:54:56,266 INFO [train.py:968] (0/2) Epoch 29, batch 31850, giga_loss[loss=0.2341, simple_loss=0.3286, pruned_loss=0.06985, over 28743.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.34, pruned_loss=0.08685, over 5657715.22 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3497, pruned_loss=0.1106, over 5698545.35 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3389, pruned_loss=0.08428, over 5659779.83 frames. ], batch size: 243, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:55:42,224 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.546e+03 1.973e+03 2.983e+03 5.061e+03, threshold=3.946e+03, percent-clipped=2.0
+2023-03-15 02:55:53,023 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307719.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 02:55:54,060 INFO [train.py:968] (0/2) Epoch 29, batch 31900, giga_loss[loss=0.288, simple_loss=0.3639, pruned_loss=0.1061, over 28709.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3391, pruned_loss=0.08602, over 5663991.38 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3499, pruned_loss=0.1108, over 5699756.38 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3377, pruned_loss=0.0832, over 5663729.39 frames. ], batch size: 243, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:56:04,396 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3042, 1.6228, 1.6458, 1.3362], device='cuda:0'), covar=tensor([0.2915, 0.2171, 0.1565, 0.2279], device='cuda:0'), in_proj_covar=tensor([0.2042, 0.2017, 0.1898, 0.2052], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 02:56:05,805 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1307730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:56:55,543 INFO [train.py:968] (0/2) Epoch 29, batch 31950, giga_loss[loss=0.2513, simple_loss=0.3323, pruned_loss=0.08511, over 29006.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3385, pruned_loss=0.08653, over 5677915.15 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3491, pruned_loss=0.1104, over 5701683.24 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.338, pruned_loss=0.08434, over 5675714.99 frames. ], batch size: 285, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:57:25,904 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1307793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:57:44,256 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.957e+02 1.397e+03 1.863e+03 2.748e+03 7.759e+03, threshold=3.726e+03, percent-clipped=8.0
+2023-03-15 02:58:04,249 INFO [train.py:968] (0/2) Epoch 29, batch 32000, giga_loss[loss=0.2454, simple_loss=0.3323, pruned_loss=0.07923, over 28400.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3379, pruned_loss=0.08739, over 5680035.60 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3491, pruned_loss=0.1103, over 5707098.32 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3371, pruned_loss=0.08489, over 5672501.75 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:58:36,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3084, 1.5994, 1.5865, 1.4541], device='cuda:0'), covar=tensor([0.1930, 0.1823, 0.1993, 0.1820], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0751, 0.0730, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 02:59:07,063 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307862.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 02:59:11,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307865.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 02:59:18,026 INFO [train.py:968] (0/2) Epoch 29, batch 32050, giga_loss[loss=0.215, simple_loss=0.3018, pruned_loss=0.0641, over 29046.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3367, pruned_loss=0.08739, over 5673393.92 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3489, pruned_loss=0.1102, over 5698673.73 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3362, pruned_loss=0.08534, over 5674659.15 frames. ], batch size: 285, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 02:59:28,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4567, 1.7582, 1.5380, 1.5703], device='cuda:0'), covar=tensor([0.0727, 0.0310, 0.0309, 0.0834], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 02:59:45,819 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307894.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 03:00:04,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.799e+02 1.541e+03 1.879e+03 2.810e+03 7.222e+03, threshold=3.757e+03, percent-clipped=12.0
+2023-03-15 03:00:16,166 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7248, 2.1946, 1.9384, 1.8085], device='cuda:0'), covar=tensor([0.2462, 0.2496, 0.2361, 0.2510], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0751, 0.0730, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 03:00:17,898 INFO [train.py:968] (0/2) Epoch 29, batch 32100, giga_loss[loss=0.2868, simple_loss=0.3536, pruned_loss=0.11, over 28109.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3337, pruned_loss=0.08638, over 5667604.41 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3486, pruned_loss=0.1103, over 5692842.57 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3329, pruned_loss=0.08372, over 5672846.87 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:00:55,618 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1307949.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 03:01:22,440 INFO [train.py:968] (0/2) Epoch 29, batch 32150, giga_loss[loss=0.2483, simple_loss=0.332, pruned_loss=0.08229, over 28478.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3319, pruned_loss=0.08528, over 5671166.00 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3488, pruned_loss=0.1105, over 5696329.47 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3308, pruned_loss=0.08255, over 5671534.35 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:01:34,712 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5325, 1.6056, 1.7380, 1.3661], device='cuda:0'), covar=tensor([0.1976, 0.2703, 0.1602, 0.1923], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0718, 0.0990, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 03:01:47,033 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-15 03:01:55,782 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1308000.pt
+2023-03-15 03:01:57,896 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:02:07,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.815e+02 1.538e+03 1.932e+03 2.568e+03 5.163e+03, threshold=3.865e+03, percent-clipped=3.0
+2023-03-15 03:02:22,185 INFO [train.py:968] (0/2) Epoch 29, batch 32200, giga_loss[loss=0.2156, simple_loss=0.3022, pruned_loss=0.06447, over 27662.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3308, pruned_loss=0.08545, over 5680610.63 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3484, pruned_loss=0.1101, over 5700822.54 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3298, pruned_loss=0.08285, over 5676339.76 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:02:35,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2800, 4.0951, 3.9429, 1.8663], device='cuda:0'), covar=tensor([0.0703, 0.0786, 0.0916, 0.2258], device='cuda:0'), in_proj_covar=tensor([0.1306, 0.1207, 0.1014, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 03:03:14,742 INFO [train.py:968] (0/2) Epoch 29, batch 32250, giga_loss[loss=0.2881, simple_loss=0.3606, pruned_loss=0.1078, over 28141.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3344, pruned_loss=0.0873, over 5675426.45 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3473, pruned_loss=0.1096, over 5697121.88 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3339, pruned_loss=0.08475, over 5675781.62 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:03:46,337 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1308100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:03:49,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:03:54,800 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.640e+03 2.184e+03 3.185e+03 8.827e+03, threshold=4.368e+03, percent-clipped=13.0
+2023-03-15 03:03:58,299 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5622, 2.0952, 1.8389, 1.7324], device='cuda:0'), covar=tensor([0.0720, 0.0266, 0.0287, 0.0840], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 03:04:08,289 INFO [train.py:968] (0/2) Epoch 29, batch 32300, giga_loss[loss=0.2322, simple_loss=0.3085, pruned_loss=0.07789, over 28891.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3344, pruned_loss=0.08849, over 5674011.94 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3469, pruned_loss=0.1095, over 5685509.70 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3339, pruned_loss=0.08566, over 5684873.44 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:04:26,957 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-15 03:04:38,678 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:04:42,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:05:05,755 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:05:07,999 INFO [train.py:968] (0/2) Epoch 29, batch 32350, giga_loss[loss=0.2967, simple_loss=0.3667, pruned_loss=0.1133, over 28967.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3346, pruned_loss=0.08964, over 5670392.29 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3465, pruned_loss=0.1092, over 5684106.81 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3344, pruned_loss=0.0874, over 5680212.40 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:05:15,642 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1308177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:05:54,798 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.592e+03 1.908e+03 2.865e+03 7.813e+03, threshold=3.817e+03, percent-clipped=10.0
+2023-03-15 03:06:08,242 INFO [train.py:968] (0/2) Epoch 29, batch 32400, giga_loss[loss=0.2556, simple_loss=0.339, pruned_loss=0.08609, over 28951.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3343, pruned_loss=0.08997, over 5664026.99 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3459, pruned_loss=0.1089, over 5677469.58 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3345, pruned_loss=0.08807, over 5678316.35 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:06:45,903 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:06:51,035 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1308251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:06:51,075 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:07:20,659 INFO [train.py:968] (0/2) Epoch 29, batch 32450, giga_loss[loss=0.2604, simple_loss=0.3479, pruned_loss=0.08646, over 28631.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3359, pruned_loss=0.08937, over 5665647.03 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3459, pruned_loss=0.1089, over 5678293.23 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3359, pruned_loss=0.08782, over 5676091.21 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:07:31,880 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1308280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:08:14,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.010e+03 1.563e+03 2.039e+03 2.567e+03 6.115e+03, threshold=4.077e+03, percent-clipped=4.0
+2023-03-15 03:08:15,144 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:08:22,769 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:08:29,253 INFO [train.py:968] (0/2) Epoch 29, batch 32500, libri_loss[loss=0.3212, simple_loss=0.3696, pruned_loss=0.1365, over 19502.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3361, pruned_loss=0.08891, over 5655457.22 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3457, pruned_loss=0.1086, over 5675721.30 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.336, pruned_loss=0.08728, over 5666513.87 frames. ], batch size: 187, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:08:34,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308324.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 03:09:02,590 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1308343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:09:37,912 INFO [train.py:968] (0/2) Epoch 29, batch 32550, giga_loss[loss=0.2349, simple_loss=0.3124, pruned_loss=0.0787, over 28651.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3329, pruned_loss=0.08714, over 5663761.08 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3457, pruned_loss=0.1087, over 5678217.51 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3327, pruned_loss=0.08563, over 5670230.14 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:10:23,941 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.094e+03 1.592e+03 2.032e+03 2.852e+03 6.036e+03, threshold=4.064e+03, percent-clipped=3.0
+2023-03-15 03:10:35,512 INFO [train.py:968] (0/2) Epoch 29, batch 32600, giga_loss[loss=0.2153, simple_loss=0.2892, pruned_loss=0.07073, over 28990.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3292, pruned_loss=0.08652, over 5651793.96 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3461, pruned_loss=0.1092, over 5660808.71 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3283, pruned_loss=0.08421, over 5672153.40 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:11:30,756 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308467.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 03:11:33,510 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308470.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 03:11:33,823 INFO [train.py:968] (0/2) Epoch 29, batch 32650, giga_loss[loss=0.248, simple_loss=0.3291, pruned_loss=0.08345, over 28901.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3266, pruned_loss=0.08597, over 5656092.40 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.346, pruned_loss=0.1091, over 5667042.00 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3252, pruned_loss=0.08328, over 5666558.12 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:11:40,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:12:06,936 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1308499.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 03:12:19,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.161e+03 1.667e+03 2.211e+03 2.868e+03 6.225e+03, threshold=4.423e+03, percent-clipped=6.0
+2023-03-15 03:12:32,275 INFO [train.py:968] (0/2) Epoch 29, batch 32700, giga_loss[loss=0.2914, simple_loss=0.3612, pruned_loss=0.1108, over 28698.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3283, pruned_loss=0.08729, over 5656535.50 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3457, pruned_loss=0.1089, over 5664018.71 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.327, pruned_loss=0.0847, over 5666779.68 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:12:37,229 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2603, 1.7290, 1.5559, 1.5222], device='cuda:0'), covar=tensor([0.2362, 0.2407, 0.2354, 0.2247], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0750, 0.0728, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 03:13:00,708 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-15 03:13:02,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0943, 1.4957, 1.5267, 1.3137], device='cuda:0'), covar=tensor([0.2342, 0.1780, 0.2371, 0.1989], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0749, 0.0727, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 03:13:27,536 INFO [train.py:968] (0/2) Epoch 29, batch 32750, giga_loss[loss=0.2757, simple_loss=0.3464, pruned_loss=0.1025, over 27609.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3299, pruned_loss=0.08791, over 5665876.49 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3459, pruned_loss=0.1091, over 5667376.22 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3281, pruned_loss=0.08498, over 5671076.85 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:13:31,106 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.56 vs. limit=5.0
+2023-03-15 03:14:09,923 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.471e+02 1.590e+03 2.024e+03 2.783e+03 1.338e+04, threshold=4.049e+03, percent-clipped=12.0
+2023-03-15 03:14:12,403 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1308613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:14:19,101 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:14:23,509 INFO [train.py:968] (0/2) Epoch 29, batch 32800, giga_loss[loss=0.2427, simple_loss=0.3088, pruned_loss=0.0883, over 24224.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3271, pruned_loss=0.08584, over 5657905.73 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3459, pruned_loss=0.1094, over 5663080.39 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3251, pruned_loss=0.08245, over 5665278.51 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:14:24,068 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:14:30,882 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:14:58,642 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1308650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:15:24,464 INFO [train.py:968] (0/2) Epoch 29, batch 32850, giga_loss[loss=0.2318, simple_loss=0.3098, pruned_loss=0.07695, over 28108.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.326, pruned_loss=0.08452, over 5653302.83 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3457, pruned_loss=0.1093, over 5658076.86 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3242, pruned_loss=0.08151, over 5663689.82 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:16:08,707 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1308706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:16:08,779 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6475, 1.8925, 1.5872, 1.7611], device='cuda:0'), covar=tensor([0.3098, 0.2571, 0.2953, 0.2175], device='cuda:0'), in_proj_covar=tensor([0.1624, 0.1167, 0.1437, 0.1021], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 03:16:15,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.545e+03 1.963e+03 2.455e+03 6.560e+03, threshold=3.927e+03, percent-clipped=4.0
+2023-03-15 03:16:28,148 INFO [train.py:968] (0/2) Epoch 29, batch 32900, giga_loss[loss=0.2239, simple_loss=0.3022, pruned_loss=0.07278, over 28802.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3253, pruned_loss=0.08498, over 5648735.12 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3457, pruned_loss=0.1094, over 5652163.59 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3235, pruned_loss=0.08192, over 5661935.37 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:17:35,218 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:17:36,430 INFO [train.py:968] (0/2) Epoch 29, batch 32950, giga_loss[loss=0.2672, simple_loss=0.3478, pruned_loss=0.09331, over 28989.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3261, pruned_loss=0.08416, over 5662180.02 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3457, pruned_loss=0.1093, over 5651149.62 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3244, pruned_loss=0.08161, over 5673431.39 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:17:40,665 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:18:14,065 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1308801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:18:24,328 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-15 03:18:24,520 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.970e+02 1.709e+03 2.074e+03 2.646e+03 6.997e+03, threshold=4.149e+03, percent-clipped=11.0
+2023-03-15 03:18:36,383 INFO [train.py:968] (0/2) Epoch 29, batch 33000, giga_loss[loss=0.2881, simple_loss=0.3564, pruned_loss=0.1099, over 28788.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3265, pruned_loss=0.08494, over 5661905.24 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3453, pruned_loss=0.1091, over 5649419.50 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3249, pruned_loss=0.08234, over 5673553.18 frames. ], batch size: 243, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:18:36,387 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 03:18:45,816 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.1925, simple_loss=0.2942, pruned_loss=0.04546, over 944034.00 frames. 
+2023-03-15 03:18:45,817 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 03:19:00,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6369, 1.9142, 1.3682, 1.4760], device='cuda:0'), covar=tensor([0.1068, 0.0533, 0.1055, 0.1085], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0450, 0.0526, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 03:19:13,665 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-15 03:19:39,113 INFO [train.py:968] (0/2) Epoch 29, batch 33050, giga_loss[loss=0.2167, simple_loss=0.3017, pruned_loss=0.06586, over 29006.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.326, pruned_loss=0.0851, over 5674983.42 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3449, pruned_loss=0.109, over 5658310.19 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3244, pruned_loss=0.0823, over 5677151.59 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:20:31,159 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.148e+02 1.313e+03 1.705e+03 2.402e+03 4.541e+03, threshold=3.411e+03, percent-clipped=2.0
+2023-03-15 03:20:42,583 INFO [train.py:968] (0/2) Epoch 29, batch 33100, giga_loss[loss=0.2464, simple_loss=0.3368, pruned_loss=0.07796, over 29007.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3247, pruned_loss=0.08362, over 5670188.46 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3449, pruned_loss=0.1089, over 5660329.39 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3233, pruned_loss=0.0813, over 5670095.26 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:21:38,468 INFO [train.py:968] (0/2) Epoch 29, batch 33150, giga_loss[loss=0.2655, simple_loss=0.3518, pruned_loss=0.08965, over 28919.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3265, pruned_loss=0.08338, over 5662004.36 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3448, pruned_loss=0.109, over 5661010.40 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.325, pruned_loss=0.08085, over 5661216.97 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:21:56,236 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:22:25,894 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.613e+02 1.609e+03 1.984e+03 2.697e+03 1.574e+04, threshold=3.968e+03, percent-clipped=14.0
+2023-03-15 03:22:34,120 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-15 03:22:34,919 INFO [train.py:968] (0/2) Epoch 29, batch 33200, libri_loss[loss=0.3131, simple_loss=0.3634, pruned_loss=0.1314, over 29534.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3294, pruned_loss=0.08449, over 5657499.48 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3444, pruned_loss=0.1089, over 5656458.62 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3281, pruned_loss=0.08191, over 5661187.05 frames. ], batch size: 80, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:23:36,303 INFO [train.py:968] (0/2) Epoch 29, batch 33250, giga_loss[loss=0.2231, simple_loss=0.3121, pruned_loss=0.06706, over 28856.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3303, pruned_loss=0.08451, over 5667862.27 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3446, pruned_loss=0.109, over 5661012.01 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3288, pruned_loss=0.0819, over 5666704.80 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:23:52,063 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1309081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:23:52,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7209, 1.9674, 1.6648, 1.7330], device='cuda:0'), covar=tensor([0.2930, 0.2892, 0.3342, 0.2522], device='cuda:0'), in_proj_covar=tensor([0.1623, 0.1165, 0.1437, 0.1021], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 03:24:08,308 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4653, 1.9682, 1.6345, 1.5928], device='cuda:0'), covar=tensor([0.0813, 0.0292, 0.0339, 0.0919], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0122, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 03:24:29,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.777e+02 1.510e+03 1.947e+03 2.616e+03 5.626e+03, threshold=3.893e+03, percent-clipped=10.0
+2023-03-15 03:24:37,698 INFO [train.py:968] (0/2) Epoch 29, batch 33300, libri_loss[loss=0.2771, simple_loss=0.3227, pruned_loss=0.1157, over 29670.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3307, pruned_loss=0.08535, over 5661923.23 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3442, pruned_loss=0.1088, over 5655747.90 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3295, pruned_loss=0.0829, over 5665997.15 frames. ], batch size: 69, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:24:47,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1309131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:24:50,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1309134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:25:12,697 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.33 vs. limit=5.0
+2023-03-15 03:25:21,781 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1309163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:25:30,955 INFO [train.py:968] (0/2) Epoch 29, batch 33350, giga_loss[loss=0.2516, simple_loss=0.3363, pruned_loss=0.08343, over 29052.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3271, pruned_loss=0.0833, over 5669500.89 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.344, pruned_loss=0.1087, over 5658292.84 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.326, pruned_loss=0.08078, over 5670520.25 frames. ], batch size: 285, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:26:11,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1658, 1.7707, 1.5658, 1.3387], device='cuda:0'), covar=tensor([0.0675, 0.0379, 0.0279, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0122, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 03:26:23,564 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.828e+02 1.642e+03 2.320e+03 4.177e+03 1.322e+04, threshold=4.640e+03, percent-clipped=26.0
+2023-03-15 03:26:32,479 INFO [train.py:968] (0/2) Epoch 29, batch 33400, giga_loss[loss=0.2273, simple_loss=0.307, pruned_loss=0.0738, over 28714.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3264, pruned_loss=0.08296, over 5669662.93 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.344, pruned_loss=0.1087, over 5660208.01 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3254, pruned_loss=0.08077, over 5668806.29 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:26:36,629 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1309224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:26:39,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1309227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:27:07,807 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1309256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:27:27,047 INFO [train.py:968] (0/2) Epoch 29, batch 33450, giga_loss[loss=0.2472, simple_loss=0.3332, pruned_loss=0.08058, over 28349.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3255, pruned_loss=0.08319, over 5678532.90 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.344, pruned_loss=0.1087, over 5668444.07 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.324, pruned_loss=0.08059, over 5670878.85 frames. ], batch size: 369, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:27:49,487 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2578, 1.5876, 1.4739, 1.2461], device='cuda:0'), covar=tensor([0.3182, 0.2653, 0.1711, 0.2489], device='cuda:0'), in_proj_covar=tensor([0.2045, 0.2010, 0.1894, 0.2049], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 03:27:59,066 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7301, 1.9508, 1.5940, 1.9471], device='cuda:0'), covar=tensor([0.2836, 0.2987, 0.3431, 0.2652], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1160, 0.1432, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 03:28:00,562 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-15 03:28:14,265 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.537e+02 1.492e+03 1.768e+03 2.375e+03 6.007e+03, threshold=3.536e+03, percent-clipped=2.0
+2023-03-15 03:28:17,459 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-15 03:28:25,184 INFO [train.py:968] (0/2) Epoch 29, batch 33500, giga_loss[loss=0.301, simple_loss=0.3693, pruned_loss=0.1163, over 27608.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3273, pruned_loss=0.084, over 5680074.92 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3438, pruned_loss=0.1086, over 5672881.78 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3259, pruned_loss=0.0813, over 5670090.34 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:28:42,640 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-15 03:29:27,716 INFO [train.py:968] (0/2) Epoch 29, batch 33550, giga_loss[loss=0.2669, simple_loss=0.3383, pruned_loss=0.09771, over 27731.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3291, pruned_loss=0.0849, over 5682666.83 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3437, pruned_loss=0.1085, over 5676526.47 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3278, pruned_loss=0.08252, over 5671417.96 frames. ], batch size: 474, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:29:43,082 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-15 03:30:21,539 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.713e+02 1.560e+03 2.013e+03 2.862e+03 6.193e+03, threshold=4.026e+03, percent-clipped=18.0
+2023-03-15 03:30:30,560 INFO [train.py:968] (0/2) Epoch 29, batch 33600, giga_loss[loss=0.2443, simple_loss=0.3324, pruned_loss=0.07807, over 28808.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3293, pruned_loss=0.08569, over 5675767.79 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3435, pruned_loss=0.1084, over 5681515.86 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3281, pruned_loss=0.08337, over 5662633.48 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:31:09,180 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1309449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:31:34,337 INFO [train.py:968] (0/2) Epoch 29, batch 33650, giga_loss[loss=0.2376, simple_loss=0.3298, pruned_loss=0.07276, over 28896.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3339, pruned_loss=0.0878, over 5667247.27 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3434, pruned_loss=0.1083, over 5683718.12 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3329, pruned_loss=0.08583, over 5655056.37 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:31:51,627 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7535, 1.8687, 1.5601, 1.9563], device='cuda:0'), covar=tensor([0.2893, 0.3011, 0.3370, 0.2630], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1158, 0.1428, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 03:32:10,793 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-15 03:32:19,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.045e+03 1.551e+03 1.932e+03 2.588e+03 7.380e+03, threshold=3.865e+03, percent-clipped=5.0
+2023-03-15 03:32:27,993 INFO [train.py:968] (0/2) Epoch 29, batch 33700, giga_loss[loss=0.2435, simple_loss=0.3372, pruned_loss=0.07492, over 28480.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3348, pruned_loss=0.0874, over 5672050.59 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.343, pruned_loss=0.1079, over 5687516.31 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3342, pruned_loss=0.08564, over 5658570.34 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:33:38,384 INFO [train.py:968] (0/2) Epoch 29, batch 33750, giga_loss[loss=0.2134, simple_loss=0.3023, pruned_loss=0.0622, over 29223.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3344, pruned_loss=0.08719, over 5669236.69 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.343, pruned_loss=0.1079, over 5682251.70 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3338, pruned_loss=0.08542, over 5663671.96 frames. ], batch size: 113, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:34:32,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.955e+02 1.561e+03 2.124e+03 3.174e+03 1.192e+04, threshold=4.248e+03, percent-clipped=12.0
+2023-03-15 03:34:41,879 INFO [train.py:968] (0/2) Epoch 29, batch 33800, giga_loss[loss=0.25, simple_loss=0.3331, pruned_loss=0.0834, over 28905.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3319, pruned_loss=0.08612, over 5669208.95 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3427, pruned_loss=0.1077, over 5678248.99 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3313, pruned_loss=0.08432, over 5668209.95 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:35:36,202 INFO [train.py:968] (0/2) Epoch 29, batch 33850, giga_loss[loss=0.2387, simple_loss=0.3265, pruned_loss=0.07541, over 28807.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3318, pruned_loss=0.08639, over 5677256.15 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3424, pruned_loss=0.1074, over 5684468.10 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3312, pruned_loss=0.08438, over 5670778.16 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:36:12,200 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5327, 1.8475, 1.4815, 1.7953], device='cuda:0'), covar=tensor([0.2865, 0.2946, 0.3400, 0.2294], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1157, 0.1428, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 03:36:32,858 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.857e+02 1.506e+03 2.000e+03 2.947e+03 7.778e+03, threshold=4.000e+03, percent-clipped=5.0
+2023-03-15 03:36:42,099 INFO [train.py:968] (0/2) Epoch 29, batch 33900, giga_loss[loss=0.2379, simple_loss=0.3185, pruned_loss=0.07866, over 28609.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3312, pruned_loss=0.0866, over 5661033.68 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3426, pruned_loss=0.1076, over 5667108.76 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3304, pruned_loss=0.08461, over 5671195.77 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:37:44,390 INFO [train.py:968] (0/2) Epoch 29, batch 33950, giga_loss[loss=0.272, simple_loss=0.3446, pruned_loss=0.09969, over 27659.00 frames. ], tot_loss[loss=0.252, simple_loss=0.33, pruned_loss=0.08698, over 5661689.67 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3426, pruned_loss=0.1076, over 5662614.37 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3292, pruned_loss=0.08498, over 5673189.00 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:37:48,630 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-15 03:38:34,229 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.189e+02 1.403e+03 1.807e+03 2.734e+03 1.150e+04, threshold=3.613e+03, percent-clipped=5.0
+2023-03-15 03:38:43,696 INFO [train.py:968] (0/2) Epoch 29, batch 34000, giga_loss[loss=0.2451, simple_loss=0.3328, pruned_loss=0.07872, over 29127.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3305, pruned_loss=0.08651, over 5673358.41 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3428, pruned_loss=0.1077, over 5665114.03 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3295, pruned_loss=0.08457, over 5680401.52 frames. ], batch size: 200, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 03:38:47,080 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1309824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:39:12,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8104, 1.1762, 2.8865, 2.8448], device='cuda:0'), covar=tensor([0.1803, 0.2661, 0.0638, 0.0963], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0678, 0.1010, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 03:39:41,294 INFO [train.py:968] (0/2) Epoch 29, batch 34050, giga_loss[loss=0.2318, simple_loss=0.3195, pruned_loss=0.07199, over 28823.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3286, pruned_loss=0.08497, over 5659570.27 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3433, pruned_loss=0.1081, over 5659574.98 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.327, pruned_loss=0.08248, over 5670401.56 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:40:03,500 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4900, 1.7833, 1.7847, 1.4981], device='cuda:0'), covar=tensor([0.1831, 0.1586, 0.2027, 0.1747], device='cuda:0'), in_proj_covar=tensor([0.0498, 0.0748, 0.0726, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 03:40:24,236 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.069e+02 1.462e+03 1.864e+03 2.427e+03 1.602e+04, threshold=3.729e+03, percent-clipped=13.0
+2023-03-15 03:40:30,322 INFO [train.py:968] (0/2) Epoch 29, batch 34100, giga_loss[loss=0.2727, simple_loss=0.3594, pruned_loss=0.09306, over 28873.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3304, pruned_loss=0.08492, over 5653247.58 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3437, pruned_loss=0.1085, over 5648556.19 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3281, pruned_loss=0.08146, over 5672004.65 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:41:23,285 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1309967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:41:25,659 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1309970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:41:28,032 INFO [train.py:968] (0/2) Epoch 29, batch 34150, giga_loss[loss=0.2363, simple_loss=0.3274, pruned_loss=0.07264, over 28537.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3322, pruned_loss=0.08454, over 5660491.73 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3439, pruned_loss=0.1086, over 5650670.67 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3298, pruned_loss=0.08111, over 5673876.00 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:41:56,761 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1309999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:41:57,210 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1310000.pt
+2023-03-15 03:42:15,914 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.096e+02 1.559e+03 1.909e+03 2.739e+03 6.714e+03, threshold=3.819e+03, percent-clipped=8.0
+2023-03-15 03:42:25,024 INFO [train.py:968] (0/2) Epoch 29, batch 34200, giga_loss[loss=0.2258, simple_loss=0.3173, pruned_loss=0.06715, over 28900.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3321, pruned_loss=0.08428, over 5668761.32 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3432, pruned_loss=0.1081, over 5656275.76 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3306, pruned_loss=0.08142, over 5674789.57 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:43:31,833 INFO [train.py:968] (0/2) Epoch 29, batch 34250, giga_loss[loss=0.2242, simple_loss=0.3148, pruned_loss=0.06683, over 28376.00 frames. ], tot_loss[loss=0.249, simple_loss=0.331, pruned_loss=0.08354, over 5670079.32 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3428, pruned_loss=0.1079, over 5660328.08 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3299, pruned_loss=0.08099, over 5671422.51 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:43:44,911 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8103, 1.1883, 1.2787, 0.9400], device='cuda:0'), covar=tensor([0.2179, 0.1478, 0.2283, 0.1875], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0747, 0.0726, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 03:44:31,950 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.513e+03 2.053e+03 2.850e+03 7.325e+03, threshold=4.107e+03, percent-clipped=10.0
+2023-03-15 03:44:36,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5003, 1.7548, 1.6034, 1.4298], device='cuda:0'), covar=tensor([0.2882, 0.2492, 0.1926, 0.2433], device='cuda:0'), in_proj_covar=tensor([0.2047, 0.2011, 0.1897, 0.2049], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 03:44:40,533 INFO [train.py:968] (0/2) Epoch 29, batch 34300, libri_loss[loss=0.3018, simple_loss=0.3616, pruned_loss=0.121, over 25579.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3309, pruned_loss=0.0833, over 5662131.44 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3428, pruned_loss=0.1079, over 5658829.65 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3299, pruned_loss=0.08105, over 5665003.86 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:45:43,952 INFO [train.py:968] (0/2) Epoch 29, batch 34350, giga_loss[loss=0.2073, simple_loss=0.2794, pruned_loss=0.06756, over 24677.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3318, pruned_loss=0.08369, over 5671230.05 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3425, pruned_loss=0.1079, over 5667092.67 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.331, pruned_loss=0.08112, over 5666074.85 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:46:42,908 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.153e+02 1.757e+03 2.435e+03 3.164e+03 1.404e+04, threshold=4.869e+03, percent-clipped=11.0
+2023-03-15 03:46:51,461 INFO [train.py:968] (0/2) Epoch 29, batch 34400, libri_loss[loss=0.2303, simple_loss=0.2968, pruned_loss=0.0819, over 28572.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3334, pruned_loss=0.08475, over 5675434.79 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3422, pruned_loss=0.1077, over 5672704.33 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3327, pruned_loss=0.08219, over 5666227.20 frames. ], batch size: 63, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:47:50,084 INFO [train.py:968] (0/2) Epoch 29, batch 34450, giga_loss[loss=0.2412, simple_loss=0.328, pruned_loss=0.07718, over 28877.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.337, pruned_loss=0.08626, over 5677231.00 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3423, pruned_loss=0.1077, over 5673696.51 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3362, pruned_loss=0.08375, over 5668955.52 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:48:35,752 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:48:46,624 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.974e+02 1.524e+03 1.820e+03 2.742e+03 6.597e+03, threshold=3.640e+03, percent-clipped=3.0
+2023-03-15 03:48:54,053 INFO [train.py:968] (0/2) Epoch 29, batch 34500, giga_loss[loss=0.2151, simple_loss=0.3045, pruned_loss=0.06282, over 28958.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3358, pruned_loss=0.08574, over 5677853.46 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3418, pruned_loss=0.1074, over 5675857.98 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3354, pruned_loss=0.08339, over 5669598.99 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:49:23,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3522, 1.9264, 1.6624, 1.6401], device='cuda:0'), covar=tensor([0.2312, 0.2120, 0.2625, 0.2385], device='cuda:0'), in_proj_covar=tensor([0.0494, 0.0741, 0.0720, 0.0684], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 03:50:00,678 INFO [train.py:968] (0/2) Epoch 29, batch 34550, giga_loss[loss=0.2323, simple_loss=0.3139, pruned_loss=0.07537, over 28969.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3358, pruned_loss=0.08658, over 5682946.42 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.342, pruned_loss=0.1074, over 5678919.66 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3353, pruned_loss=0.08437, over 5673673.17 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:50:43,834 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4589, 1.7256, 1.4374, 1.1555], device='cuda:0'), covar=tensor([0.2827, 0.2676, 0.3116, 0.2655], device='cuda:0'), in_proj_covar=tensor([0.1617, 0.1160, 0.1429, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 03:50:51,570 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:51:00,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.533e+02 1.563e+03 2.244e+03 3.366e+03 9.038e+03, threshold=4.487e+03, percent-clipped=19.0
+2023-03-15 03:51:07,686 INFO [train.py:968] (0/2) Epoch 29, batch 34600, libri_loss[loss=0.2896, simple_loss=0.3524, pruned_loss=0.1134, over 29653.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3339, pruned_loss=0.08555, over 5676878.60 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3422, pruned_loss=0.1076, over 5674824.25 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3331, pruned_loss=0.08291, over 5673524.41 frames. ], batch size: 88, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:51:57,027 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6682, 1.9026, 1.8574, 1.6744], device='cuda:0'), covar=tensor([0.3516, 0.2492, 0.2070, 0.2498], device='cuda:0'), in_proj_covar=tensor([0.2043, 0.2002, 0.1890, 0.2041], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 03:52:05,432 INFO [train.py:968] (0/2) Epoch 29, batch 34650, giga_loss[loss=0.2642, simple_loss=0.3429, pruned_loss=0.09269, over 28909.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3329, pruned_loss=0.08469, over 5688064.52 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3421, pruned_loss=0.1075, over 5682155.74 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3321, pruned_loss=0.08194, over 5678829.59 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:53:03,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.154e+02 1.274e+03 1.769e+03 2.666e+03 5.637e+03, threshold=3.537e+03, percent-clipped=4.0
+2023-03-15 03:53:07,730 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 03:53:07,920 INFO [train.py:968] (0/2) Epoch 29, batch 34700, giga_loss[loss=0.2474, simple_loss=0.3336, pruned_loss=0.08063, over 28866.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3329, pruned_loss=0.08471, over 5680631.45 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.342, pruned_loss=0.1075, over 5686186.10 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3322, pruned_loss=0.08204, over 5669285.84 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:54:02,235 INFO [train.py:968] (0/2) Epoch 29, batch 34750, giga_loss[loss=0.2261, simple_loss=0.3166, pruned_loss=0.06782, over 28979.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3358, pruned_loss=0.08662, over 5685360.76 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3415, pruned_loss=0.1072, over 5692143.13 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3354, pruned_loss=0.08401, over 5670628.00 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:54:10,239 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:54:43,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4842, 1.5532, 1.2769, 1.6097], device='cuda:0'), covar=tensor([0.0785, 0.0337, 0.0371, 0.0923], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0122, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0075, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 03:54:53,648 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.688e+03 2.298e+03 3.264e+03 8.289e+03, threshold=4.597e+03, percent-clipped=20.0
+2023-03-15 03:54:59,479 INFO [train.py:968] (0/2) Epoch 29, batch 34800, giga_loss[loss=0.2329, simple_loss=0.3114, pruned_loss=0.07725, over 28918.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3335, pruned_loss=0.08539, over 5685670.05 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.341, pruned_loss=0.107, over 5690102.33 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3335, pruned_loss=0.0831, over 5675625.55 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 03:55:24,450 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310640.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 03:55:54,955 INFO [train.py:968] (0/2) Epoch 29, batch 34850, giga_loss[loss=0.2604, simple_loss=0.3394, pruned_loss=0.09064, over 28547.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3327, pruned_loss=0.08655, over 5665343.80 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3415, pruned_loss=0.1073, over 5681895.09 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.332, pruned_loss=0.08382, over 5665243.07 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 03:56:03,885 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1310681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:56:43,183 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.800e+02 1.595e+03 2.042e+03 2.965e+03 6.151e+03, threshold=4.084e+03, percent-clipped=5.0
+2023-03-15 03:56:48,748 INFO [train.py:968] (0/2) Epoch 29, batch 34900, giga_loss[loss=0.2602, simple_loss=0.3366, pruned_loss=0.09191, over 28871.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3324, pruned_loss=0.08717, over 5659230.42 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3415, pruned_loss=0.1074, over 5677298.64 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3315, pruned_loss=0.08443, over 5662875.21 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:57:38,473 INFO [train.py:968] (0/2) Epoch 29, batch 34950, giga_loss[loss=0.2866, simple_loss=0.3692, pruned_loss=0.102, over 28495.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3365, pruned_loss=0.08936, over 5663018.12 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3411, pruned_loss=0.1071, over 5677691.27 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3359, pruned_loss=0.0868, over 5665411.87 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:57:50,661 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1310783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:57:50,732 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6991, 1.8254, 1.9080, 1.4701], device='cuda:0'), covar=tensor([0.1669, 0.2565, 0.1403, 0.1754], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0710, 0.0984, 0.0885], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 03:58:21,962 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.549e+03 1.962e+03 2.543e+03 5.854e+03, threshold=3.924e+03, percent-clipped=8.0
+2023-03-15 03:58:26,482 INFO [train.py:968] (0/2) Epoch 29, batch 35000, giga_loss[loss=0.3214, simple_loss=0.3927, pruned_loss=0.1251, over 28719.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3451, pruned_loss=0.09362, over 5664203.85 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3412, pruned_loss=0.1071, over 5679059.81 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3446, pruned_loss=0.09153, over 5664874.24 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:58:27,986 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:58:29,741 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1310824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:58:32,593 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1310827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:58:46,863 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-15 03:58:58,470 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1310856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:59:11,579 INFO [train.py:968] (0/2) Epoch 29, batch 35050, giga_loss[loss=0.265, simple_loss=0.348, pruned_loss=0.091, over 28823.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.35, pruned_loss=0.09666, over 5667728.18 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3416, pruned_loss=0.1073, over 5682384.69 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3494, pruned_loss=0.09469, over 5665192.33 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:59:47,306 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.535e+02 1.383e+03 1.777e+03 2.594e+03 6.991e+03, threshold=3.555e+03, percent-clipped=8.0
+2023-03-15 03:59:52,172 INFO [train.py:968] (0/2) Epoch 29, batch 35100, giga_loss[loss=0.261, simple_loss=0.34, pruned_loss=0.09101, over 29025.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3464, pruned_loss=0.0957, over 5676603.33 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3419, pruned_loss=0.1073, over 5684826.94 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3458, pruned_loss=0.09388, over 5671921.58 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:59:56,928 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1310926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:59:58,810 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1310929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:00:11,536 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:00:18,655 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1310953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:00:23,499 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1310958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:00:31,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([6.0401, 5.8662, 5.5306, 3.1190], device='cuda:0'), covar=tensor([0.0398, 0.0599, 0.0672, 0.1529], device='cuda:0'), in_proj_covar=tensor([0.1298, 0.1197, 0.1004, 0.0743], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 04:00:34,983 INFO [train.py:968] (0/2) Epoch 29, batch 35150, giga_loss[loss=0.2773, simple_loss=0.3393, pruned_loss=0.1076, over 27681.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3395, pruned_loss=0.0927, over 5668726.70 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.342, pruned_loss=0.1074, over 5667976.16 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3389, pruned_loss=0.09102, over 5681048.36 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:01:11,300 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311015.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:01:11,676 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.982e+02 1.132e+03 1.413e+03 1.931e+03 4.581e+03, threshold=2.826e+03, percent-clipped=6.0
+2023-03-15 04:01:15,007 INFO [train.py:968] (0/2) Epoch 29, batch 35200, giga_loss[loss=0.2384, simple_loss=0.3076, pruned_loss=0.08459, over 28991.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3332, pruned_loss=0.09063, over 5672304.71 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3415, pruned_loss=0.107, over 5673147.86 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3331, pruned_loss=0.08914, over 5677493.75 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:01:54,647 INFO [train.py:968] (0/2) Epoch 29, batch 35250, giga_loss[loss=0.1954, simple_loss=0.2794, pruned_loss=0.05569, over 29023.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3256, pruned_loss=0.0873, over 5671350.49 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3417, pruned_loss=0.107, over 5667320.26 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.325, pruned_loss=0.08567, over 5680039.67 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:02:01,697 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5610, 1.6057, 3.9590, 3.3622], device='cuda:0'), covar=tensor([0.1509, 0.2723, 0.0463, 0.1520], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0681, 0.1014, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 04:02:12,427 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:02:15,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:02:29,613 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.586e+02 1.173e+03 1.409e+03 2.055e+03 9.155e+03, threshold=2.818e+03, percent-clipped=11.0
+2023-03-15 04:02:32,140 INFO [train.py:968] (0/2) Epoch 29, batch 35300, giga_loss[loss=0.254, simple_loss=0.3135, pruned_loss=0.09724, over 26555.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3204, pruned_loss=0.08532, over 5683519.03 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3423, pruned_loss=0.1072, over 5674651.44 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3188, pruned_loss=0.08314, over 5684198.63 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:02:38,667 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:02:53,859 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0522, 2.3601, 1.9283, 2.2885], device='cuda:0'), covar=tensor([0.2496, 0.2639, 0.3039, 0.2528], device='cuda:0'), in_proj_covar=tensor([0.1619, 0.1163, 0.1430, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 04:03:04,761 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311158.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:03:07,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311161.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:03:09,503 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-15 04:03:16,429 INFO [train.py:968] (0/2) Epoch 29, batch 35350, giga_loss[loss=0.1953, simple_loss=0.2761, pruned_loss=0.05724, over 28987.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3159, pruned_loss=0.0834, over 5680581.79 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3428, pruned_loss=0.1074, over 5679161.78 frames. ], giga_tot_loss[loss=0.2378, simple_loss=0.3137, pruned_loss=0.08102, over 5677165.49 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:03:16,660 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:03:30,005 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:03:30,066 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311190.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:03:38,794 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:03:53,930 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.226e+02 1.232e+03 1.570e+03 1.981e+03 5.267e+03, threshold=3.140e+03, percent-clipped=12.0
+2023-03-15 04:03:58,094 INFO [train.py:968] (0/2) Epoch 29, batch 35400, giga_loss[loss=0.2385, simple_loss=0.3139, pruned_loss=0.08149, over 28606.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3131, pruned_loss=0.08203, over 5689826.91 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3429, pruned_loss=0.1074, over 5682917.79 frames. ], giga_tot_loss[loss=0.235, simple_loss=0.3107, pruned_loss=0.07968, over 5683751.65 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:04:38,173 INFO [train.py:968] (0/2) Epoch 29, batch 35450, giga_loss[loss=0.2192, simple_loss=0.297, pruned_loss=0.07068, over 28833.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3104, pruned_loss=0.08076, over 5698563.22 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3427, pruned_loss=0.1071, over 5685907.76 frames. ], giga_tot_loss[loss=0.2317, simple_loss=0.3073, pruned_loss=0.07807, over 5691382.66 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:05:14,447 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.357e+02 1.246e+03 1.636e+03 2.208e+03 8.736e+03, threshold=3.272e+03, percent-clipped=9.0
+2023-03-15 04:05:17,497 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:05:17,871 INFO [train.py:968] (0/2) Epoch 29, batch 35500, giga_loss[loss=0.2167, simple_loss=0.2934, pruned_loss=0.06993, over 28779.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3081, pruned_loss=0.07976, over 5700910.60 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.343, pruned_loss=0.1072, over 5681661.13 frames. ], giga_tot_loss[loss=0.2288, simple_loss=0.3043, pruned_loss=0.07661, over 5699552.31 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:05:33,677 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:05:35,502 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:06:00,953 INFO [train.py:968] (0/2) Epoch 29, batch 35550, giga_loss[loss=0.2084, simple_loss=0.2774, pruned_loss=0.06974, over 28637.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3043, pruned_loss=0.07798, over 5702449.55 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3427, pruned_loss=0.107, over 5686208.74 frames. ], giga_tot_loss[loss=0.2256, simple_loss=0.3009, pruned_loss=0.07517, over 5697808.06 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:06:01,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:06:39,085 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.223e+02 1.094e+03 1.411e+03 1.941e+03 5.024e+03, threshold=2.821e+03, percent-clipped=5.0
+2023-03-15 04:06:41,674 INFO [train.py:968] (0/2) Epoch 29, batch 35600, giga_loss[loss=0.266, simple_loss=0.3275, pruned_loss=0.1022, over 27896.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3015, pruned_loss=0.07676, over 5703589.41 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3427, pruned_loss=0.1069, over 5690593.76 frames. ], giga_tot_loss[loss=0.2229, simple_loss=0.2979, pruned_loss=0.07395, over 5696191.32 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:07:15,556 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:07:17,069 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6005, 1.8997, 1.7695, 1.6438], device='cuda:0'), covar=tensor([0.2277, 0.2537, 0.2450, 0.2506], device='cuda:0'), in_proj_covar=tensor([0.0499, 0.0747, 0.0724, 0.0688], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 04:07:17,650 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:07:23,037 INFO [train.py:968] (0/2) Epoch 29, batch 35650, giga_loss[loss=0.2451, simple_loss=0.2956, pruned_loss=0.09729, over 24052.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.2989, pruned_loss=0.07547, over 5695715.22 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3427, pruned_loss=0.1067, over 5695085.24 frames. ], giga_tot_loss[loss=0.2201, simple_loss=0.295, pruned_loss=0.07265, over 5686131.08 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:07:43,234 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:07:59,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.540e+02 1.147e+03 1.459e+03 1.919e+03 4.919e+03, threshold=2.918e+03, percent-clipped=8.0
+2023-03-15 04:08:02,002 INFO [train.py:968] (0/2) Epoch 29, batch 35700, giga_loss[loss=0.1836, simple_loss=0.2595, pruned_loss=0.05385, over 28813.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.2985, pruned_loss=0.07549, over 5696712.69 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.343, pruned_loss=0.1066, over 5694051.50 frames. ], giga_tot_loss[loss=0.2188, simple_loss=0.2933, pruned_loss=0.07213, over 5689250.09 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:08:22,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:08:41,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:08:46,031 INFO [train.py:968] (0/2) Epoch 29, batch 35750, giga_loss[loss=0.1754, simple_loss=0.2527, pruned_loss=0.04899, over 29088.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2957, pruned_loss=0.07421, over 5692580.40 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3437, pruned_loss=0.107, over 5686941.92 frames. ], giga_tot_loss[loss=0.2158, simple_loss=0.2903, pruned_loss=0.0707, over 5692615.02 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:09:17,429 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:09:26,162 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.550e+02 1.238e+03 1.607e+03 2.115e+03 6.210e+03, threshold=3.214e+03, percent-clipped=9.0
+2023-03-15 04:09:28,608 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311620.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:09:29,054 INFO [train.py:968] (0/2) Epoch 29, batch 35800, giga_loss[loss=0.2679, simple_loss=0.3446, pruned_loss=0.09555, over 28903.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3035, pruned_loss=0.0785, over 5687068.77 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3438, pruned_loss=0.107, over 5688106.06 frames. ], giga_tot_loss[loss=0.224, simple_loss=0.2981, pruned_loss=0.07494, over 5686346.73 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:09:47,780 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1419, 3.9706, 3.7227, 1.8667], device='cuda:0'), covar=tensor([0.0656, 0.0843, 0.0814, 0.2124], device='cuda:0'), in_proj_covar=tensor([0.1300, 0.1201, 0.1008, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 04:10:16,472 INFO [train.py:968] (0/2) Epoch 29, batch 35850, giga_loss[loss=0.2734, simple_loss=0.3523, pruned_loss=0.09722, over 28870.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3144, pruned_loss=0.08353, over 5691039.80 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3439, pruned_loss=0.107, over 5691300.16 frames. ], giga_tot_loss[loss=0.2352, simple_loss=0.3096, pruned_loss=0.0804, over 5687583.30 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:10:34,079 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:10:38,051 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:10:38,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6589, 1.6023, 1.8839, 1.4670], device='cuda:0'), covar=tensor([0.1657, 0.2525, 0.1362, 0.1748], device='cuda:0'), in_proj_covar=tensor([0.0945, 0.0719, 0.0996, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 04:10:53,061 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:10:55,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:11:00,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.788e+02 1.421e+03 1.834e+03 2.575e+03 5.632e+03, threshold=3.668e+03, percent-clipped=9.0
+2023-03-15 04:11:03,446 INFO [train.py:968] (0/2) Epoch 29, batch 35900, giga_loss[loss=0.2721, simple_loss=0.3495, pruned_loss=0.09736, over 28581.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3272, pruned_loss=0.08999, over 5693305.25 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3439, pruned_loss=0.107, over 5691300.16 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3234, pruned_loss=0.08756, over 5690615.02 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:11:03,627 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:11:17,905 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:11:43,654 INFO [train.py:968] (0/2) Epoch 29, batch 35950, giga_loss[loss=0.2613, simple_loss=0.3467, pruned_loss=0.08793, over 29105.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3344, pruned_loss=0.09294, over 5692416.36 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3442, pruned_loss=0.1071, over 5694372.11 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3307, pruned_loss=0.09049, over 5687637.35 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:12:10,404 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:12:17,452 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:12:17,521 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5972, 1.8301, 1.2579, 1.3852], device='cuda:0'), covar=tensor([0.1173, 0.0740, 0.1143, 0.1374], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0447, 0.0522, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 04:12:23,984 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.176e+02 1.425e+03 1.788e+03 2.257e+03 5.120e+03, threshold=3.576e+03, percent-clipped=4.0
+2023-03-15 04:12:24,918 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:12:26,178 INFO [train.py:968] (0/2) Epoch 29, batch 36000, giga_loss[loss=0.2415, simple_loss=0.3286, pruned_loss=0.0772, over 28633.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3363, pruned_loss=0.09212, over 5691859.63 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.344, pruned_loss=0.1068, over 5697712.61 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3333, pruned_loss=0.09012, over 5684628.26 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:12:26,182 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 04:12:36,686 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.1992, simple_loss=0.3059, pruned_loss=0.0462, over 944034.00 frames. 
+2023-03-15 04:12:36,687 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 04:12:42,630 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 04:13:18,353 INFO [train.py:968] (0/2) Epoch 29, batch 36050, giga_loss[loss=0.2585, simple_loss=0.3394, pruned_loss=0.08874, over 28867.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3392, pruned_loss=0.09276, over 5696641.90 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3446, pruned_loss=0.1069, over 5699171.26 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3362, pruned_loss=0.09066, over 5689140.50 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:13:40,587 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-15 04:14:01,485 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.238e+02 1.244e+03 1.629e+03 2.345e+03 4.805e+03, threshold=3.258e+03, percent-clipped=3.0
+2023-03-15 04:14:04,294 INFO [train.py:968] (0/2) Epoch 29, batch 36100, giga_loss[loss=0.2747, simple_loss=0.3484, pruned_loss=0.1005, over 28788.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3421, pruned_loss=0.09475, over 5699771.87 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3446, pruned_loss=0.1068, over 5701421.92 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3396, pruned_loss=0.09305, over 5691815.79 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:14:44,266 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4221, 4.0399, 1.7239, 1.5707], device='cuda:0'), covar=tensor([0.1100, 0.0242, 0.0888, 0.1450], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0571, 0.0412, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 04:14:46,584 INFO [train.py:968] (0/2) Epoch 29, batch 36150, giga_loss[loss=0.3165, simple_loss=0.3821, pruned_loss=0.1254, over 28236.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3453, pruned_loss=0.09736, over 5689225.21 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3443, pruned_loss=0.1066, over 5705525.12 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3435, pruned_loss=0.09596, over 5679344.56 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:14:57,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:15:08,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311995.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:15:12,502 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1312000.pt
+2023-03-15 04:15:21,192 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1312009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:15:28,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.919e+02 1.375e+03 1.725e+03 2.192e+03 7.761e+03, threshold=3.451e+03, percent-clipped=6.0
+2023-03-15 04:15:29,570 INFO [train.py:968] (0/2) Epoch 29, batch 36200, giga_loss[loss=0.2732, simple_loss=0.3545, pruned_loss=0.09599, over 28735.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3484, pruned_loss=0.09883, over 5694182.24 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3445, pruned_loss=0.1067, over 5707604.61 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3469, pruned_loss=0.09756, over 5684398.01 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:16:09,786 INFO [train.py:968] (0/2) Epoch 29, batch 36250, giga_loss[loss=0.2675, simple_loss=0.3506, pruned_loss=0.09218, over 28923.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3522, pruned_loss=0.1007, over 5690709.30 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3455, pruned_loss=0.1074, over 5701884.76 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3503, pruned_loss=0.09898, over 5688005.68 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:16:10,689 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7190, 1.9097, 1.9109, 1.4748], device='cuda:0'), covar=tensor([0.1576, 0.2559, 0.1416, 0.1766], device='cuda:0'), in_proj_covar=tensor([0.0944, 0.0717, 0.0994, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 04:16:11,212 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3643, 1.5768, 1.5858, 1.4165], device='cuda:0'), covar=tensor([0.1945, 0.1858, 0.2207, 0.1869], device='cuda:0'), in_proj_covar=tensor([0.0502, 0.0752, 0.0728, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 04:16:17,606 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7380, 3.5595, 3.3689, 1.9393], device='cuda:0'), covar=tensor([0.0720, 0.0874, 0.0797, 0.2367], device='cuda:0'), in_proj_covar=tensor([0.1291, 0.1193, 0.1000, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 04:16:22,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9067, 1.1453, 2.7807, 2.6263], device='cuda:0'), covar=tensor([0.1724, 0.2803, 0.0635, 0.1487], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0676, 0.1011, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 04:16:51,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.634e+02 1.341e+03 1.700e+03 2.226e+03 4.367e+03, threshold=3.401e+03, percent-clipped=8.0
+2023-03-15 04:16:52,328 INFO [train.py:968] (0/2) Epoch 29, batch 36300, giga_loss[loss=0.3003, simple_loss=0.3684, pruned_loss=0.1161, over 28556.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3538, pruned_loss=0.1013, over 5682159.26 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3454, pruned_loss=0.1073, over 5700052.25 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3524, pruned_loss=0.09994, over 5681122.18 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:16:55,517 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5475, 1.7029, 1.7690, 1.3582], device='cuda:0'), covar=tensor([0.1942, 0.2824, 0.1619, 0.1919], device='cuda:0'), in_proj_covar=tensor([0.0943, 0.0716, 0.0993, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 04:16:56,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:16:58,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:17:09,396 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312138.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:17:12,100 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312141.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:17:14,658 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2976, 4.1163, 3.8807, 1.9215], device='cuda:0'), covar=tensor([0.0583, 0.0780, 0.0738, 0.2142], device='cuda:0'), in_proj_covar=tensor([0.1287, 0.1191, 0.0999, 0.0741], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 04:17:23,023 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:17:32,973 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312170.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:17:33,570 INFO [train.py:968] (0/2) Epoch 29, batch 36350, giga_loss[loss=0.2844, simple_loss=0.3593, pruned_loss=0.1047, over 28801.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3548, pruned_loss=0.1007, over 5687270.53 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3455, pruned_loss=0.1072, over 5702151.60 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3537, pruned_loss=0.09966, over 5684373.67 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:17:37,452 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1312175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:17:44,014 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1312183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:17:52,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1312194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:18:12,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.271e+02 1.324e+03 1.554e+03 1.980e+03 5.762e+03, threshold=3.108e+03, percent-clipped=6.0
+2023-03-15 04:18:13,513 INFO [train.py:968] (0/2) Epoch 29, batch 36400, libri_loss[loss=0.3242, simple_loss=0.3698, pruned_loss=0.1393, over 29336.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3544, pruned_loss=0.09921, over 5699369.11 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3463, pruned_loss=0.1076, over 5706500.22 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3531, pruned_loss=0.09782, over 5692912.06 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:18:54,092 INFO [train.py:968] (0/2) Epoch 29, batch 36450, giga_loss[loss=0.2597, simple_loss=0.3456, pruned_loss=0.08691, over 28972.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3528, pruned_loss=0.09708, over 5707363.83 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3463, pruned_loss=0.1075, over 5711160.92 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3518, pruned_loss=0.09579, over 5697928.70 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:19:32,400 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:33,352 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.292e+02 1.295e+03 1.536e+03 1.846e+03 5.366e+03, threshold=3.071e+03, percent-clipped=6.0
+2023-03-15 04:19:33,903 INFO [train.py:968] (0/2) Epoch 29, batch 36500, giga_loss[loss=0.3039, simple_loss=0.3857, pruned_loss=0.1111, over 29021.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3513, pruned_loss=0.0962, over 5707531.61 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3466, pruned_loss=0.1077, over 5705180.15 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3504, pruned_loss=0.0948, over 5704705.98 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:19:34,168 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:38,550 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:40,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:47,686 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:49,476 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:56,131 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4090, 1.5837, 1.6393, 1.5155], device='cuda:0'), covar=tensor([0.1638, 0.1426, 0.1666, 0.1486], device='cuda:0'), in_proj_covar=tensor([0.0504, 0.0754, 0.0730, 0.0695], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 04:19:59,357 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:20:05,281 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:20:15,848 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:20:17,689 INFO [train.py:968] (0/2) Epoch 29, batch 36550, giga_loss[loss=0.2843, simple_loss=0.3523, pruned_loss=0.1081, over 28679.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3522, pruned_loss=0.099, over 5707224.98 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.346, pruned_loss=0.1072, over 5709539.04 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3522, pruned_loss=0.09801, over 5701290.34 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:20:17,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1312371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:20:28,611 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1312384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:20:58,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.009e+02 1.527e+03 1.872e+03 2.616e+03 8.116e+03, threshold=3.745e+03, percent-clipped=18.0
+2023-03-15 04:20:59,300 INFO [train.py:968] (0/2) Epoch 29, batch 36600, giga_loss[loss=0.2829, simple_loss=0.3544, pruned_loss=0.1057, over 28572.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3557, pruned_loss=0.1039, over 5700374.00 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3469, pruned_loss=0.1079, over 5708849.10 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.355, pruned_loss=0.1025, over 5696062.83 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:21:24,616 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 04:21:40,783 INFO [train.py:968] (0/2) Epoch 29, batch 36650, giga_loss[loss=0.2806, simple_loss=0.353, pruned_loss=0.1041, over 29030.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3554, pruned_loss=0.105, over 5697732.41 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3475, pruned_loss=0.1081, over 5706491.92 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3546, pruned_loss=0.1035, over 5695764.78 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:22:24,550 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-15 04:22:26,642 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.118e+03 1.490e+03 1.727e+03 2.263e+03 7.286e+03, threshold=3.455e+03, percent-clipped=6.0
+2023-03-15 04:22:27,255 INFO [train.py:968] (0/2) Epoch 29, batch 36700, giga_loss[loss=0.2402, simple_loss=0.3223, pruned_loss=0.07908, over 29009.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.352, pruned_loss=0.1035, over 5701303.10 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3476, pruned_loss=0.1081, over 5708562.75 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3514, pruned_loss=0.1023, over 5697804.69 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:22:30,458 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-15 04:22:30,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6199, 1.8031, 1.5106, 1.7728], device='cuda:0'), covar=tensor([0.2611, 0.2765, 0.2974, 0.2690], device='cuda:0'), in_proj_covar=tensor([0.1616, 0.1162, 0.1426, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 04:22:31,436 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:22:34,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:22:52,140 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.24 vs. limit=5.0
+2023-03-15 04:22:59,948 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:23:11,969 INFO [train.py:968] (0/2) Epoch 29, batch 36750, giga_loss[loss=0.3169, simple_loss=0.369, pruned_loss=0.1324, over 23582.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3501, pruned_loss=0.1025, over 5699537.51 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3475, pruned_loss=0.1081, over 5709583.93 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3497, pruned_loss=0.1016, over 5695968.70 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:23:54,243 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.659e+02 1.467e+03 1.863e+03 2.390e+03 4.668e+03, threshold=3.726e+03, percent-clipped=8.0
+2023-03-15 04:23:55,007 INFO [train.py:968] (0/2) Epoch 29, batch 36800, giga_loss[loss=0.2477, simple_loss=0.3289, pruned_loss=0.08324, over 28609.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3492, pruned_loss=0.1011, over 5699035.50 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3475, pruned_loss=0.108, over 5711567.99 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3489, pruned_loss=0.1004, over 5694335.01 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:24:25,691 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1312652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:24:39,941 INFO [train.py:968] (0/2) Epoch 29, batch 36850, giga_loss[loss=0.2499, simple_loss=0.3358, pruned_loss=0.08203, over 28962.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3466, pruned_loss=0.09888, over 5692759.80 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3473, pruned_loss=0.1078, over 5714753.01 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3465, pruned_loss=0.09833, over 5685879.08 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:25:25,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.707e+02 1.226e+03 1.596e+03 2.230e+03 5.485e+03, threshold=3.192e+03, percent-clipped=7.0
+2023-03-15 04:25:25,445 INFO [train.py:968] (0/2) Epoch 29, batch 36900, giga_loss[loss=0.2196, simple_loss=0.2826, pruned_loss=0.07832, over 23200.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3426, pruned_loss=0.09689, over 5673048.38 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3479, pruned_loss=0.108, over 5708165.77 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3419, pruned_loss=0.09595, over 5672978.61 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:25:43,680 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1312739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:25:47,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1312746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:26:11,042 INFO [train.py:968] (0/2) Epoch 29, batch 36950, giga_loss[loss=0.3588, simple_loss=0.3836, pruned_loss=0.167, over 26559.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.337, pruned_loss=0.09432, over 5667686.04 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3479, pruned_loss=0.1079, over 5708858.12 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3363, pruned_loss=0.09344, over 5666138.60 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:27:04,027 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.968e+02 1.130e+03 1.331e+03 2.017e+03 8.007e+03, threshold=2.662e+03, percent-clipped=9.0
+2023-03-15 04:27:04,040 INFO [train.py:968] (0/2) Epoch 29, batch 37000, libri_loss[loss=0.2755, simple_loss=0.3519, pruned_loss=0.0996, over 29688.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3332, pruned_loss=0.0928, over 5661676.76 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3481, pruned_loss=0.1078, over 5714044.42 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3321, pruned_loss=0.09179, over 5654359.13 frames. ], batch size: 88, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:27:19,959 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-15 04:27:46,953 INFO [train.py:968] (0/2) Epoch 29, batch 37050, giga_loss[loss=0.2466, simple_loss=0.3255, pruned_loss=0.08387, over 28962.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3331, pruned_loss=0.0919, over 5670241.06 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3484, pruned_loss=0.1078, over 5716455.87 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3317, pruned_loss=0.09081, over 5661209.19 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:27:53,873 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4334, 3.9867, 1.7582, 1.5433], device='cuda:0'), covar=tensor([0.1108, 0.0288, 0.0935, 0.1470], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0569, 0.0412, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 04:28:02,156 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:28:04,073 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:28:28,635 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.985e+02 1.210e+03 1.475e+03 1.998e+03 6.083e+03, threshold=2.951e+03, percent-clipped=11.0
+2023-03-15 04:28:28,648 INFO [train.py:968] (0/2) Epoch 29, batch 37100, giga_loss[loss=0.2536, simple_loss=0.3214, pruned_loss=0.09291, over 23919.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3337, pruned_loss=0.09154, over 5673206.47 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.349, pruned_loss=0.108, over 5717709.33 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3317, pruned_loss=0.09016, over 5664003.90 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:28:28,799 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:28:37,009 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3560, 1.6568, 1.5403, 1.2865], device='cuda:0'), covar=tensor([0.2534, 0.2253, 0.1501, 0.2150], device='cuda:0'), in_proj_covar=tensor([0.2063, 0.2019, 0.1920, 0.2077], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 04:28:53,905 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6370, 1.8386, 1.2846, 1.3890], device='cuda:0'), covar=tensor([0.1113, 0.0646, 0.1104, 0.1243], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0449, 0.0524, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 04:29:11,606 INFO [train.py:968] (0/2) Epoch 29, batch 37150, giga_loss[loss=0.2407, simple_loss=0.3164, pruned_loss=0.08245, over 28084.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3322, pruned_loss=0.09056, over 5688177.00 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3489, pruned_loss=0.1079, over 5718775.78 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3306, pruned_loss=0.08944, over 5679924.63 frames. ], batch size: 77, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:29:17,914 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1312979.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:29:52,668 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.065e+02 1.199e+03 1.493e+03 1.825e+03 6.170e+03, threshold=2.986e+03, percent-clipped=10.0
+2023-03-15 04:29:52,684 INFO [train.py:968] (0/2) Epoch 29, batch 37200, giga_loss[loss=0.2406, simple_loss=0.317, pruned_loss=0.08208, over 28964.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3292, pruned_loss=0.08924, over 5695717.75 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3487, pruned_loss=0.1077, over 5712761.08 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3278, pruned_loss=0.08822, over 5694340.99 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:29:56,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:30:29,748 INFO [train.py:968] (0/2) Epoch 29, batch 37250, giga_loss[loss=0.2352, simple_loss=0.3067, pruned_loss=0.08183, over 28602.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.329, pruned_loss=0.08961, over 5701950.06 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3494, pruned_loss=0.1079, over 5717948.68 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3266, pruned_loss=0.08807, over 5695661.29 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:30:59,518 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 04:31:01,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:31:06,250 INFO [train.py:968] (0/2) Epoch 29, batch 37300, giga_loss[loss=0.2007, simple_loss=0.2874, pruned_loss=0.05703, over 28794.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3271, pruned_loss=0.08878, over 5699727.72 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3501, pruned_loss=0.1081, over 5711188.61 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3239, pruned_loss=0.0868, over 5700779.08 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:31:06,856 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.501e+02 1.146e+03 1.390e+03 2.072e+03 7.586e+03, threshold=2.781e+03, percent-clipped=10.0
+2023-03-15 04:31:45,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1313170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:31:45,936 INFO [train.py:968] (0/2) Epoch 29, batch 37350, giga_loss[loss=0.2362, simple_loss=0.3118, pruned_loss=0.08028, over 28866.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.325, pruned_loss=0.08782, over 5704418.44 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3504, pruned_loss=0.1082, over 5710199.70 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3219, pruned_loss=0.08595, over 5706154.17 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:31:47,526 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1313173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:32:09,493 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1313202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:32:18,344 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4335, 3.3730, 1.6333, 1.5606], device='cuda:0'), covar=tensor([0.1036, 0.0316, 0.0873, 0.1390], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0569, 0.0412, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 04:32:24,675 INFO [train.py:968] (0/2) Epoch 29, batch 37400, giga_loss[loss=0.2275, simple_loss=0.2973, pruned_loss=0.07882, over 28509.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3226, pruned_loss=0.08633, over 5708132.87 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3509, pruned_loss=0.1082, over 5708562.05 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.319, pruned_loss=0.08425, over 5710607.71 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:32:25,273 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.605e+02 1.217e+03 1.540e+03 1.948e+03 5.527e+03, threshold=3.080e+03, percent-clipped=9.0
+2023-03-15 04:32:49,959 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 04:32:51,865 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1313257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:32:53,888 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1313260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:33:01,810 INFO [train.py:968] (0/2) Epoch 29, batch 37450, giga_loss[loss=0.2131, simple_loss=0.2916, pruned_loss=0.06731, over 28089.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3208, pruned_loss=0.08541, over 5711287.33 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3515, pruned_loss=0.1085, over 5710336.71 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3169, pruned_loss=0.08306, over 5711580.22 frames. ], batch size: 77, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:33:16,744 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1313289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:33:17,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1313290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:33:42,287 INFO [train.py:968] (0/2) Epoch 29, batch 37500, giga_loss[loss=0.2281, simple_loss=0.3076, pruned_loss=0.07432, over 29032.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3185, pruned_loss=0.0842, over 5710761.71 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3518, pruned_loss=0.1086, over 5711320.03 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.315, pruned_loss=0.08218, over 5710124.59 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:33:42,863 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.757e+02 1.107e+03 1.370e+03 1.766e+03 4.808e+03, threshold=2.740e+03, percent-clipped=5.0
+2023-03-15 04:34:08,391 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313354.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:34:23,027 INFO [train.py:968] (0/2) Epoch 29, batch 37550, giga_loss[loss=0.2501, simple_loss=0.3264, pruned_loss=0.0869, over 28015.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3175, pruned_loss=0.08344, over 5713760.00 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.352, pruned_loss=0.1085, over 5714391.09 frames. ], giga_tot_loss[loss=0.2386, simple_loss=0.3141, pruned_loss=0.08159, over 5710755.97 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:35:03,677 INFO [train.py:968] (0/2) Epoch 29, batch 37600, giga_loss[loss=0.2798, simple_loss=0.3544, pruned_loss=0.1026, over 28599.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3186, pruned_loss=0.08404, over 5717878.69 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3532, pruned_loss=0.109, over 5717453.44 frames. ], giga_tot_loss[loss=0.2386, simple_loss=0.314, pruned_loss=0.08158, over 5712752.19 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:35:04,381 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.329e+02 1.157e+03 1.420e+03 1.903e+03 9.566e+03, threshold=2.840e+03, percent-clipped=15.0
+2023-03-15 04:35:43,283 INFO [train.py:968] (0/2) Epoch 29, batch 37650, giga_loss[loss=0.2541, simple_loss=0.3272, pruned_loss=0.09049, over 28922.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3233, pruned_loss=0.08722, over 5713638.57 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3533, pruned_loss=0.109, over 5712709.59 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3187, pruned_loss=0.08463, over 5714026.96 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:36:05,308 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1313497.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:36:08,410 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1313500.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:36:27,767 INFO [train.py:968] (0/2) Epoch 29, batch 37700, giga_loss[loss=0.3187, simple_loss=0.3857, pruned_loss=0.1258, over 28587.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3303, pruned_loss=0.09169, over 5703300.08 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3538, pruned_loss=0.1092, over 5707403.97 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3257, pruned_loss=0.08898, over 5708192.41 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:36:30,013 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.402e+02 1.390e+03 1.743e+03 2.265e+03 6.056e+03, threshold=3.487e+03, percent-clipped=11.0
+2023-03-15 04:36:32,976 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1313527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:36:34,184 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1313529.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:37:12,642 INFO [train.py:968] (0/2) Epoch 29, batch 37750, giga_loss[loss=0.2599, simple_loss=0.3438, pruned_loss=0.08793, over 29001.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3387, pruned_loss=0.09725, over 5690754.30 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3543, pruned_loss=0.1094, over 5702623.99 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3339, pruned_loss=0.09441, over 5698312.80 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:37:29,847 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1313590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:37:56,996 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5503, 2.1182, 1.8239, 1.6587], device='cuda:0'), covar=tensor([0.0787, 0.0278, 0.0306, 0.0898], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 04:38:00,753 INFO [train.py:968] (0/2) Epoch 29, batch 37800, libri_loss[loss=0.3559, simple_loss=0.4077, pruned_loss=0.152, over 25871.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3435, pruned_loss=0.09981, over 5673999.65 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3547, pruned_loss=0.1097, over 5696098.82 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3388, pruned_loss=0.09694, over 5685042.05 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:38:02,767 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.621e+03 2.114e+03 2.833e+03 9.669e+03, threshold=4.229e+03, percent-clipped=12.0
+2023-03-15 04:38:13,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5547, 1.6238, 1.2027, 1.2467], device='cuda:0'), covar=tensor([0.0911, 0.0509, 0.0950, 0.1251], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0449, 0.0524, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 04:38:17,042 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3977, 3.4086, 1.5758, 1.5780], device='cuda:0'), covar=tensor([0.1069, 0.0299, 0.0930, 0.1395], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0570, 0.0412, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 04:38:35,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:38:42,108 INFO [train.py:968] (0/2) Epoch 29, batch 37850, giga_loss[loss=0.3066, simple_loss=0.3761, pruned_loss=0.1185, over 28688.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3469, pruned_loss=0.1009, over 5686666.48 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.354, pruned_loss=0.1094, over 5702392.71 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3435, pruned_loss=0.09852, over 5689239.88 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:39:02,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2501, 0.8169, 0.9447, 1.4199], device='cuda:0'), covar=tensor([0.0832, 0.0404, 0.0385, 0.0922], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 04:39:28,820 INFO [train.py:968] (0/2) Epoch 29, batch 37900, giga_loss[loss=0.2767, simple_loss=0.3564, pruned_loss=0.09844, over 28720.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3529, pruned_loss=0.1041, over 5679985.32 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3541, pruned_loss=0.1094, over 5704492.14 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3501, pruned_loss=0.1022, over 5679965.08 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:39:30,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.964e+02 1.285e+03 1.641e+03 2.034e+03 5.317e+03, threshold=3.281e+03, percent-clipped=4.0
+2023-03-15 04:39:45,145 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1313744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:40:02,235 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1313763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:40:07,470 INFO [train.py:968] (0/2) Epoch 29, batch 37950, libri_loss[loss=0.325, simple_loss=0.3914, pruned_loss=0.1293, over 29766.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3541, pruned_loss=0.1043, over 5693647.95 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3542, pruned_loss=0.1096, over 5711553.97 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3517, pruned_loss=0.1023, over 5686500.84 frames. ], batch size: 87, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:40:37,507 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1313808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:40:39,400 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1313811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:40:45,882 INFO [train.py:968] (0/2) Epoch 29, batch 38000, giga_loss[loss=0.2789, simple_loss=0.3583, pruned_loss=0.09973, over 28855.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3502, pruned_loss=0.1011, over 5694602.69 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3544, pruned_loss=0.1098, over 5711681.51 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.348, pruned_loss=0.09927, over 5688441.86 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:40:48,854 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.731e+02 1.327e+03 1.676e+03 2.454e+03 7.252e+03, threshold=3.353e+03, percent-clipped=11.0
+2023-03-15 04:40:54,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4688, 1.7623, 1.3766, 1.4106], device='cuda:0'), covar=tensor([0.2896, 0.2980, 0.3321, 0.2539], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1162, 0.1427, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 04:41:01,750 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1313840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:41:29,137 INFO [train.py:968] (0/2) Epoch 29, batch 38050, giga_loss[loss=0.2532, simple_loss=0.3322, pruned_loss=0.08706, over 28915.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3469, pruned_loss=0.09812, over 5695823.69 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3547, pruned_loss=0.1099, over 5713876.14 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3449, pruned_loss=0.09638, over 5688813.76 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:41:55,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:42:11,250 INFO [train.py:968] (0/2) Epoch 29, batch 38100, libri_loss[loss=0.2578, simple_loss=0.3342, pruned_loss=0.09065, over 29561.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3471, pruned_loss=0.09778, over 5694215.45 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3548, pruned_loss=0.11, over 5705284.53 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3454, pruned_loss=0.09626, over 5695748.62 frames. ], batch size: 76, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:42:14,715 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.396e+03 1.755e+03 2.282e+03 6.218e+03, threshold=3.510e+03, percent-clipped=7.0
+2023-03-15 04:42:50,142 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:42:54,506 INFO [train.py:968] (0/2) Epoch 29, batch 38150, giga_loss[loss=0.3055, simple_loss=0.3778, pruned_loss=0.1166, over 29014.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3495, pruned_loss=0.09907, over 5697158.17 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3548, pruned_loss=0.11, over 5708430.32 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.348, pruned_loss=0.09772, over 5695338.99 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:43:01,818 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-15 04:43:06,610 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.31 vs. limit=5.0
+2023-03-15 04:43:08,962 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5292, 1.7352, 1.4241, 1.4161], device='cuda:0'), covar=tensor([0.2889, 0.2930, 0.3323, 0.2530], device='cuda:0'), in_proj_covar=tensor([0.1621, 0.1166, 0.1432, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 04:43:18,991 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1314000.pt
+2023-03-15 04:43:21,552 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-15 04:43:36,992 INFO [train.py:968] (0/2) Epoch 29, batch 38200, giga_loss[loss=0.2938, simple_loss=0.3683, pruned_loss=0.1096, over 28458.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3513, pruned_loss=0.1003, over 5691658.52 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3551, pruned_loss=0.1101, over 5702515.76 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3498, pruned_loss=0.09899, over 5694862.50 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:43:39,327 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:43:39,770 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.372e+02 1.410e+03 1.764e+03 2.470e+03 5.070e+03, threshold=3.528e+03, percent-clipped=10.0
+2023-03-15 04:44:00,819 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:44:02,842 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1314048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:44:23,590 INFO [train.py:968] (0/2) Epoch 29, batch 38250, giga_loss[loss=0.2697, simple_loss=0.334, pruned_loss=0.1027, over 23726.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.354, pruned_loss=0.1026, over 5684144.18 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3552, pruned_loss=0.1102, over 5700011.18 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3527, pruned_loss=0.1014, over 5688915.23 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:44:29,586 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:44:40,811 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-15 04:44:52,862 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314108.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:44:52,914 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:44:55,852 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1314111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:45:03,522 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1314119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:45:04,519 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6149, 1.8849, 1.5156, 1.6728], device='cuda:0'), covar=tensor([0.2602, 0.2709, 0.3017, 0.2558], device='cuda:0'), in_proj_covar=tensor([0.1621, 0.1167, 0.1430, 0.1016], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 04:45:04,940 INFO [train.py:968] (0/2) Epoch 29, batch 38300, giga_loss[loss=0.2837, simple_loss=0.3577, pruned_loss=0.1048, over 28862.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3554, pruned_loss=0.1037, over 5685217.70 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3559, pruned_loss=0.1105, over 5691924.50 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3537, pruned_loss=0.1024, over 5695483.22 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:45:08,458 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.395e+03 1.803e+03 2.932e+03 6.227e+03, threshold=3.606e+03, percent-clipped=15.0
+2023-03-15 04:45:19,595 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1314138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:45:20,739 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:45:29,495 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-15 04:45:45,582 INFO [train.py:968] (0/2) Epoch 29, batch 38350, libri_loss[loss=0.3822, simple_loss=0.4306, pruned_loss=0.1669, over 26169.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3553, pruned_loss=0.104, over 5679949.46 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3564, pruned_loss=0.1107, over 5692132.01 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3535, pruned_loss=0.1026, over 5687970.39 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:46:20,823 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4134, 1.2578, 4.0617, 3.4525], device='cuda:0'), covar=tensor([0.1572, 0.2748, 0.0501, 0.0971], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0674, 0.1008, 0.0982], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 04:46:25,715 INFO [train.py:968] (0/2) Epoch 29, batch 38400, giga_loss[loss=0.2651, simple_loss=0.3497, pruned_loss=0.09024, over 28989.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3555, pruned_loss=0.1036, over 5690596.95 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3567, pruned_loss=0.1107, over 5695129.09 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3537, pruned_loss=0.1024, over 5694114.05 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:46:28,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.080e+02 1.318e+03 1.672e+03 2.372e+03 8.030e+03, threshold=3.343e+03, percent-clipped=6.0
+2023-03-15 04:46:56,844 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:46:58,619 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1314265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:47:02,770 INFO [train.py:968] (0/2) Epoch 29, batch 38450, giga_loss[loss=0.3022, simple_loss=0.3775, pruned_loss=0.1135, over 28974.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3554, pruned_loss=0.1029, over 5698789.25 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3573, pruned_loss=0.1111, over 5701108.24 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3534, pruned_loss=0.1013, over 5696153.32 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:47:11,951 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:47:13,287 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-15 04:47:14,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1314284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:47:22,531 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:47:37,680 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:47:43,101 INFO [train.py:968] (0/2) Epoch 29, batch 38500, giga_loss[loss=0.2524, simple_loss=0.3304, pruned_loss=0.08718, over 28758.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3546, pruned_loss=0.1015, over 5709936.62 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3566, pruned_loss=0.1106, over 5705108.77 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3536, pruned_loss=0.1005, over 5704263.18 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:47:45,688 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.732e+02 1.234e+03 1.449e+03 1.868e+03 5.028e+03, threshold=2.898e+03, percent-clipped=1.0
+2023-03-15 04:47:55,005 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-15 04:47:55,403 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3814, 1.2364, 1.3092, 1.4495], device='cuda:0'), covar=tensor([0.0797, 0.0384, 0.0348, 0.0918], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 04:48:23,443 INFO [train.py:968] (0/2) Epoch 29, batch 38550, libri_loss[loss=0.3012, simple_loss=0.355, pruned_loss=0.1237, over 29375.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3525, pruned_loss=0.1008, over 5706007.48 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3565, pruned_loss=0.1105, over 5705256.90 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3518, pruned_loss=0.09985, over 5701397.34 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:48:47,401 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1314399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:49:03,241 INFO [train.py:968] (0/2) Epoch 29, batch 38600, libri_loss[loss=0.3239, simple_loss=0.3872, pruned_loss=0.1303, over 28663.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3503, pruned_loss=0.09974, over 5707039.44 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3566, pruned_loss=0.1104, over 5708534.97 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3494, pruned_loss=0.09878, over 5700344.55 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:49:05,697 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.144e+02 1.268e+03 1.535e+03 1.965e+03 9.230e+03, threshold=3.070e+03, percent-clipped=15.0
+2023-03-15 04:49:15,170 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6742, 1.5750, 1.8453, 1.4761], device='cuda:0'), covar=tensor([0.1832, 0.2584, 0.1476, 0.1773], device='cuda:0'), in_proj_covar=tensor([0.0941, 0.0720, 0.0993, 0.0891], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 04:49:44,680 INFO [train.py:968] (0/2) Epoch 29, batch 38650, giga_loss[loss=0.275, simple_loss=0.3495, pruned_loss=0.1002, over 28955.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3484, pruned_loss=0.09866, over 5707537.44 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3566, pruned_loss=0.1104, over 5709562.30 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3477, pruned_loss=0.09789, over 5701480.45 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:49:54,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1314483.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:50:26,001 INFO [train.py:968] (0/2) Epoch 29, batch 38700, giga_loss[loss=0.2586, simple_loss=0.3368, pruned_loss=0.09015, over 28580.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3475, pruned_loss=0.09845, over 5711284.23 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3564, pruned_loss=0.1103, over 5713845.33 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3469, pruned_loss=0.09774, over 5702607.64 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:50:28,486 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:50:28,881 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.902e+02 1.253e+03 1.476e+03 1.953e+03 8.175e+03, threshold=2.952e+03, percent-clipped=7.0
+2023-03-15 04:50:38,529 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-15 04:50:44,603 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:50:46,576 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1314545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:51:07,114 INFO [train.py:968] (0/2) Epoch 29, batch 38750, giga_loss[loss=0.3251, simple_loss=0.3772, pruned_loss=0.1365, over 26702.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3472, pruned_loss=0.09813, over 5711729.27 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3563, pruned_loss=0.11, over 5718611.21 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3467, pruned_loss=0.09749, over 5700564.48 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:51:09,409 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:51:42,704 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 04:51:42,942 INFO [train.py:968] (0/2) Epoch 29, batch 38800, giga_loss[loss=0.2406, simple_loss=0.3208, pruned_loss=0.08023, over 28551.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3471, pruned_loss=0.0978, over 5718137.03 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3553, pruned_loss=0.1092, over 5723144.55 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3473, pruned_loss=0.09758, over 5704946.51 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:51:47,168 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.487e+02 1.215e+03 1.437e+03 1.818e+03 7.668e+03, threshold=2.873e+03, percent-clipped=4.0
+2023-03-15 04:51:48,141 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314626.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:51:50,009 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1314629.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:51:58,064 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:52:12,063 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314658.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:52:22,165 INFO [train.py:968] (0/2) Epoch 29, batch 38850, giga_loss[loss=0.2766, simple_loss=0.3574, pruned_loss=0.09792, over 28510.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3468, pruned_loss=0.09698, over 5722886.41 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3554, pruned_loss=0.1093, over 5725497.84 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3468, pruned_loss=0.09661, over 5710228.73 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:52:50,313 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:52:54,632 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9214, 1.3018, 1.1535, 0.1669], device='cuda:0'), covar=tensor([0.4922, 0.3518, 0.4825, 0.7164], device='cuda:0'), in_proj_covar=tensor([0.1833, 0.1712, 0.1644, 0.1489], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 04:53:02,566 INFO [train.py:968] (0/2) Epoch 29, batch 38900, giga_loss[loss=0.2937, simple_loss=0.3768, pruned_loss=0.1053, over 28245.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3466, pruned_loss=0.09705, over 5716905.71 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3551, pruned_loss=0.1091, over 5726442.46 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3466, pruned_loss=0.09672, over 5706030.26 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:53:07,652 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.594e+02 1.193e+03 1.442e+03 1.848e+03 4.022e+03, threshold=2.884e+03, percent-clipped=4.0
+2023-03-15 04:53:17,821 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:53:37,601 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5604, 1.8039, 1.3795, 1.2459], device='cuda:0'), covar=tensor([0.1177, 0.0624, 0.1070, 0.1199], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0448, 0.0522, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 04:53:43,409 INFO [train.py:968] (0/2) Epoch 29, batch 38950, giga_loss[loss=0.2589, simple_loss=0.3358, pruned_loss=0.091, over 28965.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3454, pruned_loss=0.09703, over 5707592.71 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3555, pruned_loss=0.1095, over 5721788.40 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3449, pruned_loss=0.09615, over 5702278.85 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:53:49,095 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:53:55,729 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5146, 1.8023, 1.4574, 1.4916], device='cuda:0'), covar=tensor([0.2744, 0.2815, 0.3204, 0.2552], device='cuda:0'), in_proj_covar=tensor([0.1620, 0.1165, 0.1429, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 04:54:24,207 INFO [train.py:968] (0/2) Epoch 29, batch 39000, libri_loss[loss=0.326, simple_loss=0.3824, pruned_loss=0.1348, over 29539.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3441, pruned_loss=0.09694, over 5703453.12 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.356, pruned_loss=0.1098, over 5718006.56 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3428, pruned_loss=0.0955, over 5701826.92 frames. ], batch size: 80, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:54:24,211 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 04:54:33,459 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9014, 3.7147, 3.6487, 1.6724], device='cuda:0'), covar=tensor([0.0875, 0.0979, 0.0996, 0.2434], device='cuda:0'), in_proj_covar=tensor([0.1295, 0.1201, 0.1005, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 04:54:34,599 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.202, simple_loss=0.3103, pruned_loss=0.04686, over 944034.00 frames. 
+2023-03-15 04:54:34,600 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 04:54:38,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.771e+02 1.276e+03 1.876e+03 3.081e+03 1.052e+04, threshold=3.753e+03, percent-clipped=27.0
+2023-03-15 04:55:12,877 INFO [train.py:968] (0/2) Epoch 29, batch 39050, giga_loss[loss=0.2604, simple_loss=0.3312, pruned_loss=0.09478, over 28798.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3411, pruned_loss=0.09537, over 5709976.89 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3561, pruned_loss=0.1099, over 5722776.54 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3396, pruned_loss=0.09373, over 5704025.13 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:55:33,841 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1314899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:55:54,623 INFO [train.py:968] (0/2) Epoch 29, batch 39100, giga_loss[loss=0.2524, simple_loss=0.3292, pruned_loss=0.0878, over 28795.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3402, pruned_loss=0.09512, over 5708379.29 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3563, pruned_loss=0.1099, over 5725197.02 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3387, pruned_loss=0.09361, over 5701388.63 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:55:59,875 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.550e+02 1.259e+03 1.477e+03 1.991e+03 7.228e+03, threshold=2.954e+03, percent-clipped=6.0
+2023-03-15 04:56:28,892 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3525, 1.7193, 1.0250, 1.3280], device='cuda:0'), covar=tensor([0.1412, 0.0928, 0.1677, 0.1552], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0449, 0.0524, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 04:56:34,032 INFO [train.py:968] (0/2) Epoch 29, batch 39150, giga_loss[loss=0.2707, simple_loss=0.3464, pruned_loss=0.09749, over 28766.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3386, pruned_loss=0.09446, over 5712802.39 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3559, pruned_loss=0.1098, over 5727510.22 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3374, pruned_loss=0.0931, over 5704958.90 frames. ], batch size: 66, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:56:37,370 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9394, 3.7714, 3.5727, 1.6792], device='cuda:0'), covar=tensor([0.0698, 0.0856, 0.0814, 0.2140], device='cuda:0'), in_proj_covar=tensor([0.1299, 0.1203, 0.1008, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 04:56:46,793 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1715, 1.7930, 1.4430, 0.4812], device='cuda:0'), covar=tensor([0.5215, 0.3148, 0.4429, 0.6653], device='cuda:0'), in_proj_covar=tensor([0.1844, 0.1724, 0.1656, 0.1497], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 04:56:48,947 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:56:55,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-15 04:57:05,369 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:11,173 INFO [train.py:968] (0/2) Epoch 29, batch 39200, libri_loss[loss=0.3135, simple_loss=0.3815, pruned_loss=0.1227, over 29524.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3363, pruned_loss=0.09361, over 5706896.80 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3557, pruned_loss=0.1096, over 5723537.90 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3349, pruned_loss=0.09218, over 5702973.29 frames. ], batch size: 82, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:57:14,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.064e+02 1.213e+03 1.430e+03 2.052e+03 7.995e+03, threshold=2.859e+03, percent-clipped=10.0
+2023-03-15 04:57:22,504 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:26,660 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:28,462 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:36,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3055, 1.9634, 1.4806, 0.6229], device='cuda:0'), covar=tensor([0.6427, 0.3485, 0.5166, 0.7212], device='cuda:0'), in_proj_covar=tensor([0.1844, 0.1725, 0.1656, 0.1498], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 04:57:47,524 INFO [train.py:968] (0/2) Epoch 29, batch 39250, giga_loss[loss=0.3038, simple_loss=0.3716, pruned_loss=0.118, over 28269.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3337, pruned_loss=0.09263, over 5704508.72 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3557, pruned_loss=0.1097, over 5721811.14 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3323, pruned_loss=0.09113, over 5702414.69 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:57:50,054 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:54,815 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:58:20,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:58:25,471 INFO [train.py:968] (0/2) Epoch 29, batch 39300, giga_loss[loss=0.2388, simple_loss=0.3254, pruned_loss=0.07615, over 28967.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3318, pruned_loss=0.09178, over 5716533.92 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3558, pruned_loss=0.1099, over 5727936.45 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3298, pruned_loss=0.08978, over 5708836.10 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:58:30,662 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.092e+02 1.202e+03 1.591e+03 2.171e+03 4.450e+03, threshold=3.181e+03, percent-clipped=11.0
+2023-03-15 04:58:40,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8072, 1.0114, 2.8679, 2.7023], device='cuda:0'), covar=tensor([0.1766, 0.2683, 0.0604, 0.1037], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0672, 0.1006, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 04:58:42,134 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4322, 1.3323, 3.8821, 3.3524], device='cuda:0'), covar=tensor([0.1604, 0.2733, 0.0447, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0672, 0.1006, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 04:58:51,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:58:55,378 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:58:57,221 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:59:02,750 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:59:04,499 INFO [train.py:968] (0/2) Epoch 29, batch 39350, giga_loss[loss=0.2543, simple_loss=0.3352, pruned_loss=0.08668, over 29054.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3327, pruned_loss=0.09266, over 5711782.68 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3556, pruned_loss=0.1096, over 5727491.79 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3303, pruned_loss=0.09059, over 5705561.79 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:59:18,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:59:40,994 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-15 04:59:42,159 INFO [train.py:968] (0/2) Epoch 29, batch 39400, giga_loss[loss=0.2622, simple_loss=0.3515, pruned_loss=0.08647, over 28740.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3351, pruned_loss=0.09367, over 5717617.73 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3564, pruned_loss=0.1103, over 5734873.85 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3315, pruned_loss=0.09068, over 5705332.18 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:59:44,317 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:59:47,057 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.441e+02 1.269e+03 1.536e+03 2.216e+03 6.780e+03, threshold=3.072e+03, percent-clipped=5.0
+2023-03-15 04:59:47,383 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:05,591 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-15 05:00:09,430 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:09,530 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:12,159 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:23,547 INFO [train.py:968] (0/2) Epoch 29, batch 39450, giga_loss[loss=0.2423, simple_loss=0.3328, pruned_loss=0.07588, over 28862.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3374, pruned_loss=0.09391, over 5706715.58 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3567, pruned_loss=0.1104, over 5723661.91 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3337, pruned_loss=0.09099, over 5707289.95 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:00:39,195 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:45,356 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:48,240 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:01:05,061 INFO [train.py:968] (0/2) Epoch 29, batch 39500, libri_loss[loss=0.2316, simple_loss=0.3071, pruned_loss=0.07805, over 29383.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3395, pruned_loss=0.09484, over 5694006.63 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3557, pruned_loss=0.1099, over 5716959.57 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3369, pruned_loss=0.09258, over 5700364.74 frames. ], batch size: 67, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:01:10,747 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.244e+02 1.190e+03 1.610e+03 2.161e+03 1.975e+04, threshold=3.219e+03, percent-clipped=12.0
+2023-03-15 05:01:10,944 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:01:28,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.5977, 5.4338, 5.1587, 2.7640], device='cuda:0'), covar=tensor([0.0427, 0.0573, 0.0626, 0.1675], device='cuda:0'), in_proj_covar=tensor([0.1301, 0.1206, 0.1011, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 05:01:40,435 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:01:41,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:01:42,991 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0931, 1.3179, 1.0928, 0.8910], device='cuda:0'), covar=tensor([0.1078, 0.0505, 0.1087, 0.1214], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0448, 0.0523, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 05:01:44,020 INFO [train.py:968] (0/2) Epoch 29, batch 39550, giga_loss[loss=0.2693, simple_loss=0.3387, pruned_loss=0.09992, over 28695.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3403, pruned_loss=0.09432, over 5687353.14 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3553, pruned_loss=0.1096, over 5710762.90 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.338, pruned_loss=0.09219, over 5697777.82 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:01:58,671 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3731, 1.8230, 1.4989, 1.5451], device='cuda:0'), covar=tensor([0.0768, 0.0297, 0.0339, 0.0933], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 05:02:20,067 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:02:27,071 INFO [train.py:968] (0/2) Epoch 29, batch 39600, giga_loss[loss=0.262, simple_loss=0.3443, pruned_loss=0.08982, over 28725.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3394, pruned_loss=0.09294, over 5679702.65 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3555, pruned_loss=0.1096, over 5710952.49 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3372, pruned_loss=0.09103, over 5687558.26 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:02:32,371 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.229e+02 1.319e+03 1.538e+03 1.831e+03 4.110e+03, threshold=3.076e+03, percent-clipped=4.0
+2023-03-15 05:02:37,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2993, 4.1249, 3.9794, 1.6566], device='cuda:0'), covar=tensor([0.0724, 0.0897, 0.0964, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.1299, 0.1206, 0.1011, 0.0750], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 05:03:03,713 INFO [train.py:968] (0/2) Epoch 29, batch 39650, giga_loss[loss=0.2916, simple_loss=0.3624, pruned_loss=0.1104, over 28837.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.339, pruned_loss=0.09243, over 5694382.10 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3555, pruned_loss=0.1096, over 5714481.67 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3369, pruned_loss=0.09061, over 5696997.61 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:03:37,087 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-15 05:03:39,373 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:03:41,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:03:46,420 INFO [train.py:968] (0/2) Epoch 29, batch 39700, giga_loss[loss=0.2305, simple_loss=0.3155, pruned_loss=0.07269, over 29078.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3389, pruned_loss=0.09283, over 5695057.84 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3554, pruned_loss=0.1096, over 5716573.41 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3373, pruned_loss=0.09131, over 5695148.56 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:03:52,701 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.803e+02 1.427e+03 1.679e+03 2.170e+03 8.566e+03, threshold=3.359e+03, percent-clipped=10.0
+2023-03-15 05:04:03,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4694, 1.6370, 1.6317, 1.3932], device='cuda:0'), covar=tensor([0.3642, 0.3149, 0.2479, 0.3134], device='cuda:0'), in_proj_covar=tensor([0.2060, 0.2028, 0.1922, 0.2076], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 05:04:07,187 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:04:07,696 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:04:16,283 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.0926, 3.9391, 3.7249, 1.8240], device='cuda:0'), covar=tensor([0.0683, 0.0797, 0.0696, 0.2126], device='cuda:0'), in_proj_covar=tensor([0.1298, 0.1203, 0.1009, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 05:04:17,035 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:04:19,681 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:04:31,831 INFO [train.py:968] (0/2) Epoch 29, batch 39750, giga_loss[loss=0.308, simple_loss=0.3729, pruned_loss=0.1215, over 29054.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3417, pruned_loss=0.09439, over 5692794.82 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3553, pruned_loss=0.1095, over 5718465.57 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3404, pruned_loss=0.0931, over 5690926.83 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:04:43,379 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:04:58,110 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:05:11,227 INFO [train.py:968] (0/2) Epoch 29, batch 39800, giga_loss[loss=0.3327, simple_loss=0.3918, pruned_loss=0.1368, over 26513.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3455, pruned_loss=0.097, over 5703756.97 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3555, pruned_loss=0.1095, over 5724094.82 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3439, pruned_loss=0.09552, over 5696525.94 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:05:15,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.653e+02 1.397e+03 1.772e+03 2.351e+03 4.513e+03, threshold=3.544e+03, percent-clipped=9.0
+2023-03-15 05:05:49,948 INFO [train.py:968] (0/2) Epoch 29, batch 39850, giga_loss[loss=0.2944, simple_loss=0.3708, pruned_loss=0.109, over 28655.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3475, pruned_loss=0.09717, over 5714167.28 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3557, pruned_loss=0.1095, over 5726172.03 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3457, pruned_loss=0.09566, over 5706070.26 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:05:59,241 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:06:01,066 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:06:02,378 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2732, 1.8865, 1.3951, 0.6164], device='cuda:0'), covar=tensor([0.6518, 0.3274, 0.4938, 0.7296], device='cuda:0'), in_proj_covar=tensor([0.1850, 0.1728, 0.1660, 0.1499], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 05:06:03,112 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:06:26,337 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:06:30,029 INFO [train.py:968] (0/2) Epoch 29, batch 39900, giga_loss[loss=0.2692, simple_loss=0.3484, pruned_loss=0.095, over 28217.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3495, pruned_loss=0.09845, over 5708409.03 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3565, pruned_loss=0.11, over 5718349.96 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3472, pruned_loss=0.09654, over 5708374.86 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:06:35,268 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.790e+02 1.489e+03 1.793e+03 2.640e+03 9.759e+03, threshold=3.586e+03, percent-clipped=10.0
+2023-03-15 05:06:44,243 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:07:09,739 INFO [train.py:968] (0/2) Epoch 29, batch 39950, giga_loss[loss=0.2662, simple_loss=0.3399, pruned_loss=0.09629, over 28721.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3499, pruned_loss=0.099, over 5709150.48 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3568, pruned_loss=0.1102, over 5719920.76 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3477, pruned_loss=0.097, over 5707567.29 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:07:25,877 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:07:31,821 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:07:44,390 INFO [train.py:968] (0/2) Epoch 29, batch 40000, giga_loss[loss=0.3092, simple_loss=0.3802, pruned_loss=0.1192, over 27882.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3503, pruned_loss=0.09927, over 5720909.09 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3572, pruned_loss=0.1104, over 5725680.22 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3478, pruned_loss=0.09703, over 5714011.35 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:07:51,020 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.359e+02 1.337e+03 1.687e+03 2.367e+03 7.079e+03, threshold=3.375e+03, percent-clipped=9.0
+2023-03-15 05:08:08,493 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315851.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:08:23,763 INFO [train.py:968] (0/2) Epoch 29, batch 40050, giga_loss[loss=0.2635, simple_loss=0.3474, pruned_loss=0.08981, over 28617.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3481, pruned_loss=0.09819, over 5712207.51 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.357, pruned_loss=0.1103, over 5718954.45 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3463, pruned_loss=0.0964, over 5712471.70 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:08:32,568 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:08:36,207 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:08:57,861 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:09:03,590 INFO [train.py:968] (0/2) Epoch 29, batch 40100, giga_loss[loss=0.2403, simple_loss=0.3219, pruned_loss=0.07937, over 28868.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3444, pruned_loss=0.09645, over 5714143.42 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3568, pruned_loss=0.1102, over 5724882.71 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3427, pruned_loss=0.09473, over 5708580.28 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:09:09,545 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.344e+03 1.612e+03 2.273e+03 6.246e+03, threshold=3.224e+03, percent-clipped=10.0
+2023-03-15 05:09:43,763 INFO [train.py:968] (0/2) Epoch 29, batch 40150, giga_loss[loss=0.2561, simple_loss=0.3354, pruned_loss=0.08835, over 28999.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3405, pruned_loss=0.09411, over 5716994.17 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3572, pruned_loss=0.1104, over 5726387.80 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3388, pruned_loss=0.09239, over 5711168.74 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:09:49,245 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:10:06,553 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1316000.pt
+2023-03-15 05:10:24,480 INFO [train.py:968] (0/2) Epoch 29, batch 40200, giga_loss[loss=0.2877, simple_loss=0.3575, pruned_loss=0.1089, over 28750.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3412, pruned_loss=0.09401, over 5697746.72 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3578, pruned_loss=0.111, over 5707023.21 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3389, pruned_loss=0.09176, over 5710637.42 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:10:30,972 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.385e+02 1.237e+03 1.542e+03 2.030e+03 8.694e+03, threshold=3.084e+03, percent-clipped=3.0
+2023-03-15 05:10:45,678 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:10:54,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:10:59,993 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1178, 1.3959, 1.3638, 1.0041], device='cuda:0'), covar=tensor([0.1830, 0.2522, 0.1554, 0.1775], device='cuda:0'), in_proj_covar=tensor([0.0941, 0.0717, 0.0989, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 05:11:05,789 INFO [train.py:968] (0/2) Epoch 29, batch 40250, giga_loss[loss=0.2903, simple_loss=0.3713, pruned_loss=0.1047, over 28940.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3422, pruned_loss=0.09332, over 5694835.93 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3579, pruned_loss=0.1109, over 5708652.03 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3401, pruned_loss=0.0914, over 5703470.08 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:11:31,801 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:11:46,028 INFO [train.py:968] (0/2) Epoch 29, batch 40300, giga_loss[loss=0.2609, simple_loss=0.3435, pruned_loss=0.08918, over 28712.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3439, pruned_loss=0.09493, over 5700616.70 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3579, pruned_loss=0.1109, over 5711269.90 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3419, pruned_loss=0.09312, over 5704817.03 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:11:46,338 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:11:48,539 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:11:53,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.514e+02 1.347e+03 1.690e+03 2.164e+03 4.030e+03, threshold=3.380e+03, percent-clipped=3.0
+2023-03-15 05:12:11,254 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:12:14,170 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-15 05:12:22,983 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:12:24,539 INFO [train.py:968] (0/2) Epoch 29, batch 40350, giga_loss[loss=0.2751, simple_loss=0.3449, pruned_loss=0.1027, over 28883.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.342, pruned_loss=0.09484, over 5707261.86 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3576, pruned_loss=0.1107, over 5713042.21 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3403, pruned_loss=0.0932, over 5708600.01 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:12:29,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:12:48,827 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:12:50,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:13:05,511 INFO [train.py:968] (0/2) Epoch 29, batch 40400, giga_loss[loss=0.3036, simple_loss=0.3767, pruned_loss=0.1152, over 27976.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3396, pruned_loss=0.09434, over 5707125.99 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3578, pruned_loss=0.1108, over 5705711.99 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3379, pruned_loss=0.09277, over 5715745.17 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:13:10,105 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316226.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:13:13,180 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.013e+03 1.255e+03 1.577e+03 2.158e+03 7.103e+03, threshold=3.155e+03, percent-clipped=8.0
+2023-03-15 05:13:15,950 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:13:30,877 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:13:44,941 INFO [train.py:968] (0/2) Epoch 29, batch 40450, giga_loss[loss=0.2452, simple_loss=0.3263, pruned_loss=0.0821, over 28997.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3377, pruned_loss=0.09454, over 5692602.17 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.358, pruned_loss=0.1108, over 5695621.25 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3357, pruned_loss=0.0928, over 5709076.55 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:14:18,473 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:20,288 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:24,240 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:24,646 INFO [train.py:968] (0/2) Epoch 29, batch 40500, giga_loss[loss=0.2845, simple_loss=0.3615, pruned_loss=0.1037, over 28703.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3382, pruned_loss=0.09554, over 5693538.75 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3582, pruned_loss=0.1109, over 5693928.32 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3361, pruned_loss=0.09386, over 5707915.76 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:14:27,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:33,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.045e+02 1.390e+03 1.674e+03 2.085e+03 7.315e+03, threshold=3.348e+03, percent-clipped=9.0
+2023-03-15 05:14:44,285 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:50,846 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:54,497 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3766, 1.6485, 1.4402, 1.5714], device='cuda:0'), covar=tensor([0.0739, 0.0321, 0.0332, 0.0854], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 05:15:02,688 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316369.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:15:05,259 INFO [train.py:968] (0/2) Epoch 29, batch 40550, giga_loss[loss=0.2341, simple_loss=0.3119, pruned_loss=0.07817, over 28839.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3364, pruned_loss=0.09493, over 5687169.05 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3583, pruned_loss=0.1111, over 5687098.41 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3342, pruned_loss=0.09313, over 5703902.06 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:15:06,196 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:15:13,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2984, 2.5784, 1.2977, 1.4677], device='cuda:0'), covar=tensor([0.0986, 0.0416, 0.0971, 0.1374], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0572, 0.0413, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 05:15:28,890 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316401.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:15:46,034 INFO [train.py:968] (0/2) Epoch 29, batch 40600, giga_loss[loss=0.2186, simple_loss=0.2911, pruned_loss=0.07308, over 28615.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3325, pruned_loss=0.09311, over 5692625.91 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3585, pruned_loss=0.1111, over 5690264.95 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3302, pruned_loss=0.09134, over 5703225.78 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:15:46,910 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:15:50,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4231, 3.5014, 1.6379, 1.6747], device='cuda:0'), covar=tensor([0.1024, 0.0348, 0.0909, 0.1290], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0571, 0.0413, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 05:15:51,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.939e+02 1.250e+03 1.550e+03 1.917e+03 4.727e+03, threshold=3.099e+03, percent-clipped=4.0
+2023-03-15 05:16:23,231 INFO [train.py:968] (0/2) Epoch 29, batch 40650, giga_loss[loss=0.2253, simple_loss=0.2992, pruned_loss=0.07571, over 28819.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3275, pruned_loss=0.09045, over 5694246.56 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3585, pruned_loss=0.1111, over 5682894.51 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3253, pruned_loss=0.08879, over 5708714.79 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:16:30,352 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:16:41,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3933, 1.5336, 1.4176, 1.5569], device='cuda:0'), covar=tensor([0.0758, 0.0338, 0.0351, 0.0914], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 05:16:44,624 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316499.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:17:01,340 INFO [train.py:968] (0/2) Epoch 29, batch 40700, giga_loss[loss=0.224, simple_loss=0.2939, pruned_loss=0.07703, over 28416.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3279, pruned_loss=0.09042, over 5696707.69 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3584, pruned_loss=0.1111, over 5677231.31 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3254, pruned_loss=0.08861, over 5713445.84 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:17:08,450 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.989e+02 1.296e+03 1.679e+03 2.359e+03 3.784e+03, threshold=3.358e+03, percent-clipped=5.0
+2023-03-15 05:17:29,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7325, 1.9447, 1.4432, 1.6368], device='cuda:0'), covar=tensor([0.0867, 0.0443, 0.0932, 0.0893], device='cuda:0'), in_proj_covar=tensor([0.0413, 0.0449, 0.0524, 0.0462], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 05:17:34,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4280, 1.6793, 1.4329, 1.5789], device='cuda:0'), covar=tensor([0.0757, 0.0330, 0.0341, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 05:17:36,272 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:17:38,172 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:17:38,877 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:17:40,521 INFO [train.py:968] (0/2) Epoch 29, batch 40750, giga_loss[loss=0.2722, simple_loss=0.3531, pruned_loss=0.0957, over 28833.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3319, pruned_loss=0.0922, over 5691997.14 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3582, pruned_loss=0.1111, over 5674717.21 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3294, pruned_loss=0.09023, over 5707755.61 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:18:03,784 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:18:22,355 INFO [train.py:968] (0/2) Epoch 29, batch 40800, libri_loss[loss=0.2749, simple_loss=0.3547, pruned_loss=0.09753, over 29670.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3351, pruned_loss=0.09336, over 5701262.57 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3586, pruned_loss=0.1113, over 5678038.37 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3322, pruned_loss=0.09122, over 5710973.56 frames. ], batch size: 91, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:18:22,635 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:18:24,493 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:18:25,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:18:28,247 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.124e+02 1.328e+03 1.739e+03 2.466e+03 5.572e+03, threshold=3.478e+03, percent-clipped=9.0
+2023-03-15 05:18:45,996 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:18:59,551 INFO [train.py:968] (0/2) Epoch 29, batch 40850, giga_loss[loss=0.2595, simple_loss=0.3404, pruned_loss=0.08929, over 29076.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3398, pruned_loss=0.09563, over 5693171.86 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3594, pruned_loss=0.1118, over 5675603.64 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3359, pruned_loss=0.09285, over 5704610.00 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:19:29,469 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:19:42,176 INFO [train.py:968] (0/2) Epoch 29, batch 40900, giga_loss[loss=0.2827, simple_loss=0.3571, pruned_loss=0.1042, over 29144.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3425, pruned_loss=0.09678, over 5698376.94 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3594, pruned_loss=0.1119, over 5679658.58 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3391, pruned_loss=0.09419, over 5704236.02 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:19:48,017 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.624e+02 1.256e+03 1.513e+03 1.980e+03 6.289e+03, threshold=3.026e+03, percent-clipped=5.0
+2023-03-15 05:20:02,150 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2257, 2.6279, 1.2131, 1.4582], device='cuda:0'), covar=tensor([0.1016, 0.0456, 0.0976, 0.1375], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0574, 0.0415, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 05:20:19,001 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:20:19,907 INFO [train.py:968] (0/2) Epoch 29, batch 40950, giga_loss[loss=0.2587, simple_loss=0.3351, pruned_loss=0.09112, over 28508.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3451, pruned_loss=0.09845, over 5699298.41 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3595, pruned_loss=0.112, over 5679705.74 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3416, pruned_loss=0.09571, over 5704843.22 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:20:20,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:20:25,428 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2173, 0.9818, 1.0490, 1.3762], device='cuda:0'), covar=tensor([0.0759, 0.0392, 0.0368, 0.0925], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 05:20:50,373 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9291, 1.2771, 2.8524, 2.7416], device='cuda:0'), covar=tensor([0.1598, 0.2517, 0.0655, 0.1503], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0676, 0.1015, 0.0990], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 05:20:52,050 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:21:12,592 INFO [train.py:968] (0/2) Epoch 29, batch 41000, giga_loss[loss=0.3059, simple_loss=0.38, pruned_loss=0.1159, over 28990.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3509, pruned_loss=0.1038, over 5697831.32 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3593, pruned_loss=0.1119, over 5681782.72 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3482, pruned_loss=0.1016, over 5700489.08 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:21:22,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.162e+03 1.680e+03 2.114e+03 2.816e+03 6.893e+03, threshold=4.228e+03, percent-clipped=22.0
+2023-03-15 05:21:56,186 INFO [train.py:968] (0/2) Epoch 29, batch 41050, giga_loss[loss=0.2908, simple_loss=0.362, pruned_loss=0.1098, over 28630.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3566, pruned_loss=0.108, over 5690977.59 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3593, pruned_loss=0.1119, over 5676754.26 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3542, pruned_loss=0.1061, over 5697363.19 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:22:00,875 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316874.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:22:19,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3717, 1.5178, 1.4229, 1.2561], device='cuda:0'), covar=tensor([0.2516, 0.2540, 0.2054, 0.2462], device='cuda:0'), in_proj_covar=tensor([0.2069, 0.2040, 0.1940, 0.2081], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 05:22:41,845 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:22:43,165 INFO [train.py:968] (0/2) Epoch 29, batch 41100, giga_loss[loss=0.3058, simple_loss=0.3747, pruned_loss=0.1185, over 28628.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3642, pruned_loss=0.1135, over 5676355.94 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3594, pruned_loss=0.1119, over 5666248.06 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3623, pruned_loss=0.112, over 5691792.89 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:22:51,947 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.290e+03 1.923e+03 2.504e+03 3.328e+03 7.786e+03, threshold=5.007e+03, percent-clipped=11.0
+2023-03-15 05:23:03,053 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:23:28,043 INFO [train.py:968] (0/2) Epoch 29, batch 41150, libri_loss[loss=0.3219, simple_loss=0.3841, pruned_loss=0.1298, over 29280.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.369, pruned_loss=0.1173, over 5682205.88 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3595, pruned_loss=0.1119, over 5668614.45 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3675, pruned_loss=0.1161, over 5692384.76 frames. ], batch size: 97, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:23:59,170 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1317007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:24:07,007 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317017.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:24:10,472 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317020.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:24:10,810 INFO [train.py:968] (0/2) Epoch 29, batch 41200, giga_loss[loss=0.3323, simple_loss=0.3899, pruned_loss=0.1374, over 28553.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.374, pruned_loss=0.1215, over 5684992.29 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3599, pruned_loss=0.1122, over 5673761.32 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3728, pruned_loss=0.1205, over 5689109.51 frames. ], batch size: 262, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:24:20,684 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.332e+03 2.083e+03 2.565e+03 3.231e+03 7.307e+03, threshold=5.129e+03, percent-clipped=7.0
+2023-03-15 05:24:40,687 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317049.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:24:43,920 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1317052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:25:03,685 INFO [train.py:968] (0/2) Epoch 29, batch 41250, giga_loss[loss=0.3946, simple_loss=0.4171, pruned_loss=0.1861, over 23323.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3789, pruned_loss=0.1267, over 5659920.97 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3598, pruned_loss=0.1121, over 5674619.44 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3784, pruned_loss=0.1263, over 5662252.57 frames. ], batch size: 705, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:25:13,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1317082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:25:15,841 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:25:18,164 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:25:22,776 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5701, 1.8275, 1.7991, 1.4920], device='cuda:0'), covar=tensor([0.2930, 0.2583, 0.2677, 0.2821], device='cuda:0'), in_proj_covar=tensor([0.2063, 0.2035, 0.1933, 0.2078], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 05:25:47,550 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:25:51,830 INFO [train.py:968] (0/2) Epoch 29, batch 41300, giga_loss[loss=0.3109, simple_loss=0.3744, pruned_loss=0.1237, over 28904.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.38, pruned_loss=0.1287, over 5649329.53 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3599, pruned_loss=0.1122, over 5661063.50 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3802, pruned_loss=0.1286, over 5661913.09 frames. ], batch size: 145, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:25:57,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0089, 1.2920, 1.3643, 1.1605], device='cuda:0'), covar=tensor([0.1601, 0.1118, 0.1831, 0.1256], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0767, 0.0739, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 05:26:06,645 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.328e+03 2.049e+03 2.519e+03 3.338e+03 7.835e+03, threshold=5.039e+03, percent-clipped=7.0
+2023-03-15 05:26:43,457 INFO [train.py:968] (0/2) Epoch 29, batch 41350, giga_loss[loss=0.38, simple_loss=0.4084, pruned_loss=0.1758, over 23451.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3841, pruned_loss=0.1336, over 5637521.53 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3595, pruned_loss=0.112, over 5666097.94 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.385, pruned_loss=0.1341, over 5642574.86 frames. ], batch size: 705, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:26:47,935 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1317176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:27:31,403 INFO [train.py:968] (0/2) Epoch 29, batch 41400, libri_loss[loss=0.2959, simple_loss=0.3668, pruned_loss=0.1125, over 28649.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3878, pruned_loss=0.1371, over 5637600.79 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3593, pruned_loss=0.1119, over 5671708.71 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3896, pruned_loss=0.1385, over 5635627.93 frames. ], batch size: 106, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:27:34,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:27:37,823 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:27:42,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.310e+03 2.247e+03 2.735e+03 3.684e+03 7.795e+03, threshold=5.470e+03, percent-clipped=6.0
+2023-03-15 05:28:05,459 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:28:16,397 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6151, 1.6752, 1.8075, 1.3931], device='cuda:0'), covar=tensor([0.1787, 0.2620, 0.1466, 0.1761], device='cuda:0'), in_proj_covar=tensor([0.0934, 0.0715, 0.0984, 0.0884], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 05:28:20,972 INFO [train.py:968] (0/2) Epoch 29, batch 41450, giga_loss[loss=0.3917, simple_loss=0.4163, pruned_loss=0.1836, over 23519.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3895, pruned_loss=0.1386, over 5633726.26 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3597, pruned_loss=0.1121, over 5679369.21 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3916, pruned_loss=0.1403, over 5624117.18 frames. ], batch size: 705, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:28:23,436 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6408, 1.8407, 1.5092, 1.6964], device='cuda:0'), covar=tensor([0.2513, 0.2646, 0.2965, 0.2331], device='cuda:0'), in_proj_covar=tensor([0.1621, 0.1167, 0.1433, 0.1020], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 05:28:46,422 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1317294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:29:12,713 INFO [train.py:968] (0/2) Epoch 29, batch 41500, giga_loss[loss=0.3037, simple_loss=0.3729, pruned_loss=0.1173, over 28903.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3894, pruned_loss=0.1393, over 5631614.27 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.36, pruned_loss=0.1123, over 5682152.06 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3913, pruned_loss=0.141, over 5620698.86 frames. ], batch size: 145, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:29:22,305 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.236e+03 2.027e+03 2.660e+03 3.603e+03 2.008e+04, threshold=5.320e+03, percent-clipped=13.0
+2023-03-15 05:29:57,448 INFO [train.py:968] (0/2) Epoch 29, batch 41550, giga_loss[loss=0.2992, simple_loss=0.3598, pruned_loss=0.1193, over 28823.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3877, pruned_loss=0.1383, over 5648111.19 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3604, pruned_loss=0.1128, over 5684823.50 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3898, pruned_loss=0.1401, over 5635723.70 frames. ], batch size: 112, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:30:09,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1317382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:30:16,028 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3394, 1.4639, 1.3570, 1.3012], device='cuda:0'), covar=tensor([0.1883, 0.2082, 0.2008, 0.1871], device='cuda:0'), in_proj_covar=tensor([0.2070, 0.2043, 0.1938, 0.2087], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 05:30:45,173 INFO [train.py:968] (0/2) Epoch 29, batch 41600, giga_loss[loss=0.346, simple_loss=0.4059, pruned_loss=0.143, over 28931.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3868, pruned_loss=0.1362, over 5645066.91 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3605, pruned_loss=0.1129, over 5672520.99 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3892, pruned_loss=0.1382, over 5645375.83 frames. ], batch size: 164, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:30:52,530 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1317427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:30:57,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.264e+03 1.935e+03 2.335e+03 3.249e+03 6.464e+03, threshold=4.670e+03, percent-clipped=6.0
+2023-03-15 05:31:01,537 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:31:04,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:31:33,415 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:31:35,366 INFO [train.py:968] (0/2) Epoch 29, batch 41650, giga_loss[loss=0.3364, simple_loss=0.3892, pruned_loss=0.1418, over 28775.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3859, pruned_loss=0.1345, over 5657151.31 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3602, pruned_loss=0.1127, over 5679714.06 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3887, pruned_loss=0.137, over 5650143.56 frames. ], batch size: 99, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:32:20,109 INFO [train.py:968] (0/2) Epoch 29, batch 41700, libri_loss[loss=0.2899, simple_loss=0.3627, pruned_loss=0.1086, over 29163.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3857, pruned_loss=0.1336, over 5643974.32 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3604, pruned_loss=0.1129, over 5669647.21 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.389, pruned_loss=0.1364, over 5646849.46 frames. ], batch size: 101, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:32:21,586 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9353, 3.7440, 3.5736, 1.9566], device='cuda:0'), covar=tensor([0.0938, 0.1127, 0.1193, 0.2064], device='cuda:0'), in_proj_covar=tensor([0.1320, 0.1221, 0.1024, 0.0757], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 05:32:22,933 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:32:25,505 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:32:28,874 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.105e+03 1.844e+03 2.138e+03 2.869e+03 6.996e+03, threshold=4.276e+03, percent-clipped=5.0
+2023-03-15 05:32:49,511 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1317551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:32:55,988 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:33:05,618 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3534, 3.5157, 1.5744, 1.4544], device='cuda:0'), covar=tensor([0.1007, 0.0337, 0.0903, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0576, 0.0415, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 05:33:08,253 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:33:08,630 INFO [train.py:968] (0/2) Epoch 29, batch 41750, libri_loss[loss=0.2678, simple_loss=0.3381, pruned_loss=0.09876, over 29587.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3835, pruned_loss=0.1315, over 5649469.99 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3607, pruned_loss=0.1131, over 5675965.27 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3865, pruned_loss=0.134, over 5645470.01 frames. ], batch size: 74, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:33:11,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:33:37,222 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:33:54,139 INFO [train.py:968] (0/2) Epoch 29, batch 41800, giga_loss[loss=0.3259, simple_loss=0.3849, pruned_loss=0.1335, over 28345.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3825, pruned_loss=0.1299, over 5639542.05 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3605, pruned_loss=0.113, over 5672810.94 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3857, pruned_loss=0.1324, over 5638910.45 frames. ], batch size: 368, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:34:06,369 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.232e+03 1.923e+03 2.324e+03 3.152e+03 7.415e+03, threshold=4.649e+03, percent-clipped=10.0
+2023-03-15 05:34:47,096 INFO [train.py:968] (0/2) Epoch 29, batch 41850, giga_loss[loss=0.3256, simple_loss=0.3713, pruned_loss=0.1399, over 23942.00 frames. ], tot_loss[loss=0.314, simple_loss=0.378, pruned_loss=0.125, over 5653037.26 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3603, pruned_loss=0.1129, over 5673989.89 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3807, pruned_loss=0.1271, over 5651298.03 frames. ], batch size: 705, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:34:53,472 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-15 05:34:57,681 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3496, 1.6294, 1.2175, 1.2163], device='cuda:0'), covar=tensor([0.1166, 0.0589, 0.1168, 0.1204], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0452, 0.0525, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 05:35:08,631 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:35:10,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:35:34,045 INFO [train.py:968] (0/2) Epoch 29, batch 41900, giga_loss[loss=0.304, simple_loss=0.3604, pruned_loss=0.1238, over 26527.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3743, pruned_loss=0.122, over 5654884.45 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3603, pruned_loss=0.113, over 5676317.46 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3769, pruned_loss=0.1239, over 5650863.67 frames. ], batch size: 555, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:35:39,399 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:35:45,004 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.224e+03 1.887e+03 2.247e+03 3.620e+03 9.267e+03, threshold=4.495e+03, percent-clipped=10.0
+2023-03-15 05:36:21,754 INFO [train.py:968] (0/2) Epoch 29, batch 41950, giga_loss[loss=0.3818, simple_loss=0.4146, pruned_loss=0.1745, over 26556.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3727, pruned_loss=0.1216, over 5644493.14 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3605, pruned_loss=0.113, over 5680056.12 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3748, pruned_loss=0.1233, over 5637480.37 frames. ], batch size: 555, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:37:07,313 INFO [train.py:968] (0/2) Epoch 29, batch 42000, giga_loss[loss=0.3107, simple_loss=0.3753, pruned_loss=0.1231, over 29073.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.374, pruned_loss=0.1226, over 5661254.76 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3607, pruned_loss=0.1132, over 5680554.44 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3758, pruned_loss=0.124, over 5654648.88 frames. ], batch size: 155, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:37:07,317 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 05:37:16,895 INFO [train.py:1012] (0/2) Epoch 29, validation: loss=0.2001, simple_loss=0.3077, pruned_loss=0.04619, over 944034.00 frames. 
+2023-03-15 05:37:16,896 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 05:37:29,445 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.275e+03 1.784e+03 2.252e+03 3.205e+03 1.195e+04, threshold=4.504e+03, percent-clipped=8.0
+2023-03-15 05:38:07,003 INFO [train.py:968] (0/2) Epoch 29, batch 42050, giga_loss[loss=0.3755, simple_loss=0.4134, pruned_loss=0.1688, over 27585.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3731, pruned_loss=0.1215, over 5660626.27 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3611, pruned_loss=0.1134, over 5675872.08 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3743, pruned_loss=0.1225, over 5659447.11 frames. ], batch size: 472, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:38:54,768 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1317920.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:38:55,181 INFO [train.py:968] (0/2) Epoch 29, batch 42100, giga_loss[loss=0.2767, simple_loss=0.3548, pruned_loss=0.0993, over 28942.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3712, pruned_loss=0.1188, over 5666235.45 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3611, pruned_loss=0.1134, over 5674436.32 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3726, pruned_loss=0.1199, over 5666761.31 frames. ], batch size: 174, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:39:10,002 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.729e+02 1.636e+03 2.048e+03 2.518e+03 7.042e+03, threshold=4.096e+03, percent-clipped=4.0
+2023-03-15 05:39:19,568 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9556, 2.4115, 2.1540, 2.0709], device='cuda:0'), covar=tensor([0.2071, 0.1657, 0.1720, 0.1772], device='cuda:0'), in_proj_covar=tensor([0.0516, 0.0768, 0.0740, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 05:39:44,731 INFO [train.py:968] (0/2) Epoch 29, batch 42150, giga_loss[loss=0.3167, simple_loss=0.3923, pruned_loss=0.1206, over 28863.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3714, pruned_loss=0.1167, over 5675829.98 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3604, pruned_loss=0.113, over 5677021.18 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3734, pruned_loss=0.1181, over 5673606.83 frames. ], batch size: 213, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:40:02,702 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1317987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:40:04,879 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2830, 1.9073, 5.4992, 3.8854], device='cuda:0'), covar=tensor([0.1462, 0.2554, 0.0438, 0.0883], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0686, 0.1029, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 05:40:13,797 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1318000.pt
+2023-03-15 05:40:33,770 INFO [train.py:968] (0/2) Epoch 29, batch 42200, giga_loss[loss=0.3293, simple_loss=0.3863, pruned_loss=0.1361, over 29033.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3722, pruned_loss=0.1171, over 5669064.27 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3607, pruned_loss=0.1132, over 5675866.30 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3737, pruned_loss=0.1181, over 5668384.72 frames. ], batch size: 136, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:40:44,003 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1318032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:40:45,054 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.309e+03 1.857e+03 2.579e+03 3.502e+03 7.133e+03, threshold=5.157e+03, percent-clipped=17.0
+2023-03-15 05:41:12,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4319, 4.2906, 4.1109, 1.9897], device='cuda:0'), covar=tensor([0.0643, 0.0746, 0.0770, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.1330, 0.1233, 0.1031, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 05:41:17,442 INFO [train.py:968] (0/2) Epoch 29, batch 42250, giga_loss[loss=0.3071, simple_loss=0.3842, pruned_loss=0.115, over 28659.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3715, pruned_loss=0.1174, over 5671609.97 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3599, pruned_loss=0.1129, over 5679861.86 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3738, pruned_loss=0.1185, over 5667381.79 frames. ], batch size: 262, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:42:01,202 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3981, 1.6103, 1.2443, 1.1477], device='cuda:0'), covar=tensor([0.1085, 0.0543, 0.1088, 0.1170], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0455, 0.0528, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 05:42:03,032 INFO [train.py:968] (0/2) Epoch 29, batch 42300, giga_loss[loss=0.2808, simple_loss=0.342, pruned_loss=0.1098, over 28936.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3701, pruned_loss=0.117, over 5677533.52 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3598, pruned_loss=0.1131, over 5683204.05 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3721, pruned_loss=0.1178, over 5671186.20 frames. ], batch size: 106, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:42:14,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.305e+03 1.932e+03 2.476e+03 3.965e+03 1.131e+04, threshold=4.951e+03, percent-clipped=7.0
+2023-03-15 05:42:46,639 INFO [train.py:968] (0/2) Epoch 29, batch 42350, libri_loss[loss=0.309, simple_loss=0.374, pruned_loss=0.122, over 27872.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3683, pruned_loss=0.117, over 5668029.27 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.36, pruned_loss=0.1134, over 5678222.73 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.37, pruned_loss=0.1174, over 5667213.76 frames. ], batch size: 116, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:43:32,872 INFO [train.py:968] (0/2) Epoch 29, batch 42400, giga_loss[loss=0.2606, simple_loss=0.3375, pruned_loss=0.09185, over 28612.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3667, pruned_loss=0.1164, over 5664280.52 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.36, pruned_loss=0.1133, over 5681640.38 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3681, pruned_loss=0.1169, over 5660271.47 frames. ], batch size: 85, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 05:43:46,352 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.295e+03 1.866e+03 2.273e+03 2.856e+03 7.798e+03, threshold=4.547e+03, percent-clipped=6.0
+2023-03-15 05:43:47,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5577, 2.1997, 1.6158, 0.7543], device='cuda:0'), covar=tensor([0.7302, 0.4023, 0.4799, 0.8337], device='cuda:0'), in_proj_covar=tensor([0.1871, 0.1754, 0.1673, 0.1516], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 05:44:20,126 INFO [train.py:968] (0/2) Epoch 29, batch 42450, giga_loss[loss=0.2903, simple_loss=0.3698, pruned_loss=0.1054, over 28697.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3663, pruned_loss=0.115, over 5674377.06 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1129, over 5683126.73 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3681, pruned_loss=0.1158, over 5669085.52 frames. ], batch size: 242, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 05:44:39,508 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1318295.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:44:59,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9955, 2.9009, 1.8133, 1.1497], device='cuda:0'), covar=tensor([0.8983, 0.4139, 0.4633, 0.8402], device='cuda:0'), in_proj_covar=tensor([0.1872, 0.1757, 0.1673, 0.1518], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 05:45:04,241 INFO [train.py:968] (0/2) Epoch 29, batch 42500, giga_loss[loss=0.3141, simple_loss=0.3936, pruned_loss=0.1173, over 28956.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3668, pruned_loss=0.1139, over 5687321.48 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.36, pruned_loss=0.1132, over 5687464.70 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.368, pruned_loss=0.1144, over 5679270.22 frames. ], batch size: 145, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:45:15,637 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 1.577e+03 2.017e+03 2.617e+03 4.752e+03, threshold=4.035e+03, percent-clipped=1.0
+2023-03-15 05:45:45,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1318362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:45:53,514 INFO [train.py:968] (0/2) Epoch 29, batch 42550, giga_loss[loss=0.2774, simple_loss=0.36, pruned_loss=0.09735, over 28958.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3669, pruned_loss=0.114, over 5682794.60 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3601, pruned_loss=0.1133, over 5679685.50 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3678, pruned_loss=0.1143, over 5682579.60 frames. ], batch size: 145, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:46:26,845 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1318407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:46:39,680 INFO [train.py:968] (0/2) Epoch 29, batch 42600, giga_loss[loss=0.2971, simple_loss=0.3675, pruned_loss=0.1133, over 29046.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3655, pruned_loss=0.1137, over 5678321.51 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3605, pruned_loss=0.1135, over 5680325.22 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.366, pruned_loss=0.1137, over 5677485.87 frames. ], batch size: 155, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:46:51,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.836e+02 1.778e+03 2.428e+03 3.068e+03 1.216e+04, threshold=4.856e+03, percent-clipped=15.0
+2023-03-15 05:46:55,393 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1318438.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:46:57,982 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1318441.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:47:19,895 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6894, 1.6430, 1.8652, 1.4609], device='cuda:0'), covar=tensor([0.1814, 0.2462, 0.1506, 0.1752], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0721, 0.0991, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 05:47:21,668 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1318470.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:47:22,071 INFO [train.py:968] (0/2) Epoch 29, batch 42650, giga_loss[loss=0.297, simple_loss=0.3649, pruned_loss=0.1146, over 28858.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3648, pruned_loss=0.1139, over 5680163.31 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3603, pruned_loss=0.1133, over 5685580.19 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3655, pruned_loss=0.1142, over 5674441.38 frames. ], batch size: 199, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:47:43,854 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6801, 1.8846, 1.6359, 1.5692], device='cuda:0'), covar=tensor([0.2320, 0.2251, 0.2281, 0.2219], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1164, 0.1430, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 05:47:53,920 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1318505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:47:55,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1318508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:48:07,361 INFO [train.py:968] (0/2) Epoch 29, batch 42700, giga_loss[loss=0.2878, simple_loss=0.3657, pruned_loss=0.1049, over 28948.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3629, pruned_loss=0.1135, over 5681098.96 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5691333.29 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.364, pruned_loss=0.1141, over 5671391.53 frames. ], batch size: 213, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:48:13,518 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5138, 1.4719, 1.6883, 1.3335], device='cuda:0'), covar=tensor([0.1508, 0.2322, 0.1287, 0.1570], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0721, 0.0991, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 05:48:17,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2181, 1.5331, 1.5132, 1.3938], device='cuda:0'), covar=tensor([0.2211, 0.1815, 0.2581, 0.2012], device='cuda:0'), in_proj_covar=tensor([0.0516, 0.0768, 0.0740, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 05:48:21,934 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.254e+03 2.000e+03 2.533e+03 3.183e+03 9.197e+03, threshold=5.065e+03, percent-clipped=8.0
+2023-03-15 05:48:22,859 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1318537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:48:35,540 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1318550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:48:38,406 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1318553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:48:55,098 INFO [train.py:968] (0/2) Epoch 29, batch 42750, giga_loss[loss=0.3822, simple_loss=0.4074, pruned_loss=0.1785, over 26616.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3634, pruned_loss=0.1149, over 5674933.21 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.36, pruned_loss=0.1128, over 5693620.99 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3642, pruned_loss=0.1154, over 5665066.33 frames. ], batch size: 555, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:49:06,554 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1318582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:49:40,387 INFO [train.py:968] (0/2) Epoch 29, batch 42800, giga_loss[loss=0.3025, simple_loss=0.3514, pruned_loss=0.1268, over 23882.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3614, pruned_loss=0.1138, over 5678539.10 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3594, pruned_loss=0.1124, over 5694592.21 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3627, pruned_loss=0.1146, over 5669403.78 frames. ], batch size: 705, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:49:52,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.331e+03 1.917e+03 2.803e+03 4.095e+03 8.463e+03, threshold=5.607e+03, percent-clipped=10.0
+2023-03-15 05:50:05,744 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1318647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:50:11,962 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1318655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:50:18,194 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3854, 1.5757, 1.2204, 1.1540], device='cuda:0'), covar=tensor([0.0992, 0.0488, 0.1029, 0.1070], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0454, 0.0528, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 05:50:24,265 INFO [train.py:968] (0/2) Epoch 29, batch 42850, giga_loss[loss=0.3382, simple_loss=0.3919, pruned_loss=0.1422, over 28956.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3625, pruned_loss=0.1146, over 5691758.52 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3597, pruned_loss=0.1124, over 5701330.69 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3633, pruned_loss=0.1153, over 5677707.83 frames. ], batch size: 227, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:50:30,139 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5625, 4.7166, 1.7402, 1.7925], device='cuda:0'), covar=tensor([0.1080, 0.0340, 0.0888, 0.1388], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0581, 0.0417, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 05:50:35,387 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 05:51:10,264 INFO [train.py:968] (0/2) Epoch 29, batch 42900, giga_loss[loss=0.2959, simple_loss=0.3673, pruned_loss=0.1122, over 28853.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3634, pruned_loss=0.1147, over 5684711.39 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3598, pruned_loss=0.1124, over 5693083.79 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.364, pruned_loss=0.1154, over 5679692.22 frames. ], batch size: 186, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:51:14,302 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0533, 4.8923, 4.6645, 2.3291], device='cuda:0'), covar=tensor([0.0486, 0.0610, 0.0616, 0.1859], device='cuda:0'), in_proj_covar=tensor([0.1335, 0.1237, 0.1035, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 05:51:21,653 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3517, 1.6143, 1.5652, 1.4235], device='cuda:0'), covar=tensor([0.2199, 0.2159, 0.2531, 0.2299], device='cuda:0'), in_proj_covar=tensor([0.0516, 0.0769, 0.0741, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 05:51:25,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.814e+03 2.344e+03 3.590e+03 7.434e+03, threshold=4.688e+03, percent-clipped=4.0
+2023-03-15 05:51:48,009 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-15 05:51:59,169 INFO [train.py:968] (0/2) Epoch 29, batch 42950, libri_loss[loss=0.2853, simple_loss=0.3627, pruned_loss=0.1039, over 29645.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3643, pruned_loss=0.1147, over 5683067.56 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3598, pruned_loss=0.1122, over 5697175.91 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.365, pruned_loss=0.1154, over 5674958.37 frames. ], batch size: 88, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:52:02,718 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5574, 4.3894, 4.1686, 1.9940], device='cuda:0'), covar=tensor([0.0595, 0.0731, 0.0768, 0.2084], device='cuda:0'), in_proj_covar=tensor([0.1335, 0.1237, 0.1035, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 05:52:26,198 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8483, 3.6658, 3.5131, 1.6685], device='cuda:0'), covar=tensor([0.0823, 0.0929, 0.0915, 0.2297], device='cuda:0'), in_proj_covar=tensor([0.1335, 0.1238, 0.1035, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 05:52:39,817 INFO [train.py:968] (0/2) Epoch 29, batch 43000, giga_loss[loss=0.2555, simple_loss=0.3358, pruned_loss=0.08764, over 28555.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3637, pruned_loss=0.1135, over 5681140.20 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3596, pruned_loss=0.1121, over 5699193.40 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3644, pruned_loss=0.1142, over 5672315.77 frames. ], batch size: 85, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:52:54,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.036e+03 1.669e+03 2.060e+03 2.818e+03 7.416e+03, threshold=4.119e+03, percent-clipped=5.0
+2023-03-15 05:53:26,159 INFO [train.py:968] (0/2) Epoch 29, batch 43050, giga_loss[loss=0.2799, simple_loss=0.3533, pruned_loss=0.1033, over 28990.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.364, pruned_loss=0.1138, over 5663404.10 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3596, pruned_loss=0.1121, over 5685501.63 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3647, pruned_loss=0.1144, over 5667817.12 frames. ], batch size: 145, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:54:17,086 INFO [train.py:968] (0/2) Epoch 29, batch 43100, libri_loss[loss=0.275, simple_loss=0.3539, pruned_loss=0.09804, over 29511.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3664, pruned_loss=0.117, over 5658942.05 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3593, pruned_loss=0.1119, over 5688396.62 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3673, pruned_loss=0.1178, over 5659143.74 frames. ], batch size: 83, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:54:30,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.160e+03 1.888e+03 2.524e+03 4.326e+03 1.181e+04, threshold=5.048e+03, percent-clipped=25.0
+2023-03-15 05:54:57,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2056, 1.5365, 1.5097, 1.3596], device='cuda:0'), covar=tensor([0.2238, 0.1792, 0.2475, 0.2001], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0767, 0.0740, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 05:55:02,498 INFO [train.py:968] (0/2) Epoch 29, batch 43150, giga_loss[loss=0.2398, simple_loss=0.3177, pruned_loss=0.08092, over 28518.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3685, pruned_loss=0.1196, over 5662608.98 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3594, pruned_loss=0.1119, over 5690817.62 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3694, pruned_loss=0.1204, over 5660160.18 frames. ], batch size: 60, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:55:51,119 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-15 05:55:54,415 INFO [train.py:968] (0/2) Epoch 29, batch 43200, giga_loss[loss=0.2909, simple_loss=0.3651, pruned_loss=0.1083, over 29035.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3694, pruned_loss=0.1219, over 5655531.33 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3594, pruned_loss=0.1119, over 5693027.32 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3704, pruned_loss=0.1227, over 5650478.70 frames. ], batch size: 136, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 05:55:55,360 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1319022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:56:02,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1319030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:56:07,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4967, 1.8077, 1.7168, 1.6310], device='cuda:0'), covar=tensor([0.2110, 0.1998, 0.2307, 0.2068], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0767, 0.0739, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 05:56:08,307 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.308e+03 1.998e+03 2.498e+03 3.376e+03 1.184e+04, threshold=4.995e+03, percent-clipped=8.0
+2023-03-15 05:56:39,976 INFO [train.py:968] (0/2) Epoch 29, batch 43250, giga_loss[loss=0.3309, simple_loss=0.3826, pruned_loss=0.1396, over 29014.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3714, pruned_loss=0.1235, over 5662150.82 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3598, pruned_loss=0.112, over 5693727.20 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3722, pruned_loss=0.1244, over 5656656.49 frames. ], batch size: 128, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:56:48,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9622, 3.8128, 3.6470, 1.6694], device='cuda:0'), covar=tensor([0.0768, 0.0914, 0.0840, 0.2131], device='cuda:0'), in_proj_covar=tensor([0.1336, 0.1235, 0.1034, 0.0764], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 05:57:21,308 INFO [train.py:968] (0/2) Epoch 29, batch 43300, giga_loss[loss=0.3097, simple_loss=0.3645, pruned_loss=0.1275, over 28772.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3687, pruned_loss=0.1216, over 5670004.61 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3594, pruned_loss=0.1121, over 5691755.14 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3699, pruned_loss=0.1227, over 5667103.17 frames. ], batch size: 92, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:57:27,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5307, 1.5614, 1.7293, 1.4096], device='cuda:0'), covar=tensor([0.1374, 0.2063, 0.1215, 0.1633], device='cuda:0'), in_proj_covar=tensor([0.0936, 0.0718, 0.0985, 0.0884], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 05:57:34,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.180e+03 1.947e+03 2.384e+03 3.172e+03 6.725e+03, threshold=4.769e+03, percent-clipped=4.0
+2023-03-15 05:57:58,511 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1319165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:00,730 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1319168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:03,240 INFO [train.py:968] (0/2) Epoch 29, batch 43350, giga_loss[loss=0.2699, simple_loss=0.3498, pruned_loss=0.095, over 28235.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3667, pruned_loss=0.12, over 5672691.65 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3591, pruned_loss=0.112, over 5695073.49 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3683, pruned_loss=0.1211, over 5666756.39 frames. ], batch size: 77, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:58:06,789 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1319173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:08,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1319176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:25,196 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1319197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:34,766 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1319205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:46,879 INFO [train.py:968] (0/2) Epoch 29, batch 43400, giga_loss[loss=0.309, simple_loss=0.3767, pruned_loss=0.1206, over 28896.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3674, pruned_loss=0.1186, over 5683440.49 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3596, pruned_loss=0.1123, over 5698902.87 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3683, pruned_loss=0.1193, over 5674968.66 frames. ], batch size: 213, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:58:57,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6596, 1.9247, 1.5232, 1.8899], device='cuda:0'), covar=tensor([0.2995, 0.2914, 0.3377, 0.2570], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1163, 0.1428, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 05:59:00,610 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.951e+02 1.656e+03 2.125e+03 3.193e+03 8.202e+03, threshold=4.250e+03, percent-clipped=7.0
+2023-03-15 05:59:30,199 INFO [train.py:968] (0/2) Epoch 29, batch 43450, giga_loss[loss=0.4117, simple_loss=0.4245, pruned_loss=0.1994, over 26448.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3653, pruned_loss=0.1171, over 5660227.14 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3598, pruned_loss=0.1124, over 5685916.49 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3661, pruned_loss=0.1177, over 5664423.24 frames. ], batch size: 555, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:00:12,294 INFO [train.py:968] (0/2) Epoch 29, batch 43500, giga_loss[loss=0.2866, simple_loss=0.3462, pruned_loss=0.1135, over 28698.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.363, pruned_loss=0.1162, over 5668203.37 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3595, pruned_loss=0.1123, over 5692799.32 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3641, pruned_loss=0.117, over 5664783.94 frames. ], batch size: 92, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:00:25,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.815e+03 2.401e+03 2.886e+03 7.430e+03, threshold=4.802e+03, percent-clipped=9.0
+2023-03-15 06:00:53,652 INFO [train.py:968] (0/2) Epoch 29, batch 43550, giga_loss[loss=0.283, simple_loss=0.3528, pruned_loss=0.1066, over 28823.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3624, pruned_loss=0.1164, over 5672232.48 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1127, over 5700065.15 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.363, pruned_loss=0.1169, over 5662011.30 frames. ], batch size: 284, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:01:37,772 INFO [train.py:968] (0/2) Epoch 29, batch 43600, giga_loss[loss=0.2704, simple_loss=0.3401, pruned_loss=0.1003, over 28350.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3654, pruned_loss=0.1191, over 5653141.72 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3601, pruned_loss=0.113, over 5683233.59 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3659, pruned_loss=0.1193, over 5659120.47 frames. ], batch size: 65, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 06:01:54,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.137e+03 1.773e+03 2.235e+03 3.125e+03 9.702e+03, threshold=4.471e+03, percent-clipped=6.0
+2023-03-15 06:02:07,541 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7782, 1.7140, 1.9130, 1.5334], device='cuda:0'), covar=tensor([0.1902, 0.2592, 0.1552, 0.1826], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0720, 0.0988, 0.0886], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 06:02:25,619 INFO [train.py:968] (0/2) Epoch 29, batch 43650, giga_loss[loss=0.354, simple_loss=0.4188, pruned_loss=0.1446, over 27979.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3705, pruned_loss=0.1212, over 5659156.26 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.113, over 5684649.73 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3708, pruned_loss=0.1214, over 5662264.37 frames. ], batch size: 412, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:03:16,322 INFO [train.py:968] (0/2) Epoch 29, batch 43700, libri_loss[loss=0.3239, simple_loss=0.387, pruned_loss=0.1304, over 29313.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3733, pruned_loss=0.1201, over 5656002.59 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3603, pruned_loss=0.1131, over 5685539.49 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3735, pruned_loss=0.1202, over 5657395.65 frames. ], batch size: 94, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:03:34,427 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.670e+03 2.060e+03 2.977e+03 9.240e+03, threshold=4.120e+03, percent-clipped=10.0
+2023-03-15 06:03:37,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3936, 1.4498, 1.5305, 1.1755], device='cuda:0'), covar=tensor([0.1655, 0.2808, 0.1524, 0.1948], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0720, 0.0989, 0.0888], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 06:04:06,092 INFO [train.py:968] (0/2) Epoch 29, batch 43750, giga_loss[loss=0.3692, simple_loss=0.4178, pruned_loss=0.1603, over 27568.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3753, pruned_loss=0.1216, over 5666706.09 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3601, pruned_loss=0.1131, over 5687632.49 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3759, pruned_loss=0.1218, over 5665597.49 frames. ], batch size: 472, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:04:41,458 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1319610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:04:45,926 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7205, 2.4804, 1.5712, 0.8813], device='cuda:0'), covar=tensor([0.8430, 0.3867, 0.4049, 0.7963], device='cuda:0'), in_proj_covar=tensor([0.1867, 0.1758, 0.1670, 0.1517], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 06:04:51,302 INFO [train.py:968] (0/2) Epoch 29, batch 43800, giga_loss[loss=0.3306, simple_loss=0.3963, pruned_loss=0.1324, over 28986.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3767, pruned_loss=0.1228, over 5676050.94 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3601, pruned_loss=0.1131, over 5692557.71 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3775, pruned_loss=0.1232, over 5670375.20 frames. ], batch size: 145, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:05:08,620 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.277e+03 1.980e+03 2.499e+03 3.636e+03 8.558e+03, threshold=4.999e+03, percent-clipped=15.0
+2023-03-15 06:05:21,728 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4503, 1.6693, 1.5852, 1.5433], device='cuda:0'), covar=tensor([0.1748, 0.1888, 0.2198, 0.1936], device='cuda:0'), in_proj_covar=tensor([0.0513, 0.0764, 0.0736, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 06:05:30,428 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6290, 1.9667, 1.5463, 1.6214], device='cuda:0'), covar=tensor([0.2687, 0.2730, 0.3191, 0.2510], device='cuda:0'), in_proj_covar=tensor([0.1615, 0.1163, 0.1426, 0.1015], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 06:05:37,647 INFO [train.py:968] (0/2) Epoch 29, batch 43850, giga_loss[loss=0.3441, simple_loss=0.4029, pruned_loss=0.1426, over 28669.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3769, pruned_loss=0.1238, over 5666210.84 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3603, pruned_loss=0.1132, over 5683785.91 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3778, pruned_loss=0.1243, over 5669580.95 frames. ], batch size: 284, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:05:57,501 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4745, 1.7870, 1.4700, 1.4680], device='cuda:0'), covar=tensor([0.2237, 0.2262, 0.2429, 0.2240], device='cuda:0'), in_proj_covar=tensor([0.1617, 0.1165, 0.1428, 0.1016], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 06:06:20,303 INFO [train.py:968] (0/2) Epoch 29, batch 43900, giga_loss[loss=0.4097, simple_loss=0.433, pruned_loss=0.1932, over 26591.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3761, pruned_loss=0.1241, over 5667003.51 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3606, pruned_loss=0.1134, over 5690651.55 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.377, pruned_loss=0.1246, over 5662920.39 frames. ], batch size: 555, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:06:22,149 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1319722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:06:37,711 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 1.777e+03 2.112e+03 2.665e+03 9.479e+03, threshold=4.225e+03, percent-clipped=3.0
+2023-03-15 06:07:07,763 INFO [train.py:968] (0/2) Epoch 29, batch 43950, giga_loss[loss=0.258, simple_loss=0.3279, pruned_loss=0.09405, over 28799.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3732, pruned_loss=0.1227, over 5668444.22 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3607, pruned_loss=0.1134, over 5693728.43 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.374, pruned_loss=0.1232, over 5662168.64 frames. ], batch size: 112, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:07:52,604 INFO [train.py:968] (0/2) Epoch 29, batch 44000, giga_loss[loss=0.3105, simple_loss=0.3699, pruned_loss=0.1255, over 28338.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3716, pruned_loss=0.1224, over 5669934.38 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3609, pruned_loss=0.1135, over 5696213.91 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3724, pruned_loss=0.123, over 5662118.89 frames. ], batch size: 368, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 06:08:07,656 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=2.01 vs. limit=2.0
+2023-03-15 06:08:08,493 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.994e+03 2.745e+03 3.838e+03 8.391e+03, threshold=5.491e+03, percent-clipped=20.0
+2023-03-15 06:08:20,380 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-15 06:08:41,154 INFO [train.py:968] (0/2) Epoch 29, batch 44050, giga_loss[loss=0.2654, simple_loss=0.3351, pruned_loss=0.09789, over 28394.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3721, pruned_loss=0.1236, over 5658936.94 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3611, pruned_loss=0.1138, over 5698858.50 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3727, pruned_loss=0.1239, over 5649736.10 frames. ], batch size: 65, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:08:41,620 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1319871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:09:32,905 INFO [train.py:968] (0/2) Epoch 29, batch 44100, libri_loss[loss=0.3102, simple_loss=0.3666, pruned_loss=0.1269, over 19945.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3715, pruned_loss=0.1236, over 5643197.15 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3611, pruned_loss=0.1138, over 5690309.20 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.372, pruned_loss=0.1239, over 5644312.58 frames. ], batch size: 187, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:09:46,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.481e+03 1.869e+03 2.312e+03 3.346e+03 6.638e+03, threshold=4.623e+03, percent-clipped=6.0
+2023-03-15 06:10:14,526 INFO [train.py:968] (0/2) Epoch 29, batch 44150, giga_loss[loss=0.3001, simple_loss=0.369, pruned_loss=0.1156, over 28672.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3692, pruned_loss=0.122, over 5641231.40 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5674320.68 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3697, pruned_loss=0.1225, over 5654998.89 frames. ], batch size: 307, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:10:29,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1319985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:10:40,799 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1320000.pt
+2023-03-15 06:11:01,906 INFO [train.py:968] (0/2) Epoch 29, batch 44200, giga_loss[loss=0.2778, simple_loss=0.3497, pruned_loss=0.1029, over 28936.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3676, pruned_loss=0.1209, over 5645587.77 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.361, pruned_loss=0.1136, over 5677705.63 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3684, pruned_loss=0.1216, over 5652891.68 frames. ], batch size: 112, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:11:17,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.014e+03 1.659e+03 2.129e+03 3.147e+03 7.282e+03, threshold=4.259e+03, percent-clipped=6.0
+2023-03-15 06:11:41,490 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.85 vs. limit=2.0
+2023-03-15 06:11:41,871 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:11:46,082 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 06:11:50,805 INFO [train.py:968] (0/2) Epoch 29, batch 44250, libri_loss[loss=0.2433, simple_loss=0.3222, pruned_loss=0.08223, over 29562.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3694, pruned_loss=0.1215, over 5645247.10 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3608, pruned_loss=0.1134, over 5678530.54 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3704, pruned_loss=0.1224, over 5649585.08 frames. ], batch size: 79, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:12:14,211 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:12:19,970 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:12:31,490 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5719, 2.3463, 1.7623, 0.8122], device='cuda:0'), covar=tensor([0.6404, 0.3704, 0.4508, 0.7135], device='cuda:0'), in_proj_covar=tensor([0.1868, 0.1760, 0.1671, 0.1515], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 06:12:31,838 INFO [train.py:968] (0/2) Epoch 29, batch 44300, libri_loss[loss=0.2839, simple_loss=0.3561, pruned_loss=0.1059, over 29528.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3705, pruned_loss=0.1216, over 5659036.55 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3602, pruned_loss=0.1129, over 5689362.17 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3722, pruned_loss=0.1232, over 5651172.13 frames. ], batch size: 84, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:12:32,081 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:12:35,856 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3988, 1.6892, 1.5957, 1.6641], device='cuda:0'), covar=tensor([0.0781, 0.0322, 0.0317, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 06:12:38,902 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:12:41,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:12:47,973 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.312e+03 1.908e+03 2.598e+03 3.424e+03 9.922e+03, threshold=5.196e+03, percent-clipped=13.0
+2023-03-15 06:13:10,488 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:13:19,418 INFO [train.py:968] (0/2) Epoch 29, batch 44350, libri_loss[loss=0.298, simple_loss=0.3684, pruned_loss=0.1138, over 25714.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3699, pruned_loss=0.1215, over 5653443.56 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3603, pruned_loss=0.113, over 5678423.27 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3713, pruned_loss=0.1228, over 5656227.85 frames. ], batch size: 136, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:13:41,351 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:13:41,959 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320196.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 06:14:00,699 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:14:05,764 INFO [train.py:968] (0/2) Epoch 29, batch 44400, giga_loss[loss=0.2755, simple_loss=0.353, pruned_loss=0.09905, over 28681.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3709, pruned_loss=0.1197, over 5660470.06 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3606, pruned_loss=0.1131, over 5676676.38 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3719, pruned_loss=0.1207, over 5663842.01 frames. ], batch size: 242, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 06:14:16,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1404, 2.6597, 1.9370, 1.6272], device='cuda:0'), covar=tensor([0.5134, 0.3916, 0.3600, 0.5199], device='cuda:0'), in_proj_covar=tensor([0.1868, 0.1760, 0.1670, 0.1517], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 06:14:22,248 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.698e+03 2.159e+03 2.917e+03 6.101e+03, threshold=4.319e+03, percent-clipped=2.0
+2023-03-15 06:14:22,576 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:14:24,736 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:14:28,293 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:14:48,134 INFO [train.py:968] (0/2) Epoch 29, batch 44450, giga_loss[loss=0.2978, simple_loss=0.3819, pruned_loss=0.1069, over 28941.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3721, pruned_loss=0.1179, over 5645896.03 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3609, pruned_loss=0.1136, over 5660901.44 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3727, pruned_loss=0.1183, over 5662450.69 frames. ], batch size: 145, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:14:49,882 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:14:49,942 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:15:36,917 INFO [train.py:968] (0/2) Epoch 29, batch 44500, giga_loss[loss=0.3063, simple_loss=0.3761, pruned_loss=0.1183, over 29037.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3757, pruned_loss=0.1204, over 5594157.53 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3617, pruned_loss=0.1142, over 5608768.46 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3757, pruned_loss=0.1202, over 5655202.12 frames. ], batch size: 155, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:15:53,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.778e+03 2.276e+03 3.117e+03 9.553e+03, threshold=4.552e+03, percent-clipped=10.0
+2023-03-15 06:15:58,384 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-15 06:16:24,935 INFO [train.py:968] (0/2) Epoch 29, batch 44550, giga_loss[loss=0.3779, simple_loss=0.421, pruned_loss=0.1674, over 27585.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3796, pruned_loss=0.1243, over 5594193.27 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1144, over 5592150.05 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3796, pruned_loss=0.1241, over 5655595.44 frames. ], batch size: 472, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:16:43,503 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:16:45,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:17:15,783 INFO [train.py:968] (0/2) Epoch 29, batch 44600, giga_loss[loss=0.3014, simple_loss=0.369, pruned_loss=0.1169, over 28854.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3807, pruned_loss=0.1262, over 5585665.03 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3622, pruned_loss=0.1146, over 5568150.06 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3805, pruned_loss=0.1259, over 5655551.61 frames. ], batch size: 186, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:17:15,961 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:17:31,543 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:17:31,887 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.190e+03 1.954e+03 2.462e+03 3.354e+03 7.854e+03, threshold=4.923e+03, percent-clipped=7.0
+2023-03-15 06:17:35,207 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 06:17:58,677 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-15 06:17:59,921 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-29.pt
+2023-03-15 06:18:39,467 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:18:56,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:19:16,331 INFO [train.py:968] (0/2) Epoch 30, batch 50, giga_loss[loss=0.3387, simple_loss=0.3889, pruned_loss=0.1442, over 26795.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3648, pruned_loss=0.1031, over 1264256.63 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3261, pruned_loss=0.08185, over 146539.71 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.369, pruned_loss=0.1054, over 1147001.26 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:19:35,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.318e+03 1.759e+03 2.392e+03 5.489e+03, threshold=3.517e+03, percent-clipped=4.0
+2023-03-15 06:20:00,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3431, 1.8832, 1.4933, 1.5815], device='cuda:0'), covar=tensor([0.0817, 0.0309, 0.0346, 0.0913], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 06:20:00,648 INFO [train.py:968] (0/2) Epoch 30, batch 100, giga_loss[loss=0.2468, simple_loss=0.332, pruned_loss=0.0808, over 28857.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3563, pruned_loss=0.09842, over 2247192.67 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3285, pruned_loss=0.08182, over 285957.44 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3596, pruned_loss=0.1004, over 2064280.47 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:20:03,190 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:03,811 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320571.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 06:20:12,183 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:15,154 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:19,111 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:30,027 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:39,025 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:39,074 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3039, 1.3844, 1.3432, 1.5664], device='cuda:0'), covar=tensor([0.0790, 0.0396, 0.0361, 0.0887], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 06:20:43,264 INFO [train.py:968] (0/2) Epoch 30, batch 150, giga_loss[loss=0.2708, simple_loss=0.3303, pruned_loss=0.1056, over 26635.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3421, pruned_loss=0.09181, over 3011223.34 frames. ], libri_tot_loss[loss=0.2428, simple_loss=0.3266, pruned_loss=0.07945, over 423494.25 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3444, pruned_loss=0.0935, over 2793708.84 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:20:46,539 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:48,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:58,939 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:59,397 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.704e+02 1.136e+03 1.465e+03 1.882e+03 3.365e+03, threshold=2.930e+03, percent-clipped=0.0
+2023-03-15 06:21:00,831 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:21:05,026 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:21:11,382 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:21:13,988 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2683, 1.5876, 1.5827, 1.4440], device='cuda:0'), covar=tensor([0.2461, 0.2290, 0.2618, 0.2235], device='cuda:0'), in_proj_covar=tensor([0.0516, 0.0768, 0.0738, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 06:21:21,389 INFO [train.py:968] (0/2) Epoch 30, batch 200, giga_loss[loss=0.2337, simple_loss=0.3092, pruned_loss=0.07907, over 28935.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3319, pruned_loss=0.08731, over 3618607.49 frames. ], libri_tot_loss[loss=0.2435, simple_loss=0.327, pruned_loss=0.08002, over 638796.54 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3333, pruned_loss=0.08867, over 3349715.38 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:21:23,528 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:21:56,236 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:21:57,051 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320714.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 06:21:59,809 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:22:00,521 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320717.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 06:22:00,877 INFO [train.py:968] (0/2) Epoch 30, batch 250, libri_loss[loss=0.26, simple_loss=0.333, pruned_loss=0.09347, over 29490.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.321, pruned_loss=0.08224, over 4085964.05 frames. ], libri_tot_loss[loss=0.2428, simple_loss=0.3257, pruned_loss=0.07994, over 769387.85 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3215, pruned_loss=0.08304, over 3825198.43 frames. ], batch size: 70, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:22:02,902 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-15 06:22:13,327 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-15 06:22:15,100 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:22:16,922 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:22:18,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.032e+02 1.223e+03 1.547e+03 2.183e+03 4.466e+03, threshold=3.094e+03, percent-clipped=10.0
+2023-03-15 06:22:21,342 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6225, 2.2306, 1.6433, 1.7840], device='cuda:0'), covar=tensor([0.0774, 0.0265, 0.0332, 0.0891], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 06:22:23,533 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:22:24,117 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320746.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 06:22:29,924 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3171, 1.3909, 3.6847, 3.1089], device='cuda:0'), covar=tensor([0.1617, 0.2705, 0.0507, 0.1033], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0683, 0.1027, 0.0999], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 06:22:41,216 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:22:42,050 INFO [train.py:968] (0/2) Epoch 30, batch 300, giga_loss[loss=0.2589, simple_loss=0.323, pruned_loss=0.0974, over 29140.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3119, pruned_loss=0.07829, over 4445595.52 frames. ], libri_tot_loss[loss=0.2424, simple_loss=0.3256, pruned_loss=0.07958, over 820619.49 frames. ], giga_tot_loss[loss=0.2346, simple_loss=0.3117, pruned_loss=0.07873, over 4226170.45 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:23:01,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:23:04,141 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:23:27,719 INFO [train.py:968] (0/2) Epoch 30, batch 350, giga_loss[loss=0.1982, simple_loss=0.2795, pruned_loss=0.05843, over 28894.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3061, pruned_loss=0.07629, over 4725328.63 frames. ], libri_tot_loss[loss=0.2435, simple_loss=0.3262, pruned_loss=0.08039, over 896757.83 frames. ], giga_tot_loss[loss=0.229, simple_loss=0.3053, pruned_loss=0.07636, over 4534398.92 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:23:30,699 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:23:43,979 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.175e+02 1.102e+03 1.393e+03 1.946e+03 4.492e+03, threshold=2.785e+03, percent-clipped=5.0
+2023-03-15 06:24:06,262 INFO [train.py:968] (0/2) Epoch 30, batch 400, giga_loss[loss=0.2155, simple_loss=0.2958, pruned_loss=0.06756, over 28578.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3019, pruned_loss=0.07471, over 4943384.89 frames. ], libri_tot_loss[loss=0.2452, simple_loss=0.3276, pruned_loss=0.08138, over 946731.24 frames. ], giga_tot_loss[loss=0.2248, simple_loss=0.3007, pruned_loss=0.07446, over 4783693.21 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:24:21,585 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8189, 3.6478, 3.4417, 1.8033], device='cuda:0'), covar=tensor([0.0723, 0.0891, 0.0844, 0.2056], device='cuda:0'), in_proj_covar=tensor([0.1326, 0.1227, 0.1028, 0.0759], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 06:24:47,086 INFO [train.py:968] (0/2) Epoch 30, batch 450, giga_loss[loss=0.2123, simple_loss=0.2884, pruned_loss=0.06806, over 28996.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.299, pruned_loss=0.07326, over 5117711.83 frames. ], libri_tot_loss[loss=0.2446, simple_loss=0.327, pruned_loss=0.08109, over 995980.10 frames. ], giga_tot_loss[loss=0.2219, simple_loss=0.2978, pruned_loss=0.07301, over 4984605.73 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:24:47,378 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5434, 2.2222, 1.6637, 0.8188], device='cuda:0'), covar=tensor([0.7240, 0.3641, 0.5038, 0.8318], device='cuda:0'), in_proj_covar=tensor([0.1855, 0.1747, 0.1660, 0.1506], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 06:25:02,842 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.678e+02 1.092e+03 1.328e+03 1.976e+03 6.415e+03, threshold=2.657e+03, percent-clipped=9.0
+2023-03-15 06:25:26,213 INFO [train.py:968] (0/2) Epoch 30, batch 500, libri_loss[loss=0.2742, simple_loss=0.3668, pruned_loss=0.0908, over 29522.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2978, pruned_loss=0.07238, over 5250626.63 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3326, pruned_loss=0.08323, over 1138740.44 frames. ], giga_tot_loss[loss=0.2189, simple_loss=0.2948, pruned_loss=0.07148, over 5126947.91 frames. ], batch size: 84, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:25:35,644 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:25:57,184 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-15 06:26:00,500 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 06:26:07,141 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-15 06:26:07,291 INFO [train.py:968] (0/2) Epoch 30, batch 550, giga_loss[loss=0.2258, simple_loss=0.3042, pruned_loss=0.07368, over 27869.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2971, pruned_loss=0.07227, over 5356403.30 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.332, pruned_loss=0.0832, over 1279613.17 frames. ], giga_tot_loss[loss=0.2181, simple_loss=0.2938, pruned_loss=0.07123, over 5241369.17 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:26:13,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5578, 1.8564, 1.7763, 1.5072], device='cuda:0'), covar=tensor([0.3070, 0.2437, 0.2025, 0.2639], device='cuda:0'), in_proj_covar=tensor([0.2075, 0.2040, 0.1943, 0.2094], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 06:26:26,966 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.883e+02 1.201e+03 1.513e+03 2.146e+03 5.472e+03, threshold=3.025e+03, percent-clipped=14.0
+2023-03-15 06:26:47,268 INFO [train.py:968] (0/2) Epoch 30, batch 600, libri_loss[loss=0.2702, simple_loss=0.3592, pruned_loss=0.09064, over 26171.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2973, pruned_loss=0.07229, over 5431290.98 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3338, pruned_loss=0.08396, over 1541170.74 frames. ], giga_tot_loss[loss=0.2166, simple_loss=0.292, pruned_loss=0.07057, over 5316845.77 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:27:33,586 INFO [train.py:968] (0/2) Epoch 30, batch 650, giga_loss[loss=0.2037, simple_loss=0.2789, pruned_loss=0.06429, over 28580.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2952, pruned_loss=0.0716, over 5481557.21 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3336, pruned_loss=0.08407, over 1625431.39 frames. ], giga_tot_loss[loss=0.2152, simple_loss=0.2903, pruned_loss=0.06999, over 5384858.96 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:27:36,096 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1321121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:27:39,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1321124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:27:53,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.121e+02 1.138e+03 1.454e+03 2.083e+03 1.177e+04, threshold=2.909e+03, percent-clipped=11.0
+2023-03-15 06:28:04,019 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1321153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:28:19,105 INFO [train.py:968] (0/2) Epoch 30, batch 700, giga_loss[loss=0.1972, simple_loss=0.2733, pruned_loss=0.06053, over 28926.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2924, pruned_loss=0.07011, over 5528348.50 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3335, pruned_loss=0.08383, over 1747079.61 frames. ], giga_tot_loss[loss=0.2121, simple_loss=0.2873, pruned_loss=0.06848, over 5444506.27 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:28:30,199 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 06:28:58,704 INFO [train.py:968] (0/2) Epoch 30, batch 750, giga_loss[loss=0.1827, simple_loss=0.2616, pruned_loss=0.05187, over 28767.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2903, pruned_loss=0.06892, over 5571032.45 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3329, pruned_loss=0.08358, over 1879954.81 frames. ], giga_tot_loss[loss=0.2098, simple_loss=0.2851, pruned_loss=0.06723, over 5501907.85 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:29:18,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.489e+02 1.139e+03 1.456e+03 1.910e+03 8.821e+03, threshold=2.911e+03, percent-clipped=5.0
+2023-03-15 06:29:42,779 INFO [train.py:968] (0/2) Epoch 30, batch 800, giga_loss[loss=0.2177, simple_loss=0.2893, pruned_loss=0.07301, over 28744.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2884, pruned_loss=0.06849, over 5601192.00 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3333, pruned_loss=0.08362, over 1959194.06 frames. ], giga_tot_loss[loss=0.2085, simple_loss=0.2834, pruned_loss=0.06686, over 5541129.65 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:30:30,555 INFO [train.py:968] (0/2) Epoch 30, batch 850, libri_loss[loss=0.2689, simple_loss=0.3525, pruned_loss=0.09261, over 29685.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2974, pruned_loss=0.0737, over 5607373.69 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3345, pruned_loss=0.0843, over 2084969.86 frames. ], giga_tot_loss[loss=0.2176, simple_loss=0.2917, pruned_loss=0.07181, over 5558508.82 frames. ], batch size: 91, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:30:52,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.456e+02 1.344e+03 1.604e+03 1.991e+03 4.889e+03, threshold=3.209e+03, percent-clipped=6.0
+2023-03-15 06:31:18,808 INFO [train.py:968] (0/2) Epoch 30, batch 900, giga_loss[loss=0.265, simple_loss=0.3442, pruned_loss=0.09291, over 28545.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3098, pruned_loss=0.07975, over 5621899.02 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3345, pruned_loss=0.08418, over 2119250.66 frames. ], giga_tot_loss[loss=0.2309, simple_loss=0.3051, pruned_loss=0.07828, over 5584593.54 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:31:19,238 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 06:32:00,003 INFO [train.py:968] (0/2) Epoch 30, batch 950, giga_loss[loss=0.2579, simple_loss=0.3343, pruned_loss=0.09072, over 28607.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3219, pruned_loss=0.08557, over 5644872.70 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3353, pruned_loss=0.0841, over 2245552.08 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3174, pruned_loss=0.08443, over 5610320.80 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:32:08,053 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5201, 1.3469, 4.6783, 3.4877], device='cuda:0'), covar=tensor([0.1816, 0.3057, 0.0442, 0.1044], device='cuda:0'), in_proj_covar=tensor([0.0817, 0.0681, 0.1023, 0.0996], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 06:32:17,125 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.881e+02 1.356e+03 1.777e+03 2.585e+03 5.193e+03, threshold=3.554e+03, percent-clipped=6.0
+2023-03-15 06:32:38,557 INFO [train.py:968] (0/2) Epoch 30, batch 1000, giga_loss[loss=0.2762, simple_loss=0.3574, pruned_loss=0.09754, over 28717.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3298, pruned_loss=0.08881, over 5654682.04 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3354, pruned_loss=0.08411, over 2343773.45 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3259, pruned_loss=0.08799, over 5628620.79 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:32:38,731 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1321468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:32:53,164 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9433, 1.1849, 1.0686, 0.8937], device='cuda:0'), covar=tensor([0.2559, 0.2734, 0.1800, 0.2374], device='cuda:0'), in_proj_covar=tensor([0.2072, 0.2039, 0.1941, 0.2093], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 06:33:18,527 INFO [train.py:968] (0/2) Epoch 30, batch 1050, giga_loss[loss=0.3067, simple_loss=0.3768, pruned_loss=0.1183, over 29049.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3334, pruned_loss=0.08944, over 5671313.15 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.336, pruned_loss=0.08449, over 2414587.24 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.33, pruned_loss=0.08876, over 5646070.95 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:33:42,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.102e+02 1.324e+03 1.677e+03 2.226e+03 4.246e+03, threshold=3.355e+03, percent-clipped=4.0
+2023-03-15 06:33:50,969 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1321552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:34:03,263 INFO [train.py:968] (0/2) Epoch 30, batch 1100, giga_loss[loss=0.2508, simple_loss=0.3347, pruned_loss=0.08343, over 28884.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3353, pruned_loss=0.08962, over 5665818.53 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3355, pruned_loss=0.08422, over 2463179.48 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3329, pruned_loss=0.08926, over 5645705.36 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:34:43,394 INFO [train.py:968] (0/2) Epoch 30, batch 1150, giga_loss[loss=0.2543, simple_loss=0.3379, pruned_loss=0.08529, over 28868.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3364, pruned_loss=0.08995, over 5682888.34 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3353, pruned_loss=0.08387, over 2577663.57 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3345, pruned_loss=0.08997, over 5663357.04 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:34:51,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6374, 2.0695, 1.2661, 1.6034], device='cuda:0'), covar=tensor([0.1237, 0.0637, 0.1300, 0.1188], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0453, 0.0528, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 06:35:03,295 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.875e+02 1.287e+03 1.620e+03 2.003e+03 4.077e+03, threshold=3.240e+03, percent-clipped=4.0
+2023-03-15 06:35:28,402 INFO [train.py:968] (0/2) Epoch 30, batch 1200, giga_loss[loss=0.2965, simple_loss=0.3707, pruned_loss=0.1112, over 28591.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3387, pruned_loss=0.09213, over 5672442.67 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.335, pruned_loss=0.08388, over 2611032.56 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3373, pruned_loss=0.09221, over 5654810.10 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:36:08,252 INFO [train.py:968] (0/2) Epoch 30, batch 1250, giga_loss[loss=0.2418, simple_loss=0.3273, pruned_loss=0.07813, over 28897.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3409, pruned_loss=0.09334, over 5676896.88 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3349, pruned_loss=0.08361, over 2704954.46 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.34, pruned_loss=0.09379, over 5659607.55 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:36:27,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.501e+02 1.295e+03 1.700e+03 2.265e+03 5.713e+03, threshold=3.400e+03, percent-clipped=6.0
+2023-03-15 06:36:52,787 INFO [train.py:968] (0/2) Epoch 30, batch 1300, giga_loss[loss=0.2481, simple_loss=0.3378, pruned_loss=0.07915, over 28987.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3445, pruned_loss=0.09492, over 5681627.28 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3347, pruned_loss=0.08358, over 2737037.69 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.344, pruned_loss=0.0954, over 5665801.71 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:37:29,521 INFO [train.py:968] (0/2) Epoch 30, batch 1350, giga_loss[loss=0.2709, simple_loss=0.3479, pruned_loss=0.09697, over 28971.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.346, pruned_loss=0.09465, over 5700291.50 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3349, pruned_loss=0.08358, over 2844832.07 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3459, pruned_loss=0.09541, over 5682035.14 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:37:49,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.757e+02 1.293e+03 1.685e+03 2.131e+03 4.283e+03, threshold=3.369e+03, percent-clipped=4.0
+2023-03-15 06:37:50,718 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1321843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:38:08,718 INFO [train.py:968] (0/2) Epoch 30, batch 1400, giga_loss[loss=0.2266, simple_loss=0.3215, pruned_loss=0.06585, over 28949.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3468, pruned_loss=0.09442, over 5699530.57 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3343, pruned_loss=0.08327, over 2950437.94 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3473, pruned_loss=0.09555, over 5678256.36 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:38:33,613 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 06:38:50,833 INFO [train.py:968] (0/2) Epoch 30, batch 1450, libri_loss[loss=0.2062, simple_loss=0.2914, pruned_loss=0.06048, over 29473.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3465, pruned_loss=0.09335, over 5702470.41 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3335, pruned_loss=0.0829, over 3021809.00 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3476, pruned_loss=0.09466, over 5682641.76 frames. ], batch size: 70, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:38:57,277 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1321927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:39:04,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2153, 3.6309, 1.4627, 1.4484], device='cuda:0'), covar=tensor([0.1301, 0.0378, 0.1042, 0.1639], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0573, 0.0414, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 06:39:05,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.821e+02 1.304e+03 1.527e+03 1.963e+03 3.946e+03, threshold=3.054e+03, percent-clipped=1.0
+2023-03-15 06:39:05,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1321941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:39:28,454 INFO [train.py:968] (0/2) Epoch 30, batch 1500, giga_loss[loss=0.2496, simple_loss=0.3382, pruned_loss=0.08051, over 28879.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3455, pruned_loss=0.09182, over 5707822.26 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3334, pruned_loss=0.08282, over 3050129.69 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3465, pruned_loss=0.09296, over 5690971.36 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:39:42,265 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1321986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:39:45,324 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1321989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:39:53,390 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1322000.pt
+2023-03-15 06:40:00,592 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8458, 1.1699, 1.3322, 0.9741], device='cuda:0'), covar=tensor([0.2539, 0.1741, 0.2645, 0.2144], device='cuda:0'), in_proj_covar=tensor([0.0517, 0.0765, 0.0738, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 06:40:05,803 INFO [train.py:968] (0/2) Epoch 30, batch 1550, giga_loss[loss=0.2485, simple_loss=0.3323, pruned_loss=0.08233, over 28378.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3434, pruned_loss=0.09022, over 5710657.56 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3335, pruned_loss=0.08291, over 3199328.02 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3448, pruned_loss=0.09152, over 5690379.75 frames. ], batch size: 77, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:40:06,176 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1322018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:40:25,780 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.432e+02 1.261e+03 1.519e+03 2.132e+03 5.094e+03, threshold=3.038e+03, percent-clipped=8.0
+2023-03-15 06:40:50,188 INFO [train.py:968] (0/2) Epoch 30, batch 1600, giga_loss[loss=0.2895, simple_loss=0.3569, pruned_loss=0.111, over 28938.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3457, pruned_loss=0.09298, over 5717218.28 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3337, pruned_loss=0.083, over 3238905.07 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3468, pruned_loss=0.09408, over 5699417.08 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:40:51,562 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1322070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:40:53,628 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1322073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:41:19,525 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1322102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:41:19,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3379, 1.2123, 1.1708, 1.5865], device='cuda:0'), covar=tensor([0.0800, 0.0382, 0.0356, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 06:41:31,121 INFO [train.py:968] (0/2) Epoch 30, batch 1650, giga_loss[loss=0.2762, simple_loss=0.3515, pruned_loss=0.1005, over 29053.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3488, pruned_loss=0.09774, over 5719732.44 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3335, pruned_loss=0.08284, over 3265549.15 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3499, pruned_loss=0.09878, over 5704140.35 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:41:53,416 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.417e+02 1.444e+03 1.828e+03 2.297e+03 5.874e+03, threshold=3.656e+03, percent-clipped=7.0
+2023-03-15 06:41:53,675 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6267, 3.5228, 1.7819, 1.8131], device='cuda:0'), covar=tensor([0.0971, 0.0308, 0.0822, 0.1272], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0572, 0.0414, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 06:42:12,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8645, 4.6605, 4.4250, 2.2111], device='cuda:0'), covar=tensor([0.0536, 0.0691, 0.0651, 0.2051], device='cuda:0'), in_proj_covar=tensor([0.1298, 0.1203, 0.1008, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 06:42:16,708 INFO [train.py:968] (0/2) Epoch 30, batch 1700, giga_loss[loss=0.2702, simple_loss=0.3496, pruned_loss=0.0954, over 28870.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3498, pruned_loss=0.1005, over 5701877.96 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3334, pruned_loss=0.08282, over 3291443.55 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3508, pruned_loss=0.1015, over 5688295.24 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:42:24,060 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4352, 4.0264, 1.6440, 1.6419], device='cuda:0'), covar=tensor([0.1059, 0.0290, 0.0888, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0572, 0.0414, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 06:42:52,176 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2738, 1.4096, 1.3455, 1.1568], device='cuda:0'), covar=tensor([0.2780, 0.2990, 0.2123, 0.2678], device='cuda:0'), in_proj_covar=tensor([0.2068, 0.2039, 0.1938, 0.2088], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 06:42:58,027 INFO [train.py:968] (0/2) Epoch 30, batch 1750, giga_loss[loss=0.252, simple_loss=0.3254, pruned_loss=0.08931, over 28652.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3476, pruned_loss=0.09935, over 5707578.47 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3338, pruned_loss=0.08307, over 3415591.21 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3489, pruned_loss=0.1008, over 5690825.86 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:43:12,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4466, 3.0446, 1.5244, 1.5860], device='cuda:0'), covar=tensor([0.1005, 0.0323, 0.0883, 0.1356], device='cuda:0'), in_proj_covar=tensor([0.0429, 0.0574, 0.0415, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 06:43:16,153 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.006e+02 1.324e+03 1.678e+03 2.334e+03 5.701e+03, threshold=3.357e+03, percent-clipped=7.0
+2023-03-15 06:43:31,693 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 06:43:40,043 INFO [train.py:968] (0/2) Epoch 30, batch 1800, giga_loss[loss=0.3009, simple_loss=0.3583, pruned_loss=0.1218, over 23778.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3456, pruned_loss=0.09835, over 5712902.30 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3336, pruned_loss=0.08284, over 3452807.80 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.347, pruned_loss=0.09986, over 5697014.03 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:44:18,813 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1322316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:44:19,873 INFO [train.py:968] (0/2) Epoch 30, batch 1850, giga_loss[loss=0.2527, simple_loss=0.3335, pruned_loss=0.0859, over 28986.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3455, pruned_loss=0.09751, over 5718462.56 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3339, pruned_loss=0.08279, over 3489791.79 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3465, pruned_loss=0.09893, over 5703092.40 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:44:29,433 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 06:44:38,758 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.291e+02 1.316e+03 1.573e+03 2.137e+03 6.195e+03, threshold=3.145e+03, percent-clipped=6.0
+2023-03-15 06:45:00,827 INFO [train.py:968] (0/2) Epoch 30, batch 1900, giga_loss[loss=0.2251, simple_loss=0.3082, pruned_loss=0.07101, over 28886.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3427, pruned_loss=0.09489, over 5715578.68 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3343, pruned_loss=0.08298, over 3561538.27 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3436, pruned_loss=0.0963, over 5697829.57 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:45:06,971 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5910, 1.8180, 1.9114, 1.5637], device='cuda:0'), covar=tensor([0.3485, 0.3069, 0.3043, 0.3283], device='cuda:0'), in_proj_covar=tensor([0.2076, 0.2043, 0.1941, 0.2095], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 06:45:44,507 INFO [train.py:968] (0/2) Epoch 30, batch 1950, giga_loss[loss=0.2373, simple_loss=0.3181, pruned_loss=0.07827, over 29002.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3405, pruned_loss=0.09371, over 5697369.19 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3347, pruned_loss=0.08333, over 3594696.51 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3411, pruned_loss=0.09482, over 5691992.74 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:45:49,553 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-15 06:46:03,590 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.677e+02 1.229e+03 1.563e+03 2.144e+03 5.809e+03, threshold=3.125e+03, percent-clipped=7.0
+2023-03-15 06:46:05,978 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-15 06:46:20,804 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1322459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:46:23,013 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1322462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:46:28,604 INFO [train.py:968] (0/2) Epoch 30, batch 2000, giga_loss[loss=0.2381, simple_loss=0.3006, pruned_loss=0.08777, over 23387.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3346, pruned_loss=0.09076, over 5681598.62 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3343, pruned_loss=0.08303, over 3637695.70 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3354, pruned_loss=0.09196, over 5676320.43 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:46:46,658 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1322491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:47:08,605 INFO [train.py:968] (0/2) Epoch 30, batch 2050, giga_loss[loss=0.221, simple_loss=0.297, pruned_loss=0.07246, over 28987.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3292, pruned_loss=0.0881, over 5680801.32 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3342, pruned_loss=0.08283, over 3700903.05 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3298, pruned_loss=0.08936, over 5673978.21 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:47:32,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.507e+02 1.014e+03 1.289e+03 1.665e+03 3.233e+03, threshold=2.578e+03, percent-clipped=1.0
+2023-03-15 06:47:55,255 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1322567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:47:55,613 INFO [train.py:968] (0/2) Epoch 30, batch 2100, giga_loss[loss=0.2371, simple_loss=0.3191, pruned_loss=0.07751, over 28775.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3279, pruned_loss=0.087, over 5685023.11 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3345, pruned_loss=0.08288, over 3742538.22 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.328, pruned_loss=0.08803, over 5677435.65 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:47:55,910 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6652, 1.7206, 1.4145, 1.2708], device='cuda:0'), covar=tensor([0.1127, 0.0684, 0.1098, 0.1261], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0452, 0.0528, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 06:48:16,696 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1322596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:48:35,201 INFO [train.py:968] (0/2) Epoch 30, batch 2150, giga_loss[loss=0.2314, simple_loss=0.3077, pruned_loss=0.07751, over 28769.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3299, pruned_loss=0.08781, over 5691898.64 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3346, pruned_loss=0.08283, over 3781787.33 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3298, pruned_loss=0.08875, over 5685028.24 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:48:51,694 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.226e+02 1.184e+03 1.627e+03 2.451e+03 6.024e+03, threshold=3.254e+03, percent-clipped=22.0
+2023-03-15 06:49:13,392 INFO [train.py:968] (0/2) Epoch 30, batch 2200, giga_loss[loss=0.237, simple_loss=0.3053, pruned_loss=0.08433, over 23845.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3287, pruned_loss=0.08718, over 5694237.20 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.334, pruned_loss=0.08247, over 3822578.56 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.329, pruned_loss=0.08824, over 5686114.04 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:49:52,779 INFO [train.py:968] (0/2) Epoch 30, batch 2250, giga_loss[loss=0.2131, simple_loss=0.2914, pruned_loss=0.06736, over 28667.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3268, pruned_loss=0.08614, over 5698057.50 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3343, pruned_loss=0.08248, over 3884291.09 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3266, pruned_loss=0.08713, over 5694496.08 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:49:53,875 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-15 06:50:11,946 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.459e+02 1.286e+03 1.587e+03 2.150e+03 4.654e+03, threshold=3.175e+03, percent-clipped=4.0
+2023-03-15 06:50:34,551 INFO [train.py:968] (0/2) Epoch 30, batch 2300, giga_loss[loss=0.228, simple_loss=0.3046, pruned_loss=0.07566, over 28879.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.325, pruned_loss=0.08565, over 5700003.00 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3347, pruned_loss=0.08263, over 3913515.47 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3245, pruned_loss=0.0864, over 5695352.49 frames. ], batch size: 66, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:50:36,683 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1322771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:51:11,207 INFO [train.py:968] (0/2) Epoch 30, batch 2350, giga_loss[loss=0.2426, simple_loss=0.319, pruned_loss=0.08306, over 29066.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3233, pruned_loss=0.085, over 5704872.48 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3348, pruned_loss=0.0826, over 3942675.86 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3227, pruned_loss=0.08569, over 5706326.33 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:51:17,381 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-15 06:51:31,888 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.662e+02 1.179e+03 1.411e+03 1.957e+03 5.151e+03, threshold=2.823e+03, percent-clipped=6.0
+2023-03-15 06:51:52,106 INFO [train.py:968] (0/2) Epoch 30, batch 2400, giga_loss[loss=0.2351, simple_loss=0.3078, pruned_loss=0.08121, over 28843.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3205, pruned_loss=0.08384, over 5713457.27 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3352, pruned_loss=0.08275, over 3971345.41 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3197, pruned_loss=0.08432, over 5712611.07 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:51:56,215 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4658, 1.6429, 1.2220, 1.1976], device='cuda:0'), covar=tensor([0.1200, 0.0660, 0.1159, 0.1304], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0455, 0.0530, 0.0469], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 06:52:29,136 INFO [train.py:968] (0/2) Epoch 30, batch 2450, giga_loss[loss=0.2588, simple_loss=0.3236, pruned_loss=0.09699, over 28882.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3206, pruned_loss=0.08386, over 5715685.81 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3365, pruned_loss=0.08326, over 4019716.05 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3188, pruned_loss=0.08395, over 5718881.74 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:52:46,947 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1322942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:52:47,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.135e+02 1.112e+03 1.357e+03 1.833e+03 6.113e+03, threshold=2.715e+03, percent-clipped=10.0
+2023-03-15 06:52:48,929 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-15 06:53:05,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4047, 4.2422, 4.0381, 1.7897], device='cuda:0'), covar=tensor([0.0558, 0.0705, 0.0619, 0.2337], device='cuda:0'), in_proj_covar=tensor([0.1304, 0.1208, 0.1013, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 06:53:06,782 INFO [train.py:968] (0/2) Epoch 30, batch 2500, giga_loss[loss=0.2076, simple_loss=0.284, pruned_loss=0.06556, over 28420.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3193, pruned_loss=0.08333, over 5720319.66 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3373, pruned_loss=0.08356, over 4065245.03 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.317, pruned_loss=0.08322, over 5719586.62 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:53:09,460 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1322971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:53:46,484 INFO [train.py:968] (0/2) Epoch 30, batch 2550, giga_loss[loss=0.3, simple_loss=0.3608, pruned_loss=0.1196, over 27677.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.318, pruned_loss=0.08273, over 5717063.89 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3376, pruned_loss=0.08356, over 4100522.88 frames. ], giga_tot_loss[loss=0.2404, simple_loss=0.3156, pruned_loss=0.08263, over 5714387.07 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:53:56,555 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5065, 1.5440, 1.7175, 1.3006], device='cuda:0'), covar=tensor([0.2009, 0.2686, 0.1651, 0.1870], device='cuda:0'), in_proj_covar=tensor([0.0947, 0.0724, 0.0998, 0.0895], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 06:54:05,251 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.216e+02 1.188e+03 1.465e+03 1.923e+03 4.351e+03, threshold=2.929e+03, percent-clipped=7.0
+2023-03-15 06:54:12,364 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-15 06:54:25,144 INFO [train.py:968] (0/2) Epoch 30, batch 2600, giga_loss[loss=0.2386, simple_loss=0.3137, pruned_loss=0.08173, over 29060.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3161, pruned_loss=0.08176, over 5720411.63 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3378, pruned_loss=0.08364, over 4109537.27 frames. ], giga_tot_loss[loss=0.2386, simple_loss=0.3139, pruned_loss=0.08163, over 5717519.26 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:54:25,286 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1323068.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 06:54:36,753 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:54:38,623 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1323088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:54:58,878 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:55:00,880 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1323117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:55:00,907 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1323117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:55:01,221 INFO [train.py:968] (0/2) Epoch 30, batch 2650, giga_loss[loss=0.2534, simple_loss=0.3302, pruned_loss=0.08828, over 28773.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.316, pruned_loss=0.0819, over 5713375.87 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3387, pruned_loss=0.08423, over 4152699.14 frames. ], giga_tot_loss[loss=0.2379, simple_loss=0.3131, pruned_loss=0.08135, over 5715508.96 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:55:22,866 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.905e+02 1.196e+03 1.376e+03 1.786e+03 3.554e+03, threshold=2.752e+03, percent-clipped=4.0
+2023-03-15 06:55:26,387 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1323146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:55:26,403 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1323146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:55:44,599 INFO [train.py:968] (0/2) Epoch 30, batch 2700, giga_loss[loss=0.227, simple_loss=0.3055, pruned_loss=0.07426, over 28861.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3185, pruned_loss=0.08362, over 5708396.19 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3391, pruned_loss=0.08427, over 4175015.42 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3157, pruned_loss=0.08314, over 5711028.92 frames. ], batch size: 66, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:55:59,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4743, 1.9009, 1.6819, 1.5978], device='cuda:0'), covar=tensor([0.0774, 0.0302, 0.0310, 0.0859], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 06:56:26,073 INFO [train.py:968] (0/2) Epoch 30, batch 2750, giga_loss[loss=0.2755, simple_loss=0.3566, pruned_loss=0.09721, over 28660.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3249, pruned_loss=0.08723, over 5699463.36 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3402, pruned_loss=0.08478, over 4215527.31 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3214, pruned_loss=0.08653, over 5706573.16 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:56:48,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.730e+02 1.381e+03 1.659e+03 2.287e+03 6.985e+03, threshold=3.318e+03, percent-clipped=16.0
+2023-03-15 06:57:06,714 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6148, 4.7395, 1.7707, 1.9130], device='cuda:0'), covar=tensor([0.1016, 0.0236, 0.0902, 0.1330], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0573, 0.0416, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 06:57:09,009 INFO [train.py:968] (0/2) Epoch 30, batch 2800, giga_loss[loss=0.2855, simple_loss=0.361, pruned_loss=0.105, over 28684.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3327, pruned_loss=0.09247, over 5692445.44 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3398, pruned_loss=0.08448, over 4239515.26 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3301, pruned_loss=0.09217, over 5696287.46 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:57:29,168 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:57:30,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1323292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:57:36,437 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1323300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:57:40,977 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-15 06:57:51,154 INFO [train.py:968] (0/2) Epoch 30, batch 2850, giga_loss[loss=0.2703, simple_loss=0.3417, pruned_loss=0.09947, over 28618.00 frames. ], tot_loss[loss=0.264, simple_loss=0.338, pruned_loss=0.09503, over 5677677.34 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3401, pruned_loss=0.08463, over 4277805.28 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3355, pruned_loss=0.09493, over 5685137.57 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:57:54,930 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1323321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:58:04,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.7470, 4.6041, 4.4024, 1.9837], device='cuda:0'), covar=tensor([0.0605, 0.0754, 0.0840, 0.2072], device='cuda:0'), in_proj_covar=tensor([0.1305, 0.1208, 0.1012, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 06:58:13,799 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.390e+02 1.404e+03 1.789e+03 2.573e+03 5.548e+03, threshold=3.577e+03, percent-clipped=11.0
+2023-03-15 06:58:34,817 INFO [train.py:968] (0/2) Epoch 30, batch 2900, giga_loss[loss=0.2853, simple_loss=0.3627, pruned_loss=0.1039, over 28456.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3432, pruned_loss=0.09751, over 5657870.53 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3403, pruned_loss=0.08475, over 4352990.36 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3411, pruned_loss=0.09792, over 5663813.04 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:59:14,109 INFO [train.py:968] (0/2) Epoch 30, batch 2950, giga_loss[loss=0.2709, simple_loss=0.348, pruned_loss=0.0969, over 28730.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3462, pruned_loss=0.09781, over 5675615.50 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3394, pruned_loss=0.08426, over 4405283.16 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3453, pruned_loss=0.09893, over 5673726.24 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:59:40,847 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1323443.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 06:59:41,897 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.002e+03 1.304e+03 1.522e+03 2.173e+03 5.116e+03, threshold=3.044e+03, percent-clipped=3.0
+2023-03-15 06:59:56,206 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2222, 5.0601, 4.7726, 2.3367], device='cuda:0'), covar=tensor([0.0423, 0.0532, 0.0585, 0.1885], device='cuda:0'), in_proj_covar=tensor([0.1307, 0.1209, 0.1013, 0.0756], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 07:00:01,075 INFO [train.py:968] (0/2) Epoch 30, batch 3000, giga_loss[loss=0.3352, simple_loss=0.3805, pruned_loss=0.145, over 23444.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3513, pruned_loss=0.101, over 5667043.91 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3393, pruned_loss=0.08424, over 4425541.07 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3509, pruned_loss=0.1022, over 5670536.00 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:00:01,081 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 07:00:08,172 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2781, 1.4501, 1.3847, 1.2710], device='cuda:0'), covar=tensor([0.2505, 0.2662, 0.1971, 0.2250], device='cuda:0'), in_proj_covar=tensor([0.2077, 0.2037, 0.1948, 0.2098], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 07:00:09,997 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.2082, simple_loss=0.3146, pruned_loss=0.05088, over 944034.00 frames. 
+2023-03-15 07:00:09,998 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 07:00:50,257 INFO [train.py:968] (0/2) Epoch 30, batch 3050, giga_loss[loss=0.2849, simple_loss=0.3544, pruned_loss=0.1077, over 28861.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3491, pruned_loss=0.09921, over 5658219.79 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3391, pruned_loss=0.08439, over 4454828.51 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3492, pruned_loss=0.1006, over 5673164.89 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:00:57,395 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.5223, 2.8491, 2.6738, 2.6325], device='cuda:0'), covar=tensor([0.2303, 0.2192, 0.2118, 0.1921], device='cuda:0'), in_proj_covar=tensor([0.0519, 0.0767, 0.0739, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 07:01:09,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.021e+03 1.382e+03 1.658e+03 2.224e+03 6.091e+03, threshold=3.316e+03, percent-clipped=8.0
+2023-03-15 07:01:29,736 INFO [train.py:968] (0/2) Epoch 30, batch 3100, giga_loss[loss=0.2411, simple_loss=0.3309, pruned_loss=0.07568, over 29067.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3456, pruned_loss=0.0962, over 5667052.39 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3395, pruned_loss=0.08462, over 4479863.45 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3456, pruned_loss=0.09734, over 5677870.46 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:01:45,802 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323586.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:01:47,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1323589.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:02:02,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4400, 1.6712, 1.4039, 1.0449], device='cuda:0'), covar=tensor([0.2689, 0.2856, 0.3180, 0.2513], device='cuda:0'), in_proj_covar=tensor([0.1620, 0.1168, 0.1434, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 07:02:11,290 INFO [train.py:968] (0/2) Epoch 30, batch 3150, libri_loss[loss=0.2399, simple_loss=0.3305, pruned_loss=0.07468, over 25956.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3433, pruned_loss=0.09426, over 5666486.98 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3384, pruned_loss=0.0843, over 4545571.31 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3443, pruned_loss=0.09595, over 5668540.42 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:02:11,503 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1323618.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:02:32,926 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.000e+03 1.291e+03 1.596e+03 1.946e+03 8.866e+03, threshold=3.191e+03, percent-clipped=5.0
+2023-03-15 07:02:54,262 INFO [train.py:968] (0/2) Epoch 30, batch 3200, giga_loss[loss=0.2672, simple_loss=0.3467, pruned_loss=0.09391, over 28902.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3437, pruned_loss=0.09442, over 5668306.71 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3383, pruned_loss=0.08429, over 4559191.46 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3447, pruned_loss=0.09584, over 5667922.21 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:03:00,082 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1323675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:03:10,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8796, 2.0549, 2.0706, 1.6939], device='cuda:0'), covar=tensor([0.2631, 0.2395, 0.2303, 0.2625], device='cuda:0'), in_proj_covar=tensor([0.2075, 0.2039, 0.1949, 0.2098], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 07:03:34,347 INFO [train.py:968] (0/2) Epoch 30, batch 3250, giga_loss[loss=0.2836, simple_loss=0.3605, pruned_loss=0.1033, over 28882.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3455, pruned_loss=0.0955, over 5658561.50 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3379, pruned_loss=0.08423, over 4594487.92 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3467, pruned_loss=0.09708, over 5670564.01 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:03:45,410 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 07:03:56,905 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.509e+02 1.415e+03 1.731e+03 2.291e+03 4.918e+03, threshold=3.461e+03, percent-clipped=9.0
+2023-03-15 07:04:16,334 INFO [train.py:968] (0/2) Epoch 30, batch 3300, giga_loss[loss=0.2904, simple_loss=0.3619, pruned_loss=0.1095, over 28913.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3474, pruned_loss=0.09685, over 5676190.29 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3379, pruned_loss=0.08422, over 4613814.25 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3485, pruned_loss=0.09827, over 5682715.83 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:04:57,901 INFO [train.py:968] (0/2) Epoch 30, batch 3350, giga_loss[loss=0.2356, simple_loss=0.3173, pruned_loss=0.077, over 29062.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3488, pruned_loss=0.09839, over 5670326.12 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3378, pruned_loss=0.08417, over 4633247.11 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3499, pruned_loss=0.09985, over 5681223.07 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:04:58,136 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:05:00,213 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1323821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:05:06,748 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-15 07:05:18,935 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.706e+02 1.347e+03 1.670e+03 2.265e+03 4.366e+03, threshold=3.341e+03, percent-clipped=5.0
+2023-03-15 07:05:24,762 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1323850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:05:39,262 INFO [train.py:968] (0/2) Epoch 30, batch 3400, giga_loss[loss=0.3763, simple_loss=0.4147, pruned_loss=0.1689, over 26722.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3493, pruned_loss=0.09943, over 5669890.34 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3377, pruned_loss=0.084, over 4661512.10 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3506, pruned_loss=0.1011, over 5676007.64 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:05:43,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9337, 2.1211, 1.9098, 1.8837], device='cuda:0'), covar=tensor([0.2495, 0.2813, 0.2577, 0.2447], device='cuda:0'), in_proj_covar=tensor([0.0520, 0.0768, 0.0739, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 07:06:20,052 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 07:06:20,230 INFO [train.py:968] (0/2) Epoch 30, batch 3450, giga_loss[loss=0.2582, simple_loss=0.34, pruned_loss=0.08819, over 29040.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3496, pruned_loss=0.1001, over 5670899.76 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3375, pruned_loss=0.08395, over 4679595.18 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.351, pruned_loss=0.1017, over 5672964.41 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:06:43,940 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.295e+03 1.640e+03 2.053e+03 7.274e+03, threshold=3.280e+03, percent-clipped=10.0
+2023-03-15 07:06:51,783 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5765, 1.7704, 1.4364, 1.6149], device='cuda:0'), covar=tensor([0.2824, 0.2936, 0.3336, 0.2448], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1167, 0.1431, 0.1015], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 07:07:02,234 INFO [train.py:968] (0/2) Epoch 30, batch 3500, giga_loss[loss=0.2682, simple_loss=0.3434, pruned_loss=0.09646, over 28714.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.0997, over 5674669.98 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3377, pruned_loss=0.08415, over 4700112.36 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3512, pruned_loss=0.1011, over 5679177.94 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:07:25,997 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1324000.pt
+2023-03-15 07:07:39,491 INFO [train.py:968] (0/2) Epoch 30, batch 3550, giga_loss[loss=0.2739, simple_loss=0.3314, pruned_loss=0.1082, over 23350.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3496, pruned_loss=0.0983, over 5682599.72 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3382, pruned_loss=0.08454, over 4743548.35 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3506, pruned_loss=0.0997, over 5681623.15 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:07:52,103 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9945, 3.8507, 3.6320, 1.8380], device='cuda:0'), covar=tensor([0.0644, 0.0802, 0.0751, 0.2249], device='cuda:0'), in_proj_covar=tensor([0.1306, 0.1212, 0.1014, 0.0757], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 07:08:02,094 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.546e+02 1.201e+03 1.470e+03 1.986e+03 5.634e+03, threshold=2.939e+03, percent-clipped=3.0
+2023-03-15 07:08:22,494 INFO [train.py:968] (0/2) Epoch 30, batch 3600, giga_loss[loss=0.2807, simple_loss=0.3579, pruned_loss=0.1017, over 28513.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3501, pruned_loss=0.09806, over 5688689.92 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3384, pruned_loss=0.08466, over 4760672.01 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3509, pruned_loss=0.09927, over 5685345.67 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:09:00,747 INFO [train.py:968] (0/2) Epoch 30, batch 3650, libri_loss[loss=0.2584, simple_loss=0.3399, pruned_loss=0.08849, over 29551.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3486, pruned_loss=0.09747, over 5686577.23 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3384, pruned_loss=0.0847, over 4768494.83 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3494, pruned_loss=0.09855, over 5690051.35 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:09:22,638 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.988e+02 1.200e+03 1.457e+03 1.953e+03 4.747e+03, threshold=2.914e+03, percent-clipped=4.0
+2023-03-15 07:09:38,567 INFO [train.py:968] (0/2) Epoch 30, batch 3700, libri_loss[loss=0.2507, simple_loss=0.3313, pruned_loss=0.08507, over 29532.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.346, pruned_loss=0.09636, over 5692476.83 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3382, pruned_loss=0.08472, over 4806502.50 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.347, pruned_loss=0.0976, over 5689392.07 frames. ], batch size: 81, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:10:03,291 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0251, 1.1684, 1.0906, 0.9627], device='cuda:0'), covar=tensor([0.2348, 0.2897, 0.1846, 0.2365], device='cuda:0'), in_proj_covar=tensor([0.2066, 0.2032, 0.1942, 0.2093], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 07:10:14,634 INFO [train.py:968] (0/2) Epoch 30, batch 3750, giga_loss[loss=0.2805, simple_loss=0.361, pruned_loss=0.09999, over 28600.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3435, pruned_loss=0.09467, over 5703518.86 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3381, pruned_loss=0.08463, over 4856787.71 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3447, pruned_loss=0.09623, over 5694802.16 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:10:27,095 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3846, 3.2412, 3.1504, 1.3431], device='cuda:0'), covar=tensor([0.1008, 0.1126, 0.1064, 0.2283], device='cuda:0'), in_proj_covar=tensor([0.1304, 0.1207, 0.1010, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 07:10:36,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.581e+02 1.180e+03 1.450e+03 1.792e+03 5.453e+03, threshold=2.901e+03, percent-clipped=4.0
+2023-03-15 07:10:56,636 INFO [train.py:968] (0/2) Epoch 30, batch 3800, giga_loss[loss=0.2842, simple_loss=0.3597, pruned_loss=0.1044, over 28965.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3447, pruned_loss=0.09592, over 5697947.70 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3382, pruned_loss=0.08468, over 4873770.29 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3457, pruned_loss=0.09732, over 5690911.05 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:10:57,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4667, 3.3461, 1.5224, 1.6457], device='cuda:0'), covar=tensor([0.1040, 0.0313, 0.0889, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0569, 0.0412, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 07:11:36,740 INFO [train.py:968] (0/2) Epoch 30, batch 3850, giga_loss[loss=0.2354, simple_loss=0.3286, pruned_loss=0.0711, over 28880.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3452, pruned_loss=0.09595, over 5700520.91 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3383, pruned_loss=0.08468, over 4889053.11 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.346, pruned_loss=0.09726, over 5695839.17 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:11:57,730 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.125e+02 1.135e+03 1.412e+03 1.938e+03 3.981e+03, threshold=2.825e+03, percent-clipped=5.0
+2023-03-15 07:12:12,737 INFO [train.py:968] (0/2) Epoch 30, batch 3900, giga_loss[loss=0.2543, simple_loss=0.3402, pruned_loss=0.0842, over 28597.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3452, pruned_loss=0.09515, over 5708175.29 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3386, pruned_loss=0.08475, over 4914375.98 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3458, pruned_loss=0.09639, over 5699763.64 frames. ], batch size: 65, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:12:29,105 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7432, 1.8640, 1.9402, 1.5279], device='cuda:0'), covar=tensor([0.1882, 0.2629, 0.1519, 0.1819], device='cuda:0'), in_proj_covar=tensor([0.0945, 0.0724, 0.0997, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 07:12:55,352 INFO [train.py:968] (0/2) Epoch 30, batch 3950, giga_loss[loss=0.2652, simple_loss=0.3506, pruned_loss=0.08988, over 28649.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3447, pruned_loss=0.09435, over 5713759.78 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3388, pruned_loss=0.08482, over 4932383.27 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3451, pruned_loss=0.09546, over 5705234.16 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:13:17,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.517e+02 1.224e+03 1.527e+03 1.876e+03 6.408e+03, threshold=3.054e+03, percent-clipped=10.0
+2023-03-15 07:13:35,174 INFO [train.py:968] (0/2) Epoch 30, batch 4000, giga_loss[loss=0.2694, simple_loss=0.3454, pruned_loss=0.09671, over 28243.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3451, pruned_loss=0.09532, over 5707320.83 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3387, pruned_loss=0.0848, over 4951662.26 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3456, pruned_loss=0.09638, over 5697139.17 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:13:41,958 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2638, 1.6654, 1.1257, 1.2159], device='cuda:0'), covar=tensor([0.1381, 0.0712, 0.1522, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0452, 0.0528, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 07:14:13,749 INFO [train.py:968] (0/2) Epoch 30, batch 4050, giga_loss[loss=0.2542, simple_loss=0.3336, pruned_loss=0.08743, over 28583.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3436, pruned_loss=0.09493, over 5714717.77 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3387, pruned_loss=0.08481, over 4960859.00 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3441, pruned_loss=0.09583, over 5705275.87 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:14:14,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1196, 3.1242, 2.0478, 1.3661], device='cuda:0'), covar=tensor([0.8719, 0.3042, 0.4627, 0.7298], device='cuda:0'), in_proj_covar=tensor([0.1838, 0.1718, 0.1644, 0.1496], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 07:14:35,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.268e+02 1.214e+03 1.477e+03 1.866e+03 6.613e+03, threshold=2.955e+03, percent-clipped=5.0
+2023-03-15 07:14:44,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8383, 3.6713, 3.5009, 1.7485], device='cuda:0'), covar=tensor([0.0743, 0.0872, 0.0831, 0.2211], device='cuda:0'), in_proj_covar=tensor([0.1304, 0.1208, 0.1011, 0.0755], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 07:14:50,635 INFO [train.py:968] (0/2) Epoch 30, batch 4100, libri_loss[loss=0.2789, simple_loss=0.3652, pruned_loss=0.09631, over 29556.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3412, pruned_loss=0.0938, over 5717652.19 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3391, pruned_loss=0.08502, over 4987926.13 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3413, pruned_loss=0.09461, over 5705744.92 frames. ], batch size: 89, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:14:55,084 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-15 07:15:26,749 INFO [train.py:968] (0/2) Epoch 30, batch 4150, giga_loss[loss=0.2397, simple_loss=0.3177, pruned_loss=0.08086, over 29076.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3384, pruned_loss=0.09229, over 5712084.01 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3391, pruned_loss=0.08512, over 5000214.54 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3385, pruned_loss=0.09303, over 5708004.86 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:15:48,294 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.120e+02 1.291e+03 1.656e+03 2.450e+03 8.212e+03, threshold=3.312e+03, percent-clipped=16.0
+2023-03-15 07:16:03,274 INFO [train.py:968] (0/2) Epoch 30, batch 4200, libri_loss[loss=0.2501, simple_loss=0.3332, pruned_loss=0.08353, over 29538.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3385, pruned_loss=0.09251, over 5715254.03 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3394, pruned_loss=0.0852, over 5026335.82 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3383, pruned_loss=0.09324, over 5707307.81 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:16:41,343 INFO [train.py:968] (0/2) Epoch 30, batch 4250, giga_loss[loss=0.2923, simple_loss=0.3603, pruned_loss=0.1121, over 28232.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3375, pruned_loss=0.09256, over 5718027.09 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3394, pruned_loss=0.08521, over 5054977.94 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3373, pruned_loss=0.09337, over 5707308.77 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:16:53,657 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1324730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:17:05,714 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.950e+02 1.324e+03 1.558e+03 2.123e+03 6.488e+03, threshold=3.116e+03, percent-clipped=6.0
+2023-03-15 07:17:15,570 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3764, 1.9636, 1.3726, 0.7477], device='cuda:0'), covar=tensor([0.8096, 0.3817, 0.4624, 0.7680], device='cuda:0'), in_proj_covar=tensor([0.1844, 0.1727, 0.1652, 0.1502], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 07:17:21,155 INFO [train.py:968] (0/2) Epoch 30, batch 4300, giga_loss[loss=0.227, simple_loss=0.3008, pruned_loss=0.07661, over 28733.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3367, pruned_loss=0.09266, over 5716697.62 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3392, pruned_loss=0.08524, over 5080268.57 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3366, pruned_loss=0.09349, over 5702899.73 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:17:44,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2059, 1.2609, 3.7170, 3.1634], device='cuda:0'), covar=tensor([0.1704, 0.2768, 0.0458, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0673, 0.1007, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 07:17:49,620 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4568, 2.2569, 1.7686, 0.6707], device='cuda:0'), covar=tensor([0.8250, 0.3534, 0.4777, 0.8706], device='cuda:0'), in_proj_covar=tensor([0.1845, 0.1726, 0.1653, 0.1503], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 07:17:58,462 INFO [train.py:968] (0/2) Epoch 30, batch 4350, giga_loss[loss=0.2464, simple_loss=0.3183, pruned_loss=0.08728, over 28577.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3326, pruned_loss=0.09041, over 5716326.72 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3391, pruned_loss=0.08516, over 5089108.22 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3326, pruned_loss=0.09121, over 5706181.30 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:18:20,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.489e+02 1.184e+03 1.393e+03 1.844e+03 3.508e+03, threshold=2.787e+03, percent-clipped=3.0
+2023-03-15 07:18:36,751 INFO [train.py:968] (0/2) Epoch 30, batch 4400, giga_loss[loss=0.2589, simple_loss=0.3318, pruned_loss=0.09299, over 28859.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3309, pruned_loss=0.0897, over 5711870.29 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3396, pruned_loss=0.08539, over 5104942.91 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3303, pruned_loss=0.09023, over 5700553.26 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:19:13,864 INFO [train.py:968] (0/2) Epoch 30, batch 4450, giga_loss[loss=0.3, simple_loss=0.358, pruned_loss=0.121, over 23945.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3317, pruned_loss=0.08999, over 5712696.28 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3391, pruned_loss=0.08526, over 5122668.68 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3314, pruned_loss=0.0906, over 5701514.00 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:19:37,735 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.595e+02 1.202e+03 1.519e+03 2.097e+03 1.769e+04, threshold=3.038e+03, percent-clipped=7.0
+2023-03-15 07:19:53,229 INFO [train.py:968] (0/2) Epoch 30, batch 4500, giga_loss[loss=0.3049, simple_loss=0.3747, pruned_loss=0.1176, over 27983.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3334, pruned_loss=0.09031, over 5700650.08 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3398, pruned_loss=0.08569, over 5128762.95 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3324, pruned_loss=0.09058, over 5703903.38 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:19:55,391 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1324971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:20:05,603 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4271, 1.7492, 1.3911, 1.3812], device='cuda:0'), covar=tensor([0.2846, 0.2889, 0.3333, 0.2528], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1166, 0.1431, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 07:20:34,066 INFO [train.py:968] (0/2) Epoch 30, batch 4550, giga_loss[loss=0.2809, simple_loss=0.3598, pruned_loss=0.101, over 28777.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3358, pruned_loss=0.09083, over 5704987.06 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3398, pruned_loss=0.08573, over 5136506.25 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3349, pruned_loss=0.09108, over 5708441.26 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:20:53,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4783, 1.7713, 1.4685, 1.2859], device='cuda:0'), covar=tensor([0.2363, 0.2414, 0.2641, 0.2461], device='cuda:0'), in_proj_covar=tensor([0.1617, 0.1166, 0.1430, 0.1014], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 07:20:59,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.077e+02 1.247e+03 1.472e+03 1.841e+03 7.577e+03, threshold=2.943e+03, percent-clipped=7.0
+2023-03-15 07:21:16,846 INFO [train.py:968] (0/2) Epoch 30, batch 4600, giga_loss[loss=0.2679, simple_loss=0.3523, pruned_loss=0.09178, over 28352.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3378, pruned_loss=0.09141, over 5701056.89 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.34, pruned_loss=0.08583, over 5143941.18 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3369, pruned_loss=0.09158, over 5702083.43 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:21:48,178 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1325105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:21:57,998 INFO [train.py:968] (0/2) Epoch 30, batch 4650, giga_loss[loss=0.2432, simple_loss=0.3175, pruned_loss=0.08441, over 29047.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3367, pruned_loss=0.09003, over 5700238.65 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3396, pruned_loss=0.08577, over 5162650.26 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3362, pruned_loss=0.09033, over 5696598.07 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:22:23,241 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.817e+02 1.177e+03 1.445e+03 1.854e+03 5.539e+03, threshold=2.890e+03, percent-clipped=3.0
+2023-03-15 07:22:36,597 INFO [train.py:968] (0/2) Epoch 30, batch 4700, giga_loss[loss=0.2611, simple_loss=0.3381, pruned_loss=0.09205, over 28966.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3368, pruned_loss=0.08972, over 5703684.27 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3398, pruned_loss=0.0858, over 5184327.40 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3362, pruned_loss=0.09005, over 5695550.73 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:23:17,413 INFO [train.py:968] (0/2) Epoch 30, batch 4750, giga_loss[loss=0.2281, simple_loss=0.3114, pruned_loss=0.07246, over 28238.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3387, pruned_loss=0.09107, over 5707083.44 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3403, pruned_loss=0.08609, over 5191765.56 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3378, pruned_loss=0.09114, over 5701013.01 frames. ], batch size: 77, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:23:35,591 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4812, 1.7562, 1.4281, 1.4351], device='cuda:0'), covar=tensor([0.2898, 0.3045, 0.3488, 0.2616], device='cuda:0'), in_proj_covar=tensor([0.1623, 0.1170, 0.1434, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 07:23:37,206 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-15 07:23:40,398 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1325248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:23:40,741 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.044e+03 1.371e+03 1.777e+03 2.684e+03 7.944e+03, threshold=3.553e+03, percent-clipped=22.0
+2023-03-15 07:23:42,224 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1325251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:23:47,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4444, 1.8136, 1.7538, 1.5272], device='cuda:0'), covar=tensor([0.2383, 0.2299, 0.2350, 0.2560], device='cuda:0'), in_proj_covar=tensor([0.0518, 0.0765, 0.0737, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 07:23:54,148 INFO [train.py:968] (0/2) Epoch 30, batch 4800, giga_loss[loss=0.2833, simple_loss=0.3608, pruned_loss=0.1029, over 27927.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3395, pruned_loss=0.09172, over 5714185.06 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3405, pruned_loss=0.08627, over 5216551.15 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3386, pruned_loss=0.09179, over 5703087.55 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:24:04,055 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1325280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:24:34,513 INFO [train.py:968] (0/2) Epoch 30, batch 4850, giga_loss[loss=0.2459, simple_loss=0.3331, pruned_loss=0.07941, over 29058.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3418, pruned_loss=0.09318, over 5717677.17 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3403, pruned_loss=0.08625, over 5233576.41 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3411, pruned_loss=0.09338, over 5704737.99 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:24:54,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5087, 1.7864, 1.5288, 1.3773], device='cuda:0'), covar=tensor([0.2170, 0.2211, 0.2361, 0.2260], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1166, 0.1431, 0.1016], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 07:24:57,549 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1325346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:24:59,262 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.038e+03 1.441e+03 1.732e+03 2.579e+03 4.884e+03, threshold=3.465e+03, percent-clipped=7.0
+2023-03-15 07:25:12,955 INFO [train.py:968] (0/2) Epoch 30, batch 4900, giga_loss[loss=0.2637, simple_loss=0.3383, pruned_loss=0.09449, over 28860.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3441, pruned_loss=0.09454, over 5718727.90 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3404, pruned_loss=0.08617, over 5246705.11 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3436, pruned_loss=0.0949, over 5705249.18 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:25:57,426 INFO [train.py:968] (0/2) Epoch 30, batch 4950, giga_loss[loss=0.2843, simple_loss=0.3612, pruned_loss=0.1037, over 27611.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.346, pruned_loss=0.09544, over 5714831.38 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3404, pruned_loss=0.08617, over 5246705.11 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3456, pruned_loss=0.09572, over 5704340.73 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:26:20,721 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.407e+03 1.693e+03 2.351e+03 4.833e+03, threshold=3.385e+03, percent-clipped=3.0
+2023-03-15 07:26:34,075 INFO [train.py:968] (0/2) Epoch 30, batch 5000, giga_loss[loss=0.2797, simple_loss=0.3441, pruned_loss=0.1076, over 28526.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.347, pruned_loss=0.09612, over 5719576.09 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3407, pruned_loss=0.08633, over 5256258.85 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3466, pruned_loss=0.09634, over 5709077.42 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:26:50,186 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 07:26:53,445 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1325489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:26:56,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1325492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:27:17,003 INFO [train.py:968] (0/2) Epoch 30, batch 5050, giga_loss[loss=0.2598, simple_loss=0.333, pruned_loss=0.09335, over 28691.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3483, pruned_loss=0.09737, over 5710991.67 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3405, pruned_loss=0.08636, over 5262562.21 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3481, pruned_loss=0.09761, over 5701216.15 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:27:19,778 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1325521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:27:26,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4085, 1.9124, 1.5206, 1.6299], device='cuda:0'), covar=tensor([0.0780, 0.0277, 0.0335, 0.0905], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 07:27:29,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1202, 1.2333, 1.1392, 0.8218], device='cuda:0'), covar=tensor([0.1083, 0.0565, 0.1104, 0.1181], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0451, 0.0526, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 07:27:41,030 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.869e+02 1.372e+03 1.665e+03 2.400e+03 7.150e+03, threshold=3.329e+03, percent-clipped=6.0
+2023-03-15 07:27:57,222 INFO [train.py:968] (0/2) Epoch 30, batch 5100, giga_loss[loss=0.2593, simple_loss=0.3341, pruned_loss=0.09226, over 28946.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3479, pruned_loss=0.09703, over 5715082.90 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3412, pruned_loss=0.08665, over 5271687.17 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3474, pruned_loss=0.09717, over 5707266.61 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:27:59,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1969, 1.4348, 1.4798, 1.0750], device='cuda:0'), covar=tensor([0.1800, 0.2623, 0.1522, 0.1740], device='cuda:0'), in_proj_covar=tensor([0.0946, 0.0724, 0.0997, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 07:28:36,934 INFO [train.py:968] (0/2) Epoch 30, batch 5150, giga_loss[loss=0.2408, simple_loss=0.3174, pruned_loss=0.08205, over 28877.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3458, pruned_loss=0.0962, over 5708463.17 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3414, pruned_loss=0.08671, over 5286954.09 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3453, pruned_loss=0.09649, over 5698462.24 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:29:01,932 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.131e+02 1.183e+03 1.529e+03 1.929e+03 6.746e+03, threshold=3.058e+03, percent-clipped=5.0
+2023-03-15 07:29:13,528 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1325665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:29:15,495 INFO [train.py:968] (0/2) Epoch 30, batch 5200, giga_loss[loss=0.283, simple_loss=0.3562, pruned_loss=0.1048, over 28290.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3421, pruned_loss=0.09431, over 5716440.88 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3419, pruned_loss=0.08699, over 5301900.69 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3413, pruned_loss=0.09449, over 5704334.61 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:29:53,496 INFO [train.py:968] (0/2) Epoch 30, batch 5250, giga_loss[loss=0.2535, simple_loss=0.3303, pruned_loss=0.08834, over 28928.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3403, pruned_loss=0.09255, over 5720171.09 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3416, pruned_loss=0.08694, over 5318715.66 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3398, pruned_loss=0.09291, over 5708051.09 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:29:59,431 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3438, 1.5875, 1.3286, 1.4843], device='cuda:0'), covar=tensor([0.0732, 0.0325, 0.0351, 0.0911], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 07:30:14,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3242, 1.5154, 1.5679, 1.1787], device='cuda:0'), covar=tensor([0.1897, 0.2617, 0.1634, 0.1883], device='cuda:0'), in_proj_covar=tensor([0.0944, 0.0721, 0.0994, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 07:30:17,444 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.486e+02 1.276e+03 1.684e+03 2.358e+03 8.168e+03, threshold=3.367e+03, percent-clipped=14.0
+2023-03-15 07:30:34,134 INFO [train.py:968] (0/2) Epoch 30, batch 5300, giga_loss[loss=0.2612, simple_loss=0.3491, pruned_loss=0.08661, over 28836.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3413, pruned_loss=0.09179, over 5721291.92 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3417, pruned_loss=0.08698, over 5333073.99 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3409, pruned_loss=0.09216, over 5707556.99 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:31:13,722 INFO [train.py:968] (0/2) Epoch 30, batch 5350, giga_loss[loss=0.2502, simple_loss=0.3316, pruned_loss=0.08438, over 28819.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.342, pruned_loss=0.0916, over 5725824.35 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3416, pruned_loss=0.0871, over 5343693.75 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3417, pruned_loss=0.09187, over 5712293.87 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:31:19,250 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4969, 3.5710, 1.5597, 1.6514], device='cuda:0'), covar=tensor([0.0984, 0.0383, 0.0952, 0.1327], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0574, 0.0415, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 07:31:39,383 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.060e+02 1.277e+03 1.734e+03 2.069e+03 4.076e+03, threshold=3.467e+03, percent-clipped=2.0
+2023-03-15 07:31:44,808 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1325856.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:31:52,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1360, 1.2713, 3.4052, 2.9162], device='cuda:0'), covar=tensor([0.1673, 0.2642, 0.0493, 0.1041], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0673, 0.1008, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 07:31:52,936 INFO [train.py:968] (0/2) Epoch 30, batch 5400, giga_loss[loss=0.2694, simple_loss=0.3465, pruned_loss=0.09615, over 28764.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3408, pruned_loss=0.09212, over 5727239.44 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3411, pruned_loss=0.08683, over 5362799.67 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3409, pruned_loss=0.09274, over 5710724.79 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:32:32,756 INFO [train.py:968] (0/2) Epoch 30, batch 5450, giga_loss[loss=0.2777, simple_loss=0.3529, pruned_loss=0.1013, over 28571.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3405, pruned_loss=0.09336, over 5731548.00 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3413, pruned_loss=0.08697, over 5376935.40 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3404, pruned_loss=0.09389, over 5715438.00 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:32:43,722 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1325932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:32:56,358 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.286e+02 1.336e+03 1.589e+03 2.240e+03 4.078e+03, threshold=3.177e+03, percent-clipped=1.0
+2023-03-15 07:32:59,205 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7192, 1.9077, 1.5800, 1.9758], device='cuda:0'), covar=tensor([0.2731, 0.2999, 0.3332, 0.2710], device='cuda:0'), in_proj_covar=tensor([0.1613, 0.1162, 0.1427, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 07:33:11,029 INFO [train.py:968] (0/2) Epoch 30, batch 5500, giga_loss[loss=0.2391, simple_loss=0.3063, pruned_loss=0.08596, over 28849.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3372, pruned_loss=0.09264, over 5736431.59 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3407, pruned_loss=0.08655, over 5387234.33 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3377, pruned_loss=0.09354, over 5720742.10 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:33:31,533 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 07:33:35,816 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1326000.pt
+2023-03-15 07:33:49,024 INFO [train.py:968] (0/2) Epoch 30, batch 5550, giga_loss[loss=0.2452, simple_loss=0.3231, pruned_loss=0.08363, over 29033.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3369, pruned_loss=0.09373, over 5722973.05 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3407, pruned_loss=0.08666, over 5383622.88 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3372, pruned_loss=0.0945, over 5722727.66 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:34:09,098 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:34:15,568 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.337e+03 1.649e+03 1.957e+03 5.944e+03, threshold=3.298e+03, percent-clipped=6.0
+2023-03-15 07:34:32,740 INFO [train.py:968] (0/2) Epoch 30, batch 5600, giga_loss[loss=0.2168, simple_loss=0.2964, pruned_loss=0.06863, over 29017.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3368, pruned_loss=0.09378, over 5704798.92 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3412, pruned_loss=0.08705, over 5385894.88 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3364, pruned_loss=0.09423, over 5710587.26 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:35:09,683 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:35:11,547 INFO [train.py:968] (0/2) Epoch 30, batch 5650, giga_loss[loss=0.2063, simple_loss=0.2894, pruned_loss=0.06166, over 28989.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3338, pruned_loss=0.09203, over 5706196.60 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3417, pruned_loss=0.0874, over 5398092.03 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3328, pruned_loss=0.09236, over 5713938.54 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:35:29,883 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:35:34,870 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.131e+02 1.423e+03 1.833e+03 2.633e+03 6.241e+03, threshold=3.665e+03, percent-clipped=10.0
+2023-03-15 07:35:38,516 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7439, 1.9681, 1.9332, 1.6703], device='cuda:0'), covar=tensor([0.2776, 0.2411, 0.2061, 0.2446], device='cuda:0'), in_proj_covar=tensor([0.2081, 0.2047, 0.1958, 0.2094], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 07:35:45,336 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8596, 1.0578, 2.9260, 2.8817], device='cuda:0'), covar=tensor([0.1701, 0.2627, 0.0658, 0.1016], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0673, 0.1007, 0.0987], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 07:35:48,744 INFO [train.py:968] (0/2) Epoch 30, batch 5700, giga_loss[loss=0.2817, simple_loss=0.3415, pruned_loss=0.1109, over 27578.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3303, pruned_loss=0.09026, over 5706736.19 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3417, pruned_loss=0.08743, over 5417710.80 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3291, pruned_loss=0.09063, over 5706044.22 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:36:01,211 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:36:03,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:36:25,937 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1326215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:36:27,546 INFO [train.py:968] (0/2) Epoch 30, batch 5750, giga_loss[loss=0.3023, simple_loss=0.3676, pruned_loss=0.1185, over 28897.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3284, pruned_loss=0.08947, over 5714514.11 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3417, pruned_loss=0.08745, over 5429842.09 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3272, pruned_loss=0.08981, over 5709268.49 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:36:38,854 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326231.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:36:52,152 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.267e+02 1.363e+03 1.707e+03 2.130e+03 5.618e+03, threshold=3.414e+03, percent-clipped=4.0
+2023-03-15 07:37:05,417 INFO [train.py:968] (0/2) Epoch 30, batch 5800, giga_loss[loss=0.2521, simple_loss=0.3336, pruned_loss=0.08531, over 28679.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3279, pruned_loss=0.08905, over 5718271.84 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3411, pruned_loss=0.08727, over 5438307.52 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3271, pruned_loss=0.0895, over 5711563.09 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:37:36,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:37:44,588 INFO [train.py:968] (0/2) Epoch 30, batch 5850, giga_loss[loss=0.2489, simple_loss=0.3267, pruned_loss=0.08549, over 29092.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3307, pruned_loss=0.09026, over 5719196.48 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3407, pruned_loss=0.08708, over 5447737.50 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3303, pruned_loss=0.09083, over 5710343.05 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:38:09,899 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.398e+03 1.653e+03 2.068e+03 4.774e+03, threshold=3.305e+03, percent-clipped=5.0
+2023-03-15 07:38:24,648 INFO [train.py:968] (0/2) Epoch 30, batch 5900, giga_loss[loss=0.2764, simple_loss=0.3548, pruned_loss=0.09895, over 28855.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.335, pruned_loss=0.09217, over 5717972.00 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3408, pruned_loss=0.08714, over 5454374.77 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3345, pruned_loss=0.09265, over 5708522.66 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:38:29,353 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326374.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:38:31,723 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326377.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:38:54,453 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1326406.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:39:03,612 INFO [train.py:968] (0/2) Epoch 30, batch 5950, giga_loss[loss=0.2841, simple_loss=0.3602, pruned_loss=0.104, over 27846.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3381, pruned_loss=0.09283, over 5716427.62 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.341, pruned_loss=0.08722, over 5461241.64 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3374, pruned_loss=0.09327, over 5709723.73 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:39:29,735 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5004, 1.8655, 1.2981, 1.3738], device='cuda:0'), covar=tensor([0.1112, 0.0608, 0.1096, 0.1228], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0450, 0.0524, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 07:39:31,836 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:39:32,271 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.563e+02 1.367e+03 1.680e+03 2.312e+03 5.798e+03, threshold=3.360e+03, percent-clipped=10.0
+2023-03-15 07:39:35,582 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:39:47,080 INFO [train.py:968] (0/2) Epoch 30, batch 6000, giga_loss[loss=0.2766, simple_loss=0.3484, pruned_loss=0.1024, over 28825.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3413, pruned_loss=0.09456, over 5710097.77 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3413, pruned_loss=0.08742, over 5466704.48 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3405, pruned_loss=0.09482, over 5703231.93 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:39:47,085 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 07:39:56,253 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.2064, simple_loss=0.3141, pruned_loss=0.0493, over 944034.00 frames. 
+2023-03-15 07:39:56,254 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 07:40:07,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:40:08,105 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1326482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:40:15,563 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:40:39,816 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:40:40,156 INFO [train.py:968] (0/2) Epoch 30, batch 6050, giga_loss[loss=0.3226, simple_loss=0.3888, pruned_loss=0.1282, over 28617.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3455, pruned_loss=0.09797, over 5703554.09 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3413, pruned_loss=0.08732, over 5472896.19 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3449, pruned_loss=0.0984, over 5695747.69 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:40:57,457 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:41:07,440 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.715e+02 1.627e+03 1.987e+03 2.888e+03 5.831e+03, threshold=3.975e+03, percent-clipped=16.0
+2023-03-15 07:41:08,531 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:41:23,846 INFO [train.py:968] (0/2) Epoch 30, batch 6100, giga_loss[loss=0.2697, simple_loss=0.3438, pruned_loss=0.0978, over 28737.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3509, pruned_loss=0.1028, over 5690360.83 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3409, pruned_loss=0.0874, over 5474440.90 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3509, pruned_loss=0.1034, over 5688312.49 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:42:04,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5937, 2.0330, 1.5636, 1.9814], device='cuda:0'), covar=tensor([0.2726, 0.2671, 0.3143, 0.2229], device='cuda:0'), in_proj_covar=tensor([0.1617, 0.1165, 0.1431, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 07:42:08,732 INFO [train.py:968] (0/2) Epoch 30, batch 6150, giga_loss[loss=0.3041, simple_loss=0.3818, pruned_loss=0.1132, over 28912.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3569, pruned_loss=0.1064, over 5692837.22 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3412, pruned_loss=0.08776, over 5476762.51 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3569, pruned_loss=0.107, over 5694747.84 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:42:22,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:42:25,493 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:42:41,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.200e+03 1.793e+03 2.179e+03 3.260e+03 7.533e+03, threshold=4.358e+03, percent-clipped=11.0
+2023-03-15 07:42:42,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5162, 4.6147, 1.6787, 1.7621], device='cuda:0'), covar=tensor([0.1046, 0.0344, 0.0928, 0.1334], device='cuda:0'), in_proj_covar=tensor([0.0427, 0.0572, 0.0414, 0.0449], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 07:42:48,297 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:42:51,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:42:54,966 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1326665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:42:58,686 INFO [train.py:968] (0/2) Epoch 30, batch 6200, giga_loss[loss=0.3522, simple_loss=0.4055, pruned_loss=0.1495, over 28974.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3632, pruned_loss=0.1115, over 5691478.63 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3416, pruned_loss=0.08798, over 5481775.15 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3632, pruned_loss=0.1121, over 5691157.39 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:43:19,580 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1326692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:43:26,770 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:43:42,406 INFO [train.py:968] (0/2) Epoch 30, batch 6250, giga_loss[loss=0.3158, simple_loss=0.3864, pruned_loss=0.1226, over 28685.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.371, pruned_loss=0.1182, over 5690805.86 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3414, pruned_loss=0.0879, over 5489500.27 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3717, pruned_loss=0.1193, over 5687401.93 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:44:09,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.201e+03 2.000e+03 2.624e+03 3.200e+03 7.774e+03, threshold=5.247e+03, percent-clipped=10.0
+2023-03-15 07:44:24,870 INFO [train.py:968] (0/2) Epoch 30, batch 6300, giga_loss[loss=0.3404, simple_loss=0.4036, pruned_loss=0.1386, over 28664.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3745, pruned_loss=0.1209, over 5681003.33 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3415, pruned_loss=0.088, over 5495320.91 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3761, pruned_loss=0.1228, over 5680204.37 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:44:30,278 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9801, 2.3342, 1.6929, 1.9362], device='cuda:0'), covar=tensor([0.1134, 0.0668, 0.1001, 0.1194], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0453, 0.0527, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 07:45:14,312 INFO [train.py:968] (0/2) Epoch 30, batch 6350, giga_loss[loss=0.3379, simple_loss=0.3994, pruned_loss=0.1382, over 28836.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3772, pruned_loss=0.1241, over 5668659.74 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3415, pruned_loss=0.08799, over 5503489.11 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3791, pruned_loss=0.1262, over 5664116.29 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:45:48,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.089e+03 1.838e+03 2.438e+03 3.638e+03 1.710e+04, threshold=4.877e+03, percent-clipped=8.0
+2023-03-15 07:45:50,981 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:46:04,233 INFO [train.py:968] (0/2) Epoch 30, batch 6400, giga_loss[loss=0.2918, simple_loss=0.362, pruned_loss=0.1108, over 29066.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3801, pruned_loss=0.1271, over 5670466.30 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3416, pruned_loss=0.08798, over 5509918.46 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3822, pruned_loss=0.1295, over 5664009.00 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:46:38,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.8166, 3.6589, 3.4968, 1.6789], device='cuda:0'), covar=tensor([0.0778, 0.0860, 0.0823, 0.2133], device='cuda:0'), in_proj_covar=tensor([0.1318, 0.1214, 0.1020, 0.0760], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 07:46:45,103 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-15 07:46:50,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:46:52,845 INFO [train.py:968] (0/2) Epoch 30, batch 6450, giga_loss[loss=0.3177, simple_loss=0.3775, pruned_loss=0.1289, over 28841.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3829, pruned_loss=0.1305, over 5664605.23 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3417, pruned_loss=0.08798, over 5518507.45 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3856, pruned_loss=0.1336, over 5655547.11 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:47:01,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:47:26,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.314e+03 2.056e+03 2.461e+03 3.444e+03 1.221e+04, threshold=4.922e+03, percent-clipped=8.0
+2023-03-15 07:47:41,241 INFO [train.py:968] (0/2) Epoch 30, batch 6500, giga_loss[loss=0.2948, simple_loss=0.3674, pruned_loss=0.1111, over 28819.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3854, pruned_loss=0.1329, over 5648507.52 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3418, pruned_loss=0.08801, over 5519877.30 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3887, pruned_loss=0.1364, over 5643093.63 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:48:13,580 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:48:16,318 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:48:30,326 INFO [train.py:968] (0/2) Epoch 30, batch 6550, giga_loss[loss=0.3786, simple_loss=0.423, pruned_loss=0.1671, over 27972.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3859, pruned_loss=0.1337, over 5646627.60 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3424, pruned_loss=0.08832, over 5523987.95 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3884, pruned_loss=0.1369, over 5640328.67 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:48:44,251 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:48:53,278 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1327042.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:49:00,929 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.301e+03 1.864e+03 2.317e+03 3.119e+03 8.624e+03, threshold=4.634e+03, percent-clipped=5.0
+2023-03-15 07:49:06,013 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1327057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:12,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:19,759 INFO [train.py:968] (0/2) Epoch 30, batch 6600, giga_loss[loss=0.4232, simple_loss=0.4491, pruned_loss=0.1986, over 27533.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3837, pruned_loss=0.1332, over 5638552.88 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3415, pruned_loss=0.08787, over 5527107.87 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3879, pruned_loss=0.1375, over 5634591.67 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:49:21,558 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1327070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:24,417 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:25,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1327075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:39,162 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:51,471 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:50:07,074 INFO [train.py:968] (0/2) Epoch 30, batch 6650, libri_loss[loss=0.2626, simple_loss=0.3532, pruned_loss=0.08599, over 29367.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3824, pruned_loss=0.1324, over 5637000.58 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3412, pruned_loss=0.08753, over 5536490.90 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3875, pruned_loss=0.1377, over 5628516.19 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:50:36,145 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.254e+03 2.004e+03 2.887e+03 3.822e+03 1.288e+04, threshold=5.774e+03, percent-clipped=17.0
+2023-03-15 07:50:42,216 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-15 07:50:53,903 INFO [train.py:968] (0/2) Epoch 30, batch 6700, giga_loss[loss=0.2755, simple_loss=0.3543, pruned_loss=0.09835, over 28571.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3823, pruned_loss=0.1306, over 5648034.05 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3414, pruned_loss=0.08758, over 5545173.75 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3871, pruned_loss=0.1359, over 5635703.56 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:51:00,327 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1304, 2.3797, 2.2640, 1.8370], device='cuda:0'), covar=tensor([0.3059, 0.2735, 0.2768, 0.3072], device='cuda:0'), in_proj_covar=tensor([0.2092, 0.2065, 0.1965, 0.2102], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 07:51:41,031 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-15 07:51:42,421 INFO [train.py:968] (0/2) Epoch 30, batch 6750, giga_loss[loss=0.2889, simple_loss=0.3595, pruned_loss=0.1092, over 28595.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3848, pruned_loss=0.132, over 5646690.66 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3415, pruned_loss=0.08768, over 5546970.88 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3886, pruned_loss=0.1363, over 5635760.29 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:51:42,772 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1327218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:51:47,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:52:08,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4113, 1.2952, 4.5273, 3.5937], device='cuda:0'), covar=tensor([0.1717, 0.2873, 0.0438, 0.0973], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0675, 0.1011, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 07:52:13,038 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:52:14,101 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.151e+03 1.827e+03 2.255e+03 3.333e+03 1.249e+04, threshold=4.510e+03, percent-clipped=1.0
+2023-03-15 07:52:29,592 INFO [train.py:968] (0/2) Epoch 30, batch 6800, giga_loss[loss=0.2577, simple_loss=0.3376, pruned_loss=0.08893, over 28895.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3828, pruned_loss=0.1309, over 5634896.50 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3415, pruned_loss=0.08761, over 5553761.45 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3868, pruned_loss=0.1352, over 5621619.53 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:53:18,134 INFO [train.py:968] (0/2) Epoch 30, batch 6850, giga_loss[loss=0.2968, simple_loss=0.3736, pruned_loss=0.11, over 28633.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.381, pruned_loss=0.1282, over 5643506.74 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3415, pruned_loss=0.08779, over 5561731.21 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.385, pruned_loss=0.1324, over 5627592.76 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:53:28,910 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1327328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:53:52,317 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.100e+03 1.764e+03 2.309e+03 3.884e+03 9.637e+03, threshold=4.618e+03, percent-clipped=18.0
+2023-03-15 07:54:05,786 INFO [train.py:968] (0/2) Epoch 30, batch 6900, giga_loss[loss=0.2822, simple_loss=0.3559, pruned_loss=0.1042, over 28915.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3777, pruned_loss=0.1247, over 5654767.90 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3419, pruned_loss=0.0881, over 5571345.68 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3817, pruned_loss=0.1288, over 5635359.94 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:54:46,506 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-15 07:54:48,350 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1327417.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:54:48,664 INFO [train.py:968] (0/2) Epoch 30, batch 6950, giga_loss[loss=0.3067, simple_loss=0.3767, pruned_loss=0.1183, over 28618.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3729, pruned_loss=0.1206, over 5650612.09 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3416, pruned_loss=0.08812, over 5566857.56 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3779, pruned_loss=0.1256, over 5643082.26 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:54:58,698 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5779, 1.9417, 1.7066, 1.6540], device='cuda:0'), covar=tensor([0.2442, 0.2367, 0.2759, 0.2497], device='cuda:0'), in_proj_covar=tensor([0.0517, 0.0767, 0.0739, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 07:55:25,286 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.756e+03 2.119e+03 2.819e+03 5.956e+03, threshold=4.238e+03, percent-clipped=3.0
+2023-03-15 07:55:35,396 INFO [train.py:968] (0/2) Epoch 30, batch 7000, libri_loss[loss=0.2569, simple_loss=0.3351, pruned_loss=0.08941, over 29567.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3698, pruned_loss=0.1184, over 5654659.63 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3417, pruned_loss=0.08831, over 5574786.48 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3745, pruned_loss=0.1229, over 5642930.79 frames. ], batch size: 76, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:56:20,631 INFO [train.py:968] (0/2) Epoch 30, batch 7050, giga_loss[loss=0.2758, simple_loss=0.3533, pruned_loss=0.0992, over 28856.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3673, pruned_loss=0.1165, over 5658196.21 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.08836, over 5584333.32 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3717, pruned_loss=0.1209, over 5642083.59 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:56:31,050 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1327531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:56:52,574 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.077e+03 1.642e+03 2.125e+03 2.993e+03 6.815e+03, threshold=4.251e+03, percent-clipped=13.0
+2023-03-15 07:56:59,547 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1327560.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:57:03,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327563.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:57:07,600 INFO [train.py:968] (0/2) Epoch 30, batch 7100, giga_loss[loss=0.3157, simple_loss=0.3888, pruned_loss=0.1213, over 29025.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3675, pruned_loss=0.1169, over 5654034.30 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3417, pruned_loss=0.08832, over 5591156.94 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3716, pruned_loss=0.121, over 5636447.56 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:57:17,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4504, 1.9818, 1.8000, 1.7321], device='cuda:0'), covar=tensor([0.2364, 0.1778, 0.2220, 0.1916], device='cuda:0'), in_proj_covar=tensor([0.0516, 0.0765, 0.0738, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 07:57:30,637 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327592.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:57:32,340 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.54 vs. limit=5.0
+2023-03-15 07:57:52,700 INFO [train.py:968] (0/2) Epoch 30, batch 7150, giga_loss[loss=0.3041, simple_loss=0.3803, pruned_loss=0.1139, over 28923.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3655, pruned_loss=0.115, over 5658184.00 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3414, pruned_loss=0.08814, over 5595097.51 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3696, pruned_loss=0.119, over 5642128.09 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:58:27,432 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.016e+03 1.721e+03 2.165e+03 3.256e+03 8.488e+03, threshold=4.330e+03, percent-clipped=11.0
+2023-03-15 07:58:41,912 INFO [train.py:968] (0/2) Epoch 30, batch 7200, giga_loss[loss=0.2729, simple_loss=0.3591, pruned_loss=0.0934, over 28975.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3653, pruned_loss=0.1126, over 5669457.91 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3412, pruned_loss=0.08804, over 5603177.80 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3693, pruned_loss=0.1165, over 5650851.15 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:58:43,633 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-15 07:59:08,726 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3689, 1.8249, 1.6240, 1.5790], device='cuda:0'), covar=tensor([0.2417, 0.2161, 0.2693, 0.2453], device='cuda:0'), in_proj_covar=tensor([0.0517, 0.0765, 0.0738, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 07:59:12,080 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1327703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:59:19,845 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1327712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:59:26,074 INFO [train.py:968] (0/2) Epoch 30, batch 7250, giga_loss[loss=0.3159, simple_loss=0.3909, pruned_loss=0.1204, over 28872.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3666, pruned_loss=0.1116, over 5679293.04 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3406, pruned_loss=0.08772, over 5614541.28 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3713, pruned_loss=0.1159, over 5656553.24 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:59:56,951 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.126e+03 1.711e+03 2.183e+03 3.553e+03 1.350e+04, threshold=4.365e+03, percent-clipped=14.0
+2023-03-15 08:00:09,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1788, 1.4832, 1.5354, 1.3104], device='cuda:0'), covar=tensor([0.2012, 0.1710, 0.2243, 0.1866], device='cuda:0'), in_proj_covar=tensor([0.0517, 0.0765, 0.0739, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 08:00:12,989 INFO [train.py:968] (0/2) Epoch 30, batch 7300, giga_loss[loss=0.3406, simple_loss=0.3925, pruned_loss=0.1443, over 27663.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.367, pruned_loss=0.1123, over 5680639.29 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3403, pruned_loss=0.08767, over 5625783.60 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.372, pruned_loss=0.1167, over 5654535.96 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:00:57,643 INFO [train.py:968] (0/2) Epoch 30, batch 7350, giga_loss[loss=0.2959, simple_loss=0.3668, pruned_loss=0.1125, over 28762.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3679, pruned_loss=0.1137, over 5663640.19 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3404, pruned_loss=0.08782, over 5619633.06 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3723, pruned_loss=0.1176, over 5649564.15 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:01:03,453 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5870, 3.4708, 1.7016, 1.6278], device='cuda:0'), covar=tensor([0.0943, 0.0366, 0.0857, 0.1284], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0578, 0.0416, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 08:01:22,199 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1327846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:01:24,602 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:01:29,671 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.142e+03 1.862e+03 2.348e+03 3.035e+03 7.013e+03, threshold=4.696e+03, percent-clipped=5.0
+2023-03-15 08:01:30,876 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3440, 3.7058, 1.5895, 1.5441], device='cuda:0'), covar=tensor([0.1007, 0.0363, 0.0873, 0.1275], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0579, 0.0416, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 08:01:30,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6713, 2.2583, 1.4498, 0.8039], device='cuda:0'), covar=tensor([0.8674, 0.4314, 0.4400, 0.8288], device='cuda:0'), in_proj_covar=tensor([0.1862, 0.1746, 0.1669, 0.1518], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 08:01:45,811 INFO [train.py:968] (0/2) Epoch 30, batch 7400, libri_loss[loss=0.2146, simple_loss=0.294, pruned_loss=0.06761, over 29470.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.367, pruned_loss=0.1142, over 5675204.83 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3401, pruned_loss=0.08775, over 5625401.98 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3713, pruned_loss=0.1179, over 5659706.36 frames. ], batch size: 70, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:01:54,588 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:02:19,551 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1327906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:02:30,976 INFO [train.py:968] (0/2) Epoch 30, batch 7450, giga_loss[loss=0.3105, simple_loss=0.3736, pruned_loss=0.1236, over 28854.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3665, pruned_loss=0.1154, over 5670084.07 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3406, pruned_loss=0.0881, over 5627708.50 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.37, pruned_loss=0.1185, over 5656695.62 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:03:02,062 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.212e+03 1.824e+03 2.183e+03 3.056e+03 6.566e+03, threshold=4.367e+03, percent-clipped=4.0
+2023-03-15 08:03:14,511 INFO [train.py:968] (0/2) Epoch 30, batch 7500, giga_loss[loss=0.2884, simple_loss=0.3777, pruned_loss=0.09959, over 28950.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3645, pruned_loss=0.1136, over 5672030.06 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3401, pruned_loss=0.08777, over 5637635.01 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3687, pruned_loss=0.1174, over 5653597.04 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:03:42,597 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1328000.pt
+2023-03-15 08:03:50,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4211, 1.9381, 1.3999, 0.8230], device='cuda:0'), covar=tensor([0.6256, 0.3276, 0.3680, 0.6991], device='cuda:0'), in_proj_covar=tensor([0.1865, 0.1749, 0.1673, 0.1519], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 08:03:51,388 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 08:03:59,783 INFO [train.py:968] (0/2) Epoch 30, batch 7550, giga_loss[loss=0.2896, simple_loss=0.364, pruned_loss=0.1076, over 28964.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3643, pruned_loss=0.1124, over 5672117.58 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3402, pruned_loss=0.08772, over 5640506.16 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3683, pruned_loss=0.1162, over 5655727.10 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:04:03,784 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4593, 1.6367, 1.5981, 1.3755], device='cuda:0'), covar=tensor([0.3220, 0.2698, 0.2518, 0.2832], device='cuda:0'), in_proj_covar=tensor([0.2090, 0.2060, 0.1963, 0.2100], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 08:04:05,825 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0614, 1.6380, 1.4809, 1.3488], device='cuda:0'), covar=tensor([0.2621, 0.2016, 0.2581, 0.2401], device='cuda:0'), in_proj_covar=tensor([0.0517, 0.0766, 0.0740, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 08:04:22,413 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1328042.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:04:25,337 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-15 08:04:27,284 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1328049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:04:30,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1328052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:04:31,821 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.005e+03 1.699e+03 2.203e+03 2.738e+03 5.516e+03, threshold=4.407e+03, percent-clipped=3.0
+2023-03-15 08:04:43,872 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4183, 1.5643, 1.4994, 1.5577], device='cuda:0'), covar=tensor([0.0608, 0.0298, 0.0280, 0.0639], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 08:04:47,572 INFO [train.py:968] (0/2) Epoch 30, batch 7600, giga_loss[loss=0.2691, simple_loss=0.353, pruned_loss=0.09258, over 29042.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3651, pruned_loss=0.1126, over 5672147.59 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.34, pruned_loss=0.08762, over 5643004.46 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3685, pruned_loss=0.1159, over 5657278.53 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:04:59,329 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1328081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:05:03,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1328087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:05:28,671 INFO [train.py:968] (0/2) Epoch 30, batch 7650, giga_loss[loss=0.3032, simple_loss=0.3718, pruned_loss=0.1173, over 28687.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3639, pruned_loss=0.1116, over 5687511.26 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.34, pruned_loss=0.08767, over 5645547.51 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3669, pruned_loss=0.1144, over 5674003.00 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:06:04,858 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.204e+03 2.002e+03 2.430e+03 3.476e+03 1.005e+04, threshold=4.859e+03, percent-clipped=16.0
+2023-03-15 08:06:14,333 INFO [train.py:968] (0/2) Epoch 30, batch 7700, giga_loss[loss=0.2793, simple_loss=0.3517, pruned_loss=0.1034, over 28997.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3619, pruned_loss=0.111, over 5682328.93 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3395, pruned_loss=0.08735, over 5652961.74 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3653, pruned_loss=0.1142, over 5665981.44 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:06:27,326 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-15 08:06:58,893 INFO [train.py:968] (0/2) Epoch 30, batch 7750, giga_loss[loss=0.2852, simple_loss=0.3556, pruned_loss=0.1074, over 28916.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3607, pruned_loss=0.1109, over 5671949.55 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.34, pruned_loss=0.08778, over 5653443.90 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.364, pruned_loss=0.1141, over 5658566.10 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:07:04,098 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-15 08:07:09,710 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1328230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:07:11,821 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1328233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:07:25,205 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1328248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:07:32,309 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.233e+03 1.710e+03 2.146e+03 3.431e+03 9.469e+03, threshold=4.292e+03, percent-clipped=8.0
+2023-03-15 08:07:36,336 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1328262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:07:41,088 INFO [train.py:968] (0/2) Epoch 30, batch 7800, giga_loss[loss=0.2968, simple_loss=0.367, pruned_loss=0.1133, over 28578.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3613, pruned_loss=0.1123, over 5669473.32 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.34, pruned_loss=0.08763, over 5658350.26 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3645, pruned_loss=0.1157, over 5654509.59 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:08:22,982 INFO [train.py:968] (0/2) Epoch 30, batch 7850, giga_loss[loss=0.3045, simple_loss=0.3738, pruned_loss=0.1176, over 28919.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3597, pruned_loss=0.1119, over 5666733.56 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3399, pruned_loss=0.08751, over 5659639.62 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3627, pruned_loss=0.1151, over 5653680.22 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:08:51,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6975, 3.5429, 3.3505, 1.7450], device='cuda:0'), covar=tensor([0.0796, 0.0879, 0.0841, 0.2191], device='cuda:0'), in_proj_covar=tensor([0.1330, 0.1230, 0.1028, 0.0762], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 08:08:56,272 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.233e+03 1.764e+03 2.236e+03 3.120e+03 8.965e+03, threshold=4.472e+03, percent-clipped=13.0
+2023-03-15 08:09:07,271 INFO [train.py:968] (0/2) Epoch 30, batch 7900, giga_loss[loss=0.3185, simple_loss=0.3709, pruned_loss=0.133, over 28598.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.359, pruned_loss=0.1122, over 5659709.81 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3396, pruned_loss=0.08733, over 5664273.94 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.362, pruned_loss=0.1154, over 5645335.75 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:09:29,738 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3423, 3.7882, 1.5237, 1.5262], device='cuda:0'), covar=tensor([0.1025, 0.0398, 0.0909, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0580, 0.0417, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 08:09:30,483 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6990, 1.6971, 1.8692, 1.4426], device='cuda:0'), covar=tensor([0.2015, 0.2581, 0.1631, 0.1844], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0722, 0.0990, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 08:09:49,302 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1328417.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:09:49,727 INFO [train.py:968] (0/2) Epoch 30, batch 7950, giga_loss[loss=0.3136, simple_loss=0.3774, pruned_loss=0.1249, over 28799.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3606, pruned_loss=0.1129, over 5667850.25 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3399, pruned_loss=0.08752, over 5664603.08 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3629, pruned_loss=0.1155, over 5656240.45 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:10:24,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.413e+03 1.883e+03 2.362e+03 3.385e+03 7.519e+03, threshold=4.724e+03, percent-clipped=7.0
+2023-03-15 08:10:33,043 INFO [train.py:968] (0/2) Epoch 30, batch 8000, giga_loss[loss=0.2909, simple_loss=0.3674, pruned_loss=0.1072, over 28852.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3612, pruned_loss=0.1127, over 5662177.24 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3398, pruned_loss=0.08737, over 5664051.10 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3636, pruned_loss=0.1154, over 5653197.19 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:10:52,206 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1328489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:11:13,499 INFO [train.py:968] (0/2) Epoch 30, batch 8050, giga_loss[loss=0.3222, simple_loss=0.3868, pruned_loss=0.1288, over 28963.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3611, pruned_loss=0.1115, over 5674163.58 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3396, pruned_loss=0.08729, over 5666338.10 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3635, pruned_loss=0.1142, over 5664954.82 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:11:25,230 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5838, 2.3466, 1.7159, 0.7935], device='cuda:0'), covar=tensor([0.8062, 0.3875, 0.4841, 0.8052], device='cuda:0'), in_proj_covar=tensor([0.1865, 0.1753, 0.1674, 0.1519], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 08:11:45,569 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.673e+03 2.165e+03 2.886e+03 5.676e+03, threshold=4.330e+03, percent-clipped=5.0
+2023-03-15 08:11:48,464 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1328560.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:11:52,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1328563.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:11:55,533 INFO [train.py:968] (0/2) Epoch 30, batch 8100, giga_loss[loss=0.3106, simple_loss=0.355, pruned_loss=0.1331, over 23528.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.362, pruned_loss=0.1117, over 5679389.43 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3396, pruned_loss=0.08724, over 5668744.54 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3642, pruned_loss=0.1142, over 5670061.86 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:12:16,162 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1328592.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:12:35,718 INFO [train.py:968] (0/2) Epoch 30, batch 8150, giga_loss[loss=0.3475, simple_loss=0.4193, pruned_loss=0.1378, over 28351.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3637, pruned_loss=0.1131, over 5687939.49 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3397, pruned_loss=0.08723, over 5675888.48 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3661, pruned_loss=0.1158, over 5674285.53 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:12:40,545 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1328623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:12:49,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6786, 3.5239, 3.3860, 2.2034], device='cuda:0'), covar=tensor([0.0711, 0.0859, 0.0801, 0.1566], device='cuda:0'), in_proj_covar=tensor([0.1336, 0.1235, 0.1033, 0.0766], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 08:13:10,802 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.864e+03 2.418e+03 3.507e+03 8.891e+03, threshold=4.836e+03, percent-clipped=16.0
+2023-03-15 08:13:22,570 INFO [train.py:968] (0/2) Epoch 30, batch 8200, libri_loss[loss=0.2202, simple_loss=0.3034, pruned_loss=0.0685, over 29370.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3666, pruned_loss=0.1162, over 5676780.35 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3398, pruned_loss=0.08724, over 5679946.10 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3692, pruned_loss=0.1193, over 5662044.46 frames. ], batch size: 67, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:14:08,273 INFO [train.py:968] (0/2) Epoch 30, batch 8250, giga_loss[loss=0.2949, simple_loss=0.3608, pruned_loss=0.1145, over 28757.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3677, pruned_loss=0.1189, over 5653170.81 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3398, pruned_loss=0.08729, over 5669334.39 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.37, pruned_loss=0.1216, over 5651121.46 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:14:41,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 2.151e+03 2.873e+03 4.301e+03 1.546e+04, threshold=5.746e+03, percent-clipped=17.0
+2023-03-15 08:14:50,029 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1328766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:14:52,557 INFO [train.py:968] (0/2) Epoch 30, batch 8300, giga_loss[loss=0.4388, simple_loss=0.4689, pruned_loss=0.2043, over 24278.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3696, pruned_loss=0.1213, over 5665154.07 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3399, pruned_loss=0.08726, over 5674971.73 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3721, pruned_loss=0.1242, over 5658317.64 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:14:53,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1328769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:15:20,501 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1328798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:15:38,667 INFO [train.py:968] (0/2) Epoch 30, batch 8350, giga_loss[loss=0.2721, simple_loss=0.3428, pruned_loss=0.1008, over 28931.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3717, pruned_loss=0.1232, over 5657825.44 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3402, pruned_loss=0.08739, over 5678342.28 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3737, pruned_loss=0.1258, over 5649377.94 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:15:47,310 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1328830.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:16:09,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.830e+03 2.522e+03 3.734e+03 8.530e+03, threshold=5.045e+03, percent-clipped=5.0
+2023-03-15 08:16:15,766 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1328864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:16:18,281 INFO [train.py:968] (0/2) Epoch 30, batch 8400, giga_loss[loss=0.3454, simple_loss=0.3887, pruned_loss=0.151, over 28506.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.1209, over 5673532.32 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3403, pruned_loss=0.08733, over 5683604.44 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3719, pruned_loss=0.1243, over 5661375.64 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:16:58,543 INFO [train.py:968] (0/2) Epoch 30, batch 8450, giga_loss[loss=0.2639, simple_loss=0.3399, pruned_loss=0.09394, over 28888.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 5686797.05 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3402, pruned_loss=0.08733, over 5687982.84 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3709, pruned_loss=0.1214, over 5672953.56 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:17:26,403 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-15 08:17:30,703 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.242e+03 1.806e+03 2.184e+03 2.896e+03 8.194e+03, threshold=4.367e+03, percent-clipped=4.0
+2023-03-15 08:17:39,688 INFO [train.py:968] (0/2) Epoch 30, batch 8500, giga_loss[loss=0.2999, simple_loss=0.3772, pruned_loss=0.1113, over 28983.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3672, pruned_loss=0.1166, over 5682942.80 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3408, pruned_loss=0.08762, over 5691590.56 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3696, pruned_loss=0.1199, over 5668440.89 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:18:13,578 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1329007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:18:16,039 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1329010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:18:21,770 INFO [train.py:968] (0/2) Epoch 30, batch 8550, giga_loss[loss=0.2821, simple_loss=0.353, pruned_loss=0.1056, over 28645.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3653, pruned_loss=0.116, over 5677763.63 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3415, pruned_loss=0.08798, over 5695638.90 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3671, pruned_loss=0.1188, over 5662162.34 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:18:24,637 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1329022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:18:30,504 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-15 08:18:38,710 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1329039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:18:54,731 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.074e+03 1.711e+03 2.307e+03 3.002e+03 8.397e+03, threshold=4.615e+03, percent-clipped=8.0
+2023-03-15 08:19:02,645 INFO [train.py:968] (0/2) Epoch 30, batch 8600, giga_loss[loss=0.4355, simple_loss=0.4473, pruned_loss=0.2119, over 26613.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3637, pruned_loss=0.1154, over 5685631.03 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3419, pruned_loss=0.08827, over 5699648.39 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3653, pruned_loss=0.118, over 5669244.02 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:19:50,493 INFO [train.py:968] (0/2) Epoch 30, batch 8650, giga_loss[loss=0.2983, simple_loss=0.3729, pruned_loss=0.1119, over 28566.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3657, pruned_loss=0.1168, over 5682927.57 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3418, pruned_loss=0.08823, over 5702255.19 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3675, pruned_loss=0.1195, over 5667344.23 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:20:14,807 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5003, 1.8944, 1.4393, 1.6573], device='cuda:0'), covar=tensor([0.2610, 0.2640, 0.3006, 0.2441], device='cuda:0'), in_proj_covar=tensor([0.1620, 0.1166, 0.1434, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 08:20:26,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.741e+03 2.126e+03 2.816e+03 6.301e+03, threshold=4.252e+03, percent-clipped=3.0
+2023-03-15 08:20:35,751 INFO [train.py:968] (0/2) Epoch 30, batch 8700, giga_loss[loss=0.3159, simple_loss=0.3944, pruned_loss=0.1187, over 28682.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3684, pruned_loss=0.1175, over 5683539.90 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3416, pruned_loss=0.08806, over 5704277.65 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3706, pruned_loss=0.1204, over 5668841.90 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:20:58,354 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 08:21:06,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1329205.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:21:16,469 INFO [train.py:968] (0/2) Epoch 30, batch 8750, giga_loss[loss=0.257, simple_loss=0.3411, pruned_loss=0.08649, over 28638.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3699, pruned_loss=0.1161, over 5685812.50 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3416, pruned_loss=0.08822, over 5710400.65 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3728, pruned_loss=0.1196, over 5667484.68 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:21:45,024 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2440, 1.8106, 1.3876, 0.4800], device='cuda:0'), covar=tensor([0.5254, 0.3458, 0.4742, 0.7172], device='cuda:0'), in_proj_covar=tensor([0.1863, 0.1752, 0.1673, 0.1520], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 08:21:51,779 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.668e+02 1.686e+03 2.090e+03 2.766e+03 6.833e+03, threshold=4.180e+03, percent-clipped=8.0
+2023-03-15 08:22:01,003 INFO [train.py:968] (0/2) Epoch 30, batch 8800, libri_loss[loss=0.2633, simple_loss=0.3465, pruned_loss=0.09009, over 25920.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3724, pruned_loss=0.1175, over 5685390.11 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3416, pruned_loss=0.08819, over 5708473.10 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.375, pruned_loss=0.1205, over 5672839.24 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:22:03,127 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6486, 1.7313, 1.8284, 1.4110], device='cuda:0'), covar=tensor([0.1869, 0.2541, 0.1504, 0.1803], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0722, 0.0990, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 08:22:19,496 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1329289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:22:40,735 INFO [train.py:968] (0/2) Epoch 30, batch 8850, giga_loss[loss=0.3693, simple_loss=0.4225, pruned_loss=0.1581, over 28285.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.374, pruned_loss=0.1193, over 5689647.42 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3417, pruned_loss=0.08825, over 5710137.11 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3763, pruned_loss=0.1219, over 5677995.13 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:22:43,556 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-15 08:23:06,443 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1329348.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:23:09,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1329351.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:23:14,401 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.286e+03 1.952e+03 2.444e+03 3.209e+03 1.076e+04, threshold=4.888e+03, percent-clipped=12.0
+2023-03-15 08:23:24,344 INFO [train.py:968] (0/2) Epoch 30, batch 8900, libri_loss[loss=0.2697, simple_loss=0.3602, pruned_loss=0.08964, over 29281.00 frames. ], tot_loss[loss=0.307, simple_loss=0.374, pruned_loss=0.12, over 5684753.64 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08849, over 5709945.56 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3761, pruned_loss=0.1224, over 5675186.54 frames. ], batch size: 97, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:23:33,488 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1329380.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:23:40,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8666, 2.9136, 1.9944, 1.1218], device='cuda:0'), covar=tensor([0.9587, 0.3944, 0.4343, 0.8119], device='cuda:0'), in_proj_covar=tensor([0.1865, 0.1755, 0.1675, 0.1520], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 08:23:45,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1329397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:23:46,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9764, 3.8265, 3.6406, 1.9069], device='cuda:0'), covar=tensor([0.0660, 0.0790, 0.0754, 0.2215], device='cuda:0'), in_proj_covar=tensor([0.1342, 0.1239, 0.1039, 0.0770], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 08:24:02,100 INFO [train.py:968] (0/2) Epoch 30, batch 8950, giga_loss[loss=0.3518, simple_loss=0.3982, pruned_loss=0.1528, over 28843.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3732, pruned_loss=0.12, over 5692426.63 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3423, pruned_loss=0.08852, over 5716006.57 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3759, pruned_loss=0.1233, over 5678162.93 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:24:40,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.712e+03 2.186e+03 3.406e+03 1.164e+04, threshold=4.372e+03, percent-clipped=8.0
+2023-03-15 08:24:40,809 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.4909, 5.3462, 5.0959, 2.6841], device='cuda:0'), covar=tensor([0.0451, 0.0548, 0.0626, 0.1602], device='cuda:0'), in_proj_covar=tensor([0.1343, 0.1239, 0.1040, 0.0770], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 08:24:45,317 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8137, 2.1423, 2.0409, 1.6376], device='cuda:0'), covar=tensor([0.3515, 0.2654, 0.2906, 0.3349], device='cuda:0'), in_proj_covar=tensor([0.2101, 0.2077, 0.1982, 0.2121], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 08:24:50,011 INFO [train.py:968] (0/2) Epoch 30, batch 9000, giga_loss[loss=0.2879, simple_loss=0.3661, pruned_loss=0.1049, over 28902.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3708, pruned_loss=0.1192, over 5690627.14 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3419, pruned_loss=0.08821, over 5717877.86 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3738, pruned_loss=0.1225, over 5677279.44 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:24:50,016 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 08:24:58,708 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.2044, simple_loss=0.3128, pruned_loss=0.04804, over 944034.00 frames. 
+2023-03-15 08:24:58,708 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 08:25:00,416 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5253, 2.0077, 1.7246, 1.7938], device='cuda:0'), covar=tensor([0.0772, 0.0287, 0.0312, 0.0847], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 08:25:42,872 INFO [train.py:968] (0/2) Epoch 30, batch 9050, giga_loss[loss=0.3243, simple_loss=0.3803, pruned_loss=0.1341, over 27867.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3697, pruned_loss=0.1199, over 5676988.83 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3419, pruned_loss=0.08821, over 5717373.47 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3722, pruned_loss=0.1226, over 5666860.61 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:25:44,710 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0398, 3.1166, 2.0222, 1.1467], device='cuda:0'), covar=tensor([0.9670, 0.3400, 0.4421, 0.8289], device='cuda:0'), in_proj_covar=tensor([0.1870, 0.1757, 0.1679, 0.1525], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 08:25:59,523 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6162, 1.6217, 1.7771, 1.4028], device='cuda:0'), covar=tensor([0.1920, 0.2608, 0.1542, 0.1834], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0723, 0.0991, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 08:26:01,591 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1329540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:26:04,093 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1329543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:26:05,717 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-15 08:26:15,666 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.261e+03 1.964e+03 2.481e+03 3.787e+03 8.295e+03, threshold=4.963e+03, percent-clipped=18.0
+2023-03-15 08:26:24,107 INFO [train.py:968] (0/2) Epoch 30, batch 9100, giga_loss[loss=0.2912, simple_loss=0.3545, pruned_loss=0.114, over 28649.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3671, pruned_loss=0.1181, over 5681481.31 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.342, pruned_loss=0.08826, over 5720868.76 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.37, pruned_loss=0.1214, over 5668693.05 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:26:27,579 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1329572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:27:06,740 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-15 08:27:11,190 INFO [train.py:968] (0/2) Epoch 30, batch 9150, giga_loss[loss=0.3174, simple_loss=0.3752, pruned_loss=0.1298, over 28868.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3677, pruned_loss=0.1188, over 5683893.47 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3419, pruned_loss=0.08837, over 5724741.57 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3705, pruned_loss=0.1219, over 5669344.27 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:27:44,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.188e+03 2.069e+03 2.604e+03 3.713e+03 7.838e+03, threshold=5.207e+03, percent-clipped=7.0
+2023-03-15 08:27:48,637 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1329664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:27:50,952 INFO [train.py:968] (0/2) Epoch 30, batch 9200, libri_loss[loss=0.257, simple_loss=0.352, pruned_loss=0.08098, over 29524.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3655, pruned_loss=0.1173, over 5684123.65 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3414, pruned_loss=0.08795, over 5722903.51 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3693, pruned_loss=0.1214, over 5671610.35 frames. ], batch size: 89, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:28:33,021 INFO [train.py:968] (0/2) Epoch 30, batch 9250, giga_loss[loss=0.2908, simple_loss=0.3548, pruned_loss=0.1134, over 28770.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3643, pruned_loss=0.1168, over 5678192.97 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3419, pruned_loss=0.08838, over 5717586.16 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3675, pruned_loss=0.1204, over 5671244.11 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:29:01,475 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1329753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:29:06,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.018e+03 1.760e+03 2.434e+03 3.155e+03 9.352e+03, threshold=4.868e+03, percent-clipped=6.0
+2023-03-15 08:29:07,650 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3014, 1.1263, 1.1169, 1.5221], device='cuda:0'), covar=tensor([0.0804, 0.0389, 0.0371, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 08:29:12,246 INFO [train.py:968] (0/2) Epoch 30, batch 9300, giga_loss[loss=0.2977, simple_loss=0.3731, pruned_loss=0.1111, over 28585.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3651, pruned_loss=0.1168, over 5688706.29 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3419, pruned_loss=0.08832, over 5719435.13 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3679, pruned_loss=0.1201, over 5680990.32 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:29:47,496 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1329807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:29:49,459 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1329810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:29:55,621 INFO [train.py:968] (0/2) Epoch 30, batch 9350, giga_loss[loss=0.2818, simple_loss=0.3585, pruned_loss=0.1025, over 28898.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3667, pruned_loss=0.1174, over 5679259.97 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3421, pruned_loss=0.08841, over 5719839.46 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3695, pruned_loss=0.1207, over 5671670.99 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:29:58,697 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1329821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:30:13,185 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1329839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:30:21,032 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1329848.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:30:30,380 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.125e+03 1.846e+03 2.191e+03 2.949e+03 6.857e+03, threshold=4.383e+03, percent-clipped=3.0
+2023-03-15 08:30:36,307 INFO [train.py:968] (0/2) Epoch 30, batch 9400, giga_loss[loss=0.2766, simple_loss=0.3523, pruned_loss=0.1005, over 28891.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3684, pruned_loss=0.119, over 5667888.58 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3421, pruned_loss=0.08841, over 5715028.44 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3715, pruned_loss=0.1228, over 5665447.70 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:31:13,077 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3401, 3.1710, 1.4567, 1.4603], device='cuda:0'), covar=tensor([0.1076, 0.0394, 0.0904, 0.1469], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0581, 0.0416, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 08:31:16,822 INFO [train.py:968] (0/2) Epoch 30, batch 9450, giga_loss[loss=0.3523, simple_loss=0.4151, pruned_loss=0.1447, over 28831.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3689, pruned_loss=0.1192, over 5664969.72 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3424, pruned_loss=0.08873, over 5708947.53 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3716, pruned_loss=0.1226, over 5667370.27 frames. ], batch size: 285, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:31:51,155 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.022e+03 1.697e+03 2.254e+03 3.129e+03 1.205e+04, threshold=4.507e+03, percent-clipped=9.0
+2023-03-15 08:31:58,738 INFO [train.py:968] (0/2) Epoch 30, batch 9500, giga_loss[loss=0.3614, simple_loss=0.3912, pruned_loss=0.1658, over 23422.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3699, pruned_loss=0.1172, over 5671010.80 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3425, pruned_loss=0.08884, over 5711726.14 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3725, pruned_loss=0.1203, over 5669648.39 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:32:25,404 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1330000.pt
+2023-03-15 08:32:32,669 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.67 vs. limit=5.0
+2023-03-15 08:32:37,772 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:32:38,211 INFO [train.py:968] (0/2) Epoch 30, batch 9550, giga_loss[loss=0.2884, simple_loss=0.3675, pruned_loss=0.1047, over 28966.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3713, pruned_loss=0.116, over 5676039.64 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3422, pruned_loss=0.08866, over 5711169.65 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3739, pruned_loss=0.119, over 5675017.53 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:32:54,123 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:33:16,643 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.593e+02 1.741e+03 2.073e+03 3.203e+03 9.031e+03, threshold=4.147e+03, percent-clipped=7.0
+2023-03-15 08:33:22,369 INFO [train.py:968] (0/2) Epoch 30, batch 9600, giga_loss[loss=0.2986, simple_loss=0.366, pruned_loss=0.1155, over 28384.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3757, pruned_loss=0.1193, over 5673905.34 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3424, pruned_loss=0.08878, over 5709496.84 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3779, pruned_loss=0.1218, over 5674113.38 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:33:54,146 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4603, 1.2387, 3.8588, 3.3565], device='cuda:0'), covar=tensor([0.1585, 0.2874, 0.0466, 0.1142], device='cuda:0'), in_proj_covar=tensor([0.0820, 0.0682, 0.1024, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 08:34:05,023 INFO [train.py:968] (0/2) Epoch 30, batch 9650, giga_loss[loss=0.3398, simple_loss=0.4038, pruned_loss=0.138, over 28702.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3757, pruned_loss=0.1204, over 5677216.27 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3422, pruned_loss=0.08865, over 5712236.76 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3781, pruned_loss=0.123, over 5674272.04 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:34:14,463 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 08:34:15,061 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:34:42,474 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.132e+03 1.690e+03 2.256e+03 2.953e+03 6.047e+03, threshold=4.512e+03, percent-clipped=10.0
+2023-03-15 08:34:44,391 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:34:49,201 INFO [train.py:968] (0/2) Epoch 30, batch 9700, giga_loss[loss=0.3549, simple_loss=0.4074, pruned_loss=0.1512, over 28047.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3758, pruned_loss=0.1216, over 5665900.22 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3421, pruned_loss=0.08863, over 5707932.19 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3788, pruned_loss=0.1246, over 5665711.49 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:35:02,314 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:35:15,429 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:35:33,285 INFO [train.py:968] (0/2) Epoch 30, batch 9750, giga_loss[loss=0.3032, simple_loss=0.3846, pruned_loss=0.1109, over 28638.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3743, pruned_loss=0.121, over 5651544.80 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3416, pruned_loss=0.08832, over 5709466.87 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3779, pruned_loss=0.1244, over 5648883.62 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:35:37,749 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330223.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:36:01,532 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:36:06,460 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.106e+03 1.627e+03 2.223e+03 3.640e+03 8.262e+03, threshold=4.446e+03, percent-clipped=10.0
+2023-03-15 08:36:06,786 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8357, 2.1576, 1.5225, 1.7329], device='cuda:0'), covar=tensor([0.1152, 0.0697, 0.1068, 0.1283], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0457, 0.0531, 0.0469], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 08:36:14,446 INFO [train.py:968] (0/2) Epoch 30, batch 9800, giga_loss[loss=0.4058, simple_loss=0.4345, pruned_loss=0.1886, over 26685.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3714, pruned_loss=0.118, over 5650934.60 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3404, pruned_loss=0.08782, over 5706665.07 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3763, pruned_loss=0.1222, over 5649028.87 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:36:15,551 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4777, 1.7638, 1.3829, 1.4887], device='cuda:0'), covar=tensor([0.2946, 0.3036, 0.3568, 0.2611], device='cuda:0'), in_proj_covar=tensor([0.1621, 0.1168, 0.1437, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 08:36:16,877 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:36:20,062 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:36:40,068 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-15 08:36:44,612 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:36:58,018 INFO [train.py:968] (0/2) Epoch 30, batch 9850, giga_loss[loss=0.2726, simple_loss=0.3576, pruned_loss=0.09377, over 28894.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3717, pruned_loss=0.1167, over 5667134.18 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3404, pruned_loss=0.08794, over 5711317.41 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3764, pruned_loss=0.1206, over 5660224.77 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:37:15,108 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:37:15,135 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5972, 1.7182, 1.7816, 1.3677], device='cuda:0'), covar=tensor([0.2020, 0.2701, 0.1676, 0.1951], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0723, 0.0990, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 08:37:18,155 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5286, 2.5978, 1.5850, 1.6197], device='cuda:0'), covar=tensor([0.0779, 0.0358, 0.0700, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0581, 0.0416, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 08:37:18,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:37:33,712 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.657e+03 1.947e+03 2.761e+03 5.685e+03, threshold=3.894e+03, percent-clipped=5.0
+2023-03-15 08:37:41,929 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330366.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:37:42,954 INFO [train.py:968] (0/2) Epoch 30, batch 9900, giga_loss[loss=0.2889, simple_loss=0.3648, pruned_loss=0.1065, over 28650.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3714, pruned_loss=0.1158, over 5675721.03 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.34, pruned_loss=0.08765, over 5716988.22 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3765, pruned_loss=0.1201, over 5663813.08 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:37:43,931 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330369.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:37:46,279 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:38:02,629 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:38:08,300 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330398.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:38:17,788 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:38:25,196 INFO [train.py:968] (0/2) Epoch 30, batch 9950, giga_loss[loss=0.3001, simple_loss=0.3668, pruned_loss=0.1167, over 28720.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3717, pruned_loss=0.1163, over 5671869.50 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3402, pruned_loss=0.08765, over 5718788.25 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3771, pruned_loss=0.1211, over 5658554.45 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:39:06,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.094e+03 1.866e+03 2.313e+03 3.303e+03 7.750e+03, threshold=4.626e+03, percent-clipped=12.0
+2023-03-15 08:39:13,089 INFO [train.py:968] (0/2) Epoch 30, batch 10000, giga_loss[loss=0.3996, simple_loss=0.4282, pruned_loss=0.1855, over 26508.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.373, pruned_loss=0.1179, over 5674338.84 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3404, pruned_loss=0.08775, over 5722566.92 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3778, pruned_loss=0.1222, over 5659510.63 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:39:20,399 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4817, 1.6483, 1.3300, 1.7149], device='cuda:0'), covar=tensor([0.0762, 0.0332, 0.0340, 0.0825], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 08:39:58,418 INFO [train.py:968] (0/2) Epoch 30, batch 10050, giga_loss[loss=0.3197, simple_loss=0.3762, pruned_loss=0.1315, over 27507.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3704, pruned_loss=0.1174, over 5674702.14 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3403, pruned_loss=0.08774, over 5728559.78 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3754, pruned_loss=0.1219, over 5655839.65 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:40:17,245 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:18,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:19,547 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:29,155 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:34,758 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:37,318 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:37,394 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:40,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.950e+03 2.480e+03 3.266e+03 8.992e+03, threshold=4.961e+03, percent-clipped=5.0
+2023-03-15 08:40:46,779 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:47,168 INFO [train.py:968] (0/2) Epoch 30, batch 10100, giga_loss[loss=0.3683, simple_loss=0.4049, pruned_loss=0.1658, over 26693.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3698, pruned_loss=0.1179, over 5674410.79 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3404, pruned_loss=0.08771, over 5731135.74 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3742, pruned_loss=0.1221, over 5656132.10 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:41:02,613 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:41:22,011 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2579, 1.5389, 1.5221, 1.1237], device='cuda:0'), covar=tensor([0.1680, 0.2591, 0.1407, 0.1661], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0725, 0.0993, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 08:41:30,718 INFO [train.py:968] (0/2) Epoch 30, batch 10150, giga_loss[loss=0.3179, simple_loss=0.3778, pruned_loss=0.129, over 28942.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3669, pruned_loss=0.1162, over 5685728.48 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3404, pruned_loss=0.08804, over 5735611.52 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3719, pruned_loss=0.1207, over 5664252.02 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:41:37,136 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3422, 3.0827, 1.4205, 1.6236], device='cuda:0'), covar=tensor([0.1065, 0.0392, 0.0912, 0.1361], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0581, 0.0417, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 08:41:44,264 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:41:45,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0321, 2.2286, 1.8015, 2.4056], device='cuda:0'), covar=tensor([0.2507, 0.2685, 0.3126, 0.2272], device='cuda:0'), in_proj_covar=tensor([0.1621, 0.1167, 0.1436, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 08:42:12,006 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.813e+03 2.176e+03 2.852e+03 6.466e+03, threshold=4.352e+03, percent-clipped=3.0
+2023-03-15 08:42:18,071 INFO [train.py:968] (0/2) Epoch 30, batch 10200, giga_loss[loss=0.3071, simple_loss=0.3769, pruned_loss=0.1186, over 29047.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3658, pruned_loss=0.1165, over 5684437.40 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.34, pruned_loss=0.08779, over 5740629.81 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.371, pruned_loss=0.1212, over 5661255.41 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:42:31,791 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:42:34,536 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:42:41,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1488, 1.2821, 1.1262, 1.0097], device='cuda:0'), covar=tensor([0.1095, 0.0518, 0.1151, 0.1040], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0456, 0.0529, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 08:42:46,934 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:42:49,461 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:42:56,945 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:43:02,230 INFO [train.py:968] (0/2) Epoch 30, batch 10250, libri_loss[loss=0.2499, simple_loss=0.342, pruned_loss=0.07887, over 29656.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3652, pruned_loss=0.1159, over 5687080.00 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3398, pruned_loss=0.08758, over 5745168.91 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5661467.58 frames. ], batch size: 88, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:43:13,400 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:43:31,470 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-15 08:43:32,560 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4628, 1.6307, 1.3924, 1.6030], device='cuda:0'), covar=tensor([0.0795, 0.0342, 0.0353, 0.0907], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0230], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 08:43:39,880 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.087e+03 1.685e+03 2.114e+03 2.939e+03 9.649e+03, threshold=4.227e+03, percent-clipped=8.0
+2023-03-15 08:43:45,157 INFO [train.py:968] (0/2) Epoch 30, batch 10300, giga_loss[loss=0.2585, simple_loss=0.3357, pruned_loss=0.09064, over 28835.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3632, pruned_loss=0.1137, over 5680057.85 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.34, pruned_loss=0.08774, over 5749282.03 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.368, pruned_loss=0.1184, over 5654695.65 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:43:48,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1153, 3.3847, 2.1051, 1.1368], device='cuda:0'), covar=tensor([0.9934, 0.3984, 0.4942, 0.8686], device='cuda:0'), in_proj_covar=tensor([0.1859, 0.1747, 0.1667, 0.1515], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 08:43:49,060 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:43:50,823 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:43:54,922 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2287, 2.5806, 2.5012, 2.0279], device='cuda:0'), covar=tensor([0.3217, 0.2252, 0.2445, 0.2918], device='cuda:0'), in_proj_covar=tensor([0.2098, 0.2066, 0.1980, 0.2115], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 08:44:15,029 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:44:30,388 INFO [train.py:968] (0/2) Epoch 30, batch 10350, giga_loss[loss=0.292, simple_loss=0.3635, pruned_loss=0.1103, over 28737.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.36, pruned_loss=0.1108, over 5683194.58 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3398, pruned_loss=0.08761, over 5755051.61 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3647, pruned_loss=0.1156, over 5655226.78 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:45:11,606 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.590e+03 1.925e+03 3.044e+03 8.092e+03, threshold=3.849e+03, percent-clipped=6.0
+2023-03-15 08:45:16,110 INFO [train.py:968] (0/2) Epoch 30, batch 10400, giga_loss[loss=0.3456, simple_loss=0.399, pruned_loss=0.146, over 27825.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.361, pruned_loss=0.1112, over 5681640.29 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3405, pruned_loss=0.08785, over 5756199.58 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3648, pruned_loss=0.1155, over 5655979.98 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:45:41,647 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:46:00,827 INFO [train.py:968] (0/2) Epoch 30, batch 10450, giga_loss[loss=0.3257, simple_loss=0.3659, pruned_loss=0.1428, over 23473.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3583, pruned_loss=0.1101, over 5677129.85 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3403, pruned_loss=0.08771, over 5752032.28 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3624, pruned_loss=0.1147, over 5656321.70 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:46:16,449 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:46:18,953 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4710, 3.6107, 1.6621, 1.6329], device='cuda:0'), covar=tensor([0.1058, 0.0362, 0.0882, 0.1405], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0581, 0.0417, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 08:46:43,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.266e+03 1.902e+03 2.492e+03 3.434e+03 1.648e+04, threshold=4.984e+03, percent-clipped=19.0
+2023-03-15 08:46:46,931 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9305, 1.1259, 1.0954, 0.8704], device='cuda:0'), covar=tensor([0.2446, 0.2785, 0.1862, 0.2416], device='cuda:0'), in_proj_covar=tensor([0.2103, 0.2070, 0.1984, 0.2121], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 08:46:48,853 INFO [train.py:968] (0/2) Epoch 30, batch 10500, giga_loss[loss=0.263, simple_loss=0.3393, pruned_loss=0.0933, over 28979.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3572, pruned_loss=0.1104, over 5679113.30 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3402, pruned_loss=0.08764, over 5754191.76 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3607, pruned_loss=0.1143, over 5659779.70 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:46:58,321 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-15 08:47:01,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4791, 1.8099, 1.6312, 1.4647], device='cuda:0'), covar=tensor([0.2181, 0.2079, 0.2423, 0.2280], device='cuda:0'), in_proj_covar=tensor([0.0522, 0.0771, 0.0743, 0.0708], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 08:47:15,359 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:47:34,063 INFO [train.py:968] (0/2) Epoch 30, batch 10550, libri_loss[loss=0.3024, simple_loss=0.3776, pruned_loss=0.1135, over 25365.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3602, pruned_loss=0.1118, over 5677847.88 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3399, pruned_loss=0.08745, over 5753963.81 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1157, over 5661399.02 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:48:12,696 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.237e+03 1.665e+03 2.229e+03 2.998e+03 1.055e+04, threshold=4.459e+03, percent-clipped=9.0
+2023-03-15 08:48:19,216 INFO [train.py:968] (0/2) Epoch 30, batch 10600, giga_loss[loss=0.3078, simple_loss=0.3801, pruned_loss=0.1177, over 28921.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3626, pruned_loss=0.1127, over 5679209.67 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3404, pruned_loss=0.08759, over 5754474.84 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3655, pruned_loss=0.1162, over 5663987.33 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:48:25,627 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1331075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:48:27,388 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1331078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:48:56,434 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1331107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:49:07,110 INFO [train.py:968] (0/2) Epoch 30, batch 10650, giga_loss[loss=0.3382, simple_loss=0.3853, pruned_loss=0.1455, over 27653.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3638, pruned_loss=0.1141, over 5651073.82 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3407, pruned_loss=0.08798, over 5756442.18 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3664, pruned_loss=0.1174, over 5634209.43 frames. ], batch size: 474, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:49:26,764 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4010, 1.9177, 1.3539, 0.6744], device='cuda:0'), covar=tensor([0.6401, 0.2993, 0.4010, 0.7134], device='cuda:0'), in_proj_covar=tensor([0.1856, 0.1743, 0.1665, 0.1515], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 08:49:47,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5279, 1.7458, 1.4477, 1.7536], device='cuda:0'), covar=tensor([0.2705, 0.2681, 0.2884, 0.2455], device='cuda:0'), in_proj_covar=tensor([0.1624, 0.1171, 0.1439, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 08:49:47,511 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.350e+02 1.654e+03 2.068e+03 2.905e+03 1.161e+04, threshold=4.135e+03, percent-clipped=6.0
+2023-03-15 08:49:53,489 INFO [train.py:968] (0/2) Epoch 30, batch 10700, giga_loss[loss=0.2795, simple_loss=0.3519, pruned_loss=0.1036, over 28905.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.365, pruned_loss=0.1156, over 5650211.32 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3409, pruned_loss=0.08806, over 5759497.71 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3673, pruned_loss=0.1186, over 5632229.55 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:49:57,506 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.11 vs. limit=5.0
+2023-03-15 08:50:36,610 INFO [train.py:968] (0/2) Epoch 30, batch 10750, giga_loss[loss=0.2939, simple_loss=0.3689, pruned_loss=0.1095, over 28877.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3654, pruned_loss=0.1161, over 5653988.84 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3404, pruned_loss=0.08767, over 5761248.32 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3686, pruned_loss=0.1198, over 5634292.91 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:50:45,287 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-15 08:50:58,380 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8336, 1.7719, 1.9687, 1.5673], device='cuda:0'), covar=tensor([0.1962, 0.2568, 0.1583, 0.1821], device='cuda:0'), in_proj_covar=tensor([0.0941, 0.0724, 0.0993, 0.0891], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 08:51:20,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.150e+03 1.797e+03 2.342e+03 2.841e+03 6.088e+03, threshold=4.683e+03, percent-clipped=8.0
+2023-03-15 08:51:26,022 INFO [train.py:968] (0/2) Epoch 30, batch 10800, libri_loss[loss=0.2689, simple_loss=0.3453, pruned_loss=0.09629, over 29544.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3673, pruned_loss=0.1168, over 5661472.72 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3405, pruned_loss=0.08765, over 5762802.51 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3708, pruned_loss=0.1209, over 5640064.19 frames. ], batch size: 77, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:51:28,424 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1331271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:52:09,809 INFO [train.py:968] (0/2) Epoch 30, batch 10850, giga_loss[loss=0.2926, simple_loss=0.3618, pruned_loss=0.1117, over 28854.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3684, pruned_loss=0.1177, over 5652841.41 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3407, pruned_loss=0.08773, over 5751400.59 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3719, pruned_loss=0.1218, over 5642782.22 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:52:21,294 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5784, 3.1071, 1.6464, 1.6709], device='cuda:0'), covar=tensor([0.0802, 0.0347, 0.0684, 0.1084], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0579, 0.0416, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 08:52:51,139 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.211e+03 1.774e+03 2.452e+03 3.203e+03 6.032e+03, threshold=4.904e+03, percent-clipped=6.0
+2023-03-15 08:52:52,547 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 08:52:56,548 INFO [train.py:968] (0/2) Epoch 30, batch 10900, giga_loss[loss=0.2901, simple_loss=0.3567, pruned_loss=0.1118, over 28927.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.37, pruned_loss=0.1192, over 5655115.75 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3405, pruned_loss=0.08762, over 5753818.77 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3733, pruned_loss=0.123, over 5643752.10 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:53:01,964 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1331373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:53:18,515 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0277, 2.0966, 2.1631, 1.7734], device='cuda:0'), covar=tensor([0.1863, 0.2417, 0.1497, 0.1745], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0725, 0.0994, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 08:53:42,711 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1331414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:53:44,939 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1331417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:53:45,318 INFO [train.py:968] (0/2) Epoch 30, batch 10950, giga_loss[loss=0.2919, simple_loss=0.3678, pruned_loss=0.108, over 28742.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3713, pruned_loss=0.1203, over 5655083.49 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3405, pruned_loss=0.0876, over 5756357.88 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3743, pruned_loss=0.1238, over 5642269.67 frames. ], batch size: 243, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:53:46,433 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2336, 3.2362, 1.4057, 1.4518], device='cuda:0'), covar=tensor([0.1085, 0.0489, 0.0957, 0.1435], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0580, 0.0416, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 08:54:14,621 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1331446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:54:29,036 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:54:30,134 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.198e+03 1.942e+03 2.476e+03 3.495e+03 8.052e+03, threshold=4.952e+03, percent-clipped=13.0
+2023-03-15 08:54:35,037 INFO [train.py:968] (0/2) Epoch 30, batch 11000, giga_loss[loss=0.288, simple_loss=0.3582, pruned_loss=0.1088, over 28924.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3713, pruned_loss=0.119, over 5655930.05 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3408, pruned_loss=0.08786, over 5759350.74 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.374, pruned_loss=0.1221, over 5641376.24 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:54:42,601 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:55:17,263 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1331516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:55:18,452 INFO [train.py:968] (0/2) Epoch 30, batch 11050, giga_loss[loss=0.2796, simple_loss=0.3494, pruned_loss=0.1049, over 28924.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3704, pruned_loss=0.119, over 5657932.21 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3408, pruned_loss=0.08805, over 5763354.39 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3741, pruned_loss=0.1228, over 5637878.66 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:55:20,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1331519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:55:27,577 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2938, 2.3262, 2.0168, 1.8661], device='cuda:0'), covar=tensor([0.1022, 0.0709, 0.0921, 0.1182], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0454, 0.0527, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 08:55:48,165 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1331548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:55:55,306 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4117, 1.5205, 1.4199, 1.6945], device='cuda:0'), covar=tensor([0.0783, 0.0348, 0.0331, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 08:56:05,699 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.308e+03 1.872e+03 2.462e+03 3.286e+03 9.992e+03, threshold=4.924e+03, percent-clipped=12.0
+2023-03-15 08:56:12,146 INFO [train.py:968] (0/2) Epoch 30, batch 11100, giga_loss[loss=0.2784, simple_loss=0.3537, pruned_loss=0.1015, over 28793.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3695, pruned_loss=0.1192, over 5669220.72 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3409, pruned_loss=0.0881, over 5764828.16 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3725, pruned_loss=0.1224, over 5651406.78 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:56:21,015 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-15 08:57:05,416 INFO [train.py:968] (0/2) Epoch 30, batch 11150, giga_loss[loss=0.2865, simple_loss=0.3554, pruned_loss=0.1088, over 28771.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3692, pruned_loss=0.1196, over 5662688.54 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3416, pruned_loss=0.08848, over 5766612.94 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3722, pruned_loss=0.1231, over 5643121.59 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:57:33,842 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-15 08:57:48,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.252e+03 1.881e+03 2.688e+03 3.523e+03 8.691e+03, threshold=5.376e+03, percent-clipped=8.0
+2023-03-15 08:57:53,609 INFO [train.py:968] (0/2) Epoch 30, batch 11200, giga_loss[loss=0.2645, simple_loss=0.3331, pruned_loss=0.09795, over 28817.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3682, pruned_loss=0.1192, over 5674354.09 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3418, pruned_loss=0.08863, over 5767134.87 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3705, pruned_loss=0.122, over 5658206.73 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:58:07,071 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:58:38,153 INFO [train.py:968] (0/2) Epoch 30, batch 11250, giga_loss[loss=0.2854, simple_loss=0.3533, pruned_loss=0.1088, over 28694.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3691, pruned_loss=0.1204, over 5665934.81 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3419, pruned_loss=0.08868, over 5758835.32 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3709, pruned_loss=0.1227, over 5659990.80 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:58:51,546 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3360, 1.9433, 1.4358, 0.5710], device='cuda:0'), covar=tensor([0.6460, 0.3336, 0.4751, 0.7811], device='cuda:0'), in_proj_covar=tensor([0.1862, 0.1745, 0.1666, 0.1518], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 08:59:21,689 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.225e+03 1.810e+03 2.263e+03 3.195e+03 6.594e+03, threshold=4.526e+03, percent-clipped=3.0
+2023-03-15 08:59:25,577 INFO [train.py:968] (0/2) Epoch 30, batch 11300, giga_loss[loss=0.3078, simple_loss=0.3783, pruned_loss=0.1186, over 28974.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3696, pruned_loss=0.121, over 5669057.92 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08905, over 5761293.80 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.371, pruned_loss=0.1231, over 5660311.46 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:59:30,390 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4640, 1.3525, 1.6103, 1.2593], device='cuda:0'), covar=tensor([0.1312, 0.2498, 0.1135, 0.1335], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0724, 0.0992, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 08:59:53,203 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:59:53,312 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5629, 1.9645, 1.5299, 1.6015], device='cuda:0'), covar=tensor([0.2806, 0.2851, 0.3343, 0.2637], device='cuda:0'), in_proj_covar=tensor([0.1622, 0.1171, 0.1436, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 08:59:58,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6307, 1.8048, 1.7337, 1.5819], device='cuda:0'), covar=tensor([0.2116, 0.2199, 0.2402, 0.2359], device='cuda:0'), in_proj_covar=tensor([0.0518, 0.0767, 0.0739, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 09:00:13,042 INFO [train.py:968] (0/2) Epoch 30, batch 11350, giga_loss[loss=0.3303, simple_loss=0.3831, pruned_loss=0.1388, over 28948.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3697, pruned_loss=0.1211, over 5670557.03 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3422, pruned_loss=0.08883, over 5761754.15 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3714, pruned_loss=0.1234, over 5661896.51 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:00:23,633 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 09:00:27,399 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:00:32,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1331836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:00:35,371 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1446, 1.2758, 1.1016, 0.8819], device='cuda:0'), covar=tensor([0.1015, 0.0444, 0.1001, 0.1046], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0455, 0.0528, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 09:00:44,176 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1331850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:00:50,840 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8457, 2.8101, 1.8241, 1.0036], device='cuda:0'), covar=tensor([0.8965, 0.3774, 0.4290, 0.7812], device='cuda:0'), in_proj_covar=tensor([0.1861, 0.1745, 0.1665, 0.1517], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 09:00:56,655 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.271e+03 2.014e+03 2.677e+03 3.519e+03 1.608e+04, threshold=5.355e+03, percent-clipped=14.0
+2023-03-15 09:00:59,175 INFO [train.py:968] (0/2) Epoch 30, batch 11400, giga_loss[loss=0.2912, simple_loss=0.3644, pruned_loss=0.109, over 28749.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3712, pruned_loss=0.1224, over 5670332.86 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.08847, over 5763801.56 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3734, pruned_loss=0.1249, over 5660527.86 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:01:38,557 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-15 09:01:49,646 INFO [train.py:968] (0/2) Epoch 30, batch 11450, giga_loss[loss=0.2976, simple_loss=0.3621, pruned_loss=0.1165, over 28953.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.37, pruned_loss=0.1214, over 5669217.68 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3413, pruned_loss=0.08831, over 5765925.97 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3725, pruned_loss=0.1241, over 5658372.20 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:02:36,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.843e+03 2.273e+03 2.953e+03 5.578e+03, threshold=4.546e+03, percent-clipped=2.0
+2023-03-15 09:02:38,502 INFO [train.py:968] (0/2) Epoch 30, batch 11500, giga_loss[loss=0.303, simple_loss=0.3706, pruned_loss=0.1177, over 28827.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3702, pruned_loss=0.1223, over 5659943.49 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3413, pruned_loss=0.08811, over 5766374.13 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.373, pruned_loss=0.1256, over 5647761.32 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:02:43,822 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:02:49,304 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1331979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:02:51,742 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1331982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:02,809 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1331993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:04,666 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1331996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:07,580 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1332000.pt
+2023-03-15 09:03:18,001 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:24,054 INFO [train.py:968] (0/2) Epoch 30, batch 11550, libri_loss[loss=0.2942, simple_loss=0.366, pruned_loss=0.1112, over 29565.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3698, pruned_loss=0.1218, over 5656860.46 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08853, over 5760046.82 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3723, pruned_loss=0.1249, over 5650271.89 frames. ], batch size: 75, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:03:28,593 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:29,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1332026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:57,756 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:04:03,803 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0468, 2.2833, 2.2603, 2.0016], device='cuda:0'), covar=tensor([0.2606, 0.2205, 0.2207, 0.2294], device='cuda:0'), in_proj_covar=tensor([0.2117, 0.2087, 0.1998, 0.2137], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 09:04:05,558 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.303e+03 1.831e+03 2.426e+03 3.352e+03 6.959e+03, threshold=4.853e+03, percent-clipped=9.0
+2023-03-15 09:04:08,191 INFO [train.py:968] (0/2) Epoch 30, batch 11600, giga_loss[loss=0.4201, simple_loss=0.4446, pruned_loss=0.1978, over 27463.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1206, over 5660856.77 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3417, pruned_loss=0.08877, over 5754837.52 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3722, pruned_loss=0.124, over 5657543.88 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:04:45,081 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1332106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:04:56,597 INFO [train.py:968] (0/2) Epoch 30, batch 11650, giga_loss[loss=0.2891, simple_loss=0.3588, pruned_loss=0.1097, over 28653.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3698, pruned_loss=0.1206, over 5660826.99 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3416, pruned_loss=0.08867, over 5755634.16 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3726, pruned_loss=0.1237, over 5656669.99 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:05:22,471 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1127, 1.1099, 1.1582, 1.3662], device='cuda:0'), covar=tensor([0.0812, 0.0359, 0.0306, 0.1001], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 09:05:34,589 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 09:05:43,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.121e+03 1.771e+03 2.150e+03 2.990e+03 5.921e+03, threshold=4.299e+03, percent-clipped=3.0
+2023-03-15 09:05:46,630 INFO [train.py:968] (0/2) Epoch 30, batch 11700, giga_loss[loss=0.2891, simple_loss=0.3573, pruned_loss=0.1104, over 28704.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3697, pruned_loss=0.1201, over 5679093.29 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3413, pruned_loss=0.08843, over 5758621.33 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.373, pruned_loss=0.1237, over 5670697.30 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:05:47,432 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:15,444 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:17,339 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:21,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:21,326 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 09:06:32,255 INFO [train.py:968] (0/2) Epoch 30, batch 11750, giga_loss[loss=0.2708, simple_loss=0.3462, pruned_loss=0.09776, over 29036.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3727, pruned_loss=0.1228, over 5676338.53 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3413, pruned_loss=0.08834, over 5761244.91 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3764, pruned_loss=0.1268, over 5664652.94 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:06:33,757 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1332220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:41,655 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:07:10,897 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.201e+03 1.736e+03 2.224e+03 3.257e+03 7.475e+03, threshold=4.448e+03, percent-clipped=9.0
+2023-03-15 09:07:12,925 INFO [train.py:968] (0/2) Epoch 30, batch 11800, giga_loss[loss=0.3095, simple_loss=0.3747, pruned_loss=0.1222, over 28711.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3716, pruned_loss=0.1214, over 5691430.72 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3414, pruned_loss=0.08837, over 5764996.70 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.376, pruned_loss=0.1262, over 5675006.18 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:07:53,022 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:07:56,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:07:58,618 INFO [train.py:968] (0/2) Epoch 30, batch 11850, giga_loss[loss=0.329, simple_loss=0.3951, pruned_loss=0.1315, over 28904.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3722, pruned_loss=0.1211, over 5683274.21 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3412, pruned_loss=0.08827, over 5756880.80 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3765, pruned_loss=0.1258, over 5675067.98 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:07:58,950 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9183, 2.1361, 2.2952, 1.9073], device='cuda:0'), covar=tensor([0.2572, 0.2285, 0.2222, 0.2387], device='cuda:0'), in_proj_covar=tensor([0.2109, 0.2078, 0.1985, 0.2126], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 09:08:23,444 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:08:26,427 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:08:27,665 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:08:30,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:08:43,019 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.623e+03 2.121e+03 2.608e+03 6.869e+03, threshold=4.242e+03, percent-clipped=5.0
+2023-03-15 09:08:47,758 INFO [train.py:968] (0/2) Epoch 30, batch 11900, libri_loss[loss=0.2544, simple_loss=0.3427, pruned_loss=0.08303, over 28532.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3704, pruned_loss=0.1191, over 5670461.81 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.08804, over 5755812.59 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3748, pruned_loss=0.1238, over 5662816.58 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:08:59,574 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:09:14,847 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3681, 1.4114, 3.9508, 3.4180], device='cuda:0'), covar=tensor([0.1681, 0.2768, 0.0496, 0.0792], device='cuda:0'), in_proj_covar=tensor([0.0820, 0.0683, 0.1024, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 09:09:14,858 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:09:27,938 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7777, 1.9228, 1.5627, 1.4796], device='cuda:0'), covar=tensor([0.1037, 0.0660, 0.0976, 0.1182], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0454, 0.0527, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 09:09:31,060 INFO [train.py:968] (0/2) Epoch 30, batch 11950, giga_loss[loss=0.2739, simple_loss=0.3491, pruned_loss=0.09939, over 28900.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3682, pruned_loss=0.1173, over 5679876.83 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.08802, over 5760380.70 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3729, pruned_loss=0.1221, over 5666935.98 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:10:12,388 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.136e+03 1.712e+03 2.212e+03 2.958e+03 8.389e+03, threshold=4.424e+03, percent-clipped=6.0
+2023-03-15 09:10:15,061 INFO [train.py:968] (0/2) Epoch 30, batch 12000, giga_loss[loss=0.329, simple_loss=0.3892, pruned_loss=0.1345, over 28899.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3667, pruned_loss=0.1165, over 5693064.84 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3403, pruned_loss=0.08786, over 5764751.08 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3716, pruned_loss=0.1213, over 5676583.86 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:10:15,066 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 09:10:24,610 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.2036, simple_loss=0.3112, pruned_loss=0.04797, over 944034.00 frames. 
+2023-03-15 09:10:24,611 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 09:10:37,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:10:48,673 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:10:50,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:11:12,890 INFO [train.py:968] (0/2) Epoch 30, batch 12050, giga_loss[loss=0.277, simple_loss=0.3444, pruned_loss=0.1048, over 28609.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3684, pruned_loss=0.1183, over 5663846.85 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3403, pruned_loss=0.08784, over 5757037.99 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3725, pruned_loss=0.1225, over 5656373.57 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:11:17,425 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:11:37,670 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:11:40,029 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:11:55,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.277e+03 1.873e+03 2.358e+03 3.266e+03 8.366e+03, threshold=4.717e+03, percent-clipped=7.0
+2023-03-15 09:11:57,317 INFO [train.py:968] (0/2) Epoch 30, batch 12100, giga_loss[loss=0.3039, simple_loss=0.3722, pruned_loss=0.1178, over 29102.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3691, pruned_loss=0.1178, over 5675501.92 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3406, pruned_loss=0.08792, over 5758761.10 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3732, pruned_loss=0.1221, over 5665221.17 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:12:06,165 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:12:21,940 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:12:43,119 INFO [train.py:968] (0/2) Epoch 30, batch 12150, giga_loss[loss=0.3387, simple_loss=0.3956, pruned_loss=0.1408, over 28877.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3689, pruned_loss=0.1188, over 5674249.84 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3405, pruned_loss=0.0879, over 5761508.35 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3734, pruned_loss=0.1235, over 5660321.72 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:12:49,680 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:12:53,047 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:13:01,745 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6814, 1.8632, 1.7495, 1.7301], device='cuda:0'), covar=tensor([0.2156, 0.2353, 0.2544, 0.2189], device='cuda:0'), in_proj_covar=tensor([0.0520, 0.0770, 0.0740, 0.0707], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 09:13:20,863 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:13:29,270 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.780e+03 2.053e+03 2.921e+03 8.783e+03, threshold=4.106e+03, percent-clipped=7.0
+2023-03-15 09:13:31,391 INFO [train.py:968] (0/2) Epoch 30, batch 12200, giga_loss[loss=0.3644, simple_loss=0.4008, pruned_loss=0.164, over 26474.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3696, pruned_loss=0.1201, over 5669072.15 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.08801, over 5762102.80 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3734, pruned_loss=0.1241, over 5656260.57 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:14:07,098 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3200, 3.2683, 1.3796, 1.5358], device='cuda:0'), covar=tensor([0.1057, 0.0396, 0.0959, 0.1380], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0579, 0.0415, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 09:14:21,853 INFO [train.py:968] (0/2) Epoch 30, batch 12250, giga_loss[loss=0.308, simple_loss=0.3831, pruned_loss=0.1165, over 28870.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3704, pruned_loss=0.1203, over 5676691.28 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3409, pruned_loss=0.08795, over 5765785.24 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.374, pruned_loss=0.1245, over 5660221.76 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:14:40,415 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:14:43,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:15:03,723 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.771e+03 2.304e+03 3.269e+03 9.682e+03, threshold=4.608e+03, percent-clipped=14.0
+2023-03-15 09:15:06,113 INFO [train.py:968] (0/2) Epoch 30, batch 12300, giga_loss[loss=0.3175, simple_loss=0.3822, pruned_loss=0.1264, over 29110.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3721, pruned_loss=0.1211, over 5669271.63 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3412, pruned_loss=0.08793, over 5760985.32 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3759, pruned_loss=0.1257, over 5656920.45 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:15:07,830 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:15:51,607 INFO [train.py:968] (0/2) Epoch 30, batch 12350, giga_loss[loss=0.2849, simple_loss=0.3547, pruned_loss=0.1075, over 29009.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3727, pruned_loss=0.1222, over 5654596.46 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3414, pruned_loss=0.08808, over 5757760.57 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3762, pruned_loss=0.1265, over 5645881.81 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:16:31,927 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 09:16:39,473 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.147e+03 1.724e+03 2.234e+03 2.957e+03 7.885e+03, threshold=4.467e+03, percent-clipped=4.0
+2023-03-15 09:16:40,784 INFO [train.py:968] (0/2) Epoch 30, batch 12400, giga_loss[loss=0.4337, simple_loss=0.444, pruned_loss=0.2117, over 26661.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3733, pruned_loss=0.123, over 5647083.89 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3419, pruned_loss=0.08833, over 5757105.35 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3763, pruned_loss=0.1268, over 5638550.76 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:17:23,280 INFO [train.py:968] (0/2) Epoch 30, batch 12450, giga_loss[loss=0.2858, simple_loss=0.3599, pruned_loss=0.1058, over 28831.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3724, pruned_loss=0.1215, over 5653987.93 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3416, pruned_loss=0.08804, over 5761227.54 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3758, pruned_loss=0.1258, over 5640702.96 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:18:08,734 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.157e+03 1.766e+03 2.260e+03 3.178e+03 9.507e+03, threshold=4.521e+03, percent-clipped=9.0
+2023-03-15 09:18:09,452 INFO [train.py:968] (0/2) Epoch 30, batch 12500, libri_loss[loss=0.2487, simple_loss=0.336, pruned_loss=0.08068, over 29193.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3703, pruned_loss=0.1197, over 5658072.11 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3415, pruned_loss=0.0879, over 5762961.57 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3743, pruned_loss=0.1245, over 5641686.42 frames. ], batch size: 97, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:18:55,419 INFO [train.py:968] (0/2) Epoch 30, batch 12550, giga_loss[loss=0.3365, simple_loss=0.374, pruned_loss=0.1495, over 23538.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3698, pruned_loss=0.1201, over 5664272.07 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3416, pruned_loss=0.08797, over 5766068.53 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3735, pruned_loss=0.1245, over 5646713.37 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:19:23,341 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 09:19:41,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.616e+02 1.938e+03 2.328e+03 3.591e+03 1.020e+04, threshold=4.655e+03, percent-clipped=17.0
+2023-03-15 09:19:41,483 INFO [train.py:968] (0/2) Epoch 30, batch 12600, giga_loss[loss=0.2761, simple_loss=0.3492, pruned_loss=0.1015, over 28917.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3657, pruned_loss=0.1171, over 5671632.51 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3415, pruned_loss=0.08782, over 5760297.88 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3696, pruned_loss=0.1217, over 5658881.66 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:19:54,725 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3816, 1.7022, 1.3985, 1.6495], device='cuda:0'), covar=tensor([0.0769, 0.0307, 0.0332, 0.0860], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0122, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 09:20:29,577 INFO [train.py:968] (0/2) Epoch 30, batch 12650, libri_loss[loss=0.2227, simple_loss=0.3095, pruned_loss=0.06793, over 29572.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.362, pruned_loss=0.1154, over 5664229.54 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3414, pruned_loss=0.08774, over 5760332.18 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3657, pruned_loss=0.1199, over 5651606.24 frames. ], batch size: 75, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:21:11,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-15 09:21:13,417 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.736e+03 2.582e+03 3.107e+03 7.246e+03, threshold=5.165e+03, percent-clipped=5.0
+2023-03-15 09:21:13,429 INFO [train.py:968] (0/2) Epoch 30, batch 12700, giga_loss[loss=0.2417, simple_loss=0.3198, pruned_loss=0.08182, over 28640.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3605, pruned_loss=0.1156, over 5658491.80 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3411, pruned_loss=0.08759, over 5760949.68 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.364, pruned_loss=0.1198, over 5646138.64 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:21:19,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5218, 2.1599, 1.5800, 0.7563], device='cuda:0'), covar=tensor([0.6960, 0.3643, 0.4656, 0.7425], device='cuda:0'), in_proj_covar=tensor([0.1876, 0.1761, 0.1680, 0.1526], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 09:21:59,740 INFO [train.py:968] (0/2) Epoch 30, batch 12750, giga_loss[loss=0.2519, simple_loss=0.3294, pruned_loss=0.08721, over 28910.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.36, pruned_loss=0.1158, over 5654124.61 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3417, pruned_loss=0.08781, over 5760824.49 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3628, pruned_loss=0.1195, over 5642548.28 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:22:52,113 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.300e+03 1.832e+03 2.299e+03 2.771e+03 8.154e+03, threshold=4.597e+03, percent-clipped=5.0
+2023-03-15 09:22:52,125 INFO [train.py:968] (0/2) Epoch 30, batch 12800, giga_loss[loss=0.2566, simple_loss=0.335, pruned_loss=0.08913, over 28956.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3598, pruned_loss=0.1152, over 5648376.81 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3417, pruned_loss=0.08777, over 5754255.72 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3623, pruned_loss=0.1187, over 5642532.71 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:23:38,907 INFO [train.py:968] (0/2) Epoch 30, batch 12850, giga_loss[loss=0.2535, simple_loss=0.3295, pruned_loss=0.08871, over 28600.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3588, pruned_loss=0.1124, over 5649196.11 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3415, pruned_loss=0.08773, over 5754172.26 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3613, pruned_loss=0.1158, over 5642418.17 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:24:26,479 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.048e+03 1.733e+03 2.365e+03 3.253e+03 7.791e+03, threshold=4.730e+03, percent-clipped=8.0
+2023-03-15 09:24:26,491 INFO [train.py:968] (0/2) Epoch 30, batch 12900, libri_loss[loss=0.2918, simple_loss=0.3731, pruned_loss=0.1052, over 28647.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3568, pruned_loss=0.1093, over 5641074.67 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3417, pruned_loss=0.08811, over 5748592.22 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5637175.79 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:25:10,821 INFO [train.py:968] (0/2) Epoch 30, batch 12950, giga_loss[loss=0.2874, simple_loss=0.3698, pruned_loss=0.1025, over 28715.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3541, pruned_loss=0.1059, over 5648638.06 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3417, pruned_loss=0.08827, over 5753976.64 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3567, pruned_loss=0.1093, over 5635319.91 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:25:58,632 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.447e+02 1.450e+03 1.841e+03 2.627e+03 6.400e+03, threshold=3.682e+03, percent-clipped=4.0
+2023-03-15 09:25:58,644 INFO [train.py:968] (0/2) Epoch 30, batch 13000, giga_loss[loss=0.263, simple_loss=0.3437, pruned_loss=0.09114, over 28389.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3509, pruned_loss=0.1031, over 5655584.94 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3411, pruned_loss=0.08799, over 5759996.99 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3541, pruned_loss=0.1067, over 5635592.20 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:26:45,514 INFO [train.py:968] (0/2) Epoch 30, batch 13050, giga_loss[loss=0.2844, simple_loss=0.3704, pruned_loss=0.09916, over 28830.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3494, pruned_loss=0.1002, over 5656993.52 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3412, pruned_loss=0.08804, over 5763355.60 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.352, pruned_loss=0.1033, over 5635704.27 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:26:51,111 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.7948, 1.2371, 1.3429, 0.9810], device='cuda:0'), covar=tensor([0.2488, 0.1513, 0.2351, 0.1919], device='cuda:0'), in_proj_covar=tensor([0.0517, 0.0763, 0.0735, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 09:26:51,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9317, 1.2049, 1.1347, 0.9097], device='cuda:0'), covar=tensor([0.2557, 0.2687, 0.1742, 0.2398], device='cuda:0'), in_proj_covar=tensor([0.2084, 0.2061, 0.1966, 0.2108], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 09:27:33,117 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.052e+02 1.471e+03 1.846e+03 2.744e+03 7.353e+03, threshold=3.691e+03, percent-clipped=7.0
+2023-03-15 09:27:33,130 INFO [train.py:968] (0/2) Epoch 30, batch 13100, giga_loss[loss=0.2952, simple_loss=0.3478, pruned_loss=0.1213, over 24091.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3495, pruned_loss=0.09874, over 5668176.70 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.341, pruned_loss=0.08807, over 5766010.40 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3519, pruned_loss=0.1015, over 5646777.35 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:28:24,297 INFO [train.py:968] (0/2) Epoch 30, batch 13150, giga_loss[loss=0.2684, simple_loss=0.3474, pruned_loss=0.09466, over 28771.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3501, pruned_loss=0.09928, over 5661523.01 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.08804, over 5767564.67 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3525, pruned_loss=0.1017, over 5641195.83 frames. ], batch size: 243, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:28:38,200 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1333635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:29:08,942 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.565e+03 1.978e+03 2.513e+03 4.912e+03, threshold=3.955e+03, percent-clipped=4.0
+2023-03-15 09:29:08,954 INFO [train.py:968] (0/2) Epoch 30, batch 13200, giga_loss[loss=0.2408, simple_loss=0.3249, pruned_loss=0.07836, over 28617.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.348, pruned_loss=0.09764, over 5654488.14 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.08812, over 5754590.10 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3503, pruned_loss=0.09993, over 5645860.87 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:29:18,290 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3752, 4.2359, 3.9887, 1.8911], device='cuda:0'), covar=tensor([0.0584, 0.0718, 0.0809, 0.2117], device='cuda:0'), in_proj_covar=tensor([0.1332, 0.1228, 0.1030, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 09:29:45,760 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9792, 1.1795, 1.1424, 0.9834], device='cuda:0'), covar=tensor([0.2299, 0.2562, 0.1517, 0.2023], device='cuda:0'), in_proj_covar=tensor([0.2073, 0.2049, 0.1954, 0.2095], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 09:29:53,262 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3078, 3.0951, 1.4877, 1.5240], device='cuda:0'), covar=tensor([0.1018, 0.0381, 0.0959, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0578, 0.0417, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 09:29:56,714 INFO [train.py:968] (0/2) Epoch 30, batch 13250, giga_loss[loss=0.3074, simple_loss=0.3664, pruned_loss=0.1242, over 27847.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3456, pruned_loss=0.09673, over 5645343.83 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3407, pruned_loss=0.08833, over 5759098.30 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3476, pruned_loss=0.09863, over 5631592.39 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:30:39,689 INFO [train.py:968] (0/2) Epoch 30, batch 13300, giga_loss[loss=0.3081, simple_loss=0.3827, pruned_loss=0.1167, over 28639.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3455, pruned_loss=0.09689, over 5629376.77 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3402, pruned_loss=0.08834, over 5742210.39 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3478, pruned_loss=0.09875, over 5630377.25 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:30:41,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.030e+03 1.579e+03 1.936e+03 2.860e+03 1.008e+04, threshold=3.872e+03, percent-clipped=8.0
+2023-03-15 09:31:26,062 INFO [train.py:968] (0/2) Epoch 30, batch 13350, giga_loss[loss=0.2347, simple_loss=0.3228, pruned_loss=0.07335, over 28401.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3442, pruned_loss=0.09609, over 5639536.04 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3392, pruned_loss=0.08817, over 5747287.51 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.347, pruned_loss=0.09795, over 5633004.82 frames. ], batch size: 65, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:31:59,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3296, 1.6393, 1.3443, 1.0856], device='cuda:0'), covar=tensor([0.2442, 0.2306, 0.2500, 0.2356], device='cuda:0'), in_proj_covar=tensor([0.1625, 0.1170, 0.1441, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 09:32:13,984 INFO [train.py:968] (0/2) Epoch 30, batch 13400, libri_loss[loss=0.295, simple_loss=0.3629, pruned_loss=0.1135, over 29518.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.342, pruned_loss=0.09418, over 5646903.45 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.339, pruned_loss=0.08816, over 5750299.50 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3445, pruned_loss=0.09578, over 5637195.35 frames. ], batch size: 89, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:32:15,102 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.627e+02 1.564e+03 2.008e+03 2.865e+03 8.336e+03, threshold=4.016e+03, percent-clipped=10.0
+2023-03-15 09:33:03,177 INFO [train.py:968] (0/2) Epoch 30, batch 13450, giga_loss[loss=0.2212, simple_loss=0.3123, pruned_loss=0.06506, over 28944.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3393, pruned_loss=0.09188, over 5650517.48 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.339, pruned_loss=0.08813, over 5753203.77 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3413, pruned_loss=0.09326, over 5638193.25 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:33:33,491 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1333949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:33:41,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3876, 1.5462, 1.5308, 1.4145], device='cuda:0'), covar=tensor([0.2555, 0.2093, 0.1752, 0.2160], device='cuda:0'), in_proj_covar=tensor([0.2064, 0.2040, 0.1945, 0.2085], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 09:33:44,414 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1333959.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 09:33:53,013 INFO [train.py:968] (0/2) Epoch 30, batch 13500, giga_loss[loss=0.2396, simple_loss=0.318, pruned_loss=0.08062, over 27899.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3359, pruned_loss=0.09032, over 5649287.17 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3389, pruned_loss=0.08829, over 5745680.84 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3376, pruned_loss=0.09136, over 5643322.38 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:33:53,630 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.246e+02 1.373e+03 1.767e+03 2.583e+03 9.658e+03, threshold=3.534e+03, percent-clipped=4.0
+2023-03-15 09:34:28,010 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1334000.pt
+2023-03-15 09:34:37,166 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:34:41,930 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1334015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:34:45,296 INFO [train.py:968] (0/2) Epoch 30, batch 13550, giga_loss[loss=0.2383, simple_loss=0.3191, pruned_loss=0.07879, over 28238.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3339, pruned_loss=0.08939, over 5654932.86 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3383, pruned_loss=0.08802, over 5748748.75 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3357, pruned_loss=0.09048, over 5645328.40 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:35:37,706 INFO [train.py:968] (0/2) Epoch 30, batch 13600, giga_loss[loss=0.2407, simple_loss=0.3246, pruned_loss=0.0784, over 28661.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3335, pruned_loss=0.08941, over 5646159.60 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3383, pruned_loss=0.08801, over 5751125.52 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3348, pruned_loss=0.09031, over 5634932.77 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:35:38,361 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.145e+02 1.507e+03 1.946e+03 2.706e+03 6.693e+03, threshold=3.893e+03, percent-clipped=12.0
+2023-03-15 09:36:13,951 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2703, 0.8583, 0.9049, 1.3558], device='cuda:0'), covar=tensor([0.0807, 0.0393, 0.0399, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 09:36:31,903 INFO [train.py:968] (0/2) Epoch 30, batch 13650, giga_loss[loss=0.2639, simple_loss=0.3431, pruned_loss=0.09234, over 27620.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3358, pruned_loss=0.08998, over 5656979.16 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3376, pruned_loss=0.0878, over 5754023.18 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3374, pruned_loss=0.09091, over 5643617.18 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:37:11,026 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1334153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:37:13,705 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1334156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:37:26,796 INFO [train.py:968] (0/2) Epoch 30, batch 13700, giga_loss[loss=0.268, simple_loss=0.3448, pruned_loss=0.09564, over 27569.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3365, pruned_loss=0.08923, over 5659285.17 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3367, pruned_loss=0.08733, over 5755971.26 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3386, pruned_loss=0.09049, over 5643439.48 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:37:29,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.495e+02 1.392e+03 1.901e+03 2.690e+03 5.362e+03, threshold=3.803e+03, percent-clipped=7.0
+2023-03-15 09:37:49,010 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1334185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:37:59,434 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-15 09:38:27,169 INFO [train.py:968] (0/2) Epoch 30, batch 13750, libri_loss[loss=0.234, simple_loss=0.3181, pruned_loss=0.07491, over 29543.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.337, pruned_loss=0.08935, over 5671025.63 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3365, pruned_loss=0.08716, over 5757590.25 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3388, pruned_loss=0.09053, over 5656120.93 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:39:24,239 INFO [train.py:968] (0/2) Epoch 30, batch 13800, libri_loss[loss=0.2619, simple_loss=0.3357, pruned_loss=0.09406, over 29538.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3346, pruned_loss=0.08802, over 5674697.88 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3362, pruned_loss=0.0871, over 5761384.29 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3363, pruned_loss=0.08904, over 5657313.67 frames. ], batch size: 80, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:39:26,740 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.517e+03 2.228e+03 2.900e+03 6.617e+03, threshold=4.456e+03, percent-clipped=11.0
+2023-03-15 09:40:08,771 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7796, 2.1256, 2.2387, 1.6839], device='cuda:0'), covar=tensor([0.3369, 0.2386, 0.2252, 0.2982], device='cuda:0'), in_proj_covar=tensor([0.2060, 0.2038, 0.1939, 0.2082], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 09:40:18,982 INFO [train.py:968] (0/2) Epoch 30, batch 13850, giga_loss[loss=0.281, simple_loss=0.3696, pruned_loss=0.09623, over 28740.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3341, pruned_loss=0.08663, over 5672274.38 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3357, pruned_loss=0.08685, over 5760602.71 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.336, pruned_loss=0.08769, over 5656280.98 frames. ], batch size: 243, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:40:28,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:40:38,640 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334334.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 09:40:38,685 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7923, 2.3996, 1.6430, 0.9959], device='cuda:0'), covar=tensor([0.6152, 0.3138, 0.5126, 0.7114], device='cuda:0'), in_proj_covar=tensor([0.1864, 0.1748, 0.1670, 0.1520], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 09:41:22,169 INFO [train.py:968] (0/2) Epoch 30, batch 13900, giga_loss[loss=0.2142, simple_loss=0.2959, pruned_loss=0.06632, over 24669.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3333, pruned_loss=0.08599, over 5662811.41 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3354, pruned_loss=0.08665, over 5758660.54 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.335, pruned_loss=0.08701, over 5650967.81 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:41:23,765 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.069e+02 1.524e+03 1.899e+03 2.625e+03 8.099e+03, threshold=3.799e+03, percent-clipped=4.0
+2023-03-15 09:41:43,464 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1334386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:41:48,798 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:42:20,059 INFO [train.py:968] (0/2) Epoch 30, batch 13950, giga_loss[loss=0.2695, simple_loss=0.3358, pruned_loss=0.1016, over 28426.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3317, pruned_loss=0.08634, over 5673518.66 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3352, pruned_loss=0.08663, over 5762263.38 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3331, pruned_loss=0.08716, over 5658850.13 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:42:42,734 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-15 09:43:16,791 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1334467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:43:17,159 INFO [train.py:968] (0/2) Epoch 30, batch 14000, giga_loss[loss=0.2346, simple_loss=0.3068, pruned_loss=0.08117, over 28655.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3301, pruned_loss=0.08574, over 5674611.71 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.335, pruned_loss=0.08652, over 5762997.57 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3313, pruned_loss=0.08647, over 5660905.55 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:43:18,873 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.507e+03 1.823e+03 2.581e+03 7.553e+03, threshold=3.646e+03, percent-clipped=9.0
+2023-03-15 09:43:19,151 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1334470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:43:29,459 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1334477.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 09:43:32,583 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1334480.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 09:43:37,720 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1373, 2.2422, 1.2064, 1.3143], device='cuda:0'), covar=tensor([0.1064, 0.0547, 0.0984, 0.1516], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0575, 0.0415, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 09:43:52,991 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1334499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:43:57,842 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1334504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:44:03,315 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1334509.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 09:44:04,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5422, 1.8373, 1.3300, 1.4937], device='cuda:0'), covar=tensor([0.1063, 0.0554, 0.0892, 0.1266], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0449, 0.0524, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 09:44:15,396 INFO [train.py:968] (0/2) Epoch 30, batch 14050, giga_loss[loss=0.2332, simple_loss=0.3268, pruned_loss=0.06982, over 28916.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.331, pruned_loss=0.08614, over 5667071.97 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3349, pruned_loss=0.08644, over 5765003.22 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.332, pruned_loss=0.08679, over 5653136.32 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:44:33,640 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1334533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:44:37,330 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1334536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:45:03,101 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3139, 1.7622, 1.6034, 1.5150], device='cuda:0'), covar=tensor([0.2272, 0.1999, 0.2177, 0.2008], device='cuda:0'), in_proj_covar=tensor([0.0509, 0.0754, 0.0726, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 09:45:13,348 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1334565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:45:16,010 INFO [train.py:968] (0/2) Epoch 30, batch 14100, giga_loss[loss=0.243, simple_loss=0.3253, pruned_loss=0.08037, over 28153.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3334, pruned_loss=0.08703, over 5663858.90 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3343, pruned_loss=0.08615, over 5767005.18 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3348, pruned_loss=0.08781, over 5649618.40 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:45:19,763 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.401e+03 1.710e+03 2.486e+03 5.241e+03, threshold=3.420e+03, percent-clipped=10.0
+2023-03-15 09:46:16,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9195, 1.0809, 2.8746, 2.8156], device='cuda:0'), covar=tensor([0.1642, 0.2727, 0.0589, 0.1075], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0679, 0.1012, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 09:46:20,839 INFO [train.py:968] (0/2) Epoch 30, batch 14150, giga_loss[loss=0.2268, simple_loss=0.3079, pruned_loss=0.07287, over 29027.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3299, pruned_loss=0.08455, over 5674420.37 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3334, pruned_loss=0.08569, over 5768768.88 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3317, pruned_loss=0.08558, over 5659672.01 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:47:25,445 INFO [train.py:968] (0/2) Epoch 30, batch 14200, giga_loss[loss=0.2443, simple_loss=0.3297, pruned_loss=0.07946, over 28913.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3303, pruned_loss=0.08499, over 5681655.62 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3331, pruned_loss=0.08558, over 5770340.54 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.332, pruned_loss=0.08588, over 5667500.15 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:47:29,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.046e+03 1.461e+03 1.750e+03 2.233e+03 5.734e+03, threshold=3.499e+03, percent-clipped=5.0
+2023-03-15 09:48:28,592 INFO [train.py:968] (0/2) Epoch 30, batch 14250, giga_loss[loss=0.259, simple_loss=0.354, pruned_loss=0.08197, over 28816.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3327, pruned_loss=0.08575, over 5686228.98 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3327, pruned_loss=0.08548, over 5774432.65 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3343, pruned_loss=0.08657, over 5668026.52 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:48:29,855 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1334719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:48:30,221 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.59 vs. limit=5.0
+2023-03-15 09:48:30,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2842, 2.5184, 2.3663, 2.0161], device='cuda:0'), covar=tensor([0.1977, 0.2103, 0.1897, 0.2225], device='cuda:0'), in_proj_covar=tensor([0.0507, 0.0751, 0.0724, 0.0692], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 09:49:16,458 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:49:22,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7001, 1.9109, 1.3843, 1.6211], device='cuda:0'), covar=tensor([0.1135, 0.0680, 0.1043, 0.1150], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0450, 0.0525, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 09:49:22,453 INFO [train.py:968] (0/2) Epoch 30, batch 14300, giga_loss[loss=0.2737, simple_loss=0.3645, pruned_loss=0.09142, over 28867.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3367, pruned_loss=0.08615, over 5680892.55 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3325, pruned_loss=0.08546, over 5773012.44 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3384, pruned_loss=0.08688, over 5663062.98 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:49:28,231 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.073e+03 1.698e+03 2.175e+03 2.927e+03 1.026e+04, threshold=4.351e+03, percent-clipped=14.0
+2023-03-15 09:50:08,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6239, 2.1173, 1.4622, 1.6441], device='cuda:0'), covar=tensor([0.1107, 0.0497, 0.0967, 0.1060], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0449, 0.0525, 0.0463], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 09:50:23,131 INFO [train.py:968] (0/2) Epoch 30, batch 14350, libri_loss[loss=0.2819, simple_loss=0.3659, pruned_loss=0.09896, over 29766.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3392, pruned_loss=0.08596, over 5688427.89 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3324, pruned_loss=0.08537, over 5776713.51 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3408, pruned_loss=0.08664, over 5668524.47 frames. ], batch size: 87, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:51:16,245 INFO [train.py:968] (0/2) Epoch 30, batch 14400, giga_loss[loss=0.2995, simple_loss=0.3641, pruned_loss=0.1175, over 26832.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3386, pruned_loss=0.0849, over 5684663.30 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3324, pruned_loss=0.08558, over 5778339.12 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3401, pruned_loss=0.08528, over 5664046.79 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:51:19,889 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.709e+02 1.427e+03 1.739e+03 2.503e+03 1.039e+04, threshold=3.478e+03, percent-clipped=6.0
+2023-03-15 09:51:28,271 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:51:57,619 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1334904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:52:00,085 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1334907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:52:11,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7838, 2.1292, 1.3742, 1.6589], device='cuda:0'), covar=tensor([0.1110, 0.0643, 0.1056, 0.1210], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0447, 0.0523, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 09:52:14,068 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4392, 1.9673, 1.8556, 1.6377], device='cuda:0'), covar=tensor([0.2313, 0.2066, 0.2190, 0.2178], device='cuda:0'), in_proj_covar=tensor([0.0508, 0.0752, 0.0726, 0.0693], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 09:52:14,394 INFO [train.py:968] (0/2) Epoch 30, batch 14450, giga_loss[loss=0.247, simple_loss=0.3323, pruned_loss=0.08091, over 28933.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3389, pruned_loss=0.08549, over 5684262.35 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3324, pruned_loss=0.0856, over 5779721.03 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3403, pruned_loss=0.08576, over 5664141.50 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:52:36,291 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1334936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:52:52,981 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6582, 1.9105, 1.5506, 1.6840], device='cuda:0'), covar=tensor([0.2878, 0.3012, 0.3511, 0.2607], device='cuda:0'), in_proj_covar=tensor([0.1620, 0.1166, 0.1436, 0.1016], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 09:53:12,872 INFO [train.py:968] (0/2) Epoch 30, batch 14500, giga_loss[loss=0.277, simple_loss=0.3486, pruned_loss=0.1027, over 28911.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.337, pruned_loss=0.08541, over 5690554.15 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.332, pruned_loss=0.08545, over 5783226.55 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3387, pruned_loss=0.08577, over 5668625.30 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:53:17,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.503e+03 2.079e+03 3.137e+03 9.553e+03, threshold=4.157e+03, percent-clipped=18.0
+2023-03-15 09:54:27,533 INFO [train.py:968] (0/2) Epoch 30, batch 14550, giga_loss[loss=0.2549, simple_loss=0.3356, pruned_loss=0.0871, over 28621.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3376, pruned_loss=0.08667, over 5698425.91 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3319, pruned_loss=0.08549, over 5784598.04 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3391, pruned_loss=0.08693, over 5679122.77 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:54:32,840 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1335022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:54:35,965 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1335025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:55:26,056 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1335054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:55:43,537 INFO [train.py:968] (0/2) Epoch 30, batch 14600, giga_loss[loss=0.2461, simple_loss=0.3044, pruned_loss=0.09392, over 23877.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.335, pruned_loss=0.08598, over 5688605.44 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3317, pruned_loss=0.08538, over 5785404.49 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3364, pruned_loss=0.08631, over 5671358.39 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:55:48,481 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.369e+02 1.347e+03 1.666e+03 2.390e+03 5.361e+03, threshold=3.333e+03, percent-clipped=3.0
+2023-03-15 09:56:17,509 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1335094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:56:45,240 INFO [train.py:968] (0/2) Epoch 30, batch 14650, libri_loss[loss=0.2604, simple_loss=0.3417, pruned_loss=0.08954, over 29678.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3326, pruned_loss=0.08458, over 5690538.22 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3315, pruned_loss=0.08528, over 5789208.60 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3341, pruned_loss=0.08495, over 5668856.25 frames. ], batch size: 91, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:57:47,871 INFO [train.py:968] (0/2) Epoch 30, batch 14700, giga_loss[loss=0.2134, simple_loss=0.3012, pruned_loss=0.06279, over 28824.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3297, pruned_loss=0.08314, over 5683330.50 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3316, pruned_loss=0.0854, over 5781387.40 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3308, pruned_loss=0.08329, over 5671118.07 frames. ], batch size: 243, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:57:52,193 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.238e+02 1.686e+03 2.138e+03 2.973e+03 8.741e+03, threshold=4.277e+03, percent-clipped=14.0
+2023-03-15 09:58:44,525 INFO [train.py:968] (0/2) Epoch 30, batch 14750, giga_loss[loss=0.2701, simple_loss=0.3547, pruned_loss=0.09277, over 28970.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3328, pruned_loss=0.0851, over 5675086.39 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3317, pruned_loss=0.08547, over 5771880.19 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3335, pruned_loss=0.08512, over 5670241.72 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:59:09,222 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1335237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:59:12,959 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1335240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:59:44,695 INFO [train.py:968] (0/2) Epoch 30, batch 14800, giga_loss[loss=0.2434, simple_loss=0.3234, pruned_loss=0.08171, over 29077.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3361, pruned_loss=0.0872, over 5682131.26 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3317, pruned_loss=0.0855, over 5774291.67 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3368, pruned_loss=0.08721, over 5673606.29 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:59:46,332 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1335269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:59:50,116 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.092e+03 1.638e+03 2.058e+03 2.803e+03 1.166e+04, threshold=4.116e+03, percent-clipped=12.0
+2023-03-15 10:00:41,517 INFO [train.py:968] (0/2) Epoch 30, batch 14850, giga_loss[loss=0.2644, simple_loss=0.3384, pruned_loss=0.09519, over 28928.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.334, pruned_loss=0.0873, over 5683946.25 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3315, pruned_loss=0.08556, over 5776216.45 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3349, pruned_loss=0.08731, over 5672310.02 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:00:42,161 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 10:00:50,788 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1528, 3.9825, 3.8047, 1.8094], device='cuda:0'), covar=tensor([0.0674, 0.0825, 0.0878, 0.2140], device='cuda:0'), in_proj_covar=tensor([0.1300, 0.1200, 0.1006, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 10:01:40,195 INFO [train.py:968] (0/2) Epoch 30, batch 14900, giga_loss[loss=0.2505, simple_loss=0.3299, pruned_loss=0.08556, over 28904.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3342, pruned_loss=0.08795, over 5676243.22 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3317, pruned_loss=0.08579, over 5765844.93 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3346, pruned_loss=0.0878, over 5674086.87 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:01:45,789 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.139e+03 1.588e+03 1.831e+03 2.510e+03 8.398e+03, threshold=3.663e+03, percent-clipped=6.0
+2023-03-15 10:01:59,771 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1335384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:02:39,694 INFO [train.py:968] (0/2) Epoch 30, batch 14950, giga_loss[loss=0.3395, simple_loss=0.3897, pruned_loss=0.1447, over 26883.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3353, pruned_loss=0.08885, over 5684387.86 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3315, pruned_loss=0.08574, over 5770137.56 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.336, pruned_loss=0.08884, over 5676501.00 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:02:56,662 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0456, 1.3684, 1.1660, 0.2294], device='cuda:0'), covar=tensor([0.4638, 0.4050, 0.5575, 0.8299], device='cuda:0'), in_proj_covar=tensor([0.1868, 0.1746, 0.1676, 0.1523], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 10:03:41,690 INFO [train.py:968] (0/2) Epoch 30, batch 15000, giga_loss[loss=0.2619, simple_loss=0.3462, pruned_loss=0.08885, over 28894.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3365, pruned_loss=0.08836, over 5690684.14 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.331, pruned_loss=0.08548, over 5774032.06 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3377, pruned_loss=0.08868, over 5677620.12 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:03:41,695 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 10:03:49,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9120, 3.6571, 3.5414, 1.7100], device='cuda:0'), covar=tensor([0.0844, 0.1005, 0.0995, 0.2320], device='cuda:0'), in_proj_covar=tensor([0.1306, 0.1204, 0.1010, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 10:03:51,039 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.1929, simple_loss=0.2945, pruned_loss=0.04562, over 944034.00 frames. 
+2023-03-15 10:03:51,040 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 10:03:57,278 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.478e+03 1.890e+03 2.641e+03 8.013e+03, threshold=3.780e+03, percent-clipped=10.0
+2023-03-15 10:04:38,704 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1335504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:05:01,657 INFO [train.py:968] (0/2) Epoch 30, batch 15050, giga_loss[loss=0.2762, simple_loss=0.3276, pruned_loss=0.1124, over 24366.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3361, pruned_loss=0.08794, over 5670488.02 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3309, pruned_loss=0.08549, over 5765229.23 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3373, pruned_loss=0.08826, over 5665088.09 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 10:06:15,182 INFO [train.py:968] (0/2) Epoch 30, batch 15100, giga_loss[loss=0.2875, simple_loss=0.3525, pruned_loss=0.1112, over 28774.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3334, pruned_loss=0.08734, over 5664561.90 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3307, pruned_loss=0.0854, over 5766525.60 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3346, pruned_loss=0.0877, over 5657850.78 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 10:06:23,944 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.429e+02 1.539e+03 2.174e+03 3.082e+03 6.990e+03, threshold=4.347e+03, percent-clipped=14.0
+2023-03-15 10:07:22,030 INFO [train.py:968] (0/2) Epoch 30, batch 15150, giga_loss[loss=0.2496, simple_loss=0.3189, pruned_loss=0.09015, over 28872.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3279, pruned_loss=0.08527, over 5664335.21 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3304, pruned_loss=0.08533, over 5767618.22 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3291, pruned_loss=0.08563, over 5656359.31 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 10:07:41,157 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5975, 1.8934, 2.0210, 1.6563], device='cuda:0'), covar=tensor([0.2778, 0.2313, 0.2141, 0.2456], device='cuda:0'), in_proj_covar=tensor([0.2057, 0.2025, 0.1927, 0.2075], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 10:08:11,832 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3203, 3.2887, 1.5000, 1.5122], device='cuda:0'), covar=tensor([0.1017, 0.0348, 0.0968, 0.1363], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0575, 0.0416, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 10:08:21,148 INFO [train.py:968] (0/2) Epoch 30, batch 15200, giga_loss[loss=0.2687, simple_loss=0.3492, pruned_loss=0.09409, over 28623.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.326, pruned_loss=0.08421, over 5671722.92 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3306, pruned_loss=0.08557, over 5768775.80 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3267, pruned_loss=0.08428, over 5663571.12 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:08:28,898 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.367e+02 1.632e+03 2.180e+03 3.103e+03 9.761e+03, threshold=4.359e+03, percent-clipped=10.0
+2023-03-15 10:09:15,250 INFO [train.py:968] (0/2) Epoch 30, batch 15250, giga_loss[loss=0.2224, simple_loss=0.3094, pruned_loss=0.06774, over 28879.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3274, pruned_loss=0.08547, over 5658452.66 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3307, pruned_loss=0.08574, over 5763281.81 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3278, pruned_loss=0.08533, over 5654025.52 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:09:57,058 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1335759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:10:10,463 INFO [train.py:968] (0/2) Epoch 30, batch 15300, giga_loss[loss=0.1854, simple_loss=0.2601, pruned_loss=0.0553, over 24273.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3257, pruned_loss=0.08392, over 5666596.45 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3305, pruned_loss=0.08559, over 5765899.18 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3261, pruned_loss=0.08393, over 5658164.78 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:10:15,771 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.983e+02 1.501e+03 1.917e+03 2.483e+03 5.109e+03, threshold=3.834e+03, percent-clipped=3.0
+2023-03-15 10:10:35,945 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4600, 1.7265, 1.2045, 1.2618], device='cuda:0'), covar=tensor([0.1100, 0.0474, 0.0968, 0.1244], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0447, 0.0523, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 10:11:07,022 INFO [train.py:968] (0/2) Epoch 30, batch 15350, giga_loss[loss=0.2324, simple_loss=0.3219, pruned_loss=0.07144, over 28635.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3243, pruned_loss=0.0826, over 5652594.71 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3305, pruned_loss=0.08566, over 5759301.60 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3245, pruned_loss=0.08251, over 5650143.27 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:11:15,852 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6160, 1.6565, 1.8217, 1.4183], device='cuda:0'), covar=tensor([0.1937, 0.2646, 0.1575, 0.1969], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0716, 0.0990, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 10:11:19,449 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-15 10:12:05,808 INFO [train.py:968] (0/2) Epoch 30, batch 15400, giga_loss[loss=0.2432, simple_loss=0.3301, pruned_loss=0.07814, over 28134.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3238, pruned_loss=0.08217, over 5655559.14 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3303, pruned_loss=0.08559, over 5750000.50 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3239, pruned_loss=0.08206, over 5658874.27 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:12:15,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.328e+02 1.619e+03 2.201e+03 2.941e+03 8.240e+03, threshold=4.401e+03, percent-clipped=4.0
+2023-03-15 10:12:24,170 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1335879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:12:56,526 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1335902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:12:58,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1335905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:13:07,261 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-15 10:13:14,480 INFO [train.py:968] (0/2) Epoch 30, batch 15450, giga_loss[loss=0.224, simple_loss=0.3132, pruned_loss=0.06738, over 28324.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3232, pruned_loss=0.0819, over 5648876.23 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3304, pruned_loss=0.08564, over 5751423.41 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3232, pruned_loss=0.08171, over 5648636.86 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:13:37,438 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1335934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:14:19,416 INFO [train.py:968] (0/2) Epoch 30, batch 15500, giga_loss[loss=0.2146, simple_loss=0.2963, pruned_loss=0.06644, over 28979.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.324, pruned_loss=0.0822, over 5650871.42 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3302, pruned_loss=0.08556, over 5753012.32 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.324, pruned_loss=0.08208, over 5648257.83 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:14:25,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.380e+02 1.403e+03 1.795e+03 2.590e+03 4.913e+03, threshold=3.590e+03, percent-clipped=3.0
+2023-03-15 10:14:33,747 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7420, 2.0721, 1.7273, 1.8247], device='cuda:0'), covar=tensor([0.2500, 0.2326, 0.2482, 0.2367], device='cuda:0'), in_proj_covar=tensor([0.1629, 0.1169, 0.1444, 0.1021], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 10:14:58,993 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1336000.pt
+2023-03-15 10:15:12,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1673, 2.4322, 2.2671, 2.1455], device='cuda:0'), covar=tensor([0.2148, 0.2370, 0.2091, 0.2259], device='cuda:0'), in_proj_covar=tensor([0.0503, 0.0747, 0.0721, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 10:15:17,885 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-15 10:15:20,752 INFO [train.py:968] (0/2) Epoch 30, batch 15550, libri_loss[loss=0.2684, simple_loss=0.345, pruned_loss=0.09591, over 25590.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3256, pruned_loss=0.08371, over 5656308.35 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3304, pruned_loss=0.08572, over 5750996.61 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3253, pruned_loss=0.08339, over 5654097.11 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:15:27,782 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:15:32,157 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:15:49,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3786, 1.5931, 1.2490, 1.4414], device='cuda:0'), covar=tensor([0.0796, 0.0335, 0.0378, 0.0926], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 10:16:05,761 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:16:20,009 INFO [train.py:968] (0/2) Epoch 30, batch 15600, libri_loss[loss=0.3184, simple_loss=0.3862, pruned_loss=0.1253, over 29208.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3242, pruned_loss=0.08346, over 5644472.79 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3302, pruned_loss=0.08579, over 5742790.55 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3239, pruned_loss=0.08305, over 5647195.06 frames. ], batch size: 94, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:16:28,395 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.653e+02 1.471e+03 2.036e+03 3.270e+03 7.906e+03, threshold=4.072e+03, percent-clipped=17.0
+2023-03-15 10:16:29,340 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7560, 2.0505, 1.3994, 1.6515], device='cuda:0'), covar=tensor([0.1015, 0.0531, 0.1025, 0.1049], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0449, 0.0524, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 10:16:30,116 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:17:14,884 INFO [train.py:968] (0/2) Epoch 30, batch 15650, giga_loss[loss=0.2844, simple_loss=0.3661, pruned_loss=0.1014, over 28108.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3258, pruned_loss=0.0825, over 5661301.14 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3299, pruned_loss=0.08562, over 5744680.68 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3256, pruned_loss=0.08225, over 5660073.06 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:18:17,721 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2758, 1.5261, 1.4753, 1.1829], device='cuda:0'), covar=tensor([0.1458, 0.2127, 0.1251, 0.1716], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0716, 0.0990, 0.0891], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 10:18:19,770 INFO [train.py:968] (0/2) Epoch 30, batch 15700, giga_loss[loss=0.2342, simple_loss=0.3228, pruned_loss=0.07278, over 28885.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3286, pruned_loss=0.08327, over 5654365.23 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3299, pruned_loss=0.08562, over 5744680.68 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3285, pruned_loss=0.08308, over 5653409.40 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:18:26,079 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.989e+02 1.479e+03 2.007e+03 2.773e+03 5.948e+03, threshold=4.013e+03, percent-clipped=8.0
+2023-03-15 10:18:36,253 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:18:40,470 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5047, 1.7007, 1.2524, 1.2179], device='cuda:0'), covar=tensor([0.1036, 0.0487, 0.0927, 0.1187], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0447, 0.0523, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 10:19:05,885 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3043, 1.4277, 1.3853, 1.2853], device='cuda:0'), covar=tensor([0.2330, 0.2105, 0.1938, 0.2226], device='cuda:0'), in_proj_covar=tensor([0.2049, 0.2017, 0.1920, 0.2062], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 10:19:14,236 INFO [train.py:968] (0/2) Epoch 30, batch 15750, giga_loss[loss=0.2603, simple_loss=0.3452, pruned_loss=0.08771, over 28660.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3297, pruned_loss=0.08334, over 5664244.10 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3298, pruned_loss=0.08564, over 5750248.39 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3297, pruned_loss=0.08309, over 5655456.96 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:19:53,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9704, 3.8387, 3.6093, 1.9009], device='cuda:0'), covar=tensor([0.0652, 0.0758, 0.0802, 0.2388], device='cuda:0'), in_proj_covar=tensor([0.1301, 0.1197, 0.1005, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 10:20:09,440 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:20:12,324 INFO [train.py:968] (0/2) Epoch 30, batch 15800, giga_loss[loss=0.2315, simple_loss=0.3176, pruned_loss=0.07268, over 28452.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3297, pruned_loss=0.083, over 5674391.30 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3298, pruned_loss=0.08556, over 5749927.39 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3298, pruned_loss=0.08286, over 5666878.27 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:20:20,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.470e+03 1.911e+03 2.657e+03 5.683e+03, threshold=3.821e+03, percent-clipped=5.0
+2023-03-15 10:20:30,857 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:20:38,211 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1980, 1.5044, 1.3645, 1.1481], device='cuda:0'), covar=tensor([0.3254, 0.2651, 0.2019, 0.2725], device='cuda:0'), in_proj_covar=tensor([0.2048, 0.2016, 0.1917, 0.2061], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 10:21:07,864 INFO [train.py:968] (0/2) Epoch 30, batch 15850, giga_loss[loss=0.1781, simple_loss=0.2721, pruned_loss=0.04207, over 28942.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3272, pruned_loss=0.08158, over 5684593.31 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3289, pruned_loss=0.08516, over 5753261.87 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3279, pruned_loss=0.08174, over 5673904.92 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:22:05,734 INFO [train.py:968] (0/2) Epoch 30, batch 15900, giga_loss[loss=0.2213, simple_loss=0.3114, pruned_loss=0.06557, over 28927.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3245, pruned_loss=0.07987, over 5690269.30 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3284, pruned_loss=0.08483, over 5758323.69 frames. ], giga_tot_loss[loss=0.2429, simple_loss=0.3255, pruned_loss=0.08014, over 5675033.69 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:22:14,403 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.486e+02 1.464e+03 1.880e+03 2.613e+03 8.516e+03, threshold=3.760e+03, percent-clipped=9.0
+2023-03-15 10:22:29,751 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:22:44,769 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5174, 1.8144, 1.5374, 1.6022], device='cuda:0'), covar=tensor([0.0746, 0.0307, 0.0331, 0.0856], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 10:23:00,470 INFO [train.py:968] (0/2) Epoch 30, batch 15950, giga_loss[loss=0.2549, simple_loss=0.336, pruned_loss=0.08692, over 28877.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3233, pruned_loss=0.08004, over 5681064.69 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3283, pruned_loss=0.08479, over 5751862.29 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.324, pruned_loss=0.08014, over 5673141.56 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:23:14,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9109, 3.7561, 3.5594, 2.1036], device='cuda:0'), covar=tensor([0.0636, 0.0812, 0.0826, 0.1856], device='cuda:0'), in_proj_covar=tensor([0.1304, 0.1200, 0.1006, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 10:23:39,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:23:48,022 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3584, 1.3987, 3.8881, 3.3583], device='cuda:0'), covar=tensor([0.1651, 0.2796, 0.0506, 0.0944], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0677, 0.1007, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 10:23:59,556 INFO [train.py:968] (0/2) Epoch 30, batch 16000, giga_loss[loss=0.2878, simple_loss=0.3591, pruned_loss=0.1082, over 29050.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3233, pruned_loss=0.08044, over 5681343.37 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3274, pruned_loss=0.08441, over 5755981.34 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3246, pruned_loss=0.08076, over 5668931.51 frames. ], batch size: 200, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:24:08,438 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.457e+02 1.478e+03 1.911e+03 2.803e+03 7.760e+03, threshold=3.822e+03, percent-clipped=8.0
+2023-03-15 10:25:01,952 INFO [train.py:968] (0/2) Epoch 30, batch 16050, giga_loss[loss=0.2554, simple_loss=0.3346, pruned_loss=0.08811, over 28532.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3261, pruned_loss=0.08151, over 5681000.64 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3272, pruned_loss=0.08428, over 5756836.43 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3272, pruned_loss=0.08185, over 5670238.35 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:25:26,160 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4992, 1.2925, 3.9864, 3.3859], device='cuda:0'), covar=tensor([0.1610, 0.2999, 0.0469, 0.0931], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0676, 0.1006, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 10:25:51,641 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:26:03,346 INFO [train.py:968] (0/2) Epoch 30, batch 16100, giga_loss[loss=0.2355, simple_loss=0.302, pruned_loss=0.08454, over 24393.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3274, pruned_loss=0.08307, over 5678042.89 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3271, pruned_loss=0.08419, over 5761077.64 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3285, pruned_loss=0.08337, over 5663698.26 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:26:13,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.079e+03 1.695e+03 2.161e+03 2.931e+03 6.872e+03, threshold=4.322e+03, percent-clipped=10.0
+2023-03-15 10:26:32,102 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:26:35,991 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:26:57,192 INFO [train.py:968] (0/2) Epoch 30, batch 16150, giga_loss[loss=0.2386, simple_loss=0.3328, pruned_loss=0.07221, over 28688.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.331, pruned_loss=0.08483, over 5673134.59 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3276, pruned_loss=0.08453, over 5748036.46 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3314, pruned_loss=0.08475, over 5670857.27 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:27:06,593 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:27:21,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:27:43,356 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:27:50,482 INFO [train.py:968] (0/2) Epoch 30, batch 16200, giga_loss[loss=0.2446, simple_loss=0.3284, pruned_loss=0.08039, over 28929.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3323, pruned_loss=0.08466, over 5683106.91 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3265, pruned_loss=0.08405, over 5751533.58 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3337, pruned_loss=0.08506, over 5676185.08 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:27:54,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3614, 1.6194, 1.5360, 1.2637], device='cuda:0'), covar=tensor([0.3057, 0.2415, 0.2159, 0.2797], device='cuda:0'), in_proj_covar=tensor([0.2047, 0.2013, 0.1915, 0.2061], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 10:28:00,218 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.057e+03 1.497e+03 1.914e+03 2.532e+03 4.400e+03, threshold=3.828e+03, percent-clipped=2.0
+2023-03-15 10:28:29,168 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:28:33,324 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:28:50,946 INFO [train.py:968] (0/2) Epoch 30, batch 16250, giga_loss[loss=0.2401, simple_loss=0.3258, pruned_loss=0.07724, over 28985.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3326, pruned_loss=0.08459, over 5682173.81 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3263, pruned_loss=0.08387, over 5754216.76 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3341, pruned_loss=0.0851, over 5673053.33 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:29:09,768 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:29:51,698 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:29:52,246 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3290, 4.1681, 3.9542, 1.9096], device='cuda:0'), covar=tensor([0.0640, 0.0871, 0.0849, 0.2020], device='cuda:0'), in_proj_covar=tensor([0.1302, 0.1199, 0.1006, 0.0748], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 10:29:53,959 INFO [train.py:968] (0/2) Epoch 30, batch 16300, giga_loss[loss=0.2192, simple_loss=0.3046, pruned_loss=0.06686, over 28982.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3312, pruned_loss=0.0837, over 5692325.45 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3268, pruned_loss=0.08404, over 5757050.73 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3321, pruned_loss=0.08397, over 5680722.83 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:30:06,094 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-15 10:30:06,352 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.475e+02 1.535e+03 2.017e+03 2.743e+03 1.052e+04, threshold=4.034e+03, percent-clipped=9.0
+2023-03-15 10:30:06,949 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5166, 1.9352, 1.7472, 1.6979], device='cuda:0'), covar=tensor([0.2258, 0.2546, 0.2282, 0.2250], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0744, 0.0718, 0.0689], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 10:30:15,625 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:30:20,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:30:41,150 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:30:44,258 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:30:58,379 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:31:00,801 INFO [train.py:968] (0/2) Epoch 30, batch 16350, giga_loss[loss=0.2632, simple_loss=0.3368, pruned_loss=0.09482, over 27662.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3308, pruned_loss=0.08418, over 5687016.70 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3268, pruned_loss=0.08399, over 5757941.47 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3316, pruned_loss=0.08444, over 5676898.58 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:31:21,899 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:31:59,151 INFO [train.py:968] (0/2) Epoch 30, batch 16400, giga_loss[loss=0.2491, simple_loss=0.3266, pruned_loss=0.08586, over 27793.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3292, pruned_loss=0.08344, over 5677543.77 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3265, pruned_loss=0.08386, over 5760426.86 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3302, pruned_loss=0.08376, over 5665055.02 frames. ], batch size: 474, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:32:10,625 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.008e+03 1.565e+03 1.984e+03 2.605e+03 6.877e+03, threshold=3.968e+03, percent-clipped=3.0
+2023-03-15 10:32:23,635 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-15 10:32:48,865 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:32:50,888 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3933, 1.6476, 1.6435, 1.2159], device='cuda:0'), covar=tensor([0.1826, 0.2975, 0.1643, 0.1991], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0713, 0.0987, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 10:32:52,993 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:32:59,942 INFO [train.py:968] (0/2) Epoch 30, batch 16450, libri_loss[loss=0.3012, simple_loss=0.3642, pruned_loss=0.1191, over 29497.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3287, pruned_loss=0.08443, over 5679861.47 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3264, pruned_loss=0.08398, over 5759434.10 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3295, pruned_loss=0.08458, over 5668945.80 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:33:23,478 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:33:55,898 INFO [train.py:968] (0/2) Epoch 30, batch 16500, giga_loss[loss=0.2648, simple_loss=0.3444, pruned_loss=0.09259, over 28833.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3289, pruned_loss=0.08518, over 5671769.75 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3263, pruned_loss=0.08388, over 5751094.57 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3298, pruned_loss=0.08542, over 5668028.71 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:34:00,867 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:34:05,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.456e+02 1.595e+03 2.120e+03 2.976e+03 6.467e+03, threshold=4.241e+03, percent-clipped=6.0
+2023-03-15 10:34:34,733 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-15 10:34:53,733 INFO [train.py:968] (0/2) Epoch 30, batch 16550, giga_loss[loss=0.217, simple_loss=0.2852, pruned_loss=0.07441, over 24396.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3282, pruned_loss=0.08392, over 5673243.66 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3258, pruned_loss=0.08368, over 5754652.08 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3294, pruned_loss=0.08431, over 5664808.26 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:34:58,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5619, 1.7388, 1.6591, 1.4962], device='cuda:0'), covar=tensor([0.2512, 0.2396, 0.2157, 0.2423], device='cuda:0'), in_proj_covar=tensor([0.2054, 0.2021, 0.1922, 0.2068], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 10:35:10,272 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1337031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:35:14,279 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1974, 1.8677, 1.4985, 1.5222], device='cuda:0'), covar=tensor([0.2747, 0.2231, 0.2334, 0.2316], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0742, 0.0717, 0.0687], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 10:35:48,865 INFO [train.py:968] (0/2) Epoch 30, batch 16600, libri_loss[loss=0.1992, simple_loss=0.2771, pruned_loss=0.06064, over 29408.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3278, pruned_loss=0.08245, over 5680145.89 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3252, pruned_loss=0.08333, over 5757313.61 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3294, pruned_loss=0.08307, over 5669071.87 frames. ], batch size: 67, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:35:59,836 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.608e+02 1.469e+03 2.008e+03 2.915e+03 7.433e+03, threshold=4.016e+03, percent-clipped=9.0
+2023-03-15 10:36:41,873 INFO [train.py:968] (0/2) Epoch 30, batch 16650, giga_loss[loss=0.2163, simple_loss=0.2968, pruned_loss=0.06795, over 24266.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3295, pruned_loss=0.08215, over 5675701.21 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3252, pruned_loss=0.08333, over 5759215.27 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3309, pruned_loss=0.08262, over 5662428.11 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:37:04,217 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5421, 1.8739, 1.5057, 1.5413], device='cuda:0'), covar=tensor([0.2827, 0.2902, 0.3334, 0.2641], device='cuda:0'), in_proj_covar=tensor([0.1627, 0.1167, 0.1441, 0.1020], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 10:37:32,544 INFO [train.py:968] (0/2) Epoch 30, batch 16700, giga_loss[loss=0.2525, simple_loss=0.3403, pruned_loss=0.08238, over 28591.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3306, pruned_loss=0.08208, over 5688424.45 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.325, pruned_loss=0.08326, over 5761252.08 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.332, pruned_loss=0.08247, over 5673551.52 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:37:42,755 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.050e+02 1.532e+03 2.015e+03 2.710e+03 5.681e+03, threshold=4.029e+03, percent-clipped=6.0
+2023-03-15 10:38:25,264 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5038, 1.8195, 1.4429, 1.5293], device='cuda:0'), covar=tensor([0.2840, 0.2765, 0.3227, 0.2470], device='cuda:0'), in_proj_covar=tensor([0.1625, 0.1166, 0.1440, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 10:38:31,329 INFO [train.py:968] (0/2) Epoch 30, batch 16750, libri_loss[loss=0.2446, simple_loss=0.327, pruned_loss=0.0811, over 29488.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3305, pruned_loss=0.08186, over 5692100.00 frames. ], libri_tot_loss[loss=0.2457, simple_loss=0.325, pruned_loss=0.08318, over 5762688.52 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3319, pruned_loss=0.08221, over 5676135.85 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:39:37,153 INFO [train.py:968] (0/2) Epoch 30, batch 16800, giga_loss[loss=0.2576, simple_loss=0.3401, pruned_loss=0.08761, over 28950.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.33, pruned_loss=0.08149, over 5686869.83 frames. ], libri_tot_loss[loss=0.2452, simple_loss=0.3245, pruned_loss=0.08292, over 5764903.84 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3315, pruned_loss=0.08196, over 5671394.81 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:39:49,890 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.052e+03 1.436e+03 1.881e+03 2.428e+03 9.987e+03, threshold=3.761e+03, percent-clipped=5.0
+2023-03-15 10:40:09,607 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1337290.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 10:40:47,330 INFO [train.py:968] (0/2) Epoch 30, batch 16850, giga_loss[loss=0.2311, simple_loss=0.3259, pruned_loss=0.06812, over 28502.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3309, pruned_loss=0.08181, over 5678801.79 frames. ], libri_tot_loss[loss=0.2451, simple_loss=0.3245, pruned_loss=0.08285, over 5764572.01 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3322, pruned_loss=0.08224, over 5666266.87 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:41:07,162 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3836, 1.6595, 1.3077, 1.4603], device='cuda:0'), covar=tensor([0.2810, 0.2806, 0.3324, 0.2439], device='cuda:0'), in_proj_covar=tensor([0.1622, 0.1165, 0.1437, 0.1016], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 10:41:20,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3799, 1.7145, 1.4224, 1.5267], device='cuda:0'), covar=tensor([0.0763, 0.0364, 0.0364, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 10:41:29,710 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1337347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:41:52,901 INFO [train.py:968] (0/2) Epoch 30, batch 16900, giga_loss[loss=0.3018, simple_loss=0.3576, pruned_loss=0.123, over 26858.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.331, pruned_loss=0.08127, over 5677296.83 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3242, pruned_loss=0.08278, over 5757880.31 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3324, pruned_loss=0.08166, over 5670294.69 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:42:06,749 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.264e+02 1.409e+03 1.803e+03 2.488e+03 6.633e+03, threshold=3.607e+03, percent-clipped=10.0
+2023-03-15 10:42:13,705 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6024, 1.8663, 1.3424, 1.4492], device='cuda:0'), covar=tensor([0.1199, 0.0624, 0.1052, 0.1233], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0447, 0.0524, 0.0460], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 10:42:46,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1337406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:42:47,068 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1337407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:42:53,746 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-15 10:42:57,951 INFO [train.py:968] (0/2) Epoch 30, batch 16950, giga_loss[loss=0.2809, simple_loss=0.359, pruned_loss=0.1014, over 29163.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3346, pruned_loss=0.08315, over 5678459.22 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3243, pruned_loss=0.08271, over 5756970.63 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3359, pruned_loss=0.08351, over 5671964.58 frames. ], batch size: 200, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:43:08,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9857, 2.3498, 2.2106, 1.9284], device='cuda:0'), covar=tensor([0.2215, 0.2357, 0.2074, 0.2492], device='cuda:0'), in_proj_covar=tensor([0.0497, 0.0742, 0.0716, 0.0686], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 10:43:58,051 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2988, 1.2486, 3.8249, 3.2151], device='cuda:0'), covar=tensor([0.1642, 0.2931, 0.0442, 0.1128], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0678, 0.1004, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 10:44:02,583 INFO [train.py:968] (0/2) Epoch 30, batch 17000, giga_loss[loss=0.2617, simple_loss=0.3436, pruned_loss=0.08994, over 28639.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3358, pruned_loss=0.08334, over 5686545.31 frames. ], libri_tot_loss[loss=0.245, simple_loss=0.3245, pruned_loss=0.08277, over 5760200.84 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3369, pruned_loss=0.08359, over 5677206.71 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:44:09,362 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1753, 1.3802, 1.4066, 1.0361], device='cuda:0'), covar=tensor([0.1768, 0.2707, 0.1536, 0.1786], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0713, 0.0988, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 10:44:17,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.654e+02 1.454e+03 1.969e+03 2.718e+03 1.570e+04, threshold=3.938e+03, percent-clipped=14.0
+2023-03-15 10:44:33,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1337490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:44:36,747 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1337493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:45:12,403 INFO [train.py:968] (0/2) Epoch 30, batch 17050, giga_loss[loss=0.3024, simple_loss=0.3667, pruned_loss=0.119, over 27769.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3344, pruned_loss=0.08348, over 5689716.60 frames. ], libri_tot_loss[loss=0.2448, simple_loss=0.3244, pruned_loss=0.08267, over 5761398.62 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3354, pruned_loss=0.08376, over 5680787.95 frames. ], batch size: 474, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:45:17,107 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1337522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:45:44,324 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1444, 1.3985, 3.4322, 2.9897], device='cuda:0'), covar=tensor([0.1638, 0.2597, 0.0487, 0.1011], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0679, 0.1006, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 10:45:46,863 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1337544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:45:54,916 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1337549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:46:00,746 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1337552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:46:14,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9883, 1.1480, 2.8381, 2.7814], device='cuda:0'), covar=tensor([0.1638, 0.2699, 0.0587, 0.1134], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0678, 0.1006, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 10:46:20,432 INFO [train.py:968] (0/2) Epoch 30, batch 17100, giga_loss[loss=0.2079, simple_loss=0.3101, pruned_loss=0.05285, over 28892.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3319, pruned_loss=0.08235, over 5694815.18 frames. ], libri_tot_loss[loss=0.2445, simple_loss=0.324, pruned_loss=0.08246, over 5763452.99 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3332, pruned_loss=0.08277, over 5684893.96 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:46:37,454 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.082e+03 1.671e+03 2.042e+03 3.011e+03 7.549e+03, threshold=4.084e+03, percent-clipped=13.0
+2023-03-15 10:46:40,841 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1337581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:47:30,988 INFO [train.py:968] (0/2) Epoch 30, batch 17150, libri_loss[loss=0.2491, simple_loss=0.3319, pruned_loss=0.08313, over 29753.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3306, pruned_loss=0.08079, over 5702507.61 frames. ], libri_tot_loss[loss=0.2445, simple_loss=0.324, pruned_loss=0.08246, over 5765142.67 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3317, pruned_loss=0.08109, over 5692526.25 frames. ], batch size: 87, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:48:05,404 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2892, 1.2259, 3.8789, 3.2954], device='cuda:0'), covar=tensor([0.1683, 0.3088, 0.0427, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.0811, 0.0678, 0.1005, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 10:48:24,789 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1337665.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 10:48:27,483 INFO [train.py:968] (0/2) Epoch 30, batch 17200, giga_loss[loss=0.2483, simple_loss=0.3252, pruned_loss=0.08566, over 28955.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.331, pruned_loss=0.08165, over 5693216.35 frames. ], libri_tot_loss[loss=0.245, simple_loss=0.3244, pruned_loss=0.08279, over 5766716.14 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3317, pruned_loss=0.08157, over 5682368.60 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:48:38,810 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.860e+02 1.388e+03 1.701e+03 2.457e+03 4.108e+03, threshold=3.402e+03, percent-clipped=2.0
+2023-03-15 10:49:26,527 INFO [train.py:968] (0/2) Epoch 30, batch 17250, giga_loss[loss=0.2508, simple_loss=0.3351, pruned_loss=0.08325, over 28895.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.334, pruned_loss=0.08319, over 5692381.11 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3243, pruned_loss=0.08279, over 5767567.56 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3346, pruned_loss=0.08313, over 5682918.05 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:50:18,239 INFO [train.py:968] (0/2) Epoch 30, batch 17300, libri_loss[loss=0.2268, simple_loss=0.3107, pruned_loss=0.07145, over 29541.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3334, pruned_loss=0.0832, over 5688430.52 frames. ], libri_tot_loss[loss=0.2448, simple_loss=0.3242, pruned_loss=0.08267, over 5769771.32 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3344, pruned_loss=0.0833, over 5675866.53 frames. ], batch size: 79, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:50:28,588 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.011e+03 1.522e+03 2.018e+03 2.481e+03 1.047e+04, threshold=4.035e+03, percent-clipped=12.0
+2023-03-15 10:50:31,349 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1337782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:50:56,487 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-15 10:51:04,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1337808.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 10:51:06,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1337811.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 10:51:14,384 INFO [train.py:968] (0/2) Epoch 30, batch 17350, giga_loss[loss=0.2326, simple_loss=0.3102, pruned_loss=0.07748, over 28853.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3309, pruned_loss=0.08284, over 5687836.67 frames. ], libri_tot_loss[loss=0.2446, simple_loss=0.324, pruned_loss=0.08259, over 5771815.42 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3318, pruned_loss=0.08299, over 5675317.43 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:51:41,940 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1337840.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 10:51:45,405 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1337843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:52:11,665 INFO [train.py:968] (0/2) Epoch 30, batch 17400, giga_loss[loss=0.2319, simple_loss=0.3164, pruned_loss=0.07367, over 29011.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.331, pruned_loss=0.08357, over 5677275.61 frames. ], libri_tot_loss[loss=0.2446, simple_loss=0.3242, pruned_loss=0.08255, over 5760626.11 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3318, pruned_loss=0.08374, over 5675655.87 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:52:23,830 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.549e+03 1.904e+03 2.642e+03 6.044e+03, threshold=3.808e+03, percent-clipped=5.0
+2023-03-15 10:52:30,289 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0615, 1.3183, 1.2987, 0.9999], device='cuda:0'), covar=tensor([0.1387, 0.2175, 0.1202, 0.1649], device='cuda:0'), in_proj_covar=tensor([0.0937, 0.0712, 0.0987, 0.0887], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 10:52:35,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3396, 3.3836, 1.4706, 1.5610], device='cuda:0'), covar=tensor([0.1042, 0.0378, 0.0960, 0.1359], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0571, 0.0415, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 10:53:05,717 INFO [train.py:968] (0/2) Epoch 30, batch 17450, giga_loss[loss=0.3417, simple_loss=0.4029, pruned_loss=0.1403, over 27924.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3348, pruned_loss=0.08606, over 5670943.14 frames. ], libri_tot_loss[loss=0.2443, simple_loss=0.3238, pruned_loss=0.08236, over 5752310.85 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3359, pruned_loss=0.08642, over 5676186.66 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:53:06,867 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1337919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:53:07,405 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2486, 3.1130, 2.9565, 1.4230], device='cuda:0'), covar=tensor([0.0991, 0.1041, 0.0968, 0.2380], device='cuda:0'), in_proj_covar=tensor([0.1289, 0.1184, 0.0996, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 10:53:15,717 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1337925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:53:20,231 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1337928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:53:44,114 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1337957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:53:52,810 INFO [train.py:968] (0/2) Epoch 30, batch 17500, giga_loss[loss=0.3263, simple_loss=0.397, pruned_loss=0.1278, over 28512.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3442, pruned_loss=0.09163, over 5679966.27 frames. ], libri_tot_loss[loss=0.2447, simple_loss=0.3241, pruned_loss=0.08258, over 5755559.27 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3452, pruned_loss=0.09184, over 5679512.92 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:54:03,234 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.534e+03 1.978e+03 2.890e+03 6.978e+03, threshold=3.955e+03, percent-clipped=8.0
+2023-03-15 10:54:21,278 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1338000.pt
+2023-03-15 10:54:36,735 INFO [train.py:968] (0/2) Epoch 30, batch 17550, libri_loss[loss=0.2554, simple_loss=0.3311, pruned_loss=0.08986, over 28596.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3493, pruned_loss=0.09419, over 5689403.05 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3243, pruned_loss=0.08272, over 5755260.68 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3502, pruned_loss=0.09437, over 5688501.70 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:55:11,508 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7204, 1.9303, 1.4348, 1.4121], device='cuda:0'), covar=tensor([0.1152, 0.0707, 0.1150, 0.1313], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0446, 0.0523, 0.0459], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 10:55:20,178 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1338062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:55:22,107 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1338065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:55:25,168 INFO [train.py:968] (0/2) Epoch 30, batch 17600, giga_loss[loss=0.2442, simple_loss=0.3233, pruned_loss=0.0825, over 28814.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3448, pruned_loss=0.09261, over 5691689.43 frames. ], libri_tot_loss[loss=0.2448, simple_loss=0.3243, pruned_loss=0.08263, over 5759159.27 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.346, pruned_loss=0.0931, over 5685730.41 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:55:33,820 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.427e+02 1.251e+03 1.644e+03 2.188e+03 1.165e+04, threshold=3.288e+03, percent-clipped=1.0
+2023-03-15 10:55:45,027 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1338094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:56:03,027 INFO [train.py:968] (0/2) Epoch 30, batch 17650, giga_loss[loss=0.2082, simple_loss=0.2886, pruned_loss=0.06394, over 28162.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3377, pruned_loss=0.08987, over 5692095.29 frames. ], libri_tot_loss[loss=0.2445, simple_loss=0.3242, pruned_loss=0.08246, over 5761135.17 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3395, pruned_loss=0.09074, over 5683313.03 frames. ], batch size: 77, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:56:51,901 INFO [train.py:968] (0/2) Epoch 30, batch 17700, giga_loss[loss=0.2131, simple_loss=0.2888, pruned_loss=0.06868, over 28764.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.33, pruned_loss=0.0864, over 5689135.31 frames. ], libri_tot_loss[loss=0.2446, simple_loss=0.3242, pruned_loss=0.08248, over 5762601.09 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3315, pruned_loss=0.08711, over 5680074.54 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:56:59,075 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.718e+02 1.158e+03 1.371e+03 1.867e+03 4.158e+03, threshold=2.741e+03, percent-clipped=2.0
+2023-03-15 10:57:24,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5846, 1.6564, 1.7455, 1.3917], device='cuda:0'), covar=tensor([0.1928, 0.2765, 0.1616, 0.1820], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0714, 0.0991, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 10:57:32,006 INFO [train.py:968] (0/2) Epoch 30, batch 17750, giga_loss[loss=0.2185, simple_loss=0.2867, pruned_loss=0.07512, over 28652.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3226, pruned_loss=0.08319, over 5692508.74 frames. ], libri_tot_loss[loss=0.2448, simple_loss=0.3245, pruned_loss=0.08259, over 5763997.09 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3236, pruned_loss=0.08377, over 5680820.64 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:57:32,195 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1338218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:58:01,254 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:58:11,990 INFO [train.py:968] (0/2) Epoch 30, batch 17800, giga_loss[loss=0.2569, simple_loss=0.3204, pruned_loss=0.09674, over 27678.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.317, pruned_loss=0.08099, over 5700812.55 frames. ], libri_tot_loss[loss=0.2445, simple_loss=0.3243, pruned_loss=0.08238, over 5770139.10 frames. ], giga_tot_loss[loss=0.2405, simple_loss=0.3177, pruned_loss=0.08161, over 5682791.91 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:58:21,536 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.530e+02 1.174e+03 1.551e+03 2.300e+03 5.617e+03, threshold=3.102e+03, percent-clipped=14.0
+2023-03-15 10:58:49,260 INFO [train.py:968] (0/2) Epoch 30, batch 17850, giga_loss[loss=0.2104, simple_loss=0.2833, pruned_loss=0.06872, over 28593.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3118, pruned_loss=0.07844, over 5705477.63 frames. ], libri_tot_loss[loss=0.2443, simple_loss=0.3242, pruned_loss=0.08218, over 5774710.59 frames. ], giga_tot_loss[loss=0.2351, simple_loss=0.3121, pruned_loss=0.07901, over 5684683.79 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:59:03,227 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4155, 1.4765, 1.3282, 1.5771], device='cuda:0'), covar=tensor([0.0694, 0.0437, 0.0362, 0.0805], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 10:59:17,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5912, 1.6969, 1.7857, 1.3826], device='cuda:0'), covar=tensor([0.1945, 0.2795, 0.1609, 0.1832], device='cuda:0'), in_proj_covar=tensor([0.0945, 0.0718, 0.0996, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 10:59:20,425 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:59:23,777 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1338361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:59:26,866 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1338364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:59:29,187 INFO [train.py:968] (0/2) Epoch 30, batch 17900, giga_loss[loss=0.2382, simple_loss=0.3111, pruned_loss=0.08264, over 28568.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3095, pruned_loss=0.07755, over 5708329.29 frames. ], libri_tot_loss[loss=0.2447, simple_loss=0.3246, pruned_loss=0.08238, over 5776255.00 frames. ], giga_tot_loss[loss=0.2324, simple_loss=0.3093, pruned_loss=0.07779, over 5690035.48 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:59:38,852 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.074e+02 1.210e+03 1.526e+03 1.947e+03 6.956e+03, threshold=3.052e+03, percent-clipped=3.0
+2023-03-15 10:59:52,510 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1338393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:00:08,561 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 11:00:12,474 INFO [train.py:968] (0/2) Epoch 30, batch 17950, giga_loss[loss=0.2098, simple_loss=0.2908, pruned_loss=0.06438, over 28713.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3061, pruned_loss=0.07623, over 5702377.91 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3248, pruned_loss=0.08248, over 5778319.25 frames. ], giga_tot_loss[loss=0.229, simple_loss=0.3055, pruned_loss=0.07626, over 5685147.36 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:00:54,048 INFO [train.py:968] (0/2) Epoch 30, batch 18000, giga_loss[loss=0.2132, simple_loss=0.2826, pruned_loss=0.07191, over 28763.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3024, pruned_loss=0.07436, over 5711861.71 frames. ], libri_tot_loss[loss=0.2448, simple_loss=0.3247, pruned_loss=0.08248, over 5778319.48 frames. ], giga_tot_loss[loss=0.2251, simple_loss=0.3016, pruned_loss=0.07427, over 5697291.99 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:00:54,054 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 11:01:02,016 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4053, 1.6929, 1.6392, 1.2075], device='cuda:0'), covar=tensor([0.1864, 0.3141, 0.1787, 0.2085], device='cuda:0'), in_proj_covar=tensor([0.0946, 0.0718, 0.0997, 0.0895], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 11:01:03,192 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.2004, simple_loss=0.3073, pruned_loss=0.04671, over 944034.00 frames. 
+2023-03-15 11:01:03,192 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 11:01:14,968 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.970e+02 1.118e+03 1.438e+03 1.981e+03 5.966e+03, threshold=2.876e+03, percent-clipped=9.0
+2023-03-15 11:01:17,921 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9553, 2.3185, 2.0864, 2.0664], device='cuda:0'), covar=tensor([0.2232, 0.2224, 0.2398, 0.2312], device='cuda:0'), in_proj_covar=tensor([0.0500, 0.0747, 0.0721, 0.0690], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 11:01:33,258 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8465, 4.6818, 4.4454, 1.7460], device='cuda:0'), covar=tensor([0.0602, 0.0750, 0.0977, 0.2200], device='cuda:0'), in_proj_covar=tensor([0.1296, 0.1194, 0.1001, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 11:01:46,132 INFO [train.py:968] (0/2) Epoch 30, batch 18050, giga_loss[loss=0.2047, simple_loss=0.2762, pruned_loss=0.06661, over 27711.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3006, pruned_loss=0.07368, over 5688775.33 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3249, pruned_loss=0.08245, over 5768904.47 frames. ], giga_tot_loss[loss=0.223, simple_loss=0.2992, pruned_loss=0.07341, over 5683593.93 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:02:27,021 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4097, 1.6798, 1.6599, 1.4796], device='cuda:0'), covar=tensor([0.2313, 0.2268, 0.2722, 0.2417], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0749, 0.0722, 0.0691], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 11:02:27,401 INFO [train.py:968] (0/2) Epoch 30, batch 18100, giga_loss[loss=0.1996, simple_loss=0.2779, pruned_loss=0.06062, over 28932.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2974, pruned_loss=0.07237, over 5692468.86 frames. ], libri_tot_loss[loss=0.2444, simple_loss=0.3244, pruned_loss=0.08219, over 5770428.58 frames. ], giga_tot_loss[loss=0.2204, simple_loss=0.2963, pruned_loss=0.07223, over 5685708.50 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:02:27,589 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:02:38,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.670e+02 1.216e+03 1.622e+03 2.200e+03 9.127e+03, threshold=3.244e+03, percent-clipped=14.0
+2023-03-15 11:03:08,569 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3327, 1.2626, 3.6856, 3.2676], device='cuda:0'), covar=tensor([0.1662, 0.2942, 0.0497, 0.1338], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0677, 0.1012, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 11:03:08,958 INFO [train.py:968] (0/2) Epoch 30, batch 18150, giga_loss[loss=0.1846, simple_loss=0.265, pruned_loss=0.0521, over 28626.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2953, pruned_loss=0.07176, over 5685011.42 frames. ], libri_tot_loss[loss=0.2447, simple_loss=0.3248, pruned_loss=0.08227, over 5759263.85 frames. ], giga_tot_loss[loss=0.2182, simple_loss=0.2936, pruned_loss=0.07139, over 5687441.40 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:03:19,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1338629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:03:38,914 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-15 11:03:54,795 INFO [train.py:968] (0/2) Epoch 30, batch 18200, libri_loss[loss=0.2968, simple_loss=0.3731, pruned_loss=0.1102, over 29659.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2942, pruned_loss=0.07178, over 5678508.85 frames. ], libri_tot_loss[loss=0.2452, simple_loss=0.3254, pruned_loss=0.0825, over 5763025.69 frames. ], giga_tot_loss[loss=0.2168, simple_loss=0.2915, pruned_loss=0.07101, over 5675375.07 frames. ], batch size: 91, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 11:04:05,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.352e+02 1.135e+03 1.370e+03 1.905e+03 8.071e+03, threshold=2.740e+03, percent-clipped=6.0
+2023-03-15 11:04:28,493 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3175, 2.9497, 1.4249, 1.4698], device='cuda:0'), covar=tensor([0.1061, 0.0388, 0.0959, 0.1415], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0572, 0.0415, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 11:04:34,833 INFO [train.py:968] (0/2) Epoch 30, batch 18250, giga_loss[loss=0.2428, simple_loss=0.2992, pruned_loss=0.09325, over 23890.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2913, pruned_loss=0.0706, over 5681783.76 frames. ], libri_tot_loss[loss=0.2452, simple_loss=0.3255, pruned_loss=0.08245, over 5764492.63 frames. ], giga_tot_loss[loss=0.2144, simple_loss=0.2889, pruned_loss=0.06994, over 5677313.88 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 11:04:41,080 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:04:44,201 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4343, 1.5352, 1.6168, 1.3200], device='cuda:0'), covar=tensor([0.1419, 0.2114, 0.1198, 0.1548], device='cuda:0'), in_proj_covar=tensor([0.0949, 0.0720, 0.1000, 0.0898], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 11:04:47,600 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:04:47,630 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1338732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:05:25,452 INFO [train.py:968] (0/2) Epoch 30, batch 18300, giga_loss[loss=0.2675, simple_loss=0.344, pruned_loss=0.09551, over 28604.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3001, pruned_loss=0.07527, over 5665908.32 frames. ], libri_tot_loss[loss=0.2451, simple_loss=0.3254, pruned_loss=0.0824, over 5755607.04 frames. ], giga_tot_loss[loss=0.2238, simple_loss=0.2981, pruned_loss=0.07474, over 5669302.42 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 11:05:29,308 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1338772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:05:31,857 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1338775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:05:39,377 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.239e+02 1.296e+03 1.648e+03 2.013e+03 5.321e+03, threshold=3.297e+03, percent-clipped=8.0
+2023-03-15 11:05:41,038 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:05:57,665 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1338804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:06:10,946 INFO [train.py:968] (0/2) Epoch 30, batch 18350, giga_loss[loss=0.2844, simple_loss=0.3675, pruned_loss=0.1007, over 28979.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.314, pruned_loss=0.08218, over 5675480.19 frames. ], libri_tot_loss[loss=0.2454, simple_loss=0.3256, pruned_loss=0.08263, over 5757690.89 frames. ], giga_tot_loss[loss=0.2375, simple_loss=0.3119, pruned_loss=0.0815, over 5675036.90 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 11:06:48,168 INFO [train.py:968] (0/2) Epoch 30, batch 18400, giga_loss[loss=0.2808, simple_loss=0.3398, pruned_loss=0.1108, over 23551.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3257, pruned_loss=0.08778, over 5687447.93 frames. ], libri_tot_loss[loss=0.246, simple_loss=0.3262, pruned_loss=0.08288, over 5761500.76 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3234, pruned_loss=0.08707, over 5681707.49 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:06:54,177 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1338875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:06:56,803 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1338878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:07:00,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.006e+03 1.519e+03 1.868e+03 2.493e+03 8.244e+03, threshold=3.737e+03, percent-clipped=10.0
+2023-03-15 11:07:12,341 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3060, 1.3278, 3.7803, 3.0934], device='cuda:0'), covar=tensor([0.1690, 0.2841, 0.0491, 0.1051], device='cuda:0'), in_proj_covar=tensor([0.0809, 0.0676, 0.1011, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 11:07:20,583 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1338907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:07:28,506 INFO [train.py:968] (0/2) Epoch 30, batch 18450, giga_loss[loss=0.2559, simple_loss=0.3394, pruned_loss=0.08618, over 28995.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3325, pruned_loss=0.09033, over 5689664.97 frames. ], libri_tot_loss[loss=0.246, simple_loss=0.3263, pruned_loss=0.08286, over 5763632.56 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3306, pruned_loss=0.08987, over 5682396.53 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:07:51,605 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1338943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:08:09,568 INFO [train.py:968] (0/2) Epoch 30, batch 18500, giga_loss[loss=0.2432, simple_loss=0.3335, pruned_loss=0.07647, over 28743.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3355, pruned_loss=0.09062, over 5690192.96 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3265, pruned_loss=0.08298, over 5763776.96 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.334, pruned_loss=0.09029, over 5683077.76 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:08:14,936 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8280, 1.1434, 2.8538, 2.6730], device='cuda:0'), covar=tensor([0.1770, 0.2767, 0.0598, 0.1394], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0674, 0.1007, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 11:08:22,026 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.397e+03 1.637e+03 2.129e+03 5.215e+03, threshold=3.274e+03, percent-clipped=3.0
+2023-03-15 11:08:54,767 INFO [train.py:968] (0/2) Epoch 30, batch 18550, giga_loss[loss=0.3043, simple_loss=0.3755, pruned_loss=0.1165, over 28564.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3375, pruned_loss=0.0911, over 5665546.02 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3267, pruned_loss=0.08307, over 5756259.95 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3362, pruned_loss=0.09091, over 5665194.05 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:09:38,636 INFO [train.py:968] (0/2) Epoch 30, batch 18600, giga_loss[loss=0.2664, simple_loss=0.3404, pruned_loss=0.09626, over 28484.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3408, pruned_loss=0.09383, over 5670108.40 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3268, pruned_loss=0.0831, over 5757824.03 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3399, pruned_loss=0.09381, over 5667140.11 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:09:44,839 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3743, 1.2230, 1.3241, 1.5499], device='cuda:0'), covar=tensor([0.0827, 0.0387, 0.0345, 0.0913], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 11:09:49,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.843e+02 1.339e+03 1.589e+03 1.955e+03 3.431e+03, threshold=3.178e+03, percent-clipped=1.0
+2023-03-15 11:09:53,216 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2711, 1.4178, 3.5534, 3.1084], device='cuda:0'), covar=tensor([0.1646, 0.2825, 0.0465, 0.1657], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0675, 0.1006, 0.0979], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 11:09:53,823 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1339086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:09:55,880 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1339089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:10:03,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1339099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:10:10,852 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1339107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:10:13,998 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2132, 1.4743, 1.4752, 1.0959], device='cuda:0'), covar=tensor([0.1653, 0.2478, 0.1353, 0.1677], device='cuda:0'), in_proj_covar=tensor([0.0948, 0.0721, 0.0998, 0.0897], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 11:10:21,821 INFO [train.py:968] (0/2) Epoch 30, batch 18650, giga_loss[loss=0.3246, simple_loss=0.385, pruned_loss=0.1321, over 28227.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3441, pruned_loss=0.09634, over 5671730.34 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3271, pruned_loss=0.08319, over 5758932.78 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3434, pruned_loss=0.09641, over 5667229.66 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:10:22,010 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1339118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:10:55,517 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1339159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:11:01,054 INFO [train.py:968] (0/2) Epoch 30, batch 18700, giga_loss[loss=0.2684, simple_loss=0.3426, pruned_loss=0.09704, over 28593.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3465, pruned_loss=0.09716, over 5673573.38 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3277, pruned_loss=0.08338, over 5754268.29 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.346, pruned_loss=0.09757, over 5670365.57 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:11:11,974 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.290e+03 1.659e+03 2.209e+03 7.898e+03, threshold=3.319e+03, percent-clipped=7.0
+2023-03-15 11:11:40,693 INFO [train.py:968] (0/2) Epoch 30, batch 18750, giga_loss[loss=0.2648, simple_loss=0.3527, pruned_loss=0.08847, over 28938.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3495, pruned_loss=0.09775, over 5666360.49 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3283, pruned_loss=0.08359, over 5739268.51 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3491, pruned_loss=0.09825, over 5674824.42 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:12:01,099 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1339242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:03,212 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1339245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:10,463 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1339250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:11,214 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0603, 1.2108, 3.2056, 2.9307], device='cuda:0'), covar=tensor([0.2129, 0.3184, 0.1001, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0807, 0.0674, 0.1006, 0.0980], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 11:12:12,473 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1339253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:25,028 INFO [train.py:968] (0/2) Epoch 30, batch 18800, giga_loss[loss=0.2766, simple_loss=0.3582, pruned_loss=0.09752, over 28881.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3522, pruned_loss=0.09897, over 5667829.50 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3285, pruned_loss=0.08368, over 5737916.81 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3519, pruned_loss=0.09942, over 5674961.03 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:12:29,941 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1339274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:35,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.146e+02 1.307e+03 1.623e+03 2.055e+03 4.617e+03, threshold=3.246e+03, percent-clipped=2.0
+2023-03-15 11:12:35,672 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1339282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:45,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3840, 1.5157, 1.5941, 1.2327], device='cuda:0'), covar=tensor([0.1678, 0.2480, 0.1427, 0.1694], device='cuda:0'), in_proj_covar=tensor([0.0947, 0.0720, 0.0996, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 11:12:51,747 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1339302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:54,840 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1339305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:13:07,382 INFO [train.py:968] (0/2) Epoch 30, batch 18850, giga_loss[loss=0.3043, simple_loss=0.3682, pruned_loss=0.1202, over 26632.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3523, pruned_loss=0.09776, over 5674711.02 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3286, pruned_loss=0.08369, over 5739247.44 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3522, pruned_loss=0.09825, over 5678465.93 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:13:09,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5162, 1.8416, 1.5911, 1.6868], device='cuda:0'), covar=tensor([0.0824, 0.0316, 0.0331, 0.0915], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 11:13:19,853 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1339334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:13:39,944 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1339362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:13:44,704 INFO [train.py:968] (0/2) Epoch 30, batch 18900, giga_loss[loss=0.2673, simple_loss=0.3472, pruned_loss=0.09368, over 28891.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.352, pruned_loss=0.09628, over 5684743.71 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3291, pruned_loss=0.08381, over 5732962.72 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.352, pruned_loss=0.09686, over 5692115.50 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:13:56,404 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.978e+02 1.271e+03 1.496e+03 1.871e+03 3.792e+03, threshold=2.992e+03, percent-clipped=1.0
+2023-03-15 11:14:24,933 INFO [train.py:968] (0/2) Epoch 30, batch 18950, giga_loss[loss=0.2321, simple_loss=0.3251, pruned_loss=0.06954, over 29043.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3496, pruned_loss=0.09355, over 5694099.57 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3292, pruned_loss=0.08376, over 5734676.74 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3497, pruned_loss=0.09414, over 5698043.46 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:15:05,393 INFO [train.py:968] (0/2) Epoch 30, batch 19000, libri_loss[loss=0.2605, simple_loss=0.3489, pruned_loss=0.08605, over 29358.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3509, pruned_loss=0.09503, over 5691997.42 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3296, pruned_loss=0.08404, over 5727470.56 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3509, pruned_loss=0.09537, over 5700792.62 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:15:17,878 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.648e+02 1.311e+03 1.550e+03 2.049e+03 5.983e+03, threshold=3.101e+03, percent-clipped=11.0
+2023-03-15 11:15:29,634 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1339496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:15:48,735 INFO [train.py:968] (0/2) Epoch 30, batch 19050, giga_loss[loss=0.2951, simple_loss=0.3607, pruned_loss=0.1148, over 28866.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3536, pruned_loss=0.09946, over 5699840.02 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3299, pruned_loss=0.08406, over 5728654.16 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3537, pruned_loss=0.09989, over 5705233.86 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:16:30,934 INFO [train.py:968] (0/2) Epoch 30, batch 19100, giga_loss[loss=0.3079, simple_loss=0.3751, pruned_loss=0.1203, over 29081.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3541, pruned_loss=0.1018, over 5703522.24 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3298, pruned_loss=0.08394, over 5728509.78 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3551, pruned_loss=0.1029, over 5706876.61 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:16:42,605 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.104e+03 1.665e+03 2.101e+03 2.789e+03 5.060e+03, threshold=4.201e+03, percent-clipped=19.0
+2023-03-15 11:17:11,786 INFO [train.py:968] (0/2) Epoch 30, batch 19150, giga_loss[loss=0.2939, simple_loss=0.3463, pruned_loss=0.1207, over 23837.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3533, pruned_loss=0.1025, over 5686950.88 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.33, pruned_loss=0.08405, over 5721331.83 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3541, pruned_loss=0.1035, over 5695530.76 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:17:30,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5789, 1.7114, 1.2515, 1.3020], device='cuda:0'), covar=tensor([0.1073, 0.0618, 0.1059, 0.1201], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0449, 0.0527, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 11:17:31,360 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1339643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:17:51,954 INFO [train.py:968] (0/2) Epoch 30, batch 19200, giga_loss[loss=0.2354, simple_loss=0.315, pruned_loss=0.07794, over 28440.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3504, pruned_loss=0.1013, over 5693778.23 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3307, pruned_loss=0.08438, over 5726435.26 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3512, pruned_loss=0.1024, over 5695090.36 frames. ], batch size: 65, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:18:03,586 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.294e+02 1.401e+03 1.703e+03 2.361e+03 5.865e+03, threshold=3.407e+03, percent-clipped=3.0
+2023-03-15 11:18:37,263 INFO [train.py:968] (0/2) Epoch 30, batch 19250, giga_loss[loss=0.2584, simple_loss=0.3367, pruned_loss=0.09007, over 28927.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.349, pruned_loss=0.1001, over 5692964.18 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3309, pruned_loss=0.0844, over 5714940.90 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3496, pruned_loss=0.1011, over 5703391.05 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:18:52,363 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1339737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:19:16,704 INFO [train.py:968] (0/2) Epoch 30, batch 19300, giga_loss[loss=0.2334, simple_loss=0.3128, pruned_loss=0.07704, over 28284.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3472, pruned_loss=0.09799, over 5692008.23 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3311, pruned_loss=0.08439, over 5707487.82 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3479, pruned_loss=0.09912, over 5705913.39 frames. ], batch size: 77, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:19:28,009 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.147e+02 1.322e+03 1.569e+03 2.082e+03 7.448e+03, threshold=3.138e+03, percent-clipped=4.0
+2023-03-15 11:19:44,825 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1339799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:19:45,938 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-15 11:19:59,982 INFO [train.py:968] (0/2) Epoch 30, batch 19350, giga_loss[loss=0.2327, simple_loss=0.3139, pruned_loss=0.07575, over 28705.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3444, pruned_loss=0.09603, over 5681889.28 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3314, pruned_loss=0.08445, over 5710876.61 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.345, pruned_loss=0.09718, over 5689288.60 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:20:12,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6670, 2.2017, 1.3949, 0.9455], device='cuda:0'), covar=tensor([0.9068, 0.4250, 0.4523, 0.8036], device='cuda:0'), in_proj_covar=tensor([0.1859, 0.1745, 0.1670, 0.1522], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 11:20:45,056 INFO [train.py:968] (0/2) Epoch 30, batch 19400, giga_loss[loss=0.2627, simple_loss=0.3236, pruned_loss=0.1009, over 26577.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3388, pruned_loss=0.09295, over 5684976.44 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3315, pruned_loss=0.08445, over 5711174.59 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3394, pruned_loss=0.09407, over 5689936.22 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:20:49,153 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1339871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:20:58,722 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1339880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:21:00,337 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.735e+02 1.155e+03 1.619e+03 2.007e+03 5.870e+03, threshold=3.237e+03, percent-clipped=11.0
+2023-03-15 11:21:00,669 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1339883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:21:25,518 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1339910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:21:27,821 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1339912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:21:31,689 INFO [train.py:968] (0/2) Epoch 30, batch 19450, libri_loss[loss=0.2871, simple_loss=0.3771, pruned_loss=0.09856, over 29521.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3339, pruned_loss=0.0908, over 5678556.39 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3317, pruned_loss=0.08444, over 5715072.67 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3343, pruned_loss=0.09182, over 5678286.62 frames. ], batch size: 89, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:21:48,357 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9748, 1.1841, 1.1690, 0.9110], device='cuda:0'), covar=tensor([0.2430, 0.2713, 0.1645, 0.2328], device='cuda:0'), in_proj_covar=tensor([0.2075, 0.2037, 0.1939, 0.2092], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 11:22:18,839 INFO [train.py:968] (0/2) Epoch 30, batch 19500, giga_loss[loss=0.2631, simple_loss=0.3388, pruned_loss=0.09368, over 28802.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3292, pruned_loss=0.08877, over 5658422.21 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3319, pruned_loss=0.08455, over 5716346.16 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3293, pruned_loss=0.08962, over 5655896.63 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:22:32,985 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.799e+02 1.157e+03 1.488e+03 1.914e+03 6.476e+03, threshold=2.976e+03, percent-clipped=5.0
+2023-03-15 11:22:47,994 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1340000.pt
+2023-03-15 11:22:59,511 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:02,524 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:02,988 INFO [train.py:968] (0/2) Epoch 30, batch 19550, giga_loss[loss=0.2266, simple_loss=0.3106, pruned_loss=0.07135, over 28515.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3301, pruned_loss=0.08905, over 5662243.88 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3322, pruned_loss=0.08474, over 5720275.60 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3298, pruned_loss=0.08965, over 5655244.82 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:23:03,252 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:27,487 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:31,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.5209, 4.3741, 4.1895, 1.9958], device='cuda:0'), covar=tensor([0.0684, 0.0793, 0.0881, 0.1946], device='cuda:0'), in_proj_covar=tensor([0.1294, 0.1194, 0.1000, 0.0746], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 11:23:43,530 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1340063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:48,470 INFO [train.py:968] (0/2) Epoch 30, batch 19600, giga_loss[loss=0.252, simple_loss=0.3342, pruned_loss=0.08493, over 28717.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3311, pruned_loss=0.08977, over 5653563.16 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3324, pruned_loss=0.08481, over 5712068.39 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3308, pruned_loss=0.09022, over 5654426.85 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:23:54,375 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1340076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:24:01,105 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.986e+02 1.239e+03 1.532e+03 2.234e+03 3.946e+03, threshold=3.064e+03, percent-clipped=6.0
+2023-03-15 11:24:28,316 INFO [train.py:968] (0/2) Epoch 30, batch 19650, giga_loss[loss=0.2204, simple_loss=0.3029, pruned_loss=0.06896, over 28921.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3304, pruned_loss=0.08944, over 5662361.17 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3328, pruned_loss=0.08486, over 5705308.92 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3298, pruned_loss=0.08984, over 5667651.90 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:25:00,581 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4451, 4.4180, 1.7305, 1.7444], device='cuda:0'), covar=tensor([0.1114, 0.0297, 0.0909, 0.1399], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0569, 0.0414, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 11:25:03,435 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:25:06,635 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:25:09,819 INFO [train.py:968] (0/2) Epoch 30, batch 19700, giga_loss[loss=0.2221, simple_loss=0.297, pruned_loss=0.07364, over 28237.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3288, pruned_loss=0.08862, over 5671650.26 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3328, pruned_loss=0.08471, over 5707390.47 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3283, pruned_loss=0.08913, over 5673421.85 frames. ], batch size: 77, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:25:14,140 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:25:22,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.521e+02 1.219e+03 1.441e+03 1.722e+03 4.415e+03, threshold=2.882e+03, percent-clipped=5.0
+2023-03-15 11:25:30,010 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:25:38,748 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1340204.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 11:25:42,691 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1340210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:25:49,463 INFO [train.py:968] (0/2) Epoch 30, batch 19750, giga_loss[loss=0.3032, simple_loss=0.3706, pruned_loss=0.1179, over 27965.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3277, pruned_loss=0.08778, over 5677279.92 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.333, pruned_loss=0.08455, over 5703436.20 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3269, pruned_loss=0.08845, over 5681615.26 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:25:55,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2541, 1.7618, 1.6873, 1.5335], device='cuda:0'), covar=tensor([0.2310, 0.1696, 0.2246, 0.1949], device='cuda:0'), in_proj_covar=tensor([0.0513, 0.0761, 0.0734, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 11:26:18,633 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.34 vs. limit=5.0
+2023-03-15 11:26:33,653 INFO [train.py:968] (0/2) Epoch 30, batch 19800, giga_loss[loss=0.2557, simple_loss=0.3259, pruned_loss=0.09277, over 24130.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3258, pruned_loss=0.08677, over 5681187.93 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3337, pruned_loss=0.08485, over 5701876.41 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3244, pruned_loss=0.08706, over 5686038.33 frames. ], batch size: 710, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:26:49,217 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.119e+02 1.122e+03 1.397e+03 1.860e+03 4.355e+03, threshold=2.794e+03, percent-clipped=9.0
+2023-03-15 11:26:49,471 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:27:13,980 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0566, 1.2501, 3.1906, 2.8823], device='cuda:0'), covar=tensor([0.1651, 0.2753, 0.0447, 0.2193], device='cuda:0'), in_proj_covar=tensor([0.0810, 0.0674, 0.1007, 0.0981], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 11:27:16,073 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:27:16,445 INFO [train.py:968] (0/2) Epoch 30, batch 19850, giga_loss[loss=0.2193, simple_loss=0.3033, pruned_loss=0.06761, over 29000.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3232, pruned_loss=0.08574, over 5689997.59 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3341, pruned_loss=0.08485, over 5700464.61 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3217, pruned_loss=0.08599, over 5694664.22 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:27:17,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:27:41,720 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:27:56,904 INFO [train.py:968] (0/2) Epoch 30, batch 19900, giga_loss[loss=0.2313, simple_loss=0.3127, pruned_loss=0.07492, over 28645.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3198, pruned_loss=0.08413, over 5704229.59 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3342, pruned_loss=0.08473, over 5704515.97 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3183, pruned_loss=0.08445, over 5704339.70 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:28:12,792 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.821e+02 1.162e+03 1.467e+03 1.873e+03 5.512e+03, threshold=2.933e+03, percent-clipped=6.0
+2023-03-15 11:28:41,542 INFO [train.py:968] (0/2) Epoch 30, batch 19950, giga_loss[loss=0.2293, simple_loss=0.302, pruned_loss=0.07833, over 28858.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3181, pruned_loss=0.08327, over 5712008.99 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3344, pruned_loss=0.08468, over 5709540.85 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3163, pruned_loss=0.08354, over 5707600.87 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:28:43,599 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1020, 1.0900, 3.3721, 3.0297], device='cuda:0'), covar=tensor([0.1765, 0.2926, 0.0491, 0.0994], device='cuda:0'), in_proj_covar=tensor([0.0808, 0.0672, 0.1005, 0.0978], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 11:28:49,171 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:28:51,037 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:28:58,337 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:29:10,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:29:17,302 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:29:23,539 INFO [train.py:968] (0/2) Epoch 30, batch 20000, giga_loss[loss=0.2056, simple_loss=0.2883, pruned_loss=0.06146, over 28884.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3179, pruned_loss=0.08315, over 5704446.12 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3353, pruned_loss=0.085, over 5701307.33 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3152, pruned_loss=0.08303, over 5709047.16 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:29:35,448 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.365e+02 1.139e+03 1.419e+03 1.881e+03 6.354e+03, threshold=2.837e+03, percent-clipped=11.0
+2023-03-15 11:29:38,587 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-15 11:30:02,071 INFO [train.py:968] (0/2) Epoch 30, batch 20050, giga_loss[loss=0.2321, simple_loss=0.3104, pruned_loss=0.07688, over 28874.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3181, pruned_loss=0.08297, over 5704969.78 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3361, pruned_loss=0.08534, over 5698437.73 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.3147, pruned_loss=0.08254, over 5711250.10 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:30:16,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7499, 1.9994, 1.8350, 1.7056], device='cuda:0'), covar=tensor([0.2655, 0.2861, 0.2766, 0.2995], device='cuda:0'), in_proj_covar=tensor([0.0514, 0.0761, 0.0735, 0.0702], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 11:30:34,991 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-15 11:30:43,702 INFO [train.py:968] (0/2) Epoch 30, batch 20100, giga_loss[loss=0.2372, simple_loss=0.3167, pruned_loss=0.07882, over 28973.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3189, pruned_loss=0.08317, over 5714071.81 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3362, pruned_loss=0.08506, over 5705435.95 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3156, pruned_loss=0.08299, over 5713349.25 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:30:53,388 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340579.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 11:30:55,141 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:30:57,255 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:30:58,794 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:30:59,187 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.426e+02 1.130e+03 1.447e+03 1.781e+03 8.129e+03, threshold=2.895e+03, percent-clipped=10.0
+2023-03-15 11:31:04,706 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:31:06,700 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:31:23,250 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:31:28,922 INFO [train.py:968] (0/2) Epoch 30, batch 20150, giga_loss[loss=0.2967, simple_loss=0.3617, pruned_loss=0.1159, over 28893.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3249, pruned_loss=0.08688, over 5712826.79 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3369, pruned_loss=0.08525, over 5707750.56 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3212, pruned_loss=0.08655, over 5710573.89 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:31:36,180 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:31:41,428 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-15 11:32:05,740 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6374, 1.8557, 1.3264, 1.3856], device='cuda:0'), covar=tensor([0.1120, 0.0611, 0.1064, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0451, 0.0527, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 11:32:19,288 INFO [train.py:968] (0/2) Epoch 30, batch 20200, giga_loss[loss=0.267, simple_loss=0.3345, pruned_loss=0.09978, over 28810.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3301, pruned_loss=0.09054, over 5698998.68 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3368, pruned_loss=0.08501, over 5711629.48 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3272, pruned_loss=0.09058, over 5693579.90 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:32:38,661 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.486e+03 1.771e+03 2.320e+03 5.576e+03, threshold=3.543e+03, percent-clipped=10.0
+2023-03-15 11:32:41,437 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.0837, 3.2632, 2.3322, 1.2329], device='cuda:0'), covar=tensor([0.9014, 0.3217, 0.3846, 0.7472], device='cuda:0'), in_proj_covar=tensor([0.1858, 0.1740, 0.1669, 0.1518], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 11:33:08,149 INFO [train.py:968] (0/2) Epoch 30, batch 20250, giga_loss[loss=0.249, simple_loss=0.3351, pruned_loss=0.08139, over 29007.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3379, pruned_loss=0.09548, over 5697365.30 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3371, pruned_loss=0.08508, over 5714414.10 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3352, pruned_loss=0.0958, over 5689853.46 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:33:12,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340722.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 11:33:14,171 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340725.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 11:33:17,873 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:33:20,702 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:33:40,695 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340754.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 11:33:47,075 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:33:52,093 INFO [train.py:968] (0/2) Epoch 30, batch 20300, giga_loss[loss=0.2561, simple_loss=0.3374, pruned_loss=0.08735, over 28840.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3416, pruned_loss=0.09686, over 5703534.45 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3365, pruned_loss=0.08481, over 5721692.91 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.34, pruned_loss=0.09778, over 5690280.48 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:34:08,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6475, 1.7705, 1.5221, 1.9394], device='cuda:0'), covar=tensor([0.2380, 0.2535, 0.2630, 0.2307], device='cuda:0'), in_proj_covar=tensor([0.1618, 0.1166, 0.1432, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 11:34:10,503 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.980e+02 1.310e+03 1.723e+03 2.212e+03 4.888e+03, threshold=3.445e+03, percent-clipped=7.0
+2023-03-15 11:34:37,293 INFO [train.py:968] (0/2) Epoch 30, batch 20350, giga_loss[loss=0.2615, simple_loss=0.3459, pruned_loss=0.08859, over 28658.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3466, pruned_loss=0.0988, over 5700419.60 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3369, pruned_loss=0.08489, over 5725692.82 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3452, pruned_loss=0.09984, over 5685363.19 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:35:04,254 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1340848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:35:23,001 INFO [train.py:968] (0/2) Epoch 30, batch 20400, libri_loss[loss=0.2216, simple_loss=0.3084, pruned_loss=0.06733, over 29590.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3502, pruned_loss=0.1001, over 5701239.68 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.337, pruned_loss=0.08493, over 5732493.60 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3495, pruned_loss=0.1015, over 5681679.19 frames. ], batch size: 74, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:35:39,971 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.635e+02 1.346e+03 1.666e+03 2.299e+03 5.260e+03, threshold=3.332e+03, percent-clipped=6.0
+2023-03-15 11:35:49,826 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6399, 1.8381, 1.6910, 1.7672], device='cuda:0'), covar=tensor([0.0773, 0.0300, 0.0313, 0.0865], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 11:36:07,988 INFO [train.py:968] (0/2) Epoch 30, batch 20450, giga_loss[loss=0.264, simple_loss=0.343, pruned_loss=0.09255, over 28871.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3535, pruned_loss=0.1018, over 5707610.47 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3368, pruned_loss=0.08485, over 5736241.77 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3535, pruned_loss=0.1035, over 5687951.57 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:36:50,580 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0580, 1.5476, 1.6033, 1.3626], device='cuda:0'), covar=tensor([0.2489, 0.1615, 0.2353, 0.1883], device='cuda:0'), in_proj_covar=tensor([0.0513, 0.0761, 0.0735, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 11:36:54,682 INFO [train.py:968] (0/2) Epoch 30, batch 20500, giga_loss[loss=0.247, simple_loss=0.3341, pruned_loss=0.07993, over 28954.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3495, pruned_loss=0.09914, over 5697343.99 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3369, pruned_loss=0.08508, over 5730027.75 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3496, pruned_loss=0.1005, over 5685840.25 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:37:10,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.055e+03 1.407e+03 1.656e+03 2.301e+03 8.901e+03, threshold=3.313e+03, percent-clipped=9.0
+2023-03-15 11:37:33,162 INFO [train.py:968] (0/2) Epoch 30, batch 20550, giga_loss[loss=0.2787, simple_loss=0.3536, pruned_loss=0.1019, over 28346.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3477, pruned_loss=0.09719, over 5705479.86 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.338, pruned_loss=0.08614, over 5729715.59 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3474, pruned_loss=0.09801, over 5694749.56 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:37:40,984 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2228, 2.2170, 1.7914, 1.4647], device='cuda:0'), covar=tensor([0.0869, 0.0300, 0.0294, 0.1070], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 11:38:20,341 INFO [train.py:968] (0/2) Epoch 30, batch 20600, giga_loss[loss=0.2327, simple_loss=0.3243, pruned_loss=0.07059, over 28914.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3471, pruned_loss=0.09682, over 5695392.43 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3383, pruned_loss=0.08633, over 5732847.75 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3468, pruned_loss=0.0975, over 5683564.16 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:38:38,021 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.481e+02 1.351e+03 1.736e+03 2.314e+03 5.934e+03, threshold=3.471e+03, percent-clipped=8.0
+2023-03-15 11:39:06,136 INFO [train.py:968] (0/2) Epoch 30, batch 20650, giga_loss[loss=0.2632, simple_loss=0.3454, pruned_loss=0.09045, over 29011.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3472, pruned_loss=0.09628, over 5692170.07 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3385, pruned_loss=0.08646, over 5725453.10 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3469, pruned_loss=0.09682, over 5688203.69 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:39:51,742 INFO [train.py:968] (0/2) Epoch 30, batch 20700, libri_loss[loss=0.2585, simple_loss=0.3466, pruned_loss=0.08523, over 29472.00 frames. ], tot_loss[loss=0.272, simple_loss=0.349, pruned_loss=0.09749, over 5694774.78 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3384, pruned_loss=0.08637, over 5728148.35 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.349, pruned_loss=0.09816, over 5688563.57 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:40:10,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.608e+02 1.390e+03 1.970e+03 2.840e+03 5.491e+03, threshold=3.939e+03, percent-clipped=11.0
+2023-03-15 11:40:22,785 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5313, 2.1809, 1.6497, 0.8030], device='cuda:0'), covar=tensor([0.7367, 0.3340, 0.4559, 0.7362], device='cuda:0'), in_proj_covar=tensor([0.1860, 0.1739, 0.1671, 0.1517], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 11:40:31,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2331, 1.4579, 1.4599, 1.1087], device='cuda:0'), covar=tensor([0.1659, 0.2503, 0.1364, 0.1616], device='cuda:0'), in_proj_covar=tensor([0.0944, 0.0719, 0.0993, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 11:40:34,719 INFO [train.py:968] (0/2) Epoch 30, batch 20750, giga_loss[loss=0.237, simple_loss=0.32, pruned_loss=0.07698, over 28556.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3501, pruned_loss=0.09863, over 5698661.72 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3389, pruned_loss=0.08674, over 5731203.89 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3502, pruned_loss=0.09935, over 5689140.13 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:40:38,648 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1341223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:41:01,755 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.7816, 3.6317, 3.4011, 1.8301], device='cuda:0'), covar=tensor([0.0698, 0.0803, 0.0763, 0.2466], device='cuda:0'), in_proj_covar=tensor([0.1300, 0.1200, 0.1005, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 11:41:22,458 INFO [train.py:968] (0/2) Epoch 30, batch 20800, giga_loss[loss=0.3087, simple_loss=0.3734, pruned_loss=0.122, over 28667.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3501, pruned_loss=0.09863, over 5710505.14 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3387, pruned_loss=0.08671, over 5733514.30 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3504, pruned_loss=0.09937, over 5700625.32 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:41:37,616 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.931e+02 1.430e+03 1.741e+03 2.271e+03 5.771e+03, threshold=3.483e+03, percent-clipped=6.0
+2023-03-15 11:42:07,525 INFO [train.py:968] (0/2) Epoch 30, batch 20850, giga_loss[loss=0.2821, simple_loss=0.3567, pruned_loss=0.1037, over 29023.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.352, pruned_loss=0.1005, over 5697422.88 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3392, pruned_loss=0.0871, over 5723495.33 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3521, pruned_loss=0.1011, over 5697906.75 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:42:47,662 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1341366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:42:48,602 INFO [train.py:968] (0/2) Epoch 30, batch 20900, giga_loss[loss=0.2755, simple_loss=0.3588, pruned_loss=0.0961, over 29007.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3525, pruned_loss=0.1009, over 5707829.44 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3397, pruned_loss=0.08749, over 5728911.41 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3525, pruned_loss=0.1015, over 5702543.17 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:42:49,672 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1341369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:43:02,133 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2931, 1.3503, 3.7394, 3.2335], device='cuda:0'), covar=tensor([0.1725, 0.2869, 0.0459, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0676, 0.1008, 0.0983], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 11:43:03,993 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.382e+03 1.748e+03 2.458e+03 9.311e+03, threshold=3.496e+03, percent-clipped=9.0
+2023-03-15 11:43:12,032 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1341398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:43:29,918 INFO [train.py:968] (0/2) Epoch 30, batch 20950, giga_loss[loss=0.2293, simple_loss=0.3178, pruned_loss=0.07038, over 28397.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3516, pruned_loss=0.09948, over 5707362.54 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3399, pruned_loss=0.08756, over 5731347.35 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3516, pruned_loss=0.1, over 5700809.58 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:43:39,371 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 11:44:03,120 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1341457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:44:12,843 INFO [train.py:968] (0/2) Epoch 30, batch 21000, giga_loss[loss=0.2773, simple_loss=0.3622, pruned_loss=0.09619, over 28898.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3528, pruned_loss=0.09934, over 5694630.21 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3404, pruned_loss=0.08793, over 5711879.68 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3526, pruned_loss=0.09959, over 5707076.43 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:44:12,847 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 11:44:23,034 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.2039, simple_loss=0.3125, pruned_loss=0.04764, over 944034.00 frames. 
+2023-03-15 11:44:23,034 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 11:44:38,091 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.898e+02 1.214e+03 1.503e+03 2.091e+03 5.256e+03, threshold=3.006e+03, percent-clipped=5.0
+2023-03-15 11:44:51,394 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-15 11:45:05,062 INFO [train.py:968] (0/2) Epoch 30, batch 21050, giga_loss[loss=0.2343, simple_loss=0.3181, pruned_loss=0.0752, over 28934.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3536, pruned_loss=0.0998, over 5704818.22 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3416, pruned_loss=0.08892, over 5715613.03 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3527, pruned_loss=0.09947, over 5711161.82 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:45:28,792 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.15 vs. limit=5.0
+2023-03-15 11:45:45,498 INFO [train.py:968] (0/2) Epoch 30, batch 21100, libri_loss[loss=0.2781, simple_loss=0.3629, pruned_loss=0.0966, over 29196.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3505, pruned_loss=0.09835, over 5701949.19 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3415, pruned_loss=0.08904, over 5715252.28 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.35, pruned_loss=0.09815, over 5707029.76 frames. ], batch size: 97, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:46:00,281 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.942e+02 1.258e+03 1.496e+03 1.957e+03 5.347e+03, threshold=2.992e+03, percent-clipped=5.0
+2023-03-15 11:46:25,830 INFO [train.py:968] (0/2) Epoch 30, batch 21150, giga_loss[loss=0.2429, simple_loss=0.321, pruned_loss=0.08242, over 28872.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3484, pruned_loss=0.09772, over 5700289.54 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3417, pruned_loss=0.08932, over 5709779.38 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.348, pruned_loss=0.09748, over 5708311.73 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:46:49,611 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6974, 4.9962, 1.8369, 2.0087], device='cuda:0'), covar=tensor([0.1082, 0.0192, 0.0908, 0.1293], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0567, 0.0411, 0.0448], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 11:47:05,308 INFO [train.py:968] (0/2) Epoch 30, batch 21200, giga_loss[loss=0.254, simple_loss=0.3375, pruned_loss=0.08529, over 29116.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3462, pruned_loss=0.09652, over 5701413.69 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3416, pruned_loss=0.08946, over 5704531.94 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3461, pruned_loss=0.0964, over 5711808.42 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:47:24,615 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.717e+02 1.146e+03 1.469e+03 2.060e+03 6.615e+03, threshold=2.939e+03, percent-clipped=10.0
+2023-03-15 11:47:28,122 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.7674, 0.9703, 0.7466, 0.7058], device='cuda:0'), covar=tensor([0.0922, 0.0419, 0.1061, 0.0981], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0450, 0.0527, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 11:47:49,107 INFO [train.py:968] (0/2) Epoch 30, batch 21250, giga_loss[loss=0.2835, simple_loss=0.3565, pruned_loss=0.1052, over 28856.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3469, pruned_loss=0.09721, over 5696676.00 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3419, pruned_loss=0.08965, over 5698389.93 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3466, pruned_loss=0.09704, over 5710306.22 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:48:32,915 INFO [train.py:968] (0/2) Epoch 30, batch 21300, giga_loss[loss=0.268, simple_loss=0.3481, pruned_loss=0.09395, over 28762.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3466, pruned_loss=0.09692, over 5696494.38 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3422, pruned_loss=0.08977, over 5700143.46 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3462, pruned_loss=0.09678, over 5705779.35 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:48:50,305 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.015e+02 1.166e+03 1.374e+03 1.799e+03 3.890e+03, threshold=2.748e+03, percent-clipped=3.0
+2023-03-15 11:48:55,497 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1341795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:48:57,156 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-15 11:49:15,790 INFO [train.py:968] (0/2) Epoch 30, batch 21350, giga_loss[loss=0.2654, simple_loss=0.3492, pruned_loss=0.09083, over 28345.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3458, pruned_loss=0.09555, over 5706597.65 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3425, pruned_loss=0.08992, over 5703324.71 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3453, pruned_loss=0.09537, over 5711029.42 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:49:26,663 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1341832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:49:56,095 INFO [train.py:968] (0/2) Epoch 30, batch 21400, giga_loss[loss=0.226, simple_loss=0.306, pruned_loss=0.07298, over 28663.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3455, pruned_loss=0.09548, over 5701307.61 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3432, pruned_loss=0.09071, over 5709479.13 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3445, pruned_loss=0.09481, over 5699317.56 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:50:12,928 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.332e+02 1.211e+03 1.460e+03 1.991e+03 5.080e+03, threshold=2.919e+03, percent-clipped=10.0
+2023-03-15 11:50:16,334 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5260, 2.1292, 1.5422, 0.8188], device='cuda:0'), covar=tensor([0.6310, 0.3026, 0.4503, 0.6842], device='cuda:0'), in_proj_covar=tensor([0.1851, 0.1725, 0.1657, 0.1507], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 11:50:36,574 INFO [train.py:968] (0/2) Epoch 30, batch 21450, giga_loss[loss=0.267, simple_loss=0.3229, pruned_loss=0.1056, over 23567.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3448, pruned_loss=0.09594, over 5703538.33 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3434, pruned_loss=0.0911, over 5714525.05 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3439, pruned_loss=0.09517, over 5696833.59 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:50:45,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7640, 4.8010, 2.0670, 1.8474], device='cuda:0'), covar=tensor([0.1001, 0.0209, 0.0863, 0.1387], device='cuda:0'), in_proj_covar=tensor([0.0426, 0.0565, 0.0411, 0.0447], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 11:51:13,311 INFO [train.py:968] (0/2) Epoch 30, batch 21500, giga_loss[loss=0.2454, simple_loss=0.3193, pruned_loss=0.08573, over 28720.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3432, pruned_loss=0.09539, over 5706767.18 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3437, pruned_loss=0.09152, over 5717152.29 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3422, pruned_loss=0.09456, over 5698631.20 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:51:19,050 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1341975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:51:20,330 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1341977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:51:21,127 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1341978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:51:30,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.643e+02 1.410e+03 1.815e+03 2.622e+03 7.504e+03, threshold=3.629e+03, percent-clipped=20.0
+2023-03-15 11:51:38,117 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1342000.pt
+2023-03-15 11:51:44,519 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1342007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:51:51,830 INFO [train.py:968] (0/2) Epoch 30, batch 21550, libri_loss[loss=0.2695, simple_loss=0.342, pruned_loss=0.09845, over 29531.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3405, pruned_loss=0.0943, over 5709754.63 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3439, pruned_loss=0.09191, over 5724208.70 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3394, pruned_loss=0.09337, over 5696320.01 frames. ], batch size: 80, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:52:31,371 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 11:52:34,167 INFO [train.py:968] (0/2) Epoch 30, batch 21600, giga_loss[loss=0.254, simple_loss=0.3281, pruned_loss=0.08992, over 28975.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3394, pruned_loss=0.09423, over 5704202.96 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3439, pruned_loss=0.09192, over 5725956.22 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3385, pruned_loss=0.09349, over 5691869.42 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:52:51,901 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.942e+02 1.287e+03 1.468e+03 2.101e+03 5.645e+03, threshold=2.936e+03, percent-clipped=1.0
+2023-03-15 11:53:06,096 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-15 11:53:14,594 INFO [train.py:968] (0/2) Epoch 30, batch 21650, giga_loss[loss=0.2249, simple_loss=0.3037, pruned_loss=0.07309, over 28657.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3385, pruned_loss=0.09396, over 5690882.82 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3439, pruned_loss=0.09208, over 5710970.97 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3377, pruned_loss=0.09327, over 5694106.34 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:53:25,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9756, 3.1880, 2.0413, 1.0346], device='cuda:0'), covar=tensor([1.0113, 0.3462, 0.4757, 0.8728], device='cuda:0'), in_proj_covar=tensor([0.1854, 0.1731, 0.1663, 0.1509], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 11:53:56,829 INFO [train.py:968] (0/2) Epoch 30, batch 21700, giga_loss[loss=0.2439, simple_loss=0.3209, pruned_loss=0.08348, over 29006.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3368, pruned_loss=0.09336, over 5694477.50 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3442, pruned_loss=0.09236, over 5714774.12 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3356, pruned_loss=0.09255, over 5693061.56 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:53:57,726 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1342169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:53:58,507 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1342170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:54:14,391 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.420e+02 1.198e+03 1.539e+03 2.044e+03 8.333e+03, threshold=3.078e+03, percent-clipped=14.0
+2023-03-15 11:54:37,053 INFO [train.py:968] (0/2) Epoch 30, batch 21750, giga_loss[loss=0.235, simple_loss=0.3147, pruned_loss=0.0776, over 28914.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3341, pruned_loss=0.09229, over 5700614.91 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.345, pruned_loss=0.09309, over 5717187.60 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3323, pruned_loss=0.091, over 5696924.89 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:55:08,382 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.11 vs. limit=5.0
+2023-03-15 11:55:21,894 INFO [train.py:968] (0/2) Epoch 30, batch 21800, giga_loss[loss=0.2338, simple_loss=0.3145, pruned_loss=0.07653, over 28933.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3316, pruned_loss=0.09083, over 5705726.91 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3452, pruned_loss=0.09325, over 5714411.66 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3299, pruned_loss=0.08965, over 5705500.57 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:55:39,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.969e+02 1.202e+03 1.474e+03 1.863e+03 5.819e+03, threshold=2.948e+03, percent-clipped=7.0
+2023-03-15 11:55:50,410 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5030, 1.9303, 1.5679, 1.6011], device='cuda:0'), covar=tensor([0.0763, 0.0284, 0.0346, 0.0935], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 11:55:57,209 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1342313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:55:57,268 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1342313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:55:59,122 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1342316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:56:00,120 INFO [train.py:968] (0/2) Epoch 30, batch 21850, libri_loss[loss=0.2486, simple_loss=0.3365, pruned_loss=0.08034, over 29533.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.331, pruned_loss=0.09067, over 5712559.62 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3458, pruned_loss=0.09381, over 5721180.69 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3286, pruned_loss=0.08912, over 5705882.32 frames. ], batch size: 82, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:56:23,157 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1342345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:56:29,802 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1342352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:56:42,107 INFO [train.py:968] (0/2) Epoch 30, batch 21900, giga_loss[loss=0.2513, simple_loss=0.3288, pruned_loss=0.0869, over 28866.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.332, pruned_loss=0.09079, over 5706970.55 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.346, pruned_loss=0.09416, over 5716888.16 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3294, pruned_loss=0.08913, over 5705862.56 frames. ], batch size: 66, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:56:59,634 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.900e+02 1.233e+03 1.586e+03 2.286e+03 6.490e+03, threshold=3.172e+03, percent-clipped=12.0
+2023-03-15 11:57:02,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6101, 2.0502, 1.2598, 1.5322], device='cuda:0'), covar=tensor([0.1165, 0.0664, 0.1141, 0.1366], device='cuda:0'), in_proj_covar=tensor([0.0415, 0.0447, 0.0523, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 11:57:24,645 INFO [train.py:968] (0/2) Epoch 30, batch 21950, giga_loss[loss=0.286, simple_loss=0.3634, pruned_loss=0.1043, over 29070.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3371, pruned_loss=0.09326, over 5693780.34 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3464, pruned_loss=0.09467, over 5709581.94 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3343, pruned_loss=0.09141, over 5698735.50 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:58:10,045 INFO [train.py:968] (0/2) Epoch 30, batch 22000, libri_loss[loss=0.2846, simple_loss=0.3417, pruned_loss=0.1137, over 29357.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.341, pruned_loss=0.09497, over 5694254.06 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3465, pruned_loss=0.09499, over 5717153.96 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3385, pruned_loss=0.09316, over 5690770.84 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:58:26,710 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.976e+02 1.240e+03 1.640e+03 2.044e+03 5.867e+03, threshold=3.280e+03, percent-clipped=9.0
+2023-03-15 11:58:32,252 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1342495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:58:34,856 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1342498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:58:51,134 INFO [train.py:968] (0/2) Epoch 30, batch 22050, libri_loss[loss=0.3594, simple_loss=0.4199, pruned_loss=0.1494, over 29158.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3419, pruned_loss=0.09442, over 5705259.25 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3473, pruned_loss=0.09568, over 5720546.58 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3389, pruned_loss=0.09231, over 5698639.19 frames. ], batch size: 101, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:58:53,006 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1945, 4.0570, 3.8202, 1.8717], device='cuda:0'), covar=tensor([0.0575, 0.0693, 0.0707, 0.2288], device='cuda:0'), in_proj_covar=tensor([0.1305, 0.1205, 0.1011, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 11:58:59,223 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1342527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:59:12,169 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4397, 0.9208, 4.5805, 3.5256], device='cuda:0'), covar=tensor([0.1768, 0.3324, 0.0394, 0.1014], device='cuda:0'), in_proj_covar=tensor([0.0812, 0.0678, 0.1011, 0.0985], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 11:59:13,712 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1342544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:59:20,552 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-15 11:59:34,961 INFO [train.py:968] (0/2) Epoch 30, batch 22100, giga_loss[loss=0.2339, simple_loss=0.3197, pruned_loss=0.07407, over 28703.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3414, pruned_loss=0.09365, over 5708900.42 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3474, pruned_loss=0.09597, over 5724188.26 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3388, pruned_loss=0.09168, over 5700198.36 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:59:56,028 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.796e+02 1.222e+03 1.474e+03 1.960e+03 4.955e+03, threshold=2.948e+03, percent-clipped=3.0
+2023-03-15 12:00:20,227 INFO [train.py:968] (0/2) Epoch 30, batch 22150, giga_loss[loss=0.2554, simple_loss=0.3397, pruned_loss=0.08556, over 28988.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3414, pruned_loss=0.09403, over 5705755.40 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3481, pruned_loss=0.09678, over 5725530.49 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3386, pruned_loss=0.09167, over 5697185.58 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:01:03,173 INFO [train.py:968] (0/2) Epoch 30, batch 22200, giga_loss[loss=0.2345, simple_loss=0.3187, pruned_loss=0.07517, over 29066.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3411, pruned_loss=0.09426, over 5699082.06 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3481, pruned_loss=0.0969, over 5720741.62 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3386, pruned_loss=0.0922, over 5696107.83 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:01:20,108 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1342687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:01:20,787 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1342688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:01:21,990 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.346e+02 1.346e+03 1.857e+03 2.319e+03 4.754e+03, threshold=3.715e+03, percent-clipped=14.0
+2023-03-15 12:01:22,533 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1342690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:01:40,788 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-15 12:01:43,677 INFO [train.py:968] (0/2) Epoch 30, batch 22250, giga_loss[loss=0.235, simple_loss=0.3137, pruned_loss=0.07818, over 28462.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3406, pruned_loss=0.09423, over 5697355.17 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3479, pruned_loss=0.09713, over 5719222.38 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3384, pruned_loss=0.09221, over 5695865.55 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:01:44,462 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1342719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:01:53,657 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.3692, 5.2180, 4.9425, 2.4808], device='cuda:0'), covar=tensor([0.0465, 0.0622, 0.0602, 0.1709], device='cuda:0'), in_proj_covar=tensor([0.1302, 0.1203, 0.1009, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 12:01:57,200 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-15 12:02:28,634 INFO [train.py:968] (0/2) Epoch 30, batch 22300, giga_loss[loss=0.2744, simple_loss=0.3502, pruned_loss=0.09927, over 28831.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3437, pruned_loss=0.0958, over 5702120.38 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.348, pruned_loss=0.09716, over 5720153.27 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3419, pruned_loss=0.09419, over 5700113.05 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:02:47,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.443e+03 1.705e+03 2.596e+03 5.668e+03, threshold=3.411e+03, percent-clipped=9.0
+2023-03-15 12:03:12,305 INFO [train.py:968] (0/2) Epoch 30, batch 22350, giga_loss[loss=0.257, simple_loss=0.3344, pruned_loss=0.08982, over 28813.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3475, pruned_loss=0.09756, over 5704768.03 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3484, pruned_loss=0.09744, over 5722700.23 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3457, pruned_loss=0.09601, over 5700682.97 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:03:23,917 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1342831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:03:26,626 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1342834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:03:51,025 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1342863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:03:54,013 INFO [train.py:968] (0/2) Epoch 30, batch 22400, giga_loss[loss=0.2608, simple_loss=0.3357, pruned_loss=0.09296, over 28569.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3495, pruned_loss=0.09893, over 5707737.14 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3487, pruned_loss=0.09786, over 5717961.46 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3478, pruned_loss=0.09736, over 5707600.15 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:04:13,154 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.553e+02 1.449e+03 1.781e+03 2.324e+03 6.576e+03, threshold=3.563e+03, percent-clipped=8.0
+2023-03-15 12:04:18,829 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9959, 1.0626, 3.3705, 2.9336], device='cuda:0'), covar=tensor([0.1817, 0.2991, 0.0531, 0.1106], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0680, 0.1015, 0.0991], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 12:04:35,970 INFO [train.py:968] (0/2) Epoch 30, batch 22450, giga_loss[loss=0.3062, simple_loss=0.3697, pruned_loss=0.1214, over 28849.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3503, pruned_loss=0.09926, over 5718527.08 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.349, pruned_loss=0.09812, over 5720830.54 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3487, pruned_loss=0.09779, over 5715640.26 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:04:46,583 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.8431, 1.0728, 5.0326, 3.6835], device='cuda:0'), covar=tensor([0.1610, 0.3237, 0.0403, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0679, 0.1015, 0.0990], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 12:04:59,087 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3038, 1.2467, 3.3965, 3.0685], device='cuda:0'), covar=tensor([0.1500, 0.2806, 0.0486, 0.1805], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0680, 0.1016, 0.0991], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 12:05:07,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4041, 1.5296, 1.5377, 1.3883], device='cuda:0'), covar=tensor([0.2994, 0.2966, 0.2141, 0.2793], device='cuda:0'), in_proj_covar=tensor([0.2094, 0.2061, 0.1969, 0.2109], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 12:05:20,249 INFO [train.py:968] (0/2) Epoch 30, batch 22500, giga_loss[loss=0.2802, simple_loss=0.3561, pruned_loss=0.1022, over 28916.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3521, pruned_loss=0.1006, over 5706305.31 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3497, pruned_loss=0.09865, over 5716075.58 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3503, pruned_loss=0.09898, over 5708725.64 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:05:39,239 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.351e+03 1.812e+03 2.381e+03 7.275e+03, threshold=3.624e+03, percent-clipped=6.0
+2023-03-15 12:05:50,307 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-15 12:06:03,153 INFO [train.py:968] (0/2) Epoch 30, batch 22550, giga_loss[loss=0.2577, simple_loss=0.3358, pruned_loss=0.08984, over 28971.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3503, pruned_loss=0.09974, over 5706622.04 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3502, pruned_loss=0.09908, over 5709237.76 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3485, pruned_loss=0.09813, over 5714725.50 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:06:15,527 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5604, 2.2395, 1.6805, 0.8543], device='cuda:0'), covar=tensor([0.7701, 0.3423, 0.4879, 0.8092], device='cuda:0'), in_proj_covar=tensor([0.1852, 0.1728, 0.1663, 0.1508], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 12:06:38,062 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4249, 1.3999, 3.9482, 3.4339], device='cuda:0'), covar=tensor([0.1855, 0.3068, 0.0772, 0.1191], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0677, 0.1012, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 12:06:49,473 INFO [train.py:968] (0/2) Epoch 30, batch 22600, giga_loss[loss=0.3076, simple_loss=0.3684, pruned_loss=0.1234, over 27923.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3482, pruned_loss=0.09878, over 5709452.11 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3503, pruned_loss=0.09916, over 5712373.76 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3467, pruned_loss=0.09744, over 5713010.79 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:07:10,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.819e+02 1.308e+03 1.617e+03 1.972e+03 5.224e+03, threshold=3.233e+03, percent-clipped=3.0
+2023-03-15 12:07:32,331 INFO [train.py:968] (0/2) Epoch 30, batch 22650, giga_loss[loss=0.2511, simple_loss=0.33, pruned_loss=0.08612, over 28568.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3444, pruned_loss=0.09678, over 5702479.54 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3503, pruned_loss=0.0993, over 5705083.33 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3431, pruned_loss=0.09556, over 5712034.61 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:07:46,993 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1343134.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 12:08:13,335 INFO [train.py:968] (0/2) Epoch 30, batch 22700, giga_loss[loss=0.2544, simple_loss=0.3308, pruned_loss=0.08902, over 28718.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3426, pruned_loss=0.09531, over 5709301.09 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3504, pruned_loss=0.09941, over 5708116.16 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3414, pruned_loss=0.09422, over 5714214.88 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:08:34,171 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.428e+02 1.249e+03 1.716e+03 2.353e+03 6.286e+03, threshold=3.432e+03, percent-clipped=9.0
+2023-03-15 12:09:01,280 INFO [train.py:968] (0/2) Epoch 30, batch 22750, giga_loss[loss=0.3094, simple_loss=0.3791, pruned_loss=0.1198, over 26712.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3447, pruned_loss=0.09525, over 5705094.72 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3503, pruned_loss=0.09932, over 5709179.63 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3437, pruned_loss=0.09442, over 5708127.31 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:09:20,297 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-15 12:09:43,168 INFO [train.py:968] (0/2) Epoch 30, batch 22800, libri_loss[loss=0.3088, simple_loss=0.3788, pruned_loss=0.1194, over 29209.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3441, pruned_loss=0.09439, over 5714679.97 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3504, pruned_loss=0.09945, over 5711796.12 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3431, pruned_loss=0.09354, over 5714683.47 frames. ], batch size: 101, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:09:47,048 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.6638, 4.5219, 4.2974, 2.2908], device='cuda:0'), covar=tensor([0.0631, 0.0780, 0.0731, 0.1835], device='cuda:0'), in_proj_covar=tensor([0.1300, 0.1201, 0.1008, 0.0751], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 12:09:59,373 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.216e+02 1.362e+03 1.698e+03 2.290e+03 5.216e+03, threshold=3.396e+03, percent-clipped=4.0
+2023-03-15 12:10:02,849 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1343296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:10:14,883 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-15 12:10:22,205 INFO [train.py:968] (0/2) Epoch 30, batch 22850, libri_loss[loss=0.3124, simple_loss=0.3747, pruned_loss=0.1251, over 19632.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3431, pruned_loss=0.09463, over 5707712.25 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3505, pruned_loss=0.0997, over 5700302.92 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.342, pruned_loss=0.09348, over 5719717.25 frames. ], batch size: 187, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:10:51,411 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1343352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:11:05,538 INFO [train.py:968] (0/2) Epoch 30, batch 22900, giga_loss[loss=0.2519, simple_loss=0.3248, pruned_loss=0.08957, over 29010.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3421, pruned_loss=0.09555, over 5712832.28 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3504, pruned_loss=0.0998, over 5705566.91 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3411, pruned_loss=0.09445, over 5717786.46 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:11:25,622 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.579e+02 1.379e+03 1.735e+03 2.477e+03 9.112e+03, threshold=3.470e+03, percent-clipped=14.0
+2023-03-15 12:11:47,730 INFO [train.py:968] (0/2) Epoch 30, batch 22950, giga_loss[loss=0.3009, simple_loss=0.3675, pruned_loss=0.1171, over 27993.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.341, pruned_loss=0.09624, over 5705681.29 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3507, pruned_loss=0.1001, over 5699427.85 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3398, pruned_loss=0.09503, over 5715376.53 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:11:48,003 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4783, 4.3398, 4.1268, 1.8544], device='cuda:0'), covar=tensor([0.0668, 0.0831, 0.0780, 0.2141], device='cuda:0'), in_proj_covar=tensor([0.1305, 0.1205, 0.1011, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 12:12:06,295 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5385, 2.2151, 1.6487, 0.8413], device='cuda:0'), covar=tensor([0.8174, 0.3328, 0.4479, 0.8558], device='cuda:0'), in_proj_covar=tensor([0.1853, 0.1728, 0.1667, 0.1513], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 12:12:30,776 INFO [train.py:968] (0/2) Epoch 30, batch 23000, giga_loss[loss=0.2319, simple_loss=0.3004, pruned_loss=0.08169, over 28623.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3399, pruned_loss=0.09668, over 5712518.21 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3508, pruned_loss=0.1002, over 5700832.19 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3387, pruned_loss=0.0956, over 5719101.74 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:12:50,508 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.306e+03 1.601e+03 2.047e+03 5.343e+03, threshold=3.202e+03, percent-clipped=2.0
+2023-03-15 12:12:59,814 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6327, 1.9130, 1.5394, 1.6379], device='cuda:0'), covar=tensor([0.3413, 0.3020, 0.3866, 0.2630], device='cuda:0'), in_proj_covar=tensor([0.1617, 0.1165, 0.1429, 0.1013], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 12:13:05,303 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1343509.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 12:13:13,540 INFO [train.py:968] (0/2) Epoch 30, batch 23050, giga_loss[loss=0.2574, simple_loss=0.3414, pruned_loss=0.08674, over 28748.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3381, pruned_loss=0.09614, over 5706918.36 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3508, pruned_loss=0.1002, over 5700832.19 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3372, pruned_loss=0.0953, over 5712042.41 frames. ], batch size: 243, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:13:24,846 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4746, 1.5837, 1.3975, 1.6473], device='cuda:0'), covar=tensor([0.0731, 0.0331, 0.0345, 0.0877], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 12:13:27,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4246, 1.7671, 1.5466, 1.6324], device='cuda:0'), covar=tensor([0.2050, 0.2264, 0.2248, 0.2270], device='cuda:0'), in_proj_covar=tensor([0.0516, 0.0764, 0.0736, 0.0703], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 12:13:53,337 INFO [train.py:968] (0/2) Epoch 30, batch 23100, libri_loss[loss=0.3316, simple_loss=0.3965, pruned_loss=0.1334, over 29178.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3344, pruned_loss=0.09443, over 5714594.06 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3513, pruned_loss=0.1007, over 5701294.07 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3329, pruned_loss=0.09319, over 5718230.12 frames. ], batch size: 101, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:14:00,877 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4259, 1.5857, 1.5498, 1.3782], device='cuda:0'), covar=tensor([0.3361, 0.3238, 0.2383, 0.2972], device='cuda:0'), in_proj_covar=tensor([0.2098, 0.2062, 0.1969, 0.2108], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 12:14:13,451 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.124e+02 1.297e+03 1.715e+03 2.173e+03 3.985e+03, threshold=3.430e+03, percent-clipped=3.0
+2023-03-15 12:14:18,185 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1343598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:14:21,904 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6543, 1.7393, 1.8421, 1.3851], device='cuda:0'), covar=tensor([0.1878, 0.3017, 0.1655, 0.1937], device='cuda:0'), in_proj_covar=tensor([0.0943, 0.0717, 0.0992, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 12:14:33,739 INFO [train.py:968] (0/2) Epoch 30, batch 23150, libri_loss[loss=0.2605, simple_loss=0.323, pruned_loss=0.09903, over 29348.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3313, pruned_loss=0.0931, over 5709865.05 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3516, pruned_loss=0.1013, over 5698781.39 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3292, pruned_loss=0.0913, over 5714988.48 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:14:59,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1343652.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:15:00,966 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1343655.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 12:15:10,825 INFO [train.py:968] (0/2) Epoch 30, batch 23200, giga_loss[loss=0.2133, simple_loss=0.2996, pruned_loss=0.06347, over 28873.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3293, pruned_loss=0.09168, over 5713630.25 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3522, pruned_loss=0.1019, over 5696811.15 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3263, pruned_loss=0.08934, over 5720110.34 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:15:13,375 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1343671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:15:23,791 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1343684.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 12:15:27,326 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5042, 1.7233, 1.3540, 1.6809], device='cuda:0'), covar=tensor([0.0732, 0.0308, 0.0346, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 12:15:30,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.904e+02 1.510e+03 1.962e+03 2.633e+03 7.250e+03, threshold=3.923e+03, percent-clipped=17.0
+2023-03-15 12:15:53,200 INFO [train.py:968] (0/2) Epoch 30, batch 23250, giga_loss[loss=0.2315, simple_loss=0.3126, pruned_loss=0.07522, over 28494.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3335, pruned_loss=0.09416, over 5712739.81 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3522, pruned_loss=0.1022, over 5705006.19 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3302, pruned_loss=0.0916, over 5711087.97 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:15:59,703 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1343727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:16:36,473 INFO [train.py:968] (0/2) Epoch 30, batch 23300, giga_loss[loss=0.2783, simple_loss=0.3645, pruned_loss=0.09605, over 28557.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.336, pruned_loss=0.09473, over 5711114.06 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.352, pruned_loss=0.1021, over 5707052.90 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3334, pruned_loss=0.09272, over 5708077.19 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:16:56,244 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.095e+02 1.374e+03 1.642e+03 2.375e+03 4.090e+03, threshold=3.284e+03, percent-clipped=2.0
+2023-03-15 12:17:13,954 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1343814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:17:15,899 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1343817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:17:16,588 INFO [train.py:968] (0/2) Epoch 30, batch 23350, giga_loss[loss=0.2543, simple_loss=0.3301, pruned_loss=0.0893, over 28699.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3384, pruned_loss=0.09505, over 5718706.22 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3518, pruned_loss=0.1021, over 5712212.41 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3361, pruned_loss=0.09327, over 5711697.53 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:17:18,462 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4107, 1.4816, 1.3152, 1.4750], device='cuda:0'), covar=tensor([0.0767, 0.0339, 0.0358, 0.0924], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 12:17:43,416 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1343846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:18:02,872 INFO [train.py:968] (0/2) Epoch 30, batch 23400, giga_loss[loss=0.2691, simple_loss=0.3483, pruned_loss=0.09501, over 28533.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3409, pruned_loss=0.0957, over 5718681.55 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3518, pruned_loss=0.1022, over 5714388.71 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.339, pruned_loss=0.09415, over 5711234.48 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:18:04,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1343870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:18:07,680 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1343873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:18:10,287 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3496, 3.0217, 1.4011, 1.4694], device='cuda:0'), covar=tensor([0.1011, 0.0393, 0.0991, 0.1382], device='cuda:0'), in_proj_covar=tensor([0.0428, 0.0573, 0.0415, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 12:18:26,336 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.155e+02 1.303e+03 1.574e+03 2.140e+03 7.950e+03, threshold=3.149e+03, percent-clipped=7.0
+2023-03-15 12:18:28,850 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6293, 1.9404, 1.5253, 1.6671], device='cuda:0'), covar=tensor([0.2734, 0.2798, 0.3302, 0.2731], device='cuda:0'), in_proj_covar=tensor([0.1621, 0.1168, 0.1431, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 12:18:35,868 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1343902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:18:38,930 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4840, 1.7106, 1.5331, 1.6698], device='cuda:0'), covar=tensor([0.0755, 0.0306, 0.0321, 0.0879], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:0')
+2023-03-15 12:18:48,798 INFO [train.py:968] (0/2) Epoch 30, batch 23450, giga_loss[loss=0.2447, simple_loss=0.333, pruned_loss=0.07821, over 28899.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3422, pruned_loss=0.09593, over 5722027.77 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3519, pruned_loss=0.1022, over 5713533.77 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3406, pruned_loss=0.09464, over 5717052.88 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:18:49,046 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1343918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:19:08,380 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-15 12:19:38,922 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-15 12:19:39,877 INFO [train.py:968] (0/2) Epoch 30, batch 23500, giga_loss[loss=0.2563, simple_loss=0.3358, pruned_loss=0.08839, over 28492.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3478, pruned_loss=0.1006, over 5707585.70 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3521, pruned_loss=0.1024, over 5716642.92 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3462, pruned_loss=0.09928, over 5700933.68 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:19:44,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1343973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:20:03,658 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.718e+03 2.120e+03 2.854e+03 5.958e+03, threshold=4.241e+03, percent-clipped=18.0
+2023-03-15 12:20:10,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 12:20:12,826 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1344000.pt
+2023-03-15 12:20:33,014 INFO [train.py:968] (0/2) Epoch 30, batch 23550, libri_loss[loss=0.2376, simple_loss=0.3067, pruned_loss=0.08425, over 29665.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3542, pruned_loss=0.1057, over 5703912.06 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3523, pruned_loss=0.1027, over 5721272.48 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3528, pruned_loss=0.1045, over 5694134.65 frames. ], batch size: 69, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:20:52,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4591, 1.5809, 1.5493, 1.3815], device='cuda:0'), covar=tensor([0.2740, 0.2806, 0.2379, 0.2694], device='cuda:0'), in_proj_covar=tensor([0.2093, 0.2057, 0.1967, 0.2103], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 12:21:23,659 INFO [train.py:968] (0/2) Epoch 30, batch 23600, giga_loss[loss=0.2954, simple_loss=0.3682, pruned_loss=0.1113, over 28826.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3614, pruned_loss=0.1109, over 5685065.29 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3528, pruned_loss=0.1031, over 5713738.22 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3599, pruned_loss=0.1097, over 5683477.73 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:21:39,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3592, 1.9597, 1.6550, 1.4722], device='cuda:0'), covar=tensor([0.0661, 0.0268, 0.0271, 0.0836], device='cuda:0'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:0'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 12:21:49,857 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.252e+03 1.868e+03 2.486e+03 3.444e+03 7.764e+03, threshold=4.972e+03, percent-clipped=15.0
+2023-03-15 12:22:13,309 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1344116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:22:14,326 INFO [train.py:968] (0/2) Epoch 30, batch 23650, giga_loss[loss=0.326, simple_loss=0.3907, pruned_loss=0.1306, over 28617.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3663, pruned_loss=0.1146, over 5688439.58 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3526, pruned_loss=0.103, over 5716453.56 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3655, pruned_loss=0.1139, over 5684367.09 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:22:15,878 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1344119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:22:45,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1344148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:23:05,621 INFO [train.py:968] (0/2) Epoch 30, batch 23700, giga_loss[loss=0.2928, simple_loss=0.366, pruned_loss=0.1098, over 28969.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3738, pruned_loss=0.1209, over 5687074.83 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3529, pruned_loss=0.1032, over 5720414.29 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3733, pruned_loss=0.1206, over 5679641.31 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:23:32,524 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.185e+03 1.945e+03 2.382e+03 3.524e+03 1.074e+04, threshold=4.764e+03, percent-clipped=10.0
+2023-03-15 12:23:56,342 INFO [train.py:968] (0/2) Epoch 30, batch 23750, giga_loss[loss=0.2858, simple_loss=0.3602, pruned_loss=0.1057, over 28865.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3776, pruned_loss=0.1244, over 5680713.00 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3532, pruned_loss=0.1034, over 5722430.86 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3773, pruned_loss=0.1242, over 5672571.28 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:24:48,394 INFO [train.py:968] (0/2) Epoch 30, batch 23800, giga_loss[loss=0.3891, simple_loss=0.4168, pruned_loss=0.1807, over 23489.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3791, pruned_loss=0.127, over 5669578.30 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3528, pruned_loss=0.1031, over 5724864.60 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3796, pruned_loss=0.1274, over 5660596.12 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:25:15,642 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1344293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:25:16,900 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.256e+03 1.836e+03 2.289e+03 3.181e+03 1.113e+04, threshold=4.578e+03, percent-clipped=10.0
+2023-03-15 12:25:21,178 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 12:25:41,125 INFO [train.py:968] (0/2) Epoch 30, batch 23850, giga_loss[loss=0.3052, simple_loss=0.373, pruned_loss=0.1187, over 28906.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3841, pruned_loss=0.1323, over 5653794.25 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3528, pruned_loss=0.1032, over 5719672.55 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.385, pruned_loss=0.133, over 5649824.24 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:26:31,452 INFO [train.py:968] (0/2) Epoch 30, batch 23900, giga_loss[loss=0.3964, simple_loss=0.4348, pruned_loss=0.179, over 28287.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3865, pruned_loss=0.1356, over 5647230.54 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3528, pruned_loss=0.1036, over 5718352.53 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3888, pruned_loss=0.1372, over 5642157.75 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:27:04,029 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.365e+03 2.079e+03 2.611e+03 3.301e+03 7.860e+03, threshold=5.223e+03, percent-clipped=11.0
+2023-03-15 12:27:13,761 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1344404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:27:29,983 INFO [train.py:968] (0/2) Epoch 30, batch 23950, giga_loss[loss=0.2959, simple_loss=0.3611, pruned_loss=0.1154, over 28533.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3886, pruned_loss=0.1366, over 5655284.99 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.353, pruned_loss=0.1037, over 5721904.82 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3912, pruned_loss=0.1387, over 5646194.76 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:27:50,166 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1344436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:27:53,215 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1344439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:28:24,534 INFO [train.py:968] (0/2) Epoch 30, batch 24000, giga_loss[loss=0.3271, simple_loss=0.3882, pruned_loss=0.133, over 28616.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3877, pruned_loss=0.137, over 5651458.32 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3526, pruned_loss=0.1036, over 5726270.28 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3909, pruned_loss=0.1395, over 5638761.48 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:28:24,538 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 12:28:33,774 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.2015, simple_loss=0.31, pruned_loss=0.04652, over 944034.00 frames. 
+2023-03-15 12:28:33,775 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 12:28:34,093 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1344468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:28:34,110 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1344468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:28:55,110 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.123e+03 2.132e+03 2.668e+03 3.315e+03 8.648e+03, threshold=5.336e+03, percent-clipped=10.0
+2023-03-15 12:29:17,909 INFO [train.py:968] (0/2) Epoch 30, batch 24050, giga_loss[loss=0.2909, simple_loss=0.3651, pruned_loss=0.1084, over 29140.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.387, pruned_loss=0.1371, over 5646158.86 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3538, pruned_loss=0.1048, over 5720939.85 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3894, pruned_loss=0.1389, over 5638393.76 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:29:35,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-15 12:29:37,338 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7224, 2.2481, 1.4138, 1.0090], device='cuda:0'), covar=tensor([0.7057, 0.4194, 0.3520, 0.6668], device='cuda:0'), in_proj_covar=tensor([0.1871, 0.1752, 0.1682, 0.1522], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 12:30:04,607 INFO [train.py:968] (0/2) Epoch 30, batch 24100, giga_loss[loss=0.2997, simple_loss=0.3741, pruned_loss=0.1126, over 28659.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3863, pruned_loss=0.1355, over 5651342.98 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3541, pruned_loss=0.105, over 5723191.13 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3891, pruned_loss=0.1379, over 5640380.66 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:30:17,184 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1344582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:30:32,612 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.293e+03 1.959e+03 2.422e+03 3.249e+03 8.661e+03, threshold=4.845e+03, percent-clipped=5.0
+2023-03-15 12:30:44,722 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1344606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:30:56,449 INFO [train.py:968] (0/2) Epoch 30, batch 24150, giga_loss[loss=0.2935, simple_loss=0.365, pruned_loss=0.111, over 28564.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3859, pruned_loss=0.1347, over 5648027.86 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3546, pruned_loss=0.1055, over 5727460.55 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3886, pruned_loss=0.137, over 5633385.85 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:31:26,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7177, 2.3295, 1.5422, 0.9039], device='cuda:0'), covar=tensor([0.7937, 0.3677, 0.3790, 0.7343], device='cuda:0'), in_proj_covar=tensor([0.1871, 0.1751, 0.1683, 0.1523], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 12:31:47,751 INFO [train.py:968] (0/2) Epoch 30, batch 24200, giga_loss[loss=0.2791, simple_loss=0.3571, pruned_loss=0.1005, over 29014.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.386, pruned_loss=0.1344, over 5639735.62 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3543, pruned_loss=0.1054, over 5729085.78 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3889, pruned_loss=0.1367, over 5625545.30 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:32:18,127 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.284e+03 1.934e+03 2.734e+03 3.738e+03 1.098e+04, threshold=5.468e+03, percent-clipped=10.0
+2023-03-15 12:32:41,191 INFO [train.py:968] (0/2) Epoch 30, batch 24250, giga_loss[loss=0.3659, simple_loss=0.4016, pruned_loss=0.1651, over 23741.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3828, pruned_loss=0.1316, over 5626664.82 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3538, pruned_loss=0.1052, over 5722309.72 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3862, pruned_loss=0.1343, over 5619099.82 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:33:22,554 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 12:33:32,094 INFO [train.py:968] (0/2) Epoch 30, batch 24300, giga_loss[loss=0.2719, simple_loss=0.3519, pruned_loss=0.09593, over 28764.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3793, pruned_loss=0.1273, over 5641967.56 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3539, pruned_loss=0.1055, over 5727157.42 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3827, pruned_loss=0.1298, over 5629212.36 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:33:42,385 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1344779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:33:50,438 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 12:33:57,702 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.931e+03 2.559e+03 3.785e+03 1.369e+04, threshold=5.117e+03, percent-clipped=9.0
+2023-03-15 12:34:02,748 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4347, 4.2925, 4.0744, 2.0648], device='cuda:0'), covar=tensor([0.0579, 0.0681, 0.0776, 0.2142], device='cuda:0'), in_proj_covar=tensor([0.1329, 0.1229, 0.1031, 0.0766], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 12:34:08,298 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3603, 1.1681, 1.0863, 1.5143], device='cuda:0'), covar=tensor([0.0809, 0.0386, 0.0382, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 12:34:19,682 INFO [train.py:968] (0/2) Epoch 30, batch 24350, giga_loss[loss=0.3429, simple_loss=0.3927, pruned_loss=0.1466, over 26692.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3764, pruned_loss=0.1246, over 5652540.65 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3537, pruned_loss=0.1058, over 5728647.20 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3801, pruned_loss=0.1271, over 5638142.68 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:34:41,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1344843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:35:06,079 INFO [train.py:968] (0/2) Epoch 30, batch 24400, giga_loss[loss=0.2908, simple_loss=0.362, pruned_loss=0.1098, over 27661.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3737, pruned_loss=0.1221, over 5671461.35 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3533, pruned_loss=0.1057, over 5733775.63 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3778, pruned_loss=0.1247, over 5653251.12 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:35:07,864 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5221, 1.8062, 1.4188, 1.5676], device='cuda:0'), covar=tensor([0.2884, 0.2967, 0.3388, 0.2657], device='cuda:0'), in_proj_covar=tensor([0.1620, 0.1166, 0.1431, 0.1015], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 12:35:22,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1344884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:35:35,137 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.205e+03 1.825e+03 2.308e+03 2.946e+03 8.128e+03, threshold=4.615e+03, percent-clipped=1.0
+2023-03-15 12:35:53,914 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2119, 1.5106, 1.4789, 1.3128], device='cuda:0'), covar=tensor([0.1727, 0.1452, 0.2065, 0.1684], device='cuda:0'), in_proj_covar=tensor([0.0514, 0.0762, 0.0731, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 12:35:54,916 INFO [train.py:968] (0/2) Epoch 30, batch 24450, giga_loss[loss=0.2749, simple_loss=0.3409, pruned_loss=0.1044, over 28625.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3709, pruned_loss=0.1204, over 5664776.79 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3533, pruned_loss=0.106, over 5736280.24 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3745, pruned_loss=0.1225, over 5646686.55 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:35:58,526 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1344922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:36:02,333 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1344925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:36:28,520 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1344954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:36:30,566 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1344957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:36:32,059 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3722, 1.4466, 1.4433, 1.3728], device='cuda:0'), covar=tensor([0.2444, 0.2403, 0.2099, 0.2291], device='cuda:0'), in_proj_covar=tensor([0.2102, 0.2062, 0.1974, 0.2117], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 12:36:42,514 INFO [train.py:968] (0/2) Epoch 30, batch 24500, giga_loss[loss=0.2846, simple_loss=0.3566, pruned_loss=0.1063, over 28577.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3705, pruned_loss=0.1195, over 5674799.99 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3535, pruned_loss=0.1062, over 5730822.52 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3736, pruned_loss=0.1213, over 5662765.88 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:36:45,882 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1772, 1.7195, 1.4205, 1.5353], device='cuda:0'), covar=tensor([0.2601, 0.1925, 0.2443, 0.2039], device='cuda:0'), in_proj_covar=tensor([0.0514, 0.0762, 0.0731, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 12:36:58,136 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1344981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:37:06,005 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1344986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:37:08,355 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1344989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:37:17,283 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.266e+03 1.895e+03 2.412e+03 4.243e+03 1.141e+04, threshold=4.825e+03, percent-clipped=19.0
+2023-03-15 12:37:18,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1344998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:37:41,266 INFO [train.py:968] (0/2) Epoch 30, batch 24550, giga_loss[loss=0.2783, simple_loss=0.3507, pruned_loss=0.103, over 29007.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3707, pruned_loss=0.1192, over 5671896.63 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3537, pruned_loss=0.1064, over 5730760.86 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3731, pruned_loss=0.1207, over 5662012.22 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:37:41,524 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1345018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:38:21,747 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-15 12:38:31,291 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 12:38:34,676 INFO [train.py:968] (0/2) Epoch 30, batch 24600, libri_loss[loss=0.2745, simple_loss=0.3467, pruned_loss=0.1011, over 29556.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3694, pruned_loss=0.1177, over 5674193.83 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3538, pruned_loss=0.1065, over 5734461.49 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3716, pruned_loss=0.119, over 5661778.87 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:39:05,980 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.024e+03 1.640e+03 2.099e+03 3.051e+03 7.844e+03, threshold=4.197e+03, percent-clipped=6.0
+2023-03-15 12:39:09,207 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1345100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:39:11,280 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1345103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:39:26,297 INFO [train.py:968] (0/2) Epoch 30, batch 24650, libri_loss[loss=0.2901, simple_loss=0.3485, pruned_loss=0.1159, over 29365.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3718, pruned_loss=0.1169, over 5681070.49 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3541, pruned_loss=0.1069, over 5735589.38 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3736, pruned_loss=0.1177, over 5669288.81 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:39:33,982 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1345124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:39:36,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1345127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:39:43,388 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1345132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:40:10,122 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1345156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:40:26,340 INFO [train.py:968] (0/2) Epoch 30, batch 24700, giga_loss[loss=0.3816, simple_loss=0.416, pruned_loss=0.1736, over 26743.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3741, pruned_loss=0.1184, over 5667922.14 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3542, pruned_loss=0.1068, over 5737288.15 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3756, pruned_loss=0.1192, over 5656757.81 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:40:52,060 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9278, 1.3117, 1.1045, 0.2202], device='cuda:0'), covar=tensor([0.4844, 0.3832, 0.5084, 0.7602], device='cuda:0'), in_proj_covar=tensor([0.1867, 0.1744, 0.1674, 0.1517], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 12:40:54,138 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.319e+03 1.926e+03 2.619e+03 4.370e+03 9.654e+03, threshold=5.238e+03, percent-clipped=25.0
+2023-03-15 12:41:15,090 INFO [train.py:968] (0/2) Epoch 30, batch 24750, giga_loss[loss=0.3071, simple_loss=0.3898, pruned_loss=0.1122, over 28797.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3738, pruned_loss=0.1188, over 5665475.44 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3542, pruned_loss=0.107, over 5739166.38 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3755, pruned_loss=0.1196, over 5652945.44 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:41:57,456 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1345259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:42:07,857 INFO [train.py:968] (0/2) Epoch 30, batch 24800, giga_loss[loss=0.2898, simple_loss=0.3566, pruned_loss=0.1114, over 28067.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3718, pruned_loss=0.1185, over 5652015.19 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3543, pruned_loss=0.1073, over 5741174.82 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3735, pruned_loss=0.1191, over 5638467.52 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:42:35,716 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.837e+03 2.196e+03 3.246e+03 1.306e+04, threshold=4.392e+03, percent-clipped=6.0
+2023-03-15 12:42:54,651 INFO [train.py:968] (0/2) Epoch 30, batch 24850, giga_loss[loss=0.2679, simple_loss=0.3389, pruned_loss=0.0985, over 28556.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3706, pruned_loss=0.1187, over 5666608.00 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3545, pruned_loss=0.1074, over 5741903.14 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3721, pruned_loss=0.1192, over 5654104.99 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:43:40,095 INFO [train.py:968] (0/2) Epoch 30, batch 24900, giga_loss[loss=0.2816, simple_loss=0.3631, pruned_loss=0.1001, over 28875.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3701, pruned_loss=0.1191, over 5663127.02 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3549, pruned_loss=0.108, over 5738011.31 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3713, pruned_loss=0.1194, over 5654007.51 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:43:43,484 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1345373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:44:03,730 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5980, 1.5750, 1.7923, 1.4082], device='cuda:0'), covar=tensor([0.1679, 0.2477, 0.1400, 0.1678], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0720, 0.0989, 0.0889], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 12:44:04,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.823e+02 1.907e+03 2.334e+03 3.211e+03 1.278e+04, threshold=4.668e+03, percent-clipped=15.0
+2023-03-15 12:44:09,491 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1345402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:44:12,185 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1345405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:44:22,759 INFO [train.py:968] (0/2) Epoch 30, batch 24950, giga_loss[loss=0.2658, simple_loss=0.3512, pruned_loss=0.09016, over 29037.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.369, pruned_loss=0.117, over 5670053.04 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3554, pruned_loss=0.1083, over 5729365.49 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3699, pruned_loss=0.1171, over 5669228.52 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:44:41,146 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1345434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:45:16,336 INFO [train.py:968] (0/2) Epoch 30, batch 25000, giga_loss[loss=0.3401, simple_loss=0.3931, pruned_loss=0.1436, over 28427.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.368, pruned_loss=0.1159, over 5654829.65 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3557, pruned_loss=0.1086, over 5720910.15 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3686, pruned_loss=0.1159, over 5660050.34 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:45:41,921 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.487e+02 1.640e+03 2.038e+03 2.741e+03 9.108e+03, threshold=4.077e+03, percent-clipped=4.0
+2023-03-15 12:45:59,997 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1345516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:46:02,871 INFO [train.py:968] (0/2) Epoch 30, batch 25050, giga_loss[loss=0.3081, simple_loss=0.3765, pruned_loss=0.1199, over 29070.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3682, pruned_loss=0.1156, over 5647099.61 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3561, pruned_loss=0.1088, over 5703526.50 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3687, pruned_loss=0.1157, over 5664935.79 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:46:03,999 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1345519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:46:04,697 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:46:05,469 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:46:36,055 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1345548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:46:54,929 INFO [train.py:968] (0/2) Epoch 30, batch 25100, giga_loss[loss=0.2951, simple_loss=0.3702, pruned_loss=0.1101, over 28890.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3664, pruned_loss=0.1147, over 5654149.54 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3563, pruned_loss=0.1089, over 5696396.24 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3668, pruned_loss=0.1147, over 5673350.65 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:46:57,395 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-15 12:47:07,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4288, 1.6114, 1.2265, 1.1269], device='cuda:0'), covar=tensor([0.1071, 0.0547, 0.1088, 0.1172], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0454, 0.0526, 0.0466], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 12:47:27,850 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.193e+03 1.805e+03 2.238e+03 3.332e+03 9.193e+03, threshold=4.476e+03, percent-clipped=11.0
+2023-03-15 12:47:46,946 INFO [train.py:968] (0/2) Epoch 30, batch 25150, giga_loss[loss=0.2671, simple_loss=0.3409, pruned_loss=0.09663, over 28779.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.365, pruned_loss=0.1144, over 5661388.84 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3564, pruned_loss=0.1091, over 5699252.81 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3653, pruned_loss=0.1143, over 5673296.14 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:48:36,837 INFO [train.py:968] (0/2) Epoch 30, batch 25200, giga_loss[loss=0.3447, simple_loss=0.3994, pruned_loss=0.145, over 28857.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3654, pruned_loss=0.1153, over 5674331.10 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3568, pruned_loss=0.1093, over 5701428.12 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3654, pruned_loss=0.1151, over 5681234.41 frames. ], batch size: 243, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 12:49:08,098 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.461e+03 2.031e+03 2.405e+03 3.462e+03 8.471e+03, threshold=4.809e+03, percent-clipped=11.0
+2023-03-15 12:49:26,187 INFO [train.py:968] (0/2) Epoch 30, batch 25250, giga_loss[loss=0.3482, simple_loss=0.3922, pruned_loss=0.1521, over 28866.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3642, pruned_loss=0.1149, over 5687244.20 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.357, pruned_loss=0.1093, over 5704683.16 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3643, pruned_loss=0.1149, over 5689392.94 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:49:50,348 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345740.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 12:50:03,853 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4352, 1.6649, 1.4323, 1.6271], device='cuda:0'), covar=tensor([0.0812, 0.0328, 0.0338, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 12:50:11,297 INFO [train.py:968] (0/2) Epoch 30, batch 25300, giga_loss[loss=0.3281, simple_loss=0.3874, pruned_loss=0.1344, over 27978.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3622, pruned_loss=0.1142, over 5679698.08 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3569, pruned_loss=0.1094, over 5701205.45 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3625, pruned_loss=0.1143, over 5684512.71 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:50:44,651 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.881e+03 2.316e+03 3.256e+03 6.760e+03, threshold=4.632e+03, percent-clipped=8.0
+2023-03-15 12:51:03,917 INFO [train.py:968] (0/2) Epoch 30, batch 25350, giga_loss[loss=0.2647, simple_loss=0.3419, pruned_loss=0.0938, over 28338.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3616, pruned_loss=0.1145, over 5670176.06 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3568, pruned_loss=0.1094, over 5695129.99 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.362, pruned_loss=0.1146, over 5679378.67 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:51:22,644 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:51:51,467 INFO [train.py:968] (0/2) Epoch 30, batch 25400, libri_loss[loss=0.3023, simple_loss=0.371, pruned_loss=0.1168, over 29384.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1157, over 5672800.15 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3571, pruned_loss=0.1097, over 5697694.50 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3638, pruned_loss=0.1156, over 5677005.65 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:52:16,210 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1345895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:52:16,715 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1345896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:52:20,065 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.202e+03 1.947e+03 2.858e+03 4.332e+03 9.643e+03, threshold=5.717e+03, percent-clipped=19.0
+2023-03-15 12:52:34,973 INFO [train.py:968] (0/2) Epoch 30, batch 25450, giga_loss[loss=0.2599, simple_loss=0.3416, pruned_loss=0.08915, over 29049.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3641, pruned_loss=0.1149, over 5684541.19 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3572, pruned_loss=0.1098, over 5700227.51 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3641, pruned_loss=0.1149, over 5685219.54 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:52:51,642 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3195, 1.6173, 1.6147, 1.4401], device='cuda:0'), covar=tensor([0.1989, 0.1823, 0.2167, 0.1918], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0765, 0.0734, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 12:53:24,242 INFO [train.py:968] (0/2) Epoch 30, batch 25500, giga_loss[loss=0.3378, simple_loss=0.3831, pruned_loss=0.1462, over 26490.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.364, pruned_loss=0.1143, over 5687626.61 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3571, pruned_loss=0.1097, over 5704473.21 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3643, pruned_loss=0.1145, over 5684221.27 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:53:35,808 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:53:44,675 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-15 12:53:50,874 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:53:53,843 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.809e+03 2.410e+03 3.410e+03 6.911e+03, threshold=4.820e+03, percent-clipped=1.0
+2023-03-15 12:53:54,566 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1346000.pt
+2023-03-15 12:53:56,725 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1346002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:54:09,386 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-15 12:54:11,544 INFO [train.py:968] (0/2) Epoch 30, batch 25550, giga_loss[loss=0.3249, simple_loss=0.3835, pruned_loss=0.1332, over 29060.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3646, pruned_loss=0.1148, over 5677545.14 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3571, pruned_loss=0.1098, over 5697843.88 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3649, pruned_loss=0.1149, over 5680526.23 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:54:29,189 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 12:54:32,772 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:54:33,424 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:54:33,583 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-15 12:54:35,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:54:36,232 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:54:59,336 INFO [train.py:968] (0/2) Epoch 30, batch 25600, giga_loss[loss=0.3043, simple_loss=0.3693, pruned_loss=0.1196, over 28570.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3666, pruned_loss=0.1167, over 5683045.88 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3567, pruned_loss=0.1095, over 5703543.40 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3675, pruned_loss=0.1173, over 5679552.91 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:55:02,553 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:55:02,742 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-15 12:55:03,243 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:55:28,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.326e+03 2.131e+03 2.937e+03 4.142e+03 1.113e+04, threshold=5.874e+03, percent-clipped=13.0
+2023-03-15 12:55:47,670 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346115.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:55:49,780 INFO [train.py:968] (0/2) Epoch 30, batch 25650, libri_loss[loss=0.2565, simple_loss=0.3287, pruned_loss=0.09215, over 29567.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3679, pruned_loss=0.119, over 5682497.01 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.356, pruned_loss=0.109, over 5706875.32 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3695, pruned_loss=0.12, over 5676060.80 frames. ], batch size: 74, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:56:39,627 INFO [train.py:968] (0/2) Epoch 30, batch 25700, giga_loss[loss=0.3219, simple_loss=0.3817, pruned_loss=0.131, over 28238.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3699, pruned_loss=0.122, over 5677694.47 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3562, pruned_loss=0.1092, over 5708017.35 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3713, pruned_loss=0.1231, over 5671012.11 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:56:53,609 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 12:57:11,579 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.299e+03 1.908e+03 2.335e+03 2.912e+03 7.647e+03, threshold=4.671e+03, percent-clipped=2.0
+2023-03-15 12:57:23,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:57:26,736 INFO [train.py:968] (0/2) Epoch 30, batch 25750, libri_loss[loss=0.3358, simple_loss=0.3935, pruned_loss=0.139, over 19431.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3705, pruned_loss=0.1226, over 5662696.35 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3566, pruned_loss=0.1095, over 5684286.39 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3718, pruned_loss=0.1234, over 5678678.97 frames. ], batch size: 187, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:57:50,972 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1346247.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 12:58:01,756 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346258.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:58:04,291 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346261.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:58:11,531 INFO [train.py:968] (0/2) Epoch 30, batch 25800, libri_loss[loss=0.3135, simple_loss=0.3737, pruned_loss=0.1266, over 29543.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3696, pruned_loss=0.1225, over 5648978.06 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3573, pruned_loss=0.1102, over 5676319.21 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3703, pruned_loss=0.1229, over 5668073.12 frames. ], batch size: 81, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:58:38,214 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346290.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 12:58:45,527 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.256e+03 1.758e+03 2.208e+03 2.970e+03 7.288e+03, threshold=4.416e+03, percent-clipped=5.0
+2023-03-15 12:59:02,121 INFO [train.py:968] (0/2) Epoch 30, batch 25850, giga_loss[loss=0.3089, simple_loss=0.38, pruned_loss=0.1189, over 28916.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3688, pruned_loss=0.1207, over 5661103.02 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3573, pruned_loss=0.1102, over 5678499.84 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3694, pruned_loss=0.1211, over 5673957.10 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:59:29,376 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2089, 1.4756, 1.3862, 1.1719], device='cuda:0'), covar=tensor([0.3340, 0.3157, 0.2400, 0.2952], device='cuda:0'), in_proj_covar=tensor([0.2113, 0.2078, 0.1984, 0.2127], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 12:59:34,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:59:36,845 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:59:39,819 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:59:47,680 INFO [train.py:968] (0/2) Epoch 30, batch 25900, giga_loss[loss=0.3211, simple_loss=0.3802, pruned_loss=0.131, over 28712.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3675, pruned_loss=0.1188, over 5664394.34 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3567, pruned_loss=0.1099, over 5684509.36 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3689, pruned_loss=0.1197, over 5668495.69 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:59:48,649 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:59:56,822 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:00:07,024 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:00:14,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.254e+03 1.803e+03 2.366e+03 3.060e+03 6.295e+03, threshold=4.733e+03, percent-clipped=8.0
+2023-03-15 13:00:32,603 INFO [train.py:968] (0/2) Epoch 30, batch 25950, libri_loss[loss=0.2709, simple_loss=0.3423, pruned_loss=0.09976, over 29581.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3663, pruned_loss=0.1181, over 5658039.50 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3567, pruned_loss=0.11, over 5681506.86 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3677, pruned_loss=0.119, over 5662940.56 frames. ], batch size: 76, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:00:42,492 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6280, 1.6190, 1.7999, 1.4016], device='cuda:0'), covar=tensor([0.1814, 0.2699, 0.1508, 0.1796], device='cuda:0'), in_proj_covar=tensor([0.0939, 0.0720, 0.0987, 0.0888], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 13:01:19,985 INFO [train.py:968] (0/2) Epoch 30, batch 26000, giga_loss[loss=0.3031, simple_loss=0.3605, pruned_loss=0.1229, over 28557.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3635, pruned_loss=0.1163, over 5665471.70 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3569, pruned_loss=0.1101, over 5682972.28 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3645, pruned_loss=0.117, over 5667711.19 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 13:01:53,045 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:01:56,834 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.056e+03 1.809e+03 2.234e+03 2.729e+03 7.566e+03, threshold=4.468e+03, percent-clipped=4.0
+2023-03-15 13:01:57,128 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:08,821 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:12,786 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:14,701 INFO [train.py:968] (0/2) Epoch 30, batch 26050, giga_loss[loss=0.3362, simple_loss=0.393, pruned_loss=0.1397, over 27906.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3645, pruned_loss=0.118, over 5650090.19 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.357, pruned_loss=0.1103, over 5683250.04 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3652, pruned_loss=0.1185, over 5651483.91 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 13:02:18,325 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:22,759 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:26,437 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:39,239 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:47,019 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:53,082 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1235, 1.2791, 1.1124, 0.8979], device='cuda:0'), covar=tensor([0.1140, 0.0564, 0.1146, 0.1176], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0454, 0.0526, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 13:02:59,097 INFO [train.py:968] (0/2) Epoch 30, batch 26100, giga_loss[loss=0.3505, simple_loss=0.421, pruned_loss=0.14, over 28968.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3677, pruned_loss=0.1199, over 5650791.38 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3578, pruned_loss=0.1109, over 5679945.14 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3678, pruned_loss=0.12, over 5653826.12 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:03:03,061 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1220, 1.3405, 1.1830, 0.9826], device='cuda:0'), covar=tensor([0.1163, 0.0551, 0.1108, 0.1168], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0454, 0.0526, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 13:03:18,449 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 13:03:20,303 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.2392, 5.0959, 4.7858, 2.3507], device='cuda:0'), covar=tensor([0.0525, 0.0707, 0.0850, 0.1822], device='cuda:0'), in_proj_covar=tensor([0.1336, 0.1234, 0.1035, 0.0766], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 13:03:29,123 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.799e+03 2.199e+03 2.923e+03 6.796e+03, threshold=4.397e+03, percent-clipped=6.0
+2023-03-15 13:03:46,859 INFO [train.py:968] (0/2) Epoch 30, batch 26150, giga_loss[loss=0.2557, simple_loss=0.3487, pruned_loss=0.0814, over 28950.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3702, pruned_loss=0.1189, over 5644674.72 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3578, pruned_loss=0.1109, over 5665805.29 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3705, pruned_loss=0.1191, over 5658320.22 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:03:52,265 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346622.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 13:04:38,076 INFO [train.py:968] (0/2) Epoch 30, batch 26200, giga_loss[loss=0.2836, simple_loss=0.3628, pruned_loss=0.1022, over 28410.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3723, pruned_loss=0.1192, over 5643758.84 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3576, pruned_loss=0.1109, over 5657927.00 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3728, pruned_loss=0.1194, over 5661405.78 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:05:11,148 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.307e+03 1.744e+03 2.226e+03 3.115e+03 1.001e+04, threshold=4.452e+03, percent-clipped=10.0
+2023-03-15 13:05:29,376 INFO [train.py:968] (0/2) Epoch 30, batch 26250, giga_loss[loss=0.3259, simple_loss=0.3878, pruned_loss=0.1319, over 28941.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3749, pruned_loss=0.1216, over 5635932.27 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3577, pruned_loss=0.1109, over 5663222.89 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3756, pruned_loss=0.122, over 5644805.03 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:06:01,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1346754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:06:11,110 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346765.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 13:06:12,798 INFO [train.py:968] (0/2) Epoch 30, batch 26300, libri_loss[loss=0.247, simple_loss=0.3191, pruned_loss=0.08747, over 29582.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3748, pruned_loss=0.1223, over 5653522.93 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3569, pruned_loss=0.1104, over 5671708.23 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3768, pruned_loss=0.1234, over 5652319.25 frames. ], batch size: 75, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:06:13,033 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346768.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 13:06:21,981 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-15 13:06:45,173 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346797.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 13:06:47,567 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.214e+03 2.002e+03 2.509e+03 3.773e+03 8.856e+03, threshold=5.019e+03, percent-clipped=15.0
+2023-03-15 13:07:00,311 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1346814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:07:03,239 INFO [train.py:968] (0/2) Epoch 30, batch 26350, libri_loss[loss=0.2648, simple_loss=0.3363, pruned_loss=0.09665, over 29556.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3743, pruned_loss=0.1227, over 5640773.80 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3574, pruned_loss=0.1107, over 5664611.44 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3758, pruned_loss=0.1236, over 5645968.10 frames. ], batch size: 76, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:07:51,380 INFO [train.py:968] (0/2) Epoch 30, batch 26400, giga_loss[loss=0.3226, simple_loss=0.3858, pruned_loss=0.1297, over 28587.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3727, pruned_loss=0.1221, over 5635924.17 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3576, pruned_loss=0.1109, over 5663164.35 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3743, pruned_loss=0.1231, over 5640785.89 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:08:21,034 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.338e+03 1.896e+03 2.403e+03 3.027e+03 1.525e+04, threshold=4.805e+03, percent-clipped=5.0
+2023-03-15 13:08:38,673 INFO [train.py:968] (0/2) Epoch 30, batch 26450, giga_loss[loss=0.3765, simple_loss=0.4091, pruned_loss=0.172, over 26551.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3703, pruned_loss=0.1212, over 5643726.97 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.358, pruned_loss=0.1112, over 5664194.21 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3714, pruned_loss=0.1219, over 5646321.35 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:08:38,919 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4001, 3.6025, 1.5034, 1.7379], device='cuda:0'), covar=tensor([0.1038, 0.0364, 0.0915, 0.1290], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0582, 0.0418, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 13:08:55,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4472, 1.4490, 1.2364, 1.4975], device='cuda:0'), covar=tensor([0.0774, 0.0362, 0.0359, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0124, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 13:09:02,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3532, 1.1310, 3.8413, 3.2413], device='cuda:0'), covar=tensor([0.1664, 0.2905, 0.0483, 0.1104], device='cuda:0'), in_proj_covar=tensor([0.0821, 0.0684, 0.1025, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 13:09:05,368 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-15 13:09:29,068 INFO [train.py:968] (0/2) Epoch 30, batch 26500, giga_loss[loss=0.2988, simple_loss=0.3638, pruned_loss=0.1169, over 28597.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3688, pruned_loss=0.121, over 5645294.57 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3576, pruned_loss=0.111, over 5669378.06 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3703, pruned_loss=0.1219, over 5642277.30 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:09:32,190 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4915, 1.8051, 1.2613, 1.3503], device='cuda:0'), covar=tensor([0.1117, 0.0599, 0.1116, 0.1171], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0456, 0.0528, 0.0469], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 13:09:43,742 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4141, 1.6397, 1.6335, 1.4810], device='cuda:0'), covar=tensor([0.1812, 0.1785, 0.1832, 0.1761], device='cuda:0'), in_proj_covar=tensor([0.0516, 0.0764, 0.0734, 0.0704], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 13:10:02,038 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.049e+03 1.938e+03 2.458e+03 3.293e+03 9.907e+03, threshold=4.917e+03, percent-clipped=10.0
+2023-03-15 13:10:17,815 INFO [train.py:968] (0/2) Epoch 30, batch 26550, libri_loss[loss=0.3049, simple_loss=0.3673, pruned_loss=0.1212, over 18864.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3685, pruned_loss=0.1208, over 5639096.32 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3577, pruned_loss=0.1111, over 5665420.43 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3698, pruned_loss=0.1217, over 5640275.72 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:10:22,225 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1347024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:10:24,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7883, 2.6951, 1.7304, 0.9981], device='cuda:0'), covar=tensor([0.9713, 0.3931, 0.4597, 0.8103], device='cuda:0'), in_proj_covar=tensor([0.1874, 0.1756, 0.1676, 0.1516], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 13:10:57,416 INFO [train.py:968] (0/2) Epoch 30, batch 26600, giga_loss[loss=0.2537, simple_loss=0.3301, pruned_loss=0.08869, over 28950.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.365, pruned_loss=0.1181, over 5659692.59 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3564, pruned_loss=0.1102, over 5678931.50 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3678, pruned_loss=0.1202, over 5647168.04 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:11:27,835 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.051e+03 1.934e+03 2.691e+03 3.608e+03 1.713e+04, threshold=5.381e+03, percent-clipped=11.0
+2023-03-15 13:11:28,925 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1347103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:11:39,121 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-15 13:11:44,336 INFO [train.py:968] (0/2) Epoch 30, batch 26650, giga_loss[loss=0.3034, simple_loss=0.3634, pruned_loss=0.1217, over 27673.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3635, pruned_loss=0.1175, over 5662855.65 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3561, pruned_loss=0.11, over 5670868.69 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3662, pruned_loss=0.1194, over 5659203.99 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:11:54,900 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1347129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:12:34,324 INFO [train.py:968] (0/2) Epoch 30, batch 26700, giga_loss[loss=0.325, simple_loss=0.3792, pruned_loss=0.1354, over 28024.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3631, pruned_loss=0.117, over 5669496.34 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3563, pruned_loss=0.1102, over 5674385.39 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3652, pruned_loss=0.1185, over 5663515.00 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:12:52,454 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1347189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:13:04,846 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.280e+03 1.831e+03 2.324e+03 2.952e+03 1.139e+04, threshold=4.649e+03, percent-clipped=4.0
+2023-03-15 13:13:19,841 INFO [train.py:968] (0/2) Epoch 30, batch 26750, giga_loss[loss=0.2941, simple_loss=0.3646, pruned_loss=0.1118, over 28723.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3658, pruned_loss=0.1183, over 5664843.30 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3566, pruned_loss=0.1104, over 5673624.40 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3674, pruned_loss=0.1196, over 5659993.84 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:13:29,994 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0208, 1.2987, 1.0754, 0.3004], device='cuda:0'), covar=tensor([0.3996, 0.3111, 0.4119, 0.6632], device='cuda:0'), in_proj_covar=tensor([0.1881, 0.1763, 0.1681, 0.1522], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 13:14:10,830 INFO [train.py:968] (0/2) Epoch 30, batch 26800, giga_loss[loss=0.3172, simple_loss=0.3886, pruned_loss=0.1229, over 28959.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3681, pruned_loss=0.1194, over 5666507.85 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3566, pruned_loss=0.1104, over 5674845.28 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3695, pruned_loss=0.1205, over 5661510.57 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:14:17,205 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1347272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:14:22,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1347275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:14:47,179 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.175e+03 1.805e+03 2.301e+03 3.096e+03 7.030e+03, threshold=4.602e+03, percent-clipped=4.0
+2023-03-15 13:14:50,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1347304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:14:54,182 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1143, 3.9575, 3.7320, 1.7925], device='cuda:0'), covar=tensor([0.0777, 0.0927, 0.0901, 0.2254], device='cuda:0'), in_proj_covar=tensor([0.1343, 0.1240, 0.1039, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 13:15:00,790 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3618, 1.6054, 1.5422, 1.4390], device='cuda:0'), covar=tensor([0.1990, 0.1967, 0.2450, 0.2070], device='cuda:0'), in_proj_covar=tensor([0.0516, 0.0764, 0.0736, 0.0706], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 13:15:02,557 INFO [train.py:968] (0/2) Epoch 30, batch 26850, giga_loss[loss=0.2619, simple_loss=0.3486, pruned_loss=0.08763, over 29116.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3685, pruned_loss=0.1206, over 5662244.75 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3566, pruned_loss=0.1103, over 5677301.02 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3697, pruned_loss=0.1216, over 5655937.48 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:15:16,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1347332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:15:19,757 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1347335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:15:48,923 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1347364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:15:54,140 INFO [train.py:968] (0/2) Epoch 30, batch 26900, giga_loss[loss=0.2802, simple_loss=0.36, pruned_loss=0.1002, over 28560.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3696, pruned_loss=0.1179, over 5674269.66 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3566, pruned_loss=0.1103, over 5677301.02 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3705, pruned_loss=0.1187, over 5669360.63 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:16:27,103 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1347399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:16:29,986 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.099e+03 1.714e+03 2.171e+03 2.986e+03 7.087e+03, threshold=4.341e+03, percent-clipped=6.0
+2023-03-15 13:16:36,002 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 13:16:42,512 INFO [train.py:968] (0/2) Epoch 30, batch 26950, giga_loss[loss=0.3323, simple_loss=0.4102, pruned_loss=0.1272, over 28942.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3716, pruned_loss=0.1178, over 5666344.98 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3571, pruned_loss=0.1106, over 5674927.07 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3721, pruned_loss=0.1182, over 5664271.87 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:16:58,367 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-15 13:17:01,817 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1372, 1.2026, 3.8389, 3.3223], device='cuda:0'), covar=tensor([0.2261, 0.3368, 0.0917, 0.1078], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0685, 0.1026, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 13:17:24,361 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-15 13:17:32,207 INFO [train.py:968] (0/2) Epoch 30, batch 27000, giga_loss[loss=0.3092, simple_loss=0.3818, pruned_loss=0.1183, over 28992.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.374, pruned_loss=0.1189, over 5672218.39 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3568, pruned_loss=0.1105, over 5676554.57 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3749, pruned_loss=0.1195, over 5669078.90 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:17:32,212 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 13:17:41,089 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.202, simple_loss=0.3097, pruned_loss=0.04714, over 944034.00 frames. 
+2023-03-15 13:17:41,090 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 13:17:49,216 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1347478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:18:14,074 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.193e+03 1.847e+03 2.672e+03 3.709e+03 1.125e+04, threshold=5.345e+03, percent-clipped=21.0
+2023-03-15 13:18:29,897 INFO [train.py:968] (0/2) Epoch 30, batch 27050, giga_loss[loss=0.3173, simple_loss=0.3805, pruned_loss=0.1271, over 28620.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3776, pruned_loss=0.123, over 5665096.78 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3567, pruned_loss=0.1105, over 5672604.54 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3788, pruned_loss=0.1238, over 5665752.31 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:18:54,331 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1347542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:18:56,904 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1347545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:19:22,725 INFO [train.py:968] (0/2) Epoch 30, batch 27100, giga_loss[loss=0.2963, simple_loss=0.3606, pruned_loss=0.116, over 28716.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3778, pruned_loss=0.1236, over 5677952.38 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3569, pruned_loss=0.1105, over 5676307.77 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.379, pruned_loss=0.1244, over 5675131.93 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:19:28,126 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1347574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:19:56,506 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.210e+03 1.760e+03 2.280e+03 2.779e+03 6.643e+03, threshold=4.560e+03, percent-clipped=1.0
+2023-03-15 13:19:59,191 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4100, 1.4903, 1.5075, 1.3467], device='cuda:0'), covar=tensor([0.2889, 0.2784, 0.2587, 0.2701], device='cuda:0'), in_proj_covar=tensor([0.2093, 0.2062, 0.1973, 0.2115], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 13:20:13,441 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2918, 1.1633, 3.7963, 3.4415], device='cuda:0'), covar=tensor([0.1683, 0.2920, 0.0478, 0.1079], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0684, 0.1026, 0.1001], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 13:20:13,773 INFO [train.py:968] (0/2) Epoch 30, batch 27150, giga_loss[loss=0.2902, simple_loss=0.3641, pruned_loss=0.1081, over 28676.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3779, pruned_loss=0.1246, over 5670131.80 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3571, pruned_loss=0.1107, over 5677806.80 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3789, pruned_loss=0.1252, over 5666594.05 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:20:16,148 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1347621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:20:19,331 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1347624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:20:47,299 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1347653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:20:55,790 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 13:21:00,962 INFO [train.py:968] (0/2) Epoch 30, batch 27200, giga_loss[loss=0.2757, simple_loss=0.3627, pruned_loss=0.09432, over 28627.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.375, pruned_loss=0.1216, over 5671860.38 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3572, pruned_loss=0.1108, over 5669859.84 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3761, pruned_loss=0.1223, over 5675491.40 frames. ], batch size: 78, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:21:32,249 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.49 vs. limit=5.0
+2023-03-15 13:21:33,298 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.047e+03 1.680e+03 2.064e+03 2.783e+03 8.762e+03, threshold=4.128e+03, percent-clipped=6.0
+2023-03-15 13:21:43,709 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9373, 2.2217, 2.2574, 1.7421], device='cuda:0'), covar=tensor([0.3834, 0.2808, 0.2967, 0.3506], device='cuda:0'), in_proj_covar=tensor([0.2095, 0.2063, 0.1975, 0.2116], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 13:21:46,214 INFO [train.py:968] (0/2) Epoch 30, batch 27250, giga_loss[loss=0.3269, simple_loss=0.3889, pruned_loss=0.1324, over 27927.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3737, pruned_loss=0.1191, over 5668014.88 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3568, pruned_loss=0.1105, over 5674978.06 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3753, pruned_loss=0.12, over 5666382.74 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:21:52,669 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6406, 1.7229, 1.2953, 1.5004], device='cuda:0'), covar=tensor([0.1069, 0.0739, 0.1086, 0.1239], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0458, 0.0530, 0.0469], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 13:22:11,584 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.9778, 3.7749, 3.6197, 1.8329], device='cuda:0'), covar=tensor([0.0797, 0.0951, 0.0907, 0.2066], device='cuda:0'), in_proj_covar=tensor([0.1347, 0.1241, 0.1041, 0.0768], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 13:22:25,804 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1347754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:22:40,221 INFO [train.py:968] (0/2) Epoch 30, batch 27300, giga_loss[loss=0.2779, simple_loss=0.3649, pruned_loss=0.09548, over 28868.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3755, pruned_loss=0.1197, over 5658860.15 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3572, pruned_loss=0.1109, over 5667990.35 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3766, pruned_loss=0.1202, over 5664151.74 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:23:21,047 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.148e+03 1.801e+03 2.366e+03 3.107e+03 6.685e+03, threshold=4.732e+03, percent-clipped=10.0
+2023-03-15 13:23:34,331 INFO [train.py:968] (0/2) Epoch 30, batch 27350, giga_loss[loss=0.3217, simple_loss=0.3813, pruned_loss=0.1311, over 27529.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3762, pruned_loss=0.1209, over 5652791.31 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3567, pruned_loss=0.1107, over 5671624.99 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3778, pruned_loss=0.1217, over 5653444.79 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:23:43,959 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3058, 1.2566, 3.8255, 3.2187], device='cuda:0'), covar=tensor([0.1715, 0.2935, 0.0483, 0.1273], device='cuda:0'), in_proj_covar=tensor([0.0824, 0.0687, 0.1028, 0.1005], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 13:24:21,480 INFO [train.py:968] (0/2) Epoch 30, batch 27400, giga_loss[loss=0.2802, simple_loss=0.3544, pruned_loss=0.103, over 28592.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3745, pruned_loss=0.12, over 5662334.65 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3569, pruned_loss=0.1107, over 5676009.83 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3759, pruned_loss=0.1208, over 5658632.27 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:24:58,144 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.201e+03 1.710e+03 2.314e+03 3.185e+03 6.672e+03, threshold=4.627e+03, percent-clipped=6.0
+2023-03-15 13:25:15,644 INFO [train.py:968] (0/2) Epoch 30, batch 27450, giga_loss[loss=0.3268, simple_loss=0.3814, pruned_loss=0.1361, over 27586.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3737, pruned_loss=0.1206, over 5664819.41 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3575, pruned_loss=0.111, over 5676453.37 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3748, pruned_loss=0.1212, over 5660959.04 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:26:02,443 INFO [train.py:968] (0/2) Epoch 30, batch 27500, giga_loss[loss=0.2823, simple_loss=0.3493, pruned_loss=0.1076, over 28543.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3722, pruned_loss=0.1202, over 5675791.49 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3572, pruned_loss=0.1108, over 5682331.75 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3739, pruned_loss=0.1213, over 5666780.01 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:26:17,111 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3515, 2.2059, 1.4294, 1.4575], device='cuda:0'), covar=tensor([0.0844, 0.0441, 0.0716, 0.1197], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0585, 0.0420, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 13:26:32,368 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1348000.pt
+2023-03-15 13:26:35,487 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 13:26:35,578 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.957e+03 2.599e+03 3.925e+03 1.465e+04, threshold=5.198e+03, percent-clipped=13.0
+2023-03-15 13:26:49,651 INFO [train.py:968] (0/2) Epoch 30, batch 27550, libri_loss[loss=0.2806, simple_loss=0.3616, pruned_loss=0.09983, over 29004.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.37, pruned_loss=0.1194, over 5677881.59 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3575, pruned_loss=0.1109, over 5691241.68 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3717, pruned_loss=0.1205, over 5662170.15 frames. ], batch size: 101, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:27:34,288 INFO [train.py:968] (0/2) Epoch 30, batch 27600, giga_loss[loss=0.2878, simple_loss=0.3605, pruned_loss=0.1075, over 28805.00 frames. ], tot_loss[loss=0.303, simple_loss=0.368, pruned_loss=0.119, over 5676766.83 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3568, pruned_loss=0.1106, over 5696351.16 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3704, pruned_loss=0.1205, over 5658850.55 frames. ], batch size: 285, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:28:06,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.205e+03 1.838e+03 2.286e+03 3.037e+03 8.021e+03, threshold=4.573e+03, percent-clipped=5.0
+2023-03-15 13:28:14,241 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1348112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:28:19,214 INFO [train.py:968] (0/2) Epoch 30, batch 27650, giga_loss[loss=0.3082, simple_loss=0.3724, pruned_loss=0.122, over 27948.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3682, pruned_loss=0.1197, over 5670869.80 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3572, pruned_loss=0.1109, over 5697072.10 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.37, pruned_loss=0.1208, over 5655725.08 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:28:29,071 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1348129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:28:32,943 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.8706, 1.2257, 1.3814, 1.0233], device='cuda:0'), covar=tensor([0.2277, 0.1527, 0.2383, 0.1896], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0762, 0.0736, 0.0705], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 13:29:02,336 INFO [train.py:968] (0/2) Epoch 30, batch 27700, giga_loss[loss=0.2541, simple_loss=0.3388, pruned_loss=0.08469, over 28240.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3659, pruned_loss=0.1167, over 5675849.51 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3575, pruned_loss=0.1111, over 5699686.02 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3673, pruned_loss=0.1175, over 5660865.60 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:29:06,020 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3593, 1.3624, 1.1563, 1.5524], device='cuda:0'), covar=tensor([0.0795, 0.0381, 0.0378, 0.0895], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0124, 0.0121, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0077, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 13:29:14,149 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8191, 2.8537, 1.8387, 1.0919], device='cuda:0'), covar=tensor([1.0453, 0.4166, 0.4817, 0.8694], device='cuda:0'), in_proj_covar=tensor([0.1885, 0.1764, 0.1684, 0.1525], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 13:29:20,983 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5537, 1.7803, 1.4674, 1.5330], device='cuda:0'), covar=tensor([0.2808, 0.3007, 0.3428, 0.2619], device='cuda:0'), in_proj_covar=tensor([0.1627, 0.1172, 0.1440, 0.1023], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 13:29:34,776 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.067e+03 1.675e+03 2.329e+03 3.085e+03 1.134e+04, threshold=4.659e+03, percent-clipped=7.0
+2023-03-15 13:29:51,193 INFO [train.py:968] (0/2) Epoch 30, batch 27750, giga_loss[loss=0.2707, simple_loss=0.3475, pruned_loss=0.09702, over 28983.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3636, pruned_loss=0.1144, over 5674513.34 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.358, pruned_loss=0.1114, over 5703949.90 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3644, pruned_loss=0.1149, over 5658376.13 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:30:39,801 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5939, 1.7757, 1.8417, 1.5282], device='cuda:0'), covar=tensor([0.3042, 0.2709, 0.2836, 0.2870], device='cuda:0'), in_proj_covar=tensor([0.2095, 0.2069, 0.1975, 0.2119], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 13:30:40,118 INFO [train.py:968] (0/2) Epoch 30, batch 27800, giga_loss[loss=0.3086, simple_loss=0.3736, pruned_loss=0.1218, over 27806.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3621, pruned_loss=0.1133, over 5674531.08 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3577, pruned_loss=0.1112, over 5708940.47 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3631, pruned_loss=0.114, over 5656131.03 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:30:45,586 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1348272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:30:47,416 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1348275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:31:00,815 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5388, 1.6486, 1.2453, 1.2377], device='cuda:0'), covar=tensor([0.0927, 0.0508, 0.0928, 0.1287], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0458, 0.0529, 0.0469], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 13:31:18,959 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.800e+03 2.296e+03 3.428e+03 1.058e+04, threshold=4.592e+03, percent-clipped=7.0
+2023-03-15 13:31:19,229 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1348304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:31:36,114 INFO [train.py:968] (0/2) Epoch 30, batch 27850, giga_loss[loss=0.3663, simple_loss=0.3994, pruned_loss=0.1666, over 26773.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3591, pruned_loss=0.1125, over 5665896.49 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3577, pruned_loss=0.1111, over 5711061.46 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.36, pruned_loss=0.1131, over 5649313.08 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:32:30,790 INFO [train.py:968] (0/2) Epoch 30, batch 27900, giga_loss[loss=0.382, simple_loss=0.4151, pruned_loss=0.1744, over 26801.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3585, pruned_loss=0.1131, over 5645988.17 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3575, pruned_loss=0.111, over 5694928.13 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3594, pruned_loss=0.1137, over 5645434.70 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:32:51,652 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3917, 1.3148, 3.6274, 3.2478], device='cuda:0'), covar=tensor([0.1536, 0.2682, 0.0511, 0.1047], device='cuda:0'), in_proj_covar=tensor([0.0826, 0.0687, 0.1031, 0.1009], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 13:33:04,428 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.310e+03 2.150e+03 2.737e+03 4.037e+03 1.250e+04, threshold=5.475e+03, percent-clipped=17.0
+2023-03-15 13:33:17,568 INFO [train.py:968] (0/2) Epoch 30, batch 27950, giga_loss[loss=0.2766, simple_loss=0.3582, pruned_loss=0.09751, over 28861.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3598, pruned_loss=0.1126, over 5652269.11 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3571, pruned_loss=0.1107, over 5691864.44 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3609, pruned_loss=0.1134, over 5653212.41 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:33:32,379 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5912, 1.7042, 1.8536, 1.5439], device='cuda:0'), covar=tensor([0.2755, 0.2951, 0.2734, 0.3015], device='cuda:0'), in_proj_covar=tensor([0.2101, 0.2072, 0.1979, 0.2124], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 13:34:01,148 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-15 13:34:06,384 INFO [train.py:968] (0/2) Epoch 30, batch 28000, giga_loss[loss=0.264, simple_loss=0.3476, pruned_loss=0.09018, over 28873.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3627, pruned_loss=0.1143, over 5647393.15 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3571, pruned_loss=0.1106, over 5692159.41 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3637, pruned_loss=0.1152, over 5647221.69 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:34:14,466 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1348475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:34:16,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2855, 1.8309, 1.3315, 0.5903], device='cuda:0'), covar=tensor([0.6422, 0.3195, 0.4617, 0.7507], device='cuda:0'), in_proj_covar=tensor([0.1880, 0.1760, 0.1679, 0.1521], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 13:34:27,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1348487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:34:43,501 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.635e+03 1.972e+03 2.762e+03 8.730e+03, threshold=3.945e+03, percent-clipped=2.0
+2023-03-15 13:34:44,763 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-15 13:34:54,742 INFO [train.py:968] (0/2) Epoch 30, batch 28050, giga_loss[loss=0.2713, simple_loss=0.3453, pruned_loss=0.09867, over 28660.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3634, pruned_loss=0.115, over 5647489.24 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3567, pruned_loss=0.1103, over 5693183.22 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3646, pruned_loss=0.116, over 5646028.33 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:35:43,385 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8749, 2.0754, 2.1128, 1.6688], device='cuda:0'), covar=tensor([0.3282, 0.2846, 0.2726, 0.3217], device='cuda:0'), in_proj_covar=tensor([0.2098, 0.2070, 0.1974, 0.2121], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 13:35:43,703 INFO [train.py:968] (0/2) Epoch 30, batch 28100, giga_loss[loss=0.289, simple_loss=0.3587, pruned_loss=0.1096, over 28777.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3639, pruned_loss=0.1158, over 5646418.59 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3569, pruned_loss=0.1103, over 5694482.64 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3648, pruned_loss=0.1167, over 5643049.23 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:36:12,941 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1348601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:36:15,708 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.091e+03 1.771e+03 2.273e+03 2.917e+03 5.195e+03, threshold=4.546e+03, percent-clipped=5.0
+2023-03-15 13:36:28,728 INFO [train.py:968] (0/2) Epoch 30, batch 28150, giga_loss[loss=0.2997, simple_loss=0.3688, pruned_loss=0.1153, over 28594.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3656, pruned_loss=0.1174, over 5639938.30 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3569, pruned_loss=0.1102, over 5691470.39 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3665, pruned_loss=0.1182, over 5639188.80 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:36:42,084 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1348630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:36:45,388 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1348633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:37:11,861 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1348662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:37:15,580 INFO [train.py:968] (0/2) Epoch 30, batch 28200, libri_loss[loss=0.2544, simple_loss=0.3257, pruned_loss=0.09159, over 29383.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3673, pruned_loss=0.118, over 5651829.85 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3573, pruned_loss=0.1106, over 5693602.78 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3679, pruned_loss=0.1186, over 5648017.01 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:37:15,954 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1869, 1.3259, 3.2649, 2.9741], device='cuda:0'), covar=tensor([0.1510, 0.2597, 0.0513, 0.1872], device='cuda:0'), in_proj_covar=tensor([0.0824, 0.0685, 0.1026, 0.1006], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 13:37:50,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.320e+03 1.879e+03 2.478e+03 3.528e+03 8.620e+03, threshold=4.957e+03, percent-clipped=14.0
+2023-03-15 13:38:04,709 INFO [train.py:968] (0/2) Epoch 30, batch 28250, giga_loss[loss=0.4081, simple_loss=0.4335, pruned_loss=0.1914, over 26686.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3698, pruned_loss=0.1205, over 5641799.99 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3576, pruned_loss=0.1108, over 5687966.70 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3702, pruned_loss=0.121, over 5643345.91 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:38:55,395 INFO [train.py:968] (0/2) Epoch 30, batch 28300, giga_loss[loss=0.3037, simple_loss=0.371, pruned_loss=0.1182, over 28971.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3713, pruned_loss=0.1222, over 5639870.22 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3576, pruned_loss=0.1107, over 5689971.39 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3718, pruned_loss=0.1227, over 5638659.06 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:39:34,518 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.689e+03 2.144e+03 2.922e+03 6.571e+03, threshold=4.289e+03, percent-clipped=3.0
+2023-03-15 13:39:48,625 INFO [train.py:968] (0/2) Epoch 30, batch 28350, giga_loss[loss=0.3308, simple_loss=0.3946, pruned_loss=0.1336, over 28074.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3711, pruned_loss=0.1206, over 5644483.26 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3573, pruned_loss=0.1105, over 5694266.79 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.372, pruned_loss=0.1215, over 5638911.71 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:40:23,842 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1348850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:40:37,917 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6072, 1.9014, 1.9846, 1.5870], device='cuda:0'), covar=tensor([0.3234, 0.2784, 0.2789, 0.2966], device='cuda:0'), in_proj_covar=tensor([0.2102, 0.2073, 0.1978, 0.2121], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 13:40:41,098 INFO [train.py:968] (0/2) Epoch 30, batch 28400, libri_loss[loss=0.3007, simple_loss=0.3628, pruned_loss=0.1193, over 29548.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3701, pruned_loss=0.1192, over 5655715.42 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3575, pruned_loss=0.1107, over 5697970.53 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3709, pruned_loss=0.1198, over 5647102.11 frames. ], batch size: 89, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:40:50,531 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1348877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:40:59,119 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-15 13:41:17,180 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.9298, 1.2940, 5.0626, 3.5966], device='cuda:0'), covar=tensor([0.1557, 0.2959, 0.0419, 0.0919], device='cuda:0'), in_proj_covar=tensor([0.0824, 0.0686, 0.1027, 0.1007], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 13:41:20,207 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.361e+03 1.977e+03 2.407e+03 3.360e+03 8.119e+03, threshold=4.813e+03, percent-clipped=12.0
+2023-03-15 13:41:29,623 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4911, 3.7022, 1.6335, 1.6953], device='cuda:0'), covar=tensor([0.0929, 0.0405, 0.0888, 0.1277], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0582, 0.0418, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 13:41:32,263 INFO [train.py:968] (0/2) Epoch 30, batch 28450, giga_loss[loss=0.3299, simple_loss=0.384, pruned_loss=0.138, over 27592.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3719, pruned_loss=0.1216, over 5637774.00 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3579, pruned_loss=0.1112, over 5698989.61 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3723, pruned_loss=0.1218, over 5628975.90 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:42:25,438 INFO [train.py:968] (0/2) Epoch 30, batch 28500, giga_loss[loss=0.3366, simple_loss=0.3931, pruned_loss=0.1401, over 28225.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1224, over 5640128.42 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3578, pruned_loss=0.1112, over 5702305.84 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1227, over 5628935.01 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:42:33,414 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1348976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:42:40,270 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1348982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:42:52,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5448, 3.1026, 1.6088, 1.5990], device='cuda:0'), covar=tensor([0.0816, 0.0360, 0.0728, 0.1161], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0582, 0.0419, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 13:42:53,791 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1348993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:42:57,101 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1348996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:42:59,252 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7009, 1.9320, 1.5959, 1.8625], device='cuda:0'), covar=tensor([0.2682, 0.2833, 0.3071, 0.2664], device='cuda:0'), in_proj_covar=tensor([0.1630, 0.1173, 0.1441, 0.1023], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 13:43:06,894 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.284e+03 1.863e+03 2.152e+03 3.054e+03 1.231e+04, threshold=4.303e+03, percent-clipped=7.0
+2023-03-15 13:43:16,367 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-15 13:43:22,193 INFO [train.py:968] (0/2) Epoch 30, batch 28550, giga_loss[loss=0.2688, simple_loss=0.3307, pruned_loss=0.1034, over 28916.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3697, pruned_loss=0.1218, over 5633880.14 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3575, pruned_loss=0.111, over 5710233.15 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3711, pruned_loss=0.1228, over 5615622.36 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:43:27,664 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1349025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:44:07,137 INFO [train.py:968] (0/2) Epoch 30, batch 28600, giga_loss[loss=0.3221, simple_loss=0.3734, pruned_loss=0.1354, over 28849.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3683, pruned_loss=0.1209, over 5650943.37 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3576, pruned_loss=0.111, over 5713709.12 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3695, pruned_loss=0.1218, over 5632033.30 frames. ], batch size: 66, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:44:20,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2798, 0.7430, 0.8074, 1.4077], device='cuda:0'), covar=tensor([0.0772, 0.0411, 0.0385, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0124, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 13:44:25,751 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 13:44:41,701 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3932, 4.2724, 4.0482, 2.0448], device='cuda:0'), covar=tensor([0.0625, 0.0713, 0.0769, 0.1922], device='cuda:0'), in_proj_covar=tensor([0.1342, 0.1238, 0.1038, 0.0763], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 13:44:49,050 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.121e+02 1.906e+03 2.824e+03 3.964e+03 2.247e+04, threshold=5.649e+03, percent-clipped=21.0
+2023-03-15 13:44:54,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6652, 1.9546, 1.4657, 1.5843], device='cuda:0'), covar=tensor([0.0983, 0.0477, 0.0968, 0.0998], device='cuda:0'), in_proj_covar=tensor([0.0422, 0.0459, 0.0529, 0.0470], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 13:44:58,755 INFO [train.py:968] (0/2) Epoch 30, batch 28650, giga_loss[loss=0.2786, simple_loss=0.3542, pruned_loss=0.1015, over 28977.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3678, pruned_loss=0.1208, over 5660496.37 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3572, pruned_loss=0.1107, over 5717465.43 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3693, pruned_loss=0.122, over 5640827.59 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:44:59,852 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1349119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:45:01,779 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1349122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:45:32,229 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1349151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:45:40,009 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-15 13:45:48,518 INFO [train.py:968] (0/2) Epoch 30, batch 28700, giga_loss[loss=0.336, simple_loss=0.3904, pruned_loss=0.1408, over 28742.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3681, pruned_loss=0.1213, over 5663453.28 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3577, pruned_loss=0.1112, over 5720348.39 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3691, pruned_loss=0.122, over 5644355.18 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:46:23,859 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.116e+03 1.943e+03 2.415e+03 2.971e+03 7.510e+03, threshold=4.830e+03, percent-clipped=2.0
+2023-03-15 13:46:34,285 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9168, 2.0435, 2.1560, 1.6295], device='cuda:0'), covar=tensor([0.3668, 0.2991, 0.2832, 0.3316], device='cuda:0'), in_proj_covar=tensor([0.2096, 0.2069, 0.1974, 0.2117], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 13:46:35,348 INFO [train.py:968] (0/2) Epoch 30, batch 28750, giga_loss[loss=0.2923, simple_loss=0.3688, pruned_loss=0.1079, over 28892.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3687, pruned_loss=0.1216, over 5670234.96 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3576, pruned_loss=0.111, over 5725869.27 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3699, pruned_loss=0.1226, over 5648651.50 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:46:39,137 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4035, 4.2633, 4.0531, 1.9856], device='cuda:0'), covar=tensor([0.0629, 0.0715, 0.0794, 0.2084], device='cuda:0'), in_proj_covar=tensor([0.1344, 0.1242, 0.1042, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 13:46:47,634 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4015, 4.2618, 4.0341, 1.9856], device='cuda:0'), covar=tensor([0.0629, 0.0748, 0.0818, 0.2102], device='cuda:0'), in_proj_covar=tensor([0.1344, 0.1242, 0.1042, 0.0767], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 13:47:08,018 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1349252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:47:26,803 INFO [train.py:968] (0/2) Epoch 30, batch 28800, giga_loss[loss=0.2617, simple_loss=0.3378, pruned_loss=0.09279, over 29065.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3711, pruned_loss=0.1233, over 5656074.65 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3578, pruned_loss=0.1111, over 5714284.03 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.372, pruned_loss=0.1242, over 5649268.34 frames. ], batch size: 113, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:48:05,613 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9964, 4.9484, 2.1217, 2.1819], device='cuda:0'), covar=tensor([0.0911, 0.0356, 0.0797, 0.1185], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0582, 0.0418, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 13:48:06,938 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 2.061e+03 2.709e+03 3.213e+03 1.077e+04, threshold=5.417e+03, percent-clipped=6.0
+2023-03-15 13:48:17,882 INFO [train.py:968] (0/2) Epoch 30, batch 28850, giga_loss[loss=0.3515, simple_loss=0.4012, pruned_loss=0.1509, over 28612.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3717, pruned_loss=0.124, over 5665929.73 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3578, pruned_loss=0.1111, over 5716477.08 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3725, pruned_loss=0.1247, over 5658313.99 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:48:27,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5415, 1.7548, 1.4692, 1.6934], device='cuda:0'), covar=tensor([0.2384, 0.2602, 0.2792, 0.2273], device='cuda:0'), in_proj_covar=tensor([0.1630, 0.1174, 0.1441, 0.1023], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 13:48:56,126 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1349357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:49:06,100 INFO [train.py:968] (0/2) Epoch 30, batch 28900, giga_loss[loss=0.2706, simple_loss=0.3508, pruned_loss=0.0952, over 28980.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3713, pruned_loss=0.1239, over 5670594.09 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3577, pruned_loss=0.111, over 5718537.01 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3722, pruned_loss=0.1246, over 5662563.76 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:49:07,713 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1349370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:49:32,421 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1349395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:49:35,632 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1349398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:49:43,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.465e+03 2.141e+03 2.710e+03 3.829e+03 1.433e+04, threshold=5.419e+03, percent-clipped=7.0
+2023-03-15 13:49:52,624 INFO [train.py:968] (0/2) Epoch 30, batch 28950, giga_loss[loss=0.3479, simple_loss=0.4005, pruned_loss=0.1476, over 27607.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3709, pruned_loss=0.123, over 5665104.69 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3583, pruned_loss=0.1115, over 5703076.45 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3713, pruned_loss=0.1235, over 5671196.77 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:50:01,604 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1349427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:50:39,844 INFO [train.py:968] (0/2) Epoch 30, batch 29000, giga_loss[loss=0.2699, simple_loss=0.3437, pruned_loss=0.09802, over 28777.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3719, pruned_loss=0.1235, over 5655316.11 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3584, pruned_loss=0.1115, over 5698003.54 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3726, pruned_loss=0.1242, over 5663385.02 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:50:49,035 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-15 13:51:11,432 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1349500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:51:14,124 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1349503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:51:16,582 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1349505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:51:18,131 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.113e+03 1.668e+03 2.301e+03 3.272e+03 8.317e+03, threshold=4.602e+03, percent-clipped=4.0
+2023-03-15 13:51:26,960 INFO [train.py:968] (0/2) Epoch 30, batch 29050, giga_loss[loss=0.2932, simple_loss=0.3582, pruned_loss=0.1142, over 28739.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3726, pruned_loss=0.124, over 5654892.66 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3584, pruned_loss=0.1116, over 5690373.98 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3734, pruned_loss=0.1248, over 5668073.50 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:51:41,435 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1349532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:52:15,501 INFO [train.py:968] (0/2) Epoch 30, batch 29100, giga_loss[loss=0.3664, simple_loss=0.4153, pruned_loss=0.1587, over 28269.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3739, pruned_loss=0.1252, over 5648194.74 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3584, pruned_loss=0.1116, over 5684089.99 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3748, pruned_loss=0.126, over 5664353.78 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:52:17,407 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-15 13:52:29,084 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6551, 1.9006, 1.3231, 1.4480], device='cuda:0'), covar=tensor([0.1082, 0.0616, 0.1079, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0459, 0.0530, 0.0469], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 13:52:50,969 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.259e+03 1.903e+03 2.360e+03 3.165e+03 6.876e+03, threshold=4.719e+03, percent-clipped=6.0
+2023-03-15 13:53:00,928 INFO [train.py:968] (0/2) Epoch 30, batch 29150, giga_loss[loss=0.374, simple_loss=0.4196, pruned_loss=0.1642, over 28889.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.373, pruned_loss=0.1244, over 5658369.80 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3579, pruned_loss=0.1111, over 5690767.47 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3746, pruned_loss=0.1258, over 5664618.71 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:53:43,987 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.43 vs. limit=5.0
+2023-03-15 13:53:47,318 INFO [train.py:968] (0/2) Epoch 30, batch 29200, giga_loss[loss=0.3781, simple_loss=0.4087, pruned_loss=0.1737, over 23718.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3733, pruned_loss=0.1238, over 5650499.37 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3575, pruned_loss=0.1108, over 5690865.40 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3754, pruned_loss=0.1257, over 5654373.64 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 13:54:27,646 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.114e+03 1.870e+03 2.437e+03 3.137e+03 7.282e+03, threshold=4.875e+03, percent-clipped=5.0
+2023-03-15 13:54:38,220 INFO [train.py:968] (0/2) Epoch 30, batch 29250, giga_loss[loss=0.3067, simple_loss=0.3726, pruned_loss=0.1204, over 28498.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3736, pruned_loss=0.1234, over 5646728.03 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3574, pruned_loss=0.1107, over 5694351.69 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3756, pruned_loss=0.1252, over 5645698.10 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:54:54,923 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2679, 1.1812, 3.6206, 3.2617], device='cuda:0'), covar=tensor([0.1678, 0.3041, 0.0509, 0.1608], device='cuda:0'), in_proj_covar=tensor([0.0826, 0.0685, 0.1030, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 13:55:07,320 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1349745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:55:26,918 INFO [train.py:968] (0/2) Epoch 30, batch 29300, giga_loss[loss=0.279, simple_loss=0.3556, pruned_loss=0.1012, over 28862.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3735, pruned_loss=0.123, over 5649033.05 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3575, pruned_loss=0.1107, over 5697125.24 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3753, pruned_loss=0.1247, over 5645174.73 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:55:52,352 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 13:56:05,768 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.097e+03 1.711e+03 2.131e+03 2.897e+03 5.665e+03, threshold=4.262e+03, percent-clipped=5.0
+2023-03-15 13:56:08,978 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1349812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:56:14,941 INFO [train.py:968] (0/2) Epoch 30, batch 29350, giga_loss[loss=0.3452, simple_loss=0.3787, pruned_loss=0.1559, over 23531.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3715, pruned_loss=0.1214, over 5657374.64 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3574, pruned_loss=0.1105, over 5699268.54 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3733, pruned_loss=0.1231, over 5651651.52 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:57:00,304 INFO [train.py:968] (0/2) Epoch 30, batch 29400, giga_loss[loss=0.3237, simple_loss=0.3872, pruned_loss=0.1301, over 28678.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3699, pruned_loss=0.1204, over 5659764.83 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3573, pruned_loss=0.1105, over 5699226.75 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3717, pruned_loss=0.1219, over 5654302.71 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:57:15,796 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1349880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:57:24,033 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1349888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:57:27,638 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1349891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:57:41,559 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.265e+03 1.772e+03 2.151e+03 2.778e+03 6.396e+03, threshold=4.302e+03, percent-clipped=7.0
+2023-03-15 13:57:54,957 INFO [train.py:968] (0/2) Epoch 30, batch 29450, giga_loss[loss=0.3689, simple_loss=0.3966, pruned_loss=0.1706, over 23430.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3711, pruned_loss=0.1211, over 5659610.63 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3571, pruned_loss=0.1104, over 5704565.36 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3729, pruned_loss=0.1227, over 5649476.68 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:57:57,235 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1349920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:58:05,458 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2273, 0.8474, 1.0091, 1.4501], device='cuda:0'), covar=tensor([0.0776, 0.0407, 0.0364, 0.0889], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 13:58:29,894 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-15 13:58:42,418 INFO [train.py:968] (0/2) Epoch 30, batch 29500, giga_loss[loss=0.3338, simple_loss=0.383, pruned_loss=0.1423, over 28292.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3713, pruned_loss=0.1221, over 5639057.74 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3575, pruned_loss=0.1107, over 5681054.48 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3728, pruned_loss=0.1234, over 5650901.59 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:59:03,346 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4257, 4.2765, 4.0728, 1.8940], device='cuda:0'), covar=tensor([0.0601, 0.0717, 0.0741, 0.2082], device='cuda:0'), in_proj_covar=tensor([0.1351, 0.1247, 0.1046, 0.0769], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0013], device='cuda:0')
+2023-03-15 13:59:13,327 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1350000.pt
+2023-03-15 13:59:21,431 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.043e+03 1.747e+03 2.205e+03 2.988e+03 5.863e+03, threshold=4.411e+03, percent-clipped=3.0
+2023-03-15 13:59:30,498 INFO [train.py:968] (0/2) Epoch 30, batch 29550, giga_loss[loss=0.2537, simple_loss=0.3429, pruned_loss=0.08224, over 28269.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3711, pruned_loss=0.1224, over 5650534.30 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3574, pruned_loss=0.1106, over 5683422.33 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3725, pruned_loss=0.1236, over 5657388.08 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:59:36,546 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1350023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:59:38,546 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1350026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:00:04,496 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1350055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:00:14,636 INFO [train.py:968] (0/2) Epoch 30, batch 29600, giga_loss[loss=0.2958, simple_loss=0.3645, pruned_loss=0.1135, over 29077.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3715, pruned_loss=0.1225, over 5657535.20 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3571, pruned_loss=0.1103, over 5683827.21 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3735, pruned_loss=0.1243, over 5661473.89 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 14:00:31,057 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3978, 1.8776, 1.3629, 0.8195], device='cuda:0'), covar=tensor([0.5882, 0.2925, 0.3553, 0.6510], device='cuda:0'), in_proj_covar=tensor([0.1891, 0.1776, 0.1687, 0.1533], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 14:00:46,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7077, 1.5920, 1.8907, 1.4819], device='cuda:0'), covar=tensor([0.1904, 0.2806, 0.1581, 0.1862], device='cuda:0'), in_proj_covar=tensor([0.0938, 0.0722, 0.0990, 0.0888], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 14:00:52,839 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.748e+03 2.246e+03 3.012e+03 7.041e+03, threshold=4.492e+03, percent-clipped=7.0
+2023-03-15 14:01:04,028 INFO [train.py:968] (0/2) Epoch 30, batch 29650, giga_loss[loss=0.3008, simple_loss=0.3698, pruned_loss=0.1159, over 28904.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3726, pruned_loss=0.1235, over 5648765.10 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3574, pruned_loss=0.1104, over 5687142.84 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3744, pruned_loss=0.1252, over 5647847.64 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 14:01:54,734 INFO [train.py:968] (0/2) Epoch 30, batch 29700, giga_loss[loss=0.3813, simple_loss=0.4283, pruned_loss=0.1671, over 28242.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3731, pruned_loss=0.1239, over 5651162.98 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3573, pruned_loss=0.1104, over 5689323.94 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3748, pruned_loss=0.1255, over 5647940.16 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:02:08,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8294, 2.0573, 2.1315, 1.6944], device='cuda:0'), covar=tensor([0.3156, 0.2828, 0.2867, 0.3181], device='cuda:0'), in_proj_covar=tensor([0.2107, 0.2078, 0.1979, 0.2126], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 14:02:11,465 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1350187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:02:34,182 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.224e+03 1.937e+03 2.272e+03 3.210e+03 1.023e+04, threshold=4.545e+03, percent-clipped=9.0
+2023-03-15 14:02:37,857 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2855, 3.0381, 1.3963, 1.5029], device='cuda:0'), covar=tensor([0.1103, 0.0475, 0.0987, 0.1414], device='cuda:0'), in_proj_covar=tensor([0.0436, 0.0584, 0.0419, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 14:02:40,872 INFO [train.py:968] (0/2) Epoch 30, batch 29750, libri_loss[loss=0.3289, simple_loss=0.3899, pruned_loss=0.134, over 29206.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3728, pruned_loss=0.1235, over 5647436.84 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3575, pruned_loss=0.1105, over 5691825.57 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3744, pruned_loss=0.125, over 5641443.18 frames. ], batch size: 97, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:03:26,330 INFO [train.py:968] (0/2) Epoch 30, batch 29800, giga_loss[loss=0.3027, simple_loss=0.3726, pruned_loss=0.1164, over 28800.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3717, pruned_loss=0.1218, over 5651905.41 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.357, pruned_loss=0.1103, over 5687510.33 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3739, pruned_loss=0.1237, over 5649959.86 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:04:02,267 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.205e+03 1.889e+03 2.227e+03 3.108e+03 9.902e+03, threshold=4.454e+03, percent-clipped=5.0
+2023-03-15 14:04:09,696 INFO [train.py:968] (0/2) Epoch 30, batch 29850, giga_loss[loss=0.2997, simple_loss=0.3742, pruned_loss=0.1126, over 29063.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3713, pruned_loss=0.1215, over 5655681.68 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3574, pruned_loss=0.1104, over 5693293.08 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3734, pruned_loss=0.1233, over 5647527.21 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:04:24,113 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1350330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:04:26,284 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1350333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:04:53,986 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1350362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:04:57,928 INFO [train.py:968] (0/2) Epoch 30, batch 29900, giga_loss[loss=0.2709, simple_loss=0.3423, pruned_loss=0.09974, over 28914.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3707, pruned_loss=0.121, over 5669060.87 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3576, pruned_loss=0.1104, over 5695591.93 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3724, pruned_loss=0.1226, over 5660064.61 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:05:37,626 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.143e+03 1.777e+03 2.161e+03 3.017e+03 5.688e+03, threshold=4.321e+03, percent-clipped=1.0
+2023-03-15 14:05:44,393 INFO [train.py:968] (0/2) Epoch 30, batch 29950, giga_loss[loss=0.2757, simple_loss=0.3418, pruned_loss=0.1048, over 28738.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3681, pruned_loss=0.1192, over 5675234.75 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3572, pruned_loss=0.1102, over 5699126.04 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3699, pruned_loss=0.1208, over 5664824.87 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:06:36,090 INFO [train.py:968] (0/2) Epoch 30, batch 30000, giga_loss[loss=0.3241, simple_loss=0.3796, pruned_loss=0.1343, over 27952.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.364, pruned_loss=0.1171, over 5659000.70 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.357, pruned_loss=0.1101, over 5695881.39 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3659, pruned_loss=0.1188, over 5652372.86 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:06:36,094 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 14:06:44,918 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.2021, simple_loss=0.3107, pruned_loss=0.04675, over 944034.00 frames. 
+2023-03-15 14:06:44,918 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 14:07:25,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.219e+03 2.020e+03 2.675e+03 3.666e+03 7.084e+03, threshold=5.349e+03, percent-clipped=17.0
+2023-03-15 14:07:27,113 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4251, 2.3861, 1.9176, 1.7547], device='cuda:0'), covar=tensor([0.0786, 0.0234, 0.0261, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0124, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 14:07:31,147 INFO [train.py:968] (0/2) Epoch 30, batch 30050, giga_loss[loss=0.3295, simple_loss=0.3822, pruned_loss=0.1384, over 28436.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.362, pruned_loss=0.1172, over 5651143.15 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3574, pruned_loss=0.1104, over 5687473.80 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3633, pruned_loss=0.1183, over 5652485.58 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:07:47,608 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1350537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:08:21,024 INFO [train.py:968] (0/2) Epoch 30, batch 30100, giga_loss[loss=0.3, simple_loss=0.3621, pruned_loss=0.1189, over 28625.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3619, pruned_loss=0.1178, over 5647516.65 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3574, pruned_loss=0.1104, over 5691798.96 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.363, pruned_loss=0.1189, over 5644077.72 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:08:30,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6074, 1.8380, 1.5313, 1.8944], device='cuda:0'), covar=tensor([0.2319, 0.2385, 0.2566, 0.2264], device='cuda:0'), in_proj_covar=tensor([0.1630, 0.1173, 0.1442, 0.1022], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 14:09:03,838 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.178e+03 1.958e+03 2.553e+03 3.646e+03 1.201e+04, threshold=5.105e+03, percent-clipped=9.0
+2023-03-15 14:09:10,449 INFO [train.py:968] (0/2) Epoch 30, batch 30150, libri_loss[loss=0.3403, simple_loss=0.4047, pruned_loss=0.1379, over 28466.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3603, pruned_loss=0.1163, over 5645459.46 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.357, pruned_loss=0.1101, over 5696650.61 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3616, pruned_loss=0.1176, over 5636994.66 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:09:12,049 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2141, 1.8428, 1.4600, 0.4854], device='cuda:0'), covar=tensor([0.5831, 0.3857, 0.4831, 0.7373], device='cuda:0'), in_proj_covar=tensor([0.1889, 0.1773, 0.1685, 0.1527], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 14:09:58,609 INFO [train.py:968] (0/2) Epoch 30, batch 30200, giga_loss[loss=0.2953, simple_loss=0.3682, pruned_loss=0.1112, over 28883.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3592, pruned_loss=0.1134, over 5652538.08 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3573, pruned_loss=0.1104, over 5700139.19 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3601, pruned_loss=0.1143, over 5641718.38 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:10:44,539 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3683, 1.9119, 1.3734, 0.6452], device='cuda:0'), covar=tensor([0.5636, 0.3201, 0.4570, 0.6650], device='cuda:0'), in_proj_covar=tensor([0.1882, 0.1763, 0.1680, 0.1521], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 14:10:50,343 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.096e+03 1.618e+03 1.920e+03 2.410e+03 7.371e+03, threshold=3.840e+03, percent-clipped=5.0
+2023-03-15 14:10:57,144 INFO [train.py:968] (0/2) Epoch 30, batch 30250, giga_loss[loss=0.2843, simple_loss=0.3658, pruned_loss=0.1014, over 28652.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3571, pruned_loss=0.1101, over 5651249.38 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.357, pruned_loss=0.1103, over 5702856.65 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3581, pruned_loss=0.1109, over 5639540.76 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:11:47,635 INFO [train.py:968] (0/2) Epoch 30, batch 30300, giga_loss[loss=0.2546, simple_loss=0.3376, pruned_loss=0.08582, over 28096.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3554, pruned_loss=0.1076, over 5659987.94 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3568, pruned_loss=0.1104, over 5702501.78 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3563, pruned_loss=0.1081, over 5649693.56 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:12:33,170 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.453e+03 1.830e+03 2.536e+03 6.378e+03, threshold=3.659e+03, percent-clipped=6.0
+2023-03-15 14:12:41,065 INFO [train.py:968] (0/2) Epoch 30, batch 30350, giga_loss[loss=0.2535, simple_loss=0.3328, pruned_loss=0.08708, over 27943.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3511, pruned_loss=0.1037, over 5651664.94 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3563, pruned_loss=0.1104, over 5696124.44 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3522, pruned_loss=0.104, over 5649168.83 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:12:57,664 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1350833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:13:31,479 INFO [train.py:968] (0/2) Epoch 30, batch 30400, giga_loss[loss=0.259, simple_loss=0.3516, pruned_loss=0.08317, over 28971.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3498, pruned_loss=0.1012, over 5661130.67 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.356, pruned_loss=0.1104, over 5699274.04 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3509, pruned_loss=0.1013, over 5655261.41 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:14:17,511 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.571e+02 1.620e+03 2.380e+03 3.640e+03 1.092e+04, threshold=4.759e+03, percent-clipped=23.0
+2023-03-15 14:14:18,440 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1350912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:14:24,258 INFO [train.py:968] (0/2) Epoch 30, batch 30450, libri_loss[loss=0.2537, simple_loss=0.3222, pruned_loss=0.0926, over 29581.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3497, pruned_loss=0.09993, over 5655073.28 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3558, pruned_loss=0.1105, over 5685161.32 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3506, pruned_loss=0.09953, over 5662055.26 frames. ], batch size: 75, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:15:12,217 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1350964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:15:15,265 INFO [train.py:968] (0/2) Epoch 30, batch 30500, giga_loss[loss=0.2424, simple_loss=0.3292, pruned_loss=0.07784, over 28865.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3495, pruned_loss=0.09973, over 5650883.55 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3555, pruned_loss=0.1105, over 5681040.09 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3503, pruned_loss=0.09919, over 5659234.73 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:15:47,869 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:15:56,809 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.037e+03 1.571e+03 2.016e+03 2.881e+03 1.318e+04, threshold=4.032e+03, percent-clipped=5.0
+2023-03-15 14:16:05,576 INFO [train.py:968] (0/2) Epoch 30, batch 30550, giga_loss[loss=0.2383, simple_loss=0.3238, pruned_loss=0.07643, over 28787.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3474, pruned_loss=0.09854, over 5663849.81 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3555, pruned_loss=0.1109, over 5688556.61 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3478, pruned_loss=0.09735, over 5662820.06 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:16:22,372 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 14:16:40,613 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:16:43,327 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:16:43,961 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4391, 1.3691, 3.8528, 3.3002], device='cuda:0'), covar=tensor([0.1624, 0.2855, 0.0448, 0.1136], device='cuda:0'), in_proj_covar=tensor([0.0823, 0.0682, 0.1024, 0.1002], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 14:16:46,752 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-15 14:16:53,639 INFO [train.py:968] (0/2) Epoch 30, batch 30600, giga_loss[loss=0.2673, simple_loss=0.3461, pruned_loss=0.09423, over 28881.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.344, pruned_loss=0.0962, over 5668507.18 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3542, pruned_loss=0.1102, over 5695076.83 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3452, pruned_loss=0.09548, over 5661230.45 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:17:04,200 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-15 14:17:06,096 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 14:17:16,410 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:17:38,092 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.041e+03 1.622e+03 2.024e+03 2.835e+03 5.125e+03, threshold=4.048e+03, percent-clipped=7.0
+2023-03-15 14:17:45,480 INFO [train.py:968] (0/2) Epoch 30, batch 30650, giga_loss[loss=0.2373, simple_loss=0.3303, pruned_loss=0.07215, over 28782.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3432, pruned_loss=0.09556, over 5662130.75 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3541, pruned_loss=0.1103, over 5696324.26 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.344, pruned_loss=0.09463, over 5654793.04 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:17:50,651 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5444, 1.8156, 1.7194, 1.5210], device='cuda:0'), covar=tensor([0.2848, 0.2257, 0.1873, 0.2367], device='cuda:0'), in_proj_covar=tensor([0.2068, 0.2040, 0.1943, 0.2085], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 14:18:26,046 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:18:34,795 INFO [train.py:968] (0/2) Epoch 30, batch 30700, giga_loss[loss=0.2639, simple_loss=0.3395, pruned_loss=0.09417, over 28899.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3433, pruned_loss=0.0956, over 5649836.05 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3542, pruned_loss=0.1104, over 5679510.41 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3437, pruned_loss=0.09446, over 5658784.48 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:19:00,932 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1542, 1.9318, 1.4823, 1.5407], device='cuda:0'), covar=tensor([0.2699, 0.1954, 0.2212, 0.2234], device='cuda:0'), in_proj_covar=tensor([0.0511, 0.0756, 0.0730, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 14:19:18,004 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:19:21,315 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.004e+03 1.465e+03 1.809e+03 2.625e+03 7.358e+03, threshold=3.617e+03, percent-clipped=9.0
+2023-03-15 14:19:31,242 INFO [train.py:968] (0/2) Epoch 30, batch 30750, giga_loss[loss=0.2679, simple_loss=0.3473, pruned_loss=0.09429, over 28896.00 frames. ], tot_loss[loss=0.264, simple_loss=0.341, pruned_loss=0.09354, over 5649739.96 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3539, pruned_loss=0.1103, over 5681988.44 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3415, pruned_loss=0.09266, over 5654270.55 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:20:26,983 INFO [train.py:968] (0/2) Epoch 30, batch 30800, giga_loss[loss=0.231, simple_loss=0.3189, pruned_loss=0.07158, over 28758.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3378, pruned_loss=0.09091, over 5659842.16 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3537, pruned_loss=0.1102, over 5684392.79 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3382, pruned_loss=0.09014, over 5661053.85 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 14:20:35,619 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.7089, 1.3053, 4.6932, 3.4215], device='cuda:0'), covar=tensor([0.1622, 0.2979, 0.0426, 0.1094], device='cuda:0'), in_proj_covar=tensor([0.0822, 0.0681, 0.1021, 0.0998], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 14:21:10,722 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.630e+02 1.564e+03 2.096e+03 2.900e+03 7.926e+03, threshold=4.192e+03, percent-clipped=9.0
+2023-03-15 14:21:16,887 INFO [train.py:968] (0/2) Epoch 30, batch 30850, giga_loss[loss=0.2276, simple_loss=0.3135, pruned_loss=0.07084, over 28719.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3347, pruned_loss=0.08985, over 5665052.78 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3535, pruned_loss=0.1102, over 5688204.29 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3349, pruned_loss=0.0888, over 5662116.57 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:21:39,254 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:21:53,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:21:57,045 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:22:11,758 INFO [train.py:968] (0/2) Epoch 30, batch 30900, giga_loss[loss=0.2188, simple_loss=0.29, pruned_loss=0.07377, over 23953.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3339, pruned_loss=0.0898, over 5650114.94 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3535, pruned_loss=0.1102, over 5679543.27 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3339, pruned_loss=0.08884, over 5656012.19 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:22:21,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:22:27,991 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:22:56,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3768, 1.6205, 1.3209, 1.5900], device='cuda:0'), covar=tensor([0.0784, 0.0327, 0.0357, 0.0899], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 14:23:01,861 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.093e+02 1.519e+03 2.042e+03 2.592e+03 7.424e+03, threshold=4.085e+03, percent-clipped=6.0
+2023-03-15 14:23:04,799 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:23:08,513 INFO [train.py:968] (0/2) Epoch 30, batch 30950, giga_loss[loss=0.2923, simple_loss=0.3651, pruned_loss=0.1097, over 28771.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3349, pruned_loss=0.09054, over 5645581.02 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3537, pruned_loss=0.1104, over 5681352.16 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3346, pruned_loss=0.08941, over 5648227.68 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:24:05,907 INFO [train.py:968] (0/2) Epoch 30, batch 31000, giga_loss[loss=0.2605, simple_loss=0.3482, pruned_loss=0.08635, over 28553.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3376, pruned_loss=0.09187, over 5636534.31 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3532, pruned_loss=0.1104, over 5678520.73 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3371, pruned_loss=0.09027, over 5639278.32 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:24:20,711 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:24:26,773 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:24:57,365 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.075e+03 1.599e+03 2.104e+03 2.931e+03 1.001e+04, threshold=4.208e+03, percent-clipped=5.0
+2023-03-15 14:24:59,284 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:25:03,084 INFO [train.py:968] (0/2) Epoch 30, batch 31050, giga_loss[loss=0.2528, simple_loss=0.3367, pruned_loss=0.08441, over 28773.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3371, pruned_loss=0.0906, over 5639084.48 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3525, pruned_loss=0.1099, over 5683412.89 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3369, pruned_loss=0.08923, over 5635835.82 frames. ], batch size: 243, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:25:06,733 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:25:08,738 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:25:23,316 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:25:48,274 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:26:05,101 INFO [train.py:968] (0/2) Epoch 30, batch 31100, giga_loss[loss=0.2855, simple_loss=0.3626, pruned_loss=0.1043, over 28668.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3377, pruned_loss=0.09164, over 5639127.80 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3519, pruned_loss=0.1097, over 5687516.37 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3376, pruned_loss=0.09021, over 5631416.71 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:27:04,090 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.780e+02 1.616e+03 2.159e+03 3.097e+03 8.754e+03, threshold=4.318e+03, percent-clipped=10.0
+2023-03-15 14:27:05,648 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.4044, 1.2448, 3.7924, 3.3869], device='cuda:0'), covar=tensor([0.1623, 0.3000, 0.0482, 0.1051], device='cuda:0'), in_proj_covar=tensor([0.0819, 0.0680, 0.1018, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 14:27:11,549 INFO [train.py:968] (0/2) Epoch 30, batch 31150, giga_loss[loss=0.2381, simple_loss=0.3263, pruned_loss=0.07501, over 28455.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3357, pruned_loss=0.09038, over 5647418.68 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3518, pruned_loss=0.1098, over 5693162.72 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3353, pruned_loss=0.08864, over 5635181.40 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:27:13,233 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:28:19,737 INFO [train.py:968] (0/2) Epoch 30, batch 31200, giga_loss[loss=0.2297, simple_loss=0.3255, pruned_loss=0.06697, over 28562.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.334, pruned_loss=0.08782, over 5644311.64 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3513, pruned_loss=0.1094, over 5696597.63 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3339, pruned_loss=0.08648, over 5631208.65 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:28:20,044 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2797, 0.8025, 0.8928, 1.4508], device='cuda:0'), covar=tensor([0.0719, 0.0362, 0.0371, 0.0841], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 14:28:22,638 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3668, 1.7223, 1.4655, 1.5508], device='cuda:0'), covar=tensor([0.0765, 0.0393, 0.0359, 0.0868], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 14:28:29,389 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:28:32,034 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:29:04,513 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4261, 3.4581, 1.4961, 1.5692], device='cuda:0'), covar=tensor([0.1042, 0.0335, 0.1023, 0.1421], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0579, 0.0418, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 14:29:07,183 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:29:12,582 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.110e+02 1.442e+03 1.830e+03 2.656e+03 4.670e+03, threshold=3.659e+03, percent-clipped=3.0
+2023-03-15 14:29:19,958 INFO [train.py:968] (0/2) Epoch 30, batch 31250, giga_loss[loss=0.2112, simple_loss=0.2992, pruned_loss=0.06162, over 28909.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3313, pruned_loss=0.08624, over 5649001.41 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3508, pruned_loss=0.1094, over 5697549.12 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3313, pruned_loss=0.08475, over 5636644.87 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:30:27,706 INFO [train.py:968] (0/2) Epoch 30, batch 31300, giga_loss[loss=0.2374, simple_loss=0.3226, pruned_loss=0.07604, over 27698.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3296, pruned_loss=0.08589, over 5661933.26 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3506, pruned_loss=0.1093, over 5698467.37 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3297, pruned_loss=0.0847, over 5651285.28 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:30:51,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:31:22,434 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.713e+02 1.484e+03 1.847e+03 2.384e+03 6.652e+03, threshold=3.694e+03, percent-clipped=2.0
+2023-03-15 14:31:29,191 INFO [train.py:968] (0/2) Epoch 30, batch 31350, giga_loss[loss=0.2511, simple_loss=0.3358, pruned_loss=0.08322, over 28509.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.329, pruned_loss=0.08575, over 5669328.90 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3503, pruned_loss=0.1093, over 5701229.86 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.329, pruned_loss=0.08453, over 5658056.60 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:32:03,488 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-15 14:32:21,943 INFO [train.py:968] (0/2) Epoch 30, batch 31400, giga_loss[loss=0.2583, simple_loss=0.343, pruned_loss=0.08681, over 28509.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3301, pruned_loss=0.08613, over 5664894.93 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3498, pruned_loss=0.1092, over 5699376.56 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3299, pruned_loss=0.08439, over 5657252.79 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:32:44,165 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:33:18,549 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.040e+03 1.568e+03 2.006e+03 2.639e+03 7.228e+03, threshold=4.012e+03, percent-clipped=6.0
+2023-03-15 14:33:23,372 INFO [train.py:968] (0/2) Epoch 30, batch 31450, giga_loss[loss=0.2344, simple_loss=0.3219, pruned_loss=0.07348, over 29045.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3337, pruned_loss=0.0884, over 5640267.94 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3505, pruned_loss=0.11, over 5678490.58 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3323, pruned_loss=0.08544, over 5651131.76 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:33:43,594 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:33:48,711 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:34:05,293 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:34:20,697 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:34:24,395 INFO [train.py:968] (0/2) Epoch 30, batch 31500, giga_loss[loss=0.2447, simple_loss=0.3325, pruned_loss=0.07839, over 28947.00 frames. ], tot_loss[loss=0.253, simple_loss=0.332, pruned_loss=0.08703, over 5657855.49 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3506, pruned_loss=0.1102, over 5683105.95 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3302, pruned_loss=0.08388, over 5661571.72 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:35:05,205 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:35:10,028 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1352000.pt
+2023-03-15 14:35:28,470 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.603e+02 1.503e+03 1.843e+03 2.335e+03 4.602e+03, threshold=3.686e+03, percent-clipped=2.0
+2023-03-15 14:35:35,670 INFO [train.py:968] (0/2) Epoch 30, batch 31550, giga_loss[loss=0.2651, simple_loss=0.3406, pruned_loss=0.09484, over 28370.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3321, pruned_loss=0.08726, over 5661670.42 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3502, pruned_loss=0.11, over 5681143.45 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3306, pruned_loss=0.08442, over 5666054.78 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:36:42,614 INFO [train.py:968] (0/2) Epoch 30, batch 31600, giga_loss[loss=0.2449, simple_loss=0.3433, pruned_loss=0.0732, over 28976.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3343, pruned_loss=0.08702, over 5661889.38 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3503, pruned_loss=0.1101, over 5681345.62 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3329, pruned_loss=0.08453, over 5665042.03 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 14:37:39,236 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2097, 1.6414, 1.5775, 1.3956], device='cuda:0'), covar=tensor([0.2616, 0.2460, 0.2480, 0.2503], device='cuda:0'), in_proj_covar=tensor([0.0509, 0.0756, 0.0730, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 14:37:43,596 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.408e+03 2.018e+03 3.070e+03 7.308e+03, threshold=4.037e+03, percent-clipped=11.0
+2023-03-15 14:37:47,097 INFO [train.py:968] (0/2) Epoch 30, batch 31650, giga_loss[loss=0.2806, simple_loss=0.3704, pruned_loss=0.09541, over 27705.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3381, pruned_loss=0.08703, over 5656065.37 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3502, pruned_loss=0.11, over 5682819.26 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3367, pruned_loss=0.08442, over 5656200.26 frames. ], batch size: 474, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:38:13,921 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1352137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:38:19,958 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1352140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:38:54,380 INFO [train.py:968] (0/2) Epoch 30, batch 31700, giga_loss[loss=0.2672, simple_loss=0.3592, pruned_loss=0.08762, over 28972.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.339, pruned_loss=0.08599, over 5656484.24 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3503, pruned_loss=0.1101, over 5683975.16 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3378, pruned_loss=0.08375, over 5655376.90 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:38:56,103 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1352169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:38:57,445 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-15 14:39:43,427 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4801, 1.7018, 1.4469, 1.3356], device='cuda:0'), covar=tensor([0.2827, 0.2870, 0.3213, 0.2669], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1171, 0.1444, 0.1023], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 14:39:49,086 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.175e+02 1.548e+03 1.970e+03 2.562e+03 7.445e+03, threshold=3.939e+03, percent-clipped=5.0
+2023-03-15 14:39:53,851 INFO [train.py:968] (0/2) Epoch 30, batch 31750, libri_loss[loss=0.2502, simple_loss=0.3117, pruned_loss=0.09431, over 29339.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3379, pruned_loss=0.08462, over 5666855.54 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3497, pruned_loss=0.1098, over 5685275.35 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3372, pruned_loss=0.08241, over 5664381.22 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:39:54,284 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0003, 1.3412, 1.1285, 0.2553], device='cuda:0'), covar=tensor([0.4814, 0.3975, 0.5466, 0.7913], device='cuda:0'), in_proj_covar=tensor([0.1869, 0.1748, 0.1672, 0.1517], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 14:40:04,274 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-15 14:40:46,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1352262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:40:52,844 INFO [train.py:968] (0/2) Epoch 30, batch 31800, giga_loss[loss=0.2382, simple_loss=0.3203, pruned_loss=0.07804, over 28919.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3397, pruned_loss=0.08705, over 5667248.65 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.35, pruned_loss=0.1102, over 5678922.99 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3383, pruned_loss=0.08389, over 5670991.97 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:41:50,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.019e+03 1.547e+03 2.121e+03 3.021e+03 8.924e+03, threshold=4.243e+03, percent-clipped=9.0
+2023-03-15 14:41:56,400 INFO [train.py:968] (0/2) Epoch 30, batch 31850, giga_loss[loss=0.261, simple_loss=0.341, pruned_loss=0.09055, over 28300.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3369, pruned_loss=0.08666, over 5675847.74 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3494, pruned_loss=0.1098, over 5685041.41 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.336, pruned_loss=0.08367, over 5673002.45 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:42:08,953 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1352324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:42:26,669 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-15 14:43:14,639 INFO [train.py:968] (0/2) Epoch 30, batch 31900, giga_loss[loss=0.2501, simple_loss=0.3311, pruned_loss=0.08453, over 27724.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3373, pruned_loss=0.08773, over 5670920.31 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3488, pruned_loss=0.1094, over 5681578.88 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3368, pruned_loss=0.085, over 5671491.38 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:44:12,300 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4496, 3.5293, 1.5518, 1.5625], device='cuda:0'), covar=tensor([0.1016, 0.0314, 0.1009, 0.1369], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0578, 0.0418, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 14:44:12,334 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1352405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:44:17,578 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1352408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:44:26,557 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.501e+02 1.424e+03 1.763e+03 2.442e+03 4.762e+03, threshold=3.527e+03, percent-clipped=1.0
+2023-03-15 14:44:29,817 INFO [train.py:968] (0/2) Epoch 30, batch 31950, giga_loss[loss=0.2031, simple_loss=0.2928, pruned_loss=0.05669, over 28237.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3335, pruned_loss=0.08593, over 5661905.93 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3484, pruned_loss=0.1091, over 5674514.74 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3331, pruned_loss=0.08346, over 5669528.17 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:44:58,577 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1352437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:45:01,501 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1352440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:45:33,854 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1352467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:45:34,181 INFO [train.py:968] (0/2) Epoch 30, batch 32000, giga_loss[loss=0.2634, simple_loss=0.3419, pruned_loss=0.09239, over 28517.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3315, pruned_loss=0.08483, over 5670912.05 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3481, pruned_loss=0.109, over 5677593.21 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3309, pruned_loss=0.08214, over 5673952.54 frames. ], batch size: 370, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 14:45:36,757 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3625, 1.4821, 1.3738, 1.5473], device='cuda:0'), covar=tensor([0.0731, 0.0397, 0.0358, 0.0866], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 14:45:37,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1352470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:45:55,717 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5559, 1.7970, 1.2484, 1.3729], device='cuda:0'), covar=tensor([0.1117, 0.0594, 0.1072, 0.1238], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0453, 0.0527, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 14:46:12,385 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1352499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:46:31,489 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.728e+02 1.468e+03 1.930e+03 2.519e+03 1.047e+04, threshold=3.861e+03, percent-clipped=14.0
+2023-03-15 14:46:35,120 INFO [train.py:968] (0/2) Epoch 30, batch 32050, giga_loss[loss=0.247, simple_loss=0.3129, pruned_loss=0.09056, over 24194.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3283, pruned_loss=0.0837, over 5658674.32 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3474, pruned_loss=0.1087, over 5665439.52 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3276, pruned_loss=0.08065, over 5671035.90 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:46:55,991 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 14:47:37,385 INFO [train.py:968] (0/2) Epoch 30, batch 32100, giga_loss[loss=0.2767, simple_loss=0.3627, pruned_loss=0.09533, over 28506.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3294, pruned_loss=0.08407, over 5666952.77 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.347, pruned_loss=0.1086, over 5661083.46 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3288, pruned_loss=0.08123, over 5680285.63 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:48:10,629 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5511, 3.4003, 1.6012, 1.7008], device='cuda:0'), covar=tensor([0.0970, 0.0331, 0.0921, 0.1276], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0578, 0.0418, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 14:48:23,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-15 14:48:36,040 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.128e+03 1.591e+03 2.054e+03 2.963e+03 6.148e+03, threshold=4.108e+03, percent-clipped=11.0
+2023-03-15 14:48:38,517 INFO [train.py:968] (0/2) Epoch 30, batch 32150, giga_loss[loss=0.2453, simple_loss=0.3228, pruned_loss=0.08393, over 28973.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3332, pruned_loss=0.08635, over 5676876.30 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3468, pruned_loss=0.1084, over 5667076.60 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3326, pruned_loss=0.0837, over 5682234.99 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:49:27,343 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-15 14:49:41,008 INFO [train.py:968] (0/2) Epoch 30, batch 32200, giga_loss[loss=0.2577, simple_loss=0.3324, pruned_loss=0.09146, over 28913.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.332, pruned_loss=0.08726, over 5672152.87 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3466, pruned_loss=0.1085, over 5661705.20 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3311, pruned_loss=0.08423, over 5681347.06 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:49:45,429 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1352672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:49:54,393 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-15 14:50:41,282 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.601e+03 2.094e+03 3.156e+03 8.479e+03, threshold=4.188e+03, percent-clipped=9.0
+2023-03-15 14:50:43,854 INFO [train.py:968] (0/2) Epoch 30, batch 32250, giga_loss[loss=0.3269, simple_loss=0.3828, pruned_loss=0.1356, over 26879.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3327, pruned_loss=0.08853, over 5673509.77 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3464, pruned_loss=0.1084, over 5668619.62 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3317, pruned_loss=0.08544, over 5675107.98 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:51:24,372 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1352751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:51:29,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5644, 1.8575, 1.3430, 1.4081], device='cuda:0'), covar=tensor([0.1095, 0.0533, 0.1014, 0.1058], device='cuda:0'), in_proj_covar=tensor([0.0416, 0.0449, 0.0523, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 14:51:44,069 INFO [train.py:968] (0/2) Epoch 30, batch 32300, giga_loss[loss=0.2612, simple_loss=0.3442, pruned_loss=0.08911, over 28767.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3327, pruned_loss=0.08855, over 5683375.16 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3464, pruned_loss=0.1085, over 5674652.65 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3314, pruned_loss=0.08524, over 5679533.20 frames. ], batch size: 243, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:52:02,661 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-15 14:52:20,782 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-15 14:52:53,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1352815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:52:53,660 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.020e+03 1.579e+03 2.041e+03 3.059e+03 9.503e+03, threshold=4.083e+03, percent-clipped=9.0
+2023-03-15 14:52:56,690 INFO [train.py:968] (0/2) Epoch 30, batch 32350, giga_loss[loss=0.2471, simple_loss=0.3387, pruned_loss=0.07775, over 28976.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3347, pruned_loss=0.08833, over 5681453.22 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3465, pruned_loss=0.1086, over 5677783.07 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3333, pruned_loss=0.08527, over 5675715.71 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:53:26,882 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-15 14:53:30,073 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1352838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:54:19,525 INFO [train.py:968] (0/2) Epoch 30, batch 32400, giga_loss[loss=0.2637, simple_loss=0.3423, pruned_loss=0.09259, over 28489.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3351, pruned_loss=0.0881, over 5670530.88 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3464, pruned_loss=0.1085, over 5679065.77 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3339, pruned_loss=0.08552, over 5664902.69 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:55:13,643 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6766, 2.0435, 1.9626, 1.6430], device='cuda:0'), covar=tensor([0.1980, 0.1799, 0.1940, 0.2023], device='cuda:0'), in_proj_covar=tensor([0.0506, 0.0753, 0.0728, 0.0700], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 14:55:23,045 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.737e+02 1.577e+03 1.984e+03 2.749e+03 8.335e+03, threshold=3.968e+03, percent-clipped=9.0
+2023-03-15 14:55:24,486 INFO [train.py:968] (0/2) Epoch 30, batch 32450, libri_loss[loss=0.2694, simple_loss=0.3391, pruned_loss=0.0998, over 29075.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3311, pruned_loss=0.08664, over 5671277.83 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3459, pruned_loss=0.1082, over 5676673.60 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3303, pruned_loss=0.08411, over 5668231.27 frames. ], batch size: 101, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:56:14,885 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1352958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:56:17,084 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1352961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:56:29,622 INFO [train.py:968] (0/2) Epoch 30, batch 32500, libri_loss[loss=0.2408, simple_loss=0.3148, pruned_loss=0.08341, over 29569.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3257, pruned_loss=0.08474, over 5670345.79 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.345, pruned_loss=0.1077, over 5672158.68 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3252, pruned_loss=0.08236, over 5672602.98 frames. ], batch size: 79, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:56:43,628 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6642, 2.4324, 1.8683, 0.8933], device='cuda:0'), covar=tensor([0.7259, 0.3861, 0.4529, 0.7279], device='cuda:0'), in_proj_covar=tensor([0.1882, 0.1759, 0.1681, 0.1528], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 14:56:56,513 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1352990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:57:31,599 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.228e+02 1.543e+03 1.995e+03 2.959e+03 8.754e+03, threshold=3.990e+03, percent-clipped=11.0
+2023-03-15 14:57:32,958 INFO [train.py:968] (0/2) Epoch 30, batch 32550, libri_loss[loss=0.2832, simple_loss=0.3466, pruned_loss=0.1098, over 29671.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3233, pruned_loss=0.08391, over 5668068.50 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.344, pruned_loss=0.1072, over 5677618.82 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3229, pruned_loss=0.08141, over 5664320.66 frames. ], batch size: 88, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:57:48,299 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.01 vs. limit=5.0
+2023-03-15 14:58:07,447 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1353047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:58:33,773 INFO [train.py:968] (0/2) Epoch 30, batch 32600, giga_loss[loss=0.2598, simple_loss=0.3379, pruned_loss=0.09085, over 28465.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3248, pruned_loss=0.08489, over 5673471.64 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3439, pruned_loss=0.1072, over 5679682.12 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3244, pruned_loss=0.08271, over 5668718.80 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:59:32,269 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.178e+03 1.682e+03 2.282e+03 3.303e+03 8.010e+03, threshold=4.564e+03, percent-clipped=18.0
+2023-03-15 14:59:35,903 INFO [train.py:968] (0/2) Epoch 30, batch 32650, giga_loss[loss=0.2318, simple_loss=0.3133, pruned_loss=0.07512, over 28915.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3245, pruned_loss=0.08431, over 5683137.38 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3432, pruned_loss=0.1068, over 5685605.00 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3244, pruned_loss=0.0824, over 5673875.02 frames. ], batch size: 112, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:59:45,952 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1353126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:59:49,815 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1353128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:00:38,553 INFO [train.py:968] (0/2) Epoch 30, batch 32700, giga_loss[loss=0.2242, simple_loss=0.3126, pruned_loss=0.0679, over 28910.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.323, pruned_loss=0.08283, over 5673079.47 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3432, pruned_loss=0.1069, over 5689719.08 frames. ], giga_tot_loss[loss=0.2418, simple_loss=0.3224, pruned_loss=0.0806, over 5661745.25 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:00:41,315 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-15 15:01:10,133 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1353190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:01:14,149 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1353193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:01:42,278 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1353213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:01:45,500 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.017e+03 1.527e+03 1.968e+03 2.891e+03 1.011e+04, threshold=3.936e+03, percent-clipped=8.0
+2023-03-15 15:01:47,879 INFO [train.py:968] (0/2) Epoch 30, batch 32750, giga_loss[loss=0.2529, simple_loss=0.3308, pruned_loss=0.08748, over 28887.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3218, pruned_loss=0.08197, over 5670117.50 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3429, pruned_loss=0.1067, over 5689183.48 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3214, pruned_loss=0.08025, over 5661666.40 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:01:55,129 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1353222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:02:57,427 INFO [train.py:968] (0/2) Epoch 30, batch 32800, giga_loss[loss=0.2578, simple_loss=0.3365, pruned_loss=0.08957, over 27654.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3208, pruned_loss=0.08137, over 5675805.73 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3429, pruned_loss=0.1068, over 5694790.85 frames. ], giga_tot_loss[loss=0.2393, simple_loss=0.32, pruned_loss=0.07932, over 5663525.20 frames. ], batch size: 474, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:03:00,816 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1353269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:03:04,408 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1353272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:03:44,299 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1353301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:04:02,414 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.205e+02 1.486e+03 2.097e+03 2.805e+03 1.028e+04, threshold=4.194e+03, percent-clipped=15.0
+2023-03-15 15:04:03,056 INFO [train.py:968] (0/2) Epoch 30, batch 32850, giga_loss[loss=0.2517, simple_loss=0.3288, pruned_loss=0.08729, over 28630.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3221, pruned_loss=0.08172, over 5692861.02 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.342, pruned_loss=0.1064, over 5702474.95 frames. ], giga_tot_loss[loss=0.2404, simple_loss=0.3216, pruned_loss=0.07964, over 5675428.99 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:04:53,575 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1353355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:04:54,306 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1353356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:04:57,820 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1353359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:05:10,258 INFO [train.py:968] (0/2) Epoch 30, batch 32900, giga_loss[loss=0.2492, simple_loss=0.3311, pruned_loss=0.08368, over 28549.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3234, pruned_loss=0.08282, over 5689334.36 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3417, pruned_loss=0.1061, over 5703671.04 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3229, pruned_loss=0.08092, over 5674006.43 frames. ], batch size: 370, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:05:34,589 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1353388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:06:13,322 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.595e+02 1.402e+03 1.662e+03 2.281e+03 4.919e+03, threshold=3.323e+03, percent-clipped=3.0
+2023-03-15 15:06:13,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6262, 1.8834, 1.6452, 1.6641], device='cuda:0'), covar=tensor([0.0757, 0.0289, 0.0326, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0118], device='cuda:0')
+2023-03-15 15:06:14,038 INFO [train.py:968] (0/2) Epoch 30, batch 32950, giga_loss[loss=0.2185, simple_loss=0.3028, pruned_loss=0.06711, over 28928.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3236, pruned_loss=0.08325, over 5693533.87 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3418, pruned_loss=0.1062, over 5706672.20 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3229, pruned_loss=0.08141, over 5678575.30 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:07:19,606 INFO [train.py:968] (0/2) Epoch 30, batch 33000, giga_loss[loss=0.2356, simple_loss=0.3283, pruned_loss=0.07142, over 28886.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3241, pruned_loss=0.08268, over 5663747.92 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3422, pruned_loss=0.1064, over 5697727.46 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.323, pruned_loss=0.08073, over 5660457.41 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:07:19,613 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 15:07:23,916 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4908, 1.8522, 1.5032, 1.3482], device='cuda:0'), covar=tensor([0.2923, 0.2843, 0.3149, 0.2779], device='cuda:0'), in_proj_covar=tensor([0.1637, 0.1171, 0.1447, 0.1025], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0010], device='cuda:0')
+2023-03-15 15:07:28,212 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.1937, simple_loss=0.2953, pruned_loss=0.04605, over 944034.00 frames. 
+2023-03-15 15:07:28,212 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 15:07:59,582 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1350, 2.5510, 1.2205, 1.3377], device='cuda:0'), covar=tensor([0.1099, 0.0430, 0.1018, 0.1468], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0575, 0.0418, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 15:08:09,572 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1353503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:08:14,096 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5361, 1.8235, 1.4728, 1.5441], device='cuda:0'), covar=tensor([0.2902, 0.2846, 0.3339, 0.2529], device='cuda:0'), in_proj_covar=tensor([0.1636, 0.1170, 0.1446, 0.1025], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0010], device='cuda:0')
+2023-03-15 15:08:28,314 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.081e+03 1.414e+03 1.866e+03 2.542e+03 7.166e+03, threshold=3.731e+03, percent-clipped=7.0
+2023-03-15 15:08:30,573 INFO [train.py:968] (0/2) Epoch 30, batch 33050, giga_loss[loss=0.2701, simple_loss=0.3554, pruned_loss=0.09237, over 28803.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3272, pruned_loss=0.08303, over 5661149.47 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3423, pruned_loss=0.1065, over 5699663.96 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3261, pruned_loss=0.08125, over 5656619.77 frames. ], batch size: 243, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:08:38,039 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([5.0361, 4.8751, 4.6615, 2.3399], device='cuda:0'), covar=tensor([0.0491, 0.0593, 0.0787, 0.1879], device='cuda:0'), in_proj_covar=tensor([0.1305, 0.1205, 0.1010, 0.0742], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 15:09:22,110 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7122, 2.2424, 1.5970, 1.8949], device='cuda:0'), covar=tensor([0.0703, 0.0249, 0.0317, 0.0844], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0118], device='cuda:0')
+2023-03-15 15:09:30,713 INFO [train.py:968] (0/2) Epoch 30, batch 33100, giga_loss[loss=0.222, simple_loss=0.3155, pruned_loss=0.06426, over 28738.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3287, pruned_loss=0.08351, over 5671404.56 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.342, pruned_loss=0.1062, over 5704703.55 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3277, pruned_loss=0.08168, over 5662236.03 frames. ], batch size: 243, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:10:36,549 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.168e+02 1.518e+03 1.860e+03 2.586e+03 5.686e+03, threshold=3.719e+03, percent-clipped=10.0
+2023-03-15 15:10:37,300 INFO [train.py:968] (0/2) Epoch 30, batch 33150, giga_loss[loss=0.2342, simple_loss=0.3168, pruned_loss=0.07581, over 29026.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3291, pruned_loss=0.08372, over 5671101.23 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3417, pruned_loss=0.106, over 5704478.91 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3282, pruned_loss=0.08184, over 5663339.90 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:10:44,576 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6706, 2.3339, 1.6916, 0.8044], device='cuda:0'), covar=tensor([0.7294, 0.3554, 0.4612, 0.7255], device='cuda:0'), in_proj_covar=tensor([0.1876, 0.1754, 0.1674, 0.1523], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 15:11:13,310 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1353646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:11:15,576 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1353649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:11:35,045 INFO [train.py:968] (0/2) Epoch 30, batch 33200, giga_loss[loss=0.2244, simple_loss=0.3131, pruned_loss=0.0678, over 28897.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3283, pruned_loss=0.0837, over 5672989.13 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3417, pruned_loss=0.1061, over 5705285.34 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3272, pruned_loss=0.08143, over 5665229.72 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:11:45,557 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1353678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:12:35,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.184e+02 1.572e+03 1.915e+03 2.355e+03 6.136e+03, threshold=3.830e+03, percent-clipped=6.0
+2023-03-15 15:12:36,843 INFO [train.py:968] (0/2) Epoch 30, batch 33250, giga_loss[loss=0.2428, simple_loss=0.327, pruned_loss=0.07935, over 28879.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3252, pruned_loss=0.08161, over 5679711.38 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3417, pruned_loss=0.1061, over 5708242.74 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3241, pruned_loss=0.07941, over 5670557.24 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:12:51,371 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1353730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:13:37,426 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1353766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:13:39,052 INFO [train.py:968] (0/2) Epoch 30, batch 33300, giga_loss[loss=0.1921, simple_loss=0.2722, pruned_loss=0.056, over 28536.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3234, pruned_loss=0.08124, over 5686160.68 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3415, pruned_loss=0.1061, over 5713262.37 frames. ], giga_tot_loss[loss=0.2402, simple_loss=0.3223, pruned_loss=0.07899, over 5673903.19 frames. ], batch size: 78, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:14:32,667 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.667e+02 1.444e+03 1.986e+03 2.614e+03 6.092e+03, threshold=3.972e+03, percent-clipped=11.0
+2023-03-15 15:14:33,603 INFO [train.py:968] (0/2) Epoch 30, batch 33350, giga_loss[loss=0.2417, simple_loss=0.3301, pruned_loss=0.07665, over 28861.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3251, pruned_loss=0.08272, over 5686536.62 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.342, pruned_loss=0.1063, over 5720867.25 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.3231, pruned_loss=0.07968, over 5668690.86 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:14:36,919 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-15 15:15:42,880 INFO [train.py:968] (0/2) Epoch 30, batch 33400, libri_loss[loss=0.2573, simple_loss=0.33, pruned_loss=0.09226, over 29514.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3275, pruned_loss=0.0832, over 5685170.28 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.342, pruned_loss=0.1062, over 5722692.97 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3257, pruned_loss=0.08064, over 5668975.96 frames. ], batch size: 82, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:15:48,498 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 15:15:50,298 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1353873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:15:52,902 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1353876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:16:28,773 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1353905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:16:43,011 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.084e+03 1.603e+03 2.258e+03 3.490e+03 7.361e+03, threshold=4.515e+03, percent-clipped=11.0
+2023-03-15 15:16:45,450 INFO [train.py:968] (0/2) Epoch 30, batch 33450, giga_loss[loss=0.2461, simple_loss=0.3325, pruned_loss=0.0798, over 28471.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3288, pruned_loss=0.08482, over 5687522.40 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3418, pruned_loss=0.1063, over 5724790.82 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3271, pruned_loss=0.08214, over 5671648.13 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:16:52,444 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 15:17:54,926 INFO [train.py:968] (0/2) Epoch 30, batch 33500, giga_loss[loss=0.2889, simple_loss=0.3656, pruned_loss=0.1061, over 27653.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3318, pruned_loss=0.08706, over 5674516.42 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.342, pruned_loss=0.1063, over 5728936.39 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3301, pruned_loss=0.08441, over 5657033.59 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:18:18,248 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1353987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:18:33,813 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1354000.pt
+2023-03-15 15:18:52,341 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.387e+02 1.601e+03 1.995e+03 2.484e+03 6.759e+03, threshold=3.990e+03, percent-clipped=4.0
+2023-03-15 15:18:52,354 INFO [train.py:968] (0/2) Epoch 30, batch 33550, giga_loss[loss=0.233, simple_loss=0.3241, pruned_loss=0.07094, over 28835.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3343, pruned_loss=0.0876, over 5673511.73 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3417, pruned_loss=0.1062, over 5732519.72 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.333, pruned_loss=0.08524, over 5655422.44 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:19:48,795 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-15 15:20:00,327 INFO [train.py:968] (0/2) Epoch 30, batch 33600, giga_loss[loss=0.2605, simple_loss=0.3445, pruned_loss=0.08827, over 28471.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3362, pruned_loss=0.08839, over 5671721.28 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3415, pruned_loss=0.1061, over 5734345.85 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3353, pruned_loss=0.08633, over 5655170.64 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:20:56,950 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-15 15:21:02,461 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3591, 3.3056, 1.4137, 1.5420], device='cuda:0'), covar=tensor([0.1010, 0.0370, 0.1008, 0.1342], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0574, 0.0417, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 15:21:08,474 INFO [train.py:968] (0/2) Epoch 30, batch 33650, giga_loss[loss=0.2406, simple_loss=0.3146, pruned_loss=0.08328, over 27566.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3345, pruned_loss=0.08777, over 5677034.23 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3412, pruned_loss=0.1061, over 5729318.47 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3339, pruned_loss=0.08562, over 5666907.50 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:21:08,909 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5174, 1.6552, 1.7107, 1.3251], device='cuda:0'), covar=tensor([0.1787, 0.2640, 0.1516, 0.1892], device='cuda:0'), in_proj_covar=tensor([0.0941, 0.0715, 0.0993, 0.0893], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 15:21:09,155 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.333e+02 1.716e+03 2.370e+03 3.294e+03 7.994e+03, threshold=4.739e+03, percent-clipped=17.0
+2023-03-15 15:21:43,299 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1354141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:22:18,664 INFO [train.py:968] (0/2) Epoch 30, batch 33700, giga_loss[loss=0.2655, simple_loss=0.3384, pruned_loss=0.0963, over 26895.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3327, pruned_loss=0.08707, over 5675916.55 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3415, pruned_loss=0.1063, over 5720853.43 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3318, pruned_loss=0.08492, over 5674625.86 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:22:47,036 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3506, 1.5742, 1.1151, 1.1642], device='cuda:0'), covar=tensor([0.1149, 0.0458, 0.1155, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0450, 0.0526, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 15:22:51,059 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-15 15:23:20,721 INFO [train.py:968] (0/2) Epoch 30, batch 33750, giga_loss[loss=0.2539, simple_loss=0.329, pruned_loss=0.0894, over 27908.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3317, pruned_loss=0.08649, over 5665717.71 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3415, pruned_loss=0.1063, over 5711746.44 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3307, pruned_loss=0.08423, over 5672219.62 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:23:21,303 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.384e+02 1.507e+03 2.024e+03 2.586e+03 5.746e+03, threshold=4.049e+03, percent-clipped=2.0
+2023-03-15 15:24:21,195 INFO [train.py:968] (0/2) Epoch 30, batch 33800, giga_loss[loss=0.2456, simple_loss=0.3174, pruned_loss=0.08689, over 27624.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3301, pruned_loss=0.08664, over 5664553.64 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3411, pruned_loss=0.1062, over 5707577.09 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3292, pruned_loss=0.08402, over 5671460.70 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:24:45,049 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1354284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:24:50,230 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1354287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:24:51,548 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8139, 2.4942, 1.5333, 0.9887], device='cuda:0'), covar=tensor([1.0035, 0.4779, 0.4991, 0.8390], device='cuda:0'), in_proj_covar=tensor([0.1873, 0.1753, 0.1672, 0.1522], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 15:24:56,140 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5411, 1.8802, 1.4975, 1.6909], device='cuda:0'), covar=tensor([0.2870, 0.2764, 0.3325, 0.2430], device='cuda:0'), in_proj_covar=tensor([0.1628, 0.1165, 0.1439, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 15:25:26,680 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1354316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:25:29,652 INFO [train.py:968] (0/2) Epoch 30, batch 33850, giga_loss[loss=0.217, simple_loss=0.2822, pruned_loss=0.07589, over 24468.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3273, pruned_loss=0.08557, over 5660916.79 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3413, pruned_loss=0.1064, over 5697050.00 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3263, pruned_loss=0.08312, over 5675638.56 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:25:30,631 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.085e+02 1.455e+03 1.819e+03 2.792e+03 5.895e+03, threshold=3.637e+03, percent-clipped=6.0
+2023-03-15 15:25:31,820 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2024, 1.6693, 1.6739, 1.3898], device='cuda:0'), covar=tensor([0.2430, 0.2055, 0.2233, 0.2188], device='cuda:0'), in_proj_covar=tensor([0.0501, 0.0748, 0.0723, 0.0694], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 15:26:23,542 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1354362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:26:29,698 INFO [train.py:968] (0/2) Epoch 30, batch 33900, giga_loss[loss=0.2342, simple_loss=0.3215, pruned_loss=0.07341, over 28439.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3273, pruned_loss=0.0844, over 5668728.73 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3407, pruned_loss=0.1059, over 5701454.92 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3268, pruned_loss=0.08237, over 5675651.02 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:26:32,637 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-15 15:26:38,032 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4806, 2.1142, 1.6050, 0.6147], device='cuda:0'), covar=tensor([0.5648, 0.3577, 0.4498, 0.7322], device='cuda:0'), in_proj_covar=tensor([0.1874, 0.1756, 0.1674, 0.1524], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 15:26:44,765 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5143, 1.6233, 1.2675, 1.2998], device='cuda:0'), covar=tensor([0.0921, 0.0404, 0.0801, 0.1040], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0451, 0.0527, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 15:27:31,449 INFO [train.py:968] (0/2) Epoch 30, batch 33950, giga_loss[loss=0.2604, simple_loss=0.3481, pruned_loss=0.08629, over 28960.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.326, pruned_loss=0.08285, over 5669024.50 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3405, pruned_loss=0.1059, over 5706765.59 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3253, pruned_loss=0.08064, over 5669044.69 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:27:33,125 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.880e+02 1.590e+03 2.006e+03 2.716e+03 6.961e+03, threshold=4.011e+03, percent-clipped=9.0
+2023-03-15 15:28:13,666 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-15 15:28:32,876 INFO [train.py:968] (0/2) Epoch 30, batch 34000, giga_loss[loss=0.2258, simple_loss=0.3143, pruned_loss=0.06864, over 28733.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3285, pruned_loss=0.08207, over 5676241.37 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3405, pruned_loss=0.1058, over 5708864.90 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3278, pruned_loss=0.08013, over 5674083.72 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:29:02,460 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3836, 1.6949, 1.4005, 0.9934], device='cuda:0'), covar=tensor([0.2855, 0.2912, 0.3397, 0.2722], device='cuda:0'), in_proj_covar=tensor([0.1631, 0.1166, 0.1441, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 15:29:18,918 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1354505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:29:23,269 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1354508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:29:32,638 INFO [train.py:968] (0/2) Epoch 30, batch 34050, giga_loss[loss=0.2468, simple_loss=0.3334, pruned_loss=0.08013, over 28934.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3298, pruned_loss=0.08243, over 5679817.81 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3408, pruned_loss=0.1061, over 5710104.48 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3288, pruned_loss=0.08017, over 5676157.36 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:29:35,188 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.039e+03 1.497e+03 2.062e+03 2.953e+03 1.188e+04, threshold=4.124e+03, percent-clipped=9.0
+2023-03-15 15:29:48,615 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-15 15:29:56,058 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1354537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:30:42,072 INFO [train.py:968] (0/2) Epoch 30, batch 34100, giga_loss[loss=0.2225, simple_loss=0.3081, pruned_loss=0.06847, over 29028.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3302, pruned_loss=0.0824, over 5677117.45 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3408, pruned_loss=0.106, over 5711591.23 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3292, pruned_loss=0.08021, over 5672373.62 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:31:13,375 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-15 15:31:52,653 INFO [train.py:968] (0/2) Epoch 30, batch 34150, giga_loss[loss=0.2529, simple_loss=0.3354, pruned_loss=0.08517, over 28951.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3302, pruned_loss=0.0828, over 5663693.27 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3406, pruned_loss=0.1061, over 5706566.85 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3293, pruned_loss=0.0805, over 5664435.72 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:31:54,005 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.763e+02 1.543e+03 1.888e+03 2.512e+03 6.187e+03, threshold=3.776e+03, percent-clipped=3.0
+2023-03-15 15:32:22,242 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2173, 1.0359, 1.0432, 1.3780], device='cuda:0'), covar=tensor([0.0821, 0.0370, 0.0351, 0.1077], device='cuda:0'), in_proj_covar=tensor([0.0197, 0.0123, 0.0122, 0.0234], device='cuda:0'), out_proj_covar=tensor([0.0108, 0.0076, 0.0068, 0.0118], device='cuda:0')
+2023-03-15 15:32:59,736 INFO [train.py:968] (0/2) Epoch 30, batch 34200, giga_loss[loss=0.2185, simple_loss=0.3077, pruned_loss=0.06463, over 28054.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3307, pruned_loss=0.08302, over 5653440.88 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3403, pruned_loss=0.1059, over 5695687.45 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3301, pruned_loss=0.08085, over 5662589.68 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:34:16,542 INFO [train.py:968] (0/2) Epoch 30, batch 34250, giga_loss[loss=0.2645, simple_loss=0.3472, pruned_loss=0.09087, over 28852.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3309, pruned_loss=0.08231, over 5661480.44 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.34, pruned_loss=0.1057, over 5700013.53 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3305, pruned_loss=0.08038, over 5664287.45 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:34:19,683 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.041e+02 1.381e+03 1.873e+03 2.589e+03 5.212e+03, threshold=3.745e+03, percent-clipped=8.0
+2023-03-15 15:34:38,147 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2562, 1.3980, 1.3944, 1.2019], device='cuda:0'), covar=tensor([0.2778, 0.2808, 0.1724, 0.2504], device='cuda:0'), in_proj_covar=tensor([0.2053, 0.2011, 0.1910, 0.2065], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 15:34:40,600 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6197, 1.7646, 1.8072, 1.5641], device='cuda:0'), covar=tensor([0.3004, 0.2645, 0.2148, 0.2693], device='cuda:0'), in_proj_covar=tensor([0.2053, 0.2011, 0.1909, 0.2065], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 15:35:07,533 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4576, 1.9835, 1.6071, 1.6183], device='cuda:0'), covar=tensor([0.0808, 0.0286, 0.0338, 0.0946], device='cuda:0'), in_proj_covar=tensor([0.0197, 0.0123, 0.0122, 0.0234], device='cuda:0'), out_proj_covar=tensor([0.0108, 0.0076, 0.0068, 0.0118], device='cuda:0')
+2023-03-15 15:35:21,565 INFO [train.py:968] (0/2) Epoch 30, batch 34300, giga_loss[loss=0.2449, simple_loss=0.3296, pruned_loss=0.08009, over 28456.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3338, pruned_loss=0.08418, over 5668521.81 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.34, pruned_loss=0.1058, over 5705282.70 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3332, pruned_loss=0.08184, over 5664937.07 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:36:20,040 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3414, 1.5159, 1.1570, 1.0751], device='cuda:0'), covar=tensor([0.1072, 0.0586, 0.1100, 0.1233], device='cuda:0'), in_proj_covar=tensor([0.0417, 0.0449, 0.0524, 0.0461], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 15:36:27,950 INFO [train.py:968] (0/2) Epoch 30, batch 34350, giga_loss[loss=0.2612, simple_loss=0.3509, pruned_loss=0.0857, over 28265.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3355, pruned_loss=0.08467, over 5675650.42 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3397, pruned_loss=0.1056, over 5710261.89 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3352, pruned_loss=0.08244, over 5667568.83 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:36:31,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.648e+02 1.735e+03 2.407e+03 3.811e+03 1.136e+04, threshold=4.815e+03, percent-clipped=25.0
+2023-03-15 15:37:06,394 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2777, 1.8267, 1.3663, 0.5454], device='cuda:0'), covar=tensor([0.5295, 0.3573, 0.5257, 0.7485], device='cuda:0'), in_proj_covar=tensor([0.1870, 0.1755, 0.1676, 0.1525], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 15:37:27,357 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1354860.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 15:37:36,892 INFO [train.py:968] (0/2) Epoch 30, batch 34400, giga_loss[loss=0.2488, simple_loss=0.3315, pruned_loss=0.08301, over 28396.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3341, pruned_loss=0.08492, over 5669019.28 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3393, pruned_loss=0.1053, over 5705363.31 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3341, pruned_loss=0.08277, over 5665765.51 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:38:16,676 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1354896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:38:52,187 INFO [train.py:968] (0/2) Epoch 30, batch 34450, giga_loss[loss=0.2421, simple_loss=0.3276, pruned_loss=0.07829, over 28652.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3319, pruned_loss=0.08368, over 5684442.72 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3391, pruned_loss=0.1052, over 5707936.45 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3319, pruned_loss=0.0818, over 5679186.04 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:38:57,534 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.058e+03 1.480e+03 1.833e+03 2.485e+03 7.589e+03, threshold=3.666e+03, percent-clipped=2.0
+2023-03-15 15:39:18,595 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1354932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:39:22,391 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3332, 2.6799, 2.3743, 2.5158], device='cuda:0'), covar=tensor([0.1976, 0.1935, 0.2185, 0.1827], device='cuda:0'), in_proj_covar=tensor([0.1628, 0.1163, 0.1438, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 15:40:05,107 INFO [train.py:968] (0/2) Epoch 30, batch 34500, giga_loss[loss=0.2342, simple_loss=0.3272, pruned_loss=0.07059, over 28071.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3306, pruned_loss=0.08231, over 5671711.62 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3391, pruned_loss=0.1053, over 5696815.85 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3306, pruned_loss=0.08039, over 5677003.38 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:40:37,663 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3882, 1.9849, 1.4154, 0.6542], device='cuda:0'), covar=tensor([0.6627, 0.3424, 0.4792, 0.7447], device='cuda:0'), in_proj_covar=tensor([0.1865, 0.1749, 0.1671, 0.1521], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 15:41:06,498 INFO [train.py:968] (0/2) Epoch 30, batch 34550, libri_loss[loss=0.2416, simple_loss=0.3205, pruned_loss=0.08138, over 29662.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3302, pruned_loss=0.0828, over 5664384.04 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3389, pruned_loss=0.1052, over 5698126.04 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3299, pruned_loss=0.08015, over 5665677.97 frames. ], batch size: 91, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:41:10,507 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.944e+02 1.452e+03 1.777e+03 2.263e+03 7.756e+03, threshold=3.553e+03, percent-clipped=7.0
+2023-03-15 15:41:34,063 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 15:42:04,034 INFO [train.py:968] (0/2) Epoch 30, batch 34600, libri_loss[loss=0.2353, simple_loss=0.3014, pruned_loss=0.08455, over 29620.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.333, pruned_loss=0.08536, over 5676492.81 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.339, pruned_loss=0.1055, over 5707882.85 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3323, pruned_loss=0.0819, over 5666880.26 frames. ], batch size: 74, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:42:50,653 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-15 15:43:04,938 INFO [train.py:968] (0/2) Epoch 30, batch 34650, giga_loss[loss=0.2425, simple_loss=0.3261, pruned_loss=0.07949, over 28943.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3348, pruned_loss=0.08602, over 5671233.82 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3391, pruned_loss=0.1057, over 5697000.16 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3342, pruned_loss=0.08278, over 5672509.33 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:43:10,056 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.136e+03 1.751e+03 2.185e+03 3.066e+03 1.572e+04, threshold=4.369e+03, percent-clipped=19.0
+2023-03-15 15:44:04,371 INFO [train.py:968] (0/2) Epoch 30, batch 34700, giga_loss[loss=0.2085, simple_loss=0.2983, pruned_loss=0.05931, over 29064.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3322, pruned_loss=0.08543, over 5667874.47 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3391, pruned_loss=0.1058, over 5698366.33 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3315, pruned_loss=0.08224, over 5666867.71 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:44:23,569 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 15:45:02,921 INFO [train.py:968] (0/2) Epoch 30, batch 34750, giga_loss[loss=0.2214, simple_loss=0.3052, pruned_loss=0.0688, over 28848.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3312, pruned_loss=0.08574, over 5666618.31 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3391, pruned_loss=0.1058, over 5702343.13 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3305, pruned_loss=0.08282, over 5661566.60 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:45:08,368 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.003e+03 1.605e+03 2.122e+03 3.150e+03 1.027e+04, threshold=4.245e+03, percent-clipped=14.0
+2023-03-15 15:45:25,164 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1355235.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 15:46:04,234 INFO [train.py:968] (0/2) Epoch 30, batch 34800, giga_loss[loss=0.3311, simple_loss=0.3833, pruned_loss=0.1395, over 26910.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3311, pruned_loss=0.08595, over 5663848.72 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.339, pruned_loss=0.1059, over 5702918.12 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3306, pruned_loss=0.08336, over 5659004.11 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:46:09,096 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1355271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:46:43,503 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1355307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:46:51,850 INFO [train.py:968] (0/2) Epoch 30, batch 34850, giga_loss[loss=0.2873, simple_loss=0.3702, pruned_loss=0.1022, over 28919.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3395, pruned_loss=0.09014, over 5665292.06 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3392, pruned_loss=0.106, over 5694215.90 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3388, pruned_loss=0.08766, over 5667716.75 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:46:55,382 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.072e+03 1.594e+03 1.909e+03 2.692e+03 6.679e+03, threshold=3.817e+03, percent-clipped=6.0
+2023-03-15 15:47:41,814 INFO [train.py:968] (0/2) Epoch 30, batch 34900, giga_loss[loss=0.2819, simple_loss=0.3439, pruned_loss=0.11, over 23961.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3466, pruned_loss=0.09414, over 5665959.57 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3392, pruned_loss=0.1059, over 5698349.17 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3462, pruned_loss=0.09178, over 5663480.30 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:47:42,783 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1355369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:47:51,426 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1355378.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 15:47:54,636 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1355381.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 15:48:20,294 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1355410.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 15:48:22,755 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1355414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:48:24,694 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1355417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:48:25,042 INFO [train.py:968] (0/2) Epoch 30, batch 34950, giga_loss[loss=0.2405, simple_loss=0.3294, pruned_loss=0.07583, over 28853.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3475, pruned_loss=0.09549, over 5676995.69 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3395, pruned_loss=0.1061, over 5703547.25 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3471, pruned_loss=0.09315, over 5669531.47 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:48:28,350 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.2364, 4.0770, 3.8340, 2.0567], device='cuda:0'), covar=tensor([0.0593, 0.0742, 0.0744, 0.2028], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.1203, 0.1007, 0.0744], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 15:48:29,516 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.670e+02 1.426e+03 1.912e+03 2.937e+03 8.140e+03, threshold=3.824e+03, percent-clipped=9.0
+2023-03-15 15:48:49,846 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1355446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:48:54,100 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1355450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:48:57,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1355453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:49:08,546 INFO [train.py:968] (0/2) Epoch 30, batch 35000, giga_loss[loss=0.2454, simple_loss=0.3192, pruned_loss=0.08578, over 28597.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3424, pruned_loss=0.094, over 5690862.62 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3395, pruned_loss=0.1061, over 5710629.33 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3423, pruned_loss=0.0918, over 5677835.21 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:49:21,856 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1355482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:49:51,835 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4157, 1.4220, 1.2973, 1.6396], device='cuda:0'), covar=tensor([0.0764, 0.0395, 0.0365, 0.0857], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0234], device='cuda:0'), out_proj_covar=tensor([0.0108, 0.0076, 0.0068, 0.0118], device='cuda:0')
+2023-03-15 15:49:54,887 INFO [train.py:968] (0/2) Epoch 30, batch 35050, giga_loss[loss=0.2383, simple_loss=0.3132, pruned_loss=0.08171, over 27705.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3357, pruned_loss=0.09109, over 5692374.25 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3395, pruned_loss=0.106, over 5711963.70 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3356, pruned_loss=0.08921, over 5680658.46 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:50:00,515 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.153e+02 1.281e+03 1.603e+03 2.181e+03 8.805e+03, threshold=3.207e+03, percent-clipped=10.0
+2023-03-15 15:50:07,251 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4952, 2.1810, 1.7527, 0.6930], device='cuda:0'), covar=tensor([0.7416, 0.3974, 0.4843, 0.8216], device='cuda:0'), in_proj_covar=tensor([0.1869, 0.1759, 0.1675, 0.1522], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 15:50:38,999 INFO [train.py:968] (0/2) Epoch 30, batch 35100, giga_loss[loss=0.2026, simple_loss=0.2815, pruned_loss=0.06184, over 28957.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3275, pruned_loss=0.08717, over 5695256.06 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3395, pruned_loss=0.1059, over 5714922.31 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3273, pruned_loss=0.08557, over 5683045.76 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:51:19,418 INFO [train.py:968] (0/2) Epoch 30, batch 35150, giga_loss[loss=0.236, simple_loss=0.3026, pruned_loss=0.08474, over 27571.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3222, pruned_loss=0.08526, over 5694814.63 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3397, pruned_loss=0.1057, over 5714842.91 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3214, pruned_loss=0.08351, over 5684449.17 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:51:22,243 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.982e+02 1.136e+03 1.332e+03 1.852e+03 5.727e+03, threshold=2.664e+03, percent-clipped=8.0
+2023-03-15 15:51:31,639 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.8483, 4.6768, 4.4189, 2.1401], device='cuda:0'), covar=tensor([0.0517, 0.0710, 0.0715, 0.1999], device='cuda:0'), in_proj_covar=tensor([0.1308, 0.1205, 0.1009, 0.0745], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 15:51:37,189 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-15 15:51:46,071 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4538, 2.8753, 1.5991, 1.4959], device='cuda:0'), covar=tensor([0.0946, 0.0378, 0.0886, 0.1318], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0576, 0.0418, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 15:51:47,281 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1355650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:52:02,524 INFO [train.py:968] (0/2) Epoch 30, batch 35200, giga_loss[loss=0.1982, simple_loss=0.2748, pruned_loss=0.0608, over 28521.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3169, pruned_loss=0.08334, over 5684902.99 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3404, pruned_loss=0.1062, over 5705123.96 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3151, pruned_loss=0.08089, over 5683930.63 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:52:11,206 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-15 15:52:31,276 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2526, 1.2136, 1.0601, 1.4371], device='cuda:0'), covar=tensor([0.0761, 0.0372, 0.0379, 0.0876], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0118], device='cuda:0')
+2023-03-15 15:52:45,578 INFO [train.py:968] (0/2) Epoch 30, batch 35250, giga_loss[loss=0.2212, simple_loss=0.2986, pruned_loss=0.07193, over 29053.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3137, pruned_loss=0.0819, over 5690158.40 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3406, pruned_loss=0.1061, over 5711388.71 frames. ], giga_tot_loss[loss=0.2349, simple_loss=0.3112, pruned_loss=0.07929, over 5682950.31 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:52:48,195 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.127e+02 1.171e+03 1.412e+03 1.997e+03 7.888e+03, threshold=2.823e+03, percent-clipped=12.0
+2023-03-15 15:53:10,752 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1355744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:53:16,251 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1355750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:53:26,482 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.44 vs. limit=5.0
+2023-03-15 15:53:31,990 INFO [train.py:968] (0/2) Epoch 30, batch 35300, giga_loss[loss=0.2195, simple_loss=0.2935, pruned_loss=0.07281, over 28632.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.31, pruned_loss=0.07989, over 5698217.16 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3403, pruned_loss=0.1058, over 5714815.17 frames. ], giga_tot_loss[loss=0.2315, simple_loss=0.3078, pruned_loss=0.07759, over 5689122.60 frames. ], batch size: 78, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:54:13,373 INFO [train.py:968] (0/2) Epoch 30, batch 35350, libri_loss[loss=0.2652, simple_loss=0.334, pruned_loss=0.09821, over 29575.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3074, pruned_loss=0.07879, over 5692960.38 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3406, pruned_loss=0.1059, over 5701627.88 frames. ], giga_tot_loss[loss=0.2283, simple_loss=0.3044, pruned_loss=0.07611, over 5695865.34 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:54:15,951 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-15 15:54:17,084 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.904e+02 1.161e+03 1.481e+03 2.224e+03 4.471e+03, threshold=2.962e+03, percent-clipped=9.0
+2023-03-15 15:54:23,871 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1355831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:54:57,973 INFO [train.py:968] (0/2) Epoch 30, batch 35400, giga_loss[loss=0.2353, simple_loss=0.3196, pruned_loss=0.07548, over 29044.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3046, pruned_loss=0.0771, over 5703444.16 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3406, pruned_loss=0.1056, over 5706207.12 frames. ], giga_tot_loss[loss=0.225, simple_loss=0.3012, pruned_loss=0.0744, over 5701525.10 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:55:05,097 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8283, 1.9069, 1.7253, 1.7588], device='cuda:0'), covar=tensor([0.2681, 0.2563, 0.2655, 0.2603], device='cuda:0'), in_proj_covar=tensor([0.1638, 0.1171, 0.1447, 0.1023], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0010], device='cuda:0')
+2023-03-15 15:55:15,209 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1355887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:55:17,347 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1355890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:55:40,461 INFO [train.py:968] (0/2) Epoch 30, batch 35450, giga_loss[loss=0.2154, simple_loss=0.2899, pruned_loss=0.07044, over 28862.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3014, pruned_loss=0.07561, over 5699308.31 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3404, pruned_loss=0.1053, over 5710303.50 frames. ], giga_tot_loss[loss=0.2219, simple_loss=0.2978, pruned_loss=0.07294, over 5693958.31 frames. ], batch size: 112, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:55:41,487 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1355919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:55:43,064 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.778e+02 1.211e+03 1.475e+03 1.904e+03 4.971e+03, threshold=2.950e+03, percent-clipped=7.0
+2023-03-15 15:56:21,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1355962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:56:25,796 INFO [train.py:968] (0/2) Epoch 30, batch 35500, giga_loss[loss=0.2531, simple_loss=0.3254, pruned_loss=0.09038, over 27982.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2994, pruned_loss=0.0749, over 5699399.05 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3408, pruned_loss=0.1055, over 5713096.80 frames. ], giga_tot_loss[loss=0.22, simple_loss=0.2956, pruned_loss=0.07218, over 5692500.42 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:56:54,359 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.56 vs. limit=5.0
+2023-03-15 15:56:55,860 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1356000.pt
+2023-03-15 15:57:11,980 INFO [train.py:968] (0/2) Epoch 30, batch 35550, libri_loss[loss=0.3562, simple_loss=0.406, pruned_loss=0.1532, over 19963.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.2972, pruned_loss=0.07421, over 5679767.72 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3415, pruned_loss=0.1059, over 5696888.25 frames. ], giga_tot_loss[loss=0.2176, simple_loss=0.2928, pruned_loss=0.07114, over 5689713.75 frames. ], batch size: 187, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:57:15,304 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.608e+02 1.091e+03 1.305e+03 1.758e+03 5.989e+03, threshold=2.610e+03, percent-clipped=8.0
+2023-03-15 15:57:17,493 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1356025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:57:58,133 INFO [train.py:968] (0/2) Epoch 30, batch 35600, giga_loss[loss=0.1748, simple_loss=0.2509, pruned_loss=0.04937, over 28297.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2934, pruned_loss=0.07234, over 5692539.19 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3413, pruned_loss=0.1057, over 5701451.01 frames. ], giga_tot_loss[loss=0.2139, simple_loss=0.2891, pruned_loss=0.06933, over 5696029.42 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:58:39,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6395, 1.6998, 1.8252, 1.4458], device='cuda:0'), covar=tensor([0.1909, 0.2615, 0.1573, 0.1810], device='cuda:0'), in_proj_covar=tensor([0.0949, 0.0718, 0.1001, 0.0899], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 15:58:45,860 INFO [train.py:968] (0/2) Epoch 30, batch 35650, giga_loss[loss=0.2369, simple_loss=0.3143, pruned_loss=0.07979, over 28798.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2947, pruned_loss=0.07352, over 5688270.34 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3415, pruned_loss=0.1057, over 5706045.92 frames. ], giga_tot_loss[loss=0.2156, simple_loss=0.2902, pruned_loss=0.07047, over 5686890.02 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:58:51,213 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 6.600e+02 1.136e+03 1.484e+03 1.817e+03 4.434e+03, threshold=2.968e+03, percent-clipped=8.0
+2023-03-15 15:58:52,776 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1356125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:59:34,686 INFO [train.py:968] (0/2) Epoch 30, batch 35700, giga_loss[loss=0.2821, simple_loss=0.347, pruned_loss=0.1086, over 27601.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3057, pruned_loss=0.07882, over 5684864.64 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3418, pruned_loss=0.1059, over 5702348.23 frames. ], giga_tot_loss[loss=0.2268, simple_loss=0.3015, pruned_loss=0.07602, over 5687040.04 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:59:35,847 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1356168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:59:39,040 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1356171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:00:06,020 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 16:00:08,663 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1356200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:00:13,359 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1356206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:00:24,046 INFO [train.py:968] (0/2) Epoch 30, batch 35750, giga_loss[loss=0.284, simple_loss=0.3622, pruned_loss=0.1029, over 29000.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3196, pruned_loss=0.086, over 5678681.96 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3425, pruned_loss=0.1063, over 5695493.41 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3152, pruned_loss=0.08304, over 5686172.34 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:00:30,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.119e+03 1.517e+03 1.876e+03 2.397e+03 6.694e+03, threshold=3.753e+03, percent-clipped=13.0
+2023-03-15 16:01:12,521 INFO [train.py:968] (0/2) Epoch 30, batch 35800, giga_loss[loss=0.3132, simple_loss=0.368, pruned_loss=0.1292, over 23737.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3309, pruned_loss=0.09163, over 5681031.85 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3424, pruned_loss=0.1062, over 5697715.06 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3273, pruned_loss=0.08925, over 5684701.84 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:01:12,774 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1356268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:01:14,834 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1356271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:01:33,924 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-15 16:01:39,804 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1356300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:01:53,891 INFO [train.py:968] (0/2) Epoch 30, batch 35850, libri_loss[loss=0.2868, simple_loss=0.356, pruned_loss=0.1088, over 26071.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3358, pruned_loss=0.09308, over 5672699.69 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3433, pruned_loss=0.1069, over 5678042.11 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3318, pruned_loss=0.09015, over 5693190.14 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:01:59,718 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.060e+03 1.407e+03 1.772e+03 2.428e+03 5.476e+03, threshold=3.544e+03, percent-clipped=6.0
+2023-03-15 16:02:09,704 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1356337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:02:19,047 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1356349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:02:21,312 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1356352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:02:36,435 INFO [train.py:968] (0/2) Epoch 30, batch 35900, giga_loss[loss=0.2513, simple_loss=0.3303, pruned_loss=0.08617, over 28603.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3379, pruned_loss=0.09316, over 5672303.73 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3436, pruned_loss=0.1068, over 5678168.04 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3342, pruned_loss=0.0903, over 5688473.24 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:02:49,605 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1356381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:02:58,478 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-15 16:03:21,463 INFO [train.py:968] (0/2) Epoch 30, batch 35950, giga_loss[loss=0.2905, simple_loss=0.3614, pruned_loss=0.1098, over 28759.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3406, pruned_loss=0.09439, over 5667242.67 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3437, pruned_loss=0.1067, over 5673011.58 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3373, pruned_loss=0.09168, over 5685197.35 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:03:25,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.584e+02 1.323e+03 1.736e+03 2.387e+03 5.904e+03, threshold=3.472e+03, percent-clipped=9.0
+2023-03-15 16:03:49,936 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1356452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:04:07,340 INFO [train.py:968] (0/2) Epoch 30, batch 36000, giga_loss[loss=0.2876, simple_loss=0.3381, pruned_loss=0.1186, over 23382.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3434, pruned_loss=0.09689, over 5668519.06 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3441, pruned_loss=0.107, over 5677869.52 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3404, pruned_loss=0.09431, over 5678671.80 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:04:07,344 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 16:04:16,004 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.1995, simple_loss=0.3068, pruned_loss=0.0461, over 944034.00 frames. 
+2023-03-15 16:04:16,004 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 16:04:24,667 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1356480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:04:27,538 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1356483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:04:53,283 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1356512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:04:57,834 INFO [train.py:968] (0/2) Epoch 30, batch 36050, giga_loss[loss=0.2682, simple_loss=0.3538, pruned_loss=0.09133, over 28869.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3472, pruned_loss=0.09897, over 5675279.19 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3444, pruned_loss=0.107, over 5682176.21 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3445, pruned_loss=0.09674, over 5679369.72 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:05:03,678 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.665e+02 1.413e+03 1.690e+03 2.338e+03 8.102e+03, threshold=3.381e+03, percent-clipped=7.0
+2023-03-15 16:05:39,068 INFO [train.py:968] (0/2) Epoch 30, batch 36100, giga_loss[loss=0.2764, simple_loss=0.3589, pruned_loss=0.09699, over 28902.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.35, pruned_loss=0.1003, over 5674789.47 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3446, pruned_loss=0.107, over 5675764.14 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3479, pruned_loss=0.0983, over 5683935.80 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:06:18,555 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-15 16:06:21,181 INFO [train.py:968] (0/2) Epoch 30, batch 36150, giga_loss[loss=0.2548, simple_loss=0.341, pruned_loss=0.08428, over 28631.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3533, pruned_loss=0.1015, over 5683575.37 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3453, pruned_loss=0.1074, over 5677713.36 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.351, pruned_loss=0.09945, over 5688909.55 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:06:30,374 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.196e+02 1.377e+03 1.852e+03 2.661e+03 1.255e+04, threshold=3.704e+03, percent-clipped=15.0
+2023-03-15 16:06:31,729 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-15 16:07:02,968 INFO [train.py:968] (0/2) Epoch 30, batch 36200, giga_loss[loss=0.2627, simple_loss=0.3438, pruned_loss=0.09081, over 29117.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3534, pruned_loss=0.1011, over 5684435.09 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3451, pruned_loss=0.1071, over 5684405.00 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.352, pruned_loss=0.09952, over 5682689.71 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 1.0
+2023-03-15 16:07:27,616 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9188, 1.2934, 1.1098, 0.3236], device='cuda:0'), covar=tensor([0.5591, 0.4058, 0.5581, 0.7437], device='cuda:0'), in_proj_covar=tensor([0.1870, 0.1759, 0.1677, 0.1520], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 16:07:30,185 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3600, 1.2616, 3.7994, 3.2985], device='cuda:0'), covar=tensor([0.1509, 0.2591, 0.0452, 0.0876], device='cuda:0'), in_proj_covar=tensor([0.0813, 0.0677, 0.1012, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 16:07:40,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3252, 1.6148, 1.2315, 1.1156], device='cuda:0'), covar=tensor([0.1248, 0.0559, 0.1102, 0.1255], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0451, 0.0528, 0.0466], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 16:07:44,372 INFO [train.py:968] (0/2) Epoch 30, batch 36250, libri_loss[loss=0.2069, simple_loss=0.2817, pruned_loss=0.06608, over 29632.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3544, pruned_loss=0.1009, over 5679662.46 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3454, pruned_loss=0.1073, over 5677515.84 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3534, pruned_loss=0.09933, over 5685164.92 frames. ], batch size: 69, lr: 1.05e-03, grad_scale: 1.0
+2023-03-15 16:07:52,552 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.032e+03 1.388e+03 1.681e+03 2.340e+03 5.823e+03, threshold=3.362e+03, percent-clipped=7.0
+2023-03-15 16:08:25,583 INFO [train.py:968] (0/2) Epoch 30, batch 36300, giga_loss[loss=0.2738, simple_loss=0.3586, pruned_loss=0.09449, over 29025.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3542, pruned_loss=0.09963, over 5692941.83 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3456, pruned_loss=0.1073, over 5681076.15 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3534, pruned_loss=0.09824, over 5694348.42 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 1.0
+2023-03-15 16:08:27,574 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7915, 1.1578, 2.8833, 2.6327], device='cuda:0'), covar=tensor([0.1836, 0.2811, 0.0602, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0814, 0.0678, 0.1013, 0.0988], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 16:08:41,423 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1356786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:09:08,087 INFO [train.py:968] (0/2) Epoch 30, batch 36350, giga_loss[loss=0.2482, simple_loss=0.3393, pruned_loss=0.07857, over 28563.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.353, pruned_loss=0.09767, over 5701171.72 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3461, pruned_loss=0.1077, over 5684869.43 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3519, pruned_loss=0.0961, over 5699114.48 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 1.0
+2023-03-15 16:09:14,392 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.716e+02 1.294e+03 1.499e+03 1.937e+03 6.635e+03, threshold=2.997e+03, percent-clipped=7.0
+2023-03-15 16:09:14,664 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1356827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:09:45,126 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3553, 3.4355, 1.5907, 1.5519], device='cuda:0'), covar=tensor([0.1087, 0.0311, 0.0923, 0.1441], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0575, 0.0416, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 16:09:49,465 INFO [train.py:968] (0/2) Epoch 30, batch 36400, libri_loss[loss=0.2691, simple_loss=0.3299, pruned_loss=0.1041, over 29343.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3514, pruned_loss=0.09674, over 5707778.10 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3461, pruned_loss=0.1075, over 5688554.30 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3508, pruned_loss=0.09523, over 5703515.33 frames. ], batch size: 67, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:10:27,453 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 16:10:36,607 INFO [train.py:968] (0/2) Epoch 30, batch 36450, giga_loss[loss=0.3242, simple_loss=0.3836, pruned_loss=0.1324, over 28796.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3545, pruned_loss=0.101, over 5689341.31 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3468, pruned_loss=0.108, over 5671254.96 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3535, pruned_loss=0.09917, over 5702081.78 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:10:44,674 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.657e+02 1.397e+03 1.746e+03 2.254e+03 4.649e+03, threshold=3.491e+03, percent-clipped=8.0
+2023-03-15 16:10:54,390 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 16:11:20,353 INFO [train.py:968] (0/2) Epoch 30, batch 36500, libri_loss[loss=0.3268, simple_loss=0.3906, pruned_loss=0.1315, over 25351.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3571, pruned_loss=0.105, over 5678343.27 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3472, pruned_loss=0.1083, over 5663987.30 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3562, pruned_loss=0.1032, over 5695495.52 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:11:22,737 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1356970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:11:24,610 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1356973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:11:52,598 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1357002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:12:08,284 INFO [train.py:968] (0/2) Epoch 30, batch 36550, libri_loss[loss=0.406, simple_loss=0.4384, pruned_loss=0.1868, over 29667.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3567, pruned_loss=0.106, over 5685197.72 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3477, pruned_loss=0.1087, over 5665334.36 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3556, pruned_loss=0.1041, over 5697367.37 frames. ], batch size: 88, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:12:15,149 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.110e+03 1.499e+03 1.969e+03 2.677e+03 7.730e+03, threshold=3.938e+03, percent-clipped=12.0
+2023-03-15 16:12:51,192 INFO [train.py:968] (0/2) Epoch 30, batch 36600, giga_loss[loss=0.2593, simple_loss=0.3393, pruned_loss=0.08967, over 28906.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.354, pruned_loss=0.1049, over 5692020.39 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3479, pruned_loss=0.1087, over 5668698.62 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3531, pruned_loss=0.1033, over 5699176.23 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:13:07,956 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-15 16:13:37,667 INFO [train.py:968] (0/2) Epoch 30, batch 36650, giga_loss[loss=0.2745, simple_loss=0.3474, pruned_loss=0.1008, over 28847.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3523, pruned_loss=0.104, over 5692108.32 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.348, pruned_loss=0.1088, over 5671100.68 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3515, pruned_loss=0.1027, over 5695887.45 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:13:43,548 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.432e+02 1.371e+03 1.641e+03 1.998e+03 6.628e+03, threshold=3.283e+03, percent-clipped=5.0
+2023-03-15 16:14:10,088 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5306, 1.8571, 1.4919, 1.7315], device='cuda:0'), covar=tensor([0.2792, 0.2891, 0.3170, 0.2399], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1168, 0.1441, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 16:14:12,509 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.6383, 3.4818, 3.3015, 1.7452], device='cuda:0'), covar=tensor([0.0776, 0.0852, 0.0731, 0.2285], device='cuda:0'), in_proj_covar=tensor([0.1303, 0.1205, 0.1008, 0.0747], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 16:14:13,722 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1357160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:14:14,443 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1357161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:14:20,909 INFO [train.py:968] (0/2) Epoch 30, batch 36700, libri_loss[loss=0.2682, simple_loss=0.3486, pruned_loss=0.09393, over 29354.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3511, pruned_loss=0.1027, over 5704992.84 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3483, pruned_loss=0.1088, over 5682077.47 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3503, pruned_loss=0.1013, over 5699009.15 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:15:04,603 INFO [train.py:968] (0/2) Epoch 30, batch 36750, giga_loss[loss=0.2534, simple_loss=0.3417, pruned_loss=0.08255, over 28790.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3487, pruned_loss=0.1005, over 5685688.50 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3481, pruned_loss=0.1087, over 5673241.76 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3483, pruned_loss=0.09935, over 5689519.22 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:15:14,488 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.328e+02 1.311e+03 1.651e+03 2.457e+03 7.373e+03, threshold=3.303e+03, percent-clipped=15.0
+2023-03-15 16:15:52,407 INFO [train.py:968] (0/2) Epoch 30, batch 36800, giga_loss[loss=0.2543, simple_loss=0.3253, pruned_loss=0.09161, over 27993.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3441, pruned_loss=0.09775, over 5669016.01 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3485, pruned_loss=0.1089, over 5666830.45 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3432, pruned_loss=0.09643, over 5678684.89 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:16:01,314 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4681, 1.8264, 1.8094, 1.6677], device='cuda:0'), covar=tensor([0.2590, 0.2444, 0.2572, 0.2308], device='cuda:0'), in_proj_covar=tensor([0.0509, 0.0761, 0.0731, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 16:16:27,461 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1357304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:16:30,104 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1357307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:16:41,826 INFO [train.py:968] (0/2) Epoch 30, batch 36850, giga_loss[loss=0.2198, simple_loss=0.3006, pruned_loss=0.06943, over 28593.00 frames. ], tot_loss[loss=0.265, simple_loss=0.339, pruned_loss=0.09554, over 5663275.76 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3489, pruned_loss=0.1091, over 5669374.17 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3378, pruned_loss=0.09398, over 5668379.81 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:16:52,872 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.121e+02 1.131e+03 1.426e+03 1.988e+03 6.269e+03, threshold=2.852e+03, percent-clipped=8.0
+2023-03-15 16:16:54,487 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1357328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:17:04,100 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1944, 1.1914, 3.3768, 3.0101], device='cuda:0'), covar=tensor([0.1654, 0.2911, 0.0495, 0.1108], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0679, 0.1015, 0.0990], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 16:17:04,623 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1357336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:17:10,181 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6757, 1.6630, 1.8546, 1.4818], device='cuda:0'), covar=tensor([0.1538, 0.2223, 0.1257, 0.1569], device='cuda:0'), in_proj_covar=tensor([0.0945, 0.0718, 0.0998, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 16:17:21,270 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1075, 3.9786, 3.7465, 1.7257], device='cuda:0'), covar=tensor([0.0603, 0.0700, 0.0675, 0.2147], device='cuda:0'), in_proj_covar=tensor([0.1304, 0.1207, 0.1010, 0.0749], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 16:17:38,840 INFO [train.py:968] (0/2) Epoch 30, batch 36900, giga_loss[loss=0.2583, simple_loss=0.3264, pruned_loss=0.09509, over 28741.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3342, pruned_loss=0.09341, over 5650607.13 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3488, pruned_loss=0.1089, over 5668708.05 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3331, pruned_loss=0.09204, over 5655079.90 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:18:23,675 INFO [train.py:968] (0/2) Epoch 30, batch 36950, giga_loss[loss=0.249, simple_loss=0.3294, pruned_loss=0.08424, over 28841.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3342, pruned_loss=0.09277, over 5659468.09 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3491, pruned_loss=0.1091, over 5672192.69 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3327, pruned_loss=0.09117, over 5659542.85 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:18:31,585 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1357426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:18:31,918 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.919e+02 1.172e+03 1.618e+03 2.052e+03 1.138e+04, threshold=3.236e+03, percent-clipped=9.0
+2023-03-15 16:19:05,375 INFO [train.py:968] (0/2) Epoch 30, batch 37000, giga_loss[loss=0.2351, simple_loss=0.3221, pruned_loss=0.07409, over 29047.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3336, pruned_loss=0.09146, over 5663716.31 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3495, pruned_loss=0.1093, over 5665837.48 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3319, pruned_loss=0.08984, over 5668427.03 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:19:47,241 INFO [train.py:968] (0/2) Epoch 30, batch 37050, giga_loss[loss=0.2468, simple_loss=0.3248, pruned_loss=0.08447, over 28686.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.333, pruned_loss=0.09121, over 5671237.90 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3504, pruned_loss=0.1099, over 5663078.30 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3301, pruned_loss=0.08863, over 5678285.56 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:19:54,724 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.909e+02 1.206e+03 1.464e+03 1.911e+03 1.086e+04, threshold=2.928e+03, percent-clipped=6.0
+2023-03-15 16:20:02,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1357535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:20:25,677 INFO [train.py:968] (0/2) Epoch 30, batch 37100, giga_loss[loss=0.2866, simple_loss=0.354, pruned_loss=0.1096, over 28858.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3318, pruned_loss=0.09069, over 5687153.00 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3513, pruned_loss=0.1101, over 5665709.74 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3281, pruned_loss=0.0878, over 5690864.79 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:20:30,010 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5342, 1.8424, 1.7371, 1.6655], device='cuda:0'), covar=tensor([0.2530, 0.2626, 0.2744, 0.2667], device='cuda:0'), in_proj_covar=tensor([0.0510, 0.0761, 0.0732, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 16:20:39,256 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4786, 1.6624, 1.6761, 1.2772], device='cuda:0'), covar=tensor([0.1800, 0.2854, 0.1595, 0.1890], device='cuda:0'), in_proj_covar=tensor([0.0943, 0.0718, 0.0996, 0.0895], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 16:21:06,985 INFO [train.py:968] (0/2) Epoch 30, batch 37150, libri_loss[loss=0.2899, simple_loss=0.3604, pruned_loss=0.1097, over 29511.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3298, pruned_loss=0.08978, over 5701869.24 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3514, pruned_loss=0.1098, over 5673461.93 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3261, pruned_loss=0.08696, over 5698841.78 frames. ], batch size: 80, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:21:12,813 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.406e+02 1.156e+03 1.292e+03 2.102e+03 4.745e+03, threshold=2.584e+03, percent-clipped=9.0
+2023-03-15 16:21:39,660 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2341, 1.4436, 1.3873, 1.0797], device='cuda:0'), covar=tensor([0.1254, 0.2009, 0.1116, 0.1383], device='cuda:0'), in_proj_covar=tensor([0.0944, 0.0718, 0.0997, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 16:21:43,692 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5959, 3.6652, 1.7386, 1.7218], device='cuda:0'), covar=tensor([0.1042, 0.0352, 0.0868, 0.1346], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0573, 0.0416, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 16:21:46,818 INFO [train.py:968] (0/2) Epoch 30, batch 37200, giga_loss[loss=0.246, simple_loss=0.3253, pruned_loss=0.08337, over 28790.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3277, pruned_loss=0.08901, over 5694120.21 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3519, pruned_loss=0.1101, over 5666419.80 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3239, pruned_loss=0.08632, over 5699000.54 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:21:54,804 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1357678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:21:57,780 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1357681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:22:13,535 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1357703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:22:18,673 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1357710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:22:25,919 INFO [train.py:968] (0/2) Epoch 30, batch 37250, giga_loss[loss=0.3055, simple_loss=0.3704, pruned_loss=0.1203, over 27550.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3257, pruned_loss=0.08824, over 5705875.83 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3524, pruned_loss=0.1101, over 5674877.21 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3215, pruned_loss=0.08536, over 5703347.67 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:22:35,364 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.785e+02 1.226e+03 1.497e+03 2.348e+03 6.318e+03, threshold=2.994e+03, percent-clipped=19.0
+2023-03-15 16:23:06,101 INFO [train.py:968] (0/2) Epoch 30, batch 37300, giga_loss[loss=0.2193, simple_loss=0.2896, pruned_loss=0.07451, over 28970.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3238, pruned_loss=0.08726, over 5711400.32 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3532, pruned_loss=0.1106, over 5670577.01 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3192, pruned_loss=0.08403, over 5714799.38 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:23:27,159 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-15 16:23:34,007 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1357801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:23:41,557 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1357809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:23:47,437 INFO [train.py:968] (0/2) Epoch 30, batch 37350, giga_loss[loss=0.209, simple_loss=0.2893, pruned_loss=0.06434, over 28871.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3218, pruned_loss=0.08597, over 5709799.26 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3536, pruned_loss=0.1105, over 5669413.64 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3169, pruned_loss=0.0829, over 5714080.10 frames. ], batch size: 112, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:23:55,502 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.670e+02 1.162e+03 1.433e+03 2.186e+03 7.323e+03, threshold=2.866e+03, percent-clipped=13.0
+2023-03-15 16:24:10,273 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1357846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:24:12,802 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4516, 3.1122, 1.5469, 1.6547], device='cuda:0'), covar=tensor([0.0909, 0.0316, 0.0775, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0573, 0.0416, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 16:24:13,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1357849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:24:27,554 INFO [train.py:968] (0/2) Epoch 30, batch 37400, giga_loss[loss=0.2267, simple_loss=0.3067, pruned_loss=0.07338, over 28352.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3182, pruned_loss=0.08391, over 5712781.96 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3534, pruned_loss=0.1103, over 5672821.20 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3141, pruned_loss=0.08137, over 5713624.75 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:24:35,694 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1357878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:24:39,678 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-15 16:25:10,227 INFO [train.py:968] (0/2) Epoch 30, batch 37450, giga_loss[loss=0.2185, simple_loss=0.2942, pruned_loss=0.07144, over 28528.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3173, pruned_loss=0.08318, over 5712813.21 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3536, pruned_loss=0.1102, over 5675134.22 frames. ], giga_tot_loss[loss=0.2378, simple_loss=0.3136, pruned_loss=0.081, over 5711843.40 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:25:17,860 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.460e+02 1.169e+03 1.443e+03 1.841e+03 5.922e+03, threshold=2.886e+03, percent-clipped=6.0
+2023-03-15 16:25:33,981 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1357944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:25:37,097 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1357947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:25:54,073 INFO [train.py:968] (0/2) Epoch 30, batch 37500, giga_loss[loss=0.234, simple_loss=0.3136, pruned_loss=0.07722, over 28995.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3161, pruned_loss=0.08259, over 5713785.08 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3536, pruned_loss=0.1101, over 5678450.78 frames. ], giga_tot_loss[loss=0.237, simple_loss=0.3127, pruned_loss=0.08059, over 5710658.03 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:25:55,183 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2696, 1.1335, 1.1525, 1.4926], device='cuda:0'), covar=tensor([0.0834, 0.0398, 0.0375, 0.0906], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 16:26:01,058 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1357976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:26:21,401 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1358000.pt
+2023-03-15 16:26:31,510 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1903, 2.4024, 1.2866, 1.2974], device='cuda:0'), covar=tensor([0.1062, 0.0418, 0.0935, 0.1415], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0574, 0.0417, 0.0452], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 16:26:38,506 INFO [train.py:968] (0/2) Epoch 30, batch 37550, giga_loss[loss=0.355, simple_loss=0.407, pruned_loss=0.1515, over 28351.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3211, pruned_loss=0.08571, over 5718675.87 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3539, pruned_loss=0.1101, over 5682897.99 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3175, pruned_loss=0.08369, over 5712995.73 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:26:40,277 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6130, 1.8468, 1.2460, 1.3676], device='cuda:0'), covar=tensor([0.1128, 0.0627, 0.1147, 0.1220], device='cuda:0'), in_proj_covar=tensor([0.0419, 0.0452, 0.0528, 0.0467], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 16:26:47,176 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.264e+02 1.258e+03 1.596e+03 2.067e+03 5.462e+03, threshold=3.192e+03, percent-clipped=10.0
+2023-03-15 16:27:23,401 INFO [train.py:968] (0/2) Epoch 30, batch 37600, giga_loss[loss=0.2745, simple_loss=0.3522, pruned_loss=0.09842, over 28950.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3267, pruned_loss=0.08913, over 5720738.34 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3537, pruned_loss=0.1098, over 5686095.63 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3235, pruned_loss=0.08737, over 5713833.57 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:27:30,731 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.1797, 1.6273, 5.4029, 3.7723], device='cuda:0'), covar=tensor([0.1529, 0.2788, 0.0384, 0.0840], device='cuda:0'), in_proj_covar=tensor([0.0816, 0.0678, 0.1011, 0.0989], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 16:28:11,360 INFO [train.py:968] (0/2) Epoch 30, batch 37650, giga_loss[loss=0.3066, simple_loss=0.3771, pruned_loss=0.1181, over 28679.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.336, pruned_loss=0.09548, over 5703111.98 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3536, pruned_loss=0.1096, over 5681500.33 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3329, pruned_loss=0.0937, over 5701701.12 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:28:17,203 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.2277, 3.0738, 2.9208, 1.4394], device='cuda:0'), covar=tensor([0.0957, 0.1013, 0.0870, 0.2332], device='cuda:0'), in_proj_covar=tensor([0.1309, 0.1208, 0.1013, 0.0753], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 16:28:22,346 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.070e+03 1.438e+03 1.907e+03 2.432e+03 6.949e+03, threshold=3.814e+03, percent-clipped=14.0
+2023-03-15 16:28:53,125 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([0.9677, 1.3026, 1.1185, 0.2303], device='cuda:0'), covar=tensor([0.5523, 0.4331, 0.5324, 0.8070], device='cuda:0'), in_proj_covar=tensor([0.1869, 0.1756, 0.1675, 0.1519], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 16:29:04,246 INFO [train.py:968] (0/2) Epoch 30, batch 37700, giga_loss[loss=0.2803, simple_loss=0.35, pruned_loss=0.1053, over 28682.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3409, pruned_loss=0.09797, over 5687726.62 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3536, pruned_loss=0.1096, over 5684843.57 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.338, pruned_loss=0.09626, over 5683886.95 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:29:18,843 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:29:21,653 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:29:51,977 INFO [train.py:968] (0/2) Epoch 30, batch 37750, giga_loss[loss=0.3077, simple_loss=0.3795, pruned_loss=0.118, over 28559.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3451, pruned_loss=0.0994, over 5695858.69 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3531, pruned_loss=0.1093, over 5690586.72 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.343, pruned_loss=0.0981, over 5687564.27 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:30:04,018 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.069e+03 1.387e+03 1.640e+03 1.988e+03 5.835e+03, threshold=3.280e+03, percent-clipped=5.0
+2023-03-15 16:30:04,883 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:30:42,142 INFO [train.py:968] (0/2) Epoch 30, batch 37800, giga_loss[loss=0.3078, simple_loss=0.3771, pruned_loss=0.1192, over 28172.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.351, pruned_loss=0.1026, over 5679764.53 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3533, pruned_loss=0.1094, over 5683548.03 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3492, pruned_loss=0.1013, over 5679113.01 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:31:26,654 INFO [train.py:968] (0/2) Epoch 30, batch 37850, giga_loss[loss=0.2496, simple_loss=0.3325, pruned_loss=0.08332, over 28793.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3536, pruned_loss=0.1033, over 5685957.96 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3535, pruned_loss=0.1094, over 5682299.06 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3519, pruned_loss=0.1022, over 5686500.84 frames. ], batch size: 78, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:31:35,412 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1358327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:31:36,675 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.023e+03 1.377e+03 1.700e+03 2.331e+03 7.833e+03, threshold=3.400e+03, percent-clipped=10.0
+2023-03-15 16:31:37,682 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1358330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:32:03,077 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1358359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:32:09,457 INFO [train.py:968] (0/2) Epoch 30, batch 37900, giga_loss[loss=0.2521, simple_loss=0.3353, pruned_loss=0.08444, over 28928.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3503, pruned_loss=0.1006, over 5691948.14 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3541, pruned_loss=0.1098, over 5686960.63 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3484, pruned_loss=0.09914, over 5688026.99 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:32:51,754 INFO [train.py:968] (0/2) Epoch 30, batch 37950, giga_loss[loss=0.2676, simple_loss=0.3441, pruned_loss=0.09555, over 28894.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3482, pruned_loss=0.09876, over 5699580.10 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3544, pruned_loss=0.11, over 5691164.03 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3463, pruned_loss=0.09717, over 5692879.48 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:33:03,505 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.456e+02 1.336e+03 1.733e+03 2.281e+03 6.840e+03, threshold=3.466e+03, percent-clipped=7.0
+2023-03-15 16:33:37,567 INFO [train.py:968] (0/2) Epoch 30, batch 38000, libri_loss[loss=0.2107, simple_loss=0.289, pruned_loss=0.06616, over 28146.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.346, pruned_loss=0.09688, over 5701823.98 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3538, pruned_loss=0.1096, over 5694003.09 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3449, pruned_loss=0.09577, over 5694172.98 frames. ], batch size: 62, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:34:00,928 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3753, 3.4819, 2.4145, 1.2929], device='cuda:0'), covar=tensor([0.8581, 0.2897, 0.3990, 0.7689], device='cuda:0'), in_proj_covar=tensor([0.1870, 0.1756, 0.1677, 0.1518], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 16:34:02,566 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:34:19,576 INFO [train.py:968] (0/2) Epoch 30, batch 38050, libri_loss[loss=0.2287, simple_loss=0.3022, pruned_loss=0.07754, over 29645.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3474, pruned_loss=0.09751, over 5702044.67 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3534, pruned_loss=0.1093, over 5695535.86 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3467, pruned_loss=0.09655, over 5694478.62 frames. ], batch size: 69, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:34:29,359 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.580e+02 1.374e+03 1.849e+03 2.328e+03 9.663e+03, threshold=3.698e+03, percent-clipped=7.0
+2023-03-15 16:34:50,192 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:35:01,414 INFO [train.py:968] (0/2) Epoch 30, batch 38100, giga_loss[loss=0.3183, simple_loss=0.3798, pruned_loss=0.1284, over 28650.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3514, pruned_loss=0.1003, over 5696875.34 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.354, pruned_loss=0.1095, over 5687363.64 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3503, pruned_loss=0.09914, over 5698429.65 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:35:08,369 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:35:36,673 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:35:47,969 INFO [train.py:968] (0/2) Epoch 30, batch 38150, libri_loss[loss=0.3076, simple_loss=0.3809, pruned_loss=0.1172, over 29645.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3535, pruned_loss=0.1021, over 5697637.62 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3543, pruned_loss=0.1096, over 5691809.61 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3522, pruned_loss=0.1008, over 5695103.23 frames. ], batch size: 91, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:36:00,059 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.584e+02 1.562e+03 1.949e+03 2.560e+03 7.072e+03, threshold=3.898e+03, percent-clipped=13.0
+2023-03-15 16:36:24,975 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:36:33,715 INFO [train.py:968] (0/2) Epoch 30, batch 38200, giga_loss[loss=0.2837, simple_loss=0.3585, pruned_loss=0.1044, over 28912.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3535, pruned_loss=0.1024, over 5696983.20 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3545, pruned_loss=0.1097, over 5694061.16 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3523, pruned_loss=0.1012, over 5693066.95 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:37:00,398 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1358698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:02,719 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1358701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:15,993 INFO [train.py:968] (0/2) Epoch 30, batch 38250, giga_loss[loss=0.334, simple_loss=0.3894, pruned_loss=0.1393, over 28028.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3541, pruned_loss=0.1035, over 5694299.25 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3553, pruned_loss=0.1104, over 5689172.14 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3525, pruned_loss=0.1018, over 5695609.08 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:37:25,961 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1358730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:26,324 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.012e+03 1.453e+03 1.787e+03 2.428e+03 8.315e+03, threshold=3.574e+03, percent-clipped=7.0
+2023-03-15 16:37:39,523 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1358747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:42,380 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1358750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:53,333 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:54,288 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-15 16:37:59,058 INFO [train.py:968] (0/2) Epoch 30, batch 38300, giga_loss[loss=0.2814, simple_loss=0.3591, pruned_loss=0.1018, over 28336.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3556, pruned_loss=0.1045, over 5663302.95 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3565, pruned_loss=0.1111, over 5660160.05 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.353, pruned_loss=0.1023, over 5690157.26 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:38:10,017 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1358779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:38:41,742 INFO [train.py:968] (0/2) Epoch 30, batch 38350, giga_loss[loss=0.2572, simple_loss=0.3426, pruned_loss=0.08592, over 29007.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3544, pruned_loss=0.1028, over 5679578.05 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3564, pruned_loss=0.1108, over 5663609.80 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3526, pruned_loss=0.1011, over 5697756.09 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:38:52,192 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.416e+02 1.372e+03 1.824e+03 2.503e+03 1.376e+04, threshold=3.647e+03, percent-clipped=10.0
+2023-03-15 16:39:13,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6896, 1.7440, 1.7412, 1.5594], device='cuda:0'), covar=tensor([0.3332, 0.3317, 0.2613, 0.3126], device='cuda:0'), in_proj_covar=tensor([0.2084, 0.2042, 0.1947, 0.2103], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 16:39:25,897 INFO [train.py:968] (0/2) Epoch 30, batch 38400, giga_loss[loss=0.2559, simple_loss=0.3398, pruned_loss=0.08602, over 28898.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3549, pruned_loss=0.1022, over 5678003.38 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3566, pruned_loss=0.1111, over 5653775.91 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3532, pruned_loss=0.1006, over 5702023.68 frames. ], batch size: 66, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:39:29,560 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:40:05,866 INFO [train.py:968] (0/2) Epoch 30, batch 38450, giga_loss[loss=0.2436, simple_loss=0.3243, pruned_loss=0.08146, over 28741.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3542, pruned_loss=0.1019, over 5687546.59 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3569, pruned_loss=0.111, over 5659519.56 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3525, pruned_loss=0.1003, over 5702862.70 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:40:17,463 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.994e+02 1.313e+03 1.688e+03 2.121e+03 5.263e+03, threshold=3.377e+03, percent-clipped=5.0
+2023-03-15 16:40:31,409 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:40:38,321 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9565, 2.1132, 2.1455, 1.7322], device='cuda:0'), covar=tensor([0.1872, 0.2446, 0.1450, 0.1735], device='cuda:0'), in_proj_covar=tensor([0.0946, 0.0721, 0.0999, 0.0896], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 16:40:46,337 INFO [train.py:968] (0/2) Epoch 30, batch 38500, giga_loss[loss=0.2695, simple_loss=0.3489, pruned_loss=0.0951, over 28122.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3523, pruned_loss=0.1011, over 5680110.01 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3569, pruned_loss=0.1111, over 5651031.36 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3508, pruned_loss=0.09956, over 5700787.33 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:41:26,912 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:41:27,992 INFO [train.py:968] (0/2) Epoch 30, batch 38550, giga_loss[loss=0.248, simple_loss=0.3295, pruned_loss=0.08323, over 28789.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3511, pruned_loss=0.1008, over 5669587.85 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3573, pruned_loss=0.1114, over 5638698.09 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3494, pruned_loss=0.09898, over 5699324.21 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:41:29,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:41:39,997 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.094e+02 1.260e+03 1.580e+03 2.230e+03 9.013e+03, threshold=3.161e+03, percent-clipped=9.0
+2023-03-15 16:41:41,784 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1359033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:41:43,812 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-15 16:41:45,037 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.4448, 4.3111, 4.0709, 2.0372], device='cuda:0'), covar=tensor([0.0585, 0.0701, 0.0748, 0.2042], device='cuda:0'), in_proj_covar=tensor([0.1312, 0.1212, 0.1016, 0.0754], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 16:41:52,412 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:42:08,723 INFO [train.py:968] (0/2) Epoch 30, batch 38600, giga_loss[loss=0.3528, simple_loss=0.3961, pruned_loss=0.1548, over 26703.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3504, pruned_loss=0.1007, over 5665628.77 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3582, pruned_loss=0.112, over 5626608.61 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.348, pruned_loss=0.09829, over 5701167.55 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:42:29,107 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:42:31,017 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:42:50,381 INFO [train.py:968] (0/2) Epoch 30, batch 38650, giga_loss[loss=0.2797, simple_loss=0.3534, pruned_loss=0.103, over 28624.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3506, pruned_loss=0.1009, over 5676795.73 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3583, pruned_loss=0.1122, over 5630826.40 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3485, pruned_loss=0.09874, over 5701818.48 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:42:55,431 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:42:59,823 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.693e+02 1.225e+03 1.514e+03 2.071e+03 5.477e+03, threshold=3.029e+03, percent-clipped=5.0
+2023-03-15 16:43:04,590 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1359136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:43:24,150 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:43:26,249 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8654, 1.1541, 2.9857, 2.8595], device='cuda:0'), covar=tensor([0.1596, 0.2475, 0.0645, 0.1123], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0680, 0.1015, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 16:43:29,238 INFO [train.py:968] (0/2) Epoch 30, batch 38700, giga_loss[loss=0.2822, simple_loss=0.3623, pruned_loss=0.101, over 28717.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3498, pruned_loss=0.1003, over 5689404.61 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3582, pruned_loss=0.1121, over 5639827.71 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3479, pruned_loss=0.09812, over 5703431.32 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:43:37,319 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:43:39,251 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:44:01,267 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.2969, 1.1067, 3.8213, 3.2665], device='cuda:0'), covar=tensor([0.1488, 0.2621, 0.0476, 0.0977], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0679, 0.1015, 0.0993], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 16:44:02,817 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:44:10,873 INFO [train.py:968] (0/2) Epoch 30, batch 38750, giga_loss[loss=0.2998, simple_loss=0.3725, pruned_loss=0.1136, over 28523.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3488, pruned_loss=0.09908, over 5696741.90 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3578, pruned_loss=0.1118, over 5645135.65 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3475, pruned_loss=0.09738, over 5704002.40 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:44:21,457 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.012e+02 1.340e+03 1.738e+03 2.824e+03 6.801e+03, threshold=3.477e+03, percent-clipped=21.0
+2023-03-15 16:44:40,186 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1817, 1.1231, 3.7175, 3.2706], device='cuda:0'), covar=tensor([0.1796, 0.3078, 0.0506, 0.0886], device='cuda:0'), in_proj_covar=tensor([0.0815, 0.0677, 0.1012, 0.0990], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 16:44:48,320 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-15 16:44:50,971 INFO [train.py:968] (0/2) Epoch 30, batch 38800, giga_loss[loss=0.3505, simple_loss=0.3957, pruned_loss=0.1527, over 26593.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3479, pruned_loss=0.09777, over 5706984.32 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3575, pruned_loss=0.1115, over 5650878.43 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3469, pruned_loss=0.09637, over 5709004.01 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:44:59,288 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:45:01,996 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:45:26,104 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:45:32,511 INFO [train.py:968] (0/2) Epoch 30, batch 38850, giga_loss[loss=0.2685, simple_loss=0.3408, pruned_loss=0.09806, over 28748.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3485, pruned_loss=0.09866, over 5705122.46 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.358, pruned_loss=0.1119, over 5656963.94 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3471, pruned_loss=0.09693, over 5702716.11 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:45:36,764 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:45:38,431 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-15 16:45:41,487 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.737e+02 1.160e+03 1.398e+03 1.701e+03 5.031e+03, threshold=2.796e+03, percent-clipped=5.0
+2023-03-15 16:46:10,909 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359365.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 16:46:12,949 INFO [train.py:968] (0/2) Epoch 30, batch 38900, giga_loss[loss=0.2577, simple_loss=0.3423, pruned_loss=0.08656, over 28383.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3448, pruned_loss=0.09665, over 5706467.76 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3574, pruned_loss=0.1116, over 5657430.51 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3441, pruned_loss=0.09541, over 5704520.76 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:46:33,337 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7739, 2.1169, 1.4190, 1.6302], device='cuda:0'), covar=tensor([0.1176, 0.0659, 0.1133, 0.1186], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0451, 0.0526, 0.0466], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 16:46:53,794 INFO [train.py:968] (0/2) Epoch 30, batch 38950, giga_loss[loss=0.2534, simple_loss=0.3262, pruned_loss=0.09029, over 28745.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3414, pruned_loss=0.09522, over 5692310.97 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3575, pruned_loss=0.1116, over 5648711.41 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3405, pruned_loss=0.09397, over 5698880.79 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:47:03,196 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.097e+02 1.221e+03 1.468e+03 2.004e+03 4.556e+03, threshold=2.937e+03, percent-clipped=10.0
+2023-03-15 16:47:33,161 INFO [train.py:968] (0/2) Epoch 30, batch 39000, giga_loss[loss=0.2531, simple_loss=0.3391, pruned_loss=0.08355, over 28683.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3398, pruned_loss=0.0943, over 5693941.27 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3579, pruned_loss=0.1119, over 5643618.32 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3385, pruned_loss=0.0928, over 5704239.93 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:47:33,168 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 16:47:38,475 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3749, 3.2192, 1.4826, 1.5447], device='cuda:0'), covar=tensor([0.1104, 0.0467, 0.1027, 0.1502], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0572, 0.0415, 0.0450], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 16:47:42,177 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.2055, simple_loss=0.3134, pruned_loss=0.04883, over 944034.00 frames. 
+2023-03-15 16:47:42,177 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 16:48:27,637 INFO [train.py:968] (0/2) Epoch 30, batch 39050, giga_loss[loss=0.2708, simple_loss=0.3408, pruned_loss=0.1004, over 28640.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3393, pruned_loss=0.09442, over 5699979.76 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.358, pruned_loss=0.1118, over 5646844.60 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3378, pruned_loss=0.09295, over 5706203.81 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:48:37,287 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.066e+02 1.261e+03 1.536e+03 2.007e+03 4.534e+03, threshold=3.073e+03, percent-clipped=7.0
+2023-03-15 16:48:41,599 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1359535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:49:07,016 INFO [train.py:968] (0/2) Epoch 30, batch 39100, giga_loss[loss=0.2263, simple_loss=0.3087, pruned_loss=0.07195, over 28885.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3376, pruned_loss=0.09401, over 5700142.33 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3588, pruned_loss=0.1124, over 5640023.10 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3355, pruned_loss=0.0922, over 5712547.79 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:49:47,687 INFO [train.py:968] (0/2) Epoch 30, batch 39150, giga_loss[loss=0.2452, simple_loss=0.3163, pruned_loss=0.08701, over 28624.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.336, pruned_loss=0.09384, over 5693246.40 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3592, pruned_loss=0.1127, over 5638970.35 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3333, pruned_loss=0.09152, over 5706673.99 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:49:58,727 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.383e+02 1.172e+03 1.477e+03 1.922e+03 6.471e+03, threshold=2.954e+03, percent-clipped=10.0
+2023-03-15 16:50:07,645 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4761, 1.8086, 1.3993, 1.4163], device='cuda:0'), covar=tensor([0.2832, 0.2921, 0.3372, 0.2771], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1172, 0.1439, 0.1020], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 16:50:25,696 INFO [train.py:968] (0/2) Epoch 30, batch 39200, giga_loss[loss=0.2495, simple_loss=0.3247, pruned_loss=0.08713, over 29061.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3326, pruned_loss=0.09224, over 5700135.47 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3586, pruned_loss=0.1123, over 5646511.75 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3302, pruned_loss=0.09011, over 5705562.71 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:50:34,646 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:50:36,754 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:50:52,088 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1359699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:51:00,213 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:51:04,297 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-15 16:51:06,714 INFO [train.py:968] (0/2) Epoch 30, batch 39250, giga_loss[loss=0.2272, simple_loss=0.3035, pruned_loss=0.07548, over 28591.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3303, pruned_loss=0.09088, over 5705962.23 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3587, pruned_loss=0.1123, over 5652651.26 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3275, pruned_loss=0.08855, over 5706518.48 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:51:22,408 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 7.931e+02 1.176e+03 1.534e+03 2.266e+03 1.097e+04, threshold=3.069e+03, percent-clipped=14.0
+2023-03-15 16:51:26,612 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1359740.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 16:51:30,120 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3110, 2.6272, 1.3669, 1.3561], device='cuda:0'), covar=tensor([0.1003, 0.0451, 0.0959, 0.1447], device='cuda:0'), in_proj_covar=tensor([0.0430, 0.0574, 0.0416, 0.0451], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:0')
+2023-03-15 16:51:49,774 INFO [train.py:968] (0/2) Epoch 30, batch 39300, libri_loss[loss=0.2957, simple_loss=0.3726, pruned_loss=0.1094, over 29478.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3329, pruned_loss=0.09235, over 5710489.09 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1126, over 5656886.98 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3293, pruned_loss=0.08957, over 5708927.10 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:52:15,778 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.23 vs. limit=5.0
+2023-03-15 16:52:35,883 INFO [train.py:968] (0/2) Epoch 30, batch 39350, giga_loss[loss=0.2311, simple_loss=0.3194, pruned_loss=0.07141, over 29024.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3355, pruned_loss=0.09277, over 5711568.14 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.359, pruned_loss=0.1124, over 5660315.08 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3324, pruned_loss=0.09047, over 5707868.58 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:52:41,781 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.1247, 1.0984, 1.1698, 1.3581], device='cuda:0'), covar=tensor([0.0801, 0.0350, 0.0328, 0.1106], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:0')
+2023-03-15 16:52:51,383 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.229e+02 1.252e+03 1.532e+03 2.265e+03 1.586e+04, threshold=3.065e+03, percent-clipped=16.0
+2023-03-15 16:52:58,363 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:53:00,235 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:53:11,075 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:53:20,841 INFO [train.py:968] (0/2) Epoch 30, batch 39400, giga_loss[loss=0.2957, simple_loss=0.3661, pruned_loss=0.1127, over 27605.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3384, pruned_loss=0.09386, over 5703277.62 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3587, pruned_loss=0.1121, over 5657628.22 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3355, pruned_loss=0.09157, over 5705178.61 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:53:26,157 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:53:28,723 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3067, 1.5994, 1.2965, 0.9716], device='cuda:0'), covar=tensor([0.2797, 0.2685, 0.3066, 0.2569], device='cuda:0'), in_proj_covar=tensor([0.1632, 0.1172, 0.1440, 0.1018], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 16:53:35,279 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359883.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 16:53:38,102 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 16:54:04,862 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359915.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 16:54:06,637 INFO [train.py:968] (0/2) Epoch 30, batch 39450, giga_loss[loss=0.2536, simple_loss=0.3384, pruned_loss=0.08433, over 28958.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3407, pruned_loss=0.09454, over 5702720.69 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3584, pruned_loss=0.1121, over 5663596.63 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3382, pruned_loss=0.09235, over 5699610.89 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:54:15,602 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:54:17,600 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.186e+02 1.213e+03 1.514e+03 2.272e+03 7.762e+03, threshold=3.028e+03, percent-clipped=12.0
+2023-03-15 16:54:37,470 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359957.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 16:54:46,148 INFO [train.py:968] (0/2) Epoch 30, batch 39500, giga_loss[loss=0.2375, simple_loss=0.3245, pruned_loss=0.07525, over 28709.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3409, pruned_loss=0.09416, over 5692736.05 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3585, pruned_loss=0.1121, over 5664544.54 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3383, pruned_loss=0.09192, over 5690125.34 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:55:11,631 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1360000.pt
+2023-03-15 16:55:26,511 INFO [train.py:968] (0/2) Epoch 30, batch 39550, giga_loss[loss=0.2399, simple_loss=0.3214, pruned_loss=0.07917, over 29041.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3405, pruned_loss=0.0934, over 5696173.03 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3584, pruned_loss=0.1121, over 5662785.63 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3379, pruned_loss=0.09104, over 5697166.96 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:55:39,868 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.676e+02 1.383e+03 1.660e+03 1.929e+03 4.100e+03, threshold=3.319e+03, percent-clipped=4.0
+2023-03-15 16:56:11,050 INFO [train.py:968] (0/2) Epoch 30, batch 39600, giga_loss[loss=0.3034, simple_loss=0.3758, pruned_loss=0.1155, over 28647.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3423, pruned_loss=0.09491, over 5688048.07 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3586, pruned_loss=0.1121, over 5662730.23 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3398, pruned_loss=0.09277, over 5689326.29 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:56:51,090 INFO [train.py:968] (0/2) Epoch 30, batch 39650, giga_loss[loss=0.3325, simple_loss=0.3895, pruned_loss=0.1377, over 26647.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3442, pruned_loss=0.09681, over 5680664.25 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5658965.80 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.341, pruned_loss=0.09396, over 5686286.63 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:56:57,813 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5995, 1.6364, 1.7767, 1.3914], device='cuda:0'), covar=tensor([0.1944, 0.2631, 0.1619, 0.1859], device='cuda:0'), in_proj_covar=tensor([0.0944, 0.0719, 0.0996, 0.0895], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 16:57:05,805 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.767e+02 1.500e+03 1.895e+03 2.543e+03 6.877e+03, threshold=3.791e+03, percent-clipped=11.0
+2023-03-15 16:57:14,973 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2538, 1.9395, 1.5110, 0.5549], device='cuda:0'), covar=tensor([0.6645, 0.3489, 0.4783, 0.7607], device='cuda:0'), in_proj_covar=tensor([0.1862, 0.1745, 0.1672, 0.1514], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 16:57:35,391 INFO [train.py:968] (0/2) Epoch 30, batch 39700, giga_loss[loss=0.3184, simple_loss=0.386, pruned_loss=0.1254, over 29073.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.347, pruned_loss=0.09826, over 5692094.91 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3593, pruned_loss=0.1127, over 5661572.44 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3443, pruned_loss=0.09584, over 5694448.55 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:57:36,586 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1360169.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 16:58:17,731 INFO [train.py:968] (0/2) Epoch 30, batch 39750, giga_loss[loss=0.3032, simple_loss=0.3733, pruned_loss=0.1166, over 28331.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3497, pruned_loss=0.09959, over 5703515.36 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3594, pruned_loss=0.1127, over 5666462.50 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3473, pruned_loss=0.09741, over 5701673.79 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:58:18,578 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1360219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:58:28,833 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:58:32,140 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.064e+03 1.463e+03 1.727e+03 2.196e+03 3.908e+03, threshold=3.454e+03, percent-clipped=1.0
+2023-03-15 16:59:00,634 INFO [train.py:968] (0/2) Epoch 30, batch 39800, giga_loss[loss=0.2839, simple_loss=0.3589, pruned_loss=0.1044, over 28932.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3501, pruned_loss=0.09928, over 5707407.90 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5665491.46 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3482, pruned_loss=0.09748, over 5707289.08 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:59:32,448 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1360303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:59:35,386 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:59:44,537 INFO [train.py:968] (0/2) Epoch 30, batch 39850, giga_loss[loss=0.2733, simple_loss=0.3526, pruned_loss=0.097, over 28788.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3516, pruned_loss=0.1002, over 5712122.23 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1127, over 5666903.00 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.35, pruned_loss=0.09874, over 5711089.09 frames. ], batch size: 243, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:59:58,778 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360332.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 17:00:00,499 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.948e+02 1.434e+03 1.691e+03 2.173e+03 4.384e+03, threshold=3.382e+03, percent-clipped=5.0
+2023-03-15 17:00:24,075 INFO [train.py:968] (0/2) Epoch 30, batch 39900, giga_loss[loss=0.2749, simple_loss=0.3531, pruned_loss=0.0984, over 29044.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.351, pruned_loss=0.09994, over 5707052.79 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5662939.70 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3493, pruned_loss=0.09844, over 5711694.58 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:00:28,215 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:00:31,184 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1360377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:00:52,150 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-15 17:00:54,961 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:01:03,652 INFO [train.py:968] (0/2) Epoch 30, batch 39950, giga_loss[loss=0.2432, simple_loss=0.3282, pruned_loss=0.0791, over 28801.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3494, pruned_loss=0.099, over 5708452.44 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1127, over 5666641.61 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3476, pruned_loss=0.09735, over 5710133.30 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:01:13,759 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.3905, 1.2347, 3.9755, 3.3271], device='cuda:0'), covar=tensor([0.1623, 0.2924, 0.0446, 0.0967], device='cuda:0'), in_proj_covar=tensor([0.0818, 0.0678, 0.1015, 0.0994], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 17:01:16,462 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.677e+02 1.388e+03 1.781e+03 2.361e+03 7.021e+03, threshold=3.561e+03, percent-clipped=11.0
+2023-03-15 17:01:18,426 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-15 17:01:30,367 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:01:33,552 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1360452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:01:33,566 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8201, 2.6952, 1.6823, 1.0020], device='cuda:0'), covar=tensor([1.0203, 0.4438, 0.5057, 0.8902], device='cuda:0'), in_proj_covar=tensor([0.1863, 0.1747, 0.1670, 0.1514], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 17:01:40,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7373, 1.8908, 1.3666, 1.4314], device='cuda:0'), covar=tensor([0.1069, 0.0694, 0.1111, 0.1263], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0452, 0.0527, 0.0465], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 17:01:46,621 INFO [train.py:968] (0/2) Epoch 30, batch 40000, giga_loss[loss=0.2395, simple_loss=0.3251, pruned_loss=0.07696, over 29067.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3472, pruned_loss=0.09821, over 5706860.04 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1127, over 5668521.53 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3457, pruned_loss=0.09684, over 5706940.05 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:01:52,928 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360475.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 17:01:56,885 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1360478.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 17:01:58,879 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:02:06,377 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.1564, 3.9925, 3.7802, 1.9287], device='cuda:0'), covar=tensor([0.0631, 0.0752, 0.0692, 0.2128], device='cuda:0'), in_proj_covar=tensor([0.1311, 0.1210, 0.1016, 0.0752], device='cuda:0'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:0')
+2023-03-15 17:02:18,276 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360507.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 17:02:29,338 INFO [train.py:968] (0/2) Epoch 30, batch 40050, libri_loss[loss=0.2869, simple_loss=0.3615, pruned_loss=0.1062, over 27862.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3436, pruned_loss=0.09608, over 5715417.16 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3602, pruned_loss=0.1128, over 5670193.28 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3417, pruned_loss=0.09451, over 5715022.95 frames. ], batch size: 116, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:02:41,386 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.363e+02 1.296e+03 1.857e+03 2.596e+03 9.302e+03, threshold=3.714e+03, percent-clipped=9.0
+2023-03-15 17:02:48,111 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-15 17:02:48,618 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360544.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 17:03:07,313 INFO [train.py:968] (0/2) Epoch 30, batch 40100, giga_loss[loss=0.2356, simple_loss=0.3286, pruned_loss=0.07128, over 28954.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3418, pruned_loss=0.09498, over 5711728.89 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3601, pruned_loss=0.1128, over 5672485.83 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.34, pruned_loss=0.09333, over 5710113.91 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:03:28,687 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:03:38,843 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4277, 2.1956, 1.6635, 0.7657], device='cuda:0'), covar=tensor([0.6721, 0.3394, 0.4579, 0.7475], device='cuda:0'), in_proj_covar=tensor([0.1867, 0.1750, 0.1670, 0.1514], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 17:03:48,018 INFO [train.py:968] (0/2) Epoch 30, batch 40150, giga_loss[loss=0.2694, simple_loss=0.3549, pruned_loss=0.09193, over 28591.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3427, pruned_loss=0.09438, over 5703410.53 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3599, pruned_loss=0.1128, over 5668465.19 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3411, pruned_loss=0.09271, over 5707208.18 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:04:02,640 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.791e+02 1.297e+03 1.493e+03 1.943e+03 5.989e+03, threshold=2.986e+03, percent-clipped=2.0
+2023-03-15 17:04:29,837 INFO [train.py:968] (0/2) Epoch 30, batch 40200, giga_loss[loss=0.2756, simple_loss=0.3501, pruned_loss=0.1005, over 29101.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3443, pruned_loss=0.09513, over 5700807.61 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3603, pruned_loss=0.113, over 5668852.99 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3423, pruned_loss=0.09318, over 5704250.99 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:04:38,390 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:04:45,218 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360687.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 17:04:46,147 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-15 17:04:47,286 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1360690.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 17:04:53,357 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-15 17:05:08,401 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1360716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:05:09,522 INFO [train.py:968] (0/2) Epoch 30, batch 40250, giga_loss[loss=0.2534, simple_loss=0.3278, pruned_loss=0.0895, over 28804.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3427, pruned_loss=0.09517, over 5710966.44 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3598, pruned_loss=0.1127, over 5672791.39 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3411, pruned_loss=0.09345, over 5710930.48 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:05:10,762 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360719.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 17:05:21,544 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-15 17:05:23,308 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.514e+02 1.436e+03 1.718e+03 2.147e+03 7.971e+03, threshold=3.436e+03, percent-clipped=13.0
+2023-03-15 17:05:24,904 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:05:27,707 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1360740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:05:49,190 INFO [train.py:968] (0/2) Epoch 30, batch 40300, giga_loss[loss=0.2061, simple_loss=0.2834, pruned_loss=0.06443, over 28492.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.342, pruned_loss=0.09578, over 5700006.11 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3602, pruned_loss=0.1129, over 5658948.97 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3401, pruned_loss=0.09385, over 5714322.70 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:05:50,160 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:06:21,686 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-15 17:06:32,236 INFO [train.py:968] (0/2) Epoch 30, batch 40350, giga_loss[loss=0.2683, simple_loss=0.3423, pruned_loss=0.09708, over 28579.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3394, pruned_loss=0.09559, over 5692362.10 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.113, over 5651591.42 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3375, pruned_loss=0.0936, over 5711380.57 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:06:37,266 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:06:40,123 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1360824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:06:47,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.147e+02 1.328e+03 1.720e+03 2.377e+03 1.054e+04, threshold=3.439e+03, percent-clipped=10.0
+2023-03-15 17:06:47,491 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-15 17:07:02,527 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:07:13,257 INFO [train.py:968] (0/2) Epoch 30, batch 40400, giga_loss[loss=0.2482, simple_loss=0.3215, pruned_loss=0.08748, over 28996.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3374, pruned_loss=0.09531, over 5686986.11 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3599, pruned_loss=0.1128, over 5650619.67 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3353, pruned_loss=0.09334, over 5704847.45 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:07:55,362 INFO [train.py:968] (0/2) Epoch 30, batch 40450, giga_loss[loss=0.2623, simple_loss=0.3279, pruned_loss=0.0984, over 28973.00 frames. ], tot_loss[loss=0.264, simple_loss=0.337, pruned_loss=0.09554, over 5689429.34 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3597, pruned_loss=0.1126, over 5654659.32 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3349, pruned_loss=0.09366, over 5700738.10 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:08:00,199 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4372, 1.5428, 1.2070, 1.1205], device='cuda:0'), covar=tensor([0.0888, 0.0512, 0.0943, 0.1305], device='cuda:0'), in_proj_covar=tensor([0.0418, 0.0452, 0.0526, 0.0464], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:0')
+2023-03-15 17:08:09,329 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 8.943e+02 1.270e+03 1.750e+03 2.829e+03 8.631e+03, threshold=3.500e+03, percent-clipped=19.0
+2023-03-15 17:08:16,770 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 17:08:19,336 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-15 17:08:31,746 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1360967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:08:32,166 INFO [train.py:968] (0/2) Epoch 30, batch 40500, giga_loss[loss=0.2407, simple_loss=0.3175, pruned_loss=0.08194, over 28905.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3351, pruned_loss=0.09505, over 5700067.96 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5660566.25 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3325, pruned_loss=0.09282, over 5705465.80 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:09:00,125 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:09:11,589 INFO [train.py:968] (0/2) Epoch 30, batch 40550, giga_loss[loss=0.2363, simple_loss=0.3124, pruned_loss=0.08004, over 28838.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3311, pruned_loss=0.09326, over 5704183.46 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3596, pruned_loss=0.1125, over 5665934.19 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3284, pruned_loss=0.09102, over 5704984.03 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:09:27,939 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.288e+02 1.288e+03 1.637e+03 2.085e+03 6.468e+03, threshold=3.273e+03, percent-clipped=7.0
+2023-03-15 17:09:54,161 INFO [train.py:968] (0/2) Epoch 30, batch 40600, giga_loss[loss=0.2523, simple_loss=0.3236, pruned_loss=0.09048, over 28926.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3283, pruned_loss=0.09142, over 5709891.58 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3592, pruned_loss=0.1123, over 5668806.08 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3256, pruned_loss=0.08933, over 5708797.49 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:10:11,002 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1361091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:10:34,360 INFO [train.py:968] (0/2) Epoch 30, batch 40650, giga_loss[loss=0.2536, simple_loss=0.3271, pruned_loss=0.09007, over 28787.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3288, pruned_loss=0.09099, over 5717579.32 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3589, pruned_loss=0.1121, over 5674597.91 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3262, pruned_loss=0.08895, over 5712423.42 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:10:49,141 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.027e+03 1.357e+03 1.683e+03 2.240e+03 6.431e+03, threshold=3.365e+03, percent-clipped=6.0
+2023-03-15 17:11:15,055 INFO [train.py:968] (0/2) Epoch 30, batch 40700, giga_loss[loss=0.2386, simple_loss=0.3206, pruned_loss=0.07834, over 28608.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3328, pruned_loss=0.0926, over 5716197.64 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3589, pruned_loss=0.1121, over 5679311.55 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3303, pruned_loss=0.09064, over 5708789.62 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:11:32,767 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8932, 2.1574, 1.7288, 1.8947], device='cuda:0'), covar=tensor([0.2574, 0.2755, 0.3171, 0.2793], device='cuda:0'), in_proj_covar=tensor([0.1633, 0.1171, 0.1439, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 17:11:55,085 INFO [train.py:968] (0/2) Epoch 30, batch 40750, giga_loss[loss=0.3635, simple_loss=0.4021, pruned_loss=0.1624, over 23818.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3361, pruned_loss=0.09415, over 5714031.43 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3585, pruned_loss=0.1117, over 5682923.09 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3333, pruned_loss=0.09216, over 5706426.26 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:11:55,903 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:12:07,053 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:12:08,150 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 9.124e+02 1.320e+03 1.683e+03 2.489e+03 1.175e+04, threshold=3.366e+03, percent-clipped=8.0
+2023-03-15 17:12:09,036 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:12:33,645 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1361266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:12:34,941 INFO [train.py:968] (0/2) Epoch 30, batch 40800, giga_loss[loss=0.3195, simple_loss=0.3867, pruned_loss=0.1261, over 28334.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.339, pruned_loss=0.09501, over 5701959.14 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3581, pruned_loss=0.1114, over 5680183.69 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3364, pruned_loss=0.09304, over 5700023.82 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:12:58,413 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 17:13:15,624 INFO [train.py:968] (0/2) Epoch 30, batch 40850, giga_loss[loss=0.3179, simple_loss=0.3794, pruned_loss=0.1282, over 27660.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3426, pruned_loss=0.09673, over 5709734.54 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3579, pruned_loss=0.1112, over 5683710.49 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3402, pruned_loss=0.0949, over 5705698.14 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:13:30,512 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.007e+03 1.338e+03 1.628e+03 2.196e+03 1.005e+04, threshold=3.256e+03, percent-clipped=5.0
+2023-03-15 17:13:33,174 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:13:36,223 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1361342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:14:02,061 INFO [train.py:968] (0/2) Epoch 30, batch 40900, giga_loss[loss=0.3977, simple_loss=0.4373, pruned_loss=0.1791, over 28027.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3456, pruned_loss=0.09913, over 5709161.74 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3576, pruned_loss=0.111, over 5686468.70 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3437, pruned_loss=0.09761, over 5703972.67 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:14:12,729 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1361377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:14:49,873 INFO [train.py:968] (0/2) Epoch 30, batch 40950, giga_loss[loss=0.2848, simple_loss=0.3565, pruned_loss=0.1065, over 29070.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3522, pruned_loss=0.1048, over 5702777.08 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3583, pruned_loss=0.1113, over 5685834.84 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3496, pruned_loss=0.1028, over 5700247.67 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:15:07,204 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.157e+03 1.793e+03 2.159e+03 2.854e+03 7.672e+03, threshold=4.317e+03, percent-clipped=18.0
+2023-03-15 17:15:37,650 INFO [train.py:968] (0/2) Epoch 30, batch 41000, giga_loss[loss=0.33, simple_loss=0.3898, pruned_loss=0.1351, over 28662.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3595, pruned_loss=0.1101, over 5698813.52 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3583, pruned_loss=0.1113, over 5686268.59 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3574, pruned_loss=0.1085, over 5696346.24 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:15:47,987 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:15:54,125 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:15:56,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:16:23,747 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1361517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:16:24,291 INFO [train.py:968] (0/2) Epoch 30, batch 41050, giga_loss[loss=0.3167, simple_loss=0.3766, pruned_loss=0.1284, over 28554.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3647, pruned_loss=0.114, over 5684806.49 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3576, pruned_loss=0.1109, over 5680889.29 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3637, pruned_loss=0.1131, over 5688704.07 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:16:26,862 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:16:28,826 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:16:39,017 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7594, 1.6993, 1.9066, 1.4827], device='cuda:0'), covar=tensor([0.1774, 0.2624, 0.1456, 0.1762], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0718, 0.0992, 0.0891], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 17:16:39,943 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.413e+03 1.970e+03 2.633e+03 3.466e+03 8.446e+03, threshold=5.267e+03, percent-clipped=14.0
+2023-03-15 17:16:53,784 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1361552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:17:08,342 INFO [train.py:968] (0/2) Epoch 30, batch 41100, giga_loss[loss=0.3997, simple_loss=0.4397, pruned_loss=0.1799, over 27662.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3696, pruned_loss=0.118, over 5682718.83 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3578, pruned_loss=0.111, over 5674880.61 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3689, pruned_loss=0.1173, over 5691130.40 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:17:20,145 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5519, 2.2428, 1.7335, 0.8066], device='cuda:0'), covar=tensor([0.6301, 0.3197, 0.3697, 0.6950], device='cuda:0'), in_proj_covar=tensor([0.1873, 0.1762, 0.1674, 0.1520], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 17:17:34,022 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1361594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:17:56,571 INFO [train.py:968] (0/2) Epoch 30, batch 41150, giga_loss[loss=0.4267, simple_loss=0.4526, pruned_loss=0.2003, over 27446.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3753, pruned_loss=0.1226, over 5685295.91 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3577, pruned_loss=0.1108, over 5679201.01 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.375, pruned_loss=0.1224, over 5688191.69 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:18:15,654 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.195e+03 1.935e+03 2.401e+03 3.392e+03 8.548e+03, threshold=4.802e+03, percent-clipped=7.0
+2023-03-15 17:18:23,702 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:18:28,369 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:18:33,774 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-15 17:18:49,719 INFO [train.py:968] (0/2) Epoch 30, batch 41200, giga_loss[loss=0.3455, simple_loss=0.4031, pruned_loss=0.1439, over 28626.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3787, pruned_loss=0.1267, over 5650599.22 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3581, pruned_loss=0.1111, over 5669841.69 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3787, pruned_loss=0.1265, over 5660997.36 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:19:14,208 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2881, 3.0772, 1.4554, 1.4042], device='cuda:0'), covar=tensor([0.1032, 0.0375, 0.0901, 0.1359], device='cuda:0'), in_proj_covar=tensor([0.0431, 0.0580, 0.0418, 0.0453], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 17:19:38,594 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1361713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:19:44,706 INFO [train.py:968] (0/2) Epoch 30, batch 41250, giga_loss[loss=0.3738, simple_loss=0.4197, pruned_loss=0.164, over 28572.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3807, pruned_loss=0.1294, over 5657613.76 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3581, pruned_loss=0.1111, over 5673893.39 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3811, pruned_loss=0.1296, over 5662016.28 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:20:07,187 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:20:08,474 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.377e+03 1.962e+03 2.349e+03 3.084e+03 7.598e+03, threshold=4.698e+03, percent-clipped=6.0
+2023-03-15 17:20:08,836 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6563, 3.8002, 1.6346, 1.6775], device='cuda:0'), covar=tensor([0.0950, 0.0359, 0.0869, 0.1321], device='cuda:0'), in_proj_covar=tensor([0.0432, 0.0580, 0.0419, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 17:20:10,348 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:20:39,631 INFO [train.py:968] (0/2) Epoch 30, batch 41300, giga_loss[loss=0.3409, simple_loss=0.4002, pruned_loss=0.1407, over 28862.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3849, pruned_loss=0.134, over 5641369.65 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3584, pruned_loss=0.1113, over 5677134.89 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3853, pruned_loss=0.1343, over 5641523.47 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:20:42,118 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1361769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:21:00,803 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:21:30,084 INFO [train.py:968] (0/2) Epoch 30, batch 41350, libri_loss[loss=0.2976, simple_loss=0.3643, pruned_loss=0.1155, over 29517.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3856, pruned_loss=0.1345, over 5647469.03 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3579, pruned_loss=0.1109, over 5685009.49 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3874, pruned_loss=0.1361, over 5639041.99 frames. ], batch size: 84, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:21:53,054 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.356e+03 2.036e+03 2.794e+03 3.754e+03 6.717e+03, threshold=5.589e+03, percent-clipped=11.0
+2023-03-15 17:22:05,529 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4112, 1.5309, 1.1706, 1.0908], device='cuda:0'), covar=tensor([0.0876, 0.0427, 0.0898, 0.1132], device='cuda:0'), in_proj_covar=tensor([0.0420, 0.0456, 0.0529, 0.0468], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 17:22:07,864 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1361853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:22:10,165 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:22:14,679 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:22:24,588 INFO [train.py:968] (0/2) Epoch 30, batch 41400, giga_loss[loss=0.281, simple_loss=0.3574, pruned_loss=0.1023, over 28962.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3869, pruned_loss=0.136, over 5632762.25 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3576, pruned_loss=0.1106, over 5688182.71 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.389, pruned_loss=0.1378, over 5622692.66 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:22:50,652 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1361888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:23:19,680 INFO [train.py:968] (0/2) Epoch 30, batch 41450, giga_loss[loss=0.4444, simple_loss=0.459, pruned_loss=0.2149, over 26566.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3873, pruned_loss=0.1372, over 5623500.95 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3578, pruned_loss=0.1108, over 5678910.31 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3892, pruned_loss=0.1388, over 5622803.64 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:23:41,430 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.186e+03 2.108e+03 2.768e+03 3.844e+03 1.210e+04, threshold=5.536e+03, percent-clipped=9.0
+2023-03-15 17:24:07,271 INFO [train.py:968] (0/2) Epoch 30, batch 41500, giga_loss[loss=0.3326, simple_loss=0.3893, pruned_loss=0.1379, over 28819.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3848, pruned_loss=0.1349, over 5649148.52 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3576, pruned_loss=0.1106, over 5685461.56 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3874, pruned_loss=0.1373, over 5641203.28 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:24:33,554 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:24:35,610 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:24:37,110 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1362000.pt
+2023-03-15 17:25:00,475 INFO [train.py:968] (0/2) Epoch 30, batch 41550, giga_loss[loss=0.4414, simple_loss=0.4421, pruned_loss=0.2203, over 23548.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3855, pruned_loss=0.1344, over 5641294.76 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3577, pruned_loss=0.1107, over 5681213.86 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3883, pruned_loss=0.1368, over 5637807.96 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:25:02,865 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1362020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:25:06,347 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1362024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:25:09,885 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:25:24,157 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.295e+03 1.780e+03 2.255e+03 3.143e+03 1.401e+04, threshold=4.510e+03, percent-clipped=5.0
+2023-03-15 17:25:53,527 INFO [train.py:968] (0/2) Epoch 30, batch 41600, giga_loss[loss=0.3036, simple_loss=0.3763, pruned_loss=0.1155, over 28644.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3874, pruned_loss=0.1349, over 5649665.09 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3578, pruned_loss=0.111, over 5673590.58 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3896, pruned_loss=0.1366, over 5654266.60 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:26:19,722 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6392, 1.9086, 1.6095, 1.5232], device='cuda:0'), covar=tensor([0.2206, 0.1914, 0.2060, 0.1934], device='cuda:0'), in_proj_covar=tensor([0.1630, 0.1170, 0.1438, 0.1019], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 17:26:47,112 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-15 17:26:49,119 INFO [train.py:968] (0/2) Epoch 30, batch 41650, giga_loss[loss=0.2991, simple_loss=0.3689, pruned_loss=0.1147, over 28576.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3885, pruned_loss=0.1357, over 5635686.44 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3575, pruned_loss=0.1108, over 5676104.88 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.391, pruned_loss=0.1378, over 5636731.66 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:26:50,015 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3994, 1.7297, 1.5583, 1.5148], device='cuda:0'), covar=tensor([0.0801, 0.0330, 0.0324, 0.0902], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 17:27:11,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.280e+03 1.893e+03 2.546e+03 3.273e+03 1.611e+04, threshold=5.092e+03, percent-clipped=9.0
+2023-03-15 17:27:36,806 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1362163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:27:36,836 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1362163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:27:39,218 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1362166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:27:39,880 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1362167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:27:40,194 INFO [train.py:968] (0/2) Epoch 30, batch 41700, giga_loss[loss=0.2847, simple_loss=0.3572, pruned_loss=0.1061, over 28946.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3847, pruned_loss=0.1317, over 5645551.61 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3577, pruned_loss=0.1111, over 5682498.82 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3873, pruned_loss=0.1336, over 5639777.70 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:27:42,518 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1362170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:28:07,344 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:28:12,166 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:28:29,247 INFO [train.py:968] (0/2) Epoch 30, batch 41750, giga_loss[loss=0.2739, simple_loss=0.3607, pruned_loss=0.09361, over 28654.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3827, pruned_loss=0.1293, over 5643516.56 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3576, pruned_loss=0.111, over 5676478.00 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3855, pruned_loss=0.1313, over 5644317.31 frames. ], batch size: 78, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:28:52,786 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.192e+03 1.876e+03 2.427e+03 2.954e+03 5.286e+03, threshold=4.854e+03, percent-clipped=1.0
+2023-03-15 17:29:14,940 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5007, 1.6494, 1.3162, 1.2070], device='cuda:0'), covar=tensor([0.1094, 0.0628, 0.1046, 0.1247], device='cuda:0'), in_proj_covar=tensor([0.0421, 0.0456, 0.0530, 0.0470], device='cuda:0'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:0')
+2023-03-15 17:29:20,618 INFO [train.py:968] (0/2) Epoch 30, batch 41800, libri_loss[loss=0.2813, simple_loss=0.3457, pruned_loss=0.1084, over 29586.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3783, pruned_loss=0.1252, over 5645420.91 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3578, pruned_loss=0.1113, over 5670209.24 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3807, pruned_loss=0.1268, over 5651616.22 frames. ], batch size: 75, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:29:29,919 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6970, 1.5986, 1.9032, 1.4687], device='cuda:0'), covar=tensor([0.1814, 0.2541, 0.1471, 0.1779], device='cuda:0'), in_proj_covar=tensor([0.0940, 0.0719, 0.0990, 0.0890], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 17:29:58,939 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1362306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:30:01,977 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1362309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:30:09,417 INFO [train.py:968] (0/2) Epoch 30, batch 41850, giga_loss[loss=0.2928, simple_loss=0.3623, pruned_loss=0.1117, over 28710.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3742, pruned_loss=0.1219, over 5651516.89 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3577, pruned_loss=0.1111, over 5677363.25 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3767, pruned_loss=0.1237, over 5649083.16 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:30:33,333 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:30:34,421 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.345e+03 1.877e+03 2.460e+03 3.724e+03 1.409e+04, threshold=4.920e+03, percent-clipped=14.0
+2023-03-15 17:30:49,398 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2630, 1.7509, 1.3798, 0.5433], device='cuda:0'), covar=tensor([0.4746, 0.3067, 0.4024, 0.6851], device='cuda:0'), in_proj_covar=tensor([0.1883, 0.1767, 0.1679, 0.1528], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 17:31:01,446 INFO [train.py:968] (0/2) Epoch 30, batch 41900, giga_loss[loss=0.2554, simple_loss=0.3386, pruned_loss=0.0861, over 28967.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.372, pruned_loss=0.1206, over 5650532.30 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3576, pruned_loss=0.111, over 5680675.85 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3742, pruned_loss=0.1222, over 5644989.23 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:31:07,917 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1362374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:31:51,747 INFO [train.py:968] (0/2) Epoch 30, batch 41950, giga_loss[loss=0.3567, simple_loss=0.3979, pruned_loss=0.1578, over 26742.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3724, pruned_loss=0.1209, over 5662858.96 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3576, pruned_loss=0.1108, over 5685542.23 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3747, pruned_loss=0.1226, over 5653548.38 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:32:01,305 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7345, 2.3452, 1.4281, 1.0933], device='cuda:0'), covar=tensor([0.8837, 0.4163, 0.4728, 0.7635], device='cuda:0'), in_proj_covar=tensor([0.1882, 0.1766, 0.1677, 0.1527], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 17:32:03,967 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.8389, 1.9227, 1.6811, 1.9750], device='cuda:0'), covar=tensor([0.2715, 0.2958, 0.3151, 0.2609], device='cuda:0'), in_proj_covar=tensor([0.1627, 0.1169, 0.1436, 0.1016], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 17:32:16,271 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.146e+03 1.653e+03 2.249e+03 2.975e+03 6.676e+03, threshold=4.497e+03, percent-clipped=3.0
+2023-03-15 17:32:18,702 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-15 17:32:20,465 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1362445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:32:44,941 INFO [train.py:968] (0/2) Epoch 30, batch 42000, giga_loss[loss=0.2671, simple_loss=0.3404, pruned_loss=0.09689, over 28732.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.37, pruned_loss=0.1184, over 5679018.49 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3575, pruned_loss=0.1108, over 5692233.10 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3724, pruned_loss=0.1201, over 5664791.33 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:32:44,946 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 17:32:54,621 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.1996, simple_loss=0.3068, pruned_loss=0.04618, over 944034.00 frames. 
+2023-03-15 17:32:54,622 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 17:33:49,771 INFO [train.py:968] (0/2) Epoch 30, batch 42050, giga_loss[loss=0.3081, simple_loss=0.3859, pruned_loss=0.1152, over 28344.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3692, pruned_loss=0.1161, over 5671815.74 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3575, pruned_loss=0.1108, over 5687233.47 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3712, pruned_loss=0.1175, over 5664609.34 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:34:10,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.078e+03 1.796e+03 2.363e+03 3.366e+03 8.833e+03, threshold=4.726e+03, percent-clipped=7.0
+2023-03-15 17:34:42,234 INFO [train.py:968] (0/2) Epoch 30, batch 42100, giga_loss[loss=0.3337, simple_loss=0.3941, pruned_loss=0.1367, over 27578.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3708, pruned_loss=0.1151, over 5669169.99 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3577, pruned_loss=0.111, over 5680503.89 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3724, pruned_loss=0.1162, over 5668697.18 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:35:33,050 INFO [train.py:968] (0/2) Epoch 30, batch 42150, giga_loss[loss=0.3358, simple_loss=0.3926, pruned_loss=0.1395, over 28739.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3722, pruned_loss=0.1169, over 5659741.57 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3581, pruned_loss=0.1113, over 5672707.09 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3733, pruned_loss=0.1176, over 5666430.33 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:35:54,000 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.268e+03 1.919e+03 2.395e+03 3.496e+03 1.539e+04, threshold=4.791e+03, percent-clipped=11.0
+2023-03-15 17:36:19,363 INFO [train.py:968] (0/2) Epoch 30, batch 42200, giga_loss[loss=0.2822, simple_loss=0.3608, pruned_loss=0.1018, over 28700.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.372, pruned_loss=0.1177, over 5668683.82 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3576, pruned_loss=0.111, over 5679411.22 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3737, pruned_loss=0.1187, over 5667900.70 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:36:26,218 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1362677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:37:01,907 INFO [train.py:968] (0/2) Epoch 30, batch 42250, giga_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1152, over 28417.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3721, pruned_loss=0.1188, over 5671244.56 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3584, pruned_loss=0.1115, over 5673790.67 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3733, pruned_loss=0.1193, over 5674786.47 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:37:25,729 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.342e+03 2.030e+03 2.563e+03 3.664e+03 1.081e+04, threshold=5.126e+03, percent-clipped=14.0
+2023-03-15 17:37:33,690 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1362749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:37:52,799 INFO [train.py:968] (0/2) Epoch 30, batch 42300, giga_loss[loss=0.3168, simple_loss=0.355, pruned_loss=0.1393, over 23506.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3689, pruned_loss=0.1178, over 5657695.75 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3581, pruned_loss=0.1113, over 5676549.26 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3704, pruned_loss=0.1185, over 5657745.23 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:38:22,769 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-15 17:38:27,194 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1362808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:38:38,939 INFO [train.py:968] (0/2) Epoch 30, batch 42350, giga_loss[loss=0.3092, simple_loss=0.383, pruned_loss=0.1177, over 28566.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3656, pruned_loss=0.1152, over 5671609.25 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3571, pruned_loss=0.1107, over 5684393.62 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3682, pruned_loss=0.1167, over 5664132.48 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:38:40,706 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1362820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:38:56,675 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3962, 1.7958, 1.6311, 1.5108], device='cuda:0'), covar=tensor([0.2437, 0.2206, 0.2721, 0.2616], device='cuda:0'), in_proj_covar=tensor([0.0515, 0.0766, 0.0736, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:0')
+2023-03-15 17:39:03,264 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.157e+03 1.706e+03 2.268e+03 2.970e+03 6.545e+03, threshold=4.536e+03, percent-clipped=3.0
+2023-03-15 17:39:27,908 INFO [train.py:968] (0/2) Epoch 30, batch 42400, libri_loss[loss=0.2977, simple_loss=0.3584, pruned_loss=0.1185, over 29568.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3649, pruned_loss=0.1134, over 5682742.28 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3574, pruned_loss=0.1108, over 5688673.48 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.367, pruned_loss=0.1145, over 5672504.41 frames. ], batch size: 79, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:39:49,146 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1362892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:39:54,377 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1362895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:40:13,964 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 17:40:14,885 INFO [train.py:968] (0/2) Epoch 30, batch 42450, giga_loss[loss=0.2961, simple_loss=0.3694, pruned_loss=0.1114, over 28817.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3649, pruned_loss=0.1126, over 5677028.47 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3579, pruned_loss=0.1112, over 5685261.09 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3663, pruned_loss=0.1132, over 5672238.79 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:40:15,844 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3817, 2.0022, 1.4982, 0.6785], device='cuda:0'), covar=tensor([0.6371, 0.3127, 0.3966, 0.7207], device='cuda:0'), in_proj_covar=tensor([0.1881, 0.1770, 0.1680, 0.1529], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 17:40:20,251 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:40:40,931 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.135e+03 1.760e+03 2.187e+03 2.847e+03 5.196e+03, threshold=4.375e+03, percent-clipped=4.0
+2023-03-15 17:41:03,248 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1362963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:41:04,043 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4997, 2.1846, 1.5582, 0.6885], device='cuda:0'), covar=tensor([0.7266, 0.3444, 0.4918, 0.7922], device='cuda:0'), in_proj_covar=tensor([0.1878, 0.1766, 0.1677, 0.1526], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 17:41:05,448 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1362966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:41:06,407 INFO [train.py:968] (0/2) Epoch 30, batch 42500, giga_loss[loss=0.2675, simple_loss=0.3462, pruned_loss=0.09437, over 28863.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3652, pruned_loss=0.1126, over 5690508.15 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3578, pruned_loss=0.1111, over 5688045.79 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3665, pruned_loss=0.1132, over 5684063.02 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:41:20,603 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-15 17:41:34,755 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:41:52,283 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-15 17:41:55,710 INFO [train.py:968] (0/2) Epoch 30, batch 42550, giga_loss[loss=0.299, simple_loss=0.3463, pruned_loss=0.1259, over 23698.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3631, pruned_loss=0.1122, over 5684532.76 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3573, pruned_loss=0.1108, over 5692493.47 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3647, pruned_loss=0.1129, over 5675146.04 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:42:16,764 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.112e+03 1.757e+03 2.267e+03 3.091e+03 5.841e+03, threshold=4.534e+03, percent-clipped=8.0
+2023-03-15 17:42:29,099 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1363052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:42:42,979 INFO [train.py:968] (0/2) Epoch 30, batch 42600, giga_loss[loss=0.2534, simple_loss=0.3306, pruned_loss=0.08808, over 28922.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3632, pruned_loss=0.113, over 5667567.63 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3577, pruned_loss=0.111, over 5675467.83 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3643, pruned_loss=0.1134, over 5674399.50 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:43:30,778 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5313, 1.7255, 1.3218, 1.6062], device='cuda:0'), covar=tensor([0.0720, 0.0309, 0.0346, 0.0878], device='cuda:0'), in_proj_covar=tensor([0.0196, 0.0124, 0.0122, 0.0233], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0118], device='cuda:0')
+2023-03-15 17:43:32,556 INFO [train.py:968] (0/2) Epoch 30, batch 42650, giga_loss[loss=0.3108, simple_loss=0.3747, pruned_loss=0.1235, over 28863.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3633, pruned_loss=0.1145, over 5674508.65 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3576, pruned_loss=0.1112, over 5684498.57 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3644, pruned_loss=0.1147, over 5671349.96 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:43:58,290 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 2.000e+03 2.406e+03 3.980e+03 1.143e+04, threshold=4.813e+03, percent-clipped=15.0
+2023-03-15 17:44:01,028 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([4.3264, 4.1687, 3.9801, 1.8557], device='cuda:0'), covar=tensor([0.0701, 0.0801, 0.0805, 0.2022], device='cuda:0'), in_proj_covar=tensor([0.1341, 0.1238, 0.1037, 0.0765], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 17:44:01,161 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5898, 1.6749, 1.7915, 1.3630], device='cuda:0'), covar=tensor([0.1884, 0.2751, 0.1546, 0.1810], device='cuda:0'), in_proj_covar=tensor([0.0942, 0.0721, 0.0993, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 17:44:02,839 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-15 17:44:27,290 INFO [train.py:968] (0/2) Epoch 30, batch 42700, giga_loss[loss=0.2845, simple_loss=0.353, pruned_loss=0.108, over 28515.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3624, pruned_loss=0.1145, over 5671854.49 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3575, pruned_loss=0.111, over 5686771.78 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3635, pruned_loss=0.115, over 5667155.55 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:44:29,699 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-15 17:44:41,520 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1363183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:44:52,054 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1363195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:44:55,051 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1363198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:45:00,446 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3424, 3.1603, 1.4361, 1.5378], device='cuda:0'), covar=tensor([0.1049, 0.0424, 0.0905, 0.1385], device='cuda:0'), in_proj_covar=tensor([0.0433, 0.0583, 0.0419, 0.0455], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 17:45:14,494 INFO [train.py:968] (0/2) Epoch 30, batch 42750, giga_loss[loss=0.286, simple_loss=0.3596, pruned_loss=0.1062, over 28986.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3605, pruned_loss=0.1133, over 5676663.83 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3572, pruned_loss=0.1109, over 5685964.49 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5672483.04 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:45:24,035 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1363227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:45:36,478 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.284e+03 1.907e+03 2.302e+03 3.044e+03 8.066e+03, threshold=4.605e+03, percent-clipped=8.0
+2023-03-15 17:45:38,853 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1363242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:46:03,906 INFO [train.py:968] (0/2) Epoch 30, batch 42800, giga_loss[loss=0.3175, simple_loss=0.3835, pruned_loss=0.1258, over 28547.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3613, pruned_loss=0.1143, over 5682383.24 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.357, pruned_loss=0.1108, over 5688800.73 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5676346.20 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:46:47,818 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1363311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:46:51,013 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2806, 1.4721, 1.3469, 1.5397], device='cuda:0'), covar=tensor([0.0804, 0.0366, 0.0353, 0.0900], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 17:46:53,572 INFO [train.py:968] (0/2) Epoch 30, batch 42850, libri_loss[loss=0.3392, simple_loss=0.3947, pruned_loss=0.1419, over 29650.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3617, pruned_loss=0.1136, over 5687850.96 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3566, pruned_loss=0.1106, over 5693184.28 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.363, pruned_loss=0.1143, over 5679130.20 frames. ], batch size: 91, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:47:01,167 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1363326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:47:05,030 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1363329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:47:17,607 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.107e+03 1.950e+03 2.505e+03 3.639e+03 7.002e+03, threshold=5.009e+03, percent-clipped=15.0
+2023-03-15 17:47:32,433 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1363358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:47:41,011 INFO [train.py:968] (0/2) Epoch 30, batch 42900, libri_loss[loss=0.2826, simple_loss=0.3517, pruned_loss=0.1067, over 29573.00 frames. ], tot_loss[loss=0.294, simple_loss=0.362, pruned_loss=0.1131, over 5689400.59 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3564, pruned_loss=0.1105, over 5695929.40 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3633, pruned_loss=0.1138, over 5679845.48 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:47:43,827 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-15 17:48:18,251 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1363406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:48:29,877 INFO [train.py:968] (0/2) Epoch 30, batch 42950, giga_loss[loss=0.3017, simple_loss=0.3713, pruned_loss=0.116, over 29012.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3625, pruned_loss=0.1126, over 5682743.22 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3563, pruned_loss=0.1102, over 5698821.57 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3638, pruned_loss=0.1134, over 5672765.37 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:48:31,587 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.0158, 1.2300, 1.2345, 0.9988], device='cuda:0'), covar=tensor([0.2257, 0.2651, 0.1528, 0.2231], device='cuda:0'), in_proj_covar=tensor([0.2106, 0.2065, 0.1970, 0.2120], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 17:48:55,337 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.266e+03 1.659e+03 2.023e+03 2.635e+03 6.909e+03, threshold=4.045e+03, percent-clipped=3.0
+2023-03-15 17:49:23,855 INFO [train.py:968] (0/2) Epoch 30, batch 43000, giga_loss[loss=0.319, simple_loss=0.3847, pruned_loss=0.1267, over 28299.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3637, pruned_loss=0.1141, over 5675674.26 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3567, pruned_loss=0.1104, over 5704115.33 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3645, pruned_loss=0.1147, over 5662376.65 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:50:12,187 INFO [train.py:968] (0/2) Epoch 30, batch 43050, giga_loss[loss=0.3312, simple_loss=0.3851, pruned_loss=0.1387, over 28647.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3663, pruned_loss=0.117, over 5677512.45 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3563, pruned_loss=0.1101, over 5710488.17 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3676, pruned_loss=0.1179, over 5660180.58 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:50:37,547 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.104e+03 1.974e+03 2.571e+03 3.249e+03 1.005e+04, threshold=5.142e+03, percent-clipped=10.0
+2023-03-15 17:51:06,174 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4950, 1.7050, 1.6032, 1.3051], device='cuda:0'), covar=tensor([0.3252, 0.2802, 0.2360, 0.3013], device='cuda:0'), in_proj_covar=tensor([0.2102, 0.2061, 0.1965, 0.2114], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0')
+2023-03-15 17:51:10,948 INFO [train.py:968] (0/2) Epoch 30, batch 43100, libri_loss[loss=0.2606, simple_loss=0.3404, pruned_loss=0.0904, over 29530.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3686, pruned_loss=0.1202, over 5669997.59 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3564, pruned_loss=0.11, over 5712457.30 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3697, pruned_loss=0.1211, over 5654273.05 frames. ], batch size: 83, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:52:01,609 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1363617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:52:02,113 INFO [train.py:968] (0/2) Epoch 30, batch 43150, giga_loss[loss=0.2792, simple_loss=0.3494, pruned_loss=0.1045, over 28733.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3697, pruned_loss=0.1224, over 5666671.56 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.356, pruned_loss=0.1098, over 5714625.62 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3711, pruned_loss=0.1234, over 5651573.93 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:52:29,212 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.222e+03 1.981e+03 2.471e+03 3.063e+03 8.689e+03, threshold=4.942e+03, percent-clipped=6.0
+2023-03-15 17:52:54,707 INFO [train.py:968] (0/2) Epoch 30, batch 43200, giga_loss[loss=0.2783, simple_loss=0.3517, pruned_loss=0.1024, over 28824.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3702, pruned_loss=0.1228, over 5671755.98 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3563, pruned_loss=0.11, over 5715647.71 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3712, pruned_loss=0.1236, over 5658261.20 frames. ], batch size: 112, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:52:58,211 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-15 17:53:10,851 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1363686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:53:28,265 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1363706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:53:30,310 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.9234, 1.1526, 2.8645, 2.8105], device='cuda:0'), covar=tensor([0.1659, 0.2638, 0.0632, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0823, 0.0685, 0.1025, 0.1000], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:0')
+2023-03-15 17:53:39,166 INFO [train.py:968] (0/2) Epoch 30, batch 43250, giga_loss[loss=0.286, simple_loss=0.3608, pruned_loss=0.1056, over 29045.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3673, pruned_loss=0.1209, over 5681920.65 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3557, pruned_loss=0.1096, over 5721326.86 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3691, pruned_loss=0.1222, over 5665033.15 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:53:59,465 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.298e+03 1.775e+03 2.174e+03 2.771e+03 5.258e+03, threshold=4.349e+03, percent-clipped=2.0
+2023-03-15 17:54:18,600 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1363760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:54:18,640 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7765, 1.7651, 1.9322, 1.5247], device='cuda:0'), covar=tensor([0.1973, 0.2664, 0.1622, 0.1936], device='cuda:0'), in_proj_covar=tensor([0.0943, 0.0722, 0.0993, 0.0892], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 17:54:20,596 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1363763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:54:24,955 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-15 17:54:25,125 INFO [train.py:968] (0/2) Epoch 30, batch 43300, giga_loss[loss=0.2714, simple_loss=0.3512, pruned_loss=0.09578, over 28683.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3667, pruned_loss=0.1186, over 5688181.10 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3556, pruned_loss=0.1094, over 5724976.57 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3684, pruned_loss=0.12, over 5671207.00 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:54:38,285 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1363781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:54:40,304 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1882, 3.4710, 2.3137, 1.3930], device='cuda:0'), covar=tensor([0.9759, 0.5109, 0.4193, 0.8458], device='cuda:0'), in_proj_covar=tensor([0.1882, 0.1773, 0.1683, 0.1535], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:0')
+2023-03-15 17:54:50,139 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1363792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:54:54,366 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-15 17:55:16,785 INFO [train.py:968] (0/2) Epoch 30, batch 43350, giga_loss[loss=0.3055, simple_loss=0.3739, pruned_loss=0.1185, over 29021.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3649, pruned_loss=0.1165, over 5680211.17 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.356, pruned_loss=0.1096, over 5725017.21 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.366, pruned_loss=0.1176, over 5665823.08 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:55:26,453 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1363829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:55:29,249 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1363832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:55:37,053 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-15 17:55:38,670 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.180e+03 1.884e+03 2.430e+03 3.304e+03 7.127e+03, threshold=4.860e+03, percent-clipped=13.0
+2023-03-15 17:55:59,078 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1363861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:56:04,505 INFO [train.py:968] (0/2) Epoch 30, batch 43400, giga_loss[loss=0.305, simple_loss=0.372, pruned_loss=0.119, over 28495.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3631, pruned_loss=0.116, over 5674604.71 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3561, pruned_loss=0.1097, over 5725895.52 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3641, pruned_loss=0.1169, over 5661833.40 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:56:51,446 INFO [train.py:968] (0/2) Epoch 30, batch 43450, giga_loss[loss=0.2782, simple_loss=0.3492, pruned_loss=0.1037, over 28929.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3622, pruned_loss=0.1158, over 5681349.30 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3562, pruned_loss=0.1096, over 5731402.39 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.363, pruned_loss=0.1168, over 5664774.07 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:56:57,593 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1363924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:56:59,555 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1363927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:57:15,370 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.161e+03 1.779e+03 2.545e+03 3.274e+03 1.020e+04, threshold=5.089e+03, percent-clipped=8.0
+2023-03-15 17:57:27,753 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1363956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:57:37,949 INFO [train.py:968] (0/2) Epoch 30, batch 43500, giga_loss[loss=0.2875, simple_loss=0.3627, pruned_loss=0.1061, over 29021.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.362, pruned_loss=0.1158, over 5674572.75 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3561, pruned_loss=0.1096, over 5726861.10 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3629, pruned_loss=0.1168, over 5664004.88 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:58:08,785 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/checkpoint-1364000.pt
+2023-03-15 17:58:19,210 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.4151, 2.6462, 1.4249, 1.5899], device='cuda:0'), covar=tensor([0.0928, 0.0418, 0.0895, 0.1253], device='cuda:0'), in_proj_covar=tensor([0.0435, 0.0585, 0.0420, 0.0456], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 17:58:26,201 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=3.10 vs. limit=5.0
+2023-03-15 17:58:26,350 INFO [train.py:968] (0/2) Epoch 30, batch 43550, giga_loss[loss=0.3013, simple_loss=0.3654, pruned_loss=0.1185, over 28751.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3663, pruned_loss=0.1183, over 5675812.99 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3566, pruned_loss=0.1102, over 5729793.22 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3667, pruned_loss=0.1187, over 5663256.15 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:58:50,400 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.302e+03 1.908e+03 2.363e+03 3.379e+03 8.757e+03, threshold=4.727e+03, percent-clipped=8.0
+2023-03-15 17:58:53,703 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.3639, 1.6577, 1.6977, 1.4884], device='cuda:0'), covar=tensor([0.1957, 0.1639, 0.1682, 0.1782], device='cuda:0'), in_proj_covar=tensor([0.0511, 0.0760, 0.0729, 0.0701], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:0')
+2023-03-15 17:59:03,244 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5288, 1.7260, 1.4414, 1.5866], device='cuda:0'), covar=tensor([0.2897, 0.2986, 0.3344, 0.2585], device='cuda:0'), in_proj_covar=tensor([0.1630, 0.1169, 0.1439, 0.1017], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:0')
+2023-03-15 17:59:14,327 INFO [train.py:968] (0/2) Epoch 30, batch 43600, giga_loss[loss=0.2844, simple_loss=0.37, pruned_loss=0.09945, over 29019.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.369, pruned_loss=0.1173, over 5674560.69 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3564, pruned_loss=0.11, over 5733748.52 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3699, pruned_loss=0.1181, over 5659419.11 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:59:31,130 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1364081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:59:34,780 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-15 18:00:07,186 INFO [train.py:968] (0/2) Epoch 30, batch 43650, giga_loss[loss=0.3271, simple_loss=0.3905, pruned_loss=0.1319, over 28559.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3703, pruned_loss=0.1178, over 5678000.02 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3562, pruned_loss=0.1099, over 5737832.72 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3716, pruned_loss=0.1187, over 5660185.95 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 18:00:33,058 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.062e+03 1.842e+03 2.325e+03 3.241e+03 7.277e+03, threshold=4.650e+03, percent-clipped=9.0
+2023-03-15 18:00:56,835 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-15 18:00:57,623 INFO [train.py:968] (0/2) Epoch 30, batch 43700, giga_loss[loss=0.3171, simple_loss=0.381, pruned_loss=0.1265, over 28811.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3721, pruned_loss=0.1191, over 5678490.29 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3561, pruned_loss=0.1098, over 5733308.44 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3736, pruned_loss=0.1201, over 5667196.03 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:01:29,176 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:01:42,205 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 18:01:47,608 INFO [train.py:968] (0/2) Epoch 30, batch 43750, giga_loss[loss=0.3365, simple_loss=0.3892, pruned_loss=0.1419, over 28549.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3735, pruned_loss=0.121, over 5671040.88 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.356, pruned_loss=0.1099, over 5732372.79 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.375, pruned_loss=0.1219, over 5662334.75 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:01:52,864 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1364224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:01:54,615 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1364227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:01:59,387 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.2845, 1.6043, 1.5573, 1.3952], device='cuda:0'), covar=tensor([0.0789, 0.0335, 0.0309, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:0'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:0')
+2023-03-15 18:02:08,052 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.274e+03 1.932e+03 2.493e+03 3.379e+03 7.026e+03, threshold=4.986e+03, percent-clipped=7.0
+2023-03-15 18:02:19,159 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1364256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:02:27,663 INFO [train.py:968] (0/2) Epoch 30, batch 43800, giga_loss[loss=0.2921, simple_loss=0.3542, pruned_loss=0.115, over 28805.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3725, pruned_loss=0.121, over 5679558.76 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3558, pruned_loss=0.1099, over 5730455.38 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3746, pruned_loss=0.1222, over 5671330.68 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:02:45,151 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:03:17,915 INFO [train.py:968] (0/2) Epoch 30, batch 43850, libri_loss[loss=0.2695, simple_loss=0.347, pruned_loss=0.09597, over 29526.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3717, pruned_loss=0.1214, over 5673013.45 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3556, pruned_loss=0.1096, over 5734850.13 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3741, pruned_loss=0.1229, over 5660961.14 frames. ], batch size: 84, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:03:40,783 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.095e+03 1.899e+03 2.426e+03 3.008e+03 6.987e+03, threshold=4.852e+03, percent-clipped=4.0
+2023-03-15 18:03:45,923 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:04:04,213 INFO [train.py:968] (0/2) Epoch 30, batch 43900, giga_loss[loss=0.2601, simple_loss=0.3353, pruned_loss=0.09244, over 28879.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3696, pruned_loss=0.1208, over 5668608.97 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3556, pruned_loss=0.1097, over 5731149.12 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3722, pruned_loss=0.1224, over 5659173.14 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:04:52,027 INFO [train.py:968] (0/2) Epoch 30, batch 43950, giga_loss[loss=0.3293, simple_loss=0.3616, pruned_loss=0.1485, over 23681.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3689, pruned_loss=0.1209, over 5651869.48 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3556, pruned_loss=0.1098, over 5714071.90 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3713, pruned_loss=0.1224, over 5657028.19 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:05:22,394 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.372e+03 2.032e+03 2.739e+03 3.864e+03 1.270e+04, threshold=5.477e+03, percent-clipped=15.0
+2023-03-15 18:05:46,233 INFO [train.py:968] (0/2) Epoch 30, batch 44000, giga_loss[loss=0.2895, simple_loss=0.367, pruned_loss=0.106, over 28660.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3701, pruned_loss=0.1223, over 5637965.06 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3553, pruned_loss=0.1096, over 5705988.68 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3726, pruned_loss=0.1239, over 5647563.11 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 18:06:36,115 INFO [train.py:968] (0/2) Epoch 30, batch 44050, giga_loss[loss=0.2701, simple_loss=0.3348, pruned_loss=0.1027, over 28631.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.368, pruned_loss=0.1213, over 5648235.96 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3555, pruned_loss=0.1099, over 5713146.06 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3703, pruned_loss=0.1228, over 5647666.89 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:06:57,220 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.350e+03 1.918e+03 2.516e+03 3.007e+03 6.977e+03, threshold=5.031e+03, percent-clipped=2.0
+2023-03-15 18:07:20,358 INFO [train.py:968] (0/2) Epoch 30, batch 44100, giga_loss[loss=0.328, simple_loss=0.3843, pruned_loss=0.1358, over 27690.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3672, pruned_loss=0.1206, over 5646764.27 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3558, pruned_loss=0.1101, over 5697810.96 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3691, pruned_loss=0.1219, over 5658486.10 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:07:28,191 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1364575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:08:07,407 INFO [train.py:968] (0/2) Epoch 30, batch 44150, giga_loss[loss=0.2631, simple_loss=0.3337, pruned_loss=0.09625, over 28829.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3657, pruned_loss=0.1196, over 5647472.22 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3555, pruned_loss=0.1098, over 5701856.40 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3677, pruned_loss=0.121, over 5652351.83 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:08:33,543 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.061e+03 1.765e+03 2.191e+03 2.693e+03 7.449e+03, threshold=4.381e+03, percent-clipped=4.0
+2023-03-15 18:08:53,942 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1364660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:09:01,719 INFO [train.py:968] (0/2) Epoch 30, batch 44200, giga_loss[loss=0.4138, simple_loss=0.4346, pruned_loss=0.1965, over 26767.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3679, pruned_loss=0.1201, over 5650942.57 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3557, pruned_loss=0.1099, over 5705596.71 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3696, pruned_loss=0.1214, over 5650540.63 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:09:23,974 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:09:38,274 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364706.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:09:49,583 INFO [train.py:968] (0/2) Epoch 30, batch 44250, libri_loss[loss=0.3066, simple_loss=0.3792, pruned_loss=0.117, over 29142.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3692, pruned_loss=0.1208, over 5655731.15 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3552, pruned_loss=0.1095, over 5710370.16 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3714, pruned_loss=0.1225, over 5649307.35 frames. ], batch size: 101, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:09:49,888 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1364718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:09:53,362 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1364721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:09:59,504 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1364726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:10:12,819 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.7327, 1.7354, 1.9037, 1.4800], device='cuda:0'), covar=tensor([0.1765, 0.2634, 0.1486, 0.1757], device='cuda:0'), in_proj_covar=tensor([0.0944, 0.0723, 0.0995, 0.0894], device='cuda:0'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:0')
+2023-03-15 18:10:18,542 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.108e+03 1.653e+03 2.108e+03 3.166e+03 7.677e+03, threshold=4.215e+03, percent-clipped=16.0
+2023-03-15 18:10:22,612 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1364750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:10:24,811 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:10:39,875 INFO [train.py:968] (0/2) Epoch 30, batch 44300, giga_loss[loss=0.3385, simple_loss=0.3868, pruned_loss=0.1451, over 27542.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3688, pruned_loss=0.1211, over 5656734.15 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3553, pruned_loss=0.1099, over 5704784.54 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3707, pruned_loss=0.1224, over 5655050.77 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:11:12,911 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1364803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:11:15,024 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1364806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:11:27,042 INFO [train.py:968] (0/2) Epoch 30, batch 44350, giga_loss[loss=0.3277, simple_loss=0.3923, pruned_loss=0.1315, over 28601.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3703, pruned_loss=0.1189, over 5664930.31 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3555, pruned_loss=0.1101, over 5699926.48 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.372, pruned_loss=0.12, over 5666378.97 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:11:34,234 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364826.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:11:41,020 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1364835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:11:48,222 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.233e+03 1.754e+03 2.177e+03 2.855e+03 6.710e+03, threshold=4.353e+03, percent-clipped=8.0
+2023-03-15 18:12:08,125 INFO [train.py:968] (0/2) Epoch 30, batch 44400, giga_loss[loss=0.3441, simple_loss=0.4094, pruned_loss=0.1395, over 28930.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3706, pruned_loss=0.1173, over 5660669.33 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3563, pruned_loss=0.1106, over 5697977.82 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3721, pruned_loss=0.1182, over 5661818.78 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 18:12:09,213 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1364869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:12:11,584 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1364872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:12:38,297 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1364901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:12:55,311 INFO [train.py:968] (0/2) Epoch 30, batch 44450, giga_loss[loss=0.4443, simple_loss=0.4461, pruned_loss=0.2212, over 23341.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3726, pruned_loss=0.1188, over 5662684.96 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3559, pruned_loss=0.1105, over 5706667.88 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.375, pruned_loss=0.12, over 5653777.72 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:13:22,039 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.141e+03 1.852e+03 2.348e+03 2.950e+03 7.999e+03, threshold=4.696e+03, percent-clipped=10.0
+2023-03-15 18:13:45,300 INFO [train.py:968] (0/2) Epoch 30, batch 44500, giga_loss[loss=0.3848, simple_loss=0.43, pruned_loss=0.1698, over 28376.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3762, pruned_loss=0.1221, over 5659457.60 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.356, pruned_loss=0.1107, over 5703590.66 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3783, pruned_loss=0.123, over 5654488.91 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:14:35,787 INFO [train.py:968] (0/2) Epoch 30, batch 44550, giga_loss[loss=0.3147, simple_loss=0.3798, pruned_loss=0.1248, over 28631.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3787, pruned_loss=0.1251, over 5650139.46 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3562, pruned_loss=0.1108, over 5697545.07 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3807, pruned_loss=0.126, over 5650978.97 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:14:36,661 INFO [zipformer.py:1188] (0/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1365019.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:15:04,472 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 2.054e+03 2.631e+03 3.475e+03 1.013e+04, threshold=5.263e+03, percent-clipped=7.0
+2023-03-15 18:15:21,764 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1365064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:15:24,794 INFO [train.py:968] (0/2) Epoch 30, batch 44600, giga_loss[loss=0.2935, simple_loss=0.3575, pruned_loss=0.1148, over 28838.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3768, pruned_loss=0.1239, over 5665836.19 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.356, pruned_loss=0.1107, over 5698906.72 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3789, pruned_loss=0.125, over 5664568.45 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:15:34,431 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1365081.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:16:10,340 INFO [train.py:968] (0/2) Epoch 30, batch 44650, giga_loss[loss=0.2833, simple_loss=0.3548, pruned_loss=0.1059, over 28871.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3759, pruned_loss=0.1233, over 5664127.41 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3559, pruned_loss=0.1106, over 5701501.49 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.378, pruned_loss=0.1245, over 5660553.68 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:16:20,430 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1365128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:16:35,238 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.183e+03 1.673e+03 2.114e+03 2.715e+03 6.498e+03, threshold=4.228e+03, percent-clipped=2.0
+2023-03-15 18:16:45,876 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-15 18:16:56,132 INFO [train.py:968] (0/2) Epoch 30, batch 44700, libri_loss[loss=0.2975, simple_loss=0.3638, pruned_loss=0.1156, over 29742.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.375, pruned_loss=0.1208, over 5679536.60 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3561, pruned_loss=0.1106, over 5705022.70 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3771, pruned_loss=0.1221, over 5672491.25 frames. ], batch size: 87, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:17:29,527 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1365201.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:17:36,453 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1365207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:17:38,501 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1365210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:17:44,769 INFO [train.py:968] (0/2) Epoch 30, batch 44750, giga_loss[loss=0.2813, simple_loss=0.3517, pruned_loss=0.1055, over 28808.00 frames. ], tot_loss[loss=0.307, simple_loss=0.375, pruned_loss=0.1195, over 5673487.58 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3562, pruned_loss=0.1107, over 5694859.68 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3767, pruned_loss=0.1205, over 5675666.47 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:17:49,067 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1365224.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:17:51,260 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1365227.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:17:55,740 INFO [scaling.py:679] (0/2) Whitening: num_groups=1, num_channels=128, metric=2.47 vs. limit=5.0
+2023-03-15 18:18:02,620 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1365239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:18:10,912 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.241e+03 1.848e+03 2.254e+03 2.708e+03 6.514e+03, threshold=4.507e+03, percent-clipped=4.0
+2023-03-15 18:18:21,700 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1365256.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:18:29,993 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([3.3221, 3.1453, 3.0033, 1.5003], device='cuda:0'), covar=tensor([0.1040, 0.1236, 0.1034, 0.2172], device='cuda:0'), in_proj_covar=tensor([0.1351, 0.1246, 0.1044, 0.0769], device='cuda:0'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:0')
+2023-03-15 18:18:32,861 INFO [train.py:968] (0/2) Epoch 30, batch 44800, giga_loss[loss=0.3034, simple_loss=0.3698, pruned_loss=0.1185, over 28900.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3754, pruned_loss=0.1205, over 5656470.73 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.356, pruned_loss=0.1105, over 5690663.17 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3776, pruned_loss=0.1217, over 5661776.23 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 18:18:37,267 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1365271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:18:41,600 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1365274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:19:11,882 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1365303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:19:25,096 INFO [train.py:968] (0/2) Epoch 30, batch 44850, giga_loss[loss=0.2857, simple_loss=0.3573, pruned_loss=0.1071, over 28850.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3756, pruned_loss=0.1212, over 5636718.74 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3562, pruned_loss=0.1108, over 5673561.34 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3772, pruned_loss=0.122, over 5656552.98 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 18:19:45,536 INFO [scaling.py:679] (0/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-15 18:19:48,027 INFO [zipformer.py:1188] (0/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1365344.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:19:51,052 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.5133, 4.6068, 1.7430, 1.5807], device='cuda:0'), covar=tensor([0.1051, 0.0325, 0.0926, 0.1449], device='cuda:0'), in_proj_covar=tensor([0.0434, 0.0583, 0.0419, 0.0454], device='cuda:0'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:0')
+2023-03-15 18:19:51,080 INFO [zipformer.py:1188] (0/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1365347.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:19:51,737 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.292e+03 1.923e+03 2.602e+03 3.658e+03 8.345e+03, threshold=5.204e+03, percent-clipped=11.0
+2023-03-15 18:20:11,062 INFO [train.py:968] (0/2) Epoch 30, batch 44900, giga_loss[loss=0.27, simple_loss=0.3437, pruned_loss=0.0981, over 28636.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3731, pruned_loss=0.1201, over 5589040.59 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3571, pruned_loss=0.1115, over 5611136.34 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3741, pruned_loss=0.1204, over 5659001.98 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 18:20:18,054 INFO [zipformer.py:1188] (0/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1365376.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:20:37,985 INFO [zipformer.py:1188] (0/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1365394.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:20:42,678 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([1.6601, 2.0626, 1.6392, 1.6969], device='cuda:0'), covar=tensor([0.2699, 0.2654, 0.3146, 0.2489], device='cuda:0'), in_proj_covar=tensor([0.1636, 0.1174, 0.1445, 0.1023], device='cuda:0'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0010], device='cuda:0')
+2023-03-15 18:21:02,639 INFO [train.py:968] (0/2) Epoch 30, batch 44950, giga_loss[loss=0.2686, simple_loss=0.3463, pruned_loss=0.09545, over 28932.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3725, pruned_loss=0.1212, over 5569910.08 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3576, pruned_loss=0.1119, over 5578282.42 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3731, pruned_loss=0.1212, over 5653944.50 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 18:21:17,046 INFO [zipformer.py:2441] (0/2) attn_weights_entropy = tensor([2.1483, 1.3958, 3.7476, 3.1116], device='cuda:0'), covar=tensor([0.1881, 0.2711, 0.0537, 0.0982], device='cuda:0'), in_proj_covar=tensor([0.0829, 0.0688, 0.1030, 0.1008], device='cuda:0'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:0')
+2023-03-15 18:21:33,466 INFO [optim.py:369] (0/2) Clipping_scale=2.0, grad-norm quartiles 1.042e+03 1.781e+03 2.287e+03 3.032e+03 6.000e+03, threshold=4.575e+03, percent-clipped=2.0
+2023-03-15 18:21:51,569 INFO [train.py:968] (0/2) Epoch 30, batch 45000, giga_loss[loss=0.2764, simple_loss=0.36, pruned_loss=0.09641, over 28945.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3713, pruned_loss=0.1208, over 5560278.91 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3581, pruned_loss=0.1123, over 5545094.85 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3715, pruned_loss=0.1205, over 5659049.83 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 18:21:51,575 INFO [train.py:1003] (0/2) Computing validation loss
+2023-03-15 18:22:00,673 INFO [train.py:1012] (0/2) Epoch 30, validation: loss=0.2027, simple_loss=0.3123, pruned_loss=0.04655, over 944034.00 frames. 
+2023-03-15 18:22:00,674 INFO [train.py:1013] (0/2) Maximum memory allocated so far is 19815MB
+2023-03-15 18:22:21,915 INFO [train.py:866] (0/2) libri reaches end of dataloader
+2023-03-15 18:22:23,708 INFO [checkpoint.py:75] (0/2) Saving checkpoint to pruned_transducer_stateless7_streaming_multi/exp-small-6M/epoch-30.pt
+2023-03-15 18:22:24,017 INFO [train.py:1284] (0/2) Done!